1 /* Licensed to the Apache Software Foundation (ASF) under one or more
2 * contributor license agreements. See the NOTICE file distributed with
3 * this work for additional information regarding copyright ownership.
4 * The ASF licenses this file to You under the Apache License, Version 2.0
5 * (the "License"); you may not use this file except in compliance with
6 * the License. You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 /* HTTP routines for Apache proxy */
19 #include "mod_proxy.h"
22 module AP_MODULE_DECLARE_DATA proxy_http_module;
24 static apr_status_t ap_proxy_http_cleanup(const char *scheme,
26 proxy_conn_rec *backend);
29 * Canonicalise http-like URLs.
30 * scheme is the scheme for the URL
31 * url is the URL starting with the first '/'
32 * def_port is the default port for this scheme.
34 static int proxy_http_canon(request_rec *r, char *url)
36 char *host, *path, sport[7];
40 apr_port_t port, def_port;
42 /* ap_port_of_scheme() */
43 if (strncasecmp(url, "http:", 5) == 0) {
47 else if (strncasecmp(url, "https:", 6) == 0) {
54 def_port = apr_uri_port_of_scheme(scheme);
56 ap_log_error(APLOG_MARK, APLOG_TRACE1, 0, r->server,
57 "proxy: HTTP: canonicalising URL %s", url);
60 * We break the URL into host, port, path, search
63 err = ap_proxy_canon_netloc(r->pool, &url, NULL, NULL, &host, &port);
65 ap_log_rerror(APLOG_MARK, APLOG_ERR, 0, r,
66 "error parsing URL %s: %s",
68 return HTTP_BAD_REQUEST;
72 * now parse path/search args, according to rfc1738:
75 * In a reverse proxy, our URL has been processed, so canonicalise
76 * unless proxy-nocanon is set to say it's raw
77 * In a forward proxy, we have and MUST NOT MANGLE the original.
79 switch (r->proxyreq) {
80 default: /* wtf are we doing here? */
81 case PROXYREQ_REVERSE:
82 if (apr_table_get(r->notes, "proxy-nocanon")) {
83 path = url; /* this is the raw path */
86 path = ap_proxy_canonenc(r->pool, url, strlen(url),
87 enc_path, 0, r->proxyreq);
97 return HTTP_BAD_REQUEST;
100 apr_snprintf(sport, sizeof(sport), ":%d", port);
104 if (ap_strchr_c(host, ':')) { /* if literal IPv6 address */
105 host = apr_pstrcat(r->pool, "[", host, "]", NULL);
107 r->filename = apr_pstrcat(r->pool, "proxy:", scheme, "://", host, sport,
108 "/", path, (search) ? "?" : "", (search) ? search : "", NULL);
112 /* Clear all connection-based headers from the incoming headers table */
113 typedef struct header_dptr {
118 static ap_regex_t *warn_rx;
119 static int clean_warning_headers(void *data, const char *key, const char *val)
121 apr_table_t *headers = ((header_dptr*)data)->table;
122 apr_pool_t *pool = ((header_dptr*)data)->pool;
125 apr_time_t warn_time;
126 const int nmatch = 3;
127 ap_regmatch_t pmatch[3];
129 if (headers == NULL) {
130 ((header_dptr*)data)->table = headers = apr_table_make(pool, 2);
133 * Parse this, suckers!
135 * Warning = "Warning" ":" 1#warning-value
137 * warning-value = warn-code SP warn-agent SP warn-text
141 * warn-agent = ( host [ ":" port ] ) | pseudonym
142 * ; the name or pseudonym of the server adding
143 * ; the Warning header, for use in debugging
144 * warn-text = quoted-string
145 * warn-date = <"> HTTP-date <">
147 * Buggrit, use a bloomin' regexp!
148 * (\d{3}\s+\S+\s+\".*?\"(\s+\"(.*?)\")?) --> whole in $1, date in $3
150 while (!ap_regexec(warn_rx, val, nmatch, pmatch, 0)) {
151 warning = apr_pstrndup(pool, val+pmatch[0].rm_so,
152 pmatch[0].rm_eo - pmatch[0].rm_so);
154 if (pmatch[2].rm_eo > pmatch[2].rm_so) {
155 /* OK, we have a date here */
156 date = apr_pstrndup(pool, val+pmatch[2].rm_so,
157 pmatch[2].rm_eo - pmatch[2].rm_so);
158 warn_time = apr_date_parse_http(date);
160 if (!warn_time || (warn_time == ((header_dptr*)data)->time)) {
161 apr_table_addn(headers, key, warning);
163 val += pmatch[0].rm_eo;
167 static apr_table_t *ap_proxy_clean_warnings(apr_pool_t *p, apr_table_t *headers)
172 x.time = apr_date_parse_http(apr_table_get(headers, "Date"));
173 apr_table_do(clean_warning_headers, &x, headers, "Warning", NULL);
174 if (x.table != NULL) {
175 apr_table_unset(headers, "Warning");
176 return apr_table_overlay(p, headers, x.table);
182 static int clear_conn_headers(void *data, const char *key, const char *val)
184 apr_table_t *headers = ((header_dptr*)data)->table;
185 apr_pool_t *pool = ((header_dptr*)data)->pool;
187 char *next = apr_pstrdup(pool, val);
190 while (*next && !apr_isspace(*next) && (*next != ',')) {
193 while (*next && (apr_isspace(*next) || (*next == ','))) {
196 apr_table_unset(headers, name);
200 static void ap_proxy_clear_connection(apr_pool_t *p, apr_table_t *headers)
205 apr_table_unset(headers, "Proxy-Connection");
206 apr_table_do(clear_conn_headers, &x, headers, "Connection", NULL);
207 apr_table_unset(headers, "Connection");
209 static void add_te_chunked(apr_pool_t *p,
210 apr_bucket_alloc_t *bucket_alloc,
211 apr_bucket_brigade *header_brigade)
215 const char te_hdr[] = "Transfer-Encoding: chunked" CRLF;
217 buf = apr_pmemdup(p, te_hdr, sizeof(te_hdr)-1);
218 ap_xlate_proto_to_ascii(buf, sizeof(te_hdr)-1);
220 e = apr_bucket_pool_create(buf, sizeof(te_hdr)-1, p, bucket_alloc);
221 APR_BRIGADE_INSERT_TAIL(header_brigade, e);
224 static void add_cl(apr_pool_t *p,
225 apr_bucket_alloc_t *bucket_alloc,
226 apr_bucket_brigade *header_brigade,
232 buf = apr_pstrcat(p, "Content-Length: ",
236 ap_xlate_proto_to_ascii(buf, strlen(buf));
237 e = apr_bucket_pool_create(buf, strlen(buf), p, bucket_alloc);
238 APR_BRIGADE_INSERT_TAIL(header_brigade, e);
241 #define ASCII_CRLF "\015\012"
242 #define ASCII_ZERO "\060"
244 static void terminate_headers(apr_bucket_alloc_t *bucket_alloc,
245 apr_bucket_brigade *header_brigade)
249 /* add empty line at the end of the headers */
250 e = apr_bucket_immortal_create(ASCII_CRLF, 2, bucket_alloc);
251 APR_BRIGADE_INSERT_TAIL(header_brigade, e);
254 static int pass_brigade(apr_bucket_alloc_t *bucket_alloc,
255 request_rec *r, proxy_conn_rec *conn,
256 conn_rec *origin, apr_bucket_brigade *bb,
260 apr_off_t transferred;
263 apr_bucket *e = apr_bucket_flush_create(bucket_alloc);
264 APR_BRIGADE_INSERT_TAIL(bb, e);
266 apr_brigade_length(bb, 0, &transferred);
267 if (transferred != -1)
268 conn->worker->s->transferred += transferred;
269 status = ap_pass_brigade(origin->output_filters, bb);
270 if (status != APR_SUCCESS) {
271 ap_log_error(APLOG_MARK, APLOG_ERR, status, r->server,
272 "proxy: pass request body failed to %pI (%s)",
273 conn->addr, conn->hostname);
274 if (origin->aborted) {
275 return APR_STATUS_IS_TIMEUP(status) ? HTTP_GATEWAY_TIME_OUT : HTTP_BAD_GATEWAY;
278 return HTTP_BAD_REQUEST;
281 apr_brigade_cleanup(bb);
285 #define MAX_MEM_SPOOL 16384
287 static int stream_reqbody_chunked(apr_pool_t *p,
289 proxy_conn_rec *p_conn,
291 apr_bucket_brigade *header_brigade,
292 apr_bucket_brigade *input_brigade)
294 int seen_eos = 0, rv = OK;
298 apr_bucket_alloc_t *bucket_alloc = r->connection->bucket_alloc;
299 apr_bucket_brigade *bb;
302 add_te_chunked(p, bucket_alloc, header_brigade);
303 terminate_headers(bucket_alloc, header_brigade);
305 while (!APR_BUCKET_IS_EOS(APR_BRIGADE_FIRST(input_brigade)))
307 char chunk_hdr[20]; /* must be here due to transient bucket. */
309 /* If this brigade contains EOS, either stop or remove it. */
310 if (APR_BUCKET_IS_EOS(APR_BRIGADE_LAST(input_brigade))) {
313 /* We can't pass this EOS to the output_filters. */
314 e = APR_BRIGADE_LAST(input_brigade);
315 apr_bucket_delete(e);
318 apr_brigade_length(input_brigade, 1, &bytes);
320 hdr_len = apr_snprintf(chunk_hdr, sizeof(chunk_hdr),
321 "%" APR_UINT64_T_HEX_FMT CRLF,
322 (apr_uint64_t)bytes);
324 ap_xlate_proto_to_ascii(chunk_hdr, hdr_len);
325 e = apr_bucket_transient_create(chunk_hdr, hdr_len,
327 APR_BRIGADE_INSERT_HEAD(input_brigade, e);
330 * Append the end-of-chunk CRLF
332 e = apr_bucket_immortal_create(ASCII_CRLF, 2, bucket_alloc);
333 APR_BRIGADE_INSERT_TAIL(input_brigade, e);
335 if (header_brigade) {
336 /* we never sent the header brigade, so go ahead and
337 * take care of that now
342 * Save input_brigade in bb brigade. (At least) in the SSL case
343 * input_brigade contains transient buckets whose data would get
344 * overwritten during the next call of ap_get_brigade in the loop.
