blob: e7a8303328b2eeb5a7397a94e5f279ebadebf0bf [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;
Li Zefan261507a02010-12-17 14:21:50 +0800125 int compress_type = BTRFS_COMPRESS_NONE;
Chris Masonc8b97812008-10-29 14:49:59 -0400126
127 if (compressed_size && compressed_pages) {
Li Zefan261507a02010-12-17 14:21:50 +0800128 compress_type = root->fs_info->compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -0400129 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
Li Zefan261507a02010-12-17 14:21:50 +0800162 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400163 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,
Li Zefan261507a02010-12-17 14:21:50 +0800179 compress_type);
Chris Masonc8b97812008-10-29 14:49:59 -0400180 } 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);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400255 btrfs_delalloc_release_metadata(inode, end + 1 - start);
Chris Masona1ed8352009-09-11 12:27:37 -0400256 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
Chris Masonc8b97812008-10-29 14:49:59 -0400257 return 0;
258}
259
Chris Mason771ed682008-11-06 22:02:51 -0500260struct async_extent {
261 u64 start;
262 u64 ram_size;
263 u64 compressed_size;
264 struct page **pages;
265 unsigned long nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800266 int compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500267 struct list_head list;
268};
269
270struct async_cow {
271 struct inode *inode;
272 struct btrfs_root *root;
273 struct page *locked_page;
274 u64 start;
275 u64 end;
276 struct list_head extents;
277 struct btrfs_work work;
278};
279
280static noinline int add_async_extent(struct async_cow *cow,
281 u64 start, u64 ram_size,
282 u64 compressed_size,
283 struct page **pages,
Li Zefan261507a02010-12-17 14:21:50 +0800284 unsigned long nr_pages,
285 int compress_type)
Chris Mason771ed682008-11-06 22:02:51 -0500286{
287 struct async_extent *async_extent;
288
289 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
290 async_extent->start = start;
291 async_extent->ram_size = ram_size;
292 async_extent->compressed_size = compressed_size;
293 async_extent->pages = pages;
294 async_extent->nr_pages = nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800295 async_extent->compress_type = compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500296 list_add_tail(&async_extent->list, &cow->extents);
297 return 0;
298}
299
Chris Masonc8b97812008-10-29 14:49:59 -0400300/*
Chris Mason771ed682008-11-06 22:02:51 -0500301 * we create compressed extents in two phases. The first
302 * phase compresses a range of pages that have already been
303 * locked (both pages and state bits are locked).
Chris Masonc8b97812008-10-29 14:49:59 -0400304 *
Chris Mason771ed682008-11-06 22:02:51 -0500305 * This is done inside an ordered work queue, and the compression
306 * is spread across many cpus. The actual IO submission is step
307 * two, and the ordered work queue takes care of making sure that
308 * happens in the same order things were put onto the queue by
309 * writepages and friends.
Chris Masonc8b97812008-10-29 14:49:59 -0400310 *
Chris Mason771ed682008-11-06 22:02:51 -0500311 * If this code finds it can't get good compression, it puts an
312 * entry onto the work queue to write the uncompressed bytes. This
313 * makes sure that both compressed inodes and uncompressed inodes
314 * are written in the same order that pdflush sent them down.
Chris Masond352ac62008-09-29 15:18:18 -0400315 */
Chris Mason771ed682008-11-06 22:02:51 -0500316static noinline int compress_file_range(struct inode *inode,
317 struct page *locked_page,
318 u64 start, u64 end,
319 struct async_cow *async_cow,
320 int *num_added)
Chris Masonb888db22007-08-27 16:49:44 -0400321{
322 struct btrfs_root *root = BTRFS_I(inode)->root;
323 struct btrfs_trans_handle *trans;
Chris Masondb945352007-10-15 16:15:53 -0400324 u64 num_bytes;
Chris Masondb945352007-10-15 16:15:53 -0400325 u64 blocksize = root->sectorsize;
Chris Masonc8b97812008-10-29 14:49:59 -0400326 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500327 u64 isize = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400328 int ret = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400329 struct page **pages = NULL;
330 unsigned long nr_pages;
331 unsigned long nr_pages_ret = 0;
332 unsigned long total_compressed = 0;
333 unsigned long total_in = 0;
334 unsigned long max_compressed = 128 * 1024;
Chris Mason771ed682008-11-06 22:02:51 -0500335 unsigned long max_uncompressed = 128 * 1024;
Chris Masonc8b97812008-10-29 14:49:59 -0400336 int i;
337 int will_compress;
Li Zefan261507a02010-12-17 14:21:50 +0800338 int compress_type = root->fs_info->compress_type;
Chris Masonb888db22007-08-27 16:49:44 -0400339
Chris Mason42dc7ba2008-12-15 11:44:56 -0500340 actual_end = min_t(u64, isize, end + 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400341again:
342 will_compress = 0;
343 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
344 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
345
Chris Masonf03d9301f2009-02-04 09:31:06 -0500346 /*
347 * we don't want to send crud past the end of i_size through
348 * compression, that's just a waste of CPU time. So, if the
349 * end of the file is before the start of our current
350 * requested range of bytes, we bail out to the uncompressed
351 * cleanup code that can deal with all of this.
352 *
353 * It isn't really the fastest way to fix things, but this is a
354 * very uncommon corner.
355 */
356 if (actual_end <= start)
357 goto cleanup_and_bail_uncompressed;
358
Chris Masonc8b97812008-10-29 14:49:59 -0400359 total_compressed = actual_end - start;
360
361 /* we want to make sure that amount of ram required to uncompress
362 * an extent is reasonable, so we limit the total size in ram
Chris Mason771ed682008-11-06 22:02:51 -0500363 * of a compressed extent to 128k. This is a crucial number
364 * because it also controls how easily we can spread reads across
365 * cpus for decompression.
366 *
367 * We also want to make sure the amount of IO required to do
368 * a random read is reasonably small, so we limit the size of
369 * a compressed extent to 128k.
Chris Masonc8b97812008-10-29 14:49:59 -0400370 */
371 total_compressed = min(total_compressed, max_uncompressed);
Chris Masondb945352007-10-15 16:15:53 -0400372 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
Chris Masonbe20aa92007-12-17 20:14:01 -0500373 num_bytes = max(blocksize, num_bytes);
Chris Masonc8b97812008-10-29 14:49:59 -0400374 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
Li Zefan261507a02010-12-17 14:21:50 +0800388 if (BTRFS_I(inode)->force_compress)
389 compress_type = BTRFS_I(inode)->force_compress;
390
391 ret = btrfs_compress_pages(compress_type,
392 inode->i_mapping, start,
393 total_compressed, pages,
394 nr_pages, &nr_pages_ret,
395 &total_in,
396 &total_compressed,
397 max_compressed);
Chris Masonc8b97812008-10-29 14:49:59 -0400398
399 if (!ret) {
400 unsigned long offset = total_compressed &
401 (PAGE_CACHE_SIZE - 1);
402 struct page *page = pages[nr_pages_ret - 1];
403 char *kaddr;
404
405 /* zero the tail end of the last page, we might be
406 * sending it down to disk
407 */
408 if (offset) {
409 kaddr = kmap_atomic(page, KM_USER0);
410 memset(kaddr + offset, 0,
411 PAGE_CACHE_SIZE - offset);
412 kunmap_atomic(kaddr, KM_USER0);
413 }
414 will_compress = 1;
415 }
416 }
417 if (start == 0) {
Chris Mason771ed682008-11-06 22:02:51 -0500418 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +0000419 BUG_ON(IS_ERR(trans));
Chris Mason771ed682008-11-06 22:02:51 -0500420 btrfs_set_trans_block_group(trans, inode);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400421 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Mason771ed682008-11-06 22:02:51 -0500422
Chris Masonc8b97812008-10-29 14:49:59 -0400423 /* lets try to make an inline extent */
Chris Mason771ed682008-11-06 22:02:51 -0500424 if (ret || total_in < (actual_end - start)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400425 /* we didn't compress the entire range, try
Chris Mason771ed682008-11-06 22:02:51 -0500426 * to make an uncompressed inline extent.
Chris Masonc8b97812008-10-29 14:49:59 -0400427 */
428 ret = cow_file_range_inline(trans, root, inode,
429 start, end, 0, NULL);
430 } else {
Chris Mason771ed682008-11-06 22:02:51 -0500431 /* try making a compressed inline extent */
Chris Masonc8b97812008-10-29 14:49:59 -0400432 ret = cow_file_range_inline(trans, root, inode,
433 start, end,
434 total_compressed, pages);
435 }
436 if (ret == 0) {
Chris Mason771ed682008-11-06 22:02:51 -0500437 /*
438 * inline extent creation worked, we don't need
439 * to create any more async work items. Unlock
440 * and free up our temp pages.
441 */
Chris Masonc8b97812008-10-29 14:49:59 -0400442 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400443 &BTRFS_I(inode)->io_tree,
444 start, end, NULL,
445 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
Chris Masona3429ab2009-10-08 12:30:20 -0400446 EXTENT_CLEAR_DELALLOC |
Chris Masona791e352009-10-08 11:27:10 -0400447 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000448
449 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400450 goto free_pages_out;
451 }
Yan, Zhengc2167752009-11-12 09:34:21 +0000452 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400453 }
454
455 if (will_compress) {
456 /*
457 * we aren't doing an inline extent round the compressed size
458 * up to a block size boundary so the allocator does sane
459 * things
460 */
461 total_compressed = (total_compressed + blocksize - 1) &
462 ~(blocksize - 1);
463
464 /*
465 * one last check to make sure the compression is really a
466 * win, compare the page count read with the blocks on disk
467 */
468 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
469 ~(PAGE_CACHE_SIZE - 1);
470 if (total_compressed >= total_in) {
471 will_compress = 0;
472 } else {
Chris Masonc8b97812008-10-29 14:49:59 -0400473 num_bytes = total_in;
474 }
475 }
476 if (!will_compress && pages) {
477 /*
478 * the compression code ran but failed to make things smaller,
479 * free any pages it allocated and our page pointer array
480 */
481 for (i = 0; i < nr_pages_ret; i++) {
Chris Mason70b99e62008-10-31 12:46:39 -0400482 WARN_ON(pages[i]->mapping);
Chris Masonc8b97812008-10-29 14:49:59 -0400483 page_cache_release(pages[i]);
484 }
485 kfree(pages);
486 pages = NULL;
487 total_compressed = 0;
488 nr_pages_ret = 0;
489
490 /* flag the file so we don't compress in the future */
Chris Mason1e701a32010-03-11 09:42:04 -0500491 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
492 !(BTRFS_I(inode)->force_compress)) {
Chris Masona555f812010-01-28 16:18:15 -0500493 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
Chris Mason1e701a32010-03-11 09:42:04 -0500494 }
Chris Masonc8b97812008-10-29 14:49:59 -0400495 }
Chris Mason771ed682008-11-06 22:02:51 -0500496 if (will_compress) {
497 *num_added += 1;
498
499 /* the async work queues will take care of doing actual
500 * allocation on disk for these compressed pages,
501 * and will submit them to the elevator.
502 */
503 add_async_extent(async_cow, start, num_bytes,
Li Zefan261507a02010-12-17 14:21:50 +0800504 total_compressed, pages, nr_pages_ret,
505 compress_type);
Chris Mason771ed682008-11-06 22:02:51 -0500506
Yan, Zheng24ae6362010-12-06 07:02:36 +0000507 if (start + num_bytes < end) {
Chris Mason771ed682008-11-06 22:02:51 -0500508 start += num_bytes;
509 pages = NULL;
510 cond_resched();
511 goto again;
512 }
513 } else {
Chris Masonf03d9301f2009-02-04 09:31:06 -0500514cleanup_and_bail_uncompressed:
Chris Mason771ed682008-11-06 22:02:51 -0500515 /*
516 * No compression, but we still need to write the pages in
517 * the file we've been given so far. redirty the locked
518 * page if it corresponds to our extent and set things up
519 * for the async work queue to run cow_file_range to do
520 * the normal delalloc dance
521 */
522 if (page_offset(locked_page) >= start &&
523 page_offset(locked_page) <= end) {
524 __set_page_dirty_nobuffers(locked_page);
525 /* unlocked later on in the async handlers */
526 }
Li Zefan261507a02010-12-17 14:21:50 +0800527 add_async_extent(async_cow, start, end - start + 1,
528 0, NULL, 0, BTRFS_COMPRESS_NONE);
Chris Mason771ed682008-11-06 22:02:51 -0500529 *num_added += 1;
530 }
531
532out:
533 return 0;
534
535free_pages_out:
536 for (i = 0; i < nr_pages_ret; i++) {
537 WARN_ON(pages[i]->mapping);
538 page_cache_release(pages[i]);
539 }
Chris Masond3977122009-01-05 21:25:51 -0500540 kfree(pages);
Chris Mason771ed682008-11-06 22:02:51 -0500541
542 goto out;
543}
544
545/*
546 * phase two of compressed writeback. This is the ordered portion
547 * of the code, which only gets called in the order the work was
548 * queued. We walk all the async extents created by compress_file_range
549 * and send them down to the disk.
550 */
551static noinline int submit_compressed_extents(struct inode *inode,
552 struct async_cow *async_cow)
553{
554 struct async_extent *async_extent;
555 u64 alloc_hint = 0;
556 struct btrfs_trans_handle *trans;
557 struct btrfs_key ins;
558 struct extent_map *em;
559 struct btrfs_root *root = BTRFS_I(inode)->root;
560 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
561 struct extent_io_tree *io_tree;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500562 int ret = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500563
564 if (list_empty(&async_cow->extents))
565 return 0;
566
Chris Mason771ed682008-11-06 22:02:51 -0500567
Chris Masond3977122009-01-05 21:25:51 -0500568 while (!list_empty(&async_cow->extents)) {
Chris Mason771ed682008-11-06 22:02:51 -0500569 async_extent = list_entry(async_cow->extents.next,
570 struct async_extent, list);
571 list_del(&async_extent->list);
572
573 io_tree = &BTRFS_I(inode)->io_tree;
574
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500575retry:
Chris Mason771ed682008-11-06 22:02:51 -0500576 /* did the compression code fall back to uncompressed IO? */
577 if (!async_extent->pages) {
578 int page_started = 0;
579 unsigned long nr_written = 0;
580
581 lock_extent(io_tree, async_extent->start,
Josef Bacik2ac55d42010-02-03 19:33:23 +0000582 async_extent->start +
583 async_extent->ram_size - 1, GFP_NOFS);
Chris Mason771ed682008-11-06 22:02:51 -0500584
585 /* allocate blocks */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500586 ret = cow_file_range(inode, async_cow->locked_page,
587 async_extent->start,
588 async_extent->start +
589 async_extent->ram_size - 1,
590 &page_started, &nr_written, 0);
Chris Mason771ed682008-11-06 22:02:51 -0500591
592 /*
593 * if page_started, cow_file_range inserted an
594 * inline extent and took care of all the unlocking
595 * and IO for us. Otherwise, we need to submit
596 * all those pages down to the drive.
597 */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500598 if (!page_started && !ret)
Chris Mason771ed682008-11-06 22:02:51 -0500599 extent_write_locked_range(io_tree,
600 inode, async_extent->start,
Chris Masond3977122009-01-05 21:25:51 -0500601 async_extent->start +
Chris Mason771ed682008-11-06 22:02:51 -0500602 async_extent->ram_size - 1,
603 btrfs_get_extent,
604 WB_SYNC_ALL);
605 kfree(async_extent);
606 cond_resched();
607 continue;
608 }
609
610 lock_extent(io_tree, async_extent->start,
611 async_extent->start + async_extent->ram_size - 1,
612 GFP_NOFS);
Chris Mason771ed682008-11-06 22:02:51 -0500613
Yan, Zhengc2167752009-11-12 09:34:21 +0000614 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +0000615 BUG_ON(IS_ERR(trans));
Chris Mason771ed682008-11-06 22:02:51 -0500616 ret = btrfs_reserve_extent(trans, root,
617 async_extent->compressed_size,
618 async_extent->compressed_size,
619 0, alloc_hint,
620 (u64)-1, &ins, 1);
Yan, Zhengc2167752009-11-12 09:34:21 +0000621 btrfs_end_transaction(trans, root);
622
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500623 if (ret) {
624 int i;
625 for (i = 0; i < async_extent->nr_pages; i++) {
626 WARN_ON(async_extent->pages[i]->mapping);
627 page_cache_release(async_extent->pages[i]);
628 }
629 kfree(async_extent->pages);
630 async_extent->nr_pages = 0;
631 async_extent->pages = NULL;
632 unlock_extent(io_tree, async_extent->start,
633 async_extent->start +
634 async_extent->ram_size - 1, GFP_NOFS);
635 goto retry;
636 }
637
Yan, Zhengc2167752009-11-12 09:34:21 +0000638 /*
639 * here we're doing allocation and writeback of the
640 * compressed pages
641 */
642 btrfs_drop_extent_cache(inode, async_extent->start,
643 async_extent->start +
644 async_extent->ram_size - 1, 0);
645
Chris Mason771ed682008-11-06 22:02:51 -0500646 em = alloc_extent_map(GFP_NOFS);
Tsutomu Itohc26a9202011-02-14 00:45:29 +0000647 BUG_ON(!em);
Chris Mason771ed682008-11-06 22:02:51 -0500648 em->start = async_extent->start;
649 em->len = async_extent->ram_size;
Chris Mason445a6942008-11-10 11:53:33 -0500650 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500651
652 em->block_start = ins.objectid;
653 em->block_len = ins.offset;
654 em->bdev = root->fs_info->fs_devices->latest_bdev;
Li Zefan261507a02010-12-17 14:21:50 +0800655 em->compress_type = async_extent->compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500656 set_bit(EXTENT_FLAG_PINNED, &em->flags);
657 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
658
Chris Masond3977122009-01-05 21:25:51 -0500659 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400660 write_lock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500661 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -0400662 write_unlock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500663 if (ret != -EEXIST) {
664 free_extent_map(em);
665 break;
666 }
667 btrfs_drop_extent_cache(inode, async_extent->start,
668 async_extent->start +
669 async_extent->ram_size - 1, 0);
670 }
671
Li Zefan261507a02010-12-17 14:21:50 +0800672 ret = btrfs_add_ordered_extent_compress(inode,
673 async_extent->start,
674 ins.objectid,
675 async_extent->ram_size,
676 ins.offset,
677 BTRFS_ORDERED_COMPRESSED,
678 async_extent->compress_type);
Chris Mason771ed682008-11-06 22:02:51 -0500679 BUG_ON(ret);
680
Chris Mason771ed682008-11-06 22:02:51 -0500681 /*
682 * clear dirty, set writeback and unlock the pages.
683 */
684 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400685 &BTRFS_I(inode)->io_tree,
686 async_extent->start,
687 async_extent->start +
688 async_extent->ram_size - 1,
689 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
690 EXTENT_CLEAR_UNLOCK |
Chris Masona3429ab2009-10-08 12:30:20 -0400691 EXTENT_CLEAR_DELALLOC |
Chris Masona791e352009-10-08 11:27:10 -0400692 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
Chris Mason771ed682008-11-06 22:02:51 -0500693
694 ret = btrfs_submit_compressed_write(inode,
Chris Masond3977122009-01-05 21:25:51 -0500695 async_extent->start,
696 async_extent->ram_size,
697 ins.objectid,
698 ins.offset, async_extent->pages,
699 async_extent->nr_pages);
Chris Mason771ed682008-11-06 22:02:51 -0500700
701 BUG_ON(ret);
Chris Mason771ed682008-11-06 22:02:51 -0500702 alloc_hint = ins.objectid + ins.offset;
703 kfree(async_extent);
704 cond_resched();
705 }
706
Chris Mason771ed682008-11-06 22:02:51 -0500707 return 0;
708}
709
Josef Bacik4b46fce2010-05-23 11:00:55 -0400710static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
711 u64 num_bytes)
712{
713 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
714 struct extent_map *em;
715 u64 alloc_hint = 0;
716
717 read_lock(&em_tree->lock);
718 em = search_extent_mapping(em_tree, start, num_bytes);
719 if (em) {
720 /*
721 * if block start isn't an actual block number then find the
722 * first block in this inode and use that as a hint. If that
723 * block is also bogus then just don't worry about it.
724 */
725 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
726 free_extent_map(em);
727 em = search_extent_mapping(em_tree, 0, 0);
728 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
729 alloc_hint = em->block_start;
730 if (em)
731 free_extent_map(em);
732 } else {
733 alloc_hint = em->block_start;
734 free_extent_map(em);
735 }
736 }
737 read_unlock(&em_tree->lock);
738
739 return alloc_hint;
740}
741
Chris Mason771ed682008-11-06 22:02:51 -0500742/*
743 * when extent_io.c finds a delayed allocation range in the file,
744 * the call backs end up in this code. The basic idea is to
745 * allocate extents on disk for the range, and create ordered data structs
746 * in ram to track those extents.
747 *
748 * locked_page is the page that writepage had locked already. We use
749 * it to make sure we don't do extra locks or unlocks.
750 *
751 * *page_started is set to one if we unlock locked_page and do everything
752 * required to start IO on it. It may be clean and already done with
753 * IO when we return.
754 */
755static noinline int cow_file_range(struct inode *inode,
756 struct page *locked_page,
757 u64 start, u64 end, int *page_started,
758 unsigned long *nr_written,
759 int unlock)
760{
761 struct btrfs_root *root = BTRFS_I(inode)->root;
762 struct btrfs_trans_handle *trans;
763 u64 alloc_hint = 0;
764 u64 num_bytes;
765 unsigned long ram_size;
766 u64 disk_num_bytes;
767 u64 cur_alloc_size;
768 u64 blocksize = root->sectorsize;
Chris Mason771ed682008-11-06 22:02:51 -0500769 struct btrfs_key ins;
770 struct extent_map *em;
771 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
772 int ret = 0;
773
Josef Bacik0cb59c92010-07-02 12:14:14 -0400774 BUG_ON(root == root->fs_info->tree_root);
Chris Mason771ed682008-11-06 22:02:51 -0500775 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +0000776 BUG_ON(IS_ERR(trans));
Chris Mason771ed682008-11-06 22:02:51 -0500777 btrfs_set_trans_block_group(trans, inode);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400778 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Mason771ed682008-11-06 22:02:51 -0500779
Chris Mason771ed682008-11-06 22:02:51 -0500780 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
781 num_bytes = max(blocksize, num_bytes);
782 disk_num_bytes = num_bytes;
783 ret = 0;
784
785 if (start == 0) {
786 /* lets try to make an inline extent */
787 ret = cow_file_range_inline(trans, root, inode,
788 start, end, 0, NULL);
789 if (ret == 0) {
790 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400791 &BTRFS_I(inode)->io_tree,
792 start, end, NULL,
793 EXTENT_CLEAR_UNLOCK_PAGE |
794 EXTENT_CLEAR_UNLOCK |
795 EXTENT_CLEAR_DELALLOC |
796 EXTENT_CLEAR_DIRTY |
797 EXTENT_SET_WRITEBACK |
798 EXTENT_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000799
Chris Mason771ed682008-11-06 22:02:51 -0500800 *nr_written = *nr_written +
801 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
802 *page_started = 1;
803 ret = 0;
804 goto out;
805 }
806 }
Chris Masonc8b97812008-10-29 14:49:59 -0400807
808 BUG_ON(disk_num_bytes >
809 btrfs_super_total_bytes(&root->fs_info->super_copy));
810
Josef Bacik4b46fce2010-05-23 11:00:55 -0400811 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400812 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
Chris Mason3b951512008-04-17 11:29:12 -0400813
Chris Masond3977122009-01-05 21:25:51 -0500814 while (disk_num_bytes > 0) {
Chris Masona791e352009-10-08 11:27:10 -0400815 unsigned long op;
816
Josef Bacik287a0ab2010-03-19 18:07:23 +0000817 cur_alloc_size = disk_num_bytes;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400818 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
Chris Mason771ed682008-11-06 22:02:51 -0500819 root->sectorsize, 0, alloc_hint,
Chris Masone6dcd2d2008-07-17 12:53:50 -0400820 (u64)-1, &ins, 1);
Chris Masond3977122009-01-05 21:25:51 -0500821 BUG_ON(ret);
822
Chris Masone6dcd2d2008-07-17 12:53:50 -0400823 em = alloc_extent_map(GFP_NOFS);
Tsutomu Itohc26a9202011-02-14 00:45:29 +0000824 BUG_ON(!em);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400825 em->start = start;
Chris Mason445a6942008-11-10 11:53:33 -0500826 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500827 ram_size = ins.offset;
828 em->len = ins.offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400829
Chris Masone6dcd2d2008-07-17 12:53:50 -0400830 em->block_start = ins.objectid;
Chris Masonc8b97812008-10-29 14:49:59 -0400831 em->block_len = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400832 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason7f3c74f2008-07-18 12:01:11 -0400833 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -0400834
Chris Masond3977122009-01-05 21:25:51 -0500835 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400836 write_lock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400837 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -0400838 write_unlock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400839 if (ret != -EEXIST) {
840 free_extent_map(em);
841 break;
842 }
843 btrfs_drop_extent_cache(inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -0400844 start + ram_size - 1, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400845 }
846
Chris Mason98d20f62008-04-14 09:46:10 -0400847 cur_alloc_size = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400848 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
Chris Mason771ed682008-11-06 22:02:51 -0500849 ram_size, cur_alloc_size, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400850 BUG_ON(ret);
Chris Masonc8b97812008-10-29 14:49:59 -0400851
Yan Zheng17d217f2008-12-12 10:03:38 -0500852 if (root->root_key.objectid ==
853 BTRFS_DATA_RELOC_TREE_OBJECTID) {
854 ret = btrfs_reloc_clone_csums(inode, start,
855 cur_alloc_size);
856 BUG_ON(ret);
857 }
858
Chris Masond3977122009-01-05 21:25:51 -0500859 if (disk_num_bytes < cur_alloc_size)
Chris Mason3b951512008-04-17 11:29:12 -0400860 break;
Chris Masond3977122009-01-05 21:25:51 -0500861
Chris Masonc8b97812008-10-29 14:49:59 -0400862 /* we're not doing compressed IO, don't unlock the first
863 * page (which the caller expects to stay locked), don't
864 * clear any dirty bits and don't set any writeback bits
Chris Mason8b62b722009-09-02 16:53:46 -0400865 *
866 * Do set the Private2 bit so we know this page was properly
867 * setup for writepage
Chris Masonc8b97812008-10-29 14:49:59 -0400868 */
Chris Masona791e352009-10-08 11:27:10 -0400869 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
870 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
871 EXTENT_SET_PRIVATE2;
872
Chris Masonc8b97812008-10-29 14:49:59 -0400873 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
874 start, start + ram_size - 1,
Chris Masona791e352009-10-08 11:27:10 -0400875 locked_page, op);
Chris Masonc8b97812008-10-29 14:49:59 -0400876 disk_num_bytes -= cur_alloc_size;
Chris Masonc59f8952007-12-17 20:14:04 -0500877 num_bytes -= cur_alloc_size;
878 alloc_hint = ins.objectid + ins.offset;
879 start += cur_alloc_size;
Chris Masonb888db22007-08-27 16:49:44 -0400880 }
Chris Masonb888db22007-08-27 16:49:44 -0400881out:
Chris Mason771ed682008-11-06 22:02:51 -0500882 ret = 0;
Chris Masonb888db22007-08-27 16:49:44 -0400883 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400884
Chris Masonbe20aa92007-12-17 20:14:01 -0500885 return ret;
Chris Mason771ed682008-11-06 22:02:51 -0500886}
Chris Masonc8b97812008-10-29 14:49:59 -0400887
Chris Mason771ed682008-11-06 22:02:51 -0500888/*
889 * work queue call back to started compression on a file and pages
890 */
891static noinline void async_cow_start(struct btrfs_work *work)
892{
893 struct async_cow *async_cow;
894 int num_added = 0;
895 async_cow = container_of(work, struct async_cow, work);
896
897 compress_file_range(async_cow->inode, async_cow->locked_page,
898 async_cow->start, async_cow->end, async_cow,
899 &num_added);
900 if (num_added == 0)
901 async_cow->inode = NULL;
902}
903
904/*
905 * work queue call back to submit previously compressed pages
906 */
907static noinline void async_cow_submit(struct btrfs_work *work)
908{
909 struct async_cow *async_cow;
910 struct btrfs_root *root;
911 unsigned long nr_pages;
912
913 async_cow = container_of(work, struct async_cow, work);
914
915 root = async_cow->root;
916 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
917 PAGE_CACHE_SHIFT;
918
919 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
920
921 if (atomic_read(&root->fs_info->async_delalloc_pages) <
922 5 * 1042 * 1024 &&
923 waitqueue_active(&root->fs_info->async_submit_wait))
924 wake_up(&root->fs_info->async_submit_wait);
925
Chris Masond3977122009-01-05 21:25:51 -0500926 if (async_cow->inode)
Chris Mason771ed682008-11-06 22:02:51 -0500927 submit_compressed_extents(async_cow->inode, async_cow);
Chris Mason771ed682008-11-06 22:02:51 -0500928}
Chris Masonc8b97812008-10-29 14:49:59 -0400929
Chris Mason771ed682008-11-06 22:02:51 -0500930static noinline void async_cow_free(struct btrfs_work *work)
931{
932 struct async_cow *async_cow;
933 async_cow = container_of(work, struct async_cow, work);
934 kfree(async_cow);
935}
936
937static int cow_file_range_async(struct inode *inode, struct page *locked_page,
938 u64 start, u64 end, int *page_started,
939 unsigned long *nr_written)
940{
941 struct async_cow *async_cow;
942 struct btrfs_root *root = BTRFS_I(inode)->root;
943 unsigned long nr_pages;
944 u64 cur_end;
945 int limit = 10 * 1024 * 1042;
946
Chris Masona3429ab2009-10-08 12:30:20 -0400947 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
948 1, 0, NULL, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -0500949 while (start < end) {
Chris Mason771ed682008-11-06 22:02:51 -0500950 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
951 async_cow->inode = inode;
952 async_cow->root = root;
953 async_cow->locked_page = locked_page;
954 async_cow->start = start;
955
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200956 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
Chris Mason771ed682008-11-06 22:02:51 -0500957 cur_end = end;
958 else
959 cur_end = min(end, start + 512 * 1024 - 1);
960
961 async_cow->end = cur_end;
962 INIT_LIST_HEAD(&async_cow->extents);
963
964 async_cow->work.func = async_cow_start;
965 async_cow->work.ordered_func = async_cow_submit;
966 async_cow->work.ordered_free = async_cow_free;
967 async_cow->work.flags = 0;
968
Chris Mason771ed682008-11-06 22:02:51 -0500969 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
970 PAGE_CACHE_SHIFT;
971 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
972
973 btrfs_queue_worker(&root->fs_info->delalloc_workers,
974 &async_cow->work);
975
976 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
977 wait_event(root->fs_info->async_submit_wait,
978 (atomic_read(&root->fs_info->async_delalloc_pages) <
979 limit));
980 }
981
Chris Masond3977122009-01-05 21:25:51 -0500982 while (atomic_read(&root->fs_info->async_submit_draining) &&
Chris Mason771ed682008-11-06 22:02:51 -0500983 atomic_read(&root->fs_info->async_delalloc_pages)) {
984 wait_event(root->fs_info->async_submit_wait,
985 (atomic_read(&root->fs_info->async_delalloc_pages) ==
986 0));
987 }
988
989 *nr_written += nr_pages;
990 start = cur_end + 1;
991 }
992 *page_started = 1;
993 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -0500994}
995
Chris Masond3977122009-01-05 21:25:51 -0500996static noinline int csum_exist_in_range(struct btrfs_root *root,
Yan Zheng17d217f2008-12-12 10:03:38 -0500997 u64 bytenr, u64 num_bytes)
998{
999 int ret;
1000 struct btrfs_ordered_sum *sums;
1001 LIST_HEAD(list);
1002
Yan Zheng07d400a2009-01-06 11:42:00 -05001003 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1004 bytenr + num_bytes - 1, &list);
Yan Zheng17d217f2008-12-12 10:03:38 -05001005 if (ret == 0 && list_empty(&list))
1006 return 0;
1007
1008 while (!list_empty(&list)) {
1009 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1010 list_del(&sums->list);
1011 kfree(sums);
1012 }
1013 return 1;
1014}
1015
Chris Masond352ac62008-09-29 15:18:18 -04001016/*
1017 * when nowcow writeback call back. This checks for snapshots or COW copies
1018 * of the extents that exist in the file, and COWs the file as required.
1019 *
1020 * If no cow copies or snapshots exist, we write directly to the existing
1021 * blocks on disk
1022 */
Chris Mason7f366cf2009-03-12 20:12:45 -04001023static noinline int run_delalloc_nocow(struct inode *inode,
1024 struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001025 u64 start, u64 end, int *page_started, int force,
1026 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001027{
Chris Masonbe20aa92007-12-17 20:14:01 -05001028 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001029 struct btrfs_trans_handle *trans;
Chris Masonbe20aa92007-12-17 20:14:01 -05001030 struct extent_buffer *leaf;
Chris Masonbe20aa92007-12-17 20:14:01 -05001031 struct btrfs_path *path;
Yan Zheng80ff3852008-10-30 14:20:02 -04001032 struct btrfs_file_extent_item *fi;
Chris Masonbe20aa92007-12-17 20:14:01 -05001033 struct btrfs_key found_key;
Yan Zheng80ff3852008-10-30 14:20:02 -04001034 u64 cow_start;
1035 u64 cur_offset;
1036 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001037 u64 extent_offset;
Yan Zheng80ff3852008-10-30 14:20:02 -04001038 u64 disk_bytenr;
1039 u64 num_bytes;
1040 int extent_type;
1041 int ret;
Yan Zhengd899e052008-10-30 14:25:28 -04001042 int type;
Yan Zheng80ff3852008-10-30 14:20:02 -04001043 int nocow;
1044 int check_prev = 1;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001045 bool nolock = false;
Chris Masonbe20aa92007-12-17 20:14:01 -05001046
1047 path = btrfs_alloc_path();
1048 BUG_ON(!path);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001049 if (root == root->fs_info->tree_root) {
1050 nolock = true;
1051 trans = btrfs_join_transaction_nolock(root, 1);
1052 } else {
1053 trans = btrfs_join_transaction(root, 1);
1054 }
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001055 BUG_ON(IS_ERR(trans));
Chris Masonbe20aa92007-12-17 20:14:01 -05001056
Yan Zheng80ff3852008-10-30 14:20:02 -04001057 cow_start = (u64)-1;
1058 cur_offset = start;
1059 while (1) {
1060 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1061 cur_offset, 0);
1062 BUG_ON(ret < 0);
1063 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1064 leaf = path->nodes[0];
1065 btrfs_item_key_to_cpu(leaf, &found_key,
1066 path->slots[0] - 1);
1067 if (found_key.objectid == inode->i_ino &&
1068 found_key.type == BTRFS_EXTENT_DATA_KEY)
1069 path->slots[0]--;
1070 }
1071 check_prev = 0;
1072next_slot:
1073 leaf = path->nodes[0];
1074 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1075 ret = btrfs_next_leaf(root, path);
1076 if (ret < 0)
1077 BUG_ON(1);
1078 if (ret > 0)
1079 break;
1080 leaf = path->nodes[0];
1081 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001082
Yan Zheng80ff3852008-10-30 14:20:02 -04001083 nocow = 0;
1084 disk_bytenr = 0;
Yan Zheng17d217f2008-12-12 10:03:38 -05001085 num_bytes = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001086 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05001087
Yan Zheng80ff3852008-10-30 14:20:02 -04001088 if (found_key.objectid > inode->i_ino ||
1089 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1090 found_key.offset > end)
1091 break;
Chris Masonbe20aa92007-12-17 20:14:01 -05001092
Yan Zheng80ff3852008-10-30 14:20:02 -04001093 if (found_key.offset > cur_offset) {
1094 extent_end = found_key.offset;
Chris Masone9061e22009-10-09 09:57:45 -04001095 extent_type = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001096 goto out_check;
1097 }
Chris Masonc31f8832008-01-08 15:46:31 -05001098
Yan Zheng80ff3852008-10-30 14:20:02 -04001099 fi = btrfs_item_ptr(leaf, path->slots[0],
1100 struct btrfs_file_extent_item);
1101 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Masonbe20aa92007-12-17 20:14:01 -05001102
Yan Zhengd899e052008-10-30 14:25:28 -04001103 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1104 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001105 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001106 extent_offset = btrfs_file_extent_offset(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001107 extent_end = found_key.offset +
1108 btrfs_file_extent_num_bytes(leaf, fi);
1109 if (extent_end <= start) {
1110 path->slots[0]++;
1111 goto next_slot;
1112 }
Yan Zheng17d217f2008-12-12 10:03:38 -05001113 if (disk_bytenr == 0)
1114 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001115 if (btrfs_file_extent_compression(leaf, fi) ||
1116 btrfs_file_extent_encryption(leaf, fi) ||
1117 btrfs_file_extent_other_encoding(leaf, fi))
1118 goto out_check;
Yan Zhengd899e052008-10-30 14:25:28 -04001119 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1120 goto out_check;
Yan Zhengd2fb3432008-12-11 16:30:39 -05001121 if (btrfs_extent_readonly(root, disk_bytenr))
Yan Zheng80ff3852008-10-30 14:20:02 -04001122 goto out_check;
Yan Zheng17d217f2008-12-12 10:03:38 -05001123 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001124 found_key.offset -
1125 extent_offset, disk_bytenr))
Yan Zheng17d217f2008-12-12 10:03:38 -05001126 goto out_check;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001127 disk_bytenr += extent_offset;
Yan Zheng17d217f2008-12-12 10:03:38 -05001128 disk_bytenr += cur_offset - found_key.offset;
1129 num_bytes = min(end + 1, extent_end) - cur_offset;
1130 /*
1131 * force cow if csum exists in the range.
1132 * this ensure that csum for a given extent are
1133 * either valid or do not exist.
