Index: url/url_canon_ip.cc |
=================================================================== |
--- url/url_canon_ip.cc (revision 0) |
+++ url/url_canon_ip.cc (revision 0) |
@@ -0,0 +1,730 @@ |
+// Copyright 2009, Google Inc. |
+// All rights reserved. |
+// |
+// Redistribution and use in source and binary forms, with or without |
+// modification, are permitted provided that the following conditions are |
+// met: |
+// |
+// * Redistributions of source code must retain the above copyright |
+// notice, this list of conditions and the following disclaimer. |
+// * Redistributions in binary form must reproduce the above |
+// copyright notice, this list of conditions and the following disclaimer |
+// in the documentation and/or other materials provided with the |
+// distribution. |
+// * Neither the name of Google Inc. nor the names of its |
+// contributors may be used to endorse or promote products derived from |
+// this software without specific prior written permission. |
+// |
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
+ |
+#include "googleurl/src/url_canon_ip.h" |
+ |
+#include <stdlib.h> |
+ |
+#include "base/basictypes.h" |
+#include "base/logging.h" |
+#include "googleurl/src/url_canon_internal.h" |
+ |
+namespace url_canon { |
+ |
+namespace { |
+ |
+// Converts one of the character types that represent a numerical base to the |
+// corresponding base. |
+int BaseForType(SharedCharTypes type) { |
+ switch (type) { |
+ case CHAR_HEX: |
+ return 16; |
+ case CHAR_DEC: |
+ return 10; |
+ case CHAR_OCT: |
+ return 8; |
+ default: |
+ return 0; |
+ } |
+} |
+ |
+template<typename CHAR, typename UCHAR> |
+bool DoFindIPv4Components(const CHAR* spec, |
+ const url_parse::Component& host, |
+ url_parse::Component components[4]) { |
+ if (!host.is_nonempty()) |
+ return false; |
+ |
+ 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(); |
+ for (int i = host.begin; /* nothing */; i++) { |
+ if (i >= end || spec[i] == '.') { |
+ // Found the end of the current component. |
+ int component_len = i - cur_component_begin; |
+ components[cur_component] = |
+ url_parse::Component(cur_component_begin, component_len); |
+ |
+ // The next component starts after the dot. |
+ cur_component_begin = i + 1; |
+ cur_component++; |
+ |
+ // Don't allow empty components (two dots in a row), except we may |
+ // allow an empty component at the end (this would indicate that the |
+ // input ends in a dot). We also want to error if the component is |
+ // empty and it's the only component (cur_component == 1). |
+ if (component_len == 0 && (i < end || cur_component == 1)) |
+ return false; |
+ |
+ if (i >= end) |
+ break; // End of the input. |
+ |
+ if (cur_component == 4) { |
+ // Anything else after the 4th component is an error unless it is a |
+ // dot that would otherwise be treated as the end of input. |
+ if (spec[i] == '.' && i + 1 == end) |
+ break; |
+ return false; |
+ } |
+ } else if (static_cast<UCHAR>(spec[i]) >= 0x80 || |
+ !IsIPv4Char(static_cast<unsigned char>(spec[i]))) { |
+ // Invalid character for an IPv4 address. |
+ return false; |
+ } |
+ } |
+ |
+ // Fill in any unused components. |
+ while (cur_component < 4) |
+ components[cur_component++] = url_parse::Component(); |
+ return true; |
+} |
+ |
+// Converts an IPv4 component to a 32-bit number, while checking for overflow. |
+// |
+// Possible return values: |
+// - IPV4 - The number was valid, and did not overflow. |
+// - BROKEN - The input was numeric, but too large for a 32-bit field. |
