blob: 2bfdc641d4e3ba046f3afd89400f362ec959e46b [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>
Chris Mason39279cc2007-06-12 06:35:45 -040029#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
Chris Mason9ebefb182007-06-15 13:50:00 -040035#include <linux/bit_spinlock.h>
Josef Bacik5103e942007-11-16 11:45:54 -050036#include <linux/xattr.h>
Josef Bacik33268ea2008-07-24 12:16:36 -040037#include <linux/posix_acl.h>
Yan Zhengd899e052008-10-30 14:25:28 -040038#include <linux/falloc.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090039#include <linux/slab.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
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070059static const struct inode_operations btrfs_dir_inode_operations;
60static const struct inode_operations btrfs_symlink_inode_operations;
61static const struct inode_operations btrfs_dir_ro_inode_operations;
62static const struct inode_operations btrfs_special_inode_operations;
63static const struct inode_operations btrfs_file_inode_operations;
Alexey Dobriyan7f094102009-09-21 17:01:10 -070064static const struct address_space_operations btrfs_aops;
65static const struct address_space_operations btrfs_symlink_aops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -070066static const 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
Yan, Zhengf34f57a2009-11-12 09:35:27 +000092static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
93 struct inode *inode, struct inode *dir)
Jim Owens0279b4c2009-02-04 09:29:13 -050094{
95 int err;
96
Yan, Zhengf34f57a2009-11-12 09:35:27 +000097 err = btrfs_init_acl(trans, inode, dir);
Jim Owens0279b4c2009-02-04 09:29:13 -050098 if (!err)
Yan, Zhengf34f57a2009-11-12 09:35:27 +000099 err = btrfs_xattr_security_init(trans, inode, dir);
Jim Owens0279b4c2009-02-04 09:29:13 -0500100 return err;
101}
102
Chris Masond352ac62008-09-29 15:18:18 -0400103/*
Chris Masonc8b97812008-10-29 14:49:59 -0400104 * this does all the hard work for inserting an inline extent into
105 * the btree. The caller should have done a btrfs_drop_extents so that
106 * no overlapping inline items exist in the btree
107 */
Chris Masond3977122009-01-05 21:25:51 -0500108static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
Chris Masonc8b97812008-10-29 14:49:59 -0400109 struct btrfs_root *root, struct inode *inode,
110 u64 start, size_t size, size_t compressed_size,
111 struct page **compressed_pages)
112{
113 struct btrfs_key key;
114 struct btrfs_path *path;
115 struct extent_buffer *leaf;
116 struct page *page = NULL;
117 char *kaddr;
118 unsigned long ptr;
119 struct btrfs_file_extent_item *ei;
120 int err = 0;
121 int ret;
122 size_t cur_size = size;
123 size_t datasize;
124 unsigned long offset;
125 int use_compress = 0;
126
127 if (compressed_size && compressed_pages) {
128 use_compress = 1;
129 cur_size = compressed_size;
130 }
131
Chris Masond3977122009-01-05 21:25:51 -0500132 path = btrfs_alloc_path();
133 if (!path)
Chris Masonc8b97812008-10-29 14:49:59 -0400134 return -ENOMEM;
135
Chris Masonb9473432009-03-13 11:00:37 -0400136 path->leave_spinning = 1;
Chris Masonc8b97812008-10-29 14:49:59 -0400137 btrfs_set_trans_block_group(trans, inode);
138
139 key.objectid = inode->i_ino;
140 key.offset = start;
141 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
Chris Masonc8b97812008-10-29 14:49:59 -0400142 datasize = btrfs_file_extent_calc_inline_size(cur_size);
143
144 inode_add_bytes(inode, size);
145 ret = btrfs_insert_empty_item(trans, root, path, &key,
146 datasize);
147 BUG_ON(ret);
148 if (ret) {
149 err = ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400150 goto fail;
151 }
152 leaf = path->nodes[0];
153 ei = btrfs_item_ptr(leaf, path->slots[0],
154 struct btrfs_file_extent_item);
155 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
156 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
157 btrfs_set_file_extent_encryption(leaf, ei, 0);
158 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
159 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
160 ptr = btrfs_file_extent_inline_start(ei);
161
162 if (use_compress) {
163 struct page *cpage;
164 int i = 0;
Chris Masond3977122009-01-05 21:25:51 -0500165 while (compressed_size > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400166 cpage = compressed_pages[i];
Chris Mason5b050f02008-11-11 09:34:41 -0500167 cur_size = min_t(unsigned long, compressed_size,
Chris Masonc8b97812008-10-29 14:49:59 -0400168 PAGE_CACHE_SIZE);
169
Chris Masonb9473432009-03-13 11:00:37 -0400170 kaddr = kmap_atomic(cpage, KM_USER0);
Chris Masonc8b97812008-10-29 14:49:59 -0400171 write_extent_buffer(leaf, kaddr, ptr, cur_size);
Chris Masonb9473432009-03-13 11:00:37 -0400172 kunmap_atomic(kaddr, KM_USER0);
Chris Masonc8b97812008-10-29 14:49:59 -0400173
174 i++;
175 ptr += cur_size;
176 compressed_size -= cur_size;
177 }
178 btrfs_set_file_extent_compression(leaf, ei,
179 BTRFS_COMPRESS_ZLIB);
180 } else {
181 page = find_get_page(inode->i_mapping,
182 start >> PAGE_CACHE_SHIFT);
183 btrfs_set_file_extent_compression(leaf, ei, 0);
184 kaddr = kmap_atomic(page, KM_USER0);
185 offset = start & (PAGE_CACHE_SIZE - 1);
186 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187 kunmap_atomic(kaddr, KM_USER0);
188 page_cache_release(page);
189 }
190 btrfs_mark_buffer_dirty(leaf);
191 btrfs_free_path(path);
192
Yan, Zhengc2167752009-11-12 09:34:21 +0000193 /*
194 * we're an inline extent, so nobody can
195 * extend the file past i_size without locking
196 * a page we already have locked.
197 *
198 * We must do any isize and inode updates
199 * before we unlock the pages. Otherwise we
200 * could end up racing with unlink.
201 */
Chris Masonc8b97812008-10-29 14:49:59 -0400202 BTRFS_I(inode)->disk_i_size = inode->i_size;
203 btrfs_update_inode(trans, root, inode);
Yan, Zhengc2167752009-11-12 09:34:21 +0000204
Chris Masonc8b97812008-10-29 14:49:59 -0400205 return 0;
206fail:
207 btrfs_free_path(path);
208 return err;
209}
210
211
212/*
213 * conditionally insert an inline extent into the file. This
214 * does the checks required to make sure the data is small enough
215 * to fit as an inline extent.
216 */
Chris Mason7f366cf2009-03-12 20:12:45 -0400217static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
Chris Masonc8b97812008-10-29 14:49:59 -0400218 struct btrfs_root *root,
219 struct inode *inode, u64 start, u64 end,
220 size_t compressed_size,
221 struct page **compressed_pages)
222{
223 u64 isize = i_size_read(inode);
224 u64 actual_end = min(end + 1, isize);
225 u64 inline_len = actual_end - start;
226 u64 aligned_end = (end + root->sectorsize - 1) &
227 ~((u64)root->sectorsize - 1);
228 u64 hint_byte;
229 u64 data_len = inline_len;
230 int ret;
231
232 if (compressed_size)
233 data_len = compressed_size;
234
235 if (start > 0 ||
Chris Mason70b99e62008-10-31 12:46:39 -0400236 actual_end >= PAGE_CACHE_SIZE ||
Chris Masonc8b97812008-10-29 14:49:59 -0400237 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
238 (!compressed_size &&
239 (actual_end & (root->sectorsize - 1)) == 0) ||
240 end + 1 < isize ||
241 data_len > root->fs_info->max_inline) {
242 return 1;
243 }
244
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000245 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
Chris Masona1ed8352009-09-11 12:27:37 -0400246 &hint_byte, 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400247 BUG_ON(ret);
248
249 if (isize > actual_end)
250 inline_len = min_t(u64, isize, actual_end);
251 ret = insert_inline_extent(trans, root, inode, start,
252 inline_len, compressed_size,
253 compressed_pages);
254 BUG_ON(ret);
Chris Masona1ed8352009-09-11 12:27:37 -0400255 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
Chris Masonc8b97812008-10-29 14:49:59 -0400256 return 0;
257}
258
Chris Mason771ed682008-11-06 22:02:51 -0500259struct async_extent {
260 u64 start;
261 u64 ram_size;
262 u64 compressed_size;
263 struct page **pages;
264 unsigned long nr_pages;
265 struct list_head list;
266};
267
268struct async_cow {
269 struct inode *inode;
270 struct btrfs_root *root;
271 struct page *locked_page;
272 u64 start;
273 u64 end;
274 struct list_head extents;
275 struct btrfs_work work;
276};
277
278static noinline int add_async_extent(struct async_cow *cow,
279 u64 start, u64 ram_size,
280 u64 compressed_size,
281 struct page **pages,
282 unsigned long nr_pages)
283{
284 struct async_extent *async_extent;
285
286 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
287 async_extent->start = start;
288 async_extent->ram_size = ram_size;
289 async_extent->compressed_size = compressed_size;
290 async_extent->pages = pages;
291 async_extent->nr_pages = nr_pages;
292 list_add_tail(&async_extent->list, &cow->extents);
293 return 0;
294}
295
Chris Masonc8b97812008-10-29 14:49:59 -0400296/*
Chris Mason771ed682008-11-06 22:02:51 -0500297 * we create compressed extents in two phases. The first
298 * phase compresses a range of pages that have already been
299 * locked (both pages and state bits are locked).
Chris Masonc8b97812008-10-29 14:49:59 -0400300 *
Chris Mason771ed682008-11-06 22:02:51 -0500301 * This is done inside an ordered work queue, and the compression
302 * is spread across many cpus. The actual IO submission is step
303 * two, and the ordered work queue takes care of making sure that
304 * happens in the same order things were put onto the queue by
305 * writepages and friends.
Chris Masonc8b97812008-10-29 14:49:59 -0400306 *
Chris Mason771ed682008-11-06 22:02:51 -0500307 * If this code finds it can't get good compression, it puts an
308 * entry onto the work queue to write the uncompressed bytes. This
309 * makes sure that both compressed inodes and uncompressed inodes
310 * are written in the same order that pdflush sent them down.
Chris Masond352ac62008-09-29 15:18:18 -0400311 */
Chris Mason771ed682008-11-06 22:02:51 -0500312static noinline int compress_file_range(struct inode *inode,
313 struct page *locked_page,
314 u64 start, u64 end,
315 struct async_cow *async_cow,
316 int *num_added)
Chris Masonb888db22007-08-27 16:49:44 -0400317{
318 struct btrfs_root *root = BTRFS_I(inode)->root;
319 struct btrfs_trans_handle *trans;
Chris Masondb945352007-10-15 16:15:53 -0400320 u64 num_bytes;
Chris Masonc8b97812008-10-29 14:49:59 -0400321 u64 orig_start;
322 u64 disk_num_bytes;
Chris Masondb945352007-10-15 16:15:53 -0400323 u64 blocksize = root->sectorsize;
Chris Masonc8b97812008-10-29 14:49:59 -0400324 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500325 u64 isize = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400326 int ret = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400327 struct page **pages = NULL;
328 unsigned long nr_pages;
329 unsigned long nr_pages_ret = 0;
330 unsigned long total_compressed = 0;
331 unsigned long total_in = 0;
332 unsigned long max_compressed = 128 * 1024;
Chris Mason771ed682008-11-06 22:02:51 -0500333 unsigned long max_uncompressed = 128 * 1024;
Chris Masonc8b97812008-10-29 14:49:59 -0400334 int i;
335 int will_compress;
Chris Masonb888db22007-08-27 16:49:44 -0400336
Chris Masonc8b97812008-10-29 14:49:59 -0400337 orig_start = start;
Chris Masonbe20aa92007-12-17 20:14:01 -0500338
Chris Mason42dc7ba2008-12-15 11:44:56 -0500339 actual_end = min_t(u64, isize, end + 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400340again:
341 will_compress = 0;
342 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
343 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
344
Chris Masonf03d9302009-02-04 09:31:06 -0500345 /*
346 * we don't want to send crud past the end of i_size through
347 * compression, that's just a waste of CPU time. So, if the
348 * end of the file is before the start of our current
349 * requested range of bytes, we bail out to the uncompressed
350 * cleanup code that can deal with all of this.
351 *
352 * It isn't really the fastest way to fix things, but this is a
353 * very uncommon corner.
354 */
355 if (actual_end <= start)
356 goto cleanup_and_bail_uncompressed;
357
Chris Masonc8b97812008-10-29 14:49:59 -0400358 total_compressed = actual_end - start;
359
360 /* we want to make sure that amount of ram required to uncompress
361 * an extent is reasonable, so we limit the total size in ram
Chris Mason771ed682008-11-06 22:02:51 -0500362 * of a compressed extent to 128k. This is a crucial number
363 * because it also controls how easily we can spread reads across
364 * cpus for decompression.
365 *
366 * We also want to make sure the amount of IO required to do
367 * a random read is reasonably small, so we limit the size of
368 * a compressed extent to 128k.
Chris Masonc8b97812008-10-29 14:49:59 -0400369 */
370 total_compressed = min(total_compressed, max_uncompressed);
Chris Masondb945352007-10-15 16:15:53 -0400371 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
Chris Masonbe20aa92007-12-17 20:14:01 -0500372 num_bytes = max(blocksize, num_bytes);
Chris Masonc8b97812008-10-29 14:49:59 -0400373 disk_num_bytes = num_bytes;
374 total_in = 0;
375 ret = 0;
Chris Masondb945352007-10-15 16:15:53 -0400376
Chris Mason771ed682008-11-06 22:02:51 -0500377 /*
378 * we do compression for mount -o compress and when the
379 * inode has not been flagged as nocompress. This flag can
380 * change at any time if we discover bad compression ratios.
Chris Masonc8b97812008-10-29 14:49:59 -0400381 */
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200382 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
Chris Mason1e701a32010-03-11 09:42:04 -0500383 (btrfs_test_opt(root, COMPRESS) ||
384 (BTRFS_I(inode)->force_compress))) {
Chris Masonc8b97812008-10-29 14:49:59 -0400385 WARN_ON(pages);
Chris Masoncfbc2462008-10-30 13:22:14 -0400386 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
Chris Mason179e29e2007-11-01 11:28:41 -0400387
Chris Masonc8b97812008-10-29 14:49:59 -0400388 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
389 total_compressed, pages,
390 nr_pages, &nr_pages_ret,
391 &total_in,
392 &total_compressed,
393 max_compressed);
394
395 if (!ret) {
396 unsigned long offset = total_compressed &
397 (PAGE_CACHE_SIZE - 1);
398 struct page *page = pages[nr_pages_ret - 1];
399 char *kaddr;
400
401 /* zero the tail end of the last page, we might be
402 * sending it down to disk
403 */
404 if (offset) {
405 kaddr = kmap_atomic(page, KM_USER0);
406 memset(kaddr + offset, 0,
407 PAGE_CACHE_SIZE - offset);
408 kunmap_atomic(kaddr, KM_USER0);
409 }
410 will_compress = 1;
411 }
412 }
413 if (start == 0) {
Chris Mason771ed682008-11-06 22:02:51 -0500414 trans = btrfs_join_transaction(root, 1);
415 BUG_ON(!trans);
416 btrfs_set_trans_block_group(trans, inode);
417
Chris Masonc8b97812008-10-29 14:49:59 -0400418 /* lets try to make an inline extent */
Chris Mason771ed682008-11-06 22:02:51 -0500419 if (ret || total_in < (actual_end - start)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400420 /* we didn't compress the entire range, try
Chris Mason771ed682008-11-06 22:02:51 -0500421 * to make an uncompressed inline extent.
Chris Masonc8b97812008-10-29 14:49:59 -0400422 */
423 ret = cow_file_range_inline(trans, root, inode,
424 start, end, 0, NULL);
425 } else {
Chris Mason771ed682008-11-06 22:02:51 -0500426 /* try making a compressed inline extent */
Chris Masonc8b97812008-10-29 14:49:59 -0400427 ret = cow_file_range_inline(trans, root, inode,
428 start, end,
429 total_compressed, pages);
430 }
431 if (ret == 0) {
Chris Mason771ed682008-11-06 22:02:51 -0500432 /*
433 * inline extent creation worked, we don't need
434 * to create any more async work items. Unlock
435 * and free up our temp pages.
436 */
Chris Masonc8b97812008-10-29 14:49:59 -0400437 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400438 &BTRFS_I(inode)->io_tree,
439 start, end, NULL,
440 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
Chris Masona3429ab2009-10-08 12:30:20 -0400441 EXTENT_CLEAR_DELALLOC |
Josef Bacik32c00af2009-10-08 13:34:05 -0400442 EXTENT_CLEAR_ACCOUNTING |
Chris Masona791e352009-10-08 11:27:10 -0400443 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000444
445 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400446 goto free_pages_out;
447 }
Yan, Zhengc2167752009-11-12 09:34:21 +0000448 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400449 }
450
451 if (will_compress) {
452 /*
453 * we aren't doing an inline extent round the compressed size
454 * up to a block size boundary so the allocator does sane
455 * things
456 */
457 total_compressed = (total_compressed + blocksize - 1) &
458 ~(blocksize - 1);
459
460 /*
461 * one last check to make sure the compression is really a
462 * win, compare the page count read with the blocks on disk
463 */
464 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
465 ~(PAGE_CACHE_SIZE - 1);
466 if (total_compressed >= total_in) {
467 will_compress = 0;
468 } else {
469 disk_num_bytes = total_compressed;
470 num_bytes = total_in;
471 }
472 }
473 if (!will_compress && pages) {
474 /*
475 * the compression code ran but failed to make things smaller,
476 * free any pages it allocated and our page pointer array
477 */
478 for (i = 0; i < nr_pages_ret; i++) {
Chris Mason70b99e62008-10-31 12:46:39 -0400479 WARN_ON(pages[i]->mapping);
Chris Masonc8b97812008-10-29 14:49:59 -0400480 page_cache_release(pages[i]);
481 }
482 kfree(pages);
483 pages = NULL;
484 total_compressed = 0;
485 nr_pages_ret = 0;
486
487 /* flag the file so we don't compress in the future */
Chris Mason1e701a32010-03-11 09:42:04 -0500488 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
489 !(BTRFS_I(inode)->force_compress)) {
Chris Masona555f812010-01-28 16:18:15 -0500490 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
Chris Mason1e701a32010-03-11 09:42:04 -0500491 }
Chris Masonc8b97812008-10-29 14:49:59 -0400492 }
Chris Mason771ed682008-11-06 22:02:51 -0500493 if (will_compress) {
494 *num_added += 1;
495
496 /* the async work queues will take care of doing actual
497 * allocation on disk for these compressed pages,
498 * and will submit them to the elevator.
499 */
500 add_async_extent(async_cow, start, num_bytes,
501 total_compressed, pages, nr_pages_ret);
502
Chris Mason42dc7ba2008-12-15 11:44:56 -0500503 if (start + num_bytes < end && start + num_bytes < actual_end) {
Chris Mason771ed682008-11-06 22:02:51 -0500504 start += num_bytes;
505 pages = NULL;
506 cond_resched();
507 goto again;
508 }
509 } else {
Chris Masonf03d9302009-02-04 09:31:06 -0500510cleanup_and_bail_uncompressed:
Chris Mason771ed682008-11-06 22:02:51 -0500511 /*
512 * No compression, but we still need to write the pages in
513 * the file we've been given so far. redirty the locked
514 * page if it corresponds to our extent and set things up
515 * for the async work queue to run cow_file_range to do
516 * the normal delalloc dance
517 */
518 if (page_offset(locked_page) >= start &&
519 page_offset(locked_page) <= end) {
520 __set_page_dirty_nobuffers(locked_page);
521 /* unlocked later on in the async handlers */
522 }
523 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
524 *num_added += 1;
525 }
526
527out:
528 return 0;
529
530free_pages_out:
531 for (i = 0; i < nr_pages_ret; i++) {
532 WARN_ON(pages[i]->mapping);
533 page_cache_release(pages[i]);
534 }
Chris Masond3977122009-01-05 21:25:51 -0500535 kfree(pages);
Chris Mason771ed682008-11-06 22:02:51 -0500536
537 goto out;
538}
539
540/*
541 * phase two of compressed writeback. This is the ordered portion
542 * of the code, which only gets called in the order the work was
543 * queued. We walk all the async extents created by compress_file_range
544 * and send them down to the disk.
545 */
546static noinline int submit_compressed_extents(struct inode *inode,
547 struct async_cow *async_cow)
548{
549 struct async_extent *async_extent;
550 u64 alloc_hint = 0;
551 struct btrfs_trans_handle *trans;
552 struct btrfs_key ins;
553 struct extent_map *em;
554 struct btrfs_root *root = BTRFS_I(inode)->root;
555 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
556 struct extent_io_tree *io_tree;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500557 int ret = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500558
559 if (list_empty(&async_cow->extents))
560 return 0;
561
Chris Mason771ed682008-11-06 22:02:51 -0500562
Chris Masond3977122009-01-05 21:25:51 -0500563 while (!list_empty(&async_cow->extents)) {
Chris Mason771ed682008-11-06 22:02:51 -0500564 async_extent = list_entry(async_cow->extents.next,
565 struct async_extent, list);
566 list_del(&async_extent->list);
567
568 io_tree = &BTRFS_I(inode)->io_tree;
569
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500570retry:
Chris Mason771ed682008-11-06 22:02:51 -0500571 /* did the compression code fall back to uncompressed IO? */
572 if (!async_extent->pages) {
573 int page_started = 0;
574 unsigned long nr_written = 0;
575
576 lock_extent(io_tree, async_extent->start,
Josef Bacik2ac55d42010-02-03 19:33:23 +0000577 async_extent->start +
578 async_extent->ram_size - 1, GFP_NOFS);
Chris Mason771ed682008-11-06 22:02:51 -0500579
580 /* allocate blocks */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500581 ret = cow_file_range(inode, async_cow->locked_page,
582 async_extent->start,
583 async_extent->start +
584 async_extent->ram_size - 1,
585 &page_started, &nr_written, 0);
Chris Mason771ed682008-11-06 22:02:51 -0500586
587 /*
588 * if page_started, cow_file_range inserted an
589 * inline extent and took care of all the unlocking
590 * and IO for us. Otherwise, we need to submit
591 * all those pages down to the drive.
592 */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500593 if (!page_started && !ret)
Chris Mason771ed682008-11-06 22:02:51 -0500594 extent_write_locked_range(io_tree,
595 inode, async_extent->start,
Chris Masond3977122009-01-05 21:25:51 -0500596 async_extent->start +
Chris Mason771ed682008-11-06 22:02:51 -0500597 async_extent->ram_size - 1,
598 btrfs_get_extent,
599 WB_SYNC_ALL);
600 kfree(async_extent);
601 cond_resched();
602 continue;
603 }
604
605 lock_extent(io_tree, async_extent->start,
606 async_extent->start + async_extent->ram_size - 1,
607 GFP_NOFS);
Chris Mason771ed682008-11-06 22:02:51 -0500608
Yan, Zhengc2167752009-11-12 09:34:21 +0000609 trans = btrfs_join_transaction(root, 1);
Chris Mason771ed682008-11-06 22:02:51 -0500610 ret = btrfs_reserve_extent(trans, root,
611 async_extent->compressed_size,
612 async_extent->compressed_size,
613 0, alloc_hint,
614 (u64)-1, &ins, 1);
Yan, Zhengc2167752009-11-12 09:34:21 +0000615 btrfs_end_transaction(trans, root);
616
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500617 if (ret) {
618 int i;
619 for (i = 0; i < async_extent->nr_pages; i++) {
620 WARN_ON(async_extent->pages[i]->mapping);
621 page_cache_release(async_extent->pages[i]);
622 }
623 kfree(async_extent->pages);
624 async_extent->nr_pages = 0;
625 async_extent->pages = NULL;
626 unlock_extent(io_tree, async_extent->start,
627 async_extent->start +
628 async_extent->ram_size - 1, GFP_NOFS);
629 goto retry;
630 }
631
Yan, Zhengc2167752009-11-12 09:34:21 +0000632 /*
633 * here we're doing allocation and writeback of the
634 * compressed pages
635 */
636 btrfs_drop_extent_cache(inode, async_extent->start,
637 async_extent->start +
638 async_extent->ram_size - 1, 0);
639
Chris Mason771ed682008-11-06 22:02:51 -0500640 em = alloc_extent_map(GFP_NOFS);
641 em->start = async_extent->start;
642 em->len = async_extent->ram_size;
Chris Mason445a6942008-11-10 11:53:33 -0500643 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500644
645 em->block_start = ins.objectid;
646 em->block_len = ins.offset;
647 em->bdev = root->fs_info->fs_devices->latest_bdev;
648 set_bit(EXTENT_FLAG_PINNED, &em->flags);
649 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
650
Chris Masond3977122009-01-05 21:25:51 -0500651 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400652 write_lock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500653 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -0400654 write_unlock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500655 if (ret != -EEXIST) {
656 free_extent_map(em);
657 break;
658 }
659 btrfs_drop_extent_cache(inode, async_extent->start,
660 async_extent->start +
661 async_extent->ram_size - 1, 0);
662 }
663
664 ret = btrfs_add_ordered_extent(inode, async_extent->start,
665 ins.objectid,
666 async_extent->ram_size,
667 ins.offset,
668 BTRFS_ORDERED_COMPRESSED);
669 BUG_ON(ret);
670
Chris Mason771ed682008-11-06 22:02:51 -0500671 /*
672 * clear dirty, set writeback and unlock the pages.
673 */
674 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400675 &BTRFS_I(inode)->io_tree,
676 async_extent->start,
677 async_extent->start +
678 async_extent->ram_size - 1,
679 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
680 EXTENT_CLEAR_UNLOCK |
Chris Masona3429ab2009-10-08 12:30:20 -0400681 EXTENT_CLEAR_DELALLOC |
Chris Masona791e352009-10-08 11:27:10 -0400682 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
Chris Mason771ed682008-11-06 22:02:51 -0500683
684 ret = btrfs_submit_compressed_write(inode,
Chris Masond3977122009-01-05 21:25:51 -0500685 async_extent->start,
686 async_extent->ram_size,
687 ins.objectid,
688 ins.offset, async_extent->pages,
689 async_extent->nr_pages);
Chris Mason771ed682008-11-06 22:02:51 -0500690
691 BUG_ON(ret);
Chris Mason771ed682008-11-06 22:02:51 -0500692 alloc_hint = ins.objectid + ins.offset;
693 kfree(async_extent);
694 cond_resched();
695 }
696
Chris Mason771ed682008-11-06 22:02:51 -0500697 return 0;
698}
699
700/*
701 * when extent_io.c finds a delayed allocation range in the file,
702 * the call backs end up in this code. The basic idea is to
703 * allocate extents on disk for the range, and create ordered data structs
704 * in ram to track those extents.
705 *
706 * locked_page is the page that writepage had locked already. We use
707 * it to make sure we don't do extra locks or unlocks.
708 *
709 * *page_started is set to one if we unlock locked_page and do everything
710 * required to start IO on it. It may be clean and already done with
711 * IO when we return.
712 */
713static noinline int cow_file_range(struct inode *inode,
714 struct page *locked_page,
715 u64 start, u64 end, int *page_started,
716 unsigned long *nr_written,
717 int unlock)
718{
719 struct btrfs_root *root = BTRFS_I(inode)->root;
720 struct btrfs_trans_handle *trans;
721 u64 alloc_hint = 0;
722 u64 num_bytes;
723 unsigned long ram_size;
724 u64 disk_num_bytes;
725 u64 cur_alloc_size;
726 u64 blocksize = root->sectorsize;
727 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500728 u64 isize = i_size_read(inode);
Chris Mason771ed682008-11-06 22:02:51 -0500729 struct btrfs_key ins;
730 struct extent_map *em;
731 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
732 int ret = 0;
733
734 trans = btrfs_join_transaction(root, 1);
735 BUG_ON(!trans);
736 btrfs_set_trans_block_group(trans, inode);
737
Chris Mason42dc7ba2008-12-15 11:44:56 -0500738 actual_end = min_t(u64, isize, end + 1);
Chris Mason771ed682008-11-06 22:02:51 -0500739
740 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
741 num_bytes = max(blocksize, num_bytes);
742 disk_num_bytes = num_bytes;
743 ret = 0;
744
745 if (start == 0) {
746 /* lets try to make an inline extent */
747 ret = cow_file_range_inline(trans, root, inode,
748 start, end, 0, NULL);
749 if (ret == 0) {
750 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400751 &BTRFS_I(inode)->io_tree,
752 start, end, NULL,
753 EXTENT_CLEAR_UNLOCK_PAGE |
754 EXTENT_CLEAR_UNLOCK |
755 EXTENT_CLEAR_DELALLOC |
Josef Bacik32c00af2009-10-08 13:34:05 -0400756 EXTENT_CLEAR_ACCOUNTING |
Chris Masona791e352009-10-08 11:27:10 -0400757 EXTENT_CLEAR_DIRTY |
758 EXTENT_SET_WRITEBACK |
759 EXTENT_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000760
Chris Mason771ed682008-11-06 22:02:51 -0500761 *nr_written = *nr_written +
762 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
763 *page_started = 1;
764 ret = 0;
765 goto out;
766 }
767 }
Chris Masonc8b97812008-10-29 14:49:59 -0400768
769 BUG_ON(disk_num_bytes >
770 btrfs_super_total_bytes(&root->fs_info->super_copy));
771
Chris Masonb917b7c2009-09-18 16:07:03 -0400772
773 read_lock(&BTRFS_I(inode)->extent_tree.lock);
774 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
775 start, num_bytes);
776 if (em) {
Josef Bacik6346c932009-11-10 21:23:47 -0500777 /*
778 * if block start isn't an actual block number then find the
779 * first block in this inode and use that as a hint. If that
780 * block is also bogus then just don't worry about it.
