url_canon_relative.cc revision 0529e5d033099cbfc42635f6f6183833b09dff6e
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. 4 5// Canonicalizer functions for working with and resolving relative URLs. 6 7#include "base/logging.h" 8#include "url/url_canon.h" 9#include "url/url_canon_internal.h" 10#include "url/url_file.h" 11#include "url/url_parse_internal.h" 12#include "url/url_util_internal.h" 13 14namespace url { 15 16namespace { 17 18// Firefox does a case-sensitive compare (which is probably wrong--Mozilla bug 19// 379034), whereas IE is case-insensetive. 20// 21// We choose to be more permissive like IE. We don't need to worry about 22// unescaping or anything here: neither IE or Firefox allow this. We also 23// don't have to worry about invalid scheme characters since we are comparing 24// against the canonical scheme of the base. 25// 26// The base URL should always be canonical, therefore is ASCII. 27template<typename CHAR> 28bool AreSchemesEqual(const char* base, 29 const Component& base_scheme, 30 const CHAR* cmp, 31 const Component& cmp_scheme) { 32 if (base_scheme.len != cmp_scheme.len) 33 return false; 34 for (int i = 0; i < base_scheme.len; i++) { 35 // We assume the base is already canonical, so we don't have to 36 // canonicalize it. 37 if (CanonicalSchemeChar(cmp[cmp_scheme.begin + i]) != 38 base[base_scheme.begin + i]) 39 return false; 40 } 41 return true; 42} 43 44#ifdef WIN32 45 46// Here, we also allow Windows paths to be represented as "/C:/" so we can be 47// consistent about URL paths beginning with slashes. This function is like 48// DoesBeginWindowsDrivePath except that it also requires a slash at the 49// beginning. 50template<typename CHAR> 51bool DoesBeginSlashWindowsDriveSpec(const CHAR* spec, int start_offset, 52 int spec_len) { 53 if (start_offset >= spec_len) 54 return false; 55 return IsURLSlash(spec[start_offset]) && 56 DoesBeginWindowsDriveSpec(spec, start_offset + 1, spec_len); 57} 58 59#endif // WIN32 60 61// See IsRelativeURL in the header file for usage. 62template<typename CHAR> 63bool DoIsRelativeURL(const char* base, 64 const Parsed& base_parsed, 65 const CHAR* url, 66 int url_len, 67 bool is_base_hierarchical, 68 bool* is_relative, 69 Component* relative_component) { 70 *is_relative = false; // So we can default later to not relative. 71 72 // Trim whitespace and construct a new range for the substring. 73 int begin = 0; 74 TrimURL(url, &begin, &url_len); 75 if (begin >= url_len) { 76 // Empty URLs are relative, but do nothing. 77 *relative_component = Component(begin, 0); 78 *is_relative = true; 79 return true; 80 } 81 82#ifdef WIN32 83 // We special case paths like "C:\foo" so they can link directly to the 84 // file on Windows (IE compatability). The security domain stuff should 85 // prevent a link like this from actually being followed if its on a 86 // web page. 87 // 88 // We treat "C:/foo" as an absolute URL. We can go ahead and treat "/c:/" 89 // as relative, as this will just replace the path when the base scheme 90 // is a file and the answer will still be correct. 91 // 92 // We require strict backslashes when detecting UNC since two forward 93 // shashes should be treated a a relative URL with a hostname. 94 if (DoesBeginWindowsDriveSpec(url, begin, url_len) || 95 DoesBeginUNCPath(url, begin, url_len, true)) 96 return true; 97#endif // WIN32 98 99 // See if we've got a scheme, if not, we know this is a relative URL. 100 // BUT: Just because we have a scheme, doesn't make it absolute. 101 // "http:foo.html" is a relative URL with path "foo.html". If the scheme is 102 // empty, we treat it as relative (":foo") like IE does. 103 Component scheme; 104 const bool scheme_is_empty = 105 !ExtractScheme(url, url_len, &scheme) || scheme.len == 0; 106 if (scheme_is_empty) { 107 if (url[begin] == '#') { 108 // |url| is a bare fragement (e.g. "#foo"). This can be resolved against 109 // any base. Fall-through. 110 } else if (!is_base_hierarchical) { 111 // Don't allow relative URLs if the base scheme doesn't support it. 112 return false; 113 } 114 115 *relative_component = MakeRange(begin, url_len); 116 *is_relative = true; 117 return true; 118 } 119 120 // If the scheme isn't valid, then it's relative. 121 int scheme_end = scheme.end(); 122 for (int i = scheme.begin; i < scheme_end; i++) { 123 if (!CanonicalSchemeChar(url[i])) { 124 if (!is_base_hierarchical) { 125 // Don't allow relative URLs if the base scheme doesn't support it. 126 return false; 127 } 128 *relative_component = MakeRange(begin, url_len); 129 *is_relative = true; 130 return true; 131 } 132 } 133 134 // If the scheme is not the same, then we can't count it as relative. 135 if (!AreSchemesEqual(base, base_parsed.scheme, url, scheme)) 136 return true; 137 138 // When the scheme that they both share is not hierarchical, treat the 139 // incoming scheme as absolute (this way with the base of "data:foo", 140 // "data:bar" will be reported as absolute. 141 if (!is_base_hierarchical) 142 return true; 143 144 int colon_offset = scheme.end(); 145 146 // If it's a filesystem URL, the only valid way to make it relative is not to 147 // supply a scheme. There's no equivalent to e.g. http:index.html. 148 if (CompareSchemeComponent(url, scheme, "filesystem")) 149 return true; 150 151 // ExtractScheme guarantees that the colon immediately follows what it 152 // considers to be the scheme. CountConsecutiveSlashes will handle the 153 // case where the begin offset is the end of the input. 154 int num_slashes = CountConsecutiveSlashes(url, colon_offset + 1, url_len); 155 156 if (num_slashes == 0 || num_slashes == 1) { 157 // No slashes means it's a relative path like "http:foo.html". One slash 158 // is an absolute path. "http:/home/foo.html" 159 *is_relative = true; 160 *relative_component = MakeRange(colon_offset + 1, url_len); 161 return true; 162 } 163 164 // Two or more slashes after the scheme we treat as absolute. 165 return true; 166} 167 168// Copies all characters in the range [begin, end) of |spec| to the output, 169// up until and including the last slash. There should be a slash in the 170// range, if not, nothing will be copied. 171// 172// The input is assumed to be canonical, so we search only for exact slashes 173// and not backslashes as well. We also know that it's ASCII. 174void CopyToLastSlash(const char* spec, 175 int begin, 176 int end, 177 CanonOutput* output) { 178 // Find the last slash. 179 int last_slash = -1; 180 for (int i = end - 1; i >= begin; i--) { 181 if (spec[i] == '/') { 182 last_slash = i; 183 break; 184 } 185 } 186 if (last_slash < 0) 187 return; // No slash. 188 189 // Copy. 190 for (int i = begin; i <= last_slash; i++) 191 output->push_back(spec[i]); 192} 193 194// Copies a single component from the source to the output. This is used 195// when resolving relative URLs and a given component is unchanged. Since the 196// source should already be canonical, we don't have to do anything special, 197// and the input is ASCII. 198void CopyOneComponent(const char* source, 199 const Component& source_component, 200 CanonOutput* output, 201 Component* output_component) { 202 if (source_component.len < 0) { 203 // This component is not present. 204 *output_component = Component(); 205 return; 206 } 207 208 output_component->begin = output->length(); 209 int source_end = source_component.end(); 210 for (int i = source_component.begin; i < source_end; i++) 211 output->push_back(source[i]); 212 output_component->len = output->length() - output_component->begin; 213} 214 215#ifdef WIN32 216 217// Called on Windows when the base URL is a file URL, this will copy the "C:" 218// to the output, if there is a drive letter and if that drive letter is not 219// being overridden by the relative URL. Otherwise, do nothing. 220// 221// It will return the index of the beginning of the next character in the 222// base to be processed: if there is a "C:", the slash after it, or if 223// there is no drive letter, the slash at the beginning of the path, or 224// the end of the base. This can be used as the starting offset for further 225// path processing. 226template<typename CHAR> 227int CopyBaseDriveSpecIfNecessary(const char* base_url, 228 int base_path_begin, 229 int base_path_end, 230 const CHAR* relative_url, 231 int path_start, 232 int relative_url_len, 233 CanonOutput* output) { 234 if (base_path_begin >= base_path_end) 235 return base_path_begin; // No path. 236 237 // If the relative begins with a drive spec, don't do anything. The existing 238 // drive spec in the base will be replaced. 