blob: 41132339e2ead2be931e4a1501532b4fc4f96468 [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>
David Sterba7a36dde2011-05-06 15:33:15 +020040#include <linux/ratelimit.h>
Josef Bacik22c44fe2011-11-30 10:45:38 -050041#include <linux/mount.h>
Chris Mason4b4e25f2008-11-20 10:22:27 -050042#include "compat.h"
Chris Mason39279cc2007-06-12 06:35:45 -040043#include "ctree.h"
44#include "disk-io.h"
45#include "transaction.h"
46#include "btrfs_inode.h"
47#include "ioctl.h"
48#include "print-tree.h"
Chris Masone6dcd2d2008-07-17 12:53:50 -040049#include "ordered-data.h"
Christoph Hellwig95819c02008-08-28 06:21:17 -040050#include "xattr.h"
Chris Masone02119d2008-09-05 16:13:11 -040051#include "tree-log.h"
Jan Schmidt4a54c8c2011-07-22 15:41:52 +020052#include "volumes.h"
Chris Masonc8b97812008-10-29 14:49:59 -040053#include "compression.h"
Chris Masonb4ce94d2009-02-04 09:25:08 -050054#include "locking.h"
Josef Bacikdc89e982011-01-28 17:05:48 -050055#include "free-space-cache.h"
Li Zefan581bb052011-04-20 10:06:11 +080056#include "inode-map.h"
Chris Mason39279cc2007-06-12 06:35:45 -040057
58struct btrfs_iget_args {
59 u64 ino;
60 struct btrfs_root *root;
61};
62
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070063static const struct inode_operations btrfs_dir_inode_operations;
64static const struct inode_operations btrfs_symlink_inode_operations;
65static const struct inode_operations btrfs_dir_ro_inode_operations;
66static const struct inode_operations btrfs_special_inode_operations;
67static const struct inode_operations btrfs_file_inode_operations;
Alexey Dobriyan7f094102009-09-21 17:01:10 -070068static const struct address_space_operations btrfs_aops;
69static const struct address_space_operations btrfs_symlink_aops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -070070static const struct file_operations btrfs_dir_file_operations;
Chris Masond1310b22008-01-24 16:13:08 -050071static struct extent_io_ops btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -040072
73static struct kmem_cache *btrfs_inode_cachep;
74struct kmem_cache *btrfs_trans_handle_cachep;
75struct kmem_cache *btrfs_transaction_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040076struct kmem_cache *btrfs_path_cachep;
Josef Bacikdc89e982011-01-28 17:05:48 -050077struct kmem_cache *btrfs_free_space_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040078
79#define S_SHIFT 12
80static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
82 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
83 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
84 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
85 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
86 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
87 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
88};
89
Josef Bacika41ad392011-01-31 15:30:16 -050090static int btrfs_setsize(struct inode *inode, loff_t newsize);
91static int btrfs_truncate(struct inode *inode);
Chris Masonc8b97812008-10-29 14:49:59 -040092static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
Chris Mason771ed682008-11-06 22:02:51 -050093static noinline int cow_file_range(struct inode *inode,
94 struct page *locked_page,
95 u64 start, u64 end, int *page_started,
96 unsigned long *nr_written, int unlock);
Chris Mason21151332011-11-10 20:39:08 -050097static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode);
Josef Bacik7b128762008-07-24 12:17:14 -040099
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000100static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
Eric Paris2a7dba32011-02-01 11:05:39 -0500101 struct inode *inode, struct inode *dir,
102 const struct qstr *qstr)
Jim Owens0279b4c2009-02-04 09:29:13 -0500103{
104 int err;
105
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000106 err = btrfs_init_acl(trans, inode, dir);
Jim Owens0279b4c2009-02-04 09:29:13 -0500107 if (!err)
Eric Paris2a7dba32011-02-01 11:05:39 -0500108 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
Jim Owens0279b4c2009-02-04 09:29:13 -0500109 return err;
110}
111
Chris Masond352ac62008-09-29 15:18:18 -0400112/*
Chris Masonc8b97812008-10-29 14:49:59 -0400113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
116 */
Chris Masond3977122009-01-05 21:25:51 -0500117static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
Chris Masonc8b97812008-10-29 14:49:59 -0400118 struct btrfs_root *root, struct inode *inode,
119 u64 start, size_t size, size_t compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000120 int compress_type,
Chris Masonc8b97812008-10-29 14:49:59 -0400121 struct page **compressed_pages)
122{
123 struct btrfs_key key;
124 struct btrfs_path *path;
125 struct extent_buffer *leaf;
126 struct page *page = NULL;
127 char *kaddr;
128 unsigned long ptr;
129 struct btrfs_file_extent_item *ei;
130 int err = 0;
131 int ret;
132 size_t cur_size = size;
133 size_t datasize;
134 unsigned long offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400135
Li Zefanfe3f5662011-03-28 08:30:38 +0000136 if (compressed_size && compressed_pages)
Chris Masonc8b97812008-10-29 14:49:59 -0400137 cur_size = compressed_size;
Chris Masonc8b97812008-10-29 14:49:59 -0400138
Chris Masond3977122009-01-05 21:25:51 -0500139 path = btrfs_alloc_path();
140 if (!path)
Chris Masonc8b97812008-10-29 14:49:59 -0400141 return -ENOMEM;
142
Chris Masonb9473432009-03-13 11:00:37 -0400143 path->leave_spinning = 1;
Chris Masonc8b97812008-10-29 14:49:59 -0400144
Li Zefan33345d012011-04-20 10:31:50 +0800145 key.objectid = btrfs_ino(inode);
Chris Masonc8b97812008-10-29 14:49:59 -0400146 key.offset = start;
147 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
Chris Masonc8b97812008-10-29 14:49:59 -0400148 datasize = btrfs_file_extent_calc_inline_size(cur_size);
149
150 inode_add_bytes(inode, size);
151 ret = btrfs_insert_empty_item(trans, root, path, &key,
152 datasize);
153 BUG_ON(ret);
154 if (ret) {
155 err = ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400156 goto fail;
157 }
158 leaf = path->nodes[0];
159 ei = btrfs_item_ptr(leaf, path->slots[0],
160 struct btrfs_file_extent_item);
161 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
162 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
163 btrfs_set_file_extent_encryption(leaf, ei, 0);
164 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
165 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
166 ptr = btrfs_file_extent_inline_start(ei);
167
Li Zefan261507a02010-12-17 14:21:50 +0800168 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400169 struct page *cpage;
170 int i = 0;
Chris Masond3977122009-01-05 21:25:51 -0500171 while (compressed_size > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400172 cpage = compressed_pages[i];
Chris Mason5b050f02008-11-11 09:34:41 -0500173 cur_size = min_t(unsigned long, compressed_size,
Chris Masonc8b97812008-10-29 14:49:59 -0400174 PAGE_CACHE_SIZE);
175
Chris Masonb9473432009-03-13 11:00:37 -0400176 kaddr = kmap_atomic(cpage, KM_USER0);
Chris Masonc8b97812008-10-29 14:49:59 -0400177 write_extent_buffer(leaf, kaddr, ptr, cur_size);
Chris Masonb9473432009-03-13 11:00:37 -0400178 kunmap_atomic(kaddr, KM_USER0);
Chris Masonc8b97812008-10-29 14:49:59 -0400179
180 i++;
181 ptr += cur_size;
182 compressed_size -= cur_size;
183 }
184 btrfs_set_file_extent_compression(leaf, ei,
Li Zefan261507a02010-12-17 14:21:50 +0800185 compress_type);
Chris Masonc8b97812008-10-29 14:49:59 -0400186 } else {
187 page = find_get_page(inode->i_mapping,
188 start >> PAGE_CACHE_SHIFT);
189 btrfs_set_file_extent_compression(leaf, ei, 0);
190 kaddr = kmap_atomic(page, KM_USER0);
191 offset = start & (PAGE_CACHE_SIZE - 1);
192 write_extent_buffer(leaf, kaddr + offset, ptr, size);
193 kunmap_atomic(kaddr, KM_USER0);
194 page_cache_release(page);
195 }
196 btrfs_mark_buffer_dirty(leaf);
197 btrfs_free_path(path);
198
Yan, Zhengc2167752009-11-12 09:34:21 +0000199 /*
200 * we're an inline extent, so nobody can
201 * extend the file past i_size without locking
202 * a page we already have locked.
203 *
204 * We must do any isize and inode updates
205 * before we unlock the pages. Otherwise we
206 * could end up racing with unlink.
207 */
Chris Masonc8b97812008-10-29 14:49:59 -0400208 BTRFS_I(inode)->disk_i_size = inode->i_size;
209 btrfs_update_inode(trans, root, inode);
Yan, Zhengc2167752009-11-12 09:34:21 +0000210
Chris Masonc8b97812008-10-29 14:49:59 -0400211 return 0;
212fail:
213 btrfs_free_path(path);
214 return err;
215}
216
217
218/*
219 * conditionally insert an inline extent into the file. This
220 * does the checks required to make sure the data is small enough
221 * to fit as an inline extent.
222 */
Chris Mason7f366cf2009-03-12 20:12:45 -0400223static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
Chris Masonc8b97812008-10-29 14:49:59 -0400224 struct btrfs_root *root,
225 struct inode *inode, u64 start, u64 end,
Li Zefanfe3f5662011-03-28 08:30:38 +0000226 size_t compressed_size, int compress_type,
Chris Masonc8b97812008-10-29 14:49:59 -0400227 struct page **compressed_pages)
228{
229 u64 isize = i_size_read(inode);
230 u64 actual_end = min(end + 1, isize);
231 u64 inline_len = actual_end - start;
232 u64 aligned_end = (end + root->sectorsize - 1) &
233 ~((u64)root->sectorsize - 1);
234 u64 hint_byte;
235 u64 data_len = inline_len;
236 int ret;
237
238 if (compressed_size)
239 data_len = compressed_size;
240
241 if (start > 0 ||
Chris Mason70b99e62008-10-31 12:46:39 -0400242 actual_end >= PAGE_CACHE_SIZE ||
Chris Masonc8b97812008-10-29 14:49:59 -0400243 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
244 (!compressed_size &&
245 (actual_end & (root->sectorsize - 1)) == 0) ||
246 end + 1 < isize ||
247 data_len > root->fs_info->max_inline) {
248 return 1;
249 }
250
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000251 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
Chris Masona1ed8352009-09-11 12:27:37 -0400252 &hint_byte, 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400253 BUG_ON(ret);
254
255 if (isize > actual_end)
256 inline_len = min_t(u64, isize, actual_end);
257 ret = insert_inline_extent(trans, root, inode, start,
258 inline_len, compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000259 compress_type, compressed_pages);
Chris Masonc8b97812008-10-29 14:49:59 -0400260 BUG_ON(ret);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400261 btrfs_delalloc_release_metadata(inode, end + 1 - start);
Chris Masona1ed8352009-09-11 12:27:37 -0400262 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
Chris Masonc8b97812008-10-29 14:49:59 -0400263 return 0;
264}
265
Chris Mason771ed682008-11-06 22:02:51 -0500266struct async_extent {
267 u64 start;
268 u64 ram_size;
269 u64 compressed_size;
270 struct page **pages;
271 unsigned long nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800272 int compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500273 struct list_head list;
274};
275
276struct async_cow {
277 struct inode *inode;
278 struct btrfs_root *root;
279 struct page *locked_page;
280 u64 start;
281 u64 end;
282 struct list_head extents;
283 struct btrfs_work work;
284};
285
286static noinline int add_async_extent(struct async_cow *cow,
287 u64 start, u64 ram_size,
288 u64 compressed_size,
289 struct page **pages,
Li Zefan261507a02010-12-17 14:21:50 +0800290 unsigned long nr_pages,
291 int compress_type)
Chris Mason771ed682008-11-06 22:02:51 -0500292{
293 struct async_extent *async_extent;
294
295 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
Yoshinori Sanodac97e52011-02-15 12:01:42 +0000296 BUG_ON(!async_extent);
Chris Mason771ed682008-11-06 22:02:51 -0500297 async_extent->start = start;
298 async_extent->ram_size = ram_size;
299 async_extent->compressed_size = compressed_size;
300 async_extent->pages = pages;
301 async_extent->nr_pages = nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800302 async_extent->compress_type = compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500303 list_add_tail(&async_extent->list, &cow->extents);
304 return 0;
305}
306
Chris Masonc8b97812008-10-29 14:49:59 -0400307/*
Chris Mason771ed682008-11-06 22:02:51 -0500308 * we create compressed extents in two phases. The first
309 * phase compresses a range of pages that have already been
310 * locked (both pages and state bits are locked).
Chris Masonc8b97812008-10-29 14:49:59 -0400311 *
Chris Mason771ed682008-11-06 22:02:51 -0500312 * This is done inside an ordered work queue, and the compression
313 * is spread across many cpus. The actual IO submission is step
314 * two, and the ordered work queue takes care of making sure that
315 * happens in the same order things were put onto the queue by
316 * writepages and friends.
Chris Masonc8b97812008-10-29 14:49:59 -0400317 *
Chris Mason771ed682008-11-06 22:02:51 -0500318 * If this code finds it can't get good compression, it puts an
319 * entry onto the work queue to write the uncompressed bytes. This
320 * makes sure that both compressed inodes and uncompressed inodes
321 * are written in the same order that pdflush sent them down.
Chris Masond352ac62008-09-29 15:18:18 -0400322 */
Chris Mason771ed682008-11-06 22:02:51 -0500323static noinline int compress_file_range(struct inode *inode,
324 struct page *locked_page,
325 u64 start, u64 end,
326 struct async_cow *async_cow,
327 int *num_added)
Chris Masonb888db22007-08-27 16:49:44 -0400328{
329 struct btrfs_root *root = BTRFS_I(inode)->root;
330 struct btrfs_trans_handle *trans;
Chris Masondb945352007-10-15 16:15:53 -0400331 u64 num_bytes;
Chris Masondb945352007-10-15 16:15:53 -0400332 u64 blocksize = root->sectorsize;
Chris Masonc8b97812008-10-29 14:49:59 -0400333 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500334 u64 isize = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400335 int ret = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400336 struct page **pages = NULL;
337 unsigned long nr_pages;
338 unsigned long nr_pages_ret = 0;
339 unsigned long total_compressed = 0;
340 unsigned long total_in = 0;
341 unsigned long max_compressed = 128 * 1024;
Chris Mason771ed682008-11-06 22:02:51 -0500342 unsigned long max_uncompressed = 128 * 1024;
Chris Masonc8b97812008-10-29 14:49:59 -0400343 int i;
344 int will_compress;
Li Zefan261507a02010-12-17 14:21:50 +0800345 int compress_type = root->fs_info->compress_type;
Chris Masonb888db22007-08-27 16:49:44 -0400346
Chris Mason4cb53002011-05-24 15:35:30 -0400347 /* if this is a small write inside eof, kick off a defragbot */
348 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
349 btrfs_add_inode_defrag(NULL, inode);
350
Chris Mason42dc7ba2008-12-15 11:44:56 -0500351 actual_end = min_t(u64, isize, end + 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400352again:
353 will_compress = 0;
354 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
355 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
356
Chris Masonf03d9302009-02-04 09:31:06 -0500357 /*
358 * we don't want to send crud past the end of i_size through
359 * compression, that's just a waste of CPU time. So, if the
360 * end of the file is before the start of our current
361 * requested range of bytes, we bail out to the uncompressed
362 * cleanup code that can deal with all of this.
363 *
364 * It isn't really the fastest way to fix things, but this is a
365 * very uncommon corner.
366 */
367 if (actual_end <= start)
368 goto cleanup_and_bail_uncompressed;
369
Chris Masonc8b97812008-10-29 14:49:59 -0400370 total_compressed = actual_end - start;
371
372 /* we want to make sure that amount of ram required to uncompress
373 * an extent is reasonable, so we limit the total size in ram
Chris Mason771ed682008-11-06 22:02:51 -0500374 * of a compressed extent to 128k. This is a crucial number
375 * because it also controls how easily we can spread reads across
376 * cpus for decompression.
377 *
378 * We also want to make sure the amount of IO required to do
379 * a random read is reasonably small, so we limit the size of
380 * a compressed extent to 128k.
Chris Masonc8b97812008-10-29 14:49:59 -0400381 */
382 total_compressed = min(total_compressed, max_uncompressed);
Chris Masondb945352007-10-15 16:15:53 -0400383 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
Chris Masonbe20aa92007-12-17 20:14:01 -0500384 num_bytes = max(blocksize, num_bytes);
Chris Masonc8b97812008-10-29 14:49:59 -0400385 total_in = 0;
386 ret = 0;
Chris Masondb945352007-10-15 16:15:53 -0400387
Chris Mason771ed682008-11-06 22:02:51 -0500388 /*
389 * we do compression for mount -o compress and when the
390 * inode has not been flagged as nocompress. This flag can
391 * change at any time if we discover bad compression ratios.
Chris Masonc8b97812008-10-29 14:49:59 -0400392 */
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200393 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
Chris Mason1e701a32010-03-11 09:42:04 -0500394 (btrfs_test_opt(root, COMPRESS) ||
Liu Bo75e7cb72011-03-22 10:12:20 +0000395 (BTRFS_I(inode)->force_compress) ||
396 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
Chris Masonc8b97812008-10-29 14:49:59 -0400397 WARN_ON(pages);
Chris Masoncfbc2462008-10-30 13:22:14 -0400398 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
Li Zefan560f7d72011-09-08 10:22:01 +0800399 if (!pages) {
400 /* just bail out to the uncompressed code */
401 goto cont;
402 }
Chris Mason179e29e2007-11-01 11:28:41 -0400403
Li Zefan261507a02010-12-17 14:21:50 +0800404 if (BTRFS_I(inode)->force_compress)
405 compress_type = BTRFS_I(inode)->force_compress;
406
407 ret = btrfs_compress_pages(compress_type,
408 inode->i_mapping, start,
409 total_compressed, pages,
410 nr_pages, &nr_pages_ret,
411 &total_in,
412 &total_compressed,
413 max_compressed);
Chris Masonc8b97812008-10-29 14:49:59 -0400414
415 if (!ret) {
416 unsigned long offset = total_compressed &
417 (PAGE_CACHE_SIZE - 1);
418 struct page *page = pages[nr_pages_ret - 1];
419 char *kaddr;
420
421 /* zero the tail end of the last page, we might be
422 * sending it down to disk
423 */
424 if (offset) {
425 kaddr = kmap_atomic(page, KM_USER0);
426 memset(kaddr + offset, 0,
427 PAGE_CACHE_SIZE - offset);
428 kunmap_atomic(kaddr, KM_USER0);
429 }
430 will_compress = 1;
431 }
432 }
Li Zefan560f7d72011-09-08 10:22:01 +0800433cont:
Chris Masonc8b97812008-10-29 14:49:59 -0400434 if (start == 0) {
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400435 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +0000436 BUG_ON(IS_ERR(trans));
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400437 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Mason771ed682008-11-06 22:02:51 -0500438
Chris Masonc8b97812008-10-29 14:49:59 -0400439 /* lets try to make an inline extent */
Chris Mason771ed682008-11-06 22:02:51 -0500440 if (ret || total_in < (actual_end - start)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400441 /* we didn't compress the entire range, try
Chris Mason771ed682008-11-06 22:02:51 -0500442 * to make an uncompressed inline extent.
Chris Masonc8b97812008-10-29 14:49:59 -0400443 */
444 ret = cow_file_range_inline(trans, root, inode,
Li Zefanfe3f5662011-03-28 08:30:38 +0000445 start, end, 0, 0, NULL);
Chris Masonc8b97812008-10-29 14:49:59 -0400446 } else {
Chris Mason771ed682008-11-06 22:02:51 -0500447 /* try making a compressed inline extent */
Chris Masonc8b97812008-10-29 14:49:59 -0400448 ret = cow_file_range_inline(trans, root, inode,
449 start, end,
Li Zefanfe3f5662011-03-28 08:30:38 +0000450 total_compressed,
451 compress_type, pages);
Chris Masonc8b97812008-10-29 14:49:59 -0400452 }
453 if (ret == 0) {
Chris Mason771ed682008-11-06 22:02:51 -0500454 /*
455 * inline extent creation worked, we don't need
456 * to create any more async work items. Unlock
457 * and free up our temp pages.
458 */
Chris Masonc8b97812008-10-29 14:49:59 -0400459 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400460 &BTRFS_I(inode)->io_tree,
461 start, end, NULL,
462 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
Chris Masona3429ab2009-10-08 12:30:20 -0400463 EXTENT_CLEAR_DELALLOC |
Chris Masona791e352009-10-08 11:27:10 -0400464 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000465
466 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400467 goto free_pages_out;
468 }
Yan, Zhengc2167752009-11-12 09:34:21 +0000469 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400470 }
471
472 if (will_compress) {
473 /*
474 * we aren't doing an inline extent round the compressed size
475 * up to a block size boundary so the allocator does sane
476 * things
477 */
478 total_compressed = (total_compressed + blocksize - 1) &
479 ~(blocksize - 1);
480
481 /*
482 * one last check to make sure the compression is really a
483 * win, compare the page count read with the blocks on disk
484 */
485 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
486 ~(PAGE_CACHE_SIZE - 1);
487 if (total_compressed >= total_in) {
488 will_compress = 0;
489 } else {
Chris Masonc8b97812008-10-29 14:49:59 -0400490 num_bytes = total_in;
491 }
492 }
493 if (!will_compress && pages) {
494 /*
495 * the compression code ran but failed to make things smaller,
496 * free any pages it allocated and our page pointer array
497 */
498 for (i = 0; i < nr_pages_ret; i++) {
Chris Mason70b99e62008-10-31 12:46:39 -0400499 WARN_ON(pages[i]->mapping);
Chris Masonc8b97812008-10-29 14:49:59 -0400500 page_cache_release(pages[i]);
501 }
502 kfree(pages);
503 pages = NULL;
504 total_compressed = 0;
505 nr_pages_ret = 0;
506
507 /* flag the file so we don't compress in the future */
Chris Mason1e701a32010-03-11 09:42:04 -0500508 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
509 !(BTRFS_I(inode)->force_compress)) {
Chris Masona555f812010-01-28 16:18:15 -0500510 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
Chris Mason1e701a32010-03-11 09:42:04 -0500511 }
Chris Masonc8b97812008-10-29 14:49:59 -0400512 }
Chris Mason771ed682008-11-06 22:02:51 -0500513 if (will_compress) {
514 *num_added += 1;
515
516 /* the async work queues will take care of doing actual
517 * allocation on disk for these compressed pages,
518 * and will submit them to the elevator.
519 */
520 add_async_extent(async_cow, start, num_bytes,
Li Zefan261507a02010-12-17 14:21:50 +0800521 total_compressed, pages, nr_pages_ret,
522 compress_type);
Chris Mason771ed682008-11-06 22:02:51 -0500523
Yan, Zheng24ae6362010-12-06 07:02:36 +0000524 if (start + num_bytes < end) {
Chris Mason771ed682008-11-06 22:02:51 -0500525 start += num_bytes;
526 pages = NULL;
527 cond_resched();
528 goto again;
529 }
530 } else {
Chris Masonf03d9302009-02-04 09:31:06 -0500531cleanup_and_bail_uncompressed:
Chris Mason771ed682008-11-06 22:02:51 -0500532 /*
533 * No compression, but we still need to write the pages in
534 * the file we've been given so far. redirty the locked
535 * page if it corresponds to our extent and set things up
536 * for the async work queue to run cow_file_range to do
537 * the normal delalloc dance
538 */
539 if (page_offset(locked_page) >= start &&
540 page_offset(locked_page) <= end) {
541 __set_page_dirty_nobuffers(locked_page);
542 /* unlocked later on in the async handlers */
543 }
Li Zefan261507a02010-12-17 14:21:50 +0800544 add_async_extent(async_cow, start, end - start + 1,
545 0, NULL, 0, BTRFS_COMPRESS_NONE);
Chris Mason771ed682008-11-06 22:02:51 -0500546 *num_added += 1;
547 }
548
549out:
550 return 0;
551
552free_pages_out:
553 for (i = 0; i < nr_pages_ret; i++) {
554 WARN_ON(pages[i]->mapping);
555 page_cache_release(pages[i]);
556 }
Chris Masond3977122009-01-05 21:25:51 -0500557 kfree(pages);
Chris Mason771ed682008-11-06 22:02:51 -0500558
559 goto out;
560}
561
562/*
563 * phase two of compressed writeback. This is the ordered portion
564 * of the code, which only gets called in the order the work was
565 * queued. We walk all the async extents created by compress_file_range
566 * and send them down to the disk.
567 */
568static noinline int submit_compressed_extents(struct inode *inode,
569 struct async_cow *async_cow)
570{
571 struct async_extent *async_extent;
572 u64 alloc_hint = 0;
573 struct btrfs_trans_handle *trans;
574 struct btrfs_key ins;
575 struct extent_map *em;
576 struct btrfs_root *root = BTRFS_I(inode)->root;
577 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
578 struct extent_io_tree *io_tree;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500579 int ret = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500580
581 if (list_empty(&async_cow->extents))
582 return 0;
583
Chris Mason771ed682008-11-06 22:02:51 -0500584
Chris Masond3977122009-01-05 21:25:51 -0500585 while (!list_empty(&async_cow->extents)) {
Chris Mason771ed682008-11-06 22:02:51 -0500586 async_extent = list_entry(async_cow->extents.next,
587 struct async_extent, list);
588 list_del(&async_extent->list);
589
590 io_tree = &BTRFS_I(inode)->io_tree;
591
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500592retry:
Chris Mason771ed682008-11-06 22:02:51 -0500593 /* did the compression code fall back to uncompressed IO? */
594 if (!async_extent->pages) {
595 int page_started = 0;
596 unsigned long nr_written = 0;
597
598 lock_extent(io_tree, async_extent->start,
Josef Bacik2ac55d42010-02-03 19:33:23 +0000599 async_extent->start +
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100600 async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500601
602 /* allocate blocks */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500603 ret = cow_file_range(inode, async_cow->locked_page,
604 async_extent->start,
605 async_extent->start +
606 async_extent->ram_size - 1,
607 &page_started, &nr_written, 0);
Chris Mason771ed682008-11-06 22:02:51 -0500608
609 /*
610 * if page_started, cow_file_range inserted an
611 * inline extent and took care of all the unlocking
612 * and IO for us. Otherwise, we need to submit
613 * all those pages down to the drive.
614 */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500615 if (!page_started && !ret)
Chris Mason771ed682008-11-06 22:02:51 -0500616 extent_write_locked_range(io_tree,
617 inode, async_extent->start,
Chris Masond3977122009-01-05 21:25:51 -0500618 async_extent->start +
Chris Mason771ed682008-11-06 22:02:51 -0500619 async_extent->ram_size - 1,
620 btrfs_get_extent,
621 WB_SYNC_ALL);
622 kfree(async_extent);
623 cond_resched();
624 continue;
625 }
626
627 lock_extent(io_tree, async_extent->start,
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100628 async_extent->start + async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500629
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400630 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +0000631 BUG_ON(IS_ERR(trans));
Josef Bacik74b21072011-04-13 12:02:53 -0400632 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Mason771ed682008-11-06 22:02:51 -0500633 ret = btrfs_reserve_extent(trans, root,
634 async_extent->compressed_size,
635 async_extent->compressed_size,
636 0, alloc_hint,
637 (u64)-1, &ins, 1);
Yan, Zhengc2167752009-11-12 09:34:21 +0000638 btrfs_end_transaction(trans, root);
639
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500640 if (ret) {
641 int i;
642 for (i = 0; i < async_extent->nr_pages; i++) {
643 WARN_ON(async_extent->pages[i]->mapping);
644 page_cache_release(async_extent->pages[i]);
645 }
646 kfree(async_extent->pages);
647 async_extent->nr_pages = 0;
648 async_extent->pages = NULL;
649 unlock_extent(io_tree, async_extent->start,
650 async_extent->start +
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100651 async_extent->ram_size - 1);
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500652 goto retry;
653 }
654
Yan, Zhengc2167752009-11-12 09:34:21 +0000655 /*
656 * here we're doing allocation and writeback of the
657 * compressed pages
658 */
659 btrfs_drop_extent_cache(inode, async_extent->start,
660 async_extent->start +
661 async_extent->ram_size - 1, 0);
662
David Sterba172ddd62011-04-21 00:48:27 +0200663 em = alloc_extent_map();
Tsutomu Itohc26a9202011-02-14 00:45:29 +0000664 BUG_ON(!em);
Chris Mason771ed682008-11-06 22:02:51 -0500665 em->start = async_extent->start;
666 em->len = async_extent->ram_size;
Chris Mason445a6942008-11-10 11:53:33 -0500667 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500668
669 em->block_start = ins.objectid;
670 em->block_len = ins.offset;
671 em->bdev = root->fs_info->fs_devices->latest_bdev;
Li Zefan261507a02010-12-17 14:21:50 +0800672 em->compress_type = async_extent->compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500673 set_bit(EXTENT_FLAG_PINNED, &em->flags);
674 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
675
Chris Masond3977122009-01-05 21:25:51 -0500676 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400677 write_lock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500678 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -0400679 write_unlock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500680 if (ret != -EEXIST) {
681 free_extent_map(em);
682 break;
683 }
684 btrfs_drop_extent_cache(inode, async_extent->start,
685 async_extent->start +
686 async_extent->ram_size - 1, 0);
687 }
688
Li Zefan261507a02010-12-17 14:21:50 +0800689 ret = btrfs_add_ordered_extent_compress(inode,
690 async_extent->start,
691 ins.objectid,
692 async_extent->ram_size,
693 ins.offset,
694 BTRFS_ORDERED_COMPRESSED,
695 async_extent->compress_type);
Chris Mason771ed682008-11-06 22:02:51 -0500696 BUG_ON(ret);
697
Chris Mason771ed682008-11-06 22:02:51 -0500698 /*
699 * clear dirty, set writeback and unlock the pages.
700 */
701 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400702 &BTRFS_I(inode)->io_tree,
703 async_extent->start,
704 async_extent->start +
705 async_extent->ram_size - 1,
706 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
707 EXTENT_CLEAR_UNLOCK |
Chris Masona3429ab2009-10-08 12:30:20 -0400708 EXTENT_CLEAR_DELALLOC |
Chris Masona791e352009-10-08 11:27:10 -0400709 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
Chris Mason771ed682008-11-06 22:02:51 -0500710
711 ret = btrfs_submit_compressed_write(inode,
Chris Masond3977122009-01-05 21:25:51 -0500712 async_extent->start,
713 async_extent->ram_size,
714 ins.objectid,
715 ins.offset, async_extent->pages,
716 async_extent->nr_pages);
Chris Mason771ed682008-11-06 22:02:51 -0500717
718 BUG_ON(ret);
Chris Mason771ed682008-11-06 22:02:51 -0500719 alloc_hint = ins.objectid + ins.offset;
720 kfree(async_extent);
721 cond_resched();
722 }
723
Chris Mason771ed682008-11-06 22:02:51 -0500724 return 0;
725}
726
Josef Bacik4b46fce2010-05-23 11:00:55 -0400727static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
728 u64 num_bytes)
729{
730 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
731 struct extent_map *em;
732 u64 alloc_hint = 0;
733
734 read_lock(&em_tree->lock);
735 em = search_extent_mapping(em_tree, start, num_bytes);
736 if (em) {
737 /*
738 * if block start isn't an actual block number then find the
739 * first block in this inode and use that as a hint. If that
740 * block is also bogus then just don't worry about it.
741 */
742 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
743 free_extent_map(em);
744 em = search_extent_mapping(em_tree, 0, 0);
745 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
746 alloc_hint = em->block_start;
747 if (em)
748 free_extent_map(em);
749 } else {
750 alloc_hint = em->block_start;
751 free_extent_map(em);
752 }
753 }
754 read_unlock(&em_tree->lock);
755
756 return alloc_hint;
757}
758
Chris Mason771ed682008-11-06 22:02:51 -0500759/*
760 * when extent_io.c finds a delayed allocation range in the file,
761 * the call backs end up in this code. The basic idea is to
762 * allocate extents on disk for the range, and create ordered data structs
763 * in ram to track those extents.
764 *
765 * locked_page is the page that writepage had locked already. We use
766 * it to make sure we don't do extra locks or unlocks.
767 *
768 * *page_started is set to one if we unlock locked_page and do everything
769 * required to start IO on it. It may be clean and already done with
770 * IO when we return.
771 */
772static noinline int cow_file_range(struct inode *inode,
773 struct page *locked_page,
774 u64 start, u64 end, int *page_started,
775 unsigned long *nr_written,
776 int unlock)
777{
778 struct btrfs_root *root = BTRFS_I(inode)->root;
779 struct btrfs_trans_handle *trans;
780 u64 alloc_hint = 0;
781 u64 num_bytes;
782 unsigned long ram_size;
783 u64 disk_num_bytes;
784 u64 cur_alloc_size;
785 u64 blocksize = root->sectorsize;
Chris Mason771ed682008-11-06 22:02:51 -0500786 struct btrfs_key ins;
787 struct extent_map *em;
788 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
789 int ret = 0;
790
Chris Mason2cf85722011-07-26 15:35:09 -0400791 BUG_ON(btrfs_is_free_space_inode(root, inode));
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400792 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +0000793 BUG_ON(IS_ERR(trans));
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400794 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Mason771ed682008-11-06 22:02:51 -0500795
Chris Mason771ed682008-11-06 22:02:51 -0500796 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
797 num_bytes = max(blocksize, num_bytes);
798 disk_num_bytes = num_bytes;
799 ret = 0;
800
Chris Mason4cb53002011-05-24 15:35:30 -0400801 /* if this is a small write inside eof, kick off defrag */
802 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
803 btrfs_add_inode_defrag(trans, inode);
804
Chris Mason771ed682008-11-06 22:02:51 -0500805 if (start == 0) {
806 /* lets try to make an inline extent */
807 ret = cow_file_range_inline(trans, root, inode,
Li Zefanfe3f5662011-03-28 08:30:38 +0000808 start, end, 0, 0, NULL);
Chris Mason771ed682008-11-06 22:02:51 -0500809 if (ret == 0) {
810 extent_clear_unlock_delalloc(inode,
Chris Masona791e352009-10-08 11:27:10 -0400811 &BTRFS_I(inode)->io_tree,
812 start, end, NULL,
813 EXTENT_CLEAR_UNLOCK_PAGE |
814 EXTENT_CLEAR_UNLOCK |
815 EXTENT_CLEAR_DELALLOC |
816 EXTENT_CLEAR_DIRTY |
817 EXTENT_SET_WRITEBACK |
818 EXTENT_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000819
Chris Mason771ed682008-11-06 22:02:51 -0500820 *nr_written = *nr_written +
821 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
822 *page_started = 1;
823 ret = 0;
824 goto out;
825 }
826 }
Chris Masonc8b97812008-10-29 14:49:59 -0400827
828 BUG_ON(disk_num_bytes >
David Sterba6c417612011-04-13 15:41:04 +0200829 btrfs_super_total_bytes(root->fs_info->super_copy));
Chris Masonc8b97812008-10-29 14:49:59 -0400830
Josef Bacik4b46fce2010-05-23 11:00:55 -0400831 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400832 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
Chris Mason3b951512008-04-17 11:29:12 -0400833
Chris Masond3977122009-01-05 21:25:51 -0500834 while (disk_num_bytes > 0) {
Chris Masona791e352009-10-08 11:27:10 -0400835 unsigned long op;
836
Josef Bacik287a0ab2010-03-19 18:07:23 +0000837 cur_alloc_size = disk_num_bytes;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400838 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
Chris Mason771ed682008-11-06 22:02:51 -0500839 root->sectorsize, 0, alloc_hint,
Chris Masone6dcd2d2008-07-17 12:53:50 -0400840 (u64)-1, &ins, 1);
Chris Masond3977122009-01-05 21:25:51 -0500841 BUG_ON(ret);
842
David Sterba172ddd62011-04-21 00:48:27 +0200843 em = alloc_extent_map();
Tsutomu Itohc26a9202011-02-14 00:45:29 +0000844 BUG_ON(!em);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400845 em->start = start;
Chris Mason445a6942008-11-10 11:53:33 -0500846 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500847 ram_size = ins.offset;
848 em->len = ins.offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400849
Chris Masone6dcd2d2008-07-17 12:53:50 -0400850 em->block_start = ins.objectid;
Chris Masonc8b97812008-10-29 14:49:59 -0400851 em->block_len = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400852 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason7f3c74f2008-07-18 12:01:11 -0400853 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -0400854
Chris Masond3977122009-01-05 21:25:51 -0500855 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400856 write_lock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400857 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -0400858 write_unlock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400859 if (ret != -EEXIST) {
860 free_extent_map(em);
861 break;
862 }
863 btrfs_drop_extent_cache(inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -0400864 start + ram_size - 1, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400865 }
866
Chris Mason98d20f62008-04-14 09:46:10 -0400867 cur_alloc_size = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400868 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
Chris Mason771ed682008-11-06 22:02:51 -0500869 ram_size, cur_alloc_size, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400870 BUG_ON(ret);
Chris Masonc8b97812008-10-29 14:49:59 -0400871
Yan Zheng17d217f2008-12-12 10:03:38 -0500872 if (root->root_key.objectid ==
873 BTRFS_DATA_RELOC_TREE_OBJECTID) {
874 ret = btrfs_reloc_clone_csums(inode, start,
875 cur_alloc_size);
876 BUG_ON(ret);
877 }
878
Chris Masond3977122009-01-05 21:25:51 -0500879 if (disk_num_bytes < cur_alloc_size)
Chris Mason3b951512008-04-17 11:29:12 -0400880 break;
Chris Masond3977122009-01-05 21:25:51 -0500881
Chris Masonc8b97812008-10-29 14:49:59 -0400882 /* we're not doing compressed IO, don't unlock the first
883 * page (which the caller expects to stay locked), don't
884 * clear any dirty bits and don't set any writeback bits
Chris Mason8b62b722009-09-02 16:53:46 -0400885 *
886 * Do set the Private2 bit so we know this page was properly
887 * setup for writepage
Chris Masonc8b97812008-10-29 14:49:59 -0400888 */
Chris Masona791e352009-10-08 11:27:10 -0400889 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
890 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
891 EXTENT_SET_PRIVATE2;
892
Chris Masonc8b97812008-10-29 14:49:59 -0400893 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
894 start, start + ram_size - 1,
Chris Masona791e352009-10-08 11:27:10 -0400895 locked_page, op);
Chris Masonc8b97812008-10-29 14:49:59 -0400896 disk_num_bytes -= cur_alloc_size;
Chris Masonc59f8952007-12-17 20:14:04 -0500897 num_bytes -= cur_alloc_size;
898 alloc_hint = ins.objectid + ins.offset;
899 start += cur_alloc_size;
Chris Masonb888db22007-08-27 16:49:44 -0400900 }
Chris Masonb888db22007-08-27 16:49:44 -0400901out:
Chris Mason771ed682008-11-06 22:02:51 -0500902 ret = 0;
Chris Masonb888db22007-08-27 16:49:44 -0400903 btrfs_end_transaction(trans, root);
Chris Masonc8b97812008-10-29 14:49:59 -0400904
Chris Masonbe20aa92007-12-17 20:14:01 -0500905 return ret;
Chris Mason771ed682008-11-06 22:02:51 -0500906}
Chris Masonc8b97812008-10-29 14:49:59 -0400907
Chris Mason771ed682008-11-06 22:02:51 -0500908/*
909 * work queue call back to started compression on a file and pages
910 */
911static noinline void async_cow_start(struct btrfs_work *work)
912{
913 struct async_cow *async_cow;
914 int num_added = 0;
915 async_cow = container_of(work, struct async_cow, work);
916
917 compress_file_range(async_cow->inode, async_cow->locked_page,
918 async_cow->start, async_cow->end, async_cow,
919 &num_added);
920 if (num_added == 0)
921 async_cow->inode = NULL;
922}
923
924/*
925 * work queue call back to submit previously compressed pages
926 */
927static noinline void async_cow_submit(struct btrfs_work *work)
928{
929 struct async_cow *async_cow;
930 struct btrfs_root *root;
931 unsigned long nr_pages;
932
933 async_cow = container_of(work, struct async_cow, work);
934
935 root = async_cow->root;
936 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
937 PAGE_CACHE_SHIFT;
938
939 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
940
941 if (atomic_read(&root->fs_info->async_delalloc_pages) <
942 5 * 1042 * 1024 &&
943 waitqueue_active(&root->fs_info->async_submit_wait))
944 wake_up(&root->fs_info->async_submit_wait);
945
Chris Masond3977122009-01-05 21:25:51 -0500946 if (async_cow->inode)
Chris Mason771ed682008-11-06 22:02:51 -0500947 submit_compressed_extents(async_cow->inode, async_cow);
Chris Mason771ed682008-11-06 22:02:51 -0500948}
Chris Masonc8b97812008-10-29 14:49:59 -0400949
Chris Mason771ed682008-11-06 22:02:51 -0500950static noinline void async_cow_free(struct btrfs_work *work)
951{
952 struct async_cow *async_cow;
953 async_cow = container_of(work, struct async_cow, work);
954 kfree(async_cow);
955}
956
957static int cow_file_range_async(struct inode *inode, struct page *locked_page,
958 u64 start, u64 end, int *page_started,
959 unsigned long *nr_written)
960{
961 struct async_cow *async_cow;
962 struct btrfs_root *root = BTRFS_I(inode)->root;
963 unsigned long nr_pages;
964 u64 cur_end;
965 int limit = 10 * 1024 * 1042;
966
Chris Masona3429ab2009-10-08 12:30:20 -0400967 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
968 1, 0, NULL, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -0500969 while (start < end) {
Chris Mason771ed682008-11-06 22:02:51 -0500970 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
Tsutomu Itoh8d413712011-04-25 19:43:52 -0400971 BUG_ON(!async_cow);
Chris Mason771ed682008-11-06 22:02:51 -0500972 async_cow->inode = inode;
973 async_cow->root = root;
974 async_cow->locked_page = locked_page;
975 async_cow->start = start;
976
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200977 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
Chris Mason771ed682008-11-06 22:02:51 -0500978 cur_end = end;
979 else
980 cur_end = min(end, start + 512 * 1024 - 1);
981
982 async_cow->end = cur_end;
983 INIT_LIST_HEAD(&async_cow->extents);
984
985 async_cow->work.func = async_cow_start;
986 async_cow->work.ordered_func = async_cow_submit;
987 async_cow->work.ordered_free = async_cow_free;
988 async_cow->work.flags = 0;
989
Chris Mason771ed682008-11-06 22:02:51 -0500990 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
991 PAGE_CACHE_SHIFT;
992 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
993
994 btrfs_queue_worker(&root->fs_info->delalloc_workers,
995 &async_cow->work);
996
997 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
998 wait_event(root->fs_info->async_submit_wait,
999 (atomic_read(&root->fs_info->async_delalloc_pages) <
1000 limit));
1001 }
1002
Chris Masond3977122009-01-05 21:25:51 -05001003 while (atomic_read(&root->fs_info->async_submit_draining) &&
Chris Mason771ed682008-11-06 22:02:51 -05001004 atomic_read(&root->fs_info->async_delalloc_pages)) {
1005 wait_event(root->fs_info->async_submit_wait,
1006 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1007 0));
1008 }
1009
1010 *nr_written += nr_pages;
1011 start = cur_end + 1;
1012 }
1013 *page_started = 1;
1014 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05001015}
1016
Chris Masond3977122009-01-05 21:25:51 -05001017static noinline int csum_exist_in_range(struct btrfs_root *root,
Yan Zheng17d217f2008-12-12 10:03:38 -05001018 u64 bytenr, u64 num_bytes)
1019{
1020 int ret;
1021 struct btrfs_ordered_sum *sums;
1022 LIST_HEAD(list);
1023
Yan Zheng07d400a2009-01-06 11:42:00 -05001024 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
Arne Jansena2de7332011-03-08 14:14:00 +01001025 bytenr + num_bytes - 1, &list, 0);
Yan Zheng17d217f2008-12-12 10:03:38 -05001026 if (ret == 0 && list_empty(&list))
1027 return 0;
1028
1029 while (!list_empty(&list)) {
1030 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1031 list_del(&sums->list);
1032 kfree(sums);
1033 }
1034 return 1;
1035}
1036
Chris Masond352ac62008-09-29 15:18:18 -04001037/*
1038 * when nowcow writeback call back. This checks for snapshots or COW copies
1039 * of the extents that exist in the file, and COWs the file as required.