781 */
782 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
783 free_extent_map(em);
784 em = search_extent_mapping(em_tree, 0, 0);
785 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
786 alloc_hint = em->block_start;
787 if (em)
788 free_extent_map(em);
789 } else {
790 alloc_hint = em->block_start;
791 free_extent_map(em);
792 }
Chris Masonb917b7c2009-09-18 16:07:03 -0400793 }
794 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400795 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
Chris Mason3b951512008-04-17 11:29:12 -0400796
Chris Masond3977122009-01-05 21:25:51 -0500797 while (disk_num_bytes > 0) {
Chris Masona791e352009-10-08 11:27:10 -0400798 unsigned long op;
799
Josef Bacik287a0ab2010-03-19 18:07:23 +0000800 cur_alloc_size = disk_num_bytes;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400801 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
Chris Mason771ed682008-11-06 22:02:51 -0500802 root->sectorsize, 0, alloc_hint,
Chris Masone6dcd2d2008-07-17 12:53:50 -0400803 (u64)-1, &ins, 1);
Chris Masond3977122009-01-05 21:25:51 -0500804 BUG_ON(ret);
805
Chris Masone6dcd2d2008-07-17 12:53:50 -0400806 em = alloc_extent_map(GFP_NOFS);
807 em->start = start;
Chris Mason445a6942008-11-10 11:53:33 -0500808 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500809 ram_size = ins.offset;
810 em->len = ins.offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400811
Chris Masone6dcd2d2008-07-17 12:53:50 -0400812 em->block_start = ins.objectid;
Chris Masonc8b97812008-10-29 14:49:59 -0400813 em->block_len = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400814 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason7f3c74f2008-07-18 12:01:11 -0400815 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -0400816
Chris Masond3977122009-01-05 21:25:51 -0500817 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400818 write_lock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400819 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -0400820 write_unlock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400821 if (ret != -EEXIST) {
822 free_extent_map(em);
823 break;
824 }
825 btrfs_drop_extent_cache(inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -0400826 start + ram_size - 1, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400827 }
828
Chris Mason98d20f62008-04-14 09:46:10 -0400829 cur_alloc_size = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400830 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
Chris Mason771ed682008-11-06 22:02:51 -0500831 ram_size, cur_alloc_size, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400832 BUG_ON(ret);
Chris Masonc8b97812008-10-29 14:49:59 -0400833
Yan Zheng17d217f2008-12-12 10:03:38 -0500834 if (root->root_key.objectid ==
835 BTRFS_DATA_RELOC_TREE_OBJECTID) {
836 ret = btrfs_reloc_clone_csums(inode, start,
837 cur_alloc_size);
838 BUG_ON(ret);
839 }
840
Chris Masond3977122009-01-05 21:25:51 -0500841 if (disk_num_bytes < cur_alloc_size)
Chris Mason3b951512008-04-17 11:29:12 -0400842 break;
Chris Masond3977122009-01-05 21:25:51 -0500843
Chris Masonc8b97812008-10-29 14:49:59 -0400844 /* we're not doing compressed IO, don't unlock the first
845 * page (which the caller expects to stay locked), don't
846 * clear any dirty bits and don't set any writeback bits
Chris Mason8b62b722009-09-02 16:53:46 -0400847 *
848 * Do set the Private2 bit so we know this page was properly
849 * setup for writepage
Chris Masonc8b97812008-10-29 14:49:59 -0400850 */
Chris Masona791e352009-10-08 11:27:10 -0400851 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
852 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
853 EXTENT_SET_PRIVATE2;
854
Chris Masonc8b97812008-10-29 14:49:59 -0400855 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
856 start, start + ram_size - 1,
Chris Masona791e352009-10-08 11:27:10 -0400857 locked_page, op);
Chris Masonc8b97812008-10-29 14:49:59 -0400858 disk_num_bytes -= cur_alloc_size;
Chris Masonc59f8952007-12-17 20:14:04 -0500859 num_bytes -= cur_alloc_size;
860 alloc_hint = ins.objectid + ins.offset;
861 start += cur_alloc_size;
Chris Masonb888db22007-08-27 16:49:44 -0400862 }
Chris Masonb888db22007-08-27 16:49:44 -0400863out:
Chris Mason771ed682008-11-06 22:02:51 -0500864 ret = 0;
Chris Masonb888db22007-08-27 16:49:44 -0400865 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400866
Chris Masonbe20aa92007-12-17 20:14:01 -0500867 return ret;
Chris Mason771ed682008-11-06 22:02:51 -0500868}
Chris Masonc8b97812008-10-29 14:49:59 -0400869
Chris Mason771ed682008-11-06 22:02:51 -0500870/*
871 * work queue call back to started compression on a file and pages
872 */
873static noinline void async_cow_start(struct btrfs_work *work)
874{
875 struct async_cow *async_cow;
876 int num_added = 0;
877 async_cow = container_of(work, struct async_cow, work);
878
879 compress_file_range(async_cow->inode, async_cow->locked_page,
880 async_cow->start, async_cow->end, async_cow,
881 &num_added);
882 if (num_added == 0)
883 async_cow->inode = NULL;
884}
885
886/*
887 * work queue call back to submit previously compressed pages
888 */
889static noinline void async_cow_submit(struct btrfs_work *work)
890{
891 struct async_cow *async_cow;
892 struct btrfs_root *root;
893 unsigned long nr_pages;
894
895 async_cow = container_of(work, struct async_cow, work);
896
897 root = async_cow->root;
898 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
899 PAGE_CACHE_SHIFT;
900
901 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
902
903 if (atomic_read(&root->fs_info->async_delalloc_pages) <
904 5 * 1042 * 1024 &&
905 waitqueue_active(&root->fs_info->async_submit_wait))
906 wake_up(&root->fs_info->async_submit_wait);
907
Chris Masond3977122009-01-05 21:25:51 -0500908 if (async_cow->inode)
Chris Mason771ed682008-11-06 22:02:51 -0500909 submit_compressed_extents(async_cow->inode, async_cow);
Chris Mason771ed682008-11-06 22:02:51 -0500910}
Chris Masonc8b97812008-10-29 14:49:59 -0400911
Chris Mason771ed682008-11-06 22:02:51 -0500912static noinline void async_cow_free(struct btrfs_work *work)
913{
914 struct async_cow *async_cow;
915 async_cow = container_of(work, struct async_cow, work);
916 kfree(async_cow);
917}
918
919static int cow_file_range_async(struct inode *inode, struct page *locked_page,
920 u64 start, u64 end, int *page_started,
921 unsigned long *nr_written)
922{
923 struct async_cow *async_cow;
924 struct btrfs_root *root = BTRFS_I(inode)->root;
925 unsigned long nr_pages;
926 u64 cur_end;
927 int limit = 10 * 1024 * 1042;
928
Chris Masona3429ab2009-10-08 12:30:20 -0400929 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
930 1, 0, NULL, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -0500931 while (start < end) {
Chris Mason771ed682008-11-06 22:02:51 -0500932 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
933 async_cow->inode = inode;
934 async_cow->root = root;
935 async_cow->locked_page = locked_page;
936 async_cow->start = start;
937
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200938 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
Chris Mason771ed682008-11-06 22:02:51 -0500939 cur_end = end;
940 else
941 cur_end = min(end, start + 512 * 1024 - 1);
942
943 async_cow->end = cur_end;
944 INIT_LIST_HEAD(&async_cow->extents);
945
946 async_cow->work.func = async_cow_start;
947 async_cow->work.ordered_func = async_cow_submit;
948 async_cow->work.ordered_free = async_cow_free;
949 async_cow->work.flags = 0;
950
Chris Mason771ed682008-11-06 22:02:51 -0500951 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
952 PAGE_CACHE_SHIFT;
953 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
954
955 btrfs_queue_worker(&root->fs_info->delalloc_workers,
956 &async_cow->work);
957
958 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
959 wait_event(root->fs_info->async_submit_wait,
960 (atomic_read(&root->fs_info->async_delalloc_pages) <
961 limit));
962 }
963
Chris Masond3977122009-01-05 21:25:51 -0500964 while (atomic_read(&root->fs_info->async_submit_draining) &&
Chris Mason771ed682008-11-06 22:02:51 -0500965 atomic_read(&root->fs_info->async_delalloc_pages)) {
966 wait_event(root->fs_info->async_submit_wait,
967 (atomic_read(&root->fs_info->async_delalloc_pages) ==
968 0));
969 }
970
971 *nr_written += nr_pages;
972 start = cur_end + 1;
973 }
974 *page_started = 1;
975 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -0500976}
977
Chris Masond3977122009-01-05 21:25:51 -0500978static noinline int csum_exist_in_range(struct btrfs_root *root,
Yan Zheng17d217f2008-12-12 10:03:38 -0500979 u64 bytenr, u64 num_bytes)
980{
981 int ret;
982 struct btrfs_ordered_sum *sums;
983 LIST_HEAD(list);
984
Yan Zheng07d400a2009-01-06 11:42:00 -0500985 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
986 bytenr + num_bytes - 1, &list);
Yan Zheng17d217f2008-12-12 10:03:38 -0500987 if (ret == 0 && list_empty(&list))
988 return 0;
989
990 while (!list_empty(&list)) {
991 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
992 list_del(&sums->list);
993 kfree(sums);
994 }
995 return 1;
996}
997
Chris Masond352ac62008-09-29 15:18:18 -0400998/*
999 * when nowcow writeback call back. This checks for snapshots or COW copies
1000 * of the extents that exist in the file, and COWs the file as required.
1001 *
1002 * If no cow copies or snapshots exist, we write directly to the existing
1003 * blocks on disk
1004 */
Chris Mason7f366cf2009-03-12 20:12:45 -04001005static noinline int run_delalloc_nocow(struct inode *inode,
1006 struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001007 u64 start, u64 end, int *page_started, int force,
1008 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001009{
Chris Masonbe20aa92007-12-17 20:14:01 -05001010 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001011 struct btrfs_trans_handle *trans;
Chris Masonbe20aa92007-12-17 20:14:01 -05001012 struct extent_buffer *leaf;
Chris Masonbe20aa92007-12-17 20:14:01 -05001013 struct btrfs_path *path;
Yan Zheng80ff3852008-10-30 14:20:02 -04001014 struct btrfs_file_extent_item *fi;
Chris Masonbe20aa92007-12-17 20:14:01 -05001015 struct btrfs_key found_key;
Yan Zheng80ff3852008-10-30 14:20:02 -04001016 u64 cow_start;
1017 u64 cur_offset;
1018 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001019 u64 extent_offset;
Yan Zheng80ff3852008-10-30 14:20:02 -04001020 u64 disk_bytenr;
1021 u64 num_bytes;
1022 int extent_type;
1023 int ret;
Yan Zhengd899e052008-10-30 14:25:28 -04001024 int type;
Yan Zheng80ff3852008-10-30 14:20:02 -04001025 int nocow;
1026 int check_prev = 1;
Chris Masonbe20aa92007-12-17 20:14:01 -05001027
1028 path = btrfs_alloc_path();
1029 BUG_ON(!path);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001030 trans = btrfs_join_transaction(root, 1);
1031 BUG_ON(!trans);
Chris Masonbe20aa92007-12-17 20:14:01 -05001032
Yan Zheng80ff3852008-10-30 14:20:02 -04001033 cow_start = (u64)-1;
1034 cur_offset = start;
1035 while (1) {
1036 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1037 cur_offset, 0);
1038 BUG_ON(ret < 0);
1039 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1040 leaf = path->nodes[0];
1041 btrfs_item_key_to_cpu(leaf, &found_key,
1042 path->slots[0] - 1);
1043 if (found_key.objectid == inode->i_ino &&
1044 found_key.type == BTRFS_EXTENT_DATA_KEY)
1045 path->slots[0]--;
1046 }
1047 check_prev = 0;
1048next_slot:
1049 leaf = path->nodes[0];
1050 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1051 ret = btrfs_next_leaf(root, path);
1052 if (ret < 0)
1053 BUG_ON(1);
1054 if (ret > 0)
1055 break;
1056 leaf = path->nodes[0];
1057 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001058
Yan Zheng80ff3852008-10-30 14:20:02 -04001059 nocow = 0;
1060 disk_bytenr = 0;
Yan Zheng17d217f2008-12-12 10:03:38 -05001061 num_bytes = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001062 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05001063
Yan Zheng80ff3852008-10-30 14:20:02 -04001064 if (found_key.objectid > inode->i_ino ||
1065 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1066 found_key.offset > end)
1067 break;
Chris Masonbe20aa92007-12-17 20:14:01 -05001068
Yan Zheng80ff3852008-10-30 14:20:02 -04001069 if (found_key.offset > cur_offset) {
1070 extent_end = found_key.offset;
Chris Masone9061e22009-10-09 09:57:45 -04001071 extent_type = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001072 goto out_check;
1073 }
Chris Masonc31f8832008-01-08 15:46:31 -05001074
Yan Zheng80ff3852008-10-30 14:20:02 -04001075 fi = btrfs_item_ptr(leaf, path->slots[0],
1076 struct btrfs_file_extent_item);
1077 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Masonbe20aa92007-12-17 20:14:01 -05001078
Yan Zhengd899e052008-10-30 14:25:28 -04001079 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1080 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001081 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001082 extent_offset = btrfs_file_extent_offset(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001083 extent_end = found_key.offset +
1084 btrfs_file_extent_num_bytes(leaf, fi);
1085 if (extent_end <= start) {
1086 path->slots[0]++;
1087 goto next_slot;
1088 }
Yan Zheng17d217f2008-12-12 10:03:38 -05001089 if (disk_bytenr == 0)
1090 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001091 if (btrfs_file_extent_compression(leaf, fi) ||
1092 btrfs_file_extent_encryption(leaf, fi) ||
1093 btrfs_file_extent_other_encoding(leaf, fi))
1094 goto out_check;
Yan Zhengd899e052008-10-30 14:25:28 -04001095 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1096 goto out_check;
Yan Zhengd2fb3432008-12-11 16:30:39 -05001097 if (btrfs_extent_readonly(root, disk_bytenr))
Yan Zheng80ff3852008-10-30 14:20:02 -04001098 goto out_check;
Yan Zheng17d217f2008-12-12 10:03:38 -05001099 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001100 found_key.offset -
1101 extent_offset, disk_bytenr))
Yan Zheng17d217f2008-12-12 10:03:38 -05001102 goto out_check;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001103 disk_bytenr += extent_offset;
Yan Zheng17d217f2008-12-12 10:03:38 -05001104 disk_bytenr += cur_offset - found_key.offset;
1105 num_bytes = min(end + 1, extent_end) - cur_offset;
1106 /*
1107 * force cow if csum exists in the range.
1108 * this ensure that csum for a given extent are
1109 * either valid or do not exist.
1110 */
1111 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1112 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001113 nocow = 1;
1114 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1115 extent_end = found_key.offset +
1116 btrfs_file_extent_inline_len(leaf, fi);
1117 extent_end = ALIGN(extent_end, root->sectorsize);
1118 } else {
1119 BUG_ON(1);
1120 }
1121out_check:
1122 if (extent_end <= start) {
1123 path->slots[0]++;
1124 goto next_slot;
1125 }
1126 if (!nocow) {
1127 if (cow_start == (u64)-1)
1128 cow_start = cur_offset;
1129 cur_offset = extent_end;
1130 if (cur_offset > end)
1131 break;
1132 path->slots[0]++;
1133 goto next_slot;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001134 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001135
Yan Zheng7ea394f2008-08-05 13:05:02 -04001136 btrfs_release_path(root, path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001137 if (cow_start != (u64)-1) {
1138 ret = cow_file_range(inode, locked_page, cow_start,
Chris Mason771ed682008-11-06 22:02:51 -05001139 found_key.offset - 1, page_started,
1140 nr_written, 1);
Yan Zheng80ff3852008-10-30 14:20:02 -04001141 BUG_ON(ret);
1142 cow_start = (u64)-1;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001143 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001144
Yan Zhengd899e052008-10-30 14:25:28 -04001145 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1146 struct extent_map *em;
1147 struct extent_map_tree *em_tree;
1148 em_tree = &BTRFS_I(inode)->extent_tree;
1149 em = alloc_extent_map(GFP_NOFS);
1150 em->start = cur_offset;
Chris Mason445a6942008-11-10 11:53:33 -05001151 em->orig_start = em->start;
Yan Zhengd899e052008-10-30 14:25:28 -04001152 em->len = num_bytes;
1153 em->block_len = num_bytes;
1154 em->block_start = disk_bytenr;
1155 em->bdev = root->fs_info->fs_devices->latest_bdev;
1156 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1157 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04001158 write_lock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001159 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04001160 write_unlock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001161 if (ret != -EEXIST) {
1162 free_extent_map(em);
1163 break;
1164 }
1165 btrfs_drop_extent_cache(inode, em->start,
1166 em->start + em->len - 1, 0);
1167 }
1168 type = BTRFS_ORDERED_PREALLOC;
1169 } else {
1170 type = BTRFS_ORDERED_NOCOW;
1171 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001172
1173 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
Yan Zhengd899e052008-10-30 14:25:28 -04001174 num_bytes, num_bytes, type);
1175 BUG_ON(ret);
Chris Mason771ed682008-11-06 22:02:51 -05001176
Yan Zhengd899e052008-10-30 14:25:28 -04001177 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
Chris Masona791e352009-10-08 11:27:10 -04001178 cur_offset, cur_offset + num_bytes - 1,
1179 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1180 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1181 EXTENT_SET_PRIVATE2);
Yan Zheng80ff3852008-10-30 14:20:02 -04001182 cur_offset = extent_end;
1183 if (cur_offset > end)
1184 break;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001185 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001186 btrfs_release_path(root, path);
1187
1188 if (cur_offset <= end && cow_start == (u64)-1)
1189 cow_start = cur_offset;
1190 if (cow_start != (u64)-1) {
1191 ret = cow_file_range(inode, locked_page, cow_start, end,
Chris Mason771ed682008-11-06 22:02:51 -05001192 page_started, nr_written, 1);
Yan Zheng80ff3852008-10-30 14:20:02 -04001193 BUG_ON(ret);
1194 }
1195
1196 ret = btrfs_end_transaction(trans, root);
1197 BUG_ON(ret);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001198 btrfs_free_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001199 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05001200}
1201
Chris Masond352ac62008-09-29 15:18:18 -04001202/*
1203 * extent_io.c call back to do delayed allocation processing
1204 */
Chris Masonc8b97812008-10-29 14:49:59 -04001205static int run_delalloc_range(struct inode *inode, struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001206 u64 start, u64 end, int *page_started,
1207 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001208{
Chris Masonbe20aa92007-12-17 20:14:01 -05001209 int ret;
Chris Mason7f366cf2009-03-12 20:12:45 -04001210 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masona2135012008-06-25 16:01:30 -04001211
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001212 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
Chris Masonc8b97812008-10-29 14:49:59 -04001213 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001214 page_started, 1, nr_written);
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001215 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
Yan Zhengd899e052008-10-30 14:25:28 -04001216 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001217 page_started, 0, nr_written);
Chris Mason1e701a32010-03-11 09:42:04 -05001218 else if (!btrfs_test_opt(root, COMPRESS) &&
1219 !(BTRFS_I(inode)->force_compress))
Chris Mason7f366cf2009-03-12 20:12:45 -04001220 ret = cow_file_range(inode, locked_page, start, end,
1221 page_started, nr_written, 1);
Chris Masonbe20aa92007-12-17 20:14:01 -05001222 else
Chris Mason771ed682008-11-06 22:02:51 -05001223 ret = cow_file_range_async(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001224 page_started, nr_written);
Chris Masonb888db22007-08-27 16:49:44 -04001225 return ret;
1226}
1227
Josef Bacik9ed74f22009-09-11 16:12:44 -04001228static int btrfs_split_extent_hook(struct inode *inode,
1229 struct extent_state *orig, u64 split)
1230{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001231 if (!(orig->state & EXTENT_DELALLOC))
1232 return 0;
1233
Josef Bacik32c00af2009-10-08 13:34:05 -04001234 spin_lock(&BTRFS_I(inode)->accounting_lock);
1235 BTRFS_I(inode)->outstanding_extents++;
1236 spin_unlock(&BTRFS_I(inode)->accounting_lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001237
1238 return 0;
1239}
1240
1241/*
1242 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1243 * extents so we can keep track of new extents that are just merged onto old
1244 * extents, such as when we are doing sequential writes, so we can properly
1245 * account for the metadata space we'll need.
1246 */
1247static int btrfs_merge_extent_hook(struct inode *inode,
1248 struct extent_state *new,
1249 struct extent_state *other)
1250{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001251 /* not delalloc, ignore it */
1252 if (!(other->state & EXTENT_DELALLOC))
1253 return 0;
1254
Josef Bacik32c00af2009-10-08 13:34:05 -04001255 spin_lock(&BTRFS_I(inode)->accounting_lock);
1256 BTRFS_I(inode)->outstanding_extents--;
1257 spin_unlock(&BTRFS_I(inode)->accounting_lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001258
1259 return 0;
1260}
1261
Chris Masond352ac62008-09-29 15:18:18 -04001262/*
1263 * extent_io.c set_bit_hook, used to track delayed allocation
1264 * bytes in this file, and to maintain the list of inodes that
1265 * have pending delalloc work to be done.
1266 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001267static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
Chris Masonb0c68f82008-01-31 11:05:37 -05001268 unsigned long old, unsigned long bits)
Chris Mason291d6732008-01-29 15:55:23 -05001269{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001270
Chris Mason75eff682008-12-15 15:54:40 -05001271 /*
1272 * set_bit and clear bit hooks normally require _irqsave/restore
1273 * but in this case, we are only testeing for the DELALLOC
1274 * bit, which is only set or cleared with irqs on
1275 */
Chris Masonb0c68f82008-01-31 11:05:37 -05001276 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001277 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001278
Josef Bacik32c00af2009-10-08 13:34:05 -04001279 spin_lock(&BTRFS_I(inode)->accounting_lock);
1280 BTRFS_I(inode)->outstanding_extents++;
1281 spin_unlock(&BTRFS_I(inode)->accounting_lock);
Josef Bacik6a632092009-02-20 11:00:09 -05001282 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
Josef Bacik287a0ab2010-03-19 18:07:23 +00001283
Chris Mason75eff682008-12-15 15:54:40 -05001284 spin_lock(&root->fs_info->delalloc_lock);
Chris Mason90692182008-02-08 13:49:28 -05001285 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
Chris Mason291d6732008-01-29 15:55:23 -05001286 root->fs_info->delalloc_bytes += end - start + 1;
Chris Masonea8c2812008-08-04 23:17:27 -04001287 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1288 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1289 &root->fs_info->delalloc_inodes);
1290 }
Chris Mason75eff682008-12-15 15:54:40 -05001291 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason291d6732008-01-29 15:55:23 -05001292 }
1293 return 0;
1294}
1295
Chris Masond352ac62008-09-29 15:18:18 -04001296/*
1297 * extent_io.c clear_bit_hook, see set_bit_hook for why
1298 */
Josef Bacik9ed74f22009-09-11 16:12:44 -04001299static int btrfs_clear_bit_hook(struct inode *inode,
1300 struct extent_state *state, unsigned long bits)
Chris Mason291d6732008-01-29 15:55:23 -05001301{
Chris Mason75eff682008-12-15 15:54:40 -05001302 /*
1303 * set_bit and clear bit hooks normally require _irqsave/restore
1304 * but in this case, we are only testeing for the DELALLOC
1305 * bit, which is only set or cleared with irqs on
1306 */
Josef Bacik9ed74f22009-09-11 16:12:44 -04001307 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001308 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masonbcbfce82008-04-22 13:26:47 -04001309
Josef Bacik32c00af2009-10-08 13:34:05 -04001310 if (bits & EXTENT_DO_ACCOUNTING) {
1311 spin_lock(&BTRFS_I(inode)->accounting_lock);
Josef Bacik287a0ab2010-03-19 18:07:23 +00001312 WARN_ON(!BTRFS_I(inode)->outstanding_extents);
Josef Bacik32c00af2009-10-08 13:34:05 -04001313 BTRFS_I(inode)->outstanding_extents--;
1314 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1315 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1316 }
Josef Bacik9ed74f22009-09-11 16:12:44 -04001317
Chris Mason75eff682008-12-15 15:54:40 -05001318 spin_lock(&root->fs_info->delalloc_lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001319 if (state->end - state->start + 1 >
1320 root->fs_info->delalloc_bytes) {
Chris Masond3977122009-01-05 21:25:51 -05001321 printk(KERN_INFO "btrfs warning: delalloc account "
1322 "%llu %llu\n",
Josef Bacik9ed74f22009-09-11 16:12:44 -04001323 (unsigned long long)
1324 state->end - state->start + 1,
Chris Masond3977122009-01-05 21:25:51 -05001325 (unsigned long long)
1326 root->fs_info->delalloc_bytes);
Josef Bacik6a632092009-02-20 11:00:09 -05001327 btrfs_delalloc_free_space(root, inode, (u64)-1);
Chris Masonb0c68f82008-01-31 11:05:37 -05001328 root->fs_info->delalloc_bytes = 0;
Chris Mason90692182008-02-08 13:49:28 -05001329 BTRFS_I(inode)->delalloc_bytes = 0;
Chris Masonb0c68f82008-01-31 11:05:37 -05001330 } else {
Josef Bacik6a632092009-02-20 11:00:09 -05001331 btrfs_delalloc_free_space(root, inode,
Josef Bacik9ed74f22009-09-11 16:12:44 -04001332 state->end -
1333 state->start + 1);
1334 root->fs_info->delalloc_bytes -= state->end -
1335 state->start + 1;
1336 BTRFS_I(inode)->delalloc_bytes -= state->end -
1337 state->start + 1;
Chris Masonb0c68f82008-01-31 11:05:37 -05001338 }
Chris Masonea8c2812008-08-04 23:17:27 -04001339 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1340 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1341 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1342 }
Chris Mason75eff682008-12-15 15:54:40 -05001343 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason291d6732008-01-29 15:55:23 -05001344 }
1345 return 0;
1346}
1347
Chris Masond352ac62008-09-29 15:18:18 -04001348/*
1349 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1350 * we don't create bios that span stripes or chunks
1351 */
Chris Mason239b14b2008-03-24 15:02:07 -04001352int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
Chris Masonc8b97812008-10-29 14:49:59 -04001353 size_t size, struct bio *bio,
1354 unsigned long bio_flags)
Chris Mason239b14b2008-03-24 15:02:07 -04001355{
1356 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1357 struct btrfs_mapping_tree *map_tree;
Chris Masona62b9402008-10-03 16:31:08 -04001358 u64 logical = (u64)bio->bi_sector << 9;
Chris Mason239b14b2008-03-24 15:02:07 -04001359 u64 length = 0;
1360 u64 map_length;
Chris Mason239b14b2008-03-24 15:02:07 -04001361 int ret;
1362
Chris Mason771ed682008-11-06 22:02:51 -05001363 if (bio_flags & EXTENT_BIO_COMPRESSED)
1364 return 0;
1365
Chris Masonf2d8d742008-04-21 10:03:05 -04001366 length = bio->bi_size;
Chris Mason239b14b2008-03-24 15:02:07 -04001367 map_tree = &root->fs_info->mapping_tree;
1368 map_length = length;
Chris Masoncea9e442008-04-09 16:28:12 -04001369 ret = btrfs_map_block(map_tree, READ, logical,
Chris Masonf1885912008-04-09 16:28:12 -04001370 &map_length, NULL, 0);
Chris Masoncea9e442008-04-09 16:28:12 -04001371
Chris Masond3977122009-01-05 21:25:51 -05001372 if (map_length < length + size)
Chris Mason239b14b2008-03-24 15:02:07 -04001373 return 1;
Chris Mason239b14b2008-03-24 15:02:07 -04001374 return 0;
1375}
1376
Chris Masond352ac62008-09-29 15:18:18 -04001377/*
1378 * in order to insert checksums into the metadata in large chunks,
1379 * we wait until bio submission time. All the pages in the bio are
1380 * checksummed and sums are attached onto the ordered extent record.
1381 *
1382 * At IO completion time the cums attached on the ordered extent record
1383 * are inserted into the btree
1384 */
Chris Masond3977122009-01-05 21:25:51 -05001385static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1386 struct bio *bio, int mirror_num,
1387 unsigned long bio_flags)
Chris Mason065631f2008-02-20 12:07:25 -05001388{
Chris Mason065631f2008-02-20 12:07:25 -05001389 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason065631f2008-02-20 12:07:25 -05001390 int ret = 0;
Chris Masone0156402008-04-16 11:15:20 -04001391
Chris Masond20f7042008-12-08 16:58:54 -05001392 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
Chris Mason44b8bd72008-04-16 11:14:51 -04001393 BUG_ON(ret);
Chris Mason4a69a412008-11-06 22:03:00 -05001394 return 0;
1395}
Chris Masone0156402008-04-16 11:15:20 -04001396
Chris Mason4a69a412008-11-06 22:03:00 -05001397/*
1398 * in order to insert checksums into the metadata in large chunks,
1399 * we wait until bio submission time. All the pages in the bio are
1400 * checksummed and sums are attached onto the ordered extent record.
1401 *
1402 * At IO completion time the cums attached on the ordered extent record
1403 * are inserted into the btree
1404 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001405static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
Chris Mason4a69a412008-11-06 22:03:00 -05001406 int mirror_num, unsigned long bio_flags)
1407{
1408 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason8b712842008-06-11 16:50:36 -04001409 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
Chris Mason44b8bd72008-04-16 11:14:51 -04001410}
1411
Chris Masond352ac62008-09-29 15:18:18 -04001412/*
Chris Masoncad321a2008-12-17 14:51:42 -05001413 * extent_io.c submission hook. This does the right thing for csum calculation
1414 * on write, or reading the csums from the tree before a read
Chris Masond352ac62008-09-29 15:18:18 -04001415 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001416static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
Chris Masonc8b97812008-10-29 14:49:59 -04001417 int mirror_num, unsigned long bio_flags)
Chris Mason44b8bd72008-04-16 11:14:51 -04001418{
1419 struct btrfs_root *root = BTRFS_I(inode)->root;
1420 int ret = 0;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001421 int skip_sum;
Chris Mason44b8bd72008-04-16 11:14:51 -04001422
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001423 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masoncad321a2008-12-17 14:51:42 -05001424
Chris Masone6dcd2d2008-07-17 12:53:50 -04001425 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1426 BUG_ON(ret);
Chris Mason065631f2008-02-20 12:07:25 -05001427
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001428 if (!(rw & (1 << BIO_RW))) {
Chris Masond20f7042008-12-08 16:58:54 -05001429 if (bio_flags & EXTENT_BIO_COMPRESSED) {
Chris Masonc8b97812008-10-29 14:49:59 -04001430 return btrfs_submit_compressed_read(inode, bio,
1431 mirror_num, bio_flags);
Chris Masond20f7042008-12-08 16:58:54 -05001432 } else if (!skip_sum)
1433 btrfs_lookup_bio_sums(root, inode, bio, NULL);
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001434 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001435 } else if (!skip_sum) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001436 /* csum items have already been cloned */
1437 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1438 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001439 /* we're doing a write, do the async checksumming */
1440 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
Chris Mason44b8bd72008-04-16 11:14:51 -04001441 inode, rw, bio, mirror_num,
Chris Mason4a69a412008-11-06 22:03:00 -05001442 bio_flags, __btrfs_submit_bio_start,
1443 __btrfs_submit_bio_done);
Chris Mason19b9bdb2008-10-30 14:23:13 -04001444 }
1445
Chris Mason0b86a832008-03-24 15:01:56 -04001446mapit:
Chris Mason8b712842008-06-11 16:50:36 -04001447 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
Chris Mason065631f2008-02-20 12:07:25 -05001448}
Chris Mason6885f302008-02-20 16:11:05 -05001449
Chris Masond352ac62008-09-29 15:18:18 -04001450/*
1451 * given a list of ordered sums record them in the inode. This happens
1452 * at IO completion time based on sums calculated at bio submission time.
1453 */
Chris Masonba1da2f2008-07-17 12:54:15 -04001454static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
Chris Masone6dcd2d2008-07-17 12:53:50 -04001455 struct inode *inode, u64 file_offset,
1456 struct list_head *list)
1457{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001458 struct btrfs_ordered_sum *sum;
1459
1460 btrfs_set_trans_block_group(trans, inode);
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001461
1462 list_for_each_entry(sum, list, list) {
Chris Masond20f7042008-12-08 16:58:54 -05001463 btrfs_csum_file_blocks(trans,
1464 BTRFS_I(inode)->root->fs_info->csum_root, sum);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001465 }
1466 return 0;
1467}
1468
Josef Bacik2ac55d42010-02-03 19:33:23 +00001469int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1470 struct extent_state **cached_state)
Chris Masonea8c2812008-08-04 23:17:27 -04001471{
Chris Masond3977122009-01-05 21:25:51 -05001472 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
Chris Mason771ed682008-11-06 22:02:51 -05001473 WARN_ON(1);
Chris Masonea8c2812008-08-04 23:17:27 -04001474 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00001475 cached_state, GFP_NOFS);
Chris Masonea8c2812008-08-04 23:17:27 -04001476}
1477
Chris Masond352ac62008-09-29 15:18:18 -04001478/* see btrfs_writepage_start_hook for details on why this is required */
Chris Mason247e7432008-07-17 12:53:51 -04001479struct btrfs_writepage_fixup {
1480 struct page *page;
1481 struct btrfs_work work;
1482};
1483
Christoph Hellwigb2950862008-12-02 09:54:17 -05001484static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
Chris Mason247e7432008-07-17 12:53:51 -04001485{
1486 struct btrfs_writepage_fixup *fixup;
1487 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001488 struct extent_state *cached_state = NULL;
Chris Mason247e7432008-07-17 12:53:51 -04001489 struct page *page;
1490 struct inode *inode;
1491 u64 page_start;
1492 u64 page_end;
1493
1494 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1495 page = fixup->page;
Chris Mason4a096752008-07-21 10:29:44 -04001496again:
Chris Mason247e7432008-07-17 12:53:51 -04001497 lock_page(page);
1498 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1499 ClearPageChecked(page);
1500 goto out_page;
1501 }
1502
1503 inode = page->mapping->host;
1504 page_start = page_offset(page);
1505 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1506
Josef Bacik2ac55d42010-02-03 19:33:23 +00001507 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1508 &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001509
1510 /* already ordered? We're done */
Chris Mason8b62b722009-09-02 16:53:46 -04001511 if (PagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001512 goto out;
Chris Mason4a096752008-07-21 10:29:44 -04001513
1514 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1515 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00001516 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1517 page_end, &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001518 unlock_page(page);
1519 btrfs_start_ordered_extent(inode, ordered, 1);
1520 goto again;
1521 }
Chris Mason247e7432008-07-17 12:53:51 -04001522
Josef Bacik2ac55d42010-02-03 19:33:23 +00001523 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
Chris Mason247e7432008-07-17 12:53:51 -04001524 ClearPageChecked(page);
1525out:
Josef Bacik2ac55d42010-02-03 19:33:23 +00001526 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1527 &cached_state, GFP_NOFS);
Chris Mason247e7432008-07-17 12:53:51 -04001528out_page:
1529 unlock_page(page);
1530 page_cache_release(page);
1531}
1532
1533/*
1534 * There are a few paths in the higher layers of the kernel that directly
1535 * set the page dirty bit without asking the filesystem if it is a
1536 * good idea. This causes problems because we want to make sure COW
1537 * properly happens and the data=ordered rules are followed.
1538 *
Chris Masonc8b97812008-10-29 14:49:59 -04001539 * In our case any range that doesn't have the ORDERED bit set
Chris Mason247e7432008-07-17 12:53:51 -04001540 * hasn't been properly setup for IO. We kick off an async process
1541 * to fix it up. The async helper will wait for ordered extents, set
1542 * the delalloc bit and make it safe to write the page.
1543 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001544static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
Chris Mason247e7432008-07-17 12:53:51 -04001545{
1546 struct inode *inode = page->mapping->host;
1547 struct btrfs_writepage_fixup *fixup;
1548 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason247e7432008-07-17 12:53:51 -04001549
Chris Mason8b62b722009-09-02 16:53:46 -04001550 /* this page is properly in the ordered list */
1551 if (TestClearPagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001552 return 0;
1553
1554 if (PageChecked(page))
1555 return -EAGAIN;
1556
1557 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1558 if (!fixup)
1559 return -EAGAIN;
Chris Masonf4219502008-07-22 11:18:09 -04001560
Chris Mason247e7432008-07-17 12:53:51 -04001561 SetPageChecked(page);
1562 page_cache_get(page);
1563 fixup->work.func = btrfs_writepage_fixup_worker;
1564 fixup->page = page;
1565 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1566 return -EAGAIN;
1567}
1568
Yan Zhengd899e052008-10-30 14:25:28 -04001569static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1570 struct inode *inode, u64 file_pos,
1571 u64 disk_bytenr, u64 disk_num_bytes,
1572 u64 num_bytes, u64 ram_bytes,
1573 u8 compression, u8 encryption,
1574 u16 other_encoding, int extent_type)
1575{
1576 struct btrfs_root *root = BTRFS_I(inode)->root;
1577 struct btrfs_file_extent_item *fi;
1578 struct btrfs_path *path;
1579 struct extent_buffer *leaf;
1580 struct btrfs_key ins;
1581 u64 hint;
1582 int ret;
1583
1584 path = btrfs_alloc_path();
1585 BUG_ON(!path);
1586
Chris Masonb9473432009-03-13 11:00:37 -04001587 path->leave_spinning = 1;
Chris Masona1ed8352009-09-11 12:27:37 -04001588
1589 /*
1590 * we may be replacing one extent in the tree with another.
