lzx_compress.c: avoid undefined shifts in lzx_flush_bits()

[wimlib] / src / lzx_compress.c

diff --git a/src/lzx_compress.c b/src/lzx_compress.c
index cc43fd6cc79ddf07eab9b067589a531e08b1c9a1..c682e7e9a1f26562f9c7fe63249f08d8f7217112 100644 (file)
--- a/src/lzx_compress.c
+++ b/src/lzx_compress.c
@@ -410,8 +410,10 @@ struct lzx_compressor {
 
        /* The matches and literals that the parser has chosen for the current
         * block.  The required length of this array is limited by the maximum
-        * number of matches that can ever be chosen for a single block.  */
-       struct lzx_sequence chosen_sequences[DIV_ROUND_UP(LZX_DIV_BLOCK_SIZE, LZX_MIN_MATCH_LEN)];
+        * number of matches that can ever be chosen for a single block, plus
+        * one for the special entry at the end.  */
+       struct lzx_sequence chosen_sequences[
+                      DIV_ROUND_UP(LZX_DIV_BLOCK_SIZE, LZX_MIN_MATCH_LEN) + 1];
 
        /* Tables for mapping adjusted offsets to offset slots  */
 
@@ -585,13 +587,21 @@ lzx_add_bits(struct lzx_output_bitstream *os, u32 bits, unsigned num_bits)
 static inline void
 lzx_flush_bits(struct lzx_output_bitstream *os, unsigned max_num_bits)
 {
+       /* Masking the number of bits to shift is only needed to avoid undefined
+        * behavior; we don't actually care about the results of bad shifts.  On
+        * x86, the explicit masking generates no extra code.  */
+       const u32 shift_mask = 8 * sizeof(os->bitbuf) - 1;
+
        if (os->end - os->next < 6)
                return;
-       put_unaligned_u16_le(os->bitbuf >> (os->bitcount - 16), os->next + 0);
+       put_unaligned_u16_le(os->bitbuf >> ((os->bitcount - 16) &
+                                           shift_mask), os->next + 0);
        if (max_num_bits > 16)
-               put_unaligned_u16_le(os->bitbuf >> (os->bitcount - 32), os->next + 2);
+               put_unaligned_u16_le(os->bitbuf >> ((os->bitcount - 32) &
+                                               shift_mask), os->next + 2);
        if (max_num_bits > 32)
-               put_unaligned_u16_le(os->bitbuf >> (os->bitcount - 48), os->next + 4);
+               put_unaligned_u16_le(os->bitbuf >> ((os->bitcount - 48) &
+                                               shift_mask), os->next + 4);
        os->next += (os->bitcount >> 4) << 1;
        os->bitcount &= 15;
 }
@@ -634,7 +644,7 @@ lzx_make_huffman_codes(struct lzx_compressor *c)
 
        STATIC_ASSERT(MAIN_CODEWORD_LIMIT >= 9 &&
                      MAIN_CODEWORD_LIMIT <= LZX_MAX_MAIN_CODEWORD_LEN);
-       STATIC_ASSERT(LENGTH_CODEWORD_LIMIT >= 9 &&
+       STATIC_ASSERT(LENGTH_CODEWORD_LIMIT >= 8 &&
                      LENGTH_CODEWORD_LIMIT <= LZX_MAX_LEN_CODEWORD_LEN);
        STATIC_ASSERT(ALIGNED_CODEWORD_LIMIT >= LZX_NUM_ALIGNED_OFFSET_BITS &&
                      ALIGNED_CODEWORD_LIMIT <= LZX_MAX_ALIGNED_CODEWORD_LEN);