Re: [PATCH v1 2/6] seccomp: prepare seccomp_run_filters() to support more than one listener

[Aleksandr Mikhalitsyn]

On Tue, Dec 2, 2025 at 9:26 PM Kees Cook <kees@xxxxxxxxxx> wrote:
>
> On Mon, Dec 01, 2025 at 01:23:59PM +0100, Alexander Mikhalitsyn wrote:
> > Prepare seccomp_run_filters() function to support more than one listener
> > in the seccomp tree. In this patch, we only introduce a new
> > struct seccomp_filter_matches with kdoc and modify seccomp_run_filters()
> > signature correspondingly.
> >
> > No functional change intended.
> >
> > Cc: linux-doc@xxxxxxxxxxxxxxx
> > Cc: linux-kernel@xxxxxxxxxxxxxxx
> > Cc: Kees Cook <kees@xxxxxxxxxx>
> > Cc: Andy Lutomirski <luto@xxxxxxxxxxxxxx>
> > Cc: Will Drewry <wad@xxxxxxxxxxxx>
> > Cc: Jonathan Corbet <corbet@xxxxxxx>
> > Cc: Shuah Khan <shuah@xxxxxxxxxx>
> > Cc: Tycho Andersen <tycho@tycho.pizza>
> > Cc: Andrei Vagin <avagin@xxxxxxxxx>
> > Cc: Christian Brauner <brauner@xxxxxxxxxx>
> > Cc: Stéphane Graber <stgraber@xxxxxxxxxxxx>
> > Signed-off-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@xxxxxxxxxxxxx>
> > ---
> >  kernel/seccomp.c | 35 +++++++++++++++++++++++++++++++----
> >  1 file changed, 31 insertions(+), 4 deletions(-)
> >
> > diff --git a/kernel/seccomp.c b/kernel/seccomp.c
> > index f944ea5a2716..c9a1062a53bd 100644
> > --- a/kernel/seccomp.c
> > +++ b/kernel/seccomp.c
> > @@ -237,6 +237,9 @@ struct seccomp_filter {
> >  /* Limit any path through the tree to 256KB worth of instructions. */
> >  #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
> >
> > +/* Limit number of listeners through the tree. */
> > +#define MAX_LISTENERS_PER_PATH 8
> > +
> >  /*
> >   * Endianness is explicitly ignored and left for BPF program authors to manage
> >   * as per the specific architecture.
> > @@ -391,18 +394,38 @@ static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilte
> >  }
> >  #endif /* SECCOMP_ARCH_NATIVE */
> >
> > +/**
> > + * struct seccomp_filter_matches - container for seccomp filter match results
> > + *
> > + * @n: A number of filters matched.
> > + * @filters: An array of (struct seccomp_filter) pointers.
> > + *        Holds pointers to filters that matched during evaluation.
> > + *        A first one in the array is the one with the least permissive
> > + *        action result.
> > + *
> > + * If final action result is less (or more) permissive than SECCOMP_RET_USER_NOTIF,
> > + * only the most restrictive filter is stored in the array's first element.
> > + * If final action result is SECCOMP_RET_USER_NOTIF, we need to track
> > + * all filters that resulted in the same action to support multiple listeners
> > + * in seccomp tree.
> > + */
> > +struct seccomp_filter_matches {
> > +     unsigned char n;
> > +     struct seccomp_filter *filters[MAX_LISTENERS_PER_PATH];
> > +};
> > +
> >  #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
> >  /**
> >   * seccomp_run_filters - evaluates all seccomp filters against @sd
> >   * @sd: optional seccomp data to be passed to filters
> > - * @match: stores struct seccomp_filter that resulted in the return value,
> > + * @matches: array of struct seccomp_filter pointers that resulted in the return value,
> >   *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
> >   *         be unchanged.
> >   *
> >   * Returns valid seccomp BPF response codes.
> >   */
> >  static u32 seccomp_run_filters(const struct seccomp_data *sd,
> > -                            struct seccomp_filter **match)
> > +                            struct seccomp_filter_matches *matches)
> >  {
> >       u32 ret = SECCOMP_RET_ALLOW;
> >       /* Make sure cross-thread synced filter points somewhere sane. */
> > @@ -425,7 +448,8 @@ static u32 seccomp_run_filters(const struct seccomp_data *sd,
> >
> >               if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
> >                       ret = cur_ret;
> > -                     *match = f;
> > +                     matches->n = 1;
> > +                     matches->filters[0] = f;
> >               }
> >       }
> >       return ret;
> > @@ -1252,6 +1276,7 @@ static int __seccomp_filter(int this_syscall, const bool recheck_after_trace)
> >  {
> >       u32 filter_ret, action;
> >       struct seccomp_data sd;
> > +     struct seccomp_filter_matches matches = {};
>

Hi Kees,

Thanks for looking into this stuff! ;)

> I was surprised to see this didn't induce a stack protector check (due
> to the array use). It does, however, expand the work done to clear local
> variables (i.e. this adds 9 unsigned long zeroings to the default case).

Actually, by saying this you've inspired me to look at this stuff with
a fresh mind again
and I have a feeling that I can probably make it work with just one additional
struct seccomp_filter pointer on the stack, instead of adding extra 8
pointers... :)
Let me make another iteration on this and send a -v3 then.

>I was surprised to see this didn't induce a stack protector check (due
> to the array use).

Also, sorry if my question is stupid, but what do you mean by stack
protector check in
this case? Just a check for an array index before writing to it? Or
something more generic?

>
> Regardless, I'll read this thread more closely in time for the LPC
> session; I'm not exactly opposed to allowing multiple listeners, but I
> do want to meditate on the safety logic (which I see you've spent time
> thinking about too).

Thanks, Kees!

>
> Thanks!

Kind regards,
Alex

>
> -Kees
>
> --
> Kees Cook