blob: 88f9df7bfdaee77dcc65b223b8afbc0b61ee35d2 [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
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
Chris Mason8f18cf12008-04-25 16:53:30 -040019#include <linux/kernel.h>
Chris Mason065631f2008-02-20 12:07:25 -050020#include <linux/bio.h>
Chris Mason39279cc2007-06-12 06:35:45 -040021#include <linux/buffer_head.h>
Sage Weilf2eb0a22008-05-02 14:43:14 -040022#include <linux/file.h>
Chris Mason39279cc2007-06-12 06:35:45 -040023#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>
Chris Mason9ebefb182007-06-15 13:50:00 -040036#include <linux/bit_spinlock.h>
Josef Bacik5103e942007-11-16 11:45:54 -050037#include <linux/xattr.h>
Josef Bacik33268ea2008-07-24 12:16:36 -040038#include <linux/posix_acl.h>
Yan Zhengd899e052008-10-30 14:25:28 -040039#include <linux/falloc.h>
Chris Mason4b4e25f2008-11-20 10:22:27 -050040#include "compat.h"
Chris Mason39279cc2007-06-12 06:35:45 -040041#include "ctree.h"
42#include "disk-io.h"
43#include "transaction.h"
44#include "btrfs_inode.h"
45#include "ioctl.h"
46#include "print-tree.h"
Chris Mason0b86a832008-03-24 15:01:56 -040047#include "volumes.h"
Chris Masone6dcd2d2008-07-17 12:53:50 -040048#include "ordered-data.h"
Christoph Hellwig95819c02008-08-28 06:21:17 -040049#include "xattr.h"
Chris Masone02119d2008-09-05 16:13:11 -040050#include "tree-log.h"
Chris Masonc8b97812008-10-29 14:49:59 -040051#include "compression.h"
Chris Masonb4ce94d2009-02-04 09:25:08 -050052#include "locking.h"
Chris Mason39279cc2007-06-12 06:35:45 -040053
54struct btrfs_iget_args {
55 u64 ino;
56 struct btrfs_root *root;
57};
58
59static struct inode_operations btrfs_dir_inode_operations;
60static struct inode_operations btrfs_symlink_inode_operations;
61static struct inode_operations btrfs_dir_ro_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -040062static struct inode_operations btrfs_special_inode_operations;
Chris Mason39279cc2007-06-12 06:35:45 -040063static struct inode_operations btrfs_file_inode_operations;
64static struct address_space_operations btrfs_aops;
65static struct address_space_operations btrfs_symlink_aops;
66static struct file_operations btrfs_dir_file_operations;
Chris Masond1310b22008-01-24 16:13:08 -050067static struct extent_io_ops btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -040068
69static struct kmem_cache *btrfs_inode_cachep;
70struct kmem_cache *btrfs_trans_handle_cachep;
71struct kmem_cache *btrfs_transaction_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040072struct kmem_cache *btrfs_path_cachep;
73
74#define S_SHIFT 12
75static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
77 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
78 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
79 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
80 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
81 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
82 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
83};
84
Josef Bacik7b128762008-07-24 12:17:14 -040085static void btrfs_truncate(struct inode *inode);
Chris Masonc8b97812008-10-29 14:49:59 -040086static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
Chris Mason771ed682008-11-06 22:02:51 -050087static noinline int cow_file_range(struct inode *inode,
88 struct page *locked_page,
89 u64 start, u64 end, int *page_started,
90 unsigned long *nr_written, int unlock);
Josef Bacik7b128762008-07-24 12:17:14 -040091
Jim Owens0279b4c2009-02-04 09:29:13 -050092static int btrfs_init_inode_security(struct inode *inode, struct inode *dir)
93{
94 int err;
95
96 err = btrfs_init_acl(inode, dir);
97 if (!err)
98 err = btrfs_xattr_security_init(inode, dir);
99 return err;
100}
101
Chris Masond352ac62008-09-29 15:18:18 -0400102/*
Chris Masonc8b97812008-10-29 14:49:59 -0400103 * this does all the hard work for inserting an inline extent into
104 * the btree. The caller should have done a btrfs_drop_extents so that
105 * no overlapping inline items exist in the btree
106 */
Chris Masond3977122009-01-05 21:25:51 -0500107static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
Chris Masonc8b97812008-10-29 14:49:59 -0400108 struct btrfs_root *root, struct inode *inode,
109 u64 start, size_t size, size_t compressed_size,
110 struct page **compressed_pages)
111{
112 struct btrfs_key key;
113 struct btrfs_path *path;
114 struct extent_buffer *leaf;
115 struct page *page = NULL;
116 char *kaddr;
117 unsigned long ptr;
118 struct btrfs_file_extent_item *ei;
119 int err = 0;
120 int ret;
121 size_t cur_size = size;
122 size_t datasize;
123 unsigned long offset;
124 int use_compress = 0;
125
126 if (compressed_size && compressed_pages) {
127 use_compress = 1;
128 cur_size = compressed_size;
129 }
130
Chris Masond3977122009-01-05 21:25:51 -0500131 path = btrfs_alloc_path();
132 if (!path)
Chris Masonc8b97812008-10-29 14:49:59 -0400133 return -ENOMEM;
134
Chris Masonb9473432009-03-13 11:00:37 -0400135 path->leave_spinning = 1;
Chris Masonc8b97812008-10-29 14:49:59 -0400136 btrfs_set_trans_block_group(trans, inode);
137
138 key.objectid = inode->i_ino;
139 key.offset = start;
140 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
Chris Masonc8b97812008-10-29 14:49:59 -0400141 datasize = btrfs_file_extent_calc_inline_size(cur_size);
142
143 inode_add_bytes(inode, size);
144 ret = btrfs_insert_empty_item(trans, root, path, &key,
145 datasize);
146 BUG_ON(ret);
147 if (ret) {
148 err = ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400149 goto fail;
150 }
151 leaf = path->nodes[0];
152 ei = btrfs_item_ptr(leaf, path->slots[0],
153 struct btrfs_file_extent_item);
154 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
155 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
156 btrfs_set_file_extent_encryption(leaf, ei, 0);
157 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
158 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
159 ptr = btrfs_file_extent_inline_start(ei);
160
161 if (use_compress) {
162 struct page *cpage;
163 int i = 0;
Chris Masond3977122009-01-05 21:25:51 -0500164 while (compressed_size > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400165 cpage = compressed_pages[i];
Chris Mason5b050f02008-11-11 09:34:41 -0500166 cur_size = min_t(unsigned long, compressed_size,
Chris Masonc8b97812008-10-29 14:49:59 -0400167 PAGE_CACHE_SIZE);
168
Chris Masonb9473432009-03-13 11:00:37 -0400169 kaddr = kmap_atomic(cpage, KM_USER0);
Chris Masonc8b97812008-10-29 14:49:59 -0400170 write_extent_buffer(leaf, kaddr, ptr, cur_size);
Chris Masonb9473432009-03-13 11:00:37 -0400171 kunmap_atomic(kaddr, KM_USER0);
Chris Masonc8b97812008-10-29 14:49:59 -0400172
173 i++;
174 ptr += cur_size;
175 compressed_size -= cur_size;
176 }
177 btrfs_set_file_extent_compression(leaf, ei,
178 BTRFS_COMPRESS_ZLIB);
179 } else {
180 page = find_get_page(inode->i_mapping,
181 start >> PAGE_CACHE_SHIFT);
182 btrfs_set_file_extent_compression(leaf, ei, 0);
183 kaddr = kmap_atomic(page, KM_USER0);
184 offset = start & (PAGE_CACHE_SIZE - 1);
185 write_extent_buffer(leaf, kaddr + offset, ptr, size);
186 kunmap_atomic(kaddr, KM_USER0);
187 page_cache_release(page);
188 }
189 btrfs_mark_buffer_dirty(leaf);
190 btrfs_free_path(path);
191
192 BTRFS_I(inode)->disk_i_size = inode->i_size;
193 btrfs_update_inode(trans, root, inode);
194 return 0;
195fail:
196 btrfs_free_path(path);
197 return err;
198}
199
200
201/*
202 * conditionally insert an inline extent into the file. This
203 * does the checks required to make sure the data is small enough
204 * to fit as an inline extent.
205 */
Chris Mason7f366cf2009-03-12 20:12:45 -0400206static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
Chris Masonc8b97812008-10-29 14:49:59 -0400207 struct btrfs_root *root,
208 struct inode *inode, u64 start, u64 end,
209 size_t compressed_size,
210 struct page **compressed_pages)
211{
212 u64 isize = i_size_read(inode);
213 u64 actual_end = min(end + 1, isize);
214 u64 inline_len = actual_end - start;
215 u64 aligned_end = (end + root->sectorsize - 1) &
216 ~((u64)root->sectorsize - 1);
217 u64 hint_byte;
218 u64 data_len = inline_len;
219 int ret;
220
221 if (compressed_size)
222 data_len = compressed_size;
223
224 if (start > 0 ||
Chris Mason70b99e62008-10-31 12:46:39 -0400225 actual_end >= PAGE_CACHE_SIZE ||
Chris Masonc8b97812008-10-29 14:49:59 -0400226 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
227 (!compressed_size &&
228 (actual_end & (root->sectorsize - 1)) == 0) ||
229 end + 1 < isize ||
230 data_len > root->fs_info->max_inline) {
231 return 1;
232 }
233
Chris Masonc8b97812008-10-29 14:49:59 -0400234 ret = btrfs_drop_extents(trans, root, inode, start,
Chris Masona1ed8352009-09-11 12:27:37 -0400235 aligned_end, aligned_end, start,
236 &hint_byte, 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400237 BUG_ON(ret);
238
239 if (isize > actual_end)
240 inline_len = min_t(u64, isize, actual_end);
241 ret = insert_inline_extent(trans, root, inode, start,
242 inline_len, compressed_size,
243 compressed_pages);
244 BUG_ON(ret);
Chris Masona1ed8352009-09-11 12:27:37 -0400245 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
Chris Masonc8b97812008-10-29 14:49:59 -0400246 return 0;
247}
248
Chris Mason771ed682008-11-06 22:02:51 -0500249struct async_extent {
250 u64 start;
251 u64 ram_size;
252 u64 compressed_size;
253 struct page **pages;
254 unsigned long nr_pages;
255 struct list_head list;
256};
257
258struct async_cow {
259 struct inode *inode;
260 struct btrfs_root *root;
261 struct page *locked_page;
262 u64 start;
263 u64 end;
264 struct list_head extents;
265 struct btrfs_work work;
266};
267
268static noinline int add_async_extent(struct async_cow *cow,
269 u64 start, u64 ram_size,
270 u64 compressed_size,
271 struct page **pages,
272 unsigned long nr_pages)
273{
274 struct async_extent *async_extent;
275
276 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
277 async_extent->start = start;
278 async_extent->ram_size = ram_size;
279 async_extent->compressed_size = compressed_size;
280 async_extent->pages = pages;
281 async_extent->nr_pages = nr_pages;
282 list_add_tail(&async_extent->list, &cow->extents);
283 return 0;
284}
285
Chris Masonc8b97812008-10-29 14:49:59 -0400286/*
Chris Mason771ed682008-11-06 22:02:51 -0500287 * we create compressed extents in two phases. The first
288 * phase compresses a range of pages that have already been
289 * locked (both pages and state bits are locked).
Chris Masonc8b97812008-10-29 14:49:59 -0400290 *
Chris Mason771ed682008-11-06 22:02:51 -0500291 * This is done inside an ordered work queue, and the compression
292 * is spread across many cpus. The actual IO submission is step
293 * two, and the ordered work queue takes care of making sure that
294 * happens in the same order things were put onto the queue by
295 * writepages and friends.
Chris Masonc8b97812008-10-29 14:49:59 -0400296 *
Chris Mason771ed682008-11-06 22:02:51 -0500297 * If this code finds it can't get good compression, it puts an
298 * entry onto the work queue to write the uncompressed bytes. This
299 * makes sure that both compressed inodes and uncompressed inodes
300 * are written in the same order that pdflush sent them down.
Chris Masond352ac62008-09-29 15:18:18 -0400301 */
Chris Mason771ed682008-11-06 22:02:51 -0500302static noinline int compress_file_range(struct inode *inode,
303 struct page *locked_page,
304 u64 start, u64 end,
305 struct async_cow *async_cow,
306 int *num_added)
Chris Masonb888db22007-08-27 16:49:44 -0400307{
308 struct btrfs_root *root = BTRFS_I(inode)->root;
309 struct btrfs_trans_handle *trans;
Chris Masondb945352007-10-15 16:15:53 -0400310 u64 num_bytes;
Chris Masonc8b97812008-10-29 14:49:59 -0400311 u64 orig_start;
312 u64 disk_num_bytes;
Chris Masondb945352007-10-15 16:15:53 -0400313 u64 blocksize = root->sectorsize;
Chris Masonc8b97812008-10-29 14:49:59 -0400314 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500315 u64 isize = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400316 int ret = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400317 struct page **pages = NULL;
318 unsigned long nr_pages;
319 unsigned long nr_pages_ret = 0;
320 unsigned long total_compressed = 0;
321 unsigned long total_in = 0;
322 unsigned long max_compressed = 128 * 1024;
Chris Mason771ed682008-11-06 22:02:51 -0500323 unsigned long max_uncompressed = 128 * 1024;
Chris Masonc8b97812008-10-29 14:49:59 -0400324 int i;
325 int will_compress;
Chris Masonb888db22007-08-27 16:49:44 -0400326
Chris Masonc8b97812008-10-29 14:49:59 -0400327 orig_start = start;
Chris Masonbe20aa92007-12-17 20:14:01 -0500328
Chris Mason42dc7ba2008-12-15 11:44:56 -0500329 actual_end = min_t(u64, isize, end + 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400330again:
331 will_compress = 0;
332 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
333 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
334
Chris Masonf03d9301f2009-02-04 09:31:06 -0500335 /*
336 * we don't want to send crud past the end of i_size through
337 * compression, that's just a waste of CPU time. So, if the
338 * end of the file is before the start of our current
339 * requested range of bytes, we bail out to the uncompressed
340 * cleanup code that can deal with all of this.
341 *
342 * It isn't really the fastest way to fix things, but this is a
343 * very uncommon corner.
344 */
345 if (actual_end <= start)
346 goto cleanup_and_bail_uncompressed;
347
Chris Masonc8b97812008-10-29 14:49:59 -0400348 total_compressed = actual_end - start;
349
350 /* we want to make sure that amount of ram required to uncompress
351 * an extent is reasonable, so we limit the total size in ram
Chris Mason771ed682008-11-06 22:02:51 -0500352 * of a compressed extent to 128k. This is a crucial number
353 * because it also controls how easily we can spread reads across
354 * cpus for decompression.
355 *
356 * We also want to make sure the amount of IO required to do
357 * a random read is reasonably small, so we limit the size of
358 * a compressed extent to 128k.
Chris Masonc8b97812008-10-29 14:49:59 -0400359 */
360 total_compressed = min(total_compressed, max_uncompressed);
Chris Masondb945352007-10-15 16:15:53 -0400361 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
Chris Masonbe20aa92007-12-17 20:14:01 -0500362 num_bytes = max(blocksize, num_bytes);
Chris Masonc8b97812008-10-29 14:49:59 -0400363 disk_num_bytes = num_bytes;
364 total_in = 0;
365 ret = 0;
Chris Masondb945352007-10-15 16:15:53 -0400366
Chris Mason771ed682008-11-06 22:02:51 -0500367 /*
368 * we do compression for mount -o compress and when the
369 * inode has not been flagged as nocompress. This flag can
370 * change at any time if we discover bad compression ratios.
Chris Masonc8b97812008-10-29 14:49:59 -0400371 */
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200372 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
Chris Masonc8b97812008-10-29 14:49:59 -0400373 btrfs_test_opt(root, COMPRESS)) {
374 WARN_ON(pages);
Chris Masoncfbc2462008-10-30 13:22:14 -0400375 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
Chris Mason179e29e2007-11-01 11:28:41 -0400376
Chris Masonc8b97812008-10-29 14:49:59 -0400377 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
378 total_compressed, pages,
379 nr_pages, &nr_pages_ret,
380 &total_in,
381 &total_compressed,
382 max_compressed);
383
384 if (!ret) {
385 unsigned long offset = total_compressed &
386 (PAGE_CACHE_SIZE - 1);
387 struct page *page = pages[nr_pages_ret - 1];
388 char *kaddr;
389
390 /* zero the tail end of the last page, we might be
391 * sending it down to disk
392 */
393 if (offset) {
394 kaddr = kmap_atomic(page, KM_USER0);
395 memset(kaddr + offset, 0,
396 PAGE_CACHE_SIZE - offset);
397 kunmap_atomic(kaddr, KM_USER0);
398 }
399 will_compress = 1;
400 }
401 }
402 if (start == 0) {
Chris Mason771ed682008-11-06 22:02:51 -0500403 trans = btrfs_join_transaction(root, 1);
404 BUG_ON(!trans);
405 btrfs_set_trans_block_group(trans, inode);
406
Chris Masonc8b97812008-10-29 14:49:59 -0400407 /* lets try to make an inline extent */
Chris Mason771ed682008-11-06 22:02:51 -0500408 if (ret || total_in < (actual_end - start)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400409 /* we didn't compress the entire range, try
Chris Mason771ed682008-11-06 22:02:51 -0500410 * to make an uncompressed inline extent.
Chris Masonc8b97812008-10-29 14:49:59 -0400411 */
412 ret = cow_file_range_inline(trans, root, inode,
413 start, end, 0, NULL);
414 } else {
Chris Mason771ed682008-11-06 22:02:51 -0500415 /* try making a compressed inline extent */
Chris Masonc8b97812008-10-29 14:49:59 -0400416 ret = cow_file_range_inline(trans, root, inode,
417 start, end,
418 total_compressed, pages);
419 }
Chris Mason771ed682008-11-06 22:02:51 -0500420 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400421 if (ret == 0) {
Chris Mason771ed682008-11-06 22:02:51 -0500422 /*
423 * inline extent creation worked, we don't need
424 * to create any more async work items. Unlock
425 * and free up our temp pages.
426 */
Chris Masonc8b97812008-10-29 14:49:59 -0400427 extent_clear_unlock_delalloc(inode,
428 &BTRFS_I(inode)->io_tree,
Chris Mason771ed682008-11-06 22:02:51 -0500429 start, end, NULL, 1, 0,
Chris Mason8b62b722009-09-02 16:53:46 -0400430 0, 1, 1, 1, 0);
Chris Masonc8b97812008-10-29 14:49:59 -0400431 ret = 0;
432 goto free_pages_out;
433 }
434 }
435
436 if (will_compress) {
437 /*
438 * we aren't doing an inline extent round the compressed size
439 * up to a block size boundary so the allocator does sane
440 * things
441 */
442 total_compressed = (total_compressed + blocksize - 1) &
443 ~(blocksize - 1);
444
445 /*
446 * one last check to make sure the compression is really a
447 * win, compare the page count read with the blocks on disk
448 */
449 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
450 ~(PAGE_CACHE_SIZE - 1);
451 if (total_compressed >= total_in) {
452 will_compress = 0;
453 } else {
454 disk_num_bytes = total_compressed;
455 num_bytes = total_in;
456 }
457 }
458 if (!will_compress && pages) {
459 /*
460 * the compression code ran but failed to make things smaller,
461 * free any pages it allocated and our page pointer array
462 */
463 for (i = 0; i < nr_pages_ret; i++) {
Chris Mason70b99e62008-10-31 12:46:39 -0400464 WARN_ON(pages[i]->mapping);
Chris Masonc8b97812008-10-29 14:49:59 -0400465 page_cache_release(pages[i]);
466 }
467 kfree(pages);
468 pages = NULL;
469 total_compressed = 0;
470 nr_pages_ret = 0;
471
472 /* flag the file so we don't compress in the future */
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200473 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
Chris Masonc8b97812008-10-29 14:49:59 -0400474 }
Chris Mason771ed682008-11-06 22:02:51 -0500475 if (will_compress) {
476 *num_added += 1;
477
478 /* the async work queues will take care of doing actual
479 * allocation on disk for these compressed pages,
480 * and will submit them to the elevator.
481 */
482 add_async_extent(async_cow, start, num_bytes,
483 total_compressed, pages, nr_pages_ret);
484
Chris Mason42dc7ba2008-12-15 11:44:56 -0500485 if (start + num_bytes < end && start + num_bytes < actual_end) {
Chris Mason771ed682008-11-06 22:02:51 -0500486 start += num_bytes;
487 pages = NULL;
488 cond_resched();
489 goto again;
490 }
491 } else {
Chris Masonf03d9301f2009-02-04 09:31:06 -0500492cleanup_and_bail_uncompressed:
Chris Mason771ed682008-11-06 22:02:51 -0500493 /*
494 * No compression, but we still need to write the pages in
495 * the file we've been given so far. redirty the locked
496 * page if it corresponds to our extent and set things up
497 * for the async work queue to run cow_file_range to do
498 * the normal delalloc dance
499 */
500 if (page_offset(locked_page) >= start &&
501 page_offset(locked_page) <= end) {
502 __set_page_dirty_nobuffers(locked_page);
503 /* unlocked later on in the async handlers */
504 }
505 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
506 *num_added += 1;
507 }
508
509out:
510 return 0;
511
512free_pages_out:
513 for (i = 0; i < nr_pages_ret; i++) {
514 WARN_ON(pages[i]->mapping);
515 page_cache_release(pages[i]);
516 }
Chris Masond3977122009-01-05 21:25:51 -0500517 kfree(pages);
Chris Mason771ed682008-11-06 22:02:51 -0500518
519 goto out;
520}
521
522/*
523 * phase two of compressed writeback. This is the ordered portion
524 * of the code, which only gets called in the order the work was
525 * queued. We walk all the async extents created by compress_file_range
526 * and send them down to the disk.
527 */
528static noinline int submit_compressed_extents(struct inode *inode,
529 struct async_cow *async_cow)
530{
531 struct async_extent *async_extent;
532 u64 alloc_hint = 0;
533 struct btrfs_trans_handle *trans;
534 struct btrfs_key ins;
535 struct extent_map *em;
536 struct btrfs_root *root = BTRFS_I(inode)->root;
537 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
538 struct extent_io_tree *io_tree;
539 int ret;
540
541 if (list_empty(&async_cow->extents))
542 return 0;
543
544 trans = btrfs_join_transaction(root, 1);
545
Chris Masond3977122009-01-05 21:25:51 -0500546 while (!list_empty(&async_cow->extents)) {
Chris Mason771ed682008-11-06 22:02:51 -0500547 async_extent = list_entry(async_cow->extents.next,
548 struct async_extent, list);
549 list_del(&async_extent->list);
550
551 io_tree = &BTRFS_I(inode)->io_tree;
552
553 /* did the compression code fall back to uncompressed IO? */
554 if (!async_extent->pages) {
555 int page_started = 0;
556 unsigned long nr_written = 0;
557
558 lock_extent(io_tree, async_extent->start,
Chris Masond3977122009-01-05 21:25:51 -0500559 async_extent->start +
560 async_extent->ram_size - 1, GFP_NOFS);
Chris Mason771ed682008-11-06 22:02:51 -0500561
562 /* allocate blocks */
563 cow_file_range(inode, async_cow->locked_page,
564 async_extent->start,
565 async_extent->start +
566 async_extent->ram_size - 1,
567 &page_started, &nr_written, 0);
568
569 /*
570 * if page_started, cow_file_range inserted an
571 * inline extent and took care of all the unlocking
572 * and IO for us. Otherwise, we need to submit
573 * all those pages down to the drive.
574 */
575 if (!page_started)
576 extent_write_locked_range(io_tree,
577 inode, async_extent->start,
Chris Masond3977122009-01-05 21:25:51 -0500578 async_extent->start +
Chris Mason771ed682008-11-06 22:02:51 -0500579 async_extent->ram_size - 1,
580 btrfs_get_extent,
581 WB_SYNC_ALL);
582 kfree(async_extent);
583 cond_resched();
584 continue;
585 }
586
587 lock_extent(io_tree, async_extent->start,
588 async_extent->start + async_extent->ram_size - 1,
589 GFP_NOFS);
590 /*
591 * here we're doing allocation and writeback of the
592 * compressed pages
593 */
594 btrfs_drop_extent_cache(inode, async_extent->start,
595 async_extent->start +
596 async_extent->ram_size - 1, 0);
597
598 ret = btrfs_reserve_extent(trans, root,
599 async_extent->compressed_size,
600 async_extent->compressed_size,
601 0, alloc_hint,
602 (u64)-1, &ins, 1);
603 BUG_ON(ret);
604 em = alloc_extent_map(GFP_NOFS);
605 em->start = async_extent->start;
606 em->len = async_extent->ram_size;
Chris Mason445a6942008-11-10 11:53:33 -0500607 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500608
609 em->block_start = ins.objectid;
610 em->block_len = ins.offset;
611 em->bdev = root->fs_info->fs_devices->latest_bdev;
612 set_bit(EXTENT_FLAG_PINNED, &em->flags);
613 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
614
Chris Masond3977122009-01-05 21:25:51 -0500615 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400616 write_lock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500617 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -0400618 write_unlock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500619 if (ret != -EEXIST) {
620 free_extent_map(em);
621 break;
622 }
623 btrfs_drop_extent_cache(inode, async_extent->start,
624 async_extent->start +
625 async_extent->ram_size - 1, 0);
626 }
627
628 ret = btrfs_add_ordered_extent(inode, async_extent->start,
629 ins.objectid,
630 async_extent->ram_size,
631 ins.offset,
632 BTRFS_ORDERED_COMPRESSED);
633 BUG_ON(ret);
634
635 btrfs_end_transaction(trans, root);
636
637 /*
638 * clear dirty, set writeback and unlock the pages.
639 */
640 extent_clear_unlock_delalloc(inode,
641 &BTRFS_I(inode)->io_tree,
642 async_extent->start,
643 async_extent->start +
644 async_extent->ram_size - 1,
Chris Mason8b62b722009-09-02 16:53:46 -0400645 NULL, 1, 1, 0, 1, 1, 0, 0);
Chris Mason771ed682008-11-06 22:02:51 -0500646
647 ret = btrfs_submit_compressed_write(inode,
Chris Masond3977122009-01-05 21:25:51 -0500648 async_extent->start,
649 async_extent->ram_size,
650 ins.objectid,
651 ins.offset, async_extent->pages,
652 async_extent->nr_pages);
Chris Mason771ed682008-11-06 22:02:51 -0500653
654 BUG_ON(ret);
655 trans = btrfs_join_transaction(root, 1);
656 alloc_hint = ins.objectid + ins.offset;
657 kfree(async_extent);
658 cond_resched();
659 }
660
661 btrfs_end_transaction(trans, root);
662 return 0;
663}
664
665/*
666 * when extent_io.c finds a delayed allocation range in the file,
667 * the call backs end up in this code. The basic idea is to
668 * allocate extents on disk for the range, and create ordered data structs
669 * in ram to track those extents.
670 *
671 * locked_page is the page that writepage had locked already. We use
672 * it to make sure we don't do extra locks or unlocks.
673 *
674 * *page_started is set to one if we unlock locked_page and do everything
675 * required to start IO on it. It may be clean and already done with
676 * IO when we return.