1591 * The new extent is pinned in the extent map, and we don't want
1592 * to drop it from the cache until it is completely in the btree.
1593 *
1594 * So, tell btrfs_drop_extents to leave this extent in the cache.
1595 * the caller is expected to unpin it and allow it to be merged
1596 * with the others.
1597 */
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001598 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1599 &hint, 0);
Yan Zhengd899e052008-10-30 14:25:28 -04001600 BUG_ON(ret);
1601
1602 ins.objectid = inode->i_ino;
1603 ins.offset = file_pos;
1604 ins.type = BTRFS_EXTENT_DATA_KEY;
1605 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1606 BUG_ON(ret);
1607 leaf = path->nodes[0];
1608 fi = btrfs_item_ptr(leaf, path->slots[0],
1609 struct btrfs_file_extent_item);
1610 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1611 btrfs_set_file_extent_type(leaf, fi, extent_type);
1612 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1613 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1614 btrfs_set_file_extent_offset(leaf, fi, 0);
1615 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1616 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1617 btrfs_set_file_extent_compression(leaf, fi, compression);
1618 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1619 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
Chris Masonb9473432009-03-13 11:00:37 -04001620
1621 btrfs_unlock_up_safe(path, 1);
1622 btrfs_set_lock_blocking(leaf);
1623
Yan Zhengd899e052008-10-30 14:25:28 -04001624 btrfs_mark_buffer_dirty(leaf);
1625
1626 inode_add_bytes(inode, num_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -04001627
1628 ins.objectid = disk_bytenr;
1629 ins.offset = disk_num_bytes;
1630 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001631 ret = btrfs_alloc_reserved_file_extent(trans, root,
1632 root->root_key.objectid,
1633 inode->i_ino, file_pos, &ins);
Yan Zhengd899e052008-10-30 14:25:28 -04001634 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -04001635 btrfs_free_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04001636
Yan Zhengd899e052008-10-30 14:25:28 -04001637 return 0;
1638}
1639
Chris Mason5d13a982009-03-13 11:41:46 -04001640/*
1641 * helper function for btrfs_finish_ordered_io, this
1642 * just reads in some of the csum leaves to prime them into ram
1643 * before we start the transaction. It limits the amount of btree
1644 * reads required while inside the transaction.
1645 */
Chris Masond352ac62008-09-29 15:18:18 -04001646/* as ordered data IO finishes, this gets called so we can finish
1647 * an ordered extent if the range of bytes in the file it covers are
1648 * fully written.
1649 */
Chris Mason211f90e2008-07-18 11:56:15 -04001650static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001651{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001652 struct btrfs_root *root = BTRFS_I(inode)->root;
1653 struct btrfs_trans_handle *trans;
Chris Mason5d13a982009-03-13 11:41:46 -04001654 struct btrfs_ordered_extent *ordered_extent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001655 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001656 struct extent_state *cached_state = NULL;
Yan Zhengd899e052008-10-30 14:25:28 -04001657 int compressed = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001658 int ret;
1659
Josef Bacik5a1a3df2010-02-02 20:51:14 +00001660 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1661 end - start + 1);
Chris Masonba1da2f2008-07-17 12:54:15 -04001662 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001663 return 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001664 BUG_ON(!ordered_extent);
Josef Bacikefd049f2010-02-02 20:50:10 +00001665
Yan, Zhengc2167752009-11-12 09:34:21 +00001666 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1667 BUG_ON(!list_empty(&ordered_extent->list));
1668 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1669 if (!ret) {
1670 trans = btrfs_join_transaction(root, 1);
1671 ret = btrfs_update_inode(trans, root, inode);
1672 BUG_ON(ret);
1673 btrfs_end_transaction(trans, root);
1674 }
1675 goto out;
1676 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04001677
Josef Bacik2ac55d42010-02-03 19:33:23 +00001678 lock_extent_bits(io_tree, ordered_extent->file_offset,
1679 ordered_extent->file_offset + ordered_extent->len - 1,
1680 0, &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001681
Yan, Zhengc2167752009-11-12 09:34:21 +00001682 trans = btrfs_join_transaction(root, 1);
1683
Chris Masonc8b97812008-10-29 14:49:59 -04001684 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
Yan Zhengd899e052008-10-30 14:25:28 -04001685 compressed = 1;
1686 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1687 BUG_ON(compressed);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001688 ret = btrfs_mark_extent_written(trans, inode,
Yan Zhengd899e052008-10-30 14:25:28 -04001689 ordered_extent->file_offset,
1690 ordered_extent->file_offset +
1691 ordered_extent->len);
1692 BUG_ON(ret);
1693 } else {
1694 ret = insert_reserved_file_extent(trans, inode,
1695 ordered_extent->file_offset,
1696 ordered_extent->start,
1697 ordered_extent->disk_len,
1698 ordered_extent->len,
1699 ordered_extent->len,
1700 compressed, 0, 0,
1701 BTRFS_FILE_EXTENT_REG);
Chris Masona1ed8352009-09-11 12:27:37 -04001702 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1703 ordered_extent->file_offset,
1704 ordered_extent->len);
Yan Zhengd899e052008-10-30 14:25:28 -04001705 BUG_ON(ret);
1706 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00001707 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1708 ordered_extent->file_offset +
1709 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1710
Chris Masone6dcd2d2008-07-17 12:53:50 -04001711 add_pending_csums(trans, inode, ordered_extent->file_offset,
1712 &ordered_extent->list);
1713
Yan, Zhengc2167752009-11-12 09:34:21 +00001714 /* this also removes the ordered extent from the tree */
1715 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1716 ret = btrfs_update_inode(trans, root, inode);
1717 BUG_ON(ret);
1718 btrfs_end_transaction(trans, root);
1719out:
Chris Masone6dcd2d2008-07-17 12:53:50 -04001720 /* once for us */
1721 btrfs_put_ordered_extent(ordered_extent);
1722 /* once for the tree */
1723 btrfs_put_ordered_extent(ordered_extent);
1724
Chris Masone6dcd2d2008-07-17 12:53:50 -04001725 return 0;
1726}
1727
Christoph Hellwigb2950862008-12-02 09:54:17 -05001728static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason211f90e2008-07-18 11:56:15 -04001729 struct extent_state *state, int uptodate)
1730{
Chris Mason8b62b722009-09-02 16:53:46 -04001731 ClearPagePrivate2(page);
Chris Mason211f90e2008-07-18 11:56:15 -04001732 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1733}
1734
Chris Masond352ac62008-09-29 15:18:18 -04001735/*
1736 * When IO fails, either with EIO or csum verification fails, we
1737 * try other mirrors that might have a good copy of the data. This
1738 * io_failure_record is used to record state as we go through all the
1739 * mirrors. If another mirror has good data, the page is set up to date
1740 * and things continue. If a good mirror can't be found, the original
1741 * bio end_io callback is called to indicate things have failed.
1742 */
Chris Mason7e383262008-04-09 16:28:12 -04001743struct io_failure_record {
1744 struct page *page;
1745 u64 start;
1746 u64 len;
1747 u64 logical;
Chris Masond20f7042008-12-08 16:58:54 -05001748 unsigned long bio_flags;
Chris Mason7e383262008-04-09 16:28:12 -04001749 int last_mirror;
1750};
1751
Christoph Hellwigb2950862008-12-02 09:54:17 -05001752static int btrfs_io_failed_hook(struct bio *failed_bio,
Chris Mason1259ab72008-05-12 13:39:03 -04001753 struct page *page, u64 start, u64 end,
1754 struct extent_state *state)
Chris Mason7e383262008-04-09 16:28:12 -04001755{
1756 struct io_failure_record *failrec = NULL;
1757 u64 private;
1758 struct extent_map *em;
1759 struct inode *inode = page->mapping->host;
1760 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
Chris Mason3b951512008-04-17 11:29:12 -04001761 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Mason7e383262008-04-09 16:28:12 -04001762 struct bio *bio;
1763 int num_copies;
1764 int ret;
Chris Mason1259ab72008-05-12 13:39:03 -04001765 int rw;
Chris Mason7e383262008-04-09 16:28:12 -04001766 u64 logical;
1767
1768 ret = get_state_private(failure_tree, start, &private);
1769 if (ret) {
Chris Mason7e383262008-04-09 16:28:12 -04001770 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1771 if (!failrec)
1772 return -ENOMEM;
1773 failrec->start = start;
1774 failrec->len = end - start + 1;
1775 failrec->last_mirror = 0;
Chris Masond20f7042008-12-08 16:58:54 -05001776 failrec->bio_flags = 0;
Chris Mason7e383262008-04-09 16:28:12 -04001777
Chris Mason890871b2009-09-02 16:24:52 -04001778 read_lock(&em_tree->lock);
Chris Mason3b951512008-04-17 11:29:12 -04001779 em = lookup_extent_mapping(em_tree, start, failrec->len);
1780 if (em->start > start || em->start + em->len < start) {
1781 free_extent_map(em);
1782 em = NULL;
1783 }
Chris Mason890871b2009-09-02 16:24:52 -04001784 read_unlock(&em_tree->lock);
Chris Mason7e383262008-04-09 16:28:12 -04001785
1786 if (!em || IS_ERR(em)) {
1787 kfree(failrec);
1788 return -EIO;
1789 }
1790 logical = start - em->start;
1791 logical = em->block_start + logical;
Chris Masond20f7042008-12-08 16:58:54 -05001792 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1793 logical = em->block_start;
1794 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1795 }
Chris Mason7e383262008-04-09 16:28:12 -04001796 failrec->logical = logical;
1797 free_extent_map(em);
1798 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1799 EXTENT_DIRTY, GFP_NOFS);
Chris Mason587f7702008-04-11 12:16:46 -04001800 set_state_private(failure_tree, start,
1801 (u64)(unsigned long)failrec);
Chris Mason7e383262008-04-09 16:28:12 -04001802 } else {
Chris Mason587f7702008-04-11 12:16:46 -04001803 failrec = (struct io_failure_record *)(unsigned long)private;
Chris Mason7e383262008-04-09 16:28:12 -04001804 }
1805 num_copies = btrfs_num_copies(
1806 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1807 failrec->logical, failrec->len);
1808 failrec->last_mirror++;
1809 if (!state) {
Chris Masoncad321a2008-12-17 14:51:42 -05001810 spin_lock(&BTRFS_I(inode)->io_tree.lock);
Chris Mason7e383262008-04-09 16:28:12 -04001811 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1812 failrec->start,
1813 EXTENT_LOCKED);
1814 if (state && state->start != failrec->start)
1815 state = NULL;
Chris Masoncad321a2008-12-17 14:51:42 -05001816 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
Chris Mason7e383262008-04-09 16:28:12 -04001817 }
1818 if (!state || failrec->last_mirror > num_copies) {
1819 set_state_private(failure_tree, failrec->start, 0);
1820 clear_extent_bits(failure_tree, failrec->start,
1821 failrec->start + failrec->len - 1,
1822 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1823 kfree(failrec);
1824 return -EIO;
1825 }
1826 bio = bio_alloc(GFP_NOFS, 1);
1827 bio->bi_private = state;
1828 bio->bi_end_io = failed_bio->bi_end_io;
1829 bio->bi_sector = failrec->logical >> 9;
1830 bio->bi_bdev = failed_bio->bi_bdev;
Chris Masone1c4b742008-04-22 13:26:46 -04001831 bio->bi_size = 0;
Chris Masond20f7042008-12-08 16:58:54 -05001832
Chris Mason7e383262008-04-09 16:28:12 -04001833 bio_add_page(bio, page, failrec->len, start - page_offset(page));
Chris Mason1259ab72008-05-12 13:39:03 -04001834 if (failed_bio->bi_rw & (1 << BIO_RW))
1835 rw = WRITE;
1836 else
1837 rw = READ;
1838
1839 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
Chris Masonc8b97812008-10-29 14:49:59 -04001840 failrec->last_mirror,
Chris Masond20f7042008-12-08 16:58:54 -05001841 failrec->bio_flags);
Chris Mason1259ab72008-05-12 13:39:03 -04001842 return 0;
1843}
1844
Chris Masond352ac62008-09-29 15:18:18 -04001845/*
1846 * each time an IO finishes, we do a fast check in the IO failure tree
1847 * to see if we need to process or clean up an io_failure_record
1848 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001849static int btrfs_clean_io_failures(struct inode *inode, u64 start)
Chris Mason1259ab72008-05-12 13:39:03 -04001850{
1851 u64 private;
1852 u64 private_failure;
1853 struct io_failure_record *failure;
1854 int ret;
1855
1856 private = 0;
1857 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1858 (u64)-1, 1, EXTENT_DIRTY)) {
1859 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1860 start, &private_failure);
1861 if (ret == 0) {
1862 failure = (struct io_failure_record *)(unsigned long)
1863 private_failure;
1864 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1865 failure->start, 0);
1866 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1867 failure->start,
1868 failure->start + failure->len - 1,
1869 EXTENT_DIRTY | EXTENT_LOCKED,
1870 GFP_NOFS);
1871 kfree(failure);
1872 }
1873 }
Chris Mason7e383262008-04-09 16:28:12 -04001874 return 0;
1875}
1876
Chris Masond352ac62008-09-29 15:18:18 -04001877/*
1878 * when reads are done, we need to check csums to verify the data is correct
1879 * if there's a match, we allow the bio to finish. If not, we go through
1880 * the io_failure_record routines to find good copies
1881 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001882static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason70dec802008-01-29 09:59:12 -05001883 struct extent_state *state)
Chris Mason07157aa2007-08-30 08:50:51 -04001884{
Chris Mason35ebb932007-10-30 16:56:53 -04001885 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
Chris Mason07157aa2007-08-30 08:50:51 -04001886 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05001887 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason07157aa2007-08-30 08:50:51 -04001888 char *kaddr;
Chris Masonaadfeb62008-01-29 09:10:27 -05001889 u64 private = ~(u32)0;
Chris Mason07157aa2007-08-30 08:50:51 -04001890 int ret;
Chris Masonff79f812007-10-15 16:22:25 -04001891 struct btrfs_root *root = BTRFS_I(inode)->root;
1892 u32 csum = ~(u32)0;
Chris Masond1310b22008-01-24 16:13:08 -05001893
Chris Masond20f7042008-12-08 16:58:54 -05001894 if (PageChecked(page)) {
1895 ClearPageChecked(page);
1896 goto good;
1897 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001898
1899 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Chris Masonb6cda9b2007-12-14 15:30:32 -05001900 return 0;
Chris Masond20f7042008-12-08 16:58:54 -05001901
Yan Zheng17d217f2008-12-12 10:03:38 -05001902 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
Chris Mason9655d292009-09-02 15:22:30 -04001903 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001904 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1905 GFP_NOFS);
1906 return 0;
1907 }
1908
Yanc2e639f2008-02-04 08:57:25 -05001909 if (state && state->start == start) {
Chris Mason70dec802008-01-29 09:59:12 -05001910 private = state->private;
1911 ret = 0;
1912 } else {
1913 ret = get_state_private(io_tree, start, &private);
1914 }
Chris Mason9ab86c82009-01-07 09:48:51 -05001915 kaddr = kmap_atomic(page, KM_USER0);
Chris Masond3977122009-01-05 21:25:51 -05001916 if (ret)
Chris Mason07157aa2007-08-30 08:50:51 -04001917 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05001918
Chris Masonff79f812007-10-15 16:22:25 -04001919 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1920 btrfs_csum_final(csum, (char *)&csum);
Chris Masond3977122009-01-05 21:25:51 -05001921 if (csum != private)
Chris Mason07157aa2007-08-30 08:50:51 -04001922 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05001923
Chris Mason9ab86c82009-01-07 09:48:51 -05001924 kunmap_atomic(kaddr, KM_USER0);
Chris Masond20f7042008-12-08 16:58:54 -05001925good:
Chris Mason7e383262008-04-09 16:28:12 -04001926 /* if the io failure tree for this inode is non-empty,
1927 * check to see if we've recovered from a failed IO
1928 */
Chris Mason1259ab72008-05-12 13:39:03 -04001929 btrfs_clean_io_failures(inode, start);
Chris Mason07157aa2007-08-30 08:50:51 -04001930 return 0;
1931
1932zeroit:
Chris Mason193f2842009-04-27 07:29:05 -04001933 if (printk_ratelimit()) {
1934 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1935 "private %llu\n", page->mapping->host->i_ino,
1936 (unsigned long long)start, csum,
1937 (unsigned long long)private);
1938 }
Chris Masondb945352007-10-15 16:15:53 -04001939 memset(kaddr + offset, 1, end - start + 1);
1940 flush_dcache_page(page);
Chris Mason9ab86c82009-01-07 09:48:51 -05001941 kunmap_atomic(kaddr, KM_USER0);
Chris Mason3b951512008-04-17 11:29:12 -04001942 if (private == 0)
1943 return 0;
Chris Mason7e383262008-04-09 16:28:12 -04001944 return -EIO;
Chris Mason07157aa2007-08-30 08:50:51 -04001945}
Chris Masonb888db22007-08-27 16:49:44 -04001946
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001947struct delayed_iput {
1948 struct list_head list;
1949 struct inode *inode;
1950};
1951
1952void btrfs_add_delayed_iput(struct inode *inode)
1953{
1954 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1955 struct delayed_iput *delayed;
1956
1957 if (atomic_add_unless(&inode->i_count, -1, 1))
1958 return;
1959
1960 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1961 delayed->inode = inode;
1962
1963 spin_lock(&fs_info->delayed_iput_lock);
1964 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1965 spin_unlock(&fs_info->delayed_iput_lock);
1966}
1967
1968void btrfs_run_delayed_iputs(struct btrfs_root *root)
1969{
1970 LIST_HEAD(list);
1971 struct btrfs_fs_info *fs_info = root->fs_info;
1972 struct delayed_iput *delayed;
1973 int empty;
1974
1975 spin_lock(&fs_info->delayed_iput_lock);
1976 empty = list_empty(&fs_info->delayed_iputs);
1977 spin_unlock(&fs_info->delayed_iput_lock);
1978 if (empty)
1979 return;
1980
1981 down_read(&root->fs_info->cleanup_work_sem);
1982 spin_lock(&fs_info->delayed_iput_lock);
1983 list_splice_init(&fs_info->delayed_iputs, &list);
1984 spin_unlock(&fs_info->delayed_iput_lock);
1985
1986 while (!list_empty(&list)) {
1987 delayed = list_entry(list.next, struct delayed_iput, list);
1988 list_del(&delayed->list);
1989 iput(delayed->inode);
1990 kfree(delayed);
1991 }
1992 up_read(&root->fs_info->cleanup_work_sem);
1993}
1994
Josef Bacik7b128762008-07-24 12:17:14 -04001995/*
1996 * This creates an orphan entry for the given inode in case something goes
1997 * wrong in the middle of an unlink/truncate.
1998 */
1999int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2000{
2001 struct btrfs_root *root = BTRFS_I(inode)->root;
2002 int ret = 0;
2003
Yanbcc63ab2008-07-30 16:29:20 -04002004 spin_lock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002005
2006 /* already on the orphan list, we're good */
2007 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
Yanbcc63ab2008-07-30 16:29:20 -04002008 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002009 return 0;
2010 }
2011
2012 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2013
Yanbcc63ab2008-07-30 16:29:20 -04002014 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002015
2016 /*
2017 * insert an orphan item to track this unlinked/truncated file
2018 */
2019 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2020
2021 return ret;
2022}
2023
2024/*
2025 * We have done the truncate/delete so we can go ahead and remove the orphan
2026 * item for this particular inode.
2027 */
2028int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2029{
2030 struct btrfs_root *root = BTRFS_I(inode)->root;
2031 int ret = 0;
2032
Yanbcc63ab2008-07-30 16:29:20 -04002033 spin_lock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002034
2035 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
Yanbcc63ab2008-07-30 16:29:20 -04002036 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002037 return 0;
2038 }
2039
2040 list_del_init(&BTRFS_I(inode)->i_orphan);
2041 if (!trans) {
Yanbcc63ab2008-07-30 16:29:20 -04002042 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002043 return 0;
2044 }
2045
Yanbcc63ab2008-07-30 16:29:20 -04002046 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002047
2048 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2049
2050 return ret;
2051}
2052
2053/*
2054 * this cleans up any orphans that may be left on the list from the last use
2055 * of this root.
2056 */
2057void btrfs_orphan_cleanup(struct btrfs_root *root)
2058{
2059 struct btrfs_path *path;
2060 struct extent_buffer *leaf;
2061 struct btrfs_item *item;
2062 struct btrfs_key key, found_key;
2063 struct btrfs_trans_handle *trans;
2064 struct inode *inode;
2065 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2066
Yan, Zhengc71bf092009-11-12 09:34:40 +00002067 if (!xchg(&root->clean_orphans, 0))
Josef Bacik7b128762008-07-24 12:17:14 -04002068 return;
Yan, Zhengc71bf092009-11-12 09:34:40 +00002069
2070 path = btrfs_alloc_path();
2071 BUG_ON(!path);
Josef Bacik7b128762008-07-24 12:17:14 -04002072 path->reada = -1;
2073
2074 key.objectid = BTRFS_ORPHAN_OBJECTID;
2075 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2076 key.offset = (u64)-1;
2077
Josef Bacik7b128762008-07-24 12:17:14 -04002078 while (1) {
2079 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2080 if (ret < 0) {
2081 printk(KERN_ERR "Error searching slot for orphan: %d"
2082 "\n", ret);
2083 break;
2084 }
2085
2086 /*
2087 * if ret == 0 means we found what we were searching for, which
2088 * is weird, but possible, so only screw with path if we didnt
2089 * find the key and see if we have stuff that matches
2090 */
2091 if (ret > 0) {
2092 if (path->slots[0] == 0)
2093 break;
2094 path->slots[0]--;
2095 }
2096
2097 /* pull out the item */
2098 leaf = path->nodes[0];
2099 item = btrfs_item_nr(leaf, path->slots[0]);
2100 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2101
2102 /* make sure the item matches what we want */
2103 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2104 break;
2105 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2106 break;
2107
2108 /* release the path since we're done with it */
2109 btrfs_release_path(root, path);
2110
2111 /*
2112 * this is where we are basically btrfs_lookup, without the
2113 * crossing root thing. we store the inode number in the
2114 * offset of the orphan item.
2115 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002116 found_key.objectid = found_key.offset;
2117 found_key.type = BTRFS_INODE_ITEM_KEY;
2118 found_key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +00002119 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002120 if (IS_ERR(inode))
Josef Bacik7b128762008-07-24 12:17:14 -04002121 break;
2122
Josef Bacik7b128762008-07-24 12:17:14 -04002123 /*
2124 * add this inode to the orphan list so btrfs_orphan_del does
2125 * the proper thing when we hit it
2126 */
Yanbcc63ab2008-07-30 16:29:20 -04002127 spin_lock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002128 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
Yanbcc63ab2008-07-30 16:29:20 -04002129 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002130
2131 /*
2132 * if this is a bad inode, means we actually succeeded in
2133 * removing the inode, but not the orphan record, which means
2134 * we need to manually delete the orphan since iput will just
2135 * do a destroy_inode
2136 */
2137 if (is_bad_inode(inode)) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -04002138 trans = btrfs_start_transaction(root, 1);
Josef Bacik7b128762008-07-24 12:17:14 -04002139 btrfs_orphan_del(trans, inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04002140 btrfs_end_transaction(trans, root);
Josef Bacik7b128762008-07-24 12:17:14 -04002141 iput(inode);
2142 continue;
2143 }
2144
2145 /* if we have links, this was a truncate, lets do that */
2146 if (inode->i_nlink) {
2147 nr_truncate++;
2148 btrfs_truncate(inode);
2149 } else {
2150 nr_unlink++;
2151 }
2152
2153 /* this will do delete_inode and everything for us */
2154 iput(inode);
2155 }
2156
2157 if (nr_unlink)
2158 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2159 if (nr_truncate)
2160 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2161
2162 btrfs_free_path(path);
Josef Bacik7b128762008-07-24 12:17:14 -04002163}
2164
Chris Masond352ac62008-09-29 15:18:18 -04002165/*
Chris Mason46a53cc2009-04-27 11:47:50 -04002166 * very simple check to peek ahead in the leaf looking for xattrs. If we
2167 * don't find any xattrs, we know there can't be any acls.
2168 *
2169 * slot is the slot the inode is in, objectid is the objectid of the inode
2170 */
2171static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2172 int slot, u64 objectid)
2173{
2174 u32 nritems = btrfs_header_nritems(leaf);
2175 struct btrfs_key found_key;
2176 int scanned = 0;
2177
2178 slot++;
2179 while (slot < nritems) {
2180 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2181
2182 /* we found a different objectid, there must not be acls */
2183 if (found_key.objectid != objectid)
2184 return 0;
2185
2186 /* we found an xattr, assume we've got an acl */
2187 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2188 return 1;
2189
2190 /*
2191 * we found a key greater than an xattr key, there can't
2192 * be any acls later on
2193 */
2194 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2195 return 0;
2196
2197 slot++;
2198 scanned++;
2199
2200 /*
2201 * it goes inode, inode backrefs, xattrs, extents,
2202 * so if there are a ton of hard links to an inode there can
2203 * be a lot of backrefs. Don't waste time searching too hard,
2204 * this is just an optimization
2205 */
2206 if (scanned >= 8)
2207 break;
2208 }
2209 /* we hit the end of the leaf before we found an xattr or
2210 * something larger than an xattr. We have to assume the inode
2211 * has acls
2212 */
2213 return 1;
2214}
2215
2216/*
Chris Masond352ac62008-09-29 15:18:18 -04002217 * read an inode from the btree into the in-memory inode
2218 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002219static void btrfs_read_locked_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002220{
2221 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002222 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002223 struct btrfs_inode_item *inode_item;
Chris Mason0b86a832008-03-24 15:01:56 -04002224 struct btrfs_timespec *tspec;
Chris Mason39279cc2007-06-12 06:35:45 -04002225 struct btrfs_root *root = BTRFS_I(inode)->root;
2226 struct btrfs_key location;
Chris Mason46a53cc2009-04-27 11:47:50 -04002227 int maybe_acls;
Chris Mason39279cc2007-06-12 06:35:45 -04002228 u64 alloc_group_block;
Josef Bacik618e21d2007-07-11 10:18:17 -04002229 u32 rdev;
Chris Mason39279cc2007-06-12 06:35:45 -04002230 int ret;
2231
2232 path = btrfs_alloc_path();
2233 BUG_ON(!path);
Chris Mason39279cc2007-06-12 06:35:45 -04002234 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
Chris Masondc17ff82008-01-08 15:46:30 -05002235
Chris Mason39279cc2007-06-12 06:35:45 -04002236 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04002237 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04002238 goto make_bad;
Chris Mason39279cc2007-06-12 06:35:45 -04002239
Chris Mason5f39d392007-10-15 16:14:19 -04002240 leaf = path->nodes[0];
2241 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2242 struct btrfs_inode_item);
2243
2244 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2245 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2246 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2247 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
Chris Masondbe674a2008-07-17 12:54:05 -04002248 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04002249
2250 tspec = btrfs_inode_atime(inode_item);
2251 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2252 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2253
2254 tspec = btrfs_inode_mtime(inode_item);
2255 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2256 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2257
2258 tspec = btrfs_inode_ctime(inode_item);
2259 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2260 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2261
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002262 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04002263 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
Chris Masonc3027eb2008-12-08 16:40:21 -05002264 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04002265 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik618e21d2007-07-11 10:18:17 -04002266 inode->i_rdev = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002267 rdev = btrfs_inode_rdev(leaf, inode_item);
2268
Josef Bacikaec74772008-07-24 12:12:38 -04002269 BTRFS_I(inode)->index_cnt = (u64)-1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05002270 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
Josef Bacikaec74772008-07-24 12:12:38 -04002271
Chris Mason5f39d392007-10-15 16:14:19 -04002272 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
Chris Masonb4ce94d2009-02-04 09:25:08 -05002273
Chris Mason46a53cc2009-04-27 11:47:50 -04002274 /*
2275 * try to precache a NULL acl entry for files that don't have
2276 * any xattrs or acls
2277 */
2278 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
Al Viro72c04902009-06-24 16:58:48 -04002279 if (!maybe_acls)
2280 cache_no_acl(inode);
Chris Mason46a53cc2009-04-27 11:47:50 -04002281
Yan Zhengd2fb3432008-12-11 16:30:39 -05002282 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2283 alloc_group_block, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04002284 btrfs_free_path(path);
2285 inode_item = NULL;
2286
Chris Mason39279cc2007-06-12 06:35:45 -04002287 switch (inode->i_mode & S_IFMT) {
Chris Mason39279cc2007-06-12 06:35:45 -04002288 case S_IFREG:
2289 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04002290 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05002291 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04002292 inode->i_fop = &btrfs_file_operations;
2293 inode->i_op = &btrfs_file_inode_operations;
2294 break;
2295 case S_IFDIR:
2296 inode->i_fop = &btrfs_dir_file_operations;
2297 if (root == root->fs_info->tree_root)
2298 inode->i_op = &btrfs_dir_ro_inode_operations;
2299 else
2300 inode->i_op = &btrfs_dir_inode_operations;
2301 break;
2302 case S_IFLNK:
2303 inode->i_op = &btrfs_symlink_inode_operations;
2304 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04002305 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04002306 break;
Josef Bacik618e21d2007-07-11 10:18:17 -04002307 default:
Jim Owens0279b4c2009-02-04 09:29:13 -05002308 inode->i_op = &btrfs_special_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -04002309 init_special_inode(inode, inode->i_mode, rdev);
2310 break;
Chris Mason39279cc2007-06-12 06:35:45 -04002311 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02002312
2313 btrfs_update_iflags(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002314 return;
2315
2316make_bad:
Chris Mason39279cc2007-06-12 06:35:45 -04002317 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002318 make_bad_inode(inode);
2319}
2320
Chris Masond352ac62008-09-29 15:18:18 -04002321/*
2322 * given a leaf and an inode, copy the inode fields into the leaf
2323 */
Chris Masone02119d2008-09-05 16:13:11 -04002324static void fill_inode_item(struct btrfs_trans_handle *trans,
2325 struct extent_buffer *leaf,
Chris Mason5f39d392007-10-15 16:14:19 -04002326 struct btrfs_inode_item *item,
Chris Mason39279cc2007-06-12 06:35:45 -04002327 struct inode *inode)
2328{
Chris Mason5f39d392007-10-15 16:14:19 -04002329 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2330 btrfs_set_inode_gid(leaf, item, inode->i_gid);
Chris Masondbe674a2008-07-17 12:54:05 -04002331 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
Chris Mason5f39d392007-10-15 16:14:19 -04002332 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2333 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2334
2335 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2336 inode->i_atime.tv_sec);
2337 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2338 inode->i_atime.tv_nsec);
2339
2340 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2341 inode->i_mtime.tv_sec);
2342 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2343 inode->i_mtime.tv_nsec);
2344
2345 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2346 inode->i_ctime.tv_sec);
2347 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2348 inode->i_ctime.tv_nsec);
2349
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002350 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
Chris Masone02119d2008-09-05 16:13:11 -04002351 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
Chris Masonc3027eb2008-12-08 16:40:21 -05002352 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
Chris Masone02119d2008-09-05 16:13:11 -04002353 btrfs_set_inode_transid(leaf, item, trans->transid);
Chris Mason5f39d392007-10-15 16:14:19 -04002354 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
Yanb98b6762008-01-08 15:54:37 -05002355 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
Yan Zhengd2fb3432008-12-11 16:30:39 -05002356 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
Chris Mason39279cc2007-06-12 06:35:45 -04002357}
2358
Chris Masond352ac62008-09-29 15:18:18 -04002359/*
2360 * copy everything in the in-memory inode into the btree.
