1 // Copyright 2013 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
5 // Canonicalizer functions for working with and resolving relative URLs.
9 #include "base/logging.h"
10 #include "url/url_canon.h"
11 #include "url/url_canon_internal.h"
12 #include "url/url_constants.h"
13 #include "url/url_file.h"
14 #include "url/url_parse_internal.h"
15 #include "url/url_util.h"
16 #include "url/url_util_internal.h"
22 // Firefox does a case-sensitive compare (which is probably wrong--Mozilla bug
23 // 379034), whereas IE is case-insensitive.
25 // We choose to be more permissive like IE. We don't need to worry about
26 // unescaping or anything here: neither IE or Firefox allow this. We also
27 // don't have to worry about invalid scheme characters since we are comparing
28 // against the canonical scheme of the base.
30 // The base URL should always be canonical, therefore it should be ASCII.
31 template<typename CHAR>
32 bool AreSchemesEqual(const char* base,
33 const Component& base_scheme,
35 const Component& cmp_scheme) {
36 if (base_scheme.len != cmp_scheme.len)
38 for (int i = 0; i < base_scheme.len; i++) {
39 // We assume the base is already canonical, so we don't have to
41 if (CanonicalSchemeChar(cmp[cmp_scheme.begin + i]) !=
42 base[base_scheme.begin + i])
50 // Here, we also allow Windows paths to be represented as "/C:/" so we can be
51 // consistent about URL paths beginning with slashes. This function is like
52 // DoesBeginWindowsDrivePath except that it also requires a slash at the
54 template<typename CHAR>
55 bool DoesBeginSlashWindowsDriveSpec(const CHAR* spec, int start_offset,
57 if (start_offset >= spec_len)
59 return IsURLSlash(spec[start_offset]) &&
60 DoesBeginWindowsDriveSpec(spec, start_offset + 1, spec_len);
65 // See IsRelativeURL in the header file for usage.
66 template<typename CHAR>
67 bool DoIsRelativeURL(const char* base,
68 const Parsed& base_parsed,
71 bool is_base_hierarchical,
73 Component* relative_component) {
74 *is_relative = false; // So we can default later to not relative.
76 // Trim whitespace and construct a new range for the substring.
78 TrimURL(url, &begin, &url_len);
79 if (begin >= url_len) {
80 // Empty URLs are relative, but do nothing.
81 if (!is_base_hierarchical) {
82 // Don't allow relative URLs if the base scheme doesn't support it.
85 *relative_component = Component(begin, 0);
91 // We special case paths like "C:\foo" so they can link directly to the
92 // file on Windows (IE compatibility). The security domain stuff should
93 // prevent a link like this from actually being followed if its on a
96 // We treat "C:/foo" as an absolute URL. We can go ahead and treat "/c:/"
97 // as relative, as this will just replace the path when the base scheme
98 // is a file and the answer will still be correct.
100 // We require strict backslashes when detecting UNC since two forward
101 // slashes should be treated a a relative URL with a hostname.
102 if (DoesBeginWindowsDriveSpec(url, begin, url_len) ||
103 DoesBeginUNCPath(url, begin, url_len, true))
107 // See if we've got a scheme, if not, we know this is a relative URL.
108 // BUT, just because we have a scheme, doesn't make it absolute.
109 // "http:foo.html" is a relative URL with path "foo.html". If the scheme is
110 // empty, we treat it as relative (":foo"), like IE does.
112 const bool scheme_is_empty =
113 !ExtractScheme(url, url_len, &scheme) || scheme.len == 0;
114 if (scheme_is_empty) {
115 if (url[begin] == '#') {
116 // |url| is a bare fragment (e.g. "#foo"). This can be resolved against
117 // any base. Fall-through.
118 } else if (!is_base_hierarchical) {
119 // Don't allow relative URLs if the base scheme doesn't support it.
123 *relative_component = MakeRange(begin, url_len);
128 // If the scheme isn't valid, then it's relative.
129 int scheme_end = scheme.end();
130 for (int i = scheme.begin; i < scheme_end; i++) {
131 if (!CanonicalSchemeChar(url[i])) {
132 if (!is_base_hierarchical) {
133 // Don't allow relative URLs if the base scheme doesn't support it.
