Set PCRE2 limits explicitly (to more sensible defaults), reported by Link420.

doc/RELEASE-NOTES.md

@@ -7,15 +7,21 @@ This is work in progress and may not always be a stable version.
 ### Enhancements:
 
 ### Changes:
+* Spamfilter regexes now use more sensible defaults in terms of "max effort",
+  similar to what PHP has been using for years. This means very slow regexes
+  will now raise a `SPAMFILTER_REGEX_ERROR` warning during execution if
+  this happens (should be extremely rare).
 
 ### Fixes:
-* Harden the built-in HTTPS client
+* Hardening of the built-in HTTPS client
 
 ### Developers and protocol:
 * URL API: The OutgoingWebRequest `max_size` (introduced last release) now
   also caps file-backed downloads. Default for file-backed when left at 0
   is 50MB (`DOWNLOAD_MAX_SIZE_FILE_BACKED`). For memory-backed, it stays
   at 1MB like in 6.2.5 (`DOWNLOAD_MAX_SIZE_MEMORY_BACKED`).
+* The `unreal_match()` function now has a 3rd argument `const char **error`
+  for communicating regex errors back. Just set to `NULL` if you don't care.
 * If you do something to a user that would (potentially) move the user from
   `unknown-users` to `known-users` (or vice versa) then you should call
   `update_known_user_cache(client);` to update the known users cache.

include/config.h

@@ -238,6 +238,16 @@
 #define SPAMFILTER_DETECTSLOW
 #endif
 
+/* Limits for PCRE2 regex matching (eg. spamfilter, badwords). A regex that
+ * exceeds these is aborted and treated as no match, instead of running
+ * unbounded. The match limit is honoured by JIT. The depth limit only applies
+ * to the non-JIT interpreter, since PCRE2 ignores it under JIT.
+ * We use the same defaults that PHP has been using for a long time (which is
+ * actually 10 times lower than PCRE2 defaults, as of 2026).
+ */
+#define UNREAL_PCRE2_MATCH_LIMIT	1000000
+#define UNREAL_PCRE2_DEPTH_LIMIT	100000
+
 /* Maximum number of ModData objects that may be attached to an object */
 /* UnrealIRCd 4.0.0:   8,     8, 4, 4
  * UnrealIRCd 4.0.14: 12,     8, 4, 4

include/h.h

@@ -1091,9 +1091,11 @@ extern void read_packet(int fd, int revents, void *data);
 extern int process_packet(Client *cptr, char *readbuf, int length, int killsafely);
 extern int parse_chanmode(ParseMode *pm, const char *modebuf_in, const char *parabuf_in);
 extern int dead_socket(Client *to, const char *notice);
+extern MODVAR pcre2_match_context *unreal_pcre2_match_ctx;
+extern void init_match(void);
 extern Match *unreal_create_match(MatchType type, const char *str, char **error);
 extern void unreal_delete_match(Match *m);
-extern int unreal_match(Match *m, const char *str);
+extern int unreal_match(Match *m, const char *str, const char **error);
 extern int unreal_match_method_strtoval(const char *str);
 extern char *unreal_match_method_valtostr(int val);
 #ifdef _WIN32

src/aliases.c

@@ -143,7 +143,7 @@ void cmd_alias(ClientContext *clictx, Client *client, MessageTag *mtags, int par
 
 		for (format = alias->format; format; format = format->next)
 		{
-			if (unreal_match(format->expr, ptr))
+			if (unreal_match(format->expr, ptr, NULL))
 			{
 				/* Parse the parameters */
 				int i = 0, j = 0, k = 1;

src/conf.c

@@ -1906,6 +1906,7 @@ void config_setdefaultsettings(Configuration *i)
 	add_log_throttle_config(&i->log_throttle, "BUG_CT_BUCKET_MISSING", 5, 60, 0);
 	add_log_throttle_config(&i->log_throttle, "BUG_CT_NEGATIVE_COUNTER", 5, 60, 0);
 	add_log_throttle_config(&i->log_throttle, "BUG_DECREASE_IPUSERS_BUCKET", 5, 60, 0);
+	add_log_throttle_config(&i->log_throttle, "SPAMFILTER_REGEX_ERROR", 5, 60, 0);
 
 	/* TLS options */
 	i->tls_options = safe_alloc(sizeof(TLSOptions));

src/ircd.c

@@ -547,6 +547,7 @@ int InitUnrealIRCd(int argc, char *argv[])
 
 	init_hash();
 	log_throttle_init();
+	init_match();
 
 	SetupEvents();
 

src/match.c

@@ -35,6 +35,8 @@ u_char touppertab[], tolowertab[];
 #define tolowertab2 tolowertab
 #define lc(x) tolowertab2[x]
 
+pcre2_match_context *unreal_pcre2_match_ctx = NULL;
+
 /* Match routine for special cases where escaping is needed in a normal fashion.
  * Checks a string ('name') against a globbing(+more) pattern ('mask').
  * Original by Douglas A Lewis (dalewis@acsu.buffalo.edu).
@@ -369,6 +371,17 @@ u_char char_atribs[] = {
 /* f0-ff */ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
 };
 
+/* Set up global match state. Called once at startup. */
+void init_match(void)
+{
+	unreal_pcre2_match_ctx = pcre2_match_context_create(NULL);
+	if (unreal_pcre2_match_ctx)
+	{
+		pcre2_set_match_limit(unreal_pcre2_match_ctx, UNREAL_PCRE2_MATCH_LIMIT);
+		pcre2_set_depth_limit(unreal_pcre2_match_ctx, UNREAL_PCRE2_DEPTH_LIMIT);
+	}
+}
+
 /** Free up all resources of an Match entry (including the struct itself).
  * NOTE: this function may (also) be called for Match structs that have only been
  *       setup half-way, so use special care when accessing members (NULL checks!)
@@ -441,28 +454,46 @@ Match *unreal_create_match(MatchType type, const char *str, char **error)
 }
 
 /** Try to match an Match entry ('m') against a string ('str').
+ * @param error  If non-NULL, set to an error string when the regex could not
+ *               complete (eg. a resource limit was hit), or to NULL otherwise.
+ *               Points to a static buffer, valid until the next unreal_match().
  * @returns 1 if matched, 0 if not.
  * @note These (more logical) return values are opposite to the match_simple() function.
  */
-int unreal_match(Match *m, const char *str)
+int unreal_match(Match *m, const char *str, const char **error)
 {
+	static char errbuf[256];
+
+	if (error)
+		*error = NULL;
+
 	if (m->type == MATCH_SIMPLE)
 	{
 		if (match_simple(m->str, str))
 			return 1;
 		return 0;
 	}
-	
+
 	if (m->type == MATCH_PCRE_REGEX)
 	{
 		pcre2_match_data *md = pcre2_match_data_create(9, NULL);
 		int ret;
-		
+
-		ret = pcre2_match(m->ext.pcre2_expr, str, PCRE2_ZERO_TERMINATED, 0, 0, md, NULL); /* run the regex */
+		ret = pcre2_match(m->ext.pcre2_expr, str, PCRE2_ZERO_TERMINATED, 0, 0, md, unreal_pcre2_match_ctx); /* run the regex */
 		pcre2_match_data_free(md); /* yeah, we never use it. unfortunately argument must be non-NULL for pcre2_match() */
-		
+
 		if (ret > 0)
-			return 1; /* MATCH */		
+			return 1; /* MATCH */
+
+		if (error && (ret < 0) && (ret != PCRE2_ERROR_NOMATCH) && (ret != PCRE2_ERROR_PARTIAL))
+		{
+			/* Regex did not finish (eg. hit the match, depth or JIT stack limit).
+			 * Report it so the caller can warn. We still return no-match.
+			 */
+			*errbuf = '\0';
+			pcre2_get_error_message(ret, errbuf, sizeof(errbuf));
+			*error = errbuf;
+		}
 		return 0; /* NO MATCH */
 	}
 
@@ -681,7 +712,7 @@ const char *stripbadwords(const char *str, ConfigItem_badword *start_bw, int *bl
 				pcre2_match_data *md = pcre2_match_data_create(9, NULL);
 				int ret;
 
-				ret = pcre2_match(this_word->pcre2_expr, cleanstr, PCRE2_ZERO_TERMINATED, 0, 0, md, NULL); /* run the regex */
+				ret = pcre2_match(this_word->pcre2_expr, cleanstr, PCRE2_ZERO_TERMINATED, 0, 0, md, unreal_pcre2_match_ctx); /* run the regex */
 				pcre2_match_data_free(md); /* yeah, we never use it. unfortunately argument must be non-NULL for pcre2_match() */
 				if (ret > 0)
 				{
@@ -702,7 +733,7 @@ const char *stripbadwords(const char *str, ConfigItem_badword *start_bw, int *bl
 					/* ^^ we need to free 'md' in ALL circumstances.
 					 * remember this if you break or continue in this loop!
 					 */
-					ret = pcre2_match(this_word->pcre2_expr, ptr, PCRE2_ZERO_TERMINATED, 0, 0, md, NULL); /* run the regex */
+					ret = pcre2_match(this_word->pcre2_expr, ptr, PCRE2_ZERO_TERMINATED, 0, 0, md, unreal_pcre2_match_ctx); /* run the regex */
 					if (ret > 0)
 					{
 						dd = pcre2_get_ovector_pointer(md);

