fs/sync.c: sync_file_range(2) may use WB_SYNC_ALL writeback
authorAmir Goldstein <amir73il@gmail.com>
Tue, 14 May 2019 00:22:30 +0000 (17:22 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 14 May 2019 16:47:50 +0000 (09:47 -0700)
23d0127096cb ("fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE
writeback") claims that sync_file_range(2) syscall was "created for
userspace to be able to issue background writeout and so waiting for
in-flight IO is undesirable there" and changes the writeback (back) to
WB_SYNC_NONE.

This claim is only partially true.  It is true for users that use the flag
SYNC_FILE_RANGE_WRITE by itself, as does PostgreSQL, the user that was the
reason for changing to WB_SYNC_NONE writeback.

However, that claim is not true for users that use that flag combination
SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}.  Those users explicitly
requested to wait for in-flight IO as well as to writeback of dirty pages.

Re-brand that flag combination as SYNC_FILE_RANGE_WRITE_AND_WAIT and use
WB_SYNC_ALL writeback to perform the full range sync request.

Link: http://lkml.kernel.org/r/20190409114922.30095-1-amir73il@gmail.com
Link: http://lkml.kernel.org/r/20190419072938.31320-1-amir73il@gmail.com
Fixes: 23d0127096cb ("fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE")
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Acked-by: Jan Kara <jack@suse.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/sync.c
include/uapi/linux/fs.h

index 01e82170545a3037e0499ddd896b6c0489fbca8e..4d1ff010bc5afcd6acee1e4dbd736e8bfbd9081a 100644 (file)
--- a/fs/sync.c
+++ b/fs/sync.c
@@ -292,8 +292,14 @@ int sync_file_range(struct file *file, loff_t offset, loff_t nbytes,
        }
 
        if (flags & SYNC_FILE_RANGE_WRITE) {
+               int sync_mode = WB_SYNC_NONE;
+
+               if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
+                            SYNC_FILE_RANGE_WRITE_AND_WAIT)
+                       sync_mode = WB_SYNC_ALL;
+
                ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
-                                                WB_SYNC_NONE);
+                                                sync_mode);
                if (ret < 0)
                        goto out;
        }
@@ -306,9 +312,9 @@ out:
 }
 
 /*
- * sys_sync_file_range() permits finely controlled syncing over a segment of
+ * ksys_sync_file_range() permits finely controlled syncing over a segment of
  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
- * zero then sys_sync_file_range() will operate from offset out to EOF.
+ * zero then ksys_sync_file_range() will operate from offset out to EOF.
  *
  * The flag bits are:
  *
@@ -325,7 +331,7 @@ out:
  * Useful combinations of the flag bits are:
  *
  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
- * in the range which were dirty on entry to sys_sync_file_range() are placed
+ * in the range which were dirty on entry to ksys_sync_file_range() are placed
  * under writeout.  This is a start-write-for-data-integrity operation.
  *
  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
@@ -337,10 +343,13 @@ out:
  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
  * for that operation to complete and to return the result.
  *
- * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
+ * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
+ * (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT):
  * a traditional sync() operation.  This is a write-for-data-integrity operation
  * which will ensure that all pages in the range which were dirty on entry to
- * sys_sync_file_range() are committed to disk.
+ * ksys_sync_file_range() are written to disk.  It should be noted that disk
+ * caches are not flushed by this call, so there are no guarantees here that the
+ * data will be available on disk after a crash.
  *
  *
  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
index 121e82ce296b5a4d9e85757e94e3d9521897cb56..59c71fa8c553a394d516502538c188aa01502e2c 100644 (file)
@@ -320,6 +320,9 @@ struct fscrypt_key {
 #define SYNC_FILE_RANGE_WAIT_BEFORE    1
 #define SYNC_FILE_RANGE_WRITE          2
 #define SYNC_FILE_RANGE_WAIT_AFTER     4
+#define SYNC_FILE_RANGE_WRITE_AND_WAIT (SYNC_FILE_RANGE_WRITE | \
+                                        SYNC_FILE_RANGE_WAIT_BEFORE | \
+                                        SYNC_FILE_RANGE_WAIT_AFTER)
 
 /*
  * Flags for preadv2/pwritev2: