| 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/module.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/buffer_head.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 | /* |
| 34 | * This should be safe, as we require bdi backing to actually |
| 35 | * write out data in the first place |
| 36 | */ |
| 37 | if (sb->s_bdi == &noop_backing_dev_info) |
| 38 | return 0; |
| 39 | |
| 40 | if (sb->s_qcop && sb->s_qcop->quota_sync) |
| 41 | sb->s_qcop->quota_sync(sb, -1, wait); |
| 42 | |
| 43 | if (wait) |
| 44 | sync_inodes_sb(sb); |
| 45 | else |
| 46 | writeback_inodes_sb(sb); |
| 47 | |
| 48 | if (sb->s_op->sync_fs) |
| 49 | sb->s_op->sync_fs(sb, wait); |
| 50 | return __sync_blockdev(sb->s_bdev, wait); |
| 51 | } |
| 52 | |
| 53 | /* |
| 54 | * Write out and wait upon all dirty data associated with this |
| 55 | * superblock. Filesystem data as well as the underlying block |
| 56 | * device. Takes the superblock lock. |
| 57 | */ |
| 58 | int sync_filesystem(struct super_block *sb) |
| 59 | { |
| 60 | int ret; |
| 61 | |
| 62 | /* |
| 63 | * We need to be protected against the filesystem going from |
| 64 | * r/o to r/w or vice versa. |
| 65 | */ |
| 66 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 67 | |
| 68 | /* |
| 69 | * No point in syncing out anything if the filesystem is read-only. |
| 70 | */ |
| 71 | if (sb->s_flags & MS_RDONLY) |
| 72 | return 0; |
| 73 | |
| 74 | ret = __sync_filesystem(sb, 0); |
| 75 | if (ret < 0) |
| 76 | return ret; |
| 77 | return __sync_filesystem(sb, 1); |
| 78 | } |
| 79 | EXPORT_SYMBOL_GPL(sync_filesystem); |
| 80 | |
| 81 | static void sync_one_sb(struct super_block *sb, void *arg) |
| 82 | { |
| 83 | if (!(sb->s_flags & MS_RDONLY)) |
| 84 | __sync_filesystem(sb, *(int *)arg); |
| 85 | } |
| 86 | /* |
| 87 | * Sync all the data for all the filesystems (called by sys_sync() and |
| 88 | * emergency sync) |
| 89 | */ |
| 90 | static void sync_filesystems(int wait) |
| 91 | { |
| 92 | iterate_supers(sync_one_sb, &wait); |
| 93 | } |
| 94 | |
| 95 | /* |
| 96 | * sync everything. Start out by waking pdflush, because that writes back |
| 97 | * all queues in parallel. |
| 98 | */ |
| 99 | SYSCALL_DEFINE0(sync) |
| 100 | { |
| 101 | wakeup_flusher_threads(0); |
| 102 | sync_filesystems(0); |
| 103 | sync_filesystems(1); |
| 104 | if (unlikely(laptop_mode)) |
| 105 | laptop_sync_completion(); |
| 106 | return 0; |
| 107 | } |
| 108 | |
| 109 | static void do_sync_work(struct work_struct *work) |
| 110 | { |
| 111 | /* |
| 112 | * Sync twice to reduce the possibility we skipped some inodes / pages |
| 113 | * because they were temporarily locked |
| 114 | */ |
| 115 | sync_filesystems(0); |
| 116 | sync_filesystems(0); |
| 117 | printk("Emergency Sync complete\n"); |
| 118 | kfree(work); |
| 119 | } |
| 120 | |
| 121 | void emergency_sync(void) |
| 122 | { |
| 123 | struct work_struct *work; |
| 124 | |
| 125 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
| 126 | if (work) { |
| 127 | INIT_WORK(work, do_sync_work); |
| 128 | schedule_work(work); |
| 129 | } |
| 130 | } |
| 131 | |
| 132 | /* |
| 133 | * sync a single super |
| 134 | */ |
| 135 | SYSCALL_DEFINE1(syncfs, int, fd) |
| 136 | { |
| 137 | struct file *file; |
| 138 | struct super_block *sb; |
| 139 | int ret; |
| 140 | int fput_needed; |
| 141 | |
| 142 | file = fget_light(fd, &fput_needed); |
| 143 | if (!file) |
| 144 | return -EBADF; |
| 145 | sb = file->f_dentry->d_sb; |
| 146 | |
| 147 | down_read(&sb->s_umount); |
| 148 | ret = sync_filesystem(sb); |
| 149 | up_read(&sb->s_umount); |
| 150 | |
| 151 | fput_light(file, fput_needed); |
| 152 | return ret; |
| 153 | } |
| 154 | |
| 155 | /** |
| 156 | * vfs_fsync_range - helper to sync a range of data & metadata to disk |
| 157 | * @file: file to sync |
| 158 | * @start: offset in bytes of the beginning of data range to sync |
| 159 | * @end: offset in bytes of the end of data range (inclusive) |
| 160 | * @datasync: perform only datasync |
| 161 | * |
| 162 | * Write back data in range @start..@end and metadata for @file to disk. If |
| 163 | * @datasync is set only metadata needed to access modified file data is |
| 164 | * written. |
| 165 | */ |
| 166 | int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync) |
| 167 | { |
| 168 | struct address_space *mapping = file->f_mapping; |
| 169 | int err, ret; |
| 170 | |
| 171 | if (!file->f_op || !file->f_op->fsync) { |
| 172 | ret = -EINVAL; |
| 173 | goto out; |
| 174 | } |
| 175 | |
| 176 | ret = filemap_write_and_wait_range(mapping, start, end); |
| 177 | |
| 178 | /* |
| 179 | * We need to protect against concurrent writers, which could cause |
| 180 | * livelocks in fsync_buffers_list(). |
| 181 | */ |
| 182 | mutex_lock(&mapping->host->i_mutex); |
| 183 | err = file->f_op->fsync(file, datasync); |
| 184 | if (!ret) |
| 185 | ret = err; |
| 186 | mutex_unlock(&mapping->host->i_mutex); |
| 187 | |
| 188 | out: |
| 189 | return ret; |
| 190 | } |
| 191 | EXPORT_SYMBOL(vfs_fsync_range); |
| 192 | |
| 193 | /** |
| 194 | * vfs_fsync - perform a fsync or fdatasync on a file |
| 195 | * @file: file to sync |
| 196 | * @datasync: only perform a fdatasync operation |
| 197 | * |
| 198 | * Write back data and metadata for @file to disk. If @datasync is |
| 199 | * set only metadata needed to access modified file data is written. |
| 200 | */ |
| 201 | int vfs_fsync(struct file *file, int datasync) |
| 202 | { |
| 203 | return vfs_fsync_range(file, 0, LLONG_MAX, datasync); |
| 204 | } |
| 205 | EXPORT_SYMBOL(vfs_fsync); |
| 206 | |
| 207 | static int do_fsync(unsigned int fd, int datasync) |
| 208 | { |
| 209 | struct file *file; |
| 210 | int ret = -EBADF; |
| 211 | |
| 212 | file = fget(fd); |
| 213 | if (file) { |
| 214 | ret = vfs_fsync(file, datasync); |
| 215 | fput(file); |
| 216 | } |
| 217 | return ret; |
| 218 | } |
| 219 | |
| 220 | SYSCALL_DEFINE1(fsync, unsigned int, fd) |
| 221 | { |
| 222 | return do_fsync(fd, 0); |
| 223 | } |
| 224 | |
| 225 | SYSCALL_DEFINE1(fdatasync, unsigned int, fd) |
| 226 | { |
| 227 | return do_fsync(fd, 1); |
| 228 | } |
| 229 | |
| 230 | /** |
| 231 | * generic_write_sync - perform syncing after a write if file / inode is sync |
| 232 | * @file: file to which the write happened |
| 233 | * @pos: offset where the write started |
| 234 | * @count: length of the write |
| 235 | * |
| 236 | * This is just a simple wrapper about our general syncing function. |
| 237 | */ |
| 238 | int generic_write_sync(struct file *file, loff_t pos, loff_t count) |
| 239 | { |
| 240 | if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host)) |
| 241 | return 0; |
| 242 | return vfs_fsync_range(file, pos, pos + count - 1, |
| 243 | (file->f_flags & __O_SYNC) ? 0 : 1); |
| 244 | } |
| 245 | EXPORT_SYMBOL(generic_write_sync); |
| 246 | |
| 247 | /* |
| 248 | * sys_sync_file_range() permits finely controlled syncing over a segment of |
| 249 | * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is |
| 250 | * zero then sys_sync_file_range() will operate from offset out to EOF. |
| 251 | * |
| 252 | * The flag bits are: |
| 253 | * |
| 254 | * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range |
| 255 | * before performing the write. |
| 256 | * |
| 257 | * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the |
| 258 | * range which are not presently under writeback. Note that this may block for |
| 259 | * significant periods due to exhaustion of disk request structures. |
| 260 | * |
| 261 | * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range |
| 262 | * after performing the write. |
| 263 | * |
| 264 | * Useful combinations of the flag bits are: |
| 265 | * |
| 266 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages |
| 267 | * in the range which were dirty on entry to sys_sync_file_range() are placed |
| 268 | * under writeout. This is a start-write-for-data-integrity operation. |
| 269 | * |
| 270 | * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which |
| 271 | * are not presently under writeout. This is an asynchronous flush-to-disk |
| 272 | * operation. Not suitable for data integrity operations. |
| 273 | * |
| 274 | * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for |
| 275 | * completion of writeout of all pages in the range. This will be used after an |
| 276 | * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait |
| 277 | * for that operation to complete and to return the result. |
| 278 | * |
