Improve RRTYPE and CLASS parser

Fixes #6.

Author: k0ekk0ek

CMakeLists.txt

@@ -135,7 +135,6 @@ target_include_directories(
 
 target_sources(zone PRIVATE
   src/zone.c
-  src/types.c
   src/log.c
   src/parser.c
   src/fallback/parser.c

include/zone.h

@@ -183,9 +183,6 @@ extern "C" {
 #define ZONE_DLV (32769u)
 /** @} */
 
-typedef int32_t zone_code_t;
-typedef int32_t zone_return_t;
-
 typedef struct zone_string zone_string_t;
 struct zone_string {
   size_t length;
@@ -194,7 +191,10 @@ struct zone_string {
 
 typedef struct zone_symbol zone_symbol_t;
 struct zone_symbol {
-  zone_string_t key;
+  struct {
+    char data[24]; // zero padded for convenient vectorized comparison
+    size_t length;
+  } key;
   uint32_t value;
 };
 
@@ -204,34 +204,6 @@ struct zone_table {
   const zone_symbol_t *symbols; // sorted for use with bsearch
 };
 
-// @private
-//
-// bsearch is quite slow compared to a hash table, but a hash table is either
-// quite big or there is a significant chance or collisions. a minimal perfect
-// hash table can be used instead, but there is a good chance of mispredicted
-// branches.
-//
-// the fast table provides a hybrid solution. the current incarnation uses the
-// first (upper case) character to make a first selection. the last character
-// is permuted and used as key for the smaller table. in practice, it should
-// effectively function as a one-level radix trie without branching.
-//
-// the permutation used is the following.
-//  1. use the last character as one always exists, token length is available,
-//     is very likely alphanumeric and likely does not reoccur too often for
-//     records starting with the same alphabetic character. this will provide
-//     a unique key for e.g. MB, MD, MF MG, MR, MX and e.g. NSEC, NSEC3.
-//  2. multiply the character by a given number to get a reasonably good
-//     distribution.
-//  3. increment the character by the length of the identifier to ensure
-//     unique keys for identifiers that begin and end with the same
-//     characters. e.g. A and AAAA.
-typedef struct zone_fast_table zone_fast_table_t;
-struct zone_fast_table {
-  uint8_t keys[16];
-  const zone_symbol_t *symbols[16];
-};
-
 /**
  * @brief Type of value defined by field
  *
@@ -326,6 +298,7 @@ typedef enum {
 #define ZONE_CAA_TAG (1u << 12)
 /** @} */
 
+// FIXME: drop rdata_info, just use field_info
 typedef struct zone_rdata_info zone_rdata_info_t;
 struct zone_rdata_info {
   zone_string_t name;
@@ -351,8 +324,7 @@ typedef struct zone_rdata_info zone_field_info_t;
 
 typedef struct zone_type_info zone_type_info_t;
 struct zone_type_info {
-  zone_string_t name;
-  uint16_t code;
+  zone_symbol_t name;
   uint32_t options;
   struct {
     size_t length;
@@ -508,7 +480,7 @@ struct zone_name {
 // invoked for each record (host order). header (owner, type, class and ttl)
 // fields are passed individually for convenience. rdata fields can be visited
 // individually by means of the iterator
-typedef zone_return_t(*zone_add_t)(
+typedef int32_t(*zone_add_t)(
   zone_parser_t *,
   const zone_name_t *, // owner (length + octets)
   uint16_t, // type
@@ -615,7 +587,7 @@ struct zone_parser {
 /**
  * @brief Parse zone file
  */
-ZONE_EXPORT zone_return_t
+ZONE_EXPORT int32_t
 zone_parse(
   zone_parser_t *parser,
   const zone_options_t *options,
@@ -627,7 +599,7 @@ zone_nonnull((1,2,3,4));
 /**
  * @brief Parse zone from string
  */
-ZONE_EXPORT zone_return_t
+ZONE_EXPORT int32_t
 zone_parse_string(
   zone_parser_t *parser,
   const zone_options_t *options,

