2 * Copyright (c) 2009 Eric Wong (all bugs are Eric's fault)
3 * Copyright (c) 2005 Zed A. Shaw
4 * You can redistribute it and/or modify it under the same terms as Ruby 1.8 or
5 * the GPLv2+ (GPLv3+ preferred)
11 #include <sys/types.h>
12 #include "common_field_optimization.h"
13 #include "global_variables.h"
16 void init_unicorn_httpdate(void);
18 #define UH_FL_CHUNKED 0x1
19 #define UH_FL_HASBODY 0x2
20 #define UH_FL_INBODY 0x4
21 #define UH_FL_HASTRAILER 0x8
22 #define UH_FL_INTRAILER 0x10
23 #define UH_FL_INCHUNK 0x20
24 #define UH_FL_REQEOF 0x40
25 #define UH_FL_KAVERSION 0x80
26 #define UH_FL_HASHEADER 0x100
27 #define UH_FL_TO_CLEAR 0x200
29 /* all of these flags need to be set for keepalive to be supported */
30 #define UH_FL_KEEPALIVE (UH_FL_KAVERSION | UH_FL_REQEOF | UH_FL_HASHEADER)
32 static unsigned long keepalive_requests = 100; /* same as nginx */
35 * Returns the maximum number of keepalive requests a client may make
36 * before the parser refuses to continue.
38 static VALUE ka_req(VALUE self)
40 return ULONG2NUM(keepalive_requests);
44 * Sets the maximum number of keepalive requests a client may make.
45 * A special value of +nil+ causes this to be the maximum value
46 * possible (this is architecture-dependent).
48 static VALUE set_ka_req(VALUE self, VALUE val)
50 keepalive_requests = NIL_P(val) ? ULONG_MAX : NUM2ULONG(val);
55 static size_t MAX_HEADER_LEN = 1024 * (80 + 32); /* same as Mongrel */
57 /* this is only intended for use with Rainbows! */
58 static VALUE set_maxhdrlen(VALUE self, VALUE len)
60 return SIZET2NUM(MAX_HEADER_LEN = NUM2SIZET(len));
63 /* keep this small for Rainbows! since every client has one */
65 int cs; /* Ragel internal state */
67 unsigned long nr_requests;
70 union { /* these 2 fields don't nest */
75 size_t field_len; /* only used during header processing */
76 size_t dest_offset; /* only used during body processing */
80 VALUE cont; /* Qfalse: unset, Qnil: ignored header, T_STRING: append */
87 static ID id_clear, id_set_backtrace, id_response_start_sent;
89 static void finalize_header(struct http_parser *hp);
91 static void parser_raise(VALUE klass, const char *msg)
93 VALUE exc = rb_exc_new2(klass, msg);
94 VALUE bt = rb_ary_new();
96 rb_funcall(exc, id_set_backtrace, 1, bt);
100 #define REMAINING (unsigned long)(pe - p)
101 #define LEN(AT, FPC) (FPC - buffer - hp->AT)
102 #define MARK(M,FPC) (hp->M = (FPC) - buffer)
103 #define PTR_TO(F) (buffer + hp->F)
104 #define STR_NEW(M,FPC) rb_str_new(PTR_TO(M), LEN(M, FPC))
105 #define STRIPPED_STR_NEW(M,FPC) stripped_str_new(PTR_TO(M), LEN(M, FPC))
107 #define HP_FL_TEST(hp,fl) ((hp)->flags & (UH_FL_##fl))
108 #define HP_FL_SET(hp,fl) ((hp)->flags |= (UH_FL_##fl))
109 #define HP_FL_UNSET(hp,fl) ((hp)->flags &= ~(UH_FL_##fl))
110 #define HP_FL_ALL(hp,fl) (HP_FL_TEST(hp, fl) == (UH_FL_##fl))
112 static int is_lws(char c)
114 return (c == ' ' || c == '\t');
117 static VALUE stripped_str_new(const char *str, long len)
121 for (end = len - 1; end >= 0 && is_lws(str[end]); end--);
123 return rb_str_new(str, end + 1);
127 * handles values of the "Connection:" header, keepalive is implied
128 * for HTTP/1.1 but needs to be explicitly enabled with HTTP/1.0
129 * Additionally, we require GET/HEAD requests to support keepalive.
