1 // Copyright 2013 The Chromium Authors
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.
10 #include "base/check_op.h"
11 #include "base/strings/string_util.h"
12 #include "url/url_canon.h"
13 #include "url/url_canon_internal.h"
14 #include "url/url_constants.h"
15 #include "url/url_features.h"
16 #include "url/url_file.h"
17 #include "url/url_parse_internal.h"
18 #include "url/url_util.h"
19 #include "url/url_util_internal.h"
25 // Firefox does a case-sensitive compare (which is probably wrong--Mozilla bug
26 // 379034), whereas IE is case-insensitive.
28 // We choose to be more permissive like IE. We don't need to worry about
29 // unescaping or anything here: neither IE or Firefox allow this. We also
30 // don't have to worry about invalid scheme characters since we are comparing
31 // against the canonical scheme of the base.
33 // The base URL should always be canonical, therefore it should be ASCII.
34 template<typename CHAR>
35 bool AreSchemesEqual(const char* base,
36 const Component& base_scheme,
38 const Component& cmp_scheme) {
39 if (base_scheme.len != cmp_scheme.len)
41 for (int i = 0; i < base_scheme.len; i++) {
42 // We assume the base is already canonical, so we don't have to
44 if (CanonicalSchemeChar(cmp[cmp_scheme.begin + i]) !=
45 base[base_scheme.begin + i])
53 // Here, we also allow Windows paths to be represented as "/C:/" so we can be
54 // consistent about URL paths beginning with slashes. This function is like
55 // DoesBeginWindowsDrivePath except that it also requires a slash at the
57 template<typename CHAR>
58 bool DoesBeginSlashWindowsDriveSpec(const CHAR* spec, int start_offset,
60 if (start_offset >= spec_len)
62 return IsURLSlash(spec[start_offset]) &&
63 DoesBeginWindowsDriveSpec(spec, start_offset + 1, spec_len);
68 template <typename CHAR>
69 bool IsValidScheme(const CHAR* url, const Component& scheme) {
70 // Caller should ensure that the |scheme| is not empty.
71 DCHECK_NE(0, scheme.len);
73 // From https://url.spec.whatwg.org/#scheme-start-state:
74 // scheme start state:
75 // 1. If c is an ASCII alpha, append c, lowercased, to buffer, and set
76 // state to scheme state.
77 // 2. Otherwise, if state override is not given, set state to no scheme
78 // state, and decrease pointer by one.
79 // 3. Otherwise, validation error, return failure.
80 // Note that both step 2 and step 3 mean that the scheme was not valid.
81 if (!base::IsAsciiAlpha(url[scheme.begin]))
84 // From https://url.spec.whatwg.org/#scheme-state:
86 // 1. If c is an ASCII alphanumeric, U+002B (+), U+002D (-), or U+002E
87 // (.), append c, lowercased, to buffer.
88 // 2. Otherwise, if c is U+003A (:), then [...]
90 // We begin at |scheme.begin + 1|, because the character at |scheme.begin| has
91 // already been checked by base::IsAsciiAlpha above.
92 int scheme_end = scheme.end();
93 for (int i = scheme.begin + 1; i < scheme_end; i++) {
94 if (!CanonicalSchemeChar(url[i]))
101 // See IsRelativeURL in the header file for usage.
102 template<typename CHAR>
103 bool DoIsRelativeURL(const char* base,
104 const Parsed& base_parsed,
107 bool is_base_hierarchical,
109 Component* relative_component) {
110 *is_relative = false; // So we can default later to not relative.
112 // Trim whitespace and construct a new range for the substring.
114 TrimURL(url, &begin, &url_len);
115 if (begin >= url_len) {
116 // Empty URLs are relative, but do nothing.
117 if (!is_base_hierarchical) {
118 // Don't allow relative URLs if the base scheme doesn't support it.
121 *relative_component = Component(begin, 0);
127 // We special case paths like "C:\foo" so they can link directly to the
128 // file on Windows (IE compatibility). The security domain stuff should
129 // prevent a link like this from actually being followed if its on a
132 // We treat "C:/foo" as an absolute URL. We can go ahead and treat "/c:/"
133 // as relative, as this will just replace the path when the base scheme
134 // is a file and the answer will still be correct.
136 // We require strict backslashes when detecting UNC since two forward
137 // slashes should be treated a a relative URL with a hostname.
