blob: 341a8670165f6f618c09aabb32bf9b7157187de5 [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 Masonb888db2b2007-08-27 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 Masonb888db2b2007-08-27 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 Masonf03d9301f2009-02-04 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 Masonf03d9301f2009-02-04 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 +
600 async_extent->ram_size - 1, GFP_NOFS);
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,
628 async_extent->start + async_extent->ram_size - 1,
629 GFP_NOFS);
Chris Mason771ed682008-11-06 22:02:51 -0500630
Josef Bacik7a7eaa42011-04-13 12:54:33 -0400631 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +0000632 BUG_ON(IS_ERR(trans));
Josef Bacik74b21072011-04-13 12:02:53 -0400633 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Mason771ed682008-11-06 22:02:51 -0500634 ret = btrfs_reserve_extent(trans, root,
635 async_extent->compressed_size,
636 async_extent->compressed_size,
Josef Bacik81c9ad22012-01-18 10:56:06 -0500637 0, alloc_hint, &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 +
651 async_extent->ram_size - 1, GFP_NOFS);
652 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,
Josef Bacik81c9ad22012-01-18 10:56:06 -0500840 &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 Masonb888db2b2007-08-27 16:49:44 -0400900 }
Chris Masonb888db2b2007-08-27 16:49:44 -0400901out:
Chris Mason771ed682008-11-06 22:02:51 -0500902 ret = 0;
Chris Masonb888db2b2007-08-27 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 Masonb888db2b2007-08-27 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);
Chris Masoncea9e442008-04-09 16:28:12 -04001427
Chris Masond3977122009-01-05 21:25:51 -05001428 if (map_length < length + size)
Chris Mason239b14b2008-03-24 15:02:07 -04001429 return 1;
Andi Kleen411fc6b2010-10-29 15:14:31 -04001430 return ret;
Chris Mason239b14b2008-03-24 15:02:07 -04001431}
1432
Chris Masond352ac62008-09-29 15:18:18 -04001433/*
1434 * in order to insert checksums into the metadata in large chunks,
1435 * we wait until bio submission time. All the pages in the bio are
1436 * checksummed and sums are attached onto the ordered extent record.
1437 *
1438 * At IO completion time the cums attached on the ordered extent record
1439 * are inserted into the btree
1440 */
Chris Masond3977122009-01-05 21:25:51 -05001441static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1442 struct bio *bio, int mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001443 unsigned long bio_flags,
1444 u64 bio_offset)
Chris Mason065631f2008-02-20 12:07:25 -05001445{
Chris Mason065631f2008-02-20 12:07:25 -05001446 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason065631f2008-02-20 12:07:25 -05001447 int ret = 0;
Chris Masone0156402008-04-16 11:15:20 -04001448
Chris Masond20f7042008-12-08 16:58:54 -05001449 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
Chris Mason44b8bd72008-04-16 11:14:51 -04001450 BUG_ON(ret);
Chris Mason4a69a412008-11-06 22:03:00 -05001451 return 0;
1452}
Chris Masone0156402008-04-16 11:15:20 -04001453
Chris Mason4a69a412008-11-06 22:03:00 -05001454/*
1455 * in order to insert checksums into the metadata in large chunks,
1456 * we wait until bio submission time. All the pages in the bio are
1457 * checksummed and sums are attached onto the ordered extent record.
1458 *
1459 * At IO completion time the cums attached on the ordered extent record
1460 * are inserted into the btree
1461 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001462static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001463 int mirror_num, unsigned long bio_flags,
1464 u64 bio_offset)
Chris Mason4a69a412008-11-06 22:03:00 -05001465{
1466 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason8b712842008-06-11 16:50:36 -04001467 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
Chris Mason44b8bd72008-04-16 11:14:51 -04001468}
1469
Chris Masond352ac62008-09-29 15:18:18 -04001470/*
Chris Masoncad321a2008-12-17 14:51:42 -05001471 * extent_io.c submission hook. This does the right thing for csum calculation
1472 * on write, or reading the csums from the tree before a read
Chris Masond352ac62008-09-29 15:18:18 -04001473 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001474static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001475 int mirror_num, unsigned long bio_flags,
1476 u64 bio_offset)
Chris Mason44b8bd72008-04-16 11:14:51 -04001477{
1478 struct btrfs_root *root = BTRFS_I(inode)->root;
1479 int ret = 0;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001480 int skip_sum;
Chris Mason44b8bd72008-04-16 11:14:51 -04001481
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001482 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masoncad321a2008-12-17 14:51:42 -05001483
Chris Mason2cf85722011-07-26 15:35:09 -04001484 if (btrfs_is_free_space_inode(root, inode))
Josef Bacik0cb59c92010-07-02 12:14:14 -04001485 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1486 else
1487 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001488 BUG_ON(ret);
Chris Mason065631f2008-02-20 12:07:25 -05001489
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001490 if (!(rw & REQ_WRITE)) {
Chris Masond20f7042008-12-08 16:58:54 -05001491 if (bio_flags & EXTENT_BIO_COMPRESSED) {
Chris Masonc8b97812008-10-29 14:49:59 -04001492 return btrfs_submit_compressed_read(inode, bio,
1493 mirror_num, bio_flags);
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001494 } else if (!skip_sum) {
1495 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1496 if (ret)
1497 return ret;
1498 }
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001499 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001500 } else if (!skip_sum) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001501 /* csum items have already been cloned */
1502 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1503 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001504 /* we're doing a write, do the async checksumming */
1505 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
Chris Mason44b8bd72008-04-16 11:14:51 -04001506 inode, rw, bio, mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001507 bio_flags, bio_offset,
1508 __btrfs_submit_bio_start,
Chris Mason4a69a412008-11-06 22:03:00 -05001509 __btrfs_submit_bio_done);
Chris Mason19b9bdb2008-10-30 14:23:13 -04001510 }
1511
Chris Mason0b86a832008-03-24 15:01:56 -04001512mapit:
Chris Mason8b712842008-06-11 16:50:36 -04001513 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
Chris Mason065631f2008-02-20 12:07:25 -05001514}
Chris Mason6885f302008-02-20 16:11:05 -05001515
Chris Masond352ac62008-09-29 15:18:18 -04001516/*
1517 * given a list of ordered sums record them in the inode. This happens
1518 * at IO completion time based on sums calculated at bio submission time.
1519 */
Chris Masonba1da2f2008-07-17 12:54:15 -04001520static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
Chris Masone6dcd2d2008-07-17 12:53:50 -04001521 struct inode *inode, u64 file_offset,
1522 struct list_head *list)
1523{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001524 struct btrfs_ordered_sum *sum;
1525
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001526 list_for_each_entry(sum, list, list) {
Chris Masond20f7042008-12-08 16:58:54 -05001527 btrfs_csum_file_blocks(trans,
1528 BTRFS_I(inode)->root->fs_info->csum_root, sum);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001529 }
1530 return 0;
1531}
1532
Josef Bacik2ac55d42010-02-03 19:33:23 +00001533int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1534 struct extent_state **cached_state)
Chris Masonea8c2812008-08-04 23:17:27 -04001535{
Chris Masond3977122009-01-05 21:25:51 -05001536 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
Chris Mason771ed682008-11-06 22:02:51 -05001537 WARN_ON(1);
Chris Masonea8c2812008-08-04 23:17:27 -04001538 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00001539 cached_state, GFP_NOFS);
Chris Masonea8c2812008-08-04 23:17:27 -04001540}
1541
Chris Masond352ac62008-09-29 15:18:18 -04001542/* see btrfs_writepage_start_hook for details on why this is required */
Chris Mason247e7432008-07-17 12:53:51 -04001543struct btrfs_writepage_fixup {
1544 struct page *page;
1545 struct btrfs_work work;
1546};
1547
Christoph Hellwigb2950862008-12-02 09:54:17 -05001548static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
Chris Mason247e7432008-07-17 12:53:51 -04001549{
1550 struct btrfs_writepage_fixup *fixup;
1551 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001552 struct extent_state *cached_state = NULL;
Chris Mason247e7432008-07-17 12:53:51 -04001553 struct page *page;
1554 struct inode *inode;
1555 u64 page_start;
1556 u64 page_end;
Jeff Mahoney87826df2012-02-15 16:23:57 +01001557 int ret;
Chris Mason247e7432008-07-17 12:53:51 -04001558
1559 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1560 page = fixup->page;
Chris Mason4a096752008-07-21 10:29:44 -04001561again:
Chris Mason247e7432008-07-17 12:53:51 -04001562 lock_page(page);
1563 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1564 ClearPageChecked(page);
1565 goto out_page;
1566 }
1567
1568 inode = page->mapping->host;
1569 page_start = page_offset(page);
1570 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1571
Josef Bacik2ac55d42010-02-03 19:33:23 +00001572 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1573 &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001574
1575 /* already ordered? We're done */
Chris Mason8b62b722009-09-02 16:53:46 -04001576 if (PagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001577 goto out;
Chris Mason4a096752008-07-21 10:29:44 -04001578
1579 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1580 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00001581 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1582 page_end, &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001583 unlock_page(page);
1584 btrfs_start_ordered_extent(inode, ordered, 1);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001585 btrfs_put_ordered_extent(ordered);
Chris Mason4a096752008-07-21 10:29:44 -04001586 goto again;
1587 }
Chris Mason247e7432008-07-17 12:53:51 -04001588
Jeff Mahoney87826df2012-02-15 16:23:57 +01001589 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1590 if (ret) {
1591 mapping_set_error(page->mapping, ret);
1592 end_extent_writepage(page, ret, page_start, page_end);
1593 ClearPageChecked(page);
1594 goto out;
1595 }
1596
Josef Bacik2ac55d42010-02-03 19:33:23 +00001597 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
Chris Mason247e7432008-07-17 12:53:51 -04001598 ClearPageChecked(page);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001599 set_page_dirty(page);
Chris Mason247e7432008-07-17 12:53:51 -04001600out:
Josef Bacik2ac55d42010-02-03 19:33:23 +00001601 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1602 &cached_state, GFP_NOFS);
Chris Mason247e7432008-07-17 12:53:51 -04001603out_page:
1604 unlock_page(page);
1605 page_cache_release(page);
Miao Xieb897abe2011-01-26 16:19:22 +08001606 kfree(fixup);
Chris Mason247e7432008-07-17 12:53:51 -04001607}
1608
1609/*
1610 * There are a few paths in the higher layers of the kernel that directly
1611 * set the page dirty bit without asking the filesystem if it is a
1612 * good idea. This causes problems because we want to make sure COW
1613 * properly happens and the data=ordered rules are followed.
1614 *
Chris Masonc8b97812008-10-29 14:49:59 -04001615 * In our case any range that doesn't have the ORDERED bit set
Chris Mason247e7432008-07-17 12:53:51 -04001616 * hasn't been properly setup for IO. We kick off an async process
1617 * to fix it up. The async helper will wait for ordered extents, set
1618 * the delalloc bit and make it safe to write the page.
1619 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001620static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
Chris Mason247e7432008-07-17 12:53:51 -04001621{
1622 struct inode *inode = page->mapping->host;
1623 struct btrfs_writepage_fixup *fixup;
1624 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason247e7432008-07-17 12:53:51 -04001625
Chris Mason8b62b722009-09-02 16:53:46 -04001626 /* this page is properly in the ordered list */
1627 if (TestClearPagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001628 return 0;
1629
1630 if (PageChecked(page))
1631 return -EAGAIN;
1632
1633 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1634 if (!fixup)
1635 return -EAGAIN;
Chris Masonf4219502008-07-22 11:18:09 -04001636
Chris Mason247e7432008-07-17 12:53:51 -04001637 SetPageChecked(page);
1638 page_cache_get(page);
1639 fixup->work.func = btrfs_writepage_fixup_worker;
1640 fixup->page = page;
1641 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001642 return -EBUSY;
Chris Mason247e7432008-07-17 12:53:51 -04001643}
1644
Yan Zhengd899e052008-10-30 14:25:28 -04001645static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1646 struct inode *inode, u64 file_pos,
1647 u64 disk_bytenr, u64 disk_num_bytes,
1648 u64 num_bytes, u64 ram_bytes,
1649 u8 compression, u8 encryption,
1650 u16 other_encoding, int extent_type)
1651{
1652 struct btrfs_root *root = BTRFS_I(inode)->root;
1653 struct btrfs_file_extent_item *fi;
1654 struct btrfs_path *path;
1655 struct extent_buffer *leaf;
1656 struct btrfs_key ins;
1657 u64 hint;
1658 int ret;
1659
1660 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07001661 if (!path)
1662 return -ENOMEM;
Yan Zhengd899e052008-10-30 14:25:28 -04001663
Chris Masonb9473432009-03-13 11:00:37 -04001664 path->leave_spinning = 1;
Chris Masona1ed8352009-09-11 12:27:37 -04001665
1666 /*
1667 * we may be replacing one extent in the tree with another.
1668 * The new extent is pinned in the extent map, and we don't want
1669 * to drop it from the cache until it is completely in the btree.
1670 *
1671 * So, tell btrfs_drop_extents to leave this extent in the cache.
1672 * the caller is expected to unpin it and allow it to be merged
1673 * with the others.
1674 */
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001675 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1676 &hint, 0);
Yan Zhengd899e052008-10-30 14:25:28 -04001677 BUG_ON(ret);
1678
Li Zefan33345d012011-04-20 10:31:50 +08001679 ins.objectid = btrfs_ino(inode);
Yan Zhengd899e052008-10-30 14:25:28 -04001680 ins.offset = file_pos;
1681 ins.type = BTRFS_EXTENT_DATA_KEY;
1682 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1683 BUG_ON(ret);
1684 leaf = path->nodes[0];
1685 fi = btrfs_item_ptr(leaf, path->slots[0],
1686 struct btrfs_file_extent_item);
1687 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1688 btrfs_set_file_extent_type(leaf, fi, extent_type);
1689 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1690 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1691 btrfs_set_file_extent_offset(leaf, fi, 0);
1692 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1693 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1694 btrfs_set_file_extent_compression(leaf, fi, compression);
1695 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1696 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
Chris Masonb9473432009-03-13 11:00:37 -04001697
1698 btrfs_unlock_up_safe(path, 1);
1699 btrfs_set_lock_blocking(leaf);
1700
Yan Zhengd899e052008-10-30 14:25:28 -04001701 btrfs_mark_buffer_dirty(leaf);
1702
1703 inode_add_bytes(inode, num_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -04001704
1705 ins.objectid = disk_bytenr;
1706 ins.offset = disk_num_bytes;
1707 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001708 ret = btrfs_alloc_reserved_file_extent(trans, root,
1709 root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001710 btrfs_ino(inode), file_pos, &ins);
Yan Zhengd899e052008-10-30 14:25:28 -04001711 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -04001712 btrfs_free_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04001713
Yan Zhengd899e052008-10-30 14:25:28 -04001714 return 0;
1715}
1716
Chris Mason5d13a982009-03-13 11:41:46 -04001717/*
1718 * helper function for btrfs_finish_ordered_io, this
1719 * just reads in some of the csum leaves to prime them into ram
1720 * before we start the transaction. It limits the amount of btree
1721 * reads required while inside the transaction.
1722 */
Chris Masond352ac62008-09-29 15:18:18 -04001723/* as ordered data IO finishes, this gets called so we can finish
1724 * an ordered extent if the range of bytes in the file it covers are
1725 * fully written.
1726 */
Chris Mason211f90e2008-07-18 11:56:15 -04001727static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001728{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001729 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001730 struct btrfs_trans_handle *trans = NULL;
Chris Mason5d13a982009-03-13 11:41:46 -04001731 struct btrfs_ordered_extent *ordered_extent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001732 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001733 struct extent_state *cached_state = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08001734 int compress_type = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001735 int ret;
Li Zefan82d59022011-04-20 10:33:24 +08001736 bool nolock;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001737
Josef Bacik5a1a3df2010-02-02 20:51:14 +00001738 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1739 end - start + 1);
Chris Masonba1da2f2008-07-17 12:54:15 -04001740 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -04001741 return 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001742 BUG_ON(!ordered_extent);
Josef Bacikefd049f2010-02-02 20:50:10 +00001743
Chris Mason2cf85722011-07-26 15:35:09 -04001744 nolock = btrfs_is_free_space_inode(root, inode);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001745
Yan, Zhengc2167752009-11-12 09:34:21 +00001746 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1747 BUG_ON(!list_empty(&ordered_extent->list));
1748 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1749 if (!ret) {
Josef Bacik0cb59c92010-07-02 12:14:14 -04001750 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001751 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001752 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001753 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001754 BUG_ON(IS_ERR(trans));
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001755 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Mason21151332011-11-10 20:39:08 -05001756 ret = btrfs_update_inode_fallback(trans, root, inode);
Yan, Zhengc2167752009-11-12 09:34:21 +00001757 BUG_ON(ret);
Yan, Zhengc2167752009-11-12 09:34:21 +00001758 }
1759 goto out;
1760 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04001761
Josef Bacik2ac55d42010-02-03 19:33:23 +00001762 lock_extent_bits(io_tree, ordered_extent->file_offset,
1763 ordered_extent->file_offset + ordered_extent->len - 1,
1764 0, &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001765
Josef Bacik0cb59c92010-07-02 12:14:14 -04001766 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001767 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001768 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001769 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00001770 BUG_ON(IS_ERR(trans));
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001771 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Yan, Zhengc2167752009-11-12 09:34:21 +00001772
Chris Masonc8b97812008-10-29 14:49:59 -04001773 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
Li Zefan261507a02010-12-17 14:21:50 +08001774 compress_type = ordered_extent->compress_type;
Yan Zhengd899e052008-10-30 14:25:28 -04001775 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
Li Zefan261507a02010-12-17 14:21:50 +08001776 BUG_ON(compress_type);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00001777 ret = btrfs_mark_extent_written(trans, inode,
Yan Zhengd899e052008-10-30 14:25:28 -04001778 ordered_extent->file_offset,
1779 ordered_extent->file_offset +
1780 ordered_extent->len);
1781 BUG_ON(ret);
1782 } else {
Josef Bacik0af3d002010-06-21 14:48:16 -04001783 BUG_ON(root == root->fs_info->tree_root);
Yan Zhengd899e052008-10-30 14:25:28 -04001784 ret = insert_reserved_file_extent(trans, inode,
1785 ordered_extent->file_offset,
1786 ordered_extent->start,
1787 ordered_extent->disk_len,
1788 ordered_extent->len,
1789 ordered_extent->len,
Li Zefan261507a02010-12-17 14:21:50 +08001790 compress_type, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04001791 BTRFS_FILE_EXTENT_REG);
Chris Masona1ed8352009-09-11 12:27:37 -04001792 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1793 ordered_extent->file_offset,
1794 ordered_extent->len);
Yan Zhengd899e052008-10-30 14:25:28 -04001795 BUG_ON(ret);
1796 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00001797 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1798 ordered_extent->file_offset +
1799 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1800
Chris Masone6dcd2d2008-07-17 12:53:50 -04001801 add_pending_csums(trans, inode, ordered_extent->file_offset,
1802 &ordered_extent->list);
1803
Josef Bacik1ef30be2011-04-05 19:25:36 -04001804 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
Miao Xiea39f7522011-09-11 10:52:25 -04001805 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
Chris Mason21151332011-11-10 20:39:08 -05001806 ret = btrfs_update_inode_fallback(trans, root, inode);
Josef Bacik1ef30be2011-04-05 19:25:36 -04001807 BUG_ON(ret);
1808 }
1809 ret = 0;
Yan, Zhengc2167752009-11-12 09:34:21 +00001810out:
Josef Bacik5b0e95b2011-10-06 08:58:24 -04001811 if (root != root->fs_info->tree_root)
Josef Bacik0cb59c92010-07-02 12:14:14 -04001812 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
Josef Bacik5b0e95b2011-10-06 08:58:24 -04001813 if (trans) {
1814 if (nolock)
Josef Bacik0cb59c92010-07-02 12:14:14 -04001815 btrfs_end_transaction_nolock(trans, root);
Josef Bacik5b0e95b2011-10-06 08:58:24 -04001816 else
Josef Bacik0cb59c92010-07-02 12:14:14 -04001817 btrfs_end_transaction(trans, root);
1818 }
1819
Chris Masone6dcd2d2008-07-17 12:53:50 -04001820 /* once for us */
1821 btrfs_put_ordered_extent(ordered_extent);
1822 /* once for the tree */
1823 btrfs_put_ordered_extent(ordered_extent);
1824
Chris Masone6dcd2d2008-07-17 12:53:50 -04001825 return 0;
1826}
1827
Christoph Hellwigb2950862008-12-02 09:54:17 -05001828static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason211f90e2008-07-18 11:56:15 -04001829 struct extent_state *state, int uptodate)
1830{
liubo1abe9b82011-03-24 11:18:59 +00001831 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1832
Chris Mason8b62b722009-09-02 16:53:46 -04001833 ClearPagePrivate2(page);
Chris Mason211f90e2008-07-18 11:56:15 -04001834 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1835}
1836
Chris Masond352ac62008-09-29 15:18:18 -04001837/*
Chris Masond352ac62008-09-29 15:18:18 -04001838 * when reads are done, we need to check csums to verify the data is correct
Jan Schmidt4a54c8c2011-07-22 15:41:52 +02001839 * if there's a match, we allow the bio to finish. If not, the code in
1840 * extent_io.c will try to find good copies for us.
Chris Masond352ac62008-09-29 15:18:18 -04001841 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001842static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason70dec802008-01-29 09:59:12 -05001843 struct extent_state *state)
Chris Mason07157aa2007-08-30 08:50:51 -04001844{
Chris Mason35ebb932007-10-30 16:56:53 -04001845 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
Chris Mason07157aa2007-08-30 08:50:51 -04001846 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05001847 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason07157aa2007-08-30 08:50:51 -04001848 char *kaddr;
Chris Masonaadfeb62008-01-29 09:10:27 -05001849 u64 private = ~(u32)0;
Chris Mason07157aa2007-08-30 08:50:51 -04001850 int ret;
Chris Masonff79f812007-10-15 16:22:25 -04001851 struct btrfs_root *root = BTRFS_I(inode)->root;
1852 u32 csum = ~(u32)0;
Chris Masond1310b22008-01-24 16:13:08 -05001853
Chris Masond20f7042008-12-08 16:58:54 -05001854 if (PageChecked(page)) {
1855 ClearPageChecked(page);
1856 goto good;
1857 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001858
1859 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Jan Schmidt08d2f342011-05-04 16:18:50 +02001860 goto good;
Chris Masond20f7042008-12-08 16:58:54 -05001861
Yan Zheng17d217f2008-12-12 10:03:38 -05001862 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
Chris Mason9655d292009-09-02 15:22:30 -04001863 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001864 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1865 GFP_NOFS);
1866 return 0;
1867 }
1868
Yanc2e639f2008-02-04 08:57:25 -05001869 if (state && state->start == start) {
Chris Mason70dec802008-01-29 09:59:12 -05001870 private = state->private;
1871 ret = 0;
1872 } else {
1873 ret = get_state_private(io_tree, start, &private);
1874 }
Chris Mason9ab86c82009-01-07 09:48:51 -05001875 kaddr = kmap_atomic(page, KM_USER0);
Chris Masond3977122009-01-05 21:25:51 -05001876 if (ret)
Chris Mason07157aa2007-08-30 08:50:51 -04001877 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05001878
Chris Masonff79f812007-10-15 16:22:25 -04001879 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1880 btrfs_csum_final(csum, (char *)&csum);
Chris Masond3977122009-01-05 21:25:51 -05001881 if (csum != private)
Chris Mason07157aa2007-08-30 08:50:51 -04001882 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05001883
Chris Mason9ab86c82009-01-07 09:48:51 -05001884 kunmap_atomic(kaddr, KM_USER0);
Chris Masond20f7042008-12-08 16:58:54 -05001885good:
Chris Mason07157aa2007-08-30 08:50:51 -04001886 return 0;
1887
1888zeroit:
Chris Mason945d8962011-05-22 12:33:42 -04001889 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
Li Zefan33345d012011-04-20 10:31:50 +08001890 "private %llu\n",
1891 (unsigned long long)btrfs_ino(page->mapping->host),
Chris Mason193f2842009-04-27 07:29:05 -04001892 (unsigned long long)start, csum,
1893 (unsigned long long)private);
Chris Masondb945352007-10-15 16:15:53 -04001894 memset(kaddr + offset, 1, end - start + 1);
1895 flush_dcache_page(page);
Chris Mason9ab86c82009-01-07 09:48:51 -05001896 kunmap_atomic(kaddr, KM_USER0);
Chris Mason3b951512008-04-17 11:29:12 -04001897 if (private == 0)
1898 return 0;
Chris Mason7e383262008-04-09 16:28:12 -04001899 return -EIO;
Chris Mason07157aa2007-08-30 08:50:51 -04001900}
Chris Masonb888db2b2007-08-27 16:49:44 -04001901
Yan, Zheng24bbcf02009-11-12 09:36:34 +00001902struct delayed_iput {
1903 struct list_head list;
1904 struct inode *inode;
1905};
1906
1907void btrfs_add_delayed_iput(struct inode *inode)
1908{
1909 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1910 struct delayed_iput *delayed;
1911
1912 if (atomic_add_unless(&inode->i_count, -1, 1))
1913 return;
1914
1915 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1916 delayed->inode = inode;
1917
1918 spin_lock(&fs_info->delayed_iput_lock);
1919 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1920 spin_unlock(&fs_info->delayed_iput_lock);
1921}
1922
1923void btrfs_run_delayed_iputs(struct btrfs_root *root)
1924{
1925 LIST_HEAD(list);
1926 struct btrfs_fs_info *fs_info = root->fs_info;
1927 struct delayed_iput *delayed;
1928 int empty;
1929
1930 spin_lock(&fs_info->delayed_iput_lock);
1931 empty = list_empty(&fs_info->delayed_iputs);
1932 spin_unlock(&fs_info->delayed_iput_lock);
1933 if (empty)
1934 return;
1935
1936 down_read(&root->fs_info->cleanup_work_sem);
1937 spin_lock(&fs_info->delayed_iput_lock);
1938 list_splice_init(&fs_info->delayed_iputs, &list);
1939 spin_unlock(&fs_info->delayed_iput_lock);
1940
1941 while (!list_empty(&list)) {
1942 delayed = list_entry(list.next, struct delayed_iput, list);
1943 list_del(&delayed->list);
1944 iput(delayed->inode);
1945 kfree(delayed);
1946 }
1947 up_read(&root->fs_info->cleanup_work_sem);
1948}
1949
Yan, Zhengd68fc572010-05-16 10:49:58 -04001950enum btrfs_orphan_cleanup_state {
1951 ORPHAN_CLEANUP_STARTED = 1,
1952 ORPHAN_CLEANUP_DONE = 2,
1953};
1954
1955/*
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08001956 * This is called in transaction commit time. If there are no orphan
Yan, Zhengd68fc572010-05-16 10:49:58 -04001957 * files in the subvolume, it removes orphan item and frees block_rsv
1958 * structure.
1959 */
1960void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1961 struct btrfs_root *root)
1962{
Josef Bacik90290e12011-12-02 15:44:12 -05001963 struct btrfs_block_rsv *block_rsv;
Yan, Zhengd68fc572010-05-16 10:49:58 -04001964 int ret;
1965
1966 if (!list_empty(&root->orphan_list) ||
1967 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1968 return;
1969
Josef Bacik90290e12011-12-02 15:44:12 -05001970 spin_lock(&root->orphan_lock);
1971 if (!list_empty(&root->orphan_list)) {
1972 spin_unlock(&root->orphan_lock);
1973 return;
1974 }
1975
1976 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
1977 spin_unlock(&root->orphan_lock);
1978 return;
1979 }
1980
1981 block_rsv = root->orphan_block_rsv;
1982 root->orphan_block_rsv = NULL;
1983 spin_unlock(&root->orphan_lock);
1984
Yan, Zhengd68fc572010-05-16 10:49:58 -04001985 if (root->orphan_item_inserted &&
1986 btrfs_root_refs(&root->root_item) > 0) {
1987 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1988 root->root_key.objectid);
1989 BUG_ON(ret);
1990 root->orphan_item_inserted = 0;
1991 }
1992
Josef Bacik90290e12011-12-02 15:44:12 -05001993 if (block_rsv) {
1994 WARN_ON(block_rsv->size > 0);
1995 btrfs_free_block_rsv(root, block_rsv);
Yan, Zhengd68fc572010-05-16 10:49:58 -04001996 }
1997}
1998
1999/*
Josef Bacik7b128762008-07-24 12:17:14 -04002000 * This creates an orphan entry for the given inode in case something goes
2001 * wrong in the middle of an unlink/truncate.
Yan, Zhengd68fc572010-05-16 10:49:58 -04002002 *
2003 * NOTE: caller of this function should reserve 5 units of metadata for
2004 * this function.
Josef Bacik7b128762008-07-24 12:17:14 -04002005 */
2006int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2007{
2008 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002009 struct btrfs_block_rsv *block_rsv = NULL;
2010 int reserve = 0;
2011 int insert = 0;
2012 int ret;
Josef Bacik7b128762008-07-24 12:17:14 -04002013
Yan, Zhengd68fc572010-05-16 10:49:58 -04002014 if (!root->orphan_block_rsv) {
2015 block_rsv = btrfs_alloc_block_rsv(root);
Tsutomu Itohb5324022011-07-19 07:27:20 +00002016 if (!block_rsv)
2017 return -ENOMEM;
Josef Bacik7b128762008-07-24 12:17:14 -04002018 }
2019
Yan, Zhengd68fc572010-05-16 10:49:58 -04002020 spin_lock(&root->orphan_lock);
2021 if (!root->orphan_block_rsv) {
2022 root->orphan_block_rsv = block_rsv;
2023 } else if (block_rsv) {
2024 btrfs_free_block_rsv(root, block_rsv);
2025 block_rsv = NULL;
2026 }
Josef Bacik7b128762008-07-24 12:17:14 -04002027
Yan, Zhengd68fc572010-05-16 10:49:58 -04002028 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2029 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2030#if 0
2031 /*
2032 * For proper ENOSPC handling, we should do orphan
2033 * cleanup when mounting. But this introduces backward
2034 * compatibility issue.
2035 */
2036 if (!xchg(&root->orphan_item_inserted, 1))
2037 insert = 2;
2038 else
2039 insert = 1;
2040#endif
2041 insert = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002042 }
Josef Bacik7b128762008-07-24 12:17:14 -04002043
Yan, Zhengd68fc572010-05-16 10:49:58 -04002044 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2045 BTRFS_I(inode)->orphan_meta_reserved = 1;
2046 reserve = 1;
2047 }
2048 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002049
Yan, Zhengd68fc572010-05-16 10:49:58 -04002050 /* grab metadata reservation from transaction handle */
2051 if (reserve) {
2052 ret = btrfs_orphan_reserve_metadata(trans, inode);
2053 BUG_ON(ret);
2054 }
2055
2056 /* insert an orphan item to track this unlinked/truncated file */
2057 if (insert >= 1) {
Li Zefan33345d012011-04-20 10:31:50 +08002058 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
Josef Bacikee4d89f2011-12-13 12:55:58 -05002059 BUG_ON(ret && ret != -EEXIST);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002060 }
2061
2062 /* insert an orphan item to track subvolume contains orphan files */
2063 if (insert >= 2) {
2064 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2065 root->root_key.objectid);
2066 BUG_ON(ret);
2067 }
2068 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002069}
2070
2071/*
2072 * We have done the truncate/delete so we can go ahead and remove the orphan
2073 * item for this particular inode.
2074 */
2075int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2076{
2077 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002078 int delete_item = 0;
2079 int release_rsv = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002080 int ret = 0;
2081
Yan, Zhengd68fc572010-05-16 10:49:58 -04002082 spin_lock(&root->orphan_lock);
2083 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2084 list_del_init(&BTRFS_I(inode)->i_orphan);
2085 delete_item = 1;
Josef Bacik7b128762008-07-24 12:17:14 -04002086 }
2087
Yan, Zhengd68fc572010-05-16 10:49:58 -04002088 if (BTRFS_I(inode)->orphan_meta_reserved) {
2089 BTRFS_I(inode)->orphan_meta_reserved = 0;
2090 release_rsv = 1;
2091 }
2092 spin_unlock(&root->orphan_lock);
2093
2094 if (trans && delete_item) {
Li Zefan33345d012011-04-20 10:31:50 +08002095 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
Yan, Zhengd68fc572010-05-16 10:49:58 -04002096 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04002097 }
2098
Yan, Zhengd68fc572010-05-16 10:49:58 -04002099 if (release_rsv)
2100 btrfs_orphan_release_metadata(inode);
Josef Bacik7b128762008-07-24 12:17:14 -04002101
Yan, Zhengd68fc572010-05-16 10:49:58 -04002102 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002103}
2104
2105/*
2106 * this cleans up any orphans that may be left on the list from the last use
2107 * of this root.
2108 */
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002109int btrfs_orphan_cleanup(struct btrfs_root *root)
Josef Bacik7b128762008-07-24 12:17:14 -04002110{
2111 struct btrfs_path *path;
2112 struct extent_buffer *leaf;
Josef Bacik7b128762008-07-24 12:17:14 -04002113 struct btrfs_key key, found_key;
2114 struct btrfs_trans_handle *trans;
2115 struct inode *inode;
Josef Bacik8f6d7f42011-09-26 15:55:20 -04002116 u64 last_objectid = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002117 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2118
Yan, Zhengd68fc572010-05-16 10:49:58 -04002119 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002120 return 0;
Yan, Zhengc71bf092009-11-12 09:34:40 +00002121
2122 path = btrfs_alloc_path();
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002123 if (!path) {
2124 ret = -ENOMEM;
2125 goto out;
2126 }
Josef Bacik7b128762008-07-24 12:17:14 -04002127 path->reada = -1;
2128
2129 key.objectid = BTRFS_ORPHAN_OBJECTID;
2130 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2131 key.offset = (u64)-1;
2132
Josef Bacik7b128762008-07-24 12:17:14 -04002133 while (1) {
2134 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002135 if (ret < 0)
2136 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002137
2138 /*
2139 * if ret == 0 means we found what we were searching for, which
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002140 * is weird, but possible, so only screw with path if we didn't
Josef Bacik7b128762008-07-24 12:17:14 -04002141 * find the key and see if we have stuff that matches
2142 */
2143 if (ret > 0) {
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002144 ret = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04002145 if (path->slots[0] == 0)
2146 break;
2147 path->slots[0]--;
2148 }
2149
2150 /* pull out the item */
2151 leaf = path->nodes[0];
Josef Bacik7b128762008-07-24 12:17:14 -04002152 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2153
2154 /* make sure the item matches what we want */
2155 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2156 break;
2157 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2158 break;
2159
2160 /* release the path since we're done with it */
David Sterbab3b4aa72011-04-21 01:20:15 +02002161 btrfs_release_path(path);
Josef Bacik7b128762008-07-24 12:17:14 -04002162
2163 /*
2164 * this is where we are basically btrfs_lookup, without the
2165 * crossing root thing. we store the inode number in the
2166 * offset of the orphan item.
2167 */
Josef Bacik8f6d7f42011-09-26 15:55:20 -04002168
2169 if (found_key.offset == last_objectid) {
2170 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2171 "stopping orphan cleanup\n");
2172 ret = -EINVAL;
2173 goto out;
2174 }
2175
2176 last_objectid = found_key.offset;
2177
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002178 found_key.objectid = found_key.offset;
2179 found_key.type = BTRFS_INODE_ITEM_KEY;
2180 found_key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +00002181 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
Josef Bacika8c9e572011-09-21 16:55:59 -04002182 ret = PTR_RET(inode);
2183 if (ret && ret != -ESTALE)
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002184 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04002185
Arne Jansenf8e9e0b2011-12-14 20:12:02 -05002186 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2187 struct btrfs_root *dead_root;
2188 struct btrfs_fs_info *fs_info = root->fs_info;
2189 int is_dead_root = 0;
2190
2191 /*
2192 * this is an orphan in the tree root. Currently these
2193 * could come from 2 sources:
2194 * a) a snapshot deletion in progress
2195 * b) a free space cache inode
2196 * We need to distinguish those two, as the snapshot
2197 * orphan must not get deleted.
2198 * find_dead_roots already ran before us, so if this
2199 * is a snapshot deletion, we should find the root
2200 * in the dead_roots list
2201 */
2202 spin_lock(&fs_info->trans_lock);
2203 list_for_each_entry(dead_root, &fs_info->dead_roots,
2204 root_list) {
2205 if (dead_root->root_key.objectid ==
2206 found_key.objectid) {
2207 is_dead_root = 1;
2208 break;
2209 }
2210 }
2211 spin_unlock(&fs_info->trans_lock);
2212 if (is_dead_root) {
2213 /* prevent this orphan from being found again */
2214 key.offset = found_key.objectid - 1;
2215 continue;
2216 }
2217 }
Josef Bacika8c9e572011-09-21 16:55:59 -04002218 /*
2219 * Inode is already gone but the orphan item is still there,
2220 * kill the orphan item.
2221 */
2222 if (ret == -ESTALE) {
2223 trans = btrfs_start_transaction(root, 1);
2224 if (IS_ERR(trans)) {
2225 ret = PTR_ERR(trans);
2226 goto out;
2227 }
2228 ret = btrfs_del_orphan_item(trans, root,
2229 found_key.objectid);
2230 BUG_ON(ret);
2231 btrfs_end_transaction(trans, root);
2232 continue;
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002233 }
Josef Bacik7b128762008-07-24 12:17:14 -04002234
Josef Bacik7b128762008-07-24 12:17:14 -04002235 /*
2236 * add this inode to the orphan list so btrfs_orphan_del does
2237 * the proper thing when we hit it
2238 */
Yan, Zhengd68fc572010-05-16 10:49:58 -04002239 spin_lock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002240 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002241 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002242
Josef Bacik7b128762008-07-24 12:17:14 -04002243 /* if we have links, this was a truncate, lets do that */
2244 if (inode->i_nlink) {
Josef Bacika41ad392011-01-31 15:30:16 -05002245 if (!S_ISREG(inode->i_mode)) {
2246 WARN_ON(1);
2247 iput(inode);
2248 continue;
2249 }
Josef Bacik7b128762008-07-24 12:17:14 -04002250 nr_truncate++;
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002251 ret = btrfs_truncate(inode);
Josef Bacik7b128762008-07-24 12:17:14 -04002252 } else {
2253 nr_unlink++;
2254 }
2255
2256 /* this will do delete_inode and everything for us */
2257 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002258 if (ret)
2259 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002260 }
Miao Xie3254c872011-11-10 20:45:05 -05002261 /* release the path since we're done with it */
2262 btrfs_release_path(path);
2263
Yan, Zhengd68fc572010-05-16 10:49:58 -04002264 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2265
2266 if (root->orphan_block_rsv)
2267 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2268 (u64)-1);
2269
2270 if (root->orphan_block_rsv || root->orphan_item_inserted) {
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002271 trans = btrfs_join_transaction(root);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002272 if (!IS_ERR(trans))
2273 btrfs_end_transaction(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002274 }
Josef Bacik7b128762008-07-24 12:17:14 -04002275
2276 if (nr_unlink)
2277 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2278 if (nr_truncate)
2279 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05002280
2281out:
2282 if (ret)
2283 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2284 btrfs_free_path(path);
2285 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04002286}
2287
Chris Masond352ac62008-09-29 15:18:18 -04002288/*
Chris Mason46a53cc2009-04-27 11:47:50 -04002289 * very simple check to peek ahead in the leaf looking for xattrs. If we
2290 * don't find any xattrs, we know there can't be any acls.
2291 *
2292 * slot is the slot the inode is in, objectid is the objectid of the inode
2293 */
2294static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2295 int slot, u64 objectid)
2296{
2297 u32 nritems = btrfs_header_nritems(leaf);
2298 struct btrfs_key found_key;
2299 int scanned = 0;
2300
2301 slot++;
2302 while (slot < nritems) {
2303 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2304
2305 /* we found a different objectid, there must not be acls */
2306 if (found_key.objectid != objectid)
2307 return 0;
2308
2309 /* we found an xattr, assume we've got an acl */
2310 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2311 return 1;
2312
2313 /*
2314 * we found a key greater than an xattr key, there can't
2315 * be any acls later on
2316 */
2317 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2318 return 0;
2319
2320 slot++;
2321 scanned++;
2322
2323 /*
2324 * it goes inode, inode backrefs, xattrs, extents,
2325 * so if there are a ton of hard links to an inode there can
2326 * be a lot of backrefs. Don't waste time searching too hard,
2327 * this is just an optimization
2328 */
2329 if (scanned >= 8)
2330 break;
2331 }
2332 /* we hit the end of the leaf before we found an xattr or
2333 * something larger than an xattr. We have to assume the inode
2334 * has acls
2335 */
2336 return 1;
2337}
2338
2339/*
Chris Masond352ac62008-09-29 15:18:18 -04002340 * read an inode from the btree into the in-memory inode
2341 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002342static void btrfs_read_locked_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002343{
2344 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002345 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002346 struct btrfs_inode_item *inode_item;
Chris Mason0b86a832008-03-24 15:01:56 -04002347 struct btrfs_timespec *tspec;
Chris Mason39279cc2007-06-12 06:35:45 -04002348 struct btrfs_root *root = BTRFS_I(inode)->root;
2349 struct btrfs_key location;
Chris Mason46a53cc2009-04-27 11:47:50 -04002350 int maybe_acls;
Josef Bacik618e21d2007-07-11 10:18:17 -04002351 u32 rdev;
Chris Mason39279cc2007-06-12 06:35:45 -04002352 int ret;
Miao Xie2f7e33d2011-06-23 07:27:13 +00002353 bool filled = false;
2354
2355 ret = btrfs_fill_inode(inode, &rdev);
2356 if (!ret)
2357 filled = true;
Chris Mason39279cc2007-06-12 06:35:45 -04002358
2359 path = btrfs_alloc_path();
Mark Fasheh1748f842011-07-12 11:25:31 -07002360 if (!path)
2361 goto make_bad;
2362
Josef Bacikd90c7322011-05-17 09:50:54 -04002363 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04002364 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
Chris Masondc17ff82008-01-08 15:46:30 -05002365
Chris Mason39279cc2007-06-12 06:35:45 -04002366 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04002367 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04002368 goto make_bad;
Chris Mason39279cc2007-06-12 06:35:45 -04002369
Chris Mason5f39d392007-10-15 16:14:19 -04002370 leaf = path->nodes[0];
Miao Xie2f7e33d2011-06-23 07:27:13 +00002371
2372 if (filled)
2373 goto cache_acl;
2374
Chris Mason5f39d392007-10-15 16:14:19 -04002375 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2376 struct btrfs_inode_item);
Chris Mason5f39d392007-10-15 16:14:19 -04002377 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
Miklos Szeredibfe86842011-10-28 14:13:29 +02002378 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04002379 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2380 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
Chris Masondbe674a2008-07-17 12:54:05 -04002381 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04002382
2383 tspec = btrfs_inode_atime(inode_item);
2384 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2385 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2386
2387 tspec = btrfs_inode_mtime(inode_item);
2388 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2389 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2390
2391 tspec = btrfs_inode_ctime(inode_item);
2392 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2393 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2394
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002395 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04002396 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
Chris Masonc3027eb2008-12-08 16:40:21 -05002397 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04002398 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik618e21d2007-07-11 10:18:17 -04002399 inode->i_rdev = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002400 rdev = btrfs_inode_rdev(leaf, inode_item);
2401
Josef Bacikaec74772008-07-24 12:12:38 -04002402 BTRFS_I(inode)->index_cnt = (u64)-1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05002403 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
Miao Xie2f7e33d2011-06-23 07:27:13 +00002404cache_acl:
Chris Mason46a53cc2009-04-27 11:47:50 -04002405 /*
2406 * try to precache a NULL acl entry for files that don't have
2407 * any xattrs or acls
2408 */
Li Zefan33345d012011-04-20 10:31:50 +08002409 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2410 btrfs_ino(inode));
Al Viro72c04902009-06-24 16:58:48 -04002411 if (!maybe_acls)
2412 cache_no_acl(inode);
Chris Mason46a53cc2009-04-27 11:47:50 -04002413
Chris Mason39279cc2007-06-12 06:35:45 -04002414 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002415
Chris Mason39279cc2007-06-12 06:35:45 -04002416 switch (inode->i_mode & S_IFMT) {
Chris Mason39279cc2007-06-12 06:35:45 -04002417 case S_IFREG:
2418 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04002419 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05002420 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04002421 inode->i_fop = &btrfs_file_operations;
2422 inode->i_op = &btrfs_file_inode_operations;
2423 break;
2424 case S_IFDIR:
2425 inode->i_fop = &btrfs_dir_file_operations;
2426 if (root == root->fs_info->tree_root)
2427 inode->i_op = &btrfs_dir_ro_inode_operations;
2428 else
2429 inode->i_op = &btrfs_dir_inode_operations;
2430 break;
2431 case S_IFLNK:
2432 inode->i_op = &btrfs_symlink_inode_operations;
2433 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04002434 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04002435 break;
Josef Bacik618e21d2007-07-11 10:18:17 -04002436 default:
Jim Owens0279b4c2009-02-04 09:29:13 -05002437 inode->i_op = &btrfs_special_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -04002438 init_special_inode(inode, inode->i_mode, rdev);
2439 break;
Chris Mason39279cc2007-06-12 06:35:45 -04002440 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02002441
2442 btrfs_update_iflags(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002443 return;
2444
2445make_bad:
Chris Mason39279cc2007-06-12 06:35:45 -04002446 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002447 make_bad_inode(inode);
2448}
2449
Chris Masond352ac62008-09-29 15:18:18 -04002450/*
2451 * given a leaf and an inode, copy the inode fields into the leaf
2452 */
Chris Masone02119d2008-09-05 16:13:11 -04002453static void fill_inode_item(struct btrfs_trans_handle *trans,
2454 struct extent_buffer *leaf,
Chris Mason5f39d392007-10-15 16:14:19 -04002455 struct btrfs_inode_item *item,
Chris Mason39279cc2007-06-12 06:35:45 -04002456 struct inode *inode)
2457{
Chris Mason5f39d392007-10-15 16:14:19 -04002458 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2459 btrfs_set_inode_gid(leaf, item, inode->i_gid);
Chris Masondbe674a2008-07-17 12:54:05 -04002460 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
Chris Mason5f39d392007-10-15 16:14:19 -04002461 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2462 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2463
2464 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2465 inode->i_atime.tv_sec);
2466 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2467 inode->i_atime.tv_nsec);
2468
2469 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2470 inode->i_mtime.tv_sec);
2471 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2472 inode->i_mtime.tv_nsec);
2473
2474 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2475 inode->i_ctime.tv_sec);
2476 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2477 inode->i_ctime.tv_nsec);
2478
Yan Zhenga76a3cd2008-10-09 11:46:29 -04002479 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
Chris Masone02119d2008-09-05 16:13:11 -04002480 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
Chris Masonc3027eb2008-12-08 16:40:21 -05002481 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
Chris Masone02119d2008-09-05 16:13:11 -04002482 btrfs_set_inode_transid(leaf, item, trans->transid);
Chris Mason5f39d392007-10-15 16:14:19 -04002483 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
Yanb98b6762008-01-08 15:54:37 -05002484 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04002485 btrfs_set_inode_block_group(leaf, item, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04002486}
2487
Chris Masond352ac62008-09-29 15:18:18 -04002488/*
2489 * copy everything in the in-memory inode into the btree.
2490 */
Chris Mason21151332011-11-10 20:39:08 -05002491static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
Chris Masond3977122009-01-05 21:25:51 -05002492 struct btrfs_root *root, struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04002493{
2494 struct btrfs_inode_item *inode_item;
2495 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04002496 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002497 int ret;
2498
2499 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08002500 if (!path)
2501 return -ENOMEM;
2502
Chris Masonb9473432009-03-13 11:00:37 -04002503 path->leave_spinning = 1;
Miao Xie16cdcec2011-04-22 18:12:22 +08002504 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2505 1);
Chris Mason39279cc2007-06-12 06:35:45 -04002506 if (ret) {
2507 if (ret > 0)
2508 ret = -ENOENT;
2509 goto failed;
2510 }
2511
Chris Masonb4ce94d2009-02-04 09:25:08 -05002512 btrfs_unlock_up_safe(path, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04002513 leaf = path->nodes[0];
2514 inode_item = btrfs_item_ptr(leaf, path->slots[0],
Miao Xie16cdcec2011-04-22 18:12:22 +08002515 struct btrfs_inode_item);
Chris Mason39279cc2007-06-12 06:35:45 -04002516
Chris Masone02119d2008-09-05 16:13:11 -04002517 fill_inode_item(trans, leaf, inode_item, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04002518 btrfs_mark_buffer_dirty(leaf);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04002519 btrfs_set_inode_last_trans(trans, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04002520 ret = 0;
2521failed:
Chris Mason39279cc2007-06-12 06:35:45 -04002522 btrfs_free_path(path);
2523 return ret;
2524}
2525
Chris Masond352ac62008-09-29 15:18:18 -04002526/*
Chris Mason21151332011-11-10 20:39:08 -05002527 * copy everything in the in-memory inode into the btree.
2528 */
2529noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2530 struct btrfs_root *root, struct inode *inode)
2531{
2532 int ret;
2533
2534 /*
2535 * If the inode is a free space inode, we can deadlock during commit
2536 * if we put it into the delayed code.
2537 *
2538 * The data relocation inode should also be directly updated
2539 * without delay
2540 */
2541 if (!btrfs_is_free_space_inode(root, inode)
2542 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2543 ret = btrfs_delayed_update_inode(trans, root, inode);
2544 if (!ret)
2545 btrfs_set_inode_last_trans(trans, inode);
2546 return ret;
2547 }
2548
2549 return btrfs_update_inode_item(trans, root, inode);
2550}
2551
2552static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2553 struct btrfs_root *root, struct inode *inode)
2554{
2555 int ret;
2556
2557 ret = btrfs_update_inode(trans, root, inode);
2558 if (ret == -ENOSPC)
2559 return btrfs_update_inode_item(trans, root, inode);
2560 return ret;
2561}
2562
2563/*
Chris Masond352ac62008-09-29 15:18:18 -04002564 * unlink helper that gets used here in inode.c and in the tree logging
2565 * recovery code. It remove a link in a directory with a given name, and
2566 * also drops the back refs in the inode to the directory
2567 */
Al Viro92986792011-03-04 17:14:37 +00002568static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2569 struct btrfs_root *root,
2570 struct inode *dir, struct inode *inode,
2571 const char *name, int name_len)
Chris Mason39279cc2007-06-12 06:35:45 -04002572{
2573 struct btrfs_path *path;
Chris Mason39279cc2007-06-12 06:35:45 -04002574 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002575 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04002576 struct btrfs_dir_item *di;
Chris Mason5f39d392007-10-15 16:14:19 -04002577 struct btrfs_key key;
Josef Bacikaec74772008-07-24 12:12:38 -04002578 u64 index;
Li Zefan33345d012011-04-20 10:31:50 +08002579 u64 ino = btrfs_ino(inode);
2580 u64 dir_ino = btrfs_ino(dir);
Chris Mason39279cc2007-06-12 06:35:45 -04002581
2582 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04002583 if (!path) {
2584 ret = -ENOMEM;
Tsutomu Itoh554233a2011-02-03 03:16:25 +00002585 goto out;
Chris Mason54aa1f42007-06-22 14:16:25 -04002586 }
2587
Chris Masonb9473432009-03-13 11:00:37 -04002588 path->leave_spinning = 1;
Li Zefan33345d012011-04-20 10:31:50 +08002589 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Chris Mason39279cc2007-06-12 06:35:45 -04002590 name, name_len, -1);
2591 if (IS_ERR(di)) {
2592 ret = PTR_ERR(di);
2593 goto err;
2594 }
2595 if (!di) {
2596 ret = -ENOENT;
2597 goto err;
2598 }
Chris Mason5f39d392007-10-15 16:14:19 -04002599 leaf = path->nodes[0];
2600 btrfs_dir_item_key_to_cpu(leaf, di, &key);
Chris Mason39279cc2007-06-12 06:35:45 -04002601 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason54aa1f42007-06-22 14:16:25 -04002602 if (ret)
2603 goto err;
David Sterbab3b4aa72011-04-21 01:20:15 +02002604 btrfs_release_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04002605
Li Zefan33345d012011-04-20 10:31:50 +08002606 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2607 dir_ino, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04002608 if (ret) {
Chris Masond3977122009-01-05 21:25:51 -05002609 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
Li Zefan33345d012011-04-20 10:31:50 +08002610 "inode %llu parent %llu\n", name_len, name,
2611 (unsigned long long)ino, (unsigned long long)dir_ino);
Josef Bacikaec74772008-07-24 12:12:38 -04002612 goto err;
2613 }
2614
Miao Xie16cdcec2011-04-22 18:12:22 +08002615 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2616 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04002617 goto err;
Chris Mason39279cc2007-06-12 06:35:45 -04002618
Chris Masone02119d2008-09-05 16:13:11 -04002619 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
Li Zefan33345d012011-04-20 10:31:50 +08002620 inode, dir_ino);
Chris Mason49eb7e42008-09-11 15:53:12 -04002621 BUG_ON(ret != 0 && ret != -ENOENT);
Chris Masone02119d2008-09-05 16:13:11 -04002622
2623 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2624 dir, index);
Chris Mason6418c962010-10-30 07:34:24 -04002625 if (ret == -ENOENT)
2626 ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002627err:
2628 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002629 if (ret)
2630 goto out;
2631
2632 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2633 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2634 btrfs_update_inode(trans, root, dir);
Chris Masone02119d2008-09-05 16:13:11 -04002635out:
Chris Mason39279cc2007-06-12 06:35:45 -04002636 return ret;
2637}
2638
Al Viro92986792011-03-04 17:14:37 +00002639int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2640 struct btrfs_root *root,
2641 struct inode *dir, struct inode *inode,
2642 const char *name, int name_len)
2643{
2644 int ret;
2645 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2646 if (!ret) {
2647 btrfs_drop_nlink(inode);
2648 ret = btrfs_update_inode(trans, root, inode);
2649 }
2650 return ret;
2651}
2652
2653
Yan, Zhenga22285a2010-05-16 10:48:46 -04002654/* helper to check if there is any shared block in the path */
2655static int check_path_shared(struct btrfs_root *root,
2656 struct btrfs_path *path)
2657{
2658 struct extent_buffer *eb;
2659 int level;
Dan Carpenter0e4dcbef2010-06-01 08:23:11 +00002660 u64 refs = 1;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002661
2662 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
Josef Bacikdedefd72011-01-24 21:43:18 +00002663 int ret;
2664
Yan, Zhenga22285a2010-05-16 10:48:46 -04002665 if (!path->nodes[level])
2666 break;
2667 eb = path->nodes[level];
2668 if (!btrfs_block_can_be_shared(root, eb))
2669 continue;
2670 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2671 &refs, NULL);
2672 if (refs > 1)
2673 return 1;
2674 }
Josef Bacikdedefd72011-01-24 21:43:18 +00002675 return 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002676}
2677
2678/*
2679 * helper to start transaction for unlink and rmdir.
2680 *
2681 * unlink and rmdir are special in btrfs, they do not always free space.
2682 * so in enospc case, we should make sure they will free space before
2683 * allowing them to use the global metadata reservation.
2684 */
2685static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2686 struct dentry *dentry)
2687{
2688 struct btrfs_trans_handle *trans;
2689 struct btrfs_root *root = BTRFS_I(dir)->root;
2690 struct btrfs_path *path;
2691 struct btrfs_inode_ref *ref;
2692 struct btrfs_dir_item *di;
2693 struct inode *inode = dentry->d_inode;
2694 u64 index;
2695 int check_link = 1;
2696 int err = -ENOSPC;
2697 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08002698 u64 ino = btrfs_ino(inode);
2699 u64 dir_ino = btrfs_ino(dir);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002700
Josef Bacike70bea52011-10-11 14:18:24 -04002701 /*
2702 * 1 for the possible orphan item
2703 * 1 for the dir item
2704 * 1 for the dir index
2705 * 1 for the inode ref
2706 * 1 for the inode ref in the tree log
2707 * 2 for the dir entries in the log
2708 * 1 for the inode
2709 */
2710 trans = btrfs_start_transaction(root, 8);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002711 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2712 return trans;
2713
Li Zefan33345d012011-04-20 10:31:50 +08002714 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Yan, Zhenga22285a2010-05-16 10:48:46 -04002715 return ERR_PTR(-ENOSPC);
2716
2717 /* check if there is someone else holds reference */
2718 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2719 return ERR_PTR(-ENOSPC);
2720
2721 if (atomic_read(&inode->i_count) > 2)
2722 return ERR_PTR(-ENOSPC);
2723
2724 if (xchg(&root->fs_info->enospc_unlink, 1))
2725 return ERR_PTR(-ENOSPC);
2726
2727 path = btrfs_alloc_path();
2728 if (!path) {
2729 root->fs_info->enospc_unlink = 0;
2730 return ERR_PTR(-ENOMEM);
2731 }
2732
Josef Bacik3880a1b2011-10-14 14:46:51 -04002733 /* 1 for the orphan item */
2734 trans = btrfs_start_transaction(root, 1);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002735 if (IS_ERR(trans)) {
2736 btrfs_free_path(path);
2737 root->fs_info->enospc_unlink = 0;
2738 return trans;
2739 }
2740
2741 path->skip_locking = 1;
2742 path->search_commit_root = 1;
2743
2744 ret = btrfs_lookup_inode(trans, root, path,
2745 &BTRFS_I(dir)->location, 0);
2746 if (ret < 0) {
2747 err = ret;
2748 goto out;
2749 }
2750 if (ret == 0) {
2751 if (check_path_shared(root, path))
2752 goto out;
2753 } else {
2754 check_link = 0;
2755 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002756 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002757
2758 ret = btrfs_lookup_inode(trans, root, path,
2759 &BTRFS_I(inode)->location, 0);
2760 if (ret < 0) {
2761 err = ret;
2762 goto out;
2763 }
2764 if (ret == 0) {
2765 if (check_path_shared(root, path))
2766 goto out;
2767 } else {
2768 check_link = 0;
2769 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002770 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002771
2772 if (ret == 0 && S_ISREG(inode->i_mode)) {
2773 ret = btrfs_lookup_file_extent(trans, root, path,
Li Zefan33345d012011-04-20 10:31:50 +08002774 ino, (u64)-1, 0);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002775 if (ret < 0) {
2776 err = ret;
2777 goto out;
2778 }
2779 BUG_ON(ret == 0);
2780 if (check_path_shared(root, path))
2781 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +02002782 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002783 }
2784
2785 if (!check_link) {
2786 err = 0;
2787 goto out;
2788 }
2789
Li Zefan33345d012011-04-20 10:31:50 +08002790 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Yan, Zhenga22285a2010-05-16 10:48:46 -04002791 dentry->d_name.name, dentry->d_name.len, 0);
2792 if (IS_ERR(di)) {
2793 err = PTR_ERR(di);
2794 goto out;
2795 }
2796 if (di) {
2797 if (check_path_shared(root, path))
2798 goto out;
2799 } else {
2800 err = 0;
2801 goto out;
2802 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002803 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002804
2805 ref = btrfs_lookup_inode_ref(trans, root, path,
2806 dentry->d_name.name, dentry->d_name.len,
Li Zefan33345d012011-04-20 10:31:50 +08002807 ino, dir_ino, 0);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002808 if (IS_ERR(ref)) {
2809 err = PTR_ERR(ref);
2810 goto out;
2811 }
2812 BUG_ON(!ref);
2813 if (check_path_shared(root, path))
2814 goto out;
2815 index = btrfs_inode_ref_index(path->nodes[0], ref);
David Sterbab3b4aa72011-04-21 01:20:15 +02002816 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002817
Miao Xie16cdcec2011-04-22 18:12:22 +08002818 /*
2819 * This is a commit root search, if we can lookup inode item and other
2820 * relative items in the commit root, it means the transaction of
2821 * dir/file creation has been committed, and the dir index item that we
2822 * delay to insert has also been inserted into the commit root. So
2823 * we needn't worry about the delayed insertion of the dir index item
2824 * here.
2825 */
Li Zefan33345d012011-04-20 10:31:50 +08002826 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
Yan, Zhenga22285a2010-05-16 10:48:46 -04002827 dentry->d_name.name, dentry->d_name.len, 0);
2828 if (IS_ERR(di)) {
2829 err = PTR_ERR(di);
2830 goto out;
2831 }
2832 BUG_ON(ret == -ENOENT);
2833 if (check_path_shared(root, path))
2834 goto out;
2835
2836 err = 0;
2837out:
2838 btrfs_free_path(path);
Josef Bacik3880a1b2011-10-14 14:46:51 -04002839 /* Migrate the orphan reservation over */
2840 if (!err)
2841 err = btrfs_block_rsv_migrate(trans->block_rsv,
2842 &root->fs_info->global_block_rsv,
Josef Bacik5a77d762011-11-01 14:32:23 -04002843 trans->bytes_reserved);
Josef Bacik3880a1b2011-10-14 14:46:51 -04002844
Yan, Zhenga22285a2010-05-16 10:48:46 -04002845 if (err) {
2846 btrfs_end_transaction(trans, root);
2847 root->fs_info->enospc_unlink = 0;
2848 return ERR_PTR(err);
2849 }
2850
2851 trans->block_rsv = &root->fs_info->global_block_rsv;
2852 return trans;
2853}
2854
2855static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2856 struct btrfs_root *root)
2857{
2858 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
Josef Bacik5a77d762011-11-01 14:32:23 -04002859 btrfs_block_rsv_release(root, trans->block_rsv,
2860 trans->bytes_reserved);
2861 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002862 BUG_ON(!root->fs_info->enospc_unlink);
2863 root->fs_info->enospc_unlink = 0;
2864 }
Josef Bacik7ad85bb2012-01-12 19:10:12 -05002865 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04002866}
2867
Chris Mason39279cc2007-06-12 06:35:45 -04002868static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2869{
Yan, Zhenga22285a2010-05-16 10:48:46 -04002870 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04002871 struct btrfs_trans_handle *trans;
Josef Bacik7b128762008-07-24 12:17:14 -04002872 struct inode *inode = dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04002873 int ret;
Chris Mason1832a6d2007-12-21 16:27:21 -05002874 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002875
Yan, Zhenga22285a2010-05-16 10:48:46 -04002876 trans = __unlink_start_trans(dir, dentry);
2877 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002878 return PTR_ERR(trans);
Chris Mason5f39d392007-10-15 16:14:19 -04002879
Chris Mason12fcfd22009-03-24 10:24:20 -04002880 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2881
Chris Masone02119d2008-09-05 16:13:11 -04002882 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2883 dentry->d_name.name, dentry->d_name.len);
Tsutomu Itohb5324022011-07-19 07:27:20 +00002884 if (ret)
2885 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002886
Yan, Zhenga22285a2010-05-16 10:48:46 -04002887 if (inode->i_nlink == 0) {
Josef Bacik7b128762008-07-24 12:17:14 -04002888 ret = btrfs_orphan_add(trans, inode);
Tsutomu Itohb5324022011-07-19 07:27:20 +00002889 if (ret)
2890 goto out;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002891 }
Josef Bacik7b128762008-07-24 12:17:14 -04002892
Tsutomu Itohb5324022011-07-19 07:27:20 +00002893out:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002894 nr = trans->blocks_used;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002895 __unlink_end_trans(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002896 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04002897 return ret;
2898}
2899
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002900int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2901 struct btrfs_root *root,
2902 struct inode *dir, u64 objectid,
2903 const char *name, int name_len)
2904{
2905 struct btrfs_path *path;
2906 struct extent_buffer *leaf;
2907 struct btrfs_dir_item *di;
2908 struct btrfs_key key;
2909 u64 index;
2910 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08002911 u64 dir_ino = btrfs_ino(dir);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002912
2913 path = btrfs_alloc_path();
2914 if (!path)
2915 return -ENOMEM;
2916
Li Zefan33345d012011-04-20 10:31:50 +08002917 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002918 name, name_len, -1);
David Sterbac7040052011-04-19 18:00:01 +02002919 BUG_ON(IS_ERR_OR_NULL(di));
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002920
2921 leaf = path->nodes[0];
2922 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2923 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2924 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2925 BUG_ON(ret);
David Sterbab3b4aa72011-04-21 01:20:15 +02002926 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002927
2928 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2929 objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08002930 dir_ino, &index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002931 if (ret < 0) {
2932 BUG_ON(ret != -ENOENT);
Li Zefan33345d012011-04-20 10:31:50 +08002933 di = btrfs_search_dir_index_item(root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002934 name, name_len);
David Sterbac7040052011-04-19 18:00:01 +02002935 BUG_ON(IS_ERR_OR_NULL(di));
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002936
2937 leaf = path->nodes[0];
2938 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02002939 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002940 index = key.offset;
2941 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002942 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002943
Miao Xie16cdcec2011-04-22 18:12:22 +08002944 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2945 BUG_ON(ret);
2946
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002947 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2948 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2949 ret = btrfs_update_inode(trans, root, dir);
2950 BUG_ON(ret);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002951
Josef Bacik71d7aed2011-06-14 14:24:32 -04002952 btrfs_free_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002953 return 0;
2954}
2955
Chris Mason39279cc2007-06-12 06:35:45 -04002956static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2957{
2958 struct inode *inode = dentry->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05002959 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002960 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04002961 struct btrfs_trans_handle *trans;
Chris Mason1832a6d2007-12-21 16:27:21 -05002962 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04002963
Chris Mason3394e162008-11-17 20:42:26 -05002964 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
Li Zefan33345d012011-04-20 10:31:50 +08002965 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
Yan134d4512007-10-25 15:49:25 -04002966 return -ENOTEMPTY;
2967
Yan, Zhenga22285a2010-05-16 10:48:46 -04002968 trans = __unlink_start_trans(dir, dentry);
2969 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002970 return PTR_ERR(trans);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05002971
Li Zefan33345d012011-04-20 10:31:50 +08002972 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002973 err = btrfs_unlink_subvol(trans, root, dir,
2974 BTRFS_I(inode)->location.objectid,
2975 dentry->d_name.name,
2976 dentry->d_name.len);
2977 goto out;
2978 }
2979
Josef Bacik7b128762008-07-24 12:17:14 -04002980 err = btrfs_orphan_add(trans, inode);
2981 if (err)
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002982 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04002983
Chris Mason39279cc2007-06-12 06:35:45 -04002984 /* now the directory is empty */
Chris Masone02119d2008-09-05 16:13:11 -04002985 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2986 dentry->d_name.name, dentry->d_name.len);
Chris Masond3977122009-01-05 21:25:51 -05002987 if (!err)
Chris Masondbe674a2008-07-17 12:54:05 -04002988 btrfs_i_size_write(inode, 0);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04002989out:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002990 nr = trans->blocks_used;
Yan, Zhenga22285a2010-05-16 10:48:46 -04002991 __unlink_end_trans(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04002992 btrfs_btree_balance_dirty(root, nr);
Chris Mason39544012007-12-12 14:38:19 -05002993
Chris Mason39279cc2007-06-12 06:35:45 -04002994 return err;
2995}
2996
Chris Mason323ac952008-10-01 19:05:46 -04002997/*
Chris Mason39279cc2007-06-12 06:35:45 -04002998 * this can truncate away extent items, csum items and directory items.
2999 * It starts at a high offset and removes keys until it can't find
Chris Masond352ac62008-09-29 15:18:18 -04003000 * any higher than new_size
Chris Mason39279cc2007-06-12 06:35:45 -04003001 *
3002 * csum items that cross the new i_size are truncated to the new size
3003 * as well.
Josef Bacik7b128762008-07-24 12:17:14 -04003004 *
3005 * min_type is the minimum key type to truncate down to. If set to 0, this
3006 * will kill all the items on this inode, including the INODE_ITEM_KEY.
Chris Mason39279cc2007-06-12 06:35:45 -04003007 */
Yan, Zheng80825102009-11-12 09:35:36 +00003008int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3009 struct btrfs_root *root,
3010 struct inode *inode,
3011 u64 new_size, u32 min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003012{
Chris Mason39279cc2007-06-12 06:35:45 -04003013 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003014 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003015 struct btrfs_file_extent_item *fi;
Yan, Zheng80825102009-11-12 09:35:36 +00003016 struct btrfs_key key;
3017 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04003018 u64 extent_start = 0;
Chris Masondb945352007-10-15 16:15:53 -04003019 u64 extent_num_bytes = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003020 u64 extent_offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003021 u64 item_end = 0;
Yan, Zheng80825102009-11-12 09:35:36 +00003022 u64 mask = root->sectorsize - 1;
3023 u32 found_type = (u8)-1;
Chris Mason39279cc2007-06-12 06:35:45 -04003024 int found_extent;
3025 int del_item;
Chris Mason85e21ba2008-01-29 15:11:36 -05003026 int pending_del_nr = 0;
3027 int pending_del_slot = 0;
Chris Mason179e29e2007-11-01 11:28:41 -04003028 int extent_type = -1;
Yan, Zheng80825102009-11-12 09:35:36 +00003029 int ret;
3030 int err = 0;
Li Zefan33345d012011-04-20 10:31:50 +08003031 u64 ino = btrfs_ino(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003032
3033 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
Chris Mason39279cc2007-06-12 06:35:45 -04003034
Mark Fasheh0eb0e192011-07-12 16:44:10 -07003035 path = btrfs_alloc_path();
3036 if (!path)
3037 return -ENOMEM;
3038 path->reada = -1;
3039
Josef Bacik0af3d002010-06-21 14:48:16 -04003040 if (root->ref_cows || root == root->fs_info->tree_root)
Zheng Yan5b21f2e2008-09-26 10:05:38 -04003041 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003042
Miao Xie16cdcec2011-04-22 18:12:22 +08003043 /*
3044 * This function is also used to drop the items in the log tree before
3045 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3046 * it is used to drop the loged items. So we shouldn't kill the delayed
3047 * items.
3048 */
3049 if (min_type == 0 && root == BTRFS_I(inode)->root)
3050 btrfs_kill_delayed_inode_items(inode);
3051
Li Zefan33345d012011-04-20 10:31:50 +08003052 key.objectid = ino;
Chris Mason39279cc2007-06-12 06:35:45 -04003053 key.offset = (u64)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04003054 key.type = (u8)-1;
3055
Chris Mason85e21ba2008-01-29 15:11:36 -05003056search_again:
Chris Masonb9473432009-03-13 11:00:37 -04003057 path->leave_spinning = 1;
Chris Mason85e21ba2008-01-29 15:11:36 -05003058 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Yan, Zheng80825102009-11-12 09:35:36 +00003059 if (ret < 0) {
3060 err = ret;
3061 goto out;
3062 }
Chris Masond3977122009-01-05 21:25:51 -05003063
Chris Mason85e21ba2008-01-29 15:11:36 -05003064 if (ret > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04003065 /* there are no items in the tree for us to truncate, we're
3066 * done
3067 */
Yan, Zheng80825102009-11-12 09:35:36 +00003068 if (path->slots[0] == 0)
3069 goto out;
Chris Mason85e21ba2008-01-29 15:11:36 -05003070 path->slots[0]--;
3071 }
3072
Chris Masond3977122009-01-05 21:25:51 -05003073 while (1) {
Chris Mason39279cc2007-06-12 06:35:45 -04003074 fi = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -04003075 leaf = path->nodes[0];
3076 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3077 found_type = btrfs_key_type(&found_key);
Chris Mason39279cc2007-06-12 06:35:45 -04003078
Li Zefan33345d012011-04-20 10:31:50 +08003079 if (found_key.objectid != ino)
Chris Mason39279cc2007-06-12 06:35:45 -04003080 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003081
Chris Mason85e21ba2008-01-29 15:11:36 -05003082 if (found_type < min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003083 break;
3084
Chris Mason5f39d392007-10-15 16:14:19 -04003085 item_end = found_key.offset;
Chris Mason39279cc2007-06-12 06:35:45 -04003086 if (found_type == BTRFS_EXTENT_DATA_KEY) {
Chris Mason5f39d392007-10-15 16:14:19 -04003087 fi = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04003088 struct btrfs_file_extent_item);
Chris Mason179e29e2007-11-01 11:28:41 -04003089 extent_type = btrfs_file_extent_type(leaf, fi);
3090 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04003091 item_end +=
Chris Masondb945352007-10-15 16:15:53 -04003092 btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason179e29e2007-11-01 11:28:41 -04003093 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason179e29e2007-11-01 11:28:41 -04003094 item_end += btrfs_file_extent_inline_len(leaf,
Chris Masonc8b97812008-10-29 14:49:59 -04003095 fi);
Chris Mason39279cc2007-06-12 06:35:45 -04003096 }
Yan008630c2007-11-07 13:31:09 -05003097 item_end--;
Chris Mason39279cc2007-06-12 06:35:45 -04003098 }
Yan, Zheng80825102009-11-12 09:35:36 +00003099 if (found_type > min_type) {
Chris Mason39279cc2007-06-12 06:35:45 -04003100 del_item = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00003101 } else {
3102 if (item_end < new_size)
3103 break;
3104 if (found_key.offset >= new_size)
3105 del_item = 1;
3106 else
3107 del_item = 0;
3108 }
Chris Mason39279cc2007-06-12 06:35:45 -04003109 found_extent = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003110 /* FIXME, shrink the extent if the ref count is only 1 */
Chris Mason179e29e2007-11-01 11:28:41 -04003111 if (found_type != BTRFS_EXTENT_DATA_KEY)
3112 goto delete;
3113
3114 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason39279cc2007-06-12 06:35:45 -04003115 u64 num_dec;
Chris Masondb945352007-10-15 16:15:53 -04003116 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
Miao Xief70a9a6b2012-01-12 19:10:12 -05003117 if (!del_item) {
Chris Masondb945352007-10-15 16:15:53 -04003118 u64 orig_num_bytes =
3119 btrfs_file_extent_num_bytes(leaf, fi);
Chris Masone02119d2008-09-05 16:13:11 -04003120 extent_num_bytes = new_size -
Chris Mason5f39d392007-10-15 16:14:19 -04003121 found_key.offset + root->sectorsize - 1;
Yanb1632b12008-01-30 11:54:04 -05003122 extent_num_bytes = extent_num_bytes &
3123 ~((u64)root->sectorsize - 1);
Chris Masondb945352007-10-15 16:15:53 -04003124 btrfs_set_file_extent_num_bytes(leaf, fi,
3125 extent_num_bytes);
3126 num_dec = (orig_num_bytes -
Chris Mason90692182008-02-08 13:49:28 -05003127 extent_num_bytes);
Chris Masone02119d2008-09-05 16:13:11 -04003128 if (root->ref_cows && extent_start != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003129 inode_sub_bytes(inode, num_dec);
Chris Mason5f39d392007-10-15 16:14:19 -04003130 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04003131 } else {
Chris Masondb945352007-10-15 16:15:53 -04003132 extent_num_bytes =
3133 btrfs_file_extent_disk_num_bytes(leaf,
3134 fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003135 extent_offset = found_key.offset -
3136 btrfs_file_extent_offset(leaf, fi);
3137
Chris Mason39279cc2007-06-12 06:35:45 -04003138 /* FIXME blocksize != 4096 */
Chris Mason90692182008-02-08 13:49:28 -05003139 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason39279cc2007-06-12 06:35:45 -04003140 if (extent_start != 0) {
3141 found_extent = 1;
Chris Masone02119d2008-09-05 16:13:11 -04003142 if (root->ref_cows)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003143 inode_sub_bytes(inode, num_dec);
Chris Mason39279cc2007-06-12 06:35:45 -04003144 }
3145 }
Chris Mason90692182008-02-08 13:49:28 -05003146 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -04003147 /*
3148 * we can't truncate inline items that have had
3149 * special encodings
3150 */
3151 if (!del_item &&
3152 btrfs_file_extent_compression(leaf, fi) == 0 &&
3153 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3154 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04003155 u32 size = new_size - found_key.offset;
3156
3157 if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003158 inode_sub_bytes(inode, item_end + 1 -
3159 new_size);
Chris Masone02119d2008-09-05 16:13:11 -04003160 }
3161 size =
3162 btrfs_file_extent_calc_inline_size(size);
Chris Mason90692182008-02-08 13:49:28 -05003163 ret = btrfs_truncate_item(trans, root, path,
Chris Masone02119d2008-09-05 16:13:11 -04003164 size, 1);
Chris Masone02119d2008-09-05 16:13:11 -04003165 } else if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003166 inode_sub_bytes(inode, item_end + 1 -
3167 found_key.offset);
Chris Mason90692182008-02-08 13:49:28 -05003168 }
Chris Mason39279cc2007-06-12 06:35:45 -04003169 }
Chris Mason179e29e2007-11-01 11:28:41 -04003170delete:
Chris Mason39279cc2007-06-12 06:35:45 -04003171 if (del_item) {
Chris Mason85e21ba2008-01-29 15:11:36 -05003172 if (!pending_del_nr) {
3173 /* no pending yet, add ourselves */
3174 pending_del_slot = path->slots[0];
3175 pending_del_nr = 1;
3176 } else if (pending_del_nr &&
3177 path->slots[0] + 1 == pending_del_slot) {
3178 /* hop on the pending chunk */
3179 pending_del_nr++;
3180 pending_del_slot = path->slots[0];
3181 } else {
Chris Masond3977122009-01-05 21:25:51 -05003182 BUG();
Chris Mason85e21ba2008-01-29 15:11:36 -05003183 }
Chris Mason39279cc2007-06-12 06:35:45 -04003184 } else {
3185 break;
3186 }
Josef Bacik0af3d002010-06-21 14:48:16 -04003187 if (found_extent && (root->ref_cows ||
3188 root == root->fs_info->tree_root)) {
Chris Masonb9473432009-03-13 11:00:37 -04003189 btrfs_set_path_blocking(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003190 ret = btrfs_free_extent(trans, root, extent_start,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003191 extent_num_bytes, 0,
3192 btrfs_header_owner(leaf),
Arne Jansen66d7e7f2011-09-12 15:26:38 +02003193 ino, extent_offset, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04003194 BUG_ON(ret);
3195 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003196
Yan, Zheng80825102009-11-12 09:35:36 +00003197 if (found_type == BTRFS_INODE_ITEM_KEY)
3198 break;
3199
3200 if (path->slots[0] == 0 ||
3201 path->slots[0] != pending_del_slot) {
Li Zefan82d59022011-04-20 10:33:24 +08003202 if (root->ref_cows &&
3203 BTRFS_I(inode)->location.objectid !=
3204 BTRFS_FREE_INO_OBJECTID) {
Yan, Zheng80825102009-11-12 09:35:36 +00003205 err = -EAGAIN;
3206 goto out;
3207 }
3208 if (pending_del_nr) {
3209 ret = btrfs_del_items(trans, root, path,
3210 pending_del_slot,
3211 pending_del_nr);
3212 BUG_ON(ret);
3213 pending_del_nr = 0;
3214 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003215 btrfs_release_path(path);
Chris Mason85e21ba2008-01-29 15:11:36 -05003216 goto search_again;
Yan, Zheng80825102009-11-12 09:35:36 +00003217 } else {
3218 path->slots[0]--;
Chris Mason85e21ba2008-01-29 15:11:36 -05003219 }
Chris Mason39279cc2007-06-12 06:35:45 -04003220 }
Yan, Zheng80825102009-11-12 09:35:36 +00003221out:
Chris Mason85e21ba2008-01-29 15:11:36 -05003222 if (pending_del_nr) {
3223 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3224 pending_del_nr);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003225 BUG_ON(ret);
Chris Mason85e21ba2008-01-29 15:11:36 -05003226 }
Chris Mason39279cc2007-06-12 06:35:45 -04003227 btrfs_free_path(path);
Yan, Zheng80825102009-11-12 09:35:36 +00003228 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003229}
3230
Chris Masona52d9a82007-08-27 16:49:44 -04003231/*
3232 * taken from block_truncate_page, but does cow as it zeros out
3233 * any bytes left in the last page in the file.
3234 */
3235static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3236{
3237 struct inode *inode = mapping->host;
Chris Masondb945352007-10-15 16:15:53 -04003238 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003239 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3240 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003241 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003242 char *kaddr;
Chris Masondb945352007-10-15 16:15:53 -04003243 u32 blocksize = root->sectorsize;
Chris Masona52d9a82007-08-27 16:49:44 -04003244 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3245 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3246 struct page *page;
Josef Bacik3b16a4e2011-09-21 15:05:58 -04003247 gfp_t mask = btrfs_alloc_write_mask(mapping);
Chris Masona52d9a82007-08-27 16:49:44 -04003248 int ret = 0;
3249 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04003250 u64 page_end;
Chris Masona52d9a82007-08-27 16:49:44 -04003251
3252 if ((offset & (blocksize - 1)) == 0)
3253 goto out;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003254 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003255 if (ret)
3256 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04003257
3258 ret = -ENOMEM;
Chris Mason211c17f2008-05-15 09:13:45 -04003259again:
Josef Bacik3b16a4e2011-09-21 15:05:58 -04003260 page = find_or_create_page(mapping, index, mask);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003261 if (!page) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003262 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Masona52d9a82007-08-27 16:49:44 -04003263 goto out;
Josef Bacik5d5e1032009-10-13 16:46:49 -04003264 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04003265
3266 page_start = page_offset(page);
3267 page_end = page_start + PAGE_CACHE_SIZE - 1;
3268
Chris Masona52d9a82007-08-27 16:49:44 -04003269 if (!PageUptodate(page)) {
3270 ret = btrfs_readpage(NULL, page);
3271 lock_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04003272 if (page->mapping != mapping) {
3273 unlock_page(page);
3274 page_cache_release(page);
3275 goto again;
3276 }
Chris Masona52d9a82007-08-27 16:49:44 -04003277 if (!PageUptodate(page)) {
3278 ret = -EIO;
Chris Mason89642222008-07-24 09:41:53 -04003279 goto out_unlock;
Chris Masona52d9a82007-08-27 16:49:44 -04003280 }
3281 }
Chris Mason211c17f2008-05-15 09:13:45 -04003282 wait_on_page_writeback(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003283
Josef Bacik2ac55d42010-02-03 19:33:23 +00003284 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3285 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003286 set_page_extent_mapped(page);
3287
3288 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3289 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00003290 unlock_extent_cached(io_tree, page_start, page_end,
3291 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003292 unlock_page(page);
3293 page_cache_release(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04003294 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003295 btrfs_put_ordered_extent(ordered);
3296 goto again;
3297 }
3298
Josef Bacik2ac55d42010-02-03 19:33:23 +00003299 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Josef Bacik5d5e1032009-10-13 16:46:49 -04003300 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
Josef Bacik2ac55d42010-02-03 19:33:23 +00003301 0, 0, &cached_state, GFP_NOFS);
Josef Bacik5d5e1032009-10-13 16:46:49 -04003302
Josef Bacik2ac55d42010-02-03 19:33:23 +00003303 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3304 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003305 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00003306 unlock_extent_cached(io_tree, page_start, page_end,
3307 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003308 goto out_unlock;
3309 }
3310
Chris Masone6dcd2d2008-07-17 12:53:50 -04003311 ret = 0;
3312 if (offset != PAGE_CACHE_SIZE) {
3313 kaddr = kmap(page);
3314 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3315 flush_dcache_page(page);
3316 kunmap(page);
3317 }
Chris Mason247e7432008-07-17 12:53:51 -04003318 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003319 set_page_dirty(page);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003320 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3321 GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04003322
Chris Mason89642222008-07-24 09:41:53 -04003323out_unlock:
Josef Bacik5d5e1032009-10-13 16:46:49 -04003324 if (ret)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003325 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason39279cc2007-06-12 06:35:45 -04003326 unlock_page(page);
3327 page_cache_release(page);
3328out:
3329 return ret;
3330}
3331
Josef Bacik695a0d02011-03-04 15:46:53 -05003332/*
3333 * This function puts in dummy file extents for the area we're creating a hole
3334 * for. So if we are truncating this file to a larger size we need to insert
3335 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3336 * the range between oldsize and size
3337 */
Josef Bacika41ad392011-01-31 15:30:16 -05003338int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
Yan Zheng9036c102008-10-30 14:19:41 -04003339{
3340 struct btrfs_trans_handle *trans;
3341 struct btrfs_root *root = BTRFS_I(inode)->root;
3342 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003343 struct extent_map *em = NULL;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003344 struct extent_state *cached_state = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04003345 u64 mask = root->sectorsize - 1;
Josef Bacika41ad392011-01-31 15:30:16 -05003346 u64 hole_start = (oldsize + mask) & ~mask;
Yan Zheng9036c102008-10-30 14:19:41 -04003347 u64 block_end = (size + mask) & ~mask;
3348 u64 last_byte;
3349 u64 cur_offset;
3350 u64 hole_size;
Josef Bacik9ed74f22009-09-11 16:12:44 -04003351 int err = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04003352
3353 if (size <= hole_start)
3354 return 0;
3355
Yan Zheng9036c102008-10-30 14:19:41 -04003356 while (1) {
3357 struct btrfs_ordered_extent *ordered;
3358 btrfs_wait_ordered_range(inode, hole_start,
3359 block_end - hole_start);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003360 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3361 &cached_state, GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003362 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3363 if (!ordered)
3364 break;
Josef Bacik2ac55d42010-02-03 19:33:23 +00003365 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3366 &cached_state, GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003367 btrfs_put_ordered_extent(ordered);
3368 }
3369
Yan Zheng9036c102008-10-30 14:19:41 -04003370 cur_offset = hole_start;
3371 while (1) {
3372 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3373 block_end - cur_offset, 0);
David Sterbac7040052011-04-19 18:00:01 +02003374 BUG_ON(IS_ERR_OR_NULL(em));
Yan Zheng9036c102008-10-30 14:19:41 -04003375 last_byte = min(extent_map_end(em), block_end);
3376 last_byte = (last_byte + mask) & ~mask;
Yan, Zheng80825102009-11-12 09:35:36 +00003377 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
Chris Mason771ed682008-11-06 22:02:51 -05003378 u64 hint_byte = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04003379 hole_size = last_byte - cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00003380
Miao Xie36423202011-12-14 20:12:02 -05003381 trans = btrfs_start_transaction(root, 3);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003382 if (IS_ERR(trans)) {
3383 err = PTR_ERR(trans);
Chris Mason771ed682008-11-06 22:02:51 -05003384 break;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003385 }
Yan, Zheng80825102009-11-12 09:35:36 +00003386
3387 err = btrfs_drop_extents(trans, inode, cur_offset,
3388 cur_offset + hole_size,
3389 &hint_byte, 1);
Miao Xie5b397372011-09-11 10:52:24 -04003390 if (err) {
Miao Xie36423202011-12-14 20:12:02 -05003391 btrfs_update_inode(trans, root, inode);
Miao Xie5b397372011-09-11 10:52:24 -04003392 btrfs_end_transaction(trans, root);
Josef Bacik3893e332011-01-31 16:03:11 -05003393 break;
Miao Xie5b397372011-09-11 10:52:24 -04003394 }
Josef Bacik9ed74f22009-09-11 16:12:44 -04003395
Yan Zheng9036c102008-10-30 14:19:41 -04003396 err = btrfs_insert_file_extent(trans, root,
Li Zefan33345d012011-04-20 10:31:50 +08003397 btrfs_ino(inode), cur_offset, 0,
Yan Zheng9036c102008-10-30 14:19:41 -04003398 0, hole_size, 0, hole_size,
3399 0, 0, 0);
Miao Xie5b397372011-09-11 10:52:24 -04003400 if (err) {
Miao Xie36423202011-12-14 20:12:02 -05003401 btrfs_update_inode(trans, root, inode);
Miao Xie5b397372011-09-11 10:52:24 -04003402 btrfs_end_transaction(trans, root);
Josef Bacik3893e332011-01-31 16:03:11 -05003403 break;
Miao Xie5b397372011-09-11 10:52:24 -04003404 }
Yan, Zheng80825102009-11-12 09:35:36 +00003405
Yan Zheng9036c102008-10-30 14:19:41 -04003406 btrfs_drop_extent_cache(inode, hole_start,
3407 last_byte - 1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003408
Miao Xie36423202011-12-14 20:12:02 -05003409 btrfs_update_inode(trans, root, inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003410 btrfs_end_transaction(trans, root);
Yan Zheng9036c102008-10-30 14:19:41 -04003411 }
3412 free_extent_map(em);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003413 em = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04003414 cur_offset = last_byte;
Yan, Zheng80825102009-11-12 09:35:36 +00003415 if (cur_offset >= block_end)
Yan Zheng9036c102008-10-30 14:19:41 -04003416 break;
3417 }
3418
Yan, Zhenga22285a2010-05-16 10:48:46 -04003419 free_extent_map(em);
Josef Bacik2ac55d42010-02-03 19:33:23 +00003420 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3421 GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04003422 return err;
3423}
3424
Josef Bacika41ad392011-01-31 15:30:16 -05003425static int btrfs_setsize(struct inode *inode, loff_t newsize)
Yan, Zheng80825102009-11-12 09:35:36 +00003426{
Miao Xief4a2f4c2011-12-14 20:12:01 -05003427 struct btrfs_root *root = BTRFS_I(inode)->root;
3428 struct btrfs_trans_handle *trans;
Josef Bacika41ad392011-01-31 15:30:16 -05003429 loff_t oldsize = i_size_read(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003430 int ret;
3431
Josef Bacika41ad392011-01-31 15:30:16 -05003432 if (newsize == oldsize)
Yan, Zheng80825102009-11-12 09:35:36 +00003433 return 0;
3434
Josef Bacika41ad392011-01-31 15:30:16 -05003435 if (newsize > oldsize) {
Josef Bacika41ad392011-01-31 15:30:16 -05003436 truncate_pagecache(inode, oldsize, newsize);
3437 ret = btrfs_cont_expand(inode, oldsize, newsize);
Miao Xief4a2f4c2011-12-14 20:12:01 -05003438 if (ret)
Yan, Zheng80825102009-11-12 09:35:36 +00003439 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00003440
Miao Xief4a2f4c2011-12-14 20:12:01 -05003441 trans = btrfs_start_transaction(root, 1);
3442 if (IS_ERR(trans))
3443 return PTR_ERR(trans);
3444
3445 i_size_write(inode, newsize);
3446 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3447 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik7ad85bb2012-01-12 19:10:12 -05003448 btrfs_end_transaction(trans, root);
Josef Bacika41ad392011-01-31 15:30:16 -05003449 } else {
Yan, Zheng80825102009-11-12 09:35:36 +00003450
Josef Bacika41ad392011-01-31 15:30:16 -05003451 /*
3452 * We're truncating a file that used to have good data down to
3453 * zero. Make sure it gets into the ordered flush list so that
3454 * any new writes get down to disk quickly.
3455 */
3456 if (newsize == 0)
3457 BTRFS_I(inode)->ordered_data_close = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00003458
Josef Bacika41ad392011-01-31 15:30:16 -05003459 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3460 truncate_setsize(inode, newsize);
3461 ret = btrfs_truncate(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003462 }
3463
Josef Bacika41ad392011-01-31 15:30:16 -05003464 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00003465}
3466
Chris Mason39279cc2007-06-12 06:35:45 -04003467static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3468{
3469 struct inode *inode = dentry->d_inode;
Li Zefanb83cc962010-12-20 16:04:08 +08003470 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003471 int err;
3472
Li Zefanb83cc962010-12-20 16:04:08 +08003473 if (btrfs_root_readonly(root))
3474 return -EROFS;
3475
Chris Mason39279cc2007-06-12 06:35:45 -04003476 err = inode_change_ok(inode, attr);
3477 if (err)
3478 return err;
3479
Chris Mason5a3f23d2009-03-31 13:27:11 -04003480 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Josef Bacika41ad392011-01-31 15:30:16 -05003481 err = btrfs_setsize(inode, attr->ia_size);
Yan, Zheng80825102009-11-12 09:35:36 +00003482 if (err)
3483 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003484 }
Yan Zheng9036c102008-10-30 14:19:41 -04003485
Christoph Hellwig10257742010-06-04 11:30:02 +02003486 if (attr->ia_valid) {
3487 setattr_copy(inode, attr);
Josef Bacik22c44fe2011-11-30 10:45:38 -05003488 err = btrfs_dirty_inode(inode);
Josef Bacik33268ea2008-07-24 12:16:36 -04003489
Josef Bacik22c44fe2011-11-30 10:45:38 -05003490 if (!err && attr->ia_valid & ATTR_MODE)
Christoph Hellwig10257742010-06-04 11:30:02 +02003491 err = btrfs_acl_chmod(inode);
3492 }
3493
Chris Mason39279cc2007-06-12 06:35:45 -04003494 return err;
3495}
Chris Mason61295eb2008-01-14 16:24:38 -05003496
Al Virobd555972010-06-07 11:35:40 -04003497void btrfs_evict_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003498{
3499 struct btrfs_trans_handle *trans;
3500 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik726c35f2011-09-26 15:46:06 -04003501 struct btrfs_block_rsv *rsv, *global_rsv;
Josef Bacik07127182011-08-19 10:29:59 -04003502 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003503 unsigned long nr;
Chris Mason39279cc2007-06-12 06:35:45 -04003504 int ret;
3505
liubo1abe9b82011-03-24 11:18:59 +00003506 trace_btrfs_inode_evict(inode);
3507
Chris Mason39279cc2007-06-12 06:35:45 -04003508 truncate_inode_pages(&inode->i_data, 0);
Josef Bacik0af3d002010-06-21 14:48:16 -04003509 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
Chris Mason2cf85722011-07-26 15:35:09 -04003510 btrfs_is_free_space_inode(root, inode)))
Al Virobd555972010-06-07 11:35:40 -04003511 goto no_delete;
3512
Chris Mason39279cc2007-06-12 06:35:45 -04003513 if (is_bad_inode(inode)) {
Josef Bacik7b128762008-07-24 12:17:14 -04003514 btrfs_orphan_del(NULL, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003515 goto no_delete;
3516 }
Al Virobd555972010-06-07 11:35:40 -04003517 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
Chris Mason4a096752008-07-21 10:29:44 -04003518 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Chris Mason5f39d392007-10-15 16:14:19 -04003519
Yan, Zhengc71bf092009-11-12 09:34:40 +00003520 if (root->fs_info->log_root_recovering) {
3521 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3522 goto no_delete;
3523 }
3524
Yan, Zheng76dda932009-09-21 16:00:26 -04003525 if (inode->i_nlink > 0) {
3526 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3527 goto no_delete;
3528 }
3529
Josef Bacik4289a662011-08-05 13:22:24 -04003530 rsv = btrfs_alloc_block_rsv(root);
3531 if (!rsv) {
3532 btrfs_orphan_del(NULL, inode);
3533 goto no_delete;
3534 }
Josef Bacik4a338542011-08-29 11:01:31 -04003535 rsv->size = min_size;
Josef Bacik726c35f2011-09-26 15:46:06 -04003536 global_rsv = &root->fs_info->global_block_rsv;
Josef Bacik4289a662011-08-05 13:22:24 -04003537
Chris Masondbe674a2008-07-17 12:54:05 -04003538 btrfs_i_size_write(inode, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04003539
Josef Bacik4289a662011-08-05 13:22:24 -04003540 /*
3541 * This is a bit simpler than btrfs_truncate since
3542 *
3543 * 1) We've already reserved our space for our orphan item in the
3544 * unlink.
3545 * 2) We're going to delete the inode item, so we don't need to update
3546 * it at all.
3547 *
3548 * So we just need to reserve some slack space in case we add bytes when
3549 * doing the truncate.
3550 */
Yan, Zheng80825102009-11-12 09:35:36 +00003551 while (1) {
Miao Xieaa38a712011-11-18 17:43:00 +08003552 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
Yan, Zheng80825102009-11-12 09:35:36 +00003553
Josef Bacik726c35f2011-09-26 15:46:06 -04003554 /*
3555 * Try and steal from the global reserve since we will
3556 * likely not use this space anyway, we want to try as
3557 * hard as possible to get this to work.
3558 */
3559 if (ret)
3560 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3561
Yan, Zhengd68fc572010-05-16 10:49:58 -04003562 if (ret) {
Josef Bacik4289a662011-08-05 13:22:24 -04003563 printk(KERN_WARNING "Could not get space for a "
Josef Bacik482e6dc2011-08-19 10:31:56 -04003564 "delete, will truncate on mount %d\n", ret);
Josef Bacik4289a662011-08-05 13:22:24 -04003565 btrfs_orphan_del(NULL, inode);
3566 btrfs_free_block_rsv(root, rsv);
3567 goto no_delete;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003568 }
3569
Josef Bacik4289a662011-08-05 13:22:24 -04003570 trans = btrfs_start_transaction(root, 0);
3571 if (IS_ERR(trans)) {
3572 btrfs_orphan_del(NULL, inode);
3573 btrfs_free_block_rsv(root, rsv);
3574 goto no_delete;
3575 }
3576
3577 trans->block_rsv = rsv;
3578
Yan, Zhengd68fc572010-05-16 10:49:58 -04003579 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003580 if (ret != -EAGAIN)
3581 break;
3582
3583 nr = trans->blocks_used;
3584 btrfs_end_transaction(trans, root);
3585 trans = NULL;
3586 btrfs_btree_balance_dirty(root, nr);
Josef Bacik7b128762008-07-24 12:17:14 -04003587 }
3588
Josef Bacik4289a662011-08-05 13:22:24 -04003589 btrfs_free_block_rsv(root, rsv);
3590
Yan, Zheng80825102009-11-12 09:35:36 +00003591 if (ret == 0) {
Josef Bacik4289a662011-08-05 13:22:24 -04003592 trans->block_rsv = root->orphan_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00003593 ret = btrfs_orphan_del(trans, inode);
3594 BUG_ON(ret);
3595 }
Chris Mason85e21ba2008-01-29 15:11:36 -05003596
Josef Bacik4289a662011-08-05 13:22:24 -04003597 trans->block_rsv = &root->fs_info->trans_block_rsv;
Li Zefan581bb052011-04-20 10:06:11 +08003598 if (!(root == root->fs_info->tree_root ||
3599 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
Li Zefan33345d012011-04-20 10:31:50 +08003600 btrfs_return_ino(root, btrfs_ino(inode));
Li Zefan581bb052011-04-20 10:06:11 +08003601
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003602 nr = trans->blocks_used;
Chris Mason54aa1f42007-06-22 14:16:25 -04003603 btrfs_end_transaction(trans, root);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04003604 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04003605no_delete:
Al Virobd555972010-06-07 11:35:40 -04003606 end_writeback(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003607 return;
Chris Mason39279cc2007-06-12 06:35:45 -04003608}
3609
3610/*
3611 * this returns the key found in the dir entry in the location pointer.
3612 * If no dir entries were found, location->objectid is 0.
3613 */
3614static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3615 struct btrfs_key *location)
3616{
3617 const char *name = dentry->d_name.name;
3618 int namelen = dentry->d_name.len;
3619 struct btrfs_dir_item *di;
3620 struct btrfs_path *path;
3621 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan0d9f7f32007-10-25 15:48:28 -04003622 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003623
3624 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07003625 if (!path)
3626 return -ENOMEM;
Chris Mason39544012007-12-12 14:38:19 -05003627
Li Zefan33345d012011-04-20 10:31:50 +08003628 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
Chris Mason39279cc2007-06-12 06:35:45 -04003629 namelen, 0);
Yan0d9f7f32007-10-25 15:48:28 -04003630 if (IS_ERR(di))
3631 ret = PTR_ERR(di);
Chris Masond3977122009-01-05 21:25:51 -05003632
David Sterbac7040052011-04-19 18:00:01 +02003633 if (IS_ERR_OR_NULL(di))
Chris Mason39544012007-12-12 14:38:19 -05003634 goto out_err;
Chris Masond3977122009-01-05 21:25:51 -05003635
Chris Mason5f39d392007-10-15 16:14:19 -04003636 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
Chris Mason39279cc2007-06-12 06:35:45 -04003637out:
Chris Mason39279cc2007-06-12 06:35:45 -04003638 btrfs_free_path(path);
3639 return ret;
Chris Mason39544012007-12-12 14:38:19 -05003640out_err:
3641 location->objectid = 0;
3642 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04003643}
3644
3645/*
3646 * when we hit a tree root in a directory, the btrfs part of the inode
3647 * needs to be changed to reflect the root directory of the tree root. This
3648 * is kind of like crossing a mount point.
