| 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/slab.h>
|
| 9 | #include <linux/export.h>
|
| 10 | #include <linux/namei.h>
|
| 11 | #include <linux/sched.h>
|
| 12 | #include <linux/writeback.h>
|
| 13 | #include <linux/syscalls.h>
|
| 14 | #include <linux/linkage.h>
|
| 15 | #include <linux/pagemap.h>
|
| 16 | #include <linux/quotaops.h>
|
| 17 | #include <linux/backing-dev.h>
|
| 18 | #include "internal.h"
|
| 19 |
|
| 20 | #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
|
| 21 | SYNC_FILE_RANGE_WAIT_AFTER)
|
| 22 |
|
| 23 | |
| 24 | * Do the filesystem syncing work. For simple filesystems
|
| 25 | * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
|
| 26 | * submit IO for these buffers via __sync_blockdev(). This also speeds up the
|
| 27 | * wait == 1 case since in that case write_inode() functions do
|
| 28 | * sync_dirty_buffer() and thus effectively write one block at a time.
|
| 29 |
|
| 30 | static int __sync_filesystem(struct super_block *sb, int wait,
|
| 31 | unsigned long start)
|
| 32 | {
|
| 33 | if (wait)
|
| 34 | sync_inodes_sb(sb, start);
|
| 35 | else
|
| 36 | writeback_inodes_sb(sb, WB_REASON_SYNC);
|
| 37 |
|
| 38 | if (sb->s_op->sync_fs)
|
| 39 | sb->s_op->sync_fs(sb, wait);
|
| 40 | return __sync_blockdev(sb->s_bdev, wait);
|
| 41 | }
|
| 42 |
|
| 43 | |
| 44 | * Write out and wait upon all dirty data associated with this
|
| 45 | * superblock. Filesystem data as well as the underlying block
|
| 46 | * device. Takes the superblock lock.
|
| 47 |
|
| 48 | int sync_filesystem(struct super_block *sb)
|
| 49 | {
|
| 50 | int ret;
|
| 51 | unsigned long start = jiffies;
|
| 52 |
|
| 53 | |
| 54 | * We need to be protected against the filesystem going from
|
| 55 | * r/o to r/w or vice versa.
|
| 56 |
|
| 57 | WARN_ON(!rwsem_is_locked(&sb->s_umount));
|
| 58 |
|
| 59 | |
| 60 | * No point in syncing out anything if the filesystem is read-only.
|
| 61 |
|
| 62 | if (sb->s_flags & MS_RDONLY)
|
| 63 | return 0;
|
| 64 |
|
| 65 | ret = __sync_filesystem(sb, 0, start);
|
| 66 | if (ret < 0)
|
| 67 | return ret;
|
| 68 | return __sync_filesystem(sb, 1, start);
|
| 69 | }
|
| 70 | EXPORT_SYMBOL_GPL(sync_filesystem);
|
| 71 |
|
| 72 | static void sync_inodes_one_sb(struct super_block *sb, void *arg)
|
| 73 | {
|
| 74 | if (!(sb->s_flags & MS_RDONLY))
|
| 75 | sync_inodes_sb(sb, *((unsigned long *)arg));
|
| 76 | }
|
| 77 |
|
| 78 | static void sync_fs_one_sb(struct super_block *sb, void *arg)
|
| 79 | {
|
| 80 | if (!(sb->s_flags & MS_RDONLY) && sb->s_op->sync_fs)
|
| 81 | sb->s_op->sync_fs(sb, *(int *)arg);
|
| 82 | }
|
| 83 |
|
| 84 | static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
|
| 85 | {
|
| 86 | filemap_fdatawrite(bdev->bd_inode->i_mapping);
|
| 87 | }
|
| 88 |
|
| 89 | static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
|
| 90 | {
|
| 91 | filemap_fdatawait(bdev->bd_inode->i_mapping);
|
| 92 | }
|
| 93 |
|
| 94 | |
| 95 | * Sync everything. We start by waking flusher threads so that most of
|
| 96 | * writeback runs on all devices in parallel. Then we sync all inodes reliably
|
| 97 | * which effectively also waits for all flusher threads to finish doing
|
| 98 | * writeback. At this point all data is on disk so metadata should be stable
|
| 99 | * and we tell filesystems to sync their metadata via ->sync_fs() calls.
