Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 1 | /* |
| 2 | * High-level sync()-related operations |
| 3 | */ |
| 4 | |
| 5 | #include <linux/kernel.h> |
| 6 | #include <linux/file.h> |
| 7 | #include <linux/fs.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 8 | #include <linux/slab.h> |
Paul Gortmaker | 630d9c4 | 2011-11-16 23:57:37 -0500 | [diff] [blame] | 9 | #include <linux/export.h> |
Sage Weil | b7ed78f | 2011-03-10 11:31:30 -0800 | [diff] [blame] | 10 | #include <linux/namei.h> |
Al Viro | 914e263 | 2006-10-18 13:55:46 -0400 | [diff] [blame] | 11 | #include <linux/sched.h> |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 12 | #include <linux/writeback.h> |
| 13 | #include <linux/syscalls.h> |
| 14 | #include <linux/linkage.h> |
| 15 | #include <linux/pagemap.h> |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 16 | #include <linux/quotaops.h> |
Jörn Engel | 5129a46 | 2010-04-25 08:54:42 +0200 | [diff] [blame] | 17 | #include <linux/backing-dev.h> |
Jan Kara | 5a3e5cb | 2009-04-27 16:43:48 +0200 | [diff] [blame] | 18 | #include "internal.h" |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 19 | |
| 20 | #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \ |
| 21 | SYNC_FILE_RANGE_WAIT_AFTER) |
| 22 | |
Jan Kara | c15c54f | 2009-04-27 16:43:52 +0200 | [diff] [blame] | 23 | /* |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 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. |
Jan Kara | c15c54f | 2009-04-27 16:43:52 +0200 | [diff] [blame] | 29 | */ |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 30 | static int __sync_filesystem(struct super_block *sb, int wait) |
Jan Kara | c15c54f | 2009-04-27 16:43:52 +0200 | [diff] [blame] | 31 | { |
Christoph Hellwig | 5fb324a | 2010-02-16 03:44:52 -0500 | [diff] [blame] | 32 | if (wait) |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 33 | sync_inodes_sb(sb); |
Christoph Hellwig | 5fb324a | 2010-02-16 03:44:52 -0500 | [diff] [blame] | 34 | else |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 35 | writeback_inodes_sb(sb, WB_REASON_SYNC); |
Christoph Hellwig | 5fb324a | 2010-02-16 03:44:52 -0500 | [diff] [blame] | 36 | |
Jan Kara | c15c54f | 2009-04-27 16:43:52 +0200 | [diff] [blame] | 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 | */ |
Jan Kara | 60b0680 | 2009-04-27 16:43:53 +0200 | [diff] [blame] | 47 | int sync_filesystem(struct super_block *sb) |
Jan Kara | c15c54f | 2009-04-27 16:43:52 +0200 | [diff] [blame] | 48 | { |
| 49 | int ret; |
| 50 | |
Christoph Hellwig | 5af7926 | 2009-05-05 15:41:25 +0200 | [diff] [blame] | 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 | |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 63 | ret = __sync_filesystem(sb, 0); |
Jan Kara | c15c54f | 2009-04-27 16:43:52 +0200 | [diff] [blame] | 64 | if (ret < 0) |
| 65 | return ret; |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 66 | return __sync_filesystem(sb, 1); |
Jan Kara | c15c54f | 2009-04-27 16:43:52 +0200 | [diff] [blame] | 67 | } |
Anton Altaparmakov | 10096fb | 2014-08-21 11:09:27 +0100 | [diff] [blame] | 68 | EXPORT_SYMBOL(sync_filesystem); |
Jan Kara | c15c54f | 2009-04-27 16:43:52 +0200 | [diff] [blame] | 69 | |
Jan Kara | b3de653 | 2012-07-03 16:45:30 +0200 | [diff] [blame] | 70 | static void sync_inodes_one_sb(struct super_block *sb, void *arg) |
Al Viro | 01a05b3 | 2010-03-23 06:06:58 -0400 | [diff] [blame] | 71 | { |
Jens Axboe | 95f2860 | 2011-03-17 11:13:12 +0100 | [diff] [blame] | 72 | if (!(sb->s_flags & MS_RDONLY)) |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 73 | sync_inodes_sb(sb); |
Al Viro | 01a05b3 | 2010-03-23 06:06:58 -0400 | [diff] [blame] | 74 | } |
Jan Kara | b3de653 | 2012-07-03 16:45:30 +0200 | [diff] [blame] | 75 | |
Jan Kara | b3de653 | 2012-07-03 16:45:30 +0200 | [diff] [blame] | 76 | static 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 | |
Jan Kara | d0e91b1 | 2012-07-03 16:45:33 +0200 | [diff] [blame] | 82 | static void fdatawrite_one_bdev(struct block_device *bdev, void *arg) |
Jan Kara | b3de653 | 2012-07-03 16:45:30 +0200 | [diff] [blame] | 83 | { |
Jan Kara | d0e91b1 | 2012-07-03 16:45:33 +0200 | [diff] [blame] | 84 | filemap_fdatawrite(bdev->bd_inode->i_mapping); |
Jan Kara | a8c7176 | 2012-07-03 16:45:32 +0200 | [diff] [blame] | 85 | } |
| 86 | |
Jan Kara | d0e91b1 | 2012-07-03 16:45:33 +0200 | [diff] [blame] | 87 | static void fdatawait_one_bdev(struct block_device *bdev, void *arg) |
Jan Kara | a8c7176 | 2012-07-03 16:45:32 +0200 | [diff] [blame] | 88 | { |
Junichi Nomura | aa750fd | 2015-11-05 18:47:23 -0800 | [diff] [blame] | 89 | /* |
| 90 | * We keep the error status of individual mapping so that |
| 91 | * applications can catch the writeback error using fsync(2). |
| 92 | * See filemap_fdatawait_keep_errors() for details. |
| 93 | */ |
| 94 | filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping); |
Jan Kara | c15c54f | 2009-04-27 16:43:52 +0200 | [diff] [blame] | 95 | } |
| 96 | |
Zhang, Yanmin | 3beab0b | 2009-07-05 12:08:08 -0700 | [diff] [blame] | 97 | /* |
Jan Kara | 4ea425b | 2012-07-03 16:45:34 +0200 | [diff] [blame] | 98 | * Sync everything. We start by waking flusher threads so that most of |
| 99 | * writeback runs on all devices in parallel. Then we sync all inodes reliably |
| 100 | * which effectively also waits for all flusher threads to finish doing |
| 101 | * writeback. At this point all data is on disk so metadata should be stable |
| 102 | * and we tell filesystems to sync their metadata via ->sync_fs() calls. |
| 103 | * Finally, we writeout all block devices because some filesystems (e.g. ext2) |
| 104 | * just write metadata (such as inodes or bitmaps) to block device page cache |
| 105 | * and do not sync it on their own in ->sync_fs(). |
Zhang, Yanmin | 3beab0b | 2009-07-05 12:08:08 -0700 | [diff] [blame] | 106 | */ |
Heiko Carstens | a5f8fa9 | 2009-01-14 14:14:11 +0100 | [diff] [blame] | 107 | SYSCALL_DEFINE0(sync) |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 108 | { |
Jan Kara | b3de653 | 2012-07-03 16:45:30 +0200 | [diff] [blame] | 109 | int nowait = 0, wait = 1; |
| 110 | |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 111 | wakeup_flusher_threads(0, WB_REASON_SYNC); |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 112 | iterate_supers(sync_inodes_one_sb, NULL); |
Jan Kara | 4ea425b | 2012-07-03 16:45:34 +0200 | [diff] [blame] | 113 | iterate_supers(sync_fs_one_sb, &nowait); |
Jan Kara | b3de653 | 2012-07-03 16:45:30 +0200 | [diff] [blame] | 114 | iterate_supers(sync_fs_one_sb, &wait); |
Jan Kara | d0e91b1 | 2012-07-03 16:45:33 +0200 | [diff] [blame] | 115 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
| 116 | iterate_bdevs(fdatawait_one_bdev, NULL); |
Jan Kara | 5cee581 | 2009-04-27 16:43:51 +0200 | [diff] [blame] | 117 | if (unlikely(laptop_mode)) |
| 118 | laptop_sync_completion(); |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 119 | return 0; |
| 120 | } |
| 121 | |
Jens Axboe | a2a9537 | 2009-03-17 09:38:40 +0100 | [diff] [blame] | 122 | static void do_sync_work(struct work_struct *work) |
| 123 | { |
Jan Kara | b3de653 | 2012-07-03 16:45:30 +0200 | [diff] [blame] | 124 | int nowait = 0; |
| 125 | |
Jan Kara | 5cee581 | 2009-04-27 16:43:51 +0200 | [diff] [blame] | 126 | /* |
| 127 | * Sync twice to reduce the possibility we skipped some inodes / pages |
| 128 | * because they were temporarily locked |
| 129 | */ |
Jan Kara | b3de653 | 2012-07-03 16:45:30 +0200 | [diff] [blame] | 130 | iterate_supers(sync_inodes_one_sb, &nowait); |
| 131 | iterate_supers(sync_fs_one_sb, &nowait); |
Jan Kara | d0e91b1 | 2012-07-03 16:45:33 +0200 | [diff] [blame] | 132 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
Jan Kara | b3de653 | 2012-07-03 16:45:30 +0200 | [diff] [blame] | 133 | iterate_supers(sync_inodes_one_sb, &nowait); |
| 134 | iterate_supers(sync_fs_one_sb, &nowait); |
Jan Kara | d0e91b1 | 2012-07-03 16:45:33 +0200 | [diff] [blame] | 135 | iterate_bdevs(fdatawrite_one_bdev, NULL); |
Jan Kara | 5cee581 | 2009-04-27 16:43:51 +0200 | [diff] [blame] | 136 | printk("Emergency Sync complete\n"); |
Jens Axboe | a2a9537 | 2009-03-17 09:38:40 +0100 | [diff] [blame] | 137 | kfree(work); |
| 138 | } |
| 139 | |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 140 | void emergency_sync(void) |
| 141 | { |
Jens Axboe | a2a9537 | 2009-03-17 09:38:40 +0100 | [diff] [blame] | 142 | struct work_struct *work; |
| 143 | |
| 144 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
| 145 | if (work) { |
| 146 | INIT_WORK(work, do_sync_work); |
| 147 | schedule_work(work); |
| 148 | } |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 149 | } |
| 150 | |
Sage Weil | b7ed78f | 2011-03-10 11:31:30 -0800 | [diff] [blame] | 151 | /* |
| 152 | * sync a single super |
| 153 | */ |
| 154 | SYSCALL_DEFINE1(syncfs, int, fd) |
| 155 | { |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 156 | struct fd f = fdget(fd); |
Sage Weil | b7ed78f | 2011-03-10 11:31:30 -0800 | [diff] [blame] | 157 | struct super_block *sb; |
| 158 | int ret; |
Sage Weil | b7ed78f | 2011-03-10 11:31:30 -0800 | [diff] [blame] | 159 | |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 160 | if (!f.file) |
Sage Weil | b7ed78f | 2011-03-10 11:31:30 -0800 | [diff] [blame] | 161 | return -EBADF; |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 162 | sb = f.file->f_path.dentry->d_sb; |
Sage Weil | b7ed78f | 2011-03-10 11:31:30 -0800 | [diff] [blame] | 163 | |
| 164 | down_read(&sb->s_umount); |
| 165 | ret = sync_filesystem(sb); |
| 166 | up_read(&sb->s_umount); |
| 167 | |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 168 | fdput(f); |
Sage Weil | b7ed78f | 2011-03-10 11:31:30 -0800 | [diff] [blame] | 169 | return ret; |
| 170 | } |
| 171 | |
Christoph Hellwig | 4c728ef | 2008-12-22 21:11:15 +0100 | [diff] [blame] | 172 | /** |
Jan Kara | 148f948 | 2009-08-17 19:52:36 +0200 | [diff] [blame] | 173 | * vfs_fsync_range - helper to sync a range of data & metadata to disk |
Christoph Hellwig | 4c728ef | 2008-12-22 21:11:15 +0100 | [diff] [blame] | 174 | * @file: file to sync |
Jan Kara | 148f948 | 2009-08-17 19:52:36 +0200 | [diff] [blame] | 175 | * @start: offset in bytes of the beginning of data range to sync |
| 176 | * @end: offset in bytes of the end of data range (inclusive) |
| 177 | * @datasync: perform only datasync |
Christoph Hellwig | 4c728ef | 2008-12-22 21:11:15 +0100 | [diff] [blame] | 178 | * |
Jan Kara | 148f948 | 2009-08-17 19:52:36 +0200 | [diff] [blame] | 179 | * Write back data in range @start..@end and metadata for @file to disk. If |