677 */
678static noinline int cow_file_range(struct inode *inode,
679 struct page *locked_page,
680 u64 start, u64 end, int *page_started,
681 unsigned long *nr_written,
682 int unlock)
683{
684 struct btrfs_root *root = BTRFS_I(inode)->root;
685 struct btrfs_trans_handle *trans;
686 u64 alloc_hint = 0;
687 u64 num_bytes;
688 unsigned long ram_size;
689 u64 disk_num_bytes;
690 u64 cur_alloc_size;
691 u64 blocksize = root->sectorsize;
692 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500693 u64 isize = i_size_read(inode);
Chris Mason771ed682008-11-06 22:02:51 -0500694 struct btrfs_key ins;
695 struct extent_map *em;
696 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
697 int ret = 0;
698
699 trans = btrfs_join_transaction(root, 1);
700 BUG_ON(!trans);
701 btrfs_set_trans_block_group(trans, inode);
702
Chris Mason42dc7ba2008-12-15 11:44:56 -0500703 actual_end = min_t(u64, isize, end + 1);
Chris Mason771ed682008-11-06 22:02:51 -0500704
705 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
706 num_bytes = max(blocksize, num_bytes);
707 disk_num_bytes = num_bytes;
708 ret = 0;
709
710 if (start == 0) {
711 /* lets try to make an inline extent */
712 ret = cow_file_range_inline(trans, root, inode,
713 start, end, 0, NULL);
714 if (ret == 0) {
715 extent_clear_unlock_delalloc(inode,
716 &BTRFS_I(inode)->io_tree,
717 start, end, NULL, 1, 1,
Chris Mason8b62b722009-09-02 16:53:46 -0400718 1, 1, 1, 1, 0);
Chris Mason771ed682008-11-06 22:02:51 -0500719 *nr_written = *nr_written +
720 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
721 *page_started = 1;
722 ret = 0;
723 goto out;
724 }
725 }
Chris Masonc8b97812008-10-29 14:49:59 -0400726
727 BUG_ON(disk_num_bytes >
728 btrfs_super_total_bytes(&root->fs_info->super_copy));
729
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400730 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
Chris Mason3b951512008-04-17 11:29:12 -0400731
Chris Masond3977122009-01-05 21:25:51 -0500732 while (disk_num_bytes > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400733 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400734 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
Chris Mason771ed682008-11-06 22:02:51 -0500735 root->sectorsize, 0, alloc_hint,
Chris Masone6dcd2d2008-07-17 12:53:50 -0400736 (u64)-1, &ins, 1);
Chris Masond3977122009-01-05 21:25:51 -0500737 BUG_ON(ret);
738
Chris Masone6dcd2d2008-07-17 12:53:50 -0400739 em = alloc_extent_map(GFP_NOFS);
740 em->start = start;
Chris Mason445a6942008-11-10 11:53:33 -0500741 em->orig_start = em->start;
Chris Masonc8b97812008-10-29 14:49:59 -0400742
Chris Mason771ed682008-11-06 22:02:51 -0500743 ram_size = ins.offset;
744 em->len = ins.offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400745
Chris Masone6dcd2d2008-07-17 12:53:50 -0400746 em->block_start = ins.objectid;
Chris Masonc8b97812008-10-29 14:49:59 -0400747 em->block_len = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400748 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason7f3c74f2008-07-18 12:01:11 -0400749 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -0400750
Chris Masond3977122009-01-05 21:25:51 -0500751 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400752 write_lock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400753 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -0400754 write_unlock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400755 if (ret != -EEXIST) {
756 free_extent_map(em);
757 break;
758 }
759 btrfs_drop_extent_cache(inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -0400760 start + ram_size - 1, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400761 }
762
Chris Mason98d20f62008-04-14 09:46:10 -0400763 cur_alloc_size = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400764 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
Chris Mason771ed682008-11-06 22:02:51 -0500765 ram_size, cur_alloc_size, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400766 BUG_ON(ret);
Chris Masonc8b97812008-10-29 14:49:59 -0400767
Yan Zheng17d217f2008-12-12 10:03:38 -0500768 if (root->root_key.objectid ==
769 BTRFS_DATA_RELOC_TREE_OBJECTID) {
770 ret = btrfs_reloc_clone_csums(inode, start,
771 cur_alloc_size);
772 BUG_ON(ret);
773 }
774
Chris Masond3977122009-01-05 21:25:51 -0500775 if (disk_num_bytes < cur_alloc_size)
Chris Mason3b951512008-04-17 11:29:12 -0400776 break;
Chris Masond3977122009-01-05 21:25:51 -0500777
Chris Masonc8b97812008-10-29 14:49:59 -0400778 /* we're not doing compressed IO, don't unlock the first
779 * page (which the caller expects to stay locked), don't
780 * clear any dirty bits and don't set any writeback bits
Chris Mason8b62b722009-09-02 16:53:46 -0400781 *
782 * Do set the Private2 bit so we know this page was properly
783 * setup for writepage
Chris Masonc8b97812008-10-29 14:49:59 -0400784 */
785 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
786 start, start + ram_size - 1,
Chris Mason771ed682008-11-06 22:02:51 -0500787 locked_page, unlock, 1,
Chris Mason8b62b722009-09-02 16:53:46 -0400788 1, 0, 0, 0, 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400789 disk_num_bytes -= cur_alloc_size;
Chris Masonc59f8952007-12-17 20:14:04 -0500790 num_bytes -= cur_alloc_size;
791 alloc_hint = ins.objectid + ins.offset;
792 start += cur_alloc_size;
Chris Masonb888db22007-08-27 16:49:44 -0400793 }
Chris Masonb888db22007-08-27 16:49:44 -0400794out:
Chris Mason771ed682008-11-06 22:02:51 -0500795 ret = 0;
Chris Masonb888db22007-08-27 16:49:44 -0400796 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400797
Chris Masonbe20aa92007-12-17 20:14:01 -0500798 return ret;
Chris Mason771ed682008-11-06 22:02:51 -0500799}
Chris Masonc8b97812008-10-29 14:49:59 -0400800
Chris Mason771ed682008-11-06 22:02:51 -0500801/*
802 * work queue call back to started compression on a file and pages
803 */
804static noinline void async_cow_start(struct btrfs_work *work)
805{
806 struct async_cow *async_cow;
807 int num_added = 0;
808 async_cow = container_of(work, struct async_cow, work);
809
810 compress_file_range(async_cow->inode, async_cow->locked_page,
811 async_cow->start, async_cow->end, async_cow,
812 &num_added);
813 if (num_added == 0)
814 async_cow->inode = NULL;
815}
816
817/*
818 * work queue call back to submit previously compressed pages
819 */
820static noinline void async_cow_submit(struct btrfs_work *work)
821{
822 struct async_cow *async_cow;
823 struct btrfs_root *root;
824 unsigned long nr_pages;
825
826 async_cow = container_of(work, struct async_cow, work);
827
828 root = async_cow->root;
829 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
830 PAGE_CACHE_SHIFT;
831
832 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
833
834 if (atomic_read(&root->fs_info->async_delalloc_pages) <
835 5 * 1042 * 1024 &&
836 waitqueue_active(&root->fs_info->async_submit_wait))
837 wake_up(&root->fs_info->async_submit_wait);
838
Chris Masond3977122009-01-05 21:25:51 -0500839 if (async_cow->inode)
Chris Mason771ed682008-11-06 22:02:51 -0500840 submit_compressed_extents(async_cow->inode, async_cow);
Chris Mason771ed682008-11-06 22:02:51 -0500841}
Chris Masonc8b97812008-10-29 14:49:59 -0400842
Chris Mason771ed682008-11-06 22:02:51 -0500843static noinline void async_cow_free(struct btrfs_work *work)
844{
845 struct async_cow *async_cow;
846 async_cow = container_of(work, struct async_cow, work);
847 kfree(async_cow);
848}
849
850static int cow_file_range_async(struct inode *inode, struct page *locked_page,
851 u64 start, u64 end, int *page_started,
852 unsigned long *nr_written)
853{
854 struct async_cow *async_cow;
855 struct btrfs_root *root = BTRFS_I(inode)->root;
856 unsigned long nr_pages;
857 u64 cur_end;
858 int limit = 10 * 1024 * 1042;
859
Chris Mason771ed682008-11-06 22:02:51 -0500860 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
Chris Mason2c64c532009-09-02 15:04:12 -0400861 EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -0500862 while (start < end) {
Chris Mason771ed682008-11-06 22:02:51 -0500863 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
864 async_cow->inode = inode;
865 async_cow->root = root;
866 async_cow->locked_page = locked_page;
867 async_cow->start = start;
868
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200869 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
Chris Mason771ed682008-11-06 22:02:51 -0500870 cur_end = end;
871 else
872 cur_end = min(end, start + 512 * 1024 - 1);
873
874 async_cow->end = cur_end;
875 INIT_LIST_HEAD(&async_cow->extents);
876
877 async_cow->work.func = async_cow_start;
878 async_cow->work.ordered_func = async_cow_submit;
879 async_cow->work.ordered_free = async_cow_free;
880 async_cow->work.flags = 0;
881
Chris Mason771ed682008-11-06 22:02:51 -0500882 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
883 PAGE_CACHE_SHIFT;
884 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
885
886 btrfs_queue_worker(&root->fs_info->delalloc_workers,
887 &async_cow->work);
888
889 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
890 wait_event(root->fs_info->async_submit_wait,
891 (atomic_read(&root->fs_info->async_delalloc_pages) <
892 limit));
893 }
894
Chris Masond3977122009-01-05 21:25:51 -0500895 while (atomic_read(&root->fs_info->async_submit_draining) &&
Chris Mason771ed682008-11-06 22:02:51 -0500896 atomic_read(&root->fs_info->async_delalloc_pages)) {
897 wait_event(root->fs_info->async_submit_wait,
898 (atomic_read(&root->fs_info->async_delalloc_pages) ==
899 0));
900 }
901
902 *nr_written += nr_pages;
903 start = cur_end + 1;
904 }
905 *page_started = 1;
906 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -0500907}
908
Chris Masond3977122009-01-05 21:25:51 -0500909static noinline int csum_exist_in_range(struct btrfs_root *root,
Yan Zheng17d217f2008-12-12 10:03:38 -0500910 u64 bytenr, u64 num_bytes)
911{
912 int ret;
913 struct btrfs_ordered_sum *sums;
914 LIST_HEAD(list);
915
Yan Zheng07d400a2009-01-06 11:42:00 -0500916 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
917 bytenr + num_bytes - 1, &list);
Yan Zheng17d217f2008-12-12 10:03:38 -0500918 if (ret == 0 && list_empty(&list))
919 return 0;
920
921 while (!list_empty(&list)) {
922 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
923 list_del(&sums->list);
924 kfree(sums);
925 }
926 return 1;
927}
928
Chris Masond352ac62008-09-29 15:18:18 -0400929/*
930 * when nowcow writeback call back. This checks for snapshots or COW copies
931 * of the extents that exist in the file, and COWs the file as required.
932 *
933 * If no cow copies or snapshots exist, we write directly to the existing
934 * blocks on disk
935 */
Chris Mason7f366cf2009-03-12 20:12:45 -0400936static noinline int run_delalloc_nocow(struct inode *inode,
937 struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -0500938 u64 start, u64 end, int *page_started, int force,
939 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -0500940{
Chris Masonbe20aa92007-12-17 20:14:01 -0500941 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zheng7ea394f2008-08-05 13:05:02 -0400942 struct btrfs_trans_handle *trans;
Chris Masonbe20aa92007-12-17 20:14:01 -0500943 struct extent_buffer *leaf;
Chris Masonbe20aa92007-12-17 20:14:01 -0500944 struct btrfs_path *path;
Yan Zheng80ff3852008-10-30 14:20:02 -0400945 struct btrfs_file_extent_item *fi;
Chris Masonbe20aa92007-12-17 20:14:01 -0500946 struct btrfs_key found_key;
Yan Zheng80ff3852008-10-30 14:20:02 -0400947 u64 cow_start;
948 u64 cur_offset;
949 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400950 u64 extent_offset;
Yan Zheng80ff3852008-10-30 14:20:02 -0400951 u64 disk_bytenr;
952 u64 num_bytes;
953 int extent_type;
954 int ret;
Yan Zhengd899e052008-10-30 14:25:28 -0400955 int type;
Yan Zheng80ff3852008-10-30 14:20:02 -0400956 int nocow;
957 int check_prev = 1;
Chris Masonbe20aa92007-12-17 20:14:01 -0500958
959 path = btrfs_alloc_path();
960 BUG_ON(!path);
Yan Zheng7ea394f2008-08-05 13:05:02 -0400961 trans = btrfs_join_transaction(root, 1);
962 BUG_ON(!trans);
Chris Masonbe20aa92007-12-17 20:14:01 -0500963
Yan Zheng80ff3852008-10-30 14:20:02 -0400964 cow_start = (u64)-1;
965 cur_offset = start;
966 while (1) {
967 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
968 cur_offset, 0);
969 BUG_ON(ret < 0);
970 if (ret > 0 && path->slots[0] > 0 && check_prev) {
971 leaf = path->nodes[0];
972 btrfs_item_key_to_cpu(leaf, &found_key,
973 path->slots[0] - 1);
974 if (found_key.objectid == inode->i_ino &&
975 found_key.type == BTRFS_EXTENT_DATA_KEY)
976 path->slots[0]--;
977 }
978 check_prev = 0;
979next_slot:
980 leaf = path->nodes[0];
981 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
982 ret = btrfs_next_leaf(root, path);
983 if (ret < 0)
984 BUG_ON(1);
985 if (ret > 0)
986 break;
987 leaf = path->nodes[0];
988 }
Chris Masonbe20aa92007-12-17 20:14:01 -0500989
Yan Zheng80ff3852008-10-30 14:20:02 -0400990 nocow = 0;
991 disk_bytenr = 0;
Yan Zheng17d217f2008-12-12 10:03:38 -0500992 num_bytes = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -0400993 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -0500994
Yan Zheng80ff3852008-10-30 14:20:02 -0400995 if (found_key.objectid > inode->i_ino ||
996 found_key.type > BTRFS_EXTENT_DATA_KEY ||
997 found_key.offset > end)
998 break;
Chris Masonbe20aa92007-12-17 20:14:01 -0500999
Yan Zheng80ff3852008-10-30 14:20:02 -04001000 if (found_key.offset > cur_offset) {
1001 extent_end = found_key.offset;
1002 goto out_check;
1003 }
Chris Masonc31f8832008-01-08 15:46:31 -05001004
Yan Zheng80ff3852008-10-30 14:20:02 -04001005 fi = btrfs_item_ptr(leaf, path->slots[0],
1006 struct btrfs_file_extent_item);
1007 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Masonbe20aa92007-12-17 20:14:01 -05001008
Yan Zhengd899e052008-10-30 14:25:28 -04001009 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1010 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001011 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001012 extent_offset = btrfs_file_extent_offset(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001013 extent_end = found_key.offset +
1014 btrfs_file_extent_num_bytes(leaf, fi);
1015 if (extent_end <= start) {
1016 path->slots[0]++;
1017 goto next_slot;
1018 }
Yan Zheng17d217f2008-12-12 10:03:38 -05001019 if (disk_bytenr == 0)
1020 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001021 if (btrfs_file_extent_compression(leaf, fi) ||
1022 btrfs_file_extent_encryption(leaf, fi) ||
1023 btrfs_file_extent_other_encoding(leaf, fi))
1024 goto out_check;
Yan Zhengd899e052008-10-30 14:25:28 -04001025 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1026 goto out_check;
Yan Zhengd2fb3432008-12-11 16:30:39 -05001027 if (btrfs_extent_readonly(root, disk_bytenr))
Yan Zheng80ff3852008-10-30 14:20:02 -04001028 goto out_check;
Yan Zheng17d217f2008-12-12 10:03:38 -05001029 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001030 found_key.offset -
1031 extent_offset, disk_bytenr))
Yan Zheng17d217f2008-12-12 10:03:38 -05001032 goto out_check;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001033 disk_bytenr += extent_offset;
Yan Zheng17d217f2008-12-12 10:03:38 -05001034 disk_bytenr += cur_offset - found_key.offset;
1035 num_bytes = min(end + 1, extent_end) - cur_offset;
1036 /*
1037 * force cow if csum exists in the range.
1038 * this ensure that csum for a given extent are
1039 * either valid or do not exist.
1040 */
1041 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1042 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001043 nocow = 1;
1044 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1045 extent_end = found_key.offset +
1046 btrfs_file_extent_inline_len(leaf, fi);
1047 extent_end = ALIGN(extent_end, root->sectorsize);
1048 } else {
1049 BUG_ON(1);
1050 }
1051out_check:
1052 if (extent_end <= start) {
1053 path->slots[0]++;
1054 goto next_slot;
1055 }
1056 if (!nocow) {
1057 if (cow_start == (u64)-1)
1058 cow_start = cur_offset;
1059 cur_offset = extent_end;
1060 if (cur_offset > end)
1061 break;
1062 path->slots[0]++;
1063 goto next_slot;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001064 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001065
Yan Zheng7ea394f2008-08-05 13:05:02 -04001066 btrfs_release_path(root, path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001067 if (cow_start != (u64)-1) {
1068 ret = cow_file_range(inode, locked_page, cow_start,
Chris Mason771ed682008-11-06 22:02:51 -05001069 found_key.offset - 1, page_started,
1070 nr_written, 1);
Yan Zheng80ff3852008-10-30 14:20:02 -04001071 BUG_ON(ret);
1072 cow_start = (u64)-1;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001073 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001074
Yan Zhengd899e052008-10-30 14:25:28 -04001075 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1076 struct extent_map *em;
1077 struct extent_map_tree *em_tree;
1078 em_tree = &BTRFS_I(inode)->extent_tree;
1079 em = alloc_extent_map(GFP_NOFS);
1080 em->start = cur_offset;
Chris Mason445a6942008-11-10 11:53:33 -05001081 em->orig_start = em->start;
Yan Zhengd899e052008-10-30 14:25:28 -04001082 em->len = num_bytes;
1083 em->block_len = num_bytes;
1084 em->block_start = disk_bytenr;
1085 em->bdev = root->fs_info->fs_devices->latest_bdev;
1086 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1087 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04001088 write_lock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001089 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04001090 write_unlock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001091 if (ret != -EEXIST) {
1092 free_extent_map(em);
1093 break;
1094 }
1095 btrfs_drop_extent_cache(inode, em->start,
1096 em->start + em->len - 1, 0);
1097 }
1098 type = BTRFS_ORDERED_PREALLOC;
1099 } else {
1100 type = BTRFS_ORDERED_NOCOW;
1101 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001102
1103 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
Yan Zhengd899e052008-10-30 14:25:28 -04001104 num_bytes, num_bytes, type);
1105 BUG_ON(ret);
Chris Mason771ed682008-11-06 22:02:51 -05001106
Yan Zhengd899e052008-10-30 14:25:28 -04001107 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1108 cur_offset, cur_offset + num_bytes - 1,
Chris Mason8b62b722009-09-02 16:53:46 -04001109 locked_page, 1, 1, 1, 0, 0, 0, 1);
Yan Zheng80ff3852008-10-30 14:20:02 -04001110 cur_offset = extent_end;
1111 if (cur_offset > end)
1112 break;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001113 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001114 btrfs_release_path(root, path);
1115
1116 if (cur_offset <= end && cow_start == (u64)-1)
1117 cow_start = cur_offset;
1118 if (cow_start != (u64)-1) {
1119 ret = cow_file_range(inode, locked_page, cow_start, end,
Chris Mason771ed682008-11-06 22:02:51 -05001120 page_started, nr_written, 1);
Yan Zheng80ff3852008-10-30 14:20:02 -04001121 BUG_ON(ret);
1122 }
1123
1124 ret = btrfs_end_transaction(trans, root);
1125 BUG_ON(ret);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001126 btrfs_free_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001127 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05001128}
1129
Chris Masond352ac62008-09-29 15:18:18 -04001130/*
1131 * extent_io.c call back to do delayed allocation processing
1132 */
Chris Masonc8b97812008-10-29 14:49:59 -04001133static int run_delalloc_range(struct inode *inode, struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001134 u64 start, u64 end, int *page_started,
1135 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001136{
Chris Masonbe20aa92007-12-17 20:14:01 -05001137 int ret;
Chris Mason7f366cf2009-03-12 20:12:45 -04001138 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masona2135012008-06-25 16:01:30 -04001139
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001140 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
Chris Masonc8b97812008-10-29 14:49:59 -04001141 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001142 page_started, 1, nr_written);
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001143 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
Yan Zhengd899e052008-10-30 14:25:28 -04001144 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001145 page_started, 0, nr_written);
Chris Mason7f366cf2009-03-12 20:12:45 -04001146 else if (!btrfs_test_opt(root, COMPRESS))
1147 ret = cow_file_range(inode, locked_page, start, end,
1148 page_started, nr_written, 1);
Chris Masonbe20aa92007-12-17 20:14:01 -05001149 else
Chris Mason771ed682008-11-06 22:02:51 -05001150 ret = cow_file_range_async(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001151 page_started, nr_written);
Chris Masonb888db22007-08-27 16:49:44 -04001152 return ret;
1153}
1154
Chris Masond352ac62008-09-29 15:18:18 -04001155/*
1156 * extent_io.c set_bit_hook, used to track delayed allocation
1157 * bytes in this file, and to maintain the list of inodes that
1158 * have pending delalloc work to be done.
1159 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001160static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
Chris Masonb0c68f82008-01-31 11:05:37 -05001161 unsigned long old, unsigned long bits)
Chris Mason291d6732008-01-29 15:55:23 -05001162{
Chris Mason75eff682008-12-15 15:54:40 -05001163 /*
1164 * set_bit and clear bit hooks normally require _irqsave/restore
1165 * but in this case, we are only testeing for the DELALLOC
1166 * bit, which is only set or cleared with irqs on
1167 */
Chris Masonb0c68f82008-01-31 11:05:37 -05001168 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001169 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik6a632092009-02-20 11:00:09 -05001170 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
Chris Mason75eff682008-12-15 15:54:40 -05001171 spin_lock(&root->fs_info->delalloc_lock);
Chris Mason90692182008-02-08 13:49:28 -05001172 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
Chris Mason291d6732008-01-29 15:55:23 -05001173 root->fs_info->delalloc_bytes += end - start + 1;
Chris Masonea8c2812008-08-04 23:17:27 -04001174 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1175 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1176 &root->fs_info->delalloc_inodes);
1177 }
Chris Mason75eff682008-12-15 15:54:40 -05001178 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason291d6732008-01-29 15:55:23 -05001179 }
1180 return 0;
1181}
1182
Chris Masond352ac62008-09-29 15:18:18 -04001183/*
1184 * extent_io.c clear_bit_hook, see set_bit_hook for why
1185 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001186static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
Chris Masonb0c68f82008-01-31 11:05:37 -05001187 unsigned long old, unsigned long bits)
Chris Mason291d6732008-01-29 15:55:23 -05001188{
Chris Mason75eff682008-12-15 15:54:40 -05001189 /*
1190 * set_bit and clear bit hooks normally require _irqsave/restore
1191 * but in this case, we are only testeing for the DELALLOC
1192 * bit, which is only set or cleared with irqs on
1193 */
Chris Masonb0c68f82008-01-31 11:05:37 -05001194 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001195 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masonbcbfce82008-04-22 13:26:47 -04001196
Chris Mason75eff682008-12-15 15:54:40 -05001197 spin_lock(&root->fs_info->delalloc_lock);
Chris Masonb0c68f82008-01-31 11:05:37 -05001198 if (end - start + 1 > root->fs_info->delalloc_bytes) {
Chris Masond3977122009-01-05 21:25:51 -05001199 printk(KERN_INFO "btrfs warning: delalloc account "
1200 "%llu %llu\n",
1201 (unsigned long long)end - start + 1,
1202 (unsigned long long)
1203 root->fs_info->delalloc_bytes);
Josef Bacik6a632092009-02-20 11:00:09 -05001204 btrfs_delalloc_free_space(root, inode, (u64)-1);
Chris Masonb0c68f82008-01-31 11:05:37 -05001205 root->fs_info->delalloc_bytes = 0;
Chris Mason90692182008-02-08 13:49:28 -05001206 BTRFS_I(inode)->delalloc_bytes = 0;
Chris Masonb0c68f82008-01-31 11:05:37 -05001207 } else {
Josef Bacik6a632092009-02-20 11:00:09 -05001208 btrfs_delalloc_free_space(root, inode,
1209 end - start + 1);
Chris Masonb0c68f82008-01-31 11:05:37 -05001210 root->fs_info->delalloc_bytes -= end - start + 1;
Chris Mason90692182008-02-08 13:49:28 -05001211 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
Chris Masonb0c68f82008-01-31 11:05:37 -05001212 }
Chris Masonea8c2812008-08-04 23:17:27 -04001213 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1214 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1215 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1216 }
Chris Mason75eff682008-12-15 15:54:40 -05001217 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason291d6732008-01-29 15:55:23 -05001218 }
1219 return 0;
1220}
1221
Chris Masond352ac62008-09-29 15:18:18 -04001222/*
1223 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1224 * we don't create bios that span stripes or chunks
1225 */
Chris Mason239b14b2008-03-24 15:02:07 -04001226int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
Chris Masonc8b97812008-10-29 14:49:59 -04001227 size_t size, struct bio *bio,
1228 unsigned long bio_flags)
Chris Mason239b14b2008-03-24 15:02:07 -04001229{
1230 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1231 struct btrfs_mapping_tree *map_tree;
Chris Masona62b9402008-10-03 16:31:08 -04001232 u64 logical = (u64)bio->bi_sector << 9;
Chris Mason239b14b2008-03-24 15:02:07 -04001233 u64 length = 0;
1234 u64 map_length;
Chris Mason239b14b2008-03-24 15:02:07 -04001235 int ret;
1236
Chris Mason771ed682008-11-06 22:02:51 -05001237 if (bio_flags & EXTENT_BIO_COMPRESSED)
1238 return 0;
1239
Chris Masonf2d8d742008-04-21 10:03:05 -04001240 length = bio->bi_size;
Chris Mason239b14b2008-03-24 15:02:07 -04001241 map_tree = &root->fs_info->mapping_tree;
1242 map_length = length;
Chris Masoncea9e442008-04-09 16:28:12 -04001243 ret = btrfs_map_block(map_tree, READ, logical,
Chris Masonf1885912008-04-09 16:28:12 -04001244 &map_length, NULL, 0);
Chris Masoncea9e442008-04-09 16:28:12 -04001245
Chris Masond3977122009-01-05 21:25:51 -05001246 if (map_length < length + size)
Chris Mason239b14b2008-03-24 15:02:07 -04001247 return 1;
Chris Mason239b14b2008-03-24 15:02:07 -04001248 return 0;
1249}
1250
Chris Masond352ac62008-09-29 15:18:18 -04001251/*
1252 * in order to insert checksums into the metadata in large chunks,
1253 * we wait until bio submission time. All the pages in the bio are
1254 * checksummed and sums are attached onto the ordered extent record.
1255 *
1256 * At IO completion time the cums attached on the ordered extent record
1257 * are inserted into the btree
1258 */
Chris Masond3977122009-01-05 21:25:51 -05001259static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1260 struct bio *bio, int mirror_num,
1261 unsigned long bio_flags)
Chris Mason065631f2008-02-20 12:07:25 -05001262{
Chris Mason065631f2008-02-20 12:07:25 -05001263 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason065631f2008-02-20 12:07:25 -05001264 int ret = 0;
Chris Masone0156402008-04-16 11:15:20 -04001265
Chris Masond20f7042008-12-08 16:58:54 -05001266 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
Chris Mason44b8bd72008-04-16 11:14:51 -04001267 BUG_ON(ret);
Chris Mason4a69a412008-11-06 22:03:00 -05001268 return 0;
1269}
Chris Masone0156402008-04-16 11:15:20 -04001270
Chris Mason4a69a412008-11-06 22:03:00 -05001271/*
1272 * in order to insert checksums into the metadata in large chunks,
1273 * we wait until bio submission time. All the pages in the bio are
1274 * checksummed and sums are attached onto the ordered extent record.
1275 *
1276 * At IO completion time the cums attached on the ordered extent record
1277 * are inserted into the btree
1278 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001279static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
Chris Mason4a69a412008-11-06 22:03:00 -05001280 int mirror_num, unsigned long bio_flags)
1281{
1282 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason8b712842008-06-11 16:50:36 -04001283 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
Chris Mason44b8bd72008-04-16 11:14:51 -04001284}
1285
Chris Masond352ac62008-09-29 15:18:18 -04001286/*
Chris Masoncad321a2008-12-17 14:51:42 -05001287 * extent_io.c submission hook. This does the right thing for csum calculation
1288 * on write, or reading the csums from the tree before a read
Chris Masond352ac62008-09-29 15:18:18 -04001289 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001290static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
Chris Masonc8b97812008-10-29 14:49:59 -04001291 int mirror_num, unsigned long bio_flags)
Chris Mason44b8bd72008-04-16 11:14:51 -04001292{
1293 struct btrfs_root *root = BTRFS_I(inode)->root;
1294 int ret = 0;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001295 int skip_sum;
Chris Mason44b8bd72008-04-16 11:14:51 -04001296
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001297 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masoncad321a2008-12-17 14:51:42 -05001298
Chris Masone6dcd2d2008-07-17 12:53:50 -04001299 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1300 BUG_ON(ret);
Chris Mason065631f2008-02-20 12:07:25 -05001301
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001302 if (!(rw & (1 << BIO_RW))) {
Chris Masond20f7042008-12-08 16:58:54 -05001303 if (bio_flags & EXTENT_BIO_COMPRESSED) {
Chris Masonc8b97812008-10-29 14:49:59 -04001304 return btrfs_submit_compressed_read(inode, bio,
1305 mirror_num, bio_flags);
Chris Masond20f7042008-12-08 16:58:54 -05001306 } else if (!skip_sum)
1307 btrfs_lookup_bio_sums(root, inode, bio, NULL);
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001308 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001309 } else if (!skip_sum) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001310 /* csum items have already been cloned */
1311 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1312 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001313 /* we're doing a write, do the async checksumming */
1314 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
Chris Mason44b8bd72008-04-16 11:14:51 -04001315 inode, rw, bio, mirror_num,
Chris Mason4a69a412008-11-06 22:03:00 -05001316 bio_flags, __btrfs_submit_bio_start,
1317 __btrfs_submit_bio_done);
Chris Mason19b9bdb2008-10-30 14:23:13 -04001318 }
1319
Chris Mason0b86a832008-03-24 15:01:56 -04001320mapit:
Chris Mason8b712842008-06-11 16:50:36 -04001321 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
Chris Mason065631f2008-02-20 12:07:25 -05001322}
Chris Mason6885f302008-02-20 16:11:05 -05001323
Chris Masond352ac62008-09-29 15:18:18 -04001324/*
1325 * given a list of ordered sums record them in the inode. This happens
1326 * at IO completion time based on sums calculated at bio submission time.
1327 */
Chris Masonba1da2f2008-07-17 12:54:15 -04001328static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
Chris Masone6dcd2d2008-07-17 12:53:50 -04001329 struct inode *inode, u64 file_offset,
1330 struct list_head *list)
1331{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001332 struct btrfs_ordered_sum *sum;
1333
1334 btrfs_set_trans_block_group(trans, inode);
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001335
1336 list_for_each_entry(sum, list, list) {
Chris Masond20f7042008-12-08 16:58:54 -05001337 btrfs_csum_file_blocks(trans,
1338 BTRFS_I(inode)->root->fs_info->csum_root, sum);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001339 }
1340 return 0;
1341}
1342
Chris Masonea8c2812008-08-04 23:17:27 -04001343int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1344{
Chris Masond3977122009-01-05 21:25:51 -05001345 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
Chris Mason771ed682008-11-06 22:02:51 -05001346 WARN_ON(1);
Chris Masonea8c2812008-08-04 23:17:27 -04001347 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1348 GFP_NOFS);
1349}
1350
Chris Masond352ac62008-09-29 15:18:18 -04001351/* see btrfs_writepage_start_hook for details on why this is required */
Chris Mason247e7432008-07-17 12:53:51 -04001352struct btrfs_writepage_fixup {
1353 struct page *page;
1354 struct btrfs_work work;
1355};
1356
Christoph Hellwigb2950862008-12-02 09:54:17 -05001357static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
Chris Mason247e7432008-07-17 12:53:51 -04001358{
1359 struct btrfs_writepage_fixup *fixup;
1360 struct btrfs_ordered_extent *ordered;
1361 struct page *page;
1362 struct inode *inode;
1363 u64 page_start;
1364 u64 page_end;
1365
1366 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1367 page = fixup->page;
Chris Mason4a096752008-07-21 10:29:44 -04001368again:
Chris Mason247e7432008-07-17 12:53:51 -04001369 lock_page(page);
1370 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1371 ClearPageChecked(page);
1372 goto out_page;
1373 }
1374
1375 inode = page->mapping->host;
1376 page_start = page_offset(page);
1377 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1378
1379 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001380
1381 /* already ordered? We're done */
Chris Mason8b62b722009-09-02 16:53:46 -04001382 if (PagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001383 goto out;
Chris Mason4a096752008-07-21 10:29:44 -04001384
1385 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1386 if (ordered) {
1387 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1388 page_end, GFP_NOFS);
1389 unlock_page(page);
1390 btrfs_start_ordered_extent(inode, ordered, 1);
1391 goto again;
1392 }
Chris Mason247e7432008-07-17 12:53:51 -04001393
Chris Masonea8c2812008-08-04 23:17:27 -04001394 btrfs_set_extent_delalloc(inode, page_start, page_end);
Chris Mason247e7432008-07-17 12:53:51 -04001395 ClearPageChecked(page);
1396out:
1397 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1398out_page:
1399 unlock_page(page);
1400 page_cache_release(page);
1401}
1402
1403/*
1404 * There are a few paths in the higher layers of the kernel that directly
1405 * set the page dirty bit without asking the filesystem if it is a
1406 * good idea. This causes problems because we want to make sure COW
1407 * properly happens and the data=ordered rules are followed.