1134 */
1135 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1136 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001137 nocow = 1;
1138 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1139 extent_end = found_key.offset +
1140 btrfs_file_extent_inline_len(leaf, fi);
1141 extent_end = ALIGN(extent_end, root->sectorsize);
1142 } else {
1143 BUG_ON(1);
1144 }
1145out_check:
1146 if (extent_end <= start) {
1147 path->slots[0]++;
1148 goto next_slot;
1149 }
1150 if (!nocow) {
1151 if (cow_start == (u64)-1)
1152 cow_start = cur_offset;
1153 cur_offset = extent_end;
1154 if (cur_offset > end)
1155 break;
1156 path->slots[0]++;
1157 goto next_slot;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001158 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001159
Yan Zheng7ea394f2008-08-05 13:05:02 -04001160 btrfs_release_path(root, path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001161 if (cow_start != (u64)-1) {
1162 ret = cow_file_range(inode, locked_page, cow_start,
Chris Mason771ed682008-11-06 22:02:51 -05001163 found_key.offset - 1, page_started,
1164 nr_written, 1);
Yan Zheng80ff3852008-10-30 14:20:02 -04001165 BUG_ON(ret);
1166 cow_start = (u64)-1;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001167 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001168
Yan Zhengd899e052008-10-30 14:25:28 -04001169 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1170 struct extent_map *em;
1171 struct extent_map_tree *em_tree;
1172 em_tree = &BTRFS_I(inode)->extent_tree;
1173 em = alloc_extent_map(GFP_NOFS);
Tsutomu Itohc26a9202011-02-14 00:45:29 +00001174 BUG_ON(!em);
Yan Zhengd899e052008-10-30 14:25:28 -04001175 em->start = cur_offset;
Chris Mason445a6942008-11-10 11:53:33 -05001176 em->orig_start = em->start;
Yan Zhengd899e052008-10-30 14:25:28 -04001177 em->len = num_bytes;
1178 em->block_len = num_bytes;
1179 em->block_start = disk_bytenr;
1180 em->bdev = root->fs_info->fs_devices->latest_bdev;
1181 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1182 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04001183 write_lock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001184 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04001185 write_unlock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001186 if (ret != -EEXIST) {
1187 free_extent_map(em);
1188 break;
1189 }
1190 btrfs_drop_extent_cache(inode, em->start,
1191 em->start + em->len - 1, 0);
1192 }
1193 type = BTRFS_ORDERED_PREALLOC;
1194 } else {
1195 type = BTRFS_ORDERED_NOCOW;
1196 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001197
1198 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
Yan Zhengd899e052008-10-30 14:25:28 -04001199 num_bytes, num_bytes, type);
1200 BUG_ON(ret);
Chris Mason771ed682008-11-06 22:02:51 -05001201
Yan, Zhengefa56462010-05-16 10:49:59 -04001202 if (root->root_key.objectid ==
1203 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1204 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1205 num_bytes);
1206 BUG_ON(ret);
1207 }
1208
Yan Zhengd899e052008-10-30 14:25:28 -04001209 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
Chris Masona791e352009-10-08 11:27:10 -04001210 cur_offset, cur_offset + num_bytes - 1,
1211 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1212 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1213 EXTENT_SET_PRIVATE2);
Yan Zheng80ff3852008-10-30 14:20:02 -04001214 cur_offset = extent_end;
1215 if (cur_offset > end)
1216 break;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001217 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001218 btrfs_release_path(root, path);
1219
1220 if (cur_offset <= end && cow_start == (u64)-1)
1221 cow_start = cur_offset;
1222 if (cow_start != (u64)-1) {
1223 ret = cow_file_range(inode, locked_page, cow_start, end,
Chris Mason771ed682008-11-06 22:02:51 -05001224 page_started, nr_written, 1);
Yan Zheng80ff3852008-10-30 14:20:02 -04001225 BUG_ON(ret);
1226 }
1227
Josef Bacik0cb59c92010-07-02 12:14:14 -04001228 if (nolock) {
1229 ret = btrfs_end_transaction_nolock(trans, root);
1230 BUG_ON(ret);
1231 } else {
1232 ret = btrfs_end_transaction(trans, root);
1233 BUG_ON(ret);
1234 }
Yan Zheng7ea394f2008-08-05 13:05:02 -04001235 btrfs_free_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001236 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05001237}
1238
Chris Masond352ac62008-09-29 15:18:18 -04001239/*
1240 * extent_io.c call back to do delayed allocation processing
1241 */
Chris Masonc8b97812008-10-29 14:49:59 -04001242static int run_delalloc_range(struct inode *inode, struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001243 u64 start, u64 end, int *page_started,
1244 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001245{
Chris Masonbe20aa92007-12-17 20:14:01 -05001246 int ret;
Chris Mason7f366cf2009-03-12 20:12:45 -04001247 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masona2135012008-06-25 16:01:30 -04001248
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001249 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
Chris Masonc8b97812008-10-29 14:49:59 -04001250 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001251 page_started, 1, nr_written);
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001252 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
Yan Zhengd899e052008-10-30 14:25:28 -04001253 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001254 page_started, 0, nr_written);
Chris Mason1e701a32010-03-11 09:42:04 -05001255 else if (!btrfs_test_opt(root, COMPRESS) &&
1256 !(BTRFS_I(inode)->force_compress))
Chris Mason7f366cf2009-03-12 20:12:45 -04001257 ret = cow_file_range(inode, locked_page, start, end,
1258 page_started, nr_written, 1);
Chris Masonbe20aa92007-12-17 20:14:01 -05001259 else
Chris Mason771ed682008-11-06 22:02:51 -05001260 ret = cow_file_range_async(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001261 page_started, nr_written);
Chris Masonb888db22007-08-27 16:49:44 -04001262 return ret;
1263}
1264
Josef Bacik9ed74f22009-09-11 16:12:44 -04001265static int btrfs_split_extent_hook(struct inode *inode,
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001266 struct extent_state *orig, u64 split)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001267{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001268 /* not delalloc, ignore it */
Josef Bacik9ed74f22009-09-11 16:12:44 -04001269 if (!(orig->state & EXTENT_DELALLOC))
1270 return 0;
1271
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001272 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001273 return 0;
1274}
1275
1276/*
1277 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1278 * extents so we can keep track of new extents that are just merged onto old
1279 * extents, such as when we are doing sequential writes, so we can properly
1280 * account for the metadata space we'll need.
1281 */
1282static int btrfs_merge_extent_hook(struct inode *inode,
1283 struct extent_state *new,
1284 struct extent_state *other)
1285{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001286 /* not delalloc, ignore it */
1287 if (!(other->state & EXTENT_DELALLOC))
1288 return 0;
1289
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001290 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001291 return 0;
1292}
1293
Chris Masond352ac62008-09-29 15:18:18 -04001294/*
1295 * extent_io.c set_bit_hook, used to track delayed allocation
1296 * bytes in this file, and to maintain the list of inodes that
1297 * have pending delalloc work to be done.
1298 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001299static int btrfs_set_bit_hook(struct inode *inode,
1300 struct extent_state *state, int *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001301{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001302
Chris Mason75eff682008-12-15 15:54:40 -05001303 /*
1304 * set_bit and clear bit hooks normally require _irqsave/restore
1305 * but in this case, we are only testeing for the DELALLOC
1306 * bit, which is only set or cleared with irqs on
1307 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001308 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001309 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001310 u64 len = state->end + 1 - state->start;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001311 int do_list = (root->root_key.objectid !=
1312 BTRFS_ROOT_TREE_OBJECTID);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001313
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001314 if (*bits & EXTENT_FIRST_DELALLOC)
1315 *bits &= ~EXTENT_FIRST_DELALLOC;
1316 else
1317 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
Josef Bacik287a0ab2010-03-19 18:07:23 +00001318
Chris Mason75eff682008-12-15 15:54:40 -05001319 spin_lock(&root->fs_info->delalloc_lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001320 BTRFS_I(inode)->delalloc_bytes += len;
1321 root->fs_info->delalloc_bytes += len;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001322 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
Chris Masonea8c2812008-08-04 23:17:27 -04001323 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1324 &root->fs_info->delalloc_inodes);
1325 }
Chris Mason75eff682008-12-15 15:54:40 -05001326 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason291d6732008-01-29 15:55:23 -05001327 }
1328 return 0;
1329}
1330
Chris Masond352ac62008-09-29 15:18:18 -04001331/*
1332 * extent_io.c clear_bit_hook, see set_bit_hook for why
1333 */
Josef Bacik9ed74f22009-09-11 16:12:44 -04001334static int btrfs_clear_bit_hook(struct inode *inode,
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001335 struct extent_state *state, int *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001336{
Chris Mason75eff682008-12-15 15:54:40 -05001337 /*
1338 * set_bit and clear bit hooks normally require _irqsave/restore
1339 * but in this case, we are only testeing for the DELALLOC
1340 * bit, which is only set or cleared with irqs on
1341 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001342 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001343 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001344 u64 len = state->end + 1 - state->start;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001345 int do_list = (root->root_key.objectid !=
1346 BTRFS_ROOT_TREE_OBJECTID);
Chris Masonbcbfce82008-04-22 13:26:47 -04001347
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001348 if (*bits & EXTENT_FIRST_DELALLOC)
1349 *bits &= ~EXTENT_FIRST_DELALLOC;
1350 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1351 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1352
1353 if (*bits & EXTENT_DO_ACCOUNTING)
1354 btrfs_delalloc_release_metadata(inode, len);
1355
Josef Bacik0cb59c92010-07-02 12:14:14 -04001356 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1357 && do_list)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001358 btrfs_free_reserved_data_space(inode, len);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001359
Chris Mason75eff682008-12-15 15:54:40 -05001360 spin_lock(&root->fs_info->delalloc_lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001361 root->fs_info->delalloc_bytes -= len;
1362 BTRFS_I(inode)->delalloc_bytes -= len;
1363
Josef Bacik0cb59c92010-07-02 12:14:14 -04001364 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
Chris Masonea8c2812008-08-04 23:17:27 -04001365 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1366 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1367 }
Chris Mason75eff682008-12-15 15:54:40 -05001368 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason291d6732008-01-29 15:55:23 -05001369 }
1370 return 0;
1371}
1372
Chris Masond352ac62008-09-29 15:18:18 -04001373/*
1374 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1375 * we don't create bios that span stripes or chunks
1376 */
Chris Mason239b14b2008-03-24 15:02:07 -04001377int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
Chris Masonc8b97812008-10-29 14:49:59 -04001378 size_t size, struct bio *bio,
1379 unsigned long bio_flags)
Chris Mason239b14b2008-03-24 15:02:07 -04001380{
1381 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1382 struct btrfs_mapping_tree *map_tree;
Chris Masona62b9402008-10-03 16:31:08 -04001383 u64 logical = (u64)bio->bi_sector << 9;
Chris Mason239b14b2008-03-24 15:02:07 -04001384 u64 length = 0;
1385 u64 map_length;
Chris Mason239b14b2008-03-24 15:02:07 -04001386 int ret;
1387
Chris Mason771ed682008-11-06 22:02:51 -05001388 if (bio_flags & EXTENT_BIO_COMPRESSED)
1389 return 0;
1390
Chris Masonf2d8d742008-04-21 10:03:05 -04001391 length = bio->bi_size;
Chris Mason239b14b2008-03-24 15:02:07 -04001392 map_tree = &root->fs_info->mapping_tree;
1393 map_length = length;
Chris Masoncea9e442008-04-09 16:28:12 -04001394 ret = btrfs_map_block(map_tree, READ, logical,
Chris Masonf1885912008-04-09 16:28:12 -04001395 &map_length, NULL, 0);
Chris Masoncea9e442008-04-09 16:28:12 -04001396
Chris Masond3977122009-01-05 21:25:51 -05001397 if (map_length < length + size)
Chris Mason239b14b2008-03-24 15:02:07 -04001398 return 1;
Andi Kleen411fc6b2010-10-29 15:14:31 -04001399 return ret;
Chris Mason239b14b2008-03-24 15:02:07 -04001400}
1401
Chris Masond352ac62008-09-29 15:18:18 -04001402/*
1403 * in order to insert checksums into the metadata in large chunks,
1404 * we wait until bio submission time. All the pages in the bio are
1405 * checksummed and sums are attached onto the ordered extent record.
1406 *
1407 * At IO completion time the cums attached on the ordered extent record
1408 * are inserted into the btree
1409 */
Chris Masond3977122009-01-05 21:25:51 -05001410static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1411 struct bio *bio, int mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001412 unsigned long bio_flags,
1413 u64 bio_offset)
Chris Mason065631f2008-02-20 12:07:25 -05001414{
Chris Mason065631f2008-02-20 12:07:25 -05001415 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason065631f2008-02-20 12:07:25 -05001416 int ret = 0;
Chris Masone0156402008-04-16 11:15:20 -04001417
Chris Masond20f7042008-12-08 16:58:54 -05001418 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
Chris Mason44b8bd72008-04-16 11:14:51 -04001419 BUG_ON(ret);
Chris Mason4a69a412008-11-06 22:03:00 -05001420 return 0;
1421}
Chris Masone0156402008-04-16 11:15:20 -04001422
Chris Mason4a69a412008-11-06 22:03:00 -05001423/*
1424 * in order to insert checksums into the metadata in large chunks,
1425 * we wait until bio submission time. All the pages in the bio are
1426 * checksummed and sums are attached onto the ordered extent record.
1427 *
1428 * At IO completion time the cums attached on the ordered extent record
1429 * are inserted into the btree
1430 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001431static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001432 int mirror_num, unsigned long bio_flags,
1433 u64 bio_offset)
Chris Mason4a69a412008-11-06 22:03:00 -05001434{
1435 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason8b712842008-06-11 16:50:36 -04001436 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
Chris Mason44b8bd72008-04-16 11:14:51 -04001437}
1438
Chris Masond352ac62008-09-29 15:18:18 -04001439/*
Chris Masoncad321a2008-12-17 14:51:42 -05001440 * extent_io.c submission hook. This does the right thing for csum calculation
1441 * on write, or reading the csums from the tree before a read
Chris Masond352ac62008-09-29 15:18:18 -04001442 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001443static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001444 int mirror_num, unsigned long bio_flags,
1445 u64 bio_offset)
Chris Mason44b8bd72008-04-16 11:14:51 -04001446{
1447 struct btrfs_root *root = BTRFS_I(inode)->root;
1448 int ret = 0;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001449 int skip_sum;
Chris Mason44b8bd72008-04-16 11:14:51 -04001450
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001451 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masoncad321a2008-12-17 14:51:42 -05001452
Josef Bacik0cb59c92010-07-02 12:14:14 -04001453 if (root == root->fs_info->tree_root)
1454 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1455 else
1456 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001457 BUG_ON(ret);
Chris Mason065631f2008-02-20 12:07:25 -05001458
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001459 if (!(rw & REQ_WRITE)) {
Chris Masond20f7042008-12-08 16:58:54 -05001460 if (bio_flags & EXTENT_BIO_COMPRESSED) {
Chris Masonc8b97812008-10-29 14:49:59 -04001461 return btrfs_submit_compressed_read(inode, bio,
1462 mirror_num, bio_flags);
Chris Masond20f7042008-12-08 16:58:54 -05001463 } else if (!skip_sum)
1464 btrfs_lookup_bio_sums(root, inode, bio, NULL);
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001465 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001466 } else if (!skip_sum) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001467 /* csum items have already been cloned */
1468 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1469 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001470 /* we're doing a write, do the async checksumming */
1471 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
Chris Mason44b8bd72008-04-16 11:14:51 -04001472 inode, rw, bio, mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001473 bio_flags, bio_offset,
1474 __btrfs_submit_bio_start,
Chris Mason4a69a412008-11-06 22:03:00 -05001475 __btrfs_submit_bio_done);
Chris Mason19b9bdb2008-10-30 14:23:13 -04001476 }
1477
Chris Mason0b86a832008-03-24 15:01:56 -04001478mapit:
Chris Mason8b712842008-06-11 16:50:36 -04001479 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
Chris Mason065631f2008-02-20 12:07:25 -05001480}
Chris Mason6885f302008-02-20 16:11:05 -05001481
Chris Masond352ac62008-09-29 15:18:18 -04001482/*
1483 * given a list of ordered sums record them in the inode. This happens
1484 * at IO completion time based on sums calculated at bio submission time.
1485 */
Chris Masonba1da2f2008-07-17 12:54:15 -04001486static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
Chris Masone6dcd2d2008-07-17 12:53:50 -04001487 struct inode *inode, u64 file_offset,
1488 struct list_head *list)
1489{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001490 struct btrfs_ordered_sum *sum;
1491
1492 btrfs_set_trans_block_group(trans, inode);
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001493
1494 list_for_each_entry(sum, list, list) {
Chris Masond20f7042008-12-08 16:58:54 -05001495 btrfs_csum_file_blocks(trans,
1496 BTRFS_I(inode)->root->fs_info->csum_root, sum);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001497 }
1498 return 0;
1499}
1500
Josef Bacik2ac55d42010-02-03 19:33:23 +00001501int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1502 struct extent_state **cached_state)
Chris Masonea8c2812008-08-04 23:17:27 -04001503{
Chris Masond3977122009-01-05 21:25:51 -05001504 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
Chris Mason771ed682008-11-06 22:02:51 -05001505 WARN_ON(1);
Chris Masonea8c2812008-08-04 23:17:27 -04001506 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00001507 cached_state, GFP_NOFS);
Chris Masonea8c2812008-08-04 23:17:27 -04001508}
1509
Chris Masond352ac62008-09-29 15:18:18 -04001510/* see btrfs_writepage_start_hook for details on why this is required */
Chris Mason247e7432008-07-17 12:53:51 -04001511struct btrfs_writepage_fixup {
1512 struct page *page;
1513 struct btrfs_work work;
1514};
1515
Christoph Hellwigb2950862008-12-02 09:54:17 -05001516static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
Chris Mason247e7432008-07-17 12:53:51 -04001517{
1518 struct btrfs_writepage_fixup *fixup;
1519 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001520 struct extent_state *cached_state = NULL;
Chris Mason247e7432008-07-17 12:53:51 -04001521 struct page *page;
1522 struct inode *inode;
1523 u64 page_start;
1524 u64 page_end;
1525
1526 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1527 page = fixup->page;
Chris Mason4a096752008-07-21 10:29:44 -04001528again:
Chris Mason247e7432008-07-17 12:53:51 -04001529 lock_page(page);
1530 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1531 ClearPageChecked(page);
1532 goto out_page;
1533 }
1534
1535 inode = page->mapping->host;
1536 page_start = page_offset(page);
1537 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1538
Josef Bacik2ac55d42010-02-03 19:33:23 +00001539 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1540 &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001541
1542 /* already ordered? We're done */
Chris Mason8b62b722009-09-02 16:53:46 -04001543 if (PagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001544 goto out;
Chris Mason4a096752008-07-21 10:29:44 -04001545
1546 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1547 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00001548 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1549 page_end, &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001550 unlock_page(page);
1551 btrfs_start_ordered_extent(inode, ordered, 1);
1552 goto again;
1553 }
Chris Mason247e7432008-07-17 12:53:51 -04001554
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001555 BUG();
Josef Bacik2ac55d42010-02-03 19:33:23 +00001556 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
Chris Mason247e7432008-07-17 12:53:51 -04001557 ClearPageChecked(page);
1558out:
Josef Bacik2ac55d42010-02-03 19:33:23 +00001559 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1560 &cached_state, GFP_NOFS);
Chris Mason247e7432008-07-17 12:53:51 -04001561out_page:
1562 unlock_page(page);
1563 page_cache_release(page);
Miao Xieb897abe2011-01-26 16:19:22 +08001564 kfree(fixup);
Chris Mason247e7432008-07-17 12:53:51 -04001565}
1566
1567/*
1568 * There are a few paths in the higher layers of the kernel that directly
1569 * set the page dirty bit without asking the filesystem if it is a
1570 * good idea. This causes problems because we want to make sure COW
1571 * properly happens and the data=ordered rules are followed.
1572 *
Chris Masonc8b97812008-10-29 14:49:59 -04001573 * In our case any range that doesn't have the ORDERED bit set
Chris Mason247e7432008-07-17 12:53:51 -04001574 * hasn't been properly setup for IO. We kick off an async process
1575 * to fix it up. The async helper will wait for ordered extents, set
1576 * the delalloc bit and make it safe to write the page.
1577 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001578static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
Chris Mason247e7432008-07-17 12:53:51 -04001579{
1580 struct inode *inode = page->mapping->host;
1581 struct btrfs_writepage_fixup *fixup;
1582 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason247e7432008-07-17 12:53:51 -04001583
Chris Mason8b62b722009-09-02 16:53:46 -04001584 /* this page is properly in the ordered list */
1585 if (TestClearPagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001586 return 0;
1587
1588 if (PageChecked(page))
1589 return -EAGAIN;
1590
1591 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1592 if (!fixup)
1593 return -EAGAIN;
Chris Masonf4219502008-07-22 11:18:09 -04001594
Chris Mason247e7432008-07-17 12:53:51 -04001595 SetPageChecked(page);
1596 page_cache_get(page);
1597 fixup->work.func = btrfs_writepage_fixup_worker;
1598 fixup->page = page;
1599 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1600 return -EAGAIN;
1601}
1602
Yan Zhengd899e052008-10-30 14:25:28 -04001603static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1604 struct inode *inode, u64 file_pos,
1605 u64 disk_bytenr, u64 disk_num_bytes,
1606 u64 num_bytes, u64 ram_bytes,
1607 u8 compression, u8 encryption,
1608 u16 other_encoding, int extent_type)
1609{
1610 struct btrfs_root *root = BTRFS_I(inode)->root;
1611 struct btrfs_file_extent_item *fi;
1612 struct btrfs_path *path;
1613 struct extent_buffer *leaf;
1614 struct btrfs_key ins;
1615 u64 hint;
1616 int ret;
1617
1618 path = btrfs_alloc_path();
1619 BUG_ON(!path);
1620
Chris Masonb9473432009-03-13 11:00:37 -04001621 path->leave_spinning = 1;
Chris Masona1ed8352009-09-11 12:27:37 -04001622
1623 /*
1624 * we may be replacing one extent in the tree with another.
1625 * The new extent is pinned in the extent map, and we don't want
1626 * to drop it from the cache until it is completely in the btree.
1627 *
1628 * So, tell btrfs_drop_extents to leave this extent in the cache.
1629 * the caller is expected to unpin it and allow it to be merged
1630 * with the others.
1631 */
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001632 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1633 &hint, 0);
Yan Zhengd899e052008-10-30 14:25:28 -04001634 BUG_ON(ret);
1635
1636 ins.objectid = inode->i_ino;
1637 ins.offset = file_pos;
1638 ins.type = BTRFS_EXTENT_DATA_KEY;
1639 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1640 BUG_ON(ret);
1641 leaf = path->nodes[0];
1642 fi = btrfs_item_ptr(leaf, path->slots[0],
1643 struct btrfs_file_extent_item);
1644 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1645 btrfs_set_file_extent_type(leaf, fi, extent_type);
1646 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1647 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1648 btrfs_set_file_extent_offset(leaf, fi, 0);
1649 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1650 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1651 btrfs_set_file_extent_compression(leaf, fi, compression);
1652 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1653 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
Chris Masonb9473432009-03-13 11:00:37 -04001654
1655 btrfs_unlock_up_safe(path, 1);
1656 btrfs_set_lock_blocking(leaf);
1657
Yan Zhengd899e052008-10-30 14:25:28 -04001658 btrfs_mark_buffer_dirty(leaf);
1659
1660 inode_add_bytes(inode, num_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -04001661
1662 ins.objectid = disk_bytenr;
1663 ins.offset = disk_num_bytes;
1664 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001665 ret = btrfs_alloc_reserved_file_extent(trans, root,
1666 root->root_key.objectid,
1667 inode->i_ino, file_pos, &ins);
Yan Zhengd899e052008-10-30 14:25:28 -04001668 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -04001669 btrfs_free_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04001670
Yan Zhengd899e052008-10-30 14:25:28 -04001671 return 0;
1672}
1673
Chris Mason5d13a982009-03-13 11:41:46 -04001674/*
1675 * helper function for btrfs_finish_ordered_io, this
1676 * just reads in some of the csum leaves to prime them into ram
1677 * before we start the transaction. It limits the amount of btree
1678 * reads required while inside the transaction.
1679 */
Chris Masond352ac62008-09-29 15:18:18 -04001680/* as ordered data IO finishes, this gets called so we can finish
1681 * an ordered extent if the range of bytes in the file it covers are
1682 * fully written.
1683 */
Chris Mason211f90e2008-07-18 11:56:15 -04001684static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001685{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001686 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001687 struct btrfs_trans_handle *trans = NULL;
Chris Mason5d13a982009-03-13 11:41:46 -04001688 struct btrfs_ordered_extent *ordered_extent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001689 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001690 struct extent_state *cached_state = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08001691 int compress_type = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001692 int ret;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001693 bool nolock = false;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001694
Josef Bacik5a1a3df2010-02-02 20:51:14 +00001695 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1696 end - start + 1);
Chris Masonba1da2f2008-07-17 12:54:15 -04001697 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001698 return 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001699 BUG_ON(!ordered_extent);
Josef Bacikefd049f2010-02-02 20:50:10 +00001700
Josef Bacik0cb59c92010-07-02 12:14:14 -04001701 nolock = (root == root->fs_info->tree_root);
1702
Yan, Zhengc2167752009-11-12 09:34:21 +00001703 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1704 BUG_ON(!list_empty(&ordered_extent->list));
1705 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1706 if (!ret) {
Josef Bacik0cb59c92010-07-02 12:14:14 -04001707 if (nolock)
1708 trans = btrfs_join_transaction_nolock(root, 1);
1709 else
1710 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001711 BUG_ON(IS_ERR(trans));
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001712 btrfs_set_trans_block_group(trans, inode);
1713 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Yan, Zhengc2167752009-11-12 09:34:21 +00001714 ret = btrfs_update_inode(trans, root, inode);
1715 BUG_ON(ret);
Yan, Zhengc2167752009-11-12 09:34:21 +00001716 }
1717 goto out;
1718 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04001719
Josef Bacik2ac55d42010-02-03 19:33:23 +00001720 lock_extent_bits(io_tree, ordered_extent->file_offset,
1721 ordered_extent->file_offset + ordered_extent->len - 1,
1722 0, &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001723
Josef Bacik0cb59c92010-07-02 12:14:14 -04001724 if (nolock)
1725 trans = btrfs_join_transaction_nolock(root, 1);
1726 else
1727 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001728 BUG_ON(IS_ERR(trans));
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001729 btrfs_set_trans_block_group(trans, inode);
1730 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Yan, Zhengc2167752009-11-12 09:34:21 +00001731
Chris Masonc8b97812008-10-29 14:49:59 -04001732 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
Li Zefan261507a02010-12-17 14:21:50 +08001733 compress_type = ordered_extent->compress_type;
Yan Zhengd899e052008-10-30 14:25:28 -04001734 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
Li Zefan261507a02010-12-17 14:21:50 +08001735 BUG_ON(compress_type);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001736 ret = btrfs_mark_extent_written(trans, inode,
Yan Zhengd899e052008-10-30 14:25:28 -04001737 ordered_extent->file_offset,
1738 ordered_extent->file_offset +
1739 ordered_extent->len);
1740 BUG_ON(ret);
1741 } else {
Josef Bacik0af3d002010-06-21 14:48:16 -04001742 BUG_ON(root == root->fs_info->tree_root);
Yan Zhengd899e052008-10-30 14:25:28 -04001743 ret = insert_reserved_file_extent(trans, inode,
1744 ordered_extent->file_offset,
1745 ordered_extent->start,
1746 ordered_extent->disk_len,
1747 ordered_extent->len,
1748 ordered_extent->len,
Li Zefan261507a02010-12-17 14:21:50 +08001749 compress_type, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04001750 BTRFS_FILE_EXTENT_REG);
Chris Masona1ed8352009-09-11 12:27:37 -04001751 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1752 ordered_extent->file_offset,
1753 ordered_extent->len);
Yan Zhengd899e052008-10-30 14:25:28 -04001754 BUG_ON(ret);
1755 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00001756 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1757 ordered_extent->file_offset +
1758 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1759
Chris Masone6dcd2d2008-07-17 12:53:50 -04001760 add_pending_csums(trans, inode, ordered_extent->file_offset,
1761 &ordered_extent->list);
1762
Yan, Zhengc2167752009-11-12 09:34:21 +00001763 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1764 ret = btrfs_update_inode(trans, root, inode);
1765 BUG_ON(ret);
Yan, Zhengc2167752009-11-12 09:34:21 +00001766out:
Josef Bacik0cb59c92010-07-02 12:14:14 -04001767 if (nolock) {
1768 if (trans)
1769 btrfs_end_transaction_nolock(trans, root);
1770 } else {
1771 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1772 if (trans)
1773 btrfs_end_transaction(trans, root);
1774 }
1775
Chris Masone6dcd2d2008-07-17 12:53:50 -04001776 /* once for us */
1777 btrfs_put_ordered_extent(ordered_extent);
1778 /* once for the tree */
1779 btrfs_put_ordered_extent(ordered_extent);
1780
Chris Masone6dcd2d2008-07-17 12:53:50 -04001781 return 0;
1782}
1783
Christoph Hellwigb2950862008-12-02 09:54:17 -05001784static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason211f90e2008-07-18 11:56:15 -04001785 struct extent_state *state, int uptodate)
1786{
Chris Mason8b62b722009-09-02 16:53:46 -04001787 ClearPagePrivate2(page);
Chris Mason211f90e2008-07-18 11:56:15 -04001788 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1789}
1790
Chris Masond352ac62008-09-29 15:18:18 -04001791/*
1792 * When IO fails, either with EIO or csum verification fails, we
1793 * try other mirrors that might have a good copy of the data. This
1794 * io_failure_record is used to record state as we go through all the
1795 * mirrors. If another mirror has good data, the page is set up to date
1796 * and things continue. If a good mirror can't be found, the original
1797 * bio end_io callback is called to indicate things have failed.
1798 */
Chris Mason7e383262008-04-09 16:28:12 -04001799struct io_failure_record {
1800 struct page *page;
1801 u64 start;
1802 u64 len;
1803 u64 logical;
Chris Masond20f7042008-12-08 16:58:54 -05001804 unsigned long bio_flags;
Chris Mason7e383262008-04-09 16:28:12 -04001805 int last_mirror;
1806};
1807
Christoph Hellwigb2950862008-12-02 09:54:17 -05001808static int btrfs_io_failed_hook(struct bio *failed_bio,
Chris Mason1259ab72008-05-12 13:39:03 -04001809 struct page *page, u64 start, u64 end,
1810 struct extent_state *state)
Chris Mason7e383262008-04-09 16:28:12 -04001811{
1812 struct io_failure_record *failrec = NULL;
1813 u64 private;
1814 struct extent_map *em;
1815 struct inode *inode = page->mapping->host;
1816 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
Chris Mason3b951512008-04-17 11:29:12 -04001817 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Mason7e383262008-04-09 16:28:12 -04001818 struct bio *bio;
1819 int num_copies;
1820 int ret;
Chris Mason1259ab72008-05-12 13:39:03 -04001821 int rw;
Chris Mason7e383262008-04-09 16:28:12 -04001822 u64 logical;
1823
1824 ret = get_state_private(failure_tree, start, &private);
1825 if (ret) {
Chris Mason7e383262008-04-09 16:28:12 -04001826 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1827 if (!failrec)
1828 return -ENOMEM;
1829 failrec->start = start;
1830 failrec->len = end - start + 1;
1831 failrec->last_mirror = 0;
Chris Masond20f7042008-12-08 16:58:54 -05001832 failrec->bio_flags = 0;
Chris Mason7e383262008-04-09 16:28:12 -04001833
Chris Mason890871b2009-09-02 16:24:52 -04001834 read_lock(&em_tree->lock);
Chris Mason3b951512008-04-17 11:29:12 -04001835 em = lookup_extent_mapping(em_tree, start, failrec->len);
1836 if (em->start > start || em->start + em->len < start) {
1837 free_extent_map(em);
1838 em = NULL;
1839 }
Chris Mason890871b2009-09-02 16:24:52 -04001840 read_unlock(&em_tree->lock);
Chris Mason7e383262008-04-09 16:28:12 -04001841
1842 if (!em || IS_ERR(em)) {
1843 kfree(failrec);
1844 return -EIO;
1845 }
1846 logical = start - em->start;
1847 logical = em->block_start + logical;
Chris Masond20f7042008-12-08 16:58:54 -05001848 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1849 logical = em->block_start;
1850 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
Li Zefan261507a02010-12-17 14:21:50 +08001851 extent_set_compress_type(&failrec->bio_flags,
1852 em->compress_type);
Chris Masond20f7042008-12-08 16:58:54 -05001853 }
Chris Mason7e383262008-04-09 16:28:12 -04001854 failrec->logical = logical;
1855 free_extent_map(em);
1856 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1857 EXTENT_DIRTY, GFP_NOFS);
Chris Mason587f7702008-04-11 12:16:46 -04001858 set_state_private(failure_tree, start,
1859 (u64)(unsigned long)failrec);
Chris Mason7e383262008-04-09 16:28:12 -04001860 } else {
Chris Mason587f7702008-04-11 12:16:46 -04001861 failrec = (struct io_failure_record *)(unsigned long)private;
Chris Mason7e383262008-04-09 16:28:12 -04001862 }
1863 num_copies = btrfs_num_copies(
1864 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1865 failrec->logical, failrec->len);
1866 failrec->last_mirror++;
1867 if (!state) {
Chris Masoncad321a2008-12-17 14:51:42 -05001868 spin_lock(&BTRFS_I(inode)->io_tree.lock);
Chris Mason7e383262008-04-09 16:28:12 -04001869 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1870 failrec->start,
1871 EXTENT_LOCKED);
1872 if (state && state->start != failrec->start)
1873 state = NULL;
Chris Masoncad321a2008-12-17 14:51:42 -05001874 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
Chris Mason7e383262008-04-09 16:28:12 -04001875 }
1876 if (!state || failrec->last_mirror > num_copies) {
1877 set_state_private(failure_tree, failrec->start, 0);
1878 clear_extent_bits(failure_tree, failrec->start,
1879 failrec->start + failrec->len - 1,
1880 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1881 kfree(failrec);
1882 return -EIO;
1883 }
1884 bio = bio_alloc(GFP_NOFS, 1);
1885 bio->bi_private = state;
1886 bio->bi_end_io = failed_bio->bi_end_io;
1887 bio->bi_sector = failrec->logical >> 9;
1888 bio->bi_bdev = failed_bio->bi_bdev;
Chris Masone1c4b742008-04-22 13:26:46 -04001889 bio->bi_size = 0;
Chris Masond20f7042008-12-08 16:58:54 -05001890
Chris Mason7e383262008-04-09 16:28:12 -04001891 bio_add_page(bio, page, failrec->len, start - page_offset(page));
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001892 if (failed_bio->bi_rw & REQ_WRITE)
Chris Mason1259ab72008-05-12 13:39:03 -04001893 rw = WRITE;
1894 else
1895 rw = READ;
1896
1897 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
Chris Masonc8b97812008-10-29 14:49:59 -04001898 failrec->last_mirror,
Chris Masoneaf25d92010-05-25 09:48:28 -04001899 failrec->bio_flags, 0);
Chris Mason1259ab72008-05-12 13:39:03 -04001900 return 0;
1901}
1902
Chris Masond352ac62008-09-29 15:18:18 -04001903/*
1904 * each time an IO finishes, we do a fast check in the IO failure tree
1905 * to see if we need to process or clean up an io_failure_record
1906 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001907static int btrfs_clean_io_failures(struct inode *inode, u64 start)
Chris Mason1259ab72008-05-12 13:39:03 -04001908{
1909 u64 private;
1910 u64 private_failure;
1911 struct io_failure_record *failure;
1912 int ret;
1913
1914 private = 0;
1915 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
Chris Masonec29ed52011-02-23 16:23:20 -05001916 (u64)-1, 1, EXTENT_DIRTY, 0)) {
Chris Mason1259ab72008-05-12 13:39:03 -04001917 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1918 start, &private_failure);
1919 if (ret == 0) {
1920 failure = (struct io_failure_record *)(unsigned long)
1921 private_failure;
1922 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1923 failure->start, 0);
1924 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1925 failure->start,
1926 failure->start + failure->len - 1,
1927 EXTENT_DIRTY | EXTENT_LOCKED,
1928 GFP_NOFS);
1929 kfree(failure);
1930 }
1931 }
Chris Mason7e383262008-04-09 16:28:12 -04001932 return 0;
1933}
1934
Chris Masond352ac62008-09-29 15:18:18 -04001935/*
1936 * when reads are done, we need to check csums to verify the data is correct
1937 * if there's a match, we allow the bio to finish. If not, we go through
1938 * the io_failure_record routines to find good copies
1939 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001940static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason70dec802008-01-29 09:59:12 -05001941 struct extent_state *state)
Chris Mason07157aa2007-08-30 08:50:51 -04001942{
Chris Mason35ebb932007-10-30 16:56:53 -04001943 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
Chris Mason07157aa2007-08-30 08:50:51 -04001944 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05001945 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason07157aa2007-08-30 08:50:51 -04001946 char *kaddr;
Chris Masonaadfeb62008-01-29 09:10:27 -05001947 u64 private = ~(u32)0;
Chris Mason07157aa2007-08-30 08:50:51 -04001948 int ret;
Chris Masonff79f812007-10-15 16:22:25 -04001949 struct btrfs_root *root = BTRFS_I(inode)->root;
1950 u32 csum = ~(u32)0;
Chris Masond1310b22008-01-24 16:13:08 -05001951
Chris Masond20f7042008-12-08 16:58:54 -05001952 if (PageChecked(page)) {
1953 ClearPageChecked(page);
1954 goto good;
1955 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001956
1957 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Chris Masonb6cda9b2007-12-14 15:30:32 -05001958 return 0;
Chris Masond20f7042008-12-08 16:58:54 -05001959
Yan Zheng17d217f2008-12-12 10:03:38 -05001960 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
Chris Mason9655d292009-09-02 15:22:30 -04001961 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001962 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1963 GFP_NOFS);
1964 return 0;
1965 }
1966
Yanc2e639f2008-02-04 08:57:25 -05001967 if (state && state->start == start) {
Chris Mason70dec802008-01-29 09:59:12 -05001968 private = state->private;
1969 ret = 0;
1970 } else {
1971 ret = get_state_private(io_tree, start, &private);
1972 }
Chris Mason9ab86c82009-01-07 09:48:51 -05001973 kaddr = kmap_atomic(page, KM_USER0);
Chris Masond3977122009-01-05 21:25:51 -05001974 if (ret)
Chris Mason07157aa2007-08-30 08:50:51 -04001975 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05001976
Chris Masonff79f812007-10-15 16:22:25 -04001977 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1978 btrfs_csum_final(csum, (char *)&csum);
Chris Masond3977122009-01-05 21:25:51 -05001979 if (csum != private)
Chris Mason07157aa2007-08-30 08:50:51 -04001980 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05001981
Chris Mason9ab86c82009-01-07 09:48:51 -05001982 kunmap_atomic(kaddr, KM_USER0);
Chris Masond20f7042008-12-08 16:58:54 -05001983good:
Chris Mason7e383262008-04-09 16:28:12 -04001984 /* if the io failure tree for this inode is non-empty,
1985 * check to see if we've recovered from a failed IO
1986 */
Chris Mason1259ab72008-05-12 13:39:03 -04001987 btrfs_clean_io_failures(inode, start);
Chris Mason07157aa2007-08-30 08:50:51 -04001988 return 0;
1989
1990zeroit:
Chris Mason193f2842009-04-27 07:29:05 -04001991 if (printk_ratelimit()) {
1992 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1993 "private %llu\n", page->mapping->host->i_ino,
1994 (unsigned long long)start, csum,
1995 (unsigned long long)private);
1996 }
Chris Masondb945352007-10-15 16:15:53 -04001997 memset(kaddr + offset, 1, end - start + 1);
1998 flush_dcache_page(page);
Chris Mason9ab86c82009-01-07 09:48:51 -05001999 kunmap_atomic(kaddr, KM_USER0);
Chris Mason3b951512008-04-17 11:29:12 -04002000 if (private == 0)
2001 return 0;
Chris Mason7e383262008-04-09 16:28:12 -04002002 return -EIO;
Chris Mason07157aa2007-08-30 08:50:51 -04002003}
Chris Masonb888db22007-08-27 16:49:44 -04002004
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002005struct delayed_iput {
2006 struct list_head list;
2007 struct inode *inode;
2008};
2009
2010void btrfs_add_delayed_iput(struct inode *inode)
2011{
2012 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2013 struct delayed_iput *delayed;
2014
2015 if (atomic_add_unless(&inode->i_count, -1, 1))
2016 return;
2017
2018 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2019 delayed->inode = inode;
2020
2021 spin_lock(&fs_info->delayed_iput_lock);
2022 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2023 spin_unlock(&fs_info->delayed_iput_lock);
2024}
2025
2026void btrfs_run_delayed_iputs(struct btrfs_root *root)
2027{
2028 LIST_HEAD(list);
2029 struct btrfs_fs_info *fs_info = root->fs_info;
2030 struct delayed_iput *delayed;
2031 int empty;
2032
2033 spin_lock(&fs_info->delayed_iput_lock);
2034 empty = list_empty(&fs_info->delayed_iputs);
2035 spin_unlock(&fs_info->delayed_iput_lock);
2036 if (empty)
2037 return;
2038
2039 down_read(&root->fs_info->cleanup_work_sem);
2040 spin_lock(&fs_info->delayed_iput_lock);
2041 list_splice_init(&fs_info->delayed_iputs, &list);
2042 spin_unlock(&fs_info->delayed_iput_lock);
2043
2044 while (!list_empty(&list)) {
2045 delayed = list_entry(list.next, struct delayed_iput, list);
2046 list_del(&delayed->list);
2047 iput(delayed->inode);
2048 kfree(delayed);
2049 }
2050 up_read(&root->fs_info->cleanup_work_sem);
2051}
2052
Josef Bacik7b128762008-07-24 12:17:14 -04002053/*
Yan, Zhengd68fc572010-05-16 10:49:58 -04002054 * calculate extra metadata reservation when snapshotting a subvolume
2055 * contains orphan files.
