| 1 | |
| 2 | |
| 3 | * High-level sync()-related operations
|
| 4 |
|
| 5 |
|
| 6 | #include <linux/kernel.h>
|
| 7 | #include <linux/file.h>
|
| 8 | #include <linux/fs.h>
|
| 9 | #include <linux/slab.h>
|
| 10 | #include <linux/export.h>
|
| 11 | #include <linux/namei.h>
|
| 12 | #include <linux/sched.h>
|
| 13 | #include <linux/writeback.h>
|
| 14 | #include <linux/syscalls.h>
|
| 15 | #include <linux/linkage.h>
|
| 16 | #include <linux/pagemap.h>
|
| 17 | #include <linux/quotaops.h>
|
| 18 | #include <linux/backing-dev.h>
|
| 19 | #include "internal.h"
|
| 20 |
|
| 21 | #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
|
| 22 | SYNC_FILE_RANGE_WAIT_AFTER)
|
| 23 |
|
| 24 | |
| 25 | * Do the filesystem syncing work. For simple filesystems
|
| 26 | * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
|
| 27 | * submit IO for these buffers via __sync_blockdev(). This also speeds up the
|
| 28 | * wait == 1 case since in that case write_inode() functions do
|
| 29 | * sync_dirty_buffer() and thus effectively write one block at a time.
|
| 30 |
|
| 31 | static int __sync_filesystem(struct super_block *sb, int wait)
|
| 32 | {
|
| 33 | if (wait)
|
| 34 | sync_inodes_sb(sb);
|
| 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 |
|
| 52 | |
| 53 | * We need to be protected against the filesystem going from
|
| 54 | * r/o to r/w or vice versa.
|
| 55 |
|
| 56 | WARN_ON(!rwsem_is_locked(&sb->s_umount));
|
| 57 |
|
| 58 | |
| 59 | * No point in syncing out anything if the filesystem is read-only.
|
| 60 |
|
| 61 | if (sb_rdonly(sb))
|
| 62 | return 0;
|
| 63 |
|
| 64 | ret = __sync_filesystem(sb, 0);
|
| 65 | if (ret < 0)
|
| 66 | return ret;
|
| 67 | return __sync_filesystem(sb, 1);
|
| 68 | }
|
| 69 | EXPORT_SYMBOL(sync_filesystem);
|
| 70 |
|
| 71 | static void sync_inodes_one_sb(struct super_block *sb, void *arg)
|
| 72 | {
|
| 73 | if (!sb_rdonly(sb))
|
| 74 | sync_inodes_sb(sb);
|
| 75 | }
|
| 76 |
|
| 77 | static void sync_fs_one_sb(struct super_block *sb, void *arg)
|
| 78 | {
|
| 79 | if (!sb_rdonly(sb) && !(sb->s_iflags & SB_I_SKIP_SYNC) &&
|
| 80 | 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 | |
| 92 | * We keep the error status of individual mapping so that
|
| 93 | * applications can catch the writeback error using fsync(2).
|
| 94 | * See filemap_fdatawait_keep_errors() for details.
|
| 95 |
|
| 96 | filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping);
|
| 97 | }
|
| 98 |
|
| 99 | |
| 100 | * Sync everything. We start by waking flusher threads so that most of
|
| 101 | * writeback runs on all devices in parallel. Then we sync all inodes reliably
|
| 102 | * which effectively also waits for all flusher threads to finish doing
|
| 103 | * writeback. At this point all data is on disk so metadata should be stable
|
| 104 | * and we tell filesystems to sync their metadata via ->sync_fs() calls.
|
| 105 | * Finally, we writeout all block devices because some filesystems (e.g. ext2)
|
| 106 | * just write metadata (such as inodes or bitmaps) to block device page cache
|
| 107 | * and do not sync it on their own in ->sync_fs().
