src/url.cc

   1 /*
   2  * Copyright (C) 1996-2017 The Squid Software Foundation and contributors
   3  *
   4  * Squid software is distributed under GPLv2+ license and includes
   5  * contributions from numerous individuals and organizations.
   6  * Please see the COPYING and CONTRIBUTORS files for details.
   7  */
   8
   9 /* DEBUG: section 23    URL Parsing */
  10
  11 #include "squid.h"
  12 #include "globals.h"
  13 #include "HttpRequest.h"
  14 #include "rfc1738.h"
  15 #include "SquidConfig.h"
  16 #include "SquidString.h"
  17 #include "URL.h"
  18
  19 static bool urlParseFinish(const HttpRequestMethod& method,
  20                            const AnyP::ProtocolType protocol,
  21                            const char *const protoStr,
  22                            const char *const urlpath,
  23                            const char *const host,
  24                            const SBuf &login,
  25                            const int port,
  26                            HttpRequest &request);
  27 static bool urnParse(const HttpRequestMethod& method, char *urn, HttpRequest &request);
  28 static const char valid_hostname_chars_u[] =
  29     "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
  30     "abcdefghijklmnopqrstuvwxyz"
  31     "0123456789-._"
  32     "[:]"
  33     ;
  34 static const char valid_hostname_chars[] =
  35     "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
  36     "abcdefghijklmnopqrstuvwxyz"
  37     "0123456789-."
  38     "[:]"
  39     ;
  40
  41 const SBuf &
  42 URL::Asterisk()
  43 {
  44     static SBuf star("*");
  45     return star;
  46 }
  47
  48 const SBuf &
  49 URL::SlashPath()
  50 {
  51     static SBuf slash("/");
  52     return slash;
  53 }
  54
  55 void
  56 URL::host(const char *src)
  57 {
  58     hostAddr_.setEmpty();
  59     hostAddr_ = src;
  60     if (hostAddr_.isAnyAddr()) {
  61         xstrncpy(host_, src, sizeof(host_));
  62         hostIsNumeric_ = false;
  63     } else {
  64         hostAddr_.toHostStr(host_, sizeof(host_));
  65         debugs(23, 3, "given IP: " << hostAddr_);
  66         hostIsNumeric_ = 1;
  67     }
  68     touch();
  69 }
  70
  71 const SBuf &
  72 URL::path() const
  73 {
  74     // RFC 3986 section 3.3 says path can be empty (path-abempty).
  75     // RFC 7230 sections 2.7.3, 5.3.1, 5.7.2 - says path cannot be empty, default to "/"
  76     // at least when sending and using. We must still accept path-abempty as input.
  77     if (path_.isEmpty() && (scheme_ == AnyP::PROTO_HTTP || scheme_ == AnyP::PROTO_HTTPS))
  78         return SlashPath();
  79
  80     return path_;
  81 }
  82
  83 void
  84 urlInitialize(void)
  85 {
  86     debugs(23, 5, "urlInitialize: Initializing...");
  87     /* this ensures that the number of protocol strings is the same as
  88      * the enum slots allocated because the last enum is always 'MAX'.
  89      */
  90     assert(strcmp(AnyP::ProtocolType_str[AnyP::PROTO_MAX], "MAX") == 0);
  91     /*
  92      * These test that our matchDomainName() function works the
  93      * way we expect it to.
