projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Add some debugging around the ENOSPC bugs
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
29#include <linux/smp_lock.h>
30#include <linux/backing-dev.h>
31#include <linux/mpage.h>
32#include <linux/swap.h>
33#include <linux/writeback.h>
34#include <linux/statfs.h>
35#include <linux/compat.h>
9ebefb18 36#include <linux/bit_spinlock.h>
92fee66d 37#include <linux/version.h>
5103e947 38#include <linux/xattr.h>
33268eaf 39#include <linux/posix_acl.h>
d899e052 40#include <linux/falloc.h>
39279cc3
CM		 41#include "ctree.h"
42#include "disk-io.h"
43#include "transaction.h"
44#include "btrfs_inode.h"
45#include "ioctl.h"
46#include "print-tree.h"
0b86a832 47#include "volumes.h"
e6dcd2dc 48#include "ordered-data.h"
95819c05 49#include "xattr.h"
e02119d5
CM		 50#include "compat.h"
51#include "tree-log.h"
5b84e8d6 52#include "ref-cache.h"
c8b97818 53#include "compression.h"
39279cc3
CM		 54
55struct btrfs_iget_args {
56 u64 ino;
57 struct btrfs_root *root;
58};
59
60static struct inode_operations btrfs_dir_inode_operations;
61static struct inode_operations btrfs_symlink_inode_operations;
62static struct inode_operations btrfs_dir_ro_inode_operations;
618e21d5 63static struct inode_operations btrfs_special_inode_operations;
39279cc3
CM		 64static struct inode_operations btrfs_file_inode_operations;
65static struct address_space_operations btrfs_aops;
66static struct address_space_operations btrfs_symlink_aops;
67static struct file_operations btrfs_dir_file_operations;
d1310b2e 68static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 69
70static struct kmem_cache *btrfs_inode_cachep;
71struct kmem_cache *btrfs_trans_handle_cachep;
72struct kmem_cache *btrfs_transaction_cachep;
73struct kmem_cache *btrfs_bit_radix_cachep;
74struct kmem_cache *btrfs_path_cachep;
75
76#define S_SHIFT 12
77static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
85};
86
7b128766 87static void btrfs_truncate(struct inode *inode);
c8b97818 88static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
771ed689
CM		 89static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, int *page_started,
92 unsigned long *nr_written, int unlock);
7b128766 93
d352ac68
CM		 94/*
95 * a very lame attempt at stopping writes when the FS is 85% full. There
96 * are countless ways this is incorrect, but it is better than nothing.
97 */
1832a6d5
CM		 98int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
99 int for_del)
100{
a2135011
CM		 101 u64 total;
102 u64 used;
1832a6d5 103 u64 thresh;
bcbfce8a 104 unsigned long flags;
1832a6d5
CM		 105 int ret = 0;
106
a2135011
CM		 107 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
108 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
109 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
1832a6d5 110 if (for_del)
f9ef6604 111 thresh = total * 90;
1832a6d5 112 else
f9ef6604
CM		 113 thresh = total * 85;
114
115 do_div(thresh, 100);
1832a6d5 116
1832a6d5
CM		 117 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
118 ret = -ENOSPC;
bcbfce8a 119 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1832a6d5
CM		 120 return ret;
121}
122
c8b97818
CM		 123/*
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
127 */
128static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_root *root, struct inode *inode,
130 u64 start, size_t size, size_t compressed_size,
131 struct page **compressed_pages)
132{
133 struct btrfs_key key;
134 struct btrfs_path *path;
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
137 char *kaddr;
138 unsigned long ptr;
139 struct btrfs_file_extent_item *ei;
140 int err = 0;
141 int ret;
142 size_t cur_size = size;
143 size_t datasize;
144 unsigned long offset;
145 int use_compress = 0;
146
147 if (compressed_size && compressed_pages) {
148 use_compress = 1;
149 cur_size = compressed_size;
150 }
151
152 path = btrfs_alloc_path(); if (!path)
153 return -ENOMEM;
154
155 btrfs_set_trans_block_group(trans, inode);
156
157 key.objectid = inode->i_ino;
158 key.offset = start;
159 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
160 inode_add_bytes(inode, size);
161 datasize = btrfs_file_extent_calc_inline_size(cur_size);
162
163 inode_add_bytes(inode, size);
164 ret = btrfs_insert_empty_item(trans, root, path, &key,
165 datasize);
166 BUG_ON(ret);
167 if (ret) {
168 err = ret;
169 printk("got bad ret %d\n", ret);
170 goto fail;
171 }
172 leaf = path->nodes[0];
173 ei = btrfs_item_ptr(leaf, path->slots[0],
174 struct btrfs_file_extent_item);
175 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
176 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
177 btrfs_set_file_extent_encryption(leaf, ei, 0);
178 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
179 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
180 ptr = btrfs_file_extent_inline_start(ei);
181
182 if (use_compress) {
183 struct page *cpage;
184 int i = 0;
185 while(compressed_size > 0) {
186 cpage = compressed_pages[i];
5b050f04 187 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 188 PAGE_CACHE_SIZE);
189
190 kaddr = kmap(cpage);
191 write_extent_buffer(leaf, kaddr, ptr, cur_size);
192 kunmap(cpage);
193
194 i++;
195 ptr += cur_size;
196 compressed_size -= cur_size;
197 }
198 btrfs_set_file_extent_compression(leaf, ei,
199 BTRFS_COMPRESS_ZLIB);
200 } else {
201 page = find_get_page(inode->i_mapping,
202 start >> PAGE_CACHE_SHIFT);
203 btrfs_set_file_extent_compression(leaf, ei, 0);
204 kaddr = kmap_atomic(page, KM_USER0);
205 offset = start & (PAGE_CACHE_SIZE - 1);
206 write_extent_buffer(leaf, kaddr + offset, ptr, size);
207 kunmap_atomic(kaddr, KM_USER0);
208 page_cache_release(page);
209 }
210 btrfs_mark_buffer_dirty(leaf);
211 btrfs_free_path(path);
212
213 BTRFS_I(inode)->disk_i_size = inode->i_size;
214 btrfs_update_inode(trans, root, inode);
215 return 0;
216fail:
217 btrfs_free_path(path);
218 return err;
219}
220
221
222/*
223 * conditionally insert an inline extent into the file. This
224 * does the checks required to make sure the data is small enough
225 * to fit as an inline extent.
226 */
227static int cow_file_range_inline(struct btrfs_trans_handle *trans,
228 struct btrfs_root *root,
229 struct inode *inode, u64 start, u64 end,
230 size_t compressed_size,
231 struct page **compressed_pages)
232{
233 u64 isize = i_size_read(inode);
234 u64 actual_end = min(end + 1, isize);
235 u64 inline_len = actual_end - start;
236 u64 aligned_end = (end + root->sectorsize - 1) &
237 ~((u64)root->sectorsize - 1);
238 u64 hint_byte;
239 u64 data_len = inline_len;
240 int ret;
241
242 if (compressed_size)
243 data_len = compressed_size;
244
245 if (start > 0 ||
70b99e69 246 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 247 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
248 (!compressed_size &&
249 (actual_end & (root->sectorsize - 1)) == 0) ||
250 end + 1 < isize ||
251 data_len > root->fs_info->max_inline) {
252 return 1;
253 }
254
c8b97818 255 ret = btrfs_drop_extents(trans, root, inode, start,
70b99e69 256 aligned_end, start, &hint_byte);
c8b97818
CM		 257 BUG_ON(ret);
258
259 if (isize > actual_end)
260 inline_len = min_t(u64, isize, actual_end);
261 ret = insert_inline_extent(trans, root, inode, start,
262 inline_len, compressed_size,
263 compressed_pages);
264 BUG_ON(ret);
265 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
c8b97818
CM		 266 return 0;
267}
268
771ed689
CM		 269struct async_extent {
270 u64 start;
271 u64 ram_size;
272 u64 compressed_size;
273 struct page **pages;
274 unsigned long nr_pages;
275 struct list_head list;
276};
277
278struct async_cow {
279 struct inode *inode;
280 struct btrfs_root *root;
281 struct page *locked_page;
282 u64 start;
283 u64 end;
284 struct list_head extents;
285 struct btrfs_work work;
286};
287
288static noinline int add_async_extent(struct async_cow *cow,
289 u64 start, u64 ram_size,
290 u64 compressed_size,
291 struct page **pages,
292 unsigned long nr_pages)
293{
294 struct async_extent *async_extent;
295
296 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
297 async_extent->start = start;
298 async_extent->ram_size = ram_size;
299 async_extent->compressed_size = compressed_size;
300 async_extent->pages = pages;
301 async_extent->nr_pages = nr_pages;
302 list_add_tail(&async_extent->list, &cow->extents);
303 return 0;
304}
305
d352ac68 306/*
771ed689
CM		 307 * we create compressed extents in two phases. The first
308 * phase compresses a range of pages that have already been
309 * locked (both pages and state bits are locked).
c8b97818 310 *
771ed689
CM		 311 * This is done inside an ordered work queue, and the compression
312 * is spread across many cpus. The actual IO submission is step
313 * two, and the ordered work queue takes care of making sure that
314 * happens in the same order things were put onto the queue by
315 * writepages and friends.
c8b97818 316 *
771ed689
CM		 317 * If this code finds it can't get good compression, it puts an
318 * entry onto the work queue to write the uncompressed bytes. This
319 * makes sure that both compressed inodes and uncompressed inodes
320 * are written in the same order that pdflush sent them down.
d352ac68 321 */
771ed689
CM		 322static noinline int compress_file_range(struct inode *inode,
323 struct page *locked_page,
324 u64 start, u64 end,
325 struct async_cow *async_cow,
326 int *num_added)
b888db2b
CM		 327{
328 struct btrfs_root *root = BTRFS_I(inode)->root;
329 struct btrfs_trans_handle *trans;
db94535d 330 u64 num_bytes;
c8b97818
CM		 331 u64 orig_start;
332 u64 disk_num_bytes;
db94535d 333 u64 blocksize = root->sectorsize;
c8b97818 334 u64 actual_end;
e6dcd2dc 335 int ret = 0;
c8b97818
CM		 336 struct page **pages = NULL;
337 unsigned long nr_pages;
338 unsigned long nr_pages_ret = 0;
339 unsigned long total_compressed = 0;
340 unsigned long total_in = 0;
341 unsigned long max_compressed = 128 * 1024;
771ed689 342 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 343 int i;
344 int will_compress;
b888db2b 345
c8b97818
CM		 346 orig_start = start;
347
c8b97818
CM		 348again:
349 will_compress = 0;
350 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
351 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 352
c8b97818
CM		 353 actual_end = min_t(u64, i_size_read(inode), end + 1);
354 total_compressed = actual_end - start;
355
356 /* we want to make sure that amount of ram required to uncompress
357 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 358 * of a compressed extent to 128k. This is a crucial number
359 * because it also controls how easily we can spread reads across
360 * cpus for decompression.
361 *
362 * We also want to make sure the amount of IO required to do
363 * a random read is reasonably small, so we limit the size of
364 * a compressed extent to 128k.
c8b97818
CM		 365 */
366 total_compressed = min(total_compressed, max_uncompressed);
db94535d 367 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 368 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 369 disk_num_bytes = num_bytes;
370 total_in = 0;
371 ret = 0;
db94535d 372
771ed689
CM		 373 /*
374 * we do compression for mount -o compress and when the
375 * inode has not been flagged as nocompress. This flag can
376 * change at any time if we discover bad compression ratios.
c8b97818
CM		 377 */
378 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
379 btrfs_test_opt(root, COMPRESS)) {
380 WARN_ON(pages);
cfbc246e 381 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
c8b97818 382
c8b97818
CM		 383 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
384 total_compressed, pages,
385 nr_pages, &nr_pages_ret,
386 &total_in,
387 &total_compressed,
388 max_compressed);
389
390 if (!ret) {
391 unsigned long offset = total_compressed &
392 (PAGE_CACHE_SIZE - 1);
393 struct page *page = pages[nr_pages_ret - 1];
394 char *kaddr;
395
396 /* zero the tail end of the last page, we might be
397 * sending it down to disk
398 */
399 if (offset) {
400 kaddr = kmap_atomic(page, KM_USER0);
401 memset(kaddr + offset, 0,
402 PAGE_CACHE_SIZE - offset);
403 kunmap_atomic(kaddr, KM_USER0);
404 }
405 will_compress = 1;
406 }
407 }
408 if (start == 0) {
771ed689
CM		 409 trans = btrfs_join_transaction(root, 1);
410 BUG_ON(!trans);
411 btrfs_set_trans_block_group(trans, inode);
412
c8b97818 413 /* lets try to make an inline extent */
771ed689 414 if (ret || total_in < (actual_end - start)) {
c8b97818 415 /* we didn't compress the entire range, try
771ed689 416 * to make an uncompressed inline extent.
c8b97818
CM		 417 */
418 ret = cow_file_range_inline(trans, root, inode,
419 start, end, 0, NULL);
420 } else {
771ed689 421 /* try making a compressed inline extent */
c8b97818
CM		 422 ret = cow_file_range_inline(trans, root, inode,
423 start, end,
424 total_compressed, pages);
425 }
771ed689 426 btrfs_end_transaction(trans, root);
c8b97818 427 if (ret == 0) {
771ed689
CM		 428 /*
429 * inline extent creation worked, we don't need
430 * to create any more async work items. Unlock
431 * and free up our temp pages.
432 */
c8b97818
CM		 433 extent_clear_unlock_delalloc(inode,
434 &BTRFS_I(inode)->io_tree,
771ed689
CM		 435 start, end, NULL, 1, 0,
436 0, 1, 1, 1);
c8b97818
CM		 437 ret = 0;
438 goto free_pages_out;
439 }
440 }
441
442 if (will_compress) {
443 /*
444 * we aren't doing an inline extent round the compressed size
445 * up to a block size boundary so the allocator does sane
446 * things
447 */
448 total_compressed = (total_compressed + blocksize - 1) &
449 ~(blocksize - 1);
450
451 /*
452 * one last check to make sure the compression is really a
453 * win, compare the page count read with the blocks on disk
454 */
455 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
456 ~(PAGE_CACHE_SIZE - 1);
457 if (total_compressed >= total_in) {
458 will_compress = 0;
459 } else {
460 disk_num_bytes = total_compressed;
461 num_bytes = total_in;
462 }
463 }
464 if (!will_compress && pages) {
465 /*
466 * the compression code ran but failed to make things smaller,
467 * free any pages it allocated and our page pointer array
468 */
469 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 470 WARN_ON(pages[i]->mapping);
c8b97818
CM		 471 page_cache_release(pages[i]);
472 }
473 kfree(pages);
474 pages = NULL;
475 total_compressed = 0;
476 nr_pages_ret = 0;
477
478 /* flag the file so we don't compress in the future */
479 btrfs_set_flag(inode, NOCOMPRESS);
480 }
771ed689
CM		 481 if (will_compress) {
482 *num_added += 1;
c8b97818 483
771ed689
CM		 484 /* the async work queues will take care of doing actual
485 * allocation on disk for these compressed pages,
486 * and will submit them to the elevator.
487 */
488 add_async_extent(async_cow, start, num_bytes,
489 total_compressed, pages, nr_pages_ret);
179e29e4 490
771ed689
CM		 491 if (start + num_bytes < end) {
492 start += num_bytes;
493 pages = NULL;
494 cond_resched();
495 goto again;
496 }
497 } else {
498 /*
499 * No compression, but we still need to write the pages in
500 * the file we've been given so far. redirty the locked
501 * page if it corresponds to our extent and set things up
502 * for the async work queue to run cow_file_range to do
503 * the normal delalloc dance
504 */
505 if (page_offset(locked_page) >= start &&
506 page_offset(locked_page) <= end) {
507 __set_page_dirty_nobuffers(locked_page);
508 /* unlocked later on in the async handlers */
509 }
510 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
511 *num_added += 1;
512 }
3b951516 513
771ed689
CM		 514out:
515 return 0;
516
517free_pages_out:
518 for (i = 0; i < nr_pages_ret; i++) {
519 WARN_ON(pages[i]->mapping);
520 page_cache_release(pages[i]);
521 }
522 if (pages)
523 kfree(pages);
524
525 goto out;
526}
527
528/*
529 * phase two of compressed writeback. This is the ordered portion
530 * of the code, which only gets called in the order the work was
531 * queued. We walk all the async extents created by compress_file_range
532 * and send them down to the disk.
533 */
534static noinline int submit_compressed_extents(struct inode *inode,
535 struct async_cow *async_cow)
536{
537 struct async_extent *async_extent;
538 u64 alloc_hint = 0;
539 struct btrfs_trans_handle *trans;
540 struct btrfs_key ins;
541 struct extent_map *em;
542 struct btrfs_root *root = BTRFS_I(inode)->root;
543 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
544 struct extent_io_tree *io_tree;
545 int ret;
546
547 if (list_empty(&async_cow->extents))
548 return 0;
549
550 trans = btrfs_join_transaction(root, 1);
551
552 while(!list_empty(&async_cow->extents)) {
553 async_extent = list_entry(async_cow->extents.next,
554 struct async_extent, list);
555 list_del(&async_extent->list);
c8b97818 556
771ed689
CM		 557 io_tree = &BTRFS_I(inode)->io_tree;
558
559 /* did the compression code fall back to uncompressed IO? */
560 if (!async_extent->pages) {
561 int page_started = 0;
562 unsigned long nr_written = 0;
563
564 lock_extent(io_tree, async_extent->start,
565 async_extent->start + async_extent->ram_size - 1,
566 GFP_NOFS);
567
568 /* allocate blocks */
569 cow_file_range(inode, async_cow->locked_page,
570 async_extent->start,
571 async_extent->start +
572 async_extent->ram_size - 1,
573 &page_started, &nr_written, 0);
574
575 /*
576 * if page_started, cow_file_range inserted an
577 * inline extent and took care of all the unlocking
578 * and IO for us. Otherwise, we need to submit
579 * all those pages down to the drive.
580 */
581 if (!page_started)
582 extent_write_locked_range(io_tree,
583 inode, async_extent->start,
584 async_extent->start +
585 async_extent->ram_size - 1,
586 btrfs_get_extent,
587 WB_SYNC_ALL);
588 kfree(async_extent);
589 cond_resched();
590 continue;
591 }
592
593 lock_extent(io_tree, async_extent->start,
594 async_extent->start + async_extent->ram_size - 1,
595 GFP_NOFS);
c8b97818 596 /*
771ed689
CM		 597 * here we're doing allocation and writeback of the
598 * compressed pages
c8b97818 599 */
771ed689
CM		 600 btrfs_drop_extent_cache(inode, async_extent->start,
601 async_extent->start +
602 async_extent->ram_size - 1, 0);
603
604 ret = btrfs_reserve_extent(trans, root,
605 async_extent->compressed_size,
606 async_extent->compressed_size,
607 0, alloc_hint,
608 (u64)-1, &ins, 1);
609 BUG_ON(ret);
610 em = alloc_extent_map(GFP_NOFS);
611 em->start = async_extent->start;
612 em->len = async_extent->ram_size;
445a6944 613 em->orig_start = em->start;
c8b97818 614
771ed689
CM		 615 em->block_start = ins.objectid;
616 em->block_len = ins.offset;
617 em->bdev = root->fs_info->fs_devices->latest_bdev;
618 set_bit(EXTENT_FLAG_PINNED, &em->flags);
619 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
620
621 while(1) {
622 spin_lock(&em_tree->lock);
623 ret = add_extent_mapping(em_tree, em);
624 spin_unlock(&em_tree->lock);
625 if (ret != -EEXIST) {
626 free_extent_map(em);
627 break;
628 }
629 btrfs_drop_extent_cache(inode, async_extent->start,
630 async_extent->start +
631 async_extent->ram_size - 1, 0);
632 }
633
634 ret = btrfs_add_ordered_extent(inode, async_extent->start,
635 ins.objectid,
636 async_extent->ram_size,
637 ins.offset,
638 BTRFS_ORDERED_COMPRESSED);
639 BUG_ON(ret);
640
641 btrfs_end_transaction(trans, root);
642
643 /*
644 * clear dirty, set writeback and unlock the pages.
645 */
646 extent_clear_unlock_delalloc(inode,
647 &BTRFS_I(inode)->io_tree,
648 async_extent->start,
649 async_extent->start +
650 async_extent->ram_size - 1,
651 NULL, 1, 1, 0, 1, 1, 0);
652
653 ret = btrfs_submit_compressed_write(inode,
654 async_extent->start,
655 async_extent->ram_size,
656 ins.objectid,
657 ins.offset, async_extent->pages,
658 async_extent->nr_pages);
659
660 BUG_ON(ret);
661 trans = btrfs_join_transaction(root, 1);
662 alloc_hint = ins.objectid + ins.offset;
663 kfree(async_extent);
664 cond_resched();
665 }
666
667 btrfs_end_transaction(trans, root);
668 return 0;
669}
670
671/*
672 * when extent_io.c finds a delayed allocation range in the file,
673 * the call backs end up in this code. The basic idea is to
674 * allocate extents on disk for the range, and create ordered data structs
675 * in ram to track those extents.
676 *
677 * locked_page is the page that writepage had locked already. We use
678 * it to make sure we don't do extra locks or unlocks.
679 *
680 * *page_started is set to one if we unlock locked_page and do everything
681 * required to start IO on it. It may be clean and already done with
682 * IO when we return.