3649 */
3650static int fixup_tree_root_location(struct btrfs_root *root,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003651 struct inode *dir,
3652 struct dentry *dentry,
3653 struct btrfs_key *location,
3654 struct btrfs_root **sub_root)
Chris Mason39279cc2007-06-12 06:35:45 -04003655{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003656 struct btrfs_path *path;
3657 struct btrfs_root *new_root;
3658 struct btrfs_root_ref *ref;
3659 struct extent_buffer *leaf;
3660 int ret;
3661 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003662
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003663 path = btrfs_alloc_path();
3664 if (!path) {
3665 err = -ENOMEM;
3666 goto out;
3667 }
Chris Mason39279cc2007-06-12 06:35:45 -04003668
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003669 err = -ENOENT;
3670 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3671 BTRFS_I(dir)->root->root_key.objectid,
3672 location->objectid);
3673 if (ret) {
3674 if (ret < 0)
3675 err = ret;
3676 goto out;
3677 }
Chris Mason39279cc2007-06-12 06:35:45 -04003678
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003679 leaf = path->nodes[0];
3680 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
Li Zefan33345d012011-04-20 10:31:50 +08003681 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003682 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3683 goto out;
3684
3685 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3686 (unsigned long)(ref + 1),
3687 dentry->d_name.len);
3688 if (ret)
3689 goto out;
3690
David Sterbab3b4aa72011-04-21 01:20:15 +02003691 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003692
3693 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3694 if (IS_ERR(new_root)) {
3695 err = PTR_ERR(new_root);
3696 goto out;
3697 }
3698
3699 if (btrfs_root_refs(&new_root->root_item) == 0) {
3700 err = -ENOENT;
3701 goto out;
3702 }
3703
3704 *sub_root = new_root;
3705 location->objectid = btrfs_root_dirid(&new_root->root_item);
3706 location->type = BTRFS_INODE_ITEM_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04003707 location->offset = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003708 err = 0;
3709out:
3710 btrfs_free_path(path);
3711 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04003712}
3713
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003714static void inode_tree_add(struct inode *inode)
3715{
3716 struct btrfs_root *root = BTRFS_I(inode)->root;
3717 struct btrfs_inode *entry;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003718 struct rb_node **p;
3719 struct rb_node *parent;
Li Zefan33345d012011-04-20 10:31:50 +08003720 u64 ino = btrfs_ino(inode);
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003721again:
3722 p = &root->inode_tree.rb_node;
3723 parent = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003724
Al Viro1d3382cb2010-10-23 15:19:20 -04003725 if (inode_unhashed(inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04003726 return;
3727
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003728 spin_lock(&root->inode_lock);
3729 while (*p) {
3730 parent = *p;
3731 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3732
Li Zefan33345d012011-04-20 10:31:50 +08003733 if (ino < btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003734 p = &parent->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08003735 else if (ino > btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003736 p = &parent->rb_right;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003737 else {
3738 WARN_ON(!(entry->vfs_inode.i_state &
Al Viroa4ffdde2010-06-02 17:38:30 -04003739 (I_WILL_FREE | I_FREEING)));
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003740 rb_erase(parent, &root->inode_tree);
3741 RB_CLEAR_NODE(parent);
3742 spin_unlock(&root->inode_lock);
3743 goto again;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003744 }
3745 }
3746 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3747 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3748 spin_unlock(&root->inode_lock);
3749}
3750
3751static void inode_tree_del(struct inode *inode)
3752{
3753 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng76dda932009-09-21 16:00:26 -04003754 int empty = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003755
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003756 spin_lock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003757 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003758 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003759 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Yan, Zheng76dda932009-09-21 16:00:26 -04003760 empty = RB_EMPTY_ROOT(&root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003761 }
From: Nick Piggin03e860b2009-08-21 10:09:44 +02003762 spin_unlock(&root->inode_lock);
Yan, Zheng76dda932009-09-21 16:00:26 -04003763
Josef Bacik0af3d002010-06-21 14:48:16 -04003764 /*
3765 * Free space cache has inodes in the tree root, but the tree root has a
3766 * root_refs of 0, so this could end up dropping the tree root as a
3767 * snapshot, so we need the extra !root->fs_info->tree_root check to
3768 * make sure we don't drop it.
3769 */
3770 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3771 root != root->fs_info->tree_root) {
Yan, Zheng76dda932009-09-21 16:00:26 -04003772 synchronize_srcu(&root->fs_info->subvol_srcu);
3773 spin_lock(&root->inode_lock);
3774 empty = RB_EMPTY_ROOT(&root->inode_tree);
3775 spin_unlock(&root->inode_lock);
3776 if (empty)
3777 btrfs_add_dead_root(root);
3778 }
3779}
3780
3781int btrfs_invalidate_inodes(struct btrfs_root *root)
3782{
3783 struct rb_node *node;
3784 struct rb_node *prev;
3785 struct btrfs_inode *entry;
3786 struct inode *inode;
3787 u64 objectid = 0;
3788
3789 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3790
3791 spin_lock(&root->inode_lock);
3792again:
3793 node = root->inode_tree.rb_node;
3794 prev = NULL;
3795 while (node) {
3796 prev = node;
3797 entry = rb_entry(node, struct btrfs_inode, rb_node);
3798
Li Zefan33345d012011-04-20 10:31:50 +08003799 if (objectid < btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04003800 node = node->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08003801 else if (objectid > btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04003802 node = node->rb_right;
3803 else
3804 break;
3805 }
3806 if (!node) {
3807 while (prev) {
3808 entry = rb_entry(prev, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08003809 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
Yan, Zheng76dda932009-09-21 16:00:26 -04003810 node = prev;
3811 break;
3812 }
3813 prev = rb_next(prev);
3814 }
3815 }
3816 while (node) {
3817 entry = rb_entry(node, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08003818 objectid = btrfs_ino(&entry->vfs_inode) + 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04003819 inode = igrab(&entry->vfs_inode);
3820 if (inode) {
3821 spin_unlock(&root->inode_lock);
3822 if (atomic_read(&inode->i_count) > 1)
3823 d_prune_aliases(inode);
3824 /*
Al Viro45321ac2010-06-07 13:43:19 -04003825 * btrfs_drop_inode will have it removed from
Yan, Zheng76dda932009-09-21 16:00:26 -04003826 * the inode cache when its usage count
3827 * hits zero.
3828 */
3829 iput(inode);
3830 cond_resched();
3831 spin_lock(&root->inode_lock);
3832 goto again;
3833 }
3834
3835 if (cond_resched_lock(&root->inode_lock))
3836 goto again;
3837
3838 node = rb_next(node);
3839 }
3840 spin_unlock(&root->inode_lock);
3841 return 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003842}
3843
Chris Masone02119d2008-09-05 16:13:11 -04003844static int btrfs_init_locked_inode(struct inode *inode, void *p)
3845{
3846 struct btrfs_iget_args *args = p;
3847 inode->i_ino = args->ino;
Chris Masone02119d2008-09-05 16:13:11 -04003848 BTRFS_I(inode)->root = args->root;
Josef Bacik6a632092009-02-20 11:00:09 -05003849 btrfs_set_inode_space_info(args->root, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003850 return 0;
3851}
3852
3853static int btrfs_find_actor(struct inode *inode, void *opaque)
3854{
3855 struct btrfs_iget_args *args = opaque;
Li Zefan33345d012011-04-20 10:31:50 +08003856 return args->ino == btrfs_ino(inode) &&
Chris Masond3977122009-01-05 21:25:51 -05003857 args->root == BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003858}
3859
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003860static struct inode *btrfs_iget_locked(struct super_block *s,
3861 u64 objectid,
3862 struct btrfs_root *root)
Chris Mason39279cc2007-06-12 06:35:45 -04003863{
3864 struct inode *inode;
3865 struct btrfs_iget_args args;
3866 args.ino = objectid;
3867 args.root = root;
3868
3869 inode = iget5_locked(s, objectid, btrfs_find_actor,
3870 btrfs_init_locked_inode,
3871 (void *)&args);
3872 return inode;
3873}
3874
Balaji Rao1a54ef82008-07-21 02:01:04 +05303875/* Get an inode object given its location and corresponding root.
3876 * Returns in *is_new if the inode was read from disk
3877 */
3878struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Josef Bacik73f73412009-12-04 17:38:27 +00003879 struct btrfs_root *root, int *new)
Balaji Rao1a54ef82008-07-21 02:01:04 +05303880{
3881 struct inode *inode;
3882
3883 inode = btrfs_iget_locked(s, location->objectid, root);
3884 if (!inode)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003885 return ERR_PTR(-ENOMEM);
Balaji Rao1a54ef82008-07-21 02:01:04 +05303886
3887 if (inode->i_state & I_NEW) {
3888 BTRFS_I(inode)->root = root;
3889 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3890 btrfs_read_locked_inode(inode);
Mark Fasheh1748f842011-07-12 11:25:31 -07003891 if (!is_bad_inode(inode)) {
3892 inode_tree_add(inode);
3893 unlock_new_inode(inode);
3894 if (new)
3895 *new = 1;
3896 } else {
Sergei Trofimoviche0b6d652011-09-11 10:52:24 -04003897 unlock_new_inode(inode);
3898 iput(inode);
3899 inode = ERR_PTR(-ESTALE);
Mark Fasheh1748f842011-07-12 11:25:31 -07003900 }
3901 }
3902
Balaji Rao1a54ef82008-07-21 02:01:04 +05303903 return inode;
3904}
3905
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003906static struct inode *new_simple_dir(struct super_block *s,
3907 struct btrfs_key *key,
3908 struct btrfs_root *root)
3909{
3910 struct inode *inode = new_inode(s);
3911
3912 if (!inode)
3913 return ERR_PTR(-ENOMEM);
3914
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003915 BTRFS_I(inode)->root = root;
3916 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3917 BTRFS_I(inode)->dummy_inode = 1;
3918
3919 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3920 inode->i_op = &simple_dir_inode_operations;
3921 inode->i_fop = &simple_dir_operations;
3922 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3923 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3924
3925 return inode;
3926}
3927
Chris Mason3de45862008-11-17 21:02:50 -05003928struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04003929{
Chris Masond3977122009-01-05 21:25:51 -05003930 struct inode *inode;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003931 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003932 struct btrfs_root *sub_root = root;
3933 struct btrfs_key location;
Yan, Zheng76dda932009-09-21 16:00:26 -04003934 int index;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04003935 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003936
3937 if (dentry->d_name.len > BTRFS_NAME_LEN)
3938 return ERR_PTR(-ENAMETOOLONG);
Chris Mason5f39d392007-10-15 16:14:19 -04003939
Josef Bacikb4aff1f2011-06-28 16:18:59 -04003940 if (unlikely(d_need_lookup(dentry))) {
3941 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3942 kfree(dentry->d_fsdata);
3943 dentry->d_fsdata = NULL;
Josef Bacika66e7cc2011-09-18 10:34:03 -04003944 /* This thing is hashed, drop it for now */
3945 d_drop(dentry);
Josef Bacikb4aff1f2011-06-28 16:18:59 -04003946 } else {
3947 ret = btrfs_inode_by_name(dir, dentry, &location);
3948 }
Chris Mason5f39d392007-10-15 16:14:19 -04003949
Chris Mason39279cc2007-06-12 06:35:45 -04003950 if (ret < 0)
3951 return ERR_PTR(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04003952
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003953 if (location.objectid == 0)
3954 return NULL;
3955
3956 if (location.type == BTRFS_INODE_ITEM_KEY) {
Josef Bacik73f73412009-12-04 17:38:27 +00003957 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003958 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04003959 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003960
3961 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3962
Yan, Zheng76dda932009-09-21 16:00:26 -04003963 index = srcu_read_lock(&root->fs_info->subvol_srcu);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003964 ret = fixup_tree_root_location(root, dir, dentry,
3965 &location, &sub_root);
3966 if (ret < 0) {
3967 if (ret != -ENOENT)
3968 inode = ERR_PTR(ret);
3969 else
3970 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3971 } else {
Josef Bacik73f73412009-12-04 17:38:27 +00003972 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04003973 }
Yan, Zheng76dda932009-09-21 16:00:26 -04003974 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3975
Julia Lawall34d19ba2011-01-24 19:55:19 +00003976 if (!IS_ERR(inode) && root != sub_root) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00003977 down_read(&root->fs_info->cleanup_work_sem);
3978 if (!(inode->i_sb->s_flags & MS_RDONLY))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003979 ret = btrfs_orphan_cleanup(sub_root);
Yan, Zhengc71bf092009-11-12 09:34:40 +00003980 up_read(&root->fs_info->cleanup_work_sem);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003981 if (ret)
3982 inode = ERR_PTR(ret);
Yan, Zhengc71bf092009-11-12 09:34:40 +00003983 }
3984
Chris Mason3de45862008-11-17 21:02:50 -05003985 return inode;
3986}
3987
Nick Pigginfe15ce42011-01-07 17:49:23 +11003988static int btrfs_dentry_delete(const struct dentry *dentry)
Yan, Zheng76dda932009-09-21 16:00:26 -04003989{
3990 struct btrfs_root *root;
3991
Yan, Zhengefefb142009-10-09 09:25:16 -04003992 if (!dentry->d_inode && !IS_ROOT(dentry))
3993 dentry = dentry->d_parent;
Yan, Zheng76dda932009-09-21 16:00:26 -04003994
Yan, Zhengefefb142009-10-09 09:25:16 -04003995 if (dentry->d_inode) {
3996 root = BTRFS_I(dentry->d_inode)->root;
3997 if (btrfs_root_refs(&root->root_item) == 0)
3998 return 1;
3999 }
Yan, Zheng76dda932009-09-21 16:00:26 -04004000 return 0;
4001}
4002
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004003static void btrfs_dentry_release(struct dentry *dentry)
4004{
4005 if (dentry->d_fsdata)
4006 kfree(dentry->d_fsdata);
4007}
4008
Chris Mason3de45862008-11-17 21:02:50 -05004009static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4010 struct nameidata *nd)
4011{
Josef Bacika66e7cc2011-09-18 10:34:03 -04004012 struct dentry *ret;
4013
4014 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4015 if (unlikely(d_need_lookup(dentry))) {
4016 spin_lock(&dentry->d_lock);
4017 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4018 spin_unlock(&dentry->d_lock);
4019 }
4020 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004021}
4022
Miao Xie16cdcec2011-04-22 18:12:22 +08004023unsigned char btrfs_filetype_table[] = {
Chris Mason39279cc2007-06-12 06:35:45 -04004024 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4025};
4026
David Woodhousecbdf5a22008-08-06 19:42:33 +01004027static int btrfs_real_readdir(struct file *filp, void *dirent,
4028 filldir_t filldir)
Chris Mason39279cc2007-06-12 06:35:45 -04004029{
Chris Mason6da6aba2007-12-18 16:15:09 -05004030 struct inode *inode = filp->f_dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04004031 struct btrfs_root *root = BTRFS_I(inode)->root;
4032 struct btrfs_item *item;
4033 struct btrfs_dir_item *di;
4034 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04004035 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04004036 struct btrfs_path *path;
Miao Xie16cdcec2011-04-22 18:12:22 +08004037 struct list_head ins_list;
4038 struct list_head del_list;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004039 struct qstr q;
Chris Mason39279cc2007-06-12 06:35:45 -04004040 int ret;
Chris Mason5f39d392007-10-15 16:14:19 -04004041 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04004042 int slot;
Chris Mason39279cc2007-06-12 06:35:45 -04004043 unsigned char d_type;
4044 int over = 0;
4045 u32 di_cur;
4046 u32 di_total;
4047 u32 di_len;
4048 int key_type = BTRFS_DIR_INDEX_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04004049 char tmp_name[32];
4050 char *name_ptr;
4051 int name_len;
Miao Xie16cdcec2011-04-22 18:12:22 +08004052 int is_curr = 0; /* filp->f_pos points to the current index? */
Chris Mason39279cc2007-06-12 06:35:45 -04004053
4054 /* FIXME, use a real flag for deciding about the key type */
4055 if (root->fs_info->tree_root == root)
4056 key_type = BTRFS_DIR_ITEM_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04004057
Chris Mason39544012007-12-12 14:38:19 -05004058 /* special case for "." */
4059 if (filp->f_pos == 0) {
Hidetoshi Seto3765fef2011-09-18 10:20:46 -04004060 over = filldir(dirent, ".", 1,
4061 filp->f_pos, btrfs_ino(inode), DT_DIR);
Chris Mason39544012007-12-12 14:38:19 -05004062 if (over)
4063 return 0;
4064 filp->f_pos = 1;
4065 }
Chris Mason39544012007-12-12 14:38:19 -05004066 /* special case for .., just use the back ref */
4067 if (filp->f_pos == 1) {
David Woodhouse5ecc7e52008-08-17 15:14:48 +01004068 u64 pino = parent_ino(filp->f_path.dentry);
Chris Mason39544012007-12-12 14:38:19 -05004069 over = filldir(dirent, "..", 2,
Hidetoshi Seto3765fef2011-09-18 10:20:46 -04004070 filp->f_pos, pino, DT_DIR);
Chris Mason39544012007-12-12 14:38:19 -05004071 if (over)
David Woodhouse49593bf2008-08-17 17:08:36 +01004072 return 0;
Chris Mason39544012007-12-12 14:38:19 -05004073 filp->f_pos = 2;
4074 }
David Woodhouse49593bf2008-08-17 17:08:36 +01004075 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08004076 if (!path)
4077 return -ENOMEM;
Chris Masonff5714c2011-05-28 07:00:39 -04004078
Josef Bacik026fd312011-05-13 10:32:11 -04004079 path->reada = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01004080
Miao Xie16cdcec2011-04-22 18:12:22 +08004081 if (key_type == BTRFS_DIR_INDEX_KEY) {
4082 INIT_LIST_HEAD(&ins_list);
4083 INIT_LIST_HEAD(&del_list);
4084 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4085 }
4086
Chris Mason39279cc2007-06-12 06:35:45 -04004087 btrfs_set_key_type(&key, key_type);
4088 key.offset = filp->f_pos;
Li Zefan33345d012011-04-20 10:31:50 +08004089 key.objectid = btrfs_ino(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004090
Chris Mason39279cc2007-06-12 06:35:45 -04004091 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4092 if (ret < 0)
4093 goto err;
David Woodhouse49593bf2008-08-17 17:08:36 +01004094
4095 while (1) {
Chris Mason5f39d392007-10-15 16:14:19 -04004096 leaf = path->nodes[0];
Chris Mason39279cc2007-06-12 06:35:45 -04004097 slot = path->slots[0];
Li Zefanb9e03af2011-03-23 10:43:58 +08004098 if (slot >= btrfs_header_nritems(leaf)) {
4099 ret = btrfs_next_leaf(root, path);
4100 if (ret < 0)
4101 goto err;
4102 else if (ret > 0)
4103 break;
4104 continue;
Chris Mason39279cc2007-06-12 06:35:45 -04004105 }
Chris Mason3de45862008-11-17 21:02:50 -05004106
Chris Mason5f39d392007-10-15 16:14:19 -04004107 item = btrfs_item_nr(leaf, slot);
4108 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4109
4110 if (found_key.objectid != key.objectid)
Chris Mason39279cc2007-06-12 06:35:45 -04004111 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004112 if (btrfs_key_type(&found_key) != key_type)
Chris Mason39279cc2007-06-12 06:35:45 -04004113 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004114 if (found_key.offset < filp->f_pos)
Li Zefanb9e03af2011-03-23 10:43:58 +08004115 goto next;
Miao Xie16cdcec2011-04-22 18:12:22 +08004116 if (key_type == BTRFS_DIR_INDEX_KEY &&
4117 btrfs_should_delete_dir_index(&del_list,
4118 found_key.offset))
4119 goto next;
Chris Mason5f39d392007-10-15 16:14:19 -04004120
4121 filp->f_pos = found_key.offset;
Miao Xie16cdcec2011-04-22 18:12:22 +08004122 is_curr = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01004123
Chris Mason39279cc2007-06-12 06:35:45 -04004124 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4125 di_cur = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04004126 di_total = btrfs_item_size(leaf, item);
David Woodhouse49593bf2008-08-17 17:08:36 +01004127
4128 while (di_cur < di_total) {
Chris Mason5f39d392007-10-15 16:14:19 -04004129 struct btrfs_key location;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004130 struct dentry *tmp;
Chris Mason5f39d392007-10-15 16:14:19 -04004131
Josef Bacik22a94d42011-03-16 16:47:17 -04004132 if (verify_dir_item(root, leaf, di))
4133 break;
4134
Chris Mason5f39d392007-10-15 16:14:19 -04004135 name_len = btrfs_dir_name_len(leaf, di);
David Woodhouse49593bf2008-08-17 17:08:36 +01004136 if (name_len <= sizeof(tmp_name)) {
Chris Mason5f39d392007-10-15 16:14:19 -04004137 name_ptr = tmp_name;
4138 } else {
4139 name_ptr = kmalloc(name_len, GFP_NOFS);
David Woodhouse49593bf2008-08-17 17:08:36 +01004140 if (!name_ptr) {
4141 ret = -ENOMEM;
4142 goto err;
4143 }
Chris Mason5f39d392007-10-15 16:14:19 -04004144 }
4145 read_extent_buffer(leaf, name_ptr,
4146 (unsigned long)(di + 1), name_len);
4147
4148 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4149 btrfs_dir_item_key_to_cpu(leaf, di, &location);
Chris Mason3de45862008-11-17 21:02:50 -05004150
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004151 q.name = name_ptr;
4152 q.len = name_len;
4153 q.hash = full_name_hash(q.name, q.len);
4154 tmp = d_lookup(filp->f_dentry, &q);
4155 if (!tmp) {
4156 struct btrfs_key *newkey;
4157
4158 newkey = kzalloc(sizeof(struct btrfs_key),
4159 GFP_NOFS);
4160 if (!newkey)
4161 goto no_dentry;
4162 tmp = d_alloc(filp->f_dentry, &q);
4163 if (!tmp) {
4164 kfree(newkey);
4165 dput(tmp);
4166 goto no_dentry;
4167 }
4168 memcpy(newkey, &location,
4169 sizeof(struct btrfs_key));
4170 tmp->d_fsdata = newkey;
4171 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4172 d_rehash(tmp);
4173 dput(tmp);
4174 } else {
4175 dput(tmp);
4176 }
4177no_dentry:
Chris Mason3de45862008-11-17 21:02:50 -05004178 /* is this a reference to our own snapshot? If so
4179 * skip it
4180 */
4181 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4182 location.objectid == root->root_key.objectid) {
4183 over = 0;
4184 goto skip;
4185 }
Chris Mason5f39d392007-10-15 16:14:19 -04004186 over = filldir(dirent, name_ptr, name_len,
David Woodhouse49593bf2008-08-17 17:08:36 +01004187 found_key.offset, location.objectid,
Chris Mason39279cc2007-06-12 06:35:45 -04004188 d_type);
Chris Mason5f39d392007-10-15 16:14:19 -04004189
Chris Mason3de45862008-11-17 21:02:50 -05004190skip:
Chris Mason5f39d392007-10-15 16:14:19 -04004191 if (name_ptr != tmp_name)
4192 kfree(name_ptr);
4193
Chris Mason39279cc2007-06-12 06:35:45 -04004194 if (over)
4195 goto nopos;
Josef Bacik5103e942007-11-16 11:45:54 -05004196 di_len = btrfs_dir_name_len(leaf, di) +
David Woodhouse49593bf2008-08-17 17:08:36 +01004197 btrfs_dir_data_len(leaf, di) + sizeof(*di);
Chris Mason39279cc2007-06-12 06:35:45 -04004198 di_cur += di_len;
4199 di = (struct btrfs_dir_item *)((char *)di + di_len);
4200 }
Li Zefanb9e03af2011-03-23 10:43:58 +08004201next:
4202 path->slots[0]++;
Chris Mason39279cc2007-06-12 06:35:45 -04004203 }
David Woodhouse49593bf2008-08-17 17:08:36 +01004204
Miao Xie16cdcec2011-04-22 18:12:22 +08004205 if (key_type == BTRFS_DIR_INDEX_KEY) {
4206 if (is_curr)
4207 filp->f_pos++;
4208 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4209 &ins_list);
4210 if (ret)
4211 goto nopos;
4212 }
4213
David Woodhouse49593bf2008-08-17 17:08:36 +01004214 /* Reached end of directory/root. Bump pos past the last item. */
Yan Zheng5e591a02008-02-19 11:41:02 -05004215 if (key_type == BTRFS_DIR_INDEX_KEY)
Jan Engelhardt406266a2009-12-09 22:00:38 +00004216 /*
4217 * 32-bit glibc will use getdents64, but then strtol -
4218 * so the last number we can serve is this.
4219 */
4220 filp->f_pos = 0x7fffffff;
Yan Zheng5e591a02008-02-19 11:41:02 -05004221 else
4222 filp->f_pos++;
Chris Mason39279cc2007-06-12 06:35:45 -04004223nopos:
4224 ret = 0;
4225err:
Miao Xie16cdcec2011-04-22 18:12:22 +08004226 if (key_type == BTRFS_DIR_INDEX_KEY)
4227 btrfs_put_delayed_items(&ins_list, &del_list);
Chris Mason39279cc2007-06-12 06:35:45 -04004228 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04004229 return ret;
4230}
4231
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004232int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Chris Mason39279cc2007-06-12 06:35:45 -04004233{
4234 struct btrfs_root *root = BTRFS_I(inode)->root;
4235 struct btrfs_trans_handle *trans;
4236 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04004237 bool nolock = false;
Chris Mason39279cc2007-06-12 06:35:45 -04004238
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004239 if (BTRFS_I(inode)->dummy_inode)
Chris Mason4ca8b412008-08-05 13:30:48 -04004240 return 0;
4241
Chris Mason2cf85722011-07-26 15:35:09 -04004242 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
Li Zefan82d59022011-04-20 10:33:24 +08004243 nolock = true;
Josef Bacik0af3d002010-06-21 14:48:16 -04004244
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004245 if (wbc->sync_mode == WB_SYNC_ALL) {
Josef Bacik0af3d002010-06-21 14:48:16 -04004246 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004247 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0af3d002010-06-21 14:48:16 -04004248 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004249 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00004250 if (IS_ERR(trans))
4251 return PTR_ERR(trans);
Josef Bacik0af3d002010-06-21 14:48:16 -04004252 if (nolock)
4253 ret = btrfs_end_transaction_nolock(trans, root);
4254 else
4255 ret = btrfs_commit_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004256 }
4257 return ret;
4258}
4259
4260/*
Chris Mason54aa1f42007-06-22 14:16:25 -04004261 * This is somewhat expensive, updating the tree every time the
Chris Mason39279cc2007-06-12 06:35:45 -04004262 * inode changes. But, it is most likely to find the inode in cache.
4263 * FIXME, needs more benchmarking...there are no reasons other than performance
4264 * to keep or drop this code.
4265 */
Josef Bacik22c44fe2011-11-30 10:45:38 -05004266int btrfs_dirty_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04004267{
4268 struct btrfs_root *root = BTRFS_I(inode)->root;
4269 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004270 int ret;
4271
4272 if (BTRFS_I(inode)->dummy_inode)
Josef Bacik22c44fe2011-11-30 10:45:38 -05004273 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004274
Josef Bacik7a7eaa42011-04-13 12:54:33 -04004275 trans = btrfs_join_transaction(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004276 if (IS_ERR(trans))
4277 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004278
4279 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04004280 if (ret && ret == -ENOSPC) {
4281 /* whoops, lets try again with the full transaction */
4282 btrfs_end_transaction(trans, root);
4283 trans = btrfs_start_transaction(root, 1);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004284 if (IS_ERR(trans))
4285 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004286
Chris Mason94b60442010-05-26 11:02:00 -04004287 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04004288 }
Chris Mason39279cc2007-06-12 06:35:45 -04004289 btrfs_end_transaction(trans, root);
Miao Xie16cdcec2011-04-22 18:12:22 +08004290 if (BTRFS_I(inode)->delayed_node)
4291 btrfs_balance_delayed_items(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004292
4293 return ret;
4294}
4295
4296/*
4297 * This is a copy of file_update_time. We need this so we can return error on
4298 * ENOSPC for updating the inode in the case of file write and mmap writes.
4299 */
4300int btrfs_update_time(struct file *file)
4301{
4302 struct inode *inode = file->f_path.dentry->d_inode;
4303 struct timespec now;
4304 int ret;
4305 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4306
4307 /* First try to exhaust all avenues to not sync */
4308 if (IS_NOCMTIME(inode))
4309 return 0;
4310
4311 now = current_fs_time(inode->i_sb);
4312 if (!timespec_equal(&inode->i_mtime, &now))
4313 sync_it = S_MTIME;
4314
4315 if (!timespec_equal(&inode->i_ctime, &now))
4316 sync_it |= S_CTIME;
4317
4318 if (IS_I_VERSION(inode))
4319 sync_it |= S_VERSION;
4320
4321 if (!sync_it)
4322 return 0;
4323
4324 /* Finally allowed to write? Takes lock. */
4325 if (mnt_want_write_file(file))
4326 return 0;
4327
4328 /* Only change inode inside the lock region */
4329 if (sync_it & S_VERSION)
4330 inode_inc_iversion(inode);
4331 if (sync_it & S_CTIME)
4332 inode->i_ctime = now;
4333 if (sync_it & S_MTIME)
4334 inode->i_mtime = now;
4335 ret = btrfs_dirty_inode(inode);
4336 if (!ret)
4337 mark_inode_dirty_sync(inode);
4338 mnt_drop_write(file->f_path.mnt);
4339 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004340}
4341
Chris Masond352ac62008-09-29 15:18:18 -04004342/*
4343 * find the highest existing sequence number in a directory
4344 * and then set the in-memory index_cnt variable to reflect
4345 * free sequence numbers
4346 */
Josef Bacikaec74772008-07-24 12:12:38 -04004347static int btrfs_set_inode_index_count(struct inode *inode)
4348{
4349 struct btrfs_root *root = BTRFS_I(inode)->root;
4350 struct btrfs_key key, found_key;
4351 struct btrfs_path *path;
4352 struct extent_buffer *leaf;
4353 int ret;
4354
Li Zefan33345d012011-04-20 10:31:50 +08004355 key.objectid = btrfs_ino(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004356 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4357 key.offset = (u64)-1;
4358
4359 path = btrfs_alloc_path();
4360 if (!path)
4361 return -ENOMEM;
4362
4363 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4364 if (ret < 0)
4365 goto out;
4366 /* FIXME: we should be able to handle this */
4367 if (ret == 0)
4368 goto out;
4369 ret = 0;
4370
4371 /*
4372 * MAGIC NUMBER EXPLANATION:
4373 * since we search a directory based on f_pos we have to start at 2
4374 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4375 * else has to start at 2
4376 */
4377 if (path->slots[0] == 0) {
4378 BTRFS_I(inode)->index_cnt = 2;
4379 goto out;
4380 }
4381
4382 path->slots[0]--;
4383
4384 leaf = path->nodes[0];
4385 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4386
Li Zefan33345d012011-04-20 10:31:50 +08004387 if (found_key.objectid != btrfs_ino(inode) ||
Josef Bacikaec74772008-07-24 12:12:38 -04004388 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4389 BTRFS_I(inode)->index_cnt = 2;
4390 goto out;
4391 }
4392
4393 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4394out:
4395 btrfs_free_path(path);
4396 return ret;
4397}
4398
Chris Masond352ac62008-09-29 15:18:18 -04004399/*
4400 * helper to find a free sequence number in a given directory. This current
4401 * code is very simple, later versions will do smarter things in the btree
4402 */
Chris Mason3de45862008-11-17 21:02:50 -05004403int btrfs_set_inode_index(struct inode *dir, u64 *index)
Josef Bacikaec74772008-07-24 12:12:38 -04004404{
4405 int ret = 0;
4406
4407 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
Miao Xie16cdcec2011-04-22 18:12:22 +08004408 ret = btrfs_inode_delayed_dir_index_count(dir);
4409 if (ret) {
4410 ret = btrfs_set_inode_index_count(dir);
4411 if (ret)
4412 return ret;
4413 }
Josef Bacikaec74772008-07-24 12:12:38 -04004414 }
4415
Chris Mason00e4e6b2008-08-05 11:18:09 -04004416 *index = BTRFS_I(dir)->index_cnt;
Josef Bacikaec74772008-07-24 12:12:38 -04004417 BTRFS_I(dir)->index_cnt++;
4418
4419 return ret;
4420}
4421
Chris Mason39279cc2007-06-12 06:35:45 -04004422static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4423 struct btrfs_root *root,
Josef Bacikaec74772008-07-24 12:12:38 -04004424 struct inode *dir,
Chris Mason9c583092008-01-29 15:15:18 -05004425 const char *name, int name_len,
Al Viro175a4eb2011-07-26 03:30:54 -04004426 u64 ref_objectid, u64 objectid,
4427 umode_t mode, u64 *index)
Chris Mason39279cc2007-06-12 06:35:45 -04004428{
4429 struct inode *inode;
Chris Mason5f39d392007-10-15 16:14:19 -04004430 struct btrfs_inode_item *inode_item;
Chris Mason39279cc2007-06-12 06:35:45 -04004431 struct btrfs_key *location;
Chris Mason5f39d392007-10-15 16:14:19 -04004432 struct btrfs_path *path;
Chris Mason9c583092008-01-29 15:15:18 -05004433 struct btrfs_inode_ref *ref;
4434 struct btrfs_key key[2];
4435 u32 sizes[2];
4436 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04004437 int ret;
4438 int owner;
4439
Chris Mason5f39d392007-10-15 16:14:19 -04004440 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07004441 if (!path)
4442 return ERR_PTR(-ENOMEM);
Chris Mason5f39d392007-10-15 16:14:19 -04004443
Chris Mason39279cc2007-06-12 06:35:45 -04004444 inode = new_inode(root->fs_info->sb);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00004445 if (!inode) {
4446 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04004447 return ERR_PTR(-ENOMEM);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00004448 }
Chris Mason39279cc2007-06-12 06:35:45 -04004449
Li Zefan581bb052011-04-20 10:06:11 +08004450 /*
4451 * we have to initialize this early, so we can reclaim the inode
4452 * number if we fail afterwards in this function.