  94      */
  95     assert(0 == matchDomainName("foo.com", "foo.com"));
  96     assert(0 == matchDomainName(".foo.com", "foo.com"));
  97     assert(0 == matchDomainName("foo.com", ".foo.com"));
  98     assert(0 == matchDomainName(".foo.com", ".foo.com"));
  99     assert(0 == matchDomainName("x.foo.com", ".foo.com"));
 100     assert(0 == matchDomainName("y.x.foo.com", ".foo.com"));
 101     assert(0 != matchDomainName("x.foo.com", "foo.com"));
 102     assert(0 != matchDomainName("foo.com", "x.foo.com"));
 103     assert(0 != matchDomainName("bar.com", "foo.com"));
 104     assert(0 != matchDomainName(".bar.com", "foo.com"));
 105     assert(0 != matchDomainName(".bar.com", ".foo.com"));
 106     assert(0 != matchDomainName("bar.com", ".foo.com"));
 107     assert(0 < matchDomainName("zzz.com", "foo.com"));
 108     assert(0 > matchDomainName("aaa.com", "foo.com"));
 109     assert(0 == matchDomainName("FOO.com", "foo.COM"));
 110     assert(0 < matchDomainName("bfoo.com", "afoo.com"));
 111     assert(0 > matchDomainName("afoo.com", "bfoo.com"));
 112     assert(0 < matchDomainName("x-foo.com", ".foo.com"));
 113
 114     assert(0 == matchDomainName(".foo.com", ".foo.com", mdnRejectSubsubDomains));
 115     assert(0 == matchDomainName("x.foo.com", ".foo.com", mdnRejectSubsubDomains));
 116     assert(0 != matchDomainName("y.x.foo.com", ".foo.com", mdnRejectSubsubDomains));
 117     assert(0 != matchDomainName(".x.foo.com", ".foo.com", mdnRejectSubsubDomains));
 118
 119     assert(0 == matchDomainName("*.foo.com", "x.foo.com", mdnHonorWildcards));
 120     assert(0 == matchDomainName("*.foo.com", ".x.foo.com", mdnHonorWildcards));
 121     assert(0 == matchDomainName("*.foo.com", ".foo.com", mdnHonorWildcards));
 122     assert(0 != matchDomainName("*.foo.com", "foo.com", mdnHonorWildcards));
 123
 124     /* more cases? */
 125 }
 126
 127 /**
 128  * Parse the scheme name from string b, into protocol type.
 129  * The string must be 0-terminated.
 130  */
 131 AnyP::ProtocolType
 132 urlParseProtocol(const char *b)
 133 {
 134     // make e point to the ':' character
 135     const char *e = b + strcspn(b, ":");
 136     int len = e - b;
 137
 138     /* test common stuff first */
 139
 140     if (strncasecmp(b, "http", len) == 0)
 141         return AnyP::PROTO_HTTP;
 142
 143     if (strncasecmp(b, "ftp", len) == 0)
 144         return AnyP::PROTO_FTP;
 145
 146     if (strncasecmp(b, "https", len) == 0)
 147         return AnyP::PROTO_HTTPS;
 148
 149     if (strncasecmp(b, "file", len) == 0)
 150         return AnyP::PROTO_FTP;
 151
 152     if (strncasecmp(b, "coap", len) == 0)
 153         return AnyP::PROTO_COAP;
 154
 155     if (strncasecmp(b, "coaps", len) == 0)
 156         return AnyP::PROTO_COAPS;
 157
 158     if (strncasecmp(b, "gopher", len) == 0)
 159         return AnyP::PROTO_GOPHER;
 160
 161     if (strncasecmp(b, "wais", len) == 0)
 162         return AnyP::PROTO_WAIS;
 163
 164     if (strncasecmp(b, "cache_object", len) == 0)
 165         return AnyP::PROTO_CACHE_OBJECT;
 166
 167     if (strncasecmp(b, "urn", len) == 0)
 168         return AnyP::PROTO_URN;
 169
 170     if (strncasecmp(b, "whois", len) == 0)
 171         return AnyP::PROTO_WHOIS;
 172
 173     if (len > 0)
 174         return AnyP::PROTO_UNKNOWN;
 175
 176     return AnyP::PROTO_NONE;
 177 }
 178
 179 /*
 180  * Parse a URI/URL.
 181  *
 182  * Stores parsed values in the `request` argument.
 183  *
 184  * This abuses HttpRequest as a way of representing the parsed url
 185  * and its components.
 186  * method is used to switch parsers and to init the HttpRequest.
 187  * If method is Http::METHOD_CONNECT, then rather than a URL a hostname:port is
 188  * looked for.
 189  * The url is non const so that if its too long we can NULL-terminate it in place.
 190  */
 191
 192 /*
 193  * This routine parses a URL. Its assumed that the URL is complete -
 194  * ie, the end of the string is the end of the URL. Don't pass a partial
 195  * URL here as this routine doesn't have any way of knowing whether
 196  * its partial or not (ie, it handles the case of no trailing slash as
 197  * being "end of host with implied path of /".