683 */
684static noinline int cow_file_range(struct inode *inode,
685 struct page *locked_page,
686 u64 start, u64 end, int *page_started,
687 unsigned long *nr_written,
688 int unlock)
689{
690 struct btrfs_root *root = BTRFS_I(inode)->root;
691 struct btrfs_trans_handle *trans;
692 u64 alloc_hint = 0;
693 u64 num_bytes;
694 unsigned long ram_size;
695 u64 disk_num_bytes;
696 u64 cur_alloc_size;
697 u64 blocksize = root->sectorsize;
698 u64 actual_end;
699 struct btrfs_key ins;
700 struct extent_map *em;
701 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
702 int ret = 0;
703
704 trans = btrfs_join_transaction(root, 1);
705 BUG_ON(!trans);
706 btrfs_set_trans_block_group(trans, inode);
707
708 actual_end = min_t(u64, i_size_read(inode), end + 1);
709
710 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
711 num_bytes = max(blocksize, num_bytes);
712 disk_num_bytes = num_bytes;
713 ret = 0;
714
715 if (start == 0) {
716 /* lets try to make an inline extent */
717 ret = cow_file_range_inline(trans, root, inode,
718 start, end, 0, NULL);
719 if (ret == 0) {
720 extent_clear_unlock_delalloc(inode,
721 &BTRFS_I(inode)->io_tree,
722 start, end, NULL, 1, 1,
723 1, 1, 1, 1);
724 *nr_written = *nr_written +
725 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
726 *page_started = 1;
727 ret = 0;
728 goto out;
729 }
730 }
731
732 BUG_ON(disk_num_bytes >
733 btrfs_super_total_bytes(&root->fs_info->super_copy));
734
735 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
736
737 while(disk_num_bytes > 0) {
c8b97818 738 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
e6dcd2dc 739 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 740 root->sectorsize, 0, alloc_hint,
e6dcd2dc 741 (u64)-1, &ins, 1);
c59f8951 742 if (ret) {
771ed689 743 BUG();
c59f8951 744 }
e6dcd2dc
CM		 745 em = alloc_extent_map(GFP_NOFS);
746 em->start = start;
445a6944 747 em->orig_start = em->start;
c8b97818 748
771ed689
CM		 749 ram_size = ins.offset;
750 em->len = ins.offset;
c8b97818 751
e6dcd2dc 752 em->block_start = ins.objectid;
c8b97818 753 em->block_len = ins.offset;
e6dcd2dc 754 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 755 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 756
e6dcd2dc
CM		 757 while(1) {
758 spin_lock(&em_tree->lock);
759 ret = add_extent_mapping(em_tree, em);
760 spin_unlock(&em_tree->lock);
761 if (ret != -EEXIST) {
762 free_extent_map(em);
763 break;
764 }
765 btrfs_drop_extent_cache(inode, start,
c8b97818 766 start + ram_size - 1, 0);
e6dcd2dc
CM		 767 }
768
98d20f67 769 cur_alloc_size = ins.offset;
e6dcd2dc 770 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 771 ram_size, cur_alloc_size, 0);
e6dcd2dc 772 BUG_ON(ret);
c8b97818
CM		 773
774 if (disk_num_bytes < cur_alloc_size) {
775 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
3b951516
CM		 776 cur_alloc_size);
777 break;
778 }
c8b97818
CM		 779 /* we're not doing compressed IO, don't unlock the first
780 * page (which the caller expects to stay locked), don't
781 * clear any dirty bits and don't set any writeback bits
782 */
783 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
784 start, start + ram_size - 1,
771ed689
CM		 785 locked_page, unlock, 1,
786 1, 0, 0, 0);
c8b97818 787 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 788 num_bytes -= cur_alloc_size;
789 alloc_hint = ins.objectid + ins.offset;
790 start += cur_alloc_size;
b888db2b 791 }
b888db2b 792out:
771ed689 793 ret = 0;
b888db2b 794 btrfs_end_transaction(trans, root);
c8b97818 795
be20aa9d 796 return ret;
771ed689 797}
c8b97818 798
771ed689
CM		 799/*
800 * work queue call back to started compression on a file and pages
801 */
802static noinline void async_cow_start(struct btrfs_work *work)
803{
804 struct async_cow *async_cow;
805 int num_added = 0;
806 async_cow = container_of(work, struct async_cow, work);
807
808 compress_file_range(async_cow->inode, async_cow->locked_page,
809 async_cow->start, async_cow->end, async_cow,
810 &num_added);
811 if (num_added == 0)
812 async_cow->inode = NULL;
813}
814
815/*
816 * work queue call back to submit previously compressed pages
817 */
818static noinline void async_cow_submit(struct btrfs_work *work)
819{
820 struct async_cow *async_cow;
821 struct btrfs_root *root;
822 unsigned long nr_pages;
823
824 async_cow = container_of(work, struct async_cow, work);
825
826 root = async_cow->root;
827 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
828 PAGE_CACHE_SHIFT;
829
830 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
831
832 if (atomic_read(&root->fs_info->async_delalloc_pages) <
833 5 * 1042 * 1024 &&
834 waitqueue_active(&root->fs_info->async_submit_wait))
835 wake_up(&root->fs_info->async_submit_wait);
836
837 if (async_cow->inode) {
838 submit_compressed_extents(async_cow->inode, async_cow);
70b99e69 839 }
771ed689 840}
c8b97818 841
771ed689
CM		 842static noinline void async_cow_free(struct btrfs_work *work)
843{
844 struct async_cow *async_cow;
845 async_cow = container_of(work, struct async_cow, work);
846 kfree(async_cow);
847}
848
849static int cow_file_range_async(struct inode *inode, struct page *locked_page,
850 u64 start, u64 end, int *page_started,
851 unsigned long *nr_written)
852{
853 struct async_cow *async_cow;
854 struct btrfs_root *root = BTRFS_I(inode)->root;
855 unsigned long nr_pages;
856 u64 cur_end;
857 int limit = 10 * 1024 * 1042;
858
859 if (!btrfs_test_opt(root, COMPRESS)) {
860 return cow_file_range(inode, locked_page, start, end,
861 page_started, nr_written, 1);
862 }
863
864 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
865 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
866 while(start < end) {
867 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
868 async_cow->inode = inode;
869 async_cow->root = root;
870 async_cow->locked_page = locked_page;
871 async_cow->start = start;
872
873 if (btrfs_test_flag(inode, NOCOMPRESS))
874 cur_end = end;
875 else
876 cur_end = min(end, start + 512 * 1024 - 1);
877
878 async_cow->end = cur_end;
879 INIT_LIST_HEAD(&async_cow->extents);
880
881 async_cow->work.func = async_cow_start;
882 async_cow->work.ordered_func = async_cow_submit;
883 async_cow->work.ordered_free = async_cow_free;
884 async_cow->work.flags = 0;
885
886 while(atomic_read(&root->fs_info->async_submit_draining) &&
887 atomic_read(&root->fs_info->async_delalloc_pages)) {
888 wait_event(root->fs_info->async_submit_wait,
889 (atomic_read(&root->fs_info->async_delalloc_pages)
890 == 0));
891 }
892
893 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
894 PAGE_CACHE_SHIFT;
895 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
896
897 btrfs_queue_worker(&root->fs_info->delalloc_workers,
898 &async_cow->work);
899
900 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
901 wait_event(root->fs_info->async_submit_wait,
902 (atomic_read(&root->fs_info->async_delalloc_pages) <
903 limit));
904 }
905
906 while(atomic_read(&root->fs_info->async_submit_draining) &&
907 atomic_read(&root->fs_info->async_delalloc_pages)) {
908 wait_event(root->fs_info->async_submit_wait,
909 (atomic_read(&root->fs_info->async_delalloc_pages) ==
910 0));
911 }
912
913 *nr_written += nr_pages;
914 start = cur_end + 1;
915 }
916 *page_started = 1;
917 return 0;
be20aa9d
CM		 918}
919
d352ac68
CM		 920/*
921 * when nowcow writeback call back. This checks for snapshots or COW copies
922 * of the extents that exist in the file, and COWs the file as required.
923 *
924 * If no cow copies or snapshots exist, we write directly to the existing
925 * blocks on disk
926 */
c8b97818 927static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
771ed689
CM		 928 u64 start, u64 end, int *page_started, int force,
929 unsigned long *nr_written)
be20aa9d 930{
be20aa9d 931 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 932 struct btrfs_trans_handle *trans;
be20aa9d 933 struct extent_buffer *leaf;
be20aa9d 934 struct btrfs_path *path;
80ff3856 935 struct btrfs_file_extent_item *fi;
be20aa9d 936 struct btrfs_key found_key;
80ff3856
YZ		 937 u64 cow_start;
938 u64 cur_offset;
939 u64 extent_end;
940 u64 disk_bytenr;
941 u64 num_bytes;
942 int extent_type;
943 int ret;
d899e052 944 int type;
80ff3856
YZ		 945 int nocow;
946 int check_prev = 1;
be20aa9d
CM		 947
948 path = btrfs_alloc_path();
949 BUG_ON(!path);
7ea394f1
YZ		 950 trans = btrfs_join_transaction(root, 1);
951 BUG_ON(!trans);
be20aa9d 952
80ff3856
YZ		 953 cow_start = (u64)-1;
954 cur_offset = start;
955 while (1) {
956 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
957 cur_offset, 0);
958 BUG_ON(ret < 0);
959 if (ret > 0 && path->slots[0] > 0 && check_prev) {
960 leaf = path->nodes[0];
961 btrfs_item_key_to_cpu(leaf, &found_key,
962 path->slots[0] - 1);
963 if (found_key.objectid == inode->i_ino &&
964 found_key.type == BTRFS_EXTENT_DATA_KEY)
965 path->slots[0]--;
966 }
967 check_prev = 0;
968next_slot:
969 leaf = path->nodes[0];
970 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
971 ret = btrfs_next_leaf(root, path);
972 if (ret < 0)
973 BUG_ON(1);
974 if (ret > 0)
975 break;
976 leaf = path->nodes[0];
977 }
be20aa9d 978
80ff3856
YZ		 979 nocow = 0;
980 disk_bytenr = 0;
981 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
982
983 if (found_key.objectid > inode->i_ino ||
984 found_key.type > BTRFS_EXTENT_DATA_KEY ||
985 found_key.offset > end)
986 break;
987
988 if (found_key.offset > cur_offset) {
989 extent_end = found_key.offset;
990 goto out_check;
991 }
992
993 fi = btrfs_item_ptr(leaf, path->slots[0],
994 struct btrfs_file_extent_item);
995 extent_type = btrfs_file_extent_type(leaf, fi);
996
d899e052
YZ		 997 if (extent_type == BTRFS_FILE_EXTENT_REG ||
998 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856
YZ		 999 struct btrfs_block_group_cache *block_group;
1000 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1001 extent_end = found_key.offset +
1002 btrfs_file_extent_num_bytes(leaf, fi);
1003 if (extent_end <= start) {
1004 path->slots[0]++;
1005 goto next_slot;
1006 }
1007 if (btrfs_file_extent_compression(leaf, fi) ||
1008 btrfs_file_extent_encryption(leaf, fi) ||
1009 btrfs_file_extent_other_encoding(leaf, fi))
1010 goto out_check;
1011 if (disk_bytenr == 0)
1012 goto out_check;
d899e052
YZ		 1013 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1014 goto out_check;
80ff3856
YZ		 1015 if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
1016 goto out_check;
1017 block_group = btrfs_lookup_block_group(root->fs_info,
1018 disk_bytenr);
1019 if (!block_group || block_group->ro)
1020 goto out_check;
1021 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1022 nocow = 1;
1023 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1024 extent_end = found_key.offset +
1025 btrfs_file_extent_inline_len(leaf, fi);
1026 extent_end = ALIGN(extent_end, root->sectorsize);
1027 } else {
1028 BUG_ON(1);
1029 }
1030out_check:
1031 if (extent_end <= start) {
1032 path->slots[0]++;
1033 goto next_slot;
1034 }
1035 if (!nocow) {
1036 if (cow_start == (u64)-1)
1037 cow_start = cur_offset;
1038 cur_offset = extent_end;
1039 if (cur_offset > end)
1040 break;
1041 path->slots[0]++;
1042 goto next_slot;
7ea394f1
YZ		 1043 }
1044
1045 btrfs_release_path(root, path);
80ff3856
YZ		 1046 if (cow_start != (u64)-1) {
1047 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1048 found_key.offset - 1, page_started,
1049 nr_written, 1);
80ff3856
YZ		 1050 BUG_ON(ret);
1051 cow_start = (u64)-1;
7ea394f1 1052 }
80ff3856
YZ		 1053
1054 disk_bytenr += cur_offset - found_key.offset;
1055 num_bytes = min(end + 1, extent_end) - cur_offset;
d899e052
YZ		 1056 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1057 struct extent_map *em;
1058 struct extent_map_tree *em_tree;
1059 em_tree = &BTRFS_I(inode)->extent_tree;
1060 em = alloc_extent_map(GFP_NOFS);
1061 em->start = cur_offset;
445a6944 1062 em->orig_start = em->start;
d899e052
YZ		 1063 em->len = num_bytes;
1064 em->block_len = num_bytes;
1065 em->block_start = disk_bytenr;
1066 em->bdev = root->fs_info->fs_devices->latest_bdev;
1067 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1068 while (1) {
1069 spin_lock(&em_tree->lock);
1070 ret = add_extent_mapping(em_tree, em);
1071 spin_unlock(&em_tree->lock);
1072 if (ret != -EEXIST) {
1073 free_extent_map(em);
1074 break;
1075 }
1076 btrfs_drop_extent_cache(inode, em->start,
1077 em->start + em->len - 1, 0);
1078 }
1079 type = BTRFS_ORDERED_PREALLOC;
1080 } else {
1081 type = BTRFS_ORDERED_NOCOW;
1082 }
80ff3856
YZ		 1083
1084 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 1085 num_bytes, num_bytes, type);
1086 BUG_ON(ret);
771ed689 1087
d899e052
YZ		 1088 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1089 cur_offset, cur_offset + num_bytes - 1,
771ed689 1090 locked_page, 1, 1, 1, 0, 0, 0);
80ff3856
YZ		 1091 cur_offset = extent_end;
1092 if (cur_offset > end)
1093 break;
be20aa9d 1094 }
80ff3856
YZ		 1095 btrfs_release_path(root, path);
1096
1097 if (cur_offset <= end && cow_start == (u64)-1)
1098 cow_start = cur_offset;
1099 if (cow_start != (u64)-1) {
1100 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1101 page_started, nr_written, 1);
80ff3856
YZ		 1102 BUG_ON(ret);
1103 }
1104
1105 ret = btrfs_end_transaction(trans, root);
1106 BUG_ON(ret);
7ea394f1 1107 btrfs_free_path(path);
80ff3856 1108 return 0;
be20aa9d
CM		 1109}
1110
d352ac68
CM		 1111/*
1112 * extent_io.c call back to do delayed allocation processing
1113 */
c8b97818 1114static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1115 u64 start, u64 end, int *page_started,
1116 unsigned long *nr_written)
be20aa9d
CM		 1117{
1118 struct btrfs_root *root = BTRFS_I(inode)->root;
1119 int ret;
a2135011 1120
b98b6767
Y		 1121 if (btrfs_test_opt(root, NODATACOW) ||
1122 btrfs_test_flag(inode, NODATACOW))
c8b97818 1123 ret = run_delalloc_nocow(inode, locked_page, start, end,
771ed689 1124 page_started, 0, nr_written);
d899e052
YZ		 1125 else if (btrfs_test_flag(inode, PREALLOC))
1126 ret = run_delalloc_nocow(inode, locked_page, start, end,
771ed689 1127 page_started, 1, nr_written);
be20aa9d 1128 else
771ed689
CM		 1129 ret = cow_file_range_async(inode, locked_page, start, end,
1130 page_started, nr_written);
1832a6d5 1131
b888db2b
CM		 1132 return ret;
1133}
1134
d352ac68
CM		 1135/*
1136 * extent_io.c set_bit_hook, used to track delayed allocation
1137 * bytes in this file, and to maintain the list of inodes that
1138 * have pending delalloc work to be done.
1139 */
291d673e 1140int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1141 unsigned long old, unsigned long bits)
291d673e 1142{
bcbfce8a 1143 unsigned long flags;
b0c68f8b 1144 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1145 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a 1146 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
9069218d 1147 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
291d673e 1148 root->fs_info->delalloc_bytes += end - start + 1;
ea8c2819
CM		 1149 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1150 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1151 &root->fs_info->delalloc_inodes);
1152 }
bcbfce8a 1153 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
291d673e
CM		 1154 }
1155 return 0;
1156}
1157
d352ac68
CM		 1158/*
1159 * extent_io.c clear_bit_hook, see set_bit_hook for why
1160 */
291d673e 1161int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1162 unsigned long old, unsigned long bits)
291d673e 1163{
b0c68f8b 1164 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1165 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a
CM		 1166 unsigned long flags;
1167
1168 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
b0c68f8b
CM		 1169 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1170 printk("warning: delalloc account %Lu %Lu\n",
1171 end - start + 1, root->fs_info->delalloc_bytes);
1172 root->fs_info->delalloc_bytes = 0;
9069218d 1173 BTRFS_I(inode)->delalloc_bytes = 0;
b0c68f8b
CM		 1174 } else {
1175 root->fs_info->delalloc_bytes -= end - start + 1;
9069218d 1176 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
b0c68f8b 1177 }
ea8c2819
CM		 1178 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1179 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1180 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1181 }
bcbfce8a 1182 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
291d673e
CM		 1183 }
1184 return 0;
1185}
1186
d352ac68
CM		 1187/*
1188 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1189 * we don't create bios that span stripes or chunks
1190 */
239b14b3 1191int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1192 size_t size, struct bio *bio,
1193 unsigned long bio_flags)
239b14b3
CM		 1194{
1195 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1196 struct btrfs_mapping_tree *map_tree;
a62b9401 1197 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1198 u64 length = 0;
1199 u64 map_length;
239b14b3
CM		 1200 int ret;
1201
771ed689
CM		 1202 if (bio_flags & EXTENT_BIO_COMPRESSED)
1203 return 0;
1204
f2d8d74d 1205 length = bio->bi_size;
239b14b3
CM		 1206 map_tree = &root->fs_info->mapping_tree;
1207 map_length = length;
cea9e445 1208 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1209 &map_length, NULL, 0);
cea9e445 1210
239b14b3 1211 if (map_length < length + size) {
239b14b3
CM		 1212 return 1;
1213 }
1214 return 0;
1215}
1216
d352ac68
CM		 1217/*
1218 * in order to insert checksums into the metadata in large chunks,
1219 * we wait until bio submission time. All the pages in the bio are
1220 * checksummed and sums are attached onto the ordered extent record.
1221 *
1222 * At IO completion time the cums attached on the ordered extent record
1223 * are inserted into the btree
1224 */
4a69a410 1225int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio,
c8b97818 1226 int mirror_num, unsigned long bio_flags)
065631f6 1227{
065631f6 1228 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1229 int ret = 0;
e015640f 1230
3edf7d33 1231 ret = btrfs_csum_one_bio(root, inode, bio);
44b8bd7e 1232 BUG_ON(ret);
4a69a410
CM		 1233 return 0;
1234}
e015640f 1235
4a69a410
CM		 1236/*
1237 * in order to insert checksums into the metadata in large chunks,
1238 * we wait until bio submission time. All the pages in the bio are
1239 * checksummed and sums are attached onto the ordered extent record.
1240 *
1241 * At IO completion time the cums attached on the ordered extent record
1242 * are inserted into the btree
1243 */
1244int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1245 int mirror_num, unsigned long bio_flags)
1246{
1247 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1248 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1249}
1250
d352ac68
CM		 1251/*
1252 * extent_io.c submission hook. This does the right thing for csum calculation on write,
1253 * or reading the csums from the tree before a read
1254 */
44b8bd7e 1255int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 1256 int mirror_num, unsigned long bio_flags)
44b8bd7e
CM		 1257{
1258 struct btrfs_root *root = BTRFS_I(inode)->root;
1259 int ret = 0;
19b9bdb0 1260 int skip_sum;
44b8bd7e 1261
e6dcd2dc
CM		 1262 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1263 BUG_ON(ret);
065631f6 1264
19b9bdb0
CM		 1265 skip_sum = btrfs_test_opt(root, NODATASUM) ||
1266 btrfs_test_flag(inode, NODATASUM);
7ea394f1 1267
4d1b5fb4 1268 if (!(rw & (1 << BIO_RW))) {
c8b97818 1269
19b9bdb0 1270 if (bio_flags & EXTENT_BIO_COMPRESSED)
c8b97818
CM		 1271 return btrfs_submit_compressed_read(inode, bio,
1272 mirror_num, bio_flags);
771ed689
CM		 1273 else if (!skip_sum)
1274 btrfs_lookup_bio_sums(root, inode, bio);
4d1b5fb4 1275 goto mapit;
19b9bdb0
CM		 1276 } else if (!skip_sum) {
1277 /* we're doing a write, do the async checksumming */
1278 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1279 inode, rw, bio, mirror_num,
4a69a410
CM		 1280 bio_flags, __btrfs_submit_bio_start,
1281 __btrfs_submit_bio_done);
19b9bdb0
CM		 1282 }
1283
0b86a832 1284mapit:
8b712842 1285 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1286}
6885f308 1287
d352ac68
CM		 1288/*
1289 * given a list of ordered sums record them in the inode. This happens
1290 * at IO completion time based on sums calculated at bio submission time.
