// Copyright (c) 2009 The Chromium Authors. All rights reserved. // Use of this source code is governed by a BSD-style license that can be // found in the LICENSE file. #include "chrome/browser/net/url_fixer_upper.h" #include #include "base/file_util.h" #include "base/logging.h" #include "base/string_util.h" #include "chrome/common/url_constants.h" #include "googleurl/src/gurl.h" #include "googleurl/src/url_file.h" #include "googleurl/src/url_parse.h" #include "googleurl/src/url_util.h" #include "net/base/escape.h" #include "net/base/net_util.h" #include "net/base/registry_controlled_domain.h" using std::string; using std::wstring; const char* URLFixerUpper::home_directory_override = NULL; namespace { // TODO(estade): Remove these ugly, ugly functions. They are only used in // SegmentURL. A url_parse::Parsed object keeps track of a bunch of indices into // a url string, and these need to be updated when the URL is converted from // UTF8 to UTF16. Instead of this after-the-fact adjustment, we should parse it // in the correct string format to begin with. url_parse::Component UTF8ComponentToWideComponent( const string& text_utf8, const url_parse::Component& component_utf8) { if (component_utf8.len == -1) return url_parse::Component(); string before_component_string = text_utf8.substr(0, component_utf8.begin); string component_string = text_utf8.substr(component_utf8.begin, component_utf8.len); wstring before_component_string_w = UTF8ToWide(before_component_string); wstring component_string_w = UTF8ToWide(component_string); url_parse::Component component_w(before_component_string_w.length(), component_string_w.length()); return component_w; } void UTF8PartsToWideParts(const string& text_utf8, const url_parse::Parsed& parts_utf8, url_parse::Parsed* parts) { if (IsStringASCII(text_utf8)) { *parts = parts_utf8; return; } parts->scheme = UTF8ComponentToWideComponent(text_utf8, parts_utf8.scheme); parts ->username = UTF8ComponentToWideComponent(text_utf8, parts_utf8.username); parts->password = UTF8ComponentToWideComponent(text_utf8, parts_utf8.password); parts->host = UTF8ComponentToWideComponent(text_utf8, parts_utf8.host); parts->port = UTF8ComponentToWideComponent(text_utf8, parts_utf8.port); parts->path = UTF8ComponentToWideComponent(text_utf8, parts_utf8.path); parts->query = UTF8ComponentToWideComponent(text_utf8, parts_utf8.query); parts->ref = UTF8ComponentToWideComponent(text_utf8, parts_utf8.ref); } TrimPositions TrimWhitespaceUTF8(const std::string& input, TrimPositions positions, std::string* output) { // This implementation is not so fast since it converts the text encoding // twice. Please feel free to file a bug if this function hurts the // performance of Chrome. DCHECK(IsStringUTF8(input)); std::wstring input_wide = UTF8ToWide(input); std::wstring output_wide; TrimPositions result = TrimWhitespace(input_wide, positions, &output_wide); *output = WideToUTF8(output_wide); return result; } } // namespace // does some basic fixes for input that we want to test for file-ness static void PrepareStringForFileOps(const FilePath& text, FilePath::StringType* output) { #if defined(OS_WIN) TrimWhitespace(text.value(), TRIM_ALL, output); replace(output->begin(), output->end(), '/', '\\'); #else TrimWhitespaceUTF8(text.value(), TRIM_ALL, output); #endif } // Tries to create a full path from |text|. If the result is valid and the // file exists, returns true and sets |full_path| to the result. Otherwise, // returns false and leaves |full_path| unchanged. static bool ValidPathForFile(const FilePath::StringType& text, FilePath* full_path) { FilePath file_path(text); if (!file_util::AbsolutePath(&file_path)) return false; if (!file_util::PathExists(file_path)) return false; *full_path = file_path; return true; } #if defined(OS_POSIX) // Given a path that starts with ~, return a path that starts with an // expanded-out /user/foobar directory. static string FixupHomedir(const string& text) { DCHECK(text.length() > 0 && text[0] == '~'); if (text.length() == 1 || text[1] == '/') { const char* home = getenv("HOME"); if (URLFixerUpper::home_directory_override) home = URLFixerUpper::home_directory_override; // We'll probably break elsewhere if $HOME is undefined, but check here // just in case. if (!home) return text; return home + text.substr(1); } // Otherwise, this is a path like ~foobar/baz, where we must expand to // user foobar's home directory. Officially, we should use getpwent(), // but that is a nasty blocking call. #if defined(OS_MACOSX) static const char kHome[] = "/Users/"; #else static const char kHome[] = "/home/"; #endif return kHome + text.substr(1); } #endif // Tries to