projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Fix compressed writes on truncated pages
[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;
42dc7bab 335 u64 isize = i_size_read(inode);
e6dcd2dc 336 int ret = 0;
c8b97818
CM		 337 struct page **pages = NULL;
338 unsigned long nr_pages;
339 unsigned long nr_pages_ret = 0;
340 unsigned long total_compressed = 0;
341 unsigned long total_in = 0;
342 unsigned long max_compressed = 128 * 1024;
771ed689 343 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 344 int i;
345 int will_compress;
b888db2b 346
c8b97818
CM		 347 orig_start = start;
348
42dc7bab 349 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 350again:
351 will_compress = 0;
352 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
353 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 354
c8b97818
CM		 355 total_compressed = actual_end - start;
356
357 /* we want to make sure that amount of ram required to uncompress
358 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 359 * of a compressed extent to 128k. This is a crucial number
360 * because it also controls how easily we can spread reads across
361 * cpus for decompression.
362 *
363 * We also want to make sure the amount of IO required to do
364 * a random read is reasonably small, so we limit the size of
365 * a compressed extent to 128k.
c8b97818
CM		 366 */
367 total_compressed = min(total_compressed, max_uncompressed);
db94535d 368 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 369 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 370 disk_num_bytes = num_bytes;
371 total_in = 0;
372 ret = 0;
db94535d 373
771ed689
CM		 374 /*
375 * we do compression for mount -o compress and when the
376 * inode has not been flagged as nocompress. This flag can
377 * change at any time if we discover bad compression ratios.
c8b97818
CM		 378 */
379 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
380 btrfs_test_opt(root, COMPRESS)) {
381 WARN_ON(pages);
cfbc246e 382 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
c8b97818 383
c8b97818
CM		 384 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
385 total_compressed, pages,
386 nr_pages, &nr_pages_ret,
387 &total_in,
388 &total_compressed,
389 max_compressed);
390
391 if (!ret) {
392 unsigned long offset = total_compressed &
393 (PAGE_CACHE_SIZE - 1);
394 struct page *page = pages[nr_pages_ret - 1];
395 char *kaddr;
396
397 /* zero the tail end of the last page, we might be
398 * sending it down to disk
399 */
400 if (offset) {
401 kaddr = kmap_atomic(page, KM_USER0);
402 memset(kaddr + offset, 0,
403 PAGE_CACHE_SIZE - offset);
404 kunmap_atomic(kaddr, KM_USER0);
405 }
406 will_compress = 1;
407 }
408 }
409 if (start == 0) {
771ed689
CM		 410 trans = btrfs_join_transaction(root, 1);
411 BUG_ON(!trans);
412 btrfs_set_trans_block_group(trans, inode);
413
c8b97818 414 /* lets try to make an inline extent */
771ed689 415 if (ret || total_in < (actual_end - start)) {
c8b97818 416 /* we didn't compress the entire range, try
771ed689 417 * to make an uncompressed inline extent.
c8b97818
CM		 418 */
419 ret = cow_file_range_inline(trans, root, inode,
420 start, end, 0, NULL);
421 } else {
771ed689 422 /* try making a compressed inline extent */
c8b97818
CM		 423 ret = cow_file_range_inline(trans, root, inode,
424 start, end,
425 total_compressed, pages);
426 }
771ed689 427 btrfs_end_transaction(trans, root);
c8b97818 428 if (ret == 0) {
771ed689
CM		 429 /*
430 * inline extent creation worked, we don't need
431 * to create any more async work items. Unlock
432 * and free up our temp pages.
433 */
c8b97818
CM		 434 extent_clear_unlock_delalloc(inode,
435 &BTRFS_I(inode)->io_tree,
771ed689
CM		 436 start, end, NULL, 1, 0,
437 0, 1, 1, 1);
c8b97818
CM		 438 ret = 0;
439 goto free_pages_out;
440 }
441 }
442
443 if (will_compress) {
444 /*
445 * we aren't doing an inline extent round the compressed size
446 * up to a block size boundary so the allocator does sane
447 * things
448 */
449 total_compressed = (total_compressed + blocksize - 1) &
450 ~(blocksize - 1);
451
452 /*
453 * one last check to make sure the compression is really a
454 * win, compare the page count read with the blocks on disk
455 */
456 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
457 ~(PAGE_CACHE_SIZE - 1);
458 if (total_compressed >= total_in) {
459 will_compress = 0;
460 } else {
461 disk_num_bytes = total_compressed;
462 num_bytes = total_in;
463 }
464 }
465 if (!will_compress && pages) {
466 /*
467 * the compression code ran but failed to make things smaller,
468 * free any pages it allocated and our page pointer array
469 */
470 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 471 WARN_ON(pages[i]->mapping);
c8b97818
CM		 472 page_cache_release(pages[i]);
473 }
474 kfree(pages);
475 pages = NULL;
476 total_compressed = 0;
477 nr_pages_ret = 0;
478
479 /* flag the file so we don't compress in the future */
480 btrfs_set_flag(inode, NOCOMPRESS);
481 }
771ed689
CM		 482 if (will_compress) {
483 *num_added += 1;
c8b97818 484
771ed689
CM		 485 /* the async work queues will take care of doing actual
486 * allocation on disk for these compressed pages,
487 * and will submit them to the elevator.
488 */
489 add_async_extent(async_cow, start, num_bytes,
490 total_compressed, pages, nr_pages_ret);
179e29e4 491
42dc7bab 492 if (start + num_bytes < end && start + num_bytes < actual_end) {
771ed689
CM		 493 start += num_bytes;
494 pages = NULL;
495 cond_resched();
496 goto again;
497 }
498 } else {
499 /*
500 * No compression, but we still need to write the pages in
501 * the file we've been given so far. redirty the locked
502 * page if it corresponds to our extent and set things up
503 * for the async work queue to run cow_file_range to do
504 * the normal delalloc dance
505 */
506 if (page_offset(locked_page) >= start &&
507 page_offset(locked_page) <= end) {
508 __set_page_dirty_nobuffers(locked_page);
509 /* unlocked later on in the async handlers */
510 }
511 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
512 *num_added += 1;
513 }
3b951516 514
771ed689
CM		 515out:
516 return 0;
517
518free_pages_out:
519 for (i = 0; i < nr_pages_ret; i++) {
520 WARN_ON(pages[i]->mapping);
521 page_cache_release(pages[i]);
522 }
523 if (pages)
524 kfree(pages);
525
526 goto out;
527}
528
529/*
530 * phase two of compressed writeback. This is the ordered portion
531 * of the code, which only gets called in the order the work was
532 * queued. We walk all the async extents created by compress_file_range
533 * and send them down to the disk.
534 */
535static noinline int submit_compressed_extents(struct inode *inode,
536 struct async_cow *async_cow)
537{
538 struct async_extent *async_extent;
539 u64 alloc_hint = 0;
540 struct btrfs_trans_handle *trans;
541 struct btrfs_key ins;
542 struct extent_map *em;
543 struct btrfs_root *root = BTRFS_I(inode)->root;
544 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
545 struct extent_io_tree *io_tree;
546 int ret;
547
548 if (list_empty(&async_cow->extents))
549 return 0;
550
551 trans = btrfs_join_transaction(root, 1);
552
553 while(!list_empty(&async_cow->extents)) {
554 async_extent = list_entry(async_cow->extents.next,
555 struct async_extent, list);
556 list_del(&async_extent->list);
c8b97818 557
771ed689
CM		 558 io_tree = &BTRFS_I(inode)->io_tree;
559
560 /* did the compression code fall back to uncompressed IO? */
561 if (!async_extent->pages) {
562 int page_started = 0;
563 unsigned long nr_written = 0;
564
565 lock_extent(io_tree, async_extent->start,
566 async_extent->start + async_extent->ram_size - 1,
567 GFP_NOFS);
568
569 /* allocate blocks */
570 cow_file_range(inode, async_cow->locked_page,
571 async_extent->start,
572 async_extent->start +
573 async_extent->ram_size - 1,
574 &page_started, &nr_written, 0);
575
576 /*
577 * if page_started, cow_file_range inserted an
578 * inline extent and took care of all the unlocking
579 * and IO for us. Otherwise, we need to submit
580 * all those pages down to the drive.
581 */
582 if (!page_started)
583 extent_write_locked_range(io_tree,
584 inode, async_extent->start,
585 async_extent->start +
586 async_extent->ram_size - 1,
587 btrfs_get_extent,
588 WB_SYNC_ALL);
589 kfree(async_extent);
590 cond_resched();
591 continue;
592 }
593
594 lock_extent(io_tree, async_extent->start,
595 async_extent->start + async_extent->ram_size - 1,
596 GFP_NOFS);
c8b97818 597 /*
771ed689
CM		 598 * here we're doing allocation and writeback of the
599 * compressed pages
c8b97818 600 */
771ed689
CM		 601 btrfs_drop_extent_cache(inode, async_extent->start,
602 async_extent->start +
603 async_extent->ram_size - 1, 0);
604
605 ret = btrfs_reserve_extent(trans, root,
606 async_extent->compressed_size,
607 async_extent->compressed_size,
608 0, alloc_hint,
609 (u64)-1, &ins, 1);
610 BUG_ON(ret);
611 em = alloc_extent_map(GFP_NOFS);
612 em->start = async_extent->start;
613 em->len = async_extent->ram_size;
445a6944 614 em->orig_start = em->start;
c8b97818 615
771ed689
CM		 616 em->block_start = ins.objectid;
617 em->block_len = ins.offset;
618 em->bdev = root->fs_info->fs_devices->latest_bdev;
619 set_bit(EXTENT_FLAG_PINNED, &em->flags);
620 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
621
622 while(1) {
623 spin_lock(&em_tree->lock);
624 ret = add_extent_mapping(em_tree, em);
625 spin_unlock(&em_tree->lock);
626 if (ret != -EEXIST) {
627 free_extent_map(em);
628 break;
629 }
630 btrfs_drop_extent_cache(inode, async_extent->start,
631 async_extent->start +
632 async_extent->ram_size - 1, 0);
633 }
634
635 ret = btrfs_add_ordered_extent(inode, async_extent->start,
636 ins.objectid,
637 async_extent->ram_size,
638 ins.offset,
639 BTRFS_ORDERED_COMPRESSED);
640 BUG_ON(ret);
641
642 btrfs_end_transaction(trans, root);
643
644 /*
645 * clear dirty, set writeback and unlock the pages.
646 */
647 extent_clear_unlock_delalloc(inode,
648 &BTRFS_I(inode)->io_tree,
649 async_extent->start,
650 async_extent->start +
651 async_extent->ram_size - 1,
652 NULL, 1, 1, 0, 1, 1, 0);
653
654 ret = btrfs_submit_compressed_write(inode,
655 async_extent->start,
656 async_extent->ram_size,
657 ins.objectid,
658 ins.offset, async_extent->pages,
659 async_extent->nr_pages);
660
661 BUG_ON(ret);
662 trans = btrfs_join_transaction(root, 1);
663 alloc_hint = ins.objectid + ins.offset;
664 kfree(async_extent);
665 cond_resched();
666 }
667
668 btrfs_end_transaction(trans, root);
669 return 0;
670}
671
672/*
673 * when extent_io.c finds a delayed allocation range in the file,
674 * the call backs end up in this code. The basic idea is to
675 * allocate extents on disk for the range, and create ordered data structs
676 * in ram to track those extents.
677 *
678 * locked_page is the page that writepage had locked already. We use
679 * it to make sure we don't do extra locks or unlocks.
680 *
681 * *page_started is set to one if we unlock locked_page and do everything
682 * required to start IO on it. It may be clean and already done with
683 * IO when we return.
684 */
685static noinline int cow_file_range(struct inode *inode,
686 struct page *locked_page,
687 u64 start, u64 end, int *page_started,
688 unsigned long *nr_written,
689 int unlock)
690{
691 struct btrfs_root *root = BTRFS_I(inode)->root;
692 struct btrfs_trans_handle *trans;
693 u64 alloc_hint = 0;
694 u64 num_bytes;
695 unsigned long ram_size;
696 u64 disk_num_bytes;
697 u64 cur_alloc_size;
698 u64 blocksize = root->sectorsize;
699 u64 actual_end;
42dc7bab 700 u64 isize = i_size_read(inode);
771ed689
CM		 701 struct btrfs_key ins;
702 struct extent_map *em;
703 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
704 int ret = 0;
705
706 trans = btrfs_join_transaction(root, 1);
707 BUG_ON(!trans);
708 btrfs_set_trans_block_group(trans, inode);
709
42dc7bab 710 actual_end = min_t(u64, isize, end + 1);
771ed689
CM		 711
712 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
713 num_bytes = max(blocksize, num_bytes);
714 disk_num_bytes = num_bytes;
715 ret = 0;
716
717 if (start == 0) {
718 /* lets try to make an inline extent */
719 ret = cow_file_range_inline(trans, root, inode,
720 start, end, 0, NULL);
721 if (ret == 0) {
722 extent_clear_unlock_delalloc(inode,
723 &BTRFS_I(inode)->io_tree,
724 start, end, NULL, 1, 1,
725 1, 1, 1, 1);
726 *nr_written = *nr_written +
727 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
728 *page_started = 1;
729 ret = 0;
730 goto out;
731 }
732 }
733
734 BUG_ON(disk_num_bytes >
735 btrfs_super_total_bytes(&root->fs_info->super_copy));
736
737 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
738
739 while(disk_num_bytes > 0) {
c8b97818 740 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
e6dcd2dc 741 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 742 root->sectorsize, 0, alloc_hint,
e6dcd2dc 743 (u64)-1, &ins, 1);
c59f8951 744 if (ret) {
771ed689 745 BUG();
c59f8951 746 }
e6dcd2dc
CM		 747 em = alloc_extent_map(GFP_NOFS);
748 em->start = start;
445a6944 749 em->orig_start = em->start;
c8b97818 750
771ed689
CM		 751 ram_size = ins.offset;
752 em->len = ins.offset;
c8b97818 753
e6dcd2dc 754 em->block_start = ins.objectid;
c8b97818 755 em->block_len = ins.offset;
e6dcd2dc 756 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 757 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 758
e6dcd2dc
CM		 759 while(1) {
760 spin_lock(&em_tree->lock);
761 ret = add_extent_mapping(em_tree, em);
762 spin_unlock(&em_tree->lock);
763 if (ret != -EEXIST) {
764 free_extent_map(em);
765 break;
766 }
767 btrfs_drop_extent_cache(inode, start,
c8b97818 768 start + ram_size - 1, 0);
e6dcd2dc
CM		 769 }
770
98d20f67 771 cur_alloc_size = ins.offset;
e6dcd2dc 772 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 773 ram_size, cur_alloc_size, 0);
e6dcd2dc 774 BUG_ON(ret);
c8b97818 775
17d217fe
YZ		 776 if (root->root_key.objectid ==
777 BTRFS_DATA_RELOC_TREE_OBJECTID) {
778 ret = btrfs_reloc_clone_csums(inode, start,
779 cur_alloc_size);
780 BUG_ON(ret);
781 }
782
c8b97818
CM		 783 if (disk_num_bytes < cur_alloc_size) {
784 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
3b951516
CM		 785 cur_alloc_size);
786 break;
787 }
c8b97818
CM		 788 /* we're not doing compressed IO, don't unlock the first
789 * page (which the caller expects to stay locked), don't
790 * clear any dirty bits and don't set any writeback bits
791 */
792 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
793 start, start + ram_size - 1,
771ed689
CM		 794 locked_page, unlock, 1,
795 1, 0, 0, 0);
c8b97818 796 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 797 num_bytes -= cur_alloc_size;
798 alloc_hint = ins.objectid + ins.offset;
799 start += cur_alloc_size;
b888db2b 800 }
b888db2b 801out:
771ed689 802 ret = 0;
b888db2b 803 btrfs_end_transaction(trans, root);
c8b97818 804
be20aa9d 805 return ret;
771ed689 806}
c8b97818 807
771ed689
CM		 808/*
809 * work queue call back to started compression on a file and pages
810 */
811static noinline void async_cow_start(struct btrfs_work *work)
812{
813 struct async_cow *async_cow;
814 int num_added = 0;
815 async_cow = container_of(work, struct async_cow, work);
816
817 compress_file_range(async_cow->inode, async_cow->locked_page,
818 async_cow->start, async_cow->end, async_cow,
819 &num_added);
820 if (num_added == 0)
821 async_cow->inode = NULL;
822}
823
824/*
825 * work queue call back to submit previously compressed pages
826 */
827static noinline void async_cow_submit(struct btrfs_work *work)
828{
829 struct async_cow *async_cow;
830 struct btrfs_root *root;
831 unsigned long nr_pages;
832
833 async_cow = container_of(work, struct async_cow, work);
834
835 root = async_cow->root;
836 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
837 PAGE_CACHE_SHIFT;
838
839 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
840
841 if (atomic_read(&root->fs_info->async_delalloc_pages) <
842 5 * 1042 * 1024 &&
843 waitqueue_active(&root->fs_info->async_submit_wait))
844 wake_up(&root->fs_info->async_submit_wait);
845
846 if (async_cow->inode) {
847 submit_compressed_extents(async_cow->inode, async_cow);
70b99e69 848 }
771ed689 849}
c8b97818 850
771ed689
CM		 851static noinline void async_cow_free(struct btrfs_work *work)
852{
853 struct async_cow *async_cow;
854 async_cow = container_of(work, struct async_cow, work);
855 kfree(async_cow);
856}
857
858static int cow_file_range_async(struct inode *inode, struct page *locked_page,
859 u64 start, u64 end, int *page_started,
860 unsigned long *nr_written)
861{
862 struct async_cow *async_cow;
863 struct btrfs_root *root = BTRFS_I(inode)->root;
864 unsigned long nr_pages;
865 u64 cur_end;
866 int limit = 10 * 1024 * 1042;
867
868 if (!btrfs_test_opt(root, COMPRESS)) {
869 return cow_file_range(inode, locked_page, start, end,
870 page_started, nr_written, 1);
871 }
872
873 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
874 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
875 while(start < end) {
876 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
877 async_cow->inode = inode;
878 async_cow->root = root;
879 async_cow->locked_page = locked_page;
880 async_cow->start = start;
881
882 if (btrfs_test_flag(inode, NOCOMPRESS))
883 cur_end = end;
884 else
885 cur_end = min(end, start + 512 * 1024 - 1);
886
887 async_cow->end = cur_end;
888 INIT_LIST_HEAD(&async_cow->extents);
889
890 async_cow->work.func = async_cow_start;
891 async_cow->work.ordered_func = async_cow_submit;
892 async_cow->work.ordered_free = async_cow_free;
893 async_cow->work.flags = 0;
894
771ed689
CM		 895 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
896 PAGE_CACHE_SHIFT;
897 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
898
899 btrfs_queue_worker(&root->fs_info->delalloc_workers,
900 &async_cow->work);
901
902 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
903 wait_event(root->fs_info->async_submit_wait,
904 (atomic_read(&root->fs_info->async_delalloc_pages) <
905 limit));
906 }
907
908 while(atomic_read(&root->fs_info->async_submit_draining) &&
909 atomic_read(&root->fs_info->async_delalloc_pages)) {
910 wait_event(root->fs_info->async_submit_wait,
911 (atomic_read(&root->fs_info->async_delalloc_pages) ==
912 0));
913 }
914
915 *nr_written += nr_pages;
916 start = cur_end + 1;
917 }
918 *page_started = 1;
919 return 0;
be20aa9d
CM		 920}
921
17d217fe
YZ		 922static int noinline csum_exist_in_range(struct btrfs_root *root,
923 u64 bytenr, u64 num_bytes)
924{
925 int ret;
926 struct btrfs_ordered_sum *sums;
927 LIST_HEAD(list);
928
929 ret = btrfs_lookup_csums_range(root, bytenr, bytenr + num_bytes - 1,
930 &list);
931 if (ret == 0 && list_empty(&list))
932 return 0;
933
934 while (!list_empty(&list)) {
935 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
936 list_del(&sums->list);
937 kfree(sums);
938 }
939 return 1;
940}
941
d352ac68
CM		 942/*
943 * when nowcow writeback call back. This checks for snapshots or COW copies
944 * of the extents that exist in the file, and COWs the file as required.
