projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: shared seed device
[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>
4b4e25f2 41#include "compat.h"
39279cc3
CM		 42#include "ctree.h"
43#include "disk-io.h"
44#include "transaction.h"
45#include "btrfs_inode.h"
46#include "ioctl.h"
47#include "print-tree.h"
0b86a832 48#include "volumes.h"
e6dcd2dc 49#include "ordered-data.h"
95819c05 50#include "xattr.h"
e02119d5 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
771ed689
CM		 886 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
887 PAGE_CACHE_SHIFT;
888 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
889
890 btrfs_queue_worker(&root->fs_info->delalloc_workers,
891 &async_cow->work);
892
893 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
894 wait_event(root->fs_info->async_submit_wait,
895 (atomic_read(&root->fs_info->async_delalloc_pages) <
896 limit));
897 }
898
899 while(atomic_read(&root->fs_info->async_submit_draining) &&
900 atomic_read(&root->fs_info->async_delalloc_pages)) {
901 wait_event(root->fs_info->async_submit_wait,
902 (atomic_read(&root->fs_info->async_delalloc_pages) ==
903 0));
904 }
905
906 *nr_written += nr_pages;
907 start = cur_end + 1;
908 }
909 *page_started = 1;
910 return 0;
be20aa9d
CM		 911}
912
d352ac68
CM		 913/*
914 * when nowcow writeback call back. This checks for snapshots or COW copies
915 * of the extents that exist in the file, and COWs the file as required.
916 *
917 * If no cow copies or snapshots exist, we write directly to the existing
918 * blocks on disk
919 */
c8b97818 920static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
771ed689
CM		 921 u64 start, u64 end, int *page_started, int force,
922 unsigned long *nr_written)
be20aa9d 923{
be20aa9d 924 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 925 struct btrfs_trans_handle *trans;
be20aa9d 926 struct extent_buffer *leaf;
be20aa9d 927 struct btrfs_path *path;
80ff3856 928 struct btrfs_file_extent_item *fi;
be20aa9d 929 struct btrfs_key found_key;
80ff3856
YZ		 930 u64 cow_start;
931 u64 cur_offset;
932 u64 extent_end;
933 u64 disk_bytenr;
934 u64 num_bytes;
935 int extent_type;
936 int ret;
d899e052 937 int type;
80ff3856
YZ		 938 int nocow;
939 int check_prev = 1;
be20aa9d
CM		 940
941 path = btrfs_alloc_path();
942 BUG_ON(!path);
7ea394f1
YZ		 943 trans = btrfs_join_transaction(root, 1);
944 BUG_ON(!trans);
be20aa9d 945
80ff3856
YZ		 946 cow_start = (u64)-1;
947 cur_offset = start;
948 while (1) {
949 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
950 cur_offset, 0);
951 BUG_ON(ret < 0);
952 if (ret > 0 && path->slots[0] > 0 && check_prev) {
953 leaf = path->nodes[0];
954 btrfs_item_key_to_cpu(leaf, &found_key,
955 path->slots[0] - 1);
956 if (found_key.objectid == inode->i_ino &&
957 found_key.type == BTRFS_EXTENT_DATA_KEY)
958 path->slots[0]--;
959 }
960 check_prev = 0;
961next_slot:
962 leaf = path->nodes[0];
963 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
964 ret = btrfs_next_leaf(root, path);
965 if (ret < 0)
966 BUG_ON(1);
967 if (ret > 0)
968 break;
969 leaf = path->nodes[0];
970 }
be20aa9d 971
80ff3856
YZ		 972 nocow = 0;
973 disk_bytenr = 0;
974 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
975
976 if (found_key.objectid > inode->i_ino ||
977 found_key.type > BTRFS_EXTENT_DATA_KEY ||
978 found_key.offset > end)
979 break;
980
981 if (found_key.offset > cur_offset) {
982 extent_end = found_key.offset;
983 goto out_check;
984 }
985
986 fi = btrfs_item_ptr(leaf, path->slots[0],
987 struct btrfs_file_extent_item);
988 extent_type = btrfs_file_extent_type(leaf, fi);
989
d899e052
YZ		 990 if (extent_type == BTRFS_FILE_EXTENT_REG ||
991 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856
YZ		 992 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
993 extent_end = found_key.offset +
994 btrfs_file_extent_num_bytes(leaf, fi);
995 if (extent_end <= start) {
996 path->slots[0]++;
997 goto next_slot;
998 }
999 if (btrfs_file_extent_compression(leaf, fi) ||
1000 btrfs_file_extent_encryption(leaf, fi) ||
1001 btrfs_file_extent_other_encoding(leaf, fi))
1002 goto out_check;
1003 if (disk_bytenr == 0)
1004 goto out_check;
d899e052
YZ		 1005 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1006 goto out_check;
80ff3856
YZ		 1007 if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
1008 goto out_check;
d2fb3437 1009 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856
YZ		 1010 goto out_check;
1011 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1012 nocow = 1;
1013 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1014 extent_end = found_key.offset +
1015 btrfs_file_extent_inline_len(leaf, fi);
1016 extent_end = ALIGN(extent_end, root->sectorsize);
1017 } else {
1018 BUG_ON(1);
1019 }
1020out_check:
1021 if (extent_end <= start) {
1022 path->slots[0]++;
1023 goto next_slot;
1024 }
1025 if (!nocow) {
1026 if (cow_start == (u64)-1)
1027 cow_start = cur_offset;
1028 cur_offset = extent_end;
1029 if (cur_offset > end)
1030 break;
1031 path->slots[0]++;
1032 goto next_slot;
7ea394f1
YZ		 1033 }
1034
1035 btrfs_release_path(root, path);
80ff3856
YZ		 1036 if (cow_start != (u64)-1) {
1037 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1038 found_key.offset - 1, page_started,
1039 nr_written, 1);
80ff3856
YZ		 1040 BUG_ON(ret);
1041 cow_start = (u64)-1;
7ea394f1 1042 }
80ff3856
YZ		 1043
1044 disk_bytenr += cur_offset - found_key.offset;
1045 num_bytes = min(end + 1, extent_end) - cur_offset;
d899e052
YZ		 1046 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1047 struct extent_map *em;
1048 struct extent_map_tree *em_tree;
1049 em_tree = &BTRFS_I(inode)->extent_tree;
1050 em = alloc_extent_map(GFP_NOFS);
1051 em->start = cur_offset;
445a6944 1052 em->orig_start = em->start;
d899e052
YZ		 1053 em->len = num_bytes;
1054 em->block_len = num_bytes;
1055 em->block_start = disk_bytenr;
1056 em->bdev = root->fs_info->fs_devices->latest_bdev;
1057 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1058 while (1) {
1059 spin_lock(&em_tree->lock);
1060 ret = add_extent_mapping(em_tree, em);
1061 spin_unlock(&em_tree->lock);
1062 if (ret != -EEXIST) {
1063 free_extent_map(em);
1064 break;
1065 }
1066 btrfs_drop_extent_cache(inode, em->start,
1067 em->start + em->len - 1, 0);
1068 }
1069 type = BTRFS_ORDERED_PREALLOC;
1070 } else {
1071 type = BTRFS_ORDERED_NOCOW;
1072 }
80ff3856
YZ		 1073
1074 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 1075 num_bytes, num_bytes, type);
1076 BUG_ON(ret);
771ed689 1077
d899e052
YZ		 1078 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1079 cur_offset, cur_offset + num_bytes - 1,
771ed689 1080 locked_page, 1, 1, 1, 0, 0, 0);
80ff3856
YZ		 1081 cur_offset = extent_end;
1082 if (cur_offset > end)
1083 break;
be20aa9d 1084 }
80ff3856
YZ		 1085 btrfs_release_path(root, path);
1086
1087 if (cur_offset <= end && cow_start == (u64)-1)
1088 cow_start = cur_offset;
1089 if (cow_start != (u64)-1) {
1090 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1091 page_started, nr_written, 1);
80ff3856
YZ		 1092 BUG_ON(ret);
1093 }
1094
1095 ret = btrfs_end_transaction(trans, root);
1096 BUG_ON(ret);
7ea394f1 1097 btrfs_free_path(path);
80ff3856 1098 return 0;
be20aa9d
CM		 1099}
1100
d352ac68
CM		 1101/*
1102 * extent_io.c call back to do delayed allocation processing
1103 */
c8b97818 1104static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1105 u64 start, u64 end, int *page_started,
1106 unsigned long *nr_written)
be20aa9d
CM		 1107{
1108 struct btrfs_root *root = BTRFS_I(inode)->root;
1109 int ret;
a2135011 1110
b98b6767
Y		 1111 if (btrfs_test_opt(root, NODATACOW) ||
1112 btrfs_test_flag(inode, NODATACOW))
c8b97818 1113 ret = run_delalloc_nocow(inode, locked_page, start, end,
ce397c06 1114 page_started, 1, nr_written);
d899e052
YZ		 1115 else if (btrfs_test_flag(inode, PREALLOC))
1116 ret = run_delalloc_nocow(inode, locked_page, start, end,
ce397c06 1117 page_started, 0, nr_written);
be20aa9d 1118 else
771ed689
CM		 1119 ret = cow_file_range_async(inode, locked_page, start, end,
1120 page_started, nr_written);
1832a6d5 1121
b888db2b
CM		 1122 return ret;
1123}
1124
d352ac68
CM		 1125/*
1126 * extent_io.c set_bit_hook, used to track delayed allocation
1127 * bytes in this file, and to maintain the list of inodes that
1128 * have pending delalloc work to be done.
1129 */
b2950863 1130static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1131 unsigned long old, unsigned long bits)
291d673e 1132{
bcbfce8a 1133 unsigned long flags;
b0c68f8b 1134 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1135 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a 1136 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
9069218d 1137 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
291d673e 1138 root->fs_info->delalloc_bytes += end - start + 1;
ea8c2819
CM		 1139 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1140 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1141 &root->fs_info->delalloc_inodes);
1142 }
bcbfce8a 1143 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
291d673e
CM		 1144 }
1145 return 0;
1146}
1147
d352ac68
CM		 1148/*
1149 * extent_io.c clear_bit_hook, see set_bit_hook for why
1150 */
b2950863 1151static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1152 unsigned long old, unsigned long bits)
291d673e 1153{
b0c68f8b 1154 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1155 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a
CM		 1156 unsigned long flags;
1157
1158 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
b0c68f8b
CM		 1159 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1160 printk("warning: delalloc account %Lu %Lu\n",
1161 end - start + 1, root->fs_info->delalloc_bytes);
1162 root->fs_info->delalloc_bytes = 0;
9069218d 1163 BTRFS_I(inode)->delalloc_bytes = 0;
b0c68f8b
CM		 1164 } else {
1165 root->fs_info->delalloc_bytes -= end - start + 1;
9069218d 1166 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
b0c68f8b 1167 }
ea8c2819
CM		 1168 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1169 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1170 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1171 }
bcbfce8a 1172 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
291d673e
CM		 1173 }
1174 return 0;
1175}
1176
d352ac68
CM		 1177/*
1178 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1179 * we don't create bios that span stripes or chunks
1180 */
239b14b3 1181int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1182 size_t size, struct bio *bio,
1183 unsigned long bio_flags)
239b14b3
CM		 1184{
1185 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1186 struct btrfs_mapping_tree *map_tree;
a62b9401 1187 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1188 u64 length = 0;
1189 u64 map_length;
239b14b3
CM		 1190 int ret;
1191
771ed689
CM		 1192 if (bio_flags & EXTENT_BIO_COMPRESSED)
1193 return 0;
1194
f2d8d74d 1195 length = bio->bi_size;
239b14b3
CM		 1196 map_tree = &root->fs_info->mapping_tree;
1197 map_length = length;
cea9e445 1198 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1199 &map_length, NULL, 0);
cea9e445 1200
239b14b3 1201 if (map_length < length + size) {
239b14b3
CM		 1202 return 1;
1203 }
1204 return 0;
1205}
1206
d352ac68
CM		 1207/*
1208 * in order to insert checksums into the metadata in large chunks,
1209 * we wait until bio submission time. All the pages in the bio are
1210 * checksummed and sums are attached onto the ordered extent record.
1211 *
1212 * At IO completion time the cums attached on the ordered extent record
1213 * are inserted into the btree
1214 */
b2950863 1215static int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio,
c8b97818 1216 int mirror_num, unsigned long bio_flags)
065631f6 1217{
065631f6 1218 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1219 int ret = 0;
e015640f 1220
d20f7043 1221 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
44b8bd7e 1222 BUG_ON(ret);
4a69a410
CM		 1223 return 0;
1224}
e015640f 1225
4a69a410
CM		 1226/*
1227 * in order to insert checksums into the metadata in large chunks,
1228 * we wait until bio submission time. All the pages in the bio are
1229 * checksummed and sums are attached onto the ordered extent record.
1230 *
1231 * At IO completion time the cums attached on the ordered extent record
1232 * are inserted into the btree
1233 */
b2950863 1234static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
4a69a410
CM		 1235 int mirror_num, unsigned long bio_flags)
1236{
1237 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1238 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1239}
1240
d352ac68
CM		 1241/*
1242 * extent_io.c submission hook. This does the right thing for csum calculation on write,
1243 * or reading the csums from the tree before a read
1244 */
b2950863 1245static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 1246 int mirror_num, unsigned long bio_flags)
44b8bd7e
CM		 1247{
1248 struct btrfs_root *root = BTRFS_I(inode)->root;
1249 int ret = 0;
19b9bdb0 1250 int skip_sum;
44b8bd7e 1251
e6dcd2dc
CM		 1252 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1253 BUG_ON(ret);
065631f6 1254
19b9bdb0
CM		 1255 skip_sum = btrfs_test_opt(root, NODATASUM) ||
1256 btrfs_test_flag(inode, NODATASUM);
7ea394f1 1257
4d1b5fb4 1258 if (!(rw & (1 << BIO_RW))) {
d20f7043 1259 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1260 return btrfs_submit_compressed_read(inode, bio,
1261 mirror_num, bio_flags);
d20f7043
CM		 1262 } else if (!skip_sum)
1263 btrfs_lookup_bio_sums(root, inode, bio, NULL);
4d1b5fb4 1264 goto mapit;
19b9bdb0
CM		 1265 } else if (!skip_sum) {
1266 /* we're doing a write, do the async checksumming */
1267 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1268 inode, rw, bio, mirror_num,
4a69a410
CM		 1269 bio_flags, __btrfs_submit_bio_start,
1270 __btrfs_submit_bio_done);
19b9bdb0
CM		 1271 }
1272
0b86a832 1273mapit:
8b712842 1274 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1275}
6885f308 1276
d352ac68
CM		 1277/*
1278 * given a list of ordered sums record them in the inode. This happens
1279 * at IO completion time based on sums calculated at bio submission time.
1280 */
ba1da2f4 1281static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1282 struct inode *inode, u64 file_offset,
1283 struct list_head *list)
1284{
1285 struct list_head *cur;
1286 struct btrfs_ordered_sum *sum;
1287
1288 btrfs_set_trans_block_group(trans, inode);
ba1da2f4 1289 list_for_each(cur, list) {
e6dcd2dc 1290 sum = list_entry(cur, struct btrfs_ordered_sum, list);
d20f7043
CM		 1291 btrfs_csum_file_blocks(trans,
1292 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1293 }
1294 return 0;
1295}
1296
ea8c2819
CM		 1297int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1298{
771ed689
CM		 1299 if ((end & (PAGE_CACHE_SIZE - 1)) == 0) {
1300 WARN_ON(1);
1301 }
ea8c2819
CM		 1302 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1303 GFP_NOFS);
1304}
1305
d352ac68 1306/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1307struct btrfs_writepage_fixup {
1308 struct page *page;
1309 struct btrfs_work work;
1310};
1311
b2950863 1312static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1313{
1314 struct btrfs_writepage_fixup *fixup;
1315 struct btrfs_ordered_extent *ordered;
1316 struct page *page;
1317 struct inode *inode;
1318 u64 page_start;
1319 u64 page_end;
1320
1321 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1322 page = fixup->page;
4a096752 1323again:
247e743c
CM		 1324 lock_page(page);
1325 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1326 ClearPageChecked(page);
1327 goto out_page;
1328 }
1329
1330 inode = page->mapping->host;
1331 page_start = page_offset(page);
1332 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1333
1334 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
4a096752
CM		 1335
1336 /* already ordered? We're done */
1337 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1338 EXTENT_ORDERED, 0)) {
247e743c 1339 goto out;
4a096752
CM		 1340 }
1341
1342 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1343 if (ordered) {
1344 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1345 page_end, GFP_NOFS);
1346 unlock_page(page);
1347 btrfs_start_ordered_extent(inode, ordered, 1);
1348 goto again;
1349 }
247e743c 1350
ea8c2819 1351 btrfs_set_extent_delalloc(inode, page_start, page_end);
247e743c
CM		 1352 ClearPageChecked(page);
1353out:
1354 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1355out_page:
1356 unlock_page(page);
1357 page_cache_release(page);
1358}
1359
1360/*
1361 * There are a few paths in the higher layers of the kernel that directly
1362 * set the page dirty bit without asking the filesystem if it is a
1363 * good idea. This causes problems because we want to make sure COW
1364 * properly happens and the data=ordered rules are followed.