2056 */
2057void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2058 struct btrfs_pending_snapshot *pending,
2059 u64 *bytes_to_reserve)
2060{
2061 struct btrfs_root *root;
2062 struct btrfs_block_rsv *block_rsv;
2063 u64 num_bytes;
2064 int index;
2065
2066 root = pending->root;
2067 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2068 return;
2069
2070 block_rsv = root->orphan_block_rsv;
2071
2072 /* orphan block reservation for the snapshot */
2073 num_bytes = block_rsv->size;
2074
2075 /*
2076 * after the snapshot is created, COWing tree blocks may use more
2077 * space than it frees. So we should make sure there is enough
2078 * reserved space.
2079 */
2080 index = trans->transid & 0x1;
2081 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2082 num_bytes += block_rsv->size -
2083 (block_rsv->reserved + block_rsv->freed[index]);
2084 }
2085
2086 *bytes_to_reserve += num_bytes;
2087}
2088
2089void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2090 struct btrfs_pending_snapshot *pending)
2091{
2092 struct btrfs_root *root = pending->root;
2093 struct btrfs_root *snap = pending->snap;
2094 struct btrfs_block_rsv *block_rsv;
2095 u64 num_bytes;
2096 int index;
2097 int ret;
2098
2099 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2100 return;
2101
2102 /* refill source subvolume's orphan block reservation */
2103 block_rsv = root->orphan_block_rsv;
2104 index = trans->transid & 0x1;
2105 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2106 num_bytes = block_rsv->size -
2107 (block_rsv->reserved + block_rsv->freed[index]);
2108 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2109 root->orphan_block_rsv,
2110 num_bytes);
2111 BUG_ON(ret);
2112 }
2113
2114 /* setup orphan block reservation for the snapshot */
2115 block_rsv = btrfs_alloc_block_rsv(snap);
2116 BUG_ON(!block_rsv);
2117
2118 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2119 snap->orphan_block_rsv = block_rsv;
2120
2121 num_bytes = root->orphan_block_rsv->size;
2122 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2123 block_rsv, num_bytes);
2124 BUG_ON(ret);
2125
2126#if 0
2127 /* insert orphan item for the snapshot */
2128 WARN_ON(!root->orphan_item_inserted);
2129 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2130 snap->root_key.objectid);
2131 BUG_ON(ret);
2132 snap->orphan_item_inserted = 1;
2133#endif
2134}
2135
2136enum btrfs_orphan_cleanup_state {
2137 ORPHAN_CLEANUP_STARTED = 1,
2138 ORPHAN_CLEANUP_DONE = 2,
2139};
2140
2141/*
2142 * This is called in transaction commmit time. If there are no orphan
2143 * files in the subvolume, it removes orphan item and frees block_rsv
2144 * structure.
2145 */
2146void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2147 struct btrfs_root *root)
2148{
2149 int ret;
2150
2151 if (!list_empty(&root->orphan_list) ||
2152 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2153 return;
2154
2155 if (root->orphan_item_inserted &&
2156 btrfs_root_refs(&root->root_item) > 0) {
2157 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2158 root->root_key.objectid);
2159 BUG_ON(ret);
2160 root->orphan_item_inserted = 0;
2161 }
2162
2163 if (root->orphan_block_rsv) {
2164 WARN_ON(root->orphan_block_rsv->size > 0);
2165 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2166 root->orphan_block_rsv = NULL;
2167 }
2168}
2169
2170/*
Josef Bacik7b128762008-07-24 12:17:14 -04002171 * This creates an orphan entry for the given inode in case something goes
2172 * wrong in the middle of an unlink/truncate.
Yan, Zhengd68fc572010-05-16 10:49:58 -04002173 *
2174 * NOTE: caller of this function should reserve 5 units of metadata for
2175 * this function.
Josef Bacik7b128762008-07-24 12:17:14 -04002176 */
2177int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2178{
2179 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002180 struct btrfs_block_rsv *block_rsv = NULL;
2181 int reserve = 0;
2182 int insert = 0;
2183 int ret;
Josef Bacik7b128762008-07-24 12:17:14 -04002184
Yan, Zhengd68fc572010-05-16 10:49:58 -04002185 if (!root->orphan_block_rsv) {
2186 block_rsv = btrfs_alloc_block_rsv(root);
2187 BUG_ON(!block_rsv);
Josef Bacik7b128762008-07-24 12:17:14 -04002188 }
2189
Yan, Zhengd68fc572010-05-16 10:49:58 -04002190 spin_lock(&root->orphan_lock);
2191 if (!root->orphan_block_rsv) {
2192 root->orphan_block_rsv = block_rsv;
2193 } else if (block_rsv) {
2194 btrfs_free_block_rsv(root, block_rsv);
2195 block_rsv = NULL;
2196 }
Josef Bacik7b128762008-07-24 12:17:14 -04002197
Yan, Zhengd68fc572010-05-16 10:49:58 -04002198 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2199 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2200#if 0
2201 /*
2202 * For proper ENOSPC handling, we should do orphan
2203 * cleanup when mounting. But this introduces backward
2204 * compatibility issue.
2205 */
2206 if (!xchg(&root->orphan_item_inserted, 1))
2207 insert = 2;
2208 else
2209 insert = 1;
2210#endif
2211 insert = 1;
2212 } else {
2213 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2214 }
Josef Bacik7b128762008-07-24 12:17:14 -04002215
Yan, Zhengd68fc572010-05-16 10:49:58 -04002216 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2217 BTRFS_I(inode)->orphan_meta_reserved = 1;
2218 reserve = 1;
2219 }
2220 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002221
Yan, Zhengd68fc572010-05-16 10:49:58 -04002222 if (block_rsv)
2223 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2224
2225 /* grab metadata reservation from transaction handle */
2226 if (reserve) {
2227 ret = btrfs_orphan_reserve_metadata(trans, inode);
2228 BUG_ON(ret);
2229 }
2230
2231 /* insert an orphan item to track this unlinked/truncated file */
2232 if (insert >= 1) {
2233 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2234 BUG_ON(ret);
2235 }
2236
2237 /* insert an orphan item to track subvolume contains orphan files */
2238 if (insert >= 2) {
2239 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2240 root->root_key.objectid);
2241 BUG_ON(ret);
2242 }
2243 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002244}
2245
2246/*
2247 * We have done the truncate/delete so we can go ahead and remove the orphan
2248 * item for this particular inode.
2249 */
2250int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2251{
2252 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002253 int delete_item = 0;
2254 int release_rsv = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002255 int ret = 0;
2256
Yan, Zhengd68fc572010-05-16 10:49:58 -04002257 spin_lock(&root->orphan_lock);
2258 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2259 list_del_init(&BTRFS_I(inode)->i_orphan);
2260 delete_item = 1;
Josef Bacik7b128762008-07-24 12:17:14 -04002261 }
2262
Yan, Zhengd68fc572010-05-16 10:49:58 -04002263 if (BTRFS_I(inode)->orphan_meta_reserved) {
2264 BTRFS_I(inode)->orphan_meta_reserved = 0;
2265 release_rsv = 1;
2266 }
2267 spin_unlock(&root->orphan_lock);
2268
2269 if (trans && delete_item) {
2270 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2271 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04002272 }
2273
Yan, Zhengd68fc572010-05-16 10:49:58 -04002274 if (release_rsv)
2275 btrfs_orphan_release_metadata(inode);
Josef Bacik7b128762008-07-24 12:17:14 -04002276
Yan, Zhengd68fc572010-05-16 10:49:58 -04002277 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002278}
2279
2280/*
2281 * this cleans up any orphans that may be left on the list from the last use
2282 * of this root.
2283 */
2284void btrfs_orphan_cleanup(struct btrfs_root *root)
2285{
2286 struct btrfs_path *path;
2287 struct extent_buffer *leaf;
Josef Bacik7b128762008-07-24 12:17:14 -04002288 struct btrfs_key key, found_key;
2289 struct btrfs_trans_handle *trans;
2290 struct inode *inode;
2291 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2292
Yan, Zhengd68fc572010-05-16 10:49:58 -04002293 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
Josef Bacik7b128762008-07-24 12:17:14 -04002294 return;
Yan, Zhengc71bf092009-11-12 09:34:40 +00002295
2296 path = btrfs_alloc_path();
2297 BUG_ON(!path);
Josef Bacik7b128762008-07-24 12:17:14 -04002298 path->reada = -1;
2299
2300 key.objectid = BTRFS_ORPHAN_OBJECTID;
2301 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2302 key.offset = (u64)-1;
2303
Josef Bacik7b128762008-07-24 12:17:14 -04002304 while (1) {
2305 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2306 if (ret < 0) {
2307 printk(KERN_ERR "Error searching slot for orphan: %d"
2308 "\n", ret);
2309 break;
2310 }
2311
2312 /*
2313 * if ret == 0 means we found what we were searching for, which
2314 * is weird, but possible, so only screw with path if we didnt
2315 * find the key and see if we have stuff that matches
2316 */
2317 if (ret > 0) {
2318 if (path->slots[0] == 0)
2319 break;
2320 path->slots[0]--;
2321 }
2322
2323 /* pull out the item */
2324 leaf = path->nodes[0];
Josef Bacik7b128762008-07-24 12:17:14 -04002325 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2326
2327 /* make sure the item matches what we want */
2328 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2329 break;
2330 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2331 break;
2332
2333 /* release the path since we're done with it */
2334 btrfs_release_path(root, path);
2335
2336 /*
2337 * this is where we are basically btrfs_lookup, without the
2338 * crossing root thing. we store the inode number in the
2339 * offset of the orphan item.
2340 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002341 found_key.objectid = found_key.offset;
2342 found_key.type = BTRFS_INODE_ITEM_KEY;
2343 found_key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +00002344 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002345 BUG_ON(IS_ERR(inode));
Josef Bacik7b128762008-07-24 12:17:14 -04002346
Josef Bacik7b128762008-07-24 12:17:14 -04002347 /*
2348 * add this inode to the orphan list so btrfs_orphan_del does
2349 * the proper thing when we hit it
2350 */
Yan, Zhengd68fc572010-05-16 10:49:58 -04002351 spin_lock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002352 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002353 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002354
2355 /*
2356 * if this is a bad inode, means we actually succeeded in
2357 * removing the inode, but not the orphan record, which means
2358 * we need to manually delete the orphan since iput will just
2359 * do a destroy_inode
2360 */
2361 if (is_bad_inode(inode)) {
Yan, Zhenga22285a2010-05-16 10:48:46 -04002362 trans = btrfs_start_transaction(root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00002363 BUG_ON(IS_ERR(trans));
Josef Bacik7b128762008-07-24 12:17:14 -04002364 btrfs_orphan_del(trans, inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04002365 btrfs_end_transaction(trans, root);
Josef Bacik7b128762008-07-24 12:17:14 -04002366 iput(inode);
2367 continue;
2368 }
2369
2370 /* if we have links, this was a truncate, lets do that */
2371 if (inode->i_nlink) {
2372 nr_truncate++;
2373 btrfs_truncate(inode);
2374 } else {
2375 nr_unlink++;
2376 }
2377
2378 /* this will do delete_inode and everything for us */
2379 iput(inode);
2380 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04002381 btrfs_free_path(path);
2382
2383 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2384
2385 if (root->orphan_block_rsv)
2386 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2387 (u64)-1);
2388
2389 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2390 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00002391 BUG_ON(IS_ERR(trans));
Yan, Zhengd68fc572010-05-16 10:49:58 -04002392 btrfs_end_transaction(trans, root);
2393 }
Josef Bacik7b128762008-07-24 12:17:14 -04002394
2395 if (nr_unlink)
2396 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2397 if (nr_truncate)
2398 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
Josef Bacik7b128762008-07-24 12:17:14 -04002399}
2400
Chris Masond352ac62008-09-29 15:18:18 -04002401/*
Chris Mason46a53cc2009-04-27 11:47:50 -04002402 * very simple check to peek ahead in the leaf looking for xattrs. If we
2403 * don't find any xattrs, we know there can't be any acls.
2404 *
2405 * slot is the slot the inode is in, objectid is the objectid of the inode
2406 */
2407static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2408 int slot, u64 objectid)
2409{
2410 u32 nritems = btrfs_header_nritems(leaf);
2411 struct btrfs_key found_key;
2412 int scanned = 0;
2413
2414 slot++;
2415 while (slot < nritems) {
2416 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2417
2418 /* we found a different objectid, there must not be acls */
2419 if (found_key.objectid != objectid)
2420 return 0;
2421
2422 /* we found an xattr, assume we've got an acl */
2423 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2424 return 1;
2425
2426 /*
2427 * we found a key greater than an xattr key, there can't
2428 * be any acls later on
2429 */
2430 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2431 return 0;
2432
2433 slot++;
2434 scanned++;
2435
2436 /*
2437 * it goes inode, inode backrefs, xattrs, extents,
2438 * so if there are a ton of hard links to an inode there can
2439 * be a lot of backrefs. Don't waste time searching too hard,
2440 * this is just an optimization
2441 */
2442 if (scanned >= 8)
2443 break;
2444 }
2445 /* we hit the end of the leaf before we found an xattr or
2446 * something larger than an xattr. We have to assume the inode
2447 * has acls
2448 */
2449 return 1;
2450}
2451
2452/*
Chris Masond352ac62008-09-29 15:18:18 -04002453 * read an inode from the btree into the in-memory inode
2454 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002455static void btrfs_read_locked_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002456{
2457 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002458 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002459 struct btrfs_inode_item *inode_item;
Chris Mason0b86a832008-03-24 15:01:56 -04002460 struct btrfs_timespec *tspec;
Chris Mason39279cc2007-06-12 06:35:45 -04002461 struct btrfs_root *root = BTRFS_I(inode)->root;
2462 struct btrfs_key location;
Chris Mason46a53cc2009-04-27 11:47:50 -04002463 int maybe_acls;
Chris Mason39279cc2007-06-12 06:35:45 -04002464 u64 alloc_group_block;
Josef Bacik618e21d2007-07-11 10:18:17 -04002465 u32 rdev;
Chris Mason39279cc2007-06-12 06:35:45 -04002466 int ret;
2467
2468 path = btrfs_alloc_path();
2469 BUG_ON(!path);
Chris Mason39279cc2007-06-12 06:35:45 -04002470 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
Chris Masondc17ff82008-01-08 15:46:30 -05002471
Chris Mason39279cc2007-06-12 06:35:45 -04002472 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04002473 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04002474 goto make_bad;
Chris Mason39279cc2007-06-12 06:35:45 -04002475
Chris Mason5f39d392007-10-15 16:14:19 -04002476 leaf = path->nodes[0];
2477 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2478 struct btrfs_inode_item);
2479
2480 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2481 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2482 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2483 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
Chris Masondbe674a2008-07-17 12:54:05 -04002484 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04002485
2486 tspec = btrfs_inode_atime(inode_item);
2487 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2488 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2489
2490 tspec = btrfs_inode_mtime(inode_item);
2491 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2492 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2493
2494 tspec = btrfs_inode_ctime(inode_item);
2495 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2496 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2497
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002498 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04002499 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
Chris Masonc3027eb2008-12-08 16:40:21 -05002500 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04002501 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik618e21d2007-07-11 10:18:17 -04002502 inode->i_rdev = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002503 rdev = btrfs_inode_rdev(leaf, inode_item);
2504
Josef Bacikaec74772008-07-24 12:12:38 -04002505 BTRFS_I(inode)->index_cnt = (u64)-1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05002506 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
Josef Bacikaec74772008-07-24 12:12:38 -04002507
Chris Mason5f39d392007-10-15 16:14:19 -04002508 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
Chris Masonb4ce94d2009-02-04 09:25:08 -05002509
Chris Mason46a53cc2009-04-27 11:47:50 -04002510 /*
2511 * try to precache a NULL acl entry for files that don't have
2512 * any xattrs or acls
2513 */
2514 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
Al Viro72c04902009-06-24 16:58:48 -04002515 if (!maybe_acls)
2516 cache_no_acl(inode);
Chris Mason46a53cc2009-04-27 11:47:50 -04002517
Yan Zhengd2fb3432008-12-11 16:30:39 -05002518 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2519 alloc_group_block, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04002520 btrfs_free_path(path);
2521 inode_item = NULL;
2522
Chris Mason39279cc2007-06-12 06:35:45 -04002523 switch (inode->i_mode & S_IFMT) {
Chris Mason39279cc2007-06-12 06:35:45 -04002524 case S_IFREG:
2525 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04002526 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05002527 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04002528 inode->i_fop = &btrfs_file_operations;
2529 inode->i_op = &btrfs_file_inode_operations;
2530 break;
2531 case S_IFDIR:
2532 inode->i_fop = &btrfs_dir_file_operations;
2533 if (root == root->fs_info->tree_root)
2534 inode->i_op = &btrfs_dir_ro_inode_operations;
2535 else
2536 inode->i_op = &btrfs_dir_inode_operations;
2537 break;
2538 case S_IFLNK:
2539 inode->i_op = &btrfs_symlink_inode_operations;
2540 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04002541 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04002542 break;
Josef Bacik618e21d2007-07-11 10:18:17 -04002543 default:
Jim Owens0279b4c2009-02-04 09:29:13 -05002544 inode->i_op = &btrfs_special_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -04002545 init_special_inode(inode, inode->i_mode, rdev);
2546 break;
Chris Mason39279cc2007-06-12 06:35:45 -04002547 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02002548
2549 btrfs_update_iflags(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002550 return;
2551
2552make_bad:
Chris Mason39279cc2007-06-12 06:35:45 -04002553 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002554 make_bad_inode(inode);
2555}
2556
Chris Masond352ac62008-09-29 15:18:18 -04002557/*
2558 * given a leaf and an inode, copy the inode fields into the leaf
2559 */
Chris Masone02119d2008-09-05 16:13:11 -04002560static void fill_inode_item(struct btrfs_trans_handle *trans,
2561 struct extent_buffer *leaf,
Chris Mason5f39d392007-10-15 16:14:19 -04002562 struct btrfs_inode_item *item,
Chris Mason39279cc2007-06-12 06:35:45 -04002563 struct inode *inode)
2564{
Chris Mason5f39d392007-10-15 16:14:19 -04002565 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2566 btrfs_set_inode_gid(leaf, item, inode->i_gid);
Chris Masondbe674a2008-07-17 12:54:05 -04002567 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
Chris Mason5f39d392007-10-15 16:14:19 -04002568 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2569 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2570
2571 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2572 inode->i_atime.tv_sec);
2573 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2574 inode->i_atime.tv_nsec);
2575
2576 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2577 inode->i_mtime.tv_sec);
2578 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2579 inode->i_mtime.tv_nsec);
2580
2581 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2582 inode->i_ctime.tv_sec);
2583 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2584 inode->i_ctime.tv_nsec);
2585
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002586 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
Chris Masone02119d2008-09-05 16:13:11 -04002587 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
Chris Masonc3027eb2008-12-08 16:40:21 -05002588 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
Chris Masone02119d2008-09-05 16:13:11 -04002589 btrfs_set_inode_transid(leaf, item, trans->transid);
Chris Mason5f39d392007-10-15 16:14:19 -04002590 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
Yanb98b6762008-01-08 15:54:37 -05002591 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
Yan Zhengd2fb3432008-12-11 16:30:39 -05002592 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
Chris Mason39279cc2007-06-12 06:35:45 -04002593}
2594
Chris Masond352ac62008-09-29 15:18:18 -04002595/*
2596 * copy everything in the in-memory inode into the btree.
2597 */
Chris Masond3977122009-01-05 21:25:51 -05002598noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2599 struct btrfs_root *root, struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002600{
2601 struct btrfs_inode_item *inode_item;
2602 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002603 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002604 int ret;
2605
2606 path = btrfs_alloc_path();
2607 BUG_ON(!path);
Chris Masonb9473432009-03-13 11:00:37 -04002608 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04002609 ret = btrfs_lookup_inode(trans, root, path,
2610 &BTRFS_I(inode)->location, 1);
2611 if (ret) {
2612 if (ret > 0)
2613 ret = -ENOENT;
2614 goto failed;
2615 }
2616
Chris Masonb4ce94d2009-02-04 09:25:08 -05002617 btrfs_unlock_up_safe(path, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04002618 leaf = path->nodes[0];
2619 inode_item = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04002620 struct btrfs_inode_item);
2621
Chris Masone02119d2008-09-05 16:13:11 -04002622 fill_inode_item(trans, leaf, inode_item, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04002623 btrfs_mark_buffer_dirty(leaf);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002624 btrfs_set_inode_last_trans(trans, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002625 ret = 0;
2626failed:
Chris Mason39279cc2007-06-12 06:35:45 -04002627 btrfs_free_path(path);
2628 return ret;
2629}
2630
2631
Chris Masond352ac62008-09-29 15:18:18 -04002632/*
2633 * unlink helper that gets used here in inode.c and in the tree logging
2634 * recovery code. It remove a link in a directory with a given name, and
2635 * also drops the back refs in the inode to the directory
2636 */
Chris Masone02119d2008-09-05 16:13:11 -04002637int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2638 struct btrfs_root *root,
2639 struct inode *dir, struct inode *inode,
2640 const char *name, int name_len)
Chris Mason39279cc2007-06-12 06:35:45 -04002641{
2642 struct btrfs_path *path;
Chris Mason39279cc2007-06-12 06:35:45 -04002643 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002644 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002645 struct btrfs_dir_item *di;
Chris Mason5f39d392007-10-15 16:14:19 -04002646 struct btrfs_key key;
Josef Bacikaec74772008-07-24 12:12:38 -04002647 u64 index;
Chris Mason39279cc2007-06-12 06:35:45 -04002648
2649 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04002650 if (!path) {
2651 ret = -ENOMEM;
Tsutomu Itoh554233a2011-02-03 03:16:25 +00002652 goto out;
Chris Mason54aa1f42007-06-22 14:16:25 -04002653 }
2654
Chris Masonb9473432009-03-13 11:00:37 -04002655 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04002656 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2657 name, name_len, -1);
2658 if (IS_ERR(di)) {
2659 ret = PTR_ERR(di);
2660 goto err;
2661 }
2662 if (!di) {
2663 ret = -ENOENT;
2664 goto err;
2665 }
Chris Mason5f39d392007-10-15 16:14:19 -04002666 leaf = path->nodes[0];
2667 btrfs_dir_item_key_to_cpu(leaf, di, &key);
Chris Mason39279cc2007-06-12 06:35:45 -04002668 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason54aa1f42007-06-22 14:16:25 -04002669 if (ret)
2670 goto err;
Chris Mason39279cc2007-06-12 06:35:45 -04002671 btrfs_release_path(root, path);
2672
Josef Bacikaec74772008-07-24 12:12:38 -04002673 ret = btrfs_del_inode_ref(trans, root, name, name_len,
Chris Masone02119d2008-09-05 16:13:11 -04002674 inode->i_ino,
2675 dir->i_ino, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04002676 if (ret) {
Chris Masond3977122009-01-05 21:25:51 -05002677 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
Josef Bacikaec74772008-07-24 12:12:38 -04002678 "inode %lu parent %lu\n", name_len, name,
Chris Masone02119d2008-09-05 16:13:11 -04002679 inode->i_ino, dir->i_ino);
Josef Bacikaec74772008-07-24 12:12:38 -04002680 goto err;
2681 }
2682
Chris Mason39279cc2007-06-12 06:35:45 -04002683 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
Josef Bacikaec74772008-07-24 12:12:38 -04002684 index, name, name_len, -1);
Chris Mason39279cc2007-06-12 06:35:45 -04002685 if (IS_ERR(di)) {
2686 ret = PTR_ERR(di);
2687 goto err;
2688 }
2689 if (!di) {
2690 ret = -ENOENT;
2691 goto err;
2692 }
2693 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason925baed2008-06-25 16:01:30 -04002694 btrfs_release_path(root, path);
Chris Mason39279cc2007-06-12 06:35:45 -04002695
Chris Masone02119d2008-09-05 16:13:11 -04002696 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2697 inode, dir->i_ino);
Chris Mason49eb7e42008-09-11 15:53:12 -04002698 BUG_ON(ret != 0 && ret != -ENOENT);
Chris Masone02119d2008-09-05 16:13:11 -04002699
2700 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2701 dir, index);
Chris Mason6418c962010-10-30 07:34:24 -04002702 if (ret == -ENOENT)
2703 ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002704err:
2705 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002706 if (ret)
2707 goto out;
2708
2709 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2710 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2711 btrfs_update_inode(trans, root, dir);
2712 btrfs_drop_nlink(inode);
2713 ret = btrfs_update_inode(trans, root, inode);
Chris Masone02119d2008-09-05 16:13:11 -04002714out:
Chris Mason39279cc2007-06-12 06:35:45 -04002715 return ret;
2716}
2717
Yan, Zhenga22285a2010-05-16 10:48:46 -04002718/* helper to check if there is any shared block in the path */
2719static int check_path_shared(struct btrfs_root *root,
2720 struct btrfs_path *path)
2721{
2722 struct extent_buffer *eb;
2723 int level;
Dan Carpenter0e4dcbef2010-06-01 08:23:11 +00002724 u64 refs = 1;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002725
2726 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
Josef Bacikdedefd72011-01-24 21:43:18 +00002727 int ret;
2728
Yan, Zhenga22285a2010-05-16 10:48:46 -04002729 if (!path->nodes[level])
2730 break;
2731 eb = path->nodes[level];
2732 if (!btrfs_block_can_be_shared(root, eb))
2733 continue;
2734 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2735 &refs, NULL);
2736 if (refs > 1)
2737 return 1;
2738 }
Josef Bacikdedefd72011-01-24 21:43:18 +00002739 return 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002740}
2741
2742/*
2743 * helper to start transaction for unlink and rmdir.
2744 *
2745 * unlink and rmdir are special in btrfs, they do not always free space.
2746 * so in enospc case, we should make sure they will free space before
2747 * allowing them to use the global metadata reservation.
2748 */
2749static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2750 struct dentry *dentry)
2751{
2752 struct btrfs_trans_handle *trans;
2753 struct btrfs_root *root = BTRFS_I(dir)->root;
2754 struct btrfs_path *path;
2755 struct btrfs_inode_ref *ref;
2756 struct btrfs_dir_item *di;
2757 struct inode *inode = dentry->d_inode;
2758 u64 index;
2759 int check_link = 1;
2760 int err = -ENOSPC;
2761 int ret;
2762
2763 trans = btrfs_start_transaction(root, 10);
2764 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2765 return trans;
2766
2767 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2768 return ERR_PTR(-ENOSPC);
2769
2770 /* check if there is someone else holds reference */
2771 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2772 return ERR_PTR(-ENOSPC);
2773
2774 if (atomic_read(&inode->i_count) > 2)
2775 return ERR_PTR(-ENOSPC);
2776
2777 if (xchg(&root->fs_info->enospc_unlink, 1))
2778 return ERR_PTR(-ENOSPC);
2779
2780 path = btrfs_alloc_path();
2781 if (!path) {
2782 root->fs_info->enospc_unlink = 0;
2783 return ERR_PTR(-ENOMEM);
2784 }
2785
2786 trans = btrfs_start_transaction(root, 0);
2787 if (IS_ERR(trans)) {
2788 btrfs_free_path(path);
2789 root->fs_info->enospc_unlink = 0;
2790 return trans;
2791 }
2792
2793 path->skip_locking = 1;
2794 path->search_commit_root = 1;
2795
2796 ret = btrfs_lookup_inode(trans, root, path,
2797 &BTRFS_I(dir)->location, 0);
2798 if (ret < 0) {
2799 err = ret;
2800 goto out;
2801 }
2802 if (ret == 0) {
2803 if (check_path_shared(root, path))
2804 goto out;
2805 } else {
2806 check_link = 0;
2807 }
2808 btrfs_release_path(root, path);
2809
2810 ret = btrfs_lookup_inode(trans, root, path,
2811 &BTRFS_I(inode)->location, 0);
2812 if (ret < 0) {
2813 err = ret;
2814 goto out;
2815 }
2816 if (ret == 0) {
2817 if (check_path_shared(root, path))
2818 goto out;
2819 } else {
2820 check_link = 0;
2821 }
2822 btrfs_release_path(root, path);
2823
2824 if (ret == 0 && S_ISREG(inode->i_mode)) {
2825 ret = btrfs_lookup_file_extent(trans, root, path,
2826 inode->i_ino, (u64)-1, 0);
2827 if (ret < 0) {
2828 err = ret;
2829 goto out;
2830 }
2831 BUG_ON(ret == 0);
2832 if (check_path_shared(root, path))
2833 goto out;
2834 btrfs_release_path(root, path);
2835 }
2836
2837 if (!check_link) {
2838 err = 0;
2839 goto out;
2840 }
2841
2842 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2843 dentry->d_name.name, dentry->d_name.len, 0);
2844 if (IS_ERR(di)) {
2845 err = PTR_ERR(di);
2846 goto out;
2847 }
2848 if (di) {
2849 if (check_path_shared(root, path))
2850 goto out;
2851 } else {
2852 err = 0;
2853 goto out;
2854 }
2855 btrfs_release_path(root, path);
2856
2857 ref = btrfs_lookup_inode_ref(trans, root, path,
2858 dentry->d_name.name, dentry->d_name.len,
2859 inode->i_ino, dir->i_ino, 0);
2860 if (IS_ERR(ref)) {
2861 err = PTR_ERR(ref);
2862 goto out;
2863 }
2864 BUG_ON(!ref);
2865 if (check_path_shared(root, path))
2866 goto out;
2867 index = btrfs_inode_ref_index(path->nodes[0], ref);
2868 btrfs_release_path(root, path);
2869
2870 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2871 dentry->d_name.name, dentry->d_name.len, 0);
2872 if (IS_ERR(di)) {
2873 err = PTR_ERR(di);
2874 goto out;
2875 }
2876 BUG_ON(ret == -ENOENT);
2877 if (check_path_shared(root, path))
2878 goto out;
2879
2880 err = 0;
2881out:
2882 btrfs_free_path(path);
2883 if (err) {
2884 btrfs_end_transaction(trans, root);
2885 root->fs_info->enospc_unlink = 0;
2886 return ERR_PTR(err);
2887 }
2888
2889 trans->block_rsv = &root->fs_info->global_block_rsv;
2890 return trans;
2891}
2892
2893static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2894 struct btrfs_root *root)
2895{
2896 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2897 BUG_ON(!root->fs_info->enospc_unlink);
2898 root->fs_info->enospc_unlink = 0;
2899 }
2900 btrfs_end_transaction_throttle(trans, root);
2901}
2902
Chris Mason39279cc2007-06-12 06:35:45 -04002903static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2904{
Yan, Zhenga22285a2010-05-16 10:48:46 -04002905 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04002906 struct btrfs_trans_handle *trans;
Josef Bacik7b128762008-07-24 12:17:14 -04002907 struct inode *inode = dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04002908 int ret;
Chris Mason1832a6d2007-12-21 16:27:21 -05002909 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002910
Yan, Zhenga22285a2010-05-16 10:48:46 -04002911 trans = __unlink_start_trans(dir, dentry);
2912 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002913 return PTR_ERR(trans);
Chris Mason5f39d392007-10-15 16:14:19 -04002914
Chris Mason39279cc2007-06-12 06:35:45 -04002915 btrfs_set_trans_block_group(trans, dir);
Chris Mason12fcfd22009-03-24 10:24:20 -04002916
2917 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2918
Chris Masone02119d2008-09-05 16:13:11 -04002919 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2920 dentry->d_name.name, dentry->d_name.len);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002921 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04002922
Yan, Zhenga22285a2010-05-16 10:48:46 -04002923 if (inode->i_nlink == 0) {
Josef Bacik7b128762008-07-24 12:17:14 -04002924 ret = btrfs_orphan_add(trans, inode);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002925 BUG_ON(ret);
2926 }
Josef Bacik7b128762008-07-24 12:17:14 -04002927
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002928 nr = trans->blocks_used;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002929 __unlink_end_trans(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002930 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04002931 return ret;
2932}
2933
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002934int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2935 struct btrfs_root *root,
2936 struct inode *dir, u64 objectid,
2937 const char *name, int name_len)
2938{
2939 struct btrfs_path *path;
2940 struct extent_buffer *leaf;
2941 struct btrfs_dir_item *di;
2942 struct btrfs_key key;
2943 u64 index;
2944 int ret;
2945
2946 path = btrfs_alloc_path();
2947 if (!path)
2948 return -ENOMEM;
2949
2950 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2951 name, name_len, -1);
2952 BUG_ON(!di || IS_ERR(di));
2953
2954 leaf = path->nodes[0];
2955 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2956 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2957 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2958 BUG_ON(ret);
2959 btrfs_release_path(root, path);
2960
2961 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2962 objectid, root->root_key.objectid,
2963 dir->i_ino, &index, name, name_len);
2964 if (ret < 0) {
2965 BUG_ON(ret != -ENOENT);
2966 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2967 name, name_len);
2968 BUG_ON(!di || IS_ERR(di));
2969
2970 leaf = path->nodes[0];
2971 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2972 btrfs_release_path(root, path);
2973 index = key.offset;
2974 }
2975
2976 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2977 index, name, name_len, -1);
2978 BUG_ON(!di || IS_ERR(di));
2979
2980 leaf = path->nodes[0];
2981 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2982 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2983 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2984 BUG_ON(ret);
2985 btrfs_release_path(root, path);
2986
2987 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2988 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2989 ret = btrfs_update_inode(trans, root, dir);
2990 BUG_ON(ret);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002991
2992 btrfs_free_path(path);
2993 return 0;
2994}
2995
Chris Mason39279cc2007-06-12 06:35:45 -04002996static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2997{
2998 struct inode *inode = dentry->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05002999 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003000 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003001 struct btrfs_trans_handle *trans;
Chris Mason1832a6d2007-12-21 16:27:21 -05003002 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003003
Chris Mason3394e162008-11-17 20:42:26 -05003004 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003005 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan134d4512007-10-25 15:49:25 -04003006 return -ENOTEMPTY;
3007
Yan, Zhenga22285a2010-05-16 10:48:46 -04003008 trans = __unlink_start_trans(dir, dentry);
3009 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05003010 return PTR_ERR(trans);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05003011
Chris Mason39279cc2007-06-12 06:35:45 -04003012 btrfs_set_trans_block_group(trans, dir);
Chris Mason39279cc2007-06-12 06:35:45 -04003013
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003014 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3015 err = btrfs_unlink_subvol(trans, root, dir,
3016 BTRFS_I(inode)->location.objectid,
3017 dentry->d_name.name,
3018 dentry->d_name.len);
3019 goto out;
3020 }
3021
Josef Bacik7b128762008-07-24 12:17:14 -04003022 err = btrfs_orphan_add(trans, inode);
3023 if (err)
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003024 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003025
Chris Mason39279cc2007-06-12 06:35:45 -04003026 /* now the directory is empty */
Chris Masone02119d2008-09-05 16:13:11 -04003027 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3028 dentry->d_name.name, dentry->d_name.len);
Chris Masond3977122009-01-05 21:25:51 -05003029 if (!err)
Chris Masondbe674a2008-07-17 12:54:05 -04003030 btrfs_i_size_write(inode, 0);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003031out:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003032 nr = trans->blocks_used;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003033 __unlink_end_trans(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003034 btrfs_btree_balance_dirty(root, nr);
Chris Mason39544012007-12-12 14:38:19 -05003035
Chris Mason39279cc2007-06-12 06:35:45 -04003036 return err;
3037}
3038
Chris Masond20f7042008-12-08 16:58:54 -05003039#if 0
Chris Mason39279cc2007-06-12 06:35:45 -04003040/*
Chris Mason323ac952008-10-01 19:05:46 -04003041 * when truncating bytes in a file, it is possible to avoid reading
3042 * the leaves that contain only checksum items. This can be the
3043 * majority of the IO required to delete a large file, but it must
3044 * be done carefully.
