1 /*
2 * nghttp2 - HTTP/2 C Library
3 *
4 * Copyright (c) 2012 Tatsuhiro Tsujikawa
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining
7 * a copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sublicense, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be
15 * included in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
18 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
20 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
21 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
22 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
23 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
24 */
25 #include "shrpx_client_handler.h"
26
27 #ifdef HAVE_UNISTD_H
28 # include <unistd.h>
29 #endif // HAVE_UNISTD_H
30 #ifdef HAVE_SYS_SOCKET_H
31 # include <sys/socket.h>
32 #endif // HAVE_SYS_SOCKET_H
33 #ifdef HAVE_NETDB_H
34 # include <netdb.h>
35 #endif // HAVE_NETDB_H
36
37 #include <cerrno>
38
39 #include "shrpx_upstream.h"
40 #include "shrpx_http2_upstream.h"
41 #include "shrpx_https_upstream.h"
42 #include "shrpx_config.h"
43 #include "shrpx_http_downstream_connection.h"
44 #include "shrpx_http2_downstream_connection.h"
45 #include "shrpx_tls.h"
46 #include "shrpx_worker.h"
47 #include "shrpx_downstream_connection_pool.h"
48 #include "shrpx_downstream.h"
49 #include "shrpx_http2_session.h"
50 #include "shrpx_connect_blocker.h"
51 #include "shrpx_api_downstream_connection.h"
52 #include "shrpx_health_monitor_downstream_connection.h"
53 #include "shrpx_null_downstream_connection.h"
54 #ifdef ENABLE_HTTP3
55 # include "shrpx_http3_upstream.h"
56 #endif // ENABLE_HTTP3
57 #include "shrpx_log.h"
58 #include "util.h"
59 #include "template.h"
60 #include "tls.h"
61
62 using namespace nghttp2;
63
64 namespace shrpx {
65
66 namespace {
timeoutcb(struct ev_loop * loop,ev_timer * w,int revents)67 void timeoutcb(struct ev_loop *loop, ev_timer *w, int revents) {
68 auto conn = static_cast<Connection *>(w->data);
69 auto handler = static_cast<ClientHandler *>(conn->data);
70
71 if (LOG_ENABLED(INFO)) {
72 CLOG(INFO, handler) << "Time out";
73 }
74
75 delete handler;
76 }
77 } // namespace
78
79 namespace {
shutdowncb(struct ev_loop * loop,ev_timer * w,int revents)80 void shutdowncb(struct ev_loop *loop, ev_timer *w, int revents) {
81 auto handler = static_cast<ClientHandler *>(w->data);
82
83 if (LOG_ENABLED(INFO)) {
84 CLOG(INFO, handler) << "Close connection due to TLS renegotiation";
85 }
86
87 delete handler;
88 }
89 } // namespace
90
91 namespace {
readcb(struct ev_loop * loop,ev_io * w,int revents)92 void readcb(struct ev_loop *loop, ev_io *w, int revents) {
93 auto conn = static_cast<Connection *>(w->data);
94 auto handler = static_cast<ClientHandler *>(conn->data);
95
96 if (handler->do_read() != 0) {
97 delete handler;
98 return;
99 }
100 }
101 } // namespace
102
103 namespace {
writecb(struct ev_loop * loop,ev_io * w,int revents)104 void writecb(struct ev_loop *loop, ev_io *w, int revents) {
105 auto conn = static_cast<Connection *>(w->data);
106 auto handler = static_cast<ClientHandler *>(conn->data);
107
108 if (handler->do_write() != 0) {
109 delete handler;
110 return;
111 }
112 }
113 } // namespace
114
noop()115 int ClientHandler::noop() { return 0; }
116
read_clear()117 int ClientHandler::read_clear() {
118 auto should_break = false;
119 rb_.ensure_chunk();
120 for (;;) {
121 if (rb_.rleft() && on_read() != 0) {
122 return -1;
123 }
124 if (rb_.rleft() == 0) {
125 rb_.reset();
126 } else if (rb_.wleft() == 0) {
127 conn_.rlimit.stopw();
128 return 0;
129 }
130
131 if (!ev_is_active(&conn_.rev) || should_break) {
132 return 0;
133 }
134
135 auto nread = conn_.read_clear(rb_.last(), rb_.wleft());
136
137 if (nread == 0) {
138 if (rb_.rleft() == 0) {
139 rb_.release_chunk();
140 }
141 return 0;
142 }
143
144 if (nread < 0) {
145 return -1;
146 }
147
148 rb_.write(nread);
149 should_break = true;
150 }
151 }
152
write_clear()153 int ClientHandler::write_clear() {
154 std::array<iovec, 2> iov;
155
156 for (;;) {
157 if (on_write() != 0) {
158 return -1;
159 }
160
161 auto iovcnt = upstream_->response_riovec(iov.data(), iov.size());
162 if (iovcnt == 0) {
163 break;
164 }
165
166 auto nwrite = conn_.writev_clear(iov.data(), iovcnt);
167 if (nwrite < 0) {
168 return -1;
169 }
170
171 if (nwrite == 0) {
172 return 0;
173 }
174
175 upstream_->response_drain(nwrite);
176 }
177
178 conn_.wlimit.stopw();
179 ev_timer_stop(conn_.loop, &conn_.wt);
180
181 return 0;
182 }
183
proxy_protocol_peek_clear()184 int ClientHandler::proxy_protocol_peek_clear() {
185 rb_.ensure_chunk();
186
187 assert(rb_.rleft() == 0);
188
189 auto nread = conn_.peek_clear(rb_.last(), rb_.wleft());
190 if (nread < 0) {
191 return -1;
192 }
193 if (nread == 0) {
194 return 0;
195 }
196
197 if (LOG_ENABLED(INFO)) {
198 CLOG(INFO, this) << "PROXY-protocol: Peek " << nread
199 << " bytes from socket";
200 }
201
202 rb_.write(nread);
203
204 if (on_read() != 0) {
205 return -1;
206 }
207
208 rb_.reset();
209
210 return 0;
211 }
212
tls_handshake()213 int ClientHandler::tls_handshake() {
214 ev_timer_again(conn_.loop, &conn_.rt);
215
216 ERR_clear_error();
217
218 auto rv = conn_.tls_handshake();
219
220 if (rv == SHRPX_ERR_INPROGRESS) {
221 return 0;
222 }
223
224 if (rv < 0) {
225 return -1;
226 }
227
228 if (LOG_ENABLED(INFO)) {
229 CLOG(INFO, this) << "SSL/TLS handshake completed";
230 }
231
232 if (validate_next_proto() != 0) {
233 return -1;
234 }
235
236 read_ = &ClientHandler::read_tls;
237 write_ = &ClientHandler::write_tls;
238
239 return 0;
240 }
241
read_tls()242 int ClientHandler::read_tls() {
243 auto should_break = false;
244
245 ERR_clear_error();
246
247 rb_.ensure_chunk();
248
249 for (;;) {
250 // we should process buffered data first before we read EOF.