1365 *
c8b97818 1366 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1367 * hasn't been properly setup for IO. We kick off an async process
1368 * to fix it up. The async helper will wait for ordered extents, set
1369 * the delalloc bit and make it safe to write the page.
1370 */
b2950863 1371static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1372{
1373 struct inode *inode = page->mapping->host;
1374 struct btrfs_writepage_fixup *fixup;
1375 struct btrfs_root *root = BTRFS_I(inode)->root;
1376 int ret;
1377
1378 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1379 EXTENT_ORDERED, 0);
1380 if (ret)
1381 return 0;
1382
1383 if (PageChecked(page))
1384 return -EAGAIN;
1385
1386 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1387 if (!fixup)
1388 return -EAGAIN;
f421950f 1389
247e743c
CM		 1390 SetPageChecked(page);
1391 page_cache_get(page);
1392 fixup->work.func = btrfs_writepage_fixup_worker;
1393 fixup->page = page;
1394 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1395 return -EAGAIN;
1396}
1397
d899e052
YZ		 1398static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1399 struct inode *inode, u64 file_pos,
1400 u64 disk_bytenr, u64 disk_num_bytes,
1401 u64 num_bytes, u64 ram_bytes,
1402 u8 compression, u8 encryption,
1403 u16 other_encoding, int extent_type)
1404{
1405 struct btrfs_root *root = BTRFS_I(inode)->root;
1406 struct btrfs_file_extent_item *fi;
1407 struct btrfs_path *path;
1408 struct extent_buffer *leaf;
1409 struct btrfs_key ins;
1410 u64 hint;
1411 int ret;
1412
1413 path = btrfs_alloc_path();
1414 BUG_ON(!path);
1415
1416 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1417 file_pos + num_bytes, file_pos, &hint);
1418 BUG_ON(ret);
1419
1420 ins.objectid = inode->i_ino;
1421 ins.offset = file_pos;
1422 ins.type = BTRFS_EXTENT_DATA_KEY;
1423 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1424 BUG_ON(ret);
1425 leaf = path->nodes[0];
1426 fi = btrfs_item_ptr(leaf, path->slots[0],
1427 struct btrfs_file_extent_item);
1428 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1429 btrfs_set_file_extent_type(leaf, fi, extent_type);
1430 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1431 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1432 btrfs_set_file_extent_offset(leaf, fi, 0);
1433 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1434 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1435 btrfs_set_file_extent_compression(leaf, fi, compression);
1436 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1437 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1438 btrfs_mark_buffer_dirty(leaf);
1439
1440 inode_add_bytes(inode, num_bytes);
1441 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1442
1443 ins.objectid = disk_bytenr;
1444 ins.offset = disk_num_bytes;
1445 ins.type = BTRFS_EXTENT_ITEM_KEY;
1446 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1447 root->root_key.objectid,
1448 trans->transid, inode->i_ino, &ins);
1449 BUG_ON(ret);
1450
1451 btrfs_free_path(path);
1452 return 0;
1453}
1454
d352ac68
CM		 1455/* as ordered data IO finishes, this gets called so we can finish
1456 * an ordered extent if the range of bytes in the file it covers are
1457 * fully written.
1458 */
211f90e6 1459static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1460{
e6dcd2dc
CM		 1461 struct btrfs_root *root = BTRFS_I(inode)->root;
1462 struct btrfs_trans_handle *trans;
1463 struct btrfs_ordered_extent *ordered_extent;
1464 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
d899e052 1465 int compressed = 0;
e6dcd2dc
CM		 1466 int ret;
1467
1468 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
ba1da2f4 1469 if (!ret)
e6dcd2dc 1470 return 0;
e6dcd2dc 1471
f9295749 1472 trans = btrfs_join_transaction(root, 1);
e6dcd2dc
CM		 1473
1474 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1475 BUG_ON(!ordered_extent);
7ea394f1
YZ		 1476 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1477 goto nocow;
e6dcd2dc
CM		 1478
1479 lock_extent(io_tree, ordered_extent->file_offset,
1480 ordered_extent->file_offset + ordered_extent->len - 1,
1481 GFP_NOFS);
1482
c8b97818 1483 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
d899e052
YZ		 1484 compressed = 1;
1485 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1486 BUG_ON(compressed);
1487 ret = btrfs_mark_extent_written(trans, root, inode,
1488 ordered_extent->file_offset,
1489 ordered_extent->file_offset +
1490 ordered_extent->len);
1491 BUG_ON(ret);
1492 } else {
1493 ret = insert_reserved_file_extent(trans, inode,
1494 ordered_extent->file_offset,
1495 ordered_extent->start,
1496 ordered_extent->disk_len,
1497 ordered_extent->len,
1498 ordered_extent->len,
1499 compressed, 0, 0,
1500 BTRFS_FILE_EXTENT_REG);
1501 BUG_ON(ret);
1502 }
e6dcd2dc
CM		 1503 unlock_extent(io_tree, ordered_extent->file_offset,
1504 ordered_extent->file_offset + ordered_extent->len - 1,
1505 GFP_NOFS);
7ea394f1 1506nocow:
e6dcd2dc
CM		 1507 add_pending_csums(trans, inode, ordered_extent->file_offset,
1508 &ordered_extent->list);
1509
34353029 1510 mutex_lock(&BTRFS_I(inode)->extent_mutex);
dbe674a9 1511 btrfs_ordered_update_i_size(inode, ordered_extent);
e02119d5 1512 btrfs_update_inode(trans, root, inode);
e6dcd2dc 1513 btrfs_remove_ordered_extent(inode, ordered_extent);
34353029 1514 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
7f3c74fb 1515
e6dcd2dc
CM		 1516 /* once for us */
1517 btrfs_put_ordered_extent(ordered_extent);
1518 /* once for the tree */
1519 btrfs_put_ordered_extent(ordered_extent);
1520
e6dcd2dc
CM		 1521 btrfs_end_transaction(trans, root);
1522 return 0;
1523}
1524
b2950863 1525static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1526 struct extent_state *state, int uptodate)
1527{
1528 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1529}
1530
d352ac68
CM		 1531/*
1532 * When IO fails, either with EIO or csum verification fails, we
1533 * try other mirrors that might have a good copy of the data. This
1534 * io_failure_record is used to record state as we go through all the
1535 * mirrors. If another mirror has good data, the page is set up to date
1536 * and things continue. If a good mirror can't be found, the original
1537 * bio end_io callback is called to indicate things have failed.
1538 */
7e38326f
CM		 1539struct io_failure_record {
1540 struct page *page;
1541 u64 start;
1542 u64 len;
1543 u64 logical;
d20f7043 1544 unsigned long bio_flags;
7e38326f
CM		 1545 int last_mirror;
1546};
1547
b2950863 1548static int btrfs_io_failed_hook(struct bio *failed_bio,
1259ab75
CM		 1549 struct page *page, u64 start, u64 end,
1550 struct extent_state *state)
7e38326f
CM		 1551{
1552 struct io_failure_record *failrec = NULL;
1553 u64 private;
1554 struct extent_map *em;
1555 struct inode *inode = page->mapping->host;
1556 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1557 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM		 1558 struct bio *bio;
1559 int num_copies;
1560 int ret;
1259ab75 1561 int rw;
7e38326f
CM		 1562 u64 logical;
1563
1564 ret = get_state_private(failure_tree, start, &private);
1565 if (ret) {
7e38326f
CM		 1566 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1567 if (!failrec)
1568 return -ENOMEM;
1569 failrec->start = start;
1570 failrec->len = end - start + 1;
1571 failrec->last_mirror = 0;
d20f7043 1572 failrec->bio_flags = 0;
7e38326f 1573
3b951516
CM		 1574 spin_lock(&em_tree->lock);
1575 em = lookup_extent_mapping(em_tree, start, failrec->len);
1576 if (em->start > start || em->start + em->len < start) {
1577 free_extent_map(em);
1578 em = NULL;
1579 }
1580 spin_unlock(&em_tree->lock);
7e38326f
CM		 1581
1582 if (!em || IS_ERR(em)) {
1583 kfree(failrec);
1584 return -EIO;
1585 }
1586 logical = start - em->start;
1587 logical = em->block_start + logical;
d20f7043
CM		 1588 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1589 logical = em->block_start;
1590 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1591 }
7e38326f
CM		 1592 failrec->logical = logical;
1593 free_extent_map(em);
1594 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1595 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM		 1596 set_state_private(failure_tree, start,
1597 (u64)(unsigned long)failrec);
7e38326f 1598 } else {
587f7704 1599 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM		 1600 }
1601 num_copies = btrfs_num_copies(
1602 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1603 failrec->logical, failrec->len);
1604 failrec->last_mirror++;
1605 if (!state) {
1606 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1607 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1608 failrec->start,
1609 EXTENT_LOCKED);
1610 if (state && state->start != failrec->start)
1611 state = NULL;
1612 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1613 }
1614 if (!state || failrec->last_mirror > num_copies) {
1615 set_state_private(failure_tree, failrec->start, 0);
1616 clear_extent_bits(failure_tree, failrec->start,
1617 failrec->start + failrec->len - 1,
1618 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1619 kfree(failrec);
1620 return -EIO;
1621 }
1622 bio = bio_alloc(GFP_NOFS, 1);
1623 bio->bi_private = state;
1624 bio->bi_end_io = failed_bio->bi_end_io;
1625 bio->bi_sector = failrec->logical >> 9;
1626 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1627 bio->bi_size = 0;
d20f7043 1628
7e38326f 1629 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1259ab75
CM		 1630 if (failed_bio->bi_rw & (1 << BIO_RW))
1631 rw = WRITE;
1632 else
1633 rw = READ;
1634
1635 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818 1636 failrec->last_mirror,
d20f7043 1637 failrec->bio_flags);
1259ab75
CM		 1638 return 0;
1639}
1640
d352ac68
CM		 1641/*
1642 * each time an IO finishes, we do a fast check in the IO failure tree
1643 * to see if we need to process or clean up an io_failure_record
1644 */
b2950863 1645static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1259ab75
CM		 1646{
1647 u64 private;
1648 u64 private_failure;
1649 struct io_failure_record *failure;
1650 int ret;
1651
1652 private = 0;
1653 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1654 (u64)-1, 1, EXTENT_DIRTY)) {
1655 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1656 start, &private_failure);
1657 if (ret == 0) {
1658 failure = (struct io_failure_record *)(unsigned long)
1659 private_failure;
1660 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1661 failure->start, 0);
1662 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1663 failure->start,
1664 failure->start + failure->len - 1,
1665 EXTENT_DIRTY | EXTENT_LOCKED,
1666 GFP_NOFS);
1667 kfree(failure);
1668 }
1669 }
7e38326f
CM		 1670 return 0;
1671}
1672
d352ac68
CM		 1673/*
1674 * when reads are done, we need to check csums to verify the data is correct
1675 * if there's a match, we allow the bio to finish. If not, we go through
1676 * the io_failure_record routines to find good copies
1677 */
b2950863 1678static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1679 struct extent_state *state)
07157aac 1680{
35ebb934 1681 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1682 struct inode *inode = page->mapping->host;
d1310b2e 1683 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1684 char *kaddr;
aadfeb6e 1685 u64 private = ~(u32)0;
07157aac 1686 int ret;
ff79f819
CM		 1687 struct btrfs_root *root = BTRFS_I(inode)->root;
1688 u32 csum = ~(u32)0;
bbf0d006 1689 unsigned long flags;
d1310b2e 1690
d20f7043
CM		 1691 if (PageChecked(page)) {
1692 ClearPageChecked(page);
1693 goto good;
1694 }
b98b6767
Y		 1695 if (btrfs_test_opt(root, NODATASUM) ||
1696 btrfs_test_flag(inode, NODATASUM))
b6cda9bc 1697 return 0;
d20f7043 1698
c2e639f0 1699 if (state && state->start == start) {
70dec807
CM		 1700 private = state->private;
1701 ret = 0;
1702 } else {
1703 ret = get_state_private(io_tree, start, &private);
1704 }
bbf0d006 1705 local_irq_save(flags);
07157aac
CM		 1706 kaddr = kmap_atomic(page, KM_IRQ0);
1707 if (ret) {
1708 goto zeroit;
1709 }
ff79f819
CM		 1710 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1711 btrfs_csum_final(csum, (char *)&csum);
1712 if (csum != private) {
07157aac
CM		 1713 goto zeroit;
1714 }
1715 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1716 local_irq_restore(flags);
d20f7043 1717good:
7e38326f
CM		 1718 /* if the io failure tree for this inode is non-empty,
1719 * check to see if we've recovered from a failed IO
1720 */
1259ab75 1721 btrfs_clean_io_failures(inode, start);
07157aac
CM		 1722 return 0;
1723
1724zeroit:
aadfeb6e
CM		 1725 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1726 page->mapping->host->i_ino, (unsigned long long)start, csum,
1727 private);
db94535d
CM		 1728 memset(kaddr + offset, 1, end - start + 1);
1729 flush_dcache_page(page);
07157aac 1730 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1731 local_irq_restore(flags);
3b951516
CM		 1732 if (private == 0)
1733 return 0;
7e38326f 1734 return -EIO;
07157aac 1735}
b888db2b 1736
7b128766
JB		 1737/*
1738 * This creates an orphan entry for the given inode in case something goes
1739 * wrong in the middle of an unlink/truncate.
1740 */
1741int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1742{
1743 struct btrfs_root *root = BTRFS_I(inode)->root;
1744 int ret = 0;
1745
bcc63abb 1746 spin_lock(&root->list_lock);
7b128766
JB		 1747
1748 /* already on the orphan list, we're good */
1749 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1750 spin_unlock(&root->list_lock);
7b128766
JB		 1751 return 0;
1752 }
1753
1754 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1755
bcc63abb 1756 spin_unlock(&root->list_lock);
7b128766
JB		 1757
1758 /*
1759 * insert an orphan item to track this unlinked/truncated file
1760 */
1761 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1762
1763 return ret;
1764}
1765
1766/*
1767 * We have done the truncate/delete so we can go ahead and remove the orphan
1768 * item for this particular inode.
1769 */
1770int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1771{
1772 struct btrfs_root *root = BTRFS_I(inode)->root;
1773 int ret = 0;
1774
bcc63abb 1775 spin_lock(&root->list_lock);
7b128766
JB		 1776
1777 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1778 spin_unlock(&root->list_lock);
7b128766
JB		 1779 return 0;
1780 }
1781
1782 list_del_init(&BTRFS_I(inode)->i_orphan);
1783 if (!trans) {
bcc63abb 1784 spin_unlock(&root->list_lock);
7b128766
JB		 1785 return 0;
1786 }
1787
bcc63abb 1788 spin_unlock(&root->list_lock);
7b128766
JB		 1789
1790 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1791
1792 return ret;
1793}
1794
1795/*
1796 * this cleans up any orphans that may be left on the list from the last use
1797 * of this root.
1798 */
1799void btrfs_orphan_cleanup(struct btrfs_root *root)
1800{
1801 struct btrfs_path *path;
1802 struct extent_buffer *leaf;
1803 struct btrfs_item *item;
1804 struct btrfs_key key, found_key;
1805 struct btrfs_trans_handle *trans;
1806 struct inode *inode;
1807 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1808
7b128766
JB		 1809 path = btrfs_alloc_path();
1810 if (!path)
1811 return;
1812 path->reada = -1;
1813
1814 key.objectid = BTRFS_ORPHAN_OBJECTID;
1815 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1816 key.offset = (u64)-1;
1817
7b128766
JB		 1818
1819 while (1) {
1820 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1821 if (ret < 0) {
1822 printk(KERN_ERR "Error searching slot for orphan: %d"
1823 "\n", ret);
1824 break;
1825 }
1826
1827 /*
1828 * if ret == 0 means we found what we were searching for, which
1829 * is weird, but possible, so only screw with path if we didnt
1830 * find the key and see if we have stuff that matches
1831 */
1832 if (ret > 0) {
1833 if (path->slots[0] == 0)
1834 break;
1835 path->slots[0]--;
1836 }
1837
1838 /* pull out the item */
1839 leaf = path->nodes[0];
1840 item = btrfs_item_nr(leaf, path->slots[0]);
1841 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1842
1843 /* make sure the item matches what we want */
1844 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1845 break;
1846 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1847 break;
1848
1849 /* release the path since we're done with it */
1850 btrfs_release_path(root, path);
1851
1852 /*
1853 * this is where we are basically btrfs_lookup, without the
1854 * crossing root thing. we store the inode number in the
1855 * offset of the orphan item.