345 * ap_save_brigade ensures these buckets to be set aside.
346 * Calling ap_save_brigade with NULL as filter is OK, because
347 * bb brigade already has been created and does not need to get
348 * created by ap_save_brigade.
350 status = ap_save_brigade(NULL, &bb, &input_brigade, p);
351 if (status != APR_SUCCESS) {
352 return HTTP_INTERNAL_SERVER_ERROR;
355 header_brigade = NULL;
361 /* The request is flushed below this loop with chunk EOS header */
362 rv = pass_brigade(bucket_alloc, r, p_conn, origin, bb, 0);
371 status = ap_get_brigade(r->input_filters, input_brigade,
372 AP_MODE_READBYTES, APR_BLOCK_READ,
375 if (status != APR_SUCCESS) {
376 return HTTP_BAD_REQUEST;
380 if (header_brigade) {
381 /* we never sent the header brigade because there was no request body;
387 if (!APR_BRIGADE_EMPTY(input_brigade)) {
388 /* input brigade still has an EOS which we can't pass to the output_filters. */
389 e = APR_BRIGADE_LAST(input_brigade);
390 AP_DEBUG_ASSERT(APR_BUCKET_IS_EOS(e));
391 apr_bucket_delete(e);
396 e = apr_bucket_immortal_create(ASCII_ZERO ASCII_CRLF
400 APR_BRIGADE_INSERT_TAIL(bb, e);
402 if (apr_table_get(r->subprocess_env, "proxy-sendextracrlf")) {
403 e = apr_bucket_immortal_create(ASCII_CRLF, 2, bucket_alloc);
404 APR_BRIGADE_INSERT_TAIL(bb, e);
407 /* Now we have headers-only, or the chunk EOS mark; flush it */
408 rv = pass_brigade(bucket_alloc, r, p_conn, origin, bb, 1);
412 static int stream_reqbody_cl(apr_pool_t *p,
414 proxy_conn_rec *p_conn,
416 apr_bucket_brigade *header_brigade,
417 apr_bucket_brigade *input_brigade,
418 const char *old_cl_val)
420 int seen_eos = 0, rv = 0;
421 apr_status_t status = APR_SUCCESS;
422 apr_bucket_alloc_t *bucket_alloc = r->connection->bucket_alloc;
423 apr_bucket_brigade *bb;
425 apr_off_t cl_val = 0;
427 apr_off_t bytes_streamed = 0;
432 add_cl(p, bucket_alloc, header_brigade, old_cl_val);
433 status = apr_strtoff(&cl_val, old_cl_val, &endstr, 10);
435 if (status || *endstr || endstr == old_cl_val || cl_val < 0) {
436 ap_log_rerror(APLOG_MARK, APLOG_ERR, status, r,
437 "proxy: could not parse request Content-Length (%s)",
439 return HTTP_BAD_REQUEST;
442 terminate_headers(bucket_alloc, header_brigade);
444 while (!APR_BUCKET_IS_EOS(APR_BRIGADE_FIRST(input_brigade)))
446 apr_brigade_length(input_brigade, 1, &bytes);
447 bytes_streamed += bytes;
449 /* If this brigade contains EOS, either stop or remove it. */
450 if (APR_BUCKET_IS_EOS(APR_BRIGADE_LAST(input_brigade))) {
453 /* We can't pass this EOS to the output_filters. */
454 e = APR_BRIGADE_LAST(input_brigade);
455 apr_bucket_delete(e);
457 if (apr_table_get(r->subprocess_env, "proxy-sendextracrlf")) {
458 e = apr_bucket_immortal_create(ASCII_CRLF, 2, bucket_alloc);
459 APR_BRIGADE_INSERT_TAIL(input_brigade, e);
463 /* C-L < bytes streamed?!?
464 * We will error out after the body is completely
465 * consumed, but we can't stream more bytes at the
466 * back end since they would in part be interpreted
467 * as another request! If nothing is sent, then
470 * Prevents HTTP Response Splitting.
472 if (bytes_streamed > cl_val) {
473 ap_log_rerror(APLOG_MARK, APLOG_ERR, 0, r,
474 "proxy: read more bytes of request body than expected "
475 "(got %" APR_OFF_T_FMT ", expected %" APR_OFF_T_FMT ")",
476 bytes_streamed, cl_val);
477 return HTTP_INTERNAL_SERVER_ERROR;
480 if (header_brigade) {
481 /* we never sent the header brigade, so go ahead and
482 * take care of that now
487 * Save input_brigade in bb brigade. (At least) in the SSL case
488 * input_brigade contains transient buckets whose data would get
489 * overwritten during the next call of ap_get_brigade in the loop.
490 * ap_save_brigade ensures these buckets to be set aside.
491 * Calling ap_save_brigade with NULL as filter is OK, because
492 * bb brigade already has been created and does not need to get
493 * created by ap_save_brigade.
495 status = ap_save_brigade(NULL, &bb, &input_brigade, p);
496 if (status != APR_SUCCESS) {
497 return HTTP_INTERNAL_SERVER_ERROR;
500 header_brigade = NULL;
506 /* Once we hit EOS, we are ready to flush. */
507 rv = pass_brigade(bucket_alloc, r, p_conn, origin, bb, seen_eos);
516 status = ap_get_brigade(r->input_filters, input_brigade,
517 AP_MODE_READBYTES, APR_BLOCK_READ,
520 if (status != APR_SUCCESS) {
521 return HTTP_BAD_REQUEST;
525 if (bytes_streamed != cl_val) {
526 ap_log_error(APLOG_MARK, APLOG_ERR, 0, r->server,
527 "proxy: client %s given Content-Length did not match"
528 " number of body bytes read", r->connection->remote_ip);
529 return HTTP_BAD_REQUEST;
532 if (header_brigade) {
533 /* we never sent the header brigade since there was no request
534 * body; send it now with the flush flag
537 return(pass_brigade(bucket_alloc, r, p_conn, origin, bb, 1));
543 static int spool_reqbody_cl(apr_pool_t *p,
545 proxy_conn_rec *p_conn,
547 apr_bucket_brigade *header_brigade,
548 apr_bucket_brigade *input_brigade,
553 apr_bucket_alloc_t *bucket_alloc = r->connection->bucket_alloc;
554 apr_bucket_brigade *body_brigade;
556 apr_off_t bytes, bytes_spooled = 0, fsize = 0;
557 apr_file_t *tmpfile = NULL;
559 body_brigade = apr_brigade_create(p, bucket_alloc);
561 while (!APR_BUCKET_IS_EOS(APR_BRIGADE_FIRST(input_brigade)))
563 /* If this brigade contains EOS, either stop or remove it. */
564 if (APR_BUCKET_IS_EOS(APR_BRIGADE_LAST(input_brigade))) {
567 /* We can't pass this EOS to the output_filters. */
568 e = APR_BRIGADE_LAST(input_brigade);
569 apr_bucket_delete(e);
572 apr_brigade_length(input_brigade, 1, &bytes);
574 if (bytes_spooled + bytes > MAX_MEM_SPOOL) {
575 /* can't spool any more in memory; write latest brigade to disk */
576 if (tmpfile == NULL) {
577 const char *temp_dir;
580 status = apr_temp_dir_get(&temp_dir, p);
581 if (status != APR_SUCCESS) {
582 ap_log_error(APLOG_MARK, APLOG_ERR, status, r->server,
583 "proxy: search for temporary directory failed");
584 return HTTP_INTERNAL_SERVER_ERROR;
586 apr_filepath_merge(&template, temp_dir,
587 "modproxy.tmp.XXXXXX",
588 APR_FILEPATH_NATIVE, p);
589 status = apr_file_mktemp(&tmpfile, template, 0, p);
590 if (status != APR_SUCCESS) {
591 ap_log_error(APLOG_MARK, APLOG_ERR, status, r->server,
592 "proxy: creation of temporary file in directory %s failed",
594 return HTTP_INTERNAL_SERVER_ERROR;
597 for (e = APR_BRIGADE_FIRST(input_brigade);
598 e != APR_BRIGADE_SENTINEL(input_brigade);
599 e = APR_BUCKET_NEXT(e)) {
601 apr_size_t bytes_read, bytes_written;
603 apr_bucket_read(e, &data, &bytes_read, APR_BLOCK_READ);
604 status = apr_file_write_full(tmpfile, data, bytes_read, &bytes_written);
605 if (status != APR_SUCCESS) {
606 const char *tmpfile_name;
608 if (apr_file_name_get(&tmpfile_name, tmpfile) != APR_SUCCESS) {
609 tmpfile_name = "(unknown)";
611 ap_log_error(APLOG_MARK, APLOG_ERR, status, r->server,
612 "proxy: write to temporary file %s failed",
614 return HTTP_INTERNAL_SERVER_ERROR;
616 AP_DEBUG_ASSERT(bytes_read == bytes_written);
617 fsize += bytes_written;
619 apr_brigade_cleanup(input_brigade);
624 * Save input_brigade in body_brigade. (At least) in the SSL case
625 * input_brigade contains transient buckets whose data would get
626 * overwritten during the next call of ap_get_brigade in the loop.