239 if (DoesBeginWindowsDriveSpec(relative_url, path_start, relative_url_len)) { 240 return base_path_begin; // Relative URL path is "C:/foo" 241 } 242 243 // The path should begin with a slash (as all canonical paths do). We check 244 // if it is followed by a drive letter and copy it. 245 if (DoesBeginSlashWindowsDriveSpec(base_url, 246 base_path_begin, 247 base_path_end)) { 248 // Copy the two-character drive spec to the output. It will now look like 249 // "file:///C:" so the rest of it can be treated like a standard path. 250 output->push_back('/'); 251 output->push_back(base_url[base_path_begin + 1]); 252 output->push_back(base_url[base_path_begin + 2]); 253 return base_path_begin + 3; 254 } 255 256 return base_path_begin; 257} 258 259#endif // WIN32 260 261// A subroutine of DoResolveRelativeURL, this resolves the URL knowning that 262// the input is a relative path or less (qyuery or ref). 263template<typename CHAR> 264bool DoResolveRelativePath(const char* base_url, 265 const Parsed& base_parsed, 266 bool base_is_file, 267 const CHAR* relative_url, 268 const Component& relative_component, 269 CharsetConverter* query_converter, 270 CanonOutput* output, 271 Parsed* out_parsed) { 272 bool success = true; 273 274 // We know the authority section didn't change, copy it to the output. We 275 // also know we have a path so can copy up to there. 276 Component path, query, ref; 277 ParsePathInternal(relative_url, relative_component, &path, &query, &ref); 278 // Canonical URLs always have a path, so we can use that offset. 279 output->Append(base_url, base_parsed.path.begin); 280 281 if (path.len > 0) { 282 // The path is replaced or modified. 283 int true_path_begin = output->length(); 284 285 // For file: URLs on Windows, we don't want to treat the drive letter and 286 // colon as part of the path for relative file resolution when the 287 // incoming URL does not provide a drive spec. We save the true path 288 // beginning so we can fix it up after we are done. 289 int base_path_begin = base_parsed.path.begin; 290#ifdef WIN32 291 if (base_is_file) { 292 base_path_begin = CopyBaseDriveSpecIfNecessary( 293 base_url, base_parsed.path.begin, base_parsed.path.end(), 294 relative_url, relative_component.begin, relative_component.end(), 295 output); 296 // Now the output looks like either "file://" or "file:///C:" 297 // and we can start appending the rest of the path. |base_path_begin| 298 // points to the character in the base that comes next. 299 } 300#endif // WIN32 301 302 if (IsURLSlash(relative_url[path.begin])) { 303 // Easy case: the path is an absolute path on the server, so we can 304 // just replace everything from the path on with the new versions. 305 // Since the input should be canonical hierarchical URL, we should 306 // always have a path. 307 success &= CanonicalizePath(relative_url, path, 308 output, &out_parsed->path); 309 } else { 310 // Relative path, replace the query, and reference. We take the 311 // original path with the file part stripped, and append the new path. 312 // The canonicalizer will take care of resolving ".." and "." 313 int path_begin = output->length(); 314 CopyToLastSlash(base_url, base_path_begin, base_parsed.path.end(), 315 output); 316 success &= CanonicalizePartialPath(relative_url, path, path_begin, 317 output); 318 out_parsed->path = MakeRange(path_begin, output->length()); 319 320 // Copy the rest of the stuff after the path from the relative path. 321 } 322 323 // Finish with the query and reference part (these can't fail). 324 CanonicalizeQuery(relative_url, query, query_converter, 325 output, &out_parsed->query); 326 CanonicalizeRef(relative_url, ref, output, &out_parsed->ref); 327 328 // Fix the path beginning to add back the "C:" we may have written above. 329 out_parsed->path = MakeRange(true_path_begin, out_parsed->path.end()); 330 return success; 331 } 332 333 // If we get here, the path is unchanged: copy to output. 334 CopyOneComponent(base_url, base_parsed.path, output, &out_parsed->path); 335 336 if (query.is_valid()) { 337 // Just the query specified, replace the query and reference (ignore 338 // failures for refs) 339 CanonicalizeQuery(relative_url, query, query_converter, 340 output, &out_parsed->query); 341 CanonicalizeRef(relative_url, ref, output, &out_parsed->ref); 342 return success; 343 } 344 345 // If we get here, the query is unchanged: copy to output. Note that the 346 // range of the query parameter doesn't include the question mark, so we 347 // have to add it manually if there is a component. 