136 *relative_component = MakeRange(begin, url_len);
142 // If the scheme is not the same, then we can't count it as relative.
143 if (!AreSchemesEqual(base, base_parsed.scheme, url, scheme))
146 // When the scheme that they both share is not hierarchical, treat the
147 // incoming scheme as absolute (this way with the base of "data:foo",
148 // "data:bar" will be reported as absolute.
149 if (!is_base_hierarchical)
152 int colon_offset = scheme.end();
154 // If it's a filesystem URL, the only valid way to make it relative is not to
155 // supply a scheme. There's no equivalent to e.g. http:index.html.
156 if (CompareSchemeComponent(url, scheme, kFileSystemScheme))
159 // ExtractScheme guarantees that the colon immediately follows what it
160 // considers to be the scheme. CountConsecutiveSlashes will handle the
161 // case where the begin offset is the end of the input.
162 int num_slashes = CountConsecutiveSlashes(url, colon_offset + 1, url_len);
164 if (num_slashes == 0 || num_slashes == 1) {
165 // No slashes means it's a relative path like "http:foo.html". One slash
166 // is an absolute path. "http:/home/foo.html"
168 *relative_component = MakeRange(colon_offset + 1, url_len);
172 // Two or more slashes after the scheme we treat as absolute.
176 // Copies all characters in the range [begin, end) of |spec| to the output,
177 // up until and including the last slash. There should be a slash in the
178 // range, if not, nothing will be copied.
180 // For stardard URLs the input should be canonical, but when resolving relative
181 // URLs on a non-standard base (like "data:") the input can be anything.
182 void CopyToLastSlash(const char* spec,
185 CanonOutput* output) {
186 // Find the last slash.
188 for (int i = end - 1; i >= begin; i--) {
189 if (spec[i] == '/' || spec[i] == '\\') {
198 for (int i = begin; i <= last_slash; i++)
199 output->push_back(spec[i]);
202 // Copies a single component from the source to the output. This is used
203 // when resolving relative URLs and a given component is unchanged. Since the
204 // source should already be canonical, we don't have to do anything special,
205 // and the input is ASCII.
206 void CopyOneComponent(const char* source,
207 const Component& source_component,
209 Component* output_component) {
210 if (source_component.len < 0) {
211 // This component is not present.
212 *output_component = Component();
216 output_component->begin = output->length();
217 int source_end = source_component.end();
218 for (int i = source_component.begin; i < source_end; i++)
219 output->push_back(source[i]);
220 output_component->len = output->length() - output_component->begin;
225 // Called on Windows when the base URL is a file URL, this will copy the "C:"
226 // to the output, if there is a drive letter and if that drive letter is not
227 // being overridden by the relative URL. Otherwise, do nothing.
229 // It will return the index of the beginning of the next character in the
230 // base to be processed: if there is a "C:", the slash after it, or if
231 // there is no drive letter, the slash at the beginning of the path, or
232 // the end of the base. This can be used as the starting offset for further
234 template<typename CHAR>
235 int CopyBaseDriveSpecIfNecessary(const char* base_url,
238 const CHAR* relative_url,
240 int relative_url_len,
241 CanonOutput* output) {
242 if (base_path_begin >= base_path_end)
243 return base_path_begin; // No path.
245 // If the relative begins with a drive spec, don't do anything. The existing
246 // drive spec in the base will be replaced.
247 if (DoesBeginWindowsDriveSpec(relative_url, path_start, relative_url_len)) {
248 return base_path_begin; // Relative URL path is "C:/foo"
251 // The path should begin with a slash (as all canonical paths do). We check
252 // if it is followed by a drive letter and copy it.
253 if (DoesBeginSlashWindowsDriveSpec(base_url,
256 // Copy the two-character drive spec to the output. It will now look like
257 // "file:///C:" so the rest of it can be treated like a standard path.
