| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 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_op->sync_fs) |
| 80 | sb->s_op->sync_fs(sb, *(int *)arg); |
| 81 | } |
| 82 | |
| 83 | static void fdatawrite_one_bdev(struct block_device *bdev, void *arg) |
| 84 | { |
| 85 | filemap_fdatawrite(bdev->bd_inode->i_mapping); |
| 86 | } |
| 87 | |
| 88 | static void fdatawait_one_bdev(struct block_device *bdev, void *arg) |
| 89 | { |
| 90 | /* |
| 91 | * We keep the error status of individual mapping so that |
| 92 | * applications can catch the writeback error using fsync(2). |
| 93 | * See filemap_fdatawait_keep_errors() for details. |
| 94 | */ |
| 95 | filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping); |
| 96 | } |
| 97 | |
| 98 | /* |
| 99 | * Sync everything. We start by waking flusher threads so that most of |
| 100 | * writeback runs on all devices in parallel. Then we sync all inodes reliably |
| 101 | * which effectively also waits for all flusher threads to finish doing |
| 102 | * writeback. At this point all data is on disk so metadata should be stable |
| 103 | * and we tell filesystems to sync their metadata via ->sync_fs() calls. |
| 104 | * Finally, we writeout all block devices because some filesystems (e.g. ext2) |
| 105 | * just write metadata (such as inodes or bitmaps) to block device page cache |
| 106 | * and do not sync it on their own in ->sync_fs(). |
| 107 | */ |
| 108 | void ksys_sync(void) |
| 109 | { |
| 110 | int nowait = 0, wait = 1; |
| 111 | |
| 112 | wakeup_flusher_threads(WB_REASON_SYNC); |
| 113 | iterate_supers(sync_inodes_one_sb, NULL); |
| 114 | iterate_supers(sync_fs_one_sb, &nowait); |
| 115 | iterate_supers(sync_fs_one_sb, &wait); |
| 116 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
| 117 | iterate_bdevs(fdatawait_one_bdev, NULL); |
| 118 | if (unlikely(laptop_mode)) |
| 119 | laptop_sync_completion(); |
| 120 | } |
| 121 | |
| 122 | SYSCALL_DEFINE0(sync) |
| 123 | { |
| 124 | ksys_sync(); |
| 125 | return 0; |
| 126 | } |
| 127 | |
| 128 | static void do_sync_work(struct work_struct *work) |
| 129 | { |
| 130 | int nowait = 0; |
| 131 | |
| 132 | /* |
| 133 | * Sync twice to reduce the possibility we skipped some inodes / pages |
| 134 | * because they were temporarily locked |
| 135 | */ |
| 136 | iterate_supers(sync_inodes_one_sb, &nowait); |
| 137 | iterate_supers(sync_fs_one_sb, &nowait); |
| 138 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
| 139 | iterate_supers(sync_inodes_one_sb, &nowait); |
| 140 | iterate_supers(sync_fs_one_sb, &nowait); |
| 141 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
| 142 | printk("Emergency Sync complete\n"); |
| 143 | kfree(work); |
| 144 | } |
| 145 | |
| 146 | void emergency_sync(void) |
| 147 | { |
| 148 | struct work_struct *work; |
| 149 | |
| 150 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
| 151 | if (work) { |
| 152 | INIT_WORK(work, do_sync_work); |
| 153 | schedule_work(work); |
| 154 | } |
| 155 | } |
| 156 | |
| 157 | /* |
| 158 | * sync a single super |
| 159 | */ |
| 160 | SYSCALL_DEFINE1(syncfs, int, fd) |
| 161 | { |
| 162 | struct fd f = fdget(fd); |
| 163 | struct super_block *sb; |
| 164 | int ret; |
| 165 | |
| 166 | if (!f.file) |
| 167 | return -EBADF; |
| 168 | sb = f.file->f_path.dentry->d_sb; |
| 169 | |
| 170 | down_read(&sb->s_umount); |
| 171 | ret = sync_filesystem(sb); |
| 172 | up_read(&sb->s_umount); |
| 173 | |
| 174 | fdput(f); |
| 175 | return ret; |
| 176 | } |
| 177 | |
| 178 | /** |
| 179 | * vfs_fsync_range - helper to sync a range of data & metadata to disk |
| 180 | * @file: file to sync |
| 181 | * @start: offset in bytes of the beginning of data range to sync |
| 182 | * @end: offset in bytes of the end of data range (inclusive) |
| 183 | * @datasync: perform only datasync |