131 static void hp_keepalive_connection(struct http_parser *hp, VALUE val)
133 if (STR_CSTR_CASE_EQ(val, "keep-alive")) {
134 /* basically have HTTP/1.0 masquerade as HTTP/1.1+ */
135 HP_FL_SET(hp, KAVERSION);
136 } else if (STR_CSTR_CASE_EQ(val, "close")) {
138 * it doesn't matter what HTTP version or request method we have,
139 * if a client says "Connection: close", we disable keepalive
141 HP_FL_UNSET(hp, KAVERSION);
144 * client could've sent anything, ignore it for now. Maybe
145 * "HP_FL_UNSET(hp, KAVERSION);" just in case?
146 * Raising an exception might be too mean...
152 request_method(struct http_parser *hp, const char *ptr, size_t len)
154 VALUE v = rb_str_new(ptr, len);
156 rb_hash_aset(hp->env, g_request_method, v);
160 http_version(struct http_parser *hp, const char *ptr, size_t len)
164 HP_FL_SET(hp, HASHEADER);
166 if (CONST_MEM_EQ("HTTP/1.1", ptr, len)) {
167 /* HTTP/1.1 implies keepalive unless "Connection: close" is set */
168 HP_FL_SET(hp, KAVERSION);
170 } else if (CONST_MEM_EQ("HTTP/1.0", ptr, len)) {
173 v = rb_str_new(ptr, len);
175 rb_hash_aset(hp->env, g_server_protocol, v);
176 rb_hash_aset(hp->env, g_http_version, v);
179 static inline void hp_invalid_if_trailer(struct http_parser *hp)
181 if (HP_FL_TEST(hp, INTRAILER))
182 parser_raise(eHttpParserError, "invalid Trailer");
185 static void write_cont_value(struct http_parser *hp,
186 char *buffer, const char *p)
190 long len = LEN(mark, p);
193 if (hp->cont == Qfalse)
194 parser_raise(eHttpParserError, "invalid continuation line");
196 return; /* we're ignoring this header (probably Host:) */
198 assert(TYPE(hp->cont) == T_STRING && "continuation line is not a string");
199 assert(hp->mark > 0 && "impossible continuation line offset");
204 cont_len = RSTRING_LEN(hp->cont);
211 /* normalize tab to space */
213 assert((' ' == *vptr || '\t' == *vptr) && "invalid leading white space");
217 for (end = len - 1; end >= 0 && is_lws(vptr[end]); end--);
218 rb_str_buf_cat(hp->cont, vptr, end + 1);
221 static void write_value(struct http_parser *hp,
222 const char *buffer, const char *p)
224 VALUE f = find_common_field(PTR_TO(start.field), hp->s.field_len);
228 VALIDATE_MAX_LENGTH(LEN(mark, p), FIELD_VALUE);
229 v = LEN(mark, p) == 0 ? rb_str_buf_new(128) : STRIPPED_STR_NEW(mark, p);
231 const char *field = PTR_TO(start.field);
232 size_t flen = hp->s.field_len;
234 VALIDATE_MAX_LENGTH(flen, FIELD_NAME);
237 * ignore "Version" headers since they conflict with the HTTP_VERSION
240 if (CONST_MEM_EQ("VERSION", field, flen)) {
244 f = uncommon_field(field, flen);
245 } else if (f == g_http_connection) {
246 hp_keepalive_connection(hp, v);
247 } else if (f == g_content_length) {
248 hp->len.content = parse_length(RSTRING_PTR(v), RSTRING_LEN(v));
249 if (hp->len.content < 0)
250 parser_raise(eHttpParserError, "invalid Content-Length");
251 if (hp->len.content != 0)
252 HP_FL_SET(hp, HASBODY);
253 hp_invalid_if_trailer(hp);
254 } else if (f == g_http_transfer_encoding) {