| 180 | * @datasync is set only metadata needed to access modified file data is |
| 181 | * written. |
Christoph Hellwig | 4c728ef | 2008-12-22 21:11:15 +0100 | [diff] [blame] | 182 | */ |
Christoph Hellwig | 8018ab0 | 2010-03-22 17:32:25 +0100 | [diff] [blame] | 183 | int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync) |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 184 | { |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 185 | struct inode *inode = file->f_mapping->host; |
| 186 | |
Al Viro | 72c2d53 | 2013-09-22 16:27:52 -0400 | [diff] [blame] | 187 | if (!file->f_op->fsync) |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 188 | return -EINVAL; |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 189 | if (!datasync && (inode->i_state & I_DIRTY_TIME)) { |
| 190 | spin_lock(&inode->i_lock); |
| 191 | inode->i_state &= ~I_DIRTY_TIME; |
| 192 | spin_unlock(&inode->i_lock); |
| 193 | mark_inode_dirty_sync(inode); |
| 194 | } |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 195 | return file->f_op->fsync(file, start, end, datasync); |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 196 | } |
Jan Kara | 148f948 | 2009-08-17 19:52:36 +0200 | [diff] [blame] | 197 | EXPORT_SYMBOL(vfs_fsync_range); |
| 198 | |
| 199 | /** |
| 200 | * vfs_fsync - perform a fsync or fdatasync on a file |
| 201 | * @file: file to sync |
Jan Kara | 148f948 | 2009-08-17 19:52:36 +0200 | [diff] [blame] | 202 | * @datasync: only perform a fdatasync operation |
| 203 | * |
| 204 | * Write back data and metadata for @file to disk. If @datasync is |
| 205 | * set only metadata needed to access modified file data is written. |
Jan Kara | 148f948 | 2009-08-17 19:52:36 +0200 | [diff] [blame] | 206 | */ |
Christoph Hellwig | 8018ab0 | 2010-03-22 17:32:25 +0100 | [diff] [blame] | 207 | int vfs_fsync(struct file *file, int datasync) |
Jan Kara | 148f948 | 2009-08-17 19:52:36 +0200 | [diff] [blame] | 208 | { |
Christoph Hellwig | 8018ab0 | 2010-03-22 17:32:25 +0100 | [diff] [blame] | 209 | return vfs_fsync_range(file, 0, LLONG_MAX, datasync); |
Jan Kara | 148f948 | 2009-08-17 19:52:36 +0200 | [diff] [blame] | 210 | } |
Christoph Hellwig | 4c728ef | 2008-12-22 21:11:15 +0100 | [diff] [blame] | 211 | EXPORT_SYMBOL(vfs_fsync); |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 212 | |
Christoph Hellwig | 4c728ef | 2008-12-22 21:11:15 +0100 | [diff] [blame] | 213 | static int do_fsync(unsigned int fd, int datasync) |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 214 | { |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 215 | struct fd f = fdget(fd); |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 216 | int ret = -EBADF; |
| 217 | |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 218 | if (f.file) { |
| 219 | ret = vfs_fsync(f.file, datasync); |
| 220 | fdput(f); |
Jin Qian | e61abd8 | 2017-03-02 13:32:59 -0800 | [diff] [blame] | 221 | inc_syscfs(current); |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 222 | } |
| 223 | return ret; |
| 224 | } |
| 225 | |
Heiko Carstens | a5f8fa9 | 2009-01-14 14:14:11 +0100 | [diff] [blame] | 226 | SYSCALL_DEFINE1(fsync, unsigned int, fd) |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 227 | { |
Christoph Hellwig | 4c728ef | 2008-12-22 21:11:15 +0100 | [diff] [blame] | 228 | return do_fsync(fd, 0); |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 229 | } |
| 230 | |