|
| 108 |
|
| 109 | void ksys_sync(void)
|
| 110 | {
|
| 111 | int nowait = 0, wait = 1;
|
| 112 |
|
| 113 | wakeup_flusher_threads(WB_REASON_SYNC);
|
| 114 | iterate_supers(sync_inodes_one_sb, NULL);
|
| 115 | iterate_supers(sync_fs_one_sb, &nowait);
|
| 116 | iterate_supers(sync_fs_one_sb, &wait);
|
| 117 | iterate_bdevs(fdatawrite_one_bdev, NULL);
|
| 118 | iterate_bdevs(fdatawait_one_bdev, NULL);
|
| 119 | if (unlikely(laptop_mode))
|
| 120 | laptop_sync_completion();
|
| 121 | }
|
| 122 |
|
| 123 | SYSCALL_DEFINE0(sync)
|
| 124 | {
|
| 125 | ksys_sync();
|
| 126 | return 0;
|
| 127 | }
|
| 128 |
|
| 129 | static void do_sync_work(struct work_struct *work)
|
| 130 | {
|
| 131 | int nowait = 0;
|
| 132 |
|
| 133 | |
| 134 | * Sync twice to reduce the possibility we skipped some inodes / pages
|
| 135 | * because they were temporarily locked
|
| 136 |
|
| 137 | iterate_supers(sync_inodes_one_sb, &nowait);
|
| 138 | iterate_supers(sync_fs_one_sb, &nowait);
|
| 139 | iterate_bdevs(fdatawrite_one_bdev, NULL);
|
| 140 | iterate_supers(sync_inodes_one_sb, &nowait);
|
| 141 | iterate_supers(sync_fs_one_sb, &nowait);
|
| 142 | iterate_bdevs(fdatawrite_one_bdev, NULL);
|
| 143 | printk("Emergency Sync complete\n");
|
| 144 | kfree(work);
|
| 145 | }
|
| 146 |
|
| 147 | void emergency_sync(void)
|
| 148 | {
|
| 149 | struct work_struct *work;
|
| 150 |
|
| 151 | work = kmalloc(sizeof(*work), GFP_ATOMIC);
|
| 152 | if (work) {
|
| 153 | INIT_WORK(work, do_sync_work);
|
| 154 | schedule_work(work);
|
| 155 | }
|
| 156 | }
|
| 157 |
|
| 158 | |
| 159 | * sync a single super
|
| 160 |
|
| 161 | SYSCALL_DEFINE1(syncfs, int, fd)
|
| 162 | {
|
| 163 | struct fd f = fdget(fd);
|
| 164 | struct super_block *sb;
|
| 165 | int ret, ret2;
|
| 166 |
|
| 167 | if (!f.file)
|
| 168 | return -EBADF;
|
| 169 | sb = f.file->f_path.dentry->d_sb;
|
| 170 |
|
| 171 | down_read(&sb->s_umount);
|
| 172 | ret = sync_filesystem(sb);
|
| 173 | up_read(&sb->s_umount);
|
| 174 |
|
| 175 | ret2 = errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err);
|
| 176 |
|
| 177 | fdput(f);
|
| 178 | return ret ? ret : ret2;
|
| 179 | }
|
| 180 |
|
| 181 | |
| 182 | * vfs_fsync_range - helper to sync a range of data & metadata to disk
|
| 183 | * @file: file to sync
|
| 184 | * @start: offset in bytes of the beginning of data range to sync
|
| 185 | * @end: offset in bytes of the end of data range (inclusive)
|
| 186 | * @datasync: perform only datasync
|
| 187 | *
|
| 188 | * Write back data in range @start..@end and metadata for @file to disk. If
|
| 189 | * @datasync is set only metadata needed to access modified file data is
|
| 190 | * written.
