sync.c 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378
  1. /*
  2. * High-level sync()-related operations
  3. */
  4. #include <linux/kernel.h>
  5. #include <linux/file.h>
  6. #include <linux/fs.h>
  7. #include <linux/slab.h>
  8. #include <linux/export.h>
  9. #include <linux/namei.h>
  10. #include <linux/sched.h>
  11. #include <linux/writeback.h>
  12. #include <linux/syscalls.h>
  13. #include <linux/linkage.h>
  14. #include <linux/pagemap.h>
  15. #include <linux/quotaops.h>
  16. #include <linux/backing-dev.h>
  17. #include "internal.h"
  18. #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
  19. SYNC_FILE_RANGE_WAIT_AFTER)
  20. /*
  21. * Do the filesystem syncing work. For simple filesystems
  22. * writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
  23. * submit IO for these buffers via __sync_blockdev(). This also speeds up the
  24. * wait == 1 case since in that case write_inode() functions do
  25. * sync_dirty_buffer() and thus effectively write one block at a time.
  26. */
  27. static int __sync_filesystem(struct super_block *sb, int wait,
  28. unsigned long start)
  29. {
  30. if (wait)
  31. sync_inodes_sb(sb, start);
  32. else
  33. writeback_inodes_sb(sb, WB_REASON_SYNC);
  34. if (sb->s_op->sync_fs)
  35. sb->s_op->sync_fs(sb, wait);
  36. return __sync_blockdev(sb->s_bdev, wait);
  37. }
  38. /*
  39. * Write out and wait upon all dirty data associated with this
  40. * superblock. Filesystem data as well as the underlying block
  41. * device. Takes the superblock lock.
  42. */
  43. int sync_filesystem(struct super_block *sb)
  44. {
  45. int ret;
  46. unsigned long start = jiffies;
  47. /*
  48. * We need to be protected against the filesystem going from
  49. * r/o to r/w or vice versa.
  50. */
  51. WARN_ON(!rwsem_is_locked(&sb->s_umount));
  52. /*
  53. * No point in syncing out anything if the filesystem is read-only.
  54. */
  55. if (sb->s_flags & MS_RDONLY)
  56. return 0;
  57. ret = __sync_filesystem(sb, 0, start);
  58. if (ret < 0)
  59. return ret;
  60. return __sync_filesystem(sb, 1, start);
  61. }
  62. EXPORT_SYMBOL_GPL(sync_filesystem);
  63. static void sync_inodes_one_sb(struct super_block *sb, void *arg)
  64. {
  65. if (!(sb->s_flags & MS_RDONLY))
  66. sync_inodes_sb(sb, *((unsigned long *)arg));
  67. }
  68. static void sync_fs_one_sb(struct super_block *sb, void *arg)
  69. {
  70. if (!(sb->s_flags & MS_RDONLY) && sb->s_op->sync_fs)
  71. sb->s_op->sync_fs(sb, *(int *)arg);
  72. }
  73. static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
  74. {
  75. filemap_fdatawrite(bdev->bd_inode->i_mapping);
  76. }
  77. static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
  78. {
  79. filemap_fdatawait(bdev->bd_inode->i_mapping);
  80. }
  81. /*
  82. * Sync everything. We start by waking flusher threads so that most of
  83. * writeback runs on all devices in parallel. Then we sync all inodes reliably
  84. * which effectively also waits for all flusher threads to finish doing
  85. * writeback. At this point all data is on disk so metadata should be stable
  86. * and we tell filesystems to sync their metadata via ->sync_fs() calls.
  87. * Finally, we writeout all block devices because some filesystems (e.g. ext2)
  88. * just write metadata (such as inodes or bitmaps) to block device page cache
  89. * and do not sync it on their own in ->sync_fs().