945 *
946 * If no cow copies or snapshots exist, we write directly to the existing
947 * blocks on disk
948 */
c8b97818 949static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
771ed689
CM		 950 u64 start, u64 end, int *page_started, int force,
951 unsigned long *nr_written)
be20aa9d 952{
be20aa9d 953 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 954 struct btrfs_trans_handle *trans;
be20aa9d 955 struct extent_buffer *leaf;
be20aa9d 956 struct btrfs_path *path;
80ff3856 957 struct btrfs_file_extent_item *fi;
be20aa9d 958 struct btrfs_key found_key;
80ff3856
YZ		 959 u64 cow_start;
960 u64 cur_offset;
961 u64 extent_end;
962 u64 disk_bytenr;
963 u64 num_bytes;
964 int extent_type;
965 int ret;
d899e052 966 int type;
80ff3856
YZ		 967 int nocow;
968 int check_prev = 1;
be20aa9d
CM		 969
970 path = btrfs_alloc_path();
971 BUG_ON(!path);
7ea394f1
YZ		 972 trans = btrfs_join_transaction(root, 1);
973 BUG_ON(!trans);
be20aa9d 974
80ff3856
YZ		 975 cow_start = (u64)-1;
976 cur_offset = start;
977 while (1) {
978 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
979 cur_offset, 0);
980 BUG_ON(ret < 0);
981 if (ret > 0 && path->slots[0] > 0 && check_prev) {
982 leaf = path->nodes[0];
983 btrfs_item_key_to_cpu(leaf, &found_key,
984 path->slots[0] - 1);
985 if (found_key.objectid == inode->i_ino &&
986 found_key.type == BTRFS_EXTENT_DATA_KEY)
987 path->slots[0]--;
988 }
989 check_prev = 0;
990next_slot:
991 leaf = path->nodes[0];
992 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
993 ret = btrfs_next_leaf(root, path);
994 if (ret < 0)
995 BUG_ON(1);
996 if (ret > 0)
997 break;
998 leaf = path->nodes[0];
999 }
be20aa9d 1000
80ff3856
YZ		 1001 nocow = 0;
1002 disk_bytenr = 0;
17d217fe 1003 num_bytes = 0;
80ff3856
YZ		 1004 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1005
1006 if (found_key.objectid > inode->i_ino ||
1007 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1008 found_key.offset > end)
1009 break;
1010
1011 if (found_key.offset > cur_offset) {
1012 extent_end = found_key.offset;
1013 goto out_check;
1014 }
1015
1016 fi = btrfs_item_ptr(leaf, path->slots[0],
1017 struct btrfs_file_extent_item);
1018 extent_type = btrfs_file_extent_type(leaf, fi);
1019
d899e052
YZ		 1020 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1021 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856
YZ		 1022 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1023 extent_end = found_key.offset +
1024 btrfs_file_extent_num_bytes(leaf, fi);
1025 if (extent_end <= start) {
1026 path->slots[0]++;
1027 goto next_slot;
1028 }
17d217fe
YZ		 1029 if (disk_bytenr == 0)
1030 goto out_check;
80ff3856
YZ		 1031 if (btrfs_file_extent_compression(leaf, fi) ||
1032 btrfs_file_extent_encryption(leaf, fi) ||
1033 btrfs_file_extent_other_encoding(leaf, fi))
1034 goto out_check;
d899e052
YZ		 1035 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1036 goto out_check;
d2fb3437 1037 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1038 goto out_check;
17d217fe
YZ		 1039 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1040 disk_bytenr))
1041 goto out_check;
80ff3856 1042 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
17d217fe
YZ		 1043 disk_bytenr += cur_offset - found_key.offset;
1044 num_bytes = min(end + 1, extent_end) - cur_offset;
1045 /*
1046 * force cow if csum exists in the range.
1047 * this ensure that csum for a given extent are
1048 * either valid or do not exist.
1049 */
1050 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1051 goto out_check;
80ff3856
YZ		 1052 nocow = 1;
1053 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1054 extent_end = found_key.offset +
1055 btrfs_file_extent_inline_len(leaf, fi);
1056 extent_end = ALIGN(extent_end, root->sectorsize);
1057 } else {
1058 BUG_ON(1);
1059 }
1060out_check:
1061 if (extent_end <= start) {
1062 path->slots[0]++;
1063 goto next_slot;
1064 }
1065 if (!nocow) {
1066 if (cow_start == (u64)-1)
1067 cow_start = cur_offset;
1068 cur_offset = extent_end;
1069 if (cur_offset > end)
1070 break;
1071 path->slots[0]++;
1072 goto next_slot;
7ea394f1
YZ		 1073 }
1074
1075 btrfs_release_path(root, path);
80ff3856
YZ		 1076 if (cow_start != (u64)-1) {
1077 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1078 found_key.offset - 1, page_started,
1079 nr_written, 1);
80ff3856
YZ		 1080 BUG_ON(ret);
1081 cow_start = (u64)-1;
7ea394f1 1082 }
80ff3856 1083
d899e052
YZ		 1084 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1085 struct extent_map *em;
1086 struct extent_map_tree *em_tree;
1087 em_tree = &BTRFS_I(inode)->extent_tree;
1088 em = alloc_extent_map(GFP_NOFS);
1089 em->start = cur_offset;
445a6944 1090 em->orig_start = em->start;
d899e052
YZ		 1091 em->len = num_bytes;
1092 em->block_len = num_bytes;
1093 em->block_start = disk_bytenr;
1094 em->bdev = root->fs_info->fs_devices->latest_bdev;
1095 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1096 while (1) {
1097 spin_lock(&em_tree->lock);
1098 ret = add_extent_mapping(em_tree, em);
1099 spin_unlock(&em_tree->lock);
1100 if (ret != -EEXIST) {
1101 free_extent_map(em);
1102 break;
1103 }
1104 btrfs_drop_extent_cache(inode, em->start,
1105 em->start + em->len - 1, 0);
1106 }
1107 type = BTRFS_ORDERED_PREALLOC;
1108 } else {
1109 type = BTRFS_ORDERED_NOCOW;
1110 }
80ff3856
YZ		 1111
1112 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 1113 num_bytes, num_bytes, type);
1114 BUG_ON(ret);
771ed689 1115
d899e052
YZ		 1116 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1117 cur_offset, cur_offset + num_bytes - 1,
771ed689 1118 locked_page, 1, 1, 1, 0, 0, 0);
80ff3856
YZ		 1119 cur_offset = extent_end;
1120 if (cur_offset > end)
1121 break;
be20aa9d 1122 }
80ff3856
YZ		 1123 btrfs_release_path(root, path);
1124
1125 if (cur_offset <= end && cow_start == (u64)-1)
1126 cow_start = cur_offset;
1127 if (cow_start != (u64)-1) {
1128 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1129 page_started, nr_written, 1);
80ff3856
YZ		 1130 BUG_ON(ret);
1131 }
1132
1133 ret = btrfs_end_transaction(trans, root);
1134 BUG_ON(ret);
7ea394f1 1135 btrfs_free_path(path);
80ff3856 1136 return 0;
be20aa9d
CM		 1137}
1138
d352ac68
CM		 1139/*
1140 * extent_io.c call back to do delayed allocation processing
1141 */
c8b97818 1142static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1143 u64 start, u64 end, int *page_started,
1144 unsigned long *nr_written)
be20aa9d 1145{
be20aa9d 1146 int ret;
a2135011 1147
17d217fe 1148 if (btrfs_test_flag(inode, NODATACOW))
c8b97818 1149 ret = run_delalloc_nocow(inode, locked_page, start, end,
ce397c06 1150 page_started, 1, nr_written);
d899e052
YZ		 1151 else if (btrfs_test_flag(inode, PREALLOC))
1152 ret = run_delalloc_nocow(inode, locked_page, start, end,
ce397c06 1153 page_started, 0, nr_written);
be20aa9d 1154 else
771ed689
CM		 1155 ret = cow_file_range_async(inode, locked_page, start, end,
1156 page_started, nr_written);
1832a6d5 1157
b888db2b
CM		 1158 return ret;
1159}
1160
d352ac68
CM		 1161/*
1162 * extent_io.c set_bit_hook, used to track delayed allocation
1163 * bytes in this file, and to maintain the list of inodes that
1164 * have pending delalloc work to be done.
1165 */
b2950863 1166static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1167 unsigned long old, unsigned long bits)
291d673e 1168{
bcbfce8a 1169 unsigned long flags;
b0c68f8b 1170 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1171 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a 1172 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
9069218d 1173 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
291d673e 1174 root->fs_info->delalloc_bytes += end - start + 1;
ea8c2819
CM		 1175 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1176 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1177 &root->fs_info->delalloc_inodes);
1178 }
bcbfce8a 1179 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
291d673e
CM		 1180 }
1181 return 0;
1182}
1183
d352ac68
CM		 1184/*
1185 * extent_io.c clear_bit_hook, see set_bit_hook for why
1186 */
b2950863 1187static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1188 unsigned long old, unsigned long bits)
291d673e 1189{
b0c68f8b 1190 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1191 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a
CM		 1192 unsigned long flags;
1193
1194 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
b0c68f8b
CM		 1195 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1196 printk("warning: delalloc account %Lu %Lu\n",
1197 end - start + 1, root->fs_info->delalloc_bytes);
1198 root->fs_info->delalloc_bytes = 0;
9069218d 1199 BTRFS_I(inode)->delalloc_bytes = 0;
b0c68f8b
CM		 1200 } else {
1201 root->fs_info->delalloc_bytes -= end - start + 1;
9069218d 1202 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
b0c68f8b 1203 }
ea8c2819
CM		 1204 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1205 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1206 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1207 }
bcbfce8a 1208 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
291d673e
CM		 1209 }
1210 return 0;
1211}
1212
d352ac68
CM		 1213/*
1214 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1215 * we don't create bios that span stripes or chunks
1216 */
239b14b3 1217int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1218 size_t size, struct bio *bio,
1219 unsigned long bio_flags)
239b14b3
CM		 1220{
1221 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1222 struct btrfs_mapping_tree *map_tree;
a62b9401 1223 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1224 u64 length = 0;
1225 u64 map_length;
239b14b3
CM		 1226 int ret;
1227
771ed689
CM		 1228 if (bio_flags & EXTENT_BIO_COMPRESSED)
1229 return 0;
1230
f2d8d74d 1231 length = bio->bi_size;
239b14b3
CM		 1232 map_tree = &root->fs_info->mapping_tree;
1233 map_length = length;
cea9e445 1234 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1235 &map_length, NULL, 0);
cea9e445 1236
239b14b3 1237 if (map_length < length + size) {
239b14b3
CM		 1238 return 1;
1239 }
1240 return 0;
1241}
1242
d352ac68
CM		 1243/*
1244 * in order to insert checksums into the metadata in large chunks,
1245 * we wait until bio submission time. All the pages in the bio are
1246 * checksummed and sums are attached onto the ordered extent record.
1247 *
1248 * At IO completion time the cums attached on the ordered extent record
1249 * are inserted into the btree
1250 */
b2950863 1251static int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio,
c8b97818 1252 int mirror_num, unsigned long bio_flags)
065631f6 1253{
065631f6 1254 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1255 int ret = 0;
e015640f 1256
d20f7043 1257 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
44b8bd7e 1258 BUG_ON(ret);
4a69a410
CM		 1259 return 0;
1260}
e015640f 1261
4a69a410
CM		 1262/*
1263 * in order to insert checksums into the metadata in large chunks,
1264 * we wait until bio submission time. All the pages in the bio are
1265 * checksummed and sums are attached onto the ordered extent record.
1266 *
1267 * At IO completion time the cums attached on the ordered extent record
1268 * are inserted into the btree
1269 */
b2950863 1270static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
4a69a410
CM		 1271 int mirror_num, unsigned long bio_flags)
1272{
1273 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1274 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1275}
1276
d352ac68
CM		 1277/*
1278 * extent_io.c submission hook. This does the right thing for csum calculation on write,
1279 * or reading the csums from the tree before a read
1280 */
b2950863 1281static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 1282 int mirror_num, unsigned long bio_flags)
44b8bd7e
CM		 1283{
1284 struct btrfs_root *root = BTRFS_I(inode)->root;
1285 int ret = 0;
19b9bdb0 1286 int skip_sum;
44b8bd7e 1287
e6dcd2dc
CM		 1288 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1289 BUG_ON(ret);
065631f6 1290
17d217fe 1291 skip_sum = btrfs_test_flag(inode, NODATASUM);
7ea394f1 1292
4d1b5fb4 1293 if (!(rw & (1 << BIO_RW))) {
d20f7043 1294 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1295 return btrfs_submit_compressed_read(inode, bio,
1296 mirror_num, bio_flags);
d20f7043
CM		 1297 } else if (!skip_sum)
1298 btrfs_lookup_bio_sums(root, inode, bio, NULL);
4d1b5fb4 1299 goto mapit;
19b9bdb0 1300 } else if (!skip_sum) {
17d217fe
YZ		 1301 /* csum items have already been cloned */
1302 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1303 goto mapit;
19b9bdb0
CM		 1304 /* we're doing a write, do the async checksumming */
1305 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1306 inode, rw, bio, mirror_num,
4a69a410
CM		 1307 bio_flags, __btrfs_submit_bio_start,
1308 __btrfs_submit_bio_done);
19b9bdb0
CM		 1309 }
1310
0b86a832 1311mapit:
8b712842 1312 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1313}
6885f308 1314
d352ac68
CM		 1315/*
1316 * given a list of ordered sums record them in the inode. This happens
1317 * at IO completion time based on sums calculated at bio submission time.
1318 */
ba1da2f4 1319static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1320 struct inode *inode, u64 file_offset,
1321 struct list_head *list)
1322{
1323 struct list_head *cur;
1324 struct btrfs_ordered_sum *sum;
1325
1326 btrfs_set_trans_block_group(trans, inode);
ba1da2f4 1327 list_for_each(cur, list) {
e6dcd2dc 1328 sum = list_entry(cur, struct btrfs_ordered_sum, list);
d20f7043
CM		 1329 btrfs_csum_file_blocks(trans,
1330 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1331 }
1332 return 0;
1333}
1334
ea8c2819
CM		 1335int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1336{
771ed689
CM		 1337 if ((end & (PAGE_CACHE_SIZE - 1)) == 0) {
1338 WARN_ON(1);
1339 }
ea8c2819
CM		 1340 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1341 GFP_NOFS);
1342}
1343
d352ac68 1344/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1345struct btrfs_writepage_fixup {
1346 struct page *page;
1347 struct btrfs_work work;
1348};
1349
b2950863 1350static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1351{
1352 struct btrfs_writepage_fixup *fixup;
1353 struct btrfs_ordered_extent *ordered;
1354 struct page *page;
1355 struct inode *inode;
1356 u64 page_start;
1357 u64 page_end;
1358
1359 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1360 page = fixup->page;
4a096752 1361again:
247e743c
CM		 1362 lock_page(page);
1363 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1364 ClearPageChecked(page);
1365 goto out_page;
1366 }
1367
1368 inode = page->mapping->host;
1369 page_start = page_offset(page);
1370 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1371
1372 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
4a096752
CM		 1373
1374 /* already ordered? We're done */
1375 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1376 EXTENT_ORDERED, 0)) {
247e743c 1377 goto out;
4a096752
CM		 1378 }
1379
1380 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1381 if (ordered) {
1382 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1383 page_end, GFP_NOFS);
1384 unlock_page(page);
1385 btrfs_start_ordered_extent(inode, ordered, 1);
1386 goto again;
1387 }
247e743c 1388
ea8c2819 1389 btrfs_set_extent_delalloc(inode, page_start, page_end);
247e743c
CM		 1390 ClearPageChecked(page);
1391out:
1392 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1393out_page:
1394 unlock_page(page);
1395 page_cache_release(page);
1396}
1397
1398/*
1399 * There are a few paths in the higher layers of the kernel that directly
1400 * set the page dirty bit without asking the filesystem if it is a
1401 * good idea. This causes problems because we want to make sure COW
1402 * properly happens and the data=ordered rules are followed.
1403 *
c8b97818 1404 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1405 * hasn't been properly setup for IO. We kick off an async process
1406 * to fix it up. The async helper will wait for ordered extents, set
1407 * the delalloc bit and make it safe to write the page.
1408 */
b2950863 1409static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1410{
1411 struct inode *inode = page->mapping->host;
1412 struct btrfs_writepage_fixup *fixup;
1413 struct btrfs_root *root = BTRFS_I(inode)->root;
1414 int ret;
1415
1416 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1417 EXTENT_ORDERED, 0);
1418 if (ret)
1419 return 0;
1420
1421 if (PageChecked(page))
1422 return -EAGAIN;
1423
1424 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1425 if (!fixup)
1426 return -EAGAIN;
f421950f 1427
247e743c
CM		 1428 SetPageChecked(page);
1429 page_cache_get(page);
1430 fixup->work.func = btrfs_writepage_fixup_worker;
1431 fixup->page = page;
1432 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1433 return -EAGAIN;
1434}
1435
d899e052
YZ		 1436static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1437 struct inode *inode, u64 file_pos,
1438 u64 disk_bytenr, u64 disk_num_bytes,
1439 u64 num_bytes, u64 ram_bytes,
1440 u8 compression, u8 encryption,
1441 u16 other_encoding, int extent_type)
1442{
1443 struct btrfs_root *root = BTRFS_I(inode)->root;
1444 struct btrfs_file_extent_item *fi;
1445 struct btrfs_path *path;
1446 struct extent_buffer *leaf;
1447 struct btrfs_key ins;
1448 u64 hint;
1449 int ret;
1450
1451 path = btrfs_alloc_path();
1452 BUG_ON(!path);
1453
1454 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1455 file_pos + num_bytes, file_pos, &hint);
1456 BUG_ON(ret);
1457
1458 ins.objectid = inode->i_ino;
1459 ins.offset = file_pos;
1460 ins.type = BTRFS_EXTENT_DATA_KEY;
1461 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1462 BUG_ON(ret);
1463 leaf = path->nodes[0];
1464 fi = btrfs_item_ptr(leaf, path->slots[0],
1465 struct btrfs_file_extent_item);
1466 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1467 btrfs_set_file_extent_type(leaf, fi, extent_type);
1468 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1469 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1470 btrfs_set_file_extent_offset(leaf, fi, 0);
1471 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1472 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1473 btrfs_set_file_extent_compression(leaf, fi, compression);
1474 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1475 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1476 btrfs_mark_buffer_dirty(leaf);
1477
1478 inode_add_bytes(inode, num_bytes);
1479 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1480
1481 ins.objectid = disk_bytenr;
1482 ins.offset = disk_num_bytes;
1483 ins.type = BTRFS_EXTENT_ITEM_KEY;
1484 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1485 root->root_key.objectid,
1486 trans->transid, inode->i_ino, &ins);
1487 BUG_ON(ret);
1488
1489 btrfs_free_path(path);
1490 return 0;
1491}
1492
d352ac68
CM		 1493/* as ordered data IO finishes, this gets called so we can finish
1494 * an ordered extent if the range of bytes in the file it covers are
1495 * fully written.
