Browse Source

totemsrp: Fix integer overflow in memb_join_sanity

This commit addresses an integer overflow (wraparound) vulnerability
in the check_memb_join_sanity function.

Previously, the 32-bit unsigned network values proc_list_entries and
failed_list_entries were added together before being promoted to
size_t. This allowed the addition to wrap around in 32-bit arithmetic
(e.g., 0x80000000 + 0x80000000 = 0), resulting in a required_len
calculation that was incorrectly small.

The solution is to cast the list entries to size_t and verify that
neither exceeds the maximum allowed value before the addition occurs.

Fixes: CVE-2026-35092

Reported-by: Sebastián Alba Vives (@Sebasteuo / 0xS4bb1) <sebasjosue84@gmail.com>
Signed-off-by: Jan Friesse <jfriesse@redhat.com>
Also-proposed-by: nicholasyang <nicholas.yang@suse.com>
Reviewed-by: Christine Caulfield <ccaulfie@redhat.com>
Jan Friesse 19 hours ago
parent
commit
4082294f50
1 changed files with 11 additions and 1 deletions
  1. 11 1
      exec/totemsrp.c

+ 11 - 1
exec/totemsrp.c

@@ -3786,7 +3786,17 @@ static int check_memb_join_sanity(
 		failed_list_entries = swab32(failed_list_entries);
 		failed_list_entries = swab32(failed_list_entries);
 	}
 	}
 
 
-	required_len = sizeof(struct memb_join) + ((proc_list_entries + failed_list_entries) * sizeof(struct srp_addr));
+	if (proc_list_entries > PROCESSOR_COUNT_MAX ||
+	    failed_list_entries > PROCESSOR_COUNT_MAX) {
+		log_printf (instance->totemsrp_log_level_security,
+		    "Received memb_join message list_entries exceeds the maximum "
+		    "allowed value...  ignoring.");
+
+		return (-1);
+	}
+
+	required_len = sizeof(struct memb_join) +
+	  (((size_t)proc_list_entries + (size_t)failed_list_entries) * sizeof(struct srp_addr));
 	if (msg_len < required_len) {
 	if (msg_len < required_len) {
 		log_printf (instance->totemsrp_log_level_security,
 		log_printf (instance->totemsrp_log_level_security,
 		    "Received memb_join message is too short...  ignoring.");
 		    "Received memb_join message is too short...  ignoring.");