1408 *
Chris Masonc8b97812008-10-29 14:49:59 -04001409 * In our case any range that doesn't have the ORDERED bit set
Chris Mason247e7432008-07-17 12:53:51 -04001410 * hasn't been properly setup for IO. We kick off an async process
1411 * to fix it up. The async helper will wait for ordered extents, set
1412 * the delalloc bit and make it safe to write the page.
1413 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001414static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
Chris Mason247e7432008-07-17 12:53:51 -04001415{
1416 struct inode *inode = page->mapping->host;
1417 struct btrfs_writepage_fixup *fixup;
1418 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason247e7432008-07-17 12:53:51 -04001419
Chris Mason8b62b722009-09-02 16:53:46 -04001420 /* this page is properly in the ordered list */
1421 if (TestClearPagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001422 return 0;
1423
1424 if (PageChecked(page))
1425 return -EAGAIN;
1426
1427 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1428 if (!fixup)
1429 return -EAGAIN;
Chris Masonf4219502008-07-22 11:18:09 -04001430
Chris Mason247e7432008-07-17 12:53:51 -04001431 SetPageChecked(page);
1432 page_cache_get(page);
1433 fixup->work.func = btrfs_writepage_fixup_worker;
1434 fixup->page = page;
1435 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1436 return -EAGAIN;
1437}
1438
Yan Zhengd899e052008-10-30 14:25:28 -04001439static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1440 struct inode *inode, u64 file_pos,
1441 u64 disk_bytenr, u64 disk_num_bytes,
1442 u64 num_bytes, u64 ram_bytes,
Chris Masone980b502009-04-24 14:39:24 -04001443 u64 locked_end,
Yan Zhengd899e052008-10-30 14:25:28 -04001444 u8 compression, u8 encryption,
1445 u16 other_encoding, int extent_type)
1446{
1447 struct btrfs_root *root = BTRFS_I(inode)->root;
1448 struct btrfs_file_extent_item *fi;
1449 struct btrfs_path *path;
1450 struct extent_buffer *leaf;
1451 struct btrfs_key ins;
1452 u64 hint;
1453 int ret;
1454
1455 path = btrfs_alloc_path();
1456 BUG_ON(!path);
1457
Chris Masonb9473432009-03-13 11:00:37 -04001458 path->leave_spinning = 1;
Chris Masona1ed8352009-09-11 12:27:37 -04001459
1460 /*
1461 * we may be replacing one extent in the tree with another.
1462 * The new extent is pinned in the extent map, and we don't want
1463 * to drop it from the cache until it is completely in the btree.
1464 *
1465 * So, tell btrfs_drop_extents to leave this extent in the cache.
1466 * the caller is expected to unpin it and allow it to be merged
1467 * with the others.
1468 */
Yan Zhengd899e052008-10-30 14:25:28 -04001469 ret = btrfs_drop_extents(trans, root, inode, file_pos,
Chris Masone980b502009-04-24 14:39:24 -04001470 file_pos + num_bytes, locked_end,
Chris Masona1ed8352009-09-11 12:27:37 -04001471 file_pos, &hint, 0);
Yan Zhengd899e052008-10-30 14:25:28 -04001472 BUG_ON(ret);
1473
1474 ins.objectid = inode->i_ino;
1475 ins.offset = file_pos;
1476 ins.type = BTRFS_EXTENT_DATA_KEY;
1477 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1478 BUG_ON(ret);
1479 leaf = path->nodes[0];
1480 fi = btrfs_item_ptr(leaf, path->slots[0],
1481 struct btrfs_file_extent_item);
1482 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1483 btrfs_set_file_extent_type(leaf, fi, extent_type);
1484 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1485 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1486 btrfs_set_file_extent_offset(leaf, fi, 0);
1487 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1488 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1489 btrfs_set_file_extent_compression(leaf, fi, compression);
1490 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1491 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
Chris Masonb9473432009-03-13 11:00:37 -04001492
1493 btrfs_unlock_up_safe(path, 1);
1494 btrfs_set_lock_blocking(leaf);
1495
Yan Zhengd899e052008-10-30 14:25:28 -04001496 btrfs_mark_buffer_dirty(leaf);
1497
1498 inode_add_bytes(inode, num_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -04001499
1500 ins.objectid = disk_bytenr;
1501 ins.offset = disk_num_bytes;
1502 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001503 ret = btrfs_alloc_reserved_file_extent(trans, root,
1504 root->root_key.objectid,
1505 inode->i_ino, file_pos, &ins);
Yan Zhengd899e052008-10-30 14:25:28 -04001506 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -04001507 btrfs_free_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04001508
Yan Zhengd899e052008-10-30 14:25:28 -04001509 return 0;
1510}
1511
Chris Mason5d13a982009-03-13 11:41:46 -04001512/*
1513 * helper function for btrfs_finish_ordered_io, this
1514 * just reads in some of the csum leaves to prime them into ram
1515 * before we start the transaction. It limits the amount of btree
1516 * reads required while inside the transaction.
1517 */
1518static noinline void reada_csum(struct btrfs_root *root,
1519 struct btrfs_path *path,
1520 struct btrfs_ordered_extent *ordered_extent)
1521{
1522 struct btrfs_ordered_sum *sum;
1523 u64 bytenr;
1524
1525 sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1526 list);
1527 bytenr = sum->sums[0].bytenr;
1528
1529 /*
1530 * we don't care about the results, the point of this search is
1531 * just to get the btree leaves into ram
1532 */
1533 btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1534}
1535
Chris Masond352ac62008-09-29 15:18:18 -04001536/* as ordered data IO finishes, this gets called so we can finish
1537 * an ordered extent if the range of bytes in the file it covers are
1538 * fully written.
1539 */
Chris Mason211f90e2008-07-18 11:56:15 -04001540static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001541{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001542 struct btrfs_root *root = BTRFS_I(inode)->root;
1543 struct btrfs_trans_handle *trans;
Chris Mason5d13a982009-03-13 11:41:46 -04001544 struct btrfs_ordered_extent *ordered_extent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001545 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masonb7ec40d2009-03-12 20:12:45 -04001546 struct btrfs_path *path;
Yan Zhengd899e052008-10-30 14:25:28 -04001547 int compressed = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001548 int ret;
1549
1550 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
Chris Masonba1da2f2008-07-17 12:54:15 -04001551 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001552 return 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001553
Chris Masonb7ec40d2009-03-12 20:12:45 -04001554 /*
1555 * before we join the transaction, try to do some of our IO.
1556 * This will limit the amount of IO that we have to do with
1557 * the transaction running. We're unlikely to need to do any
1558 * IO if the file extents are new, the disk_i_size checks
1559 * covers the most common case.
1560 */
1561 if (start < BTRFS_I(inode)->disk_i_size) {
1562 path = btrfs_alloc_path();
1563 if (path) {
1564 ret = btrfs_lookup_file_extent(NULL, root, path,
1565 inode->i_ino,
1566 start, 0);
Chris Mason5d13a982009-03-13 11:41:46 -04001567 ordered_extent = btrfs_lookup_ordered_extent(inode,
1568 start);
1569 if (!list_empty(&ordered_extent->list)) {
1570 btrfs_release_path(root, path);
1571 reada_csum(root, path, ordered_extent);
1572 }
Chris Masonb7ec40d2009-03-12 20:12:45 -04001573 btrfs_free_path(path);
1574 }
1575 }
1576
Chris Masonf9295742008-07-17 12:54:14 -04001577 trans = btrfs_join_transaction(root, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001578
Chris Mason5d13a982009-03-13 11:41:46 -04001579 if (!ordered_extent)
1580 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001581 BUG_ON(!ordered_extent);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001582 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1583 goto nocow;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001584
1585 lock_extent(io_tree, ordered_extent->file_offset,
1586 ordered_extent->file_offset + ordered_extent->len - 1,
1587 GFP_NOFS);
1588
Chris Masonc8b97812008-10-29 14:49:59 -04001589 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
Yan Zhengd899e052008-10-30 14:25:28 -04001590 compressed = 1;
1591 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1592 BUG_ON(compressed);
1593 ret = btrfs_mark_extent_written(trans, root, inode,
1594 ordered_extent->file_offset,
1595 ordered_extent->file_offset +
1596 ordered_extent->len);
1597 BUG_ON(ret);
1598 } else {
1599 ret = insert_reserved_file_extent(trans, inode,
1600 ordered_extent->file_offset,
1601 ordered_extent->start,
1602 ordered_extent->disk_len,
1603 ordered_extent->len,
1604 ordered_extent->len,
Chris Masone980b502009-04-24 14:39:24 -04001605 ordered_extent->file_offset +
1606 ordered_extent->len,
Yan Zhengd899e052008-10-30 14:25:28 -04001607 compressed, 0, 0,
1608 BTRFS_FILE_EXTENT_REG);
Chris Masona1ed8352009-09-11 12:27:37 -04001609 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1610 ordered_extent->file_offset,
1611 ordered_extent->len);
Yan Zhengd899e052008-10-30 14:25:28 -04001612 BUG_ON(ret);
1613 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04001614 unlock_extent(io_tree, ordered_extent->file_offset,
1615 ordered_extent->file_offset + ordered_extent->len - 1,
1616 GFP_NOFS);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001617nocow:
Chris Masone6dcd2d2008-07-17 12:53:50 -04001618 add_pending_csums(trans, inode, ordered_extent->file_offset,
1619 &ordered_extent->list);
1620
Chris Mason34353022008-09-23 20:19:49 -04001621 mutex_lock(&BTRFS_I(inode)->extent_mutex);
Chris Masondbe674a2008-07-17 12:54:05 -04001622 btrfs_ordered_update_i_size(inode, ordered_extent);
Chris Masone02119d2008-09-05 16:13:11 -04001623 btrfs_update_inode(trans, root, inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001624 btrfs_remove_ordered_extent(inode, ordered_extent);
Chris Mason34353022008-09-23 20:19:49 -04001625 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
Chris Mason7f3c74f2008-07-18 12:01:11 -04001626
Chris Masone6dcd2d2008-07-17 12:53:50 -04001627 /* once for us */
1628 btrfs_put_ordered_extent(ordered_extent);
1629 /* once for the tree */
1630 btrfs_put_ordered_extent(ordered_extent);
1631
Chris Masone6dcd2d2008-07-17 12:53:50 -04001632 btrfs_end_transaction(trans, root);
1633 return 0;
1634}
1635
Christoph Hellwigb2950862008-12-02 09:54:17 -05001636static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason211f90e2008-07-18 11:56:15 -04001637 struct extent_state *state, int uptodate)
1638{
Chris Mason8b62b722009-09-02 16:53:46 -04001639 ClearPagePrivate2(page);
Chris Mason211f90e2008-07-18 11:56:15 -04001640 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1641}
1642
Chris Masond352ac62008-09-29 15:18:18 -04001643/*
1644 * When IO fails, either with EIO or csum verification fails, we
1645 * try other mirrors that might have a good copy of the data. This
1646 * io_failure_record is used to record state as we go through all the
1647 * mirrors. If another mirror has good data, the page is set up to date
1648 * and things continue. If a good mirror can't be found, the original
1649 * bio end_io callback is called to indicate things have failed.
1650 */
Chris Mason7e383262008-04-09 16:28:12 -04001651struct io_failure_record {
1652 struct page *page;
1653 u64 start;
1654 u64 len;
1655 u64 logical;
Chris Masond20f7042008-12-08 16:58:54 -05001656 unsigned long bio_flags;
Chris Mason7e383262008-04-09 16:28:12 -04001657 int last_mirror;
1658};
1659
Christoph Hellwigb2950862008-12-02 09:54:17 -05001660static int btrfs_io_failed_hook(struct bio *failed_bio,
Chris Mason1259ab72008-05-12 13:39:03 -04001661 struct page *page, u64 start, u64 end,
1662 struct extent_state *state)
Chris Mason7e383262008-04-09 16:28:12 -04001663{
1664 struct io_failure_record *failrec = NULL;
1665 u64 private;
1666 struct extent_map *em;
1667 struct inode *inode = page->mapping->host;
1668 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
Chris Mason3b951512008-04-17 11:29:12 -04001669 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Mason7e383262008-04-09 16:28:12 -04001670 struct bio *bio;
1671 int num_copies;
1672 int ret;
Chris Mason1259ab72008-05-12 13:39:03 -04001673 int rw;
Chris Mason7e383262008-04-09 16:28:12 -04001674 u64 logical;
1675
1676 ret = get_state_private(failure_tree, start, &private);
1677 if (ret) {
Chris Mason7e383262008-04-09 16:28:12 -04001678 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1679 if (!failrec)
1680 return -ENOMEM;
1681 failrec->start = start;
1682 failrec->len = end - start + 1;
1683 failrec->last_mirror = 0;
Chris Masond20f7042008-12-08 16:58:54 -05001684 failrec->bio_flags = 0;
Chris Mason7e383262008-04-09 16:28:12 -04001685
Chris Mason890871b2009-09-02 16:24:52 -04001686 read_lock(&em_tree->lock);
Chris Mason3b951512008-04-17 11:29:12 -04001687 em = lookup_extent_mapping(em_tree, start, failrec->len);
1688 if (em->start > start || em->start + em->len < start) {
1689 free_extent_map(em);
1690 em = NULL;
1691 }
Chris Mason890871b2009-09-02 16:24:52 -04001692 read_unlock(&em_tree->lock);
Chris Mason7e383262008-04-09 16:28:12 -04001693
1694 if (!em || IS_ERR(em)) {
1695 kfree(failrec);
1696 return -EIO;
1697 }
1698 logical = start - em->start;
1699 logical = em->block_start + logical;
Chris Masond20f7042008-12-08 16:58:54 -05001700 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1701 logical = em->block_start;
1702 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1703 }
Chris Mason7e383262008-04-09 16:28:12 -04001704 failrec->logical = logical;
1705 free_extent_map(em);
1706 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1707 EXTENT_DIRTY, GFP_NOFS);
Chris Mason587f7702008-04-11 12:16:46 -04001708 set_state_private(failure_tree, start,
1709 (u64)(unsigned long)failrec);
Chris Mason7e383262008-04-09 16:28:12 -04001710 } else {
Chris Mason587f7702008-04-11 12:16:46 -04001711 failrec = (struct io_failure_record *)(unsigned long)private;
Chris Mason7e383262008-04-09 16:28:12 -04001712 }
1713 num_copies = btrfs_num_copies(
1714 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1715 failrec->logical, failrec->len);
1716 failrec->last_mirror++;
1717 if (!state) {
Chris Masoncad321a2008-12-17 14:51:42 -05001718 spin_lock(&BTRFS_I(inode)->io_tree.lock);
Chris Mason7e383262008-04-09 16:28:12 -04001719 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1720 failrec->start,
1721 EXTENT_LOCKED);
1722 if (state && state->start != failrec->start)
1723 state = NULL;
Chris Masoncad321a2008-12-17 14:51:42 -05001724 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
Chris Mason7e383262008-04-09 16:28:12 -04001725 }
1726 if (!state || failrec->last_mirror > num_copies) {
1727 set_state_private(failure_tree, failrec->start, 0);
1728 clear_extent_bits(failure_tree, failrec->start,
1729 failrec->start + failrec->len - 1,
1730 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1731 kfree(failrec);
1732 return -EIO;
1733 }
1734 bio = bio_alloc(GFP_NOFS, 1);
1735 bio->bi_private = state;
1736 bio->bi_end_io = failed_bio->bi_end_io;
1737 bio->bi_sector = failrec->logical >> 9;
1738 bio->bi_bdev = failed_bio->bi_bdev;
Chris Masone1c4b742008-04-22 13:26:46 -04001739 bio->bi_size = 0;
Chris Masond20f7042008-12-08 16:58:54 -05001740
Chris Mason7e383262008-04-09 16:28:12 -04001741 bio_add_page(bio, page, failrec->len, start - page_offset(page));
Chris Mason1259ab72008-05-12 13:39:03 -04001742 if (failed_bio->bi_rw & (1 << BIO_RW))
1743 rw = WRITE;
1744 else
1745 rw = READ;
1746
1747 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
Chris Masonc8b97812008-10-29 14:49:59 -04001748 failrec->last_mirror,
Chris Masond20f7042008-12-08 16:58:54 -05001749 failrec->bio_flags);
Chris Mason1259ab72008-05-12 13:39:03 -04001750 return 0;
1751}
1752
Chris Masond352ac62008-09-29 15:18:18 -04001753/*
1754 * each time an IO finishes, we do a fast check in the IO failure tree
1755 * to see if we need to process or clean up an io_failure_record
1756 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001757static int btrfs_clean_io_failures(struct inode *inode, u64 start)
Chris Mason1259ab72008-05-12 13:39:03 -04001758{
1759 u64 private;
1760 u64 private_failure;
1761 struct io_failure_record *failure;
1762 int ret;
1763
1764 private = 0;
1765 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1766 (u64)-1, 1, EXTENT_DIRTY)) {
1767 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1768 start, &private_failure);
1769 if (ret == 0) {
1770 failure = (struct io_failure_record *)(unsigned long)
1771 private_failure;
1772 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1773 failure->start, 0);
1774 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1775 failure->start,
1776 failure->start + failure->len - 1,
1777 EXTENT_DIRTY | EXTENT_LOCKED,
1778 GFP_NOFS);
1779 kfree(failure);
1780 }
1781 }
Chris Mason7e383262008-04-09 16:28:12 -04001782 return 0;
1783}
1784
Chris Masond352ac62008-09-29 15:18:18 -04001785/*
1786 * when reads are done, we need to check csums to verify the data is correct
1787 * if there's a match, we allow the bio to finish. If not, we go through
1788 * the io_failure_record routines to find good copies
1789 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001790static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason70dec802008-01-29 09:59:12 -05001791 struct extent_state *state)
Chris Mason07157aa2007-08-30 08:50:51 -04001792{
Chris Mason35ebb932007-10-30 16:56:53 -04001793 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
Chris Mason07157aa2007-08-30 08:50:51 -04001794 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05001795 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason07157aa2007-08-30 08:50:51 -04001796 char *kaddr;
Chris Masonaadfeb62008-01-29 09:10:27 -05001797 u64 private = ~(u32)0;
Chris Mason07157aa2007-08-30 08:50:51 -04001798 int ret;
Chris Masonff79f812007-10-15 16:22:25 -04001799 struct btrfs_root *root = BTRFS_I(inode)->root;
1800 u32 csum = ~(u32)0;
Chris Masond1310b22008-01-24 16:13:08 -05001801
Chris Masond20f7042008-12-08 16:58:54 -05001802 if (PageChecked(page)) {
1803 ClearPageChecked(page);
1804 goto good;
1805 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001806
1807 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Chris Masonb6cda9b2007-12-14 15:30:32 -05001808 return 0;
Chris Masond20f7042008-12-08 16:58:54 -05001809
Yan Zheng17d217f2008-12-12 10:03:38 -05001810 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
Chris Mason9655d292009-09-02 15:22:30 -04001811 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001812 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1813 GFP_NOFS);
1814 return 0;
1815 }
1816
Yanc2e639f2008-02-04 08:57:25 -05001817 if (state && state->start == start) {
Chris Mason70dec802008-01-29 09:59:12 -05001818 private = state->private;
1819 ret = 0;
1820 } else {
1821 ret = get_state_private(io_tree, start, &private);
1822 }
Chris Mason9ab86c82009-01-07 09:48:51 -05001823 kaddr = kmap_atomic(page, KM_USER0);
Chris Masond3977122009-01-05 21:25:51 -05001824 if (ret)
Chris Mason07157aa2007-08-30 08:50:51 -04001825 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05001826
Chris Masonff79f812007-10-15 16:22:25 -04001827 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1828 btrfs_csum_final(csum, (char *)&csum);
Chris Masond3977122009-01-05 21:25:51 -05001829 if (csum != private)
Chris Mason07157aa2007-08-30 08:50:51 -04001830 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05001831
Chris Mason9ab86c82009-01-07 09:48:51 -05001832 kunmap_atomic(kaddr, KM_USER0);
Chris Masond20f7042008-12-08 16:58:54 -05001833good:
Chris Mason7e383262008-04-09 16:28:12 -04001834 /* if the io failure tree for this inode is non-empty,
1835 * check to see if we've recovered from a failed IO
1836 */
Chris Mason1259ab72008-05-12 13:39:03 -04001837 btrfs_clean_io_failures(inode, start);
Chris Mason07157aa2007-08-30 08:50:51 -04001838 return 0;
1839
1840zeroit:
Chris Mason193f2842009-04-27 07:29:05 -04001841 if (printk_ratelimit()) {
1842 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1843 "private %llu\n", page->mapping->host->i_ino,
1844 (unsigned long long)start, csum,
1845 (unsigned long long)private);
1846 }
Chris Masondb945352007-10-15 16:15:53 -04001847 memset(kaddr + offset, 1, end - start + 1);
1848 flush_dcache_page(page);
Chris Mason9ab86c82009-01-07 09:48:51 -05001849 kunmap_atomic(kaddr, KM_USER0);
Chris Mason3b951512008-04-17 11:29:12 -04001850 if (private == 0)
1851 return 0;
Chris Mason7e383262008-04-09 16:28:12 -04001852 return -EIO;
Chris Mason07157aa2007-08-30 08:50:51 -04001853}
Chris Masonb888db22007-08-27 16:49:44 -04001854
Josef Bacik7b128762008-07-24 12:17:14 -04001855/*
1856 * This creates an orphan entry for the given inode in case something goes
1857 * wrong in the middle of an unlink/truncate.
1858 */
1859int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1860{
1861 struct btrfs_root *root = BTRFS_I(inode)->root;
1862 int ret = 0;
1863
Yanbcc63ab2008-07-30 16:29:20 -04001864 spin_lock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04001865
1866 /* already on the orphan list, we're good */
1867 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
Yanbcc63ab2008-07-30 16:29:20 -04001868 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04001869 return 0;
1870 }
1871
1872 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1873
Yanbcc63ab2008-07-30 16:29:20 -04001874 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04001875
1876 /*
1877 * insert an orphan item to track this unlinked/truncated file
1878 */
1879 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1880
1881 return ret;
1882}
1883
1884/*
1885 * We have done the truncate/delete so we can go ahead and remove the orphan
1886 * item for this particular inode.
1887 */
1888int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1889{
1890 struct btrfs_root *root = BTRFS_I(inode)->root;
1891 int ret = 0;
1892
Yanbcc63ab2008-07-30 16:29:20 -04001893 spin_lock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04001894
1895 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
Yanbcc63ab2008-07-30 16:29:20 -04001896 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04001897 return 0;
1898 }
1899
1900 list_del_init(&BTRFS_I(inode)->i_orphan);
1901 if (!trans) {
Yanbcc63ab2008-07-30 16:29:20 -04001902 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04001903 return 0;
1904 }
1905
Yanbcc63ab2008-07-30 16:29:20 -04001906 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04001907
1908 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1909
1910 return ret;
1911}
1912
1913/*
1914 * this cleans up any orphans that may be left on the list from the last use
1915 * of this root.
1916 */
1917void btrfs_orphan_cleanup(struct btrfs_root *root)
1918{
1919 struct btrfs_path *path;
1920 struct extent_buffer *leaf;
1921 struct btrfs_item *item;
1922 struct btrfs_key key, found_key;
1923 struct btrfs_trans_handle *trans;
1924 struct inode *inode;
1925 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1926
Josef Bacik7b128762008-07-24 12:17:14 -04001927 path = btrfs_alloc_path();
1928 if (!path)
1929 return;
1930 path->reada = -1;
1931
1932 key.objectid = BTRFS_ORPHAN_OBJECTID;
1933 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1934 key.offset = (u64)-1;
1935
Josef Bacik7b128762008-07-24 12:17:14 -04001936
1937 while (1) {
1938 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1939 if (ret < 0) {
1940 printk(KERN_ERR "Error searching slot for orphan: %d"
1941 "\n", ret);
1942 break;
1943 }
1944
1945 /*
1946 * if ret == 0 means we found what we were searching for, which
1947 * is weird, but possible, so only screw with path if we didnt
1948 * find the key and see if we have stuff that matches
1949 */
1950 if (ret > 0) {
1951 if (path->slots[0] == 0)
1952 break;
1953 path->slots[0]--;
1954 }
1955
1956 /* pull out the item */
1957 leaf = path->nodes[0];
1958 item = btrfs_item_nr(leaf, path->slots[0]);
1959 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1960
1961 /* make sure the item matches what we want */
1962 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1963 break;
1964 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1965 break;
1966
1967 /* release the path since we're done with it */
1968 btrfs_release_path(root, path);
1969
1970 /*
1971 * this is where we are basically btrfs_lookup, without the
1972 * crossing root thing. we store the inode number in the
1973 * offset of the orphan item.
1974 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001975 found_key.objectid = found_key.offset;
1976 found_key.type = BTRFS_INODE_ITEM_KEY;
1977 found_key.offset = 0;
1978 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
1979 if (IS_ERR(inode))
Josef Bacik7b128762008-07-24 12:17:14 -04001980 break;
1981
Josef Bacik7b128762008-07-24 12:17:14 -04001982 /*
1983 * add this inode to the orphan list so btrfs_orphan_del does
1984 * the proper thing when we hit it
1985 */
Yanbcc63ab2008-07-30 16:29:20 -04001986 spin_lock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04001987 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
Yanbcc63ab2008-07-30 16:29:20 -04001988 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04001989
1990 /*
1991 * if this is a bad inode, means we actually succeeded in
1992 * removing the inode, but not the orphan record, which means
1993 * we need to manually delete the orphan since iput will just
1994 * do a destroy_inode
1995 */
1996 if (is_bad_inode(inode)) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -04001997 trans = btrfs_start_transaction(root, 1);
Josef Bacik7b128762008-07-24 12:17:14 -04001998 btrfs_orphan_del(trans, inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04001999 btrfs_end_transaction(trans, root);
Josef Bacik7b128762008-07-24 12:17:14 -04002000 iput(inode);
2001 continue;
2002 }
2003
2004 /* if we have links, this was a truncate, lets do that */
2005 if (inode->i_nlink) {
2006 nr_truncate++;
2007 btrfs_truncate(inode);
2008 } else {
2009 nr_unlink++;
2010 }
2011
2012 /* this will do delete_inode and everything for us */
2013 iput(inode);
2014 }
2015
2016 if (nr_unlink)
2017 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2018 if (nr_truncate)
2019 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2020
2021 btrfs_free_path(path);
Josef Bacik7b128762008-07-24 12:17:14 -04002022}
2023
Chris Masond352ac62008-09-29 15:18:18 -04002024/*
Chris Mason46a53cc2009-04-27 11:47:50 -04002025 * very simple check to peek ahead in the leaf looking for xattrs. If we
2026 * don't find any xattrs, we know there can't be any acls.