4453 */
4454 inode->i_ino = objectid;
4455
Josef Bacikaec74772008-07-24 12:12:38 -04004456 if (dir) {
liubo1abe9b82011-03-24 11:18:59 +00004457 trace_btrfs_inode_request(dir);
4458
Chris Mason3de45862008-11-17 21:02:50 -05004459 ret = btrfs_set_inode_index(dir, index);
Shen Feng09771432009-04-02 16:46:06 -04004460 if (ret) {
Yoshinori Sano8fb27642011-04-09 02:30:07 +00004461 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04004462 iput(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004463 return ERR_PTR(ret);
Shen Feng09771432009-04-02 16:46:06 -04004464 }
Josef Bacikaec74772008-07-24 12:12:38 -04004465 }
4466 /*
4467 * index_cnt is ignored for everything but a dir,
4468 * btrfs_get_inode_index_count has an explanation for the magic
4469 * number
4470 */
4471 BTRFS_I(inode)->index_cnt = 2;
Chris Mason39279cc2007-06-12 06:35:45 -04004472 BTRFS_I(inode)->root = root;
Chris Masone02119d2008-09-05 16:13:11 -04004473 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik76195852010-11-19 02:18:02 +00004474 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik6a632092009-02-20 11:00:09 -05004475 btrfs_set_inode_space_info(root, inode);
Chris Masonb888db2b2007-08-27 16:49:44 -04004476
Al Viro569254b2011-07-24 17:08:40 -04004477 if (S_ISDIR(mode))
Chris Mason39279cc2007-06-12 06:35:45 -04004478 owner = 0;
4479 else
4480 owner = 1;
Chris Mason9c583092008-01-29 15:15:18 -05004481
4482 key[0].objectid = objectid;
4483 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4484 key[0].offset = 0;
4485
4486 key[1].objectid = objectid;
4487 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4488 key[1].offset = ref_objectid;
4489
4490 sizes[0] = sizeof(struct btrfs_inode_item);
4491 sizes[1] = name_len + sizeof(*ref);
4492
Chris Masonb9473432009-03-13 11:00:37 -04004493 path->leave_spinning = 1;
Chris Mason9c583092008-01-29 15:15:18 -05004494 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4495 if (ret != 0)
Chris Mason5f39d392007-10-15 16:14:19 -04004496 goto fail;
4497
Dmitry Monakhovecc11fab2010-03-04 17:31:47 +03004498 inode_init_owner(inode, dir, mode);
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004499 inode_set_bytes(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004500 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
Chris Mason5f39d392007-10-15 16:14:19 -04004501 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4502 struct btrfs_inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04004503 fill_inode_item(trans, path->nodes[0], inode_item, inode);
Chris Mason9c583092008-01-29 15:15:18 -05004504
4505 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4506 struct btrfs_inode_ref);
4507 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
Chris Mason00e4e6b2008-08-05 11:18:09 -04004508 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
Chris Mason9c583092008-01-29 15:15:18 -05004509 ptr = (unsigned long)(ref + 1);
4510 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4511
Chris Mason5f39d392007-10-15 16:14:19 -04004512 btrfs_mark_buffer_dirty(path->nodes[0]);
4513 btrfs_free_path(path);
4514
Chris Mason39279cc2007-06-12 06:35:45 -04004515 location = &BTRFS_I(inode)->location;
4516 location->objectid = objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04004517 location->offset = 0;
4518 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4519
Christoph Hellwig6cbff002009-04-17 10:37:41 +02004520 btrfs_inherit_iflags(inode, dir);
4521
Al Viro569254b2011-07-24 17:08:40 -04004522 if (S_ISREG(mode)) {
Chris Mason94272162009-07-02 12:26:06 -04004523 if (btrfs_test_opt(root, NODATASUM))
4524 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
Liu Bo75e7cb72011-03-22 10:12:20 +00004525 if (btrfs_test_opt(root, NODATACOW) ||
4526 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
Chris Mason94272162009-07-02 12:26:06 -04004527 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4528 }
4529
Chris Mason39279cc2007-06-12 06:35:45 -04004530 insert_inode_hash(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004531 inode_tree_add(inode);
liubo1abe9b82011-03-24 11:18:59 +00004532
4533 trace_btrfs_inode_new(inode);
Chris Mason1973f0f2011-06-24 13:13:29 -04004534 btrfs_set_inode_last_trans(trans, inode);
liubo1abe9b82011-03-24 11:18:59 +00004535
Chris Mason39279cc2007-06-12 06:35:45 -04004536 return inode;
Chris Mason5f39d392007-10-15 16:14:19 -04004537fail:
Josef Bacikaec74772008-07-24 12:12:38 -04004538 if (dir)
4539 BTRFS_I(dir)->index_cnt--;
Chris Mason5f39d392007-10-15 16:14:19 -04004540 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04004541 iput(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004542 return ERR_PTR(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04004543}
4544
4545static inline u8 btrfs_inode_type(struct inode *inode)
4546{
4547 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4548}
4549
Chris Masond352ac62008-09-29 15:18:18 -04004550/*
4551 * utility function to add 'inode' into 'parent_inode' with
4552 * a give name and a given sequence number.
4553 * if 'add_backref' is true, also insert a backref from the
4554 * inode to the parent directory.
4555 */
Chris Masone02119d2008-09-05 16:13:11 -04004556int btrfs_add_link(struct btrfs_trans_handle *trans,
4557 struct inode *parent_inode, struct inode *inode,
4558 const char *name, int name_len, int add_backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04004559{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004560 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004561 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04004562 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
Li Zefan33345d012011-04-20 10:31:50 +08004563 u64 ino = btrfs_ino(inode);
4564 u64 parent_ino = btrfs_ino(parent_inode);
Chris Mason5f39d392007-10-15 16:14:19 -04004565
Li Zefan33345d012011-04-20 10:31:50 +08004566 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004567 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4568 } else {
Li Zefan33345d012011-04-20 10:31:50 +08004569 key.objectid = ino;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004570 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4571 key.offset = 0;
4572 }
Chris Mason39279cc2007-06-12 06:35:45 -04004573
Li Zefan33345d012011-04-20 10:31:50 +08004574 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004575 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4576 key.objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08004577 parent_ino, index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004578 } else if (add_backref) {
Li Zefan33345d012011-04-20 10:31:50 +08004579 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4580 parent_ino, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004581 }
4582
Chris Mason39279cc2007-06-12 06:35:45 -04004583 if (ret == 0) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004584 ret = btrfs_insert_dir_item(trans, root, name, name_len,
Miao Xie16cdcec2011-04-22 18:12:22 +08004585 parent_inode, &key,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004586 btrfs_inode_type(inode), index);
Chris Masonfe66a052012-02-20 08:40:56 -05004587 if (ret)
4588 goto fail_dir_item;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004589
Chris Masondbe674a2008-07-17 12:54:05 -04004590 btrfs_i_size_write(parent_inode, parent_inode->i_size +
Chris Masone02119d2008-09-05 16:13:11 -04004591 name_len * 2);
Chris Mason79c44582007-06-25 10:09:33 -04004592 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
Chris Masone02119d2008-09-05 16:13:11 -04004593 ret = btrfs_update_inode(trans, root, parent_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004594 }
4595 return ret;
Chris Masonfe66a052012-02-20 08:40:56 -05004596
4597fail_dir_item:
4598 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4599 u64 local_index;
4600 int err;
4601 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4602 key.objectid, root->root_key.objectid,
4603 parent_ino, &local_index, name, name_len);
4604
4605 } else if (add_backref) {
4606 u64 local_index;
4607 int err;
4608
4609 err = btrfs_del_inode_ref(trans, root, name, name_len,
4610 ino, parent_ino, &local_index);
4611 }
4612 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04004613}
4614
4615static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
Josef Bacika1b075d2010-11-19 20:36:11 +00004616 struct inode *dir, struct dentry *dentry,
4617 struct inode *inode, int backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04004618{
Josef Bacika1b075d2010-11-19 20:36:11 +00004619 int err = btrfs_add_link(trans, dir, inode,
4620 dentry->d_name.name, dentry->d_name.len,
4621 backref, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004622 if (err > 0)
4623 err = -EEXIST;
4624 return err;
4625}
4626
Josef Bacik618e21d2007-07-11 10:18:17 -04004627static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -04004628 umode_t mode, dev_t rdev)
Josef Bacik618e21d2007-07-11 10:18:17 -04004629{
4630 struct btrfs_trans_handle *trans;
4631 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05004632 struct inode *inode = NULL;
Josef Bacik618e21d2007-07-11 10:18:17 -04004633 int err;
4634 int drop_inode = 0;
4635 u64 objectid;
Chris Mason1832a6d2007-12-21 16:27:21 -05004636 unsigned long nr = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004637 u64 index = 0;
Josef Bacik618e21d2007-07-11 10:18:17 -04004638
4639 if (!new_valid_dev(rdev))
4640 return -EINVAL;
4641
Josef Bacik9ed74f22009-09-11 16:12:44 -04004642 /*
4643 * 2 for inode item and ref
4644 * 2 for dir items
4645 * 1 for xattr if selinux is on
4646 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004647 trans = btrfs_start_transaction(root, 5);
4648 if (IS_ERR(trans))
4649 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05004650
Li Zefan581bb052011-04-20 10:06:11 +08004651 err = btrfs_find_free_ino(root, &objectid);
4652 if (err)
4653 goto out_unlock;
4654
Josef Bacikaec74772008-07-24 12:12:38 -04004655 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08004656 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04004657 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04004658 if (IS_ERR(inode)) {
4659 err = PTR_ERR(inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04004660 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04004661 }
Josef Bacik618e21d2007-07-11 10:18:17 -04004662
Eric Paris2a7dba32011-02-01 11:05:39 -05004663 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04004664 if (err) {
4665 drop_inode = 1;
4666 goto out_unlock;
4667 }
4668
Casey Schauflerad19db72011-12-15 10:09:07 -05004669 /*
4670 * If the active LSM wants to access the inode during
4671 * d_instantiate it needs these. Smack checks to see
4672 * if the filesystem supports xattrs by looking at the
4673 * ops vector.
4674 */
4675
4676 inode->i_op = &btrfs_special_inode_operations;
Josef Bacika1b075d2010-11-19 20:36:11 +00004677 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Josef Bacik618e21d2007-07-11 10:18:17 -04004678 if (err)
4679 drop_inode = 1;
4680 else {
Josef Bacik618e21d2007-07-11 10:18:17 -04004681 init_special_inode(inode, inode->i_mode, rdev);
Yan1b4ab1b2007-08-29 09:11:44 -04004682 btrfs_update_inode(trans, root, inode);
Al Viro08c422c2011-12-23 07:58:13 -05004683 d_instantiate(dentry, inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04004684 }
Josef Bacik618e21d2007-07-11 10:18:17 -04004685out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004686 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004687 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004688 btrfs_btree_balance_dirty(root, nr);
Josef Bacik618e21d2007-07-11 10:18:17 -04004689 if (drop_inode) {
4690 inode_dec_link_count(inode);
4691 iput(inode);
4692 }
Josef Bacik618e21d2007-07-11 10:18:17 -04004693 return err;
4694}
4695
Chris Mason39279cc2007-06-12 06:35:45 -04004696static int btrfs_create(struct inode *dir, struct dentry *dentry,
Al Viro4acdaf22011-07-26 01:42:34 -04004697 umode_t mode, struct nameidata *nd)
Chris Mason39279cc2007-06-12 06:35:45 -04004698{
4699 struct btrfs_trans_handle *trans;
4700 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05004701 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04004702 int drop_inode = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004703 int err;
Chris Mason1832a6d2007-12-21 16:27:21 -05004704 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004705 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004706 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004707
Josef Bacik9ed74f22009-09-11 16:12:44 -04004708 /*
4709 * 2 for inode item and ref
4710 * 2 for dir items
4711 * 1 for xattr if selinux is on
4712 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004713 trans = btrfs_start_transaction(root, 5);
4714 if (IS_ERR(trans))
4715 return PTR_ERR(trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004716
Li Zefan581bb052011-04-20 10:06:11 +08004717 err = btrfs_find_free_ino(root, &objectid);
4718 if (err)
4719 goto out_unlock;
4720
Josef Bacikaec74772008-07-24 12:12:38 -04004721 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08004722 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04004723 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04004724 if (IS_ERR(inode)) {
4725 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004726 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04004727 }
Chris Mason39279cc2007-06-12 06:35:45 -04004728
Eric Paris2a7dba32011-02-01 11:05:39 -05004729 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04004730 if (err) {
4731 drop_inode = 1;
4732 goto out_unlock;
4733 }
4734
Casey Schauflerad19db72011-12-15 10:09:07 -05004735 /*
4736 * If the active LSM wants to access the inode during
4737 * d_instantiate it needs these. Smack checks to see
4738 * if the filesystem supports xattrs by looking at the
4739 * ops vector.
4740 */
4741 inode->i_fop = &btrfs_file_operations;
4742 inode->i_op = &btrfs_file_inode_operations;
4743
Josef Bacika1b075d2010-11-19 20:36:11 +00004744 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004745 if (err)
4746 drop_inode = 1;
4747 else {
4748 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04004749 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05004750 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Al Viro08c422c2011-12-23 07:58:13 -05004751 d_instantiate(dentry, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004752 }
Chris Mason39279cc2007-06-12 06:35:45 -04004753out_unlock:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004754 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004755 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004756 if (drop_inode) {
4757 inode_dec_link_count(inode);
4758 iput(inode);
4759 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004760 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004761 return err;
4762}
4763
4764static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4765 struct dentry *dentry)
4766{
4767 struct btrfs_trans_handle *trans;
4768 struct btrfs_root *root = BTRFS_I(dir)->root;
4769 struct inode *inode = old_dentry->d_inode;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004770 u64 index;
Chris Mason1832a6d2007-12-21 16:27:21 -05004771 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004772 int err;
4773 int drop_inode = 0;
4774
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00004775 /* do not allow sys_link's with other subvols of the same device */
4776 if (root->objectid != BTRFS_I(inode)->root->objectid)
Mark Fasheh3ab35642011-03-22 17:20:26 +00004777 return -EXDEV;
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00004778
Al Viroc055e992011-03-04 17:15:18 +00004779 if (inode->i_nlink == ~0U)
4780 return -EMLINK;
Josef Bacik9ed74f22009-09-11 16:12:44 -04004781
Chris Mason3de45862008-11-17 21:02:50 -05004782 err = btrfs_set_inode_index(dir, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04004783 if (err)
4784 goto fail;
4785
Yan, Zhenga22285a2010-05-16 10:48:46 -04004786 /*
Miao Xie7e6b6462011-02-18 09:21:17 +00004787 * 2 items for inode and inode ref
Yan, Zhenga22285a2010-05-16 10:48:46 -04004788 * 2 items for dir items
Miao Xie7e6b6462011-02-18 09:21:17 +00004789 * 1 item for parent inode
Yan, Zhenga22285a2010-05-16 10:48:46 -04004790 */
Miao Xie7e6b6462011-02-18 09:21:17 +00004791 trans = btrfs_start_transaction(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004792 if (IS_ERR(trans)) {
4793 err = PTR_ERR(trans);
4794 goto fail;
4795 }
Chris Mason5f39d392007-10-15 16:14:19 -04004796
Miao Xie31534952011-04-13 13:19:21 +08004797 btrfs_inc_nlink(inode);
4798 inode->i_ctime = CURRENT_TIME;
Al Viro7de9c6ee2010-10-23 11:11:40 -04004799 ihold(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04004800
Josef Bacika1b075d2010-11-19 20:36:11 +00004801 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
Chris Mason5f39d392007-10-15 16:14:19 -04004802
Yan, Zhenga5719522009-09-24 09:17:31 -04004803 if (err) {
Chris Mason39279cc2007-06-12 06:35:45 -04004804 drop_inode = 1;
Yan, Zhenga5719522009-09-24 09:17:31 -04004805 } else {
Al Viro10d9f302011-07-16 23:09:10 -04004806 struct dentry *parent = dentry->d_parent;
Yan, Zhenga5719522009-09-24 09:17:31 -04004807 err = btrfs_update_inode(trans, root, inode);
4808 BUG_ON(err);
Al Viro08c422c2011-12-23 07:58:13 -05004809 d_instantiate(dentry, inode);
Josef Bacik6a912212010-11-20 09:48:00 +00004810 btrfs_log_new_name(trans, inode, NULL, parent);
Yan, Zhenga5719522009-09-24 09:17:31 -04004811 }
Chris Mason39279cc2007-06-12 06:35:45 -04004812
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004813 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004814 btrfs_end_transaction(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05004815fail:
Chris Mason39279cc2007-06-12 06:35:45 -04004816 if (drop_inode) {
4817 inode_dec_link_count(inode);
4818 iput(inode);
4819 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004820 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004821 return err;
4822}
4823
Al Viro18bb1db2011-07-26 01:41:39 -04004824static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Chris Mason39279cc2007-06-12 06:35:45 -04004825{
Chris Masonb9d86662008-05-02 16:13:49 -04004826 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04004827 struct btrfs_trans_handle *trans;
4828 struct btrfs_root *root = BTRFS_I(dir)->root;
4829 int err = 0;
4830 int drop_on_err = 0;
Chris Masonb9d86662008-05-02 16:13:49 -04004831 u64 objectid = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04004832 u64 index = 0;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004833 unsigned long nr = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04004834
Josef Bacik9ed74f22009-09-11 16:12:44 -04004835 /*
4836 * 2 items for inode and ref
4837 * 2 items for dir items
4838 * 1 for xattr if selinux is on
4839 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04004840 trans = btrfs_start_transaction(root, 5);
4841 if (IS_ERR(trans))
4842 return PTR_ERR(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04004843
Li Zefan581bb052011-04-20 10:06:11 +08004844 err = btrfs_find_free_ino(root, &objectid);
4845 if (err)
4846 goto out_fail;
4847
Josef Bacikaec74772008-07-24 12:12:38 -04004848 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08004849 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04004850 S_IFDIR | mode, &index);
Chris Mason39279cc2007-06-12 06:35:45 -04004851 if (IS_ERR(inode)) {
4852 err = PTR_ERR(inode);
4853 goto out_fail;
4854 }
Chris Mason5f39d392007-10-15 16:14:19 -04004855
Chris Mason39279cc2007-06-12 06:35:45 -04004856 drop_on_err = 1;
Josef Bacik33268ea2008-07-24 12:16:36 -04004857
Eric Paris2a7dba32011-02-01 11:05:39 -05004858 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04004859 if (err)
4860 goto out_fail;
4861
Chris Mason39279cc2007-06-12 06:35:45 -04004862 inode->i_op = &btrfs_dir_inode_operations;
4863 inode->i_fop = &btrfs_dir_file_operations;
Chris Mason39279cc2007-06-12 06:35:45 -04004864
Chris Masondbe674a2008-07-17 12:54:05 -04004865 btrfs_i_size_write(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004866 err = btrfs_update_inode(trans, root, inode);
4867 if (err)
4868 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04004869
Josef Bacika1b075d2010-11-19 20:36:11 +00004870 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4871 dentry->d_name.len, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04004872 if (err)
4873 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04004874
Chris Mason39279cc2007-06-12 06:35:45 -04004875 d_instantiate(dentry, inode);
4876 drop_on_err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004877
4878out_fail:
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004879 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004880 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04004881 if (drop_on_err)
4882 iput(inode);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04004883 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04004884 return err;
4885}
4886
Chris Masond352ac62008-09-29 15:18:18 -04004887/* helper for btfs_get_extent. Given an existing extent in the tree,
4888 * and an extent that you want to insert, deal with overlap and insert
4889 * the new extent into the tree.
4890 */
Chris Mason3b951512008-04-17 11:29:12 -04004891static int merge_extent_mapping(struct extent_map_tree *em_tree,
4892 struct extent_map *existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04004893 struct extent_map *em,
4894 u64 map_start, u64 map_len)
Chris Mason3b951512008-04-17 11:29:12 -04004895{
4896 u64 start_diff;
Chris Mason3b951512008-04-17 11:29:12 -04004897
Chris Masone6dcd2d2008-07-17 12:53:50 -04004898 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4899 start_diff = map_start - em->start;
4900 em->start = map_start;
4901 em->len = map_len;
Chris Masonc8b97812008-10-29 14:49:59 -04004902 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4903 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04004904 em->block_start += start_diff;
Chris Masonc8b97812008-10-29 14:49:59 -04004905 em->block_len -= start_diff;
4906 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004907 return add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04004908}
4909
Chris Masonc8b97812008-10-29 14:49:59 -04004910static noinline int uncompress_inline(struct btrfs_path *path,
4911 struct inode *inode, struct page *page,
4912 size_t pg_offset, u64 extent_offset,
4913 struct btrfs_file_extent_item *item)
4914{
4915 int ret;
4916 struct extent_buffer *leaf = path->nodes[0];
4917 char *tmp;
4918 size_t max_size;
4919 unsigned long inline_size;
4920 unsigned long ptr;
Li Zefan261507a02010-12-17 14:21:50 +08004921 int compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04004922
4923 WARN_ON(pg_offset != 0);
Li Zefan261507a02010-12-17 14:21:50 +08004924 compress_type = btrfs_file_extent_compression(leaf, item);
Chris Masonc8b97812008-10-29 14:49:59 -04004925 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4926 inline_size = btrfs_file_extent_inline_item_len(leaf,
4927 btrfs_item_nr(leaf, path->slots[0]));
4928 tmp = kmalloc(inline_size, GFP_NOFS);
Tsutomu Itoh8d413712011-04-25 19:43:52 -04004929 if (!tmp)
4930 return -ENOMEM;
Chris Masonc8b97812008-10-29 14:49:59 -04004931 ptr = btrfs_file_extent_inline_start(item);
4932
4933 read_extent_buffer(leaf, tmp, ptr, inline_size);
4934
Chris Mason5b050f02008-11-11 09:34:41 -05004935 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
Li Zefan261507a02010-12-17 14:21:50 +08004936 ret = btrfs_decompress(compress_type, tmp, page,
4937 extent_offset, inline_size, max_size);
Chris Masonc8b97812008-10-29 14:49:59 -04004938 if (ret) {
4939 char *kaddr = kmap_atomic(page, KM_USER0);
4940 unsigned long copy_size = min_t(u64,
4941 PAGE_CACHE_SIZE - pg_offset,
4942 max_size - extent_offset);
4943 memset(kaddr + pg_offset, 0, copy_size);
4944 kunmap_atomic(kaddr, KM_USER0);
4945 }
4946 kfree(tmp);
4947 return 0;
4948}
4949
Chris Masond352ac62008-09-29 15:18:18 -04004950/*
4951 * a bit scary, this does extent mapping from logical file offset to the disk.
Chris Masond3977122009-01-05 21:25:51 -05004952 * the ugly parts come from merging extents from the disk with the in-ram
4953 * representation. This gets more complex because of the data=ordered code,
Chris Masond352ac62008-09-29 15:18:18 -04004954 * where the in-ram extents might be locked pending data=ordered completion.
4955 *
4956 * This also copies inline extents directly into the page.
4957 */
Chris Masond3977122009-01-05 21:25:51 -05004958
Chris Masona52d9a82007-08-27 16:49:44 -04004959struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
Chris Mason70dec802008-01-29 09:59:12 -05004960 size_t pg_offset, u64 start, u64 len,
Chris Masona52d9a82007-08-27 16:49:44 -04004961 int create)
4962{
4963 int ret;
4964 int err = 0;
Chris Masondb945352007-10-15 16:15:53 -04004965 u64 bytenr;
Chris Masona52d9a82007-08-27 16:49:44 -04004966 u64 extent_start = 0;
4967 u64 extent_end = 0;
Li Zefan33345d012011-04-20 10:31:50 +08004968 u64 objectid = btrfs_ino(inode);
Chris Masona52d9a82007-08-27 16:49:44 -04004969 u32 found_type;
Chris Masonf4219502008-07-22 11:18:09 -04004970 struct btrfs_path *path = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -04004971 struct btrfs_root *root = BTRFS_I(inode)->root;
4972 struct btrfs_file_extent_item *item;
Chris Mason5f39d392007-10-15 16:14:19 -04004973 struct extent_buffer *leaf;
4974 struct btrfs_key found_key;
Chris Masona52d9a82007-08-27 16:49:44 -04004975 struct extent_map *em = NULL;
4976 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Masond1310b22008-01-24 16:13:08 -05004977 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04004978 struct btrfs_trans_handle *trans = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08004979 int compress_type;
Chris Masona52d9a82007-08-27 16:49:44 -04004980
Chris Masona52d9a82007-08-27 16:49:44 -04004981again:
Chris Mason890871b2009-09-02 16:24:52 -04004982 read_lock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05004983 em = lookup_extent_mapping(em_tree, start, len);
Chris Masona061fc82008-05-07 11:43:44 -04004984 if (em)
4985 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason890871b2009-09-02 16:24:52 -04004986 read_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05004987
Chris Masona52d9a82007-08-27 16:49:44 -04004988 if (em) {
Chris Masone1c4b742008-04-22 13:26:46 -04004989 if (em->start > start || em->start + em->len <= start)
4990 free_extent_map(em);
4991 else if (em->block_start == EXTENT_MAP_INLINE && page)
Chris Mason70dec802008-01-29 09:59:12 -05004992 free_extent_map(em);
4993 else
4994 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004995 }
David Sterba172ddd62011-04-21 00:48:27 +02004996 em = alloc_extent_map();
Chris Masona52d9a82007-08-27 16:49:44 -04004997 if (!em) {
Chris Masond1310b22008-01-24 16:13:08 -05004998 err = -ENOMEM;
4999 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04005000 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04005001 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Masond1310b22008-01-24 16:13:08 -05005002 em->start = EXTENT_MAP_HOLE;
Chris Mason445a6942008-11-10 11:53:33 -05005003 em->orig_start = EXTENT_MAP_HOLE;
Chris Masond1310b22008-01-24 16:13:08 -05005004 em->len = (u64)-1;
Chris Masonc8b97812008-10-29 14:49:59 -04005005 em->block_len = (u64)-1;
Chris Masonf4219502008-07-22 11:18:09 -04005006
5007 if (!path) {
5008 path = btrfs_alloc_path();
Josef Bacik026fd312011-05-13 10:32:11 -04005009 if (!path) {
5010 err = -ENOMEM;
5011 goto out;
5012 }
5013 /*
5014 * Chances are we'll be called again, so go ahead and do
5015 * readahead
5016 */
5017 path->reada = 1;
Chris Masonf4219502008-07-22 11:18:09 -04005018 }
5019
Chris Mason179e29e2007-11-01 11:28:41 -04005020 ret = btrfs_lookup_file_extent(trans, root, path,
5021 objectid, start, trans != NULL);
Chris Masona52d9a82007-08-27 16:49:44 -04005022 if (ret < 0) {
5023 err = ret;
5024 goto out;
5025 }
5026
5027 if (ret != 0) {
5028 if (path->slots[0] == 0)
5029 goto not_found;
5030 path->slots[0]--;
5031 }
5032
Chris Mason5f39d392007-10-15 16:14:19 -04005033 leaf = path->nodes[0];
5034 item = btrfs_item_ptr(leaf, path->slots[0],
Chris Masona52d9a82007-08-27 16:49:44 -04005035 struct btrfs_file_extent_item);
Chris Masona52d9a82007-08-27 16:49:44 -04005036 /* are we inside the extent that was found? */
Chris Mason5f39d392007-10-15 16:14:19 -04005037 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5038 found_type = btrfs_key_type(&found_key);
5039 if (found_key.objectid != objectid ||
Chris Masona52d9a82007-08-27 16:49:44 -04005040 found_type != BTRFS_EXTENT_DATA_KEY) {
5041 goto not_found;
5042 }
5043
Chris Mason5f39d392007-10-15 16:14:19 -04005044 found_type = btrfs_file_extent_type(leaf, item);
5045 extent_start = found_key.offset;
Li Zefan261507a02010-12-17 14:21:50 +08005046 compress_type = btrfs_file_extent_compression(leaf, item);
Yan Zhengd899e052008-10-30 14:25:28 -04005047 if (found_type == BTRFS_FILE_EXTENT_REG ||
5048 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04005049 extent_end = extent_start +
Chris Masondb945352007-10-15 16:15:53 -04005050 btrfs_file_extent_num_bytes(leaf, item);
Yan Zheng9036c102008-10-30 14:19:41 -04005051 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5052 size_t size;
5053 size = btrfs_file_extent_inline_len(leaf, item);
5054 extent_end = (extent_start + size + root->sectorsize - 1) &
5055 ~((u64)root->sectorsize - 1);
5056 }
5057
5058 if (start >= extent_end) {
5059 path->slots[0]++;
5060 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5061 ret = btrfs_next_leaf(root, path);
5062 if (ret < 0) {
5063 err = ret;
5064 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04005065 }
Yan Zheng9036c102008-10-30 14:19:41 -04005066 if (ret > 0)
5067 goto not_found;
5068 leaf = path->nodes[0];
Chris Masona52d9a82007-08-27 16:49:44 -04005069 }
Yan Zheng9036c102008-10-30 14:19:41 -04005070 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5071 if (found_key.objectid != objectid ||
5072 found_key.type != BTRFS_EXTENT_DATA_KEY)
5073 goto not_found;
5074 if (start + len <= found_key.offset)
5075 goto not_found;
5076 em->start = start;
5077 em->len = found_key.offset - start;
5078 goto not_found_em;
5079 }
5080
Yan Zhengd899e052008-10-30 14:25:28 -04005081 if (found_type == BTRFS_FILE_EXTENT_REG ||
5082 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng9036c102008-10-30 14:19:41 -04005083 em->start = extent_start;
5084 em->len = extent_end - extent_start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -05005085 em->orig_start = extent_start -
5086 btrfs_file_extent_offset(leaf, item);
Chris Masondb945352007-10-15 16:15:53 -04005087 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5088 if (bytenr == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04005089 em->block_start = EXTENT_MAP_HOLE;
Chris Masona52d9a82007-08-27 16:49:44 -04005090 goto insert;
5091 }
Li Zefan261507a02010-12-17 14:21:50 +08005092 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04005093 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Li Zefan261507a02010-12-17 14:21:50 +08005094 em->compress_type = compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04005095 em->block_start = bytenr;
5096 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5097 item);
5098 } else {
5099 bytenr += btrfs_file_extent_offset(leaf, item);
5100 em->block_start = bytenr;
5101 em->block_len = em->len;
Yan Zhengd899e052008-10-30 14:25:28 -04005102 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5103 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -04005104 }
Chris Masona52d9a82007-08-27 16:49:44 -04005105 goto insert;
5106 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04005107 unsigned long ptr;
Chris Masona52d9a82007-08-27 16:49:44 -04005108 char *map;
Chris Mason3326d1b2007-10-15 16:18:25 -04005109 size_t size;
5110 size_t extent_offset;
5111 size_t copy_size;
Chris Masona52d9a82007-08-27 16:49:44 -04005112
Chris Masona52d9a82007-08-27 16:49:44 -04005113 em->block_start = EXTENT_MAP_INLINE;
Chris Masonc8b97812008-10-29 14:49:59 -04005114 if (!page || create) {
Yan689f9342007-10-29 11:41:07 -04005115 em->start = extent_start;
Yan Zheng9036c102008-10-30 14:19:41 -04005116 em->len = extent_end - extent_start;
Yan689f9342007-10-29 11:41:07 -04005117 goto out;
5118 }
5119
Yan Zheng9036c102008-10-30 14:19:41 -04005120 size = btrfs_file_extent_inline_len(leaf, item);
5121 extent_offset = page_offset(page) + pg_offset - extent_start;
Chris Mason70dec802008-01-29 09:59:12 -05005122 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
Yan689f9342007-10-29 11:41:07 -04005123 size - extent_offset);
Chris Mason3326d1b2007-10-15 16:18:25 -04005124 em->start = extent_start + extent_offset;
Chris Mason70dec802008-01-29 09:59:12 -05005125 em->len = (copy_size + root->sectorsize - 1) &
5126 ~((u64)root->sectorsize - 1);
Yan Zhengff5b7ee2008-11-10 07:34:43 -05005127 em->orig_start = EXTENT_MAP_INLINE;
Li Zefan261507a02010-12-17 14:21:50 +08005128 if (compress_type) {
Chris Masonc8b97812008-10-29 14:49:59 -04005129 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Li Zefan261507a02010-12-17 14:21:50 +08005130 em->compress_type = compress_type;
5131 }
Yan689f9342007-10-29 11:41:07 -04005132 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
Chris Mason179e29e2007-11-01 11:28:41 -04005133 if (create == 0 && !PageUptodate(page)) {
Li Zefan261507a02010-12-17 14:21:50 +08005134 if (btrfs_file_extent_compression(leaf, item) !=
5135 BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04005136 ret = uncompress_inline(path, inode, page,
5137 pg_offset,
5138 extent_offset, item);
5139 BUG_ON(ret);
5140 } else {
5141 map = kmap(page);
5142 read_extent_buffer(leaf, map + pg_offset, ptr,
5143 copy_size);
Chris Mason93c82d52009-09-11 12:36:29 -04005144 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5145 memset(map + pg_offset + copy_size, 0,
5146 PAGE_CACHE_SIZE - pg_offset -
5147 copy_size);
5148 }
Chris Masonc8b97812008-10-29 14:49:59 -04005149 kunmap(page);
5150 }
Chris Mason179e29e2007-11-01 11:28:41 -04005151 flush_dcache_page(page);
5152 } else if (create && PageUptodate(page)) {
Jan Schmidt6bf7e082011-12-01 14:35:19 +01005153 BUG();
Chris Mason179e29e2007-11-01 11:28:41 -04005154 if (!trans) {
5155 kunmap(page);
5156 free_extent_map(em);
5157 em = NULL;
Chris Masonff5714c2011-05-28 07:00:39 -04005158
David Sterbab3b4aa72011-04-21 01:20:15 +02005159 btrfs_release_path(path);
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005160 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04005161
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005162 if (IS_ERR(trans))
5163 return ERR_CAST(trans);
Chris Mason179e29e2007-11-01 11:28:41 -04005164 goto again;
5165 }
Chris Masonc8b97812008-10-29 14:49:59 -04005166 map = kmap(page);
Chris Mason70dec802008-01-29 09:59:12 -05005167 write_extent_buffer(leaf, map + pg_offset, ptr,
Chris Mason179e29e2007-11-01 11:28:41 -04005168 copy_size);
Chris Masonc8b97812008-10-29 14:49:59 -04005169 kunmap(page);
Chris Mason179e29e2007-11-01 11:28:41 -04005170 btrfs_mark_buffer_dirty(leaf);
Chris Masona52d9a82007-08-27 16:49:44 -04005171 }
Chris Masond1310b22008-01-24 16:13:08 -05005172 set_extent_uptodate(io_tree, em->start,
Arne Jansen507903b2011-04-06 10:02:20 +00005173 extent_map_end(em) - 1, NULL, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04005174 goto insert;
5175 } else {
Chris Masond3977122009-01-05 21:25:51 -05005176 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
Chris Masona52d9a82007-08-27 16:49:44 -04005177 WARN_ON(1);
5178 }
5179not_found:
5180 em->start = start;
Chris Masond1310b22008-01-24 16:13:08 -05005181 em->len = len;
Chris Masona52d9a82007-08-27 16:49:44 -04005182not_found_em:
Chris Mason5f39d392007-10-15 16:14:19 -04005183 em->block_start = EXTENT_MAP_HOLE;
Yan Zheng9036c102008-10-30 14:19:41 -04005184 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -04005185insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02005186 btrfs_release_path(path);
Chris Masond1310b22008-01-24 16:13:08 -05005187 if (em->start > start || extent_map_end(em) <= start) {
Chris Masond3977122009-01-05 21:25:51 -05005188 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5189 "[%llu %llu]\n", (unsigned long long)em->start,
5190 (unsigned long long)em->len,
5191 (unsigned long long)start,
5192 (unsigned long long)len);
Chris Masona52d9a82007-08-27 16:49:44 -04005193 err = -EIO;
5194 goto out;
5195 }
Chris Masond1310b22008-01-24 16:13:08 -05005196
5197 err = 0;
Chris Mason890871b2009-09-02 16:24:52 -04005198 write_lock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04005199 ret = add_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -04005200 /* it is possible that someone inserted the extent into the tree
5201 * while we had the lock dropped. It is also possible that
5202 * an overlapping map exists in the tree
5203 */
Chris Masona52d9a82007-08-27 16:49:44 -04005204 if (ret == -EEXIST) {
Chris Mason3b951512008-04-17 11:29:12 -04005205 struct extent_map *existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005206
5207 ret = 0;
5208
Chris Mason3b951512008-04-17 11:29:12 -04005209 existing = lookup_extent_mapping(em_tree, start, len);
Chris Masone1c4b742008-04-22 13:26:46 -04005210 if (existing && (existing->start > start ||
5211 existing->start + existing->len <= start)) {
5212 free_extent_map(existing);
5213 existing = NULL;
5214 }
Chris Mason3b951512008-04-17 11:29:12 -04005215 if (!existing) {
5216 existing = lookup_extent_mapping(em_tree, em->start,
5217 em->len);
5218 if (existing) {
5219 err = merge_extent_mapping(em_tree, existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04005220 em, start,
5221 root->sectorsize);
Chris Mason3b951512008-04-17 11:29:12 -04005222 free_extent_map(existing);
5223 if (err) {
5224 free_extent_map(em);
5225 em = NULL;
5226 }
5227 } else {
5228 err = -EIO;
Chris Mason3b951512008-04-17 11:29:12 -04005229 free_extent_map(em);
5230 em = NULL;
5231 }
5232 } else {
5233 free_extent_map(em);
5234 em = existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005235 err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04005236 }
Chris Masona52d9a82007-08-27 16:49:44 -04005237 }
Chris Mason890871b2009-09-02 16:24:52 -04005238 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04005239out:
liubo1abe9b82011-03-24 11:18:59 +00005240
5241 trace_btrfs_get_extent(root, em);
5242
Chris Masonf4219502008-07-22 11:18:09 -04005243 if (path)
5244 btrfs_free_path(path);
Chris Masona52d9a82007-08-27 16:49:44 -04005245 if (trans) {
5246 ret = btrfs_end_transaction(trans, root);
Chris Masond3977122009-01-05 21:25:51 -05005247 if (!err)
Chris Masona52d9a82007-08-27 16:49:44 -04005248 err = ret;
5249 }
Chris Masona52d9a82007-08-27 16:49:44 -04005250 if (err) {
5251 free_extent_map(em);
Chris Masona52d9a82007-08-27 16:49:44 -04005252 return ERR_PTR(err);
5253 }
5254 return em;
5255}
5256
Chris Masonec29ed52011-02-23 16:23:20 -05005257struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5258 size_t pg_offset, u64 start, u64 len,
5259 int create)
5260{
5261 struct extent_map *em;
5262 struct extent_map *hole_em = NULL;
5263 u64 range_start = start;
5264 u64 end;
5265 u64 found;
5266 u64 found_end;
5267 int err = 0;
5268
5269 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5270 if (IS_ERR(em))
5271 return em;
5272 if (em) {
5273 /*
5274 * if our em maps to a hole, there might
5275 * actually be delalloc bytes behind it
5276 */
5277 if (em->block_start != EXTENT_MAP_HOLE)
5278 return em;
5279 else
5280 hole_em = em;
5281 }
5282
5283 /* check to see if we've wrapped (len == -1 or similar) */
5284 end = start + len;
5285 if (end < start)
5286 end = (u64)-1;
5287 else
5288 end -= 1;
5289
5290 em = NULL;
5291
5292 /* ok, we didn't find anything, lets look for delalloc */
5293 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5294 end, len, EXTENT_DELALLOC, 1);
5295 found_end = range_start + found;
5296 if (found_end < range_start)
5297 found_end = (u64)-1;
5298
5299 /*
5300 * we didn't find anything useful, return
5301 * the original results from get_extent()
5302 */
5303 if (range_start > end || found_end <= start) {
5304 em = hole_em;
5305 hole_em = NULL;
5306 goto out;
5307 }
5308
5309 /* adjust the range_start to make sure it doesn't
5310 * go backwards from the start they passed in
5311 */
5312 range_start = max(start,range_start);
5313 found = found_end - range_start;
5314
5315 if (found > 0) {
5316 u64 hole_start = start;
5317 u64 hole_len = len;
5318
David Sterba172ddd62011-04-21 00:48:27 +02005319 em = alloc_extent_map();
Chris Masonec29ed52011-02-23 16:23:20 -05005320 if (!em) {
5321 err = -ENOMEM;
5322 goto out;
5323 }
5324 /*
5325 * when btrfs_get_extent can't find anything it
5326 * returns one huge hole
5327 *
5328 * make sure what it found really fits our range, and
5329 * adjust to make sure it is based on the start from
5330 * the caller
5331 */
5332 if (hole_em) {
5333 u64 calc_end = extent_map_end(hole_em);
5334
5335 if (calc_end <= start || (hole_em->start > end)) {
5336 free_extent_map(hole_em);
5337 hole_em = NULL;
5338 } else {
5339 hole_start = max(hole_em->start, start);
5340 hole_len = calc_end - hole_start;
5341 }
5342 }
5343 em->bdev = NULL;
5344 if (hole_em && range_start > hole_start) {
5345 /* our hole starts before our delalloc, so we
5346 * have to return just the parts of the hole
5347 * that go until the delalloc starts
5348 */
5349 em->len = min(hole_len,
5350 range_start - hole_start);
5351 em->start = hole_start;
5352 em->orig_start = hole_start;
5353 /*
5354 * don't adjust block start at all,
5355 * it is fixed at EXTENT_MAP_HOLE
5356 */
5357 em->block_start = hole_em->block_start;
5358 em->block_len = hole_len;
5359 } else {
5360 em->start = range_start;
5361 em->len = found;
5362 em->orig_start = range_start;
5363 em->block_start = EXTENT_MAP_DELALLOC;
5364 em->block_len = found;
5365 }
5366 } else if (hole_em) {
5367 return hole_em;
5368 }
5369out:
5370
5371 free_extent_map(hole_em);
5372 if (err) {
5373 free_extent_map(em);
5374 return ERR_PTR(err);
5375 }
5376 return em;
5377}
5378
Josef Bacik4b46fce2010-05-23 11:00:55 -04005379static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
Josef Bacik16d299a2011-04-06 14:53:07 -04005380 struct extent_map *em,
Josef Bacik4b46fce2010-05-23 11:00:55 -04005381 u64 start, u64 len)
5382{
5383 struct btrfs_root *root = BTRFS_I(inode)->root;
5384 struct btrfs_trans_handle *trans;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005385 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5386 struct btrfs_key ins;
5387 u64 alloc_hint;
5388 int ret;
Josef Bacik16d299a2011-04-06 14:53:07 -04005389 bool insert = false;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005390
Josef Bacik16d299a2011-04-06 14:53:07 -04005391 /*
5392 * Ok if the extent map we looked up is a hole and is for the exact
5393 * range we want, there is no reason to allocate a new one, however if
5394 * it is not right then we need to free this one and drop the cache for
5395 * our range.