1496 */
211f90e6 1497static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1498{
e6dcd2dc
CM		 1499 struct btrfs_root *root = BTRFS_I(inode)->root;
1500 struct btrfs_trans_handle *trans;
1501 struct btrfs_ordered_extent *ordered_extent;
1502 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
d899e052 1503 int compressed = 0;
e6dcd2dc
CM		 1504 int ret;
1505
1506 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
ba1da2f4 1507 if (!ret)
e6dcd2dc 1508 return 0;
e6dcd2dc 1509
f9295749 1510 trans = btrfs_join_transaction(root, 1);
e6dcd2dc
CM		 1511
1512 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1513 BUG_ON(!ordered_extent);
7ea394f1
YZ		 1514 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1515 goto nocow;
e6dcd2dc
CM		 1516
1517 lock_extent(io_tree, ordered_extent->file_offset,
1518 ordered_extent->file_offset + ordered_extent->len - 1,
1519 GFP_NOFS);
1520
c8b97818 1521 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
d899e052
YZ		 1522 compressed = 1;
1523 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1524 BUG_ON(compressed);
1525 ret = btrfs_mark_extent_written(trans, root, inode,
1526 ordered_extent->file_offset,
1527 ordered_extent->file_offset +
1528 ordered_extent->len);
1529 BUG_ON(ret);
1530 } else {
1531 ret = insert_reserved_file_extent(trans, inode,
1532 ordered_extent->file_offset,
1533 ordered_extent->start,
1534 ordered_extent->disk_len,
1535 ordered_extent->len,
1536 ordered_extent->len,
1537 compressed, 0, 0,
1538 BTRFS_FILE_EXTENT_REG);
1539 BUG_ON(ret);
1540 }
e6dcd2dc
CM		 1541 unlock_extent(io_tree, ordered_extent->file_offset,
1542 ordered_extent->file_offset + ordered_extent->len - 1,
1543 GFP_NOFS);
7ea394f1 1544nocow:
e6dcd2dc
CM		 1545 add_pending_csums(trans, inode, ordered_extent->file_offset,
1546 &ordered_extent->list);
1547
34353029 1548 mutex_lock(&BTRFS_I(inode)->extent_mutex);
dbe674a9 1549 btrfs_ordered_update_i_size(inode, ordered_extent);
e02119d5 1550 btrfs_update_inode(trans, root, inode);
e6dcd2dc 1551 btrfs_remove_ordered_extent(inode, ordered_extent);
34353029 1552 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
7f3c74fb 1553
e6dcd2dc
CM		 1554 /* once for us */
1555 btrfs_put_ordered_extent(ordered_extent);
1556 /* once for the tree */
1557 btrfs_put_ordered_extent(ordered_extent);
1558
e6dcd2dc
CM		 1559 btrfs_end_transaction(trans, root);
1560 return 0;
1561}
1562
b2950863 1563static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1564 struct extent_state *state, int uptodate)
1565{
1566 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1567}
1568
d352ac68
CM		 1569/*
1570 * When IO fails, either with EIO or csum verification fails, we
1571 * try other mirrors that might have a good copy of the data. This
1572 * io_failure_record is used to record state as we go through all the
1573 * mirrors. If another mirror has good data, the page is set up to date
1574 * and things continue. If a good mirror can't be found, the original
1575 * bio end_io callback is called to indicate things have failed.
1576 */
7e38326f
CM		 1577struct io_failure_record {
1578 struct page *page;
1579 u64 start;
1580 u64 len;
1581 u64 logical;
d20f7043 1582 unsigned long bio_flags;
7e38326f
CM		 1583 int last_mirror;
1584};
1585
b2950863 1586static int btrfs_io_failed_hook(struct bio *failed_bio,
1259ab75
CM		 1587 struct page *page, u64 start, u64 end,
1588 struct extent_state *state)
7e38326f
CM		 1589{
1590 struct io_failure_record *failrec = NULL;
1591 u64 private;
1592 struct extent_map *em;
1593 struct inode *inode = page->mapping->host;
1594 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1595 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM		 1596 struct bio *bio;
1597 int num_copies;
1598 int ret;
1259ab75 1599 int rw;
7e38326f
CM		 1600 u64 logical;
1601
1602 ret = get_state_private(failure_tree, start, &private);
1603 if (ret) {
7e38326f
CM		 1604 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1605 if (!failrec)
1606 return -ENOMEM;
1607 failrec->start = start;
1608 failrec->len = end - start + 1;
1609 failrec->last_mirror = 0;
d20f7043 1610 failrec->bio_flags = 0;
7e38326f 1611
3b951516
CM		 1612 spin_lock(&em_tree->lock);
1613 em = lookup_extent_mapping(em_tree, start, failrec->len);
1614 if (em->start > start || em->start + em->len < start) {
1615 free_extent_map(em);
1616 em = NULL;
1617 }
1618 spin_unlock(&em_tree->lock);
7e38326f
CM		 1619
1620 if (!em || IS_ERR(em)) {
1621 kfree(failrec);
1622 return -EIO;
1623 }
1624 logical = start - em->start;
1625 logical = em->block_start + logical;
d20f7043
CM		 1626 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1627 logical = em->block_start;
1628 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1629 }
7e38326f
CM		 1630 failrec->logical = logical;
1631 free_extent_map(em);
1632 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1633 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM		 1634 set_state_private(failure_tree, start,
1635 (u64)(unsigned long)failrec);
7e38326f 1636 } else {
587f7704 1637 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM		 1638 }
1639 num_copies = btrfs_num_copies(
1640 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1641 failrec->logical, failrec->len);
1642 failrec->last_mirror++;
1643 if (!state) {
1644 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1645 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1646 failrec->start,
1647 EXTENT_LOCKED);
1648 if (state && state->start != failrec->start)
1649 state = NULL;
1650 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1651 }
1652 if (!state || failrec->last_mirror > num_copies) {
1653 set_state_private(failure_tree, failrec->start, 0);
1654 clear_extent_bits(failure_tree, failrec->start,
1655 failrec->start + failrec->len - 1,
1656 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1657 kfree(failrec);
1658 return -EIO;
1659 }
1660 bio = bio_alloc(GFP_NOFS, 1);
1661 bio->bi_private = state;
1662 bio->bi_end_io = failed_bio->bi_end_io;
1663 bio->bi_sector = failrec->logical >> 9;
1664 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1665 bio->bi_size = 0;
d20f7043 1666
7e38326f 1667 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1259ab75
CM		 1668 if (failed_bio->bi_rw & (1 << BIO_RW))
1669 rw = WRITE;
1670 else
1671 rw = READ;
1672
1673 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818 1674 failrec->last_mirror,
d20f7043 1675 failrec->bio_flags);
1259ab75
CM		 1676 return 0;
1677}
1678
d352ac68
CM		 1679/*
1680 * each time an IO finishes, we do a fast check in the IO failure tree
1681 * to see if we need to process or clean up an io_failure_record
1682 */
b2950863 1683static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1259ab75
CM		 1684{
1685 u64 private;
1686 u64 private_failure;
1687 struct io_failure_record *failure;
1688 int ret;
1689
1690 private = 0;
1691 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1692 (u64)-1, 1, EXTENT_DIRTY)) {
1693 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1694 start, &private_failure);
1695 if (ret == 0) {
1696 failure = (struct io_failure_record *)(unsigned long)
1697 private_failure;
1698 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1699 failure->start, 0);
1700 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1701 failure->start,
1702 failure->start + failure->len - 1,
1703 EXTENT_DIRTY | EXTENT_LOCKED,
1704 GFP_NOFS);
1705 kfree(failure);
1706 }
1707 }
7e38326f
CM		 1708 return 0;
1709}
1710
d352ac68
CM		 1711/*
1712 * when reads are done, we need to check csums to verify the data is correct
1713 * if there's a match, we allow the bio to finish. If not, we go through
1714 * the io_failure_record routines to find good copies
1715 */
b2950863 1716static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1717 struct extent_state *state)
07157aac 1718{
35ebb934 1719 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1720 struct inode *inode = page->mapping->host;
d1310b2e 1721 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1722 char *kaddr;
aadfeb6e 1723 u64 private = ~(u32)0;
07157aac 1724 int ret;
ff79f819
CM		 1725 struct btrfs_root *root = BTRFS_I(inode)->root;
1726 u32 csum = ~(u32)0;
bbf0d006 1727 unsigned long flags;
d1310b2e 1728
d20f7043
CM		 1729 if (PageChecked(page)) {
1730 ClearPageChecked(page);
1731 goto good;
1732 }
17d217fe
YZ		 1733 if (btrfs_test_flag(inode, NODATASUM))
1734 return 0;
1735
1736 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1737 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1738 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1739 GFP_NOFS);
b6cda9bc 1740 return 0;
17d217fe 1741 }
d20f7043 1742
c2e639f0 1743 if (state && state->start == start) {
70dec807
CM		 1744 private = state->private;
1745 ret = 0;
1746 } else {
1747 ret = get_state_private(io_tree, start, &private);
1748 }
bbf0d006 1749 local_irq_save(flags);
07157aac
CM		 1750 kaddr = kmap_atomic(page, KM_IRQ0);
1751 if (ret) {
1752 goto zeroit;
1753 }
ff79f819
CM		 1754 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1755 btrfs_csum_final(csum, (char *)&csum);
1756 if (csum != private) {
07157aac
CM		 1757 goto zeroit;
1758 }
1759 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1760 local_irq_restore(flags);
d20f7043 1761good:
7e38326f
CM		 1762 /* if the io failure tree for this inode is non-empty,
1763 * check to see if we've recovered from a failed IO
1764 */
1259ab75 1765 btrfs_clean_io_failures(inode, start);
07157aac
CM		 1766 return 0;
1767
1768zeroit:
aadfeb6e
CM		 1769 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1770 page->mapping->host->i_ino, (unsigned long long)start, csum,
1771 private);
db94535d
CM		 1772 memset(kaddr + offset, 1, end - start + 1);
1773 flush_dcache_page(page);
07157aac 1774 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1775 local_irq_restore(flags);
3b951516
CM		 1776 if (private == 0)
1777 return 0;
7e38326f 1778 return -EIO;
07157aac 1779}
b888db2b 1780
7b128766
JB		 1781/*
1782 * This creates an orphan entry for the given inode in case something goes
1783 * wrong in the middle of an unlink/truncate.
1784 */
1785int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1786{
1787 struct btrfs_root *root = BTRFS_I(inode)->root;
1788 int ret = 0;
1789
bcc63abb 1790 spin_lock(&root->list_lock);
7b128766
JB		 1791
1792 /* already on the orphan list, we're good */
1793 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1794 spin_unlock(&root->list_lock);
7b128766
JB		 1795 return 0;
1796 }
1797
1798 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1799
bcc63abb 1800 spin_unlock(&root->list_lock);
7b128766
JB		 1801
1802 /*
1803 * insert an orphan item to track this unlinked/truncated file
1804 */
1805 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1806
1807 return ret;
1808}
1809
1810/*
1811 * We have done the truncate/delete so we can go ahead and remove the orphan
1812 * item for this particular inode.
1813 */
1814int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1815{
1816 struct btrfs_root *root = BTRFS_I(inode)->root;
1817 int ret = 0;
1818
bcc63abb 1819 spin_lock(&root->list_lock);
7b128766
JB		 1820
1821 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1822 spin_unlock(&root->list_lock);
7b128766
JB		 1823 return 0;
1824 }
1825
1826 list_del_init(&BTRFS_I(inode)->i_orphan);
1827 if (!trans) {
bcc63abb 1828 spin_unlock(&root->list_lock);
7b128766
JB		 1829 return 0;
1830 }
1831
bcc63abb 1832 spin_unlock(&root->list_lock);
7b128766
JB		 1833
1834 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1835
1836 return ret;
1837}
1838
1839/*
1840 * this cleans up any orphans that may be left on the list from the last use
1841 * of this root.
1842 */
1843void btrfs_orphan_cleanup(struct btrfs_root *root)
1844{
1845 struct btrfs_path *path;
1846 struct extent_buffer *leaf;
1847 struct btrfs_item *item;
1848 struct btrfs_key key, found_key;
1849 struct btrfs_trans_handle *trans;
1850 struct inode *inode;
1851 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1852
7b128766
JB		 1853 path = btrfs_alloc_path();
1854 if (!path)
1855 return;
1856 path->reada = -1;
1857
1858 key.objectid = BTRFS_ORPHAN_OBJECTID;
1859 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1860 key.offset = (u64)-1;
1861
7b128766
JB		 1862
1863 while (1) {
1864 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1865 if (ret < 0) {
1866 printk(KERN_ERR "Error searching slot for orphan: %d"
1867 "\n", ret);
1868 break;
1869 }
1870
1871 /*
1872 * if ret == 0 means we found what we were searching for, which
1873 * is weird, but possible, so only screw with path if we didnt
1874 * find the key and see if we have stuff that matches
1875 */
1876 if (ret > 0) {
1877 if (path->slots[0] == 0)
1878 break;
1879 path->slots[0]--;
1880 }
1881
1882 /* pull out the item */
1883 leaf = path->nodes[0];
1884 item = btrfs_item_nr(leaf, path->slots[0]);
1885 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1886
1887 /* make sure the item matches what we want */
1888 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1889 break;
1890 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1891 break;
1892
1893 /* release the path since we're done with it */
1894 btrfs_release_path(root, path);
1895
1896 /*
1897 * this is where we are basically btrfs_lookup, without the
1898 * crossing root thing. we store the inode number in the
1899 * offset of the orphan item.
1900 */
5b21f2ed 1901 inode = btrfs_iget_locked(root->fs_info->sb,
7b128766
JB		 1902 found_key.offset, root);
1903 if (!inode)
1904 break;
1905
1906 if (inode->i_state & I_NEW) {
1907 BTRFS_I(inode)->root = root;
1908
1909 /* have to set the location manually */
1910 BTRFS_I(inode)->location.objectid = inode->i_ino;
1911 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1912 BTRFS_I(inode)->location.offset = 0;
1913
1914 btrfs_read_locked_inode(inode);
1915 unlock_new_inode(inode);
1916 }
1917
1918 /*
1919 * add this inode to the orphan list so btrfs_orphan_del does
1920 * the proper thing when we hit it
1921 */
bcc63abb 1922 spin_lock(&root->list_lock);
7b128766 1923 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
bcc63abb 1924 spin_unlock(&root->list_lock);
7b128766
JB		 1925
1926 /*
1927 * if this is a bad inode, means we actually succeeded in
1928 * removing the inode, but not the orphan record, which means
1929 * we need to manually delete the orphan since iput will just
1930 * do a destroy_inode
1931 */
1932 if (is_bad_inode(inode)) {
5b21f2ed 1933 trans = btrfs_start_transaction(root, 1);
7b128766 1934 btrfs_orphan_del(trans, inode);
5b21f2ed 1935 btrfs_end_transaction(trans, root);
7b128766
JB		 1936 iput(inode);
1937 continue;
1938 }
1939
1940 /* if we have links, this was a truncate, lets do that */
1941 if (inode->i_nlink) {
1942 nr_truncate++;
1943 btrfs_truncate(inode);
1944 } else {
1945 nr_unlink++;
1946 }
1947
1948 /* this will do delete_inode and everything for us */
1949 iput(inode);
1950 }
1951
1952 if (nr_unlink)
1953 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1954 if (nr_truncate)
1955 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1956
1957 btrfs_free_path(path);
7b128766
JB		 1958}
1959
d352ac68
CM		 1960/*
1961 * read an inode from the btree into the in-memory inode
1962 */
39279cc3
CM		 1963void btrfs_read_locked_inode(struct inode *inode)
1964{
1965 struct btrfs_path *path;
5f39d397 1966 struct extent_buffer *leaf;
39279cc3 1967 struct btrfs_inode_item *inode_item;
0b86a832 1968 struct btrfs_timespec *tspec;
39279cc3
CM		 1969 struct btrfs_root *root = BTRFS_I(inode)->root;
1970 struct btrfs_key location;
1971 u64 alloc_group_block;
618e21d5 1972 u32 rdev;
39279cc3
CM		 1973 int ret;
1974
1975 path = btrfs_alloc_path();
1976 BUG_ON(!path);
39279cc3 1977 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 1978
39279cc3 1979 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 1980 if (ret)
39279cc3 1981 goto make_bad;
39279cc3 1982
5f39d397
CM		 1983 leaf = path->nodes[0];
1984 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1985 struct btrfs_inode_item);
1986
1987 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1988 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1989 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1990 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 1991 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 1992
1993 tspec = btrfs_inode_atime(inode_item);
1994 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1995 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1996
1997 tspec = btrfs_inode_mtime(inode_item);
1998 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1999 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2000
2001 tspec = btrfs_inode_ctime(inode_item);
2002 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2003 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2004
a76a3cd4 2005 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2006 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2007 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2008 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2009 inode->i_rdev = 0;
5f39d397
CM		 2010 rdev = btrfs_inode_rdev(leaf, inode_item);
2011
aec7477b 2012 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2013 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
aec7477b 2014
5f39d397 2015 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
d2fb3437
YZ		 2016 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2017 alloc_group_block, 0);
39279cc3
CM		 2018 btrfs_free_path(path);
2019 inode_item = NULL;
2020
39279cc3 2021 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2022 case S_IFREG:
2023 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2024 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2025 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2026 inode->i_fop = &btrfs_file_operations;
2027 inode->i_op = &btrfs_file_inode_operations;
2028 break;
2029 case S_IFDIR:
2030 inode->i_fop = &btrfs_dir_file_operations;
2031 if (root == root->fs_info->tree_root)
2032 inode->i_op = &btrfs_dir_ro_inode_operations;
2033 else
2034 inode->i_op = &btrfs_dir_inode_operations;
2035 break;
2036 case S_IFLNK:
2037 inode->i_op = &btrfs_symlink_inode_operations;
2038 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2039 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2040 break;
618e21d5
JB		 2041 default:
2042 init_special_inode(inode, inode->i_mode, rdev);
2043 break;
39279cc3
CM		 2044 }
2045 return;
2046
2047make_bad:
39279cc3 2048 btrfs_free_path(path);
39279cc3
CM		 2049 make_bad_inode(inode);
2050}
2051
d352ac68
CM		 2052/*
2053 * given a leaf and an inode, copy the inode fields into the leaf
2054 */
e02119d5
CM		 2055static void fill_inode_item(struct btrfs_trans_handle *trans,
2056 struct extent_buffer *leaf,
5f39d397 2057 struct btrfs_inode_item *item,
39279cc3
CM		 2058 struct inode *inode)
2059{
5f39d397
CM		 2060 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2061 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2062 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2063 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2064 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2065
2066 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2067 inode->i_atime.tv_sec);
2068 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2069 inode->i_atime.tv_nsec);
2070
2071 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2072 inode->i_mtime.tv_sec);
2073 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2074 inode->i_mtime.tv_nsec);
2075
2076 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2077 inode->i_ctime.tv_sec);
2078 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2079 inode->i_ctime.tv_nsec);
2080
a76a3cd4 2081 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2082 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2083 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2084 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2085 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2086 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d2fb3437 2087 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
39279cc3
CM		 2088}
2089
d352ac68
CM		 2090/*
2091 * copy everything in the in-memory inode into the btree.
