linux/fs/sync.c
<<
>>
Prefs
   1/*
   2 * High-level sync()-related operations
   3 */
   4
   5#include <linux/kernel.h>
   6#include <linux/file.h>
   7#include <linux/fs.h>
   8#include <linux/slab.h>
   9#include <linux/export.h>
  10#include <linux/namei.h>
  11#include <linux/sched.h>
  12#include <linux/writeback.h>
  13#include <linux/syscalls.h>
  14#include <linux/linkage.h>
  15#include <linux/pagemap.h>
  16#include <linux/quotaops.h>
  17#include <linux/backing-dev.h>
  18#include "internal.h"
  19
  20#define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
  21                        SYNC_FILE_RANGE_WAIT_AFTER)
  22
  23/*
  24 * Do the filesystem syncing work. For simple filesystems
  25 * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
  26 * submit IO for these buffers via __sync_blockdev(). This also speeds up the
  27 * wait == 1 case since in that case write_inode() functions do
  28 * sync_dirty_buffer() and thus effectively write one block at a time.
  29 */
  30static int __sync_filesystem(struct super_block *sb, int wait)
  31{
  32        if (wait)
  33                sync_inodes_sb(sb);
  34        else
  35                writeback_inodes_sb(sb, WB_REASON_SYNC);
  36
  37        if (sb->s_op->sync_fs)
  38                sb->s_op->sync_fs(sb, wait);
  39        return __sync_blockdev(sb->s_bdev, wait);
  40}
  41
  42/*
  43 * Write out and wait upon all dirty data associated with this
  44 * superblock.  Filesystem data as well as the underlying block
  45 * device.  Takes the superblock lock.
  46 */
  47int sync_filesystem(struct super_block *sb)
  48{
  49        int ret;
  50
  51        /*
  52         * We need to be protected against the filesystem going from
  53         * r/o to r/w or vice versa.
  54         */
  55        WARN_ON(!rwsem_is_locked(&sb->s_umount));
  56
  57        /*
  58         * No point in syncing out anything if the filesystem is read-only.
  59         */
  60        if (sb->s_flags & MS_RDONLY)
  61                return 0;
  62
  63        ret = __sync_filesystem(sb, 0);
  64        if (ret < 0)
  65                return ret;
  66        return __sync_filesystem(sb, 1);
  67}
  68EXPORT_SYMBOL(sync_filesystem);
  69
  70static void sync_inodes_one_sb(struct super_block *sb, void *arg)
  71{
  72        if (!(sb->s_flags & MS_RDONLY))
  73                sync_inodes_sb(sb);
  74}
  75
  76static void sync_fs_one_sb(struct super_block *sb, void *arg)
  77{
  78        if (!(sb->s_flags & MS_RDONLY) && sb->s_op->sync_fs)
  79                sb->s_op->sync_fs(sb, *(int *)arg);
  80}
  81
  82static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
  83{
  84        filemap_fdatawrite(bdev->bd_inode->i_mapping);
  85}
  86
  87static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
  88{
  89        filemap_fdatawait(bdev->bd_inode->i_mapping);
  90}
  91
  92/*
  93 * Sync everything. We start by waking flusher threads so that most of
  94 * writeback runs on all devices in parallel. Then we sync all inodes reliably
  95 * which effectively also waits for all flusher threads to finish doing
  96 * writeback. At this point all data is on disk so metadata should be stable
  97 * and we tell filesystems to sync their metadata via ->sync_fs() calls.
  98 * Finally, we writeout all block devices because some filesystems (e.g. ext2)
  99 * just write metadata (such as inodes or bitmaps) to block device page cache
 100 * and do not sync it on their own in ->sync_fs().
 101 */
 102SYSCALL_DEFINE0(sync)
 103{
 104        int nowait = 0, wait = 1;
 105
 106        wakeup_flusher_threads(0, WB_REASON_SYNC);
 107        iterate_supers(sync_inodes_one_sb, NULL);
 108        iterate_supers(sync_fs_one_sb, &nowait);
 109        iterate_supers(sync_fs_one_sb, &wait);
 110        iterate_bdevs(fdatawrite_one_bdev, NULL);
 111        iterate_bdevs(fdatawait_one_bdev, NULL);
 112        if (unlikely(laptop_mode))
 113                laptop_sync_completion();
 114        return 0;
 115}
 116
 117static 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        iterate_bdevs(fdatawrite_one_bdev, NULL);
 128        iterate_supers(sync_inodes_one_sb, &nowait);
 129        iterate_supers(sync_fs_one_sb, &nowait);
 130        iterate_bdevs(fdatawrite_one_bdev, NULL);
 131        printk("Emergency Sync complete\n");
 132        kfree(work);
 133}
 134
 135void 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 */
 149SYSCALL_DEFINE1(syncfs, int, fd)
 150{
 151        struct fd f = fdget(fd);
 152        struct super_block *sb;
 153        int ret;
 154
 155        if (!f.file)
 156                return -EBADF;
 157        sb = f.file->f_dentry->d_sb;
 158
 159        down_read(&sb->s_umount);
 160        ret = sync_filesystem(sb);
 161        up_read(&sb->s_umount);
 162
 163        fdput(f);
 164        return ret;
 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 */
 178int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
 179{
 180        if (!file->f_op->fsync)
 181                return -EINVAL;
 182        return file->f_op->fsync(file, start, end, datasync);
 183}
 184EXPORT_SYMBOL(vfs_fsync_range);
 185
 186/**
 187 * vfs_fsync - perform a fsync or fdatasync on a file
 188 * @file:               file to sync
 189 * @datasync:           only perform a fdatasync operation
 190 *
 191 * Write back data and metadata for @file to disk.  If @datasync is
 192 * set only metadata needed to access modified file data is written.