1291 */
ba1da2f4 1292static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1293 struct inode *inode, u64 file_offset,
1294 struct list_head *list)
1295{
1296 struct list_head *cur;
1297 struct btrfs_ordered_sum *sum;
1298
1299 btrfs_set_trans_block_group(trans, inode);
ba1da2f4 1300 list_for_each(cur, list) {
e6dcd2dc 1301 sum = list_entry(cur, struct btrfs_ordered_sum, list);
e6dcd2dc
CM		 1302 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
1303 inode, sum);
e6dcd2dc
CM		 1304 }
1305 return 0;
1306}
1307
ea8c2819
CM		 1308int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1309{
771ed689
CM		 1310 if ((end & (PAGE_CACHE_SIZE - 1)) == 0) {
1311 WARN_ON(1);
1312 }
ea8c2819
CM		 1313 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1314 GFP_NOFS);
1315}
1316
d352ac68 1317/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1318struct btrfs_writepage_fixup {
1319 struct page *page;
1320 struct btrfs_work work;
1321};
1322
247e743c
CM		 1323void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1324{
1325 struct btrfs_writepage_fixup *fixup;
1326 struct btrfs_ordered_extent *ordered;
1327 struct page *page;
1328 struct inode *inode;
1329 u64 page_start;
1330 u64 page_end;
1331
1332 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1333 page = fixup->page;
4a096752 1334again:
247e743c
CM		 1335 lock_page(page);
1336 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1337 ClearPageChecked(page);
1338 goto out_page;
1339 }
1340
1341 inode = page->mapping->host;
1342 page_start = page_offset(page);
1343 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1344
1345 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
4a096752
CM		 1346
1347 /* already ordered? We're done */
1348 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1349 EXTENT_ORDERED, 0)) {
247e743c 1350 goto out;
4a096752
CM		 1351 }
1352
1353 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1354 if (ordered) {
1355 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1356 page_end, GFP_NOFS);
1357 unlock_page(page);
1358 btrfs_start_ordered_extent(inode, ordered, 1);
1359 goto again;
1360 }
247e743c 1361
ea8c2819 1362 btrfs_set_extent_delalloc(inode, page_start, page_end);
247e743c
CM		 1363 ClearPageChecked(page);
1364out:
1365 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1366out_page:
1367 unlock_page(page);
1368 page_cache_release(page);
1369}
1370
1371/*
1372 * There are a few paths in the higher layers of the kernel that directly
1373 * set the page dirty bit without asking the filesystem if it is a
1374 * good idea. This causes problems because we want to make sure COW
1375 * properly happens and the data=ordered rules are followed.
1376 *
c8b97818 1377 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1378 * hasn't been properly setup for IO. We kick off an async process
1379 * to fix it up. The async helper will wait for ordered extents, set
1380 * the delalloc bit and make it safe to write the page.
1381 */
1382int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1383{
1384 struct inode *inode = page->mapping->host;
1385 struct btrfs_writepage_fixup *fixup;
1386 struct btrfs_root *root = BTRFS_I(inode)->root;
1387 int ret;
1388
1389 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1390 EXTENT_ORDERED, 0);
1391 if (ret)
1392 return 0;
1393
1394 if (PageChecked(page))
1395 return -EAGAIN;
1396
1397 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1398 if (!fixup)
1399 return -EAGAIN;
f421950f 1400
247e743c
CM		 1401 SetPageChecked(page);
1402 page_cache_get(page);
1403 fixup->work.func = btrfs_writepage_fixup_worker;
1404 fixup->page = page;
1405 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1406 return -EAGAIN;
1407}
1408
d899e052
YZ		 1409static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1410 struct inode *inode, u64 file_pos,
1411 u64 disk_bytenr, u64 disk_num_bytes,
1412 u64 num_bytes, u64 ram_bytes,
1413 u8 compression, u8 encryption,
1414 u16 other_encoding, int extent_type)
1415{
1416 struct btrfs_root *root = BTRFS_I(inode)->root;
1417 struct btrfs_file_extent_item *fi;
1418 struct btrfs_path *path;
1419 struct extent_buffer *leaf;
1420 struct btrfs_key ins;
1421 u64 hint;
1422 int ret;
1423
1424 path = btrfs_alloc_path();
1425 BUG_ON(!path);
1426
1427 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1428 file_pos + num_bytes, file_pos, &hint);
1429 BUG_ON(ret);
1430
1431 ins.objectid = inode->i_ino;
1432 ins.offset = file_pos;
1433 ins.type = BTRFS_EXTENT_DATA_KEY;
1434 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1435 BUG_ON(ret);
1436 leaf = path->nodes[0];
1437 fi = btrfs_item_ptr(leaf, path->slots[0],
1438 struct btrfs_file_extent_item);
1439 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1440 btrfs_set_file_extent_type(leaf, fi, extent_type);
1441 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1442 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1443 btrfs_set_file_extent_offset(leaf, fi, 0);
1444 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1445 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1446 btrfs_set_file_extent_compression(leaf, fi, compression);
1447 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1448 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1449 btrfs_mark_buffer_dirty(leaf);
1450
1451 inode_add_bytes(inode, num_bytes);
1452 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1453
1454 ins.objectid = disk_bytenr;
1455 ins.offset = disk_num_bytes;
1456 ins.type = BTRFS_EXTENT_ITEM_KEY;
1457 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1458 root->root_key.objectid,
1459 trans->transid, inode->i_ino, &ins);
1460 BUG_ON(ret);
1461
1462 btrfs_free_path(path);
1463 return 0;
1464}
1465
d352ac68
CM		 1466/* as ordered data IO finishes, this gets called so we can finish
1467 * an ordered extent if the range of bytes in the file it covers are
1468 * fully written.
1469 */
211f90e6 1470static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1471{
e6dcd2dc
CM		 1472 struct btrfs_root *root = BTRFS_I(inode)->root;
1473 struct btrfs_trans_handle *trans;
1474 struct btrfs_ordered_extent *ordered_extent;
1475 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
d899e052 1476 int compressed = 0;
e6dcd2dc
CM		 1477 int ret;
1478
1479 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
ba1da2f4 1480 if (!ret)
e6dcd2dc 1481 return 0;
e6dcd2dc 1482
f9295749 1483 trans = btrfs_join_transaction(root, 1);
e6dcd2dc
CM		 1484
1485 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1486 BUG_ON(!ordered_extent);
7ea394f1
YZ		 1487 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1488 goto nocow;
e6dcd2dc
CM		 1489
1490 lock_extent(io_tree, ordered_extent->file_offset,
1491 ordered_extent->file_offset + ordered_extent->len - 1,
1492 GFP_NOFS);
1493
c8b97818 1494 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
d899e052
YZ		 1495 compressed = 1;
1496 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1497 BUG_ON(compressed);
1498 ret = btrfs_mark_extent_written(trans, root, inode,
1499 ordered_extent->file_offset,
1500 ordered_extent->file_offset +
1501 ordered_extent->len);
1502 BUG_ON(ret);
1503 } else {
1504 ret = insert_reserved_file_extent(trans, inode,
1505 ordered_extent->file_offset,
1506 ordered_extent->start,
1507 ordered_extent->disk_len,
1508 ordered_extent->len,
1509 ordered_extent->len,
1510 compressed, 0, 0,
1511 BTRFS_FILE_EXTENT_REG);
1512 BUG_ON(ret);
1513 }
e6dcd2dc
CM		 1514 unlock_extent(io_tree, ordered_extent->file_offset,
1515 ordered_extent->file_offset + ordered_extent->len - 1,
1516 GFP_NOFS);
7ea394f1 1517nocow:
e6dcd2dc
CM		 1518 add_pending_csums(trans, inode, ordered_extent->file_offset,
1519 &ordered_extent->list);
1520
34353029 1521 mutex_lock(&BTRFS_I(inode)->extent_mutex);
dbe674a9 1522 btrfs_ordered_update_i_size(inode, ordered_extent);
e02119d5 1523 btrfs_update_inode(trans, root, inode);
e6dcd2dc 1524 btrfs_remove_ordered_extent(inode, ordered_extent);
34353029 1525 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
7f3c74fb 1526
e6dcd2dc
CM		 1527 /* once for us */
1528 btrfs_put_ordered_extent(ordered_extent);
1529 /* once for the tree */
1530 btrfs_put_ordered_extent(ordered_extent);
1531
e6dcd2dc
CM		 1532 btrfs_end_transaction(trans, root);
1533 return 0;
1534}
1535
211f90e6
CM		 1536int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1537 struct extent_state *state, int uptodate)
1538{
1539 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1540}
1541
d352ac68
CM		 1542/*
1543 * When IO fails, either with EIO or csum verification fails, we
1544 * try other mirrors that might have a good copy of the data. This
1545 * io_failure_record is used to record state as we go through all the
1546 * mirrors. If another mirror has good data, the page is set up to date
1547 * and things continue. If a good mirror can't be found, the original
1548 * bio end_io callback is called to indicate things have failed.
1549 */
7e38326f
CM		 1550struct io_failure_record {
1551 struct page *page;
1552 u64 start;
1553 u64 len;
1554 u64 logical;
1555 int last_mirror;
1556};
1557
1259ab75
CM		 1558int btrfs_io_failed_hook(struct bio *failed_bio,
1559 struct page *page, u64 start, u64 end,
1560 struct extent_state *state)
7e38326f
CM		 1561{
1562 struct io_failure_record *failrec = NULL;
1563 u64 private;
1564 struct extent_map *em;
1565 struct inode *inode = page->mapping->host;
1566 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1567 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM		 1568 struct bio *bio;
1569 int num_copies;
1570 int ret;
1259ab75 1571 int rw;
7e38326f 1572 u64 logical;
c8b97818 1573 unsigned long bio_flags = 0;
7e38326f
CM		 1574
1575 ret = get_state_private(failure_tree, start, &private);
1576 if (ret) {
7e38326f
CM		 1577 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1578 if (!failrec)
1579 return -ENOMEM;
1580 failrec->start = start;
1581 failrec->len = end - start + 1;
1582 failrec->last_mirror = 0;
1583
3b951516
CM		 1584 spin_lock(&em_tree->lock);
1585 em = lookup_extent_mapping(em_tree, start, failrec->len);
1586 if (em->start > start || em->start + em->len < start) {
1587 free_extent_map(em);
1588 em = NULL;
1589 }
1590 spin_unlock(&em_tree->lock);
7e38326f
CM		 1591
1592 if (!em || IS_ERR(em)) {
1593 kfree(failrec);
1594 return -EIO;
1595 }
1596 logical = start - em->start;
1597 logical = em->block_start + logical;
c8b97818
CM		 1598 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1599 bio_flags = EXTENT_BIO_COMPRESSED;
7e38326f
CM		 1600 failrec->logical = logical;
1601 free_extent_map(em);
1602 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1603 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM		 1604 set_state_private(failure_tree, start,
1605 (u64)(unsigned long)failrec);
7e38326f 1606 } else {
587f7704 1607 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM		 1608 }
1609 num_copies = btrfs_num_copies(
1610 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1611 failrec->logical, failrec->len);
1612 failrec->last_mirror++;
1613 if (!state) {
1614 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1615 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1616 failrec->start,
1617 EXTENT_LOCKED);
1618 if (state && state->start != failrec->start)
1619 state = NULL;
1620 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1621 }
1622 if (!state || failrec->last_mirror > num_copies) {
1623 set_state_private(failure_tree, failrec->start, 0);
1624 clear_extent_bits(failure_tree, failrec->start,
1625 failrec->start + failrec->len - 1,
1626 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1627 kfree(failrec);
1628 return -EIO;
1629 }
1630 bio = bio_alloc(GFP_NOFS, 1);
1631 bio->bi_private = state;
1632 bio->bi_end_io = failed_bio->bi_end_io;
1633 bio->bi_sector = failrec->logical >> 9;
1634 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1635 bio->bi_size = 0;
7e38326f 1636 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1259ab75
CM		 1637 if (failed_bio->bi_rw & (1 << BIO_RW))
1638 rw = WRITE;
1639 else
1640 rw = READ;
1641
1642 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818
CM		 1643 failrec->last_mirror,
1644 bio_flags);
1259ab75
CM		 1645 return 0;
1646}
1647
d352ac68
CM		 1648/*
1649 * each time an IO finishes, we do a fast check in the IO failure tree
1650 * to see if we need to process or clean up an io_failure_record
1651 */
1259ab75
CM		 1652int btrfs_clean_io_failures(struct inode *inode, u64 start)
1653{
1654 u64 private;
1655 u64 private_failure;
1656 struct io_failure_record *failure;
1657 int ret;
1658
1659 private = 0;
1660 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1661 (u64)-1, 1, EXTENT_DIRTY)) {
1662 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1663 start, &private_failure);
1664 if (ret == 0) {
1665 failure = (struct io_failure_record *)(unsigned long)
1666 private_failure;
1667 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1668 failure->start, 0);
1669 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1670 failure->start,
1671 failure->start + failure->len - 1,
1672 EXTENT_DIRTY | EXTENT_LOCKED,
1673 GFP_NOFS);
1674 kfree(failure);
1675 }
1676 }
7e38326f
CM		 1677 return 0;
1678}
1679
d352ac68
CM		 1680/*
1681 * when reads are done, we need to check csums to verify the data is correct
1682 * if there's a match, we allow the bio to finish. If not, we go through
1683 * the io_failure_record routines to find good copies
1684 */
70dec807
CM		 1685int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1686 struct extent_state *state)
07157aac 1687{
35ebb934 1688 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1689 struct inode *inode = page->mapping->host;
d1310b2e 1690 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1691 char *kaddr;
aadfeb6e 1692 u64 private = ~(u32)0;
07157aac 1693 int ret;
ff79f819
CM		 1694 struct btrfs_root *root = BTRFS_I(inode)->root;
1695 u32 csum = ~(u32)0;
bbf0d006 1696 unsigned long flags;
d1310b2e 1697
b98b6767
Y		 1698 if (btrfs_test_opt(root, NODATASUM) ||
1699 btrfs_test_flag(inode, NODATASUM))
b6cda9bc 1700 return 0;
c2e639f0 1701 if (state && state->start == start) {
70dec807
CM		 1702 private = state->private;
1703 ret = 0;
1704 } else {
1705 ret = get_state_private(io_tree, start, &private);
1706 }
bbf0d006 1707 local_irq_save(flags);
07157aac
CM		 1708 kaddr = kmap_atomic(page, KM_IRQ0);
1709 if (ret) {
1710 goto zeroit;
1711 }
ff79f819
CM		 1712 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1713 btrfs_csum_final(csum, (char *)&csum);
1714 if (csum != private) {
07157aac
CM		 1715 goto zeroit;
1716 }
1717 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1718 local_irq_restore(flags);
7e38326f
CM		 1719
1720 /* if the io failure tree for this inode is non-empty,
1721 * check to see if we've recovered from a failed IO
1722 */
1259ab75 1723 btrfs_clean_io_failures(inode, start);
07157aac
CM		 1724 return 0;
1725
1726zeroit:
aadfeb6e
CM		 1727 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1728 page->mapping->host->i_ino, (unsigned long long)start, csum,
1729 private);
db94535d
CM		 1730 memset(kaddr + offset, 1, end - start + 1);
1731 flush_dcache_page(page);
07157aac 1732 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1733 local_irq_restore(flags);
3b951516
CM		 1734 if (private == 0)
1735 return 0;
7e38326f 1736 return -EIO;
07157aac 1737}
b888db2b 1738
7b128766
JB		 1739/*
1740 * This creates an orphan entry for the given inode in case something goes
1741 * wrong in the middle of an unlink/truncate.
1742 */
1743int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1744{
1745 struct btrfs_root *root = BTRFS_I(inode)->root;
1746 int ret = 0;
1747
bcc63abb 1748 spin_lock(&root->list_lock);
7b128766
JB		 1749
1750 /* already on the orphan list, we're good */
1751 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1752 spin_unlock(&root->list_lock);
7b128766
JB		 1753 return 0;
1754 }
1755
1756 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1757
bcc63abb 1758 spin_unlock(&root->list_lock);
7b128766
JB		 1759
1760 /*
1761 * insert an orphan item to track this unlinked/truncated file
1762 */
1763 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1764
1765 return ret;
1766}
1767
1768/*
1769 * We have done the truncate/delete so we can go ahead and remove the orphan
1770 * item for this particular inode.
1771 */
1772int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1773{
1774 struct btrfs_root *root = BTRFS_I(inode)->root;
1775 int ret = 0;
1776
bcc63abb 1777 spin_lock(&root->list_lock);
7b128766
JB		 1778
1779 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1780 spin_unlock(&root->list_lock);
7b128766
JB		 1781 return 0;
1782 }
1783
1784 list_del_init(&BTRFS_I(inode)->i_orphan);
1785 if (!trans) {
bcc63abb 1786 spin_unlock(&root->list_lock);
7b128766
JB		 1787 return 0;
1788 }
1789
bcc63abb 1790 spin_unlock(&root->list_lock);
7b128766
JB		 1791
1792 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1793
1794 return ret;
1795}
1796
1797/*
1798 * this cleans up any orphans that may be left on the list from the last use
1799 * of this root.
1800 */
1801void btrfs_orphan_cleanup(struct btrfs_root *root)
1802{
1803 struct btrfs_path *path;
1804 struct extent_buffer *leaf;
1805 struct btrfs_item *item;
1806 struct btrfs_key key, found_key;
1807 struct btrfs_trans_handle *trans;
1808 struct inode *inode;
1809 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1810
7b128766
JB		 1811 path = btrfs_alloc_path();
1812 if (!path)
1813 return;
1814 path->reada = -1;
1815
1816 key.objectid = BTRFS_ORPHAN_OBJECTID;
1817 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1818 key.offset = (u64)-1;
1819
7b128766
JB		 1820
1821 while (1) {
1822 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1823 if (ret < 0) {
1824 printk(KERN_ERR "Error searching slot for orphan: %d"
1825 "\n", ret);
1826 break;
1827 }
1828
1829 /*
1830 * if ret == 0 means we found what we were searching for, which
1831 * is weird, but possible, so only screw with path if we didnt
1832 * find the key and see if we have stuff that matches
1833 */
1834 if (ret > 0) {
1835 if (path->slots[0] == 0)
1836 break;
1837 path->slots[0]--;
1838 }
1839
1840 /* pull out the item */
1841 leaf = path->nodes[0];
1842 item = btrfs_item_nr(leaf, path->slots[0]);
1843 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1844
1845 /* make sure the item matches what we want */
1846 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1847 break;
1848 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1849 break;
1850
1851 /* release the path since we're done with it */
1852 btrfs_release_path(root, path);
1853
1854 /*
1855 * this is where we are basically btrfs_lookup, without the
1856 * crossing root thing. we store the inode number in the
1857 * offset of the orphan item.
1858 */
5b21f2ed 1859 inode = btrfs_iget_locked(root->fs_info->sb,
7b128766
JB		 1860 found_key.offset, root);
1861 if (!inode)
1862 break;
1863
1864 if (inode->i_state & I_NEW) {
1865 BTRFS_I(inode)->root = root;
1866
1867 /* have to set the location manually */
1868 BTRFS_I(inode)->location.objectid = inode->i_ino;
1869 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1870 BTRFS_I(inode)->location.offset = 0;
1871
1872 btrfs_read_locked_inode(inode);
1873 unlock_new_inode(inode);
1874 }
1875
1876 /*
1877 * add this inode to the orphan list so btrfs_orphan_del does
1878 * the proper thing when we hit it
1879 */
bcc63abb 1880 spin_lock(&root->list_lock);
7b128766 1881 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
bcc63abb 1882 spin_unlock(&root->list_lock);
7b128766
JB		 1883
1884 /*
1885 * if this is a bad inode, means we actually succeeded in
1886 * removing the inode, but not the orphan record, which means
1887 * we need to manually delete the orphan since iput will just
1888 * do a destroy_inode
1889 */
1890 if (is_bad_inode(inode)) {
5b21f2ed 1891 trans = btrfs_start_transaction(root, 1);
7b128766 1892 btrfs_orphan_del(trans, inode);
5b21f2ed 1893 btrfs_end_transaction(trans, root);
7b128766
JB		 1894 iput(inode);
1895 continue;
1896 }
1897
1898 /* if we have links, this was a truncate, lets do that */
1899 if (inode->i_nlink) {
1900 nr_truncate++;
1901 btrfs_truncate(inode);
1902 } else {
1903 nr_unlink++;
1904 }
1905
1906 /* this will do delete_inode and everything for us */
1907 iput(inode);
1908 }
1909
1910 if (nr_unlink)
1911 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1912 if (nr_truncate)
1913 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1914
1915 btrfs_free_path(path);
7b128766
JB		 1916}
1917
d352ac68
CM		 1918/*
1919 * read an inode from the btree into the in-memory inode
1920 */
39279cc3
CM		 1921void btrfs_read_locked_inode(struct inode *inode)
1922{
1923 struct btrfs_path *path;
5f39d397 1924 struct extent_buffer *leaf;
39279cc3 1925 struct btrfs_inode_item *inode_item;
0b86a832 1926 struct btrfs_timespec *tspec;
39279cc3
CM		 1927 struct btrfs_root *root = BTRFS_I(inode)->root;
1928 struct btrfs_key location;
1929 u64 alloc_group_block;
618e21d5 1930 u32 rdev;
39279cc3
CM		 1931 int ret;
1932
1933 path = btrfs_alloc_path();
1934 BUG_ON(!path);
39279cc3 1935 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 1936
39279cc3 1937 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 1938 if (ret)
39279cc3 1939 goto make_bad;
39279cc3 1940
5f39d397
CM		 1941 leaf = path->nodes[0];
1942 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1943 struct btrfs_inode_item);
1944
1945 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1946 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1947 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1948 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 1949 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 1950
1951 tspec = btrfs_inode_atime(inode_item);
1952 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1953 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1954
1955 tspec = btrfs_inode_mtime(inode_item);
1956 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1957 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1958
1959 tspec = btrfs_inode_ctime(inode_item);
1960 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1961 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1962
a76a3cd4 1963 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5
CM		 1964 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1965 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 1966 inode->i_rdev = 0;
5f39d397
CM		 1967 rdev = btrfs_inode_rdev(leaf, inode_item);
1968
aec7477b
JB		 1969 BTRFS_I(inode)->index_cnt = (u64)-1;
1970
5f39d397 1971 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
39279cc3
CM		 1972 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1973 alloc_group_block);
b98b6767 1974 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
e52ec0eb
CM		 1975 if (!BTRFS_I(inode)->block_group) {
1976 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
0b86a832
CM		 1977 NULL, 0,
1978 BTRFS_BLOCK_GROUP_METADATA, 0);
e52ec0eb 1979 }
39279cc3
CM		 1980 btrfs_free_path(path);
1981 inode_item = NULL;
1982
39279cc3 1983 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 1984 case S_IFREG:
1985 inode->i_mapping->a_ops = &btrfs_aops;
04160088 1986 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 1987 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 1988 inode->i_fop = &btrfs_file_operations;
1989 inode->i_op = &btrfs_file_inode_operations;
1990 break;
1991 case S_IFDIR:
1992 inode->i_fop = &btrfs_dir_file_operations;
1993 if (root == root->fs_info->tree_root)
1994 inode->i_op = &btrfs_dir_ro_inode_operations;
1995 else
1996 inode->i_op = &btrfs_dir_inode_operations;
1997 break;
1998 case S_IFLNK:
1999 inode->i_op = &btrfs_symlink_inode_operations;
2000 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2001 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2002 break;
618e21d5
JB		 2003 default:
2004 init_special_inode(inode, inode->i_mode, rdev);
2005 break;
39279cc3
CM		 2006 }
2007 return;
2008
2009make_bad:
39279cc3 2010 btrfs_free_path(path);
39279cc3
CM		 2011 make_bad_inode(inode);
2012}
2013
d352ac68
CM		 2014/*
2015 * given a leaf and an inode, copy the inode fields into the leaf
2016 */
e02119d5
CM		 2017static void fill_inode_item(struct btrfs_trans_handle *trans,
2018 struct extent_buffer *leaf,
5f39d397 2019 struct btrfs_inode_item *item,
39279cc3
CM		 2020 struct inode *inode)
2021{
5f39d397
CM		 2022 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2023 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2024 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2025 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2026 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2027
2028 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2029 inode->i_atime.tv_sec);
2030 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2031 inode->i_atime.tv_nsec);
2032
2033 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2034 inode->i_mtime.tv_sec);
2035 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2036 inode->i_mtime.tv_nsec);
2037
2038 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2039 inode->i_ctime.tv_sec);
2040 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2041 inode->i_ctime.tv_nsec);
2042
a76a3cd4 2043 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5
CM		 2044 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2045 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2046 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2047 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
5f39d397 2048 btrfs_set_inode_block_group(leaf, item,
39279cc3
CM		 2049 BTRFS_I(inode)->block_group->key.objectid);
2050}
2051
d352ac68
CM		 2052/*
2053 * copy everything in the in-memory inode into the btree.