1856 */
5b21f2ed 1857 inode = btrfs_iget_locked(root->fs_info->sb,
7b128766
JB		 1858 found_key.offset, root);
1859 if (!inode)
1860 break;
1861
1862 if (inode->i_state & I_NEW) {
1863 BTRFS_I(inode)->root = root;
1864
1865 /* have to set the location manually */
1866 BTRFS_I(inode)->location.objectid = inode->i_ino;
1867 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1868 BTRFS_I(inode)->location.offset = 0;
1869
1870 btrfs_read_locked_inode(inode);
1871 unlock_new_inode(inode);
1872 }
1873
1874 /*
1875 * add this inode to the orphan list so btrfs_orphan_del does
1876 * the proper thing when we hit it
1877 */
bcc63abb 1878 spin_lock(&root->list_lock);
7b128766 1879 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
bcc63abb 1880 spin_unlock(&root->list_lock);
7b128766
JB		 1881
1882 /*
1883 * if this is a bad inode, means we actually succeeded in
1884 * removing the inode, but not the orphan record, which means
1885 * we need to manually delete the orphan since iput will just
1886 * do a destroy_inode
1887 */
1888 if (is_bad_inode(inode)) {
5b21f2ed 1889 trans = btrfs_start_transaction(root, 1);
7b128766 1890 btrfs_orphan_del(trans, inode);
5b21f2ed 1891 btrfs_end_transaction(trans, root);
7b128766
JB		 1892 iput(inode);
1893 continue;
1894 }
1895
1896 /* if we have links, this was a truncate, lets do that */
1897 if (inode->i_nlink) {
1898 nr_truncate++;
1899 btrfs_truncate(inode);
1900 } else {
1901 nr_unlink++;
1902 }
1903
1904 /* this will do delete_inode and everything for us */
1905 iput(inode);
1906 }
1907
1908 if (nr_unlink)
1909 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1910 if (nr_truncate)
1911 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1912
1913 btrfs_free_path(path);
7b128766
JB		 1914}
1915
d352ac68
CM		 1916/*
1917 * read an inode from the btree into the in-memory inode
1918 */
39279cc3
CM		 1919void btrfs_read_locked_inode(struct inode *inode)
1920{
1921 struct btrfs_path *path;
5f39d397 1922 struct extent_buffer *leaf;
39279cc3 1923 struct btrfs_inode_item *inode_item;
0b86a832 1924 struct btrfs_timespec *tspec;
39279cc3
CM		 1925 struct btrfs_root *root = BTRFS_I(inode)->root;
1926 struct btrfs_key location;
1927 u64 alloc_group_block;
618e21d5 1928 u32 rdev;
39279cc3
CM		 1929 int ret;
1930
1931 path = btrfs_alloc_path();
1932 BUG_ON(!path);
39279cc3 1933 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 1934
39279cc3 1935 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 1936 if (ret)
39279cc3 1937 goto make_bad;
39279cc3 1938
5f39d397
CM		 1939 leaf = path->nodes[0];
1940 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1941 struct btrfs_inode_item);
1942
1943 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1944 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1945 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1946 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 1947 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 1948
1949 tspec = btrfs_inode_atime(inode_item);
1950 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1951 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1952
1953 tspec = btrfs_inode_mtime(inode_item);
1954 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1955 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1956
1957 tspec = btrfs_inode_ctime(inode_item);
1958 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1959 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1960
a76a3cd4 1961 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 1962 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 1963 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 1964 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 1965 inode->i_rdev = 0;
5f39d397
CM		 1966 rdev = btrfs_inode_rdev(leaf, inode_item);
1967
aec7477b 1968 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 1969 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
aec7477b 1970
5f39d397 1971 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
d2fb3437
YZ		 1972 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
1973 alloc_group_block, 0);
39279cc3
CM		 1974 btrfs_free_path(path);
1975 inode_item = NULL;
1976
39279cc3 1977 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 1978 case S_IFREG:
1979 inode->i_mapping->a_ops = &btrfs_aops;
04160088 1980 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 1981 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 1982 inode->i_fop = &btrfs_file_operations;
1983 inode->i_op = &btrfs_file_inode_operations;
1984 break;
1985 case S_IFDIR:
1986 inode->i_fop = &btrfs_dir_file_operations;
1987 if (root == root->fs_info->tree_root)
1988 inode->i_op = &btrfs_dir_ro_inode_operations;
1989 else
1990 inode->i_op = &btrfs_dir_inode_operations;
1991 break;
1992 case S_IFLNK:
1993 inode->i_op = &btrfs_symlink_inode_operations;
1994 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 1995 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 1996 break;
618e21d5
JB		 1997 default:
1998 init_special_inode(inode, inode->i_mode, rdev);
1999 break;
39279cc3
CM		 2000 }
2001 return;
2002
2003make_bad:
39279cc3 2004 btrfs_free_path(path);
39279cc3
CM		 2005 make_bad_inode(inode);
2006}
2007
d352ac68
CM		 2008/*
2009 * given a leaf and an inode, copy the inode fields into the leaf
2010 */
e02119d5
CM		 2011static void fill_inode_item(struct btrfs_trans_handle *trans,
2012 struct extent_buffer *leaf,
5f39d397 2013 struct btrfs_inode_item *item,
39279cc3
CM		 2014 struct inode *inode)
2015{
5f39d397
CM		 2016 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2017 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2018 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2019 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2020 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2021
2022 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2023 inode->i_atime.tv_sec);
2024 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2025 inode->i_atime.tv_nsec);
2026
2027 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2028 inode->i_mtime.tv_sec);
2029 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2030 inode->i_mtime.tv_nsec);
2031
2032 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2033 inode->i_ctime.tv_sec);
2034 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2035 inode->i_ctime.tv_nsec);
2036
a76a3cd4 2037 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2038 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2039 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2040 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2041 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2042 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d2fb3437 2043 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
39279cc3
CM		 2044}
2045
d352ac68
CM		 2046/*
2047 * copy everything in the in-memory inode into the btree.
2048 */
ba1da2f4 2049int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
39279cc3
CM		 2050 struct btrfs_root *root,
2051 struct inode *inode)
2052{
2053 struct btrfs_inode_item *inode_item;
2054 struct btrfs_path *path;
5f39d397 2055 struct extent_buffer *leaf;
39279cc3
CM		 2056 int ret;
2057
2058 path = btrfs_alloc_path();
2059 BUG_ON(!path);
39279cc3
CM		 2060 ret = btrfs_lookup_inode(trans, root, path,
2061 &BTRFS_I(inode)->location, 1);
2062 if (ret) {
2063 if (ret > 0)
2064 ret = -ENOENT;
2065 goto failed;
2066 }
2067
5f39d397
CM		 2068 leaf = path->nodes[0];
2069 inode_item = btrfs_item_ptr(leaf, path->slots[0],
39279cc3
CM		 2070 struct btrfs_inode_item);
2071
e02119d5 2072 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2073 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2074 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2075 ret = 0;
2076failed:
39279cc3
CM		 2077 btrfs_free_path(path);
2078 return ret;
2079}
2080
2081
d352ac68
CM		 2082/*
2083 * unlink helper that gets used here in inode.c and in the tree logging
2084 * recovery code. It remove a link in a directory with a given name, and
2085 * also drops the back refs in the inode to the directory
2086 */
e02119d5
CM		 2087int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2088 struct btrfs_root *root,
2089 struct inode *dir, struct inode *inode,
2090 const char *name, int name_len)
39279cc3
CM		 2091{
2092 struct btrfs_path *path;
39279cc3 2093 int ret = 0;
5f39d397 2094 struct extent_buffer *leaf;
39279cc3 2095 struct btrfs_dir_item *di;
5f39d397 2096 struct btrfs_key key;
aec7477b 2097 u64 index;
39279cc3
CM		 2098
2099 path = btrfs_alloc_path();
54aa1f4d
CM		 2100 if (!path) {
2101 ret = -ENOMEM;
2102 goto err;
2103 }
2104
39279cc3
CM		 2105 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2106 name, name_len, -1);
2107 if (IS_ERR(di)) {
2108 ret = PTR_ERR(di);
2109 goto err;
2110 }
2111 if (!di) {
2112 ret = -ENOENT;
2113 goto err;
2114 }
5f39d397
CM		 2115 leaf = path->nodes[0];
2116 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2117 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2118 if (ret)
2119 goto err;
39279cc3
CM		 2120 btrfs_release_path(root, path);
2121
aec7477b 2122 ret = btrfs_del_inode_ref(trans, root, name, name_len,
e02119d5
CM		 2123 inode->i_ino,
2124 dir->i_ino, &index);
aec7477b
JB		 2125 if (ret) {
2126 printk("failed to delete reference to %.*s, "
2127 "inode %lu parent %lu\n", name_len, name,
e02119d5 2128 inode->i_ino, dir->i_ino);
aec7477b
JB		 2129 goto err;
2130 }
2131
39279cc3 2132 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
aec7477b 2133 index, name, name_len, -1);
39279cc3
CM		 2134 if (IS_ERR(di)) {
2135 ret = PTR_ERR(di);
2136 goto err;
2137 }
2138 if (!di) {
2139 ret = -ENOENT;
2140 goto err;
2141 }
2142 ret = btrfs_delete_one_dir_name(trans, root, path, di);
925baedd 2143 btrfs_release_path(root, path);
39279cc3 2144
e02119d5
CM		 2145 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2146 inode, dir->i_ino);
49eb7e46
CM		 2147 BUG_ON(ret != 0 && ret != -ENOENT);
2148 if (ret != -ENOENT)
2149 BTRFS_I(dir)->log_dirty_trans = trans->transid;
e02119d5
CM		 2150
2151 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2152 dir, index);
2153 BUG_ON(ret);
39279cc3
CM		 2154err:
2155 btrfs_free_path(path);
e02119d5
CM		 2156 if (ret)
2157 goto out;
2158
2159 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2160 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2161 btrfs_update_inode(trans, root, dir);
2162 btrfs_drop_nlink(inode);
2163 ret = btrfs_update_inode(trans, root, inode);
2164 dir->i_sb->s_dirt = 1;
2165out:
39279cc3
CM		 2166 return ret;
2167}
2168
2169static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2170{
2171 struct btrfs_root *root;
2172 struct btrfs_trans_handle *trans;
7b128766 2173 struct inode *inode = dentry->d_inode;
39279cc3 2174 int ret;
1832a6d5 2175 unsigned long nr = 0;
39279cc3
CM		 2176
2177 root = BTRFS_I(dir)->root;
1832a6d5
CM		 2178
2179 ret = btrfs_check_free_space(root, 1, 1);
2180 if (ret)
2181 goto fail;
2182
39279cc3 2183 trans = btrfs_start_transaction(root, 1);
5f39d397 2184
39279cc3 2185 btrfs_set_trans_block_group(trans, dir);
e02119d5
CM		 2186 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2187 dentry->d_name.name, dentry->d_name.len);
7b128766
JB		 2188
2189 if (inode->i_nlink == 0)
2190 ret = btrfs_orphan_add(trans, inode);
2191
d3c2fdcf 2192 nr = trans->blocks_used;
5f39d397 2193
89ce8a63 2194 btrfs_end_transaction_throttle(trans, root);
1832a6d5 2195fail:
d3c2fdcf 2196 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2197 return ret;
2198}
2199
2200static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2201{
2202 struct inode *inode = dentry->d_inode;
1832a6d5 2203 int err = 0;
39279cc3
CM		 2204 int ret;
2205 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2206 struct btrfs_trans_handle *trans;
1832a6d5 2207 unsigned long nr = 0;
39279cc3 2208
3394e160
CM		 2209 /*
2210 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2211 * the root of a subvolume or snapshot
2212 */
2213 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2214 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
134d4512 2215 return -ENOTEMPTY;
925baedd 2216 }
134d4512 2217
1832a6d5
CM		 2218 ret = btrfs_check_free_space(root, 1, 1);
2219 if (ret)
2220 goto fail;
2221
39279cc3
CM		 2222 trans = btrfs_start_transaction(root, 1);
2223 btrfs_set_trans_block_group(trans, dir);
39279cc3 2224
7b128766
JB		 2225 err = btrfs_orphan_add(trans, inode);
2226 if (err)
2227 goto fail_trans;
2228
39279cc3 2229 /* now the directory is empty */
e02119d5
CM		 2230 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2231 dentry->d_name.name, dentry->d_name.len);
39279cc3 2232 if (!err) {
dbe674a9 2233 btrfs_i_size_write(inode, 0);
39279cc3 2234 }
3954401f 2235
7b128766 2236fail_trans:
d3c2fdcf 2237 nr = trans->blocks_used;
89ce8a63 2238 ret = btrfs_end_transaction_throttle(trans, root);
1832a6d5 2239fail:
d3c2fdcf 2240 btrfs_btree_balance_dirty(root, nr);
3954401f 2241
39279cc3
CM		 2242 if (ret && !err)
2243 err = ret;
2244 return err;
2245}
2246
d20f7043 2247#if 0
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
d20f7043
CM		 2415#endif
2416
39279cc3
CM		 2417/*
2418 * this can truncate away extent items, csum items and directory items.
2419 * It starts at a high offset and removes keys until it can't find
d352ac68 2420 * any higher than new_size
39279cc3
CM		 2421 *
2422 * csum items that cross the new i_size are truncated to the new size
2423 * as well.
7b128766
JB		 2424 *
2425 * min_type is the minimum key type to truncate down to. If set to 0, this
2426 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2427 */
e02119d5
CM		 2428noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2429 struct btrfs_root *root,
2430 struct inode *inode,
2431 u64 new_size, u32 min_type)
39279cc3
CM		 2432{
2433 int ret;
2434 struct btrfs_path *path;
2435 struct btrfs_key key;
5f39d397 2436 struct btrfs_key found_key;
39279cc3 2437 u32 found_type;
5f39d397 2438 struct extent_buffer *leaf;
39279cc3
CM		 2439 struct btrfs_file_extent_item *fi;
2440 u64 extent_start = 0;
db94535d 2441 u64 extent_num_bytes = 0;
39279cc3 2442 u64 item_end = 0;
7bb86316 2443 u64 root_gen = 0;
d8d5f3e1 2444 u64 root_owner = 0;
39279cc3
CM		 2445 int found_extent;
2446 int del_item;
85e21bac
CM		 2447 int pending_del_nr = 0;
2448 int pending_del_slot = 0;
179e29e4 2449 int extent_type = -1;
771ed689 2450 int encoding;
3b951516 2451 u64 mask = root->sectorsize - 1;
39279cc3 2452
e02119d5 2453 if (root->ref_cows)
5b21f2ed 2454 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
39279cc3 2455 path = btrfs_alloc_path();
3c69faec 2456 path->reada = -1;
39279cc3 2457 BUG_ON(!path);
5f39d397 2458
39279cc3
CM		 2459 /* FIXME, add redo link to tree so we don't leak on crash */
2460 key.objectid = inode->i_ino;
2461 key.offset = (u64)-1;
5f39d397
CM		 2462 key.type = (u8)-1;
2463
85e21bac 2464 btrfs_init_path(path);
323ac95b 2465
85e21bac
CM		 2466search_again:
2467 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2468 if (ret < 0) {
2469 goto error;
2470 }
2471 if (ret > 0) {
e02119d5
CM		 2472 /* there are no items in the tree for us to truncate, we're
2473 * done
2474 */
2475 if (path->slots[0] == 0) {
2476 ret = 0;
2477 goto error;
2478 }
85e21bac
CM		 2479 path->slots[0]--;
2480 }
2481
39279cc3 2482 while(1) {
39279cc3 2483 fi = NULL;
5f39d397
CM		 2484 leaf = path->nodes[0];
2485 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2486 found_type = btrfs_key_type(&found_key);
771ed689 2487 encoding = 0;
39279cc3 2488
5f39d397 2489 if (found_key.objectid != inode->i_ino)
39279cc3 2490 break;
5f39d397 2491
85e21bac 2492 if (found_type < min_type)
39279cc3
CM		 2493 break;
2494
5f39d397 2495 item_end = found_key.offset;
39279cc3 2496 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 2497 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 2498 struct btrfs_file_extent_item);
179e29e4 2499 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM		 2500 encoding = btrfs_file_extent_compression(leaf, fi);
2501 encoding |= btrfs_file_extent_encryption(leaf, fi);
2502 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2503
179e29e4 2504 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 2505 item_end +=
db94535d 2506 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 2507 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 2508 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 2509 fi);
39279cc3 2510 }
008630c1 2511 item_end--;
39279cc3 2512 }
e02119d5 2513 if (item_end < new_size) {
b888db2b
CM		 2514 if (found_type == BTRFS_DIR_ITEM_KEY) {
2515 found_type = BTRFS_INODE_ITEM_KEY;
2516 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
d20f7043 2517 found_type = BTRFS_EXTENT_DATA_KEY;
85e21bac
CM		 2518 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2519 found_type = BTRFS_XATTR_ITEM_KEY;
2520 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2521 found_type = BTRFS_INODE_REF_KEY;
b888db2b
CM		 2522 } else if (found_type) {
2523 found_type--;
2524 } else {
2525 break;
39279cc3 2526 }
a61721d5 2527 btrfs_set_key_type(&key, found_type);
85e21bac 2528 goto next;
39279cc3 2529 }
e02119d5 2530 if (found_key.offset >= new_size)
39279cc3
CM		 2531 del_item = 1;
2532 else
2533 del_item = 0;
2534 found_extent = 0;
2535
2536 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 2537 if (found_type != BTRFS_EXTENT_DATA_KEY)
2538 goto delete;
2539
2540 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 2541 u64 num_dec;
db94535d 2542 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 2543 if (!del_item && !encoding) {
db94535d
CM		 2544 u64 orig_num_bytes =
2545 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 2546 extent_num_bytes = new_size -
5f39d397 2547 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 2548 extent_num_bytes = extent_num_bytes &
2549 ~((u64)root->sectorsize - 1);
db94535d
CM		 2550 btrfs_set_file_extent_num_bytes(leaf, fi,
2551 extent_num_bytes);
2552 num_dec = (orig_num_bytes -
9069218d 2553 extent_num_bytes);
e02119d5 2554 if (root->ref_cows && extent_start != 0)
a76a3cd4 2555 inode_sub_bytes(inode, num_dec);
5f39d397 2556 btrfs_mark_buffer_dirty(leaf);
39279cc3 2557 } else {
db94535d
CM		 2558 extent_num_bytes =
2559 btrfs_file_extent_disk_num_bytes(leaf,
2560 fi);
39279cc3 2561 /* FIXME blocksize != 4096 */
9069218d 2562 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 2563 if (extent_start != 0) {
2564 found_extent = 1;
e02119d5 2565 if (root->ref_cows)
a76a3cd4 2566 inode_sub_bytes(inode, num_dec);
e02119d5 2567 }
31840ae1 2568 root_gen = btrfs_header_generation(leaf);
d8d5f3e1 2569 root_owner = btrfs_header_owner(leaf);
39279cc3 2570 }
9069218d 2571 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 2572 /*
2573 * we can't truncate inline items that have had
2574 * special encodings
2575 */
2576 if (!del_item &&
2577 btrfs_file_extent_compression(leaf, fi) == 0 &&
2578 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2579 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 2580 u32 size = new_size - found_key.offset;
2581
2582 if (root->ref_cows) {
a76a3cd4
YZ		 2583 inode_sub_bytes(inode, item_end + 1 -
2584 new_size);
e02119d5
CM		 2585 }
2586 size =
2587 btrfs_file_extent_calc_inline_size(size);
9069218d 2588 ret = btrfs_truncate_item(trans, root, path,
e02119d5 2589 size, 1);
9069218d 2590 BUG_ON(ret);
e02119d5 2591 } else if (root->ref_cows) {
a76a3cd4
YZ		 2592 inode_sub_bytes(inode, item_end + 1 -
2593 found_key.offset);
9069218d 2594 }
39279cc3 2595 }
179e29e4 2596delete:
39279cc3 2597 if (del_item) {
85e21bac
CM		 2598 if (!pending_del_nr) {
2599 /* no pending yet, add ourselves */
2600 pending_del_slot = path->slots[0];
2601 pending_del_nr = 1;
2602 } else if (pending_del_nr &&
2603 path->slots[0] + 1 == pending_del_slot) {
2604 /* hop on the pending chunk */
2605 pending_del_nr++;
2606 pending_del_slot = path->slots[0];
2607 } else {
2608 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2609 }
39279cc3
CM		 2610 } else {
2611 break;
2612 }
39279cc3
CM		 2613 if (found_extent) {
2614 ret = btrfs_free_extent(trans, root, extent_start,
7bb86316 2615 extent_num_bytes,
31840ae1 2616 leaf->start, root_owner,
3bb1a1bc 2617 root_gen, inode->i_ino, 0);
39279cc3
CM		 2618 BUG_ON(ret);
2619 }
85e21bac
CM		 2620next:
2621 if (path->slots[0] == 0) {
2622 if (pending_del_nr)
2623 goto del_pending;
2624 btrfs_release_path(root, path);
2625 goto search_again;
2626 }
2627
2628 path->slots[0]--;
2629 if (pending_del_nr &&
2630 path->slots[0] + 1 != pending_del_slot) {
2631 struct btrfs_key debug;
2632del_pending:
2633 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2634 pending_del_slot);
2635 ret = btrfs_del_items(trans, root, path,
2636 pending_del_slot,
2637 pending_del_nr);
2638 BUG_ON(ret);
2639 pending_del_nr = 0;
2640 btrfs_release_path(root, path);
2641 goto search_again;
2642 }
39279cc3
CM		 2643 }
2644 ret = 0;
2645error:
85e21bac
CM		 2646 if (pending_del_nr) {
2647 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2648 pending_del_nr);
2649 }
39279cc3
CM		 2650 btrfs_free_path(path);
2651 inode->i_sb->s_dirt = 1;
2652 return ret;
2653}
2654
2655/*
2656 * taken from block_truncate_page, but does cow as it zeros out
2657 * any bytes left in the last page in the file.
2658 */
2659static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2660{
2661 struct inode *inode = mapping->host;
db94535d 2662 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 2663 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2664 struct btrfs_ordered_extent *ordered;
2665 char *kaddr;
db94535d 2666 u32 blocksize = root->sectorsize;
39279cc3
CM		 2667 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2668 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2669 struct page *page;
39279cc3 2670 int ret = 0;
a52d9a80 2671 u64 page_start;
e6dcd2dc 2672 u64 page_end;
39279cc3
CM		 2673
2674 if ((offset & (blocksize - 1)) == 0)
2675 goto out;
2676
2677 ret = -ENOMEM;
211c17f5 2678again:
39279cc3
CM		 2679 page = grab_cache_page(mapping, index);
2680 if (!page)
2681 goto out;
e6dcd2dc
CM		 2682
2683 page_start = page_offset(page);
2684 page_end = page_start + PAGE_CACHE_SIZE - 1;
2685
39279cc3 2686 if (!PageUptodate(page)) {
9ebefb18 2687 ret = btrfs_readpage(NULL, page);
39279cc3 2688 lock_page(page);
211c17f5
CM		 2689 if (page->mapping != mapping) {
2690 unlock_page(page);
2691 page_cache_release(page);
2692 goto again;
2693 }
39279cc3
CM		 2694 if (!PageUptodate(page)) {
2695 ret = -EIO;
89642229 2696 goto out_unlock;
39279cc3
CM		 2697 }
2698 }
211c17f5 2699 wait_on_page_writeback(page);
e6dcd2dc
CM		 2700
2701 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2702 set_page_extent_mapped(page);
2703
2704 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2705 if (ordered) {
2706 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2707 unlock_page(page);
2708 page_cache_release(page);
eb84ae03 2709 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 2710 btrfs_put_ordered_extent(ordered);
2711 goto again;
2712 }
2713
ea8c2819 2714 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc
CM		 2715 ret = 0;
2716 if (offset != PAGE_CACHE_SIZE) {
2717 kaddr = kmap(page);
2718 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2719 flush_dcache_page(page);
2720 kunmap(page);
2721 }
247e743c 2722 ClearPageChecked(page);
e6dcd2dc
CM		 2723 set_page_dirty(page);
2724 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
39279cc3 2725
89642229 2726out_unlock:
39279cc3
CM		 2727 unlock_page(page);
2728 page_cache_release(page);
2729out:
2730 return ret;
2731}
2732
9036c102 2733int btrfs_cont_expand(struct inode *inode, loff_t size)
39279cc3 2734{
9036c102
YZ		 2735 struct btrfs_trans_handle *trans;
2736 struct btrfs_root *root = BTRFS_I(inode)->root;
2737 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2738 struct extent_map *em;
2739 u64 mask = root->sectorsize - 1;
2740 u64 hole_start = (inode->i_size + mask) & ~mask;
2741 u64 block_end = (size + mask) & ~mask;
2742 u64 last_byte;
2743 u64 cur_offset;
2744 u64 hole_size;
39279cc3
CM		 2745 int err;
2746
9036c102
YZ		 2747 if (size <= hole_start)
2748 return 0;
2749
2750 err = btrfs_check_free_space(root, 1, 0);
39279cc3
CM		 2751 if (err)
2752 return err;
2753
9036c102 2754 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2bf5a725 2755
9036c102
YZ		 2756 while (1) {
2757 struct btrfs_ordered_extent *ordered;
2758 btrfs_wait_ordered_range(inode, hole_start,
2759 block_end - hole_start);
2760 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2761 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2762 if (!ordered)
2763 break;
2764 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2765 btrfs_put_ordered_extent(ordered);
2766 }
39279cc3 2767
9036c102
YZ		 2768 trans = btrfs_start_transaction(root, 1);
2769 btrfs_set_trans_block_group(trans, inode);
39279cc3 2770
9036c102
YZ		 2771 cur_offset = hole_start;
2772 while (1) {
2773 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2774 block_end - cur_offset, 0);
2775 BUG_ON(IS_ERR(em) || !em);
2776 last_byte = min(extent_map_end(em), block_end);
2777 last_byte = (last_byte + mask) & ~mask;
2778 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
771ed689 2779 u64 hint_byte = 0;
9036c102 2780 hole_size = last_byte - cur_offset;
771ed689
CM		 2781 err = btrfs_drop_extents(trans, root, inode,
2782 cur_offset,
2783 cur_offset + hole_size,
2784 cur_offset, &hint_byte);
2785 if (err)
2786 break;
9036c102
YZ		 2787 err = btrfs_insert_file_extent(trans, root,
2788 inode->i_ino, cur_offset, 0,
2789 0, hole_size, 0, hole_size,
2790 0, 0, 0);
2791 btrfs_drop_extent_cache(inode, hole_start,
2792 last_byte - 1, 0);
2793 }
2794 free_extent_map(em);
2795 cur_offset = last_byte;
2796 if (err || cur_offset >= block_end)
2797 break;
2798 }
1832a6d5 2799
9036c102
YZ		 2800 btrfs_end_transaction(trans, root);
2801 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2802 return err;
2803}
39279cc3 2804
9036c102
YZ		 2805static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2806{
2807 struct inode *inode = dentry->d_inode;
2808 int err;
39279cc3 2809
9036c102
YZ		 2810 err = inode_change_ok(inode, attr);
2811 if (err)
2812 return err;
2bf5a725 2813
9036c102
YZ		 2814 if (S_ISREG(inode->i_mode) &&
2815 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2816 err = btrfs_cont_expand(inode, attr->ia_size);
54aa1f4d
CM		 2817 if (err)
2818 return err;
39279cc3 2819 }
9036c102 2820
39279cc3 2821 err = inode_setattr(inode, attr);
33268eaf
JB		 2822
2823 if (!err && ((attr->ia_valid & ATTR_MODE)))
2824 err = btrfs_acl_chmod(inode);
39279cc3
CM		 2825 return err;
2826}
61295eb8 2827
39279cc3
CM		 2828void btrfs_delete_inode(struct inode *inode)
2829{
2830 struct btrfs_trans_handle *trans;
2831 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 2832 unsigned long nr;
39279cc3
CM		 2833 int ret;
2834
2835 truncate_inode_pages(&inode->i_data, 0);
2836 if (is_bad_inode(inode)) {
7b128766 2837 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 2838 goto no_delete;
2839 }
4a096752 2840 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 2841
dbe674a9 2842 btrfs_i_size_write(inode, 0);
39279cc3 2843 trans = btrfs_start_transaction(root, 1);
5f39d397 2844
39279cc3 2845 btrfs_set_trans_block_group(trans, inode);
e02119d5 2846 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
7b128766
JB		 2847 if (ret) {
2848 btrfs_orphan_del(NULL, inode);
54aa1f4d 2849 goto no_delete_lock;
7b128766
JB		 2850 }
2851
2852 btrfs_orphan_del(trans, inode);
85e21bac 2853
d3c2fdcf 2854 nr = trans->blocks_used;
85e21bac 2855 clear_inode(inode);
5f39d397 2856
39279cc3 2857 btrfs_end_transaction(trans, root);
d3c2fdcf 2858 btrfs_btree_balance_dirty(root, nr);
39279cc3 2859 return;
54aa1f4d
CM		 2860
2861no_delete_lock:
d3c2fdcf 2862 nr = trans->blocks_used;
54aa1f4d 2863 btrfs_end_transaction(trans, root);
d3c2fdcf 2864 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2865no_delete:
2866 clear_inode(inode);
2867}
2868
2869/*
2870 * this returns the key found in the dir entry in the location pointer.
2871 * If no dir entries were found, location->objectid is 0.
2872 */
2873static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2874 struct btrfs_key *location)
2875{
2876 const char *name = dentry->d_name.name;
2877 int namelen = dentry->d_name.len;
2878 struct btrfs_dir_item *di;
2879 struct btrfs_path *path;
2880 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 2881 int ret = 0;
39279cc3
CM		 2882
2883 path = btrfs_alloc_path();
2884 BUG_ON(!path);
3954401f 2885
39279cc3
CM		 2886 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2887 namelen, 0);
0d9f7f3e
Y		 2888 if (IS_ERR(di))
2889 ret = PTR_ERR(di);
39279cc3 2890 if (!di || IS_ERR(di)) {
3954401f 2891 goto out_err;
39279cc3 2892 }
5f39d397 2893 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 2894out:
39279cc3
CM		 2895 btrfs_free_path(path);
2896 return ret;
3954401f
CM		 2897out_err:
2898 location->objectid = 0;
2899 goto out;
39279cc3
CM		 2900}
2901
2902/*
2903 * when we hit a tree root in a directory, the btrfs part of the inode
2904 * needs to be changed to reflect the root directory of the tree root. This
2905 * is kind of like crossing a mount point.
2906 */
2907static int fixup_tree_root_location(struct btrfs_root *root,
2908 struct btrfs_key *location,
58176a96
JB		 2909 struct btrfs_root **sub_root,
2910 struct dentry *dentry)
39279cc3 2911{
39279cc3
CM		 2912 struct btrfs_root_item *ri;
2913
2914 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2915 return 0;
2916 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2917 return 0;
2918
58176a96
JB		 2919 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2920 dentry->d_name.name,
2921 dentry->d_name.len);
39279cc3
CM		 2922 if (IS_ERR(*sub_root))
2923 return PTR_ERR(*sub_root);
2924
2925 ri = &(*sub_root)->root_item;
2926 location->objectid = btrfs_root_dirid(ri);
39279cc3
CM		 2927 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2928 location->offset = 0;
2929
39279cc3
CM		 2930 return 0;
2931}
2932
e02119d5 2933static noinline void init_btrfs_i(struct inode *inode)
39279cc3 2934{
e02119d5
CM		 2935 struct btrfs_inode *bi = BTRFS_I(inode);
2936
2937 bi->i_acl = NULL;
2938 bi->i_default_acl = NULL;
2939
2940 bi->generation = 0;
c3027eb5 2941 bi->sequence = 0;
e02119d5
CM		 2942 bi->last_trans = 0;
2943 bi->logged_trans = 0;
2944 bi->delalloc_bytes = 0;
2945 bi->disk_i_size = 0;
2946 bi->flags = 0;
2947 bi->index_cnt = (u64)-1;
49eb7e46 2948 bi->log_dirty_trans = 0;
d1310b2e
CM		 2949 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2950 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
b888db2b 2951 inode->i_mapping, GFP_NOFS);
7e38326f
CM		 2952 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2953 inode->i_mapping, GFP_NOFS);
ea8c2819 2954 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
ba1da2f4 2955 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
ee6e6504 2956 mutex_init(&BTRFS_I(inode)->extent_mutex);
e02119d5
CM		 2957 mutex_init(&BTRFS_I(inode)->log_mutex);
2958}
2959
2960static int btrfs_init_locked_inode(struct inode *inode, void *p)
2961{
2962 struct btrfs_iget_args *args = p;
2963 inode->i_ino = args->ino;
2964 init_btrfs_i(inode);
2965 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 2966 return 0;
2967}
2968
2969static int btrfs_find_actor(struct inode *inode, void *opaque)
2970{
2971 struct btrfs_iget_args *args = opaque;
2972 return (args->ino == inode->i_ino &&
2973 args->root == BTRFS_I(inode)->root);
2974}
2975
5b21f2ed
ZY		 2976struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2977 struct btrfs_root *root, int wait)
2978{
2979 struct inode *inode;
2980 struct btrfs_iget_args args;
2981 args.ino = objectid;
2982 args.root = root;
2983
2984 if (wait) {
2985 inode = ilookup5(s, objectid, btrfs_find_actor,
2986 (void *)&args);
2987 } else {
2988 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
2989 (void *)&args);
2990 }
2991 return inode;
2992}
2993
39279cc3
CM		 2994struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
2995 struct btrfs_root *root)
2996{
2997 struct inode *inode;
2998 struct btrfs_iget_args args;
2999 args.ino = objectid;
3000 args.root = root;
3001
3002 inode = iget5_locked(s, objectid, btrfs_find_actor,
3003 btrfs_init_locked_inode,
3004 (void *)&args);
3005 return inode;
3006}
3007
1a54ef8c
BR		 3008/* Get an inode object given its location and corresponding root.