1040 *
1041 * If no cow copies or snapshots exist, we write directly to the existing
1042 * blocks on disk
1043 */
Chris Mason7f366cf2009-03-12 20:12:45 -04001044static noinline int run_delalloc_nocow(struct inode *inode,
1045 struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001046 u64 start, u64 end, int *page_started, int force,
1047 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001048{
Chris Masonbe20aa92007-12-17 20:14:01 -05001049 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001050 struct btrfs_trans_handle *trans;
Chris Masonbe20aa92007-12-17 20:14:01 -05001051 struct extent_buffer *leaf;
Chris Masonbe20aa92007-12-17 20:14:01 -05001052 struct btrfs_path *path;
Yan Zheng80ff3852008-10-30 14:20:02 -04001053 struct btrfs_file_extent_item *fi;
Chris Masonbe20aa92007-12-17 20:14:01 -05001054 struct btrfs_key found_key;
Yan Zheng80ff3852008-10-30 14:20:02 -04001055 u64 cow_start;
1056 u64 cur_offset;
1057 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001058 u64 extent_offset;
Yan Zheng80ff3852008-10-30 14:20:02 -04001059 u64 disk_bytenr;
1060 u64 num_bytes;
1061 int extent_type;
1062 int ret;
Yan Zhengd899e052008-10-30 14:25:28 -04001063 int type;
Yan Zheng80ff3852008-10-30 14:20:02 -04001064 int nocow;
1065 int check_prev = 1;
Li Zefan82d59022011-04-20 10:33:24 +08001066 bool nolock;
Li Zefan33345d012011-04-20 10:31:50 +08001067 u64 ino = btrfs_ino(inode);
Chris Masonbe20aa92007-12-17 20:14:01 -05001068
1069 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07001070 if (!path)
1071 return -ENOMEM;
Li Zefan82d59022011-04-20 10:33:24 +08001072
Chris Mason2cf85722011-07-26 15:35:09 -04001073 nolock = btrfs_is_free_space_inode(root, inode);
Li Zefan82d59022011-04-20 10:33:24 +08001074
1075 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001076 trans = btrfs_join_transaction_nolock(root);
Li Zefan82d59022011-04-20 10:33:24 +08001077 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001078 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04001079
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001080 BUG_ON(IS_ERR(trans));
Josef Bacik74b21072011-04-13 12:02:53 -04001081 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Masonbe20aa92007-12-17 20:14:01 -05001082
Yan Zheng80ff3852008-10-30 14:20:02 -04001083 cow_start = (u64)-1;
1084 cur_offset = start;
1085 while (1) {
Li Zefan33345d012011-04-20 10:31:50 +08001086 ret = btrfs_lookup_file_extent(trans, root, path, ino,
Yan Zheng80ff3852008-10-30 14:20:02 -04001087 cur_offset, 0);
1088 BUG_ON(ret < 0);
1089 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1090 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key,
1092 path->slots[0] - 1);
Li Zefan33345d012011-04-20 10:31:50 +08001093 if (found_key.objectid == ino &&
Yan Zheng80ff3852008-10-30 14:20:02 -04001094 found_key.type == BTRFS_EXTENT_DATA_KEY)
1095 path->slots[0]--;
1096 }
1097 check_prev = 0;
1098next_slot:
1099 leaf = path->nodes[0];
1100 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1101 ret = btrfs_next_leaf(root, path);
1102 if (ret < 0)
1103 BUG_ON(1);
1104 if (ret > 0)
1105 break;
1106 leaf = path->nodes[0];
1107 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001108
Yan Zheng80ff3852008-10-30 14:20:02 -04001109 nocow = 0;
1110 disk_bytenr = 0;
Yan Zheng17d217f2008-12-12 10:03:38 -05001111 num_bytes = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001112 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05001113
Li Zefan33345d012011-04-20 10:31:50 +08001114 if (found_key.objectid > ino ||
Yan Zheng80ff3852008-10-30 14:20:02 -04001115 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1116 found_key.offset > end)
1117 break;
Chris Masonbe20aa92007-12-17 20:14:01 -05001118
Yan Zheng80ff3852008-10-30 14:20:02 -04001119 if (found_key.offset > cur_offset) {
1120 extent_end = found_key.offset;
Chris Masone9061e22009-10-09 09:57:45 -04001121 extent_type = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001122 goto out_check;
1123 }
Chris Masonc31f8832008-01-08 15:46:31 -05001124
Yan Zheng80ff3852008-10-30 14:20:02 -04001125 fi = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_file_extent_item);
1127 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Masonbe20aa92007-12-17 20:14:01 -05001128
Yan Zhengd899e052008-10-30 14:25:28 -04001129 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1130 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001131 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001132 extent_offset = btrfs_file_extent_offset(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001133 extent_end = found_key.offset +
1134 btrfs_file_extent_num_bytes(leaf, fi);
1135 if (extent_end <= start) {
1136 path->slots[0]++;
1137 goto next_slot;
1138 }
Yan Zheng17d217f2008-12-12 10:03:38 -05001139 if (disk_bytenr == 0)
1140 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001141 if (btrfs_file_extent_compression(leaf, fi) ||
1142 btrfs_file_extent_encryption(leaf, fi) ||
1143 btrfs_file_extent_other_encoding(leaf, fi))
1144 goto out_check;
Yan Zhengd899e052008-10-30 14:25:28 -04001145 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1146 goto out_check;
Yan Zhengd2fb3432008-12-11 16:30:39 -05001147 if (btrfs_extent_readonly(root, disk_bytenr))
Yan Zheng80ff3852008-10-30 14:20:02 -04001148 goto out_check;
Li Zefan33345d012011-04-20 10:31:50 +08001149 if (btrfs_cross_ref_exist(trans, root, ino,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001150 found_key.offset -
1151 extent_offset, disk_bytenr))
Yan Zheng17d217f2008-12-12 10:03:38 -05001152 goto out_check;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001153 disk_bytenr += extent_offset;
Yan Zheng17d217f2008-12-12 10:03:38 -05001154 disk_bytenr += cur_offset - found_key.offset;
1155 num_bytes = min(end + 1, extent_end) - cur_offset;
1156 /*
1157 * force cow if csum exists in the range.
1158 * this ensure that csum for a given extent are
1159 * either valid or do not exist.
1160 */
1161 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1162 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001163 nocow = 1;
1164 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1165 extent_end = found_key.offset +
1166 btrfs_file_extent_inline_len(leaf, fi);
1167 extent_end = ALIGN(extent_end, root->sectorsize);
1168 } else {
1169 BUG_ON(1);
1170 }
1171out_check:
1172 if (extent_end <= start) {
1173 path->slots[0]++;
1174 goto next_slot;
1175 }
1176 if (!nocow) {
1177 if (cow_start == (u64)-1)
1178 cow_start = cur_offset;
1179 cur_offset = extent_end;
1180 if (cur_offset > end)
1181 break;
1182 path->slots[0]++;
1183 goto next_slot;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001184 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001185
David Sterbab3b4aa72011-04-21 01:20:15 +02001186 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001187 if (cow_start != (u64)-1) {
1188 ret = cow_file_range(inode, locked_page, cow_start,
Chris Mason771ed682008-11-06 22:02:51 -05001189 found_key.offset - 1, page_started,
1190 nr_written, 1);
Yan Zheng80ff3852008-10-30 14:20:02 -04001191 BUG_ON(ret);
1192 cow_start = (u64)-1;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001193 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001194
Yan Zhengd899e052008-10-30 14:25:28 -04001195 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1196 struct extent_map *em;
1197 struct extent_map_tree *em_tree;
1198 em_tree = &BTRFS_I(inode)->extent_tree;
David Sterba172ddd62011-04-21 00:48:27 +02001199 em = alloc_extent_map();
Tsutomu Itohc26a9202011-02-14 00:45:29 +00001200 BUG_ON(!em);
Yan Zhengd899e052008-10-30 14:25:28 -04001201 em->start = cur_offset;
Chris Mason445a6942008-11-10 11:53:33 -05001202 em->orig_start = em->start;
Yan Zhengd899e052008-10-30 14:25:28 -04001203 em->len = num_bytes;
1204 em->block_len = num_bytes;
1205 em->block_start = disk_bytenr;
1206 em->bdev = root->fs_info->fs_devices->latest_bdev;
1207 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1208 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04001209 write_lock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001210 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04001211 write_unlock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001212 if (ret != -EEXIST) {
1213 free_extent_map(em);
1214 break;
1215 }
1216 btrfs_drop_extent_cache(inode, em->start,
1217 em->start + em->len - 1, 0);
1218 }
1219 type = BTRFS_ORDERED_PREALLOC;
1220 } else {
1221 type = BTRFS_ORDERED_NOCOW;
1222 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001223
1224 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
Yan Zhengd899e052008-10-30 14:25:28 -04001225 num_bytes, num_bytes, type);
1226 BUG_ON(ret);
Chris Mason771ed682008-11-06 22:02:51 -05001227
Yan, Zhengefa56462010-05-16 10:49:59 -04001228 if (root->root_key.objectid ==
1229 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1230 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1231 num_bytes);
1232 BUG_ON(ret);
1233 }
1234
Yan Zhengd899e052008-10-30 14:25:28 -04001235 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
Chris Masona791e352009-10-08 11:27:10 -04001236 cur_offset, cur_offset + num_bytes - 1,
1237 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1238 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1239 EXTENT_SET_PRIVATE2);
Yan Zheng80ff3852008-10-30 14:20:02 -04001240 cur_offset = extent_end;
1241 if (cur_offset > end)
1242 break;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001243 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001244 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001245
1246 if (cur_offset <= end && cow_start == (u64)-1)
1247 cow_start = cur_offset;
1248 if (cow_start != (u64)-1) {
1249 ret = cow_file_range(inode, locked_page, cow_start, end,
Chris Mason771ed682008-11-06 22:02:51 -05001250 page_started, nr_written, 1);
Yan Zheng80ff3852008-10-30 14:20:02 -04001251 BUG_ON(ret);
1252 }
1253
Josef Bacik0cb59c92010-07-02 12:14:14 -04001254 if (nolock) {
1255 ret = btrfs_end_transaction_nolock(trans, root);
1256 BUG_ON(ret);
1257 } else {
1258 ret = btrfs_end_transaction(trans, root);
1259 BUG_ON(ret);
1260 }
Yan Zheng7ea394f2008-08-05 13:05:02 -04001261 btrfs_free_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001262 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05001263}
1264
Chris Masond352ac62008-09-29 15:18:18 -04001265/*
1266 * extent_io.c call back to do delayed allocation processing
1267 */
Chris Masonc8b97812008-10-29 14:49:59 -04001268static int run_delalloc_range(struct inode *inode, struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001269 u64 start, u64 end, int *page_started,
1270 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001271{
Chris Masonbe20aa92007-12-17 20:14:01 -05001272 int ret;
Chris Mason7f366cf2009-03-12 20:12:45 -04001273 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masona2135012008-06-25 16:01:30 -04001274
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001275 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
Chris Masonc8b97812008-10-29 14:49:59 -04001276 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001277 page_started, 1, nr_written);
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001278 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
Yan Zhengd899e052008-10-30 14:25:28 -04001279 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001280 page_started, 0, nr_written);
Chris Mason1e701a32010-03-11 09:42:04 -05001281 else if (!btrfs_test_opt(root, COMPRESS) &&
Liu Bo75e7cb72011-03-22 10:12:20 +00001282 !(BTRFS_I(inode)->force_compress) &&
1283 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
Chris Mason7f366cf2009-03-12 20:12:45 -04001284 ret = cow_file_range(inode, locked_page, start, end,
1285 page_started, nr_written, 1);
Chris Masonbe20aa92007-12-17 20:14:01 -05001286 else
Chris Mason771ed682008-11-06 22:02:51 -05001287 ret = cow_file_range_async(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001288 page_started, nr_written);
Chris Masonb888db22007-08-27 16:49:44 -04001289 return ret;
1290}
1291
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001292static void btrfs_split_extent_hook(struct inode *inode,
1293 struct extent_state *orig, u64 split)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001294{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001295 /* not delalloc, ignore it */
Josef Bacik9ed74f22009-09-11 16:12:44 -04001296 if (!(orig->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001297 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001298
Josef Bacik9e0baf62011-07-15 15:16:44 +00001299 spin_lock(&BTRFS_I(inode)->lock);
1300 BTRFS_I(inode)->outstanding_extents++;
1301 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001302}
1303
1304/*
1305 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1306 * extents so we can keep track of new extents that are just merged onto old
1307 * extents, such as when we are doing sequential writes, so we can properly
1308 * account for the metadata space we'll need.
1309 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001310static void btrfs_merge_extent_hook(struct inode *inode,
1311 struct extent_state *new,
1312 struct extent_state *other)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001313{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001314 /* not delalloc, ignore it */
1315 if (!(other->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001316 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001317
Josef Bacik9e0baf62011-07-15 15:16:44 +00001318 spin_lock(&BTRFS_I(inode)->lock);
1319 BTRFS_I(inode)->outstanding_extents--;
1320 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001321}
1322
Chris Masond352ac62008-09-29 15:18:18 -04001323/*
1324 * extent_io.c set_bit_hook, used to track delayed allocation
1325 * bytes in this file, and to maintain the list of inodes that
1326 * have pending delalloc work to be done.
1327 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001328static void btrfs_set_bit_hook(struct inode *inode,
1329 struct extent_state *state, int *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001330{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001331
Chris Mason75eff682008-12-15 15:54:40 -05001332 /*
1333 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001334 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001335 * bit, which is only set or cleared with irqs on
1336 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001337 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001338 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001339 u64 len = state->end + 1 - state->start;
Chris Mason2cf85722011-07-26 15:35:09 -04001340 bool do_list = !btrfs_is_free_space_inode(root, inode);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001341
Josef Bacik9e0baf62011-07-15 15:16:44 +00001342 if (*bits & EXTENT_FIRST_DELALLOC) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001343 *bits &= ~EXTENT_FIRST_DELALLOC;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001344 } else {
1345 spin_lock(&BTRFS_I(inode)->lock);
1346 BTRFS_I(inode)->outstanding_extents++;
1347 spin_unlock(&BTRFS_I(inode)->lock);
1348 }
Josef Bacik287a0ab2010-03-19 18:07:23 +00001349
Chris Mason75eff682008-12-15 15:54:40 -05001350 spin_lock(&root->fs_info->delalloc_lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001351 BTRFS_I(inode)->delalloc_bytes += len;
1352 root->fs_info->delalloc_bytes += len;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001353 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
Chris Masonea8c2812008-08-04 23:17:27 -04001354 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1355 &root->fs_info->delalloc_inodes);
1356 }
Chris Mason75eff682008-12-15 15:54:40 -05001357 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason291d6732008-01-29 15:55:23 -05001358 }
Chris Mason291d6732008-01-29 15:55:23 -05001359}
1360
Chris Masond352ac62008-09-29 15:18:18 -04001361/*
1362 * extent_io.c clear_bit_hook, see set_bit_hook for why
1363 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001364static void btrfs_clear_bit_hook(struct inode *inode,
1365 struct extent_state *state, int *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001366{
Chris Mason75eff682008-12-15 15:54:40 -05001367 /*
1368 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001369 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001370 * bit, which is only set or cleared with irqs on
1371 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001372 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001373 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001374 u64 len = state->end + 1 - state->start;
Chris Mason2cf85722011-07-26 15:35:09 -04001375 bool do_list = !btrfs_is_free_space_inode(root, inode);
Chris Masonbcbfce82008-04-22 13:26:47 -04001376
Josef Bacik9e0baf62011-07-15 15:16:44 +00001377 if (*bits & EXTENT_FIRST_DELALLOC) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001378 *bits &= ~EXTENT_FIRST_DELALLOC;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001379 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1380 spin_lock(&BTRFS_I(inode)->lock);
1381 BTRFS_I(inode)->outstanding_extents--;
1382 spin_unlock(&BTRFS_I(inode)->lock);
1383 }
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001384
1385 if (*bits & EXTENT_DO_ACCOUNTING)
1386 btrfs_delalloc_release_metadata(inode, len);
1387
Josef Bacik0cb59c92010-07-02 12:14:14 -04001388 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1389 && do_list)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001390 btrfs_free_reserved_data_space(inode, len);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001391
Chris Mason75eff682008-12-15 15:54:40 -05001392 spin_lock(&root->fs_info->delalloc_lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001393 root->fs_info->delalloc_bytes -= len;
1394 BTRFS_I(inode)->delalloc_bytes -= len;
1395
Josef Bacik0cb59c92010-07-02 12:14:14 -04001396 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
Chris Masonea8c2812008-08-04 23:17:27 -04001397 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1398 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1399 }
Chris Mason75eff682008-12-15 15:54:40 -05001400 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason291d6732008-01-29 15:55:23 -05001401 }
Chris Mason291d6732008-01-29 15:55:23 -05001402}
1403
Chris Masond352ac62008-09-29 15:18:18 -04001404/*
1405 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1406 * we don't create bios that span stripes or chunks
1407 */
Chris Mason239b14b2008-03-24 15:02:07 -04001408int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
Chris Masonc8b97812008-10-29 14:49:59 -04001409 size_t size, struct bio *bio,
1410 unsigned long bio_flags)
Chris Mason239b14b2008-03-24 15:02:07 -04001411{
1412 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1413 struct btrfs_mapping_tree *map_tree;
Chris Masona62b9402008-10-03 16:31:08 -04001414 u64 logical = (u64)bio->bi_sector << 9;
Chris Mason239b14b2008-03-24 15:02:07 -04001415 u64 length = 0;
1416 u64 map_length;
Chris Mason239b14b2008-03-24 15:02:07 -04001417 int ret;
1418
Chris Mason771ed682008-11-06 22:02:51 -05001419 if (bio_flags & EXTENT_BIO_COMPRESSED)
1420 return 0;
1421
Chris Masonf2d8d742008-04-21 10:03:05 -04001422 length = bio->bi_size;
Chris Mason239b14b2008-03-24 15:02:07 -04001423 map_tree = &root->fs_info->mapping_tree;
1424 map_length = length;
Chris Masoncea9e442008-04-09 16:28:12 -04001425 ret = btrfs_map_block(map_tree, READ, logical,
Chris Masonf1885912008-04-09 16:28:12 -04001426 &map_length, NULL, 0);
Jeff Mahoney3444a972011-10-03 23:23:13 -04001427 /* Will always return 0 or 1 with map_multi == NULL */
1428 BUG_ON(ret < 0);
Chris Masond3977122009-01-05 21:25:51 -05001429 if (map_length < length + size)
Chris Mason239b14b2008-03-24 15:02:07 -04001430 return 1;
Jeff Mahoney3444a972011-10-03 23:23:13 -04001431 return 0;
Chris Mason239b14b2008-03-24 15:02:07 -04001432}
1433
Chris Masond352ac62008-09-29 15:18:18 -04001434/*
1435 * in order to insert checksums into the metadata in large chunks,
1436 * we wait until bio submission time. All the pages in the bio are
1437 * checksummed and sums are attached onto the ordered extent record.
1438 *
1439 * At IO completion time the cums attached on the ordered extent record
1440 * are inserted into the btree
1441 */
Chris Masond3977122009-01-05 21:25:51 -05001442static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1443 struct bio *bio, int mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001444 unsigned long bio_flags,
1445 u64 bio_offset)
Chris Mason065631f2008-02-20 12:07:25 -05001446{
Chris Mason065631f2008-02-20 12:07:25 -05001447 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason065631f2008-02-20 12:07:25 -05001448 int ret = 0;
Chris Masone0156402008-04-16 11:15:20 -04001449
Chris Masond20f7042008-12-08 16:58:54 -05001450 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
Chris Mason44b8bd72008-04-16 11:14:51 -04001451 BUG_ON(ret);
Chris Mason4a69a412008-11-06 22:03:00 -05001452 return 0;
1453}
Chris Masone0156402008-04-16 11:15:20 -04001454
Chris Mason4a69a412008-11-06 22:03:00 -05001455/*
1456 * in order to insert checksums into the metadata in large chunks,
1457 * we wait until bio submission time. All the pages in the bio are
1458 * checksummed and sums are attached onto the ordered extent record.
1459 *
1460 * At IO completion time the cums attached on the ordered extent record
1461 * are inserted into the btree
1462 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001463static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001464 int mirror_num, unsigned long bio_flags,
1465 u64 bio_offset)
Chris Mason4a69a412008-11-06 22:03:00 -05001466{
1467 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason8b712842008-06-11 16:50:36 -04001468 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
Chris Mason44b8bd72008-04-16 11:14:51 -04001469}
1470
Chris Masond352ac62008-09-29 15:18:18 -04001471/*
Chris Masoncad321a2008-12-17 14:51:42 -05001472 * extent_io.c submission hook. This does the right thing for csum calculation
1473 * on write, or reading the csums from the tree before a read
Chris Masond352ac62008-09-29 15:18:18 -04001474 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001475static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001476 int mirror_num, unsigned long bio_flags,
1477 u64 bio_offset)
Chris Mason44b8bd72008-04-16 11:14:51 -04001478{
1479 struct btrfs_root *root = BTRFS_I(inode)->root;
1480 int ret = 0;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001481 int skip_sum;
Jeff Mahoney04173412011-10-03 23:23:12 -04001482 int metadata = 0;
Chris Mason44b8bd72008-04-16 11:14:51 -04001483
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001484 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masoncad321a2008-12-17 14:51:42 -05001485
Chris Mason2cf85722011-07-26 15:35:09 -04001486 if (btrfs_is_free_space_inode(root, inode))
Jeff Mahoney04173412011-10-03 23:23:12 -04001487 metadata = 2;
1488
1489 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
Jeff Mahoney355808c2011-10-03 23:23:14 -04001490 if (ret)
1491 return ret;
Chris Mason065631f2008-02-20 12:07:25 -05001492
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001493 if (!(rw & REQ_WRITE)) {
Chris Masond20f7042008-12-08 16:58:54 -05001494 if (bio_flags & EXTENT_BIO_COMPRESSED) {
Chris Masonc8b97812008-10-29 14:49:59 -04001495 return btrfs_submit_compressed_read(inode, bio,
1496 mirror_num, bio_flags);
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001497 } else if (!skip_sum) {
1498 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1499 if (ret)
1500 return ret;
1501 }
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001502 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001503 } else if (!skip_sum) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001504 /* csum items have already been cloned */
1505 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1506 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001507 /* we're doing a write, do the async checksumming */
1508 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
Chris Mason44b8bd72008-04-16 11:14:51 -04001509 inode, rw, bio, mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001510 bio_flags, bio_offset,
1511 __btrfs_submit_bio_start,
Chris Mason4a69a412008-11-06 22:03:00 -05001512 __btrfs_submit_bio_done);
Chris Mason19b9bdb2008-10-30 14:23:13 -04001513 }
1514
Chris Mason0b86a832008-03-24 15:01:56 -04001515mapit:
Chris Mason8b712842008-06-11 16:50:36 -04001516 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
Chris Mason065631f2008-02-20 12:07:25 -05001517}
Chris Mason6885f302008-02-20 16:11:05 -05001518
Chris Masond352ac62008-09-29 15:18:18 -04001519/*
1520 * given a list of ordered sums record them in the inode. This happens
1521 * at IO completion time based on sums calculated at bio submission time.
1522 */
Chris Masonba1da2f2008-07-17 12:54:15 -04001523static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
Chris Masone6dcd2d2008-07-17 12:53:50 -04001524 struct inode *inode, u64 file_offset,
1525 struct list_head *list)
1526{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001527 struct btrfs_ordered_sum *sum;
1528
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001529 list_for_each_entry(sum, list, list) {
Chris Masond20f7042008-12-08 16:58:54 -05001530 btrfs_csum_file_blocks(trans,
1531 BTRFS_I(inode)->root->fs_info->csum_root, sum);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001532 }
1533 return 0;
1534}
1535
Josef Bacik2ac55d42010-02-03 19:33:23 +00001536int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1537 struct extent_state **cached_state)
Chris Masonea8c2812008-08-04 23:17:27 -04001538{
Chris Masond3977122009-01-05 21:25:51 -05001539 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
Chris Mason771ed682008-11-06 22:02:51 -05001540 WARN_ON(1);
Chris Masonea8c2812008-08-04 23:17:27 -04001541 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00001542 cached_state, GFP_NOFS);
Chris Masonea8c2812008-08-04 23:17:27 -04001543}
1544
Chris Masond352ac62008-09-29 15:18:18 -04001545/* see btrfs_writepage_start_hook for details on why this is required */
Chris Mason247e7432008-07-17 12:53:51 -04001546struct btrfs_writepage_fixup {
1547 struct page *page;
1548 struct btrfs_work work;
1549};
1550
Christoph Hellwigb2950862008-12-02 09:54:17 -05001551static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
Chris Mason247e7432008-07-17 12:53:51 -04001552{
1553 struct btrfs_writepage_fixup *fixup;
1554 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001555 struct extent_state *cached_state = NULL;
Chris Mason247e7432008-07-17 12:53:51 -04001556 struct page *page;
1557 struct inode *inode;
1558 u64 page_start;
1559 u64 page_end;
Jeff Mahoney87826df2012-02-15 16:23:57 +01001560 int ret;
Chris Mason247e7432008-07-17 12:53:51 -04001561
1562 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1563 page = fixup->page;
Chris Mason4a096752008-07-21 10:29:44 -04001564again:
Chris Mason247e7432008-07-17 12:53:51 -04001565 lock_page(page);
1566 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1567 ClearPageChecked(page);
1568 goto out_page;
1569 }
1570
1571 inode = page->mapping->host;
1572 page_start = page_offset(page);
1573 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1574
Josef Bacik2ac55d42010-02-03 19:33:23 +00001575 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001576 &cached_state);
Chris Mason4a096752008-07-21 10:29:44 -04001577
1578 /* already ordered? We're done */
Chris Mason8b62b722009-09-02 16:53:46 -04001579 if (PagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001580 goto out;
Chris Mason4a096752008-07-21 10:29:44 -04001581
1582 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1583 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00001584 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1585 page_end, &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001586 unlock_page(page);
1587 btrfs_start_ordered_extent(inode, ordered, 1);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001588 btrfs_put_ordered_extent(ordered);
Chris Mason4a096752008-07-21 10:29:44 -04001589 goto again;
1590 }
Chris Mason247e7432008-07-17 12:53:51 -04001591
Jeff Mahoney87826df2012-02-15 16:23:57 +01001592 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1593 if (ret) {
1594 mapping_set_error(page->mapping, ret);
1595 end_extent_writepage(page, ret, page_start, page_end);
1596 ClearPageChecked(page);
1597 goto out;
1598 }
1599
Josef Bacik2ac55d42010-02-03 19:33:23 +00001600 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
Chris Mason247e7432008-07-17 12:53:51 -04001601 ClearPageChecked(page);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001602 set_page_dirty(page);
Chris Mason247e7432008-07-17 12:53:51 -04001603out:
Josef Bacik2ac55d42010-02-03 19:33:23 +00001604 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1605 &cached_state, GFP_NOFS);
Chris Mason247e7432008-07-17 12:53:51 -04001606out_page:
1607 unlock_page(page);
1608 page_cache_release(page);
Miao Xieb897abe2011-01-26 16:19:22 +08001609 kfree(fixup);
Chris Mason247e7432008-07-17 12:53:51 -04001610}
1611
1612/*
1613 * There are a few paths in the higher layers of the kernel that directly
1614 * set the page dirty bit without asking the filesystem if it is a
1615 * good idea. This causes problems because we want to make sure COW
1616 * properly happens and the data=ordered rules are followed.
1617 *
Chris Masonc8b97812008-10-29 14:49:59 -04001618 * In our case any range that doesn't have the ORDERED bit set
Chris Mason247e7432008-07-17 12:53:51 -04001619 * hasn't been properly setup for IO. We kick off an async process
1620 * to fix it up. The async helper will wait for ordered extents, set
1621 * the delalloc bit and make it safe to write the page.
1622 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001623static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
Chris Mason247e7432008-07-17 12:53:51 -04001624{
1625 struct inode *inode = page->mapping->host;
1626 struct btrfs_writepage_fixup *fixup;
1627 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason247e7432008-07-17 12:53:51 -04001628
Chris Mason8b62b722009-09-02 16:53:46 -04001629 /* this page is properly in the ordered list */
1630 if (TestClearPagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001631 return 0;
1632
1633 if (PageChecked(page))
1634 return -EAGAIN;
1635
1636 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1637 if (!fixup)
1638 return -EAGAIN;
Chris Masonf4219502008-07-22 11:18:09 -04001639
Chris Mason247e7432008-07-17 12:53:51 -04001640 SetPageChecked(page);
1641 page_cache_get(page);
1642 fixup->work.func = btrfs_writepage_fixup_worker;
1643 fixup->page = page;
1644 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001645 return -EBUSY;
Chris Mason247e7432008-07-17 12:53:51 -04001646}
1647
Yan Zhengd899e052008-10-30 14:25:28 -04001648static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1649 struct inode *inode, u64 file_pos,
1650 u64 disk_bytenr, u64 disk_num_bytes,
1651 u64 num_bytes, u64 ram_bytes,
1652 u8 compression, u8 encryption,
1653 u16 other_encoding, int extent_type)
1654{
1655 struct btrfs_root *root = BTRFS_I(inode)->root;
1656 struct btrfs_file_extent_item *fi;
1657 struct btrfs_path *path;
1658 struct extent_buffer *leaf;
1659 struct btrfs_key ins;
1660 u64 hint;
1661 int ret;
1662
1663 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07001664 if (!path)
1665 return -ENOMEM;
Yan Zhengd899e052008-10-30 14:25:28 -04001666
Chris Masonb9473432009-03-13 11:00:37 -04001667 path->leave_spinning = 1;
Chris Masona1ed8352009-09-11 12:27:37 -04001668
1669 /*
1670 * we may be replacing one extent in the tree with another.
1671 * The new extent is pinned in the extent map, and we don't want
1672 * to drop it from the cache until it is completely in the btree.
1673 *
1674 * So, tell btrfs_drop_extents to leave this extent in the cache.
1675 * the caller is expected to unpin it and allow it to be merged
1676 * with the others.
1677 */
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001678 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1679 &hint, 0);
Yan Zhengd899e052008-10-30 14:25:28 -04001680 BUG_ON(ret);
1681
Li Zefan33345d012011-04-20 10:31:50 +08001682 ins.objectid = btrfs_ino(inode);
Yan Zhengd899e052008-10-30 14:25:28 -04001683 ins.offset = file_pos;
1684 ins.type = BTRFS_EXTENT_DATA_KEY;
1685 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1686 BUG_ON(ret);
1687 leaf = path->nodes[0];
1688 fi = btrfs_item_ptr(leaf, path->slots[0],
1689 struct btrfs_file_extent_item);
1690 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1691 btrfs_set_file_extent_type(leaf, fi, extent_type);
1692 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1693 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1694 btrfs_set_file_extent_offset(leaf, fi, 0);
1695 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1696 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1697 btrfs_set_file_extent_compression(leaf, fi, compression);
1698 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1699 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
Chris Masonb9473432009-03-13 11:00:37 -04001700
1701 btrfs_unlock_up_safe(path, 1);
1702 btrfs_set_lock_blocking(leaf);
1703
Yan Zhengd899e052008-10-30 14:25:28 -04001704 btrfs_mark_buffer_dirty(leaf);
1705
1706 inode_add_bytes(inode, num_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -04001707
1708 ins.objectid = disk_bytenr;
1709 ins.offset = disk_num_bytes;
1710 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001711 ret = btrfs_alloc_reserved_file_extent(trans, root,
1712 root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001713 btrfs_ino(inode), file_pos, &ins);
Yan Zhengd899e052008-10-30 14:25:28 -04001714 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -04001715 btrfs_free_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04001716
Yan Zhengd899e052008-10-30 14:25:28 -04001717 return 0;
1718}
1719
Chris Mason5d13a982009-03-13 11:41:46 -04001720/*
1721 * helper function for btrfs_finish_ordered_io, this
1722 * just reads in some of the csum leaves to prime them into ram
1723 * before we start the transaction. It limits the amount of btree
1724 * reads required while inside the transaction.
1725 */
Chris Masond352ac62008-09-29 15:18:18 -04001726/* as ordered data IO finishes, this gets called so we can finish
1727 * an ordered extent if the range of bytes in the file it covers are
1728 * fully written.
1729 */
Chris Mason211f90e2008-07-18 11:56:15 -04001730static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001731{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001732 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001733 struct btrfs_trans_handle *trans = NULL;
Chris Mason5d13a982009-03-13 11:41:46 -04001734 struct btrfs_ordered_extent *ordered_extent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001735 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001736 struct extent_state *cached_state = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08001737 int compress_type = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001738 int ret;
Li Zefan82d59022011-04-20 10:33:24 +08001739 bool nolock;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001740
Josef Bacik5a1a3df2010-02-02 20:51:14 +00001741 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1742 end - start + 1);
Chris Masonba1da2f2008-07-17 12:54:15 -04001743 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001744 return 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001745 BUG_ON(!ordered_extent);
Josef Bacikefd049f2010-02-02 20:50:10 +00001746
Chris Mason2cf85722011-07-26 15:35:09 -04001747 nolock = btrfs_is_free_space_inode(root, inode);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001748
Yan, Zhengc2167752009-11-12 09:34:21 +00001749 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1750 BUG_ON(!list_empty(&ordered_extent->list));
1751 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1752 if (!ret) {
Josef Bacik0cb59c92010-07-02 12:14:14 -04001753 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001754 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001755 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001756 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001757 BUG_ON(IS_ERR(trans));
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001758 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Mason21151332011-11-10 20:39:08 -05001759 ret = btrfs_update_inode_fallback(trans, root, inode);
Yan, Zhengc2167752009-11-12 09:34:21 +00001760 BUG_ON(ret);
Yan, Zhengc2167752009-11-12 09:34:21 +00001761 }
1762 goto out;
1763 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04001764
Josef Bacik2ac55d42010-02-03 19:33:23 +00001765 lock_extent_bits(io_tree, ordered_extent->file_offset,
1766 ordered_extent->file_offset + ordered_extent->len - 1,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001767 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001768
Josef Bacik0cb59c92010-07-02 12:14:14 -04001769 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001770 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001771 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001772 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001773 BUG_ON(IS_ERR(trans));
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001774 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Yan, Zhengc2167752009-11-12 09:34:21 +00001775
Chris Masonc8b97812008-10-29 14:49:59 -04001776 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
Li Zefan261507a02010-12-17 14:21:50 +08001777 compress_type = ordered_extent->compress_type;
Yan Zhengd899e052008-10-30 14:25:28 -04001778 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
Li Zefan261507a02010-12-17 14:21:50 +08001779 BUG_ON(compress_type);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001780 ret = btrfs_mark_extent_written(trans, inode,
Yan Zhengd899e052008-10-30 14:25:28 -04001781 ordered_extent->file_offset,
1782 ordered_extent->file_offset +
1783 ordered_extent->len);
1784 BUG_ON(ret);
1785 } else {
Josef Bacik0af3d002010-06-21 14:48:16 -04001786 BUG_ON(root == root->fs_info->tree_root);
Yan Zhengd899e052008-10-30 14:25:28 -04001787 ret = insert_reserved_file_extent(trans, inode,
1788 ordered_extent->file_offset,
1789 ordered_extent->start,
1790 ordered_extent->disk_len,
1791 ordered_extent->len,
1792 ordered_extent->len,
Li Zefan261507a02010-12-17 14:21:50 +08001793 compress_type, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04001794 BTRFS_FILE_EXTENT_REG);
Chris Masona1ed8352009-09-11 12:27:37 -04001795 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1796 ordered_extent->file_offset,
1797 ordered_extent->len);
Yan Zhengd899e052008-10-30 14:25:28 -04001798 BUG_ON(ret);
1799 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00001800 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1801 ordered_extent->file_offset +
1802 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1803
Chris Masone6dcd2d2008-07-17 12:53:50 -04001804 add_pending_csums(trans, inode, ordered_extent->file_offset,
1805 &ordered_extent->list);
1806
Josef Bacik1ef30be2011-04-05 19:25:36 -04001807 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
Miao Xiea39f7522011-09-11 10:52:25 -04001808 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
Chris Mason21151332011-11-10 20:39:08 -05001809 ret = btrfs_update_inode_fallback(trans, root, inode);
Josef Bacik1ef30be2011-04-05 19:25:36 -04001810 BUG_ON(ret);
1811 }
1812 ret = 0;
Yan, Zhengc2167752009-11-12 09:34:21 +00001813out:
Josef Bacik5b0e95b2011-10-06 08:58:24 -04001814 if (root != root->fs_info->tree_root)
Josef Bacik0cb59c92010-07-02 12:14:14 -04001815 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
Josef Bacik5b0e95b2011-10-06 08:58:24 -04001816 if (trans) {
1817 if (nolock)
Josef Bacik0cb59c92010-07-02 12:14:14 -04001818 btrfs_end_transaction_nolock(trans, root);
Josef Bacik5b0e95b2011-10-06 08:58:24 -04001819 else
Josef Bacik0cb59c92010-07-02 12:14:14 -04001820 btrfs_end_transaction(trans, root);
1821 }
1822
Chris Masone6dcd2d2008-07-17 12:53:50 -04001823 /* once for us */
1824 btrfs_put_ordered_extent(ordered_extent);
1825 /* once for the tree */
1826 btrfs_put_ordered_extent(ordered_extent);
1827
Chris Masone6dcd2d2008-07-17 12:53:50 -04001828 return 0;
1829}
1830
Christoph Hellwigb2950862008-12-02 09:54:17 -05001831static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason211f90e2008-07-18 11:56:15 -04001832 struct extent_state *state, int uptodate)
1833{
liubo1abe9b82011-03-24 11:18:59 +00001834 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1835
Chris Mason8b62b722009-09-02 16:53:46 -04001836 ClearPagePrivate2(page);
Chris Mason211f90e2008-07-18 11:56:15 -04001837 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1838}
1839
Chris Masond352ac62008-09-29 15:18:18 -04001840/*
Chris Masond352ac62008-09-29 15:18:18 -04001841 * when reads are done, we need to check csums to verify the data is correct
Jan Schmidt4a54c8c2011-07-22 15:41:52 +02001842 * if there's a match, we allow the bio to finish. If not, the code in
1843 * extent_io.c will try to find good copies for us.