 193 */
 194int vfs_fsync(struct file *file, int datasync)
 195{
 196        return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
 197}
 198EXPORT_SYMBOL(vfs_fsync);
 199
 200static int do_fsync(unsigned int fd, int datasync)
 201{
 202        struct fd f = fdget(fd);
 203        int ret = -EBADF;
 204
 205        if (f.file) {
 206                ret = vfs_fsync(f.file, datasync);
 207                fdput(f);
 208        }
 209        return ret;
 210}
 211
 212SYSCALL_DEFINE1(fsync, unsigned int, fd)
 213{
 214        return do_fsync(fd, 0);
 215}
 216
 217SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 218{
 219        return do_fsync(fd, 1);
 220}
 221
 222/*
 223 * sys_sync_file_range() permits finely controlled syncing over a segment of
 224 * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 225 * zero then sys_sync_file_range() will operate from offset out to EOF.
 226 *
 227 * The flag bits are:
 228 *
 229 * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 230 * before performing the write.
 231 *
 232 * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 233 * range which are not presently under writeback. Note that this may block for
 234 * significant periods due to exhaustion of disk request structures.
 235 *
 236 * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 237 * after performing the write.
 238 *
 239 * Useful combinations of the flag bits are:
 240 *
 241 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 242 * in the range which were dirty on entry to sys_sync_file_range() are placed
 243 * under writeout.  This is a start-write-for-data-integrity operation.
 244 *
 245 * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 246 * are not presently under writeout.  This is an asynchronous flush-to-disk
 247 * operation.  Not suitable for data integrity operations.
 248 *
 249 * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 250 * completion of writeout of all pages in the range.  This will be used after an
 251 * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 252 * for that operation to complete and to return the result.
 253 *
 254 * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 255 * a traditional sync() operation.  This is a write-for-data-integrity operation
 256 * which will ensure that all pages in the range which were dirty on entry to
 257 * sys_sync_file_range() are committed to disk.
 258 *
 259 *
 260 * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 261 * I/O errors or ENOSPC conditions and will return those to the caller, after
 262 * clearing the EIO and ENOSPC flags in the address_space.
 263 *
 264 * It should be noted that none of these operations write out the file's
 265 * metadata.  So unless the application is strictly performing overwrites of
 266 * already-instantiated disk blocks, there are no guarantees here that the data
 267 * will be available after a crash.
 268 */
 269SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
 270                                unsigned int, flags)
 271{
 272        int ret;
 273        struct fd f;
 274        struct address_space *mapping;
 275        loff_t endbyte;                 /* inclusive */
 276        umode_t i_mode;
 277
 278        ret = -EINVAL;
 279        if (flags & ~VALID_FLAGS)
 280                goto out;
 281
 282        endbyte = offset + nbytes;
 283
 284        if ((s64)offset < 0)
 285                goto out;
 286        if ((s64)endbyte < 0)
 287                goto out;
 288        if (endbyte < offset)
 289                goto out;
 290
 291        if (sizeof(pgoff_t) == 4) {
 292                if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 293                        /*
 294                         * The range starts outside a 32 bit machine's
 295                         * pagecache addressing capabilities.  Let it "succeed"
 296                         */
 297                        ret = 0;
 298                        goto out;
 299                }
 300                if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 301                        /*
 302                         * Out to EOF
 303                         */
 304                        nbytes = 0;
 305                }
 306        }
 307
 308        if (nbytes == 0)
 309                endbyte = LLONG_MAX;
 310        else
 311                endbyte--;              /* inclusive */
 312
 313        ret = -EBADF;
 314        f = fdget(fd);
 315        if (!f.file)
 316                goto out;
 317
 318        i_mode = file_inode(f.file)->i_mode;
 319        ret = -ESPIPE;
 320        if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 321                        !S_ISLNK(i_mode))
 322                goto out_put;
 323
 324        mapping = f.file->f_mapping;
 325        if (!mapping) {
 326                ret = -EINVAL;
 327                goto out_put;
 328        }
 329
 330        ret = 0;
 331        if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 332                ret = filemap_fdatawait_range(mapping, offset, endbyte);
 333                if (ret < 0)
 334                        goto out_put;
 335        }
 336
 337        if (flags & SYNC_FILE_RANGE_WRITE) {
 338                ret = filemap_fdatawrite_range(mapping, offset, endbyte);
 339                if (ret < 0)
 340                        goto out_put;
 341        }
 342
 343        if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
 344                ret = filemap_fdatawait_range(mapping, offset, endbyte);
 345
 346out_put:
 347        fdput(f);
 348out:
 349        return ret;
 350}
 351
 352/* It would be nice if people remember that not all the world's an i386
 353   when they introduce new system calls */
 354SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
 355                                 loff_t, offset, loff_t, nbytes)
 356{
 357        return sys_sync_file_range(fd, offset, nbytes, flags);
 358}
 359