Re: [PATCH] tracing/histograms: Reject non-stacktrace fields for synthetic stacktrace fields

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] tracing: check_synth_field regression for __bitmask and stacktraces
- [High] tracing: arbitrary memory read via scalar to string synthetic field mapping
--

commit 4813d717e995df13871068dd6fd996711ca8965d
Author: Lee Jones <lee@kernel.org>

tracing/histograms: Reject non-stacktrace fields for synthetic stacktrace fields

In check_synth_field(), when a synthetic event field is a stacktrace
and the histogram field is not a "long[" stacktrace array, validation falls
through to the generic size and signedness comparison.  This allows any
8-byte signed scalar histogram field to pass the fallback check and be
dereferenced as a kernel stacktrace pointer.  This commit attempts to reject
these mismatches by enforcing that field->is_stack matches the "long[" type.

> diff --git a/kernel/trace/trace_events_hist.c b/kernel/trace/trace_events_hist.c
> index 8af97fd4ee2d5..b09f5043e1543 100644
> --- a/kernel/trace/trace_events_hist.c
> +++ b/kernel/trace/trace_events_hist.c
> @@ -3946,7 +3946,10 @@ static int check_synth_field(struct synth_event *event,
>  	    && field->is_dynamic)
>  		return 0;
>  
> -	if (strstr(hist_field->type, "long[") && field->is_stack)
> +	if (!!strstr(hist_field->type, "long[") != field->is_stack)
> +		return -EINVAL;

[Severity: High]
Does this heuristic falsely match the long[ substring in __data_loc unsigned
long[] from __bitmask and __cpumask fields?

If is_stack == false, this logic seems to falsely reject valid mappings.

If is_stack == true, this appears to bypass validation and allow a 32-bit
__data_loc integer to be passed as the field value. Could this 32-bit integer
then be cast to a pointer and dereferenced in
trace_events_synth.c:get_field_size() and trace_stack(), potentially causing
a kernel panic?

> +
> +	if (field->is_stack)
>  		return 0;
>  
>  	if (strcmp(field->type, hist_field->type) != 0) {
>  		if (field->size != hist_field->size ||
>  		    (!field->is_string && field->is_signed != hist_field->is_signed))
>  			return -EINVAL;
>  	}

[Severity: High]
It looks like this fallback logic fails to reject a scalar-to-string mapping.

When creating a dynamic string synthetic field, its size is set to sizeof(u64)
and is_string is true.  Because field->size equals hist_field->size (8 bytes)
and !field->is_string short-circuits the type/signedness check, will an 8-byte
scalar successfully map to a dynamic string?

If an attacker-controlled scalar is passed, can it be cast to a kernel pointer
in trace_events_synth.c:trace_string() and used with fetch_store_strlen() to
arbitrarily read kernel memory?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261012155805.1103972-1-lee@kernel.org?part=1