2027 *
2028 * slot is the slot the inode is in, objectid is the objectid of the inode
2029 */
2030static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2031 int slot, u64 objectid)
2032{
2033 u32 nritems = btrfs_header_nritems(leaf);
2034 struct btrfs_key found_key;
2035 int scanned = 0;
2036
2037 slot++;
2038 while (slot < nritems) {
2039 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2040
2041 /* we found a different objectid, there must not be acls */
2042 if (found_key.objectid != objectid)
2043 return 0;
2044
2045 /* we found an xattr, assume we've got an acl */
2046 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2047 return 1;
2048
2049 /*
2050 * we found a key greater than an xattr key, there can't
2051 * be any acls later on
2052 */
2053 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2054 return 0;
2055
2056 slot++;
2057 scanned++;
2058
2059 /*
2060 * it goes inode, inode backrefs, xattrs, extents,
2061 * so if there are a ton of hard links to an inode there can
2062 * be a lot of backrefs. Don't waste time searching too hard,
2063 * this is just an optimization
2064 */
2065 if (scanned >= 8)
2066 break;
2067 }
2068 /* we hit the end of the leaf before we found an xattr or
2069 * something larger than an xattr. We have to assume the inode
2070 * has acls
2071 */
2072 return 1;
2073}
2074
2075/*
Chris Masond352ac62008-09-29 15:18:18 -04002076 * read an inode from the btree into the in-memory inode
2077 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002078static void btrfs_read_locked_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002079{
2080 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002081 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002082 struct btrfs_inode_item *inode_item;
Chris Mason0b86a832008-03-24 15:01:56 -04002083 struct btrfs_timespec *tspec;
Chris Mason39279cc2007-06-12 06:35:45 -04002084 struct btrfs_root *root = BTRFS_I(inode)->root;
2085 struct btrfs_key location;
Chris Mason46a53cc2009-04-27 11:47:50 -04002086 int maybe_acls;
Chris Mason39279cc2007-06-12 06:35:45 -04002087 u64 alloc_group_block;
Josef Bacik618e21d2007-07-11 10:18:17 -04002088 u32 rdev;
Chris Mason39279cc2007-06-12 06:35:45 -04002089 int ret;
2090
2091 path = btrfs_alloc_path();
2092 BUG_ON(!path);
Chris Mason39279cc2007-06-12 06:35:45 -04002093 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
Chris Masondc17ff82008-01-08 15:46:30 -05002094
Chris Mason39279cc2007-06-12 06:35:45 -04002095 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04002096 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04002097 goto make_bad;
Chris Mason39279cc2007-06-12 06:35:45 -04002098
Chris Mason5f39d392007-10-15 16:14:19 -04002099 leaf = path->nodes[0];
2100 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2101 struct btrfs_inode_item);
2102
2103 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2104 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2105 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2106 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
Chris Masondbe674a2008-07-17 12:54:05 -04002107 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04002108
2109 tspec = btrfs_inode_atime(inode_item);
2110 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2111 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2112
2113 tspec = btrfs_inode_mtime(inode_item);
2114 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2115 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2116
2117 tspec = btrfs_inode_ctime(inode_item);
2118 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2119 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2120
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002121 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04002122 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
Chris Masonc3027eb2008-12-08 16:40:21 -05002123 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04002124 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik618e21d2007-07-11 10:18:17 -04002125 inode->i_rdev = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002126 rdev = btrfs_inode_rdev(leaf, inode_item);
2127
Josef Bacikaec74772008-07-24 12:12:38 -04002128 BTRFS_I(inode)->index_cnt = (u64)-1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05002129 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
Josef Bacikaec74772008-07-24 12:12:38 -04002130
Chris Mason5f39d392007-10-15 16:14:19 -04002131 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
Chris Masonb4ce94d2009-02-04 09:25:08 -05002132
Chris Mason46a53cc2009-04-27 11:47:50 -04002133 /*
2134 * try to precache a NULL acl entry for files that don't have
2135 * any xattrs or acls
2136 */
2137 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2138 if (!maybe_acls) {
2139 BTRFS_I(inode)->i_acl = NULL;
2140 BTRFS_I(inode)->i_default_acl = NULL;
2141 }
2142
Yan Zhengd2fb3432008-12-11 16:30:39 -05002143 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2144 alloc_group_block, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04002145 btrfs_free_path(path);
2146 inode_item = NULL;
2147
Chris Mason39279cc2007-06-12 06:35:45 -04002148 switch (inode->i_mode & S_IFMT) {
Chris Mason39279cc2007-06-12 06:35:45 -04002149 case S_IFREG:
2150 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04002151 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05002152 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04002153 inode->i_fop = &btrfs_file_operations;
2154 inode->i_op = &btrfs_file_inode_operations;
2155 break;
2156 case S_IFDIR:
2157 inode->i_fop = &btrfs_dir_file_operations;
2158 if (root == root->fs_info->tree_root)
2159 inode->i_op = &btrfs_dir_ro_inode_operations;
2160 else
2161 inode->i_op = &btrfs_dir_inode_operations;
2162 break;
2163 case S_IFLNK:
2164 inode->i_op = &btrfs_symlink_inode_operations;
2165 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04002166 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04002167 break;
Josef Bacik618e21d2007-07-11 10:18:17 -04002168 default:
Jim Owens0279b4c2009-02-04 09:29:13 -05002169 inode->i_op = &btrfs_special_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -04002170 init_special_inode(inode, inode->i_mode, rdev);
2171 break;
Chris Mason39279cc2007-06-12 06:35:45 -04002172 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02002173
2174 btrfs_update_iflags(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002175 return;
2176
2177make_bad:
Chris Mason39279cc2007-06-12 06:35:45 -04002178 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002179 make_bad_inode(inode);
2180}
2181
Chris Masond352ac62008-09-29 15:18:18 -04002182/*
2183 * given a leaf and an inode, copy the inode fields into the leaf
2184 */
Chris Masone02119d2008-09-05 16:13:11 -04002185static void fill_inode_item(struct btrfs_trans_handle *trans,
2186 struct extent_buffer *leaf,
Chris Mason5f39d392007-10-15 16:14:19 -04002187 struct btrfs_inode_item *item,
Chris Mason39279cc2007-06-12 06:35:45 -04002188 struct inode *inode)
2189{
Chris Mason5f39d392007-10-15 16:14:19 -04002190 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2191 btrfs_set_inode_gid(leaf, item, inode->i_gid);
Chris Masondbe674a2008-07-17 12:54:05 -04002192 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
Chris Mason5f39d392007-10-15 16:14:19 -04002193 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2194 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2195
2196 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2197 inode->i_atime.tv_sec);
2198 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2199 inode->i_atime.tv_nsec);
2200
2201 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2202 inode->i_mtime.tv_sec);
2203 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2204 inode->i_mtime.tv_nsec);
2205
2206 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2207 inode->i_ctime.tv_sec);
2208 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2209 inode->i_ctime.tv_nsec);
2210
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002211 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
Chris Masone02119d2008-09-05 16:13:11 -04002212 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
Chris Masonc3027eb2008-12-08 16:40:21 -05002213 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
Chris Masone02119d2008-09-05 16:13:11 -04002214 btrfs_set_inode_transid(leaf, item, trans->transid);
Chris Mason5f39d392007-10-15 16:14:19 -04002215 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
Yanb98b6762008-01-08 15:54:37 -05002216 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
Yan Zhengd2fb3432008-12-11 16:30:39 -05002217 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
Chris Mason39279cc2007-06-12 06:35:45 -04002218}
2219
Chris Masond352ac62008-09-29 15:18:18 -04002220/*
2221 * copy everything in the in-memory inode into the btree.
2222 */
Chris Masond3977122009-01-05 21:25:51 -05002223noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2224 struct btrfs_root *root, struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002225{
2226 struct btrfs_inode_item *inode_item;
2227 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002228 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002229 int ret;
2230
2231 path = btrfs_alloc_path();
2232 BUG_ON(!path);
Chris Masonb9473432009-03-13 11:00:37 -04002233 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04002234 ret = btrfs_lookup_inode(trans, root, path,
2235 &BTRFS_I(inode)->location, 1);
2236 if (ret) {
2237 if (ret > 0)
2238 ret = -ENOENT;
2239 goto failed;
2240 }
2241
Chris Masonb4ce94d2009-02-04 09:25:08 -05002242 btrfs_unlock_up_safe(path, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04002243 leaf = path->nodes[0];
2244 inode_item = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04002245 struct btrfs_inode_item);
2246
Chris Masone02119d2008-09-05 16:13:11 -04002247 fill_inode_item(trans, leaf, inode_item, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04002248 btrfs_mark_buffer_dirty(leaf);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002249 btrfs_set_inode_last_trans(trans, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002250 ret = 0;
2251failed:
Chris Mason39279cc2007-06-12 06:35:45 -04002252 btrfs_free_path(path);
2253 return ret;
2254}
2255
2256
Chris Masond352ac62008-09-29 15:18:18 -04002257/*
2258 * unlink helper that gets used here in inode.c and in the tree logging
2259 * recovery code. It remove a link in a directory with a given name, and
2260 * also drops the back refs in the inode to the directory
2261 */
Chris Masone02119d2008-09-05 16:13:11 -04002262int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2263 struct btrfs_root *root,
2264 struct inode *dir, struct inode *inode,
2265 const char *name, int name_len)
Chris Mason39279cc2007-06-12 06:35:45 -04002266{
2267 struct btrfs_path *path;
Chris Mason39279cc2007-06-12 06:35:45 -04002268 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002269 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002270 struct btrfs_dir_item *di;
Chris Mason5f39d392007-10-15 16:14:19 -04002271 struct btrfs_key key;
Josef Bacikaec74772008-07-24 12:12:38 -04002272 u64 index;
Chris Mason39279cc2007-06-12 06:35:45 -04002273
2274 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04002275 if (!path) {
2276 ret = -ENOMEM;
2277 goto err;
2278 }
2279
Chris Masonb9473432009-03-13 11:00:37 -04002280 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04002281 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2282 name, name_len, -1);
2283 if (IS_ERR(di)) {
2284 ret = PTR_ERR(di);
2285 goto err;
2286 }
2287 if (!di) {
2288 ret = -ENOENT;
2289 goto err;
2290 }
Chris Mason5f39d392007-10-15 16:14:19 -04002291 leaf = path->nodes[0];
2292 btrfs_dir_item_key_to_cpu(leaf, di, &key);
Chris Mason39279cc2007-06-12 06:35:45 -04002293 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason54aa1f42007-06-22 14:16:25 -04002294 if (ret)
2295 goto err;
Chris Mason39279cc2007-06-12 06:35:45 -04002296 btrfs_release_path(root, path);
2297
Josef Bacikaec74772008-07-24 12:12:38 -04002298 ret = btrfs_del_inode_ref(trans, root, name, name_len,
Chris Masone02119d2008-09-05 16:13:11 -04002299 inode->i_ino,
2300 dir->i_ino, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04002301 if (ret) {
Chris Masond3977122009-01-05 21:25:51 -05002302 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
Josef Bacikaec74772008-07-24 12:12:38 -04002303 "inode %lu parent %lu\n", name_len, name,
Chris Masone02119d2008-09-05 16:13:11 -04002304 inode->i_ino, dir->i_ino);
Josef Bacikaec74772008-07-24 12:12:38 -04002305 goto err;
2306 }
2307
Chris Mason39279cc2007-06-12 06:35:45 -04002308 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
Josef Bacikaec74772008-07-24 12:12:38 -04002309 index, name, name_len, -1);
Chris Mason39279cc2007-06-12 06:35:45 -04002310 if (IS_ERR(di)) {
2311 ret = PTR_ERR(di);
2312 goto err;
2313 }
2314 if (!di) {
2315 ret = -ENOENT;
2316 goto err;
2317 }
2318 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason925baed2008-06-25 16:01:30 -04002319 btrfs_release_path(root, path);
Chris Mason39279cc2007-06-12 06:35:45 -04002320
Chris Masone02119d2008-09-05 16:13:11 -04002321 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2322 inode, dir->i_ino);
Chris Mason49eb7e42008-09-11 15:53:12 -04002323 BUG_ON(ret != 0 && ret != -ENOENT);
Chris Masone02119d2008-09-05 16:13:11 -04002324
2325 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2326 dir, index);
2327 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04002328err:
2329 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002330 if (ret)
2331 goto out;
2332
2333 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2334 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2335 btrfs_update_inode(trans, root, dir);
2336 btrfs_drop_nlink(inode);
2337 ret = btrfs_update_inode(trans, root, inode);
2338 dir->i_sb->s_dirt = 1;
2339out:
Chris Mason39279cc2007-06-12 06:35:45 -04002340 return ret;
2341}
2342
2343static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2344{
2345 struct btrfs_root *root;
2346 struct btrfs_trans_handle *trans;
Josef Bacik7b128762008-07-24 12:17:14 -04002347 struct inode *inode = dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04002348 int ret;
Chris Mason1832a6d2007-12-21 16:27:21 -05002349 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002350
2351 root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05002352
Chris Mason39279cc2007-06-12 06:35:45 -04002353 trans = btrfs_start_transaction(root, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04002354
Chris Mason39279cc2007-06-12 06:35:45 -04002355 btrfs_set_trans_block_group(trans, dir);
Chris Mason12fcfd22009-03-24 10:24:20 -04002356
2357 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2358
Chris Masone02119d2008-09-05 16:13:11 -04002359 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2360 dentry->d_name.name, dentry->d_name.len);
Josef Bacik7b128762008-07-24 12:17:14 -04002361
2362 if (inode->i_nlink == 0)
2363 ret = btrfs_orphan_add(trans, inode);
2364
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002365 nr = trans->blocks_used;
Chris Mason5f39d392007-10-15 16:14:19 -04002366
Chris Mason89ce8a62008-06-25 16:01:31 -04002367 btrfs_end_transaction_throttle(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002368 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04002369 return ret;
2370}
2371
2372static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2373{
2374 struct inode *inode = dentry->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05002375 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002376 int ret;
2377 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04002378 struct btrfs_trans_handle *trans;
Chris Mason1832a6d2007-12-21 16:27:21 -05002379 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002380
Chris Mason3394e162008-11-17 20:42:26 -05002381 /*
2382 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2383 * the root of a subvolume or snapshot
2384 */
2385 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2386 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
Yan134d4512007-10-25 15:49:25 -04002387 return -ENOTEMPTY;
Chris Mason925baed2008-06-25 16:01:30 -04002388 }
Yan134d4512007-10-25 15:49:25 -04002389
Chris Mason39279cc2007-06-12 06:35:45 -04002390 trans = btrfs_start_transaction(root, 1);
2391 btrfs_set_trans_block_group(trans, dir);
Chris Mason39279cc2007-06-12 06:35:45 -04002392
Josef Bacik7b128762008-07-24 12:17:14 -04002393 err = btrfs_orphan_add(trans, inode);
2394 if (err)
2395 goto fail_trans;
2396
Chris Mason39279cc2007-06-12 06:35:45 -04002397 /* now the directory is empty */
Chris Masone02119d2008-09-05 16:13:11 -04002398 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2399 dentry->d_name.name, dentry->d_name.len);
Chris Masond3977122009-01-05 21:25:51 -05002400 if (!err)
Chris Masondbe674a2008-07-17 12:54:05 -04002401 btrfs_i_size_write(inode, 0);
Chris Mason39544012007-12-12 14:38:19 -05002402
Josef Bacik7b128762008-07-24 12:17:14 -04002403fail_trans:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002404 nr = trans->blocks_used;
Chris Mason89ce8a62008-06-25 16:01:31 -04002405 ret = btrfs_end_transaction_throttle(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002406 btrfs_btree_balance_dirty(root, nr);
Chris Mason39544012007-12-12 14:38:19 -05002407
Chris Mason39279cc2007-06-12 06:35:45 -04002408 if (ret && !err)
2409 err = ret;
2410 return err;
2411}
2412
Chris Masond20f7042008-12-08 16:58:54 -05002413#if 0
Chris Mason39279cc2007-06-12 06:35:45 -04002414/*
Chris Mason323ac952008-10-01 19:05:46 -04002415 * when truncating bytes in a file, it is possible to avoid reading
2416 * the leaves that contain only checksum items. This can be the
2417 * majority of the IO required to delete a large file, but it must
2418 * be done carefully.
2419 *
2420 * The keys in the level just above the leaves are checked to make sure
2421 * the lowest key in a given leaf is a csum key, and starts at an offset
2422 * after the new size.
2423 *
2424 * Then the key for the next leaf is checked to make sure it also has
2425 * a checksum item for the same file. If it does, we know our target leaf
2426 * contains only checksum items, and it can be safely freed without reading
2427 * it.
2428 *
2429 * This is just an optimization targeted at large files. It may do
2430 * nothing. It will return 0 unless things went badly.
2431 */
2432static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2433 struct btrfs_root *root,
2434 struct btrfs_path *path,
2435 struct inode *inode, u64 new_size)
2436{
2437 struct btrfs_key key;
2438 int ret;
2439 int nritems;
2440 struct btrfs_key found_key;
2441 struct btrfs_key other_key;
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002442 struct btrfs_leaf_ref *ref;
2443 u64 leaf_gen;
2444 u64 leaf_start;
Chris Mason323ac952008-10-01 19:05:46 -04002445
2446 path->lowest_level = 1;
2447 key.objectid = inode->i_ino;
2448 key.type = BTRFS_CSUM_ITEM_KEY;
2449 key.offset = new_size;
2450again:
2451 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2452 if (ret < 0)
2453 goto out;
2454
2455 if (path->nodes[1] == NULL) {
2456 ret = 0;
2457 goto out;
2458 }
2459 ret = 0;
2460 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2461 nritems = btrfs_header_nritems(path->nodes[1]);
2462
2463 if (!nritems)
2464 goto out;
2465
2466 if (path->slots[1] >= nritems)
2467 goto next_node;
2468
2469 /* did we find a key greater than anything we want to delete? */
2470 if (found_key.objectid > inode->i_ino ||
2471 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2472 goto out;
2473
2474 /* we check the next key in the node to make sure the leave contains
2475 * only checksum items. This comparison doesn't work if our
2476 * leaf is the last one in the node
2477 */
2478 if (path->slots[1] + 1 >= nritems) {
2479next_node:
2480 /* search forward from the last key in the node, this
2481 * will bring us into the next node in the tree
2482 */
2483 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2484
2485 /* unlikely, but we inc below, so check to be safe */
2486 if (found_key.offset == (u64)-1)
2487 goto out;
2488
2489 /* search_forward needs a path with locks held, do the
2490 * search again for the original key. It is possible
2491 * this will race with a balance and return a path that
2492 * we could modify, but this drop is just an optimization
2493 * and is allowed to miss some leaves.
2494 */
2495 btrfs_release_path(root, path);
2496 found_key.offset++;
2497
2498 /* setup a max key for search_forward */
2499 other_key.offset = (u64)-1;
2500 other_key.type = key.type;
2501 other_key.objectid = key.objectid;
2502
2503 path->keep_locks = 1;
2504 ret = btrfs_search_forward(root, &found_key, &other_key,
2505 path, 0, 0);
2506 path->keep_locks = 0;
2507 if (ret || found_key.objectid != key.objectid ||
2508 found_key.type != key.type) {
2509 ret = 0;
2510 goto out;
2511 }
2512
2513 key.offset = found_key.offset;
2514 btrfs_release_path(root, path);
2515 cond_resched();
2516 goto again;
2517 }
2518
2519 /* we know there's one more slot after us in the tree,
2520 * read that key so we can verify it is also a checksum item
2521 */
2522 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2523
2524 if (found_key.objectid < inode->i_ino)
2525 goto next_key;
2526
2527 if (found_key.type != key.type || found_key.offset < new_size)
2528 goto next_key;
2529
2530 /*
2531 * if the key for the next leaf isn't a csum key from this objectid,
2532 * we can't be sure there aren't good items inside this leaf.
2533 * Bail out
2534 */
2535 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2536 goto out;
2537
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002538 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2539 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
Chris Mason323ac952008-10-01 19:05:46 -04002540 /*
2541 * it is safe to delete this leaf, it contains only
2542 * csum items from this inode at an offset >= new_size
2543 */
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002544 ret = btrfs_del_leaf(trans, root, path, leaf_start);
Chris Mason323ac952008-10-01 19:05:46 -04002545 BUG_ON(ret);
2546
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002547 if (root->ref_cows && leaf_gen < trans->transid) {
2548 ref = btrfs_alloc_leaf_ref(root, 0);
2549 if (ref) {
2550 ref->root_gen = root->root_key.offset;
2551 ref->bytenr = leaf_start;
2552 ref->owner = 0;
2553 ref->generation = leaf_gen;
2554 ref->nritems = 0;
2555
Chris Masonbd56b302009-02-04 09:27:02 -05002556 btrfs_sort_leaf_ref(ref);
2557
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002558 ret = btrfs_add_leaf_ref(root, ref, 0);
2559 WARN_ON(ret);
2560 btrfs_free_leaf_ref(root, ref);
2561 } else {
2562 WARN_ON(1);
2563 }
2564 }
Chris Mason323ac952008-10-01 19:05:46 -04002565next_key:
2566 btrfs_release_path(root, path);
2567
2568 if (other_key.objectid == inode->i_ino &&
2569 other_key.type == key.type && other_key.offset > key.offset) {
2570 key.offset = other_key.offset;
2571 cond_resched();
2572 goto again;
2573 }
2574 ret = 0;
2575out:
2576 /* fixup any changes we've made to the path */
2577 path->lowest_level = 0;
2578 path->keep_locks = 0;
2579 btrfs_release_path(root, path);
2580 return ret;
2581}
2582
Chris Masond20f7042008-12-08 16:58:54 -05002583#endif
2584
Chris Mason323ac952008-10-01 19:05:46 -04002585/*
Chris Mason39279cc2007-06-12 06:35:45 -04002586 * this can truncate away extent items, csum items and directory items.
2587 * It starts at a high offset and removes keys until it can't find
Chris Masond352ac62008-09-29 15:18:18 -04002588 * any higher than new_size
Chris Mason39279cc2007-06-12 06:35:45 -04002589 *
2590 * csum items that cross the new i_size are truncated to the new size
2591 * as well.
Josef Bacik7b128762008-07-24 12:17:14 -04002592 *
2593 * min_type is the minimum key type to truncate down to. If set to 0, this
2594 * will kill all the items on this inode, including the INODE_ITEM_KEY.
Chris Mason39279cc2007-06-12 06:35:45 -04002595 */
Chris Masone02119d2008-09-05 16:13:11 -04002596noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2597 struct btrfs_root *root,
2598 struct inode *inode,
2599 u64 new_size, u32 min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04002600{
2601 int ret;
2602 struct btrfs_path *path;
2603 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04002604 struct btrfs_key found_key;
Chris Mason06d9a8d2009-02-04 09:30:58 -05002605 u32 found_type = (u8)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04002606 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002607 struct btrfs_file_extent_item *fi;
2608 u64 extent_start = 0;
Chris Masondb945352007-10-15 16:15:53 -04002609 u64 extent_num_bytes = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002610 u64 extent_offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002611 u64 item_end = 0;
2612 int found_extent;
2613 int del_item;
Chris Mason85e21ba2008-01-29 15:11:36 -05002614 int pending_del_nr = 0;
2615 int pending_del_slot = 0;
Chris Mason179e29e2007-11-01 11:28:41 -04002616 int extent_type = -1;
Chris Mason771ed682008-11-06 22:02:51 -05002617 int encoding;
Chris Mason3b951512008-04-17 11:29:12 -04002618 u64 mask = root->sectorsize - 1;
Chris Mason39279cc2007-06-12 06:35:45 -04002619
Chris Masone02119d2008-09-05 16:13:11 -04002620 if (root->ref_cows)
Zheng Yan5b21f2e2008-09-26 10:05:38 -04002621 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04002622 path = btrfs_alloc_path();
2623 BUG_ON(!path);
Julia Lawall33c17ad2009-07-22 16:49:01 -04002624 path->reada = -1;
Chris Mason5f39d392007-10-15 16:14:19 -04002625
Chris Mason39279cc2007-06-12 06:35:45 -04002626 /* FIXME, add redo link to tree so we don't leak on crash */
2627 key.objectid = inode->i_ino;
2628 key.offset = (u64)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04002629 key.type = (u8)-1;
2630
Chris Mason85e21ba2008-01-29 15:11:36 -05002631search_again:
Chris Masonb9473432009-03-13 11:00:37 -04002632 path->leave_spinning = 1;
Chris Mason85e21ba2008-01-29 15:11:36 -05002633 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Chris Masond3977122009-01-05 21:25:51 -05002634 if (ret < 0)
Chris Mason85e21ba2008-01-29 15:11:36 -05002635 goto error;
Chris Masond3977122009-01-05 21:25:51 -05002636
Chris Mason85e21ba2008-01-29 15:11:36 -05002637 if (ret > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002638 /* there are no items in the tree for us to truncate, we're
2639 * done
2640 */
2641 if (path->slots[0] == 0) {
2642 ret = 0;
2643 goto error;
2644 }
Chris Mason85e21ba2008-01-29 15:11:36 -05002645 path->slots[0]--;
2646 }
2647
Chris Masond3977122009-01-05 21:25:51 -05002648 while (1) {
Chris Mason39279cc2007-06-12 06:35:45 -04002649 fi = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -04002650 leaf = path->nodes[0];
2651 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2652 found_type = btrfs_key_type(&found_key);
Chris Mason771ed682008-11-06 22:02:51 -05002653 encoding = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002654
Chris Mason5f39d392007-10-15 16:14:19 -04002655 if (found_key.objectid != inode->i_ino)
Chris Mason39279cc2007-06-12 06:35:45 -04002656 break;
Chris Mason5f39d392007-10-15 16:14:19 -04002657
Chris Mason85e21ba2008-01-29 15:11:36 -05002658 if (found_type < min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04002659 break;
2660
Chris Mason5f39d392007-10-15 16:14:19 -04002661 item_end = found_key.offset;
Chris Mason39279cc2007-06-12 06:35:45 -04002662 if (found_type == BTRFS_EXTENT_DATA_KEY) {
Chris Mason5f39d392007-10-15 16:14:19 -04002663 fi = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04002664 struct btrfs_file_extent_item);
Chris Mason179e29e2007-11-01 11:28:41 -04002665 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Mason771ed682008-11-06 22:02:51 -05002666 encoding = btrfs_file_extent_compression(leaf, fi);
2667 encoding |= btrfs_file_extent_encryption(leaf, fi);
2668 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2669
Chris Mason179e29e2007-11-01 11:28:41 -04002670 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04002671 item_end +=
Chris Masondb945352007-10-15 16:15:53 -04002672 btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason179e29e2007-11-01 11:28:41 -04002673 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason179e29e2007-11-01 11:28:41 -04002674 item_end += btrfs_file_extent_inline_len(leaf,
Chris Masonc8b97812008-10-29 14:49:59 -04002675 fi);
Chris Mason39279cc2007-06-12 06:35:45 -04002676 }
Yan008630c2007-11-07 13:31:09 -05002677 item_end--;
Chris Mason39279cc2007-06-12 06:35:45 -04002678 }
Chris Masone02119d2008-09-05 16:13:11 -04002679 if (item_end < new_size) {
Chris Masond3977122009-01-05 21:25:51 -05002680 if (found_type == BTRFS_DIR_ITEM_KEY)
Chris Masonb888db22007-08-27 16:49:44 -04002681 found_type = BTRFS_INODE_ITEM_KEY;
Chris Masond3977122009-01-05 21:25:51 -05002682 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
Chris Masond20f7042008-12-08 16:58:54 -05002683 found_type = BTRFS_EXTENT_DATA_KEY;
Chris Masond3977122009-01-05 21:25:51 -05002684 else if (found_type == BTRFS_EXTENT_DATA_KEY)
Chris Mason85e21ba2008-01-29 15:11:36 -05002685 found_type = BTRFS_XATTR_ITEM_KEY;
Chris Masond3977122009-01-05 21:25:51 -05002686 else if (found_type == BTRFS_XATTR_ITEM_KEY)
Chris Mason85e21ba2008-01-29 15:11:36 -05002687 found_type = BTRFS_INODE_REF_KEY;
Chris Masond3977122009-01-05 21:25:51 -05002688 else if (found_type)
Chris Masonb888db22007-08-27 16:49:44 -04002689 found_type--;
Chris Masond3977122009-01-05 21:25:51 -05002690 else
Chris Masonb888db22007-08-27 16:49:44 -04002691 break;
Yana61721d2007-09-17 11:08:38 -04002692 btrfs_set_key_type(&key, found_type);
Chris Mason85e21ba2008-01-29 15:11:36 -05002693 goto next;
Chris Mason39279cc2007-06-12 06:35:45 -04002694 }
Chris Masone02119d2008-09-05 16:13:11 -04002695 if (found_key.offset >= new_size)
Chris Mason39279cc2007-06-12 06:35:45 -04002696 del_item = 1;
2697 else
2698 del_item = 0;
2699 found_extent = 0;
2700
2701 /* FIXME, shrink the extent if the ref count is only 1 */
Chris Mason179e29e2007-11-01 11:28:41 -04002702 if (found_type != BTRFS_EXTENT_DATA_KEY)
2703 goto delete;
2704
2705 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason39279cc2007-06-12 06:35:45 -04002706 u64 num_dec;
Chris Masondb945352007-10-15 16:15:53 -04002707 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
Chris Mason771ed682008-11-06 22:02:51 -05002708 if (!del_item && !encoding) {
Chris Masondb945352007-10-15 16:15:53 -04002709 u64 orig_num_bytes =
2710 btrfs_file_extent_num_bytes(leaf, fi);
Chris Masone02119d2008-09-05 16:13:11 -04002711 extent_num_bytes = new_size -
Chris Mason5f39d392007-10-15 16:14:19 -04002712 found_key.offset + root->sectorsize - 1;
Yanb1632b12008-01-30 11:54:04 -05002713 extent_num_bytes = extent_num_bytes &
2714 ~((u64)root->sectorsize - 1);
Chris Masondb945352007-10-15 16:15:53 -04002715 btrfs_set_file_extent_num_bytes(leaf, fi,
2716 extent_num_bytes);
2717 num_dec = (orig_num_bytes -
Chris Mason90692182008-02-08 13:49:28 -05002718 extent_num_bytes);
Chris Masone02119d2008-09-05 16:13:11 -04002719 if (root->ref_cows && extent_start != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002720 inode_sub_bytes(inode, num_dec);
Chris Mason5f39d392007-10-15 16:14:19 -04002721 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04002722 } else {
Chris Masondb945352007-10-15 16:15:53 -04002723 extent_num_bytes =
2724 btrfs_file_extent_disk_num_bytes(leaf,
2725 fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002726 extent_offset = found_key.offset -
2727 btrfs_file_extent_offset(leaf, fi);
2728
Chris Mason39279cc2007-06-12 06:35:45 -04002729 /* FIXME blocksize != 4096 */
Chris Mason90692182008-02-08 13:49:28 -05002730 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason39279cc2007-06-12 06:35:45 -04002731 if (extent_start != 0) {
2732 found_extent = 1;
Chris Masone02119d2008-09-05 16:13:11 -04002733 if (root->ref_cows)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002734 inode_sub_bytes(inode, num_dec);
Chris Mason39279cc2007-06-12 06:35:45 -04002735 }
2736 }
Chris Mason90692182008-02-08 13:49:28 -05002737 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -04002738 /*
2739 * we can't truncate inline items that have had
2740 * special encodings
2741 */
2742 if (!del_item &&
2743 btrfs_file_extent_compression(leaf, fi) == 0 &&
2744 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2745 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002746 u32 size = new_size - found_key.offset;
2747
2748 if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002749 inode_sub_bytes(inode, item_end + 1 -
2750 new_size);
Chris Masone02119d2008-09-05 16:13:11 -04002751 }
2752 size =
2753 btrfs_file_extent_calc_inline_size(size);
Chris Mason90692182008-02-08 13:49:28 -05002754 ret = btrfs_truncate_item(trans, root, path,
Chris Masone02119d2008-09-05 16:13:11 -04002755 size, 1);
Chris Mason90692182008-02-08 13:49:28 -05002756 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04002757 } else if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002758 inode_sub_bytes(inode, item_end + 1 -
2759 found_key.offset);
Chris Mason90692182008-02-08 13:49:28 -05002760 }
Chris Mason39279cc2007-06-12 06:35:45 -04002761 }
Chris Mason179e29e2007-11-01 11:28:41 -04002762delete:
Chris Mason39279cc2007-06-12 06:35:45 -04002763 if (del_item) {
Chris Mason85e21ba2008-01-29 15:11:36 -05002764 if (!pending_del_nr) {
2765 /* no pending yet, add ourselves */
2766 pending_del_slot = path->slots[0];
2767 pending_del_nr = 1;
2768 } else if (pending_del_nr &&
2769 path->slots[0] + 1 == pending_del_slot) {
2770 /* hop on the pending chunk */
2771 pending_del_nr++;
2772 pending_del_slot = path->slots[0];
2773 } else {
Chris Masond3977122009-01-05 21:25:51 -05002774 BUG();
Chris Mason85e21ba2008-01-29 15:11:36 -05002775 }
Chris Mason39279cc2007-06-12 06:35:45 -04002776 } else {
2777 break;
2778 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002779 if (found_extent && root->ref_cows) {
Chris Masonb9473432009-03-13 11:00:37 -04002780 btrfs_set_path_blocking(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002781 ret = btrfs_free_extent(trans, root, extent_start,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002782 extent_num_bytes, 0,
2783 btrfs_header_owner(leaf),
2784 inode->i_ino, extent_offset);
Chris Mason39279cc2007-06-12 06:35:45 -04002785 BUG_ON(ret);
2786 }
Chris Mason85e21ba2008-01-29 15:11:36 -05002787next:
2788 if (path->slots[0] == 0) {
2789 if (pending_del_nr)
2790 goto del_pending;
2791 btrfs_release_path(root, path);
Chris Mason06d9a8d2009-02-04 09:30:58 -05002792 if (found_type == BTRFS_INODE_ITEM_KEY)
2793 break;
Chris Mason85e21ba2008-01-29 15:11:36 -05002794 goto search_again;
2795 }
2796
2797 path->slots[0]--;
2798 if (pending_del_nr &&
2799 path->slots[0] + 1 != pending_del_slot) {
2800 struct btrfs_key debug;
2801del_pending:
2802 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2803 pending_del_slot);
2804 ret = btrfs_del_items(trans, root, path,
2805 pending_del_slot,
2806 pending_del_nr);
2807 BUG_ON(ret);
2808 pending_del_nr = 0;
2809 btrfs_release_path(root, path);
Chris Mason06d9a8d2009-02-04 09:30:58 -05002810 if (found_type == BTRFS_INODE_ITEM_KEY)
2811 break;
Chris Mason85e21ba2008-01-29 15:11:36 -05002812 goto search_again;
2813 }
Chris Mason39279cc2007-06-12 06:35:45 -04002814 }
2815 ret = 0;
2816error:
Chris Mason85e21ba2008-01-29 15:11:36 -05002817 if (pending_del_nr) {
2818 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2819 pending_del_nr);
2820 }
Chris Mason39279cc2007-06-12 06:35:45 -04002821 btrfs_free_path(path);
2822 inode->i_sb->s_dirt = 1;
2823 return ret;
2824}
2825
Chris Masona52d9a82007-08-27 16:49:44 -04002826/*
2827 * taken from block_truncate_page, but does cow as it zeros out
2828 * any bytes left in the last page in the file.