| 184 | * |
| 185 | * Write back data in range @start..@end and metadata for @file to disk. If |
| 186 | * @datasync is set only metadata needed to access modified file data is |
| 187 | * written. |
| 188 | */ |
| 189 | int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync) |
| 190 | { |
| 191 | struct inode *inode = file->f_mapping->host; |
| 192 | |
| 193 | if (!file->f_op->fsync) |
| 194 | return -EINVAL; |
| 195 | if (!datasync && (inode->i_state & I_DIRTY_TIME)) |
| 196 | mark_inode_dirty_sync(inode); |
| 197 | return file->f_op->fsync(file, start, end, datasync); |
| 198 | } |
| 199 | EXPORT_SYMBOL(vfs_fsync_range); |
| 200 | |
| 201 | /** |
| 202 | * vfs_fsync - perform a fsync or fdatasync on a file |
| 203 | * @file: file to sync |
| 204 | * @datasync: only perform a fdatasync operation |
| 205 | * |
| 206 | * Write back data and metadata for @file to disk. If @datasync is |
| 207 | * set only metadata needed to access modified file data is written. |
| 208 | */ |
| 209 | int vfs_fsync(struct file *file, int datasync) |
| 210 | { |
| 211 | return vfs_fsync_range(file, 0, LLONG_MAX, datasync); |
| 212 | } |
| 213 | EXPORT_SYMBOL(vfs_fsync); |
| 214 | |
| 215 | static int do_fsync(unsigned int fd, int datasync) |
| 216 | { |
| 217 | struct fd f = fdget(fd); |
| 218 | int ret = -EBADF; |
| 219 | |
| 220 | if (f.file) { |
| 221 | ret = vfs_fsync(f.file, datasync); |
| 222 | fdput(f); |
| 223 | } |
| 224 | return ret; |
| 225 | } |
| 226 | |
| 227 | SYSCALL_DEFINE1(fsync, unsigned int, fd) |
| 228 | { |
| 229 | return do_fsync(fd, 0); |
| 230 | } |
| 231 | |
| 232 | SYSCALL_DEFINE1(fdatasync, unsigned int, fd) |
| 233 | { |
| 234 | return do_fsync(fd, 1); |
| 235 | } |
| 236 | |
| 237 | int sync_file_range(struct file *file, loff_t offset, loff_t nbytes, |
| 238 | unsigned int flags) |
| 239 | { |
| 240 | int ret; |
| 241 | struct address_space *mapping; |
| 242 | loff_t endbyte; /* inclusive */ |
| 243 | umode_t i_mode; |
| 244 | |
| 245 | ret = -EINVAL; |
| 246 | if (flags & ~VALID_FLAGS) |
| 247 | goto out; |
| 248 | |
| 249 | endbyte = offset + nbytes; |
| 250 | |
| 251 | if ((s64)offset < 0) |
| 252 | goto out; |
| 253 | if ((s64)endbyte < 0) |
| 254 | goto out; |
| 255 | if (endbyte < offset) |
| 256 | goto out; |
| 257 | |
| 258 | if (sizeof(pgoff_t) == 4) { |
| 259 | if (offset >= (0x100000000ULL << PAGE_SHIFT)) { |
| 260 | /* |
| 261 | * The range starts outside a 32 bit machine's |
| 262 | * pagecache addressing capabilities. Let it "succeed" |
| 263 | */ |
| 264 | ret = 0; |
| 265 | goto out; |
| 266 | } |
| 267 | if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) { |
| 268 | /* |
| 269 | * Out to EOF |
| 270 | */ |
| 271 | nbytes = 0; |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | if (nbytes == 0) |
| 276 | endbyte = LLONG_MAX; |
| 277 | else |
| 278 | endbyte--; /* inclusive */ |
| 279 | |
| 280 | i_mode = file_inode(file)->i_mode; |
| 281 | ret = -ESPIPE; |
| 282 | if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) && |
| 283 | !S_ISLNK(i_mode)) |
| 284 | goto out; |
| 285 | |
| 286 | mapping = file->f_mapping; |
| 287 | ret = 0; |
| 288 | if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) { |
| 289 | ret = file_fdatawait_range(file, offset, endbyte); |
| 290 | if (ret < 0) |
| 291 | goto out; |
| 292 | } |
| 293 | |
| 294 | if (flags & SYNC_FILE_RANGE_WRITE) { |
| 295 | int sync_mode = WB_SYNC_NONE; |
| 296 | |
| 297 | if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) == |
| 298 | SYNC_FILE_RANGE_WRITE_AND_WAIT) |
| 299 | sync_mode = WB_SYNC_ALL; |
| 300 | |
| 301 | ret = __filemap_fdatawrite_range(mapping, offset, endbyte, |
| 302 | sync_mode); |
| 303 | if (ret < 0) |
| 304 | goto out; |
| 305 | } |
| 306 | |
| 307 | if (flags & SYNC_FILE_RANGE_WAIT_AFTER) |