348 if (base_parsed.query.is_valid()) 349 output->push_back('?'); 350 CopyOneComponent(base_url, base_parsed.query, output, &out_parsed->query); 351 352 if (ref.is_valid()) { 353 // Just the reference specified: replace it (ignoring failures). 354 CanonicalizeRef(relative_url, ref, output, &out_parsed->ref); 355 return success; 356 } 357 358 // We should always have something to do in this function, the caller checks 359 // that some component is being replaced. 360 DCHECK(false) << "Not reached"; 361 return success; 362} 363 364// Resolves a relative URL that contains a host. Typically, these will 365// be of the form "//www.google.com/foo/bar?baz#ref" and the only thing which 366// should be kept from the original URL is the scheme. 367template<typename CHAR> 368bool DoResolveRelativeHost(const char* base_url, 369 const Parsed& base_parsed, 370 const CHAR* relative_url, 371 const Component& relative_component, 372 CharsetConverter* query_converter, 373 CanonOutput* output, 374 Parsed* out_parsed) { 375 // Parse the relative URL, just like we would for anything following a 376 // scheme. 377 Parsed relative_parsed; // Everything but the scheme is valid. 378 ParseAfterScheme(relative_url, relative_component.end(), 379 relative_component.begin, &relative_parsed); 380 381 // Now we can just use the replacement function to replace all the necessary 382 // parts of the old URL with the new one. 383 Replacements<CHAR> replacements; 384 replacements.SetUsername(relative_url, relative_parsed.username); 385 replacements.SetPassword(relative_url, relative_parsed.password); 386 replacements.SetHost(relative_url, relative_parsed.host); 387 replacements.SetPort(relative_url, relative_parsed.port); 388 replacements.SetPath(relative_url, relative_parsed.path); 389 replacements.SetQuery(relative_url, relative_parsed.query); 390 replacements.SetRef(relative_url, relative_parsed.ref); 391 392 return ReplaceStandardURL(base_url, base_parsed, replacements, 393 query_converter, output, out_parsed); 394} 395 396// Resolves a relative URL that happens to be an absolute file path. Examples 397// include: "//hostname/path", "/c:/foo", and "//hostname/c:/foo". 398template<typename CHAR> 399bool DoResolveAbsoluteFile(const CHAR* relative_url, 400 const Component& relative_component, 401 CharsetConverter* query_converter, 402 CanonOutput* output, 403 Parsed* out_parsed) { 404 // Parse the file URL. The file URl parsing function uses the same logic 405 // as we do for determining if the file is absolute, in which case it will 406 // not bother to look for a scheme. 407 Parsed relative_parsed; 408 ParseFileURL(&relative_url[relative_component.begin], relative_component.len, 409 &relative_parsed); 410 411 return CanonicalizeFileURL(&relative_url[relative_component.begin], 412 relative_component.len, relative_parsed, 413 query_converter, output, out_parsed); 414} 415 416// TODO(brettw) treat two slashes as root like Mozilla for FTP? 417template<typename CHAR> 418bool DoResolveRelativeURL(const char* base_url, 419 const Parsed& base_parsed, 420 bool base_is_file, 421 const CHAR* relative_url, 422 const Component& relative_component, 423 CharsetConverter* query_converter, 424 CanonOutput* output, 425 Parsed* out_parsed) { 426 // Starting point for our output parsed. We'll fix what we change. 427 *out_parsed = base_parsed; 428 429 // Sanity check: the input should have a host or we'll break badly below. 430 // We can only resolve relative URLs with base URLs that have hosts and 431 // paths (even the default path of "/" is OK). 432 // 433 // We allow hosts with no length so we can handle file URLs, for example. 434 if (base_parsed.path.len <= 0) { 435 // On error, return the input (resolving a relative URL on a non-relative 436 // base = the base). 437 int base_len = base_parsed.Length(); 438 for (int i = 0; i < base_len; i++) 439 output->push_back(base_url[i]); 440 return false; 441 } 442 443 if (relative_component.len <= 0) { 444 // Empty relative URL, leave unchanged, only removing the ref component. 