627 * ap_save_brigade ensures these buckets to be set aside.
628 * Calling ap_save_brigade with NULL as filter is OK, because
629 * body_brigade already has been created and does not need to get
630 * created by ap_save_brigade.
632 status = ap_save_brigade(NULL, &body_brigade, &input_brigade, p);
633 if (status != APR_SUCCESS) {
634 return HTTP_INTERNAL_SERVER_ERROR;
639 bytes_spooled += bytes;
645 status = ap_get_brigade(r->input_filters, input_brigade,
646 AP_MODE_READBYTES, APR_BLOCK_READ,
649 if (status != APR_SUCCESS) {
650 return HTTP_BAD_REQUEST;
654 if (bytes_spooled || force_cl) {
655 add_cl(p, bucket_alloc, header_brigade, apr_off_t_toa(p, bytes_spooled));
657 terminate_headers(bucket_alloc, header_brigade);
658 APR_BRIGADE_CONCAT(header_brigade, body_brigade);
660 apr_brigade_insert_file(header_brigade, tmpfile, 0, fsize, p);
662 if (apr_table_get(r->subprocess_env, "proxy-sendextracrlf")) {
663 e = apr_bucket_immortal_create(ASCII_CRLF, 2, bucket_alloc);
664 APR_BRIGADE_INSERT_TAIL(header_brigade, e);
666 /* This is all a single brigade, pass with flush flagged */
667 return(pass_brigade(bucket_alloc, r, p_conn, origin, header_brigade, 1));
671 int ap_proxy_http_request(apr_pool_t *p, request_rec *r,
672 proxy_conn_rec *p_conn, proxy_worker *worker,
673 proxy_server_conf *conf,
675 char *url, char *server_portstr)
677 conn_rec *c = r->connection;
678 apr_bucket_alloc_t *bucket_alloc = c->bucket_alloc;
679 apr_bucket_brigade *header_brigade;
680 apr_bucket_brigade *input_brigade;
681 apr_bucket_brigade *temp_brigade;
684 const apr_array_header_t *headers_in_array;
685 const apr_table_entry_t *headers_in;
688 enum rb_methods {RB_INIT, RB_STREAM_CL, RB_STREAM_CHUNKED, RB_SPOOL_CL};
689 enum rb_methods rb_method = RB_INIT;
690 const char *old_cl_val = NULL;
691 const char *old_te_val = NULL;
692 apr_off_t bytes_read = 0;
695 apr_table_t *headers_in_copy;
696 proxy_dir_conf *dconf;
697 conn_rec *origin = p_conn->connection;
700 dconf = ap_get_module_config(r->per_dir_config, &proxy_module);
701 header_brigade = apr_brigade_create(p, origin->bucket_alloc);
704 * Send the HTTP/1.1 request to the remote server
707 do_100_continue = (worker->ping_timeout_set
709 && (PROXYREQ_REVERSE == r->proxyreq)
710 && !(apr_table_get(r->subprocess_env, "force-proxy-request-1.0")));
712 if (apr_table_get(r->subprocess_env, "force-proxy-request-1.0")) {
714 * According to RFC 2616 8.2.3 we are not allowed to forward an
715 * Expect: 100-continue to an HTTP/1.0 server. Instead we MUST return
716 * a HTTP_EXPECTATION_FAILED
718 if (r->expecting_100) {
719 return HTTP_EXPECTATION_FAILED;
721 buf = apr_pstrcat(p, r->method, " ", url, " HTTP/1.0" CRLF, NULL);
725 buf = apr_pstrcat(p, r->method, " ", url, " HTTP/1.1" CRLF, NULL);
728 if (apr_table_get(r->subprocess_env, "proxy-nokeepalive")) {
729 origin->keepalive = AP_CONN_CLOSE;
732 ap_xlate_proto_to_ascii(buf, strlen(buf));
733 e = apr_bucket_pool_create(buf, strlen(buf), p, c->bucket_alloc);
734 APR_BRIGADE_INSERT_TAIL(header_brigade, e);
735 if (dconf->preserve_host == 0) {
736 if (ap_strchr_c(uri->hostname, ':')) { /* if literal IPv6 address */
737 if (uri->port_str && uri->port != DEFAULT_HTTP_PORT) {
738 buf = apr_pstrcat(p, "Host: [", uri->hostname, "]:",
739 uri->port_str, CRLF, NULL);
741 buf = apr_pstrcat(p, "Host: [", uri->hostname, "]", CRLF, NULL);
744 if (uri->port_str && uri->port != DEFAULT_HTTP_PORT) {
745 buf = apr_pstrcat(p, "Host: ", uri->hostname, ":",
746 uri->port_str, CRLF, NULL);
748 buf = apr_pstrcat(p, "Host: ", uri->hostname, CRLF, NULL);
753 /* don't want to use r->hostname, as the incoming header might have a
756 const char* hostname = apr_table_get(r->headers_in,"Host");
758 hostname = r->server->server_hostname;
759 ap_log_rerror(APLOG_MARK, APLOG_WARNING, 0, r,
760 "proxy: no HTTP 0.9 request (with no host line) "
761 "on incoming request and preserve host set "
762 "forcing hostname to be %s for uri %s",
766 buf = apr_pstrcat(p, "Host: ", hostname, CRLF, NULL);
768 ap_xlate_proto_to_ascii(buf, strlen(buf));
769 e = apr_bucket_pool_create(buf, strlen(buf), p, c->bucket_alloc);
770 APR_BRIGADE_INSERT_TAIL(header_brigade, e);
773 if (conf->viaopt == via_block) {
774 /* Block all outgoing Via: headers */
775 apr_table_unset(r->headers_in, "Via");
776 } else if (conf->viaopt != via_off) {
777 const char *server_name = ap_get_server_name(r);
778 /* If USE_CANONICAL_NAME_OFF was configured for the proxy virtual host,
779 * then the server name returned by ap_get_server_name() is the
780 * origin server name (which does make too much sense with Via: headers)
781 * so we use the proxy vhost's name instead.
783 if (server_name == r->hostname)
784 server_name = r->server->server_hostname;
785 /* Create a "Via:" request header entry and merge it */
786 /* Generate outgoing Via: header with/without server comment: */
787 apr_table_mergen(r->headers_in, "Via",
788 (conf->viaopt == via_full)
789 ? apr_psprintf(p, "%d.%d %s%s (%s)",
790 HTTP_VERSION_MAJOR(r->proto_num),
791 HTTP_VERSION_MINOR(r->proto_num),
792 server_name, server_portstr,
793 AP_SERVER_BASEVERSION)
794 : apr_psprintf(p, "%d.%d %s%s",
795 HTTP_VERSION_MAJOR(r->proto_num),
796 HTTP_VERSION_MINOR(r->proto_num),
797 server_name, server_portstr)
801 /* Use HTTP/1.1 100-Continue as quick "HTTP ping" test
804 if (do_100_continue) {
805 apr_table_mergen(r->headers_in, "Expect", "100-Continue");
806 r->expecting_100 = 1;
809 /* X-Forwarded-*: handling
814 * These request headers are only really useful when the mod_proxy
815 * is used in a reverse proxy configuration, so that useful info
816 * about the client can be passed through the reverse proxy and on
817 * to the backend server, which may require the information to
820 * In a forward proxy situation, these options are a potential
821 * privacy violation, as information about clients behind the proxy
822 * are revealed to arbitrary servers out there on the internet.
824 * The HTTP/1.1 Via: header is designed for passing client
825 * information through proxies to a server, and should be used in
826 * a forward proxy configuation instead of X-Forwarded-*. See the
827 * ProxyVia option for details.
830 if (PROXYREQ_REVERSE == r->proxyreq) {
833 /* Add X-Forwarded-For: so that the upstream has a chance to
834 * determine, where the original request came from.
836 apr_table_mergen(r->headers_in, "X-Forwarded-For",
839 /* Add X-Forwarded-Host: so that upstream knows what the
840 * original request hostname was.
