1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * High-level sync()-related operations 4 */ 5 6 #include <linux/blkdev.h> 7 #include <linux/kernel.h> 8 #include <linux/file.h> 9 #include <linux/fs.h> 10 #include <linux/slab.h> 11 #include <linux/export.h> 12 #include <linux/namei.h> 13 #include <linux/sched.h> 14 #include <linux/writeback.h> 15 #include <linux/syscalls.h> 16 #include <linux/linkage.h> 17 #include <linux/pagemap.h> 18 #include <linux/quotaops.h> 19 #include <linux/backing-dev.h> 20 #include "internal.h" 21 22 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \ 23 SYNC_FILE_RANGE_WAIT_AFTER) 24 25 /* 26 * Write out and wait upon all dirty data associated with this 27 * superblock. Filesystem data as well as the underlying block 28 * device. Takes the superblock lock. 29 */ 30 int sync_filesystem(struct super_block *sb) 31 { 32 int ret = 0; 33 34 /* 35 * We need to be protected against the filesystem going from 36 * r/o to r/w or vice versa. 37 */ 38 WARN_ON(!rwsem_is_locked(&sb->s_umount)); 39 40 /* 41 * No point in syncing out anything if the filesystem is read-only. 42 */ 43 if (sb_rdonly(sb)) 44 return 0; 45 46 /* 47 * Do the filesystem syncing work. For simple filesystems 48 * writeback_inodes_sb(sb) just dirties buffers with inodes so we have 49 * to submit I/O for these buffers via sync_blockdev(). This also 50 * speeds up the wait == 1 case since in that case write_inode() 51 * methods call sync_dirty_buffer() and thus effectively write one block 52 * at a time. 53 */ 54 writeback_inodes_sb(sb, WB_REASON_SYNC); 55 if (sb->s_op->sync_fs) { 56 ret = sb->s_op->sync_fs(sb, 0); 57 if (ret) 58 return ret; 59 } 60 ret = sync_blockdev_nowait(sb->s_bdev); 61 if (ret) 62 return ret; 63 64 sync_inodes_sb(sb); 65 if (sb->s_op->sync_fs) { 66 ret = sb->s_op->sync_fs(sb, 1); 67 if (ret) 68 return ret; 69 } 70 return sync_blockdev(sb->s_bdev); 71 } 72 EXPORT_SYMBOL(sync_filesystem); 73 74 static void sync_inodes_one_sb(struct super_block *sb, void *arg) 75 { 76 if (!sb_rdonly(sb)) 77 sync_inodes_sb(sb); 78 } 79 80 static void sync_fs_one_sb(struct super_block *sb, void *arg) 81 { 82 if (!sb_rdonly(sb) && !(sb->s_iflags & SB_I_SKIP_SYNC) && 83 sb->s_op->sync_fs) 84 sb->s_op->sync_fs(sb, *(int *)arg); 85 } 86 87 /* 88 * Sync everything. We start by waking flusher threads so that most of 89 * writeback runs on all devices in parallel. Then we sync all inodes reliably 90 * which effectively also waits for all flusher threads to finish doing 91 * writeback. At this point all data is on disk so metadata should be stable 92 * and we tell filesystems to sync their metadata via ->sync_fs() calls. 93 * Finally, we writeout all block devices because some filesystems (e.g. ext2) 94 * just write metadata (such as inodes or bitmaps) to block device page cache 95 * and do not sync it on their own in ->sync_fs(). 96 */ 97 void ksys_sync(void) 98 { 99 int nowait = 0, wait = 1; 100 101 wakeup_flusher_threads(WB_REASON_SYNC); 102 iterate_supers(sync_inodes_one_sb, NULL); 103 iterate_supers(sync_fs_one_sb, &nowait); 104 iterate_supers(sync_fs_one_sb, &wait); 105 sync_bdevs(false); 106 sync_bdevs(true); 107 if (unlikely(laptop_mode)) 108 laptop_sync_completion(); 109 } 110 111 SYSCALL_DEFINE0(sync) 112 { 113 ksys_sync(); 114 return 0; 115 } 116 117 static void do_sync_work(struct work_struct *work) 118 { 119 int nowait = 0; 120 121 /* 122 * Sync twice to reduce the possibility we skipped some inodes / pages 123 * because they were temporarily locked 124 */ 125 iterate_supers(sync_inodes_one_sb, &nowait); 126 iterate_supers(sync_fs_one_sb, &nowait); 127 sync_bdevs(false); 128 iterate_supers(sync_inodes_one_sb, &nowait); 129 iterate_supers(sync_fs_one_sb, &nowait); 130 