2054 */
ba1da2f4 2055int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
39279cc3
CM		 2056 struct btrfs_root *root,
2057 struct inode *inode)
2058{
2059 struct btrfs_inode_item *inode_item;
2060 struct btrfs_path *path;
5f39d397 2061 struct extent_buffer *leaf;
39279cc3
CM		 2062 int ret;
2063
2064 path = btrfs_alloc_path();
2065 BUG_ON(!path);
39279cc3
CM		 2066 ret = btrfs_lookup_inode(trans, root, path,
2067 &BTRFS_I(inode)->location, 1);
2068 if (ret) {
2069 if (ret > 0)
2070 ret = -ENOENT;
2071 goto failed;
2072 }
2073
5f39d397
CM		 2074 leaf = path->nodes[0];
2075 inode_item = btrfs_item_ptr(leaf, path->slots[0],
39279cc3
CM		 2076 struct btrfs_inode_item);
2077
e02119d5 2078 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2079 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2080 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2081 ret = 0;
2082failed:
39279cc3
CM		 2083 btrfs_free_path(path);
2084 return ret;
2085}
2086
2087
d352ac68
CM		 2088/*
2089 * unlink helper that gets used here in inode.c and in the tree logging
2090 * recovery code. It remove a link in a directory with a given name, and
2091 * also drops the back refs in the inode to the directory
2092 */
e02119d5
CM		 2093int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2094 struct btrfs_root *root,
2095 struct inode *dir, struct inode *inode,
2096 const char *name, int name_len)
39279cc3
CM		 2097{
2098 struct btrfs_path *path;
39279cc3 2099 int ret = 0;
5f39d397 2100 struct extent_buffer *leaf;
39279cc3 2101 struct btrfs_dir_item *di;
5f39d397 2102 struct btrfs_key key;
aec7477b 2103 u64 index;
39279cc3
CM		 2104
2105 path = btrfs_alloc_path();
54aa1f4d
CM		 2106 if (!path) {
2107 ret = -ENOMEM;
2108 goto err;
2109 }
2110
39279cc3
CM		 2111 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2112 name, name_len, -1);
2113 if (IS_ERR(di)) {
2114 ret = PTR_ERR(di);
2115 goto err;
2116 }
2117 if (!di) {
2118 ret = -ENOENT;
2119 goto err;
2120 }
5f39d397
CM		 2121 leaf = path->nodes[0];
2122 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2123 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2124 if (ret)
2125 goto err;
39279cc3
CM		 2126 btrfs_release_path(root, path);
2127
aec7477b 2128 ret = btrfs_del_inode_ref(trans, root, name, name_len,
e02119d5
CM		 2129 inode->i_ino,
2130 dir->i_ino, &index);
aec7477b
JB		 2131 if (ret) {
2132 printk("failed to delete reference to %.*s, "
2133 "inode %lu parent %lu\n", name_len, name,
e02119d5 2134 inode->i_ino, dir->i_ino);
aec7477b
JB		 2135 goto err;
2136 }
2137
39279cc3 2138 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
aec7477b 2139 index, name, name_len, -1);
39279cc3
CM		 2140 if (IS_ERR(di)) {
2141 ret = PTR_ERR(di);
2142 goto err;
2143 }
2144 if (!di) {
2145 ret = -ENOENT;
2146 goto err;
2147 }
2148 ret = btrfs_delete_one_dir_name(trans, root, path, di);
925baedd 2149 btrfs_release_path(root, path);
39279cc3 2150
e02119d5
CM		 2151 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2152 inode, dir->i_ino);
49eb7e46
CM		 2153 BUG_ON(ret != 0 && ret != -ENOENT);
2154 if (ret != -ENOENT)
2155 BTRFS_I(dir)->log_dirty_trans = trans->transid;
e02119d5
CM		 2156
2157 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2158 dir, index);
2159 BUG_ON(ret);
39279cc3
CM		 2160err:
2161 btrfs_free_path(path);
e02119d5
CM		 2162 if (ret)
2163 goto out;
2164
2165 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2166 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2167 btrfs_update_inode(trans, root, dir);
2168 btrfs_drop_nlink(inode);
2169 ret = btrfs_update_inode(trans, root, inode);
2170 dir->i_sb->s_dirt = 1;
2171out:
39279cc3
CM		 2172 return ret;
2173}
2174
2175static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2176{
2177 struct btrfs_root *root;
2178 struct btrfs_trans_handle *trans;
7b128766 2179 struct inode *inode = dentry->d_inode;
39279cc3 2180 int ret;
1832a6d5 2181 unsigned long nr = 0;
39279cc3
CM		 2182
2183 root = BTRFS_I(dir)->root;
1832a6d5
CM		 2184
2185 ret = btrfs_check_free_space(root, 1, 1);
2186 if (ret)
2187 goto fail;
2188
39279cc3 2189 trans = btrfs_start_transaction(root, 1);
5f39d397 2190
39279cc3 2191 btrfs_set_trans_block_group(trans, dir);
e02119d5
CM		 2192 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2193 dentry->d_name.name, dentry->d_name.len);
7b128766
JB		 2194
2195 if (inode->i_nlink == 0)
2196 ret = btrfs_orphan_add(trans, inode);
2197
d3c2fdcf 2198 nr = trans->blocks_used;
5f39d397 2199
89ce8a63 2200 btrfs_end_transaction_throttle(trans, root);
1832a6d5 2201fail:
d3c2fdcf 2202 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2203 return ret;
2204}
2205
2206static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2207{
2208 struct inode *inode = dentry->d_inode;
1832a6d5 2209 int err = 0;
39279cc3
CM		 2210 int ret;
2211 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2212 struct btrfs_trans_handle *trans;
1832a6d5 2213 unsigned long nr = 0;
39279cc3 2214
925baedd 2215 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
134d4512 2216 return -ENOTEMPTY;
925baedd 2217 }
134d4512 2218
1832a6d5
CM		 2219 ret = btrfs_check_free_space(root, 1, 1);
2220 if (ret)
2221 goto fail;
2222
39279cc3
CM		 2223 trans = btrfs_start_transaction(root, 1);
2224 btrfs_set_trans_block_group(trans, dir);
39279cc3 2225
7b128766
JB		 2226 err = btrfs_orphan_add(trans, inode);
2227 if (err)
2228 goto fail_trans;
2229
39279cc3 2230 /* now the directory is empty */
e02119d5
CM		 2231 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2232 dentry->d_name.name, dentry->d_name.len);
39279cc3 2233 if (!err) {
dbe674a9 2234 btrfs_i_size_write(inode, 0);
39279cc3 2235 }
3954401f 2236
7b128766 2237fail_trans:
d3c2fdcf 2238 nr = trans->blocks_used;
89ce8a63 2239 ret = btrfs_end_transaction_throttle(trans, root);
1832a6d5 2240fail:
d3c2fdcf 2241 btrfs_btree_balance_dirty(root, nr);
3954401f 2242
39279cc3
CM		 2243 if (ret && !err)
2244 err = ret;
2245 return err;
2246}
2247
323ac95b
CM		 2248/*
2249 * when truncating bytes in a file, it is possible to avoid reading
2250 * the leaves that contain only checksum items. This can be the
2251 * majority of the IO required to delete a large file, but it must
2252 * be done carefully.
2253 *
2254 * The keys in the level just above the leaves are checked to make sure
2255 * the lowest key in a given leaf is a csum key, and starts at an offset
2256 * after the new size.
2257 *
2258 * Then the key for the next leaf is checked to make sure it also has
2259 * a checksum item for the same file. If it does, we know our target leaf
2260 * contains only checksum items, and it can be safely freed without reading
2261 * it.
2262 *
2263 * This is just an optimization targeted at large files. It may do
2264 * nothing. It will return 0 unless things went badly.
2265 */
2266static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2267 struct btrfs_root *root,
2268 struct btrfs_path *path,
2269 struct inode *inode, u64 new_size)
2270{
2271 struct btrfs_key key;
2272 int ret;
2273 int nritems;
2274 struct btrfs_key found_key;
2275 struct btrfs_key other_key;
5b84e8d6
YZ		 2276 struct btrfs_leaf_ref *ref;
2277 u64 leaf_gen;
2278 u64 leaf_start;
323ac95b
CM		 2279
2280 path->lowest_level = 1;
2281 key.objectid = inode->i_ino;
2282 key.type = BTRFS_CSUM_ITEM_KEY;
2283 key.offset = new_size;
2284again:
2285 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2286 if (ret < 0)
2287 goto out;
2288
2289 if (path->nodes[1] == NULL) {
2290 ret = 0;
2291 goto out;
2292 }
2293 ret = 0;
2294 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2295 nritems = btrfs_header_nritems(path->nodes[1]);
2296
2297 if (!nritems)
2298 goto out;
2299
2300 if (path->slots[1] >= nritems)
2301 goto next_node;
2302
2303 /* did we find a key greater than anything we want to delete? */
2304 if (found_key.objectid > inode->i_ino ||
2305 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2306 goto out;
2307
2308 /* we check the next key in the node to make sure the leave contains
2309 * only checksum items. This comparison doesn't work if our
2310 * leaf is the last one in the node
2311 */
2312 if (path->slots[1] + 1 >= nritems) {
2313next_node:
2314 /* search forward from the last key in the node, this
2315 * will bring us into the next node in the tree
2316 */
2317 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2318
2319 /* unlikely, but we inc below, so check to be safe */
2320 if (found_key.offset == (u64)-1)
2321 goto out;
2322
2323 /* search_forward needs a path with locks held, do the
2324 * search again for the original key. It is possible
2325 * this will race with a balance and return a path that
2326 * we could modify, but this drop is just an optimization
2327 * and is allowed to miss some leaves.
2328 */
2329 btrfs_release_path(root, path);
2330 found_key.offset++;
2331
2332 /* setup a max key for search_forward */
2333 other_key.offset = (u64)-1;
2334 other_key.type = key.type;
2335 other_key.objectid = key.objectid;
2336
2337 path->keep_locks = 1;
2338 ret = btrfs_search_forward(root, &found_key, &other_key,
2339 path, 0, 0);
2340 path->keep_locks = 0;
2341 if (ret || found_key.objectid != key.objectid ||
2342 found_key.type != key.type) {
2343 ret = 0;
2344 goto out;
2345 }
2346
2347 key.offset = found_key.offset;
2348 btrfs_release_path(root, path);
2349 cond_resched();
2350 goto again;
2351 }
2352
2353 /* we know there's one more slot after us in the tree,
2354 * read that key so we can verify it is also a checksum item
2355 */
2356 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2357
2358 if (found_key.objectid < inode->i_ino)
2359 goto next_key;
2360
2361 if (found_key.type != key.type || found_key.offset < new_size)
2362 goto next_key;
2363
2364 /*
2365 * if the key for the next leaf isn't a csum key from this objectid,
2366 * we can't be sure there aren't good items inside this leaf.
2367 * Bail out
2368 */
2369 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2370 goto out;
2371
5b84e8d6
YZ		 2372 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2373 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
323ac95b
CM		 2374 /*
2375 * it is safe to delete this leaf, it contains only
2376 * csum items from this inode at an offset >= new_size
2377 */
5b84e8d6 2378 ret = btrfs_del_leaf(trans, root, path, leaf_start);
323ac95b
CM		 2379 BUG_ON(ret);
2380
5b84e8d6
YZ		 2381 if (root->ref_cows && leaf_gen < trans->transid) {
2382 ref = btrfs_alloc_leaf_ref(root, 0);
2383 if (ref) {
2384 ref->root_gen = root->root_key.offset;
2385 ref->bytenr = leaf_start;
2386 ref->owner = 0;
2387 ref->generation = leaf_gen;
2388 ref->nritems = 0;
2389
2390 ret = btrfs_add_leaf_ref(root, ref, 0);
2391 WARN_ON(ret);
2392 btrfs_free_leaf_ref(root, ref);
2393 } else {
2394 WARN_ON(1);
2395 }
2396 }
323ac95b
CM		 2397next_key:
2398 btrfs_release_path(root, path);
2399
2400 if (other_key.objectid == inode->i_ino &&
2401 other_key.type == key.type && other_key.offset > key.offset) {
2402 key.offset = other_key.offset;
2403 cond_resched();
2404 goto again;
2405 }
2406 ret = 0;
2407out:
2408 /* fixup any changes we've made to the path */
2409 path->lowest_level = 0;
2410 path->keep_locks = 0;
2411 btrfs_release_path(root, path);
2412 return ret;
2413}
2414
39279cc3
CM		 2415/*
2416 * this can truncate away extent items, csum items and directory items.
2417 * It starts at a high offset and removes keys until it can't find
d352ac68 2418 * any higher than new_size
39279cc3
CM		 2419 *
2420 * csum items that cross the new i_size are truncated to the new size
2421 * as well.
7b128766
JB		 2422 *
2423 * min_type is the minimum key type to truncate down to. If set to 0, this
2424 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2425 */
e02119d5
CM		 2426noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2427 struct btrfs_root *root,
2428 struct inode *inode,
2429 u64 new_size, u32 min_type)
39279cc3
CM		 2430{
2431 int ret;
2432 struct btrfs_path *path;
2433 struct btrfs_key key;
5f39d397 2434 struct btrfs_key found_key;
39279cc3 2435 u32 found_type;
5f39d397 2436 struct extent_buffer *leaf;
39279cc3
CM		 2437 struct btrfs_file_extent_item *fi;
2438 u64 extent_start = 0;
db94535d 2439 u64 extent_num_bytes = 0;
39279cc3 2440 u64 item_end = 0;
7bb86316 2441 u64 root_gen = 0;
d8d5f3e1 2442 u64 root_owner = 0;
39279cc3
CM		 2443 int found_extent;
2444 int del_item;
85e21bac
CM		 2445 int pending_del_nr = 0;
2446 int pending_del_slot = 0;
179e29e4 2447 int extent_type = -1;
771ed689 2448 int encoding;
3b951516 2449 u64 mask = root->sectorsize - 1;
39279cc3 2450
e02119d5 2451 if (root->ref_cows)
5b21f2ed 2452 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
39279cc3 2453 path = btrfs_alloc_path();
3c69faec 2454 path->reada = -1;
39279cc3 2455 BUG_ON(!path);
5f39d397 2456
39279cc3
CM		 2457 /* FIXME, add redo link to tree so we don't leak on crash */
2458 key.objectid = inode->i_ino;
2459 key.offset = (u64)-1;
5f39d397
CM		 2460 key.type = (u8)-1;
2461
85e21bac 2462 btrfs_init_path(path);
323ac95b
CM		 2463
2464 ret = drop_csum_leaves(trans, root, path, inode, new_size);
2465 BUG_ON(ret);
2466
85e21bac
CM		 2467search_again:
2468 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2469 if (ret < 0) {
2470 goto error;
2471 }
2472 if (ret > 0) {
e02119d5
CM		 2473 /* there are no items in the tree for us to truncate, we're
2474 * done
2475 */
2476 if (path->slots[0] == 0) {
2477 ret = 0;
2478 goto error;
2479 }
85e21bac
CM		 2480 path->slots[0]--;
2481 }
2482
39279cc3 2483 while(1) {
39279cc3 2484 fi = NULL;
5f39d397
CM		 2485 leaf = path->nodes[0];
2486 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2487 found_type = btrfs_key_type(&found_key);
771ed689 2488 encoding = 0;
39279cc3 2489
5f39d397 2490 if (found_key.objectid != inode->i_ino)
39279cc3 2491 break;
5f39d397 2492
85e21bac 2493 if (found_type < min_type)
39279cc3
CM		 2494 break;
2495
5f39d397 2496 item_end = found_key.offset;
39279cc3 2497 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 2498 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 2499 struct btrfs_file_extent_item);
179e29e4 2500 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM		 2501 encoding = btrfs_file_extent_compression(leaf, fi);
2502 encoding |= btrfs_file_extent_encryption(leaf, fi);
2503 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2504
179e29e4 2505 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 2506 item_end +=
db94535d 2507 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 2508 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 2509 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 2510 fi);
39279cc3 2511 }
008630c1 2512 item_end--;
39279cc3
CM		 2513 }
2514 if (found_type == BTRFS_CSUM_ITEM_KEY) {
2515 ret = btrfs_csum_truncate(trans, root, path,
e02119d5 2516 new_size);
39279cc3
CM		 2517 BUG_ON(ret);
2518 }
e02119d5 2519 if (item_end < new_size) {
b888db2b
CM		 2520 if (found_type == BTRFS_DIR_ITEM_KEY) {
2521 found_type = BTRFS_INODE_ITEM_KEY;
2522 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2523 found_type = BTRFS_CSUM_ITEM_KEY;
85e21bac
CM		 2524 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2525 found_type = BTRFS_XATTR_ITEM_KEY;
2526 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2527 found_type = BTRFS_INODE_REF_KEY;
b888db2b
CM		 2528 } else if (found_type) {
2529 found_type--;
2530 } else {
2531 break;
39279cc3 2532 }
a61721d5 2533 btrfs_set_key_type(&key, found_type);
85e21bac 2534 goto next;
39279cc3 2535 }
e02119d5 2536 if (found_key.offset >= new_size)
39279cc3
CM		 2537 del_item = 1;
2538 else
2539 del_item = 0;
2540 found_extent = 0;
2541
2542 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 2543 if (found_type != BTRFS_EXTENT_DATA_KEY)
2544 goto delete;
2545
2546 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 2547 u64 num_dec;
db94535d 2548 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 2549 if (!del_item && !encoding) {
db94535d
CM		 2550 u64 orig_num_bytes =
2551 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 2552 extent_num_bytes = new_size -
5f39d397 2553 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 2554 extent_num_bytes = extent_num_bytes &
2555 ~((u64)root->sectorsize - 1);
db94535d
CM		 2556 btrfs_set_file_extent_num_bytes(leaf, fi,
2557 extent_num_bytes);
2558 num_dec = (orig_num_bytes -
9069218d 2559 extent_num_bytes);
e02119d5 2560 if (root->ref_cows && extent_start != 0)
a76a3cd4 2561 inode_sub_bytes(inode, num_dec);
5f39d397 2562 btrfs_mark_buffer_dirty(leaf);
39279cc3 2563 } else {
db94535d
CM		 2564 extent_num_bytes =
2565 btrfs_file_extent_disk_num_bytes(leaf,
2566 fi);
39279cc3 2567 /* FIXME blocksize != 4096 */
9069218d 2568 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 2569 if (extent_start != 0) {
2570 found_extent = 1;
e02119d5 2571 if (root->ref_cows)
a76a3cd4 2572 inode_sub_bytes(inode, num_dec);
e02119d5 2573 }
31840ae1 2574 root_gen = btrfs_header_generation(leaf);
d8d5f3e1 2575 root_owner = btrfs_header_owner(leaf);
39279cc3 2576 }
9069218d 2577 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 2578 /*
2579 * we can't truncate inline items that have had
2580 * special encodings
2581 */
2582 if (!del_item &&
2583 btrfs_file_extent_compression(leaf, fi) == 0 &&
2584 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2585 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 2586 u32 size = new_size - found_key.offset;
2587
2588 if (root->ref_cows) {
a76a3cd4
YZ		 2589 inode_sub_bytes(inode, item_end + 1 -
2590 new_size);
e02119d5
CM		 2591 }
2592 size =
2593 btrfs_file_extent_calc_inline_size(size);
9069218d 2594 ret = btrfs_truncate_item(trans, root, path,
e02119d5 2595 size, 1);
9069218d 2596 BUG_ON(ret);
e02119d5 2597 } else if (root->ref_cows) {
a76a3cd4
YZ		 2598 inode_sub_bytes(inode, item_end + 1 -
2599 found_key.offset);
9069218d 2600 }
39279cc3 2601 }
179e29e4 2602delete:
39279cc3 2603 if (del_item) {
85e21bac
CM		 2604 if (!pending_del_nr) {
2605 /* no pending yet, add ourselves */
2606 pending_del_slot = path->slots[0];
2607 pending_del_nr = 1;
2608 } else if (pending_del_nr &&
2609 path->slots[0] + 1 == pending_del_slot) {
2610 /* hop on the pending chunk */
2611 pending_del_nr++;
2612 pending_del_slot = path->slots[0];
2613 } else {
2614 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2615 }
39279cc3
CM		 2616 } else {
2617 break;
2618 }
39279cc3
CM		 2619 if (found_extent) {
2620 ret = btrfs_free_extent(trans, root, extent_start,
7bb86316 2621 extent_num_bytes,
31840ae1 2622 leaf->start, root_owner,
3bb1a1bc 2623 root_gen, inode->i_ino, 0);
39279cc3
CM		 2624 BUG_ON(ret);
2625 }
85e21bac
CM		 2626next:
2627 if (path->slots[0] == 0) {
2628 if (pending_del_nr)
2629 goto del_pending;
2630 btrfs_release_path(root, path);
2631 goto search_again;
2632 }
2633
2634 path->slots[0]--;
2635 if (pending_del_nr &&
2636 path->slots[0] + 1 != pending_del_slot) {
2637 struct btrfs_key debug;
2638del_pending:
2639 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2640 pending_del_slot);
2641 ret = btrfs_del_items(trans, root, path,
2642 pending_del_slot,
2643 pending_del_nr);
2644 BUG_ON(ret);
2645 pending_del_nr = 0;
2646 btrfs_release_path(root, path);
2647 goto search_again;
2648 }
39279cc3
CM		 2649 }
2650 ret = 0;
2651error:
85e21bac
CM		 2652 if (pending_del_nr) {
2653 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2654 pending_del_nr);
2655 }
39279cc3
CM		 2656 btrfs_free_path(path);
2657 inode->i_sb->s_dirt = 1;
2658 return ret;
2659}
2660
2661/*
2662 * taken from block_truncate_page, but does cow as it zeros out
2663 * any bytes left in the last page in the file.