2092 */
ba1da2f4 2093int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
39279cc3
CM		 2094 struct btrfs_root *root,
2095 struct inode *inode)
2096{
2097 struct btrfs_inode_item *inode_item;
2098 struct btrfs_path *path;
5f39d397 2099 struct extent_buffer *leaf;
39279cc3
CM		 2100 int ret;
2101
2102 path = btrfs_alloc_path();
2103 BUG_ON(!path);
39279cc3
CM		 2104 ret = btrfs_lookup_inode(trans, root, path,
2105 &BTRFS_I(inode)->location, 1);
2106 if (ret) {
2107 if (ret > 0)
2108 ret = -ENOENT;
2109 goto failed;
2110 }
2111
5f39d397
CM		 2112 leaf = path->nodes[0];
2113 inode_item = btrfs_item_ptr(leaf, path->slots[0],
39279cc3
CM		 2114 struct btrfs_inode_item);
2115
e02119d5 2116 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2117 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2118 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2119 ret = 0;
2120failed:
39279cc3
CM		 2121 btrfs_free_path(path);
2122 return ret;
2123}
2124
2125
d352ac68
CM		 2126/*
2127 * unlink helper that gets used here in inode.c and in the tree logging
2128 * recovery code. It remove a link in a directory with a given name, and
2129 * also drops the back refs in the inode to the directory
2130 */
e02119d5
CM		 2131int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2132 struct btrfs_root *root,
2133 struct inode *dir, struct inode *inode,
2134 const char *name, int name_len)
39279cc3
CM		 2135{
2136 struct btrfs_path *path;
39279cc3 2137 int ret = 0;
5f39d397 2138 struct extent_buffer *leaf;
39279cc3 2139 struct btrfs_dir_item *di;
5f39d397 2140 struct btrfs_key key;
aec7477b 2141 u64 index;
39279cc3
CM		 2142
2143 path = btrfs_alloc_path();
54aa1f4d
CM		 2144 if (!path) {
2145 ret = -ENOMEM;
2146 goto err;
2147 }
2148
39279cc3
CM		 2149 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2150 name, name_len, -1);
2151 if (IS_ERR(di)) {
2152 ret = PTR_ERR(di);
2153 goto err;
2154 }
2155 if (!di) {
2156 ret = -ENOENT;
2157 goto err;
2158 }
5f39d397
CM		 2159 leaf = path->nodes[0];
2160 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2161 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2162 if (ret)
2163 goto err;
39279cc3
CM		 2164 btrfs_release_path(root, path);
2165
aec7477b 2166 ret = btrfs_del_inode_ref(trans, root, name, name_len,
e02119d5
CM		 2167 inode->i_ino,
2168 dir->i_ino, &index);
aec7477b
JB		 2169 if (ret) {
2170 printk("failed to delete reference to %.*s, "
2171 "inode %lu parent %lu\n", name_len, name,
e02119d5 2172 inode->i_ino, dir->i_ino);
aec7477b
JB		 2173 goto err;
2174 }
2175
39279cc3 2176 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
aec7477b 2177 index, name, name_len, -1);
39279cc3
CM		 2178 if (IS_ERR(di)) {
2179 ret = PTR_ERR(di);
2180 goto err;
2181 }
2182 if (!di) {
2183 ret = -ENOENT;
2184 goto err;
2185 }
2186 ret = btrfs_delete_one_dir_name(trans, root, path, di);
925baedd 2187 btrfs_release_path(root, path);
39279cc3 2188
e02119d5
CM		 2189 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2190 inode, dir->i_ino);
49eb7e46
CM		 2191 BUG_ON(ret != 0 && ret != -ENOENT);
2192 if (ret != -ENOENT)
2193 BTRFS_I(dir)->log_dirty_trans = trans->transid;
e02119d5
CM		 2194
2195 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2196 dir, index);
2197 BUG_ON(ret);
39279cc3
CM		 2198err:
2199 btrfs_free_path(path);
e02119d5
CM		 2200 if (ret)
2201 goto out;
2202
2203 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2204 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2205 btrfs_update_inode(trans, root, dir);
2206 btrfs_drop_nlink(inode);
2207 ret = btrfs_update_inode(trans, root, inode);
2208 dir->i_sb->s_dirt = 1;
2209out:
39279cc3
CM		 2210 return ret;
2211}
2212
2213static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2214{
2215 struct btrfs_root *root;
2216 struct btrfs_trans_handle *trans;
7b128766 2217 struct inode *inode = dentry->d_inode;
39279cc3 2218 int ret;
1832a6d5 2219 unsigned long nr = 0;
39279cc3
CM		 2220
2221 root = BTRFS_I(dir)->root;
1832a6d5
CM		 2222
2223 ret = btrfs_check_free_space(root, 1, 1);
2224 if (ret)
2225 goto fail;
2226
39279cc3 2227 trans = btrfs_start_transaction(root, 1);
5f39d397 2228
39279cc3 2229 btrfs_set_trans_block_group(trans, dir);
e02119d5
CM		 2230 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2231 dentry->d_name.name, dentry->d_name.len);
7b128766
JB		 2232
2233 if (inode->i_nlink == 0)
2234 ret = btrfs_orphan_add(trans, inode);
2235
d3c2fdcf 2236 nr = trans->blocks_used;
5f39d397 2237
89ce8a63 2238 btrfs_end_transaction_throttle(trans, root);
1832a6d5 2239fail:
d3c2fdcf 2240 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2241 return ret;
2242}
2243
2244static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2245{
2246 struct inode *inode = dentry->d_inode;
1832a6d5 2247 int err = 0;
39279cc3
CM		 2248 int ret;
2249 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2250 struct btrfs_trans_handle *trans;
1832a6d5 2251 unsigned long nr = 0;
39279cc3 2252
3394e160
CM		 2253 /*
2254 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2255 * the root of a subvolume or snapshot
2256 */
2257 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2258 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
134d4512 2259 return -ENOTEMPTY;
925baedd 2260 }
134d4512 2261
1832a6d5
CM		 2262 ret = btrfs_check_free_space(root, 1, 1);
2263 if (ret)
2264 goto fail;
2265
39279cc3
CM		 2266 trans = btrfs_start_transaction(root, 1);
2267 btrfs_set_trans_block_group(trans, dir);
39279cc3 2268
7b128766
JB		 2269 err = btrfs_orphan_add(trans, inode);
2270 if (err)
2271 goto fail_trans;
2272
39279cc3 2273 /* now the directory is empty */
e02119d5
CM		 2274 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2275 dentry->d_name.name, dentry->d_name.len);
39279cc3 2276 if (!err) {
dbe674a9 2277 btrfs_i_size_write(inode, 0);
39279cc3 2278 }
3954401f 2279
7b128766 2280fail_trans:
d3c2fdcf 2281 nr = trans->blocks_used;
89ce8a63 2282 ret = btrfs_end_transaction_throttle(trans, root);
1832a6d5 2283fail:
d3c2fdcf 2284 btrfs_btree_balance_dirty(root, nr);
3954401f 2285
39279cc3
CM		 2286 if (ret && !err)
2287 err = ret;
2288 return err;
2289}
2290
d20f7043 2291#if 0
323ac95b
CM		 2292/*
2293 * when truncating bytes in a file, it is possible to avoid reading
2294 * the leaves that contain only checksum items. This can be the
2295 * majority of the IO required to delete a large file, but it must
2296 * be done carefully.
2297 *
2298 * The keys in the level just above the leaves are checked to make sure
2299 * the lowest key in a given leaf is a csum key, and starts at an offset
2300 * after the new size.
2301 *
2302 * Then the key for the next leaf is checked to make sure it also has
2303 * a checksum item for the same file. If it does, we know our target leaf
2304 * contains only checksum items, and it can be safely freed without reading
2305 * it.
2306 *
2307 * This is just an optimization targeted at large files. It may do
2308 * nothing. It will return 0 unless things went badly.
2309 */
2310static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2311 struct btrfs_root *root,
2312 struct btrfs_path *path,
2313 struct inode *inode, u64 new_size)
2314{
2315 struct btrfs_key key;
2316 int ret;
2317 int nritems;
2318 struct btrfs_key found_key;
2319 struct btrfs_key other_key;
5b84e8d6
YZ		 2320 struct btrfs_leaf_ref *ref;
2321 u64 leaf_gen;
2322 u64 leaf_start;
323ac95b
CM		 2323
2324 path->lowest_level = 1;
2325 key.objectid = inode->i_ino;
2326 key.type = BTRFS_CSUM_ITEM_KEY;
2327 key.offset = new_size;
2328again:
2329 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2330 if (ret < 0)
2331 goto out;
2332
2333 if (path->nodes[1] == NULL) {
2334 ret = 0;
2335 goto out;
2336 }
2337 ret = 0;
2338 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2339 nritems = btrfs_header_nritems(path->nodes[1]);
2340
2341 if (!nritems)
2342 goto out;
2343
2344 if (path->slots[1] >= nritems)
2345 goto next_node;
2346
2347 /* did we find a key greater than anything we want to delete? */
2348 if (found_key.objectid > inode->i_ino ||
2349 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2350 goto out;
2351
2352 /* we check the next key in the node to make sure the leave contains
2353 * only checksum items. This comparison doesn't work if our
2354 * leaf is the last one in the node
2355 */
2356 if (path->slots[1] + 1 >= nritems) {
2357next_node:
2358 /* search forward from the last key in the node, this
2359 * will bring us into the next node in the tree
2360 */
2361 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2362
2363 /* unlikely, but we inc below, so check to be safe */
2364 if (found_key.offset == (u64)-1)
2365 goto out;
2366
2367 /* search_forward needs a path with locks held, do the
2368 * search again for the original key. It is possible
2369 * this will race with a balance and return a path that
2370 * we could modify, but this drop is just an optimization
2371 * and is allowed to miss some leaves.
2372 */
2373 btrfs_release_path(root, path);
2374 found_key.offset++;
2375
2376 /* setup a max key for search_forward */
2377 other_key.offset = (u64)-1;
2378 other_key.type = key.type;
2379 other_key.objectid = key.objectid;
2380
2381 path->keep_locks = 1;
2382 ret = btrfs_search_forward(root, &found_key, &other_key,
2383 path, 0, 0);
2384 path->keep_locks = 0;
2385 if (ret || found_key.objectid != key.objectid ||
2386 found_key.type != key.type) {
2387 ret = 0;
2388 goto out;
2389 }
2390
2391 key.offset = found_key.offset;
2392 btrfs_release_path(root, path);
2393 cond_resched();
2394 goto again;
2395 }
2396
2397 /* we know there's one more slot after us in the tree,
2398 * read that key so we can verify it is also a checksum item
2399 */
2400 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2401
2402 if (found_key.objectid < inode->i_ino)
2403 goto next_key;
2404
2405 if (found_key.type != key.type || found_key.offset < new_size)
2406 goto next_key;
2407
2408 /*
2409 * if the key for the next leaf isn't a csum key from this objectid,
2410 * we can't be sure there aren't good items inside this leaf.
2411 * Bail out
2412 */
2413 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2414 goto out;
2415
5b84e8d6
YZ		 2416 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2417 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
323ac95b
CM		 2418 /*
2419 * it is safe to delete this leaf, it contains only
2420 * csum items from this inode at an offset >= new_size
2421 */
5b84e8d6 2422 ret = btrfs_del_leaf(trans, root, path, leaf_start);
323ac95b
CM		 2423 BUG_ON(ret);
2424
5b84e8d6
YZ		 2425 if (root->ref_cows && leaf_gen < trans->transid) {
2426 ref = btrfs_alloc_leaf_ref(root, 0);
2427 if (ref) {
2428 ref->root_gen = root->root_key.offset;
2429 ref->bytenr = leaf_start;
2430 ref->owner = 0;
2431 ref->generation = leaf_gen;
2432 ref->nritems = 0;
2433
2434 ret = btrfs_add_leaf_ref(root, ref, 0);
2435 WARN_ON(ret);
2436 btrfs_free_leaf_ref(root, ref);
2437 } else {
2438 WARN_ON(1);
2439 }
2440 }
323ac95b
CM		 2441next_key:
2442 btrfs_release_path(root, path);
2443
2444 if (other_key.objectid == inode->i_ino &&
2445 other_key.type == key.type && other_key.offset > key.offset) {
2446 key.offset = other_key.offset;
2447 cond_resched();
2448 goto again;
2449 }
2450 ret = 0;
2451out:
2452 /* fixup any changes we've made to the path */
2453 path->lowest_level = 0;
2454 path->keep_locks = 0;
2455 btrfs_release_path(root, path);
2456 return ret;
2457}
2458
d20f7043
CM		 2459#endif
2460
39279cc3
CM		 2461/*
2462 * this can truncate away extent items, csum items and directory items.
2463 * It starts at a high offset and removes keys until it can't find
d352ac68 2464 * any higher than new_size
39279cc3
CM		 2465 *
2466 * csum items that cross the new i_size are truncated to the new size
2467 * as well.
7b128766
JB		 2468 *
2469 * min_type is the minimum key type to truncate down to. If set to 0, this
2470 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2471 */
e02119d5
CM		 2472noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2473 struct btrfs_root *root,
2474 struct inode *inode,
2475 u64 new_size, u32 min_type)
39279cc3
CM		 2476{
2477 int ret;
2478 struct btrfs_path *path;
2479 struct btrfs_key key;
5f39d397 2480 struct btrfs_key found_key;
39279cc3 2481 u32 found_type;
5f39d397 2482 struct extent_buffer *leaf;
39279cc3
CM		 2483 struct btrfs_file_extent_item *fi;
2484 u64 extent_start = 0;
db94535d 2485 u64 extent_num_bytes = 0;
39279cc3 2486 u64 item_end = 0;
7bb86316 2487 u64 root_gen = 0;
d8d5f3e1 2488 u64 root_owner = 0;
39279cc3
CM		 2489 int found_extent;
2490 int del_item;
85e21bac
CM		 2491 int pending_del_nr = 0;
2492 int pending_del_slot = 0;
179e29e4 2493 int extent_type = -1;
771ed689 2494 int encoding;
3b951516 2495 u64 mask = root->sectorsize - 1;
39279cc3 2496
e02119d5 2497 if (root->ref_cows)
5b21f2ed 2498 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
39279cc3 2499 path = btrfs_alloc_path();
3c69faec 2500 path->reada = -1;
39279cc3 2501 BUG_ON(!path);
5f39d397 2502
39279cc3
CM		 2503 /* FIXME, add redo link to tree so we don't leak on crash */
2504 key.objectid = inode->i_ino;
2505 key.offset = (u64)-1;
5f39d397
CM		 2506 key.type = (u8)-1;
2507
85e21bac 2508 btrfs_init_path(path);
323ac95b 2509
85e21bac
CM		 2510search_again:
2511 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2512 if (ret < 0) {
2513 goto error;
2514 }
2515 if (ret > 0) {
e02119d5
CM		 2516 /* there are no items in the tree for us to truncate, we're
2517 * done
2518 */
2519 if (path->slots[0] == 0) {
2520 ret = 0;
2521 goto error;
2522 }
85e21bac
CM		 2523 path->slots[0]--;
2524 }
2525
39279cc3 2526 while(1) {
39279cc3 2527 fi = NULL;
5f39d397
CM		 2528 leaf = path->nodes[0];
2529 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2530 found_type = btrfs_key_type(&found_key);
771ed689 2531 encoding = 0;
39279cc3 2532
5f39d397 2533 if (found_key.objectid != inode->i_ino)
39279cc3 2534 break;
5f39d397 2535
85e21bac 2536 if (found_type < min_type)
39279cc3
CM		 2537 break;
2538
5f39d397 2539 item_end = found_key.offset;
39279cc3 2540 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 2541 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 2542 struct btrfs_file_extent_item);
179e29e4 2543 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM		 2544 encoding = btrfs_file_extent_compression(leaf, fi);
2545 encoding |= btrfs_file_extent_encryption(leaf, fi);
2546 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2547
179e29e4 2548 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 2549 item_end +=
db94535d 2550 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 2551 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 2552 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 2553 fi);
39279cc3 2554 }
008630c1 2555 item_end--;
39279cc3 2556 }
e02119d5 2557 if (item_end < new_size) {
b888db2b
CM		 2558 if (found_type == BTRFS_DIR_ITEM_KEY) {
2559 found_type = BTRFS_INODE_ITEM_KEY;
2560 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
d20f7043 2561 found_type = BTRFS_EXTENT_DATA_KEY;
85e21bac
CM		 2562 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2563 found_type = BTRFS_XATTR_ITEM_KEY;
2564 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2565 found_type = BTRFS_INODE_REF_KEY;
b888db2b
CM		 2566 } else if (found_type) {
2567 found_type--;
2568 } else {
2569 break;
39279cc3 2570 }
a61721d5 2571 btrfs_set_key_type(&key, found_type);
85e21bac 2572 goto next;
39279cc3 2573 }
e02119d5 2574 if (found_key.offset >= new_size)
39279cc3
CM		 2575 del_item = 1;
2576 else
2577 del_item = 0;
2578 found_extent = 0;
2579
2580 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 2581 if (found_type != BTRFS_EXTENT_DATA_KEY)
2582 goto delete;
2583
2584 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 2585 u64 num_dec;
db94535d 2586 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 2587 if (!del_item && !encoding) {
db94535d
CM		 2588 u64 orig_num_bytes =
2589 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 2590 extent_num_bytes = new_size -
5f39d397 2591 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 2592 extent_num_bytes = extent_num_bytes &
2593 ~((u64)root->sectorsize - 1);
db94535d
CM		 2594 btrfs_set_file_extent_num_bytes(leaf, fi,
2595 extent_num_bytes);
2596 num_dec = (orig_num_bytes -
9069218d 2597 extent_num_bytes);
e02119d5 2598 if (root->ref_cows && extent_start != 0)
a76a3cd4 2599 inode_sub_bytes(inode, num_dec);
5f39d397 2600 btrfs_mark_buffer_dirty(leaf);
39279cc3 2601 } else {
db94535d
CM		 2602 extent_num_bytes =
2603 btrfs_file_extent_disk_num_bytes(leaf,
2604 fi);
39279cc3 2605 /* FIXME blocksize != 4096 */
9069218d 2606 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 2607 if (extent_start != 0) {
2608 found_extent = 1;
e02119d5 2609 if (root->ref_cows)
a76a3cd4 2610 inode_sub_bytes(inode, num_dec);
e02119d5 2611 }
31840ae1 2612 root_gen = btrfs_header_generation(leaf);
d8d5f3e1 2613 root_owner = btrfs_header_owner(leaf);
39279cc3 2614 }
9069218d 2615 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 2616 /*
2617 * we can't truncate inline items that have had
2618 * special encodings
2619 */
2620 if (!del_item &&
2621 btrfs_file_extent_compression(leaf, fi) == 0 &&
2622 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2623 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 2624 u32 size = new_size - found_key.offset;
2625
2626 if (root->ref_cows) {
a76a3cd4
YZ		 2627 inode_sub_bytes(inode, item_end + 1 -
2628 new_size);
e02119d5
CM		 2629 }
2630 size =
2631 btrfs_file_extent_calc_inline_size(size);
9069218d 2632 ret = btrfs_truncate_item(trans, root, path,
e02119d5 2633 size, 1);
9069218d 2634 BUG_ON(ret);
e02119d5 2635 } else if (root->ref_cows) {
a76a3cd4
YZ		 2636 inode_sub_bytes(inode, item_end + 1 -
2637 found_key.offset);
9069218d 2638 }
39279cc3 2639 }
179e29e4 2640delete:
39279cc3 2641 if (del_item) {
85e21bac
CM		 2642 if (!pending_del_nr) {
2643 /* no pending yet, add ourselves */
2644 pending_del_slot = path->slots[0];
2645 pending_del_nr = 1;
2646 } else if (pending_del_nr &&
2647 path->slots[0] + 1 == pending_del_slot) {
2648 /* hop on the pending chunk */
2649 pending_del_nr++;
2650 pending_del_slot = path->slots[0];
2651 } else {
2652 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2653 }
39279cc3
CM		 2654 } else {
2655 break;
2656 }
39279cc3
CM		 2657 if (found_extent) {
2658 ret = btrfs_free_extent(trans, root, extent_start,
7bb86316 2659 extent_num_bytes,
31840ae1 2660 leaf->start, root_owner,
3bb1a1bc 2661 root_gen, inode->i_ino, 0);
39279cc3
CM		 2662 BUG_ON(ret);
2663 }
85e21bac
CM		 2664next:
2665 if (path->slots[0] == 0) {
2666 if (pending_del_nr)
2667 goto del_pending;
2668 btrfs_release_path(root, path);
2669 goto search_again;
2670 }
2671
2672 path->slots[0]--;
2673 if (pending_del_nr &&
2674 path->slots[0] + 1 != pending_del_slot) {
2675 struct btrfs_key debug;
2676del_pending:
2677 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2678 pending_del_slot);
2679 ret = btrfs_del_items(trans, root, path,
2680 pending_del_slot,
2681 pending_del_nr);
2682 BUG_ON(ret);
2683 pending_del_nr = 0;
2684 btrfs_release_path(root, path);
2685 goto search_again;
2686 }
39279cc3
CM		 2687 }
2688 ret = 0;
2689error:
85e21bac
CM		 2690 if (pending_del_nr) {
2691 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2692 pending_del_nr);
2693 }
39279cc3
CM		 2694 btrfs_free_path(path);
2695 inode->i_sb->s_dirt = 1;
2696 return ret;
2697}
2698
2699/*
2700 * taken from block_truncate_page, but does cow as it zeros out
2701 * any bytes left in the last page in the file.