5396 */
5397 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5398 em->len != len) {
5399 free_extent_map(em);
5400 em = NULL;
5401 insert = true;
5402 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5403 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04005404
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005405 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005406 if (IS_ERR(trans))
5407 return ERR_CAST(trans);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005408
Chris Mason4cb53002011-05-24 15:35:30 -04005409 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5410 btrfs_add_inode_defrag(trans, inode);
5411
Josef Bacik4b46fce2010-05-23 11:00:55 -04005412 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5413
5414 alloc_hint = get_extent_allocation_hint(inode, start, len);
5415 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
Josef Bacik81c9ad22012-01-18 10:56:06 -05005416 alloc_hint, &ins, 1);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005417 if (ret) {
5418 em = ERR_PTR(ret);
5419 goto out;
5420 }
5421
Josef Bacik4b46fce2010-05-23 11:00:55 -04005422 if (!em) {
David Sterba172ddd62011-04-21 00:48:27 +02005423 em = alloc_extent_map();
Josef Bacik16d299a2011-04-06 14:53:07 -04005424 if (!em) {
5425 em = ERR_PTR(-ENOMEM);
5426 goto out;
5427 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04005428 }
5429
5430 em->start = start;
5431 em->orig_start = em->start;
5432 em->len = ins.offset;
5433
5434 em->block_start = ins.objectid;
5435 em->block_len = ins.offset;
5436 em->bdev = root->fs_info->fs_devices->latest_bdev;
Josef Bacik16d299a2011-04-06 14:53:07 -04005437
5438 /*
5439 * We need to do this because if we're using the original em we searched
5440 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5441 */
5442 em->flags = 0;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005443 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5444
Josef Bacik16d299a2011-04-06 14:53:07 -04005445 while (insert) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04005446 write_lock(&em_tree->lock);
5447 ret = add_extent_mapping(em_tree, em);
5448 write_unlock(&em_tree->lock);
5449 if (ret != -EEXIST)
5450 break;
5451 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5452 }
5453
5454 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5455 ins.offset, ins.offset, 0);
5456 if (ret) {
5457 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5458 em = ERR_PTR(ret);
5459 }
5460out:
5461 btrfs_end_transaction(trans, root);
5462 return em;
5463}
5464
Chris Mason46bfbb52010-05-26 11:04:10 -04005465/*
5466 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5467 * block must be cow'd
5468 */
5469static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5470 struct inode *inode, u64 offset, u64 len)
5471{
5472 struct btrfs_path *path;
5473 int ret;
5474 struct extent_buffer *leaf;
5475 struct btrfs_root *root = BTRFS_I(inode)->root;
5476 struct btrfs_file_extent_item *fi;
5477 struct btrfs_key key;
5478 u64 disk_bytenr;
5479 u64 backref_offset;
5480 u64 extent_end;
5481 u64 num_bytes;
5482 int slot;
5483 int found_type;
5484
5485 path = btrfs_alloc_path();
5486 if (!path)
5487 return -ENOMEM;
5488
Li Zefan33345d012011-04-20 10:31:50 +08005489 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
Chris Mason46bfbb52010-05-26 11:04:10 -04005490 offset, 0);
5491 if (ret < 0)
5492 goto out;
5493
5494 slot = path->slots[0];
5495 if (ret == 1) {
5496 if (slot == 0) {
5497 /* can't find the item, must cow */
5498 ret = 0;
5499 goto out;
5500 }
5501 slot--;
5502 }
5503 ret = 0;
5504 leaf = path->nodes[0];
5505 btrfs_item_key_to_cpu(leaf, &key, slot);
Li Zefan33345d012011-04-20 10:31:50 +08005506 if (key.objectid != btrfs_ino(inode) ||
Chris Mason46bfbb52010-05-26 11:04:10 -04005507 key.type != BTRFS_EXTENT_DATA_KEY) {
5508 /* not our file or wrong item type, must cow */
5509 goto out;
5510 }
5511
5512 if (key.offset > offset) {
5513 /* Wrong offset, must cow */
5514 goto out;
5515 }
5516
5517 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5518 found_type = btrfs_file_extent_type(leaf, fi);
5519 if (found_type != BTRFS_FILE_EXTENT_REG &&
5520 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5521 /* not a regular extent, must cow */
5522 goto out;
5523 }
5524 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5525 backref_offset = btrfs_file_extent_offset(leaf, fi);
5526
5527 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5528 if (extent_end < offset + len) {
5529 /* extent doesn't include our full range, must cow */
5530 goto out;
5531 }
5532
5533 if (btrfs_extent_readonly(root, disk_bytenr))
5534 goto out;
5535
5536 /*
5537 * look for other files referencing this extent, if we
5538 * find any we must cow
5539 */
Li Zefan33345d012011-04-20 10:31:50 +08005540 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
Chris Mason46bfbb52010-05-26 11:04:10 -04005541 key.offset - backref_offset, disk_bytenr))
5542 goto out;
5543
5544 /*
5545 * adjust disk_bytenr and num_bytes to cover just the bytes
5546 * in this extent we are about to write. If there
5547 * are any csums in that range we have to cow in order
5548 * to keep the csums correct
5549 */
5550 disk_bytenr += backref_offset;
5551 disk_bytenr += offset - key.offset;
5552 num_bytes = min(offset + len, extent_end) - offset;
5553 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5554 goto out;
5555 /*
5556 * all of the above have passed, it is safe to overwrite this extent
5557 * without cow
5558 */
5559 ret = 1;
5560out:
5561 btrfs_free_path(path);
5562 return ret;
5563}
5564
Josef Bacik4b46fce2010-05-23 11:00:55 -04005565static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5566 struct buffer_head *bh_result, int create)
5567{
5568 struct extent_map *em;
5569 struct btrfs_root *root = BTRFS_I(inode)->root;
5570 u64 start = iblock << inode->i_blkbits;
5571 u64 len = bh_result->b_size;
Chris Mason46bfbb52010-05-26 11:04:10 -04005572 struct btrfs_trans_handle *trans;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005573
5574 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5575 if (IS_ERR(em))
5576 return PTR_ERR(em);
5577
5578 /*
5579 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5580 * io. INLINE is special, and we could probably kludge it in here, but
5581 * it's still buffered so for safety lets just fall back to the generic
5582 * buffered path.
5583 *
5584 * For COMPRESSED we _have_ to read the entire extent in so we can
5585 * decompress it, so there will be buffering required no matter what we
5586 * do, so go ahead and fallback to buffered.
5587 *
5588 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5589 * to buffered IO. Don't blame me, this is the price we pay for using
5590 * the generic code.
5591 */
5592 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5593 em->block_start == EXTENT_MAP_INLINE) {
5594 free_extent_map(em);
5595 return -ENOTBLK;
5596 }
5597
5598 /* Just a good old fashioned hole, return */
5599 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5600 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5601 free_extent_map(em);
5602 /* DIO will do one hole at a time, so just unlock a sector */
5603 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5604 start + root->sectorsize - 1, GFP_NOFS);
5605 return 0;
5606 }
5607
5608 /*
5609 * We don't allocate a new extent in the following cases
5610 *
5611 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5612 * existing extent.
5613 * 2) The extent is marked as PREALLOC. We're good to go here and can
5614 * just use the extent.
5615 *
5616 */
Chris Mason46bfbb52010-05-26 11:04:10 -04005617 if (!create) {
5618 len = em->len - (start - em->start);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005619 goto map;
Chris Mason46bfbb52010-05-26 11:04:10 -04005620 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04005621
5622 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5623 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5624 em->block_start != EXTENT_MAP_HOLE)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04005625 int type;
5626 int ret;
Chris Mason46bfbb52010-05-26 11:04:10 -04005627 u64 block_start;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005628
5629 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5630 type = BTRFS_ORDERED_PREALLOC;
5631 else
5632 type = BTRFS_ORDERED_NOCOW;
Chris Mason46bfbb52010-05-26 11:04:10 -04005633 len = min(len, em->len - (start - em->start));
Josef Bacik4b46fce2010-05-23 11:00:55 -04005634 block_start = em->block_start + (start - em->start);
Chris Mason46bfbb52010-05-26 11:04:10 -04005635
5636 /*
5637 * we're not going to log anything, but we do need
5638 * to make sure the current transaction stays open
5639 * while we look for nocow cross refs
5640 */
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005641 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005642 if (IS_ERR(trans))
Chris Mason46bfbb52010-05-26 11:04:10 -04005643 goto must_cow;
5644
5645 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5646 ret = btrfs_add_ordered_extent_dio(inode, start,
5647 block_start, len, len, type);
5648 btrfs_end_transaction(trans, root);
5649 if (ret) {
5650 free_extent_map(em);
5651 return ret;
5652 }
5653 goto unlock;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005654 }
Chris Mason46bfbb52010-05-26 11:04:10 -04005655 btrfs_end_transaction(trans, root);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005656 }
Chris Mason46bfbb52010-05-26 11:04:10 -04005657must_cow:
5658 /*
5659 * this will cow the extent, reset the len in case we changed
5660 * it above
5661 */
5662 len = bh_result->b_size;
Josef Bacik16d299a2011-04-06 14:53:07 -04005663 em = btrfs_new_extent_direct(inode, em, start, len);
Chris Mason46bfbb52010-05-26 11:04:10 -04005664 if (IS_ERR(em))
5665 return PTR_ERR(em);
5666 len = min(len, em->len - (start - em->start));
5667unlock:
Chris Mason4845e442010-05-25 20:56:50 -04005668 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5669 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5670 0, NULL, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005671map:
5672 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5673 inode->i_blkbits;
Chris Mason46bfbb52010-05-26 11:04:10 -04005674 bh_result->b_size = len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005675 bh_result->b_bdev = em->bdev;
5676 set_buffer_mapped(bh_result);
5677 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5678 set_buffer_new(bh_result);
5679
5680 free_extent_map(em);
5681
5682 return 0;
5683}
5684
5685struct btrfs_dio_private {
5686 struct inode *inode;
5687 u64 logical_offset;
5688 u64 disk_bytenr;
5689 u64 bytes;
5690 u32 *csums;
5691 void *private;
Miao Xiee65e1532010-11-22 03:04:43 +00005692
5693 /* number of bios pending for this dio */
5694 atomic_t pending_bios;
5695
5696 /* IO errors */
5697 int errors;
5698
5699 struct bio *orig_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005700};
5701
5702static void btrfs_endio_direct_read(struct bio *bio, int err)
5703{
Miao Xiee65e1532010-11-22 03:04:43 +00005704 struct btrfs_dio_private *dip = bio->bi_private;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005705 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5706 struct bio_vec *bvec = bio->bi_io_vec;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005707 struct inode *inode = dip->inode;
5708 struct btrfs_root *root = BTRFS_I(inode)->root;
5709 u64 start;
5710 u32 *private = dip->csums;
5711
5712 start = dip->logical_offset;
5713 do {
5714 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5715 struct page *page = bvec->bv_page;
5716 char *kaddr;
5717 u32 csum = ~(u32)0;
5718 unsigned long flags;
5719
5720 local_irq_save(flags);
5721 kaddr = kmap_atomic(page, KM_IRQ0);
5722 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5723 csum, bvec->bv_len);
5724 btrfs_csum_final(csum, (char *)&csum);
5725 kunmap_atomic(kaddr, KM_IRQ0);
5726 local_irq_restore(flags);
5727
5728 flush_dcache_page(bvec->bv_page);
5729 if (csum != *private) {
Li Zefan33345d012011-04-20 10:31:50 +08005730 printk(KERN_ERR "btrfs csum failed ino %llu off"
Josef Bacik4b46fce2010-05-23 11:00:55 -04005731 " %llu csum %u private %u\n",
Li Zefan33345d012011-04-20 10:31:50 +08005732 (unsigned long long)btrfs_ino(inode),
5733 (unsigned long long)start,
Josef Bacik4b46fce2010-05-23 11:00:55 -04005734 csum, *private);
5735 err = -EIO;
5736 }
5737 }
5738
5739 start += bvec->bv_len;
5740 private++;
5741 bvec++;
5742 } while (bvec <= bvec_end);
5743
5744 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5745 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5746 bio->bi_private = dip->private;
5747
5748 kfree(dip->csums);
5749 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04005750
5751 /* If we had a csum failure make sure to clear the uptodate flag */
5752 if (err)
5753 clear_bit(BIO_UPTODATE, &bio->bi_flags);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005754 dio_end_io(bio, err);
5755}
5756
5757static void btrfs_endio_direct_write(struct bio *bio, int err)
5758{
5759 struct btrfs_dio_private *dip = bio->bi_private;
5760 struct inode *inode = dip->inode;
5761 struct btrfs_root *root = BTRFS_I(inode)->root;
5762 struct btrfs_trans_handle *trans;
5763 struct btrfs_ordered_extent *ordered = NULL;
5764 struct extent_state *cached_state = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05005765 u64 ordered_offset = dip->logical_offset;
5766 u64 ordered_bytes = dip->bytes;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005767 int ret;
5768
5769 if (err)
5770 goto out_done;
Chris Mason163cf092010-11-28 19:56:33 -05005771again:
5772 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5773 &ordered_offset,
5774 ordered_bytes);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005775 if (!ret)
Chris Mason163cf092010-11-28 19:56:33 -05005776 goto out_test;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005777
5778 BUG_ON(!ordered);
5779
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005780 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005781 if (IS_ERR(trans)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04005782 err = -ENOMEM;
5783 goto out;
5784 }
5785 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5786
5787 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5788 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5789 if (!ret)
Chris Mason21151332011-11-10 20:39:08 -05005790 err = btrfs_update_inode_fallback(trans, root, inode);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005791 goto out;
5792 }
5793
5794 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5795 ordered->file_offset + ordered->len - 1, 0,
5796 &cached_state, GFP_NOFS);
5797
5798 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5799 ret = btrfs_mark_extent_written(trans, inode,
5800 ordered->file_offset,
5801 ordered->file_offset +
5802 ordered->len);
5803 if (ret) {
5804 err = ret;
5805 goto out_unlock;
5806 }
5807 } else {
5808 ret = insert_reserved_file_extent(trans, inode,
5809 ordered->file_offset,
5810 ordered->start,
5811 ordered->disk_len,
5812 ordered->len,
5813 ordered->len,
5814 0, 0, 0,
5815 BTRFS_FILE_EXTENT_REG);
5816 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5817 ordered->file_offset, ordered->len);
5818 if (ret) {
5819 err = ret;
5820 WARN_ON(1);
5821 goto out_unlock;
5822 }
5823 }
5824
5825 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
Josef Bacik1ef30be2011-04-05 19:25:36 -04005826 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
Miao Xiea39f7522011-09-11 10:52:25 -04005827 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
Chris Mason21151332011-11-10 20:39:08 -05005828 btrfs_update_inode_fallback(trans, root, inode);
Josef Bacik1ef30be2011-04-05 19:25:36 -04005829 ret = 0;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005830out_unlock:
5831 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5832 ordered->file_offset + ordered->len - 1,
5833 &cached_state, GFP_NOFS);
5834out:
5835 btrfs_delalloc_release_metadata(inode, ordered->len);
5836 btrfs_end_transaction(trans, root);
Chris Mason163cf092010-11-28 19:56:33 -05005837 ordered_offset = ordered->file_offset + ordered->len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04005838 btrfs_put_ordered_extent(ordered);
5839 btrfs_put_ordered_extent(ordered);
Chris Mason163cf092010-11-28 19:56:33 -05005840
5841out_test:
5842 /*
5843 * our bio might span multiple ordered extents. If we haven't
5844 * completed the accounting for the whole dio, go back and try again
5845 */
5846 if (ordered_offset < dip->logical_offset + dip->bytes) {
5847 ordered_bytes = dip->logical_offset + dip->bytes -
5848 ordered_offset;
5849 goto again;
5850 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04005851out_done:
5852 bio->bi_private = dip->private;
5853
5854 kfree(dip->csums);
5855 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04005856
5857 /* If we had an error make sure to clear the uptodate flag */
5858 if (err)
5859 clear_bit(BIO_UPTODATE, &bio->bi_flags);
Josef Bacik4b46fce2010-05-23 11:00:55 -04005860 dio_end_io(bio, err);
5861}
5862
Chris Masoneaf25d92010-05-25 09:48:28 -04005863static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5864 struct bio *bio, int mirror_num,
5865 unsigned long bio_flags, u64 offset)
5866{
5867 int ret;
5868 struct btrfs_root *root = BTRFS_I(inode)->root;
5869 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5870 BUG_ON(ret);
5871 return 0;
5872}
5873
Miao Xiee65e1532010-11-22 03:04:43 +00005874static void btrfs_end_dio_bio(struct bio *bio, int err)
5875{
5876 struct btrfs_dio_private *dip = bio->bi_private;
5877
5878 if (err) {
Li Zefan33345d012011-04-20 10:31:50 +08005879 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
Jan Beulich3dd14622010-12-07 14:54:09 +00005880 "sector %#Lx len %u err no %d\n",
Li Zefan33345d012011-04-20 10:31:50 +08005881 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
Jan Beulich3dd14622010-12-07 14:54:09 +00005882 (unsigned long long)bio->bi_sector, bio->bi_size, err);
Miao Xiee65e1532010-11-22 03:04:43 +00005883 dip->errors = 1;
5884
5885 /*
5886 * before atomic variable goto zero, we must make sure
5887 * dip->errors is perceived to be set.
5888 */
5889 smp_mb__before_atomic_dec();
5890 }
5891
5892 /* if there are more bios still pending for this dio, just exit */
5893 if (!atomic_dec_and_test(&dip->pending_bios))
5894 goto out;
5895
5896 if (dip->errors)
5897 bio_io_error(dip->orig_bio);
5898 else {
5899 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5900 bio_endio(dip->orig_bio, 0);
5901 }
5902out:
5903 bio_put(bio);
5904}
5905
5906static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5907 u64 first_sector, gfp_t gfp_flags)
5908{
5909 int nr_vecs = bio_get_nr_vecs(bdev);
5910 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5911}
5912
5913static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5914 int rw, u64 file_offset, int skip_sum,
Josef Bacik1ae39932011-04-06 14:41:34 -04005915 u32 *csums, int async_submit)
Miao Xiee65e1532010-11-22 03:04:43 +00005916{
5917 int write = rw & REQ_WRITE;
5918 struct btrfs_root *root = BTRFS_I(inode)->root;
5919 int ret;
5920
5921 bio_get(bio);
5922 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5923 if (ret)
5924 goto err;
5925
Josef Bacik1ae39932011-04-06 14:41:34 -04005926 if (skip_sum)
5927 goto map;
5928
5929 if (write && async_submit) {
Miao Xiee65e1532010-11-22 03:04:43 +00005930 ret = btrfs_wq_submit_bio(root->fs_info,
5931 inode, rw, bio, 0, 0,
5932 file_offset,
5933 __btrfs_submit_bio_start_direct_io,
5934 __btrfs_submit_bio_done);
5935 goto err;
Josef Bacik1ae39932011-04-06 14:41:34 -04005936 } else if (write) {
5937 /*
5938 * If we aren't doing async submit, calculate the csum of the
5939 * bio now.
5940 */
5941 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5942 if (ret)
5943 goto err;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00005944 } else if (!skip_sum) {
5945 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
Miao Xiee65e1532010-11-22 03:04:43 +00005946 file_offset, csums);
Tsutomu Itohc2db1072011-03-01 06:48:31 +00005947 if (ret)
5948 goto err;
5949 }
Miao Xiee65e1532010-11-22 03:04:43 +00005950
Josef Bacik1ae39932011-04-06 14:41:34 -04005951map:
5952 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00005953err:
5954 bio_put(bio);
5955 return ret;
5956}
5957
5958static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5959 int skip_sum)
5960{
5961 struct inode *inode = dip->inode;
5962 struct btrfs_root *root = BTRFS_I(inode)->root;
5963 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5964 struct bio *bio;
5965 struct bio *orig_bio = dip->orig_bio;
5966 struct bio_vec *bvec = orig_bio->bi_io_vec;
5967 u64 start_sector = orig_bio->bi_sector;
5968 u64 file_offset = dip->logical_offset;
5969 u64 submit_len = 0;
5970 u64 map_length;
5971 int nr_pages = 0;
5972 u32 *csums = dip->csums;
5973 int ret = 0;
Josef Bacik1ae39932011-04-06 14:41:34 -04005974 int async_submit = 0;
Josef Bacik98bc3142011-03-22 11:00:46 -04005975 int write = rw & REQ_WRITE;
Miao Xiee65e1532010-11-22 03:04:43 +00005976
Miao Xiee65e1532010-11-22 03:04:43 +00005977 map_length = orig_bio->bi_size;
5978 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5979 &map_length, NULL, 0);
5980 if (ret) {
Josef Bacik64728bb2011-04-25 19:43:52 -04005981 bio_put(orig_bio);
Miao Xiee65e1532010-11-22 03:04:43 +00005982 return -EIO;
5983 }
5984
Josef Bacik02f57c72011-04-06 14:25:44 -04005985 if (map_length >= orig_bio->bi_size) {
5986 bio = orig_bio;
5987 goto submit;
5988 }
5989
Josef Bacik1ae39932011-04-06 14:41:34 -04005990 async_submit = 1;
Josef Bacik02f57c72011-04-06 14:25:44 -04005991 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5992 if (!bio)
5993 return -ENOMEM;
5994 bio->bi_private = dip;
5995 bio->bi_end_io = btrfs_end_dio_bio;
5996 atomic_inc(&dip->pending_bios);
5997
Miao Xiee65e1532010-11-22 03:04:43 +00005998 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5999 if (unlikely(map_length < submit_len + bvec->bv_len ||
6000 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6001 bvec->bv_offset) < bvec->bv_len)) {
6002 /*
6003 * inc the count before we submit the bio so
6004 * we know the end IO handler won't happen before
6005 * we inc the count. Otherwise, the dip might get freed
6006 * before we're done setting it up
6007 */
6008 atomic_inc(&dip->pending_bios);
6009 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6010 file_offset, skip_sum,
Josef Bacik1ae39932011-04-06 14:41:34 -04006011 csums, async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00006012 if (ret) {
6013 bio_put(bio);
6014 atomic_dec(&dip->pending_bios);
6015 goto out_err;
6016 }
6017
Josef Bacik98bc3142011-03-22 11:00:46 -04006018 /* Write's use the ordered csums */
6019 if (!write && !skip_sum)
Miao Xiee65e1532010-11-22 03:04:43 +00006020 csums = csums + nr_pages;
6021 start_sector += submit_len >> 9;
6022 file_offset += submit_len;
6023
6024 submit_len = 0;
6025 nr_pages = 0;
6026
6027 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6028 start_sector, GFP_NOFS);
6029 if (!bio)
6030 goto out_err;
6031 bio->bi_private = dip;
6032 bio->bi_end_io = btrfs_end_dio_bio;
6033
6034 map_length = orig_bio->bi_size;
6035 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6036 &map_length, NULL, 0);
6037 if (ret) {
6038 bio_put(bio);
6039 goto out_err;
6040 }
6041 } else {
6042 submit_len += bvec->bv_len;
6043 nr_pages ++;
6044 bvec++;
6045 }
6046 }
6047
Josef Bacik02f57c72011-04-06 14:25:44 -04006048submit:
Miao Xiee65e1532010-11-22 03:04:43 +00006049 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
Josef Bacik1ae39932011-04-06 14:41:34 -04006050 csums, async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00006051 if (!ret)
6052 return 0;
6053
6054 bio_put(bio);
6055out_err:
6056 dip->errors = 1;
6057 /*
6058 * before atomic variable goto zero, we must
6059 * make sure dip->errors is perceived to be set.
6060 */
6061 smp_mb__before_atomic_dec();
6062 if (atomic_dec_and_test(&dip->pending_bios))
6063 bio_io_error(dip->orig_bio);
6064
6065 /* bio_end_io() will handle error, so we needn't return it */
6066 return 0;
6067}
6068
Josef Bacik4b46fce2010-05-23 11:00:55 -04006069static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6070 loff_t file_offset)
6071{
6072 struct btrfs_root *root = BTRFS_I(inode)->root;
6073 struct btrfs_dio_private *dip;
6074 struct bio_vec *bvec = bio->bi_io_vec;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006075 int skip_sum;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02006076 int write = rw & REQ_WRITE;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006077 int ret = 0;
6078
6079 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6080
6081 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6082 if (!dip) {
6083 ret = -ENOMEM;
6084 goto free_ordered;
6085 }
6086 dip->csums = NULL;
6087
Josef Bacik98bc3142011-03-22 11:00:46 -04006088 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6089 if (!write && !skip_sum) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04006090 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6091 if (!dip->csums) {
Daniel J Bluemanb4966b72011-03-09 16:46:42 +00006092 kfree(dip);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006093 ret = -ENOMEM;
6094 goto free_ordered;
6095 }
6096 }
6097
6098 dip->private = bio->bi_private;
6099 dip->inode = inode;
6100 dip->logical_offset = file_offset;
6101
Josef Bacik4b46fce2010-05-23 11:00:55 -04006102 dip->bytes = 0;
6103 do {
6104 dip->bytes += bvec->bv_len;
6105 bvec++;
6106 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6107
Chris Mason46bfbb52010-05-26 11:04:10 -04006108 dip->disk_bytenr = (u64)bio->bi_sector << 9;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006109 bio->bi_private = dip;
Miao Xiee65e1532010-11-22 03:04:43 +00006110 dip->errors = 0;
6111 dip->orig_bio = bio;
6112 atomic_set(&dip->pending_bios, 0);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006113
6114 if (write)
6115 bio->bi_end_io = btrfs_endio_direct_write;
6116 else
6117 bio->bi_end_io = btrfs_endio_direct_read;
6118
Miao Xiee65e1532010-11-22 03:04:43 +00006119 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6120 if (!ret)
Chris Masoneaf25d92010-05-25 09:48:28 -04006121 return;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006122free_ordered:
6123 /*
6124 * If this is a write, we need to clean up the reserved space and kill
6125 * the ordered extent.
