Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 16 Feb 2018 21:44:37 +0000
From:      bugzilla-noreply@freebsd.org
To:        freebsd-bugs@FreeBSD.org
Subject:   [Bug 225960] zfs: g_access leak when unmounting UFS on a zvol
Message-ID:  <bug-225960-8-oIvLJBZGy1@https.bugs.freebsd.org/bugzilla/>
In-Reply-To: <bug-225960-8@https.bugs.freebsd.org/bugzilla/>
References:  <bug-225960-8@https.bugs.freebsd.org/bugzilla/>

next in thread | previous in thread | raw e-mail | index | archive | help
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=3D225960

--- Comment #5 from Andriy Gapon <avg@FreeBSD.org> ---
I now think that this could be a / the deep problem in GEOM.
The problem is that g_access() must be called with the GEOM topology lock h=
eld.
And that gives a false impression that the lock is indeed held across the c=
all.
But this isn't true because many classes, ZVOL being one of the many, need =
to
perform I/O in their access method.  So, they must drop and pick up the
topology lock.
That, of course, can break many assumptions.

Specifically, looking at g_slice_access() we can see that the code assumes =
that
all invocations are serialized.  Indeed, if another call to g_slice_access(=
) is
permitted while the consumer's access bits are not updated yet, then the
following condition can be true multiple times:

        /* On first open, grab an extra "exclusive" bit */
        if (cp->acr =3D=3D 0 && cp->acw =3D=3D 0 && cp->ace =3D=3D 0)
                de++;

And if that ever happens and we have extra grabs on cp->ace, then the follo=
wing
condition will never be true (because cp->ace + de > 1):

        /* ... and let go of it on last close */
        if ((cp->acr + dr) =3D=3D 0 && (cp->acw + dw) =3D=3D 0 && (cp->ace =
+ de) =3D=3D 1)
                de--;

--=20
You are receiving this mail because:
You are the assignee for the bug.=



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?bug-225960-8-oIvLJBZGy1>