2702 */
2703static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2704{
2705 struct inode *inode = mapping->host;
db94535d 2706 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 2707 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2708 struct btrfs_ordered_extent *ordered;
2709 char *kaddr;
db94535d 2710 u32 blocksize = root->sectorsize;
39279cc3
CM		 2711 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2712 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2713 struct page *page;
39279cc3 2714 int ret = 0;
a52d9a80 2715 u64 page_start;
e6dcd2dc 2716 u64 page_end;
39279cc3
CM		 2717
2718 if ((offset & (blocksize - 1)) == 0)
2719 goto out;
2720
2721 ret = -ENOMEM;
211c17f5 2722again:
39279cc3
CM		 2723 page = grab_cache_page(mapping, index);
2724 if (!page)
2725 goto out;
e6dcd2dc
CM		 2726
2727 page_start = page_offset(page);
2728 page_end = page_start + PAGE_CACHE_SIZE - 1;
2729
39279cc3 2730 if (!PageUptodate(page)) {
9ebefb18 2731 ret = btrfs_readpage(NULL, page);
39279cc3 2732 lock_page(page);
211c17f5
CM		 2733 if (page->mapping != mapping) {
2734 unlock_page(page);
2735 page_cache_release(page);
2736 goto again;
2737 }
39279cc3
CM		 2738 if (!PageUptodate(page)) {
2739 ret = -EIO;
89642229 2740 goto out_unlock;
39279cc3
CM		 2741 }
2742 }
211c17f5 2743 wait_on_page_writeback(page);
e6dcd2dc
CM		 2744
2745 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2746 set_page_extent_mapped(page);
2747
2748 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2749 if (ordered) {
2750 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2751 unlock_page(page);
2752 page_cache_release(page);
eb84ae03 2753 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 2754 btrfs_put_ordered_extent(ordered);
2755 goto again;
2756 }
2757
ea8c2819 2758 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc
CM		 2759 ret = 0;
2760 if (offset != PAGE_CACHE_SIZE) {
2761 kaddr = kmap(page);
2762 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2763 flush_dcache_page(page);
2764 kunmap(page);
2765 }
247e743c 2766 ClearPageChecked(page);
e6dcd2dc
CM		 2767 set_page_dirty(page);
2768 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
39279cc3 2769
89642229 2770out_unlock:
39279cc3
CM		 2771 unlock_page(page);
2772 page_cache_release(page);
2773out:
2774 return ret;
2775}
2776
9036c102 2777int btrfs_cont_expand(struct inode *inode, loff_t size)
39279cc3 2778{
9036c102
YZ		 2779 struct btrfs_trans_handle *trans;
2780 struct btrfs_root *root = BTRFS_I(inode)->root;
2781 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2782 struct extent_map *em;
2783 u64 mask = root->sectorsize - 1;
2784 u64 hole_start = (inode->i_size + mask) & ~mask;
2785 u64 block_end = (size + mask) & ~mask;
2786 u64 last_byte;
2787 u64 cur_offset;
2788 u64 hole_size;
39279cc3
CM		 2789 int err;
2790
9036c102
YZ		 2791 if (size <= hole_start)
2792 return 0;
2793
2794 err = btrfs_check_free_space(root, 1, 0);
39279cc3
CM		 2795 if (err)
2796 return err;
2797
9036c102 2798 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2bf5a725 2799
9036c102
YZ		 2800 while (1) {
2801 struct btrfs_ordered_extent *ordered;
2802 btrfs_wait_ordered_range(inode, hole_start,
2803 block_end - hole_start);
2804 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2805 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2806 if (!ordered)
2807 break;
2808 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2809 btrfs_put_ordered_extent(ordered);
2810 }
39279cc3 2811
9036c102
YZ		 2812 trans = btrfs_start_transaction(root, 1);
2813 btrfs_set_trans_block_group(trans, inode);
39279cc3 2814
9036c102
YZ		 2815 cur_offset = hole_start;
2816 while (1) {
2817 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2818 block_end - cur_offset, 0);
2819 BUG_ON(IS_ERR(em) || !em);
2820 last_byte = min(extent_map_end(em), block_end);
2821 last_byte = (last_byte + mask) & ~mask;
2822 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
771ed689 2823 u64 hint_byte = 0;
9036c102 2824 hole_size = last_byte - cur_offset;
771ed689
CM		 2825 err = btrfs_drop_extents(trans, root, inode,
2826 cur_offset,
2827 cur_offset + hole_size,
2828 cur_offset, &hint_byte);
2829 if (err)
2830 break;
9036c102
YZ		 2831 err = btrfs_insert_file_extent(trans, root,
2832 inode->i_ino, cur_offset, 0,
2833 0, hole_size, 0, hole_size,
2834 0, 0, 0);
2835 btrfs_drop_extent_cache(inode, hole_start,
2836 last_byte - 1, 0);
2837 }
2838 free_extent_map(em);
2839 cur_offset = last_byte;
2840 if (err || cur_offset >= block_end)
2841 break;
2842 }
1832a6d5 2843
9036c102
YZ		 2844 btrfs_end_transaction(trans, root);
2845 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2846 return err;
2847}
39279cc3 2848
9036c102
YZ		 2849static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2850{
2851 struct inode *inode = dentry->d_inode;
2852 int err;
39279cc3 2853
9036c102
YZ		 2854 err = inode_change_ok(inode, attr);
2855 if (err)
2856 return err;
2bf5a725 2857
9036c102
YZ		 2858 if (S_ISREG(inode->i_mode) &&
2859 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2860 err = btrfs_cont_expand(inode, attr->ia_size);
54aa1f4d
CM		 2861 if (err)
2862 return err;
39279cc3 2863 }
9036c102 2864
39279cc3 2865 err = inode_setattr(inode, attr);
33268eaf
JB		 2866
2867 if (!err && ((attr->ia_valid & ATTR_MODE)))
2868 err = btrfs_acl_chmod(inode);
39279cc3
CM		 2869 return err;
2870}
61295eb8 2871
39279cc3
CM		 2872void btrfs_delete_inode(struct inode *inode)
2873{
2874 struct btrfs_trans_handle *trans;
2875 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 2876 unsigned long nr;
39279cc3
CM		 2877 int ret;
2878
2879 truncate_inode_pages(&inode->i_data, 0);
2880 if (is_bad_inode(inode)) {
7b128766 2881 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 2882 goto no_delete;
2883 }
4a096752 2884 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 2885
dbe674a9 2886 btrfs_i_size_write(inode, 0);
39279cc3 2887 trans = btrfs_start_transaction(root, 1);
5f39d397 2888
39279cc3 2889 btrfs_set_trans_block_group(trans, inode);
e02119d5 2890 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
7b128766
JB		 2891 if (ret) {
2892 btrfs_orphan_del(NULL, inode);
54aa1f4d 2893 goto no_delete_lock;
7b128766
JB		 2894 }
2895
2896 btrfs_orphan_del(trans, inode);
85e21bac 2897
d3c2fdcf 2898 nr = trans->blocks_used;
85e21bac 2899 clear_inode(inode);
5f39d397 2900
39279cc3 2901 btrfs_end_transaction(trans, root);
d3c2fdcf 2902 btrfs_btree_balance_dirty(root, nr);
39279cc3 2903 return;
54aa1f4d
CM		 2904
2905no_delete_lock:
d3c2fdcf 2906 nr = trans->blocks_used;
54aa1f4d 2907 btrfs_end_transaction(trans, root);
d3c2fdcf 2908 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2909no_delete:
2910 clear_inode(inode);
2911}
2912
2913/*
2914 * this returns the key found in the dir entry in the location pointer.
2915 * If no dir entries were found, location->objectid is 0.
2916 */
2917static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2918 struct btrfs_key *location)
2919{
2920 const char *name = dentry->d_name.name;
2921 int namelen = dentry->d_name.len;
2922 struct btrfs_dir_item *di;
2923 struct btrfs_path *path;
2924 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 2925 int ret = 0;
39279cc3
CM		 2926
2927 path = btrfs_alloc_path();
2928 BUG_ON(!path);
3954401f 2929
39279cc3
CM		 2930 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2931 namelen, 0);
0d9f7f3e
Y		 2932 if (IS_ERR(di))
2933 ret = PTR_ERR(di);
39279cc3 2934 if (!di || IS_ERR(di)) {
3954401f 2935 goto out_err;
39279cc3 2936 }
5f39d397 2937 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 2938out:
39279cc3
CM		 2939 btrfs_free_path(path);
2940 return ret;
3954401f
CM		 2941out_err:
2942 location->objectid = 0;
2943 goto out;
39279cc3
CM		 2944}
2945
2946/*
2947 * when we hit a tree root in a directory, the btrfs part of the inode
2948 * needs to be changed to reflect the root directory of the tree root. This
2949 * is kind of like crossing a mount point.
2950 */
2951static int fixup_tree_root_location(struct btrfs_root *root,
2952 struct btrfs_key *location,
58176a96
JB		 2953 struct btrfs_root **sub_root,
2954 struct dentry *dentry)
39279cc3 2955{
39279cc3
CM		 2956 struct btrfs_root_item *ri;
2957
2958 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2959 return 0;
2960 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2961 return 0;
2962
58176a96
JB		 2963 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2964 dentry->d_name.name,
2965 dentry->d_name.len);
39279cc3
CM		 2966 if (IS_ERR(*sub_root))
2967 return PTR_ERR(*sub_root);
2968
2969 ri = &(*sub_root)->root_item;
2970 location->objectid = btrfs_root_dirid(ri);
39279cc3
CM		 2971 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2972 location->offset = 0;
2973
39279cc3
CM		 2974 return 0;
2975}
2976
e02119d5 2977static noinline void init_btrfs_i(struct inode *inode)
39279cc3 2978{
e02119d5
CM		 2979 struct btrfs_inode *bi = BTRFS_I(inode);
2980
2981 bi->i_acl = NULL;
2982 bi->i_default_acl = NULL;
2983
2984 bi->generation = 0;
c3027eb5 2985 bi->sequence = 0;
e02119d5
CM		 2986 bi->last_trans = 0;
2987 bi->logged_trans = 0;
2988 bi->delalloc_bytes = 0;
2989 bi->disk_i_size = 0;
2990 bi->flags = 0;
2991 bi->index_cnt = (u64)-1;
49eb7e46 2992 bi->log_dirty_trans = 0;
d1310b2e
CM		 2993 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2994 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
b888db2b 2995 inode->i_mapping, GFP_NOFS);
7e38326f
CM		 2996 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2997 inode->i_mapping, GFP_NOFS);
ea8c2819 2998 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
ba1da2f4 2999 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
ee6e6504 3000 mutex_init(&BTRFS_I(inode)->extent_mutex);
e02119d5
CM		 3001 mutex_init(&BTRFS_I(inode)->log_mutex);
3002}
3003
3004static int btrfs_init_locked_inode(struct inode *inode, void *p)
3005{
3006 struct btrfs_iget_args *args = p;
3007 inode->i_ino = args->ino;
3008 init_btrfs_i(inode);
3009 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 3010 return 0;
3011}
3012
3013static int btrfs_find_actor(struct inode *inode, void *opaque)
3014{
3015 struct btrfs_iget_args *args = opaque;
3016 return (args->ino == inode->i_ino &&
3017 args->root == BTRFS_I(inode)->root);
3018}
3019
5b21f2ed
ZY		 3020struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3021 struct btrfs_root *root, int wait)
3022{
3023 struct inode *inode;
3024 struct btrfs_iget_args args;
3025 args.ino = objectid;
3026 args.root = root;
3027
3028 if (wait) {
3029 inode = ilookup5(s, objectid, btrfs_find_actor,
3030 (void *)&args);
3031 } else {
3032 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3033 (void *)&args);
3034 }
3035 return inode;
3036}
3037
39279cc3
CM		 3038struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3039 struct btrfs_root *root)
3040{
3041 struct inode *inode;
3042 struct btrfs_iget_args args;
3043 args.ino = objectid;
3044 args.root = root;
3045
3046 inode = iget5_locked(s, objectid, btrfs_find_actor,
3047 btrfs_init_locked_inode,
3048 (void *)&args);
3049 return inode;
3050}
3051
1a54ef8c
BR		 3052/* Get an inode object given its location and corresponding root.
3053 * Returns in *is_new if the inode was read from disk
3054 */
3055struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3056 struct btrfs_root *root, int *is_new)
3057{
3058 struct inode *inode;
3059
3060 inode = btrfs_iget_locked(s, location->objectid, root);
3061 if (!inode)
3062 return ERR_PTR(-EACCES);
3063
3064 if (inode->i_state & I_NEW) {
3065 BTRFS_I(inode)->root = root;
3066 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3067 btrfs_read_locked_inode(inode);
3068 unlock_new_inode(inode);
3069 if (is_new)
3070 *is_new = 1;
3071 } else {
3072 if (is_new)
3073 *is_new = 0;
3074 }
3075
3076 return inode;
3077}
3078
3de4586c 3079struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3
CM		 3080{
3081 struct inode * inode;
3082 struct btrfs_inode *bi = BTRFS_I(dir);
3083 struct btrfs_root *root = bi->root;
3084 struct btrfs_root *sub_root = root;
3085 struct btrfs_key location;
c146afad 3086 int ret, new;
39279cc3
CM		 3087
3088 if (dentry->d_name.len > BTRFS_NAME_LEN)
3089 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3090
39279cc3 3091 ret = btrfs_inode_by_name(dir, dentry, &location);
5f39d397 3092
39279cc3
CM		 3093 if (ret < 0)
3094 return ERR_PTR(ret);
5f39d397 3095
39279cc3
CM		 3096 inode = NULL;
3097 if (location.objectid) {
58176a96
JB		 3098 ret = fixup_tree_root_location(root, &location, &sub_root,
3099 dentry);
39279cc3
CM		 3100 if (ret < 0)
3101 return ERR_PTR(ret);
3102 if (ret > 0)
3103 return ERR_PTR(-ENOENT);
1a54ef8c
BR		 3104 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3105 if (IS_ERR(inode))
3106 return ERR_CAST(inode);
39279cc3 3107 }
3de4586c
CM		 3108 return inode;
3109}
3110
3111static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3112 struct nameidata *nd)
3113{
3114 struct inode *inode;
3115
3116 if (dentry->d_name.len > BTRFS_NAME_LEN)
3117 return ERR_PTR(-ENAMETOOLONG);
3118
3119 inode = btrfs_lookup_dentry(dir, dentry);
3120 if (IS_ERR(inode))
3121 return ERR_CAST(inode);
7b128766 3122
39279cc3
CM		 3123 return d_splice_alias(inode, dentry);
3124}
3125
39279cc3
CM		 3126static unsigned char btrfs_filetype_table[] = {
3127 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3128};
3129
cbdf5a24
DW		 3130static int btrfs_real_readdir(struct file *filp, void *dirent,
3131 filldir_t filldir)
39279cc3 3132{
6da6abae 3133 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 3134 struct btrfs_root *root = BTRFS_I(inode)->root;
3135 struct btrfs_item *item;
3136 struct btrfs_dir_item *di;
3137 struct btrfs_key key;
5f39d397 3138 struct btrfs_key found_key;
39279cc3
CM		 3139 struct btrfs_path *path;
3140 int ret;
3141 u32 nritems;
5f39d397 3142 struct extent_buffer *leaf;
39279cc3
CM		 3143 int slot;
3144 int advance;
3145 unsigned char d_type;
3146 int over = 0;
3147 u32 di_cur;
3148 u32 di_total;
3149 u32 di_len;
3150 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 3151 char tmp_name[32];
3152 char *name_ptr;
3153 int name_len;
39279cc3
CM		 3154
3155 /* FIXME, use a real flag for deciding about the key type */
3156 if (root->fs_info->tree_root == root)
3157 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 3158
3954401f
CM		 3159 /* special case for "." */
3160 if (filp->f_pos == 0) {
3161 over = filldir(dirent, ".", 1,
3162 1, inode->i_ino,
3163 DT_DIR);
3164 if (over)
3165 return 0;
3166 filp->f_pos = 1;
3167 }
3954401f
CM		 3168 /* special case for .., just use the back ref */
3169 if (filp->f_pos == 1) {
5ecc7e5d 3170 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 3171 over = filldir(dirent, "..", 2,
5ecc7e5d 3172 2, pino, DT_DIR);
3954401f 3173 if (over)
49593bfa 3174 return 0;
3954401f
CM		 3175 filp->f_pos = 2;
3176 }
49593bfa
DW		 3177 path = btrfs_alloc_path();
3178 path->reada = 2;
3179
39279cc3
CM		 3180 btrfs_set_key_type(&key, key_type);
3181 key.offset = filp->f_pos;
49593bfa 3182 key.objectid = inode->i_ino;
5f39d397 3183
39279cc3
CM		 3184 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3185 if (ret < 0)
3186 goto err;
3187 advance = 0;
49593bfa
DW		 3188
3189 while (1) {
5f39d397
CM		 3190 leaf = path->nodes[0];
3191 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3192 slot = path->slots[0];
3193 if (advance || slot >= nritems) {
49593bfa 3194 if (slot >= nritems - 1) {
39279cc3
CM		 3195 ret = btrfs_next_leaf(root, path);
3196 if (ret)
3197 break;
5f39d397
CM		 3198 leaf = path->nodes[0];
3199 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3200 slot = path->slots[0];
3201 } else {
3202 slot++;
3203 path->slots[0]++;
3204 }
3205 }
3de4586c 3206
39279cc3 3207 advance = 1;
5f39d397
CM		 3208 item = btrfs_item_nr(leaf, slot);
3209 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3210
3211 if (found_key.objectid != key.objectid)
39279cc3 3212 break;
5f39d397 3213 if (btrfs_key_type(&found_key) != key_type)
39279cc3 3214 break;
5f39d397 3215 if (found_key.offset < filp->f_pos)
39279cc3 3216 continue;
5f39d397
CM		 3217
3218 filp->f_pos = found_key.offset;
49593bfa 3219
39279cc3
CM		 3220 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3221 di_cur = 0;
5f39d397 3222 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 3223
3224 while (di_cur < di_total) {
5f39d397
CM		 3225 struct btrfs_key location;
3226
3227 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 3228 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 3229 name_ptr = tmp_name;
3230 } else {
3231 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 3232 if (!name_ptr) {
3233 ret = -ENOMEM;
3234 goto err;
3235 }
5f39d397
CM		 3236 }
3237 read_extent_buffer(leaf, name_ptr,
3238 (unsigned long)(di + 1), name_len);
3239
3240 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3241 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c
CM		 3242
3243 /* is this a reference to our own snapshot? If so
3244 * skip it
3245 */
3246 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3247 location.objectid == root->root_key.objectid) {
3248 over = 0;
3249 goto skip;
3250 }
5f39d397 3251 over = filldir(dirent, name_ptr, name_len,
49593bfa 3252 found_key.offset, location.objectid,
39279cc3 3253 d_type);
5f39d397 3254
3de4586c 3255skip:
5f39d397
CM		 3256 if (name_ptr != tmp_name)
3257 kfree(name_ptr);
3258
39279cc3
CM		 3259 if (over)
3260 goto nopos;
5103e947 3261 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 3262 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 3263 di_cur += di_len;
3264 di = (struct btrfs_dir_item *)((char *)di + di_len);
3265 }
3266 }
49593bfa
DW		 3267
3268 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07
YZ		 3269 if (key_type == BTRFS_DIR_INDEX_KEY)
3270 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3271 else
3272 filp->f_pos++;
39279cc3
CM		 3273nopos:
3274 ret = 0;
3275err:
39279cc3 3276 btrfs_free_path(path);
39279cc3
CM		 3277 return ret;
3278}
3279
3280int btrfs_write_inode(struct inode *inode, int wait)
3281{
3282 struct btrfs_root *root = BTRFS_I(inode)->root;
3283 struct btrfs_trans_handle *trans;
3284 int ret = 0;
3285
c146afad 3286 if (root->fs_info->btree_inode == inode)
4ca8b41e
CM		 3287 return 0;
3288
39279cc3 3289 if (wait) {
f9295749 3290 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3291 btrfs_set_trans_block_group(trans, inode);
3292 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 3293 }
3294 return ret;
3295}
3296
3297/*
54aa1f4d 3298 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 3299 * inode changes. But, it is most likely to find the inode in cache.
3300 * FIXME, needs more benchmarking...there are no reasons other than performance
3301 * to keep or drop this code.
3302 */
3303void btrfs_dirty_inode(struct inode *inode)
3304{
3305 struct btrfs_root *root = BTRFS_I(inode)->root;
3306 struct btrfs_trans_handle *trans;
3307
f9295749 3308 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3309 btrfs_set_trans_block_group(trans, inode);
3310 btrfs_update_inode(trans, root, inode);
3311 btrfs_end_transaction(trans, root);
39279cc3
CM		 3312}
3313
d352ac68
CM		 3314/*
3315 * find the highest existing sequence number in a directory
3316 * and then set the in-memory index_cnt variable to reflect
3317 * free sequence numbers
3318 */
aec7477b
JB		 3319static int btrfs_set_inode_index_count(struct inode *inode)
3320{
3321 struct btrfs_root *root = BTRFS_I(inode)->root;
3322 struct btrfs_key key, found_key;
3323 struct btrfs_path *path;
3324 struct extent_buffer *leaf;
3325 int ret;
3326
3327 key.objectid = inode->i_ino;
3328 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3329 key.offset = (u64)-1;
3330
3331 path = btrfs_alloc_path();
3332 if (!path)
3333 return -ENOMEM;
3334
3335 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3336 if (ret < 0)
3337 goto out;
3338 /* FIXME: we should be able to handle this */
3339 if (ret == 0)
3340 goto out;
3341 ret = 0;
3342
3343 /*
3344 * MAGIC NUMBER EXPLANATION:
3345 * since we search a directory based on f_pos we have to start at 2
3346 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3347 * else has to start at 2
3348 */
3349 if (path->slots[0] == 0) {
3350 BTRFS_I(inode)->index_cnt = 2;
3351 goto out;
3352 }
3353
3354 path->slots[0]--;
3355
3356 leaf = path->nodes[0];
3357 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3358
3359 if (found_key.objectid != inode->i_ino ||
3360 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3361 BTRFS_I(inode)->index_cnt = 2;
3362 goto out;
3363 }
3364
3365 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3366out:
3367 btrfs_free_path(path);
3368 return ret;
3369}
3370
d352ac68
CM		 3371/*
3372 * helper to find a free sequence number in a given directory. This current
3373 * code is very simple, later versions will do smarter things in the btree
3374 */
3de4586c 3375int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 3376{
3377 int ret = 0;
3378
3379 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3380 ret = btrfs_set_inode_index_count(dir);
8d5bf1cb 3381 if (ret) {
aec7477b 3382 return ret;
8d5bf1cb 3383 }
aec7477b
JB		 3384 }
3385
00e4e6b3 3386 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 3387 BTRFS_I(dir)->index_cnt++;
3388
3389 return ret;
3390}
3391
39279cc3
CM		 3392static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3393 struct btrfs_root *root,
aec7477b 3394 struct inode *dir,
9c58309d 3395 const char *name, int name_len,
d2fb3437
YZ		 3396 u64 ref_objectid, u64 objectid,
3397 u64 alloc_hint, int mode, u64 *index)
39279cc3
CM		 3398{
3399 struct inode *inode;
5f39d397 3400 struct btrfs_inode_item *inode_item;
39279cc3 3401 struct btrfs_key *location;
5f39d397 3402 struct btrfs_path *path;
9c58309d
CM		 3403 struct btrfs_inode_ref *ref;
3404 struct btrfs_key key[2];
3405 u32 sizes[2];
3406 unsigned long ptr;
39279cc3
CM		 3407 int ret;
3408 int owner;
3409
5f39d397
CM		 3410 path = btrfs_alloc_path();
3411 BUG_ON(!path);
3412
39279cc3
CM		 3413 inode = new_inode(root->fs_info->sb);
3414 if (!inode)
3415 return ERR_PTR(-ENOMEM);
3416
aec7477b 3417 if (dir) {
3de4586c 3418 ret = btrfs_set_inode_index(dir, index);
aec7477b
JB		 3419 if (ret)
3420 return ERR_PTR(ret);
aec7477b
JB		 3421 }
3422 /*
3423 * index_cnt is ignored for everything but a dir,
3424 * btrfs_get_inode_index_count has an explanation for the magic
3425 * number
3426 */
e02119d5 3427 init_btrfs_i(inode);
aec7477b 3428 BTRFS_I(inode)->index_cnt = 2;
39279cc3 3429 BTRFS_I(inode)->root = root;
e02119d5 3430 BTRFS_I(inode)->generation = trans->transid;
b888db2b 3431
39279cc3
CM		 3432 if (mode & S_IFDIR)
3433 owner = 0;
3434 else
3435 owner = 1;
d2fb3437
YZ		 3436 BTRFS_I(inode)->block_group =
3437 btrfs_find_block_group(root, 0, alloc_hint, owner);
17d217fe
YZ		 3438 if ((mode & S_IFREG)) {
3439 if (btrfs_test_opt(root, NODATASUM))
3440 btrfs_set_flag(inode, NODATASUM);
3441 if (btrfs_test_opt(root, NODATACOW))
3442 btrfs_set_flag(inode, NODATACOW);
3443 }
9c58309d
CM		 3444
3445 key[0].objectid = objectid;
3446 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3447 key[0].offset = 0;
3448
3449 key[1].objectid = objectid;
3450 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3451 key[1].offset = ref_objectid;
3452
3453 sizes[0] = sizeof(struct btrfs_inode_item);
3454 sizes[1] = name_len + sizeof(*ref);
3455
3456 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3457 if (ret != 0)
5f39d397
CM		 3458 goto fail;
3459
9c58309d
CM		 3460 if (objectid > root->highest_inode)
3461 root->highest_inode = objectid;
3462
79683f2d
CM		 3463 inode->i_uid = current_fsuid();
3464 inode->i_gid = current_fsgid();
39279cc3
CM		 3465 inode->i_mode = mode;
3466 inode->i_ino = objectid;
a76a3cd4 3467 inode_set_bytes(inode, 0);
39279cc3 3468 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 3469 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3470 struct btrfs_inode_item);
e02119d5 3471 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 3472
3473 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3474 struct btrfs_inode_ref);
3475 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 3476 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 3477 ptr = (unsigned long)(ref + 1);
3478 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3479
5f39d397
CM		 3480 btrfs_mark_buffer_dirty(path->nodes[0]);
3481 btrfs_free_path(path);
3482
39279cc3
CM		 3483 location = &BTRFS_I(inode)->location;
3484 location->objectid = objectid;
39279cc3
CM		 3485 location->offset = 0;
3486 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3487
39279cc3
CM		 3488 insert_inode_hash(inode);
3489 return inode;
5f39d397 3490fail:
aec7477b
JB		 3491 if (dir)
3492 BTRFS_I(dir)->index_cnt--;
5f39d397
CM		 3493 btrfs_free_path(path);
3494 return ERR_PTR(ret);
39279cc3
CM		 3495}
3496
3497static inline u8 btrfs_inode_type(struct inode *inode)
3498{
3499 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3500}
3501
d352ac68
CM		 3502/*
3503 * utility function to add 'inode' into 'parent_inode' with
3504 * a give name and a given sequence number.
3505 * if 'add_backref' is true, also insert a backref from the
3506 * inode to the parent directory.
3507 */
e02119d5
CM		 3508int btrfs_add_link(struct btrfs_trans_handle *trans,
3509 struct inode *parent_inode, struct inode *inode,
3510 const char *name, int name_len, int add_backref, u64 index)
39279cc3
CM		 3511{
3512 int ret;
3513 struct btrfs_key key;
e02119d5 3514 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5f39d397 3515
39279cc3 3516 key.objectid = inode->i_ino;
39279cc3
CM		 3517 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3518 key.offset = 0;
3519
e02119d5
CM		 3520 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3521 parent_inode->i_ino,
aec7477b 3522 &key, btrfs_inode_type(inode),
00e4e6b3 3523 index);
39279cc3 3524 if (ret == 0) {
9c58309d
CM		 3525 if (add_backref) {
3526 ret = btrfs_insert_inode_ref(trans, root,
e02119d5
CM		 3527 name, name_len,
3528 inode->i_ino,
3529 parent_inode->i_ino,
3530 index);
9c58309d 3531 }
dbe674a9 3532 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 3533 name_len * 2);
79c44584 3534 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 3535 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 3536 }
3537 return ret;
3538}
3539
3540static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
9c58309d 3541 struct dentry *dentry, struct inode *inode,
00e4e6b3 3542 int backref, u64 index)
39279cc3 3543{
e02119d5
CM		 3544 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3545 inode, dentry->d_name.name,
3546 dentry->d_name.len, backref, index);
39279cc3
CM		 3547 if (!err) {
3548 d_instantiate(dentry, inode);
3549 return 0;
3550 }
3551 if (err > 0)
3552 err = -EEXIST;
3553 return err;
3554}
3555
618e21d5
JB		 3556static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3557 int mode, dev_t rdev)
3558{
3559 struct btrfs_trans_handle *trans;
3560 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3561 struct inode *inode = NULL;
618e21d5
JB		 3562 int err;
3563 int drop_inode = 0;
3564 u64 objectid;
1832a6d5 3565 unsigned long nr = 0;
00e4e6b3 3566 u64 index = 0;
618e21d5
JB		 3567
3568 if (!new_valid_dev(rdev))
3569 return -EINVAL;
3570
1832a6d5
CM		 3571 err = btrfs_check_free_space(root, 1, 0);
3572 if (err)
3573 goto fail;
3574
618e21d5
JB		 3575 trans = btrfs_start_transaction(root, 1);
3576 btrfs_set_trans_block_group(trans, dir);
3577
3578 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3579 if (err) {
3580 err = -ENOSPC;
3581 goto out_unlock;
3582 }
3583
aec7477b 3584 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3585 dentry->d_name.len,
3586 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3 3587 BTRFS_I(dir)->block_group, mode, &index);
618e21d5
JB		 3588 err = PTR_ERR(inode);
3589 if (IS_ERR(inode))
3590 goto out_unlock;
3591
33268eaf
JB		 3592 err = btrfs_init_acl(inode, dir);
3593 if (err) {
3594 drop_inode = 1;
3595 goto out_unlock;
3596 }
3597
618e21d5 3598 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3599 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
618e21d5
JB		 3600 if (err)
3601 drop_inode = 1;
3602 else {
3603 inode->i_op = &btrfs_special_inode_operations;
3604 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 3605 btrfs_update_inode(trans, root, inode);
618e21d5
JB		 3606 }
3607 dir->i_sb->s_dirt = 1;
3608 btrfs_update_inode_block_group(trans, inode);
3609 btrfs_update_inode_block_group(trans, dir);
3610out_unlock:
d3c2fdcf 3611 nr = trans->blocks_used;
89ce8a63 3612 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3613fail:
618e21d5
JB		 3614 if (drop_inode) {
3615 inode_dec_link_count(inode);
3616 iput(inode);
3617 }
d3c2fdcf 3618 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 3619 return err;
3620}
3621
39279cc3
CM		 3622static int btrfs_create(struct inode *dir, struct dentry *dentry,
3623 int mode, struct nameidata *nd)
3624{
3625 struct btrfs_trans_handle *trans;
3626 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3627 struct inode *inode = NULL;
39279cc3
CM		 3628 int err;
3629 int drop_inode = 0;
1832a6d5 3630 unsigned long nr = 0;
39279cc3 3631 u64 objectid;
00e4e6b3 3632 u64 index = 0;
39279cc3 3633
1832a6d5
CM		 3634 err = btrfs_check_free_space(root, 1, 0);
3635 if (err)
3636 goto fail;
39279cc3
CM		 3637 trans = btrfs_start_transaction(root, 1);
3638 btrfs_set_trans_block_group(trans, dir);
3639
3640 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3641 if (err) {
3642 err = -ENOSPC;
3643 goto out_unlock;
3644 }
3645
aec7477b 3646 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3647 dentry->d_name.len,
3648 dentry->d_parent->d_inode->i_ino,
00e4e6b3
CM		 3649 objectid, BTRFS_I(dir)->block_group, mode,
3650 &index);
39279cc3
CM		 3651 err = PTR_ERR(inode);
3652 if (IS_ERR(inode))
3653 goto out_unlock;
3654
33268eaf
JB		 3655 err = btrfs_init_acl(inode, dir);
3656 if (err) {
3657 drop_inode = 1;
3658 goto out_unlock;
3659 }
3660
39279cc3 3661 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3662 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 3663 if (err)
3664 drop_inode = 1;
3665 else {
3666 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3667 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 3668 inode->i_fop = &btrfs_file_operations;
3669 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 3670 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3671 }
3672 dir->i_sb->s_dirt = 1;
3673 btrfs_update_inode_block_group(trans, inode);
3674 btrfs_update_inode_block_group(trans, dir);
3675out_unlock:
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
3687static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3688 struct dentry *dentry)
3689{
3690 struct btrfs_trans_handle *trans;
3691 struct btrfs_root *root = BTRFS_I(dir)->root;
3692 struct inode *inode = old_dentry->d_inode;
00e4e6b3 3693 u64 index;
1832a6d5 3694 unsigned long nr = 0;
39279cc3
CM		 3695 int err;
3696 int drop_inode = 0;
3697
3698 if (inode->i_nlink == 0)
3699 return -ENOENT;
3700
e02119d5 3701 btrfs_inc_nlink(inode);
1832a6d5
CM		 3702 err = btrfs_check_free_space(root, 1, 0);
3703 if (err)
3704 goto fail;
3de4586c 3705 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 3706 if (err)
3707 goto fail;
3708
39279cc3 3709 trans = btrfs_start_transaction(root, 1);
5f39d397 3710
39279cc3
CM		 3711 btrfs_set_trans_block_group(trans, dir);
3712 atomic_inc(&inode->i_count);
aec7477b 3713
00e4e6b3 3714 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
5f39d397 3715
39279cc3
CM		 3716 if (err)
3717 drop_inode = 1;
5f39d397 3718
39279cc3
CM		 3719 dir->i_sb->s_dirt = 1;
3720 btrfs_update_inode_block_group(trans, dir);
54aa1f4d 3721 err = btrfs_update_inode(trans, root, inode);
5f39d397 3722
54aa1f4d
CM		 3723 if (err)
3724 drop_inode = 1;
39279cc3 3725
d3c2fdcf 3726 nr = trans->blocks_used;
ab78c84d 3727 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3728fail:
39279cc3
CM		 3729 if (drop_inode) {
3730 inode_dec_link_count(inode);
3731 iput(inode);
3732 }
d3c2fdcf 3733 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3734 return err;
3735}
3736
39279cc3
CM		 3737static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3738{
b9d86667 3739 struct inode *inode = NULL;
39279cc3
CM		 3740 struct btrfs_trans_handle *trans;
3741 struct btrfs_root *root = BTRFS_I(dir)->root;
3742 int err = 0;
3743 int drop_on_err = 0;
b9d86667 3744 u64 objectid = 0;
00e4e6b3 3745 u64 index = 0;
d3c2fdcf 3746 unsigned long nr = 1;
39279cc3 3747
1832a6d5
CM		 3748 err = btrfs_check_free_space(root, 1, 0);
3749 if (err)
3750 goto out_unlock;
3751
39279cc3
CM		 3752 trans = btrfs_start_transaction(root, 1);
3753 btrfs_set_trans_block_group(trans, dir);
5f39d397 3754
39279cc3
CM		 3755 if (IS_ERR(trans)) {
3756 err = PTR_ERR(trans);
3757 goto out_unlock;
3758 }
3759
3760 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3761 if (err) {
3762 err = -ENOSPC;
3763 goto out_unlock;
3764 }
3765
aec7477b 3766 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3767 dentry->d_name.len,
3768 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 3769 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3770 &index);
39279cc3
CM		 3771 if (IS_ERR(inode)) {
3772 err = PTR_ERR(inode);
3773 goto out_fail;
3774 }
5f39d397 3775
39279cc3 3776 drop_on_err = 1;
33268eaf
JB		 3777
3778 err = btrfs_init_acl(inode, dir);
3779 if (err)
3780 goto out_fail;
3781
39279cc3
CM		 3782 inode->i_op = &btrfs_dir_inode_operations;
3783 inode->i_fop = &btrfs_dir_file_operations;
3784 btrfs_set_trans_block_group(trans, inode);
3785
dbe674a9 3786 btrfs_i_size_write(inode, 0);
39279cc3
CM		 3787 err = btrfs_update_inode(trans, root, inode);
3788 if (err)
3789 goto out_fail;
5f39d397 3790
e02119d5
CM		 3791 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3792 inode, dentry->d_name.name,
3793 dentry->d_name.len, 0, index);
39279cc3
CM		 3794 if (err)
3795 goto out_fail;
5f39d397 3796
39279cc3
CM		 3797 d_instantiate(dentry, inode);
3798 drop_on_err = 0;
3799 dir->i_sb->s_dirt = 1;
3800 btrfs_update_inode_block_group(trans, inode);
3801 btrfs_update_inode_block_group(trans, dir);
3802
3803out_fail:
d3c2fdcf 3804 nr = trans->blocks_used;
ab78c84d 3805 btrfs_end_transaction_throttle(trans, root);
5f39d397 3806
39279cc3 3807out_unlock:
39279cc3
CM		 3808 if (drop_on_err)
3809 iput(inode);
d3c2fdcf 3810 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3811 return err;
3812}
3813
d352ac68
CM		 3814/* helper for btfs_get_extent. Given an existing extent in the tree,
3815 * and an extent that you want to insert, deal with overlap and insert
3816 * the new extent into the tree.
3817 */
3b951516
CM		 3818static int merge_extent_mapping(struct extent_map_tree *em_tree,
3819 struct extent_map *existing,
e6dcd2dc
CM		 3820 struct extent_map *em,
3821 u64 map_start, u64 map_len)
3b951516
CM		 3822{
3823 u64 start_diff;
3b951516 3824
e6dcd2dc
CM		 3825 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3826 start_diff = map_start - em->start;
3827 em->start = map_start;
3828 em->len = map_len;
c8b97818
CM		 3829 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3830 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 3831 em->block_start += start_diff;
c8b97818
CM		 3832 em->block_len -= start_diff;
3833 }
e6dcd2dc 3834 return add_extent_mapping(em_tree, em);
3b951516
CM		 3835}
3836
c8b97818
CM		 3837static noinline int uncompress_inline(struct btrfs_path *path,
3838 struct inode *inode, struct page *page,
3839 size_t pg_offset, u64 extent_offset,
3840 struct btrfs_file_extent_item *item)
3841{
3842 int ret;
3843 struct extent_buffer *leaf = path->nodes[0];
3844 char *tmp;
3845 size_t max_size;
3846 unsigned long inline_size;
3847 unsigned long ptr;
3848
3849 WARN_ON(pg_offset != 0);
3850 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3851 inline_size = btrfs_file_extent_inline_item_len(leaf,
3852 btrfs_item_nr(leaf, path->slots[0]));
3853 tmp = kmalloc(inline_size, GFP_NOFS);
3854 ptr = btrfs_file_extent_inline_start(item);
3855
3856 read_extent_buffer(leaf, tmp, ptr, inline_size);
3857
5b050f04 3858 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
c8b97818
CM		 3859 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3860 inline_size, max_size);
3861 if (ret) {
3862 char *kaddr = kmap_atomic(page, KM_USER0);
3863 unsigned long copy_size = min_t(u64,
3864 PAGE_CACHE_SIZE - pg_offset,
3865 max_size - extent_offset);
3866 memset(kaddr + pg_offset, 0, copy_size);
3867 kunmap_atomic(kaddr, KM_USER0);
3868 }
3869 kfree(tmp);
3870 return 0;
3871}
3872
d352ac68
CM		 3873/*
3874 * a bit scary, this does extent mapping from logical file offset to the disk.
3875 * the ugly parts come from merging extents from the disk with the
3876 * in-ram representation. This gets more complex because of the data=ordered code,
3877 * where the in-ram extents might be locked pending data=ordered completion.
3878 *
3879 * This also copies inline extents directly into the page.
