[V3,15/18] multipath: Fix kpartx and udevd race

Commit Message

Benjamin Marzinski Jan. 12, 2013, 6:04 a.m. UTC

Sometimes when kpartx is used to view partition data on disk image files,
udev still has the loop device open when kpartx is trying to tear it down.
This causes the LOOP_CLR_FD ioctl to fail with EBUSY.  kpartx now retries
in this case.

Signed-off-by: Benjamin Marzinski <[e-mail]>
---
 kpartx/lopart.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

Patch

diff --git a/kpartx/lopart.c b/kpartx/lopart.c
index 79d8328..b1007e1 100644
--- a/kpartx/lopart.c
+++ b/kpartx/lopart.c
@@ -286,6 +286,7 @@  set_loop (const char *device, const char *file, int offset, int *loopro)
 extern int 
 del_loop (const char *device)
 {
+	int retries = 3;
 	int fd;
 
 	if ((fd = open (device, O_RDONLY)) < 0) {
@@ -295,10 +296,17 @@  del_loop (const char *device)
 		return 1;
 	}
 
-	if (ioctl (fd, LOOP_CLR_FD, 0) < 0) {
-		perror ("ioctl: LOOP_CLR_FD");
-		close (fd);
-		return 1;
+	while (ioctl (fd, LOOP_CLR_FD, 0) < 0) {
+		if (errno != EBUSY || retries-- <= 0) {
+			perror ("ioctl: LOOP_CLR_FD");
+			close (fd);
+			return 1;
+		}
+		fprintf(stderr,
+			"loop: device %s still in use, retrying delete\n",
+			device);
+		sleep(1);
+		continue;
 	}
 
 	close (fd);