 198  */
 199 bool
 200 urlParse(const HttpRequestMethod& method, char *url, HttpRequest &request)
 201 {
 202     LOCAL_ARRAY(char, proto, MAX_URL);
 203     LOCAL_ARRAY(char, login, MAX_URL);
 204     LOCAL_ARRAY(char, host, MAX_URL);
 205     LOCAL_ARRAY(char, urlpath, MAX_URL);
 206     char *t = NULL;
 207     char *q = NULL;
 208     int port;
 209     AnyP::ProtocolType protocol = AnyP::PROTO_NONE;
 210     int l;
 211     int i;
 212     const char *src;
 213     char *dst;
 214     proto[0] = host[0] = urlpath[0] = login[0] = '\0';
 215
 216     if ((l = strlen(url)) + Config.appendDomainLen > (MAX_URL - 1)) {
 217         /* terminate so it doesn't overflow other buffers */
 218         *(url + (MAX_URL >> 1)) = '\0';
 219         debugs(23, DBG_IMPORTANT, "urlParse: URL too large (" << l << " bytes)");
 220         return false;
 221     }
 222     if (method == Http::METHOD_CONNECT) {
 223         port = CONNECT_PORT;
 224
 225         if (sscanf(url, "[%[^]]]:%d", host, &port) < 1)
 226             if (sscanf(url, "%[^:]:%d", host, &port) < 1)
 227                 return false;
 228
 229     } else if ((method == Http::METHOD_OPTIONS || method == Http::METHOD_TRACE) &&
 230                URL::Asterisk().cmp(url) == 0) {
 231         protocol = AnyP::PROTO_HTTP;
 232         port = 80; // or the slow way ...  AnyP::UriScheme(protocol,"http").defaultPort();
 233         return urlParseFinish(method, protocol, "http", url, host, SBuf(), port, request);
 234     } else if (!strncmp(url, "urn:", 4)) {
 235         return urnParse(method, url, request);
 236     } else {
 237         /* Parse the URL: */
 238         src = url;
 239         i = 0;
 240         /* Find first : - everything before is protocol */
 241         for (i = 0, dst = proto; i < l && *src != ':'; ++i, ++src, ++dst) {
 242             *dst = *src;
 243         }
 244         if (i >= l)
 245             return false;
 246         *dst = '\0';
 247
 248         /* Then its :// */
 249         if ((i+3) > l || *src != ':' || *(src + 1) != '/' || *(src + 2) != '/')
 250             return false;
 251         i += 3;
 252         src += 3;
 253
 254         /* Then everything until first /; thats host (and port; which we'll look for here later) */
 255         // bug 1881: If we don't get a "/" then we imply it was there
 256         // bug 3074: We could just be given a "?" or "#". These also imply "/"
 257         // bug 3233: whitespace is also a hostname delimiter.
 258         for (dst = host; i < l && *src != '/' && *src != '?' && *src != '#' && *src != '\0' && !xisspace(*src); ++i, ++src, ++dst) {
 259             *dst = *src;
 260         }
 261
 262         /*
 263          * We can't check for "i >= l" here because we could be at the end of the line
 264          * and have a perfectly valid URL w/ no trailing '/'. In this case we assume we've
 265          * been -given- a valid URL and the path is just '/'.
 266          */
 267         if (i > l)
 268             return false;
 269         *dst = '\0';
 270
 271         // bug 3074: received 'path' starting with '?', '#', or '\0' implies '/'
 272         if (*src == '?' || *src == '#' || *src == '\0') {
 273             urlpath[0] = '/';
 274             dst = &urlpath[1];
 275         } else {
 276             dst = urlpath;
 277         }
 278         /* Then everything from / (inclusive) until \r\n or \0 - thats urlpath */
 279         for (; i < l && *src != '\r' && *src != '\n' && *src != '\0'; ++i, ++src, ++dst) {
 280             *dst = *src;
 281         }
 282
 283         /* We -could- be at the end of the buffer here */
 284         if (i > l)
 285             return false;
 286         /* If the URL path is empty we set it to be "/" */
 287         if (dst == urlpath) {
 288             *dst = '/';
 289             ++dst;
 290         }
 291         *dst = '\0';
 292
 293         protocol = urlParseProtocol(proto);
 294         port = AnyP::UriScheme(protocol).defaultPort();
 295
 296         /* Is there any login information? (we should eventually parse it above) */
 297         t = strrchr(host, '@');
 298         if (t != NULL) {
 299             strncpy((char *) login, (char *) host, sizeof(login)-1);
 300             login[sizeof(login)-1] = '\0';
 301             t = strrchr(login, '@');
 302             *t = 0;
 303             strncpy((char *) host, t + 1, sizeof(host)-1);
 304             host[sizeof(host)-1] = '\0';
 305             // Bug 4498: URL-unescape the login info after extraction
 306             rfc1738_unescape(login);
 307         }
 308
 309         /* Is there any host information? (we should eventually parse it above) */
 310         if (*host == '[') {
 311             /* strip any IPA brackets. valid under IPv6. */
 312             dst = host;
 313             /* only for IPv6 sadly, pre-IPv6/URL code can't handle the clean result properly anyway. */
