autofs-5.1.4 - set bind mount as propagation slave
From: Ian Kent <raven@themaw.net>
When using a multi-mount with mount targets that are on the
local machine and resulting bind mounts are made to a file
system that is mount propagation shared (such as the root
file system on systemd managed systems) the autofs offset
mount triggers made within the bind mount will be propagated
back to the target file system.
When this happens the target of the offset (the unwanted
propagated mount) is itself an autofs trigger mount and
accessing the path results in a deadlock.
In order for these multi-mounts to function properly in this
case bind mounts that contain mount triggers must be set to
propagation shared or private so the backward propagation
doesn't occur.
Signed-off-by: Ian Kent <raven@themaw.net>
---
CHANGELOG | 1 +
modules/mount_bind.c | 16 +++++++++++++++-
2 files changed, 16 insertions(+), 1 deletion(-)
--- autofs-5.0.7.orig/CHANGELOG
+++ autofs-5.0.7/CHANGELOG
@@ -307,6 +307,7 @@
- add port parameter to rpc_ping().
- dont probe NFSv2 by default.
- add version parameter to rpc_ping().
+- set bind mount as propagation slave.
25/07/2012 autofs-5.0.7
=======================
--- autofs-5.0.7.orig/modules/mount_bind.c
+++ autofs-5.0.7/modules/mount_bind.c
@@ -20,6 +20,7 @@
#include <sys/param.h>
#include <sys/types.h>
#include <sys/stat.h>
+#include <sys/mount.h>
#define MODULE_MOUNT
#include "automount.h"
@@ -184,8 +185,21 @@ int mount_mount(struct autofs_point *ap,
debug(ap->logopt,
MODPREFIX "mounted %s type %s on %s",
what, fstype, fullpath);
- return 0;
}
+
+ /* The bind mount has succeeded but if the target
+ * mount is propagation shared propagation of child
+ * mounts (autofs offset mounts for example) back to
+ * the target of the bind mount must be avoided or
+ * autofs trigger mounts will deadlock.
+ */
+ err = mount(NULL, fullpath, NULL, MS_SLAVE, NULL);
+ if (err)
+ warn(ap->logopt,
+ "failed to set propagation type for %s",
+ fullpath);
+
+ return 0;
} else {
char *cp;
char basepath[PATH_MAX];