138 if (DoesBeginWindowsDriveSpec(url, begin, url_len) ||
139 DoesBeginUNCPath(url, begin, url_len, true))
143 // See if we've got a scheme, if not, we know this is a relative URL.
144 // BUT, just because we have a scheme, doesn't make it absolute.
145 // "http:foo.html" is a relative URL with path "foo.html". If the scheme is
146 // empty, we treat it as relative (":foo"), like IE does.
148 const bool scheme_is_empty =
149 !ExtractScheme(url, url_len, &scheme) || scheme.len == 0;
150 if (scheme_is_empty) {
151 if (url[begin] == '#') {
152 // |url| is a bare fragment (e.g. "#foo"). This can be resolved against
153 // any base. Fall-through.
154 } else if (!is_base_hierarchical) {
155 // Don't allow relative URLs if the base scheme doesn't support it.
159 *relative_component = MakeRange(begin, url_len);
164 // If the scheme isn't valid, then it's relative.
165 if (!IsValidScheme(url, scheme)) {
166 if (url[begin] == '#' &&
167 base::FeatureList::IsEnabled(
168 kResolveBareFragmentWithColonOnNonHierarchical)) {
169 // |url| is a bare fragment (e.g. "#foo:bar"). This can be resolved
170 // against any base. Fall-through.
171 } else if (!is_base_hierarchical) {
172 // Don't allow relative URLs if the base scheme doesn't support it.
175 *relative_component = MakeRange(begin, url_len);
180 // If the scheme is not the same, then we can't count it as relative.
181 if (!AreSchemesEqual(base, base_parsed.scheme, url, scheme))
184 // When the scheme that they both share is not hierarchical, treat the
185 // incoming scheme as absolute (this way with the base of "data:foo",
186 // "data:bar" will be reported as absolute.
187 if (!is_base_hierarchical)
190 int colon_offset = scheme.end();
192 // If it's a filesystem URL, the only valid way to make it relative is not to
193 // supply a scheme. There's no equivalent to e.g. http:index.html.
194 if (CompareSchemeComponent(url, scheme, kFileSystemScheme))
197 // ExtractScheme guarantees that the colon immediately follows what it
198 // considers to be the scheme. CountConsecutiveSlashes will handle the
199 // case where the begin offset is the end of the input.
200 int num_slashes = CountConsecutiveSlashes(url, colon_offset + 1, url_len);
202 if (num_slashes == 0 || num_slashes == 1) {
203 // No slashes means it's a relative path like "http:foo.html". One slash
204 // is an absolute path. "http:/home/foo.html"
206 *relative_component = MakeRange(colon_offset + 1, url_len);
210 // Two or more slashes after the scheme we treat as absolute.
214 // Copies all characters in the range [begin, end) of |spec| to the output,
215 // up until and including the last slash. There should be a slash in the
216 // range, if not, nothing will be copied.
218 // For stardard URLs the input should be canonical, but when resolving relative
219 // URLs on a non-standard base (like "data:") the input can be anything.
220 void CopyToLastSlash(const char* spec,
223 CanonOutput* output) {
224 // Find the last slash.
226 for (int i = end - 1; i >= begin; i--) {
227 if (spec[i] == '/' || spec[i] == '\\') {
236 for (int i = begin; i <= last_slash; i++)
237 output->push_back(spec[i]);
240 // Copies a single component from the source to the output. This is used
241 // when resolving relative URLs and a given component is unchanged. Since the
242 // source should already be canonical, we don't have to do anything special,
243 // and the input is ASCII.
244 void CopyOneComponent(const char* source,
245 const Component& source_component,
247 Component* output_component) {
248 if (!source_component.is_valid()) {
249 // This component is not present.
250 *output_component = Component();
254 output_component->begin = output->length();
255 int source_end = source_component.end();
256 for (int i = source_component.begin; i < source_end; i++)
257 output->push_back(source[i]);
258 output_component->len = output->length() - output_component->begin;
263 // Called on Windows when the base URL is a file URL, this will copy the "C:"
264 // to the output, if there is a drive letter and if that drive letter is not
265 // being overridden by the relative URL. Otherwise, do nothing.
267 // It will return the index of the beginning of the next character in the
268 // base to be processed: if there is a "C:", the slash after it, or if
269 // there is no drive letter, the slash at the beginning of the path, or
270 // the end of the base. This can be used as the starting offset for further
272 template<typename CHAR>
273 int CopyBaseDriveSpecIfNecessary(const char* base_url,
276 const CHAR* relative_url,
278 int relative_url_len,
279 CanonOutput* output) {
280 if (base_path_begin >= base_path_end)
281 return base_path_begin; // No path.
283 // If the relative begins with a drive spec, don't do anything. The existing
284 // drive spec in the base will be replaced.
285 if (DoesBeginWindowsDriveSpec(relative_url, path_start, relative_url_len)) {
286 return base_path_begin; // Relative URL path is "C:/foo"
289 // The path should begin with a slash (as all canonical paths do). We check
290 // if it is followed by a drive letter and copy it.
291 if (DoesBeginSlashWindowsDriveSpec(base_url,
294 // Copy the two-character drive spec to the output. It will now look like
295 // "file:///C:" so the rest of it can be treated like a standard path.
296 output->push_back('/');
297 output->push_back(base_url[base_path_begin + 1]);
298 output->push_back(base_url[base_path_begin + 2]);
299 return base_path_begin + 3;
302 return base_path_begin;
307 // A subroutine of DoResolveRelativeURL, this resolves the URL knowning that
308 // the input is a relative path or less (query or ref).
309 template<typename CHAR>
310 bool DoResolveRelativePath(const char* base_url,
311 const Parsed& base_parsed,
313 const CHAR* relative_url,
314 const Component& relative_component,
315 CharsetConverter* query_converter,
317 Parsed* out_parsed) {
320 // We know the authority section didn't change, copy it to the output. We
321 // also know we have a path so can copy up to there.
322 Component path, query, ref;
323 ParsePathInternal(relative_url, relative_component, &path, &query, &ref);
325 // Canonical URLs always have a path, so we can use that offset. Reserve
326 // enough room for the base URL, the new path, and some extra bytes for
327 // possible escaped characters.
328 output->ReserveSizeIfNeeded(base_parsed.path.begin +
329 std::max({path.end(), query.end(), ref.end()}));
330 output->Append(base_url, base_parsed.path.begin);
332 if (path.is_nonempty()) {
333 // The path is replaced or modified.
334 int true_path_begin = output->length();
336 // For file: URLs on Windows, we don't want to treat the drive letter and
337 // colon as part of the path for relative file resolution when the
338 // incoming URL does not provide a drive spec. We save the true path
339 // beginning so we can fix it up after we are done.
340 int base_path_begin = base_parsed.path.begin;
343 base_path_begin = CopyBaseDriveSpecIfNecessary(
344 base_url, base_parsed.path.begin, base_parsed.path.end(),
345 relative_url, relative_component.begin, relative_component.end(),
347 // Now the output looks like either "file://" or "file:///C:"
348 // and we can start appending the rest of the path. |base_path_begin|
349 // points to the character in the base that comes next.
353 if (IsURLSlash(relative_url[path.begin])) {
354 // Easy case: the path is an absolute path on the server, so we can
355 // just replace everything from the path on with the new versions.
356 // Since the input should be canonical hierarchical URL, we should
357 // always have a path.
358 success &= CanonicalizePath(relative_url, path,
359 output, &out_parsed->path);
361 // Relative path, replace the query, and reference. We take the
362 // original path with the file part stripped, and append the new path.
363 // The canonicalizer will take care of resolving ".." and "."
364 size_t path_begin = output->length();
365 CopyToLastSlash(base_url, base_path_begin, base_parsed.path.end(),
367 success &= CanonicalizePartialPathInternal(relative_url, path, path_begin,
369 out_parsed->path = MakeRange(path_begin, output->length());
371 // Copy the rest of the stuff after the path from the relative path.
374 // Finish with the query and reference part (these can't fail).
375 CanonicalizeQuery(relative_url, query, query_converter,
376 output, &out_parsed->query);
377 CanonicalizeRef(relative_url, ref, output, &out_parsed->ref);
379 // Fix the path beginning to add back the "C:" we may have written above.
380 out_parsed->path = MakeRange(true_path_begin, out_parsed->path.end());
384 // If we get here, the path is unchanged: copy to output.
385 CopyOneComponent(base_url, base_parsed.path, output, &out_parsed->path);
387 if (query.is_valid()) {
388 // Just the query specified, replace the query and reference (ignore
389 // failures for refs)
390 CanonicalizeQuery(relative_url, query, query_converter,
391 output, &out_parsed->query);
392 CanonicalizeRef(relative_url, ref, output, &out_parsed->ref);
396 // If we get here, the query is unchanged: copy to output. Note that the
397 // range of the query parameter doesn't include the question mark, so we
398 // have to add it manually if there is a component.
