On Sun, Jul 14, 2013 at 12:54 AM, Rafael Neves <[email protected]> wrote:
> The patch below fixes a bug on sysctl(8) introduced by revision 1.191
> of sysctl.c. After rev1.191, `sysctl vfs' mangles information about
> filesystems (mounted instances of ffs are attributed to nfs, of nfs
> are atrributed to mfs, and so on). As a consequence, `sysctl
> vfs.mounts.nfs' reports 0 mounted instances on a diskless(8) setup,
> thus /etc/rc script (lines 335 to 342) doesn't add pf rules that allow
> NFS, and system hangs when it enables pf.
First off: thank you for (a) noticing this, and (b) tracking down the mismatch.
...
> --- sysctl.c 9 Jun 2013 12:54:38 -0000 1.192
> +++ sysctl.c 14 Jul 2013 07:09:28 -0000
> @@ -1175,8 +1175,8 @@ vfsinit(void)
>
> vfsname[0].ctl_name = "mounts";
> vfsname[0].ctl_type = CTLTYPE_NODE;
> - vfsvars[0].list = vfsname + 1;
> - vfsvars[0].size = maxtypenum - 1;
> + vfsvars[0].list = vfsname;
> + vfsvars[0].size = maxtypenum;
Soooooo close...
While this fixes the observed problem, it's not 100% correct. The
glitch is that it fails a negative test: the command
sysctl vfs.mounts.mounts
should fail with the error
sysctl: third level name mounts in vfs.mounts.mounts is invalid
but with your patch it silently succeeds. The vfsname list is offset
by one in vfsvars[0].list to prevent that, so the fix that avoids the
unwanted match against vfsname[0] is to keep the offset, but undo it
in the lookup:
--- sysctl.c 9 Jun 2013 12:54:38 -0000 1.192
+++ sysctl.c 15 Jul 2013 03:43:27 -0000
@@ -1200,7 +1200,7 @@ sysctl_vfsgen(char *string, char **bufpp
mib[1] = VFS_GENERIC;
mib[2] = VFS_CONF;
- mib[3] = indx;
+ mib[3] = indx + 1;
size = sizeof vfc;
if (sysctl(mib, 4, &vfc, &size, (void *)0, (size_t)0) < 0) {
if (errno != EOPNOTSUPP)
That make sense?
Philip Guenther