842 if ((buf = apr_table_get(r->headers_in, "Host"))) {
843 apr_table_mergen(r->headers_in, "X-Forwarded-Host", buf);
846 /* Add X-Forwarded-Server: so that upstream knows what the
847 * name of this proxy server is (if there are more than one)
848 * XXX: This duplicates Via: - do we strictly need it?
850 apr_table_mergen(r->headers_in, "X-Forwarded-Server",
851 r->server->server_hostname);
856 * Make a copy of the headers_in table before clearing the connection
857 * headers as we need the connection headers later in the http output
858 * filter to prepare the correct response headers.
860 * Note: We need to take r->pool for apr_table_copy as the key / value
861 * pairs in r->headers_in have been created out of r->pool and
862 * p might be (and actually is) a longer living pool.
863 * This would trigger the bad pool ancestry abort in apr_table_copy if
864 * apr is compiled with APR_POOL_DEBUG.
866 headers_in_copy = apr_table_copy(r->pool, r->headers_in);
867 ap_proxy_clear_connection(p, headers_in_copy);
868 /* send request headers */
869 headers_in_array = apr_table_elts(headers_in_copy);
870 headers_in = (const apr_table_entry_t *) headers_in_array->elts;
871 for (counter = 0; counter < headers_in_array->nelts; counter++) {
872 if (headers_in[counter].key == NULL
873 || headers_in[counter].val == NULL
876 || !strcasecmp(headers_in[counter].key, "Host")
878 /* Clear out hop-by-hop request headers not to send
879 * RFC2616 13.5.1 says we should strip these headers
881 || !strcasecmp(headers_in[counter].key, "Keep-Alive")
882 || !strcasecmp(headers_in[counter].key, "TE")
883 || !strcasecmp(headers_in[counter].key, "Trailer")
884 || !strcasecmp(headers_in[counter].key, "Upgrade")
889 /* Do we want to strip Proxy-Authorization ?
890 * If we haven't used it, then NO
891 * If we have used it then MAYBE: RFC2616 says we MAY propagate it.
892 * So let's make it configurable by env.
894 if (!strcasecmp(headers_in[counter].key,"Proxy-Authorization")) {
895 if (r->user != NULL) { /* we've authenticated */
896 if (!apr_table_get(r->subprocess_env, "Proxy-Chain-Auth")) {
903 /* Skip Transfer-Encoding and Content-Length for now.
905 if (!strcasecmp(headers_in[counter].key, "Transfer-Encoding")) {
906 old_te_val = headers_in[counter].val;
909 if (!strcasecmp(headers_in[counter].key, "Content-Length")) {
910 old_cl_val = headers_in[counter].val;
914 /* for sub-requests, ignore freshness/expiry headers */
916 if ( !strcasecmp(headers_in[counter].key, "If-Match")
917 || !strcasecmp(headers_in[counter].key, "If-Modified-Since")
918 || !strcasecmp(headers_in[counter].key, "If-Range")
919 || !strcasecmp(headers_in[counter].key, "If-Unmodified-Since")
920 || !strcasecmp(headers_in[counter].key, "If-None-Match")) {
925 buf = apr_pstrcat(p, headers_in[counter].key, ": ",
926 headers_in[counter].val, CRLF,
928 ap_xlate_proto_to_ascii(buf, strlen(buf));
929 e = apr_bucket_pool_create(buf, strlen(buf), p, c->bucket_alloc);
930 APR_BRIGADE_INSERT_TAIL(header_brigade, e);
933 /* We have headers, let's figure out our request body... */
934 input_brigade = apr_brigade_create(p, bucket_alloc);
936 /* sub-requests never use keepalives, and mustn't pass request bodies.
937 * Because the new logic looks at input_brigade, we will self-terminate
938 * input_brigade and jump past all of the request body logic...
939 * Reading anything with ap_get_brigade is likely to consume the
940 * main request's body or read beyond EOS - which would be unplesant.
942 * An exception: when a kept_body is present, then subrequest CAN use
943 * pass request bodies, and we DONT skip the body.
945 if (!r->kept_body && r->main) {
946 /* XXX: Why DON'T sub-requests use keepalives? */
950 apr_table_unset(r->headers_in, "Content-Length");
954 apr_table_unset(r->headers_in, "Transfer-Encoding");
956 rb_method = RB_STREAM_CL;
957 e = apr_bucket_eos_create(input_brigade->bucket_alloc);
958 APR_BRIGADE_INSERT_TAIL(input_brigade, e);
962 /* WE only understand chunked. Other modules might inject
963 * (and therefore, decode) other flavors but we don't know
964 * that the can and have done so unless they they remove
965 * their decoding from the headers_in T-E list.
966 * XXX: Make this extensible, but in doing so, presume the
967 * encoding has been done by the extensions' handler, and
968 * do not modify add_te_chunked's logic
970 if (old_te_val && strcasecmp(old_te_val, "chunked") != 0) {
971 ap_log_error(APLOG_MARK, APLOG_ERR, 0, r->server,
972 "proxy: %s Transfer-Encoding is not supported",
974 return HTTP_INTERNAL_SERVER_ERROR;
977 if (old_cl_val && old_te_val) {
978 ap_log_error(APLOG_MARK, APLOG_DEBUG, APR_ENOTIMPL, r->server,
979 "proxy: client %s (%s) requested Transfer-Encoding "
980 "chunked body with Content-Length (C-L ignored)",
981 c->remote_ip, c->remote_host ? c->remote_host: "");
982 apr_table_unset(r->headers_in, "Content-Length");
984 origin->keepalive = AP_CONN_CLOSE;
988 /* Prefetch MAX_MEM_SPOOL bytes
990 * This helps us avoid any election of C-L v.s. T-E
991 * request bodies, since we are willing to keep in
992 * memory this much data, in any case. This gives
993 * us an instant C-L election if the body is of some
996 temp_brigade = apr_brigade_create(p, bucket_alloc);
998 status = ap_get_brigade(r->input_filters, temp_brigade,
999 AP_MODE_READBYTES, APR_BLOCK_READ,
1000 MAX_MEM_SPOOL - bytes_read);
1001 if (status != APR_SUCCESS) {
1002 ap_log_error(APLOG_MARK, APLOG_ERR, status, r->server,
1003 "proxy: prefetch request body failed to %pI (%s)"
1005 p_conn->addr, p_conn->hostname ? p_conn->hostname: "",
1006 c->remote_ip, c->remote_host ? c->remote_host: "");
1007 return HTTP_BAD_REQUEST;
1010 apr_brigade_length(temp_brigade, 1, &bytes);
1011 bytes_read += bytes;
1014 * Save temp_brigade in input_brigade. (At least) in the SSL case
1015 * temp_brigade contains transient buckets whose data would get
1016 * overwritten during the next call of ap_get_brigade in the loop.
1017 * ap_save_brigade ensures these buckets to be set aside.
1018 * Calling ap_save_brigade with NULL as filter is OK, because
1019 * input_brigade already has been created and does not need to get
1020 * created by ap_save_brigade.
1022 status = ap_save_brigade(NULL, &input_brigade, &temp_brigade, p);
1023 if (status != APR_SUCCESS) {
1024 ap_log_error(APLOG_MARK, APLOG_ERR, status, r->server,
1025 "proxy: processing prefetched request body failed"
1026 " to %pI (%s) from %s (%s)",
1027 p_conn->addr, p_conn->hostname ? p_conn->hostname: "",
1028 c->remote_ip, c->remote_host ? c->remote_host: "");
1029 return HTTP_INTERNAL_SERVER_ERROR;
1032 /* Ensure we don't hit a wall where we have a buffer too small
1033 * for ap_get_brigade's filters to fetch us another bucket,
1034 * surrender once we hit 80 bytes less than MAX_MEM_SPOOL
1035 * (an arbitrary value.)
1037 } while ((bytes_read < MAX_MEM_SPOOL - 80)
1038 && !APR_BUCKET_IS_EOS(APR_BRIGADE_LAST(input_brigade)));
1040 /* Use chunked request body encoding or send a content-length body?
1044 * We have no request body (handled by RB_STREAM_CL)
1046 * We have a request body length <= MAX_MEM_SPOOL
1048 * The administrator has setenv force-proxy-request-1.0
1050 * The client sent a C-L body, and the administrator has
1051 * not setenv proxy-sendchunked or has set setenv proxy-sendcl
1053 * The client sent a T-E body, and the administrator has
1054 * setenv proxy-sendcl, and not setenv proxy-sendchunked
1056 * If both proxy-sendcl and proxy-sendchunked are set, the
1057 * behavior is the same as if neither were set, large bodies
1058 * that can't be read will be forwarded in their original
1059 * form of C-L, or T-E.
1061 * To ensure maximum compatibility, setenv proxy-sendcl
1062 * To reduce server resource use, setenv proxy-sendchunked
1064 * Then address specific servers with conditional setenv
1065 * options to restore the default behavior where desireable.
1067 * We have to compute content length by reading the entire request
1068 * body; if request body is not small, we'll spool the remaining
1069 * input to a temporary file. Chunked is always preferable.
1071 * We can only trust the client-provided C-L if the T-E header
1072 * is absent, and the filters are unchanged (the body won't
1073 * be resized by another content filter).
