Date: Thu, 20 Jul 2023 21:37:45 -0500 From: Mike Karels <mike@karels.net> To: Mateusz Guzik <mjguzik@gmail.com> Cc: Current FreeBSD <freebsd-current@freebsd.org> Subject: Re: confusion about root partition causes panic during startup Message-ID: <202307210237.36L2bjqk049891@mail2.karels.net> In-Reply-To: Your message of Thu, 20 Jul 2023 23:13:31 %2B0200. <CAGudoHEgVf6=kFrrCQjce0wjVRQgWJZyZ-bWDz87gse9qFgJQA@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Mateusz Guzik <mjguzik@gmail.com> wrote:
> On 7/20/23, Mike Karels <mike@karels.net> wrote:
> > I installed an additional NVME drive on a system, and then booted. It
> > turns
> > out that the new drive became nda0, renumbering the other drives. The
> > loader
> > found the correct partition to boot (the only choice), and loaded the
> > kernel
> > correctly. However, /etc/fstab still had the old name (nvd1p2), which=
is
> > now drive 2. I expected it to drop into single user, but instead the
> > system
> > panicked in vfs_mountroot_shuffle trying to switch root devices (see
> > below).
> > It doesn't seem that having the wrong root device in /etc/fstab should
> > cause
> > a panic; it makes it harder to patch the system. I was unable to get =
the
> > system to boot using boot-to-single-user or setting currdev, but I man=
aged
> > to remember doing "boot -a" from a loader prompt to get the system to =
ask
> > the root device before mounting it. I can easily reproduce this to te=
st.
> > Probably the NDFREE_PNBUF() shouldn't happen if namei() returned an er=
ror.
> >
> ye, this should do it (untested):
> diff --git a/sys/kern/vfs_mountroot.c b/sys/kern/vfs_mountroot.c
> index 956d29e3f084..85398ff781e4 100644
> --- a/sys/kern/vfs_mountroot.c
> +++ b/sys/kern/vfs_mountroot.c
> @@ -352,13 +352,13 @@ vfs_mountroot_shuffle(struct thread *td, struct
> mount *mpdevfs)
> NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fsp=
ath);
> error =3D namei(&nd);
> if (error) {
> - NDFREE_PNBUF(&nd);
> fspath =3D "/mnt";
> NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSP=
ACE,
> fspath);
> error =3D namei(&nd);
> }
> if (!error) {
> + NDFREE_PNBUF(&nd);
> vp =3D nd.ni_vp;
> error =3D (vp->v_type =3D=3D VDIR) ? 0 : ENOTDIR=
;
> if (!error)
> @@ -376,7 +376,6 @@ vfs_mountroot_shuffle(struct thread *td, struct
> mount *mpdevfs)
> } else
> vput(vp);
> }
> - NDFREE_PNBUF(&nd);
> if (error)
> printf("mountroot: unable to remount previous ro=
ot "
> @@ -387,6 +386,7 @@ vfs_mountroot_shuffle(struct thread *td, struct
> mount *mpdevfs)
> NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev");
> error =3D namei(&nd);
> if (!error) {
> + NDFREE_PNBUF(&nd);
> vp =3D nd.ni_vp;
> error =3D (vp->v_type =3D=3D VDIR) ? 0 : ENOTDIR;
> if (!error)
That was missing the last change, and tabs were expanded. I put it in
by hand, and the patch works, at least to avoid this panic. It still
insisted on remounting root on nda1p2, which is not a root file system.
Remounting /dev still failed without panicking, then it panicked because
there was no /sbin/init. Apparently it is necessary to use "boot -a"
in this situation. Too bad the loader option menu doesn't include that.
Just to be clear what I tested, my patch follows.
Mike
diff --git a/sys/kern/vfs_mountroot.c b/sys/kern/vfs_mountroot.c
index 956d29e3f084..b08b2a3200f8 100644
--- a/sys/kern/vfs_mountroot.c
+++ b/sys/kern/vfs_mountroot.c
@@ -352,13 +352,13 @@ vfs_mountroot_shuffle(struct thread *td, struct moun=
t *mpdevfs)
NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath);
error =3D namei(&nd);
if (error) {
- NDFREE_PNBUF(&nd);
fspath =3D "/mnt";
NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
fspath);
error =3D namei(&nd);
}
if (!error) {
+ NDFREE_PNBUF(&nd);
vp =3D nd.ni_vp;
error =3D (vp->v_type =3D=3D VDIR) ? 0 : ENOTDIR;
if (!error)
@@ -376,7 +376,6 @@ vfs_mountroot_shuffle(struct thread *td, struct mount =
*mpdevfs)
} else
vput(vp);
}
- NDFREE_PNBUF(&nd);
=
if (error)
printf("mountroot: unable to remount previous root "
@@ -387,6 +386,7 @@ vfs_mountroot_shuffle(struct thread *td, struct mount =
*mpdevfs)
NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev");
error =3D namei(&nd);
if (!error) {
+ NDFREE_PNBUF(&nd);
vp =3D nd.ni_vp;
error =3D (vp->v_type =3D=3D VDIR) ? 0 : ENOTDIR;
if (!error)
@@ -413,7 +413,6 @@ vfs_mountroot_shuffle(struct thread *td, struct mount =
*mpdevfs)
if (error)
printf("mountroot: unable to remount devfs under /dev "
"(error %d)\n", error);
- NDFREE_PNBUF(&nd);
=
if (mporoot =3D=3D mpdevfs) {
vfs_unbusy(mpdevfs);
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202307210237.36L2bjqk049891>