445 int base_len = base_parsed.Length(); 446 base_len -= base_parsed.ref.len + 1; 447 out_parsed->ref.reset(); 448 output->Append(base_url, base_len); 449 return true; 450 } 451 452 int num_slashes = CountConsecutiveSlashes( 453 relative_url, relative_component.begin, relative_component.end()); 454 455#ifdef WIN32 456 // On Windows, two slashes for a file path (regardless of which direction 457 // they are) means that it's UNC. Two backslashes on any base scheme mean 458 // that it's an absolute UNC path (we use the base_is_file flag to control 459 // how strict the UNC finder is). 460 // 461 // We also allow Windows absolute drive specs on any scheme (for example 462 // "c:\foo") like IE does. There must be no preceeding slashes in this 463 // case (we reject anything like "/c:/foo") because that should be treated 464 // as a path. For file URLs, we allow any number of slashes since that would 465 // be setting the path. 466 // 467 // This assumes the absolute path resolver handles absolute URLs like this 468 // properly. DoCanonicalize does this. 469 int after_slashes = relative_component.begin + num_slashes; 470 if (DoesBeginUNCPath(relative_url, relative_component.begin, 471 relative_component.end(), !base_is_file) || 472 ((num_slashes == 0 || base_is_file) && 473 DoesBeginWindowsDriveSpec( 474 relative_url, after_slashes, relative_component.end()))) { 475 return DoResolveAbsoluteFile(relative_url, relative_component, 476 query_converter, output, out_parsed); 477 } 478#else 479 // Other platforms need explicit handling for file: URLs with multiple 480 // slashes because the generic scheme parsing always extracts a host, but a 481 // file: URL only has a host if it has exactly 2 slashes. Even if it does 482 // have a host, we want to use the special host detection logic for file 483 // URLs provided by DoResolveAbsoluteFile(), as opposed to the generic host 484 // detection logic, for consistency with parsing file URLs from scratch. 485 // This also handles the special case where the URL is only slashes, 486 // since that doesn't have a host part either. 487 if (base_is_file && 488 (num_slashes >= 2 || num_slashes == relative_component.len)) { 489 return DoResolveAbsoluteFile(relative_url, relative_component, 490 query_converter, output, out_parsed); 491 } 492#endif 493 494 // Any other double-slashes mean that this is relative to the scheme. 495 if (num_slashes >= 2) { 496 return DoResolveRelativeHost(base_url, base_parsed, 497 relative_url, relative_component, 498 query_converter, output, out_parsed); 499 } 500 501 // When we get here, we know that the relative URL is on the same host. 502 return DoResolveRelativePath(base_url, base_parsed, base_is_file, 503 relative_url, relative_component, 504 query_converter, output, out_parsed); 505} 506 507} // namespace 508 509bool IsRelativeURL(const char* base, 510 const Parsed& base_parsed, 511 const char* fragment, 512 int fragment_len, 513 bool is_base_hierarchical, 514 bool* is_relative, 515 Component* relative_component) { 516 return DoIsRelativeURL<char>( 517 base, base_parsed, fragment, fragment_len, is_base_hierarchical, 518 is_relative, relative_component); 519} 520 521bool IsRelativeURL(const char* base, 522 const Parsed& base_parsed, 523 const base::char16* fragment, 524 int fragment_len, 525 bool is_base_hierarchical, 526 bool* is_relative, 527 Component* relative_component) { 528 return DoIsRelativeURL<base::char16>( 529 base, base_parsed, fragment, fragment_len, is_base_hierarchical, 530 is_relative, relative_component); 531} 532 533bool ResolveRelativeURL(const char* base_url, 534 const Parsed& base_parsed, 535 bool base_is_file, 536 const char* relative_url, 537 const Component& relative_component, 538 CharsetConverter* query_converter, 539 CanonOutput* output, 540 Parsed* out_parsed) { 541 return DoResolveRelativeURL<char>( 542 base_url, base_parsed, base_is_file, relative_url, 543 relative_component, query_converter, output, out_parsed); 544} 545 546bool ResolveRelativeURL(const char* base_url, 547 const Parsed& base_parsed, 548 bool base_is_file, 549 const base::char16* relative_url, 550 const Component& relative_component, 551 CharsetConverter* query_converter, 552 CanonOutput* output, 553 Parsed* out_parsed) { 554 return DoResolveRelativeURL<base::char16>( 555 base_url, base_parsed, base_is_file, relative_url, 556 relative_component, query_converter, output, out_parsed); 557} 558 559} // namespace url 560