Chris Masond352ac62008-09-29 15:18:18 -04001844 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001845static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason70dec802008-01-29 09:59:12 -05001846 struct extent_state *state)
Chris Mason07157aa2007-08-30 08:50:51 -04001847{
Chris Mason35ebb932007-10-30 16:56:53 -04001848 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
Chris Mason07157aa2007-08-30 08:50:51 -04001849 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05001850 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason07157aa2007-08-30 08:50:51 -04001851 char *kaddr;
Chris Masonaadfeb62008-01-29 09:10:27 -05001852 u64 private = ~(u32)0;
Chris Mason07157aa2007-08-30 08:50:51 -04001853 int ret;
Chris Masonff79f812007-10-15 16:22:25 -04001854 struct btrfs_root *root = BTRFS_I(inode)->root;
1855 u32 csum = ~(u32)0;
Chris Masond1310b22008-01-24 16:13:08 -05001856
Chris Masond20f7042008-12-08 16:58:54 -05001857 if (PageChecked(page)) {
1858 ClearPageChecked(page);
1859 goto good;
1860 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001861
1862 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Jan Schmidt08d2f342011-05-04 16:18:50 +02001863 goto good;
Chris Masond20f7042008-12-08 16:58:54 -05001864
Yan Zheng17d217f2008-12-12 10:03:38 -05001865 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
Chris Mason9655d292009-09-02 15:22:30 -04001866 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001867 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1868 GFP_NOFS);
1869 return 0;
1870 }
1871
Yanc2e639f2008-02-04 08:57:25 -05001872 if (state && state->start == start) {
Chris Mason70dec802008-01-29 09:59:12 -05001873 private = state->private;
1874 ret = 0;
1875 } else {
1876 ret = get_state_private(io_tree, start, &private);
1877 }
Chris Mason9ab86c82009-01-07 09:48:51 -05001878 kaddr = kmap_atomic(page, KM_USER0);
Chris Masond3977122009-01-05 21:25:51 -05001879 if (ret)
Chris Mason07157aa2007-08-30 08:50:51 -04001880 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05001881
Chris Masonff79f812007-10-15 16:22:25 -04001882 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1883 btrfs_csum_final(csum, (char *)&csum);
Chris Masond3977122009-01-05 21:25:51 -05001884 if (csum != private)
Chris Mason07157aa2007-08-30 08:50:51 -04001885 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05001886
Chris Mason9ab86c82009-01-07 09:48:51 -05001887 kunmap_atomic(kaddr, KM_USER0);
Chris Masond20f7042008-12-08 16:58:54 -05001888good:
Chris Mason07157aa2007-08-30 08:50:51 -04001889 return 0;
1890
1891zeroit:
Chris Mason945d8962011-05-22 12:33:42 -04001892 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
Li Zefan33345d012011-04-20 10:31:50 +08001893 "private %llu\n",
1894 (unsigned long long)btrfs_ino(page->mapping->host),
Chris Mason193f2842009-04-27 07:29:05 -04001895 (unsigned long long)start, csum,
1896 (unsigned long long)private);
Chris Masondb945352007-10-15 16:15:53 -04001897 memset(kaddr + offset, 1, end - start + 1);
1898 flush_dcache_page(page);
Chris Mason9ab86c82009-01-07 09:48:51 -05001899 kunmap_atomic(kaddr, KM_USER0);
Chris Mason3b951512008-04-17 11:29:12 -04001900 if (private == 0)
1901 return 0;
Chris Mason7e383262008-04-09 16:28:12 -04001902 return -EIO;
Chris Mason07157aa2007-08-30 08:50:51 -04001903}
Chris Masonb888db22007-08-27 16:49:44 -04001904
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001905struct delayed_iput {
1906 struct list_head list;
1907 struct inode *inode;
1908};
1909
1910void btrfs_add_delayed_iput(struct inode *inode)
1911{
1912 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1913 struct delayed_iput *delayed;
1914
1915 if (atomic_add_unless(&inode->i_count, -1, 1))
1916 return;
1917
1918 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1919 delayed->inode = inode;
1920
1921 spin_lock(&fs_info->delayed_iput_lock);
1922 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1923 spin_unlock(&fs_info->delayed_iput_lock);
1924}
1925
1926void btrfs_run_delayed_iputs(struct btrfs_root *root)
1927{
1928 LIST_HEAD(list);
1929 struct btrfs_fs_info *fs_info = root->fs_info;
1930 struct delayed_iput *delayed;
1931 int empty;
1932
1933 spin_lock(&fs_info->delayed_iput_lock);
1934 empty = list_empty(&fs_info->delayed_iputs);
1935 spin_unlock(&fs_info->delayed_iput_lock);
1936 if (empty)
1937 return;
1938
1939 down_read(&root->fs_info->cleanup_work_sem);
1940 spin_lock(&fs_info->delayed_iput_lock);
1941 list_splice_init(&fs_info->delayed_iputs, &list);
1942 spin_unlock(&fs_info->delayed_iput_lock);
1943
1944 while (!list_empty(&list)) {
1945 delayed = list_entry(list.next, struct delayed_iput, list);
1946 list_del(&delayed->list);
1947 iput(delayed->inode);
1948 kfree(delayed);
1949 }
1950 up_read(&root->fs_info->cleanup_work_sem);
1951}
1952
Yan, Zhengd68fc572010-05-16 10:49:58 -04001953enum btrfs_orphan_cleanup_state {
1954 ORPHAN_CLEANUP_STARTED = 1,
1955 ORPHAN_CLEANUP_DONE = 2,
1956};
1957
1958/*
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08001959 * This is called in transaction commit time. If there are no orphan
Yan, Zhengd68fc572010-05-16 10:49:58 -04001960 * files in the subvolume, it removes orphan item and frees block_rsv
1961 * structure.
1962 */
1963void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1964 struct btrfs_root *root)
1965{
Josef Bacik90290e12011-12-02 15:44:12 -05001966 struct btrfs_block_rsv *block_rsv;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001967 int ret;
1968
1969 if (!list_empty(&root->orphan_list) ||
1970 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1971 return;
1972
Josef Bacik90290e12011-12-02 15:44:12 -05001973 spin_lock(&root->orphan_lock);
1974 if (!list_empty(&root->orphan_list)) {
1975 spin_unlock(&root->orphan_lock);
1976 return;
1977 }
1978
1979 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
1980 spin_unlock(&root->orphan_lock);
1981 return;
1982 }
1983
1984 block_rsv = root->orphan_block_rsv;
1985 root->orphan_block_rsv = NULL;
1986 spin_unlock(&root->orphan_lock);
1987
Yan, Zhengd68fc572010-05-16 10:49:58 -04001988 if (root->orphan_item_inserted &&
1989 btrfs_root_refs(&root->root_item) > 0) {
1990 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1991 root->root_key.objectid);
1992 BUG_ON(ret);
1993 root->orphan_item_inserted = 0;
1994 }
1995
Josef Bacik90290e12011-12-02 15:44:12 -05001996 if (block_rsv) {
1997 WARN_ON(block_rsv->size > 0);
1998 btrfs_free_block_rsv(root, block_rsv);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001999 }
2000}
2001
2002/*
Josef Bacik7b128762008-07-24 12:17:14 -04002003 * This creates an orphan entry for the given inode in case something goes
2004 * wrong in the middle of an unlink/truncate.
Yan, Zhengd68fc572010-05-16 10:49:58 -04002005 *
2006 * NOTE: caller of this function should reserve 5 units of metadata for
2007 * this function.
Josef Bacik7b128762008-07-24 12:17:14 -04002008 */
2009int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2010{
2011 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002012 struct btrfs_block_rsv *block_rsv = NULL;
2013 int reserve = 0;
2014 int insert = 0;
2015 int ret;
Josef Bacik7b128762008-07-24 12:17:14 -04002016
Yan, Zhengd68fc572010-05-16 10:49:58 -04002017 if (!root->orphan_block_rsv) {
2018 block_rsv = btrfs_alloc_block_rsv(root);
Tsutomu Itohb5324022011-07-19 07:27:20 +00002019 if (!block_rsv)
2020 return -ENOMEM;
Josef Bacik7b128762008-07-24 12:17:14 -04002021 }
2022
Yan, Zhengd68fc572010-05-16 10:49:58 -04002023 spin_lock(&root->orphan_lock);
2024 if (!root->orphan_block_rsv) {
2025 root->orphan_block_rsv = block_rsv;
2026 } else if (block_rsv) {
2027 btrfs_free_block_rsv(root, block_rsv);
2028 block_rsv = NULL;
2029 }
Josef Bacik7b128762008-07-24 12:17:14 -04002030
Yan, Zhengd68fc572010-05-16 10:49:58 -04002031 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2032 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2033#if 0
2034 /*
2035 * For proper ENOSPC handling, we should do orphan
2036 * cleanup when mounting. But this introduces backward
2037 * compatibility issue.
2038 */
2039 if (!xchg(&root->orphan_item_inserted, 1))
2040 insert = 2;
2041 else
2042 insert = 1;
2043#endif
2044 insert = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002045 }
Josef Bacik7b128762008-07-24 12:17:14 -04002046
Yan, Zhengd68fc572010-05-16 10:49:58 -04002047 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2048 BTRFS_I(inode)->orphan_meta_reserved = 1;
2049 reserve = 1;
2050 }
2051 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002052
Yan, Zhengd68fc572010-05-16 10:49:58 -04002053 /* grab metadata reservation from transaction handle */
2054 if (reserve) {
2055 ret = btrfs_orphan_reserve_metadata(trans, inode);
2056 BUG_ON(ret);
2057 }
2058
2059 /* insert an orphan item to track this unlinked/truncated file */
2060 if (insert >= 1) {
Li Zefan33345d012011-04-20 10:31:50 +08002061 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
Josef Bacikee4d89f2011-12-13 12:55:58 -05002062 BUG_ON(ret && ret != -EEXIST);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002063 }
2064
2065 /* insert an orphan item to track subvolume contains orphan files */
2066 if (insert >= 2) {
2067 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2068 root->root_key.objectid);
2069 BUG_ON(ret);
2070 }
2071 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002072}
2073
2074/*
2075 * We have done the truncate/delete so we can go ahead and remove the orphan
2076 * item for this particular inode.
2077 */
2078int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2079{
2080 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002081 int delete_item = 0;
2082 int release_rsv = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002083 int ret = 0;
2084
Yan, Zhengd68fc572010-05-16 10:49:58 -04002085 spin_lock(&root->orphan_lock);
2086 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2087 list_del_init(&BTRFS_I(inode)->i_orphan);
2088 delete_item = 1;
Josef Bacik7b128762008-07-24 12:17:14 -04002089 }
2090
Yan, Zhengd68fc572010-05-16 10:49:58 -04002091 if (BTRFS_I(inode)->orphan_meta_reserved) {
2092 BTRFS_I(inode)->orphan_meta_reserved = 0;
2093 release_rsv = 1;
2094 }
2095 spin_unlock(&root->orphan_lock);
2096
2097 if (trans && delete_item) {
Li Zefan33345d012011-04-20 10:31:50 +08002098 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
Yan, Zhengd68fc572010-05-16 10:49:58 -04002099 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04002100 }
2101
Yan, Zhengd68fc572010-05-16 10:49:58 -04002102 if (release_rsv)
2103 btrfs_orphan_release_metadata(inode);
Josef Bacik7b128762008-07-24 12:17:14 -04002104
Yan, Zhengd68fc572010-05-16 10:49:58 -04002105 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002106}
2107
2108/*
2109 * this cleans up any orphans that may be left on the list from the last use
2110 * of this root.
2111 */
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002112int btrfs_orphan_cleanup(struct btrfs_root *root)
Josef Bacik7b128762008-07-24 12:17:14 -04002113{
2114 struct btrfs_path *path;
2115 struct extent_buffer *leaf;
Josef Bacik7b128762008-07-24 12:17:14 -04002116 struct btrfs_key key, found_key;
2117 struct btrfs_trans_handle *trans;
2118 struct inode *inode;
Josef Bacik8f6d7f42011-09-26 15:55:20 -04002119 u64 last_objectid = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002120 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2121
Yan, Zhengd68fc572010-05-16 10:49:58 -04002122 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002123 return 0;
Yan, Zhengc71bf092009-11-12 09:34:40 +00002124
2125 path = btrfs_alloc_path();
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002126 if (!path) {
2127 ret = -ENOMEM;
2128 goto out;
2129 }
Josef Bacik7b128762008-07-24 12:17:14 -04002130 path->reada = -1;
2131
2132 key.objectid = BTRFS_ORPHAN_OBJECTID;
2133 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2134 key.offset = (u64)-1;
2135
Josef Bacik7b128762008-07-24 12:17:14 -04002136 while (1) {
2137 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002138 if (ret < 0)
2139 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002140
2141 /*
2142 * if ret == 0 means we found what we were searching for, which
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002143 * is weird, but possible, so only screw with path if we didn't
Josef Bacik7b128762008-07-24 12:17:14 -04002144 * find the key and see if we have stuff that matches
2145 */
2146 if (ret > 0) {
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002147 ret = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002148 if (path->slots[0] == 0)
2149 break;
2150 path->slots[0]--;
2151 }
2152
2153 /* pull out the item */
2154 leaf = path->nodes[0];
Josef Bacik7b128762008-07-24 12:17:14 -04002155 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2156
2157 /* make sure the item matches what we want */
2158 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2159 break;
2160 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2161 break;
2162
2163 /* release the path since we're done with it */
David Sterbab3b4aa72011-04-21 01:20:15 +02002164 btrfs_release_path(path);
Josef Bacik7b128762008-07-24 12:17:14 -04002165
2166 /*
2167 * this is where we are basically btrfs_lookup, without the
2168 * crossing root thing. we store the inode number in the
2169 * offset of the orphan item.
2170 */
Josef Bacik8f6d7f42011-09-26 15:55:20 -04002171
2172 if (found_key.offset == last_objectid) {
2173 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2174 "stopping orphan cleanup\n");
2175 ret = -EINVAL;
2176 goto out;
2177 }
2178
2179 last_objectid = found_key.offset;
2180
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002181 found_key.objectid = found_key.offset;
2182 found_key.type = BTRFS_INODE_ITEM_KEY;
2183 found_key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +00002184 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
Josef Bacika8c9e572011-09-21 16:55:59 -04002185 ret = PTR_RET(inode);
2186 if (ret && ret != -ESTALE)
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002187 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04002188
Arne Jansenf8e9e0b2011-12-14 20:12:02 -05002189 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2190 struct btrfs_root *dead_root;
2191 struct btrfs_fs_info *fs_info = root->fs_info;
2192 int is_dead_root = 0;
2193
2194 /*
2195 * this is an orphan in the tree root. Currently these
2196 * could come from 2 sources:
2197 * a) a snapshot deletion in progress
2198 * b) a free space cache inode
2199 * We need to distinguish those two, as the snapshot
2200 * orphan must not get deleted.
2201 * find_dead_roots already ran before us, so if this
2202 * is a snapshot deletion, we should find the root
2203 * in the dead_roots list
2204 */
2205 spin_lock(&fs_info->trans_lock);
2206 list_for_each_entry(dead_root, &fs_info->dead_roots,
2207 root_list) {
2208 if (dead_root->root_key.objectid ==
2209 found_key.objectid) {
2210 is_dead_root = 1;
2211 break;
2212 }
2213 }
2214 spin_unlock(&fs_info->trans_lock);
2215 if (is_dead_root) {
2216 /* prevent this orphan from being found again */
2217 key.offset = found_key.objectid - 1;
2218 continue;
2219 }
2220 }
Josef Bacika8c9e572011-09-21 16:55:59 -04002221 /*
2222 * Inode is already gone but the orphan item is still there,
2223 * kill the orphan item.
2224 */
2225 if (ret == -ESTALE) {
2226 trans = btrfs_start_transaction(root, 1);
2227 if (IS_ERR(trans)) {
2228 ret = PTR_ERR(trans);
2229 goto out;
2230 }
2231 ret = btrfs_del_orphan_item(trans, root,
2232 found_key.objectid);
2233 BUG_ON(ret);
2234 btrfs_end_transaction(trans, root);
2235 continue;
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002236 }
Josef Bacik7b128762008-07-24 12:17:14 -04002237
Josef Bacik7b128762008-07-24 12:17:14 -04002238 /*
2239 * add this inode to the orphan list so btrfs_orphan_del does
2240 * the proper thing when we hit it
2241 */
Yan, Zhengd68fc572010-05-16 10:49:58 -04002242 spin_lock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002243 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002244 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002245
Josef Bacik7b128762008-07-24 12:17:14 -04002246 /* if we have links, this was a truncate, lets do that */
2247 if (inode->i_nlink) {
Josef Bacika41ad392011-01-31 15:30:16 -05002248 if (!S_ISREG(inode->i_mode)) {
2249 WARN_ON(1);
2250 iput(inode);
2251 continue;
2252 }
Josef Bacik7b128762008-07-24 12:17:14 -04002253 nr_truncate++;
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002254 ret = btrfs_truncate(inode);
Josef Bacik7b128762008-07-24 12:17:14 -04002255 } else {
2256 nr_unlink++;
2257 }
2258
2259 /* this will do delete_inode and everything for us */
2260 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002261 if (ret)
2262 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002263 }
Miao Xie3254c872011-11-10 20:45:05 -05002264 /* release the path since we're done with it */
2265 btrfs_release_path(path);
2266
Yan, Zhengd68fc572010-05-16 10:49:58 -04002267 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2268
2269 if (root->orphan_block_rsv)
2270 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2271 (u64)-1);
2272
2273 if (root->orphan_block_rsv || root->orphan_item_inserted) {
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002274 trans = btrfs_join_transaction(root);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002275 if (!IS_ERR(trans))
2276 btrfs_end_transaction(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002277 }
Josef Bacik7b128762008-07-24 12:17:14 -04002278
2279 if (nr_unlink)
2280 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2281 if (nr_truncate)
2282 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002283
2284out:
2285 if (ret)
2286 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2287 btrfs_free_path(path);
2288 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04002289}
2290
Chris Masond352ac62008-09-29 15:18:18 -04002291/*
Chris Mason46a53cc2009-04-27 11:47:50 -04002292 * very simple check to peek ahead in the leaf looking for xattrs. If we
2293 * don't find any xattrs, we know there can't be any acls.
2294 *
2295 * slot is the slot the inode is in, objectid is the objectid of the inode
2296 */
2297static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2298 int slot, u64 objectid)
2299{
2300 u32 nritems = btrfs_header_nritems(leaf);
2301 struct btrfs_key found_key;
2302 int scanned = 0;
2303
2304 slot++;
2305 while (slot < nritems) {
2306 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2307
2308 /* we found a different objectid, there must not be acls */
2309 if (found_key.objectid != objectid)
2310 return 0;
2311
2312 /* we found an xattr, assume we've got an acl */
2313 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2314 return 1;
2315
2316 /*
2317 * we found a key greater than an xattr key, there can't
2318 * be any acls later on
2319 */
2320 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2321 return 0;
2322
2323 slot++;
2324 scanned++;
2325
2326 /*
2327 * it goes inode, inode backrefs, xattrs, extents,
2328 * so if there are a ton of hard links to an inode there can
2329 * be a lot of backrefs. Don't waste time searching too hard,
2330 * this is just an optimization
2331 */
2332 if (scanned >= 8)
2333 break;
2334 }
2335 /* we hit the end of the leaf before we found an xattr or
2336 * something larger than an xattr. We have to assume the inode
2337 * has acls
2338 */
2339 return 1;
2340}
2341
2342/*
Chris Masond352ac62008-09-29 15:18:18 -04002343 * read an inode from the btree into the in-memory inode
2344 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002345static void btrfs_read_locked_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002346{
2347 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002348 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002349 struct btrfs_inode_item *inode_item;
Chris Mason0b86a832008-03-24 15:01:56 -04002350 struct btrfs_timespec *tspec;
Chris Mason39279cc2007-06-12 06:35:45 -04002351 struct btrfs_root *root = BTRFS_I(inode)->root;
2352 struct btrfs_key location;
Chris Mason46a53cc2009-04-27 11:47:50 -04002353 int maybe_acls;
Josef Bacik618e21d2007-07-11 10:18:17 -04002354 u32 rdev;
Chris Mason39279cc2007-06-12 06:35:45 -04002355 int ret;
Miao Xie2f7e33d2011-06-23 07:27:13 +00002356 bool filled = false;
2357
2358 ret = btrfs_fill_inode(inode, &rdev);
2359 if (!ret)
2360 filled = true;
Chris Mason39279cc2007-06-12 06:35:45 -04002361
2362 path = btrfs_alloc_path();
Mark Fasheh1748f842011-07-12 11:25:31 -07002363 if (!path)
2364 goto make_bad;
2365
Josef Bacikd90c7322011-05-17 09:50:54 -04002366 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04002367 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
Chris Masondc17ff82008-01-08 15:46:30 -05002368
Chris Mason39279cc2007-06-12 06:35:45 -04002369 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04002370 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04002371 goto make_bad;
Chris Mason39279cc2007-06-12 06:35:45 -04002372
Chris Mason5f39d392007-10-15 16:14:19 -04002373 leaf = path->nodes[0];
Miao Xie2f7e33d2011-06-23 07:27:13 +00002374
2375 if (filled)
2376 goto cache_acl;
2377
Chris Mason5f39d392007-10-15 16:14:19 -04002378 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2379 struct btrfs_inode_item);
Chris Mason5f39d392007-10-15 16:14:19 -04002380 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
Miklos Szeredibfe86842011-10-28 14:13:29 +02002381 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04002382 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2383 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
Chris Masondbe674a2008-07-17 12:54:05 -04002384 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04002385
2386 tspec = btrfs_inode_atime(inode_item);
2387 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2388 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2389
2390 tspec = btrfs_inode_mtime(inode_item);
2391 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2392 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2393
2394 tspec = btrfs_inode_ctime(inode_item);
2395 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2396 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2397
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002398 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04002399 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
Chris Masonc3027eb2008-12-08 16:40:21 -05002400 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04002401 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik618e21d2007-07-11 10:18:17 -04002402 inode->i_rdev = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002403 rdev = btrfs_inode_rdev(leaf, inode_item);
2404
Josef Bacikaec74772008-07-24 12:12:38 -04002405 BTRFS_I(inode)->index_cnt = (u64)-1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05002406 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
Miao Xie2f7e33d2011-06-23 07:27:13 +00002407cache_acl:
Chris Mason46a53cc2009-04-27 11:47:50 -04002408 /*
2409 * try to precache a NULL acl entry for files that don't have
2410 * any xattrs or acls
2411 */
Li Zefan33345d012011-04-20 10:31:50 +08002412 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2413 btrfs_ino(inode));
Al Viro72c04902009-06-24 16:58:48 -04002414 if (!maybe_acls)
2415 cache_no_acl(inode);
Chris Mason46a53cc2009-04-27 11:47:50 -04002416
Chris Mason39279cc2007-06-12 06:35:45 -04002417 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002418
Chris Mason39279cc2007-06-12 06:35:45 -04002419 switch (inode->i_mode & S_IFMT) {
Chris Mason39279cc2007-06-12 06:35:45 -04002420 case S_IFREG:
2421 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04002422 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05002423 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04002424 inode->i_fop = &btrfs_file_operations;
2425 inode->i_op = &btrfs_file_inode_operations;
2426 break;
2427 case S_IFDIR:
2428 inode->i_fop = &btrfs_dir_file_operations;
2429 if (root == root->fs_info->tree_root)
2430 inode->i_op = &btrfs_dir_ro_inode_operations;
2431 else
2432 inode->i_op = &btrfs_dir_inode_operations;
2433 break;
2434 case S_IFLNK:
2435 inode->i_op = &btrfs_symlink_inode_operations;
2436 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04002437 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04002438 break;
Josef Bacik618e21d2007-07-11 10:18:17 -04002439 default:
Jim Owens0279b4c2009-02-04 09:29:13 -05002440 inode->i_op = &btrfs_special_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -04002441 init_special_inode(inode, inode->i_mode, rdev);
2442 break;
Chris Mason39279cc2007-06-12 06:35:45 -04002443 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02002444
2445 btrfs_update_iflags(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002446 return;
2447
2448make_bad:
Chris Mason39279cc2007-06-12 06:35:45 -04002449 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002450 make_bad_inode(inode);
2451}
2452
Chris Masond352ac62008-09-29 15:18:18 -04002453/*
2454 * given a leaf and an inode, copy the inode fields into the leaf
2455 */
Chris Masone02119d2008-09-05 16:13:11 -04002456static void fill_inode_item(struct btrfs_trans_handle *trans,
2457 struct extent_buffer *leaf,
Chris Mason5f39d392007-10-15 16:14:19 -04002458 struct btrfs_inode_item *item,
Chris Mason39279cc2007-06-12 06:35:45 -04002459 struct inode *inode)
2460{
Chris Mason5f39d392007-10-15 16:14:19 -04002461 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2462 btrfs_set_inode_gid(leaf, item, inode->i_gid);
Chris Masondbe674a2008-07-17 12:54:05 -04002463 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
Chris Mason5f39d392007-10-15 16:14:19 -04002464 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2465 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2466
2467 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2468 inode->i_atime.tv_sec);
2469 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2470 inode->i_atime.tv_nsec);
2471
2472 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2473 inode->i_mtime.tv_sec);
2474 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2475 inode->i_mtime.tv_nsec);
2476
2477 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2478 inode->i_ctime.tv_sec);
2479 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2480 inode->i_ctime.tv_nsec);
2481
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002482 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
Chris Masone02119d2008-09-05 16:13:11 -04002483 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
Chris Masonc3027eb2008-12-08 16:40:21 -05002484 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
Chris Masone02119d2008-09-05 16:13:11 -04002485 btrfs_set_inode_transid(leaf, item, trans->transid);
Chris Mason5f39d392007-10-15 16:14:19 -04002486 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
Yanb98b6762008-01-08 15:54:37 -05002487 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
Josef Bacikd82a6f12011-05-11 15:26:06 -04002488 btrfs_set_inode_block_group(leaf, item, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04002489}
2490
Chris Masond352ac62008-09-29 15:18:18 -04002491/*
2492 * copy everything in the in-memory inode into the btree.
2493 */
Chris Mason21151332011-11-10 20:39:08 -05002494static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
Chris Masond3977122009-01-05 21:25:51 -05002495 struct btrfs_root *root, struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002496{
2497 struct btrfs_inode_item *inode_item;
2498 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002499 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002500 int ret;
2501
2502 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08002503 if (!path)
2504 return -ENOMEM;
2505
Chris Masonb9473432009-03-13 11:00:37 -04002506 path->leave_spinning = 1;
Miao Xie16cdcec2011-04-22 18:12:22 +08002507 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2508 1);
Chris Mason39279cc2007-06-12 06:35:45 -04002509 if (ret) {
2510 if (ret > 0)
2511 ret = -ENOENT;
2512 goto failed;
2513 }
2514
Chris Masonb4ce94d2009-02-04 09:25:08 -05002515 btrfs_unlock_up_safe(path, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04002516 leaf = path->nodes[0];
2517 inode_item = btrfs_item_ptr(leaf, path->slots[0],
Miao Xie16cdcec2011-04-22 18:12:22 +08002518 struct btrfs_inode_item);
Chris Mason39279cc2007-06-12 06:35:45 -04002519
Chris Masone02119d2008-09-05 16:13:11 -04002520 fill_inode_item(trans, leaf, inode_item, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04002521 btrfs_mark_buffer_dirty(leaf);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002522 btrfs_set_inode_last_trans(trans, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002523 ret = 0;
2524failed:
Chris Mason39279cc2007-06-12 06:35:45 -04002525 btrfs_free_path(path);
2526 return ret;
2527}
2528
Chris Masond352ac62008-09-29 15:18:18 -04002529/*
Chris Mason21151332011-11-10 20:39:08 -05002530 * copy everything in the in-memory inode into the btree.
2531 */
2532noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2533 struct btrfs_root *root, struct inode *inode)
2534{
2535 int ret;
2536
2537 /*
2538 * If the inode is a free space inode, we can deadlock during commit
2539 * if we put it into the delayed code.
2540 *
2541 * The data relocation inode should also be directly updated
2542 * without delay
2543 */
2544 if (!btrfs_is_free_space_inode(root, inode)
2545 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2546 ret = btrfs_delayed_update_inode(trans, root, inode);
2547 if (!ret)
2548 btrfs_set_inode_last_trans(trans, inode);
2549 return ret;
2550 }
2551
2552 return btrfs_update_inode_item(trans, root, inode);
2553}
2554
2555static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2556 struct btrfs_root *root, struct inode *inode)
2557{
2558 int ret;
2559
2560 ret = btrfs_update_inode(trans, root, inode);
2561 if (ret == -ENOSPC)
2562 return btrfs_update_inode_item(trans, root, inode);
2563 return ret;
2564}
2565
2566/*
Chris Masond352ac62008-09-29 15:18:18 -04002567 * unlink helper that gets used here in inode.c and in the tree logging
2568 * recovery code. It remove a link in a directory with a given name, and
2569 * also drops the back refs in the inode to the directory
2570 */
Al Viro92986792011-03-04 17:14:37 +00002571static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2572 struct btrfs_root *root,
2573 struct inode *dir, struct inode *inode,
2574 const char *name, int name_len)
Chris Mason39279cc2007-06-12 06:35:45 -04002575{
2576 struct btrfs_path *path;
Chris Mason39279cc2007-06-12 06:35:45 -04002577 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002578 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002579 struct btrfs_dir_item *di;
Chris Mason5f39d392007-10-15 16:14:19 -04002580 struct btrfs_key key;
Josef Bacikaec74772008-07-24 12:12:38 -04002581 u64 index;
Li Zefan33345d012011-04-20 10:31:50 +08002582 u64 ino = btrfs_ino(inode);
2583 u64 dir_ino = btrfs_ino(dir);
Chris Mason39279cc2007-06-12 06:35:45 -04002584
2585 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04002586 if (!path) {
2587 ret = -ENOMEM;
Tsutomu Itoh554233a2011-02-03 03:16:25 +00002588 goto out;
Chris Mason54aa1f42007-06-22 14:16:25 -04002589 }
2590
Chris Masonb9473432009-03-13 11:00:37 -04002591 path->leave_spinning = 1;
Li Zefan33345d012011-04-20 10:31:50 +08002592 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Chris Mason39279cc2007-06-12 06:35:45 -04002593 name, name_len, -1);
2594 if (IS_ERR(di)) {
2595 ret = PTR_ERR(di);
2596 goto err;
2597 }
2598 if (!di) {
2599 ret = -ENOENT;
2600 goto err;
2601 }
Chris Mason5f39d392007-10-15 16:14:19 -04002602 leaf = path->nodes[0];
2603 btrfs_dir_item_key_to_cpu(leaf, di, &key);
Chris Mason39279cc2007-06-12 06:35:45 -04002604 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason54aa1f42007-06-22 14:16:25 -04002605 if (ret)
2606 goto err;
David Sterbab3b4aa72011-04-21 01:20:15 +02002607 btrfs_release_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002608
Li Zefan33345d012011-04-20 10:31:50 +08002609 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2610 dir_ino, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04002611 if (ret) {
Chris Masond3977122009-01-05 21:25:51 -05002612 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
Li Zefan33345d012011-04-20 10:31:50 +08002613 "inode %llu parent %llu\n", name_len, name,
2614 (unsigned long long)ino, (unsigned long long)dir_ino);
Josef Bacikaec74772008-07-24 12:12:38 -04002615 goto err;
2616 }
2617
Miao Xie16cdcec2011-04-22 18:12:22 +08002618 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2619 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04002620 goto err;
Chris Mason39279cc2007-06-12 06:35:45 -04002621
Chris Masone02119d2008-09-05 16:13:11 -04002622 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
Li Zefan33345d012011-04-20 10:31:50 +08002623 inode, dir_ino);
Chris Mason49eb7e42008-09-11 15:53:12 -04002624 BUG_ON(ret != 0 && ret != -ENOENT);
Chris Masone02119d2008-09-05 16:13:11 -04002625
2626 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2627 dir, index);
Chris Mason6418c962010-10-30 07:34:24 -04002628 if (ret == -ENOENT)
2629 ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002630err:
2631 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002632 if (ret)
2633 goto out;
2634
2635 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2636 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2637 btrfs_update_inode(trans, root, dir);
Chris Masone02119d2008-09-05 16:13:11 -04002638out:
Chris Mason39279cc2007-06-12 06:35:45 -04002639 return ret;
2640}
2641
Al Viro92986792011-03-04 17:14:37 +00002642int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2643 struct btrfs_root *root,
2644 struct inode *dir, struct inode *inode,
2645 const char *name, int name_len)
2646{
2647 int ret;
2648 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2649 if (!ret) {
2650 btrfs_drop_nlink(inode);
2651 ret = btrfs_update_inode(trans, root, inode);
2652 }
2653 return ret;
2654}
2655
2656
Yan, Zhenga22285a2010-05-16 10:48:46 -04002657/* helper to check if there is any shared block in the path */
2658static int check_path_shared(struct btrfs_root *root,
2659 struct btrfs_path *path)
2660{
2661 struct extent_buffer *eb;
2662 int level;
Dan Carpenter0e4dcbe2010-06-01 08:23:11 +00002663 u64 refs = 1;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002664
2665 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
Josef Bacikdedefd72011-01-24 21:43:18 +00002666 int ret;
2667
Yan, Zhenga22285a2010-05-16 10:48:46 -04002668 if (!path->nodes[level])
2669 break;
2670 eb = path->nodes[level];
2671 if (!btrfs_block_can_be_shared(root, eb))
2672 continue;
2673 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2674 &refs, NULL);
2675 if (refs > 1)
2676 return 1;
2677 }
Josef Bacikdedefd72011-01-24 21:43:18 +00002678 return 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002679}
2680
2681/*
2682 * helper to start transaction for unlink and rmdir.
2683 *
2684 * unlink and rmdir are special in btrfs, they do not always free space.
2685 * so in enospc case, we should make sure they will free space before
2686 * allowing them to use the global metadata reservation.
2687 */
2688static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2689 struct dentry *dentry)
2690{
2691 struct btrfs_trans_handle *trans;
2692 struct btrfs_root *root = BTRFS_I(dir)->root;
2693 struct btrfs_path *path;
2694 struct btrfs_inode_ref *ref;
2695 struct btrfs_dir_item *di;
2696 struct inode *inode = dentry->d_inode;
2697 u64 index;
2698 int check_link = 1;
2699 int err = -ENOSPC;
2700 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08002701 u64 ino = btrfs_ino(inode);
2702 u64 dir_ino = btrfs_ino(dir);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002703
Josef Bacike70bea52011-10-11 14:18:24 -04002704 /*
2705 * 1 for the possible orphan item
2706 * 1 for the dir item
2707 * 1 for the dir index
2708 * 1 for the inode ref
2709 * 1 for the inode ref in the tree log
2710 * 2 for the dir entries in the log
2711 * 1 for the inode
2712 */
2713 trans = btrfs_start_transaction(root, 8);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002714 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2715 return trans;
2716
Li Zefan33345d012011-04-20 10:31:50 +08002717 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Yan, Zhenga22285a2010-05-16 10:48:46 -04002718 return ERR_PTR(-ENOSPC);
2719
2720 /* check if there is someone else holds reference */
2721 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2722 return ERR_PTR(-ENOSPC);
2723
2724 if (atomic_read(&inode->i_count) > 2)
2725 return ERR_PTR(-ENOSPC);
2726
2727 if (xchg(&root->fs_info->enospc_unlink, 1))
2728 return ERR_PTR(-ENOSPC);
2729
2730 path = btrfs_alloc_path();
2731 if (!path) {
2732 root->fs_info->enospc_unlink = 0;
2733 return ERR_PTR(-ENOMEM);
2734 }
2735
Josef Bacik3880a1b2011-10-14 14:46:51 -04002736 /* 1 for the orphan item */
2737 trans = btrfs_start_transaction(root, 1);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002738 if (IS_ERR(trans)) {
2739 btrfs_free_path(path);
2740 root->fs_info->enospc_unlink = 0;
2741 return trans;
2742 }
2743
2744 path->skip_locking = 1;
2745 path->search_commit_root = 1;
2746
2747 ret = btrfs_lookup_inode(trans, root, path,
2748 &BTRFS_I(dir)->location, 0);
2749 if (ret < 0) {
2750 err = ret;
2751 goto out;
2752 }
2753 if (ret == 0) {
2754 if (check_path_shared(root, path))
2755 goto out;
2756 } else {
2757 check_link = 0;
2758 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002759 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002760
2761 ret = btrfs_lookup_inode(trans, root, path,
2762 &BTRFS_I(inode)->location, 0);
2763 if (ret < 0) {
2764 err = ret;
2765 goto out;
2766 }
2767 if (ret == 0) {
2768 if (check_path_shared(root, path))
2769 goto out;
2770 } else {
2771 check_link = 0;
2772 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002773 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002774
2775 if (ret == 0 && S_ISREG(inode->i_mode)) {
2776 ret = btrfs_lookup_file_extent(trans, root, path,
Li Zefan33345d012011-04-20 10:31:50 +08002777 ino, (u64)-1, 0);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002778 if (ret < 0) {
2779 err = ret;
2780 goto out;
2781 }
2782 BUG_ON(ret == 0);
2783 if (check_path_shared(root, path))
2784 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +02002785 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002786 }
2787
2788 if (!check_link) {
2789 err = 0;
2790 goto out;
2791 }
2792
Li Zefan33345d012011-04-20 10:31:50 +08002793 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Yan, Zhenga22285a2010-05-16 10:48:46 -04002794 dentry->d_name.name, dentry->d_name.len, 0);
2795 if (IS_ERR(di)) {
2796 err = PTR_ERR(di);
2797 goto out;
2798 }
2799 if (di) {
2800 if (check_path_shared(root, path))
2801 goto out;
2802 } else {
2803 err = 0;
2804 goto out;
2805 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002806 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002807
2808 ref = btrfs_lookup_inode_ref(trans, root, path,
2809 dentry->d_name.name, dentry->d_name.len,
Li Zefan33345d012011-04-20 10:31:50 +08002810 ino, dir_ino, 0);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002811 if (IS_ERR(ref)) {
2812 err = PTR_ERR(ref);
2813 goto out;
2814 }
2815 BUG_ON(!ref);
2816 if (check_path_shared(root, path))
2817 goto out;
2818 index = btrfs_inode_ref_index(path->nodes[0], ref);
David Sterbab3b4aa72011-04-21 01:20:15 +02002819 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002820
Miao Xie16cdcec2011-04-22 18:12:22 +08002821 /*
2822 * This is a commit root search, if we can lookup inode item and other
2823 * relative items in the commit root, it means the transaction of
2824 * dir/file creation has been committed, and the dir index item that we
2825 * delay to insert has also been inserted into the commit root. So
2826 * we needn't worry about the delayed insertion of the dir index item
2827 * here.
