mirror of
https://github.com/mozilla/gecko-dev.git
synced 2024-11-02 15:15:23 +00:00
364 lines
8.6 KiB
C++
364 lines
8.6 KiB
C++
/* This Source Code Form is subject to the terms of the Mozilla Public
|
|
* License, v. 2.0. If a copy of the MPL was not distributed with this
|
|
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
|
|
|
|
#include "nsEscape.h"
|
|
#include "nsString.h"
|
|
#include "nsIURI.h"
|
|
#include "nsUrlClassifierUtils.h"
|
|
#include "nsTArray.h"
|
|
#include "nsReadableUtils.h"
|
|
#include "plbase64.h"
|
|
#include "prprf.h"
|
|
|
|
static char int_to_hex_digit(int32_t i)
|
|
{
|
|
NS_ASSERTION((i >= 0) && (i <= 15), "int too big in int_to_hex_digit");
|
|
return static_cast<char>(((i < 10) ? (i + '0') : ((i - 10) + 'A')));
|
|
}
|
|
|
|
static bool
|
|
IsDecimal(const nsACString & num)
|
|
{
|
|
for (uint32_t i = 0; i < num.Length(); i++) {
|
|
if (!isdigit(num[i])) {
|
|
return false;
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
IsHex(const nsACString & num)
|
|
{
|
|
if (num.Length() < 3) {
|
|
return false;
|
|
}
|
|
|
|
if (num[0] != '0' || !(num[1] == 'x' || num[1] == 'X')) {
|
|
return false;
|
|
}
|
|
|
|
for (uint32_t i = 2; i < num.Length(); i++) {
|
|
if (!isxdigit(num[i])) {
|
|
return false;
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
IsOctal(const nsACString & num)
|
|
{
|
|
if (num.Length() < 2) {
|
|
return false;
|
|
}
|
|
|
|
if (num[0] != '0') {
|
|
return false;
|
|
}
|
|
|
|
for (uint32_t i = 1; i < num.Length(); i++) {
|
|
if (!isdigit(num[i]) || num[i] == '8' || num[i] == '9') {
|
|
return false;
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
nsUrlClassifierUtils::nsUrlClassifierUtils() : mEscapeCharmap(nullptr)
|
|
{
|
|
}
|
|
|
|
nsresult
|
|
nsUrlClassifierUtils::Init()
|
|
{
|
|
// Everything but alpha numerics, - and .
|
|
mEscapeCharmap = new Charmap(0xffffffff, 0xfc009fff, 0xf8000001, 0xf8000001,
|
|
0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff);
|
|
if (!mEscapeCharmap)
|
|
return NS_ERROR_OUT_OF_MEMORY;
|
|
return NS_OK;
|
|
}
|
|
|
|
NS_IMPL_ISUPPORTS(nsUrlClassifierUtils, nsIUrlClassifierUtils)
|
|
|
|
/////////////////////////////////////////////////////////////////////////////
|
|
// nsIUrlClassifierUtils
|
|
|
|
NS_IMETHODIMP
|
|
nsUrlClassifierUtils::GetKeyForURI(nsIURI * uri, nsACString & _retval)
|
|
{
|
|
nsCOMPtr<nsIURI> innerURI = NS_GetInnermostURI(uri);
|
|
if (!innerURI)
|
|
innerURI = uri;
|
|
|
|
nsAutoCString host;
|
|
innerURI->GetAsciiHost(host);
|
|
|
|
if (host.IsEmpty()) {
|
|
return NS_ERROR_MALFORMED_URI;
|
|
}
|
|
|
|
nsresult rv = CanonicalizeHostname(host, _retval);
|
|
NS_ENSURE_SUCCESS(rv, rv);
|
|
|
|
nsAutoCString path;
|
|
rv = innerURI->GetPath(path);
|
|
NS_ENSURE_SUCCESS(rv, rv);
|
|
|
|
// strip out anchors
|
|
int32_t ref = path.FindChar('#');
|
|
if (ref != kNotFound)
|
|
path.SetLength(ref);
|
|
|
|
nsAutoCString temp;
|
|
rv = CanonicalizePath(path, temp);
|
|
NS_ENSURE_SUCCESS(rv, rv);
|
|
|
|
_retval.Append(temp);
|
|
|
|
return NS_OK;
|
|
}
|
|
|
|
/////////////////////////////////////////////////////////////////////////////
|
|
// non-interface methods
|
|
|
|
nsresult
|
|
nsUrlClassifierUtils::CanonicalizeHostname(const nsACString & hostname,
|
|
nsACString & _retval)
|
|
{
|
|
nsAutoCString unescaped;
|
|
if (!NS_UnescapeURL(PromiseFlatCString(hostname).get(),
|
|
PromiseFlatCString(hostname).Length(),
|
|
0, unescaped)) {
|
|
unescaped.Assign(hostname);
|
|
}
|
|
|
|
nsAutoCString cleaned;
|
|
CleanupHostname(unescaped, cleaned);
|
|
|
|
nsAutoCString temp;
|
|
ParseIPAddress(cleaned, temp);
|
|
if (!temp.IsEmpty()) {
|
|
cleaned.Assign(temp);
|
|
}
|
|
|
|
ToLowerCase(cleaned);
|
|
SpecialEncode(cleaned, false, _retval);
|
|
|
|
return NS_OK;
|
|
}
|
|
|
|
|
|
nsresult
|
|
nsUrlClassifierUtils::CanonicalizePath(const nsACString & path,
|
|
nsACString & _retval)
|
|
{
|
|
_retval.Truncate();
|
|
|
|
nsAutoCString decodedPath(path);
|
|
nsAutoCString temp;
|
|
while (NS_UnescapeURL(decodedPath.get(), decodedPath.Length(), 0, temp)) {
|
|
decodedPath.Assign(temp);
|
|
temp.Truncate();
|
|
}
|
|
|
|
SpecialEncode(decodedPath, true, _retval);
|
|
// XXX: lowercase the path?
|
|
|
|
return NS_OK;
|
|
}
|
|
|
|
void
|
|
nsUrlClassifierUtils::CleanupHostname(const nsACString & hostname,
|
|
nsACString & _retval)
|
|
{
|
|
_retval.Truncate();
|
|
|
|
const char* curChar = hostname.BeginReading();
|
|
const char* end = hostname.EndReading();
|
|
char lastChar = '\0';
|
|
while (curChar != end) {
|
|
unsigned char c = static_cast<unsigned char>(*curChar);
|
|
if (c == '.' && (lastChar == '\0' || lastChar == '.')) {
|
|
// skip
|
|
} else {
|
|
_retval.Append(*curChar);
|
|
}
|
|
lastChar = c;
|
|
++curChar;
|
|
}
|
|
|
|
// cut off trailing dots
|
|
while (_retval.Length() > 0 && _retval[_retval.Length() - 1] == '.') {
|
|
_retval.SetLength(_retval.Length() - 1);
|
|
}
|
|
}
|
|
|
|
void
|
|
nsUrlClassifierUtils::ParseIPAddress(const nsACString & host,
|
|
nsACString & _retval)
|
|
{
|
|
_retval.Truncate();
|
|
nsACString::const_iterator iter, end;
|
|
host.BeginReading(iter);
|
|
host.EndReading(end);
|
|
|
|
if (host.Length() <= 15) {
|
|
// The Windows resolver allows a 4-part dotted decimal IP address to
|
|
// have a space followed by any old rubbish, so long as the total length
|
|
// of the string doesn't get above 15 characters. So, "10.192.95.89 xy"
|
|
// is resolved to 10.192.95.89.
|
|
// If the string length is greater than 15 characters, e.g.
|
|
// "10.192.95.89 xy.wildcard.example.com", it will be resolved through
|
|
// DNS.
|
|
|
|
if (FindCharInReadable(' ', iter, end)) {
|
|
end = iter;
|
|
}
|
|
}
|
|
|
|
for (host.BeginReading(iter); iter != end; iter++) {
|
|
if (!(isxdigit(*iter) || *iter == 'x' || *iter == 'X' || *iter == '.')) {
|
|
// not an IP
|
|
return;
|
|
}
|
|
}
|
|
|
|
host.BeginReading(iter);
|
|
nsTArray<nsCString> parts;
|
|
ParseString(PromiseFlatCString(Substring(iter, end)), '.', parts);
|
|
if (parts.Length() > 4) {
|
|
return;
|
|
}
|
|
|
|
// If any potentially-octal numbers (start with 0 but not hex) have
|
|
// non-octal digits, no part of the ip can be in octal
|
|
// XXX: this came from the old javascript implementation, is it really
|
|
// supposed to be like this?