create a file: URL from |text| if it looks like a filename, even if // it doesn't resolve as a valid path or to an existing file. Returns a // (possibly invalid) file: URL in |fixed_up_url| for input beginning // with a drive specifier or "\\". Returns the unchanged input in other cases // (including file: URLs: these don't look like filenames). static string FixupPath(const string& text) { DCHECK(!text.empty()); FilePath::StringType filename; #if defined(OS_WIN) FilePath input_path(UTF8ToWide(text)); PrepareStringForFileOps(input_path, &filename); // Fixup Windows-style drive letters, where "C:" gets rewritten to "C|". if (filename.length() > 1 && filename[1] == '|') filename[1] = ':'; #elif defined(OS_POSIX) FilePath input_path(text); PrepareStringForFileOps(input_path, &filename); if (filename.length() > 0 && filename[0] == '~') filename = FixupHomedir(filename); #endif // Here, we know the input looks like a file. GURL file_url = net::FilePathToFileURL(FilePath(filename)); if (file_url.is_valid()) { return WideToUTF8(net::FormatUrl(file_url, std::wstring(), true, UnescapeRule::NORMAL, NULL, NULL, NULL)); } // Invalid file URL, just return the input. return text; } // Checks |domain| to see if a valid TLD is already present. If not, appends // |desired_tld| to the domain, and prepends "www." unless it's already present. static void AddDesiredTLD(const string& desired_tld, string* domain) { if (desired_tld.empty() || domain->empty()) return; // Check the TLD. If the return value is positive, we already have a TLD, so // abort; if the return value is string::npos, there's no valid host (e.g. if // the user pasted in garbage for which HistoryURLProvider is trying to // suggest an exact match), so adding a TLD makes no sense. The only useful // case is where the return value is 0 (there's a valid host with no known // TLD). We disallow unknown registries here so users can input "mail.yahoo" // and hit ctrl-enter to get "www.mail.yahoo.com". const size_t registry_length = net::RegistryControlledDomainService::GetRegistryLength(*domain, false); if (registry_length != 0) return; // Add the suffix at the end of the domain. const size_t domain_length(domain->length()); DCHECK(domain_length > 0); DCHECK(desired_tld[0] != '.'); if ((*domain)[domain_length - 1] != '.') domain->push_back('.'); domain->append(desired_tld); // Now, if the domain begins with "www.", stop. const string prefix("www."); if (domain->compare(0, prefix.length(), prefix) != 0) { // Otherwise, add www. to the beginning of the URL. domain->insert(0, prefix); } } static inline void FixupUsername(const string& text, const url_parse::Component& part, string* url) { if (!part.is_valid()) return; // We don't fix up the username at the moment. url->append(text, part.begin, part.len); // Do not append the trailing '@' because we might need to include the user's // password. FixupURL itself will append the '@' for us. } static inline void FixupPassword(const string& text, const url_parse::Component& part, string* url) { if (!part.is_valid()) return; // We don't fix up the password at the moment. url->append(":"); url->append(text, part.begin, part.len); } static void FixupHost(const string& text, const url_parse::Component& part, bool has_scheme, const string& desired_tld, string* url) { if (!part.is_valid()) return; // Make domain valid. // Strip all leading dots and all but one trailing dot, unless the user only // typed dots, in which case their input is totally invalid and we should just // leave it unchanged. string domain(text, part.begin, part.len); const size_t first_nondot(domain.find_first_not_of('.')); if (first_nondot != string::npos) { domain.erase(0, first_nondot); size_t last_nondot(domain.find_last_not_of('.')); DCHECK(last_nondot != string::npos); last_nondot += 2; // Point at second period in ending string if (last_nondot < domain.length()) domain.erase(last_nondot); } // Add any user-specified TLD, if applicable. AddDesiredTLD(desired_tld, &domain); url->append(domain); } static void FixupPort(const string& text, const url_parse::Component& part, string* url) { if (!part.is_valid()) return; // We don't fix up the port at the moment. url->append(":"); url->append(text, part.begin, part.len); } static inline void FixupPath(const string& text, const url_parse::Component& part, string* url) { if (!part.is_valid() || part.len == 0) { // We should always have a path. url->append("/"); return; } // Append the path as is. url->append(text, part.begin, part.len); } static inline void FixupQuery(const string& text, const url_parse::Component& part, string* url) { if (!part.is_valid()) return; // We don't fix