fs/sync.c

   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 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 EXPORT_SYMBOL_GPL(__sync_filesystem);
  43
  44 /*
  45  * Write out and wait upon all dirty data associated with this
  46  * superblock.  Filesystem data as well as the underlying block
  47  * device.  Takes the superblock lock.
  48  */
  49 int sync_filesystem(struct super_block *sb)
  50 {
  51         int ret;
  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_rdonly(sb))
  63                 return 0;
  64
  65         ret = __sync_filesystem(sb, 0);
  66         if (ret < 0)
  67                 return ret;
  68         return __sync_filesystem(sb, 1);
  69 }
  70 EXPORT_SYMBOL(sync_filesystem);
  71
  72 static void sync_inodes_one_sb(struct super_block *sb, void *arg)
  73 {
  74         if (!sb_rdonly(sb))
  75                 sync_inodes_sb(sb);
  76 }
  77
  78 static void sync_fs_one_sb(struct super_block *sb, void *arg)
  79 {
  80         if (!sb_rdonly(sb) && !(sb->s_iflags & SB_I_SKIP_SYNC) &&
  81             sb->s_op->sync_fs)
  82                 sb->s_op->sync_fs(sb, *(int *)arg);
  83 }
  84
  85 static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
  86 {
  87         filemap_fdatawrite(bdev->bd_inode->i_mapping);
  88 }
  89
  90 static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
  91 {
  92         /*
  93          * We keep the error status of individual mapping so that
  94          * applications can catch the writeback error using fsync(2).
  95          * See filemap_fdatawait_keep_errors() for details.
  96          */
  97         filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping);
  98 }
  99
 100 /*
 101  * Sync everything. We start by waking flusher threads so that most of
 102  * writeback runs on all devices in parallel. Then we sync all inodes reliably
 103  * which effectively also waits for all flusher threads to finish doing
 104  * writeback. At this point all data is on disk so metadata should be stable
 105  * and we tell filesystems to sync their metadata via ->sync_fs() calls.
 106  * Finally, we writeout all block devices because some filesystems (e.g. ext2)
 107  * just write metadata (such as inodes or bitmaps) to block device page cache
 108  * and do not sync it on their own in ->sync_fs().
 109  */
 110 void ksys_sync(void)
 111 {
 112         int nowait = 0, wait = 1;
 113
 114         wakeup_flusher_threads(WB_REASON_SYNC);
 115         iterate_supers(sync_inodes_one_sb, NULL);
 116         iterate_supers(sync_fs_one_sb, &nowait);
 117         iterate_supers(sync_fs_one_sb, &wait);
 118         iterate_bdevs(fdatawrite_one_bdev, NULL);
 119         iterate_bdevs(fdatawait_one_bdev, NULL);
 120         if (unlikely(laptop_mode))
 121                 laptop_sync_completion();
 122 }
 123
 124 SYSCALL_DEFINE0(sync)
 125 {
 126         ksys_sync();
 127         return 0;
 128 }
 129
 130 static void do_sync_work(struct work_struct *work)
 131 {
 132         int nowait = 0;
 133
 134         /*
 135          * Sync twice to reduce the possibility we skipped some inodes / pages
 136          * because they were temporarily locked
 137          */
 138         iterate_supers(sync_inodes_one_sb, &nowait);
 139         iterate_supers(sync_fs_one_sb, &nowait);
 140         iterate_bdevs(fdatawrite_one_bdev, NULL);
 141         iterate_supers(sync_inodes_one_sb, &nowait);
 142         iterate_supers(sync_fs_one_sb, &nowait);
 143         iterate_bdevs(fdatawrite_one_bdev, NULL);
 144         printk("Emergency Sync complete\n");
 145         kfree(work);
 146 }
 147
 148 void emergency_sync(void)
 149 {
 150         struct work_struct *work;
 151
 152         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 153         if (work) {
 154                 INIT_WORK(work, do_sync_work);
 155                 schedule_work(work);
 156         }
 157 }
 158
 159 /*
 160  * sync a single super
 161  */
 162 SYSCALL_DEFINE1(syncfs, int, fd)
 163 {
 164         struct fd f = fdget(fd);
 165         struct super_block *sb;
 166         int ret, ret2;
 167
 168         if (!f.file)
 169                 return -EBADF;
 170         sb = f.file->f_path.dentry->d_sb;
 171
 172         down_read(&sb->s_umount);
 173         ret = sync_filesystem(sb);
 174         up_read(&sb->s_umount);
 175
 176         ret2 = errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err);
 177
 178         fdput(f);
 179         return ret ? ret : ret2;
 180 }
 181
 182 /**
 183  * vfs_fsync_range - helper to sync a range of data & metadata to disk
 184  * @file:               file to sync
 185  * @start:              offset in bytes of the beginning of data range to sync
 186  * @end:                offset in bytes of the end of data range (inclusive)
 187  * @datasync:           perform only datasync
 188  *
 189  * Write back data in range @start..@end and metadata for @file to disk.  If
 190  * @datasync is set only metadata needed to access modified file data is
 191  * written.
 192  */
 193 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
 194 {
 195         struct inode *inode = file->f_mapping->host;
 196
 197         if (!file->f_op->fsync)
 198                 return -EINVAL;
 199         if (!datasync && (inode->i_state & I_DIRTY_TIME))
 200                 mark_inode_dirty_sync(inode);
 201         return file->f_op->fsync(file, start, end, datasync);
 202 }
 203 EXPORT_SYMBOL(vfs_fsync_range);
 204
 205 /**
 206  * vfs_fsync - perform a fsync or fdatasync on a file
 207  * @file:               file to sync
 208  * @datasync:           only perform a fdatasync operation
 209  *
 210  * Write back data and metadata for @file to disk.  If @datasync is
 211  * set only metadata needed to access modified file data is written.
