projects/linux·blob v2.6.21 · refs

278 lines · 6.8 KiB

1/*
2 * High-level sync()-related operations
3 */
4
5#include <linux/kernel.h>
6#include <linux/file.h>
7#include <linux/fs.h>
8#include <linux/module.h>
9#include <linux/sched.h>
10#include <linux/writeback.h>
11#include <linux/syscalls.h>
12#include <linux/linkage.h>
13#include <linux/pagemap.h>
14#include <linux/quotaops.h>
15#include <linux/buffer_head.h>
16
17#define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
18 SYNC_FILE_RANGE_WAIT_AFTER)
19
20/*
21 * sync everything. Start out by waking pdflush, because that writes back
22 * all queues in parallel.
23 */
24static void do_sync(unsigned long wait)
25{
26 wakeup_pdflush(0);
27 sync_inodes(0); /* All mappings, inodes and their blockdevs */
28 DQUOT_SYNC(NULL);
29 sync_supers(); /* Write the superblocks */
30 sync_filesystems(0); /* Start syncing the filesystems */
31 sync_filesystems(wait); /* Waitingly sync the filesystems */
32 sync_inodes(wait); /* Mappings, inodes and blockdevs, again. */
33 if (!wait)
34 printk("Emergency Sync complete\n");
35 if (unlikely(laptop_mode))
36 laptop_sync_completion();
37}
38
39asmlinkage long sys_sync(void)
40{
41 do_sync(1);
42 return 0;
43}
44
45void emergency_sync(void)
46{
47 pdflush_operation(do_sync, 0);
48}
49
50/*
51 * Generic function to fsync a file.
52 *
53 * filp may be NULL if called via the msync of a vma.
54 */
55int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
56{
57 struct inode * inode = dentry->d_inode;
58 struct super_block * sb;
59 int ret, err;
60
61 /* sync the inode to buffers */
62 ret = write_inode_now(inode, 0);
63
64 /* sync the superblock to buffers */
65 sb = inode->i_sb;
66 lock_super(sb);
67 if (sb->s_op->write_super)
68 sb->s_op->write_super(sb);
69 unlock_super(sb);
70
71 /* .. finally sync the buffers to disk */
72 err = sync_blockdev(sb->s_bdev);
73 if (!ret)
74 ret = err;
75 return ret;
76}
77
78long do_fsync(struct file *file, int datasync)
79{
80 int ret;
81 int err;
82 struct address_space *mapping = file->f_mapping;
83
84 if (!file->f_op || !file->f_op->fsync) {
85 /* Why? We can still call filemap_fdatawrite */
86 ret = -EINVAL;
87 goto out;
88 }
89
90 ret = filemap_fdatawrite(mapping);
91
92 /*
93 * We need to protect against concurrent writers, which could cause
94 * livelocks in fsync_buffers_list().
95 */
96 mutex_lock(&mapping->host->i_mutex);
97 err = file->f_op->fsync(file, file->f_path.dentry, datasync);
98 if (!ret)
99 ret = err;
100 mutex_unlock(&mapping->host->i_mutex);
101 err = filemap_fdatawait(mapping);
102 if (!ret)
103 ret = err;
104out:
105 return ret;
106}
107
108static long __do_fsync(unsigned int fd, int datasync)
109{
110 struct file *file;
111 int ret = -EBADF;
112
113 file = fget(fd);
114 if (file) {
115 ret = do_fsync(file, datasync);
116 fput(file);
117 }
118 return ret;
119}
120
121asmlinkage long sys_fsync(unsigned int fd)
122{
123 return __do_fsync(fd, 0);
124}
125
126asmlinkage long sys_fdatasync(unsigned int fd)
127{
128 return __do_fsync(fd, 1);
129}
130
131/*
132 * sys_sync_file_range() permits finely controlled syncing over a segment of
133 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
134 * zero then sys_sync_file_range() will operate from offset out to EOF.
