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