2828 */
Li Zefan33345d012011-04-20 10:31:50 +08002829 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04002830 dentry->d_name.name, dentry->d_name.len, 0);
2831 if (IS_ERR(di)) {
2832 err = PTR_ERR(di);
2833 goto out;
2834 }
2835 BUG_ON(ret == -ENOENT);
2836 if (check_path_shared(root, path))
2837 goto out;
2838
2839 err = 0;
2840out:
2841 btrfs_free_path(path);
Josef Bacik3880a1b2011-10-14 14:46:51 -04002842 /* Migrate the orphan reservation over */
2843 if (!err)
2844 err = btrfs_block_rsv_migrate(trans->block_rsv,
2845 &root->fs_info->global_block_rsv,
Josef Bacik5a77d762011-11-01 14:32:23 -04002846 trans->bytes_reserved);
Josef Bacik3880a1b2011-10-14 14:46:51 -04002847
Yan, Zhenga22285a2010-05-16 10:48:46 -04002848 if (err) {
2849 btrfs_end_transaction(trans, root);
2850 root->fs_info->enospc_unlink = 0;
2851 return ERR_PTR(err);
2852 }
2853
2854 trans->block_rsv = &root->fs_info->global_block_rsv;
2855 return trans;
2856}
2857
2858static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2859 struct btrfs_root *root)
2860{
2861 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
Josef Bacik5a77d762011-11-01 14:32:23 -04002862 btrfs_block_rsv_release(root, trans->block_rsv,
2863 trans->bytes_reserved);
2864 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002865 BUG_ON(!root->fs_info->enospc_unlink);
2866 root->fs_info->enospc_unlink = 0;
2867 }
Josef Bacik7ad85bb2012-01-12 19:10:12 -05002868 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002869}
2870
Chris Mason39279cc2007-06-12 06:35:45 -04002871static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2872{
Yan, Zhenga22285a2010-05-16 10:48:46 -04002873 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04002874 struct btrfs_trans_handle *trans;
Josef Bacik7b128762008-07-24 12:17:14 -04002875 struct inode *inode = dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04002876 int ret;
Chris Mason1832a6d2007-12-21 16:27:21 -05002877 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002878
Yan, Zhenga22285a2010-05-16 10:48:46 -04002879 trans = __unlink_start_trans(dir, dentry);
2880 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002881 return PTR_ERR(trans);
Chris Mason5f39d392007-10-15 16:14:19 -04002882
Chris Mason12fcfd22009-03-24 10:24:20 -04002883 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2884
Chris Masone02119d2008-09-05 16:13:11 -04002885 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2886 dentry->d_name.name, dentry->d_name.len);
Tsutomu Itohb5324022011-07-19 07:27:20 +00002887 if (ret)
2888 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002889
Yan, Zhenga22285a2010-05-16 10:48:46 -04002890 if (inode->i_nlink == 0) {
Josef Bacik7b128762008-07-24 12:17:14 -04002891 ret = btrfs_orphan_add(trans, inode);
Tsutomu Itohb5324022011-07-19 07:27:20 +00002892 if (ret)
2893 goto out;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002894 }
Josef Bacik7b128762008-07-24 12:17:14 -04002895
Tsutomu Itohb5324022011-07-19 07:27:20 +00002896out:
Chris Masond3c2fdc2007-09-17 10:58:06 -04002897 nr = trans->blocks_used;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002898 __unlink_end_trans(trans, root);
Chris Masond3c2fdc2007-09-17 10:58:06 -04002899 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04002900 return ret;
2901}
2902
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002903int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2904 struct btrfs_root *root,
2905 struct inode *dir, u64 objectid,
2906 const char *name, int name_len)
2907{
2908 struct btrfs_path *path;
2909 struct extent_buffer *leaf;
2910 struct btrfs_dir_item *di;
2911 struct btrfs_key key;
2912 u64 index;
2913 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08002914 u64 dir_ino = btrfs_ino(dir);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002915
2916 path = btrfs_alloc_path();
2917 if (!path)
2918 return -ENOMEM;
2919
Li Zefan33345d012011-04-20 10:31:50 +08002920 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002921 name, name_len, -1);
David Sterbac7040052011-04-19 18:00:01 +02002922 BUG_ON(IS_ERR_OR_NULL(di));
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002923
2924 leaf = path->nodes[0];
2925 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2926 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2927 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2928 BUG_ON(ret);
David Sterbab3b4aa72011-04-21 01:20:15 +02002929 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002930
2931 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2932 objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08002933 dir_ino, &index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002934 if (ret < 0) {
2935 BUG_ON(ret != -ENOENT);
Li Zefan33345d012011-04-20 10:31:50 +08002936 di = btrfs_search_dir_index_item(root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002937 name, name_len);
David Sterbac7040052011-04-19 18:00:01 +02002938 BUG_ON(IS_ERR_OR_NULL(di));
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002939
2940 leaf = path->nodes[0];
2941 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02002942 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002943 index = key.offset;
2944 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002945 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002946
Miao Xie16cdcec2011-04-22 18:12:22 +08002947 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2948 BUG_ON(ret);
2949
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002950 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2951 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2952 ret = btrfs_update_inode(trans, root, dir);
2953 BUG_ON(ret);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002954
Josef Bacik71d7aed2011-06-14 14:24:32 -04002955 btrfs_free_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002956 return 0;
2957}
2958
Chris Mason39279cc2007-06-12 06:35:45 -04002959static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2960{
2961 struct inode *inode = dentry->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05002962 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002963 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04002964 struct btrfs_trans_handle *trans;
Chris Mason1832a6d2007-12-21 16:27:21 -05002965 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002966
Chris Mason3394e162008-11-17 20:42:26 -05002967 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
Li Zefan33345d012011-04-20 10:31:50 +08002968 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
Yan134d4512007-10-25 15:49:25 -04002969 return -ENOTEMPTY;
2970
Yan, Zhenga22285a2010-05-16 10:48:46 -04002971 trans = __unlink_start_trans(dir, dentry);
2972 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002973 return PTR_ERR(trans);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002974
Li Zefan33345d012011-04-20 10:31:50 +08002975 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002976 err = btrfs_unlink_subvol(trans, root, dir,
2977 BTRFS_I(inode)->location.objectid,
2978 dentry->d_name.name,
2979 dentry->d_name.len);
2980 goto out;
2981 }
2982
Josef Bacik7b128762008-07-24 12:17:14 -04002983 err = btrfs_orphan_add(trans, inode);
2984 if (err)
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002985 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002986
Chris Mason39279cc2007-06-12 06:35:45 -04002987 /* now the directory is empty */
Chris Masone02119d2008-09-05 16:13:11 -04002988 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2989 dentry->d_name.name, dentry->d_name.len);
Chris Masond3977122009-01-05 21:25:51 -05002990 if (!err)
Chris Masondbe674a2008-07-17 12:54:05 -04002991 btrfs_i_size_write(inode, 0);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002992out:
Chris Masond3c2fdc2007-09-17 10:58:06 -04002993 nr = trans->blocks_used;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002994 __unlink_end_trans(trans, root);
Chris Masond3c2fdc2007-09-17 10:58:06 -04002995 btrfs_btree_balance_dirty(root, nr);
Chris Mason39544012007-12-12 14:38:19 -05002996
Chris Mason39279cc2007-06-12 06:35:45 -04002997 return err;
2998}
2999
Chris Mason323ac952008-10-01 19:05:46 -04003000/*
Chris Mason39279cc2007-06-12 06:35:45 -04003001 * this can truncate away extent items, csum items and directory items.
3002 * It starts at a high offset and removes keys until it can't find
Chris Masond352ac62008-09-29 15:18:18 -04003003 * any higher than new_size
Chris Mason39279cc2007-06-12 06:35:45 -04003004 *
3005 * csum items that cross the new i_size are truncated to the new size
3006 * as well.
Josef Bacik7b128762008-07-24 12:17:14 -04003007 *
3008 * min_type is the minimum key type to truncate down to. If set to 0, this
3009 * will kill all the items on this inode, including the INODE_ITEM_KEY.
Chris Mason39279cc2007-06-12 06:35:45 -04003010 */
Yan, Zheng80825102009-11-12 09:35:36 +00003011int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3012 struct btrfs_root *root,
3013 struct inode *inode,
3014 u64 new_size, u32 min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003015{
Chris Mason39279cc2007-06-12 06:35:45 -04003016 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003017 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003018 struct btrfs_file_extent_item *fi;
Yan, Zheng80825102009-11-12 09:35:36 +00003019 struct btrfs_key key;
3020 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04003021 u64 extent_start = 0;
Chris Masondb945352007-10-15 16:15:53 -04003022 u64 extent_num_bytes = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003023 u64 extent_offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003024 u64 item_end = 0;
Yan, Zheng80825102009-11-12 09:35:36 +00003025 u64 mask = root->sectorsize - 1;
3026 u32 found_type = (u8)-1;
Chris Mason39279cc2007-06-12 06:35:45 -04003027 int found_extent;
3028 int del_item;
Chris Mason85e21ba2008-01-29 15:11:36 -05003029 int pending_del_nr = 0;
3030 int pending_del_slot = 0;
Chris Mason179e29e2007-11-01 11:28:41 -04003031 int extent_type = -1;
Yan, Zheng80825102009-11-12 09:35:36 +00003032 int ret;
3033 int err = 0;
Li Zefan33345d012011-04-20 10:31:50 +08003034 u64 ino = btrfs_ino(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003035
3036 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
Chris Mason39279cc2007-06-12 06:35:45 -04003037
Mark Fasheh0eb0e192011-07-12 16:44:10 -07003038 path = btrfs_alloc_path();
3039 if (!path)
3040 return -ENOMEM;
3041 path->reada = -1;
3042
Josef Bacik0af3d002010-06-21 14:48:16 -04003043 if (root->ref_cows || root == root->fs_info->tree_root)
Zheng Yan5b21f2e2008-09-26 10:05:38 -04003044 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003045
Miao Xie16cdcec2011-04-22 18:12:22 +08003046 /*
3047 * This function is also used to drop the items in the log tree before
3048 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3049 * it is used to drop the loged items. So we shouldn't kill the delayed
3050 * items.
3051 */
3052 if (min_type == 0 && root == BTRFS_I(inode)->root)
3053 btrfs_kill_delayed_inode_items(inode);
3054
Li Zefan33345d012011-04-20 10:31:50 +08003055 key.objectid = ino;
Chris Mason39279cc2007-06-12 06:35:45 -04003056 key.offset = (u64)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04003057 key.type = (u8)-1;
3058
Chris Mason85e21ba2008-01-29 15:11:36 -05003059search_again:
Chris Masonb9473432009-03-13 11:00:37 -04003060 path->leave_spinning = 1;
Chris Mason85e21ba2008-01-29 15:11:36 -05003061 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Yan, Zheng80825102009-11-12 09:35:36 +00003062 if (ret < 0) {
3063 err = ret;
3064 goto out;
3065 }
Chris Masond3977122009-01-05 21:25:51 -05003066
Chris Mason85e21ba2008-01-29 15:11:36 -05003067 if (ret > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04003068 /* there are no items in the tree for us to truncate, we're
3069 * done
3070 */
Yan, Zheng80825102009-11-12 09:35:36 +00003071 if (path->slots[0] == 0)
3072 goto out;
Chris Mason85e21ba2008-01-29 15:11:36 -05003073 path->slots[0]--;
3074 }
3075
Chris Masond3977122009-01-05 21:25:51 -05003076 while (1) {
Chris Mason39279cc2007-06-12 06:35:45 -04003077 fi = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -04003078 leaf = path->nodes[0];
3079 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3080 found_type = btrfs_key_type(&found_key);
Chris Mason39279cc2007-06-12 06:35:45 -04003081
Li Zefan33345d012011-04-20 10:31:50 +08003082 if (found_key.objectid != ino)
Chris Mason39279cc2007-06-12 06:35:45 -04003083 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003084
Chris Mason85e21ba2008-01-29 15:11:36 -05003085 if (found_type < min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003086 break;
3087
Chris Mason5f39d392007-10-15 16:14:19 -04003088 item_end = found_key.offset;
Chris Mason39279cc2007-06-12 06:35:45 -04003089 if (found_type == BTRFS_EXTENT_DATA_KEY) {
Chris Mason5f39d392007-10-15 16:14:19 -04003090 fi = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04003091 struct btrfs_file_extent_item);
Chris Mason179e29e2007-11-01 11:28:41 -04003092 extent_type = btrfs_file_extent_type(leaf, fi);
3093 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04003094 item_end +=
Chris Masondb945352007-10-15 16:15:53 -04003095 btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason179e29e2007-11-01 11:28:41 -04003096 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason179e29e2007-11-01 11:28:41 -04003097 item_end += btrfs_file_extent_inline_len(leaf,
Chris Masonc8b97812008-10-29 14:49:59 -04003098 fi);
Chris Mason39279cc2007-06-12 06:35:45 -04003099 }
Yan008630c2007-11-07 13:31:09 -05003100 item_end--;
Chris Mason39279cc2007-06-12 06:35:45 -04003101 }
Yan, Zheng80825102009-11-12 09:35:36 +00003102 if (found_type > min_type) {
Chris Mason39279cc2007-06-12 06:35:45 -04003103 del_item = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00003104 } else {
3105 if (item_end < new_size)
3106 break;
3107 if (found_key.offset >= new_size)
3108 del_item = 1;
3109 else
3110 del_item = 0;
3111 }
Chris Mason39279cc2007-06-12 06:35:45 -04003112 found_extent = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003113 /* FIXME, shrink the extent if the ref count is only 1 */
Chris Mason179e29e2007-11-01 11:28:41 -04003114 if (found_type != BTRFS_EXTENT_DATA_KEY)
3115 goto delete;
3116
3117 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason39279cc2007-06-12 06:35:45 -04003118 u64 num_dec;
Chris Masondb945352007-10-15 16:15:53 -04003119 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
Miao Xief70a9a62012-01-12 19:10:12 -05003120 if (!del_item) {
Chris Masondb945352007-10-15 16:15:53 -04003121 u64 orig_num_bytes =
3122 btrfs_file_extent_num_bytes(leaf, fi);
Chris Masone02119d2008-09-05 16:13:11 -04003123 extent_num_bytes = new_size -
Chris Mason5f39d392007-10-15 16:14:19 -04003124 found_key.offset + root->sectorsize - 1;
Yanb1632b12008-01-30 11:54:04 -05003125 extent_num_bytes = extent_num_bytes &
3126 ~((u64)root->sectorsize - 1);
Chris Masondb945352007-10-15 16:15:53 -04003127 btrfs_set_file_extent_num_bytes(leaf, fi,
3128 extent_num_bytes);
3129 num_dec = (orig_num_bytes -
Chris Mason90692182008-02-08 13:49:28 -05003130 extent_num_bytes);
Chris Masone02119d2008-09-05 16:13:11 -04003131 if (root->ref_cows && extent_start != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003132 inode_sub_bytes(inode, num_dec);
Chris Mason5f39d392007-10-15 16:14:19 -04003133 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04003134 } else {
Chris Masondb945352007-10-15 16:15:53 -04003135 extent_num_bytes =
3136 btrfs_file_extent_disk_num_bytes(leaf,
3137 fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003138 extent_offset = found_key.offset -
3139 btrfs_file_extent_offset(leaf, fi);
3140
Chris Mason39279cc2007-06-12 06:35:45 -04003141 /* FIXME blocksize != 4096 */
Chris Mason90692182008-02-08 13:49:28 -05003142 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason39279cc2007-06-12 06:35:45 -04003143 if (extent_start != 0) {
3144 found_extent = 1;
Chris Masone02119d2008-09-05 16:13:11 -04003145 if (root->ref_cows)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003146 inode_sub_bytes(inode, num_dec);
Chris Mason39279cc2007-06-12 06:35:45 -04003147 }
3148 }
Chris Mason90692182008-02-08 13:49:28 -05003149 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -04003150 /*
3151 * we can't truncate inline items that have had
3152 * special encodings
3153 */
3154 if (!del_item &&
3155 btrfs_file_extent_compression(leaf, fi) == 0 &&
3156 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3157 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04003158 u32 size = new_size - found_key.offset;
3159
3160 if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003161 inode_sub_bytes(inode, item_end + 1 -
3162 new_size);
Chris Masone02119d2008-09-05 16:13:11 -04003163 }
3164 size =
3165 btrfs_file_extent_calc_inline_size(size);
Jeff Mahoney143bede2012-03-01 14:56:26 +01003166 btrfs_truncate_item(trans, root, path,
3167 size, 1);
Chris Masone02119d2008-09-05 16:13:11 -04003168 } else if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003169 inode_sub_bytes(inode, item_end + 1 -
3170 found_key.offset);
Chris Mason90692182008-02-08 13:49:28 -05003171 }
Chris Mason39279cc2007-06-12 06:35:45 -04003172 }
Chris Mason179e29e2007-11-01 11:28:41 -04003173delete:
Chris Mason39279cc2007-06-12 06:35:45 -04003174 if (del_item) {
Chris Mason85e21ba2008-01-29 15:11:36 -05003175 if (!pending_del_nr) {
3176 /* no pending yet, add ourselves */
3177 pending_del_slot = path->slots[0];
3178 pending_del_nr = 1;
3179 } else if (pending_del_nr &&
3180 path->slots[0] + 1 == pending_del_slot) {
3181 /* hop on the pending chunk */
3182 pending_del_nr++;
3183 pending_del_slot = path->slots[0];
3184 } else {
Chris Masond3977122009-01-05 21:25:51 -05003185 BUG();
Chris Mason85e21ba2008-01-29 15:11:36 -05003186 }
Chris Mason39279cc2007-06-12 06:35:45 -04003187 } else {
3188 break;
3189 }
Josef Bacik0af3d002010-06-21 14:48:16 -04003190 if (found_extent && (root->ref_cows ||
3191 root == root->fs_info->tree_root)) {
Chris Masonb9473432009-03-13 11:00:37 -04003192 btrfs_set_path_blocking(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003193 ret = btrfs_free_extent(trans, root, extent_start,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003194 extent_num_bytes, 0,
3195 btrfs_header_owner(leaf),
Arne Jansen66d7e7f2011-09-12 15:26:38 +02003196 ino, extent_offset, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04003197 BUG_ON(ret);
3198 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003199
Yan, Zheng80825102009-11-12 09:35:36 +00003200 if (found_type == BTRFS_INODE_ITEM_KEY)
3201 break;
3202
3203 if (path->slots[0] == 0 ||
3204 path->slots[0] != pending_del_slot) {
Li Zefan82d59022011-04-20 10:33:24 +08003205 if (root->ref_cows &&
3206 BTRFS_I(inode)->location.objectid !=
3207 BTRFS_FREE_INO_OBJECTID) {
Yan, Zheng80825102009-11-12 09:35:36 +00003208 err = -EAGAIN;
3209 goto out;
3210 }
3211 if (pending_del_nr) {
3212 ret = btrfs_del_items(trans, root, path,
3213 pending_del_slot,
3214 pending_del_nr);
3215 BUG_ON(ret);
3216 pending_del_nr = 0;
3217 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003218 btrfs_release_path(path);
Chris Mason85e21ba2008-01-29 15:11:36 -05003219 goto search_again;
Yan, Zheng80825102009-11-12 09:35:36 +00003220 } else {
3221 path->slots[0]--;
Chris Mason85e21ba2008-01-29 15:11:36 -05003222 }
Chris Mason39279cc2007-06-12 06:35:45 -04003223 }
Yan, Zheng80825102009-11-12 09:35:36 +00003224out:
Chris Mason85e21ba2008-01-29 15:11:36 -05003225 if (pending_del_nr) {
3226 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3227 pending_del_nr);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003228 BUG_ON(ret);
Chris Mason85e21ba2008-01-29 15:11:36 -05003229 }
Chris Mason39279cc2007-06-12 06:35:45 -04003230 btrfs_free_path(path);
Yan, Zheng80825102009-11-12 09:35:36 +00003231 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003232}
3233
Chris Masona52d9a82007-08-27 16:49:44 -04003234/*
3235 * taken from block_truncate_page, but does cow as it zeros out
3236 * any bytes left in the last page in the file.
3237 */
3238static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3239{
3240 struct inode *inode = mapping->host;
Chris Masondb945352007-10-15 16:15:53 -04003241 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003242 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3243 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003244 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003245 char *kaddr;
Chris Masondb945352007-10-15 16:15:53 -04003246 u32 blocksize = root->sectorsize;
Chris Masona52d9a82007-08-27 16:49:44 -04003247 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3248 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3249 struct page *page;
Josef Bacik3b16a4e2011-09-21 15:05:58 -04003250 gfp_t mask = btrfs_alloc_write_mask(mapping);
Chris Masona52d9a82007-08-27 16:49:44 -04003251 int ret = 0;
3252 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003253 u64 page_end;
Chris Masona52d9a82007-08-27 16:49:44 -04003254
3255 if ((offset & (blocksize - 1)) == 0)
3256 goto out;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003257 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003258 if (ret)
3259 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04003260
3261 ret = -ENOMEM;
Chris Mason211c17f2008-05-15 09:13:45 -04003262again:
Josef Bacik3b16a4e2011-09-21 15:05:58 -04003263 page = find_or_create_page(mapping, index, mask);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003264 if (!page) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003265 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Masona52d9a82007-08-27 16:49:44 -04003266 goto out;
Josef Bacik5d5e1032009-10-13 16:46:49 -04003267 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04003268
3269 page_start = page_offset(page);
3270 page_end = page_start + PAGE_CACHE_SIZE - 1;
3271
Chris Masona52d9a82007-08-27 16:49:44 -04003272 if (!PageUptodate(page)) {
3273 ret = btrfs_readpage(NULL, page);
3274 lock_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04003275 if (page->mapping != mapping) {
3276 unlock_page(page);
3277 page_cache_release(page);
3278 goto again;
3279 }
Chris Masona52d9a82007-08-27 16:49:44 -04003280 if (!PageUptodate(page)) {
3281 ret = -EIO;
Chris Mason89642222008-07-24 09:41:53 -04003282 goto out_unlock;
Chris Masona52d9a82007-08-27 16:49:44 -04003283 }
3284 }
Chris Mason211c17f2008-05-15 09:13:45 -04003285 wait_on_page_writeback(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003286
Jeff Mahoneyd0082372012-03-01 14:57:19 +01003287 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003288 set_page_extent_mapped(page);
3289
3290 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3291 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00003292 unlock_extent_cached(io_tree, page_start, page_end,
3293 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003294 unlock_page(page);
3295 page_cache_release(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04003296 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003297 btrfs_put_ordered_extent(ordered);
3298 goto again;
3299 }
3300
Josef Bacik2ac55d42010-02-03 19:33:23 +00003301 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Josef Bacik5d5e1032009-10-13 16:46:49 -04003302 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
Josef Bacik2ac55d42010-02-03 19:33:23 +00003303 0, 0, &cached_state, GFP_NOFS);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003304
Josef Bacik2ac55d42010-02-03 19:33:23 +00003305 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3306 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003307 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00003308 unlock_extent_cached(io_tree, page_start, page_end,
3309 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003310 goto out_unlock;
3311 }
3312
Chris Masone6dcd2d2008-07-17 12:53:50 -04003313 ret = 0;
3314 if (offset != PAGE_CACHE_SIZE) {
3315 kaddr = kmap(page);
3316 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3317 flush_dcache_page(page);
3318 kunmap(page);
3319 }
Chris Mason247e7432008-07-17 12:53:51 -04003320 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003321 set_page_dirty(page);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003322 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3323 GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04003324
Chris Mason89642222008-07-24 09:41:53 -04003325out_unlock:
Josef Bacik5d5e1032009-10-13 16:46:49 -04003326 if (ret)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003327 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason39279cc2007-06-12 06:35:45 -04003328 unlock_page(page);
3329 page_cache_release(page);
3330out:
3331 return ret;
3332}
3333
Josef Bacik695a0d02011-03-04 15:46:53 -05003334/*
3335 * This function puts in dummy file extents for the area we're creating a hole
3336 * for. So if we are truncating this file to a larger size we need to insert
3337 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3338 * the range between oldsize and size
3339 */
Josef Bacika41ad392011-01-31 15:30:16 -05003340int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
Yan Zheng9036c102008-10-30 14:19:41 -04003341{
3342 struct btrfs_trans_handle *trans;
3343 struct btrfs_root *root = BTRFS_I(inode)->root;
3344 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003345 struct extent_map *em = NULL;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003346 struct extent_state *cached_state = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04003347 u64 mask = root->sectorsize - 1;
Josef Bacika41ad392011-01-31 15:30:16 -05003348 u64 hole_start = (oldsize + mask) & ~mask;
Yan Zheng9036c102008-10-30 14:19:41 -04003349 u64 block_end = (size + mask) & ~mask;
3350 u64 last_byte;
3351 u64 cur_offset;
3352 u64 hole_size;
Josef Bacik9ed74f22009-09-11 16:12:44 -04003353 int err = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04003354
3355 if (size <= hole_start)
3356 return 0;
3357
Yan Zheng9036c102008-10-30 14:19:41 -04003358 while (1) {
3359 struct btrfs_ordered_extent *ordered;
3360 btrfs_wait_ordered_range(inode, hole_start,
3361 block_end - hole_start);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003362 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01003363 &cached_state);
Yan Zheng9036c102008-10-30 14:19:41 -04003364 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3365 if (!ordered)
3366 break;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003367 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3368 &cached_state, GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003369 btrfs_put_ordered_extent(ordered);
3370 }
3371
Yan Zheng9036c102008-10-30 14:19:41 -04003372 cur_offset = hole_start;
3373 while (1) {
3374 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3375 block_end - cur_offset, 0);
David Sterbac7040052011-04-19 18:00:01 +02003376 BUG_ON(IS_ERR_OR_NULL(em));
Yan Zheng9036c102008-10-30 14:19:41 -04003377 last_byte = min(extent_map_end(em), block_end);
3378 last_byte = (last_byte + mask) & ~mask;
Yan, Zheng80825102009-11-12 09:35:36 +00003379 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
Chris Mason771ed682008-11-06 22:02:51 -05003380 u64 hint_byte = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04003381 hole_size = last_byte - cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00003382
Miao Xie36423202011-12-14 20:12:02 -05003383 trans = btrfs_start_transaction(root, 3);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003384 if (IS_ERR(trans)) {
3385 err = PTR_ERR(trans);
Chris Mason771ed682008-11-06 22:02:51 -05003386 break;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003387 }
Yan, Zheng80825102009-11-12 09:35:36 +00003388
3389 err = btrfs_drop_extents(trans, inode, cur_offset,
3390 cur_offset + hole_size,
3391 &hint_byte, 1);
Miao Xie5b397372011-09-11 10:52:24 -04003392 if (err) {
Miao Xie36423202011-12-14 20:12:02 -05003393 btrfs_update_inode(trans, root, inode);
Miao Xie5b397372011-09-11 10:52:24 -04003394 btrfs_end_transaction(trans, root);
Josef Bacik3893e332011-01-31 16:03:11 -05003395 break;
Miao Xie5b397372011-09-11 10:52:24 -04003396 }
Josef Bacik9ed74f22009-09-11 16:12:44 -04003397
Yan Zheng9036c102008-10-30 14:19:41 -04003398 err = btrfs_insert_file_extent(trans, root,
Li Zefan33345d012011-04-20 10:31:50 +08003399 btrfs_ino(inode), cur_offset, 0,
Yan Zheng9036c102008-10-30 14:19:41 -04003400 0, hole_size, 0, hole_size,
3401 0, 0, 0);
Miao Xie5b397372011-09-11 10:52:24 -04003402 if (err) {
Miao Xie36423202011-12-14 20:12:02 -05003403 btrfs_update_inode(trans, root, inode);
Miao Xie5b397372011-09-11 10:52:24 -04003404 btrfs_end_transaction(trans, root);
Josef Bacik3893e332011-01-31 16:03:11 -05003405 break;
Miao Xie5b397372011-09-11 10:52:24 -04003406 }
Yan, Zheng80825102009-11-12 09:35:36 +00003407
Yan Zheng9036c102008-10-30 14:19:41 -04003408 btrfs_drop_extent_cache(inode, hole_start,
3409 last_byte - 1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003410
Miao Xie36423202011-12-14 20:12:02 -05003411 btrfs_update_inode(trans, root, inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003412 btrfs_end_transaction(trans, root);
Yan Zheng9036c102008-10-30 14:19:41 -04003413 }
3414 free_extent_map(em);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003415 em = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04003416 cur_offset = last_byte;
Yan, Zheng80825102009-11-12 09:35:36 +00003417 if (cur_offset >= block_end)
Yan Zheng9036c102008-10-30 14:19:41 -04003418 break;
3419 }
3420
Yan, Zhenga22285a2010-05-16 10:48:46 -04003421 free_extent_map(em);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003422 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3423 GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003424 return err;
3425}
3426
Josef Bacika41ad392011-01-31 15:30:16 -05003427static int btrfs_setsize(struct inode *inode, loff_t newsize)
Yan, Zheng80825102009-11-12 09:35:36 +00003428{
Miao Xief4a2f4c2011-12-14 20:12:01 -05003429 struct btrfs_root *root = BTRFS_I(inode)->root;
3430 struct btrfs_trans_handle *trans;
Josef Bacika41ad392011-01-31 15:30:16 -05003431 loff_t oldsize = i_size_read(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003432 int ret;
3433
Josef Bacika41ad392011-01-31 15:30:16 -05003434 if (newsize == oldsize)
Yan, Zheng80825102009-11-12 09:35:36 +00003435 return 0;
3436
Josef Bacika41ad392011-01-31 15:30:16 -05003437 if (newsize > oldsize) {
Josef Bacika41ad392011-01-31 15:30:16 -05003438 truncate_pagecache(inode, oldsize, newsize);
3439 ret = btrfs_cont_expand(inode, oldsize, newsize);
Miao Xief4a2f4c2011-12-14 20:12:01 -05003440 if (ret)
Yan, Zheng80825102009-11-12 09:35:36 +00003441 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00003442
Miao Xief4a2f4c2011-12-14 20:12:01 -05003443 trans = btrfs_start_transaction(root, 1);
3444 if (IS_ERR(trans))
3445 return PTR_ERR(trans);
3446
3447 i_size_write(inode, newsize);
3448 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3449 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik7ad85bb2012-01-12 19:10:12 -05003450 btrfs_end_transaction(trans, root);
Josef Bacika41ad392011-01-31 15:30:16 -05003451 } else {
Yan, Zheng80825102009-11-12 09:35:36 +00003452
Josef Bacika41ad392011-01-31 15:30:16 -05003453 /*
3454 * We're truncating a file that used to have good data down to
3455 * zero. Make sure it gets into the ordered flush list so that
3456 * any new writes get down to disk quickly.
3457 */
3458 if (newsize == 0)
3459 BTRFS_I(inode)->ordered_data_close = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00003460
Josef Bacika41ad392011-01-31 15:30:16 -05003461 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3462 truncate_setsize(inode, newsize);
3463 ret = btrfs_truncate(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003464 }
3465
Josef Bacika41ad392011-01-31 15:30:16 -05003466 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00003467}
3468
Chris Mason39279cc2007-06-12 06:35:45 -04003469static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3470{
3471 struct inode *inode = dentry->d_inode;
Li Zefanb83cc962010-12-20 16:04:08 +08003472 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003473 int err;
3474
Li Zefanb83cc962010-12-20 16:04:08 +08003475 if (btrfs_root_readonly(root))
3476 return -EROFS;
3477
Chris Mason39279cc2007-06-12 06:35:45 -04003478 err = inode_change_ok(inode, attr);
3479 if (err)
3480 return err;
3481
Chris Mason5a3f23d2009-03-31 13:27:11 -04003482 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Josef Bacika41ad392011-01-31 15:30:16 -05003483 err = btrfs_setsize(inode, attr->ia_size);
Yan, Zheng80825102009-11-12 09:35:36 +00003484 if (err)
3485 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003486 }
Yan Zheng9036c102008-10-30 14:19:41 -04003487
Christoph Hellwig10257742010-06-04 11:30:02 +02003488 if (attr->ia_valid) {
3489 setattr_copy(inode, attr);
Josef Bacik22c44fe2011-11-30 10:45:38 -05003490 err = btrfs_dirty_inode(inode);
Josef Bacik33268ea2008-07-24 12:16:36 -04003491
Josef Bacik22c44fe2011-11-30 10:45:38 -05003492 if (!err && attr->ia_valid & ATTR_MODE)
Christoph Hellwig10257742010-06-04 11:30:02 +02003493 err = btrfs_acl_chmod(inode);
3494 }
3495
Chris Mason39279cc2007-06-12 06:35:45 -04003496 return err;
3497}
Chris Mason61295eb2008-01-14 16:24:38 -05003498
Al Virobd555972010-06-07 11:35:40 -04003499void btrfs_evict_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003500{
3501 struct btrfs_trans_handle *trans;
3502 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik726c35f2011-09-26 15:46:06 -04003503 struct btrfs_block_rsv *rsv, *global_rsv;
Josef Bacik07127182011-08-19 10:29:59 -04003504 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Masond3c2fdc2007-09-17 10:58:06 -04003505 unsigned long nr;
Chris Mason39279cc2007-06-12 06:35:45 -04003506 int ret;
3507
liubo1abe9b82011-03-24 11:18:59 +00003508 trace_btrfs_inode_evict(inode);
3509
Chris Mason39279cc2007-06-12 06:35:45 -04003510 truncate_inode_pages(&inode->i_data, 0);
Josef Bacik0af3d002010-06-21 14:48:16 -04003511 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
Chris Mason2cf85722011-07-26 15:35:09 -04003512 btrfs_is_free_space_inode(root, inode)))
Al Virobd555972010-06-07 11:35:40 -04003513 goto no_delete;
3514
Chris Mason39279cc2007-06-12 06:35:45 -04003515 if (is_bad_inode(inode)) {
Josef Bacik7b128762008-07-24 12:17:14 -04003516 btrfs_orphan_del(NULL, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003517 goto no_delete;
3518 }
Al Virobd555972010-06-07 11:35:40 -04003519 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
Chris Mason4a096752008-07-21 10:29:44 -04003520 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Chris Mason5f39d392007-10-15 16:14:19 -04003521
Yan, Zhengc71bf092009-11-12 09:34:40 +00003522 if (root->fs_info->log_root_recovering) {
3523 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3524 goto no_delete;
3525 }
3526
Yan, Zheng76dda932009-09-21 16:00:26 -04003527 if (inode->i_nlink > 0) {
3528 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3529 goto no_delete;
3530 }
3531
Josef Bacik4289a662011-08-05 13:22:24 -04003532 rsv = btrfs_alloc_block_rsv(root);
3533 if (!rsv) {
3534 btrfs_orphan_del(NULL, inode);
3535 goto no_delete;
3536 }
Josef Bacik4a338542011-08-29 11:01:31 -04003537 rsv->size = min_size;
Josef Bacik726c35f2011-09-26 15:46:06 -04003538 global_rsv = &root->fs_info->global_block_rsv;
Josef Bacik4289a662011-08-05 13:22:24 -04003539
Chris Masondbe674a2008-07-17 12:54:05 -04003540 btrfs_i_size_write(inode, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04003541
Josef Bacik4289a662011-08-05 13:22:24 -04003542 /*
3543 * This is a bit simpler than btrfs_truncate since
3544 *
3545 * 1) We've already reserved our space for our orphan item in the
3546 * unlink.
3547 * 2) We're going to delete the inode item, so we don't need to update
3548 * it at all.
3549 *
3550 * So we just need to reserve some slack space in case we add bytes when
3551 * doing the truncate.
3552 */
Yan, Zheng80825102009-11-12 09:35:36 +00003553 while (1) {
Miao Xieaa38a712011-11-18 17:43:00 +08003554 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
Yan, Zheng80825102009-11-12 09:35:36 +00003555
Josef Bacik726c35f2011-09-26 15:46:06 -04003556 /*
3557 * Try and steal from the global reserve since we will
3558 * likely not use this space anyway, we want to try as
3559 * hard as possible to get this to work.
3560 */
3561 if (ret)
3562 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3563
Yan, Zhengd68fc572010-05-16 10:49:58 -04003564 if (ret) {
Josef Bacik4289a662011-08-05 13:22:24 -04003565 printk(KERN_WARNING "Could not get space for a "
Josef Bacik482e6dc2011-08-19 10:31:56 -04003566 "delete, will truncate on mount %d\n", ret);
Josef Bacik4289a662011-08-05 13:22:24 -04003567 btrfs_orphan_del(NULL, inode);
3568 btrfs_free_block_rsv(root, rsv);
3569 goto no_delete;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003570 }
3571
Josef Bacik4289a662011-08-05 13:22:24 -04003572 trans = btrfs_start_transaction(root, 0);
3573 if (IS_ERR(trans)) {
3574 btrfs_orphan_del(NULL, inode);
3575 btrfs_free_block_rsv(root, rsv);
3576 goto no_delete;
3577 }
3578
3579 trans->block_rsv = rsv;
3580
Yan, Zhengd68fc572010-05-16 10:49:58 -04003581 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003582 if (ret != -EAGAIN)
3583 break;
3584
3585 nr = trans->blocks_used;
3586 btrfs_end_transaction(trans, root);
3587 trans = NULL;
3588 btrfs_btree_balance_dirty(root, nr);
Josef Bacik7b128762008-07-24 12:17:14 -04003589 }
3590
Josef Bacik4289a662011-08-05 13:22:24 -04003591 btrfs_free_block_rsv(root, rsv);
3592
Yan, Zheng80825102009-11-12 09:35:36 +00003593 if (ret == 0) {
Josef Bacik4289a662011-08-05 13:22:24 -04003594 trans->block_rsv = root->orphan_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00003595 ret = btrfs_orphan_del(trans, inode);
3596 BUG_ON(ret);
3597 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003598
Josef Bacik4289a662011-08-05 13:22:24 -04003599 trans->block_rsv = &root->fs_info->trans_block_rsv;
Li Zefan581bb052011-04-20 10:06:11 +08003600 if (!(root == root->fs_info->tree_root ||
3601 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
Li Zefan33345d012011-04-20 10:31:50 +08003602 btrfs_return_ino(root, btrfs_ino(inode));
Li Zefan581bb052011-04-20 10:06:11 +08003603
Chris Masond3c2fdc2007-09-17 10:58:06 -04003604 nr = trans->blocks_used;
Chris Mason54aa1f42007-06-22 14:16:25 -04003605 btrfs_end_transaction(trans, root);
Chris Masond3c2fdc2007-09-17 10:58:06 -04003606 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04003607no_delete:
Al Virobd555972010-06-07 11:35:40 -04003608 end_writeback(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003609 return;
Chris Mason39279cc2007-06-12 06:35:45 -04003610}
3611
3612/*
3613 * this returns the key found in the dir entry in the location pointer.
3614 * If no dir entries were found, location->objectid is 0.
3615 */
3616static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3617 struct btrfs_key *location)
3618{
3619 const char *name = dentry->d_name.name;
3620 int namelen = dentry->d_name.len;
3621 struct btrfs_dir_item *di;
3622 struct btrfs_path *path;
3623 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan0d9f7f32007-10-25 15:48:28 -04003624 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003625
3626 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07003627 if (!path)
3628 return -ENOMEM;
Chris Mason39544012007-12-12 14:38:19 -05003629
Li Zefan33345d012011-04-20 10:31:50 +08003630 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
Chris Mason39279cc2007-06-12 06:35:45 -04003631 namelen, 0);
Yan0d9f7f32007-10-25 15:48:28 -04003632 if (IS_ERR(di))
3633 ret = PTR_ERR(di);
Chris Masond3977122009-01-05 21:25:51 -05003634
David Sterbac7040052011-04-19 18:00:01 +02003635 if (IS_ERR_OR_NULL(di))
Chris Mason39544012007-12-12 14:38:19 -05003636 goto out_err;
Chris Masond3977122009-01-05 21:25:51 -05003637
Chris Mason5f39d392007-10-15 16:14:19 -04003638 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
Chris Mason39279cc2007-06-12 06:35:45 -04003639out:
Chris Mason39279cc2007-06-12 06:35:45 -04003640 btrfs_free_path(path);
3641 return ret;
Chris Mason39544012007-12-12 14:38:19 -05003642out_err:
3643 location->objectid = 0;
3644 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04003645}
3646
3647/*
3648 * when we hit a tree root in a directory, the btrfs part of the inode
3649 * needs to be changed to reflect the root directory of the tree root. This
3650 * is kind of like crossing a mount point.
3651 */
3652static int fixup_tree_root_location(struct btrfs_root *root,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003653 struct inode *dir,
3654 struct dentry *dentry,
3655 struct btrfs_key *location,
3656 struct btrfs_root **sub_root)
Chris Mason39279cc2007-06-12 06:35:45 -04003657{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003658 struct btrfs_path *path;
3659 struct btrfs_root *new_root;
3660 struct btrfs_root_ref *ref;
3661 struct extent_buffer *leaf;
3662 int ret;
3663 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003664
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003665 path = btrfs_alloc_path();
3666 if (!path) {
3667 err = -ENOMEM;
3668 goto out;
3669 }
Chris Mason39279cc2007-06-12 06:35:45 -04003670
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003671 err = -ENOENT;
3672 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3673 BTRFS_I(dir)->root->root_key.objectid,
3674 location->objectid);
3675 if (ret) {
3676 if (ret < 0)
3677 err = ret;
3678 goto out;
3679 }
Chris Mason39279cc2007-06-12 06:35:45 -04003680
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003681 leaf = path->nodes[0];
3682 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
Li Zefan33345d012011-04-20 10:31:50 +08003683 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003684 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3685 goto out;
3686
3687 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3688 (unsigned long)(ref + 1),
3689 dentry->d_name.len);
3690 if (ret)
3691 goto out;
3692
David Sterbab3b4aa72011-04-21 01:20:15 +02003693 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003694
3695 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3696 if (IS_ERR(new_root)) {
3697 err = PTR_ERR(new_root);
3698 goto out;
3699 }
3700
3701 if (btrfs_root_refs(&new_root->root_item) == 0) {
3702 err = -ENOENT;
3703 goto out;
3704 }
3705
3706 *sub_root = new_root;
3707 location->objectid = btrfs_root_dirid(&new_root->root_item);
3708 location->type = BTRFS_INODE_ITEM_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04003709 location->offset = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003710 err = 0;
3711out:
3712 btrfs_free_path(path);
3713 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003714}
3715
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003716static void inode_tree_add(struct inode *inode)
3717{
3718 struct btrfs_root *root = BTRFS_I(inode)->root;
3719 struct btrfs_inode *entry;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003720 struct rb_node **p;
3721 struct rb_node *parent;
Li Zefan33345d012011-04-20 10:31:50 +08003722 u64 ino = btrfs_ino(inode);
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003723again:
3724 p = &root->inode_tree.rb_node;
3725 parent = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003726
Al Viro1d3382c2010-10-23 15:19:20 -04003727 if (inode_unhashed(inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04003728 return;
3729
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003730 spin_lock(&root->inode_lock);
3731 while (*p) {
3732 parent = *p;
3733 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3734
Li Zefan33345d012011-04-20 10:31:50 +08003735 if (ino < btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003736 p = &parent->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08003737 else if (ino > btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003738 p = &parent->rb_right;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003739 else {
3740 WARN_ON(!(entry->vfs_inode.i_state &
Al Viroa4ffdde2010-06-02 17:38:30 -04003741 (I_WILL_FREE | I_FREEING)));
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003742 rb_erase(parent, &root->inode_tree);
3743 RB_CLEAR_NODE(parent);
3744 spin_unlock(&root->inode_lock);
3745 goto again;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003746 }
3747 }
3748 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3749 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3750 spin_unlock(&root->inode_lock);
3751}
3752
3753static void inode_tree_del(struct inode *inode)
3754{
3755 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng76dda932009-09-21 16:00:26 -04003756 int empty = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003757
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003758 spin_lock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003759 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003760 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003761 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Yan, Zheng76dda932009-09-21 16:00:26 -04003762 empty = RB_EMPTY_ROOT(&root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003763 }
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003764 spin_unlock(&root->inode_lock);
Yan, Zheng76dda932009-09-21 16:00:26 -04003765
Josef Bacik0af3d002010-06-21 14:48:16 -04003766 /*
3767 * Free space cache has inodes in the tree root, but the tree root has a
3768 * root_refs of 0, so this could end up dropping the tree root as a
3769 * snapshot, so we need the extra !root->fs_info->tree_root check to
3770 * make sure we don't drop it.
