| /* |
| * UPnP XML helper routines |
| * Copyright (c) 2000-2003 Intel Corporation |
| * Copyright (c) 2006-2007 Sony Corporation |
| * Copyright (c) 2008-2009 Atheros Communications |
| * Copyright (c) 2009, Jouni Malinen <j@w1.fi> |
| * |
| * See wps_upnp.c for more details on licensing and code history. |
| */ |
| |
| #include "includes.h" |
| |
| #include "common.h" |
| #include "base64.h" |
| #include "http.h" |
| #include "upnp_xml.h" |
| |
| |
| /* |
| * XML parsing and formatting |
| * |
| * XML is a markup language based on unicode; usually (and in our case, |
| * always!) based on utf-8. utf-8 uses a variable number of bytes per |
| * character. utf-8 has the advantage that all non-ASCII unicode characters are |
| * represented by sequences of non-ascii (high bit set) bytes, whereas ASCII |
| * characters are single ascii bytes, thus we can use typical text processing. |
| * |
| * (One other interesting thing about utf-8 is that it is possible to look at |
| * any random byte and determine if it is the first byte of a character as |
| * versus a continuation byte). |
| * |
| * The base syntax of XML uses a few ASCII punctionation characters; any |
| * characters that would appear in the payload data are rewritten using |
| * sequences, e.g., & for ampersand(&) and < for left angle bracket (<). |
| * Five such escapes total (more can be defined but that does not apply to our |
| * case). Thus we can safely parse for angle brackets etc. |
| * |
| * XML describes tree structures of tagged data, with each element beginning |
| * with an opening tag <label> and ending with a closing tag </label> with |
| * matching label. (There is also a self-closing tag <label/> which is supposed |
| * to be equivalent to <label></label>, i.e., no payload, but we are unlikely |
| * to see it for our purpose). |
| * |
| * Actually the opening tags are a little more complicated because they can |
| * contain "attributes" after the label (delimited by ascii space or tab chars) |
| * of the form attribute_label="value" or attribute_label='value'; as it turns |
| * out we do not have to read any of these attributes, just ignore them. |
| * |
| * Labels are any sequence of chars other than space, tab, right angle bracket |
| * (and ?), but may have an inner structure of <namespace><colon><plain_label>. |
| * As it turns out, we can ignore the namespaces, in fact we can ignore the |
| * entire tree hierarchy, because the plain labels we are looking for will be |
| * unique (not in general, but for this application). We do however have to be |
| * careful to skip over the namespaces. |
| * |
| * In generating XML we have to be more careful, but that is easy because |
| * everything we do is pretty canned. The only real care to take is to escape |
| * any special chars in our payload. |
| */ |
| |
| /** |
| * xml_next_tag - Advance to next tag |
| * @in: Input |
| * @out: OUT: start of tag just after '<' |
| * @out_tagname: OUT: start of name of tag, skipping namespace |
| * @end: OUT: one after tag |
| * Returns: 0 on success, 1 on failure |
| * |
| * A tag has form: |
| * <left angle bracket><...><right angle bracket> |
| * Within the angle brackets, there is an optional leading forward slash (which |
| * makes the tag an ending tag), then an optional leading label (followed by |
| * colon) and then the tag name itself. |
| * |
| * Note that angle brackets present in the original data must have been encoded |
| * as < and > so they will not trouble us. |
| */ |
| int xml_next_tag(const char *in, const char **out, |
| const char **out_tagname, const char **end) |
| { |
| while (*in && *in != '<') |
| in++; |
| if (*in != '<') |
| return 1; |
| *out = ++in; |
| if (*in == '/') |
| in++; |
| *out_tagname = in; /* maybe */ |
| while (isalnum(*in) || *in == '-') |
| in++; |
| if (*in == ':') |
| *out_tagname = ++in; |
| while (*in && *in != '>') |
| in++; |
| if (*in != '>') |
| return 1; |
| *end = ++in; |
| return 0; |
| } |
| |
| |
| /* xml_data_encode -- format data for xml file, escaping special characters. |
| * |
| * Note that we assume we are using utf8 both as input and as output! |
| * In utf8, characters may be classed as follows: |
| * 0xxxxxxx(2) -- 1 byte ascii char |