|
|
bool allowOctal = true;
|
|
uint32_t i;
|
|
|
|
for (i = 0; i < parts.Length(); i++) {
|
|
const nsCString& part = parts[i];
|
|
if (part[0] == '0') {
|
|
for (uint32_t j = 1; j < part.Length(); j++) {
|
|
if (part[j] == 'x') {
|
|
break;
|
|
}
|
|
if (part[j] == '8' || part[j] == '9') {
|
|
allowOctal = false;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
for (i = 0; i < parts.Length(); i++) {
|
|
nsAutoCString canonical;
|
|
|
|
if (i == parts.Length() - 1) {
|
|
CanonicalNum(parts[i], 5 - parts.Length(), allowOctal, canonical);
|
|
} else {
|
|
CanonicalNum(parts[i], 1, allowOctal, canonical);
|
|
}
|
|
|
|
if (canonical.IsEmpty()) {
|
|
_retval.Truncate();
|
|
return;
|
|
}
|
|
|
|
if (_retval.IsEmpty()) {
|
|
_retval.Assign(canonical);
|
|
} else {
|
|
_retval.Append('.');
|
|
_retval.Append(canonical);
|
|
}
|
|
}
|
|
return;
|
|
}
|
|
|
|
void
|
|
nsUrlClassifierUtils::CanonicalNum(const nsACString& num,
|
|
uint32_t bytes,
|
|
bool allowOctal,
|
|
nsACString& _retval)
|
|
{
|
|
_retval.Truncate();
|
|
|
|
if (num.Length() < 1) {
|
|
return;
|
|
}
|
|
|
|
uint32_t val;
|
|
if (allowOctal && IsOctal(num)) {
|
|
if (PR_sscanf(PromiseFlatCString(num).get(), "%o", &val) != 1) {
|
|
return;
|
|
}
|
|
} else if (IsDecimal(num)) {
|
|
if (PR_sscanf(PromiseFlatCString(num).get(), "%u", &val) != 1) {
|
|
return;
|
|
}
|
|
} else if (IsHex(num)) {
|
|
if (PR_sscanf(PromiseFlatCString(num).get(), num[1] == 'X' ? "0X%x" : "0x%x",
|
|
&val) != 1) {
|
|
return;
|
|
}
|
|
} else {
|
|
return;
|
|
}
|
|
|
|
while (bytes--) {
|
|
char buf[20];
|
|
PR_snprintf(buf, sizeof(buf), "%u", val & 0xff);
|
|
if (_retval.IsEmpty()) {
|
|
_retval.Assign(buf);
|
|
} else {
|
|
_retval = nsDependentCString(buf) + NS_LITERAL_CSTRING(".") + _retval;
|
|
}
|
|
val >>= 8;
|
|
}
|
|
}
|
|
|
|
// This function will encode all "special" characters in typical url
|
|
// encoding, that is %hh where h is a valid hex digit. It will also fold
|
|
// any duplicated slashes.
|
|
bool
|
|
nsUrlClassifierUtils::SpecialEncode(const nsACString & url,
|
|
bool foldSlashes,
|
|
nsACString & _retval)
|
|
{
|
|
bool changed = false;
|
|
const char* curChar = url.BeginReading();
|
|
const char* end = url.EndReading();
|
|
|
|
unsigned char lastChar = '\0';
|
|
while (curChar != end) {
|
|
unsigned char c = static_cast<unsigned char>(*curChar);
|
|
if (ShouldURLEscape(c)) {
|
|
_retval.Append('%');
|
|
_retval.Append(int_to_hex_digit(c / 16));
|
|
_retval.Append(int_to_hex_digit(c % 16));
|
|
|
|
changed = true;
|
|
} else if (foldSlashes && (c == '/' && lastChar == '/')) {
|
|
// skip
|
|
} else {
|
|
_retval.Append(*curChar);
|
|
}
|
|
lastChar = c;
|
|
curChar++;
|
|
}
|
|
return changed;
|
|
}
|
|
|
|
bool
|
|
nsUrlClassifierUtils::ShouldURLEscape(const unsigned char c) const
|
|
{
|
|
return c <= 32 || c == '%' || c >=127;
|
|
}
|