+// - NEUTRAL - Input was not numeric. |
+// |
+// The input is assumed to be ASCII. FindIPv4Components should have stripped |
+// out any input that is greater than 7 bits. The components are assumed |
+// to be non-empty. |
+template<typename CHAR> |
+CanonHostInfo::Family IPv4ComponentToNumber( |
+ const CHAR* spec, |
+ const url_parse::Component& component, |
+ uint32* number) { |
+ // Figure out the base |
+ SharedCharTypes base; |
+ int base_prefix_len = 0; // Size of the prefix for this base. |
+ if (spec[component.begin] == '0') { |
+ // Either hex or dec, or a standalone zero. |
+ if (component.len == 1) { |
+ base = CHAR_DEC; |
+ } else if (spec[component.begin + 1] == 'X' || |
+ spec[component.begin + 1] == 'x') { |
+ base = CHAR_HEX; |
+ base_prefix_len = 2; |
+ } else { |
+ base = CHAR_OCT; |
+ base_prefix_len = 1; |
+ } |
+ } else { |
+ base = CHAR_DEC; |
+ } |
+ |
+ // Extend the prefix to consume all leading zeros. |
+ while (base_prefix_len < component.len && |
+ spec[component.begin + base_prefix_len] == '0') |
+ base_prefix_len++; |
+ |
+ // Put the component, minus any base prefix, into a NULL-terminated buffer so |
+ // we can call the standard library. Because leading zeros have already been |
+ // discarded, filling the entire buffer is guaranteed to trigger the 32-bit |
+ // overflow check. |
+ const int kMaxComponentLen = 16; |
+ char buf[kMaxComponentLen + 1]; // digits + '\0' |
+ int dest_i = 0; |
+ for (int i = component.begin + base_prefix_len; i < component.end(); i++) { |
+ // We know the input is 7-bit, so convert to narrow (if this is the wide |
+ // version of the template) by casting. |
+ char input = static_cast<char>(spec[i]); |
+ |
+ // Validate that this character is OK for the given base. |
+ if (!IsCharOfType(input, base)) |
+ return CanonHostInfo::NEUTRAL; |
+ |
+ // Fill the buffer, if there's space remaining. This check allows us to |
+ // verify that all characters are numeric, even those that don't fit. |
+ if (dest_i < kMaxComponentLen) |
+ buf[dest_i++] = input; |
+ } |
+ |
+ buf[dest_i] = '\0'; |
+ |
+ // Use the 64-bit strtoi so we get a big number (no hex, decimal, or octal |
+ // number can overflow a 64-bit number in <= 16 characters). |
+ uint64 num = _strtoui64(buf, NULL, BaseForType(base)); |
+ |
+ // Check for 32-bit overflow. |
+ if (num > kuint32max) |
+ return CanonHostInfo::BROKEN; |
+ |
+ // No overflow. Success! |
+ *number = static_cast<uint32>(num); |
+ return CanonHostInfo::IPV4; |
+} |
+ |
+// See declaration of IPv4AddressToNumber for documentation. |
+template<typename CHAR> |
+CanonHostInfo::Family DoIPv4AddressToNumber(const CHAR* spec, |
+ const url_parse::Component& host, |
+ unsigned char address[4], |
+ int* num_ipv4_components) { |
+ // The identified components. Not all may exist. |
+ url_parse::Component components[4]; |
+ if (!FindIPv4Components(spec, host, components)) |
+ return CanonHostInfo::NEUTRAL; |
+ |
+ // Convert existing components to digits. Values up to |
+ // |existing_components| will be valid. |
+ uint32 component_values[4]; |
+ int existing_components = 0; |
+ |
+ // Set to true if one or more components are BROKEN. BROKEN is only |
+ // returned if all components are IPV4 or BROKEN, so, for example, |
+ // 12345678912345.de returns NEUTRAL rather than broken. |
+ bool broken = false; |
+ for (int i = 0; i < 4; i++) { |
+ if (components[i].len <= 0) |
+ continue; |
+ CanonHostInfo::Family family = IPv4ComponentToNumber( |
+ spec, components[i], &component_values[existing_components]); |
+ |
+ if (family == CanonHostInfo::BROKEN) { |
+ broken = true; |
+ } else if (family != CanonHostInfo::IPV4) { |
+ // Stop if we hit a non-BROKEN invalid non-empty component. |
+ return family; |
+ } |
+ |
+ existing_components++; |
+ } |
+ |
+ if (broken) |
+ return CanonHostInfo::BROKEN; |
+ |
+ // Use that sequence of numbers to fill out the 4-component IP address. |
+ |
+ // First, process all components but the last, while making sure each fits |
+ // within an 8-bit field. |
+ for (int i = 0; i < existing_components - 1; i++) { |
+ if (component_values[i] > kuint8max) |
+ return CanonHostInfo::BROKEN; |
+ address[i] = static_cast<unsigned char>(component_values[i]); |
+ } |
+ |
+ // Next, consume the last component to fill in the remaining bytes. |
+ uint32 last_value = component_values[existing_components - 1]; |
+ for (int i = 3; i >= existing_components - 1; i--) { |
+ address[i] = static_cast<unsigned char>(last_value); |
+ last_value >>= 8; |
+ } |
+ |
+ // If the last component has residual bits, report overflow. |
+ if (last_value != 0) |
+ return CanonHostInfo::BROKEN; |
+ |
+ // Tell the caller how many components we saw. |
+ *num_ipv4_components = existing_components; |
+ |
+ // Success! |
+ return CanonHostInfo::IPV4; |
+} |
+ |
+// Return true if we've made a final IPV4/BROKEN decision, false if the result |
+// is NEUTRAL, and we could use a second opinion. |
+template<typename CHAR, typename UCHAR> |
+bool DoCanonicalizeIPv4Address(const CHAR* spec, |
+ const url_parse::Component& host, |
+ CanonOutput* output, |
+ CanonHostInfo* host_info) { |
+ host_info->family = IPv4AddressToNumber( |
+ spec, host, host_info->address, &host_info->num_ipv4_components); |
+ |
+ switch (host_info->family) { |
+ case CanonHostInfo::IPV4: |
+ // Definitely an IPv4 address. |
+ host_info->out_host.begin = output->length(); |
+ AppendIPv4Address(host_info->address, output); |
+ host_info->out_host.len = output->length() - host_info->out_host.begin; |
+ return true; |
+ case CanonHostInfo::BROKEN: |
+ // Definitely broken. |
+ return true; |
+ default: |
+ // Could be IPv6 or a hostname. |
+ return false; |
+ } |
+} |
+ |
+// 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 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>(_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 DoIPv6AddressToNumber(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; |
+ |
+ // Exclude the square brackets. |
+ url_parse::Component ipv6_comp(host.begin + 1, host.len - 2); |
+ |
+ // 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 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()) { |
+ // Append the 32-bit number to |address|. |
+ int ignored_num_ipv4_components; |
+ if (CanonHostInfo::IPV4 != |
+ IPv4AddressToNumber(spec, |
+ ipv6_parsed.ipv4_component, |
+ &address[cur_index_in_address], |
+ &ignored_num_ipv4_components)) |
+ 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(); |
+ } |
+ } |
+ *contraction_range = max_range; |
+} |
+ |
+// Return true if we've made a final IPV6/BROKEN decision, false if the result |
+// is NEUTRAL, and we could use a second opinion. |
+template<typename CHAR, typename UCHAR> |
+bool DoCanonicalizeIPv6Address(const CHAR* spec, |
+ const url_parse::Component& host, |
+ CanonOutput* output, |
+ CanonHostInfo* host_info) { |
+ // Turn the IP address into a 128 bit number. |
+ if (!IPv6AddressToNumber(spec, host, host_info->address)) { |
+ // If it's not an IPv6 address, scan for characters that should *only* |
+ // exist in an IPv6 address. |
