On Sat, Sep 17, 2016 at 09:20:31PM +0200, Theo Buehler wrote:
> While this patch avoids the bug, it isn't quite right: pstat -ft should
> be equivalent to "pstat -f && pstat -t" but...
>
> $ pstat -ft
> pstat: kvm_openfiles: /dev/mem: Permission denied
>
> The actual problem is in ttymode(): if kd != NULL (which is the case if
> -f is combined with -t), there are accesses of the kvm datastructures
> without them being properly initialized:
>
> 879 if (kd != NULL)
> 880 KGET(TTY_NTTY, ntty);
>
> and
>
> 896 KGET(TTY_TTYLIST, tty_head);
> 897 for (tp = TAILQ_FIRST(_head); tp;
> 898 tp = TAILQ_NEXT(, tty_link)) {
> 899 KGET2(tp, , sizeof tty, "tty struct");
> 900 tty2itty(, );
> 901 ttyprt();
> 902 }
>
> These code paths must not be hit unless need_nlist is true, so expose it
> globally and replace 'kd != NULL' with the proper condition need_nlist
>
> The following patch works in all affected cases, pstat -t, pstat -f,
> pstat -tf, both as root and non-root as well as pstat -tv as root
> (combinations with -d aren't interesting since pstat exits before
> doing anything else but -d).
>
> Index: pstat.c
> ===
> RCS file: /var/cvs/src/usr.sbin/pstat/pstat.c,v
> retrieving revision 1.108
> diff -u -p -r1.108 pstat.c
> --- pstat.c 14 Aug 2016 22:47:26 - 1.108
> +++ pstat.c 17 Sep 2016 17:57:38 -
> @@ -89,6 +89,7 @@ struct e_vnode {
> struct vnode vnode;
> };
>
> +int need_nlist;
> int usenumflag;
> int totalflag;
> int kflag;
> @@ -141,7 +142,7 @@ main(int argc, char *argv[])
> const char *dformat = NULL;
> extern char *optarg;
> extern int optind;
> - int i, need_nlist;
> + int i;
>
> hideroot = getuid();
>
> @@ -869,7 +870,7 @@ ttymode(void)
> struct itty itty, *itp;
> size_t nlen;
>
> - if (kd == 0) {
> + if (!need_nlist) {
> mib[0] = CTL_KERN;
> mib[1] = KERN_TTYCOUNT;
> nlen = sizeof(ntty);
> @@ -879,7 +880,7 @@ ttymode(void)
> KGET(TTY_NTTY, ntty);
> (void)printf("%d terminal device%s\n", ntty, ntty == 1 ? "" : "s");
> (void)printf("%s", hdr);
> - if (kd == 0) {
> + if (!need_nlist) {
> mib[0] = CTL_KERN;
> mib[1] = KERN_TTY;
> mib[2] = KERN_TTY_INFO;
>
Thanks Theo.
Fixes the regress and works well with your pending pledge diff.
Rob
