add ICU changeset-39671 fix for CVE-2017-7867 CVE-2017-7868

http://bugs.icu-project.org/trac/changeset/39671
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=213
https://bugzilla.redhat.com/show_bug.cgi?id=1444101

Reviewed-on: https://gerrit.libreoffice.org/36754
Reviewed-by: Eike Rathke <erack@redhat.com>
Tested-by: Jenkins <ci@libreoffice.org>
(cherry picked from commit c7de8233d15ed0c90fef6c49a54d60cf10119f58)

Backported to older MSVC using the UGLY_SIZEOF_MAPTOUCHARS macro instead
of sizeof(UTF8Buf::mapToUChars).

Change-Id: I4e776ad4fe63c77057b0c823f8672a2b6703346f
Reviewed-on: https://gerrit.libreoffice.org/36776
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Michael Stahl <mstahl@redhat.com>
(cherry picked from commit 91f5d002884cae1a60768e9caa9d182f41fb7be6)
(cherry picked from commit 3cdac6bb2defce45342dff04400c7a37bb8a2453)

2 files changed, 190 insertions, 0 deletions

diff --git a/external/icu/UnpackedTarball_icu.mk b/external/icu/UnpackedTarball_icu.mk
index 4a6a11477af3..86369ef8e43b 100644
--- a/external/icu/UnpackedTarball_icu.mk
+++ b/external/icu/UnpackedTarball_icu.mk
@@ -28,6 +28,7 @@ $(eval $(call gb_UnpackedTarball_add_patches,icu,\
 	$(if $(filter-out ANDROID,$(OS)),external/icu/icu4c-icudata-stdlibs.diff) \
 	$(if $(filter EMSCRIPTEN,$(OS)),external/icu/icu4c-emscripten.patch.1) \
 	external/icu/khmerbreakengine.patch \
+	external/icu/icu4c-changeset-39671.patch.1 \
 ))
 
