blob: d515b6d281aae1f560ca485448df573a036cdc29 [file] [log] [blame]
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001/*
2 * QUIC protocol implementation. Lower layer with internal features implemented
3 * here such as QUIC encryption, idle timeout, acknowledgement and
4 * retransmission.
5 *
6 * Copyright 2020 HAProxy Technologies, Frederic Lecaille <flecaille@haproxy.com>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 */
14
15#include <haproxy/quic_conn.h>
16
17#define _GNU_SOURCE
18#include <errno.h>
19#include <stdio.h>
20#include <stdlib.h>
21
22#include <sys/socket.h>
23#include <sys/stat.h>
24#include <sys/types.h>
25
26#include <netinet/tcp.h>
27
28#include <import/ebmbtree.h>
29
30#include <haproxy/buf-t.h>
31#include <haproxy/compat.h>
32#include <haproxy/api.h>
33#include <haproxy/debug.h>
34#include <haproxy/tools.h>
35#include <haproxy/ticks.h>
36
37#include <haproxy/connection.h>
38#include <haproxy/fd.h>
39#include <haproxy/freq_ctr.h>
40#include <haproxy/global.h>
41#include <haproxy/h3.h>
42#include <haproxy/hq_interop.h>
43#include <haproxy/log.h>
44#include <haproxy/mux_quic.h>
45#include <haproxy/ncbuf.h>
46#include <haproxy/pipe.h>
47#include <haproxy/proxy.h>
48#include <haproxy/quic_cc.h>
49#include <haproxy/quic_frame.h>
50#include <haproxy/quic_enc.h>
51#include <haproxy/quic_loss.h>
52#include <haproxy/quic_sock.h>
53#include <haproxy/quic_stats.h>
54#include <haproxy/quic_stream.h>
55#include <haproxy/quic_tp.h>
56#include <haproxy/cbuf.h>
57#include <haproxy/proto_quic.h>
58#include <haproxy/quic_tls.h>
59#include <haproxy/ssl_sock.h>
60#include <haproxy/task.h>
61#include <haproxy/trace.h>
62
63/* list of supported QUIC versions by this implementation */
64const struct quic_version quic_versions[] = {
65 {
66 .num = QUIC_PROTOCOL_VERSION_DRAFT_29,
67 .initial_salt = initial_salt_draft_29,
68 .initial_salt_len = sizeof initial_salt_draft_29,
69 .key_label = (const unsigned char *)QUIC_HKDF_KEY_LABEL_V1,
70 .key_label_len = sizeof(QUIC_HKDF_KEY_LABEL_V1) - 1,
71 .iv_label = (const unsigned char *)QUIC_HKDF_IV_LABEL_V1,
72 .iv_label_len = sizeof(QUIC_HKDF_IV_LABEL_V1) - 1,
73 .hp_label = (const unsigned char *)QUIC_HKDF_HP_LABEL_V1,
74 .hp_label_len = sizeof(QUIC_HKDF_HP_LABEL_V1) - 1,
75 .ku_label = (const unsigned char *)QUIC_HKDF_KU_LABEL_V1,
76 .ku_label_len = sizeof(QUIC_HKDF_KU_LABEL_V1) - 1,
77 .retry_tag_key = (const unsigned char *)QUIC_TLS_RETRY_KEY_DRAFT,
78 .retry_tag_nonce = (const unsigned char *)QUIC_TLS_RETRY_NONCE_DRAFT,
79 },
80 {
81 .num = QUIC_PROTOCOL_VERSION_1,
82 .initial_salt = initial_salt_v1,
83 .initial_salt_len = sizeof initial_salt_v1,
84 .key_label = (const unsigned char *)QUIC_HKDF_KEY_LABEL_V1,
85 .key_label_len = sizeof(QUIC_HKDF_KEY_LABEL_V1) - 1,
86 .iv_label = (const unsigned char *)QUIC_HKDF_IV_LABEL_V1,
87 .iv_label_len = sizeof(QUIC_HKDF_IV_LABEL_V1) - 1,
88 .hp_label = (const unsigned char *)QUIC_HKDF_HP_LABEL_V1,
89 .hp_label_len = sizeof(QUIC_HKDF_HP_LABEL_V1) - 1,
90 .ku_label = (const unsigned char *)QUIC_HKDF_KU_LABEL_V1,
91 .ku_label_len = sizeof(QUIC_HKDF_KU_LABEL_V1) - 1,
92 .retry_tag_key = (const unsigned char *)QUIC_TLS_RETRY_KEY_V1,
93 .retry_tag_nonce = (const unsigned char *)QUIC_TLS_RETRY_NONCE_V1,
94 },
95 {
Frédéric Lécaille21c4c9b2023-01-13 16:37:02 +010096 .num = QUIC_PROTOCOL_VERSION_2,
97 .initial_salt = initial_salt_v2,
98 .initial_salt_len = sizeof initial_salt_v2,
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +020099 .key_label = (const unsigned char *)QUIC_HKDF_KEY_LABEL_V2,
100 .key_label_len = sizeof(QUIC_HKDF_KEY_LABEL_V2) - 1,
101 .iv_label = (const unsigned char *)QUIC_HKDF_IV_LABEL_V2,
102 .iv_label_len = sizeof(QUIC_HKDF_IV_LABEL_V2) - 1,
103 .hp_label = (const unsigned char *)QUIC_HKDF_HP_LABEL_V2,
104 .hp_label_len = sizeof(QUIC_HKDF_HP_LABEL_V2) - 1,
105 .ku_label = (const unsigned char *)QUIC_HKDF_KU_LABEL_V2,
106 .ku_label_len = sizeof(QUIC_HKDF_KU_LABEL_V2) - 1,
Frédéric Lécaille21c4c9b2023-01-13 16:37:02 +0100107 .retry_tag_key = (const unsigned char *)QUIC_TLS_RETRY_KEY_V2,
108 .retry_tag_nonce = (const unsigned char *)QUIC_TLS_RETRY_NONCE_V2,
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200109 },
110};
111
112/* The total number of supported versions */
113const size_t quic_versions_nb = sizeof quic_versions / sizeof *quic_versions;
114/* Listener only preferred version */
115const struct quic_version *preferred_version;
116
117/* trace source and events */
118static void quic_trace(enum trace_level level, uint64_t mask, \
119 const struct trace_source *src,
120 const struct ist where, const struct ist func,
121 const void *a1, const void *a2, const void *a3, const void *a4);
122
123static const struct trace_event quic_trace_events[] = {
124 { .mask = QUIC_EV_CONN_NEW, .name = "new_conn", .desc = "new QUIC connection" },
125 { .mask = QUIC_EV_CONN_INIT, .name = "new_conn_init", .desc = "new QUIC connection initialization" },
126 { .mask = QUIC_EV_CONN_ISEC, .name = "init_secs", .desc = "initial secrets derivation" },
127 { .mask = QUIC_EV_CONN_RSEC, .name = "read_secs", .desc = "read secrets derivation" },
128 { .mask = QUIC_EV_CONN_WSEC, .name = "write_secs", .desc = "write secrets derivation" },
129 { .mask = QUIC_EV_CONN_LPKT, .name = "lstnr_packet", .desc = "new listener received packet" },
130 { .mask = QUIC_EV_CONN_SPKT, .name = "srv_packet", .desc = "new server received packet" },
131 { .mask = QUIC_EV_CONN_ENCPKT, .name = "enc_hdshk_pkt", .desc = "handhshake packet encryption" },
132 { .mask = QUIC_EV_CONN_TXPKT, .name = "tx_pkt", .desc = "TX packet" },
133 { .mask = QUIC_EV_CONN_PAPKT, .name = "phdshk_apkt", .desc = "post handhshake application packet preparation" },
134 { .mask = QUIC_EV_CONN_PAPKTS, .name = "phdshk_apkts", .desc = "post handhshake application packets preparation" },
135 { .mask = QUIC_EV_CONN_IO_CB, .name = "qc_io_cb", .desc = "QUIC conn. I/O processing" },
136 { .mask = QUIC_EV_CONN_RMHP, .name = "rm_hp", .desc = "Remove header protection" },
137 { .mask = QUIC_EV_CONN_PRSHPKT, .name = "parse_hpkt", .desc = "parse handshake packet" },
138 { .mask = QUIC_EV_CONN_PRSAPKT, .name = "parse_apkt", .desc = "parse application packet" },
139 { .mask = QUIC_EV_CONN_PRSFRM, .name = "parse_frm", .desc = "parse frame" },
140 { .mask = QUIC_EV_CONN_PRSAFRM, .name = "parse_ack_frm", .desc = "parse ACK frame" },
141 { .mask = QUIC_EV_CONN_BFRM, .name = "build_frm", .desc = "build frame" },
142 { .mask = QUIC_EV_CONN_PHPKTS, .name = "phdshk_pkts", .desc = "handhshake packets preparation" },
143 { .mask = QUIC_EV_CONN_TRMHP, .name = "rm_hp_try", .desc = "header protection removing try" },
144 { .mask = QUIC_EV_CONN_ELRMHP, .name = "el_rm_hp", .desc = "handshake enc. level header protection removing" },
145 { .mask = QUIC_EV_CONN_RXPKT, .name = "rx_pkt", .desc = "RX packet" },
146 { .mask = QUIC_EV_CONN_SSLDATA, .name = "ssl_provide_data", .desc = "CRYPTO data provision to TLS stack" },
147 { .mask = QUIC_EV_CONN_RXCDATA, .name = "el_treat_rx_cfrms",.desc = "enc. level RX CRYPTO frames processing"},
148 { .mask = QUIC_EV_CONN_ADDDATA, .name = "add_hdshk_data", .desc = "TLS stack ->add_handshake_data() call"},
149 { .mask = QUIC_EV_CONN_FFLIGHT, .name = "flush_flight", .desc = "TLS stack ->flush_flight() call"},
150 { .mask = QUIC_EV_CONN_SSLALERT, .name = "send_alert", .desc = "TLS stack ->send_alert() call"},
151 { .mask = QUIC_EV_CONN_RTTUPDT, .name = "rtt_updt", .desc = "RTT sampling" },
152 { .mask = QUIC_EV_CONN_SPPKTS, .name = "sppkts", .desc = "send prepared packets" },
153 { .mask = QUIC_EV_CONN_PKTLOSS, .name = "pktloss", .desc = "detect packet loss" },
154 { .mask = QUIC_EV_CONN_STIMER, .name = "stimer", .desc = "set timer" },
155 { .mask = QUIC_EV_CONN_PTIMER, .name = "ptimer", .desc = "process timer" },
156 { .mask = QUIC_EV_CONN_SPTO, .name = "spto", .desc = "set PTO" },
157 { .mask = QUIC_EV_CONN_BCFRMS, .name = "bcfrms", .desc = "build CRYPTO data frames" },
158 { .mask = QUIC_EV_CONN_XPRTSEND, .name = "xprt_send", .desc = "sending XRPT subscription" },
159 { .mask = QUIC_EV_CONN_XPRTRECV, .name = "xprt_recv", .desc = "receiving XRPT subscription" },
160 { .mask = QUIC_EV_CONN_FREED, .name = "conn_freed", .desc = "releasing conn. memory" },
161 { .mask = QUIC_EV_CONN_CLOSE, .name = "conn_close", .desc = "closing conn." },
162 { .mask = QUIC_EV_CONN_ACKSTRM, .name = "ack_strm", .desc = "STREAM ack."},
163 { .mask = QUIC_EV_CONN_FRMLIST, .name = "frm_list", .desc = "frame list"},
164 { .mask = QUIC_EV_STATELESS_RST, .name = "stateless_reset", .desc = "stateless reset sent"},
165 { .mask = QUIC_EV_TRANSP_PARAMS, .name = "transport_params", .desc = "transport parameters"},
166 { .mask = QUIC_EV_CONN_IDLE_TIMER, .name = "idle_timer", .desc = "idle timer task"},
167 { .mask = QUIC_EV_CONN_SUB, .name = "xprt_sub", .desc = "RX/TX subcription or unsubscription to QUIC xprt"},
Amaury Denoyelle5b414862022-10-24 17:40:37 +0200168 { .mask = QUIC_EV_CONN_RCV, .name = "conn_recv", .desc = "RX on connection" },
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200169 { /* end */ }
170};
171
172static const struct name_desc quic_trace_lockon_args[4] = {
173 /* arg1 */ { /* already used by the connection */ },
174 /* arg2 */ { .name="quic", .desc="QUIC transport" },
175 /* arg3 */ { },
176 /* arg4 */ { }
177};
178
179static const struct name_desc quic_trace_decoding[] = {
180#define QUIC_VERB_CLEAN 1
181 { .name="clean", .desc="only user-friendly stuff, generally suitable for level \"user\"" },
182 { /* end */ }
183};
184
185
186struct trace_source trace_quic = {
187 .name = IST("quic"),
188 .desc = "QUIC xprt",
189 .arg_def = TRC_ARG1_QCON, /* TRACE()'s first argument is always a quic_conn */
190 .default_cb = quic_trace,
191 .known_events = quic_trace_events,
192 .lockon_args = quic_trace_lockon_args,
193 .decoding = quic_trace_decoding,
194 .report_events = ~0, /* report everything by default */
195};
196
197#define TRACE_SOURCE &trace_quic
198INITCALL1(STG_REGISTER, trace_register_source, TRACE_SOURCE);
199
200static BIO_METHOD *ha_quic_meth;
201
202DECLARE_POOL(pool_head_quic_tx_ring, "quic_tx_ring", QUIC_TX_RING_BUFSZ);
203DECLARE_POOL(pool_head_quic_conn_rxbuf, "quic_conn_rxbuf", QUIC_CONN_RX_BUFSZ);
204DECLARE_STATIC_POOL(pool_head_quic_conn_ctx,
205 "quic_conn_ctx", sizeof(struct ssl_sock_ctx));
206DECLARE_STATIC_POOL(pool_head_quic_conn, "quic_conn", sizeof(struct quic_conn));
207DECLARE_POOL(pool_head_quic_connection_id,
208 "quic_connnection_id", sizeof(struct quic_connection_id));
209DECLARE_POOL(pool_head_quic_dgram, "quic_dgram", sizeof(struct quic_dgram));
210DECLARE_POOL(pool_head_quic_rx_packet, "quic_rx_packet", sizeof(struct quic_rx_packet));
211DECLARE_POOL(pool_head_quic_tx_packet, "quic_tx_packet", sizeof(struct quic_tx_packet));
212DECLARE_STATIC_POOL(pool_head_quic_rx_crypto_frm, "quic_rx_crypto_frm", sizeof(struct quic_rx_crypto_frm));
213DECLARE_STATIC_POOL(pool_head_quic_crypto_buf, "quic_crypto_buf", sizeof(struct quic_crypto_buf));
Frédéric Lécaille7e3f7c42022-09-09 18:05:45 +0200214DECLARE_STATIC_POOL(pool_head_quic_cstream, "quic_cstream", sizeof(struct quic_cstream));
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200215DECLARE_POOL(pool_head_quic_frame, "quic_frame", sizeof(struct quic_frame));
216DECLARE_STATIC_POOL(pool_head_quic_arng, "quic_arng", sizeof(struct quic_arng_node));
217
218static struct quic_tx_packet *qc_build_pkt(unsigned char **pos, const unsigned char *buf_end,
219 struct quic_enc_level *qel, struct quic_tls_ctx *ctx,
220 struct list *frms, struct quic_conn *qc,
221 const struct quic_version *ver, size_t dglen, int pkt_type,
222 int force_ack, int padding, int probe, int cc, int *err);
223struct task *quic_conn_app_io_cb(struct task *t, void *context, unsigned int state);
224static void qc_idle_timer_do_rearm(struct quic_conn *qc);
225static void qc_idle_timer_rearm(struct quic_conn *qc, int read);
226static int qc_conn_alloc_ssl_ctx(struct quic_conn *qc);
227static int quic_conn_init_timer(struct quic_conn *qc);
228static int quic_conn_init_idle_timer_task(struct quic_conn *qc);
229
230/* Only for debug purpose */
231struct enc_debug_info {
232 unsigned char *payload;
233 size_t payload_len;
234 unsigned char *aad;
235 size_t aad_len;
236 uint64_t pn;
237};
238
239/* Initializes a enc_debug_info struct (only for debug purpose) */
240static inline void enc_debug_info_init(struct enc_debug_info *edi,
241 unsigned char *payload, size_t payload_len,
242 unsigned char *aad, size_t aad_len, uint64_t pn)
243{
244 edi->payload = payload;
245 edi->payload_len = payload_len;
246 edi->aad = aad;
247 edi->aad_len = aad_len;
248 edi->pn = pn;
249}
250
251/* Trace callback for QUIC.
252 * These traces always expect that arg1, if non-null, is of type connection.
253 */
254static void quic_trace(enum trace_level level, uint64_t mask, const struct trace_source *src,
255 const struct ist where, const struct ist func,
256 const void *a1, const void *a2, const void *a3, const void *a4)
257{
258 const struct quic_conn *qc = a1;
259
260 if (qc) {
261 const struct quic_tls_ctx *tls_ctx;
262
263 chunk_appendf(&trace_buf, " : qc@%p", qc);
264 if (mask & QUIC_EV_CONN_INIT) {
265 chunk_appendf(&trace_buf, "\n odcid");
266 quic_cid_dump(&trace_buf, &qc->odcid);
267 chunk_appendf(&trace_buf, "\n dcid");
268 quic_cid_dump(&trace_buf, &qc->dcid);
269 chunk_appendf(&trace_buf, "\n scid");
270 quic_cid_dump(&trace_buf, &qc->scid);
271 }
272
273 if (mask & QUIC_EV_TRANSP_PARAMS) {
274 const struct quic_transport_params *p = a2;
Frédéric Lécaille0aa79952023-02-03 16:15:08 +0100275
276 if (p)
277 quic_transport_params_dump(&trace_buf, qc, p);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200278 }
279
280 if (mask & QUIC_EV_CONN_ADDDATA) {
281 const enum ssl_encryption_level_t *level = a2;
282 const size_t *len = a3;
283
284 if (level) {
285 enum quic_tls_enc_level lvl = ssl_to_quic_enc_level(*level);
286
287 chunk_appendf(&trace_buf, " el=%c(%d)", quic_enc_level_char(lvl), lvl);
288 }
289 if (len)
290 chunk_appendf(&trace_buf, " len=%llu", (unsigned long long)*len);
291 }
292 if ((mask & QUIC_EV_CONN_ISEC) && qc) {
293 /* Initial read & write secrets. */
294 enum quic_tls_enc_level level = QUIC_TLS_ENC_LEVEL_INITIAL;
295 const unsigned char *rx_sec = a2;
296 const unsigned char *tx_sec = a3;
297
298 tls_ctx = &qc->els[level].tls_ctx;
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200299 chunk_appendf(&trace_buf, "\n RX el=%c", quic_enc_level_char(level));
300 if (rx_sec)
301 quic_tls_secret_hexdump(&trace_buf, rx_sec, 32);
302 quic_tls_keys_hexdump(&trace_buf, &tls_ctx->rx);
303 chunk_appendf(&trace_buf, "\n TX el=%c", quic_enc_level_char(level));
304 if (tx_sec)
305 quic_tls_secret_hexdump(&trace_buf, tx_sec, 32);
306 quic_tls_keys_hexdump(&trace_buf, &tls_ctx->tx);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200307 }
308 if (mask & (QUIC_EV_CONN_RSEC|QUIC_EV_CONN_RWSEC)) {
309 const enum ssl_encryption_level_t *level = a2;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200310
311 if (level) {
312 enum quic_tls_enc_level lvl = ssl_to_quic_enc_level(*level);
313
314 chunk_appendf(&trace_buf, "\n RX el=%c", quic_enc_level_char(lvl));
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200315 if (quic_tls_has_rx_sec(&qc->els[lvl])) {
316 tls_ctx = &qc->els[lvl].tls_ctx;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200317 quic_tls_keys_hexdump(&trace_buf, &tls_ctx->rx);
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200318 }
319 else
320 chunk_appendf(&trace_buf, " (none)");
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200321 }
322 }
323
324 if (mask & (QUIC_EV_CONN_WSEC|QUIC_EV_CONN_RWSEC)) {
325 const enum ssl_encryption_level_t *level = a2;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200326
327 if (level) {
328 enum quic_tls_enc_level lvl = ssl_to_quic_enc_level(*level);
329
330 chunk_appendf(&trace_buf, "\n TX el=%c", quic_enc_level_char(lvl));
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200331 if (quic_tls_has_tx_sec(&qc->els[lvl])) {
332 tls_ctx = &qc->els[lvl].tls_ctx;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200333 quic_tls_keys_hexdump(&trace_buf, &tls_ctx->tx);
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200334 }
335 else
336 chunk_appendf(&trace_buf, " (none)");
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200337 }
338
339 }
340
341 if (mask & QUIC_EV_CONN_FRMLIST) {
342 const struct list *l = a2;
343
344 if (l) {
345 const struct quic_frame *frm;
346 list_for_each_entry(frm, l, list) {
347 chunk_appendf(&trace_buf, " frm@%p", frm);
348 chunk_frm_appendf(&trace_buf, frm);
349 }
350 }
351 }
352
353 if (mask & (QUIC_EV_CONN_TXPKT|QUIC_EV_CONN_PAPKT)) {
354 const struct quic_tx_packet *pkt = a2;
355 const struct quic_enc_level *qel = a3;
356 const ssize_t *room = a4;
357
358 if (qel) {
359 const struct quic_pktns *pktns = qel->pktns;
360 chunk_appendf(&trace_buf, " qel=%c cwnd=%llu ppif=%lld pif=%llu "
361 "if=%llu pp=%u",
362 quic_enc_level_char_from_qel(qel, qc),
363 (unsigned long long)qc->path->cwnd,
364 (unsigned long long)qc->path->prep_in_flight,
365 (unsigned long long)qc->path->in_flight,
366 (unsigned long long)pktns->tx.in_flight,
367 pktns->tx.pto_probe);
368 }
369 if (pkt) {
370 const struct quic_frame *frm;
371 if (pkt->pn_node.key != (uint64_t)-1)
372 chunk_appendf(&trace_buf, " pn=%llu",(ull)pkt->pn_node.key);
373 list_for_each_entry(frm, &pkt->frms, list) {
374 chunk_appendf(&trace_buf, " frm@%p", frm);
375 chunk_frm_appendf(&trace_buf, frm);
376 }
377 }
378
379 if (room) {
380 chunk_appendf(&trace_buf, " room=%lld", (long long)*room);
381 chunk_appendf(&trace_buf, " dcid.len=%llu scid.len=%llu",
382 (unsigned long long)qc->dcid.len, (unsigned long long)qc->scid.len);
383 }
384 }
385
386 if (mask & QUIC_EV_CONN_IO_CB) {
387 const enum quic_handshake_state *state = a2;
388 const int *err = a3;
389
390 if (state)
391 chunk_appendf(&trace_buf, " state=%s", quic_hdshk_state_str(*state));
392 if (err)
393 chunk_appendf(&trace_buf, " err=%s", ssl_error_str(*err));
394 }
395
396 if (mask & (QUIC_EV_CONN_TRMHP|QUIC_EV_CONN_ELRMHP|QUIC_EV_CONN_SPKT)) {
397 const struct quic_rx_packet *pkt = a2;
398 const unsigned long *pktlen = a3;
399 const SSL *ssl = a4;
400
401 if (pkt) {
402 chunk_appendf(&trace_buf, " pkt@%p", pkt);
403 if (pkt->type == QUIC_PACKET_TYPE_SHORT && pkt->data)
404 chunk_appendf(&trace_buf, " kp=%d",
405 !!(*pkt->data & QUIC_PACKET_KEY_PHASE_BIT));
406 chunk_appendf(&trace_buf, " el=%c",
407 quic_packet_type_enc_level_char(pkt->type));
408 if (pkt->pnl)
409 chunk_appendf(&trace_buf, " pnl=%u pn=%llu", pkt->pnl,
410 (unsigned long long)pkt->pn);
411 if (pkt->token_len)
412 chunk_appendf(&trace_buf, " toklen=%llu",
413 (unsigned long long)pkt->token_len);
414 if (pkt->aad_len)
415 chunk_appendf(&trace_buf, " aadlen=%llu",
416 (unsigned long long)pkt->aad_len);
417 chunk_appendf(&trace_buf, " flags=0x%x len=%llu",
418 pkt->flags, (unsigned long long)pkt->len);
419 }
420 if (pktlen)
421 chunk_appendf(&trace_buf, " (%ld)", *pktlen);
422 if (ssl) {
423 enum ssl_encryption_level_t level = SSL_quic_read_level(ssl);
424 chunk_appendf(&trace_buf, " el=%c",
425 quic_enc_level_char(ssl_to_quic_enc_level(level)));
426 }
427 }
428
429 if (mask & (QUIC_EV_CONN_RXPKT|QUIC_EV_CONN_PRSHPKT|QUIC_EV_CONN_SSLDATA)) {
430 const struct quic_rx_packet *pkt = a2;
431 const struct quic_rx_crypto_frm *cf = a3;
432 const SSL *ssl = a4;
433
434 if (pkt)
435 chunk_appendf(&trace_buf, " pkt@%p el=%c pn=%llu", pkt,
436 quic_packet_type_enc_level_char(pkt->type),
437 (unsigned long long)pkt->pn);
438 if (cf)
439 chunk_appendf(&trace_buf, " cfoff=%llu cflen=%llu",
440 (unsigned long long)cf->offset_node.key,
441 (unsigned long long)cf->len);
442 if (ssl) {
443 enum ssl_encryption_level_t level = SSL_quic_read_level(ssl);
444 chunk_appendf(&trace_buf, " rel=%c",
445 quic_enc_level_char(ssl_to_quic_enc_level(level)));
446 }
447
448 if (qc->err.code)
449 chunk_appendf(&trace_buf, " err_code=0x%llx", (ull)qc->err.code);
450 }
451
452 if (mask & (QUIC_EV_CONN_PRSFRM|QUIC_EV_CONN_BFRM)) {
453 const struct quic_frame *frm = a2;
454
455 if (frm)
456 chunk_appendf(&trace_buf, " %s", quic_frame_type_string(frm->type));
457 }
458
459 if (mask & QUIC_EV_CONN_PHPKTS) {
460 const struct quic_enc_level *qel = a2;
461
462 if (qel) {
463 const struct quic_pktns *pktns = qel->pktns;
464 chunk_appendf(&trace_buf,
Amaury Denoyelle2f668f02022-11-18 15:24:08 +0100465 " qel=%c state=%s ack?%d cwnd=%llu ppif=%lld pif=%llu if=%llu pp=%u off=%llu",
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200466 quic_enc_level_char_from_qel(qel, qc),
467 quic_hdshk_state_str(qc->state),
468 !!(qel->pktns->flags & QUIC_FL_PKTNS_ACK_REQUIRED),
469 (unsigned long long)qc->path->cwnd,
470 (unsigned long long)qc->path->prep_in_flight,
471 (unsigned long long)qc->path->in_flight,
472 (unsigned long long)pktns->tx.in_flight,
Amaury Denoyelle2f668f02022-11-18 15:24:08 +0100473 pktns->tx.pto_probe,
474 qel->cstream ? (unsigned long long)qel->cstream->rx.offset : 0);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200475 }
476 }
477
478 if (mask & QUIC_EV_CONN_ENCPKT) {
479 const struct enc_debug_info *edi = a2;
480
481 if (edi)
482 chunk_appendf(&trace_buf,
483 " payload=@%p payload_len=%llu"
484 " aad=@%p aad_len=%llu pn=%llu",
485 edi->payload, (unsigned long long)edi->payload_len,
486 edi->aad, (unsigned long long)edi->aad_len,
487 (unsigned long long)edi->pn);
488 }
489
490 if (mask & QUIC_EV_CONN_RMHP) {
491 const struct quic_rx_packet *pkt = a2;
492
493 if (pkt) {
494 const int *ret = a3;
495
496 chunk_appendf(&trace_buf, " pkt@%p", pkt);
497 if (ret && *ret)
498 chunk_appendf(&trace_buf, " pnl=%u pn=%llu",
499 pkt->pnl, (unsigned long long)pkt->pn);
500 }
501 }
502
503 if (mask & QUIC_EV_CONN_PRSAFRM) {
504 const struct quic_frame *frm = a2;
505 const unsigned long *val1 = a3;
506 const unsigned long *val2 = a4;
507
508 if (frm) {
509 chunk_appendf(&trace_buf, " frm@%p", frm);
510 chunk_frm_appendf(&trace_buf, frm);
511 }
512 if (val1)
513 chunk_appendf(&trace_buf, " %lu", *val1);
514 if (val2)
515 chunk_appendf(&trace_buf, "..%lu", *val2);
516 }
517
518 if (mask & QUIC_EV_CONN_ACKSTRM) {
519 const struct quic_stream *s = a2;
520 const struct qc_stream_desc *stream = a3;
521
522 if (s)
523 chunk_appendf(&trace_buf, " off=%llu len=%llu", (ull)s->offset.key, (ull)s->len);
524 if (stream)
525 chunk_appendf(&trace_buf, " ack_offset=%llu", (ull)stream->ack_offset);
526 }
527
528 if (mask & QUIC_EV_CONN_RTTUPDT) {
529 const unsigned int *rtt_sample = a2;
530 const unsigned int *ack_delay = a3;
531 const struct quic_loss *ql = a4;
532
533 if (rtt_sample)
534 chunk_appendf(&trace_buf, " rtt_sample=%ums", *rtt_sample);
535 if (ack_delay)
536 chunk_appendf(&trace_buf, " ack_delay=%ums", *ack_delay);
537 if (ql)
538 chunk_appendf(&trace_buf,
539 " srtt=%ums rttvar=%ums min_rtt=%ums",
540 ql->srtt >> 3, ql->rtt_var >> 2, ql->rtt_min);
541 }
542 if (mask & QUIC_EV_CONN_CC) {
543 const struct quic_cc_event *ev = a2;
544 const struct quic_cc *cc = a3;
545
546 if (a2)
547 quic_cc_event_trace(&trace_buf, ev);
548 if (a3)
549 quic_cc_state_trace(&trace_buf, cc);
550 }
551
552 if (mask & QUIC_EV_CONN_PKTLOSS) {
553 const struct quic_pktns *pktns = a2;
554 const struct list *lost_pkts = a3;
555
556 if (pktns) {
557 chunk_appendf(&trace_buf, " pktns=%s",
558 pktns == &qc->pktns[QUIC_TLS_PKTNS_INITIAL] ? "I" :
559 pktns == &qc->pktns[QUIC_TLS_PKTNS_01RTT] ? "01RTT": "H");
560 if (pktns->tx.loss_time)
561 chunk_appendf(&trace_buf, " loss_time=%dms",
562 TICKS_TO_MS(tick_remain(now_ms, pktns->tx.loss_time)));
563 }
564 if (lost_pkts && !LIST_ISEMPTY(lost_pkts)) {
565 struct quic_tx_packet *pkt;
566
567 chunk_appendf(&trace_buf, " lost_pkts:");
568 list_for_each_entry(pkt, lost_pkts, list)
569 chunk_appendf(&trace_buf, " %lu", (unsigned long)pkt->pn_node.key);
570 }
571 }
572
573 if (mask & (QUIC_EV_CONN_STIMER|QUIC_EV_CONN_PTIMER|QUIC_EV_CONN_SPTO)) {
574 const struct quic_pktns *pktns = a2;
575 const int *duration = a3;
576 const uint64_t *ifae_pkts = a4;
577
578 if (ifae_pkts)
579 chunk_appendf(&trace_buf, " ifae_pkts=%llu",
580 (unsigned long long)*ifae_pkts);
581 if (pktns) {
582 chunk_appendf(&trace_buf, " pktns=%s pp=%d",
583 pktns == &qc->pktns[QUIC_TLS_PKTNS_INITIAL] ? "I" :
584 pktns == &qc->pktns[QUIC_TLS_PKTNS_01RTT] ? "01RTT": "H",
585 pktns->tx.pto_probe);
586 if (mask & (QUIC_EV_CONN_STIMER|QUIC_EV_CONN_SPTO)) {
587 if (pktns->tx.in_flight)
588 chunk_appendf(&trace_buf, " if=%llu", (ull)pktns->tx.in_flight);
589 if (pktns->tx.loss_time)
590 chunk_appendf(&trace_buf, " loss_time=%dms",
591 TICKS_TO_MS(pktns->tx.loss_time - now_ms));
592 }
593 if (mask & QUIC_EV_CONN_SPTO) {
594 if (pktns->tx.time_of_last_eliciting)
595 chunk_appendf(&trace_buf, " tole=%dms",
596 TICKS_TO_MS(pktns->tx.time_of_last_eliciting - now_ms));
597 if (duration)
598 chunk_appendf(&trace_buf, " dur=%dms", TICKS_TO_MS(*duration));
599 }
600 }
601
602 if (!(mask & (QUIC_EV_CONN_SPTO|QUIC_EV_CONN_PTIMER)) && qc->timer_task) {
603 chunk_appendf(&trace_buf,
604 " expire=%dms", TICKS_TO_MS(qc->timer - now_ms));
605 }
606 }
607
608 if (mask & QUIC_EV_CONN_SPPKTS) {
609 const struct quic_tx_packet *pkt = a2;
610
611 chunk_appendf(&trace_buf, " cwnd=%llu ppif=%llu pif=%llu",
612 (unsigned long long)qc->path->cwnd,
613 (unsigned long long)qc->path->prep_in_flight,
614 (unsigned long long)qc->path->in_flight);
615 if (pkt) {
616 const struct quic_frame *frm;
617 chunk_appendf(&trace_buf, " pn=%lu(%s) iflen=%llu",
618 (unsigned long)pkt->pn_node.key,
619 pkt->pktns == &qc->pktns[QUIC_TLS_PKTNS_INITIAL] ? "I" :
620 pkt->pktns == &qc->pktns[QUIC_TLS_PKTNS_01RTT] ? "01RTT": "H",
621 (unsigned long long)pkt->in_flight_len);
622 chunk_appendf(&trace_buf, " rx.bytes=%llu tx.bytes=%llu",
623 (unsigned long long)qc->rx.bytes,
624 (unsigned long long)qc->tx.bytes);
625 list_for_each_entry(frm, &pkt->frms, list) {
626 chunk_appendf(&trace_buf, " frm@%p", frm);
627 chunk_frm_appendf(&trace_buf, frm);
628 }
629 }
630 }
631
632 if (mask & QUIC_EV_CONN_SSLALERT) {
633 const uint8_t *alert = a2;
634 const enum ssl_encryption_level_t *level = a3;
635
636 if (alert)
637 chunk_appendf(&trace_buf, " alert=0x%02x", *alert);
638 if (level)
639 chunk_appendf(&trace_buf, " el=%c",
640 quic_enc_level_char(ssl_to_quic_enc_level(*level)));
641 }
642
643 if (mask & QUIC_EV_CONN_BCFRMS) {
644 const size_t *sz1 = a2;
645 const size_t *sz2 = a3;
646 const size_t *sz3 = a4;
647
648 if (sz1)
649 chunk_appendf(&trace_buf, " %llu", (unsigned long long)*sz1);
650 if (sz2)
651 chunk_appendf(&trace_buf, " %llu", (unsigned long long)*sz2);
652 if (sz3)
653 chunk_appendf(&trace_buf, " %llu", (unsigned long long)*sz3);
654 }
655
656 if (mask & QUIC_EV_CONN_PSTRM) {
657 const struct quic_frame *frm = a2;
658
659 if (frm) {
660 chunk_appendf(&trace_buf, " frm@%p", frm);
661 chunk_frm_appendf(&trace_buf, frm);
662 }
663 }
Frédéric Lécaille4aa7d812022-09-16 10:15:58 +0200664
665 if (mask & QUIC_EV_CONN_ELEVELSEL) {
666 const enum quic_handshake_state *state = a2;
667 const enum quic_tls_enc_level *level = a3;
668 const enum quic_tls_enc_level *next_level = a4;
669
670 if (state)
671 chunk_appendf(&trace_buf, " state=%s", quic_hdshk_state_str(qc->state));
672 if (level)
673 chunk_appendf(&trace_buf, " level=%c", quic_enc_level_char(*level));
674 if (next_level)
675 chunk_appendf(&trace_buf, " next_level=%c", quic_enc_level_char(*next_level));
676
677 }
Amaury Denoyelle5b414862022-10-24 17:40:37 +0200678
679 if (mask & QUIC_EV_CONN_RCV) {
680 const struct quic_dgram *dgram = a2;
681
682 if (dgram)
683 chunk_appendf(&trace_buf, " dgram.len=%zu", dgram->len);
684 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200685 }
686 if (mask & QUIC_EV_CONN_LPKT) {
687 const struct quic_rx_packet *pkt = a2;
688 const uint64_t *len = a3;
689 const struct quic_version *ver = a4;
690
691 if (pkt) {
692 chunk_appendf(&trace_buf, " pkt@%p type=0x%02x %s",
693 pkt, pkt->type, qc_pkt_long(pkt) ? "long" : "short");
694 if (pkt->pn_node.key != (uint64_t)-1)
695 chunk_appendf(&trace_buf, " pn=%llu", pkt->pn_node.key);
696 }
697
698 if (len)
699 chunk_appendf(&trace_buf, " len=%llu", (ull)*len);
700
701 if (ver)
702 chunk_appendf(&trace_buf, " ver=0x%08x", ver->num);
703 }
704
705 if (mask & QUIC_EV_STATELESS_RST) {
706 const struct quic_cid *cid = a2;
707
708 if (cid)
709 quic_cid_dump(&trace_buf, cid);
710 }
711
712}
713
714/* Returns 1 if the peer has validated <qc> QUIC connection address, 0 if not. */
715static inline int quic_peer_validated_addr(struct quic_conn *qc)
716{
717 struct quic_pktns *hdshk_pktns, *app_pktns;
718
719 if (!qc_is_listener(qc))
720 return 1;
721
722 hdshk_pktns = qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE].pktns;
723 app_pktns = qc->els[QUIC_TLS_ENC_LEVEL_APP].pktns;
724 if ((hdshk_pktns->flags & QUIC_FL_PKTNS_PKT_RECEIVED) ||
725 (app_pktns->flags & QUIC_FL_PKTNS_PKT_RECEIVED) ||
726 qc->state >= QUIC_HS_ST_COMPLETE)
727 return 1;
728
729 return 0;
730}
731
Frédéric Lécaille0aa79952023-02-03 16:15:08 +0100732/* To be called to kill a connection as soon as possible (without sending any packet). */
733void qc_kill_conn(struct quic_conn *qc)
734{
735 qc->flags |= QUIC_FL_CONN_TO_KILL;
736 task_wakeup(qc->idle_timer_task, TASK_WOKEN_OTHER);
737}
738
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200739/* Set the timer attached to the QUIC connection with <ctx> as I/O handler and used for
740 * both loss detection and PTO and schedule the task assiated to this timer if needed.
741 */
742static inline void qc_set_timer(struct quic_conn *qc)
743{
744 struct quic_pktns *pktns;
745 unsigned int pto;
Frédéric Lécailleb75eecc2023-01-26 15:18:17 +0100746 int handshake_confirmed;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200747
748 TRACE_ENTER(QUIC_EV_CONN_STIMER, qc,
749 NULL, NULL, &qc->path->ifae_pkts);
750
751 pktns = quic_loss_pktns(qc);
752 if (tick_isset(pktns->tx.loss_time)) {
753 qc->timer = pktns->tx.loss_time;
754 goto out;
755 }
756
757 /* anti-amplification: the timer must be
758 * cancelled for a server which reached the anti-amplification limit.