3009 * Returns in *is_new if the inode was read from disk
3010 */
3011struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3012 struct btrfs_root *root, int *is_new)
3013{
3014 struct inode *inode;
3015
3016 inode = btrfs_iget_locked(s, location->objectid, root);
3017 if (!inode)
3018 return ERR_PTR(-EACCES);
3019
3020 if (inode->i_state & I_NEW) {
3021 BTRFS_I(inode)->root = root;
3022 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3023 btrfs_read_locked_inode(inode);
3024 unlock_new_inode(inode);
3025 if (is_new)
3026 *is_new = 1;
3027 } else {
3028 if (is_new)
3029 *is_new = 0;
3030 }
3031
3032 return inode;
3033}
3034
3de4586c 3035struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3
CM		 3036{
3037 struct inode * inode;
3038 struct btrfs_inode *bi = BTRFS_I(dir);
3039 struct btrfs_root *root = bi->root;
3040 struct btrfs_root *sub_root = root;
3041 struct btrfs_key location;
c146afad 3042 int ret, new;
39279cc3
CM		 3043
3044 if (dentry->d_name.len > BTRFS_NAME_LEN)
3045 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3046
39279cc3 3047 ret = btrfs_inode_by_name(dir, dentry, &location);
5f39d397 3048
39279cc3
CM		 3049 if (ret < 0)
3050 return ERR_PTR(ret);
5f39d397 3051
39279cc3
CM		 3052 inode = NULL;
3053 if (location.objectid) {
58176a96
JB		 3054 ret = fixup_tree_root_location(root, &location, &sub_root,
3055 dentry);
39279cc3
CM		 3056 if (ret < 0)
3057 return ERR_PTR(ret);
3058 if (ret > 0)
3059 return ERR_PTR(-ENOENT);
1a54ef8c
BR		 3060 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3061 if (IS_ERR(inode))
3062 return ERR_CAST(inode);
39279cc3 3063 }
3de4586c
CM		 3064 return inode;
3065}
3066
3067static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3068 struct nameidata *nd)
3069{
3070 struct inode *inode;
3071
3072 if (dentry->d_name.len > BTRFS_NAME_LEN)
3073 return ERR_PTR(-ENAMETOOLONG);
3074
3075 inode = btrfs_lookup_dentry(dir, dentry);
3076 if (IS_ERR(inode))
3077 return ERR_CAST(inode);
7b128766 3078
39279cc3
CM		 3079 return d_splice_alias(inode, dentry);
3080}
3081
39279cc3
CM		 3082static unsigned char btrfs_filetype_table[] = {
3083 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3084};
3085
cbdf5a24
DW		 3086static int btrfs_real_readdir(struct file *filp, void *dirent,
3087 filldir_t filldir)
39279cc3 3088{
6da6abae 3089 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 3090 struct btrfs_root *root = BTRFS_I(inode)->root;
3091 struct btrfs_item *item;
3092 struct btrfs_dir_item *di;
3093 struct btrfs_key key;
5f39d397 3094 struct btrfs_key found_key;
39279cc3
CM		 3095 struct btrfs_path *path;
3096 int ret;
3097 u32 nritems;
5f39d397 3098 struct extent_buffer *leaf;
39279cc3
CM		 3099 int slot;
3100 int advance;
3101 unsigned char d_type;
3102 int over = 0;
3103 u32 di_cur;
3104 u32 di_total;
3105 u32 di_len;
3106 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 3107 char tmp_name[32];
3108 char *name_ptr;
3109 int name_len;
39279cc3
CM		 3110
3111 /* FIXME, use a real flag for deciding about the key type */
3112 if (root->fs_info->tree_root == root)
3113 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 3114
3954401f
CM		 3115 /* special case for "." */
3116 if (filp->f_pos == 0) {
3117 over = filldir(dirent, ".", 1,
3118 1, inode->i_ino,
3119 DT_DIR);
3120 if (over)
3121 return 0;
3122 filp->f_pos = 1;
3123 }
3954401f
CM		 3124 /* special case for .., just use the back ref */
3125 if (filp->f_pos == 1) {
5ecc7e5d 3126 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 3127 over = filldir(dirent, "..", 2,
5ecc7e5d 3128 2, pino, DT_DIR);
3954401f 3129 if (over)
49593bfa 3130 return 0;
3954401f
CM		 3131 filp->f_pos = 2;
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 }
3de4586c 3162
39279cc3 3163 advance = 1;
5f39d397
CM		 3164 item = btrfs_item_nr(leaf, slot);
3165 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3166
3167 if (found_key.objectid != key.objectid)
39279cc3 3168 break;
5f39d397 3169 if (btrfs_key_type(&found_key) != key_type)
39279cc3 3170 break;
5f39d397 3171 if (found_key.offset < filp->f_pos)
39279cc3 3172 continue;
5f39d397
CM		 3173
3174 filp->f_pos = found_key.offset;
49593bfa 3175
39279cc3
CM		 3176 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3177 di_cur = 0;
5f39d397 3178 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 3179
3180 while (di_cur < di_total) {
5f39d397
CM		 3181 struct btrfs_key location;
3182
3183 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 3184 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 3185 name_ptr = tmp_name;
3186 } else {
3187 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 3188 if (!name_ptr) {
3189 ret = -ENOMEM;
3190 goto err;
3191 }
5f39d397
CM		 3192 }
3193 read_extent_buffer(leaf, name_ptr,
3194 (unsigned long)(di + 1), name_len);
3195
3196 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3197 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c
CM		 3198
3199 /* is this a reference to our own snapshot? If so
3200 * skip it
3201 */
3202 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3203 location.objectid == root->root_key.objectid) {
3204 over = 0;
3205 goto skip;
3206 }
5f39d397 3207 over = filldir(dirent, name_ptr, name_len,
49593bfa 3208 found_key.offset, location.objectid,
39279cc3 3209 d_type);
5f39d397 3210
3de4586c 3211skip:
5f39d397
CM		 3212 if (name_ptr != tmp_name)
3213 kfree(name_ptr);
3214
39279cc3
CM		 3215 if (over)
3216 goto nopos;
5103e947 3217 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 3218 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 3219 di_cur += di_len;
3220 di = (struct btrfs_dir_item *)((char *)di + di_len);
3221 }
3222 }
49593bfa
DW		 3223
3224 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07
YZ		 3225 if (key_type == BTRFS_DIR_INDEX_KEY)
3226 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3227 else
3228 filp->f_pos++;
39279cc3
CM		 3229nopos:
3230 ret = 0;
3231err:
39279cc3 3232 btrfs_free_path(path);
39279cc3
CM		 3233 return ret;
3234}
3235
3236int btrfs_write_inode(struct inode *inode, int wait)
3237{
3238 struct btrfs_root *root = BTRFS_I(inode)->root;
3239 struct btrfs_trans_handle *trans;
3240 int ret = 0;
3241
c146afad 3242 if (root->fs_info->btree_inode == inode)
4ca8b41e
CM		 3243 return 0;
3244
39279cc3 3245 if (wait) {
f9295749 3246 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3247 btrfs_set_trans_block_group(trans, inode);
3248 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 3249 }
3250 return ret;
3251}
3252
3253/*
54aa1f4d 3254 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 3255 * inode changes. But, it is most likely to find the inode in cache.
3256 * FIXME, needs more benchmarking...there are no reasons other than performance
3257 * to keep or drop this code.
3258 */
3259void btrfs_dirty_inode(struct inode *inode)
3260{
3261 struct btrfs_root *root = BTRFS_I(inode)->root;
3262 struct btrfs_trans_handle *trans;
3263
f9295749 3264 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3265 btrfs_set_trans_block_group(trans, inode);
3266 btrfs_update_inode(trans, root, inode);
3267 btrfs_end_transaction(trans, root);
39279cc3
CM		 3268}
3269
d352ac68
CM		 3270/*
3271 * find the highest existing sequence number in a directory
3272 * and then set the in-memory index_cnt variable to reflect
3273 * free sequence numbers
3274 */
aec7477b
JB		 3275static int btrfs_set_inode_index_count(struct inode *inode)
3276{
3277 struct btrfs_root *root = BTRFS_I(inode)->root;
3278 struct btrfs_key key, found_key;
3279 struct btrfs_path *path;
3280 struct extent_buffer *leaf;
3281 int ret;
3282
3283 key.objectid = inode->i_ino;
3284 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3285 key.offset = (u64)-1;
3286
3287 path = btrfs_alloc_path();
3288 if (!path)
3289 return -ENOMEM;
3290
3291 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3292 if (ret < 0)
3293 goto out;
3294 /* FIXME: we should be able to handle this */
3295 if (ret == 0)
3296 goto out;
3297 ret = 0;
3298
3299 /*
3300 * MAGIC NUMBER EXPLANATION:
3301 * since we search a directory based on f_pos we have to start at 2
3302 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3303 * else has to start at 2
3304 */
3305 if (path->slots[0] == 0) {
3306 BTRFS_I(inode)->index_cnt = 2;
3307 goto out;
3308 }
3309
3310 path->slots[0]--;
3311
3312 leaf = path->nodes[0];
3313 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3314
3315 if (found_key.objectid != inode->i_ino ||
3316 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3317 BTRFS_I(inode)->index_cnt = 2;
3318 goto out;
3319 }
3320
3321 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3322out:
3323 btrfs_free_path(path);
3324 return ret;
3325}
3326
d352ac68
CM		 3327/*
3328 * helper to find a free sequence number in a given directory. This current
3329 * code is very simple, later versions will do smarter things in the btree
3330 */
3de4586c 3331int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 3332{
3333 int ret = 0;
3334
3335 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3336 ret = btrfs_set_inode_index_count(dir);
8d5bf1cb 3337 if (ret) {
aec7477b 3338 return ret;
8d5bf1cb 3339 }
aec7477b
JB		 3340 }
3341
00e4e6b3 3342 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 3343 BTRFS_I(dir)->index_cnt++;
3344
3345 return ret;
3346}
3347
39279cc3
CM		 3348static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3349 struct btrfs_root *root,
aec7477b 3350 struct inode *dir,
9c58309d 3351 const char *name, int name_len,
d2fb3437
YZ		 3352 u64 ref_objectid, u64 objectid,
3353 u64 alloc_hint, int mode, u64 *index)
39279cc3
CM		 3354{
3355 struct inode *inode;
5f39d397 3356 struct btrfs_inode_item *inode_item;
39279cc3 3357 struct btrfs_key *location;
5f39d397 3358 struct btrfs_path *path;
9c58309d
CM		 3359 struct btrfs_inode_ref *ref;
3360 struct btrfs_key key[2];
3361 u32 sizes[2];
3362 unsigned long ptr;
39279cc3
CM		 3363 int ret;
3364 int owner;
3365
5f39d397
CM		 3366 path = btrfs_alloc_path();
3367 BUG_ON(!path);
3368
39279cc3
CM		 3369 inode = new_inode(root->fs_info->sb);
3370 if (!inode)
3371 return ERR_PTR(-ENOMEM);
3372
aec7477b 3373 if (dir) {
3de4586c 3374 ret = btrfs_set_inode_index(dir, index);
aec7477b
JB		 3375 if (ret)
3376 return ERR_PTR(ret);
aec7477b
JB		 3377 }
3378 /*
3379 * index_cnt is ignored for everything but a dir,
3380 * btrfs_get_inode_index_count has an explanation for the magic
3381 * number
3382 */
e02119d5 3383 init_btrfs_i(inode);
aec7477b 3384 BTRFS_I(inode)->index_cnt = 2;
39279cc3 3385 BTRFS_I(inode)->root = root;
e02119d5 3386 BTRFS_I(inode)->generation = trans->transid;
b888db2b 3387
39279cc3
CM		 3388 if (mode & S_IFDIR)
3389 owner = 0;
3390 else
3391 owner = 1;
d2fb3437
YZ		 3392 BTRFS_I(inode)->block_group =
3393 btrfs_find_block_group(root, 0, alloc_hint, owner);
9c58309d
CM		 3394
3395 key[0].objectid = objectid;
3396 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3397 key[0].offset = 0;
3398
3399 key[1].objectid = objectid;
3400 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3401 key[1].offset = ref_objectid;
3402
3403 sizes[0] = sizeof(struct btrfs_inode_item);
3404 sizes[1] = name_len + sizeof(*ref);
3405
3406 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3407 if (ret != 0)
5f39d397
CM		 3408 goto fail;
3409
9c58309d
CM		 3410 if (objectid > root->highest_inode)
3411 root->highest_inode = objectid;
3412
79683f2d
CM		 3413 inode->i_uid = current_fsuid();
3414 inode->i_gid = current_fsgid();
39279cc3
CM		 3415 inode->i_mode = mode;
3416 inode->i_ino = objectid;
a76a3cd4 3417 inode_set_bytes(inode, 0);
39279cc3 3418 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 3419 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3420 struct btrfs_inode_item);
e02119d5 3421 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 3422
3423 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3424 struct btrfs_inode_ref);
3425 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 3426 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 3427 ptr = (unsigned long)(ref + 1);
3428 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3429
5f39d397
CM		 3430 btrfs_mark_buffer_dirty(path->nodes[0]);
3431 btrfs_free_path(path);
3432
39279cc3
CM		 3433 location = &BTRFS_I(inode)->location;
3434 location->objectid = objectid;
39279cc3
CM		 3435 location->offset = 0;
3436 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3437
39279cc3
CM		 3438 insert_inode_hash(inode);
3439 return inode;
5f39d397 3440fail:
aec7477b
JB		 3441 if (dir)
3442 BTRFS_I(dir)->index_cnt--;
5f39d397
CM		 3443 btrfs_free_path(path);
3444 return ERR_PTR(ret);
39279cc3
CM		 3445}
3446
3447static inline u8 btrfs_inode_type(struct inode *inode)
3448{
3449 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3450}
3451
d352ac68
CM		 3452/*
3453 * utility function to add 'inode' into 'parent_inode' with
3454 * a give name and a given sequence number.
3455 * if 'add_backref' is true, also insert a backref from the
3456 * inode to the parent directory.