3045 *
3046 * The keys in the level just above the leaves are checked to make sure
3047 * the lowest key in a given leaf is a csum key, and starts at an offset
3048 * after the new size.
3049 *
3050 * Then the key for the next leaf is checked to make sure it also has
3051 * a checksum item for the same file. If it does, we know our target leaf
3052 * contains only checksum items, and it can be safely freed without reading
3053 * it.
3054 *
3055 * This is just an optimization targeted at large files. It may do
3056 * nothing. It will return 0 unless things went badly.
3057 */
3058static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3059 struct btrfs_root *root,
3060 struct btrfs_path *path,
3061 struct inode *inode, u64 new_size)
3062{
3063 struct btrfs_key key;
3064 int ret;
3065 int nritems;
3066 struct btrfs_key found_key;
3067 struct btrfs_key other_key;
Yan Zheng5b84e8d2008-10-09 11:46:19 -04003068 struct btrfs_leaf_ref *ref;
3069 u64 leaf_gen;
3070 u64 leaf_start;
Chris Mason323ac952008-10-01 19:05:46 -04003071
3072 path->lowest_level = 1;
3073 key.objectid = inode->i_ino;
3074 key.type = BTRFS_CSUM_ITEM_KEY;
3075 key.offset = new_size;
3076again:
3077 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3078 if (ret < 0)
3079 goto out;
3080
3081 if (path->nodes[1] == NULL) {
3082 ret = 0;
3083 goto out;
3084 }
3085 ret = 0;
3086 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3087 nritems = btrfs_header_nritems(path->nodes[1]);
3088
3089 if (!nritems)
3090 goto out;
3091
3092 if (path->slots[1] >= nritems)
3093 goto next_node;
3094
3095 /* did we find a key greater than anything we want to delete? */
3096 if (found_key.objectid > inode->i_ino ||
3097 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3098 goto out;
3099
3100 /* we check the next key in the node to make sure the leave contains
3101 * only checksum items. This comparison doesn't work if our
3102 * leaf is the last one in the node
3103 */
3104 if (path->slots[1] + 1 >= nritems) {
3105next_node:
3106 /* search forward from the last key in the node, this
3107 * will bring us into the next node in the tree
3108 */
3109 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3110
3111 /* unlikely, but we inc below, so check to be safe */
3112 if (found_key.offset == (u64)-1)
3113 goto out;
3114
3115 /* search_forward needs a path with locks held, do the
3116 * search again for the original key. It is possible
3117 * this will race with a balance and return a path that
3118 * we could modify, but this drop is just an optimization
3119 * and is allowed to miss some leaves.
3120 */
3121 btrfs_release_path(root, path);
3122 found_key.offset++;
3123
3124 /* setup a max key for search_forward */
3125 other_key.offset = (u64)-1;
3126 other_key.type = key.type;
3127 other_key.objectid = key.objectid;
3128
3129 path->keep_locks = 1;
3130 ret = btrfs_search_forward(root, &found_key, &other_key,
3131 path, 0, 0);
3132 path->keep_locks = 0;
3133 if (ret || found_key.objectid != key.objectid ||
3134 found_key.type != key.type) {
3135 ret = 0;
3136 goto out;
3137 }
3138
3139 key.offset = found_key.offset;
3140 btrfs_release_path(root, path);
3141 cond_resched();
3142 goto again;
3143 }
3144
3145 /* we know there's one more slot after us in the tree,
3146 * read that key so we can verify it is also a checksum item
3147 */
3148 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3149
3150 if (found_key.objectid < inode->i_ino)
3151 goto next_key;
3152
3153 if (found_key.type != key.type || found_key.offset < new_size)
3154 goto next_key;
3155
3156 /*
3157 * if the key for the next leaf isn't a csum key from this objectid,
3158 * we can't be sure there aren't good items inside this leaf.
3159 * Bail out
3160 */
3161 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3162 goto out;
3163
Yan Zheng5b84e8d2008-10-09 11:46:19 -04003164 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3165 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
Chris Mason323ac952008-10-01 19:05:46 -04003166 /*
3167 * it is safe to delete this leaf, it contains only
3168 * csum items from this inode at an offset >= new_size
3169 */
Yan Zheng5b84e8d2008-10-09 11:46:19 -04003170 ret = btrfs_del_leaf(trans, root, path, leaf_start);
Chris Mason323ac952008-10-01 19:05:46 -04003171 BUG_ON(ret);
3172
Yan Zheng5b84e8d2008-10-09 11:46:19 -04003173 if (root->ref_cows && leaf_gen < trans->transid) {
3174 ref = btrfs_alloc_leaf_ref(root, 0);
3175 if (ref) {
3176 ref->root_gen = root->root_key.offset;
3177 ref->bytenr = leaf_start;
3178 ref->owner = 0;
3179 ref->generation = leaf_gen;
3180 ref->nritems = 0;
3181
Chris Masonbd56b302009-02-04 09:27:02 -05003182 btrfs_sort_leaf_ref(ref);
3183
Yan Zheng5b84e8d2008-10-09 11:46:19 -04003184 ret = btrfs_add_leaf_ref(root, ref, 0);
3185 WARN_ON(ret);
3186 btrfs_free_leaf_ref(root, ref);
3187 } else {
3188 WARN_ON(1);
3189 }
3190 }
Chris Mason323ac952008-10-01 19:05:46 -04003191next_key:
3192 btrfs_release_path(root, path);
3193
3194 if (other_key.objectid == inode->i_ino &&
3195 other_key.type == key.type && other_key.offset > key.offset) {
3196 key.offset = other_key.offset;
3197 cond_resched();
3198 goto again;
3199 }
3200 ret = 0;
3201out:
3202 /* fixup any changes we've made to the path */
3203 path->lowest_level = 0;
3204 path->keep_locks = 0;
3205 btrfs_release_path(root, path);
3206 return ret;
3207}
3208
Chris Masond20f7042008-12-08 16:58:54 -05003209#endif
3210
Chris Mason323ac952008-10-01 19:05:46 -04003211/*
Chris Mason39279cc2007-06-12 06:35:45 -04003212 * this can truncate away extent items, csum items and directory items.
3213 * It starts at a high offset and removes keys until it can't find
Chris Masond352ac62008-09-29 15:18:18 -04003214 * any higher than new_size
Chris Mason39279cc2007-06-12 06:35:45 -04003215 *
3216 * csum items that cross the new i_size are truncated to the new size
3217 * as well.
Josef Bacik7b128762008-07-24 12:17:14 -04003218 *
3219 * min_type is the minimum key type to truncate down to. If set to 0, this
3220 * will kill all the items on this inode, including the INODE_ITEM_KEY.
Chris Mason39279cc2007-06-12 06:35:45 -04003221 */
Yan, Zheng80825102009-11-12 09:35:36 +00003222int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3223 struct btrfs_root *root,
3224 struct inode *inode,
3225 u64 new_size, u32 min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003226{
Chris Mason39279cc2007-06-12 06:35:45 -04003227 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003228 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003229 struct btrfs_file_extent_item *fi;
Yan, Zheng80825102009-11-12 09:35:36 +00003230 struct btrfs_key key;
3231 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04003232 u64 extent_start = 0;
Chris Masondb945352007-10-15 16:15:53 -04003233 u64 extent_num_bytes = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003234 u64 extent_offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003235 u64 item_end = 0;
Yan, Zheng80825102009-11-12 09:35:36 +00003236 u64 mask = root->sectorsize - 1;
3237 u32 found_type = (u8)-1;
Chris Mason39279cc2007-06-12 06:35:45 -04003238 int found_extent;
3239 int del_item;
Chris Mason85e21ba2008-01-29 15:11:36 -05003240 int pending_del_nr = 0;
3241 int pending_del_slot = 0;
Chris Mason179e29e2007-11-01 11:28:41 -04003242 int extent_type = -1;
Chris Mason771ed682008-11-06 22:02:51 -05003243 int encoding;
Yan, Zheng80825102009-11-12 09:35:36 +00003244 int ret;
3245 int err = 0;
3246
3247 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
Chris Mason39279cc2007-06-12 06:35:45 -04003248
Josef Bacik0af3d002010-06-21 14:48:16 -04003249 if (root->ref_cows || root == root->fs_info->tree_root)
Zheng Yan5b21f2e2008-09-26 10:05:38 -04003250 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003251
Chris Mason39279cc2007-06-12 06:35:45 -04003252 path = btrfs_alloc_path();
3253 BUG_ON(!path);
Julia Lawall33c17ad2009-07-22 16:49:01 -04003254 path->reada = -1;
Chris Mason5f39d392007-10-15 16:14:19 -04003255
Chris Mason39279cc2007-06-12 06:35:45 -04003256 key.objectid = inode->i_ino;
3257 key.offset = (u64)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04003258 key.type = (u8)-1;
3259
Chris Mason85e21ba2008-01-29 15:11:36 -05003260search_again:
Chris Masonb9473432009-03-13 11:00:37 -04003261 path->leave_spinning = 1;
Chris Mason85e21ba2008-01-29 15:11:36 -05003262 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Yan, Zheng80825102009-11-12 09:35:36 +00003263 if (ret < 0) {
3264 err = ret;
3265 goto out;
3266 }
Chris Masond3977122009-01-05 21:25:51 -05003267
Chris Mason85e21ba2008-01-29 15:11:36 -05003268 if (ret > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04003269 /* there are no items in the tree for us to truncate, we're
3270 * done
3271 */
Yan, Zheng80825102009-11-12 09:35:36 +00003272 if (path->slots[0] == 0)
3273 goto out;
Chris Mason85e21ba2008-01-29 15:11:36 -05003274 path->slots[0]--;
3275 }
3276
Chris Masond3977122009-01-05 21:25:51 -05003277 while (1) {
Chris Mason39279cc2007-06-12 06:35:45 -04003278 fi = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -04003279 leaf = path->nodes[0];
3280 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3281 found_type = btrfs_key_type(&found_key);
Chris Mason771ed682008-11-06 22:02:51 -05003282 encoding = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003283
Chris Mason5f39d392007-10-15 16:14:19 -04003284 if (found_key.objectid != inode->i_ino)
Chris Mason39279cc2007-06-12 06:35:45 -04003285 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003286
Chris Mason85e21ba2008-01-29 15:11:36 -05003287 if (found_type < min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003288 break;
3289
Chris Mason5f39d392007-10-15 16:14:19 -04003290 item_end = found_key.offset;
Chris Mason39279cc2007-06-12 06:35:45 -04003291 if (found_type == BTRFS_EXTENT_DATA_KEY) {
Chris Mason5f39d392007-10-15 16:14:19 -04003292 fi = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04003293 struct btrfs_file_extent_item);
Chris Mason179e29e2007-11-01 11:28:41 -04003294 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Mason771ed682008-11-06 22:02:51 -05003295 encoding = btrfs_file_extent_compression(leaf, fi);
3296 encoding |= btrfs_file_extent_encryption(leaf, fi);
3297 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3298
Chris Mason179e29e2007-11-01 11:28:41 -04003299 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04003300 item_end +=
Chris Masondb945352007-10-15 16:15:53 -04003301 btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason179e29e2007-11-01 11:28:41 -04003302 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason179e29e2007-11-01 11:28:41 -04003303 item_end += btrfs_file_extent_inline_len(leaf,
Chris Masonc8b97812008-10-29 14:49:59 -04003304 fi);
Chris Mason39279cc2007-06-12 06:35:45 -04003305 }
Yan008630c2007-11-07 13:31:09 -05003306 item_end--;
Chris Mason39279cc2007-06-12 06:35:45 -04003307 }
Yan, Zheng80825102009-11-12 09:35:36 +00003308 if (found_type > min_type) {
Chris Mason39279cc2007-06-12 06:35:45 -04003309 del_item = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00003310 } else {
3311 if (item_end < new_size)
3312 break;
3313 if (found_key.offset >= new_size)
3314 del_item = 1;
3315 else
3316 del_item = 0;
3317 }
Chris Mason39279cc2007-06-12 06:35:45 -04003318 found_extent = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003319 /* FIXME, shrink the extent if the ref count is only 1 */
Chris Mason179e29e2007-11-01 11:28:41 -04003320 if (found_type != BTRFS_EXTENT_DATA_KEY)
3321 goto delete;
3322
3323 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason39279cc2007-06-12 06:35:45 -04003324 u64 num_dec;
Chris Masondb945352007-10-15 16:15:53 -04003325 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
Chris Mason771ed682008-11-06 22:02:51 -05003326 if (!del_item && !encoding) {
Chris Masondb945352007-10-15 16:15:53 -04003327 u64 orig_num_bytes =
3328 btrfs_file_extent_num_bytes(leaf, fi);
Chris Masone02119d2008-09-05 16:13:11 -04003329 extent_num_bytes = new_size -
Chris Mason5f39d392007-10-15 16:14:19 -04003330 found_key.offset + root->sectorsize - 1;
Yanb1632b12008-01-30 11:54:04 -05003331 extent_num_bytes = extent_num_bytes &
3332 ~((u64)root->sectorsize - 1);
Chris Masondb945352007-10-15 16:15:53 -04003333 btrfs_set_file_extent_num_bytes(leaf, fi,
3334 extent_num_bytes);
3335 num_dec = (orig_num_bytes -
Chris Mason90692182008-02-08 13:49:28 -05003336 extent_num_bytes);
Chris Masone02119d2008-09-05 16:13:11 -04003337 if (root->ref_cows && extent_start != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003338 inode_sub_bytes(inode, num_dec);
Chris Mason5f39d392007-10-15 16:14:19 -04003339 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04003340 } else {
Chris Masondb945352007-10-15 16:15:53 -04003341 extent_num_bytes =
3342 btrfs_file_extent_disk_num_bytes(leaf,
3343 fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003344 extent_offset = found_key.offset -
3345 btrfs_file_extent_offset(leaf, fi);
3346
Chris Mason39279cc2007-06-12 06:35:45 -04003347 /* FIXME blocksize != 4096 */
Chris Mason90692182008-02-08 13:49:28 -05003348 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason39279cc2007-06-12 06:35:45 -04003349 if (extent_start != 0) {
3350 found_extent = 1;
Chris Masone02119d2008-09-05 16:13:11 -04003351 if (root->ref_cows)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003352 inode_sub_bytes(inode, num_dec);
Chris Mason39279cc2007-06-12 06:35:45 -04003353 }
3354 }
Chris Mason90692182008-02-08 13:49:28 -05003355 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -04003356 /*
3357 * we can't truncate inline items that have had
3358 * special encodings
3359 */
3360 if (!del_item &&
3361 btrfs_file_extent_compression(leaf, fi) == 0 &&
3362 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3363 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04003364 u32 size = new_size - found_key.offset;
3365
3366 if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003367 inode_sub_bytes(inode, item_end + 1 -
3368 new_size);
Chris Masone02119d2008-09-05 16:13:11 -04003369 }
3370 size =
3371 btrfs_file_extent_calc_inline_size(size);
Chris Mason90692182008-02-08 13:49:28 -05003372 ret = btrfs_truncate_item(trans, root, path,
Chris Masone02119d2008-09-05 16:13:11 -04003373 size, 1);
Chris Mason90692182008-02-08 13:49:28 -05003374 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04003375 } else if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003376 inode_sub_bytes(inode, item_end + 1 -
3377 found_key.offset);
Chris Mason90692182008-02-08 13:49:28 -05003378 }
Chris Mason39279cc2007-06-12 06:35:45 -04003379 }
Chris Mason179e29e2007-11-01 11:28:41 -04003380delete:
Chris Mason39279cc2007-06-12 06:35:45 -04003381 if (del_item) {
Chris Mason85e21ba2008-01-29 15:11:36 -05003382 if (!pending_del_nr) {
3383 /* no pending yet, add ourselves */
3384 pending_del_slot = path->slots[0];
3385 pending_del_nr = 1;
3386 } else if (pending_del_nr &&
3387 path->slots[0] + 1 == pending_del_slot) {
3388 /* hop on the pending chunk */
3389 pending_del_nr++;
3390 pending_del_slot = path->slots[0];
3391 } else {
Chris Masond3977122009-01-05 21:25:51 -05003392 BUG();
Chris Mason85e21ba2008-01-29 15:11:36 -05003393 }
Chris Mason39279cc2007-06-12 06:35:45 -04003394 } else {
3395 break;
3396 }
Josef Bacik0af3d002010-06-21 14:48:16 -04003397 if (found_extent && (root->ref_cows ||
3398 root == root->fs_info->tree_root)) {
Chris Masonb9473432009-03-13 11:00:37 -04003399 btrfs_set_path_blocking(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003400 ret = btrfs_free_extent(trans, root, extent_start,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003401 extent_num_bytes, 0,
3402 btrfs_header_owner(leaf),
3403 inode->i_ino, extent_offset);
Chris Mason39279cc2007-06-12 06:35:45 -04003404 BUG_ON(ret);
3405 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003406
Yan, Zheng80825102009-11-12 09:35:36 +00003407 if (found_type == BTRFS_INODE_ITEM_KEY)
3408 break;
3409
3410 if (path->slots[0] == 0 ||
3411 path->slots[0] != pending_del_slot) {
3412 if (root->ref_cows) {
3413 err = -EAGAIN;
3414 goto out;
3415 }
3416 if (pending_del_nr) {
3417 ret = btrfs_del_items(trans, root, path,
3418 pending_del_slot,
3419 pending_del_nr);
3420 BUG_ON(ret);
3421 pending_del_nr = 0;
3422 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003423 btrfs_release_path(root, path);
3424 goto search_again;
Yan, Zheng80825102009-11-12 09:35:36 +00003425 } else {
3426 path->slots[0]--;
Chris Mason85e21ba2008-01-29 15:11:36 -05003427 }
Chris Mason39279cc2007-06-12 06:35:45 -04003428 }
Yan, Zheng80825102009-11-12 09:35:36 +00003429out:
Chris Mason85e21ba2008-01-29 15:11:36 -05003430 if (pending_del_nr) {
3431 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3432 pending_del_nr);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003433 BUG_ON(ret);
Chris Mason85e21ba2008-01-29 15:11:36 -05003434 }
Chris Mason39279cc2007-06-12 06:35:45 -04003435 btrfs_free_path(path);
Yan, Zheng80825102009-11-12 09:35:36 +00003436 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003437}
3438
Chris Masona52d9a82007-08-27 16:49:44 -04003439/*
3440 * taken from block_truncate_page, but does cow as it zeros out
3441 * any bytes left in the last page in the file.
3442 */
3443static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3444{
3445 struct inode *inode = mapping->host;
Chris Masondb945352007-10-15 16:15:53 -04003446 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003447 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3448 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003449 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003450 char *kaddr;
Chris Masondb945352007-10-15 16:15:53 -04003451 u32 blocksize = root->sectorsize;
Chris Masona52d9a82007-08-27 16:49:44 -04003452 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3453 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3454 struct page *page;
3455 int ret = 0;
3456 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003457 u64 page_end;
Chris Masona52d9a82007-08-27 16:49:44 -04003458
3459 if ((offset & (blocksize - 1)) == 0)
3460 goto out;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003461 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003462 if (ret)
3463 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04003464
3465 ret = -ENOMEM;
Chris Mason211c17f2008-05-15 09:13:45 -04003466again:
Chris Masona52d9a82007-08-27 16:49:44 -04003467 page = grab_cache_page(mapping, index);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003468 if (!page) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003469 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Masona52d9a82007-08-27 16:49:44 -04003470 goto out;
Josef Bacik5d5e1032009-10-13 16:46:49 -04003471 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04003472
3473 page_start = page_offset(page);
3474 page_end = page_start + PAGE_CACHE_SIZE - 1;
3475
Chris Masona52d9a82007-08-27 16:49:44 -04003476 if (!PageUptodate(page)) {
3477 ret = btrfs_readpage(NULL, page);
3478 lock_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04003479 if (page->mapping != mapping) {
3480 unlock_page(page);
3481 page_cache_release(page);
3482 goto again;
3483 }
Chris Masona52d9a82007-08-27 16:49:44 -04003484 if (!PageUptodate(page)) {
3485 ret = -EIO;
Chris Mason89642222008-07-24 09:41:53 -04003486 goto out_unlock;
Chris Masona52d9a82007-08-27 16:49:44 -04003487 }
3488 }
Chris Mason211c17f2008-05-15 09:13:45 -04003489 wait_on_page_writeback(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003490
Josef Bacik2ac55d42010-02-03 19:33:23 +00003491 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3492 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003493 set_page_extent_mapped(page);
3494
3495 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3496 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00003497 unlock_extent_cached(io_tree, page_start, page_end,
3498 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003499 unlock_page(page);
3500 page_cache_release(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04003501 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003502 btrfs_put_ordered_extent(ordered);
3503 goto again;
3504 }
3505
Josef Bacik2ac55d42010-02-03 19:33:23 +00003506 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Josef Bacik5d5e1032009-10-13 16:46:49 -04003507 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
Josef Bacik2ac55d42010-02-03 19:33:23 +00003508 0, 0, &cached_state, GFP_NOFS);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003509
Josef Bacik2ac55d42010-02-03 19:33:23 +00003510 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3511 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003512 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00003513 unlock_extent_cached(io_tree, page_start, page_end,
3514 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003515 goto out_unlock;
3516 }
3517
Chris Masone6dcd2d2008-07-17 12:53:50 -04003518 ret = 0;
3519 if (offset != PAGE_CACHE_SIZE) {
3520 kaddr = kmap(page);
3521 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3522 flush_dcache_page(page);
3523 kunmap(page);
3524 }
Chris Mason247e7432008-07-17 12:53:51 -04003525 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003526 set_page_dirty(page);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003527 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3528 GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04003529
Chris Mason89642222008-07-24 09:41:53 -04003530out_unlock:
Josef Bacik5d5e1032009-10-13 16:46:49 -04003531 if (ret)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003532 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason39279cc2007-06-12 06:35:45 -04003533 unlock_page(page);
3534 page_cache_release(page);
3535out:
3536 return ret;
3537}
3538
Yan Zheng9036c102008-10-30 14:19:41 -04003539int btrfs_cont_expand(struct inode *inode, loff_t size)
3540{
3541 struct btrfs_trans_handle *trans;
3542 struct btrfs_root *root = BTRFS_I(inode)->root;
3543 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003544 struct extent_map *em = NULL;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003545 struct extent_state *cached_state = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04003546 u64 mask = root->sectorsize - 1;
3547 u64 hole_start = (inode->i_size + mask) & ~mask;
3548 u64 block_end = (size + mask) & ~mask;
3549 u64 last_byte;
3550 u64 cur_offset;
3551 u64 hole_size;
Josef Bacik9ed74f22009-09-11 16:12:44 -04003552 int err = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04003553
3554 if (size <= hole_start)
3555 return 0;
3556
Yan Zheng9036c102008-10-30 14:19:41 -04003557 while (1) {
3558 struct btrfs_ordered_extent *ordered;
3559 btrfs_wait_ordered_range(inode, hole_start,
3560 block_end - hole_start);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003561 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3562 &cached_state, GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003563 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3564 if (!ordered)
3565 break;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003566 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3567 &cached_state, GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003568 btrfs_put_ordered_extent(ordered);
3569 }
3570
Yan Zheng9036c102008-10-30 14:19:41 -04003571 cur_offset = hole_start;
3572 while (1) {
3573 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3574 block_end - cur_offset, 0);
3575 BUG_ON(IS_ERR(em) || !em);
3576 last_byte = min(extent_map_end(em), block_end);
3577 last_byte = (last_byte + mask) & ~mask;
Yan, Zheng80825102009-11-12 09:35:36 +00003578 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
Chris Mason771ed682008-11-06 22:02:51 -05003579 u64 hint_byte = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04003580 hole_size = last_byte - cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00003581
Yan, Zhenga22285a2010-05-16 10:48:46 -04003582 trans = btrfs_start_transaction(root, 2);
3583 if (IS_ERR(trans)) {
3584 err = PTR_ERR(trans);
Chris Mason771ed682008-11-06 22:02:51 -05003585 break;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003586 }
Yan, Zheng80825102009-11-12 09:35:36 +00003587 btrfs_set_trans_block_group(trans, inode);
3588
3589 err = btrfs_drop_extents(trans, inode, cur_offset,
3590 cur_offset + hole_size,
3591 &hint_byte, 1);
3592 BUG_ON(err);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003593
Yan Zheng9036c102008-10-30 14:19:41 -04003594 err = btrfs_insert_file_extent(trans, root,
3595 inode->i_ino, cur_offset, 0,
3596 0, hole_size, 0, hole_size,
3597 0, 0, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003598 BUG_ON(err);
3599
Yan Zheng9036c102008-10-30 14:19:41 -04003600 btrfs_drop_extent_cache(inode, hole_start,
3601 last_byte - 1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003602
3603 btrfs_end_transaction(trans, root);
Yan Zheng9036c102008-10-30 14:19:41 -04003604 }
3605 free_extent_map(em);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003606 em = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04003607 cur_offset = last_byte;
Yan, Zheng80825102009-11-12 09:35:36 +00003608 if (cur_offset >= block_end)
Yan Zheng9036c102008-10-30 14:19:41 -04003609 break;
3610 }
3611
Yan, Zhenga22285a2010-05-16 10:48:46 -04003612 free_extent_map(em);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003613 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3614 GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003615 return err;
3616}
3617
Yan, Zheng80825102009-11-12 09:35:36 +00003618static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3619{
3620 struct btrfs_root *root = BTRFS_I(inode)->root;
3621 struct btrfs_trans_handle *trans;
3622 unsigned long nr;
3623 int ret;
3624
3625 if (attr->ia_size == inode->i_size)
3626 return 0;
3627
3628 if (attr->ia_size > inode->i_size) {
3629 unsigned long limit;
3630 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3631 if (attr->ia_size > inode->i_sb->s_maxbytes)
3632 return -EFBIG;
3633 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3634 send_sig(SIGXFSZ, current, 0);
3635 return -EFBIG;
3636 }
3637 }
3638
Yan, Zhengd68fc572010-05-16 10:49:58 -04003639 trans = btrfs_start_transaction(root, 5);
3640 if (IS_ERR(trans))
3641 return PTR_ERR(trans);
Yan, Zheng80825102009-11-12 09:35:36 +00003642
Yan, Zheng80825102009-11-12 09:35:36 +00003643 btrfs_set_trans_block_group(trans, inode);
3644
3645 ret = btrfs_orphan_add(trans, inode);
3646 BUG_ON(ret);
3647
3648 nr = trans->blocks_used;
3649 btrfs_end_transaction(trans, root);
Yan, Zheng80825102009-11-12 09:35:36 +00003650 btrfs_btree_balance_dirty(root, nr);
3651
3652 if (attr->ia_size > inode->i_size) {
3653 ret = btrfs_cont_expand(inode, attr->ia_size);
3654 if (ret) {
3655 btrfs_truncate(inode);
3656 return ret;
3657 }
3658
3659 i_size_write(inode, attr->ia_size);
3660 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3661
Yan, Zhengd68fc572010-05-16 10:49:58 -04003662 trans = btrfs_start_transaction(root, 0);
3663 BUG_ON(IS_ERR(trans));
Yan, Zheng80825102009-11-12 09:35:36 +00003664 btrfs_set_trans_block_group(trans, inode);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003665 trans->block_rsv = root->orphan_block_rsv;
3666 BUG_ON(!trans->block_rsv);
Yan, Zheng80825102009-11-12 09:35:36 +00003667
3668 ret = btrfs_update_inode(trans, root, inode);
3669 BUG_ON(ret);
3670 if (inode->i_nlink > 0) {
3671 ret = btrfs_orphan_del(trans, inode);
3672 BUG_ON(ret);
3673 }
3674 nr = trans->blocks_used;
3675 btrfs_end_transaction(trans, root);
3676 btrfs_btree_balance_dirty(root, nr);
3677 return 0;
3678 }
3679
3680 /*
3681 * We're truncating a file that used to have good data down to
3682 * zero. Make sure it gets into the ordered flush list so that
3683 * any new writes get down to disk quickly.
3684 */
3685 if (attr->ia_size == 0)
3686 BTRFS_I(inode)->ordered_data_close = 1;
3687
3688 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3689 ret = vmtruncate(inode, attr->ia_size);
3690 BUG_ON(ret);
3691
3692 return 0;
3693}
3694
Chris Mason39279cc2007-06-12 06:35:45 -04003695static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3696{
3697 struct inode *inode = dentry->d_inode;
Li Zefanb83cc962010-12-20 16:04:08 +08003698 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003699 int err;
3700
Li Zefanb83cc962010-12-20 16:04:08 +08003701 if (btrfs_root_readonly(root))
3702 return -EROFS;
3703
Chris Mason39279cc2007-06-12 06:35:45 -04003704 err = inode_change_ok(inode, attr);
3705 if (err)
3706 return err;
3707
Chris Mason5a3f23d2009-03-31 13:27:11 -04003708 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Yan, Zheng80825102009-11-12 09:35:36 +00003709 err = btrfs_setattr_size(inode, attr);
3710 if (err)
3711 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003712 }
Yan Zheng9036c102008-10-30 14:19:41 -04003713
Christoph Hellwig10257742010-06-04 11:30:02 +02003714 if (attr->ia_valid) {
3715 setattr_copy(inode, attr);
3716 mark_inode_dirty(inode);
Josef Bacik33268ea2008-07-24 12:16:36 -04003717
Christoph Hellwig10257742010-06-04 11:30:02 +02003718 if (attr->ia_valid & ATTR_MODE)
3719 err = btrfs_acl_chmod(inode);
3720 }
3721
Chris Mason39279cc2007-06-12 06:35:45 -04003722 return err;
3723}
Chris Mason61295eb2008-01-14 16:24:38 -05003724
Al Virobd555972010-06-07 11:35:40 -04003725void btrfs_evict_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003726{
3727 struct btrfs_trans_handle *trans;
3728 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003729 unsigned long nr;
Chris Mason39279cc2007-06-12 06:35:45 -04003730 int ret;
3731
3732 truncate_inode_pages(&inode->i_data, 0);
Josef Bacik0af3d002010-06-21 14:48:16 -04003733 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3734 root == root->fs_info->tree_root))
Al Virobd555972010-06-07 11:35:40 -04003735 goto no_delete;
3736
Chris Mason39279cc2007-06-12 06:35:45 -04003737 if (is_bad_inode(inode)) {
Josef Bacik7b128762008-07-24 12:17:14 -04003738 btrfs_orphan_del(NULL, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003739 goto no_delete;
3740 }
Al Virobd555972010-06-07 11:35:40 -04003741 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
Chris Mason4a096752008-07-21 10:29:44 -04003742 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Chris Mason5f39d392007-10-15 16:14:19 -04003743
Yan, Zhengc71bf092009-11-12 09:34:40 +00003744 if (root->fs_info->log_root_recovering) {
3745 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3746 goto no_delete;
3747 }
3748
Yan, Zheng76dda932009-09-21 16:00:26 -04003749 if (inode->i_nlink > 0) {
3750 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3751 goto no_delete;
3752 }
3753
Chris Masondbe674a2008-07-17 12:54:05 -04003754 btrfs_i_size_write(inode, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04003755
Yan, Zheng80825102009-11-12 09:35:36 +00003756 while (1) {
Yan, Zhengd68fc572010-05-16 10:49:58 -04003757 trans = btrfs_start_transaction(root, 0);
3758 BUG_ON(IS_ERR(trans));
Yan, Zheng80825102009-11-12 09:35:36 +00003759 btrfs_set_trans_block_group(trans, inode);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003760 trans->block_rsv = root->orphan_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00003761
Yan, Zhengd68fc572010-05-16 10:49:58 -04003762 ret = btrfs_block_rsv_check(trans, root,
3763 root->orphan_block_rsv, 0, 5);
3764 if (ret) {
3765 BUG_ON(ret != -EAGAIN);
3766 ret = btrfs_commit_transaction(trans, root);
3767 BUG_ON(ret);
3768 continue;
3769 }
3770
3771 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003772 if (ret != -EAGAIN)
3773 break;
3774
3775 nr = trans->blocks_used;
3776 btrfs_end_transaction(trans, root);
3777 trans = NULL;
3778 btrfs_btree_balance_dirty(root, nr);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003779
Josef Bacik7b128762008-07-24 12:17:14 -04003780 }
3781
Yan, Zheng80825102009-11-12 09:35:36 +00003782 if (ret == 0) {
3783 ret = btrfs_orphan_del(trans, inode);
3784 BUG_ON(ret);
3785 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003786
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003787 nr = trans->blocks_used;
Chris Mason54aa1f42007-06-22 14:16:25 -04003788 btrfs_end_transaction(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003789 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04003790no_delete:
Al Virobd555972010-06-07 11:35:40 -04003791 end_writeback(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003792 return;
Chris Mason39279cc2007-06-12 06:35:45 -04003793}
3794
3795/*
3796 * this returns the key found in the dir entry in the location pointer.
3797 * If no dir entries were found, location->objectid is 0.
3798 */
3799static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3800 struct btrfs_key *location)
3801{
3802 const char *name = dentry->d_name.name;
3803 int namelen = dentry->d_name.len;
3804 struct btrfs_dir_item *di;
3805 struct btrfs_path *path;
3806 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan0d9f7f32007-10-25 15:48:28 -04003807 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003808
3809 path = btrfs_alloc_path();
3810 BUG_ON(!path);
Chris Mason39544012007-12-12 14:38:19 -05003811
Chris Mason39279cc2007-06-12 06:35:45 -04003812 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3813 namelen, 0);
Yan0d9f7f32007-10-25 15:48:28 -04003814 if (IS_ERR(di))
3815 ret = PTR_ERR(di);
Chris Masond3977122009-01-05 21:25:51 -05003816
3817 if (!di || IS_ERR(di))
Chris Mason39544012007-12-12 14:38:19 -05003818 goto out_err;
Chris Masond3977122009-01-05 21:25:51 -05003819
Chris Mason5f39d392007-10-15 16:14:19 -04003820 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
Chris Mason39279cc2007-06-12 06:35:45 -04003821out:
Chris Mason39279cc2007-06-12 06:35:45 -04003822 btrfs_free_path(path);
3823 return ret;
Chris Mason39544012007-12-12 14:38:19 -05003824out_err:
3825 location->objectid = 0;
3826 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04003827}
3828
3829/*
3830 * when we hit a tree root in a directory, the btrfs part of the inode
3831 * needs to be changed to reflect the root directory of the tree root. This
3832 * is kind of like crossing a mount point.
3833 */
3834static int fixup_tree_root_location(struct btrfs_root *root,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003835 struct inode *dir,
3836 struct dentry *dentry,
3837 struct btrfs_key *location,
3838 struct btrfs_root **sub_root)
Chris Mason39279cc2007-06-12 06:35:45 -04003839{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003840 struct btrfs_path *path;
3841 struct btrfs_root *new_root;
3842 struct btrfs_root_ref *ref;
3843 struct extent_buffer *leaf;
3844 int ret;
3845 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003846
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003847 path = btrfs_alloc_path();
3848 if (!path) {
3849 err = -ENOMEM;
3850 goto out;
3851 }
Chris Mason39279cc2007-06-12 06:35:45 -04003852
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003853 err = -ENOENT;
3854 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3855 BTRFS_I(dir)->root->root_key.objectid,
3856 location->objectid);
3857 if (ret) {
3858 if (ret < 0)
3859 err = ret;
3860 goto out;
3861 }
Chris Mason39279cc2007-06-12 06:35:45 -04003862
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003863 leaf = path->nodes[0];
3864 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3865 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3866 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3867 goto out;
3868
3869 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3870 (unsigned long)(ref + 1),
3871 dentry->d_name.len);
3872 if (ret)
3873 goto out;
3874
3875 btrfs_release_path(root->fs_info->tree_root, path);
3876
3877 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3878 if (IS_ERR(new_root)) {
3879 err = PTR_ERR(new_root);
3880 goto out;
3881 }
3882
3883 if (btrfs_root_refs(&new_root->root_item) == 0) {
3884 err = -ENOENT;
3885 goto out;
3886 }
3887
3888 *sub_root = new_root;
3889 location->objectid = btrfs_root_dirid(&new_root->root_item);
3890 location->type = BTRFS_INODE_ITEM_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04003891 location->offset = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003892 err = 0;
3893out:
3894 btrfs_free_path(path);
3895 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003896}
3897
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003898static void inode_tree_add(struct inode *inode)
3899{
3900 struct btrfs_root *root = BTRFS_I(inode)->root;
3901 struct btrfs_inode *entry;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003902 struct rb_node **p;
3903 struct rb_node *parent;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003904again:
3905 p = &root->inode_tree.rb_node;
3906 parent = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003907
Yan, Zheng76dda932009-09-21 16:00:26 -04003908 if (hlist_unhashed(&inode->i_hash))
3909 return;
3910
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003911 spin_lock(&root->inode_lock);
3912 while (*p) {
3913 parent = *p;
3914 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3915
3916 if (inode->i_ino < entry->vfs_inode.i_ino)
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003917 p = &parent->rb_left;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003918 else if (inode->i_ino > entry->vfs_inode.i_ino)
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003919 p = &parent->rb_right;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003920 else {
3921 WARN_ON(!(entry->vfs_inode.i_state &
Al Viroa4ffdde2010-06-02 17:38:30 -04003922 (I_WILL_FREE | I_FREEING)));
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003923 rb_erase(parent, &root->inode_tree);
3924 RB_CLEAR_NODE(parent);
3925 spin_unlock(&root->inode_lock);
3926 goto again;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003927 }
3928 }
3929 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3930 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3931 spin_unlock(&root->inode_lock);
3932}
3933
3934static void inode_tree_del(struct inode *inode)
3935{
3936 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng76dda932009-09-21 16:00:26 -04003937 int empty = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003938
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003939 spin_lock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003940 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003941 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003942 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Yan, Zheng76dda932009-09-21 16:00:26 -04003943 empty = RB_EMPTY_ROOT(&root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003944 }
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003945 spin_unlock(&root->inode_lock);
Yan, Zheng76dda932009-09-21 16:00:26 -04003946
Josef Bacik0af3d002010-06-21 14:48:16 -04003947 /*
3948 * Free space cache has inodes in the tree root, but the tree root has a
3949 * root_refs of 0, so this could end up dropping the tree root as a
3950 * snapshot, so we need the extra !root->fs_info->tree_root check to
3951 * make sure we don't drop it.