sync_bdevs(false); 131 printk("Emergency Sync complete\n"); 132 kfree(work); 133 } 134 135 void emergency_sync(void) 136 { 137 struct work_struct *work; 138 139 work = kmalloc(sizeof(*work), GFP_ATOMIC); 140 if (work) { 141 INIT_WORK(work, do_sync_work); 142 schedule_work(work); 143 } 144 } 145 146 /* 147 * sync a single super 148 */ 149 SYSCALL_DEFINE1(syncfs, int, fd) 150 { 151 CLASS(fd, f)(fd); 152 struct super_block *sb; 153 int ret, ret2; 154 155 if (fd_empty(f)) 156 return -EBADF; 157 sb = fd_file(f)->f_path.dentry->d_sb; 158 159 down_read(&sb->s_umount); 160 ret = sync_filesystem(sb); 161 up_read(&sb->s_umount); 162 163 ret2 = errseq_check_and_advance(&sb->s_wb_err, &fd_file(f)->f_sb_err); 164 165 return ret ? ret : ret2; 166 } 167 168 /** 169 * vfs_fsync_range - helper to sync a range of data & metadata to disk 170 * @file: file to sync 171 * @start: offset in bytes of the beginning of data range to sync 172 * @end: offset in bytes of the end of data range (inclusive) 173 * @datasync: perform only datasync 174 * 175 * Write back data in range @start..@end and metadata for @file to disk. If 176 * @datasync is set only metadata needed to access modified file data is 177 * written. 178 */ 179 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync) 180 { 181 struct inode *inode = file->f_mapping->host; 182 183 if (!file->f_op->fsync) 184 return -EINVAL; 185 if (!datasync && (inode->i_state & I_DIRTY_TIME)) 186 mark_inode_dirty_sync(inode); 187 return file->f_op->fsync(file, start, end, datasync); 188 } 189 EXPORT_SYMBOL(vfs_fsync_range); 190 191 /** 192 * vfs_fsync - perform a fsync or fdatasync on a file 193 * @file: file to sync 194 * @datasync: only perform a fdatasync operation 195 * 196 * Write back data and metadata for @file to disk. If @datasync is 197 * set only metadata needed to access modified file data is written. 198 */ 199 int vfs_fsync(struct file *file, int datasync) 200 { 201 return vfs_fsync_range(file, 0, LLONG_MAX, datasync); 202 } 203 EXPORT_SYMBOL(vfs_fsync); 204 205 static int do_fsync(unsigned int fd, int datasync) 206 { 207 CLASS(fd, f)(fd); 208 209 if (fd_empty(f)) 210 return -EBADF; 211 212 return vfs_fsync(fd_file(f), datasync); 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 int sync_file_range(struct file *file, loff_t offset, loff_t nbytes, 226 unsigned int flags) 227 { 228 int ret; 229 struct address_space *mapping; 230 loff_t endbyte; /* inclusive */ 231 umode_t i_mode; 232 233 ret = -EINVAL; 234 if (flags & ~VALID_FLAGS) 235 goto out; 236 237 endbyte = offset + nbytes; 238 239 if ((s64)offset < 0) 240 goto out; 241 if ((s64)endbyte < 0) 242 goto out; 243 if (endbyte < offset) 244 goto out; 245 246 if (sizeof(pgoff_t) == 4) { 247 if (offset >= (0x100000000ULL << PAGE_SHIFT)) { 248 /* 249 * The range starts outside a 32 bit machine's 250 * pagecache addressing capabilities. Let it "succeed" 251 */ 252 ret = 0; 253 goto out; 254 } 255 if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) { 256 /* 257 * Out to EOF 258 */ 259 nbytes = 0; 260 } 261 } 262 263 if (nbytes == 0) 264 endbyte = LLONG_MAX; 265 else 266 endbyte--; /* inclusive */ 267 268 i_mode = file_inode(file)->i_mode; 269 ret = -ESPIPE; 270 if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) && 271 !S_ISLNK(i_mode)) 272 goto out; 273 274 mapping = file->f_mapping; 275 ret = 0; 276 if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) { 277 ret = file_fdatawait_range(file, offset, endbyte); 278 if (ret < 0) 279 goto out; 280 } 281 282 if (flags & SYNC_FILE_RANGE_WRITE) { 283 int sync_mode = WB_SYNC_NONE; 284 285 if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) == 286 SYNC_FILE_RANGE_WRITE_AND_WAIT) 287 sync_mode = WB_SYNC_ALL; 288 289 ret = __filemap_fdatawrite_range(mapping, offset, endbyte, 290 sync_mode); 291 if (ret < 0) 292 goto out; 293 } 294 295 if (flags & SYNC_FILE_RANGE_WAIT_AFTER) 296 ret = file_fdatawait_range(file, offset, endbyte); 297 298 out: 299 return ret; 300 } 301 302 /* 303 * ksys_sync_file_range() permits finely controlled syncing over a segment of 304 * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is 305 * zero then ksys_sync_file_range() will operate from offset out to EOF. 306 * 307 * The flag bits are: 308 * 309 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range 310 * before performing the write. 311 * 312 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the 313 * range which are not presently under writeback. Note that this may block for 314 * significant periods due to exhaustion of disk request structures. 315 * 316 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range 317 * after performing the write. 318 * 319 * Useful combinations of the flag bits are: 320 * 321 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages 322 * in the range which were dirty on entry to ksys_sync_file_range() are placed 323 * under writeout. This is a start-write-for-data-integrity operation. 324 * 325 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which 326 * are not presently under writeout. This is an asynchronous flush-to-disk 327 * operation. Not suitable for data integrity operations. 328 * 329 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for 330 * completion of writeout of all pages in the range. This will be used after an 331 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait 332 * for that operation to complete and to return the result. 333 * 334 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER 335 * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT): 336 * a traditional sync() operation. This is a write-for-data-integrity operation 337 * which will ensure that all pages in the range which were dirty on entry to 338 * ksys_sync_file_range() are written to disk. It should be noted that disk 339 * caches are not flushed by this call, so there are no guarantees here that the 340 * data will be available on disk after a crash. 341 * 342 * 343 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any 344 * I/O errors or ENOSPC conditions and will return those to the caller, after 345 * clearing the EIO and ENOSPC flags in the address_space. 346 * 347 * It should be noted that none of these operations write out the file's 348 * metadata. So unless the application is strictly performing overwrites of 349 * already-instantiated disk blocks, there are no guarantees here that the data 350 * will be available after a crash. 351 */ 352 int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes, 353 unsigned int flags) 354 { 355 CLASS(fd, f)(fd); 356 357 if (fd_empty(f)) 358 return -EBADF; 359 360 return sync_file_range(fd_file(f), offset, nbytes, flags); 361 } 362 363 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes, 364 unsigned int, flags) 365 { 366 return ksys_sync_file_range(fd, offset, nbytes, flags); 367 } 368 369 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_SYNC_FILE_RANGE) 370 COMPAT_SYSCALL_DEFINE6(sync_file_range, int, fd, compat_arg_u64_dual(offset), 371 compat_arg_u64_dual(nbytes), unsigned int, flags) 372 { 373 return ksys_sync_file_range(fd, compat_arg_u64_glue(offset), 374 compat_arg_u64_glue(nbytes), flags); 375 } 376 #endif 377 378 /* It would be nice if people remember that not all the world's an i386 379 when they introduce new system calls */ 380 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags, 381 loff_t, offset, loff_t, nbytes) 382 { 383 return ksys_sync_file_range(fd, offset, nbytes, flags); 384 } 385