blob: ec95a69d17aa73e058bedd5c61e42d23f87a4f91 [file] [log] [blame]
Andrew Mortonf79e2ab2006-03-31 02:30:42 -08001/*
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>
Al Viro914e2632006-10-18 13:55:46 -04009#include <linux/sched.h>
Andrew Mortonf79e2ab2006-03-31 02:30:42 -080010#include <linux/writeback.h>
11#include <linux/syscalls.h>
12#include <linux/linkage.h>
13#include <linux/pagemap.h>
David Howellscf9a2ae2006-08-29 19:05:54 +010014#include <linux/quotaops.h>
15#include <linux/buffer_head.h>
Andrew Mortonf79e2ab2006-03-31 02:30:42 -080016
17#define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
18 SYNC_FILE_RANGE_WAIT_AFTER)
19
20/*
David Howellscf9a2ae2006-08-29 19:05:54 +010021 * 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
Heiko Carstensa5f8fa92009-01-14 14:14:11 +010039SYSCALL_DEFINE0(sync)
David Howellscf9a2ae2006-08-29 19:05:54 +010040{
41 do_sync(1);
42 return 0;
43}
44
Jens Axboea2a95372009-03-17 09:38:40 +010045static void do_sync_work(struct work_struct *work)
46{
47 do_sync(0);
48 kfree(work);
49}
50
David Howellscf9a2ae2006-08-29 19:05:54 +010051void emergency_sync(void)
52{
Jens Axboea2a95372009-03-17 09:38:40 +010053 struct work_struct *work;
54
55 work = kmalloc(sizeof(*work), GFP_ATOMIC);
56 if (work) {
57 INIT_WORK(work, do_sync_work);
58 schedule_work(work);
59 }
David Howellscf9a2ae2006-08-29 19:05:54 +010060}
61
62/*
63 * Generic function to fsync a file.
64 *
65 * filp may be NULL if called via the msync of a vma.
66 */
67int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
68{
69 struct inode * inode = dentry->d_inode;
70 struct super_block * sb;
71 int ret, err;
72
73 /* sync the inode to buffers */
74 ret = write_inode_now(inode, 0);
75
76 /* sync the superblock to buffers */
77 sb = inode->i_sb;
78 lock_super(sb);
OGAWA Hirofumi762873c2008-04-29 00:59:42 -070079 if (sb->s_dirt && sb->s_op->write_super)
David Howellscf9a2ae2006-08-29 19:05:54 +010080 sb->s_op->write_super(sb);
81 unlock_super(sb);
82
83 /* .. finally sync the buffers to disk */
84 err = sync_blockdev(sb->s_bdev);
85 if (!ret)
86 ret = err;
87 return ret;
88}
89
Christoph Hellwig4c728ef2008-12-22 21:11:15 +010090/**
91 * vfs_fsync - perform a fsync or fdatasync on a file
92 * @file: file to sync
93 * @dentry: dentry of @file
94 * @data: only perform a fdatasync operation
95 *
96 * Write back data and metadata for @file to disk. If @datasync is
97 * set only metadata needed to access modified file data is written.
98 *
99 * In case this function is called from nfsd @file may be %NULL and
100 * only @dentry is set. This can only happen when the filesystem
101 * implements the export_operations API.
102 */
103int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100104{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100105 const struct file_operations *fop;
106 struct address_space *mapping;
107 int err, ret;
David Howellscf9a2ae2006-08-29 19:05:54 +0100108
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100109 /*
110 * Get mapping and operations from the file in case we have
111 * as file, or get the default values for them in case we
112 * don't have a struct file available. Damn nfsd..
113 */
114 if (file) {
115 mapping = file->f_mapping;
116 fop = file->f_op;
117 } else {
118 mapping = dentry->d_inode->i_mapping;
119 fop = dentry->d_inode->i_fop;
120 }
121
122 if (!fop || !fop->fsync) {
David Howellscf9a2ae2006-08-29 19:05:54 +0100123 ret = -EINVAL;
124 goto out;
125 }
126
127 ret = filemap_fdatawrite(mapping);
128
129 /*
130 * We need to protect against concurrent writers, which could cause
131 * livelocks in fsync_buffers_list().
132 */
133 mutex_lock(&mapping->host->i_mutex);
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100134 err = fop->fsync(file, dentry, datasync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100135 if (!ret)
136 ret = err;
137 mutex_unlock(&mapping->host->i_mutex);
138 err = filemap_fdatawait(mapping);
139 if (!ret)
140 ret = err;
141out:
142 return ret;
143}
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100144EXPORT_SYMBOL(vfs_fsync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100145
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100146static int do_fsync(unsigned int fd, int datasync)
David Howellscf9a2ae2006-08-29 19:05:54 +0100147{
148 struct file *file;
149 int ret = -EBADF;
150
151 file = fget(fd);
152 if (file) {
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100153 ret = vfs_fsync(file, file->f_path.dentry, datasync);
David Howellscf9a2ae2006-08-29 19:05:54 +0100154 fput(file);
155 }
156 return ret;
157}
158
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100159SYSCALL_DEFINE1(fsync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100160{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100161 return do_fsync(fd, 0);
David Howellscf9a2ae2006-08-29 19:05:54 +0100162}
163
Heiko Carstensa5f8fa92009-01-14 14:14:11 +0100164SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
David Howellscf9a2ae2006-08-29 19:05:54 +0100165{
Christoph Hellwig4c728ef2008-12-22 21:11:15 +0100166 return do_fsync(fd, 1);
David Howellscf9a2ae2006-08-29 19:05:54 +0100167}
168
169/*
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800170 * sys_sync_file_range() permits finely controlled syncing over a segment of
171 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
172 * zero then sys_sync_file_range() will operate from offset out to EOF.
