Pick up extract.h stuff from tcpdump, and use it in bpf_filter.c.

No point in duplicating the unaligned-big-endian-extraction stuff.

1 files changed, 242 insertions, 50 deletions

diff --git a/extract.h b/extract.h
index aa3ff991..941ccc2a 100644
--- a/extract.h
+++ b/extract.h
@@ -25,15 +25,98 @@
 
 #include <pcap/pcap-inttypes.h>
 #include <pcap/compiler-tests.h>
+#include "portability.h"
 
 /*
- * Macros to extract possibly-unaligned big-endian integral values.
+ * If we have versions of GCC or Clang that support an __attribute__
+ * to say "if we're building with unsigned behavior sanitization,
+ * don't complain about undefined behavior in this function", we
+ * label these functions with that attribute - we *know* it's undefined
+ * in the C standard, but we *also* know it does what we want with
+ * the ISA we're targeting and the compiler we're using.
+ *
+ * For GCC 4.9.0 and later, we use __attribute__((no_sanitize_undefined));
+ * pre-5.0 GCC doesn't have __has_attribute, and I'm not sure whether
+ * GCC or Clang first had __attribute__((no_sanitize(XXX)).
+ *
+ * For Clang, we check for __attribute__((no_sanitize(XXX)) with
+ * __has_attribute, as there are versions of Clang that support
+ * __attribute__((no_sanitize("undefined")) but don't support
+ * __attribute__((no_sanitize_undefined)).
+ *
+ * We define this here, rather than in funcattrs.h, because we
+ * only want it used here, we don't want it to be broadly used.
+ * (Any printer will get this defined, but this should at least
+ * make it harder for people to find.)
  */
-#ifdef LBL_ALIGN
+#if defined(__GNUC__) && ((__GNUC__ * 100 + __GNUC_MINOR__) >= 409)
+#define UNALIGNED_OK	__attribute__((no_sanitize_undefined))
+#elif __has_attribute(no_sanitize)
+#define UNALIGNED_OK	__attribute__((no_sanitize("undefined")))
+#else
+#define UNALIGNED_OK
+#endif
+
+#if (defined(__i386__) || defined(_M_IX86) || defined(__X86__) || defined(__x86_64__) || defined(_M_X64)) || \
+    (defined(__arm__) || defined(_M_ARM) || defined(__aarch64__)) || \
+    (defined(__m68k__) && (!defined(__mc68000__) && !defined(__mc68010__))) || \
+    (defined(__ppc__) || defined(__ppc64__) || defined(_M_PPC) || defined(_ARCH_PPC) || defined(_ARCH_PPC64)) || \
+    (defined(__s390__) || defined(__s390x__) || defined(__zarch__))
 /*
- * The processor doesn't natively handle unaligned loads.
+ * The processor natively handles unaligned loads, so we can just
+ * cast the pointer and fetch through it.
+ *
+ * XXX - are those all the x86 tests we need?
+ * XXX - do we need to worry about ARMv1 through ARMv5, which didn't
+ * support unaligned loads, and, if so, do we need to worry about all
+ * of them, or just some of them, e.g. ARMv5?
+ * XXX - are those the only 68k tests we need not to generated
+ * unaligned accesses if the target is the 68000 or 68010?
+ * XXX - are there any tests we don't need, because some definitions are for
+ * compilers that also predefine the GCC symbols?
+ * XXX - do we need to test for both 32-bit and 64-bit versions of those
+ * architectures in all cases?
  */
-#if PCAP_IS_AT_LEAST_GNUC_VERSION(2,0) && \
+UNALIGNED_OK static inline uint16_t
+EXTRACT_BE_U_2(const void *p)
+{
+	return ((uint16_t)ntohs(*(const uint16_t *)(p)));
+}
+
+UNALIGNED_OK static inline int16_t
+EXTRACT_BE_S_2(const void *p)
+{
+	return ((int16_t)ntohs(*(const int16_t *)(p)));
+}
+
+UNALIGNED_OK static inline uint32_t
+EXTRACT_BE_U_4(const void *p)
+{
+	return ((uint32_t)ntohl(*(const uint32_t *)(p)));
+}
+
+UNALIGNED_OK static inline int32_t
+EXTRACT_BE_S_4(const void *p)
+{
+	return ((int32_t)ntohl(*(const int32_t *)(p)));
+}
+
+UNALIGNED_OK static inline uint64_t
+EXTRACT_BE_U_8(const void *p)
+{
+	return ((uint64_t)(((uint64_t)ntohl(*((const uint32_t *)(p) + 0))) << 32 |
+		((uint64_t)ntohl(*((const uint32_t *)(p) + 1))) << 0));
+
+}
+
+UNALIGNED_OK static inline int64_t
+EXTRACT_BE_S_8(const void *p)
+{
+	return ((int64_t)(((int64_t)ntohl(*((const uint32_t *)(p) + 0))) << 32 |
+		((uint64_t)ntohl(*((const uint32_t *)(p) + 1))) << 0));
+
+}
+#elif ND_IS_AT_LEAST_GNUC_VERSION(2,0) && \
     (defined(__alpha) || defined(__alpha__) || \
      defined(__mips) || defined(__mips__))
 /*
@@ -93,45 +176,80 @@ typedef struct {
 } __attribute__((packed)) unaligned_uint16_t;
 