|
| 100 | * Finally, we writeout all block devices because some filesystems (e.g. ext2)
|
| 101 | * just write metadata (such as inodes or bitmaps) to block device page cache
|
| 102 | * and do not sync it on their own in ->sync_fs().
|
| 103 |
|
| 104 | SYSCALL_DEFINE0(sync)
|
| 105 | {
|
| 106 | int nowait = 0, wait = 1;
|
| 107 | unsigned long start = jiffies;
|
| 108 |
|
| 109 | wakeup_flusher_threads(0, WB_REASON_SYNC);
|
| 110 | iterate_supers(sync_inodes_one_sb, &start);
|
| 111 | iterate_supers(sync_fs_one_sb, &nowait);
|
| 112 | iterate_supers(sync_fs_one_sb, &wait);
|
| 113 | iterate_bdevs(fdatawrite_one_bdev, NULL);
|
| 114 | iterate_bdevs(fdatawait_one_bdev, NULL);
|
| 115 | if (unlikely(laptop_mode))
|
| 116 | laptop_sync_completion();
|
| 117 | return 0;
|
| 118 | }
|
| 119 |
|
| 120 | static void do_sync_work(struct work_struct *work)
|
| 121 | {
|
| 122 | int nowait = 0;
|
| 123 |
|
| 124 | |
| 125 | * Sync twice to reduce the possibility we skipped some inodes / pages
|
| 126 | * because they were temporarily locked
|
| 127 |
|
| 128 | iterate_supers(sync_inodes_one_sb, &nowait);
|
| 129 | iterate_supers(sync_fs_one_sb, &nowait);
|
| 130 | iterate_bdevs(fdatawrite_one_bdev, NULL);
|
| 131 | iterate_supers(sync_inodes_one_sb, &nowait);
|
| 132 | iterate_supers(sync_fs_one_sb, &nowait);
|
| 133 | iterate_bdevs(fdatawrite_one_bdev, NULL);
|
| 134 | printk("Emergency Sync complete\n");
|
| 135 | kfree(work);
|
| 136 | }
|
| 137 |
|
| 138 | void emergency_sync(void)
|
| 139 | {
|
| 140 | struct work_struct *work;
|
| 141 |
|
| 142 | work = kmalloc(sizeof(*work), GFP_ATOMIC);
|
| 143 | if (work) {
|
| 144 | INIT_WORK(work, do_sync_work);
|
| 145 | schedule_work(work);
|
| 146 | }
|
| 147 | }
|
| 148 |
|
| 149 | |
| 150 | * sync a single super
|
| 151 |
|
| 152 | SYSCALL_DEFINE1(syncfs, int, fd)
|
| 153 | {
|
| 154 | struct fd f = fdget(fd);
|
| 155 | struct super_block *sb;
|
| 156 | int ret;
|
| 157 |
|
| 158 | if (!f.file)
|
| 159 | return -EBADF;
|
| 160 | sb = f.file->f_dentry->d_sb;
|
| 161 |
|
| 162 | down_read(&sb->s_umount);
|
| 163 | ret = sync_filesystem(sb);
|
| 164 | up_read(&sb->s_umount);
|
| 165 |
|
| 166 | fdput(f);
|
| 167 | return ret;
|
| 168 | }
|
| 169 |
|
| 170 | |
| 171 | * vfs_fsync_range - helper to sync a range of data & metadata to disk
|
| 172 | * @file: file to sync
|
| 173 | * @start: offset in bytes of the beginning of data range to sync
|
| 174 | * @end: offset in bytes of the end of data range (inclusive)
|
| 175 | * @datasync: perform only datasync
|
| 176 | *
|
| 177 | * Write back data in range @start..@end and metadata for @file to disk. If
|
| 178 | * @datasync is set only metadata needed to access modified file data is
|
| 179 | * written.
