2 * Copyright (C) 1996-2014 The Squid Software Foundation and contributors
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.
11 #include "http/one/RequestParser.h"
12 #include "http/ProtocolVersion.h"
13 #include "mime_header.h"
14 #include "profiler/Profiler.h"
15 #include "SquidConfig.h"
17 Http::One::RequestParser::RequestParser() :
19 request_parse_status(Http::scNone
)
21 req
.start
= req
.end
= -1;
22 req
.m_start
= req
.m_end
= -1;
23 req
.u_start
= req
.u_end
= -1;
24 req
.v_start
= req
.v_end
= -1;
28 * Attempt to parse the first line of a new request message.
30 * Governed by RFC 7230 section 3.5
32 * In the interest of robustness, a server that is expecting to receive
33 * and parse a request-line SHOULD ignore at least one empty line (CRLF)
34 * received prior to the request-line.
37 * Parsing state is stored between calls to avoid repeating buffer scans.
38 * If garbage is found the parsing offset is incremented.
41 Http::One::RequestParser::skipGarbageLines()
43 if (Config
.onoff
.relaxed_header_parser
) {
44 if (Config
.onoff
.relaxed_header_parser
< 0 && (buf_
[0] == '\r' || buf_
[0] == '\n'))
45 debugs(74, DBG_IMPORTANT
, "WARNING: Invalid HTTP Request: " <<
46 "CRLF bytes received ahead of request-line. " <<
47 "Ignored due to relaxed_header_parser.");
48 // Be tolerant of prefix empty lines
49 // ie any series of either \n or \r\n with no other characters and no repeated \r
50 while (!buf_
.isEmpty() && (buf_
[0] == '\n' || (buf_
[0] == '\r' && buf_
[1] == '\n'))) {
55 /* XXX: this is a Squid-specific tolerance
56 * it appears never to have been relevant outside out unit-tests
57 * because the ConnStateData parser loop starts with consumeWhitespace()
58 * which absorbs any SP HTAB VTAB CR LF characters.
59 * But unit-tests called the HttpParser method directly without that pruning.
61 #if USE_HTTP_VIOLATIONS
62 if (Config
.onoff
.relaxed_header_parser
) {
63 if (Config
.onoff
.relaxed_header_parser
< 0 && buf_
[0] == ' ')
64 debugs(74, DBG_IMPORTANT
, "WARNING: Invalid HTTP Request: " <<
65 "Whitespace bytes received ahead of method. " <<
66 "Ignored due to relaxed_header_parser.");
67 // Be tolerant of prefix spaces (other bytes are valid method values)
68 while (!buf_
.isEmpty() && buf_
[0] == ' ') {
76 * Attempt to parse the first line of a new request message.
79 * RFC 1945 section 5.1
80 * RFC 7230 section 3.1 and 3.5
82 * Parsing state is stored between calls. However the current implementation
83 * begins parsing from scratch on every call.
84 * The return value tells you whether the parsing state fields are valid or not.
86 * \retval -1 an error occurred. request_parse_status indicates HTTP status result.
87 * \retval 1 successful parse. member fields contain the request-line items
88 * \retval 0 more data is needed to complete the parse
91 Http::One::RequestParser::parseRequestFirstLine()
93 int second_word
= -1; // track the suspected URI start
94 int first_whitespace
= -1, last_whitespace
= -1; // track the first and last SP byte
95 int line_end
= -1; // tracks the last byte BEFORE terminal \r\n or \n sequence
97 debugs(74, 5, "parsing possible request: buf.length=" << buf_
.length());
98 debugs(74, DBG_DATA
, buf_
);
100 // Single-pass parse: (provided we have the whole line anyways)
104 for (SBuf::size_type i
= 0; i
< buf_
.length(); ++i
) {
105 // track first and last whitespace (SP only)
106 if (buf_
[i
] == ' ') {
108 if (first_whitespace
< req
.start
)
109 first_whitespace
= i
;
112 // track next non-SP/non-HT byte after first_whitespace
113 if (second_word
< first_whitespace
&& buf_
[i
] != ' ' && buf_
[i
] != '\t') {
117 // locate line terminator
118 if (buf_
[i
] == '\n') {
123 if (i
< buf_
.length() - 1 && buf_
[i
] == '\r') {
124 if (Config
.onoff
.relaxed_header_parser
) {
125 if (Config
.onoff
.relaxed_header_parser
< 0 && buf_
[i
+ 1] == '\r')
126 debugs(74, DBG_IMPORTANT
, "WARNING: Invalid HTTP Request: " <<
127 "Series of carriage-return bytes received prior to line terminator. " <<
128 "Ignored due to relaxed_header_parser.");
130 // Be tolerant of invalid multiple \r prior to terminal \n
131 if (buf_
[i
+ 1] == '\n' || buf_
[i
+ 1] == '\r')
133 while (i
< buf_
.length() - 1 && buf_
[i
+ 1] == '\r')
136 if (buf_
[i
+ 1] == '\n') {
141 if (buf_
[i
+ 1] == '\n') {
148 // RFC 7230 section 3.1.1 does not prohibit embeded CR like RFC 2616 used to.
149 // However it does explicitly state an exact syntax which omits un-encoded CR
150 // and defines 400 (Bad Request) as the required action when
151 // handed an invalid request-line.
152 request_parse_status
= Http::scBadRequest
;
158 // DoS protection against long first-line
159 if ((size_t)buf_
.length() >= Config
.maxRequestHeaderSize
) {
160 debugs(33, 5, "Too large request-line");
161 // RFC 7230 section 3.1.1 mandatory 414 response if URL longer than acceptible.
162 request_parse_status
= Http::scUriTooLong
;
166 debugs(74, 5, "Parser: retval 0: from " << req
.start
<<
167 "->" << req
.end
<< ": needs more data to complete first line.");
171 // NP: we have now seen EOL, more-data (0) cannot occur.
172 // From here on any failure is -1, success is 1
176 // DoS protection against long first-line
177 if ((size_t)(req
.end
-req
.start
) >= Config
.maxRequestHeaderSize
) {
178 debugs(33, 5, "Too large request-line");
179 request_parse_status
= Http::scUriTooLong
;
183 // Process what we now know about the line structure into field offsets
184 // generating HTTP status for any aborts as we go.
186 // First non-whitespace = beginning of method
187 if (req
.start
> line_end
) {
188 request_parse_status
= Http::scBadRequest
;
191 req
.m_start
= req
.start
;
193 // First whitespace = end of method
194 if (first_whitespace
> line_end
|| first_whitespace
< req
.start
) {
195 request_parse_status
= Http::scBadRequest
; // no method
198 req
.m_end
= first_whitespace
- 1;
199 if (req
.m_end
< req
.m_start
) {
200 request_parse_status
= Http::scBadRequest
; // missing URI?
205 const SBuf tmp
= buf_
.substr(req
.m_start
, req
.m_end
- req
.m_start
+ 1);
206 method_
= HttpRequestMethod(tmp
);
208 // First non-whitespace after first SP = beginning of URL+Version
209 if (second_word
> line_end
|| second_word
< req
.start
) {
210 request_parse_status
= Http::scBadRequest
; // missing URI
213 req
.u_start
= second_word
;
215 // RFC 1945: SP and version following URI are optional, marking version 0.9
216 // we identify this by the last whitespace being earlier than URI start
217 if (last_whitespace
< second_word
&& last_whitespace
>= req
.start
) {
218 msgProtocol_
= Http::ProtocolVersion(0,9);
219 req
.u_end
= line_end
;
220 uri_
= buf_
.substr(req
.u_start
, req
.u_end
- req
.u_start
+ 1);
221 request_parse_status
= Http::scOkay
; // HTTP/0.9
224 // otherwise last whitespace is somewhere after end of URI.
225 req
.u_end
= last_whitespace
;
226 // crop any trailing whitespace in the area we think of as URI
227 for (; req
.u_end
>= req
.u_start
&& xisspace(buf_
[req
.u_end
]); --req
.u_end
);
229 if (req
.u_end
< req
.u_start
) {
230 request_parse_status
= Http::scBadRequest
; // missing URI
233 uri_
= buf_
.substr(req
.u_start
, req
.u_end
- req
.u_start
+ 1);
235 // Last whitespace SP = before start of protocol/version
236 if (last_whitespace
>= line_end
) {
237 request_parse_status
= Http::scBadRequest
; // missing version
240 req
.v_start
= last_whitespace
+ 1;
241 req
.v_end
= line_end
;
243 /* RFC 7230 section 2.6 : handle unsupported HTTP major versions cleanly. */
244 if ((req
.v_end
- req
.v_start
+1) < (int)Http1magic
.length() || !buf_
.substr(req
.v_start
, SBuf::npos
).startsWith(Http1magic
)) {
245 // non-HTTP/1 protocols not supported / implemented.
246 request_parse_status
= Http::scHttpVersionNotSupported
;
249 // NP: magic octets include the protocol name and major version DIGIT.