255 if (STR_CSTR_CASE_EQ(v, "chunked")) {
256 HP_FL_SET(hp, CHUNKED);
257 HP_FL_SET(hp, HASBODY);
259 hp_invalid_if_trailer(hp);
260 } else if (f == g_http_trailer) {
261 HP_FL_SET(hp, HASTRAILER);
262 hp_invalid_if_trailer(hp);
264 assert(TYPE(f) == T_STRING && "memoized object is not a string");
268 e = rb_hash_aref(hp->env, f);
270 hp->cont = rb_hash_aset(hp->env, f, v);
271 } else if (f == g_http_host) {
273 * ignored, absolute URLs in REQUEST_URI take precedence over
274 * the Host: header (ref: rfc 2616, section 5.2.1)
278 rb_str_buf_cat(e, ",", 1);
279 hp->cont = rb_str_buf_append(e, v);
288 action mark {MARK(mark, fpc); }
290 action start_field { MARK(start.field, fpc); }
291 action snake_upcase_field { snake_upcase_char(deconst(fpc)); }
292 action downcase_char { downcase_char(deconst(fpc)); }
293 action write_field { hp->s.field_len = LEN(start.field, fpc); }
294 action start_value { MARK(mark, fpc); }
295 action write_value { write_value(hp, buffer, fpc); }
296 action write_cont_value { write_cont_value(hp, buffer, fpc); }
297 action request_method { request_method(hp, PTR_TO(mark), LEN(mark, fpc)); }
299 rb_hash_aset(hp->env, g_rack_url_scheme, STR_NEW(mark, fpc));
301 action host { rb_hash_aset(hp->env, g_http_host, STR_NEW(mark, fpc)); }
305 VALIDATE_MAX_URI_LENGTH(LEN(mark, fpc), REQUEST_URI);
306 str = rb_hash_aset(hp->env, g_request_uri, STR_NEW(mark, fpc));
308 * "OPTIONS * HTTP/1.1\r\n" is a valid request, but we can't have '*'
309 * in REQUEST_PATH or PATH_INFO or else Rack::Lint will complain
311 if (STR_CSTR_EQ(str, "*")) {
312 str = rb_str_new(NULL, 0);
313 rb_hash_aset(hp->env, g_path_info, str);
314 rb_hash_aset(hp->env, g_request_path, str);
318 VALIDATE_MAX_URI_LENGTH(LEN(mark, fpc), FRAGMENT);
319 rb_hash_aset(hp->env, g_fragment, STR_NEW(mark, fpc));
321 action start_query {MARK(start.query, fpc); }
322 action query_string {
323 VALIDATE_MAX_URI_LENGTH(LEN(start.query, fpc), QUERY_STRING);
324 rb_hash_aset(hp->env, g_query_string, STR_NEW(start.query, fpc));
326 action http_version { http_version(hp, PTR_TO(mark), LEN(mark, fpc)); }
327 action request_path {
330 VALIDATE_MAX_URI_LENGTH(LEN(mark, fpc), REQUEST_PATH);
331 val = rb_hash_aset(hp->env, g_request_path, STR_NEW(mark, fpc));
333 /* rack says PATH_INFO must start with "/" or be empty */
334 if (!STR_CSTR_EQ(val, "*"))
335 rb_hash_aset(hp->env, g_path_info, val);
337 action add_to_chunk_size {
338 hp->len.chunk = step_incr(hp->len.chunk, fc, 16);
339 if (hp->len.chunk < 0)
340 parser_raise(eHttpParserError, "invalid chunk size");
345 cs = http_parser_first_final;
346 if (HP_FL_TEST(hp, HASBODY)) {
347 HP_FL_SET(hp, INBODY);
348 if (HP_FL_TEST(hp, CHUNKED))
349 cs = http_parser_en_ChunkedBody;
351 HP_FL_SET(hp, REQEOF);
352 assert(!HP_FL_TEST(hp, CHUNKED) && "chunked encoding without body!");
355 * go back to Ruby so we can call the Rack application, we'll reenter
356 * the parser iff the body needs to be processed.