2829 */
2830static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2831{
2832 struct inode *inode = mapping->host;
Chris Masondb945352007-10-15 16:15:53 -04002833 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002834 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2835 struct btrfs_ordered_extent *ordered;
2836 char *kaddr;
Chris Masondb945352007-10-15 16:15:53 -04002837 u32 blocksize = root->sectorsize;
Chris Masona52d9a82007-08-27 16:49:44 -04002838 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2839 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2840 struct page *page;
2841 int ret = 0;
2842 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002843 u64 page_end;
Chris Masona52d9a82007-08-27 16:49:44 -04002844
2845 if ((offset & (blocksize - 1)) == 0)
2846 goto out;
2847
2848 ret = -ENOMEM;
Chris Mason211c17f2008-05-15 09:13:45 -04002849again:
Chris Masona52d9a82007-08-27 16:49:44 -04002850 page = grab_cache_page(mapping, index);
2851 if (!page)
2852 goto out;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002853
2854 page_start = page_offset(page);
2855 page_end = page_start + PAGE_CACHE_SIZE - 1;
2856
Chris Masona52d9a82007-08-27 16:49:44 -04002857 if (!PageUptodate(page)) {
2858 ret = btrfs_readpage(NULL, page);
2859 lock_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04002860 if (page->mapping != mapping) {
2861 unlock_page(page);
2862 page_cache_release(page);
2863 goto again;
2864 }
Chris Masona52d9a82007-08-27 16:49:44 -04002865 if (!PageUptodate(page)) {
2866 ret = -EIO;
Chris Mason89642222008-07-24 09:41:53 -04002867 goto out_unlock;
Chris Masona52d9a82007-08-27 16:49:44 -04002868 }
2869 }
Chris Mason211c17f2008-05-15 09:13:45 -04002870 wait_on_page_writeback(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002871
2872 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2873 set_page_extent_mapped(page);
2874
2875 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2876 if (ordered) {
2877 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2878 unlock_page(page);
2879 page_cache_release(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04002880 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002881 btrfs_put_ordered_extent(ordered);
2882 goto again;
2883 }
2884
Chris Masonea8c2812008-08-04 23:17:27 -04002885 btrfs_set_extent_delalloc(inode, page_start, page_end);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002886 ret = 0;
2887 if (offset != PAGE_CACHE_SIZE) {
2888 kaddr = kmap(page);
2889 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2890 flush_dcache_page(page);
2891 kunmap(page);
2892 }
Chris Mason247e7432008-07-17 12:53:51 -04002893 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002894 set_page_dirty(page);
2895 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04002896
Chris Mason89642222008-07-24 09:41:53 -04002897out_unlock:
Chris Mason39279cc2007-06-12 06:35:45 -04002898 unlock_page(page);
2899 page_cache_release(page);
2900out:
2901 return ret;
2902}
2903
Yan Zheng9036c102008-10-30 14:19:41 -04002904int btrfs_cont_expand(struct inode *inode, loff_t size)
2905{
2906 struct btrfs_trans_handle *trans;
2907 struct btrfs_root *root = BTRFS_I(inode)->root;
2908 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2909 struct extent_map *em;
2910 u64 mask = root->sectorsize - 1;
2911 u64 hole_start = (inode->i_size + mask) & ~mask;
2912 u64 block_end = (size + mask) & ~mask;
2913 u64 last_byte;
2914 u64 cur_offset;
2915 u64 hole_size;
2916 int err;
2917
2918 if (size <= hole_start)
2919 return 0;
2920
Josef Bacik6a632092009-02-20 11:00:09 -05002921 err = btrfs_check_metadata_free_space(root);
Yan Zheng9036c102008-10-30 14:19:41 -04002922 if (err)
2923 return err;
2924
2925 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2926
2927 while (1) {
2928 struct btrfs_ordered_extent *ordered;
2929 btrfs_wait_ordered_range(inode, hole_start,
2930 block_end - hole_start);
2931 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2932 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2933 if (!ordered)
2934 break;
2935 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2936 btrfs_put_ordered_extent(ordered);
2937 }
2938
2939 trans = btrfs_start_transaction(root, 1);
2940 btrfs_set_trans_block_group(trans, inode);
2941
2942 cur_offset = hole_start;
2943 while (1) {
2944 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2945 block_end - cur_offset, 0);
2946 BUG_ON(IS_ERR(em) || !em);
2947 last_byte = min(extent_map_end(em), block_end);
2948 last_byte = (last_byte + mask) & ~mask;
2949 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
Chris Mason771ed682008-11-06 22:02:51 -05002950 u64 hint_byte = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04002951 hole_size = last_byte - cur_offset;
Chris Mason771ed682008-11-06 22:02:51 -05002952 err = btrfs_drop_extents(trans, root, inode,
2953 cur_offset,
2954 cur_offset + hole_size,
Chris Masone980b502009-04-24 14:39:24 -04002955 block_end,
Chris Masona1ed8352009-09-11 12:27:37 -04002956 cur_offset, &hint_byte, 1);
Chris Mason771ed682008-11-06 22:02:51 -05002957 if (err)
2958 break;
Yan Zheng9036c102008-10-30 14:19:41 -04002959 err = btrfs_insert_file_extent(trans, root,
2960 inode->i_ino, cur_offset, 0,
2961 0, hole_size, 0, hole_size,
2962 0, 0, 0);
2963 btrfs_drop_extent_cache(inode, hole_start,
2964 last_byte - 1, 0);
2965 }
2966 free_extent_map(em);
2967 cur_offset = last_byte;
2968 if (err || cur_offset >= block_end)
2969 break;
2970 }
2971
2972 btrfs_end_transaction(trans, root);
2973 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2974 return err;
2975}
2976
Chris Mason39279cc2007-06-12 06:35:45 -04002977static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2978{
2979 struct inode *inode = dentry->d_inode;
2980 int err;
2981
2982 err = inode_change_ok(inode, attr);
2983 if (err)
2984 return err;
2985
Chris Mason5a3f23d2009-03-31 13:27:11 -04002986 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2987 if (attr->ia_size > inode->i_size) {
2988 err = btrfs_cont_expand(inode, attr->ia_size);
2989 if (err)
2990 return err;
2991 } else if (inode->i_size > 0 &&
2992 attr->ia_size == 0) {
2993
2994 /* we're truncating a file that used to have good
2995 * data down to zero. Make sure it gets into
2996 * the ordered flush list so that any new writes
2997 * get down to disk quickly.
2998 */
2999 BTRFS_I(inode)->ordered_data_close = 1;
3000 }
Chris Mason39279cc2007-06-12 06:35:45 -04003001 }
Yan Zheng9036c102008-10-30 14:19:41 -04003002
Chris Mason39279cc2007-06-12 06:35:45 -04003003 err = inode_setattr(inode, attr);
Josef Bacik33268ea2008-07-24 12:16:36 -04003004
3005 if (!err && ((attr->ia_valid & ATTR_MODE)))
3006 err = btrfs_acl_chmod(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003007 return err;
3008}
Chris Mason61295eb2008-01-14 16:24:38 -05003009
Chris Mason39279cc2007-06-12 06:35:45 -04003010void btrfs_delete_inode(struct inode *inode)
3011{
3012 struct btrfs_trans_handle *trans;
3013 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003014 unsigned long nr;
Chris Mason39279cc2007-06-12 06:35:45 -04003015 int ret;
3016
3017 truncate_inode_pages(&inode->i_data, 0);
3018 if (is_bad_inode(inode)) {
Josef Bacik7b128762008-07-24 12:17:14 -04003019 btrfs_orphan_del(NULL, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003020 goto no_delete;
3021 }
Chris Mason4a096752008-07-21 10:29:44 -04003022 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Chris Mason5f39d392007-10-15 16:14:19 -04003023
Chris Masondbe674a2008-07-17 12:54:05 -04003024 btrfs_i_size_write(inode, 0);
Yan Zheng180591b2009-01-06 09:58:06 -05003025 trans = btrfs_join_transaction(root, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04003026
Chris Mason39279cc2007-06-12 06:35:45 -04003027 btrfs_set_trans_block_group(trans, inode);
Chris Masone02119d2008-09-05 16:13:11 -04003028 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
Josef Bacik7b128762008-07-24 12:17:14 -04003029 if (ret) {
3030 btrfs_orphan_del(NULL, inode);
Chris Mason54aa1f42007-06-22 14:16:25 -04003031 goto no_delete_lock;
Josef Bacik7b128762008-07-24 12:17:14 -04003032 }
3033
3034 btrfs_orphan_del(trans, inode);
Chris Mason85e21ba2008-01-29 15:11:36 -05003035
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003036 nr = trans->blocks_used;
Chris Mason85e21ba2008-01-29 15:11:36 -05003037 clear_inode(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04003038
Chris Mason39279cc2007-06-12 06:35:45 -04003039 btrfs_end_transaction(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003040 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04003041 return;
Chris Mason54aa1f42007-06-22 14:16:25 -04003042
3043no_delete_lock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003044 nr = trans->blocks_used;
Chris Mason54aa1f42007-06-22 14:16:25 -04003045 btrfs_end_transaction(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003046 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04003047no_delete:
3048 clear_inode(inode);
3049}
3050
3051/*
3052 * this returns the key found in the dir entry in the location pointer.
3053 * If no dir entries were found, location->objectid is 0.
3054 */
3055static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3056 struct btrfs_key *location)
3057{
3058 const char *name = dentry->d_name.name;
3059 int namelen = dentry->d_name.len;
3060 struct btrfs_dir_item *di;
3061 struct btrfs_path *path;
3062 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan0d9f7f32007-10-25 15:48:28 -04003063 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003064
3065 path = btrfs_alloc_path();
3066 BUG_ON(!path);
Chris Mason39544012007-12-12 14:38:19 -05003067
Chris Mason39279cc2007-06-12 06:35:45 -04003068 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3069 namelen, 0);
Yan0d9f7f32007-10-25 15:48:28 -04003070 if (IS_ERR(di))
3071 ret = PTR_ERR(di);
Chris Masond3977122009-01-05 21:25:51 -05003072
3073 if (!di || IS_ERR(di))
Chris Mason39544012007-12-12 14:38:19 -05003074 goto out_err;
Chris Masond3977122009-01-05 21:25:51 -05003075
Chris Mason5f39d392007-10-15 16:14:19 -04003076 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
Chris Mason39279cc2007-06-12 06:35:45 -04003077out:
Chris Mason39279cc2007-06-12 06:35:45 -04003078 btrfs_free_path(path);
3079 return ret;
Chris Mason39544012007-12-12 14:38:19 -05003080out_err:
3081 location->objectid = 0;
3082 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04003083}
3084
3085/*
3086 * when we hit a tree root in a directory, the btrfs part of the inode
3087 * needs to be changed to reflect the root directory of the tree root. This
3088 * is kind of like crossing a mount point.
3089 */
3090static int fixup_tree_root_location(struct btrfs_root *root,
3091 struct btrfs_key *location,
Josef Bacik58176a92007-08-29 15:47:34 -04003092 struct btrfs_root **sub_root,
3093 struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04003094{
Chris Mason39279cc2007-06-12 06:35:45 -04003095 struct btrfs_root_item *ri;
3096
3097 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3098 return 0;
3099 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3100 return 0;
3101
Josef Bacik58176a92007-08-29 15:47:34 -04003102 *sub_root = btrfs_read_fs_root(root->fs_info, location,
3103 dentry->d_name.name,
3104 dentry->d_name.len);
Chris Mason39279cc2007-06-12 06:35:45 -04003105 if (IS_ERR(*sub_root))
3106 return PTR_ERR(*sub_root);
3107
3108 ri = &(*sub_root)->root_item;
3109 location->objectid = btrfs_root_dirid(ri);
Chris Mason39279cc2007-06-12 06:35:45 -04003110 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3111 location->offset = 0;
3112
Chris Mason39279cc2007-06-12 06:35:45 -04003113 return 0;
3114}
3115
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003116static void inode_tree_add(struct inode *inode)
3117{
3118 struct btrfs_root *root = BTRFS_I(inode)->root;
3119 struct btrfs_inode *entry;
3120 struct rb_node **p = &root->inode_tree.rb_node;
3121 struct rb_node *parent = NULL;
3122
3123 spin_lock(&root->inode_lock);
3124 while (*p) {
3125 parent = *p;
3126 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3127
3128 if (inode->i_ino < entry->vfs_inode.i_ino)
3129 p = &(*p)->rb_left;
3130 else if (inode->i_ino > entry->vfs_inode.i_ino)
3131 p = &(*p)->rb_right;
3132 else {
3133 WARN_ON(!(entry->vfs_inode.i_state &
3134 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3135 break;
3136 }
3137 }
3138 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3139 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3140 spin_unlock(&root->inode_lock);
3141}
3142
3143static void inode_tree_del(struct inode *inode)
3144{
3145 struct btrfs_root *root = BTRFS_I(inode)->root;
3146
3147 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3148 spin_lock(&root->inode_lock);
3149 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3150 spin_unlock(&root->inode_lock);
3151 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3152 }
3153}
3154
Chris Masone02119d2008-09-05 16:13:11 -04003155static noinline void init_btrfs_i(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003156{
Chris Masone02119d2008-09-05 16:13:11 -04003157 struct btrfs_inode *bi = BTRFS_I(inode);
3158
Chris Mason7b1a14b2009-04-27 10:49:53 -04003159 bi->i_acl = BTRFS_ACL_NOT_CACHED;
3160 bi->i_default_acl = BTRFS_ACL_NOT_CACHED;
Chris Masone02119d2008-09-05 16:13:11 -04003161
3162 bi->generation = 0;
Chris Masonc3027eb2008-12-08 16:40:21 -05003163 bi->sequence = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003164 bi->last_trans = 0;
3165 bi->logged_trans = 0;
3166 bi->delalloc_bytes = 0;
Josef Bacik6a632092009-02-20 11:00:09 -05003167 bi->reserved_bytes = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003168 bi->disk_i_size = 0;
3169 bi->flags = 0;
3170 bi->index_cnt = (u64)-1;
Chris Mason12fcfd22009-03-24 10:24:20 -04003171 bi->last_unlink_trans = 0;
Chris Mason27574952009-05-14 13:10:02 -04003172 bi->ordered_data_close = 0;
Chris Masond1310b22008-01-24 16:13:08 -05003173 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3174 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
Chris Masonb888db22007-08-27 16:49:44 -04003175 inode->i_mapping, GFP_NOFS);
Chris Mason7e383262008-04-09 16:28:12 -04003176 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3177 inode->i_mapping, GFP_NOFS);
Chris Masonea8c2812008-08-04 23:17:27 -04003178 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
Chris Mason5a3f23d2009-03-31 13:27:11 -04003179 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003180 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Chris Masonba1da2f2008-07-17 12:54:15 -04003181 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
Chris Masonee6e6502008-07-17 12:54:40 -04003182 mutex_init(&BTRFS_I(inode)->extent_mutex);
Chris Masone02119d2008-09-05 16:13:11 -04003183 mutex_init(&BTRFS_I(inode)->log_mutex);
3184}
3185
3186static int btrfs_init_locked_inode(struct inode *inode, void *p)
3187{
3188 struct btrfs_iget_args *args = p;
3189 inode->i_ino = args->ino;
3190 init_btrfs_i(inode);
3191 BTRFS_I(inode)->root = args->root;
Josef Bacik6a632092009-02-20 11:00:09 -05003192 btrfs_set_inode_space_info(args->root, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003193 return 0;
3194}
3195
3196static int btrfs_find_actor(struct inode *inode, void *opaque)
3197{
3198 struct btrfs_iget_args *args = opaque;
Chris Masond3977122009-01-05 21:25:51 -05003199 return args->ino == inode->i_ino &&
3200 args->root == BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003201}
3202
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003203static struct inode *btrfs_iget_locked(struct super_block *s,
3204 u64 objectid,
3205 struct btrfs_root *root)
Chris Mason39279cc2007-06-12 06:35:45 -04003206{
3207 struct inode *inode;
3208 struct btrfs_iget_args args;
3209 args.ino = objectid;
3210 args.root = root;
3211
3212 inode = iget5_locked(s, objectid, btrfs_find_actor,
3213 btrfs_init_locked_inode,
3214 (void *)&args);
3215 return inode;
3216}
3217
Balaji Rao1a54ef82008-07-21 02:01:04 +05303218/* Get an inode object given its location and corresponding root.
3219 * Returns in *is_new if the inode was read from disk
3220 */
3221struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003222 struct btrfs_root *root)
Balaji Rao1a54ef82008-07-21 02:01:04 +05303223{
3224 struct inode *inode;
3225
3226 inode = btrfs_iget_locked(s, location->objectid, root);
3227 if (!inode)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003228 return ERR_PTR(-ENOMEM);
Balaji Rao1a54ef82008-07-21 02:01:04 +05303229
3230 if (inode->i_state & I_NEW) {
3231 BTRFS_I(inode)->root = root;
3232 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3233 btrfs_read_locked_inode(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003234
3235 inode_tree_add(inode);
Balaji Rao1a54ef82008-07-21 02:01:04 +05303236 unlock_new_inode(inode);
Balaji Rao1a54ef82008-07-21 02:01:04 +05303237 }
3238
3239 return inode;
3240}
3241
Chris Mason3de45862008-11-17 21:02:50 -05003242struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04003243{
Chris Masond3977122009-01-05 21:25:51 -05003244 struct inode *inode;
Chris Mason39279cc2007-06-12 06:35:45 -04003245 struct btrfs_inode *bi = BTRFS_I(dir);
3246 struct btrfs_root *root = bi->root;
3247 struct btrfs_root *sub_root = root;
3248 struct btrfs_key location;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003249 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04003250
3251 if (dentry->d_name.len > BTRFS_NAME_LEN)
3252 return ERR_PTR(-ENAMETOOLONG);
Chris Mason5f39d392007-10-15 16:14:19 -04003253
Chris Mason39279cc2007-06-12 06:35:45 -04003254 ret = btrfs_inode_by_name(dir, dentry, &location);
Chris Mason5f39d392007-10-15 16:14:19 -04003255
Chris Mason39279cc2007-06-12 06:35:45 -04003256 if (ret < 0)
3257 return ERR_PTR(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04003258
Chris Mason39279cc2007-06-12 06:35:45 -04003259 inode = NULL;
3260 if (location.objectid) {
Josef Bacik58176a92007-08-29 15:47:34 -04003261 ret = fixup_tree_root_location(root, &location, &sub_root,
3262 dentry);
Chris Mason39279cc2007-06-12 06:35:45 -04003263 if (ret < 0)
3264 return ERR_PTR(ret);
3265 if (ret > 0)
3266 return ERR_PTR(-ENOENT);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003267 inode = btrfs_iget(dir->i_sb, &location, sub_root);
Balaji Rao1a54ef82008-07-21 02:01:04 +05303268 if (IS_ERR(inode))
3269 return ERR_CAST(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003270 }
Chris Mason3de45862008-11-17 21:02:50 -05003271 return inode;
3272}
3273
3274static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3275 struct nameidata *nd)
3276{
3277 struct inode *inode;
3278
3279 if (dentry->d_name.len > BTRFS_NAME_LEN)
3280 return ERR_PTR(-ENAMETOOLONG);
3281
3282 inode = btrfs_lookup_dentry(dir, dentry);
3283 if (IS_ERR(inode))
3284 return ERR_CAST(inode);
Josef Bacik7b128762008-07-24 12:17:14 -04003285
Chris Mason39279cc2007-06-12 06:35:45 -04003286 return d_splice_alias(inode, dentry);
3287}
3288
Chris Mason39279cc2007-06-12 06:35:45 -04003289static unsigned char btrfs_filetype_table[] = {
3290 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3291};
3292
David Woodhousecbdf5a22008-08-06 19:42:33 +01003293static int btrfs_real_readdir(struct file *filp, void *dirent,
3294 filldir_t filldir)
Chris Mason39279cc2007-06-12 06:35:45 -04003295{
Chris Mason6da6aba2007-12-18 16:15:09 -05003296 struct inode *inode = filp->f_dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04003297 struct btrfs_root *root = BTRFS_I(inode)->root;
3298 struct btrfs_item *item;
3299 struct btrfs_dir_item *di;
3300 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04003301 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04003302 struct btrfs_path *path;
3303 int ret;
3304 u32 nritems;
Chris Mason5f39d392007-10-15 16:14:19 -04003305 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003306 int slot;
3307 int advance;
3308 unsigned char d_type;
3309 int over = 0;
3310 u32 di_cur;
3311 u32 di_total;
3312 u32 di_len;
3313 int key_type = BTRFS_DIR_INDEX_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04003314 char tmp_name[32];
3315 char *name_ptr;
3316 int name_len;
Chris Mason39279cc2007-06-12 06:35:45 -04003317
3318 /* FIXME, use a real flag for deciding about the key type */
3319 if (root->fs_info->tree_root == root)
3320 key_type = BTRFS_DIR_ITEM_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04003321
Chris Mason39544012007-12-12 14:38:19 -05003322 /* special case for "." */
3323 if (filp->f_pos == 0) {
3324 over = filldir(dirent, ".", 1,
3325 1, inode->i_ino,
3326 DT_DIR);
3327 if (over)
3328 return 0;
3329 filp->f_pos = 1;
3330 }
Chris Mason39544012007-12-12 14:38:19 -05003331 /* special case for .., just use the back ref */
3332 if (filp->f_pos == 1) {
David Woodhouse5ecc7e52008-08-17 15:14:48 +01003333 u64 pino = parent_ino(filp->f_path.dentry);
Chris Mason39544012007-12-12 14:38:19 -05003334 over = filldir(dirent, "..", 2,
David Woodhouse5ecc7e52008-08-17 15:14:48 +01003335 2, pino, DT_DIR);
Chris Mason39544012007-12-12 14:38:19 -05003336 if (over)
David Woodhouse49593bf2008-08-17 17:08:36 +01003337 return 0;
Chris Mason39544012007-12-12 14:38:19 -05003338 filp->f_pos = 2;
3339 }
David Woodhouse49593bf2008-08-17 17:08:36 +01003340 path = btrfs_alloc_path();
3341 path->reada = 2;
3342
Chris Mason39279cc2007-06-12 06:35:45 -04003343 btrfs_set_key_type(&key, key_type);
3344 key.offset = filp->f_pos;
David Woodhouse49593bf2008-08-17 17:08:36 +01003345 key.objectid = inode->i_ino;
Chris Mason5f39d392007-10-15 16:14:19 -04003346
Chris Mason39279cc2007-06-12 06:35:45 -04003347 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3348 if (ret < 0)
3349 goto err;
3350 advance = 0;
David Woodhouse49593bf2008-08-17 17:08:36 +01003351
3352 while (1) {
Chris Mason5f39d392007-10-15 16:14:19 -04003353 leaf = path->nodes[0];
3354 nritems = btrfs_header_nritems(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04003355 slot = path->slots[0];
3356 if (advance || slot >= nritems) {
David Woodhouse49593bf2008-08-17 17:08:36 +01003357 if (slot >= nritems - 1) {
Chris Mason39279cc2007-06-12 06:35:45 -04003358 ret = btrfs_next_leaf(root, path);
3359 if (ret)
3360 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003361 leaf = path->nodes[0];
3362 nritems = btrfs_header_nritems(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04003363 slot = path->slots[0];
3364 } else {
3365 slot++;
3366 path->slots[0]++;
3367 }
3368 }
Chris Mason3de45862008-11-17 21:02:50 -05003369
Chris Mason39279cc2007-06-12 06:35:45 -04003370 advance = 1;
Chris Mason5f39d392007-10-15 16:14:19 -04003371 item = btrfs_item_nr(leaf, slot);
3372 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3373
3374 if (found_key.objectid != key.objectid)
Chris Mason39279cc2007-06-12 06:35:45 -04003375 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003376 if (btrfs_key_type(&found_key) != key_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003377 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003378 if (found_key.offset < filp->f_pos)
Chris Mason39279cc2007-06-12 06:35:45 -04003379 continue;
Chris Mason5f39d392007-10-15 16:14:19 -04003380
3381 filp->f_pos = found_key.offset;
David Woodhouse49593bf2008-08-17 17:08:36 +01003382
Chris Mason39279cc2007-06-12 06:35:45 -04003383 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3384 di_cur = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003385 di_total = btrfs_item_size(leaf, item);
David Woodhouse49593bf2008-08-17 17:08:36 +01003386
3387 while (di_cur < di_total) {
Chris Mason5f39d392007-10-15 16:14:19 -04003388 struct btrfs_key location;
3389
3390 name_len = btrfs_dir_name_len(leaf, di);
David Woodhouse49593bf2008-08-17 17:08:36 +01003391 if (name_len <= sizeof(tmp_name)) {
Chris Mason5f39d392007-10-15 16:14:19 -04003392 name_ptr = tmp_name;
3393 } else {
3394 name_ptr = kmalloc(name_len, GFP_NOFS);
David Woodhouse49593bf2008-08-17 17:08:36 +01003395 if (!name_ptr) {
3396 ret = -ENOMEM;
3397 goto err;
3398 }
Chris Mason5f39d392007-10-15 16:14:19 -04003399 }
3400 read_extent_buffer(leaf, name_ptr,
3401 (unsigned long)(di + 1), name_len);
3402
3403 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3404 btrfs_dir_item_key_to_cpu(leaf, di, &location);
Chris Mason3de45862008-11-17 21:02:50 -05003405
3406 /* is this a reference to our own snapshot? If so
3407 * skip it
3408 */
3409 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3410 location.objectid == root->root_key.objectid) {
3411 over = 0;
3412 goto skip;
3413 }
Chris Mason5f39d392007-10-15 16:14:19 -04003414 over = filldir(dirent, name_ptr, name_len,
David Woodhouse49593bf2008-08-17 17:08:36 +01003415 found_key.offset, location.objectid,
Chris Mason39279cc2007-06-12 06:35:45 -04003416 d_type);
Chris Mason5f39d392007-10-15 16:14:19 -04003417
Chris Mason3de45862008-11-17 21:02:50 -05003418skip:
Chris Mason5f39d392007-10-15 16:14:19 -04003419 if (name_ptr != tmp_name)
3420 kfree(name_ptr);
3421
Chris Mason39279cc2007-06-12 06:35:45 -04003422 if (over)
3423 goto nopos;
Josef Bacik5103e942007-11-16 11:45:54 -05003424 di_len = btrfs_dir_name_len(leaf, di) +
David Woodhouse49593bf2008-08-17 17:08:36 +01003425 btrfs_dir_data_len(leaf, di) + sizeof(*di);
Chris Mason39279cc2007-06-12 06:35:45 -04003426 di_cur += di_len;
3427 di = (struct btrfs_dir_item *)((char *)di + di_len);
3428 }
3429 }
David Woodhouse49593bf2008-08-17 17:08:36 +01003430
3431 /* Reached end of directory/root. Bump pos past the last item. */
Yan Zheng5e591a02008-02-19 11:41:02 -05003432 if (key_type == BTRFS_DIR_INDEX_KEY)
Chris Mason89f135d2009-01-28 15:34:27 -05003433 filp->f_pos = INT_LIMIT(off_t);
Yan Zheng5e591a02008-02-19 11:41:02 -05003434 else
3435 filp->f_pos++;
Chris Mason39279cc2007-06-12 06:35:45 -04003436nopos:
3437 ret = 0;
3438err:
Chris Mason39279cc2007-06-12 06:35:45 -04003439 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003440 return ret;
3441}
3442
3443int btrfs_write_inode(struct inode *inode, int wait)
3444{
3445 struct btrfs_root *root = BTRFS_I(inode)->root;
3446 struct btrfs_trans_handle *trans;
3447 int ret = 0;
3448
Yan Zhengc146afa2008-11-12 14:34:12 -05003449 if (root->fs_info->btree_inode == inode)
Chris Mason4ca8b412008-08-05 13:30:48 -04003450 return 0;
3451
Chris Mason39279cc2007-06-12 06:35:45 -04003452 if (wait) {
Chris Masonf9295742008-07-17 12:54:14 -04003453 trans = btrfs_join_transaction(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04003454 btrfs_set_trans_block_group(trans, inode);
3455 ret = btrfs_commit_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04003456 }
3457 return ret;
3458}
3459
3460/*
Chris Mason54aa1f42007-06-22 14:16:25 -04003461 * This is somewhat expensive, updating the tree every time the
Chris Mason39279cc2007-06-12 06:35:45 -04003462 * inode changes. But, it is most likely to find the inode in cache.
3463 * FIXME, needs more benchmarking...there are no reasons other than performance
3464 * to keep or drop this code.