3457 */
e02119d5
CM		 3458int btrfs_add_link(struct btrfs_trans_handle *trans,
3459 struct inode *parent_inode, struct inode *inode,
3460 const char *name, int name_len, int add_backref, u64 index)
39279cc3
CM		 3461{
3462 int ret;
3463 struct btrfs_key key;
e02119d5 3464 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5f39d397 3465
39279cc3 3466 key.objectid = inode->i_ino;
39279cc3
CM		 3467 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3468 key.offset = 0;
3469
e02119d5
CM		 3470 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3471 parent_inode->i_ino,
aec7477b 3472 &key, btrfs_inode_type(inode),
00e4e6b3 3473 index);
39279cc3 3474 if (ret == 0) {
9c58309d
CM		 3475 if (add_backref) {
3476 ret = btrfs_insert_inode_ref(trans, root,
e02119d5
CM		 3477 name, name_len,
3478 inode->i_ino,
3479 parent_inode->i_ino,
3480 index);
9c58309d 3481 }
dbe674a9 3482 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 3483 name_len * 2);
79c44584 3484 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 3485 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 3486 }
3487 return ret;
3488}
3489
3490static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
9c58309d 3491 struct dentry *dentry, struct inode *inode,
00e4e6b3 3492 int backref, u64 index)
39279cc3 3493{
e02119d5
CM		 3494 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3495 inode, dentry->d_name.name,
3496 dentry->d_name.len, backref, index);
39279cc3
CM		 3497 if (!err) {
3498 d_instantiate(dentry, inode);
3499 return 0;
3500 }
3501 if (err > 0)
3502 err = -EEXIST;
3503 return err;
3504}
3505
618e21d5
JB		 3506static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3507 int mode, dev_t rdev)
3508{
3509 struct btrfs_trans_handle *trans;
3510 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3511 struct inode *inode = NULL;
618e21d5
JB		 3512 int err;
3513 int drop_inode = 0;
3514 u64 objectid;
1832a6d5 3515 unsigned long nr = 0;
00e4e6b3 3516 u64 index = 0;
618e21d5
JB		 3517
3518 if (!new_valid_dev(rdev))
3519 return -EINVAL;
3520
1832a6d5
CM		 3521 err = btrfs_check_free_space(root, 1, 0);
3522 if (err)
3523 goto fail;
3524
618e21d5
JB		 3525 trans = btrfs_start_transaction(root, 1);
3526 btrfs_set_trans_block_group(trans, dir);
3527
3528 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3529 if (err) {
3530 err = -ENOSPC;
3531 goto out_unlock;
3532 }
3533
aec7477b 3534 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3535 dentry->d_name.len,
3536 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3 3537 BTRFS_I(dir)->block_group, mode, &index);
618e21d5
JB		 3538 err = PTR_ERR(inode);
3539 if (IS_ERR(inode))
3540 goto out_unlock;
3541
33268eaf
JB		 3542 err = btrfs_init_acl(inode, dir);
3543 if (err) {
3544 drop_inode = 1;
3545 goto out_unlock;
3546 }
3547
618e21d5 3548 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3549 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
618e21d5
JB		 3550 if (err)
3551 drop_inode = 1;
3552 else {
3553 inode->i_op = &btrfs_special_inode_operations;
3554 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 3555 btrfs_update_inode(trans, root, inode);
618e21d5
JB		 3556 }
3557 dir->i_sb->s_dirt = 1;
3558 btrfs_update_inode_block_group(trans, inode);
3559 btrfs_update_inode_block_group(trans, dir);
3560out_unlock:
d3c2fdcf 3561 nr = trans->blocks_used;
89ce8a63 3562 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3563fail:
618e21d5
JB		 3564 if (drop_inode) {
3565 inode_dec_link_count(inode);
3566 iput(inode);
3567 }
d3c2fdcf 3568 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 3569 return err;
3570}
3571
39279cc3
CM		 3572static int btrfs_create(struct inode *dir, struct dentry *dentry,
3573 int mode, struct nameidata *nd)
3574{
3575 struct btrfs_trans_handle *trans;
3576 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3577 struct inode *inode = NULL;
39279cc3
CM		 3578 int err;
3579 int drop_inode = 0;
1832a6d5 3580 unsigned long nr = 0;
39279cc3 3581 u64 objectid;
00e4e6b3 3582 u64 index = 0;
39279cc3 3583
1832a6d5
CM		 3584 err = btrfs_check_free_space(root, 1, 0);
3585 if (err)
3586 goto fail;
39279cc3
CM		 3587 trans = btrfs_start_transaction(root, 1);
3588 btrfs_set_trans_block_group(trans, dir);
3589
3590 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3591 if (err) {
3592 err = -ENOSPC;
3593 goto out_unlock;
3594 }
3595
aec7477b 3596 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3597 dentry->d_name.len,
3598 dentry->d_parent->d_inode->i_ino,
00e4e6b3
CM		 3599 objectid, BTRFS_I(dir)->block_group, mode,
3600 &index);
39279cc3
CM		 3601 err = PTR_ERR(inode);
3602 if (IS_ERR(inode))
3603 goto out_unlock;
3604
33268eaf
JB		 3605 err = btrfs_init_acl(inode, dir);
3606 if (err) {
3607 drop_inode = 1;
3608 goto out_unlock;
3609 }
3610
39279cc3 3611 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3612 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 3613 if (err)
3614 drop_inode = 1;
3615 else {
3616 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3617 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 3618 inode->i_fop = &btrfs_file_operations;
3619 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 3620 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3621 }
3622 dir->i_sb->s_dirt = 1;
3623 btrfs_update_inode_block_group(trans, inode);
3624 btrfs_update_inode_block_group(trans, dir);
3625out_unlock:
d3c2fdcf 3626 nr = trans->blocks_used;
ab78c84d 3627 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3628fail:
39279cc3
CM		 3629 if (drop_inode) {
3630 inode_dec_link_count(inode);
3631 iput(inode);
3632 }
d3c2fdcf 3633 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3634 return err;
3635}
3636
3637static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3638 struct dentry *dentry)
3639{
3640 struct btrfs_trans_handle *trans;
3641 struct btrfs_root *root = BTRFS_I(dir)->root;
3642 struct inode *inode = old_dentry->d_inode;
00e4e6b3 3643 u64 index;
1832a6d5 3644 unsigned long nr = 0;
39279cc3
CM		 3645 int err;
3646 int drop_inode = 0;
3647
3648 if (inode->i_nlink == 0)
3649 return -ENOENT;
3650
e02119d5 3651 btrfs_inc_nlink(inode);
1832a6d5
CM		 3652 err = btrfs_check_free_space(root, 1, 0);
3653 if (err)
3654 goto fail;
3de4586c 3655 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 3656 if (err)
3657 goto fail;
3658
39279cc3 3659 trans = btrfs_start_transaction(root, 1);
5f39d397 3660
39279cc3
CM		 3661 btrfs_set_trans_block_group(trans, dir);
3662 atomic_inc(&inode->i_count);
aec7477b 3663
00e4e6b3 3664 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
5f39d397 3665
39279cc3
CM		 3666 if (err)
3667 drop_inode = 1;
5f39d397 3668
39279cc3
CM		 3669 dir->i_sb->s_dirt = 1;
3670 btrfs_update_inode_block_group(trans, dir);
54aa1f4d 3671 err = btrfs_update_inode(trans, root, inode);
5f39d397 3672
54aa1f4d
CM		 3673 if (err)
3674 drop_inode = 1;
39279cc3 3675
d3c2fdcf 3676 nr = trans->blocks_used;
ab78c84d 3677 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3678fail:
39279cc3
CM		 3679 if (drop_inode) {
3680 inode_dec_link_count(inode);
3681 iput(inode);
3682 }
d3c2fdcf 3683 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3684 return err;
3685}
3686
39279cc3
CM		 3687static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3688{
b9d86667 3689 struct inode *inode = NULL;
39279cc3
CM		 3690 struct btrfs_trans_handle *trans;
3691 struct btrfs_root *root = BTRFS_I(dir)->root;
3692 int err = 0;
3693 int drop_on_err = 0;
b9d86667 3694 u64 objectid = 0;
00e4e6b3 3695 u64 index = 0;
d3c2fdcf 3696 unsigned long nr = 1;
39279cc3 3697
1832a6d5
CM		 3698 err = btrfs_check_free_space(root, 1, 0);
3699 if (err)
3700 goto out_unlock;
3701
39279cc3
CM		 3702 trans = btrfs_start_transaction(root, 1);
3703 btrfs_set_trans_block_group(trans, dir);
5f39d397 3704
39279cc3
CM		 3705 if (IS_ERR(trans)) {
3706 err = PTR_ERR(trans);
3707 goto out_unlock;
3708 }
3709
3710 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3711 if (err) {
3712 err = -ENOSPC;
3713 goto out_unlock;
3714 }
3715
aec7477b 3716 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3717 dentry->d_name.len,
3718 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 3719 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3720 &index);
39279cc3
CM		 3721 if (IS_ERR(inode)) {
3722 err = PTR_ERR(inode);
3723 goto out_fail;
3724 }
5f39d397 3725
39279cc3 3726 drop_on_err = 1;
33268eaf
JB		 3727
3728 err = btrfs_init_acl(inode, dir);
3729 if (err)
3730 goto out_fail;
3731
39279cc3
CM		 3732 inode->i_op = &btrfs_dir_inode_operations;
3733 inode->i_fop = &btrfs_dir_file_operations;
3734 btrfs_set_trans_block_group(trans, inode);
3735
dbe674a9 3736 btrfs_i_size_write(inode, 0);
39279cc3
CM		 3737 err = btrfs_update_inode(trans, root, inode);
3738 if (err)
3739 goto out_fail;
5f39d397 3740
e02119d5
CM		 3741 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3742 inode, dentry->d_name.name,
3743 dentry->d_name.len, 0, index);
39279cc3
CM		 3744 if (err)
3745 goto out_fail;
5f39d397 3746
39279cc3
CM		 3747 d_instantiate(dentry, inode);
3748 drop_on_err = 0;
3749 dir->i_sb->s_dirt = 1;
3750 btrfs_update_inode_block_group(trans, inode);
3751 btrfs_update_inode_block_group(trans, dir);
3752
3753out_fail:
d3c2fdcf 3754 nr = trans->blocks_used;
ab78c84d 3755 btrfs_end_transaction_throttle(trans, root);
5f39d397 3756
39279cc3 3757out_unlock:
39279cc3
CM		 3758 if (drop_on_err)
3759 iput(inode);
d3c2fdcf 3760 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3761 return err;
3762}
3763
d352ac68
CM		 3764/* helper for btfs_get_extent. Given an existing extent in the tree,
3765 * and an extent that you want to insert, deal with overlap and insert
3766 * the new extent into the tree.
3767 */
3b951516
CM		 3768static int merge_extent_mapping(struct extent_map_tree *em_tree,
3769 struct extent_map *existing,
e6dcd2dc
CM		 3770 struct extent_map *em,
3771 u64 map_start, u64 map_len)
3b951516
CM		 3772{
3773 u64 start_diff;
3b951516 3774
e6dcd2dc
CM		 3775 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3776 start_diff = map_start - em->start;
3777 em->start = map_start;
3778 em->len = map_len;
c8b97818
CM		 3779 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3780 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 3781 em->block_start += start_diff;
c8b97818
CM		 3782 em->block_len -= start_diff;
3783 }
e6dcd2dc 3784 return add_extent_mapping(em_tree, em);
3b951516
CM		 3785}
3786
c8b97818
CM		 3787static noinline int uncompress_inline(struct btrfs_path *path,
3788 struct inode *inode, struct page *page,
3789 size_t pg_offset, u64 extent_offset,
3790 struct btrfs_file_extent_item *item)
3791{
3792 int ret;
3793 struct extent_buffer *leaf = path->nodes[0];
3794 char *tmp;
3795 size_t max_size;
3796 unsigned long inline_size;
3797 unsigned long ptr;
3798
3799 WARN_ON(pg_offset != 0);
3800 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3801 inline_size = btrfs_file_extent_inline_item_len(leaf,
3802 btrfs_item_nr(leaf, path->slots[0]));
3803 tmp = kmalloc(inline_size, GFP_NOFS);
3804 ptr = btrfs_file_extent_inline_start(item);
3805
3806 read_extent_buffer(leaf, tmp, ptr, inline_size);
3807
5b050f04 3808 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
c8b97818
CM		 3809 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3810 inline_size, max_size);
3811 if (ret) {
3812 char *kaddr = kmap_atomic(page, KM_USER0);
3813 unsigned long copy_size = min_t(u64,
3814 PAGE_CACHE_SIZE - pg_offset,
3815 max_size - extent_offset);
3816 memset(kaddr + pg_offset, 0, copy_size);
3817 kunmap_atomic(kaddr, KM_USER0);
3818 }
3819 kfree(tmp);
3820 return 0;
3821}
3822
d352ac68
CM		 3823/*
3824 * a bit scary, this does extent mapping from logical file offset to the disk.
3825 * the ugly parts come from merging extents from the disk with the
3826 * in-ram representation. This gets more complex because of the data=ordered code,
3827 * where the in-ram extents might be locked pending data=ordered completion.
3828 *
3829 * This also copies inline extents directly into the page.
3830 */
a52d9a80 3831struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 3832 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 3833 int create)
3834{
3835 int ret;
3836 int err = 0;
db94535d 3837 u64 bytenr;
a52d9a80
CM		 3838 u64 extent_start = 0;
3839 u64 extent_end = 0;
3840 u64 objectid = inode->i_ino;
3841 u32 found_type;
f421950f 3842 struct btrfs_path *path = NULL;
a52d9a80
CM		 3843 struct btrfs_root *root = BTRFS_I(inode)->root;
3844 struct btrfs_file_extent_item *item;
5f39d397
CM		 3845 struct extent_buffer *leaf;
3846 struct btrfs_key found_key;
a52d9a80
CM		 3847 struct extent_map *em = NULL;
3848 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 3849 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 3850 struct btrfs_trans_handle *trans = NULL;
c8b97818 3851 int compressed;
a52d9a80 3852
a52d9a80 3853again:
d1310b2e
CM		 3854 spin_lock(&em_tree->lock);
3855 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 3856 if (em)
3857 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 3858 spin_unlock(&em_tree->lock);
3859
a52d9a80 3860 if (em) {
e1c4b745
CM		 3861 if (em->start > start || em->start + em->len <= start)
3862 free_extent_map(em);
3863 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 3864 free_extent_map(em);
3865 else
3866 goto out;
a52d9a80 3867 }
d1310b2e 3868 em = alloc_extent_map(GFP_NOFS);
a52d9a80 3869 if (!em) {
d1310b2e
CM		 3870 err = -ENOMEM;
3871 goto out;
a52d9a80 3872 }
e6dcd2dc 3873 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 3874 em->start = EXTENT_MAP_HOLE;
445a6944 3875 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 3876 em->len = (u64)-1;
c8b97818 3877 em->block_len = (u64)-1;
f421950f
CM		 3878
3879 if (!path) {
3880 path = btrfs_alloc_path();
3881 BUG_ON(!path);
3882 }
3883
179e29e4
CM		 3884 ret = btrfs_lookup_file_extent(trans, root, path,
3885 objectid, start, trans != NULL);
a52d9a80
CM		 3886 if (ret < 0) {
3887 err = ret;
3888 goto out;
3889 }
3890
3891 if (ret != 0) {
3892 if (path->slots[0] == 0)
3893 goto not_found;
3894 path->slots[0]--;
3895 }
3896
5f39d397
CM		 3897 leaf = path->nodes[0];
3898 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 3899 struct btrfs_file_extent_item);
a52d9a80 3900 /* are we inside the extent that was found? */
5f39d397
CM		 3901 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3902 found_type = btrfs_key_type(&found_key);
3903 if (found_key.objectid != objectid ||
a52d9a80
CM		 3904 found_type != BTRFS_EXTENT_DATA_KEY) {
3905 goto not_found;
3906 }
3907
5f39d397
CM		 3908 found_type = btrfs_file_extent_type(leaf, item);
3909 extent_start = found_key.offset;
c8b97818 3910 compressed = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 3911 if (found_type == BTRFS_FILE_EXTENT_REG ||
3912 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 3913 extent_end = extent_start +
db94535d 3914 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 3915 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3916 size_t size;
3917 size = btrfs_file_extent_inline_len(leaf, item);
3918 extent_end = (extent_start + size + root->sectorsize - 1) &
3919 ~((u64)root->sectorsize - 1);
3920 }
3921
3922 if (start >= extent_end) {
3923 path->slots[0]++;
3924 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3925 ret = btrfs_next_leaf(root, path);
3926 if (ret < 0) {
3927 err = ret;
3928 goto out;
a52d9a80 3929 }
9036c102
YZ		 3930 if (ret > 0)
3931 goto not_found;
3932 leaf = path->nodes[0];
a52d9a80 3933 }
9036c102
YZ		 3934 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3935 if (found_key.objectid != objectid ||
3936 found_key.type != BTRFS_EXTENT_DATA_KEY)
3937 goto not_found;
3938 if (start + len <= found_key.offset)
3939 goto not_found;
3940 em->start = start;
3941 em->len = found_key.offset - start;
3942 goto not_found_em;
3943 }
3944
d899e052
YZ		 3945 if (found_type == BTRFS_FILE_EXTENT_REG ||
3946 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 3947 em->start = extent_start;
3948 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 3949 em->orig_start = extent_start -
3950 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 3951 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3952 if (bytenr == 0) {
5f39d397 3953 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 3954 goto insert;
3955 }
c8b97818
CM		 3956 if (compressed) {
3957 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3958 em->block_start = bytenr;
3959 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3960 item);
3961 } else {
3962 bytenr += btrfs_file_extent_offset(leaf, item);
3963 em->block_start = bytenr;
3964 em->block_len = em->len;
d899e052
YZ		 3965 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3966 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 3967 }
a52d9a80
CM		 3968 goto insert;
3969 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3970 unsigned long ptr;
a52d9a80 3971 char *map;
3326d1b0
CM		 3972 size_t size;
3973 size_t extent_offset;
3974 size_t copy_size;
a52d9a80 3975
689f9346 3976 em->block_start = EXTENT_MAP_INLINE;
c8b97818 3977 if (!page || create) {
689f9346 3978 em->start = extent_start;
9036c102 3979 em->len = extent_end - extent_start;
689f9346
Y		 3980 goto out;
3981 }
5f39d397 3982
9036c102
YZ		 3983 size = btrfs_file_extent_inline_len(leaf, item);
3984 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 3985 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 3986 size - extent_offset);
3326d1b0 3987 em->start = extent_start + extent_offset;
70dec807
CM		 3988 em->len = (copy_size + root->sectorsize - 1) &
3989 ~((u64)root->sectorsize - 1);
ff5b7ee3 3990 em->orig_start = EXTENT_MAP_INLINE;
c8b97818
CM		 3991 if (compressed)
3992 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
689f9346 3993 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 3994 if (create == 0 && !PageUptodate(page)) {
c8b97818
CM		 3995 if (btrfs_file_extent_compression(leaf, item) ==
3996 BTRFS_COMPRESS_ZLIB) {
3997 ret = uncompress_inline(path, inode, page,
3998 pg_offset,
3999 extent_offset, item);
4000 BUG_ON(ret);
4001 } else {
4002 map = kmap(page);
4003 read_extent_buffer(leaf, map + pg_offset, ptr,
4004 copy_size);
4005 kunmap(page);
4006 }
179e29e4
CM		 4007 flush_dcache_page(page);
4008 } else if (create && PageUptodate(page)) {
4009 if (!trans) {
4010 kunmap(page);
4011 free_extent_map(em);
4012 em = NULL;
4013 btrfs_release_path(root, path);
f9295749 4014 trans = btrfs_join_transaction(root, 1);
179e29e4
CM		 4015 goto again;
4016 }
c8b97818 4017 map = kmap(page);
70dec807 4018 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 4019 copy_size);
c8b97818 4020 kunmap(page);
179e29e4 4021 btrfs_mark_buffer_dirty(leaf);
a52d9a80 4022 }
d1310b2e
CM		 4023 set_extent_uptodate(io_tree, em->start,
4024 extent_map_end(em) - 1, GFP_NOFS);
a52d9a80
CM		 4025 goto insert;
4026 } else {
4027 printk("unkknown found_type %d\n", found_type);
4028 WARN_ON(1);
4029 }
4030not_found:
4031 em->start = start;
d1310b2e 4032 em->len = len;
a52d9a80 4033not_found_em:
5f39d397 4034 em->block_start = EXTENT_MAP_HOLE;
9036c102 4035 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80
CM		 4036insert:
4037 btrfs_release_path(root, path);
d1310b2e