3771 */
3772 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3773 root != root->fs_info->tree_root) {
Yan, Zheng76dda932009-09-21 16:00:26 -04003774 synchronize_srcu(&root->fs_info->subvol_srcu);
3775 spin_lock(&root->inode_lock);
3776 empty = RB_EMPTY_ROOT(&root->inode_tree);
3777 spin_unlock(&root->inode_lock);
3778 if (empty)
3779 btrfs_add_dead_root(root);
3780 }
3781}
3782
Jeff Mahoney143bede2012-03-01 14:56:26 +01003783void btrfs_invalidate_inodes(struct btrfs_root *root)
Yan, Zheng76dda932009-09-21 16:00:26 -04003784{
3785 struct rb_node *node;
3786 struct rb_node *prev;
3787 struct btrfs_inode *entry;
3788 struct inode *inode;
3789 u64 objectid = 0;
3790
3791 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3792
3793 spin_lock(&root->inode_lock);
3794again:
3795 node = root->inode_tree.rb_node;
3796 prev = NULL;
3797 while (node) {
3798 prev = node;
3799 entry = rb_entry(node, struct btrfs_inode, rb_node);
3800
Li Zefan33345d012011-04-20 10:31:50 +08003801 if (objectid < btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04003802 node = node->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08003803 else if (objectid > btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04003804 node = node->rb_right;
3805 else
3806 break;
3807 }
3808 if (!node) {
3809 while (prev) {
3810 entry = rb_entry(prev, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08003811 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
Yan, Zheng76dda932009-09-21 16:00:26 -04003812 node = prev;
3813 break;
3814 }
3815 prev = rb_next(prev);
3816 }
3817 }
3818 while (node) {
3819 entry = rb_entry(node, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08003820 objectid = btrfs_ino(&entry->vfs_inode) + 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04003821 inode = igrab(&entry->vfs_inode);
3822 if (inode) {
3823 spin_unlock(&root->inode_lock);
3824 if (atomic_read(&inode->i_count) > 1)
3825 d_prune_aliases(inode);
3826 /*
Al Viro45321ac2010-06-07 13:43:19 -04003827 * btrfs_drop_inode will have it removed from
Yan, Zheng76dda932009-09-21 16:00:26 -04003828 * the inode cache when its usage count
3829 * hits zero.
3830 */
3831 iput(inode);
3832 cond_resched();
3833 spin_lock(&root->inode_lock);
3834 goto again;
3835 }
3836
3837 if (cond_resched_lock(&root->inode_lock))
3838 goto again;
3839
3840 node = rb_next(node);
3841 }
3842 spin_unlock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003843}
3844
Chris Masone02119d2008-09-05 16:13:11 -04003845static int btrfs_init_locked_inode(struct inode *inode, void *p)
3846{
3847 struct btrfs_iget_args *args = p;
3848 inode->i_ino = args->ino;
Chris Masone02119d2008-09-05 16:13:11 -04003849 BTRFS_I(inode)->root = args->root;
Josef Bacik6a632092009-02-20 11:00:09 -05003850 btrfs_set_inode_space_info(args->root, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003851 return 0;
3852}
3853
3854static int btrfs_find_actor(struct inode *inode, void *opaque)
3855{
3856 struct btrfs_iget_args *args = opaque;
Li Zefan33345d012011-04-20 10:31:50 +08003857 return args->ino == btrfs_ino(inode) &&
Chris Masond3977122009-01-05 21:25:51 -05003858 args->root == BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003859}
3860
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003861static struct inode *btrfs_iget_locked(struct super_block *s,
3862 u64 objectid,
3863 struct btrfs_root *root)
Chris Mason39279cc2007-06-12 06:35:45 -04003864{
3865 struct inode *inode;
3866 struct btrfs_iget_args args;
3867 args.ino = objectid;
3868 args.root = root;
3869
3870 inode = iget5_locked(s, objectid, btrfs_find_actor,
3871 btrfs_init_locked_inode,
3872 (void *)&args);
3873 return inode;
3874}
3875
Balaji Rao1a54ef82008-07-21 02:01:04 +05303876/* Get an inode object given its location and corresponding root.
3877 * Returns in *is_new if the inode was read from disk
3878 */
3879struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Josef Bacik73f73412009-12-04 17:38:27 +00003880 struct btrfs_root *root, int *new)
Balaji Rao1a54ef82008-07-21 02:01:04 +05303881{
3882 struct inode *inode;
3883
3884 inode = btrfs_iget_locked(s, location->objectid, root);
3885 if (!inode)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003886 return ERR_PTR(-ENOMEM);
Balaji Rao1a54ef82008-07-21 02:01:04 +05303887
3888 if (inode->i_state & I_NEW) {
3889 BTRFS_I(inode)->root = root;
3890 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3891 btrfs_read_locked_inode(inode);
Mark Fasheh1748f842011-07-12 11:25:31 -07003892 if (!is_bad_inode(inode)) {
3893 inode_tree_add(inode);
3894 unlock_new_inode(inode);
3895 if (new)
3896 *new = 1;
3897 } else {
Sergei Trofimoviche0b6d652011-09-11 10:52:24 -04003898 unlock_new_inode(inode);
3899 iput(inode);
3900 inode = ERR_PTR(-ESTALE);
Mark Fasheh1748f842011-07-12 11:25:31 -07003901 }
3902 }
3903
Balaji Rao1a54ef82008-07-21 02:01:04 +05303904 return inode;
3905}
3906
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003907static struct inode *new_simple_dir(struct super_block *s,
3908 struct btrfs_key *key,
3909 struct btrfs_root *root)
3910{
3911 struct inode *inode = new_inode(s);
3912
3913 if (!inode)
3914 return ERR_PTR(-ENOMEM);
3915
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003916 BTRFS_I(inode)->root = root;
3917 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3918 BTRFS_I(inode)->dummy_inode = 1;
3919
3920 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3921 inode->i_op = &simple_dir_inode_operations;
3922 inode->i_fop = &simple_dir_operations;
3923 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3924 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3925
3926 return inode;
3927}
3928
Chris Mason3de45862008-11-17 21:02:50 -05003929struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04003930{
Chris Masond3977122009-01-05 21:25:51 -05003931 struct inode *inode;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003932 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003933 struct btrfs_root *sub_root = root;
3934 struct btrfs_key location;
Yan, Zheng76dda932009-09-21 16:00:26 -04003935 int index;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04003936 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003937
3938 if (dentry->d_name.len > BTRFS_NAME_LEN)
3939 return ERR_PTR(-ENAMETOOLONG);
Chris Mason5f39d392007-10-15 16:14:19 -04003940
Josef Bacikb4aff1f2011-06-28 16:18:59 -04003941 if (unlikely(d_need_lookup(dentry))) {
3942 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3943 kfree(dentry->d_fsdata);
3944 dentry->d_fsdata = NULL;
Josef Bacika66e7cc2011-09-18 10:34:03 -04003945 /* This thing is hashed, drop it for now */
3946 d_drop(dentry);
Josef Bacikb4aff1f2011-06-28 16:18:59 -04003947 } else {
3948 ret = btrfs_inode_by_name(dir, dentry, &location);
3949 }
Chris Mason5f39d392007-10-15 16:14:19 -04003950
Chris Mason39279cc2007-06-12 06:35:45 -04003951 if (ret < 0)
3952 return ERR_PTR(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04003953
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003954 if (location.objectid == 0)
3955 return NULL;
3956
3957 if (location.type == BTRFS_INODE_ITEM_KEY) {
Josef Bacik73f73412009-12-04 17:38:27 +00003958 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003959 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04003960 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003961
3962 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3963
Yan, Zheng76dda932009-09-21 16:00:26 -04003964 index = srcu_read_lock(&root->fs_info->subvol_srcu);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003965 ret = fixup_tree_root_location(root, dir, dentry,
3966 &location, &sub_root);
3967 if (ret < 0) {
3968 if (ret != -ENOENT)
3969 inode = ERR_PTR(ret);
3970 else
3971 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3972 } else {
Josef Bacik73f73412009-12-04 17:38:27 +00003973 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04003974 }
Yan, Zheng76dda932009-09-21 16:00:26 -04003975 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3976
Julia Lawall34d19ba2011-01-24 19:55:19 +00003977 if (!IS_ERR(inode) && root != sub_root) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00003978 down_read(&root->fs_info->cleanup_work_sem);
3979 if (!(inode->i_sb->s_flags & MS_RDONLY))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003980 ret = btrfs_orphan_cleanup(sub_root);
Yan, Zhengc71bf092009-11-12 09:34:40 +00003981 up_read(&root->fs_info->cleanup_work_sem);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003982 if (ret)
3983 inode = ERR_PTR(ret);
Yan, Zhengc71bf092009-11-12 09:34:40 +00003984 }
3985
Chris Mason3de45862008-11-17 21:02:50 -05003986 return inode;
3987}
3988
Nick Pigginfe15ce42011-01-07 17:49:23 +11003989static int btrfs_dentry_delete(const struct dentry *dentry)
Yan, Zheng76dda932009-09-21 16:00:26 -04003990{
3991 struct btrfs_root *root;
3992
Yan, Zhengefefb142009-10-09 09:25:16 -04003993 if (!dentry->d_inode && !IS_ROOT(dentry))
3994 dentry = dentry->d_parent;
Yan, Zheng76dda932009-09-21 16:00:26 -04003995
Yan, Zhengefefb142009-10-09 09:25:16 -04003996 if (dentry->d_inode) {
3997 root = BTRFS_I(dentry->d_inode)->root;
3998 if (btrfs_root_refs(&root->root_item) == 0)
3999 return 1;
4000 }
Yan, Zheng76dda932009-09-21 16:00:26 -04004001 return 0;
4002}
4003
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004004static void btrfs_dentry_release(struct dentry *dentry)
4005{
4006 if (dentry->d_fsdata)
4007 kfree(dentry->d_fsdata);
4008}
4009
Chris Mason3de45862008-11-17 21:02:50 -05004010static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4011 struct nameidata *nd)
4012{
Josef Bacika66e7cc2011-09-18 10:34:03 -04004013 struct dentry *ret;
4014
4015 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4016 if (unlikely(d_need_lookup(dentry))) {
4017 spin_lock(&dentry->d_lock);
4018 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4019 spin_unlock(&dentry->d_lock);
4020 }
4021 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004022}
4023
Miao Xie16cdcec2011-04-22 18:12:22 +08004024unsigned char btrfs_filetype_table[] = {
Chris Mason39279cc2007-06-12 06:35:45 -04004025 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4026};
4027
David Woodhousecbdf5a22008-08-06 19:42:33 +01004028static int btrfs_real_readdir(struct file *filp, void *dirent,
4029 filldir_t filldir)
Chris Mason39279cc2007-06-12 06:35:45 -04004030{
Chris Mason6da6aba2007-12-18 16:15:09 -05004031 struct inode *inode = filp->f_dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04004032 struct btrfs_root *root = BTRFS_I(inode)->root;
4033 struct btrfs_item *item;
4034 struct btrfs_dir_item *di;
4035 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04004036 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04004037 struct btrfs_path *path;
Miao Xie16cdcec2011-04-22 18:12:22 +08004038 struct list_head ins_list;
4039 struct list_head del_list;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004040 struct qstr q;
Chris Mason39279cc2007-06-12 06:35:45 -04004041 int ret;
Chris Mason5f39d392007-10-15 16:14:19 -04004042 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04004043 int slot;
Chris Mason39279cc2007-06-12 06:35:45 -04004044 unsigned char d_type;
4045 int over = 0;
4046 u32 di_cur;
4047 u32 di_total;
4048 u32 di_len;
4049 int key_type = BTRFS_DIR_INDEX_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04004050 char tmp_name[32];
4051 char *name_ptr;
4052 int name_len;
Miao Xie16cdcec2011-04-22 18:12:22 +08004053 int is_curr = 0; /* filp->f_pos points to the current index? */
Chris Mason39279cc2007-06-12 06:35:45 -04004054
4055 /* FIXME, use a real flag for deciding about the key type */
4056 if (root->fs_info->tree_root == root)
4057 key_type = BTRFS_DIR_ITEM_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04004058
Chris Mason39544012007-12-12 14:38:19 -05004059 /* special case for "." */
4060 if (filp->f_pos == 0) {
Hidetoshi Seto3765fef2011-09-18 10:20:46 -04004061 over = filldir(dirent, ".", 1,
4062 filp->f_pos, btrfs_ino(inode), DT_DIR);
Chris Mason39544012007-12-12 14:38:19 -05004063 if (over)
4064 return 0;
4065 filp->f_pos = 1;
4066 }
Chris Mason39544012007-12-12 14:38:19 -05004067 /* special case for .., just use the back ref */
4068 if (filp->f_pos == 1) {
David Woodhouse5ecc7e52008-08-17 15:14:48 +01004069 u64 pino = parent_ino(filp->f_path.dentry);
Chris Mason39544012007-12-12 14:38:19 -05004070 over = filldir(dirent, "..", 2,
Hidetoshi Seto3765fef2011-09-18 10:20:46 -04004071 filp->f_pos, pino, DT_DIR);
Chris Mason39544012007-12-12 14:38:19 -05004072 if (over)
David Woodhouse49593bf2008-08-17 17:08:36 +01004073 return 0;
Chris Mason39544012007-12-12 14:38:19 -05004074 filp->f_pos = 2;
4075 }
David Woodhouse49593bf2008-08-17 17:08:36 +01004076 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08004077 if (!path)
4078 return -ENOMEM;
Chris Masonff5714c2011-05-28 07:00:39 -04004079
Josef Bacik026fd312011-05-13 10:32:11 -04004080 path->reada = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01004081
Miao Xie16cdcec2011-04-22 18:12:22 +08004082 if (key_type == BTRFS_DIR_INDEX_KEY) {
4083 INIT_LIST_HEAD(&ins_list);
4084 INIT_LIST_HEAD(&del_list);
4085 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4086 }
4087
Chris Mason39279cc2007-06-12 06:35:45 -04004088 btrfs_set_key_type(&key, key_type);
4089 key.offset = filp->f_pos;
Li Zefan33345d012011-04-20 10:31:50 +08004090 key.objectid = btrfs_ino(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004091
Chris Mason39279cc2007-06-12 06:35:45 -04004092 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4093 if (ret < 0)
4094 goto err;
David Woodhouse49593bf2008-08-17 17:08:36 +01004095
4096 while (1) {
Chris Mason5f39d392007-10-15 16:14:19 -04004097 leaf = path->nodes[0];
Chris Mason39279cc2007-06-12 06:35:45 -04004098 slot = path->slots[0];
Li Zefanb9e03af2011-03-23 10:43:58 +08004099 if (slot >= btrfs_header_nritems(leaf)) {
4100 ret = btrfs_next_leaf(root, path);
4101 if (ret < 0)
4102 goto err;
4103 else if (ret > 0)
4104 break;
4105 continue;
Chris Mason39279cc2007-06-12 06:35:45 -04004106 }
Chris Mason3de45862008-11-17 21:02:50 -05004107
Chris Mason5f39d392007-10-15 16:14:19 -04004108 item = btrfs_item_nr(leaf, slot);
4109 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4110
4111 if (found_key.objectid != key.objectid)
Chris Mason39279cc2007-06-12 06:35:45 -04004112 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004113 if (btrfs_key_type(&found_key) != key_type)
Chris Mason39279cc2007-06-12 06:35:45 -04004114 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004115 if (found_key.offset < filp->f_pos)
Li Zefanb9e03af2011-03-23 10:43:58 +08004116 goto next;
Miao Xie16cdcec2011-04-22 18:12:22 +08004117 if (key_type == BTRFS_DIR_INDEX_KEY &&
4118 btrfs_should_delete_dir_index(&del_list,
4119 found_key.offset))
4120 goto next;
Chris Mason5f39d392007-10-15 16:14:19 -04004121
4122 filp->f_pos = found_key.offset;
Miao Xie16cdcec2011-04-22 18:12:22 +08004123 is_curr = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01004124
Chris Mason39279cc2007-06-12 06:35:45 -04004125 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4126 di_cur = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04004127 di_total = btrfs_item_size(leaf, item);
David Woodhouse49593bf2008-08-17 17:08:36 +01004128
4129 while (di_cur < di_total) {
Chris Mason5f39d392007-10-15 16:14:19 -04004130 struct btrfs_key location;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004131 struct dentry *tmp;
Chris Mason5f39d392007-10-15 16:14:19 -04004132
Josef Bacik22a94d42011-03-16 16:47:17 -04004133 if (verify_dir_item(root, leaf, di))
4134 break;
4135
Chris Mason5f39d392007-10-15 16:14:19 -04004136 name_len = btrfs_dir_name_len(leaf, di);
David Woodhouse49593bf2008-08-17 17:08:36 +01004137 if (name_len <= sizeof(tmp_name)) {
Chris Mason5f39d392007-10-15 16:14:19 -04004138 name_ptr = tmp_name;
4139 } else {
4140 name_ptr = kmalloc(name_len, GFP_NOFS);
David Woodhouse49593bf2008-08-17 17:08:36 +01004141 if (!name_ptr) {
4142 ret = -ENOMEM;
4143 goto err;
4144 }
Chris Mason5f39d392007-10-15 16:14:19 -04004145 }
4146 read_extent_buffer(leaf, name_ptr,
4147 (unsigned long)(di + 1), name_len);
4148
4149 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4150 btrfs_dir_item_key_to_cpu(leaf, di, &location);
Chris Mason3de45862008-11-17 21:02:50 -05004151
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004152 q.name = name_ptr;
4153 q.len = name_len;
4154 q.hash = full_name_hash(q.name, q.len);
4155 tmp = d_lookup(filp->f_dentry, &q);
4156 if (!tmp) {
4157 struct btrfs_key *newkey;
4158
4159 newkey = kzalloc(sizeof(struct btrfs_key),
4160 GFP_NOFS);
4161 if (!newkey)
4162 goto no_dentry;
4163 tmp = d_alloc(filp->f_dentry, &q);
4164 if (!tmp) {
4165 kfree(newkey);
4166 dput(tmp);
4167 goto no_dentry;
4168 }
4169 memcpy(newkey, &location,
4170 sizeof(struct btrfs_key));
4171 tmp->d_fsdata = newkey;
4172 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4173 d_rehash(tmp);
4174 dput(tmp);
4175 } else {
4176 dput(tmp);
4177 }
4178no_dentry:
Chris Mason3de45862008-11-17 21:02:50 -05004179 /* is this a reference to our own snapshot? If so
4180 * skip it
4181 */
4182 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4183 location.objectid == root->root_key.objectid) {
4184 over = 0;
4185 goto skip;
4186 }
Chris Mason5f39d392007-10-15 16:14:19 -04004187 over = filldir(dirent, name_ptr, name_len,
David Woodhouse49593bf2008-08-17 17:08:36 +01004188 found_key.offset, location.objectid,
Chris Mason39279cc2007-06-12 06:35:45 -04004189 d_type);
Chris Mason5f39d392007-10-15 16:14:19 -04004190
Chris Mason3de45862008-11-17 21:02:50 -05004191skip:
Chris Mason5f39d392007-10-15 16:14:19 -04004192 if (name_ptr != tmp_name)
4193 kfree(name_ptr);
4194
Chris Mason39279cc2007-06-12 06:35:45 -04004195 if (over)
4196 goto nopos;
Josef Bacik5103e942007-11-16 11:45:54 -05004197 di_len = btrfs_dir_name_len(leaf, di) +
David Woodhouse49593bf2008-08-17 17:08:36 +01004198 btrfs_dir_data_len(leaf, di) + sizeof(*di);
Chris Mason39279cc2007-06-12 06:35:45 -04004199 di_cur += di_len;
4200 di = (struct btrfs_dir_item *)((char *)di + di_len);
4201 }
Li Zefanb9e03af2011-03-23 10:43:58 +08004202next:
4203 path->slots[0]++;
Chris Mason39279cc2007-06-12 06:35:45 -04004204 }
David Woodhouse49593bf2008-08-17 17:08:36 +01004205
Miao Xie16cdcec2011-04-22 18:12:22 +08004206 if (key_type == BTRFS_DIR_INDEX_KEY) {
4207 if (is_curr)
4208 filp->f_pos++;
4209 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4210 &ins_list);
4211 if (ret)
4212 goto nopos;
4213 }
4214
David Woodhouse49593bf2008-08-17 17:08:36 +01004215 /* Reached end of directory/root. Bump pos past the last item. */
Yan Zheng5e591a02008-02-19 11:41:02 -05004216 if (key_type == BTRFS_DIR_INDEX_KEY)
Jan Engelhardt406266a2009-12-09 22:00:38 +00004217 /*
4218 * 32-bit glibc will use getdents64, but then strtol -
4219 * so the last number we can serve is this.
4220 */
4221 filp->f_pos = 0x7fffffff;
Yan Zheng5e591a02008-02-19 11:41:02 -05004222 else
4223 filp->f_pos++;
Chris Mason39279cc2007-06-12 06:35:45 -04004224nopos:
4225 ret = 0;
4226err:
Miao Xie16cdcec2011-04-22 18:12:22 +08004227 if (key_type == BTRFS_DIR_INDEX_KEY)
4228 btrfs_put_delayed_items(&ins_list, &del_list);
Chris Mason39279cc2007-06-12 06:35:45 -04004229 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04004230 return ret;
4231}
4232
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004233int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Chris Mason39279cc2007-06-12 06:35:45 -04004234{
4235 struct btrfs_root *root = BTRFS_I(inode)->root;
4236 struct btrfs_trans_handle *trans;
4237 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04004238 bool nolock = false;
Chris Mason39279cc2007-06-12 06:35:45 -04004239
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004240 if (BTRFS_I(inode)->dummy_inode)
Chris Mason4ca8b412008-08-05 13:30:48 -04004241 return 0;
4242
Chris Mason2cf85722011-07-26 15:35:09 -04004243 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
Li Zefan82d59022011-04-20 10:33:24 +08004244 nolock = true;
Josef Bacik0af3d002010-06-21 14:48:16 -04004245
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004246 if (wbc->sync_mode == WB_SYNC_ALL) {
Josef Bacik0af3d002010-06-21 14:48:16 -04004247 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004248 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0af3d002010-06-21 14:48:16 -04004249 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004250 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00004251 if (IS_ERR(trans))
4252 return PTR_ERR(trans);
Josef Bacik0af3d002010-06-21 14:48:16 -04004253 if (nolock)
4254 ret = btrfs_end_transaction_nolock(trans, root);
4255 else
4256 ret = btrfs_commit_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004257 }
4258 return ret;
4259}
4260
4261/*
Chris Mason54aa1f42007-06-22 14:16:25 -04004262 * This is somewhat expensive, updating the tree every time the
Chris Mason39279cc2007-06-12 06:35:45 -04004263 * inode changes. But, it is most likely to find the inode in cache.
4264 * FIXME, needs more benchmarking...there are no reasons other than performance
4265 * to keep or drop this code.
4266 */
Josef Bacik22c44fe2011-11-30 10:45:38 -05004267int btrfs_dirty_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04004268{
4269 struct btrfs_root *root = BTRFS_I(inode)->root;
4270 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004271 int ret;
4272
4273 if (BTRFS_I(inode)->dummy_inode)
Josef Bacik22c44fe2011-11-30 10:45:38 -05004274 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004275
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004276 trans = btrfs_join_transaction(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004277 if (IS_ERR(trans))
4278 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004279
4280 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04004281 if (ret && ret == -ENOSPC) {
4282 /* whoops, lets try again with the full transaction */
4283 btrfs_end_transaction(trans, root);
4284 trans = btrfs_start_transaction(root, 1);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004285 if (IS_ERR(trans))
4286 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004287
Chris Mason94b60442010-05-26 11:02:00 -04004288 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04004289 }
Chris Mason39279cc2007-06-12 06:35:45 -04004290 btrfs_end_transaction(trans, root);
Miao Xie16cdcec2011-04-22 18:12:22 +08004291 if (BTRFS_I(inode)->delayed_node)
4292 btrfs_balance_delayed_items(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004293
4294 return ret;
4295}
4296
4297/*
4298 * This is a copy of file_update_time. We need this so we can return error on
4299 * ENOSPC for updating the inode in the case of file write and mmap writes.
4300 */
4301int btrfs_update_time(struct file *file)
4302{
4303 struct inode *inode = file->f_path.dentry->d_inode;
4304 struct timespec now;
4305 int ret;
4306 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4307
4308 /* First try to exhaust all avenues to not sync */
4309 if (IS_NOCMTIME(inode))
4310 return 0;
4311
4312 now = current_fs_time(inode->i_sb);
4313 if (!timespec_equal(&inode->i_mtime, &now))
4314 sync_it = S_MTIME;
4315
4316 if (!timespec_equal(&inode->i_ctime, &now))
4317 sync_it |= S_CTIME;
4318
4319 if (IS_I_VERSION(inode))
4320 sync_it |= S_VERSION;
4321
4322 if (!sync_it)
4323 return 0;
4324
4325 /* Finally allowed to write? Takes lock. */
4326 if (mnt_want_write_file(file))
4327 return 0;
4328
4329 /* Only change inode inside the lock region */
4330 if (sync_it & S_VERSION)
4331 inode_inc_iversion(inode);
4332 if (sync_it & S_CTIME)
4333 inode->i_ctime = now;
4334 if (sync_it & S_MTIME)
4335 inode->i_mtime = now;
4336 ret = btrfs_dirty_inode(inode);
4337 if (!ret)
4338 mark_inode_dirty_sync(inode);
4339 mnt_drop_write(file->f_path.mnt);
4340 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004341}
4342
Chris Masond352ac62008-09-29 15:18:18 -04004343/*
4344 * find the highest existing sequence number in a directory
4345 * and then set the in-memory index_cnt variable to reflect
4346 * free sequence numbers
4347 */
Josef Bacikaec74772008-07-24 12:12:38 -04004348static int btrfs_set_inode_index_count(struct inode *inode)
4349{
4350 struct btrfs_root *root = BTRFS_I(inode)->root;
4351 struct btrfs_key key, found_key;
4352 struct btrfs_path *path;
4353 struct extent_buffer *leaf;
4354 int ret;
4355
Li Zefan33345d012011-04-20 10:31:50 +08004356 key.objectid = btrfs_ino(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004357 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4358 key.offset = (u64)-1;
4359
4360 path = btrfs_alloc_path();
4361 if (!path)
4362 return -ENOMEM;
4363
4364 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4365 if (ret < 0)
4366 goto out;
4367 /* FIXME: we should be able to handle this */
4368 if (ret == 0)
4369 goto out;
4370 ret = 0;
4371
4372 /*
4373 * MAGIC NUMBER EXPLANATION:
4374 * since we search a directory based on f_pos we have to start at 2
4375 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4376 * else has to start at 2
4377 */
4378 if (path->slots[0] == 0) {
4379 BTRFS_I(inode)->index_cnt = 2;
4380 goto out;
4381 }
4382
4383 path->slots[0]--;
4384
4385 leaf = path->nodes[0];
4386 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4387
Li Zefan33345d012011-04-20 10:31:50 +08004388 if (found_key.objectid != btrfs_ino(inode) ||
Josef Bacikaec74772008-07-24 12:12:38 -04004389 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4390 BTRFS_I(inode)->index_cnt = 2;
4391 goto out;
4392 }
4393
4394 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4395out:
4396 btrfs_free_path(path);
4397 return ret;
4398}
4399
Chris Masond352ac62008-09-29 15:18:18 -04004400/*
4401 * helper to find a free sequence number in a given directory. This current
4402 * code is very simple, later versions will do smarter things in the btree
4403 */
Chris Mason3de45862008-11-17 21:02:50 -05004404int btrfs_set_inode_index(struct inode *dir, u64 *index)
Josef Bacikaec74772008-07-24 12:12:38 -04004405{
4406 int ret = 0;
4407
4408 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
Miao Xie16cdcec2011-04-22 18:12:22 +08004409 ret = btrfs_inode_delayed_dir_index_count(dir);
4410 if (ret) {
4411 ret = btrfs_set_inode_index_count(dir);
4412 if (ret)
4413 return ret;
4414 }
Josef Bacikaec74772008-07-24 12:12:38 -04004415 }
4416
Chris Mason00e4e6b2008-08-05 11:18:09 -04004417 *index = BTRFS_I(dir)->index_cnt;
Josef Bacikaec74772008-07-24 12:12:38 -04004418 BTRFS_I(dir)->index_cnt++;
4419
4420 return ret;
4421}
4422
Chris Mason39279cc2007-06-12 06:35:45 -04004423static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4424 struct btrfs_root *root,
Josef Bacikaec74772008-07-24 12:12:38 -04004425 struct inode *dir,
Chris Mason9c583092008-01-29 15:15:18 -05004426 const char *name, int name_len,
Al Viro175a4eb2011-07-26 03:30:54 -04004427 u64 ref_objectid, u64 objectid,
4428 umode_t mode, u64 *index)
Chris Mason39279cc2007-06-12 06:35:45 -04004429{
4430 struct inode *inode;
Chris Mason5f39d392007-10-15 16:14:19 -04004431 struct btrfs_inode_item *inode_item;
Chris Mason39279cc2007-06-12 06:35:45 -04004432 struct btrfs_key *location;
Chris Mason5f39d392007-10-15 16:14:19 -04004433 struct btrfs_path *path;
Chris Mason9c583092008-01-29 15:15:18 -05004434 struct btrfs_inode_ref *ref;
4435 struct btrfs_key key[2];
4436 u32 sizes[2];
4437 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04004438 int ret;
4439 int owner;
4440
Chris Mason5f39d392007-10-15 16:14:19 -04004441 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07004442 if (!path)
4443 return ERR_PTR(-ENOMEM);
Chris Mason5f39d392007-10-15 16:14:19 -04004444
Chris Mason39279cc2007-06-12 06:35:45 -04004445 inode = new_inode(root->fs_info->sb);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00004446 if (!inode) {
4447 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04004448 return ERR_PTR(-ENOMEM);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00004449 }
Chris Mason39279cc2007-06-12 06:35:45 -04004450
Li Zefan581bb052011-04-20 10:06:11 +08004451 /*
4452 * we have to initialize this early, so we can reclaim the inode
4453 * number if we fail afterwards in this function.
4454 */
4455 inode->i_ino = objectid;
4456
Josef Bacikaec74772008-07-24 12:12:38 -04004457 if (dir) {
liubo1abe9b82011-03-24 11:18:59 +00004458 trace_btrfs_inode_request(dir);
4459
Chris Mason3de45862008-11-17 21:02:50 -05004460 ret = btrfs_set_inode_index(dir, index);
Shen Feng09771432009-04-02 16:46:06 -04004461 if (ret) {
Yoshinori Sano8fb27642011-04-09 02:30:07 +00004462 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04004463 iput(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004464 return ERR_PTR(ret);
Shen Feng09771432009-04-02 16:46:06 -04004465 }
Josef Bacikaec74772008-07-24 12:12:38 -04004466 }
4467 /*
4468 * index_cnt is ignored for everything but a dir,
4469 * btrfs_get_inode_index_count has an explanation for the magic
4470 * number
4471 */
4472 BTRFS_I(inode)->index_cnt = 2;
Chris Mason39279cc2007-06-12 06:35:45 -04004473 BTRFS_I(inode)->root = root;
Chris Masone02119d2008-09-05 16:13:11 -04004474 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik76195852010-11-19 02:18:02 +00004475 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik6a632092009-02-20 11:00:09 -05004476 btrfs_set_inode_space_info(root, inode);
Chris Masonb888db22007-08-27 16:49:44 -04004477
Al Viro569254b2011-07-24 17:08:40 -04004478 if (S_ISDIR(mode))
Chris Mason39279cc2007-06-12 06:35:45 -04004479 owner = 0;
4480 else
4481 owner = 1;
Chris Mason9c583092008-01-29 15:15:18 -05004482
4483 key[0].objectid = objectid;
4484 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4485 key[0].offset = 0;
4486
4487 key[1].objectid = objectid;
4488 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4489 key[1].offset = ref_objectid;
4490
4491 sizes[0] = sizeof(struct btrfs_inode_item);
4492 sizes[1] = name_len + sizeof(*ref);
4493
Chris Masonb9473432009-03-13 11:00:37 -04004494 path->leave_spinning = 1;
Chris Mason9c583092008-01-29 15:15:18 -05004495 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4496 if (ret != 0)
Chris Mason5f39d392007-10-15 16:14:19 -04004497 goto fail;
4498
Dmitry Monakhovecc11fab2010-03-04 17:31:47 +03004499 inode_init_owner(inode, dir, mode);
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004500 inode_set_bytes(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004501 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
Chris Mason5f39d392007-10-15 16:14:19 -04004502 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4503 struct btrfs_inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04004504 fill_inode_item(trans, path->nodes[0], inode_item, inode);
Chris Mason9c583092008-01-29 15:15:18 -05004505
4506 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4507 struct btrfs_inode_ref);
4508 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
Chris Mason00e4e6b2008-08-05 11:18:09 -04004509 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
Chris Mason9c583092008-01-29 15:15:18 -05004510 ptr = (unsigned long)(ref + 1);
4511 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4512
Chris Mason5f39d392007-10-15 16:14:19 -04004513 btrfs_mark_buffer_dirty(path->nodes[0]);
4514 btrfs_free_path(path);
4515
Chris Mason39279cc2007-06-12 06:35:45 -04004516 location = &BTRFS_I(inode)->location;
4517 location->objectid = objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04004518 location->offset = 0;
4519 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4520
Christoph Hellwig6cbff002009-04-17 10:37:41 +02004521 btrfs_inherit_iflags(inode, dir);
4522
Al Viro569254b2011-07-24 17:08:40 -04004523 if (S_ISREG(mode)) {
Chris Mason94272162009-07-02 12:26:06 -04004524 if (btrfs_test_opt(root, NODATASUM))
4525 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
Liu Bo75e7cb72011-03-22 10:12:20 +00004526 if (btrfs_test_opt(root, NODATACOW) ||
4527 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
Chris Mason94272162009-07-02 12:26:06 -04004528 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4529 }
4530
Chris Mason39279cc2007-06-12 06:35:45 -04004531 insert_inode_hash(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004532 inode_tree_add(inode);
liubo1abe9b82011-03-24 11:18:59 +00004533
4534 trace_btrfs_inode_new(inode);
Chris Mason1973f0f2011-06-24 13:13:29 -04004535 btrfs_set_inode_last_trans(trans, inode);
liubo1abe9b82011-03-24 11:18:59 +00004536
Chris Mason39279cc2007-06-12 06:35:45 -04004537 return inode;
Chris Mason5f39d392007-10-15 16:14:19 -04004538fail:
Josef Bacikaec74772008-07-24 12:12:38 -04004539 if (dir)
4540 BTRFS_I(dir)->index_cnt--;
Chris Mason5f39d392007-10-15 16:14:19 -04004541 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04004542 iput(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004543 return ERR_PTR(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04004544}
4545
4546static inline u8 btrfs_inode_type(struct inode *inode)
4547{
4548 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4549}
4550
Chris Masond352ac62008-09-29 15:18:18 -04004551/*
4552 * utility function to add 'inode' into 'parent_inode' with
4553 * a give name and a given sequence number.
4554 * if 'add_backref' is true, also insert a backref from the
4555 * inode to the parent directory.
4556 */
Chris Masone02119d2008-09-05 16:13:11 -04004557int btrfs_add_link(struct btrfs_trans_handle *trans,
4558 struct inode *parent_inode, struct inode *inode,
4559 const char *name, int name_len, int add_backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04004560{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004561 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004562 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04004563 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
Li Zefan33345d012011-04-20 10:31:50 +08004564 u64 ino = btrfs_ino(inode);
4565 u64 parent_ino = btrfs_ino(parent_inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004566
Li Zefan33345d012011-04-20 10:31:50 +08004567 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004568 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4569 } else {
Li Zefan33345d012011-04-20 10:31:50 +08004570 key.objectid = ino;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004571 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4572 key.offset = 0;
4573 }
Chris Mason39279cc2007-06-12 06:35:45 -04004574
Li Zefan33345d012011-04-20 10:31:50 +08004575 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004576 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4577 key.objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08004578 parent_ino, index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004579 } else if (add_backref) {
Li Zefan33345d012011-04-20 10:31:50 +08004580 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4581 parent_ino, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004582 }
4583
Chris Mason39279cc2007-06-12 06:35:45 -04004584 if (ret == 0) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004585 ret = btrfs_insert_dir_item(trans, root, name, name_len,
Miao Xie16cdcec2011-04-22 18:12:22 +08004586 parent_inode, &key,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004587 btrfs_inode_type(inode), index);
Chris Masonfe66a052012-02-20 08:40:56 -05004588 if (ret)
4589 goto fail_dir_item;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004590
Chris Masondbe674a2008-07-17 12:54:05 -04004591 btrfs_i_size_write(parent_inode, parent_inode->i_size +
Chris Masone02119d2008-09-05 16:13:11 -04004592 name_len * 2);
Chris Mason79c44582007-06-25 10:09:33 -04004593 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
Chris Masone02119d2008-09-05 16:13:11 -04004594 ret = btrfs_update_inode(trans, root, parent_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004595 }
4596 return ret;
Chris Masonfe66a052012-02-20 08:40:56 -05004597
4598fail_dir_item:
4599 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4600 u64 local_index;
4601 int err;
4602 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4603 key.objectid, root->root_key.objectid,
4604 parent_ino, &local_index, name, name_len);
4605
4606 } else if (add_backref) {
4607 u64 local_index;
4608 int err;
4609
4610 err = btrfs_del_inode_ref(trans, root, name, name_len,
4611 ino, parent_ino, &local_index);
4612 }
4613 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004614}
4615
4616static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
Josef Bacika1b075d2010-11-19 20:36:11 +00004617 struct inode *dir, struct dentry *dentry,
4618 struct inode *inode, int backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04004619{
Josef Bacika1b075d2010-11-19 20:36:11 +00004620 int err = btrfs_add_link(trans, dir, inode,
4621 dentry->d_name.name, dentry->d_name.len,
4622 backref, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004623 if (err > 0)
4624 err = -EEXIST;
4625 return err;
4626}
4627
Josef Bacik618e21d2007-07-11 10:18:17 -04004628static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -04004629 umode_t mode, dev_t rdev)
Josef Bacik618e21d2007-07-11 10:18:17 -04004630{
4631 struct btrfs_trans_handle *trans;
4632 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05004633 struct inode *inode = NULL;
Josef Bacik618e21d2007-07-11 10:18:17 -04004634 int err;
4635 int drop_inode = 0;
4636 u64 objectid;
Chris Mason1832a6d2007-12-21 16:27:21 -05004637 unsigned long nr = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004638 u64 index = 0;
Josef Bacik618e21d2007-07-11 10:18:17 -04004639
4640 if (!new_valid_dev(rdev))
4641 return -EINVAL;
4642
Josef Bacik9ed74f22009-09-11 16:12:44 -04004643 /*
4644 * 2 for inode item and ref
4645 * 2 for dir items
4646 * 1 for xattr if selinux is on
4647 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004648 trans = btrfs_start_transaction(root, 5);
4649 if (IS_ERR(trans))
4650 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05004651
Li Zefan581bb052011-04-20 10:06:11 +08004652 err = btrfs_find_free_ino(root, &objectid);
4653 if (err)
4654 goto out_unlock;
4655
Josef Bacikaec74772008-07-24 12:12:38 -04004656 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08004657 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04004658 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04004659 if (IS_ERR(inode)) {
4660 err = PTR_ERR(inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04004661 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04004662 }
Josef Bacik618e21d2007-07-11 10:18:17 -04004663
Eric Paris2a7dba32011-02-01 11:05:39 -05004664 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04004665 if (err) {
4666 drop_inode = 1;
4667 goto out_unlock;
4668 }
4669
Casey Schauflerad19db72011-12-15 10:09:07 -05004670 /*
4671 * If the active LSM wants to access the inode during
4672 * d_instantiate it needs these. Smack checks to see
4673 * if the filesystem supports xattrs by looking at the
4674 * ops vector.
4675 */
4676
4677 inode->i_op = &btrfs_special_inode_operations;
Josef Bacika1b075d2010-11-19 20:36:11 +00004678 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Josef Bacik618e21d2007-07-11 10:18:17 -04004679 if (err)
4680 drop_inode = 1;
4681 else {
Josef Bacik618e21d2007-07-11 10:18:17 -04004682 init_special_inode(inode, inode->i_mode, rdev);
Yan1b4ab1b2007-08-29 09:11:44 -04004683 btrfs_update_inode(trans, root, inode);
Al Viro08c422c2011-12-23 07:58:13 -05004684 d_instantiate(dentry, inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04004685 }
Josef Bacik618e21d2007-07-11 10:18:17 -04004686out_unlock:
Chris Masond3c2fdc2007-09-17 10:58:06 -04004687 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004688 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004689 btrfs_btree_balance_dirty(root, nr);
Josef Bacik618e21d2007-07-11 10:18:17 -04004690 if (drop_inode) {
4691 inode_dec_link_count(inode);
4692 iput(inode);
4693 }
Josef Bacik618e21d2007-07-11 10:18:17 -04004694 return err;
4695}
4696
Chris Mason39279cc2007-06-12 06:35:45 -04004697static int btrfs_create(struct inode *dir, struct dentry *dentry,
Al Viro4acdaf22011-07-26 01:42:34 -04004698 umode_t mode, struct nameidata *nd)
Chris Mason39279cc2007-06-12 06:35:45 -04004699{
4700 struct btrfs_trans_handle *trans;
4701 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05004702 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04004703 int drop_inode = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004704 int err;
Chris Mason1832a6d2007-12-21 16:27:21 -05004705 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004706 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004707 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004708
Josef Bacik9ed74f22009-09-11 16:12:44 -04004709 /*
4710 * 2 for inode item and ref
4711 * 2 for dir items
4712 * 1 for xattr if selinux is on
4713 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004714 trans = btrfs_start_transaction(root, 5);
4715 if (IS_ERR(trans))
4716 return PTR_ERR(trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004717
Li Zefan581bb052011-04-20 10:06:11 +08004718 err = btrfs_find_free_ino(root, &objectid);
4719 if (err)
4720 goto out_unlock;
4721
Josef Bacikaec74772008-07-24 12:12:38 -04004722 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08004723 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04004724 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04004725 if (IS_ERR(inode)) {
4726 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004727 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04004728 }
Chris Mason39279cc2007-06-12 06:35:45 -04004729
Eric Paris2a7dba32011-02-01 11:05:39 -05004730 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04004731 if (err) {
4732 drop_inode = 1;
4733 goto out_unlock;
4734 }
4735
Casey Schauflerad19db72011-12-15 10:09:07 -05004736 /*
4737 * If the active LSM wants to access the inode during
4738 * d_instantiate it needs these. Smack checks to see
4739 * if the filesystem supports xattrs by looking at the
4740 * ops vector.