3465 */
3466void btrfs_dirty_inode(struct inode *inode)
3467{
3468 struct btrfs_root *root = BTRFS_I(inode)->root;
3469 struct btrfs_trans_handle *trans;
3470
Chris Masonf9295742008-07-17 12:54:14 -04003471 trans = btrfs_join_transaction(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04003472 btrfs_set_trans_block_group(trans, inode);
3473 btrfs_update_inode(trans, root, inode);
3474 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04003475}
3476
Chris Masond352ac62008-09-29 15:18:18 -04003477/*
3478 * find the highest existing sequence number in a directory
3479 * and then set the in-memory index_cnt variable to reflect
3480 * free sequence numbers
3481 */
Josef Bacikaec74772008-07-24 12:12:38 -04003482static int btrfs_set_inode_index_count(struct inode *inode)
3483{
3484 struct btrfs_root *root = BTRFS_I(inode)->root;
3485 struct btrfs_key key, found_key;
3486 struct btrfs_path *path;
3487 struct extent_buffer *leaf;
3488 int ret;
3489
3490 key.objectid = inode->i_ino;
3491 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3492 key.offset = (u64)-1;
3493
3494 path = btrfs_alloc_path();
3495 if (!path)
3496 return -ENOMEM;
3497
3498 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3499 if (ret < 0)
3500 goto out;
3501 /* FIXME: we should be able to handle this */
3502 if (ret == 0)
3503 goto out;
3504 ret = 0;
3505
3506 /*
3507 * MAGIC NUMBER EXPLANATION:
3508 * since we search a directory based on f_pos we have to start at 2
3509 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3510 * else has to start at 2
3511 */
3512 if (path->slots[0] == 0) {
3513 BTRFS_I(inode)->index_cnt = 2;
3514 goto out;
3515 }
3516
3517 path->slots[0]--;
3518
3519 leaf = path->nodes[0];
3520 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3521
3522 if (found_key.objectid != inode->i_ino ||
3523 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3524 BTRFS_I(inode)->index_cnt = 2;
3525 goto out;
3526 }
3527
3528 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3529out:
3530 btrfs_free_path(path);
3531 return ret;
3532}
3533
Chris Masond352ac62008-09-29 15:18:18 -04003534/*
3535 * helper to find a free sequence number in a given directory. This current
3536 * code is very simple, later versions will do smarter things in the btree
3537 */
Chris Mason3de45862008-11-17 21:02:50 -05003538int btrfs_set_inode_index(struct inode *dir, u64 *index)
Josef Bacikaec74772008-07-24 12:12:38 -04003539{
3540 int ret = 0;
3541
3542 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3543 ret = btrfs_set_inode_index_count(dir);
Chris Masond3977122009-01-05 21:25:51 -05003544 if (ret)
Josef Bacikaec74772008-07-24 12:12:38 -04003545 return ret;
3546 }
3547
Chris Mason00e4e6b2008-08-05 11:18:09 -04003548 *index = BTRFS_I(dir)->index_cnt;
Josef Bacikaec74772008-07-24 12:12:38 -04003549 BTRFS_I(dir)->index_cnt++;
3550
3551 return ret;
3552}
3553
Chris Mason39279cc2007-06-12 06:35:45 -04003554static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3555 struct btrfs_root *root,
Josef Bacikaec74772008-07-24 12:12:38 -04003556 struct inode *dir,
Chris Mason9c583092008-01-29 15:15:18 -05003557 const char *name, int name_len,
Yan Zhengd2fb3432008-12-11 16:30:39 -05003558 u64 ref_objectid, u64 objectid,
3559 u64 alloc_hint, int mode, u64 *index)
Chris Mason39279cc2007-06-12 06:35:45 -04003560{
3561 struct inode *inode;
Chris Mason5f39d392007-10-15 16:14:19 -04003562 struct btrfs_inode_item *inode_item;
Chris Mason39279cc2007-06-12 06:35:45 -04003563 struct btrfs_key *location;
Chris Mason5f39d392007-10-15 16:14:19 -04003564 struct btrfs_path *path;
Chris Mason9c583092008-01-29 15:15:18 -05003565 struct btrfs_inode_ref *ref;
3566 struct btrfs_key key[2];
3567 u32 sizes[2];
3568 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04003569 int ret;
3570 int owner;
3571
Chris Mason5f39d392007-10-15 16:14:19 -04003572 path = btrfs_alloc_path();
3573 BUG_ON(!path);
3574
Chris Mason39279cc2007-06-12 06:35:45 -04003575 inode = new_inode(root->fs_info->sb);
3576 if (!inode)
3577 return ERR_PTR(-ENOMEM);
3578
Josef Bacikaec74772008-07-24 12:12:38 -04003579 if (dir) {
Chris Mason3de45862008-11-17 21:02:50 -05003580 ret = btrfs_set_inode_index(dir, index);
Shen Feng09771432009-04-02 16:46:06 -04003581 if (ret) {
3582 iput(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04003583 return ERR_PTR(ret);
Shen Feng09771432009-04-02 16:46:06 -04003584 }
Josef Bacikaec74772008-07-24 12:12:38 -04003585 }
3586 /*
3587 * index_cnt is ignored for everything but a dir,
3588 * btrfs_get_inode_index_count has an explanation for the magic
3589 * number
3590 */
Chris Masone02119d2008-09-05 16:13:11 -04003591 init_btrfs_i(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04003592 BTRFS_I(inode)->index_cnt = 2;
Chris Mason39279cc2007-06-12 06:35:45 -04003593 BTRFS_I(inode)->root = root;
Chris Masone02119d2008-09-05 16:13:11 -04003594 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik6a632092009-02-20 11:00:09 -05003595 btrfs_set_inode_space_info(root, inode);
Chris Masonb888db22007-08-27 16:49:44 -04003596
Chris Mason39279cc2007-06-12 06:35:45 -04003597 if (mode & S_IFDIR)
3598 owner = 0;
3599 else
3600 owner = 1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05003601 BTRFS_I(inode)->block_group =
3602 btrfs_find_block_group(root, 0, alloc_hint, owner);
Chris Mason9c583092008-01-29 15:15:18 -05003603
3604 key[0].objectid = objectid;
3605 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3606 key[0].offset = 0;
3607
3608 key[1].objectid = objectid;
3609 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3610 key[1].offset = ref_objectid;
3611
3612 sizes[0] = sizeof(struct btrfs_inode_item);
3613 sizes[1] = name_len + sizeof(*ref);
3614
Chris Masonb9473432009-03-13 11:00:37 -04003615 path->leave_spinning = 1;
Chris Mason9c583092008-01-29 15:15:18 -05003616 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3617 if (ret != 0)
Chris Mason5f39d392007-10-15 16:14:19 -04003618 goto fail;
3619
Chris Mason9c583092008-01-29 15:15:18 -05003620 if (objectid > root->highest_inode)
3621 root->highest_inode = objectid;
3622
Chris Mason79683f22008-11-19 22:00:53 -05003623 inode->i_uid = current_fsuid();
Chris Ball8c087b52009-02-04 09:29:54 -05003624
Chris Mason42f15d72009-02-06 11:35:57 -05003625 if (dir && (dir->i_mode & S_ISGID)) {
Chris Ball8c087b52009-02-04 09:29:54 -05003626 inode->i_gid = dir->i_gid;
3627 if (S_ISDIR(mode))
3628 mode |= S_ISGID;
3629 } else
3630 inode->i_gid = current_fsgid();
3631
Chris Mason39279cc2007-06-12 06:35:45 -04003632 inode->i_mode = mode;
3633 inode->i_ino = objectid;
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003634 inode_set_bytes(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04003635 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
Chris Mason5f39d392007-10-15 16:14:19 -04003636 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3637 struct btrfs_inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04003638 fill_inode_item(trans, path->nodes[0], inode_item, inode);
Chris Mason9c583092008-01-29 15:15:18 -05003639
3640 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3641 struct btrfs_inode_ref);
3642 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
Chris Mason00e4e6b2008-08-05 11:18:09 -04003643 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
Chris Mason9c583092008-01-29 15:15:18 -05003644 ptr = (unsigned long)(ref + 1);
3645 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3646
Chris Mason5f39d392007-10-15 16:14:19 -04003647 btrfs_mark_buffer_dirty(path->nodes[0]);
3648 btrfs_free_path(path);
3649
Chris Mason39279cc2007-06-12 06:35:45 -04003650 location = &BTRFS_I(inode)->location;
3651 location->objectid = objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04003652 location->offset = 0;
3653 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3654
Christoph Hellwig6cbff002009-04-17 10:37:41 +02003655 btrfs_inherit_iflags(inode, dir);
3656
Chris Mason94272162009-07-02 12:26:06 -04003657 if ((mode & S_IFREG)) {
3658 if (btrfs_test_opt(root, NODATASUM))
3659 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
3660 if (btrfs_test_opt(root, NODATACOW))
3661 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
3662 }
3663
Chris Mason39279cc2007-06-12 06:35:45 -04003664 insert_inode_hash(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003665 inode_tree_add(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003666 return inode;
Chris Mason5f39d392007-10-15 16:14:19 -04003667fail:
Josef Bacikaec74772008-07-24 12:12:38 -04003668 if (dir)
3669 BTRFS_I(dir)->index_cnt--;
Chris Mason5f39d392007-10-15 16:14:19 -04003670 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04003671 iput(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04003672 return ERR_PTR(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04003673}
3674
3675static inline u8 btrfs_inode_type(struct inode *inode)
3676{
3677 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3678}
3679
Chris Masond352ac62008-09-29 15:18:18 -04003680/*
3681 * utility function to add 'inode' into 'parent_inode' with
3682 * a give name and a given sequence number.
3683 * if 'add_backref' is true, also insert a backref from the
3684 * inode to the parent directory.
3685 */
Chris Masone02119d2008-09-05 16:13:11 -04003686int btrfs_add_link(struct btrfs_trans_handle *trans,
3687 struct inode *parent_inode, struct inode *inode,
3688 const char *name, int name_len, int add_backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04003689{
3690 int ret;
3691 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04003692 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
Chris Mason5f39d392007-10-15 16:14:19 -04003693
Chris Mason39279cc2007-06-12 06:35:45 -04003694 key.objectid = inode->i_ino;
Chris Mason39279cc2007-06-12 06:35:45 -04003695 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3696 key.offset = 0;
3697
Chris Masone02119d2008-09-05 16:13:11 -04003698 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3699 parent_inode->i_ino,
Josef Bacikaec74772008-07-24 12:12:38 -04003700 &key, btrfs_inode_type(inode),
Chris Mason00e4e6b2008-08-05 11:18:09 -04003701 index);
Chris Mason39279cc2007-06-12 06:35:45 -04003702 if (ret == 0) {
Chris Mason9c583092008-01-29 15:15:18 -05003703 if (add_backref) {
3704 ret = btrfs_insert_inode_ref(trans, root,
Chris Masone02119d2008-09-05 16:13:11 -04003705 name, name_len,
3706 inode->i_ino,
3707 parent_inode->i_ino,
3708 index);
Chris Mason9c583092008-01-29 15:15:18 -05003709 }
Chris Masondbe674a2008-07-17 12:54:05 -04003710 btrfs_i_size_write(parent_inode, parent_inode->i_size +
Chris Masone02119d2008-09-05 16:13:11 -04003711 name_len * 2);
Chris Mason79c44582007-06-25 10:09:33 -04003712 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
Chris Masone02119d2008-09-05 16:13:11 -04003713 ret = btrfs_update_inode(trans, root, parent_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003714 }
3715 return ret;
3716}
3717
3718static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
Chris Mason9c583092008-01-29 15:15:18 -05003719 struct dentry *dentry, struct inode *inode,
Chris Mason00e4e6b2008-08-05 11:18:09 -04003720 int backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04003721{
Chris Masone02119d2008-09-05 16:13:11 -04003722 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3723 inode, dentry->d_name.name,
3724 dentry->d_name.len, backref, index);
Chris Mason39279cc2007-06-12 06:35:45 -04003725 if (!err) {
3726 d_instantiate(dentry, inode);
3727 return 0;
3728 }
3729 if (err > 0)
3730 err = -EEXIST;
3731 return err;
3732}
3733
Josef Bacik618e21d2007-07-11 10:18:17 -04003734static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3735 int mode, dev_t rdev)
3736{
3737 struct btrfs_trans_handle *trans;
3738 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05003739 struct inode *inode = NULL;
Josef Bacik618e21d2007-07-11 10:18:17 -04003740 int err;
3741 int drop_inode = 0;
3742 u64 objectid;
Chris Mason1832a6d2007-12-21 16:27:21 -05003743 unsigned long nr = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04003744 u64 index = 0;
Josef Bacik618e21d2007-07-11 10:18:17 -04003745
3746 if (!new_valid_dev(rdev))
3747 return -EINVAL;
3748
Josef Bacik6a632092009-02-20 11:00:09 -05003749 err = btrfs_check_metadata_free_space(root);
Chris Mason1832a6d2007-12-21 16:27:21 -05003750 if (err)
3751 goto fail;
3752
Josef Bacik618e21d2007-07-11 10:18:17 -04003753 trans = btrfs_start_transaction(root, 1);
3754 btrfs_set_trans_block_group(trans, dir);
3755
3756 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3757 if (err) {
3758 err = -ENOSPC;
3759 goto out_unlock;
3760 }
3761
Josef Bacikaec74772008-07-24 12:12:38 -04003762 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Chris Mason9c583092008-01-29 15:15:18 -05003763 dentry->d_name.len,
3764 dentry->d_parent->d_inode->i_ino, objectid,
Chris Mason00e4e6b2008-08-05 11:18:09 -04003765 BTRFS_I(dir)->block_group, mode, &index);
Josef Bacik618e21d2007-07-11 10:18:17 -04003766 err = PTR_ERR(inode);
3767 if (IS_ERR(inode))
3768 goto out_unlock;
3769
Jim Owens0279b4c2009-02-04 09:29:13 -05003770 err = btrfs_init_inode_security(inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04003771 if (err) {
3772 drop_inode = 1;
3773 goto out_unlock;
3774 }
3775
Josef Bacik618e21d2007-07-11 10:18:17 -04003776 btrfs_set_trans_block_group(trans, inode);
Chris Mason00e4e6b2008-08-05 11:18:09 -04003777 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
Josef Bacik618e21d2007-07-11 10:18:17 -04003778 if (err)
3779 drop_inode = 1;
3780 else {
3781 inode->i_op = &btrfs_special_inode_operations;
3782 init_special_inode(inode, inode->i_mode, rdev);
Yan1b4ab1b2007-08-29 09:11:44 -04003783 btrfs_update_inode(trans, root, inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04003784 }
3785 dir->i_sb->s_dirt = 1;
3786 btrfs_update_inode_block_group(trans, inode);
3787 btrfs_update_inode_block_group(trans, dir);
3788out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003789 nr = trans->blocks_used;
Chris Mason89ce8a62008-06-25 16:01:31 -04003790 btrfs_end_transaction_throttle(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05003791fail:
Josef Bacik618e21d2007-07-11 10:18:17 -04003792 if (drop_inode) {
3793 inode_dec_link_count(inode);
3794 iput(inode);
3795 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003796 btrfs_btree_balance_dirty(root, nr);
Josef Bacik618e21d2007-07-11 10:18:17 -04003797 return err;
3798}
3799
Chris Mason39279cc2007-06-12 06:35:45 -04003800static int btrfs_create(struct inode *dir, struct dentry *dentry,
3801 int mode, struct nameidata *nd)
3802{
3803 struct btrfs_trans_handle *trans;
3804 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05003805 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04003806 int err;
3807 int drop_inode = 0;
Chris Mason1832a6d2007-12-21 16:27:21 -05003808 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003809 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04003810 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003811
Josef Bacik6a632092009-02-20 11:00:09 -05003812 err = btrfs_check_metadata_free_space(root);
Chris Mason1832a6d2007-12-21 16:27:21 -05003813 if (err)
3814 goto fail;
Chris Mason39279cc2007-06-12 06:35:45 -04003815 trans = btrfs_start_transaction(root, 1);
3816 btrfs_set_trans_block_group(trans, dir);
3817
3818 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3819 if (err) {
3820 err = -ENOSPC;
3821 goto out_unlock;
3822 }
3823
Josef Bacikaec74772008-07-24 12:12:38 -04003824 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Chris Mason9c583092008-01-29 15:15:18 -05003825 dentry->d_name.len,
3826 dentry->d_parent->d_inode->i_ino,
Chris Mason00e4e6b2008-08-05 11:18:09 -04003827 objectid, BTRFS_I(dir)->block_group, mode,
3828 &index);
Chris Mason39279cc2007-06-12 06:35:45 -04003829 err = PTR_ERR(inode);
3830 if (IS_ERR(inode))
3831 goto out_unlock;
3832
Jim Owens0279b4c2009-02-04 09:29:13 -05003833 err = btrfs_init_inode_security(inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04003834 if (err) {
3835 drop_inode = 1;
3836 goto out_unlock;
3837 }
3838
Chris Mason39279cc2007-06-12 06:35:45 -04003839 btrfs_set_trans_block_group(trans, inode);
Chris Mason00e4e6b2008-08-05 11:18:09 -04003840 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04003841 if (err)
3842 drop_inode = 1;
3843 else {
3844 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04003845 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04003846 inode->i_fop = &btrfs_file_operations;
3847 inode->i_op = &btrfs_file_inode_operations;
Chris Masond1310b22008-01-24 16:13:08 -05003848 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04003849 }
3850 dir->i_sb->s_dirt = 1;
3851 btrfs_update_inode_block_group(trans, inode);
3852 btrfs_update_inode_block_group(trans, dir);
3853out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003854 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04003855 btrfs_end_transaction_throttle(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05003856fail:
Chris Mason39279cc2007-06-12 06:35:45 -04003857 if (drop_inode) {
3858 inode_dec_link_count(inode);
3859 iput(inode);
3860 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003861 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04003862 return err;
3863}
3864
3865static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3866 struct dentry *dentry)
3867{
3868 struct btrfs_trans_handle *trans;
3869 struct btrfs_root *root = BTRFS_I(dir)->root;
3870 struct inode *inode = old_dentry->d_inode;
Chris Mason00e4e6b2008-08-05 11:18:09 -04003871 u64 index;
Chris Mason1832a6d2007-12-21 16:27:21 -05003872 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003873 int err;
3874 int drop_inode = 0;
3875
3876 if (inode->i_nlink == 0)
3877 return -ENOENT;
3878
Chris Masone02119d2008-09-05 16:13:11 -04003879 btrfs_inc_nlink(inode);
Josef Bacik6a632092009-02-20 11:00:09 -05003880 err = btrfs_check_metadata_free_space(root);
Chris Mason1832a6d2007-12-21 16:27:21 -05003881 if (err)
3882 goto fail;
Chris Mason3de45862008-11-17 21:02:50 -05003883 err = btrfs_set_inode_index(dir, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04003884 if (err)
3885 goto fail;
3886
Chris Mason39279cc2007-06-12 06:35:45 -04003887 trans = btrfs_start_transaction(root, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04003888
Chris Mason39279cc2007-06-12 06:35:45 -04003889 btrfs_set_trans_block_group(trans, dir);
3890 atomic_inc(&inode->i_count);
Josef Bacikaec74772008-07-24 12:12:38 -04003891
Chris Mason00e4e6b2008-08-05 11:18:09 -04003892 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
Chris Mason5f39d392007-10-15 16:14:19 -04003893
Chris Mason39279cc2007-06-12 06:35:45 -04003894 if (err)
3895 drop_inode = 1;
Chris Mason5f39d392007-10-15 16:14:19 -04003896
Chris Mason39279cc2007-06-12 06:35:45 -04003897 dir->i_sb->s_dirt = 1;
3898 btrfs_update_inode_block_group(trans, dir);
Chris Mason54aa1f42007-06-22 14:16:25 -04003899 err = btrfs_update_inode(trans, root, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04003900
Chris Mason54aa1f42007-06-22 14:16:25 -04003901 if (err)
3902 drop_inode = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04003903
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003904 nr = trans->blocks_used;
Chris Mason12fcfd22009-03-24 10:24:20 -04003905
3906 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
Chris Masonab78c842008-07-29 16:15:18 -04003907 btrfs_end_transaction_throttle(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05003908fail:
Chris Mason39279cc2007-06-12 06:35:45 -04003909 if (drop_inode) {
3910 inode_dec_link_count(inode);
3911 iput(inode);
3912 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003913 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04003914 return err;
3915}
3916
Chris Mason39279cc2007-06-12 06:35:45 -04003917static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3918{
Chris Masonb9d86662008-05-02 16:13:49 -04003919 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04003920 struct btrfs_trans_handle *trans;
3921 struct btrfs_root *root = BTRFS_I(dir)->root;
3922 int err = 0;
3923 int drop_on_err = 0;
Chris Masonb9d86662008-05-02 16:13:49 -04003924 u64 objectid = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04003925 u64 index = 0;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003926 unsigned long nr = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04003927
Josef Bacik6a632092009-02-20 11:00:09 -05003928 err = btrfs_check_metadata_free_space(root);
Chris Mason1832a6d2007-12-21 16:27:21 -05003929 if (err)
3930 goto out_unlock;
3931
Chris Mason39279cc2007-06-12 06:35:45 -04003932 trans = btrfs_start_transaction(root, 1);
3933 btrfs_set_trans_block_group(trans, dir);
Chris Mason5f39d392007-10-15 16:14:19 -04003934
Chris Mason39279cc2007-06-12 06:35:45 -04003935 if (IS_ERR(trans)) {
3936 err = PTR_ERR(trans);
3937 goto out_unlock;
3938 }
3939
3940 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3941 if (err) {
3942 err = -ENOSPC;
3943 goto out_unlock;
3944 }
3945
Josef Bacikaec74772008-07-24 12:12:38 -04003946 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Chris Mason9c583092008-01-29 15:15:18 -05003947 dentry->d_name.len,
3948 dentry->d_parent->d_inode->i_ino, objectid,
Chris Mason00e4e6b2008-08-05 11:18:09 -04003949 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3950 &index);
Chris Mason39279cc2007-06-12 06:35:45 -04003951 if (IS_ERR(inode)) {
3952 err = PTR_ERR(inode);
3953 goto out_fail;
3954 }
Chris Mason5f39d392007-10-15 16:14:19 -04003955
Chris Mason39279cc2007-06-12 06:35:45 -04003956 drop_on_err = 1;
Josef Bacik33268ea2008-07-24 12:16:36 -04003957
Jim Owens0279b4c2009-02-04 09:29:13 -05003958 err = btrfs_init_inode_security(inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04003959 if (err)
3960 goto out_fail;
3961
Chris Mason39279cc2007-06-12 06:35:45 -04003962 inode->i_op = &btrfs_dir_inode_operations;
3963 inode->i_fop = &btrfs_dir_file_operations;
3964 btrfs_set_trans_block_group(trans, inode);
3965
Chris Masondbe674a2008-07-17 12:54:05 -04003966 btrfs_i_size_write(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04003967 err = btrfs_update_inode(trans, root, inode);
3968 if (err)
3969 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04003970
Chris Masone02119d2008-09-05 16:13:11 -04003971 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3972 inode, dentry->d_name.name,
3973 dentry->d_name.len, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04003974 if (err)
3975 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04003976
Chris Mason39279cc2007-06-12 06:35:45 -04003977 d_instantiate(dentry, inode);
3978 drop_on_err = 0;
3979 dir->i_sb->s_dirt = 1;
3980 btrfs_update_inode_block_group(trans, inode);
3981 btrfs_update_inode_block_group(trans, dir);
3982
3983out_fail:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003984 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04003985 btrfs_end_transaction_throttle(trans, root);
Chris Mason5f39d392007-10-15 16:14:19 -04003986
Chris Mason39279cc2007-06-12 06:35:45 -04003987out_unlock:
Chris Mason39279cc2007-06-12 06:35:45 -04003988 if (drop_on_err)
3989 iput(inode);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003990 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04003991 return err;
3992}
3993
Chris Masond352ac62008-09-29 15:18:18 -04003994/* helper for btfs_get_extent. Given an existing extent in the tree,
3995 * and an extent that you want to insert, deal with overlap and insert
3996 * the new extent into the tree.
3997 */
Chris Mason3b951512008-04-17 11:29:12 -04003998static int merge_extent_mapping(struct extent_map_tree *em_tree,
3999 struct extent_map *existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04004000 struct extent_map *em,
4001 u64 map_start, u64 map_len)
Chris Mason3b951512008-04-17 11:29:12 -04004002{
4003 u64 start_diff;
Chris Mason3b951512008-04-17 11:29:12 -04004004
Chris Masone6dcd2d2008-07-17 12:53:50 -04004005 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4006 start_diff = map_start - em->start;
4007 em->start = map_start;
4008 em->len = map_len;
Chris Masonc8b97812008-10-29 14:49:59 -04004009 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4010 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04004011 em->block_start += start_diff;
Chris Masonc8b97812008-10-29 14:49:59 -04004012 em->block_len -= start_diff;
4013 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004014 return add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04004015}
4016
Chris Masonc8b97812008-10-29 14:49:59 -04004017static noinline int uncompress_inline(struct btrfs_path *path,
4018 struct inode *inode, struct page *page,
4019 size_t pg_offset, u64 extent_offset,
4020 struct btrfs_file_extent_item *item)
4021{
4022 int ret;
4023 struct extent_buffer *leaf = path->nodes[0];
4024 char *tmp;
4025 size_t max_size;
4026 unsigned long inline_size;
4027 unsigned long ptr;
4028
4029 WARN_ON(pg_offset != 0);
4030 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4031 inline_size = btrfs_file_extent_inline_item_len(leaf,
4032 btrfs_item_nr(leaf, path->slots[0]));
4033 tmp = kmalloc(inline_size, GFP_NOFS);
4034 ptr = btrfs_file_extent_inline_start(item);
4035
4036 read_extent_buffer(leaf, tmp, ptr, inline_size);
4037
Chris Mason5b050f02008-11-11 09:34:41 -05004038 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
Chris Masonc8b97812008-10-29 14:49:59 -04004039 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4040 inline_size, max_size);
4041 if (ret) {
4042 char *kaddr = kmap_atomic(page, KM_USER0);
4043 unsigned long copy_size = min_t(u64,
4044 PAGE_CACHE_SIZE - pg_offset,
4045 max_size - extent_offset);
4046 memset(kaddr + pg_offset, 0, copy_size);
4047 kunmap_atomic(kaddr, KM_USER0);
4048 }
4049 kfree(tmp);
4050 return 0;
4051}
4052
Chris Masond352ac62008-09-29 15:18:18 -04004053/*
4054 * a bit scary, this does extent mapping from logical file offset to the disk.
Chris Masond3977122009-01-05 21:25:51 -05004055 * the ugly parts come from merging extents from the disk with the in-ram
4056 * representation. This gets more complex because of the data=ordered code,
Chris Masond352ac62008-09-29 15:18:18 -04004057 * where the in-ram extents might be locked pending data=ordered completion.
4058 *
4059 * This also copies inline extents directly into the page.