3880 */
a52d9a80 3881struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 3882 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 3883 int create)
3884{
3885 int ret;
3886 int err = 0;
db94535d 3887 u64 bytenr;
a52d9a80
CM		 3888 u64 extent_start = 0;
3889 u64 extent_end = 0;
3890 u64 objectid = inode->i_ino;
3891 u32 found_type;
f421950f 3892 struct btrfs_path *path = NULL;
a52d9a80
CM		 3893 struct btrfs_root *root = BTRFS_I(inode)->root;
3894 struct btrfs_file_extent_item *item;
5f39d397
CM		 3895 struct extent_buffer *leaf;
3896 struct btrfs_key found_key;
a52d9a80
CM		 3897 struct extent_map *em = NULL;
3898 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 3899 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 3900 struct btrfs_trans_handle *trans = NULL;
c8b97818 3901 int compressed;
a52d9a80 3902
a52d9a80 3903again:
d1310b2e
CM		 3904 spin_lock(&em_tree->lock);
3905 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 3906 if (em)
3907 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 3908 spin_unlock(&em_tree->lock);
3909
a52d9a80 3910 if (em) {
e1c4b745
CM		 3911 if (em->start > start || em->start + em->len <= start)
3912 free_extent_map(em);
3913 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 3914 free_extent_map(em);
3915 else
3916 goto out;
a52d9a80 3917 }
d1310b2e 3918 em = alloc_extent_map(GFP_NOFS);
a52d9a80 3919 if (!em) {
d1310b2e
CM		 3920 err = -ENOMEM;
3921 goto out;
a52d9a80 3922 }
e6dcd2dc 3923 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 3924 em->start = EXTENT_MAP_HOLE;
445a6944 3925 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 3926 em->len = (u64)-1;
c8b97818 3927 em->block_len = (u64)-1;
f421950f
CM		 3928
3929 if (!path) {
3930 path = btrfs_alloc_path();
3931 BUG_ON(!path);
3932 }
3933
179e29e4
CM		 3934 ret = btrfs_lookup_file_extent(trans, root, path,
3935 objectid, start, trans != NULL);
a52d9a80
CM		 3936 if (ret < 0) {
3937 err = ret;
3938 goto out;
3939 }
3940
3941 if (ret != 0) {
3942 if (path->slots[0] == 0)
3943 goto not_found;
3944 path->slots[0]--;
3945 }
3946
5f39d397
CM		 3947 leaf = path->nodes[0];
3948 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 3949 struct btrfs_file_extent_item);
a52d9a80 3950 /* are we inside the extent that was found? */
5f39d397
CM		 3951 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3952 found_type = btrfs_key_type(&found_key);
3953 if (found_key.objectid != objectid ||
a52d9a80
CM		 3954 found_type != BTRFS_EXTENT_DATA_KEY) {
3955 goto not_found;
3956 }
3957
5f39d397
CM		 3958 found_type = btrfs_file_extent_type(leaf, item);
3959 extent_start = found_key.offset;
c8b97818 3960 compressed = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 3961 if (found_type == BTRFS_FILE_EXTENT_REG ||
3962 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 3963 extent_end = extent_start +
db94535d 3964 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 3965 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3966 size_t size;
3967 size = btrfs_file_extent_inline_len(leaf, item);
3968 extent_end = (extent_start + size + root->sectorsize - 1) &
3969 ~((u64)root->sectorsize - 1);
3970 }
3971
3972 if (start >= extent_end) {
3973 path->slots[0]++;
3974 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3975 ret = btrfs_next_leaf(root, path);
3976 if (ret < 0) {
3977 err = ret;
3978 goto out;
a52d9a80 3979 }
9036c102
YZ		 3980 if (ret > 0)
3981 goto not_found;
3982 leaf = path->nodes[0];
a52d9a80 3983 }
9036c102
YZ		 3984 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3985 if (found_key.objectid != objectid ||
3986 found_key.type != BTRFS_EXTENT_DATA_KEY)
3987 goto not_found;
3988 if (start + len <= found_key.offset)
3989 goto not_found;
3990 em->start = start;
3991 em->len = found_key.offset - start;
3992 goto not_found_em;
3993 }
3994
d899e052
YZ		 3995 if (found_type == BTRFS_FILE_EXTENT_REG ||
3996 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 3997 em->start = extent_start;
3998 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 3999 em->orig_start = extent_start -
4000 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 4001 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4002 if (bytenr == 0) {
5f39d397 4003 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 4004 goto insert;
4005 }
c8b97818
CM		 4006 if (compressed) {
4007 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4008 em->block_start = bytenr;
4009 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4010 item);
4011 } else {
4012 bytenr += btrfs_file_extent_offset(leaf, item);
4013 em->block_start = bytenr;
4014 em->block_len = em->len;
d899e052
YZ		 4015 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4016 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 4017 }
a52d9a80
CM		 4018 goto insert;
4019 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4020 unsigned long ptr;
a52d9a80 4021 char *map;
3326d1b0
CM		 4022 size_t size;
4023 size_t extent_offset;
4024 size_t copy_size;
a52d9a80 4025
689f9346 4026 em->block_start = EXTENT_MAP_INLINE;
c8b97818 4027 if (!page || create) {
689f9346 4028 em->start = extent_start;
9036c102 4029 em->len = extent_end - extent_start;
689f9346
Y		 4030 goto out;
4031 }
5f39d397 4032
9036c102
YZ		 4033 size = btrfs_file_extent_inline_len(leaf, item);
4034 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 4035 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 4036 size - extent_offset);
3326d1b0 4037 em->start = extent_start + extent_offset;
70dec807
CM		 4038 em->len = (copy_size + root->sectorsize - 1) &
4039 ~((u64)root->sectorsize - 1);
ff5b7ee3 4040 em->orig_start = EXTENT_MAP_INLINE;
c8b97818
CM		 4041 if (compressed)
4042 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
689f9346 4043 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 4044 if (create == 0 && !PageUptodate(page)) {
c8b97818
CM		 4045 if (btrfs_file_extent_compression(leaf, item) ==
4046 BTRFS_COMPRESS_ZLIB) {
4047 ret = uncompress_inline(path, inode, page,
4048 pg_offset,
4049 extent_offset, item);
4050 BUG_ON(ret);
4051 } else {
4052 map = kmap(page);
4053 read_extent_buffer(leaf, map + pg_offset, ptr,
4054 copy_size);
4055 kunmap(page);
4056 }
179e29e4
CM		 4057 flush_dcache_page(page);
4058 } else if (create && PageUptodate(page)) {
4059 if (!trans) {
4060 kunmap(page);
4061 free_extent_map(em);
4062 em = NULL;
4063 btrfs_release_path(root, path);
f9295749 4064 trans = btrfs_join_transaction(root, 1);
179e29e4
CM		 4065 goto again;
4066 }
c8b97818 4067 map = kmap(page);
70dec807 4068 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 4069 copy_size);
c8b97818 4070 kunmap(page);
179e29e4 4071 btrfs_mark_buffer_dirty(leaf);
a52d9a80 4072 }
d1310b2e
CM		 4073 set_extent_uptodate(io_tree, em->start,
4074 extent_map_end(em) - 1, GFP_NOFS);
a52d9a80
CM		 4075 goto insert;
4076 } else {
4077 printk("unkknown found_type %d\n", found_type);
4078 WARN_ON(1);
4079 }
4080not_found:
4081 em->start = start;
d1310b2e 4082 em->len = len;
a52d9a80 4083not_found_em:
5f39d397 4084 em->block_start = EXTENT_MAP_HOLE;
9036c102 4085 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80
CM		 4086insert:
4087 btrfs_release_path(root, path);
d1310b2e
CM		 4088 if (em->start > start || extent_map_end(em) <= start) {
4089 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
a52d9a80
CM		 4090 err = -EIO;
4091 goto out;
4092 }
d1310b2e
CM		 4093
4094 err = 0;
4095 spin_lock(&em_tree->lock);
a52d9a80 4096 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 4097 /* it is possible that someone inserted the extent into the tree
4098 * while we had the lock dropped. It is also possible that
4099 * an overlapping map exists in the tree
4100 */
a52d9a80 4101 if (ret == -EEXIST) {
3b951516 4102 struct extent_map *existing;
e6dcd2dc
CM		 4103
4104 ret = 0;
4105
3b951516 4106 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 4107 if (existing && (existing->start > start ||
4108 existing->start + existing->len <= start)) {
4109 free_extent_map(existing);
4110 existing = NULL;
4111 }
3b951516
CM		 4112 if (!existing) {
4113 existing = lookup_extent_mapping(em_tree, em->start,
4114 em->len);
4115 if (existing) {
4116 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 4117 em, start,
4118 root->sectorsize);
3b951516
CM		 4119 free_extent_map(existing);
4120 if (err) {
4121 free_extent_map(em);
4122 em = NULL;
4123 }
4124 } else {
4125 err = -EIO;
4126 printk("failing to insert %Lu %Lu\n",
4127 start, len);
4128 free_extent_map(em);
4129 em = NULL;
4130 }
4131 } else {
4132 free_extent_map(em);
4133 em = existing;
e6dcd2dc 4134 err = 0;
a52d9a80 4135 }
a52d9a80 4136 }
d1310b2e 4137 spin_unlock(&em_tree->lock);
a52d9a80 4138out:
f421950f
CM		 4139 if (path)
4140 btrfs_free_path(path);
a52d9a80
CM		 4141 if (trans) {
4142 ret = btrfs_end_transaction(trans, root);
e6dcd2dc 4143 if (!err) {
a52d9a80 4144 err = ret;
e6dcd2dc 4145 }
a52d9a80 4146 }
a52d9a80
CM		 4147 if (err) {
4148 free_extent_map(em);
4149 WARN_ON(1);
4150 return ERR_PTR(err);
4151 }
4152 return em;
4153}
4154
16432985
CM		 4155static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4156 const struct iovec *iov, loff_t offset,
4157 unsigned long nr_segs)
4158{
e1c4b745 4159 return -EINVAL;
16432985
CM		 4160}
4161
d396c6f5 4162static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
39279cc3 4163{
d396c6f5 4164 return extent_bmap(mapping, iblock, btrfs_get_extent);
39279cc3
CM		 4165}
4166
a52d9a80 4167int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 4168{
d1310b2e
CM		 4169 struct extent_io_tree *tree;
4170 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4171 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 4172}
1832a6d5 4173
a52d9a80 4174static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 4175{
d1310b2e 4176 struct extent_io_tree *tree;
b888db2b
CM		 4177
4178
4179 if (current->flags & PF_MEMALLOC) {
4180 redirty_page_for_writepage(wbc, page);
4181 unlock_page(page);
4182 return 0;
4183 }
d1310b2e 4184 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4185 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 4186}
4187
f421950f
CM		 4188int btrfs_writepages(struct address_space *mapping,
4189 struct writeback_control *wbc)
b293f02e 4190{
d1310b2e 4191 struct extent_io_tree *tree;
771ed689 4192
d1310b2e 4193 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 4194 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4195}
4196
3ab2fb5a
CM		 4197static int
4198btrfs_readpages(struct file *file, struct address_space *mapping,
4199 struct list_head *pages, unsigned nr_pages)
4200{
d1310b2e
CM		 4201 struct extent_io_tree *tree;
4202 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 4203 return extent_readpages(tree, mapping, pages, nr_pages,
4204 btrfs_get_extent);
4205}
e6dcd2dc 4206static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 4207{
d1310b2e
CM		 4208 struct extent_io_tree *tree;
4209 struct extent_map_tree *map;
a52d9a80 4210 int ret;
8c2383c3 4211
d1310b2e
CM		 4212 tree = &BTRFS_I(page->mapping->host)->io_tree;
4213 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 4214 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 4215 if (ret == 1) {
4216 ClearPagePrivate(page);
4217 set_page_private(page, 0);
4218 page_cache_release(page);
39279cc3 4219 }
a52d9a80 4220 return ret;
39279cc3
CM		 4221}
4222
e6dcd2dc
CM		 4223static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4224{
98509cfc
CM		 4225 if (PageWriteback(page) || PageDirty(page))
4226 return 0;
e6dcd2dc
CM		 4227 return __btrfs_releasepage(page, gfp_flags);
4228}
4229
a52d9a80 4230static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 4231{
d1310b2e 4232 struct extent_io_tree *tree;
e6dcd2dc
CM		 4233 struct btrfs_ordered_extent *ordered;
4234 u64 page_start = page_offset(page);
4235 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 4236
e6dcd2dc 4237 wait_on_page_writeback(page);
d1310b2e 4238 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 4239 if (offset) {
4240 btrfs_releasepage(page, GFP_NOFS);
4241 return;
4242 }
4243
4244 lock_extent(tree, page_start, page_end, GFP_NOFS);
4245 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4246 page_offset(page));
4247 if (ordered) {
eb84ae03
CM		 4248 /*
4249 * IO on this page will never be started, so we need
4250 * to account for any ordered extents now
4251 */
e6dcd2dc
CM		 4252 clear_extent_bit(tree, page_start, page_end,
4253 EXTENT_DIRTY | EXTENT_DELALLOC |
4254 EXTENT_LOCKED, 1, 0, GFP_NOFS);
211f90e6
CM		 4255 btrfs_finish_ordered_io(page->mapping->host,
4256 page_start, page_end);
e6dcd2dc
CM		 4257 btrfs_put_ordered_extent(ordered);
4258 lock_extent(tree, page_start, page_end, GFP_NOFS);
4259 }
4260 clear_extent_bit(tree, page_start, page_end,
4261 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4262 EXTENT_ORDERED,
4263 1, 1, GFP_NOFS);
4264 __btrfs_releasepage(page, GFP_NOFS);
4265
4a096752 4266 ClearPageChecked(page);
9ad6b7bc 4267 if (PagePrivate(page)) {
9ad6b7bc
CM		 4268 ClearPagePrivate(page);
4269 set_page_private(page, 0);
4270 page_cache_release(page);
4271 }
39279cc3
CM		 4272}
4273
9ebefb18
CM		 4274/*
4275 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4276 * called from a page fault handler when a page is first dirtied. Hence we must
4277 * be careful to check for EOF conditions here. We set the page up correctly
4278 * for a written page which means we get ENOSPC checking when writing into
4279 * holes and correct delalloc and unwritten extent mapping on filesystems that
4280 * support these features.
4281 *
4282 * We are not allowed to take the i_mutex here so we have to play games to
4283 * protect against truncate races as the page could now be beyond EOF. Because
4284 * vmtruncate() writes the inode size before removing pages, once we have the
4285 * page lock we can determine safely if the page is beyond EOF. If it is not
4286 * beyond EOF, then the page is guaranteed safe against truncation until we
4287 * unlock the page.
4288 */
4289int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4290{
6da6abae 4291 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 4292 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4293 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4294 struct btrfs_ordered_extent *ordered;
4295 char *kaddr;
4296 unsigned long zero_start;
9ebefb18 4297 loff_t size;
1832a6d5 4298 int ret;
a52d9a80 4299 u64 page_start;
e6dcd2dc 4300 u64 page_end;
9ebefb18 4301
1832a6d5 4302 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
1832a6d5
CM		 4303 if (ret)
4304 goto out;
4305
4306 ret = -EINVAL;
e6dcd2dc 4307again:
9ebefb18 4308 lock_page(page);
9ebefb18 4309 size = i_size_read(inode);
e6dcd2dc
CM		 4310 page_start = page_offset(page);
4311 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 4312
9ebefb18 4313 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 4314 (page_start >= size)) {
9ebefb18
CM		 4315 /* page got truncated out from underneath us */
4316 goto out_unlock;
4317 }
e6dcd2dc
CM		 4318 wait_on_page_writeback(page);
4319
4320 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4321 set_page_extent_mapped(page);
4322
eb84ae03
CM		 4323 /*
4324 * we can't set the delalloc bits if there are pending ordered
4325 * extents. Drop our locks and wait for them to finish
4326 */
e6dcd2dc
CM		 4327 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4328 if (ordered) {
4329 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4330 unlock_page(page);
eb84ae03 4331 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4332 btrfs_put_ordered_extent(ordered);
4333 goto again;
4334 }
4335
ea8c2819 4336 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc 4337 ret = 0;
9ebefb18
CM		 4338
4339 /* page is wholly or partially inside EOF */
a52d9a80 4340 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 4341 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 4342 else
e6dcd2dc 4343 zero_start = PAGE_CACHE_SIZE;
9ebefb18 4344
e6dcd2dc
CM		 4345 if (zero_start != PAGE_CACHE_SIZE) {
4346 kaddr = kmap(page);
4347 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4348 flush_dcache_page(page);
4349 kunmap(page);
4350 }
247e743c 4351 ClearPageChecked(page);
e6dcd2dc
CM		 4352 set_page_dirty(page);
4353 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
9ebefb18
CM		 4354
4355out_unlock:
4356 unlock_page(page);
1832a6d5 4357out:
9ebefb18
CM		 4358 return ret;
4359}
4360
39279cc3
CM		 4361static void btrfs_truncate(struct inode *inode)
4362{
4363 struct btrfs_root *root = BTRFS_I(inode)->root;
4364 int ret;
4365 struct btrfs_trans_handle *trans;
d3c2fdcf 4366 unsigned long nr;
dbe674a9 4367 u64 mask = root->sectorsize - 1;
39279cc3
CM		 4368
4369 if (!S_ISREG(inode->i_mode))
4370 return;
4371 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4372 return;
4373
4374 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4a096752 4375 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 4376
39279cc3
CM		 4377 trans = btrfs_start_transaction(root, 1);
4378 btrfs_set_trans_block_group(trans, inode);
dbe674a9 4379 btrfs_i_size_write(inode, inode->i_size);
39279cc3 4380
7b128766
JB		 4381 ret = btrfs_orphan_add(trans, inode);
4382 if (ret)
4383 goto out;
39279cc3 4384 /* FIXME, add redo link to tree so we don't leak on crash */
e02119d5 4385 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
85e21bac 4386 BTRFS_EXTENT_DATA_KEY);
39279cc3 4387 btrfs_update_inode(trans, root, inode);
5f39d397 4388
7b128766
JB		 4389 ret = btrfs_orphan_del(trans, inode);
4390 BUG_ON(ret);
4391
4392out:
4393 nr = trans->blocks_used;
89ce8a63 4394 ret = btrfs_end_transaction_throttle(trans, root);
39279cc3 4395 BUG_ON(ret);
d3c2fdcf 4396 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4397}
4398
d352ac68
CM		 4399/*
4400 * create a new subvolume directory/inode (helper for the ioctl).