135 *
136 * The flag bits are:
137 *
138 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
139 * before performing the write.
140 *
141 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
142 * range which are not presently under writeback.
143 *
144 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
145 * after performing the write.
146 *
147 * Useful combinations of the flag bits are:
148 *
149 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
150 * in the range which were dirty on entry to sys_sync_file_range() are placed
151 * under writeout. This is a start-write-for-data-integrity operation.
152 *
153 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
154 * are not presently under writeout. This is an asynchronous flush-to-disk
155 * operation. Not suitable for data integrity operations.
156 *
157 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
158 * completion of writeout of all pages in the range. This will be used after an
159 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
160 * for that operation to complete and to return the result.
161 *
162 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
163 * a traditional sync() operation. This is a write-for-data-integrity operation
164 * which will ensure that all pages in the range which were dirty on entry to
165 * sys_sync_file_range() are committed to disk.
166 *
167 *
168 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
169 * I/O errors or ENOSPC conditions and will return those to the caller, after
170 * clearing the EIO and ENOSPC flags in the address_space.
171 *
172 * It should be noted that none of these operations write out the file's
173 * metadata. So unless the application is strictly performing overwrites of
174 * already-instantiated disk blocks, there are no guarantees here that the data
175 * will be available after a crash.
176 */
177asmlinkage long sys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
178 unsigned int flags)
179{
180 int ret;
181 struct file *file;
182 loff_t endbyte; /* inclusive */
183 int fput_needed;
184 umode_t i_mode;
185
186 ret = -EINVAL;
187 if (flags & ~VALID_FLAGS)
188 goto out;
189
190 endbyte = offset + nbytes;
191
192 if ((s64)offset < 0)
193 goto out;
194 if ((s64)endbyte < 0)
195 goto out;
196 if (endbyte < offset)
197 goto out;
198
199 if (sizeof(pgoff_t) == 4) {
200 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
201 /*
202 * The range starts outside a 32 bit machine's
203 * pagecache addressing capabilities. Let it "succeed"
204 */
205 ret = 0;
206 goto out;
207 }
208 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
209 /*
210 * Out to EOF
211 */
212 nbytes = 0;
213 }
214 }
215
216 if (nbytes == 0)
217 endbyte = LLONG_MAX;
218 else
219 endbyte--; /* inclusive */
220
221 ret = -EBADF;
222 file = fget_light(fd, &fput_needed);
223 if (!file)
224 goto out;
225
226 i_mode = file->f_path.dentry->d_inode->i_mode;
227 ret = -ESPIPE;
228 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
229 !S_ISLNK(i_mode))
230 goto out_put;
231
232 ret = do_sync_file_range(file, offset, endbyte, flags);
233out_put:
234 fput_light(file, fput_needed);
235out:
236 return ret;
237}
238
239/*
240 * `endbyte' is inclusive
241 */
242int do_sync_file_range(struct file *file, loff_t offset, loff_t endbyte,
243 unsigned int flags)
244{
245 int ret;
246 struct address_space *mapping;
247
248 mapping = file->f_mapping;
249 if (!mapping) {
250 ret = -EINVAL;
251 goto out;
252 }
253
254 ret = 0;
255 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
256 ret = wait_on_page_writeback_range(mapping,
257 offset >> PAGE_CACHE_SHIFT,
258 endbyte >> PAGE_CACHE_SHIFT);
259 if (ret < 0)
260 goto out;
261 }
262
263 if (flags & SYNC_FILE_RANGE_WRITE) {
264 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
265 WB_SYNC_NONE);
266 if (ret < 0)
267 goto out;
268 }
269
270 if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
271 ret = wait_on_page_writeback_range(mapping,
272 offset >> PAGE_CACHE_SHIFT,
273 endbyte >> PAGE_CACHE_SHIFT);
274 }
275out:
276 return ret;
277}
278EXPORT_SYMBOL_GPL(do_sync_file_range);