2361 */
Chris Masond3977122009-01-05 21:25:51 -05002362noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2363 struct btrfs_root *root, struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002364{
2365 struct btrfs_inode_item *inode_item;
2366 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002367 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002368 int ret;
2369
2370 path = btrfs_alloc_path();
2371 BUG_ON(!path);
Chris Masonb9473432009-03-13 11:00:37 -04002372 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04002373 ret = btrfs_lookup_inode(trans, root, path,
2374 &BTRFS_I(inode)->location, 1);
2375 if (ret) {
2376 if (ret > 0)
2377 ret = -ENOENT;
2378 goto failed;
2379 }
2380
Chris Masonb4ce94d2009-02-04 09:25:08 -05002381 btrfs_unlock_up_safe(path, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04002382 leaf = path->nodes[0];
2383 inode_item = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04002384 struct btrfs_inode_item);
2385
Chris Masone02119d2008-09-05 16:13:11 -04002386 fill_inode_item(trans, leaf, inode_item, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04002387 btrfs_mark_buffer_dirty(leaf);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002388 btrfs_set_inode_last_trans(trans, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002389 ret = 0;
2390failed:
Chris Mason39279cc2007-06-12 06:35:45 -04002391 btrfs_free_path(path);
2392 return ret;
2393}
2394
2395
Chris Masond352ac62008-09-29 15:18:18 -04002396/*
2397 * unlink helper that gets used here in inode.c and in the tree logging
2398 * recovery code. It remove a link in a directory with a given name, and
2399 * also drops the back refs in the inode to the directory
2400 */
Chris Masone02119d2008-09-05 16:13:11 -04002401int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2402 struct btrfs_root *root,
2403 struct inode *dir, struct inode *inode,
2404 const char *name, int name_len)
Chris Mason39279cc2007-06-12 06:35:45 -04002405{
2406 struct btrfs_path *path;
Chris Mason39279cc2007-06-12 06:35:45 -04002407 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002408 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002409 struct btrfs_dir_item *di;
Chris Mason5f39d392007-10-15 16:14:19 -04002410 struct btrfs_key key;
Josef Bacikaec74772008-07-24 12:12:38 -04002411 u64 index;
Chris Mason39279cc2007-06-12 06:35:45 -04002412
2413 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04002414 if (!path) {
2415 ret = -ENOMEM;
2416 goto err;
2417 }
2418
Chris Masonb9473432009-03-13 11:00:37 -04002419 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04002420 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2421 name, name_len, -1);
2422 if (IS_ERR(di)) {
2423 ret = PTR_ERR(di);
2424 goto err;
2425 }
2426 if (!di) {
2427 ret = -ENOENT;
2428 goto err;
2429 }
Chris Mason5f39d392007-10-15 16:14:19 -04002430 leaf = path->nodes[0];
2431 btrfs_dir_item_key_to_cpu(leaf, di, &key);
Chris Mason39279cc2007-06-12 06:35:45 -04002432 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason54aa1f42007-06-22 14:16:25 -04002433 if (ret)
2434 goto err;
Chris Mason39279cc2007-06-12 06:35:45 -04002435 btrfs_release_path(root, path);
2436
Josef Bacikaec74772008-07-24 12:12:38 -04002437 ret = btrfs_del_inode_ref(trans, root, name, name_len,
Chris Masone02119d2008-09-05 16:13:11 -04002438 inode->i_ino,
2439 dir->i_ino, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04002440 if (ret) {
Chris Masond3977122009-01-05 21:25:51 -05002441 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
Josef Bacikaec74772008-07-24 12:12:38 -04002442 "inode %lu parent %lu\n", name_len, name,
Chris Masone02119d2008-09-05 16:13:11 -04002443 inode->i_ino, dir->i_ino);
Josef Bacikaec74772008-07-24 12:12:38 -04002444 goto err;
2445 }
2446
Chris Mason39279cc2007-06-12 06:35:45 -04002447 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
Josef Bacikaec74772008-07-24 12:12:38 -04002448 index, name, name_len, -1);
Chris Mason39279cc2007-06-12 06:35:45 -04002449 if (IS_ERR(di)) {
2450 ret = PTR_ERR(di);
2451 goto err;
2452 }
2453 if (!di) {
2454 ret = -ENOENT;
2455 goto err;
2456 }
2457 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason925baed2008-06-25 16:01:30 -04002458 btrfs_release_path(root, path);
Chris Mason39279cc2007-06-12 06:35:45 -04002459
Chris Masone02119d2008-09-05 16:13:11 -04002460 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2461 inode, dir->i_ino);
Chris Mason49eb7e42008-09-11 15:53:12 -04002462 BUG_ON(ret != 0 && ret != -ENOENT);
Chris Masone02119d2008-09-05 16:13:11 -04002463
2464 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2465 dir, index);
2466 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04002467err:
2468 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002469 if (ret)
2470 goto out;
2471
2472 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2473 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2474 btrfs_update_inode(trans, root, dir);
2475 btrfs_drop_nlink(inode);
2476 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04002477out:
Chris Mason39279cc2007-06-12 06:35:45 -04002478 return ret;
2479}
2480
2481static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2482{
2483 struct btrfs_root *root;
2484 struct btrfs_trans_handle *trans;
Josef Bacik7b128762008-07-24 12:17:14 -04002485 struct inode *inode = dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04002486 int ret;
Chris Mason1832a6d2007-12-21 16:27:21 -05002487 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002488
2489 root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05002490
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002491 /*
2492 * 5 items for unlink inode
2493 * 1 for orphan
2494 */
2495 ret = btrfs_reserve_metadata_space(root, 6);
2496 if (ret)
2497 return ret;
2498
Chris Mason39279cc2007-06-12 06:35:45 -04002499 trans = btrfs_start_transaction(root, 1);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002500 if (IS_ERR(trans)) {
2501 btrfs_unreserve_metadata_space(root, 6);
2502 return PTR_ERR(trans);
2503 }
Chris Mason5f39d392007-10-15 16:14:19 -04002504
Chris Mason39279cc2007-06-12 06:35:45 -04002505 btrfs_set_trans_block_group(trans, dir);
Chris Mason12fcfd22009-03-24 10:24:20 -04002506
2507 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2508
Chris Masone02119d2008-09-05 16:13:11 -04002509 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2510 dentry->d_name.name, dentry->d_name.len);
Josef Bacik7b128762008-07-24 12:17:14 -04002511
2512 if (inode->i_nlink == 0)
2513 ret = btrfs_orphan_add(trans, inode);
2514
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002515 nr = trans->blocks_used;
Chris Mason5f39d392007-10-15 16:14:19 -04002516
Chris Mason89ce8a62008-06-25 16:01:31 -04002517 btrfs_end_transaction_throttle(trans, root);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002518 btrfs_unreserve_metadata_space(root, 6);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002519 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04002520 return ret;
2521}
2522
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002523int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2524 struct btrfs_root *root,
2525 struct inode *dir, u64 objectid,
2526 const char *name, int name_len)
2527{
2528 struct btrfs_path *path;
2529 struct extent_buffer *leaf;
2530 struct btrfs_dir_item *di;
2531 struct btrfs_key key;
2532 u64 index;
2533 int ret;
2534
2535 path = btrfs_alloc_path();
2536 if (!path)
2537 return -ENOMEM;
2538
2539 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2540 name, name_len, -1);
2541 BUG_ON(!di || IS_ERR(di));
2542
2543 leaf = path->nodes[0];
2544 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2545 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2546 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2547 BUG_ON(ret);
2548 btrfs_release_path(root, path);
2549
2550 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2551 objectid, root->root_key.objectid,
2552 dir->i_ino, &index, name, name_len);
2553 if (ret < 0) {
2554 BUG_ON(ret != -ENOENT);
2555 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2556 name, name_len);
2557 BUG_ON(!di || IS_ERR(di));
2558
2559 leaf = path->nodes[0];
2560 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2561 btrfs_release_path(root, path);
2562 index = key.offset;
2563 }
2564
2565 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2566 index, name, name_len, -1);
2567 BUG_ON(!di || IS_ERR(di));
2568
2569 leaf = path->nodes[0];
2570 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2571 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2572 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2573 BUG_ON(ret);
2574 btrfs_release_path(root, path);
2575
2576 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2577 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2578 ret = btrfs_update_inode(trans, root, dir);
2579 BUG_ON(ret);
2580 dir->i_sb->s_dirt = 1;
2581
2582 btrfs_free_path(path);
2583 return 0;
2584}
2585
Chris Mason39279cc2007-06-12 06:35:45 -04002586static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2587{
2588 struct inode *inode = dentry->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05002589 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002590 int ret;
2591 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04002592 struct btrfs_trans_handle *trans;
Chris Mason1832a6d2007-12-21 16:27:21 -05002593 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002594
Chris Mason3394e162008-11-17 20:42:26 -05002595 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002596 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan134d4512007-10-25 15:49:25 -04002597 return -ENOTEMPTY;
2598
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002599 ret = btrfs_reserve_metadata_space(root, 5);
2600 if (ret)
2601 return ret;
2602
Chris Mason39279cc2007-06-12 06:35:45 -04002603 trans = btrfs_start_transaction(root, 1);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002604 if (IS_ERR(trans)) {
2605 btrfs_unreserve_metadata_space(root, 5);
2606 return PTR_ERR(trans);
2607 }
2608
Chris Mason39279cc2007-06-12 06:35:45 -04002609 btrfs_set_trans_block_group(trans, dir);
Chris Mason39279cc2007-06-12 06:35:45 -04002610
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002611 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2612 err = btrfs_unlink_subvol(trans, root, dir,
2613 BTRFS_I(inode)->location.objectid,
2614 dentry->d_name.name,
2615 dentry->d_name.len);
2616 goto out;
2617 }
2618
Josef Bacik7b128762008-07-24 12:17:14 -04002619 err = btrfs_orphan_add(trans, inode);
2620 if (err)
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002621 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002622
Chris Mason39279cc2007-06-12 06:35:45 -04002623 /* now the directory is empty */
Chris Masone02119d2008-09-05 16:13:11 -04002624 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2625 dentry->d_name.name, dentry->d_name.len);
Chris Masond3977122009-01-05 21:25:51 -05002626 if (!err)
Chris Masondbe674a2008-07-17 12:54:05 -04002627 btrfs_i_size_write(inode, 0);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002628out:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002629 nr = trans->blocks_used;
Chris Mason89ce8a62008-06-25 16:01:31 -04002630 ret = btrfs_end_transaction_throttle(trans, root);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002631 btrfs_unreserve_metadata_space(root, 5);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002632 btrfs_btree_balance_dirty(root, nr);
Chris Mason39544012007-12-12 14:38:19 -05002633
Chris Mason39279cc2007-06-12 06:35:45 -04002634 if (ret && !err)
2635 err = ret;
2636 return err;
2637}
2638
Chris Masond20f7042008-12-08 16:58:54 -05002639#if 0
Chris Mason39279cc2007-06-12 06:35:45 -04002640/*
Chris Mason323ac952008-10-01 19:05:46 -04002641 * when truncating bytes in a file, it is possible to avoid reading
2642 * the leaves that contain only checksum items. This can be the
2643 * majority of the IO required to delete a large file, but it must
2644 * be done carefully.
2645 *
2646 * The keys in the level just above the leaves are checked to make sure
2647 * the lowest key in a given leaf is a csum key, and starts at an offset
2648 * after the new size.
2649 *
2650 * Then the key for the next leaf is checked to make sure it also has
2651 * a checksum item for the same file. If it does, we know our target leaf
2652 * contains only checksum items, and it can be safely freed without reading
2653 * it.
2654 *
2655 * This is just an optimization targeted at large files. It may do
2656 * nothing. It will return 0 unless things went badly.
2657 */
2658static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2659 struct btrfs_root *root,
2660 struct btrfs_path *path,
2661 struct inode *inode, u64 new_size)
2662{
2663 struct btrfs_key key;
2664 int ret;
2665 int nritems;
2666 struct btrfs_key found_key;
2667 struct btrfs_key other_key;
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002668 struct btrfs_leaf_ref *ref;
2669 u64 leaf_gen;
2670 u64 leaf_start;
Chris Mason323ac952008-10-01 19:05:46 -04002671
2672 path->lowest_level = 1;
2673 key.objectid = inode->i_ino;
2674 key.type = BTRFS_CSUM_ITEM_KEY;
2675 key.offset = new_size;
2676again:
2677 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2678 if (ret < 0)
2679 goto out;
2680
2681 if (path->nodes[1] == NULL) {
2682 ret = 0;
2683 goto out;
2684 }
2685 ret = 0;
2686 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2687 nritems = btrfs_header_nritems(path->nodes[1]);
2688
2689 if (!nritems)
2690 goto out;
2691
2692 if (path->slots[1] >= nritems)
2693 goto next_node;
2694
2695 /* did we find a key greater than anything we want to delete? */
2696 if (found_key.objectid > inode->i_ino ||
2697 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2698 goto out;
2699
2700 /* we check the next key in the node to make sure the leave contains
2701 * only checksum items. This comparison doesn't work if our
2702 * leaf is the last one in the node
2703 */
2704 if (path->slots[1] + 1 >= nritems) {
2705next_node:
2706 /* search forward from the last key in the node, this
2707 * will bring us into the next node in the tree
2708 */
2709 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2710
2711 /* unlikely, but we inc below, so check to be safe */
2712 if (found_key.offset == (u64)-1)
2713 goto out;
2714
2715 /* search_forward needs a path with locks held, do the
2716 * search again for the original key. It is possible
2717 * this will race with a balance and return a path that
2718 * we could modify, but this drop is just an optimization
2719 * and is allowed to miss some leaves.
2720 */
2721 btrfs_release_path(root, path);
2722 found_key.offset++;
2723
2724 /* setup a max key for search_forward */
2725 other_key.offset = (u64)-1;
2726 other_key.type = key.type;
2727 other_key.objectid = key.objectid;
2728
2729 path->keep_locks = 1;
2730 ret = btrfs_search_forward(root, &found_key, &other_key,
2731 path, 0, 0);
2732 path->keep_locks = 0;
2733 if (ret || found_key.objectid != key.objectid ||
2734 found_key.type != key.type) {
2735 ret = 0;
2736 goto out;
2737 }
2738
2739 key.offset = found_key.offset;
2740 btrfs_release_path(root, path);
2741 cond_resched();
2742 goto again;
2743 }
2744
2745 /* we know there's one more slot after us in the tree,
2746 * read that key so we can verify it is also a checksum item
2747 */
2748 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2749
2750 if (found_key.objectid < inode->i_ino)
2751 goto next_key;
2752
2753 if (found_key.type != key.type || found_key.offset < new_size)
2754 goto next_key;
2755
2756 /*
2757 * if the key for the next leaf isn't a csum key from this objectid,
2758 * we can't be sure there aren't good items inside this leaf.
2759 * Bail out
2760 */
2761 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2762 goto out;
2763
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002764 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2765 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
Chris Mason323ac952008-10-01 19:05:46 -04002766 /*
2767 * it is safe to delete this leaf, it contains only
2768 * csum items from this inode at an offset >= new_size
2769 */
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002770 ret = btrfs_del_leaf(trans, root, path, leaf_start);
Chris Mason323ac952008-10-01 19:05:46 -04002771 BUG_ON(ret);
2772
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002773 if (root->ref_cows && leaf_gen < trans->transid) {
2774 ref = btrfs_alloc_leaf_ref(root, 0);
2775 if (ref) {
2776 ref->root_gen = root->root_key.offset;
2777 ref->bytenr = leaf_start;
2778 ref->owner = 0;
2779 ref->generation = leaf_gen;
2780 ref->nritems = 0;
2781
Chris Masonbd56b302009-02-04 09:27:02 -05002782 btrfs_sort_leaf_ref(ref);
2783
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002784 ret = btrfs_add_leaf_ref(root, ref, 0);
2785 WARN_ON(ret);
2786 btrfs_free_leaf_ref(root, ref);
2787 } else {
2788 WARN_ON(1);
2789 }
2790 }
Chris Mason323ac952008-10-01 19:05:46 -04002791next_key:
2792 btrfs_release_path(root, path);
2793
2794 if (other_key.objectid == inode->i_ino &&
2795 other_key.type == key.type && other_key.offset > key.offset) {
2796 key.offset = other_key.offset;
2797 cond_resched();
2798 goto again;
2799 }
2800 ret = 0;
2801out:
2802 /* fixup any changes we've made to the path */
2803 path->lowest_level = 0;
2804 path->keep_locks = 0;
2805 btrfs_release_path(root, path);
2806 return ret;
2807}
2808
Chris Masond20f7042008-12-08 16:58:54 -05002809#endif
2810
Chris Mason323ac952008-10-01 19:05:46 -04002811/*
Chris Mason39279cc2007-06-12 06:35:45 -04002812 * this can truncate away extent items, csum items and directory items.
2813 * It starts at a high offset and removes keys until it can't find
Chris Masond352ac62008-09-29 15:18:18 -04002814 * any higher than new_size
Chris Mason39279cc2007-06-12 06:35:45 -04002815 *
2816 * csum items that cross the new i_size are truncated to the new size
2817 * as well.
Josef Bacik7b128762008-07-24 12:17:14 -04002818 *
2819 * min_type is the minimum key type to truncate down to. If set to 0, this
2820 * will kill all the items on this inode, including the INODE_ITEM_KEY.
Chris Mason39279cc2007-06-12 06:35:45 -04002821 */
Yan, Zheng80825102009-11-12 09:35:36 +00002822int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2823 struct btrfs_root *root,
2824 struct inode *inode,
2825 u64 new_size, u32 min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04002826{
Chris Mason39279cc2007-06-12 06:35:45 -04002827 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002828 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002829 struct btrfs_file_extent_item *fi;
Yan, Zheng80825102009-11-12 09:35:36 +00002830 struct btrfs_key key;
2831 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04002832 u64 extent_start = 0;
Chris Masondb945352007-10-15 16:15:53 -04002833 u64 extent_num_bytes = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002834 u64 extent_offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002835 u64 item_end = 0;
Yan, Zheng80825102009-11-12 09:35:36 +00002836 u64 mask = root->sectorsize - 1;
2837 u32 found_type = (u8)-1;
Chris Mason39279cc2007-06-12 06:35:45 -04002838 int found_extent;
2839 int del_item;
Chris Mason85e21ba2008-01-29 15:11:36 -05002840 int pending_del_nr = 0;
2841 int pending_del_slot = 0;
Chris Mason179e29e2007-11-01 11:28:41 -04002842 int extent_type = -1;
Chris Mason771ed682008-11-06 22:02:51 -05002843 int encoding;
Yan, Zheng80825102009-11-12 09:35:36 +00002844 int ret;
2845 int err = 0;
2846
2847 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
Chris Mason39279cc2007-06-12 06:35:45 -04002848
Chris Masone02119d2008-09-05 16:13:11 -04002849 if (root->ref_cows)
Zheng Yan5b21f2e2008-09-26 10:05:38 -04002850 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00002851
Chris Mason39279cc2007-06-12 06:35:45 -04002852 path = btrfs_alloc_path();
2853 BUG_ON(!path);
Julia Lawall33c17ad2009-07-22 16:49:01 -04002854 path->reada = -1;
Chris Mason5f39d392007-10-15 16:14:19 -04002855
Chris Mason39279cc2007-06-12 06:35:45 -04002856 key.objectid = inode->i_ino;
2857 key.offset = (u64)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04002858 key.type = (u8)-1;
2859
Chris Mason85e21ba2008-01-29 15:11:36 -05002860search_again:
Chris Masonb9473432009-03-13 11:00:37 -04002861 path->leave_spinning = 1;
Chris Mason85e21ba2008-01-29 15:11:36 -05002862 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Yan, Zheng80825102009-11-12 09:35:36 +00002863 if (ret < 0) {
2864 err = ret;
2865 goto out;
2866 }
Chris Masond3977122009-01-05 21:25:51 -05002867
Chris Mason85e21ba2008-01-29 15:11:36 -05002868 if (ret > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002869 /* there are no items in the tree for us to truncate, we're
2870 * done
2871 */
Yan, Zheng80825102009-11-12 09:35:36 +00002872 if (path->slots[0] == 0)
2873 goto out;
Chris Mason85e21ba2008-01-29 15:11:36 -05002874 path->slots[0]--;
2875 }
2876
Chris Masond3977122009-01-05 21:25:51 -05002877 while (1) {
Chris Mason39279cc2007-06-12 06:35:45 -04002878 fi = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -04002879 leaf = path->nodes[0];
2880 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2881 found_type = btrfs_key_type(&found_key);
Chris Mason771ed682008-11-06 22:02:51 -05002882 encoding = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002883
Chris Mason5f39d392007-10-15 16:14:19 -04002884 if (found_key.objectid != inode->i_ino)
Chris Mason39279cc2007-06-12 06:35:45 -04002885 break;
Chris Mason5f39d392007-10-15 16:14:19 -04002886
Chris Mason85e21ba2008-01-29 15:11:36 -05002887 if (found_type < min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04002888 break;
2889
Chris Mason5f39d392007-10-15 16:14:19 -04002890 item_end = found_key.offset;
Chris Mason39279cc2007-06-12 06:35:45 -04002891 if (found_type == BTRFS_EXTENT_DATA_KEY) {
Chris Mason5f39d392007-10-15 16:14:19 -04002892 fi = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04002893 struct btrfs_file_extent_item);
Chris Mason179e29e2007-11-01 11:28:41 -04002894 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Mason771ed682008-11-06 22:02:51 -05002895 encoding = btrfs_file_extent_compression(leaf, fi);
2896 encoding |= btrfs_file_extent_encryption(leaf, fi);
2897 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2898
Chris Mason179e29e2007-11-01 11:28:41 -04002899 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04002900 item_end +=
Chris Masondb945352007-10-15 16:15:53 -04002901 btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason179e29e2007-11-01 11:28:41 -04002902 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason179e29e2007-11-01 11:28:41 -04002903 item_end += btrfs_file_extent_inline_len(leaf,
Chris Masonc8b97812008-10-29 14:49:59 -04002904 fi);
Chris Mason39279cc2007-06-12 06:35:45 -04002905 }
Yan008630c2007-11-07 13:31:09 -05002906 item_end--;
Chris Mason39279cc2007-06-12 06:35:45 -04002907 }
Yan, Zheng80825102009-11-12 09:35:36 +00002908 if (found_type > min_type) {
Chris Mason39279cc2007-06-12 06:35:45 -04002909 del_item = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00002910 } else {
2911 if (item_end < new_size)
2912 break;
2913 if (found_key.offset >= new_size)
2914 del_item = 1;
2915 else
2916 del_item = 0;
2917 }
Chris Mason39279cc2007-06-12 06:35:45 -04002918 found_extent = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002919 /* FIXME, shrink the extent if the ref count is only 1 */
Chris Mason179e29e2007-11-01 11:28:41 -04002920 if (found_type != BTRFS_EXTENT_DATA_KEY)
2921 goto delete;
2922
2923 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason39279cc2007-06-12 06:35:45 -04002924 u64 num_dec;
Chris Masondb945352007-10-15 16:15:53 -04002925 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
Chris Mason771ed682008-11-06 22:02:51 -05002926 if (!del_item && !encoding) {
Chris Masondb945352007-10-15 16:15:53 -04002927 u64 orig_num_bytes =
2928 btrfs_file_extent_num_bytes(leaf, fi);
Chris Masone02119d2008-09-05 16:13:11 -04002929 extent_num_bytes = new_size -
Chris Mason5f39d392007-10-15 16:14:19 -04002930 found_key.offset + root->sectorsize - 1;
Yanb1632b12008-01-30 11:54:04 -05002931 extent_num_bytes = extent_num_bytes &
2932 ~((u64)root->sectorsize - 1);
Chris Masondb945352007-10-15 16:15:53 -04002933 btrfs_set_file_extent_num_bytes(leaf, fi,
2934 extent_num_bytes);
2935 num_dec = (orig_num_bytes -
Chris Mason90692182008-02-08 13:49:28 -05002936 extent_num_bytes);
Chris Masone02119d2008-09-05 16:13:11 -04002937 if (root->ref_cows && extent_start != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002938 inode_sub_bytes(inode, num_dec);
Chris Mason5f39d392007-10-15 16:14:19 -04002939 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04002940 } else {
Chris Masondb945352007-10-15 16:15:53 -04002941 extent_num_bytes =
2942 btrfs_file_extent_disk_num_bytes(leaf,
2943 fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002944 extent_offset = found_key.offset -
2945 btrfs_file_extent_offset(leaf, fi);
2946
Chris Mason39279cc2007-06-12 06:35:45 -04002947 /* FIXME blocksize != 4096 */
Chris Mason90692182008-02-08 13:49:28 -05002948 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason39279cc2007-06-12 06:35:45 -04002949 if (extent_start != 0) {
2950 found_extent = 1;
Chris Masone02119d2008-09-05 16:13:11 -04002951 if (root->ref_cows)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002952 inode_sub_bytes(inode, num_dec);
Chris Mason39279cc2007-06-12 06:35:45 -04002953 }
2954 }
Chris Mason90692182008-02-08 13:49:28 -05002955 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -04002956 /*
2957 * we can't truncate inline items that have had
2958 * special encodings
2959 */
2960 if (!del_item &&
2961 btrfs_file_extent_compression(leaf, fi) == 0 &&
2962 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2963 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04002964 u32 size = new_size - found_key.offset;
2965
2966 if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002967 inode_sub_bytes(inode, item_end + 1 -
2968 new_size);
Chris Masone02119d2008-09-05 16:13:11 -04002969 }
2970 size =
2971 btrfs_file_extent_calc_inline_size(size);
Chris Mason90692182008-02-08 13:49:28 -05002972 ret = btrfs_truncate_item(trans, root, path,
Chris Masone02119d2008-09-05 16:13:11 -04002973 size, 1);
Chris Mason90692182008-02-08 13:49:28 -05002974 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04002975 } else if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002976 inode_sub_bytes(inode, item_end + 1 -
2977 found_key.offset);
Chris Mason90692182008-02-08 13:49:28 -05002978 }
Chris Mason39279cc2007-06-12 06:35:45 -04002979 }
Chris Mason179e29e2007-11-01 11:28:41 -04002980delete:
Chris Mason39279cc2007-06-12 06:35:45 -04002981 if (del_item) {
Chris Mason85e21ba2008-01-29 15:11:36 -05002982 if (!pending_del_nr) {
2983 /* no pending yet, add ourselves */
2984 pending_del_slot = path->slots[0];
2985 pending_del_nr = 1;
2986 } else if (pending_del_nr &&
2987 path->slots[0] + 1 == pending_del_slot) {
2988 /* hop on the pending chunk */
2989 pending_del_nr++;
2990 pending_del_slot = path->slots[0];
2991 } else {
Chris Masond3977122009-01-05 21:25:51 -05002992 BUG();
Chris Mason85e21ba2008-01-29 15:11:36 -05002993 }
Chris Mason39279cc2007-06-12 06:35:45 -04002994 } else {
2995 break;
2996 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002997 if (found_extent && root->ref_cows) {
Chris Masonb9473432009-03-13 11:00:37 -04002998 btrfs_set_path_blocking(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002999 ret = btrfs_free_extent(trans, root, extent_start,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003000 extent_num_bytes, 0,
3001 btrfs_header_owner(leaf),
3002 inode->i_ino, extent_offset);
Chris Mason39279cc2007-06-12 06:35:45 -04003003 BUG_ON(ret);
3004 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003005
Yan, Zheng80825102009-11-12 09:35:36 +00003006 if (found_type == BTRFS_INODE_ITEM_KEY)
3007 break;
3008
3009 if (path->slots[0] == 0 ||
3010 path->slots[0] != pending_del_slot) {
3011 if (root->ref_cows) {
3012 err = -EAGAIN;
3013 goto out;
3014 }
3015 if (pending_del_nr) {
3016 ret = btrfs_del_items(trans, root, path,
3017 pending_del_slot,
3018 pending_del_nr);
3019 BUG_ON(ret);
3020 pending_del_nr = 0;
3021 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003022 btrfs_release_path(root, path);
3023 goto search_again;
Yan, Zheng80825102009-11-12 09:35:36 +00003024 } else {
3025 path->slots[0]--;
Chris Mason85e21ba2008-01-29 15:11:36 -05003026 }
Chris Mason39279cc2007-06-12 06:35:45 -04003027 }
Yan, Zheng80825102009-11-12 09:35:36 +00003028out:
Chris Mason85e21ba2008-01-29 15:11:36 -05003029 if (pending_del_nr) {
3030 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3031 pending_del_nr);
3032 }
Chris Mason39279cc2007-06-12 06:35:45 -04003033 btrfs_free_path(path);
Yan, Zheng80825102009-11-12 09:35:36 +00003034 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003035}
3036
Chris Masona52d9a82007-08-27 16:49:44 -04003037/*
3038 * taken from block_truncate_page, but does cow as it zeros out
3039 * any bytes left in the last page in the file.
3040 */
3041static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3042{
3043 struct inode *inode = mapping->host;
Chris Masondb945352007-10-15 16:15:53 -04003044 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003045 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3046 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003047 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003048 char *kaddr;
Chris Masondb945352007-10-15 16:15:53 -04003049 u32 blocksize = root->sectorsize;
Chris Masona52d9a82007-08-27 16:49:44 -04003050 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3051 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3052 struct page *page;
3053 int ret = 0;
3054 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003055 u64 page_end;
Chris Masona52d9a82007-08-27 16:49:44 -04003056
3057 if ((offset & (blocksize - 1)) == 0)
3058 goto out;
Josef Bacik5d5e1032009-10-13 16:46:49 -04003059 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
3060 if (ret)
3061 goto out;
3062
3063 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
3064 if (ret)
3065 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04003066
3067 ret = -ENOMEM;
Chris Mason211c17f2008-05-15 09:13:45 -04003068again:
Chris Masona52d9a82007-08-27 16:49:44 -04003069 page = grab_cache_page(mapping, index);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003070 if (!page) {
3071 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3072 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
Chris Masona52d9a82007-08-27 16:49:44 -04003073 goto out;
Josef Bacik5d5e1032009-10-13 16:46:49 -04003074 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04003075
3076 page_start = page_offset(page);
3077 page_end = page_start + PAGE_CACHE_SIZE - 1;
3078
Chris Masona52d9a82007-08-27 16:49:44 -04003079 if (!PageUptodate(page)) {
3080 ret = btrfs_readpage(NULL, page);
3081 lock_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04003082 if (page->mapping != mapping) {
3083 unlock_page(page);
3084 page_cache_release(page);
3085 goto again;
3086 }
Chris Masona52d9a82007-08-27 16:49:44 -04003087 if (!PageUptodate(page)) {
3088 ret = -EIO;
Chris Mason89642222008-07-24 09:41:53 -04003089 goto out_unlock;
Chris Masona52d9a82007-08-27 16:49:44 -04003090 }
3091 }
Chris Mason211c17f2008-05-15 09:13:45 -04003092 wait_on_page_writeback(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003093
Josef Bacik2ac55d42010-02-03 19:33:23 +00003094 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3095 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003096 set_page_extent_mapped(page);
3097
3098 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3099 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00003100 unlock_extent_cached(io_tree, page_start, page_end,
3101 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003102 unlock_page(page);
3103 page_cache_release(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04003104 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003105 btrfs_put_ordered_extent(ordered);
3106 goto again;
3107 }
3108
Josef Bacik2ac55d42010-02-03 19:33:23 +00003109 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Josef Bacik5d5e1032009-10-13 16:46:49 -04003110 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
Josef Bacik2ac55d42010-02-03 19:33:23 +00003111 0, 0, &cached_state, GFP_NOFS);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003112
Josef Bacik2ac55d42010-02-03 19:33:23 +00003113 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3114 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003115 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00003116 unlock_extent_cached(io_tree, page_start, page_end,
3117 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003118 goto out_unlock;
3119 }
3120
Chris Masone6dcd2d2008-07-17 12:53:50 -04003121 ret = 0;
3122 if (offset != PAGE_CACHE_SIZE) {
3123 kaddr = kmap(page);
3124 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3125 flush_dcache_page(page);
3126 kunmap(page);
3127 }
Chris Mason247e7432008-07-17 12:53:51 -04003128 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003129 set_page_dirty(page);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003130 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3131 GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04003132
Chris Mason89642222008-07-24 09:41:53 -04003133out_unlock:
Josef Bacik5d5e1032009-10-13 16:46:49 -04003134 if (ret)
3135 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3136 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04003137 unlock_page(page);
3138 page_cache_release(page);
3139out:
3140 return ret;
3141}
3142
Yan Zheng9036c102008-10-30 14:19:41 -04003143int btrfs_cont_expand(struct inode *inode, loff_t size)
3144{
3145 struct btrfs_trans_handle *trans;
3146 struct btrfs_root *root = BTRFS_I(inode)->root;
3147 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3148 struct extent_map *em;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003149 struct extent_state *cached_state = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04003150 u64 mask = root->sectorsize - 1;
3151 u64 hole_start = (inode->i_size + mask) & ~mask;
3152 u64 block_end = (size + mask) & ~mask;
3153 u64 last_byte;
3154 u64 cur_offset;
3155 u64 hole_size;
Josef Bacik9ed74f22009-09-11 16:12:44 -04003156 int err = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04003157
3158 if (size <= hole_start)
3159 return 0;
3160
Yan Zheng9036c102008-10-30 14:19:41 -04003161 while (1) {
3162 struct btrfs_ordered_extent *ordered;
3163 btrfs_wait_ordered_range(inode, hole_start,
3164 block_end - hole_start);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003165 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3166 &cached_state, GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003167 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3168 if (!ordered)
3169 break;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003170 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3171 &cached_state, GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003172 btrfs_put_ordered_extent(ordered);
3173 }
3174
Yan Zheng9036c102008-10-30 14:19:41 -04003175 cur_offset = hole_start;
3176 while (1) {
3177 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3178 block_end - cur_offset, 0);
3179 BUG_ON(IS_ERR(em) || !em);
3180 last_byte = min(extent_map_end(em), block_end);
3181 last_byte = (last_byte + mask) & ~mask;
Yan, Zheng80825102009-11-12 09:35:36 +00003182 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
Chris Mason771ed682008-11-06 22:02:51 -05003183 u64 hint_byte = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04003184 hole_size = last_byte - cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00003185
3186 err = btrfs_reserve_metadata_space(root, 2);
Chris Mason771ed682008-11-06 22:02:51 -05003187 if (err)
3188 break;
Josef Bacik9ed74f22009-09-11 16:12:44 -04003189
Yan, Zheng80825102009-11-12 09:35:36 +00003190 trans = btrfs_start_transaction(root, 1);
3191 btrfs_set_trans_block_group(trans, inode);
3192
3193 err = btrfs_drop_extents(trans, inode, cur_offset,
3194 cur_offset + hole_size,
3195 &hint_byte, 1);
3196 BUG_ON(err);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003197
Yan Zheng9036c102008-10-30 14:19:41 -04003198 err = btrfs_insert_file_extent(trans, root,
3199 inode->i_ino, cur_offset, 0,
3200 0, hole_size, 0, hole_size,
3201 0, 0, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003202 BUG_ON(err);
3203
Yan Zheng9036c102008-10-30 14:19:41 -04003204 btrfs_drop_extent_cache(inode, hole_start,
3205 last_byte - 1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003206
3207 btrfs_end_transaction(trans, root);
3208 btrfs_unreserve_metadata_space(root, 2);
Yan Zheng9036c102008-10-30 14:19:41 -04003209 }
3210 free_extent_map(em);
3211 cur_offset = last_byte;
Yan, Zheng80825102009-11-12 09:35:36 +00003212 if (cur_offset >= block_end)
Yan Zheng9036c102008-10-30 14:19:41 -04003213 break;
3214 }
3215
Josef Bacik2ac55d42010-02-03 19:33:23 +00003216 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3217 GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003218 return err;
3219}
3220
Yan, Zheng80825102009-11-12 09:35:36 +00003221static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3222{
3223 struct btrfs_root *root = BTRFS_I(inode)->root;
3224 struct btrfs_trans_handle *trans;
3225 unsigned long nr;
3226 int ret;
3227
3228 if (attr->ia_size == inode->i_size)
3229 return 0;
3230
3231 if (attr->ia_size > inode->i_size) {
3232 unsigned long limit;
3233 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3234 if (attr->ia_size > inode->i_sb->s_maxbytes)
3235 return -EFBIG;
3236 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3237 send_sig(SIGXFSZ, current, 0);
3238 return -EFBIG;
3239 }
3240 }
3241
3242 ret = btrfs_reserve_metadata_space(root, 1);
3243 if (ret)
3244 return ret;
3245
3246 trans = btrfs_start_transaction(root, 1);
3247 btrfs_set_trans_block_group(trans, inode);
3248
3249 ret = btrfs_orphan_add(trans, inode);
3250 BUG_ON(ret);
3251
3252 nr = trans->blocks_used;
3253 btrfs_end_transaction(trans, root);
3254 btrfs_unreserve_metadata_space(root, 1);
3255 btrfs_btree_balance_dirty(root, nr);
3256
3257 if (attr->ia_size > inode->i_size) {
3258 ret = btrfs_cont_expand(inode, attr->ia_size);
3259 if (ret) {
3260 btrfs_truncate(inode);
3261 return ret;
3262 }
3263
3264 i_size_write(inode, attr->ia_size);
3265 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3266
3267 trans = btrfs_start_transaction(root, 1);
3268 btrfs_set_trans_block_group(trans, inode);
3269
3270 ret = btrfs_update_inode(trans, root, inode);
3271 BUG_ON(ret);
3272 if (inode->i_nlink > 0) {
3273 ret = btrfs_orphan_del(trans, inode);
3274 BUG_ON(ret);
3275 }
3276 nr = trans->blocks_used;
3277 btrfs_end_transaction(trans, root);
3278 btrfs_btree_balance_dirty(root, nr);
3279 return 0;
3280 }
3281
3282 /*
3283 * We're truncating a file that used to have good data down to
3284 * zero. Make sure it gets into the ordered flush list so that
3285 * any new writes get down to disk quickly.