258 output->push_back('/');
259 output->push_back(base_url[base_path_begin + 1]);
260 output->push_back(base_url[base_path_begin + 2]);
261 return base_path_begin + 3;
264 return base_path_begin;
269 // A subroutine of DoResolveRelativeURL, this resolves the URL knowning that
270 // the input is a relative path or less (query or ref).
271 template<typename CHAR>
272 bool DoResolveRelativePath(const char* base_url,
273 const Parsed& base_parsed,
275 const CHAR* relative_url,
276 const Component& relative_component,
277 CharsetConverter* query_converter,
279 Parsed* out_parsed) {
282 // We know the authority section didn't change, copy it to the output. We
283 // also know we have a path so can copy up to there.
284 Component path, query, ref;
285 ParsePathInternal(relative_url, relative_component, &path, &query, &ref);
287 // Canonical URLs always have a path, so we can use that offset. Reserve
288 // enough room for the base URL, the new path, and some extra bytes for
289 // possible escaped characters.
290 output->ReserveSizeIfNeeded(
291 base_parsed.path.begin +
292 std::max(path.end(), std::max(query.end(), ref.end())));
293 output->Append(base_url, base_parsed.path.begin);
296 // The path is replaced or modified.
297 int true_path_begin = output->length();
299 // For file: URLs on Windows, we don't want to treat the drive letter and
300 // colon as part of the path for relative file resolution when the
301 // incoming URL does not provide a drive spec. We save the true path
302 // beginning so we can fix it up after we are done.
303 int base_path_begin = base_parsed.path.begin;
306 base_path_begin = CopyBaseDriveSpecIfNecessary(
307 base_url, base_parsed.path.begin, base_parsed.path.end(),
308 relative_url, relative_component.begin, relative_component.end(),
310 // Now the output looks like either "file://" or "file:///C:"
311 // and we can start appending the rest of the path. |base_path_begin|
312 // points to the character in the base that comes next.
316 if (IsURLSlash(relative_url[path.begin])) {
317 // Easy case: the path is an absolute path on the server, so we can
318 // just replace everything from the path on with the new versions.
319 // Since the input should be canonical hierarchical URL, we should
320 // always have a path.
321 success &= CanonicalizePath(relative_url, path,
322 output, &out_parsed->path);
324 // Relative path, replace the query, and reference. We take the
325 // original path with the file part stripped, and append the new path.
326 // The canonicalizer will take care of resolving ".." and "."
327 int path_begin = output->length();
328 CopyToLastSlash(base_url, base_path_begin, base_parsed.path.end(),
330 success &= CanonicalizePartialPath(relative_url, path, path_begin,
332 out_parsed->path = MakeRange(path_begin, output->length());
334 // Copy the rest of the stuff after the path from the relative path.
337 // Finish with the query and reference part (these can't fail).
338 CanonicalizeQuery(relative_url, query, query_converter,
339 output, &out_parsed->query);
340 CanonicalizeRef(relative_url, ref, output, &out_parsed->ref);
342 // Fix the path beginning to add back the "C:" we may have written above.
343 out_parsed->path = MakeRange(true_path_begin, out_parsed->path.end());
347 // If we get here, the path is unchanged: copy to output.
348 CopyOneComponent(base_url, base_parsed.path, output, &out_parsed->path);
350 if (query.is_valid()) {
351 // Just the query specified, replace the query and reference (ignore
352 // failures for refs)
353 CanonicalizeQuery(relative_url, query, query_converter,
354 output, &out_parsed->query);
355 CanonicalizeRef(relative_url, ref, output, &out_parsed->ref);
359 // If we get here, the query is unchanged: copy to output. Note that the
360 // range of the query parameter doesn't include the question mark, so we
361 // have to add it manually if there is a component.
362 if (base_parsed.query.is_valid())
363 output->push_back('?');
364 CopyOneComponent(base_url, base_parsed.query, output, &out_parsed->query);
366 if (ref.is_valid()) {
367 // Just the reference specified: replace it (ignoring failures).
368 CanonicalizeRef(relative_url, ref, output, &out_parsed->ref);
372 // We should always have something to do in this function, the caller checks
373 // that some component is being replaced.
