Canonicalize IPv6 addresses.

src/url_canon_ip.cc

Index: src/url_canon_ip.cc
===================================================================
--- src/url_canon_ip.cc	(revision 103)
+++ src/url_canon_ip.cc	(working copy)
@@ -31,6 +31,7 @@
 
 #include <stdlib.h>
 
+#include "base/logging.h"
 #include "googleurl/src/url_canon_internal.h"
 
 namespace url_canon {
@@ -54,8 +55,8 @@
 
 template<typename CHAR, typename UCHAR>
 bool DoFindIPv4Components(const CHAR* spec,
-                        const url_parse::Component& host,
-                        url_parse::Component components[4]) {
+                          const url_parse::Component& host,
+                          url_parse::Component components[4]) {
   int cur_component = 0;  // Index of the component we're working on.
   int cur_component_begin = host.begin;  // Start of the current component.
   int end = host.end();
@@ -179,11 +180,12 @@
   out_host->len = output->length() - out_host->begin;
 }
 
-template<typename CHAR, typename UCHAR>
-bool DoCanonicalizeIPv4Address(const CHAR* spec,
-                               const url_parse::Component& host,
-                               CanonOutput* output,
-                               url_parse::Component* out_host) {
+// Converts an IPv4 address to a 32-bit number (network byte order), returning
+// true on success. False means that the input is not a valid IPv4 address.
+template<typename CHAR>
+bool IPv4AddressToNumber(const CHAR* spec,
+                         const url_parse::Component& host,
+                         unsigned char address[4]) {
   // The identified components. Not all may exist.
   url_parse::Component components[4];
   if (!FindIPv4Components(spec, host, components))
@@ -203,7 +205,6 @@
   }
 
   // Use that sequence of numbers to fill out the 4-component IP address.
-  unsigned char address[4];
 
   // ...first fill all but the last component by truncating to one byte.
   for (int i = 0; i < existing_components - 1; i++)
@@ -220,68 +221,382 @@
     address[2] = (last_value & 0x0000FF00) >> 8;
   address[3] = last_value & 0xFF;
 
-  AppendIPv4Address(address, output, out_host);
   return true;
 }
 
-// This function does NO canonicalization.  It does _some_ validation
-// and then copies the component as is to the output.
-// TODO: Actual canonicalization!
 template<typename CHAR, typename UCHAR>
-bool DoCanonicalizeIPv6Address(const CHAR* spec,
+bool DoCanonicalizeIPv4Address(const CHAR* spec,
                                const url_parse::Component& host,
                                CanonOutput* output,
                                url_parse::Component* out_host) {
+  unsigned char address[4];
+  if (!IPv4AddressToNumber<CHAR>(spec, host, address))
+    return false;
+
+  AppendIPv4Address(address, output, out_host);
+  return true;
+}
+
+// Helper class that describes the main components of an IPv6 input string.
+// See the following examples to understand how it breaks up an input string:
+//
+// [Example 1]: input = "[::aa:bb]"
+//  ==> num_hex_components = 2
+//  ==> hex_components[0] = Component(3,2) "aa"
+//  ==> hex_components[1] = Component(6,2) "bb"
+//  ==> index_of_contraction = 0
+//  ==> ipv4_component = Component(0, -1)
+//
+// [Example 2]: input = "[1:2::3:4:5]"
+//  ==> num_hex_components = 5
+//  ==> hex_components[0] = Component(1,1) "1"
+//  ==> hex_components[1] = Component(3,1) "2"
+//  ==> hex_components[2] = Component(6,1) "3"
+//  ==> hex_components[3] = Component(8,1) "4"
+//  ==> hex_components[4] = Component(10,1) "5"
+//  ==> index_of_contraction = 2
+//  ==> ipv4_component = Component(0, -1)
+//
+// [Example 3]: input = "[::ffff:192.168.0.1]"
+//  ==> num_hex_components = 1
+//  ==> hex_components[0] = Component(3,4) "ffff"
+//  ==> index_of_contraction = 0
+//  ==> ipv4_component = Component(8, 11) "192.168.0.1"
+//
+// [Example 4]: input = "[1::]"
+//  ==> num_hex_components = 1
+//  ==> hex_components[0] = Component(1,1) "1"
+//  ==> index_of_contraction = 1
+//  ==> ipv4_component = Component(0, -1)
+//
+// [Example 5]: input = "[::192.168.0.1]"
+//  ==> num_hex_components = 0
+//  ==> index_of_contraction = 0
+//  ==> ipv4_component = Component(8, 11) "192.168.0.1"
+//
+struct IPv6Parsed {
+  // Zero-out the parse information.
+  void reset() {
+    num_hex_components = 0;
+    index_of_contraction = -1;
+    ipv4_component.reset();
+  }
+
+  // There can be up to 8 hex components (colon separated) in the literal.
+  url_parse::Component hex_components[8];
+
+  // The count of hex components present. Ranges from [0,8].
+  int num_hex_components;
+
+  // The index of the hex component that the "::" contraction precedes, or
+  // -1 if there is no contraction.
+  int index_of_contraction;
+
+  // The range of characters which are an IPv4 literal.
+  url_parse::Component ipv4_component;
+};
+
+// Parse the IPv6 input string. If parsing succeeded returns true and fills
+// |parsed| with the information. If parsing failed (because the input is
+// invalid) returns false.
+template<typename CHAR, typename UCHAR>
+bool DoParseIPv6(const CHAR* spec,
+                 const url_parse::Component& host,
+                 IPv6Parsed* parsed) {
+  // Zero-out the info.
+  parsed->reset();
+
+  if (!host.is_nonempty())
+    return false;
+
+  // The index for start and end of address range (no brackets).
+  int begin = host.begin;
+  int end = host.end();
+
+  int cur_component_begin = begin;  // Start of the current component.
+
+  // Scan through the input, searching for hex components, "::" contractions,
+  // and IPv4 components.
+  for (int i = begin; /* i <= end */; i++) {
+    bool is_colon = spec[i] == ':';
+    bool is_contraction = is_colon && i < end - 1 && spec[i + 1] == ':';
+
+    // We reached the end of the current component if we encounter a colon
+    // (separator between hex components, or start of a contraction), or end of
+    // input.
+    if (is_colon || i == end) {
+      int component_len = i - cur_component_begin;
+
+      // A component should not have more than 4 hex digits.
+      if (component_len > 4)
+        return false;
+
+      // Don't allow empty components.
+      if (component_len == 0) {
+        // The exception is when contractions appear at beginning of the
+        // input or at the end of the input.
+        if (!((is_contraction && i == begin) || (i == end &&
+            parsed->index_of_contraction == parsed->num_hex_components)))
+          return false;
+      }
+
+      // Add the hex component we just found to running list.
+      if (component_len > 0) {
+        // Can't have more than 8 components!
+        if (parsed->num_hex_components >= 8)
+          return false;
+
+        parsed->hex_components[parsed->num_hex_components++] =
+            url_parse::Component(cur_component_begin, component_len);
+      }
+    }
+
+    if (i == end)
+      break;  // Reached the end of the input, DONE.
+
+    // We found a "::" contraction.
+    if (is_contraction) {
+      // There can be at most one contraction in the literal.
+      if (parsed->index_of_contraction != -1)
+        return false;
+      parsed->index_of_contraction = parsed->num_hex_components;
+      ++i;  // Consume the colon we peeked.
+    }
+
+    if (is_colon) {
+      // Colons are separators between components, keep track of where the
+      // current component started (after this colon).
+      cur_component_begin = i + 1;
+    } else {
+      if (static_cast<UCHAR>(spec[i]) >= 0x80)
+        return false;  // Not ASCII.
+
+      if (!IsHexChar(static_cast<unsigned char>(spec[i]))) {
+        // Regular components are hex numbers. It is also possible for
+        // a component to be an IPv4 address in dotted form.
+        if (IsIPv4Char(static_cast<unsigned char>(spec[i]))) {
+          // Since IPv4 address can only appear at the end, assume the rest
+          // of the string is an IPv4 address. (We will parse this separately
+          // later).
+          parsed->ipv4_component = url_parse::Component(
+              cur_component_begin, end - cur_component_begin);
+          break;
+        } else {
+          // The character was neither a hex digit, nor an IPv4 character.
+          return false;
+        }
+      }
+    }
+  }
+
+  return true;
+}
+
+// Verifies the parsed IPv6 information, checking that the various components
+// add up to the right number of bits (hex components are 16 bits, while
+// embedded IPv4 formats are 32 bits, and contractions are placeholdes for
+// 16 or more bits). Returns true if sizes match up, false otherwise. On
+// success writes the length of the contraction (if any) to
+// |out_num_bytes_of_contraction|.
+bool CheckIPv6ComponentsSize(const IPv6Parsed& parsed,
+                             int* out_num_bytes_of_contraction) {
+  // Each group of four hex digits contributes 16 bits.
+  int num_bytes_without_contraction = parsed.num_hex_components * 2;
+
+  // If an IPv4 address was embedded at the end, it contributes 32 bits.
+  if (parsed.ipv4_component.is_valid())
+    num_bytes_without_contraction += 4;
+
+  // If there was a "::" contraction, its size is going to be:
+  // MAX([16bits], [128bits] - num_bytes_without_contraction).
+  int num_bytes_of_contraction = 0;
+  if (parsed.index_of_contraction != -1) {
+    num_bytes_of_contraction = 16 - num_bytes_without_contraction;
+    if (num_bytes_of_contraction < 2)
+      num_bytes_of_contraction = 2;
+  }
+
+  // Check that the numbers add up.
+  if (num_bytes_without_contraction + num_bytes_of_contraction != 16)
+    return false;
+
+  *out_num_bytes_of_contraction = num_bytes_of_contraction;
+  return true;
+}
+
+// Converts a hex comonent into a number. This cannot fail since the caller has
+// already verified that each character in the string was a hex digit, and
+// that there were no more than 4 characters.
+template<typename CHAR>
+uint16_t IPv6HexComponentToNumber(const CHAR* spec,
+                                  const url_parse::Component& component) {
+  DCHECK(component.len <= 4);
+
+  // Copy the hex string into a C-string.
+  char buf[5];
+  for (int i = 0; i < component.len; ++i)
+    buf[i] = static_cast<char>(spec[component.begin + i]);
+  buf[component.len] = '\0';
+
+  // Convert it to a number (overflow is not possible, since with 4 hex
+  // characters we can at most have a 16 bit number).
+  return static_cast<uint16_t>(_strtoui64(buf, NULL, 16));
+}
+
+// Converts an IPv6 address to a 128-bit number (network byte order), returning
+// true on success. False means that the input was not a valid IPv6 address.
+template<typename CHAR, typename UCHAR>
+bool IPv6AddressToNumber(const CHAR* spec,
+                         const url_parse::Component& host,
+                         unsigned char address[16]) {
   // Make sure the component is bounded by '[' and ']'.
   int end = host.end();
   if (!host.is_nonempty() || spec[host.begin] != '[' || spec[end - 1] != ']')
     return false;
 