4741 */
4742 inode->i_fop = &btrfs_file_operations;
4743 inode->i_op = &btrfs_file_inode_operations;
4744
Josef Bacika1b075d2010-11-19 20:36:11 +00004745 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004746 if (err)
4747 drop_inode = 1;
4748 else {
4749 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04004750 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05004751 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Al Viro08c422c2011-12-23 07:58:13 -05004752 d_instantiate(dentry, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004753 }
Chris Mason39279cc2007-06-12 06:35:45 -04004754out_unlock:
Chris Masond3c2fdc2007-09-17 10:58:06 -04004755 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004756 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004757 if (drop_inode) {
4758 inode_dec_link_count(inode);
4759 iput(inode);
4760 }
Chris Masond3c2fdc2007-09-17 10:58:06 -04004761 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004762 return err;
4763}
4764
4765static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4766 struct dentry *dentry)
4767{
4768 struct btrfs_trans_handle *trans;
4769 struct btrfs_root *root = BTRFS_I(dir)->root;
4770 struct inode *inode = old_dentry->d_inode;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004771 u64 index;
Chris Mason1832a6d2007-12-21 16:27:21 -05004772 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004773 int err;
4774 int drop_inode = 0;
4775
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00004776 /* do not allow sys_link's with other subvols of the same device */
4777 if (root->objectid != BTRFS_I(inode)->root->objectid)
Mark Fasheh3ab35642011-03-22 17:20:26 +00004778 return -EXDEV;
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00004779
Al Viroc055e992011-03-04 17:15:18 +00004780 if (inode->i_nlink == ~0U)
4781 return -EMLINK;
Josef Bacik9ed74f22009-09-11 16:12:44 -04004782
Chris Mason3de45862008-11-17 21:02:50 -05004783 err = btrfs_set_inode_index(dir, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04004784 if (err)
4785 goto fail;
4786
Yan, Zhenga22285a2010-05-16 10:48:46 -04004787 /*
Miao Xie7e6b6462011-02-18 09:21:17 +00004788 * 2 items for inode and inode ref
Yan, Zhenga22285a2010-05-16 10:48:46 -04004789 * 2 items for dir items
Miao Xie7e6b6462011-02-18 09:21:17 +00004790 * 1 item for parent inode
Yan, Zhenga22285a2010-05-16 10:48:46 -04004791 */
Miao Xie7e6b6462011-02-18 09:21:17 +00004792 trans = btrfs_start_transaction(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004793 if (IS_ERR(trans)) {
4794 err = PTR_ERR(trans);
4795 goto fail;
4796 }
Chris Mason5f39d392007-10-15 16:14:19 -04004797
Miao Xie31534952011-04-13 13:19:21 +08004798 btrfs_inc_nlink(inode);
4799 inode->i_ctime = CURRENT_TIME;
Al Viro7de9c6e2010-10-23 11:11:40 -04004800 ihold(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004801
Josef Bacika1b075d2010-11-19 20:36:11 +00004802 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
Chris Mason5f39d392007-10-15 16:14:19 -04004803
Yan, Zhenga5719522009-09-24 09:17:31 -04004804 if (err) {
Chris Mason39279cc2007-06-12 06:35:45 -04004805 drop_inode = 1;
Yan, Zhenga5719522009-09-24 09:17:31 -04004806 } else {
Al Viro10d9f302011-07-16 23:09:10 -04004807 struct dentry *parent = dentry->d_parent;
Yan, Zhenga5719522009-09-24 09:17:31 -04004808 err = btrfs_update_inode(trans, root, inode);
4809 BUG_ON(err);
Al Viro08c422c2011-12-23 07:58:13 -05004810 d_instantiate(dentry, inode);
Josef Bacik6a912212010-11-20 09:48:00 +00004811 btrfs_log_new_name(trans, inode, NULL, parent);
Yan, Zhenga5719522009-09-24 09:17:31 -04004812 }
Chris Mason39279cc2007-06-12 06:35:45 -04004813
Chris Masond3c2fdc2007-09-17 10:58:06 -04004814 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004815 btrfs_end_transaction(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05004816fail:
Chris Mason39279cc2007-06-12 06:35:45 -04004817 if (drop_inode) {
4818 inode_dec_link_count(inode);
4819 iput(inode);
4820 }
Chris Masond3c2fdc2007-09-17 10:58:06 -04004821 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004822 return err;
4823}
4824
Al Viro18bb1db2011-07-26 01:41:39 -04004825static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Chris Mason39279cc2007-06-12 06:35:45 -04004826{
Chris Masonb9d86662008-05-02 16:13:49 -04004827 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04004828 struct btrfs_trans_handle *trans;
4829 struct btrfs_root *root = BTRFS_I(dir)->root;
4830 int err = 0;
4831 int drop_on_err = 0;
Chris Masonb9d86662008-05-02 16:13:49 -04004832 u64 objectid = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004833 u64 index = 0;
Chris Masond3c2fdc2007-09-17 10:58:06 -04004834 unsigned long nr = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04004835
Josef Bacik9ed74f22009-09-11 16:12:44 -04004836 /*
4837 * 2 items for inode and ref
4838 * 2 items for dir items
4839 * 1 for xattr if selinux is on
4840 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004841 trans = btrfs_start_transaction(root, 5);
4842 if (IS_ERR(trans))
4843 return PTR_ERR(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04004844
Li Zefan581bb052011-04-20 10:06:11 +08004845 err = btrfs_find_free_ino(root, &objectid);
4846 if (err)
4847 goto out_fail;
4848
Josef Bacikaec74772008-07-24 12:12:38 -04004849 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08004850 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04004851 S_IFDIR | mode, &index);
Chris Mason39279cc2007-06-12 06:35:45 -04004852 if (IS_ERR(inode)) {
4853 err = PTR_ERR(inode);
4854 goto out_fail;
4855 }
Chris Mason5f39d392007-10-15 16:14:19 -04004856
Chris Mason39279cc2007-06-12 06:35:45 -04004857 drop_on_err = 1;
Josef Bacik33268ea2008-07-24 12:16:36 -04004858
Eric Paris2a7dba32011-02-01 11:05:39 -05004859 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04004860 if (err)
4861 goto out_fail;
4862
Chris Mason39279cc2007-06-12 06:35:45 -04004863 inode->i_op = &btrfs_dir_inode_operations;
4864 inode->i_fop = &btrfs_dir_file_operations;
Chris Mason39279cc2007-06-12 06:35:45 -04004865
Chris Masondbe674a2008-07-17 12:54:05 -04004866 btrfs_i_size_write(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004867 err = btrfs_update_inode(trans, root, inode);
4868 if (err)
4869 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04004870
Josef Bacika1b075d2010-11-19 20:36:11 +00004871 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4872 dentry->d_name.len, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004873 if (err)
4874 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04004875
Chris Mason39279cc2007-06-12 06:35:45 -04004876 d_instantiate(dentry, inode);
4877 drop_on_err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004878
4879out_fail:
Chris Masond3c2fdc2007-09-17 10:58:06 -04004880 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004881 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004882 if (drop_on_err)
4883 iput(inode);
Chris Masond3c2fdc2007-09-17 10:58:06 -04004884 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004885 return err;
4886}
4887
Chris Masond352ac62008-09-29 15:18:18 -04004888/* helper for btfs_get_extent. Given an existing extent in the tree,
4889 * and an extent that you want to insert, deal with overlap and insert
4890 * the new extent into the tree.
4891 */
Chris Mason3b951512008-04-17 11:29:12 -04004892static int merge_extent_mapping(struct extent_map_tree *em_tree,
4893 struct extent_map *existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04004894 struct extent_map *em,
4895 u64 map_start, u64 map_len)
Chris Mason3b951512008-04-17 11:29:12 -04004896{
4897 u64 start_diff;
Chris Mason3b951512008-04-17 11:29:12 -04004898
Chris Masone6dcd2d2008-07-17 12:53:50 -04004899 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4900 start_diff = map_start - em->start;
4901 em->start = map_start;
4902 em->len = map_len;
Chris Masonc8b97812008-10-29 14:49:59 -04004903 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4904 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04004905 em->block_start += start_diff;
Chris Masonc8b97812008-10-29 14:49:59 -04004906 em->block_len -= start_diff;
4907 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004908 return add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04004909}
4910
Chris Masonc8b97812008-10-29 14:49:59 -04004911static noinline int uncompress_inline(struct btrfs_path *path,
4912 struct inode *inode, struct page *page,
4913 size_t pg_offset, u64 extent_offset,
4914 struct btrfs_file_extent_item *item)
4915{
4916 int ret;
4917 struct extent_buffer *leaf = path->nodes[0];
4918 char *tmp;
4919 size_t max_size;
4920 unsigned long inline_size;
4921 unsigned long ptr;
Li Zefan261507a02010-12-17 14:21:50 +08004922 int compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04004923
4924 WARN_ON(pg_offset != 0);
Li Zefan261507a02010-12-17 14:21:50 +08004925 compress_type = btrfs_file_extent_compression(leaf, item);
Chris Masonc8b97812008-10-29 14:49:59 -04004926 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4927 inline_size = btrfs_file_extent_inline_item_len(leaf,
4928 btrfs_item_nr(leaf, path->slots[0]));
4929 tmp = kmalloc(inline_size, GFP_NOFS);
Tsutomu Itoh8d413712011-04-25 19:43:52 -04004930 if (!tmp)
4931 return -ENOMEM;
Chris Masonc8b97812008-10-29 14:49:59 -04004932 ptr = btrfs_file_extent_inline_start(item);
4933
4934 read_extent_buffer(leaf, tmp, ptr, inline_size);
4935
Chris Mason5b050f02008-11-11 09:34:41 -05004936 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
Li Zefan261507a02010-12-17 14:21:50 +08004937 ret = btrfs_decompress(compress_type, tmp, page,
4938 extent_offset, inline_size, max_size);
Chris Masonc8b97812008-10-29 14:49:59 -04004939 if (ret) {
4940 char *kaddr = kmap_atomic(page, KM_USER0);
4941 unsigned long copy_size = min_t(u64,
4942 PAGE_CACHE_SIZE - pg_offset,
4943 max_size - extent_offset);
4944 memset(kaddr + pg_offset, 0, copy_size);
4945 kunmap_atomic(kaddr, KM_USER0);
4946 }
4947 kfree(tmp);
4948 return 0;
4949}
4950
Chris Masond352ac62008-09-29 15:18:18 -04004951/*
4952 * a bit scary, this does extent mapping from logical file offset to the disk.
Chris Masond3977122009-01-05 21:25:51 -05004953 * the ugly parts come from merging extents from the disk with the in-ram
4954 * representation. This gets more complex because of the data=ordered code,
Chris Masond352ac62008-09-29 15:18:18 -04004955 * where the in-ram extents might be locked pending data=ordered completion.
4956 *
4957 * This also copies inline extents directly into the page.
4958 */
Chris Masond3977122009-01-05 21:25:51 -05004959
Chris Masona52d9a82007-08-27 16:49:44 -04004960struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
Chris Mason70dec802008-01-29 09:59:12 -05004961 size_t pg_offset, u64 start, u64 len,
Chris Masona52d9a82007-08-27 16:49:44 -04004962 int create)
4963{
4964 int ret;
4965 int err = 0;
Chris Masondb945352007-10-15 16:15:53 -04004966 u64 bytenr;
Chris Masona52d9a82007-08-27 16:49:44 -04004967 u64 extent_start = 0;
4968 u64 extent_end = 0;
Li Zefan33345d012011-04-20 10:31:50 +08004969 u64 objectid = btrfs_ino(inode);
Chris Masona52d9a82007-08-27 16:49:44 -04004970 u32 found_type;
Chris Masonf4219502008-07-22 11:18:09 -04004971 struct btrfs_path *path = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -04004972 struct btrfs_root *root = BTRFS_I(inode)->root;
4973 struct btrfs_file_extent_item *item;
Chris Mason5f39d392007-10-15 16:14:19 -04004974 struct extent_buffer *leaf;
4975 struct btrfs_key found_key;
Chris Masona52d9a82007-08-27 16:49:44 -04004976 struct extent_map *em = NULL;
4977 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Masond1310b22008-01-24 16:13:08 -05004978 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04004979 struct btrfs_trans_handle *trans = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08004980 int compress_type;
Chris Masona52d9a82007-08-27 16:49:44 -04004981
Chris Masona52d9a82007-08-27 16:49:44 -04004982again:
Chris Mason890871b2009-09-02 16:24:52 -04004983 read_lock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05004984 em = lookup_extent_mapping(em_tree, start, len);
Chris Masona061fc82008-05-07 11:43:44 -04004985 if (em)
4986 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason890871b2009-09-02 16:24:52 -04004987 read_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05004988
Chris Masona52d9a82007-08-27 16:49:44 -04004989 if (em) {
Chris Masone1c4b742008-04-22 13:26:46 -04004990 if (em->start > start || em->start + em->len <= start)
4991 free_extent_map(em);
4992 else if (em->block_start == EXTENT_MAP_INLINE && page)
Chris Mason70dec802008-01-29 09:59:12 -05004993 free_extent_map(em);
4994 else
4995 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004996 }
David Sterba172ddd62011-04-21 00:48:27 +02004997 em = alloc_extent_map();
Chris Masona52d9a82007-08-27 16:49:44 -04004998 if (!em) {
Chris Masond1310b22008-01-24 16:13:08 -05004999 err = -ENOMEM;
5000 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04005001 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04005002 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Masond1310b22008-01-24 16:13:08 -05005003 em->start = EXTENT_MAP_HOLE;
Chris Mason445a6942008-11-10 11:53:33 -05005004 em->orig_start = EXTENT_MAP_HOLE;
Chris Masond1310b22008-01-24 16:13:08 -05005005 em->len = (u64)-1;
Chris Masonc8b97812008-10-29 14:49:59 -04005006 em->block_len = (u64)-1;
Chris Masonf4219502008-07-22 11:18:09 -04005007
5008 if (!path) {
5009 path = btrfs_alloc_path();
Josef Bacik026fd312011-05-13 10:32:11 -04005010 if (!path) {
5011 err = -ENOMEM;
5012 goto out;
5013 }
5014 /*
5015 * Chances are we'll be called again, so go ahead and do
5016 * readahead
5017 */
5018 path->reada = 1;
Chris Masonf4219502008-07-22 11:18:09 -04005019 }
5020
Chris Mason179e29e2007-11-01 11:28:41 -04005021 ret = btrfs_lookup_file_extent(trans, root, path,
5022 objectid, start, trans != NULL);
Chris Masona52d9a82007-08-27 16:49:44 -04005023 if (ret < 0) {
5024 err = ret;
5025 goto out;
5026 }
5027
5028 if (ret != 0) {
5029 if (path->slots[0] == 0)
5030 goto not_found;
5031 path->slots[0]--;
5032 }
5033
Chris Mason5f39d392007-10-15 16:14:19 -04005034 leaf = path->nodes[0];
5035 item = btrfs_item_ptr(leaf, path->slots[0],
Chris Masona52d9a82007-08-27 16:49:44 -04005036 struct btrfs_file_extent_item);
Chris Masona52d9a82007-08-27 16:49:44 -04005037 /* are we inside the extent that was found? */
Chris Mason5f39d392007-10-15 16:14:19 -04005038 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5039 found_type = btrfs_key_type(&found_key);
5040 if (found_key.objectid != objectid ||
Chris Masona52d9a82007-08-27 16:49:44 -04005041 found_type != BTRFS_EXTENT_DATA_KEY) {
5042 goto not_found;
5043 }
5044
Chris Mason5f39d392007-10-15 16:14:19 -04005045 found_type = btrfs_file_extent_type(leaf, item);
5046 extent_start = found_key.offset;
Li Zefan261507a02010-12-17 14:21:50 +08005047 compress_type = btrfs_file_extent_compression(leaf, item);
Yan Zhengd899e052008-10-30 14:25:28 -04005048 if (found_type == BTRFS_FILE_EXTENT_REG ||
5049 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04005050 extent_end = extent_start +
Chris Masondb945352007-10-15 16:15:53 -04005051 btrfs_file_extent_num_bytes(leaf, item);
Yan Zheng9036c102008-10-30 14:19:41 -04005052 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5053 size_t size;
5054 size = btrfs_file_extent_inline_len(leaf, item);
5055 extent_end = (extent_start + size + root->sectorsize - 1) &
5056 ~((u64)root->sectorsize - 1);
5057 }
5058
5059 if (start >= extent_end) {
5060 path->slots[0]++;
5061 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5062 ret = btrfs_next_leaf(root, path);
5063 if (ret < 0) {
5064 err = ret;
5065 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04005066 }
Yan Zheng9036c102008-10-30 14:19:41 -04005067 if (ret > 0)
5068 goto not_found;
5069 leaf = path->nodes[0];
Chris Masona52d9a82007-08-27 16:49:44 -04005070 }
Yan Zheng9036c102008-10-30 14:19:41 -04005071 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5072 if (found_key.objectid != objectid ||
5073 found_key.type != BTRFS_EXTENT_DATA_KEY)
5074 goto not_found;
5075 if (start + len <= found_key.offset)
5076 goto not_found;
5077 em->start = start;
5078 em->len = found_key.offset - start;
5079 goto not_found_em;
5080 }
5081
Yan Zhengd899e052008-10-30 14:25:28 -04005082 if (found_type == BTRFS_FILE_EXTENT_REG ||
5083 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng9036c102008-10-30 14:19:41 -04005084 em->start = extent_start;
5085 em->len = extent_end - extent_start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -05005086 em->orig_start = extent_start -
5087 btrfs_file_extent_offset(leaf, item);
Chris Masondb945352007-10-15 16:15:53 -04005088 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5089 if (bytenr == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04005090 em->block_start = EXTENT_MAP_HOLE;
Chris Masona52d9a82007-08-27 16:49:44 -04005091 goto insert;
5092 }
Li Zefan261507a02010-12-17 14:21:50 +08005093 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04005094 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Li Zefan261507a02010-12-17 14:21:50 +08005095 em->compress_type = compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04005096 em->block_start = bytenr;
5097 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5098 item);
5099 } else {
5100 bytenr += btrfs_file_extent_offset(leaf, item);
5101 em->block_start = bytenr;
5102 em->block_len = em->len;
Yan Zhengd899e052008-10-30 14:25:28 -04005103 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5104 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -04005105 }
Chris Masona52d9a82007-08-27 16:49:44 -04005106 goto insert;
5107 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04005108 unsigned long ptr;
Chris Masona52d9a82007-08-27 16:49:44 -04005109 char *map;
Chris Mason3326d1b2007-10-15 16:18:25 -04005110 size_t size;
5111 size_t extent_offset;
5112 size_t copy_size;
Chris Masona52d9a82007-08-27 16:49:44 -04005113
Chris Masona52d9a82007-08-27 16:49:44 -04005114 em->block_start = EXTENT_MAP_INLINE;
Chris Masonc8b97812008-10-29 14:49:59 -04005115 if (!page || create) {
Yan689f9342007-10-29 11:41:07 -04005116 em->start = extent_start;
Yan Zheng9036c102008-10-30 14:19:41 -04005117 em->len = extent_end - extent_start;
Yan689f9342007-10-29 11:41:07 -04005118 goto out;
5119 }
5120
Yan Zheng9036c102008-10-30 14:19:41 -04005121 size = btrfs_file_extent_inline_len(leaf, item);
5122 extent_offset = page_offset(page) + pg_offset - extent_start;
Chris Mason70dec802008-01-29 09:59:12 -05005123 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
Yan689f9342007-10-29 11:41:07 -04005124 size - extent_offset);
Chris Mason3326d1b2007-10-15 16:18:25 -04005125 em->start = extent_start + extent_offset;
Chris Mason70dec802008-01-29 09:59:12 -05005126 em->len = (copy_size + root->sectorsize - 1) &
5127 ~((u64)root->sectorsize - 1);
Yan Zhengff5b7ee2008-11-10 07:34:43 -05005128 em->orig_start = EXTENT_MAP_INLINE;
Li Zefan261507a02010-12-17 14:21:50 +08005129 if (compress_type) {
Chris Masonc8b97812008-10-29 14:49:59 -04005130 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Li Zefan261507a02010-12-17 14:21:50 +08005131 em->compress_type = compress_type;
5132 }
Yan689f9342007-10-29 11:41:07 -04005133 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
Chris Mason179e29e2007-11-01 11:28:41 -04005134 if (create == 0 && !PageUptodate(page)) {
Li Zefan261507a02010-12-17 14:21:50 +08005135 if (btrfs_file_extent_compression(leaf, item) !=
5136 BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04005137 ret = uncompress_inline(path, inode, page,
5138 pg_offset,
5139 extent_offset, item);
5140 BUG_ON(ret);
5141 } else {
5142 map = kmap(page);
5143 read_extent_buffer(leaf, map + pg_offset, ptr,
5144 copy_size);
Chris Mason93c82d52009-09-11 12:36:29 -04005145 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5146 memset(map + pg_offset + copy_size, 0,
5147 PAGE_CACHE_SIZE - pg_offset -
5148 copy_size);
5149 }
Chris Masonc8b97812008-10-29 14:49:59 -04005150 kunmap(page);
5151 }
Chris Mason179e29e2007-11-01 11:28:41 -04005152 flush_dcache_page(page);
5153 } else if (create && PageUptodate(page)) {
Jan Schmidt6bf7e082011-12-01 14:35:19 +01005154 BUG();
Chris Mason179e29e2007-11-01 11:28:41 -04005155 if (!trans) {
5156 kunmap(page);
5157 free_extent_map(em);
5158 em = NULL;
Chris Masonff5714c2011-05-28 07:00:39 -04005159
David Sterbab3b4aa72011-04-21 01:20:15 +02005160 btrfs_release_path(path);
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005161 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04005162
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005163 if (IS_ERR(trans))
5164 return ERR_CAST(trans);
Chris Mason179e29e2007-11-01 11:28:41 -04005165 goto again;
5166 }
Chris Masonc8b97812008-10-29 14:49:59 -04005167 map = kmap(page);
Chris Mason70dec802008-01-29 09:59:12 -05005168 write_extent_buffer(leaf, map + pg_offset, ptr,
Chris Mason179e29e2007-11-01 11:28:41 -04005169 copy_size);
Chris Masonc8b97812008-10-29 14:49:59 -04005170 kunmap(page);
Chris Mason179e29e2007-11-01 11:28:41 -04005171 btrfs_mark_buffer_dirty(leaf);
Chris Masona52d9a82007-08-27 16:49:44 -04005172 }
Chris Masond1310b22008-01-24 16:13:08 -05005173 set_extent_uptodate(io_tree, em->start,
Arne Jansen507903b2011-04-06 10:02:20 +00005174 extent_map_end(em) - 1, NULL, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04005175 goto insert;
5176 } else {
Chris Masond3977122009-01-05 21:25:51 -05005177 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
Chris Masona52d9a82007-08-27 16:49:44 -04005178 WARN_ON(1);
5179 }
5180not_found:
5181 em->start = start;
Chris Masond1310b22008-01-24 16:13:08 -05005182 em->len = len;
Chris Masona52d9a82007-08-27 16:49:44 -04005183not_found_em:
Chris Mason5f39d392007-10-15 16:14:19 -04005184 em->block_start = EXTENT_MAP_HOLE;
Yan Zheng9036c102008-10-30 14:19:41 -04005185 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -04005186insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02005187 btrfs_release_path(path);
Chris Masond1310b22008-01-24 16:13:08 -05005188 if (em->start > start || extent_map_end(em) <= start) {
Chris Masond3977122009-01-05 21:25:51 -05005189 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5190 "[%llu %llu]\n", (unsigned long long)em->start,
5191 (unsigned long long)em->len,
5192 (unsigned long long)start,
5193 (unsigned long long)len);
Chris Masona52d9a82007-08-27 16:49:44 -04005194 err = -EIO;
5195 goto out;
5196 }
Chris Masond1310b22008-01-24 16:13:08 -05005197
5198 err = 0;
Chris Mason890871b2009-09-02 16:24:52 -04005199 write_lock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04005200 ret = add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04005201 /* it is possible that someone inserted the extent into the tree
5202 * while we had the lock dropped. It is also possible that
5203 * an overlapping map exists in the tree
5204 */
Chris Masona52d9a82007-08-27 16:49:44 -04005205 if (ret == -EEXIST) {
Chris Mason3b951512008-04-17 11:29:12 -04005206 struct extent_map *existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005207
5208 ret = 0;
5209
Chris Mason3b951512008-04-17 11:29:12 -04005210 existing = lookup_extent_mapping(em_tree, start, len);
Chris Masone1c4b742008-04-22 13:26:46 -04005211 if (existing && (existing->start > start ||
5212 existing->start + existing->len <= start)) {
5213 free_extent_map(existing);
5214 existing = NULL;
5215 }
Chris Mason3b951512008-04-17 11:29:12 -04005216 if (!existing) {
5217 existing = lookup_extent_mapping(em_tree, em->start,
5218 em->len);
5219 if (existing) {
5220 err = merge_extent_mapping(em_tree, existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04005221 em, start,
5222 root->sectorsize);
Chris Mason3b951512008-04-17 11:29:12 -04005223 free_extent_map(existing);
5224 if (err) {
5225 free_extent_map(em);
5226 em = NULL;
5227 }
5228 } else {
5229 err = -EIO;
Chris Mason3b951512008-04-17 11:29:12 -04005230 free_extent_map(em);
5231 em = NULL;
5232 }
5233 } else {
5234 free_extent_map(em);
5235 em = existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005236 err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04005237 }
Chris Masona52d9a82007-08-27 16:49:44 -04005238 }
Chris Mason890871b2009-09-02 16:24:52 -04005239 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04005240out:
liubo1abe9b82011-03-24 11:18:59 +00005241
5242 trace_btrfs_get_extent(root, em);
5243
Chris Masonf4219502008-07-22 11:18:09 -04005244 if (path)
5245 btrfs_free_path(path);
Chris Masona52d9a82007-08-27 16:49:44 -04005246 if (trans) {
5247 ret = btrfs_end_transaction(trans, root);
Chris Masond3977122009-01-05 21:25:51 -05005248 if (!err)
Chris Masona52d9a82007-08-27 16:49:44 -04005249 err = ret;
5250 }
Chris Masona52d9a82007-08-27 16:49:44 -04005251 if (err) {
5252 free_extent_map(em);
Chris Masona52d9a82007-08-27 16:49:44 -04005253 return ERR_PTR(err);
5254 }
5255 return em;
5256}
5257
Chris Masonec29ed52011-02-23 16:23:20 -05005258struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5259 size_t pg_offset, u64 start, u64 len,
5260 int create)
5261{
5262 struct extent_map *em;
5263 struct extent_map *hole_em = NULL;
5264 u64 range_start = start;
5265 u64 end;
5266 u64 found;
5267 u64 found_end;
5268 int err = 0;
5269
5270 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5271 if (IS_ERR(em))
5272 return em;
5273 if (em) {
5274 /*
5275 * if our em maps to a hole, there might
5276 * actually be delalloc bytes behind it
5277 */
5278 if (em->block_start != EXTENT_MAP_HOLE)
5279 return em;
5280 else
5281 hole_em = em;
5282 }
5283
5284 /* check to see if we've wrapped (len == -1 or similar) */
5285 end = start + len;
5286 if (end < start)
5287 end = (u64)-1;
5288 else
5289 end -= 1;
5290
5291 em = NULL;
5292
5293 /* ok, we didn't find anything, lets look for delalloc */
5294 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5295 end, len, EXTENT_DELALLOC, 1);
5296 found_end = range_start + found;
5297 if (found_end < range_start)
5298 found_end = (u64)-1;
5299
5300 /*
5301 * we didn't find anything useful, return
5302 * the original results from get_extent()
5303 */
5304 if (range_start > end || found_end <= start) {
5305 em = hole_em;
5306 hole_em = NULL;
5307 goto out;
5308 }
5309
5310 /* adjust the range_start to make sure it doesn't
5311 * go backwards from the start they passed in
5312 */
5313 range_start = max(start,range_start);
5314 found = found_end - range_start;
5315
5316 if (found > 0) {
5317 u64 hole_start = start;
5318 u64 hole_len = len;
5319
David Sterba172ddd62011-04-21 00:48:27 +02005320 em = alloc_extent_map();
Chris Masonec29ed52011-02-23 16:23:20 -05005321 if (!em) {
5322 err = -ENOMEM;
5323 goto out;
5324 }
5325 /*
5326 * when btrfs_get_extent can't find anything it
5327 * returns one huge hole
5328 *
5329 * make sure what it found really fits our range, and
5330 * adjust to make sure it is based on the start from
5331 * the caller
5332 */
5333 if (hole_em) {
5334 u64 calc_end = extent_map_end(hole_em);
5335
5336 if (calc_end <= start || (hole_em->start > end)) {
5337 free_extent_map(hole_em);
5338 hole_em = NULL;
5339 } else {
5340 hole_start = max(hole_em->start, start);
5341 hole_len = calc_end - hole_start;
5342 }
5343 }
5344 em->bdev = NULL;
5345 if (hole_em && range_start > hole_start) {
5346 /* our hole starts before our delalloc, so we
5347 * have to return just the parts of the hole
5348 * that go until the delalloc starts
5349 */
5350 em->len = min(hole_len,
5351 range_start - hole_start);
5352 em->start = hole_start;
5353 em->orig_start = hole_start;
5354 /*
5355 * don't adjust block start at all,
5356 * it is fixed at EXTENT_MAP_HOLE
5357 */
5358 em->block_start = hole_em->block_start;
5359 em->block_len = hole_len;
5360 } else {
5361 em->start = range_start;
5362 em->len = found;
5363 em->orig_start = range_start;
5364 em->block_start = EXTENT_MAP_DELALLOC;
5365 em->block_len = found;
5366 }
5367 } else if (hole_em) {
5368 return hole_em;
5369 }
5370out:
5371
5372 free_extent_map(hole_em);
5373 if (err) {
5374 free_extent_map(em);
5375 return ERR_PTR(err);
5376 }
5377 return em;
5378}
5379
Josef Bacik4b46fce2010-05-23 11:00:55 -04005380static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
Josef Bacik16d299a2011-04-06 14:53:07 -04005381 struct extent_map *em,
Josef Bacik4b46fce2010-05-23 11:00:55 -04005382 u64 start, u64 len)
5383{
5384 struct btrfs_root *root = BTRFS_I(inode)->root;
5385 struct btrfs_trans_handle *trans;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005386 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5387 struct btrfs_key ins;
5388 u64 alloc_hint;
5389 int ret;
Josef Bacik16d299a2011-04-06 14:53:07 -04005390 bool insert = false;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005391
Josef Bacik16d299a2011-04-06 14:53:07 -04005392 /*
5393 * Ok if the extent map we looked up is a hole and is for the exact
5394 * range we want, there is no reason to allocate a new one, however if
5395 * it is not right then we need to free this one and drop the cache for
5396 * our range.
5397 */
5398 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5399 em->len != len) {
5400 free_extent_map(em);
5401 em = NULL;
5402 insert = true;
5403 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5404 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04005405
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005406 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005407 if (IS_ERR(trans))
5408 return ERR_CAST(trans);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005409
Chris Mason4cb53002011-05-24 15:35:30 -04005410 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5411 btrfs_add_inode_defrag(trans, inode);
5412
Josef Bacik4b46fce2010-05-23 11:00:55 -04005413 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5414
5415 alloc_hint = get_extent_allocation_hint(inode, start, len);
5416 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5417 alloc_hint, (u64)-1, &ins, 1);
5418 if (ret) {
5419 em = ERR_PTR(ret);
5420 goto out;
5421 }
5422
Josef Bacik4b46fce2010-05-23 11:00:55 -04005423 if (!em) {
David Sterba172ddd62011-04-21 00:48:27 +02005424 em = alloc_extent_map();
Josef Bacik16d299a2011-04-06 14:53:07 -04005425 if (!em) {
5426 em = ERR_PTR(-ENOMEM);
5427 goto out;
5428 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04005429 }
5430
5431 em->start = start;
5432 em->orig_start = em->start;
5433 em->len = ins.offset;
5434
5435 em->block_start = ins.objectid;
5436 em->block_len = ins.offset;
5437 em->bdev = root->fs_info->fs_devices->latest_bdev;
Josef Bacik16d299a2011-04-06 14:53:07 -04005438
5439 /*
5440 * We need to do this because if we're using the original em we searched
5441 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5442 */
5443 em->flags = 0;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005444 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5445
Josef Bacik16d299a2011-04-06 14:53:07 -04005446 while (insert) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04005447 write_lock(&em_tree->lock);
5448 ret = add_extent_mapping(em_tree, em);
5449 write_unlock(&em_tree->lock);
5450 if (ret != -EEXIST)
5451 break;
5452 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5453 }
5454
5455 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5456 ins.offset, ins.offset, 0);
5457 if (ret) {
5458 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5459 em = ERR_PTR(ret);
5460 }
5461out:
5462 btrfs_end_transaction(trans, root);
5463 return em;
5464}
5465
Chris Mason46bfbb52010-05-26 11:04:10 -04005466/*
5467 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5468 * block must be cow'd
5469 */
5470static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5471 struct inode *inode, u64 offset, u64 len)
5472{
5473 struct btrfs_path *path;
5474 int ret;
5475 struct extent_buffer *leaf;
5476 struct btrfs_root *root = BTRFS_I(inode)->root;
5477 struct btrfs_file_extent_item *fi;
5478 struct btrfs_key key;
5479 u64 disk_bytenr;
5480 u64 backref_offset;
5481 u64 extent_end;
5482 u64 num_bytes;
5483 int slot;
5484 int found_type;
5485
5486 path = btrfs_alloc_path();
5487 if (!path)
5488 return -ENOMEM;
5489
Li Zefan33345d012011-04-20 10:31:50 +08005490 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
Chris Mason46bfbb52010-05-26 11:04:10 -04005491 offset, 0);
5492 if (ret < 0)
5493 goto out;
5494
5495 slot = path->slots[0];
5496 if (ret == 1) {
5497 if (slot == 0) {
5498 /* can't find the item, must cow */
5499 ret = 0;
5500 goto out;
5501 }
5502 slot--;
5503 }
5504 ret = 0;
5505 leaf = path->nodes[0];
5506 btrfs_item_key_to_cpu(leaf, &key, slot);
Li Zefan33345d012011-04-20 10:31:50 +08005507 if (key.objectid != btrfs_ino(inode) ||
Chris Mason46bfbb52010-05-26 11:04:10 -04005508 key.type != BTRFS_EXTENT_DATA_KEY) {
5509 /* not our file or wrong item type, must cow */
5510 goto out;
5511 }
5512
5513 if (key.offset > offset) {
5514 /* Wrong offset, must cow */
5515 goto out;
5516 }
5517
5518 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5519 found_type = btrfs_file_extent_type(leaf, fi);
5520 if (found_type != BTRFS_FILE_EXTENT_REG &&
5521 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5522 /* not a regular extent, must cow */
5523 goto out;
5524 }
5525 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5526 backref_offset = btrfs_file_extent_offset(leaf, fi);
5527
5528 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5529 if (extent_end < offset + len) {
5530 /* extent doesn't include our full range, must cow */
5531 goto out;
5532 }
5533
5534 if (btrfs_extent_readonly(root, disk_bytenr))
5535 goto out;
5536
5537 /*
5538 * look for other files referencing this extent, if we
5539 * find any we must cow
5540 */
Li Zefan33345d012011-04-20 10:31:50 +08005541 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
Chris Mason46bfbb52010-05-26 11:04:10 -04005542 key.offset - backref_offset, disk_bytenr))
5543 goto out;
5544
5545 /*
5546 * adjust disk_bytenr and num_bytes to cover just the bytes
5547 * in this extent we are about to write. If there
5548 * are any csums in that range we have to cow in order
5549 * to keep the csums correct
5550 */
5551 disk_bytenr += backref_offset;
5552 disk_bytenr += offset - key.offset;
5553 num_bytes = min(offset + len, extent_end) - offset;
5554 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5555 goto out;
5556 /*
5557 * all of the above have passed, it is safe to overwrite this extent
5558 * without cow
5559 */
5560 ret = 1;
5561out:
5562 btrfs_free_path(path);
5563 return ret;
5564}
5565
Josef Bacik4b46fce2010-05-23 11:00:55 -04005566static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5567 struct buffer_head *bh_result, int create)
5568{
5569 struct extent_map *em;
5570 struct btrfs_root *root = BTRFS_I(inode)->root;
5571 u64 start = iblock << inode->i_blkbits;
5572 u64 len = bh_result->b_size;
Chris Mason46bfbb52010-05-26 11:04:10 -04005573 struct btrfs_trans_handle *trans;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005574
5575 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5576 if (IS_ERR(em))
5577 return PTR_ERR(em);
5578
5579 /*
5580 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5581 * io. INLINE is special, and we could probably kludge it in here, but
5582 * it's still buffered so for safety lets just fall back to the generic
5583 * buffered path.
5584 *
5585 * For COMPRESSED we _have_ to read the entire extent in so we can
5586 * decompress it, so there will be buffering required no matter what we
5587 * do, so go ahead and fallback to buffered.
5588 *
5589 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5590 * to buffered IO. Don't blame me, this is the price we pay for using
5591 * the generic code.
5592 */
5593 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5594 em->block_start == EXTENT_MAP_INLINE) {
5595 free_extent_map(em);
5596 return -ENOTBLK;
5597 }
5598
5599 /* Just a good old fashioned hole, return */
5600 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5601 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5602 free_extent_map(em);
5603 /* DIO will do one hole at a time, so just unlock a sector */
5604 unlock_extent(&BTRFS_I(inode)->io_tree, start,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01005605 start + root->sectorsize - 1);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005606 return 0;
5607 }
5608
5609 /*
5610 * We don't allocate a new extent in the following cases
5611 *
5612 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5613 * existing extent.
5614 * 2) The extent is marked as PREALLOC. We're good to go here and can
5615 * just use the extent.
