[Pcre-svn] [659] code/trunk: Reduce recursion limit for aut…

          http://www.exim.org/viewvc/pcre2?view=rev&revision=659
Author:   ph10
Date:     2017-02-08 17:03:30 +0000 (Wed, 08 Feb 2017)
Log Message:
-----------
Reduce recursion limit for auto-possessify to reduce stack usage at compile 
time. Fixes oss-fuzz issue 553.

Modified Paths:
--------------
    code/trunk/ChangeLog
    code/trunk/src/pcre2_auto_possess.c
    code/trunk/testdata/testinput1
    code/trunk/testdata/testoutput1

Modified: code/trunk/ChangeLog
===================================================================
--- code/trunk/ChangeLog    2017-02-04 16:27:20 UTC (rev 658)
+++ code/trunk/ChangeLog    2017-02-08 17:03:30 UTC (rev 659)
@@ -338,6 +338,10 @@
     * LC_ALL was displayed as "LCC_ALL";
     * numbers 11, 12 & 13 should end in "th";
     * use double quotes in usage message.
+    
+53. When autopossessifying, skip empty branches without recursion, to reduce 
+stack usage for the benefit of clang with -fsanitize-address, which uses huge 
+stack frames. Example pattern: /X?(R||){3335}/. Fixes oss-fuzz issue 553. 

Modified: code/trunk/src/pcre2_auto_possess.c
===================================================================
--- code/trunk/src/pcre2_auto_possess.c    2017-02-04 16:27:20 UTC (rev 658)
+++ code/trunk/src/pcre2_auto_possess.c    2017-02-08 17:03:30 UTC (rev 659)
@@ -589,6 +589,7 @@
       case OP_ASSERTBACK_NOT:
       case OP_ONCE:
       case OP_ONCE_NC:
+
       /* Atomic sub-patterns and assertions can always auto-possessify their
       last iterator. However, if the group was entered as a result of checking
       a previous iterator, this is not possible. */
@@ -606,6 +607,9 @@
     next_code = code + GET(code, 1);
     code += PRIV(OP_lengths)[c];

+    /* Check each branch. We have to recurse a level for all but the last
+    branch. */
+
     while (*next_code == OP_ALT)
       {
       if (!compare_opcodes(code, utf, cb, base_list, base_end, rec_limit))
@@ -1067,7 +1071,7 @@
 PCRE2_SPTR end;
 PCRE2_UCHAR *repeat_opcode;
 uint32_t list[8];
-int rec_limit = 10000;
+int rec_limit = 1000;  /* Was 10,000 but clang+ASAN uses a lot of stack. */

Modified: code/trunk/testdata/testinput1
===================================================================
--- code/trunk/testdata/testinput1    2017-02-04 16:27:20 UTC (rev 658)
+++ code/trunk/testdata/testinput1    2017-02-08 17:03:30 UTC (rev 659)
@@ -5826,4 +5826,9 @@
 /[s[:digit:]\Q\E-H]+/
     s09-H

+/a+(?:|b)a/
+    aaaa
+
+/X?(R||){3335}/
+
 # End of testinput1 

Modified: code/trunk/testdata/testoutput1
===================================================================
--- code/trunk/testdata/testoutput1    2017-02-04 16:27:20 UTC (rev 658)
+++ code/trunk/testdata/testoutput1    2017-02-08 17:03:30 UTC (rev 659)
@@ -9305,4 +9305,10 @@
     s09-H
  0: s09-H

+/a+(?:|b)a/
+    aaaa
+ 0: aaaa
+
+/X?(R||){3335}/
+
 # End of testinput1