-  int num_colons = 0;
-  int num_dots = 0;
-  int num_hex = 0;
-  for (int i = host.begin + 1; i < end - 1; i++) {
-    if (static_cast<UCHAR>(spec[i]) >= 0x80)
-      return false;
+  // Exclude the square brackets.
+  url_parse::Component ipv6_comp(host.begin + 1, host.len - 2);
 
-    unsigned char u = static_cast<unsigned char>(spec[i]);
-    if (IsHexChar(u)) {
-      // No block between ':'s can be more than 4 hex characters.
-      if (num_hex > 3)
+  // Parse the IPv6 address -- identify where all the colon separated hex
+  // components are, the "::" contraction, and the embedded IPv4 address.
+  IPv6Parsed ipv6_parsed;
+  if (!DoParseIPv6<CHAR, UCHAR>(spec, ipv6_comp, &ipv6_parsed))
+    return false;
+
+  // Do some basic size checks to make sure that the address doesn't
+  // specify more than 128 bits or fewer than 128 bits. This also resolves
+  // how may zero bytes the "::" contraction represents.
+  int num_bytes_of_contraction;
+  if (!CheckIPv6ComponentsSize(ipv6_parsed, &num_bytes_of_contraction))
+    return false;
+
+  int cur_index_in_address = 0;
+
+  // Loop through each hex components, and contraction in order.
+  for (int i = 0; i <= ipv6_parsed.num_hex_components; ++i) {
+    // Append the contraction if it appears before this component.
+    if (i == ipv6_parsed.index_of_contraction) {
+      for (int j = 0; j < num_bytes_of_contraction; ++j)
+        address[cur_index_in_address++] = 0;
+    }
+    // Append the hex component's value.
+    if (i != ipv6_parsed.num_hex_components) {
+      // Get the 16-bit value for this hex component.
+      uint16_t number = IPv6HexComponentToNumber<CHAR>(
+          spec, ipv6_parsed.hex_components[i]);
+      // Append to |address|, in network byte order.
+      address[cur_index_in_address++] = (number & 0xFF00) >> 8;
+      address[cur_index_in_address++] = (number & 0x00FF);
+    }
+  }
+
+  // If there was an IPv4 section, convert it into a 32-bit number and append
+  // it to |address|.
+  if (ipv6_parsed.ipv4_component.is_valid()) {
+    // We only allow the embedded IPv4 syntax to be used for "compat" and
+    // "mapped" formats:
+    //     "compat" ==>  0:0:0:0:0:ffff:<IPv4-literal>
+    //     "mapped" ==>  0:0:0:0:0:0000:<IPv4-literal>
+    for (int j = 0; j < 10; ++j) {
+      if (address[j] != 0)
         return false;
-      num_hex++;
-    } else if (u == ':') {
-      // No ':'s can appear after '.'s have appeared and there can be no
-      // more than 7 ':'s separating the 8 hex shorts.
-      if (num_dots > 0 || num_colons > 6)
-        return false;
-      num_colons++;
-      num_hex = 0;
-    } else if (u == '.') {
-      // No hex chars between ':'s is fine (signifies successive
-      // zeroed shorts concatentated, but can only be used once).  Not
-      // valid for embedded IPv4 addresses, however.
-      if (num_hex < 1)
-        return false;
-      num_dots++;
-      num_hex = 0;
-    } else {
-      // Invalid characters for an IPv6 address.
+    }
+    if (!((address[10] == 0 && address[11] == 0) ||
+          (address[10] == 0xFF && address[11] == 0xFF)))
       return false;
+
+    // Append the 32-bit number to |address|.
+    if (!IPv4AddressToNumber(spec,
+                             ipv6_parsed.ipv4_component,
+                             &address[cur_index_in_address]))
+      return false;
+  }
+
+  return true;
+}
+
+// Searches for the longest sequence of zeros in |address|, and writes the
+// range into |contraction_range|. The run of zeros must be at least 16 bits,
+// and if there is a tie the first is chosen.
+void ChooseIPv6ContractionRange(const unsigned char address[16],
+                                url_parse::Component* contraction_range) {
+  // The longest run of zeros in |address| seen so far.
+  url_parse::Component max_range;
+
+  // The current run of zeros in |address| being iterated over.
+  url_parse::Component cur_range;
+
+  for (int i = 0; i < 16; i += 2) {
+    // Test for 16 bits worth of zero.
+    bool is_zero = (address[i] == 0 && address[i + 1] == 0);
+
+    if (is_zero) {
+      // Add the zero to the current range (or start a new one).
+      if (!cur_range.is_valid())
+        cur_range = url_parse::Component(i, 0);
+      cur_range.len += 2;
     }
+
+    if (!is_zero || i == 14) {
+      // Just completed a run of zeros. If the run is greater than 16 bits,
+      // it is a candidate for the contraction.
+      if (cur_range.len > 2 && cur_range.len > max_range.len) {
+        max_range = cur_range;
+      }
+      cur_range.reset();
+    }
   }
-  if (num_colons < 2)
+  *contraction_range = max_range;
+}
+
+template<typename CHAR, typename UCHAR>
+bool DoCanonicalizeIPv6Address(const CHAR* spec,
+                               const url_parse::Component& host,
+                               CanonOutput* output,
+                               url_parse::Component* out_host) {
+  // Turn the IP address into a 128 bit number.
+  unsigned char address[16];
+  if (!IPv6AddressToNumber<CHAR, UCHAR>(spec, host, address))
     return false;
-  if (num_dots != 0 && num_dots != 3)
-    return false;
 
-  // This passed all the checks thus far, so just copy input to output.
-  // NOTE: It may still be invalid, and it's definitely not canonicalized.
-  // TODO: Actually canonicalize.
   out_host->begin = output->length();
-  for (int i = host.begin; i < end; i++)
-    output->push_back(static_cast<char>(spec[i]));
+  output->push_back('[');
+
+  // We will now output the address according to the rules in:
+  // http://tools.ietf.org/html/draft-kawamura-ipv6-text-representation-01#section-4
+
+  // Start by finding where to place the "::" contraction (if any).
+  url_parse::Component contraction_range;
+  ChooseIPv6ContractionRange(address, &contraction_range);
+
+  for (int i = 0; i < 16;) {
+    if (i == contraction_range.begin && contraction_range.len > 0) {
+      // Jump over the contraction.
+      if (i == 0)
+        output->push_back(':');
+      output->push_back(':');
+      i = contraction_range.end();
+    } else {
+      // Consume the next 16 bits from |address|.
+      int x = address[i] << 8 | address[i + 1];
+
+      i += 2;
+
+      // Stringify the 16 bit number (at most requires 4 hex digits).
+      char str[5];
+      _itoa_s(x, str, 16);
+      for (int ch = 0; str[ch] != 0; ++ch)
+        output->push_back(str[ch]);
+
+      // Put a colon after each number, except the last.
+      if (i < 16)
+        output->push_back(':');
+    }
+  }
+
+  output->push_back(']');
   out_host->len = output->length() - out_host->begin;
+
   return true;
 }