251 if (rb_.rleft() && on_read() != 0) {
252 return -1;
253 }
254 if (rb_.rleft() == 0) {
255 rb_.reset();
256 } else if (rb_.wleft() == 0) {
257 conn_.rlimit.stopw();
258 return 0;
259 }
260
261 if (!ev_is_active(&conn_.rev) || should_break) {
262 return 0;
263 }
264
265 auto nread = conn_.read_tls(rb_.last(), rb_.wleft());
266
267 if (nread == 0) {
268 if (rb_.rleft() == 0) {
269 rb_.release_chunk();
270 }
271 return 0;
272 }
273
274 if (nread < 0) {
275 return -1;
276 }
277
278 rb_.write(nread);
279 should_break = true;
280 }
281 }
282
write_tls()283 int ClientHandler::write_tls() {
284 struct iovec iov;
285
286 ERR_clear_error();
287
288 if (on_write() != 0) {
289 return -1;
290 }
291
292 auto iovcnt = upstream_->response_riovec(&iov, 1);
293 if (iovcnt == 0) {
294 conn_.start_tls_write_idle();
295
296 conn_.wlimit.stopw();
297 ev_timer_stop(conn_.loop, &conn_.wt);
298
299 return 0;
300 }
301
302 for (;;) {
303 auto nwrite = conn_.write_tls(iov.iov_base, iov.iov_len);
304 if (nwrite < 0) {
305 return -1;
306 }
307
308 if (nwrite == 0) {
309 return 0;
310 }
311
312 upstream_->response_drain(nwrite);
313
314 iovcnt = upstream_->response_riovec(&iov, 1);
315 if (iovcnt == 0) {
316 return 0;
317 }
318 }
319 }
320
321 #ifdef ENABLE_HTTP3
read_quic(const UpstreamAddr * faddr,const Address & remote_addr,const Address & local_addr,const ngtcp2_pkt_info & pi,std::span<const uint8_t> data)322 int ClientHandler::read_quic(const UpstreamAddr *faddr,
323 const Address &remote_addr,
324 const Address &local_addr,
325 const ngtcp2_pkt_info &pi,
326 std::span<const uint8_t> data) {
327 auto upstream = static_cast<Http3Upstream *>(upstream_.get());
328
329 return upstream->on_read(faddr, remote_addr, local_addr, pi, data);
330 }
331
write_quic()332 int ClientHandler::write_quic() { return upstream_->on_write(); }
333 #endif // ENABLE_HTTP3
334
upstream_noop()335 int ClientHandler::upstream_noop() { return 0; }
336
upstream_read()337 int ClientHandler::upstream_read() {
338 assert(upstream_);
339 if (upstream_->on_read() != 0) {
340 return -1;
341 }
342 return 0;
343 }
344
upstream_write()345 int ClientHandler::upstream_write() {
346 assert(upstream_);
347 if (upstream_->on_write() != 0) {
348 return -1;
349 }
350
351 if (get_should_close_after_write() && upstream_->response_empty()) {
352 return -1;
353 }
354
355 return 0;
356 }
357
upstream_http2_connhd_read()358 int ClientHandler::upstream_http2_connhd_read() {
359 auto nread = std::min(left_connhd_len_, rb_.rleft());
360 if (memcmp(&NGHTTP2_CLIENT_MAGIC[NGHTTP2_CLIENT_MAGIC_LEN - left_connhd_len_],
361 rb_.pos(), nread) != 0) {
362 // There is no downgrade path here. Just drop the connection.
363 if (LOG_ENABLED(INFO)) {
364 CLOG(INFO, this) << "invalid client connection header";
365 }
366
367 return -1;
368 }
369
370 left_connhd_len_ -= nread;
371 rb_.drain(nread);
372 conn_.rlimit.startw();
373
374 if (left_connhd_len_ == 0) {
375 on_read_ = &ClientHandler::upstream_read;
376 // Run on_read to process data left in buffer since they are not
377 // notified further
378 if (on_read() != 0) {
379 return -1;
380 }
381 return 0;
382 }
383
384 return 0;
385 }
386
upstream_http1_connhd_read()387 int ClientHandler::upstream_http1_connhd_read() {
388 auto nread = std::min(left_connhd_len_, rb_.rleft());
389 if (memcmp(&NGHTTP2_CLIENT_MAGIC[NGHTTP2_CLIENT_MAGIC_LEN - left_connhd_len_],
390 rb_.pos(), nread) != 0) {
391 if (LOG_ENABLED(INFO)) {
392 CLOG(INFO, this) << "This is HTTP/1.1 connection, "
393 << "but may be upgraded to HTTP/2 later.";
394 }
395
396 // Reset header length for later HTTP/2 upgrade
397 left_connhd_len_ = NGHTTP2_CLIENT_MAGIC_LEN;
398 on_read_ = &ClientHandler::upstream_read;
399 on_write_ = &ClientHandler::upstream_write;
400
401 if (on_read() != 0) {
402 return -1;
403 }
404
405 return 0;
406 }
407
408 left_connhd_len_ -= nread;
409 rb_.drain(nread);
410 conn_.rlimit.startw();
411
412 if (left_connhd_len_ == 0) {
413 if (LOG_ENABLED(INFO)) {
414 CLOG(INFO, this) << "direct HTTP/2 connection";
415 }
416
417 direct_http2_upgrade();
418 on_read_ = &ClientHandler::upstream_read;
419 on_write_ = &ClientHandler::upstream_write;
420
421 // Run on_read to process data left in buffer since they are not
422 // notified further
423 if (on_read() != 0) {
424 return -1;
425 }
426
427 return 0;
428 }
429
430 return 0;
431 }
432
ClientHandler(Worker * worker,int fd,SSL * ssl,const StringRef & ipaddr,const StringRef & port,int family,const UpstreamAddr * faddr)433 ClientHandler::ClientHandler(Worker *worker, int fd, SSL *ssl,
434 const StringRef &ipaddr, const StringRef &port,
435 int family, const UpstreamAddr *faddr)
436 : // We use balloc_ for TLS session ID (64), ipaddr (IPv6) (39),
437 // port (5), forwarded-for (IPv6) (41), alpn (5), proxyproto
438 // ipaddr (15), proxyproto port (5), sni (32, estimated). we
439 // need terminal NULL byte for each. We also require 8 bytes
440 // header for each allocation. We align at 16 bytes boundary,
441 // so the required space is 64 + 48 + 16 + 48 + 16 + 16 + 16 +
442 // 32 + 8 + 8 * 8 = 328.
443 balloc_(512, 512),
444 rb_(worker->get_mcpool()),
445 conn_(worker->get_loop(), fd, ssl, worker->get_mcpool(),
446 get_config()->conn.upstream.timeout.write,
447 get_config()->conn.upstream.timeout.idle,
448 get_config()->conn.upstream.ratelimit.write,
449 get_config()->conn.upstream.ratelimit.read, writecb, readcb,
450 timeoutcb, this, get_config()->tls.dyn_rec.warmup_threshold,
451 get_config()->tls.dyn_rec.idle_timeout,
452 faddr->quic ? Proto::HTTP3 : Proto::NONE),
453 ipaddr_(make_string_ref(balloc_, ipaddr)),
454 port_(make_string_ref(balloc_, port)),
455 faddr_(faddr),
456 worker_(worker),
457 left_connhd_len_(NGHTTP2_CLIENT_MAGIC_LEN),
458 affinity_hash_(0),
459 should_close_after_write_(false),
460 affinity_hash_computed_(false) {
461
462 ++worker_->get_worker_stat()->num_connections;
463
464 ev_timer_init(&reneg_shutdown_timer_, shutdowncb, 0., 0.);
465
466 reneg_shutdown_timer_.data = this;
467
468 if (!faddr->quic) {
469 conn_.rlimit.startw();
470 }
471 ev_timer_again(conn_.loop, &conn_.rt);
472
473 auto config = get_config();
474
475 if (!faddr->quic) {
476 if (faddr_->accept_proxy_protocol ||
477 config->conn.upstream.accept_proxy_protocol) {
478 read_ = &ClientHandler::proxy_protocol_peek_clear;
479 write_ = &ClientHandler::noop;
480 on_read_ = &ClientHandler::proxy_protocol_read;
481 on_write_ = &ClientHandler::upstream_noop;
482 } else {
483 setup_upstream_io_callback();
484 }
485 }
486
487 auto &fwdconf = config->http.forwarded;
488
489 if (fwdconf.params & FORWARDED_FOR) {
490 if (fwdconf.for_node_type == ForwardedNode::OBFUSCATED) {
491 // 1 for '_'
492 auto len = SHRPX_OBFUSCATED_NODE_LENGTH + 1;
493 // 1 for terminating NUL.