250 msgProtocol_
.protocol
= AnyP::PROTO_HTTP
;
251 msgProtocol_
.major
= 1;
253 int i
= req
.v_start
+ Http1magic
.length() -1;
255 // catch missing minor part
256 if (++i
> line_end
) {
257 request_parse_status
= Http::scHttpVersionNotSupported
;
260 /* next should be one or more digits */
261 if (!isdigit(buf_
[i
])) {
262 request_parse_status
= Http::scHttpVersionNotSupported
;
266 for (; i
<= line_end
&& (isdigit(buf_
[i
])) && min
< 65536; ++i
) {
268 min
= min
+ (buf_
[i
]) - '0';
270 // catch too-big values or trailing garbage
271 if (min
>= 65536 || i
< line_end
) {
272 request_parse_status
= Http::scHttpVersionNotSupported
;
275 msgProtocol_
.minor
= min
;
278 * Rightio - we have all the schtuff. Return true; we've got enough.
280 request_parse_status
= Http::scOkay
;
285 Http::One::RequestParser::parse(const SBuf
&aBuf
)
288 debugs(74, DBG_DATA
, "Parse buf={length=" << aBuf
.length() << ", data='" << aBuf
<< "'}");
290 // stage 1: locate the request-line
291 if (parsingStage_
== HTTP_PARSE_NONE
) {
294 // if we hit something before EOS treat it as a message
296 parsingStage_
= HTTP_PARSE_FIRST
;
301 // stage 2: parse the request-line
302 if (parsingStage_
== HTTP_PARSE_FIRST
) {
303 PROF_start(HttpParserParseReqLine
);
304 const int retcode
= parseRequestFirstLine();
306 // first-line (or a look-alike) found successfully.
308 buf_
.consume(firstLineSize()); // first line bytes including CRLF terminator are now done.
309 parsingStage_
= HTTP_PARSE_MIME
;
312 debugs(74, 5, "request-line: retval " << retcode
<< ": from " << req
.start
<< "->" << req
.end
<<
313 " line={" << aBuf
.length() << ", data='" << aBuf
<< "'}");
314 debugs(74, 5, "request-line: method " << req
.m_start
<< "->" << req
.m_end
<< " (" << method_
<< ")");
315 debugs(74, 5, "request-line: url " << req
.u_start
<< "->" << req
.u_end
<< " (" << uri_
<< ")");
316 debugs(74, 5, "request-line: proto " << req
.v_start
<< "->" << req
.v_end
<< " (" << msgProtocol_
<< ")");
317 debugs(74, 5, "Parser: bytes processed=" << (aBuf
.length()-buf_
.length()));
318 PROF_stop(HttpParserParseReqLine
);
320 // syntax errors already
322 parsingStage_
= HTTP_PARSE_DONE
;
327 // stage 3: locate the mime header block
328 if (parsingStage_
== HTTP_PARSE_MIME
) {
329 // HTTP/1.x request-line is valid and parsing completed.
330 if (msgProtocol_
.major
== 1) {
331 /* NOTE: HTTP/0.9 requests do not have a mime header block.
332 * So the rest of the code will need to deal with '0'-byte headers
333 * (ie, none, so don't try parsing em)
335 int64_t mimeHeaderBytes
= 0;
336 // XXX: c_str() reallocates. performance regression.
337 if ((mimeHeaderBytes
= headersEnd(buf_
.c_str(), buf_
.length())) == 0) {
338 if (buf_
.length()+firstLineSize() >= Config
.maxRequestHeaderSize
) {
339 debugs(33, 5, "Too large request");
340 request_parse_status
= Http::scRequestHeaderFieldsTooLarge
;
341 parsingStage_
= HTTP_PARSE_DONE
;
343 debugs(33, 5, "Incomplete request, waiting for end of headers");
346 mimeHeaderBlock_
= buf_
.consume(mimeHeaderBytes
);
347 debugs(74, 5, "mime header (0-" << mimeHeaderBytes
<< ") {" << mimeHeaderBlock_
<< "}");
350 debugs(33, 3, "Missing HTTP/1.x identifier");
352 // NP: we do not do any further stages here yet so go straight to DONE
353 parsingStage_
= HTTP_PARSE_DONE
;
355 // Squid could handle these headers, but admin does not want to
356 if (messageHeaderSize() >= Config
.maxRequestHeaderSize
) {
357 debugs(33, 5, "Too large request");
358 request_parse_status
= Http::scRequestHeaderFieldsTooLarge
;
363 return !needsMoreData();