CM		 4038 if (em->start > start || extent_map_end(em) <= start) {
4039 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
a52d9a80
CM		 4040 err = -EIO;
4041 goto out;
4042 }
d1310b2e
CM		 4043
4044 err = 0;
4045 spin_lock(&em_tree->lock);
a52d9a80 4046 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 4047 /* it is possible that someone inserted the extent into the tree
4048 * while we had the lock dropped. It is also possible that
4049 * an overlapping map exists in the tree
4050 */
a52d9a80 4051 if (ret == -EEXIST) {
3b951516 4052 struct extent_map *existing;
e6dcd2dc
CM		 4053
4054 ret = 0;
4055
3b951516 4056 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 4057 if (existing && (existing->start > start ||
4058 existing->start + existing->len <= start)) {
4059 free_extent_map(existing);
4060 existing = NULL;
4061 }
3b951516
CM		 4062 if (!existing) {
4063 existing = lookup_extent_mapping(em_tree, em->start,
4064 em->len);
4065 if (existing) {
4066 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 4067 em, start,
4068 root->sectorsize);
3b951516
CM		 4069 free_extent_map(existing);
4070 if (err) {
4071 free_extent_map(em);
4072 em = NULL;
4073 }
4074 } else {
4075 err = -EIO;
4076 printk("failing to insert %Lu %Lu\n",
4077 start, len);
4078 free_extent_map(em);
4079 em = NULL;
4080 }
4081 } else {
4082 free_extent_map(em);
4083 em = existing;
e6dcd2dc 4084 err = 0;
a52d9a80 4085 }
a52d9a80 4086 }
d1310b2e 4087 spin_unlock(&em_tree->lock);
a52d9a80 4088out:
f421950f
CM		 4089 if (path)
4090 btrfs_free_path(path);
a52d9a80
CM		 4091 if (trans) {
4092 ret = btrfs_end_transaction(trans, root);
e6dcd2dc 4093 if (!err) {
a52d9a80 4094 err = ret;
e6dcd2dc 4095 }
a52d9a80 4096 }
a52d9a80
CM		 4097 if (err) {
4098 free_extent_map(em);
4099 WARN_ON(1);
4100 return ERR_PTR(err);
4101 }
4102 return em;
4103}
4104
16432985
CM		 4105static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4106 const struct iovec *iov, loff_t offset,
4107 unsigned long nr_segs)
4108{
e1c4b745 4109 return -EINVAL;
16432985
CM		 4110}
4111
d396c6f5 4112static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
39279cc3 4113{
d396c6f5 4114 return extent_bmap(mapping, iblock, btrfs_get_extent);
39279cc3
CM		 4115}
4116
a52d9a80 4117int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 4118{
d1310b2e
CM		 4119 struct extent_io_tree *tree;
4120 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4121 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 4122}
1832a6d5 4123
a52d9a80 4124static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 4125{
d1310b2e 4126 struct extent_io_tree *tree;
b888db2b
CM		 4127
4128
4129 if (current->flags & PF_MEMALLOC) {
4130 redirty_page_for_writepage(wbc, page);
4131 unlock_page(page);
4132 return 0;
4133 }
d1310b2e 4134 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4135 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 4136}
4137
f421950f
CM		 4138int btrfs_writepages(struct address_space *mapping,
4139 struct writeback_control *wbc)
b293f02e 4140{
d1310b2e 4141 struct extent_io_tree *tree;
771ed689 4142
d1310b2e 4143 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 4144 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4145}
4146
3ab2fb5a
CM		 4147static int
4148btrfs_readpages(struct file *file, struct address_space *mapping,
4149 struct list_head *pages, unsigned nr_pages)
4150{
d1310b2e
CM		 4151 struct extent_io_tree *tree;
4152 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 4153 return extent_readpages(tree, mapping, pages, nr_pages,
4154 btrfs_get_extent);
4155}
e6dcd2dc 4156static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 4157{
d1310b2e
CM		 4158 struct extent_io_tree *tree;
4159 struct extent_map_tree *map;
a52d9a80 4160 int ret;
8c2383c3 4161
d1310b2e
CM		 4162 tree = &BTRFS_I(page->mapping->host)->io_tree;
4163 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 4164 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 4165 if (ret == 1) {
4166 ClearPagePrivate(page);
4167 set_page_private(page, 0);
4168 page_cache_release(page);
39279cc3 4169 }
a52d9a80 4170 return ret;
39279cc3
CM		 4171}
4172
e6dcd2dc
CM		 4173static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4174{
98509cfc
CM		 4175 if (PageWriteback(page) || PageDirty(page))
4176 return 0;
e6dcd2dc
CM		 4177 return __btrfs_releasepage(page, gfp_flags);
4178}
4179
a52d9a80 4180static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 4181{
d1310b2e 4182 struct extent_io_tree *tree;
e6dcd2dc
CM		 4183 struct btrfs_ordered_extent *ordered;
4184 u64 page_start = page_offset(page);
4185 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 4186
e6dcd2dc 4187 wait_on_page_writeback(page);
d1310b2e 4188 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 4189 if (offset) {
4190 btrfs_releasepage(page, GFP_NOFS);
4191 return;
4192 }
4193
4194 lock_extent(tree, page_start, page_end, GFP_NOFS);
4195 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4196 page_offset(page));
4197 if (ordered) {
eb84ae03
CM		 4198 /*
4199 * IO on this page will never be started, so we need
4200 * to account for any ordered extents now
4201 */
e6dcd2dc
CM		 4202 clear_extent_bit(tree, page_start, page_end,
4203 EXTENT_DIRTY | EXTENT_DELALLOC |
4204 EXTENT_LOCKED, 1, 0, GFP_NOFS);
211f90e6
CM		 4205 btrfs_finish_ordered_io(page->mapping->host,
4206 page_start, page_end);
e6dcd2dc
CM		 4207 btrfs_put_ordered_extent(ordered);
4208 lock_extent(tree, page_start, page_end, GFP_NOFS);
4209 }
4210 clear_extent_bit(tree, page_start, page_end,
4211 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4212 EXTENT_ORDERED,
4213 1, 1, GFP_NOFS);
4214 __btrfs_releasepage(page, GFP_NOFS);
4215
4a096752 4216 ClearPageChecked(page);
9ad6b7bc 4217 if (PagePrivate(page)) {
9ad6b7bc
CM		 4218 ClearPagePrivate(page);
4219 set_page_private(page, 0);
4220 page_cache_release(page);
4221 }
39279cc3
CM		 4222}
4223
9ebefb18
CM		 4224/*
4225 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4226 * called from a page fault handler when a page is first dirtied. Hence we must
4227 * be careful to check for EOF conditions here. We set the page up correctly
4228 * for a written page which means we get ENOSPC checking when writing into
4229 * holes and correct delalloc and unwritten extent mapping on filesystems that
4230 * support these features.
4231 *
4232 * We are not allowed to take the i_mutex here so we have to play games to
4233 * protect against truncate races as the page could now be beyond EOF. Because
4234 * vmtruncate() writes the inode size before removing pages, once we have the
4235 * page lock we can determine safely if the page is beyond EOF. If it is not
4236 * beyond EOF, then the page is guaranteed safe against truncation until we
4237 * unlock the page.
4238 */
4239int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4240{
6da6abae 4241 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 4242 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4243 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4244 struct btrfs_ordered_extent *ordered;
4245 char *kaddr;
4246 unsigned long zero_start;
9ebefb18 4247 loff_t size;
1832a6d5 4248 int ret;
a52d9a80 4249 u64 page_start;
e6dcd2dc 4250 u64 page_end;
9ebefb18 4251
1832a6d5 4252 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
1832a6d5
CM		 4253 if (ret)
4254 goto out;
4255
4256 ret = -EINVAL;
e6dcd2dc 4257again:
9ebefb18 4258 lock_page(page);
9ebefb18 4259 size = i_size_read(inode);
e6dcd2dc
CM		 4260 page_start = page_offset(page);
4261 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 4262
9ebefb18 4263 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 4264 (page_start >= size)) {
9ebefb18
CM		 4265 /* page got truncated out from underneath us */
4266 goto out_unlock;
4267 }
e6dcd2dc
CM		 4268 wait_on_page_writeback(page);
4269
4270 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4271 set_page_extent_mapped(page);
4272
eb84ae03
CM		 4273 /*
4274 * we can't set the delalloc bits if there are pending ordered
4275 * extents. Drop our locks and wait for them to finish
4276 */
e6dcd2dc
CM		 4277 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4278 if (ordered) {
4279 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4280 unlock_page(page);
eb84ae03 4281 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4282 btrfs_put_ordered_extent(ordered);
4283 goto again;
4284 }
4285
ea8c2819 4286 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc 4287 ret = 0;
9ebefb18
CM		 4288
4289 /* page is wholly or partially inside EOF */
a52d9a80 4290 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 4291 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 4292 else
e6dcd2dc 4293 zero_start = PAGE_CACHE_SIZE;
9ebefb18 4294
e6dcd2dc
CM		 4295 if (zero_start != PAGE_CACHE_SIZE) {
4296 kaddr = kmap(page);
4297 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4298 flush_dcache_page(page);
4299 kunmap(page);
4300 }
247e743c 4301 ClearPageChecked(page);
e6dcd2dc
CM		 4302 set_page_dirty(page);
4303 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
9ebefb18
CM		 4304
4305out_unlock:
4306 unlock_page(page);
1832a6d5 4307out:
9ebefb18
CM		 4308 return ret;
4309}
4310
39279cc3
CM		 4311static void btrfs_truncate(struct inode *inode)
4312{
4313 struct btrfs_root *root = BTRFS_I(inode)->root;
4314 int ret;
4315 struct btrfs_trans_handle *trans;
d3c2fdcf 4316 unsigned long nr;
dbe674a9 4317 u64 mask = root->sectorsize - 1;
39279cc3
CM		 4318
4319 if (!S_ISREG(inode->i_mode))
4320 return;
4321 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4322 return;
4323
4324 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4a096752 4325 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 4326
39279cc3
CM		 4327 trans = btrfs_start_transaction(root, 1);
4328 btrfs_set_trans_block_group(trans, inode);
dbe674a9 4329 btrfs_i_size_write(inode, inode->i_size);
39279cc3 4330
7b128766
JB		 4331 ret = btrfs_orphan_add(trans, inode);
4332 if (ret)
4333 goto out;
39279cc3 4334 /* FIXME, add redo link to tree so we don't leak on crash */
e02119d5 4335 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
85e21bac 4336 BTRFS_EXTENT_DATA_KEY);
39279cc3 4337 btrfs_update_inode(trans, root, inode);
5f39d397 4338
7b128766
JB		 4339 ret = btrfs_orphan_del(trans, inode);
4340 BUG_ON(ret);
4341
4342out:
4343 nr = trans->blocks_used;
89ce8a63 4344 ret = btrfs_end_transaction_throttle(trans, root);
39279cc3 4345 BUG_ON(ret);
d3c2fdcf 4346 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4347}
4348
d352ac68
CM		 4349/*
4350 * create a new subvolume directory/inode (helper for the ioctl).
4351 */
d2fb3437
YZ		 4352int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4353 struct btrfs_root *new_root, struct dentry *dentry,
4354 u64 new_dirid, u64 alloc_hint)
39279cc3 4355{
39279cc3 4356 struct inode *inode;
cb8e7090 4357 int error;
00e4e6b3 4358 u64 index = 0;
39279cc3 4359
aec7477b 4360 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
d2fb3437 4361 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
54aa1f4d 4362 if (IS_ERR(inode))
f46b5a66 4363 return PTR_ERR(inode);
39279cc3
CM		 4364 inode->i_op = &btrfs_dir_inode_operations;
4365 inode->i_fop = &btrfs_dir_file_operations;
4366
39279cc3 4367 inode->i_nlink = 1;
dbe674a9 4368 btrfs_i_size_write(inode, 0);
3b96362c 4369
cb8e7090
CH		 4370 error = btrfs_update_inode(trans, new_root, inode);
4371 if (error)
4372 return error;
4373
4374 d_instantiate(dentry, inode);
4375 return 0;
39279cc3
CM		 4376}
4377
d352ac68
CM		 4378/* helper function for file defrag and space balancing. This
4379 * forces readahead on a given range of bytes in an inode
4380 */
edbd8d4e 4381unsigned long btrfs_force_ra(struct address_space *mapping,
86479a04
CM		 4382 struct file_ra_state *ra, struct file *file,
4383 pgoff_t offset, pgoff_t last_index)
4384{
8e7bf94f 4385 pgoff_t req_size = last_index - offset + 1;
86479a04 4386
86479a04
CM		 4387 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4388 return offset + req_size;
86479a04
CM		 4389}
4390
39279cc3
CM		 4391struct inode *btrfs_alloc_inode(struct super_block *sb)
4392{
4393 struct btrfs_inode *ei;
4394
4395 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4396 if (!ei)
4397 return NULL;
15ee9bc7 4398 ei->last_trans = 0;
e02119d5 4399 ei->logged_trans = 0;
e6dcd2dc 4400 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
33268eaf
JB		 4401 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4402 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
7b128766 4403 INIT_LIST_HEAD(&ei->i_orphan);
39279cc3
CM		 4404 return &ei->vfs_inode;
4405}
4406
4407void btrfs_destroy_inode(struct inode *inode)
4408{
e6dcd2dc 4409 struct btrfs_ordered_extent *ordered;
39279cc3
CM		 4410 WARN_ON(!list_empty(&inode->i_dentry));
4411 WARN_ON(inode->i_data.nrpages);
4412
33268eaf
JB		 4413 if (BTRFS_I(inode)->i_acl &&
4414 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4415 posix_acl_release(BTRFS_I(inode)->i_acl);
4416 if (BTRFS_I(inode)->i_default_acl &&
4417 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4418 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4419
bcc63abb 4420 spin_lock(&BTRFS_I(inode)->root->list_lock);
7b128766
JB		 4421 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4422 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4423 " list\n", inode->i_ino);
4424 dump_stack();
4425 }
bcc63abb 4426 spin_unlock(&BTRFS_I(inode)->root->list_lock);
7b128766 4427
e6dcd2dc
CM		 4428 while(1) {
4429 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4430 if (!ordered)
4431 break;
4432 else {
4433 printk("found ordered extent %Lu %Lu\n",
4434 ordered->file_offset, ordered->len);
4435 btrfs_remove_ordered_extent(inode, ordered);
4436 btrfs_put_ordered_extent(ordered);
4437 btrfs_put_ordered_extent(ordered);
4438 }
4439 }
5b21f2ed 4440 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
39279cc3
CM		 4441 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4442}
4443
0ee0fda0 4444static void init_once(void *foo)
39279cc3
CM		 4445{
4446 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4447
4448 inode_init_once(&ei->vfs_inode);
4449}
4450
4451void btrfs_destroy_cachep(void)
4452{
4453 if (btrfs_inode_cachep)
4454 kmem_cache_destroy(btrfs_inode_cachep);
4455 if (btrfs_trans_handle_cachep)
4456 kmem_cache_destroy(btrfs_trans_handle_cachep);
4457 if (btrfs_transaction_cachep)
4458 kmem_cache_destroy(btrfs_transaction_cachep);
4459 if (btrfs_bit_radix_cachep)
4460 kmem_cache_destroy(btrfs_bit_radix_cachep);
4461 if (btrfs_path_cachep)
4462 kmem_cache_destroy(btrfs_path_cachep);
4463}
4464
86479a04 4465struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
92fee66d 4466 unsigned long extra_flags,
2b1f55b0 4467 void (*ctor)(void *))
92fee66d
CM		 4468{
4469 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2b1f55b0 4470 SLAB_MEM_SPREAD | extra_flags), ctor);
92fee66d
CM		 4471}
4472
39279cc3
CM		 4473int btrfs_init_cachep(void)
4474{
86479a04 4475 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
92fee66d
CM		 4476 sizeof(struct btrfs_inode),
4477 0, init_once);
39279cc3
CM		 4478 if (!btrfs_inode_cachep)
4479 goto fail;
86479a04
CM		 4480 btrfs_trans_handle_cachep =
4481 btrfs_cache_create("btrfs_trans_handle_cache",
4482 sizeof(struct btrfs_trans_handle),
4483 0, NULL);
39279cc3
CM		 4484 if (!btrfs_trans_handle_cachep)
4485 goto fail;
86479a04 4486 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
39279cc3 4487 sizeof(struct btrfs_transaction),
92fee66d 4488 0, NULL);
39279cc3
CM		 4489 if (!btrfs_transaction_cachep)
4490 goto fail;
86479a04 4491 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
23223584 4492 sizeof(struct btrfs_path),
92fee66d 4493 0, NULL);
39279cc3
CM		 4494 if (!btrfs_path_cachep)
4495 goto fail;
86479a04 4496 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
92fee66d 4497 SLAB_DESTROY_BY_RCU, NULL);
39279cc3
CM		 4498 if (!btrfs_bit_radix_cachep)
4499 goto fail;
4500 return 0;
4501fail:
4502 btrfs_destroy_cachep();
4503 return -ENOMEM;
4504}
4505
4506static int btrfs_getattr(struct vfsmount *mnt,
4507 struct dentry *dentry, struct kstat *stat)
4508{
4509 struct inode *inode = dentry->d_inode;
4510 generic_fillattr(inode, stat);
3394e160 4511 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
d6667462 4512 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 4513 stat->blocks = (inode_get_bytes(inode) +
4514 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 4515 return 0;
4516}
4517
4518static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4519 struct inode * new_dir,struct dentry *new_dentry)
4520{
4521 struct btrfs_trans_handle *trans;
4522 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4523 struct inode *new_inode = new_dentry->d_inode;
4524 struct inode *old_inode = old_dentry->d_inode;
4525 struct timespec ctime = CURRENT_TIME;
00e4e6b3 4526 u64 index = 0;
39279cc3
CM		 4527 int ret;
4528
3394e160
CM		 4529 /* we're not allowed to rename between subvolumes */
4530 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4531 BTRFS_I(new_dir)->root->root_key.objectid)
4532 return -EXDEV;
4533
39279cc3
CM		 4534 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4535 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4536 return -ENOTEMPTY;
4537 }
5f39d397 4538
0660b5af
CM		 4539 /* to rename a snapshot or subvolume, we need to juggle the
4540 * backrefs. This isn't coded yet
4541 */
4542 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4543 return -EXDEV;
4544
1832a6d5
CM		 4545 ret = btrfs_check_free_space(root, 1, 0);
4546 if (ret)
4547 goto out_unlock;
4548
39279cc3 4549 trans = btrfs_start_transaction(root, 1);
5f39d397 4550
39279cc3 4551 btrfs_set_trans_block_group(trans, new_dir);
39279cc3 4552
e02119d5 4553 btrfs_inc_nlink(old_dentry->d_inode);
39279cc3
CM		 4554 old_dir->i_ctime = old_dir->i_mtime = ctime;
4555 new_dir->i_ctime = new_dir->i_mtime = ctime;
4556 old_inode->i_ctime = ctime;
5f39d397 4557
e02119d5
CM		 4558 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4559 old_dentry->d_name.name,
4560 old_dentry->d_name.len);
39279cc3
CM		 4561 if (ret)
4562 goto out_fail;
4563
4564 if (new_inode) {
4565 new_inode->i_ctime = CURRENT_TIME;
e02119d5
CM		 4566 ret = btrfs_unlink_inode(trans, root, new_dir,
4567 new_dentry->d_inode,
4568 new_dentry->d_name.name,
4569 new_dentry->d_name.len);
39279cc3
CM		 4570 if (ret)
4571 goto out_fail;
7b128766 4572 if (new_inode->i_nlink == 0) {
e02119d5 4573 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7b128766
JB		 4574 if (ret)
4575 goto out_fail;
4576 }
e02119d5 4577
39279cc3 4578 }
3de4586c 4579 ret = btrfs_set_inode_index(new_dir, &index);
aec7477b
JB		 4580 if (ret)
4581 goto out_fail;
4582
e02119d5
CM		 4583 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4584 old_inode, new_dentry->d_name.name,
4585 new_dentry->d_name.len, 1, index);
39279cc3
CM		 4586 if (ret)
4587 goto out_fail;
4588
4589out_fail:
ab78c84d 4590 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4591out_unlock:
39279cc3
CM		 4592 return ret;
4593}
4594
d352ac68
CM		 4595/*
4596 * some fairly slow code that needs optimization. This walks the list
4597 * of all the inodes with pending delalloc and forces them to disk.