| * 11xxxxxx(2) -- 1st byte of multi-byte char w/ unicode value >= 0x80 |
| * 110xxxxx(2) -- 1st byte of 2 byte sequence (5 payload bits here) |
| * 1110xxxx(2) -- 1st byte of 3 byte sequence (4 payload bits here) |
| * 11110xxx(2) -- 1st byte of 4 byte sequence (3 payload bits here) |
| * 10xxxxxx(2) -- extension byte (6 payload bits per byte) |
| * Some values implied by the above are however illegal because they |
| * do not represent unicode chars or are not the shortest encoding. |
| * Actually, we can almost entirely ignore the above and just do |
| * text processing same as for ascii text. |
| * |
| * XML is written with arbitrary unicode characters, except that five |
| * characters have special meaning and so must be escaped where they |
| * appear in payload data... which we do here. |
| */ |
| void xml_data_encode(struct wpabuf *buf, const char *data, int len) |
| { |
| int i; |
| for (i = 0; i < len; i++) { |
| u8 c = ((u8 *) data)[i]; |
| if (c == '<') { |
| wpabuf_put_str(buf, "<"); |
| continue; |
| } |
| if (c == '>') { |
| wpabuf_put_str(buf, ">"); |
| continue; |
| } |
| if (c == '&') { |
| wpabuf_put_str(buf, "&"); |
| continue; |
| } |
| if (c == '\'') { |
| wpabuf_put_str(buf, "'"); |
| continue; |
| } |
| if (c == '"') { |
| wpabuf_put_str(buf, """); |
| continue; |
| } |
| /* |
| * We could try to represent control characters using the |
| * sequence: &#x; where x is replaced by a hex numeral, but not |
| * clear why we would do this. |
| */ |
| wpabuf_put_u8(buf, c); |
| } |
| } |
| |
| |
| /* xml_add_tagged_data -- format tagged data as a new xml line. |
| * |
| * tag must not have any special chars. |
| * data may have special chars, which are escaped. |
| */ |
| void xml_add_tagged_data(struct wpabuf *buf, const char *tag, const char *data) |
| { |
| wpabuf_printf(buf, "<%s>", tag); |
| xml_data_encode(buf, data, os_strlen(data)); |
| wpabuf_printf(buf, "</%s>\n", tag); |
| } |
| |
| |
| /* A POST body looks something like (per upnp spec): |
| * <?xml version="1.0"?> |
| * <s:Envelope |
| * xmlns:s="http://schemas.xmlsoap.org/soap/envelope/" |
| * s:encodingStyle="http://schemas.xmlsoap.org/soap/encoding/"> |
| * <s:Body> |
| * <u:actionName xmlns:u="urn:schemas-upnp-org:service:serviceType:v"> |
| * <argumentName>in arg value</argumentName> |
| * other in args and their values go here, if any |
| * </u:actionName> |
| * </s:Body> |
| * </s:Envelope> |
| * |
| * where : |
| * s: might be some other namespace name followed by colon |
| * u: might be some other namespace name followed by colon |
| * actionName will be replaced according to action requested |
| * schema following actionName will be WFA scheme instead |
| * argumentName will be actual argument name |
| * (in arg value) will be actual argument value |
| */ |
| char * xml_get_first_item(const char *doc, const char *item) |
| { |
| const char *match = item; |
| int match_len = os_strlen(item); |
| const char *tag, *tagname, *end; |
| char *value; |
| |
| /* |
| * This is crude: ignore any possible tag name conflicts and go right |
| * to the first tag of this name. This should be ok for the limited |
| * domain of UPnP messages. |
| */ |
| for (;;) { |
| if (xml_next_tag(doc, &tag, &tagname, &end)) |
| return NULL; |
| doc = end; |
| if (!os_strncasecmp(tagname, match, match_len) && |
| *tag != '/' && |
| (tagname[match_len] == '>' || |
| !isgraph(tagname[match_len]))) { |
| break; |
| } |
| } |
| end = doc; |
| while (*end && *end != '<') |
| end++; |
| value = os_zalloc(1 + (end - doc)); |
| if (value == NULL) |
| return NULL; |
| os_memcpy(value, doc, end - doc); |
| return value; |
| } |
| |
| |
| struct wpabuf * xml_get_base64_item(const char *data, const char *name, |
| enum http_reply_code *ret) |
| { |
| char *msg; |
| struct wpabuf *buf; |
| unsigned char *decoded; |
| size_t len; |
| |
| msg = xml_get_first_item(data, name); |
| if (msg == NULL) { |
| *ret = UPNP_ARG_VALUE_INVALID; |
| return NULL; |
| } |
| |
| decoded = base64_decode(msg, os_strlen(msg), &len); |
| os_free(msg); |
| if (decoded == NULL) { |
| *ret = UPNP_OUT_OF_MEMORY; |
| return NULL; |
| } |
| |
| buf = wpabuf_alloc_ext_data(decoded, len); |
| if (buf == NULL) { |
| os_free(decoded); |
| *ret = UPNP_OUT_OF_MEMORY; |
| return NULL; |
| } |
| return buf; |
| } |