 $(eval $(call gb_UnpackedTarball_add_file,icu,source/data/brkitr/khmerdict.dict,external/icu/khmerdict.dict))
diff --git a/external/icu/icu4c-changeset-39671.patch.1 b/external/icu/icu4c-changeset-39671.patch.1
new file mode 100644
index 000000000000..b8ac1385364e
--- /dev/null
+++ b/external/icu/icu4c-changeset-39671.patch.1
@@ -0,0 +1,189 @@
+diff -ur icu.org/source/common/utext.cpp icu/source/common/utext.cpp
+--- icu.org/source/common/utext.cpp	2016-06-15 20:58:17.000000000 +0200
++++ icu/source/common/utext.cpp	2017-04-21 16:38:15.993398034 +0200
+@@ -847,9 +847,15 @@
+ //------------------------------------------------------------------------------
+ 
+ // Chunk size.
+-//     Must be less than 85, because of byte mapping from UChar indexes to native indexes.
+-//     Worst case is three native bytes to one UChar.  (Supplemenaries are 4 native bytes
+-//     to two UChars.)
++//     Must be less than 42  (256/6), because of byte mapping from UChar indexes to native indexes.
++//     Worst case there are six UTF-8 bytes per UChar.
++//         obsolete 6 byte form fd + 5 trails maps to fffd
++//         obsolete 5 byte form fc + 4 trails maps to fffd
++//         non-shortest 4 byte forms maps to fffd
++//         normal supplementaries map to a pair of utf-16, two utf8 bytes per utf-16 unit
++//     mapToUChars array size must allow for the worst case, 6.
++//     This could be brought down to 4, by treating fd and fc as pure illegal,
++//     rather than obsolete lead bytes. But that is not compatible with the utf-8 access macros.
+ //
+ enum { UTF8_TEXT_CHUNK_SIZE=32 };
+ 
+@@ -867,6 +873,15 @@
+ //     pair.  Doing this is simpler than checking for the edge case.
+ //
+ 
++// erAck: older MSVC used on libreoffice-5-3 and 5-2 bails out with
++// error C2070: 'unknown': illegal sizeof operand
++// for sizeof(UTF8Buf::mapToUChars)
++// so have an ugly workaround:
++// First define a macro of the original size expression, so a follow-up patch
++// on the original code would fail..
++#define UGLY_MAPTOUCHARS_SIZE (UTF8_TEXT_CHUNK_SIZE*6+6)
++#define UGLY_SIZEOF_MAPTOUCHARS (sizeof(uint8_t)*(UGLY_MAPTOUCHARS_SIZE))
++
+ struct UTF8Buf {
+     int32_t   bufNativeStart;                        // Native index of first char in UChar buf
+     int32_t   bufNativeLimit;                        // Native index following last char in buf.
+@@ -889,7 +904,7 @@
+                                                      //  Requires two extra slots,
+                                                      //    one for a supplementary starting in the last normal position,
+                                                      //    and one for an entry for the buffer limit position.
+-    uint8_t   mapToUChars[UTF8_TEXT_CHUNK_SIZE*3+6]; // Map native offset from bufNativeStart to
++    uint8_t   mapToUChars[UGLY_MAPTOUCHARS_SIZE];    // Map native offset from bufNativeStart to
+                                                      //   correspoding offset in filled part of buf.
+     int32_t   align;
+ };
+@@ -1032,6 +1047,7 @@
+             // Requested index is in this buffer.
+             u8b = (UTF8Buf *)ut->p;   // the current buffer
+             mapIndex = ix - u8b->toUCharsMapStart;
++            U_ASSERT(mapIndex < (int32_t)UGLY_SIZEOF_MAPTOUCHARS);
+             ut->chunkOffset = u8b->mapToUChars[mapIndex] - u8b->bufStartIdx;
+             return TRUE;
+ 
+@@ -1298,6 +1314,10 @@
+         // Can only do this if the incoming index is somewhere in the interior of the string.
+         //   If index is at the end, there is no character there to look at.
+         if (ix != ut->b) {
++            // Note: this function will only move the index back if it is on a trail byte
++            //       and there is a preceding lead byte and the sequence from the lead 
++            //       through this trail could be part of a valid UTF-8 sequence
++            //       Otherwise the index remains unchanged.
+             U8_SET_CP_START(s8, 0, ix);
+         }
+ 
+@@ -1311,7 +1331,10 @@
+         UChar   *buf = u8b->buf;
+         uint8_t *mapToNative = u8b->mapToNative;
+         uint8_t *mapToUChars = u8b->mapToUChars;
+-        int32_t  toUCharsMapStart = ix - (UTF8_TEXT_CHUNK_SIZE*3 + 1);
++        int32_t  toUCharsMapStart = ix - UGLY_SIZEOF_MAPTOUCHARS + 1;
++        // Note that toUCharsMapStart can be negative. Happens when the remaining