759 */
760 if (!quic_peer_validated_addr(qc) &&
761 (qc->flags & QUIC_FL_CONN_ANTI_AMPLIFICATION_REACHED)) {
762 TRACE_PROTO("anti-amplification reached", QUIC_EV_CONN_STIMER, qc);
763 qc->timer = TICK_ETERNITY;
764 goto out;
765 }
766
767 if (!qc->path->ifae_pkts && quic_peer_validated_addr(qc)) {
768 TRACE_PROTO("timer cancellation", QUIC_EV_CONN_STIMER, qc);
769 /* Timer cancellation. */
770 qc->timer = TICK_ETERNITY;
771 goto out;
772 }
773
Frédéric Lécailleb75eecc2023-01-26 15:18:17 +0100774 handshake_confirmed = qc->state >= QUIC_HS_ST_CONFIRMED;
775 pktns = quic_pto_pktns(qc, handshake_confirmed, &pto);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200776 if (tick_isset(pto))
777 qc->timer = pto;
778 out:
779 if (qc->timer_task && qc->timer != TICK_ETERNITY) {
780 if (tick_is_expired(qc->timer, now_ms)) {
781 TRACE_DEVEL("wakeup asap timer task", QUIC_EV_CONN_STIMER, qc);
782 task_wakeup(qc->timer_task, TASK_WOKEN_MSG);
783 }
784 else {
785 TRACE_DEVEL("timer task scheduling", QUIC_EV_CONN_STIMER, qc);
786 task_schedule(qc->timer_task, qc->timer);
787 }
788 }
789 TRACE_LEAVE(QUIC_EV_CONN_STIMER, qc, pktns);
790}
791
792/* Derive new keys and ivs required for Key Update feature for <qc> QUIC
793 * connection.
794 * Return 1 if succeeded, 0 if not.
795 */
796static int quic_tls_key_update(struct quic_conn *qc)
797{
798 struct quic_tls_ctx *tls_ctx = &qc->els[QUIC_TLS_ENC_LEVEL_APP].tls_ctx;
799 struct quic_tls_secrets *rx, *tx;
800 struct quic_tls_kp *nxt_rx = &qc->ku.nxt_rx;
801 struct quic_tls_kp *nxt_tx = &qc->ku.nxt_tx;
802 const struct quic_version *ver =
803 qc->negotiated_version ? qc->negotiated_version : qc->original_version;
804 int ret = 0;
805
806 TRACE_ENTER(QUIC_EV_CONN_RWSEC, qc);
807
808 tls_ctx = &qc->els[QUIC_TLS_ENC_LEVEL_APP].tls_ctx;
809 rx = &tls_ctx->rx;
810 tx = &tls_ctx->tx;
811 nxt_rx = &qc->ku.nxt_rx;
812 nxt_tx = &qc->ku.nxt_tx;
813
814 /* Prepare new RX secrets */
815 if (!quic_tls_sec_update(rx->md, ver, nxt_rx->secret, nxt_rx->secretlen,
816 rx->secret, rx->secretlen)) {
817 TRACE_ERROR("New RX secret update failed", QUIC_EV_CONN_RWSEC, qc);
818 goto leave;
819 }
820
821 if (!quic_tls_derive_keys(rx->aead, NULL, rx->md, ver,
822 nxt_rx->key, nxt_rx->keylen,
823 nxt_rx->iv, nxt_rx->ivlen, NULL, 0,
824 nxt_rx->secret, nxt_rx->secretlen)) {
825 TRACE_ERROR("New RX key derivation failed", QUIC_EV_CONN_RWSEC, qc);
826 goto leave;
827 }
828
829 /* Prepare new TX secrets */
830 if (!quic_tls_sec_update(tx->md, ver, nxt_tx->secret, nxt_tx->secretlen,
831 tx->secret, tx->secretlen)) {
832 TRACE_ERROR("New TX secret update failed", QUIC_EV_CONN_RWSEC, qc);
833 goto leave;
834 }
835
836 if (!quic_tls_derive_keys(tx->aead, NULL, tx->md, ver,
837 nxt_tx->key, nxt_tx->keylen,
838 nxt_tx->iv, nxt_tx->ivlen, NULL, 0,
839 nxt_tx->secret, nxt_tx->secretlen)) {
840 TRACE_ERROR("New TX key derivation failed", QUIC_EV_CONN_RWSEC, qc);
841 goto leave;
842 }
843
844 if (nxt_rx->ctx) {
845 EVP_CIPHER_CTX_free(nxt_rx->ctx);
846 nxt_rx->ctx = NULL;
847 }
848
849 if (!quic_tls_rx_ctx_init(&nxt_rx->ctx, tls_ctx->rx.aead, nxt_rx->key)) {
850 TRACE_ERROR("could not initial RX TLS cipher context", QUIC_EV_CONN_RWSEC, qc);
851 goto leave;
852 }
853
854 if (nxt_tx->ctx) {
855 EVP_CIPHER_CTX_free(nxt_tx->ctx);
856 nxt_tx->ctx = NULL;
857 }
858
859 if (!quic_tls_rx_ctx_init(&nxt_tx->ctx, tls_ctx->tx.aead, nxt_tx->key)) {
860 TRACE_ERROR("could not initial RX TLS cipher context", QUIC_EV_CONN_RWSEC, qc);
861 goto leave;
862 }
863
864 ret = 1;
865 leave:
866 TRACE_LEAVE(QUIC_EV_CONN_RWSEC, qc);
867 return ret;
868}
869
870/* Rotate the Key Update information for <qc> QUIC connection.
871 * Must be used after having updated them.
872 * Always succeeds.
873 */
874static void quic_tls_rotate_keys(struct quic_conn *qc)
875{
876 struct quic_tls_ctx *tls_ctx = &qc->els[QUIC_TLS_ENC_LEVEL_APP].tls_ctx;
877 unsigned char *curr_secret, *curr_iv, *curr_key;
878 EVP_CIPHER_CTX *curr_ctx;
879
880 TRACE_ENTER(QUIC_EV_CONN_RXPKT, qc);
881
882 /* Rotate the RX secrets */
883 curr_ctx = tls_ctx->rx.ctx;
884 curr_secret = tls_ctx->rx.secret;
885 curr_iv = tls_ctx->rx.iv;
886 curr_key = tls_ctx->rx.key;
887
888 tls_ctx->rx.ctx = qc->ku.nxt_rx.ctx;
889 tls_ctx->rx.secret = qc->ku.nxt_rx.secret;
890 tls_ctx->rx.iv = qc->ku.nxt_rx.iv;
891 tls_ctx->rx.key = qc->ku.nxt_rx.key;
892
893 qc->ku.nxt_rx.ctx = qc->ku.prv_rx.ctx;
894 qc->ku.nxt_rx.secret = qc->ku.prv_rx.secret;
895 qc->ku.nxt_rx.iv = qc->ku.prv_rx.iv;
896 qc->ku.nxt_rx.key = qc->ku.prv_rx.key;
897
898 qc->ku.prv_rx.ctx = curr_ctx;
899 qc->ku.prv_rx.secret = curr_secret;
900 qc->ku.prv_rx.iv = curr_iv;
901 qc->ku.prv_rx.key = curr_key;
902 qc->ku.prv_rx.pn = tls_ctx->rx.pn;
903
904 /* Update the TX secrets */
905 curr_ctx = tls_ctx->tx.ctx;
906 curr_secret = tls_ctx->tx.secret;
907 curr_iv = tls_ctx->tx.iv;
908 curr_key = tls_ctx->tx.key;
909
910 tls_ctx->tx.ctx = qc->ku.nxt_tx.ctx;
911 tls_ctx->tx.secret = qc->ku.nxt_tx.secret;
912 tls_ctx->tx.iv = qc->ku.nxt_tx.iv;
913 tls_ctx->tx.key = qc->ku.nxt_tx.key;
914
915 qc->ku.nxt_tx.ctx = curr_ctx;
916 qc->ku.nxt_tx.secret = curr_secret;
917 qc->ku.nxt_tx.iv = curr_iv;
918 qc->ku.nxt_tx.key = curr_key;
919
920 TRACE_LEAVE(QUIC_EV_CONN_RXPKT, qc);
921}
922
923/* returns 0 on error, 1 on success */
924int ha_quic_set_encryption_secrets(SSL *ssl, enum ssl_encryption_level_t level,
925 const uint8_t *read_secret,
926 const uint8_t *write_secret, size_t secret_len)
927{
928 struct quic_conn *qc = SSL_get_ex_data(ssl, ssl_qc_app_data_index);
929 struct quic_tls_ctx *tls_ctx = &qc->els[ssl_to_quic_enc_level(level)].tls_ctx;
930 const SSL_CIPHER *cipher = SSL_get_current_cipher(ssl);
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200931 struct quic_tls_secrets *rx = NULL, *tx = NULL;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200932 const struct quic_version *ver =
933 qc->negotiated_version ? qc->negotiated_version : qc->original_version;
934 int ret = 0;
935
936 TRACE_ENTER(QUIC_EV_CONN_RWSEC, qc);
937 BUG_ON(secret_len > QUIC_TLS_SECRET_LEN);
Frédéric Lécaille0aa79952023-02-03 16:15:08 +0100938
939 if (qc->flags & QUIC_FL_CONN_TO_KILL) {
940 TRACE_PROTO("connection to be killed", QUIC_EV_CONN_ADDDATA, qc);
941 goto out;
942 }
943
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200944 if (qc->flags & QUIC_FL_CONN_IMMEDIATE_CLOSE) {
945 TRACE_PROTO("CC required", QUIC_EV_CONN_RWSEC, qc);
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200946 goto out;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200947 }
948
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200949 if (!read_secret)
950 goto write;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200951
952 rx = &tls_ctx->rx;
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200953 if (!quic_tls_secrets_keys_alloc(rx)) {
954 TRACE_ERROR("RX keys allocation failed", QUIC_EV_CONN_RWSEC, qc);
955 goto leave;
956 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200957
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200958 rx->aead = tls_aead(cipher);
959 rx->md = tls_md(cipher);
960 rx->hp = tls_hp(cipher);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200961
962 if (!quic_tls_derive_keys(rx->aead, rx->hp, rx->md, ver, rx->key, rx->keylen,
963 rx->iv, rx->ivlen, rx->hp_key, sizeof rx->hp_key,
964 read_secret, secret_len)) {
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200965 TRACE_ERROR("TX key derivation failed", QUIC_EV_CONN_RWSEC, qc);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +0200966 goto leave;
967 }
968
969 if (!quic_tls_rx_ctx_init(&rx->ctx, rx->aead, rx->key)) {
970 TRACE_ERROR("could not initial RX TLS cipher context", QUIC_EV_CONN_RWSEC, qc);
971 goto leave;
972 }
973
974 if (!quic_tls_dec_aes_ctx_init(&rx->hp_ctx, rx->hp, rx->hp_key)) {
975 TRACE_ERROR("could not initial RX TLS cipher context for HP", QUIC_EV_CONN_RWSEC, qc);
976 goto leave;
977 }
978
979 /* Enqueue this connection asap if we could derive O-RTT secrets as
980 * listener. Note that a listener derives only RX secrets for this
981 * level.
982 */
983 if (qc_is_listener(qc) && level == ssl_encryption_early_data) {
984 TRACE_DEVEL("pushing connection into accept queue", QUIC_EV_CONN_RWSEC, qc);
985 quic_accept_push_qc(qc);
986 }
987
988write:
989
990 if (!write_secret)
991 goto out;
992
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +0200993 tx = &tls_ctx->tx;
994 if (!quic_tls_secrets_keys_alloc(tx)) {
995 TRACE_ERROR("TX keys allocation failed", QUIC_EV_CONN_RWSEC, qc);
996 goto leave;
997 }
998
999 tx->aead = tls_aead(cipher);
1000 tx->md = tls_md(cipher);
1001 tx->hp = tls_hp(cipher);
1002
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001003 if (!quic_tls_derive_keys(tx->aead, tx->hp, tx->md, ver, tx->key, tx->keylen,
1004 tx->iv, tx->ivlen, tx->hp_key, sizeof tx->hp_key,
1005 write_secret, secret_len)) {
1006 TRACE_ERROR("TX key derivation failed", QUIC_EV_CONN_RWSEC, qc);
1007 goto leave;
1008 }
1009
1010 if (!quic_tls_tx_ctx_init(&tx->ctx, tx->aead, tx->key)) {
1011 TRACE_ERROR("could not initial RX TLS cipher context", QUIC_EV_CONN_RWSEC, qc);
1012 goto leave;
1013 }
1014
1015 if (!quic_tls_enc_aes_ctx_init(&tx->hp_ctx, tx->hp, tx->hp_key)) {
1016 TRACE_ERROR("could not initial TX TLS cipher context for HP", QUIC_EV_CONN_RWSEC, qc);
1017 goto leave;
1018 }
1019
Frédéric Lécailleaf25a692023-02-01 17:56:57 +01001020 if (level == ssl_encryption_handshake && qc_is_listener(qc)) {
1021 qc->enc_params_len =
1022 quic_transport_params_encode(qc->enc_params,
1023 qc->enc_params + sizeof qc->enc_params,
1024 &qc->rx.params, ver, 1);
1025 if (!qc->enc_params_len) {
1026 TRACE_ERROR("quic_transport_params_encode() failed", QUIC_EV_CONN_RWSEC);
1027 goto leave;
1028 }
1029
1030 if (!SSL_set_quic_transport_params(qc->xprt_ctx->ssl, qc->enc_params, qc->enc_params_len)) {
1031 TRACE_ERROR("SSL_set_quic_transport_params() failed", QUIC_EV_CONN_RWSEC);
1032 goto leave;
1033 }
1034 }
1035
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001036 if (level == ssl_encryption_application) {
1037 struct quic_tls_kp *prv_rx = &qc->ku.prv_rx;
1038 struct quic_tls_kp *nxt_rx = &qc->ku.nxt_rx;
1039 struct quic_tls_kp *nxt_tx = &qc->ku.nxt_tx;
1040
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02001041 if (rx) {
1042 if (!(rx->secret = pool_alloc(pool_head_quic_tls_secret))) {
1043 TRACE_ERROR("Could not allocate RX Application secrete keys", QUIC_EV_CONN_RWSEC, qc);
1044 goto leave;
1045 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001046
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001047 memcpy(rx->secret, read_secret, secret_len);
1048 rx->secretlen = secret_len;
1049 }
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02001050
1051 if (tx) {
1052 if (!(tx->secret = pool_alloc(pool_head_quic_tls_secret))) {
1053 TRACE_ERROR("Could not allocate TX Application secrete keys", QUIC_EV_CONN_RWSEC, qc);
1054 goto leave;
1055 }
1056
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001057 memcpy(tx->secret, write_secret, secret_len);
1058 tx->secretlen = secret_len;
1059 }
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02001060
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001061 /* Initialize all the secret keys lengths */
1062 prv_rx->secretlen = nxt_rx->secretlen = nxt_tx->secretlen = secret_len;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001063 }
1064
1065 out:
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001066 ret = 1;
1067 leave:
1068 TRACE_LEAVE(QUIC_EV_CONN_RWSEC, qc, &level);
1069 return ret;
1070}
1071
1072/* This function copies the CRYPTO data provided by the TLS stack found at <data>
1073 * with <len> as size in CRYPTO buffers dedicated to store the information about
1074 * outgoing CRYPTO frames so that to be able to replay the CRYPTO data streams.
1075 * It fails (returns 0) only if it could not managed to allocate enough CRYPTO
1076 * buffers to store all the data.
1077 * Note that CRYPTO data may exist at any encryption level except at 0-RTT.
1078 */
1079static int quic_crypto_data_cpy(struct quic_conn *qc, struct quic_enc_level *qel,
1080 const unsigned char *data, size_t len)
1081{
1082 struct quic_crypto_buf **qcb;
1083 /* The remaining byte to store in CRYPTO buffers. */
1084 size_t cf_offset, cf_len, *nb_buf;
1085 unsigned char *pos;
1086 int ret = 0;
1087
1088 nb_buf = &qel->tx.crypto.nb_buf;
1089 qcb = &qel->tx.crypto.bufs[*nb_buf - 1];
1090 cf_offset = (*nb_buf - 1) * QUIC_CRYPTO_BUF_SZ + (*qcb)->sz;
1091 cf_len = len;
1092
1093 TRACE_ENTER(QUIC_EV_CONN_ADDDATA, qc);
1094
1095 while (len) {
1096 size_t to_copy, room;
1097
1098 pos = (*qcb)->data + (*qcb)->sz;
1099 room = QUIC_CRYPTO_BUF_SZ - (*qcb)->sz;
1100 to_copy = len > room ? room : len;
1101 if (to_copy) {
1102 memcpy(pos, data, to_copy);
1103 /* Increment the total size of this CRYPTO buffers by <to_copy>. */
1104 qel->tx.crypto.sz += to_copy;
1105 (*qcb)->sz += to_copy;
1106 len -= to_copy;
1107 data += to_copy;
1108 }
1109 else {
1110 struct quic_crypto_buf **tmp;
1111
1112 // FIXME: realloc!
1113 tmp = realloc(qel->tx.crypto.bufs,
1114 (*nb_buf + 1) * sizeof *qel->tx.crypto.bufs);
1115 if (tmp) {
1116 qel->tx.crypto.bufs = tmp;
1117 qcb = &qel->tx.crypto.bufs[*nb_buf];
1118 *qcb = pool_alloc(pool_head_quic_crypto_buf);
1119 if (!*qcb) {
1120 TRACE_ERROR("Could not allocate crypto buf", QUIC_EV_CONN_ADDDATA, qc);
1121 goto leave;
1122 }
1123
1124 (*qcb)->sz = 0;
1125 ++*nb_buf;
1126 }
1127 else {
1128 break;
1129 }
1130 }
1131 }
1132
1133 /* Allocate a TX CRYPTO frame only if all the CRYPTO data
1134 * have been buffered.
1135 */
1136 if (!len) {
1137 struct quic_frame *frm;
1138 struct quic_frame *found = NULL;
1139
1140 /* There is at most one CRYPTO frame in this packet number
1141 * space. Let's look for it.
1142 */
1143 list_for_each_entry(frm, &qel->pktns->tx.frms, list) {
1144 if (frm->type != QUIC_FT_CRYPTO)
1145 continue;
1146
1147 /* Found */
1148 found = frm;
1149 break;
1150 }
1151
1152 if (found) {
1153 found->crypto.len += cf_len;
1154 }
1155 else {
Amaury Denoyelle40c24f12023-01-27 17:47:49 +01001156 frm = qc_frm_alloc(QUIC_FT_CRYPTO);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001157 if (!frm) {
1158 TRACE_ERROR("Could not allocate quic frame", QUIC_EV_CONN_ADDDATA, qc);
1159 goto leave;
1160 }
1161
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001162 frm->crypto.offset = cf_offset;
1163 frm->crypto.len = cf_len;
1164 frm->crypto.qel = qel;
1165 LIST_APPEND(&qel->pktns->tx.frms, &frm->list);
1166 }
1167 }
1168 ret = len == 0;
1169 leave:
1170 TRACE_LEAVE(QUIC_EV_CONN_ADDDATA, qc);
1171 return ret;
1172}
1173
1174/* Prepare the emission of CONNECTION_CLOSE with error <err>. All send/receive
1175 * activity for <qc> will be interrupted.
1176 */
1177void quic_set_connection_close(struct quic_conn *qc, const struct quic_err err)
1178{
1179 TRACE_ENTER(QUIC_EV_CONN_CLOSE, qc);
1180 if (qc->flags & QUIC_FL_CONN_IMMEDIATE_CLOSE)
1181 goto leave;
1182
1183 TRACE_STATE("setting immediate close", QUIC_EV_CONN_CLOSE, qc);
1184 qc->flags |= QUIC_FL_CONN_IMMEDIATE_CLOSE;
1185 qc->err.code = err.code;
1186 qc->err.app = err.app;
1187 leave:
1188 TRACE_LEAVE(QUIC_EV_CONN_CLOSE, qc);
1189}
1190
1191/* Set <alert> TLS alert as QUIC CRYPTO_ERROR error */
1192void quic_set_tls_alert(struct quic_conn *qc, int alert)
1193{
1194 TRACE_ENTER(QUIC_EV_CONN_SSLALERT, qc);
1195
1196 if (!(qc->flags & QUIC_FL_CONN_HALF_OPEN_CNT_DECREMENTED)) {
1197 qc->flags |= QUIC_FL_CONN_HALF_OPEN_CNT_DECREMENTED;
1198 TRACE_DEVEL("dec half open counter", QUIC_EV_CONN_SSLALERT, qc);
1199 HA_ATOMIC_DEC(&qc->prx_counters->half_open_conn);
1200 }
1201 quic_set_connection_close(qc, quic_err_tls(alert));
1202 qc->flags |= QUIC_FL_CONN_TLS_ALERT;
1203 TRACE_STATE("Alert set", QUIC_EV_CONN_SSLALERT, qc);
1204
1205 TRACE_LEAVE(QUIC_EV_CONN_SSLALERT, qc);
1206}
1207
1208/* Set the application for <qc> QUIC connection.
1209 * Return 1 if succeeded, 0 if not.
1210 */
1211int quic_set_app_ops(struct quic_conn *qc, const unsigned char *alpn, size_t alpn_len)
1212{
1213 if (alpn_len >= 2 && memcmp(alpn, "h3", 2) == 0)
1214 qc->app_ops = &h3_ops;
1215 else if (alpn_len >= 10 && memcmp(alpn, "hq-interop", 10) == 0)
1216 qc->app_ops = &hq_interop_ops;
1217 else
1218 return 0;
1219
1220 return 1;
1221}
1222
1223/* ->add_handshake_data QUIC TLS callback used by the QUIC TLS stack when it
1224 * wants to provide the QUIC layer with CRYPTO data.
1225 * Returns 1 if succeeded, 0 if not.
1226 */
1227int ha_quic_add_handshake_data(SSL *ssl, enum ssl_encryption_level_t level,
1228 const uint8_t *data, size_t len)
1229{
1230 struct quic_conn *qc;
1231 enum quic_tls_enc_level tel;
1232 struct quic_enc_level *qel;
1233 int ret = 0;
1234
1235 qc = SSL_get_ex_data(ssl, ssl_qc_app_data_index);
1236 TRACE_ENTER(QUIC_EV_CONN_ADDDATA, qc);
1237
Frédéric Lécaille0aa79952023-02-03 16:15:08 +01001238 if (qc->flags & QUIC_FL_CONN_TO_KILL) {
1239 TRACE_PROTO("connection to be killed", QUIC_EV_CONN_ADDDATA, qc);
1240 goto out;
1241 }
1242
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001243 if (qc->flags & QUIC_FL_CONN_IMMEDIATE_CLOSE) {
1244 TRACE_PROTO("CC required", QUIC_EV_CONN_ADDDATA, qc);
1245 goto out;
1246 }
1247
1248 tel = ssl_to_quic_enc_level(level);
1249 if (tel == -1) {
1250 TRACE_ERROR("Wrong encryption level", QUIC_EV_CONN_ADDDATA, qc);
1251 goto leave;
1252 }
1253
1254 qel = &qc->els[tel];
1255 if (!quic_crypto_data_cpy(qc, qel, data, len)) {
1256 TRACE_ERROR("Could not bufferize", QUIC_EV_CONN_ADDDATA, qc);
1257 goto leave;
1258 }
1259
1260 TRACE_DEVEL("CRYPTO data buffered", QUIC_EV_CONN_ADDDATA,
1261 qc, &level, &len);
1262 out:
1263 ret = 1;
1264 leave:
1265 TRACE_LEAVE(QUIC_EV_CONN_ADDDATA, qc);
1266 return ret;
1267}
1268
1269int ha_quic_flush_flight(SSL *ssl)
1270{
1271 struct quic_conn *qc = SSL_get_ex_data(ssl, ssl_qc_app_data_index);
1272
1273 TRACE_ENTER(QUIC_EV_CONN_FFLIGHT, qc);
1274 TRACE_LEAVE(QUIC_EV_CONN_FFLIGHT, qc);
1275
1276 return 1;
1277}
1278
1279int ha_quic_send_alert(SSL *ssl, enum ssl_encryption_level_t level, uint8_t alert)
1280{
1281 struct quic_conn *qc = SSL_get_ex_data(ssl, ssl_qc_app_data_index);
1282
1283 TRACE_ENTER(QUIC_EV_CONN_SSLALERT, qc);
1284
1285 TRACE_PROTO("Received TLS alert", QUIC_EV_CONN_SSLALERT, qc, &alert, &level);
1286
1287 quic_set_tls_alert(qc, alert);
1288 TRACE_LEAVE(QUIC_EV_CONN_SSLALERT, qc);
1289 return 1;
1290}
1291
1292/* QUIC TLS methods */
1293static SSL_QUIC_METHOD ha_quic_method = {
1294 .set_encryption_secrets = ha_quic_set_encryption_secrets,
1295 .add_handshake_data = ha_quic_add_handshake_data,
1296 .flush_flight = ha_quic_flush_flight,
1297 .send_alert = ha_quic_send_alert,
1298};
1299
1300/* Initialize the TLS context of a listener with <bind_conf> as configuration.
1301 * Returns an error count.
1302 */
1303int ssl_quic_initial_ctx(struct bind_conf *bind_conf)
1304{
1305 struct ssl_bind_conf __maybe_unused *ssl_conf_cur;
1306 int cfgerr = 0;
1307
1308 long options =
1309 (SSL_OP_ALL & ~SSL_OP_DONT_INSERT_EMPTY_FRAGMENTS) |
1310 SSL_OP_SINGLE_ECDH_USE |
1311 SSL_OP_CIPHER_SERVER_PREFERENCE;
1312 SSL_CTX *ctx;
1313
1314 ctx = SSL_CTX_new(TLS_server_method());
1315 bind_conf->initial_ctx = ctx;
1316
1317 SSL_CTX_set_options(ctx, options);
1318 SSL_CTX_set_mode(ctx, SSL_MODE_RELEASE_BUFFERS);
1319 SSL_CTX_set_min_proto_version(ctx, TLS1_3_VERSION);
1320 SSL_CTX_set_max_proto_version(ctx, TLS1_3_VERSION);
1321
1322#ifdef SSL_CTRL_SET_TLSEXT_HOSTNAME
1323# if defined(HAVE_SSL_CLIENT_HELLO_CB)
1324# if defined(SSL_OP_NO_ANTI_REPLAY)
1325 if (bind_conf->ssl_conf.early_data) {
1326 SSL_CTX_set_options(ctx, SSL_OP_NO_ANTI_REPLAY);
1327 SSL_CTX_set_max_early_data(ctx, 0xffffffff);
1328 }
1329# endif /* !SSL_OP_NO_ANTI_REPLAY */
1330 SSL_CTX_set_client_hello_cb(ctx, ssl_sock_switchctx_cbk, NULL);
1331 SSL_CTX_set_tlsext_servername_callback(ctx, ssl_sock_switchctx_err_cbk);
1332# else /* ! HAVE_SSL_CLIENT_HELLO_CB */
1333 SSL_CTX_set_tlsext_servername_callback(ctx, ssl_sock_switchctx_cbk);
1334# endif
1335 SSL_CTX_set_tlsext_servername_arg(ctx, bind_conf);
1336#endif
1337 SSL_CTX_set_quic_method(ctx, &ha_quic_method);
1338
1339 return cfgerr;
1340}
1341
1342/* Decode an expected packet number from <truncated_on> its truncated value,
1343 * depending on <largest_pn> the largest received packet number, and <pn_nbits>
1344 * the number of bits used to encode this packet number (its length in bytes * 8).
1345 * See https://quicwg.org/base-drafts/draft-ietf-quic-transport.html#packet-encoding
1346 */
1347static uint64_t decode_packet_number(uint64_t largest_pn,
1348 uint32_t truncated_pn, unsigned int pn_nbits)
1349{
1350 uint64_t expected_pn = largest_pn + 1;
1351 uint64_t pn_win = (uint64_t)1 << pn_nbits;
1352 uint64_t pn_hwin = pn_win / 2;
1353 uint64_t pn_mask = pn_win - 1;
1354 uint64_t candidate_pn;
1355
1356
1357 candidate_pn = (expected_pn & ~pn_mask) | truncated_pn;
1358 /* Note that <pn_win> > <pn_hwin>. */
1359 if (candidate_pn < QUIC_MAX_PACKET_NUM - pn_win &&
1360 candidate_pn + pn_hwin <= expected_pn)
1361 return candidate_pn + pn_win;
1362
1363 if (candidate_pn > expected_pn + pn_hwin && candidate_pn >= pn_win)
1364 return candidate_pn - pn_win;
1365
1366 return candidate_pn;
1367}
1368
1369/* Remove the header protection of <pkt> QUIC packet using <tls_ctx> as QUIC TLS
1370 * cryptographic context.
1371 * <largest_pn> is the largest received packet number and <pn> the address of
1372 * the packet number field for this packet with <byte0> address of its first byte.
1373 * <end> points to one byte past the end of this packet.
1374 * Returns 1 if succeeded, 0 if not.
1375 */
1376static int qc_do_rm_hp(struct quic_conn *qc,
1377 struct quic_rx_packet *pkt, struct quic_tls_ctx *tls_ctx,
1378 int64_t largest_pn, unsigned char *pn, unsigned char *byte0)
1379{
1380 int ret, i, pnlen;
1381 uint64_t packet_number;
1382 uint32_t truncated_pn = 0;
1383 unsigned char mask[5] = {0};
1384 unsigned char *sample;
1385 EVP_CIPHER_CTX *cctx = NULL;
1386
1387 TRACE_ENTER(QUIC_EV_CONN_RMHP, qc);
1388
1389 ret = 0;
1390
1391 /* Check there is enough data in this packet. */
1392 if (pkt->len - (pn - byte0) < QUIC_PACKET_PN_MAXLEN + sizeof mask) {
1393 TRACE_PROTO("too short packet", QUIC_EV_CONN_RMHP, qc, pkt);
1394 goto leave;
1395 }
1396
1397 cctx = EVP_CIPHER_CTX_new();
1398 if (!cctx) {
1399 TRACE_ERROR("memory allocation failed", QUIC_EV_CONN_RMHP, qc, pkt);
1400 goto leave;
1401 }
1402
1403 sample = pn + QUIC_PACKET_PN_MAXLEN;
1404
1405 if (!quic_tls_aes_decrypt(mask, sample, sizeof mask, tls_ctx->rx.hp_ctx)) {
1406 TRACE_ERROR("HP removing failed", QUIC_EV_CONN_RMHP, qc, pkt);
1407 goto leave;
1408 }
1409
1410 *byte0 ^= mask[0] & (*byte0 & QUIC_PACKET_LONG_HEADER_BIT ? 0xf : 0x1f);
1411 pnlen = (*byte0 & QUIC_PACKET_PNL_BITMASK) + 1;
1412 for (i = 0; i < pnlen; i++) {
1413 pn[i] ^= mask[i + 1];
1414 truncated_pn = (truncated_pn << 8) | pn[i];
1415 }
1416
1417 packet_number = decode_packet_number(largest_pn, truncated_pn, pnlen * 8);
1418 /* Store remaining information for this unprotected header */
1419 pkt->pn = packet_number;
1420 pkt->pnl = pnlen;
1421
1422 ret = 1;
1423 leave:
1424 if (cctx)
1425 EVP_CIPHER_CTX_free(cctx);
1426 TRACE_LEAVE(QUIC_EV_CONN_RMHP, qc);
1427 return ret;
1428}
1429
1430/* Encrypt the payload of a QUIC packet with <pn> as number found at <payload>
1431 * address, with <payload_len> as payload length, <aad> as address of
1432 * the ADD and <aad_len> as AAD length depending on the <tls_ctx> QUIC TLS
1433 * context.
1434 * Returns 1 if succeeded, 0 if not.
1435 */
1436static int quic_packet_encrypt(unsigned char *payload, size_t payload_len,
1437 unsigned char *aad, size_t aad_len, uint64_t pn,
1438 struct quic_tls_ctx *tls_ctx, struct quic_conn *qc)
1439{
1440 int ret = 0;
1441 unsigned char iv[QUIC_TLS_IV_LEN];
1442 unsigned char *tx_iv = tls_ctx->tx.iv;
1443 size_t tx_iv_sz = tls_ctx->tx.ivlen;
1444 struct enc_debug_info edi;
1445
1446 TRACE_ENTER(QUIC_EV_CONN_ENCPKT, qc);
1447
1448 if (!quic_aead_iv_build(iv, sizeof iv, tx_iv, tx_iv_sz, pn)) {
1449 TRACE_ERROR("AEAD IV building for encryption failed", QUIC_EV_CONN_ENCPKT, qc);
1450 goto err;
1451 }
1452
1453 if (!quic_tls_encrypt(payload, payload_len, aad, aad_len,
1454 tls_ctx->tx.ctx, tls_ctx->tx.aead, tls_ctx->tx.key, iv)) {
1455 TRACE_ERROR("QUIC packet encryption failed", QUIC_EV_CONN_ENCPKT, qc);
1456 goto err;
1457 }
1458
1459 ret = 1;
1460 leave:
1461 TRACE_LEAVE(QUIC_EV_CONN_ENCPKT, qc);
1462 return ret;
1463
1464 err:
1465 enc_debug_info_init(&edi, payload, payload_len, aad, aad_len, pn);
1466 goto leave;
1467}
1468
Amaury Denoyelle518c98f2022-11-24 17:12:25 +01001469/* Decrypt <pkt> packet using encryption level <qel> for <qc> connection.
1470 * Decryption is done in place in packet buffer.
1471 *
Ilya Shipitsin5fa29b82022-12-07 09:46:19 +05001472 * Returns 1 on success else 0.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001473 */
Amaury Denoyelle518c98f2022-11-24 17:12:25 +01001474static int qc_pkt_decrypt(struct quic_conn *qc, struct quic_enc_level *qel,
1475 struct quic_rx_packet *pkt)
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001476{
1477 int ret, kp_changed;
1478 unsigned char iv[QUIC_TLS_IV_LEN];
1479 struct quic_tls_ctx *tls_ctx = &qel->tls_ctx;
1480 EVP_CIPHER_CTX *rx_ctx = tls_ctx->rx.ctx;
1481 unsigned char *rx_iv = tls_ctx->rx.iv;
1482 size_t rx_iv_sz = tls_ctx->rx.ivlen;
1483 unsigned char *rx_key = tls_ctx->rx.key;
1484
1485 TRACE_ENTER(QUIC_EV_CONN_RXPKT, qc);
1486
1487 ret = 0;
1488 kp_changed = 0;
1489
1490 if (pkt->type == QUIC_PACKET_TYPE_SHORT) {
1491 /* The two tested bits are not at the same position,
1492 * this is why they are first both inversed.
1493 */
1494 if (!(*pkt->data & QUIC_PACKET_KEY_PHASE_BIT) ^ !(tls_ctx->flags & QUIC_FL_TLS_KP_BIT_SET)) {
1495 if (pkt->pn < tls_ctx->rx.pn) {
1496 /* The lowest packet number of a previous key phase
1497 * cannot be null if it really stores previous key phase
1498 * secrets.
1499 */
1500 // TODO: check if BUG_ON() more suitable
Amaury Denoyelle518c98f2022-11-24 17:12:25 +01001501 if (!qc->ku.prv_rx.pn) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001502 TRACE_ERROR("null previous packet number", QUIC_EV_CONN_RXPKT, qc);
1503 goto leave;
1504 }
1505
Amaury Denoyelle518c98f2022-11-24 17:12:25 +01001506 rx_ctx = qc->ku.prv_rx.ctx;
1507 rx_iv = qc->ku.prv_rx.iv;
1508 rx_key = qc->ku.prv_rx.key;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001509 }
1510 else if (pkt->pn > qel->pktns->rx.largest_pn) {
1511 /* Next key phase */
1512 kp_changed = 1;
Amaury Denoyelle518c98f2022-11-24 17:12:25 +01001513 rx_ctx = qc->ku.nxt_rx.ctx;
1514 rx_iv = qc->ku.nxt_rx.iv;
1515 rx_key = qc->ku.nxt_rx.key;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001516 }
1517 }
1518 }
1519
1520 if (!quic_aead_iv_build(iv, sizeof iv, rx_iv, rx_iv_sz, pkt->pn)) {
1521 TRACE_ERROR("quic_aead_iv_build() failed", QUIC_EV_CONN_RXPKT, qc);
1522 goto leave;
1523 }
1524
1525 ret = quic_tls_decrypt(pkt->data + pkt->aad_len, pkt->len - pkt->aad_len,
1526 pkt->data, pkt->aad_len,
1527 rx_ctx, tls_ctx->rx.aead, rx_key, iv);
1528 if (!ret) {
1529 TRACE_ERROR("quic_tls_decrypt() failed", QUIC_EV_CONN_RXPKT, qc);
1530 goto leave;
1531 }
1532
1533 /* Update the keys only if the packet decryption succeeded. */
1534 if (kp_changed) {
Amaury Denoyelle518c98f2022-11-24 17:12:25 +01001535 quic_tls_rotate_keys(qc);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001536 /* Toggle the Key Phase bit */
1537 tls_ctx->flags ^= QUIC_FL_TLS_KP_BIT_SET;
1538 /* Store the lowest packet number received for the current key phase */
1539 tls_ctx->rx.pn = pkt->pn;
1540 /* Prepare the next key update */
Amaury Denoyelle518c98f2022-11-24 17:12:25 +01001541 if (!quic_tls_key_update(qc)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001542 TRACE_ERROR("quic_tls_key_update() failed", QUIC_EV_CONN_RXPKT, qc);
1543 goto leave;
1544 }
1545 }
1546
1547 /* Update the packet length (required to parse the frames). */
1548 pkt->len -= QUIC_TLS_TAG_LEN;
1549 ret = 1;
1550 leave:
1551 TRACE_LEAVE(QUIC_EV_CONN_RXPKT, qc);
1552 return ret;
1553}
1554
1555
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001556/* Release <frm> frame and mark its copies as acknowledged */
1557void qc_release_frm(struct quic_conn *qc, struct quic_frame *frm)
1558{
1559 uint64_t pn;
1560 struct quic_frame *origin, *f, *tmp;
1561
1562 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc);
1563
1564 /* Identify this frame: a frame copy or one of its copies */
1565 origin = frm->origin ? frm->origin : frm;
1566 /* Ensure the source of the copies is flagged as acked, <frm> being
1567 * possibly a copy of <origin>
1568 */
1569 origin->flags |= QUIC_FL_TX_FRAME_ACKED;
1570 /* Mark all the copy of <origin> as acknowledged. We must
1571 * not release the packets (releasing the frames) at this time as
1572 * they are possibly also to be acknowledged alongside the
1573 * the current one.
1574 */
1575 list_for_each_entry_safe(f, tmp, &origin->reflist, ref) {
1576 if (f->pkt) {
1577 f->flags |= QUIC_FL_TX_FRAME_ACKED;
1578 f->origin = NULL;
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001579 LIST_DEL_INIT(&f->ref);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001580 pn = f->pkt->pn_node.key;
1581 TRACE_DEVEL("mark frame as acked from packet",
1582 QUIC_EV_CONN_PRSAFRM, qc, f, &pn);
1583 }
1584 else {
1585 TRACE_DEVEL("freeing unsent frame",
1586 QUIC_EV_CONN_PRSAFRM, qc, f);
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001587 LIST_DEL_INIT(&f->ref);
1588 qc_frm_free(&f);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001589 }
1590 }
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001591 LIST_DEL_INIT(&frm->list);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001592 pn = frm->pkt->pn_node.key;
1593 quic_tx_packet_refdec(frm->pkt);
1594 TRACE_DEVEL("freeing frame from packet",
1595 QUIC_EV_CONN_PRSAFRM, qc, frm, &pn);
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001596 qc_frm_free(&frm);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001597
1598 TRACE_LEAVE(QUIC_EV_CONN_PRSAFRM, qc);
1599}
1600
1601/* Schedule a CONNECTION_CLOSE emission on <qc> if the MUX has been released
1602 * and all STREAM data are acknowledged. The MUX is responsible to have set
1603 * <qc.err> before as it is reused for the CONNECTION_CLOSE frame.