 typedef struct {
+	int16_t		val;
+} __attribute__((packed)) unaligned_int16_t;
+
+typedef struct {
 	uint32_t	val;
 } __attribute__((packed)) unaligned_uint32_t;
 
-static inline uint16_t
-EXTRACT_16BITS(const void *p)
+typedef struct {
+	int32_t		val;
+} __attribute__((packed)) unaligned_int32_t;
+
+UNALIGNED_OK static inline uint16_t
+EXTRACT_BE_U_2(const void *p)
 {
 	return ((uint16_t)ntohs(((const unaligned_uint16_t *)(p))->val));
 }
 
-static inline uint32_t
-EXTRACT_32BITS(const void *p)
+UNALIGNED_OK static inline int16_t
+EXTRACT_BE_S_2(const void *p)
+{
+	return ((int16_t)ntohs(((const unaligned_int16_t *)(p))->val));
+}
+
+UNALIGNED_OK static inline uint32_t
+EXTRACT_BE_U_4(const void *p)
 {
 	return ((uint32_t)ntohl(((const unaligned_uint32_t *)(p))->val));
 }
 
-static inline uint64_t
-EXTRACT_64BITS(const void *p)
+UNALIGNED_OK static inline int32_t
+EXTRACT_BE_S_4(const void *p)
 {
-	return ((uint64_t)(((uint64_t)ntohl(((const unaligned_uint32_t *)(p) + 0)->val)) << 32 | \
+	return ((int32_t)ntohl(((const unaligned_int32_t *)(p))->val));
+}
+
+UNALIGNED_OK static inline uint64_t
+EXTRACT_BE_U_8(const void *p)
+{
+	return ((uint64_t)(((uint64_t)ntohl(((const unaligned_uint32_t *)(p) + 0)->val)) << 32 |
 		((uint64_t)ntohl(((const unaligned_uint32_t *)(p) + 1)->val)) << 0));
 }
 