494 auto buf = make_byte_ref(balloc_, len + 1);
495 auto p = std::begin(buf);
496 *p++ = '_';
497 p = util::random_alpha_digit(p, p + SHRPX_OBFUSCATED_NODE_LENGTH,
498 worker_->get_randgen());
499 *p = '\0';
500
501 forwarded_for_ = StringRef{std::span{std::begin(buf), p}};
502 } else {
503 init_forwarded_for(family, ipaddr_);
504 }
505 }
506 }
507
init_forwarded_for(int family,const StringRef & ipaddr)508 void ClientHandler::init_forwarded_for(int family, const StringRef &ipaddr) {
509 if (family == AF_INET6) {
510 // 2 for '[' and ']'
511 auto len = 2 + ipaddr.size();
512 // 1 for terminating NUL.
513 auto buf = make_byte_ref(balloc_, len + 1);
514 auto p = std::begin(buf);
515 *p++ = '[';
516 p = std::copy(std::begin(ipaddr), std::end(ipaddr), p);
517 *p++ = ']';
518 *p = '\0';
519
520 forwarded_for_ = StringRef{std::span{std::begin(buf), p}};
521 } else {
522 // family == AF_INET or family == AF_UNIX
523 forwarded_for_ = ipaddr;
524 }
525 }
526
setup_upstream_io_callback()527 void ClientHandler::setup_upstream_io_callback() {
528 if (conn_.tls.ssl) {
529 conn_.prepare_server_handshake();
530 read_ = write_ = &ClientHandler::tls_handshake;
531 on_read_ = &ClientHandler::upstream_noop;
532 on_write_ = &ClientHandler::upstream_write;
533 } else {
534 // For non-TLS version, first create HttpsUpstream. It may be
535 // upgraded to HTTP/2 through HTTP Upgrade or direct HTTP/2
536 // connection.
537 upstream_ = std::make_unique<HttpsUpstream>(this);
538 alpn_ = "http/1.1"_sr;
539 read_ = &ClientHandler::read_clear;
540 write_ = &ClientHandler::write_clear;
541 on_read_ = &ClientHandler::upstream_http1_connhd_read;
542 on_write_ = &ClientHandler::upstream_noop;
543 }
544 }
545
546 #ifdef ENABLE_HTTP3
setup_http3_upstream(std::unique_ptr<Http3Upstream> && upstream)547 void ClientHandler::setup_http3_upstream(
548 std::unique_ptr<Http3Upstream> &&upstream) {
549 upstream_ = std::move(upstream);
550 write_ = &ClientHandler::write_quic;
551
552 auto config = get_config();
553
554 reset_upstream_read_timeout(config->conn.upstream.timeout.http3_idle);
555 }
556 #endif // ENABLE_HTTP3
557
~ClientHandler()558 ClientHandler::~ClientHandler() {
559 if (LOG_ENABLED(INFO)) {
560 CLOG(INFO, this) << "Deleting";
561 }
562
563 if (upstream_) {
564 upstream_->on_handler_delete();
565 }
566
567 auto worker_stat = worker_->get_worker_stat();
568 --worker_stat->num_connections;
569
570 if (worker_stat->num_connections == 0) {
571 worker_->schedule_clear_mcpool();
572 }
573
574 ev_timer_stop(conn_.loop, &reneg_shutdown_timer_);
575
576 // TODO If backend is http/2, and it is in CONNECTED state, signal
577 // it and make it loopbreak when output is zero.
578 if (worker_->get_graceful_shutdown() && worker_stat->num_connections == 0 &&
579 worker_stat->num_close_waits == 0) {
580 ev_break(conn_.loop);
581 }
582
583 if (LOG_ENABLED(INFO)) {
584 CLOG(INFO, this) << "Deleted";
585 }
586 }
587
get_upstream()588 Upstream *ClientHandler::get_upstream() { return upstream_.get(); }
589
get_loop() const590 struct ev_loop *ClientHandler::get_loop() const { return conn_.loop; }
591
reset_upstream_read_timeout(ev_tstamp t)592 void ClientHandler::reset_upstream_read_timeout(ev_tstamp t) {
593 conn_.rt.repeat = t;
594
595 ev_timer_again(conn_.loop, &conn_.rt);
596 }
597
reset_upstream_write_timeout(ev_tstamp t)598 void ClientHandler::reset_upstream_write_timeout(ev_tstamp t) {
599 conn_.wt.repeat = t;
600
601 ev_timer_again(conn_.loop, &conn_.wt);
602 }
603
repeat_read_timer()604 void ClientHandler::repeat_read_timer() {
605 ev_timer_again(conn_.loop, &conn_.rt);
606 }
607
stop_read_timer()608 void ClientHandler::stop_read_timer() { ev_timer_stop(conn_.loop, &conn_.rt); }
609
validate_next_proto()610 int ClientHandler::validate_next_proto() {
611 const unsigned char *next_proto = nullptr;
612 unsigned int next_proto_len = 0;
613
614 // First set callback for catch all cases
615 on_read_ = &ClientHandler::upstream_read;
616
617 SSL_get0_alpn_selected(conn_.tls.ssl, &next_proto, &next_proto_len);
618
619 StringRef proto;
620
621 if (next_proto) {
622 proto = StringRef{next_proto, next_proto_len};
623
624 if (LOG_ENABLED(INFO)) {
625 CLOG(INFO, this) << "The negotiated next protocol: " << proto;
626 }
627 } else {
628 if (LOG_ENABLED(INFO)) {
629 CLOG(INFO, this) << "No protocol negotiated. Fallback to HTTP/1.1";
630 }
631
632 proto = "http/1.1"_sr;
633 }
634
635 if (!tls::in_proto_list(get_config()->tls.alpn_list, proto)) {
636 if (LOG_ENABLED(INFO)) {
637 CLOG(INFO, this) << "The negotiated protocol is not supported: " << proto;
638 }
639 return -1;
640 }
641
642 if (util::check_h2_is_selected(proto)) {
643 on_read_ = &ClientHandler::upstream_http2_connhd_read;
644
645 auto http2_upstream = std::make_unique<Http2Upstream>(this);
646
647 upstream_ = std::move(http2_upstream);
648 alpn_ = make_string_ref(balloc_, proto);
649
650 // At this point, input buffer is already filled with some bytes.
651 // The read callback is not called until new data come. So consume
652 // input buffer here.
653 if (on_read() != 0) {
654 return -1;
655 }
656
657 return 0;
658 }
659
660 if (proto == "http/1.1"_sr) {
661 upstream_ = std::make_unique<HttpsUpstream>(this);
662 alpn_ = "http/1.1"_sr;
663
664 // At this point, input buffer is already filled with some bytes.
665 // The read callback is not called until new data come. So consume
666 // input buffer here.
667 if (on_read() != 0) {
668 return -1;
669 }
670
671 return 0;
672 }
673 if (LOG_ENABLED(INFO)) {
674 CLOG(INFO, this) << "The negotiated protocol is not supported";
675 }
676 return -1;
677 }
678
do_read()679 int ClientHandler::do_read() { return read_(*this); }
do_write()680 int ClientHandler::do_write() { return write_(*this); }
681
on_read()682 int ClientHandler::on_read() {
683 if (rb_.chunk_avail()) {
684 auto rv = on_read_(*this);
685 if (rv != 0) {
686 return rv;
687 }
688 }
689 conn_.handle_tls_pending_read();
690 return 0;
691 }
on_write()692 int ClientHandler::on_write() { return on_write_(*this); }
693
get_ipaddr() const694 const StringRef &ClientHandler::get_ipaddr() const { return ipaddr_; }
695
get_should_close_after_write() const696 bool ClientHandler::get_should_close_after_write() const {
697 return should_close_after_write_;
698 }
699
set_should_close_after_write(bool f)700 void ClientHandler::set_should_close_after_write(bool f) {
701 should_close_after_write_ = f;
702 }
703
pool_downstream_connection(std::unique_ptr<DownstreamConnection> dconn)704 void ClientHandler::pool_downstream_connection(
705 std::unique_ptr<DownstreamConnection> dconn) {
706 if (!dconn->poolable()) {
707 return;
708 }
709
710 dconn->set_client_handler(nullptr);
711
712 auto &group = dconn->get_downstream_addr_group();
713
714 if (LOG_ENABLED(INFO)) {
715 CLOG(INFO, this) << "Pooling downstream connection DCONN:" << dconn.get()
716 << " in group " << group;
717 }
718
719 auto addr = dconn->get_addr();
720 auto &dconn_pool = addr->dconn_pool;
721 dconn_pool->add_downstream_connection(std::move(dconn));
722 }
723
724 namespace {
725 // Computes 32bits hash for session affinity for IP address |ip|.
compute_affinity_from_ip(const StringRef & ip)726 uint32_t compute_affinity_from_ip(const StringRef &ip) {
727 int rv;
728 std::array<uint8_t, 32> buf;
729
730 rv = util::sha256(buf.data(), ip);
731 if (rv != 0) {
732 // Not sure when sha256 failed. Just fall back to another
733 // function.
734 return util::hash32(ip);
735 }
736
737 return (static_cast<uint32_t>(buf[0]) << 24) |
738 (static_cast<uint32_t>(buf[1]) << 16) |
739 (static_cast<uint32_t>(buf[2]) << 8) | static_cast<uint32_t>(buf[3]);
740 }
741 } // namespace
742
get_http2_session(const std::shared_ptr<DownstreamAddrGroup> & group,DownstreamAddr * addr)743 Http2Session *ClientHandler::get_http2_session(
744 const std::shared_ptr<DownstreamAddrGroup> &group, DownstreamAddr *addr) {
745 auto &shared_addr = group->shared_addr;
746
747 if (LOG_ENABLED(INFO)) {
748 CLOG(INFO, this) << "Selected DownstreamAddr=" << addr
749 << ", index=" << (addr - shared_addr->addrs.data());
750 }
751
752 for (auto session = addr->http2_extra_freelist.head; session;) {
753 auto next = session->dlnext;
754
755 if (session->max_concurrency_reached(0)) {
756 if (LOG_ENABLED(INFO)) {
757 CLOG(INFO, this)
758 << "Maximum streams have been reached for Http2Session(" << session
759 << "). Skip it";
760 }
761
762 session->remove_from_freelist();
763 session = next;
764
765 continue;
766 }
767
768 if (LOG_ENABLED(INFO)) {
769 CLOG(INFO, this) << "Use Http2Session " << session
770 << " from http2_extra_freelist";
771 }
772
773 if (session->max_concurrency_reached(1)) {
774 if (LOG_ENABLED(INFO)) {
775 CLOG(INFO, this) << "Maximum streams are reached for Http2Session("
776 << session << ").";
777 }
778
779 session->remove_from_freelist();
780 }
781 return session;
782 }
783
784 auto session = new Http2Session(conn_.loop, worker_->get_cl_ssl_ctx(),
785 worker_, group, addr);
786
787 if (LOG_ENABLED(INFO)) {
788 CLOG(INFO, this) << "Create new Http2Session " << session;
789 }
790
791 session->add_to_extra_freelist();
792
793 return session;
794 }
795
get_affinity_cookie(Downstream * downstream,const StringRef & cookie_name)796 uint32_t ClientHandler::get_affinity_cookie(Downstream *downstream,
797 const StringRef &cookie_name) {
798 auto h = downstream->find_affinity_cookie(cookie_name);
799 if (h) {
800 return h;
801 }
802
803 auto d = std::uniform_int_distribution<uint32_t>(1);
804 auto rh = d(worker_->get_randgen());
805 h = util::hash32(StringRef{reinterpret_cast<char *>(&rh), sizeof(rh)});
806
807 downstream->renew_affinity_cookie(h);
808
809 return h;
810 }
811
812 namespace {
reschedule_addr(std::priority_queue<DownstreamAddrEntry,std::vector<DownstreamAddrEntry>,DownstreamAddrEntryGreater> & pq,DownstreamAddr * addr)813 void reschedule_addr(
814 std::priority_queue<DownstreamAddrEntry, std::vector<DownstreamAddrEntry>,
815 DownstreamAddrEntryGreater> &pq,
816 DownstreamAddr *addr) {
817 auto penalty = MAX_DOWNSTREAM_ADDR_WEIGHT + addr->pending_penalty;
818 addr->cycle += penalty / addr->weight;
819 addr->pending_penalty = penalty % addr->weight;
820
821 pq.push(DownstreamAddrEntry{addr, addr->seq, addr->cycle});
822 addr->queued = true;
823 }
824 } // namespace
825
826 namespace {
reschedule_wg(std::priority_queue<WeightGroupEntry,std::vector<WeightGroupEntry>,WeightGroupEntryGreater> & pq,WeightGroup * wg)827 void reschedule_wg(
828 std::priority_queue<WeightGroupEntry, std::vector<WeightGroupEntry>,
829 WeightGroupEntryGreater> &pq,
830 WeightGroup *wg) {
831 auto penalty = MAX_DOWNSTREAM_ADDR_WEIGHT + wg->pending_penalty;
832 wg->cycle += penalty / wg->weight;
833 wg->pending_penalty = penalty % wg->weight;
834
835 pq.push(WeightGroupEntry{wg, wg->seq, wg->cycle});
836 wg->queued = true;
837 }
838 } // namespace
839
get_downstream_addr(int & err,DownstreamAddrGroup * group,Downstream * downstream)840 DownstreamAddr *ClientHandler::get_downstream_addr(int &err,
841 DownstreamAddrGroup *group,
842 Downstream *downstream) {
843 err = 0;
844
845 switch (faddr_->alt_mode) {
846 case UpstreamAltMode::API:
847 case UpstreamAltMode::HEALTHMON:
848 assert(0);
849 default:
850 break;
851 }
852
853 auto &shared_addr = group->shared_addr;
854
855 if (shared_addr->affinity.type != SessionAffinity::NONE) {
856 uint32_t hash;
857 switch (shared_addr->affinity.type) {
858 case SessionAffinity::IP:
859 if (!affinity_hash_computed_) {
860 affinity_hash_ = compute_affinity_from_ip(ipaddr_);
861 affinity_hash_computed_ = true;
862 }
863 hash = affinity_hash_;
864 break;
865 case SessionAffinity::COOKIE:
866 if (shared_addr->affinity.cookie.stickiness ==
867 SessionAffinityCookieStickiness::STRICT) {
868 return get_downstream_addr_strict_affinity(err, shared_addr,
869 downstream);
870 }
871
872 hash = get_affinity_cookie(downstream, shared_addr->affinity.cookie.name);
873 break;
874 default:
875 assert(0);
876 }
877
878 const auto &affinity_hash = shared_addr->affinity_hash;
879
880 auto it = std::lower_bound(
881 std::begin(affinity_hash), std::end(affinity_hash), hash,
882 [](const AffinityHash &lhs, uint32_t rhs) { return lhs.hash < rhs; });
883
884 if (it == std::end(affinity_hash)) {
885 it = std::begin(affinity_hash);
886 }
887
888 auto aff_idx =
889 static_cast<size_t>(std::distance(std::begin(affinity_hash), it));
890 auto idx = (*it).idx;
891 auto addr = &shared_addr->addrs[idx];
892
893 if (addr->connect_blocker->blocked()) {
894 size_t i;
895 for (i = aff_idx + 1; i != aff_idx; ++i) {
896 if (i == shared_addr->affinity_hash.size()) {
897 i = 0;
898 }
899 addr = &shared_addr->addrs[shared_addr->affinity_hash[i].idx];
900 if (addr->connect_blocker->blocked()) {
901 continue;
902 }
903 break;
904 }
905 if (i == aff_idx) {
906 err = -1;
907 return nullptr;
908 }
909 }
910
911 return addr;
912 }
913
914 auto &wgpq = shared_addr->pq;
915
916 for (;;) {
917 if (wgpq.empty()) {
918 CLOG(INFO, this) << "No working downstream address found";
919 err = -1;
920 return nullptr;
921 }
922
923 auto wg = wgpq.top().wg;
924 wgpq.pop();
925 wg->queued = false;
926
927 for (;;) {
928 if (wg->pq.empty()) {
929 break;
930 }
931
932 auto addr = wg->pq.top().addr;
933 wg->pq.pop();
934 addr->queued = false;
935
936 if (addr->connect_blocker->blocked()) {
937 continue;
938 }
939
940 reschedule_addr(wg->pq, addr);
941 reschedule_wg(wgpq, wg);
942
943 return addr;
944 }
945 }
946 }
947
get_downstream_addr_strict_affinity(int & err,const std::shared_ptr<SharedDownstreamAddr> & shared_addr,Downstream * downstream)948 DownstreamAddr *ClientHandler::get_downstream_addr_strict_affinity(
949 int &err, const std::shared_ptr<SharedDownstreamAddr> &shared_addr,
950 Downstream *downstream) {
951 const auto &affinity_hash = shared_addr->affinity_hash;
952
953 auto h = downstream->find_affinity_cookie(shared_addr->affinity.cookie.name);
954 if (h) {
955 auto it = shared_addr->affinity_hash_map.find(h);
956 if (it != std::end(shared_addr->affinity_hash_map)) {
957 auto addr = &shared_addr->addrs[(*it).second];
958 if (!addr->connect_blocker->blocked()) {
959 return addr;
960 }
961 }
962 } else {
963 auto d = std::uniform_int_distribution<uint32_t>(1);
964 auto rh = d(worker_->get_randgen());
965 h = util::hash32(StringRef{reinterpret_cast<char *>(&rh), sizeof(rh)});
966 }
967
968 // Client is not bound to a particular backend, or the bound backend
969 // is not found, or is blocked. Find new backend using h. Using
970 // existing h allows us to find new server in a deterministic way.
971 // It is preferable because multiple concurrent requests with the
972 // stale cookie might be in-flight.
973 auto it = std::lower_bound(
974 std::begin(affinity_hash), std::end(affinity_hash), h,
975 [](const AffinityHash &lhs, uint32_t rhs) { return lhs.hash < rhs; });
976
977 if (it == std::end(affinity_hash)) {
978 it = std::begin(affinity_hash);
979 }
980
981 auto aff_idx =
982 static_cast<size_t>(std::distance(std::begin(affinity_hash), it));
983 auto idx = (*it).idx;
984 auto addr = &shared_addr->addrs[idx];
985
986 if (addr->connect_blocker->blocked()) {
987 size_t i;
988 for (i = aff_idx + 1; i != aff_idx; ++i) {
989 if (i == shared_addr->affinity_hash.size()) {
990 i = 0;
991 }
992 addr = &shared_addr->addrs[shared_addr->affinity_hash[i].idx];
993 if (addr->connect_blocker->blocked()) {
994 continue;
995 }
996 break;
997 }
998 if (i == aff_idx) {
999 err = -1;
1000 return nullptr;
1001 }
1002 }
1003
1004 downstream->renew_affinity_cookie(addr->affinity_hash);
1005
1006 return addr;
1007 }
1008
1009 std::unique_ptr<DownstreamConnection>
get_downstream_connection(int & err,Downstream * downstream)1010 ClientHandler::get_downstream_connection(int &err, Downstream *downstream) {
1011 size_t group_idx;
1012 auto &downstreamconf = *worker_->get_downstream_config();
1013 auto &routerconf = downstreamconf.router;
1014
1015 auto catch_all = downstreamconf.addr_group_catch_all;
1016 auto &groups = worker_->get_downstream_addr_groups();
1017
1018 auto &req = downstream->request();
1019
1020 err = 0;
1021
1022 switch (faddr_->alt_mode) {
1023 case UpstreamAltMode::API: {
1024 auto dconn = std::make_unique<APIDownstreamConnection>(worker_);
1025 dconn->set_client_handler(this);
1026 return dconn;
1027 }
1028 case UpstreamAltMode::HEALTHMON: {
1029 auto dconn = std::make_unique<HealthMonitorDownstreamConnection>();
1030 dconn->set_client_handler(this);
1031 return dconn;
1032 }
1033 default:
1034 break;
1035 }
1036
1037 auto &balloc = downstream->get_block_allocator();
1038
1039 StringRef authority, path;
1040
1041 if (req.forwarded_once) {
1042 if (groups.size() != 1) {
1043 authority = req.orig_authority;
1044 path = req.orig_path;
1045 }
1046 } else {
1047 if (faddr_->sni_fwd) {
1048 authority = sni_;
1049 } else if (!req.authority.empty()) {
1050 authority = req.authority;
1051 } else {
1052 auto h = req.fs.header(http2::HD_HOST);
1053 if (h) {
1054 authority = h->value;
1055 }
1056 }
1057
1058 // CONNECT method does not have path. But we requires path in
1059 // host-path mapping. As workaround, we assume that path is
1060 // "/".
1061 if (!req.regular_connect_method()) {
1062 path = req.path;
1063 }
1064
1065 // Cache the authority and path used for the first-time backend
1066 // selection because per-pattern mruby script can change them.
1067 req.orig_authority = authority;
1068 req.orig_path = path;
1069 req.forwarded_once = true;
1070 }
1071
1072 // Fast path. If we have one group, it must be catch-all group.