++        // text from current position to the beginning is less than the buffer size.
++        // + 1 because mapToUChars must have a slot at the end for the bufNativeLimit entry.
+         int32_t  destIx = UTF8_TEXT_CHUNK_SIZE+2;   // Start in the overflow region
+                                                     //   at end of buffer to leave room
+                                                     //   for a surrogate pair at the
+@@ -1338,6 +1361,7 @@
+             if (c<0x80) {
+                 // Special case ASCII range for speed.
+                 buf[destIx] = (UChar)c;
++                U_ASSERT(toUCharsMapStart <= srcIx);
+                 mapToUChars[srcIx - toUCharsMapStart] = (uint8_t)destIx;
+                 mapToNative[destIx] = (uint8_t)(srcIx - toUCharsMapStart);
+             } else {
+@@ -1367,6 +1391,7 @@
+                 do {
+                     mapToUChars[sIx-- - toUCharsMapStart] = (uint8_t)destIx;
+                 } while (sIx >= srcIx);
++                U_ASSERT(toUCharsMapStart <= (srcIx+1));
+ 
+                 // Set native indexing limit to be the current position.
+                 //   We are processing a non-ascii, non-native-indexing char now;
+@@ -1541,6 +1566,7 @@
+     U_ASSERT(index>=ut->chunkNativeStart+ut->nativeIndexingLimit);
+     U_ASSERT(index<=ut->chunkNativeLimit);
+     int32_t mapIndex = index - u8b->toUCharsMapStart;
++    U_ASSERT(mapIndex < (int32_t)UGLY_SIZEOF_MAPTOUCHARS);
+     int32_t offset = u8b->mapToUChars[mapIndex] - u8b->bufStartIdx;
+     U_ASSERT(offset>=0 && offset<=ut->chunkLength);
+     return offset;
+diff -ur icu.org/source/test/intltest/utxttest.cpp icu/source/test/intltest/utxttest.cpp
+--- icu.org/source/test/intltest/utxttest.cpp	2016-06-15 20:58:17.000000000 +0200
++++ icu/source/test/intltest/utxttest.cpp	2017-04-21 16:14:57.383814739 +0200
+@@ -67,6 +67,8 @@
+             if (exec) Ticket10983();  break;
+         case 7: name = "Ticket12130";
+             if (exec) Ticket12130(); break;
++        case 8: name = "Ticket12888";
++            if (exec) Ticket12888(); break;
+         default: name = "";          break;
+     }
+ }
+@@ -1583,3 +1585,63 @@
+     }
+     utext_close(&ut);
+ }
++
++// Ticket 12888: bad handling of illegal utf-8 containing many instances of the archaic, now illegal,
++//               six byte utf-8 forms. Original implementation had an assumption that
++//               there would be at most three utf-8 bytes per UTF-16 code unit.
++//               The five and six byte sequences map to a single replacement character.
++
++void UTextTest::Ticket12888() {
++    const char *badString = 
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
++            "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80";
++
++    UErrorCode status = U_ZERO_ERROR;
++    LocalUTextPointer ut(utext_openUTF8(NULL, badString, -1, &status));
++    TEST_SUCCESS(status);
++    for (;;) {
++        UChar32 c = utext_next32(ut.getAlias());
++        if (c == U_SENTINEL) {
++            break;
++        }
++    }
++    int32_t endIdx = utext_getNativeIndex(ut.getAlias());
++    if (endIdx != (int32_t)strlen(badString)) {
++        errln("%s:%d expected=%d, actual=%d", __FILE__, __LINE__, strlen(badString), endIdx);
++        return;
++    }
++
++    for (int32_t prevIndex = endIdx; prevIndex>0;) {
++        UChar32 c = utext_previous32(ut.getAlias());
++        int32_t currentIndex = utext_getNativeIndex(ut.getAlias());
++        if (c != 0xfffd) {
++            errln("%s:%d (expected, actual, index) = (%d, %d, %d)\n",
++                    __FILE__, __LINE__, 0xfffd, c, currentIndex);
++            break;
++        }
++        if (currentIndex != prevIndex - 6) {
++            errln("%s:%d: wrong index. Expected, actual = %d, %d",
++                    __FILE__, __LINE__, prevIndex - 6, currentIndex);
++            break;
++        }
++        prevIndex = currentIndex;
++    }
++}
+diff -ur icu.org/source/test/intltest/utxttest.h icu/source/test/intltest/utxttest.h
+--- icu.org/source/test/intltest/utxttest.h	2016-06-15 20:58:17.000000000 +0200
++++ icu/source/test/intltest/utxttest.h	2017-04-21 16:14:57.383814739 +0200
+@@ -38,6 +38,7 @@
+     void Ticket10562();
+     void Ticket10983();
+     void Ticket12130();
++    void Ticket12888();
+ 
+ private:
+     struct m {                              // Map between native indices & code points.