-#else /* have to do it a byte at a time */
+UNALIGNED_OK static inline int64_t
+EXTRACT_BE_S_8(const void *p)
+{
+	return ((int64_t)(((uint64_t)ntohl(((const unaligned_uint32_t *)(p) + 0)->val)) << 32 |
+		((uint64_t)ntohl(((const unaligned_uint32_t *)(p) + 1)->val)) << 0));
+}
+#else
 /*
- * This isn't a GCC-compatible compiler, we don't have __attribute__,
+ * This architecture doesn't natively support unaligned loads, and either
+ * this isn't a GCC-compatible compiler, we don't have __attribute__,
  * or we do but we don't know of any better way with this instruction
  * set to do unaligned loads, so do unaligned loads of big-endian
  * quantities the hard way - fetch the bytes one at a time and
  * assemble them.
  */
-#define EXTRACT_16BITS(p) \
+#define EXTRACT_BE_U_2(p) \
 	((uint16_t)(((uint16_t)(*((const uint8_t *)(p) + 0)) << 8) | \
 	            ((uint16_t)(*((const uint8_t *)(p) + 1)) << 0)))
-#define EXTRACT_32BITS(p) \
+#define EXTRACT_BE_S_2(p) \
+	((int16_t)(((uint16_t)(*((const uint8_t *)(p) + 0)) << 8) | \
+	           ((uint16_t)(*((const uint8_t *)(p) + 1)) << 0)))
+#define EXTRACT_BE_U_4(p) \
 	((uint32_t)(((uint32_t)(*((const uint8_t *)(p) + 0)) << 24) | \
 	            ((uint32_t)(*((const uint8_t *)(p) + 1)) << 16) | \
 	            ((uint32_t)(*((const uint8_t *)(p) + 2)) << 8) | \
 	            ((uint32_t)(*((const uint8_t *)(p) + 3)) << 0)))
-#define EXTRACT_64BITS(p) \
+#define EXTRACT_BE_S_4(p) \
+	((int32_t)(((uint32_t)(*((const uint8_t *)(p) + 0)) << 24) | \
+	           ((uint32_t)(*((const uint8_t *)(p) + 1)) << 16) | \
+	           ((uint32_t)(*((const uint8_t *)(p) + 2)) << 8) | \
+	           ((uint32_t)(*((const uint8_t *)(p) + 3)) << 0)))
+#define EXTRACT_BE_U_8(p) \
 	((uint64_t)(((uint64_t)(*((const uint8_t *)(p) + 0)) << 56) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 1)) << 48) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 2)) << 40) | \
@@ -140,47 +258,67 @@ EXTRACT_64BITS(const void *p)
 	            ((uint64_t)(*((const uint8_t *)(p) + 5)) << 16) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 6)) << 8) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 7)) << 0)))
-#endif /* must special-case unaligned accesses */
-#else /* LBL_ALIGN */
+#define EXTRACT_BE_S_8(p) \
+	((int64_t)(((uint64_t)(*((const uint8_t *)(p) + 0)) << 56) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 1)) << 48) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 2)) << 40) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 3)) << 32) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 4)) << 24) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 5)) << 16) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 6)) << 8) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 7)) << 0)))
+
 /*
- * The processor natively handles unaligned loads, so we can just
- * cast the pointer and fetch through it.
+ * Extract an IPv4 address, which is in network byte order, and not
+ * necessarily aligned, and provide the result in host byte order.
  */
-static inline uint16_t
-EXTRACT_16BITS(const void *p)
-{
-	return ((uint16_t)ntohs(*(const uint16_t *)(p)));
-}
-
-static inline uint32_t
-EXTRACT_32BITS(const void *p)
-{
-	return ((uint32_t)ntohl(*(const uint32_t *)(p)));
-}
-
-static inline uint64_t
-EXTRACT_64BITS(const void *p)
-{
-	return ((uint64_t)(((uint64_t)ntohl(*((const uint32_t *)(p) + 0))) << 32 | \
-		((uint64_t)ntohl(*((const uint32_t *)(p) + 1))) << 0));
-
-}
-
-#endif /* LBL_ALIGN */
+#define EXTRACT_IPV4_TO_HOST_ORDER(p) \
+	((uint32_t)(((uint32_t)(*((const uint8_t *)(p) + 0)) << 24) | \
+	            ((uint32_t)(*((const uint8_t *)(p) + 1)) << 16) | \
+	            ((uint32_t)(*((const uint8_t *)(p) + 2)) << 8) | \
+	            ((uint32_t)(*((const uint8_t *)(p) + 3)) << 0)))
+#endif /* unaligned access checks */
 