399 if (base_parsed.query.is_valid())
400 output->push_back('?');
401 CopyOneComponent(base_url, base_parsed.query, output, &out_parsed->query);
403 if (ref.is_valid()) {
404 // Just the reference specified: replace it (ignoring failures).
405 CanonicalizeRef(relative_url, ref, output, &out_parsed->ref);
409 // We should always have something to do in this function, the caller checks
410 // that some component is being replaced.
411 DCHECK(false) << "Not reached";
415 // Resolves a relative URL that contains a host. Typically, these will
416 // be of the form "//www.google.com/foo/bar?baz#ref" and the only thing which
417 // should be kept from the original URL is the scheme.
418 template<typename CHAR>
419 bool DoResolveRelativeHost(const char* base_url,
420 const Parsed& base_parsed,
421 const CHAR* relative_url,
422 const Component& relative_component,
423 CharsetConverter* query_converter,
425 Parsed* out_parsed) {
426 // Parse the relative URL, just like we would for anything following a
428 Parsed relative_parsed; // Everything but the scheme is valid.
429 ParseAfterScheme(relative_url, relative_component.end(),
430 relative_component.begin, &relative_parsed);
432 // Now we can just use the replacement function to replace all the necessary
433 // parts of the old URL with the new one.
434 Replacements<CHAR> replacements;
435 replacements.SetUsername(relative_url, relative_parsed.username);
436 replacements.SetPassword(relative_url, relative_parsed.password);
437 replacements.SetHost(relative_url, relative_parsed.host);
438 replacements.SetPort(relative_url, relative_parsed.port);
439 replacements.SetPath(relative_url, relative_parsed.path);
440 replacements.SetQuery(relative_url, relative_parsed.query);
441 replacements.SetRef(relative_url, relative_parsed.ref);
443 // Length() does not include the old scheme, so make sure to add it from the
445 output->ReserveSizeIfNeeded(
446 replacements.components().Length() +
447 base_parsed.CountCharactersBefore(Parsed::USERNAME, false));
448 SchemeType scheme_type = SCHEME_WITH_HOST_PORT_AND_USER_INFORMATION;
449 if (!GetStandardSchemeType(base_url, base_parsed.scheme, &scheme_type)) {
450 // A path with an authority section gets canonicalized under standard URL
451 // rules, even though the base was not known to be standard.
452 scheme_type = SCHEME_WITH_HOST_PORT_AND_USER_INFORMATION;
454 return ReplaceStandardURL(base_url, base_parsed, replacements, scheme_type,
455 query_converter, output, out_parsed);
458 // Resolves a relative URL that happens to be an absolute file path. Examples
459 // include: "//hostname/path", "/c:/foo", and "//hostname/c:/foo".
460 template<typename CHAR>
461 bool DoResolveAbsoluteFile(const CHAR* relative_url,
462 const Component& relative_component,
463 CharsetConverter* query_converter,
465 Parsed* out_parsed) {
466 // Parse the file URL. The file URl parsing function uses the same logic
467 // as we do for determining if the file is absolute, in which case it will
468 // not bother to look for a scheme.
469 Parsed relative_parsed;
470 ParseFileURL(&relative_url[relative_component.begin], relative_component.len,
473 return CanonicalizeFileURL(&relative_url[relative_component.begin],
474 relative_component.len, relative_parsed,
475 query_converter, output, out_parsed);
478 // TODO(brettw) treat two slashes as root like Mozilla for FTP?
479 template<typename CHAR>
480 bool DoResolveRelativeURL(const char* base_url,
481 const Parsed& base_parsed,
483 const CHAR* relative_url,
484 const Component& relative_component,
485 CharsetConverter* query_converter,
487 Parsed* out_parsed) {
488 // |base_parsed| is the starting point for our output. Since we may have
489 // removed whitespace from |relative_url| before entering this method, we'll
490 // carry over the |potentially_dangling_markup| flag.
491 bool potentially_dangling_markup = out_parsed->potentially_dangling_markup;
492 *out_parsed = base_parsed;
493 if (potentially_dangling_markup)
494 out_parsed->potentially_dangling_markup = true;
496 // Sanity check: the input should have a host or we'll break badly below.