3952 */
3953 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3954 root != root->fs_info->tree_root) {
Yan, Zheng76dda932009-09-21 16:00:26 -04003955 synchronize_srcu(&root->fs_info->subvol_srcu);
3956 spin_lock(&root->inode_lock);
3957 empty = RB_EMPTY_ROOT(&root->inode_tree);
3958 spin_unlock(&root->inode_lock);
3959 if (empty)
3960 btrfs_add_dead_root(root);
3961 }
3962}
3963
3964int btrfs_invalidate_inodes(struct btrfs_root *root)
3965{
3966 struct rb_node *node;
3967 struct rb_node *prev;
3968 struct btrfs_inode *entry;
3969 struct inode *inode;
3970 u64 objectid = 0;
3971
3972 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3973
3974 spin_lock(&root->inode_lock);
3975again:
3976 node = root->inode_tree.rb_node;
3977 prev = NULL;
3978 while (node) {
3979 prev = node;
3980 entry = rb_entry(node, struct btrfs_inode, rb_node);
3981
3982 if (objectid < entry->vfs_inode.i_ino)
3983 node = node->rb_left;
3984 else if (objectid > entry->vfs_inode.i_ino)
3985 node = node->rb_right;
3986 else
3987 break;
3988 }
3989 if (!node) {
3990 while (prev) {
3991 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3992 if (objectid <= entry->vfs_inode.i_ino) {
3993 node = prev;
3994 break;
3995 }
3996 prev = rb_next(prev);
3997 }
3998 }
3999 while (node) {
4000 entry = rb_entry(node, struct btrfs_inode, rb_node);
4001 objectid = entry->vfs_inode.i_ino + 1;
4002 inode = igrab(&entry->vfs_inode);
4003 if (inode) {
4004 spin_unlock(&root->inode_lock);
4005 if (atomic_read(&inode->i_count) > 1)
4006 d_prune_aliases(inode);
4007 /*
Al Viro45321ac2010-06-07 13:43:19 -04004008 * btrfs_drop_inode will have it removed from
Yan, Zheng76dda932009-09-21 16:00:26 -04004009 * the inode cache when its usage count
4010 * hits zero.
4011 */
4012 iput(inode);
4013 cond_resched();
4014 spin_lock(&root->inode_lock);
4015 goto again;
4016 }
4017
4018 if (cond_resched_lock(&root->inode_lock))
4019 goto again;
4020
4021 node = rb_next(node);
4022 }
4023 spin_unlock(&root->inode_lock);
4024 return 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004025}
4026
Chris Masone02119d2008-09-05 16:13:11 -04004027static int btrfs_init_locked_inode(struct inode *inode, void *p)
4028{
4029 struct btrfs_iget_args *args = p;
4030 inode->i_ino = args->ino;
Chris Masone02119d2008-09-05 16:13:11 -04004031 BTRFS_I(inode)->root = args->root;
Josef Bacik6a632092009-02-20 11:00:09 -05004032 btrfs_set_inode_space_info(args->root, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004033 return 0;
4034}
4035
4036static int btrfs_find_actor(struct inode *inode, void *opaque)
4037{
4038 struct btrfs_iget_args *args = opaque;
Chris Masond3977122009-01-05 21:25:51 -05004039 return args->ino == inode->i_ino &&
4040 args->root == BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004041}
4042
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004043static struct inode *btrfs_iget_locked(struct super_block *s,
4044 u64 objectid,
4045 struct btrfs_root *root)
Chris Mason39279cc2007-06-12 06:35:45 -04004046{
4047 struct inode *inode;
4048 struct btrfs_iget_args args;
4049 args.ino = objectid;
4050 args.root = root;
4051
4052 inode = iget5_locked(s, objectid, btrfs_find_actor,
4053 btrfs_init_locked_inode,
4054 (void *)&args);
4055 return inode;
4056}
4057
Balaji Rao1a54ef82008-07-21 02:01:04 +05304058/* Get an inode object given its location and corresponding root.
4059 * Returns in *is_new if the inode was read from disk
4060 */
4061struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Josef Bacik73f73412009-12-04 17:38:27 +00004062 struct btrfs_root *root, int *new)
Balaji Rao1a54ef82008-07-21 02:01:04 +05304063{
4064 struct inode *inode;
4065
4066 inode = btrfs_iget_locked(s, location->objectid, root);
4067 if (!inode)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004068 return ERR_PTR(-ENOMEM);
Balaji Rao1a54ef82008-07-21 02:01:04 +05304069
4070 if (inode->i_state & I_NEW) {
4071 BTRFS_I(inode)->root = root;
4072 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4073 btrfs_read_locked_inode(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004074
4075 inode_tree_add(inode);
Balaji Rao1a54ef82008-07-21 02:01:04 +05304076 unlock_new_inode(inode);
Josef Bacik73f73412009-12-04 17:38:27 +00004077 if (new)
4078 *new = 1;
Balaji Rao1a54ef82008-07-21 02:01:04 +05304079 }
4080
4081 return inode;
4082}
4083
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004084static struct inode *new_simple_dir(struct super_block *s,
4085 struct btrfs_key *key,
4086 struct btrfs_root *root)
4087{
4088 struct inode *inode = new_inode(s);
4089
4090 if (!inode)
4091 return ERR_PTR(-ENOMEM);
4092
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004093 BTRFS_I(inode)->root = root;
4094 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4095 BTRFS_I(inode)->dummy_inode = 1;
4096
4097 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4098 inode->i_op = &simple_dir_inode_operations;
4099 inode->i_fop = &simple_dir_operations;
4100 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4101 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4102
4103 return inode;
4104}
4105
Chris Mason3de45862008-11-17 21:02:50 -05004106struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04004107{
Chris Masond3977122009-01-05 21:25:51 -05004108 struct inode *inode;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004109 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004110 struct btrfs_root *sub_root = root;
4111 struct btrfs_key location;
Yan, Zheng76dda932009-09-21 16:00:26 -04004112 int index;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004113 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004114
Yan, Zheng76dda932009-09-21 16:00:26 -04004115 dentry->d_op = &btrfs_dentry_operations;
4116
Chris Mason39279cc2007-06-12 06:35:45 -04004117 if (dentry->d_name.len > BTRFS_NAME_LEN)
4118 return ERR_PTR(-ENAMETOOLONG);
Chris Mason5f39d392007-10-15 16:14:19 -04004119
Chris Mason39279cc2007-06-12 06:35:45 -04004120 ret = btrfs_inode_by_name(dir, dentry, &location);
Chris Mason5f39d392007-10-15 16:14:19 -04004121
Chris Mason39279cc2007-06-12 06:35:45 -04004122 if (ret < 0)
4123 return ERR_PTR(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04004124
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004125 if (location.objectid == 0)
4126 return NULL;
4127
4128 if (location.type == BTRFS_INODE_ITEM_KEY) {
Josef Bacik73f73412009-12-04 17:38:27 +00004129 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004130 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04004131 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004132
4133 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4134
Yan, Zheng76dda932009-09-21 16:00:26 -04004135 index = srcu_read_lock(&root->fs_info->subvol_srcu);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004136 ret = fixup_tree_root_location(root, dir, dentry,
4137 &location, &sub_root);
4138 if (ret < 0) {
4139 if (ret != -ENOENT)
4140 inode = ERR_PTR(ret);
4141 else
4142 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4143 } else {
Josef Bacik73f73412009-12-04 17:38:27 +00004144 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04004145 }
Yan, Zheng76dda932009-09-21 16:00:26 -04004146 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4147
Julia Lawall34d19ba2011-01-24 19:55:19 +00004148 if (!IS_ERR(inode) && root != sub_root) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00004149 down_read(&root->fs_info->cleanup_work_sem);
4150 if (!(inode->i_sb->s_flags & MS_RDONLY))
4151 btrfs_orphan_cleanup(sub_root);
4152 up_read(&root->fs_info->cleanup_work_sem);
4153 }
4154
Chris Mason3de45862008-11-17 21:02:50 -05004155 return inode;
4156}
4157
Yan, Zheng76dda932009-09-21 16:00:26 -04004158static int btrfs_dentry_delete(struct dentry *dentry)
4159{
4160 struct btrfs_root *root;
4161
Yan, Zhengefefb142009-10-09 09:25:16 -04004162 if (!dentry->d_inode && !IS_ROOT(dentry))
4163 dentry = dentry->d_parent;
Yan, Zheng76dda932009-09-21 16:00:26 -04004164
Yan, Zhengefefb142009-10-09 09:25:16 -04004165 if (dentry->d_inode) {
4166 root = BTRFS_I(dentry->d_inode)->root;
4167 if (btrfs_root_refs(&root->root_item) == 0)
4168 return 1;
4169 }
Yan, Zheng76dda932009-09-21 16:00:26 -04004170 return 0;
4171}
4172
Chris Mason3de45862008-11-17 21:02:50 -05004173static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4174 struct nameidata *nd)
4175{
4176 struct inode *inode;
4177
Chris Mason3de45862008-11-17 21:02:50 -05004178 inode = btrfs_lookup_dentry(dir, dentry);
4179 if (IS_ERR(inode))
4180 return ERR_CAST(inode);
Josef Bacik7b128762008-07-24 12:17:14 -04004181
Chris Mason39279cc2007-06-12 06:35:45 -04004182 return d_splice_alias(inode, dentry);
4183}
4184
Chris Mason39279cc2007-06-12 06:35:45 -04004185static unsigned char btrfs_filetype_table[] = {
4186 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4187};
4188
David Woodhousecbdf5a22008-08-06 19:42:33 +01004189static int btrfs_real_readdir(struct file *filp, void *dirent,
4190 filldir_t filldir)
Chris Mason39279cc2007-06-12 06:35:45 -04004191{
Chris Mason6da6aba2007-12-18 16:15:09 -05004192 struct inode *inode = filp->f_dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04004193 struct btrfs_root *root = BTRFS_I(inode)->root;
4194 struct btrfs_item *item;
4195 struct btrfs_dir_item *di;
4196 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04004197 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04004198 struct btrfs_path *path;
4199 int ret;
4200 u32 nritems;
Chris Mason5f39d392007-10-15 16:14:19 -04004201 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04004202 int slot;
4203 int advance;
4204 unsigned char d_type;
4205 int over = 0;
4206 u32 di_cur;
4207 u32 di_total;
4208 u32 di_len;
4209 int key_type = BTRFS_DIR_INDEX_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04004210 char tmp_name[32];
4211 char *name_ptr;
4212 int name_len;
Chris Mason39279cc2007-06-12 06:35:45 -04004213
4214 /* FIXME, use a real flag for deciding about the key type */
4215 if (root->fs_info->tree_root == root)
4216 key_type = BTRFS_DIR_ITEM_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04004217
Chris Mason39544012007-12-12 14:38:19 -05004218 /* special case for "." */
4219 if (filp->f_pos == 0) {
4220 over = filldir(dirent, ".", 1,
4221 1, inode->i_ino,
4222 DT_DIR);
4223 if (over)
4224 return 0;
4225 filp->f_pos = 1;
4226 }
Chris Mason39544012007-12-12 14:38:19 -05004227 /* special case for .., just use the back ref */
4228 if (filp->f_pos == 1) {
David Woodhouse5ecc7e52008-08-17 15:14:48 +01004229 u64 pino = parent_ino(filp->f_path.dentry);
Chris Mason39544012007-12-12 14:38:19 -05004230 over = filldir(dirent, "..", 2,
David Woodhouse5ecc7e52008-08-17 15:14:48 +01004231 2, pino, DT_DIR);
Chris Mason39544012007-12-12 14:38:19 -05004232 if (over)
David Woodhouse49593bf2008-08-17 17:08:36 +01004233 return 0;
Chris Mason39544012007-12-12 14:38:19 -05004234 filp->f_pos = 2;
4235 }
David Woodhouse49593bf2008-08-17 17:08:36 +01004236 path = btrfs_alloc_path();
4237 path->reada = 2;
4238
Chris Mason39279cc2007-06-12 06:35:45 -04004239 btrfs_set_key_type(&key, key_type);
4240 key.offset = filp->f_pos;
David Woodhouse49593bf2008-08-17 17:08:36 +01004241 key.objectid = inode->i_ino;
Chris Mason5f39d392007-10-15 16:14:19 -04004242
Chris Mason39279cc2007-06-12 06:35:45 -04004243 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4244 if (ret < 0)
4245 goto err;
4246 advance = 0;
David Woodhouse49593bf2008-08-17 17:08:36 +01004247
4248 while (1) {
Chris Mason5f39d392007-10-15 16:14:19 -04004249 leaf = path->nodes[0];
4250 nritems = btrfs_header_nritems(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04004251 slot = path->slots[0];
4252 if (advance || slot >= nritems) {
David Woodhouse49593bf2008-08-17 17:08:36 +01004253 if (slot >= nritems - 1) {
Chris Mason39279cc2007-06-12 06:35:45 -04004254 ret = btrfs_next_leaf(root, path);
4255 if (ret)
4256 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004257 leaf = path->nodes[0];
4258 nritems = btrfs_header_nritems(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04004259 slot = path->slots[0];
4260 } else {
4261 slot++;
4262 path->slots[0]++;
4263 }
4264 }
Chris Mason3de45862008-11-17 21:02:50 -05004265
Chris Mason39279cc2007-06-12 06:35:45 -04004266 advance = 1;
Chris Mason5f39d392007-10-15 16:14:19 -04004267 item = btrfs_item_nr(leaf, slot);
4268 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4269
4270 if (found_key.objectid != key.objectid)
Chris Mason39279cc2007-06-12 06:35:45 -04004271 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004272 if (btrfs_key_type(&found_key) != key_type)
Chris Mason39279cc2007-06-12 06:35:45 -04004273 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004274 if (found_key.offset < filp->f_pos)
Chris Mason39279cc2007-06-12 06:35:45 -04004275 continue;
Chris Mason5f39d392007-10-15 16:14:19 -04004276
4277 filp->f_pos = found_key.offset;
David Woodhouse49593bf2008-08-17 17:08:36 +01004278
Chris Mason39279cc2007-06-12 06:35:45 -04004279 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4280 di_cur = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04004281 di_total = btrfs_item_size(leaf, item);
David Woodhouse49593bf2008-08-17 17:08:36 +01004282
4283 while (di_cur < di_total) {
Chris Mason5f39d392007-10-15 16:14:19 -04004284 struct btrfs_key location;
4285
4286 name_len = btrfs_dir_name_len(leaf, di);
David Woodhouse49593bf2008-08-17 17:08:36 +01004287 if (name_len <= sizeof(tmp_name)) {
Chris Mason5f39d392007-10-15 16:14:19 -04004288 name_ptr = tmp_name;
4289 } else {
4290 name_ptr = kmalloc(name_len, GFP_NOFS);
David Woodhouse49593bf2008-08-17 17:08:36 +01004291 if (!name_ptr) {
4292 ret = -ENOMEM;
4293 goto err;
4294 }
Chris Mason5f39d392007-10-15 16:14:19 -04004295 }
4296 read_extent_buffer(leaf, name_ptr,
4297 (unsigned long)(di + 1), name_len);
4298
4299 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4300 btrfs_dir_item_key_to_cpu(leaf, di, &location);
Chris Mason3de45862008-11-17 21:02:50 -05004301
4302 /* is this a reference to our own snapshot? If so
4303 * skip it
4304 */
4305 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4306 location.objectid == root->root_key.objectid) {
4307 over = 0;
4308 goto skip;
4309 }
Chris Mason5f39d392007-10-15 16:14:19 -04004310 over = filldir(dirent, name_ptr, name_len,
David Woodhouse49593bf2008-08-17 17:08:36 +01004311 found_key.offset, location.objectid,
Chris Mason39279cc2007-06-12 06:35:45 -04004312 d_type);
Chris Mason5f39d392007-10-15 16:14:19 -04004313
Chris Mason3de45862008-11-17 21:02:50 -05004314skip:
Chris Mason5f39d392007-10-15 16:14:19 -04004315 if (name_ptr != tmp_name)
4316 kfree(name_ptr);
4317
Chris Mason39279cc2007-06-12 06:35:45 -04004318 if (over)
4319 goto nopos;
Josef Bacik5103e942007-11-16 11:45:54 -05004320 di_len = btrfs_dir_name_len(leaf, di) +
David Woodhouse49593bf2008-08-17 17:08:36 +01004321 btrfs_dir_data_len(leaf, di) + sizeof(*di);
Chris Mason39279cc2007-06-12 06:35:45 -04004322 di_cur += di_len;
4323 di = (struct btrfs_dir_item *)((char *)di + di_len);
4324 }
4325 }
David Woodhouse49593bf2008-08-17 17:08:36 +01004326
4327 /* Reached end of directory/root. Bump pos past the last item. */
Yan Zheng5e591a02008-02-19 11:41:02 -05004328 if (key_type == BTRFS_DIR_INDEX_KEY)
Jan Engelhardt406266a2009-12-09 22:00:38 +00004329 /*
4330 * 32-bit glibc will use getdents64, but then strtol -
4331 * so the last number we can serve is this.
4332 */
4333 filp->f_pos = 0x7fffffff;
Yan Zheng5e591a02008-02-19 11:41:02 -05004334 else
4335 filp->f_pos++;
Chris Mason39279cc2007-06-12 06:35:45 -04004336nopos:
4337 ret = 0;
4338err:
Chris Mason39279cc2007-06-12 06:35:45 -04004339 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04004340 return ret;
4341}
4342
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004343int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Chris Mason39279cc2007-06-12 06:35:45 -04004344{
4345 struct btrfs_root *root = BTRFS_I(inode)->root;
4346 struct btrfs_trans_handle *trans;
4347 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04004348 bool nolock = false;
Chris Mason39279cc2007-06-12 06:35:45 -04004349
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004350 if (BTRFS_I(inode)->dummy_inode)
Chris Mason4ca8b412008-08-05 13:30:48 -04004351 return 0;
4352
Josef Bacik0af3d002010-06-21 14:48:16 -04004353 smp_mb();
4354 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4355
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004356 if (wbc->sync_mode == WB_SYNC_ALL) {
Josef Bacik0af3d002010-06-21 14:48:16 -04004357 if (nolock)
4358 trans = btrfs_join_transaction_nolock(root, 1);
4359 else
4360 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00004361 if (IS_ERR(trans))
4362 return PTR_ERR(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04004363 btrfs_set_trans_block_group(trans, inode);
Josef Bacik0af3d002010-06-21 14:48:16 -04004364 if (nolock)
4365 ret = btrfs_end_transaction_nolock(trans, root);
4366 else
4367 ret = btrfs_commit_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004368 }
4369 return ret;
4370}
4371
4372/*
Chris Mason54aa1f42007-06-22 14:16:25 -04004373 * This is somewhat expensive, updating the tree every time the
Chris Mason39279cc2007-06-12 06:35:45 -04004374 * inode changes. But, it is most likely to find the inode in cache.
4375 * FIXME, needs more benchmarking...there are no reasons other than performance
4376 * to keep or drop this code.
4377 */
4378void btrfs_dirty_inode(struct inode *inode)
4379{
4380 struct btrfs_root *root = BTRFS_I(inode)->root;
4381 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004382 int ret;
4383
4384 if (BTRFS_I(inode)->dummy_inode)
4385 return;
Chris Mason39279cc2007-06-12 06:35:45 -04004386
Chris Masonf9295742008-07-17 12:54:14 -04004387 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00004388 BUG_ON(IS_ERR(trans));
Chris Mason39279cc2007-06-12 06:35:45 -04004389 btrfs_set_trans_block_group(trans, inode);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004390
4391 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04004392 if (ret && ret == -ENOSPC) {
4393 /* whoops, lets try again with the full transaction */
4394 btrfs_end_transaction(trans, root);
4395 trans = btrfs_start_transaction(root, 1);
Chris Mason9aeead72010-05-27 10:23:00 -04004396 if (IS_ERR(trans)) {
4397 if (printk_ratelimit()) {
4398 printk(KERN_ERR "btrfs: fail to "
4399 "dirty inode %lu error %ld\n",
4400 inode->i_ino, PTR_ERR(trans));
4401 }
4402 return;
4403 }
Chris Mason94b60442010-05-26 11:02:00 -04004404 btrfs_set_trans_block_group(trans, inode);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004405
Chris Mason94b60442010-05-26 11:02:00 -04004406 ret = btrfs_update_inode(trans, root, inode);
4407 if (ret) {
Chris Mason9aeead72010-05-27 10:23:00 -04004408 if (printk_ratelimit()) {
4409 printk(KERN_ERR "btrfs: fail to "
4410 "dirty inode %lu error %d\n",
4411 inode->i_ino, ret);
4412 }
Chris Mason94b60442010-05-26 11:02:00 -04004413 }
4414 }
Chris Mason39279cc2007-06-12 06:35:45 -04004415 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004416}
4417
Chris Masond352ac62008-09-29 15:18:18 -04004418/*
4419 * find the highest existing sequence number in a directory
4420 * and then set the in-memory index_cnt variable to reflect
4421 * free sequence numbers
4422 */
Josef Bacikaec74772008-07-24 12:12:38 -04004423static int btrfs_set_inode_index_count(struct inode *inode)
4424{
4425 struct btrfs_root *root = BTRFS_I(inode)->root;
4426 struct btrfs_key key, found_key;
4427 struct btrfs_path *path;
4428 struct extent_buffer *leaf;
4429 int ret;
4430
4431 key.objectid = inode->i_ino;
4432 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4433 key.offset = (u64)-1;
4434
4435 path = btrfs_alloc_path();
4436 if (!path)
4437 return -ENOMEM;
4438
4439 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4440 if (ret < 0)
4441 goto out;
4442 /* FIXME: we should be able to handle this */
4443 if (ret == 0)
4444 goto out;
4445 ret = 0;
4446
4447 /*
4448 * MAGIC NUMBER EXPLANATION:
4449 * since we search a directory based on f_pos we have to start at 2
4450 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4451 * else has to start at 2
4452 */
4453 if (path->slots[0] == 0) {
4454 BTRFS_I(inode)->index_cnt = 2;
4455 goto out;
4456 }
4457
4458 path->slots[0]--;
4459
4460 leaf = path->nodes[0];
4461 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4462
4463 if (found_key.objectid != inode->i_ino ||
4464 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4465 BTRFS_I(inode)->index_cnt = 2;
4466 goto out;
4467 }
4468
4469 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4470out:
4471 btrfs_free_path(path);
4472 return ret;
4473}
4474
Chris Masond352ac62008-09-29 15:18:18 -04004475/*
4476 * helper to find a free sequence number in a given directory. This current
4477 * code is very simple, later versions will do smarter things in the btree
4478 */
Chris Mason3de45862008-11-17 21:02:50 -05004479int btrfs_set_inode_index(struct inode *dir, u64 *index)
Josef Bacikaec74772008-07-24 12:12:38 -04004480{
4481 int ret = 0;
4482
4483 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4484 ret = btrfs_set_inode_index_count(dir);
Chris Masond3977122009-01-05 21:25:51 -05004485 if (ret)
Josef Bacikaec74772008-07-24 12:12:38 -04004486 return ret;
4487 }
4488
Chris Mason00e4e6b2008-08-05 11:18:09 -04004489 *index = BTRFS_I(dir)->index_cnt;
Josef Bacikaec74772008-07-24 12:12:38 -04004490 BTRFS_I(dir)->index_cnt++;
4491
4492 return ret;
4493}
4494
Chris Mason39279cc2007-06-12 06:35:45 -04004495static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4496 struct btrfs_root *root,
Josef Bacikaec74772008-07-24 12:12:38 -04004497 struct inode *dir,
Chris Mason9c583092008-01-29 15:15:18 -05004498 const char *name, int name_len,
Yan Zhengd2fb3432008-12-11 16:30:39 -05004499 u64 ref_objectid, u64 objectid,
4500 u64 alloc_hint, int mode, u64 *index)
Chris Mason39279cc2007-06-12 06:35:45 -04004501{
4502 struct inode *inode;
Chris Mason5f39d392007-10-15 16:14:19 -04004503 struct btrfs_inode_item *inode_item;
Chris Mason39279cc2007-06-12 06:35:45 -04004504 struct btrfs_key *location;
Chris Mason5f39d392007-10-15 16:14:19 -04004505 struct btrfs_path *path;
Chris Mason9c583092008-01-29 15:15:18 -05004506 struct btrfs_inode_ref *ref;
4507 struct btrfs_key key[2];
4508 u32 sizes[2];
4509 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04004510 int ret;
4511 int owner;
4512
Chris Mason5f39d392007-10-15 16:14:19 -04004513 path = btrfs_alloc_path();
4514 BUG_ON(!path);
4515
Chris Mason39279cc2007-06-12 06:35:45 -04004516 inode = new_inode(root->fs_info->sb);
4517 if (!inode)
4518 return ERR_PTR(-ENOMEM);
4519
Josef Bacikaec74772008-07-24 12:12:38 -04004520 if (dir) {
Chris Mason3de45862008-11-17 21:02:50 -05004521 ret = btrfs_set_inode_index(dir, index);
Shen Feng09771432009-04-02 16:46:06 -04004522 if (ret) {
4523 iput(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004524 return ERR_PTR(ret);
Shen Feng09771432009-04-02 16:46:06 -04004525 }
Josef Bacikaec74772008-07-24 12:12:38 -04004526 }
4527 /*
4528 * index_cnt is ignored for everything but a dir,
4529 * btrfs_get_inode_index_count has an explanation for the magic
4530 * number
4531 */
4532 BTRFS_I(inode)->index_cnt = 2;
Chris Mason39279cc2007-06-12 06:35:45 -04004533 BTRFS_I(inode)->root = root;
Chris Masone02119d2008-09-05 16:13:11 -04004534 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik76195852010-11-19 02:18:02 +00004535 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik6a632092009-02-20 11:00:09 -05004536 btrfs_set_inode_space_info(root, inode);
Chris Masonb888db22007-08-27 16:49:44 -04004537
Chris Mason39279cc2007-06-12 06:35:45 -04004538 if (mode & S_IFDIR)
4539 owner = 0;
4540 else
4541 owner = 1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05004542 BTRFS_I(inode)->block_group =
4543 btrfs_find_block_group(root, 0, alloc_hint, owner);
Chris Mason9c583092008-01-29 15:15:18 -05004544
4545 key[0].objectid = objectid;
4546 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4547 key[0].offset = 0;
4548
4549 key[1].objectid = objectid;
4550 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4551 key[1].offset = ref_objectid;
4552
4553 sizes[0] = sizeof(struct btrfs_inode_item);
4554 sizes[1] = name_len + sizeof(*ref);
4555
Chris Masonb9473432009-03-13 11:00:37 -04004556 path->leave_spinning = 1;
Chris Mason9c583092008-01-29 15:15:18 -05004557 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4558 if (ret != 0)
Chris Mason5f39d392007-10-15 16:14:19 -04004559 goto fail;
4560
Dmitry Monakhovecc11fab2010-03-04 17:31:47 +03004561 inode_init_owner(inode, dir, mode);
Chris Mason39279cc2007-06-12 06:35:45 -04004562 inode->i_ino = objectid;
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004563 inode_set_bytes(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004564 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
Chris Mason5f39d392007-10-15 16:14:19 -04004565 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4566 struct btrfs_inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04004567 fill_inode_item(trans, path->nodes[0], inode_item, inode);
Chris Mason9c583092008-01-29 15:15:18 -05004568
4569 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4570 struct btrfs_inode_ref);
4571 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
Chris Mason00e4e6b2008-08-05 11:18:09 -04004572 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
Chris Mason9c583092008-01-29 15:15:18 -05004573 ptr = (unsigned long)(ref + 1);
4574 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4575
Chris Mason5f39d392007-10-15 16:14:19 -04004576 btrfs_mark_buffer_dirty(path->nodes[0]);
4577 btrfs_free_path(path);
4578
Chris Mason39279cc2007-06-12 06:35:45 -04004579 location = &BTRFS_I(inode)->location;
4580 location->objectid = objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04004581 location->offset = 0;
4582 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4583
Christoph Hellwig6cbff002009-04-17 10:37:41 +02004584 btrfs_inherit_iflags(inode, dir);
4585
Chris Mason94272162009-07-02 12:26:06 -04004586 if ((mode & S_IFREG)) {
4587 if (btrfs_test_opt(root, NODATASUM))
4588 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4589 if (btrfs_test_opt(root, NODATACOW))
4590 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4591 }
4592
Chris Mason39279cc2007-06-12 06:35:45 -04004593 insert_inode_hash(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004594 inode_tree_add(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004595 return inode;
Chris Mason5f39d392007-10-15 16:14:19 -04004596fail:
Josef Bacikaec74772008-07-24 12:12:38 -04004597 if (dir)
4598 BTRFS_I(dir)->index_cnt--;
Chris Mason5f39d392007-10-15 16:14:19 -04004599 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04004600 iput(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004601 return ERR_PTR(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04004602}
4603
4604static inline u8 btrfs_inode_type(struct inode *inode)
4605{
4606 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4607}
4608
Chris Masond352ac62008-09-29 15:18:18 -04004609/*
4610 * utility function to add 'inode' into 'parent_inode' with
4611 * a give name and a given sequence number.
4612 * if 'add_backref' is true, also insert a backref from the
4613 * inode to the parent directory.
4614 */
Chris Masone02119d2008-09-05 16:13:11 -04004615int btrfs_add_link(struct btrfs_trans_handle *trans,
4616 struct inode *parent_inode, struct inode *inode,
4617 const char *name, int name_len, int add_backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04004618{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004619 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004620 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04004621 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
Chris Mason5f39d392007-10-15 16:14:19 -04004622
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004623 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4624 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4625 } else {
4626 key.objectid = inode->i_ino;
4627 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4628 key.offset = 0;
4629 }
Chris Mason39279cc2007-06-12 06:35:45 -04004630
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004631 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4632 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4633 key.objectid, root->root_key.objectid,
4634 parent_inode->i_ino,
4635 index, name, name_len);
4636 } else if (add_backref) {
4637 ret = btrfs_insert_inode_ref(trans, root,
4638 name, name_len, inode->i_ino,
4639 parent_inode->i_ino, index);
4640 }
4641
Chris Mason39279cc2007-06-12 06:35:45 -04004642 if (ret == 0) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004643 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4644 parent_inode->i_ino, &key,
4645 btrfs_inode_type(inode), index);
4646 BUG_ON(ret);
4647
Chris Masondbe674a2008-07-17 12:54:05 -04004648 btrfs_i_size_write(parent_inode, parent_inode->i_size +
Chris Masone02119d2008-09-05 16:13:11 -04004649 name_len * 2);
Chris Mason79c44582007-06-25 10:09:33 -04004650 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
Chris Masone02119d2008-09-05 16:13:11 -04004651 ret = btrfs_update_inode(trans, root, parent_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004652 }
4653 return ret;
4654}
4655
4656static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
Josef Bacika1b075d2010-11-19 20:36:11 +00004657 struct inode *dir, struct dentry *dentry,
4658 struct inode *inode, int backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04004659{
Josef Bacika1b075d2010-11-19 20:36:11 +00004660 int err = btrfs_add_link(trans, dir, inode,
4661 dentry->d_name.name, dentry->d_name.len,
4662 backref, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004663 if (!err) {
4664 d_instantiate(dentry, inode);
4665 return 0;
4666 }
4667 if (err > 0)
4668 err = -EEXIST;
4669 return err;
4670}
4671
Josef Bacik618e21d2007-07-11 10:18:17 -04004672static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4673 int mode, dev_t rdev)
4674{
4675 struct btrfs_trans_handle *trans;
4676 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05004677 struct inode *inode = NULL;
Josef Bacik618e21d2007-07-11 10:18:17 -04004678 int err;
4679 int drop_inode = 0;
4680 u64 objectid;
Chris Mason1832a6d2007-12-21 16:27:21 -05004681 unsigned long nr = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004682 u64 index = 0;
Josef Bacik618e21d2007-07-11 10:18:17 -04004683
4684 if (!new_valid_dev(rdev))
4685 return -EINVAL;
4686
Yan, Zhenga22285a2010-05-16 10:48:46 -04004687 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4688 if (err)
4689 return err;
4690
Josef Bacik9ed74f22009-09-11 16:12:44 -04004691 /*
4692 * 2 for inode item and ref
4693 * 2 for dir items
4694 * 1 for xattr if selinux is on
4695 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004696 trans = btrfs_start_transaction(root, 5);
4697 if (IS_ERR(trans))
4698 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05004699
Josef Bacik618e21d2007-07-11 10:18:17 -04004700 btrfs_set_trans_block_group(trans, dir);
4701
Josef Bacikaec74772008-07-24 12:12:38 -04004702 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Josef Bacika1b075d2010-11-19 20:36:11 +00004703 dentry->d_name.len, dir->i_ino, objectid,
Chris Mason00e4e6b2008-08-05 11:18:09 -04004704 BTRFS_I(dir)->block_group, mode, &index);
Josef Bacik618e21d2007-07-11 10:18:17 -04004705 err = PTR_ERR(inode);
4706 if (IS_ERR(inode))
4707 goto out_unlock;
4708
Yan, Zhengf34f57a2009-11-12 09:35:27 +00004709 err = btrfs_init_inode_security(trans, inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04004710 if (err) {
4711 drop_inode = 1;
4712 goto out_unlock;
4713 }
4714
Josef Bacik618e21d2007-07-11 10:18:17 -04004715 btrfs_set_trans_block_group(trans, inode);
Josef Bacika1b075d2010-11-19 20:36:11 +00004716 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Josef Bacik618e21d2007-07-11 10:18:17 -04004717 if (err)
4718 drop_inode = 1;
4719 else {
4720 inode->i_op = &btrfs_special_inode_operations;
4721 init_special_inode(inode, inode->i_mode, rdev);
Yan1b4ab1b2007-08-29 09:11:44 -04004722 btrfs_update_inode(trans, root, inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04004723 }
Josef Bacik618e21d2007-07-11 10:18:17 -04004724 btrfs_update_inode_block_group(trans, inode);
4725 btrfs_update_inode_block_group(trans, dir);
4726out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004727 nr = trans->blocks_used;
Chris Mason89ce8a62008-06-25 16:01:31 -04004728 btrfs_end_transaction_throttle(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004729 btrfs_btree_balance_dirty(root, nr);
Josef Bacik618e21d2007-07-11 10:18:17 -04004730 if (drop_inode) {
4731 inode_dec_link_count(inode);
4732 iput(inode);
4733 }
Josef Bacik618e21d2007-07-11 10:18:17 -04004734 return err;
4735}
4736
Chris Mason39279cc2007-06-12 06:35:45 -04004737static int btrfs_create(struct inode *dir, struct dentry *dentry,
4738 int mode, struct nameidata *nd)
4739{
4740 struct btrfs_trans_handle *trans;
4741 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05004742 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04004743 int drop_inode = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004744 int err;
Chris Mason1832a6d2007-12-21 16:27:21 -05004745 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004746 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004747 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004748
Yan, Zhenga22285a2010-05-16 10:48:46 -04004749 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4750 if (err)
4751 return err;
Josef Bacik9ed74f22009-09-11 16:12:44 -04004752 /*
4753 * 2 for inode item and ref
4754 * 2 for dir items
4755 * 1 for xattr if selinux is on
4756 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004757 trans = btrfs_start_transaction(root, 5);
4758 if (IS_ERR(trans))
4759 return PTR_ERR(trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004760
Chris Mason39279cc2007-06-12 06:35:45 -04004761 btrfs_set_trans_block_group(trans, dir);
4762
Josef Bacikaec74772008-07-24 12:12:38 -04004763 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Josef Bacika1b075d2010-11-19 20:36:11 +00004764 dentry->d_name.len, dir->i_ino, objectid,
4765 BTRFS_I(dir)->block_group, mode, &index);
Chris Mason39279cc2007-06-12 06:35:45 -04004766 err = PTR_ERR(inode);
4767 if (IS_ERR(inode))
4768 goto out_unlock;
4769
Yan, Zhengf34f57a2009-11-12 09:35:27 +00004770 err = btrfs_init_inode_security(trans, inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04004771 if (err) {
4772 drop_inode = 1;
4773 goto out_unlock;
4774 }
4775
Chris Mason39279cc2007-06-12 06:35:45 -04004776 btrfs_set_trans_block_group(trans, inode);
Josef Bacika1b075d2010-11-19 20:36:11 +00004777 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004778 if (err)
4779 drop_inode = 1;
4780 else {
4781 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04004782 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04004783 inode->i_fop = &btrfs_file_operations;
4784 inode->i_op = &btrfs_file_inode_operations;
Chris Masond1310b22008-01-24 16:13:08 -05004785 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04004786 }
Chris Mason39279cc2007-06-12 06:35:45 -04004787 btrfs_update_inode_block_group(trans, inode);
4788 btrfs_update_inode_block_group(trans, dir);
4789out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004790 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04004791 btrfs_end_transaction_throttle(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004792 if (drop_inode) {
4793 inode_dec_link_count(inode);
4794 iput(inode);
4795 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004796 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004797 return err;
4798}
4799
4800static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4801 struct dentry *dentry)
4802{
4803 struct btrfs_trans_handle *trans;
4804 struct btrfs_root *root = BTRFS_I(dir)->root;
4805 struct inode *inode = old_dentry->d_inode;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004806 u64 index;
Chris Mason1832a6d2007-12-21 16:27:21 -05004807 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004808 int err;
4809 int drop_inode = 0;
4810
4811 if (inode->i_nlink == 0)
4812 return -ENOENT;
4813
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00004814 /* do not allow sys_link's with other subvols of the same device */
4815 if (root->objectid != BTRFS_I(inode)->root->objectid)
4816 return -EPERM;
4817
Josef Bacik9ed74f22009-09-11 16:12:44 -04004818 btrfs_inc_nlink(inode);
Josef Bacikbc1cbf12010-11-23 19:50:59 +00004819 inode->i_ctime = CURRENT_TIME;
Josef Bacik9ed74f22009-09-11 16:12:44 -04004820
Chris Mason3de45862008-11-17 21:02:50 -05004821 err = btrfs_set_inode_index(dir, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04004822 if (err)
4823 goto fail;
4824
Yan, Zhenga22285a2010-05-16 10:48:46 -04004825 /*
4826 * 1 item for inode ref
4827 * 2 items for dir items
4828 */
4829 trans = btrfs_start_transaction(root, 3);
4830 if (IS_ERR(trans)) {
4831 err = PTR_ERR(trans);
4832 goto fail;
4833 }
Chris Mason5f39d392007-10-15 16:14:19 -04004834
Chris Mason39279cc2007-06-12 06:35:45 -04004835 btrfs_set_trans_block_group(trans, dir);
4836 atomic_inc(&inode->i_count);
Josef Bacikaec74772008-07-24 12:12:38 -04004837
Josef Bacika1b075d2010-11-19 20:36:11 +00004838 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
Chris Mason5f39d392007-10-15 16:14:19 -04004839
Yan, Zhenga5719522009-09-24 09:17:31 -04004840 if (err) {
Chris Mason39279cc2007-06-12 06:35:45 -04004841 drop_inode = 1;
Yan, Zhenga5719522009-09-24 09:17:31 -04004842 } else {
Josef Bacik6a912212010-11-20 09:48:00 +00004843 struct dentry *parent = dget_parent(dentry);
Yan, Zhenga5719522009-09-24 09:17:31 -04004844 btrfs_update_inode_block_group(trans, dir);
4845 err = btrfs_update_inode(trans, root, inode);
4846 BUG_ON(err);
Josef Bacik6a912212010-11-20 09:48:00 +00004847 btrfs_log_new_name(trans, inode, NULL, parent);
4848 dput(parent);
Yan, Zhenga5719522009-09-24 09:17:31 -04004849 }
Chris Mason39279cc2007-06-12 06:35:45 -04004850
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004851 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04004852 btrfs_end_transaction_throttle(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05004853fail:
Chris Mason39279cc2007-06-12 06:35:45 -04004854 if (drop_inode) {
4855 inode_dec_link_count(inode);
4856 iput(inode);
4857 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004858 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004859 return err;
4860}
4861
Chris Mason39279cc2007-06-12 06:35:45 -04004862static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4863{
Chris Masonb9d86662008-05-02 16:13:49 -04004864 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04004865 struct btrfs_trans_handle *trans;
4866 struct btrfs_root *root = BTRFS_I(dir)->root;
4867 int err = 0;
4868 int drop_on_err = 0;
Chris Masonb9d86662008-05-02 16:13:49 -04004869 u64 objectid = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004870 u64 index = 0;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004871 unsigned long nr = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04004872
Yan, Zhenga22285a2010-05-16 10:48:46 -04004873 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4874 if (err)
4875 return err;
4876
Josef Bacik9ed74f22009-09-11 16:12:44 -04004877 /*
4878 * 2 items for inode and ref
4879 * 2 items for dir items
4880 * 1 for xattr if selinux is on
4881 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004882 trans = btrfs_start_transaction(root, 5);
4883 if (IS_ERR(trans))
4884 return PTR_ERR(trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004885 btrfs_set_trans_block_group(trans, dir);
Chris Mason39279cc2007-06-12 06:35:45 -04004886
Josef Bacikaec74772008-07-24 12:12:38 -04004887 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Josef Bacika1b075d2010-11-19 20:36:11 +00004888 dentry->d_name.len, dir->i_ino, objectid,
Chris Mason00e4e6b2008-08-05 11:18:09 -04004889 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4890 &index);
Chris Mason39279cc2007-06-12 06:35:45 -04004891 if (IS_ERR(inode)) {
4892 err = PTR_ERR(inode);
4893 goto out_fail;
4894 }
Chris Mason5f39d392007-10-15 16:14:19 -04004895
Chris Mason39279cc2007-06-12 06:35:45 -04004896 drop_on_err = 1;
Josef Bacik33268ea2008-07-24 12:16:36 -04004897
Yan, Zhengf34f57a2009-11-12 09:35:27 +00004898 err = btrfs_init_inode_security(trans, inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04004899 if (err)
4900 goto out_fail;
4901
Chris Mason39279cc2007-06-12 06:35:45 -04004902 inode->i_op = &btrfs_dir_inode_operations;
4903 inode->i_fop = &btrfs_dir_file_operations;
4904 btrfs_set_trans_block_group(trans, inode);
4905
Chris Masondbe674a2008-07-17 12:54:05 -04004906 btrfs_i_size_write(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004907 err = btrfs_update_inode(trans, root, inode);
4908 if (err)
4909 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04004910
Josef Bacika1b075d2010-11-19 20:36:11 +00004911 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4912 dentry->d_name.len, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004913 if (err)
4914 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04004915
Chris Mason39279cc2007-06-12 06:35:45 -04004916 d_instantiate(dentry, inode);
4917 drop_on_err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004918 btrfs_update_inode_block_group(trans, inode);
4919 btrfs_update_inode_block_group(trans, dir);
4920
4921out_fail:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004922 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04004923 btrfs_end_transaction_throttle(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004924 if (drop_on_err)
4925 iput(inode);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004926 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004927 return err;
4928}
4929
Chris Masond352ac62008-09-29 15:18:18 -04004930/* helper for btfs_get_extent. Given an existing extent in the tree,
4931 * and an extent that you want to insert, deal with overlap and insert
4932 * the new extent into the tree.
4933 */
Chris Mason3b951512008-04-17 11:29:12 -04004934static int merge_extent_mapping(struct extent_map_tree *em_tree,
4935 struct extent_map *existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04004936 struct extent_map *em,
4937 u64 map_start, u64 map_len)
Chris Mason3b951512008-04-17 11:29:12 -04004938{
4939 u64 start_diff;
Chris Mason3b951512008-04-17 11:29:12 -04004940
Chris Masone6dcd2d2008-07-17 12:53:50 -04004941 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4942 start_diff = map_start - em->start;
4943 em->start = map_start;
4944 em->len = map_len;
Chris Masonc8b97812008-10-29 14:49:59 -04004945 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4946 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04004947 em->block_start += start_diff;
Chris Masonc8b97812008-10-29 14:49:59 -04004948 em->block_len -= start_diff;
4949 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004950 return add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04004951}
4952
Chris Masonc8b97812008-10-29 14:49:59 -04004953static noinline int uncompress_inline(struct btrfs_path *path,
4954 struct inode *inode, struct page *page,
4955 size_t pg_offset, u64 extent_offset,
4956 struct btrfs_file_extent_item *item)
4957{
4958 int ret;
4959 struct extent_buffer *leaf = path->nodes[0];
4960 char *tmp;
4961 size_t max_size;
4962 unsigned long inline_size;
4963 unsigned long ptr;
Li Zefan261507a02010-12-17 14:21:50 +08004964 int compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04004965
4966 WARN_ON(pg_offset != 0);
Li Zefan261507a02010-12-17 14:21:50 +08004967 compress_type = btrfs_file_extent_compression(leaf, item);
Chris Masonc8b97812008-10-29 14:49:59 -04004968 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4969 inline_size = btrfs_file_extent_inline_item_len(leaf,
4970 btrfs_item_nr(leaf, path->slots[0]));
4971 tmp = kmalloc(inline_size, GFP_NOFS);
4972 ptr = btrfs_file_extent_inline_start(item);
4973
4974 read_extent_buffer(leaf, tmp, ptr, inline_size);
4975
Chris Mason5b050f02008-11-11 09:34:41 -05004976 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
Li Zefan261507a02010-12-17 14:21:50 +08004977 ret = btrfs_decompress(compress_type, tmp, page,
4978 extent_offset, inline_size, max_size);
Chris Masonc8b97812008-10-29 14:49:59 -04004979 if (ret) {
4980 char *kaddr = kmap_atomic(page, KM_USER0);
4981 unsigned long copy_size = min_t(u64,
4982 PAGE_CACHE_SIZE - pg_offset,
4983 max_size - extent_offset);
4984 memset(kaddr + pg_offset, 0, copy_size);
4985 kunmap_atomic(kaddr, KM_USER0);
4986 }
4987 kfree(tmp);
4988 return 0;
4989}
4990
Chris Masond352ac62008-09-29 15:18:18 -04004991/*
4992 * a bit scary, this does extent mapping from logical file offset to the disk.
Chris Masond3977122009-01-05 21:25:51 -05004993 * the ugly parts come from merging extents from the disk with the in-ram
4994 * representation. This gets more complex because of the data=ordered code,
Chris Masond352ac62008-09-29 15:18:18 -04004995 * where the in-ram extents might be locked pending data=ordered completion.
4996 *
4997 * This also copies inline extents directly into the page.
4998 */
Chris Masond3977122009-01-05 21:25:51 -05004999
Chris Masona52d9a82007-08-27 16:49:44 -04005000struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
Chris Mason70dec802008-01-29 09:59:12 -05005001 size_t pg_offset, u64 start, u64 len,
Chris Masona52d9a82007-08-27 16:49:44 -04005002 int create)
5003{
5004 int ret;
5005 int err = 0;
Chris Masondb945352007-10-15 16:15:53 -04005006 u64 bytenr;
Chris Masona52d9a82007-08-27 16:49:44 -04005007 u64 extent_start = 0;
5008 u64 extent_end = 0;
5009 u64 objectid = inode->i_ino;
5010 u32 found_type;
Chris Masonf4219502008-07-22 11:18:09 -04005011 struct btrfs_path *path = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -04005012 struct btrfs_root *root = BTRFS_I(inode)->root;
5013 struct btrfs_file_extent_item *item;
Chris Mason5f39d392007-10-15 16:14:19 -04005014 struct extent_buffer *leaf;
5015 struct btrfs_key found_key;
Chris Masona52d9a82007-08-27 16:49:44 -04005016 struct extent_map *em = NULL;
5017 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Masond1310b22008-01-24 16:13:08 -05005018 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04005019 struct btrfs_trans_handle *trans = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08005020 int compress_type;
Chris Masona52d9a82007-08-27 16:49:44 -04005021
Chris Masona52d9a82007-08-27 16:49:44 -04005022again:
Chris Mason890871b2009-09-02 16:24:52 -04005023 read_lock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05005024 em = lookup_extent_mapping(em_tree, start, len);
Chris Masona061fc82008-05-07 11:43:44 -04005025 if (em)
5026 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason890871b2009-09-02 16:24:52 -04005027 read_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05005028
Chris Masona52d9a82007-08-27 16:49:44 -04005029 if (em) {
Chris Masone1c4b742008-04-22 13:26:46 -04005030 if (em->start > start || em->start + em->len <= start)
5031 free_extent_map(em);
5032 else if (em->block_start == EXTENT_MAP_INLINE && page)
Chris Mason70dec802008-01-29 09:59:12 -05005033 free_extent_map(em);
5034 else
5035 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04005036 }
Chris Masond1310b22008-01-24 16:13:08 -05005037 em = alloc_extent_map(GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04005038 if (!em) {
Chris Masond1310b22008-01-24 16:13:08 -05005039 err = -ENOMEM;
5040 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04005041 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04005042 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Masond1310b22008-01-24 16:13:08 -05005043 em->start = EXTENT_MAP_HOLE;
Chris Mason445a6942008-11-10 11:53:33 -05005044 em->orig_start = EXTENT_MAP_HOLE;
Chris Masond1310b22008-01-24 16:13:08 -05005045 em->len = (u64)-1;
Chris Masonc8b97812008-10-29 14:49:59 -04005046 em->block_len = (u64)-1;
Chris Masonf4219502008-07-22 11:18:09 -04005047
5048 if (!path) {
5049 path = btrfs_alloc_path();
5050 BUG_ON(!path);
5051 }
5052
Chris Mason179e29e2007-11-01 11:28:41 -04005053 ret = btrfs_lookup_file_extent(trans, root, path,
5054 objectid, start, trans != NULL);
Chris Masona52d9a82007-08-27 16:49:44 -04005055 if (ret < 0) {
5056 err = ret;
5057 goto out;
5058 }
5059
5060 if (ret != 0) {
5061 if (path->slots[0] == 0)
5062 goto not_found;
5063 path->slots[0]--;
5064 }
5065
Chris Mason5f39d392007-10-15 16:14:19 -04005066 leaf = path->nodes[0];
5067 item = btrfs_item_ptr(leaf, path->slots[0],
Chris Masona52d9a82007-08-27 16:49:44 -04005068 struct btrfs_file_extent_item);
Chris Masona52d9a82007-08-27 16:49:44 -04005069 /* are we inside the extent that was found? */
Chris Mason5f39d392007-10-15 16:14:19 -04005070 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5071 found_type = btrfs_key_type(&found_key);
5072 if (found_key.objectid != objectid ||
Chris Masona52d9a82007-08-27 16:49:44 -04005073 found_type != BTRFS_EXTENT_DATA_KEY) {
5074 goto not_found;
5075 }
5076
Chris Mason5f39d392007-10-15 16:14:19 -04005077 found_type = btrfs_file_extent_type(leaf, item);
5078 extent_start = found_key.offset;
Li Zefan261507a02010-12-17 14:21:50 +08005079 compress_type = btrfs_file_extent_compression(leaf, item);
Yan Zhengd899e052008-10-30 14:25:28 -04005080 if (found_type == BTRFS_FILE_EXTENT_REG ||
5081 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04005082 extent_end = extent_start +
Chris Masondb945352007-10-15 16:15:53 -04005083 btrfs_file_extent_num_bytes(leaf, item);
Yan Zheng9036c102008-10-30 14:19:41 -04005084 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5085 size_t size;
5086 size = btrfs_file_extent_inline_len(leaf, item);
5087 extent_end = (extent_start + size + root->sectorsize - 1) &
5088 ~((u64)root->sectorsize - 1);
5089 }
5090
5091 if (start >= extent_end) {
5092 path->slots[0]++;
5093 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5094 ret = btrfs_next_leaf(root, path);
5095 if (ret < 0) {
5096 err = ret;
5097 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04005098 }
Yan Zheng9036c102008-10-30 14:19:41 -04005099 if (ret > 0)
5100 goto not_found;
5101 leaf = path->nodes[0];
Chris Masona52d9a82007-08-27 16:49:44 -04005102 }
Yan Zheng9036c102008-10-30 14:19:41 -04005103 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5104 if (found_key.objectid != objectid ||
5105 found_key.type != BTRFS_EXTENT_DATA_KEY)
5106 goto not_found;
5107 if (start + len <= found_key.offset)
5108 goto not_found;
5109 em->start = start;
5110 em->len = found_key.offset - start;
5111 goto not_found_em;
5112 }
5113
Yan Zhengd899e052008-10-30 14:25:28 -04005114 if (found_type == BTRFS_FILE_EXTENT_REG ||
5115 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng9036c102008-10-30 14:19:41 -04005116 em->start = extent_start;
5117 em->len = extent_end - extent_start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -05005118 em->orig_start = extent_start -
5119 btrfs_file_extent_offset(leaf, item);
Chris Masondb945352007-10-15 16:15:53 -04005120 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5121 if (bytenr == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04005122 em->block_start = EXTENT_MAP_HOLE;
Chris Masona52d9a82007-08-27 16:49:44 -04005123 goto insert;
5124 }
Li Zefan261507a02010-12-17 14:21:50 +08005125 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04005126 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Li Zefan261507a02010-12-17 14:21:50 +08005127 em->compress_type = compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04005128 em->block_start = bytenr;
5129 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5130 item);
5131 } else {
5132 bytenr += btrfs_file_extent_offset(leaf, item);
5133 em->block_start = bytenr;
5134 em->block_len = em->len;
Yan Zhengd899e052008-10-30 14:25:28 -04005135 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5136 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -04005137 }
Chris Masona52d9a82007-08-27 16:49:44 -04005138 goto insert;
5139 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04005140 unsigned long ptr;
Chris Masona52d9a82007-08-27 16:49:44 -04005141 char *map;
Chris Mason3326d1b2007-10-15 16:18:25 -04005142 size_t size;
5143 size_t extent_offset;
5144 size_t copy_size;
Chris Masona52d9a82007-08-27 16:49:44 -04005145
Chris Masona52d9a82007-08-27 16:49:44 -04005146 em->block_start = EXTENT_MAP_INLINE;
Chris Masonc8b97812008-10-29 14:49:59 -04005147 if (!page || create) {
Yan689f9342007-10-29 11:41:07 -04005148 em->start = extent_start;
Yan Zheng9036c102008-10-30 14:19:41 -04005149 em->len = extent_end - extent_start;
Yan689f9342007-10-29 11:41:07 -04005150 goto out;
5151 }
5152
Yan Zheng9036c102008-10-30 14:19:41 -04005153 size = btrfs_file_extent_inline_len(leaf, item);
5154 extent_offset = page_offset(page) + pg_offset - extent_start;
Chris Mason70dec802008-01-29 09:59:12 -05005155 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
Yan689f9342007-10-29 11:41:07 -04005156 size - extent_offset);
Chris Mason3326d1b2007-10-15 16:18:25 -04005157 em->start = extent_start + extent_offset;
Chris Mason70dec802008-01-29 09:59:12 -05005158 em->len = (copy_size + root->sectorsize - 1) &
5159 ~((u64)root->sectorsize - 1);
Yan Zhengff5b7ee2008-11-10 07:34:43 -05005160 em->orig_start = EXTENT_MAP_INLINE;
Li Zefan261507a02010-12-17 14:21:50 +08005161 if (compress_type) {
Chris Masonc8b97812008-10-29 14:49:59 -04005162 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Li Zefan261507a02010-12-17 14:21:50 +08005163 em->compress_type = compress_type;
5164 }
Yan689f9342007-10-29 11:41:07 -04005165 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
Chris Mason179e29e2007-11-01 11:28:41 -04005166 if (create == 0 && !PageUptodate(page)) {
Li Zefan261507a02010-12-17 14:21:50 +08005167 if (btrfs_file_extent_compression(leaf, item) !=
5168 BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04005169 ret = uncompress_inline(path, inode, page,
5170 pg_offset,
5171 extent_offset, item);
5172 BUG_ON(ret);
5173 } else {
5174 map = kmap(page);
5175 read_extent_buffer(leaf, map + pg_offset, ptr,
5176 copy_size);
Chris Mason93c82d52009-09-11 12:36:29 -04005177 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5178 memset(map + pg_offset + copy_size, 0,
5179 PAGE_CACHE_SIZE - pg_offset -
5180 copy_size);
5181 }
Chris Masonc8b97812008-10-29 14:49:59 -04005182 kunmap(page);
5183 }
Chris Mason179e29e2007-11-01 11:28:41 -04005184 flush_dcache_page(page);
5185 } else if (create && PageUptodate(page)) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04005186 WARN_ON(1);
Chris Mason179e29e2007-11-01 11:28:41 -04005187 if (!trans) {
5188 kunmap(page);
5189 free_extent_map(em);
5190 em = NULL;
5191 btrfs_release_path(root, path);
Chris Masonf9295742008-07-17 12:54:14 -04005192 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005193 if (IS_ERR(trans))
5194 return ERR_CAST(trans);
Chris Mason179e29e2007-11-01 11:28:41 -04005195 goto again;
5196 }
Chris Masonc8b97812008-10-29 14:49:59 -04005197 map = kmap(page);
Chris Mason70dec802008-01-29 09:59:12 -05005198 write_extent_buffer(leaf, map + pg_offset, ptr,
Chris Mason179e29e2007-11-01 11:28:41 -04005199 copy_size);
Chris Masonc8b97812008-10-29 14:49:59 -04005200 kunmap(page);
Chris Mason179e29e2007-11-01 11:28:41 -04005201 btrfs_mark_buffer_dirty(leaf);
Chris Masona52d9a82007-08-27 16:49:44 -04005202 }
Chris Masond1310b22008-01-24 16:13:08 -05005203 set_extent_uptodate(io_tree, em->start,
5204 extent_map_end(em) - 1, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04005205 goto insert;
5206 } else {
Chris Masond3977122009-01-05 21:25:51 -05005207 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
Chris Masona52d9a82007-08-27 16:49:44 -04005208 WARN_ON(1);
5209 }
5210not_found:
5211 em->start = start;
Chris Masond1310b22008-01-24 16:13:08 -05005212 em->len = len;
Chris Masona52d9a82007-08-27 16:49:44 -04005213not_found_em:
Chris Mason5f39d392007-10-15 16:14:19 -04005214 em->block_start = EXTENT_MAP_HOLE;
Yan Zheng9036c102008-10-30 14:19:41 -04005215 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -04005216insert:
5217 btrfs_release_path(root, path);
Chris Masond1310b22008-01-24 16:13:08 -05005218 if (em->start > start || extent_map_end(em) <= start) {
Chris Masond3977122009-01-05 21:25:51 -05005219 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5220 "[%llu %llu]\n", (unsigned long long)em->start,
5221 (unsigned long long)em->len,
5222 (unsigned long long)start,
5223 (unsigned long long)len);
Chris Masona52d9a82007-08-27 16:49:44 -04005224 err = -EIO;
5225 goto out;
5226 }
Chris Masond1310b22008-01-24 16:13:08 -05005227
5228 err = 0;
Chris Mason890871b2009-09-02 16:24:52 -04005229 write_lock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04005230 ret = add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04005231 /* it is possible that someone inserted the extent into the tree
5232 * while we had the lock dropped. It is also possible that
5233 * an overlapping map exists in the tree
5234 */
Chris Masona52d9a82007-08-27 16:49:44 -04005235 if (ret == -EEXIST) {
Chris Mason3b951512008-04-17 11:29:12 -04005236 struct extent_map *existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005237
5238 ret = 0;
5239
Chris Mason3b951512008-04-17 11:29:12 -04005240 existing = lookup_extent_mapping(em_tree, start, len);
Chris Masone1c4b742008-04-22 13:26:46 -04005241 if (existing && (existing->start > start ||
5242 existing->start + existing->len <= start)) {
5243 free_extent_map(existing);
5244 existing = NULL;
5245 }
Chris Mason3b951512008-04-17 11:29:12 -04005246 if (!existing) {
5247 existing = lookup_extent_mapping(em_tree, em->start,
5248 em->len);
5249 if (existing) {
5250 err = merge_extent_mapping(em_tree, existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04005251 em, start,
5252 root->sectorsize);
Chris Mason3b951512008-04-17 11:29:12 -04005253 free_extent_map(existing);
5254 if (err) {
5255 free_extent_map(em);
5256 em = NULL;
5257 }
5258 } else {
5259 err = -EIO;
Chris Mason3b951512008-04-17 11:29:12 -04005260 free_extent_map(em);
5261 em = NULL;
5262 }
5263 } else {
5264 free_extent_map(em);
5265 em = existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005266 err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04005267 }
Chris Masona52d9a82007-08-27 16:49:44 -04005268 }
Chris Mason890871b2009-09-02 16:24:52 -04005269 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04005270out:
Chris Masonf4219502008-07-22 11:18:09 -04005271 if (path)
5272 btrfs_free_path(path);
Chris Masona52d9a82007-08-27 16:49:44 -04005273 if (trans) {
5274 ret = btrfs_end_transaction(trans, root);
Chris Masond3977122009-01-05 21:25:51 -05005275 if (!err)
Chris Masona52d9a82007-08-27 16:49:44 -04005276 err = ret;
5277 }
Chris Masona52d9a82007-08-27 16:49:44 -04005278 if (err) {
5279 free_extent_map(em);
Chris Masona52d9a82007-08-27 16:49:44 -04005280 return ERR_PTR(err);
5281 }
5282 return em;
5283}
5284
Chris Masonec29ed52011-02-23 16:23:20 -05005285struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5286 size_t pg_offset, u64 start, u64 len,
5287 int create)
5288{
5289 struct extent_map *em;
5290 struct extent_map *hole_em = NULL;
5291 u64 range_start = start;
5292 u64 end;
5293 u64 found;
5294 u64 found_end;
5295 int err = 0;
5296
5297 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5298 if (IS_ERR(em))
5299 return em;
5300 if (em) {
5301 /*
5302 * if our em maps to a hole, there might
5303 * actually be delalloc bytes behind it
5304 */
5305 if (em->block_start != EXTENT_MAP_HOLE)
5306 return em;
5307 else
5308 hole_em = em;
5309 }
5310
5311 /* check to see if we've wrapped (len == -1 or similar) */
5312 end = start + len;
5313 if (end < start)
5314 end = (u64)-1;
5315 else
5316 end -= 1;
5317
5318 em = NULL;
5319
5320 /* ok, we didn't find anything, lets look for delalloc */
5321 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5322 end, len, EXTENT_DELALLOC, 1);
5323 found_end = range_start + found;
5324 if (found_end < range_start)
5325 found_end = (u64)-1;
5326
5327 /*
5328 * we didn't find anything useful, return
5329 * the original results from get_extent()
5330 */
5331 if (range_start > end || found_end <= start) {
5332 em = hole_em;
5333 hole_em = NULL;
5334 goto out;
5335 }
5336
5337 /* adjust the range_start to make sure it doesn't
5338 * go backwards from the start they passed in
5339 */
5340 range_start = max(start,range_start);
5341 found = found_end - range_start;
5342
5343 if (found > 0) {
5344 u64 hole_start = start;
5345 u64 hole_len = len;
5346
5347 em = alloc_extent_map(GFP_NOFS);
5348 if (!em) {
5349 err = -ENOMEM;
5350 goto out;
5351 }
5352 /*
5353 * when btrfs_get_extent can't find anything it
5354 * returns one huge hole
5355 *
5356 * make sure what it found really fits our range, and
5357 * adjust to make sure it is based on the start from
5358 * the caller
5359 */
5360 if (hole_em) {
5361 u64 calc_end = extent_map_end(hole_em);
5362
5363 if (calc_end <= start || (hole_em->start > end)) {
5364 free_extent_map(hole_em);
5365 hole_em = NULL;
5366 } else {
5367 hole_start = max(hole_em->start, start);
5368 hole_len = calc_end - hole_start;
5369 }
5370 }
5371 em->bdev = NULL;
5372 if (hole_em && range_start > hole_start) {
5373 /* our hole starts before our delalloc, so we
5374 * have to return just the parts of the hole
5375 * that go until the delalloc starts
5376 */
5377 em->len = min(hole_len,
5378 range_start - hole_start);
5379 em->start = hole_start;
5380 em->orig_start = hole_start;
5381 /*
5382 * don't adjust block start at all,
5383 * it is fixed at EXTENT_MAP_HOLE
5384 */
5385 em->block_start = hole_em->block_start;
5386 em->block_len = hole_len;
5387 } else {
5388 em->start = range_start;
5389 em->len = found;
5390 em->orig_start = range_start;
5391 em->block_start = EXTENT_MAP_DELALLOC;
5392 em->block_len = found;
5393 }
5394 } else if (hole_em) {
5395 return hole_em;
5396 }
5397out:
5398
5399 free_extent_map(hole_em);
5400 if (err) {
5401 free_extent_map(em);
5402 return ERR_PTR(err);
5403 }
5404 return em;
5405}
5406
Josef Bacik4b46fce2010-05-23 11:00:55 -04005407static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5408 u64 start, u64 len)
5409{
5410 struct btrfs_root *root = BTRFS_I(inode)->root;
5411 struct btrfs_trans_handle *trans;
5412 struct extent_map *em;
5413 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5414 struct btrfs_key ins;
5415 u64 alloc_hint;
5416 int ret;
5417
5418 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5419
5420 trans = btrfs_join_transaction(root, 0);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005421 if (IS_ERR(trans))
5422 return ERR_CAST(trans);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005423
5424 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5425
5426 alloc_hint = get_extent_allocation_hint(inode, start, len);
5427 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5428 alloc_hint, (u64)-1, &ins, 1);
5429 if (ret) {
5430 em = ERR_PTR(ret);
5431 goto out;
5432 }
5433
5434 em = alloc_extent_map(GFP_NOFS);
5435 if (!em) {
5436 em = ERR_PTR(-ENOMEM);
5437 goto out;
5438 }
5439
5440 em->start = start;
5441 em->orig_start = em->start;
5442 em->len = ins.offset;
5443
5444 em->block_start = ins.objectid;
5445 em->block_len = ins.offset;
5446 em->bdev = root->fs_info->fs_devices->latest_bdev;
5447 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5448
5449 while (1) {
5450 write_lock(&em_tree->lock);
5451 ret = add_extent_mapping(em_tree, em);
5452 write_unlock(&em_tree->lock);
5453 if (ret != -EEXIST)
5454 break;
5455 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5456 }
5457
5458 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5459 ins.offset, ins.offset, 0);
5460 if (ret) {
5461 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5462 em = ERR_PTR(ret);
5463 }
5464out:
5465 btrfs_end_transaction(trans, root);
5466 return em;
5467}
5468
Chris Mason46bfbb52010-05-26 11:04:10 -04005469/*
5470 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5471 * block must be cow'd
5472 */
5473static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5474 struct inode *inode, u64 offset, u64 len)
5475{
5476 struct btrfs_path *path;
5477 int ret;
5478 struct extent_buffer *leaf;
5479 struct btrfs_root *root = BTRFS_I(inode)->root;
5480 struct btrfs_file_extent_item *fi;
5481 struct btrfs_key key;
5482 u64 disk_bytenr;
5483 u64 backref_offset;
5484 u64 extent_end;
5485 u64 num_bytes;
5486 int slot;
5487 int found_type;
5488
5489 path = btrfs_alloc_path();
5490 if (!path)
5491 return -ENOMEM;
5492
5493 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5494 offset, 0);
5495 if (ret < 0)
5496 goto out;
5497
5498 slot = path->slots[0];
5499 if (ret == 1) {
5500 if (slot == 0) {
5501 /* can't find the item, must cow */
5502 ret = 0;
5503 goto out;
5504 }
5505 slot--;
5506 }
5507 ret = 0;
5508 leaf = path->nodes[0];
5509 btrfs_item_key_to_cpu(leaf, &key, slot);
5510 if (key.objectid != inode->i_ino ||
5511 key.type != BTRFS_EXTENT_DATA_KEY) {
5512 /* not our file or wrong item type, must cow */
5513 goto out;
5514 }
5515
5516 if (key.offset > offset) {
5517 /* Wrong offset, must cow */
5518 goto out;
5519 }
5520
5521 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5522 found_type = btrfs_file_extent_type(leaf, fi);
5523 if (found_type != BTRFS_FILE_EXTENT_REG &&
5524 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5525 /* not a regular extent, must cow */
5526 goto out;
5527 }
5528 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5529 backref_offset = btrfs_file_extent_offset(leaf, fi);
5530
5531 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5532 if (extent_end < offset + len) {
5533 /* extent doesn't include our full range, must cow */
5534 goto out;
5535 }
5536
5537 if (btrfs_extent_readonly(root, disk_bytenr))
5538 goto out;
5539
5540 /*
5541 * look for other files referencing this extent, if we
5542 * find any we must cow
5543 */
5544 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5545 key.offset - backref_offset, disk_bytenr))
5546 goto out;
5547
5548 /*
5549 * adjust disk_bytenr and num_bytes to cover just the bytes
5550 * in this extent we are about to write. If there
5551 * are any csums in that range we have to cow in order
5552 * to keep the csums correct
5553 */
5554 disk_bytenr += backref_offset;
5555 disk_bytenr += offset - key.offset;
5556 num_bytes = min(offset + len, extent_end) - offset;
5557 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5558 goto out;
5559 /*
5560 * all of the above have passed, it is safe to overwrite this extent
5561 * without cow
5562 */
5563 ret = 1;
5564out:
5565 btrfs_free_path(path);
5566 return ret;
5567}
5568
Josef Bacik4b46fce2010-05-23 11:00:55 -04005569static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5570 struct buffer_head *bh_result, int create)
5571{
5572 struct extent_map *em;
5573 struct btrfs_root *root = BTRFS_I(inode)->root;
5574 u64 start = iblock << inode->i_blkbits;
5575 u64 len = bh_result->b_size;
Chris Mason46bfbb52010-05-26 11:04:10 -04005576 struct btrfs_trans_handle *trans;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005577
5578 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5579 if (IS_ERR(em))
5580 return PTR_ERR(em);
5581
5582 /*
5583 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5584 * io. INLINE is special, and we could probably kludge it in here, but
5585 * it's still buffered so for safety lets just fall back to the generic
5586 * buffered path.
5587 *
5588 * For COMPRESSED we _have_ to read the entire extent in so we can
5589 * decompress it, so there will be buffering required no matter what we
5590 * do, so go ahead and fallback to buffered.
5591 *
5592 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5593 * to buffered IO. Don't blame me, this is the price we pay for using
5594 * the generic code.
5595 */
5596 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5597 em->block_start == EXTENT_MAP_INLINE) {
5598 free_extent_map(em);
5599 return -ENOTBLK;
5600 }
5601
5602 /* Just a good old fashioned hole, return */
5603 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5604 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5605 free_extent_map(em);
5606 /* DIO will do one hole at a time, so just unlock a sector */
5607 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5608 start + root->sectorsize - 1, GFP_NOFS);
5609 return 0;
5610 }
5611
5612 /*
5613 * We don't allocate a new extent in the following cases
5614 *
5615 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5616 * existing extent.
5617 * 2) The extent is marked as PREALLOC. We're good to go here and can
5618 * just use the extent.
5619 *
5620 */
Chris Mason46bfbb52010-05-26 11:04:10 -04005621 if (!create) {
5622 len = em->len - (start - em->start);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005623 goto map;
Chris Mason46bfbb52010-05-26 11:04:10 -04005624 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04005625
5626 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5627 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5628 em->block_start != EXTENT_MAP_HOLE)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04005629 int type;
5630 int ret;
Chris Mason46bfbb52010-05-26 11:04:10 -04005631 u64 block_start;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005632
5633 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5634 type = BTRFS_ORDERED_PREALLOC;
5635 else
5636 type = BTRFS_ORDERED_NOCOW;
Chris Mason46bfbb52010-05-26 11:04:10 -04005637 len = min(len, em->len - (start - em->start));
Josef Bacik4b46fce2010-05-23 11:00:55 -04005638 block_start = em->block_start + (start - em->start);
Chris Mason46bfbb52010-05-26 11:04:10 -04005639
5640 /*
5641 * we're not going to log anything, but we do need
5642 * to make sure the current transaction stays open
5643 * while we look for nocow cross refs
5644 */
5645 trans = btrfs_join_transaction(root, 0);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005646 if (IS_ERR(trans))
Chris Mason46bfbb52010-05-26 11:04:10 -04005647 goto must_cow;
5648
5649 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5650 ret = btrfs_add_ordered_extent_dio(inode, start,
5651 block_start, len, len, type);
5652 btrfs_end_transaction(trans, root);
5653 if (ret) {
5654 free_extent_map(em);
5655 return ret;
5656 }
5657 goto unlock;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005658 }
Chris Mason46bfbb52010-05-26 11:04:10 -04005659 btrfs_end_transaction(trans, root);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005660 }
Chris Mason46bfbb52010-05-26 11:04:10 -04005661must_cow:
5662 /*
5663 * this will cow the extent, reset the len in case we changed
5664 * it above
5665 */
5666 len = bh_result->b_size;
5667 free_extent_map(em);
5668 em = btrfs_new_extent_direct(inode, start, len);
5669 if (IS_ERR(em))
5670 return PTR_ERR(em);
5671 len = min(len, em->len - (start - em->start));
5672unlock:
Chris Mason4845e442010-05-25 20:56:50 -04005673 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5674 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5675 0, NULL, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005676map:
5677 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5678 inode->i_blkbits;
Chris Mason46bfbb52010-05-26 11:04:10 -04005679 bh_result->b_size = len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005680 bh_result->b_bdev = em->bdev;
5681 set_buffer_mapped(bh_result);
5682 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5683 set_buffer_new(bh_result);
5684
5685 free_extent_map(em);
5686
5687 return 0;
5688}
5689
5690struct btrfs_dio_private {
5691 struct inode *inode;
5692 u64 logical_offset;
5693 u64 disk_bytenr;
5694 u64 bytes;
5695 u32 *csums;
5696 void *private;
Miao Xiee65e1532010-11-22 03:04:43 +00005697
5698 /* number of bios pending for this dio */
5699 atomic_t pending_bios;
5700
5701 /* IO errors */
5702 int errors;
5703
5704 struct bio *orig_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005705};
5706
5707static void btrfs_endio_direct_read(struct bio *bio, int err)
5708{
Miao Xiee65e1532010-11-22 03:04:43 +00005709 struct btrfs_dio_private *dip = bio->bi_private;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005710 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5711 struct bio_vec *bvec = bio->bi_io_vec;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005712 struct inode *inode = dip->inode;
5713 struct btrfs_root *root = BTRFS_I(inode)->root;
5714 u64 start;
5715 u32 *private = dip->csums;
5716
5717 start = dip->logical_offset;
5718 do {
5719 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5720 struct page *page = bvec->bv_page;
5721 char *kaddr;
5722 u32 csum = ~(u32)0;
5723 unsigned long flags;
5724
5725 local_irq_save(flags);
5726 kaddr = kmap_atomic(page, KM_IRQ0);
5727 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5728 csum, bvec->bv_len);
5729 btrfs_csum_final(csum, (char *)&csum);
5730 kunmap_atomic(kaddr, KM_IRQ0);
5731 local_irq_restore(flags);
5732
5733 flush_dcache_page(bvec->bv_page);
5734 if (csum != *private) {
5735 printk(KERN_ERR "btrfs csum failed ino %lu off"
5736 " %llu csum %u private %u\n",
5737 inode->i_ino, (unsigned long long)start,
5738 csum, *private);
5739 err = -EIO;
5740 }
5741 }
5742
5743 start += bvec->bv_len;
5744 private++;
5745 bvec++;
5746 } while (bvec <= bvec_end);
5747
5748 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5749 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5750 bio->bi_private = dip->private;
5751
5752 kfree(dip->csums);
5753 kfree(dip);
5754 dio_end_io(bio, err);
5755}
5756
5757static void btrfs_endio_direct_write(struct bio *bio, int err)
5758{
5759 struct btrfs_dio_private *dip = bio->bi_private;
5760 struct inode *inode = dip->inode;
5761 struct btrfs_root *root = BTRFS_I(inode)->root;
5762 struct btrfs_trans_handle *trans;
5763 struct btrfs_ordered_extent *ordered = NULL;
5764 struct extent_state *cached_state = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05005765 u64 ordered_offset = dip->logical_offset;
5766 u64 ordered_bytes = dip->bytes;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005767 int ret;
5768
5769 if (err)
5770 goto out_done;
Chris Mason163cf092010-11-28 19:56:33 -05005771again:
5772 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5773 &ordered_offset,
5774 ordered_bytes);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005775 if (!ret)
Chris Mason163cf092010-11-28 19:56:33 -05005776 goto out_test;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005777
5778 BUG_ON(!ordered);
5779
5780 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005781 if (IS_ERR(trans)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04005782 err = -ENOMEM;
5783 goto out;
5784 }
5785 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5786
5787 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5788 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5789 if (!ret)
5790 ret = btrfs_update_inode(trans, root, inode);
5791 err = ret;
5792 goto out;
5793 }
5794
5795 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5796 ordered->file_offset + ordered->len - 1, 0,
5797 &cached_state, GFP_NOFS);
5798
5799 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5800 ret = btrfs_mark_extent_written(trans, inode,
5801 ordered->file_offset,
5802 ordered->file_offset +
5803 ordered->len);
5804 if (ret) {
5805 err = ret;
5806 goto out_unlock;
5807 }
5808 } else {
5809 ret = insert_reserved_file_extent(trans, inode,
5810 ordered->file_offset,
5811 ordered->start,
5812 ordered->disk_len,
5813 ordered->len,
5814 ordered->len,
5815 0, 0, 0,
5816 BTRFS_FILE_EXTENT_REG);
5817 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5818 ordered->file_offset, ordered->len);
5819 if (ret) {
5820 err = ret;
5821 WARN_ON(1);
5822 goto out_unlock;
5823 }
5824 }
5825
5826 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5827 btrfs_ordered_update_i_size(inode, 0, ordered);
5828 btrfs_update_inode(trans, root, inode);
5829out_unlock:
5830 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5831 ordered->file_offset + ordered->len - 1,
5832 &cached_state, GFP_NOFS);
5833out:
5834 btrfs_delalloc_release_metadata(inode, ordered->len);
5835 btrfs_end_transaction(trans, root);
Chris Mason163cf092010-11-28 19:56:33 -05005836 ordered_offset = ordered->file_offset + ordered->len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005837 btrfs_put_ordered_extent(ordered);
5838 btrfs_put_ordered_extent(ordered);
Chris Mason163cf092010-11-28 19:56:33 -05005839
5840out_test:
5841 /*
5842 * our bio might span multiple ordered extents. If we haven't
5843 * completed the accounting for the whole dio, go back and try again
5844 */
5845 if (ordered_offset < dip->logical_offset + dip->bytes) {
5846 ordered_bytes = dip->logical_offset + dip->bytes -
5847 ordered_offset;
5848 goto again;
5849 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04005850out_done:
5851 bio->bi_private = dip->private;
5852
5853 kfree(dip->csums);
5854 kfree(dip);
5855 dio_end_io(bio, err);
5856}
5857
Chris Masoneaf25d92010-05-25 09:48:28 -04005858static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5859 struct bio *bio, int mirror_num,
5860 unsigned long bio_flags, u64 offset)
5861{
5862 int ret;
5863 struct btrfs_root *root = BTRFS_I(inode)->root;
5864 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5865 BUG_ON(ret);
5866 return 0;
5867}
5868
Miao Xiee65e1532010-11-22 03:04:43 +00005869static void btrfs_end_dio_bio(struct bio *bio, int err)
5870{
5871 struct btrfs_dio_private *dip = bio->bi_private;
5872
5873 if (err) {
5874 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
Jan Beulich3dd14622010-12-07 14:54:09 +00005875 "sector %#Lx len %u err no %d\n",
5876 dip->inode->i_ino, bio->bi_rw,
5877 (unsigned long long)bio->bi_sector, bio->bi_size, err);
Miao Xiee65e1532010-11-22 03:04:43 +00005878 dip->errors = 1;
5879
5880 /*
5881 * before atomic variable goto zero, we must make sure
5882 * dip->errors is perceived to be set.