374 DCHECK(false) << "Not reached";
378 // Resolves a relative URL that contains a host. Typically, these will
379 // be of the form "//www.google.com/foo/bar?baz#ref" and the only thing which
380 // should be kept from the original URL is the scheme.
381 template<typename CHAR>
382 bool DoResolveRelativeHost(const char* base_url,
383 const Parsed& base_parsed,
384 const CHAR* relative_url,
385 const Component& relative_component,
386 CharsetConverter* query_converter,
388 Parsed* out_parsed) {
389 // Parse the relative URL, just like we would for anything following a
391 Parsed relative_parsed; // Everything but the scheme is valid.
392 ParseAfterScheme(relative_url, relative_component.end(),
393 relative_component.begin, &relative_parsed);
395 // Now we can just use the replacement function to replace all the necessary
396 // parts of the old URL with the new one.
397 Replacements<CHAR> replacements;
398 replacements.SetUsername(relative_url, relative_parsed.username);
399 replacements.SetPassword(relative_url, relative_parsed.password);
400 replacements.SetHost(relative_url, relative_parsed.host);
401 replacements.SetPort(relative_url, relative_parsed.port);
402 replacements.SetPath(relative_url, relative_parsed.path);
403 replacements.SetQuery(relative_url, relative_parsed.query);
404 replacements.SetRef(relative_url, relative_parsed.ref);
406 // Length() does not include the old scheme, so make sure to add it from the
408 output->ReserveSizeIfNeeded(
409 replacements.components().Length() +
410 base_parsed.CountCharactersBefore(Parsed::USERNAME, false));
411 SchemeType scheme_type = SCHEME_WITH_HOST_PORT_AND_USER_INFORMATION;
412 if (!GetStandardSchemeType(base_url, base_parsed.scheme, &scheme_type)) {
413 // A path with an authority section gets canonicalized under standard URL
414 // rules, even though the base was not known to be standard.
415 scheme_type = SCHEME_WITH_HOST_PORT_AND_USER_INFORMATION;
417 return ReplaceStandardURL(base_url, base_parsed, replacements, scheme_type,
418 query_converter, output, out_parsed);
421 // Resolves a relative URL that happens to be an absolute file path. Examples
422 // include: "//hostname/path", "/c:/foo", and "//hostname/c:/foo".
423 template<typename CHAR>
424 bool DoResolveAbsoluteFile(const CHAR* relative_url,
425 const Component& relative_component,
426 CharsetConverter* query_converter,
428 Parsed* out_parsed) {
429 // Parse the file URL. The file URl parsing function uses the same logic
430 // as we do for determining if the file is absolute, in which case it will
431 // not bother to look for a scheme.
432 Parsed relative_parsed;
433 ParseFileURL(&relative_url[relative_component.begin], relative_component.len,
436 return CanonicalizeFileURL(&relative_url[relative_component.begin],
437 relative_component.len, relative_parsed,
438 query_converter, output, out_parsed);
441 // TODO(brettw) treat two slashes as root like Mozilla for FTP?
442 template<typename CHAR>
443 bool DoResolveRelativeURL(const char* base_url,
444 const Parsed& base_parsed,
446 const CHAR* relative_url,
447 const Component& relative_component,
448 CharsetConverter* query_converter,
450 Parsed* out_parsed) {
451 // |base_parsed| is the starting point for our output. Since we may have
452 // removed whitespace from |relative_url| before entering this method, we'll
453 // carry over the |potentially_dangling_markup| flag.
454 bool potentially_dangling_markup = out_parsed->potentially_dangling_markup;
455 *out_parsed = base_parsed;
456 if (potentially_dangling_markup)
457 out_parsed->potentially_dangling_markup = true;
459 // Sanity check: the input should have a host or we'll break badly below.
460 // We can only resolve relative URLs with base URLs that have hosts and
461 // paths (even the default path of "/" is OK).
463 // We allow hosts with no length so we can handle file URLs, for example.
464 if (base_parsed.path.len <= 0) {
465 // On error, return the input (resolving a relative URL on a non-relative
467 int base_len = base_parsed.Length();
468 for (int i = 0; i < base_len; i++)
469 output->push_back(base_url[i]);
473 if (relative_component.len <= 0) {
474 // Empty relative URL, leave unchanged, only removing the ref component.