-#define EXTRACT_24BITS(p) \
+/*
+ * Non-power-of-2 sizes.
+ */
+#define EXTRACT_BE_U_3(p) \
 	((uint32_t)(((uint32_t)(*((const uint8_t *)(p) + 0)) << 16) | \
 	            ((uint32_t)(*((const uint8_t *)(p) + 1)) << 8) | \
 	            ((uint32_t)(*((const uint8_t *)(p) + 2)) << 0)))
 
-#define EXTRACT_40BITS(p) \
+#define EXTRACT_BE_S_3(p) \
+	(((*((const uint8_t *)(p) + 0)) & 0x80) ? \
+	  ((int32_t)(((uint32_t)(*((const uint8_t *)(p) + 0)) << 16) | \
+	             ((uint32_t)(*((const uint8_t *)(p) + 1)) << 8) | \
+	             ((uint32_t)(*((const uint8_t *)(p) + 2)) << 0))) : \
+	  ((int32_t)(0xFF000000U | \
+	             ((uint32_t)(*((const uint8_t *)(p) + 0)) << 16) | \
+	             ((uint32_t)(*((const uint8_t *)(p) + 1)) << 8) | \
+	             ((uint32_t)(*((const uint8_t *)(p) + 2)) << 0))))
+
+#define EXTRACT_BE_U_5(p) \
 	((uint64_t)(((uint64_t)(*((const uint8_t *)(p) + 0)) << 32) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 1)) << 24) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 2)) << 16) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 3)) << 8) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 4)) << 0)))
 
-#define EXTRACT_48BITS(p) \
+#define EXTRACT_BE_S_5(p) \
+	(((*((const uint8_t *)(p) + 0)) & 0x80) ? \
+	  ((int64_t)(((uint64_t)(*((const uint8_t *)(p) + 0)) << 32) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 1)) << 24) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 2)) << 16) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 3)) << 8) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 4)) << 0))) : \
+	  ((int64_t)(INT64_T_CONSTANT(0xFFFFFF0000000000U) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 0)) << 32) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 1)) << 24) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 2)) << 16) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 3)) << 8) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 4)) << 0))))
+
+#define EXTRACT_BE_U_6(p) \
 	((uint64_t)(((uint64_t)(*((const uint8_t *)(p) + 0)) << 40) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 1)) << 32) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 2)) << 24) | \
@@ -188,7 +326,23 @@ EXTRACT_64BITS(const void *p)
 	            ((uint64_t)(*((const uint8_t *)(p) + 4)) << 8) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 5)) << 0)))
 
-#define EXTRACT_56BITS(p) \
+#define EXTRACT_BE_S_6(p) \
+	(((*((const uint8_t *)(p) + 0)) & 0x80) ? \
+	   ((int64_t)(((uint64_t)(*((const uint8_t *)(p) + 0)) << 40) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 1)) << 32) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 2)) << 24) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 3)) << 16) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 4)) << 8) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 5)) << 0))) : \
+	  ((int64_t)(INT64_T_CONSTANT(0xFFFFFFFF00000000U) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 0)) << 40) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 1)) << 32) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 2)) << 24) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 3)) << 16) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 4)) << 8) | \
+	              ((uint64_t)(*((const uint8_t *)(p) + 5)) << 0))))
+
+#define EXTRACT_BE_U_7(p) \
 	((uint64_t)(((uint64_t)(*((const uint8_t *)(p) + 0)) << 48) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 1)) << 40) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 2)) << 32) | \
@@ -197,24 +351,53 @@ EXTRACT_64BITS(const void *p)
 	            ((uint64_t)(*((const uint8_t *)(p) + 5)) << 8) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 6)) << 0)))
 