5883 */
5884 smp_mb__before_atomic_dec();
5885 }
5886
5887 /* if there are more bios still pending for this dio, just exit */
5888 if (!atomic_dec_and_test(&dip->pending_bios))
5889 goto out;
5890
5891 if (dip->errors)
5892 bio_io_error(dip->orig_bio);
5893 else {
5894 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5895 bio_endio(dip->orig_bio, 0);
5896 }
5897out:
5898 bio_put(bio);
5899}
5900
5901static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5902 u64 first_sector, gfp_t gfp_flags)
5903{
5904 int nr_vecs = bio_get_nr_vecs(bdev);
5905 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5906}
5907
5908static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5909 int rw, u64 file_offset, int skip_sum,
5910 u32 *csums)
5911{
5912 int write = rw & REQ_WRITE;
5913 struct btrfs_root *root = BTRFS_I(inode)->root;
5914 int ret;
5915
5916 bio_get(bio);
5917 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5918 if (ret)
5919 goto err;
5920
5921 if (write && !skip_sum) {
5922 ret = btrfs_wq_submit_bio(root->fs_info,
5923 inode, rw, bio, 0, 0,
5924 file_offset,
5925 __btrfs_submit_bio_start_direct_io,
5926 __btrfs_submit_bio_done);
5927 goto err;
5928 } else if (!skip_sum)
5929 btrfs_lookup_bio_sums_dio(root, inode, bio,
5930 file_offset, csums);
5931
5932 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5933err:
5934 bio_put(bio);
5935 return ret;
5936}
5937
5938static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5939 int skip_sum)
5940{
5941 struct inode *inode = dip->inode;
5942 struct btrfs_root *root = BTRFS_I(inode)->root;
5943 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5944 struct bio *bio;
5945 struct bio *orig_bio = dip->orig_bio;
5946 struct bio_vec *bvec = orig_bio->bi_io_vec;
5947 u64 start_sector = orig_bio->bi_sector;
5948 u64 file_offset = dip->logical_offset;
5949 u64 submit_len = 0;
5950 u64 map_length;
5951 int nr_pages = 0;
5952 u32 *csums = dip->csums;
5953 int ret = 0;
5954
5955 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5956 if (!bio)
5957 return -ENOMEM;
5958 bio->bi_private = dip;
5959 bio->bi_end_io = btrfs_end_dio_bio;
5960 atomic_inc(&dip->pending_bios);
5961
5962 map_length = orig_bio->bi_size;
5963 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5964 &map_length, NULL, 0);
5965 if (ret) {
5966 bio_put(bio);
5967 return -EIO;
5968 }
5969
5970 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5971 if (unlikely(map_length < submit_len + bvec->bv_len ||
5972 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5973 bvec->bv_offset) < bvec->bv_len)) {
5974 /*
5975 * inc the count before we submit the bio so
5976 * we know the end IO handler won't happen before
5977 * we inc the count. Otherwise, the dip might get freed
5978 * before we're done setting it up
5979 */
5980 atomic_inc(&dip->pending_bios);
5981 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5982 file_offset, skip_sum,
5983 csums);
5984 if (ret) {
5985 bio_put(bio);
5986 atomic_dec(&dip->pending_bios);
5987 goto out_err;
5988 }
5989
5990 if (!skip_sum)
5991 csums = csums + nr_pages;
5992 start_sector += submit_len >> 9;
5993 file_offset += submit_len;
5994
5995 submit_len = 0;
5996 nr_pages = 0;
5997
5998 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5999 start_sector, GFP_NOFS);
6000 if (!bio)
6001 goto out_err;
6002 bio->bi_private = dip;
6003 bio->bi_end_io = btrfs_end_dio_bio;
6004
6005 map_length = orig_bio->bi_size;
6006 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6007 &map_length, NULL, 0);
6008 if (ret) {
6009 bio_put(bio);
6010 goto out_err;
6011 }
6012 } else {
6013 submit_len += bvec->bv_len;
6014 nr_pages ++;
6015 bvec++;
6016 }
6017 }
6018
6019 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6020 csums);
6021 if (!ret)
6022 return 0;
6023
6024 bio_put(bio);
6025out_err:
6026 dip->errors = 1;
6027 /*
6028 * before atomic variable goto zero, we must
6029 * make sure dip->errors is perceived to be set.
6030 */
6031 smp_mb__before_atomic_dec();
6032 if (atomic_dec_and_test(&dip->pending_bios))
6033 bio_io_error(dip->orig_bio);
6034
6035 /* bio_end_io() will handle error, so we needn't return it */
6036 return 0;
6037}
6038
Josef Bacik4b46fce2010-05-23 11:00:55 -04006039static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6040 loff_t file_offset)
6041{
6042 struct btrfs_root *root = BTRFS_I(inode)->root;
6043 struct btrfs_dio_private *dip;
6044 struct bio_vec *bvec = bio->bi_io_vec;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006045 int skip_sum;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02006046 int write = rw & REQ_WRITE;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006047 int ret = 0;
6048
6049 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6050
6051 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6052 if (!dip) {
6053 ret = -ENOMEM;
6054 goto free_ordered;
6055 }
6056 dip->csums = NULL;
6057
6058 if (!skip_sum) {
6059 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6060 if (!dip->csums) {
Daniel J Bluemanb4966b72011-03-09 16:46:42 +00006061 kfree(dip);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006062 ret = -ENOMEM;
6063 goto free_ordered;
6064 }
6065 }
6066
6067 dip->private = bio->bi_private;
6068 dip->inode = inode;
6069 dip->logical_offset = file_offset;
6070
Josef Bacik4b46fce2010-05-23 11:00:55 -04006071 dip->bytes = 0;
6072 do {
6073 dip->bytes += bvec->bv_len;
6074 bvec++;
6075 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6076
Chris Mason46bfbb52010-05-26 11:04:10 -04006077 dip->disk_bytenr = (u64)bio->bi_sector << 9;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006078 bio->bi_private = dip;
Miao Xiee65e1532010-11-22 03:04:43 +00006079 dip->errors = 0;
6080 dip->orig_bio = bio;
6081 atomic_set(&dip->pending_bios, 0);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006082
6083 if (write)
6084 bio->bi_end_io = btrfs_endio_direct_write;
6085 else
6086 bio->bi_end_io = btrfs_endio_direct_read;
6087
Miao Xiee65e1532010-11-22 03:04:43 +00006088 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6089 if (!ret)
Chris Masoneaf25d92010-05-25 09:48:28 -04006090 return;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006091free_ordered:
6092 /*
6093 * If this is a write, we need to clean up the reserved space and kill
6094 * the ordered extent.
6095 */
6096 if (write) {
6097 struct btrfs_ordered_extent *ordered;
Josef Bacik955256f2010-11-19 09:41:10 -05006098 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006099 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6100 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6101 btrfs_free_reserved_extent(root, ordered->start,
6102 ordered->disk_len);
6103 btrfs_put_ordered_extent(ordered);
6104 btrfs_put_ordered_extent(ordered);
6105 }
6106 bio_endio(bio, ret);
6107}
6108
Chris Mason5a5f79b2010-05-26 21:33:37 -04006109static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6110 const struct iovec *iov, loff_t offset,
6111 unsigned long nr_segs)
6112{
6113 int seg;
6114 size_t size;
6115 unsigned long addr;
6116 unsigned blocksize_mask = root->sectorsize - 1;
6117 ssize_t retval = -EINVAL;
6118 loff_t end = offset;
6119
6120 if (offset & blocksize_mask)
6121 goto out;
6122
6123 /* Check the memory alignment. Blocks cannot straddle pages */
6124 for (seg = 0; seg < nr_segs; seg++) {
6125 addr = (unsigned long)iov[seg].iov_base;
6126 size = iov[seg].iov_len;
6127 end += size;
6128 if ((addr & blocksize_mask) || (size & blocksize_mask))
6129 goto out;
6130 }
6131 retval = 0;
6132out:
6133 return retval;
6134}
Chris Mason16432982008-04-10 10:23:21 -04006135static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6136 const struct iovec *iov, loff_t offset,
6137 unsigned long nr_segs)
6138{
Josef Bacik4b46fce2010-05-23 11:00:55 -04006139 struct file *file = iocb->ki_filp;
6140 struct inode *inode = file->f_mapping->host;
6141 struct btrfs_ordered_extent *ordered;
Chris Mason4845e442010-05-25 20:56:50 -04006142 struct extent_state *cached_state = NULL;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006143 u64 lockstart, lockend;
6144 ssize_t ret;
Chris Mason4845e442010-05-25 20:56:50 -04006145 int writing = rw & WRITE;
6146 int write_bits = 0;
Chris Mason3f7c5792010-05-26 10:59:53 -04006147 size_t count = iov_length(iov, nr_segs);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006148
Chris Mason5a5f79b2010-05-26 21:33:37 -04006149 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6150 offset, nr_segs)) {
6151 return 0;
6152 }
6153
Josef Bacik4b46fce2010-05-23 11:00:55 -04006154 lockstart = offset;
Chris Mason3f7c5792010-05-26 10:59:53 -04006155 lockend = offset + count - 1;
6156
6157 if (writing) {
6158 ret = btrfs_delalloc_reserve_space(inode, count);
6159 if (ret)
6160 goto out;
6161 }
Chris Mason4845e442010-05-25 20:56:50 -04006162
Josef Bacik4b46fce2010-05-23 11:00:55 -04006163 while (1) {
Chris Mason4845e442010-05-25 20:56:50 -04006164 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6165 0, &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006166 /*
6167 * We're concerned with the entire range that we're going to be
6168 * doing DIO to, so we need to make sure theres no ordered
6169 * extents in this range.
6170 */
6171 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6172 lockend - lockstart + 1);
6173 if (!ordered)
6174 break;
Chris Mason4845e442010-05-25 20:56:50 -04006175 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6176 &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006177 btrfs_start_ordered_extent(inode, ordered, 1);
6178 btrfs_put_ordered_extent(ordered);
6179 cond_resched();
6180 }
6181
Chris Mason4845e442010-05-25 20:56:50 -04006182 /*
6183 * we don't use btrfs_set_extent_delalloc because we don't want
6184 * the dirty or uptodate bits
6185 */
6186 if (writing) {
6187 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6188 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6189 EXTENT_DELALLOC, 0, NULL, &cached_state,
6190 GFP_NOFS);
6191 if (ret) {
6192 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6193 lockend, EXTENT_LOCKED | write_bits,
6194 1, 0, &cached_state, GFP_NOFS);
6195 goto out;
6196 }
6197 }
6198
6199 free_extent_state(cached_state);
6200 cached_state = NULL;
6201
Chris Mason5a5f79b2010-05-26 21:33:37 -04006202 ret = __blockdev_direct_IO(rw, iocb, inode,
6203 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6204 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6205 btrfs_submit_direct, 0);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006206
6207 if (ret < 0 && ret != -EIOCBQUEUED) {
Chris Mason4845e442010-05-25 20:56:50 -04006208 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6209 offset + iov_length(iov, nr_segs) - 1,
6210 EXTENT_LOCKED | write_bits, 1, 0,
6211 &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006212 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6213 /*
6214 * We're falling back to buffered, unlock the section we didn't
6215 * do IO on.
6216 */
Chris Mason4845e442010-05-25 20:56:50 -04006217 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6218 offset + iov_length(iov, nr_segs) - 1,
6219 EXTENT_LOCKED | write_bits, 1, 0,
6220 &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006221 }
Chris Mason4845e442010-05-25 20:56:50 -04006222out:
6223 free_extent_state(cached_state);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006224 return ret;
Chris Mason16432982008-04-10 10:23:21 -04006225}
6226
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05006227static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6228 __u64 start, __u64 len)
6229{
Chris Masonec29ed52011-02-23 16:23:20 -05006230 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05006231}
6232
Chris Mason9ebefb182007-06-15 13:50:00 -04006233int btrfs_readpage(struct file *file, struct page *page)
6234{
Chris Masond1310b22008-01-24 16:13:08 -05006235 struct extent_io_tree *tree;
6236 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04006237 return extent_read_full_page(tree, page, btrfs_get_extent);
Chris Mason39279cc2007-06-12 06:35:45 -04006238}
Chris Mason1832a6d2007-12-21 16:27:21 -05006239
Chris Mason39279cc2007-06-12 06:35:45 -04006240static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6241{
Chris Masond1310b22008-01-24 16:13:08 -05006242 struct extent_io_tree *tree;
Chris Masonb888db22007-08-27 16:49:44 -04006243
6244
6245 if (current->flags & PF_MEMALLOC) {
6246 redirty_page_for_writepage(wbc, page);
6247 unlock_page(page);
6248 return 0;
6249 }
Chris Masond1310b22008-01-24 16:13:08 -05006250 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04006251 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6252}
Chris Mason39279cc2007-06-12 06:35:45 -04006253
Chris Masonf4219502008-07-22 11:18:09 -04006254int btrfs_writepages(struct address_space *mapping,
6255 struct writeback_control *wbc)
Chris Masonb293f022007-11-01 19:45:34 -04006256{
Chris Masond1310b22008-01-24 16:13:08 -05006257 struct extent_io_tree *tree;
Chris Mason771ed682008-11-06 22:02:51 -05006258
Chris Masond1310b22008-01-24 16:13:08 -05006259 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Masonb293f022007-11-01 19:45:34 -04006260 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6261}
6262
Chris Mason3ab2fb52007-11-08 10:59:22 -05006263static int
6264btrfs_readpages(struct file *file, struct address_space *mapping,
6265 struct list_head *pages, unsigned nr_pages)
6266{
Chris Masond1310b22008-01-24 16:13:08 -05006267 struct extent_io_tree *tree;
6268 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Mason3ab2fb52007-11-08 10:59:22 -05006269 return extent_readpages(tree, mapping, pages, nr_pages,
6270 btrfs_get_extent);
6271}
Chris Masone6dcd2d2008-07-17 12:53:50 -04006272static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
Chris Masona52d9a82007-08-27 16:49:44 -04006273{
Chris Masond1310b22008-01-24 16:13:08 -05006274 struct extent_io_tree *tree;
6275 struct extent_map_tree *map;
Chris Masona52d9a82007-08-27 16:49:44 -04006276 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04006277
Chris Masond1310b22008-01-24 16:13:08 -05006278 tree = &BTRFS_I(page->mapping->host)->io_tree;
6279 map = &BTRFS_I(page->mapping->host)->extent_tree;
Chris Mason70dec802008-01-29 09:59:12 -05006280 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
Chris Masona52d9a82007-08-27 16:49:44 -04006281 if (ret == 1) {
6282 ClearPagePrivate(page);
6283 set_page_private(page, 0);
6284 page_cache_release(page);
6285 }
6286 return ret;
6287}
Chris Mason39279cc2007-06-12 06:35:45 -04006288
Chris Masone6dcd2d2008-07-17 12:53:50 -04006289static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6290{
Chris Mason98509cf2008-09-11 15:51:43 -04006291 if (PageWriteback(page) || PageDirty(page))
6292 return 0;
Yan Zhengb335b002009-02-12 10:06:04 -05006293 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006294}
6295
Chris Masona52d9a82007-08-27 16:49:44 -04006296static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6297{
Chris Masond1310b22008-01-24 16:13:08 -05006298 struct extent_io_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006299 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00006300 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006301 u64 page_start = page_offset(page);
6302 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04006303
Chris Mason8b62b722009-09-02 16:53:46 -04006304
6305 /*
6306 * we have the page locked, so new writeback can't start,
6307 * and the dirty bit won't be cleared while we are here.
6308 *
6309 * Wait for IO on this page so that we can safely clear
6310 * the PagePrivate2 bit and do ordered accounting
6311 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006312 wait_on_page_writeback(page);
Chris Mason8b62b722009-09-02 16:53:46 -04006313
Chris Masond1310b22008-01-24 16:13:08 -05006314 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006315 if (offset) {
6316 btrfs_releasepage(page, GFP_NOFS);
6317 return;
6318 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00006319 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6320 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006321 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6322 page_offset(page));
6323 if (ordered) {
Chris Masoneb84ae02008-07-17 13:53:27 -04006324 /*
6325 * IO on this page will never be started, so we need
6326 * to account for any ordered extents now
6327 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006328 clear_extent_bit(tree, page_start, page_end,
6329 EXTENT_DIRTY | EXTENT_DELALLOC |
Josef Bacik32c00af2009-10-08 13:34:05 -04006330 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
Josef Bacik2ac55d42010-02-03 19:33:23 +00006331 &cached_state, GFP_NOFS);
Chris Mason8b62b722009-09-02 16:53:46 -04006332 /*
6333 * whoever cleared the private bit is responsible
6334 * for the finish_ordered_io
6335 */
6336 if (TestClearPagePrivate2(page)) {
6337 btrfs_finish_ordered_io(page->mapping->host,
6338 page_start, page_end);
6339 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006340 btrfs_put_ordered_extent(ordered);
Josef Bacik2ac55d42010-02-03 19:33:23 +00006341 cached_state = NULL;
6342 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6343 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006344 }
6345 clear_extent_bit(tree, page_start, page_end,
Josef Bacik32c00af2009-10-08 13:34:05 -04006346 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
Josef Bacik2ac55d42010-02-03 19:33:23 +00006347 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006348 __btrfs_releasepage(page, GFP_NOFS);
6349
Chris Mason4a096752008-07-21 10:29:44 -04006350 ClearPageChecked(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04006351 if (PagePrivate(page)) {
Chris Mason9ad6b7b2008-04-18 16:11:30 -04006352 ClearPagePrivate(page);
6353 set_page_private(page, 0);
6354 page_cache_release(page);
6355 }
Chris Mason39279cc2007-06-12 06:35:45 -04006356}
6357
Chris Mason9ebefb182007-06-15 13:50:00 -04006358/*
6359 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6360 * called from a page fault handler when a page is first dirtied. Hence we must
6361 * be careful to check for EOF conditions here. We set the page up correctly
6362 * for a written page which means we get ENOSPC checking when writing into
6363 * holes and correct delalloc and unwritten extent mapping on filesystems that
6364 * support these features.
6365 *
6366 * We are not allowed to take the i_mutex here so we have to play games to
6367 * protect against truncate races as the page could now be beyond EOF. Because
6368 * vmtruncate() writes the inode size before removing pages, once we have the
6369 * page lock we can determine safely if the page is beyond EOF. If it is not
6370 * beyond EOF, then the page is guaranteed safe against truncation until we
6371 * unlock the page.
6372 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07006373int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Chris Mason9ebefb182007-06-15 13:50:00 -04006374{
Nick Pigginc2ec1752009-03-31 15:23:21 -07006375 struct page *page = vmf->page;
Chris Mason6da6aba2007-12-18 16:15:09 -05006376 struct inode *inode = fdentry(vma->vm_file)->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05006377 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006378 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6379 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00006380 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006381 char *kaddr;
6382 unsigned long zero_start;
Chris Mason9ebefb182007-06-15 13:50:00 -04006383 loff_t size;
Chris Mason1832a6d2007-12-21 16:27:21 -05006384 int ret;
Chris Masona52d9a82007-08-27 16:49:44 -04006385 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006386 u64 page_end;
Chris Mason9ebefb182007-06-15 13:50:00 -04006387
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04006388 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Nick Piggin56a76f82009-03-31 15:23:23 -07006389 if (ret) {
6390 if (ret == -ENOMEM)
6391 ret = VM_FAULT_OOM;
6392 else /* -ENOSPC, -EIO, etc */
6393 ret = VM_FAULT_SIGBUS;
Chris Mason1832a6d2007-12-21 16:27:21 -05006394 goto out;
Nick Piggin56a76f82009-03-31 15:23:23 -07006395 }
Chris Mason1832a6d2007-12-21 16:27:21 -05006396
Nick Piggin56a76f82009-03-31 15:23:23 -07006397 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006398again:
Chris Mason9ebefb182007-06-15 13:50:00 -04006399 lock_page(page);
Chris Mason9ebefb182007-06-15 13:50:00 -04006400 size = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006401 page_start = page_offset(page);
6402 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04006403
Chris Mason9ebefb182007-06-15 13:50:00 -04006404 if ((page->mapping != inode->i_mapping) ||
Chris Masone6dcd2d2008-07-17 12:53:50 -04006405 (page_start >= size)) {
Chris Mason9ebefb182007-06-15 13:50:00 -04006406 /* page got truncated out from underneath us */
6407 goto out_unlock;
6408 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006409 wait_on_page_writeback(page);
6410
Josef Bacik2ac55d42010-02-03 19:33:23 +00006411 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6412 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006413 set_page_extent_mapped(page);
6414
Chris Masoneb84ae02008-07-17 13:53:27 -04006415 /*
6416 * we can't set the delalloc bits if there are pending ordered
6417 * extents. Drop our locks and wait for them to finish
6418 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006419 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6420 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00006421 unlock_extent_cached(io_tree, page_start, page_end,
6422 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006423 unlock_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04006424 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006425 btrfs_put_ordered_extent(ordered);
6426 goto again;
6427 }
6428
Josef Bacikfbf19082009-10-01 17:10:23 -04006429 /*
6430 * XXX - page_mkwrite gets called every time the page is dirtied, even
6431 * if it was already dirty, so for space accounting reasons we need to
6432 * clear any delalloc bits for the range we are fixing to save. There
6433 * is probably a better way to do this, but for now keep consistent with
6434 * prepare_pages in the normal write path.
6435 */
Josef Bacik2ac55d42010-02-03 19:33:23 +00006436 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Josef Bacik32c00af2009-10-08 13:34:05 -04006437 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
Josef Bacik2ac55d42010-02-03 19:33:23 +00006438 0, 0, &cached_state, GFP_NOFS);
Josef Bacikfbf19082009-10-01 17:10:23 -04006439
Josef Bacik2ac55d42010-02-03 19:33:23 +00006440 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6441 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006442 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00006443 unlock_extent_cached(io_tree, page_start, page_end,
6444 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006445 ret = VM_FAULT_SIGBUS;
6446 goto out_unlock;
6447 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006448 ret = 0;
Chris Mason9ebefb182007-06-15 13:50:00 -04006449
6450 /* page is wholly or partially inside EOF */
Chris Masona52d9a82007-08-27 16:49:44 -04006451 if (page_start + PAGE_CACHE_SIZE > size)
Chris Masone6dcd2d2008-07-17 12:53:50 -04006452 zero_start = size & ~PAGE_CACHE_MASK;
Chris Mason9ebefb182007-06-15 13:50:00 -04006453 else
Chris Masone6dcd2d2008-07-17 12:53:50 -04006454 zero_start = PAGE_CACHE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04006455
Chris Masone6dcd2d2008-07-17 12:53:50 -04006456 if (zero_start != PAGE_CACHE_SIZE) {
6457 kaddr = kmap(page);
6458 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6459 flush_dcache_page(page);
6460 kunmap(page);
6461 }
Chris Mason247e7432008-07-17 12:53:51 -04006462 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006463 set_page_dirty(page);
Chris Mason50a9b212009-09-11 12:33:12 -04006464 SetPageUptodate(page);
Chris Mason5a3f23d2009-03-31 13:27:11 -04006465
Chris Mason257c62e2009-10-13 13:21:08 -04006466 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6467 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6468
Josef Bacik2ac55d42010-02-03 19:33:23 +00006469 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
Chris Mason9ebefb182007-06-15 13:50:00 -04006470
6471out_unlock:
Chris Mason50a9b212009-09-11 12:33:12 -04006472 if (!ret)
6473 return VM_FAULT_LOCKED;
Chris Mason9ebefb182007-06-15 13:50:00 -04006474 unlock_page(page);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04006475 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason1832a6d2007-12-21 16:27:21 -05006476out:
Chris Mason9ebefb182007-06-15 13:50:00 -04006477 return ret;
6478}
6479
Chris Mason39279cc2007-06-12 06:35:45 -04006480static void btrfs_truncate(struct inode *inode)
6481{
6482 struct btrfs_root *root = BTRFS_I(inode)->root;
6483 int ret;
6484 struct btrfs_trans_handle *trans;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04006485 unsigned long nr;
Chris Masondbe674a2008-07-17 12:54:05 -04006486 u64 mask = root->sectorsize - 1;
Chris Mason39279cc2007-06-12 06:35:45 -04006487
Yan, Zheng80825102009-11-12 09:35:36 +00006488 if (!S_ISREG(inode->i_mode)) {
6489 WARN_ON(1);
Chris Mason39279cc2007-06-12 06:35:45 -04006490 return;
Yan, Zheng80825102009-11-12 09:35:36 +00006491 }
Chris Mason39279cc2007-06-12 06:35:45 -04006492
Josef Bacik5d5e1032009-10-13 16:46:49 -04006493 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6494 if (ret)
6495 return;
Yan, Zheng80825102009-11-12 09:35:36 +00006496
Chris Mason4a096752008-07-21 10:29:44 -04006497 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
Yan, Zheng80825102009-11-12 09:35:36 +00006498 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006499
Yan, Zhengd68fc572010-05-16 10:49:58 -04006500 trans = btrfs_start_transaction(root, 0);
6501 BUG_ON(IS_ERR(trans));
Yan, Zheng80825102009-11-12 09:35:36 +00006502 btrfs_set_trans_block_group(trans, inode);
Yan, Zhengd68fc572010-05-16 10:49:58 -04006503 trans->block_rsv = root->orphan_block_rsv;
Chris Mason5a3f23d2009-03-31 13:27:11 -04006504
6505 /*
6506 * setattr is responsible for setting the ordered_data_close flag,
6507 * but that is only tested during the last file release. That
6508 * could happen well after the next commit, leaving a great big
6509 * window where new writes may get lost if someone chooses to write
6510 * to this file after truncating to zero
6511 *
6512 * The inode doesn't have any dirty data here, and so if we commit
6513 * this is a noop. If someone immediately starts writing to the inode
6514 * it is very likely we'll catch some of their writes in this
6515 * transaction, and the commit will find this file on the ordered
6516 * data list with good things to send down.
6517 *
6518 * This is a best effort solution, there is still a window where
6519 * using truncate to replace the contents of the file will
6520 * end up with a zero length file after a crash.
6521 */
6522 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6523 btrfs_add_ordered_operation(trans, root, inode);
6524
Yan, Zheng80825102009-11-12 09:35:36 +00006525 while (1) {
Yan, Zhengd68fc572010-05-16 10:49:58 -04006526 if (!trans) {
6527 trans = btrfs_start_transaction(root, 0);
6528 BUG_ON(IS_ERR(trans));
6529 btrfs_set_trans_block_group(trans, inode);
6530 trans->block_rsv = root->orphan_block_rsv;
6531 }
6532
6533 ret = btrfs_block_rsv_check(trans, root,
6534 root->orphan_block_rsv, 0, 5);
6535 if (ret) {
6536 BUG_ON(ret != -EAGAIN);
6537 ret = btrfs_commit_transaction(trans, root);
6538 BUG_ON(ret);
6539 trans = NULL;
6540 continue;
6541 }
6542
Yan, Zheng80825102009-11-12 09:35:36 +00006543 ret = btrfs_truncate_inode_items(trans, root, inode,
6544 inode->i_size,
6545 BTRFS_EXTENT_DATA_KEY);
6546 if (ret != -EAGAIN)
6547 break;
Chris Mason39279cc2007-06-12 06:35:45 -04006548
Yan, Zheng80825102009-11-12 09:35:36 +00006549 ret = btrfs_update_inode(trans, root, inode);
6550 BUG_ON(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04006551
Yan, Zheng80825102009-11-12 09:35:36 +00006552 nr = trans->blocks_used;
6553 btrfs_end_transaction(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04006554 trans = NULL;
Yan, Zheng80825102009-11-12 09:35:36 +00006555 btrfs_btree_balance_dirty(root, nr);
Yan, Zheng80825102009-11-12 09:35:36 +00006556 }
6557
6558 if (ret == 0 && inode->i_nlink > 0) {
6559 ret = btrfs_orphan_del(trans, inode);
6560 BUG_ON(ret);
6561 }
6562
6563 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik7b128762008-07-24 12:17:14 -04006564 BUG_ON(ret);
6565
Josef Bacik7b128762008-07-24 12:17:14 -04006566 nr = trans->blocks_used;
Chris Mason89ce8a62008-06-25 16:01:31 -04006567 ret = btrfs_end_transaction_throttle(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04006568 BUG_ON(ret);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04006569 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04006570}
6571
Sven Wegener3b963622008-06-09 21:57:42 -04006572/*
Chris Masond352ac62008-09-29 15:18:18 -04006573 * create a new subvolume directory/inode (helper for the ioctl).
6574 */
Yan Zhengd2fb3432008-12-11 16:30:39 -05006575int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
Yan, Zheng76dda932009-09-21 16:00:26 -04006576 struct btrfs_root *new_root,
Yan Zhengd2fb3432008-12-11 16:30:39 -05006577 u64 new_dirid, u64 alloc_hint)
Chris Mason39279cc2007-06-12 06:35:45 -04006578{
Chris Mason39279cc2007-06-12 06:35:45 -04006579 struct inode *inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04006580 int err;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006581 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006582
Josef Bacikaec74772008-07-24 12:12:38 -04006583 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
Yan Zhengd2fb3432008-12-11 16:30:39 -05006584 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
Chris Mason54aa1f42007-06-22 14:16:25 -04006585 if (IS_ERR(inode))
Christoph Hellwigf46b5a62008-06-11 21:53:53 -04006586 return PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006587 inode->i_op = &btrfs_dir_inode_operations;
6588 inode->i_fop = &btrfs_dir_file_operations;
6589
Chris Mason39279cc2007-06-12 06:35:45 -04006590 inode->i_nlink = 1;
Chris Masondbe674a2008-07-17 12:54:05 -04006591 btrfs_i_size_write(inode, 0);
Sven Wegener3b963622008-06-09 21:57:42 -04006592
Yan, Zheng76dda932009-09-21 16:00:26 -04006593 err = btrfs_update_inode(trans, new_root, inode);
6594 BUG_ON(err);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04006595
Yan, Zheng76dda932009-09-21 16:00:26 -04006596 iput(inode);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04006597 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006598}
6599
Chris Masond352ac62008-09-29 15:18:18 -04006600/* helper function for file defrag and space balancing. This
6601 * forces readahead on a given range of bytes in an inode
6602 */
Chris Masonedbd8d42007-12-21 16:27:24 -05006603unsigned long btrfs_force_ra(struct address_space *mapping,
Chris Mason86479a02007-09-10 19:58:16 -04006604 struct file_ra_state *ra, struct file *file,
6605 pgoff_t offset, pgoff_t last_index)
6606{
Chris Mason8e7bf942008-04-28 09:02:36 -04006607 pgoff_t req_size = last_index - offset + 1;
Chris Mason86479a02007-09-10 19:58:16 -04006608
Chris Mason86479a02007-09-10 19:58:16 -04006609 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6610 return offset + req_size;
Chris Mason86479a02007-09-10 19:58:16 -04006611}
6612
Chris Mason39279cc2007-06-12 06:35:45 -04006613struct inode *btrfs_alloc_inode(struct super_block *sb)
6614{
6615 struct btrfs_inode *ei;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006616 struct inode *inode;
Chris Mason39279cc2007-06-12 06:35:45 -04006617
6618 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6619 if (!ei)
6620 return NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006621
6622 ei->root = NULL;
6623 ei->space_info = NULL;
6624 ei->generation = 0;
6625 ei->sequence = 0;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04006626 ei->last_trans = 0;
Chris Mason257c62e2009-10-13 13:21:08 -04006627 ei->last_sub_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04006628 ei->logged_trans = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006629 ei->delalloc_bytes = 0;
6630 ei->reserved_bytes = 0;
6631 ei->disk_i_size = 0;
6632 ei->flags = 0;
6633 ei->index_cnt = (u64)-1;
6634 ei->last_unlink_trans = 0;
6635
Josef Bacik32c00af2009-10-08 13:34:05 -04006636 spin_lock_init(&ei->accounting_lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04006637 atomic_set(&ei->outstanding_extents, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04006638 ei->reserved_extents = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006639
6640 ei->ordered_data_close = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04006641 ei->orphan_meta_reserved = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006642 ei->dummy_inode = 0;
Li Zefan261507a02010-12-17 14:21:50 +08006643 ei->force_compress = BTRFS_COMPRESS_NONE;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006644
6645 inode = &ei->vfs_inode;
6646 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6647 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6648 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6649 mutex_init(&ei->log_mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006650 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Josef Bacik7b128762008-07-24 12:17:14 -04006651 INIT_LIST_HEAD(&ei->i_orphan);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006652 INIT_LIST_HEAD(&ei->delalloc_inodes);
Chris Mason5a3f23d2009-03-31 13:27:11 -04006653 INIT_LIST_HEAD(&ei->ordered_operations);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006654 RB_CLEAR_NODE(&ei->rb_node);
6655
6656 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04006657}
6658
6659void btrfs_destroy_inode(struct inode *inode)
6660{
Chris Masone6dcd2d2008-07-17 12:53:50 -04006661 struct btrfs_ordered_extent *ordered;
Chris Mason5a3f23d2009-03-31 13:27:11 -04006662 struct btrfs_root *root = BTRFS_I(inode)->root;
6663
Chris Mason39279cc2007-06-12 06:35:45 -04006664 WARN_ON(!list_empty(&inode->i_dentry));
6665 WARN_ON(inode->i_data.nrpages);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04006666 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6667 WARN_ON(BTRFS_I(inode)->reserved_extents);
Chris Mason39279cc2007-06-12 06:35:45 -04006668
Chris Mason5a3f23d2009-03-31 13:27:11 -04006669 /*
Josef Bacika6dbd422009-11-11 15:53:34 -05006670 * This can happen where we create an inode, but somebody else also
6671 * created the same inode and we need to destroy the one we already
6672 * created.
6673 */
6674 if (!root)
6675 goto free;
6676
6677 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04006678 * Make sure we're properly removed from the ordered operation
6679 * lists.