|
| 180 |
|
| 181 | int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
|
| 182 | {
|
| 183 | if (!file->f_op->fsync)
|
| 184 | return -EINVAL;
|
| 185 | return file->f_op->fsync(file, start, end, datasync);
|
| 186 | }
|
| 187 | EXPORT_SYMBOL(vfs_fsync_range);
|
| 188 |
|
| 189 | |
| 190 | * vfs_fsync - perform a fsync or fdatasync on a file
|
| 191 | * @file: file to sync
|
| 192 | * @datasync: only perform a fdatasync operation
|
| 193 | *
|
| 194 | * Write back data and metadata for @file to disk. If @datasync is
|
| 195 | * set only metadata needed to access modified file data is written.
|
| 196 |
|
| 197 | int vfs_fsync(struct file *file, int datasync)
|
| 198 | {
|
| 199 | return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
|
| 200 | }
|
| 201 | EXPORT_SYMBOL(vfs_fsync);
|
| 202 |
|
| 203 | static int do_fsync(unsigned int fd, int datasync)
|
| 204 | {
|
| 205 | struct fd f = fdget(fd);
|
| 206 | int ret = -EBADF;
|
| 207 |
|
| 208 | if (f.file) {
|
| 209 | ret = vfs_fsync(f.file, datasync);
|
| 210 | fdput(f);
|
| 211 | }
|
| 212 | return ret;
|
| 213 | }
|
| 214 |
|
| 215 | SYSCALL_DEFINE1(fsync, unsigned int, fd)
|
| 216 | {
|
| 217 | return do_fsync(fd, 0);
|
| 218 | }
|
| 219 |
|
| 220 | SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
|
| 221 | {
|
| 222 | return do_fsync(fd, 1);
|
| 223 | }
|
| 224 |
|
| 225 | |
| 226 | * generic_write_sync - perform syncing after a write if file / inode is sync
|
| 227 | * @file: file to which the write happened
|
| 228 | * @pos: offset where the write started
|
| 229 | * @count: length of the write
|
| 230 | *
|
| 231 | * This is just a simple wrapper about our general syncing function.
|
| 232 |
|
| 233 | int generic_write_sync(struct file *file, loff_t pos, loff_t count)
|
| 234 | {
|
| 235 | if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
|
| 236 | return 0;
|
| 237 | return vfs_fsync_range(file, pos, pos + count - 1,
|
| 238 | (file->f_flags & __O_SYNC) ? 0 : 1);
|
| 239 | }
|
| 240 | EXPORT_SYMBOL(generic_write_sync);
|
| 241 |
|
| 242 | |
| 243 | * sys_sync_file_range() permits finely controlled syncing over a segment of
|
| 244 | * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
|
| 245 | * zero then sys_sync_file_range() will operate from offset out to EOF.
|
| 246 | *
|
| 247 | * The flag bits are:
|
| 248 | *
|
| 249 | * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
|
| 250 | * before performing the write.
|
| 251 | *
|
| 252 | * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
|
| 253 | * range which are not presently under writeback. Note that this may block for
|
| 254 | * significant periods due to exhaustion of disk request structures.
|
| 255 | *
|
| 256 | * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
|
| 257 | * after performing the write.
|
| 258 | *
|
| 259 | * Useful combinations of the flag bits are:
|
| 260 | *
|
| 261 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
|
| 262 | * in the range which were dirty on entry to sys_sync_file_range() are placed
|
| 263 | * under writeout. This is a start-write-for-data-integrity operation.
|
| 264 | *
|
| 265 | * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
|
| 266 | * are not presently under writeout. This is an asynchronous flush-to-disk
|
| 267 | * operation. Not suitable for data integrity operations.
|
| 268 | *
|
| 269 | * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
|
| 270 | * completion of writeout of all pages in the range. This will be used after an
|
| 271 | * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
|
| 272 | * for that operation to complete and to return the result.