  90. */
  91. SYSCALL_DEFINE0(sync)
  92. {
  93. int nowait = 0, wait = 1;
  94. unsigned long start = jiffies;
  95. wakeup_flusher_threads(0, WB_REASON_SYNC);
  96. iterate_supers(sync_inodes_one_sb, &start);
  97. iterate_supers(sync_fs_one_sb, &nowait);
  98. iterate_supers(sync_fs_one_sb, &wait);
  99. iterate_bdevs(fdatawrite_one_bdev, NULL);
  100. iterate_bdevs(fdatawait_one_bdev, NULL);
  101. if (unlikely(laptop_mode))
  102. laptop_sync_completion();
  103. return 0;
  104. }
  105. static void do_sync_work(struct work_struct *work)
  106. {
  107. int nowait = 0;
  108. /*
  109. * Sync twice to reduce the possibility we skipped some inodes / pages
  110. * because they were temporarily locked
  111. */
  112. iterate_supers(sync_inodes_one_sb, &nowait);
  113. iterate_supers(sync_fs_one_sb, &nowait);
  114. iterate_bdevs(fdatawrite_one_bdev, NULL);
  115. iterate_supers(sync_inodes_one_sb, &nowait);
  116. iterate_supers(sync_fs_one_sb, &nowait);
  117. iterate_bdevs(fdatawrite_one_bdev, NULL);
  118. printk("Emergency Sync complete\n");
  119. kfree(work);
  120. }
  121. void emergency_sync(void)
  122. {
  123. struct work_struct *work;
  124. work = kmalloc(sizeof(*work), GFP_ATOMIC);
  125. if (work) {
  126. INIT_WORK(work, do_sync_work);
  127. schedule_work(work);
  128. }
  129. }
  130. /*
  131. * sync a single super
  132. */
  133. SYSCALL_DEFINE1(syncfs, int, fd)
  134. {
  135. struct fd f = fdget(fd);
  136. struct super_block *sb;
  137. int ret;
  138. if (!f.file)
  139. return -EBADF;
  140. sb = f.file->f_dentry->d_sb;
  141. down_read(&sb->s_umount);
  142. ret = sync_filesystem(sb);
  143. up_read(&sb->s_umount);
  144. fdput(f);
  145. return ret;
  146. }
  147. /**
  148. * vfs_fsync_range - helper to sync a range of data & metadata to disk
  149. * @file: file to sync
  150. * @start: offset in bytes of the beginning of data range to sync
  151. * @end: offset in bytes of the end of data range (inclusive)
  152. * @datasync: perform only datasync
  153. *
  154. * Write back data in range @start..@end and metadata for @file to disk. If
  155. * @datasync is set only metadata needed to access modified file data is
  156. * written.
  157. */
  158. int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
  159. {
  160. if (!file->f_op->fsync)
  161. return -EINVAL;
  162. return file->f_op->fsync(file, start, end, datasync);
  163. }
  164. EXPORT_SYMBOL(vfs_fsync_range);
  165. /**
  166. * vfs_fsync - perform a fsync or fdatasync on a file
  167. * @file: file to sync
  168. * @datasync: only perform a fdatasync operation
  169. *
  170. * Write back data and metadata for @file to disk. If @datasync is
  171. * set only metadata needed to access modified file data is written.
  172. */
  173. int vfs_fsync(struct file *file, int datasync)
  174. {
  175. return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
  176. }
  177. EXPORT_SYMBOL(vfs_fsync);
  178. static int do_fsync(unsigned int fd, int datasync)
  179. {
  180. struct fd f = fdget(fd);
  181. int ret = -EBADF;
  182. if (f.file) {
  183. ret = vfs_fsync(f.file, datasync);
  184. fdput(f);
  185. }
  186. return ret;
  187. }
  188. SYSCALL_DEFINE1(fsync, unsigned int, fd)
  189. {
  190. return do_fsync(fd, 0);
  191. }
  192. SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
  193. {
  194. return do_fsync(fd, 1);
  195. }
  196. /**
  197. * generic_write_sync - perform syncing after a write if file / inode is sync
  198. * @file: file to which the write happened
  199. * @pos: offset where the write started
  200. * @count: length of the write
  201. *
  202. * This is just a simple wrapper about our general syncing function.
  203. */
  204. int generic_write_sync(struct file *file, loff_t pos, loff_t count)
  205. {
  206. if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
  207. return 0;
  208. return vfs_fsync_range(file, pos, pos + count - 1,
  209. (file->f_flags & __O_SYNC) ? 0 : 1);
  210. }
  211. EXPORT_SYMBOL(generic_write_sync);
  212. /*
  213. * sys_sync_file_range() permits finely controlled syncing over a segment of
  214. * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
  215. * zero then sys_sync_file_range() will operate from offset out to EOF.