1073 if (groups.size() == 1) {
1074 group_idx = 0;
1075 } else {
1076 group_idx = match_downstream_addr_group(routerconf, authority, path, groups,
1077 catch_all, balloc);
1078 }
1079
1080 if (LOG_ENABLED(INFO)) {
1081 CLOG(INFO, this) << "Downstream address group_idx: " << group_idx;
1082 }
1083
1084 if (groups[group_idx]->shared_addr->redirect_if_not_tls && !conn_.tls.ssl) {
1085 if (LOG_ENABLED(INFO)) {
1086 CLOG(INFO, this) << "Downstream address group " << group_idx
1087 << " requires frontend TLS connection.";
1088 }
1089 err = SHRPX_ERR_TLS_REQUIRED;
1090 return nullptr;
1091 }
1092
1093 auto &group = groups[group_idx];
1094
1095 if (group->shared_addr->dnf) {
1096 auto dconn = std::make_unique<NullDownstreamConnection>(group);
1097 dconn->set_client_handler(this);
1098 return dconn;
1099 }
1100
1101 auto addr = get_downstream_addr(err, group.get(), downstream);
1102 if (addr == nullptr) {
1103 return nullptr;
1104 }
1105
1106 if (addr->proto == Proto::HTTP1) {
1107 auto dconn = addr->dconn_pool->pop_downstream_connection();
1108 if (dconn) {
1109 dconn->set_client_handler(this);
1110 return dconn;
1111 }
1112
1113 if (worker_->get_connect_blocker()->blocked()) {
1114 if (LOG_ENABLED(INFO)) {
1115 DCLOG(INFO, this)
1116 << "Worker wide backend connection was blocked temporarily";
1117 }
1118 return nullptr;
1119 }
1120
1121 if (LOG_ENABLED(INFO)) {
1122 CLOG(INFO, this) << "Downstream connection pool is empty."
1123 << " Create new one";
1124 }
1125
1126 dconn = std::make_unique<HttpDownstreamConnection>(group, addr, conn_.loop,
1127 worker_);
1128 dconn->set_client_handler(this);
1129 return dconn;
1130 }
1131
1132 if (LOG_ENABLED(INFO)) {
1133 CLOG(INFO, this) << "Downstream connection pool is empty."
1134 << " Create new one";
1135 }
1136
1137 auto http2session = get_http2_session(group, addr);
1138 auto dconn = std::make_unique<Http2DownstreamConnection>(http2session);
1139 dconn->set_client_handler(this);
1140 return dconn;
1141 }
1142
get_mcpool()1143 MemchunkPool *ClientHandler::get_mcpool() { return worker_->get_mcpool(); }
1144
get_ssl() const1145 SSL *ClientHandler::get_ssl() const { return conn_.tls.ssl; }
1146
direct_http2_upgrade()1147 void ClientHandler::direct_http2_upgrade() {
1148 upstream_ = std::make_unique<Http2Upstream>(this);
1149 alpn_ = NGHTTP2_CLEARTEXT_PROTO_VERSION_ID ""_sr;
1150 on_read_ = &ClientHandler::upstream_read;
1151 write_ = &ClientHandler::write_clear;
1152 }
1153
perform_http2_upgrade(HttpsUpstream * http)1154 int ClientHandler::perform_http2_upgrade(HttpsUpstream *http) {
1155 auto upstream = std::make_unique<Http2Upstream>(this);
1156
1157 auto output = upstream->get_response_buf();
1158
1159 // We might have written non-final header in response_buf, in this
1160 // case, response_state is still INITIAL. If this non-final header
1161 // and upgrade header fit in output buffer, do upgrade. Otherwise,
1162 // to avoid to send this non-final header as response body in HTTP/2
1163 // upstream, fail upgrade.
1164 auto downstream = http->get_downstream();
1165 auto input = downstream->get_response_buf();
1166
1167 if (upstream->upgrade_upstream(http) != 0) {
1168 return -1;
1169 }
1170 // http pointer is now owned by upstream.
1171 upstream_.release();
1172 // TODO We might get other version id in HTTP2-settings, if we
1173 // support aliasing for h2, but we just use library default for now.
1174 alpn_ = NGHTTP2_CLEARTEXT_PROTO_VERSION_ID ""_sr;
1175 on_read_ = &ClientHandler::upstream_http2_connhd_read;
1176 write_ = &ClientHandler::write_clear;
1177
1178 input->remove(*output, input->rleft());
1179
1180 constexpr auto res = "HTTP/1.1 101 Switching Protocols\r\n"
1181 "Connection: Upgrade\r\n"
1182 "Upgrade: " NGHTTP2_CLEARTEXT_PROTO_VERSION_ID "\r\n"
1183 "\r\n"_sr;
1184
1185 output->append(res);
1186 upstream_ = std::move(upstream);
1187
1188 signal_write();
1189 return 0;
1190 }
1191
get_http2_upgrade_allowed() const1192 bool ClientHandler::get_http2_upgrade_allowed() const { return !conn_.tls.ssl; }
1193
get_upstream_scheme() const1194 StringRef ClientHandler::get_upstream_scheme() const {
1195 if (conn_.tls.ssl) {
1196 return "https"_sr;
1197 } else {
1198 return "http"_sr;
1199 }
1200 }
1201
start_immediate_shutdown()1202 void ClientHandler::start_immediate_shutdown() {
1203 ev_timer_start(conn_.loop, &reneg_shutdown_timer_);
1204 }
1205
write_accesslog(Downstream * downstream)1206 void ClientHandler::write_accesslog(Downstream *downstream) {
1207 auto &req = downstream->request();
1208
1209 auto config = get_config();
1210
1211 if (!req.tstamp) {
1212 auto lgconf = log_config();
1213 lgconf->update_tstamp(std::chrono::system_clock::now());
1214 req.tstamp = lgconf->tstamp;
1215 }
1216
1217 upstream_accesslog(
1218 config->logging.access.format,
1219 LogSpec{
1220 downstream,
1221 ipaddr_,
1222 alpn_,
1223 sni_,
1224 conn_.tls.ssl,
1225 std::chrono::high_resolution_clock::now(), // request_end_time
1226 port_,
1227 faddr_->port,
1228 config->pid,
1229 });
1230 }
1231
get_rb()1232 ClientHandler::ReadBuf *ClientHandler::get_rb() { return &rb_; }
1233
signal_write()1234 void ClientHandler::signal_write() { conn_.wlimit.startw(); }
1235
get_rlimit()1236 RateLimit *ClientHandler::get_rlimit() { return &conn_.rlimit; }
get_wlimit()1237 RateLimit *ClientHandler::get_wlimit() { return &conn_.wlimit; }
1238
get_wev()1239 ev_io *ClientHandler::get_wev() { return &conn_.wev; }
1240
get_worker() const1241 Worker *ClientHandler::get_worker() const { return worker_; }
1242
1243 namespace {
parse_proxy_line_port(const uint8_t * first,const uint8_t * last)1244 ssize_t parse_proxy_line_port(const uint8_t *first, const uint8_t *last) {
1245 auto p = first;
1246 int32_t port = 0;
1247
1248 if (p == last) {
1249 return -1;
1250 }
1251
1252 if (*p == '0') {
1253 if (p + 1 != last && util::is_digit(*(p + 1))) {
1254 return -1;
1255 }
1256 return 1;
1257 }
1258
1259 for (; p != last && util::is_digit(*p); ++p) {
1260 port *= 10;
1261 port += *p - '0';
1262
1263 if (port > 65535) {
1264 return -1;
1265 }
1266 }
1267
1268 return p - first;
1269 }
1270 } // namespace
1271
on_proxy_protocol_finish()1272 int ClientHandler::on_proxy_protocol_finish() {
1273 auto len = rb_.pos() - rb_.begin();
1274
1275 assert(len);
1276
1277 if (LOG_ENABLED(INFO)) {
1278 CLOG(INFO, this) << "PROXY-protocol: Draining " << len
1279 << " bytes from socket";
1280 }
1281
1282 rb_.reset();
1283
1284 if (conn_.read_nolim_clear(rb_.pos(), len) < 0) {
1285 return -1;
1286 }
1287
1288 rb_.reset();
1289
1290 setup_upstream_io_callback();
1291
1292 return 0;
1293 }
1294
1295 namespace {
1296 // PROXY-protocol v2 header signature
1297 constexpr uint8_t PROXY_PROTO_V2_SIG[] =
1298 "\x0D\x0A\x0D\x0A\x00\x0D\x0A\x51\x55\x49\x54\x0A";
1299
1300 // PROXY-protocol v2 header length
1301 constexpr size_t PROXY_PROTO_V2_HDLEN =
1302 str_size(PROXY_PROTO_V2_SIG) + /* ver_cmd(1) + fam(1) + len(2) = */ 4;
1303 } // namespace
1304
1305 // http://www.haproxy.org/download/1.5/doc/proxy-protocol.txt
proxy_protocol_read()1306 int ClientHandler::proxy_protocol_read() {
1307 if (LOG_ENABLED(INFO)) {
1308 CLOG(INFO, this) << "PROXY-protocol: Started";
1309 }
1310
1311 auto first = rb_.pos();
1312
1313 if (rb_.rleft() >= PROXY_PROTO_V2_HDLEN &&
1314 (*(first + str_size(PROXY_PROTO_V2_SIG)) & 0xf0) == 0x20) {
1315 if (LOG_ENABLED(INFO)) {
1316 CLOG(INFO, this) << "PROXY-protocol: Detected v2 header signature";
1317 }
1318 return proxy_protocol_v2_read();
1319 }
1320
1321 // NULL character really destroys functions which expects NULL
1322 // terminated string. We won't expect it in PROXY protocol line, so
1323 // find it here.