3286 */
3287 if (attr->ia_size == 0)
3288 BTRFS_I(inode)->ordered_data_close = 1;
3289
3290 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3291 ret = vmtruncate(inode, attr->ia_size);
3292 BUG_ON(ret);
3293
3294 return 0;
3295}
3296
Chris Mason39279cc2007-06-12 06:35:45 -04003297static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3298{
3299 struct inode *inode = dentry->d_inode;
3300 int err;
3301
3302 err = inode_change_ok(inode, attr);
3303 if (err)
3304 return err;
3305
Chris Mason5a3f23d2009-03-31 13:27:11 -04003306 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Yan, Zheng80825102009-11-12 09:35:36 +00003307 err = btrfs_setattr_size(inode, attr);
3308 if (err)
3309 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003310 }
Yan, Zheng80825102009-11-12 09:35:36 +00003311 attr->ia_valid &= ~ATTR_SIZE;
Yan Zheng9036c102008-10-30 14:19:41 -04003312
Yan, Zheng80825102009-11-12 09:35:36 +00003313 if (attr->ia_valid)
3314 err = inode_setattr(inode, attr);
Josef Bacik33268ea2008-07-24 12:16:36 -04003315
3316 if (!err && ((attr->ia_valid & ATTR_MODE)))
3317 err = btrfs_acl_chmod(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003318 return err;
3319}
Chris Mason61295eb2008-01-14 16:24:38 -05003320
Chris Mason39279cc2007-06-12 06:35:45 -04003321void btrfs_delete_inode(struct inode *inode)
3322{
3323 struct btrfs_trans_handle *trans;
3324 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003325 unsigned long nr;
Chris Mason39279cc2007-06-12 06:35:45 -04003326 int ret;
3327
3328 truncate_inode_pages(&inode->i_data, 0);
3329 if (is_bad_inode(inode)) {
Josef Bacik7b128762008-07-24 12:17:14 -04003330 btrfs_orphan_del(NULL, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003331 goto no_delete;
3332 }
Chris Mason4a096752008-07-21 10:29:44 -04003333 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Chris Mason5f39d392007-10-15 16:14:19 -04003334
Yan, Zhengc71bf092009-11-12 09:34:40 +00003335 if (root->fs_info->log_root_recovering) {
3336 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3337 goto no_delete;
3338 }
3339
Yan, Zheng76dda932009-09-21 16:00:26 -04003340 if (inode->i_nlink > 0) {
3341 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3342 goto no_delete;
3343 }
3344
Chris Masondbe674a2008-07-17 12:54:05 -04003345 btrfs_i_size_write(inode, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04003346
Yan, Zheng80825102009-11-12 09:35:36 +00003347 while (1) {
3348 trans = btrfs_start_transaction(root, 1);
3349 btrfs_set_trans_block_group(trans, inode);
3350 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3351
3352 if (ret != -EAGAIN)
3353 break;
3354
3355 nr = trans->blocks_used;
3356 btrfs_end_transaction(trans, root);
3357 trans = NULL;
3358 btrfs_btree_balance_dirty(root, nr);
Josef Bacik7b128762008-07-24 12:17:14 -04003359 }
3360
Yan, Zheng80825102009-11-12 09:35:36 +00003361 if (ret == 0) {
3362 ret = btrfs_orphan_del(trans, inode);
3363 BUG_ON(ret);
3364 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003365
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003366 nr = trans->blocks_used;
Chris Mason54aa1f42007-06-22 14:16:25 -04003367 btrfs_end_transaction(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003368 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04003369no_delete:
3370 clear_inode(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003371 return;
Chris Mason39279cc2007-06-12 06:35:45 -04003372}
3373
3374/*
3375 * this returns the key found in the dir entry in the location pointer.
3376 * If no dir entries were found, location->objectid is 0.
3377 */
3378static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3379 struct btrfs_key *location)
3380{
3381 const char *name = dentry->d_name.name;
3382 int namelen = dentry->d_name.len;
3383 struct btrfs_dir_item *di;
3384 struct btrfs_path *path;
3385 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan0d9f7f32007-10-25 15:48:28 -04003386 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003387
3388 path = btrfs_alloc_path();
3389 BUG_ON(!path);
Chris Mason39544012007-12-12 14:38:19 -05003390
Chris Mason39279cc2007-06-12 06:35:45 -04003391 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3392 namelen, 0);
Yan0d9f7f32007-10-25 15:48:28 -04003393 if (IS_ERR(di))
3394 ret = PTR_ERR(di);
Chris Masond3977122009-01-05 21:25:51 -05003395
3396 if (!di || IS_ERR(di))
Chris Mason39544012007-12-12 14:38:19 -05003397 goto out_err;
Chris Masond3977122009-01-05 21:25:51 -05003398
Chris Mason5f39d392007-10-15 16:14:19 -04003399 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
Chris Mason39279cc2007-06-12 06:35:45 -04003400out:
Chris Mason39279cc2007-06-12 06:35:45 -04003401 btrfs_free_path(path);
3402 return ret;
Chris Mason39544012007-12-12 14:38:19 -05003403out_err:
3404 location->objectid = 0;
3405 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04003406}
3407
3408/*
3409 * when we hit a tree root in a directory, the btrfs part of the inode
3410 * needs to be changed to reflect the root directory of the tree root. This
3411 * is kind of like crossing a mount point.
3412 */
3413static int fixup_tree_root_location(struct btrfs_root *root,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003414 struct inode *dir,
3415 struct dentry *dentry,
3416 struct btrfs_key *location,
3417 struct btrfs_root **sub_root)
Chris Mason39279cc2007-06-12 06:35:45 -04003418{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003419 struct btrfs_path *path;
3420 struct btrfs_root *new_root;
3421 struct btrfs_root_ref *ref;
3422 struct extent_buffer *leaf;
3423 int ret;
3424 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003425
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003426 path = btrfs_alloc_path();
3427 if (!path) {
3428 err = -ENOMEM;
3429 goto out;
3430 }
Chris Mason39279cc2007-06-12 06:35:45 -04003431
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003432 err = -ENOENT;
3433 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3434 BTRFS_I(dir)->root->root_key.objectid,
3435 location->objectid);
3436 if (ret) {
3437 if (ret < 0)
3438 err = ret;
3439 goto out;
3440 }
Chris Mason39279cc2007-06-12 06:35:45 -04003441
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003442 leaf = path->nodes[0];
3443 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3444 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3445 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3446 goto out;
3447
3448 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3449 (unsigned long)(ref + 1),
3450 dentry->d_name.len);
3451 if (ret)
3452 goto out;
3453
3454 btrfs_release_path(root->fs_info->tree_root, path);
3455
3456 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3457 if (IS_ERR(new_root)) {
3458 err = PTR_ERR(new_root);
3459 goto out;
3460 }
3461
3462 if (btrfs_root_refs(&new_root->root_item) == 0) {
3463 err = -ENOENT;
3464 goto out;
3465 }
3466
3467 *sub_root = new_root;
3468 location->objectid = btrfs_root_dirid(&new_root->root_item);
3469 location->type = BTRFS_INODE_ITEM_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04003470 location->offset = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003471 err = 0;
3472out:
3473 btrfs_free_path(path);
3474 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003475}
3476
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003477static void inode_tree_add(struct inode *inode)
3478{
3479 struct btrfs_root *root = BTRFS_I(inode)->root;
3480 struct btrfs_inode *entry;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003481 struct rb_node **p;
3482 struct rb_node *parent;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003483again:
3484 p = &root->inode_tree.rb_node;
3485 parent = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003486
Yan, Zheng76dda932009-09-21 16:00:26 -04003487 if (hlist_unhashed(&inode->i_hash))
3488 return;
3489
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003490 spin_lock(&root->inode_lock);
3491 while (*p) {
3492 parent = *p;
3493 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3494
3495 if (inode->i_ino < entry->vfs_inode.i_ino)
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003496 p = &parent->rb_left;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003497 else if (inode->i_ino > entry->vfs_inode.i_ino)
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003498 p = &parent->rb_right;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003499 else {
3500 WARN_ON(!(entry->vfs_inode.i_state &
3501 (I_WILL_FREE | I_FREEING | I_CLEAR)));
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003502 rb_erase(parent, &root->inode_tree);
3503 RB_CLEAR_NODE(parent);
3504 spin_unlock(&root->inode_lock);
3505 goto again;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003506 }
3507 }
3508 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3509 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3510 spin_unlock(&root->inode_lock);
3511}
3512
3513static void inode_tree_del(struct inode *inode)
3514{
3515 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng76dda932009-09-21 16:00:26 -04003516 int empty = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003517
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003518 spin_lock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003519 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003520 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003521 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Yan, Zheng76dda932009-09-21 16:00:26 -04003522 empty = RB_EMPTY_ROOT(&root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003523 }
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003524 spin_unlock(&root->inode_lock);
Yan, Zheng76dda932009-09-21 16:00:26 -04003525
3526 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3527 synchronize_srcu(&root->fs_info->subvol_srcu);
3528 spin_lock(&root->inode_lock);
3529 empty = RB_EMPTY_ROOT(&root->inode_tree);
3530 spin_unlock(&root->inode_lock);
3531 if (empty)
3532 btrfs_add_dead_root(root);
3533 }
3534}
3535
3536int btrfs_invalidate_inodes(struct btrfs_root *root)
3537{
3538 struct rb_node *node;
3539 struct rb_node *prev;
3540 struct btrfs_inode *entry;
3541 struct inode *inode;
3542 u64 objectid = 0;
3543
3544 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3545
3546 spin_lock(&root->inode_lock);
3547again:
3548 node = root->inode_tree.rb_node;
3549 prev = NULL;
3550 while (node) {
3551 prev = node;
3552 entry = rb_entry(node, struct btrfs_inode, rb_node);
3553
3554 if (objectid < entry->vfs_inode.i_ino)
3555 node = node->rb_left;
3556 else if (objectid > entry->vfs_inode.i_ino)
3557 node = node->rb_right;
3558 else
3559 break;
3560 }
3561 if (!node) {
3562 while (prev) {
3563 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3564 if (objectid <= entry->vfs_inode.i_ino) {
3565 node = prev;
3566 break;
3567 }
3568 prev = rb_next(prev);
3569 }
3570 }
3571 while (node) {
3572 entry = rb_entry(node, struct btrfs_inode, rb_node);
3573 objectid = entry->vfs_inode.i_ino + 1;
3574 inode = igrab(&entry->vfs_inode);
3575 if (inode) {
3576 spin_unlock(&root->inode_lock);
3577 if (atomic_read(&inode->i_count) > 1)
3578 d_prune_aliases(inode);
3579 /*
3580 * btrfs_drop_inode will remove it from
3581 * the inode cache when its usage count
3582 * hits zero.
3583 */
3584 iput(inode);
3585 cond_resched();
3586 spin_lock(&root->inode_lock);
3587 goto again;
3588 }
3589
3590 if (cond_resched_lock(&root->inode_lock))
3591 goto again;
3592
3593 node = rb_next(node);
3594 }
3595 spin_unlock(&root->inode_lock);
3596 return 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003597}
3598
Chris Masone02119d2008-09-05 16:13:11 -04003599static noinline void init_btrfs_i(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003600{
Chris Masone02119d2008-09-05 16:13:11 -04003601 struct btrfs_inode *bi = BTRFS_I(inode);
3602
Chris Masone02119d2008-09-05 16:13:11 -04003603 bi->generation = 0;
Chris Masonc3027eb2008-12-08 16:40:21 -05003604 bi->sequence = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003605 bi->last_trans = 0;
Chris Mason257c62e2009-10-13 13:21:08 -04003606 bi->last_sub_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003607 bi->logged_trans = 0;
3608 bi->delalloc_bytes = 0;
Josef Bacik6a632092009-02-20 11:00:09 -05003609 bi->reserved_bytes = 0;
Chris Masone02119d2008-09-05 16:13:11 -04003610 bi->disk_i_size = 0;
3611 bi->flags = 0;
3612 bi->index_cnt = (u64)-1;
Chris Mason12fcfd22009-03-24 10:24:20 -04003613 bi->last_unlink_trans = 0;
Chris Mason27574952009-05-14 13:10:02 -04003614 bi->ordered_data_close = 0;
Chris Mason1e701a32010-03-11 09:42:04 -05003615 bi->force_compress = 0;
Chris Masond1310b22008-01-24 16:13:08 -05003616 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3617 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
Chris Masonb888db22007-08-27 16:49:44 -04003618 inode->i_mapping, GFP_NOFS);
Chris Mason7e383262008-04-09 16:28:12 -04003619 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3620 inode->i_mapping, GFP_NOFS);
Chris Masonea8c2812008-08-04 23:17:27 -04003621 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
Chris Mason5a3f23d2009-03-31 13:27:11 -04003622 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003623 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Chris Masonba1da2f2008-07-17 12:54:15 -04003624 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
Chris Masone02119d2008-09-05 16:13:11 -04003625 mutex_init(&BTRFS_I(inode)->log_mutex);
3626}
3627
3628static int btrfs_init_locked_inode(struct inode *inode, void *p)
3629{
3630 struct btrfs_iget_args *args = p;
3631 inode->i_ino = args->ino;
3632 init_btrfs_i(inode);
3633 BTRFS_I(inode)->root = args->root;
Josef Bacik6a632092009-02-20 11:00:09 -05003634 btrfs_set_inode_space_info(args->root, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003635 return 0;
3636}
3637
3638static int btrfs_find_actor(struct inode *inode, void *opaque)
3639{
3640 struct btrfs_iget_args *args = opaque;
Chris Masond3977122009-01-05 21:25:51 -05003641 return args->ino == inode->i_ino &&
3642 args->root == BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003643}
3644
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003645static struct inode *btrfs_iget_locked(struct super_block *s,
3646 u64 objectid,
3647 struct btrfs_root *root)
Chris Mason39279cc2007-06-12 06:35:45 -04003648{
3649 struct inode *inode;
3650 struct btrfs_iget_args args;
3651 args.ino = objectid;
3652 args.root = root;
3653
3654 inode = iget5_locked(s, objectid, btrfs_find_actor,
3655 btrfs_init_locked_inode,
3656 (void *)&args);
3657 return inode;
3658}
3659
Balaji Rao1a54ef82008-07-21 02:01:04 +05303660/* Get an inode object given its location and corresponding root.
3661 * Returns in *is_new if the inode was read from disk
3662 */
3663struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Josef Bacik73f73412009-12-04 17:38:27 +00003664 struct btrfs_root *root, int *new)
Balaji Rao1a54ef82008-07-21 02:01:04 +05303665{
3666 struct inode *inode;
3667
3668 inode = btrfs_iget_locked(s, location->objectid, root);
3669 if (!inode)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003670 return ERR_PTR(-ENOMEM);
Balaji Rao1a54ef82008-07-21 02:01:04 +05303671
3672 if (inode->i_state & I_NEW) {
3673 BTRFS_I(inode)->root = root;
3674 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3675 btrfs_read_locked_inode(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003676
3677 inode_tree_add(inode);
Balaji Rao1a54ef82008-07-21 02:01:04 +05303678 unlock_new_inode(inode);
Josef Bacik73f73412009-12-04 17:38:27 +00003679 if (new)
3680 *new = 1;
Balaji Rao1a54ef82008-07-21 02:01:04 +05303681 }
3682
3683 return inode;
3684}
3685
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003686static struct inode *new_simple_dir(struct super_block *s,
3687 struct btrfs_key *key,
3688 struct btrfs_root *root)
3689{
3690 struct inode *inode = new_inode(s);
3691
3692 if (!inode)
3693 return ERR_PTR(-ENOMEM);
3694
3695 init_btrfs_i(inode);
3696
3697 BTRFS_I(inode)->root = root;
3698 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3699 BTRFS_I(inode)->dummy_inode = 1;
3700
3701 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3702 inode->i_op = &simple_dir_inode_operations;
3703 inode->i_fop = &simple_dir_operations;
3704 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3705 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3706
3707 return inode;
3708}
3709
Chris Mason3de45862008-11-17 21:02:50 -05003710struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04003711{
Chris Masond3977122009-01-05 21:25:51 -05003712 struct inode *inode;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003713 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003714 struct btrfs_root *sub_root = root;
3715 struct btrfs_key location;
Yan, Zheng76dda932009-09-21 16:00:26 -04003716 int index;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003717 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04003718
Yan, Zheng76dda932009-09-21 16:00:26 -04003719 dentry->d_op = &btrfs_dentry_operations;
3720
Chris Mason39279cc2007-06-12 06:35:45 -04003721 if (dentry->d_name.len > BTRFS_NAME_LEN)
3722 return ERR_PTR(-ENAMETOOLONG);
Chris Mason5f39d392007-10-15 16:14:19 -04003723
Chris Mason39279cc2007-06-12 06:35:45 -04003724 ret = btrfs_inode_by_name(dir, dentry, &location);
Chris Mason5f39d392007-10-15 16:14:19 -04003725
Chris Mason39279cc2007-06-12 06:35:45 -04003726 if (ret < 0)
3727 return ERR_PTR(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04003728
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003729 if (location.objectid == 0)
3730 return NULL;
3731
3732 if (location.type == BTRFS_INODE_ITEM_KEY) {
Josef Bacik73f73412009-12-04 17:38:27 +00003733 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003734 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04003735 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003736
3737 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3738
Yan, Zheng76dda932009-09-21 16:00:26 -04003739 index = srcu_read_lock(&root->fs_info->subvol_srcu);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003740 ret = fixup_tree_root_location(root, dir, dentry,
3741 &location, &sub_root);
3742 if (ret < 0) {
3743 if (ret != -ENOENT)
3744 inode = ERR_PTR(ret);
3745 else
3746 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3747 } else {
Josef Bacik73f73412009-12-04 17:38:27 +00003748 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04003749 }
Yan, Zheng76dda932009-09-21 16:00:26 -04003750 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3751
Yan, Zhengc71bf092009-11-12 09:34:40 +00003752 if (root != sub_root) {
3753 down_read(&root->fs_info->cleanup_work_sem);
3754 if (!(inode->i_sb->s_flags & MS_RDONLY))
3755 btrfs_orphan_cleanup(sub_root);
3756 up_read(&root->fs_info->cleanup_work_sem);
3757 }
3758
Chris Mason3de45862008-11-17 21:02:50 -05003759 return inode;
3760}
3761
Yan, Zheng76dda932009-09-21 16:00:26 -04003762static int btrfs_dentry_delete(struct dentry *dentry)
3763{
3764 struct btrfs_root *root;
3765
Yan, Zhengefefb142009-10-09 09:25:16 -04003766 if (!dentry->d_inode && !IS_ROOT(dentry))
3767 dentry = dentry->d_parent;
Yan, Zheng76dda932009-09-21 16:00:26 -04003768
Yan, Zhengefefb142009-10-09 09:25:16 -04003769 if (dentry->d_inode) {
3770 root = BTRFS_I(dentry->d_inode)->root;
3771 if (btrfs_root_refs(&root->root_item) == 0)
3772 return 1;
3773 }
Yan, Zheng76dda932009-09-21 16:00:26 -04003774 return 0;
3775}
3776
Chris Mason3de45862008-11-17 21:02:50 -05003777static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3778 struct nameidata *nd)
3779{
3780 struct inode *inode;
3781
Chris Mason3de45862008-11-17 21:02:50 -05003782 inode = btrfs_lookup_dentry(dir, dentry);
3783 if (IS_ERR(inode))
3784 return ERR_CAST(inode);
Josef Bacik7b128762008-07-24 12:17:14 -04003785
Chris Mason39279cc2007-06-12 06:35:45 -04003786 return d_splice_alias(inode, dentry);
3787}
3788
Chris Mason39279cc2007-06-12 06:35:45 -04003789static unsigned char btrfs_filetype_table[] = {
3790 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3791};
3792
David Woodhousecbdf5a22008-08-06 19:42:33 +01003793static int btrfs_real_readdir(struct file *filp, void *dirent,
3794 filldir_t filldir)
Chris Mason39279cc2007-06-12 06:35:45 -04003795{
Chris Mason6da6aba2007-12-18 16:15:09 -05003796 struct inode *inode = filp->f_dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04003797 struct btrfs_root *root = BTRFS_I(inode)->root;
3798 struct btrfs_item *item;
3799 struct btrfs_dir_item *di;
3800 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04003801 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04003802 struct btrfs_path *path;
3803 int ret;
3804 u32 nritems;
Chris Mason5f39d392007-10-15 16:14:19 -04003805 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003806 int slot;
3807 int advance;
3808 unsigned char d_type;
3809 int over = 0;
3810 u32 di_cur;
3811 u32 di_total;
3812 u32 di_len;
3813 int key_type = BTRFS_DIR_INDEX_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04003814 char tmp_name[32];
3815 char *name_ptr;
3816 int name_len;
Chris Mason39279cc2007-06-12 06:35:45 -04003817
3818 /* FIXME, use a real flag for deciding about the key type */
3819 if (root->fs_info->tree_root == root)
3820 key_type = BTRFS_DIR_ITEM_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04003821
Chris Mason39544012007-12-12 14:38:19 -05003822 /* special case for "." */
3823 if (filp->f_pos == 0) {
3824 over = filldir(dirent, ".", 1,
3825 1, inode->i_ino,
3826 DT_DIR);
3827 if (over)
3828 return 0;
3829 filp->f_pos = 1;
3830 }
Chris Mason39544012007-12-12 14:38:19 -05003831 /* special case for .., just use the back ref */
3832 if (filp->f_pos == 1) {
David Woodhouse5ecc7e52008-08-17 15:14:48 +01003833 u64 pino = parent_ino(filp->f_path.dentry);
Chris Mason39544012007-12-12 14:38:19 -05003834 over = filldir(dirent, "..", 2,
David Woodhouse5ecc7e52008-08-17 15:14:48 +01003835 2, pino, DT_DIR);
Chris Mason39544012007-12-12 14:38:19 -05003836 if (over)
David Woodhouse49593bf2008-08-17 17:08:36 +01003837 return 0;
Chris Mason39544012007-12-12 14:38:19 -05003838 filp->f_pos = 2;
3839 }
David Woodhouse49593bf2008-08-17 17:08:36 +01003840 path = btrfs_alloc_path();
3841 path->reada = 2;
3842
Chris Mason39279cc2007-06-12 06:35:45 -04003843 btrfs_set_key_type(&key, key_type);
3844 key.offset = filp->f_pos;
David Woodhouse49593bf2008-08-17 17:08:36 +01003845 key.objectid = inode->i_ino;
Chris Mason5f39d392007-10-15 16:14:19 -04003846
Chris Mason39279cc2007-06-12 06:35:45 -04003847 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3848 if (ret < 0)
3849 goto err;
3850 advance = 0;
David Woodhouse49593bf2008-08-17 17:08:36 +01003851
3852 while (1) {
Chris Mason5f39d392007-10-15 16:14:19 -04003853 leaf = path->nodes[0];
3854 nritems = btrfs_header_nritems(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04003855 slot = path->slots[0];
3856 if (advance || slot >= nritems) {
David Woodhouse49593bf2008-08-17 17:08:36 +01003857 if (slot >= nritems - 1) {
Chris Mason39279cc2007-06-12 06:35:45 -04003858 ret = btrfs_next_leaf(root, path);
3859 if (ret)
3860 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003861 leaf = path->nodes[0];
3862 nritems = btrfs_header_nritems(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04003863 slot = path->slots[0];
3864 } else {
3865 slot++;
3866 path->slots[0]++;
3867 }
3868 }
Chris Mason3de45862008-11-17 21:02:50 -05003869
Chris Mason39279cc2007-06-12 06:35:45 -04003870 advance = 1;
Chris Mason5f39d392007-10-15 16:14:19 -04003871 item = btrfs_item_nr(leaf, slot);
3872 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3873
3874 if (found_key.objectid != key.objectid)
Chris Mason39279cc2007-06-12 06:35:45 -04003875 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003876 if (btrfs_key_type(&found_key) != key_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003877 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003878 if (found_key.offset < filp->f_pos)
Chris Mason39279cc2007-06-12 06:35:45 -04003879 continue;
Chris Mason5f39d392007-10-15 16:14:19 -04003880
3881 filp->f_pos = found_key.offset;
David Woodhouse49593bf2008-08-17 17:08:36 +01003882
Chris Mason39279cc2007-06-12 06:35:45 -04003883 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3884 di_cur = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003885 di_total = btrfs_item_size(leaf, item);
David Woodhouse49593bf2008-08-17 17:08:36 +01003886
3887 while (di_cur < di_total) {
Chris Mason5f39d392007-10-15 16:14:19 -04003888 struct btrfs_key location;
3889
3890 name_len = btrfs_dir_name_len(leaf, di);
David Woodhouse49593bf2008-08-17 17:08:36 +01003891 if (name_len <= sizeof(tmp_name)) {
Chris Mason5f39d392007-10-15 16:14:19 -04003892 name_ptr = tmp_name;
3893 } else {
3894 name_ptr = kmalloc(name_len, GFP_NOFS);
David Woodhouse49593bf2008-08-17 17:08:36 +01003895 if (!name_ptr) {
3896 ret = -ENOMEM;
3897 goto err;
3898 }
Chris Mason5f39d392007-10-15 16:14:19 -04003899 }
3900 read_extent_buffer(leaf, name_ptr,
3901 (unsigned long)(di + 1), name_len);
3902
3903 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3904 btrfs_dir_item_key_to_cpu(leaf, di, &location);
Chris Mason3de45862008-11-17 21:02:50 -05003905
3906 /* is this a reference to our own snapshot? If so
3907 * skip it
3908 */
3909 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3910 location.objectid == root->root_key.objectid) {
3911 over = 0;
3912 goto skip;
3913 }
Chris Mason5f39d392007-10-15 16:14:19 -04003914 over = filldir(dirent, name_ptr, name_len,
David Woodhouse49593bf2008-08-17 17:08:36 +01003915 found_key.offset, location.objectid,
Chris Mason39279cc2007-06-12 06:35:45 -04003916 d_type);
Chris Mason5f39d392007-10-15 16:14:19 -04003917
Chris Mason3de45862008-11-17 21:02:50 -05003918skip:
Chris Mason5f39d392007-10-15 16:14:19 -04003919 if (name_ptr != tmp_name)
3920 kfree(name_ptr);
3921
Chris Mason39279cc2007-06-12 06:35:45 -04003922 if (over)
3923 goto nopos;
Josef Bacik5103e942007-11-16 11:45:54 -05003924 di_len = btrfs_dir_name_len(leaf, di) +
David Woodhouse49593bf2008-08-17 17:08:36 +01003925 btrfs_dir_data_len(leaf, di) + sizeof(*di);
Chris Mason39279cc2007-06-12 06:35:45 -04003926 di_cur += di_len;
3927 di = (struct btrfs_dir_item *)((char *)di + di_len);
3928 }
3929 }
David Woodhouse49593bf2008-08-17 17:08:36 +01003930
3931 /* Reached end of directory/root. Bump pos past the last item. */
Yan Zheng5e591a02008-02-19 11:41:02 -05003932 if (key_type == BTRFS_DIR_INDEX_KEY)
Jan Engelhardt406266a2009-12-09 22:00:38 +00003933 /*
3934 * 32-bit glibc will use getdents64, but then strtol -
3935 * so the last number we can serve is this.
3936 */
3937 filp->f_pos = 0x7fffffff;
Yan Zheng5e591a02008-02-19 11:41:02 -05003938 else
3939 filp->f_pos++;
Chris Mason39279cc2007-06-12 06:35:45 -04003940nopos:
3941 ret = 0;
3942err:
Chris Mason39279cc2007-06-12 06:35:45 -04003943 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003944 return ret;
3945}
3946
Christoph Hellwiga9185b42010-03-05 09:21:37 +01003947int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Chris Mason39279cc2007-06-12 06:35:45 -04003948{
3949 struct btrfs_root *root = BTRFS_I(inode)->root;
3950 struct btrfs_trans_handle *trans;
3951 int ret = 0;
3952
Yan Zhengc146afa2008-11-12 14:34:12 -05003953 if (root->fs_info->btree_inode == inode)
Chris Mason4ca8b412008-08-05 13:30:48 -04003954 return 0;
3955
Christoph Hellwiga9185b42010-03-05 09:21:37 +01003956 if (wbc->sync_mode == WB_SYNC_ALL) {
Chris Masonf9295742008-07-17 12:54:14 -04003957 trans = btrfs_join_transaction(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04003958 btrfs_set_trans_block_group(trans, inode);
3959 ret = btrfs_commit_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04003960 }
3961 return ret;
3962}
3963
3964/*
Chris Mason54aa1f42007-06-22 14:16:25 -04003965 * This is somewhat expensive, updating the tree every time the
Chris Mason39279cc2007-06-12 06:35:45 -04003966 * inode changes. But, it is most likely to find the inode in cache.
3967 * FIXME, needs more benchmarking...there are no reasons other than performance
3968 * to keep or drop this code.