  216. *
  217. * The flag bits are:
  218. *
  219. * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
  220. * before performing the write.
  221. *
  222. * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
  223. * range which are not presently under writeback. Note that this may block for
  224. * significant periods due to exhaustion of disk request structures.
  225. *
  226. * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
  227. * after performing the write.
  228. *
  229. * Useful combinations of the flag bits are:
  230. *
  231. * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
  232. * in the range which were dirty on entry to sys_sync_file_range() are placed
  233. * under writeout. This is a start-write-for-data-integrity operation.
  234. *
  235. * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
  236. * are not presently under writeout. This is an asynchronous flush-to-disk
  237. * operation. Not suitable for data integrity operations.
  238. *
  239. * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
  240. * completion of writeout of all pages in the range. This will be used after an
  241. * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
  242. * for that operation to complete and to return the result.
  243. *
  244. * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
  245. * a traditional sync() operation. This is a write-for-data-integrity operation
  246. * which will ensure that all pages in the range which were dirty on entry to
  247. * sys_sync_file_range() are committed to disk.
  248. *
  249. *
  250. * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
  251. * I/O errors or ENOSPC conditions and will return those to the caller, after
  252. * clearing the EIO and ENOSPC flags in the address_space.
  253. *
  254. * It should be noted that none of these operations write out the file's
  255. * metadata. So unless the application is strictly performing overwrites of
  256. * already-instantiated disk blocks, there are no guarantees here that the data
  257. * will be available after a crash.
  258. */
  259. SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
  260. unsigned int, flags)
  261. {
  262. int ret;
  263. struct fd f;
  264. struct address_space *mapping;
  265. loff_t endbyte; /* inclusive */
  266. umode_t i_mode;
  267. ret = -EINVAL;
  268. if (flags & ~VALID_FLAGS)
  269. goto out;
  270. endbyte = offset + nbytes;
  271. if ((s64)offset < 0)
  272. goto out;
  273. if ((s64)endbyte < 0)
  274. goto out;
  275. if (endbyte < offset)
  276. goto out;
  277. if (sizeof(pgoff_t) == 4) {
  278. if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
  279. /*
  280. * The range starts outside a 32 bit machine's
  281. * pagecache addressing capabilities. Let it "succeed"
  282. */
  283. ret = 0;
  284. goto out;
  285. }
  286. if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
  287. /*
  288. * Out to EOF
  289. */
  290. nbytes = 0;
  291. }
  292. }
  293. if (nbytes == 0)
  294. endbyte = LLONG_MAX;
  295. else
  296. endbyte--; /* inclusive */
  297. ret = -EBADF;
  298. f = fdget(fd);
  299. if (!f.file)
  300. goto out;
  301. i_mode = file_inode(f.file)->i_mode;
  302. ret = -ESPIPE;
  303. if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
  304. !S_ISLNK(i_mode))
  305. goto out_put;
  306. mapping = f.file->f_mapping;
  307. if (!mapping) {
  308. ret = -EINVAL;
  309. goto out_put;
  310. }
  311. ret = 0;
  312. if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
  313. ret = filemap_fdatawait_range(mapping, offset, endbyte);
  314. if (ret < 0)
  315. goto out_put;
  316. }
  317. if (flags & SYNC_FILE_RANGE_WRITE) {
  318. ret = filemap_fdatawrite_range(mapping, offset, endbyte);
  319. if (ret < 0)
  320. goto out_put;
  321. }
  322. if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
  323. ret = filemap_fdatawait_range(mapping, offset, endbyte);
  324. out_put:
  325. fdput(f);
  326. out:
  327. return ret;
  328. }
  329. /* It would be nice if people remember that not all the world's an i386
  330. when they introduce new system calls */
  331. SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
  332. loff_t, offset, loff_t, nbytes)
  333. {
  334. return sys_sync_file_range(fd, offset, nbytes, flags);
  335. }