1324 auto chrs = std::to_array({'\n', '\0'});
1325
1326 constexpr size_t MAX_PROXY_LINELEN = 107;
1327
1328 auto bufend = rb_.pos() + std::min(MAX_PROXY_LINELEN, rb_.rleft());
1329
1330 auto end =
1331 std::find_first_of(rb_.pos(), bufend, std::begin(chrs), std::end(chrs));
1332
1333 if (end == bufend || *end == '\0' || end == rb_.pos() || *(end - 1) != '\r') {
1334 if (LOG_ENABLED(INFO)) {
1335 CLOG(INFO, this) << "PROXY-protocol-v1: No ending CR LF sequence found";
1336 }
1337 return -1;
1338 }
1339
1340 --end;
1341
1342 constexpr auto HEADER = "PROXY "_sr;
1343
1344 if (static_cast<size_t>(end - rb_.pos()) < HEADER.size()) {
1345 if (LOG_ENABLED(INFO)) {
1346 CLOG(INFO, this) << "PROXY-protocol-v1: PROXY version 1 ID not found";
1347 }
1348 return -1;
1349 }
1350
1351 if (HEADER != StringRef{rb_.pos(), HEADER.size()}) {
1352 if (LOG_ENABLED(INFO)) {
1353 CLOG(INFO, this) << "PROXY-protocol-v1: Bad PROXY protocol version 1 ID";
1354 }
1355 return -1;
1356 }
1357
1358 rb_.drain(HEADER.size());
1359
1360 int family;
1361
1362 if (rb_.pos()[0] == 'T') {
1363 if (end - rb_.pos() < 5) {
1364 if (LOG_ENABLED(INFO)) {
1365 CLOG(INFO, this) << "PROXY-protocol-v1: INET protocol family not found";
1366 }
1367 return -1;
1368 }
1369
1370 if (rb_.pos()[1] != 'C' || rb_.pos()[2] != 'P') {
1371 if (LOG_ENABLED(INFO)) {
1372 CLOG(INFO, this) << "PROXY-protocol-v1: Unknown INET protocol family";
1373 }
1374 return -1;
1375 }
1376
1377 switch (rb_.pos()[3]) {
1378 case '4':
1379 family = AF_INET;
1380 break;
1381 case '6':
1382 family = AF_INET6;
1383 break;
1384 default:
1385 if (LOG_ENABLED(INFO)) {
1386 CLOG(INFO, this) << "PROXY-protocol-v1: Unknown INET protocol family";
1387 }
1388 return -1;
1389 }
1390
1391 rb_.drain(5);
1392 } else {
1393 if (end - rb_.pos() < 7) {
1394 if (LOG_ENABLED(INFO)) {
1395 CLOG(INFO, this) << "PROXY-protocol-v1: INET protocol family not found";
1396 }
1397 return -1;
1398 }
1399 if ("UNKNOWN"_sr != StringRef{rb_.pos(), 7}) {
1400 if (LOG_ENABLED(INFO)) {
1401 CLOG(INFO, this) << "PROXY-protocol-v1: Unknown INET protocol family";
1402 }
1403 return -1;
1404 }
1405
1406 rb_.drain(end + 2 - rb_.pos());
1407
1408 return on_proxy_protocol_finish();
1409 }
1410
1411 // source address
1412 auto token_end = std::find(rb_.pos(), end, ' ');
1413 if (token_end == end) {
1414 if (LOG_ENABLED(INFO)) {
1415 CLOG(INFO, this) << "PROXY-protocol-v1: Source address not found";
1416 }
1417 return -1;
1418 }
1419
1420 *token_end = '\0';
1421 if (!util::numeric_host(reinterpret_cast<const char *>(rb_.pos()), family)) {
1422 if (LOG_ENABLED(INFO)) {
1423 CLOG(INFO, this) << "PROXY-protocol-v1: Invalid source address";
1424 }
1425 return -1;
1426 }
1427
1428 auto src_addr = rb_.pos();
1429 auto src_addrlen = token_end - rb_.pos();
1430
1431 rb_.drain(token_end - rb_.pos() + 1);
1432
1433 // destination address
1434 token_end = std::find(rb_.pos(), end, ' ');
1435 if (token_end == end) {
1436 if (LOG_ENABLED(INFO)) {
1437 CLOG(INFO, this) << "PROXY-protocol-v1: Destination address not found";
1438 }
1439 return -1;
1440 }
1441
1442 *token_end = '\0';
1443 if (!util::numeric_host(reinterpret_cast<const char *>(rb_.pos()), family)) {
1444 if (LOG_ENABLED(INFO)) {
1445 CLOG(INFO, this) << "PROXY-protocol-v1: Invalid destination address";
1446 }
1447 return -1;
1448 }
1449
1450 // Currently we don't use destination address
1451
1452 rb_.drain(token_end - rb_.pos() + 1);
1453
1454 // source port
1455 auto n = parse_proxy_line_port(rb_.pos(), end);
1456 if (n <= 0 || *(rb_.pos() + n) != ' ') {
1457 if (LOG_ENABLED(INFO)) {
1458 CLOG(INFO, this) << "PROXY-protocol-v1: Invalid source port";
1459 }
1460 return -1;
1461 }
1462
1463 rb_.pos()[n] = '\0';
1464 auto src_port = rb_.pos();
1465 auto src_portlen = n;
1466
1467 rb_.drain(n + 1);
1468
1469 // destination port
1470 n = parse_proxy_line_port(rb_.pos(), end);
1471 if (n <= 0 || rb_.pos() + n != end) {
1472 if (LOG_ENABLED(INFO)) {
1473 CLOG(INFO, this) << "PROXY-protocol-v1: Invalid destination port";
1474 }
1475 return -1;
1476 }
1477
1478 // Currently we don't use destination port
1479
1480 rb_.drain(end + 2 - rb_.pos());
1481
1482 ipaddr_ = make_string_ref(
1483 balloc_, StringRef{src_addr, static_cast<size_t>(src_addrlen)});
1484 port_ = make_string_ref(
1485 balloc_, StringRef{src_port, static_cast<size_t>(src_portlen)});
1486
1487 if (LOG_ENABLED(INFO)) {
1488 CLOG(INFO, this) << "PROXY-protocol-v1: Finished, " << (rb_.pos() - first)
1489 << " bytes read";
1490 }
1491
1492 auto config = get_config();
1493 auto &fwdconf = config->http.forwarded;
1494
1495 if ((fwdconf.params & FORWARDED_FOR) &&
1496 fwdconf.for_node_type == ForwardedNode::IP) {
1497 init_forwarded_for(family, ipaddr_);
1498 }
1499
1500 return on_proxy_protocol_finish();
1501 }
1502
proxy_protocol_v2_read()1503 int ClientHandler::proxy_protocol_v2_read() {
1504 // Assume that first str_size(PROXY_PROTO_V2_SIG) octets match v2
1505 // protocol signature and followed by the bytes which indicates v2.