361 action end_trailers {
362 cs = http_parser_first_final;
366 action end_chunked_body {
367 HP_FL_SET(hp, INTRAILER);
368 cs = http_parser_en_Trailers;
370 assert(p <= pe && "buffer overflow after chunked body");
374 action skip_chunk_data {
375 skip_chunk_data_hack: {
376 size_t nr = MIN((size_t)hp->len.chunk, REMAINING);
377 memcpy(RSTRING_PTR(hp->cont) + hp->s.dest_offset, fpc, nr);
378 hp->s.dest_offset += nr;
381 assert(hp->len.chunk >= 0 && "negative chunk length");
382 if ((size_t)hp->len.chunk > REMAINING) {
383 HP_FL_SET(hp, INCHUNK);
391 include unicorn_http_common "unicorn_http_common.rl";
397 static void http_parser_init(struct http_parser *hp)
406 hp->cont = Qfalse; /* zero on MRI, should be optimized away by above */
413 http_parser_execute(struct http_parser *hp, char *buffer, size_t len)
417 size_t off = hp->offset;
419 if (cs == http_parser_first_final)
422 assert(off <= len && "offset past end of buffer");
427 assert((void *)(pe - p) == (void *)(len - off) &&
428 "pointers aren't same distance");
430 if (HP_FL_TEST(hp, INCHUNK)) {
431 HP_FL_UNSET(hp, INCHUNK);
432 goto skip_chunk_data_hack;
435 post_exec: /* "_out:" also goes here */
436 if (hp->cs != http_parser_error)
438 hp->offset = p - buffer;
440 assert(p <= pe && "buffer overflow after parsing execute");
441 assert(hp->offset <= len && "offset longer than length");
444 static struct http_parser *data_get(VALUE self)
446 struct http_parser *hp;
448 Data_Get_Struct(self, struct http_parser, hp);
449 assert(hp && "failed to extract http_parser struct");
454 * set rack.url_scheme to "https" or "http", no others are allowed by Rack
455 * this resembles the Rack::Request#scheme method as of rack commit
456 * 35bb5ba6746b5d346de9202c004cc926039650c7
458 static void set_url_scheme(VALUE env, VALUE *server_port)
460 VALUE scheme = rb_hash_aref(env, g_rack_url_scheme);
464 * would anybody be horribly opposed to removing the X-Forwarded-SSL
465 * and X-Forwarded-Proto handling from this parser? We've had it
466 * forever and nobody has said anything against it, either.
467 * Anyways, please send comments to our public mailing list:
468 * unicorn-public@bogomips.org (no HTML mail, no subscription necessary)
470 scheme = rb_hash_aref(env, g_http_x_forwarded_ssl);
471 if (!NIL_P(scheme) && STR_CSTR_EQ(scheme, "on")) {
472 *server_port = g_port_443;
475 scheme = rb_hash_aref(env, g_http_x_forwarded_proto);
479 long len = RSTRING_LEN(scheme);
480 if (len >= 5 && !memcmp(RSTRING_PTR(scheme), "https", 5)) {
483 *server_port = g_port_443;
489 rb_hash_aset(env, g_rack_url_scheme, scheme);
490 } else if (STR_CSTR_EQ(scheme, "https")) {
491 *server_port = g_port_443;
493 assert(*server_port == g_port_80 && "server_port not set");
498 * Parse and set the SERVER_NAME and SERVER_PORT variables
499 * Not supporting X-Forwarded-Host/X-Forwarded-Port in here since
500 * anybody who needs them is using an unsupported configuration and/or
501 * incompetent. Rack::Request will handle X-Forwarded-{Port,Host} just
504 static void set_server_vars(VALUE env, VALUE *server_port)
506 VALUE server_name = g_localhost;
507 VALUE host = rb_hash_aref(env, g_http_host);
510 char *host_ptr = RSTRING_PTR(host);
511 long host_len = RSTRING_LEN(host);
514 if (*host_ptr == '[') { /* ipv6 address format */
515 char *rbracket = memchr(host_ptr + 1, ']', host_len - 1);
518 colon = (rbracket[1] == ':') ? rbracket + 1 : NULL;
520 colon = memchr(host_ptr + 1, ':', host_len - 1);
522 colon = memchr(host_ptr, ':', host_len);
526 long port_start = colon - host_ptr + 1;
528 server_name = rb_str_substr(host, 0, colon - host_ptr);
529 if ((host_len - port_start) > 0)
530 *server_port = rb_str_substr(host, port_start, host_len);
535 rb_hash_aset(env, g_server_name, server_name);
536 rb_hash_aset(env, g_server_port, *server_port);
539 static void finalize_header(struct http_parser *hp)
541 VALUE server_port = g_port_80;
543 set_url_scheme(hp->env, &server_port);
544 set_server_vars(hp->env, &server_port);
546 if (!HP_FL_TEST(hp, HASHEADER))
547 rb_hash_aset(hp->env, g_server_protocol, g_http_09);
549 /* rack requires QUERY_STRING */
550 if (NIL_P(rb_hash_aref(hp->env, g_query_string)))
551 rb_hash_aset(hp->env, g_query_string, rb_str_new(NULL, 0));
554 static void hp_mark(void *ptr)
556 struct http_parser *hp = ptr;
560 rb_gc_mark(hp->cont);
563 static VALUE HttpParser_alloc(VALUE klass)
565 struct http_parser *hp;
566 return Data_Make_Struct(klass, struct http_parser, hp_mark, -1, hp);
572 * parser.new => parser
574 * Creates a new parser.
576 static VALUE HttpParser_init(VALUE self)
578 struct http_parser *hp = data_get(self);
580 http_parser_init(hp);
581 hp->buf = rb_str_new(NULL, 0);
582 hp->env = rb_hash_new();
583 hp->nr_requests = keepalive_requests;
590 * parser.clear => parser
592 * Resets the parser to it's initial state so that you can reuse it
593 * rather than making new ones.
595 static VALUE HttpParser_clear(VALUE self)
597 struct http_parser *hp = data_get(self);
599 http_parser_init(hp);
600 rb_funcall(hp->env, id_clear, 0);
601 rb_ivar_set(self, id_response_start_sent, Qfalse);
608 * parser.dechunk! => parser
610 * Resets the parser to a state suitable for dechunking response bodies
613 static VALUE HttpParser_dechunk_bang(VALUE self)
615 struct http_parser *hp = data_get(self);
617 http_parser_init(hp);
620 * we don't care about trailers in dechunk-only mode,
621 * but if we did we'd set UH_FL_HASTRAILER and clear hp->env
624 rb_funcall(hp->env, id_clear, 0);
625 hp->flags = UH_FL_HASTRAILER;
628 hp->flags |= UH_FL_HASBODY | UH_FL_INBODY | UH_FL_CHUNKED;
629 hp->cs = http_parser_en_ChunkedBody;
636 * parser.reset => nil
638 * Resets the parser to it's initial state so that you can reuse it
639 * rather than making new ones.
641 * This method is deprecated and to be removed in Unicorn 4.x
643 static VALUE HttpParser_reset(VALUE self)
648 rb_warn("Unicorn::HttpParser#reset is deprecated; "
649 "use Unicorn::HttpParser#clear instead");
651 HttpParser_clear(self);
655 static void advance_str(VALUE str, off_t nr)
657 long len = RSTRING_LEN(str);
664 assert(nr <= len && "trying to advance past end of buffer");
666 if (len > 0) /* unlikely, len is usually 0 */
667 memmove(RSTRING_PTR(str), RSTRING_PTR(str) + nr, len);
668 rb_str_set_len(str, len);
673 * parser.content_length => nil or Integer
675 * Returns the number of bytes left to run through HttpParser#filter_body.
676 * This will initially be the value of the "Content-Length" HTTP header
677 * after header parsing is complete and will decrease in value as
678 * HttpParser#filter_body is called for each chunk. This should return
679 * zero for requests with no body.
681 * This will return nil on "Transfer-Encoding: chunked" requests.
683 static VALUE HttpParser_content_length(VALUE self)
685 struct http_parser *hp = data_get(self);
687 return HP_FL_TEST(hp, CHUNKED) ? Qnil : OFFT2NUM(hp->len.content);
691 * Document-method: parse
693 * parser.parse => env or nil
695 * Takes a Hash and a String of data, parses the String of data filling
696 * in the Hash returning the Hash if parsing is finished, nil otherwise
697 * When returning the env Hash, it may modify data to point to where
698 * body processing should begin.
700 * Raises HttpParserError if there are parsing errors.
702 static VALUE HttpParser_parse(VALUE self)
704 struct http_parser *hp = data_get(self);
705 VALUE data = hp->buf;
707 if (HP_FL_TEST(hp, TO_CLEAR))
708 HttpParser_clear(self);
710 http_parser_execute(hp, RSTRING_PTR(data), RSTRING_LEN(data));
711 if (hp->offset > MAX_HEADER_LEN)
712 parser_raise(e413, "HTTP header is too large");
714 if (hp->cs == http_parser_first_final ||
715 hp->cs == http_parser_en_ChunkedBody) {
716 advance_str(data, hp->offset + 1);
718 if (HP_FL_TEST(hp, INTRAILER))
719 HP_FL_SET(hp, REQEOF);
724 if (hp->cs == http_parser_error)
725 parser_raise(eHttpParserError, "Invalid HTTP format, parsing fails.");
731 * Document-method: parse
733 * parser.add_parse(buffer) => env or nil
735 * adds the contents of +buffer+ to the internal buffer and attempts to
736 * continue parsing. Returns the +env+ Hash on success or nil if more
739 * Raises HttpParserError if there are parsing errors.
741 static VALUE HttpParser_add_parse(VALUE self, VALUE buffer)
743 struct http_parser *hp = data_get(self);
745 Check_Type(buffer, T_STRING);
746 rb_str_buf_append(hp->buf, buffer);
748 return HttpParser_parse(self);
752 * Document-method: trailers
754 * parser.trailers(req, data) => req or nil
756 * This is an alias for HttpParser#headers
760 * Document-method: headers
762 static VALUE HttpParser_headers(VALUE self, VALUE env, VALUE buf)
764 struct http_parser *hp = data_get(self);
769 return HttpParser_parse(self);
772 static int chunked_eof(struct http_parser *hp)
774 return ((hp->cs == http_parser_first_final) || HP_FL_TEST(hp, INTRAILER));
779 * parser.body_eof? => true or false
781 * Detects if we're done filtering the body or not. This can be used
782 * to detect when to stop calling HttpParser#filter_body.
784 static VALUE HttpParser_body_eof(VALUE self)
786 struct http_parser *hp = data_get(self);
788 if (HP_FL_TEST(hp, CHUNKED))
789 return chunked_eof(hp) ? Qtrue : Qfalse;
791 return hp->len.content == 0 ? Qtrue : Qfalse;
796 * parser.keepalive? => true or false
798 * This should be used to detect if a request can really handle
799 * keepalives and pipelining. Currently, the rules are:
801 * 1. MUST be a GET or HEAD request
802 * 2. MUST be HTTP/1.1 +or+ HTTP/1.0 with "Connection: keep-alive"
803 * 3. MUST NOT have "Connection: close" set
805 static VALUE HttpParser_keepalive(VALUE self)
807 struct http_parser *hp = data_get(self);
809 return HP_FL_ALL(hp, KEEPALIVE) ? Qtrue : Qfalse;
814 * parser.next? => true or false
816 * Exactly like HttpParser#keepalive?, except it will reset the internal
817 * parser state on next parse if it returns true. It will also respect
818 * the maximum *keepalive_requests* value and return false if that is
821 static VALUE HttpParser_next(VALUE self)
823 struct http_parser *hp = data_get(self);
825 if ((HP_FL_ALL(hp, KEEPALIVE)) && (hp->nr_requests-- != 0)) {
826 HP_FL_SET(hp, TO_CLEAR);
834 * parser.headers? => true or false
836 * This should be used to detect if a request has headers (and if
837 * the response will have headers as well). HTTP/0.9 requests
838 * should return false, all subsequent HTTP versions will return true
840 static VALUE HttpParser_has_headers(VALUE self)
842 struct http_parser *hp = data_get(self);
844 return HP_FL_TEST(hp, HASHEADER) ? Qtrue : Qfalse;
847 static VALUE HttpParser_buf(VALUE self)
849 return data_get(self)->buf;
852 static VALUE HttpParser_env(VALUE self)
854 return data_get(self)->env;
859 * parser.filter_body(dst, src) => nil/src
861 * Takes a String of +src+, will modify data if dechunking is done.
862 * Returns +nil+ if there is more data left to process. Returns
863 * +src+ if body processing is complete. When returning +src+,
864 * it may modify +src+ so the start of the string points to where
865 * the body ended so that trailer processing can begin.
867 * Raises HttpParserError if there are dechunking errors.
868 * Basically this is a glorified memcpy(3) that copies +src+
869 * into +buf+ while filtering it through the dechunker.
871 static VALUE HttpParser_filter_body(VALUE self, VALUE dst, VALUE src)
873 struct http_parser *hp = data_get(self);
877 srcptr = RSTRING_PTR(src);
878 srclen = RSTRING_LEN(src);
882 if (HP_FL_TEST(hp, CHUNKED)) {
883 if (!chunked_eof(hp)) {
885 rb_str_resize(dst, srclen); /* we can never copy more than srclen bytes */
887 hp->s.dest_offset = 0;
890 http_parser_execute(hp, srcptr, srclen);
891 if (hp->cs == http_parser_error)
892 parser_raise(eHttpParserError, "Invalid HTTP format, parsing fails.");
894 assert(hp->s.dest_offset <= hp->offset &&
895 "destination buffer overflow");
896 advance_str(src, hp->offset);
897 rb_str_set_len(dst, hp->s.dest_offset);
899 if (RSTRING_LEN(dst) == 0 && chunked_eof(hp)) {
900 assert(hp->len.chunk == 0 && "chunk at EOF but more to parse");
906 /* no need to enter the Ragel machine for unchunked transfers */
907 assert(hp->len.content >= 0 && "negative Content-Length");
908 if (hp->len.content > 0) {
909 long nr = MIN(srclen, hp->len.content);
912 rb_str_resize(dst, nr);
914 * using rb_str_replace() to avoid memcpy() doesn't help in
915 * most cases because a GC-aware programmer will pass an explicit
916 * buffer to env["rack.input"].read and reuse the buffer in a loop.
917 * This causes copy-on-write behavior to be triggered anyways
918 * when the +src+ buffer is modified (when reading off the socket).