Heiko Carstens | a5f8fa9 | 2009-01-14 14:14:11 +0100 | [diff] [blame] | 231 | SYSCALL_DEFINE1(fdatasync, unsigned int, fd) |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 232 | { |
Christoph Hellwig | 4c728ef | 2008-12-22 21:11:15 +0100 | [diff] [blame] | 233 | return do_fsync(fd, 1); |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 234 | } |
| 235 | |
| 236 | /* |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 237 | * sys_sync_file_range() permits finely controlled syncing over a segment of |
| 238 | * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is |
| 239 | * zero then sys_sync_file_range() will operate from offset out to EOF. |
| 240 | * |
| 241 | * The flag bits are: |
| 242 | * |
| 243 | * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range |
| 244 | * before performing the write. |
| 245 | * |
| 246 | * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the |
Pavel Machek | cce7708 | 2008-07-23 21:27:36 -0700 | [diff] [blame] | 247 | * range which are not presently under writeback. Note that this may block for |
| 248 | * significant periods due to exhaustion of disk request structures. |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 249 | * |
| 250 | * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range |
| 251 | * after performing the write. |
| 252 | * |
| 253 | * Useful combinations of the flag bits are: |
| 254 | * |
| 255 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages |
| 256 | * in the range which were dirty on entry to sys_sync_file_range() are placed |
| 257 | * under writeout. This is a start-write-for-data-integrity operation. |
| 258 | * |
| 259 | * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which |
| 260 | * are not presently under writeout. This is an asynchronous flush-to-disk |
| 261 | * operation. Not suitable for data integrity operations. |
| 262 | * |
| 263 | * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for |
| 264 | * completion of writeout of all pages in the range. This will be used after an |
| 265 | * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait |
| 266 | * for that operation to complete and to return the result. |
| 267 | * |
| 268 | * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER: |
| 269 | * a traditional sync() operation. This is a write-for-data-integrity operation |
| 270 | * which will ensure that all pages in the range which were dirty on entry to |
| 271 | * sys_sync_file_range() are committed to disk. |
| 272 | * |
| 273 | * |
| 274 | * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any |
| 275 | * I/O errors or ENOSPC conditions and will return those to the caller, after |
| 276 | * clearing the EIO and ENOSPC flags in the address_space. |
| 277 | * |
| 278 | * It should be noted that none of these operations write out the file's |
| 279 | * metadata. So unless the application is strictly performing overwrites of |
| 280 | * already-instantiated disk blocks, there are no guarantees here that the data |
| 281 | * will be available after a crash. |
| 282 | */ |
Al Viro | 4a0fd5b | 2013-01-21 15:16:58 -0500 | [diff] [blame] | 283 | SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes, |
| 284 | unsigned int, flags) |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 285 | { |
| 286 | int ret; |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 287 | struct fd f; |
Christoph Hellwig | 7a0ad10 | 2009-12-17 14:24:40 +0100 | [diff] [blame] | 288 | struct address_space *mapping; |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 289 | loff_t endbyte; /* inclusive */ |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 290 | umode_t i_mode; |