1604 *
1605 * TODO this should also be called on lost packet detection
1606 */
1607void qc_check_close_on_released_mux(struct quic_conn *qc)
1608{
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001609 TRACE_ENTER(QUIC_EV_CONN_CLOSE, qc);
1610
1611 if (qc->mux_state == QC_MUX_RELEASED && eb_is_empty(&qc->streams_by_id)) {
1612 /* Reuse errcode which should have been previously set by the MUX on release. */
1613 quic_set_connection_close(qc, qc->err);
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02001614 tasklet_wakeup(qc->wait_event.tasklet);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001615 }
1616
1617 TRACE_LEAVE(QUIC_EV_CONN_CLOSE, qc);
1618}
1619
1620/* Remove from <stream> the acknowledged frames.
1621 *
1622 * Returns 1 if at least one frame was removed else 0.
1623 */
1624static int quic_stream_try_to_consume(struct quic_conn *qc,
1625 struct qc_stream_desc *stream)
1626{
1627 int ret;
1628 struct eb64_node *frm_node;
1629
1630 TRACE_ENTER(QUIC_EV_CONN_ACKSTRM, qc);
1631
1632 ret = 0;
1633 frm_node = eb64_first(&stream->acked_frms);
1634 while (frm_node) {
1635 struct quic_stream *strm;
1636 struct quic_frame *frm;
1637 size_t offset, len;
1638
1639 strm = eb64_entry(frm_node, struct quic_stream, offset);
1640 offset = strm->offset.key;
1641 len = strm->len;
1642
1643 if (offset > stream->ack_offset)
1644 break;
1645
1646 if (qc_stream_desc_ack(&stream, offset, len)) {
1647 /* cf. next comment : frame may be freed at this stage. */
1648 TRACE_DEVEL("stream consumed", QUIC_EV_CONN_ACKSTRM,
1649 qc, stream ? strm : NULL, stream);
1650 ret = 1;
1651 }
1652
1653 /* If stream is NULL after qc_stream_desc_ack(), it means frame
1654 * has been freed. with the stream frames tree. Nothing to do
1655 * anymore in here.
1656 */
1657 if (!stream) {
1658 qc_check_close_on_released_mux(qc);
1659 ret = 1;
1660 goto leave;
1661 }
1662
1663 frm_node = eb64_next(frm_node);
1664 eb64_delete(&strm->offset);
1665
1666 frm = container_of(strm, struct quic_frame, stream);
1667 qc_release_frm(qc, frm);
1668 }
1669
1670 leave:
1671 TRACE_LEAVE(QUIC_EV_CONN_ACKSTRM, qc);
1672 return ret;
1673}
1674
1675/* Treat <frm> frame whose packet it is attached to has just been acknowledged. */
1676static inline void qc_treat_acked_tx_frm(struct quic_conn *qc,
1677 struct quic_frame *frm)
1678{
1679 int stream_acked;
1680
1681 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc, frm);
1682
1683 stream_acked = 0;
1684 switch (frm->type) {
1685 case QUIC_FT_STREAM_8 ... QUIC_FT_STREAM_F:
1686 {
1687 struct quic_stream *strm_frm = &frm->stream;
1688 struct eb64_node *node = NULL;
1689 struct qc_stream_desc *stream = NULL;
1690 const size_t offset = strm_frm->offset.key;
1691 const size_t len = strm_frm->len;
1692
1693 /* do not use strm_frm->stream as the qc_stream_desc instance
1694 * might be freed at this stage. Use the id to do a proper
1695 * lookup.
1696 *
1697 * TODO if lookup operation impact on the perf is noticeable,
1698 * implement a refcount on qc_stream_desc instances.
1699 */
1700 node = eb64_lookup(&qc->streams_by_id, strm_frm->id);
1701 if (!node) {
1702 TRACE_DEVEL("acked stream for released stream", QUIC_EV_CONN_ACKSTRM, qc, strm_frm);
1703 qc_release_frm(qc, frm);
1704 /* early return */
1705 goto leave;
1706 }
1707 stream = eb64_entry(node, struct qc_stream_desc, by_id);
1708
1709 TRACE_DEVEL("acked stream", QUIC_EV_CONN_ACKSTRM, qc, strm_frm, stream);
1710 if (offset <= stream->ack_offset) {
1711 if (qc_stream_desc_ack(&stream, offset, len)) {
1712 stream_acked = 1;
1713 TRACE_DEVEL("stream consumed", QUIC_EV_CONN_ACKSTRM,
1714 qc, strm_frm, stream);
1715 }
1716
1717 if (!stream) {
1718 /* no need to continue if stream freed. */
1719 TRACE_DEVEL("stream released and freed", QUIC_EV_CONN_ACKSTRM, qc);
1720 qc_release_frm(qc, frm);
1721 qc_check_close_on_released_mux(qc);
1722 break;
1723 }
1724
1725 TRACE_DEVEL("stream consumed", QUIC_EV_CONN_ACKSTRM,
1726 qc, strm_frm, stream);
1727 qc_release_frm(qc, frm);
1728 }
1729 else {
1730 eb64_insert(&stream->acked_frms, &strm_frm->offset);
1731 }
1732
1733 stream_acked |= quic_stream_try_to_consume(qc, stream);
1734 }
1735 break;
1736 default:
1737 qc_release_frm(qc, frm);
1738 }
1739
Amaury Denoyellebbb1c682022-09-28 15:15:51 +02001740 if (stream_acked) {
1741 if (qc->subs && qc->subs->events & SUB_RETRY_SEND) {
1742 tasklet_wakeup(qc->subs->tasklet);
1743 qc->subs->events &= ~SUB_RETRY_SEND;
1744 if (!qc->subs->events)
1745 qc->subs = NULL;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001746 }
1747 }
1748 leave:
1749 TRACE_LEAVE(QUIC_EV_CONN_PRSAFRM, qc);
1750}
1751
1752/* Remove <largest> down to <smallest> node entries from <pkts> tree of TX packet,
1753 * deallocating them, and their TX frames.
1754 * Returns the last node reached to be used for the next range.
1755 * May be NULL if <largest> node could not be found.
1756 */
1757static inline struct eb64_node *qc_ackrng_pkts(struct quic_conn *qc,
1758 struct eb_root *pkts,
1759 unsigned int *pkt_flags,
1760 struct list *newly_acked_pkts,
1761 struct eb64_node *largest_node,
1762 uint64_t largest, uint64_t smallest)
1763{
1764 struct eb64_node *node;
1765 struct quic_tx_packet *pkt;
1766
1767 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc);
1768
1769 node = largest_node ? largest_node : eb64_lookup_le(pkts, largest);
1770 while (node && node->key >= smallest) {
1771 struct quic_frame *frm, *frmbak;
1772
1773 pkt = eb64_entry(node, struct quic_tx_packet, pn_node);
1774 *pkt_flags |= pkt->flags;
1775 LIST_INSERT(newly_acked_pkts, &pkt->list);
1776 TRACE_DEVEL("Removing packet #", QUIC_EV_CONN_PRSAFRM, qc, NULL, &pkt->pn_node.key);
1777 list_for_each_entry_safe(frm, frmbak, &pkt->frms, list)
1778 qc_treat_acked_tx_frm(qc, frm);
Frédéric Lécaille814645f2022-11-18 18:15:28 +01001779 /* If there are others packet in the same datagram <pkt> is attached to,
1780 * detach the previous one and the next one from <pkt>.
1781 */
Frédéric Lécaille74b5f7b2022-11-20 18:35:35 +01001782 quic_tx_packet_dgram_detach(pkt);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001783 node = eb64_prev(node);
1784 eb64_delete(&pkt->pn_node);
1785 }
1786
1787 TRACE_LEAVE(QUIC_EV_CONN_PRSAFRM, qc);
1788 return node;
1789}
1790
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01001791/* Remove all frames from <pkt_frm_list> and reinsert them in the same order
1792 * they have been sent into <pktns_frm_list>. The loss counter of each frame is
1793 * incremented and checked if it does not exceed retransmission limit.
1794 *
1795 * Returns 1 on success, 0 if a frame loss limit is exceeded. A
1796 * CONNECTION_CLOSE is scheduled in this case.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001797 */
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01001798static inline int qc_requeue_nacked_pkt_tx_frms(struct quic_conn *qc,
1799 struct quic_tx_packet *pkt,
1800 struct list *pktns_frm_list)
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001801{
1802 struct quic_frame *frm, *frmbak;
1803 struct list tmp = LIST_HEAD_INIT(tmp);
1804 struct list *pkt_frm_list = &pkt->frms;
1805 uint64_t pn = pkt->pn_node.key;
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01001806 int close = 0;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001807
1808 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc);
1809
1810 list_for_each_entry_safe(frm, frmbak, pkt_frm_list, list) {
1811 /* First remove this frame from the packet it was attached to */
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001812 LIST_DEL_INIT(&frm->list);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001813 quic_tx_packet_refdec(pkt);
1814 /* At this time, this frame is not freed but removed from its packet */
1815 frm->pkt = NULL;
1816 /* Remove any reference to this frame */
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001817 qc_frm_unref(frm, qc);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001818 switch (frm->type) {
1819 case QUIC_FT_STREAM_8 ... QUIC_FT_STREAM_F:
1820 {
1821 struct quic_stream *strm_frm = &frm->stream;
1822 struct eb64_node *node = NULL;
1823 struct qc_stream_desc *stream_desc;
1824
1825 node = eb64_lookup(&qc->streams_by_id, strm_frm->id);
1826 if (!node) {
1827 TRACE_DEVEL("released stream", QUIC_EV_CONN_PRSAFRM, qc, frm);
1828 TRACE_DEVEL("freeing frame from packet", QUIC_EV_CONN_PRSAFRM,
1829 qc, frm, &pn);
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001830 qc_frm_free(&frm);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001831 continue;
1832 }
1833
1834 stream_desc = eb64_entry(node, struct qc_stream_desc, by_id);
1835 /* Do not resend this frame if in the "already acked range" */
1836 if (strm_frm->offset.key + strm_frm->len <= stream_desc->ack_offset) {
1837 TRACE_DEVEL("ignored frame in already acked range",
1838 QUIC_EV_CONN_PRSAFRM, qc, frm);
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001839 qc_frm_free(&frm);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001840 continue;
1841 }
1842 else if (strm_frm->offset.key < stream_desc->ack_offset) {
1843 strm_frm->offset.key = stream_desc->ack_offset;
1844 TRACE_DEVEL("updated partially acked frame",
1845 QUIC_EV_CONN_PRSAFRM, qc, frm);
1846 }
1847 break;
1848 }
1849
1850 default:
1851 break;
1852 }
1853
1854 /* Do not resend probing packet with old data */
1855 if (pkt->flags & QUIC_FL_TX_PACKET_PROBE_WITH_OLD_DATA) {
1856 TRACE_DEVEL("ignored frame with old data from packet", QUIC_EV_CONN_PRSAFRM,
1857 qc, frm, &pn);
1858 if (frm->origin)
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001859 LIST_DEL_INIT(&frm->ref);
1860 qc_frm_free(&frm);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001861 continue;
1862 }
1863
1864 if (frm->flags & QUIC_FL_TX_FRAME_ACKED) {
1865 TRACE_DEVEL("already acked frame", QUIC_EV_CONN_PRSAFRM, qc, frm);
1866 TRACE_DEVEL("freeing frame from packet", QUIC_EV_CONN_PRSAFRM,
1867 qc, frm, &pn);
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001868 qc_frm_free(&frm);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001869 }
1870 else {
Amaury Denoyelle24d5b722023-01-31 11:44:50 +01001871 if (++frm->loss_count >= global.tune.quic_max_frame_loss) {
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01001872 TRACE_ERROR("retransmission limit reached, closing the connection", QUIC_EV_CONN_PRSAFRM, qc);
1873 quic_set_connection_close(qc, quic_err_transport(QC_ERR_INTERNAL_ERROR));
1874 close = 1;
1875 }
1876
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001877 if (QUIC_FT_STREAM_8 <= frm->type && frm->type <= QUIC_FT_STREAM_F) {
1878 /* Mark this STREAM frame as lost. A look up their stream descriptor
1879 * will be performed to check the stream is not consumed or released.
1880 */
1881 frm->flags |= QUIC_FL_TX_FRAME_LOST;
1882 }
1883 LIST_APPEND(&tmp, &frm->list);
1884 TRACE_DEVEL("frame requeued", QUIC_EV_CONN_PRSAFRM, qc, frm);
1885 }
1886 }
1887
1888 LIST_SPLICE(pktns_frm_list, &tmp);
1889
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01001890 end:
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001891 TRACE_LEAVE(QUIC_EV_CONN_PRSAFRM, qc);
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01001892 return !close;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001893}
1894
1895/* Free <pkt> TX packet and its attached frames.
1896 * This is the responsibility of the caller to remove this packet of
1897 * any data structure it was possibly attached to.
1898 */
1899static inline void free_quic_tx_packet(struct quic_conn *qc,
1900 struct quic_tx_packet *pkt)
1901{
1902 struct quic_frame *frm, *frmbak;
1903
1904 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
1905
1906 if (!pkt)
1907 goto leave;
1908
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01001909 list_for_each_entry_safe(frm, frmbak, &pkt->frms, list)
1910 qc_frm_free(&frm);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02001911 pool_free(pool_head_quic_tx_packet, pkt);
1912
1913 leave:
1914 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
1915}
1916
1917/* Free the TX packets of <pkts> list */
1918static inline void free_quic_tx_pkts(struct quic_conn *qc, struct list *pkts)
1919{
1920 struct quic_tx_packet *pkt, *tmp;
1921
1922 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
1923
1924 list_for_each_entry_safe(pkt, tmp, pkts, list) {
1925 LIST_DELETE(&pkt->list);
1926 eb64_delete(&pkt->pn_node);
1927 free_quic_tx_packet(qc, pkt);
1928 }
1929
1930 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
1931}
1932
1933/* Remove already sent ranges of acknowledged packet numbers from
1934 * <pktns> packet number space tree below <largest_acked_pn> possibly
1935 * updating the range which contains <largest_acked_pn>.
1936 * Never fails.
1937 */
1938static void qc_treat_ack_of_ack(struct quic_conn *qc,
1939 struct quic_pktns *pktns,
1940 int64_t largest_acked_pn)
1941{
1942 struct eb64_node *ar, *next_ar;
1943 struct quic_arngs *arngs = &pktns->rx.arngs;
1944
1945 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc);
1946
1947 ar = eb64_first(&arngs->root);
1948 while (ar) {
1949 struct quic_arng_node *ar_node;
1950
1951 next_ar = eb64_next(ar);
1952 ar_node = eb64_entry(ar, struct quic_arng_node, first);
1953
1954 if ((int64_t)ar_node->first.key > largest_acked_pn) {
1955 TRACE_DEVEL("first.key > largest", QUIC_EV_CONN_PRSAFRM, qc);
1956 break;
1957 }
1958
1959 if (largest_acked_pn < ar_node->last) {
1960 eb64_delete(ar);
1961 ar_node->first.key = largest_acked_pn + 1;
1962 eb64_insert(&arngs->root, ar);
1963 break;
1964 }
1965
1966 eb64_delete(ar);
1967 pool_free(pool_head_quic_arng, ar_node);
1968 arngs->sz--;
1969 ar = next_ar;
1970 }
1971
1972 TRACE_LEAVE(QUIC_EV_CONN_PRSAFRM, qc);
1973}
1974
1975/* Send a packet ack event nofication for each newly acked packet of
1976 * <newly_acked_pkts> list and free them.
1977 * Always succeeds.
1978 */
1979static inline void qc_treat_newly_acked_pkts(struct quic_conn *qc,
1980 struct list *newly_acked_pkts)
1981{
1982 struct quic_tx_packet *pkt, *tmp;
1983 struct quic_cc_event ev = { .type = QUIC_CC_EVT_ACK, };
1984
1985 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc);
1986
1987 list_for_each_entry_safe(pkt, tmp, newly_acked_pkts, list) {
1988 pkt->pktns->tx.in_flight -= pkt->in_flight_len;
1989 qc->path->prep_in_flight -= pkt->in_flight_len;
1990 qc->path->in_flight -= pkt->in_flight_len;
1991 if (pkt->flags & QUIC_FL_TX_PACKET_ACK_ELICITING)
1992 qc->path->ifae_pkts--;
1993 /* If this packet contained an ACK frame, proceed to the
1994 * acknowledging of range of acks from the largest acknowledged
1995 * packet number which was sent in an ACK frame by this packet.
1996 */
1997 if (pkt->largest_acked_pn != -1)
1998 qc_treat_ack_of_ack(qc, pkt->pktns, pkt->largest_acked_pn);
1999 ev.ack.acked = pkt->in_flight_len;
2000 ev.ack.time_sent = pkt->time_sent;
2001 quic_cc_event(&qc->path->cc, &ev);
2002 LIST_DELETE(&pkt->list);
2003 eb64_delete(&pkt->pn_node);
2004 quic_tx_packet_refdec(pkt);
2005 }
2006
2007 TRACE_LEAVE(QUIC_EV_CONN_PRSAFRM, qc);
2008
2009}
2010
2011/* Release all the frames attached to <pktns> packet number space */
2012static inline void qc_release_pktns_frms(struct quic_conn *qc,
2013 struct quic_pktns *pktns)
2014{
2015 struct quic_frame *frm, *frmbak;
2016
2017 TRACE_ENTER(QUIC_EV_CONN_PHPKTS, qc);
2018
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01002019 list_for_each_entry_safe(frm, frmbak, &pktns->tx.frms, list)
2020 qc_frm_free(&frm);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002021
2022 TRACE_LEAVE(QUIC_EV_CONN_PHPKTS, qc);
2023}
2024
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002025/* Handle <pkts> list of lost packets detected at <now_us> handling their TX
2026 * frames. Send a packet loss event to the congestion controller if in flight
2027 * packet have been lost. Also frees the packet in <pkts> list.
2028 *
2029 * Returns 1 on success else 0 if loss limit has been exceeded. A
2030 * CONNECTION_CLOSE was prepared to close the connection ASAP.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002031 */
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002032static inline int qc_release_lost_pkts(struct quic_conn *qc,
2033 struct quic_pktns *pktns,
2034 struct list *pkts,
2035 uint64_t now_us)
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002036{
2037 struct quic_tx_packet *pkt, *tmp, *oldest_lost, *newest_lost;
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002038 int close = 0;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002039
2040 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc);
2041
2042 if (LIST_ISEMPTY(pkts))
2043 goto leave;
2044
2045 oldest_lost = newest_lost = NULL;
2046 list_for_each_entry_safe(pkt, tmp, pkts, list) {
2047 struct list tmp = LIST_HEAD_INIT(tmp);
2048
2049 pkt->pktns->tx.in_flight -= pkt->in_flight_len;
2050 qc->path->prep_in_flight -= pkt->in_flight_len;
2051 qc->path->in_flight -= pkt->in_flight_len;
2052 if (pkt->flags & QUIC_FL_TX_PACKET_ACK_ELICITING)
2053 qc->path->ifae_pkts--;
2054 /* Treat the frames of this lost packet. */
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002055 if (!qc_requeue_nacked_pkt_tx_frms(qc, pkt, &pktns->tx.frms))
2056 close = 1;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002057 LIST_DELETE(&pkt->list);
2058 if (!oldest_lost) {
2059 oldest_lost = newest_lost = pkt;
2060 }
2061 else {
2062 if (newest_lost != oldest_lost)
2063 quic_tx_packet_refdec(newest_lost);
2064 newest_lost = pkt;
2065 }
2066 }
2067
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002068 if (!close) {
2069 if (newest_lost) {
2070 /* Sent a congestion event to the controller */
2071 struct quic_cc_event ev = { };
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002072
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002073 ev.type = QUIC_CC_EVT_LOSS;
2074 ev.loss.time_sent = newest_lost->time_sent;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002075
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002076 quic_cc_event(&qc->path->cc, &ev);
2077 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002078
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002079 /* If an RTT have been already sampled, <rtt_min> has been set.
2080 * We must check if we are experiencing a persistent congestion.
2081 * If this is the case, the congestion controller must re-enter
2082 * slow start state.
2083 */
2084 if (qc->path->loss.rtt_min && newest_lost != oldest_lost) {
2085 unsigned int period = newest_lost->time_sent - oldest_lost->time_sent;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002086
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002087 if (quic_loss_persistent_congestion(&qc->path->loss, period,
2088 now_ms, qc->max_ack_delay))
2089 qc->path->cc.algo->slow_start(&qc->path->cc);
2090 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002091 }
2092
Amaury Denoyelle3a72ba22022-11-14 11:41:34 +01002093 /* <oldest_lost> cannot be NULL at this stage because we have ensured
2094 * that <pkts> list is not empty. Without this, GCC 12.2.0 reports a
2095 * possible overflow on a 0 byte region with O2 optimization.
2096 */
2097 ALREADY_CHECKED(oldest_lost);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002098 quic_tx_packet_refdec(oldest_lost);
2099 if (newest_lost != oldest_lost)
2100 quic_tx_packet_refdec(newest_lost);
2101
2102 leave:
2103 TRACE_LEAVE(QUIC_EV_CONN_PRSAFRM, qc);
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002104 return !close;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002105}
2106
2107/* Parse ACK frame into <frm> from a buffer at <buf> address with <end> being at
2108 * one byte past the end of this buffer. Also update <rtt_sample> if needed, i.e.
2109 * if the largest acked packet was newly acked and if there was at least one newly
2110 * acked ack-eliciting packet.
2111 * Return 1, if succeeded, 0 if not.
2112 */
2113static inline int qc_parse_ack_frm(struct quic_conn *qc,
2114 struct quic_frame *frm,
2115 struct quic_enc_level *qel,
2116 unsigned int *rtt_sample,
2117 const unsigned char **pos, const unsigned char *end)
2118{
2119 struct quic_ack *ack = &frm->ack;
2120 uint64_t smallest, largest;
2121 struct eb_root *pkts;
2122 struct eb64_node *largest_node;
2123 unsigned int time_sent, pkt_flags;
2124 struct list newly_acked_pkts = LIST_HEAD_INIT(newly_acked_pkts);
2125 struct list lost_pkts = LIST_HEAD_INIT(lost_pkts);
2126 int ret = 0;
2127
2128 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc);
2129
2130 if (ack->largest_ack > qel->pktns->tx.next_pn) {
2131 TRACE_DEVEL("ACK for not sent packet", QUIC_EV_CONN_PRSAFRM,
2132 qc, NULL, &ack->largest_ack);
2133 goto err;
2134 }
2135
2136 if (ack->first_ack_range > ack->largest_ack) {
2137 TRACE_DEVEL("too big first ACK range", QUIC_EV_CONN_PRSAFRM,
2138 qc, NULL, &ack->first_ack_range);
2139 goto err;
2140 }
2141
2142 largest = ack->largest_ack;
2143 smallest = largest - ack->first_ack_range;
2144 pkts = &qel->pktns->tx.pkts;
2145 pkt_flags = 0;
2146 largest_node = NULL;
2147 time_sent = 0;
2148
2149 if ((int64_t)ack->largest_ack > qel->pktns->rx.largest_acked_pn) {
2150 largest_node = eb64_lookup(pkts, largest);
2151 if (!largest_node) {
2152 TRACE_DEVEL("Largest acked packet not found",
2153 QUIC_EV_CONN_PRSAFRM, qc);
2154 }
2155 else {
2156 time_sent = eb64_entry(largest_node,
2157 struct quic_tx_packet, pn_node)->time_sent;
2158 }
2159 }
2160
2161 TRACE_PROTO("rcvd ack range", QUIC_EV_CONN_PRSAFRM,
2162 qc, NULL, &largest, &smallest);
2163 do {
2164 uint64_t gap, ack_range;
2165
2166 qc_ackrng_pkts(qc, pkts, &pkt_flags, &newly_acked_pkts,
2167 largest_node, largest, smallest);
2168 if (!ack->ack_range_num--)
2169 break;
2170
2171 if (!quic_dec_int(&gap, pos, end)) {
2172 TRACE_ERROR("quic_dec_int(gap) failed", QUIC_EV_CONN_PRSAFRM, qc);
2173 goto err;
2174 }
2175
2176 if (smallest < gap + 2) {
2177 TRACE_DEVEL("wrong gap value", QUIC_EV_CONN_PRSAFRM,
2178 qc, NULL, &gap, &smallest);
2179 goto err;
2180 }
2181
2182 largest = smallest - gap - 2;
2183 if (!quic_dec_int(&ack_range, pos, end)) {
2184 TRACE_ERROR("quic_dec_int(ack_range) failed", QUIC_EV_CONN_PRSAFRM, qc);
2185 goto err;
2186 }
2187
2188 if (largest < ack_range) {
2189 TRACE_DEVEL("wrong ack range value", QUIC_EV_CONN_PRSAFRM,
2190 qc, NULL, &largest, &ack_range);
2191 goto err;
2192 }
2193
2194 /* Do not use this node anymore. */
2195 largest_node = NULL;
2196 /* Next range */
2197 smallest = largest - ack_range;
2198
2199 TRACE_PROTO("rcvd next ack range", QUIC_EV_CONN_PRSAFRM,
2200 qc, NULL, &largest, &smallest);
2201 } while (1);
2202
2203 if (time_sent && (pkt_flags & QUIC_FL_TX_PACKET_ACK_ELICITING)) {
2204 *rtt_sample = tick_remain(time_sent, now_ms);
2205 qel->pktns->rx.largest_acked_pn = ack->largest_ack;
2206 }
2207
2208 if (!LIST_ISEMPTY(&newly_acked_pkts)) {
2209 if (!eb_is_empty(&qel->pktns->tx.pkts)) {
2210 qc_packet_loss_lookup(qel->pktns, qc, &lost_pkts);
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01002211 if (!qc_release_lost_pkts(qc, qel->pktns, &lost_pkts, now_ms))
2212 goto leave;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002213 }
2214 qc_treat_newly_acked_pkts(qc, &newly_acked_pkts);
2215 if (quic_peer_validated_addr(qc))
2216 qc->path->loss.pto_count = 0;
2217 qc_set_timer(qc);
2218 }
2219
2220 ret = 1;
2221 leave:
2222 TRACE_LEAVE(QUIC_EV_CONN_PRSAFRM, qc);
2223 return ret;
2224
2225 err:
2226 free_quic_tx_pkts(qc, &newly_acked_pkts);
2227 goto leave;
2228}
2229
2230/* This function gives the detail of the SSL error. It is used only
2231 * if the debug mode and the verbose mode are activated. It dump all
2232 * the SSL error until the stack was empty.
2233 */
2234static forceinline void qc_ssl_dump_errors(struct connection *conn)
2235{
2236 if (unlikely(global.mode & MODE_DEBUG)) {
2237 while (1) {
2238 const char *func = NULL;
2239 unsigned long ret;
2240
2241 ERR_peek_error_func(&func);
2242 ret = ERR_get_error();
2243 if (!ret)
2244 return;
2245
2246 fprintf(stderr, "conn. @%p OpenSSL error[0x%lx] %s: %s\n", conn, ret,
2247 func, ERR_reason_error_string(ret));
2248 }
2249 }
2250}
2251
2252int ssl_sock_get_alpn(const struct connection *conn, void *xprt_ctx,
2253 const char **str, int *len);
2254
Frédéric Lécailleaf25a692023-02-01 17:56:57 +01002255/* Finalize <qc> QUIC connection:
2256 * - initialize the Initial QUIC TLS context for negotiated version,
2257 * - derive the secrets for this context,
2258 * - set them into the TLS stack,
2259 *
2260 * MUST be called after having received the remote transport parameters which
2261 * are parsed when the TLS callback for the ClientHello message is called upon
2262 * SSL_do_handshake() calls, not necessarily at the first time as this TLS
2263 * message may be splitted between packets
2264 * Return 1 if succeeded, 0 if not.
2265 */
2266static int qc_conn_finalize(struct quic_conn *qc, int server)
2267{
2268 int ret = 0;
2269
2270 TRACE_ENTER(QUIC_EV_CONN_NEW, qc);
2271
2272 if (qc->flags & QUIC_FL_CONN_FINALIZED)
2273 goto finalized;
2274
2275 if (qc->negotiated_version &&
2276 !qc_new_isecs(qc, &qc->negotiated_ictx, qc->negotiated_version,
2277 qc->odcid.data, qc->odcid.len, server))
2278 goto out;
2279
2280 /* This connection is functional (ready to send/receive) */
2281 qc->flags |= QUIC_FL_CONN_FINALIZED;
2282
2283 finalized:
2284 ret = 1;
2285 out:
2286 TRACE_LEAVE(QUIC_EV_CONN_NEW, qc);
2287 return ret;
2288}
2289
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002290/* Provide CRYPTO data to the TLS stack found at <data> with <len> as length
2291 * from <qel> encryption level with <ctx> as QUIC connection context.
2292 * Remaining parameter are there for debugging purposes.
2293 * Return 1 if succeeded, 0 if not.
2294 */
2295static inline int qc_provide_cdata(struct quic_enc_level *el,
2296 struct ssl_sock_ctx *ctx,
2297 const unsigned char *data, size_t len,
2298 struct quic_rx_packet *pkt,
2299 struct quic_rx_crypto_frm *cf)
2300{
Amaury Denoyelle2f668f02022-11-18 15:24:08 +01002301#ifdef DEBUG_STRICT
2302 enum ncb_ret ncb_ret;
2303#endif
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002304 int ssl_err, state;
2305 struct quic_conn *qc;
2306 int ret = 0;
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002307 struct ncbuf *ncbuf = &el->cstream->rx.ncbuf;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002308
2309 ssl_err = SSL_ERROR_NONE;
2310 qc = ctx->qc;
2311
2312 TRACE_ENTER(QUIC_EV_CONN_SSLDATA, qc);
2313
2314 if (SSL_provide_quic_data(ctx->ssl, el->level, data, len) != 1) {
2315 TRACE_ERROR("SSL_provide_quic_data() error",
2316 QUIC_EV_CONN_SSLDATA, qc, pkt, cf, ctx->ssl);
2317 goto leave;
2318 }
2319
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002320 TRACE_PROTO("in order CRYPTO data",
2321 QUIC_EV_CONN_SSLDATA, qc, NULL, cf, ctx->ssl);
2322
2323 state = qc->state;
2324 if (state < QUIC_HS_ST_COMPLETE) {
2325 ssl_err = SSL_do_handshake(ctx->ssl);
Frédéric Lécailleaf25a692023-02-01 17:56:57 +01002326
Frédéric Lécaille0aa79952023-02-03 16:15:08 +01002327 if (qc->flags & QUIC_FL_CONN_TO_KILL) {
2328 TRACE_DEVEL("connection to be killed", QUIC_EV_CONN_IO_CB, qc);
2329 goto leave;
2330 }
2331
Frédéric Lécailleaf25a692023-02-01 17:56:57 +01002332 /* Finalize the connection as soon as possible if the peer transport parameters
2333 * have been received. This may be useful to send packets even if this
2334 * handshake fails.
2335 */
2336 if ((qc->flags & QUIC_FL_CONN_TX_TP_RECEIVED) && !qc_conn_finalize(qc, 1)) {
2337 TRACE_ERROR("connection finalization failed", QUIC_EV_CONN_IO_CB, qc, &state);
2338 goto leave;
2339 }
2340
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002341 if (ssl_err != 1) {
2342 ssl_err = SSL_get_error(ctx->ssl, ssl_err);
2343 if (ssl_err == SSL_ERROR_WANT_READ || ssl_err == SSL_ERROR_WANT_WRITE) {
2344 TRACE_PROTO("SSL handshake in progress",
2345 QUIC_EV_CONN_IO_CB, qc, &state, &ssl_err);
2346 goto out;
2347 }
2348
2349 /* TODO: Should close the connection asap */
2350 if (!(qc->flags & QUIC_FL_CONN_HALF_OPEN_CNT_DECREMENTED)) {
2351 qc->flags |= QUIC_FL_CONN_HALF_OPEN_CNT_DECREMENTED;
2352 HA_ATOMIC_DEC(&qc->prx_counters->half_open_conn);
2353 HA_ATOMIC_INC(&qc->prx_counters->hdshk_fail);
2354 }
2355 TRACE_ERROR("SSL handshake error", QUIC_EV_CONN_IO_CB, qc, &state, &ssl_err);
2356 qc_ssl_dump_errors(ctx->conn);
2357 ERR_clear_error();
2358 goto leave;
2359 }
2360
2361 TRACE_PROTO("SSL handshake OK", QUIC_EV_CONN_IO_CB, qc, &state);
2362
2363 /* Check the alpn could be negotiated */
2364 if (!qc->app_ops) {
2365 TRACE_ERROR("No negotiated ALPN", QUIC_EV_CONN_IO_CB, qc, &state);
2366 quic_set_tls_alert(qc, SSL_AD_NO_APPLICATION_PROTOCOL);
2367 goto leave;
2368 }
2369
2370 if (!(qc->flags & QUIC_FL_CONN_HALF_OPEN_CNT_DECREMENTED)) {
2371 TRACE_DEVEL("dec half open counter", QUIC_EV_CONN_IO_CB, qc, &state);
2372 qc->flags |= QUIC_FL_CONN_HALF_OPEN_CNT_DECREMENTED;
2373 HA_ATOMIC_DEC(&qc->prx_counters->half_open_conn);
2374 }
2375 /* I/O callback switch */
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02002376 qc->wait_event.tasklet->process = quic_conn_app_io_cb;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002377 if (qc_is_listener(ctx->qc)) {
2378 qc->state = QUIC_HS_ST_CONFIRMED;
2379 /* The connection is ready to be accepted. */
2380 quic_accept_push_qc(qc);
2381 }
2382 else {
2383 qc->state = QUIC_HS_ST_COMPLETE;
2384 }
2385
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02002386 /* Prepare the next key update */
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002387 if (!quic_tls_key_update(qc)) {
2388 TRACE_ERROR("quic_tls_key_update() failed", QUIC_EV_CONN_IO_CB, qc);
2389 goto leave;
2390 }
2391 } else {
2392 ssl_err = SSL_process_quic_post_handshake(ctx->ssl);
2393 if (ssl_err != 1) {
2394 ssl_err = SSL_get_error(ctx->ssl, ssl_err);
2395 if (ssl_err == SSL_ERROR_WANT_READ || ssl_err == SSL_ERROR_WANT_WRITE) {
2396 TRACE_PROTO("SSL post handshake in progress",
2397 QUIC_EV_CONN_IO_CB, qc, &state, &ssl_err);
2398 goto out;
2399 }
2400
2401 TRACE_ERROR("SSL post handshake error",
2402 QUIC_EV_CONN_IO_CB, qc, &state, &ssl_err);
2403 goto leave;
2404 }
2405
2406 TRACE_STATE("SSL post handshake succeeded", QUIC_EV_CONN_IO_CB, qc, &state);
2407 }
2408
2409 out:
2410 ret = 1;
2411 leave:
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002412 /* The CRYPTO data are consumed even in case of an error to release
2413 * the memory asap.
2414 */
Amaury Denoyelle2f668f02022-11-18 15:24:08 +01002415 if (!ncb_is_null(ncbuf)) {
2416#ifdef DEBUG_STRICT
2417 ncb_ret = ncb_advance(ncbuf, len);
2418 /* ncb_advance() must always succeed. This is guaranteed as
2419 * this is only done inside a data block. If false, this will
2420 * lead to handshake failure with quic_enc_level offset shifted
2421 * from buffer data.
2422 */
2423 BUG_ON(ncb_ret != NCB_RET_OK);
2424#else
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002425 ncb_advance(ncbuf, len);
Amaury Denoyelle2f668f02022-11-18 15:24:08 +01002426#endif
2427 }
2428
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002429 TRACE_LEAVE(QUIC_EV_CONN_SSLDATA, qc);
2430 return ret;
2431}
2432
Amaury Denoyelle2216b082023-02-02 14:59:36 +01002433/* Parse a STREAM frame <strm_frm> received in <pkt> packet for <qc>
2434 * connection. <fin> is true if FIN bit is set on frame type.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002435 *
2436 * Return 1 on success. On error, 0 is returned. In this case, the packet
2437 * containing the frame must not be acknowledged.
2438 */
2439static inline int qc_handle_strm_frm(struct quic_rx_packet *pkt,
2440 struct quic_stream *strm_frm,
Amaury Denoyelle2216b082023-02-02 14:59:36 +01002441 struct quic_conn *qc, char fin)
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002442{
2443 int ret;
2444
2445 /* RFC9000 13.1. Packet Processing
2446 *
2447 * A packet MUST NOT be acknowledged until packet protection has been
2448 * successfully removed and all frames contained in the packet have
2449 * been processed. For STREAM frames, this means the data has been
2450 * enqueued in preparation to be received by the application protocol,
2451 * but it does not require that data be delivered and consumed.
2452 */
2453 TRACE_ENTER(QUIC_EV_CONN_PRSFRM, qc);
2454
2455 ret = qcc_recv(qc->qcc, strm_frm->id, strm_frm->len,
Amaury Denoyelle2216b082023-02-02 14:59:36 +01002456 strm_frm->offset.key, fin, (char *)strm_frm->data);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002457
2458 /* frame rejected - packet must not be acknowledeged */
2459 TRACE_LEAVE(QUIC_EV_CONN_PRSFRM, qc);
2460 return !ret;
2461}
2462
2463/* Duplicate all frames from <pkt_frm_list> list into <out_frm_list> list
2464 * for <qc> QUIC connection.
2465 * This is a best effort function which never fails even if no memory could be
2466 * allocated to duplicate these frames.
2467 */
2468static void qc_dup_pkt_frms(struct quic_conn *qc,
2469 struct list *pkt_frm_list, struct list *out_frm_list)
2470{
2471 struct quic_frame *frm, *frmbak;
2472 struct list tmp = LIST_HEAD_INIT(tmp);
2473
2474 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc);
2475
2476 list_for_each_entry_safe(frm, frmbak, pkt_frm_list, list) {
2477 struct quic_frame *dup_frm, *origin;
2478
2479 switch (frm->type) {
2480 case QUIC_FT_STREAM_8 ... QUIC_FT_STREAM_F:
2481 {
2482 struct quic_stream *strm_frm = &frm->stream;
2483 struct eb64_node *node = NULL;
2484 struct qc_stream_desc *stream_desc;
2485
2486 node = eb64_lookup(&qc->streams_by_id, strm_frm->id);
2487 if (!node) {
2488 TRACE_DEVEL("ignored frame for a released stream", QUIC_EV_CONN_PRSAFRM, qc, frm);
2489 continue;
2490 }
2491
2492 stream_desc = eb64_entry(node, struct qc_stream_desc, by_id);
2493 /* Do not resend this frame if in the "already acked range" */
2494 if (strm_frm->offset.key + strm_frm->len <= stream_desc->ack_offset) {
2495 TRACE_DEVEL("ignored frame in already acked range",
2496 QUIC_EV_CONN_PRSAFRM, qc, frm);
2497 continue;
2498 }
2499 else if (strm_frm->offset.key < stream_desc->ack_offset) {
2500 strm_frm->offset.key = stream_desc->ack_offset;
2501 TRACE_DEVEL("updated partially acked frame",
2502 QUIC_EV_CONN_PRSAFRM, qc, frm);
2503 }
2504 break;
2505 }
2506
2507 default:
2508 break;
2509 }
2510
Amaury Denoyelle40c24f12023-01-27 17:47:49 +01002511 /* If <frm> is already a copy of another frame, we must take
2512 * its original frame as source for the copy.