1075 if (APR_BUCKET_IS_EOS(APR_BRIGADE_LAST(input_brigade))) {
1076 /* The whole thing fit, so our decision is trivial, use
1077 * the filtered bytes read from the client for the request
1078 * body Content-Length.
1080 * If we expected no body, and read no body, do not set
1081 * the Content-Length.
1083 if (old_cl_val || old_te_val || bytes_read) {
1084 old_cl_val = apr_off_t_toa(r->pool, bytes_read);
1086 rb_method = RB_STREAM_CL;
1088 else if (old_te_val) {
1090 || (apr_table_get(r->subprocess_env, "proxy-sendcl")
1091 && !apr_table_get(r->subprocess_env, "proxy-sendchunks")
1092 && !apr_table_get(r->subprocess_env, "proxy-sendchunked"))) {
1093 rb_method = RB_SPOOL_CL;
1096 rb_method = RB_STREAM_CHUNKED;
1099 else if (old_cl_val) {
1100 if (r->input_filters == r->proto_input_filters) {
1101 rb_method = RB_STREAM_CL;
1104 && (apr_table_get(r->subprocess_env, "proxy-sendchunks")
1105 || apr_table_get(r->subprocess_env, "proxy-sendchunked"))
1106 && !apr_table_get(r->subprocess_env, "proxy-sendcl")) {
1107 rb_method = RB_STREAM_CHUNKED;
1110 rb_method = RB_SPOOL_CL;
1114 /* This is an appropriate default; very efficient for no-body
1115 * requests, and has the behavior that it will not add any C-L
1116 * when the old_cl_val is NULL.
1118 rb_method = RB_SPOOL_CL;
1121 /* Yes I hate gotos. This is the subrequest shortcut */
1124 * Handle Connection: header if we do HTTP/1.1 request:
1125 * If we plan to close the backend connection sent Connection: close
1126 * otherwise sent Connection: Keep-Alive.
1129 if (p_conn->close) {
1130 buf = apr_pstrdup(p, "Connection: close" CRLF);
1133 buf = apr_pstrdup(p, "Connection: Keep-Alive" CRLF);
1135 ap_xlate_proto_to_ascii(buf, strlen(buf));
1136 e = apr_bucket_pool_create(buf, strlen(buf), p, c->bucket_alloc);
1137 APR_BRIGADE_INSERT_TAIL(header_brigade, e);
1140 /* send the request body, if any. */
1142 case RB_STREAM_CHUNKED:
1143 rv = stream_reqbody_chunked(p, r, p_conn, origin, header_brigade,
1147 rv = stream_reqbody_cl(p, r, p_conn, origin, header_brigade,
1148 input_brigade, old_cl_val);
1151 rv = spool_reqbody_cl(p, r, p_conn, origin, header_brigade,
1152 input_brigade, (old_cl_val != NULL)
1153 || (old_te_val != NULL)
1154 || (bytes_read > 0));
1157 /* shouldn't be possible */
1158 rv = HTTP_INTERNAL_SERVER_ERROR ;
1163 /* apr_status_t value has been logged in lower level method */
1164 ap_log_error(APLOG_MARK, APLOG_ERR, 0, r->server,
1165 "proxy: pass request body failed to %pI (%s)"
1168 p_conn->hostname ? p_conn->hostname: "",
1170 c->remote_host ? c->remote_host: "");
1177 static void process_proxy_header(request_rec *r, proxy_dir_conf *c,
1178 const char *key, const char *value)
1180 static const char *date_hdrs[]
1181 = { "Date", "Expires", "Last-Modified", NULL };
1182 static const struct {
1184 ap_proxy_header_reverse_map_fn func;
1185 } transform_hdrs[] = {
1186 { "Location", ap_proxy_location_reverse_map },
1187 { "Content-Location", ap_proxy_location_reverse_map },
1188 { "URI", ap_proxy_location_reverse_map },
1189 { "Destination", ap_proxy_location_reverse_map },
1190 { "Set-Cookie", ap_proxy_cookie_reverse_map },
1194 for (i = 0; date_hdrs[i]; ++i) {
1195 if (!strcasecmp(date_hdrs[i], key)) {
1196 apr_table_add(r->headers_out, key,
1197 ap_proxy_date_canon(r->pool, value));
1201 for (i = 0; transform_hdrs[i].name; ++i) {
1202 if (!strcasecmp(transform_hdrs[i].name, key)) {
1203 apr_table_add(r->headers_out, key,
1204 (*transform_hdrs[i].func)(r, c, value));
1208 apr_table_add(r->headers_out, key, value);
1213 * Note: pread_len is the length of the response that we've mistakenly
1214 * read (assuming that we don't consider that an error via
1215 * ProxyBadHeader StartBody). This depends on buffer actually being
1216 * local storage to the calling code in order for pread_len to make
1217 * any sense at all, since we depend on buffer still containing
1218 * what was read by ap_getline() upon return.
1220 static void ap_proxy_read_headers(request_rec *r, request_rec *rr,
1221 char *buffer, int size,
1222 conn_rec *c, int *pread_len)
1226 char field[MAX_STRING_LEN];
1227 int saw_headers = 0;
1228 void *sconf = r->server->module_config;
1229 proxy_server_conf *psc;
1230 proxy_dir_conf *dconf;
1232 dconf = ap_get_module_config(r->per_dir_config, &proxy_module);
1233 psc = (proxy_server_conf *) ap_get_module_config(sconf, &proxy_module);
1235 r->headers_out = apr_table_make(r->pool, 20);
1239 * Read header lines until we get the empty separator line, a read error,
1240 * the connection closes (EOF), or we timeout.
1242 ap_log_rerror(APLOG_MARK, APLOG_TRACE4, 0, r,
1243 "Headers received from backend:");
1244 while ((len = ap_getline(buffer, size, rr, 1)) > 0) {
1245 ap_log_rerror(APLOG_MARK, APLOG_TRACE4, 0, r, "%s", buffer);
1247 if (!(value = strchr(buffer, ':'))) { /* Find the colon separator */
1249 /* We may encounter invalid headers, usually from buggy
1250 * MS IIS servers, so we need to determine just how to handle
1251 * them. We can either ignore them, assume that they mark the
1252 * start-of-body (eg: a missing CRLF) or (the default) mark
1253 * the headers as totally bogus and return a 500. The sole
1254 * exception is an extra "HTTP/1.0 200, OK" line sprinkled
1255 * in between the usual MIME headers, which is a favorite
1258 /* XXX: The mask check is buggy if we ever see an HTTP/1.10 */
1260 if (!apr_date_checkmask(buffer, "HTTP/#.# ###*")) {
1261 if (psc->badopt == bad_error) {
1262 /* Nope, it wasn't even an extra HTTP header. Give up. */
1263 r->headers_out = NULL;
1266 else if (psc->badopt == bad_body) {
1267 /* if we've already started loading headers_out, then
1268 * return what we've accumulated so far, in the hopes
1269 * that they are useful; also note that we likely pre-read
1270 * the first line of the response.
1273 ap_log_error(APLOG_MARK, APLOG_WARNING, 0, r->server,
1274 "proxy: Starting body due to bogus non-header in headers "
1275 "returned by %s (%s)", r->uri, r->method);
1279 ap_log_error(APLOG_MARK, APLOG_WARNING, 0, r->server,
1280 "proxy: No HTTP headers "
1281 "returned by %s (%s)", r->uri, r->method);
1286 /* this is the psc->badopt == bad_ignore case */
1287 ap_log_error(APLOG_MARK, APLOG_WARNING, 0, r->server,
1288 "proxy: Ignoring bogus HTTP header "
1289 "returned by %s (%s)", r->uri, r->method);
1295 /* XXX: RFC2068 defines only SP and HT as whitespace, this test is
1296 * wrong... and so are many others probably.
1298 while (apr_isspace(*value))
1299 ++value; /* Skip to start of value */
1301 /* should strip trailing whitespace as well */
1302 for (end = &value[strlen(value)-1]; end > value && apr_isspace(*end); --
1306 /* make sure we add so as not to destroy duplicated headers
1307 * Modify headers requiring canonicalisation and/or affected
1308 * by ProxyPassReverse and family with process_proxy_header
1310 process_proxy_header(r, dconf, buffer, value) ;
1313 /* the header was too long; at the least we should skip extra data */
1314 if (len >= size - 1) {
1315 while ((len = ap_getline(field, MAX_STRING_LEN, rr, 1))
1316 >= MAX_STRING_LEN - 1) {
1317 /* soak up the extra data */
1319 if (len == 0) /* time to exit the larger loop as well */
1327 static int addit_dammit(void *v, const char *key, const char *val)
1329 apr_table_addn(v, key, val);
1334 apr_status_t ap_proxygetline(apr_bucket_brigade *bb, char *s, int n, request_rec *r,
1335 int fold, int *writen)
1341 rv = ap_rgetline(&tmp_s, n, &len, r, fold, bb);
1342 apr_brigade_cleanup(bb);
1344 if (rv == APR_SUCCESS) {
1345 *writen = (int) len;
1346 } else if (rv == APR_ENOSPC) {
1356 * Limit the number of interim respones we sent back to the client. Otherwise
1357 * we suffer from a memory build up. Besides there is NO sense in sending back
1358 * an unlimited number of interim responses to the client. Thus if we cross
1359 * this limit send back a 502 (Bad Gateway).