 212  */
 213 int vfs_fsync(struct file *file, int datasync)
 214 {
 215         return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
 216 }
 217 EXPORT_SYMBOL(vfs_fsync);
 218
 219 static int do_fsync(unsigned int fd, int datasync)
 220 {
 221         struct fd f = fdget(fd);
 222         int ret = -EBADF;
 223
 224         if (f.file) {
 225                 ret = vfs_fsync(f.file, datasync);
 226                 fdput(f);
 227         }
 228         return ret;
 229 }
 230
 231 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 232 {
 233         return do_fsync(fd, 0);
 234 }
 235
 236 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 237 {
 238         return do_fsync(fd, 1);
 239 }
 240
 241 int sync_file_range(struct file *file, loff_t offset, loff_t nbytes,
 242                     unsigned int flags)
 243 {
 244         int ret;
 245         struct address_space *mapping;
 246         loff_t endbyte;                 /* inclusive */
 247         umode_t i_mode;
 248
 249         ret = -EINVAL;
 250         if (flags & ~VALID_FLAGS)
 251                 goto out;
 252
 253         endbyte = offset + nbytes;
 254
 255         if ((s64)offset < 0)
 256                 goto out;
 257         if ((s64)endbyte < 0)
 258                 goto out;
 259         if (endbyte < offset)
 260                 goto out;
 261
 262         if (sizeof(pgoff_t) == 4) {
 263                 if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
 264                         /*
 265                          * The range starts outside a 32 bit machine's
 266                          * pagecache addressing capabilities.  Let it "succeed"
 267                          */
 268                         ret = 0;
 269                         goto out;
 270                 }
 271                 if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
 272                         /*
 273                          * Out to EOF
 274                          */
 275                         nbytes = 0;
 276                 }
 277         }
 278
 279         if (nbytes == 0)
 280                 endbyte = LLONG_MAX;
 281         else
 282                 endbyte--;              /* inclusive */
 283
 284         i_mode = file_inode(file)->i_mode;
 285         ret = -ESPIPE;
 286         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 287                         !S_ISLNK(i_mode))
 288                 goto out;
 289
 290         mapping = file->f_mapping;
 291         ret = 0;
 292         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 293                 ret = file_fdatawait_range(file, offset, endbyte);
 294                 if (ret < 0)
 295                         goto out;
 296         }
 297
 298         if (flags & SYNC_FILE_RANGE_WRITE) {
 299                 int sync_mode = WB_SYNC_NONE;
 300
 301                 if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
 302                              SYNC_FILE_RANGE_WRITE_AND_WAIT)
 303                         sync_mode = WB_SYNC_ALL;
 304
 305                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 306                                                  sync_mode);
 307                 if (ret < 0)
 308                         goto out;
 309         }
 310
 311         if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
 312                 ret = file_fdatawait_range(file, offset, endbyte);
 313
 314 out:
 315         return ret;
 316 }
 317
 318 /*
 319  * ksys_sync_file_range() permits finely controlled syncing over a segment of
 320  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 321  * zero then ksys_sync_file_range() will operate from offset out to EOF.
 322  *
 323  * The flag bits are:
 324  *
 325  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 326  * before performing the write.
 327  *
 328  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 329  * range which are not presently under writeback. Note that this may block for
 330  * significant periods due to exhaustion of disk request structures.
 331  *
 332  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 333  * after performing the write.
 334  *
 335  * Useful combinations of the flag bits are:
 336  *
 337  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 338  * in the range which were dirty on entry to ksys_sync_file_range() are placed
 339  * under writeout.  This is a start-write-for-data-integrity operation.
 340  *
 341  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 342  * are not presently under writeout.  This is an asynchronous flush-to-disk
 343  * operation.  Not suitable for data integrity operations.
 344  *
 345  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 346  * completion of writeout of all pages in the range.  This will be used after an
 347  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 348  * for that operation to complete and to return the result.
 349  *
 350  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
 351  * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
 352  * a traditional sync() operation.  This is a write-for-data-integrity operation
 353  * which will ensure that all pages in the range which were dirty on entry to
 354  * ksys_sync_file_range() are written to disk.  It should be noted that disk
 355  * caches are not flushed by this call, so there are no guarantees here that the
 356  * data will be available on disk after a crash.
 357  *
 358  *
 359  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 360  * I/O errors or ENOSPC conditions and will return those to the caller, after
 361  * clearing the EIO and ENOSPC flags in the address_space.
 362  *
 363  * It should be noted that none of these operations write out the file's
 364  * metadata.  So unless the application is strictly performing overwrites of
 365  * already-instantiated disk blocks, there are no guarantees here that the data
 366  * will be available after a crash.
 367  */
 368 int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
 369                          unsigned int flags)
 370 {
 371         int ret;
 372         struct fd f;
 373
 374         ret = -EBADF;
 375         f = fdget(fd);
 376         if (f.file)
 377                 ret = sync_file_range(f.file, offset, nbytes, flags);
 378
 379         fdput(f);
 380         return ret;
 381 }
 382
 383 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
 384                                 unsigned int, flags)
 385 {
 386         return ksys_sync_file_range(fd, offset, nbytes, flags);
 387 }
 388
 389 /* It would be nice if people remember that not all the world's an i386
 390    when they introduce new system calls */
 391 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
 392                                  loff_t, offset, loff_t, nbytes)
 393 {
 394         return ksys_sync_file_range(fd, offset, nbytes, flags);
 395 }