2513 */
2514 origin = frm->origin ? frm->origin : frm;
2515 dup_frm = qc_frm_dup(origin);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002516 if (!dup_frm) {
2517 TRACE_ERROR("could not duplicate frame", QUIC_EV_CONN_PRSAFRM, qc, frm);
2518 break;
2519 }
2520
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002521 TRACE_DEVEL("built probing frame", QUIC_EV_CONN_PRSAFRM, qc, origin);
Amaury Denoyelle40c24f12023-01-27 17:47:49 +01002522 if (origin->pkt) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002523 TRACE_DEVEL("duplicated from packet", QUIC_EV_CONN_PRSAFRM,
2524 qc, NULL, &origin->pkt->pn_node.key);
Amaury Denoyelle40c24f12023-01-27 17:47:49 +01002525 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002526 else {
2527 /* <origin> is a frame which was sent from a packet detected as lost. */
2528 TRACE_DEVEL("duplicated from lost packet", QUIC_EV_CONN_PRSAFRM, qc);
2529 }
Amaury Denoyelle40c24f12023-01-27 17:47:49 +01002530
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002531 LIST_APPEND(&tmp, &dup_frm->list);
2532 }
2533
2534 LIST_SPLICE(out_frm_list, &tmp);
2535
2536 TRACE_LEAVE(QUIC_EV_CONN_PRSAFRM, qc);
2537}
2538
2539/* Prepare a fast retransmission from <qel> encryption level */
2540static void qc_prep_fast_retrans(struct quic_conn *qc,
2541 struct quic_enc_level *qel,
2542 struct list *frms1, struct list *frms2)
2543{
2544 struct eb_root *pkts = &qel->pktns->tx.pkts;
2545 struct list *frms = frms1;
2546 struct eb64_node *node;
2547 struct quic_tx_packet *pkt;
2548
2549 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc);
2550
2551 BUG_ON(frms1 == frms2);
2552
2553 pkt = NULL;
2554 node = eb64_first(pkts);
2555 start:
2556 while (node) {
2557 pkt = eb64_entry(node, struct quic_tx_packet, pn_node);
2558 node = eb64_next(node);
2559 /* Skip the empty and coalesced packets */
Frédéric Lécaille6dead912023-01-30 17:27:32 +01002560 TRACE_PRINTF(TRACE_LEVEL_DEVELOPER, QUIC_EV_CONN_SPPKTS, qc, 0, 0, 0,
2561 "--> pn=%llu (%d %d)", (ull)pkt->pn_node.key,
2562 LIST_ISEMPTY(&pkt->frms), !!(pkt->flags & QUIC_FL_TX_PACKET_COALESCED));
Frédéric Lécaille055e8262023-01-31 10:10:06 +01002563 if (!LIST_ISEMPTY(&pkt->frms))
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002564 break;
2565 }
2566
2567 if (!pkt)
2568 goto leave;
2569
2570 /* When building a packet from another one, the field which may increase the
2571 * packet size is the packet number. And the maximum increase is 4 bytes.
2572 */
2573 if (!quic_peer_validated_addr(qc) && qc_is_listener(qc) &&
2574 pkt->len + 4 > 3 * qc->rx.bytes - qc->tx.prep_bytes) {
2575 TRACE_PROTO("anti-amplification limit would be reached", QUIC_EV_CONN_SPPKTS, qc, pkt);
2576 goto leave;
2577 }
2578
2579 TRACE_DEVEL("duplicating packet", QUIC_EV_CONN_SPPKTS, qc, pkt);
2580 qc_dup_pkt_frms(qc, &pkt->frms, frms);
2581 if (frms == frms1 && frms2) {
2582 frms = frms2;
2583 goto start;
2584 }
2585 leave:
2586 TRACE_LEAVE(QUIC_EV_CONN_SPPKTS, qc);
2587}
2588
2589/* Prepare a fast retransmission during a handshake after a client
2590 * has resent Initial packets. According to the RFC a server may retransmit
2591 * Initial packets send them coalescing with others (Handshake here).
2592 * (Listener only function).
2593 */
2594static void qc_prep_hdshk_fast_retrans(struct quic_conn *qc,
2595 struct list *ifrms, struct list *hfrms)
2596{
2597 struct list itmp = LIST_HEAD_INIT(itmp);
2598 struct list htmp = LIST_HEAD_INIT(htmp);
2599
2600 struct quic_enc_level *iqel = &qc->els[QUIC_TLS_ENC_LEVEL_INITIAL];
2601 struct quic_enc_level *hqel = &qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE];
2602 struct quic_enc_level *qel = iqel;
2603 struct eb_root *pkts;
2604 struct eb64_node *node;
2605 struct quic_tx_packet *pkt;
2606 struct list *tmp = &itmp;
2607
2608 TRACE_ENTER(QUIC_EV_CONN_PRSAFRM, qc);
2609 start:
2610 pkt = NULL;
2611 pkts = &qel->pktns->tx.pkts;
2612 node = eb64_first(pkts);
2613 /* Skip the empty packet (they have already been retransmitted) */
2614 while (node) {
2615 pkt = eb64_entry(node, struct quic_tx_packet, pn_node);
Frédéric Lécaille6dead912023-01-30 17:27:32 +01002616 TRACE_PRINTF(TRACE_LEVEL_DEVELOPER, QUIC_EV_CONN_SPPKTS, qc, 0, 0, 0,
2617 "--> pn=%llu (%d %d)", (ull)pkt->pn_node.key,
2618 LIST_ISEMPTY(&pkt->frms), !!(pkt->flags & QUIC_FL_TX_PACKET_COALESCED));
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002619 if (!LIST_ISEMPTY(&pkt->frms) && !(pkt->flags & QUIC_FL_TX_PACKET_COALESCED))
2620 break;
2621 node = eb64_next(node);
2622 }
2623
2624 if (!pkt)
2625 goto end;
2626
2627 /* When building a packet from another one, the field which may increase the
2628 * packet size is the packet number. And the maximum increase is 4 bytes.
2629 */
2630 if (!quic_peer_validated_addr(qc) && qc_is_listener(qc) &&
2631 pkt->len + 4 > 3 * qc->rx.bytes - qc->tx.prep_bytes) {
2632 TRACE_PROTO("anti-amplification limit would be reached", QUIC_EV_CONN_PRSAFRM, qc);
2633 goto end;
2634 }
2635
2636 qel->pktns->tx.pto_probe += 1;
2637
2638 /* No risk to loop here, #packet per datagram is bounded */
2639 requeue:
2640 TRACE_DEVEL("duplicating packet", QUIC_EV_CONN_PRSAFRM, qc, NULL, &pkt->pn_node.key);
2641 qc_dup_pkt_frms(qc, &pkt->frms, tmp);
2642 if (qel == iqel) {
2643 if (pkt->next && pkt->next->type == QUIC_PACKET_TYPE_HANDSHAKE) {
2644 pkt = pkt->next;
2645 tmp = &htmp;
2646 hqel->pktns->tx.pto_probe += 1;
2647 TRACE_DEVEL("looping for next packet", QUIC_EV_CONN_PRSAFRM, qc);
2648 goto requeue;
2649 }
2650 }
2651
2652 end:
2653 LIST_SPLICE(ifrms, &itmp);
2654 LIST_SPLICE(hfrms, &htmp);
2655
2656 TRACE_LEAVE(QUIC_EV_CONN_PRSAFRM, qc);
2657}
2658
2659static void qc_cc_err_count_inc(struct quic_conn *qc, struct quic_frame *frm)
2660{
2661 TRACE_ENTER(QUIC_EV_CONN_CLOSE, qc);
2662
2663 if (frm->type == QUIC_FT_CONNECTION_CLOSE)
2664 quic_stats_transp_err_count_inc(qc->prx_counters, frm->connection_close.error_code);
2665 else if (frm->type == QUIC_FT_CONNECTION_CLOSE_APP) {
2666 if (qc->mux_state != QC_MUX_READY || !qc->qcc->app_ops->inc_err_cnt)
2667 goto out;
2668
2669 qc->qcc->app_ops->inc_err_cnt(qc->qcc->ctx, frm->connection_close_app.error_code);
2670 }
2671
2672 out:
2673 TRACE_LEAVE(QUIC_EV_CONN_CLOSE, qc);
2674}
2675
2676/* Enqueue a STOP_SENDING frame to send into 1RTT packet number space
2677 * frame list to send.
2678 * Return 1 if succeeded, 0 if not.
2679 */
2680static int qc_stop_sending_frm_enqueue(struct quic_conn *qc, uint64_t id)
2681{
2682 int ret = 0;
2683 struct quic_frame *frm;
2684 struct quic_enc_level *qel = &qc->els[QUIC_TLS_ENC_LEVEL_APP];
2685 uint64_t app_error_code;
2686
2687 TRACE_ENTER(QUIC_EV_CONN_PRSHPKT, qc);
2688
2689 /* TODO: the mux may be released, we cannot have more
2690 * information about the application error code to send
2691 * at this time.
2692 */
2693 app_error_code = H3_REQUEST_REJECTED;
Amaury Denoyelle40c24f12023-01-27 17:47:49 +01002694 frm = qc_frm_alloc(QUIC_FT_STOP_SENDING);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002695 if (!frm) {
2696 TRACE_ERROR("failed to allocate quic_frame", QUIC_EV_CONN_PRSHPKT, qc);
2697 goto out;
2698 }
2699
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002700 frm->stop_sending.id = id;
2701 frm->stop_sending.app_error_code = app_error_code;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002702 LIST_APPEND(&qel->pktns->tx.frms, &frm->list);
2703 ret = 1;
2704 out:
2705 TRACE_LEAVE(QUIC_EV_CONN_PRSHPKT, qc);
2706 return ret;
2707}
2708
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002709/* Release the underlying memory use by <ncbuf> non-contiguous buffer */
2710static void quic_free_ncbuf(struct ncbuf *ncbuf)
2711{
2712 struct buffer buf;
2713
2714 if (ncb_is_null(ncbuf))
2715 return;
2716
2717 buf = b_make(ncbuf->area, ncbuf->size, 0, 0);
2718 b_free(&buf);
2719 offer_buffers(NULL, 1);
2720
2721 *ncbuf = NCBUF_NULL;
2722}
2723
2724/* Allocate the underlying required memory for <ncbuf> non-contiguous buffer */
2725static struct ncbuf *quic_get_ncbuf(struct ncbuf *ncbuf)
2726{
2727 struct buffer buf = BUF_NULL;
2728
2729 if (!ncb_is_null(ncbuf))
2730 return ncbuf;
2731
2732 b_alloc(&buf);
2733 BUG_ON(b_is_null(&buf));
2734
2735 *ncbuf = ncb_make(buf.area, buf.size, 0);
2736 ncb_init(ncbuf, 0);
2737
2738 return ncbuf;
2739}
2740
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002741/* Parse <frm> CRYPTO frame coming with <pkt> packet at <qel> <qc> connectionn.
2742 * Returns 1 if succeeded, 0 if not. Also set <*fast_retrans> to 1 if the
2743 * speed up handshake completion may be run after having received duplicated
2744 * CRYPTO data.
2745 */
2746static int qc_handle_crypto_frm(struct quic_conn *qc,
2747 struct quic_crypto *frm, struct quic_rx_packet *pkt,
2748 struct quic_enc_level *qel, int *fast_retrans)
2749{
2750 int ret = 0;
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002751 enum ncb_ret ncb_ret;
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002752 /* XXX TO DO: <cfdebug> is used only for the traces. */
2753 struct quic_rx_crypto_frm cfdebug = {
2754 .offset_node.key = frm->offset,
2755 .len = frm->len,
2756 };
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002757 struct quic_cstream *cstream = qel->cstream;
2758 struct ncbuf *ncbuf = &qel->cstream->rx.ncbuf;
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002759
2760 TRACE_ENTER(QUIC_EV_CONN_PRSHPKT, qc);
2761 if (unlikely(qel->tls_ctx.flags & QUIC_FL_TLS_SECRETS_DCD)) {
2762 TRACE_PROTO("CRYPTO data discarded",
2763 QUIC_EV_CONN_RXPKT, qc, pkt, &cfdebug);
2764 goto done;
2765 }
2766
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002767 if (unlikely(frm->offset < cstream->rx.offset)) {
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002768 size_t diff;
2769
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002770 if (frm->offset + frm->len <= cstream->rx.offset) {
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002771 /* Nothing to do */
2772 TRACE_PROTO("Already received CRYPTO data",
2773 QUIC_EV_CONN_RXPKT, qc, pkt, &cfdebug);
2774 if (qc_is_listener(qc) && qel == &qc->els[QUIC_TLS_ENC_LEVEL_INITIAL] &&
2775 !(qc->flags & QUIC_FL_CONN_HANDSHAKE_SPEED_UP))
2776 *fast_retrans = 1;
2777 goto done;
2778 }
2779
2780 TRACE_PROTO("Partially already received CRYPTO data",
2781 QUIC_EV_CONN_RXPKT, qc, pkt, &cfdebug);
2782
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002783 diff = cstream->rx.offset - frm->offset;
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002784 frm->len -= diff;
2785 frm->data += diff;
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002786 frm->offset = cstream->rx.offset;
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002787 }
2788
Amaury Denoyelleff95f2d2022-11-18 14:50:06 +01002789 if (frm->offset == cstream->rx.offset && ncb_is_empty(ncbuf)) {
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002790 if (!qc_provide_cdata(qel, qc->xprt_ctx, frm->data, frm->len,
2791 pkt, &cfdebug)) {
2792 // trace already emitted by function above
2793 goto leave;
2794 }
2795
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002796 cstream->rx.offset += frm->len;
Amaury Denoyelle2f668f02022-11-18 15:24:08 +01002797 TRACE_DEVEL("increment crypto level offset", QUIC_EV_CONN_PHPKTS, qc, qel);
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002798 goto done;
2799 }
2800
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002801 if (!quic_get_ncbuf(ncbuf) ||
2802 ncb_is_null(ncbuf)) {
2803 TRACE_ERROR("CRYPTO ncbuf allocation failed", QUIC_EV_CONN_PRSHPKT, qc);
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002804 goto leave;
2805 }
2806
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02002807 /* frm->offset > cstream-trx.offset */
2808 ncb_ret = ncb_add(ncbuf, frm->offset - cstream->rx.offset,
2809 (const char *)frm->data, frm->len, NCB_ADD_COMPARE);
2810 if (ncb_ret != NCB_RET_OK) {
2811 if (ncb_ret == NCB_RET_DATA_REJ) {
2812 TRACE_ERROR("overlapping data rejected", QUIC_EV_CONN_PRSHPKT, qc);
2813 quic_set_connection_close(qc, quic_err_transport(QC_ERR_PROTOCOL_VIOLATION));
2814 }
2815 else if (ncb_ret == NCB_RET_GAP_SIZE) {
2816 TRACE_ERROR("cannot bufferize frame due to gap size limit",
2817 QUIC_EV_CONN_PRSHPKT, qc);
2818 }
2819 goto leave;
2820 }
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002821
2822 done:
2823 ret = 1;
2824 leave:
2825 TRACE_LEAVE(QUIC_EV_CONN_PRSHPKT, qc);
2826 return ret;
2827}
2828
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02002829/* Parse all the frames of <pkt> QUIC packet for QUIC connection <qc> and <qel>
2830 * as encryption level.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002831 * Returns 1 if succeeded, 0 if failed.
2832 */
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02002833static int qc_parse_pkt_frms(struct quic_conn *qc, struct quic_rx_packet *pkt,
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002834 struct quic_enc_level *qel)
2835{
2836 struct quic_frame frm;
2837 const unsigned char *pos, *end;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002838 int fast_retrans = 0, ret = 0;
2839
2840 TRACE_ENTER(QUIC_EV_CONN_PRSHPKT, qc);
2841 /* Skip the AAD */
2842 pos = pkt->data + pkt->aad_len;
2843 end = pkt->data + pkt->len;
2844
2845 while (pos < end) {
2846 if (!qc_parse_frm(&frm, pkt, &pos, end, qc)) {
2847 // trace already emitted by function above
2848 goto leave;
2849 }
2850
2851 TRACE_PROTO("RX frame", QUIC_EV_CONN_PSTRM, qc, &frm);
2852 switch (frm.type) {
2853 case QUIC_FT_PADDING:
2854 break;
2855 case QUIC_FT_PING:
2856 break;
2857 case QUIC_FT_ACK:
2858 {
2859 unsigned int rtt_sample;
2860
2861 rtt_sample = 0;
2862 if (!qc_parse_ack_frm(qc, &frm, qel, &rtt_sample, &pos, end)) {
2863 // trace already emitted by function above
2864 goto leave;
2865 }
2866
2867 if (rtt_sample) {
2868 unsigned int ack_delay;
2869
2870 ack_delay = !quic_application_pktns(qel->pktns, qc) ? 0 :
2871 qc->state >= QUIC_HS_ST_CONFIRMED ?
2872 MS_TO_TICKS(QUIC_MIN(quic_ack_delay_ms(&frm.ack, qc), qc->max_ack_delay)) :
2873 MS_TO_TICKS(quic_ack_delay_ms(&frm.ack, qc));
2874 quic_loss_srtt_update(&qc->path->loss, rtt_sample, ack_delay, qc);
2875 }
2876 break;
2877 }
2878 case QUIC_FT_RESET_STREAM:
Amaury Denoyelle5854fc02022-12-09 16:25:48 +01002879 if (qc->mux_state == QC_MUX_READY) {
2880 struct quic_reset_stream *rs = &frm.reset_stream;
2881 qcc_recv_reset_stream(qc->qcc, rs->id, rs->app_error_code, rs->final_size);
2882 }
2883 break;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002884 case QUIC_FT_STOP_SENDING:
2885 {
2886 struct quic_stop_sending *stop_sending = &frm.stop_sending;
2887 if (qc->mux_state == QC_MUX_READY) {
2888 if (qcc_recv_stop_sending(qc->qcc, stop_sending->id,
2889 stop_sending->app_error_code)) {
2890 TRACE_ERROR("qcc_recv_stop_sending() failed", QUIC_EV_CONN_PRSHPKT, qc);
2891 goto leave;
2892 }
2893 }
2894 break;
2895 }
2896 case QUIC_FT_CRYPTO:
Frédéric Lécaillea20c93e2022-09-12 14:54:45 +02002897 if (!qc_handle_crypto_frm(qc, &frm.crypto, pkt, qel, &fast_retrans))
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002898 goto leave;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002899 break;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002900 case QUIC_FT_STREAM_8 ... QUIC_FT_STREAM_F:
2901 {
2902 struct quic_stream *stream = &frm.stream;
2903 unsigned nb_streams = qc->rx.strms[qcs_id_type(stream->id)].nb_streams;
Amaury Denoyelle2216b082023-02-02 14:59:36 +01002904 const char fin = frm.type & QUIC_STREAM_FRAME_TYPE_FIN_BIT;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002905
2906 /* The upper layer may not be allocated. */
2907 if (qc->mux_state != QC_MUX_READY) {
2908 if ((stream->id >> QCS_ID_TYPE_SHIFT) < nb_streams) {
2909 TRACE_DATA("Already closed stream", QUIC_EV_CONN_PRSHPKT, qc);
2910 break;
2911 }
2912 else {
2913 TRACE_DEVEL("No mux for new stream", QUIC_EV_CONN_PRSHPKT, qc);
Frédéric Lécailled18025e2023-01-20 15:33:50 +01002914 if (qc->app_ops == &h3_ops && quic_stream_is_uni(stream->id)) {
2915 /* Do not send STOP_SENDING frames for h3 unidirectional streams.
2916 * TODO: this test should be removed when the connection closure
2917 * will be more clean.
2918 * At quic_conn level there is no mean to know that an application
2919 * want to forbid stream closure requests to receivers. This is the
2920 * case for the Control and QPACK h3 unidirectional streams.
2921 */
2922 goto leave;
2923 }
2924
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002925 if (!qc_stop_sending_frm_enqueue(qc, stream->id))
2926 TRACE_ERROR("could not enqueue STOP_SENDING frame", QUIC_EV_CONN_PRSHPKT, qc);
2927 /* This packet will not be acknowledged */
2928 goto leave;
2929 }
2930 }
2931
Amaury Denoyelle2216b082023-02-02 14:59:36 +01002932 if (!qc_handle_strm_frm(pkt, stream, qc, fin)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02002933 TRACE_ERROR("qc_handle_strm_frm() failed", QUIC_EV_CONN_PRSHPKT, qc);
2934 goto leave;
2935 }
2936
2937 break;
2938 }
2939 case QUIC_FT_MAX_DATA:
2940 if (qc->mux_state == QC_MUX_READY) {
2941 struct quic_max_data *data = &frm.max_data;
2942 qcc_recv_max_data(qc->qcc, data->max_data);
2943 }
2944 break;
2945 case QUIC_FT_MAX_STREAM_DATA:
2946 if (qc->mux_state == QC_MUX_READY) {
2947 struct quic_max_stream_data *data = &frm.max_stream_data;
2948 if (qcc_recv_max_stream_data(qc->qcc, data->id,
2949 data->max_stream_data)) {
2950 TRACE_ERROR("qcc_recv_max_stream_data() failed", QUIC_EV_CONN_PRSHPKT, qc);
2951 goto leave;
2952 }
2953 }
2954 break;
2955 case QUIC_FT_MAX_STREAMS_BIDI:
2956 case QUIC_FT_MAX_STREAMS_UNI:
2957 break;
2958 case QUIC_FT_DATA_BLOCKED:
2959 HA_ATOMIC_INC(&qc->prx_counters->data_blocked);
2960 break;
2961 case QUIC_FT_STREAM_DATA_BLOCKED:
2962 HA_ATOMIC_INC(&qc->prx_counters->stream_data_blocked);
2963 break;
2964 case QUIC_FT_STREAMS_BLOCKED_BIDI:
2965 HA_ATOMIC_INC(&qc->prx_counters->streams_data_blocked_bidi);
2966 break;
2967 case QUIC_FT_STREAMS_BLOCKED_UNI:
2968 HA_ATOMIC_INC(&qc->prx_counters->streams_data_blocked_uni);
2969 break;
2970 case QUIC_FT_NEW_CONNECTION_ID:
2971 case QUIC_FT_RETIRE_CONNECTION_ID:
2972 /* XXX TO DO XXX */
2973 break;
2974 case QUIC_FT_CONNECTION_CLOSE:
2975 case QUIC_FT_CONNECTION_CLOSE_APP:
2976 /* Increment the error counters */
2977 qc_cc_err_count_inc(qc, &frm);
2978 if (!(qc->flags & QUIC_FL_CONN_DRAINING)) {
2979 if (!(qc->flags & QUIC_FL_CONN_HALF_OPEN_CNT_DECREMENTED)) {
2980 qc->flags |= QUIC_FL_CONN_HALF_OPEN_CNT_DECREMENTED;
2981 HA_ATOMIC_DEC(&qc->prx_counters->half_open_conn);
2982 }
2983 TRACE_STATE("Entering draining state", QUIC_EV_CONN_PRSHPKT, qc);
2984 /* RFC 9000 10.2. Immediate Close:
2985 * The closing and draining connection states exist to ensure
2986 * that connections close cleanly and that delayed or reordered
2987 * packets are properly discarded. These states SHOULD persist
2988 * for at least three times the current PTO interval...
2989 *
2990 * Rearm the idle timeout only one time when entering draining
2991 * state.
2992 */
2993 qc_idle_timer_do_rearm(qc);
2994 qc->flags |= QUIC_FL_CONN_DRAINING|QUIC_FL_CONN_IMMEDIATE_CLOSE;
2995 qc_notify_close(qc);
2996 }
2997 break;
2998 case QUIC_FT_HANDSHAKE_DONE:
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02002999 if (qc_is_listener(qc)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003000 TRACE_ERROR("non accepted QUIC_FT_HANDSHAKE_DONE frame",
3001 QUIC_EV_CONN_PRSHPKT, qc);
3002 goto leave;
3003 }
3004
3005 qc->state = QUIC_HS_ST_CONFIRMED;
3006 break;
3007 default:
3008 TRACE_ERROR("unknosw frame type", QUIC_EV_CONN_PRSHPKT, qc);
3009 goto leave;
3010 }
3011 }
3012
3013 /* Flag this packet number space as having received a packet. */
3014 qel->pktns->flags |= QUIC_FL_PKTNS_PKT_RECEIVED;
3015
3016 if (fast_retrans) {
3017 struct quic_enc_level *iqel = &qc->els[QUIC_TLS_ENC_LEVEL_INITIAL];
3018 struct quic_enc_level *hqel = &qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE];
3019
3020 TRACE_PROTO("speeding up handshake completion", QUIC_EV_CONN_PRSHPKT, qc);
3021 qc_prep_hdshk_fast_retrans(qc, &iqel->pktns->tx.frms, &hqel->pktns->tx.frms);
3022 qc->flags |= QUIC_FL_CONN_HANDSHAKE_SPEED_UP;
3023 }
3024
3025 /* The server must switch from INITIAL to HANDSHAKE handshake state when it
3026 * has successfully parse a Handshake packet. The Initial encryption must also
3027 * be discarded.
3028 */
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02003029 if (pkt->type == QUIC_PACKET_TYPE_HANDSHAKE && qc_is_listener(qc)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003030 if (qc->state >= QUIC_HS_ST_SERVER_INITIAL) {
3031 if (!(qc->els[QUIC_TLS_ENC_LEVEL_INITIAL].tls_ctx.flags &
3032 QUIC_FL_TLS_SECRETS_DCD)) {
3033 quic_tls_discard_keys(&qc->els[QUIC_TLS_ENC_LEVEL_INITIAL]);
3034 TRACE_PROTO("discarding Initial pktns", QUIC_EV_CONN_PRSHPKT, qc);
3035 quic_pktns_discard(qc->els[QUIC_TLS_ENC_LEVEL_INITIAL].pktns, qc);
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02003036 qc_set_timer(qc);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003037 qc_el_rx_pkts_del(&qc->els[QUIC_TLS_ENC_LEVEL_INITIAL]);
3038 qc_release_pktns_frms(qc, qc->els[QUIC_TLS_ENC_LEVEL_INITIAL].pktns);
3039 }
3040 if (qc->state < QUIC_HS_ST_SERVER_HANDSHAKE)
3041 qc->state = QUIC_HS_ST_SERVER_HANDSHAKE;
3042 }
3043 }
3044
3045 ret = 1;
3046 leave:
3047 TRACE_LEAVE(QUIC_EV_CONN_PRSHPKT, qc);
3048 return ret;
3049}
3050
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02003051
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003052/* Allocate Tx buffer from <qc> quic-conn if needed.
3053 *
3054 * Returns allocated buffer or NULL on error.
3055 */
3056static struct buffer *qc_txb_alloc(struct quic_conn *qc)
3057{
3058 struct buffer *buf = &qc->tx.buf;
3059 if (!b_alloc(buf))
3060 return NULL;
3061
3062 return buf;
3063}
3064
3065/* Free Tx buffer from <qc> if it is empty. */
3066static void qc_txb_release(struct quic_conn *qc)
3067{
3068 struct buffer *buf = &qc->tx.buf;
3069
3070 /* For the moment sending function is responsible to purge the buffer
3071 * entirely. It may change in the future but this requires to be able
3072 * to reuse old data.
3073 */
3074 BUG_ON_HOT(buf && b_data(buf));
3075
3076 if (!b_data(buf)) {
3077 b_free(buf);
3078 offer_buffers(NULL, 1);
3079 }
3080}
3081
3082/* Commit a datagram payload written into <buf> of length <length>. <first_pkt>
3083 * must contains the address of the first packet stored in the payload.
3084 *
3085 * Caller is responsible that there is enough space in the buffer.
3086 */
3087static void qc_txb_store(struct buffer *buf, uint16_t length,
3088 struct quic_tx_packet *first_pkt)
3089{
3090 const size_t hdlen = sizeof(uint16_t) + sizeof(void *);
3091 BUG_ON_HOT(b_contig_space(buf) < hdlen); /* this must not happen */
3092
3093 write_u16(b_tail(buf), length);
3094 write_ptr(b_tail(buf) + sizeof(length), first_pkt);
3095 b_add(buf, hdlen + length);
3096}
3097
3098/* Returns 1 if a packet may be built for <qc> from <qel> encryption level
3099 * with <frms> as ack-eliciting frame list to send, 0 if not.
3100 * <cc> must equal to 1 if an immediate close was asked, 0 if not.
3101 * <probe> must equalt to 1 if a probing packet is required, 0 if not.
3102 * <force_ack> may be set to 1 if you want to force an ack.
3103 */
3104static int qc_may_build_pkt(struct quic_conn *qc, struct list *frms,
3105 struct quic_enc_level *qel, int cc, int probe, int force_ack)
3106{
3107 unsigned int must_ack = force_ack ||
3108 (LIST_ISEMPTY(frms) && (qel->pktns->flags & QUIC_FL_PKTNS_ACK_REQUIRED));
3109
3110 /* Do not build any more packet if the TX secrets are not available or
3111 * if there is nothing to send, i.e. if no CONNECTION_CLOSE or ACK are required
3112 * and if there is no more packets to send upon PTO expiration
3113 * and if there is no more ack-eliciting frames to send or in flight
3114 * congestion control limit is reached for prepared data
3115 */
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02003116 if (!quic_tls_has_tx_sec(qel) ||
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003117 (!cc && !probe && !must_ack &&
3118 (LIST_ISEMPTY(frms) || qc->path->prep_in_flight >= qc->path->cwnd))) {
3119 return 0;
3120 }
3121
3122 return 1;
3123}
3124
3125/* Prepare as much as possible QUIC packets for sending from prebuilt frames
3126 * <frms>. Each packet is stored in a distinct datagram written to <buf>.
3127 *
3128 * Each datagram is prepended by a two fields header : the datagram length and
3129 * the address of the packet contained in the datagram.
3130 *
3131 * Returns the number of bytes prepared in packets if succeeded (may be 0), or
3132 * -1 if something wrong happened.
3133 */
3134static int qc_prep_app_pkts(struct quic_conn *qc, struct buffer *buf,
3135 struct list *frms)
3136{
3137 int ret = -1;
3138 struct quic_enc_level *qel;
3139 unsigned char *end, *pos;
3140 struct quic_tx_packet *pkt;
3141 size_t total;
3142 /* Each datagram is prepended with its length followed by the address
3143 * of the first packet in the datagram.
3144 */
3145 const size_t dg_headlen = sizeof(uint16_t) + sizeof(pkt);
3146
3147 TRACE_ENTER(QUIC_EV_CONN_PHPKTS, qc);
3148
3149 qel = &qc->els[QUIC_TLS_ENC_LEVEL_APP];
3150 total = 0;
3151 pos = (unsigned char *)b_tail(buf);
3152 while (b_contig_space(buf) >= (int)qc->path->mtu + dg_headlen) {
3153 int err, probe, cc;
3154
3155 TRACE_POINT(QUIC_EV_CONN_PHPKTS, qc, qel);
3156 probe = 0;
3157 cc = qc->flags & QUIC_FL_CONN_IMMEDIATE_CLOSE;
3158 /* We do not probe if an immediate close was asked */
3159 if (!cc)
3160 probe = qel->pktns->tx.pto_probe;
3161
3162 if (!qc_may_build_pkt(qc, frms, qel, cc, probe, 0))
3163 break;
3164
3165 /* Leave room for the datagram header */
3166 pos += dg_headlen;
3167 if (!quic_peer_validated_addr(qc) && qc_is_listener(qc)) {
3168 end = pos + QUIC_MIN((uint64_t)qc->path->mtu, 3 * qc->rx.bytes - qc->tx.prep_bytes);
3169 }
3170 else {
3171 end = pos + qc->path->mtu;
3172 }
3173
3174 pkt = qc_build_pkt(&pos, end, qel, &qel->tls_ctx, frms, qc, NULL, 0,
3175 QUIC_PACKET_TYPE_SHORT, 0, 0, probe, cc, &err);
3176 switch (err) {
3177 case -2:
3178 // trace already emitted by function above
3179 goto leave;
3180 case -1:
3181 /* As we provide qc_build_pkt() with an enough big buffer to fulfill an
3182 * MTU, we are here because of the congestion control window. There is
3183 * no need to try to reuse this buffer.
3184 */
3185 TRACE_DEVEL("could not prepare anymore packet", QUIC_EV_CONN_PHPKTS, qc);
3186 goto out;
3187 default:
3188 break;
3189 }
3190
3191 /* This is to please to GCC. We cannot have (err >= 0 && !pkt) */
3192 BUG_ON(!pkt);
3193
3194 if (qc->flags & QUIC_FL_CONN_RETRANS_OLD_DATA)
3195 pkt->flags |= QUIC_FL_TX_PACKET_PROBE_WITH_OLD_DATA;
3196
3197 total += pkt->len;
3198
3199 /* Write datagram header. */
3200 qc_txb_store(buf, pkt->len, pkt);
3201 }
3202
3203 out:
3204 ret = total;
3205 leave:
3206 TRACE_LEAVE(QUIC_EV_CONN_PHPKTS, qc);
3207 return ret;
3208}
3209
3210/* Prepare as much as possible QUIC packets for sending from prebuilt frames
3211 * <frms>. Several packets can be regrouped in a single datagram. The result is
3212 * written into <buf>.
3213 *
3214 * Each datagram is prepended by a two fields header : the datagram length and
3215 * the address of first packet in the datagram.
3216 *
3217 * Returns the number of bytes prepared in packets if succeeded (may be 0), or
3218 * -1 if something wrong happened.
3219 */
3220static int qc_prep_pkts(struct quic_conn *qc, struct buffer *buf,
3221 enum quic_tls_enc_level tel, struct list *tel_frms,
3222 enum quic_tls_enc_level next_tel, struct list *next_tel_frms)
3223{
3224 struct quic_enc_level *qel;
3225 unsigned char *end, *pos;
3226 struct quic_tx_packet *first_pkt, *cur_pkt, *prv_pkt;
3227 /* length of datagrams */
3228 uint16_t dglen;
3229 size_t total;
3230 int ret = -1, padding;
3231 /* Each datagram is prepended with its length followed by the address
3232 * of the first packet in the datagram.
3233 */
3234 const size_t dg_headlen = sizeof(uint16_t) + sizeof(first_pkt);
3235 struct list *frms;
3236
3237 TRACE_ENTER(QUIC_EV_CONN_PHPKTS, qc);
3238
3239 /* Currently qc_prep_pkts() does not handle buffer wrapping so the
Ilya Shipitsin4a689da2022-10-29 09:34:32 +05003240 * caller must ensure that buf is reset.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003241 */
3242 BUG_ON_HOT(buf->head || buf->data);
3243
3244 total = 0;
3245 qel = &qc->els[tel];
3246 frms = tel_frms;
3247 dglen = 0;
3248 padding = 0;
3249 pos = (unsigned char *)b_head(buf);
3250 first_pkt = prv_pkt = NULL;
3251 while (b_contig_space(buf) >= (int)qc->path->mtu + dg_headlen || prv_pkt) {
3252 int err, probe, cc;
3253 enum quic_pkt_type pkt_type;
3254 struct quic_tls_ctx *tls_ctx;
3255 const struct quic_version *ver;
3256 int force_ack = (qel->pktns->flags & QUIC_FL_PKTNS_ACK_REQUIRED) &&
3257 (qel == &qc->els[QUIC_TLS_ENC_LEVEL_INITIAL] ||
3258 qel == &qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE]);
3259
3260 TRACE_POINT(QUIC_EV_CONN_PHPKTS, qc, qel);
3261 probe = 0;
3262 cc = qc->flags & QUIC_FL_CONN_IMMEDIATE_CLOSE;
3263 /* We do not probe if an immediate close was asked */
3264 if (!cc)
3265 probe = qel->pktns->tx.pto_probe;
3266
3267 if (!qc_may_build_pkt(qc, frms, qel, cc, probe, force_ack)) {
3268 if (prv_pkt)
3269 qc_txb_store(buf, dglen, first_pkt);
3270 /* Let's select the next encryption level */
3271 if (tel != next_tel && next_tel != QUIC_TLS_ENC_LEVEL_NONE) {
3272 tel = next_tel;
3273 frms = next_tel_frms;
3274 qel = &qc->els[tel];
3275 /* Build a new datagram */
3276 prv_pkt = NULL;
3277 TRACE_DEVEL("next encryption level selected", QUIC_EV_CONN_PHPKTS, qc);
3278 continue;
3279 }
3280 break;
3281 }
3282
3283 pkt_type = quic_tls_level_pkt_type(tel);
3284 if (!prv_pkt) {
3285 /* Leave room for the datagram header */
3286 pos += dg_headlen;
3287 if (!quic_peer_validated_addr(qc) && qc_is_listener(qc)) {
3288 end = pos + QUIC_MIN((uint64_t)qc->path->mtu, 3 * qc->rx.bytes - qc->tx.prep_bytes);
3289 }
3290 else {
3291 end = pos + qc->path->mtu;
3292 }
3293 }
3294
3295 if (qc->negotiated_version) {
3296 ver = qc->negotiated_version;
3297 if (qel == &qc->els[QUIC_TLS_ENC_LEVEL_INITIAL])
3298 tls_ctx = &qc->negotiated_ictx;
3299 else
3300 tls_ctx = &qel->tls_ctx;
3301 }
3302 else {
3303 ver = qc->original_version;
3304 tls_ctx = &qel->tls_ctx;
3305 }
3306
3307 cur_pkt = qc_build_pkt(&pos, end, qel, tls_ctx, frms,
3308 qc, ver, dglen, pkt_type,
3309 force_ack, padding, probe, cc, &err);
3310 switch (err) {
3311 case -2:
3312 // trace already emitted by function above
3313 goto leave;
3314 case -1:
3315 /* If there was already a correct packet present, set the
3316 * current datagram as prepared into <cbuf>.
3317 */
3318 if (prv_pkt)
3319 qc_txb_store(buf, dglen, first_pkt);
3320 TRACE_DEVEL("could not prepare anymore packet", QUIC_EV_CONN_PHPKTS, qc);
3321 goto out;
3322 default:
3323 break;
3324 }
3325
3326 /* This is to please to GCC. We cannot have (err >= 0 && !cur_pkt) */
3327 BUG_ON(!cur_pkt);
3328
3329 if (qc->flags & QUIC_FL_CONN_RETRANS_OLD_DATA)
3330 cur_pkt->flags |= QUIC_FL_TX_PACKET_PROBE_WITH_OLD_DATA;
3331
3332 total += cur_pkt->len;
3333 /* keep trace of the first packet in the datagram */
3334 if (!first_pkt)
3335 first_pkt = cur_pkt;
Frédéric Lécaille74b5f7b2022-11-20 18:35:35 +01003336 /* Attach the current one to the previous one and vice versa */
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003337 if (prv_pkt) {
3338 prv_pkt->next = cur_pkt;
Frédéric Lécaille814645f2022-11-18 18:15:28 +01003339 cur_pkt->prev = prv_pkt;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003340 cur_pkt->flags |= QUIC_FL_TX_PACKET_COALESCED;
3341 }
3342 /* Let's say we have to build a new dgram */
3343 prv_pkt = NULL;
3344 dglen += cur_pkt->len;
3345 /* Client: discard the Initial encryption keys as soon as
3346 * a handshake packet could be built.