 314             src = host;
 315             ++src;
 316             l = strlen(host);
 317             i = 1;
 318             for (; i < l && *src != ']' && *src != '\0'; ++i, ++src, ++dst) {
 319                 *dst = *src;
 320             }
 321
 322             /* we moved in-place, so truncate the actual hostname found */
 323             *dst = '\0';
 324             ++dst;
 325
 326             /* skip ahead to either start of port, or original EOS */
 327             while (*dst != '\0' && *dst != ':')
 328                 ++dst;
 329             t = dst;
 330         } else {
 331             t = strrchr(host, ':');
 332
 333             if (t != strchr(host,':') ) {
 334                 /* RFC 2732 states IPv6 "SHOULD" be bracketed. allowing for times when its not. */
 335                 /* RFC 3986 'update' simply modifies this to an "is" with no emphasis at all! */
 336                 /* therefore we MUST accept the case where they are not bracketed at all. */
 337                 t = NULL;
 338             }
 339         }
 340
 341         // Bug 3183 sanity check: If scheme is present, host must be too.
 342         if (protocol != AnyP::PROTO_NONE && host[0] == '\0') {
 343             debugs(23, DBG_IMPORTANT, "SECURITY ALERT: Missing hostname in URL '" << url << "'. see access.log for details.");
 344             return false;
 345         }
 346
 347         if (t && *t == ':') {
 348             *t = '\0';
 349             ++t;
 350             port = atoi(t);
 351         }
 352     }
 353
 354     for (t = host; *t; ++t)
 355         *t = xtolower(*t);
 356
 357     if (stringHasWhitespace(host)) {
 358         if (URI_WHITESPACE_STRIP == Config.uri_whitespace) {
 359             t = q = host;
 360             while (*t) {
 361                 if (!xisspace(*t)) {
 362                     *q = *t;
 363                     ++q;
 364                 }
 365                 ++t;
 366             }
 367             *q = '\0';
 368         }
 369     }
 370
 371     debugs(23, 3, "urlParse: Split URL '" << url << "' into proto='" << proto << "', host='" << host << "', port='" << port << "', path='" << urlpath << "'");
 372
 373     if (Config.onoff.check_hostnames && strspn(host, Config.onoff.allow_underscore ? valid_hostname_chars_u : valid_hostname_chars) != strlen(host)) {
 374         debugs(23, DBG_IMPORTANT, "urlParse: Illegal character in hostname '" << host << "'");
 375         return false;
 376     }
 377
 378     /* For IPV6 addresses also check for a colon */
 379     if (Config.appendDomain && !strchr(host, '.') && !strchr(host, ':'))
 380         strncat(host, Config.appendDomain, SQUIDHOSTNAMELEN - strlen(host) - 1);
 381
 382     /* remove trailing dots from hostnames */
 383     while ((l = strlen(host)) > 0 && host[--l] == '.')
 384         host[l] = '\0';
 385
 386     /* reject duplicate or leading dots */
 387     if (strstr(host, "..") || *host == '.') {
 388         debugs(23, DBG_IMPORTANT, "urlParse: Illegal hostname '" << host << "'");
 389         return false;
 390     }
 391
 392     if (port < 1 || port > 65535) {
 393         debugs(23, 3, "urlParse: Invalid port '" << port << "'");
 394         return false;
 395     }
 396
 397 #if HARDCODE_DENY_PORTS
 398     /* These ports are filtered in the default squid.conf, but
 399      * maybe someone wants them hardcoded... */
 400     if (port == 7 || port == 9 || port == 19) {
 401         debugs(23, DBG_CRITICAL, "urlParse: Deny access to port " << port);
 402         return false;
 403     }
 404 #endif
 405
 406     if (stringHasWhitespace(urlpath)) {
 407         debugs(23, 2, "urlParse: URI has whitespace: {" << url << "}");
 408
 409         switch (Config.uri_whitespace) {
 410
 411         case URI_WHITESPACE_DENY:
 412             return false;
 413
 414         case URI_WHITESPACE_ALLOW:
 415             break;
 416
 417         case URI_WHITESPACE_ENCODE:
 418             t = rfc1738_escape_unescaped(urlpath);
 419             xstrncpy(urlpath, t, MAX_URL);
 420             break;
 421
 422         case URI_WHITESPACE_CHOP:
 423             *(urlpath + strcspn(urlpath, w_space)) = '\0';
 424             break;
 425
 426         case URI_WHITESPACE_STRIP:
 427         default:
 428             t = q = urlpath;
 429             while (*t) {
 430                 if (!xisspace(*t)) {
 431                     *q = *t;
 432                     ++q;
 433                 }
 434                 ++t;
 435             }
 436             *q = '\0';
 437         }
 438     }
 439
 440     return urlParseFinish(method, protocol, proto, urlpath, host, SBuf(login), port, request);
 441 }
 442
 443 /**
 444  * Update request with parsed URI data.  If the request arg is
 445  * non-NULL, put parsed values there instead of allocating a new
 446  * HttpRequest.
