[PATCH v3 12/13] tracing: Check for printable characters when printing field dyn strings

Steven Rostedt posted 13 patches 2 months ago
There is a newer version of this series
[PATCH v3 12/13] tracing: Check for printable characters when printing field dyn strings
Posted by Steven Rostedt 2 months ago
From: Steven Rostedt <rostedt@goodmis.org>

When the "fields" option is enabled, it prints each trace event field
based on its type. But a dynamic array and a dynamic string can both have
a "char *" type. Printing it as a string can cause escape characters to be
printed and mess up the output of the trace.

For dynamic strings, test if there are any non-printable characters, and
if so, print both the string with the non printable characters as '.', and
the print the hex value of the array.

Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
---
 kernel/trace/trace_output.c | 27 +++++++++++++++++++++++++--
 1 file changed, 25 insertions(+), 2 deletions(-)

diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c
index 97db0b0ccf3e..718b255b6fd8 100644
--- a/kernel/trace/trace_output.c
+++ b/kernel/trace/trace_output.c
@@ -950,7 +950,9 @@ static void print_fields(struct trace_iterator *iter, struct trace_event_call *c
 	int offset;
 	int len;
 	int ret;
+	int i;
 	void *pos;
+	char *str;
 
 	list_for_each_entry_reverse(field, head, link) {
 		trace_seq_printf(&iter->seq, " %s=", field->name);
@@ -977,8 +979,29 @@ static void print_fields(struct trace_iterator *iter, struct trace_event_call *c
 				trace_seq_puts(&iter->seq, "<OVERFLOW>");
 				break;
 			}
-			pos = (void *)iter->ent + offset;
-			trace_seq_printf(&iter->seq, "%.*s", len, (char *)pos);
+			str = (char *)iter->ent + offset;
+			/* Check if there's any non printable strings */
+			for (i = 0; i < len; i++) {
+				if (str[i] && !(isascii(str[i]) && isprint(str[i])))
+					break;
+			}
+			if (i < len) {
+				for (i = 0; i < len; i++) {
+					if (isascii(str[i]) && isprint(str[i]))
+						trace_seq_putc(&iter->seq, str[i]);
+					else
+						trace_seq_putc(&iter->seq, '.');
+				}
+				trace_seq_puts(&iter->seq, " (");
+				for (i = 0; i < len; i++) {
+					if (i)
+						trace_seq_putc(&iter->seq, ':');
+					trace_seq_printf(&iter->seq, "%02x", str[i]);
+				}
+				trace_seq_putc(&iter->seq, ')');
+			} else {
+				trace_seq_printf(&iter->seq, "%.*s", len, str);
+			}
 			break;
 		case FILTER_PTR_STRING:
 			if (!iter->fmt_size)
-- 
2.51.0
Re: [PATCH v3 12/13] tracing: Check for printable characters when printing field dyn strings
Posted by Douglas Raillard 2 months ago
Hi Steve,

On 15-10-2025 18:32, Steven Rostedt wrote:
> From: Steven Rostedt <rostedt@goodmis.org>
> 
> When the "fields" option is enabled, it prints each trace event field
> based on its type. But a dynamic array and a dynamic string can both have
> a "char *" type. Printing it as a string can cause escape characters to be
> printed and mess up the output of the trace.

We faced the same issue when converting trace.dat to other formats that distinguish
between a byte buffer and an actual string. The current solution we have is to
reserve "char []" for actual string and use "u8 []"/"uint8_t []" for byte buffers.

Obviously it does not solve the problem for existing events but that could be worth
establishing a convention like that for new code and support it in trace_output.c ?
Most uses of trace data beyond direct printing would require knowing the data schema
from the header info, so value-based criteria are typically not suitable for that.

> For dynamic strings, test if there are any non-printable characters, and
> if so, print both the string with the non printable characters as '.', and
> the print the hex value of the array.
> 
> Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
> ---
>   kernel/trace/trace_output.c | 27 +++++++++++++++++++++++++--
>   1 file changed, 25 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c
> index 97db0b0ccf3e..718b255b6fd8 100644
> --- a/kernel/trace/trace_output.c
> +++ b/kernel/trace/trace_output.c
> @@ -950,7 +950,9 @@ static void print_fields(struct trace_iterator *iter, struct trace_event_call *c
>   	int offset;
>   	int len;
>   	int ret;
> +	int i;
>   	void *pos;
> +	char *str;
>   
>   	list_for_each_entry_reverse(field, head, link) {
>   		trace_seq_printf(&iter->seq, " %s=", field->name);
> @@ -977,8 +979,29 @@ static void print_fields(struct trace_iterator *iter, struct trace_event_call *c
>   				trace_seq_puts(&iter->seq, "<OVERFLOW>");
>   				break;
>   			}
> -			pos = (void *)iter->ent + offset;
> -			trace_seq_printf(&iter->seq, "%.*s", len, (char *)pos);
> +			str = (char *)iter->ent + offset;
> +			/* Check if there's any non printable strings */
> +			for (i = 0; i < len; i++) {
> +				if (str[i] && !(isascii(str[i]) && isprint(str[i])))
> +					break;
> +			}
> +			if (i < len) {
> +				for (i = 0; i < len; i++) {
> +					if (isascii(str[i]) && isprint(str[i]))
> +						trace_seq_putc(&iter->seq, str[i]);
> +					else
> +						trace_seq_putc(&iter->seq, '.');
> +				}
> +				trace_seq_puts(&iter->seq, " (");
> +				for (i = 0; i < len; i++) {
> +					if (i)
> +						trace_seq_putc(&iter->seq, ':');
> +					trace_seq_printf(&iter->seq, "%02x", str[i]);
> +				}
> +				trace_seq_putc(&iter->seq, ')');
> +			} else {
> +				trace_seq_printf(&iter->seq, "%.*s", len, str);
> +			}
>   			break;
>   		case FILTER_PTR_STRING:
>   			if (!iter->fmt_size)

--

Douglas
Re: [PATCH v3 12/13] tracing: Check for printable characters when printing field dyn strings
Posted by Steven Rostedt 2 months ago
On Mon, 20 Oct 2025 13:19:46 +0100
Douglas Raillard <douglas.raillard@arm.com> wrote:

> Hi Steve,
> 
> On 15-10-2025 18:32, Steven Rostedt wrote:
> > From: Steven Rostedt <rostedt@goodmis.org>
> > 
> > When the "fields" option is enabled, it prints each trace event field
> > based on its type. But a dynamic array and a dynamic string can both have
> > a "char *" type. Printing it as a string can cause escape characters to be
> > printed and mess up the output of the trace.  
> 
> We faced the same issue when converting trace.dat to other formats that distinguish
> between a byte buffer and an actual string. The current solution we have is to
> reserve "char []" for actual string and use "u8 []"/"uint8_t []" for byte buffers.
> 
> Obviously it does not solve the problem for existing events but that could be worth
> establishing a convention like that for new code and support it in trace_output.c ?
> Most uses of trace data beyond direct printing would require knowing the data schema
> from the header info, so value-based criteria are typically not suitable for that.

Well, what the types are is not quite controlled by me, but by whoever
writes a trace event. Thus, it's not going to be something I can control in trace_output.c

-- Steve