ticket:12888 UText, problems with handling of bad UTF-8

Fixes:

CVE-2017-7867 - International Components for Unicode (ICU) for C/C++ before
2017-02-13 has an out-of-bounds write caused by a heap-based buffer overflow
related to the utf8TextAccess function in common/utext.cpp and the
utext_setNativeIndex* function.

CVE-2017-7868 - International Components for Unicode (ICU) for C/C++ before
2017-02-13 has an out-of-bounds write caused by a heap-based buffer overflow
related to the utf8TextAccess function in common/utext.cpp and the
utext_moveIndex32* function.

Upstream: http://bugs.icu-project.org/trac/changeset/39671

Signed-off-by: Peter Korsgaard <peter@korsgaard.com>

Index: /trunk/icu4c/source/common/utext.cpp
===================================================================
--- icu4c/source/common/utext.cpp	(revision 39670)
+++ icu4c/source/common/utext.cpp	(revision 39671)
@@ -848,7 +848,13 @@
 
 // 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 };
@@ -890,5 +896,5 @@
                                                      //    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[UTF8_TEXT_CHUNK_SIZE*6+6]; // Map native offset from bufNativeStart to
                                                      //   correspoding offset in filled part of buf.
     int32_t   align;
@@ -1033,4 +1039,5 @@
             u8b = (UTF8Buf *)ut->p;   // the current buffer
             mapIndex = ix - u8b->toUCharsMapStart;
+            U_ASSERT(mapIndex < (int32_t)sizeof(UTF8Buf::mapToUChars));
             ut->chunkOffset = u8b->mapToUChars[mapIndex] - u8b->bufStartIdx;
             return TRUE;
@@ -1299,4 +1306,8 @@
         //   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);
         }
@@ -1312,5 +1323,8 @@
         uint8_t *mapToNative = u8b->mapToNative;
         uint8_t *mapToUChars = u8b->mapToUChars;
-        int32_t  toUCharsMapStart = ix - (UTF8_TEXT_CHUNK_SIZE*3 + 1);
+        int32_t  toUCharsMapStart = ix - sizeof(UTF8Buf::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
@@ -1339,4 +1353,5 @@
                 // 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);
@@ -1368,4 +1383,5 @@
                     mapToUChars[sIx-- - toUCharsMapStart] = (uint8_t)destIx;
                 } while (sIx >= srcIx);
+                U_ASSERT(toUCharsMapStart <= (srcIx+1));
 
                 // Set native indexing limit to be the current position.
@@ -1542,4 +1558,5 @@
     U_ASSERT(index<=ut->chunkNativeLimit);
     int32_t mapIndex = index - u8b->toUCharsMapStart;
+    U_ASSERT(mapIndex < (int32_t)sizeof(UTF8Buf::mapToUChars));
     int32_t offset = u8b->mapToUChars[mapIndex] - u8b->bufStartIdx;
     U_ASSERT(offset>=0 && offset<=ut->chunkLength);
Index: /trunk/icu4c/source/test/intltest/utxttest.cpp
===================================================================
--- icu4c/source/test/intltest/utxttest.cpp	(revision 39670)
+++ icu4c/source/test/intltest/utxttest.cpp	(revision 39671)
@@ -68,4 +68,6 @@
         case 7: name = "Ticket12130";
             if (exec) Ticket12130(); break;
+        case 8: name = "Ticket12888";
+            if (exec) Ticket12888(); break;
         default: name = "";          break;
     }
@@ -1584,2 +1586,62 @@
     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;
+    }
+}
Index: /trunk/icu4c/source/test/intltest/utxttest.h
===================================================================
--- icu4c/source/test/intltest/utxttest.h	(revision 39670)
+++ icu4c/source/test/intltest/utxttest.h	(revision 39671)
@@ -39,4 +39,5 @@
     void Ticket10983();
     void Ticket12130();
+    void Ticket12888();
 
 private: