uri.c

一个很有名的浏览器

				}

				mem_free(host);
			}
		}

#endif
		if (add_host)
			add_bytes_to_string(string, uri->host, uri->hostlen);

#ifdef CONFIG_IPV6
		if (uri->ipv6 && wants(URI_PORT)) add_char_to_string(string, ']');
#endif
 	}

 	if (wants(URI_PORT) || wants(URI_DEFAULT_PORT)) {
 		if (uri->portlen) {
			add_char_to_string(string, ':');
			add_bytes_to_string(string, uri->port, uri->portlen);

		} else if (wants(URI_DEFAULT_PORT)
			   && uri->protocol != PROTOCOL_USER) {
			/* For user protocols we don't know a default port.
			 * Should user protocols ports be configurable? */
			int port = get_protocol_port(uri->protocol);

			add_char_to_string(string, ':');
			add_long_to_string(string, port);
		}
	}

	/* Only add slash if we need to separate */
	if ((wants(URI_DATA) || wants(URI_POST))
	    && wants(~(URI_DATA | URI_PORT))
	    && get_protocol_need_slash_after_host(uri->protocol))
		add_char_to_string(string, '/');

	if (wants(URI_DATA) && uri->datalen)
		add_bytes_to_string(string, uri->data, uri->datalen);

	/* We can not test uri->datalen here since we need to always
	 * add '/'. */
	if (wants(URI_PATH) || wants(URI_FILENAME)) {
		unsigned char *filename = uri->data;
		unsigned char *pos;

		assertm(!wants(URI_FILENAME) || components == URI_FILENAME,
			"URI_FILENAME should be used alone %d", components);

		if (wants(URI_PATH) && !is_uri_dir_sep(uri, *filename)) {
			/* FIXME: Add correct separator */
			add_char_to_string(string, '/');
		}

		if (!uri->datalen) return string;

		for (pos = filename; *pos && !end_of_dir(*pos); pos++)
			if (wants(URI_FILENAME) && is_uri_dir_sep(uri, *pos))
				filename = pos + 1;

		return add_bytes_to_string(string, filename, pos - filename);
	}

	if (wants(URI_QUERY) && uri->datalen) {
		unsigned char *query = memchr(uri->data, '?', uri->datalen);

		assertm(URI_QUERY == components,
			"URI_QUERY should be used alone %d", components);

		if (!query) return string;

		query++;
		/* Check fragment and POST_CHAR */
		return add_bytes_to_string(string, query, strcspn(query, "#" POST_CHAR_S));
	}

	if (wants(URI_FRAGMENT) && uri->fragmentlen) {
		add_char_to_string(string, '#');
		add_bytes_to_string(string, uri->fragment, uri->fragmentlen);
	}

	if (wants(URI_POST) && uri->post) {
		add_char_to_string(string, POST_CHAR);
		add_to_string(string, uri->post);

	} else if (wants(URI_POST_INFO) && uri->post) {
		if (!strncmp(uri->post, "text/plain", 10)) {
			add_to_string(string, " (PLAIN TEXT DATA)");

		} else if (!strncmp(uri->post, "multipart/form-data;", 20)) {
			add_to_string(string, " (MULTIPART FORM DATA)");

		} else {
			add_to_string(string, " (POST DATA)");
		}

	}

	return string;
}

#undef wants

unsigned char *
get_uri_string(struct uri *uri, enum uri_component components)
{
	struct string string;

	if (init_string(&string)
	    && add_uri_to_string(&string, uri, components))
		return string.source;

	done_string(&string);
	return NULL;
}


struct string *
add_string_uri_to_string(struct string *string, unsigned char *uristring,
			 enum uri_component components)
{
	struct uri uri;

	if (parse_uri(&uri, uristring) != URI_ERRNO_OK)
		return NULL;

	return add_uri_to_string(string, &uri, components);
}


#define normalize_uri_reparse(str)	normalize_uri(NULL, str)
#define normalize_uri_noparse(uri)	normalize_uri(uri, struri(uri))

static unsigned char *
normalize_uri(struct uri *uri, unsigned char *uristring)
{
	unsigned char *parse_string = uristring;
	unsigned char *src, *dest, *path;
	int need_slash = 0;
	int parse = (uri == NULL);
	struct uri uri_struct;

	if (!uri) uri = &uri_struct;

	/* We need to get the real (proxied) URI but lowercase relevant URI
	 * parts along the way. */
	do {
		if (parse && parse_uri(uri, parse_string) != URI_ERRNO_OK)
			return uristring;

		assert(uri->data);

		/* This is a maybe not the right place but both join_urls() and
		 * get_translated_uri() through translate_url() calls this
		 * function and then it already works on and modifies an
		 * allocated copy. */
		convert_to_lowercase(uri->string, uri->protocollen);
		if (uri->hostlen) convert_to_lowercase(uri->host, uri->hostlen);

		parse = 1;
		parse_string = uri->data;
	} while (uri->protocol == PROTOCOL_PROXY);

	if (get_protocol_free_syntax(uri->protocol))
		return uristring;

	if (uri->protocol != PROTOCOL_UNKNOWN)
		need_slash = get_protocol_need_slash_after_host(uri->protocol);

	/* We want to start at the first slash to also reduce URIs like
	 * http://host//index.html to http://host/index.html */
	path = uri->data - need_slash;
	dest = src = path;

	/* This loop mangles the URI string by removing directory elevators and
	 * other cruft. Example: /.././etc////..//usr/ -> /usr/ */
	while (*dest) {
		/* If the following pieces are the LAST parts of URL, we remove
		 * them as well. See RFC 1808 for details. */

		if (end_of_dir(src[0])) {
			/* URL data contains no more path. */
			memmove(dest, src, strlen(src) + 1);
			break;
		}

		if (!is_uri_dir_sep(uri, src[0])) {
			/* This is to reduce indentation */

		} else if (src[1] == '.') {
			if (!src[2]) {
				/* /. - skip the dot */
				*dest++ = *src;
				*dest = 0;
				break;

			} else if (is_uri_dir_sep(uri, src[2])) {
				/* /./ - strip that.. */
				src += 2;
				continue;

			} else if (src[2] == '.'
				   && (is_uri_dir_sep(uri, src[3]) || !src[3])) {
				/* /../ or /.. - skip it and preceding element. */

				/* First back out the last incrementation of
				 * @dest (dest++) to get the position that was
				 * last asigned to. */
				if (dest > path) dest--;

				/* @dest might be pointing to a dir separator
				 * so we decrement before any testing. */
				while (dest > path) {
					dest--;
					if (is_uri_dir_sep(uri, *dest)) break;
				}

				if (!src[3]) {
					/* /.. - add ending slash and stop */
					*dest++ = *src;
					*dest = 0;
					break;
				}

				src += 3;
				continue;
			}

		} else if (is_uri_dir_sep(uri, src[1])) {
			/* // - ignore first '/'. */
			src += 1;
			continue;
		}

		/* We don't want to access memory past the NUL char. */
		*dest = *src++;
		if (*dest) dest++;
	}

	return uristring;
}

/* The 'file' scheme URI comes in and bastardized URI comes out which consists
 * of just the complete path to file/directory, which the dumb 'file' protocol
 * backend can understand. No host parts etc, that is what this function is
 * supposed to chew. */
static struct uri *
transform_file_url(struct uri *uri, unsigned char *cwd)
{
	unsigned char *path = uri->data;

	assert(uri->protocol == PROTOCOL_FILE && uri->data);

	/* Sort out the host part. We currently support only host "localhost"
	 * (plus empty host part will be assumed to be "localhost" as well).
	 * As our extensions, '.' will reference to the cwd on localhost
	 * (originally, when the first thing after file:// wasn't "localhost/",
	 * we assumed the cwd as well, and pretended that there's no host part
	 * at all) and '..' to the directory parent to cwd. Another extension
	 * is that if this is a DOS-like system, the first char in two-char
	 * host part is uppercase letter and the second char is a colon, it is
	 * assumed to be a local disk specification. */
	/* TODO: Use FTP for non-localhost hosts. --pasky */

	/* For URL "file://", we open the current directory. Some other
	 * browsers instead open root directory, but AFAIK the standard does
	 * not specify that and this was the original behaviour and it is more
	 * consistent with our file://./ notation. */

	/* Who would name their file/dir '...' ? */
	if (*path == '.' || !*path) {
		struct string dir;

		if (!init_string(&dir))
			return NULL;

		encode_uri_string(&dir, cwd, 0);

		/* Either we will end up with '//' and translate_directories()
		 * will shorten it or the '/' will mark the inserted cwd as a
		 * directory. */
		if (*path == '.') *path = '/';

		/* Insert the current working directory. */
		/* The offset is 7 == sizeof("file://") - 1. */
		insert_in_string(&struri(uri), 7, dir.source, dir.length);

		done_string(&dir);
		return uri;
	}

#ifdef DOS_FS
	if (isasciialpha(path[0]) && path[1] == ':' && dir_sep(path[2]))
		return NULL;
#endif

	for (; *path && !dir_sep(*path); path++);

	/* FIXME: We will in fact assume localhost even for non-local hosts,
	 * until we will support the FTP transformation. --pasky */

	memmove(uri->data, path, strlen(path) + 1);
	return uri;
}

static unsigned char *translate_url(unsigned char *url, unsigned char *cwd);

unsigned char *
join_urls(struct uri *base, unsigned char *rel)
{
	unsigned char *uristring, *path;
	int add_slash = 0;
	int translate = 0;
	int length = 0;

	/* See RFC 1808 */
	/* TODO: Support for ';' ? (see the RFC) --pasky */

	/* For '#', '?' and '//' we could use get_uri_string() but it might be
	 * too expensive since it uses granular allocation scheme. I wouldn't
	 * personally mind tho' because it would be cleaner. --jonas */
	if (rel[0] == '#') {
		/* Strip fragment and post part from the base URI and append
		 * the fragment string in @rel. */
		length  = base->fragment
			? base->fragment - struri(base) - 1
			: get_real_uri_length(base);

	} else if (rel[0] == '?') {
		/* Strip query, fragment and post part from the base URI and
		 * append the query string in @rel. */
		length  = base->fragment ? base->fragment - struri(base) - 1
					 : get_real_uri_length(base);

		uristring = memchr(base->data, '?', base->datalen);
		if (uristring) length = uristring - struri(base);

	} else if (rel[0] == '/' && rel[1] == '/') {
		if (!get_protocol_need_slashes(base->protocol))
			return NULL;

		/* Get `<protocol>:' from the base URI and append the `//' part
		 * from @rel. */
		length = base->protocollen + 1;

		/* We need to sanitize the relative part and add stuff like
		 * host slash. */
		translate = 1;
	}

	/* If one of the tests above set @length to something useful */
	if (length) {
		uristring = memacpy(struri(base), length);
		if (!uristring) return NULL;

		add_to_strn(&uristring, rel);

		if (translate) {
			unsigned char *translated;

			translated = translate_url(uristring, NULL);
			mem_free(uristring);
			return translated;
		}
		return normalize_uri_reparse(uristring);
	}

	/* Check if there is some protocol name to go for */
	length = get_protocol_length(rel);
	if (length) {
		switch (get_protocol(rel, length)) {
		case PROTOCOL_UNKNOWN:
		case PROTOCOL_PROXY:
			/* Mysteriously proxy URIs are breaking here ... */
			break;

		case PROTOCOL_FILE:
			/* FIXME: Use get_uri_string(base, URI_PATH) as cwd arg
			 * to translate_url(). */
		default:
			uristring = translate_url(rel, NULL);
			if (uristring) return uristring;
		}
	}

	assertm(base->data, "bad base url");
	if_assert_failed return NULL;

	path = base->data;

	/* Either is path blank, but we've slash char before, or path is not
	 * blank, but doesn't start by a slash (if we'd just stay along with
	 * is_uri_dir_sep(&uri, path[-1]) w/o all the surrounding crap, it
	 * should be enough, but I'm not sure and I don't want to break
	 * anything --pasky). */
	/* We skip first char of URL ('/') in parse_url() (ARGH). This
	 * is reason of all this bug-bearing magic.. */
	if (*path) {
		if (!is_uri_dir_sep(base, *path)) path--;
	} else {
		if (is_uri_dir_sep(base, path[-1])) path--;
	}

	if (!is_uri_dir_sep(base, rel[0])) {
		unsigned char *path_end;

		/* The URL is relative. */

		if (!*path) {
			/* There's no path in the URL, but we're going to add
			 * something there, and the something doesn't start by
			 * a slash. So we need to insert a slash after the base
			 * URL. Clever, eh? ;) */
			add_slash = 1;
		}

		for (path_end = path; *path_end; path_end++) {
			if (end_of_dir(*path_end)) break;
			/* Modify the path pointer, so that it'll always point
			 * above the last '/' in the URL; later, we'll copy the
			 * URL only _TO_ this point, and anything after last
			 * slash will be substituted by 'rel'. */
			if (is_uri_dir_sep(base, *path_end))
				path = path_end + 1;
		}
	}

	length = path - struri(base);
	uristring = mem_alloc(length + strlen(rel) + add_slash + 1);
	if (!uristring) return NULL;

	memcpy(uristring, struri(base), length);
	if (add_slash) uristring[length] = '/';
	strcpy(uristring + length + add_slash, rel);

	return normalize_uri_reparse(uristring);
}


/* Tries to figure out what protocol @newurl might be specifying by checking if
 * it exists as a file locally or by checking parts of the host name. */
static inline enum protocol
find_uri_protocol(unsigned char *newurl)
{
	unsigned char *ch;

	/* First see if it is a file so filenames that look like hostnames
	 * won't confuse us below. */
	if (check_uri_file(newurl) >= 0) return PROTOCOL_FILE;

	/* Yes, it would be simpler to make test for IPv6 address first,
	 * but it would result in confusing mix of ifdefs ;-). */
	/* FIXME: Ideas for improve protocol detection
	 *
	 * - Handle common hostnames. It could be part of the protocol backend
	 *   structure. [ www -> http, irc -> irc, news -> nntp, ... ]
	 *
	 * - Resolve using port number. [ 119 -> nntp, 443 -> https, ... ]
	 */

	ch = newurl + strcspn(newurl, ".:/@");
	if (*ch == '@' || (*ch == ':' && *newurl != '[')
		|| !strncasecmp(newurl, "ftp.", 4)) {
		/* Contains user/password/ftp-hostname */
		return PROTOCOL_FTP;

#ifdef CONFIG_IPV6
	} else if (*newurl == '[' && *ch == ':') {
		/* Candidate for IPv6 address */
		unsigned char *bracket2, *colon2;

		ch++;
		bracket2 = strchr(ch, ']');
		colon2 = strchr(ch, ':');
		if (bracket2 && colon2 && bracket2 > colon2)
			return PROTOCOL_HTTP;
#endif

	} else if (*newurl != '.' && *ch == '.') {
		/* Contains domain name? */
		unsigned char *host_end, *domain;
		unsigned char *ipscan;

		/* Process the hostname */
		for (domain = ch + 1;
			*(host_end = domain + strcspn(domain, ".:/?")) == '.';
			domain = host_end + 1);

		/* It's IP? */
		for (ipscan = ch; isdigit(*ipscan) || *ipscan == '.';
			ipscan++);

		if (!*ipscan || *ipscan == ':' || *ipscan == '/')
			return PROTOCOL_HTTP;

		/* It's two-letter or known TLD? */
		if (host_end - domain == 2
		    || end_with_known_tld(domain, host_end - domain) >= 0)
			return PROTOCOL_HTTP;
	}

	/* We default to file:// even though we already tested if the file
	 * existed since it will give a "No such file or directory" error.
	 * which might better hint the user that there was problem figuring out
	 * the URI. */
	return PROTOCOL_FILE;
}


#define MAX_TRANSLATION_ATTEMPTS	32

/* Returns an URI string that can be used internally. Adding protocol prefix,
 * missing slashes etc. */
static unsigned char *
translate_url(unsigned char *url, unsigned char *cwd)
{
	unsigned char *newurl;
	struct uri uri;
	enum uri_errno uri_errno, prev_errno = URI_ERRNO_EMPTY;