3969 */
3970void btrfs_dirty_inode(struct inode *inode)
3971{
3972 struct btrfs_root *root = BTRFS_I(inode)->root;
3973 struct btrfs_trans_handle *trans;
3974
Chris Masonf9295742008-07-17 12:54:14 -04003975 trans = btrfs_join_transaction(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04003976 btrfs_set_trans_block_group(trans, inode);
3977 btrfs_update_inode(trans, root, inode);
3978 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04003979}
3980
Chris Masond352ac62008-09-29 15:18:18 -04003981/*
3982 * find the highest existing sequence number in a directory
3983 * and then set the in-memory index_cnt variable to reflect
3984 * free sequence numbers
3985 */
Josef Bacikaec74772008-07-24 12:12:38 -04003986static int btrfs_set_inode_index_count(struct inode *inode)
3987{
3988 struct btrfs_root *root = BTRFS_I(inode)->root;
3989 struct btrfs_key key, found_key;
3990 struct btrfs_path *path;
3991 struct extent_buffer *leaf;
3992 int ret;
3993
3994 key.objectid = inode->i_ino;
3995 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3996 key.offset = (u64)-1;
3997
3998 path = btrfs_alloc_path();
3999 if (!path)
4000 return -ENOMEM;
4001
4002 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4003 if (ret < 0)
4004 goto out;
4005 /* FIXME: we should be able to handle this */
4006 if (ret == 0)
4007 goto out;
4008 ret = 0;
4009
4010 /*
4011 * MAGIC NUMBER EXPLANATION:
4012 * since we search a directory based on f_pos we have to start at 2
4013 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4014 * else has to start at 2
4015 */
4016 if (path->slots[0] == 0) {
4017 BTRFS_I(inode)->index_cnt = 2;
4018 goto out;
4019 }
4020
4021 path->slots[0]--;
4022
4023 leaf = path->nodes[0];
4024 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4025
4026 if (found_key.objectid != inode->i_ino ||
4027 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4028 BTRFS_I(inode)->index_cnt = 2;
4029 goto out;
4030 }
4031
4032 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4033out:
4034 btrfs_free_path(path);
4035 return ret;
4036}
4037
Chris Masond352ac62008-09-29 15:18:18 -04004038/*
4039 * helper to find a free sequence number in a given directory. This current
4040 * code is very simple, later versions will do smarter things in the btree
4041 */
Chris Mason3de45862008-11-17 21:02:50 -05004042int btrfs_set_inode_index(struct inode *dir, u64 *index)
Josef Bacikaec74772008-07-24 12:12:38 -04004043{
4044 int ret = 0;
4045
4046 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4047 ret = btrfs_set_inode_index_count(dir);
Chris Masond3977122009-01-05 21:25:51 -05004048 if (ret)
Josef Bacikaec74772008-07-24 12:12:38 -04004049 return ret;
4050 }
4051
Chris Mason00e4e6b2008-08-05 11:18:09 -04004052 *index = BTRFS_I(dir)->index_cnt;
Josef Bacikaec74772008-07-24 12:12:38 -04004053 BTRFS_I(dir)->index_cnt++;
4054
4055 return ret;
4056}
4057
Chris Mason39279cc2007-06-12 06:35:45 -04004058static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4059 struct btrfs_root *root,
Josef Bacikaec74772008-07-24 12:12:38 -04004060 struct inode *dir,
Chris Mason9c583092008-01-29 15:15:18 -05004061 const char *name, int name_len,
Yan Zhengd2fb3432008-12-11 16:30:39 -05004062 u64 ref_objectid, u64 objectid,
4063 u64 alloc_hint, int mode, u64 *index)
Chris Mason39279cc2007-06-12 06:35:45 -04004064{
4065 struct inode *inode;
Chris Mason5f39d392007-10-15 16:14:19 -04004066 struct btrfs_inode_item *inode_item;
Chris Mason39279cc2007-06-12 06:35:45 -04004067 struct btrfs_key *location;
Chris Mason5f39d392007-10-15 16:14:19 -04004068 struct btrfs_path *path;
Chris Mason9c583092008-01-29 15:15:18 -05004069 struct btrfs_inode_ref *ref;
4070 struct btrfs_key key[2];
4071 u32 sizes[2];
4072 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04004073 int ret;
4074 int owner;
4075
Chris Mason5f39d392007-10-15 16:14:19 -04004076 path = btrfs_alloc_path();
4077 BUG_ON(!path);
4078
Chris Mason39279cc2007-06-12 06:35:45 -04004079 inode = new_inode(root->fs_info->sb);
4080 if (!inode)
4081 return ERR_PTR(-ENOMEM);
4082
Josef Bacikaec74772008-07-24 12:12:38 -04004083 if (dir) {
Chris Mason3de45862008-11-17 21:02:50 -05004084 ret = btrfs_set_inode_index(dir, index);
Shen Feng09771432009-04-02 16:46:06 -04004085 if (ret) {
4086 iput(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004087 return ERR_PTR(ret);
Shen Feng09771432009-04-02 16:46:06 -04004088 }
Josef Bacikaec74772008-07-24 12:12:38 -04004089 }
4090 /*
4091 * index_cnt is ignored for everything but a dir,
4092 * btrfs_get_inode_index_count has an explanation for the magic
4093 * number
4094 */
Chris Masone02119d2008-09-05 16:13:11 -04004095 init_btrfs_i(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004096 BTRFS_I(inode)->index_cnt = 2;
Chris Mason39279cc2007-06-12 06:35:45 -04004097 BTRFS_I(inode)->root = root;
Chris Masone02119d2008-09-05 16:13:11 -04004098 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik6a632092009-02-20 11:00:09 -05004099 btrfs_set_inode_space_info(root, inode);
Chris Masonb888db22007-08-27 16:49:44 -04004100
Chris Mason39279cc2007-06-12 06:35:45 -04004101 if (mode & S_IFDIR)
4102 owner = 0;
4103 else
4104 owner = 1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05004105 BTRFS_I(inode)->block_group =
4106 btrfs_find_block_group(root, 0, alloc_hint, owner);
Chris Mason9c583092008-01-29 15:15:18 -05004107
4108 key[0].objectid = objectid;
4109 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4110 key[0].offset = 0;
4111
4112 key[1].objectid = objectid;
4113 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4114 key[1].offset = ref_objectid;
4115
4116 sizes[0] = sizeof(struct btrfs_inode_item);
4117 sizes[1] = name_len + sizeof(*ref);
4118
Chris Masonb9473432009-03-13 11:00:37 -04004119 path->leave_spinning = 1;
Chris Mason9c583092008-01-29 15:15:18 -05004120 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4121 if (ret != 0)
Chris Mason5f39d392007-10-15 16:14:19 -04004122 goto fail;
4123
Chris Mason79683f22008-11-19 22:00:53 -05004124 inode->i_uid = current_fsuid();
Chris Ball8c087b52009-02-04 09:29:54 -05004125
Chris Mason42f15d72009-02-06 11:35:57 -05004126 if (dir && (dir->i_mode & S_ISGID)) {
Chris Ball8c087b52009-02-04 09:29:54 -05004127 inode->i_gid = dir->i_gid;
4128 if (S_ISDIR(mode))
4129 mode |= S_ISGID;
4130 } else
4131 inode->i_gid = current_fsgid();
4132
Chris Mason39279cc2007-06-12 06:35:45 -04004133 inode->i_mode = mode;
4134 inode->i_ino = objectid;
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004135 inode_set_bytes(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004136 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
Chris Mason5f39d392007-10-15 16:14:19 -04004137 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4138 struct btrfs_inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04004139 fill_inode_item(trans, path->nodes[0], inode_item, inode);
Chris Mason9c583092008-01-29 15:15:18 -05004140
4141 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4142 struct btrfs_inode_ref);
4143 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
Chris Mason00e4e6b2008-08-05 11:18:09 -04004144 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
Chris Mason9c583092008-01-29 15:15:18 -05004145 ptr = (unsigned long)(ref + 1);
4146 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4147
Chris Mason5f39d392007-10-15 16:14:19 -04004148 btrfs_mark_buffer_dirty(path->nodes[0]);
4149 btrfs_free_path(path);
4150
Chris Mason39279cc2007-06-12 06:35:45 -04004151 location = &BTRFS_I(inode)->location;
4152 location->objectid = objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04004153 location->offset = 0;
4154 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4155
Christoph Hellwig6cbff002009-04-17 10:37:41 +02004156 btrfs_inherit_iflags(inode, dir);
4157
Chris Mason94272162009-07-02 12:26:06 -04004158 if ((mode & S_IFREG)) {
4159 if (btrfs_test_opt(root, NODATASUM))
4160 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4161 if (btrfs_test_opt(root, NODATACOW))
4162 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4163 }
4164
Chris Mason39279cc2007-06-12 06:35:45 -04004165 insert_inode_hash(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004166 inode_tree_add(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004167 return inode;
Chris Mason5f39d392007-10-15 16:14:19 -04004168fail:
Josef Bacikaec74772008-07-24 12:12:38 -04004169 if (dir)
4170 BTRFS_I(dir)->index_cnt--;
Chris Mason5f39d392007-10-15 16:14:19 -04004171 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04004172 iput(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004173 return ERR_PTR(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04004174}
4175
4176static inline u8 btrfs_inode_type(struct inode *inode)
4177{
4178 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4179}
4180
Chris Masond352ac62008-09-29 15:18:18 -04004181/*
4182 * utility function to add 'inode' into 'parent_inode' with
4183 * a give name and a given sequence number.
4184 * if 'add_backref' is true, also insert a backref from the
4185 * inode to the parent directory.
4186 */
Chris Masone02119d2008-09-05 16:13:11 -04004187int btrfs_add_link(struct btrfs_trans_handle *trans,
4188 struct inode *parent_inode, struct inode *inode,
4189 const char *name, int name_len, int add_backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04004190{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004191 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004192 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04004193 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
Chris Mason5f39d392007-10-15 16:14:19 -04004194
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004195 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4196 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4197 } else {
4198 key.objectid = inode->i_ino;
4199 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4200 key.offset = 0;
4201 }
Chris Mason39279cc2007-06-12 06:35:45 -04004202
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004203 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4204 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4205 key.objectid, root->root_key.objectid,
4206 parent_inode->i_ino,
4207 index, name, name_len);
4208 } else if (add_backref) {
4209 ret = btrfs_insert_inode_ref(trans, root,
4210 name, name_len, inode->i_ino,
4211 parent_inode->i_ino, index);
4212 }
4213
Chris Mason39279cc2007-06-12 06:35:45 -04004214 if (ret == 0) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004215 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4216 parent_inode->i_ino, &key,
4217 btrfs_inode_type(inode), index);
4218 BUG_ON(ret);
4219
Chris Masondbe674a2008-07-17 12:54:05 -04004220 btrfs_i_size_write(parent_inode, parent_inode->i_size +
Chris Masone02119d2008-09-05 16:13:11 -04004221 name_len * 2);
Chris Mason79c44582007-06-25 10:09:33 -04004222 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
Chris Masone02119d2008-09-05 16:13:11 -04004223 ret = btrfs_update_inode(trans, root, parent_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004224 }
4225 return ret;
4226}
4227
4228static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
Chris Mason9c583092008-01-29 15:15:18 -05004229 struct dentry *dentry, struct inode *inode,
Chris Mason00e4e6b2008-08-05 11:18:09 -04004230 int backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04004231{
Chris Masone02119d2008-09-05 16:13:11 -04004232 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4233 inode, dentry->d_name.name,
4234 dentry->d_name.len, backref, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004235 if (!err) {
4236 d_instantiate(dentry, inode);
4237 return 0;
4238 }
4239 if (err > 0)
4240 err = -EEXIST;
4241 return err;
4242}
4243
Josef Bacik618e21d2007-07-11 10:18:17 -04004244static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4245 int mode, dev_t rdev)
4246{
4247 struct btrfs_trans_handle *trans;
4248 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05004249 struct inode *inode = NULL;
Josef Bacik618e21d2007-07-11 10:18:17 -04004250 int err;
4251 int drop_inode = 0;
4252 u64 objectid;
Chris Mason1832a6d2007-12-21 16:27:21 -05004253 unsigned long nr = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004254 u64 index = 0;
Josef Bacik618e21d2007-07-11 10:18:17 -04004255
4256 if (!new_valid_dev(rdev))
4257 return -EINVAL;
4258
Josef Bacik9ed74f22009-09-11 16:12:44 -04004259 /*
4260 * 2 for inode item and ref
4261 * 2 for dir items
4262 * 1 for xattr if selinux is on
4263 */
4264 err = btrfs_reserve_metadata_space(root, 5);
Chris Mason1832a6d2007-12-21 16:27:21 -05004265 if (err)
Josef Bacik9ed74f22009-09-11 16:12:44 -04004266 return err;
Chris Mason1832a6d2007-12-21 16:27:21 -05004267
Josef Bacik618e21d2007-07-11 10:18:17 -04004268 trans = btrfs_start_transaction(root, 1);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004269 if (!trans)
4270 goto fail;
Josef Bacik618e21d2007-07-11 10:18:17 -04004271 btrfs_set_trans_block_group(trans, dir);
4272
4273 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4274 if (err) {
4275 err = -ENOSPC;
4276 goto out_unlock;
4277 }
4278
Josef Bacikaec74772008-07-24 12:12:38 -04004279 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Chris Mason9c583092008-01-29 15:15:18 -05004280 dentry->d_name.len,
4281 dentry->d_parent->d_inode->i_ino, objectid,
Chris Mason00e4e6b2008-08-05 11:18:09 -04004282 BTRFS_I(dir)->block_group, mode, &index);
Josef Bacik618e21d2007-07-11 10:18:17 -04004283 err = PTR_ERR(inode);
4284 if (IS_ERR(inode))
4285 goto out_unlock;
4286
Yan, Zhengf34f57a2009-11-12 09:35:27 +00004287 err = btrfs_init_inode_security(trans, inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04004288 if (err) {
4289 drop_inode = 1;
4290 goto out_unlock;
4291 }
4292
Josef Bacik618e21d2007-07-11 10:18:17 -04004293 btrfs_set_trans_block_group(trans, inode);
Chris Mason00e4e6b2008-08-05 11:18:09 -04004294 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
Josef Bacik618e21d2007-07-11 10:18:17 -04004295 if (err)
4296 drop_inode = 1;
4297 else {
4298 inode->i_op = &btrfs_special_inode_operations;
4299 init_special_inode(inode, inode->i_mode, rdev);
Yan1b4ab1b2007-08-29 09:11:44 -04004300 btrfs_update_inode(trans, root, inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04004301 }
Josef Bacik618e21d2007-07-11 10:18:17 -04004302 btrfs_update_inode_block_group(trans, inode);
4303 btrfs_update_inode_block_group(trans, dir);
4304out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004305 nr = trans->blocks_used;
Chris Mason89ce8a62008-06-25 16:01:31 -04004306 btrfs_end_transaction_throttle(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05004307fail:
Josef Bacik9ed74f22009-09-11 16:12:44 -04004308 btrfs_unreserve_metadata_space(root, 5);
Josef Bacik618e21d2007-07-11 10:18:17 -04004309 if (drop_inode) {
4310 inode_dec_link_count(inode);
4311 iput(inode);
4312 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004313 btrfs_btree_balance_dirty(root, nr);
Josef Bacik618e21d2007-07-11 10:18:17 -04004314 return err;
4315}
4316
Chris Mason39279cc2007-06-12 06:35:45 -04004317static int btrfs_create(struct inode *dir, struct dentry *dentry,
4318 int mode, struct nameidata *nd)
4319{
4320 struct btrfs_trans_handle *trans;
4321 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05004322 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04004323 int err;
4324 int drop_inode = 0;
Chris Mason1832a6d2007-12-21 16:27:21 -05004325 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004326 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004327 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004328
Josef Bacik9ed74f22009-09-11 16:12:44 -04004329 /*
4330 * 2 for inode item and ref
4331 * 2 for dir items
4332 * 1 for xattr if selinux is on
4333 */
4334 err = btrfs_reserve_metadata_space(root, 5);
Chris Mason1832a6d2007-12-21 16:27:21 -05004335 if (err)
Josef Bacik9ed74f22009-09-11 16:12:44 -04004336 return err;
4337
Chris Mason39279cc2007-06-12 06:35:45 -04004338 trans = btrfs_start_transaction(root, 1);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004339 if (!trans)
4340 goto fail;
Chris Mason39279cc2007-06-12 06:35:45 -04004341 btrfs_set_trans_block_group(trans, dir);
4342
4343 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4344 if (err) {
4345 err = -ENOSPC;
4346 goto out_unlock;
4347 }
4348
Josef Bacikaec74772008-07-24 12:12:38 -04004349 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Chris Mason9c583092008-01-29 15:15:18 -05004350 dentry->d_name.len,
4351 dentry->d_parent->d_inode->i_ino,
Chris Mason00e4e6b2008-08-05 11:18:09 -04004352 objectid, BTRFS_I(dir)->block_group, mode,
4353 &index);
Chris Mason39279cc2007-06-12 06:35:45 -04004354 err = PTR_ERR(inode);
4355 if (IS_ERR(inode))
4356 goto out_unlock;
4357
Yan, Zhengf34f57a2009-11-12 09:35:27 +00004358 err = btrfs_init_inode_security(trans, inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04004359 if (err) {
4360 drop_inode = 1;
4361 goto out_unlock;
4362 }
4363
Chris Mason39279cc2007-06-12 06:35:45 -04004364 btrfs_set_trans_block_group(trans, inode);
Chris Mason00e4e6b2008-08-05 11:18:09 -04004365 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004366 if (err)
4367 drop_inode = 1;
4368 else {
4369 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04004370 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04004371 inode->i_fop = &btrfs_file_operations;
4372 inode->i_op = &btrfs_file_inode_operations;
Chris Masond1310b22008-01-24 16:13:08 -05004373 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04004374 }
Chris Mason39279cc2007-06-12 06:35:45 -04004375 btrfs_update_inode_block_group(trans, inode);
4376 btrfs_update_inode_block_group(trans, dir);
4377out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004378 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04004379 btrfs_end_transaction_throttle(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05004380fail:
Josef Bacik9ed74f22009-09-11 16:12:44 -04004381 btrfs_unreserve_metadata_space(root, 5);
Chris Mason39279cc2007-06-12 06:35:45 -04004382 if (drop_inode) {
4383 inode_dec_link_count(inode);
4384 iput(inode);
4385 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004386 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004387 return err;
4388}
4389
4390static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4391 struct dentry *dentry)
4392{
4393 struct btrfs_trans_handle *trans;
4394 struct btrfs_root *root = BTRFS_I(dir)->root;
4395 struct inode *inode = old_dentry->d_inode;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004396 u64 index;
Chris Mason1832a6d2007-12-21 16:27:21 -05004397 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004398 int err;
4399 int drop_inode = 0;
4400
4401 if (inode->i_nlink == 0)
4402 return -ENOENT;
4403
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00004404 /* do not allow sys_link's with other subvols of the same device */
4405 if (root->objectid != BTRFS_I(inode)->root->objectid)
4406 return -EPERM;
4407
Josef Bacik9ed74f22009-09-11 16:12:44 -04004408 /*
4409 * 1 item for inode ref
4410 * 2 items for dir items
4411 */
4412 err = btrfs_reserve_metadata_space(root, 3);
Chris Mason1832a6d2007-12-21 16:27:21 -05004413 if (err)
Josef Bacik9ed74f22009-09-11 16:12:44 -04004414 return err;
4415
4416 btrfs_inc_nlink(inode);
4417
Chris Mason3de45862008-11-17 21:02:50 -05004418 err = btrfs_set_inode_index(dir, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04004419 if (err)
4420 goto fail;
4421
Chris Mason39279cc2007-06-12 06:35:45 -04004422 trans = btrfs_start_transaction(root, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04004423
Chris Mason39279cc2007-06-12 06:35:45 -04004424 btrfs_set_trans_block_group(trans, dir);
4425 atomic_inc(&inode->i_count);
Josef Bacikaec74772008-07-24 12:12:38 -04004426
Chris Mason00e4e6b2008-08-05 11:18:09 -04004427 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
Chris Mason5f39d392007-10-15 16:14:19 -04004428
Yan, Zhenga5719522009-09-24 09:17:31 -04004429 if (err) {
Chris Mason39279cc2007-06-12 06:35:45 -04004430 drop_inode = 1;
Yan, Zhenga5719522009-09-24 09:17:31 -04004431 } else {
4432 btrfs_update_inode_block_group(trans, dir);
4433 err = btrfs_update_inode(trans, root, inode);
4434 BUG_ON(err);
4435 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4436 }
Chris Mason39279cc2007-06-12 06:35:45 -04004437
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004438 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04004439 btrfs_end_transaction_throttle(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05004440fail:
Josef Bacik9ed74f22009-09-11 16:12:44 -04004441 btrfs_unreserve_metadata_space(root, 3);
Chris Mason39279cc2007-06-12 06:35:45 -04004442 if (drop_inode) {
4443 inode_dec_link_count(inode);
4444 iput(inode);
4445 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004446 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004447 return err;
4448}
4449
Chris Mason39279cc2007-06-12 06:35:45 -04004450static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4451{
Chris Masonb9d86662008-05-02 16:13:49 -04004452 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04004453 struct btrfs_trans_handle *trans;
4454 struct btrfs_root *root = BTRFS_I(dir)->root;
4455 int err = 0;
4456 int drop_on_err = 0;
Chris Masonb9d86662008-05-02 16:13:49 -04004457 u64 objectid = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004458 u64 index = 0;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004459 unsigned long nr = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04004460
Josef Bacik9ed74f22009-09-11 16:12:44 -04004461 /*
4462 * 2 items for inode and ref
4463 * 2 items for dir items
4464 * 1 for xattr if selinux is on
4465 */
4466 err = btrfs_reserve_metadata_space(root, 5);
Chris Mason1832a6d2007-12-21 16:27:21 -05004467 if (err)
Josef Bacik9ed74f22009-09-11 16:12:44 -04004468 return err;
Chris Mason1832a6d2007-12-21 16:27:21 -05004469
Chris Mason39279cc2007-06-12 06:35:45 -04004470 trans = btrfs_start_transaction(root, 1);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004471 if (!trans) {
4472 err = -ENOMEM;
Chris Mason39279cc2007-06-12 06:35:45 -04004473 goto out_unlock;
4474 }
Josef Bacik9ed74f22009-09-11 16:12:44 -04004475 btrfs_set_trans_block_group(trans, dir);
Chris Mason39279cc2007-06-12 06:35:45 -04004476
4477 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4478 if (err) {
4479 err = -ENOSPC;
Miao Xie0be2e982010-02-11 08:06:58 +00004480 goto out_fail;
Chris Mason39279cc2007-06-12 06:35:45 -04004481 }
4482
Josef Bacikaec74772008-07-24 12:12:38 -04004483 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Chris Mason9c583092008-01-29 15:15:18 -05004484 dentry->d_name.len,
4485 dentry->d_parent->d_inode->i_ino, objectid,
Chris Mason00e4e6b2008-08-05 11:18:09 -04004486 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4487 &index);
Chris Mason39279cc2007-06-12 06:35:45 -04004488 if (IS_ERR(inode)) {
4489 err = PTR_ERR(inode);
4490 goto out_fail;
4491 }
Chris Mason5f39d392007-10-15 16:14:19 -04004492
Chris Mason39279cc2007-06-12 06:35:45 -04004493 drop_on_err = 1;
Josef Bacik33268ea2008-07-24 12:16:36 -04004494
Yan, Zhengf34f57a2009-11-12 09:35:27 +00004495 err = btrfs_init_inode_security(trans, inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04004496 if (err)
4497 goto out_fail;
4498
Chris Mason39279cc2007-06-12 06:35:45 -04004499 inode->i_op = &btrfs_dir_inode_operations;
4500 inode->i_fop = &btrfs_dir_file_operations;
4501 btrfs_set_trans_block_group(trans, inode);
4502
Chris Masondbe674a2008-07-17 12:54:05 -04004503 btrfs_i_size_write(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004504 err = btrfs_update_inode(trans, root, inode);
4505 if (err)
4506 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04004507
Chris Masone02119d2008-09-05 16:13:11 -04004508 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4509 inode, dentry->d_name.name,
4510 dentry->d_name.len, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004511 if (err)
4512 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04004513
Chris Mason39279cc2007-06-12 06:35:45 -04004514 d_instantiate(dentry, inode);
4515 drop_on_err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004516 btrfs_update_inode_block_group(trans, inode);
4517 btrfs_update_inode_block_group(trans, dir);
4518
4519out_fail:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004520 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04004521 btrfs_end_transaction_throttle(trans, root);
Chris Mason5f39d392007-10-15 16:14:19 -04004522
Chris Mason39279cc2007-06-12 06:35:45 -04004523out_unlock:
Josef Bacik9ed74f22009-09-11 16:12:44 -04004524 btrfs_unreserve_metadata_space(root, 5);
Chris Mason39279cc2007-06-12 06:35:45 -04004525 if (drop_on_err)
4526 iput(inode);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004527 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004528 return err;
4529}
4530
Chris Masond352ac62008-09-29 15:18:18 -04004531/* helper for btfs_get_extent. Given an existing extent in the tree,
4532 * and an extent that you want to insert, deal with overlap and insert
4533 * the new extent into the tree.
4534 */
Chris Mason3b951512008-04-17 11:29:12 -04004535static int merge_extent_mapping(struct extent_map_tree *em_tree,
4536 struct extent_map *existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04004537 struct extent_map *em,
4538 u64 map_start, u64 map_len)
Chris Mason3b951512008-04-17 11:29:12 -04004539{
4540 u64 start_diff;
Chris Mason3b951512008-04-17 11:29:12 -04004541
Chris Masone6dcd2d2008-07-17 12:53:50 -04004542 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4543 start_diff = map_start - em->start;
4544 em->start = map_start;
4545 em->len = map_len;
Chris Masonc8b97812008-10-29 14:49:59 -04004546 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4547 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04004548 em->block_start += start_diff;
Chris Masonc8b97812008-10-29 14:49:59 -04004549 em->block_len -= start_diff;
4550 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004551 return add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04004552}
4553
Chris Masonc8b97812008-10-29 14:49:59 -04004554static noinline int uncompress_inline(struct btrfs_path *path,
4555 struct inode *inode, struct page *page,
4556 size_t pg_offset, u64 extent_offset,
4557 struct btrfs_file_extent_item *item)
4558{
4559 int ret;
4560 struct extent_buffer *leaf = path->nodes[0];
4561 char *tmp;
4562 size_t max_size;
4563 unsigned long inline_size;
4564 unsigned long ptr;
4565
4566 WARN_ON(pg_offset != 0);
4567 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4568 inline_size = btrfs_file_extent_inline_item_len(leaf,
4569 btrfs_item_nr(leaf, path->slots[0]));
4570 tmp = kmalloc(inline_size, GFP_NOFS);
4571 ptr = btrfs_file_extent_inline_start(item);
4572
4573 read_extent_buffer(leaf, tmp, ptr, inline_size);
4574
Chris Mason5b050f02008-11-11 09:34:41 -05004575 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
Chris Masonc8b97812008-10-29 14:49:59 -04004576 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4577 inline_size, max_size);
4578 if (ret) {
4579 char *kaddr = kmap_atomic(page, KM_USER0);
4580 unsigned long copy_size = min_t(u64,
4581 PAGE_CACHE_SIZE - pg_offset,
4582 max_size - extent_offset);
4583 memset(kaddr + pg_offset, 0, copy_size);
4584 kunmap_atomic(kaddr, KM_USER0);
4585 }
4586 kfree(tmp);
4587 return 0;
4588}
4589
Chris Masond352ac62008-09-29 15:18:18 -04004590/*
4591 * a bit scary, this does extent mapping from logical file offset to the disk.
Chris Masond3977122009-01-05 21:25:51 -05004592 * the ugly parts come from merging extents from the disk with the in-ram
4593 * representation. This gets more complex because of the data=ordered code,
Chris Masond352ac62008-09-29 15:18:18 -04004594 * where the in-ram extents might be locked pending data=ordered completion.
4595 *
4596 * This also copies inline extents directly into the page.