1361 #ifndef AP_MAX_INTERIM_RESPONSES
1362 #define AP_MAX_INTERIM_RESPONSES 10
1366 apr_status_t ap_proxy_http_process_response(apr_pool_t * p, request_rec *r,
1367 proxy_conn_rec *backend,
1368 proxy_worker *worker,
1369 proxy_server_conf *conf,
1370 char *server_portstr) {
1371 conn_rec *c = r->connection;
1372 char buffer[HUGE_STRING_LEN];
1377 apr_bucket_brigade *bb, *tmp_bb;
1378 apr_bucket_brigade *pass_bb;
1379 int len, backasswards;
1380 int interim_response = 0; /* non-zero whilst interim 1xx responses
1381 * are being read. */
1383 apr_table_t *save_table;
1384 int backend_broke = 0;
1385 static const char *hop_by_hop_hdrs[] =
1386 {"Keep-Alive", "Proxy-Authenticate", "TE", "Trailer", "Upgrade", NULL};
1388 const char *te = NULL;
1389 int original_status = r->status;
1390 int proxy_status = OK;
1391 const char *original_status_line = r->status_line;
1392 const char *proxy_status_line = NULL;
1393 conn_rec *origin = backend->connection;
1394 apr_interval_time_t old_timeout = 0;
1396 int do_100_continue;
1398 do_100_continue = (worker->ping_timeout_set
1400 && (PROXYREQ_REVERSE == r->proxyreq)
1401 && !(apr_table_get(r->subprocess_env, "force-proxy-request-1.0")));
1404 bb = apr_brigade_create(p, c->bucket_alloc);
1405 pass_bb = apr_brigade_create(p, c->bucket_alloc);
1407 /* Setup for 100-Continue timeout if appropriate */
1408 if (do_100_continue) {
1409 apr_socket_timeout_get(backend->sock, &old_timeout);
1410 if (worker->ping_timeout != old_timeout) {
1412 rc = apr_socket_timeout_set(backend->sock, worker->ping_timeout);
1413 if (rc != APR_SUCCESS) {
1414 ap_log_error(APLOG_MARK, APLOG_ERR, rc, r->server,
1415 "proxy: could not set 100-Continue timeout");
1420 /* Get response from the remote server, and pass it up the
1424 rp = ap_proxy_make_fake_req(origin, r);
1425 /* In case anyone needs to know, this is a fake request that is really a
1428 rp->proxyreq = PROXYREQ_RESPONSE;
1429 tmp_bb = apr_brigade_create(p, c->bucket_alloc);
1433 apr_brigade_cleanup(bb);
1435 rc = ap_proxygetline(tmp_bb, buffer, sizeof(buffer), rp, 0, &len);
1437 /* handle one potential stray CRLF */
1438 rc = ap_proxygetline(tmp_bb, buffer, sizeof(buffer), rp, 0, &len);
1441 ap_log_rerror(APLOG_MARK, APLOG_ERR, rc, r,
1442 "proxy: error reading status line from remote "
1443 "server %s:%d", backend->hostname, backend->port);
1444 if (APR_STATUS_IS_TIMEUP(rc)) {
1445 ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r,
1446 "proxy: read timeout");
1447 if (do_100_continue) {
1448 return ap_proxyerror(r, HTTP_SERVICE_UNAVAILABLE, "Timeout on 100-Continue");
1452 * If we are a reverse proxy request shutdown the connection
1453 * WITHOUT ANY response to trigger a retry by the client
1454 * if allowed (as for idempotent requests).
1455 * BUT currently we should not do this if the request is the
1456 * first request on a keepalive connection as browsers like
1457 * seamonkey only display an empty page in this case and do
1458 * not do a retry. We should also not do this on a
1459 * connection which times out; instead handle as
1460 * we normally would handle timeouts
1462 if (r->proxyreq == PROXYREQ_REVERSE && c->keepalives &&
1463 !APR_STATUS_IS_TIMEUP(rc)) {
1466 ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r,
1467 "proxy: Closing connection to client because"
1468 " reading from backend server %s:%d failed."
1469 " Number of keepalives %i", backend->hostname,
1470 backend->port, c->keepalives);
1471 ap_proxy_backend_broke(r, bb);
1473 * Add an EOC bucket to signal the ap_http_header_filter
1474 * that it should get out of our way, BUT ensure that the
1475 * EOC bucket is inserted BEFORE an EOS bucket in bb as
1476 * some resource filters like mod_deflate pass everything
1477 * up to the EOS down the chain immediately and sent the
1478 * remainder of the brigade later (or even never). But in
1479 * this case the ap_http_header_filter does not get out of
1480 * our way soon enough.
1482 e = ap_bucket_eoc_create(c->bucket_alloc);
1483 eos = APR_BRIGADE_LAST(bb);
1484 while ((APR_BRIGADE_SENTINEL(bb) != eos)
1485 && !APR_BUCKET_IS_EOS(eos)) {
1486 eos = APR_BUCKET_PREV(eos);
1488 if (eos == APR_BRIGADE_SENTINEL(bb)) {
1489 APR_BRIGADE_INSERT_TAIL(bb, e);
1492 APR_BUCKET_INSERT_BEFORE(eos, e);
1494 ap_pass_brigade(r->output_filters, bb);
1495 /* Mark the backend connection for closing */
1497 /* Need to return OK to avoid sending an error message */
1500 else if (!c->keepalives) {
1501 ap_log_rerror(APLOG_MARK, APLOG_DEBUG, 0, r,
1502 "proxy: NOT Closing connection to client"
1503 " although reading from backend server %s:%d"
1504 " failed.", backend->hostname,
1507 return ap_proxyerror(r, HTTP_BAD_GATEWAY,
1508 "Error reading from remote server");
1510 /* XXX: Is this a real headers length send from remote? */
1511 backend->worker->s->read += len;
1513 /* Is it an HTTP/1 response?
1514 * This is buggy if we ever see an HTTP/1.10
1516 if (apr_date_checkmask(buffer, "HTTP/#.# ###*")) {
1519 if (2 != sscanf(buffer, "HTTP/%u.%u", &major, &minor)) {
1523 /* If not an HTTP/1 message or
1524 * if the status line was > 8192 bytes
1526 else if ((buffer[5] != '1') || (len >= sizeof(buffer)-1)) {
1527 return ap_proxyerror(r, HTTP_BAD_GATEWAY,
1528 apr_pstrcat(p, "Corrupt status line returned by remote "
1529 "server: ", buffer, NULL));
1533 keepchar = buffer[12];
1535 proxy_status = atoi(&buffer[9]);
1537 if (keepchar != '\0') {
1538 buffer[12] = keepchar;
1540 /* 2616 requires the space in Status-Line; the origin
1541 * server may have sent one but ap_rgetline_core will
1542 * have stripped it. */
1546 proxy_status_line = apr_pstrdup(p, &buffer[9]);
1548 /* The status out of the front is the same as the status coming in
1549 * from the back, until further notice.
1551 r->status = proxy_status;
1552 r->status_line = proxy_status_line;
1554 ap_log_rerror(APLOG_MARK, APLOG_TRACE3, 0, r,
1555 "Status from backend: %d", proxy_status);
1557 /* read the headers. */
1558 /* N.B. for HTTP/1.0 clients, we have to fold line-wrapped headers*/
1559 /* Also, take care with headers with multiple occurences. */
1561 /* First, tuck away all already existing cookies */
1562 save_table = apr_table_make(r->pool, 2);
1563 apr_table_do(addit_dammit, save_table, r->headers_out,
1564 "Set-Cookie", NULL);
1566 /* shove the headers direct into r->headers_out */
1567 ap_proxy_read_headers(r, rp, buffer, sizeof(buffer), origin,
1570 if (r->headers_out == NULL) {
1571 ap_log_error(APLOG_MARK, APLOG_WARNING, 0,
1572 r->server, "proxy: bad HTTP/%d.%d header "
1573 "returned by %s (%s)", major, minor, r->uri,
1575 backend->close += 1;
1577 * ap_send_error relies on a headers_out to be present. we
1578 * are in a bad position here.. so force everything we send out
1579 * to have nothing to do with the incoming packet
1581 r->headers_out = apr_table_make(r->pool,1);
1582 r->status = HTTP_BAD_GATEWAY;
1583 r->status_line = "bad gateway";
1587 /* Now, add in the just read cookies */
1588 apr_table_do(addit_dammit, save_table, r->headers_out,
1589 "Set-Cookie", NULL);
1591 /* and now load 'em all in */
1592 if (!apr_is_empty_table(save_table)) {
1593 apr_table_unset(r->headers_out, "Set-Cookie");
1594 r->headers_out = apr_table_overlay(r->pool,
1599 /* can't have both Content-Length and Transfer-Encoding */
1600 if (apr_table_get(r->headers_out, "Transfer-Encoding")
1601 && apr_table_get(r->headers_out, "Content-Length")) {
1603 * 2616 section 4.4, point 3: "if both Transfer-Encoding
1604 * and Content-Length are received, the latter MUST be
1607 * To help mitigate HTTP Splitting, unset Content-Length
1608 * and shut down the backend server connection
1609 * XXX: We aught to treat such a response as uncachable
1611 apr_table_unset(r->headers_out, "Content-Length");
1612 ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, r->server,
1613 "proxy: server %s:%d returned Transfer-Encoding"
1614 " and Content-Length", backend->hostname,
1616 backend->close += 1;
1620 * Save a possible Transfer-Encoding header as we need it later for
1621 * ap_http_filter to know where to end.