4060 */
Chris Masond3977122009-01-05 21:25:51 -05004061
Chris Masona52d9a82007-08-27 16:49:44 -04004062struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
Chris Mason70dec802008-01-29 09:59:12 -05004063 size_t pg_offset, u64 start, u64 len,
Chris Masona52d9a82007-08-27 16:49:44 -04004064 int create)
4065{
4066 int ret;
4067 int err = 0;
Chris Masondb945352007-10-15 16:15:53 -04004068 u64 bytenr;
Chris Masona52d9a82007-08-27 16:49:44 -04004069 u64 extent_start = 0;
4070 u64 extent_end = 0;
4071 u64 objectid = inode->i_ino;
4072 u32 found_type;
Chris Masonf4219502008-07-22 11:18:09 -04004073 struct btrfs_path *path = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -04004074 struct btrfs_root *root = BTRFS_I(inode)->root;
4075 struct btrfs_file_extent_item *item;
Chris Mason5f39d392007-10-15 16:14:19 -04004076 struct extent_buffer *leaf;
4077 struct btrfs_key found_key;
Chris Masona52d9a82007-08-27 16:49:44 -04004078 struct extent_map *em = NULL;
4079 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Masond1310b22008-01-24 16:13:08 -05004080 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04004081 struct btrfs_trans_handle *trans = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -04004082 int compressed;
Chris Masona52d9a82007-08-27 16:49:44 -04004083
Chris Masona52d9a82007-08-27 16:49:44 -04004084again:
Chris Mason890871b2009-09-02 16:24:52 -04004085 read_lock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05004086 em = lookup_extent_mapping(em_tree, start, len);
Chris Masona061fc82008-05-07 11:43:44 -04004087 if (em)
4088 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason890871b2009-09-02 16:24:52 -04004089 read_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05004090
Chris Masona52d9a82007-08-27 16:49:44 -04004091 if (em) {
Chris Masone1c4b742008-04-22 13:26:46 -04004092 if (em->start > start || em->start + em->len <= start)
4093 free_extent_map(em);
4094 else if (em->block_start == EXTENT_MAP_INLINE && page)
Chris Mason70dec802008-01-29 09:59:12 -05004095 free_extent_map(em);
4096 else
4097 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004098 }
Chris Masond1310b22008-01-24 16:13:08 -05004099 em = alloc_extent_map(GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04004100 if (!em) {
Chris Masond1310b22008-01-24 16:13:08 -05004101 err = -ENOMEM;
4102 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004103 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004104 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Masond1310b22008-01-24 16:13:08 -05004105 em->start = EXTENT_MAP_HOLE;
Chris Mason445a6942008-11-10 11:53:33 -05004106 em->orig_start = EXTENT_MAP_HOLE;
Chris Masond1310b22008-01-24 16:13:08 -05004107 em->len = (u64)-1;
Chris Masonc8b97812008-10-29 14:49:59 -04004108 em->block_len = (u64)-1;
Chris Masonf4219502008-07-22 11:18:09 -04004109
4110 if (!path) {
4111 path = btrfs_alloc_path();
4112 BUG_ON(!path);
4113 }
4114
Chris Mason179e29e2007-11-01 11:28:41 -04004115 ret = btrfs_lookup_file_extent(trans, root, path,
4116 objectid, start, trans != NULL);
Chris Masona52d9a82007-08-27 16:49:44 -04004117 if (ret < 0) {
4118 err = ret;
4119 goto out;
4120 }
4121
4122 if (ret != 0) {
4123 if (path->slots[0] == 0)
4124 goto not_found;
4125 path->slots[0]--;
4126 }
4127
Chris Mason5f39d392007-10-15 16:14:19 -04004128 leaf = path->nodes[0];
4129 item = btrfs_item_ptr(leaf, path->slots[0],
Chris Masona52d9a82007-08-27 16:49:44 -04004130 struct btrfs_file_extent_item);
Chris Masona52d9a82007-08-27 16:49:44 -04004131 /* are we inside the extent that was found? */
Chris Mason5f39d392007-10-15 16:14:19 -04004132 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4133 found_type = btrfs_key_type(&found_key);
4134 if (found_key.objectid != objectid ||
Chris Masona52d9a82007-08-27 16:49:44 -04004135 found_type != BTRFS_EXTENT_DATA_KEY) {
4136 goto not_found;
4137 }
4138
Chris Mason5f39d392007-10-15 16:14:19 -04004139 found_type = btrfs_file_extent_type(leaf, item);
4140 extent_start = found_key.offset;
Chris Masonc8b97812008-10-29 14:49:59 -04004141 compressed = btrfs_file_extent_compression(leaf, item);
Yan Zhengd899e052008-10-30 14:25:28 -04004142 if (found_type == BTRFS_FILE_EXTENT_REG ||
4143 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04004144 extent_end = extent_start +
Chris Masondb945352007-10-15 16:15:53 -04004145 btrfs_file_extent_num_bytes(leaf, item);
Yan Zheng9036c102008-10-30 14:19:41 -04004146 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4147 size_t size;
4148 size = btrfs_file_extent_inline_len(leaf, item);
4149 extent_end = (extent_start + size + root->sectorsize - 1) &
4150 ~((u64)root->sectorsize - 1);
4151 }
4152
4153 if (start >= extent_end) {
4154 path->slots[0]++;
4155 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4156 ret = btrfs_next_leaf(root, path);
4157 if (ret < 0) {
4158 err = ret;
4159 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004160 }
Yan Zheng9036c102008-10-30 14:19:41 -04004161 if (ret > 0)
4162 goto not_found;
4163 leaf = path->nodes[0];
Chris Masona52d9a82007-08-27 16:49:44 -04004164 }
Yan Zheng9036c102008-10-30 14:19:41 -04004165 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4166 if (found_key.objectid != objectid ||
4167 found_key.type != BTRFS_EXTENT_DATA_KEY)
4168 goto not_found;
4169 if (start + len <= found_key.offset)
4170 goto not_found;
4171 em->start = start;
4172 em->len = found_key.offset - start;
4173 goto not_found_em;
4174 }
4175
Yan Zhengd899e052008-10-30 14:25:28 -04004176 if (found_type == BTRFS_FILE_EXTENT_REG ||
4177 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng9036c102008-10-30 14:19:41 -04004178 em->start = extent_start;
4179 em->len = extent_end - extent_start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -05004180 em->orig_start = extent_start -
4181 btrfs_file_extent_offset(leaf, item);
Chris Masondb945352007-10-15 16:15:53 -04004182 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4183 if (bytenr == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04004184 em->block_start = EXTENT_MAP_HOLE;
Chris Masona52d9a82007-08-27 16:49:44 -04004185 goto insert;
4186 }
Chris Masonc8b97812008-10-29 14:49:59 -04004187 if (compressed) {
4188 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4189 em->block_start = bytenr;
4190 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4191 item);
4192 } else {
4193 bytenr += btrfs_file_extent_offset(leaf, item);
4194 em->block_start = bytenr;
4195 em->block_len = em->len;
Yan Zhengd899e052008-10-30 14:25:28 -04004196 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4197 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -04004198 }
Chris Masona52d9a82007-08-27 16:49:44 -04004199 goto insert;
4200 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04004201 unsigned long ptr;
Chris Masona52d9a82007-08-27 16:49:44 -04004202 char *map;
Chris Mason3326d1b2007-10-15 16:18:25 -04004203 size_t size;
4204 size_t extent_offset;
4205 size_t copy_size;
Chris Masona52d9a82007-08-27 16:49:44 -04004206
Chris Masona52d9a82007-08-27 16:49:44 -04004207 em->block_start = EXTENT_MAP_INLINE;
Chris Masonc8b97812008-10-29 14:49:59 -04004208 if (!page || create) {
Yan689f9342007-10-29 11:41:07 -04004209 em->start = extent_start;
Yan Zheng9036c102008-10-30 14:19:41 -04004210 em->len = extent_end - extent_start;
Yan689f9342007-10-29 11:41:07 -04004211 goto out;
4212 }
4213
Yan Zheng9036c102008-10-30 14:19:41 -04004214 size = btrfs_file_extent_inline_len(leaf, item);
4215 extent_offset = page_offset(page) + pg_offset - extent_start;
Chris Mason70dec802008-01-29 09:59:12 -05004216 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
Yan689f9342007-10-29 11:41:07 -04004217 size - extent_offset);
Chris Mason3326d1b2007-10-15 16:18:25 -04004218 em->start = extent_start + extent_offset;
Chris Mason70dec802008-01-29 09:59:12 -05004219 em->len = (copy_size + root->sectorsize - 1) &
4220 ~((u64)root->sectorsize - 1);
Yan Zhengff5b7ee2008-11-10 07:34:43 -05004221 em->orig_start = EXTENT_MAP_INLINE;
Chris Masonc8b97812008-10-29 14:49:59 -04004222 if (compressed)
4223 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Yan689f9342007-10-29 11:41:07 -04004224 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
Chris Mason179e29e2007-11-01 11:28:41 -04004225 if (create == 0 && !PageUptodate(page)) {
Chris Masonc8b97812008-10-29 14:49:59 -04004226 if (btrfs_file_extent_compression(leaf, item) ==
4227 BTRFS_COMPRESS_ZLIB) {
4228 ret = uncompress_inline(path, inode, page,
4229 pg_offset,
4230 extent_offset, item);
4231 BUG_ON(ret);
4232 } else {
4233 map = kmap(page);
4234 read_extent_buffer(leaf, map + pg_offset, ptr,
4235 copy_size);
Chris Mason93c82d52009-09-11 12:36:29 -04004236 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4237 memset(map + pg_offset + copy_size, 0,
4238 PAGE_CACHE_SIZE - pg_offset -
4239 copy_size);
4240 }
Chris Masonc8b97812008-10-29 14:49:59 -04004241 kunmap(page);
4242 }
Chris Mason179e29e2007-11-01 11:28:41 -04004243 flush_dcache_page(page);
4244 } else if (create && PageUptodate(page)) {
4245 if (!trans) {
4246 kunmap(page);
4247 free_extent_map(em);
4248 em = NULL;
4249 btrfs_release_path(root, path);
Chris Masonf9295742008-07-17 12:54:14 -04004250 trans = btrfs_join_transaction(root, 1);
Chris Mason179e29e2007-11-01 11:28:41 -04004251 goto again;
4252 }
Chris Masonc8b97812008-10-29 14:49:59 -04004253 map = kmap(page);
Chris Mason70dec802008-01-29 09:59:12 -05004254 write_extent_buffer(leaf, map + pg_offset, ptr,
Chris Mason179e29e2007-11-01 11:28:41 -04004255 copy_size);
Chris Masonc8b97812008-10-29 14:49:59 -04004256 kunmap(page);
Chris Mason179e29e2007-11-01 11:28:41 -04004257 btrfs_mark_buffer_dirty(leaf);
Chris Masona52d9a82007-08-27 16:49:44 -04004258 }
Chris Masond1310b22008-01-24 16:13:08 -05004259 set_extent_uptodate(io_tree, em->start,
4260 extent_map_end(em) - 1, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04004261 goto insert;
4262 } else {
Chris Masond3977122009-01-05 21:25:51 -05004263 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
Chris Masona52d9a82007-08-27 16:49:44 -04004264 WARN_ON(1);
4265 }
4266not_found:
4267 em->start = start;
Chris Masond1310b22008-01-24 16:13:08 -05004268 em->len = len;
Chris Masona52d9a82007-08-27 16:49:44 -04004269not_found_em:
Chris Mason5f39d392007-10-15 16:14:19 -04004270 em->block_start = EXTENT_MAP_HOLE;
Yan Zheng9036c102008-10-30 14:19:41 -04004271 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -04004272insert:
4273 btrfs_release_path(root, path);
Chris Masond1310b22008-01-24 16:13:08 -05004274 if (em->start > start || extent_map_end(em) <= start) {
Chris Masond3977122009-01-05 21:25:51 -05004275 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4276 "[%llu %llu]\n", (unsigned long long)em->start,
4277 (unsigned long long)em->len,
4278 (unsigned long long)start,
4279 (unsigned long long)len);
Chris Masona52d9a82007-08-27 16:49:44 -04004280 err = -EIO;
4281 goto out;
4282 }
Chris Masond1310b22008-01-24 16:13:08 -05004283
4284 err = 0;
Chris Mason890871b2009-09-02 16:24:52 -04004285 write_lock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04004286 ret = add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04004287 /* it is possible that someone inserted the extent into the tree
4288 * while we had the lock dropped. It is also possible that
4289 * an overlapping map exists in the tree
4290 */
Chris Masona52d9a82007-08-27 16:49:44 -04004291 if (ret == -EEXIST) {
Chris Mason3b951512008-04-17 11:29:12 -04004292 struct extent_map *existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004293
4294 ret = 0;
4295
Chris Mason3b951512008-04-17 11:29:12 -04004296 existing = lookup_extent_mapping(em_tree, start, len);
Chris Masone1c4b742008-04-22 13:26:46 -04004297 if (existing && (existing->start > start ||
4298 existing->start + existing->len <= start)) {
4299 free_extent_map(existing);
4300 existing = NULL;
4301 }
Chris Mason3b951512008-04-17 11:29:12 -04004302 if (!existing) {
4303 existing = lookup_extent_mapping(em_tree, em->start,
4304 em->len);
4305 if (existing) {
4306 err = merge_extent_mapping(em_tree, existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04004307 em, start,
4308 root->sectorsize);
Chris Mason3b951512008-04-17 11:29:12 -04004309 free_extent_map(existing);
4310 if (err) {
4311 free_extent_map(em);
4312 em = NULL;
4313 }
4314 } else {
4315 err = -EIO;
Chris Mason3b951512008-04-17 11:29:12 -04004316 free_extent_map(em);
4317 em = NULL;
4318 }
4319 } else {
4320 free_extent_map(em);
4321 em = existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004322 err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04004323 }
Chris Masona52d9a82007-08-27 16:49:44 -04004324 }
Chris Mason890871b2009-09-02 16:24:52 -04004325 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04004326out:
Chris Masonf4219502008-07-22 11:18:09 -04004327 if (path)
4328 btrfs_free_path(path);
Chris Masona52d9a82007-08-27 16:49:44 -04004329 if (trans) {
4330 ret = btrfs_end_transaction(trans, root);
Chris Masond3977122009-01-05 21:25:51 -05004331 if (!err)
Chris Masona52d9a82007-08-27 16:49:44 -04004332 err = ret;
4333 }
Chris Masona52d9a82007-08-27 16:49:44 -04004334 if (err) {
4335 free_extent_map(em);
Chris Masona52d9a82007-08-27 16:49:44 -04004336 return ERR_PTR(err);
4337 }
4338 return em;
4339}
4340
Chris Mason16432982008-04-10 10:23:21 -04004341static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4342 const struct iovec *iov, loff_t offset,
4343 unsigned long nr_segs)
4344{
Chris Masone1c4b742008-04-22 13:26:46 -04004345 return -EINVAL;
Chris Mason16432982008-04-10 10:23:21 -04004346}
4347
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05004348static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4349 __u64 start, __u64 len)
4350{
4351 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4352}
4353
Chris Mason9ebefb182007-06-15 13:50:00 -04004354int btrfs_readpage(struct file *file, struct page *page)
4355{
Chris Masond1310b22008-01-24 16:13:08 -05004356 struct extent_io_tree *tree;
4357 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04004358 return extent_read_full_page(tree, page, btrfs_get_extent);
Chris Mason39279cc2007-06-12 06:35:45 -04004359}
Chris Mason1832a6d2007-12-21 16:27:21 -05004360
Chris Mason39279cc2007-06-12 06:35:45 -04004361static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4362{
Chris Masond1310b22008-01-24 16:13:08 -05004363 struct extent_io_tree *tree;
Chris Masonb888db22007-08-27 16:49:44 -04004364
4365
4366 if (current->flags & PF_MEMALLOC) {
4367 redirty_page_for_writepage(wbc, page);
4368 unlock_page(page);
4369 return 0;
4370 }
Chris Masond1310b22008-01-24 16:13:08 -05004371 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04004372 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4373}
Chris Mason39279cc2007-06-12 06:35:45 -04004374
Chris Masonf4219502008-07-22 11:18:09 -04004375int btrfs_writepages(struct address_space *mapping,
4376 struct writeback_control *wbc)
Chris Masonb293f022007-11-01 19:45:34 -04004377{
Chris Masond1310b22008-01-24 16:13:08 -05004378 struct extent_io_tree *tree;
Chris Mason771ed682008-11-06 22:02:51 -05004379
Chris Masond1310b22008-01-24 16:13:08 -05004380 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Masonb293f022007-11-01 19:45:34 -04004381 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4382}
4383
Chris Mason3ab2fb52007-11-08 10:59:22 -05004384static int
4385btrfs_readpages(struct file *file, struct address_space *mapping,
4386 struct list_head *pages, unsigned nr_pages)
4387{
Chris Masond1310b22008-01-24 16:13:08 -05004388 struct extent_io_tree *tree;
4389 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Mason3ab2fb52007-11-08 10:59:22 -05004390 return extent_readpages(tree, mapping, pages, nr_pages,
4391 btrfs_get_extent);
4392}
Chris Masone6dcd2d2008-07-17 12:53:50 -04004393static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
Chris Masona52d9a82007-08-27 16:49:44 -04004394{
Chris Masond1310b22008-01-24 16:13:08 -05004395 struct extent_io_tree *tree;
4396 struct extent_map_tree *map;
Chris Masona52d9a82007-08-27 16:49:44 -04004397 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004398
Chris Masond1310b22008-01-24 16:13:08 -05004399 tree = &BTRFS_I(page->mapping->host)->io_tree;
4400 map = &BTRFS_I(page->mapping->host)->extent_tree;
Chris Mason70dec802008-01-29 09:59:12 -05004401 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
Chris Masona52d9a82007-08-27 16:49:44 -04004402 if (ret == 1) {
4403 ClearPagePrivate(page);
4404 set_page_private(page, 0);
4405 page_cache_release(page);
4406 }
4407 return ret;
4408}
Chris Mason39279cc2007-06-12 06:35:45 -04004409
Chris Masone6dcd2d2008-07-17 12:53:50 -04004410static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4411{
Chris Mason98509cf2008-09-11 15:51:43 -04004412 if (PageWriteback(page) || PageDirty(page))
4413 return 0;
Yan Zhengb335b002009-02-12 10:06:04 -05004414 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004415}
4416
Chris Masona52d9a82007-08-27 16:49:44 -04004417static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4418{
Chris Masond1310b22008-01-24 16:13:08 -05004419 struct extent_io_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004420 struct btrfs_ordered_extent *ordered;
4421 u64 page_start = page_offset(page);
4422 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04004423
Chris Mason8b62b722009-09-02 16:53:46 -04004424
4425 /*
4426 * we have the page locked, so new writeback can't start,
4427 * and the dirty bit won't be cleared while we are here.
4428 *
4429 * Wait for IO on this page so that we can safely clear
4430 * the PagePrivate2 bit and do ordered accounting
4431 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04004432 wait_on_page_writeback(page);
Chris Mason8b62b722009-09-02 16:53:46 -04004433
Chris Masond1310b22008-01-24 16:13:08 -05004434 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004435 if (offset) {
4436 btrfs_releasepage(page, GFP_NOFS);
4437 return;
4438 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004439 lock_extent(tree, page_start, page_end, GFP_NOFS);
4440 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4441 page_offset(page));
4442 if (ordered) {
Chris Masoneb84ae02008-07-17 13:53:27 -04004443 /*
4444 * IO on this page will never be started, so we need
4445 * to account for any ordered extents now
4446 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04004447 clear_extent_bit(tree, page_start, page_end,
4448 EXTENT_DIRTY | EXTENT_DELALLOC |
Chris Mason2c64c532009-09-02 15:04:12 -04004449 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
Chris Mason8b62b722009-09-02 16:53:46 -04004450 /*
4451 * whoever cleared the private bit is responsible
4452 * for the finish_ordered_io
4453 */
4454 if (TestClearPagePrivate2(page)) {
4455 btrfs_finish_ordered_io(page->mapping->host,
4456 page_start, page_end);
4457 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004458 btrfs_put_ordered_extent(ordered);
4459 lock_extent(tree, page_start, page_end, GFP_NOFS);
4460 }
4461 clear_extent_bit(tree, page_start, page_end,
Chris Mason8b62b722009-09-02 16:53:46 -04004462 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
Chris Mason2c64c532009-09-02 15:04:12 -04004463 1, 1, NULL, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004464 __btrfs_releasepage(page, GFP_NOFS);
4465
Chris Mason4a096752008-07-21 10:29:44 -04004466 ClearPageChecked(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04004467 if (PagePrivate(page)) {
Chris Mason9ad6b7b2008-04-18 16:11:30 -04004468 ClearPagePrivate(page);
4469 set_page_private(page, 0);
4470 page_cache_release(page);
4471 }
Chris Mason39279cc2007-06-12 06:35:45 -04004472}
4473
Chris Mason9ebefb182007-06-15 13:50:00 -04004474/*
4475 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4476 * called from a page fault handler when a page is first dirtied. Hence we must
4477 * be careful to check for EOF conditions here. We set the page up correctly
4478 * for a written page which means we get ENOSPC checking when writing into
4479 * holes and correct delalloc and unwritten extent mapping on filesystems that
4480 * support these features.
4481 *
4482 * We are not allowed to take the i_mutex here so we have to play games to
4483 * protect against truncate races as the page could now be beyond EOF. Because
4484 * vmtruncate() writes the inode size before removing pages, once we have the
4485 * page lock we can determine safely if the page is beyond EOF. If it is not
4486 * beyond EOF, then the page is guaranteed safe against truncation until we
4487 * unlock the page.
4488 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07004489int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Chris Mason9ebefb182007-06-15 13:50:00 -04004490{
Nick Pigginc2ec1752009-03-31 15:23:21 -07004491 struct page *page = vmf->page;
Chris Mason6da6aba2007-12-18 16:15:09 -05004492 struct inode *inode = fdentry(vma->vm_file)->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05004493 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004494 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4495 struct btrfs_ordered_extent *ordered;
4496 char *kaddr;
4497 unsigned long zero_start;
Chris Mason9ebefb182007-06-15 13:50:00 -04004498 loff_t size;
Chris Mason1832a6d2007-12-21 16:27:21 -05004499 int ret;
Chris Masona52d9a82007-08-27 16:49:44 -04004500 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004501 u64 page_end;
Chris Mason9ebefb182007-06-15 13:50:00 -04004502
Josef Bacik6a632092009-02-20 11:00:09 -05004503 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
Nick Piggin56a76f82009-03-31 15:23:23 -07004504 if (ret) {
4505 if (ret == -ENOMEM)
4506 ret = VM_FAULT_OOM;
4507 else /* -ENOSPC, -EIO, etc */
4508 ret = VM_FAULT_SIGBUS;
Chris Mason1832a6d2007-12-21 16:27:21 -05004509 goto out;
Nick Piggin56a76f82009-03-31 15:23:23 -07004510 }
Chris Mason1832a6d2007-12-21 16:27:21 -05004511
Nick Piggin56a76f82009-03-31 15:23:23 -07004512 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
Chris Masone6dcd2d2008-07-17 12:53:50 -04004513again:
Chris Mason9ebefb182007-06-15 13:50:00 -04004514 lock_page(page);
Chris Mason9ebefb182007-06-15 13:50:00 -04004515 size = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004516 page_start = page_offset(page);
4517 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04004518
Chris Mason9ebefb182007-06-15 13:50:00 -04004519 if ((page->mapping != inode->i_mapping) ||
Chris Masone6dcd2d2008-07-17 12:53:50 -04004520 (page_start >= size)) {
Josef Bacik6a632092009-02-20 11:00:09 -05004521 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
Chris Mason9ebefb182007-06-15 13:50:00 -04004522 /* page got truncated out from underneath us */
4523 goto out_unlock;
4524 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004525 wait_on_page_writeback(page);
4526
4527 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4528 set_page_extent_mapped(page);
4529
Chris Masoneb84ae02008-07-17 13:53:27 -04004530 /*
4531 * we can't set the delalloc bits if there are pending ordered
4532 * extents. Drop our locks and wait for them to finish
4533 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04004534 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4535 if (ordered) {
4536 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4537 unlock_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04004538 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004539 btrfs_put_ordered_extent(ordered);
4540 goto again;
4541 }
4542
Chris Masonea8c2812008-08-04 23:17:27 -04004543 btrfs_set_extent_delalloc(inode, page_start, page_end);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004544 ret = 0;
Chris Mason9ebefb182007-06-15 13:50:00 -04004545
4546 /* page is wholly or partially inside EOF */
Chris Masona52d9a82007-08-27 16:49:44 -04004547 if (page_start + PAGE_CACHE_SIZE > size)
Chris Masone6dcd2d2008-07-17 12:53:50 -04004548 zero_start = size & ~PAGE_CACHE_MASK;
Chris Mason9ebefb182007-06-15 13:50:00 -04004549 else
Chris Masone6dcd2d2008-07-17 12:53:50 -04004550 zero_start = PAGE_CACHE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04004551
Chris Masone6dcd2d2008-07-17 12:53:50 -04004552 if (zero_start != PAGE_CACHE_SIZE) {
4553 kaddr = kmap(page);
4554 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4555 flush_dcache_page(page);
4556 kunmap(page);
4557 }
Chris Mason247e7432008-07-17 12:53:51 -04004558 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004559 set_page_dirty(page);
Chris Mason50a9b212009-09-11 12:33:12 -04004560 SetPageUptodate(page);
Chris Mason5a3f23d2009-03-31 13:27:11 -04004561
4562 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004563 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
Chris Mason9ebefb182007-06-15 13:50:00 -04004564
4565out_unlock:
Chris Mason50a9b212009-09-11 12:33:12 -04004566 if (!ret)
4567 return VM_FAULT_LOCKED;
Chris Mason9ebefb182007-06-15 13:50:00 -04004568 unlock_page(page);
Chris Mason1832a6d2007-12-21 16:27:21 -05004569out:
Chris Mason9ebefb182007-06-15 13:50:00 -04004570 return ret;
4571}
4572
Chris Mason39279cc2007-06-12 06:35:45 -04004573static void btrfs_truncate(struct inode *inode)
4574{
4575 struct btrfs_root *root = BTRFS_I(inode)->root;
4576 int ret;
4577 struct btrfs_trans_handle *trans;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004578 unsigned long nr;
Chris Masondbe674a2008-07-17 12:54:05 -04004579 u64 mask = root->sectorsize - 1;
Chris Mason39279cc2007-06-12 06:35:45 -04004580
4581 if (!S_ISREG(inode->i_mode))
4582 return;
4583 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4584 return;
4585
4586 btrfs_truncate_page(inode->i_mapping, inode->i_size);
Chris Mason4a096752008-07-21 10:29:44 -04004587 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
Chris Mason39279cc2007-06-12 06:35:45 -04004588
Chris Mason39279cc2007-06-12 06:35:45 -04004589 trans = btrfs_start_transaction(root, 1);
Chris Mason5a3f23d2009-03-31 13:27:11 -04004590
4591 /*
4592 * setattr is responsible for setting the ordered_data_close flag,
4593 * but that is only tested during the last file release. That
4594 * could happen well after the next commit, leaving a great big
4595 * window where new writes may get lost if someone chooses to write
4596 * to this file after truncating to zero
4597 *
4598 * The inode doesn't have any dirty data here, and so if we commit
4599 * this is a noop. If someone immediately starts writing to the inode
4600 * it is very likely we'll catch some of their writes in this
4601 * transaction, and the commit will find this file on the ordered
4602 * data list with good things to send down.
4603 *
4604 * This is a best effort solution, there is still a window where
4605 * using truncate to replace the contents of the file will
4606 * end up with a zero length file after a crash.
4607 */
4608 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4609 btrfs_add_ordered_operation(trans, root, inode);
4610
Chris Mason39279cc2007-06-12 06:35:45 -04004611 btrfs_set_trans_block_group(trans, inode);
Chris Masondbe674a2008-07-17 12:54:05 -04004612 btrfs_i_size_write(inode, inode->i_size);
Chris Mason39279cc2007-06-12 06:35:45 -04004613
Josef Bacik7b128762008-07-24 12:17:14 -04004614 ret = btrfs_orphan_add(trans, inode);
4615 if (ret)
4616 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04004617 /* FIXME, add redo link to tree so we don't leak on crash */
Chris Masone02119d2008-09-05 16:13:11 -04004618 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
Chris Mason85e21ba2008-01-29 15:11:36 -05004619 BTRFS_EXTENT_DATA_KEY);
Chris Mason39279cc2007-06-12 06:35:45 -04004620 btrfs_update_inode(trans, root, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004621
Josef Bacik7b128762008-07-24 12:17:14 -04004622 ret = btrfs_orphan_del(trans, inode);
4623 BUG_ON(ret);
4624
4625out:
4626 nr = trans->blocks_used;
Chris Mason89ce8a62008-06-25 16:01:31 -04004627 ret = btrfs_end_transaction_throttle(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004628 BUG_ON(ret);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004629 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004630}
4631
Sven Wegener3b963622008-06-09 21:57:42 -04004632/*
Chris Masond352ac62008-09-29 15:18:18 -04004633 * create a new subvolume directory/inode (helper for the ioctl).
4634 */
Yan Zhengd2fb3432008-12-11 16:30:39 -05004635int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4636 struct btrfs_root *new_root, struct dentry *dentry,
4637 u64 new_dirid, u64 alloc_hint)
Chris Mason39279cc2007-06-12 06:35:45 -04004638{
Chris Mason39279cc2007-06-12 06:35:45 -04004639 struct inode *inode;
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04004640 int error;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004641 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004642
Josef Bacikaec74772008-07-24 12:12:38 -04004643 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
Yan Zhengd2fb3432008-12-11 16:30:39 -05004644 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
Chris Mason54aa1f42007-06-22 14:16:25 -04004645 if (IS_ERR(inode))
Christoph Hellwigf46b5a62008-06-11 21:53:53 -04004646 return PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004647 inode->i_op = &btrfs_dir_inode_operations;
4648 inode->i_fop = &btrfs_dir_file_operations;
4649
Chris Mason39279cc2007-06-12 06:35:45 -04004650 inode->i_nlink = 1;
Chris Masondbe674a2008-07-17 12:54:05 -04004651 btrfs_i_size_write(inode, 0);
Sven Wegener3b963622008-06-09 21:57:42 -04004652
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04004653 error = btrfs_update_inode(trans, new_root, inode);
4654 if (error)
4655 return error;
4656
4657 d_instantiate(dentry, inode);
4658 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004659}
4660
Chris Masond352ac62008-09-29 15:18:18 -04004661/* helper function for file defrag and space balancing. This
4662 * forces readahead on a given range of bytes in an inode
4663 */
Chris Masonedbd8d42007-12-21 16:27:24 -05004664unsigned long btrfs_force_ra(struct address_space *mapping,
Chris Mason86479a02007-09-10 19:58:16 -04004665 struct file_ra_state *ra, struct file *file,
4666 pgoff_t offset, pgoff_t last_index)
4667{
Chris Mason8e7bf942008-04-28 09:02:36 -04004668 pgoff_t req_size = last_index - offset + 1;
Chris Mason86479a02007-09-10 19:58:16 -04004669
Chris Mason86479a02007-09-10 19:58:16 -04004670 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4671 return offset + req_size;
Chris Mason86479a02007-09-10 19:58:16 -04004672}
4673
Chris Mason39279cc2007-06-12 06:35:45 -04004674struct inode *btrfs_alloc_inode(struct super_block *sb)
4675{
4676 struct btrfs_inode *ei;
4677
4678 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4679 if (!ei)
4680 return NULL;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04004681 ei->last_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04004682 ei->logged_trans = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004683 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Josef Bacik33268ea2008-07-24 12:16:36 -04004684 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4685 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
Josef Bacik7b128762008-07-24 12:17:14 -04004686 INIT_LIST_HEAD(&ei->i_orphan);
Chris Mason5a3f23d2009-03-31 13:27:11 -04004687 INIT_LIST_HEAD(&ei->ordered_operations);
Chris Mason39279cc2007-06-12 06:35:45 -04004688 return &ei->vfs_inode;
4689}
4690
4691void btrfs_destroy_inode(struct inode *inode)
4692{
Chris Masone6dcd2d2008-07-17 12:53:50 -04004693 struct btrfs_ordered_extent *ordered;
Chris Mason5a3f23d2009-03-31 13:27:11 -04004694 struct btrfs_root *root = BTRFS_I(inode)->root;
4695
Chris Mason39279cc2007-06-12 06:35:45 -04004696 WARN_ON(!list_empty(&inode->i_dentry));
4697 WARN_ON(inode->i_data.nrpages);
4698
Josef Bacik33268ea2008-07-24 12:16:36 -04004699 if (BTRFS_I(inode)->i_acl &&
4700 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4701 posix_acl_release(BTRFS_I(inode)->i_acl);
4702 if (BTRFS_I(inode)->i_default_acl &&
4703 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4704 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4705
Chris Mason5a3f23d2009-03-31 13:27:11 -04004706 /*
4707 * Make sure we're properly removed from the ordered operation
4708 * lists.