3347 */
3348 if (qc->state == QUIC_HS_ST_CLIENT_INITIAL &&
3349 pkt_type == QUIC_PACKET_TYPE_HANDSHAKE) {
3350 quic_tls_discard_keys(&qc->els[QUIC_TLS_ENC_LEVEL_INITIAL]);
3351 TRACE_PROTO("discarding Initial pktns", QUIC_EV_CONN_PHPKTS, qc);
3352 quic_pktns_discard(qc->els[QUIC_TLS_ENC_LEVEL_INITIAL].pktns, qc);
3353 qc_set_timer(qc);
3354 qc_el_rx_pkts_del(&qc->els[QUIC_TLS_ENC_LEVEL_INITIAL]);
3355 qc_release_pktns_frms(qc, qc->els[QUIC_TLS_ENC_LEVEL_INITIAL].pktns);
3356 qc->state = QUIC_HS_ST_CLIENT_HANDSHAKE;
3357 }
3358 /* If the data for the current encryption level have all been sent,
3359 * select the next level.
3360 */
3361 if ((tel == QUIC_TLS_ENC_LEVEL_INITIAL || tel == QUIC_TLS_ENC_LEVEL_HANDSHAKE) &&
3362 next_tel != QUIC_TLS_ENC_LEVEL_NONE && (LIST_ISEMPTY(frms) && !qel->pktns->tx.pto_probe)) {
3363 /* If QUIC_TLS_ENC_LEVEL_HANDSHAKE was already reached let's try QUIC_TLS_ENC_LEVEL_APP */
3364 if (tel == QUIC_TLS_ENC_LEVEL_HANDSHAKE && next_tel == tel)
3365 next_tel = QUIC_TLS_ENC_LEVEL_APP;
3366 tel = next_tel;
3367 if (tel == QUIC_TLS_ENC_LEVEL_APP)
3368 frms = &qc->els[tel].pktns->tx.frms;
3369 else
3370 frms = next_tel_frms;
3371 qel = &qc->els[tel];
3372 if (!LIST_ISEMPTY(frms)) {
3373 /* If there is data for the next level, do not
3374 * consume a datagram.
3375 */
3376 prv_pkt = cur_pkt;
3377 }
3378 }
3379
3380 /* If we have to build a new datagram, set the current datagram as
3381 * prepared into <cbuf>.
3382 */
3383 if (!prv_pkt) {
3384 qc_txb_store(buf, dglen, first_pkt);
3385 first_pkt = NULL;
3386 dglen = 0;
3387 padding = 0;
3388 }
3389 else if (prv_pkt->type == QUIC_TLS_ENC_LEVEL_INITIAL &&
3390 (!qc_is_listener(qc) ||
3391 prv_pkt->flags & QUIC_FL_TX_PACKET_ACK_ELICITING)) {
3392 padding = 1;
3393 }
3394 }
3395
3396 out:
3397 ret = total;
3398 leave:
3399 TRACE_LEAVE(QUIC_EV_CONN_PHPKTS, qc);
3400 return ret;
3401}
3402
3403/* Send datagrams stored in <buf>.
3404 *
3405 * This function always returns 1 for success. Even if sendto() syscall failed,
3406 * buffer is drained and packets are considered as emitted. QUIC loss detection
3407 * mechanism is used as a back door way to retry sending.
3408 */
3409int qc_send_ppkts(struct buffer *buf, struct ssl_sock_ctx *ctx)
3410{
3411 struct quic_conn *qc;
3412 char skip_sendto = 0;
3413
3414 qc = ctx->qc;
3415 TRACE_ENTER(QUIC_EV_CONN_SPPKTS, qc);
3416 while (b_contig_data(buf, 0)) {
3417 unsigned char *pos;
3418 struct buffer tmpbuf = { };
3419 struct quic_tx_packet *first_pkt, *pkt, *next_pkt;
3420 uint16_t dglen;
3421 size_t headlen = sizeof dglen + sizeof first_pkt;
3422 unsigned int time_sent;
3423
3424 pos = (unsigned char *)b_head(buf);
3425 dglen = read_u16(pos);
3426 BUG_ON_HOT(!dglen); /* this should not happen */
3427
3428 pos += sizeof dglen;
3429 first_pkt = read_ptr(pos);
3430 pos += sizeof first_pkt;
3431 tmpbuf.area = (char *)pos;
3432 tmpbuf.size = tmpbuf.data = dglen;
3433
3434 TRACE_DATA("send dgram", QUIC_EV_CONN_SPPKTS, qc);
3435 /* If sendto is on error just skip the call to it for the rest
3436 * of the loop but continue to purge the buffer. Data will be
3437 * transmitted when QUIC packets are detected as lost on our
3438 * side.
3439 *
3440 * TODO use fd-monitoring to detect when send operation can be
3441 * retry. This should improve the bandwidth without relying on
3442 * retransmission timer. However, it requires a major rework on
3443 * quic-conn fd management.
3444 */
3445 if (!skip_sendto) {
3446 if (qc_snd_buf(qc, &tmpbuf, tmpbuf.data, 0)) {
3447 skip_sendto = 1;
3448 TRACE_ERROR("sendto error, simulate sending for the rest of data", QUIC_EV_CONN_SPPKTS, qc);
3449 }
3450 }
3451
3452 b_del(buf, dglen + headlen);
3453 qc->tx.bytes += tmpbuf.data;
3454 time_sent = now_ms;
3455
3456 for (pkt = first_pkt; pkt; pkt = next_pkt) {
3457 pkt->time_sent = time_sent;
3458 if (pkt->flags & QUIC_FL_TX_PACKET_ACK_ELICITING) {
3459 pkt->pktns->tx.time_of_last_eliciting = time_sent;
3460 qc->path->ifae_pkts++;
3461 if (qc->flags & QUIC_FL_CONN_IDLE_TIMER_RESTARTED_AFTER_READ)
3462 qc_idle_timer_rearm(qc, 0);
3463 }
3464 if (!(qc->flags & QUIC_FL_CONN_CLOSING) &&
3465 (pkt->flags & QUIC_FL_TX_PACKET_CC)) {
3466 qc->flags |= QUIC_FL_CONN_CLOSING;
3467 qc_notify_close(qc);
3468
3469 /* RFC 9000 10.2. Immediate Close:
3470 * The closing and draining connection states exist to ensure
3471 * that connections close cleanly and that delayed or reordered
3472 * packets are properly discarded. These states SHOULD persist
3473 * for at least three times the current PTO interval...
3474 *
3475 * Rearm the idle timeout only one time when entering closing
3476 * state.
3477 */
3478 qc_idle_timer_do_rearm(qc);
3479 if (qc->timer_task) {
3480 task_destroy(qc->timer_task);
3481 qc->timer_task = NULL;
3482 }
3483 }
3484 qc->path->in_flight += pkt->in_flight_len;
3485 pkt->pktns->tx.in_flight += pkt->in_flight_len;
3486 if (pkt->in_flight_len)
3487 qc_set_timer(qc);
3488 TRACE_DATA("sent pkt", QUIC_EV_CONN_SPPKTS, qc, pkt);
3489 next_pkt = pkt->next;
3490 quic_tx_packet_refinc(pkt);
3491 eb64_insert(&pkt->pktns->tx.pkts, &pkt->pn_node);
3492 }
3493 }
3494
3495 TRACE_LEAVE(QUIC_EV_CONN_SPPKTS, qc);
3496
3497 return 1;
3498}
3499
3500/* Copy into <buf> buffer a stateless reset token depending on the
3501 * <salt> salt input. This is the cluster secret which will be derived
3502 * as HKDF input secret to generate this token.
3503 * Return 1 if succeeded, 0 if not.
3504 */
3505static int quic_stateless_reset_token_cpy(struct quic_conn *qc,
3506 unsigned char *buf, size_t len,
3507 const unsigned char *salt, size_t saltlen)
3508{
3509 /* Input secret */
3510 const unsigned char *key = (const unsigned char *)global.cluster_secret;
3511 size_t keylen = strlen(global.cluster_secret);
3512 /* Info */
3513 const unsigned char label[] = "stateless token";
3514 size_t labellen = sizeof label - 1;
3515 int ret;
3516
3517 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
3518
3519 ret = quic_hkdf_extract_and_expand(EVP_sha256(), buf, len,
3520 key, keylen, salt, saltlen, label, labellen);
3521 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
3522 return ret;
3523}
3524
3525/* Initialize the stateless reset token attached to <cid> connection ID.
3526 * Returns 1 if succeeded, 0 if not.
3527 */
3528static int quic_stateless_reset_token_init(struct quic_conn *qc,
3529 struct quic_connection_id *quic_cid)
3530{
3531 int ret;
3532
3533 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
3534
3535 if (global.cluster_secret) {
3536 /* Output secret */
3537 unsigned char *token = quic_cid->stateless_reset_token;
3538 size_t tokenlen = sizeof quic_cid->stateless_reset_token;
3539 /* Salt */
3540 const unsigned char *cid = quic_cid->cid.data;
3541 size_t cidlen = quic_cid->cid.len;
3542
3543 ret = quic_stateless_reset_token_cpy(qc, token, tokenlen, cid, cidlen);
3544 }
3545 else {
3546 /* TODO: RAND_bytes() should be replaced */
3547 ret = RAND_bytes(quic_cid->stateless_reset_token,
3548 sizeof quic_cid->stateless_reset_token) == 1;
3549 }
3550
3551 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
3552 return ret;
3553}
3554
3555/* Allocate a new CID with <seq_num> as sequence number and attach it to <root>
3556 * ebtree.
3557 *
3558 * The CID is randomly generated in part with the result altered to be
3559 * associated with the current thread ID. This means this function must only
3560 * be called by the quic_conn thread.
3561 *
3562 * Returns the new CID if succeeded, NULL if not.
3563 */
3564static struct quic_connection_id *new_quic_cid(struct eb_root *root,
3565 struct quic_conn *qc,
3566 int seq_num)
3567{
3568 struct quic_connection_id *cid;
3569
3570 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
3571
3572 cid = pool_alloc(pool_head_quic_connection_id);
3573 if (!cid) {
3574 TRACE_ERROR("cid allocation failed", QUIC_EV_CONN_TXPKT, qc);
3575 goto err;
3576 }
3577
3578 cid->cid.len = QUIC_HAP_CID_LEN;
3579 /* TODO: RAND_bytes() should be replaced */
3580 if (RAND_bytes(cid->cid.data, cid->cid.len) != 1) {
3581 TRACE_ERROR("RAND_bytes() failed", QUIC_EV_CONN_TXPKT, qc);
3582 goto err;
3583 }
3584
3585 quic_pin_cid_to_tid(cid->cid.data, tid);
3586 if (quic_stateless_reset_token_init(qc, cid) != 1) {
3587 TRACE_ERROR("quic_stateless_reset_token_init() failed", QUIC_EV_CONN_TXPKT, qc);
3588 goto err;
3589 }
3590
3591 cid->qc = qc;
3592
3593 cid->seq_num.key = seq_num;
3594 cid->retire_prior_to = 0;
3595 /* insert the allocated CID in the quic_conn tree */
3596 eb64_insert(root, &cid->seq_num);
3597
3598 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
3599 return cid;
3600
3601 err:
3602 pool_free(pool_head_quic_connection_id, cid);
3603 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
3604 return NULL;
3605}
3606
3607/* Build all the frames which must be sent just after the handshake have succeeded.
3608 * This is essentially NEW_CONNECTION_ID frames. A QUIC server must also send
3609 * a HANDSHAKE_DONE frame.
3610 * Return 1 if succeeded, 0 if not.
3611 */
3612static int quic_build_post_handshake_frames(struct quic_conn *qc)
3613{
3614 int ret = 0, i, first, max;
3615 struct quic_enc_level *qel;
3616 struct quic_frame *frm, *frmbak;
3617 struct list frm_list = LIST_HEAD_INIT(frm_list);
3618 struct eb64_node *node;
3619
3620 TRACE_ENTER(QUIC_EV_CONN_IO_CB, qc);
3621
3622 qel = &qc->els[QUIC_TLS_ENC_LEVEL_APP];
3623 /* Only servers must send a HANDSHAKE_DONE frame. */
3624 if (qc_is_listener(qc)) {
Amaury Denoyelle40c24f12023-01-27 17:47:49 +01003625 frm = qc_frm_alloc(QUIC_FT_HANDSHAKE_DONE);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003626 if (!frm) {
3627 TRACE_ERROR("frame allocation error", QUIC_EV_CONN_IO_CB, qc);
3628 goto leave;
3629 }
3630
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003631 LIST_APPEND(&frm_list, &frm->list);
3632 }
3633
3634 /* Initialize <max> connection IDs minus one: there is
3635 * already one connection ID used for the current connection.
3636 */
3637 first = 1;
3638 max = qc->tx.params.active_connection_id_limit;
3639
3640 /* TODO: check limit */
3641 for (i = first; i < max; i++) {
3642 struct quic_connection_id *cid;
3643
Amaury Denoyelle40c24f12023-01-27 17:47:49 +01003644 frm = qc_frm_alloc(QUIC_FT_NEW_CONNECTION_ID);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003645 if (!frm) {
3646 TRACE_ERROR("frame allocation error", QUIC_EV_CONN_IO_CB, qc);
3647 goto err;
3648 }
3649
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003650 cid = new_quic_cid(&qc->cids, qc, i);
3651 if (!cid) {
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01003652 qc_frm_free(&frm);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003653 TRACE_ERROR("CID allocation error", QUIC_EV_CONN_IO_CB, qc);
3654 goto err;
3655 }
3656
3657 /* insert the allocated CID in the receiver datagram handler tree */
3658 ebmb_insert(&quic_dghdlrs[tid].cids, &cid->node, cid->cid.len);
3659
3660 quic_connection_id_to_frm_cpy(frm, cid);
3661 LIST_APPEND(&frm_list, &frm->list);
3662 }
3663
3664 LIST_SPLICE(&qel->pktns->tx.frms, &frm_list);
3665 qc->flags |= QUIC_FL_CONN_POST_HANDSHAKE_FRAMES_BUILT;
3666
3667 ret = 1;
3668 leave:
3669 TRACE_LEAVE(QUIC_EV_CONN_IO_CB, qc);
3670 return ret;
3671
3672 err:
3673 /* free the frames */
3674 list_for_each_entry_safe(frm, frmbak, &frm_list, list)
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01003675 qc_frm_free(&frm);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003676
3677 node = eb64_lookup_ge(&qc->cids, first);
3678 while (node) {
3679 struct quic_connection_id *cid;
3680
3681 cid = eb64_entry(node, struct quic_connection_id, seq_num);
3682 if (cid->seq_num.key >= max)
3683 break;
3684
3685 node = eb64_next(node);
3686 ebmb_delete(&cid->node);
3687 eb64_delete(&cid->seq_num);
3688 pool_free(pool_head_quic_connection_id, cid);
3689 }
3690 goto leave;
3691}
3692
3693/* Deallocate <l> list of ACK ranges. */
3694void quic_free_arngs(struct quic_conn *qc, struct quic_arngs *arngs)
3695{
3696 struct eb64_node *n;
3697 struct quic_arng_node *ar;
3698
3699 TRACE_ENTER(QUIC_EV_CONN_CLOSE, qc);
3700
3701 n = eb64_first(&arngs->root);
3702 while (n) {
3703 struct eb64_node *next;
3704
3705 ar = eb64_entry(n, struct quic_arng_node, first);
3706 next = eb64_next(n);
3707 eb64_delete(n);
3708 pool_free(pool_head_quic_arng, ar);
3709 n = next;
3710 }
3711
3712 TRACE_LEAVE(QUIC_EV_CONN_CLOSE, qc);
3713}
3714
3715/* Return the gap value between <p> and <q> ACK ranges where <q> follows <p> in
3716 * descending order.
3717 */
3718static inline size_t sack_gap(struct quic_arng_node *p,
3719 struct quic_arng_node *q)
3720{
3721 return p->first.key - q->last - 2;
3722}
3723
3724
3725/* Remove the last elements of <ack_ranges> list of ack range updating its
3726 * encoded size until it goes below <limit>.
3727 * Returns 1 if succeeded, 0 if not (no more element to remove).
3728 */
3729static int quic_rm_last_ack_ranges(struct quic_conn *qc,
3730 struct quic_arngs *arngs, size_t limit)
3731{
3732 int ret = 0;
3733 struct eb64_node *last, *prev;
3734
3735 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
3736
3737 last = eb64_last(&arngs->root);
3738 while (last && arngs->enc_sz > limit) {
3739 struct quic_arng_node *last_node, *prev_node;
3740
3741 prev = eb64_prev(last);
3742 if (!prev) {
3743 TRACE_DEVEL("<last> not found", QUIC_EV_CONN_TXPKT, qc);
3744 goto out;
3745 }
3746
3747 last_node = eb64_entry(last, struct quic_arng_node, first);
3748 prev_node = eb64_entry(prev, struct quic_arng_node, first);
3749 arngs->enc_sz -= quic_int_getsize(last_node->last - last_node->first.key);
3750 arngs->enc_sz -= quic_int_getsize(sack_gap(prev_node, last_node));
3751 arngs->enc_sz -= quic_decint_size_diff(arngs->sz);
3752 --arngs->sz;
3753 eb64_delete(last);
3754 pool_free(pool_head_quic_arng, last);
3755 last = prev;
3756 }
3757
3758 ret = 1;
3759 out:
3760 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
3761 return ret;
3762}
3763
3764/* Set the encoded size of <arngs> QUIC ack ranges. */
3765static void quic_arngs_set_enc_sz(struct quic_conn *qc, struct quic_arngs *arngs)
3766{
3767 struct eb64_node *node, *next;
3768 struct quic_arng_node *ar, *ar_next;
3769
3770 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
3771
3772 node = eb64_last(&arngs->root);
3773 if (!node)
3774 goto leave;
3775
3776 ar = eb64_entry(node, struct quic_arng_node, first);
3777 arngs->enc_sz = quic_int_getsize(ar->last) +
3778 quic_int_getsize(ar->last - ar->first.key) + quic_int_getsize(arngs->sz - 1);
3779
3780 while ((next = eb64_prev(node))) {
3781 ar_next = eb64_entry(next, struct quic_arng_node, first);
3782 arngs->enc_sz += quic_int_getsize(sack_gap(ar, ar_next)) +
3783 quic_int_getsize(ar_next->last - ar_next->first.key);
3784 node = next;
3785 ar = eb64_entry(node, struct quic_arng_node, first);
3786 }
3787
3788 leave:
3789 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
3790}
3791
3792/* Insert <ar> ack range into <argns> tree of ack ranges.
3793 * Returns the ack range node which has been inserted if succeeded, NULL if not.
3794 */
3795static inline
3796struct quic_arng_node *quic_insert_new_range(struct quic_conn *qc,
3797 struct quic_arngs *arngs,
3798 struct quic_arng *ar)
3799{
3800 struct quic_arng_node *new_ar;
3801
3802 TRACE_ENTER(QUIC_EV_CONN_RXPKT, qc);
3803
3804 new_ar = pool_alloc(pool_head_quic_arng);
3805 if (!new_ar) {
3806 TRACE_ERROR("ack range allocation failed", QUIC_EV_CONN_RXPKT, qc);
3807 goto leave;
3808 }
3809
3810 new_ar->first.key = ar->first;
3811 new_ar->last = ar->last;
3812 eb64_insert(&arngs->root, &new_ar->first);
3813 arngs->sz++;
3814
3815 leave:
3816 TRACE_LEAVE(QUIC_EV_CONN_RXPKT, qc);
3817 return new_ar;
3818}
3819
3820/* Update <arngs> tree of ACK ranges with <ar> as new ACK range value.
3821 * Note that this function computes the number of bytes required to encode
3822 * this tree of ACK ranges in descending order.
3823 *
3824 * Descending order
3825 * ------------->
3826 * range1 range2
3827 * ..........|--------|..............|--------|
3828 * ^ ^ ^ ^
3829 * | | | |
3830 * last1 first1 last2 first2
3831 * ..........+--------+--------------+--------+......
3832 * diff1 gap12 diff2
3833 *
3834 * To encode the previous list of ranges we must encode integers as follows in
3835 * descending order:
3836 * enc(last2),enc(diff2),enc(gap12),enc(diff1)
3837 * with diff1 = last1 - first1
3838 * diff2 = last2 - first2
3839 * gap12 = first1 - last2 - 2 (>= 0)
3840 *
3841
3842returns 0 on error
3843
3844 */
3845int quic_update_ack_ranges_list(struct quic_conn *qc,
3846 struct quic_arngs *arngs,
3847 struct quic_arng *ar)
3848{
3849 int ret = 0;
3850 struct eb64_node *le;
3851 struct quic_arng_node *new_node;
3852 struct eb64_node *new;
3853
3854 TRACE_ENTER(QUIC_EV_CONN_RXPKT, qc);
3855
3856 new = NULL;
3857 if (eb_is_empty(&arngs->root)) {
3858 new_node = quic_insert_new_range(qc, arngs, ar);
3859 if (new_node)
3860 ret = 1;
3861
3862 goto leave;
3863 }
3864
3865 le = eb64_lookup_le(&arngs->root, ar->first);
3866 if (!le) {
3867 new_node = quic_insert_new_range(qc, arngs, ar);
3868 if (!new_node)
3869 goto leave;
3870
3871 new = &new_node->first;
3872 }
3873 else {
3874 struct quic_arng_node *le_ar =
3875 eb64_entry(le, struct quic_arng_node, first);
3876
3877 /* Already existing range */
3878 if (le_ar->last >= ar->last) {
3879 ret = 1;
3880 }
3881 else if (le_ar->last + 1 >= ar->first) {
3882 le_ar->last = ar->last;
3883 new = le;
3884 new_node = le_ar;
3885 }
3886 else {
3887 new_node = quic_insert_new_range(qc, arngs, ar);
3888 if (!new_node)
3889 goto leave;
3890
3891 new = &new_node->first;
3892 }
3893 }
3894
3895 /* Verify that the new inserted node does not overlap the nodes
3896 * which follow it.
3897 */
3898 if (new) {
3899 struct eb64_node *next;
3900 struct quic_arng_node *next_node;
3901
3902 while ((next = eb64_next(new))) {
3903 next_node =
3904 eb64_entry(next, struct quic_arng_node, first);
3905 if (new_node->last + 1 < next_node->first.key)
3906 break;
3907
3908 if (next_node->last > new_node->last)
3909 new_node->last = next_node->last;
3910 eb64_delete(next);
3911 pool_free(pool_head_quic_arng, next_node);
3912 /* Decrement the size of these ranges. */
3913 arngs->sz--;
3914 }
3915 }
3916
3917 ret = 1;
3918 leave:
3919 quic_arngs_set_enc_sz(qc, arngs);
3920 TRACE_LEAVE(QUIC_EV_CONN_RXPKT, qc);
3921 return ret;
3922}
3923/* Remove the header protection of packets at <el> encryption level.
3924 * Always succeeds.
3925 */
3926static inline void qc_rm_hp_pkts(struct quic_conn *qc, struct quic_enc_level *el)
3927{
3928 struct quic_tls_ctx *tls_ctx;
3929 struct quic_rx_packet *pqpkt, *pkttmp;
3930 struct quic_enc_level *app_qel;
3931
3932 TRACE_ENTER(QUIC_EV_CONN_ELRMHP, qc);
3933 app_qel = &qc->els[QUIC_TLS_ENC_LEVEL_APP];
3934 /* A server must not process incoming 1-RTT packets before the handshake is complete. */
3935 if (el == app_qel && qc_is_listener(qc) && qc->state < QUIC_HS_ST_COMPLETE) {
3936 TRACE_DEVEL("hp not removed (handshake not completed)",
3937 QUIC_EV_CONN_ELRMHP, qc);
3938 goto out;
3939 }
3940 tls_ctx = &el->tls_ctx;
3941 list_for_each_entry_safe(pqpkt, pkttmp, &el->rx.pqpkts, list) {
3942 if (!qc_do_rm_hp(qc, pqpkt, tls_ctx, el->pktns->rx.largest_pn,
3943 pqpkt->data + pqpkt->pn_offset, pqpkt->data)) {
3944 TRACE_ERROR("hp removing error", QUIC_EV_CONN_ELRMHP, qc);
3945 }
3946 else {
3947 /* The AAD includes the packet number field */
3948 pqpkt->aad_len = pqpkt->pn_offset + pqpkt->pnl;
3949 /* Store the packet into the tree of packets to decrypt. */
3950 pqpkt->pn_node.key = pqpkt->pn;
3951 eb64_insert(&el->rx.pkts, &pqpkt->pn_node);
3952 quic_rx_packet_refinc(pqpkt);
3953 TRACE_DEVEL("hp removed", QUIC_EV_CONN_ELRMHP, qc, pqpkt);
3954 }
3955 LIST_DELETE(&pqpkt->list);
3956 quic_rx_packet_refdec(pqpkt);
3957 }
3958
3959 out:
3960 TRACE_LEAVE(QUIC_EV_CONN_ELRMHP, qc);
3961}
3962
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02003963/* Process all the CRYPTO frame at <el> encryption level. This is the
Ilya Shipitsin4a689da2022-10-29 09:34:32 +05003964 * responsibility of the called to ensure there exists a CRYPTO data
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02003965 * stream for this level.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003966 * Return 1 if succeeded, 0 if not.
3967 */
3968static inline int qc_treat_rx_crypto_frms(struct quic_conn *qc,
3969 struct quic_enc_level *el,
3970 struct ssl_sock_ctx *ctx)
3971{
3972 int ret = 0;
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02003973 struct ncbuf *ncbuf;
3974 struct quic_cstream *cstream = el->cstream;
3975 ncb_sz_t data;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003976
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02003977 TRACE_ENTER(QUIC_EV_CONN_PHPKTS, qc, el);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003978
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02003979 BUG_ON(!cstream);
3980 ncbuf = &cstream->rx.ncbuf;
3981 if (ncb_is_null(ncbuf))
3982 goto done;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003983
Amaury Denoyelle2f668f02022-11-18 15:24:08 +01003984 /* TODO not working if buffer is wrapping */
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02003985 while ((data = ncb_data(ncbuf, 0))) {
3986 const unsigned char *cdata = (const unsigned char *)ncb_head(ncbuf);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003987
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02003988 if (!qc_provide_cdata(el, ctx, cdata, data, NULL, NULL))
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003989 goto leave;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003990
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02003991 cstream->rx.offset += data;
Amaury Denoyelle2f668f02022-11-18 15:24:08 +01003992 TRACE_DEVEL("buffered crypto data were provided to TLS stack",
3993 QUIC_EV_CONN_PHPKTS, qc, el);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003994 }
3995
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02003996 done:
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02003997 ret = 1;
3998 leave:
Amaury Denoyelle2f668f02022-11-18 15:24:08 +01003999 if (!ncb_is_null(ncbuf) && ncb_is_empty(ncbuf)) {
4000 TRACE_DEVEL("freeing crypto buf", QUIC_EV_CONN_PHPKTS, qc, el);
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02004001 quic_free_ncbuf(ncbuf);
Amaury Denoyelle2f668f02022-11-18 15:24:08 +01004002 }
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02004003 TRACE_LEAVE(QUIC_EV_CONN_PHPKTS, qc);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004004 return ret;
4005}
4006
4007/* Process all the packets at <el> and <next_el> encryption level.
4008 * This is the caller responsibility to check that <cur_el> is different of <next_el>
4009 * as pointer value.
4010 * Return 1 if succeeded, 0 if not.
4011 */
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004012int qc_treat_rx_pkts(struct quic_conn *qc, struct quic_enc_level *cur_el,
4013 struct quic_enc_level *next_el, int force_ack)
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004014{
4015 int ret = 0;
4016 struct eb64_node *node;
4017 int64_t largest_pn = -1;
4018 unsigned int largest_pn_time_received = 0;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004019 struct quic_enc_level *qel = cur_el;
4020
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004021 TRACE_ENTER(QUIC_EV_CONN_RXPKT, qc);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004022 qel = cur_el;
4023 next_tel:
4024 if (!qel)
4025 goto out;
4026
4027 node = eb64_first(&qel->rx.pkts);
4028 while (node) {
4029 struct quic_rx_packet *pkt;
4030
4031 pkt = eb64_entry(node, struct quic_rx_packet, pn_node);
4032 TRACE_DATA("new packet", QUIC_EV_CONN_RXPKT,
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004033 qc, pkt, NULL, qc->xprt_ctx->ssl);
Amaury Denoyelle518c98f2022-11-24 17:12:25 +01004034 if (!qc_pkt_decrypt(qc, qel, pkt)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004035 /* Drop the packet */
4036 TRACE_ERROR("packet decryption failed -> dropped",
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004037 QUIC_EV_CONN_RXPKT, qc, pkt);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004038 }
4039 else {
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004040 if (!qc_parse_pkt_frms(qc, pkt, qel)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004041 /* Drop the packet */
4042 TRACE_ERROR("packet parsing failed -> dropped",
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004043 QUIC_EV_CONN_RXPKT, qc, pkt);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004044 HA_ATOMIC_INC(&qc->prx_counters->dropped_parsing);
4045 }
4046 else {
4047 struct quic_arng ar = { .first = pkt->pn, .last = pkt->pn };
4048
4049 if (pkt->flags & QUIC_FL_RX_PACKET_ACK_ELICITING || force_ack) {
4050 qel->pktns->flags |= QUIC_FL_PKTNS_ACK_REQUIRED;
4051 qel->pktns->rx.nb_aepkts_since_last_ack++;
4052 qc_idle_timer_rearm(qc, 1);
4053 }
4054 if (pkt->pn > largest_pn) {
4055 largest_pn = pkt->pn;
4056 largest_pn_time_received = pkt->time_received;
4057 }
4058 /* Update the list of ranges to acknowledge. */
4059 if (!quic_update_ack_ranges_list(qc, &qel->pktns->rx.arngs, &ar))
4060 TRACE_ERROR("Could not update ack range list",
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004061 QUIC_EV_CONN_RXPKT, qc);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004062 }
4063 }
4064 node = eb64_next(node);
4065 eb64_delete(&pkt->pn_node);
4066 quic_rx_packet_refdec(pkt);
4067 }
4068
4069 if (largest_pn != -1 && largest_pn > qel->pktns->rx.largest_pn) {
4070 /* Update the largest packet number. */
4071 qel->pktns->rx.largest_pn = largest_pn;
4072 /* Update the largest acknowledged packet timestamps */
4073 qel->pktns->rx.largest_time_received = largest_pn_time_received;
4074 qel->pktns->flags |= QUIC_FL_PKTNS_NEW_LARGEST_PN;
4075 }
4076
Frédéric Lécaille9f9263e2022-09-13 14:36:44 +02004077 if (qel->cstream && !qc_treat_rx_crypto_frms(qc, qel, qc->xprt_ctx)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004078 // trace already emitted by function above
4079 goto leave;
4080 }
4081
4082 if (qel == cur_el) {
4083 BUG_ON(qel == next_el);
4084 qel = next_el;
4085 largest_pn = -1;
4086 goto next_tel;
4087 }
4088
4089 out:
4090 ret = 1;
4091 leave:
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004092 TRACE_LEAVE(QUIC_EV_CONN_RXPKT, qc);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004093 return ret;
4094}
4095
4096/* Check if it's possible to remove header protection for packets related to
4097 * encryption level <qel>. If <qel> is NULL, assume it's false.
4098 *
4099 * Return true if the operation is possible else false.
4100 */
4101static int qc_qel_may_rm_hp(struct quic_conn *qc, struct quic_enc_level *qel)
4102{
4103 int ret = 0;
4104 enum quic_tls_enc_level tel;
4105
4106 TRACE_ENTER(QUIC_EV_CONN_TRMHP, qc);
4107
4108 if (!qel)
4109 goto cant_rm_hp;
4110
4111 tel = ssl_to_quic_enc_level(qel->level);
4112
4113 /* check if tls secrets are available */
4114 if (qel->tls_ctx.flags & QUIC_FL_TLS_SECRETS_DCD) {
4115 TRACE_DEVEL("Discarded keys", QUIC_EV_CONN_TRMHP, qc);
4116 goto cant_rm_hp;
4117 }
4118
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02004119 if (!quic_tls_has_rx_sec(qel)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004120 TRACE_DEVEL("non available secrets", QUIC_EV_CONN_TRMHP, qc);
4121 goto cant_rm_hp;
4122 }
4123
Frédéric Lécaille8417beb2023-02-01 10:31:35 +01004124 if (tel == QUIC_TLS_ENC_LEVEL_APP && qc->state < QUIC_HS_ST_COMPLETE) {
4125 TRACE_DEVEL("handshake not complete", QUIC_EV_CONN_TRMHP, qc);
4126 goto cant_rm_hp;
4127 }
4128
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004129 /* check if the connection layer is ready before using app level */
4130 if ((tel == QUIC_TLS_ENC_LEVEL_APP || tel == QUIC_TLS_ENC_LEVEL_EARLY_DATA) &&
4131 qc->mux_state == QC_MUX_NULL) {
4132 TRACE_DEVEL("connection layer not ready", QUIC_EV_CONN_TRMHP, qc);
4133 goto cant_rm_hp;
4134 }
4135
4136 ret = 1;
4137 cant_rm_hp:
4138 TRACE_LEAVE(QUIC_EV_CONN_TRMHP, qc);
4139 return ret;
4140}
4141
4142/* Try to send application frames from list <frms> on connection <qc>.
4143 *
4144 * Use qc_send_app_probing wrapper when probing with old data.
4145 *
4146 * Returns 1 on success. Some data might not have been sent due to congestion,
4147 * in this case they are left in <frms> input list. The caller may subscribe on
4148 * quic-conn to retry later.
4149 *
4150 * Returns 0 on critical error.
4151 * TODO review and classify more distinctly transient from definitive errors to
4152 * allow callers to properly handle it.
4153 */
4154static int qc_send_app_pkts(struct quic_conn *qc, struct list *frms)
4155{
4156 int status = 0;
4157 struct buffer *buf;
4158
4159 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
4160
4161 buf = qc_txb_alloc(qc);
4162 if (!buf) {
4163 TRACE_ERROR("buffer allocation failed", QUIC_EV_CONN_TXPKT, qc);
4164 goto leave;
4165 }
4166
4167 /* Prepare and send packets until we could not further prepare packets. */
4168 while (1) {
4169 int ret;
4170 /* Currently buf cannot be non-empty at this stage. Even if a
4171 * previous sendto() has failed it is emptied to simulate
4172 * packet emission and rely on QUIC lost detection to try to
4173 * emit it.
4174 */
4175 BUG_ON_HOT(b_data(buf));
4176 b_reset(buf);
4177
4178 ret = qc_prep_app_pkts(qc, buf, frms);
4179 if (ret == -1)
4180 goto err;
4181 else if (ret == 0)
4182 goto out;
4183
4184 if (!qc_send_ppkts(buf, qc->xprt_ctx))
4185 goto err;
4186 }
4187
4188 out:
4189 status = 1;
4190 qc_txb_release(qc);
4191 leave:
4192 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
4193 return status;
4194
4195 err:
4196 qc_txb_release(qc);
4197 goto leave;
4198}
4199
4200/* Try to send application frames from list <frms> on connection <qc>. Use this
4201 * function when probing is required.
4202 *
4203 * Returns the result from qc_send_app_pkts function.
4204 */
4205static forceinline int qc_send_app_probing(struct quic_conn *qc,
4206 struct list *frms)
4207{
4208 int ret;
4209
4210 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
4211
4212 TRACE_STATE("preparing old data (probing)", QUIC_EV_CONN_TXPKT, qc);
4213 qc->flags |= QUIC_FL_CONN_RETRANS_OLD_DATA;
4214 ret = qc_send_app_pkts(qc, frms);
4215 qc->flags &= ~QUIC_FL_CONN_RETRANS_OLD_DATA;
4216
4217 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
4218 return ret;
4219}
4220
4221/* Try to send application frames from list <frms> on connection <qc>. This
4222 * function is provided for MUX upper layer usage only.
4223 *
4224 * Returns the result from qc_send_app_pkts function.
4225 */
4226int qc_send_mux(struct quic_conn *qc, struct list *frms)
4227{
4228 int ret;
4229
4230 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
4231 BUG_ON(qc->mux_state != QC_MUX_READY); /* Only MUX can uses this function so it must be ready. */
4232
4233 TRACE_STATE("preparing data (from MUX)", QUIC_EV_CONN_TXPKT, qc);
4234 qc->flags |= QUIC_FL_CONN_TX_MUX_CONTEXT;
4235 ret = qc_send_app_pkts(qc, frms);
4236 qc->flags &= ~QUIC_FL_CONN_TX_MUX_CONTEXT;
4237
4238 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
4239 return ret;
4240}
4241
4242/* Sends handshake packets from up to two encryption levels <tel> and <next_te>
4243 * with <tel_frms> and <next_tel_frms> as frame list respectively for <qc>
4244 * QUIC connection. <old_data> is used as boolean to send data already sent but
4245 * not already acknowledged (in flight).
4246 * Returns 1 if succeeded, 0 if not.
4247 */
4248int qc_send_hdshk_pkts(struct quic_conn *qc, int old_data,
4249 enum quic_tls_enc_level tel, struct list *tel_frms,
4250 enum quic_tls_enc_level next_tel, struct list *next_tel_frms)
4251{
4252 int ret, status = 0;
4253 struct buffer *buf = qc_txb_alloc(qc);
4254
4255 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
4256
4257 if (!buf) {
4258 TRACE_ERROR("buffer allocation failed", QUIC_EV_CONN_TXPKT, qc);
4259 goto leave;
4260 }
4261
4262 /* Currently buf cannot be non-empty at this stage. Even if a previous
4263 * sendto() has failed it is emptied to simulate packet emission and
4264 * rely on QUIC lost detection to try to emit it.