5616 *
5617 */
Chris Mason46bfbb52010-05-26 11:04:10 -04005618 if (!create) {
5619 len = em->len - (start - em->start);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005620 goto map;
Chris Mason46bfbb52010-05-26 11:04:10 -04005621 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04005622
5623 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5624 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5625 em->block_start != EXTENT_MAP_HOLE)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04005626 int type;
5627 int ret;
Chris Mason46bfbb52010-05-26 11:04:10 -04005628 u64 block_start;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005629
5630 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5631 type = BTRFS_ORDERED_PREALLOC;
5632 else
5633 type = BTRFS_ORDERED_NOCOW;
Chris Mason46bfbb52010-05-26 11:04:10 -04005634 len = min(len, em->len - (start - em->start));
Josef Bacik4b46fce2010-05-23 11:00:55 -04005635 block_start = em->block_start + (start - em->start);
Chris Mason46bfbb52010-05-26 11:04:10 -04005636
5637 /*
5638 * we're not going to log anything, but we do need
5639 * to make sure the current transaction stays open
5640 * while we look for nocow cross refs
5641 */
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005642 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005643 if (IS_ERR(trans))
Chris Mason46bfbb52010-05-26 11:04:10 -04005644 goto must_cow;
5645
5646 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5647 ret = btrfs_add_ordered_extent_dio(inode, start,
5648 block_start, len, len, type);
5649 btrfs_end_transaction(trans, root);
5650 if (ret) {
5651 free_extent_map(em);
5652 return ret;
5653 }
5654 goto unlock;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005655 }
Chris Mason46bfbb52010-05-26 11:04:10 -04005656 btrfs_end_transaction(trans, root);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005657 }
Chris Mason46bfbb52010-05-26 11:04:10 -04005658must_cow:
5659 /*
5660 * this will cow the extent, reset the len in case we changed
5661 * it above
5662 */
5663 len = bh_result->b_size;
Josef Bacik16d299a2011-04-06 14:53:07 -04005664 em = btrfs_new_extent_direct(inode, em, start, len);
Chris Mason46bfbb52010-05-26 11:04:10 -04005665 if (IS_ERR(em))
5666 return PTR_ERR(em);
5667 len = min(len, em->len - (start - em->start));
5668unlock:
Chris Mason4845e442010-05-25 20:56:50 -04005669 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5670 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5671 0, NULL, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005672map:
5673 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5674 inode->i_blkbits;
Chris Mason46bfbb52010-05-26 11:04:10 -04005675 bh_result->b_size = len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005676 bh_result->b_bdev = em->bdev;
5677 set_buffer_mapped(bh_result);
5678 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5679 set_buffer_new(bh_result);
5680
5681 free_extent_map(em);
5682
5683 return 0;
5684}
5685
5686struct btrfs_dio_private {
5687 struct inode *inode;
5688 u64 logical_offset;
5689 u64 disk_bytenr;
5690 u64 bytes;
5691 u32 *csums;
5692 void *private;
Miao Xiee65e1532010-11-22 03:04:43 +00005693
5694 /* number of bios pending for this dio */
5695 atomic_t pending_bios;
5696
5697 /* IO errors */
5698 int errors;
5699
5700 struct bio *orig_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005701};
5702
5703static void btrfs_endio_direct_read(struct bio *bio, int err)
5704{
Miao Xiee65e1532010-11-22 03:04:43 +00005705 struct btrfs_dio_private *dip = bio->bi_private;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005706 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5707 struct bio_vec *bvec = bio->bi_io_vec;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005708 struct inode *inode = dip->inode;
5709 struct btrfs_root *root = BTRFS_I(inode)->root;
5710 u64 start;
5711 u32 *private = dip->csums;
5712
5713 start = dip->logical_offset;
5714 do {
5715 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5716 struct page *page = bvec->bv_page;
5717 char *kaddr;
5718 u32 csum = ~(u32)0;
5719 unsigned long flags;
5720
5721 local_irq_save(flags);
5722 kaddr = kmap_atomic(page, KM_IRQ0);
5723 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5724 csum, bvec->bv_len);
5725 btrfs_csum_final(csum, (char *)&csum);
5726 kunmap_atomic(kaddr, KM_IRQ0);
5727 local_irq_restore(flags);
5728
5729 flush_dcache_page(bvec->bv_page);
5730 if (csum != *private) {
Li Zefan33345d012011-04-20 10:31:50 +08005731 printk(KERN_ERR "btrfs csum failed ino %llu off"
Josef Bacik4b46fce2010-05-23 11:00:55 -04005732 " %llu csum %u private %u\n",
Li Zefan33345d012011-04-20 10:31:50 +08005733 (unsigned long long)btrfs_ino(inode),
5734 (unsigned long long)start,
Josef Bacik4b46fce2010-05-23 11:00:55 -04005735 csum, *private);
5736 err = -EIO;
5737 }
5738 }
5739
5740 start += bvec->bv_len;
5741 private++;
5742 bvec++;
5743 } while (bvec <= bvec_end);
5744
5745 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01005746 dip->logical_offset + dip->bytes - 1);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005747 bio->bi_private = dip->private;
5748
5749 kfree(dip->csums);
5750 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04005751
5752 /* If we had a csum failure make sure to clear the uptodate flag */
5753 if (err)
5754 clear_bit(BIO_UPTODATE, &bio->bi_flags);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005755 dio_end_io(bio, err);
5756}
5757
5758static void btrfs_endio_direct_write(struct bio *bio, int err)
5759{
5760 struct btrfs_dio_private *dip = bio->bi_private;
5761 struct inode *inode = dip->inode;
5762 struct btrfs_root *root = BTRFS_I(inode)->root;
5763 struct btrfs_trans_handle *trans;
5764 struct btrfs_ordered_extent *ordered = NULL;
5765 struct extent_state *cached_state = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05005766 u64 ordered_offset = dip->logical_offset;
5767 u64 ordered_bytes = dip->bytes;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005768 int ret;
5769
5770 if (err)
5771 goto out_done;
Chris Mason163cf092010-11-28 19:56:33 -05005772again:
5773 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5774 &ordered_offset,
5775 ordered_bytes);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005776 if (!ret)
Chris Mason163cf092010-11-28 19:56:33 -05005777 goto out_test;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005778
5779 BUG_ON(!ordered);
5780
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005781 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005782 if (IS_ERR(trans)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04005783 err = -ENOMEM;
5784 goto out;
5785 }
5786 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5787
5788 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5789 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5790 if (!ret)
Chris Mason21151332011-11-10 20:39:08 -05005791 err = btrfs_update_inode_fallback(trans, root, inode);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005792 goto out;
5793 }
5794
5795 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5796 ordered->file_offset + ordered->len - 1, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01005797 &cached_state);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005798
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);
Josef Bacik1ef30be2011-04-05 19:25:36 -04005827 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
Miao Xiea39f7522011-09-11 10:52:25 -04005828 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
Chris Mason21151332011-11-10 20:39:08 -05005829 btrfs_update_inode_fallback(trans, root, inode);
Josef Bacik1ef30be2011-04-05 19:25:36 -04005830 ret = 0;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005831out_unlock:
5832 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5833 ordered->file_offset + ordered->len - 1,
5834 &cached_state, GFP_NOFS);
5835out:
5836 btrfs_delalloc_release_metadata(inode, ordered->len);
5837 btrfs_end_transaction(trans, root);
Chris Mason163cf092010-11-28 19:56:33 -05005838 ordered_offset = ordered->file_offset + ordered->len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005839 btrfs_put_ordered_extent(ordered);
5840 btrfs_put_ordered_extent(ordered);
Chris Mason163cf092010-11-28 19:56:33 -05005841
5842out_test:
5843 /*
5844 * our bio might span multiple ordered extents. If we haven't
5845 * completed the accounting for the whole dio, go back and try again
5846 */
5847 if (ordered_offset < dip->logical_offset + dip->bytes) {
5848 ordered_bytes = dip->logical_offset + dip->bytes -
5849 ordered_offset;
5850 goto again;
5851 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04005852out_done:
5853 bio->bi_private = dip->private;
5854
5855 kfree(dip->csums);
5856 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04005857
5858 /* If we had an error make sure to clear the uptodate flag */
5859 if (err)
5860 clear_bit(BIO_UPTODATE, &bio->bi_flags);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005861 dio_end_io(bio, err);
5862}
5863
Chris Masoneaf25d92010-05-25 09:48:28 -04005864static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5865 struct bio *bio, int mirror_num,
5866 unsigned long bio_flags, u64 offset)
5867{
5868 int ret;
5869 struct btrfs_root *root = BTRFS_I(inode)->root;
5870 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5871 BUG_ON(ret);
5872 return 0;
5873}
5874
Miao Xiee65e1532010-11-22 03:04:43 +00005875static void btrfs_end_dio_bio(struct bio *bio, int err)
5876{
5877 struct btrfs_dio_private *dip = bio->bi_private;
5878
5879 if (err) {
Li Zefan33345d012011-04-20 10:31:50 +08005880 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
Jan Beulich3dd14622010-12-07 14:54:09 +00005881 "sector %#Lx len %u err no %d\n",
Li Zefan33345d012011-04-20 10:31:50 +08005882 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
Jan Beulich3dd14622010-12-07 14:54:09 +00005883 (unsigned long long)bio->bi_sector, bio->bi_size, err);
Miao Xiee65e1532010-11-22 03:04:43 +00005884 dip->errors = 1;
5885
5886 /*
5887 * before atomic variable goto zero, we must make sure
5888 * dip->errors is perceived to be set.
5889 */
5890 smp_mb__before_atomic_dec();
5891 }
5892
5893 /* if there are more bios still pending for this dio, just exit */
5894 if (!atomic_dec_and_test(&dip->pending_bios))
5895 goto out;
5896
5897 if (dip->errors)
5898 bio_io_error(dip->orig_bio);
5899 else {
5900 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5901 bio_endio(dip->orig_bio, 0);
5902 }
5903out:
5904 bio_put(bio);
5905}
5906
5907static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5908 u64 first_sector, gfp_t gfp_flags)
5909{
5910 int nr_vecs = bio_get_nr_vecs(bdev);
5911 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5912}
5913
5914static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5915 int rw, u64 file_offset, int skip_sum,
Josef Bacik1ae39932011-04-06 14:41:34 -04005916 u32 *csums, int async_submit)
Miao Xiee65e1532010-11-22 03:04:43 +00005917{
5918 int write = rw & REQ_WRITE;
5919 struct btrfs_root *root = BTRFS_I(inode)->root;
5920 int ret;
5921
5922 bio_get(bio);
5923 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5924 if (ret)
5925 goto err;
5926
Josef Bacik1ae39932011-04-06 14:41:34 -04005927 if (skip_sum)
5928 goto map;
5929
5930 if (write && async_submit) {
Miao Xiee65e1532010-11-22 03:04:43 +00005931 ret = btrfs_wq_submit_bio(root->fs_info,
5932 inode, rw, bio, 0, 0,
5933 file_offset,
5934 __btrfs_submit_bio_start_direct_io,
5935 __btrfs_submit_bio_done);
5936 goto err;
Josef Bacik1ae39932011-04-06 14:41:34 -04005937 } else if (write) {
5938 /*
5939 * If we aren't doing async submit, calculate the csum of the
5940 * bio now.
5941 */
5942 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5943 if (ret)
5944 goto err;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00005945 } else if (!skip_sum) {
5946 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
Miao Xiee65e1532010-11-22 03:04:43 +00005947 file_offset, csums);
Tsutomu Itohc2db1072011-03-01 06:48:31 +00005948 if (ret)
5949 goto err;
5950 }
Miao Xiee65e1532010-11-22 03:04:43 +00005951
Josef Bacik1ae39932011-04-06 14:41:34 -04005952map:
5953 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00005954err:
5955 bio_put(bio);
5956 return ret;
5957}
5958
5959static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5960 int skip_sum)
5961{
5962 struct inode *inode = dip->inode;
5963 struct btrfs_root *root = BTRFS_I(inode)->root;
5964 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5965 struct bio *bio;
5966 struct bio *orig_bio = dip->orig_bio;
5967 struct bio_vec *bvec = orig_bio->bi_io_vec;
5968 u64 start_sector = orig_bio->bi_sector;
5969 u64 file_offset = dip->logical_offset;
5970 u64 submit_len = 0;
5971 u64 map_length;
5972 int nr_pages = 0;
5973 u32 *csums = dip->csums;
5974 int ret = 0;
Josef Bacik1ae39932011-04-06 14:41:34 -04005975 int async_submit = 0;
Josef Bacik98bc3142011-03-22 11:00:46 -04005976 int write = rw & REQ_WRITE;
Miao Xiee65e1532010-11-22 03:04:43 +00005977
Miao Xiee65e1532010-11-22 03:04:43 +00005978 map_length = orig_bio->bi_size;
5979 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5980 &map_length, NULL, 0);
5981 if (ret) {
Josef Bacik64728bb2011-04-25 19:43:52 -04005982 bio_put(orig_bio);
Miao Xiee65e1532010-11-22 03:04:43 +00005983 return -EIO;
5984 }
5985
Josef Bacik02f57c72011-04-06 14:25:44 -04005986 if (map_length >= orig_bio->bi_size) {
5987 bio = orig_bio;
5988 goto submit;
5989 }
5990
Josef Bacik1ae39932011-04-06 14:41:34 -04005991 async_submit = 1;
Josef Bacik02f57c72011-04-06 14:25:44 -04005992 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5993 if (!bio)
5994 return -ENOMEM;
5995 bio->bi_private = dip;
5996 bio->bi_end_io = btrfs_end_dio_bio;
5997 atomic_inc(&dip->pending_bios);
5998
Miao Xiee65e1532010-11-22 03:04:43 +00005999 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6000 if (unlikely(map_length < submit_len + bvec->bv_len ||
6001 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6002 bvec->bv_offset) < bvec->bv_len)) {
6003 /*
6004 * inc the count before we submit the bio so
6005 * we know the end IO handler won't happen before
6006 * we inc the count. Otherwise, the dip might get freed
6007 * before we're done setting it up
6008 */
6009 atomic_inc(&dip->pending_bios);
6010 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6011 file_offset, skip_sum,
Josef Bacik1ae39932011-04-06 14:41:34 -04006012 csums, async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00006013 if (ret) {
6014 bio_put(bio);
6015 atomic_dec(&dip->pending_bios);
6016 goto out_err;
6017 }
6018
Josef Bacik98bc3142011-03-22 11:00:46 -04006019 /* Write's use the ordered csums */
6020 if (!write && !skip_sum)
Miao Xiee65e1532010-11-22 03:04:43 +00006021 csums = csums + nr_pages;
6022 start_sector += submit_len >> 9;
6023 file_offset += submit_len;
6024
6025 submit_len = 0;
6026 nr_pages = 0;
6027
6028 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6029 start_sector, GFP_NOFS);
6030 if (!bio)
6031 goto out_err;
6032 bio->bi_private = dip;
6033 bio->bi_end_io = btrfs_end_dio_bio;
6034
6035 map_length = orig_bio->bi_size;
6036 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6037 &map_length, NULL, 0);
6038 if (ret) {
6039 bio_put(bio);
6040 goto out_err;
6041 }
6042 } else {
6043 submit_len += bvec->bv_len;
6044 nr_pages ++;
6045 bvec++;
6046 }
6047 }
6048
Josef Bacik02f57c72011-04-06 14:25:44 -04006049submit:
Miao Xiee65e1532010-11-22 03:04:43 +00006050 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
Josef Bacik1ae39932011-04-06 14:41:34 -04006051 csums, async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00006052 if (!ret)
6053 return 0;
6054
6055 bio_put(bio);
6056out_err:
6057 dip->errors = 1;
6058 /*
6059 * before atomic variable goto zero, we must
6060 * make sure dip->errors is perceived to be set.
6061 */
6062 smp_mb__before_atomic_dec();
6063 if (atomic_dec_and_test(&dip->pending_bios))
6064 bio_io_error(dip->orig_bio);
6065
6066 /* bio_end_io() will handle error, so we needn't return it */
6067 return 0;
6068}
6069
Josef Bacik4b46fce2010-05-23 11:00:55 -04006070static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6071 loff_t file_offset)
6072{
6073 struct btrfs_root *root = BTRFS_I(inode)->root;
6074 struct btrfs_dio_private *dip;
6075 struct bio_vec *bvec = bio->bi_io_vec;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006076 int skip_sum;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02006077 int write = rw & REQ_WRITE;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006078 int ret = 0;
6079
6080 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6081
6082 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6083 if (!dip) {
6084 ret = -ENOMEM;
6085 goto free_ordered;
6086 }
6087 dip->csums = NULL;
6088
Josef Bacik98bc3142011-03-22 11:00:46 -04006089 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6090 if (!write && !skip_sum) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04006091 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6092 if (!dip->csums) {
Daniel J Bluemanb4966b72011-03-09 16:46:42 +00006093 kfree(dip);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006094 ret = -ENOMEM;
6095 goto free_ordered;
6096 }
6097 }
6098
6099 dip->private = bio->bi_private;
6100 dip->inode = inode;
6101 dip->logical_offset = file_offset;
6102
Josef Bacik4b46fce2010-05-23 11:00:55 -04006103 dip->bytes = 0;
6104 do {
6105 dip->bytes += bvec->bv_len;
6106 bvec++;
6107 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6108
Chris Mason46bfbb52010-05-26 11:04:10 -04006109 dip->disk_bytenr = (u64)bio->bi_sector << 9;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006110 bio->bi_private = dip;
Miao Xiee65e1532010-11-22 03:04:43 +00006111 dip->errors = 0;
6112 dip->orig_bio = bio;
6113 atomic_set(&dip->pending_bios, 0);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006114
6115 if (write)
6116 bio->bi_end_io = btrfs_endio_direct_write;
6117 else
6118 bio->bi_end_io = btrfs_endio_direct_read;
6119
Miao Xiee65e1532010-11-22 03:04:43 +00006120 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6121 if (!ret)
Chris Masoneaf25d92010-05-25 09:48:28 -04006122 return;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006123free_ordered:
6124 /*
6125 * If this is a write, we need to clean up the reserved space and kill
6126 * the ordered extent.
6127 */
6128 if (write) {
6129 struct btrfs_ordered_extent *ordered;
Josef Bacik955256f2010-11-19 09:41:10 -05006130 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006131 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6132 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6133 btrfs_free_reserved_extent(root, ordered->start,
6134 ordered->disk_len);
6135 btrfs_put_ordered_extent(ordered);
6136 btrfs_put_ordered_extent(ordered);
6137 }
6138 bio_endio(bio, ret);
6139}
6140
Chris Mason5a5f79b2010-05-26 21:33:37 -04006141static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6142 const struct iovec *iov, loff_t offset,
6143 unsigned long nr_segs)
6144{
6145 int seg;
Josef Bacika1b75f72011-04-08 15:51:18 +00006146 int i;
Chris Mason5a5f79b2010-05-26 21:33:37 -04006147 size_t size;
6148 unsigned long addr;
6149 unsigned blocksize_mask = root->sectorsize - 1;
6150 ssize_t retval = -EINVAL;
6151 loff_t end = offset;
6152
6153 if (offset & blocksize_mask)
6154 goto out;
6155
6156 /* Check the memory alignment. Blocks cannot straddle pages */
6157 for (seg = 0; seg < nr_segs; seg++) {
6158 addr = (unsigned long)iov[seg].iov_base;
6159 size = iov[seg].iov_len;
6160 end += size;
Josef Bacika1b75f72011-04-08 15:51:18 +00006161 if ((addr & blocksize_mask) || (size & blocksize_mask))
Chris Mason5a5f79b2010-05-26 21:33:37 -04006162 goto out;
Josef Bacika1b75f72011-04-08 15:51:18 +00006163
6164 /* If this is a write we don't need to check anymore */
6165 if (rw & WRITE)
6166 continue;
6167
6168 /*
6169 * Check to make sure we don't have duplicate iov_base's in this
6170 * iovec, if so return EINVAL, otherwise we'll get csum errors
6171 * when reading back.
6172 */
6173 for (i = seg + 1; i < nr_segs; i++) {
6174 if (iov[seg].iov_base == iov[i].iov_base)
6175 goto out;
6176 }
Chris Mason5a5f79b2010-05-26 21:33:37 -04006177 }
6178 retval = 0;
6179out:
6180 return retval;
6181}
Chris Mason16432982008-04-10 10:23:21 -04006182static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6183 const struct iovec *iov, loff_t offset,
6184 unsigned long nr_segs)
6185{
Josef Bacik4b46fce2010-05-23 11:00:55 -04006186 struct file *file = iocb->ki_filp;
6187 struct inode *inode = file->f_mapping->host;
6188 struct btrfs_ordered_extent *ordered;
Chris Mason4845e442010-05-25 20:56:50 -04006189 struct extent_state *cached_state = NULL;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006190 u64 lockstart, lockend;
6191 ssize_t ret;
Chris Mason4845e442010-05-25 20:56:50 -04006192 int writing = rw & WRITE;
6193 int write_bits = 0;
Chris Mason3f7c5792010-05-26 10:59:53 -04006194 size_t count = iov_length(iov, nr_segs);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006195
Chris Mason5a5f79b2010-05-26 21:33:37 -04006196 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6197 offset, nr_segs)) {
6198 return 0;
6199 }
6200
Josef Bacik4b46fce2010-05-23 11:00:55 -04006201 lockstart = offset;
Chris Mason3f7c5792010-05-26 10:59:53 -04006202 lockend = offset + count - 1;
6203
6204 if (writing) {
6205 ret = btrfs_delalloc_reserve_space(inode, count);
6206 if (ret)
6207 goto out;
6208 }
Chris Mason4845e442010-05-25 20:56:50 -04006209
Josef Bacik4b46fce2010-05-23 11:00:55 -04006210 while (1) {
Chris Mason4845e442010-05-25 20:56:50 -04006211 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01006212 0, &cached_state);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006213 /*
6214 * We're concerned with the entire range that we're going to be
6215 * doing DIO to, so we need to make sure theres no ordered
6216 * extents in this range.
6217 */
6218 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6219 lockend - lockstart + 1);
6220 if (!ordered)
6221 break;
Chris Mason4845e442010-05-25 20:56:50 -04006222 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6223 &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006224 btrfs_start_ordered_extent(inode, ordered, 1);
6225 btrfs_put_ordered_extent(ordered);
6226 cond_resched();
6227 }
6228
Chris Mason4845e442010-05-25 20:56:50 -04006229 /*
6230 * we don't use btrfs_set_extent_delalloc because we don't want
6231 * the dirty or uptodate bits
6232 */
6233 if (writing) {
6234 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6235 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
Jeff Mahoney3fbe5c02012-03-01 14:57:19 +01006236 EXTENT_DELALLOC, NULL, &cached_state,
Chris Mason4845e442010-05-25 20:56:50 -04006237 GFP_NOFS);
6238 if (ret) {
6239 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6240 lockend, EXTENT_LOCKED | write_bits,
6241 1, 0, &cached_state, GFP_NOFS);
6242 goto out;
6243 }
6244 }
6245
6246 free_extent_state(cached_state);
6247 cached_state = NULL;
6248
Chris Mason5a5f79b2010-05-26 21:33:37 -04006249 ret = __blockdev_direct_IO(rw, iocb, inode,
6250 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6251 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6252 btrfs_submit_direct, 0);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006253
6254 if (ret < 0 && ret != -EIOCBQUEUED) {
Chris Mason4845e442010-05-25 20:56:50 -04006255 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6256 offset + iov_length(iov, nr_segs) - 1,
6257 EXTENT_LOCKED | write_bits, 1, 0,
6258 &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006259 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6260 /*
6261 * We're falling back to buffered, unlock the section we didn't
6262 * do IO on.
6263 */
Chris Mason4845e442010-05-25 20:56:50 -04006264 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6265 offset + iov_length(iov, nr_segs) - 1,
6266 EXTENT_LOCKED | write_bits, 1, 0,
6267 &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006268 }
Chris Mason4845e442010-05-25 20:56:50 -04006269out:
6270 free_extent_state(cached_state);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006271 return ret;
Chris Mason16432982008-04-10 10:23:21 -04006272}
6273
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05006274static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6275 __u64 start, __u64 len)
6276{
Chris Masonec29ed52011-02-23 16:23:20 -05006277 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05006278}
6279
Chris Mason9ebefb182007-06-15 13:50:00 -04006280int btrfs_readpage(struct file *file, struct page *page)
6281{
Chris Masond1310b22008-01-24 16:13:08 -05006282 struct extent_io_tree *tree;
6283 tree = &BTRFS_I(page->mapping->host)->io_tree;
Jan Schmidt8ddc7d92011-06-13 20:02:58 +02006284 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04006285}
Chris Mason1832a6d2007-12-21 16:27:21 -05006286
Chris Mason39279cc2007-06-12 06:35:45 -04006287static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6288{
Chris Masond1310b22008-01-24 16:13:08 -05006289 struct extent_io_tree *tree;
Chris Masonb888db22007-08-27 16:49:44 -04006290
6291
6292 if (current->flags & PF_MEMALLOC) {
6293 redirty_page_for_writepage(wbc, page);
6294 unlock_page(page);
6295 return 0;
6296 }
Chris Masond1310b22008-01-24 16:13:08 -05006297 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04006298 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6299}
Chris Mason39279cc2007-06-12 06:35:45 -04006300
Chris Masonf4219502008-07-22 11:18:09 -04006301int btrfs_writepages(struct address_space *mapping,
6302 struct writeback_control *wbc)
Chris Masonb293f022007-11-01 19:45:34 -04006303{
Chris Masond1310b22008-01-24 16:13:08 -05006304 struct extent_io_tree *tree;
Chris Mason771ed682008-11-06 22:02:51 -05006305
Chris Masond1310b22008-01-24 16:13:08 -05006306 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Masonb293f022007-11-01 19:45:34 -04006307 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6308}
6309
Chris Mason3ab2fb52007-11-08 10:59:22 -05006310static int
6311btrfs_readpages(struct file *file, struct address_space *mapping,
6312 struct list_head *pages, unsigned nr_pages)
6313{
Chris Masond1310b22008-01-24 16:13:08 -05006314 struct extent_io_tree *tree;
6315 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Mason3ab2fb52007-11-08 10:59:22 -05006316 return extent_readpages(tree, mapping, pages, nr_pages,
6317 btrfs_get_extent);
6318}
Chris Masone6dcd2d2008-07-17 12:53:50 -04006319static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
Chris Masona52d9a82007-08-27 16:49:44 -04006320{
Chris Masond1310b22008-01-24 16:13:08 -05006321 struct extent_io_tree *tree;
6322 struct extent_map_tree *map;
Chris Masona52d9a82007-08-27 16:49:44 -04006323 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04006324
Chris Masond1310b22008-01-24 16:13:08 -05006325 tree = &BTRFS_I(page->mapping->host)->io_tree;
6326 map = &BTRFS_I(page->mapping->host)->extent_tree;
Chris Mason70dec802008-01-29 09:59:12 -05006327 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
Chris Masona52d9a82007-08-27 16:49:44 -04006328 if (ret == 1) {
6329 ClearPagePrivate(page);
6330 set_page_private(page, 0);
6331 page_cache_release(page);
6332 }
6333 return ret;
6334}
Chris Mason39279cc2007-06-12 06:35:45 -04006335
Chris Masone6dcd2d2008-07-17 12:53:50 -04006336static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6337{
Chris Mason98509cf2008-09-11 15:51:43 -04006338 if (PageWriteback(page) || PageDirty(page))
6339 return 0;
Yan Zhengb335b002009-02-12 10:06:04 -05006340 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006341}
6342
Chris Masona52d9a82007-08-27 16:49:44 -04006343static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6344{
Chris Masond1310b22008-01-24 16:13:08 -05006345 struct extent_io_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006346 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00006347 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006348 u64 page_start = page_offset(page);
6349 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04006350
Chris Mason8b62b722009-09-02 16:53:46 -04006351
6352 /*
6353 * we have the page locked, so new writeback can't start,
6354 * and the dirty bit won't be cleared while we are here.
6355 *
6356 * Wait for IO on this page so that we can safely clear
6357 * the PagePrivate2 bit and do ordered accounting
6358 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006359 wait_on_page_writeback(page);
Chris Mason8b62b722009-09-02 16:53:46 -04006360
Chris Masond1310b22008-01-24 16:13:08 -05006361 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006362 if (offset) {
6363 btrfs_releasepage(page, GFP_NOFS);
6364 return;
6365 }
Jeff Mahoneyd0082372012-03-01 14:57:19 +01006366 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006367 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6368 page_offset(page));
6369 if (ordered) {
Chris Masoneb84ae02008-07-17 13:53:27 -04006370 /*
6371 * IO on this page will never be started, so we need
6372 * to account for any ordered extents now
6373 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006374 clear_extent_bit(tree, page_start, page_end,
6375 EXTENT_DIRTY | EXTENT_DELALLOC |
Josef Bacik32c00af2009-10-08 13:34:05 -04006376 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
Josef Bacik2ac55d42010-02-03 19:33:23 +00006377 &cached_state, GFP_NOFS);
Chris Mason8b62b722009-09-02 16:53:46 -04006378 /*
6379 * whoever cleared the private bit is responsible
6380 * for the finish_ordered_io
6381 */
6382 if (TestClearPagePrivate2(page)) {
6383 btrfs_finish_ordered_io(page->mapping->host,
6384 page_start, page_end);
6385 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006386 btrfs_put_ordered_extent(ordered);
Josef Bacik2ac55d42010-02-03 19:33:23 +00006387 cached_state = NULL;
Jeff Mahoneyd0082372012-03-01 14:57:19 +01006388 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006389 }
6390 clear_extent_bit(tree, page_start, page_end,
Josef Bacik32c00af2009-10-08 13:34:05 -04006391 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
Josef Bacik2ac55d42010-02-03 19:33:23 +00006392 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006393 __btrfs_releasepage(page, GFP_NOFS);
6394
Chris Mason4a096752008-07-21 10:29:44 -04006395 ClearPageChecked(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04006396 if (PagePrivate(page)) {
Chris Mason9ad6b7b2008-04-18 16:11:30 -04006397 ClearPagePrivate(page);
6398 set_page_private(page, 0);
6399 page_cache_release(page);
6400 }
Chris Mason39279cc2007-06-12 06:35:45 -04006401}
6402
Chris Mason9ebefb182007-06-15 13:50:00 -04006403/*
6404 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6405 * called from a page fault handler when a page is first dirtied. Hence we must
6406 * be careful to check for EOF conditions here. We set the page up correctly
6407 * for a written page which means we get ENOSPC checking when writing into
6408 * holes and correct delalloc and unwritten extent mapping on filesystems that
6409 * support these features.
6410 *
6411 * We are not allowed to take the i_mutex here so we have to play games to
6412 * protect against truncate races as the page could now be beyond EOF. Because
6413 * vmtruncate() writes the inode size before removing pages, once we have the
6414 * page lock we can determine safely if the page is beyond EOF. If it is not
6415 * beyond EOF, then the page is guaranteed safe against truncation until we
6416 * unlock the page.
6417 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07006418int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Chris Mason9ebefb182007-06-15 13:50:00 -04006419{
Nick Pigginc2ec1752009-03-31 15:23:21 -07006420 struct page *page = vmf->page;
Chris Mason6da6aba2007-12-18 16:15:09 -05006421 struct inode *inode = fdentry(vma->vm_file)->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05006422 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006423 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6424 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00006425 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006426 char *kaddr;
6427 unsigned long zero_start;
Chris Mason9ebefb182007-06-15 13:50:00 -04006428 loff_t size;
Chris Mason1832a6d2007-12-21 16:27:21 -05006429 int ret;
Chris Mason9998eb72012-01-25 13:47:40 -05006430 int reserved = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04006431 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006432 u64 page_end;
Chris Mason9ebefb182007-06-15 13:50:00 -04006433
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04006434 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Chris Mason9998eb72012-01-25 13:47:40 -05006435 if (!ret) {
Josef Bacik22c44fe2011-11-30 10:45:38 -05006436 ret = btrfs_update_time(vma->vm_file);
Chris Mason9998eb72012-01-25 13:47:40 -05006437 reserved = 1;
6438 }
Nick Piggin56a76f82009-03-31 15:23:23 -07006439 if (ret) {
6440 if (ret == -ENOMEM)
6441 ret = VM_FAULT_OOM;
6442 else /* -ENOSPC, -EIO, etc */
6443 ret = VM_FAULT_SIGBUS;
Chris Mason9998eb72012-01-25 13:47:40 -05006444 if (reserved)
6445 goto out;
6446 goto out_noreserve;
Nick Piggin56a76f82009-03-31 15:23:23 -07006447 }
Chris Mason1832a6d2007-12-21 16:27:21 -05006448
Nick Piggin56a76f82009-03-31 15:23:23 -07006449 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006450again:
Chris Mason9ebefb182007-06-15 13:50:00 -04006451 lock_page(page);
Chris Mason9ebefb182007-06-15 13:50:00 -04006452 size = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006453 page_start = page_offset(page);
6454 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04006455
Chris Mason9ebefb182007-06-15 13:50:00 -04006456 if ((page->mapping != inode->i_mapping) ||
Chris Masone6dcd2d2008-07-17 12:53:50 -04006457 (page_start >= size)) {
Chris Mason9ebefb182007-06-15 13:50:00 -04006458 /* page got truncated out from underneath us */
6459 goto out_unlock;
6460 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006461 wait_on_page_writeback(page);
6462
Jeff Mahoneyd0082372012-03-01 14:57:19 +01006463 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006464 set_page_extent_mapped(page);
6465
Chris Masoneb84ae02008-07-17 13:53:27 -04006466 /*
6467 * we can't set the delalloc bits if there are pending ordered
6468 * extents. Drop our locks and wait for them to finish
6469 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006470 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6471 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00006472 unlock_extent_cached(io_tree, page_start, page_end,
6473 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006474 unlock_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04006475 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006476 btrfs_put_ordered_extent(ordered);
6477 goto again;
6478 }
6479
Josef Bacikfbf19082009-10-01 17:10:23 -04006480 /*
6481 * XXX - page_mkwrite gets called every time the page is dirtied, even
6482 * if it was already dirty, so for space accounting reasons we need to
6483 * clear any delalloc bits for the range we are fixing to save. There
6484 * is probably a better way to do this, but for now keep consistent with
6485 * prepare_pages in the normal write path.
6486 */
Josef Bacik2ac55d42010-02-03 19:33:23 +00006487 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Josef Bacik32c00af2009-10-08 13:34:05 -04006488 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
Josef Bacik2ac55d42010-02-03 19:33:23 +00006489 0, 0, &cached_state, GFP_NOFS);
Josef Bacikfbf19082009-10-01 17:10:23 -04006490
Josef Bacik2ac55d42010-02-03 19:33:23 +00006491 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6492 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006493 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00006494 unlock_extent_cached(io_tree, page_start, page_end,
6495 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006496 ret = VM_FAULT_SIGBUS;
6497 goto out_unlock;
6498 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006499 ret = 0;
Chris Mason9ebefb182007-06-15 13:50:00 -04006500
6501 /* page is wholly or partially inside EOF */
Chris Masona52d9a82007-08-27 16:49:44 -04006502 if (page_start + PAGE_CACHE_SIZE > size)
Chris Masone6dcd2d2008-07-17 12:53:50 -04006503 zero_start = size & ~PAGE_CACHE_MASK;
Chris Mason9ebefb182007-06-15 13:50:00 -04006504 else
Chris Masone6dcd2d2008-07-17 12:53:50 -04006505 zero_start = PAGE_CACHE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04006506
Chris Masone6dcd2d2008-07-17 12:53:50 -04006507 if (zero_start != PAGE_CACHE_SIZE) {
6508 kaddr = kmap(page);
6509 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6510 flush_dcache_page(page);
6511 kunmap(page);
6512 }
Chris Mason247e7432008-07-17 12:53:51 -04006513 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006514 set_page_dirty(page);
Chris Mason50a9b212009-09-11 12:33:12 -04006515 SetPageUptodate(page);
Chris Mason5a3f23d2009-03-31 13:27:11 -04006516
Chris Mason257c62e2009-10-13 13:21:08 -04006517 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6518 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6519
Josef Bacik2ac55d42010-02-03 19:33:23 +00006520 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
Chris Mason9ebefb182007-06-15 13:50:00 -04006521
6522out_unlock:
Chris Mason50a9b212009-09-11 12:33:12 -04006523 if (!ret)
6524 return VM_FAULT_LOCKED;
Chris Mason9ebefb182007-06-15 13:50:00 -04006525 unlock_page(page);
Chris Mason1832a6d2007-12-21 16:27:21 -05006526out:
Josef Bacikec39e182012-01-12 19:10:12 -05006527 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason9998eb72012-01-25 13:47:40 -05006528out_noreserve:
Chris Mason9ebefb182007-06-15 13:50:00 -04006529 return ret;
6530}
6531
Josef Bacika41ad392011-01-31 15:30:16 -05006532static int btrfs_truncate(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04006533{
6534 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikfcb80c22011-05-03 10:40:22 -04006535 struct btrfs_block_rsv *rsv;
Chris Mason39279cc2007-06-12 06:35:45 -04006536 int ret;
Josef Bacik3893e332011-01-31 16:03:11 -05006537 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006538 struct btrfs_trans_handle *trans;
Chris Masond3c2fdc2007-09-17 10:58:06 -04006539 unsigned long nr;
Chris Masondbe674a2008-07-17 12:54:05 -04006540 u64 mask = root->sectorsize - 1;
Josef Bacik07127182011-08-19 10:29:59 -04006541 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04006542
Josef Bacik5d5e1032009-10-13 16:46:49 -04006543 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6544 if (ret)
Josef Bacika41ad392011-01-31 15:30:16 -05006545 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00006546
Chris Mason4a096752008-07-21 10:29:44 -04006547 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
Yan, Zheng80825102009-11-12 09:35:36 +00006548 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006549
Josef Bacikfcb80c22011-05-03 10:40:22 -04006550 /*
6551 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6552 * 3 things going on here
6553 *
6554 * 1) We need to reserve space for our orphan item and the space to
6555 * delete our orphan item. Lord knows we don't want to have a dangling
6556 * orphan item because we didn't reserve space to remove it.
6557 *
6558 * 2) We need to reserve space to update our inode.
6559 *
6560 * 3) We need to have something to cache all the space that is going to
6561 * be free'd up by the truncate operation, but also have some slack
6562 * space reserved in case it uses space during the truncate (thank you
6563 * very much snapshotting).
6564 *
6565 * And we need these to all be seperate. The fact is we can use alot of
6566 * space doing the truncate, and we have no earthly idea how much space
6567 * we will use, so we need the truncate reservation to be seperate so it
6568 * doesn't end up using space reserved for updating the inode or
6569 * removing the orphan item. We also need to be able to stop the
6570 * transaction and start a new one, which means we need to be able to
6571 * update the inode several times, and we have no idea of knowing how
6572 * many times that will be, so we can't just reserve 1 item for the
6573 * entirety of the opration, so that has to be done seperately as well.
6574 * Then there is the orphan item, which does indeed need to be held on
6575 * to for the whole operation, and we need nobody to touch this reserved
6576 * space except the orphan code.
6577 *
6578 * So that leaves us with
6579 *
6580 * 1) root->orphan_block_rsv - for the orphan deletion.
6581 * 2) rsv - for the truncate reservation, which we will steal from the
6582 * transaction reservation.
6583 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6584 * updating the inode.
6585 */
6586 rsv = btrfs_alloc_block_rsv(root);
6587 if (!rsv)
6588 return -ENOMEM;
Josef Bacik4a338542011-08-29 11:01:31 -04006589 rsv->size = min_size;
Josef Bacikf0cd8462011-03-04 14:37:08 -05006590
Josef Bacik907cbce2011-08-08 13:46:15 -04006591 /*
Josef Bacik07127182011-08-19 10:29:59 -04006592 * 1 for the truncate slack space
Josef Bacik907cbce2011-08-08 13:46:15 -04006593 * 1 for the orphan item we're going to add
6594 * 1 for the orphan item deletion
6595 * 1 for updating the inode.
6596 */
Josef Bacikfcb80c22011-05-03 10:40:22 -04006597 trans = btrfs_start_transaction(root, 4);
6598 if (IS_ERR(trans)) {
6599 err = PTR_ERR(trans);
6600 goto out;
6601 }
Josef Bacikf0cd8462011-03-04 14:37:08 -05006602
Josef Bacik907cbce2011-08-08 13:46:15 -04006603 /* Migrate the slack space for the truncate to our reserve */
6604 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6605 min_size);
Josef Bacikfcb80c22011-05-03 10:40:22 -04006606 BUG_ON(ret);
Josef Bacikf0cd8462011-03-04 14:37:08 -05006607
6608 ret = btrfs_orphan_add(trans, inode);
6609 if (ret) {
6610 btrfs_end_transaction(trans, root);
Josef Bacikfcb80c22011-05-03 10:40:22 -04006611 goto out;
Josef Bacikf0cd8462011-03-04 14:37:08 -05006612 }
6613
Chris Mason5a3f23d2009-03-31 13:27:11 -04006614 /*
6615 * setattr is responsible for setting the ordered_data_close flag,
6616 * but that is only tested during the last file release. That
6617 * could happen well after the next commit, leaving a great big
6618 * window where new writes may get lost if someone chooses to write
6619 * to this file after truncating to zero
6620 *
6621 * The inode doesn't have any dirty data here, and so if we commit
6622 * this is a noop. If someone immediately starts writing to the inode
6623 * it is very likely we'll catch some of their writes in this
6624 * transaction, and the commit will find this file on the ordered
6625 * data list with good things to send down.
6626 *
6627 * This is a best effort solution, there is still a window where
6628 * using truncate to replace the contents of the file will
6629 * end up with a zero length file after a crash.
6630 */
6631 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6632 btrfs_add_ordered_operation(trans, root, inode);
6633
Yan, Zheng80825102009-11-12 09:35:36 +00006634 while (1) {
Josef Bacik36ba0222011-10-18 12:15:48 -04006635 ret = btrfs_block_rsv_refill(root, rsv, min_size);
Josef Bacik907cbce2011-08-08 13:46:15 -04006636 if (ret) {
6637 /*
6638 * This can only happen with the original transaction we
6639 * started above, every other time we shouldn't have a
6640 * transaction started yet.
6641 */
6642 if (ret == -EAGAIN)
6643 goto end_trans;
6644 err = ret;
6645 break;
6646 }
6647
Yan, Zhengd68fc572010-05-16 10:49:58 -04006648 if (!trans) {
Josef Bacik907cbce2011-08-08 13:46:15 -04006649 /* Just need the 1 for updating the inode */
6650 trans = btrfs_start_transaction(root, 1);
Josef Bacikfcb80c22011-05-03 10:40:22 -04006651 if (IS_ERR(trans)) {
Josef Bacik7041ee92011-12-09 13:26:22 -05006652 ret = err = PTR_ERR(trans);
6653 trans = NULL;
6654 break;
Josef Bacikfcb80c22011-05-03 10:40:22 -04006655 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04006656 }
6657
Josef Bacik907cbce2011-08-08 13:46:15 -04006658 trans->block_rsv = rsv;
6659
Yan, Zheng80825102009-11-12 09:35:36 +00006660 ret = btrfs_truncate_inode_items(trans, root, inode,
6661 inode->i_size,
6662 BTRFS_EXTENT_DATA_KEY);
Josef Bacik3893e332011-01-31 16:03:11 -05006663 if (ret != -EAGAIN) {
6664 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00006665 break;
Josef Bacik3893e332011-01-31 16:03:11 -05006666 }
Chris Mason39279cc2007-06-12 06:35:45 -04006667
Josef Bacikfcb80c22011-05-03 10:40:22 -04006668 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00006669 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05006670 if (ret) {
6671 err = ret;
6672 break;
6673 }
Josef Bacik907cbce2011-08-08 13:46:15 -04006674end_trans:
Yan, Zheng80825102009-11-12 09:35:36 +00006675 nr = trans->blocks_used;
6676 btrfs_end_transaction(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04006677 trans = NULL;
Yan, Zheng80825102009-11-12 09:35:36 +00006678 btrfs_btree_balance_dirty(root, nr);
Yan, Zheng80825102009-11-12 09:35:36 +00006679 }
6680
6681 if (ret == 0 && inode->i_nlink > 0) {
Josef Bacikfcb80c22011-05-03 10:40:22 -04006682 trans->block_rsv = root->orphan_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00006683 ret = btrfs_orphan_del(trans, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05006684 if (ret)
6685 err = ret;
Josef Bacikded5db92011-03-04 14:09:46 -05006686 } else if (ret && inode->i_nlink > 0) {
6687 /*
6688 * Failed to do the truncate, remove us from the in memory
6689 * orphan list.
6690 */
6691 ret = btrfs_orphan_del(NULL, inode);
Yan, Zheng80825102009-11-12 09:35:36 +00006692 }
6693
Chris Mason917c16b2011-11-08 14:49:59 -05006694 if (trans) {
6695 trans->block_rsv = &root->fs_info->trans_block_rsv;
6696 ret = btrfs_update_inode(trans, root, inode);
6697 if (ret && !err)
6698 err = ret;
Josef Bacik7b128762008-07-24 12:17:14 -04006699
Chris Mason917c16b2011-11-08 14:49:59 -05006700 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05006701 ret = btrfs_end_transaction(trans, root);
Chris Mason917c16b2011-11-08 14:49:59 -05006702 btrfs_btree_balance_dirty(root, nr);
6703 }
Josef Bacikfcb80c22011-05-03 10:40:22 -04006704
6705out:
6706 btrfs_free_block_rsv(root, rsv);
6707
Josef Bacik3893e332011-01-31 16:03:11 -05006708 if (ret && !err)
6709 err = ret;
Josef Bacika41ad392011-01-31 15:30:16 -05006710
Josef Bacik3893e332011-01-31 16:03:11 -05006711 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04006712}
6713
Sven Wegener3b963622008-06-09 21:57:42 -04006714/*
Chris Masond352ac62008-09-29 15:18:18 -04006715 * create a new subvolume directory/inode (helper for the ioctl).
6716 */
Yan Zhengd2fb3432008-12-11 16:30:39 -05006717int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
Josef Bacikd82a6f12011-05-11 15:26:06 -04006718 struct btrfs_root *new_root, u64 new_dirid)
Chris Mason39279cc2007-06-12 06:35:45 -04006719{
Chris Mason39279cc2007-06-12 06:35:45 -04006720 struct inode *inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04006721 int err;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006722 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006723
Florian Albrechtskirchinger12fc9d02012-02-10 22:15:54 +01006724 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6725 new_dirid, new_dirid,
6726 S_IFDIR | (~current_umask() & S_IRWXUGO),
6727 &index);
Chris Mason54aa1f42007-06-22 14:16:25 -04006728 if (IS_ERR(inode))
Christoph Hellwigf46b5a62008-06-11 21:53:53 -04006729 return PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006730 inode->i_op = &btrfs_dir_inode_operations;
6731 inode->i_fop = &btrfs_dir_file_operations;
6732
Miklos Szeredibfe86842011-10-28 14:13:29 +02006733 set_nlink(inode, 1);
Chris Masondbe674a2008-07-17 12:54:05 -04006734 btrfs_i_size_write(inode, 0);
Sven Wegener3b963622008-06-09 21:57:42 -04006735
Yan, Zheng76dda932009-09-21 16:00:26 -04006736 err = btrfs_update_inode(trans, new_root, inode);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04006737
Yan, Zheng76dda932009-09-21 16:00:26 -04006738 iput(inode);
Mark Fashehce598972011-07-26 11:32:23 -07006739 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04006740}
6741
Chris Mason39279cc2007-06-12 06:35:45 -04006742struct inode *btrfs_alloc_inode(struct super_block *sb)
6743{
6744 struct btrfs_inode *ei;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006745 struct inode *inode;
Chris Mason39279cc2007-06-12 06:35:45 -04006746
6747 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6748 if (!ei)
6749 return NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006750
6751 ei->root = NULL;
6752 ei->space_info = NULL;
6753 ei->generation = 0;
6754 ei->sequence = 0;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04006755 ei->last_trans = 0;
Chris Mason257c62e2009-10-13 13:21:08 -04006756 ei->last_sub_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04006757 ei->logged_trans = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006758 ei->delalloc_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006759 ei->disk_i_size = 0;
6760 ei->flags = 0;
Josef Bacik7709cde2011-08-04 10:25:02 -04006761 ei->csum_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006762 ei->index_cnt = (u64)-1;
6763 ei->last_unlink_trans = 0;
6764
Josef Bacik9e0baf62011-07-15 15:16:44 +00006765 spin_lock_init(&ei->lock);
6766 ei->outstanding_extents = 0;
6767 ei->reserved_extents = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006768
6769 ei->ordered_data_close = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04006770 ei->orphan_meta_reserved = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006771 ei->dummy_inode = 0;
Chris Mason4cb53002011-05-24 15:35:30 -04006772 ei->in_defrag = 0;
Josef Bacik7fd2ae22011-11-08 15:47:34 -05006773 ei->delalloc_meta_reserved = 0;
Li Zefan261507a02010-12-17 14:21:50 +08006774 ei->force_compress = BTRFS_COMPRESS_NONE;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006775
Miao Xie16cdcec2011-04-22 18:12:22 +08006776 ei->delayed_node = NULL;
6777
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006778 inode = &ei->vfs_inode;
David Sterbaa8067e02011-04-21 00:34:43 +02006779 extent_map_tree_init(&ei->extent_tree);
David Sterbaf993c882011-04-20 23:35:57 +02006780 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6781 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006782 mutex_init(&ei->log_mutex);
Josef Bacikf2486792012-01-13 12:09:22 -05006783 mutex_init(&ei->delalloc_mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006784 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Josef Bacik7b128762008-07-24 12:17:14 -04006785 INIT_LIST_HEAD(&ei->i_orphan);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006786 INIT_LIST_HEAD(&ei->delalloc_inodes);
Chris Mason5a3f23d2009-03-31 13:27:11 -04006787 INIT_LIST_HEAD(&ei->ordered_operations);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006788 RB_CLEAR_NODE(&ei->rb_node);
6789
6790 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04006791}
6792
Nick Pigginfa0d7e32011-01-07 17:49:49 +11006793static void btrfs_i_callback(struct rcu_head *head)
6794{
6795 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e32011-01-07 17:49:49 +11006796 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6797}
6798
Chris Mason39279cc2007-06-12 06:35:45 -04006799void btrfs_destroy_inode(struct inode *inode)
6800{
Chris Masone6dcd2d2008-07-17 12:53:50 -04006801 struct btrfs_ordered_extent *ordered;
Chris Mason5a3f23d2009-03-31 13:27:11 -04006802 struct btrfs_root *root = BTRFS_I(inode)->root;
6803
Chris Mason39279cc2007-06-12 06:35:45 -04006804 WARN_ON(!list_empty(&inode->i_dentry));
6805 WARN_ON(inode->i_data.nrpages);
Josef Bacik9e0baf62011-07-15 15:16:44 +00006806 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6807 WARN_ON(BTRFS_I(inode)->reserved_extents);
Josef Bacik7709cde2011-08-04 10:25:02 -04006808 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6809 WARN_ON(BTRFS_I(inode)->csum_bytes);
Chris Mason39279cc2007-06-12 06:35:45 -04006810
Chris Mason5a3f23d2009-03-31 13:27:11 -04006811 /*
Josef Bacika6dbd422009-11-11 15:53:34 -05006812 * This can happen where we create an inode, but somebody else also
6813 * created the same inode and we need to destroy the one we already
6814 * created.
6815 */
6816 if (!root)
6817 goto free;
6818
6819 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04006820 * Make sure we're properly removed from the ordered operation
6821 * lists.
6822 */
6823 smp_mb();
6824 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6825 spin_lock(&root->fs_info->ordered_extent_lock);
6826 list_del_init(&BTRFS_I(inode)->ordered_operations);
6827 spin_unlock(&root->fs_info->ordered_extent_lock);
6828 }
6829
Yan, Zhengd68fc572010-05-16 10:49:58 -04006830 spin_lock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04006831 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
Li Zefan33345d012011-04-20 10:31:50 +08006832 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6833 (unsigned long long)btrfs_ino(inode));
Yan, Zheng80825102009-11-12 09:35:36 +00006834 list_del_init(&BTRFS_I(inode)->i_orphan);
Josef Bacik7b128762008-07-24 12:17:14 -04006835 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04006836 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04006837
Chris Masond3977122009-01-05 21:25:51 -05006838 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04006839 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6840 if (!ordered)
6841 break;
6842 else {
Chris Masond3977122009-01-05 21:25:51 -05006843 printk(KERN_ERR "btrfs found ordered "
6844 "extent %llu %llu on inode cleanup\n",
6845 (unsigned long long)ordered->file_offset,
6846 (unsigned long long)ordered->len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006847 btrfs_remove_ordered_extent(inode, ordered);
6848 btrfs_put_ordered_extent(ordered);
6849 btrfs_put_ordered_extent(ordered);
6850 }
6851 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006852 inode_tree_del(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04006853 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
Josef Bacika6dbd422009-11-11 15:53:34 -05006854free:
Miao Xie16cdcec2011-04-22 18:12:22 +08006855 btrfs_remove_delayed_node(inode);
Nick Pigginfa0d7e32011-01-07 17:49:49 +11006856 call_rcu(&inode->i_rcu, btrfs_i_callback);
Chris Mason39279cc2007-06-12 06:35:45 -04006857}
6858
Al Viro45321ac2010-06-07 13:43:19 -04006859int btrfs_drop_inode(struct inode *inode)
Yan, Zheng76dda932009-09-21 16:00:26 -04006860{
6861 struct btrfs_root *root = BTRFS_I(inode)->root;
Al Viro45321ac2010-06-07 13:43:19 -04006862
Josef Bacik0af3d002010-06-21 14:48:16 -04006863 if (btrfs_root_refs(&root->root_item) == 0 &&
Chris Mason2cf85722011-07-26 15:35:09 -04006864 !btrfs_is_free_space_inode(root, inode))
Al Viro45321ac2010-06-07 13:43:19 -04006865 return 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04006866 else
Al Viro45321ac2010-06-07 13:43:19 -04006867 return generic_drop_inode(inode);
Yan, Zheng76dda932009-09-21 16:00:26 -04006868}
6869
Sven Wegener0ee0fda2008-07-30 16:54:26 -04006870static void init_once(void *foo)
Chris Mason39279cc2007-06-12 06:35:45 -04006871{
6872 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6873
6874 inode_init_once(&ei->vfs_inode);
6875}
6876
6877void btrfs_destroy_cachep(void)
6878{
6879 if (btrfs_inode_cachep)
6880 kmem_cache_destroy(btrfs_inode_cachep);
6881 if (btrfs_trans_handle_cachep)
6882 kmem_cache_destroy(btrfs_trans_handle_cachep);
6883 if (btrfs_transaction_cachep)
6884 kmem_cache_destroy(btrfs_transaction_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04006885 if (btrfs_path_cachep)
6886 kmem_cache_destroy(btrfs_path_cachep);
Josef Bacikdc89e982011-01-28 17:05:48 -05006887 if (btrfs_free_space_cachep)
6888 kmem_cache_destroy(btrfs_free_space_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04006889}
6890
6891int btrfs_init_cachep(void)
6892{
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006893 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6894 sizeof(struct btrfs_inode), 0,
6895 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
Chris Mason39279cc2007-06-12 06:35:45 -04006896 if (!btrfs_inode_cachep)
6897 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006898
6899 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6900 sizeof(struct btrfs_trans_handle), 0,
6901 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006902 if (!btrfs_trans_handle_cachep)
6903 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006904
6905 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6906 sizeof(struct btrfs_transaction), 0,
6907 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006908 if (!btrfs_transaction_cachep)
6909 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006910
6911 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6912 sizeof(struct btrfs_path), 0,
6913 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006914 if (!btrfs_path_cachep)
6915 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006916
Josef Bacikdc89e982011-01-28 17:05:48 -05006917 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6918 sizeof(struct btrfs_free_space), 0,
6919 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6920 if (!btrfs_free_space_cachep)
6921 goto fail;
6922
Chris Mason39279cc2007-06-12 06:35:45 -04006923 return 0;
6924fail:
6925 btrfs_destroy_cachep();
6926 return -ENOMEM;
6927}
6928
6929static int btrfs_getattr(struct vfsmount *mnt,
6930 struct dentry *dentry, struct kstat *stat)
6931{
6932 struct inode *inode = dentry->d_inode;
David Sterbafadc0d82011-11-20 07:33:38 -05006933 u32 blocksize = inode->i_sb->s_blocksize;
6934
Chris Mason39279cc2007-06-12 06:35:45 -04006935 generic_fillattr(inode, stat);
Al Viro0ee5dc62011-07-07 15:44:25 -04006936 stat->dev = BTRFS_I(inode)->root->anon_dev;
Chris Masond6667462008-01-03 14:51:00 -05006937 stat->blksize = PAGE_CACHE_SIZE;
David Sterbafadc0d82011-11-20 07:33:38 -05006938 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6939 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
Chris Mason39279cc2007-06-12 06:35:45 -04006940 return 0;
6941}
6942
Liu Bo75e7cb72011-03-22 10:12:20 +00006943/*
6944 * If a file is moved, it will inherit the cow and compression flags of the new
6945 * directory.
6946 */
6947static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6948{
6949 struct btrfs_inode *b_dir = BTRFS_I(dir);
6950 struct btrfs_inode *b_inode = BTRFS_I(inode);
6951
6952 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6953 b_inode->flags |= BTRFS_INODE_NODATACOW;
6954 else
6955 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6956
6957 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6958 b_inode->flags |= BTRFS_INODE_COMPRESS;
6959 else
6960 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6961}
6962
Chris Masond3977122009-01-05 21:25:51 -05006963static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6964 struct inode *new_dir, struct dentry *new_dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04006965{
6966 struct btrfs_trans_handle *trans;
6967 struct btrfs_root *root = BTRFS_I(old_dir)->root;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006968 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04006969 struct inode *new_inode = new_dentry->d_inode;
6970 struct inode *old_inode = old_dentry->d_inode;
6971 struct timespec ctime = CURRENT_TIME;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006972 u64 index = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006973 u64 root_objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04006974 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08006975 u64 old_ino = btrfs_ino(old_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006976
Li Zefan33345d012011-04-20 10:31:50 +08006977 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Yan, Zhengf679a842009-09-24 09:17:31 -04006978 return -EPERM;
6979
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006980 /* we only allow rename subvolume link between subvolumes */
Li Zefan33345d012011-04-20 10:31:50 +08006981 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
Chris Mason3394e162008-11-17 20:42:26 -05006982 return -EXDEV;
6983
Li Zefan33345d012011-04-20 10:31:50 +08006984 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6985 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006986 return -ENOTEMPTY;
6987
Chris Mason39279cc2007-06-12 06:35:45 -04006988 if (S_ISDIR(old_inode->i_mode) && new_inode &&
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006989 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Chris Mason39279cc2007-06-12 06:35:45 -04006990 return -ENOTEMPTY;
Chris Mason5a3f23d2009-03-31 13:27:11 -04006991 /*
6992 * we're using rename to replace one file with another.
6993 * and the replacement file is large. Start IO on it now so
6994 * we don't add too much work to the end of the transaction
6995 */
Bartlomiej Zolnierkiewicz4baf8c92009-08-07 13:47:08 -04006996 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
Chris Mason5a3f23d2009-03-31 13:27:11 -04006997 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6998 filemap_flush(old_inode->i_mapping);
6999
Yan, Zheng76dda932009-09-21 16:00:26 -04007000 /* close the racy window with snapshot create/destroy ioctl */
Li Zefan33345d012011-04-20 10:31:50 +08007001 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04007002 down_read(&root->fs_info->subvol_sem);
Yan, Zhenga22285a2010-05-16 10:48:46 -04007003 /*
7004 * We want to reserve the absolute worst case amount of items. So if
7005 * both inodes are subvols and we need to unlink them then that would
7006 * require 4 item modifications, but if they are both normal inodes it
7007 * would require 5 item modifications, so we'll assume their normal
7008 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7009 * should cover the worst case number of items we'll modify.
7010 */
7011 trans = btrfs_start_transaction(root, 20);
Johann Lombardib44c59a2011-03-31 13:23:47 +00007012 if (IS_ERR(trans)) {
7013 ret = PTR_ERR(trans);
7014 goto out_notrans;
7015 }
Chris Mason5f39d392007-10-15 16:14:19 -04007016
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007017 if (dest != root)
7018 btrfs_record_root_in_trans(trans, dest);
7019
Yan, Zhenga5719522009-09-24 09:17:31 -04007020 ret = btrfs_set_inode_index(new_dir, &index);
7021 if (ret)
7022 goto out_fail;
Chris Mason5a3f23d2009-03-31 13:27:11 -04007023
Li Zefan33345d012011-04-20 10:31:50 +08007024 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007025 /* force full log commit if subvolume involved. */
7026 root->fs_info->last_trans_log_full_commit = trans->transid;
7027 } else {
Yan, Zhenga5719522009-09-24 09:17:31 -04007028 ret = btrfs_insert_inode_ref(trans, dest,
7029 new_dentry->d_name.name,
7030 new_dentry->d_name.len,
Li Zefan33345d012011-04-20 10:31:50 +08007031 old_ino,
7032 btrfs_ino(new_dir), index);
Yan, Zhenga5719522009-09-24 09:17:31 -04007033 if (ret)
7034 goto out_fail;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007035 /*
7036 * this is an ugly little race, but the rename is required
7037 * to make sure that if we crash, the inode is either at the
7038 * old name or the new one. pinning the log transaction lets
7039 * us make sure we don't allow a log commit to come in after
7040 * we unlink the name but before we add the new name back in.
7041 */
7042 btrfs_pin_log_trans(root);
7043 }
Chris Mason12fcfd22009-03-24 10:24:20 -04007044 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04007045 * make sure the inode gets flushed if it is replacing
7046 * something.
7047 */
Li Zefan33345d012011-04-20 10:31:50 +08007048 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
Chris Mason5a3f23d2009-03-31 13:27:11 -04007049 btrfs_add_ordered_operation(trans, root, old_inode);
Chris Mason5a3f23d2009-03-31 13:27:11 -04007050
Chris Mason39279cc2007-06-12 06:35:45 -04007051 old_dir->i_ctime = old_dir->i_mtime = ctime;
7052 new_dir->i_ctime = new_dir->i_mtime = ctime;
7053 old_inode->i_ctime = ctime;
Chris Mason5f39d392007-10-15 16:14:19 -04007054
Chris Mason12fcfd22009-03-24 10:24:20 -04007055 if (old_dentry->d_parent != new_dentry->d_parent)
7056 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7057
Li Zefan33345d012011-04-20 10:31:50 +08007058 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007059 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7060 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7061 old_dentry->d_name.name,
7062 old_dentry->d_name.len);
7063 } else {
Al Viro92986792011-03-04 17:14:37 +00007064 ret = __btrfs_unlink_inode(trans, root, old_dir,
7065 old_dentry->d_inode,
7066 old_dentry->d_name.name,
7067 old_dentry->d_name.len);
7068 if (!ret)
7069 ret = btrfs_update_inode(trans, root, old_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007070 }
7071 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04007072
7073 if (new_inode) {
7074 new_inode->i_ctime = CURRENT_TIME;
Li Zefan33345d012011-04-20 10:31:50 +08007075 if (unlikely(btrfs_ino(new_inode) ==
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007076 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7077 root_objectid = BTRFS_I(new_inode)->location.objectid;
7078 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7079 root_objectid,
7080 new_dentry->d_name.name,
7081 new_dentry->d_name.len);
7082 BUG_ON(new_inode->i_nlink == 0);
7083 } else {
7084 ret = btrfs_unlink_inode(trans, dest, new_dir,
7085 new_dentry->d_inode,
7086 new_dentry->d_name.name,
7087 new_dentry->d_name.len);
7088 }
7089 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04007090 if (new_inode->i_nlink == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04007091 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007092 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04007093 }
Chris Mason39279cc2007-06-12 06:35:45 -04007094 }
Josef Bacikaec74772008-07-24 12:12:38 -04007095
Liu Bo75e7cb72011-03-22 10:12:20 +00007096 fixup_inode_flags(new_dir, old_inode);
7097
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007098 ret = btrfs_add_link(trans, new_dir, old_inode,
7099 new_dentry->d_name.name,
Yan, Zhenga5719522009-09-24 09:17:31 -04007100 new_dentry->d_name.len, 0, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007101 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04007102
Li Zefan33345d012011-04-20 10:31:50 +08007103 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
Al Viro10d9f302011-07-16 23:09:10 -04007104 struct dentry *parent = new_dentry->d_parent;
Josef Bacik6a912212010-11-20 09:48:00 +00007105 btrfs_log_new_name(trans, old_inode, old_dir, parent);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007106 btrfs_end_log_trans(root);
7107 }
Chris Mason39279cc2007-06-12 06:35:45 -04007108out_fail:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05007109 btrfs_end_transaction(trans, root);
Johann Lombardib44c59a2011-03-31 13:23:47 +00007110out_notrans:
Li Zefan33345d012011-04-20 10:31:50 +08007111 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04007112 up_read(&root->fs_info->subvol_sem);
Josef Bacik9ed74f22009-09-11 16:12:44 -04007113
Chris Mason39279cc2007-06-12 06:35:45 -04007114 return ret;
7115}
7116
Chris Masond352ac62008-09-29 15:18:18 -04007117/*
7118 * some fairly slow code that needs optimization. This walks the list
7119 * of all the inodes with pending delalloc and forces them to disk.
7120 */
Yan, Zheng24bbcf02009-11-12 09:36:34 +00007121int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
Chris Masonea8c2812008-08-04 23:17:27 -04007122{
7123 struct list_head *head = &root->fs_info->delalloc_inodes;
7124 struct btrfs_inode *binode;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007125 struct inode *inode;
Chris Masonea8c2812008-08-04 23:17:27 -04007126
Yan Zhengc146afa2008-11-12 14:34:12 -05007127 if (root->fs_info->sb->s_flags & MS_RDONLY)
7128 return -EROFS;
7129
Chris Mason75eff682008-12-15 15:54:40 -05007130 spin_lock(&root->fs_info->delalloc_lock);
Chris Masond3977122009-01-05 21:25:51 -05007131 while (!list_empty(head)) {
Chris Masonea8c2812008-08-04 23:17:27 -04007132 binode = list_entry(head->next, struct btrfs_inode,
7133 delalloc_inodes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007134 inode = igrab(&binode->vfs_inode);
7135 if (!inode)
7136 list_del_init(&binode->delalloc_inodes);
Chris Mason75eff682008-12-15 15:54:40 -05007137 spin_unlock(&root->fs_info->delalloc_lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007138 if (inode) {
Chris Mason8c8bee12008-09-29 11:19:10 -04007139 filemap_flush(inode->i_mapping);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00007140 if (delay_iput)
7141 btrfs_add_delayed_iput(inode);
7142 else
7143 iput(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007144 }
7145 cond_resched();
Chris Mason75eff682008-12-15 15:54:40 -05007146 spin_lock(&root->fs_info->delalloc_lock);
Chris Masonea8c2812008-08-04 23:17:27 -04007147 }
Chris Mason75eff682008-12-15 15:54:40 -05007148 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason8c8bee12008-09-29 11:19:10 -04007149
7150 /* the filemap_flush will queue IO into the worker threads, but
7151 * we have to make sure the IO is actually started and that
7152 * ordered extents get created before we return
7153 */
7154 atomic_inc(&root->fs_info->async_submit_draining);
Chris Masond3977122009-01-05 21:25:51 -05007155 while (atomic_read(&root->fs_info->nr_async_submits) ||
Chris Mason771ed682008-11-06 22:02:51 -05007156 atomic_read(&root->fs_info->async_delalloc_pages)) {
Chris Mason8c8bee12008-09-29 11:19:10 -04007157 wait_event(root->fs_info->async_submit_wait,
Chris Mason771ed682008-11-06 22:02:51 -05007158 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7159 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
Chris Mason8c8bee12008-09-29 11:19:10 -04007160 }
7161 atomic_dec(&root->fs_info->async_submit_draining);
Chris Masonea8c2812008-08-04 23:17:27 -04007162 return 0;
7163}
7164
Chris Mason39279cc2007-06-12 06:35:45 -04007165static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7166 const char *symname)
7167{
7168 struct btrfs_trans_handle *trans;
7169 struct btrfs_root *root = BTRFS_I(dir)->root;
7170 struct btrfs_path *path;
7171 struct btrfs_key key;
Chris Mason1832a6d2007-12-21 16:27:21 -05007172 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04007173 int err;
7174 int drop_inode = 0;
7175 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04007176 u64 index = 0 ;
Chris Mason39279cc2007-06-12 06:35:45 -04007177 int name_len;
7178 int datasize;
Chris Mason5f39d392007-10-15 16:14:19 -04007179 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04007180 struct btrfs_file_extent_item *ei;
Chris Mason5f39d392007-10-15 16:14:19 -04007181 struct extent_buffer *leaf;
Chris Mason1832a6d2007-12-21 16:27:21 -05007182 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04007183
7184 name_len = strlen(symname) + 1;
7185 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7186 return -ENAMETOOLONG;
Chris Mason1832a6d2007-12-21 16:27:21 -05007187
Josef Bacik9ed74f22009-09-11 16:12:44 -04007188 /*
7189 * 2 items for inode item and ref
7190 * 2 items for dir items
7191 * 1 item for xattr if selinux is on
7192 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04007193 trans = btrfs_start_transaction(root, 5);
7194 if (IS_ERR(trans))
7195 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05007196
Li Zefan581bb052011-04-20 10:06:11 +08007197 err = btrfs_find_free_ino(root, &objectid);
7198 if (err)
7199 goto out_unlock;
7200
Josef Bacikaec74772008-07-24 12:12:38 -04007201 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08007202 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04007203 S_IFLNK|S_IRWXUGO, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04007204 if (IS_ERR(inode)) {
7205 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007206 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04007207 }
Chris Mason39279cc2007-06-12 06:35:45 -04007208
Eric Paris2a7dba32011-02-01 11:05:39 -05007209 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04007210 if (err) {
7211 drop_inode = 1;
7212 goto out_unlock;
7213 }
7214
Casey Schauflerad19db72011-12-15 10:09:07 -05007215 /*
7216 * If the active LSM wants to access the inode during
7217 * d_instantiate it needs these. Smack checks to see
7218 * if the filesystem supports xattrs by looking at the
7219 * ops vector.
7220 */
7221 inode->i_fop = &btrfs_file_operations;
7222 inode->i_op = &btrfs_file_inode_operations;
7223
Josef Bacika1b075d2010-11-19 20:36:11 +00007224 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04007225 if (err)
7226 drop_inode = 1;
7227 else {
7228 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04007229 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05007230 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04007231 }
Chris Mason39279cc2007-06-12 06:35:45 -04007232 if (drop_inode)
7233 goto out_unlock;
7234
7235 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07007236 if (!path) {
7237 err = -ENOMEM;
7238 drop_inode = 1;
7239 goto out_unlock;
7240 }
Li Zefan33345d012011-04-20 10:31:50 +08007241 key.objectid = btrfs_ino(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007242 key.offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04007243 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7244 datasize = btrfs_file_extent_calc_inline_size(name_len);
7245 err = btrfs_insert_empty_item(trans, root, path, &key,
7246 datasize);
Chris Mason54aa1f42007-06-22 14:16:25 -04007247 if (err) {
7248 drop_inode = 1;
Julia Lawallb0839162011-05-14 07:10:51 +00007249 btrfs_free_path(path);
Chris Mason54aa1f42007-06-22 14:16:25 -04007250 goto out_unlock;
7251 }
Chris Mason5f39d392007-10-15 16:14:19 -04007252 leaf = path->nodes[0];
7253 ei = btrfs_item_ptr(leaf, path->slots[0],
7254 struct btrfs_file_extent_item);
7255 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7256 btrfs_set_file_extent_type(leaf, ei,
Chris Mason39279cc2007-06-12 06:35:45 -04007257 BTRFS_FILE_EXTENT_INLINE);
Chris Masonc8b97812008-10-29 14:49:59 -04007258 btrfs_set_file_extent_encryption(leaf, ei, 0);
7259 btrfs_set_file_extent_compression(leaf, ei, 0);
7260 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7261 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7262
Chris Mason39279cc2007-06-12 06:35:45 -04007263 ptr = btrfs_file_extent_inline_start(ei);
Chris Mason5f39d392007-10-15 16:14:19 -04007264 write_extent_buffer(leaf, symname, ptr, name_len);
7265 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04007266 btrfs_free_path(path);
Chris Mason5f39d392007-10-15 16:14:19 -04007267
Chris Mason39279cc2007-06-12 06:35:45 -04007268 inode->i_op = &btrfs_symlink_inode_operations;
7269 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04007270 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Yan Zhengd899e052008-10-30 14:25:28 -04007271 inode_set_bytes(inode, name_len);
Chris Masondbe674a2008-07-17 12:54:05 -04007272 btrfs_i_size_write(inode, name_len - 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04007273 err = btrfs_update_inode(trans, root, inode);
7274 if (err)
7275 drop_inode = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04007276
7277out_unlock:
Al Viro08c422c2011-12-23 07:58:13 -05007278 if (!err)
7279 d_instantiate(dentry, inode);
Chris Masond3c2fdc2007-09-17 10:58:06 -04007280 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05007281 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04007282 if (drop_inode) {
7283 inode_dec_link_count(inode);
7284 iput(inode);
7285 }
Chris Masond3c2fdc2007-09-17 10:58:06 -04007286 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04007287 return err;
7288}
Chris Mason16432982008-04-10 10:23:21 -04007289
Josef Bacik0af3d002010-06-21 14:48:16 -04007290static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7291 u64 start, u64 num_bytes, u64 min_size,
7292 loff_t actual_len, u64 *alloc_hint,
7293 struct btrfs_trans_handle *trans)
Yan Zhengd899e052008-10-30 14:25:28 -04007294{
Yan Zhengd899e052008-10-30 14:25:28 -04007295 struct btrfs_root *root = BTRFS_I(inode)->root;
7296 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -04007297 u64 cur_offset = start;
Josef Bacik55a61d12010-11-22 18:50:32 +00007298 u64 i_size;
Yan Zhengd899e052008-10-30 14:25:28 -04007299 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04007300 bool own_trans = true;
Yan Zhengd899e052008-10-30 14:25:28 -04007301
Josef Bacik0af3d002010-06-21 14:48:16 -04007302 if (trans)
7303 own_trans = false;
Yan Zhengd899e052008-10-30 14:25:28 -04007304 while (num_bytes > 0) {
Josef Bacik0af3d002010-06-21 14:48:16 -04007305 if (own_trans) {
7306 trans = btrfs_start_transaction(root, 3);
7307 if (IS_ERR(trans)) {
7308 ret = PTR_ERR(trans);
7309 break;
7310 }
Yan Zhengd899e052008-10-30 14:25:28 -04007311 }
Yan, Zheng5a303d52009-11-12 09:34:52 +00007312
Yan, Zhengefa56462010-05-16 10:49:59 -04007313 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7314 0, *alloc_hint, (u64)-1, &ins, 1);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007315 if (ret) {
Josef Bacik0af3d002010-06-21 14:48:16 -04007316 if (own_trans)
7317 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04007318 break;
Yan, Zheng5a303d52009-11-12 09:34:52 +00007319 }
7320
Yan Zhengd899e052008-10-30 14:25:28 -04007321 ret = insert_reserved_file_extent(trans, inode,
7322 cur_offset, ins.objectid,
7323 ins.offset, ins.offset,
Yan, Zheng920bbbf2009-11-12 09:34:08 +00007324 ins.offset, 0, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04007325 BTRFS_FILE_EXTENT_PREALLOC);
7326 BUG_ON(ret);
Chris Masona1ed8352009-09-11 12:27:37 -04007327 btrfs_drop_extent_cache(inode, cur_offset,
7328 cur_offset + ins.offset -1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007329
Yan Zhengd899e052008-10-30 14:25:28 -04007330 num_bytes -= ins.offset;
7331 cur_offset += ins.offset;
Yan, Zhengefa56462010-05-16 10:49:59 -04007332 *alloc_hint = ins.objectid + ins.offset;
Yan, Zheng5a303d52009-11-12 09:34:52 +00007333
Yan Zhengd899e052008-10-30 14:25:28 -04007334 inode->i_ctime = CURRENT_TIME;
Christoph Hellwig6cbff002009-04-17 10:37:41 +02007335 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Yan Zhengd899e052008-10-30 14:25:28 -04007336 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
Yan, Zhengefa56462010-05-16 10:49:59 -04007337 (actual_len > inode->i_size) &&
7338 (cur_offset > inode->i_size)) {
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00007339 if (cur_offset > actual_len)
Josef Bacik55a61d12010-11-22 18:50:32 +00007340 i_size = actual_len;
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00007341 else
Josef Bacik55a61d12010-11-22 18:50:32 +00007342 i_size = cur_offset;
7343 i_size_write(inode, i_size);
7344 btrfs_ordered_update_i_size(inode, i_size, NULL);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007345 }
7346
Yan Zhengd899e052008-10-30 14:25:28 -04007347 ret = btrfs_update_inode(trans, root, inode);
7348 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -04007349
Josef Bacik0af3d002010-06-21 14:48:16 -04007350 if (own_trans)
7351 btrfs_end_transaction(trans, root);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007352 }
Yan Zhengd899e052008-10-30 14:25:28 -04007353 return ret;
7354}
7355
Josef Bacik0af3d002010-06-21 14:48:16 -04007356int btrfs_prealloc_file_range(struct inode *inode, int mode,
7357 u64 start, u64 num_bytes, u64 min_size,
7358 loff_t actual_len, u64 *alloc_hint)
7359{
7360 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7361 min_size, actual_len, alloc_hint,
7362 NULL);
7363}
7364
7365int btrfs_prealloc_file_range_trans(struct inode *inode,
7366 struct btrfs_trans_handle *trans, int mode,
7367 u64 start, u64 num_bytes, u64 min_size,
7368 loff_t actual_len, u64 *alloc_hint)
7369{
7370 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7371 min_size, actual_len, alloc_hint, trans);
7372}
7373
Chris Masone6dcd2d2008-07-17 12:53:50 -04007374static int btrfs_set_page_dirty(struct page *page)
7375{
Chris Masone6dcd2d2008-07-17 12:53:50 -04007376 return __set_page_dirty_nobuffers(page);
7377}
7378
Al Viro10556cb2011-06-20 19:28:19 -04007379static int btrfs_permission(struct inode *inode, int mask)
Yanfdebe2b2008-01-14 13:26:08 -05007380{
Li Zefanb83cc962010-12-20 16:04:08 +08007381 struct btrfs_root *root = BTRFS_I(inode)->root;
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +00007382 umode_t mode = inode->i_mode;
Li Zefanb83cc962010-12-20 16:04:08 +08007383
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +00007384 if (mask & MAY_WRITE &&
7385 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7386 if (btrfs_root_readonly(root))
7387 return -EROFS;
7388 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7389 return -EACCES;
7390 }
Al Viro2830ba72011-06-20 19:16:29 -04007391 return generic_permission(inode, mask);
Yanfdebe2b2008-01-14 13:26:08 -05007392}
Chris Mason39279cc2007-06-12 06:35:45 -04007393
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007394static const struct inode_operations btrfs_dir_inode_operations = {
Chris Mason3394e162008-11-17 20:42:26 -05007395 .getattr = btrfs_getattr,
Chris Mason39279cc2007-06-12 06:35:45 -04007396 .lookup = btrfs_lookup,
7397 .create = btrfs_create,
7398 .unlink = btrfs_unlink,
7399 .link = btrfs_link,
7400 .mkdir = btrfs_mkdir,
7401 .rmdir = btrfs_rmdir,
7402 .rename = btrfs_rename,
7403 .symlink = btrfs_symlink,
7404 .setattr = btrfs_setattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04007405 .mknod = btrfs_mknod,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007406 .setxattr = btrfs_setxattr,
7407 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05007408 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007409 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05007410 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007411 .get_acl = btrfs_get_acl,
Chris Mason39279cc2007-06-12 06:35:45 -04007412};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007413static const struct inode_operations btrfs_dir_ro_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007414 .lookup = btrfs_lookup,
Yanfdebe2b2008-01-14 13:26:08 -05007415 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007416 .get_acl = btrfs_get_acl,
Chris Mason39279cc2007-06-12 06:35:45 -04007417};
Yan, Zheng76dda932009-09-21 16:00:26 -04007418
Alexey Dobriyan828c0952009-10-01 15:43:56 -07007419static const struct file_operations btrfs_dir_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007420 .llseek = generic_file_llseek,
7421 .read = generic_read_dir,
David Woodhousecbdf5a22008-08-06 19:42:33 +01007422 .readdir = btrfs_real_readdir,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04007423 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04007424#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04007425 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04007426#endif
Sage Weil6bf13c02008-06-10 10:07:39 -04007427 .release = btrfs_release_file,
Chris Masone02119d2008-09-05 16:13:11 -04007428 .fsync = btrfs_sync_file,
Chris Mason39279cc2007-06-12 06:35:45 -04007429};
7430
Chris Masond1310b22008-01-24 16:13:08 -05007431static struct extent_io_ops btrfs_extent_io_ops = {
Chris Mason07157aa2007-08-30 08:50:51 -04007432 .fill_delalloc = run_delalloc_range,
Chris Mason065631f2008-02-20 12:07:25 -05007433 .submit_bio_hook = btrfs_submit_bio_hook,
Chris Mason239b14b2008-03-24 15:02:07 -04007434 .merge_bio_hook = btrfs_merge_bio_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04007435 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
Chris Masone6dcd2d2008-07-17 12:53:50 -04007436 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
Chris Mason247e7432008-07-17 12:53:51 -04007437 .writepage_start_hook = btrfs_writepage_start_hook,
Chris Masonb0c68f82008-01-31 11:05:37 -05007438 .set_bit_hook = btrfs_set_bit_hook,
7439 .clear_bit_hook = btrfs_clear_bit_hook,
Josef Bacik9ed74f22009-09-11 16:12:44 -04007440 .merge_extent_hook = btrfs_merge_extent_hook,
7441 .split_extent_hook = btrfs_split_extent_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04007442};
7443
Chris Mason35054392009-01-21 13:11:13 -05007444/*
7445 * btrfs doesn't support the bmap operation because swapfiles
7446 * use bmap to make a mapping of extents in the file. They assume
7447 * these extents won't change over the life of the file and they
7448 * use the bmap result to do IO directly to the drive.
7449 *
7450 * the btrfs bmap call would return logical addresses that aren't
7451 * suitable for IO and they also will change frequently as COW
7452 * operations happen. So, swapfile + btrfs == corruption.
7453 *
7454 * For now we're avoiding this by dropping bmap.
7455 */
Alexey Dobriyan7f094102009-09-21 17:01:10 -07007456static const struct address_space_operations btrfs_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04007457 .readpage = btrfs_readpage,
7458 .writepage = btrfs_writepage,
Chris Masonb293f022007-11-01 19:45:34 -04007459 .writepages = btrfs_writepages,
Chris Mason3ab2fb52007-11-08 10:59:22 -05007460 .readpages = btrfs_readpages,
Chris Mason16432982008-04-10 10:23:21 -04007461 .direct_IO = btrfs_direct_IO,
Chris Masona52d9a82007-08-27 16:49:44 -04007462 .invalidatepage = btrfs_invalidatepage,
7463 .releasepage = btrfs_releasepage,
Chris Masone6dcd2d2008-07-17 12:53:50 -04007464 .set_page_dirty = btrfs_set_page_dirty,
Andi Kleen465fdd92009-09-16 11:50:18 +02007465 .error_remove_page = generic_error_remove_page,
Chris Mason39279cc2007-06-12 06:35:45 -04007466};
7467
Alexey Dobriyan7f094102009-09-21 17:01:10 -07007468static const struct address_space_operations btrfs_symlink_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04007469 .readpage = btrfs_readpage,
7470 .writepage = btrfs_writepage,
Chris Mason2bf5a722007-08-30 11:54:02 -04007471 .invalidatepage = btrfs_invalidatepage,
7472 .releasepage = btrfs_releasepage,
Chris Mason39279cc2007-06-12 06:35:45 -04007473};
7474
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007475static const struct inode_operations btrfs_file_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007476 .getattr = btrfs_getattr,
7477 .setattr = btrfs_setattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007478 .setxattr = btrfs_setxattr,
7479 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05007480 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007481 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05007482 .permission = btrfs_permission,
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05007483 .fiemap = btrfs_fiemap,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007484 .get_acl = btrfs_get_acl,
Chris Mason39279cc2007-06-12 06:35:45 -04007485};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007486static const struct inode_operations btrfs_special_inode_operations = {
Josef Bacik618e21d2007-07-11 10:18:17 -04007487 .getattr = btrfs_getattr,
7488 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05007489 .permission = btrfs_permission,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007490 .setxattr = btrfs_setxattr,
7491 .getxattr = btrfs_getxattr,
Josef Bacik33268ea2008-07-24 12:16:36 -04007492 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007493 .removexattr = btrfs_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007494 .get_acl = btrfs_get_acl,
Josef Bacik618e21d2007-07-11 10:18:17 -04007495};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007496static const struct inode_operations btrfs_symlink_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007497 .readlink = generic_readlink,
7498 .follow_link = page_follow_link_light,
7499 .put_link = page_put_link,
Li Zefanf2095612010-11-19 02:05:24 +00007500 .getattr = btrfs_getattr,
Josef Bacik22c44fe2011-11-30 10:45:38 -05007501 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05007502 .permission = btrfs_permission,
Jim Owens0279b4c2009-02-04 09:29:13 -05007503 .setxattr = btrfs_setxattr,
7504 .getxattr = btrfs_getxattr,
7505 .listxattr = btrfs_listxattr,
7506 .removexattr = btrfs_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007507 .get_acl = btrfs_get_acl,
Chris Mason39279cc2007-06-12 06:35:45 -04007508};
Yan, Zheng76dda932009-09-21 16:00:26 -04007509
Alexey Dobriyan82d339d2009-10-09 09:54:36 -04007510const struct dentry_operations btrfs_dentry_operations = {
Yan, Zheng76dda932009-09-21 16:00:26 -04007511 .d_delete = btrfs_dentry_delete,
Josef Bacikb4aff1f2011-06-28 16:18:59 -04007512 .d_release = btrfs_dentry_release,
Yan, Zheng76dda932009-09-21 16:00:26 -04007513};