The parse_ns_duration() function currently uses prefix matching for
detecting time units. This approach is problematic as it silently
accepts malformed strings such as "100nsx" or "100us_invalid" by
ignoring the trailing characters, leading to potential configuration
errors.

Introduce a match_time_unit() helper that checks the suffix matches
exactly and is followed by either end-of-string or a ':' delimiter.
The ':' is needed because parse_ns_duration() is also called from
get_long_ns_after_colon() when parsing SCHED_DEADLINE priority
specifications in the format "d:runtime:period" (e.g., "d:10ms:100ms").

A plain strcmp() would reject valid deadline strings because the suffix
"ms" is followed by ":100ms", not end-of-string. Similarly,
strncmp_static() would fail because ARRAY_SIZE() includes the NUL
terminator, making it equivalent to strcmp() for this comparison.

The match_time_unit() helper solves both problems: it rejects malformed
input like "100msx" while correctly handling the colon-delimited
deadline format.

Signed-off-by: Wander Lairson Costa <wander@redhat.com>
---
 tools/tracing/rtla/src/utils.c | 23 +++++++++++++++++++----
 1 file changed, 19 insertions(+), 4 deletions(-)

diff --git a/tools/tracing/rtla/src/utils.c b/tools/tracing/rtla/src/utils.c
index 486d96e8290fb..c3ed9089e7260 100644
--- a/tools/tracing/rtla/src/utils.c
+++ b/tools/tracing/rtla/src/utils.c
@@ -200,6 +200,21 @@ long parse_seconds_duration(char *val)
 	return t;
 }
 
+/*
+ * match_time_unit - check if str starts with unit followed by end-of-string or ':'
+ *
+ * This allows the time unit parser to work both in standalone duration strings
+ * like "100ms" and in colon-delimited SCHED_DEADLINE specifications like
+ * "d:10ms:100ms", while still rejecting malformed input like "100msx".
+ */
+static bool match_time_unit(const char *str, const char *unit)
+{
+	size_t len = strlen(unit);
+
+	return strncmp(str, unit, len) == 0 &&
+	       (str[len] == '\0' || str[len] == ':');
+}
+
 /*
  * parse_ns_duration - parse duration with ns/us/ms/s converting it to nanoseconds
  */
@@ -211,15 +226,15 @@ long parse_ns_duration(char *val)
 	t = strtol(val, &end, 10);
 
 	if (end) {
-		if (!strncmp(end, "ns", 2)) {
+		if (match_time_unit(end, "ns")) {
 			return t;
-		} else if (!strncmp(end, "us", 2)) {
+		} else if (match_time_unit(end, "us")) {
 			t *= 1000;
 			return t;
-		} else if (!strncmp(end, "ms", 2)) {
+		} else if (match_time_unit(end, "ms")) {
 			t *= 1000 * 1000;
 			return t;
-		} else if (!strncmp(end, "s", 1)) {
+		} else if (match_time_unit(end, "s")) {
 			t *= 1000 * 1000 * 1000;
 			return t;
 		}
-- 
2.53.0

po 9. 3. 2026 v 20:58 odesílatel Wander Lairson Costa
<wander@redhat.com> napsal:
>
> The parse_ns_duration() function currently uses prefix matching for
> detecting time units. This approach is problematic as it silently
> accepts malformed strings such as "100nsx" or "100us_invalid" by
> ignoring the trailing characters, leading to potential configuration
> errors.
>
> Introduce a match_time_unit() helper that checks the suffix matches
> exactly and is followed by either end-of-string or a ':' delimiter.
> The ':' is needed because parse_ns_duration() is also called from
> get_long_ns_after_colon() when parsing SCHED_DEADLINE priority
> specifications in the format "d:runtime:period" (e.g., "d:10ms:100ms").
>
> A plain strcmp() would reject valid deadline strings because the suffix
> "ms" is followed by ":100ms", not end-of-string. Similarly,
> strncmp_static() would fail because ARRAY_SIZE() includes the NUL
> terminator, making it equivalent to strcmp() for this comparison.
>

This fixes both the command and the corresponding unit test, thanks!

> The match_time_unit() helper solves both problems: it rejects malformed
> input like "100msx" while correctly handling the colon-delimited
> deadline format.
>

It now fails to reject this kind of input:

$  rtla timerlat -P d:10ms:100ms:somethingsomething
// this is also parsed as valid

But that can be fixed in a future patchset, perhaps together with also migrating
parse_seconds_duration() from prefix matching to match_time_unit() as well.

That would prevent the same imprecise parsing issue for -d:

$ rtla timerlat -d 1somethingsomething
// this is also parsed as valid

Tomas