Merge branch 'PHP-7.3' into PHP-7.4

* PHP-7.3:
  Fixed bug #79188

NEWS

@@ -1,5 +1,12 @@
 PHP                                                                        NEWS
 |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
+
+?? ??? ????, PHP 7.4.4
+
+- PCRE:
+  . Fixed bug #79188 (Memory corruption in preg_replace/preg_replace_callback
+    and unicode). (Nikita)
+
 ?? ??? ????, PHP 7.4.3
 
 - Core:

ext/pcre/php_pcre.c

@@ -1582,6 +1582,7 @@ PHPAPI zend_string *php_pcre_replace_impl(pcre_cache_entry *pce, zend_string *su
 	size_t			 match_len;			/* Length of the current match */
 	int				 backref;			/* Backreference number */
 	PCRE2_SIZE		 start_offset;		/* Where the new search starts */
+	size_t			 last_end_offset;	/* Where the last search ended */
 	char			*walkbuf,			/* Location of current replacement in the result */
 					*walk,				/* Used to walk the replacement string */
 					*match,				/* The current match */
@@ -1600,6 +1601,7 @@ PHPAPI zend_string *php_pcre_replace_impl(pcre_cache_entry *pce, zend_string *su
 	/* Initialize */
 	match = NULL;
 	start_offset = 0;
+	last_end_offset = 0;
 	result_len = 0;
 	PCRE_G(error_code) = PHP_PCRE_NO_ERROR;
 
@@ -1626,7 +1628,7 @@ PHPAPI zend_string *php_pcre_replace_impl(pcre_cache_entry *pce, zend_string *su
 			options, match_data, mctx);
 
 	while (1) {
-		piece = subject + start_offset;
+		piece = subject + last_end_offset;
 
 		if (count >= 0 && limit > 0) {
 			zend_bool simple_string;
@@ -1656,7 +1658,7 @@ matched:
 			/* Set the match location in subject */
 			match = subject + offsets[0];
 
-			new_len = result_len + offsets[0] - start_offset; /* part before the match */
+			new_len = result_len + offsets[0] - last_end_offset; /* part before the match */
 
 			walk = ZSTR_VAL(replace_str);
 			replace_end = walk + ZSTR_LEN(replace_str);
@@ -1733,7 +1735,7 @@ matched:
 			limit--;
 
 			/* Advance to the next piece. */
-			start_offset = offsets[1];
+			start_offset = last_end_offset = offsets[1];
 
 			/* If we have matched an empty string, mimic what Perl's /g options does.
 			   This turns out to be rather cunning. First we set PCRE2_NOTEMPTY_ATSTART and try
@@ -1753,10 +1755,7 @@ matched:
 					   to achieve this, unless we're already at the end of the string. */
 					if (start_offset < subject_len) {
 						size_t unit_len = calculate_unit_length(pce, piece);
-
 						start_offset += unit_len;
-						memcpy(ZSTR_VAL(result) + result_len, piece, unit_len);
-						result_len += unit_len;
 					} else {
 						goto not_matched;
 					}
@@ -1771,7 +1770,7 @@ not_matched:
 				result = zend_string_copy(subject_str);
 				break;
 			}
-			new_len = result_len + subject_len - start_offset;
+			new_len = result_len + subject_len - last_end_offset;
 			if (new_len >= alloc_len) {
 				alloc_len = new_len; /* now we know exactly how long it is */
 				if (NULL != result) {
@@ -1781,8 +1780,8 @@ not_matched:
 				}
 			}
 			/* stick that last bit of string on our output */
-			memcpy(ZSTR_VAL(result) + result_len, piece, subject_len - start_offset);
-			result_len += subject_len - start_offset;
+			memcpy(ZSTR_VAL(result) + result_len, piece, subject_len - last_end_offset);
+			result_len += subject_len - last_end_offset;
 			ZSTR_VAL(result)[result_len] = '\0';
 			ZSTR_LEN(result) = result_len;
 			break;
@@ -1824,6 +1823,7 @@ static zend_string *php_pcre_replace_func_impl(pcre_cache_entry *pce, zend_strin
 	size_t			 new_len;			/* Length of needed storage */
 	size_t			 alloc_len;			/* Actual allocated length */
 	PCRE2_SIZE		 start_offset;		/* Where the new search starts */
+	size_t			 last_end_offset;	/* Where the last search ended */
 	char			*match,				/* The current match */
 					*piece;				/* The current piece of subject */
 	size_t			result_len; 		/* Length of result */
@@ -1853,6 +1853,7 @@ static zend_string *php_pcre_replace_func_impl(pcre_cache_entry *pce, zend_strin
 	/* Initialize */
 	match = NULL;
 	start_offset = 0;
+	last_end_offset = 0;
 	result_len = 0;
 	PCRE_G(error_code) = PHP_PCRE_NO_ERROR;
 