173 *
174 * The flag bits are:
175 *
176 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
177 * before performing the write.
178 *
179 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
Pavel Machekcce77082008-07-23 21:27:36 -0700180 * range which are not presently under writeback. Note that this may block for
181 * significant periods due to exhaustion of disk request structures.
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800182 *
183 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
184 * after performing the write.
185 *
186 * Useful combinations of the flag bits are:
187 *
188 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
189 * in the range which were dirty on entry to sys_sync_file_range() are placed
190 * under writeout. This is a start-write-for-data-integrity operation.
191 *
192 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
193 * are not presently under writeout. This is an asynchronous flush-to-disk
194 * operation. Not suitable for data integrity operations.
195 *
196 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
197 * completion of writeout of all pages in the range. This will be used after an
198 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
199 * for that operation to complete and to return the result.
200 *
201 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
202 * a traditional sync() operation. This is a write-for-data-integrity operation
203 * which will ensure that all pages in the range which were dirty on entry to
204 * sys_sync_file_range() are committed to disk.
205 *
206 *
207 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
208 * I/O errors or ENOSPC conditions and will return those to the caller, after
209 * clearing the EIO and ENOSPC flags in the address_space.
210 *
211 * It should be noted that none of these operations write out the file's
212 * metadata. So unless the application is strictly performing overwrites of
213 * already-instantiated disk blocks, there are no guarantees here that the data
214 * will be available after a crash.
215 */
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100216SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
217 unsigned int flags)
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800218{
219 int ret;
220 struct file *file;
221 loff_t endbyte; /* inclusive */
222 int fput_needed;
223 umode_t i_mode;
224
225 ret = -EINVAL;
226 if (flags & ~VALID_FLAGS)
227 goto out;
228
229 endbyte = offset + nbytes;
230
231 if ((s64)offset < 0)
232 goto out;
233 if ((s64)endbyte < 0)
234 goto out;
235 if (endbyte < offset)
236 goto out;
237
238 if (sizeof(pgoff_t) == 4) {
239 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
240 /*
241 * The range starts outside a 32 bit machine's
242 * pagecache addressing capabilities. Let it "succeed"
243 */
244 ret = 0;
245 goto out;
246 }
247 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
248 /*
249 * Out to EOF
250 */
251 nbytes = 0;
252 }
253 }
254
255 if (nbytes == 0)
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700256 endbyte = LLONG_MAX;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800257 else
258 endbyte--; /* inclusive */
259
260 ret = -EBADF;
261 file = fget_light(fd, &fput_needed);
262 if (!file)
263 goto out;
264
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800265 i_mode = file->f_path.dentry->d_inode->i_mode;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800266 ret = -ESPIPE;
267 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
268 !S_ISLNK(i_mode))
269 goto out_put;
270
Mark Fashehef51c972007-05-08 00:27:10 -0700271 ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800272out_put:
273 fput_light(file, fput_needed);
274out:
275 return ret;
276}
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100277#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
278asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
279 long flags)
280{
281 return SYSC_sync_file_range((int) fd, offset, nbytes,
282 (unsigned int) flags);
283}
284SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
285#endif
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800286
David Woodhouseedd5cd42007-06-27 14:10:09 -0700287/* It would be nice if people remember that not all the world's an i386
288 when they introduce new system calls */
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100289SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
290 loff_t offset, loff_t nbytes)
David Woodhouseedd5cd42007-06-27 14:10:09 -0700291{
292 return sys_sync_file_range(fd, offset, nbytes, flags);
293}
Heiko Carstens6673e0c2009-01-14 14:14:02 +0100294#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
295asmlinkage long SyS_sync_file_range2(long fd, long flags,
296 loff_t offset, loff_t nbytes)
297{
298 return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
299 offset, nbytes);
300}
301SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
302#endif
David Woodhouseedd5cd42007-06-27 14:10:09 -0700303
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800304/*
305 * `endbyte' is inclusive
306 */
Mark Fasheh5b04aa32007-03-01 11:01:55 -0800307int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
308 loff_t endbyte, unsigned int flags)
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800309{
310 int ret;
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800311
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800312 if (!mapping) {
313 ret = -EINVAL;
314 goto out;
315 }
316
317 ret = 0;
318 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
319 ret = wait_on_page_writeback_range(mapping,
320 offset >> PAGE_CACHE_SHIFT,
321 endbyte >> PAGE_CACHE_SHIFT);
322 if (ret < 0)
323 goto out;
324 }
325
326 if (flags & SYNC_FILE_RANGE_WRITE) {
327 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
Nick Pigginee53a892009-01-06 14:39:12 -0800328 WB_SYNC_ALL);
Andrew Mortonf79e2ab2006-03-31 02:30:42 -0800329 if (ret < 0)
330 goto out;
331 }
332
333 if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
334 ret = wait_on_page_writeback_range(mapping,
335 offset >> PAGE_CACHE_SHIFT,
336 endbyte >> PAGE_CACHE_SHIFT);
337 }
338out:
339 return ret;
340}
Mark Fasheh5b04aa32007-03-01 11:01:55 -0800341EXPORT_SYMBOL_GPL(do_sync_mapping_range);