 447  */
 448 static bool
 449 urlParseFinish(const HttpRequestMethod& method,
 450                const AnyP::ProtocolType protocol,
 451                const char *const protoStr, // for unknown protocols
 452                const char *const urlpath,
 453                const char *const host,
 454                const SBuf &login,
 455                const int port,
 456                HttpRequest &request)
 457 {
 458     request.initHTTP(method, protocol, protoStr, urlpath);
 459     request.url.host(host);
 460     request.url.userInfo(login);
 461     request.url.port(port);
 462     return true;
 463 }
 464
 465 static bool
 466 urnParse(const HttpRequestMethod& method, char *urn, HttpRequest &request)
 467 {
 468     debugs(50, 5, "urnParse: " << urn);
 469     request.initHTTP(method, AnyP::PROTO_URN, "urn", urn + 4);
 470     return true;
 471 }
 472
 473 void
 474 URL::touch()
 475 {
 476     absolute_.clear();
 477     authorityHttp_.clear();
 478     authorityWithPort_.clear();
 479 }
 480
 481 SBuf &
 482 URL::authority(bool requirePort) const
 483 {
 484     if (authorityHttp_.isEmpty()) {
 485
 486         // both formats contain Host/IP
 487         authorityWithPort_.append(host());
 488         authorityHttp_ = authorityWithPort_;
 489
 490         // authorityForm_ only has :port if it is non-default
 491         authorityWithPort_.appendf(":%u",port());
 492         if (port() != getScheme().defaultPort())
 493             authorityHttp_ = authorityWithPort_;
 494     }
 495
 496     return requirePort ? authorityWithPort_ : authorityHttp_;
 497 }
 498
 499 SBuf &
 500 URL::absolute() const
 501 {
 502     if (absolute_.isEmpty()) {
 503         // TODO: most URL will be much shorter, avoid allocating this much
 504         absolute_.reserveCapacity(MAX_URL);
 505
 506         absolute_.append(getScheme().image());
 507         absolute_.append(":",1);
 508         if (getScheme() != AnyP::PROTO_URN) {
 509             absolute_.append("//", 2);
 510             const bool omitUserInfo = getScheme() == AnyP::PROTO_HTTP ||
 511                                       getScheme() != AnyP::PROTO_HTTPS ||
 512                                       userInfo().isEmpty();
 513             if (!omitUserInfo) {
 514                 absolute_.append(userInfo());
 515                 absolute_.append("@", 1);
 516             }
 517             absolute_.append(authority());
 518         }
 519         absolute_.append(path());
 520     }
 521
 522     return absolute_;
 523 }
 524
 525 /** \todo AYJ: Performance: This is an *almost* duplicate of HttpRequest::effectiveRequestUri(). But elides the query-string.
 526  *        After copying it on in the first place! Would be less code to merge the two with a flag parameter.
 527  *        and never copy the query-string part in the first place
 528  */
 529 char *
 530 urlCanonicalClean(const HttpRequest * request)
 531 {
 532     LOCAL_ARRAY(char, buf, MAX_URL);
 533
 534     snprintf(buf, sizeof(buf), SQUIDSBUFPH, SQUIDSBUFPRINT(request->effectiveRequestUri()));
 535     buf[sizeof(buf)-1] = '\0';
 536
 537     // URN, CONNECT method, and non-stripped URIs can go straight out
 538     if (Config.onoff.strip_query_terms && !(request->method == Http::METHOD_CONNECT || request->url.getScheme() == AnyP::PROTO_URN)) {
 539         // strip anything AFTER a question-mark
 540         // leaving the '?' in place
 541         if (auto t = strchr(buf, '?')) {
 542             *(++t) = '\0';
 543         }
 544     }
 545
 546     if (stringHasCntl(buf))
 547         xstrncpy(buf, rfc1738_escape_unescaped(buf), MAX_URL);
 548
 549     return buf;
 550 }
 551
 552 /**
 553  * Yet another alternative to urlCanonical.