src/modules/tkl.c

@@ -1281,6 +1281,29 @@ char *spamfilter_id(TKL *tk)
 	return buf;
 }
 
+/* Warn opers when a spamfilter regex could not finish (eg. it hit the
+ * PCRE2 match or depth limit). The match is treated as no-match, so we
+ * only warn and do not remove the spamfilter.
+ */
+static void spamfilter_regex_error(TKL *tkl, const char *regex_error)
+{
+	if (tkl->type & TKL_GLOBAL)
+	{
+		unreal_log(ULOG_WARNING, "tkl", "SPAMFILTER_REGEX_ERROR", NULL,
+		           "[Spamfilter] Regex aborted ($regex_error) for '$tkl'. Possibly too complex regex? "
+		           "To delete, use: /SPAMFILTER del $spamfilter_id",
+		           log_data_string("regex_error", regex_error),
+		           log_data_string("spamfilter_id", spamfilter_id(tkl)),
+		           log_data_tkl("tkl", tkl));
+	} else {
+		unreal_log(ULOG_WARNING, "tkl", "SPAMFILTER_REGEX_ERROR", NULL,
+		           "[Spamfilter] Regex aborted ($regex_error) for '$tkl'. Possibly too complex regex? "
+		           "To remove it, edit your config file",
+		           log_data_string("regex_error", regex_error),
+		           log_data_tkl("tkl", tkl));
+	}
+}
+
 int tkl_ip_change(Client *client, const char *oldip)
 {
 	TKL *tkl;
@@ -3774,9 +3797,15 @@ int spamfilter_check_users(TKL *tkl)
 	{
 		if (MyUser(client))
 		{
+			const char *regex_error = NULL;
+
 			spamfilter_build_user_string(spamfilter_user, client->name, client);
-			if (!unreal_match(tkl->ptr.spamfilter->match, spamfilter_user))
+			if (!unreal_match(tkl->ptr.spamfilter->match, spamfilter_user, &regex_error))
+			{
+				if (regex_error)
+					spamfilter_regex_error(tkl, regex_error);
 				continue; /* No match */
+			}
 
 			/* matched! */
 			unreal_log(ULOG_INFO, "tkl", "SPAMFILTER_MATCH", client,
@@ -5589,6 +5618,8 @@ int _match_spamfilter(Client *client, const char *str_in, int target, const char
 
 		if (tkl->ptr.spamfilter->match && (tkl->ptr.spamfilter->match->type != MATCH_NONE))
 		{
+			const char *regex_error = NULL;
+
 #ifdef SPAMFILTER_DETECTSLOW
 			if (tkl->ptr.spamfilter->match->type == MATCH_PCRE_REGEX)
 			{
@@ -5600,11 +5631,11 @@ int _match_spamfilter(Client *client, const char *str_in, int target, const char
 #endif
 
 			if (tkl->ptr.spamfilter->input_conversion == INPUT_CONVERSION_STRIP_CONTROL_CODES)
-				ret = unreal_match(tkl->ptr.spamfilter->match, str); /* StripControlCodes() */
+				ret = unreal_match(tkl->ptr.spamfilter->match, str, &regex_error); /* StripControlCodes() */
 			else if (tkl->ptr.spamfilter->input_conversion == INPUT_CONVERSION_CONFUSABLES)
-				ret = unreal_match(tkl->ptr.spamfilter->match, str_deconfused ? str_deconfused : str); /* utf8_convert_confusables(), with fallback */
+				ret = unreal_match(tkl->ptr.spamfilter->match, str_deconfused ? str_deconfused : str, &regex_error); /* utf8_convert_confusables(), with fallback */
 			else
-				ret = unreal_match(tkl->ptr.spamfilter->match, str_in); /* raw */
+				ret = unreal_match(tkl->ptr.spamfilter->match, str_in, &regex_error); /* raw */
 
 #ifdef SPAMFILTER_DETECTSLOW
 			if (tkl->ptr.spamfilter->match->type == MATCH_PCRE_REGEX)
@@ -5633,6 +5664,9 @@ int _match_spamfilter(Client *client, const char *str_in, int target, const char
 				}
 			}
 #endif
+
+			if (regex_error)
+				spamfilter_regex_error(tkl, regex_error);
 		} else {
 			/* There is no ::match but there was a ::rule, and that is enough for a match.. */
 			if (tkl->ptr.spamfilter->rule)