@@ -1885,7 +1886,7 @@ static zend_string *php_pcre_replace_func_impl(pcre_cache_entry *pce, zend_strin
 			options, match_data, mctx);
 
 	while (1) {
-		piece = subject + start_offset;
+		piece = subject + last_end_offset;
 
 		if (count >= 0 && limit) {
 			/* Check for too many substrings condition. */
@@ -1913,7 +1914,7 @@ matched:
 			/* Set the match location in subject */
 			match = subject + offsets[0];
 
-			new_len = result_len + offsets[0] - start_offset; /* part before the match */
+			new_len = result_len + offsets[0] - last_end_offset; /* part before the match */
 
 			/* Use custom function to get replacement string and its length. */
 			eval_result = preg_do_repl_func(
@@ -1945,7 +1946,7 @@ matched:
 			limit--;
 
 			/* Advance to the next piece. */
-			start_offset = offsets[1];
+			start_offset = last_end_offset = offsets[1];
 
 			/* If we have matched an empty string, mimic what Perl's /g options does.
 			   This turns out to be rather cunning. First we set PCRE2_NOTEMPTY_ATSTART and try
@@ -1965,10 +1966,7 @@ matched:
 					   to achieve this, unless we're already at the end of the string. */
 					if (start_offset < subject_len) {
 						size_t unit_len = calculate_unit_length(pce, piece);
-
 						start_offset += unit_len;
-						memcpy(ZSTR_VAL(result) + result_len, piece, unit_len);
-						result_len += unit_len;
 					} else {
 						goto not_matched;
 					}
@@ -1983,7 +1981,7 @@ not_matched:
 				result = zend_string_copy(subject_str);
 				break;
 			}
-			new_len = result_len + subject_len - start_offset;
+			new_len = result_len + subject_len - last_end_offset;
 			if (new_len >= alloc_len) {
 				alloc_len = new_len; /* now we know exactly how long it is */
 				if (NULL != result) {
@@ -1993,8 +1991,8 @@ not_matched:
 				}
 			}
 			/* stick that last bit of string on our output */
-			memcpy(ZSTR_VAL(result) + result_len, piece, subject_len - start_offset);
-			result_len += subject_len - start_offset;
+			memcpy(ZSTR_VAL(result) + result_len, piece, subject_len - last_end_offset);
+			result_len += subject_len - last_end_offset;
 			ZSTR_VAL(result)[result_len] = '\0';
 			ZSTR_LEN(result) = result_len;
 			break;

ext/pcre/tests/bug79188.phpt

@@ -0,0 +1,13 @@
+--TEST--
+Bug #79188: Memory corruption in preg_replace/preg_replace_callback and unicode
+--FILE--
+<?php
+
+var_dump(preg_replace("//u", "", "a" . str_repeat("\u{1f612}", 10)));
+var_dump(preg_replace_callback(
+    "//u", function() { return ""; }, "a" . str_repeat("\u{1f612}", 10)));
+
+?>
+--EXPECT--
+string(41) "a😒😒😒😒😒😒😒😒😒😒"
+string(41) "a😒😒😒😒😒😒😒😒😒😒"