 554  * This one adds the https:// parts to Http::METHOD_CONNECT URL
 555  * for use in error page outputs.
 556  * Luckily we can leverage the others instead of duplicating.
 557  */
 558 const char *
 559 urlCanonicalFakeHttps(const HttpRequest * request)
 560 {
 561     LOCAL_ARRAY(char, buf, MAX_URL);
 562
 563     // method CONNECT and port HTTPS
 564     if (request->method == Http::METHOD_CONNECT && request->url.port() == 443) {
 565         snprintf(buf, MAX_URL, "https://%s/*", request->url.host());
 566         return buf;
 567     }
 568
 569     // else do the normal complete canonical thing.
 570     return urlCanonicalClean(request);
 571 }
 572
 573 /*
 574  * Test if a URL is relative.
 575  *
 576  * RFC 2396, Section 5 (Page 17) implies that in a relative URL, a '/' will
 577  * appear before a ':'.
 578  */
 579 bool
 580 urlIsRelative(const char *url)
 581 {
 582     const char *p;
 583
 584     if (url == NULL) {
 585         return (false);
 586     }
 587     if (*url == '\0') {
 588         return (false);
 589     }
 590
 591     for (p = url; *p != '\0' && *p != ':' && *p != '/'; ++p);
 592
 593     if (*p == ':') {
 594         return (false);
 595     }
 596     return (true);
 597 }
 598
 599 /*
 600  * Convert a relative URL to an absolute URL using the context of a given
 601  * request.
 602  *
 603  * It is assumed that you have already ensured that the URL is relative.
 604  *
 605  * If NULL is returned it is an indication that the method in use in the
 606  * request does not distinguish between relative and absolute and you should
 607  * use the url unchanged.
 608  *
 609  * If non-NULL is returned, it is up to the caller to free the resulting
 610  * memory using safe_free().
 611  */
 612 char *
 613 urlMakeAbsolute(const HttpRequest * req, const char *relUrl)
 614 {
 615
 616     if (req->method.id() == Http::METHOD_CONNECT) {
 617         return (NULL);
 618     }
 619
 620     char *urlbuf = (char *)xmalloc(MAX_URL * sizeof(char));
 621
 622     if (req->url.getScheme() == AnyP::PROTO_URN) {
 623         // XXX: this is what the original code did, but it seems to break the
 624         // intended behaviour of this function. It returns the stored URN path,
 625         // not converting the given one into a URN...
 626         snprintf(urlbuf, MAX_URL, SQUIDSBUFPH, SQUIDSBUFPRINT(req->url.absolute()));
 627         return (urlbuf);
 628     }
 629
 630     SBuf authorityForm = req->url.authority(); // host[:port]
 631     const SBuf &scheme = req->url.getScheme().image();
 632     size_t urllen = snprintf(urlbuf, MAX_URL, SQUIDSBUFPH "://" SQUIDSBUFPH "%s" SQUIDSBUFPH,
 633                              SQUIDSBUFPRINT(scheme),
 634                              SQUIDSBUFPRINT(req->url.userInfo()),
 635                              !req->url.userInfo().isEmpty() ? "@" : "",
 636                              SQUIDSBUFPRINT(authorityForm));
 637
 638     // if the first char is '/' assume its a relative path
 639     // XXX: this breaks on scheme-relative URLs,
 640     // but we should not see those outside ESI, and rarely there.
 641     // XXX: also breaks on any URL containing a '/' in the query-string portion
 642     if (relUrl[0] == '/') {
 643         xstrncpy(&urlbuf[urllen], relUrl, MAX_URL - urllen - 1);
 644     } else {
 645         SBuf path = req->url.path();
 646         SBuf::size_type lastSlashPos = path.rfind('/');
 647
 648         if (lastSlashPos == SBuf::npos) {
 649             // replace the whole path with the given bit(s)
 650             urlbuf[urllen] = '/';
 651             ++urllen;
 652             xstrncpy(&urlbuf[urllen], relUrl, MAX_URL - urllen - 1);
 653         } else {
 654             // replace only the last (file?) segment with the given bit(s)
 655             ++lastSlashPos;
 656             if (lastSlashPos > MAX_URL - urllen - 1) {
 657                 // XXX: crops bits in the middle of the combined URL.