|
| 191 |
|
| 192 | int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
|
| 193 | {
|
| 194 | struct inode *inode = file->f_mapping->host;
|
| 195 |
|
| 196 | if (!file->f_op->fsync)
|
| 197 | return -EINVAL;
|
| 198 | if (!datasync && (inode->i_state & I_DIRTY_TIME))
|
| 199 | mark_inode_dirty_sync(inode);
|
| 200 | return file->f_op->fsync(file, start, end, datasync);
|
| 201 | }
|
| 202 | EXPORT_SYMBOL(vfs_fsync_range);
|
| 203 |
|
| 204 | |
| 205 | * vfs_fsync - perform a fsync or fdatasync on a file
|
| 206 | * @file: file to sync
|
| 207 | * @datasync: only perform a fdatasync operation
|
| 208 | *
|
| 209 | * Write back data and metadata for @file to disk. If @datasync is
|
| 210 | * set only metadata needed to access modified file data is written.
|
| 211 |
|
| 212 | int vfs_fsync(struct file *file, int datasync)
|
| 213 | {
|
| 214 | return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
|
| 215 | }
|
| 216 | EXPORT_SYMBOL(vfs_fsync);
|
| 217 |
|
| 218 | static int do_fsync(unsigned int fd, int datasync)
|
| 219 | {
|
| 220 | struct fd f = fdget(fd);
|
| 221 | int ret = -EBADF;
|
| 222 |
|
| 223 | if (f.file) {
|
| 224 | ret = vfs_fsync(f.file, datasync);
|
| 225 | fdput(f);
|
| 226 | }
|
| 227 | return ret;
|
| 228 | }
|
| 229 |
|
| 230 | SYSCALL_DEFINE1(fsync, unsigned int, fd)
|
| 231 | {
|
| 232 | return do_fsync(fd, 0);
|
| 233 | }
|
| 234 |
|
| 235 | SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
|
| 236 | {
|
| 237 | return do_fsync(fd, 1);
|
| 238 | }
|
| 239 |
|
| 240 | int sync_file_range(struct file *file, loff_t offset, loff_t nbytes,
|
| 241 | unsigned int flags)
|
| 242 | {
|
| 243 | int ret;
|
| 244 | struct address_space *mapping;
|
| 245 | loff_t endbyte;
|
| 246 | umode_t i_mode;
|
| 247 |
|
| 248 | ret = -EINVAL;
|
| 249 | if (flags & ~VALID_FLAGS)
|
| 250 | goto out;
|
| 251 |
|
| 252 | endbyte = offset + nbytes;
|
| 253 |
|
| 254 | if ((s64)offset < 0)
|
| 255 | goto out;
|
| 256 | if ((s64)endbyte < 0)
|
| 257 | goto out;
|
| 258 | if (endbyte < offset)
|
| 259 | goto out;
|
| 260 |
|
| 261 | if (sizeof(pgoff_t) == 4) {
|
| 262 | if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
|
| 263 | |
| 264 | * The range starts outside a 32 bit machine's
|
| 265 | * pagecache addressing capabilities. Let it "succeed"
|
| 266 |
|
| 267 | ret = 0;
|
| 268 | goto out;
|
| 269 | }
|
| 270 | if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
|
| 271 | |
| 272 | * Out to EOF
|
| 273 |
|
| 274 | nbytes = 0;
|
| 275 | }
|
| 276 | }
|
| 277 |
|
| 278 | if (nbytes == 0)
|
| 279 | endbyte = LLONG_MAX;
|
| 280 | else
|
| 281 | endbyte--;
|
| 282 |
|
| 283 | i_mode = file_inode(file)->i_mode;
|
| 284 | ret = -ESPIPE;
|
| 285 | if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
|
| 286 | !S_ISLNK(i_mode))
|
| 287 | goto out;
|
| 288 |
|
| 289 | mapping = file->f_mapping;
|
| 290 | ret = 0;
|
| 291 | if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
|
| 292 | ret = file_fdatawait_range(file, offset, endbyte);
|
| 293 | if (ret < 0)
|
| 294 | goto out;
|
| 295 | }
|
| 296 |
|
| 297 | if (flags & SYNC_FILE_RANGE_WRITE) {
|
| 298 | int sync_mode = WB_SYNC_NONE;
|
| 299 |
|
| 300 | if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
|
| 301 | SYNC_FILE_RANGE_WRITE_AND_WAIT)
|
| 302 | sync_mode = WB_SYNC_ALL;
|
| 303 |
|
| 304 | ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
|
| 305 | sync_mode);
|
| 306 | if (ret < 0)
|
| 307 | goto out;
|
| 308 | }
|
| 309 |
|
| 310 | if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
|
| 311 | ret = file_fdatawait_range(file, offset, endbyte);
|
| 312 |
|
| 313 | out:
|
| 314 | return ret;
|
| 315 | }
|
| 316 |
|
| 317 | |
| 318 | * ksys_sync_file_range() permits finely controlled syncing over a segment of
|
| 319 | * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
|
| 320 | * zero then ksys_sync_file_range() will operate from offset out to EOF.
|
| 321 | *
|
| 322 | * The flag bits are:
|
| 323 | *
|
| 324 | * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
|
| 325 | * before performing the write.
|
| 326 | *
|
| 327 | * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
|
| 328 | * range which are not presently under writeback. Note that this may block for
|
| 329 | * significant periods due to exhaustion of disk request structures.
|
| 330 | *
|
| 331 | * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
|
| 332 | * after performing the write.
|
| 333 | *
|
| 334 | * Useful combinations of the flag bits are:
|
| 335 | *
|
| 336 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
|
| 337 | * in the range which were dirty on entry to ksys_sync_file_range() are placed
|
| 338 | * under writeout. This is a start-write-for-data-integrity operation.
|
| 339 | *
|
| 340 | * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
|
| 341 | * are not presently under writeout. This is an asynchronous flush-to-disk
|
| 342 | * operation. Not suitable for data integrity operations.
|
| 343 | *
|
| 344 | * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
|
| 345 | * completion of writeout of all pages in the range. This will be used after an
|
| 346 | * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
|
| 347 | * for that operation to complete and to return the result.
|
| 348 | *
|
| 349 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
|
| 350 | * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
|
| 351 | * a traditional sync() operation. This is a write-for-data-integrity operation
|
| 352 | * which will ensure that all pages in the range which were dirty on entry to
|
| 353 | * ksys_sync_file_range() are written to disk. It should be noted that disk
|
| 354 | * caches are not flushed by this call, so there are no guarantees here that the
|
| 355 | * data will be available on disk after a crash.
|
| 356 | *
|
| 357 | *
|
| 358 | * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
|
| 359 | * I/O errors or ENOSPC conditions and will return those to the caller, after
|
| 360 | * clearing the EIO and ENOSPC flags in the address_space.
|
| 361 | *
|
| 362 | * It should be noted that none of these operations write out the file's
|
| 363 | * metadata. So unless the application is strictly performing overwrites of
|
| 364 | * already-instantiated disk blocks, there are no guarantees here that the data
|
| 365 | * will be available after a crash.
|
| 366 |
|
| 367 | int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
|
| 368 | unsigned int flags)
|
| 369 | {
|
| 370 | int ret;
|
| 371 | struct fd f;
|
| 372 |
|
| 373 | ret = -EBADF;
|
| 374 | f = fdget(fd);
|
| 375 | if (f.file)
|
| 376 | ret = sync_file_range(f.file, offset, nbytes, flags);
|
| 377 |
|
| 378 | fdput(f);
|
| 379 | return ret;
|
| 380 | }
|
| 381 |
|
| 382 | SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
|
| 383 | unsigned int, flags)
|
| 384 | {
|
| 385 | return ksys_sync_file_range(fd, offset, nbytes, flags);
|
| 386 | }
|
| 387 |
|
| 388 | |
| 389 | when they introduce new system calls
|
| 390 | SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
|
| 391 | loff_t, offset, loff_t, nbytes)
|
| 392 | {
|
| 393 | return ksys_sync_file_range(fd, offset, nbytes, flags);
|
| 394 | }
|