+ for (int i = host.begin; i < host.end(); i++) { |
+ switch (spec[i]) { |
+ case '[': |
+ case ']': |
+ case ':': |
+ host_info->family = CanonHostInfo::BROKEN; |
+ return true; |
+ } |
+ } |
+ |
+ // No invalid characters. Could still be IPv4 or a hostname. |
+ host_info->family = CanonHostInfo::NEUTRAL; |
+ return false; |
+ } |
+ |
+ host_info->out_host.begin = output->length(); |
+ output->push_back('['); |
+ AppendIPv6Address(host_info->address, output); |
+ output->push_back(']'); |
+ host_info->out_host.len = output->length() - host_info->out_host.begin; |
+ |
+ host_info->family = CanonHostInfo::IPV6; |
+ return true; |
+} |
+ |
+} // namespace |
+ |
+void AppendIPv4Address(const unsigned char address[4], CanonOutput* output) { |
+ for (int i = 0; i < 4; i++) { |
+ char str[16]; |
+ _itoa_s(address[i], str, 10); |
+ |
+ for (int ch = 0; str[ch] != 0; ch++) |
+ output->push_back(str[ch]); |
+ |
+ if (i != 3) |
+ output->push_back('.'); |
+ } |
+} |
+ |
+void AppendIPv6Address(const unsigned char address[16], CanonOutput* output) { |
+ // We will 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 <= 14;) { |
+ // We check 2 bytes at a time, from bytes (0, 1) to (14, 15), inclusive. |
+ DCHECK(i % 2 == 0); |
+ 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(':'); |
+ } |
+ } |
+} |
+ |
+bool FindIPv4Components(const char* spec, |
+ const url_parse::Component& host, |
+ url_parse::Component components[4]) { |
+ return DoFindIPv4Components<char, unsigned char>(spec, host, components); |
+} |
+ |
+bool FindIPv4Components(const char16* spec, |
+ const url_parse::Component& host, |
+ url_parse::Component components[4]) { |
+ return DoFindIPv4Components<char16, char16>(spec, host, components); |
+} |
+ |
+void CanonicalizeIPAddress(const char* spec, |
+ const url_parse::Component& host, |
+ CanonOutput* output, |
+ CanonHostInfo* host_info) { |
+ if (DoCanonicalizeIPv4Address<char, unsigned char>( |
+ spec, host, output, host_info)) |
+ return; |
+ if (DoCanonicalizeIPv6Address<char, unsigned char>( |
+ spec, host, output, host_info)) |
+ return; |
+} |
+ |
+void CanonicalizeIPAddress(const char16* spec, |
+ const url_parse::Component& host, |
+ CanonOutput* output, |
+ CanonHostInfo* host_info) { |
+ if (DoCanonicalizeIPv4Address<char16, char16>( |
+ spec, host, output, host_info)) |
+ return; |
+ if (DoCanonicalizeIPv6Address<char16, char16>( |
+ spec, host, output, host_info)) |
+ return; |
+} |
+ |
+CanonHostInfo::Family IPv4AddressToNumber(const char* spec, |
+ const url_parse::Component& host, |
+ unsigned char address[4], |
+ int* num_ipv4_components) { |
+ return DoIPv4AddressToNumber<char>(spec, host, address, num_ipv4_components); |
+} |
+ |
+CanonHostInfo::Family IPv4AddressToNumber(const char16* spec, |
+ const url_parse::Component& host, |
+ unsigned char address[4], |
+ int* num_ipv4_components) { |
+ return DoIPv4AddressToNumber<char16>( |
+ spec, host, address, num_ipv4_components); |
+} |
+ |
+bool IPv6AddressToNumber(const char* spec, |
+ const url_parse::Component& host, |
+ unsigned char address[16]) { |
+ return DoIPv6AddressToNumber<char, unsigned char>(spec, host, address); |
+} |
+ |
+bool IPv6AddressToNumber(const char16* spec, |
+ const url_parse::Component& host, |
+ unsigned char address[16]) { |
+ return DoIPv6AddressToNumber<char16, char16>(spec, host, address); |
+} |
+ |
+} // namespace url_canon |
Property changes on: url/url_canon_ip.cc |
___________________________________________________________________ |
Added: svn:eol-style |
+ LF |