 658                 lastSlashPos = MAX_URL - urllen - 1;
 659             }
 660             SBufToCstring(&urlbuf[urllen], path.substr(0,lastSlashPos));
 661             urllen += lastSlashPos;
 662             if (urllen + 1 < MAX_URL) {
 663                 xstrncpy(&urlbuf[urllen], relUrl, MAX_URL - urllen - 1);
 664             }
 665         }
 666     }
 667
 668     return (urlbuf);
 669 }
 670
 671 int
 672 matchDomainName(const char *h, const char *d, uint flags)
 673 {
 674     int dl;
 675     int hl;
 676
 677     const bool hostIncludesSubdomains = (*h == '.');
 678     while ('.' == *h)
 679         ++h;
 680
 681     hl = strlen(h);
 682
 683     if (hl == 0)
 684         return -1;
 685
 686     dl = strlen(d);
 687
 688     /*
 689      * Start at the ends of the two strings and work towards the
 690      * beginning.
 691      */
 692     while (xtolower(h[--hl]) == xtolower(d[--dl])) {
 693         if (hl == 0 && dl == 0) {
 694             /*
 695              * We made it all the way to the beginning of both
 696              * strings without finding any difference.
 697              */
 698             return 0;
 699         }
 700
 701         if (0 == hl) {
 702             /*
 703              * The host string is shorter than the domain string.
 704              * There is only one case when this can be a match.
 705              * If the domain is just one character longer, and if
 706              * that character is a leading '.' then we call it a
 707              * match.
 708              */
 709
 710             if (1 == dl && '.' == d[0])
 711                 return 0;
 712             else
 713                 return -1;
 714         }
 715
 716         if (0 == dl) {
 717             /*
 718              * The domain string is shorter than the host string.
 719              * This is a match only if the first domain character
 720              * is a leading '.'.
 721              */
 722
 723             if ('.' == d[0]) {
 724                 if (flags & mdnRejectSubsubDomains) {
 725                     // Check for sub-sub domain and reject
 726                     while(--hl >= 0 && h[hl] != '.');
 727                     if (hl < 0) {
 728                         // No sub-sub domain found, but reject if there is a
 729                         // leading dot in given host string (which is removed
 730                         // before the check is started).
 731                         return hostIncludesSubdomains ? 1 : 0;
 732                     } else
 733                         return 1; // sub-sub domain, reject
 734                 } else
 735                     return 0;
 736             } else
 737                 return 1;
 738         }
 739     }
 740
 741     /*
 742      * We found different characters in the same position (from the end).
 743      */
 744
 745     // If the h has a form of "*.foo.com" and d has a form of "x.foo.com"
 746     // then the h[hl] points to '*', h[hl+1] to '.' and d[dl] to 'x'
 747     // The following checks are safe, the "h[hl + 1]" in the worst case is '\0'.
 748     if ((flags & mdnHonorWildcards) && h[hl] == '*' && h[hl + 1] == '.')
 749         return 0;
 750
 751     /*
 752      * If one of those character is '.' then its special.  In order
 753      * for splay tree sorting to work properly, "x-foo.com" must
 754      * be greater than ".foo.com" even though '-' is less than '.'.
 755      */
 756     if ('.' == d[dl])
 757         return 1;
 758
 759     if ('.' == h[hl])
 760         return -1;
 761
 762     return (xtolower(h[hl]) - xtolower(d[dl]));
 763 }
 764
 765 /*
 766  * return true if we can serve requests for this method.
 767  */
 768 int
 769 urlCheckRequest(const HttpRequest * r)
 770 {
 771     int rc = 0;
 772     /* protocol "independent" methods
 773      *
 774      * actually these methods are specific to HTTP:
 775      * they are methods we recieve on our HTTP port,
 776      * and if we had a FTP listener would not be relevant
 777      * there.
 778      *
 779      * So, we should delegate them to HTTP. The problem is that we
 780      * do not have a default protocol from the client side of HTTP.