475 int base_len = base_parsed.Length();
476 base_len -= base_parsed.ref.len + 1;
477 out_parsed->ref.reset();
478 output->Append(base_url, base_len);
482 int num_slashes = CountConsecutiveSlashes(
483 relative_url, relative_component.begin, relative_component.end());
486 // On Windows, two slashes for a file path (regardless of which direction
487 // they are) means that it's UNC. Two backslashes on any base scheme mean
488 // that it's an absolute UNC path (we use the base_is_file flag to control
489 // how strict the UNC finder is).
491 // We also allow Windows absolute drive specs on any scheme (for example
492 // "c:\foo") like IE does. There must be no preceding slashes in this
493 // case (we reject anything like "/c:/foo") because that should be treated
494 // as a path. For file URLs, we allow any number of slashes since that would
495 // be setting the path.
497 // This assumes the absolute path resolver handles absolute URLs like this
498 // properly. DoCanonicalize does this.
499 int after_slashes = relative_component.begin + num_slashes;
500 if (DoesBeginUNCPath(relative_url, relative_component.begin,
501 relative_component.end(), !base_is_file) ||
502 ((num_slashes == 0 || base_is_file) &&
503 DoesBeginWindowsDriveSpec(
504 relative_url, after_slashes, relative_component.end()))) {
505 return DoResolveAbsoluteFile(relative_url, relative_component,
506 query_converter, output, out_parsed);
509 // Other platforms need explicit handling for file: URLs with multiple
510 // slashes because the generic scheme parsing always extracts a host, but a
511 // file: URL only has a host if it has exactly 2 slashes. Even if it does
512 // have a host, we want to use the special host detection logic for file
513 // URLs provided by DoResolveAbsoluteFile(), as opposed to the generic host
514 // detection logic, for consistency with parsing file URLs from scratch.
515 // This also handles the special case where the URL is only slashes,
516 // since that doesn't have a host part either.
518 (num_slashes >= 2 || num_slashes == relative_component.len)) {
519 return DoResolveAbsoluteFile(relative_url, relative_component,
520 query_converter, output, out_parsed);
524 // Any other double-slashes mean that this is relative to the scheme.
525 if (num_slashes >= 2) {
526 return DoResolveRelativeHost(base_url, base_parsed,
527 relative_url, relative_component,
528 query_converter, output, out_parsed);
531 // When we get here, we know that the relative URL is on the same host.
532 return DoResolveRelativePath(base_url, base_parsed, base_is_file,
533 relative_url, relative_component,
534 query_converter, output, out_parsed);
539 bool IsRelativeURL(const char* base,
540 const Parsed& base_parsed,
541 const char* fragment,
543 bool is_base_hierarchical,
545 Component* relative_component) {
546 return DoIsRelativeURL<char>(
547 base, base_parsed, fragment, fragment_len, is_base_hierarchical,
548 is_relative, relative_component);
551 bool IsRelativeURL(const char* base,
552 const Parsed& base_parsed,
553 const base::char16* fragment,
555 bool is_base_hierarchical,
557 Component* relative_component) {
558 return DoIsRelativeURL<base::char16>(
559 base, base_parsed, fragment, fragment_len, is_base_hierarchical,
560 is_relative, relative_component);
563 bool ResolveRelativeURL(const char* base_url,
564 const Parsed& base_parsed,
566 const char* relative_url,
567 const Component& relative_component,
568 CharsetConverter* query_converter,
570 Parsed* out_parsed) {
571 return DoResolveRelativeURL<char>(
572 base_url, base_parsed, base_is_file, relative_url,
573 relative_component, query_converter, output, out_parsed);
576 bool ResolveRelativeURL(const char* base_url,
577 const Parsed& base_parsed,
579 const base::char16* relative_url,
580 const Component& relative_component,
581 CharsetConverter* query_converter,
583 Parsed* out_parsed) {
584 return DoResolveRelativeURL<base::char16>(
585 base_url, base_parsed, base_is_file, relative_url,
586 relative_component, query_converter, output, out_parsed);