4401 */
d2fb3437
YZ		 4402int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4403 struct btrfs_root *new_root, struct dentry *dentry,
4404 u64 new_dirid, u64 alloc_hint)
39279cc3 4405{
39279cc3 4406 struct inode *inode;
cb8e7090 4407 int error;
00e4e6b3 4408 u64 index = 0;
39279cc3 4409
aec7477b 4410 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
d2fb3437 4411 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
54aa1f4d 4412 if (IS_ERR(inode))
f46b5a66 4413 return PTR_ERR(inode);
39279cc3
CM		 4414 inode->i_op = &btrfs_dir_inode_operations;
4415 inode->i_fop = &btrfs_dir_file_operations;
4416
39279cc3 4417 inode->i_nlink = 1;
dbe674a9 4418 btrfs_i_size_write(inode, 0);
3b96362c 4419
cb8e7090
CH		 4420 error = btrfs_update_inode(trans, new_root, inode);
4421 if (error)
4422 return error;
4423
4424 d_instantiate(dentry, inode);
4425 return 0;
39279cc3
CM		 4426}
4427
d352ac68
CM		 4428/* helper function for file defrag and space balancing. This
4429 * forces readahead on a given range of bytes in an inode
4430 */
edbd8d4e 4431unsigned long btrfs_force_ra(struct address_space *mapping,
86479a04
CM		 4432 struct file_ra_state *ra, struct file *file,
4433 pgoff_t offset, pgoff_t last_index)
4434{
8e7bf94f 4435 pgoff_t req_size = last_index - offset + 1;
86479a04 4436
86479a04
CM		 4437 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4438 return offset + req_size;
86479a04
CM		 4439}
4440
39279cc3
CM		 4441struct inode *btrfs_alloc_inode(struct super_block *sb)
4442{
4443 struct btrfs_inode *ei;
4444
4445 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4446 if (!ei)
4447 return NULL;
15ee9bc7 4448 ei->last_trans = 0;
e02119d5 4449 ei->logged_trans = 0;
e6dcd2dc 4450 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
33268eaf
JB		 4451 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4452 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
7b128766 4453 INIT_LIST_HEAD(&ei->i_orphan);
39279cc3
CM		 4454 return &ei->vfs_inode;
4455}
4456
4457void btrfs_destroy_inode(struct inode *inode)
4458{
e6dcd2dc 4459 struct btrfs_ordered_extent *ordered;
39279cc3
CM		 4460 WARN_ON(!list_empty(&inode->i_dentry));
4461 WARN_ON(inode->i_data.nrpages);
4462
33268eaf
JB		 4463 if (BTRFS_I(inode)->i_acl &&
4464 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4465 posix_acl_release(BTRFS_I(inode)->i_acl);
4466 if (BTRFS_I(inode)->i_default_acl &&
4467 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4468 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4469
bcc63abb 4470 spin_lock(&BTRFS_I(inode)->root->list_lock);
7b128766
JB		 4471 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4472 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4473 " list\n", inode->i_ino);
4474 dump_stack();
4475 }
bcc63abb 4476 spin_unlock(&BTRFS_I(inode)->root->list_lock);
7b128766 4477
e6dcd2dc
CM		 4478 while(1) {
4479 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4480 if (!ordered)
4481 break;
4482 else {
4483 printk("found ordered extent %Lu %Lu\n",
4484 ordered->file_offset, ordered->len);
4485 btrfs_remove_ordered_extent(inode, ordered);
4486 btrfs_put_ordered_extent(ordered);
4487 btrfs_put_ordered_extent(ordered);
4488 }
4489 }
5b21f2ed 4490 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
39279cc3
CM		 4491 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4492}
4493
0ee0fda0 4494static void init_once(void *foo)
39279cc3
CM		 4495{
4496 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4497
4498 inode_init_once(&ei->vfs_inode);
4499}
4500
4501void btrfs_destroy_cachep(void)
4502{
4503 if (btrfs_inode_cachep)
4504 kmem_cache_destroy(btrfs_inode_cachep);
4505 if (btrfs_trans_handle_cachep)
4506 kmem_cache_destroy(btrfs_trans_handle_cachep);
4507 if (btrfs_transaction_cachep)
4508 kmem_cache_destroy(btrfs_transaction_cachep);
4509 if (btrfs_bit_radix_cachep)
4510 kmem_cache_destroy(btrfs_bit_radix_cachep);
4511 if (btrfs_path_cachep)
4512 kmem_cache_destroy(btrfs_path_cachep);
4513}
4514
86479a04 4515struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
92fee66d 4516 unsigned long extra_flags,
2b1f55b0 4517 void (*ctor)(void *))
92fee66d
CM		 4518{
4519 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2b1f55b0 4520 SLAB_MEM_SPREAD | extra_flags), ctor);
92fee66d
CM		 4521}
4522
39279cc3
CM		 4523int btrfs_init_cachep(void)
4524{
86479a04 4525 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
92fee66d
CM		 4526 sizeof(struct btrfs_inode),
4527 0, init_once);
39279cc3
CM		 4528 if (!btrfs_inode_cachep)
4529 goto fail;
86479a04
CM		 4530 btrfs_trans_handle_cachep =
4531 btrfs_cache_create("btrfs_trans_handle_cache",
4532 sizeof(struct btrfs_trans_handle),
4533 0, NULL);
39279cc3
CM		 4534 if (!btrfs_trans_handle_cachep)
4535 goto fail;
86479a04 4536 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
39279cc3 4537 sizeof(struct btrfs_transaction),
92fee66d 4538 0, NULL);
39279cc3
CM		 4539 if (!btrfs_transaction_cachep)
4540 goto fail;
86479a04 4541 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
23223584 4542 sizeof(struct btrfs_path),
92fee66d 4543 0, NULL);
39279cc3
CM		 4544 if (!btrfs_path_cachep)
4545 goto fail;
86479a04 4546 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
92fee66d 4547 SLAB_DESTROY_BY_RCU, NULL);
39279cc3
CM		 4548 if (!btrfs_bit_radix_cachep)
4549 goto fail;
4550 return 0;
4551fail:
4552 btrfs_destroy_cachep();
4553 return -ENOMEM;
4554}
4555
4556static int btrfs_getattr(struct vfsmount *mnt,
4557 struct dentry *dentry, struct kstat *stat)
4558{
4559 struct inode *inode = dentry->d_inode;
4560 generic_fillattr(inode, stat);
3394e160 4561 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
d6667462 4562 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 4563 stat->blocks = (inode_get_bytes(inode) +
4564 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 4565 return 0;
4566}
4567
4568static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4569 struct inode * new_dir,struct dentry *new_dentry)
4570{
4571 struct btrfs_trans_handle *trans;
4572 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4573 struct inode *new_inode = new_dentry->d_inode;
4574 struct inode *old_inode = old_dentry->d_inode;
4575 struct timespec ctime = CURRENT_TIME;
00e4e6b3 4576 u64 index = 0;
39279cc3
CM		 4577 int ret;
4578
3394e160
CM		 4579 /* we're not allowed to rename between subvolumes */
4580 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4581 BTRFS_I(new_dir)->root->root_key.objectid)
4582 return -EXDEV;
4583
39279cc3
CM		 4584 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4585 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4586 return -ENOTEMPTY;
4587 }
5f39d397 4588
0660b5af
CM		 4589 /* to rename a snapshot or subvolume, we need to juggle the
4590 * backrefs. This isn't coded yet
4591 */
4592 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4593 return -EXDEV;
4594
1832a6d5
CM		 4595 ret = btrfs_check_free_space(root, 1, 0);
4596 if (ret)
4597 goto out_unlock;
4598
39279cc3 4599 trans = btrfs_start_transaction(root, 1);
5f39d397 4600
39279cc3 4601 btrfs_set_trans_block_group(trans, new_dir);
39279cc3 4602
e02119d5 4603 btrfs_inc_nlink(old_dentry->d_inode);
39279cc3
CM		 4604 old_dir->i_ctime = old_dir->i_mtime = ctime;
4605 new_dir->i_ctime = new_dir->i_mtime = ctime;
4606 old_inode->i_ctime = ctime;
5f39d397 4607
e02119d5
CM		 4608 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4609 old_dentry->d_name.name,
4610 old_dentry->d_name.len);
39279cc3
CM		 4611 if (ret)
4612 goto out_fail;
4613
4614 if (new_inode) {
4615 new_inode->i_ctime = CURRENT_TIME;
e02119d5
CM		 4616 ret = btrfs_unlink_inode(trans, root, new_dir,
4617 new_dentry->d_inode,
4618 new_dentry->d_name.name,
4619 new_dentry->d_name.len);
39279cc3
CM		 4620 if (ret)
4621 goto out_fail;
7b128766 4622 if (new_inode->i_nlink == 0) {
e02119d5 4623 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7b128766
JB		 4624 if (ret)
4625 goto out_fail;
4626 }
e02119d5 4627
39279cc3 4628 }
3de4586c 4629 ret = btrfs_set_inode_index(new_dir, &index);
aec7477b
JB		 4630 if (ret)
4631 goto out_fail;
4632
e02119d5
CM		 4633 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4634 old_inode, new_dentry->d_name.name,
4635 new_dentry->d_name.len, 1, index);
39279cc3
CM		 4636 if (ret)
4637 goto out_fail;
4638
4639out_fail:
ab78c84d 4640 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4641out_unlock:
39279cc3
CM		 4642 return ret;
4643}
4644
d352ac68
CM		 4645/*
4646 * some fairly slow code that needs optimization. This walks the list
4647 * of all the inodes with pending delalloc and forces them to disk.
4648 */
ea8c2819
CM		 4649int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4650{
4651 struct list_head *head = &root->fs_info->delalloc_inodes;
4652 struct btrfs_inode *binode;
5b21f2ed 4653 struct inode *inode;
ea8c2819
CM		 4654 unsigned long flags;
4655
c146afad
YZ		 4656 if (root->fs_info->sb->s_flags & MS_RDONLY)
4657 return -EROFS;
4658
ea8c2819
CM		 4659 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4660 while(!list_empty(head)) {
4661 binode = list_entry(head->next, struct btrfs_inode,
4662 delalloc_inodes);
5b21f2ed
ZY		 4663 inode = igrab(&binode->vfs_inode);
4664 if (!inode)
4665 list_del_init(&binode->delalloc_inodes);
ea8c2819 4666 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
5b21f2ed 4667 if (inode) {
8c8bee1d 4668 filemap_flush(inode->i_mapping);
5b21f2ed
ZY		 4669 iput(inode);
4670 }
4671 cond_resched();
ea8c2819
CM		 4672 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4673 }
4674 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
8c8bee1d
CM		 4675
4676 /* the filemap_flush will queue IO into the worker threads, but
4677 * we have to make sure the IO is actually started and that
4678 * ordered extents get created before we return
4679 */
4680 atomic_inc(&root->fs_info->async_submit_draining);
771ed689
CM		 4681 while(atomic_read(&root->fs_info->nr_async_submits) ||
4682 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 4683 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 4684 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4685 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 4686 }
4687 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 4688 return 0;
4689}
4690
39279cc3
CM		 4691static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4692 const char *symname)
4693{
4694 struct btrfs_trans_handle *trans;
4695 struct btrfs_root *root = BTRFS_I(dir)->root;
4696 struct btrfs_path *path;
4697 struct btrfs_key key;
1832a6d5 4698 struct inode *inode = NULL;
39279cc3
CM		 4699 int err;
4700 int drop_inode = 0;
4701 u64 objectid;
00e4e6b3 4702 u64 index = 0 ;
39279cc3
CM		 4703 int name_len;
4704 int datasize;
5f39d397 4705 unsigned long ptr;
39279cc3 4706 struct btrfs_file_extent_item *ei;
5f39d397 4707 struct extent_buffer *leaf;
1832a6d5 4708 unsigned long nr = 0;
39279cc3
CM		 4709
4710 name_len = strlen(symname) + 1;
4711 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4712 return -ENAMETOOLONG;
1832a6d5 4713
1832a6d5
CM		 4714 err = btrfs_check_free_space(root, 1, 0);
4715 if (err)
4716 goto out_fail;
4717
39279cc3
CM		 4718 trans = btrfs_start_transaction(root, 1);
4719 btrfs_set_trans_block_group(trans, dir);
4720
4721 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4722 if (err) {
4723 err = -ENOSPC;
4724 goto out_unlock;
4725 }
4726
aec7477b 4727 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 4728 dentry->d_name.len,
4729 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 4730 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4731 &index);
39279cc3
CM		 4732 err = PTR_ERR(inode);
4733 if (IS_ERR(inode))
4734 goto out_unlock;
4735
33268eaf
JB		 4736 err = btrfs_init_acl(inode, dir);
4737 if (err) {
4738 drop_inode = 1;
4739 goto out_unlock;
4740 }
4741
39279cc3 4742 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 4743 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 4744 if (err)
4745 drop_inode = 1;
4746 else {
4747 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4748 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4749 inode->i_fop = &btrfs_file_operations;
4750 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4751 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 4752 }
4753 dir->i_sb->s_dirt = 1;
4754 btrfs_update_inode_block_group(trans, inode);
4755 btrfs_update_inode_block_group(trans, dir);
4756 if (drop_inode)
4757 goto out_unlock;
4758
4759 path = btrfs_alloc_path();
4760 BUG_ON(!path);
4761 key.objectid = inode->i_ino;
4762 key.offset = 0;
39279cc3
CM		 4763 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4764 datasize = btrfs_file_extent_calc_inline_size(name_len);
4765 err = btrfs_insert_empty_item(trans, root, path, &key,
4766 datasize);
54aa1f4d
CM		 4767 if (err) {
4768 drop_inode = 1;
4769 goto out_unlock;
4770 }
5f39d397
CM		 4771 leaf = path->nodes[0];
4772 ei = btrfs_item_ptr(leaf, path->slots[0],
4773 struct btrfs_file_extent_item);
4774 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4775 btrfs_set_file_extent_type(leaf, ei,
39279cc3 4776 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 4777 btrfs_set_file_extent_encryption(leaf, ei, 0);
4778 btrfs_set_file_extent_compression(leaf, ei, 0);
4779 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4780 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4781
39279cc3 4782 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 4783 write_extent_buffer(leaf, symname, ptr, name_len);
4784 btrfs_mark_buffer_dirty(leaf);
39279cc3 4785 btrfs_free_path(path);
5f39d397 4786
39279cc3
CM		 4787 inode->i_op = &btrfs_symlink_inode_operations;
4788 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 4789 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 4790 inode_set_bytes(inode, name_len);
dbe674a9 4791 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 4792 err = btrfs_update_inode(trans, root, inode);
4793 if (err)
4794 drop_inode = 1;
39279cc3
CM		 4795
4796out_unlock:
d3c2fdcf 4797 nr = trans->blocks_used;
ab78c84d 4798 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4799out_fail:
39279cc3
CM		 4800 if (drop_inode) {
4801 inode_dec_link_count(inode);
4802 iput(inode);
4803 }
d3c2fdcf 4804 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4805 return err;
4806}
16432985 4807
d899e052
YZ		 4808static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4809 u64 alloc_hint, int mode)
4810{
4811 struct btrfs_trans_handle *trans;
4812 struct btrfs_root *root = BTRFS_I(inode)->root;
4813 struct btrfs_key ins;
4814 u64 alloc_size;
4815 u64 cur_offset = start;
4816 u64 num_bytes = end - start;
4817 int ret = 0;
4818
4819 trans = btrfs_join_transaction(root, 1);
4820 BUG_ON(!trans);
4821 btrfs_set_trans_block_group(trans, inode);
4822
4823 while (num_bytes > 0) {
4824 alloc_size = min(num_bytes, root->fs_info->max_extent);
4825 ret = btrfs_reserve_extent(trans, root, alloc_size,
4826 root->sectorsize, 0, alloc_hint,
4827 (u64)-1, &ins, 1);
4828 if (ret) {
4829 WARN_ON(1);
4830 goto out;
4831 }
4832 ret = insert_reserved_file_extent(trans, inode,
4833 cur_offset, ins.objectid,
4834 ins.offset, ins.offset,
4835 ins.offset, 0, 0, 0,
4836 BTRFS_FILE_EXTENT_PREALLOC);
4837 BUG_ON(ret);
4838 num_bytes -= ins.offset;
4839 cur_offset += ins.offset;
4840 alloc_hint = ins.objectid + ins.offset;
4841 }
4842out:
4843 if (cur_offset > start) {
4844 inode->i_ctime = CURRENT_TIME;
4845 btrfs_set_flag(inode, PREALLOC);
4846 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4847 cur_offset > i_size_read(inode))
4848 btrfs_i_size_write(inode, cur_offset);
4849 ret = btrfs_update_inode(trans, root, inode);
4850 BUG_ON(ret);
4851 }
4852
4853 btrfs_end_transaction(trans, root);
4854 return ret;
4855}
4856
4857static long btrfs_fallocate(struct inode *inode, int mode,
4858 loff_t offset, loff_t len)
4859{
4860 u64 cur_offset;
4861 u64 last_byte;
4862 u64 alloc_start;
4863 u64 alloc_end;
4864 u64 alloc_hint = 0;
4865 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4866 struct extent_map *em;
4867 int ret;
4868
4869 alloc_start = offset & ~mask;
4870 alloc_end = (offset + len + mask) & ~mask;
4871
4872 mutex_lock(&inode->i_mutex);
4873 if (alloc_start > inode->i_size) {
4874 ret = btrfs_cont_expand(inode, alloc_start);
4875 if (ret)
4876 goto out;
4877 }
4878
4879 while (1) {
4880 struct btrfs_ordered_extent *ordered;
4881 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4882 alloc_end - 1, GFP_NOFS);
4883 ordered = btrfs_lookup_first_ordered_extent(inode,
4884 alloc_end - 1);
4885 if (ordered &&
4886 ordered->file_offset + ordered->len > alloc_start &&
4887 ordered->file_offset < alloc_end) {
4888 btrfs_put_ordered_extent(ordered);
4889 unlock_extent(&BTRFS_I(inode)->io_tree,
4890 alloc_start, alloc_end - 1, GFP_NOFS);
4891 btrfs_wait_ordered_range(inode, alloc_start,
4892 alloc_end - alloc_start);
4893 } else {
4894 if (ordered)
4895 btrfs_put_ordered_extent(ordered);
4896 break;
4897 }
4898 }
4899
4900 cur_offset = alloc_start;
4901 while (1) {
4902 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4903 alloc_end - cur_offset, 0);
4904 BUG_ON(IS_ERR(em) || !em);
4905 last_byte = min(extent_map_end(em), alloc_end);
4906 last_byte = (last_byte + mask) & ~mask;
4907 if (em->block_start == EXTENT_MAP_HOLE) {
4908 ret = prealloc_file_range(inode, cur_offset,
4909 last_byte, alloc_hint, mode);
4910 if (ret < 0) {
4911 free_extent_map(em);
4912 break;
4913 }
4914 }
4915 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4916 alloc_hint = em->block_start;
4917 free_extent_map(em);
4918
4919 cur_offset = last_byte;
4920 if (cur_offset >= alloc_end) {
4921 ret = 0;
4922 break;
4923 }
4924 }
4925 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4926 GFP_NOFS);
4927out:
4928 mutex_unlock(&inode->i_mutex);
4929 return ret;
4930}
4931
e6dcd2dc
CM		 4932static int btrfs_set_page_dirty(struct page *page)
4933{
e6dcd2dc
CM		 4934 return __set_page_dirty_nobuffers(page);
4935}
4936
0ee0fda0 4937static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd
Y		 4938{
4939 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4940 return -EACCES;
33268eaf 4941 return generic_permission(inode, mask, btrfs_check_acl);
fdebe2bd 4942}
39279cc3
CM		 4943
4944static struct inode_operations btrfs_dir_inode_operations = {
3394e160 4945 .getattr = btrfs_getattr,
39279cc3
CM		 4946 .lookup = btrfs_lookup,
4947 .create = btrfs_create,
4948 .unlink = btrfs_unlink,
4949 .link = btrfs_link,
4950 .mkdir = btrfs_mkdir,
4951 .rmdir = btrfs_rmdir,
4952 .rename = btrfs_rename,
4953 .symlink = btrfs_symlink,
4954 .setattr = btrfs_setattr,
618e21d5 4955 .mknod = btrfs_mknod,
95819c05
CH		 4956 .setxattr = btrfs_setxattr,
4957 .getxattr = btrfs_getxattr,
5103e947 4958 .listxattr = btrfs_listxattr,
95819c05 4959 .removexattr = btrfs_removexattr,
fdebe2bd 4960 .permission = btrfs_permission,
39279cc3 4961};
39279cc3
CM		 4962static struct inode_operations btrfs_dir_ro_inode_operations = {
4963 .lookup = btrfs_lookup,
fdebe2bd 4964 .permission = btrfs_permission,
39279cc3 4965};
39279cc3
CM		 4966static struct file_operations btrfs_dir_file_operations = {
4967 .llseek = generic_file_llseek,
4968 .read = generic_read_dir,
cbdf5a24 4969 .readdir = btrfs_real_readdir,
34287aa3 4970 .unlocked_ioctl = btrfs_ioctl,
39279cc3 4971#ifdef CONFIG_COMPAT
34287aa3 4972 .compat_ioctl = btrfs_ioctl,
39279cc3 4973#endif
6bf13c0c 4974 .release = btrfs_release_file,
e02119d5 4975 .fsync = btrfs_sync_file,
39279cc3
CM		 4976};
4977
d1310b2e 4978static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 4979 .fill_delalloc = run_delalloc_range,
065631f6 4980 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 4981 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 4982 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 4983 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 4984 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 4985 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 4986 .set_bit_hook = btrfs_set_bit_hook,
4987 .clear_bit_hook = btrfs_clear_bit_hook,
07157aac
CM		 4988};
4989
39279cc3
CM		 4990static struct address_space_operations btrfs_aops = {
4991 .readpage = btrfs_readpage,
4992 .writepage = btrfs_writepage,
b293f02e 4993 .writepages = btrfs_writepages,
3ab2fb5a 4994 .readpages = btrfs_readpages,
39279cc3 4995 .sync_page = block_sync_page,
39279cc3 4996 .bmap = btrfs_bmap,
16432985 4997 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 4998 .invalidatepage = btrfs_invalidatepage,
4999 .releasepage = btrfs_releasepage,
e6dcd2dc 5000 .set_page_dirty = btrfs_set_page_dirty,
39279cc3
CM		 5001};
5002
5003static struct address_space_operations btrfs_symlink_aops = {
5004 .readpage = btrfs_readpage,
5005 .writepage = btrfs_writepage,
2bf5a725
CM		 5006 .invalidatepage = btrfs_invalidatepage,
5007 .releasepage = btrfs_releasepage,
39279cc3
CM		 5008};
5009
5010static struct inode_operations btrfs_file_inode_operations = {
5011 .truncate = btrfs_truncate,
5012 .getattr = btrfs_getattr,
5013 .setattr = btrfs_setattr,
95819c05
CH		 5014 .setxattr = btrfs_setxattr,
5015 .getxattr = btrfs_getxattr,
5103e947 5016 .listxattr = btrfs_listxattr,
95819c05 5017 .removexattr = btrfs_removexattr,
fdebe2bd 5018 .permission = btrfs_permission,
d899e052 5019 .fallocate = btrfs_fallocate,
39279cc3 5020};
618e21d5
JB		 5021static struct inode_operations btrfs_special_inode_operations = {
5022 .getattr = btrfs_getattr,
5023 .setattr = btrfs_setattr,
fdebe2bd 5024 .permission = btrfs_permission,
95819c05
CH		 5025 .setxattr = btrfs_setxattr,
5026 .getxattr = btrfs_getxattr,
33268eaf 5027 .listxattr = btrfs_listxattr,
95819c05 5028 .removexattr = btrfs_removexattr,
618e21d5 5029};
39279cc3
CM		 5030static struct inode_operations btrfs_symlink_inode_operations = {
5031 .readlink = generic_readlink,
5032 .follow_link = page_follow_link_light,
5033 .put_link = page_put_link,
fdebe2bd 5034 .permission = btrfs_permission,
39279cc3 5035};
Linux 6.x block layer and io_uring tree(s)