linux-kernel - Re: [PATCH] tracing: fix va_list breakage in trace_check

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20211117183827.4989cfab@gandalf.local.home>
Date:   Wed, 17 Nov 2021 18:38:27 -0500
From:   Steven Rostedt <rostedt@...dmis.org>
To:     Nikita Yushchenko <nikita.yushchenko@...tuozzo.com>
Cc:     Ingo Molnar <mingo@...hat.com>, linux-kernel@...r.kernel.org,
        kernel@...nvz.org
Subject: Re: [PATCH] tracing: fix va_list breakage in trace_check_vprintf()

On Wed, 17 Nov 2021 21:37:20 +0300
Nikita Yushchenko <nikita.yushchenko@...tuozzo.com> wrote:

> When trace_check_vprintf() extracts parts of the format string and
> passes those to trace_seq_printf() together with va_list, it expects
> that trace_seq_printf() consumes arguments from va_list, as defined
> by the passed format substring.
> 
> However, trace_seq_printf() has a special path for overflow handling,
> that does not consume any arguments from va_list. This causes va_list
> to get out of sync with format string processing, the next va_arg()
> inside trace_check_vprintf() gets wrong argument, and WARN_ON_ONCE()
> hits.
> 
> This situation easily triggers by ftrace_stress_test from LTP.
> 
> Fix that by adding a dummy vsnprintf() call to the overflow path inside
> trace_seq_printf() to ensure args from va_list are still consumed.

Hi Nikita,

> 
> Signed-off-by: Nikita Yushchenko <nikita.yushchenko@...tuozzo.com>
> ---
>  kernel/trace/trace_seq.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/trace/trace_seq.c b/kernel/trace/trace_seq.c
> index 9c90b3a7dce2..3551b5e18aa2 100644
> --- a/kernel/trace/trace_seq.c
> +++ b/kernel/trace/trace_seq.c
> @@ -141,9 +141,15 @@ EXPORT_SYMBOL_GPL(trace_seq_bitmask);
>  void trace_seq_vprintf(struct trace_seq *s, const char *fmt, va_list args)
>  {
>  	unsigned int save_len = s->seq.len;
> +	char buf[1];
>  
> -	if (s->full)
> +	if (s->full) {
> +		/* Consume args from va_list before returning, some callers
> +		 * expect that.

First, only the networking code uses the /* comment ... for multiline
comments. The rest of the kernel uses:

		/*
		 * Consume ...
		 */

format for multi line comments.

But regardless. Consumers do not expect va_list to be consumed if it is
full. The one use case that does is buggy.

> +		 */
> +		vsnprintf(buf, sizeof(buf), fmt, args);
>  		return;
> +	}
>  
>  	__trace_seq_init(s);
>  

The real fix is:

diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index f9139dc1262c..7aa5ea5ca912 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -3654,6 +3654,10 @@ static bool trace_safe_str(struct trace_iterator *iter, const char *str)
 	struct trace_event *trace_event;
 	struct trace_event_call *event;
 
+	/* if seq is full, then we can't test it */
+	if (iter->seq->full)
+		return true;
+
 	/* OK if part of the event data */
 	if ((addr >= (unsigned long)iter->ent) &&
 	    (addr < (unsigned long)iter->ent + iter->ent_size))


Cheers,

-- Steve