6126 */
6127 if (write) {
6128 struct btrfs_ordered_extent *ordered;
Josef Bacik955256f2010-11-19 09:41:10 -05006129 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006130 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6131 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6132 btrfs_free_reserved_extent(root, ordered->start,
6133 ordered->disk_len);
6134 btrfs_put_ordered_extent(ordered);
6135 btrfs_put_ordered_extent(ordered);
6136 }
6137 bio_endio(bio, ret);
6138}
6139
Chris Mason5a5f79b2010-05-26 21:33:37 -04006140static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6141 const struct iovec *iov, loff_t offset,
6142 unsigned long nr_segs)
6143{
6144 int seg;
Josef Bacika1b75f72011-04-08 15:51:18 +00006145 int i;
Chris Mason5a5f79b2010-05-26 21:33:37 -04006146 size_t size;
6147 unsigned long addr;
6148 unsigned blocksize_mask = root->sectorsize - 1;
6149 ssize_t retval = -EINVAL;
6150 loff_t end = offset;
6151
6152 if (offset & blocksize_mask)
6153 goto out;
6154
6155 /* Check the memory alignment. Blocks cannot straddle pages */
6156 for (seg = 0; seg < nr_segs; seg++) {
6157 addr = (unsigned long)iov[seg].iov_base;
6158 size = iov[seg].iov_len;
6159 end += size;
Josef Bacika1b75f72011-04-08 15:51:18 +00006160 if ((addr & blocksize_mask) || (size & blocksize_mask))
Chris Mason5a5f79b2010-05-26 21:33:37 -04006161 goto out;
Josef Bacika1b75f72011-04-08 15:51:18 +00006162
6163 /* If this is a write we don't need to check anymore */
6164 if (rw & WRITE)
6165 continue;
6166
6167 /*
6168 * Check to make sure we don't have duplicate iov_base's in this
6169 * iovec, if so return EINVAL, otherwise we'll get csum errors
6170 * when reading back.
6171 */
6172 for (i = seg + 1; i < nr_segs; i++) {
6173 if (iov[seg].iov_base == iov[i].iov_base)
6174 goto out;
6175 }
Chris Mason5a5f79b2010-05-26 21:33:37 -04006176 }
6177 retval = 0;
6178out:
6179 return retval;
6180}
Chris Mason16432982008-04-10 10:23:21 -04006181static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6182 const struct iovec *iov, loff_t offset,
6183 unsigned long nr_segs)
6184{
Josef Bacik4b46fce2010-05-23 11:00:55 -04006185 struct file *file = iocb->ki_filp;
6186 struct inode *inode = file->f_mapping->host;
6187 struct btrfs_ordered_extent *ordered;
Chris Mason4845e442010-05-25 20:56:50 -04006188 struct extent_state *cached_state = NULL;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006189 u64 lockstart, lockend;
6190 ssize_t ret;
Chris Mason4845e442010-05-25 20:56:50 -04006191 int writing = rw & WRITE;
6192 int write_bits = 0;
Chris Mason3f7c5792010-05-26 10:59:53 -04006193 size_t count = iov_length(iov, nr_segs);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006194
Chris Mason5a5f79b2010-05-26 21:33:37 -04006195 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6196 offset, nr_segs)) {
6197 return 0;
6198 }
6199
Josef Bacik4b46fce2010-05-23 11:00:55 -04006200 lockstart = offset;
Chris Mason3f7c5792010-05-26 10:59:53 -04006201 lockend = offset + count - 1;
6202
6203 if (writing) {
6204 ret = btrfs_delalloc_reserve_space(inode, count);
6205 if (ret)
6206 goto out;
6207 }
Chris Mason4845e442010-05-25 20:56:50 -04006208
Josef Bacik4b46fce2010-05-23 11:00:55 -04006209 while (1) {
Chris Mason4845e442010-05-25 20:56:50 -04006210 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6211 0, &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006212 /*
6213 * We're concerned with the entire range that we're going to be
6214 * doing DIO to, so we need to make sure theres no ordered
6215 * extents in this range.
6216 */
6217 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6218 lockend - lockstart + 1);
6219 if (!ordered)
6220 break;
Chris Mason4845e442010-05-25 20:56:50 -04006221 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6222 &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006223 btrfs_start_ordered_extent(inode, ordered, 1);
6224 btrfs_put_ordered_extent(ordered);
6225 cond_resched();
6226 }
6227
Chris Mason4845e442010-05-25 20:56:50 -04006228 /*
6229 * we don't use btrfs_set_extent_delalloc because we don't want
6230 * the dirty or uptodate bits
6231 */
6232 if (writing) {
6233 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6234 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6235 EXTENT_DELALLOC, 0, NULL, &cached_state,
6236 GFP_NOFS);
6237 if (ret) {
6238 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6239 lockend, EXTENT_LOCKED | write_bits,
6240 1, 0, &cached_state, GFP_NOFS);
6241 goto out;
6242 }
6243 }
6244
6245 free_extent_state(cached_state);
6246 cached_state = NULL;
6247
Chris Mason5a5f79b2010-05-26 21:33:37 -04006248 ret = __blockdev_direct_IO(rw, iocb, inode,
6249 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6250 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6251 btrfs_submit_direct, 0);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006252
6253 if (ret < 0 && ret != -EIOCBQUEUED) {
Chris Mason4845e442010-05-25 20:56:50 -04006254 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6255 offset + iov_length(iov, nr_segs) - 1,
6256 EXTENT_LOCKED | write_bits, 1, 0,
6257 &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006258 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6259 /*
6260 * We're falling back to buffered, unlock the section we didn't
6261 * do IO on.
6262 */
Chris Mason4845e442010-05-25 20:56:50 -04006263 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6264 offset + iov_length(iov, nr_segs) - 1,
6265 EXTENT_LOCKED | write_bits, 1, 0,
6266 &cached_state, GFP_NOFS);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006267 }
Chris Mason4845e442010-05-25 20:56:50 -04006268out:
6269 free_extent_state(cached_state);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006270 return ret;
Chris Mason16432982008-04-10 10:23:21 -04006271}
6272
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05006273static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6274 __u64 start, __u64 len)
6275{
Chris Masonec29ed52011-02-23 16:23:20 -05006276 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05006277}
6278
Chris Mason9ebefb182007-06-15 13:50:00 -04006279int btrfs_readpage(struct file *file, struct page *page)
6280{
Chris Masond1310b22008-01-24 16:13:08 -05006281 struct extent_io_tree *tree;
6282 tree = &BTRFS_I(page->mapping->host)->io_tree;
Jan Schmidt8ddc7d92011-06-13 20:02:58 +02006283 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04006284}
Chris Mason1832a6d2007-12-21 16:27:21 -05006285
Chris Mason39279cc2007-06-12 06:35:45 -04006286static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6287{
Chris Masond1310b22008-01-24 16:13:08 -05006288 struct extent_io_tree *tree;
Chris Masonb888db2b2007-08-27 16:49:44 -04006289
6290
6291 if (current->flags & PF_MEMALLOC) {
6292 redirty_page_for_writepage(wbc, page);
6293 unlock_page(page);
6294 return 0;
6295 }
Chris Masond1310b22008-01-24 16:13:08 -05006296 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04006297 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6298}
Chris Mason39279cc2007-06-12 06:35:45 -04006299
Chris Masonf4219502008-07-22 11:18:09 -04006300int btrfs_writepages(struct address_space *mapping,
6301 struct writeback_control *wbc)
Chris Masonb293f02e2007-11-01 19:45:34 -04006302{
Chris Masond1310b22008-01-24 16:13:08 -05006303 struct extent_io_tree *tree;
Chris Mason771ed682008-11-06 22:02:51 -05006304
Chris Masond1310b22008-01-24 16:13:08 -05006305 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Masonb293f02e2007-11-01 19:45:34 -04006306 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6307}
6308
Chris Mason3ab2fb52007-11-08 10:59:22 -05006309static int
6310btrfs_readpages(struct file *file, struct address_space *mapping,
6311 struct list_head *pages, unsigned nr_pages)
6312{
Chris Masond1310b22008-01-24 16:13:08 -05006313 struct extent_io_tree *tree;
6314 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Mason3ab2fb52007-11-08 10:59:22 -05006315 return extent_readpages(tree, mapping, pages, nr_pages,
6316 btrfs_get_extent);
6317}
Chris Masone6dcd2d2008-07-17 12:53:50 -04006318static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
Chris Masona52d9a82007-08-27 16:49:44 -04006319{
Chris Masond1310b22008-01-24 16:13:08 -05006320 struct extent_io_tree *tree;
6321 struct extent_map_tree *map;
Chris Masona52d9a82007-08-27 16:49:44 -04006322 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04006323
Chris Masond1310b22008-01-24 16:13:08 -05006324 tree = &BTRFS_I(page->mapping->host)->io_tree;
6325 map = &BTRFS_I(page->mapping->host)->extent_tree;
Chris Mason70dec802008-01-29 09:59:12 -05006326 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
Chris Masona52d9a82007-08-27 16:49:44 -04006327 if (ret == 1) {
6328 ClearPagePrivate(page);
6329 set_page_private(page, 0);
6330 page_cache_release(page);
6331 }
6332 return ret;
6333}
Chris Mason39279cc2007-06-12 06:35:45 -04006334
Chris Masone6dcd2d2008-07-17 12:53:50 -04006335static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6336{
Chris Mason98509cf2008-09-11 15:51:43 -04006337 if (PageWriteback(page) || PageDirty(page))
6338 return 0;
Yan Zhengb335b002009-02-12 10:06:04 -05006339 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006340}
6341
Chris Masona52d9a82007-08-27 16:49:44 -04006342static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6343{
Chris Masond1310b22008-01-24 16:13:08 -05006344 struct extent_io_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006345 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00006346 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006347 u64 page_start = page_offset(page);
6348 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04006349
Chris Mason8b62b722009-09-02 16:53:46 -04006350
6351 /*
6352 * we have the page locked, so new writeback can't start,
6353 * and the dirty bit won't be cleared while we are here.
6354 *
6355 * Wait for IO on this page so that we can safely clear
6356 * the PagePrivate2 bit and do ordered accounting
6357 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006358 wait_on_page_writeback(page);
Chris Mason8b62b722009-09-02 16:53:46 -04006359
Chris Masond1310b22008-01-24 16:13:08 -05006360 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006361 if (offset) {
6362 btrfs_releasepage(page, GFP_NOFS);
6363 return;
6364 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00006365 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6366 GFP_NOFS);
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;
6388 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6389 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006390 }
6391 clear_extent_bit(tree, page_start, page_end,
Josef Bacik32c00af2009-10-08 13:34:05 -04006392 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
Josef Bacik2ac55d42010-02-03 19:33:23 +00006393 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006394 __btrfs_releasepage(page, GFP_NOFS);
6395
Chris Mason4a096752008-07-21 10:29:44 -04006396 ClearPageChecked(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04006397 if (PagePrivate(page)) {
Chris Mason9ad6b7b2008-04-18 16:11:30 -04006398 ClearPagePrivate(page);
6399 set_page_private(page, 0);
6400 page_cache_release(page);
6401 }
Chris Mason39279cc2007-06-12 06:35:45 -04006402}
6403
Chris Mason9ebefb182007-06-15 13:50:00 -04006404/*
6405 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6406 * called from a page fault handler when a page is first dirtied. Hence we must
6407 * be careful to check for EOF conditions here. We set the page up correctly
6408 * for a written page which means we get ENOSPC checking when writing into
6409 * holes and correct delalloc and unwritten extent mapping on filesystems that
6410 * support these features.
6411 *
6412 * We are not allowed to take the i_mutex here so we have to play games to
6413 * protect against truncate races as the page could now be beyond EOF. Because
6414 * vmtruncate() writes the inode size before removing pages, once we have the
6415 * page lock we can determine safely if the page is beyond EOF. If it is not
6416 * beyond EOF, then the page is guaranteed safe against truncation until we
6417 * unlock the page.
6418 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07006419int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Chris Mason9ebefb182007-06-15 13:50:00 -04006420{
Nick Pigginc2ec1752009-03-31 15:23:21 -07006421 struct page *page = vmf->page;
Chris Mason6da6aba2007-12-18 16:15:09 -05006422 struct inode *inode = fdentry(vma->vm_file)->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05006423 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006424 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6425 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00006426 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006427 char *kaddr;
6428 unsigned long zero_start;
Chris Mason9ebefb182007-06-15 13:50:00 -04006429 loff_t size;
Chris Mason1832a6d2007-12-21 16:27:21 -05006430 int ret;
Chris Mason9998eb72012-01-25 13:47:40 -05006431 int reserved = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04006432 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006433 u64 page_end;
Chris Mason9ebefb182007-06-15 13:50:00 -04006434
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04006435 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Chris Mason9998eb72012-01-25 13:47:40 -05006436 if (!ret) {
Josef Bacik22c44fe2011-11-30 10:45:38 -05006437 ret = btrfs_update_time(vma->vm_file);
Chris Mason9998eb72012-01-25 13:47:40 -05006438 reserved = 1;
6439 }
Nick Piggin56a76f82009-03-31 15:23:23 -07006440 if (ret) {
6441 if (ret == -ENOMEM)
6442 ret = VM_FAULT_OOM;
6443 else /* -ENOSPC, -EIO, etc */
6444 ret = VM_FAULT_SIGBUS;
Chris Mason9998eb72012-01-25 13:47:40 -05006445 if (reserved)
6446 goto out;
6447 goto out_noreserve;
Nick Piggin56a76f82009-03-31 15:23:23 -07006448 }
Chris Mason1832a6d2007-12-21 16:27:21 -05006449
Nick Piggin56a76f82009-03-31 15:23:23 -07006450 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006451again:
Chris Mason9ebefb182007-06-15 13:50:00 -04006452 lock_page(page);
Chris Mason9ebefb182007-06-15 13:50:00 -04006453 size = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006454 page_start = page_offset(page);
6455 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04006456
Chris Mason9ebefb182007-06-15 13:50:00 -04006457 if ((page->mapping != inode->i_mapping) ||
Chris Masone6dcd2d2008-07-17 12:53:50 -04006458 (page_start >= size)) {
Chris Mason9ebefb182007-06-15 13:50:00 -04006459 /* page got truncated out from underneath us */
6460 goto out_unlock;
6461 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006462 wait_on_page_writeback(page);
6463
Josef Bacik2ac55d42010-02-03 19:33:23 +00006464 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6465 GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006466 set_page_extent_mapped(page);
6467
Chris Masoneb84ae02008-07-17 13:53:27 -04006468 /*
6469 * we can't set the delalloc bits if there are pending ordered
6470 * extents. Drop our locks and wait for them to finish
6471 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04006472 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6473 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00006474 unlock_extent_cached(io_tree, page_start, page_end,
6475 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006476 unlock_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04006477 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006478 btrfs_put_ordered_extent(ordered);
6479 goto again;
6480 }
6481
Josef Bacikfbf19082009-10-01 17:10:23 -04006482 /*
6483 * XXX - page_mkwrite gets called every time the page is dirtied, even
6484 * if it was already dirty, so for space accounting reasons we need to
6485 * clear any delalloc bits for the range we are fixing to save. There
6486 * is probably a better way to do this, but for now keep consistent with
6487 * prepare_pages in the normal write path.
6488 */
Josef Bacik2ac55d42010-02-03 19:33:23 +00006489 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Josef Bacik32c00af2009-10-08 13:34:05 -04006490 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
Josef Bacik2ac55d42010-02-03 19:33:23 +00006491 0, 0, &cached_state, GFP_NOFS);
Josef Bacikfbf19082009-10-01 17:10:23 -04006492
Josef Bacik2ac55d42010-02-03 19:33:23 +00006493 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6494 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006495 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00006496 unlock_extent_cached(io_tree, page_start, page_end,
6497 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006498 ret = VM_FAULT_SIGBUS;
6499 goto out_unlock;
6500 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006501 ret = 0;
Chris Mason9ebefb182007-06-15 13:50:00 -04006502
6503 /* page is wholly or partially inside EOF */
Chris Masona52d9a82007-08-27 16:49:44 -04006504 if (page_start + PAGE_CACHE_SIZE > size)
Chris Masone6dcd2d2008-07-17 12:53:50 -04006505 zero_start = size & ~PAGE_CACHE_MASK;
Chris Mason9ebefb182007-06-15 13:50:00 -04006506 else
Chris Masone6dcd2d2008-07-17 12:53:50 -04006507 zero_start = PAGE_CACHE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04006508
Chris Masone6dcd2d2008-07-17 12:53:50 -04006509 if (zero_start != PAGE_CACHE_SIZE) {
6510 kaddr = kmap(page);
6511 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6512 flush_dcache_page(page);
6513 kunmap(page);
6514 }
Chris Mason247e7432008-07-17 12:53:51 -04006515 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006516 set_page_dirty(page);
Chris Mason50a9b212009-09-11 12:33:12 -04006517 SetPageUptodate(page);
Chris Mason5a3f23d2009-03-31 13:27:11 -04006518
Chris Mason257c62e2009-10-13 13:21:08 -04006519 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6520 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6521
Josef Bacik2ac55d42010-02-03 19:33:23 +00006522 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
Chris Mason9ebefb182007-06-15 13:50:00 -04006523
6524out_unlock:
Chris Mason50a9b212009-09-11 12:33:12 -04006525 if (!ret)
6526 return VM_FAULT_LOCKED;
Chris Mason9ebefb182007-06-15 13:50:00 -04006527 unlock_page(page);
Chris Mason1832a6d2007-12-21 16:27:21 -05006528out:
Josef Bacikec39e182012-01-12 19:10:12 -05006529 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason9998eb72012-01-25 13:47:40 -05006530out_noreserve:
Chris Mason9ebefb182007-06-15 13:50:00 -04006531 return ret;
6532}
6533
Josef Bacika41ad392011-01-31 15:30:16 -05006534static int btrfs_truncate(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04006535{
6536 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikfcb80c22011-05-03 10:40:22 -04006537 struct btrfs_block_rsv *rsv;
Chris Mason39279cc2007-06-12 06:35:45 -04006538 int ret;
Josef Bacik3893e332011-01-31 16:03:11 -05006539 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006540 struct btrfs_trans_handle *trans;
Chris Masond3c2fdcf2007-09-17 10:58:06 -04006541 unsigned long nr;
Chris Masondbe674a2008-07-17 12:54:05 -04006542 u64 mask = root->sectorsize - 1;
Josef Bacik07127182011-08-19 10:29:59 -04006543 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04006544
Josef Bacik5d5e1032009-10-13 16:46:49 -04006545 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6546 if (ret)
Josef Bacika41ad392011-01-31 15:30:16 -05006547 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00006548
Chris Mason4a096752008-07-21 10:29:44 -04006549 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
Yan, Zheng80825102009-11-12 09:35:36 +00006550 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006551
Josef Bacikfcb80c22011-05-03 10:40:22 -04006552 /*
6553 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6554 * 3 things going on here
6555 *
6556 * 1) We need to reserve space for our orphan item and the space to
6557 * delete our orphan item. Lord knows we don't want to have a dangling
6558 * orphan item because we didn't reserve space to remove it.
6559 *
6560 * 2) We need to reserve space to update our inode.
6561 *
6562 * 3) We need to have something to cache all the space that is going to
6563 * be free'd up by the truncate operation, but also have some slack
6564 * space reserved in case it uses space during the truncate (thank you
6565 * very much snapshotting).
6566 *
6567 * And we need these to all be seperate. The fact is we can use alot of
6568 * space doing the truncate, and we have no earthly idea how much space
6569 * we will use, so we need the truncate reservation to be seperate so it
6570 * doesn't end up using space reserved for updating the inode or
6571 * removing the orphan item. We also need to be able to stop the
6572 * transaction and start a new one, which means we need to be able to
6573 * update the inode several times, and we have no idea of knowing how
6574 * many times that will be, so we can't just reserve 1 item for the
6575 * entirety of the opration, so that has to be done seperately as well.
6576 * Then there is the orphan item, which does indeed need to be held on
6577 * to for the whole operation, and we need nobody to touch this reserved
6578 * space except the orphan code.
6579 *
6580 * So that leaves us with
6581 *
6582 * 1) root->orphan_block_rsv - for the orphan deletion.
6583 * 2) rsv - for the truncate reservation, which we will steal from the
6584 * transaction reservation.
6585 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6586 * updating the inode.
6587 */
6588 rsv = btrfs_alloc_block_rsv(root);
6589 if (!rsv)
6590 return -ENOMEM;
Josef Bacik4a338542011-08-29 11:01:31 -04006591 rsv->size = min_size;
Josef Bacikf0cd8462011-03-04 14:37:08 -05006592
Josef Bacik907cbce2011-08-08 13:46:15 -04006593 /*
Josef Bacik07127182011-08-19 10:29:59 -04006594 * 1 for the truncate slack space
Josef Bacik907cbce2011-08-08 13:46:15 -04006595 * 1 for the orphan item we're going to add
6596 * 1 for the orphan item deletion
6597 * 1 for updating the inode.
6598 */
Josef Bacikfcb80c22011-05-03 10:40:22 -04006599 trans = btrfs_start_transaction(root, 4);
6600 if (IS_ERR(trans)) {
6601 err = PTR_ERR(trans);
6602 goto out;
6603 }
Josef Bacikf0cd8462011-03-04 14:37:08 -05006604
Josef Bacik907cbce2011-08-08 13:46:15 -04006605 /* Migrate the slack space for the truncate to our reserve */
6606 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6607 min_size);
Josef Bacikfcb80c22011-05-03 10:40:22 -04006608 BUG_ON(ret);
Josef Bacikf0cd8462011-03-04 14:37:08 -05006609
6610 ret = btrfs_orphan_add(trans, inode);
6611 if (ret) {
6612 btrfs_end_transaction(trans, root);
Josef Bacikfcb80c22011-05-03 10:40:22 -04006613 goto out;
Josef Bacikf0cd8462011-03-04 14:37:08 -05006614 }
6615
Chris Mason5a3f23d2009-03-31 13:27:11 -04006616 /*
6617 * setattr is responsible for setting the ordered_data_close flag,
6618 * but that is only tested during the last file release. That
6619 * could happen well after the next commit, leaving a great big
6620 * window where new writes may get lost if someone chooses to write
6621 * to this file after truncating to zero
6622 *
6623 * The inode doesn't have any dirty data here, and so if we commit
6624 * this is a noop. If someone immediately starts writing to the inode
6625 * it is very likely we'll catch some of their writes in this
6626 * transaction, and the commit will find this file on the ordered
6627 * data list with good things to send down.
6628 *
6629 * This is a best effort solution, there is still a window where
6630 * using truncate to replace the contents of the file will
6631 * end up with a zero length file after a crash.
6632 */
6633 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6634 btrfs_add_ordered_operation(trans, root, inode);
6635
Yan, Zheng80825102009-11-12 09:35:36 +00006636 while (1) {
Josef Bacik36ba0222011-10-18 12:15:48 -04006637 ret = btrfs_block_rsv_refill(root, rsv, min_size);
Josef Bacik907cbce2011-08-08 13:46:15 -04006638 if (ret) {
6639 /*
6640 * This can only happen with the original transaction we
6641 * started above, every other time we shouldn't have a
6642 * transaction started yet.
6643 */
6644 if (ret == -EAGAIN)
6645 goto end_trans;
6646 err = ret;
6647 break;
6648 }
6649
Yan, Zhengd68fc572010-05-16 10:49:58 -04006650 if (!trans) {
Josef Bacik907cbce2011-08-08 13:46:15 -04006651 /* Just need the 1 for updating the inode */
6652 trans = btrfs_start_transaction(root, 1);
Josef Bacikfcb80c22011-05-03 10:40:22 -04006653 if (IS_ERR(trans)) {
Josef Bacik7041ee92011-12-09 13:26:22 -05006654 ret = err = PTR_ERR(trans);
6655 trans = NULL;
6656 break;
Josef Bacikfcb80c22011-05-03 10:40:22 -04006657 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04006658 }
6659
Josef Bacik907cbce2011-08-08 13:46:15 -04006660 trans->block_rsv = rsv;
6661
Yan, Zheng80825102009-11-12 09:35:36 +00006662 ret = btrfs_truncate_inode_items(trans, root, inode,
6663 inode->i_size,
6664 BTRFS_EXTENT_DATA_KEY);
Josef Bacik3893e332011-01-31 16:03:11 -05006665 if (ret != -EAGAIN) {
6666 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00006667 break;
Josef Bacik3893e332011-01-31 16:03:11 -05006668 }
Chris Mason39279cc2007-06-12 06:35:45 -04006669
Josef Bacikfcb80c22011-05-03 10:40:22 -04006670 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00006671 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05006672 if (ret) {
6673 err = ret;
6674 break;
6675 }
Josef Bacik907cbce2011-08-08 13:46:15 -04006676end_trans:
Yan, Zheng80825102009-11-12 09:35:36 +00006677 nr = trans->blocks_used;
6678 btrfs_end_transaction(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04006679 trans = NULL;
Yan, Zheng80825102009-11-12 09:35:36 +00006680 btrfs_btree_balance_dirty(root, nr);
Yan, Zheng80825102009-11-12 09:35:36 +00006681 }
6682
6683 if (ret == 0 && inode->i_nlink > 0) {
Josef Bacikfcb80c22011-05-03 10:40:22 -04006684 trans->block_rsv = root->orphan_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00006685 ret = btrfs_orphan_del(trans, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05006686 if (ret)
6687 err = ret;
Josef Bacikded5db92011-03-04 14:09:46 -05006688 } else if (ret && inode->i_nlink > 0) {
6689 /*
6690 * Failed to do the truncate, remove us from the in memory
6691 * orphan list.
6692 */
6693 ret = btrfs_orphan_del(NULL, inode);
Yan, Zheng80825102009-11-12 09:35:36 +00006694 }
6695
Chris Mason917c16b2011-11-08 14:49:59 -05006696 if (trans) {
6697 trans->block_rsv = &root->fs_info->trans_block_rsv;
6698 ret = btrfs_update_inode(trans, root, inode);
6699 if (ret && !err)
6700 err = ret;
Josef Bacik7b128762008-07-24 12:17:14 -04006701
Chris Mason917c16b2011-11-08 14:49:59 -05006702 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05006703 ret = btrfs_end_transaction(trans, root);
Chris Mason917c16b2011-11-08 14:49:59 -05006704 btrfs_btree_balance_dirty(root, nr);
6705 }
Josef Bacikfcb80c22011-05-03 10:40:22 -04006706
6707out:
6708 btrfs_free_block_rsv(root, rsv);
6709
Josef Bacik3893e332011-01-31 16:03:11 -05006710 if (ret && !err)
6711 err = ret;
Josef Bacika41ad392011-01-31 15:30:16 -05006712
Josef Bacik3893e332011-01-31 16:03:11 -05006713 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04006714}
6715
Sven Wegener3b963622008-06-09 21:57:42 -04006716/*
Chris Masond352ac62008-09-29 15:18:18 -04006717 * create a new subvolume directory/inode (helper for the ioctl).
6718 */
Yan Zhengd2fb3432008-12-11 16:30:39 -05006719int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04006720 struct btrfs_root *new_root, u64 new_dirid)
Chris Mason39279cc2007-06-12 06:35:45 -04006721{
Chris Mason39279cc2007-06-12 06:35:45 -04006722 struct inode *inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04006723 int err;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006724 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006725
Florian Albrechtskirchinger12fc9d02012-02-10 22:15:54 +01006726 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6727 new_dirid, new_dirid,
6728 S_IFDIR | (~current_umask() & S_IRWXUGO),
6729 &index);
Chris Mason54aa1f42007-06-22 14:16:25 -04006730 if (IS_ERR(inode))
Christoph Hellwigf46b5a62008-06-11 21:53:53 -04006731 return PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006732 inode->i_op = &btrfs_dir_inode_operations;
6733 inode->i_fop = &btrfs_dir_file_operations;
6734
Miklos Szeredibfe86842011-10-28 14:13:29 +02006735 set_nlink(inode, 1);
Chris Masondbe674a2008-07-17 12:54:05 -04006736 btrfs_i_size_write(inode, 0);
Sven Wegener3b963622008-06-09 21:57:42 -04006737
Yan, Zheng76dda932009-09-21 16:00:26 -04006738 err = btrfs_update_inode(trans, new_root, inode);
6739 BUG_ON(err);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04006740
Yan, Zheng76dda932009-09-21 16:00:26 -04006741 iput(inode);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04006742 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006743}
6744
Chris Mason39279cc2007-06-12 06:35:45 -04006745struct inode *btrfs_alloc_inode(struct super_block *sb)
6746{
6747 struct btrfs_inode *ei;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006748 struct inode *inode;
Chris Mason39279cc2007-06-12 06:35:45 -04006749
6750 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6751 if (!ei)
6752 return NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006753
6754 ei->root = NULL;
6755 ei->space_info = NULL;
6756 ei->generation = 0;
6757 ei->sequence = 0;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04006758 ei->last_trans = 0;
Chris Mason257c62e2009-10-13 13:21:08 -04006759 ei->last_sub_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04006760 ei->logged_trans = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006761 ei->delalloc_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006762 ei->disk_i_size = 0;
6763 ei->flags = 0;
Josef Bacik7709cde2011-08-04 10:25:02 -04006764 ei->csum_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006765 ei->index_cnt = (u64)-1;
6766 ei->last_unlink_trans = 0;
6767
Josef Bacik9e0baf62011-07-15 15:16:44 +00006768 spin_lock_init(&ei->lock);
6769 ei->outstanding_extents = 0;
6770 ei->reserved_extents = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006771
6772 ei->ordered_data_close = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04006773 ei->orphan_meta_reserved = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006774 ei->dummy_inode = 0;
Chris Mason4cb53002011-05-24 15:35:30 -04006775 ei->in_defrag = 0;
Josef Bacik7fd2ae22011-11-08 15:47:34 -05006776 ei->delalloc_meta_reserved = 0;
Li Zefan261507a02010-12-17 14:21:50 +08006777 ei->force_compress = BTRFS_COMPRESS_NONE;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006778
Miao Xie16cdcec2011-04-22 18:12:22 +08006779 ei->delayed_node = NULL;
6780
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006781 inode = &ei->vfs_inode;
David Sterbaa8067e02011-04-21 00:34:43 +02006782 extent_map_tree_init(&ei->extent_tree);
David Sterbaf993c882011-04-20 23:35:57 +02006783 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6784 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
Josef Bacik0b32f4b2012-03-13 09:38:00 -04006785 ei->io_tree.track_uptodate = 1;
6786 ei->io_failure_tree.track_uptodate = 1;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006787 mutex_init(&ei->log_mutex);
Josef Bacikf2486792012-01-13 12:09:22 -05006788 mutex_init(&ei->delalloc_mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006789 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Josef Bacik7b128762008-07-24 12:17:14 -04006790 INIT_LIST_HEAD(&ei->i_orphan);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006791 INIT_LIST_HEAD(&ei->delalloc_inodes);
Chris Mason5a3f23d2009-03-31 13:27:11 -04006792 INIT_LIST_HEAD(&ei->ordered_operations);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04006793 RB_CLEAR_NODE(&ei->rb_node);
6794
6795 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04006796}
6797
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11006798static void btrfs_i_callback(struct rcu_head *head)
6799{
6800 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11006801 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6802}
6803
Chris Mason39279cc2007-06-12 06:35:45 -04006804void btrfs_destroy_inode(struct inode *inode)
6805{
Chris Masone6dcd2d2008-07-17 12:53:50 -04006806 struct btrfs_ordered_extent *ordered;
Chris Mason5a3f23d2009-03-31 13:27:11 -04006807 struct btrfs_root *root = BTRFS_I(inode)->root;
6808
Chris Mason39279cc2007-06-12 06:35:45 -04006809 WARN_ON(!list_empty(&inode->i_dentry));
6810 WARN_ON(inode->i_data.nrpages);
Josef Bacik9e0baf62011-07-15 15:16:44 +00006811 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6812 WARN_ON(BTRFS_I(inode)->reserved_extents);
Josef Bacik7709cde2011-08-04 10:25:02 -04006813 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6814 WARN_ON(BTRFS_I(inode)->csum_bytes);
Chris Mason39279cc2007-06-12 06:35:45 -04006815
Chris Mason5a3f23d2009-03-31 13:27:11 -04006816 /*
Josef Bacika6dbd422009-11-11 15:53:34 -05006817 * This can happen where we create an inode, but somebody else also
6818 * created the same inode and we need to destroy the one we already
6819 * created.
6820 */
6821 if (!root)
6822 goto free;
6823
6824 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04006825 * Make sure we're properly removed from the ordered operation
6826 * lists.