4709 */
4710 smp_mb();
4711 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4712 spin_lock(&root->fs_info->ordered_extent_lock);
4713 list_del_init(&BTRFS_I(inode)->ordered_operations);
4714 spin_unlock(&root->fs_info->ordered_extent_lock);
4715 }
4716
4717 spin_lock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04004718 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4719 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4720 " list\n", inode->i_ino);
4721 dump_stack();
4722 }
Chris Mason5a3f23d2009-03-31 13:27:11 -04004723 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04004724
Chris Masond3977122009-01-05 21:25:51 -05004725 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04004726 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4727 if (!ordered)
4728 break;
4729 else {
Chris Masond3977122009-01-05 21:25:51 -05004730 printk(KERN_ERR "btrfs found ordered "
4731 "extent %llu %llu on inode cleanup\n",
4732 (unsigned long long)ordered->file_offset,
4733 (unsigned long long)ordered->len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004734 btrfs_remove_ordered_extent(inode, ordered);
4735 btrfs_put_ordered_extent(ordered);
4736 btrfs_put_ordered_extent(ordered);
4737 }
4738 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004739 inode_tree_del(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04004740 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004741 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4742}
4743
Sven Wegener0ee0fda2008-07-30 16:54:26 -04004744static void init_once(void *foo)
Chris Mason39279cc2007-06-12 06:35:45 -04004745{
4746 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4747
4748 inode_init_once(&ei->vfs_inode);
4749}
4750
4751void btrfs_destroy_cachep(void)
4752{
4753 if (btrfs_inode_cachep)
4754 kmem_cache_destroy(btrfs_inode_cachep);
4755 if (btrfs_trans_handle_cachep)
4756 kmem_cache_destroy(btrfs_trans_handle_cachep);
4757 if (btrfs_transaction_cachep)
4758 kmem_cache_destroy(btrfs_transaction_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04004759 if (btrfs_path_cachep)
4760 kmem_cache_destroy(btrfs_path_cachep);
4761}
4762
4763int btrfs_init_cachep(void)
4764{
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02004765 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
4766 sizeof(struct btrfs_inode), 0,
4767 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
Chris Mason39279cc2007-06-12 06:35:45 -04004768 if (!btrfs_inode_cachep)
4769 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02004770
4771 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
4772 sizeof(struct btrfs_trans_handle), 0,
4773 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04004774 if (!btrfs_trans_handle_cachep)
4775 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02004776
4777 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
4778 sizeof(struct btrfs_transaction), 0,
4779 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04004780 if (!btrfs_transaction_cachep)
4781 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02004782
4783 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
4784 sizeof(struct btrfs_path), 0,
4785 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04004786 if (!btrfs_path_cachep)
4787 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02004788
Chris Mason39279cc2007-06-12 06:35:45 -04004789 return 0;
4790fail:
4791 btrfs_destroy_cachep();
4792 return -ENOMEM;
4793}
4794
4795static int btrfs_getattr(struct vfsmount *mnt,
4796 struct dentry *dentry, struct kstat *stat)
4797{
4798 struct inode *inode = dentry->d_inode;
4799 generic_fillattr(inode, stat);
Chris Mason3394e162008-11-17 20:42:26 -05004800 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
Chris Masond6667462008-01-03 14:51:00 -05004801 stat->blksize = PAGE_CACHE_SIZE;
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004802 stat->blocks = (inode_get_bytes(inode) +
4803 BTRFS_I(inode)->delalloc_bytes) >> 9;
Chris Mason39279cc2007-06-12 06:35:45 -04004804 return 0;
4805}
4806
Chris Masond3977122009-01-05 21:25:51 -05004807static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4808 struct inode *new_dir, struct dentry *new_dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04004809{
4810 struct btrfs_trans_handle *trans;
4811 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4812 struct inode *new_inode = new_dentry->d_inode;
4813 struct inode *old_inode = old_dentry->d_inode;
4814 struct timespec ctime = CURRENT_TIME;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004815 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004816 int ret;
4817
Chris Mason3394e162008-11-17 20:42:26 -05004818 /* we're not allowed to rename between subvolumes */
4819 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4820 BTRFS_I(new_dir)->root->root_key.objectid)
4821 return -EXDEV;
4822
Chris Mason39279cc2007-06-12 06:35:45 -04004823 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4824 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4825 return -ENOTEMPTY;
4826 }
Chris Mason5f39d392007-10-15 16:14:19 -04004827
Chris Mason0660b5a2008-11-17 20:37:39 -05004828 /* to rename a snapshot or subvolume, we need to juggle the
4829 * backrefs. This isn't coded yet
4830 */
4831 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4832 return -EXDEV;
4833
Josef Bacik6a632092009-02-20 11:00:09 -05004834 ret = btrfs_check_metadata_free_space(root);
Chris Mason1832a6d2007-12-21 16:27:21 -05004835 if (ret)
4836 goto out_unlock;
4837
Chris Mason5a3f23d2009-03-31 13:27:11 -04004838 /*
4839 * we're using rename to replace one file with another.
4840 * and the replacement file is large. Start IO on it now so
4841 * we don't add too much work to the end of the transaction
4842 */
Bartlomiej Zolnierkiewicz4baf8c92009-08-07 13:47:08 -04004843 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
Chris Mason5a3f23d2009-03-31 13:27:11 -04004844 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4845 filemap_flush(old_inode->i_mapping);
4846
Chris Mason39279cc2007-06-12 06:35:45 -04004847 trans = btrfs_start_transaction(root, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04004848
Chris Mason12fcfd22009-03-24 10:24:20 -04004849 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04004850 * make sure the inode gets flushed if it is replacing
4851 * something.
4852 */
4853 if (new_inode && new_inode->i_size &&
4854 old_inode && S_ISREG(old_inode->i_mode)) {
4855 btrfs_add_ordered_operation(trans, root, old_inode);
4856 }
4857
4858 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04004859 * this is an ugly little race, but the rename is required to make
4860 * sure that if we crash, the inode is either at the old name
4861 * or the new one. pinning the log transaction lets us make sure
4862 * we don't allow a log commit to come in after we unlink the
4863 * name but before we add the new name back in.
4864 */
4865 btrfs_pin_log_trans(root);
4866
Chris Mason39279cc2007-06-12 06:35:45 -04004867 btrfs_set_trans_block_group(trans, new_dir);
Chris Mason39279cc2007-06-12 06:35:45 -04004868
Chris Masone02119d2008-09-05 16:13:11 -04004869 btrfs_inc_nlink(old_dentry->d_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004870 old_dir->i_ctime = old_dir->i_mtime = ctime;
4871 new_dir->i_ctime = new_dir->i_mtime = ctime;
4872 old_inode->i_ctime = ctime;
Chris Mason5f39d392007-10-15 16:14:19 -04004873
Chris Mason12fcfd22009-03-24 10:24:20 -04004874 if (old_dentry->d_parent != new_dentry->d_parent)
4875 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4876
Chris Masone02119d2008-09-05 16:13:11 -04004877 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4878 old_dentry->d_name.name,
4879 old_dentry->d_name.len);
Chris Mason39279cc2007-06-12 06:35:45 -04004880 if (ret)
4881 goto out_fail;
4882
4883 if (new_inode) {
4884 new_inode->i_ctime = CURRENT_TIME;
Chris Masone02119d2008-09-05 16:13:11 -04004885 ret = btrfs_unlink_inode(trans, root, new_dir,
4886 new_dentry->d_inode,
4887 new_dentry->d_name.name,
4888 new_dentry->d_name.len);
Chris Mason39279cc2007-06-12 06:35:45 -04004889 if (ret)
4890 goto out_fail;
Josef Bacik7b128762008-07-24 12:17:14 -04004891 if (new_inode->i_nlink == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04004892 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
Josef Bacik7b128762008-07-24 12:17:14 -04004893 if (ret)
4894 goto out_fail;
4895 }
Chris Masone02119d2008-09-05 16:13:11 -04004896
Chris Mason39279cc2007-06-12 06:35:45 -04004897 }
Chris Mason3de45862008-11-17 21:02:50 -05004898 ret = btrfs_set_inode_index(new_dir, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04004899 if (ret)
4900 goto out_fail;
4901
Chris Masone02119d2008-09-05 16:13:11 -04004902 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4903 old_inode, new_dentry->d_name.name,
4904 new_dentry->d_name.len, 1, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004905 if (ret)
4906 goto out_fail;
4907
Chris Mason12fcfd22009-03-24 10:24:20 -04004908 btrfs_log_new_name(trans, old_inode, old_dir,
4909 new_dentry->d_parent);
Chris Mason39279cc2007-06-12 06:35:45 -04004910out_fail:
Chris Mason12fcfd22009-03-24 10:24:20 -04004911
4912 /* this btrfs_end_log_trans just allows the current
4913 * log-sub transaction to complete
4914 */
4915 btrfs_end_log_trans(root);
Chris Masonab78c842008-07-29 16:15:18 -04004916 btrfs_end_transaction_throttle(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05004917out_unlock:
Chris Mason39279cc2007-06-12 06:35:45 -04004918 return ret;
4919}
4920
Chris Masond352ac62008-09-29 15:18:18 -04004921/*
4922 * some fairly slow code that needs optimization. This walks the list
4923 * of all the inodes with pending delalloc and forces them to disk.
4924 */
Chris Masonea8c2812008-08-04 23:17:27 -04004925int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4926{
4927 struct list_head *head = &root->fs_info->delalloc_inodes;
4928 struct btrfs_inode *binode;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04004929 struct inode *inode;
Chris Masonea8c2812008-08-04 23:17:27 -04004930
Yan Zhengc146afa2008-11-12 14:34:12 -05004931 if (root->fs_info->sb->s_flags & MS_RDONLY)
4932 return -EROFS;
4933
Chris Mason75eff682008-12-15 15:54:40 -05004934 spin_lock(&root->fs_info->delalloc_lock);
Chris Masond3977122009-01-05 21:25:51 -05004935 while (!list_empty(head)) {
Chris Masonea8c2812008-08-04 23:17:27 -04004936 binode = list_entry(head->next, struct btrfs_inode,
4937 delalloc_inodes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04004938 inode = igrab(&binode->vfs_inode);
4939 if (!inode)
4940 list_del_init(&binode->delalloc_inodes);
Chris Mason75eff682008-12-15 15:54:40 -05004941 spin_unlock(&root->fs_info->delalloc_lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04004942 if (inode) {
Chris Mason8c8bee12008-09-29 11:19:10 -04004943 filemap_flush(inode->i_mapping);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04004944 iput(inode);
4945 }
4946 cond_resched();
Chris Mason75eff682008-12-15 15:54:40 -05004947 spin_lock(&root->fs_info->delalloc_lock);
Chris Masonea8c2812008-08-04 23:17:27 -04004948 }
Chris Mason75eff682008-12-15 15:54:40 -05004949 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason8c8bee12008-09-29 11:19:10 -04004950
4951 /* the filemap_flush will queue IO into the worker threads, but
4952 * we have to make sure the IO is actually started and that
4953 * ordered extents get created before we return
4954 */
4955 atomic_inc(&root->fs_info->async_submit_draining);
Chris Masond3977122009-01-05 21:25:51 -05004956 while (atomic_read(&root->fs_info->nr_async_submits) ||
Chris Mason771ed682008-11-06 22:02:51 -05004957 atomic_read(&root->fs_info->async_delalloc_pages)) {
Chris Mason8c8bee12008-09-29 11:19:10 -04004958 wait_event(root->fs_info->async_submit_wait,
Chris Mason771ed682008-11-06 22:02:51 -05004959 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4960 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
Chris Mason8c8bee12008-09-29 11:19:10 -04004961 }
4962 atomic_dec(&root->fs_info->async_submit_draining);
Chris Masonea8c2812008-08-04 23:17:27 -04004963 return 0;
4964}
4965
Chris Mason39279cc2007-06-12 06:35:45 -04004966static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4967 const char *symname)
4968{
4969 struct btrfs_trans_handle *trans;
4970 struct btrfs_root *root = BTRFS_I(dir)->root;
4971 struct btrfs_path *path;
4972 struct btrfs_key key;
Chris Mason1832a6d2007-12-21 16:27:21 -05004973 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04004974 int err;
4975 int drop_inode = 0;
4976 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004977 u64 index = 0 ;
Chris Mason39279cc2007-06-12 06:35:45 -04004978 int name_len;
4979 int datasize;
Chris Mason5f39d392007-10-15 16:14:19 -04004980 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04004981 struct btrfs_file_extent_item *ei;
Chris Mason5f39d392007-10-15 16:14:19 -04004982 struct extent_buffer *leaf;
Chris Mason1832a6d2007-12-21 16:27:21 -05004983 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004984
4985 name_len = strlen(symname) + 1;
4986 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4987 return -ENAMETOOLONG;
Chris Mason1832a6d2007-12-21 16:27:21 -05004988
Josef Bacik6a632092009-02-20 11:00:09 -05004989 err = btrfs_check_metadata_free_space(root);
Chris Mason1832a6d2007-12-21 16:27:21 -05004990 if (err)
4991 goto out_fail;
4992
Chris Mason39279cc2007-06-12 06:35:45 -04004993 trans = btrfs_start_transaction(root, 1);
4994 btrfs_set_trans_block_group(trans, dir);
4995
4996 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4997 if (err) {
4998 err = -ENOSPC;
4999 goto out_unlock;
5000 }
5001
Josef Bacikaec74772008-07-24 12:12:38 -04005002 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Chris Mason9c583092008-01-29 15:15:18 -05005003 dentry->d_name.len,
5004 dentry->d_parent->d_inode->i_ino, objectid,
Chris Mason00e4e6b2008-08-05 11:18:09 -04005005 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5006 &index);
Chris Mason39279cc2007-06-12 06:35:45 -04005007 err = PTR_ERR(inode);
5008 if (IS_ERR(inode))
5009 goto out_unlock;
5010
Jim Owens0279b4c2009-02-04 09:29:13 -05005011 err = btrfs_init_inode_security(inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04005012 if (err) {
5013 drop_inode = 1;
5014 goto out_unlock;
5015 }
5016
Chris Mason39279cc2007-06-12 06:35:45 -04005017 btrfs_set_trans_block_group(trans, inode);
Chris Mason00e4e6b2008-08-05 11:18:09 -04005018 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04005019 if (err)
5020 drop_inode = 1;
5021 else {
5022 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04005023 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04005024 inode->i_fop = &btrfs_file_operations;
5025 inode->i_op = &btrfs_file_inode_operations;
Chris Masond1310b22008-01-24 16:13:08 -05005026 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04005027 }
5028 dir->i_sb->s_dirt = 1;
5029 btrfs_update_inode_block_group(trans, inode);
5030 btrfs_update_inode_block_group(trans, dir);
5031 if (drop_inode)
5032 goto out_unlock;
5033
5034 path = btrfs_alloc_path();
5035 BUG_ON(!path);
5036 key.objectid = inode->i_ino;
5037 key.offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005038 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5039 datasize = btrfs_file_extent_calc_inline_size(name_len);
5040 err = btrfs_insert_empty_item(trans, root, path, &key,
5041 datasize);
Chris Mason54aa1f42007-06-22 14:16:25 -04005042 if (err) {
5043 drop_inode = 1;
5044 goto out_unlock;
5045 }
Chris Mason5f39d392007-10-15 16:14:19 -04005046 leaf = path->nodes[0];
5047 ei = btrfs_item_ptr(leaf, path->slots[0],
5048 struct btrfs_file_extent_item);
5049 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5050 btrfs_set_file_extent_type(leaf, ei,
Chris Mason39279cc2007-06-12 06:35:45 -04005051 BTRFS_FILE_EXTENT_INLINE);
Chris Masonc8b97812008-10-29 14:49:59 -04005052 btrfs_set_file_extent_encryption(leaf, ei, 0);
5053 btrfs_set_file_extent_compression(leaf, ei, 0);
5054 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5055 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5056
Chris Mason39279cc2007-06-12 06:35:45 -04005057 ptr = btrfs_file_extent_inline_start(ei);
Chris Mason5f39d392007-10-15 16:14:19 -04005058 write_extent_buffer(leaf, symname, ptr, name_len);
5059 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04005060 btrfs_free_path(path);
Chris Mason5f39d392007-10-15 16:14:19 -04005061
Chris Mason39279cc2007-06-12 06:35:45 -04005062 inode->i_op = &btrfs_symlink_inode_operations;
5063 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04005064 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Yan Zhengd899e052008-10-30 14:25:28 -04005065 inode_set_bytes(inode, name_len);
Chris Masondbe674a2008-07-17 12:54:05 -04005066 btrfs_i_size_write(inode, name_len - 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04005067 err = btrfs_update_inode(trans, root, inode);
5068 if (err)
5069 drop_inode = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04005070
5071out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04005072 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04005073 btrfs_end_transaction_throttle(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05005074out_fail:
Chris Mason39279cc2007-06-12 06:35:45 -04005075 if (drop_inode) {
5076 inode_dec_link_count(inode);
5077 iput(inode);
5078 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04005079 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04005080 return err;
5081}
Chris Mason16432982008-04-10 10:23:21 -04005082
Chris Mason546888d2009-04-21 11:53:38 -04005083static int prealloc_file_range(struct btrfs_trans_handle *trans,
5084 struct inode *inode, u64 start, u64 end,
Chris Masone980b502009-04-24 14:39:24 -04005085 u64 locked_end, u64 alloc_hint, int mode)
Yan Zhengd899e052008-10-30 14:25:28 -04005086{
Yan Zhengd899e052008-10-30 14:25:28 -04005087 struct btrfs_root *root = BTRFS_I(inode)->root;
5088 struct btrfs_key ins;
5089 u64 alloc_size;
5090 u64 cur_offset = start;
5091 u64 num_bytes = end - start;
5092 int ret = 0;
5093
Yan Zhengd899e052008-10-30 14:25:28 -04005094 while (num_bytes > 0) {
5095 alloc_size = min(num_bytes, root->fs_info->max_extent);
5096 ret = btrfs_reserve_extent(trans, root, alloc_size,
5097 root->sectorsize, 0, alloc_hint,
5098 (u64)-1, &ins, 1);
5099 if (ret) {
5100 WARN_ON(1);
5101 goto out;
5102 }
5103 ret = insert_reserved_file_extent(trans, inode,
5104 cur_offset, ins.objectid,
5105 ins.offset, ins.offset,
Chris Masone980b502009-04-24 14:39:24 -04005106 ins.offset, locked_end,
5107 0, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04005108 BTRFS_FILE_EXTENT_PREALLOC);
5109 BUG_ON(ret);
Chris Masona1ed8352009-09-11 12:27:37 -04005110 btrfs_drop_extent_cache(inode, cur_offset,
5111 cur_offset + ins.offset -1, 0);
Yan Zhengd899e052008-10-30 14:25:28 -04005112 num_bytes -= ins.offset;
5113 cur_offset += ins.offset;
5114 alloc_hint = ins.objectid + ins.offset;
5115 }
5116out:
5117 if (cur_offset > start) {
5118 inode->i_ctime = CURRENT_TIME;
Christoph Hellwig6cbff002009-04-17 10:37:41 +02005119 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Yan Zhengd899e052008-10-30 14:25:28 -04005120 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5121 cur_offset > i_size_read(inode))
5122 btrfs_i_size_write(inode, cur_offset);
5123 ret = btrfs_update_inode(trans, root, inode);
5124 BUG_ON(ret);
5125 }
5126
Yan Zhengd899e052008-10-30 14:25:28 -04005127 return ret;
5128}
5129
5130static long btrfs_fallocate(struct inode *inode, int mode,
5131 loff_t offset, loff_t len)
5132{
5133 u64 cur_offset;
5134 u64 last_byte;
5135 u64 alloc_start;
5136 u64 alloc_end;
5137 u64 alloc_hint = 0;
Chris Masone980b502009-04-24 14:39:24 -04005138 u64 locked_end;
Yan Zhengd899e052008-10-30 14:25:28 -04005139 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5140 struct extent_map *em;
Chris Mason546888d2009-04-21 11:53:38 -04005141 struct btrfs_trans_handle *trans;
Josef Bacika970b0a2009-06-27 21:07:34 -04005142 struct btrfs_root *root;
Yan Zhengd899e052008-10-30 14:25:28 -04005143 int ret;
5144
5145 alloc_start = offset & ~mask;
5146 alloc_end = (offset + len + mask) & ~mask;
5147
Chris Mason546888d2009-04-21 11:53:38 -04005148 /*
5149 * wait for ordered IO before we have any locks. We'll loop again
5150 * below with the locks held.
5151 */
5152 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5153
Yan Zhengd899e052008-10-30 14:25:28 -04005154 mutex_lock(&inode->i_mutex);
5155 if (alloc_start > inode->i_size) {
5156 ret = btrfs_cont_expand(inode, alloc_start);
5157 if (ret)
5158 goto out;
5159 }
5160
Josef Bacika970b0a2009-06-27 21:07:34 -04005161 root = BTRFS_I(inode)->root;
5162
5163 ret = btrfs_check_data_free_space(root, inode,
5164 alloc_end - alloc_start);
5165 if (ret)
5166 goto out;
5167
Chris Masone980b502009-04-24 14:39:24 -04005168 locked_end = alloc_end - 1;
Yan Zhengd899e052008-10-30 14:25:28 -04005169 while (1) {
5170 struct btrfs_ordered_extent *ordered;
Chris Mason546888d2009-04-21 11:53:38 -04005171
5172 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5173 if (!trans) {
5174 ret = -EIO;
Josef Bacika970b0a2009-06-27 21:07:34 -04005175 goto out_free;
Chris Mason546888d2009-04-21 11:53:38 -04005176 }
5177
5178 /* the extent lock is ordered inside the running
5179 * transaction
5180 */
Chris Masone980b502009-04-24 14:39:24 -04005181 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5182 GFP_NOFS);
Yan Zhengd899e052008-10-30 14:25:28 -04005183 ordered = btrfs_lookup_first_ordered_extent(inode,
5184 alloc_end - 1);
5185 if (ordered &&
5186 ordered->file_offset + ordered->len > alloc_start &&
5187 ordered->file_offset < alloc_end) {
5188 btrfs_put_ordered_extent(ordered);
5189 unlock_extent(&BTRFS_I(inode)->io_tree,
Chris Masone980b502009-04-24 14:39:24 -04005190 alloc_start, locked_end, GFP_NOFS);
Chris Mason546888d2009-04-21 11:53:38 -04005191 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5192
5193 /*
5194 * we can't wait on the range with the transaction
5195 * running or with the extent lock held
5196 */
Yan Zhengd899e052008-10-30 14:25:28 -04005197 btrfs_wait_ordered_range(inode, alloc_start,
5198 alloc_end - alloc_start);
5199 } else {
5200 if (ordered)
5201 btrfs_put_ordered_extent(ordered);
5202 break;
5203 }
5204 }
5205
5206 cur_offset = alloc_start;
5207 while (1) {
5208 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5209 alloc_end - cur_offset, 0);
5210 BUG_ON(IS_ERR(em) || !em);
5211 last_byte = min(extent_map_end(em), alloc_end);
5212 last_byte = (last_byte + mask) & ~mask;
5213 if (em->block_start == EXTENT_MAP_HOLE) {
Chris Mason546888d2009-04-21 11:53:38 -04005214 ret = prealloc_file_range(trans, inode, cur_offset,
Chris Masone980b502009-04-24 14:39:24 -04005215 last_byte, locked_end + 1,
5216 alloc_hint, mode);
Yan Zhengd899e052008-10-30 14:25:28 -04005217 if (ret < 0) {
5218 free_extent_map(em);
5219 break;
5220 }
5221 }
5222 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5223 alloc_hint = em->block_start;
5224 free_extent_map(em);
5225
5226 cur_offset = last_byte;
5227 if (cur_offset >= alloc_end) {
5228 ret = 0;
5229 break;
5230 }
5231 }
Chris Masone980b502009-04-24 14:39:24 -04005232 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
Yan Zhengd899e052008-10-30 14:25:28 -04005233 GFP_NOFS);
Chris Mason546888d2009-04-21 11:53:38 -04005234
5235 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
Josef Bacika970b0a2009-06-27 21:07:34 -04005236out_free:
5237 btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start);
Yan Zhengd899e052008-10-30 14:25:28 -04005238out:
5239 mutex_unlock(&inode->i_mutex);
5240 return ret;
5241}
5242
Chris Masone6dcd2d2008-07-17 12:53:50 -04005243static int btrfs_set_page_dirty(struct page *page)
5244{
Chris Masone6dcd2d2008-07-17 12:53:50 -04005245 return __set_page_dirty_nobuffers(page);
5246}
5247
Sven Wegener0ee0fda2008-07-30 16:54:26 -04005248static int btrfs_permission(struct inode *inode, int mask)
Yanfdebe2b2008-01-14 13:26:08 -05005249{
Christoph Hellwig6cbff002009-04-17 10:37:41 +02005250 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
Yanfdebe2b2008-01-14 13:26:08 -05005251 return -EACCES;
Josef Bacik33268ea2008-07-24 12:16:36 -04005252 return generic_permission(inode, mask, btrfs_check_acl);
Yanfdebe2b2008-01-14 13:26:08 -05005253}
Chris Mason39279cc2007-06-12 06:35:45 -04005254
5255static struct inode_operations btrfs_dir_inode_operations = {
Chris Mason3394e162008-11-17 20:42:26 -05005256 .getattr = btrfs_getattr,
Chris Mason39279cc2007-06-12 06:35:45 -04005257 .lookup = btrfs_lookup,
5258 .create = btrfs_create,
5259 .unlink = btrfs_unlink,
5260 .link = btrfs_link,
5261 .mkdir = btrfs_mkdir,
5262 .rmdir = btrfs_rmdir,
5263 .rename = btrfs_rename,
5264 .symlink = btrfs_symlink,
5265 .setattr = btrfs_setattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04005266 .mknod = btrfs_mknod,
Christoph Hellwig95819c02008-08-28 06:21:17 -04005267 .setxattr = btrfs_setxattr,
5268 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05005269 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04005270 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05005271 .permission = btrfs_permission,
Chris Mason39279cc2007-06-12 06:35:45 -04005272};
Chris Mason39279cc2007-06-12 06:35:45 -04005273static struct inode_operations btrfs_dir_ro_inode_operations = {
5274 .lookup = btrfs_lookup,
Yanfdebe2b2008-01-14 13:26:08 -05005275 .permission = btrfs_permission,
Chris Mason39279cc2007-06-12 06:35:45 -04005276};
Chris Mason39279cc2007-06-12 06:35:45 -04005277static struct file_operations btrfs_dir_file_operations = {
5278 .llseek = generic_file_llseek,
5279 .read = generic_read_dir,
David Woodhousecbdf5a22008-08-06 19:42:33 +01005280 .readdir = btrfs_real_readdir,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04005281 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04005282#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04005283 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04005284#endif
Sage Weil6bf13c02008-06-10 10:07:39 -04005285 .release = btrfs_release_file,
Chris Masone02119d2008-09-05 16:13:11 -04005286 .fsync = btrfs_sync_file,
Chris Mason39279cc2007-06-12 06:35:45 -04005287};
5288
Chris Masond1310b22008-01-24 16:13:08 -05005289static struct extent_io_ops btrfs_extent_io_ops = {
Chris Mason07157aa2007-08-30 08:50:51 -04005290 .fill_delalloc = run_delalloc_range,
Chris Mason065631f2008-02-20 12:07:25 -05005291 .submit_bio_hook = btrfs_submit_bio_hook,
Chris Mason239b14b2008-03-24 15:02:07 -04005292 .merge_bio_hook = btrfs_merge_bio_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04005293 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
Chris Masone6dcd2d2008-07-17 12:53:50 -04005294 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
Chris Mason247e7432008-07-17 12:53:51 -04005295 .writepage_start_hook = btrfs_writepage_start_hook,
Chris Mason1259ab72008-05-12 13:39:03 -04005296 .readpage_io_failed_hook = btrfs_io_failed_hook,
Chris Masonb0c68f82008-01-31 11:05:37 -05005297 .set_bit_hook = btrfs_set_bit_hook,
5298 .clear_bit_hook = btrfs_clear_bit_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04005299};
5300
Chris Mason35054392009-01-21 13:11:13 -05005301/*
5302 * btrfs doesn't support the bmap operation because swapfiles
5303 * use bmap to make a mapping of extents in the file. They assume
5304 * these extents won't change over the life of the file and they
5305 * use the bmap result to do IO directly to the drive.
5306 *
5307 * the btrfs bmap call would return logical addresses that aren't
5308 * suitable for IO and they also will change frequently as COW
5309 * operations happen. So, swapfile + btrfs == corruption.
5310 *
5311 * For now we're avoiding this by dropping bmap.
5312 */
Chris Mason39279cc2007-06-12 06:35:45 -04005313static struct address_space_operations btrfs_aops = {
5314 .readpage = btrfs_readpage,
5315 .writepage = btrfs_writepage,
Chris Masonb293f022007-11-01 19:45:34 -04005316 .writepages = btrfs_writepages,
Chris Mason3ab2fb52007-11-08 10:59:22 -05005317 .readpages = btrfs_readpages,
Chris Mason39279cc2007-06-12 06:35:45 -04005318 .sync_page = block_sync_page,
Chris Mason16432982008-04-10 10:23:21 -04005319 .direct_IO = btrfs_direct_IO,
Chris Masona52d9a82007-08-27 16:49:44 -04005320 .invalidatepage = btrfs_invalidatepage,
5321 .releasepage = btrfs_releasepage,
Chris Masone6dcd2d2008-07-17 12:53:50 -04005322 .set_page_dirty = btrfs_set_page_dirty,
Chris Mason39279cc2007-06-12 06:35:45 -04005323};
5324
5325static struct address_space_operations btrfs_symlink_aops = {
5326 .readpage = btrfs_readpage,
5327 .writepage = btrfs_writepage,
Chris Mason2bf5a722007-08-30 11:54:02 -04005328 .invalidatepage = btrfs_invalidatepage,
5329 .releasepage = btrfs_releasepage,
Chris Mason39279cc2007-06-12 06:35:45 -04005330};
5331
5332static struct inode_operations btrfs_file_inode_operations = {
5333 .truncate = btrfs_truncate,
5334 .getattr = btrfs_getattr,
5335 .setattr = btrfs_setattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04005336 .setxattr = btrfs_setxattr,
5337 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05005338 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04005339 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05005340 .permission = btrfs_permission,
Yan Zhengd899e052008-10-30 14:25:28 -04005341 .fallocate = btrfs_fallocate,
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05005342 .fiemap = btrfs_fiemap,
Chris Mason39279cc2007-06-12 06:35:45 -04005343};
Josef Bacik618e21d2007-07-11 10:18:17 -04005344static struct inode_operations btrfs_special_inode_operations = {
5345 .getattr = btrfs_getattr,
5346 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05005347 .permission = btrfs_permission,
Christoph Hellwig95819c02008-08-28 06:21:17 -04005348 .setxattr = btrfs_setxattr,
5349 .getxattr = btrfs_getxattr,
Josef Bacik33268ea2008-07-24 12:16:36 -04005350 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04005351 .removexattr = btrfs_removexattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04005352};
Chris Mason39279cc2007-06-12 06:35:45 -04005353static struct inode_operations btrfs_symlink_inode_operations = {
5354 .readlink = generic_readlink,
5355 .follow_link = page_follow_link_light,
5356 .put_link = page_put_link,
Yanfdebe2b2008-01-14 13:26:08 -05005357 .permission = btrfs_permission,
Jim Owens0279b4c2009-02-04 09:29:13 -05005358 .setxattr = btrfs_setxattr,
5359 .getxattr = btrfs_getxattr,
5360 .listxattr = btrfs_listxattr,
5361 .removexattr = btrfs_removexattr,
Chris Mason39279cc2007-06-12 06:35:45 -04005362};