1623 te = apr_table_get(r->headers_out, "Transfer-Encoding");
1624 /* strip connection listed hop-by-hop headers from response */
1625 backend->close += ap_proxy_liststr(apr_table_get(r->headers_out,
1628 ap_proxy_clear_connection(p, r->headers_out);
1629 if ((buf = apr_table_get(r->headers_out, "Content-Type"))) {
1630 ap_set_content_type(r, apr_pstrdup(p, buf));
1632 if (!ap_is_HTTP_INFO(proxy_status)) {
1633 ap_proxy_pre_http_request(origin, rp);
1636 /* Clear hop-by-hop headers */
1637 for (i=0; hop_by_hop_hdrs[i]; ++i) {
1638 apr_table_unset(r->headers_out, hop_by_hop_hdrs[i]);
1640 /* Delete warnings with wrong date */
1641 r->headers_out = ap_proxy_clean_warnings(p, r->headers_out);
1643 /* handle Via header in response */
1644 if (conf->viaopt != via_off && conf->viaopt != via_block) {
1645 const char *server_name = ap_get_server_name(r);
1646 /* If USE_CANONICAL_NAME_OFF was configured for the proxy virtual host,
1647 * then the server name returned by ap_get_server_name() is the
1648 * origin server name (which does make too much sense with Via: headers)
1649 * so we use the proxy vhost's name instead.
1651 if (server_name == r->hostname)
1652 server_name = r->server->server_hostname;
1653 /* create a "Via:" response header entry and merge it */
1654 apr_table_addn(r->headers_out, "Via",
1655 (conf->viaopt == via_full)
1656 ? apr_psprintf(p, "%d.%d %s%s (%s)",
1657 HTTP_VERSION_MAJOR(r->proto_num),
1658 HTTP_VERSION_MINOR(r->proto_num),
1661 AP_SERVER_BASEVERSION)
1662 : apr_psprintf(p, "%d.%d %s%s",
1663 HTTP_VERSION_MAJOR(r->proto_num),
1664 HTTP_VERSION_MINOR(r->proto_num),
1670 /* cancel keepalive if HTTP/1.0 or less */
1671 if ((major < 1) || (minor < 1)) {
1672 backend->close += 1;
1673 origin->keepalive = AP_CONN_CLOSE;
1676 /* an http/0.9 response */
1679 r->status_line = "200 OK";
1680 backend->close += 1;
1683 if (ap_is_HTTP_INFO(proxy_status)) {
1685 /* Reset to old timeout iff we've adjusted it */
1687 && (r->status == HTTP_CONTINUE)
1688 && (worker->ping_timeout != old_timeout)) {
1689 apr_socket_timeout_set(backend->sock, old_timeout);
1693 interim_response = 0;
1695 if (interim_response) {
1696 /* RFC2616 tells us to forward this.
1698 * OTOH, an interim response here may mean the backend
1699 * is playing sillybuggers. The Client didn't ask for
1700 * it within the defined HTTP/1.1 mechanisms, and if
1701 * it's an extension, it may also be unsupported by us.
1703 * There's also the possibility that changing existing
1704 * behaviour here might break something.
1706 * So let's make it configurable.
1708 const char *policy = apr_table_get(r->subprocess_env,
1709 "proxy-interim-response");
1710 ap_log_rerror(APLOG_MARK, APLOG_TRACE2, 0, r,
1711 "proxy: HTTP: received interim %d response",
1713 if (!policy || !strcasecmp(policy, "RFC")) {
1714 ap_send_interim_response(r, 1);
1716 /* FIXME: refine this to be able to specify per-response-status
1717 * policies and maybe also add option to bail out with 502
1719 else if (strcasecmp(policy, "Suppress")) {
1720 ap_log_rerror(APLOG_MARK, APLOG_WARNING, 0, r,
1721 "undefined proxy interim response policy");
1724 /* Moved the fixups of Date headers and those affected by
1725 * ProxyPassReverse/etc from here to ap_proxy_read_headers
1728 if ((proxy_status == 401) && (conf->error_override)) {
1730 const char *wa = "WWW-Authenticate";
1731 if ((buf = apr_table_get(r->headers_out, wa))) {
1732 apr_table_set(r->err_headers_out, wa, buf);
1734 ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, r->server,
1735 "proxy: origin server sent 401 without WWW-Authenticate header");
1741 * Is it an HTTP/0.9 response or did we maybe preread the 1st line of
1742 * the response? If so, load the extra data. These are 2 mutually
1743 * exclusive possibilities, that just happen to require very
1746 if (backasswards || pread_len) {
1747 apr_ssize_t cntr = (apr_ssize_t)pread_len;
1750 * At this point in response processing of a 0.9 response,
1751 * we don't know yet whether data is binary or not.
1752 * mod_charset_lite will get control later on, so it cannot
1753 * decide on the conversion of this buffer full of data.
1754 * However, chances are that we are not really talking to an
1755 * HTTP/0.9 server, but to some different protocol, therefore
1756 * the best guess IMHO is to always treat the buffer as "text/x":
1758 ap_xlate_proto_to_ascii(buffer, len);
1759 cntr = (apr_ssize_t)len;
1761 e = apr_bucket_heap_create(buffer, cntr, NULL, c->bucket_alloc);
1762 APR_BRIGADE_INSERT_TAIL(bb, e);
1764 /* PR 41646: get HEAD right with ProxyErrorOverride */
1765 if (ap_is_HTTP_ERROR(r->status) && conf->error_override) {
1766 /* clear r->status for override error, otherwise ErrorDocument
1767 * thinks that this is a recursive error, and doesn't find the
1770 r->status = HTTP_OK;
1771 /* Discard body, if one is expected */
1772 if (!r->header_only && /* not HEAD request */
1773 (proxy_status != HTTP_NO_CONTENT) && /* not 204 */
1774 (proxy_status != HTTP_NOT_MODIFIED)) { /* not 304 */
1775 ap_discard_request_body(rp);
1777 return proxy_status;
1780 /* send body - but only if a body is expected */
1781 if ((!r->header_only) && /* not HEAD request */
1782 !interim_response && /* not any 1xx response */
1783 (proxy_status != HTTP_NO_CONTENT) && /* not 204 */
1784 (proxy_status != HTTP_NOT_MODIFIED)) { /* not 304 */
1786 /* We need to copy the output headers and treat them as input
1787 * headers as well. BUT, we need to do this before we remove
1788 * TE, so that they are preserved accordingly for
1789 * ap_http_filter to know where to end.
1791 rp->headers_in = apr_table_copy(r->pool, r->headers_out);
1793 * Restore Transfer-Encoding header from response if we saved
1794 * one before and there is none left. We need it for the
1795 * ap_http_filter. See above.
1797 if (te && !apr_table_get(rp->headers_in, "Transfer-Encoding")) {
1798 apr_table_add(rp->headers_in, "Transfer-Encoding", te);
1801 apr_table_unset(r->headers_out,"Transfer-Encoding");
1803 ap_log_error(APLOG_MARK, APLOG_TRACE3, 0, r->server,
1804 "proxy: start body send");
1807 * if we are overriding the errors, we can't put the content
1808 * of the page into the brigade
1810 if (!conf->error_override || !ap_is_HTTP_ERROR(proxy_status)) {
1811 /* read the body, pass it to the output filters */
1812 apr_read_type_e mode = APR_NONBLOCK_READ;
1815 /* Handle the case where the error document is itself reverse
1816 * proxied and was successful. We must maintain any previous
1817 * error status so that an underlying error (eg HTTP_NOT_FOUND)
1818 * doesn't become an HTTP_OK.