4598 */
ea8c2819
CM		 4599int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4600{
4601 struct list_head *head = &root->fs_info->delalloc_inodes;
4602 struct btrfs_inode *binode;
5b21f2ed 4603 struct inode *inode;
ea8c2819
CM		 4604 unsigned long flags;
4605
c146afad
YZ		 4606 if (root->fs_info->sb->s_flags & MS_RDONLY)
4607 return -EROFS;
4608
ea8c2819
CM		 4609 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4610 while(!list_empty(head)) {
4611 binode = list_entry(head->next, struct btrfs_inode,
4612 delalloc_inodes);
5b21f2ed
ZY		 4613 inode = igrab(&binode->vfs_inode);
4614 if (!inode)
4615 list_del_init(&binode->delalloc_inodes);
ea8c2819 4616 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
5b21f2ed 4617 if (inode) {
8c8bee1d 4618 filemap_flush(inode->i_mapping);
5b21f2ed
ZY		 4619 iput(inode);
4620 }
4621 cond_resched();
ea8c2819
CM		 4622 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4623 }
4624 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
8c8bee1d
CM		 4625
4626 /* the filemap_flush will queue IO into the worker threads, but
4627 * we have to make sure the IO is actually started and that
4628 * ordered extents get created before we return
4629 */
4630 atomic_inc(&root->fs_info->async_submit_draining);
771ed689
CM		 4631 while(atomic_read(&root->fs_info->nr_async_submits) ||
4632 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 4633 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 4634 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4635 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 4636 }
4637 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 4638 return 0;
4639}
4640
39279cc3
CM		 4641static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4642 const char *symname)
4643{
4644 struct btrfs_trans_handle *trans;
4645 struct btrfs_root *root = BTRFS_I(dir)->root;
4646 struct btrfs_path *path;
4647 struct btrfs_key key;
1832a6d5 4648 struct inode *inode = NULL;
39279cc3
CM		 4649 int err;
4650 int drop_inode = 0;
4651 u64 objectid;
00e4e6b3 4652 u64 index = 0 ;
39279cc3
CM		 4653 int name_len;
4654 int datasize;
5f39d397 4655 unsigned long ptr;
39279cc3 4656 struct btrfs_file_extent_item *ei;
5f39d397 4657 struct extent_buffer *leaf;
1832a6d5 4658 unsigned long nr = 0;
39279cc3
CM		 4659
4660 name_len = strlen(symname) + 1;
4661 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4662 return -ENAMETOOLONG;
1832a6d5 4663
1832a6d5
CM		 4664 err = btrfs_check_free_space(root, 1, 0);
4665 if (err)
4666 goto out_fail;
4667
39279cc3
CM		 4668 trans = btrfs_start_transaction(root, 1);
4669 btrfs_set_trans_block_group(trans, dir);
4670
4671 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4672 if (err) {
4673 err = -ENOSPC;
4674 goto out_unlock;
4675 }
4676
aec7477b 4677 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 4678 dentry->d_name.len,
4679 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 4680 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4681 &index);
39279cc3
CM		 4682 err = PTR_ERR(inode);
4683 if (IS_ERR(inode))
4684 goto out_unlock;
4685
33268eaf
JB		 4686 err = btrfs_init_acl(inode, dir);
4687 if (err) {
4688 drop_inode = 1;
4689 goto out_unlock;
4690 }
4691
39279cc3 4692 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 4693 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 4694 if (err)
4695 drop_inode = 1;
4696 else {
4697 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4698 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4699 inode->i_fop = &btrfs_file_operations;
4700 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4701 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 4702 }
4703 dir->i_sb->s_dirt = 1;
4704 btrfs_update_inode_block_group(trans, inode);
4705 btrfs_update_inode_block_group(trans, dir);
4706 if (drop_inode)
4707 goto out_unlock;
4708
4709 path = btrfs_alloc_path();
4710 BUG_ON(!path);
4711 key.objectid = inode->i_ino;
4712 key.offset = 0;
39279cc3
CM		 4713 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4714 datasize = btrfs_file_extent_calc_inline_size(name_len);
4715 err = btrfs_insert_empty_item(trans, root, path, &key,
4716 datasize);
54aa1f4d
CM		 4717 if (err) {
4718 drop_inode = 1;
4719 goto out_unlock;
4720 }
5f39d397
CM		 4721 leaf = path->nodes[0];
4722 ei = btrfs_item_ptr(leaf, path->slots[0],
4723 struct btrfs_file_extent_item);
4724 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4725 btrfs_set_file_extent_type(leaf, ei,
39279cc3 4726 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 4727 btrfs_set_file_extent_encryption(leaf, ei, 0);
4728 btrfs_set_file_extent_compression(leaf, ei, 0);
4729 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4730 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4731
39279cc3 4732 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 4733 write_extent_buffer(leaf, symname, ptr, name_len);
4734 btrfs_mark_buffer_dirty(leaf);
39279cc3 4735 btrfs_free_path(path);
5f39d397 4736
39279cc3
CM		 4737 inode->i_op = &btrfs_symlink_inode_operations;
4738 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 4739 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 4740 inode_set_bytes(inode, name_len);
dbe674a9 4741 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 4742 err = btrfs_update_inode(trans, root, inode);
4743 if (err)
4744 drop_inode = 1;
39279cc3
CM		 4745
4746out_unlock:
d3c2fdcf 4747 nr = trans->blocks_used;
ab78c84d 4748 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4749out_fail:
39279cc3
CM		 4750 if (drop_inode) {
4751 inode_dec_link_count(inode);
4752 iput(inode);
4753 }
d3c2fdcf 4754 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4755 return err;
4756}
16432985 4757
d899e052
YZ		 4758static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4759 u64 alloc_hint, int mode)
4760{
4761 struct btrfs_trans_handle *trans;
4762 struct btrfs_root *root = BTRFS_I(inode)->root;
4763 struct btrfs_key ins;
4764 u64 alloc_size;
4765 u64 cur_offset = start;
4766 u64 num_bytes = end - start;
4767 int ret = 0;
4768
4769 trans = btrfs_join_transaction(root, 1);
4770 BUG_ON(!trans);
4771 btrfs_set_trans_block_group(trans, inode);
4772
4773 while (num_bytes > 0) {
4774 alloc_size = min(num_bytes, root->fs_info->max_extent);
4775 ret = btrfs_reserve_extent(trans, root, alloc_size,
4776 root->sectorsize, 0, alloc_hint,
4777 (u64)-1, &ins, 1);
4778 if (ret) {
4779 WARN_ON(1);
4780 goto out;
4781 }
4782 ret = insert_reserved_file_extent(trans, inode,
4783 cur_offset, ins.objectid,
4784 ins.offset, ins.offset,
4785 ins.offset, 0, 0, 0,
4786 BTRFS_FILE_EXTENT_PREALLOC);
4787 BUG_ON(ret);
4788 num_bytes -= ins.offset;
4789 cur_offset += ins.offset;
4790 alloc_hint = ins.objectid + ins.offset;
4791 }
4792out:
4793 if (cur_offset > start) {
4794 inode->i_ctime = CURRENT_TIME;
4795 btrfs_set_flag(inode, PREALLOC);
4796 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4797 cur_offset > i_size_read(inode))
4798 btrfs_i_size_write(inode, cur_offset);
4799 ret = btrfs_update_inode(trans, root, inode);
4800 BUG_ON(ret);
4801 }
4802
4803 btrfs_end_transaction(trans, root);
4804 return ret;
4805}
4806
4807static long btrfs_fallocate(struct inode *inode, int mode,
4808 loff_t offset, loff_t len)
4809{
4810 u64 cur_offset;
4811 u64 last_byte;
4812 u64 alloc_start;
4813 u64 alloc_end;
4814 u64 alloc_hint = 0;
4815 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4816 struct extent_map *em;
4817 int ret;
4818
4819 alloc_start = offset & ~mask;
4820 alloc_end = (offset + len + mask) & ~mask;
4821
4822 mutex_lock(&inode->i_mutex);
4823 if (alloc_start > inode->i_size) {
4824 ret = btrfs_cont_expand(inode, alloc_start);
4825 if (ret)
4826 goto out;
4827 }
4828
4829 while (1) {
4830 struct btrfs_ordered_extent *ordered;
4831 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4832 alloc_end - 1, GFP_NOFS);
4833 ordered = btrfs_lookup_first_ordered_extent(inode,
4834 alloc_end - 1);
4835 if (ordered &&
4836 ordered->file_offset + ordered->len > alloc_start &&
4837 ordered->file_offset < alloc_end) {
4838 btrfs_put_ordered_extent(ordered);
4839 unlock_extent(&BTRFS_I(inode)->io_tree,
4840 alloc_start, alloc_end - 1, GFP_NOFS);
4841 btrfs_wait_ordered_range(inode, alloc_start,
4842 alloc_end - alloc_start);
4843 } else {
4844 if (ordered)
4845 btrfs_put_ordered_extent(ordered);
4846 break;
4847 }
4848 }
4849
4850 cur_offset = alloc_start;
4851 while (1) {
4852 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4853 alloc_end - cur_offset, 0);
4854 BUG_ON(IS_ERR(em) || !em);
4855 last_byte = min(extent_map_end(em), alloc_end);
4856 last_byte = (last_byte + mask) & ~mask;
4857 if (em->block_start == EXTENT_MAP_HOLE) {
4858 ret = prealloc_file_range(inode, cur_offset,
4859 last_byte, alloc_hint, mode);
4860 if (ret < 0) {
4861 free_extent_map(em);
4862 break;
4863 }
4864 }
4865 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4866 alloc_hint = em->block_start;
4867 free_extent_map(em);
4868
4869 cur_offset = last_byte;
4870 if (cur_offset >= alloc_end) {
4871 ret = 0;
4872 break;
4873 }
4874 }
4875 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4876 GFP_NOFS);
4877out:
4878 mutex_unlock(&inode->i_mutex);
4879 return ret;
4880}
4881
e6dcd2dc
CM		 4882static int btrfs_set_page_dirty(struct page *page)
4883{
e6dcd2dc
CM		 4884 return __set_page_dirty_nobuffers(page);
4885}
4886
0ee0fda0 4887static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd
Y		 4888{
4889 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4890 return -EACCES;
33268eaf 4891 return generic_permission(inode, mask, btrfs_check_acl);
fdebe2bd 4892}
39279cc3
CM		 4893
4894static struct inode_operations btrfs_dir_inode_operations = {
3394e160 4895 .getattr = btrfs_getattr,
39279cc3
CM		 4896 .lookup = btrfs_lookup,
4897 .create = btrfs_create,
4898 .unlink = btrfs_unlink,
4899 .link = btrfs_link,
4900 .mkdir = btrfs_mkdir,
4901 .rmdir = btrfs_rmdir,
4902 .rename = btrfs_rename,
4903 .symlink = btrfs_symlink,
4904 .setattr = btrfs_setattr,
618e21d5 4905 .mknod = btrfs_mknod,
95819c05
CH		 4906 .setxattr = btrfs_setxattr,
4907 .getxattr = btrfs_getxattr,
5103e947 4908 .listxattr = btrfs_listxattr,
95819c05 4909 .removexattr = btrfs_removexattr,
fdebe2bd 4910 .permission = btrfs_permission,
39279cc3 4911};
39279cc3
CM		 4912static struct inode_operations btrfs_dir_ro_inode_operations = {
4913 .lookup = btrfs_lookup,
fdebe2bd 4914 .permission = btrfs_permission,
39279cc3 4915};
39279cc3
CM		 4916static struct file_operations btrfs_dir_file_operations = {
4917 .llseek = generic_file_llseek,
4918 .read = generic_read_dir,
cbdf5a24 4919 .readdir = btrfs_real_readdir,
34287aa3 4920 .unlocked_ioctl = btrfs_ioctl,
39279cc3 4921#ifdef CONFIG_COMPAT
34287aa3 4922 .compat_ioctl = btrfs_ioctl,
39279cc3 4923#endif
6bf13c0c 4924 .release = btrfs_release_file,
e02119d5 4925 .fsync = btrfs_sync_file,
39279cc3
CM		 4926};
4927
d1310b2e 4928static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 4929 .fill_delalloc = run_delalloc_range,
065631f6 4930 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 4931 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 4932 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 4933 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 4934 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 4935 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 4936 .set_bit_hook = btrfs_set_bit_hook,
4937 .clear_bit_hook = btrfs_clear_bit_hook,
07157aac
CM		 4938};
4939
39279cc3
CM		 4940static struct address_space_operations btrfs_aops = {
4941 .readpage = btrfs_readpage,
4942 .writepage = btrfs_writepage,
b293f02e 4943 .writepages = btrfs_writepages,
3ab2fb5a 4944 .readpages = btrfs_readpages,
39279cc3 4945 .sync_page = block_sync_page,
39279cc3 4946 .bmap = btrfs_bmap,
16432985 4947 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 4948 .invalidatepage = btrfs_invalidatepage,
4949 .releasepage = btrfs_releasepage,
e6dcd2dc 4950 .set_page_dirty = btrfs_set_page_dirty,
39279cc3
CM		 4951};
4952
4953static struct address_space_operations btrfs_symlink_aops = {
4954 .readpage = btrfs_readpage,
4955 .writepage = btrfs_writepage,
2bf5a725
CM		 4956 .invalidatepage = btrfs_invalidatepage,
4957 .releasepage = btrfs_releasepage,
39279cc3
CM		 4958};
4959
4960static struct inode_operations btrfs_file_inode_operations = {
4961 .truncate = btrfs_truncate,
4962 .getattr = btrfs_getattr,
4963 .setattr = btrfs_setattr,
95819c05
CH		 4964 .setxattr = btrfs_setxattr,
4965 .getxattr = btrfs_getxattr,
5103e947 4966 .listxattr = btrfs_listxattr,
95819c05 4967 .removexattr = btrfs_removexattr,
fdebe2bd 4968 .permission = btrfs_permission,
d899e052 4969 .fallocate = btrfs_fallocate,
39279cc3 4970};
618e21d5
JB		 4971static struct inode_operations btrfs_special_inode_operations = {
4972 .getattr = btrfs_getattr,
4973 .setattr = btrfs_setattr,
fdebe2bd 4974 .permission = btrfs_permission,
95819c05
CH		 4975 .setxattr = btrfs_setxattr,
4976 .getxattr = btrfs_getxattr,
33268eaf 4977 .listxattr = btrfs_listxattr,
95819c05 4978 .removexattr = btrfs_removexattr,
618e21d5 4979};
39279cc3
CM		 4980static struct inode_operations btrfs_symlink_inode_operations = {
4981 .readlink = generic_readlink,
4982 .follow_link = page_follow_link_light,
4983 .put_link = page_put_link,
fdebe2bd 4984 .permission = btrfs_permission,
39279cc3 4985};
Linux 6.x block layer and io_uring tree(s)