921 memcpy(RSTRING_PTR(dst), srcptr, nr);
922 hp->len.content -= nr;
923 if (hp->len.content == 0) {
924 HP_FL_SET(hp, REQEOF);
925 hp->cs = http_parser_first_final;
927 advance_str(src, nr);
931 hp->offset = 0; /* for trailer parsing */
935 #define SET_GLOBAL(var,str) do { \
936 var = find_common_field(str, sizeof(str) - 1); \
937 assert(!NIL_P(var) && "missed global field"); \
940 void Init_unicorn_http(void)
942 VALUE mUnicorn, cHttpParser;
944 mUnicorn = rb_const_get(rb_cObject, rb_intern("Unicorn"));
945 cHttpParser = rb_define_class_under(mUnicorn, "HttpParser", rb_cObject);
947 rb_define_class_under(mUnicorn, "HttpParserError", rb_eIOError);
948 e413 = rb_define_class_under(mUnicorn, "RequestEntityTooLargeError",
950 e414 = rb_define_class_under(mUnicorn, "RequestURITooLongError",
954 rb_define_alloc_func(cHttpParser, HttpParser_alloc);
955 rb_define_method(cHttpParser, "initialize", HttpParser_init, 0);
956 rb_define_method(cHttpParser, "clear", HttpParser_clear, 0);
957 rb_define_method(cHttpParser, "reset", HttpParser_reset, 0);
958 rb_define_method(cHttpParser, "dechunk!", HttpParser_dechunk_bang, 0);
959 rb_define_method(cHttpParser, "parse", HttpParser_parse, 0);
960 rb_define_method(cHttpParser, "add_parse", HttpParser_add_parse, 1);
961 rb_define_method(cHttpParser, "headers", HttpParser_headers, 2);
962 rb_define_method(cHttpParser, "trailers", HttpParser_headers, 2);
963 rb_define_method(cHttpParser, "filter_body", HttpParser_filter_body, 2);
964 rb_define_method(cHttpParser, "content_length", HttpParser_content_length, 0);
965 rb_define_method(cHttpParser, "body_eof?", HttpParser_body_eof, 0);
966 rb_define_method(cHttpParser, "keepalive?", HttpParser_keepalive, 0);
967 rb_define_method(cHttpParser, "headers?", HttpParser_has_headers, 0);
968 rb_define_method(cHttpParser, "next?", HttpParser_next, 0);
969 rb_define_method(cHttpParser, "buf", HttpParser_buf, 0);
970 rb_define_method(cHttpParser, "env", HttpParser_env, 0);
973 * The maximum size a single chunk when using chunked transfer encoding.
974 * This is only a theoretical maximum used to detect errors in clients,
975 * it is highly unlikely to encounter clients that send more than
976 * several kilobytes at once.
978 rb_define_const(cHttpParser, "CHUNK_MAX", OFFT2NUM(UH_OFF_T_MAX));
981 * The maximum size of the body as specified by Content-Length.
982 * This is only a theoretical maximum, the actual limit is subject
983 * to the limits of the file system used for +Dir.tmpdir+.
985 rb_define_const(cHttpParser, "LENGTH_MAX", OFFT2NUM(UH_OFF_T_MAX));
987 /* default value for keepalive_requests */
988 rb_define_const(cHttpParser, "KEEPALIVE_REQUESTS_DEFAULT",
989 ULONG2NUM(keepalive_requests));
991 rb_define_singleton_method(cHttpParser, "keepalive_requests", ka_req, 0);
992 rb_define_singleton_method(cHttpParser, "keepalive_requests=", set_ka_req, 1);
993 rb_define_singleton_method(cHttpParser, "max_header_len=", set_maxhdrlen, 1);
995 init_common_fields();
996 SET_GLOBAL(g_http_host, "HOST");
997 SET_GLOBAL(g_http_trailer, "TRAILER");
998 SET_GLOBAL(g_http_transfer_encoding, "TRANSFER_ENCODING");
999 SET_GLOBAL(g_content_length, "CONTENT_LENGTH");
1000 SET_GLOBAL(g_http_connection, "CONNECTION");
1001 id_clear = rb_intern("clear");
1002 id_set_backtrace = rb_intern("set_backtrace");
1003 id_response_start_sent = rb_intern("@response_start_sent");
1004 init_unicorn_httpdate();