Accept as many selectors as the file format allows.

But ignore any larger than the theoretical maximum, BZ_MAX_SELECTORS.

The theoretical maximum number of selectors depends on the maximum
blocksize (900000 bytes) and the number of symbols (50) that can be
encoded with a different Huffman tree. BZ_MAX_SELECTORS is 18002.

But the bzip2 file format allows the number of selectors to be encoded
with 15 bits (because 18002 isn't a factor of 2 and doesn't fit in
14 bits). So the file format maximum is 32767 selectors.

Some bzip2 encoders might actually have written out more selectors
than the theoretical maximum because they rounded up the number of
selectors to some convenient factor of 8.

The extra 14766 selectors can never be validly used by the decompression
algorithm. So we can read them, but then discard them.

This is effectively what was done (by accident) before we added a
check for nSelectors to be at most BZ_MAX_SELECTORS to mitigate
CVE-2019-12900.

The extra selectors were written out after the array inside the
EState struct. But the struct has extra space allocated after the
selector arrays of 18060 bytes (which is larger than 14766).
All of which will be initialized later (so the overwrite of that
space with extra selector values would have been harmless).

diff --git a/compress.c b/compress.c
index 237620d85891e045b9a40c19bef0a896f91d9dca..76adee6589b4d0243e00ec2a87a2798d2b6fd032 100644 (file)
--- a/compress.c
+++ b/compress.c
@@ -454,7 +454,7 @@ void sendMTFValues ( EState* s )
 
    AssertH( nGroups < 8, 3002 );
    AssertH( nSelectors < 32768 &&
-            nSelectors <= (2 + (900000 / BZ_G_SIZE)),
+            nSelectors <= BZ_MAX_SELECTORS,
             3003 );
 
 

diff --git a/decompress.c b/decompress.c
index 20ce4936cdfb15e3a7125c01a6ceffa22a349dcb..33034996f99f6a3992441b7d49d38facfedf62e8 100644 (file)
--- a/decompress.c
+++ b/decompress.c
@@ -287,7 +287,7 @@ Int32 BZ2_decompress ( DState* s )
       GET_BITS(BZ_X_SELECTOR_1, nGroups, 3);
       if (nGroups < 2 || nGroups > BZ_N_GROUPS) RETURN(BZ_DATA_ERROR);
       GET_BITS(BZ_X_SELECTOR_2, nSelectors, 15);
-      if (nSelectors < 1 || nSelectors > BZ_MAX_SELECTORS) RETURN(BZ_DATA_ERROR);
+      if (nSelectors < 1) RETURN(BZ_DATA_ERROR);
       for (i = 0; i < nSelectors; i++) {
          j = 0;
          while (True) {
@@ -296,8 +296,14 @@ Int32 BZ2_decompress ( DState* s )
             j++;
             if (j >= nGroups) RETURN(BZ_DATA_ERROR);
          }
-         s->selectorMtf[i] = j;
+         /* Having more than BZ_MAX_SELECTORS doesn't make much sense
+            since they will never be used, but some implementations might
+            "round up" the number of selectors, so just ignore those. */
+         if (i < BZ_MAX_SELECTORS)
+           s->selectorMtf[i] = j;
       }
+      if (nSelectors > BZ_MAX_SELECTORS)
+        nSelectors = BZ_MAX_SELECTORS;
 
       /*--- Undo the MTF values for the selectors. ---*/
       {