2664 */
2665static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2666{
2667 struct inode *inode = mapping->host;
db94535d 2668 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 2669 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2670 struct btrfs_ordered_extent *ordered;
2671 char *kaddr;
db94535d 2672 u32 blocksize = root->sectorsize;
39279cc3
CM		 2673 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2674 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2675 struct page *page;
39279cc3 2676 int ret = 0;
a52d9a80 2677 u64 page_start;
e6dcd2dc 2678 u64 page_end;
39279cc3
CM		 2679
2680 if ((offset & (blocksize - 1)) == 0)
2681 goto out;
2682
2683 ret = -ENOMEM;
211c17f5 2684again:
39279cc3
CM		 2685 page = grab_cache_page(mapping, index);
2686 if (!page)
2687 goto out;
e6dcd2dc
CM		 2688
2689 page_start = page_offset(page);
2690 page_end = page_start + PAGE_CACHE_SIZE - 1;
2691
39279cc3 2692 if (!PageUptodate(page)) {
9ebefb18 2693 ret = btrfs_readpage(NULL, page);
39279cc3 2694 lock_page(page);
211c17f5
CM		 2695 if (page->mapping != mapping) {
2696 unlock_page(page);
2697 page_cache_release(page);
2698 goto again;
2699 }
39279cc3
CM		 2700 if (!PageUptodate(page)) {
2701 ret = -EIO;
89642229 2702 goto out_unlock;
39279cc3
CM		 2703 }
2704 }
211c17f5 2705 wait_on_page_writeback(page);
e6dcd2dc
CM		 2706
2707 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2708 set_page_extent_mapped(page);
2709
2710 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2711 if (ordered) {
2712 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2713 unlock_page(page);
2714 page_cache_release(page);
eb84ae03 2715 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 2716 btrfs_put_ordered_extent(ordered);
2717 goto again;
2718 }
2719
ea8c2819 2720 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc
CM		 2721 ret = 0;
2722 if (offset != PAGE_CACHE_SIZE) {
2723 kaddr = kmap(page);
2724 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2725 flush_dcache_page(page);
2726 kunmap(page);
2727 }
247e743c 2728 ClearPageChecked(page);
e6dcd2dc
CM		 2729 set_page_dirty(page);
2730 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
39279cc3 2731
89642229 2732out_unlock:
39279cc3
CM		 2733 unlock_page(page);
2734 page_cache_release(page);
2735out:
2736 return ret;
2737}
2738
9036c102 2739int btrfs_cont_expand(struct inode *inode, loff_t size)
39279cc3 2740{
9036c102
YZ		 2741 struct btrfs_trans_handle *trans;
2742 struct btrfs_root *root = BTRFS_I(inode)->root;
2743 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2744 struct extent_map *em;
2745 u64 mask = root->sectorsize - 1;
2746 u64 hole_start = (inode->i_size + mask) & ~mask;
2747 u64 block_end = (size + mask) & ~mask;
2748 u64 last_byte;
2749 u64 cur_offset;
2750 u64 hole_size;
39279cc3
CM		 2751 int err;
2752
9036c102
YZ		 2753 if (size <= hole_start)
2754 return 0;
2755
2756 err = btrfs_check_free_space(root, 1, 0);
39279cc3
CM		 2757 if (err)
2758 return err;
2759
9036c102 2760 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2bf5a725 2761
9036c102
YZ		 2762 while (1) {
2763 struct btrfs_ordered_extent *ordered;
2764 btrfs_wait_ordered_range(inode, hole_start,
2765 block_end - hole_start);
2766 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2767 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2768 if (!ordered)
2769 break;
2770 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2771 btrfs_put_ordered_extent(ordered);
2772 }
39279cc3 2773
9036c102
YZ		 2774 trans = btrfs_start_transaction(root, 1);
2775 btrfs_set_trans_block_group(trans, inode);
39279cc3 2776
9036c102
YZ		 2777 cur_offset = hole_start;
2778 while (1) {
2779 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2780 block_end - cur_offset, 0);
2781 BUG_ON(IS_ERR(em) || !em);
2782 last_byte = min(extent_map_end(em), block_end);
2783 last_byte = (last_byte + mask) & ~mask;
2784 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
771ed689 2785 u64 hint_byte = 0;
9036c102 2786 hole_size = last_byte - cur_offset;
771ed689
CM		 2787 err = btrfs_drop_extents(trans, root, inode,
2788 cur_offset,
2789 cur_offset + hole_size,
2790 cur_offset, &hint_byte);
2791 if (err)
2792 break;
9036c102
YZ		 2793 err = btrfs_insert_file_extent(trans, root,
2794 inode->i_ino, cur_offset, 0,
2795 0, hole_size, 0, hole_size,
2796 0, 0, 0);
2797 btrfs_drop_extent_cache(inode, hole_start,
2798 last_byte - 1, 0);
2799 }
2800 free_extent_map(em);
2801 cur_offset = last_byte;
2802 if (err || cur_offset >= block_end)
2803 break;
2804 }
1832a6d5 2805
9036c102
YZ		 2806 btrfs_end_transaction(trans, root);
2807 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2808 return err;
2809}
39279cc3 2810
9036c102
YZ		 2811static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2812{
2813 struct inode *inode = dentry->d_inode;
2814 int err;
39279cc3 2815
9036c102
YZ		 2816 err = inode_change_ok(inode, attr);
2817 if (err)
2818 return err;
2bf5a725 2819
9036c102
YZ		 2820 if (S_ISREG(inode->i_mode) &&
2821 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2822 err = btrfs_cont_expand(inode, attr->ia_size);
54aa1f4d
CM		 2823 if (err)
2824 return err;
39279cc3 2825 }
9036c102 2826
39279cc3 2827 err = inode_setattr(inode, attr);
33268eaf
JB		 2828
2829 if (!err && ((attr->ia_valid & ATTR_MODE)))
2830 err = btrfs_acl_chmod(inode);
39279cc3
CM		 2831 return err;
2832}
61295eb8 2833
39279cc3
CM		 2834void btrfs_delete_inode(struct inode *inode)
2835{
2836 struct btrfs_trans_handle *trans;
2837 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 2838 unsigned long nr;
39279cc3
CM		 2839 int ret;
2840
2841 truncate_inode_pages(&inode->i_data, 0);
2842 if (is_bad_inode(inode)) {
7b128766 2843 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 2844 goto no_delete;
2845 }
4a096752 2846 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 2847
dbe674a9 2848 btrfs_i_size_write(inode, 0);
39279cc3 2849 trans = btrfs_start_transaction(root, 1);
5f39d397 2850
39279cc3 2851 btrfs_set_trans_block_group(trans, inode);
e02119d5 2852 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
7b128766
JB		 2853 if (ret) {
2854 btrfs_orphan_del(NULL, inode);
54aa1f4d 2855 goto no_delete_lock;
7b128766
JB		 2856 }
2857
2858 btrfs_orphan_del(trans, inode);
85e21bac 2859
d3c2fdcf 2860 nr = trans->blocks_used;
85e21bac 2861 clear_inode(inode);
5f39d397 2862
39279cc3 2863 btrfs_end_transaction(trans, root);
d3c2fdcf 2864 btrfs_btree_balance_dirty(root, nr);
39279cc3 2865 return;
54aa1f4d
CM		 2866
2867no_delete_lock:
d3c2fdcf 2868 nr = trans->blocks_used;
54aa1f4d 2869 btrfs_end_transaction(trans, root);
d3c2fdcf 2870 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2871no_delete:
2872 clear_inode(inode);
2873}
2874
2875/*
2876 * this returns the key found in the dir entry in the location pointer.
2877 * If no dir entries were found, location->objectid is 0.
2878 */
2879static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2880 struct btrfs_key *location)
2881{
2882 const char *name = dentry->d_name.name;
2883 int namelen = dentry->d_name.len;
2884 struct btrfs_dir_item *di;
2885 struct btrfs_path *path;
2886 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 2887 int ret = 0;
39279cc3
CM		 2888
2889 path = btrfs_alloc_path();
2890 BUG_ON(!path);
3954401f 2891
39279cc3
CM		 2892 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2893 namelen, 0);
0d9f7f3e
Y		 2894 if (IS_ERR(di))
2895 ret = PTR_ERR(di);
39279cc3 2896 if (!di || IS_ERR(di)) {
3954401f 2897 goto out_err;
39279cc3 2898 }
5f39d397 2899 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 2900out:
39279cc3
CM		 2901 btrfs_free_path(path);
2902 return ret;
3954401f
CM		 2903out_err:
2904 location->objectid = 0;
2905 goto out;
39279cc3
CM		 2906}
2907
2908/*
2909 * when we hit a tree root in a directory, the btrfs part of the inode
2910 * needs to be changed to reflect the root directory of the tree root. This
2911 * is kind of like crossing a mount point.
2912 */
2913static int fixup_tree_root_location(struct btrfs_root *root,
2914 struct btrfs_key *location,
58176a96
JB		 2915 struct btrfs_root **sub_root,
2916 struct dentry *dentry)
39279cc3 2917{
39279cc3
CM		 2918 struct btrfs_root_item *ri;
2919
2920 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2921 return 0;
2922 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2923 return 0;
2924
58176a96
JB		 2925 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2926 dentry->d_name.name,
2927 dentry->d_name.len);
39279cc3
CM		 2928 if (IS_ERR(*sub_root))
2929 return PTR_ERR(*sub_root);
2930
2931 ri = &(*sub_root)->root_item;
2932 location->objectid = btrfs_root_dirid(ri);
39279cc3
CM		 2933 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2934 location->offset = 0;
2935
39279cc3
CM		 2936 return 0;
2937}
2938
e02119d5 2939static noinline void init_btrfs_i(struct inode *inode)
39279cc3 2940{
e02119d5
CM		 2941 struct btrfs_inode *bi = BTRFS_I(inode);
2942
2943 bi->i_acl = NULL;
2944 bi->i_default_acl = NULL;
2945
2946 bi->generation = 0;
2947 bi->last_trans = 0;
2948 bi->logged_trans = 0;
2949 bi->delalloc_bytes = 0;
2950 bi->disk_i_size = 0;
2951 bi->flags = 0;
2952 bi->index_cnt = (u64)-1;
49eb7e46 2953 bi->log_dirty_trans = 0;
d1310b2e
CM		 2954 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2955 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
b888db2b 2956 inode->i_mapping, GFP_NOFS);
7e38326f
CM		 2957 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2958 inode->i_mapping, GFP_NOFS);
ea8c2819 2959 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
ba1da2f4 2960 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1b1e2135 2961 mutex_init(&BTRFS_I(inode)->csum_mutex);
ee6e6504 2962 mutex_init(&BTRFS_I(inode)->extent_mutex);
e02119d5
CM		 2963 mutex_init(&BTRFS_I(inode)->log_mutex);
2964}
2965
2966static int btrfs_init_locked_inode(struct inode *inode, void *p)
2967{
2968 struct btrfs_iget_args *args = p;
2969 inode->i_ino = args->ino;
2970 init_btrfs_i(inode);
2971 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 2972 return 0;
2973}
2974
2975static int btrfs_find_actor(struct inode *inode, void *opaque)
2976{
2977 struct btrfs_iget_args *args = opaque;
2978 return (args->ino == inode->i_ino &&
2979 args->root == BTRFS_I(inode)->root);
2980}
2981
5b21f2ed
ZY		 2982struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2983 struct btrfs_root *root, int wait)
2984{
2985 struct inode *inode;
2986 struct btrfs_iget_args args;
2987 args.ino = objectid;
2988 args.root = root;
2989
2990 if (wait) {
2991 inode = ilookup5(s, objectid, btrfs_find_actor,
2992 (void *)&args);
2993 } else {
2994 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
2995 (void *)&args);
2996 }
2997 return inode;
2998}
2999
39279cc3
CM		 3000struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3001 struct btrfs_root *root)
3002{
3003 struct inode *inode;
3004 struct btrfs_iget_args args;
3005 args.ino = objectid;
3006 args.root = root;
3007
3008 inode = iget5_locked(s, objectid, btrfs_find_actor,
3009 btrfs_init_locked_inode,
3010 (void *)&args);
3011 return inode;
3012}
3013
1a54ef8c
BR		 3014/* Get an inode object given its location and corresponding root.
3015 * Returns in *is_new if the inode was read from disk
3016 */
3017struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3018 struct btrfs_root *root, int *is_new)
3019{
3020 struct inode *inode;
3021
3022 inode = btrfs_iget_locked(s, location->objectid, root);
3023 if (!inode)
3024 return ERR_PTR(-EACCES);
3025
3026 if (inode->i_state & I_NEW) {
3027 BTRFS_I(inode)->root = root;
3028 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3029 btrfs_read_locked_inode(inode);
3030 unlock_new_inode(inode);
3031 if (is_new)
3032 *is_new = 1;
3033 } else {
3034 if (is_new)
3035 *is_new = 0;
3036 }
3037
3038 return inode;
3039}
3040
39279cc3
CM		 3041static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3042 struct nameidata *nd)
3043{
3044 struct inode * inode;
3045 struct btrfs_inode *bi = BTRFS_I(dir);
3046 struct btrfs_root *root = bi->root;
3047 struct btrfs_root *sub_root = root;
3048 struct btrfs_key location;
c146afad 3049 int ret, new;
39279cc3
CM		 3050
3051 if (dentry->d_name.len > BTRFS_NAME_LEN)
3052 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3053
39279cc3 3054 ret = btrfs_inode_by_name(dir, dentry, &location);
5f39d397 3055
39279cc3
CM		 3056 if (ret < 0)
3057 return ERR_PTR(ret);
5f39d397 3058
39279cc3
CM		 3059 inode = NULL;
3060 if (location.objectid) {
58176a96
JB		 3061 ret = fixup_tree_root_location(root, &location, &sub_root,
3062 dentry);
39279cc3
CM		 3063 if (ret < 0)
3064 return ERR_PTR(ret);
3065 if (ret > 0)
3066 return ERR_PTR(-ENOENT);
1a54ef8c
BR		 3067 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3068 if (IS_ERR(inode))
3069 return ERR_CAST(inode);
3070
3071 /* the inode and parent dir are two different roots */
3072 if (new && root != sub_root) {
3073 igrab(inode);
3074 sub_root->inode = inode;
39279cc3
CM		 3075 }
3076 }
7b128766 3077
39279cc3
CM		 3078 return d_splice_alias(inode, dentry);
3079}
3080
39279cc3
CM		 3081static unsigned char btrfs_filetype_table[] = {
3082 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3083};
3084
cbdf5a24
DW		 3085static int btrfs_real_readdir(struct file *filp, void *dirent,
3086 filldir_t filldir)
39279cc3 3087{
6da6abae 3088 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 3089 struct btrfs_root *root = BTRFS_I(inode)->root;
3090 struct btrfs_item *item;
3091 struct btrfs_dir_item *di;
3092 struct btrfs_key key;
5f39d397 3093 struct btrfs_key found_key;
39279cc3
CM		 3094 struct btrfs_path *path;
3095 int ret;
3096 u32 nritems;
5f39d397 3097 struct extent_buffer *leaf;
39279cc3
CM		 3098 int slot;
3099 int advance;
3100 unsigned char d_type;
3101 int over = 0;
3102 u32 di_cur;
3103 u32 di_total;
3104 u32 di_len;
3105 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 3106 char tmp_name[32];
3107 char *name_ptr;
3108 int name_len;
39279cc3
CM		 3109
3110 /* FIXME, use a real flag for deciding about the key type */
3111 if (root->fs_info->tree_root == root)
3112 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 3113
3954401f
CM		 3114 /* special case for "." */
3115 if (filp->f_pos == 0) {
3116 over = filldir(dirent, ".", 1,
3117 1, inode->i_ino,
3118 DT_DIR);
3119 if (over)
3120 return 0;
3121 filp->f_pos = 1;
3122 }
3954401f
CM		 3123 /* special case for .., just use the back ref */
3124 if (filp->f_pos == 1) {
5ecc7e5d 3125 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 3126 over = filldir(dirent, "..", 2,
5ecc7e5d 3127 2, pino, DT_DIR);
3954401f 3128 if (over)
49593bfa 3129 return 0;
3954401f
CM		 3130 filp->f_pos = 2;
3131 }
3132
49593bfa
DW		 3133 path = btrfs_alloc_path();
3134 path->reada = 2;
3135
39279cc3
CM		 3136 btrfs_set_key_type(&key, key_type);
3137 key.offset = filp->f_pos;
49593bfa 3138 key.objectid = inode->i_ino;
5f39d397 3139
39279cc3
CM		 3140 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3141 if (ret < 0)
3142 goto err;
3143 advance = 0;
49593bfa
DW		 3144
3145 while (1) {
5f39d397
CM		 3146 leaf = path->nodes[0];
3147 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3148 slot = path->slots[0];
3149 if (advance || slot >= nritems) {
49593bfa 3150 if (slot >= nritems - 1) {
39279cc3
CM		 3151 ret = btrfs_next_leaf(root, path);
3152 if (ret)
3153 break;
5f39d397
CM		 3154 leaf = path->nodes[0];
3155 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3156 slot = path->slots[0];
3157 } else {
3158 slot++;
3159 path->slots[0]++;
3160 }
3161 }
3162 advance = 1;
5f39d397
CM		 3163 item = btrfs_item_nr(leaf, slot);
3164 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3165
3166 if (found_key.objectid != key.objectid)
39279cc3 3167 break;
5f39d397 3168 if (btrfs_key_type(&found_key) != key_type)
39279cc3 3169 break;
5f39d397 3170 if (found_key.offset < filp->f_pos)
39279cc3 3171 continue;
5f39d397
CM		 3172
3173 filp->f_pos = found_key.offset;
49593bfa 3174
39279cc3
CM		 3175 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3176 di_cur = 0;
5f39d397 3177 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 3178
3179 while (di_cur < di_total) {
5f39d397
CM		 3180 struct btrfs_key location;
3181
3182 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 3183 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 3184 name_ptr = tmp_name;
3185 } else {
3186 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 3187 if (!name_ptr) {
3188 ret = -ENOMEM;
3189 goto err;
3190 }
5f39d397
CM		 3191 }
3192 read_extent_buffer(leaf, name_ptr,
3193 (unsigned long)(di + 1), name_len);
3194
3195 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3196 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5f39d397 3197 over = filldir(dirent, name_ptr, name_len,
49593bfa 3198 found_key.offset, location.objectid,
39279cc3 3199 d_type);
5f39d397
CM		 3200
3201 if (name_ptr != tmp_name)
3202 kfree(name_ptr);
3203
39279cc3
CM		 3204 if (over)
3205 goto nopos;
49593bfa 3206
5103e947 3207 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 3208 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 3209 di_cur += di_len;
3210 di = (struct btrfs_dir_item *)((char *)di + di_len);
3211 }
3212 }
49593bfa
DW		 3213
3214 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07
YZ		 3215 if (key_type == BTRFS_DIR_INDEX_KEY)
3216 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3217 else
3218 filp->f_pos++;
39279cc3
CM		 3219nopos:
3220 ret = 0;
3221err:
39279cc3 3222 btrfs_free_path(path);
39279cc3
CM		 3223 return ret;
3224}
3225
3226int btrfs_write_inode(struct inode *inode, int wait)
3227{
3228 struct btrfs_root *root = BTRFS_I(inode)->root;
3229 struct btrfs_trans_handle *trans;
3230 int ret = 0;
3231
c146afad 3232 if (root->fs_info->btree_inode == inode)
4ca8b41e
CM		 3233 return 0;
3234
39279cc3 3235 if (wait) {
f9295749 3236 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3237 btrfs_set_trans_block_group(trans, inode);
3238 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 3239 }
3240 return ret;
3241}
3242
3243/*
54aa1f4d 3244 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 3245 * inode changes. But, it is most likely to find the inode in cache.
3246 * FIXME, needs more benchmarking...there are no reasons other than performance
3247 * to keep or drop this code.
3248 */
3249void btrfs_dirty_inode(struct inode *inode)
3250{
3251 struct btrfs_root *root = BTRFS_I(inode)->root;
3252 struct btrfs_trans_handle *trans;
3253
f9295749 3254 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3255 btrfs_set_trans_block_group(trans, inode);
3256 btrfs_update_inode(trans, root, inode);
3257 btrfs_end_transaction(trans, root);
39279cc3
CM		 3258}
3259
d352ac68
CM		 3260/*
3261 * find the highest existing sequence number in a directory
3262 * and then set the in-memory index_cnt variable to reflect
3263 * free sequence numbers
3264 */
aec7477b
JB		 3265static int btrfs_set_inode_index_count(struct inode *inode)
3266{
3267 struct btrfs_root *root = BTRFS_I(inode)->root;
3268 struct btrfs_key key, found_key;
3269 struct btrfs_path *path;
3270 struct extent_buffer *leaf;
3271 int ret;
3272
3273 key.objectid = inode->i_ino;
3274 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3275 key.offset = (u64)-1;
3276
3277 path = btrfs_alloc_path();
3278 if (!path)
3279 return -ENOMEM;
3280
3281 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3282 if (ret < 0)
3283 goto out;
3284 /* FIXME: we should be able to handle this */
3285 if (ret == 0)
3286 goto out;
3287 ret = 0;
3288
3289 /*
3290 * MAGIC NUMBER EXPLANATION:
3291 * since we search a directory based on f_pos we have to start at 2
3292 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3293 * else has to start at 2
3294 */
3295 if (path->slots[0] == 0) {
3296 BTRFS_I(inode)->index_cnt = 2;
3297 goto out;
3298 }
3299
3300 path->slots[0]--;
3301
3302 leaf = path->nodes[0];
3303 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3304
3305 if (found_key.objectid != inode->i_ino ||
3306 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3307 BTRFS_I(inode)->index_cnt = 2;
3308 goto out;
3309 }
3310
3311 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3312out:
3313 btrfs_free_path(path);
3314 return ret;
3315}
3316
d352ac68
CM		 3317/*
3318 * helper to find a free sequence number in a given directory. This current
3319 * code is very simple, later versions will do smarter things in the btree
3320 */
00e4e6b3
CM		 3321static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
3322 u64 *index)
aec7477b
JB		 3323{
3324 int ret = 0;
3325
3326 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3327 ret = btrfs_set_inode_index_count(dir);
8d5bf1cb 3328 if (ret) {
aec7477b 3329 return ret;
8d5bf1cb 3330 }
aec7477b
JB		 3331 }
3332
00e4e6b3 3333 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 3334 BTRFS_I(dir)->index_cnt++;
3335
3336 return ret;
3337}
3338
39279cc3
CM		 3339static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3340 struct btrfs_root *root,
aec7477b 3341 struct inode *dir,
9c58309d
CM		 3342 const char *name, int name_len,
3343 u64 ref_objectid,
39279cc3
CM		 3344 u64 objectid,
3345 struct btrfs_block_group_cache *group,
00e4e6b3 3346 int mode, u64 *index)
39279cc3
CM		 3347{
3348 struct inode *inode;
5f39d397 3349 struct btrfs_inode_item *inode_item;
6324fbf3 3350 struct btrfs_block_group_cache *new_inode_group;
39279cc3 3351 struct btrfs_key *location;
5f39d397 3352 struct btrfs_path *path;
9c58309d
CM		 3353 struct btrfs_inode_ref *ref;
3354 struct btrfs_key key[2];
3355 u32 sizes[2];
3356 unsigned long ptr;
39279cc3
CM		 3357 int ret;
3358 int owner;
3359
5f39d397
CM		 3360 path = btrfs_alloc_path();
3361 BUG_ON(!path);
3362
39279cc3
CM		 3363 inode = new_inode(root->fs_info->sb);
3364 if (!inode)
3365 return ERR_PTR(-ENOMEM);
3366
aec7477b 3367 if (dir) {
00e4e6b3 3368 ret = btrfs_set_inode_index(dir, inode, index);
aec7477b
JB		 3369 if (ret)
3370 return ERR_PTR(ret);
aec7477b
JB		 3371 }
3372 /*
3373 * index_cnt is ignored for everything but a dir,
3374 * btrfs_get_inode_index_count has an explanation for the magic
3375 * number
3376 */
e02119d5 3377 init_btrfs_i(inode);
aec7477b 3378 BTRFS_I(inode)->index_cnt = 2;
39279cc3 3379 BTRFS_I(inode)->root = root;
e02119d5 3380 BTRFS_I(inode)->generation = trans->transid;
b888db2b 3381
39279cc3
CM		 3382 if (mode & S_IFDIR)
3383 owner = 0;
3384 else
3385 owner = 1;
6324fbf3 3386 new_inode_group = btrfs_find_block_group(root, group, 0,
0b86a832 3387 BTRFS_BLOCK_GROUP_METADATA, owner);
6324fbf3
CM		 3388 if (!new_inode_group) {
3389 printk("find_block group failed\n");
3390 new_inode_group = group;
3391 }
3392 BTRFS_I(inode)->block_group = new_inode_group;
9c58309d
CM		 3393
3394 key[0].objectid = objectid;
3395 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3396 key[0].offset = 0;
3397
3398 key[1].objectid = objectid;
3399 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3400 key[1].offset = ref_objectid;
3401
3402 sizes[0] = sizeof(struct btrfs_inode_item);
3403 sizes[1] = name_len + sizeof(*ref);
3404
3405 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3406 if (ret != 0)
5f39d397
CM		 3407 goto fail;
3408
9c58309d
CM		 3409 if (objectid > root->highest_inode)
3410 root->highest_inode = objectid;
3411
39279cc3
CM		 3412 inode->i_uid = current->fsuid;
3413 inode->i_gid = current->fsgid;
3414 inode->i_mode = mode;
3415 inode->i_ino = objectid;
a76a3cd4 3416 inode_set_bytes(inode, 0);
39279cc3 3417 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 3418 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3419 struct btrfs_inode_item);
e02119d5 3420 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 3421
3422 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3423 struct btrfs_inode_ref);
3424 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 3425 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 3426 ptr = (unsigned long)(ref + 1);
3427 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3428
5f39d397
CM		 3429 btrfs_mark_buffer_dirty(path->nodes[0]);
3430 btrfs_free_path(path);
3431
39279cc3
CM		 3432 location = &BTRFS_I(inode)->location;
3433 location->objectid = objectid;
39279cc3
CM		 3434 location->offset = 0;
3435 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3436
39279cc3
CM		 3437 insert_inode_hash(inode);
3438 return inode;
5f39d397 3439fail:
aec7477b
JB		 3440 if (dir)
3441 BTRFS_I(dir)->index_cnt--;
5f39d397
CM		 3442 btrfs_free_path(path);
3443 return ERR_PTR(ret);
39279cc3
CM		 3444}
3445
3446static inline u8 btrfs_inode_type(struct inode *inode)
3447{
3448 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3449}
3450
d352ac68
CM		 3451/*
3452 * utility function to add 'inode' into 'parent_inode' with
3453 * a give name and a given sequence number.
3454 * if 'add_backref' is true, also insert a backref from the
3455 * inode to the parent directory.
3456 */
e02119d5
CM		 3457int btrfs_add_link(struct btrfs_trans_handle *trans,
3458 struct inode *parent_inode, struct inode *inode,
3459 const char *name, int name_len, int add_backref, u64 index)
39279cc3
CM		 3460{
3461 int ret;
3462 struct btrfs_key key;
e02119d5 3463 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5f39d397 3464
39279cc3 3465 key.objectid = inode->i_ino;
39279cc3
CM		 3466 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3467 key.offset = 0;
3468
e02119d5
CM		 3469 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3470 parent_inode->i_ino,
aec7477b 3471 &key, btrfs_inode_type(inode),
00e4e6b3 3472 index);
39279cc3 3473 if (ret == 0) {
9c58309d
CM		 3474 if (add_backref) {
3475 ret = btrfs_insert_inode_ref(trans, root,
e02119d5
CM		 3476 name, name_len,
3477 inode->i_ino,
3478 parent_inode->i_ino,
3479 index);
9c58309d 3480 }
dbe674a9 3481 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 3482 name_len * 2);
79c44584 3483 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 3484 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 3485 }
3486 return ret;
3487}
3488
3489static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
9c58309d 3490 struct dentry *dentry, struct inode *inode,
00e4e6b3 3491 int backref, u64 index)
39279cc3 3492{
e02119d5
CM		 3493 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3494 inode, dentry->d_name.name,
3495 dentry->d_name.len, backref, index);
39279cc3
CM		 3496 if (!err) {
3497 d_instantiate(dentry, inode);
3498 return 0;
3499 }
3500 if (err > 0)
3501 err = -EEXIST;
3502 return err;
3503}
3504
618e21d5
JB		 3505static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3506 int mode, dev_t rdev)
3507{
3508 struct btrfs_trans_handle *trans;
3509 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3510 struct inode *inode = NULL;
618e21d5
JB		 3511 int err;
3512 int drop_inode = 0;
3513 u64 objectid;
1832a6d5 3514 unsigned long nr = 0;
00e4e6b3 3515 u64 index = 0;
618e21d5
JB		 3516
3517 if (!new_valid_dev(rdev))
3518 return -EINVAL;
3519
1832a6d5
CM		 3520 err = btrfs_check_free_space(root, 1, 0);
3521 if (err)
3522 goto fail;
3523
618e21d5
JB		 3524 trans = btrfs_start_transaction(root, 1);
3525 btrfs_set_trans_block_group(trans, dir);
3526
3527 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3528 if (err) {
3529 err = -ENOSPC;
3530 goto out_unlock;
3531 }
3532
aec7477b 3533 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3534 dentry->d_name.len,
3535 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3 3536 BTRFS_I(dir)->block_group, mode, &index);
618e21d5
JB		 3537 err = PTR_ERR(inode);
3538 if (IS_ERR(inode))
3539 goto out_unlock;
3540
33268eaf
JB		 3541 err = btrfs_init_acl(inode, dir);
3542 if (err) {
3543 drop_inode = 1;
3544 goto out_unlock;
3545 }
3546
618e21d5 3547 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3548 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
618e21d5
JB		 3549 if (err)
3550 drop_inode = 1;
3551 else {
3552 inode->i_op = &btrfs_special_inode_operations;
3553 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 3554 btrfs_update_inode(trans, root, inode);
618e21d5
JB		 3555 }
3556 dir->i_sb->s_dirt = 1;
3557 btrfs_update_inode_block_group(trans, inode);
3558 btrfs_update_inode_block_group(trans, dir);
3559out_unlock:
d3c2fdcf 3560 nr = trans->blocks_used;
89ce8a63 3561 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3562fail:
618e21d5
JB		 3563 if (drop_inode) {
3564 inode_dec_link_count(inode);
3565 iput(inode);
3566 }
d3c2fdcf 3567 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 3568 return err;
3569}
3570
39279cc3
CM		 3571static int btrfs_create(struct inode *dir, struct dentry *dentry,
3572 int mode, struct nameidata *nd)
3573{
3574 struct btrfs_trans_handle *trans;
3575 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3576 struct inode *inode = NULL;
39279cc3
CM		 3577 int err;
3578 int drop_inode = 0;
1832a6d5 3579 unsigned long nr = 0;
39279cc3 3580 u64 objectid;
00e4e6b3 3581 u64 index = 0;
39279cc3 3582
1832a6d5
CM		 3583 err = btrfs_check_free_space(root, 1, 0);
3584 if (err)
3585 goto fail;
39279cc3
CM		 3586 trans = btrfs_start_transaction(root, 1);
3587 btrfs_set_trans_block_group(trans, dir);
3588
3589 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3590 if (err) {
3591 err = -ENOSPC;
3592 goto out_unlock;
3593 }
3594
aec7477b 3595 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3596 dentry->d_name.len,
3597 dentry->d_parent->d_inode->i_ino,
00e4e6b3
CM		 3598 objectid, BTRFS_I(dir)->block_group, mode,
3599 &index);
39279cc3
CM		 3600 err = PTR_ERR(inode);
3601 if (IS_ERR(inode))
3602 goto out_unlock;
3603
33268eaf
JB		 3604 err = btrfs_init_acl(inode, dir);
3605 if (err) {
3606 drop_inode = 1;
3607 goto out_unlock;
3608 }
3609
39279cc3 3610 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3611 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 3612 if (err)
3613 drop_inode = 1;
3614 else {
3615 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3616 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 3617 inode->i_fop = &btrfs_file_operations;
3618 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 3619 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3620 }
3621 dir->i_sb->s_dirt = 1;
3622 btrfs_update_inode_block_group(trans, inode);
3623 btrfs_update_inode_block_group(trans, dir);
3624out_unlock:
d3c2fdcf 3625 nr = trans->blocks_used;
ab78c84d 3626 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3627fail:
39279cc3
CM		 3628 if (drop_inode) {
3629 inode_dec_link_count(inode);
3630 iput(inode);
3631 }
d3c2fdcf 3632 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3633 return err;
3634}
3635
3636static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3637 struct dentry *dentry)
3638{
3639 struct btrfs_trans_handle *trans;
3640 struct btrfs_root *root = BTRFS_I(dir)->root;
3641 struct inode *inode = old_dentry->d_inode;
00e4e6b3 3642 u64 index;
1832a6d5 3643 unsigned long nr = 0;
39279cc3
CM		 3644 int err;
3645 int drop_inode = 0;
3646
3647 if (inode->i_nlink == 0)
3648 return -ENOENT;
3649
e02119d5 3650 btrfs_inc_nlink(inode);
1832a6d5
CM		 3651 err = btrfs_check_free_space(root, 1, 0);
3652 if (err)
3653 goto fail;
00e4e6b3 3654 err = btrfs_set_inode_index(dir, inode, &index);
aec7477b
JB		 3655 if (err)
3656 goto fail;
3657
39279cc3 3658 trans = btrfs_start_transaction(root, 1);
5f39d397 3659
39279cc3
CM		 3660 btrfs_set_trans_block_group(trans, dir);
3661 atomic_inc(&inode->i_count);
aec7477b 3662
00e4e6b3 3663 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
5f39d397 3664
39279cc3
CM		 3665 if (err)
3666 drop_inode = 1;
5f39d397 3667
39279cc3
CM		 3668 dir->i_sb->s_dirt = 1;
3669 btrfs_update_inode_block_group(trans, dir);
54aa1f4d 3670 err = btrfs_update_inode(trans, root, inode);
5f39d397 3671
54aa1f4d
CM		 3672 if (err)
3673 drop_inode = 1;
39279cc3 3674
d3c2fdcf 3675 nr = trans->blocks_used;
ab78c84d 3676 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3677fail:
39279cc3
CM		 3678 if (drop_inode) {
3679 inode_dec_link_count(inode);
3680 iput(inode);
3681 }
d3c2fdcf 3682 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3683 return err;
3684}
3685
39279cc3
CM		 3686static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3687{
b9d86667 3688 struct inode *inode = NULL;
39279cc3
CM		 3689 struct btrfs_trans_handle *trans;
3690 struct btrfs_root *root = BTRFS_I(dir)->root;
3691 int err = 0;
3692 int drop_on_err = 0;
b9d86667 3693 u64 objectid = 0;
00e4e6b3 3694 u64 index = 0;
d3c2fdcf 3695 unsigned long nr = 1;
39279cc3 3696
1832a6d5
CM		 3697 err = btrfs_check_free_space(root, 1, 0);
3698 if (err)
3699 goto out_unlock;
3700
39279cc3
CM		 3701 trans = btrfs_start_transaction(root, 1);
3702 btrfs_set_trans_block_group(trans, dir);
5f39d397 3703
39279cc3
CM		 3704 if (IS_ERR(trans)) {
3705 err = PTR_ERR(trans);
3706 goto out_unlock;
3707 }
3708
3709 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3710 if (err) {
3711 err = -ENOSPC;
3712 goto out_unlock;
3713 }
3714
aec7477b 3715 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3716 dentry->d_name.len,
3717 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 3718 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3719 &index);
39279cc3
CM		 3720 if (IS_ERR(inode)) {
3721 err = PTR_ERR(inode);
3722 goto out_fail;
3723 }
5f39d397 3724
39279cc3 3725 drop_on_err = 1;
33268eaf
JB		 3726
3727 err = btrfs_init_acl(inode, dir);
3728 if (err)
3729 goto out_fail;
3730
39279cc3
CM		 3731 inode->i_op = &btrfs_dir_inode_operations;
3732 inode->i_fop = &btrfs_dir_file_operations;
3733 btrfs_set_trans_block_group(trans, inode);
3734
dbe674a9 3735 btrfs_i_size_write(inode, 0);
39279cc3
CM		 3736 err = btrfs_update_inode(trans, root, inode);
3737 if (err)
3738 goto out_fail;
5f39d397 3739
e02119d5
CM		 3740 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3741 inode, dentry->d_name.name,
3742 dentry->d_name.len, 0, index);
39279cc3
CM		 3743 if (err)
3744 goto out_fail;
5f39d397 3745
39279cc3
CM		 3746 d_instantiate(dentry, inode);
3747 drop_on_err = 0;
3748 dir->i_sb->s_dirt = 1;
3749 btrfs_update_inode_block_group(trans, inode);
3750 btrfs_update_inode_block_group(trans, dir);
3751
3752out_fail:
d3c2fdcf 3753 nr = trans->blocks_used;
ab78c84d 3754 btrfs_end_transaction_throttle(trans, root);
5f39d397 3755
39279cc3 3756out_unlock:
39279cc3
CM		 3757 if (drop_on_err)
3758 iput(inode);
d3c2fdcf 3759 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3760 return err;
3761}
3762
d352ac68
CM		 3763/* helper for btfs_get_extent. Given an existing extent in the tree,
3764 * and an extent that you want to insert, deal with overlap and insert
3765 * the new extent into the tree.
3766 */
3b951516
CM		 3767static int merge_extent_mapping(struct extent_map_tree *em_tree,
3768 struct extent_map *existing,
e6dcd2dc
CM		 3769 struct extent_map *em,
3770 u64 map_start, u64 map_len)
3b951516
CM		 3771{
3772 u64 start_diff;
3b951516 3773
e6dcd2dc
CM		 3774 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3775 start_diff = map_start - em->start;
3776 em->start = map_start;
3777 em->len = map_len;
c8b97818
CM		 3778 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3779 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 3780 em->block_start += start_diff;
c8b97818
CM		 3781 em->block_len -= start_diff;
3782 }
e6dcd2dc 3783 return add_extent_mapping(em_tree, em);
3b951516
CM		 3784}
3785
c8b97818
CM		 3786static noinline int uncompress_inline(struct btrfs_path *path,
3787 struct inode *inode, struct page *page,
3788 size_t pg_offset, u64 extent_offset,
3789 struct btrfs_file_extent_item *item)
3790{
3791 int ret;
3792 struct extent_buffer *leaf = path->nodes[0];
3793 char *tmp;
3794 size_t max_size;
3795 unsigned long inline_size;
3796 unsigned long ptr;
3797
3798 WARN_ON(pg_offset != 0);
3799 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3800 inline_size = btrfs_file_extent_inline_item_len(leaf,
3801 btrfs_item_nr(leaf, path->slots[0]));
3802 tmp = kmalloc(inline_size, GFP_NOFS);
3803 ptr = btrfs_file_extent_inline_start(item);
3804
3805 read_extent_buffer(leaf, tmp, ptr, inline_size);
3806
5b050f04 3807 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
c8b97818
CM		 3808 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3809 inline_size, max_size);
3810 if (ret) {
3811 char *kaddr = kmap_atomic(page, KM_USER0);
3812 unsigned long copy_size = min_t(u64,
3813 PAGE_CACHE_SIZE - pg_offset,
3814 max_size - extent_offset);
3815 memset(kaddr + pg_offset, 0, copy_size);
3816 kunmap_atomic(kaddr, KM_USER0);
3817 }
3818 kfree(tmp);
3819 return 0;
3820}
3821
d352ac68
CM		 3822/*
3823 * a bit scary, this does extent mapping from logical file offset to the disk.
3824 * the ugly parts come from merging extents from the disk with the
3825 * in-ram representation. This gets more complex because of the data=ordered code,
3826 * where the in-ram extents might be locked pending data=ordered completion.