+#define EXTRACT_BE_S_7(p) \
+	(((*((const uint8_t *)(p) + 0)) & 0x80) ? \
+	  ((int64_t)(((uint64_t)(*((const uint8_t *)(p) + 0)) << 48) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 1)) << 40) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 2)) << 32) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 3)) << 24) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 4)) << 16) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 5)) << 8) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 6)) << 0))) : \
+	    ((int64_t)(INT64_T_CONSTANT(0xFFFFFFFFFF000000U) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 0)) << 48) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 1)) << 40) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 2)) << 32) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 3)) << 24) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 4)) << 16) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 5)) << 8) | \
+	             ((uint64_t)(*((const uint8_t *)(p) + 6)) << 0))))
+
 /*
  * Macros to extract possibly-unaligned little-endian integral values.
  * XXX - do loads on little-endian machines that support unaligned loads?
  */
-#define EXTRACT_LE_8BITS(p) (*(p))
-#define EXTRACT_LE_16BITS(p) \
+#define EXTRACT_LE_U_2(p) \
 	((uint16_t)(((uint16_t)(*((const uint8_t *)(p) + 1)) << 8) | \
 	            ((uint16_t)(*((const uint8_t *)(p) + 0)) << 0)))
-#define EXTRACT_LE_32BITS(p) \
+#define EXTRACT_LE_S_2(p) \
+	((int16_t)(((uint16_t)(*((const uint8_t *)(p) + 1)) << 8) | \
+	           ((uint16_t)(*((const uint8_t *)(p) + 0)) << 0)))
+#define EXTRACT_LE_U_4(p) \
 	((uint32_t)(((uint32_t)(*((const uint8_t *)(p) + 3)) << 24) | \
 	            ((uint32_t)(*((const uint8_t *)(p) + 2)) << 16) | \
 	            ((uint32_t)(*((const uint8_t *)(p) + 1)) << 8) | \
 	            ((uint32_t)(*((const uint8_t *)(p) + 0)) << 0)))
-#define EXTRACT_LE_24BITS(p) \
+#define EXTRACT_LE_S_4(p) \
+	((int32_t)(((uint32_t)(*((const uint8_t *)(p) + 3)) << 24) | \
+	           ((uint32_t)(*((const uint8_t *)(p) + 2)) << 16) | \
+	           ((uint32_t)(*((const uint8_t *)(p) + 1)) << 8) | \
+	           ((uint32_t)(*((const uint8_t *)(p) + 0)) << 0)))
+#define EXTRACT_LE_U_3(p) \
 	((uint32_t)(((uint32_t)(*((const uint8_t *)(p) + 2)) << 16) | \
 	            ((uint32_t)(*((const uint8_t *)(p) + 1)) << 8) | \
 	            ((uint32_t)(*((const uint8_t *)(p) + 0)) << 0)))
-#define EXTRACT_LE_64BITS(p) \
+#define EXTRACT_LE_S_3(p) \
+	((int32_t)(((uint32_t)(*((const uint8_t *)(p) + 2)) << 16) | \
+	           ((uint32_t)(*((const uint8_t *)(p) + 1)) << 8) | \
+	           ((uint32_t)(*((const uint8_t *)(p) + 0)) << 0)))
+#define EXTRACT_LE_U_8(p) \
 	((uint64_t)(((uint64_t)(*((const uint8_t *)(p) + 7)) << 56) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 6)) << 48) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 5)) << 40) | \
@@ -223,3 +406,12 @@ EXTRACT_64BITS(const void *p)
 	            ((uint64_t)(*((const uint8_t *)(p) + 2)) << 16) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 1)) << 8) | \
 	            ((uint64_t)(*((const uint8_t *)(p) + 0)) << 0)))
+#define EXTRACT_LE_S_8(p) \
+	((int64_t)(((uint64_t)(*((const uint8_t *)(p) + 7)) << 56) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 6)) << 48) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 5)) << 40) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 4)) << 32) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 3)) << 24) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 2)) << 16) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 1)) << 8) | \
+	           ((uint64_t)(*((const uint8_t *)(p) + 0)) << 0)))