Fix integer overflow for patterns whose minimum matching length is large

This commit is contained in:
Petr Písař 2015-10-12 13:00:35 +02:00
parent d9624ccacb
commit 4c42ca60bd
2 changed files with 116 additions and 0 deletions

View File

@ -0,0 +1,110 @@
From 3d9c33b0c83722c90ca091389aff9056d97a8e93 Mon Sep 17 00:00:00 2001
From: ph10 <ph10@6239d852-aaf2-0410-a92c-79f79f948069>
Date: Fri, 9 Oct 2015 16:54:29 +0000
Subject: [PATCH] Fix integer overflow for patterns whose minimum matching
length is very, very large.
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Upstream commit ported to 10.20:
commit 91cb973c4537327c18217b50f20e82283affe976
Author: ph10 <ph10@6239d852-aaf2-0410-a92c-79f79f948069>
Date: Fri Oct 9 16:54:29 2015 +0000
Fix integer overflow for patterns whose minimum matching length is very, very
large.
git-svn-id: svn://vcs.exim.org/pcre2/code/trunk@385 6239d852-aaf2-0410-a92c-79f79f948069
The "Subject length lower bound" in testoutput should be 65535, but
it's not because the lower bound guess was improved in:
commit 209880186ae777d2404560fff04d4e3c36b55ffe
Author: ph10 <ph10@6239d852-aaf2-0410-a92c-79f79f948069>
Date: Sat Aug 1 09:11:28 2015 +0000
Fix issues with minimum length finding.
git-svn-id: svn://vcs.exim.org/pcre2/code/trunk@331 6239d852-aaf2-0410-a92c-
79f79f948069
that is quite invasive, thus not ported. Therefore the test returns
suboptimal value.
Signed-off-by: Petr Písař <ppisar@redhat.com>
---
src/pcre2_study.c | 12 +++++++++---
testdata/testinput2 | 2 ++
testdata/testoutput2 | 8 ++++++++
3 files changed, 19 insertions(+), 3 deletions(-)
diff --git a/src/pcre2_study.c b/src/pcre2_study.c
index 25d7e51..a5c2cb4 100644
--- a/src/pcre2_study.c
+++ b/src/pcre2_study.c
@@ -66,8 +66,11 @@ enum { SSB_FAIL, SSB_DONE, SSB_CONTINUE, SSB_UNKNOWN };
/* Scan a parenthesized group and compute the minimum length of subject that
is needed to match it. This is a lower bound; it does not mean there is a
-string of that length that matches. In UTF8 mode, the result is in characters
-rather than bytes.
+string of that length that matches. In UTF mode, the result is in characters
+rather than code units. The field in a compiled pattern for storing the minimum
+length is 16-bits long (on the grounds that anything longer than that is
+pathological), so we give up when we reach that amount. This also means that
+integer overflow for really crazy patterns cannot happen.
Arguments:
re compiled pattern block
@@ -97,7 +100,8 @@ if (*code == OP_CBRA || *code == OP_SCBRA ||
*code == OP_CBRAPOS || *code == OP_SCBRAPOS) cc += IMM2_SIZE;
/* Scan along the opcodes for this branch. If we get to the end of the
-branch, check the length against that of the other branches. */
+branch, check the length against that of the other branches. If the accumulated
+length passes 16-bits, stop and return it. */
for (;;)
{
@@ -105,6 +109,8 @@ for (;;)
PCRE2_UCHAR *cs, *ce;
register PCRE2_UCHAR op = *cc;
+ if (branchlength > UINT16_MAX) return branchlength;
+
switch (op)
{
case OP_COND:
diff --git a/testdata/testinput2 b/testdata/testinput2
index e0e149f..dc64019 100644
--- a/testdata/testinput2
+++ b/testdata/testinput2
@@ -4327,4 +4327,6 @@ a random value. /Ix
/^(?:(?(1)x|)+)+$()/B
+/^(o(\1{72}{\"{\\{00000059079}\d*){74}}){19}/I
+
# End of testinput2
diff --git a/testdata/testoutput2 b/testdata/testoutput2
index b628466..cfc3ef0 100644
--- a/testdata/testoutput2
+++ b/testdata/testoutput2
@@ -14470,4 +14470,12 @@ Failed: error 161 at offset 32: number is too big
End
------------------------------------------------------------------
+/^(o(\1{72}{\"{\\{00000059079}\d*){74}}){19}/I
+Capturing subpattern count = 2
+Max back reference = 1
+Compile options: <none>
+Overall options: anchored
+Last code unit = '}'
+Subject length lower bound = 52629
+
# End of testinput2
--
2.4.3

View File

@ -19,6 +19,9 @@ Patch0: pcre2-10.10-Fix-multilib.patch
# Fix compiling classes with a negative escape and a property escape,
# upstream bug #1697, fixed in upstream after 10.20.
Patch1: pcre2-10.20-Fix-compiler-bug-for-classes-such-as-W-p-Any.patch
# Fix integer overflow for patterns whose minimum matching length is large,
# upstream bug #1699, fixed in upstream after 10.20.
Patch2: pcre2-10.20-Fix-integer-overflow-for-patterns-whose-minimum-matc.patch
# New libtool to get rid of RPATH and to use distribution autotools
BuildRequires: autoconf
@ -80,6 +83,7 @@ Utilities demonstrating PCRE2 capabilities like pcre2grep or pcre2test.
%setup -q -n %{name}-%{myversion}
%patch0 -p1
%patch1 -p1
%patch2 -p1
# Because of multilib patch
libtoolize --copy --force
autoreconf -vif
@ -161,6 +165,8 @@ make %{?_smp_mflags} check VERBOSE=yes
* Mon Oct 12 2015 Petr Pisar <ppisar@redhat.com> - 10.20-2
- Fix compiling classes with a negative escape and a property escape
(upstream bug #1697)
- Fix integer overflow for patterns whose minimum matching length is large
(upstream bug #1699)
* Fri Jul 03 2015 Petr Pisar <ppisar@redhat.com> - 10.20-1
- 10.20 bump