3827 *
3828 * This also copies inline extents directly into the page.
3829 */
a52d9a80 3830struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 3831 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 3832 int create)
3833{
3834 int ret;
3835 int err = 0;
db94535d 3836 u64 bytenr;
a52d9a80
CM		 3837 u64 extent_start = 0;
3838 u64 extent_end = 0;
3839 u64 objectid = inode->i_ino;
3840 u32 found_type;
f421950f 3841 struct btrfs_path *path = NULL;
a52d9a80
CM		 3842 struct btrfs_root *root = BTRFS_I(inode)->root;
3843 struct btrfs_file_extent_item *item;
5f39d397
CM		 3844 struct extent_buffer *leaf;
3845 struct btrfs_key found_key;
a52d9a80
CM		 3846 struct extent_map *em = NULL;
3847 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 3848 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 3849 struct btrfs_trans_handle *trans = NULL;
c8b97818 3850 int compressed;
a52d9a80 3851
a52d9a80 3852again:
d1310b2e
CM		 3853 spin_lock(&em_tree->lock);
3854 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 3855 if (em)
3856 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 3857 spin_unlock(&em_tree->lock);
3858
a52d9a80 3859 if (em) {
e1c4b745
CM		 3860 if (em->start > start || em->start + em->len <= start)
3861 free_extent_map(em);
3862 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 3863 free_extent_map(em);
3864 else
3865 goto out;
a52d9a80 3866 }
d1310b2e 3867 em = alloc_extent_map(GFP_NOFS);
a52d9a80 3868 if (!em) {
d1310b2e
CM		 3869 err = -ENOMEM;
3870 goto out;
a52d9a80 3871 }
e6dcd2dc 3872 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 3873 em->start = EXTENT_MAP_HOLE;
445a6944 3874 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 3875 em->len = (u64)-1;
c8b97818 3876 em->block_len = (u64)-1;
f421950f
CM		 3877
3878 if (!path) {
3879 path = btrfs_alloc_path();
3880 BUG_ON(!path);
3881 }
3882
179e29e4
CM		 3883 ret = btrfs_lookup_file_extent(trans, root, path,
3884 objectid, start, trans != NULL);
a52d9a80
CM		 3885 if (ret < 0) {
3886 err = ret;
3887 goto out;
3888 }
3889
3890 if (ret != 0) {
3891 if (path->slots[0] == 0)
3892 goto not_found;
3893 path->slots[0]--;
3894 }
3895
5f39d397
CM		 3896 leaf = path->nodes[0];
3897 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 3898 struct btrfs_file_extent_item);
a52d9a80 3899 /* are we inside the extent that was found? */
5f39d397
CM		 3900 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3901 found_type = btrfs_key_type(&found_key);
3902 if (found_key.objectid != objectid ||
a52d9a80
CM		 3903 found_type != BTRFS_EXTENT_DATA_KEY) {
3904 goto not_found;
3905 }
3906
5f39d397
CM		 3907 found_type = btrfs_file_extent_type(leaf, item);
3908 extent_start = found_key.offset;
c8b97818 3909 compressed = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 3910 if (found_type == BTRFS_FILE_EXTENT_REG ||
3911 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 3912 extent_end = extent_start +
db94535d 3913 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 3914 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3915 size_t size;
3916 size = btrfs_file_extent_inline_len(leaf, item);
3917 extent_end = (extent_start + size + root->sectorsize - 1) &
3918 ~((u64)root->sectorsize - 1);
3919 }
3920
3921 if (start >= extent_end) {
3922 path->slots[0]++;
3923 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3924 ret = btrfs_next_leaf(root, path);
3925 if (ret < 0) {
3926 err = ret;
3927 goto out;
a52d9a80 3928 }
9036c102
YZ		 3929 if (ret > 0)
3930 goto not_found;
3931 leaf = path->nodes[0];
a52d9a80 3932 }
9036c102
YZ		 3933 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3934 if (found_key.objectid != objectid ||
3935 found_key.type != BTRFS_EXTENT_DATA_KEY)
3936 goto not_found;
3937 if (start + len <= found_key.offset)
3938 goto not_found;
3939 em->start = start;
3940 em->len = found_key.offset - start;
3941 goto not_found_em;
3942 }
3943
d899e052
YZ		 3944 if (found_type == BTRFS_FILE_EXTENT_REG ||
3945 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 3946 em->start = extent_start;
3947 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 3948 em->orig_start = extent_start -
3949 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 3950 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3951 if (bytenr == 0) {
5f39d397 3952 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 3953 goto insert;
3954 }
c8b97818
CM		 3955 if (compressed) {
3956 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3957 em->block_start = bytenr;
3958 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3959 item);
3960 } else {
3961 bytenr += btrfs_file_extent_offset(leaf, item);
3962 em->block_start = bytenr;
3963 em->block_len = em->len;
d899e052
YZ		 3964 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3965 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 3966 }
a52d9a80
CM		 3967 goto insert;
3968 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3969 unsigned long ptr;
a52d9a80 3970 char *map;
3326d1b0
CM		 3971 size_t size;
3972 size_t extent_offset;
3973 size_t copy_size;
a52d9a80 3974
689f9346 3975 em->block_start = EXTENT_MAP_INLINE;
c8b97818 3976 if (!page || create) {
689f9346 3977 em->start = extent_start;
9036c102 3978 em->len = extent_end - extent_start;
689f9346
Y		 3979 goto out;
3980 }
5f39d397 3981
9036c102
YZ		 3982 size = btrfs_file_extent_inline_len(leaf, item);
3983 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 3984 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 3985 size - extent_offset);
3326d1b0 3986 em->start = extent_start + extent_offset;
70dec807
CM		 3987 em->len = (copy_size + root->sectorsize - 1) &
3988 ~((u64)root->sectorsize - 1);
ff5b7ee3 3989 em->orig_start = EXTENT_MAP_INLINE;
c8b97818
CM		 3990 if (compressed)
3991 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
689f9346 3992 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 3993 if (create == 0 && !PageUptodate(page)) {
c8b97818
CM		 3994 if (btrfs_file_extent_compression(leaf, item) ==
3995 BTRFS_COMPRESS_ZLIB) {
3996 ret = uncompress_inline(path, inode, page,
3997 pg_offset,
3998 extent_offset, item);
3999 BUG_ON(ret);
4000 } else {
4001 map = kmap(page);
4002 read_extent_buffer(leaf, map + pg_offset, ptr,
4003 copy_size);
4004 kunmap(page);
4005 }
179e29e4
CM		 4006 flush_dcache_page(page);
4007 } else if (create && PageUptodate(page)) {
4008 if (!trans) {
4009 kunmap(page);
4010 free_extent_map(em);
4011 em = NULL;
4012 btrfs_release_path(root, path);
f9295749 4013 trans = btrfs_join_transaction(root, 1);
179e29e4
CM		 4014 goto again;
4015 }
c8b97818 4016 map = kmap(page);
70dec807 4017 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 4018 copy_size);
c8b97818 4019 kunmap(page);
179e29e4 4020 btrfs_mark_buffer_dirty(leaf);
a52d9a80 4021 }
d1310b2e
CM		 4022 set_extent_uptodate(io_tree, em->start,
4023 extent_map_end(em) - 1, GFP_NOFS);
a52d9a80
CM		 4024 goto insert;
4025 } else {
4026 printk("unkknown found_type %d\n", found_type);
4027 WARN_ON(1);
4028 }
4029not_found:
4030 em->start = start;
d1310b2e 4031 em->len = len;
a52d9a80 4032not_found_em:
5f39d397 4033 em->block_start = EXTENT_MAP_HOLE;
9036c102 4034 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80
CM		 4035insert:
4036 btrfs_release_path(root, path);
d1310b2e
CM		 4037 if (em->start > start || extent_map_end(em) <= start) {
4038 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
a52d9a80
CM		 4039 err = -EIO;
4040 goto out;
4041 }
d1310b2e
CM		 4042
4043 err = 0;
4044 spin_lock(&em_tree->lock);
a52d9a80 4045 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 4046 /* it is possible that someone inserted the extent into the tree
4047 * while we had the lock dropped. It is also possible that
4048 * an overlapping map exists in the tree
4049 */
a52d9a80 4050 if (ret == -EEXIST) {
3b951516 4051 struct extent_map *existing;
e6dcd2dc
CM		 4052
4053 ret = 0;
4054
3b951516 4055 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 4056 if (existing && (existing->start > start ||
4057 existing->start + existing->len <= start)) {
4058 free_extent_map(existing);
4059 existing = NULL;
4060 }
3b951516
CM		 4061 if (!existing) {
4062 existing = lookup_extent_mapping(em_tree, em->start,
4063 em->len);
4064 if (existing) {
4065 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 4066 em, start,
4067 root->sectorsize);
3b951516
CM		 4068 free_extent_map(existing);
4069 if (err) {
4070 free_extent_map(em);
4071 em = NULL;
4072 }
4073 } else {
4074 err = -EIO;
4075 printk("failing to insert %Lu %Lu\n",
4076 start, len);
4077 free_extent_map(em);
4078 em = NULL;
4079 }
4080 } else {
4081 free_extent_map(em);
4082 em = existing;
e6dcd2dc 4083 err = 0;
a52d9a80 4084 }
a52d9a80 4085 }
d1310b2e 4086 spin_unlock(&em_tree->lock);
a52d9a80 4087out:
f421950f
CM		 4088 if (path)
4089 btrfs_free_path(path);
a52d9a80
CM		 4090 if (trans) {
4091 ret = btrfs_end_transaction(trans, root);
e6dcd2dc 4092 if (!err) {
a52d9a80 4093 err = ret;
e6dcd2dc 4094 }
a52d9a80 4095 }
a52d9a80
CM		 4096 if (err) {
4097 free_extent_map(em);
4098 WARN_ON(1);
4099 return ERR_PTR(err);
4100 }
4101 return em;
4102}
4103
16432985
CM		 4104static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4105 const struct iovec *iov, loff_t offset,
4106 unsigned long nr_segs)
4107{
e1c4b745 4108 return -EINVAL;
16432985
CM		 4109}
4110
d396c6f5 4111static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
39279cc3 4112{
d396c6f5 4113 return extent_bmap(mapping, iblock, btrfs_get_extent);
39279cc3
CM		 4114}
4115
a52d9a80 4116int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 4117{
d1310b2e
CM		 4118 struct extent_io_tree *tree;
4119 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4120 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 4121}
1832a6d5 4122
a52d9a80 4123static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 4124{
d1310b2e 4125 struct extent_io_tree *tree;
b888db2b
CM		 4126
4127
4128 if (current->flags & PF_MEMALLOC) {
4129 redirty_page_for_writepage(wbc, page);
4130 unlock_page(page);
4131 return 0;
4132 }
d1310b2e 4133 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4134 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 4135}
4136
f421950f
CM		 4137int btrfs_writepages(struct address_space *mapping,
4138 struct writeback_control *wbc)
b293f02e 4139{
d1310b2e 4140 struct extent_io_tree *tree;
771ed689 4141
d1310b2e 4142 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 4143 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4144}
4145
3ab2fb5a
CM		 4146static int
4147btrfs_readpages(struct file *file, struct address_space *mapping,
4148 struct list_head *pages, unsigned nr_pages)
4149{
d1310b2e
CM		 4150 struct extent_io_tree *tree;
4151 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 4152 return extent_readpages(tree, mapping, pages, nr_pages,
4153 btrfs_get_extent);
4154}
e6dcd2dc 4155static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 4156{
d1310b2e
CM		 4157 struct extent_io_tree *tree;
4158 struct extent_map_tree *map;
a52d9a80 4159 int ret;
8c2383c3 4160
d1310b2e
CM		 4161 tree = &BTRFS_I(page->mapping->host)->io_tree;
4162 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 4163 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 4164 if (ret == 1) {
4165 ClearPagePrivate(page);
4166 set_page_private(page, 0);
4167 page_cache_release(page);
39279cc3 4168 }
a52d9a80 4169 return ret;
39279cc3
CM		 4170}
4171
e6dcd2dc
CM		 4172static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4173{
98509cfc
CM		 4174 if (PageWriteback(page) || PageDirty(page))
4175 return 0;
e6dcd2dc
CM		 4176 return __btrfs_releasepage(page, gfp_flags);
4177}
4178
a52d9a80 4179static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 4180{
d1310b2e 4181 struct extent_io_tree *tree;
e6dcd2dc
CM		 4182 struct btrfs_ordered_extent *ordered;
4183 u64 page_start = page_offset(page);
4184 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 4185
e6dcd2dc 4186 wait_on_page_writeback(page);
d1310b2e 4187 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 4188 if (offset) {
4189 btrfs_releasepage(page, GFP_NOFS);
4190 return;
4191 }
4192
4193 lock_extent(tree, page_start, page_end, GFP_NOFS);
4194 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4195 page_offset(page));
4196 if (ordered) {
eb84ae03
CM		 4197 /*
4198 * IO on this page will never be started, so we need
4199 * to account for any ordered extents now
4200 */
e6dcd2dc
CM		 4201 clear_extent_bit(tree, page_start, page_end,
4202 EXTENT_DIRTY | EXTENT_DELALLOC |
4203 EXTENT_LOCKED, 1, 0, GFP_NOFS);
211f90e6
CM		 4204 btrfs_finish_ordered_io(page->mapping->host,
4205 page_start, page_end);
e6dcd2dc
CM		 4206 btrfs_put_ordered_extent(ordered);
4207 lock_extent(tree, page_start, page_end, GFP_NOFS);
4208 }
4209 clear_extent_bit(tree, page_start, page_end,
4210 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4211 EXTENT_ORDERED,
4212 1, 1, GFP_NOFS);
4213 __btrfs_releasepage(page, GFP_NOFS);
4214
4a096752 4215 ClearPageChecked(page);
9ad6b7bc 4216 if (PagePrivate(page)) {
9ad6b7bc
CM		 4217 ClearPagePrivate(page);
4218 set_page_private(page, 0);
4219 page_cache_release(page);
4220 }
39279cc3
CM		 4221}
4222
9ebefb18
CM		 4223/*
4224 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4225 * called from a page fault handler when a page is first dirtied. Hence we must
4226 * be careful to check for EOF conditions here. We set the page up correctly
4227 * for a written page which means we get ENOSPC checking when writing into
4228 * holes and correct delalloc and unwritten extent mapping on filesystems that
4229 * support these features.
4230 *
4231 * We are not allowed to take the i_mutex here so we have to play games to
4232 * protect against truncate races as the page could now be beyond EOF. Because
4233 * vmtruncate() writes the inode size before removing pages, once we have the
4234 * page lock we can determine safely if the page is beyond EOF. If it is not
4235 * beyond EOF, then the page is guaranteed safe against truncation until we
4236 * unlock the page.
4237 */
4238int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4239{
6da6abae 4240 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 4241 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4242 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4243 struct btrfs_ordered_extent *ordered;
4244 char *kaddr;
4245 unsigned long zero_start;
9ebefb18 4246 loff_t size;
1832a6d5 4247 int ret;
a52d9a80 4248 u64 page_start;
e6dcd2dc 4249 u64 page_end;
9ebefb18 4250
1832a6d5 4251 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
1832a6d5
CM		 4252 if (ret)
4253 goto out;
4254
4255 ret = -EINVAL;
e6dcd2dc 4256again:
9ebefb18 4257 lock_page(page);
9ebefb18 4258 size = i_size_read(inode);
e6dcd2dc
CM		 4259 page_start = page_offset(page);
4260 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 4261
9ebefb18 4262 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 4263 (page_start >= size)) {
9ebefb18
CM		 4264 /* page got truncated out from underneath us */
4265 goto out_unlock;
4266 }
e6dcd2dc
CM		 4267 wait_on_page_writeback(page);
4268
4269 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4270 set_page_extent_mapped(page);
4271
eb84ae03
CM		 4272 /*
4273 * we can't set the delalloc bits if there are pending ordered
4274 * extents. Drop our locks and wait for them to finish
4275 */
e6dcd2dc
CM		 4276 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4277 if (ordered) {
4278 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4279 unlock_page(page);
eb84ae03 4280 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4281 btrfs_put_ordered_extent(ordered);
4282 goto again;
4283 }
4284
ea8c2819 4285 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc 4286 ret = 0;
9ebefb18
CM		 4287
4288 /* page is wholly or partially inside EOF */
a52d9a80 4289 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 4290 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 4291 else
e6dcd2dc 4292 zero_start = PAGE_CACHE_SIZE;
9ebefb18 4293
e6dcd2dc
CM		 4294 if (zero_start != PAGE_CACHE_SIZE) {
4295 kaddr = kmap(page);
4296 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4297 flush_dcache_page(page);
4298 kunmap(page);
4299 }
247e743c 4300 ClearPageChecked(page);
e6dcd2dc
CM		 4301 set_page_dirty(page);
4302 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
9ebefb18
CM		 4303
4304out_unlock:
4305 unlock_page(page);
1832a6d5 4306out:
9ebefb18
CM		 4307 return ret;
4308}
4309
39279cc3
CM		 4310static void btrfs_truncate(struct inode *inode)
4311{
4312 struct btrfs_root *root = BTRFS_I(inode)->root;
4313 int ret;
4314 struct btrfs_trans_handle *trans;
d3c2fdcf 4315 unsigned long nr;
dbe674a9 4316 u64 mask = root->sectorsize - 1;
39279cc3
CM		 4317
4318 if (!S_ISREG(inode->i_mode))
4319 return;
4320 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4321 return;
4322
4323 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4a096752 4324 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 4325
39279cc3
CM		 4326 trans = btrfs_start_transaction(root, 1);
4327 btrfs_set_trans_block_group(trans, inode);
dbe674a9 4328 btrfs_i_size_write(inode, inode->i_size);
39279cc3 4329
7b128766
JB		 4330 ret = btrfs_orphan_add(trans, inode);
4331 if (ret)
4332 goto out;
39279cc3 4333 /* FIXME, add redo link to tree so we don't leak on crash */
e02119d5 4334 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
85e21bac 4335 BTRFS_EXTENT_DATA_KEY);
39279cc3 4336 btrfs_update_inode(trans, root, inode);
5f39d397 4337
7b128766
JB		 4338 ret = btrfs_orphan_del(trans, inode);
4339 BUG_ON(ret);
4340
4341out:
4342 nr = trans->blocks_used;
89ce8a63 4343 ret = btrfs_end_transaction_throttle(trans, root);
39279cc3 4344 BUG_ON(ret);
d3c2fdcf 4345 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4346}
4347
3b96362c
SW		 4348/*
4349 * Invalidate a single dcache entry at the root of the filesystem.
4350 * Needed after creation of snapshot or subvolume.
4351 */
4352void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
4353 int namelen)
4354{
4355 struct dentry *alias, *entry;
4356 struct qstr qstr;
4357
4358 alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
4359 if (alias) {
4360 qstr.name = name;
4361 qstr.len = namelen;
4362 /* change me if btrfs ever gets a d_hash operation */
4363 qstr.hash = full_name_hash(qstr.name, qstr.len);
4364 entry = d_lookup(alias, &qstr);
4365 dput(alias);
4366 if (entry) {
4367 d_invalidate(entry);
4368 dput(entry);
4369 }
4370 }
4371}
4372
d352ac68
CM		 4373/*
4374 * create a new subvolume directory/inode (helper for the ioctl).
