[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openbsd-tech
Subject:    ensure device revoke unmounts
From:       Alexander Bluhm <alexander.bluhm () gmx ! net>
Date:       2016-01-29 14:50:43
Message-ID: 20160129145043.GC11559 () t430s ! bluhm ! invalid
[Download RAW message or body]

Hi,

When removing an umass USB stick with a mounted file system, it
must get unmounted.  Otherwise the kernel will crash later.  While
reading the code, I discovered that vfs_busy() is using RW_SLEEPFAIL.
So if another regular non-forced unmount is unsuccessful while
vop_generic_revoke() does not unmount because the mount point is
busy, this will result in a mount point without a valid device.

So I think it is better to check and sleep in a loop, although I
did not see this problem in practice.

ok?

bluhm

Index: kern/vfs_default.c
===================================================================
RCS file: /data/mirror/openbsd/cvs/src/sys/kern/vfs_default.c,v
retrieving revision 1.41
diff -u -p -r1.41 vfs_default.c
--- kern/vfs_default.c	14 Mar 2015 03:38:51 -0000	1.41
+++ kern/vfs_default.c	27 Jan 2016 17:53:10 -0000
@@ -66,7 +66,8 @@ vop_generic_revoke(void *v)
 
 	vp = ap->a_vp;
  
-	if (vp->v_type == VBLK && vp->v_specinfo != 0) {
+	while (vp->v_type == VBLK && vp->v_specinfo != NULL &&
+	    vp->v_specmountpoint != NULL) {
 		struct mount *mp = vp->v_specmountpoint;
 
 		/*
@@ -74,8 +75,10 @@ vop_generic_revoke(void *v)
 		 * flush it out now, as to not leave a dangling zombie mount
 		 * point laying around in VFS.
 		 */
-		if (mp != NULL && !vfs_busy(mp, VB_WRITE|VB_WAIT))
+		if (!vfs_busy(mp, VB_WRITE|VB_WAIT)) {
 			dounmount(mp, MNT_FORCE | MNT_DOOMED, p, NULL);
+			break;
+		}
 	}
 
 	if (vp->v_flag & VALIASED) {

[prev in list] [next in list] [prev in thread] [next in thread]