up the query at the moment. url->append("?"); url->append(text, part.begin, part.len); } static inline void FixupRef(const string& text, const url_parse::Component& part, string* url) { if (!part.is_valid()) return; // We don't fix up the ref at the moment. url->append("#"); url->append(text, part.begin, part.len); } static void OffsetComponent(int offset, url_parse::Component* part) { DCHECK(part); if (part->is_valid()) { // Offset the location of this component. part->begin += offset; // This part might not have existed in the original text. if (part->begin < 0) part->reset(); } } static bool HasPort(const std::string& original_text, const url_parse::Component& scheme_component) { // Find the range between the ":" and the "/". size_t port_start = scheme_component.end() + 1; size_t port_end = port_start; while ((port_end < original_text.length()) && !url_parse::IsAuthorityTerminator(original_text[port_end])) ++port_end; if (port_end == port_start) return false; // Scan the range to see if it is entirely digits. for (size_t i = port_start; i < port_end; ++i) { if (!IsAsciiDigit(original_text[i])) return false; } return true; } // Try to extract a valid scheme from the beginning of |text|. // If successful, set |scheme_component| to the text range where the scheme // was located, and fill |canon_scheme| with its canonicalized form. // Otherwise, return false and leave the outputs in an indeterminate state. static bool GetValidScheme(const string &text, url_parse::Component *scheme_component, string *canon_scheme) { // Locate everything up to (but not including) the first ':' if (!url_parse::ExtractScheme(text.data(), static_cast(text.length()), scheme_component)) return false; // Make sure the scheme contains only valid characters, and convert // to lowercase. This also catches IPv6 literals like [::1], because // brackets are not in the whitelist. url_canon::StdStringCanonOutput canon_scheme_output(canon_scheme); url_parse::Component canon_scheme_component; if (!url_canon::CanonicalizeScheme(text.data(), *scheme_component, &canon_scheme_output, &canon_scheme_component)) return false; // Strip the ':', and any trailing buffer space. DCHECK_EQ(0, canon_scheme_component.begin); canon_scheme->erase(canon_scheme_component.len); // We need to fix up the segmentation for "www.example.com:/". For this // case, we guess that schemes with a "." are not actually schemes. if (canon_scheme->find('.') != string::npos) return false; // We need to fix up the segmentation for "www:123/". For this case, we // will add an HTTP scheme later and make the URL parser happy. // TODO(pkasting): Maybe we should try to use GURL's parser for this? if (HasPort(text, *scheme_component)) return false; // Everything checks out. return true; } string URLFixerUpper::SegmentURL(const string& text, url_parse::Parsed* parts) { // Initialize the result. *parts = url_parse::Parsed(); string trimmed; TrimWhitespaceUTF8(text, TRIM_ALL, &trimmed); if (trimmed.empty()) return string(); // Nothing to segment. #if defined(OS_WIN) int trimmed_length = static_cast(trimmed.length()); if (url_parse::DoesBeginWindowsDriveSpec(trimmed.data(), 0, trimmed_length) || url_parse::DoesBeginUNCPath(trimmed.data(), 0, trimmed_length, false)) return "file"; #elif defined(OS_POSIX) if (FilePath::IsSeparator(trimmed.data()[0]) || trimmed.data()[0] == '~') return "file"; #endif // Otherwise, we need to look at things carefully. string scheme; if (!GetValidScheme(text, &parts->scheme, &scheme)) { // Couldn't determine the scheme, so just pick one. parts->scheme.reset(); scheme.assign(StartsWithASCII(text, "ftp.", false) ? chrome::kFtpScheme : chrome::kHttpScheme); } // Not segmenting file schemes or nonstandard schemes. if ((scheme == chrome::kFileScheme) || !url_util::IsStandard(scheme.c_str(), static_cast(scheme.length()), url_parse::Component(0, static_cast(scheme.length())))) return scheme; if (parts->scheme.is_valid()) { // Have the GURL parser do the heavy lifting for us. url_parse::ParseStandardURL(text.data(), static_cast(text.length()), parts); return scheme; } // We need to add a scheme in order for ParseStandardURL to be happy. // Find the first non-whitespace character. string::const_iterator first_nonwhite = text.begin(); while ((first_nonwhite != text.end()) && IsWhitespace(*first_nonwhite)) ++first_nonwhite; // Construct the text to parse by inserting the scheme. string inserted_text(scheme); inserted_text.append("://"); string text_to_parse(text.begin(), first_nonwhite); text_to_parse.append(inserted_text); text_to_parse.append(first_nonwhite, text.end()); // Have the GURL parser do the heavy lifting for