6827 */
6828 smp_mb();
6829 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6830 spin_lock(&root->fs_info->ordered_extent_lock);
6831 list_del_init(&BTRFS_I(inode)->ordered_operations);
6832 spin_unlock(&root->fs_info->ordered_extent_lock);
6833 }
6834
Yan, Zhengd68fc572010-05-16 10:49:58 -04006835 spin_lock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04006836 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
Li Zefan33345d012011-04-20 10:31:50 +08006837 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6838 (unsigned long long)btrfs_ino(inode));
Yan, Zheng80825102009-11-12 09:35:36 +00006839 list_del_init(&BTRFS_I(inode)->i_orphan);
Josef Bacik7b128762008-07-24 12:17:14 -04006840 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04006841 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04006842
Chris Masond3977122009-01-05 21:25:51 -05006843 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04006844 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6845 if (!ordered)
6846 break;
6847 else {
Chris Masond3977122009-01-05 21:25:51 -05006848 printk(KERN_ERR "btrfs found ordered "
6849 "extent %llu %llu on inode cleanup\n",
6850 (unsigned long long)ordered->file_offset,
6851 (unsigned long long)ordered->len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006852 btrfs_remove_ordered_extent(inode, ordered);
6853 btrfs_put_ordered_extent(ordered);
6854 btrfs_put_ordered_extent(ordered);
6855 }
6856 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006857 inode_tree_del(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04006858 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
Josef Bacika6dbd422009-11-11 15:53:34 -05006859free:
Miao Xie16cdcec2011-04-22 18:12:22 +08006860 btrfs_remove_delayed_node(inode);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11006861 call_rcu(&inode->i_rcu, btrfs_i_callback);
Chris Mason39279cc2007-06-12 06:35:45 -04006862}
6863
Al Viro45321ac2010-06-07 13:43:19 -04006864int btrfs_drop_inode(struct inode *inode)
Yan, Zheng76dda932009-09-21 16:00:26 -04006865{
6866 struct btrfs_root *root = BTRFS_I(inode)->root;
Al Viro45321ac2010-06-07 13:43:19 -04006867
Josef Bacik0af3d002010-06-21 14:48:16 -04006868 if (btrfs_root_refs(&root->root_item) == 0 &&
Chris Mason2cf85722011-07-26 15:35:09 -04006869 !btrfs_is_free_space_inode(root, inode))
Al Viro45321ac2010-06-07 13:43:19 -04006870 return 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04006871 else
Al Viro45321ac2010-06-07 13:43:19 -04006872 return generic_drop_inode(inode);
Yan, Zheng76dda932009-09-21 16:00:26 -04006873}
6874
Sven Wegener0ee0fda2008-07-30 16:54:26 -04006875static void init_once(void *foo)
Chris Mason39279cc2007-06-12 06:35:45 -04006876{
6877 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6878
6879 inode_init_once(&ei->vfs_inode);
6880}
6881
6882void btrfs_destroy_cachep(void)
6883{
6884 if (btrfs_inode_cachep)
6885 kmem_cache_destroy(btrfs_inode_cachep);
6886 if (btrfs_trans_handle_cachep)
6887 kmem_cache_destroy(btrfs_trans_handle_cachep);
6888 if (btrfs_transaction_cachep)
6889 kmem_cache_destroy(btrfs_transaction_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04006890 if (btrfs_path_cachep)
6891 kmem_cache_destroy(btrfs_path_cachep);
Josef Bacikdc89e982011-01-28 17:05:48 -05006892 if (btrfs_free_space_cachep)
6893 kmem_cache_destroy(btrfs_free_space_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04006894}
6895
6896int btrfs_init_cachep(void)
6897{
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006898 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6899 sizeof(struct btrfs_inode), 0,
6900 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
Chris Mason39279cc2007-06-12 06:35:45 -04006901 if (!btrfs_inode_cachep)
6902 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006903
6904 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6905 sizeof(struct btrfs_trans_handle), 0,
6906 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006907 if (!btrfs_trans_handle_cachep)
6908 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006909
6910 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6911 sizeof(struct btrfs_transaction), 0,
6912 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006913 if (!btrfs_transaction_cachep)
6914 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006915
6916 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6917 sizeof(struct btrfs_path), 0,
6918 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04006919 if (!btrfs_path_cachep)
6920 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02006921
Josef Bacikdc89e982011-01-28 17:05:48 -05006922 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6923 sizeof(struct btrfs_free_space), 0,
6924 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6925 if (!btrfs_free_space_cachep)
6926 goto fail;
6927
Chris Mason39279cc2007-06-12 06:35:45 -04006928 return 0;
6929fail:
6930 btrfs_destroy_cachep();
6931 return -ENOMEM;
6932}
6933
6934static int btrfs_getattr(struct vfsmount *mnt,
6935 struct dentry *dentry, struct kstat *stat)
6936{
6937 struct inode *inode = dentry->d_inode;
David Sterbafadc0d82011-11-20 07:33:38 -05006938 u32 blocksize = inode->i_sb->s_blocksize;
6939
Chris Mason39279cc2007-06-12 06:35:45 -04006940 generic_fillattr(inode, stat);
Al Viro0ee5dc62011-07-07 15:44:25 -04006941 stat->dev = BTRFS_I(inode)->root->anon_dev;
Chris Masond6667462008-01-03 14:51:00 -05006942 stat->blksize = PAGE_CACHE_SIZE;
David Sterbafadc0d82011-11-20 07:33:38 -05006943 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6944 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
Chris Mason39279cc2007-06-12 06:35:45 -04006945 return 0;
6946}
6947
Liu Bo75e7cb72011-03-22 10:12:20 +00006948/*
6949 * If a file is moved, it will inherit the cow and compression flags of the new
6950 * directory.
6951 */
6952static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6953{
6954 struct btrfs_inode *b_dir = BTRFS_I(dir);
6955 struct btrfs_inode *b_inode = BTRFS_I(inode);
6956
6957 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6958 b_inode->flags |= BTRFS_INODE_NODATACOW;
6959 else
6960 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6961
6962 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6963 b_inode->flags |= BTRFS_INODE_COMPRESS;
6964 else
6965 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6966}
6967
Chris Masond3977122009-01-05 21:25:51 -05006968static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6969 struct inode *new_dir, struct dentry *new_dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04006970{
6971 struct btrfs_trans_handle *trans;
6972 struct btrfs_root *root = BTRFS_I(old_dir)->root;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006973 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04006974 struct inode *new_inode = new_dentry->d_inode;
6975 struct inode *old_inode = old_dentry->d_inode;
6976 struct timespec ctime = CURRENT_TIME;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006977 u64 index = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006978 u64 root_objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04006979 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08006980 u64 old_ino = btrfs_ino(old_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006981
Li Zefan33345d012011-04-20 10:31:50 +08006982 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Yan, Zhengf679a842009-09-24 09:17:31 -04006983 return -EPERM;
6984
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006985 /* we only allow rename subvolume link between subvolumes */
Li Zefan33345d012011-04-20 10:31:50 +08006986 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
Chris Mason3394e162008-11-17 20:42:26 -05006987 return -EXDEV;
6988
Li Zefan33345d012011-04-20 10:31:50 +08006989 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6990 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006991 return -ENOTEMPTY;
6992
Chris Mason39279cc2007-06-12 06:35:45 -04006993 if (S_ISDIR(old_inode->i_mode) && new_inode &&
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006994 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Chris Mason39279cc2007-06-12 06:35:45 -04006995 return -ENOTEMPTY;
Chris Mason5a3f23d2009-03-31 13:27:11 -04006996 /*
6997 * we're using rename to replace one file with another.
6998 * and the replacement file is large. Start IO on it now so
6999 * we don't add too much work to the end of the transaction
7000 */
Bartlomiej Zolnierkiewicz4baf8c92009-08-07 13:47:08 -04007001 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
Chris Mason5a3f23d2009-03-31 13:27:11 -04007002 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7003 filemap_flush(old_inode->i_mapping);
7004
Yan, Zheng76dda932009-09-21 16:00:26 -04007005 /* close the racy window with snapshot create/destroy ioctl */
Li Zefan33345d012011-04-20 10:31:50 +08007006 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04007007 down_read(&root->fs_info->subvol_sem);
Yan, Zhenga22285a2010-05-16 10:48:46 -04007008 /*
7009 * We want to reserve the absolute worst case amount of items. So if
7010 * both inodes are subvols and we need to unlink them then that would
7011 * require 4 item modifications, but if they are both normal inodes it
7012 * would require 5 item modifications, so we'll assume their normal
7013 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7014 * should cover the worst case number of items we'll modify.
7015 */
7016 trans = btrfs_start_transaction(root, 20);
Johann Lombardib44c59a2011-03-31 13:23:47 +00007017 if (IS_ERR(trans)) {
7018 ret = PTR_ERR(trans);
7019 goto out_notrans;
7020 }
Chris Mason5f39d392007-10-15 16:14:19 -04007021
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007022 if (dest != root)
7023 btrfs_record_root_in_trans(trans, dest);
7024
Yan, Zhenga5719522009-09-24 09:17:31 -04007025 ret = btrfs_set_inode_index(new_dir, &index);
7026 if (ret)
7027 goto out_fail;
Chris Mason5a3f23d2009-03-31 13:27:11 -04007028
Li Zefan33345d012011-04-20 10:31:50 +08007029 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007030 /* force full log commit if subvolume involved. */
7031 root->fs_info->last_trans_log_full_commit = trans->transid;
7032 } else {
Yan, Zhenga5719522009-09-24 09:17:31 -04007033 ret = btrfs_insert_inode_ref(trans, dest,
7034 new_dentry->d_name.name,
7035 new_dentry->d_name.len,
Li Zefan33345d012011-04-20 10:31:50 +08007036 old_ino,
7037 btrfs_ino(new_dir), index);
Yan, Zhenga5719522009-09-24 09:17:31 -04007038 if (ret)
7039 goto out_fail;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007040 /*
7041 * this is an ugly little race, but the rename is required
7042 * to make sure that if we crash, the inode is either at the
7043 * old name or the new one. pinning the log transaction lets
7044 * us make sure we don't allow a log commit to come in after
7045 * we unlink the name but before we add the new name back in.
7046 */
7047 btrfs_pin_log_trans(root);
7048 }
Chris Mason12fcfd22009-03-24 10:24:20 -04007049 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04007050 * make sure the inode gets flushed if it is replacing
7051 * something.
7052 */
Li Zefan33345d012011-04-20 10:31:50 +08007053 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
Chris Mason5a3f23d2009-03-31 13:27:11 -04007054 btrfs_add_ordered_operation(trans, root, old_inode);
Chris Mason5a3f23d2009-03-31 13:27:11 -04007055
Chris Mason39279cc2007-06-12 06:35:45 -04007056 old_dir->i_ctime = old_dir->i_mtime = ctime;
7057 new_dir->i_ctime = new_dir->i_mtime = ctime;
7058 old_inode->i_ctime = ctime;
Chris Mason5f39d392007-10-15 16:14:19 -04007059
Chris Mason12fcfd22009-03-24 10:24:20 -04007060 if (old_dentry->d_parent != new_dentry->d_parent)
7061 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7062
Li Zefan33345d012011-04-20 10:31:50 +08007063 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007064 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7065 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7066 old_dentry->d_name.name,
7067 old_dentry->d_name.len);
7068 } else {
Al Viro92986792011-03-04 17:14:37 +00007069 ret = __btrfs_unlink_inode(trans, root, old_dir,
7070 old_dentry->d_inode,
7071 old_dentry->d_name.name,
7072 old_dentry->d_name.len);
7073 if (!ret)
7074 ret = btrfs_update_inode(trans, root, old_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007075 }
7076 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04007077
7078 if (new_inode) {
7079 new_inode->i_ctime = CURRENT_TIME;
Li Zefan33345d012011-04-20 10:31:50 +08007080 if (unlikely(btrfs_ino(new_inode) ==
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007081 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7082 root_objectid = BTRFS_I(new_inode)->location.objectid;
7083 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7084 root_objectid,
7085 new_dentry->d_name.name,
7086 new_dentry->d_name.len);
7087 BUG_ON(new_inode->i_nlink == 0);
7088 } else {
7089 ret = btrfs_unlink_inode(trans, dest, new_dir,
7090 new_dentry->d_inode,
7091 new_dentry->d_name.name,
7092 new_dentry->d_name.len);
7093 }
7094 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04007095 if (new_inode->i_nlink == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04007096 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007097 BUG_ON(ret);
Josef Bacik7b128762008-07-24 12:17:14 -04007098 }
Chris Mason39279cc2007-06-12 06:35:45 -04007099 }
Josef Bacikaec74772008-07-24 12:12:38 -04007100
Liu Bo75e7cb72011-03-22 10:12:20 +00007101 fixup_inode_flags(new_dir, old_inode);
7102
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007103 ret = btrfs_add_link(trans, new_dir, old_inode,
7104 new_dentry->d_name.name,
Yan, Zhenga5719522009-09-24 09:17:31 -04007105 new_dentry->d_name.len, 0, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007106 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04007107
Li Zefan33345d012011-04-20 10:31:50 +08007108 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
Al Viro10d9f302011-07-16 23:09:10 -04007109 struct dentry *parent = new_dentry->d_parent;
Josef Bacik6a912212010-11-20 09:48:00 +00007110 btrfs_log_new_name(trans, old_inode, old_dir, parent);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04007111 btrfs_end_log_trans(root);
7112 }
Chris Mason39279cc2007-06-12 06:35:45 -04007113out_fail:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05007114 btrfs_end_transaction(trans, root);
Johann Lombardib44c59a2011-03-31 13:23:47 +00007115out_notrans:
Li Zefan33345d012011-04-20 10:31:50 +08007116 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04007117 up_read(&root->fs_info->subvol_sem);
Josef Bacik9ed74f22009-09-11 16:12:44 -04007118
Chris Mason39279cc2007-06-12 06:35:45 -04007119 return ret;
7120}
7121
Chris Masond352ac62008-09-29 15:18:18 -04007122/*
7123 * some fairly slow code that needs optimization. This walks the list
7124 * of all the inodes with pending delalloc and forces them to disk.
7125 */
Yan, Zheng24bbcf02009-11-12 09:36:34 +00007126int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
Chris Masonea8c2812008-08-04 23:17:27 -04007127{
7128 struct list_head *head = &root->fs_info->delalloc_inodes;
7129 struct btrfs_inode *binode;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007130 struct inode *inode;
Chris Masonea8c2812008-08-04 23:17:27 -04007131
Yan Zhengc146afa2008-11-12 14:34:12 -05007132 if (root->fs_info->sb->s_flags & MS_RDONLY)
7133 return -EROFS;
7134
Chris Mason75eff682008-12-15 15:54:40 -05007135 spin_lock(&root->fs_info->delalloc_lock);
Chris Masond3977122009-01-05 21:25:51 -05007136 while (!list_empty(head)) {
Chris Masonea8c2812008-08-04 23:17:27 -04007137 binode = list_entry(head->next, struct btrfs_inode,
7138 delalloc_inodes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007139 inode = igrab(&binode->vfs_inode);
7140 if (!inode)
7141 list_del_init(&binode->delalloc_inodes);
Chris Mason75eff682008-12-15 15:54:40 -05007142 spin_unlock(&root->fs_info->delalloc_lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007143 if (inode) {
Chris Mason8c8bee12008-09-29 11:19:10 -04007144 filemap_flush(inode->i_mapping);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00007145 if (delay_iput)
7146 btrfs_add_delayed_iput(inode);
7147 else
7148 iput(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007149 }
7150 cond_resched();
Chris Mason75eff682008-12-15 15:54:40 -05007151 spin_lock(&root->fs_info->delalloc_lock);
Chris Masonea8c2812008-08-04 23:17:27 -04007152 }
Chris Mason75eff682008-12-15 15:54:40 -05007153 spin_unlock(&root->fs_info->delalloc_lock);
Chris Mason8c8bee12008-09-29 11:19:10 -04007154
7155 /* the filemap_flush will queue IO into the worker threads, but
7156 * we have to make sure the IO is actually started and that
7157 * ordered extents get created before we return
7158 */
7159 atomic_inc(&root->fs_info->async_submit_draining);
Chris Masond3977122009-01-05 21:25:51 -05007160 while (atomic_read(&root->fs_info->nr_async_submits) ||
Chris Mason771ed682008-11-06 22:02:51 -05007161 atomic_read(&root->fs_info->async_delalloc_pages)) {
Chris Mason8c8bee12008-09-29 11:19:10 -04007162 wait_event(root->fs_info->async_submit_wait,
Chris Mason771ed682008-11-06 22:02:51 -05007163 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7164 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
Chris Mason8c8bee12008-09-29 11:19:10 -04007165 }
7166 atomic_dec(&root->fs_info->async_submit_draining);
Chris Masonea8c2812008-08-04 23:17:27 -04007167 return 0;
7168}
7169
Chris Mason39279cc2007-06-12 06:35:45 -04007170static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7171 const char *symname)
7172{
7173 struct btrfs_trans_handle *trans;
7174 struct btrfs_root *root = BTRFS_I(dir)->root;
7175 struct btrfs_path *path;
7176 struct btrfs_key key;
Chris Mason1832a6d2007-12-21 16:27:21 -05007177 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04007178 int err;
7179 int drop_inode = 0;
7180 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04007181 u64 index = 0 ;
Chris Mason39279cc2007-06-12 06:35:45 -04007182 int name_len;
7183 int datasize;
Chris Mason5f39d392007-10-15 16:14:19 -04007184 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04007185 struct btrfs_file_extent_item *ei;
Chris Mason5f39d392007-10-15 16:14:19 -04007186 struct extent_buffer *leaf;
Chris Mason1832a6d2007-12-21 16:27:21 -05007187 unsigned long nr = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04007188
7189 name_len = strlen(symname) + 1;
7190 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7191 return -ENAMETOOLONG;
Chris Mason1832a6d2007-12-21 16:27:21 -05007192
Josef Bacik9ed74f22009-09-11 16:12:44 -04007193 /*
7194 * 2 items for inode item and ref
7195 * 2 items for dir items
7196 * 1 item for xattr if selinux is on
7197 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04007198 trans = btrfs_start_transaction(root, 5);
7199 if (IS_ERR(trans))
7200 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05007201
Li Zefan581bb052011-04-20 10:06:11 +08007202 err = btrfs_find_free_ino(root, &objectid);
7203 if (err)
7204 goto out_unlock;
7205
Josef Bacikaec74772008-07-24 12:12:38 -04007206 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08007207 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f1d2011-05-11 15:26:06 -04007208 S_IFLNK|S_IRWXUGO, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04007209 if (IS_ERR(inode)) {
7210 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007211 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04007212 }
Chris Mason39279cc2007-06-12 06:35:45 -04007213
Eric Paris2a7dba32011-02-01 11:05:39 -05007214 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04007215 if (err) {
7216 drop_inode = 1;
7217 goto out_unlock;
7218 }
7219
Casey Schauflerad19db72011-12-15 10:09:07 -05007220 /*
7221 * If the active LSM wants to access the inode during
7222 * d_instantiate it needs these. Smack checks to see
7223 * if the filesystem supports xattrs by looking at the
7224 * ops vector.
7225 */
7226 inode->i_fop = &btrfs_file_operations;
7227 inode->i_op = &btrfs_file_inode_operations;
7228
Josef Bacika1b075d2010-11-19 20:36:11 +00007229 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04007230 if (err)
7231 drop_inode = 1;
7232 else {
7233 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04007234 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05007235 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04007236 }
Chris Mason39279cc2007-06-12 06:35:45 -04007237 if (drop_inode)
7238 goto out_unlock;
7239
7240 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07007241 if (!path) {
7242 err = -ENOMEM;
7243 drop_inode = 1;
7244 goto out_unlock;
7245 }
Li Zefan33345d012011-04-20 10:31:50 +08007246 key.objectid = btrfs_ino(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007247 key.offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04007248 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7249 datasize = btrfs_file_extent_calc_inline_size(name_len);
7250 err = btrfs_insert_empty_item(trans, root, path, &key,
7251 datasize);
Chris Mason54aa1f42007-06-22 14:16:25 -04007252 if (err) {
7253 drop_inode = 1;
Julia Lawallb0839162011-05-14 07:10:51 +00007254 btrfs_free_path(path);
Chris Mason54aa1f42007-06-22 14:16:25 -04007255 goto out_unlock;
7256 }
Chris Mason5f39d392007-10-15 16:14:19 -04007257 leaf = path->nodes[0];
7258 ei = btrfs_item_ptr(leaf, path->slots[0],
7259 struct btrfs_file_extent_item);
7260 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7261 btrfs_set_file_extent_type(leaf, ei,
Chris Mason39279cc2007-06-12 06:35:45 -04007262 BTRFS_FILE_EXTENT_INLINE);
Chris Masonc8b97812008-10-29 14:49:59 -04007263 btrfs_set_file_extent_encryption(leaf, ei, 0);
7264 btrfs_set_file_extent_compression(leaf, ei, 0);
7265 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7266 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7267
Chris Mason39279cc2007-06-12 06:35:45 -04007268 ptr = btrfs_file_extent_inline_start(ei);
Chris Mason5f39d392007-10-15 16:14:19 -04007269 write_extent_buffer(leaf, symname, ptr, name_len);
7270 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04007271 btrfs_free_path(path);
Chris Mason5f39d392007-10-15 16:14:19 -04007272
Chris Mason39279cc2007-06-12 06:35:45 -04007273 inode->i_op = &btrfs_symlink_inode_operations;
7274 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04007275 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Yan Zhengd899e052008-10-30 14:25:28 -04007276 inode_set_bytes(inode, name_len);
Chris Masondbe674a2008-07-17 12:54:05 -04007277 btrfs_i_size_write(inode, name_len - 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04007278 err = btrfs_update_inode(trans, root, inode);
7279 if (err)
7280 drop_inode = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04007281
7282out_unlock:
Al Viro08c422c2011-12-23 07:58:13 -05007283 if (!err)
7284 d_instantiate(dentry, inode);
Chris Masond3c2fdcf2007-09-17 10:58:06 -04007285 nr = trans->blocks_used;
Josef Bacik7ad85bb2012-01-12 19:10:12 -05007286 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04007287 if (drop_inode) {
7288 inode_dec_link_count(inode);
7289 iput(inode);
7290 }
Chris Masond3c2fdcf2007-09-17 10:58:06 -04007291 btrfs_btree_balance_dirty(root, nr);
Chris Mason39279cc2007-06-12 06:35:45 -04007292 return err;
7293}
Chris Mason16432982008-04-10 10:23:21 -04007294
Josef Bacik0af3d002010-06-21 14:48:16 -04007295static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7296 u64 start, u64 num_bytes, u64 min_size,
7297 loff_t actual_len, u64 *alloc_hint,
7298 struct btrfs_trans_handle *trans)
Yan Zhengd899e052008-10-30 14:25:28 -04007299{
Yan Zhengd899e052008-10-30 14:25:28 -04007300 struct btrfs_root *root = BTRFS_I(inode)->root;
7301 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -04007302 u64 cur_offset = start;
Josef Bacik55a61d12010-11-22 18:50:32 +00007303 u64 i_size;
Yan Zhengd899e052008-10-30 14:25:28 -04007304 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04007305 bool own_trans = true;
Yan Zhengd899e052008-10-30 14:25:28 -04007306
Josef Bacik0af3d002010-06-21 14:48:16 -04007307 if (trans)
7308 own_trans = false;
Yan Zhengd899e052008-10-30 14:25:28 -04007309 while (num_bytes > 0) {
Josef Bacik0af3d002010-06-21 14:48:16 -04007310 if (own_trans) {
7311 trans = btrfs_start_transaction(root, 3);
7312 if (IS_ERR(trans)) {
7313 ret = PTR_ERR(trans);
7314 break;
7315 }
Yan Zhengd899e052008-10-30 14:25:28 -04007316 }
Yan, Zheng5a303d52009-11-12 09:34:52 +00007317
Yan, Zhengefa56462010-05-16 10:49:59 -04007318 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
Josef Bacik81c9ad22012-01-18 10:56:06 -05007319 0, *alloc_hint, &ins, 1);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007320 if (ret) {
Josef Bacik0af3d002010-06-21 14:48:16 -04007321 if (own_trans)
7322 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04007323 break;
Yan, Zheng5a303d52009-11-12 09:34:52 +00007324 }
7325
Yan Zhengd899e052008-10-30 14:25:28 -04007326 ret = insert_reserved_file_extent(trans, inode,
7327 cur_offset, ins.objectid,
7328 ins.offset, ins.offset,
Yan, Zheng920bbbf2009-11-12 09:34:08 +00007329 ins.offset, 0, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04007330 BTRFS_FILE_EXTENT_PREALLOC);
7331 BUG_ON(ret);
Chris Masona1ed8352009-09-11 12:27:37 -04007332 btrfs_drop_extent_cache(inode, cur_offset,
7333 cur_offset + ins.offset -1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007334
Yan Zhengd899e052008-10-30 14:25:28 -04007335 num_bytes -= ins.offset;
7336 cur_offset += ins.offset;
Yan, Zhengefa56462010-05-16 10:49:59 -04007337 *alloc_hint = ins.objectid + ins.offset;
Yan, Zheng5a303d52009-11-12 09:34:52 +00007338
Yan Zhengd899e052008-10-30 14:25:28 -04007339 inode->i_ctime = CURRENT_TIME;
Christoph Hellwig6cbff002009-04-17 10:37:41 +02007340 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Yan Zhengd899e052008-10-30 14:25:28 -04007341 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
Yan, Zhengefa56462010-05-16 10:49:59 -04007342 (actual_len > inode->i_size) &&
7343 (cur_offset > inode->i_size)) {
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00007344 if (cur_offset > actual_len)
Josef Bacik55a61d12010-11-22 18:50:32 +00007345 i_size = actual_len;
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00007346 else
Josef Bacik55a61d12010-11-22 18:50:32 +00007347 i_size = cur_offset;
7348 i_size_write(inode, i_size);
7349 btrfs_ordered_update_i_size(inode, i_size, NULL);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007350 }
7351
Yan Zhengd899e052008-10-30 14:25:28 -04007352 ret = btrfs_update_inode(trans, root, inode);
7353 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -04007354
Josef Bacik0af3d002010-06-21 14:48:16 -04007355 if (own_trans)
7356 btrfs_end_transaction(trans, root);
Yan, Zheng5a303d52009-11-12 09:34:52 +00007357 }
Yan Zhengd899e052008-10-30 14:25:28 -04007358 return ret;
7359}
7360
Josef Bacik0af3d002010-06-21 14:48:16 -04007361int btrfs_prealloc_file_range(struct inode *inode, int mode,
7362 u64 start, u64 num_bytes, u64 min_size,
7363 loff_t actual_len, u64 *alloc_hint)
7364{
7365 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7366 min_size, actual_len, alloc_hint,
7367 NULL);
7368}
7369
7370int btrfs_prealloc_file_range_trans(struct inode *inode,
7371 struct btrfs_trans_handle *trans, int mode,
7372 u64 start, u64 num_bytes, u64 min_size,
7373 loff_t actual_len, u64 *alloc_hint)
7374{
7375 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7376 min_size, actual_len, alloc_hint, trans);
7377}
7378
Chris Masone6dcd2d2008-07-17 12:53:50 -04007379static int btrfs_set_page_dirty(struct page *page)
7380{
Chris Masone6dcd2d2008-07-17 12:53:50 -04007381 return __set_page_dirty_nobuffers(page);
7382}
7383
Al Viro10556cb2011-06-20 19:28:19 -04007384static int btrfs_permission(struct inode *inode, int mask)
Yanfdebe2b2008-01-14 13:26:08 -05007385{
Li Zefanb83cc962010-12-20 16:04:08 +08007386 struct btrfs_root *root = BTRFS_I(inode)->root;
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +00007387 umode_t mode = inode->i_mode;
Li Zefanb83cc962010-12-20 16:04:08 +08007388
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +00007389 if (mask & MAY_WRITE &&
7390 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7391 if (btrfs_root_readonly(root))
7392 return -EROFS;
7393 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7394 return -EACCES;
7395 }
Al Viro2830ba72011-06-20 19:16:29 -04007396 return generic_permission(inode, mask);
Yanfdebe2b2008-01-14 13:26:08 -05007397}
Chris Mason39279cc2007-06-12 06:35:45 -04007398
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007399static const struct inode_operations btrfs_dir_inode_operations = {
Chris Mason3394e162008-11-17 20:42:26 -05007400 .getattr = btrfs_getattr,
Chris Mason39279cc2007-06-12 06:35:45 -04007401 .lookup = btrfs_lookup,
7402 .create = btrfs_create,
7403 .unlink = btrfs_unlink,
7404 .link = btrfs_link,
7405 .mkdir = btrfs_mkdir,
7406 .rmdir = btrfs_rmdir,
7407 .rename = btrfs_rename,
7408 .symlink = btrfs_symlink,
7409 .setattr = btrfs_setattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04007410 .mknod = btrfs_mknod,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007411 .setxattr = btrfs_setxattr,
7412 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05007413 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007414 .removexattr = btrfs_removexattr,
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};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007418static const struct inode_operations btrfs_dir_ro_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007419 .lookup = btrfs_lookup,
Yanfdebe2b2008-01-14 13:26:08 -05007420 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007421 .get_acl = btrfs_get_acl,
Chris Mason39279cc2007-06-12 06:35:45 -04007422};
Yan, Zheng76dda932009-09-21 16:00:26 -04007423
Alexey Dobriyan828c0952009-10-01 15:43:56 -07007424static const struct file_operations btrfs_dir_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007425 .llseek = generic_file_llseek,
7426 .read = generic_read_dir,
David Woodhousecbdf5a22008-08-06 19:42:33 +01007427 .readdir = btrfs_real_readdir,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04007428 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04007429#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04007430 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04007431#endif
Sage Weil6bf13c02008-06-10 10:07:39 -04007432 .release = btrfs_release_file,
Chris Masone02119d2008-09-05 16:13:11 -04007433 .fsync = btrfs_sync_file,
Chris Mason39279cc2007-06-12 06:35:45 -04007434};
7435
Chris Masond1310b22008-01-24 16:13:08 -05007436static struct extent_io_ops btrfs_extent_io_ops = {
Chris Mason07157aa2007-08-30 08:50:51 -04007437 .fill_delalloc = run_delalloc_range,
Chris Mason065631f2008-02-20 12:07:25 -05007438 .submit_bio_hook = btrfs_submit_bio_hook,
Chris Mason239b14b2008-03-24 15:02:07 -04007439 .merge_bio_hook = btrfs_merge_bio_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04007440 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
Chris Masone6dcd2d2008-07-17 12:53:50 -04007441 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
Chris Mason247e7432008-07-17 12:53:51 -04007442 .writepage_start_hook = btrfs_writepage_start_hook,
Chris Masonb0c68f82008-01-31 11:05:37 -05007443 .set_bit_hook = btrfs_set_bit_hook,
7444 .clear_bit_hook = btrfs_clear_bit_hook,
Josef Bacik9ed74f22009-09-11 16:12:44 -04007445 .merge_extent_hook = btrfs_merge_extent_hook,
7446 .split_extent_hook = btrfs_split_extent_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04007447};
7448
Chris Mason35054392009-01-21 13:11:13 -05007449/*
7450 * btrfs doesn't support the bmap operation because swapfiles
7451 * use bmap to make a mapping of extents in the file. They assume
7452 * these extents won't change over the life of the file and they
7453 * use the bmap result to do IO directly to the drive.
7454 *
7455 * the btrfs bmap call would return logical addresses that aren't
7456 * suitable for IO and they also will change frequently as COW
7457 * operations happen. So, swapfile + btrfs == corruption.
7458 *
7459 * For now we're avoiding this by dropping bmap.
7460 */
Alexey Dobriyan7f094102009-09-21 17:01:10 -07007461static const struct address_space_operations btrfs_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04007462 .readpage = btrfs_readpage,
7463 .writepage = btrfs_writepage,
Chris Masonb293f02e2007-11-01 19:45:34 -04007464 .writepages = btrfs_writepages,
Chris Mason3ab2fb52007-11-08 10:59:22 -05007465 .readpages = btrfs_readpages,
Chris Mason16432982008-04-10 10:23:21 -04007466 .direct_IO = btrfs_direct_IO,
Chris Masona52d9a82007-08-27 16:49:44 -04007467 .invalidatepage = btrfs_invalidatepage,
7468 .releasepage = btrfs_releasepage,
Chris Masone6dcd2d2008-07-17 12:53:50 -04007469 .set_page_dirty = btrfs_set_page_dirty,
Andi Kleen465fdd92009-09-16 11:50:18 +02007470 .error_remove_page = generic_error_remove_page,
Chris Mason39279cc2007-06-12 06:35:45 -04007471};
7472
Alexey Dobriyan7f094102009-09-21 17:01:10 -07007473static const struct address_space_operations btrfs_symlink_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04007474 .readpage = btrfs_readpage,
7475 .writepage = btrfs_writepage,
Chris Mason2bf5a722007-08-30 11:54:02 -04007476 .invalidatepage = btrfs_invalidatepage,
7477 .releasepage = btrfs_releasepage,
Chris Mason39279cc2007-06-12 06:35:45 -04007478};
7479
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007480static const struct inode_operations btrfs_file_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007481 .getattr = btrfs_getattr,
7482 .setattr = btrfs_setattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007483 .setxattr = btrfs_setxattr,
7484 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05007485 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007486 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05007487 .permission = btrfs_permission,
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05007488 .fiemap = btrfs_fiemap,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007489 .get_acl = btrfs_get_acl,
Chris Mason39279cc2007-06-12 06:35:45 -04007490};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007491static const struct inode_operations btrfs_special_inode_operations = {
Josef Bacik618e21d2007-07-11 10:18:17 -04007492 .getattr = btrfs_getattr,
7493 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05007494 .permission = btrfs_permission,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007495 .setxattr = btrfs_setxattr,
7496 .getxattr = btrfs_getxattr,
Josef Bacik33268ea2008-07-24 12:16:36 -04007497 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04007498 .removexattr = btrfs_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007499 .get_acl = btrfs_get_acl,
Josef Bacik618e21d2007-07-11 10:18:17 -04007500};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07007501static const struct inode_operations btrfs_symlink_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04007502 .readlink = generic_readlink,
7503 .follow_link = page_follow_link_light,
7504 .put_link = page_put_link,
Li Zefanf2095612010-11-19 02:05:24 +00007505 .getattr = btrfs_getattr,
Josef Bacik22c44fe2011-11-30 10:45:38 -05007506 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05007507 .permission = btrfs_permission,
Jim Owens0279b4c2009-02-04 09:29:13 -05007508 .setxattr = btrfs_setxattr,
7509 .getxattr = btrfs_getxattr,
7510 .listxattr = btrfs_listxattr,
7511 .removexattr = btrfs_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02007512 .get_acl = btrfs_get_acl,
Chris Mason39279cc2007-06-12 06:35:45 -04007513};
Yan, Zheng76dda932009-09-21 16:00:26 -04007514
Alexey Dobriyan82d339d2009-10-09 09:54:36 -04007515const struct dentry_operations btrfs_dentry_operations = {
Yan, Zheng76dda932009-09-21 16:00:26 -04007516 .d_delete = btrfs_dentry_delete,
Josef Bacikb4aff1f2011-06-28 16:18:59 -04007517 .d_release = btrfs_dentry_release,
Yan, Zheng76dda932009-09-21 16:00:26 -04007518};