4265 */
4266 BUG_ON_HOT(b_data(buf));
4267 b_reset(buf);
4268
4269 if (old_data) {
4270 TRACE_STATE("old data for probing asked", QUIC_EV_CONN_TXPKT, qc);
4271 qc->flags |= QUIC_FL_CONN_RETRANS_OLD_DATA;
4272 }
4273
4274 ret = qc_prep_pkts(qc, buf, tel, tel_frms, next_tel, next_tel_frms);
4275 if (ret == -1)
4276 goto out;
4277 else if (ret == 0)
4278 goto skip_send;
4279
4280 if (!qc_send_ppkts(buf, qc->xprt_ctx))
4281 goto out;
4282
4283 skip_send:
4284 status = 1;
4285 out:
4286 TRACE_STATE("no more need old data for probing", QUIC_EV_CONN_TXPKT, qc);
4287 qc->flags &= ~QUIC_FL_CONN_RETRANS_OLD_DATA;
4288 qc_txb_release(qc);
4289 leave:
4290 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
4291 return status;
4292}
4293
4294/* Retransmit up to two datagrams depending on packet number space */
4295static void qc_dgrams_retransmit(struct quic_conn *qc)
4296{
4297 struct quic_enc_level *iqel = &qc->els[QUIC_TLS_ENC_LEVEL_INITIAL];
4298 struct quic_enc_level *hqel = &qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE];
4299 struct quic_enc_level *aqel = &qc->els[QUIC_TLS_ENC_LEVEL_APP];
4300
4301 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
4302
4303 if (iqel->pktns->flags & QUIC_FL_PKTNS_PROBE_NEEDED) {
Frédéric Lécaille37ed4a32023-01-31 17:32:06 +01004304 int i;
4305
4306 for (i = 0; i < QUIC_MAX_NB_PTO_DGRAMS; i++) {
4307 struct list ifrms = LIST_HEAD_INIT(ifrms);
4308 struct list hfrms = LIST_HEAD_INIT(hfrms);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004309
Frédéric Lécaille37ed4a32023-01-31 17:32:06 +01004310 qc_prep_hdshk_fast_retrans(qc, &ifrms, &hfrms);
4311 TRACE_DEVEL("Avail. ack eliciting frames", QUIC_EV_CONN_FRMLIST, qc, &ifrms);
4312 TRACE_DEVEL("Avail. ack eliciting frames", QUIC_EV_CONN_FRMLIST, qc, &hfrms);
4313 if (!LIST_ISEMPTY(&ifrms)) {
4314 iqel->pktns->tx.pto_probe = 1;
4315 if (!LIST_ISEMPTY(&hfrms))
4316 hqel->pktns->tx.pto_probe = 1;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004317 qc_send_hdshk_pkts(qc, 1, QUIC_TLS_ENC_LEVEL_INITIAL, &ifrms,
4318 QUIC_TLS_ENC_LEVEL_HANDSHAKE, &hfrms);
4319 /* Put back unsent frames in their packet number spaces */
4320 LIST_SPLICE(&iqel->pktns->tx.frms, &ifrms);
4321 LIST_SPLICE(&hqel->pktns->tx.frms, &hfrms);
4322 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004323 }
4324 TRACE_STATE("no more need to probe Initial packet number space",
4325 QUIC_EV_CONN_TXPKT, qc);
4326 iqel->pktns->flags &= ~QUIC_FL_PKTNS_PROBE_NEEDED;
Frédéric Lécaille37ed4a32023-01-31 17:32:06 +01004327 hqel->pktns->flags &= ~QUIC_FL_PKTNS_PROBE_NEEDED;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004328 }
4329 else {
4330 int i;
4331
4332 if (hqel->pktns->flags & QUIC_FL_PKTNS_PROBE_NEEDED) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004333 hqel->pktns->tx.pto_probe = 0;
4334 for (i = 0; i < QUIC_MAX_NB_PTO_DGRAMS; i++) {
Frédéric Lécaille7b5d9b12022-11-28 17:21:45 +01004335 struct list frms1 = LIST_HEAD_INIT(frms1);
4336
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004337 qc_prep_fast_retrans(qc, hqel, &frms1, NULL);
4338 TRACE_DEVEL("Avail. ack eliciting frames", QUIC_EV_CONN_FRMLIST, qc, &frms1);
4339 if (!LIST_ISEMPTY(&frms1)) {
4340 hqel->pktns->tx.pto_probe = 1;
4341 qc_send_hdshk_pkts(qc, 1, QUIC_TLS_ENC_LEVEL_HANDSHAKE, &frms1,
4342 QUIC_TLS_ENC_LEVEL_NONE, NULL);
4343 /* Put back unsent frames into their packet number spaces */
4344 LIST_SPLICE(&hqel->pktns->tx.frms, &frms1);
4345 }
4346 }
4347 TRACE_STATE("no more need to probe Handshake packet number space",
4348 QUIC_EV_CONN_TXPKT, qc);
4349 hqel->pktns->flags &= ~QUIC_FL_PKTNS_PROBE_NEEDED;
4350 }
4351 else if (aqel->pktns->flags & QUIC_FL_PKTNS_PROBE_NEEDED) {
4352 struct list frms2 = LIST_HEAD_INIT(frms2);
4353 struct list frms1 = LIST_HEAD_INIT(frms1);
4354
4355 aqel->pktns->tx.pto_probe = 0;
4356 qc_prep_fast_retrans(qc, aqel, &frms1, &frms2);
4357 TRACE_PROTO("Avail. ack eliciting frames", QUIC_EV_CONN_FRMLIST, qc, &frms1);
4358 TRACE_PROTO("Avail. ack eliciting frames", QUIC_EV_CONN_FRMLIST, qc, &frms2);
4359 if (!LIST_ISEMPTY(&frms1)) {
4360 aqel->pktns->tx.pto_probe = 1;
4361 qc_send_app_probing(qc, &frms1);
4362 /* Put back unsent frames into their packet number spaces */
4363 LIST_SPLICE(&aqel->pktns->tx.frms, &frms1);
4364 }
4365 if (!LIST_ISEMPTY(&frms2)) {
4366 aqel->pktns->tx.pto_probe = 1;
4367 qc_send_app_probing(qc, &frms2);
4368 /* Put back unsent frames into their packet number spaces */
4369 LIST_SPLICE(&aqel->pktns->tx.frms, &frms2);
4370 }
4371 TRACE_STATE("no more need to probe 01RTT packet number space",
4372 QUIC_EV_CONN_TXPKT, qc);
4373 aqel->pktns->flags &= ~QUIC_FL_PKTNS_PROBE_NEEDED;
4374 }
4375 }
4376 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
4377}
4378
4379/* QUIC connection packet handler task (post handshake) */
4380struct task *quic_conn_app_io_cb(struct task *t, void *context, unsigned int state)
4381{
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004382 struct quic_conn *qc = context;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004383 struct quic_enc_level *qel;
4384
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004385 qel = &qc->els[QUIC_TLS_ENC_LEVEL_APP];
4386
4387 TRACE_ENTER(QUIC_EV_CONN_IO_CB, qc);
4388 TRACE_STATE("connection handshake state", QUIC_EV_CONN_IO_CB, qc, &qc->state);
4389
Amaury Denoyelle7c9fdd92022-11-16 11:01:02 +01004390 if (qc_test_fd(qc))
4391 qc_rcv_buf(qc);
4392
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004393 /* Retranmissions */
4394 if (qc->flags & QUIC_FL_CONN_RETRANS_NEEDED) {
4395 TRACE_STATE("retransmission needed", QUIC_EV_CONN_IO_CB, qc);
4396 qc->flags &= ~QUIC_FL_CONN_RETRANS_NEEDED;
4397 qc_dgrams_retransmit(qc);
4398 }
4399
4400 if (!LIST_ISEMPTY(&qel->rx.pqpkts) && qc_qel_may_rm_hp(qc, qel))
4401 qc_rm_hp_pkts(qc, qel);
4402
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004403 if (!qc_treat_rx_pkts(qc, qel, NULL, 0)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004404 TRACE_DEVEL("qc_treat_rx_pkts() failed", QUIC_EV_CONN_IO_CB, qc);
4405 goto out;
4406 }
4407
Frédéric Lécaille0aa79952023-02-03 16:15:08 +01004408 if (qc->flags & QUIC_FL_CONN_TO_KILL) {
4409 TRACE_DEVEL("connection to be killed", QUIC_EV_CONN_IO_CB, qc);
4410 goto out;
4411 }
4412
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004413 if ((qc->flags & QUIC_FL_CONN_DRAINING) &&
4414 !(qc->flags & QUIC_FL_CONN_IMMEDIATE_CLOSE)) {
4415 TRACE_STATE("draining connection (must not send packets)", QUIC_EV_CONN_IO_CB, qc);
4416 goto out;
4417 }
4418
4419 /* XXX TODO: how to limit the list frames to send */
4420 if (!qc_send_app_pkts(qc, &qel->pktns->tx.frms)) {
4421 TRACE_DEVEL("qc_send_app_pkts() failed", QUIC_EV_CONN_IO_CB, qc);
4422 goto out;
4423 }
4424
4425 out:
4426 TRACE_LEAVE(QUIC_EV_CONN_IO_CB, qc);
4427 return t;
4428}
4429
4430/* Returns a boolean if <qc> needs to emit frames for <qel> encryption level. */
4431static int qc_need_sending(struct quic_conn *qc, struct quic_enc_level *qel)
4432{
4433 return (qc->flags & QUIC_FL_CONN_IMMEDIATE_CLOSE) ||
4434 (qel->pktns->flags & QUIC_FL_PKTNS_ACK_REQUIRED) ||
4435 qel->pktns->tx.pto_probe ||
4436 !LIST_ISEMPTY(&qel->pktns->tx.frms);
4437}
4438
4439/* QUIC connection packet handler task. */
4440struct task *quic_conn_io_cb(struct task *t, void *context, unsigned int state)
4441{
4442 int ret, ssl_err;
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004443 struct quic_conn *qc = context;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004444 enum quic_tls_enc_level tel, next_tel;
4445 struct quic_enc_level *qel, *next_qel;
Frédéric Lécaille4aa7d812022-09-16 10:15:58 +02004446 /* Early-data encryption level */
4447 struct quic_enc_level *eqel;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004448 struct buffer *buf = NULL;
4449 int st, force_ack, zero_rtt;
4450
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004451 TRACE_ENTER(QUIC_EV_CONN_IO_CB, qc);
Frédéric Lécaille4aa7d812022-09-16 10:15:58 +02004452 eqel = &qc->els[QUIC_TLS_ENC_LEVEL_EARLY_DATA];
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004453 st = qc->state;
4454 TRACE_PROTO("connection state", QUIC_EV_CONN_IO_CB, qc, &st);
4455
4456 /* Retranmissions */
4457 if (qc->flags & QUIC_FL_CONN_RETRANS_NEEDED) {
4458 TRACE_DEVEL("retransmission needed", QUIC_EV_CONN_PHPKTS, qc);
4459 qc->flags &= ~QUIC_FL_CONN_RETRANS_NEEDED;
4460 qc_dgrams_retransmit(qc);
4461 }
4462
4463 if (qc->flags & QUIC_FL_CONN_IO_CB_WAKEUP) {
4464 qc->flags &= ~QUIC_FL_CONN_IO_CB_WAKEUP;
4465 TRACE_DEVEL("needs to wakeup the timer task after the anti-amplicaiton limit was reached",
4466 QUIC_EV_CONN_IO_CB, qc);
4467 /* The I/O handler has been woken up by the dgram parser (qc_lstnr_pkt_rcv())
4468 * after the anti-amplification was reached.
4469 *
4470 * TODO: this part should be removed. This was there because the
4471 * datagram parser was not executed by only one thread.
4472 */
4473 qc_set_timer(qc);
Amaury Denoyelle5ac6b3b2022-12-14 18:04:06 +01004474 if (qc->timer_task && tick_isset(qc->timer) && tick_is_lt(qc->timer, now_ms))
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004475 task_wakeup(qc->timer_task, TASK_WOKEN_MSG);
4476 }
4477 ssl_err = SSL_ERROR_NONE;
4478 zero_rtt = st < QUIC_HS_ST_COMPLETE &&
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02004479 quic_tls_has_rx_sec(eqel) &&
Frédéric Lécaille4aa7d812022-09-16 10:15:58 +02004480 (!LIST_ISEMPTY(&eqel->rx.pqpkts) || qc_el_rx_pkts(eqel));
Amaury Denoyelle7c9fdd92022-11-16 11:01:02 +01004481
4482 if (qc_test_fd(qc))
4483 qc_rcv_buf(qc);
4484
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004485 if (st >= QUIC_HS_ST_COMPLETE &&
4486 qc_el_rx_pkts(&qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE])) {
4487 TRACE_DEVEL("remaining Handshake packets", QUIC_EV_CONN_PHPKTS, qc);
4488 /* There may be remaining Handshake packets to treat and acknowledge. */
4489 tel = QUIC_TLS_ENC_LEVEL_HANDSHAKE;
4490 next_tel = QUIC_TLS_ENC_LEVEL_APP;
4491 }
Frédéric Lécaille4aa7d812022-09-16 10:15:58 +02004492 else if (!quic_get_tls_enc_levels(&tel, &next_tel, qc, st, zero_rtt))
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004493 goto out;
4494
4495 qel = &qc->els[tel];
4496 next_qel = next_tel == QUIC_TLS_ENC_LEVEL_NONE ? NULL : &qc->els[next_tel];
4497
4498 next_level:
4499 /* Treat packets waiting for header packet protection decryption */
4500 if (!LIST_ISEMPTY(&qel->rx.pqpkts) && qc_qel_may_rm_hp(qc, qel))
4501 qc_rm_hp_pkts(qc, qel);
4502
4503 force_ack = qel == &qc->els[QUIC_TLS_ENC_LEVEL_INITIAL] ||
4504 qel == &qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE];
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004505 if (!qc_treat_rx_pkts(qc, qel, next_qel, force_ack))
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004506 goto out;
4507
Frédéric Lécaille0aa79952023-02-03 16:15:08 +01004508 if (qc->flags & QUIC_FL_CONN_TO_KILL) {
4509 TRACE_DEVEL("connection to be killed", QUIC_EV_CONN_PHPKTS, qc);
4510 goto out;
4511 }
4512
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004513 if ((qc->flags & QUIC_FL_CONN_DRAINING) &&
4514 !(qc->flags & QUIC_FL_CONN_IMMEDIATE_CLOSE))
4515 goto out;
4516
Frédéric Lécaille4aa7d812022-09-16 10:15:58 +02004517 zero_rtt = st < QUIC_HS_ST_COMPLETE &&
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02004518 quic_tls_has_rx_sec(eqel) &&
Frédéric Lécaille4aa7d812022-09-16 10:15:58 +02004519 (!LIST_ISEMPTY(&eqel->rx.pqpkts) || qc_el_rx_pkts(eqel));
4520 if (next_qel && next_qel == eqel && zero_rtt) {
4521 TRACE_DEVEL("select 0RTT as next encryption level",
4522 QUIC_EV_CONN_PHPKTS, qc);
4523 qel = next_qel;
4524 next_qel = NULL;
4525 goto next_level;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004526 }
4527
4528 st = qc->state;
4529 if (st >= QUIC_HS_ST_COMPLETE) {
4530 if (!(qc->flags & QUIC_FL_CONN_POST_HANDSHAKE_FRAMES_BUILT) &&
4531 !quic_build_post_handshake_frames(qc))
4532 goto out;
4533
4534 if (!(qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE].tls_ctx.flags &
4535 QUIC_FL_TLS_SECRETS_DCD)) {
4536 /* Discard the Handshake keys. */
4537 quic_tls_discard_keys(&qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE]);
4538 TRACE_PROTO("discarding Handshake pktns", QUIC_EV_CONN_PHPKTS, qc);
4539 quic_pktns_discard(qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE].pktns, qc);
4540 qc_set_timer(qc);
4541 qc_el_rx_pkts_del(&qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE]);
4542 qc_release_pktns_frms(qc, qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE].pktns);
4543 }
4544
4545 if (qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE].pktns->flags & QUIC_FL_PKTNS_ACK_REQUIRED) {
4546 /* There may be remaining handshake to build (acks) */
4547 st = QUIC_HS_ST_SERVER_HANDSHAKE;
4548 }
4549 }
4550
4551 /* A listener does not send any O-RTT packet. O-RTT packet number space must not
4552 * be considered.
4553 */
Frédéric Lécaille4aa7d812022-09-16 10:15:58 +02004554 if (!quic_get_tls_enc_levels(&tel, &next_tel, qc, st, 0))
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004555 goto out;
4556
4557 if (!qc_need_sending(qc, qel) &&
4558 (!next_qel || !qc_need_sending(qc, next_qel))) {
4559 goto skip_send;
4560 }
4561
4562 buf = qc_txb_alloc(qc);
4563 if (!buf)
4564 goto out;
4565
4566 /* Currently buf cannot be non-empty at this stage. Even if a previous
4567 * sendto() has failed it is emptied to simulate packet emission and
4568 * rely on QUIC lost detection to try to emit it.
4569 */
4570 BUG_ON_HOT(b_data(buf));
4571 b_reset(buf);
4572
4573 ret = qc_prep_pkts(qc, buf, tel, &qc->els[tel].pktns->tx.frms,
4574 next_tel, &qc->els[next_tel].pktns->tx.frms);
4575 if (ret == -1)
4576 goto out;
4577 else if (ret == 0)
4578 goto skip_send;
4579
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004580 if (!qc_send_ppkts(buf, qc->xprt_ctx))
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004581 goto out;
4582
4583 skip_send:
4584 /* Check if there is something to do for the next level.
4585 */
4586 if (next_qel && next_qel != qel &&
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02004587 quic_tls_has_rx_sec(next_qel) &&
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004588 (!LIST_ISEMPTY(&next_qel->rx.pqpkts) || qc_el_rx_pkts(next_qel))) {
4589 qel = next_qel;
4590 next_qel = NULL;
4591 goto next_level;
4592 }
4593
4594 out:
4595 qc_txb_release(qc);
4596 TRACE_LEAVE(QUIC_EV_CONN_IO_CB, qc, &st, &ssl_err);
4597 return t;
4598}
4599
Frédéric Lécaille7e3f7c42022-09-09 18:05:45 +02004600/* Release the memory allocated for <cs> CRYPTO stream */
4601void quic_cstream_free(struct quic_cstream *cs)
4602{
4603 if (!cs) {
4604 /* This is the case for ORTT encryption level */
4605 return;
4606 }
4607
Amaury Denoyellebc174b22022-11-17 10:12:52 +01004608 quic_free_ncbuf(&cs->rx.ncbuf);
4609
Frédéric Lécaille7e3f7c42022-09-09 18:05:45 +02004610 qc_stream_desc_release(cs->desc);
4611 pool_free(pool_head_quic_cstream, cs);
4612}
4613
4614/* Allocate a new QUIC stream for <qc>.
4615 * Return it if succeeded, NULL if not.
4616 */
4617struct quic_cstream *quic_cstream_new(struct quic_conn *qc)
4618{
4619 struct quic_cstream *cs, *ret_cs = NULL;
4620
4621 TRACE_ENTER(QUIC_EV_CONN_LPKT, qc);
4622 cs = pool_alloc(pool_head_quic_cstream);
4623 if (!cs) {
4624 TRACE_ERROR("crypto stream allocation failed", QUIC_EV_CONN_INIT, qc);
4625 goto leave;
4626 }
4627
4628 cs->rx.offset = 0;
4629 cs->rx.ncbuf = NCBUF_NULL;
4630 cs->rx.offset = 0;
4631
4632 cs->tx.offset = 0;
4633 cs->tx.sent_offset = 0;
4634 cs->tx.buf = BUF_NULL;
4635 cs->desc = qc_stream_desc_new((uint64_t)-1, -1, cs, qc);
4636 if (!cs->desc) {
4637 TRACE_ERROR("crypto stream allocation failed", QUIC_EV_CONN_INIT, qc);
4638 goto err;
4639 }
4640
4641 ret_cs = cs;
4642 leave:
4643 TRACE_LEAVE(QUIC_EV_CONN_LPKT, qc);
4644 return ret_cs;
4645
4646 err:
4647 pool_free(pool_head_quic_cstream, cs);
4648 goto leave;
4649}
4650
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004651/* Uninitialize <qel> QUIC encryption level. Never fails. */
4652static void quic_conn_enc_level_uninit(struct quic_conn *qc, struct quic_enc_level *qel)
4653{
4654 int i;
4655
4656 TRACE_ENTER(QUIC_EV_CONN_CLOSE, qc);
4657
4658 for (i = 0; i < qel->tx.crypto.nb_buf; i++) {
4659 if (qel->tx.crypto.bufs[i]) {
4660 pool_free(pool_head_quic_crypto_buf, qel->tx.crypto.bufs[i]);
4661 qel->tx.crypto.bufs[i] = NULL;
4662 }
4663 }
4664 ha_free(&qel->tx.crypto.bufs);
Frédéric Lécaille7e3f7c42022-09-09 18:05:45 +02004665 quic_cstream_free(qel->cstream);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004666
4667 TRACE_LEAVE(QUIC_EV_CONN_CLOSE, qc);
4668}
4669
4670/* Initialize QUIC TLS encryption level with <level<> as level for <qc> QUIC
4671 * connection allocating everything needed.
Amaury Denoyelledbf6ad42022-12-12 11:22:42 +01004672 *
4673 * Returns 1 if succeeded, 0 if not. On error the caller is responsible to use
4674 * quic_conn_enc_level_uninit() to cleanup partially allocated content.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004675 */
4676static int quic_conn_enc_level_init(struct quic_conn *qc,
4677 enum quic_tls_enc_level level)
4678{
4679 int ret = 0;
4680 struct quic_enc_level *qel;
4681
4682 TRACE_ENTER(QUIC_EV_CONN_CLOSE, qc);
4683
4684 qel = &qc->els[level];
4685 qel->level = quic_to_ssl_enc_level(level);
4686 qel->tls_ctx.rx.aead = qel->tls_ctx.tx.aead = NULL;
4687 qel->tls_ctx.rx.md = qel->tls_ctx.tx.md = NULL;
4688 qel->tls_ctx.rx.hp = qel->tls_ctx.tx.hp = NULL;
4689 qel->tls_ctx.flags = 0;
4690
4691 qel->rx.pkts = EB_ROOT;
4692 LIST_INIT(&qel->rx.pqpkts);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004693
4694 /* Allocate only one buffer. */
4695 /* TODO: use a pool */
4696 qel->tx.crypto.bufs = malloc(sizeof *qel->tx.crypto.bufs);
4697 if (!qel->tx.crypto.bufs)
Amaury Denoyelledbf6ad42022-12-12 11:22:42 +01004698 goto leave;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004699
4700 qel->tx.crypto.bufs[0] = pool_alloc(pool_head_quic_crypto_buf);
4701 if (!qel->tx.crypto.bufs[0])
Amaury Denoyelledbf6ad42022-12-12 11:22:42 +01004702 goto leave;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004703
4704 qel->tx.crypto.bufs[0]->sz = 0;
4705 qel->tx.crypto.nb_buf = 1;
4706
4707 qel->tx.crypto.sz = 0;
4708 qel->tx.crypto.offset = 0;
Frédéric Lécaille7e3f7c42022-09-09 18:05:45 +02004709 /* No CRYPTO data for early data TLS encryption level */
4710 if (level == QUIC_TLS_ENC_LEVEL_EARLY_DATA)
4711 qel->cstream = NULL;
4712 else {
4713 qel->cstream = quic_cstream_new(qc);
4714 if (!qel->cstream)
Amaury Denoyelledbf6ad42022-12-12 11:22:42 +01004715 goto leave;
Frédéric Lécaille7e3f7c42022-09-09 18:05:45 +02004716 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004717
4718 ret = 1;
4719 leave:
4720 TRACE_LEAVE(QUIC_EV_CONN_CLOSE, qc);
4721 return ret;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004722}
4723
4724/* Callback called upon loss detection and PTO timer expirations. */
4725struct task *qc_process_timer(struct task *task, void *ctx, unsigned int state)
4726{
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004727 struct quic_conn *qc = ctx;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004728 struct quic_pktns *pktns;
4729
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004730 TRACE_ENTER(QUIC_EV_CONN_PTIMER, qc,
4731 NULL, NULL, &qc->path->ifae_pkts);
4732 task->expire = TICK_ETERNITY;
4733 pktns = quic_loss_pktns(qc);
4734 if (tick_isset(pktns->tx.loss_time)) {
4735 struct list lost_pkts = LIST_HEAD_INIT(lost_pkts);
4736
4737 qc_packet_loss_lookup(pktns, qc, &lost_pkts);
4738 if (!LIST_ISEMPTY(&lost_pkts))
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004739 tasklet_wakeup(qc->wait_event.tasklet);
Amaury Denoyellee4abb1f2023-01-27 17:54:15 +01004740 if (qc_release_lost_pkts(qc, pktns, &lost_pkts, now_ms))
4741 qc_set_timer(qc);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004742 goto out;
4743 }
4744
4745 if (qc->path->in_flight) {
Frédéric Lécailleb75eecc2023-01-26 15:18:17 +01004746 pktns = quic_pto_pktns(qc, qc->state >= QUIC_HS_ST_CONFIRMED, NULL);
Amaury Denoyellebbb1c682022-09-28 15:15:51 +02004747 if (qc->subs && qc->subs->events & SUB_RETRY_SEND) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004748 pktns->tx.pto_probe = QUIC_MAX_NB_PTO_DGRAMS;
Amaury Denoyellebbb1c682022-09-28 15:15:51 +02004749 tasklet_wakeup(qc->subs->tasklet);
4750 qc->subs->events &= ~SUB_RETRY_SEND;
4751 if (!qc->subs->events)
4752 qc->subs = NULL;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004753 }
4754 else {
4755 qc->flags |= QUIC_FL_CONN_RETRANS_NEEDED;
4756 pktns->flags |= QUIC_FL_PKTNS_PROBE_NEEDED;
4757 if (pktns == &qc->pktns[QUIC_TLS_PKTNS_INITIAL]) {
4758 TRACE_STATE("needs to probe Initial packet number space", QUIC_EV_CONN_TXPKT, qc);
4759 if (qc->pktns[QUIC_TLS_PKTNS_HANDSHAKE].tx.in_flight) {
4760 qc->pktns[QUIC_TLS_PKTNS_HANDSHAKE].flags |= QUIC_FL_PKTNS_PROBE_NEEDED;
4761 TRACE_STATE("needs to probe Handshake packet number space", QUIC_EV_CONN_TXPKT, qc);
4762 }
4763 }
4764 else if (pktns == &qc->pktns[QUIC_TLS_PKTNS_HANDSHAKE]) {
4765 TRACE_STATE("needs to probe Handshake packet number space", QUIC_EV_CONN_TXPKT, qc);
Frédéric Lécaille37ed4a32023-01-31 17:32:06 +01004766 if (qc->pktns[QUIC_TLS_PKTNS_INITIAL].tx.in_flight) {
4767 qc->pktns[QUIC_TLS_PKTNS_INITIAL].flags |= QUIC_FL_PKTNS_PROBE_NEEDED;
4768 TRACE_STATE("needs to probe Initial packet number space", QUIC_EV_CONN_TXPKT, qc);
4769 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004770 }
4771 else if (pktns == &qc->pktns[QUIC_TLS_PKTNS_01RTT]) {
4772 TRACE_STATE("needs to probe 01RTT packet number space", QUIC_EV_CONN_TXPKT, qc);
4773 }
4774 }
4775 }
4776 else if (!qc_is_listener(qc) && qc->state <= QUIC_HS_ST_COMPLETE) {
4777 struct quic_enc_level *iel = &qc->els[QUIC_TLS_ENC_LEVEL_INITIAL];
4778 struct quic_enc_level *hel = &qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE];
4779
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02004780 if (quic_tls_has_tx_sec(hel))
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004781 hel->pktns->tx.pto_probe = 1;
Frédéric Lécaillee1a49cf2022-09-16 16:24:47 +02004782 if (quic_tls_has_tx_sec(iel))
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004783 iel->pktns->tx.pto_probe = 1;
4784 }
4785
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02004786 tasklet_wakeup(qc->wait_event.tasklet);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004787 qc->path->loss.pto_count++;
4788
4789 out:
4790 TRACE_LEAVE(QUIC_EV_CONN_PTIMER, qc, pktns);
4791
4792 return task;
4793}
4794
4795/* Parse the Retry token from buffer <token> with <end> a pointer to
4796 * one byte past the end of this buffer. This will extract the ODCID
4797 * which will be stored into <odcid>
4798 *
4799 * Returns 0 on success else non-zero.
4800 */
4801static int parse_retry_token(struct quic_conn *qc,
4802 const unsigned char *token, const unsigned char *end,
4803 struct quic_cid *odcid)
4804{
4805 int ret = 0;
4806 uint64_t odcid_len;
4807 uint32_t timestamp;
4808
4809 TRACE_ENTER(QUIC_EV_CONN_LPKT, qc);
4810
4811 if (!quic_dec_int(&odcid_len, &token, end)) {
4812 TRACE_ERROR("quic_dec_int() error", QUIC_EV_CONN_LPKT, qc);
4813 goto leave;
4814 }
4815
4816 /* RFC 9000 7.2. Negotiating Connection IDs:
4817 * When an Initial packet is sent by a client that has not previously
4818 * received an Initial or Retry packet from the server, the client
4819 * populates the Destination Connection ID field with an unpredictable
4820 * value. This Destination Connection ID MUST be at least 8 bytes in length.
4821 */
4822 if (odcid_len < QUIC_ODCID_MINLEN || odcid_len > QUIC_CID_MAXLEN) {
4823 TRACE_ERROR("wrong ODCID length", QUIC_EV_CONN_LPKT, qc);
4824 goto leave;
4825 }
4826
4827 if (end - token < odcid_len + sizeof timestamp) {
4828 TRACE_ERROR("too long ODCID length", QUIC_EV_CONN_LPKT, qc);
4829 goto leave;
4830 }
4831
4832 timestamp = ntohl(read_u32(token + odcid_len));
4833 if (timestamp + MS_TO_TICKS(QUIC_RETRY_DURATION_MS) <= now_ms) {
4834 TRACE_ERROR("token has expired", QUIC_EV_CONN_LPKT, qc);
4835 goto leave;
4836 }
4837
4838 ret = 1;
4839 memcpy(odcid->data, token, odcid_len);
4840 odcid->len = odcid_len;
4841 leave:
4842 TRACE_LEAVE(QUIC_EV_CONN_LPKT, qc);
4843 return !ret;
4844}
4845
4846/* Allocate a new QUIC connection with <version> as QUIC version. <ipv4>
4847 * boolean is set to 1 for IPv4 connection, 0 for IPv6. <server> is set to 1
4848 * for QUIC servers (or haproxy listeners).
4849 * <dcid> is the destination connection ID, <scid> is the source connection ID,
4850 * <token> the token found to be used for this connection with <token_len> as
Amaury Denoyelle97ecc7a2022-09-23 17:15:58 +02004851 * length. Endpoints addresses are specified via <local_addr> and <peer_addr>.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004852 * Returns the connection if succeeded, NULL if not.
4853 */
4854static struct quic_conn *qc_new_conn(const struct quic_version *qv, int ipv4,
4855 struct quic_cid *dcid, struct quic_cid *scid,
4856 const struct quic_cid *token_odcid,
Amaury Denoyelle97ecc7a2022-09-23 17:15:58 +02004857 struct sockaddr_storage *local_addr,
4858 struct sockaddr_storage *peer_addr,
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004859 int server, int token, void *owner)
4860{
4861 int i;
4862 struct quic_conn *qc;
4863 /* Initial CID. */
4864 struct quic_connection_id *icid;
4865 char *buf_area = NULL;
4866 struct listener *l = NULL;
4867 struct quic_cc_algo *cc_algo = NULL;
4868 struct quic_tls_ctx *ictx;
4869 TRACE_ENTER(QUIC_EV_CONN_INIT);
Amaury Denoyelledbf6ad42022-12-12 11:22:42 +01004870 /* TODO replace pool_zalloc by pool_alloc(). This requires special care
4871 * to properly initialized internal quic_conn members to safely use
4872 * quic_conn_release() on alloc failure.
4873 */
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004874 qc = pool_zalloc(pool_head_quic_conn);
4875 if (!qc) {
4876 TRACE_ERROR("Could not allocate a new connection", QUIC_EV_CONN_INIT);
4877 goto err;
4878 }
4879
Amaury Denoyelledbf6ad42022-12-12 11:22:42 +01004880 /* Initialize in priority qc members required for a safe dealloc. */
4881
4882 /* required to use MTLIST_IN_LIST */
4883 MT_LIST_INIT(&qc->accept_list);
4884
4885 LIST_INIT(&qc->rx.pkt_list);
4886
Amaury Denoyelle42448332022-12-12 11:24:05 +01004887 qc_init_fd(qc);
4888
Amaury Denoyelledbf6ad42022-12-12 11:22:42 +01004889 /* Now proceeds to allocation of qc members. */
4890
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004891 buf_area = pool_alloc(pool_head_quic_conn_rxbuf);
4892 if (!buf_area) {
4893 TRACE_ERROR("Could not allocate a new RX buffer", QUIC_EV_CONN_INIT, qc);
4894 goto err;
4895 }
4896
4897 qc->cids = EB_ROOT;
4898 /* QUIC Server (or listener). */
4899 if (server) {
4900 struct proxy *prx;
4901
4902 l = owner;
4903 prx = l->bind_conf->frontend;
4904 cc_algo = l->bind_conf->quic_cc_algo;
4905
4906 qc->prx_counters = EXTRA_COUNTERS_GET(prx->extra_counters_fe,
4907 &quic_stats_module);
4908 qc->flags |= QUIC_FL_CONN_LISTENER;
4909 qc->state = QUIC_HS_ST_SERVER_INITIAL;
4910 /* Copy the initial DCID with the address. */
4911 qc->odcid.len = dcid->len;
4912 qc->odcid.addrlen = dcid->addrlen;
4913 memcpy(qc->odcid.data, dcid->data, dcid->len + dcid->addrlen);
4914
4915 /* copy the packet SCID to reuse it as DCID for sending */
4916 if (scid->len)
4917 memcpy(qc->dcid.data, scid->data, scid->len);
4918 qc->dcid.len = scid->len;
4919 qc->tx.buf = BUF_NULL;
4920 qc->li = l;
4921 }
4922 /* QUIC Client (outgoing connection to servers) */
4923 else {
4924 qc->state = QUIC_HS_ST_CLIENT_INITIAL;
4925 if (dcid->len)
4926 memcpy(qc->dcid.data, dcid->data, dcid->len);
4927 qc->dcid.len = dcid->len;
4928 }
4929 qc->mux_state = QC_MUX_NULL;
4930 qc->err = quic_err_transport(QC_ERR_NO_ERROR);
4931
4932 icid = new_quic_cid(&qc->cids, qc, 0);
4933 if (!icid) {
4934 TRACE_ERROR("Could not allocate a new connection ID", QUIC_EV_CONN_INIT, qc);
4935 goto err;
4936 }
4937
Amaury Denoyelle40909df2022-10-24 17:08:43 +02004938 if ((global.tune.options & GTUNE_QUIC_SOCK_PER_CONN) &&
4939 is_addr(local_addr)) {
4940 TRACE_USER("Allocate a socket for QUIC connection", QUIC_EV_CONN_INIT, qc);
4941 qc_alloc_fd(qc, local_addr, peer_addr);
4942 }
Amaury Denoyelle40909df2022-10-24 17:08:43 +02004943
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004944 /* insert the allocated CID in the receiver datagram handler tree */
4945 if (server)
4946 ebmb_insert(&quic_dghdlrs[tid].cids, &icid->node, icid->cid.len);
4947
4948 /* Select our SCID which is the first CID with 0 as sequence number. */
4949 qc->scid = icid->cid;
4950
4951 /* Packet number spaces initialization. */
4952 for (i = 0; i < QUIC_TLS_PKTNS_MAX; i++)
4953 quic_pktns_init(&qc->pktns[i]);
4954 /* QUIC encryption level context initialization. */
4955 for (i = 0; i < QUIC_TLS_ENC_LEVEL_MAX; i++) {
4956 if (!quic_conn_enc_level_init(qc, i)) {
4957 TRACE_ERROR("Could not initialize an encryption level", QUIC_EV_CONN_INIT, qc);
4958 goto err;
4959 }
4960 /* Initialize the packet number space. */
4961 qc->els[i].pktns = &qc->pktns[quic_tls_pktns(i)];
4962 }
4963
4964 qc->original_version = qv;
4965 qc->tps_tls_ext = (qc->original_version->num & 0xff000000) == 0xff000000 ?
4966 TLS_EXTENSION_QUIC_TRANSPORT_PARAMETERS_DRAFT:
4967 TLS_EXTENSION_QUIC_TRANSPORT_PARAMETERS;
4968 /* TX part. */
4969 LIST_INIT(&qc->tx.frms_to_send);
4970 qc->tx.nb_buf = QUIC_CONN_TX_BUFS_NB;
4971 qc->tx.wbuf = qc->tx.rbuf = 0;
4972 qc->tx.bytes = 0;
4973 qc->tx.buf = BUF_NULL;
4974 /* RX part. */
4975 qc->rx.bytes = 0;
4976 qc->rx.buf = b_make(buf_area, QUIC_CONN_RX_BUFSZ, 0, 0);
4977 for (i = 0; i < QCS_MAX_TYPES; i++)
4978 qc->rx.strms[i].nb_streams = 0;
4979
4980 qc->nb_pkt_for_cc = 1;
4981 qc->nb_pkt_since_cc = 0;
4982
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004983 if (!quic_tls_ku_init(qc)) {
4984 TRACE_ERROR("Key update initialization failed", QUIC_EV_CONN_INIT, qc);
4985 goto err;
4986 }
4987
4988 /* XXX TO DO: Only one path at this time. */
4989 qc->path = &qc->paths[0];
4990 quic_path_init(qc->path, ipv4, cc_algo ? cc_algo : default_quic_cc_algo, qc);
4991
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004992 qc->streams_by_id = EB_ROOT_UNIQUE;
4993 qc->stream_buf_count = 0;
Amaury Denoyelle97ecc7a2022-09-23 17:15:58 +02004994 memcpy(&qc->local_addr, local_addr, sizeof(qc->local_addr));
4995 memcpy(&qc->peer_addr, peer_addr, sizeof qc->peer_addr);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02004996
4997 if (server && !qc_lstnr_params_init(qc, &l->bind_conf->quic_params,
4998 icid->stateless_reset_token,
4999 dcid->data, dcid->len,
5000 qc->scid.data, qc->scid.len, token_odcid))
5001 goto err;
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02005002
5003 qc->wait_event.tasklet = tasklet_new();
5004 if (!qc->wait_event.tasklet) {
5005 TRACE_ERROR("tasklet_new() failed", QUIC_EV_CONN_TXPKT);
5006 goto err;
5007 }
5008 qc->wait_event.tasklet->process = quic_conn_io_cb;
5009 qc->wait_event.tasklet->context = qc;
5010 qc->wait_event.events = 0;
5011 /* Set tasklet tid based on the SCID selected by us for this
5012 * connection. The upper layer will also be binded on the same thread.
5013 */
Willy Tarreaueed78262022-12-21 09:09:19 +01005014 qc->tid = quic_get_cid_tid(qc->scid.data, &l->rx);
Willy Tarreauf5a0c8a2022-10-13 16:14:11 +02005015 qc->wait_event.tasklet->tid = qc->tid;
Amaury Denoyellebbb1c682022-09-28 15:15:51 +02005016 qc->subs = NULL;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005017
5018 if (qc_conn_alloc_ssl_ctx(qc) ||
5019 !quic_conn_init_timer(qc) ||
5020 !quic_conn_init_idle_timer_task(qc))
5021 goto err;
5022
5023 ictx = &qc->els[QUIC_TLS_ENC_LEVEL_INITIAL].tls_ctx;
5024 if (!qc_new_isecs(qc, ictx,qc->original_version, dcid->data, dcid->len, 1))
5025 goto err;
5026
5027 TRACE_LEAVE(QUIC_EV_CONN_INIT, qc);
5028
5029 return qc;
5030
5031 err:
5032 pool_free(pool_head_quic_conn_rxbuf, buf_area);
Amaury Denoyelledbf6ad42022-12-12 11:22:42 +01005033 if (qc) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005034 qc->rx.buf.area = NULL;
Amaury Denoyelledbf6ad42022-12-12 11:22:42 +01005035 quic_conn_release(qc);
5036 }
5037 TRACE_LEAVE(QUIC_EV_CONN_INIT);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005038 return NULL;
5039}
5040
5041/* Release the quic_conn <qc>. The connection is removed from the CIDs tree.
5042 * The connection tasklet is killed.
5043 *
5044 * This function must only be called by the thread responsible of the quic_conn
5045 * tasklet.
5046 */
5047void quic_conn_release(struct quic_conn *qc)
5048{
5049 int i;
5050 struct ssl_sock_ctx *conn_ctx;
5051 struct eb64_node *node;
5052 struct quic_tls_ctx *app_tls_ctx;
5053 struct quic_rx_packet *pkt, *pktback;
5054
5055 TRACE_ENTER(QUIC_EV_CONN_CLOSE, qc);
5056
5057 /* We must not free the quic-conn if the MUX is still allocated. */
5058 BUG_ON(qc->mux_state == QC_MUX_READY);
5059
Amaury Denoyelle40909df2022-10-24 17:08:43 +02005060 /* Close quic-conn socket fd. */
Amaury Denoyelled3083c92022-12-01 16:20:06 +01005061 qc_release_fd(qc, 0);
Amaury Denoyelle40909df2022-10-24 17:08:43 +02005062
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005063 /* in the unlikely (but possible) case the connection was just added to
5064 * the accept_list we must delete it from there.
5065 */
5066 MT_LIST_DELETE(&qc->accept_list);
5067
5068 /* free remaining stream descriptors */
5069 node = eb64_first(&qc->streams_by_id);
5070 while (node) {
5071 struct qc_stream_desc *stream;
5072
5073 stream = eb64_entry(node, struct qc_stream_desc, by_id);
5074 node = eb64_next(node);
5075
5076 /* all streams attached to the quic-conn are released, so
5077 * qc_stream_desc_free will liberate the stream instance.
5078 */
5079 BUG_ON(!stream->release);
5080 qc_stream_desc_free(stream, 1);
5081 }
5082
5083 /* Purge Rx packet list. */
5084 list_for_each_entry_safe(pkt, pktback, &qc->rx.pkt_list, qc_rx_pkt_list) {
5085 LIST_DELETE(&pkt->qc_rx_pkt_list);
5086 pool_free(pool_head_quic_rx_packet, pkt);
5087 }
5088
5089 if (qc->idle_timer_task) {
5090 task_destroy(qc->idle_timer_task);
5091 qc->idle_timer_task = NULL;
5092 }
5093
5094 if (qc->timer_task) {
5095 task_destroy(qc->timer_task);
5096 qc->timer_task = NULL;
5097 }
5098
Amaury Denoyelledbf6ad42022-12-12 11:22:42 +01005099 if (qc->wait_event.tasklet)
5100 tasklet_free(qc->wait_event.tasklet);
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02005101
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005102 /* remove the connection from receiver cids trees */
5103 ebmb_delete(&qc->odcid_node);
5104 ebmb_delete(&qc->scid_node);
5105 free_quic_conn_cids(qc);
5106
5107 conn_ctx = qc->xprt_ctx;
5108 if (conn_ctx) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005109 SSL_free(conn_ctx->ssl);
5110 pool_free(pool_head_quic_conn_ctx, conn_ctx);
5111 }
5112
5113 quic_tls_ku_free(qc);
5114 for (i = 0; i < QUIC_TLS_ENC_LEVEL_MAX; i++) {
5115 quic_tls_ctx_secs_free(&qc->els[i].tls_ctx);
5116 quic_conn_enc_level_uninit(qc, &qc->els[i]);
5117 }
5118 quic_tls_ctx_secs_free(&qc->negotiated_ictx);
5119
5120 app_tls_ctx = &qc->els[QUIC_TLS_ENC_LEVEL_APP].tls_ctx;
5121 pool_free(pool_head_quic_tls_secret, app_tls_ctx->rx.secret);
5122 pool_free(pool_head_quic_tls_secret, app_tls_ctx->tx.secret);
5123
5124 for (i = 0; i < QUIC_TLS_PKTNS_MAX; i++) {
5125 quic_pktns_tx_pkts_release(&qc->pktns[i], qc);
5126 quic_free_arngs(qc, &qc->pktns[i].rx.arngs);
5127 }
5128
5129 pool_free(pool_head_quic_conn_rxbuf, qc->rx.buf.area);
5130 pool_free(pool_head_quic_conn, qc);
5131 TRACE_PROTO("QUIC conn. freed", QUIC_EV_CONN_FREED, qc);
5132
5133 TRACE_LEAVE(QUIC_EV_CONN_CLOSE, qc);
5134}
5135
5136/* Initialize the timer task of <qc> QUIC connection.
5137 * Returns 1 if succeeded, 0 if not.
5138 */
5139static int quic_conn_init_timer(struct quic_conn *qc)
5140{
5141 int ret = 0;
5142 /* Attach this task to the same thread ID used for the connection */
5143 TRACE_ENTER(QUIC_EV_CONN_NEW, qc);
5144
5145 qc->timer_task = task_new_on(qc->tid);
5146 if (!qc->timer_task) {
5147 TRACE_ERROR("timer task allocation failed", QUIC_EV_CONN_NEW, qc);
5148 goto leave;
5149 }
5150
5151 qc->timer = TICK_ETERNITY;
5152 qc->timer_task->process = qc_process_timer;
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02005153 qc->timer_task->context = qc;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005154
5155 ret = 1;
5156 leave:
5157 TRACE_LEAVE(QUIC_EV_CONN_NEW, qc);
5158 return ret;
5159}
5160
5161/* Rearm the idle timer for <qc> QUIC connection. */
5162static void qc_idle_timer_do_rearm(struct quic_conn *qc)
5163{
5164 unsigned int expire;
5165
5166 expire = QUIC_MAX(3 * quic_pto(qc), qc->max_idle_timeout);
5167 qc->idle_timer_task->expire = tick_add(now_ms, MS_TO_TICKS(expire));
5168}
5169
5170/* Rearm the idle timer for <qc> QUIC connection depending on <read> boolean
5171 * which is set to 1 when receiving a packet , and 0 when sending packet
5172 */
5173static void qc_idle_timer_rearm(struct quic_conn *qc, int read)
5174{
5175 TRACE_ENTER(QUIC_EV_CONN_IDLE_TIMER, qc);
5176
5177 if (read) {
5178 qc->flags |= QUIC_FL_CONN_IDLE_TIMER_RESTARTED_AFTER_READ;
5179 }
5180 else {
5181 qc->flags &= ~QUIC_FL_CONN_IDLE_TIMER_RESTARTED_AFTER_READ;
5182 }
5183 qc_idle_timer_do_rearm(qc);
5184
5185 TRACE_LEAVE(QUIC_EV_CONN_IDLE_TIMER, qc);
5186}
5187
5188/* The task handling the idle timeout */
5189struct task *qc_idle_timer_task(struct task *t, void *ctx, unsigned int state)
5190{
5191 struct quic_conn *qc = ctx;
5192 struct quic_counters *prx_counters = qc->prx_counters;
5193 unsigned int qc_flags = qc->flags;
5194
5195 TRACE_ENTER(QUIC_EV_CONN_IDLE_TIMER, qc);
5196
5197 /* Notify the MUX before settings QUIC_FL_CONN_EXP_TIMER or the MUX
5198 * might free the quic-conn too early via quic_close().
5199 */
5200 qc_notify_close(qc);
5201
5202 /* If the MUX is still alive, keep the quic-conn. The MUX is
5203 * responsible to call quic_close to release it.
5204 */
5205 qc->flags |= QUIC_FL_CONN_EXP_TIMER;
5206 if (qc->mux_state != QC_MUX_READY)
5207 quic_conn_release(qc);
5208
5209 /* TODO if the quic-conn cannot be freed because of the MUX, we may at
5210 * least clean some parts of it such as the tasklet.
5211 */
5212
5213 if (!(qc_flags & QUIC_FL_CONN_HALF_OPEN_CNT_DECREMENTED)) {
5214 qc_flags |= QUIC_FL_CONN_HALF_OPEN_CNT_DECREMENTED;
5215 TRACE_DEVEL("dec half open counter", QUIC_EV_CONN_SSLALERT, qc);
5216 HA_ATOMIC_DEC(&prx_counters->half_open_conn);
5217 }
5218
5219 TRACE_LEAVE(QUIC_EV_CONN_IDLE_TIMER, qc);
5220 return NULL;
5221}
5222
5223/* Initialize the idle timeout task for <qc>.
5224 * Returns 1 if succeeded, 0 if not.
5225 */
5226static int quic_conn_init_idle_timer_task(struct quic_conn *qc)
5227{
5228 int ret = 0;
5229
5230 TRACE_ENTER(QUIC_EV_CONN_NEW, qc);
5231
5232 qc->idle_timer_task = task_new_here();
5233 if (!qc->idle_timer_task) {
5234 TRACE_ERROR("Idle timer task allocation failed", QUIC_EV_CONN_NEW, qc);
5235 goto leave;
5236 }
5237
5238 qc->idle_timer_task->process = qc_idle_timer_task;
5239 qc->idle_timer_task->context = qc;
5240 qc_idle_timer_rearm(qc, 1);
5241 task_queue(qc->idle_timer_task);
5242
5243 ret = 1;
5244 leave:
5245 TRACE_LEAVE(QUIC_EV_CONN_NEW, qc);
5246 return ret;
5247}
5248
5249/* Parse into <pkt> a long header located at <*buf> buffer, <end> begin a pointer to the end
5250 * past one byte of this buffer.
5251 */
5252static inline int quic_packet_read_long_header(unsigned char **buf, const unsigned char *end,
5253 struct quic_rx_packet *pkt)
5254{
5255 int ret = 0;
5256 unsigned char dcid_len, scid_len;
5257
5258 TRACE_ENTER(QUIC_EV_CONN_RXPKT);
5259
5260 if (end == *buf) {
5261 TRACE_ERROR("buffer data consumed", QUIC_EV_CONN_RXPKT);
5262 goto leave;
5263 }
5264
5265 /* Destination Connection ID Length */
5266 dcid_len = *(*buf)++;
5267 /* We want to be sure we can read <dcid_len> bytes and one more for <scid_len> value */
5268 if (dcid_len > QUIC_CID_MAXLEN || end - *buf < dcid_len + 1) {
5269 TRACE_ERROR("too long DCID", QUIC_EV_CONN_RXPKT);
5270 goto leave;
5271 }
5272
5273 if (dcid_len) {
5274 /* Check that the length of this received DCID matches the CID lengths
5275 * of our implementation for non Initials packets only.
5276 */
5277 if (pkt->type != QUIC_PACKET_TYPE_INITIAL &&
5278 pkt->type != QUIC_PACKET_TYPE_0RTT &&
5279 dcid_len != QUIC_HAP_CID_LEN) {
5280 TRACE_ERROR("wrong DCID length", QUIC_EV_CONN_RXPKT);
5281 goto leave;
5282 }
5283
5284 memcpy(pkt->dcid.data, *buf, dcid_len);
5285 }
5286
5287 pkt->dcid.len = dcid_len;
5288 *buf += dcid_len;
5289
5290 /* Source Connection ID Length */
5291 scid_len = *(*buf)++;
5292 if (scid_len > QUIC_CID_MAXLEN || end - *buf < scid_len) {
5293 TRACE_ERROR("too long SCID", QUIC_EV_CONN_RXPKT);
5294 goto leave;
5295 }
5296
5297 if (scid_len)
5298 memcpy(pkt->scid.data, *buf, scid_len);
5299 pkt->scid.len = scid_len;
5300 *buf += scid_len;
5301
5302 ret = 1;
5303 leave:
5304 TRACE_LEAVE(QUIC_EV_CONN_RXPKT);
5305 return ret;
5306}
5307
5308/* Insert <pkt> RX packet in its <qel> RX packets tree */
5309static void qc_pkt_insert(struct quic_conn *qc,
5310 struct quic_rx_packet *pkt, struct quic_enc_level *qel)
5311{
5312 TRACE_ENTER(QUIC_EV_CONN_RXPKT, qc);
5313
5314 pkt->pn_node.key = pkt->pn;
5315 quic_rx_packet_refinc(pkt);
5316 eb64_insert(&qel->rx.pkts, &pkt->pn_node);
5317
5318 TRACE_LEAVE(QUIC_EV_CONN_RXPKT, qc);
5319}
5320
Amaury Denoyelle845169d2022-10-17 18:05:26 +02005321/* Try to remove the header protection of <pkt> QUIC packet with <beg> the
5322 * address of the packet first byte, using the keys from encryption level <el>.
5323 *
5324 * If header protection has been successfully removed, packet data are copied
5325 * into <qc> Rx buffer. If <el> secrets are not yet available, the copy is also
5326 * proceeded, and the packet is inserted into <qc> protected packets tree. In
5327 * both cases, packet can now be considered handled by the <qc> connection.
5328 *
5329 * If header protection cannot be removed due to <el> secrets already
5330 * discarded, no operation is conducted.
5331 *
5332 * Returns 1 on success : packet data is now handled by the connection. On
5333 * error 0 is returned : packet should be dropped by the caller.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005334 */
5335static inline int qc_try_rm_hp(struct quic_conn *qc,
5336 struct quic_rx_packet *pkt,
Amaury Denoyelle845169d2022-10-17 18:05:26 +02005337 unsigned char *beg,
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005338 struct quic_enc_level **el)
5339{
5340 int ret = 0;
5341 unsigned char *pn = NULL; /* Packet number field */
5342 enum quic_tls_enc_level tel;
5343 struct quic_enc_level *qel;
5344 /* Only for traces. */
5345 struct quic_rx_packet *qpkt_trace;
5346
5347 qpkt_trace = NULL;
5348 TRACE_ENTER(QUIC_EV_CONN_TRMHP, qc);
Amaury Denoyelle845169d2022-10-17 18:05:26 +02005349 BUG_ON(!pkt->pn_offset);
5350
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005351 /* The packet number is here. This is also the start minus
5352 * QUIC_PACKET_PN_MAXLEN of the sample used to add/remove the header
5353 * protection.
5354 */
Amaury Denoyelle845169d2022-10-17 18:05:26 +02005355 pn = beg + pkt->pn_offset;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005356
5357 tel = quic_packet_type_enc_level(pkt->type);
5358 qel = &qc->els[tel];
5359
5360 if (qc_qel_may_rm_hp(qc, qel)) {
5361 /* Note that the following function enables us to unprotect the packet
5362 * number and its length subsequently used to decrypt the entire
5363 * packets.
5364 */
5365 if (!qc_do_rm_hp(qc, pkt, &qel->tls_ctx,
5366 qel->pktns->rx.largest_pn, pn, beg)) {
5367 TRACE_PROTO("hp error", QUIC_EV_CONN_TRMHP, qc);
5368 goto out;
5369 }
5370
Amaury Denoyelle845169d2022-10-17 18:05:26 +02005371 /* The AAD includes the packet number field. */
5372 pkt->aad_len = pkt->pn_offset + pkt->pnl;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005373 if (pkt->len - pkt->aad_len < QUIC_TLS_TAG_LEN) {
5374 TRACE_PROTO("Too short packet", QUIC_EV_CONN_TRMHP, qc);
5375 goto out;
5376 }
5377
5378 qpkt_trace = pkt;
5379 }
5380 else {
5381 if (qel->tls_ctx.flags & QUIC_FL_TLS_SECRETS_DCD) {
5382 /* If the packet number space has been discarded, this packet
5383 * will be not parsed.
5384 */
5385 TRACE_PROTO("Discarded pktns", QUIC_EV_CONN_TRMHP, qc, pkt);
5386 goto out;
5387 }
5388
5389 TRACE_PROTO("hp not removed", QUIC_EV_CONN_TRMHP, qc, pkt);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005390 LIST_APPEND(&qel->rx.pqpkts, &pkt->list);
5391 quic_rx_packet_refinc(pkt);
5392 }
5393
5394 *el = qel;
5395 /* No reference counter incrementation here!!! */
5396 LIST_APPEND(&qc->rx.pkt_list, &pkt->qc_rx_pkt_list);
5397 memcpy(b_tail(&qc->rx.buf), beg, pkt->len);
5398 pkt->data = (unsigned char *)b_tail(&qc->rx.buf);
5399 b_add(&qc->rx.buf, pkt->len);
5400
5401 ret = 1;
5402 out:
5403 TRACE_LEAVE(QUIC_EV_CONN_TRMHP, qc, qpkt_trace);
5404 return ret;
5405}
5406
5407/* Parse the header form from <byte0> first byte of <pkt> packet to set its type.
5408 * Also set <*long_header> to 1 if this form is long, 0 if not and the version
5409 * of this packet into <*version>.
5410 */
5411static inline int qc_parse_hd_form(struct quic_rx_packet *pkt,
5412 unsigned char **buf, const unsigned char *end,
5413 int *long_header, uint32_t *version)
5414{
5415 int ret = 0;
5416 const unsigned char byte0 = **buf;
5417
5418 TRACE_ENTER(QUIC_EV_CONN_RXPKT);
5419
5420 (*buf)++;
5421 if (byte0 & QUIC_PACKET_LONG_HEADER_BIT) {
5422 unsigned char type =
5423 (byte0 >> QUIC_PACKET_TYPE_SHIFT) & QUIC_PACKET_TYPE_BITMASK;
5424
5425 *long_header = 1;
5426 /* Version */
5427 if (!quic_read_uint32(version, (const unsigned char **)buf, end)) {
5428 TRACE_ERROR("could not read the packet version", QUIC_EV_CONN_RXPKT);
5429 goto out;
5430 }
5431
Frédéric Lécaille21c4c9b2023-01-13 16:37:02 +01005432 if (*version != QUIC_PROTOCOL_VERSION_2) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005433 pkt->type = type;
5434 }
5435 else {
5436 switch (type) {
5437 case 0:
5438 pkt->type = QUIC_PACKET_TYPE_RETRY;
5439 break;
5440 case 1:
5441 pkt->type = QUIC_PACKET_TYPE_INITIAL;
5442 break;
5443 case 2:
5444 pkt->type = QUIC_PACKET_TYPE_0RTT;
5445 break;
5446 case 3:
5447 pkt->type = QUIC_PACKET_TYPE_HANDSHAKE;
5448 break;
5449 }
5450 }
5451 }
5452 else {
5453 pkt->type = QUIC_PACKET_TYPE_SHORT;
5454 *long_header = 0;
5455 }
5456
5457 ret = 1;
5458 out:
5459 TRACE_LEAVE(QUIC_EV_CONN_RXPKT);
5460 return ret;
5461}
5462
5463/* Return the QUIC version (quic_version struct) with <version> as version number
5464 * if supported or NULL if not.
5465 */
5466static inline const struct quic_version *qc_supported_version(uint32_t version)
5467{
5468 int i;
5469
5470 for (i = 0; i < quic_versions_nb; i++)
5471 if (quic_versions[i].num == version)
5472 return &quic_versions[i];
5473
5474 return NULL;
5475}
5476
5477/*
5478 * Send a Version Negotiation packet on response to <pkt> on socket <fd> to
5479 * address <addr>.
5480 * Implementation of RFC9000 6. Version Negotiation
5481 *
5482 * TODO implement a rate-limiting sending of Version Negotiation packets
5483 *
5484 * Returns 0 on success else non-zero
5485 */
5486static int send_version_negotiation(int fd, struct sockaddr_storage *addr,
5487 struct quic_rx_packet *pkt)
5488{
5489 char buf[256];
5490 int ret = 0, i = 0, j;
5491 uint32_t version;
5492 const socklen_t addrlen = get_addr_len(addr);
5493
5494 TRACE_ENTER(QUIC_EV_CONN_TXPKT);
5495 /*
5496 * header form
5497 * long header, fixed bit to 0 for Version Negotiation
5498 */
5499 /* TODO: RAND_bytes() should be replaced? */
5500 if (RAND_bytes((unsigned char *)buf, 1) != 1) {
5501 TRACE_ERROR("RAND_bytes() error", QUIC_EV_CONN_TXPKT);
5502 goto out;
5503 }
5504
5505 buf[i++] |= '\x80';
5506 /* null version for Version Negotiation */
5507 buf[i++] = '\x00';
5508 buf[i++] = '\x00';
5509 buf[i++] = '\x00';
5510 buf[i++] = '\x00';
5511
5512 /* source connection id */
5513 buf[i++] = pkt->scid.len;
5514 memcpy(&buf[i], pkt->scid.data, pkt->scid.len);
5515 i += pkt->scid.len;
5516
5517 /* destination connection id */
5518 buf[i++] = pkt->dcid.len;
5519 memcpy(&buf[i], pkt->dcid.data, pkt->dcid.len);
5520 i += pkt->dcid.len;
5521
5522 /* supported version */
5523 for (j = 0; j < quic_versions_nb; j++) {
5524 version = htonl(quic_versions[j].num);
5525 memcpy(&buf[i], &version, sizeof(version));
5526 i += sizeof(version);
5527 }
5528
5529 if (sendto(fd, buf, i, 0, (struct sockaddr *)addr, addrlen) < 0)
5530 goto out;
5531
5532 ret = 1;
5533 out:
5534 TRACE_LEAVE(QUIC_EV_CONN_TXPKT);
5535 return !ret;
5536}
5537
5538/* Send a stateless reset packet depending on <pkt> RX packet information
5539 * from <fd> UDP socket to <dst>
5540 * Return 1 if succeeded, 0 if not.
5541 */
5542static int send_stateless_reset(struct listener *l, struct sockaddr_storage *dstaddr,
5543 struct quic_rx_packet *rxpkt)
5544{
5545 int ret = 0, pktlen, rndlen;
5546 unsigned char pkt[64];
5547 const socklen_t addrlen = get_addr_len(dstaddr);
5548 struct proxy *prx;
5549 struct quic_counters *prx_counters;
5550
5551 TRACE_ENTER(QUIC_EV_STATELESS_RST);
5552
5553 prx = l->bind_conf->frontend;
5554 prx_counters = EXTRA_COUNTERS_GET(prx->extra_counters_fe, &quic_stats_module);
5555 /* 10.3 Stateless Reset (https://www.rfc-editor.org/rfc/rfc9000.html#section-10.3)
5556 * The resulting minimum size of 21 bytes does not guarantee that a Stateless
5557 * Reset is difficult to distinguish from other packets if the recipient requires
5558 * the use of a connection ID. To achieve that end, the endpoint SHOULD ensure
5559 * that all packets it sends are at least 22 bytes longer than the minimum
5560 * connection ID length that it requests the peer to include in its packets,
5561 * adding PADDING frames as necessary. This ensures that any Stateless Reset
5562 * sent by the peer is indistinguishable from a valid packet sent to the endpoint.
5563 * An endpoint that sends a Stateless Reset in response to a packet that is
5564 * 43 bytes or shorter SHOULD send a Stateless Reset that is one byte shorter
5565 * than the packet it responds to.
5566 */
5567
5568 /* Note that we build at most a 42 bytes QUIC packet to mimic a short packet */
5569 pktlen = rxpkt->len <= 43 ? rxpkt->len - 1 : 0;
5570 pktlen = QUIC_MAX(QUIC_STATELESS_RESET_PACKET_MINLEN, pktlen);
5571 rndlen = pktlen - QUIC_STATELESS_RESET_TOKEN_LEN;
5572
5573 /* Put a header of random bytes */
5574 /* TODO: RAND_bytes() should be replaced */
5575 if (RAND_bytes(pkt, rndlen) != 1) {
5576 TRACE_ERROR("RAND_bytes() failed", QUIC_EV_STATELESS_RST);
5577 goto leave;
5578 }
5579
5580 /* Clear the most significant bit, and set the second one */
5581 *pkt = (*pkt & ~0x80) | 0x40;
5582 if (!quic_stateless_reset_token_cpy(NULL, pkt + rndlen, QUIC_STATELESS_RESET_TOKEN_LEN,
5583 rxpkt->dcid.data, rxpkt->dcid.len))
5584 goto leave;
5585
5586 if (sendto(l->rx.fd, pkt, pktlen, 0, (struct sockaddr *)dstaddr, addrlen) < 0)
5587 goto leave;
5588
5589 ret = 1;
5590 HA_ATOMIC_INC(&prx_counters->stateless_reset_sent);
5591 TRACE_PROTO("stateless reset sent", QUIC_EV_STATELESS_RST, NULL, &rxpkt->dcid);
5592 leave:
5593 TRACE_LEAVE(QUIC_EV_STATELESS_RST);
5594 return ret;
5595}
5596
5597/* QUIC server only function.
5598 * Add AAD to <add> buffer from <cid> connection ID and <addr> socket address.
5599 * This is the responsibility of the caller to check <aad> size is big enough
5600 * to contain these data.
5601 * Return the number of bytes copied to <aad>.
5602 */
5603static int quic_generate_retry_token_aad(unsigned char *aad,
5604 uint32_t version,
5605 const struct quic_cid *cid,
5606 const struct sockaddr_storage *addr)
5607{
5608 unsigned char *p;
5609
5610 p = aad;
5611 memcpy(p, &version, sizeof version);
5612 p += sizeof version;
5613 p += quic_saddr_cpy(p, addr);
5614 memcpy(p, cid->data, cid->len);
5615 p += cid->len;
5616
5617 return p - aad;
5618}
5619
5620/* QUIC server only function.
5621 * Generate the token to be used in Retry packets. The token is written to
Ilya Shipitsin4a689da2022-10-29 09:34:32 +05005622 * <buf> with <len> as length. <odcid> is the original destination connection
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005623 * ID and <dcid> is our side destination connection ID (or client source
5624 * connection ID).
5625 * Returns the length of the encoded token or 0 on error.
5626 */
5627static int quic_generate_retry_token(unsigned char *buf, size_t len,
5628 const uint32_t version,
5629 const struct quic_cid *odcid,
5630 const struct quic_cid *dcid,
5631 struct sockaddr_storage *addr)
5632{
5633 int ret = 0;
5634 unsigned char *p;
5635 unsigned char aad[sizeof(uint32_t) + sizeof(in_port_t) +
Amaury Denoyelle6c940562022-10-18 11:05:02 +02005636 sizeof(struct in6_addr) + QUIC_CID_MAXLEN];
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005637 size_t aadlen;
5638 unsigned char salt[QUIC_RETRY_TOKEN_SALTLEN];
5639 unsigned char key[QUIC_TLS_KEY_LEN];
5640 unsigned char iv[QUIC_TLS_IV_LEN];
5641 const unsigned char *sec = (const unsigned char *)global.cluster_secret;
5642 size_t seclen = strlen(global.cluster_secret);
5643 EVP_CIPHER_CTX *ctx = NULL;
5644 const EVP_CIPHER *aead = EVP_aes_128_gcm();
5645 uint32_t timestamp = now_ms;
5646
5647 TRACE_ENTER(QUIC_EV_CONN_TXPKT);
5648
5649 /* We copy the odcid into the token, prefixed by its one byte
5650 * length, the format token byte. It is followed by an AEAD TAG, and finally
5651 * the random bytes used to derive the secret to encrypt the token.
5652 */
5653 if (1 + dcid->len + 1 + QUIC_TLS_TAG_LEN + sizeof salt > len)
5654 goto err;
5655
5656 aadlen = quic_generate_retry_token_aad(aad, version, dcid, addr);
5657 /* TODO: RAND_bytes() should be replaced */
5658 if (RAND_bytes(salt, sizeof salt) != 1) {
5659 TRACE_ERROR("RAND_bytes()", QUIC_EV_CONN_TXPKT);
5660 goto err;
5661 }
5662
5663 if (!quic_tls_derive_retry_token_secret(EVP_sha256(), key, sizeof key, iv, sizeof iv,
5664 salt, sizeof salt, sec, seclen)) {
5665 TRACE_ERROR("quic_tls_derive_retry_token_secret() failed", QUIC_EV_CONN_TXPKT);
5666 goto err;
5667 }
5668
5669 if (!quic_tls_tx_ctx_init(&ctx, aead, key)) {
5670 TRACE_ERROR("quic_tls_tx_ctx_init() failed", QUIC_EV_CONN_TXPKT);
5671 goto err;
5672 }
5673
5674 /* Token build */
5675 p = buf;
5676 *p++ = QUIC_TOKEN_FMT_RETRY,
5677 *p++ = odcid->len;
5678 memcpy(p, odcid->data, odcid->len);
5679 p += odcid->len;
5680 write_u32(p, htonl(timestamp));
5681 p += sizeof timestamp;
5682
5683 /* Do not encrypt the QUIC_TOKEN_FMT_RETRY byte */
5684 if (!quic_tls_encrypt(buf + 1, p - buf - 1, aad, aadlen, ctx, aead, key, iv)) {
5685 TRACE_ERROR("quic_tls_encrypt() failed", QUIC_EV_CONN_TXPKT);
5686 goto err;
5687 }
5688
5689 p += QUIC_TLS_TAG_LEN;
5690 memcpy(p, salt, sizeof salt);
5691 p += sizeof salt;
5692 EVP_CIPHER_CTX_free(ctx);
5693
5694 ret = p - buf;
5695 leave:
5696 TRACE_LEAVE(QUIC_EV_CONN_TXPKT);
5697 return ret;
5698
5699 err:
5700 if (ctx)
5701 EVP_CIPHER_CTX_free(ctx);
5702 goto leave;
5703}
5704
5705/* QUIC server only function.
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02005706 *
5707 * Check the validity of the Retry token from Initial packet <pkt>. <dgram> is
5708 * the UDP datagram containing <pkt> and <l> is the listener instance on which
5709 * it was received. If the token is valid, the ODCID of <qc> QUIC connection
5710 * will be put into <odcid>. <qc> is used to retrieve the QUIC version needed
5711 * to validate the token but it can be NULL : in this case the version will be
5712 * retrieved from the packet.
5713 *
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005714 * Return 1 if succeeded, 0 if not.
5715 */
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02005716
5717static int quic_retry_token_check(struct quic_rx_packet *pkt,
5718 struct quic_dgram *dgram,
5719 struct listener *l,
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005720 struct quic_conn *qc,
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02005721 struct quic_cid *odcid)
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005722{
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02005723 struct proxy *prx;
5724 struct quic_counters *prx_counters;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005725 int ret = 0;
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02005726 unsigned char *token = pkt->token;
5727 const uint64_t tokenlen = pkt->token_len;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005728 unsigned char buf[128];
5729 unsigned char aad[sizeof(uint32_t) + sizeof(in_port_t) +
Amaury Denoyelle6c940562022-10-18 11:05:02 +02005730 sizeof(struct in6_addr) + QUIC_CID_MAXLEN];
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005731 size_t aadlen;
5732 const unsigned char *salt;
5733 unsigned char key[QUIC_TLS_KEY_LEN];
5734 unsigned char iv[QUIC_TLS_IV_LEN];
5735 const unsigned char *sec = (const unsigned char *)global.cluster_secret;
5736 size_t seclen = strlen(global.cluster_secret);
5737 EVP_CIPHER_CTX *ctx = NULL;
5738 const EVP_CIPHER *aead = EVP_aes_128_gcm();
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02005739 const struct quic_version *qv = qc ? qc->original_version :
5740 pkt->version;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005741
5742 TRACE_ENTER(QUIC_EV_CONN_LPKT, qc);
5743
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02005744 /* The caller must ensure this. */
5745 BUG_ON(!global.cluster_secret || !pkt->token_len);
5746
5747 prx = l->bind_conf->frontend;
5748 prx_counters = EXTRA_COUNTERS_GET(prx->extra_counters_fe, &quic_stats_module);
5749
5750 if (*pkt->token != QUIC_TOKEN_FMT_RETRY) {
5751 /* TODO: New token check */
5752 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT, qc, NULL, NULL, pkt->version);
5753 goto leave;
5754 }
5755
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005756 if (sizeof buf < tokenlen) {
5757 TRACE_ERROR("too short buffer", QUIC_EV_CONN_LPKT, qc);
5758 goto err;
5759 }
5760
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02005761 aadlen = quic_generate_retry_token_aad(aad, qv->num, &pkt->scid, &dgram->saddr);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005762 salt = token + tokenlen - QUIC_RETRY_TOKEN_SALTLEN;
5763 if (!quic_tls_derive_retry_token_secret(EVP_sha256(), key, sizeof key, iv, sizeof iv,
5764 salt, QUIC_RETRY_TOKEN_SALTLEN, sec, seclen)) {
5765 TRACE_ERROR("Could not derive retry secret", QUIC_EV_CONN_LPKT, qc);
5766 goto err;
5767 }
5768
5769 if (!quic_tls_rx_ctx_init(&ctx, aead, key)) {
5770 TRACE_ERROR("quic_tls_rx_ctx_init() failed", QUIC_EV_CONN_LPKT, qc);
5771 goto err;
5772 }
5773
5774 /* Do not decrypt the QUIC_TOKEN_FMT_RETRY byte */
5775 if (!quic_tls_decrypt2(buf, token + 1, tokenlen - QUIC_RETRY_TOKEN_SALTLEN - 1, aad, aadlen,
5776 ctx, aead, key, iv)) {
5777 TRACE_ERROR("Could not decrypt retry token", QUIC_EV_CONN_LPKT, qc);
5778 goto err;
5779 }
5780
5781 if (parse_retry_token(qc, buf, buf + tokenlen - QUIC_RETRY_TOKEN_SALTLEN - 1, odcid)) {
5782 TRACE_ERROR("Error during Initial token parsing", QUIC_EV_CONN_LPKT, qc);
5783 goto err;
5784 }
5785
5786 EVP_CIPHER_CTX_free(ctx);
5787
5788 ret = 1;
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02005789 HA_ATOMIC_INC(&prx_counters->retry_validated);
5790
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005791 leave:
5792 TRACE_LEAVE(QUIC_EV_CONN_LPKT, qc);
5793 return ret;
5794
5795 err:
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02005796 HA_ATOMIC_INC(&prx_counters->retry_error);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005797 if (ctx)
5798 EVP_CIPHER_CTX_free(ctx);
5799 goto leave;
5800}
5801
5802/* Generate a Retry packet and send it on <fd> socket to <addr> in response to
5803 * the Initial <pkt> packet.
5804 *
5805 * Returns 0 on success else non-zero.
5806 */
5807static int send_retry(int fd, struct sockaddr_storage *addr,
5808 struct quic_rx_packet *pkt, const struct quic_version *qv)
5809{
5810 int ret = 0;
5811 unsigned char buf[128];
5812 int i = 0, token_len;
5813 const socklen_t addrlen = get_addr_len(addr);
5814 struct quic_cid scid;
5815
5816 TRACE_ENTER(QUIC_EV_CONN_TXPKT);
5817
5818 /* long header + fixed bit + packet type QUIC_PACKET_TYPE_RETRY */
5819 buf[i++] = (QUIC_PACKET_LONG_HEADER_BIT | QUIC_PACKET_FIXED_BIT) |
5820 (quic_pkt_type(QUIC_PACKET_TYPE_RETRY, qv->num) << QUIC_PACKET_TYPE_SHIFT);
5821 /* version */
5822 buf[i++] = *((unsigned char *)&qv->num + 3);
5823 buf[i++] = *((unsigned char *)&qv->num + 2);
5824 buf[i++] = *((unsigned char *)&qv->num + 1);
5825 buf[i++] = *(unsigned char *)&qv->num;
5826
5827 /* Use the SCID from <pkt> for Retry DCID. */
5828 buf[i++] = pkt->scid.len;
5829 memcpy(&buf[i], pkt->scid.data, pkt->scid.len);
5830 i += pkt->scid.len;
5831
5832 /* Generate a new CID to be used as SCID for the Retry packet. */
5833 scid.len = QUIC_HAP_CID_LEN;
5834 /* TODO: RAND_bytes() should be replaced */
5835 if (RAND_bytes(scid.data, scid.len) != 1) {
5836 TRACE_ERROR("RAND_bytes() failed", QUIC_EV_CONN_TXPKT);
5837 goto out;
5838 }
5839
5840 buf[i++] = scid.len;
5841 memcpy(&buf[i], scid.data, scid.len);
5842 i += scid.len;
5843
5844 /* token */
5845 if (!(token_len = quic_generate_retry_token(&buf[i], sizeof(buf) - i, qv->num,
5846 &pkt->dcid, &pkt->scid, addr))) {
5847 TRACE_ERROR("quic_generate_retry_token() failed", QUIC_EV_CONN_TXPKT);
5848 goto out;
5849 }
5850
5851 i += token_len;
5852
5853 /* token integrity tag */
5854 if ((&buf[i] - buf < QUIC_TLS_TAG_LEN) ||
5855 !quic_tls_generate_retry_integrity_tag(pkt->dcid.data,
5856 pkt->dcid.len, buf, i, qv)) {
5857 TRACE_ERROR("quic_tls_generate_retry_integrity_tag() failed", QUIC_EV_CONN_TXPKT);
5858 goto out;
5859 }
5860
5861 i += QUIC_TLS_TAG_LEN;
5862
5863 if (sendto(fd, buf, i, 0, (struct sockaddr *)addr, addrlen) < 0) {
5864 TRACE_ERROR("quic_tls_generate_retry_integrity_tag() failed", QUIC_EV_CONN_TXPKT);
5865 goto out;
5866 }
5867
5868 ret = 1;
5869 out:
5870 TRACE_LEAVE(QUIC_EV_CONN_TXPKT);
5871 return !ret;
5872}
5873
5874/* Retrieve a quic_conn instance from the <pkt> DCID field. If the packet is of
5875 * type INITIAL, the ODCID tree is first used. In this case, <saddr> is
5876 * concatenated to the <pkt> DCID field.
5877 *
5878 * Returns the instance or NULL if not found.
5879 */
5880static struct quic_conn *retrieve_qc_conn_from_cid(struct quic_rx_packet *pkt,
5881 struct listener *l,
5882 struct sockaddr_storage *saddr)
5883{
5884 struct quic_conn *qc = NULL;
5885 struct ebmb_node *node;
5886 struct quic_connection_id *id;
5887 /* set if the quic_conn is found in the second DCID tree */
5888
5889 TRACE_ENTER(QUIC_EV_CONN_RXPKT);
5890
5891 /* Look first into ODCIDs tree for INITIAL/0-RTT packets. */
5892 if (pkt->type == QUIC_PACKET_TYPE_INITIAL ||
5893 pkt->type == QUIC_PACKET_TYPE_0RTT) {
5894 /* DCIDs of first packets coming from multiple clients may have
5895 * the same values. Let's distinguish them by concatenating the
5896 * socket addresses.
5897 */
5898 quic_cid_saddr_cat(&pkt->dcid, saddr);
5899 node = ebmb_lookup(&quic_dghdlrs[tid].odcids, pkt->dcid.data,
5900 pkt->dcid.len + pkt->dcid.addrlen);
5901 if (node) {
5902 qc = ebmb_entry(node, struct quic_conn, odcid_node);
5903 goto end;
5904 }
5905 }
5906
5907 /* Look into DCIDs tree for non-INITIAL/0-RTT packets. This may be used
5908 * also for INITIAL/0-RTT non-first packets with the final DCID in
5909 * used.
5910 */
5911 node = ebmb_lookup(&quic_dghdlrs[tid].cids, pkt->dcid.data, pkt->dcid.len);
5912 if (!node)
5913 goto end;
5914
5915 id = ebmb_entry(node, struct quic_connection_id, node);
5916 qc = id->qc;
5917
5918 /* If found in DCIDs tree, remove the quic_conn from the ODCIDs tree.
5919 * If already done, this is a noop.
5920 */
5921 if (qc)
5922 ebmb_delete(&qc->odcid_node);
5923
5924 end:
5925 TRACE_LEAVE(QUIC_EV_CONN_RXPKT, qc);
5926 return qc;
5927}
5928
5929/* Try to allocate the <*ssl> SSL session object for <qc> QUIC connection
5930 * with <ssl_ctx> as SSL context inherited settings. Also set the transport
5931 * parameters of this session.
5932 * This is the responsibility of the caller to check the validity of all the
5933 * pointers passed as parameter to this function.
5934 * Return 0 if succeeded, -1 if not. If failed, sets the ->err_code member of <qc->conn> to
5935 * CO_ER_SSL_NO_MEM.
5936 */
5937static int qc_ssl_sess_init(struct quic_conn *qc, SSL_CTX *ssl_ctx, SSL **ssl,
5938 unsigned char *params, size_t params_len)
5939{
5940 int retry, ret = -1;
5941
5942 TRACE_ENTER(QUIC_EV_CONN_NEW, qc);
5943
5944 retry = 1;
5945 retry:
5946 *ssl = SSL_new(ssl_ctx);
5947 if (!*ssl) {
5948 if (!retry--)
5949 goto err;
5950
5951 pool_gc(NULL);
5952 goto retry;
5953 }
5954
5955 if (!SSL_set_quic_method(*ssl, &ha_quic_method) ||
5956 !SSL_set_ex_data(*ssl, ssl_qc_app_data_index, qc)) {
5957 SSL_free(*ssl);
5958 *ssl = NULL;
5959 if (!retry--)
5960 goto err;
5961
5962 pool_gc(NULL);
5963 goto retry;
5964 }
5965
5966 ret = 0;
5967 leave:
5968 TRACE_LEAVE(QUIC_EV_CONN_NEW, qc);
5969 return ret;
5970
5971 err:
5972 qc->conn->err_code = CO_ER_SSL_NO_MEM;
5973 goto leave;
5974}
5975
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005976/* Allocate the ssl_sock_ctx from connection <qc>. This creates the tasklet
5977 * used to process <qc> received packets. The allocated context is stored in
5978 * <qc.xprt_ctx>.