| 291 | |
| 292 | ret = -EINVAL; |
| 293 | if (flags & ~VALID_FLAGS) |
| 294 | goto out; |
| 295 | |
| 296 | endbyte = offset + nbytes; |
| 297 | |
| 298 | if ((s64)offset < 0) |
| 299 | goto out; |
| 300 | if ((s64)endbyte < 0) |
| 301 | goto out; |
| 302 | if (endbyte < offset) |
| 303 | goto out; |
| 304 | |
| 305 | if (sizeof(pgoff_t) == 4) { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 306 | if (offset >= (0x100000000ULL << PAGE_SHIFT)) { |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 307 | /* |
| 308 | * The range starts outside a 32 bit machine's |
| 309 | * pagecache addressing capabilities. Let it "succeed" |
| 310 | */ |
| 311 | ret = 0; |
| 312 | goto out; |
| 313 | } |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 314 | if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) { |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 315 | /* |
| 316 | * Out to EOF |
| 317 | */ |
| 318 | nbytes = 0; |
| 319 | } |
| 320 | } |
| 321 | |
| 322 | if (nbytes == 0) |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 323 | endbyte = LLONG_MAX; |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 324 | else |
| 325 | endbyte--; /* inclusive */ |
| 326 | |
| 327 | ret = -EBADF; |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 328 | f = fdget(fd); |
| 329 | if (!f.file) |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 330 | goto out; |
| 331 | |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 332 | i_mode = file_inode(f.file)->i_mode; |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 333 | ret = -ESPIPE; |
| 334 | if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) && |
| 335 | !S_ISLNK(i_mode)) |
| 336 | goto out_put; |
| 337 | |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 338 | mapping = f.file->f_mapping; |
Christoph Hellwig | 7a0ad10 | 2009-12-17 14:24:40 +0100 | [diff] [blame] | 339 | if (!mapping) { |
| 340 | ret = -EINVAL; |
| 341 | goto out_put; |
| 342 | } |
| 343 | |
| 344 | ret = 0; |
| 345 | if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) { |
| 346 | ret = filemap_fdatawait_range(mapping, offset, endbyte); |
| 347 | if (ret < 0) |
| 348 | goto out_put; |
| 349 | } |
| 350 | |
| 351 | if (flags & SYNC_FILE_RANGE_WRITE) { |
Jan Kara | 23d0127 | 2015-11-06 16:28:55 -0800 | [diff] [blame] | 352 | ret = __filemap_fdatawrite_range(mapping, offset, endbyte, |
| 353 | WB_SYNC_NONE); |
Christoph Hellwig | 7a0ad10 | 2009-12-17 14:24:40 +0100 | [diff] [blame] | 354 | if (ret < 0) |
| 355 | goto out_put; |
| 356 | } |
| 357 | |
| 358 | if (flags & SYNC_FILE_RANGE_WAIT_AFTER) |
| 359 | ret = filemap_fdatawait_range(mapping, offset, endbyte); |
| 360 | |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 361 | out_put: |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 362 | fdput(f); |
Andrew Morton | f79e2ab | 2006-03-31 02:30:42 -0800 | [diff] [blame] | 363 | out: |
| 364 | return ret; |
| 365 | } |
| 366 | |
David Woodhouse | edd5cd4 | 2007-06-27 14:10:09 -0700 | [diff] [blame] | 367 | /* It would be nice if people remember that not all the world's an i386 |
| 368 | when they introduce new system calls */ |
Al Viro | 4a0fd5b | 2013-01-21 15:16:58 -0500 | [diff] [blame] | 369 | SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags, |
| 370 | loff_t, offset, loff_t, nbytes) |
David Woodhouse | edd5cd4 | 2007-06-27 14:10:09 -0700 | [diff] [blame] | 371 | { |
| 372 | return sys_sync_file_range(fd, offset, nbytes, flags); |
| 373 | } |