 781      */
 782
 783     if (r->method == Http::METHOD_CONNECT)
 784         return 1;
 785
 786     // we support OPTIONS and TRACE directed at us (with a 501 reply, for now)
 787     // we also support forwarding OPTIONS and TRACE, except for the *-URI ones
 788     if (r->method == Http::METHOD_OPTIONS || r->method == Http::METHOD_TRACE)
 789         return (r->header.getInt64(Http::HdrType::MAX_FORWARDS) == 0 || r->url.path() != URL::Asterisk());
 790
 791     if (r->method == Http::METHOD_PURGE)
 792         return 1;
 793
 794     /* does method match the protocol? */
 795     switch (r->url.getScheme()) {
 796
 797     case AnyP::PROTO_URN:
 798
 799     case AnyP::PROTO_HTTP:
 800
 801     case AnyP::PROTO_CACHE_OBJECT:
 802         rc = 1;
 803         break;
 804
 805     case AnyP::PROTO_FTP:
 806
 807         if (r->method == Http::METHOD_PUT)
 808             rc = 1;
 809
 810     case AnyP::PROTO_GOPHER:
 811
 812     case AnyP::PROTO_WAIS:
 813
 814     case AnyP::PROTO_WHOIS:
 815         if (r->method == Http::METHOD_GET)
 816             rc = 1;
 817         else if (r->method == Http::METHOD_HEAD)
 818             rc = 1;
 819
 820         break;
 821
 822     case AnyP::PROTO_HTTPS:
 823 #if USE_OPENSSL
 824         rc = 1;
 825 #elif USE_GNUTLS
 826         rc = 1;
 827 #else
 828         /*
 829         * Squid can't originate an SSL connection, so it should
 830         * never receive an "https:" URL.  It should always be
 831         * CONNECT instead.
 832         */
 833         rc = 0;
 834 #endif
 835         break;
 836
 837     default:
 838         break;
 839     }
 840
 841     return rc;
 842 }
 843
 844 /*
 845  * Quick-n-dirty host extraction from a URL.  Steps:
 846  *      Look for a colon
 847  *      Skip any '/' after the colon
 848  *      Copy the next SQUID_MAXHOSTNAMELEN bytes to host[]
 849  *      Look for an ending '/' or ':' and terminate
 850  *      Look for login info preceeded by '@'
 851  */
 852
 853 class URLHostName
 854 {
 855
 856 public:
 857     char * extract(char const *url);
 858
 859 private:
 860     static char Host [SQUIDHOSTNAMELEN];
 861     void init(char const *);
 862     void findHostStart();
 863     void trimTrailingChars();
 864     void trimAuth();
 865     char const *hostStart;
 866     char const *url;
 867 };
 868
 869 char *
 870 urlHostname(const char *url)
 871 {
 872     return URLHostName().extract(url);
 873 }
 874
 875 char URLHostName::Host[SQUIDHOSTNAMELEN];
 876
 877 void
 878 URLHostName::init(char const *aUrl)
 879 {
 880     Host[0] = '\0';
 881     url = aUrl;
 882 }
 883
 884 void
 885 URLHostName::findHostStart()
 886 {
 887     if (NULL == (hostStart = strchr(url, ':')))
 888         return;
 889
 890     ++hostStart;
 891
 892     while (*hostStart != '\0' && *hostStart == '/')
 893         ++hostStart;
 894
 895     if (*hostStart == ']')
 896         ++hostStart;
 897 }
 898
 899 void
 900 URLHostName::trimTrailingChars()
 901 {
 902     char *t;
 903
 904     if ((t = strchr(Host, '/')))
 905         *t = '\0';
 906
 907     if ((t = strrchr(Host, ':')))
 908         *t = '\0';
 909
 910     if ((t = strchr(Host, ']')))
 911         *t = '\0';
 912 }
 913
 914 void
 915 URLHostName::trimAuth()
 916 {
 917     char *t;
 918
 919     if ((t = strrchr(Host, '@'))) {
 920         ++t;
 921         memmove(Host, t, strlen(t) + 1);
 922     }
 923 }
 924
 925 char *
 926 URLHostName::extract(char const *aUrl)
 927 {
 928     init(aUrl);
 929     findHostStart();
 930
 931     if (hostStart == NULL)
 932         return NULL;
 933
 934     xstrncpy(Host, hostStart, SQUIDHOSTNAMELEN);
 935
 936     trimTrailingChars();
 937
 938     trimAuth();
 939
 940     return Host;
 941 }
 942
 943 URL::URL(AnyP::UriScheme const &aScheme) :
 944     scheme_(aScheme),
 945     hostIsNumeric_(false),
 946     port_(0)
 947 {
 948     *host_=0;
 949 }
 950