4375 */
cb8e7090 4376int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
f46b5a66
CH		 4377 struct btrfs_trans_handle *trans, u64 new_dirid,
4378 struct btrfs_block_group_cache *block_group)
39279cc3 4379{
39279cc3 4380 struct inode *inode;
cb8e7090 4381 int error;
00e4e6b3 4382 u64 index = 0;
39279cc3 4383
aec7477b 4384 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
00e4e6b3 4385 new_dirid, block_group, S_IFDIR | 0700, &index);
54aa1f4d 4386 if (IS_ERR(inode))
f46b5a66 4387 return PTR_ERR(inode);
39279cc3
CM		 4388 inode->i_op = &btrfs_dir_inode_operations;
4389 inode->i_fop = &btrfs_dir_file_operations;
34088780 4390 new_root->inode = inode;
39279cc3 4391
39279cc3 4392 inode->i_nlink = 1;
dbe674a9 4393 btrfs_i_size_write(inode, 0);
3b96362c 4394
cb8e7090
CH		 4395 error = btrfs_update_inode(trans, new_root, inode);
4396 if (error)
4397 return error;
4398
d899e052 4399 atomic_inc(&inode->i_count);
cb8e7090
CH		 4400 d_instantiate(dentry, inode);
4401 return 0;
39279cc3
CM		 4402}
4403
d352ac68
CM		 4404/* helper function for file defrag and space balancing. This
4405 * forces readahead on a given range of bytes in an inode
4406 */
edbd8d4e 4407unsigned long btrfs_force_ra(struct address_space *mapping,
86479a04
CM		 4408 struct file_ra_state *ra, struct file *file,
4409 pgoff_t offset, pgoff_t last_index)
4410{
8e7bf94f 4411 pgoff_t req_size = last_index - offset + 1;
86479a04 4412
86479a04
CM		 4413 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4414 return offset + req_size;
86479a04
CM		 4415}
4416
39279cc3
CM		 4417struct inode *btrfs_alloc_inode(struct super_block *sb)
4418{
4419 struct btrfs_inode *ei;
4420
4421 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4422 if (!ei)
4423 return NULL;
15ee9bc7 4424 ei->last_trans = 0;
e02119d5 4425 ei->logged_trans = 0;
e6dcd2dc 4426 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
33268eaf
JB		 4427 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4428 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
7b128766 4429 INIT_LIST_HEAD(&ei->i_orphan);
39279cc3
CM		 4430 return &ei->vfs_inode;
4431}
4432
4433void btrfs_destroy_inode(struct inode *inode)
4434{
e6dcd2dc 4435 struct btrfs_ordered_extent *ordered;
39279cc3
CM		 4436 WARN_ON(!list_empty(&inode->i_dentry));
4437 WARN_ON(inode->i_data.nrpages);
4438
33268eaf
JB		 4439 if (BTRFS_I(inode)->i_acl &&
4440 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4441 posix_acl_release(BTRFS_I(inode)->i_acl);
4442 if (BTRFS_I(inode)->i_default_acl &&
4443 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4444 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4445
bcc63abb 4446 spin_lock(&BTRFS_I(inode)->root->list_lock);
7b128766
JB		 4447 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4448 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4449 " list\n", inode->i_ino);
4450 dump_stack();
4451 }
bcc63abb 4452 spin_unlock(&BTRFS_I(inode)->root->list_lock);
7b128766 4453
e6dcd2dc
CM		 4454 while(1) {
4455 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4456 if (!ordered)
4457 break;
4458 else {
4459 printk("found ordered extent %Lu %Lu\n",
4460 ordered->file_offset, ordered->len);
4461 btrfs_remove_ordered_extent(inode, ordered);
4462 btrfs_put_ordered_extent(ordered);
4463 btrfs_put_ordered_extent(ordered);
4464 }
4465 }
5b21f2ed 4466 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
39279cc3
CM		 4467 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4468}
4469
0ee0fda0 4470static void init_once(void *foo)
39279cc3
CM		 4471{
4472 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4473
4474 inode_init_once(&ei->vfs_inode);
4475}
4476
4477void btrfs_destroy_cachep(void)
4478{
4479 if (btrfs_inode_cachep)
4480 kmem_cache_destroy(btrfs_inode_cachep);
4481 if (btrfs_trans_handle_cachep)
4482 kmem_cache_destroy(btrfs_trans_handle_cachep);
4483 if (btrfs_transaction_cachep)
4484 kmem_cache_destroy(btrfs_transaction_cachep);
4485 if (btrfs_bit_radix_cachep)
4486 kmem_cache_destroy(btrfs_bit_radix_cachep);
4487 if (btrfs_path_cachep)
4488 kmem_cache_destroy(btrfs_path_cachep);
4489}
4490
86479a04 4491struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
92fee66d 4492 unsigned long extra_flags,
2b1f55b0 4493 void (*ctor)(void *))
92fee66d
CM		 4494{
4495 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2b1f55b0 4496 SLAB_MEM_SPREAD | extra_flags), ctor);
92fee66d
CM		 4497}
4498
39279cc3
CM		 4499int btrfs_init_cachep(void)
4500{
86479a04 4501 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
92fee66d
CM		 4502 sizeof(struct btrfs_inode),
4503 0, init_once);
39279cc3
CM		 4504 if (!btrfs_inode_cachep)
4505 goto fail;
86479a04
CM		 4506 btrfs_trans_handle_cachep =
4507 btrfs_cache_create("btrfs_trans_handle_cache",
4508 sizeof(struct btrfs_trans_handle),
4509 0, NULL);
39279cc3
CM		 4510 if (!btrfs_trans_handle_cachep)
4511 goto fail;
86479a04 4512 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
39279cc3 4513 sizeof(struct btrfs_transaction),
92fee66d 4514 0, NULL);
39279cc3
CM		 4515 if (!btrfs_transaction_cachep)
4516 goto fail;
86479a04 4517 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
23223584 4518 sizeof(struct btrfs_path),
92fee66d 4519 0, NULL);
39279cc3
CM		 4520 if (!btrfs_path_cachep)
4521 goto fail;
86479a04 4522 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
92fee66d 4523 SLAB_DESTROY_BY_RCU, NULL);
39279cc3
CM		 4524 if (!btrfs_bit_radix_cachep)
4525 goto fail;
4526 return 0;
4527fail:
4528 btrfs_destroy_cachep();
4529 return -ENOMEM;
4530}
4531
4532static int btrfs_getattr(struct vfsmount *mnt,
4533 struct dentry *dentry, struct kstat *stat)
4534{
4535 struct inode *inode = dentry->d_inode;
4536 generic_fillattr(inode, stat);
d6667462 4537 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 4538 stat->blocks = (inode_get_bytes(inode) +
4539 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 4540 return 0;
4541}
4542
4543static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4544 struct inode * new_dir,struct dentry *new_dentry)
4545{
4546 struct btrfs_trans_handle *trans;
4547 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4548 struct inode *new_inode = new_dentry->d_inode;
4549 struct inode *old_inode = old_dentry->d_inode;
4550 struct timespec ctime = CURRENT_TIME;
00e4e6b3 4551 u64 index = 0;
39279cc3
CM		 4552 int ret;
4553
4554 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4555 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4556 return -ENOTEMPTY;
4557 }
5f39d397 4558
1832a6d5
CM		 4559 ret = btrfs_check_free_space(root, 1, 0);
4560 if (ret)
4561 goto out_unlock;
4562
39279cc3 4563 trans = btrfs_start_transaction(root, 1);
5f39d397 4564
39279cc3 4565 btrfs_set_trans_block_group(trans, new_dir);
39279cc3 4566
e02119d5 4567 btrfs_inc_nlink(old_dentry->d_inode);
39279cc3
CM		 4568 old_dir->i_ctime = old_dir->i_mtime = ctime;
4569 new_dir->i_ctime = new_dir->i_mtime = ctime;
4570 old_inode->i_ctime = ctime;
5f39d397 4571
e02119d5
CM		 4572 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4573 old_dentry->d_name.name,
4574 old_dentry->d_name.len);
39279cc3
CM		 4575 if (ret)
4576 goto out_fail;
4577
4578 if (new_inode) {
4579 new_inode->i_ctime = CURRENT_TIME;
e02119d5
CM		 4580 ret = btrfs_unlink_inode(trans, root, new_dir,
4581 new_dentry->d_inode,
4582 new_dentry->d_name.name,
4583 new_dentry->d_name.len);
39279cc3
CM		 4584 if (ret)
4585 goto out_fail;
7b128766 4586 if (new_inode->i_nlink == 0) {
e02119d5 4587 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7b128766
JB		 4588 if (ret)
4589 goto out_fail;
4590 }
e02119d5 4591
39279cc3 4592 }
00e4e6b3 4593 ret = btrfs_set_inode_index(new_dir, old_inode, &index);
aec7477b
JB		 4594 if (ret)
4595 goto out_fail;
4596
e02119d5
CM		 4597 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4598 old_inode, new_dentry->d_name.name,
4599 new_dentry->d_name.len, 1, index);
39279cc3
CM		 4600 if (ret)
4601 goto out_fail;
4602
4603out_fail:
ab78c84d 4604 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4605out_unlock:
39279cc3
CM		 4606 return ret;
4607}
4608
d352ac68
CM		 4609/*
4610 * some fairly slow code that needs optimization. This walks the list
4611 * of all the inodes with pending delalloc and forces them to disk.
4612 */
ea8c2819
CM		 4613int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4614{
4615 struct list_head *head = &root->fs_info->delalloc_inodes;
4616 struct btrfs_inode *binode;
5b21f2ed 4617 struct inode *inode;
ea8c2819
CM		 4618 unsigned long flags;
4619
c146afad
YZ		 4620 if (root->fs_info->sb->s_flags & MS_RDONLY)
4621 return -EROFS;
4622
ea8c2819
CM		 4623 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4624 while(!list_empty(head)) {
4625 binode = list_entry(head->next, struct btrfs_inode,
4626 delalloc_inodes);
5b21f2ed
ZY		 4627 inode = igrab(&binode->vfs_inode);
4628 if (!inode)
4629 list_del_init(&binode->delalloc_inodes);
ea8c2819 4630 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
5b21f2ed 4631 if (inode) {
8c8bee1d 4632 filemap_flush(inode->i_mapping);
5b21f2ed
ZY		 4633 iput(inode);
4634 }
4635 cond_resched();
ea8c2819
CM		 4636 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4637 }
4638 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
8c8bee1d
CM		 4639
4640 /* the filemap_flush will queue IO into the worker threads, but
4641 * we have to make sure the IO is actually started and that
4642 * ordered extents get created before we return
4643 */
4644 atomic_inc(&root->fs_info->async_submit_draining);
771ed689
CM		 4645 while(atomic_read(&root->fs_info->nr_async_submits) ||
4646 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 4647 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 4648 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4649 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 4650 }
4651 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 4652 return 0;
4653}
4654
39279cc3
CM		 4655static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4656 const char *symname)
4657{
4658 struct btrfs_trans_handle *trans;
4659 struct btrfs_root *root = BTRFS_I(dir)->root;
4660 struct btrfs_path *path;
4661 struct btrfs_key key;
1832a6d5 4662 struct inode *inode = NULL;
39279cc3
CM		 4663 int err;
4664 int drop_inode = 0;
4665 u64 objectid;
00e4e6b3 4666 u64 index = 0 ;
39279cc3
CM		 4667 int name_len;
4668 int datasize;
5f39d397 4669 unsigned long ptr;
39279cc3 4670 struct btrfs_file_extent_item *ei;
5f39d397 4671 struct extent_buffer *leaf;
1832a6d5 4672 unsigned long nr = 0;
39279cc3
CM		 4673
4674 name_len = strlen(symname) + 1;
4675 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4676 return -ENAMETOOLONG;
1832a6d5 4677
1832a6d5
CM		 4678 err = btrfs_check_free_space(root, 1, 0);
4679 if (err)
4680 goto out_fail;
4681
39279cc3
CM		 4682 trans = btrfs_start_transaction(root, 1);
4683 btrfs_set_trans_block_group(trans, dir);
4684
4685 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4686 if (err) {
4687 err = -ENOSPC;
4688 goto out_unlock;
4689 }
4690
aec7477b 4691 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 4692 dentry->d_name.len,
4693 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 4694 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4695 &index);
39279cc3
CM		 4696 err = PTR_ERR(inode);
4697 if (IS_ERR(inode))
4698 goto out_unlock;
4699
33268eaf
JB		 4700 err = btrfs_init_acl(inode, dir);
4701 if (err) {
4702 drop_inode = 1;
4703 goto out_unlock;
4704 }
4705
39279cc3 4706 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 4707 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 4708 if (err)
4709 drop_inode = 1;
4710 else {
4711 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4712 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4713 inode->i_fop = &btrfs_file_operations;
4714 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4715 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 4716 }
4717 dir->i_sb->s_dirt = 1;
4718 btrfs_update_inode_block_group(trans, inode);
4719 btrfs_update_inode_block_group(trans, dir);
4720 if (drop_inode)
4721 goto out_unlock;
4722
4723 path = btrfs_alloc_path();
4724 BUG_ON(!path);
4725 key.objectid = inode->i_ino;
4726 key.offset = 0;
39279cc3
CM		 4727 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4728 datasize = btrfs_file_extent_calc_inline_size(name_len);
4729 err = btrfs_insert_empty_item(trans, root, path, &key,
4730 datasize);
54aa1f4d
CM		 4731 if (err) {
4732 drop_inode = 1;
4733 goto out_unlock;
4734 }
5f39d397
CM		 4735 leaf = path->nodes[0];
4736 ei = btrfs_item_ptr(leaf, path->slots[0],
4737 struct btrfs_file_extent_item);
4738 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4739 btrfs_set_file_extent_type(leaf, ei,
39279cc3 4740 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 4741 btrfs_set_file_extent_encryption(leaf, ei, 0);
4742 btrfs_set_file_extent_compression(leaf, ei, 0);
4743 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4744 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4745
39279cc3 4746 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 4747 write_extent_buffer(leaf, symname, ptr, name_len);
4748 btrfs_mark_buffer_dirty(leaf);
39279cc3 4749 btrfs_free_path(path);
5f39d397 4750
39279cc3
CM		 4751 inode->i_op = &btrfs_symlink_inode_operations;
4752 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 4753 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 4754 inode_set_bytes(inode, name_len);
dbe674a9 4755 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 4756 err = btrfs_update_inode(trans, root, inode);
4757 if (err)
4758 drop_inode = 1;
39279cc3
CM		 4759
4760out_unlock:
d3c2fdcf 4761 nr = trans->blocks_used;
ab78c84d 4762 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4763out_fail:
39279cc3
CM		 4764 if (drop_inode) {
4765 inode_dec_link_count(inode);
4766 iput(inode);
4767 }
d3c2fdcf 4768 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4769 return err;
4770}
16432985 4771
d899e052
YZ		 4772static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4773 u64 alloc_hint, int mode)
4774{
4775 struct btrfs_trans_handle *trans;
4776 struct btrfs_root *root = BTRFS_I(inode)->root;
4777 struct btrfs_key ins;
4778 u64 alloc_size;
4779 u64 cur_offset = start;
4780 u64 num_bytes = end - start;
4781 int ret = 0;
4782
4783 trans = btrfs_join_transaction(root, 1);
4784 BUG_ON(!trans);
4785 btrfs_set_trans_block_group(trans, inode);
4786
4787 while (num_bytes > 0) {
4788 alloc_size = min(num_bytes, root->fs_info->max_extent);
4789 ret = btrfs_reserve_extent(trans, root, alloc_size,
4790 root->sectorsize, 0, alloc_hint,
4791 (u64)-1, &ins, 1);
4792 if (ret) {
4793 WARN_ON(1);
4794 goto out;
4795 }
4796 ret = insert_reserved_file_extent(trans, inode,
4797 cur_offset, ins.objectid,
4798 ins.offset, ins.offset,
4799 ins.offset, 0, 0, 0,
4800 BTRFS_FILE_EXTENT_PREALLOC);
4801 BUG_ON(ret);
4802 num_bytes -= ins.offset;
4803 cur_offset += ins.offset;
4804 alloc_hint = ins.objectid + ins.offset;
4805 }
4806out:
4807 if (cur_offset > start) {
4808 inode->i_ctime = CURRENT_TIME;
4809 btrfs_set_flag(inode, PREALLOC);
4810 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4811 cur_offset > i_size_read(inode))
4812 btrfs_i_size_write(inode, cur_offset);
4813 ret = btrfs_update_inode(trans, root, inode);
4814 BUG_ON(ret);
4815 }
4816
4817 btrfs_end_transaction(trans, root);
4818 return ret;
4819}
4820
4821static long btrfs_fallocate(struct inode *inode, int mode,
4822 loff_t offset, loff_t len)
4823{
4824 u64 cur_offset;
4825 u64 last_byte;
4826 u64 alloc_start;
4827 u64 alloc_end;
4828 u64 alloc_hint = 0;
4829 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4830 struct extent_map *em;
4831 int ret;
4832
4833 alloc_start = offset & ~mask;
4834 alloc_end = (offset + len + mask) & ~mask;
4835
4836 mutex_lock(&inode->i_mutex);
4837 if (alloc_start > inode->i_size) {
4838 ret = btrfs_cont_expand(inode, alloc_start);
4839 if (ret)
4840 goto out;
4841 }
4842
4843 while (1) {
4844 struct btrfs_ordered_extent *ordered;
4845 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4846 alloc_end - 1, GFP_NOFS);
4847 ordered = btrfs_lookup_first_ordered_extent(inode,
4848 alloc_end - 1);
4849 if (ordered &&
4850 ordered->file_offset + ordered->len > alloc_start &&
4851 ordered->file_offset < alloc_end) {
4852 btrfs_put_ordered_extent(ordered);
4853 unlock_extent(&BTRFS_I(inode)->io_tree,
4854 alloc_start, alloc_end - 1, GFP_NOFS);
4855 btrfs_wait_ordered_range(inode, alloc_start,
4856 alloc_end - alloc_start);
4857 } else {
4858 if (ordered)
4859 btrfs_put_ordered_extent(ordered);
4860 break;
4861 }
4862 }
4863
4864 cur_offset = alloc_start;
4865 while (1) {
4866 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4867 alloc_end - cur_offset, 0);
4868 BUG_ON(IS_ERR(em) || !em);
4869 last_byte = min(extent_map_end(em), alloc_end);
4870 last_byte = (last_byte + mask) & ~mask;
4871 if (em->block_start == EXTENT_MAP_HOLE) {
4872 ret = prealloc_file_range(inode, cur_offset,
4873 last_byte, alloc_hint, mode);
4874 if (ret < 0) {
4875 free_extent_map(em);
4876 break;
4877 }
4878 }
4879 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4880 alloc_hint = em->block_start;
4881 free_extent_map(em);
4882
4883 cur_offset = last_byte;
4884 if (cur_offset >= alloc_end) {
4885 ret = 0;
4886 break;
4887 }
4888 }
4889 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4890 GFP_NOFS);
4891out:
4892 mutex_unlock(&inode->i_mutex);
4893 return ret;
4894}
4895
e6dcd2dc
CM		 4896static int btrfs_set_page_dirty(struct page *page)
4897{
e6dcd2dc
CM		 4898 return __set_page_dirty_nobuffers(page);
4899}
4900
0ee0fda0 4901static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd
Y		 4902{
4903 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4904 return -EACCES;
33268eaf 4905 return generic_permission(inode, mask, btrfs_check_acl);
fdebe2bd 4906}
39279cc3
CM		 4907
4908static struct inode_operations btrfs_dir_inode_operations = {
4909 .lookup = btrfs_lookup,
4910 .create = btrfs_create,
4911 .unlink = btrfs_unlink,
4912 .link = btrfs_link,
4913 .mkdir = btrfs_mkdir,
4914 .rmdir = btrfs_rmdir,
4915 .rename = btrfs_rename,
4916 .symlink = btrfs_symlink,
4917 .setattr = btrfs_setattr,
618e21d5 4918 .mknod = btrfs_mknod,
95819c05
CH		 4919 .setxattr = btrfs_setxattr,
4920 .getxattr = btrfs_getxattr,
5103e947 4921 .listxattr = btrfs_listxattr,
95819c05 4922 .removexattr = btrfs_removexattr,
fdebe2bd 4923 .permission = btrfs_permission,
39279cc3 4924};
39279cc3
CM		 4925static struct inode_operations btrfs_dir_ro_inode_operations = {
4926 .lookup = btrfs_lookup,
fdebe2bd 4927 .permission = btrfs_permission,
39279cc3 4928};
39279cc3
CM		 4929static struct file_operations btrfs_dir_file_operations = {
4930 .llseek = generic_file_llseek,
4931 .read = generic_read_dir,
cbdf5a24 4932 .readdir = btrfs_real_readdir,
34287aa3 4933 .unlocked_ioctl = btrfs_ioctl,
39279cc3 4934#ifdef CONFIG_COMPAT
34287aa3 4935 .compat_ioctl = btrfs_ioctl,
39279cc3 4936#endif
6bf13c0c 4937 .release = btrfs_release_file,
e02119d5 4938 .fsync = btrfs_sync_file,
39279cc3
CM		 4939};
4940
d1310b2e 4941static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 4942 .fill_delalloc = run_delalloc_range,
065631f6 4943 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 4944 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 4945 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 4946 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 4947 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 4948 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 4949 .set_bit_hook = btrfs_set_bit_hook,
4950 .clear_bit_hook = btrfs_clear_bit_hook,
07157aac
CM		 4951};
4952
39279cc3
CM		 4953static struct address_space_operations btrfs_aops = {
4954 .readpage = btrfs_readpage,
4955 .writepage = btrfs_writepage,
b293f02e 4956 .writepages = btrfs_writepages,
3ab2fb5a 4957 .readpages = btrfs_readpages,
39279cc3 4958 .sync_page = block_sync_page,
39279cc3 4959 .bmap = btrfs_bmap,
16432985 4960 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 4961 .invalidatepage = btrfs_invalidatepage,
4962 .releasepage = btrfs_releasepage,
e6dcd2dc 4963 .set_page_dirty = btrfs_set_page_dirty,
39279cc3
CM		 4964};
4965
4966static struct address_space_operations btrfs_symlink_aops = {
4967 .readpage = btrfs_readpage,
4968 .writepage = btrfs_writepage,
2bf5a725
CM		 4969 .invalidatepage = btrfs_invalidatepage,
4970 .releasepage = btrfs_releasepage,
39279cc3
CM		 4971};
4972
4973static struct inode_operations btrfs_file_inode_operations = {
4974 .truncate = btrfs_truncate,
4975 .getattr = btrfs_getattr,
4976 .setattr = btrfs_setattr,
95819c05
CH		 4977 .setxattr = btrfs_setxattr,
4978 .getxattr = btrfs_getxattr,
5103e947 4979 .listxattr = btrfs_listxattr,
95819c05 4980 .removexattr = btrfs_removexattr,
fdebe2bd 4981 .permission = btrfs_permission,
d899e052 4982 .fallocate = btrfs_fallocate,
39279cc3 4983};
618e21d5
JB		 4984static struct inode_operations btrfs_special_inode_operations = {
4985 .getattr = btrfs_getattr,
4986 .setattr = btrfs_setattr,
fdebe2bd 4987 .permission = btrfs_permission,
95819c05
CH		 4988 .setxattr = btrfs_setxattr,
4989 .getxattr = btrfs_getxattr,
33268eaf 4990 .listxattr = btrfs_listxattr,
95819c05 4991 .removexattr = btrfs_removexattr,
618e21d5 4992};
39279cc3
CM		 4993static struct inode_operations btrfs_symlink_inode_operations = {
4994 .readlink = generic_readlink,
4995 .follow_link = page_follow_link_light,
4996 .put_link = page_put_link,
fdebe2bd 4997 .permission = btrfs_permission,
39279cc3 4998};
Linux 6.x block layer and io_uring tree(s)