4597 */
Chris Masond3977122009-01-05 21:25:51 -05004598
Chris Masona52d9a82007-08-27 16:49:44 -04004599struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
Chris Mason70dec802008-01-29 09:59:12 -05004600 size_t pg_offset, u64 start, u64 len,
Chris Masona52d9a82007-08-27 16:49:44 -04004601 int create)
4602{
4603 int ret;
4604 int err = 0;
Chris Masondb945352007-10-15 16:15:53 -04004605 u64 bytenr;
Chris Masona52d9a82007-08-27 16:49:44 -04004606 u64 extent_start = 0;
4607 u64 extent_end = 0;
4608 u64 objectid = inode->i_ino;
4609 u32 found_type;
Chris Masonf4219502008-07-22 11:18:09 -04004610 struct btrfs_path *path = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -04004611 struct btrfs_root *root = BTRFS_I(inode)->root;
4612 struct btrfs_file_extent_item *item;
Chris Mason5f39d392007-10-15 16:14:19 -04004613 struct extent_buffer *leaf;
4614 struct btrfs_key found_key;
Chris Masona52d9a82007-08-27 16:49:44 -04004615 struct extent_map *em = NULL;
4616 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Masond1310b22008-01-24 16:13:08 -05004617 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04004618 struct btrfs_trans_handle *trans = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -04004619 int compressed;
Chris Masona52d9a82007-08-27 16:49:44 -04004620
Chris Masona52d9a82007-08-27 16:49:44 -04004621again:
Chris Mason890871b2009-09-02 16:24:52 -04004622 read_lock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05004623 em = lookup_extent_mapping(em_tree, start, len);
Chris Masona061fc82008-05-07 11:43:44 -04004624 if (em)
4625 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason890871b2009-09-02 16:24:52 -04004626 read_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05004627
Chris Masona52d9a82007-08-27 16:49:44 -04004628 if (em) {
Chris Masone1c4b742008-04-22 13:26:46 -04004629 if (em->start > start || em->start + em->len <= start)
4630 free_extent_map(em);
4631 else if (em->block_start == EXTENT_MAP_INLINE && page)
Chris Mason70dec802008-01-29 09:59:12 -05004632 free_extent_map(em);
4633 else
4634 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004635 }
Chris Masond1310b22008-01-24 16:13:08 -05004636 em = alloc_extent_map(GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04004637 if (!em) {
Chris Masond1310b22008-01-24 16:13:08 -05004638 err = -ENOMEM;
4639 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004640 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004641 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Masond1310b22008-01-24 16:13:08 -05004642 em->start = EXTENT_MAP_HOLE;
Chris Mason445a6942008-11-10 11:53:33 -05004643 em->orig_start = EXTENT_MAP_HOLE;
Chris Masond1310b22008-01-24 16:13:08 -05004644 em->len = (u64)-1;
Chris Masonc8b97812008-10-29 14:49:59 -04004645 em->block_len = (u64)-1;
Chris Masonf4219502008-07-22 11:18:09 -04004646
4647 if (!path) {
4648 path = btrfs_alloc_path();
4649 BUG_ON(!path);
4650 }
4651
Chris Mason179e29e2007-11-01 11:28:41 -04004652 ret = btrfs_lookup_file_extent(trans, root, path,
4653 objectid, start, trans != NULL);
Chris Masona52d9a82007-08-27 16:49:44 -04004654 if (ret < 0) {
4655 err = ret;
4656 goto out;
4657 }
4658
4659 if (ret != 0) {
4660 if (path->slots[0] == 0)
4661 goto not_found;
4662 path->slots[0]--;
4663 }
4664
Chris Mason5f39d392007-10-15 16:14:19 -04004665 leaf = path->nodes[0];
4666 item = btrfs_item_ptr(leaf, path->slots[0],
Chris Masona52d9a82007-08-27 16:49:44 -04004667 struct btrfs_file_extent_item);
Chris Masona52d9a82007-08-27 16:49:44 -04004668 /* are we inside the extent that was found? */
Chris Mason5f39d392007-10-15 16:14:19 -04004669 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4670 found_type = btrfs_key_type(&found_key);
4671 if (found_key.objectid != objectid ||
Chris Masona52d9a82007-08-27 16:49:44 -04004672 found_type != BTRFS_EXTENT_DATA_KEY) {
4673 goto not_found;
4674 }
4675
Chris Mason5f39d392007-10-15 16:14:19 -04004676 found_type = btrfs_file_extent_type(leaf, item);
4677 extent_start = found_key.offset;
Chris Masonc8b97812008-10-29 14:49:59 -04004678 compressed = btrfs_file_extent_compression(leaf, item);
Yan Zhengd899e052008-10-30 14:25:28 -04004679 if (found_type == BTRFS_FILE_EXTENT_REG ||
4680 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04004681 extent_end = extent_start +
Chris Masondb945352007-10-15 16:15:53 -04004682 btrfs_file_extent_num_bytes(leaf, item);
Yan Zheng9036c102008-10-30 14:19:41 -04004683 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4684 size_t size;
4685 size = btrfs_file_extent_inline_len(leaf, item);
4686 extent_end = (extent_start + size + root->sectorsize - 1) &
4687 ~((u64)root->sectorsize - 1);
4688 }
4689
4690 if (start >= extent_end) {
4691 path->slots[0]++;
4692 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4693 ret = btrfs_next_leaf(root, path);
4694 if (ret < 0) {
4695 err = ret;
4696 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004697 }
Yan Zheng9036c102008-10-30 14:19:41 -04004698 if (ret > 0)
4699 goto not_found;
4700 leaf = path->nodes[0];
Chris Masona52d9a82007-08-27 16:49:44 -04004701 }
Yan Zheng9036c102008-10-30 14:19:41 -04004702 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4703 if (found_key.objectid != objectid ||
4704 found_key.type != BTRFS_EXTENT_DATA_KEY)
4705 goto not_found;
4706 if (start + len <= found_key.offset)
4707 goto not_found;
4708 em->start = start;
4709 em->len = found_key.offset - start;
4710 goto not_found_em;
4711 }
4712
Yan Zhengd899e052008-10-30 14:25:28 -04004713 if (found_type == BTRFS_FILE_EXTENT_REG ||
4714 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng9036c102008-10-30 14:19:41 -04004715 em->start = extent_start;
4716 em->len = extent_end - extent_start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -05004717 em->orig_start = extent_start -
4718 btrfs_file_extent_offset(leaf, item);
Chris Masondb945352007-10-15 16:15:53 -04004719 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4720 if (bytenr == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04004721 em->block_start = EXTENT_MAP_HOLE;
Chris Masona52d9a82007-08-27 16:49:44 -04004722 goto insert;
4723 }
Chris Masonc8b97812008-10-29 14:49:59 -04004724 if (compressed) {
4725 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4726 em->block_start = bytenr;
4727 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4728 item);
4729 } else {
4730 bytenr += btrfs_file_extent_offset(leaf, item);
4731 em->block_start = bytenr;
4732 em->block_len = em->len;
Yan Zhengd899e052008-10-30 14:25:28 -04004733 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4734 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -04004735 }
Chris Masona52d9a82007-08-27 16:49:44 -04004736 goto insert;
4737 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04004738 unsigned long ptr;
Chris Masona52d9a82007-08-27 16:49:44 -04004739 char *map;
Chris Mason3326d1b2007-10-15 16:18:25 -04004740 size_t size;
4741 size_t extent_offset;
4742 size_t copy_size;
Chris Masona52d9a82007-08-27 16:49:44 -04004743
Chris Masona52d9a82007-08-27 16:49:44 -04004744 em->block_start = EXTENT_MAP_INLINE;
Chris Masonc8b97812008-10-29 14:49:59 -04004745 if (!page || create) {
Yan689f9342007-10-29 11:41:07 -04004746 em->start = extent_start;
Yan Zheng9036c102008-10-30 14:19:41 -04004747 em->len = extent_end - extent_start;
Yan689f9342007-10-29 11:41:07 -04004748 goto out;
4749 }
4750
Yan Zheng9036c102008-10-30 14:19:41 -04004751 size = btrfs_file_extent_inline_len(leaf, item);
4752 extent_offset = page_offset(page) + pg_offset - extent_start;
Chris Mason70dec802008-01-29 09:59:12 -05004753 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
Yan689f9342007-10-29 11:41:07 -04004754 size - extent_offset);
Chris Mason3326d1b2007-10-15 16:18:25 -04004755 em->start = extent_start + extent_offset;
Chris Mason70dec802008-01-29 09:59:12 -05004756 em->len = (copy_size + root->sectorsize - 1) &
4757 ~((u64)root->sectorsize - 1);
Yan Zhengff5b7ee2008-11-10 07:34:43 -05004758 em->orig_start = EXTENT_MAP_INLINE;
Chris Masonc8b97812008-10-29 14:49:59 -04004759 if (compressed)
4760 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Yan689f9342007-10-29 11:41:07 -04004761 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
Chris Mason179e29e2007-11-01 11:28:41 -04004762 if (create == 0 && !PageUptodate(page)) {
Chris Masonc8b97812008-10-29 14:49:59 -04004763 if (btrfs_file_extent_compression(leaf, item) ==
4764 BTRFS_COMPRESS_ZLIB) {
4765 ret = uncompress_inline(path, inode, page,
4766 pg_offset,
4767 extent_offset, item);
4768 BUG_ON(ret);
4769 } else {
4770 map = kmap(page);
4771 read_extent_buffer(leaf, map + pg_offset, ptr,
4772 copy_size);
Chris Mason93c82d52009-09-11 12:36:29 -04004773 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4774 memset(map + pg_offset + copy_size, 0,
4775 PAGE_CACHE_SIZE - pg_offset -
4776 copy_size);
4777 }
Chris Masonc8b97812008-10-29 14:49:59 -04004778 kunmap(page);
4779 }
Chris Mason179e29e2007-11-01 11:28:41 -04004780 flush_dcache_page(page);
4781 } else if (create && PageUptodate(page)) {
4782 if (!trans) {
4783 kunmap(page);
4784 free_extent_map(em);
4785 em = NULL;
4786 btrfs_release_path(root, path);
Chris Masonf9295742008-07-17 12:54:14 -04004787 trans = btrfs_join_transaction(root, 1);
Chris Mason179e29e2007-11-01 11:28:41 -04004788 goto again;
4789 }
Chris Masonc8b97812008-10-29 14:49:59 -04004790 map = kmap(page);
Chris Mason70dec802008-01-29 09:59:12 -05004791 write_extent_buffer(leaf, map + pg_offset, ptr,
Chris Mason179e29e2007-11-01 11:28:41 -04004792 copy_size);
Chris Masonc8b97812008-10-29 14:49:59 -04004793 kunmap(page);
Chris Mason179e29e2007-11-01 11:28:41 -04004794 btrfs_mark_buffer_dirty(leaf);
Chris Masona52d9a82007-08-27 16:49:44 -04004795 }
Chris Masond1310b22008-01-24 16:13:08 -05004796 set_extent_uptodate(io_tree, em->start,
4797 extent_map_end(em) - 1, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04004798 goto insert;
4799 } else {
Chris Masond3977122009-01-05 21:25:51 -05004800 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
Chris Masona52d9a82007-08-27 16:49:44 -04004801 WARN_ON(1);
4802 }
4803not_found:
4804 em->start = start;
Chris Masond1310b22008-01-24 16:13:08 -05004805 em->len = len;
Chris Masona52d9a82007-08-27 16:49:44 -04004806not_found_em:
Chris Mason5f39d392007-10-15 16:14:19 -04004807 em->block_start = EXTENT_MAP_HOLE;
Yan Zheng9036c102008-10-30 14:19:41 -04004808 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -04004809insert:
4810 btrfs_release_path(root, path);
Chris Masond1310b22008-01-24 16:13:08 -05004811 if (em->start > start || extent_map_end(em) <= start) {
Chris Masond3977122009-01-05 21:25:51 -05004812 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4813 "[%llu %llu]\n", (unsigned long long)em->start,
4814 (unsigned long long)em->len,
4815 (unsigned long long)start,
4816 (unsigned long long)len);
Chris Masona52d9a82007-08-27 16:49:44 -04004817 err = -EIO;
4818 goto out;
4819 }
Chris Masond1310b22008-01-24 16:13:08 -05004820
4821 err = 0;
Chris Mason890871b2009-09-02 16:24:52 -04004822 write_lock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04004823 ret = add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04004824 /* it is possible that someone inserted the extent into the tree
4825 * while we had the lock dropped. It is also possible that
4826 * an overlapping map exists in the tree
4827 */
Chris Masona52d9a82007-08-27 16:49:44 -04004828 if (ret == -EEXIST) {
Chris Mason3b951512008-04-17 11:29:12 -04004829 struct extent_map *existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004830
4831 ret = 0;
4832
Chris Mason3b951512008-04-17 11:29:12 -04004833 existing = lookup_extent_mapping(em_tree, start, len);
Chris Masone1c4b742008-04-22 13:26:46 -04004834 if (existing && (existing->start > start ||
4835 existing->start + existing->len <= start)) {
4836 free_extent_map(existing);
4837 existing = NULL;
4838 }
Chris Mason3b951512008-04-17 11:29:12 -04004839 if (!existing) {
4840 existing = lookup_extent_mapping(em_tree, em->start,
4841 em->len);
4842 if (existing) {
4843 err = merge_extent_mapping(em_tree, existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04004844 em, start,
4845 root->sectorsize);
Chris Mason3b951512008-04-17 11:29:12 -04004846 free_extent_map(existing);
4847 if (err) {
4848 free_extent_map(em);
4849 em = NULL;
4850 }
4851 } else {
4852 err = -EIO;
Chris Mason3b951512008-04-17 11:29:12 -04004853 free_extent_map(em);
4854 em = NULL;
4855 }
4856 } else {
4857 free_extent_map(em);
4858 em = existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004859 err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04004860 }
Chris Masona52d9a82007-08-27 16:49:44 -04004861 }
Chris Mason890871b2009-09-02 16:24:52 -04004862 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04004863out:
Chris Masonf4219502008-07-22 11:18:09 -04004864 if (path)
4865 btrfs_free_path(path);
Chris Masona52d9a82007-08-27 16:49:44 -04004866 if (trans) {
4867 ret = btrfs_end_transaction(trans, root);
Chris Masond3977122009-01-05 21:25:51 -05004868 if (!err)
Chris Masona52d9a82007-08-27 16:49:44 -04004869 err = ret;
4870 }
Chris Masona52d9a82007-08-27 16:49:44 -04004871 if (err) {
4872 free_extent_map(em);
Chris Masona52d9a82007-08-27 16:49:44 -04004873 return ERR_PTR(err);
4874 }
4875 return em;
4876}
4877
Chris Mason16432982008-04-10 10:23:21 -04004878static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4879 const struct iovec *iov, loff_t offset,
4880 unsigned long nr_segs)
4881{
Chris Masone1c4b742008-04-22 13:26:46 -04004882 return -EINVAL;
Chris Mason16432982008-04-10 10:23:21 -04004883}
4884
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05004885static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4886 __u64 start, __u64 len)
4887{
4888 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4889}
4890
Chris Mason9ebefb182007-06-15 13:50:00 -04004891int btrfs_readpage(struct file *file, struct page *page)
4892{
Chris Masond1310b22008-01-24 16:13:08 -05004893 struct extent_io_tree *tree;
4894 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04004895 return extent_read_full_page(tree, page, btrfs_get_extent);
Chris Mason39279cc2007-06-12 06:35:45 -04004896}
Chris Mason1832a6d2007-12-21 16:27:21 -05004897
Chris Mason39279cc2007-06-12 06:35:45 -04004898static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4899{
Chris Masond1310b22008-01-24 16:13:08 -05004900 struct extent_io_tree *tree;
Chris Masonb888db22007-08-27 16:49:44 -04004901
4902
4903 if (current->flags & PF_MEMALLOC) {
4904 redirty_page_for_writepage(wbc, page);
4905 unlock_page(page);
4906 return 0;
4907 }
Chris Masond1310b22008-01-24 16:13:08 -05004908 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04004909 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4910}
Chris Mason39279cc2007-06-12 06:35:45 -04004911
Chris Masonf4219502008-07-22 11:18:09 -04004912int btrfs_writepages(struct address_space *mapping,
4913 struct writeback_control *wbc)
Chris Masonb293f022007-11-01 19:45:34 -04004914{
Chris Masond1310b22008-01-24 16:13:08 -05004915 struct extent_io_tree *tree;
Chris Mason771ed682008-11-06 22:02:51 -05004916
Chris Masond1310b22008-01-24 16:13:08 -05004917 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Masonb293f022007-11-01 19:45:34 -04004918 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4919}
4920
Chris Mason3ab2fb52007-11-08 10:59:22 -05004921static int
4922btrfs_readpages(struct file *file, struct address_space *mapping,
4923 struct list_head *pages, unsigned nr_pages)
4924{
Chris Masond1310b22008-01-24 16:13:08 -05004925 struct extent_io_tree *tree;
4926 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Mason3ab2fb52007-11-08 10:59:22 -05004927 return extent_readpages(tree, mapping, pages, nr_pages,
4928 btrfs_get_extent);
4929}
Chris Masone6dcd2d2008-07-17 12:53:50 -04004930static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
Chris Masona52d9a82007-08-27 16:49:44 -04004931{
Chris Masond1310b22008-01-24 16:13:08 -05004932 struct extent_io_tree *tree;
4933 struct extent_map_tree *map;
Chris Masona52d9a82007-08-27 16:49:44 -04004934 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004935
Chris Masond1310b22008-01-24 16:13:08 -05004936 tree = &BTRFS_I(page->mapping->host)->io_tree;
4937 map = &BTRFS_I(page->mapping->host)->extent_tree;
Chris Mason70dec802008-01-29 09:59:12 -05004938 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
Chris Masona52d9a82007-08-27 16:49:44 -04004939 if (ret == 1) {
4940 ClearPagePrivate(page);
4941 set_page_private(page, 0);
4942 page_cache_release(page);
4943 }
4944 return ret;
4945}
Chris Mason39279cc2007-06-12 06:35:45 -04004946
Chris Masone6dcd2d2008-07-17 12:53:50 -04004947static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4948{
Chris Mason98509cf2008-09-11 15:51:43 -04004949 if (PageWriteback(page) || PageDirty(page))
4950 return 0;
Yan Zhengb335b002009-02-12 10:06:04 -05004951 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004952}
4953
Chris Masona52d9a82007-08-27 16:49:44 -04004954static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4955{
Chris Masond1310b22008-01-24 16:13:08 -05004956 struct extent_io_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004957 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004958 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004959 u64 page_start = page_offset(page);
4960 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04004961
Chris Mason8b62b722009-09-02 16:53:46 -04004962
4963 /*
4964 * we have the page locked, so new writeback can't start,
4965 * and the dirty bit won't be cleared while we are here.
4966 *
4967 * Wait for IO on this page so that we can safely clear
4968 * the PagePrivate2 bit and do ordered accounting
4969 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04004970 wait_on_page_writeback(page);
Chris Mason8b62b722009-09-02 16:53:46 -04004971
Chris Masond1310b22008-01-24 16:13:08 -05004972 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004973 if (offset) {
4974 btrfs_releasepage(page, GFP_NOFS);
4975 return;
4976 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00004977 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
4978 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004979 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4980 page_offset(page));
4981 if (ordered) {
Chris Masoneb84ae02008-07-17 13:53:27 -04004982 /*
4983 * IO on this page will never be started, so we need
4984 * to account for any ordered extents now
4985 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04004986 clear_extent_bit(tree, page_start, page_end,
4987 EXTENT_DIRTY | EXTENT_DELALLOC |
Josef Bacik32c00af2009-10-08 13:34:05 -04004988 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004989 &cached_state, GFP_NOFS);
Chris Mason8b62b722009-09-02 16:53:46 -04004990 /*
4991 * whoever cleared the private bit is responsible
4992 * for the finish_ordered_io
4993 */
4994 if (TestClearPagePrivate2(page)) {
4995 btrfs_finish_ordered_io(page->mapping->host,
4996 page_start, page_end);
4997 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004998 btrfs_put_ordered_extent(ordered);
Josef Bacik2ac55d42010-02-03 19:33:23 +00004999 cached_state = NULL;
5000 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5001 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005002 }
5003 clear_extent_bit(tree, page_start, page_end,
Josef Bacik32c00af2009-10-08 13:34:05 -04005004 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
Josef Bacik2ac55d42010-02-03 19:33:23 +00005005 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005006 __btrfs_releasepage(page, GFP_NOFS);
5007
Chris Mason4a096752008-07-21 10:29:44 -04005008 ClearPageChecked(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04005009 if (PagePrivate(page)) {
Chris Mason9ad6b7b2008-04-18 16:11:30 -04005010 ClearPagePrivate(page);
5011 set_page_private(page, 0);
5012 page_cache_release(page);
5013 }
Chris Mason39279cc2007-06-12 06:35:45 -04005014}
5015
Chris Mason9ebefb182007-06-15 13:50:00 -04005016/*
5017 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5018 * called from a page fault handler when a page is first dirtied. Hence we must
5019 * be careful to check for EOF conditions here. We set the page up correctly
5020 * for a written page which means we get ENOSPC checking when writing into
5021 * holes and correct delalloc and unwritten extent mapping on filesystems that
5022 * support these features.
5023 *
5024 * We are not allowed to take the i_mutex here so we have to play games to
5025 * protect against truncate races as the page could now be beyond EOF. Because
5026 * vmtruncate() writes the inode size before removing pages, once we have the
5027 * page lock we can determine safely if the page is beyond EOF. If it is not
5028 * beyond EOF, then the page is guaranteed safe against truncation until we
5029 * unlock the page.
5030 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07005031int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Chris Mason9ebefb182007-06-15 13:50:00 -04005032{
Nick Pigginc2ec1752009-03-31 15:23:21 -07005033 struct page *page = vmf->page;
Chris Mason6da6aba2007-12-18 16:15:09 -05005034 struct inode *inode = fdentry(vma->vm_file)->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05005035 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005036 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5037 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00005038 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005039 char *kaddr;
5040 unsigned long zero_start;
Chris Mason9ebefb182007-06-15 13:50:00 -04005041 loff_t size;
Chris Mason1832a6d2007-12-21 16:27:21 -05005042 int ret;
Chris Masona52d9a82007-08-27 16:49:44 -04005043 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005044 u64 page_end;
Chris Mason9ebefb182007-06-15 13:50:00 -04005045
Josef Bacik6a632092009-02-20 11:00:09 -05005046 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
Nick Piggin56a76f82009-03-31 15:23:23 -07005047 if (ret) {
5048 if (ret == -ENOMEM)
5049 ret = VM_FAULT_OOM;
5050 else /* -ENOSPC, -EIO, etc */
5051 ret = VM_FAULT_SIGBUS;
Chris Mason1832a6d2007-12-21 16:27:21 -05005052 goto out;
Nick Piggin56a76f82009-03-31 15:23:23 -07005053 }
Chris Mason1832a6d2007-12-21 16:27:21 -05005054
Josef Bacik9ed74f22009-09-11 16:12:44 -04005055 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
5056 if (ret) {
5057 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5058 ret = VM_FAULT_SIGBUS;
5059 goto out;
5060 }
5061
Nick Piggin56a76f82009-03-31 15:23:23 -07005062 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
Chris Masone6dcd2d2008-07-17 12:53:50 -04005063again:
Chris Mason9ebefb182007-06-15 13:50:00 -04005064 lock_page(page);
Chris Mason9ebefb182007-06-15 13:50:00 -04005065 size = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005066 page_start = page_offset(page);
5067 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04005068
Chris Mason9ebefb182007-06-15 13:50:00 -04005069 if ((page->mapping != inode->i_mapping) ||
Chris Masone6dcd2d2008-07-17 12:53:50 -04005070 (page_start >= size)) {
Josef Bacik6a632092009-02-20 11:00:09 -05005071 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
Chris Mason9ebefb182007-06-15 13:50:00 -04005072 /* page got truncated out from underneath us */
5073 goto out_unlock;
5074 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04005075 wait_on_page_writeback(page);
5076
Josef Bacik2ac55d42010-02-03 19:33:23 +00005077 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
5078 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005079 set_page_extent_mapped(page);
5080
Chris Masoneb84ae02008-07-17 13:53:27 -04005081 /*
5082 * we can't set the delalloc bits if there are pending ordered
5083 * extents. Drop our locks and wait for them to finish
5084 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04005085 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5086 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00005087 unlock_extent_cached(io_tree, page_start, page_end,
5088 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005089 unlock_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04005090 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005091 btrfs_put_ordered_extent(ordered);
5092 goto again;
5093 }
5094
Josef Bacikfbf19082009-10-01 17:10:23 -04005095 /*
5096 * XXX - page_mkwrite gets called every time the page is dirtied, even
5097 * if it was already dirty, so for space accounting reasons we need to
5098 * clear any delalloc bits for the range we are fixing to save. There
5099 * is probably a better way to do this, but for now keep consistent with
5100 * prepare_pages in the normal write path.
5101 */
Josef Bacik2ac55d42010-02-03 19:33:23 +00005102 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Josef Bacik32c00af2009-10-08 13:34:05 -04005103 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
Josef Bacik2ac55d42010-02-03 19:33:23 +00005104 0, 0, &cached_state, GFP_NOFS);
Josef Bacikfbf19082009-10-01 17:10:23 -04005105
Josef Bacik2ac55d42010-02-03 19:33:23 +00005106 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
5107 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005108 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00005109 unlock_extent_cached(io_tree, page_start, page_end,
5110 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005111 ret = VM_FAULT_SIGBUS;
Josef Bacikfbf19082009-10-01 17:10:23 -04005112 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005113 goto out_unlock;
5114 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04005115 ret = 0;
Chris Mason9ebefb182007-06-15 13:50:00 -04005116
5117 /* page is wholly or partially inside EOF */
Chris Masona52d9a82007-08-27 16:49:44 -04005118 if (page_start + PAGE_CACHE_SIZE > size)
Chris Masone6dcd2d2008-07-17 12:53:50 -04005119 zero_start = size & ~PAGE_CACHE_MASK;
Chris Mason9ebefb182007-06-15 13:50:00 -04005120 else
Chris Masone6dcd2d2008-07-17 12:53:50 -04005121 zero_start = PAGE_CACHE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04005122
Chris Masone6dcd2d2008-07-17 12:53:50 -04005123 if (zero_start != PAGE_CACHE_SIZE) {
5124 kaddr = kmap(page);
5125 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
5126 flush_dcache_page(page);
5127 kunmap(page);
5128 }
Chris Mason247e7432008-07-17 12:53:51 -04005129 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005130 set_page_dirty(page);
Chris Mason50a9b212009-09-11 12:33:12 -04005131 SetPageUptodate(page);
Chris Mason5a3f23d2009-03-31 13:27:11 -04005132
Chris Mason257c62e2009-10-13 13:21:08 -04005133 BTRFS_I(inode)->last_trans = root->fs_info->generation;
5134 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
5135
Josef Bacik2ac55d42010-02-03 19:33:23 +00005136 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
Chris Mason9ebefb182007-06-15 13:50:00 -04005137
5138out_unlock:
Josef Bacik9ed74f22009-09-11 16:12:44 -04005139 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
Chris Mason50a9b212009-09-11 12:33:12 -04005140 if (!ret)
5141 return VM_FAULT_LOCKED;
Chris Mason9ebefb182007-06-15 13:50:00 -04005142 unlock_page(page);
Chris Mason1832a6d2007-12-21 16:27:21 -05005143out:
Chris Mason9ebefb182007-06-15 13:50:00 -04005144 return ret;
5145}
5146
Chris Mason39279cc2007-06-12 06:35:45 -04005147static void btrfs_truncate(struct inode *inode)
5148{
5149 struct btrfs_root *root = BTRFS_I(inode)->root;
5150 int ret;
5151 struct btrfs_trans_handle *trans;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04005152 unsigned long nr;
Chris Masondbe674a2008-07-17 12:54:05 -04005153 u64 mask = root->sectorsize - 1;
Chris Mason39279cc2007-06-12 06:35:45 -04005154
Yan, Zheng80825102009-11-12 09:35:36 +00005155 if (!S_ISREG(inode->i_mode)) {
5156 WARN_ON(1);
Chris Mason39279cc2007-06-12 06:35:45 -04005157 return;
Yan, Zheng80825102009-11-12 09:35:36 +00005158 }
Chris Mason39279cc2007-06-12 06:35:45 -04005159
Josef Bacik5d5e1032009-10-13 16:46:49 -04005160 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
5161 if (ret)
5162 return;
Yan, Zheng80825102009-11-12 09:35:36 +00005163
Chris Mason4a096752008-07-21 10:29:44 -04005164 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
Yan, Zheng80825102009-11-12 09:35:36 +00005165 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04005166
Chris Mason39279cc2007-06-12 06:35:45 -04005167 trans = btrfs_start_transaction(root, 1);
Yan, Zheng80825102009-11-12 09:35:36 +00005168 btrfs_set_trans_block_group(trans, inode);
Chris Mason5a3f23d2009-03-31 13:27:11 -04005169
5170 /*
5171 * setattr is responsible for setting the ordered_data_close flag,
5172 * but that is only tested during the last file release. That
5173 * could happen well after the next commit, leaving a great big
5174 * window where new writes may get lost if someone chooses to write
5175 * to this file after truncating to zero
5176 *
5177 * The inode doesn't have any dirty data here, and so if we commit
5178 * this is a noop. If someone immediately starts writing to the inode
5179 * it is very likely we'll catch some of their writes in this
5180 * transaction, and the commit will find this file on the ordered
5181 * data list with good things to send down.
5182 *
5183 * This is a best effort solution, there is still a window where
5184 * using truncate to replace the contents of the file will
5185 * end up with a zero length file after a crash.
5186 */
5187 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5188 btrfs_add_ordered_operation(trans, root, inode);
5189
Yan, Zheng80825102009-11-12 09:35:36 +00005190 while (1) {
5191 ret = btrfs_truncate_inode_items(trans, root, inode,
5192 inode->i_size,
5193 BTRFS_EXTENT_DATA_KEY);
5194 if (ret != -EAGAIN)
5195 break;
Chris Mason39279cc2007-06-12 06:35:45 -04005196
Yan, Zheng80825102009-11-12 09:35:36 +00005197 ret = btrfs_update_inode(trans, root, inode);
5198 BUG_ON(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04005199
Yan, Zheng80825102009-11-12 09:35:36 +00005200 nr = trans->blocks_used;
5201 btrfs_end_transaction(trans, root);
5202 btrfs_btree_balance_dirty(root, nr);
5203
5204 trans = btrfs_start_transaction(root, 1);
5205 btrfs_set_trans_block_group(trans, inode);
5206 }
5207
5208 if (ret == 0 && inode->i_nlink > 0) {
5209 ret = btrfs_orphan_del(trans, inode);
5210 BUG_ON(ret);
5211 }
5212
5213 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik7b128762008-07-24 12:17:14 -04005214 BUG_ON(ret);
5215
Josef Bacik7b128762008-07-24 12:17:14 -04005216 nr = trans->blocks_used;
Chris Mason89ce8a62008-06-25 16:01:31 -04005217 ret = btrfs_end_transaction_throttle(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04005218 BUG_ON(ret);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04005219 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04005220}
5221
Sven Wegener3b963622008-06-09 21:57:42 -04005222/*
Chris Masond352ac62008-09-29 15:18:18 -04005223 * create a new subvolume directory/inode (helper for the ioctl).
5224 */
Yan Zhengd2fb3432008-12-11 16:30:39 -05005225int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
Yan, Zheng76dda932009-09-21 16:00:26 -04005226 struct btrfs_root *new_root,
Yan Zhengd2fb3432008-12-11 16:30:39 -05005227 u64 new_dirid, u64 alloc_hint)
Chris Mason39279cc2007-06-12 06:35:45 -04005228{
Chris Mason39279cc2007-06-12 06:35:45 -04005229 struct inode *inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04005230 int err;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005231 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005232
Josef Bacikaec74772008-07-24 12:12:38 -04005233 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
Yan Zhengd2fb3432008-12-11 16:30:39 -05005234 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
Chris Mason54aa1f42007-06-22 14:16:25 -04005235 if (IS_ERR(inode))
Christoph Hellwigf46b5a62008-06-11 21:53:53 -04005236 return PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04005237 inode->i_op = &btrfs_dir_inode_operations;
5238 inode->i_fop = &btrfs_dir_file_operations;
5239
Chris Mason39279cc2007-06-12 06:35:45 -04005240 inode->i_nlink = 1;
Chris Masondbe674a2008-07-17 12:54:05 -04005241 btrfs_i_size_write(inode, 0);
Sven Wegener3b963622008-06-09 21:57:42 -04005242
Yan, Zheng76dda932009-09-21 16:00:26 -04005243 err = btrfs_update_inode(trans, new_root, inode);
5244 BUG_ON(err);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04005245
Yan, Zheng76dda932009-09-21 16:00:26 -04005246 iput(inode);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04005247 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005248}
5249
Chris Masond352ac62008-09-29 15:18:18 -04005250/* helper function for file defrag and space balancing. This
5251 * forces readahead on a given range of bytes in an inode
5252 */
Chris Masonedbd8d42007-12-21 16:27:24 -05005253unsigned long btrfs_force_ra(struct address_space *mapping,
Chris Mason86479a02007-09-10 19:58:16 -04005254 struct file_ra_state *ra, struct file *file,
5255 pgoff_t offset, pgoff_t last_index)
5256{
Chris Mason8e7bf942008-04-28 09:02:36 -04005257 pgoff_t req_size = last_index - offset + 1;
Chris Mason86479a02007-09-10 19:58:16 -04005258
Chris Mason86479a02007-09-10 19:58:16 -04005259 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5260 return offset + req_size;
Chris Mason86479a02007-09-10 19:58:16 -04005261}
5262
Chris Mason39279cc2007-06-12 06:35:45 -04005263struct inode *btrfs_alloc_inode(struct super_block *sb)
5264{
5265 struct btrfs_inode *ei;
5266
5267 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5268 if (!ei)
5269 return NULL;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04005270 ei->last_trans = 0;
Chris Mason257c62e2009-10-13 13:21:08 -04005271 ei->last_sub_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04005272 ei->logged_trans = 0;
Josef Bacik32c00af2009-10-08 13:34:05 -04005273 ei->outstanding_extents = 0;
5274 ei->reserved_extents = 0;
Josef Bacika6dbd422009-11-11 15:53:34 -05005275 ei->root = NULL;
Josef Bacik32c00af2009-10-08 13:34:05 -04005276 spin_lock_init(&ei->accounting_lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005277 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Josef Bacik7b128762008-07-24 12:17:14 -04005278 INIT_LIST_HEAD(&ei->i_orphan);
Chris Mason5a3f23d2009-03-31 13:27:11 -04005279 INIT_LIST_HEAD(&ei->ordered_operations);
Chris Mason39279cc2007-06-12 06:35:45 -04005280 return &ei->vfs_inode;
5281}
5282
5283void btrfs_destroy_inode(struct inode *inode)
5284{
Chris Masone6dcd2d2008-07-17 12:53:50 -04005285 struct btrfs_ordered_extent *ordered;
Chris Mason5a3f23d2009-03-31 13:27:11 -04005286 struct btrfs_root *root = BTRFS_I(inode)->root;
5287
Chris Mason39279cc2007-06-12 06:35:45 -04005288 WARN_ON(!list_empty(&inode->i_dentry));
5289 WARN_ON(inode->i_data.nrpages);
5290
Chris Mason5a3f23d2009-03-31 13:27:11 -04005291 /*
Josef Bacika6dbd422009-11-11 15:53:34 -05005292 * This can happen where we create an inode, but somebody else also
5293 * created the same inode and we need to destroy the one we already
5294 * created.
5295 */
5296 if (!root)
5297 goto free;
5298
5299 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04005300 * Make sure we're properly removed from the ordered operation
5301 * lists.
5302 */
5303 smp_mb();
5304 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5305 spin_lock(&root->fs_info->ordered_extent_lock);
5306 list_del_init(&BTRFS_I(inode)->ordered_operations);
5307 spin_unlock(&root->fs_info->ordered_extent_lock);
5308 }
5309
5310 spin_lock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04005311 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
Yan, Zheng80825102009-11-12 09:35:36 +00005312 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
5313 inode->i_ino);
5314 list_del_init(&BTRFS_I(inode)->i_orphan);
Josef Bacik7b128762008-07-24 12:17:14 -04005315 }
Chris Mason5a3f23d2009-03-31 13:27:11 -04005316 spin_unlock(&root->list_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04005317
Chris Masond3977122009-01-05 21:25:51 -05005318 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04005319 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5320 if (!ordered)
5321 break;
5322 else {
Chris Masond3977122009-01-05 21:25:51 -05005323 printk(KERN_ERR "btrfs found ordered "
5324 "extent %llu %llu on inode cleanup\n",
5325 (unsigned long long)ordered->file_offset,
5326 (unsigned long long)ordered->len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005327 btrfs_remove_ordered_extent(inode, ordered);
5328 btrfs_put_ordered_extent(ordered);
5329 btrfs_put_ordered_extent(ordered);
5330 }
5331 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005332 inode_tree_del(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04005333 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
Josef Bacika6dbd422009-11-11 15:53:34 -05005334free:
Chris Mason39279cc2007-06-12 06:35:45 -04005335 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5336}
5337
Yan, Zheng76dda932009-09-21 16:00:26 -04005338void btrfs_drop_inode(struct inode *inode)
5339{
5340 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng76dda932009-09-21 16:00:26 -04005341 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5342 generic_delete_inode(inode);
5343 else
5344 generic_drop_inode(inode);
5345}
5346
Sven Wegener0ee0fda2008-07-30 16:54:26 -04005347static void init_once(void *foo)
Chris Mason39279cc2007-06-12 06:35:45 -04005348{
5349 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5350
5351 inode_init_once(&ei->vfs_inode);
5352}
5353
5354void btrfs_destroy_cachep(void)
5355{
5356 if (btrfs_inode_cachep)
5357 kmem_cache_destroy(btrfs_inode_cachep);
5358 if (btrfs_trans_handle_cachep)
5359 kmem_cache_destroy(btrfs_trans_handle_cachep);
5360 if (btrfs_transaction_cachep)
5361 kmem_cache_destroy(btrfs_transaction_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04005362 if (btrfs_path_cachep)
5363 kmem_cache_destroy(btrfs_path_cachep);
5364}
5365
5366int btrfs_init_cachep(void)
5367{
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02005368 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5369 sizeof(struct btrfs_inode), 0,
5370 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
Chris Mason39279cc2007-06-12 06:35:45 -04005371 if (!btrfs_inode_cachep)
5372 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02005373
5374 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5375 sizeof(struct btrfs_trans_handle), 0,
5376 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04005377 if (!btrfs_trans_handle_cachep)
5378 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02005379
5380 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5381 sizeof(struct btrfs_transaction), 0,
5382 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04005383 if (!btrfs_transaction_cachep)
5384 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02005385
5386 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5387 sizeof(struct btrfs_path), 0,
5388 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04005389 if (!btrfs_path_cachep)
5390 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02005391
Chris Mason39279cc2007-06-12 06:35:45 -04005392 return 0;
5393fail:
5394 btrfs_destroy_cachep();
5395 return -ENOMEM;
5396}
5397
5398static int btrfs_getattr(struct vfsmount *mnt,
5399 struct dentry *dentry, struct kstat *stat)
5400{
5401 struct inode *inode = dentry->d_inode;
5402 generic_fillattr(inode, stat);
Chris Mason3394e162008-11-17 20:42:26 -05005403 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
Chris Masond6667462008-01-03 14:51:00 -05005404 stat->blksize = PAGE_CACHE_SIZE;
Yan Zhenga76a3cd2008-10-09 11:46:29 -04005405 stat->blocks = (inode_get_bytes(inode) +
5406 BTRFS_I(inode)->delalloc_bytes) >> 9;
Chris Mason39279cc2007-06-12 06:35:45 -04005407 return 0;
5408}
5409
Chris Masond3977122009-01-05 21:25:51 -05005410static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5411 struct inode *new_dir, struct dentry *new_dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04005412{
5413 struct btrfs_trans_handle *trans;
5414 struct btrfs_root *root = BTRFS_I(old_dir)->root;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005415 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005416 struct inode *new_inode = new_dentry->d_inode;
5417 struct inode *old_inode = old_dentry->d_inode;
5418 struct timespec ctime = CURRENT_TIME;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005419 u64 index = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005420 u64 root_objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04005421 int ret;
5422
Yan, Zhengf679a842009-09-24 09:17:31 -04005423 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5424 return -EPERM;
5425
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005426 /* we only allow rename subvolume link between subvolumes */
5427 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
Chris Mason3394e162008-11-17 20:42:26 -05005428 return -EXDEV;
5429
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005430 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5431 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5432 return -ENOTEMPTY;
5433
Chris Mason39279cc2007-06-12 06:35:45 -04005434 if (S_ISDIR(old_inode->i_mode) && new_inode &&
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005435 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Chris Mason39279cc2007-06-12 06:35:45 -04005436 return -ENOTEMPTY;
Chris Mason0660b5a2008-11-17 20:37:39 -05005437
Josef Bacik9ed74f22009-09-11 16:12:44 -04005438 /*
Josef Bacik5df6a9f2009-11-10 21:23:48 -05005439 * We want to reserve the absolute worst case amount of items. So if
5440 * both inodes are subvols and we need to unlink them then that would
5441 * require 4 item modifications, but if they are both normal inodes it
5442 * would require 5 item modifications, so we'll assume their normal
5443 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5444 * should cover the worst case number of items we'll modify.
Josef Bacik9ed74f22009-09-11 16:12:44 -04005445 */
Josef Bacik5df6a9f2009-11-10 21:23:48 -05005446 ret = btrfs_reserve_metadata_space(root, 11);
Chris Mason1832a6d2007-12-21 16:27:21 -05005447 if (ret)
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005448 return ret;
Chris Mason1832a6d2007-12-21 16:27:21 -05005449
Chris Mason5a3f23d2009-03-31 13:27:11 -04005450 /*
5451 * we're using rename to replace one file with another.
5452 * and the replacement file is large. Start IO on it now so
5453 * we don't add too much work to the end of the transaction
5454 */
Bartlomiej Zolnierkiewicz4baf8c92009-08-07 13:47:08 -04005455 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
Chris Mason5a3f23d2009-03-31 13:27:11 -04005456 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5457 filemap_flush(old_inode->i_mapping);
5458
Yan, Zheng76dda932009-09-21 16:00:26 -04005459 /* close the racy window with snapshot create/destroy ioctl */
5460 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5461 down_read(&root->fs_info->subvol_sem);
Chris Mason5f39d392007-10-15 16:14:19 -04005462
Chris Mason39279cc2007-06-12 06:35:45 -04005463 trans = btrfs_start_transaction(root, 1);
Yan, Zhenga5719522009-09-24 09:17:31 -04005464 btrfs_set_trans_block_group(trans, new_dir);
Chris Mason39279cc2007-06-12 06:35:45 -04005465
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005466 if (dest != root)
5467 btrfs_record_root_in_trans(trans, dest);
5468
Yan, Zhenga5719522009-09-24 09:17:31 -04005469 ret = btrfs_set_inode_index(new_dir, &index);
5470 if (ret)
5471 goto out_fail;
Chris Mason5a3f23d2009-03-31 13:27:11 -04005472
Yan, Zhenga5719522009-09-24 09:17:31 -04005473 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005474 /* force full log commit if subvolume involved. */
5475 root->fs_info->last_trans_log_full_commit = trans->transid;
5476 } else {
Yan, Zhenga5719522009-09-24 09:17:31 -04005477 ret = btrfs_insert_inode_ref(trans, dest,
5478 new_dentry->d_name.name,
5479 new_dentry->d_name.len,
5480 old_inode->i_ino,
5481 new_dir->i_ino, index);
5482 if (ret)
5483 goto out_fail;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005484 /*
5485 * this is an ugly little race, but the rename is required
5486 * to make sure that if we crash, the inode is either at the
5487 * old name or the new one. pinning the log transaction lets
5488 * us make sure we don't allow a log commit to come in after
5489 * we unlink the name but before we add the new name back in.
5490 */
5491 btrfs_pin_log_trans(root);
5492 }
Chris Mason12fcfd22009-03-24 10:24:20 -04005493 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04005494 * make sure the inode gets flushed if it is replacing
5495 * something.
5496 */
5497 if (new_inode && new_inode->i_size &&
5498 old_inode && S_ISREG(old_inode->i_mode)) {
5499 btrfs_add_ordered_operation(trans, root, old_inode);
5500 }
5501
Chris Mason39279cc2007-06-12 06:35:45 -04005502 old_dir->i_ctime = old_dir->i_mtime = ctime;
5503 new_dir->i_ctime = new_dir->i_mtime = ctime;
5504 old_inode->i_ctime = ctime;
Chris Mason5f39d392007-10-15 16:14:19 -04005505
Chris Mason12fcfd22009-03-24 10:24:20 -04005506 if (old_dentry->d_parent != new_dentry->d_parent)
5507 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5508
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005509 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5510 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5511 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5512 old_dentry->d_name.name,
5513 old_dentry->d_name.len);
5514 } else {
5515 btrfs_inc_nlink(old_dentry->d_inode);
5516 ret = btrfs_unlink_inode(trans, root, old_dir,
5517 old_dentry->d_inode,
5518 old_dentry->d_name.name,
5519 old_dentry->d_name.len);
5520 }
5521 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04005522
5523 if (new_inode) {
5524 new_inode->i_ctime = CURRENT_TIME;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005525 if (unlikely(new_inode->i_ino ==
5526 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5527 root_objectid = BTRFS_I(new_inode)->location.objectid;
5528 ret = btrfs_unlink_subvol(trans, dest, new_dir,
5529 root_objectid,
5530 new_dentry->d_name.name,
5531 new_dentry->d_name.len);
5532 BUG_ON(new_inode->i_nlink == 0);
5533 } else {
5534 ret = btrfs_unlink_inode(trans, dest, new_dir,
5535 new_dentry->d_inode,
5536 new_dentry->d_name.name,
5537 new_dentry->d_name.len);
5538 }
5539 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04005540 if (new_inode->i_nlink == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04005541 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005542 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04005543 }
Chris Mason39279cc2007-06-12 06:35:45 -04005544 }
Josef Bacikaec74772008-07-24 12:12:38 -04005545
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005546 ret = btrfs_add_link(trans, new_dir, old_inode,
5547 new_dentry->d_name.name,
Yan, Zhenga5719522009-09-24 09:17:31 -04005548 new_dentry->d_name.len, 0, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005549 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04005550
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005551 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5552 btrfs_log_new_name(trans, old_inode, old_dir,
5553 new_dentry->d_parent);
5554 btrfs_end_log_trans(root);
5555 }
Chris Mason39279cc2007-06-12 06:35:45 -04005556out_fail:
Chris Masonab78c842008-07-29 16:15:18 -04005557 btrfs_end_transaction_throttle(trans, root);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005558
Yan, Zheng76dda932009-09-21 16:00:26 -04005559 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5560 up_read(&root->fs_info->subvol_sem);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005561
Josef Bacik5df6a9f2009-11-10 21:23:48 -05005562 btrfs_unreserve_metadata_space(root, 11);
Chris Mason39279cc2007-06-12 06:35:45 -04005563 return ret;
5564}
5565
Chris Masond352ac62008-09-29 15:18:18 -04005566/*
5567 * some fairly slow code that needs optimization. This walks the list
5568 * of all the inodes with pending delalloc and forces them to disk.
5569 */
Yan, Zheng24bbcf02009-11-12 09:36:34 +00005570int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
Chris Masonea8c2812008-08-04 23:17:27 -04005571{
5572 struct list_head *head = &root->fs_info->delalloc_inodes;
5573 struct btrfs_inode *binode;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04005574 struct inode *inode;
Chris Masonea8c2812008-08-04 23:17:27 -04005575
Yan Zhengc146afa2008-11-12 14:34:12 -05005576 if (root->fs_info->sb->s_flags & MS_RDONLY)
5577 return -EROFS;
5578
Chris Mason75eff682008-12-15 15:54:40 -05005579 spin_lock(&root->fs_info->delalloc_lock);
Chris Masond3977122009-01-05 21:25:51 -05005580 while (!list_empty(head)) {
Chris Masonea8c2812008-08-04 23:17:27 -04005581 binode = list_entry(head->next, struct btrfs_inode,
5582 delalloc_inodes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04005583 inode = igrab(&binode->vfs_inode);
5584 if (!inode)
5585 list_del_init(&binode->delalloc_inodes);
Chris Mason75eff682008-12-15 15:54:40 -05005586 spin_unlock(&root->fs_info->delalloc_lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04005587 if (inode) {
Chris Mason8c8bee12008-09-29 11:19:10 -04005588 filemap_flush(inode->i_mapping);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00005589 if (delay_iput)
5590 btrfs_add_delayed_iput(inode);
5591 else
5592 iput(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04005593 }
5594 cond_resched();
Chris Mason75eff682008-12-15 15:54:40 -05005595 spin_lock(&root->fs_info->delalloc_lock);
Chris Masonea8c2812008-08-04 23:17:27 -04005596 }
Chris Mason75eff682008-12-15 15:54:40 -05005597 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason8c8bee12008-09-29 11:19:10 -04005598
5599 /* the filemap_flush will queue IO into the worker threads, but
5600 * we have to make sure the IO is actually started and that
5601 * ordered extents get created before we return
5602 */
5603 atomic_inc(&root->fs_info->async_submit_draining);
Chris Masond3977122009-01-05 21:25:51 -05005604 while (atomic_read(&root->fs_info->nr_async_submits) ||
Chris Mason771ed682008-11-06 22:02:51 -05005605 atomic_read(&root->fs_info->async_delalloc_pages)) {
Chris Mason8c8bee12008-09-29 11:19:10 -04005606 wait_event(root->fs_info->async_submit_wait,
Chris Mason771ed682008-11-06 22:02:51 -05005607 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5608 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
Chris Mason8c8bee12008-09-29 11:19:10 -04005609 }
5610 atomic_dec(&root->fs_info->async_submit_draining);
Chris Masonea8c2812008-08-04 23:17:27 -04005611 return 0;
5612}
5613
Chris Mason39279cc2007-06-12 06:35:45 -04005614static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5615 const char *symname)
5616{
5617 struct btrfs_trans_handle *trans;
5618 struct btrfs_root *root = BTRFS_I(dir)->root;
5619 struct btrfs_path *path;
5620 struct btrfs_key key;
Chris Mason1832a6d2007-12-21 16:27:21 -05005621 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04005622 int err;
5623 int drop_inode = 0;
5624 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005625 u64 index = 0 ;
Chris Mason39279cc2007-06-12 06:35:45 -04005626 int name_len;
5627 int datasize;
Chris Mason5f39d392007-10-15 16:14:19 -04005628 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04005629 struct btrfs_file_extent_item *ei;
Chris Mason5f39d392007-10-15 16:14:19 -04005630 struct extent_buffer *leaf;
Chris Mason1832a6d2007-12-21 16:27:21 -05005631 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005632
5633 name_len = strlen(symname) + 1;
5634 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5635 return -ENAMETOOLONG;
Chris Mason1832a6d2007-12-21 16:27:21 -05005636
Josef Bacik9ed74f22009-09-11 16:12:44 -04005637 /*
5638 * 2 items for inode item and ref
5639 * 2 items for dir items
5640 * 1 item for xattr if selinux is on
5641 */
5642 err = btrfs_reserve_metadata_space(root, 5);
Chris Mason1832a6d2007-12-21 16:27:21 -05005643 if (err)
Josef Bacik9ed74f22009-09-11 16:12:44 -04005644 return err;
Chris Mason1832a6d2007-12-21 16:27:21 -05005645
Chris Mason39279cc2007-06-12 06:35:45 -04005646 trans = btrfs_start_transaction(root, 1);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005647 if (!trans)
5648 goto out_fail;
Chris Mason39279cc2007-06-12 06:35:45 -04005649 btrfs_set_trans_block_group(trans, dir);
5650
5651 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
5652 if (err) {
5653 err = -ENOSPC;
5654 goto out_unlock;
5655 }
5656
Josef Bacikaec74772008-07-24 12:12:38 -04005657 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Chris Mason9c583092008-01-29 15:15:18 -05005658 dentry->d_name.len,
5659 dentry->d_parent->d_inode->i_ino, objectid,
Chris Mason00e4e6b2008-08-05 11:18:09 -04005660 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5661 &index);
Chris Mason39279cc2007-06-12 06:35:45 -04005662 err = PTR_ERR(inode);
5663 if (IS_ERR(inode))
5664 goto out_unlock;
5665
Yan, Zhengf34f57a2009-11-12 09:35:27 +00005666 err = btrfs_init_inode_security(trans, inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04005667 if (err) {
5668 drop_inode = 1;
5669 goto out_unlock;
5670 }
5671
Chris Mason39279cc2007-06-12 06:35:45 -04005672 btrfs_set_trans_block_group(trans, inode);
Chris Mason00e4e6b2008-08-05 11:18:09 -04005673 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04005674 if (err)
5675 drop_inode = 1;
5676 else {
5677 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04005678 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04005679 inode->i_fop = &btrfs_file_operations;
5680 inode->i_op = &btrfs_file_inode_operations;
Chris Masond1310b22008-01-24 16:13:08 -05005681 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04005682 }
Chris Mason39279cc2007-06-12 06:35:45 -04005683 btrfs_update_inode_block_group(trans, inode);
5684 btrfs_update_inode_block_group(trans, dir);
5685 if (drop_inode)
5686 goto out_unlock;
5687
5688 path = btrfs_alloc_path();
5689 BUG_ON(!path);
5690 key.objectid = inode->i_ino;
5691 key.offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005692 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5693 datasize = btrfs_file_extent_calc_inline_size(name_len);
5694 err = btrfs_insert_empty_item(trans, root, path, &key,
5695 datasize);
Chris Mason54aa1f42007-06-22 14:16:25 -04005696 if (err) {
5697 drop_inode = 1;
5698 goto out_unlock;
5699 }
Chris Mason5f39d392007-10-15 16:14:19 -04005700 leaf = path->nodes[0];
5701 ei = btrfs_item_ptr(leaf, path->slots[0],
5702 struct btrfs_file_extent_item);
5703 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5704 btrfs_set_file_extent_type(leaf, ei,
Chris Mason39279cc2007-06-12 06:35:45 -04005705 BTRFS_FILE_EXTENT_INLINE);
Chris Masonc8b97812008-10-29 14:49:59 -04005706 btrfs_set_file_extent_encryption(leaf, ei, 0);
5707 btrfs_set_file_extent_compression(leaf, ei, 0);
5708 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5709 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5710
Chris Mason39279cc2007-06-12 06:35:45 -04005711 ptr = btrfs_file_extent_inline_start(ei);
Chris Mason5f39d392007-10-15 16:14:19 -04005712 write_extent_buffer(leaf, symname, ptr, name_len);
5713 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04005714 btrfs_free_path(path);
Chris Mason5f39d392007-10-15 16:14:19 -04005715
Chris Mason39279cc2007-06-12 06:35:45 -04005716 inode->i_op = &btrfs_symlink_inode_operations;
5717 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04005718 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Yan Zhengd899e052008-10-30 14:25:28 -04005719 inode_set_bytes(inode, name_len);
Chris Masondbe674a2008-07-17 12:54:05 -04005720 btrfs_i_size_write(inode, name_len - 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04005721 err = btrfs_update_inode(trans, root, inode);
5722 if (err)
5723 drop_inode = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04005724
5725out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04005726 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04005727 btrfs_end_transaction_throttle(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05005728out_fail:
Josef Bacik9ed74f22009-09-11 16:12:44 -04005729 btrfs_unreserve_metadata_space(root, 5);
Chris Mason39279cc2007-06-12 06:35:45 -04005730 if (drop_inode) {
5731 inode_dec_link_count(inode);
5732 iput(inode);
5733 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04005734 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04005735 return err;
5736}
Chris Mason16432982008-04-10 10:23:21 -04005737
Yan, Zheng5a303d52009-11-12 09:34:52 +00005738static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00005739 u64 alloc_hint, int mode, loff_t actual_len)
Yan Zhengd899e052008-10-30 14:25:28 -04005740{
Yan, Zheng5a303d52009-11-12 09:34:52 +00005741 struct btrfs_trans_handle *trans;
Yan Zhengd899e052008-10-30 14:25:28 -04005742 struct btrfs_root *root = BTRFS_I(inode)->root;
5743 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -04005744 u64 cur_offset = start;
5745 u64 num_bytes = end - start;
5746 int ret = 0;
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00005747 u64 i_size;
Yan Zhengd899e052008-10-30 14:25:28 -04005748
Yan Zhengd899e052008-10-30 14:25:28 -04005749 while (num_bytes > 0) {
Chris Mason3a1abec2009-12-17 15:47:17 -05005750 trans = btrfs_start_transaction(root, 1);
5751
Josef Bacik287a0ab2010-03-19 18:07:23 +00005752 ret = btrfs_reserve_extent(trans, root, num_bytes,
Yan Zhengd899e052008-10-30 14:25:28 -04005753 root->sectorsize, 0, alloc_hint,
5754 (u64)-1, &ins, 1);
5755 if (ret) {
5756 WARN_ON(1);
Chris Mason3a1abec2009-12-17 15:47:17 -05005757 goto stop_trans;
Yan Zhengd899e052008-10-30 14:25:28 -04005758 }
Yan, Zheng5a303d52009-11-12 09:34:52 +00005759
5760 ret = btrfs_reserve_metadata_space(root, 3);
5761 if (ret) {
5762 btrfs_free_reserved_extent(root, ins.objectid,
5763 ins.offset);
Chris Mason3a1abec2009-12-17 15:47:17 -05005764 goto stop_trans;
Yan, Zheng5a303d52009-11-12 09:34:52 +00005765 }
5766
Yan Zhengd899e052008-10-30 14:25:28 -04005767 ret = insert_reserved_file_extent(trans, inode,
5768 cur_offset, ins.objectid,
5769 ins.offset, ins.offset,
Yan, Zheng920bbbf2009-11-12 09:34:08 +00005770 ins.offset, 0, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04005771 BTRFS_FILE_EXTENT_PREALLOC);
5772 BUG_ON(ret);
Chris Masona1ed8352009-09-11 12:27:37 -04005773 btrfs_drop_extent_cache(inode, cur_offset,
5774 cur_offset + ins.offset -1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +00005775
Yan Zhengd899e052008-10-30 14:25:28 -04005776 num_bytes -= ins.offset;
5777 cur_offset += ins.offset;
5778 alloc_hint = ins.objectid + ins.offset;
Yan, Zheng5a303d52009-11-12 09:34:52 +00005779
Yan Zhengd899e052008-10-30 14:25:28 -04005780 inode->i_ctime = CURRENT_TIME;
Christoph Hellwig6cbff002009-04-17 10:37:41 +02005781 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Yan Zhengd899e052008-10-30 14:25:28 -04005782 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
Aneesh Kumar K.V23b5c502010-02-04 11:33:03 -05005783 (actual_len > inode->i_size) &&
5784 (cur_offset > inode->i_size)) {
5785
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00005786 if (cur_offset > actual_len)
5787 i_size = actual_len;
5788 else
5789 i_size = cur_offset;
5790 i_size_write(inode, i_size);
5791 btrfs_ordered_update_i_size(inode, i_size, NULL);
Yan, Zheng5a303d52009-11-12 09:34:52 +00005792 }
5793
Yan Zhengd899e052008-10-30 14:25:28 -04005794 ret = btrfs_update_inode(trans, root, inode);
5795 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -04005796
Yan, Zheng5a303d52009-11-12 09:34:52 +00005797 btrfs_end_transaction(trans, root);
5798 btrfs_unreserve_metadata_space(root, 3);
5799 }
Yan Zhengd899e052008-10-30 14:25:28 -04005800 return ret;
Chris Mason3a1abec2009-12-17 15:47:17 -05005801
5802stop_trans:
5803 btrfs_end_transaction(trans, root);
5804 return ret;
5805
Yan Zhengd899e052008-10-30 14:25:28 -04005806}
5807
5808static long btrfs_fallocate(struct inode *inode, int mode,
5809 loff_t offset, loff_t len)
5810{
Josef Bacik2ac55d42010-02-03 19:33:23 +00005811 struct extent_state *cached_state = NULL;
Yan Zhengd899e052008-10-30 14:25:28 -04005812 u64 cur_offset;
5813 u64 last_byte;
5814 u64 alloc_start;
5815 u64 alloc_end;
5816 u64 alloc_hint = 0;
Chris Masone980b502009-04-24 14:39:24 -04005817 u64 locked_end;
Yan Zhengd899e052008-10-30 14:25:28 -04005818 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5819 struct extent_map *em;
5820 int ret;
5821
5822 alloc_start = offset & ~mask;
5823 alloc_end = (offset + len + mask) & ~mask;
5824
Chris Mason546888d2009-04-21 11:53:38 -04005825 /*
5826 * wait for ordered IO before we have any locks. We'll loop again
5827 * below with the locks held.
5828 */
5829 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5830
Yan Zhengd899e052008-10-30 14:25:28 -04005831 mutex_lock(&inode->i_mutex);
5832 if (alloc_start > inode->i_size) {
5833 ret = btrfs_cont_expand(inode, alloc_start);
5834 if (ret)
5835 goto out;
5836 }
5837
Yan, Zheng5a303d52009-11-12 09:34:52 +00005838 ret = btrfs_check_data_free_space(BTRFS_I(inode)->root, inode,
Josef Bacika970b0a2009-06-27 21:07:34 -04005839 alloc_end - alloc_start);
5840 if (ret)
5841 goto out;
5842
Chris Masone980b502009-04-24 14:39:24 -04005843 locked_end = alloc_end - 1;
Yan Zhengd899e052008-10-30 14:25:28 -04005844 while (1) {
5845 struct btrfs_ordered_extent *ordered;
Chris Mason546888d2009-04-21 11:53:38 -04005846
Chris Mason546888d2009-04-21 11:53:38 -04005847 /* the extent lock is ordered inside the running
5848 * transaction
5849 */
Josef Bacik2ac55d42010-02-03 19:33:23 +00005850 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
5851 locked_end, 0, &cached_state, GFP_NOFS);
Yan Zhengd899e052008-10-30 14:25:28 -04005852 ordered = btrfs_lookup_first_ordered_extent(inode,
5853 alloc_end - 1);
5854 if (ordered &&
5855 ordered->file_offset + ordered->len > alloc_start &&
5856 ordered->file_offset < alloc_end) {
5857 btrfs_put_ordered_extent(ordered);
Josef Bacik2ac55d42010-02-03 19:33:23 +00005858 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
5859 alloc_start, locked_end,
5860 &cached_state, GFP_NOFS);
Chris Mason546888d2009-04-21 11:53:38 -04005861 /*
5862 * we can't wait on the range with the transaction
5863 * running or with the extent lock held
5864 */
Yan Zhengd899e052008-10-30 14:25:28 -04005865 btrfs_wait_ordered_range(inode, alloc_start,
5866 alloc_end - alloc_start);
5867 } else {
5868 if (ordered)
5869 btrfs_put_ordered_extent(ordered);
5870 break;
5871 }
5872 }
5873
5874 cur_offset = alloc_start;
5875 while (1) {
5876 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5877 alloc_end - cur_offset, 0);
5878 BUG_ON(IS_ERR(em) || !em);
5879 last_byte = min(extent_map_end(em), alloc_end);
5880 last_byte = (last_byte + mask) & ~mask;
Yan, Zheng5a303d52009-11-12 09:34:52 +00005881 if (em->block_start == EXTENT_MAP_HOLE ||
5882 (cur_offset >= inode->i_size &&
5883 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5884 ret = prealloc_file_range(inode,
5885 cur_offset, last_byte,
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00005886 alloc_hint, mode, offset+len);
Yan Zhengd899e052008-10-30 14:25:28 -04005887 if (ret < 0) {
5888 free_extent_map(em);
5889 break;
5890 }
5891 }
5892 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5893 alloc_hint = em->block_start;
5894 free_extent_map(em);
5895
5896 cur_offset = last_byte;
5897 if (cur_offset >= alloc_end) {
5898 ret = 0;
5899 break;
5900 }
5901 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00005902 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5903 &cached_state, GFP_NOFS);
Chris Mason546888d2009-04-21 11:53:38 -04005904
Yan, Zheng5a303d52009-11-12 09:34:52 +00005905 btrfs_free_reserved_data_space(BTRFS_I(inode)->root, inode,
5906 alloc_end - alloc_start);
Yan Zhengd899e052008-10-30 14:25:28 -04005907out:
5908 mutex_unlock(&inode->i_mutex);
5909 return ret;
5910}
5911
Chris Masone6dcd2d2008-07-17 12:53:50 -04005912static int btrfs_set_page_dirty(struct page *page)
5913{
Chris Masone6dcd2d2008-07-17 12:53:50 -04005914 return __set_page_dirty_nobuffers(page);
5915}
5916
Sven Wegener0ee0fda2008-07-30 16:54:26 -04005917static int btrfs_permission(struct inode *inode, int mask)
Yanfdebe2b2008-01-14 13:26:08 -05005918{
Christoph Hellwig6cbff002009-04-17 10:37:41 +02005919 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
Yanfdebe2b2008-01-14 13:26:08 -05005920 return -EACCES;
Josef Bacik33268ea2008-07-24 12:16:36 -04005921 return generic_permission(inode, mask, btrfs_check_acl);
Yanfdebe2b2008-01-14 13:26:08 -05005922}
Chris Mason39279cc2007-06-12 06:35:45 -04005923
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07005924static const struct inode_operations btrfs_dir_inode_operations = {
Chris Mason3394e162008-11-17 20:42:26 -05005925 .getattr = btrfs_getattr,
Chris Mason39279cc2007-06-12 06:35:45 -04005926 .lookup = btrfs_lookup,
5927 .create = btrfs_create,
5928 .unlink = btrfs_unlink,
5929 .link = btrfs_link,
5930 .mkdir = btrfs_mkdir,
5931 .rmdir = btrfs_rmdir,
5932 .rename = btrfs_rename,
5933 .symlink = btrfs_symlink,
5934 .setattr = btrfs_setattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04005935 .mknod = btrfs_mknod,
Christoph Hellwig95819c02008-08-28 06:21:17 -04005936 .setxattr = btrfs_setxattr,
5937 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05005938 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04005939 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05005940 .permission = btrfs_permission,
Chris Mason39279cc2007-06-12 06:35:45 -04005941};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07005942static const struct inode_operations btrfs_dir_ro_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04005943 .lookup = btrfs_lookup,
Yanfdebe2b2008-01-14 13:26:08 -05005944 .permission = btrfs_permission,
Chris Mason39279cc2007-06-12 06:35:45 -04005945};
Yan, Zheng76dda932009-09-21 16:00:26 -04005946
Alexey Dobriyan828c0952009-10-01 15:43:56 -07005947static const struct file_operations btrfs_dir_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04005948 .llseek = generic_file_llseek,
5949 .read = generic_read_dir,
David Woodhousecbdf5a22008-08-06 19:42:33 +01005950 .readdir = btrfs_real_readdir,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04005951 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04005952#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04005953 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04005954#endif
Sage Weil6bf13c02008-06-10 10:07:39 -04005955 .release = btrfs_release_file,
Chris Masone02119d2008-09-05 16:13:11 -04005956 .fsync = btrfs_sync_file,
Chris Mason39279cc2007-06-12 06:35:45 -04005957};
5958
Chris Masond1310b22008-01-24 16:13:08 -05005959static struct extent_io_ops btrfs_extent_io_ops = {
Chris Mason07157aa2007-08-30 08:50:51 -04005960 .fill_delalloc = run_delalloc_range,
Chris Mason065631f2008-02-20 12:07:25 -05005961 .submit_bio_hook = btrfs_submit_bio_hook,
Chris Mason239b14b2008-03-24 15:02:07 -04005962 .merge_bio_hook = btrfs_merge_bio_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04005963 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
Chris Masone6dcd2d2008-07-17 12:53:50 -04005964 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
Chris Mason247e7432008-07-17 12:53:51 -04005965 .writepage_start_hook = btrfs_writepage_start_hook,
Chris Mason1259ab72008-05-12 13:39:03 -04005966 .readpage_io_failed_hook = btrfs_io_failed_hook,
Chris Masonb0c68f82008-01-31 11:05:37 -05005967 .set_bit_hook = btrfs_set_bit_hook,
5968 .clear_bit_hook = btrfs_clear_bit_hook,
Josef Bacik9ed74f22009-09-11 16:12:44 -04005969 .merge_extent_hook = btrfs_merge_extent_hook,
5970 .split_extent_hook = btrfs_split_extent_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04005971};
5972
Chris Mason35054392009-01-21 13:11:13 -05005973/*
5974 * btrfs doesn't support the bmap operation because swapfiles
5975 * use bmap to make a mapping of extents in the file. They assume
5976 * these extents won't change over the life of the file and they
5977 * use the bmap result to do IO directly to the drive.
5978 *
5979 * the btrfs bmap call would return logical addresses that aren't
5980 * suitable for IO and they also will change frequently as COW
5981 * operations happen. So, swapfile + btrfs == corruption.
5982 *
5983 * For now we're avoiding this by dropping bmap.
5984 */
Alexey Dobriyan7f094102009-09-21 17:01:10 -07005985static const struct address_space_operations btrfs_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04005986 .readpage = btrfs_readpage,
5987 .writepage = btrfs_writepage,
Chris Masonb293f022007-11-01 19:45:34 -04005988 .writepages = btrfs_writepages,
Chris Mason3ab2fb52007-11-08 10:59:22 -05005989 .readpages = btrfs_readpages,
Chris Mason39279cc2007-06-12 06:35:45 -04005990 .sync_page = block_sync_page,
Chris Mason16432982008-04-10 10:23:21 -04005991 .direct_IO = btrfs_direct_IO,
Chris Masona52d9a82007-08-27 16:49:44 -04005992 .invalidatepage = btrfs_invalidatepage,
5993 .releasepage = btrfs_releasepage,
Chris Masone6dcd2d2008-07-17 12:53:50 -04005994 .set_page_dirty = btrfs_set_page_dirty,
Andi Kleen465fdd92009-09-16 11:50:18 +02005995 .error_remove_page = generic_error_remove_page,
Chris Mason39279cc2007-06-12 06:35:45 -04005996};
5997
Alexey Dobriyan7f094102009-09-21 17:01:10 -07005998static const struct address_space_operations btrfs_symlink_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04005999 .readpage = btrfs_readpage,
6000 .writepage = btrfs_writepage,
Chris Mason2bf5a722007-08-30 11:54:02 -04006001 .invalidatepage = btrfs_invalidatepage,
6002 .releasepage = btrfs_releasepage,
Chris Mason39279cc2007-06-12 06:35:45 -04006003};
6004
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07006005static const struct inode_operations btrfs_file_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04006006 .truncate = btrfs_truncate,
6007 .getattr = btrfs_getattr,
6008 .setattr = btrfs_setattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04006009 .setxattr = btrfs_setxattr,
6010 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05006011 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04006012 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05006013 .permission = btrfs_permission,
Yan Zhengd899e052008-10-30 14:25:28 -04006014 .fallocate = btrfs_fallocate,
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05006015 .fiemap = btrfs_fiemap,
Chris Mason39279cc2007-06-12 06:35:45 -04006016};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07006017static const struct inode_operations btrfs_special_inode_operations = {
Josef Bacik618e21d2007-07-11 10:18:17 -04006018 .getattr = btrfs_getattr,
6019 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05006020 .permission = btrfs_permission,
Christoph Hellwig95819c02008-08-28 06:21:17 -04006021 .setxattr = btrfs_setxattr,
6022 .getxattr = btrfs_getxattr,
Josef Bacik33268ea2008-07-24 12:16:36 -04006023 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04006024 .removexattr = btrfs_removexattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04006025};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07006026static const struct inode_operations btrfs_symlink_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04006027 .readlink = generic_readlink,
6028 .follow_link = page_follow_link_light,
6029 .put_link = page_put_link,
Yanfdebe2b2008-01-14 13:26:08 -05006030 .permission = btrfs_permission,
Jim Owens0279b4c2009-02-04 09:29:13 -05006031 .setxattr = btrfs_setxattr,
6032 .getxattr = btrfs_getxattr,
6033 .listxattr = btrfs_listxattr,
6034 .removexattr = btrfs_removexattr,
Chris Mason39279cc2007-06-12 06:35:45 -04006035};
Yan, Zheng76dda932009-09-21 16:00:26 -04006036
Alexey Dobriyan82d339d2009-10-09 09:54:36 -04006037const struct dentry_operations btrfs_dentry_operations = {
Yan, Zheng76dda932009-09-21 16:00:26 -04006038 .d_delete = btrfs_dentry_delete,
6039};