go1.16 (#14783)

vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm.s

@@ -140,25 +140,22 @@ readMatchlenLoop:
 	BEQ     readMatchlenLoop
 
 readMatchlenDone:
-	ADD minMatch, len
-
-	// Bounds check dst+len and match = dst-offset.
+	// Bounds check dst+len+minMatch and match = dst-offset.
 	ADD    dst, len, tmp1
+	ADD    minMatch, tmp1
 	CMP    dstend, tmp1
 	//BHI  shortDst	// Uncomment for distinct error codes.
 	SUB    offset, dst, match
 	CMP.LS match, dstorig
 	BHI    corrupt
 
-	// If the offset is at least four (len is, because of minMatch),
-	// do a four-way unrolled byte copy loop. Using MOVD instead of four
-	// byte loads is much faster, but to remain portable we'd have to
-	// align match first, which in turn is too expensive.
-	CMP $4, offset
-	BLO copyMatch
-
-	SUB $4, len
+	// Since len+minMatch is at least four, we can do a 4× unrolled
+	// byte copy loop. Using MOVW instead of four byte loads is faster,
+	// but to remain portable we'd have to align match first, which is
+	// too expensive. By alternating loads and stores, we also handle
+	// the case offset < 4.
 copyMatch4:
+	SUB.S   $4, len
 	MOVBU.P 4(match), tmp1
 	MOVB.P  tmp1, 4(dst)
 	MOVBU   -3(match), tmp2
@@ -167,7 +164,6 @@ copyMatch4:
 	MOVB    tmp3, -2(dst)
 	MOVBU   -1(match), tmp1
 	MOVB    tmp1, -1(dst)
-	SUB.S   $4, len
 	BPL     copyMatch4
 
 	// Restore len, which is now negative.
@@ -175,7 +171,7 @@ copyMatch4:
 	BEQ   copyMatchDone
 
 copyMatch:
-	// Simple byte-at-a-time copy.
+	// Finish with a byte-at-a-time copy.
 	SUB.S   $1, len
 	MOVBU.P 1(match), tmp2
 	MOVB.P  tmp2, 1(dst)

vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_other.go

@@ -2,7 +2,13 @@
 
 package lz4block
 
+import "encoding/binary"
+
 func decodeBlock(dst, src []byte) (ret int) {
+	// Restrict capacities so we don't read or write out of bounds.
+	dst = dst[:len(dst):len(dst)]
+	src = src[:len(src):len(src)]
+
 	const hasError = -2
 	defer func() {
 		if recover() != nil {
@@ -32,7 +38,7 @@ func decodeBlock(dst, src []byte) (ret int) {
 					// if the match length (4..18) fits within the literals, then copy
 					// all 18 bytes, even if not all are part of the literals.
 					mLen += 4
-					if offset := uint(src[si]) | uint(src[si+1])<<8; mLen <= offset {
+					if offset := u16(src[si:]); mLen <= offset {
 						i := di - offset
 						end := i + 18
 						if end > uint(len(dst)) {
@@ -66,7 +72,7 @@ func decodeBlock(dst, src []byte) (ret int) {
 			return hasError
 		}
 
-		offset := uint(src[si]) | uint(src[si+1])<<8
+		offset := u16(src[si:])
 		if offset == 0 {
 			return hasError
 		}
@@ -98,3 +104,5 @@ func decodeBlock(dst, src []byte) (ret int) {
 		di += uint(copy(dst[di:di+mLen], expanded[:mLen]))
 	}
 }
+
+func u16(p []byte) uint { return uint(binary.LittleEndian.Uint16(p)) }