6680 */
6681 smp_mb();
6682 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6683 spin_lock(&root->fs_info->ordered_extent_lock);
6684 list_del_init(&BTRFS_I(inode)->ordered_operations);
6685 spin_unlock(&root->fs_info->ordered_extent_lock);
6686 }
6687
Josef Bacik0af3d002010-06-21 14:48:16 -04006688 if (root == root->fs_info->tree_root) {
6689 struct btrfs_block_group_cache *block_group;
6690
6691 block_group = btrfs_lookup_block_group(root->fs_info,
6692 BTRFS_I(inode)->block_group);
6693 if (block_group && block_group->inode == inode) {
6694 spin_lock(&block_group->lock);
6695 block_group->inode = NULL;
6696 spin_unlock(&block_group->lock);
6697 btrfs_put_block_group(block_group);
6698 } else if (block_group) {
6699 btrfs_put_block_group(block_group);
6700 }
6701 }
6702
Yan, Zhengd68fc572010-05-16 10:49:58 -04006703 spin_lock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04006704 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
Yan, Zheng80825102009-11-12 09:35:36 +00006705 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6706 inode->i_ino);
6707 list_del_init(&BTRFS_I(inode)->i_orphan);
Josef Bacik7b128762008-07-24 12:17:14 -04006708 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04006709 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04006710
Chris Masond3977122009-01-05 21:25:51 -05006711 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04006712 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6713 if (!ordered)
6714 break;
6715 else {
Chris Masond3977122009-01-05 21:25:51 -05006716 printk(KERN_ERR "btrfs found ordered "
6717 "extent %llu %llu on inode cleanup\n",
6718 (unsigned long long)ordered->file_offset,
6719 (unsigned long long)ordered->len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006720 btrfs_remove_ordered_extent(inode, ordered);
6721 btrfs_put_ordered_extent(ordered);
6722 btrfs_put_ordered_extent(ordered);
6723 }
6724 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006725 inode_tree_del(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04006726 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
Josef Bacika6dbd422009-11-11 15:53:34 -05006727free:
Chris Mason39279cc2007-06-12 06:35:45 -04006728 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6729}
6730
Al Viro45321ac2010-06-07 13:43:19 -04006731int btrfs_drop_inode(struct inode *inode)
Yan, Zheng76dda932009-09-21 16:00:26 -04006732{
6733 struct btrfs_root *root = BTRFS_I(inode)->root;
Al Viro45321ac2010-06-07 13:43:19 -04006734
Josef Bacik0af3d002010-06-21 14:48:16 -04006735 if (btrfs_root_refs(&root->root_item) == 0 &&
6736 root != root->fs_info->tree_root)
Al Viro45321ac2010-06-07 13:43:19 -04006737 return 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04006738 else
Al Viro45321ac2010-06-07 13:43:19 -04006739 return generic_drop_inode(inode);
Yan, Zheng76dda932009-09-21 16:00:26 -04006740}
6741
Sven Wegener0ee0fda2008-07-30 16:54:26 -04006742static void init_once(void *foo)
Chris Mason39279cc2007-06-12 06:35:45 -04006743{
6744 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6745
6746 inode_init_once(&ei->vfs_inode);
6747}
6748
6749void btrfs_destroy_cachep(void)
6750{
6751 if (btrfs_inode_cachep)
6752 kmem_cache_destroy(btrfs_inode_cachep);
6753 if (btrfs_trans_handle_cachep)
6754 kmem_cache_destroy(btrfs_trans_handle_cachep);
6755 if (btrfs_transaction_cachep)
6756 kmem_cache_destroy(btrfs_transaction_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04006757 if (btrfs_path_cachep)
6758 kmem_cache_destroy(btrfs_path_cachep);
6759}
6760
6761int btrfs_init_cachep(void)
6762{
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006763 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6764 sizeof(struct btrfs_inode), 0,
6765 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
Chris Mason39279cc2007-06-12 06:35:45 -04006766 if (!btrfs_inode_cachep)
6767 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006768
6769 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6770 sizeof(struct btrfs_trans_handle), 0,
6771 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006772 if (!btrfs_trans_handle_cachep)
6773 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006774
6775 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6776 sizeof(struct btrfs_transaction), 0,
6777 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006778 if (!btrfs_transaction_cachep)
6779 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006780
6781 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6782 sizeof(struct btrfs_path), 0,
6783 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006784 if (!btrfs_path_cachep)
6785 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006786
Chris Mason39279cc2007-06-12 06:35:45 -04006787 return 0;
6788fail:
6789 btrfs_destroy_cachep();
6790 return -ENOMEM;
6791}
6792
6793static int btrfs_getattr(struct vfsmount *mnt,
6794 struct dentry *dentry, struct kstat *stat)
6795{
6796 struct inode *inode = dentry->d_inode;
6797 generic_fillattr(inode, stat);
Chris Mason3394e162008-11-17 20:42:26 -05006798 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
Chris Masond6667462008-01-03 14:51:00 -05006799 stat->blksize = PAGE_CACHE_SIZE;
Yan Zhenga76a3cd2008-10-09 11:46:29 -04006800 stat->blocks = (inode_get_bytes(inode) +
6801 BTRFS_I(inode)->delalloc_bytes) >> 9;
Chris Mason39279cc2007-06-12 06:35:45 -04006802 return 0;
6803}
6804
Chris Masond3977122009-01-05 21:25:51 -05006805static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6806 struct inode *new_dir, struct dentry *new_dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04006807{
6808 struct btrfs_trans_handle *trans;
6809 struct btrfs_root *root = BTRFS_I(old_dir)->root;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006810 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04006811 struct inode *new_inode = new_dentry->d_inode;
6812 struct inode *old_inode = old_dentry->d_inode;
6813 struct timespec ctime = CURRENT_TIME;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006814 u64 index = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006815 u64 root_objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04006816 int ret;
6817
Yan, Zhengf679a842009-09-24 09:17:31 -04006818 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6819 return -EPERM;
6820
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006821 /* we only allow rename subvolume link between subvolumes */
6822 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
Chris Mason3394e162008-11-17 20:42:26 -05006823 return -EXDEV;
6824
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006825 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6826 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6827 return -ENOTEMPTY;
6828
Chris Mason39279cc2007-06-12 06:35:45 -04006829 if (S_ISDIR(old_inode->i_mode) && new_inode &&
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006830 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Chris Mason39279cc2007-06-12 06:35:45 -04006831 return -ENOTEMPTY;
Chris Mason5a3f23d2009-03-31 13:27:11 -04006832 /*
6833 * we're using rename to replace one file with another.
6834 * and the replacement file is large. Start IO on it now so
6835 * we don't add too much work to the end of the transaction
6836 */
Bartlomiej Zolnierkiewicz4baf8c92009-08-07 13:47:08 -04006837 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
Chris Mason5a3f23d2009-03-31 13:27:11 -04006838 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6839 filemap_flush(old_inode->i_mapping);
6840
Yan, Zheng76dda932009-09-21 16:00:26 -04006841 /* close the racy window with snapshot create/destroy ioctl */
6842 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6843 down_read(&root->fs_info->subvol_sem);
Yan, Zhenga22285a2010-05-16 10:48:46 -04006844 /*
6845 * We want to reserve the absolute worst case amount of items. So if
6846 * both inodes are subvols and we need to unlink them then that would
6847 * require 4 item modifications, but if they are both normal inodes it
6848 * would require 5 item modifications, so we'll assume their normal
6849 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6850 * should cover the worst case number of items we'll modify.
6851 */
6852 trans = btrfs_start_transaction(root, 20);
6853 if (IS_ERR(trans))
6854 return PTR_ERR(trans);
Chris Mason5f39d392007-10-15 16:14:19 -04006855
Yan, Zhenga5719522009-09-24 09:17:31 -04006856 btrfs_set_trans_block_group(trans, new_dir);
Chris Mason39279cc2007-06-12 06:35:45 -04006857
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006858 if (dest != root)
6859 btrfs_record_root_in_trans(trans, dest);
6860
Yan, Zhenga5719522009-09-24 09:17:31 -04006861 ret = btrfs_set_inode_index(new_dir, &index);
6862 if (ret)
6863 goto out_fail;
Chris Mason5a3f23d2009-03-31 13:27:11 -04006864
Yan, Zhenga5719522009-09-24 09:17:31 -04006865 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006866 /* force full log commit if subvolume involved. */
6867 root->fs_info->last_trans_log_full_commit = trans->transid;
6868 } else {
Yan, Zhenga5719522009-09-24 09:17:31 -04006869 ret = btrfs_insert_inode_ref(trans, dest,
6870 new_dentry->d_name.name,
6871 new_dentry->d_name.len,
6872 old_inode->i_ino,
6873 new_dir->i_ino, index);
6874 if (ret)
6875 goto out_fail;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006876 /*
6877 * this is an ugly little race, but the rename is required
6878 * to make sure that if we crash, the inode is either at the
6879 * old name or the new one. pinning the log transaction lets
6880 * us make sure we don't allow a log commit to come in after
6881 * we unlink the name but before we add the new name back in.
6882 */
6883 btrfs_pin_log_trans(root);
6884 }
Chris Mason12fcfd22009-03-24 10:24:20 -04006885 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04006886 * make sure the inode gets flushed if it is replacing
6887 * something.
6888 */
6889 if (new_inode && new_inode->i_size &&
6890 old_inode && S_ISREG(old_inode->i_mode)) {
6891 btrfs_add_ordered_operation(trans, root, old_inode);
6892 }
6893
Chris Mason39279cc2007-06-12 06:35:45 -04006894 old_dir->i_ctime = old_dir->i_mtime = ctime;
6895 new_dir->i_ctime = new_dir->i_mtime = ctime;
6896 old_inode->i_ctime = ctime;
Chris Mason5f39d392007-10-15 16:14:19 -04006897
Chris Mason12fcfd22009-03-24 10:24:20 -04006898 if (old_dentry->d_parent != new_dentry->d_parent)
6899 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6900
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006901 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6902 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6903 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6904 old_dentry->d_name.name,
6905 old_dentry->d_name.len);
6906 } else {
6907 btrfs_inc_nlink(old_dentry->d_inode);
6908 ret = btrfs_unlink_inode(trans, root, old_dir,
6909 old_dentry->d_inode,
6910 old_dentry->d_name.name,
6911 old_dentry->d_name.len);
6912 }
6913 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04006914
6915 if (new_inode) {
6916 new_inode->i_ctime = CURRENT_TIME;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006917 if (unlikely(new_inode->i_ino ==
6918 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6919 root_objectid = BTRFS_I(new_inode)->location.objectid;
6920 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6921 root_objectid,
6922 new_dentry->d_name.name,
6923 new_dentry->d_name.len);
6924 BUG_ON(new_inode->i_nlink == 0);
6925 } else {
6926 ret = btrfs_unlink_inode(trans, dest, new_dir,
6927 new_dentry->d_inode,
6928 new_dentry->d_name.name,
6929 new_dentry->d_name.len);
6930 }
6931 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04006932 if (new_inode->i_nlink == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04006933 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006934 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04006935 }
Chris Mason39279cc2007-06-12 06:35:45 -04006936 }
Josef Bacikaec74772008-07-24 12:12:38 -04006937
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006938 ret = btrfs_add_link(trans, new_dir, old_inode,
6939 new_dentry->d_name.name,
Yan, Zhenga5719522009-09-24 09:17:31 -04006940 new_dentry->d_name.len, 0, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006941 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04006942
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006943 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
Josef Bacik6a912212010-11-20 09:48:00 +00006944 struct dentry *parent = dget_parent(new_dentry);
6945 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6946 dput(parent);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006947 btrfs_end_log_trans(root);
6948 }
Chris Mason39279cc2007-06-12 06:35:45 -04006949out_fail:
Chris Masonab78c842008-07-29 16:15:18 -04006950 btrfs_end_transaction_throttle(trans, root);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006951
Yan, Zheng76dda932009-09-21 16:00:26 -04006952 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6953 up_read(&root->fs_info->subvol_sem);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006954
Chris Mason39279cc2007-06-12 06:35:45 -04006955 return ret;
6956}
6957
Chris Masond352ac62008-09-29 15:18:18 -04006958/*
6959 * some fairly slow code that needs optimization. This walks the list
6960 * of all the inodes with pending delalloc and forces them to disk.
6961 */
Yan, Zheng24bbcf02009-11-12 09:36:34 +00006962int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
Chris Masonea8c2812008-08-04 23:17:27 -04006963{
6964 struct list_head *head = &root->fs_info->delalloc_inodes;
6965 struct btrfs_inode *binode;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04006966 struct inode *inode;
Chris Masonea8c2812008-08-04 23:17:27 -04006967
Yan Zhengc146afa2008-11-12 14:34:12 -05006968 if (root->fs_info->sb->s_flags & MS_RDONLY)
6969 return -EROFS;
6970
Chris Mason75eff682008-12-15 15:54:40 -05006971 spin_lock(&root->fs_info->delalloc_lock);
Chris Masond3977122009-01-05 21:25:51 -05006972 while (!list_empty(head)) {
Chris Masonea8c2812008-08-04 23:17:27 -04006973 binode = list_entry(head->next, struct btrfs_inode,
6974 delalloc_inodes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04006975 inode = igrab(&binode->vfs_inode);
6976 if (!inode)
6977 list_del_init(&binode->delalloc_inodes);
Chris Mason75eff682008-12-15 15:54:40 -05006978 spin_unlock(&root->fs_info->delalloc_lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04006979 if (inode) {
Chris Mason8c8bee12008-09-29 11:19:10 -04006980 filemap_flush(inode->i_mapping);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00006981 if (delay_iput)
6982 btrfs_add_delayed_iput(inode);
6983 else
6984 iput(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04006985 }
6986 cond_resched();
Chris Mason75eff682008-12-15 15:54:40 -05006987 spin_lock(&root->fs_info->delalloc_lock);
Chris Masonea8c2812008-08-04 23:17:27 -04006988 }
Chris Mason75eff682008-12-15 15:54:40 -05006989 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason8c8bee12008-09-29 11:19:10 -04006990
6991 /* the filemap_flush will queue IO into the worker threads, but
6992 * we have to make sure the IO is actually started and that
6993 * ordered extents get created before we return
6994 */
6995 atomic_inc(&root->fs_info->async_submit_draining);
Chris Masond3977122009-01-05 21:25:51 -05006996 while (atomic_read(&root->fs_info->nr_async_submits) ||
Chris Mason771ed682008-11-06 22:02:51 -05006997 atomic_read(&root->fs_info->async_delalloc_pages)) {
Chris Mason8c8bee12008-09-29 11:19:10 -04006998 wait_event(root->fs_info->async_submit_wait,
Chris Mason771ed682008-11-06 22:02:51 -05006999 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7000 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
Chris Mason8c8bee12008-09-29 11:19:10 -04007001 }
7002 atomic_dec(&root->fs_info->async_submit_draining);
Chris Masonea8c2812008-08-04 23:17:27 -04007003 return 0;
7004}
7005
Josef Bacik0019f102010-10-15 15:18:40 -04007006int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7007 int sync)
Yan, Zheng5da9d012010-05-16 10:46:25 -04007008{
7009 struct btrfs_inode *binode;
7010 struct inode *inode = NULL;
7011
7012 spin_lock(&root->fs_info->delalloc_lock);
7013 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7014 binode = list_entry(root->fs_info->delalloc_inodes.next,
7015 struct btrfs_inode, delalloc_inodes);
7016 inode = igrab(&binode->vfs_inode);
7017 if (inode) {
7018 list_move_tail(&binode->delalloc_inodes,
7019 &root->fs_info->delalloc_inodes);
7020 break;
7021 }
7022
7023 list_del_init(&binode->delalloc_inodes);
7024 cond_resched_lock(&root->fs_info->delalloc_lock);
7025 }
7026 spin_unlock(&root->fs_info->delalloc_lock);
7027
7028 if (inode) {
Josef Bacik0019f102010-10-15 15:18:40 -04007029 if (sync) {
7030 filemap_write_and_wait(inode->i_mapping);
7031 /*
7032 * We have to do this because compression doesn't
7033 * actually set PG_writeback until it submits the pages
7034 * for IO, which happens in an async thread, so we could
7035 * race and not actually wait for any writeback pages
7036 * because they've not been submitted yet. Technically
7037 * this could still be the case for the ordered stuff
7038 * since the async thread may not have started to do its
7039 * work yet. If this becomes the case then we need to
7040 * figure out a way to make sure that in writepage we
7041 * wait for any async pages to be submitted before
7042 * returning so that fdatawait does what its supposed to
7043 * do.
7044 */
7045 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7046 } else {
7047 filemap_flush(inode->i_mapping);
7048 }
Yan, Zheng5da9d012010-05-16 10:46:25 -04007049 if (delay_iput)
7050 btrfs_add_delayed_iput(inode);
7051 else
7052 iput(inode);
7053 return 1;
7054 }
7055 return 0;
7056}
7057
Chris Mason39279cc2007-06-12 06:35:45 -04007058static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7059 const char *symname)
7060{
7061 struct btrfs_trans_handle *trans;
7062 struct btrfs_root *root = BTRFS_I(dir)->root;
7063 struct btrfs_path *path;
7064 struct btrfs_key key;
Chris Mason1832a6d2007-12-21 16:27:21 -05007065 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04007066 int err;
7067 int drop_inode = 0;
7068 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04007069 u64 index = 0 ;
Chris Mason39279cc2007-06-12 06:35:45 -04007070 int name_len;
7071 int datasize;
Chris Mason5f39d392007-10-15 16:14:19 -04007072 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04007073 struct btrfs_file_extent_item *ei;
Chris Mason5f39d392007-10-15 16:14:19 -04007074 struct extent_buffer *leaf;
Chris Mason1832a6d2007-12-21 16:27:21 -05007075 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04007076
7077 name_len = strlen(symname) + 1;
7078 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7079 return -ENAMETOOLONG;
Chris Mason1832a6d2007-12-21 16:27:21 -05007080
Yan, Zhenga22285a2010-05-16 10:48:46 -04007081 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7082 if (err)
7083 return err;
Josef Bacik9ed74f22009-09-11 16:12:44 -04007084 /*
7085 * 2 items for inode item and ref
7086 * 2 items for dir items
7087 * 1 item for xattr if selinux is on
7088 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04007089 trans = btrfs_start_transaction(root, 5);
7090 if (IS_ERR(trans))
7091 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05007092
Chris Mason39279cc2007-06-12 06:35:45 -04007093 btrfs_set_trans_block_group(trans, dir);
7094
Josef Bacikaec74772008-07-24 12:12:38 -04007095 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Josef Bacika1b075d2010-11-19 20:36:11 +00007096 dentry->d_name.len, dir->i_ino, objectid,
Chris Mason00e4e6b2008-08-05 11:18:09 -04007097 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7098 &index);
Chris Mason39279cc2007-06-12 06:35:45 -04007099 err = PTR_ERR(inode);
7100 if (IS_ERR(inode))
7101 goto out_unlock;
7102
Yan, Zhengf34f57a2009-11-12 09:35:27 +00007103 err = btrfs_init_inode_security(trans, inode, dir);
Josef Bacik33268ea2008-07-24 12:16:36 -04007104 if (err) {
7105 drop_inode = 1;
7106 goto out_unlock;
7107 }
7108
Chris Mason39279cc2007-06-12 06:35:45 -04007109 btrfs_set_trans_block_group(trans, inode);
Josef Bacika1b075d2010-11-19 20:36:11 +00007110 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04007111 if (err)
7112 drop_inode = 1;
7113 else {
7114 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04007115 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04007116 inode->i_fop = &btrfs_file_operations;
7117 inode->i_op = &btrfs_file_inode_operations;
Chris Masond1310b22008-01-24 16:13:08 -05007118 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04007119 }
Chris Mason39279cc2007-06-12 06:35:45 -04007120 btrfs_update_inode_block_group(trans, inode);
7121 btrfs_update_inode_block_group(trans, dir);
7122 if (drop_inode)
7123 goto out_unlock;
7124
7125 path = btrfs_alloc_path();
7126 BUG_ON(!path);
7127 key.objectid = inode->i_ino;
7128 key.offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04007129 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7130 datasize = btrfs_file_extent_calc_inline_size(name_len);
7131 err = btrfs_insert_empty_item(trans, root, path, &key,
7132 datasize);
Chris Mason54aa1f42007-06-22 14:16:25 -04007133 if (err) {
7134 drop_inode = 1;
7135 goto out_unlock;
7136 }
Chris Mason5f39d392007-10-15 16:14:19 -04007137 leaf = path->nodes[0];
7138 ei = btrfs_item_ptr(leaf, path->slots[0],
7139 struct btrfs_file_extent_item);
7140 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7141 btrfs_set_file_extent_type(leaf, ei,
Chris Mason39279cc2007-06-12 06:35:45 -04007142 BTRFS_FILE_EXTENT_INLINE);
Chris Masonc8b97812008-10-29 14:49:59 -04007143 btrfs_set_file_extent_encryption(leaf, ei, 0);
7144 btrfs_set_file_extent_compression(leaf, ei, 0);
7145 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7146 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7147
Chris Mason39279cc2007-06-12 06:35:45 -04007148 ptr = btrfs_file_extent_inline_start(ei);
Chris Mason5f39d392007-10-15 16:14:19 -04007149 write_extent_buffer(leaf, symname, ptr, name_len);
7150 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04007151 btrfs_free_path(path);
Chris Mason5f39d392007-10-15 16:14:19 -04007152
Chris Mason39279cc2007-06-12 06:35:45 -04007153 inode->i_op = &btrfs_symlink_inode_operations;
7154 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04007155 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Yan Zhengd899e052008-10-30 14:25:28 -04007156 inode_set_bytes(inode, name_len);
Chris Masondbe674a2008-07-17 12:54:05 -04007157 btrfs_i_size_write(inode, name_len - 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04007158 err = btrfs_update_inode(trans, root, inode);
7159 if (err)
7160 drop_inode = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04007161
7162out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04007163 nr = trans->blocks_used;
Chris Masonab78c842008-07-29 16:15:18 -04007164 btrfs_end_transaction_throttle(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04007165 if (drop_inode) {
7166 inode_dec_link_count(inode);
7167 iput(inode);
7168 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04007169 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04007170 return err;
7171}
Chris Mason16432982008-04-10 10:23:21 -04007172
Josef Bacik0af3d002010-06-21 14:48:16 -04007173static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7174 u64 start, u64 num_bytes, u64 min_size,
7175 loff_t actual_len, u64 *alloc_hint,
7176 struct btrfs_trans_handle *trans)
Yan Zhengd899e052008-10-30 14:25:28 -04007177{
Yan Zhengd899e052008-10-30 14:25:28 -04007178 struct btrfs_root *root = BTRFS_I(inode)->root;
7179 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -04007180 u64 cur_offset = start;
Josef Bacik55a61d12010-11-22 18:50:32 +00007181 u64 i_size;
Yan Zhengd899e052008-10-30 14:25:28 -04007182 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04007183 bool own_trans = true;
Yan Zhengd899e052008-10-30 14:25:28 -04007184
Josef Bacik0af3d002010-06-21 14:48:16 -04007185 if (trans)
7186 own_trans = false;
Yan Zhengd899e052008-10-30 14:25:28 -04007187 while (num_bytes > 0) {
Josef Bacik0af3d002010-06-21 14:48:16 -04007188 if (own_trans) {
7189 trans = btrfs_start_transaction(root, 3);
7190 if (IS_ERR(trans)) {
7191 ret = PTR_ERR(trans);
7192 break;
7193 }
Yan Zhengd899e052008-10-30 14:25:28 -04007194 }
Yan, Zheng5a303d52009-11-12 09:34:52 +00007195
Yan, Zhengefa56462010-05-16 10:49:59 -04007196 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7197 0, *alloc_hint, (u64)-1, &ins, 1);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007198 if (ret) {
Josef Bacik0af3d002010-06-21 14:48:16 -04007199 if (own_trans)
7200 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04007201 break;
Yan, Zheng5a303d52009-11-12 09:34:52 +00007202 }
7203
Yan Zhengd899e052008-10-30 14:25:28 -04007204 ret = insert_reserved_file_extent(trans, inode,
7205 cur_offset, ins.objectid,
7206 ins.offset, ins.offset,
Yan, Zheng920bbbf2009-11-12 09:34:08 +00007207 ins.offset, 0, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04007208 BTRFS_FILE_EXTENT_PREALLOC);
7209 BUG_ON(ret);
Chris Masona1ed8352009-09-11 12:27:37 -04007210 btrfs_drop_extent_cache(inode, cur_offset,
7211 cur_offset + ins.offset -1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007212
Yan Zhengd899e052008-10-30 14:25:28 -04007213 num_bytes -= ins.offset;
7214 cur_offset += ins.offset;
Yan, Zhengefa56462010-05-16 10:49:59 -04007215 *alloc_hint = ins.objectid + ins.offset;
Yan, Zheng5a303d52009-11-12 09:34:52 +00007216
Yan Zhengd899e052008-10-30 14:25:28 -04007217 inode->i_ctime = CURRENT_TIME;
Christoph Hellwig6cbff002009-04-17 10:37:41 +02007218 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Yan Zhengd899e052008-10-30 14:25:28 -04007219 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
Yan, Zhengefa56462010-05-16 10:49:59 -04007220 (actual_len > inode->i_size) &&
7221 (cur_offset > inode->i_size)) {
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00007222 if (cur_offset > actual_len)
Josef Bacik55a61d12010-11-22 18:50:32 +00007223 i_size = actual_len;
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00007224 else
Josef Bacik55a61d12010-11-22 18:50:32 +00007225 i_size = cur_offset;
7226 i_size_write(inode, i_size);
7227 btrfs_ordered_update_i_size(inode, i_size, NULL);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007228 }
7229
Yan Zhengd899e052008-10-30 14:25:28 -04007230 ret = btrfs_update_inode(trans, root, inode);
7231 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -04007232
Josef Bacik0af3d002010-06-21 14:48:16 -04007233 if (own_trans)
7234 btrfs_end_transaction(trans, root);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007235 }
Yan Zhengd899e052008-10-30 14:25:28 -04007236 return ret;
7237}
7238
Josef Bacik0af3d002010-06-21 14:48:16 -04007239int btrfs_prealloc_file_range(struct inode *inode, int mode,
7240 u64 start, u64 num_bytes, u64 min_size,
7241 loff_t actual_len, u64 *alloc_hint)
7242{
7243 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7244 min_size, actual_len, alloc_hint,
7245 NULL);
7246}
7247
7248int btrfs_prealloc_file_range_trans(struct inode *inode,
7249 struct btrfs_trans_handle *trans, int mode,
7250 u64 start, u64 num_bytes, u64 min_size,
7251 loff_t actual_len, u64 *alloc_hint)
7252{
7253 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7254 min_size, actual_len, alloc_hint, trans);
7255}
7256
Yan Zhengd899e052008-10-30 14:25:28 -04007257static long btrfs_fallocate(struct inode *inode, int mode,
7258 loff_t offset, loff_t len)
7259{
Josef Bacik2ac55d42010-02-03 19:33:23 +00007260 struct extent_state *cached_state = NULL;
Yan Zhengd899e052008-10-30 14:25:28 -04007261 u64 cur_offset;
7262 u64 last_byte;
7263 u64 alloc_start;
7264 u64 alloc_end;
7265 u64 alloc_hint = 0;
Chris Masone980b502009-04-24 14:39:24 -04007266 u64 locked_end;
Yan Zhengd899e052008-10-30 14:25:28 -04007267 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
7268 struct extent_map *em;
7269 int ret;
7270
7271 alloc_start = offset & ~mask;
7272 alloc_end = (offset + len + mask) & ~mask;
7273
Chris Mason546888d2009-04-21 11:53:38 -04007274 /*
7275 * wait for ordered IO before we have any locks. We'll loop again
7276 * below with the locks held.
7277 */
7278 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
7279
Yan Zhengd899e052008-10-30 14:25:28 -04007280 mutex_lock(&inode->i_mutex);
Josef Bacik0ed42a62010-11-22 18:55:39 +00007281 ret = inode_newsize_ok(inode, alloc_end);
7282 if (ret)
7283 goto out;
7284
Yan Zhengd899e052008-10-30 14:25:28 -04007285 if (alloc_start > inode->i_size) {
7286 ret = btrfs_cont_expand(inode, alloc_start);
7287 if (ret)
7288 goto out;
7289 }
7290
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04007291 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
Josef Bacika970b0a2009-06-27 21:07:34 -04007292 if (ret)
7293 goto out;
7294
Chris Masone980b502009-04-24 14:39:24 -04007295 locked_end = alloc_end - 1;
Yan Zhengd899e052008-10-30 14:25:28 -04007296 while (1) {
7297 struct btrfs_ordered_extent *ordered;
Chris Mason546888d2009-04-21 11:53:38 -04007298
Chris Mason546888d2009-04-21 11:53:38 -04007299 /* the extent lock is ordered inside the running
7300 * transaction
7301 */
Josef Bacik2ac55d42010-02-03 19:33:23 +00007302 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
7303 locked_end, 0, &cached_state, GFP_NOFS);
Yan Zhengd899e052008-10-30 14:25:28 -04007304 ordered = btrfs_lookup_first_ordered_extent(inode,
7305 alloc_end - 1);
7306 if (ordered &&
7307 ordered->file_offset + ordered->len > alloc_start &&
7308 ordered->file_offset < alloc_end) {
7309 btrfs_put_ordered_extent(ordered);
Josef Bacik2ac55d42010-02-03 19:33:23 +00007310 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7311 alloc_start, locked_end,
7312 &cached_state, GFP_NOFS);
Chris Mason546888d2009-04-21 11:53:38 -04007313 /*
7314 * we can't wait on the range with the transaction
7315 * running or with the extent lock held
7316 */
Yan Zhengd899e052008-10-30 14:25:28 -04007317 btrfs_wait_ordered_range(inode, alloc_start,
7318 alloc_end - alloc_start);
7319 } else {
7320 if (ordered)
7321 btrfs_put_ordered_extent(ordered);
7322 break;
7323 }
7324 }
7325
7326 cur_offset = alloc_start;
7327 while (1) {
7328 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
7329 alloc_end - cur_offset, 0);
7330 BUG_ON(IS_ERR(em) || !em);
7331 last_byte = min(extent_map_end(em), alloc_end);
7332 last_byte = (last_byte + mask) & ~mask;
Yan, Zheng5a303d52009-11-12 09:34:52 +00007333 if (em->block_start == EXTENT_MAP_HOLE ||
7334 (cur_offset >= inode->i_size &&
7335 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
Josef Bacik83609772010-06-07 18:26:37 +00007336 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
Yan, Zhengefa56462010-05-16 10:49:59 -04007337 last_byte - cur_offset,
7338 1 << inode->i_blkbits,
7339 offset + len,
7340 &alloc_hint);
Yan Zhengd899e052008-10-30 14:25:28 -04007341 if (ret < 0) {
7342 free_extent_map(em);
7343 break;
7344 }
7345 }
Yan Zhengd899e052008-10-30 14:25:28 -04007346 free_extent_map(em);
7347
7348 cur_offset = last_byte;
7349 if (cur_offset >= alloc_end) {
7350 ret = 0;
7351 break;
7352 }
7353 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00007354 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
7355 &cached_state, GFP_NOFS);
Chris Mason546888d2009-04-21 11:53:38 -04007356
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04007357 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
Yan Zhengd899e052008-10-30 14:25:28 -04007358out:
7359 mutex_unlock(&inode->i_mutex);
7360 return ret;
7361}
7362
Chris Masone6dcd2d2008-07-17 12:53:50 -04007363static int btrfs_set_page_dirty(struct page *page)
7364{
Chris Masone6dcd2d2008-07-17 12:53:50 -04007365 return __set_page_dirty_nobuffers(page);
7366}
7367
Sven Wegener0ee0fda2008-07-30 16:54:26 -04007368static int btrfs_permission(struct inode *inode, int mask)
Yanfdebe2b2008-01-14 13:26:08 -05007369{
Li Zefanb83cc962010-12-20 16:04:08 +08007370 struct btrfs_root *root = BTRFS_I(inode)->root;
7371
7372 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7373 return -EROFS;
Christoph Hellwig6cbff002009-04-17 10:37:41 +02007374 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
Yanfdebe2b2008-01-14 13:26:08 -05007375 return -EACCES;
Josef Bacik33268ea2008-07-24 12:16:36 -04007376 return generic_permission(inode, mask, btrfs_check_acl);
Yanfdebe2b2008-01-14 13:26:08 -05007377}
Chris Mason39279cc2007-06-12 06:35:45 -04007378
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007379static const struct inode_operations btrfs_dir_inode_operations = {
Chris Mason3394e162008-11-17 20:42:26 -05007380 .getattr = btrfs_getattr,
Chris Mason39279cc2007-06-12 06:35:45 -04007381 .lookup = btrfs_lookup,
7382 .create = btrfs_create,
7383 .unlink = btrfs_unlink,
7384 .link = btrfs_link,
7385 .mkdir = btrfs_mkdir,
7386 .rmdir = btrfs_rmdir,
7387 .rename = btrfs_rename,
7388 .symlink = btrfs_symlink,
7389 .setattr = btrfs_setattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04007390 .mknod = btrfs_mknod,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007391 .setxattr = btrfs_setxattr,
7392 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05007393 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007394 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05007395 .permission = btrfs_permission,
Chris Mason39279cc2007-06-12 06:35:45 -04007396};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007397static const struct inode_operations btrfs_dir_ro_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007398 .lookup = btrfs_lookup,
Yanfdebe2b2008-01-14 13:26:08 -05007399 .permission = btrfs_permission,
Chris Mason39279cc2007-06-12 06:35:45 -04007400};
Yan, Zheng76dda932009-09-21 16:00:26 -04007401
Alexey Dobriyan828c0952009-10-01 15:43:56 -07007402static const struct file_operations btrfs_dir_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007403 .llseek = generic_file_llseek,
7404 .read = generic_read_dir,
David Woodhousecbdf5a22008-08-06 19:42:33 +01007405 .readdir = btrfs_real_readdir,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04007406 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04007407#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04007408 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04007409#endif
Sage Weil6bf13c02008-06-10 10:07:39 -04007410 .release = btrfs_release_file,
Chris Masone02119d2008-09-05 16:13:11 -04007411 .fsync = btrfs_sync_file,
Chris Mason39279cc2007-06-12 06:35:45 -04007412};
7413
Chris Masond1310b22008-01-24 16:13:08 -05007414static struct extent_io_ops btrfs_extent_io_ops = {
Chris Mason07157aa2007-08-30 08:50:51 -04007415 .fill_delalloc = run_delalloc_range,
Chris Mason065631f2008-02-20 12:07:25 -05007416 .submit_bio_hook = btrfs_submit_bio_hook,
Chris Mason239b14b2008-03-24 15:02:07 -04007417 .merge_bio_hook = btrfs_merge_bio_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04007418 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
Chris Masone6dcd2d2008-07-17 12:53:50 -04007419 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
Chris Mason247e7432008-07-17 12:53:51 -04007420 .writepage_start_hook = btrfs_writepage_start_hook,
Chris Mason1259ab72008-05-12 13:39:03 -04007421 .readpage_io_failed_hook = btrfs_io_failed_hook,
Chris Masonb0c68f82008-01-31 11:05:37 -05007422 .set_bit_hook = btrfs_set_bit_hook,
7423 .clear_bit_hook = btrfs_clear_bit_hook,
Josef Bacik9ed74f22009-09-11 16:12:44 -04007424 .merge_extent_hook = btrfs_merge_extent_hook,
7425 .split_extent_hook = btrfs_split_extent_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04007426};
7427
Chris Mason35054392009-01-21 13:11:13 -05007428/*
7429 * btrfs doesn't support the bmap operation because swapfiles
7430 * use bmap to make a mapping of extents in the file. They assume
7431 * these extents won't change over the life of the file and they
7432 * use the bmap result to do IO directly to the drive.
7433 *
7434 * the btrfs bmap call would return logical addresses that aren't
7435 * suitable for IO and they also will change frequently as COW
7436 * operations happen. So, swapfile + btrfs == corruption.
7437 *
7438 * For now we're avoiding this by dropping bmap.
7439 */
Alexey Dobriyan7f094102009-09-21 17:01:10 -07007440static const struct address_space_operations btrfs_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04007441 .readpage = btrfs_readpage,
7442 .writepage = btrfs_writepage,
Chris Masonb293f022007-11-01 19:45:34 -04007443 .writepages = btrfs_writepages,
Chris Mason3ab2fb52007-11-08 10:59:22 -05007444 .readpages = btrfs_readpages,
Chris Mason39279cc2007-06-12 06:35:45 -04007445 .sync_page = block_sync_page,
Chris Mason16432982008-04-10 10:23:21 -04007446 .direct_IO = btrfs_direct_IO,
Chris Masona52d9a82007-08-27 16:49:44 -04007447 .invalidatepage = btrfs_invalidatepage,
7448 .releasepage = btrfs_releasepage,
Chris Masone6dcd2d2008-07-17 12:53:50 -04007449 .set_page_dirty = btrfs_set_page_dirty,
Andi Kleen465fdd92009-09-16 11:50:18 +02007450 .error_remove_page = generic_error_remove_page,
Chris Mason39279cc2007-06-12 06:35:45 -04007451};
7452
Alexey Dobriyan7f094102009-09-21 17:01:10 -07007453static const struct address_space_operations btrfs_symlink_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04007454 .readpage = btrfs_readpage,
7455 .writepage = btrfs_writepage,
Chris Mason2bf5a722007-08-30 11:54:02 -04007456 .invalidatepage = btrfs_invalidatepage,
7457 .releasepage = btrfs_releasepage,
Chris Mason39279cc2007-06-12 06:35:45 -04007458};
7459
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007460static const struct inode_operations btrfs_file_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007461 .truncate = btrfs_truncate,
7462 .getattr = btrfs_getattr,
7463 .setattr = btrfs_setattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007464 .setxattr = btrfs_setxattr,
7465 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05007466 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007467 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05007468 .permission = btrfs_permission,
Yan Zhengd899e052008-10-30 14:25:28 -04007469 .fallocate = btrfs_fallocate,
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05007470 .fiemap = btrfs_fiemap,
Chris Mason39279cc2007-06-12 06:35:45 -04007471};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007472static const struct inode_operations btrfs_special_inode_operations = {
Josef Bacik618e21d2007-07-11 10:18:17 -04007473 .getattr = btrfs_getattr,
7474 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05007475 .permission = btrfs_permission,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007476 .setxattr = btrfs_setxattr,
7477 .getxattr = btrfs_getxattr,
Josef Bacik33268ea2008-07-24 12:16:36 -04007478 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007479 .removexattr = btrfs_removexattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04007480};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007481static const struct inode_operations btrfs_symlink_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007482 .readlink = generic_readlink,
7483 .follow_link = page_follow_link_light,
7484 .put_link = page_put_link,
Li Zefanf2095612010-11-19 02:05:24 +00007485 .getattr = btrfs_getattr,
Yanfdebe2b2008-01-14 13:26:08 -05007486 .permission = btrfs_permission,
Jim Owens0279b4c2009-02-04 09:29:13 -05007487 .setxattr = btrfs_setxattr,
7488 .getxattr = btrfs_getxattr,
7489 .listxattr = btrfs_listxattr,
7490 .removexattr = btrfs_removexattr,
Chris Mason39279cc2007-06-12 06:35:45 -04007491};
Yan, Zheng76dda932009-09-21 16:00:26 -04007492
Alexey Dobriyan82d339d2009-10-09 09:54:36 -04007493const struct dentry_operations btrfs_dentry_operations = {
Yan, Zheng76dda932009-09-21 16:00:26 -04007494 .d_delete = btrfs_dentry_delete,
7495};