scripts/hash.c

@@ -0,0 +1,161 @@
+/*
+ * hash.c -- Calculate perfect hash for TYPEs and CLASSes
+ *
+ * Copyright (c) 2023, NLnet Labs. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+#include <stdio.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <string.h>
+#include <inttypes.h>
+
+typedef struct tuple tuple_t;
+struct tuple {
+  char name[16];
+  uint16_t code;
+  bool type;
+};
+
+static const tuple_t types_and_classes[] = {
+  // classes
+  { "IN", 1, false },
+  { "CS", 2, false },
+  { "CH", 3, false },
+  { "HS", 4, false },
+  // types
+  { "A", 1, true },
+  { "NS", 2, true },
+  { "MD", 3, true },
+  { "MF", 4, true },
+  { "CNAME", 5, true },
+  { "SOA", 6, true },
+  { "MB", 7, true },
+  { "MG", 8, true },
+  { "MR", 9, true },
+  { "NULL", 10, true },
+  { "WKS", 11, true },
+  { "PTR", 12, true },
+  { "HINFO", 13, true },
+  { "MINFO", 14, true },
+  { "MX", 15, true },
+  { "TXT", 16, true },
+  { "RP", 17, true },
+  { "AFSDB", 18, true },
+  { "X25", 19, true },
+  { "ISDN", 20, true },
+  { "RT", 21, true },
+  { "NSAP", 22, true },
+  { "NSAP-PTR", 23, true },
+  { "SIG", 24, true },
+  { "KEY", 25, true },
+  { "PX", 26, true },
+  { "GPOS", 27, true },
+  { "AAAA", 28, true },
+  { "LOC", 29, true },
+  { "NXT", 30, true },
+  { "SRV", 33, true },
+  { "NAPTR", 35, true },
+  { "KX", 36, true },
+  { "CERT", 37, true },
+  { "A6", 38, true },
+  { "DNAME", 39, true },
+  { "APL", 42, true },
+  { "DS", 43, true },
+  { "SSHFP", 44, true },
+  { "IPSECKEY", 45, true },
+  { "RRSIG", 46, true },
+  { "NSEC", 47, true },
+  { "DNSKEY", 48, true },
+  { "DHCID", 49, true },
+  { "NSEC3", 50, true },
+  { "NSEC3PARAM", 51, true },
+  { "TLSA", 52, true },
+  { "SMIMEA", 53, true },
+  { "HIP", 55, true },
+  { "CDS", 59, true },
+  { "CDNSKEY", 60, true },
+  { "OPENPGPKEY", 61, true },
+  { "CSYNC", 62, true },
+  { "ZONEMD", 63, true },
+  { "SVCB", 64, true },
+  { "HTTPS", 65, true },
+  { "SPF", 99, true },
+  { "NID", 104, true },
+  { "L32", 105, true },
+  { "L64", 106, true },
+  { "LP", 107, true },
+  { "EUI48", 108, true },
+  { "EUI64", 109, true },
+  { "URI", 256, true },
+  { "CAA", 257, true },
+  { "AVC", 258, true },
+  { "DLV", 32769, true }
+};
+
+const uint64_t original_magic = 3523216699ull; // original hash from hash.cpp
+
+static uint8_t hash(uint64_t magic, uint64_t value)
+{
+  uint32_t value32 = ((value >> 32) ^ value);
+  return (value32 * magic) >> 32;
+}
+
+static void print_table(uint64_t magic)
+{
+  struct { uint16_t code; bool type; } keys[256];
+  memset(keys, 0, sizeof(keys));
+  const size_t n = sizeof(types_and_classes)/sizeof(types_and_classes[0]);
+  for (size_t i=0; i < n; i++) {
+    uint64_t value;
+    memcpy(&value, types_and_classes[i].name, 8);
+    uint8_t key = hash(magic, value);
+    keys[key].code = types_and_classes[i].code;
+    keys[key].type = types_and_classes[i].type;
+  }
+
+  printf("static const symbol_t *hash_to_symbol[256] = {\n");
+  for (size_t i=0; i < 256; ) {
+    for (size_t j=i+8; i < j; i++) {
+      uint16_t code = keys[i].code;
+      char macro = !code || keys[i].type ? 'T' : 'C';
+      printf("%c(%u), ", macro, code);
+    }
+    printf("\n");
+  }
+  printf("};\n");
+}
+
+int main(int argc, char *argv[])
+{
+  const size_t n = sizeof(types_and_classes)/sizeof(types_and_classes[0]);
+  for (uint64_t magic = original_magic; magic < UINT64_MAX; magic++) {
+    size_t i;
+    uint16_t keys[256] = { 0 };
+    for (i=0; i < n; i++) {
+      uint64_t value;
+      memcpy(&value, types_and_classes[i].name, 8);
+
+      uint8_t key = hash(magic, value);
+      if (keys[key])
+        break;
+      keys[key] = 1;
+    }
+
+    if (i == n) {
+      printf("i: %zu, magic: %" PRIu64 "\n", i, magic);
+      for (i=0; i < n; i++) {
+        uint64_t value;
+        memcpy(&value, types_and_classes[i].name, 8);
+        uint8_t key = hash(magic, value);
+        printf("TYPE_%s: %" PRIu8 " (%" PRIu16 ")\n", types_and_classes[i].name, key, types_and_classes[i].code);
+      }
+      print_table(magic);
+      return 0;
+    }
+  }
+
+  printf("no magic value\n");
+  return 1;
+}

scripts/keys.py

@@ -34,7 +34,7 @@ static zone_really_inline int32_t parse_ip4(
       n = n * 10 + (uint8_t)d;
     } else {
       if (!(p - ps) || p - ps > 3 || n < m[(p - ps)] || n > 255 || o - os > 3)
-        SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), NAME(type));
+        SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), TNAME(type));
       ps = p + 1;
       *o++ = (uint8_t)n;
       if (*p != '.')
@@ -44,7 +44,7 @@ static zone_really_inline int32_t parse_ip4(
   }
 
   if (is_contiguous((uint8_t)*p) || o - os != 4)
-    SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), NAME(type));
+    SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), TNAME(type));
 
   parser->rdata->length += 4;
   return ZONE_IP4;

src/fallback/ip4.h

@@ -39,20 +39,20 @@ static zone_really_inline int32_t scan_name(
         d[1] = (uint8_t)s[2] - '0';
         d[2] = (uint8_t)s[3] - '0';
         if (d[1] > m || d[2] > m)
-          SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), NAME(type));
+          SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), TNAME(type));
         b[0] = d[0] * 100 + d[1] * 10 + d[0];
         b += 1; s += 4;
       }
     } else if (c == '.') {
       if ((b - 1) - l > 63 || (b - 1) - l == 0)
-        SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), NAME(type));
+        SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), TNAME(type));
       l[0] = (uint8_t)((b - 1) - l);
       l = b;
       l[0] = 0;
       b += 1; s += 1;
     } else if (delimiters[c] != token->code) {
       if ((b - 1) - l > 63)
-        SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), NAME(type));
+        SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), TNAME(type));
       l[0] = (uint8_t)((b - 1) - l);
       break;
     } else {
@@ -62,7 +62,7 @@ static zone_really_inline int32_t scan_name(
   }
 
   if (delimiters[(uint8_t)*s] == token->code)
-    SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), NAME(type));
+    SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), TNAME(type));
 
   *length = (size_t)(b - octets);
   return l[0] == 0 ? 0 : ZONE_NAME;
@@ -124,7 +124,7 @@ static zone_really_inline int32_t parse_name(
 
 relative:
   if (n > 255 - parser->file->origin.length)
-    SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), NAME(type));
+    SYNTAX_ERROR(parser, "Invalid %s in %s", NAME(field), TNAME(type));
   memcpy(o+n, parser->file->origin.octets, parser->file->origin.length);
   parser->rdata->length += n + parser->file->origin.length;
   return ZONE_NAME;