| 308 | ret = file_fdatawait_range(file, offset, endbyte); |
| 309 | |
| 310 | out: |
| 311 | return ret; |
| 312 | } |
| 313 | |
| 314 | /* |
| 315 | * ksys_sync_file_range() permits finely controlled syncing over a segment of |
| 316 | * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is |
| 317 | * zero then ksys_sync_file_range() will operate from offset out to EOF. |
| 318 | * |
| 319 | * The flag bits are: |
| 320 | * |
| 321 | * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range |
| 322 | * before performing the write. |
| 323 | * |
| 324 | * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the |
| 325 | * range which are not presently under writeback. Note that this may block for |
| 326 | * significant periods due to exhaustion of disk request structures. |
| 327 | * |
| 328 | * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range |
| 329 | * after performing the write. |
| 330 | * |
| 331 | * Useful combinations of the flag bits are: |
| 332 | * |
| 333 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages |
| 334 | * in the range which were dirty on entry to ksys_sync_file_range() are placed |
| 335 | * under writeout. This is a start-write-for-data-integrity operation. |
| 336 | * |
| 337 | * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which |
| 338 | * are not presently under writeout. This is an asynchronous flush-to-disk |
| 339 | * operation. Not suitable for data integrity operations. |
| 340 | * |
| 341 | * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for |
| 342 | * completion of writeout of all pages in the range. This will be used after an |
| 343 | * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait |
| 344 | * for that operation to complete and to return the result. |
| 345 | * |
| 346 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER |
| 347 | * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT): |
| 348 | * a traditional sync() operation. This is a write-for-data-integrity operation |
| 349 | * which will ensure that all pages in the range which were dirty on entry to |
| 350 | * ksys_sync_file_range() are written to disk. It should be noted that disk |
| 351 | * caches are not flushed by this call, so there are no guarantees here that the |
| 352 | * data will be available on disk after a crash. |
| 353 | * |
| 354 | * |
| 355 | * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any |
| 356 | * I/O errors or ENOSPC conditions and will return those to the caller, after |
| 357 | * clearing the EIO and ENOSPC flags in the address_space. |
| 358 | * |
| 359 | * It should be noted that none of these operations write out the file's |
| 360 | * metadata. So unless the application is strictly performing overwrites of |
| 361 | * already-instantiated disk blocks, there are no guarantees here that the data |
| 362 | * will be available after a crash. |
| 363 | */ |
| 364 | int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes, |
| 365 | unsigned int flags) |
| 366 | { |
| 367 | int ret; |
| 368 | struct fd f; |
| 369 | |
| 370 | ret = -EBADF; |
| 371 | f = fdget(fd); |
| 372 | if (f.file) |
| 373 | ret = sync_file_range(f.file, offset, nbytes, flags); |
| 374 | |
| 375 | fdput(f); |
| 376 | return ret; |
| 377 | } |
| 378 | |
| 379 | SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes, |
| 380 | unsigned int, flags) |
| 381 | { |
| 382 | return ksys_sync_file_range(fd, offset, nbytes, flags); |
| 383 | } |
| 384 | |
| 385 | /* It would be nice if people remember that not all the world's an i386 |
| 386 | when they introduce new system calls */ |
| 387 | SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags, |
| 388 | loff_t, offset, loff_t, nbytes) |
| 389 | { |
| 390 | return ksys_sync_file_range(fd, offset, nbytes, flags); |
| 391 | } |