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_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;
 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         fdput(f);
 176         return ret;
 177 }
 178
 179 /**
 180  * vfs_fsync_range - helper to sync a range of data & metadata to disk
 181  * @file:               file to sync
 182  * @start:              offset in bytes of the beginning of data range to sync
 183  * @end:                offset in bytes of the end of data range (inclusive)
 184  * @datasync:           perform only datasync
 185  *
 186  * Write back data in range @start..@end and metadata for @file to disk.  If
 187  * @datasync is set only metadata needed to access modified file data is
 188  * written.
 189  */
 190 int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
 191 {
 192         struct inode *inode = file->f_mapping->host;
 193
 194         if (!file->f_op->fsync)
 195                 return -EINVAL;
 196         if (!datasync && (inode->i_state & I_DIRTY_TIME))
 197                 mark_inode_dirty_sync(inode);
 198         return file->f_op->fsync(file, start, end, datasync);
 199 }
 200 EXPORT_SYMBOL(vfs_fsync_range);
 201
 202 /**
 203  * vfs_fsync - perform a fsync or fdatasync on a file
 204  * @file:               file to sync
 205  * @datasync:           only perform a fdatasync operation
 206  *
 207  * Write back data and metadata for @file to disk.  If @datasync is
 208  * set only metadata needed to access modified file data is written.
 209  */
 210 int vfs_fsync(struct file *file, int datasync)
 211 {
 212         return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
 213 }
 214 EXPORT_SYMBOL(vfs_fsync);
 215
 216 static int do_fsync(unsigned int fd, int datasync)
 217 {
 218         struct fd f = fdget(fd);
 219         int ret = -EBADF;
 220
 221         if (f.file) {
 222                 ret = vfs_fsync(f.file, datasync);
 223                 fdput(f);
 224         }
 225         return ret;
 226 }
 227
 228 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 229 {
 230         return do_fsync(fd, 0);
 231 }
 232
 233 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 234 {
 235         return do_fsync(fd, 1);
 236 }
 237
 238 int sync_file_range(struct file *file, loff_t offset, loff_t nbytes,
 239                     unsigned int flags)
 240 {
 241         int ret;
 242         struct address_space *mapping;
 243         loff_t endbyte;                 /* inclusive */
 244         umode_t i_mode;
 245
 246         ret = -EINVAL;
 247         if (flags & ~VALID_FLAGS)
 248                 goto out;
 249
 250         endbyte = offset + nbytes;
 251
 252         if ((s64)offset < 0)
 253                 goto out;
 254         if ((s64)endbyte < 0)
 255                 goto out;
 256         if (endbyte < offset)
 257                 goto out;
 258
 259         if (sizeof(pgoff_t) == 4) {
 260                 if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
 261                         /*
 262                          * The range starts outside a 32 bit machine's
 263                          * pagecache addressing capabilities.  Let it "succeed"
 264                          */
 265                         ret = 0;
 266                         goto out;
 267                 }
 268                 if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
 269                         /*
 270                          * Out to EOF
 271                          */
 272                         nbytes = 0;
 273                 }
 274         }
 275
 276         if (nbytes == 0)
 277                 endbyte = LLONG_MAX;
 278         else
 279                 endbyte--;              /* inclusive */
 280
 281         i_mode = file_inode(file)->i_mode;
 282         ret = -ESPIPE;
 283         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 284                         !S_ISLNK(i_mode))
 285                 goto out;
 286
 287         mapping = file->f_mapping;
 288         ret = 0;
 289         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 290                 ret = file_fdatawait_range(file, offset, endbyte);
 291                 if (ret < 0)
 292                         goto out;
 293         }
 294
 295         if (flags & SYNC_FILE_RANGE_WRITE) {
 296                 int sync_mode = WB_SYNC_NONE;
 297
 298                 if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
 299                              SYNC_FILE_RANGE_WRITE_AND_WAIT)
 300                         sync_mode = WB_SYNC_ALL;
 301
 302                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 303                                                  sync_mode);
 304                 if (ret < 0)
 305                         goto out;
 306         }
 307
 308         if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
 309                 ret = file_fdatawait_range(file, offset, endbyte);
 310
 311 out:
 312         return ret;
 313 }
 314
 315 /*
 316  * ksys_sync_file_range() permits finely controlled syncing over a segment of
 317  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 318  * zero then ksys_sync_file_range() will operate from offset out to EOF.
 319  *
 320  * The flag bits are:
 321  *
 322  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 323  * before performing the write.
 324  *
 325  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 326  * range which are not presently under writeback. Note that this may block for
 327  * significant periods due to exhaustion of disk request structures.
 328  *
 329  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 330  * after performing the write.
 331  *
 332  * Useful combinations of the flag bits are:
 333  *
 334  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 335  * in the range which were dirty on entry to ksys_sync_file_range() are placed
 336  * under writeout.  This is a start-write-for-data-integrity operation.
 337  *
 338  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 339  * are not presently under writeout.  This is an asynchronous flush-to-disk
 340  * operation.  Not suitable for data integrity operations.
 341  *
 342  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 343  * completion of writeout of all pages in the range.  This will be used after an
 344  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 345  * for that operation to complete and to return the result.
 346  *
 347  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
 348  * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
 349  * a traditional sync() operation.  This is a write-for-data-integrity operation
 350  * which will ensure that all pages in the range which were dirty on entry to
 351  * ksys_sync_file_range() are written to disk.  It should be noted that disk
 352  * caches are not flushed by this call, so there are no guarantees here that the
 353  * data will be available on disk after a crash.
 354  *
 355  *
 356  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 357  * I/O errors or ENOSPC conditions and will return those to the caller, after
 358  * clearing the EIO and ENOSPC flags in the address_space.
 359  *
 360  * It should be noted that none of these operations write out the file's
 361  * metadata.  So unless the application is strictly performing overwrites of
 362  * already-instantiated disk blocks, there are no guarantees here that the data
 363  * will be available after a crash.
 364  */
 365 int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
 366                          unsigned int flags)
 367 {
 368         int ret;
 369         struct fd f;
 370
 371         ret = -EBADF;
 372         f = fdget(fd);
 373         if (f.file)
 374                 ret = sync_file_range(f.file, offset, nbytes, flags);
 375
 376         fdput(f);
 377         return ret;
 378 }
 379
 380 SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
 381                                 unsigned int, flags)
 382 {
 383         return ksys_sync_file_range(fd, offset, nbytes, flags);
 384 }
 385
 386 /* It would be nice if people remember that not all the world's an i386
 387    when they introduce new system calls */
 388 SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
 389                                  loff_t, offset, loff_t, nbytes)
 390 {
 391         return ksys_sync_file_range(fd, offset, nbytes, flags);
 392 }