497 // We can only resolve relative URLs with base URLs that have hosts and
498 // paths (even the default path of "/" is OK).
500 // We allow hosts with no length so we can handle file URLs, for example.
501 if (base_parsed.path.is_empty()) {
502 // On error, return the input (resolving a relative URL on a non-relative
504 int base_len = base_parsed.Length();
505 for (int i = 0; i < base_len; i++)
506 output->push_back(base_url[i]);
510 if (relative_component.is_empty()) {
511 // Empty relative URL, leave unchanged, only removing the ref component.
512 int base_len = base_parsed.Length();
513 base_len -= base_parsed.ref.len + 1;
514 out_parsed->ref.reset();
515 output->Append(base_url, base_len);
519 int num_slashes = CountConsecutiveSlashes(
520 relative_url, relative_component.begin, relative_component.end());
523 // On Windows, two slashes for a file path (regardless of which direction
524 // they are) means that it's UNC. Two backslashes on any base scheme mean
525 // that it's an absolute UNC path (we use the base_is_file flag to control
526 // how strict the UNC finder is).
528 // We also allow Windows absolute drive specs on any scheme (for example
529 // "c:\foo") like IE does. There must be no preceding slashes in this
530 // case (we reject anything like "/c:/foo") because that should be treated
531 // as a path. For file URLs, we allow any number of slashes since that would
532 // be setting the path.
534 // This assumes the absolute path resolver handles absolute URLs like this
535 // properly. DoCanonicalize does this.
536 int after_slashes = relative_component.begin + num_slashes;
537 if (DoesBeginUNCPath(relative_url, relative_component.begin,
538 relative_component.end(), !base_is_file) ||
539 ((num_slashes == 0 || base_is_file) &&
540 DoesBeginWindowsDriveSpec(
541 relative_url, after_slashes, relative_component.end()))) {
542 return DoResolveAbsoluteFile(relative_url, relative_component,
543 query_converter, output, out_parsed);
546 // Other platforms need explicit handling for file: URLs with multiple
547 // slashes because the generic scheme parsing always extracts a host, but a
548 // file: URL only has a host if it has exactly 2 slashes. Even if it does
549 // have a host, we want to use the special host detection logic for file
550 // URLs provided by DoResolveAbsoluteFile(), as opposed to the generic host
551 // detection logic, for consistency with parsing file URLs from scratch.
552 if (base_is_file && num_slashes >= 2) {
553 return DoResolveAbsoluteFile(relative_url, relative_component,
554 query_converter, output, out_parsed);
558 // Any other double-slashes mean that this is relative to the scheme.
559 if (num_slashes >= 2) {
560 return DoResolveRelativeHost(base_url, base_parsed,
561 relative_url, relative_component,
562 query_converter, output, out_parsed);
565 // When we get here, we know that the relative URL is on the same host.
566 return DoResolveRelativePath(base_url, base_parsed, base_is_file,
567 relative_url, relative_component,
568 query_converter, output, out_parsed);
573 bool IsRelativeURL(const char* base,
574 const Parsed& base_parsed,
575 const char* fragment,
577 bool is_base_hierarchical,
579 Component* relative_component) {
580 return DoIsRelativeURL<char>(
581 base, base_parsed, fragment, fragment_len, is_base_hierarchical,
582 is_relative, relative_component);
585 bool IsRelativeURL(const char* base,
586 const Parsed& base_parsed,
587 const char16_t* fragment,
589 bool is_base_hierarchical,
591 Component* relative_component) {
592 return DoIsRelativeURL<char16_t>(base, base_parsed, fragment, fragment_len,
593 is_base_hierarchical, is_relative,
597 bool ResolveRelativeURL(const char* base_url,
598 const Parsed& base_parsed,
600 const char* relative_url,
601 const Component& relative_component,
602 CharsetConverter* query_converter,
604 Parsed* out_parsed) {
605 return DoResolveRelativeURL<char>(
606 base_url, base_parsed, base_is_file, relative_url,
607 relative_component, query_converter, output, out_parsed);
610 bool ResolveRelativeURL(const char* base_url,
611 const Parsed& base_parsed,
613 const char16_t* relative_url,
614 const Component& relative_component,
615 CharsetConverter* query_converter,
617 Parsed* out_parsed) {
618 return DoResolveRelativeURL<char16_t>(base_url, base_parsed, base_is_file,
619 relative_url, relative_component,
620 query_converter, output, out_parsed);