On Fri, Nov 01, 2024 at 11:50:54PM +0000, Roman Gushchin wrote:
> On Sat, Nov 02, 2024 at 12:28:38AM +0100, Alexey Gladkov wrote:
> > On Thu, Oct 31, 2024 at 08:04:38PM +0000, Roman Gushchin wrote:
> > > Prior to commit d64696905554 ("Reimplement RLIMIT_SIGPENDING on top of
> > > ucounts") UCOUNT_RLIMIT_SIGPENDING rlimit was not enforced for a class
> > > of signals. However now it's enforced unconditionally, even if
> > > override_rlimit is set. This behavior change caused production issues.
> > >
> > > For example, if the limit is reached and a process receives a SIGSEGV
> > > signal, sigqueue_alloc fails to allocate the necessary resources for the
> > > signal delivery, preventing the signal from being delivered with
> > > siginfo. This prevents the process from correctly identifying the fault
> > > address and handling the error. From the user-space perspective,
> > > applications are unaware that the limit has been reached and that the
> > > siginfo is effectively 'corrupted'. This can lead to unpredictable
> > > behavior and crashes, as we observed with java applications.
> > >
> > > Fix this by passing override_rlimit into inc_rlimit_get_ucounts() and
> > > skip the comparison to max there if override_rlimit is set. This
> > > effectively restores the old behavior.
> > >
> > > Fixes: d64696905554 ("Reimplement RLIMIT_SIGPENDING on top of ucounts")
> > > Signed-off-by: Roman Gushchin <roman.gushchin@xxxxxxxxx>
> > > Co-developed-by: Andrei Vagin <avagin@xxxxxxxxxx>
> > > Signed-off-by: Andrei Vagin <avagin@xxxxxxxxxx>
> > > Cc: Kees Cook <kees@xxxxxxxxxx>
> > > Cc: "Eric W. Biederman" <ebiederm@xxxxxxxxxxxx>
> > > Cc: Alexey Gladkov <legion@xxxxxxxxxx>
> > > Cc: <stable@xxxxxxxxxxxxxxx>
> > > ---
> > > include/linux/user_namespace.h | 3 ++-
> > > kernel/signal.c | 3 ++-
> > > kernel/ucount.c | 5 +++--
> > > 3 files changed, 7 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/include/linux/user_namespace.h b/include/linux/user_namespace.h
> > > index 3625096d5f85..7183e5aca282 100644
> > > --- a/include/linux/user_namespace.h
> > > +++ b/include/linux/user_namespace.h
> > > @@ -141,7 +141,8 @@ static inline long get_rlimit_value(struct ucounts *ucounts, enum rlimit_type ty
> > >
> > > long inc_rlimit_ucounts(struct ucounts *ucounts, enum rlimit_type type, long v);
> > > bool dec_rlimit_ucounts(struct ucounts *ucounts, enum rlimit_type type, long v);
> > > -long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum rlimit_type type);
> > > +long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum rlimit_type type,
> > > + bool override_rlimit);
> > > void dec_rlimit_put_ucounts(struct ucounts *ucounts, enum rlimit_type type);
> > > bool is_rlimit_overlimit(struct ucounts *ucounts, enum rlimit_type type, unsigned long max);
> > >
> > > diff --git a/kernel/signal.c b/kernel/signal.c
> > > index 4344860ffcac..cbabb2d05e0a 100644
> > > --- a/kernel/signal.c
> > > +++ b/kernel/signal.c
> > > @@ -419,7 +419,8 @@ __sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
> > > */
> > > rcu_read_lock();
> > > ucounts = task_ucounts(t);
> > > - sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
> > > + sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING,
> > > + override_rlimit);
> > > rcu_read_unlock();
> > > if (!sigpending)
> > > return NULL;
> > > diff --git a/kernel/ucount.c b/kernel/ucount.c
> > > index 16c0ea1cb432..046b3d57ebb4 100644
> > > --- a/kernel/ucount.c
> > > +++ b/kernel/ucount.c
> > > @@ -307,7 +307,8 @@ void dec_rlimit_put_ucounts(struct ucounts *ucounts, enum rlimit_type type)
> > > do_dec_rlimit_put_ucounts(ucounts, NULL, type);
> > > }
> > >
> > > -long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum rlimit_type type)
> > > +long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum rlimit_type type,
> > > + bool override_rlimit)
> > > {
> > > /* Caller must hold a reference to ucounts */
> > > struct ucounts *iter;
> > > @@ -316,7 +317,7 @@ long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum rlimit_type type)
> > >
> > > for (iter = ucounts; iter; iter = iter->ns->ucounts) {
> > > long new = atomic_long_add_return(1, &iter->rlimit[type]);
> > > - if (new < 0 || new > max)
> > > + if (new < 0 || (!override_rlimit && (new > max)))
> > > goto unwind;
> > > if (iter == ucounts)
> > > ret = new;
> >
> > It's a bad patch. If we do as you suggest, it will
> > do_dec_rlimit_put_ucounts() in case of overflow. This means you'll
> > break the counter and there will be an extra decrement in __sigqueue_free().
> > We can't just ignore the overflow here.
>
> Hm, I don't think my code is changing anything in terms of the overflow handling.
> The (new < 0) handling is exactly the same as it was, the only difference is
> that (new > max) is allowed if override_rlimit is set. But new physically
> can't be larger than LONG_MAX, so there is no actual change if the limit
> is LONG_MAX.
>
> Maybe I'm missing something here, please, clarify.
I re-read your patch one more time. Sorry. Yes, you're right, i am wrong.
You're just allow overlimit.
But one thing confuses me.
Now the maximum rlimits of the upper userns are being forced. Changing
rlimit to RLIM_INFINITY affects only the current userns and child userns.
But after this patch, this is not the case for RLIMIT_SIGPENDING and
within userns it is possible to ignore the restrictions of upper-level
userns which ruins the whole idea.
I agree with Eric. If you don't need upper-level userns limits, you don't
need to set them.
--
Rgrds, legion
