Re: [PATCH v4 3/4] rseq: Make rseq work with protection keys

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

On 2025-02-24 08:20, Dmitry Vyukov wrote:
> If an application registers rseq, and ever switches to another pkey
> protection (such that the rseq becomes inaccessible), then any
> context switch will cause failure in __rseq_handle_notify_resume()
> attempting to read/write struct rseq and/or rseq_cs. Since context
> switches are asynchronous and are outside of the application control
> (not part of the restricted code scope), temporarily switch to
> pkey value that allows access to the 0 (default) PKEY.

This is a good start, but the plan Dave and I discussed went further
than this. Those additions are needed:

1) Add validation at rseq registration that the struct rseq is indeed
    pkey-0 memory (return failure if not).

2) The pkey-0 requirement is only for struct rseq, which we can check
    for at rseq registration, and happens to be the fast path. For struct
    rseq_cs, this is not the same tradeoff: we cannot easily check its
    associated pkey because the rseq_cs pointer is updated by userspace
    when entering a critical section. But the good news is that reading
    the content of struct rseq_cs is *not* a fast-path: it's only done
    when preempting/delivering a signal over a thread which has a
    non-NULL rseq_cs pointer.

    Therefore reading the struct rseq_cs content should be done with
    write_permissive_pkey_val(), giving access to all pkeys.

Thanks,

Mathieu

> 
> Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: "Paul E. McKenney" <paulmck@kernel.org>
> Cc: Boqun Feng <boqun.feng@gmail.com>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Borislav Petkov <bp@alien8.de>
> Cc: Dave Hansen <dave.hansen@linux.intel.com>
> Cc: "H. Peter Anvin" <hpa@zytor.com>
> Cc: Aruna Ramakrishna <aruna.ramakrishna@oracle.com>
> Cc: x86@kernel.org
> Cc: linux-kernel@vger.kernel.org
> Fixes: d7822b1e24f2 ("rseq: Introduce restartable sequences system call")
> 
> ---
> Changes in v4:
>   - Added Fixes tag
> 
> Changes in v3:
>   - simplify control flow to always enable access to 0 pkey
> 
> Changes in v2:
>   - fixed typos and reworded the comment
> ---
>   kernel/rseq.c | 11 +++++++++++
>   1 file changed, 11 insertions(+)
> 
> diff --git a/kernel/rseq.c b/kernel/rseq.c
> index 2cb16091ec0ae..9d9c976d3b78c 100644
> --- a/kernel/rseq.c
> +++ b/kernel/rseq.c
> @@ -10,6 +10,7 @@
>   
>   #include <linux/sched.h>
>   #include <linux/uaccess.h>
> +#include <linux/pkeys.h>
>   #include <linux/syscalls.h>
>   #include <linux/rseq.h>
>   #include <linux/types.h>
> @@ -402,11 +403,19 @@ static int rseq_ip_fixup(struct pt_regs *regs)
>   void __rseq_handle_notify_resume(struct ksignal *ksig, struct pt_regs *regs)
>   {
>   	struct task_struct *t = current;
> +	pkey_reg_t saved_pkey;
>   	int ret, sig;
>   
>   	if (unlikely(t->flags & PF_EXITING))
>   		return;
>   
> +	/*
> +	 * Enable access to the default (0) pkey in case the thread has
> +	 * currently disabled access to it and struct rseq/rseq_cs has
> +	 * 0 pkey assigned (the only supported value for now).
> +	 */
> +	saved_pkey = enable_zero_pkey_val();
> +
>   	/*
>   	 * regs is NULL if and only if the caller is in a syscall path.  Skip
>   	 * fixup and leave rseq_cs as is so that rseq_sycall() will detect and
> @@ -419,9 +428,11 @@ void __rseq_handle_notify_resume(struct ksignal *ksig, struct pt_regs *regs)
>   	}
>   	if (unlikely(rseq_update_cpu_node_id(t)))
>   		goto error;
> +	write_pkey_val(saved_pkey);
>   	return;
>   
>   error:
> +	write_pkey_val(saved_pkey);
>   	sig = ksig ? ksig->sig : 0;
>   	force_sigsegv(sig);
>   }


-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com