1820 if (conf->error_override && !ap_is_HTTP_ERROR(proxy_status)
1821 && ap_is_HTTP_ERROR(original_status)) {
1822 r->status = original_status;
1823 r->status_line = original_status_line;
1827 apr_off_t readbytes;
1830 rv = ap_get_brigade(rp->input_filters, bb,
1831 AP_MODE_READBYTES, mode,
1832 conf->io_buffer_size);
1834 /* ap_get_brigade will return success with an empty brigade
1835 * for a non-blocking read which would block: */
1836 if (APR_STATUS_IS_EAGAIN(rv)
1837 || (rv == APR_SUCCESS && APR_BRIGADE_EMPTY(bb))) {
1838 /* flush to the client and switch to blocking mode */
1839 e = apr_bucket_flush_create(c->bucket_alloc);
1840 APR_BRIGADE_INSERT_TAIL(bb, e);
1841 if (ap_pass_brigade(r->output_filters, bb)
1846 apr_brigade_cleanup(bb);
1847 mode = APR_BLOCK_READ;
1850 else if (rv == APR_EOF) {
1853 else if (rv != APR_SUCCESS) {
1854 /* In this case, we are in real trouble because
1855 * our backend bailed on us. Pass along a 502 error
1858 ap_log_cerror(APLOG_MARK, APLOG_ERR, rv, c,
1859 "proxy: error reading response");
1860 ap_proxy_backend_broke(r, bb);
1861 ap_pass_brigade(r->output_filters, bb);
1866 /* next time try a non-blocking read */
1867 mode = APR_NONBLOCK_READ;
1869 apr_brigade_length(bb, 0, &readbytes);
1870 backend->worker->s->read += readbytes;
1873 ap_log_error(APLOG_MARK, APLOG_DEBUG, 0,
1874 r->server, "proxy (PID %d): readbytes: %#x",
1875 getpid(), readbytes);
1879 if (APR_BRIGADE_EMPTY(bb)) {
1880 apr_brigade_cleanup(bb);
1884 /* Switch the allocator lifetime of the buckets */
1885 ap_proxy_buckets_lifetime_transform(r, bb, pass_bb);
1887 /* found the last brigade? */
1888 if (APR_BUCKET_IS_EOS(APR_BRIGADE_LAST(bb))) {
1889 /* signal that we must leave */
1893 /* try send what we read */
1894 if (ap_pass_brigade(r->output_filters, pass_bb) != APR_SUCCESS
1896 /* Ack! Phbtt! Die! User aborted! */
1897 backend->close = 1; /* this causes socket close below */
1901 /* make sure we always clean up after ourselves */
1902 apr_brigade_cleanup(bb);
1903 apr_brigade_cleanup(pass_bb);
1907 ap_log_error(APLOG_MARK, APLOG_TRACE2, 0, r->server,
1908 "proxy: end body send");
1910 else if (!interim_response) {
1911 ap_log_error(APLOG_MARK, APLOG_TRACE2, 0, r->server,
1912 "proxy: header only");
1914 /* Pass EOS bucket down the filter chain. */
1915 e = apr_bucket_eos_create(c->bucket_alloc);
1916 APR_BRIGADE_INSERT_TAIL(bb, e);
1917 if (ap_pass_brigade(r->output_filters, bb) != APR_SUCCESS
1919 /* Ack! Phbtt! Die! User aborted! */
1920 backend->close = 1; /* this causes socket close below */
1923 apr_brigade_cleanup(bb);
1925 } while (interim_response && (interim_response < AP_MAX_INTERIM_RESPONSES));
1927 /* See define of AP_MAX_INTERIM_RESPONSES for why */
1928 if (interim_response >= AP_MAX_INTERIM_RESPONSES) {
1929 return ap_proxyerror(r, HTTP_BAD_GATEWAY,
1931 "Too many (%d) interim responses from origin server",
1935 /* If our connection with the client is to be aborted, return DONE. */
1936 if (c->aborted || backend_broke) {
1944 apr_status_t ap_proxy_http_cleanup(const char *scheme, request_rec *r,
1945 proxy_conn_rec *backend)
1947 ap_proxy_release_connection(scheme, backend, r->server);
1952 * This handles http:// URLs, and other URLs using a remote proxy over http
1953 * If proxyhost is NULL, then contact the server directly, otherwise
1955 * Note that if a proxy is used, then URLs other than http: can be accessed,
1956 * also, if we have trouble which is clearly specific to the proxy, then
1957 * we return DECLINED so that we can try another proxy. (Or the direct
1960 static int proxy_http_handler(request_rec *r, proxy_worker *worker,
1961 proxy_server_conf *conf,
1962 char *url, const char *proxyname,
1963 apr_port_t proxyport)
1966 char server_portstr[32];
1968 const char *proxy_function;
1970 proxy_conn_rec *backend = NULL;
1972 conn_rec *c = r->connection;
1975 * Use a shorter-lived pool to reduce memory usage
1976 * and avoid a memory leak
1978 apr_pool_t *p = r->pool;
1979 apr_uri_t *uri = apr_palloc(p, sizeof(*uri));
1981 /* find the scheme */
1982 u = strchr(url, ':');
1983 if (u == NULL || u[1] != '/' || u[2] != '/' || u[3] == '\0')
1986 return HTTP_BAD_REQUEST;
1987 scheme = apr_pstrndup(p, url, u - url);
1988 /* scheme is lowercase */
1989 ap_str_tolower(scheme);
1991 if (strcmp(scheme, "https") == 0) {
1992 if (!ap_proxy_ssl_enable(NULL)) {
1993 ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, r->server,
1994 "proxy: HTTPS: declining URL %s"
1995 " (mod_ssl not configured?)", url);
1999 proxy_function = "HTTPS";
2001 else if (!(strcmp(scheme, "http") == 0 || (strcmp(scheme, "ftp") == 0 && proxyname))) {
2002 ap_log_error(APLOG_MARK, APLOG_DEBUG, 0, r->server,
2003 "proxy: HTTP: declining URL %s", url);
2004 return DECLINED; /* only interested in HTTP, or FTP via proxy */
2008 proxy_function = "HTTP";
2010 proxy_function = "FTP";
2012 ap_log_error(APLOG_MARK, APLOG_TRACE1, 0, r->server,
2013 "proxy: HTTP: serving URL %s", url);
2016 /* create space for state information */
2017 if ((status = ap_proxy_acquire_connection(proxy_function, &backend,
2018 worker, r->server)) != OK)
2022 backend->is_ssl = is_ssl;
2025 ap_proxy_ssl_connection_cleanup(backend, r);
2029 * In the case that we are handling a reverse proxy connection and this
2030 * is not a request that is coming over an already kept alive connection
2031 * with the client, do NOT reuse the connection to the backend, because
2032 * we cannot forward a failure to the client in this case as the client
2033 * does NOT expect this in this situation.
2034 * Yes, this creates a performance penalty.
2036 if ((r->proxyreq == PROXYREQ_REVERSE) && (!c->keepalives)
2037 && (apr_table_get(r->subprocess_env, "proxy-initial-not-pooled"))) {
2044 /* Step One: Determine Who To Connect To */
2045 if ((status = ap_proxy_determine_connection(p, r, conf, worker, backend,
2046 uri, &locurl, proxyname,
2047 proxyport, server_portstr,
2048 sizeof(server_portstr))) != OK)
2051 /* Step Two: Make the Connection */
2052 if (ap_proxy_connect_backend(proxy_function, backend, worker, r->server)) {
2053 ap_log_error(APLOG_MARK, APLOG_ERR, 0, r->server,
2054 "proxy: HTTP: failed to make connection to backend: %s",
2056 status = HTTP_SERVICE_UNAVAILABLE;
2060 /* Step Three: Create conn_rec */
2061 if (!backend->connection) {
2062 if ((status = ap_proxy_connection_create(proxy_function, backend,
2063 c, r->server)) != OK)
2066 * On SSL connections set a note on the connection what CN is
2067 * requested, such that mod_ssl can check if it is requested to do
2071 apr_table_set(backend->connection->notes, "proxy-request-hostname",
2076 /* Step Four: Send the Request
2077 * On the off-chance that we forced a 100-Continue as a
2078 * kinda HTTP ping test, allow for retries
2080 if ((status = ap_proxy_http_request(p, r, backend, worker,
2081 conf, uri, locurl, server_portstr)) != OK) {
2082 if ((status == HTTP_SERVICE_UNAVAILABLE) && worker->ping_timeout_set) {
2084 ap_log_error(APLOG_MARK, APLOG_INFO, status, r->server,
2085 "proxy: HTTP: 100-Continue failed to %pI (%s)",
2086 worker->cp->addr, worker->hostname);
2095 /* Step Five: Receive the Response... Fall thru to cleanup */
2096 status = ap_proxy_http_process_response(p, r, backend, worker,
2097 conf, server_portstr);
2102 /* Step Six: Clean Up */
2107 ap_proxy_http_cleanup(proxy_function, r, backend);
2111 static apr_status_t warn_rx_free(void *p)
2113 ap_pregfree((apr_pool_t*)p, warn_rx);
2116 static void ap_proxy_http_register_hook(apr_pool_t *p)
2118 proxy_hook_scheme_handler(proxy_http_handler, NULL, NULL, APR_HOOK_FIRST);
2119 proxy_hook_canon_handler(proxy_http_canon, NULL, NULL, APR_HOOK_FIRST);
2120 warn_rx = ap_pregcomp(p, "[0-9]{3}[ \t]+[^ \t]+[ \t]+\"[^\"]*\"([ \t]+\"([^\"]+)\")?", 0);
2121 apr_pool_cleanup_register(p, p, warn_rx_free, apr_pool_cleanup_null);
2124 AP_DECLARE_MODULE(proxy_http) = {
2125 STANDARD20_MODULE_STUFF,
2126 NULL, /* create per-directory config structure */
2127 NULL, /* merge per-directory config structures */
2128 NULL, /* create per-server config structure */
2129 NULL, /* merge per-server config structures */
2130 NULL, /* command apr_table_t */
2131 ap_proxy_http_register_hook/* register hooks */