| 279 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER: |
| 280 | * a traditional sync() operation. This is a write-for-data-integrity operation |
| 281 | * which will ensure that all pages in the range which were dirty on entry to |
| 282 | * sys_sync_file_range() are committed to disk. |
| 283 | * |
| 284 | * |
| 285 | * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any |
| 286 | * I/O errors or ENOSPC conditions and will return those to the caller, after |
| 287 | * clearing the EIO and ENOSPC flags in the address_space. |
| 288 | * |
| 289 | * It should be noted that none of these operations write out the file's |
| 290 | * metadata. So unless the application is strictly performing overwrites of |
| 291 | * already-instantiated disk blocks, there are no guarantees here that the data |
| 292 | * will be available after a crash. |
| 293 | */ |
| 294 | SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes, |
| 295 | unsigned int flags) |
| 296 | { |
| 297 | int ret; |
| 298 | struct file *file; |
| 299 | struct address_space *mapping; |
| 300 | loff_t endbyte; /* inclusive */ |
| 301 | int fput_needed; |
| 302 | umode_t i_mode; |
| 303 | |
| 304 | ret = -EINVAL; |
| 305 | if (flags & ~VALID_FLAGS) |
| 306 | goto out; |
| 307 | |
| 308 | endbyte = offset + nbytes; |
| 309 | |
| 310 | if ((s64)offset < 0) |
| 311 | goto out; |
| 312 | if ((s64)endbyte < 0) |
| 313 | goto out; |
| 314 | if (endbyte < offset) |
| 315 | goto out; |
| 316 | |
| 317 | if (sizeof(pgoff_t) == 4) { |
| 318 | if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) { |
| 319 | /* |
| 320 | * The range starts outside a 32 bit machine's |
| 321 | * pagecache addressing capabilities. Let it "succeed" |
| 322 | */ |
| 323 | ret = 0; |
| 324 | goto out; |
| 325 | } |
| 326 | if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) { |
| 327 | /* |
| 328 | * Out to EOF |
| 329 | */ |
| 330 | nbytes = 0; |
| 331 | } |
| 332 | } |
| 333 | |
| 334 | if (nbytes == 0) |
| 335 | endbyte = LLONG_MAX; |
| 336 | else |
| 337 | endbyte--; /* inclusive */ |
| 338 | |
| 339 | ret = -EBADF; |
| 340 | file = fget_light(fd, &fput_needed); |
| 341 | if (!file) |
| 342 | goto out; |
| 343 | |
| 344 | i_mode = file->f_path.dentry->d_inode->i_mode; |
| 345 | ret = -ESPIPE; |
| 346 | if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) && |
| 347 | !S_ISLNK(i_mode)) |
| 348 | goto out_put; |
| 349 | |
| 350 | mapping = file->f_mapping; |
| 351 | if (!mapping) { |
| 352 | ret = -EINVAL; |
| 353 | goto out_put; |
| 354 | } |
| 355 | |
| 356 | ret = 0; |
| 357 | if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) { |
| 358 | ret = filemap_fdatawait_range(mapping, offset, endbyte); |
| 359 | if (ret < 0) |
| 360 | goto out_put; |
| 361 | } |
| 362 | |
| 363 | if (flags & SYNC_FILE_RANGE_WRITE) { |
| 364 | ret = filemap_fdatawrite_range(mapping, offset, endbyte); |
| 365 | if (ret < 0) |
| 366 | goto out_put; |
| 367 | } |
| 368 | |
| 369 | if (flags & SYNC_FILE_RANGE_WAIT_AFTER) |
| 370 | ret = filemap_fdatawait_range(mapping, offset, endbyte); |
| 371 | |
| 372 | out_put: |
| 373 | fput_light(file, fput_needed); |
| 374 | out: |
| 375 | return ret; |
| 376 | } |
| 377 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
| 378 | asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes, |
| 379 | long flags) |
| 380 | { |
| 381 | return SYSC_sync_file_range((int) fd, offset, nbytes, |
| 382 | (unsigned int) flags); |
| 383 | } |
| 384 | SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range); |
| 385 | #endif |
| 386 | |
| 387 | /* It would be nice if people remember that not all the world's an i386 |
| 388 | when they introduce new system calls */ |
| 389 | SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags, |
| 390 | loff_t offset, loff_t nbytes) |
| 391 | { |
| 392 | return sys_sync_file_range(fd, offset, nbytes, flags); |
| 393 | } |
| 394 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
| 395 | asmlinkage long SyS_sync_file_range2(long fd, long flags, |
| 396 | loff_t offset, loff_t nbytes) |
| 397 | { |
| 398 | return SYSC_sync_file_range2((int) fd, (unsigned int) flags, |
| 399 | offset, nbytes); |
| 400 | } |
| 401 | SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2); |
| 402 | #endif |