1506 assert(rb_.rleft() >= PROXY_PROTO_V2_HDLEN);
1507
1508 auto p = rb_.pos() + str_size(PROXY_PROTO_V2_SIG);
1509
1510 assert(((*p) & 0xf0) == 0x20);
1511
1512 enum { LOCAL, PROXY } cmd;
1513
1514 auto cmd_bits = (*p++) & 0xf;
1515 switch (cmd_bits) {
1516 case 0x0:
1517 cmd = LOCAL;
1518 break;
1519 case 0x01:
1520 cmd = PROXY;
1521 break;
1522 default:
1523 if (LOG_ENABLED(INFO)) {
1524 CLOG(INFO, this) << "PROXY-protocol-v2: Unknown command " << log::hex
1525 << cmd_bits;
1526 }
1527 return -1;
1528 }
1529
1530 auto fam = *p++;
1531 uint16_t len;
1532 memcpy(&len, p, sizeof(len));
1533 len = ntohs(len);
1534
1535 p += sizeof(len);
1536
1537 if (LOG_ENABLED(INFO)) {
1538 CLOG(INFO, this) << "PROXY-protocol-v2: Detected family=" << log::hex << fam
1539 << ", len=" << log::dec << len;
1540 }
1541
1542 if (rb_.last() - p < len) {
1543 if (LOG_ENABLED(INFO)) {
1544 CLOG(INFO, this)
1545 << "PROXY-protocol-v2: Prematurely truncated header block; require "
1546 << len << " bytes, " << rb_.last() - p << " bytes left";
1547 }
1548 return -1;
1549 }
1550
1551 int family;
1552 std::array<char, std::max(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)> src_addr,
1553 dst_addr;
1554 size_t addrlen;
1555
1556 switch (fam) {
1557 case 0x11:
1558 case 0x12:
1559 if (len < 12) {
1560 if (LOG_ENABLED(INFO)) {
1561 CLOG(INFO, this) << "PROXY-protocol-v2: Too short AF_INET addresses";
1562 }
1563 return -1;
1564 }
1565 family = AF_INET;
1566 addrlen = 4;
1567 break;
1568 case 0x21:
1569 case 0x22:
1570 if (len < 36) {
1571 if (LOG_ENABLED(INFO)) {
1572 CLOG(INFO, this) << "PROXY-protocol-v2: Too short AF_INET6 addresses";
1573 }
1574 return -1;
1575 }
1576 family = AF_INET6;
1577 addrlen = 16;
1578 break;
1579 case 0x31:
1580 case 0x32:
1581 if (len < 216) {
1582 if (LOG_ENABLED(INFO)) {
1583 CLOG(INFO, this) << "PROXY-protocol-v2: Too short AF_UNIX addresses";
1584 }
1585 return -1;
1586 }
1587 // fall through
1588 case 0x00: {
1589 // UNSPEC and UNIX are just ignored.
1590 if (LOG_ENABLED(INFO)) {
1591 CLOG(INFO, this) << "PROXY-protocol-v2: Ignore combination of address "
1592 "family and protocol "
1593 << log::hex << fam;
1594 }
1595 rb_.drain(PROXY_PROTO_V2_HDLEN + len);
1596 return on_proxy_protocol_finish();
1597 }
1598 default:
1599 if (LOG_ENABLED(INFO)) {
1600 CLOG(INFO, this) << "PROXY-protocol-v2: Unknown combination of address "
1601 "family and protocol "
1602 << log::hex << fam;
1603 }
1604 return -1;
1605 }
1606
1607 if (cmd != PROXY) {
1608 if (LOG_ENABLED(INFO)) {
1609 CLOG(INFO, this) << "PROXY-protocol-v2: Ignore non-PROXY command";
1610 }
1611 rb_.drain(PROXY_PROTO_V2_HDLEN + len);
1612 return on_proxy_protocol_finish();
1613 }
1614
1615 if (inet_ntop(family, p, src_addr.data(), src_addr.size()) == nullptr) {
1616 if (LOG_ENABLED(INFO)) {
1617 CLOG(INFO, this) << "PROXY-protocol-v2: Unable to parse source address";
1618 }
1619 return -1;
1620 }
1621
1622 p += addrlen;
1623
1624 if (inet_ntop(family, p, dst_addr.data(), dst_addr.size()) == nullptr) {
1625 if (LOG_ENABLED(INFO)) {
1626 CLOG(INFO, this)
1627 << "PROXY-protocol-v2: Unable to parse destination address";
1628 }
1629 return -1;
1630 }
1631
1632 p += addrlen;
1633
1634 uint16_t src_port;
1635
1636 memcpy(&src_port, p, sizeof(src_port));
1637 src_port = ntohs(src_port);
1638
1639 // We don't use destination port.
1640 p += 4;
1641
1642 ipaddr_ = make_string_ref(balloc_, StringRef{src_addr.data()});
1643 port_ = util::make_string_ref_uint(balloc_, src_port);
1644
1645 if (LOG_ENABLED(INFO)) {
1646 CLOG(INFO, this) << "PROXY-protocol-v2: Finished reading proxy addresses, "
1647 << p - rb_.pos() << " bytes read, "
1648 << PROXY_PROTO_V2_HDLEN + len - (p - rb_.pos())
1649 << " bytes left";
1650 }
1651
1652 auto config = get_config();
1653 auto &fwdconf = config->http.forwarded;
1654
1655 if ((fwdconf.params & FORWARDED_FOR) &&
1656 fwdconf.for_node_type == ForwardedNode::IP) {
1657 init_forwarded_for(family, ipaddr_);
1658 }
1659
1660 rb_.drain(PROXY_PROTO_V2_HDLEN + len);
1661 return on_proxy_protocol_finish();
1662 }
1663
get_forwarded_by() const1664 StringRef ClientHandler::get_forwarded_by() const {
1665 auto &fwdconf = get_config()->http.forwarded;
1666
1667 if (fwdconf.by_node_type == ForwardedNode::OBFUSCATED) {
1668 return fwdconf.by_obfuscated;
1669 }
1670
1671 return faddr_->hostport;
1672 }
1673
get_forwarded_for() const1674 StringRef ClientHandler::get_forwarded_for() const { return forwarded_for_; }
1675
get_upstream_addr() const1676 const UpstreamAddr *ClientHandler::get_upstream_addr() const { return faddr_; }
1677
get_connection()1678 Connection *ClientHandler::get_connection() { return &conn_; };
1679
set_tls_sni(const StringRef & sni)1680 void ClientHandler::set_tls_sni(const StringRef &sni) {
1681 sni_ = make_string_ref(balloc_, sni);
1682 }
1683
get_tls_sni() const1684 StringRef ClientHandler::get_tls_sni() const { return sni_; }
1685
get_alpn() const1686 StringRef ClientHandler::get_alpn() const { return alpn_; }
1687
get_block_allocator()1688 BlockAllocator &ClientHandler::get_block_allocator() { return balloc_; }
1689
set_alpn_from_conn()1690 void ClientHandler::set_alpn_from_conn() {
1691 const unsigned char *alpn;
1692 unsigned int alpnlen;
1693
1694 SSL_get0_alpn_selected(conn_.tls.ssl, &alpn, &alpnlen);
1695
1696 alpn_ = make_string_ref(balloc_, StringRef{alpn, alpnlen});
1697 }
1698
1699 } // namespace shrpx
1700