us. url_parse::ParseStandardURL(text_to_parse.data(), static_cast(text_to_parse.length()), parts); // Offset the results of the parse to match the original text. const int offset = -static_cast(inserted_text.length()); OffsetComponent(offset, &parts->scheme); OffsetComponent(offset, &parts->username); OffsetComponent(offset, &parts->password); OffsetComponent(offset, &parts->host); OffsetComponent(offset, &parts->port); OffsetComponent(offset, &parts->path); OffsetComponent(offset, &parts->query); OffsetComponent(offset, &parts->ref); return scheme; } string URLFixerUpper::FixupURL(const string& text, const string& desired_tld) { string trimmed; TrimWhitespaceUTF8(text, TRIM_ALL, &trimmed); if (trimmed.empty()) return string(); // Nothing here. // Segment the URL. url_parse::Parsed parts; string scheme(SegmentURL(trimmed, &parts)); // We handle the file scheme separately. if (scheme == "file") return (parts.scheme.is_valid() ? text : FixupPath(text)); // For some schemes whose layouts we understand, we rebuild it. if (url_util::IsStandard(scheme.c_str(), static_cast(scheme.length()), url_parse::Component(0, static_cast(scheme.length())))) { string url(scheme); url.append("://"); // We need to check whether the |username| is valid because it is our // responsibility to append the '@' to delineate the user information from // the host portion of the URL. if (parts.username.is_valid()) { FixupUsername(trimmed, parts.username, &url); FixupPassword(trimmed, parts.password, &url); url.append("@"); } FixupHost(trimmed, parts.host, parts.scheme.is_valid(), desired_tld, &url); FixupPort(trimmed, parts.port, &url); FixupPath(trimmed, parts.path, &url); FixupQuery(trimmed, parts.query, &url); FixupRef(trimmed, parts.ref, &url); return url; } // In the worst-case, we insert a scheme if the URL lacks one. if (!parts.scheme.is_valid()) { string fixed_scheme(scheme); fixed_scheme.append("://"); trimmed.insert(0, fixed_scheme); } return trimmed; } // The rules are different here than for regular fixup, since we need to handle // input like "hello.html" and know to look in the current directory. Regular // fixup will look for cues that it is actually a file path before trying to // figure out what file it is. If our logic doesn't work, we will fall back on // regular fixup. string URLFixerUpper::FixupRelativeFile(const FilePath& base_dir, const FilePath& text) { FilePath old_cur_directory; if (!base_dir.empty()) { // Save the old current directory before we move to the new one. file_util::GetCurrentDirectory(&old_cur_directory); file_util::SetCurrentDirectory(base_dir); } // Allow funny input with extra whitespace and the wrong kind of slashes. FilePath::StringType trimmed; PrepareStringForFileOps(text, &trimmed); bool is_file = true; FilePath full_path; if (!ValidPathForFile(trimmed, &full_path)) { // Not a path as entered, try unescaping it in case the user has // escaped things. We need to go through 8-bit since the escaped values // only represent 8-bit values. #if defined(OS_WIN) std::wstring unescaped = UTF8ToWide(UnescapeURLComponent( WideToUTF8(trimmed), UnescapeRule::SPACES | UnescapeRule::URL_SPECIAL_CHARS)); #elif defined(OS_POSIX) std::string unescaped = UnescapeURLComponent( trimmed, UnescapeRule::SPACES | UnescapeRule::URL_SPECIAL_CHARS); #endif if (!ValidPathForFile(unescaped, &full_path)) is_file = false; } // Put back the current directory if we saved it. if (!base_dir.empty()) { file_util::SetCurrentDirectory(old_cur_directory); } if (is_file) { GURL file_url = net::FilePathToFileURL(full_path); if (file_url.is_valid()) return WideToUTF8(net::FormatUrl(file_url, std::wstring(), true, UnescapeRule::NORMAL, NULL, NULL, NULL)); // Invalid files fall through to regular processing. } // Fall back on regular fixup for this input. #if defined(OS_WIN) string text_utf8 = WideToUTF8(text.value()); #elif defined(OS_POSIX) string text_utf8 = text.value(); #endif return FixupURL(text_utf8, ""); } // Deprecated functions. To be removed when all callers are updated. wstring URLFixerUpper::SegmentURL(const wstring& text, url_parse::Parsed* parts) { string text_utf8 = WideToUTF8(text); url_parse::Parsed parts_utf8; string scheme_utf8 = SegmentURL(text_utf8, &parts_utf8); UTF8PartsToWideParts(text_utf8, parts_utf8, parts); return UTF8ToWide(scheme_utf8); } wstring URLFixerUpper::FixupURL(const wstring& text, const wstring& desired_tld) { return UTF8ToWide(FixupURL(WideToUTF8(text), WideToUTF8(desired_tld))); } wstring URLFixerUpper::FixupRelativeFile(const wstring& base_dir, const wstring& text) { return UTF8ToWide(FixupRelativeFile(FilePath::FromWStringHack(base_dir), FilePath::FromWStringHack(text))); }