|
| 273 | *
|
| 274 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
|
| 275 | * a traditional sync() operation. This is a write-for-data-integrity operation
|
| 276 | * which will ensure that all pages in the range which were dirty on entry to
|
| 277 | * sys_sync_file_range() are committed to disk.
|
| 278 | *
|
| 279 | *
|
| 280 | * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
|
| 281 | * I/O errors or ENOSPC conditions and will return those to the caller, after
|
| 282 | * clearing the EIO and ENOSPC flags in the address_space.
|
| 283 | *
|
| 284 | * It should be noted that none of these operations write out the file's
|
| 285 | * metadata. So unless the application is strictly performing overwrites of
|
| 286 | * already-instantiated disk blocks, there are no guarantees here that the data
|
| 287 | * will be available after a crash.
|
| 288 |
|
| 289 | SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
|
| 290 | unsigned int, flags)
|
| 291 | {
|
| 292 | int ret;
|
| 293 | struct fd f;
|
| 294 | struct address_space *mapping;
|
| 295 | loff_t endbyte;
|
| 296 | umode_t i_mode;
|
| 297 |
|
| 298 | ret = -EINVAL;
|
| 299 | if (flags & ~VALID_FLAGS)
|
| 300 | goto out;
|
| 301 |
|
| 302 | endbyte = offset + nbytes;
|
| 303 |
|
| 304 | if ((s64)offset < 0)
|
| 305 | goto out;
|
| 306 | if ((s64)endbyte < 0)
|
| 307 | goto out;
|
| 308 | if (endbyte < offset)
|
| 309 | goto out;
|
| 310 |
|
| 311 | if (sizeof(pgoff_t) == 4) {
|
| 312 | if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
|
| 313 | |
| 314 | * The range starts outside a 32 bit machine's
|
| 315 | * pagecache addressing capabilities. Let it "succeed"
|
| 316 |
|
| 317 | ret = 0;
|
| 318 | goto out;
|
| 319 | }
|
| 320 | if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
|
| 321 | |
| 322 | * Out to EOF
|
| 323 |
|
| 324 | nbytes = 0;
|
| 325 | }
|
| 326 | }
|
| 327 |
|
| 328 | if (nbytes == 0)
|
| 329 | endbyte = LLONG_MAX;
|
| 330 | else
|
| 331 | endbyte--;
|
| 332 |
|
| 333 | ret = -EBADF;
|
| 334 | f = fdget(fd);
|
| 335 | if (!f.file)
|
| 336 | goto out;
|
| 337 |
|
| 338 | i_mode = file_inode(f.file)->i_mode;
|
| 339 | ret = -ESPIPE;
|
| 340 | if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
|
| 341 | !S_ISLNK(i_mode))
|
| 342 | goto out_put;
|
| 343 |
|
| 344 | mapping = f.file->f_mapping;
|
| 345 | if (!mapping) {
|
| 346 | ret = -EINVAL;
|
| 347 | goto out_put;
|
| 348 | }
|
| 349 |
|
| 350 | ret = 0;
|
| 351 | if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
|
| 352 | ret = filemap_fdatawait_range(mapping, offset, endbyte);
|
| 353 | if (ret < 0)
|
| 354 | goto out_put;
|
| 355 | }
|
| 356 |
|
| 357 | if (flags & SYNC_FILE_RANGE_WRITE) {
|
| 358 | ret = filemap_fdatawrite_range(mapping, offset, endbyte);
|
| 359 | if (ret < 0)
|
| 360 | goto out_put;
|
| 361 | }
|
| 362 |
|
| 363 | if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
|
| 364 | ret = filemap_fdatawait_range(mapping, offset, endbyte);
|
| 365 |
|
| 366 | out_put:
|
| 367 | fdput(f);
|
| 368 | out:
|
| 369 | return ret;
|
| 370 | }
|
| 371 |
|
| 372 | |
| 373 | when they introduce new system calls
|
| 374 | SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
|
| 375 | loff_t, offset, loff_t, nbytes)
|
| 376 | {
|
| 377 | return sys_sync_file_range(fd, offset, nbytes, flags);
|
| 378 | }
|