5979 *
5980 * Returns 0 on success else non-zero.
5981 */
5982static int qc_conn_alloc_ssl_ctx(struct quic_conn *qc)
5983{
5984 int ret = 0;
5985 struct bind_conf *bc = qc->li->bind_conf;
5986 struct ssl_sock_ctx *ctx = NULL;
5987
5988 TRACE_ENTER(QUIC_EV_CONN_NEW, qc);
5989
5990 ctx = pool_zalloc(pool_head_quic_conn_ctx);
5991 if (!ctx) {
5992 TRACE_ERROR("SSL context allocation failed", QUIC_EV_CONN_TXPKT);
5993 goto err;
5994 }
5995
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02005996 ctx->subs = NULL;
5997 ctx->xprt_ctx = NULL;
5998 ctx->qc = qc;
5999
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006000 if (qc_is_listener(qc)) {
6001 if (qc_ssl_sess_init(qc, bc->initial_ctx, &ctx->ssl,
6002 qc->enc_params, qc->enc_params_len) == -1) {
6003 goto err;
6004 }
6005#if (HA_OPENSSL_VERSION_NUMBER >= 0x10101000L)
6006 /* Enabling 0-RTT */
6007 if (bc->ssl_conf.early_data)
6008 SSL_set_quic_early_data_enabled(ctx->ssl, 1);
6009#endif
6010
6011 SSL_set_accept_state(ctx->ssl);
6012 }
6013
6014 ctx->xprt = xprt_get(XPRT_QUIC);
6015
6016 /* Store the allocated context in <qc>. */
6017 qc->xprt_ctx = ctx;
6018
6019 ret = 1;
6020 leave:
6021 TRACE_LEAVE(QUIC_EV_CONN_NEW, qc);
6022 return !ret;
6023
6024 err:
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006025 pool_free(pool_head_quic_conn_ctx, ctx);
6026 goto leave;
6027}
6028
6029/* Check that all the bytes between <buf> included and <end> address
6030 * excluded are null. This is the responsibility of the caller to
6031 * check that there is at least one byte between <buf> end <end>.
6032 * Return 1 if this all the bytes are null, 0 if not.
6033 */
6034static inline int quic_padding_check(const unsigned char *buf,
6035 const unsigned char *end)
6036{
6037 while (buf < end && !*buf)
6038 buf++;
6039
6040 return buf == end;
6041}
6042
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006043/* Find the associated connection to the packet <pkt> or create a new one if
6044 * this is an Initial packet. <dgram> is the datagram containing the packet and
6045 * <l> is the listener instance on which it was received.
6046 *
6047 * Returns the quic-conn instance or NULL.
6048 */
6049static struct quic_conn *quic_rx_pkt_retrieve_conn(struct quic_rx_packet *pkt,
6050 struct quic_dgram *dgram,
6051 struct listener *l)
6052{
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02006053 struct quic_cid token_odcid = { .len = 0 };
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006054 struct quic_conn *qc = NULL;
6055 struct proxy *prx;
6056 struct quic_counters *prx_counters;
6057
6058 TRACE_ENTER(QUIC_EV_CONN_LPKT);
6059
6060 prx = l->bind_conf->frontend;
6061 prx_counters = EXTRA_COUNTERS_GET(prx->extra_counters_fe, &quic_stats_module);
6062
6063 qc = retrieve_qc_conn_from_cid(pkt, l, &dgram->saddr);
6064
6065 if (pkt->type == QUIC_PACKET_TYPE_INITIAL) {
6066 BUG_ON(!pkt->version); /* This must not happen. */
6067
6068 if (global.cluster_secret && pkt->token_len) {
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02006069 if (!quic_retry_token_check(pkt, dgram, l, qc, &token_odcid))
6070 goto err;
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006071 }
6072
6073 if (!qc) {
6074 int ipv4;
6075
6076 if (global.cluster_secret && !pkt->token_len && !(l->bind_conf->options & BC_O_QUIC_FORCE_RETRY) &&
6077 HA_ATOMIC_LOAD(&prx_counters->half_open_conn) >= global.tune.quic_retry_threshold) {
6078 TRACE_PROTO("Initial without token, sending retry",
6079 QUIC_EV_CONN_LPKT, NULL, NULL, NULL, pkt->version);
6080 if (send_retry(l->rx.fd, &dgram->saddr, pkt, pkt->version)) {
6081 TRACE_ERROR("Error during Retry generation",
6082 QUIC_EV_CONN_LPKT, NULL, NULL, NULL, pkt->version);
6083 goto out;
6084 }
6085
6086 HA_ATOMIC_INC(&prx_counters->retry_sent);
6087 goto out;
6088 }
6089
6090 /* RFC 9000 7.2. Negotiating Connection IDs:
6091 * When an Initial packet is sent by a client that has not previously
6092 * received an Initial or Retry packet from the server, the client
6093 * populates the Destination Connection ID field with an unpredictable
6094 * value. This Destination Connection ID MUST be at least 8 bytes in length.
6095 */
6096 if (pkt->dcid.len < QUIC_ODCID_MINLEN) {
6097 TRACE_PROTO("dropped packet",
6098 QUIC_EV_CONN_LPKT, NULL, NULL, NULL, pkt->version);
6099 goto err;
6100 }
6101
6102 pkt->saddr = dgram->saddr;
6103 ipv4 = dgram->saddr.ss_family == AF_INET;
6104
Amaury Denoyelle9e3026c2022-10-17 11:13:07 +02006105 qc = qc_new_conn(pkt->version, ipv4, &pkt->dcid, &pkt->scid, &token_odcid,
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006106 &dgram->daddr, &pkt->saddr, 1,
6107 !!pkt->token_len, l);
6108 if (qc == NULL)
6109 goto err;
6110
6111 HA_ATOMIC_INC(&prx_counters->half_open_conn);
6112 /* Insert the DCID the QUIC client has chosen (only for listeners) */
6113 ebmb_insert(&quic_dghdlrs[tid].odcids, &qc->odcid_node,
6114 qc->odcid.len + qc->odcid.addrlen);
6115 }
6116 }
6117 else if (!qc) {
6118 TRACE_PROTO("No connection on a non Initial packet", QUIC_EV_CONN_LPKT, NULL, NULL, NULL, pkt->version);
6119 if (global.cluster_secret && !send_stateless_reset(l, &dgram->saddr, pkt))
6120 TRACE_ERROR("stateless reset not sent", QUIC_EV_CONN_LPKT, qc);
6121 goto err;
6122 }
6123
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006124 out:
6125 TRACE_LEAVE(QUIC_EV_CONN_LPKT, qc);
6126 return qc;
6127
6128 err:
6129 HA_ATOMIC_INC(&prx_counters->dropped_pkt);
6130 TRACE_LEAVE(QUIC_EV_CONN_LPKT);
6131 return NULL;
6132}
6133
Amaury Denoyelle98289692022-10-19 15:37:44 +02006134/* Parse a QUIC packet starting at <buf>. Data won't be read after <end> even
6135 * if the packet is incomplete. This function will populate fields of <pkt>
6136 * instance, most notably its length. <dgram> is the UDP datagram which
6137 * contains the parsed packet. <l> is the listener instance on which it was
6138 * received.
6139 *
6140 * Returns 0 on success else non-zero. Packet length is guaranteed to be set to
6141 * the real packet value or to cover all data between <buf> and <end> : this is
6142 * useful to reject a whole datagram.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006143 */
Amaury Denoyelle98289692022-10-19 15:37:44 +02006144static int quic_rx_pkt_parse(struct quic_rx_packet *pkt,
6145 unsigned char *buf, const unsigned char *end,
6146 struct quic_dgram *dgram, struct listener *l)
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006147{
Amaury Denoyelle98289692022-10-19 15:37:44 +02006148 const unsigned char *beg = buf;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006149 struct proxy *prx;
6150 struct quic_counters *prx_counters;
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006151 int long_header = 0;
Willy Tarreau33a68702022-11-24 09:16:41 +01006152 uint32_t version = 0;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006153 const struct quic_version *qv = NULL;
6154
6155 TRACE_ENTER(QUIC_EV_CONN_LPKT);
6156
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006157 prx = l->bind_conf->frontend;
6158 prx_counters = EXTRA_COUNTERS_GET(prx->extra_counters_fe, &quic_stats_module);
6159 /* This ist only to please to traces and distinguish the
6160 * packet with parsed packet number from others.
6161 */
6162 pkt->pn_node.key = (uint64_t)-1;
6163 if (end <= buf) {
6164 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT);
6165 goto drop;
6166 }
6167
6168 /* Fixed bit */
6169 if (!(*buf & QUIC_PACKET_FIXED_BIT)) {
Amaury Denoyelledeb7c872022-10-19 17:14:28 +02006170 if (!(pkt->flags & QUIC_FL_RX_PACKET_DGRAM_FIRST) &&
6171 quic_padding_check(buf, end)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006172 /* Some browsers may pad the remaining datagram space with null bytes.
6173 * That is what we called add padding out of QUIC packets. Such
6174 * datagrams must be considered as valid. But we can only consume
6175 * the remaining space.
6176 */
6177 pkt->len = end - buf;
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006178 goto drop_silent;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006179 }
6180
6181 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT);
6182 goto drop;
6183 }
6184
6185 /* Header form */
6186 if (!qc_parse_hd_form(pkt, &buf, end, &long_header, &version)) {
6187 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT);
6188 goto drop;
6189 }
6190
6191 if (long_header) {
6192 uint64_t len;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006193
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006194 TRACE_PROTO("long header packet received", QUIC_EV_CONN_LPKT);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006195 if (!quic_packet_read_long_header(&buf, end, pkt)) {
6196 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT);
6197 goto drop;
6198 }
6199
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006200 if (pkt->type == QUIC_PACKET_TYPE_INITIAL &&
6201 dgram->len < QUIC_INITIAL_PACKET_MINLEN) {
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006202 TRACE_PROTO("Too short datagram with an Initial packet", QUIC_EV_CONN_LPKT);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006203 HA_ATOMIC_INC(&prx_counters->too_short_initial_dgram);
6204 goto drop;
6205 }
6206
6207 /* When multiple QUIC packets are coalesced on the same UDP datagram,
6208 * they must have the same DCID.
6209 */
Amaury Denoyelledeb7c872022-10-19 17:14:28 +02006210 if (!(pkt->flags & QUIC_FL_RX_PACKET_DGRAM_FIRST) &&
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006211 (pkt->dcid.len != dgram->dcid_len ||
6212 memcmp(dgram->dcid, pkt->dcid.data, pkt->dcid.len))) {
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006213 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006214 goto drop;
6215 }
6216
6217 /* Retry of Version Negotiation packets are only sent by servers */
6218 if (pkt->type == QUIC_PACKET_TYPE_RETRY || !version) {
6219 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT);
6220 goto drop;
6221 }
6222
6223 /* RFC9000 6. Version Negotiation */
6224 qv = qc_supported_version(version);
6225 if (!qv) {
6226 /* unsupported version, send Negotiation packet */
6227 if (send_version_negotiation(l->rx.fd, &dgram->saddr, pkt)) {
6228 TRACE_ERROR("VN packet not sent", QUIC_EV_CONN_LPKT);
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006229 goto drop_silent;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006230 }
6231
6232 TRACE_PROTO("VN packet sent", QUIC_EV_CONN_LPKT);
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006233 goto drop_silent;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006234 }
Amaury Denoyelle0eae5722022-10-17 18:05:18 +02006235 pkt->version = qv;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006236
6237 /* For Initial packets, and for servers (QUIC clients connections),
6238 * there is no Initial connection IDs storage.
6239 */
6240 if (pkt->type == QUIC_PACKET_TYPE_INITIAL) {
6241 uint64_t token_len;
6242
6243 if (!quic_dec_int(&token_len, (const unsigned char **)&buf, end) ||
6244 end - buf < token_len) {
6245 TRACE_PROTO("Packet dropped",
6246 QUIC_EV_CONN_LPKT, NULL, NULL, NULL, qv);
6247 goto drop;
6248 }
6249
6250 /* TODO Retry should be automatically activated if
6251 * suspect network usage is detected.
6252 */
6253 if (global.cluster_secret && !token_len) {
6254 if (l->bind_conf->options & BC_O_QUIC_FORCE_RETRY) {
6255 TRACE_PROTO("Initial without token, sending retry",
Amaury Denoyelle90121b32022-09-27 10:35:29 +02006256 QUIC_EV_CONN_LPKT, NULL, NULL, NULL, qv);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006257 if (send_retry(l->rx.fd, &dgram->saddr, pkt, qv)) {
6258 TRACE_PROTO("Error during Retry generation",
Amaury Denoyelle90121b32022-09-27 10:35:29 +02006259 QUIC_EV_CONN_LPKT, NULL, NULL, NULL, qv);
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006260 goto drop_silent;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006261 }
6262
6263 HA_ATOMIC_INC(&prx_counters->retry_sent);
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006264 goto drop_silent;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006265 }
6266 }
6267 else if (!global.cluster_secret && token_len) {
6268 /* Impossible case: a token was received without configured
6269 * cluster secret.
6270 */
6271 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT,
6272 NULL, NULL, NULL, qv);
6273 goto drop;
6274 }
6275
6276 pkt->token = buf;
6277 pkt->token_len = token_len;
6278 buf += pkt->token_len;
6279 }
6280 else if (pkt->type != QUIC_PACKET_TYPE_0RTT) {
6281 if (pkt->dcid.len != QUIC_HAP_CID_LEN) {
6282 TRACE_PROTO("Packet dropped",
6283 QUIC_EV_CONN_LPKT, NULL, NULL, NULL, qv);
6284 goto drop;
6285 }
6286 }
6287
6288 if (!quic_dec_int(&len, (const unsigned char **)&buf, end) ||
6289 end - buf < len) {
6290 TRACE_PROTO("Packet dropped",
6291 QUIC_EV_CONN_LPKT, NULL, NULL, NULL, qv);
6292 goto drop;
6293 }
6294
Amaury Denoyelle845169d2022-10-17 18:05:26 +02006295 /* Packet Number is stored here. Packet Length totalizes the
6296 * rest of the content.
6297 */
6298 pkt->pn_offset = buf - beg;
6299 pkt->len = pkt->pn_offset + len;
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006300
6301 /* Interrupt parsing after packet length retrieval : this
6302 * ensures that only the packet is dropped but not the whole
6303 * datagram.
6304 */
6305 if (pkt->type == QUIC_PACKET_TYPE_0RTT && !l->bind_conf->ssl_conf.early_data) {
6306 TRACE_PROTO("0-RTT packet not supported", QUIC_EV_CONN_LPKT);
6307 goto drop;
6308 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006309 }
6310 else {
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006311 TRACE_PROTO("short header packet received", QUIC_EV_CONN_LPKT);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006312 if (end - buf < QUIC_HAP_CID_LEN) {
6313 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT);
6314 goto drop;
6315 }
6316
6317 memcpy(pkt->dcid.data, buf, QUIC_HAP_CID_LEN);
6318 pkt->dcid.len = QUIC_HAP_CID_LEN;
6319
6320 /* When multiple QUIC packets are coalesced on the same UDP datagram,
6321 * they must have the same DCID.
6322 */
Amaury Denoyelledeb7c872022-10-19 17:14:28 +02006323 if (!(pkt->flags & QUIC_FL_RX_PACKET_DGRAM_FIRST) &&
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006324 (pkt->dcid.len != dgram->dcid_len ||
6325 memcmp(dgram->dcid, pkt->dcid.data, pkt->dcid.len))) {
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006326 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006327 goto drop;
6328 }
6329
6330 buf += QUIC_HAP_CID_LEN;
6331
Amaury Denoyelle845169d2022-10-17 18:05:26 +02006332 pkt->pn_offset = buf - beg;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006333 /* A short packet is the last one of a UDP datagram. */
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006334 pkt->len = end - beg;
Amaury Denoyelle449b1a82022-10-19 15:28:44 +02006335 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006336
Amaury Denoyelle98289692022-10-19 15:37:44 +02006337 TRACE_LEAVE(QUIC_EV_CONN_LPKT, NULL, pkt, NULL, qv);
6338 return 0;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006339
Amaury Denoyelle98289692022-10-19 15:37:44 +02006340 drop:
6341 HA_ATOMIC_INC(&prx_counters->dropped_pkt);
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006342 drop_silent:
Amaury Denoyelle98289692022-10-19 15:37:44 +02006343 if (!pkt->len)
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006344 pkt->len = end - beg;
Amaury Denoyelle98289692022-10-19 15:37:44 +02006345 TRACE_LEAVE(QUIC_EV_CONN_LPKT, NULL, pkt, NULL, qv);
6346 return -1;
6347}
6348
6349/* Check if received packet <pkt> should be drop due to <qc> already in closing
6350 * state. This can be true if a CONNECTION_CLOSE has already been emitted for
6351 * this connection.
6352 *
6353 * Returns false if connection is not in closing state else true. The caller
6354 * should drop the whole datagram in the last case to not mess up <qc>
6355 * CONNECTION_CLOSE rate limit counter.
6356 */
6357static int qc_rx_check_closing(struct quic_conn *qc,
6358 struct quic_rx_packet *pkt)
6359{
6360 if (!(qc->flags & QUIC_FL_CONN_CLOSING))
6361 return 0;
6362
6363 TRACE_STATE("Closing state connection", QUIC_EV_CONN_LPKT, qc, NULL, NULL, pkt->version);
6364
6365 /* Check if CONNECTION_CLOSE rate reemission is reached. */
6366 if (++qc->nb_pkt_since_cc >= qc->nb_pkt_for_cc) {
6367 qc->flags |= QUIC_FL_CONN_IMMEDIATE_CLOSE;
6368 qc->nb_pkt_for_cc++;
6369 qc->nb_pkt_since_cc = 0;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006370 }
6371
Amaury Denoyelle98289692022-10-19 15:37:44 +02006372 return 1;
6373}
6374
Amaury Denoyelleeec0b3c2022-12-02 09:57:32 +01006375/* React to a connection migration initiated on <qc> by a client with the new
6376 * path addresses <peer_addr>/<local_addr>.
6377 *
6378 * Returns 0 on success else non-zero.
6379 */
6380static int qc_handle_conn_migration(struct quic_conn *qc,
6381 const struct sockaddr_storage *peer_addr,
6382 const struct sockaddr_storage *local_addr)
6383{
6384 TRACE_ENTER(QUIC_EV_CONN_LPKT, qc);
6385
Frédéric Lécaille6fc86972023-01-12 08:29:23 +01006386 /* RFC 9000. Connection Migration
6387 *
6388 * If the peer sent the disable_active_migration transport parameter,
6389 * an endpoint also MUST NOT send packets (including probing packets;
6390 * see Section 9.1) from a different local address to the address the peer
6391 * used during the handshake, unless the endpoint has acted on a
6392 * preferred_address transport parameter from the peer.
6393 */
6394 if (qc->li->bind_conf->quic_params.disable_active_migration) {
6395 TRACE_ERROR("Active migration was disabled, datagram dropped", QUIC_EV_CONN_LPKT, qc);
6396 goto err;
6397 }
6398
Amaury Denoyelleeec0b3c2022-12-02 09:57:32 +01006399 /* RFC 9000 9. Connection Migration
6400 *
Amaury Denoyelleeb6be982022-11-21 11:14:45 +01006401 * The design of QUIC relies on endpoints retaining a stable address for
6402 * the duration of the handshake. An endpoint MUST NOT initiate
6403 * connection migration before the handshake is confirmed, as defined in
6404 * Section 4.1.2 of [QUIC-TLS].
6405 */
6406 if (qc->state < QUIC_HS_ST_COMPLETE) {
6407 TRACE_STATE("Connection migration during handshake rejected", QUIC_EV_CONN_LPKT, qc);
6408 goto err;
6409 }
6410
6411 /* RFC 9000 9. Connection Migration
6412 *
Amaury Denoyelleeec0b3c2022-12-02 09:57:32 +01006413 * TODO
6414 * An endpoint MUST
6415 * perform path validation (Section 8.2) if it detects any change to a
6416 * peer's address, unless it has previously validated that address.
6417 */
6418
Amaury Denoyelled3083c92022-12-01 16:20:06 +01006419 /* Update quic-conn owned socket if in used.
6420 * TODO try to reuse it instead of closing and opening a new one.
6421 */
6422 if (qc_test_fd(qc)) {
6423 /* TODO try to reuse socket instead of closing it and opening a new one. */
6424 TRACE_STATE("Connection migration detected, allocate a new connection socket", QUIC_EV_CONN_LPKT, qc);
6425 qc_release_fd(qc, 1);
6426 qc_alloc_fd(qc, local_addr, peer_addr);
6427 }
6428
Amaury Denoyelleeec0b3c2022-12-02 09:57:32 +01006429 qc->local_addr = *local_addr;
6430 qc->peer_addr = *peer_addr;
6431 HA_ATOMIC_INC(&qc->prx_counters->conn_migration_done);
6432
6433 TRACE_LEAVE(QUIC_EV_CONN_LPKT, qc);
6434 return 0;
6435
6436 err:
6437 TRACE_LEAVE(QUIC_EV_CONN_LPKT, qc);
6438 return 1;
6439}
6440
Amaury Denoyelle98289692022-10-19 15:37:44 +02006441/* Handle a parsed packet <pkt> by the connection <qc>. Data will be copied
6442 * into <qc> receive buffer after header protection removal procedure.
6443 *
6444 * <dgram> must be set to the datagram which contains the QUIC packet. <beg>
6445 * must point to packet buffer first byte.
6446 *
6447 * <tasklist_head> may be non-NULL when the caller treat several datagrams for
6448 * different quic-conn. In this case, each quic-conn tasklet will be appended
6449 * to it in order to be woken up after the current task.
6450 *
6451 * The caller can safely removed the packet data. If packet refcount was not
6452 * incremented by this function, it means that the connection did not handled
6453 * it and it should be freed by the caller.
6454 */
6455static void qc_rx_pkt_handle(struct quic_conn *qc, struct quic_rx_packet *pkt,
6456 struct quic_dgram *dgram, unsigned char *beg,
6457 struct list **tasklist_head)
6458{
6459 const struct quic_version *qv = pkt->version;
6460 struct quic_enc_level *qel = NULL;
6461 size_t b_cspace;
6462 int io_cb_wakeup = 1;
6463
Amaury Denoyelle3f474e62022-11-24 17:15:08 +01006464 TRACE_ENTER(QUIC_EV_CONN_LPKT, qc, pkt, NULL, qv);
6465
Amaury Denoyelledeb7c872022-10-19 17:14:28 +02006466 if (pkt->flags & QUIC_FL_RX_PACKET_DGRAM_FIRST &&
6467 !quic_peer_validated_addr(qc) &&
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006468 qc->flags & QUIC_FL_CONN_ANTI_AMPLIFICATION_REACHED) {
6469 TRACE_PROTO("PTO timer must be armed after anti-amplication was reached",
6470 QUIC_EV_CONN_LPKT, qc, NULL, NULL, qv);
6471 /* Reset the anti-amplification bit. It will be set again
6472 * when sending the next packet if reached again.
6473 */
6474 qc->flags &= ~QUIC_FL_CONN_ANTI_AMPLIFICATION_REACHED;
6475 qc->flags |= QUIC_FL_CONN_IO_CB_WAKEUP;
6476 io_cb_wakeup = 1;
6477 }
6478
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006479 if (qc->flags & QUIC_FL_CONN_IMMEDIATE_CLOSE) {
6480 TRACE_PROTO("Connection error",
6481 QUIC_EV_CONN_LPKT, qc, NULL, NULL, qv);
6482 goto out;
6483 }
6484
6485 pkt->raw_len = pkt->len;
6486 quic_rx_pkts_del(qc);
6487 b_cspace = b_contig_space(&qc->rx.buf);
6488 if (b_cspace < pkt->len) {
6489 /* Do not consume buf if space not at the end. */
6490 if (b_tail(&qc->rx.buf) + b_cspace < b_wrap(&qc->rx.buf)) {
6491 TRACE_PROTO("Packet dropped",
6492 QUIC_EV_CONN_LPKT, qc, NULL, NULL, qv);
Amaury Denoyelle98289692022-10-19 15:37:44 +02006493 HA_ATOMIC_INC(&qc->prx_counters->dropped_pkt_bufoverrun);
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006494 goto drop_silent;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006495 }
6496
6497 /* Let us consume the remaining contiguous space. */
6498 if (b_cspace) {
6499 b_putchr(&qc->rx.buf, 0x00);
6500 b_cspace--;
6501 }
6502 b_add(&qc->rx.buf, b_cspace);
6503 if (b_contig_space(&qc->rx.buf) < pkt->len) {
6504 TRACE_PROTO("Too big packet",
6505 QUIC_EV_CONN_LPKT, qc, pkt, &pkt->len, qv);
Amaury Denoyelle98289692022-10-19 15:37:44 +02006506 HA_ATOMIC_INC(&qc->prx_counters->dropped_pkt_bufoverrun);
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006507 goto drop_silent;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006508 }
6509 }
6510
Amaury Denoyelle845169d2022-10-17 18:05:26 +02006511 if (!qc_try_rm_hp(qc, pkt, beg, &qel)) {
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006512 TRACE_PROTO("Packet dropped", QUIC_EV_CONN_LPKT, qc, NULL, NULL, qv);
6513 goto drop;
6514 }
6515
6516 TRACE_DATA("New packet", QUIC_EV_CONN_LPKT, qc, pkt, NULL, qv);
6517 if (pkt->aad_len)
6518 qc_pkt_insert(qc, pkt, qel);
6519 out:
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02006520 *tasklist_head = tasklet_wakeup_after(*tasklist_head,
6521 qc->wait_event.tasklet);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006522
Amaury Denoyelle6e56a9e2022-10-17 12:04:49 +02006523 drop_silent:
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006524 TRACE_LEAVE(QUIC_EV_CONN_LPKT, qc ? qc : NULL, pkt, NULL, qv);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006525 return;
6526
6527 drop:
Amaury Denoyelle98289692022-10-19 15:37:44 +02006528 HA_ATOMIC_INC(&qc->prx_counters->dropped_pkt);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006529 err:
6530 /* Wakeup the I/O handler callback if the PTO timer must be armed.
6531 * This cannot be done by this thread.
6532 */
Amaury Denoyelle2ed84002022-09-26 14:53:59 +02006533 if (io_cb_wakeup)
6534 tasklet_wakeup(qc->wait_event.tasklet);
6535
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006536 TRACE_LEAVE(QUIC_EV_CONN_LPKT, qc ? qc : NULL, pkt, NULL, qv);
6537}
6538
6539/* This function builds into <buf> buffer a QUIC long packet header.
6540 * Return 1 if enough room to build this header, 0 if not.
6541 */
6542static int quic_build_packet_long_header(unsigned char **buf, const unsigned char *end,
6543 int type, size_t pn_len,
6544 struct quic_conn *qc, const struct quic_version *ver)
6545{
6546 int ret = 0;
6547
6548 TRACE_ENTER(QUIC_EV_CONN_LPKT, qc);
6549
6550 if (end - *buf < sizeof ver->num + qc->dcid.len + qc->scid.len + 3) {
6551 TRACE_DEVEL("not enough room", QUIC_EV_CONN_LPKT, qc);
6552 goto leave;
6553 }
6554
6555 type = quic_pkt_type(type, ver->num);
6556 /* #0 byte flags */
6557 *(*buf)++ = QUIC_PACKET_FIXED_BIT | QUIC_PACKET_LONG_HEADER_BIT |
6558 (type << QUIC_PACKET_TYPE_SHIFT) | (pn_len - 1);
6559 /* Version */
6560 quic_write_uint32(buf, end, ver->num);
6561 *(*buf)++ = qc->dcid.len;
6562 /* Destination connection ID */
6563 if (qc->dcid.len) {
6564 memcpy(*buf, qc->dcid.data, qc->dcid.len);
6565 *buf += qc->dcid.len;
6566 }
6567 /* Source connection ID */
6568 *(*buf)++ = qc->scid.len;
6569 if (qc->scid.len) {
6570 memcpy(*buf, qc->scid.data, qc->scid.len);
6571 *buf += qc->scid.len;
6572 }
6573
6574 ret = 1;
6575 leave:
6576 TRACE_LEAVE(QUIC_EV_CONN_LPKT, qc);
6577 return ret;
6578}
6579
6580/* This function builds into <buf> buffer a QUIC short packet header.
6581 * Return 1 if enough room to build this header, 0 if not.
6582 */
6583static int quic_build_packet_short_header(unsigned char **buf, const unsigned char *end,
6584 size_t pn_len, struct quic_conn *qc,
6585 unsigned char tls_flags)
6586{
6587 int ret = 0;
6588
6589 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
6590
6591 if (end - *buf < 1 + qc->dcid.len) {
6592 TRACE_DEVEL("not enough room", QUIC_EV_CONN_LPKT, qc);
6593 goto leave;
6594 }
6595
6596 /* #0 byte flags */
6597 *(*buf)++ = QUIC_PACKET_FIXED_BIT |
6598 ((tls_flags & QUIC_FL_TLS_KP_BIT_SET) ? QUIC_PACKET_KEY_PHASE_BIT : 0) | (pn_len - 1);
6599 /* Destination connection ID */
6600 if (qc->dcid.len) {
6601 memcpy(*buf, qc->dcid.data, qc->dcid.len);
6602 *buf += qc->dcid.len;
6603 }
6604
6605 ret = 1;
6606 leave:
6607 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
6608 return ret;
6609}
6610
6611/* Apply QUIC header protection to the packet with <buf> as first byte address,
6612 * <pn> as address of the Packet number field, <pnlen> being this field length
6613 * with <aead> as AEAD cipher and <key> as secret key.
6614 * Returns 1 if succeeded or 0 if failed.
6615 */
6616static int quic_apply_header_protection(struct quic_conn *qc, unsigned char *buf,
6617 unsigned char *pn, size_t pnlen,
6618 struct quic_tls_ctx *tls_ctx)
6619
6620{
6621 int i, ret = 0;
6622 /* We need an IV of at least 5 bytes: one byte for bytes #0
6623 * and at most 4 bytes for the packet number
6624 */
6625 unsigned char mask[5] = {0};
6626 EVP_CIPHER_CTX *aes_ctx = tls_ctx->tx.hp_ctx;
6627
6628 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
6629
6630 if (!quic_tls_aes_encrypt(mask, pn + QUIC_PACKET_PN_MAXLEN, sizeof mask, aes_ctx)) {
6631 TRACE_ERROR("could not apply header protection", QUIC_EV_CONN_TXPKT, qc);
6632 goto out;
6633 }
6634
6635 *buf ^= mask[0] & (*buf & QUIC_PACKET_LONG_HEADER_BIT ? 0xf : 0x1f);
6636 for (i = 0; i < pnlen; i++)
6637 pn[i] ^= mask[i + 1];
6638
6639 ret = 1;
6640 out:
6641 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
6642 return ret;
6643}
6644
6645/* Reduce the encoded size of <ack_frm> ACK frame removing the last
6646 * ACK ranges if needed to a value below <limit> in bytes.
6647 * Return 1 if succeeded, 0 if not.
6648 */
6649static int quic_ack_frm_reduce_sz(struct quic_conn *qc,
6650 struct quic_frame *ack_frm, size_t limit)
6651{
6652 size_t room, ack_delay_sz;
6653 int ret = 0;
6654
6655 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
6656
6657 ack_delay_sz = quic_int_getsize(ack_frm->tx_ack.ack_delay);
6658 /* A frame is made of 1 byte for the frame type. */
6659 room = limit - ack_delay_sz - 1;
6660 if (!quic_rm_last_ack_ranges(qc, ack_frm->tx_ack.arngs, room))
6661 goto leave;
6662
6663 ret = 1 + ack_delay_sz + ack_frm->tx_ack.arngs->enc_sz;
6664 leave:
6665 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
6666 return ret;
6667}
6668
6669/* Prepare into <outlist> as most as possible ack-eliciting frame from their
6670 * <inlist> prebuilt frames for <qel> encryption level to be encoded in a buffer
6671 * with <room> as available room, and <*len> the packet Length field initialized
6672 * with the number of bytes already present in this buffer which must be taken
6673 * into an account for the Length packet field value. <headlen> is the number of
6674 * bytes already present in this packet before building frames.
6675 *
6676 * Update consequently <*len> to reflect the size of these frames built
6677 * by this function. Also attach these frames to <l> frame list.
6678 * Return 1 if at least one ack-eleciting frame could be built, 0 if not.
6679 */
6680static inline int qc_build_frms(struct list *outlist, struct list *inlist,
6681 size_t room, size_t *len, size_t headlen,
6682 struct quic_enc_level *qel,
6683 struct quic_conn *qc)
6684{
6685 int ret;
6686 struct quic_frame *cf, *cfbak;
6687
6688 TRACE_ENTER(QUIC_EV_CONN_BCFRMS, qc);
6689
6690 ret = 0;
6691 if (*len > room)
6692 goto leave;
6693
6694 /* If we are not probing we must take into an account the congestion
6695 * control window.
6696 */
6697 if (!qel->pktns->tx.pto_probe) {
6698 size_t remain = quic_path_prep_data(qc->path);
6699
6700 if (headlen > remain)
6701 goto leave;
6702
6703 room = QUIC_MIN(room, remain - headlen);
6704 }
6705
6706 TRACE_PROTO("************** frames build (headlen)",
6707 QUIC_EV_CONN_BCFRMS, qc, &headlen);
6708
6709 /* NOTE: switch/case block inside a loop, a successful status must be
6710 * returned by this function only if at least one frame could be built
6711 * in the switch/case block.
6712 */
6713 list_for_each_entry_safe(cf, cfbak, inlist, list) {
6714 /* header length, data length, frame length. */
6715 size_t hlen, dlen, dlen_sz, avail_room, flen;
6716
6717 if (!room)
6718 break;
6719
6720 switch (cf->type) {
6721 case QUIC_FT_CRYPTO:
6722 TRACE_DEVEL(" New CRYPTO frame build (room, len)",
6723 QUIC_EV_CONN_BCFRMS, qc, &room, len);
6724 /* Compute the length of this CRYPTO frame header */
6725 hlen = 1 + quic_int_getsize(cf->crypto.offset);
6726 /* Compute the data length of this CRyPTO frame. */
6727 dlen = max_stream_data_size(room, *len + hlen, cf->crypto.len);
6728 TRACE_DEVEL(" CRYPTO data length (hlen, crypto.len, dlen)",
6729 QUIC_EV_CONN_BCFRMS, qc, &hlen, &cf->crypto.len, &dlen);
6730 if (!dlen)
6731 continue;
6732
6733 /* CRYPTO frame length. */
6734 flen = hlen + quic_int_getsize(dlen) + dlen;
6735 TRACE_DEVEL(" CRYPTO frame length (flen)",
6736 QUIC_EV_CONN_BCFRMS, qc, &flen);
6737 /* Add the CRYPTO data length and its encoded length to the packet
6738 * length and the length of this length.
6739 */
6740 *len += flen;
6741 room -= flen;
6742 if (dlen == cf->crypto.len) {
6743 /* <cf> CRYPTO data have been consumed. */
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01006744 LIST_DEL_INIT(&cf->list);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006745 LIST_APPEND(outlist, &cf->list);
6746 }
6747 else {
6748 struct quic_frame *new_cf;
6749
Amaury Denoyelle40c24f12023-01-27 17:47:49 +01006750 new_cf = qc_frm_alloc(QUIC_FT_CRYPTO);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006751 if (!new_cf) {
6752 TRACE_ERROR("No memory for new crypto frame", QUIC_EV_CONN_BCFRMS, qc);
6753 continue;
6754 }
6755
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006756 new_cf->crypto.len = dlen;
6757 new_cf->crypto.offset = cf->crypto.offset;
6758 new_cf->crypto.qel = qel;
Ilya Shipitsin4a689da2022-10-29 09:34:32 +05006759 TRACE_DEVEL("split frame", QUIC_EV_CONN_PRSAFRM, qc, new_cf);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006760 if (cf->origin) {
6761 TRACE_DEVEL("duplicated frame", QUIC_EV_CONN_PRSAFRM, qc);
6762 /* This <cf> frame was duplicated */
6763 LIST_APPEND(&cf->origin->reflist, &new_cf->ref);
6764 new_cf->origin = cf->origin;
Frédéric Lécailledd419462023-01-26 15:07:39 +01006765 /* Detach the remaining CRYPTO frame from its original frame */
6766 LIST_DEL_INIT(&cf->ref);
6767 cf->origin = NULL;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006768 }
6769 LIST_APPEND(outlist, &new_cf->list);
6770 /* Consume <dlen> bytes of the current frame. */
6771 cf->crypto.len -= dlen;
6772 cf->crypto.offset += dlen;
6773 }
6774 break;
6775
6776 case QUIC_FT_STREAM_8 ... QUIC_FT_STREAM_F:
6777 if (cf->flags & QUIC_FL_TX_FRAME_LOST) {
6778 struct eb64_node *node = NULL;
6779 struct qc_stream_desc *stream_desc = NULL;
6780 struct quic_stream *strm = &cf->stream;
6781
6782 /* As this frame has been already lost, ensure the stream is always
6783 * available or the range of this frame is not consumed before
6784 * resending it.
6785 */
6786 node = eb64_lookup(&qc->streams_by_id, strm->id);
6787 if (!node) {
6788 TRACE_DEVEL("released stream", QUIC_EV_CONN_PRSAFRM, qc, cf);
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01006789 qc_frm_free(&cf);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006790 continue;
6791 }
6792
6793 stream_desc = eb64_entry(node, struct qc_stream_desc, by_id);
6794 if (strm->offset.key + strm->len <= stream_desc->ack_offset) {
6795 TRACE_DEVEL("ignored frame frame in already acked range",
6796 QUIC_EV_CONN_PRSAFRM, qc, cf);
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01006797 qc_frm_free(&cf);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006798 continue;
6799 }
6800 else if (strm->offset.key < stream_desc->ack_offset) {
6801 strm->offset.key = stream_desc->ack_offset;
6802 TRACE_DEVEL("updated partially acked frame",
6803 QUIC_EV_CONN_PRSAFRM, qc, cf);
6804 }
6805 }
6806 /* Note that these frames are accepted in short packets only without
6807 * "Length" packet field. Here, <*len> is used only to compute the
6808 * sum of the lengths of the already built frames for this packet.
6809 *
6810 * Compute the length of this STREAM frame "header" made a all the field
6811 * excepting the variable ones. Note that +1 is for the type of this frame.
6812 */
6813 hlen = 1 + quic_int_getsize(cf->stream.id) +
6814 ((cf->type & QUIC_STREAM_FRAME_TYPE_OFF_BIT) ? quic_int_getsize(cf->stream.offset.key) : 0);
6815 /* Compute the data length of this STREAM frame. */
6816 avail_room = room - hlen - *len;
6817 if ((ssize_t)avail_room <= 0)
6818 continue;
6819
6820 TRACE_DEVEL(" New STREAM frame build (room, len)",
6821 QUIC_EV_CONN_BCFRMS, qc, &room, len);
6822 if (cf->type & QUIC_STREAM_FRAME_TYPE_LEN_BIT) {
6823 dlen = max_available_room(avail_room, &dlen_sz);
6824 if (dlen > cf->stream.len) {
6825 dlen = cf->stream.len;
6826 }
6827 dlen_sz = quic_int_getsize(dlen);
6828 flen = hlen + dlen_sz + dlen;
6829 }
6830 else {
6831 dlen = QUIC_MIN((uint64_t)avail_room, cf->stream.len);
6832 flen = hlen + dlen;
6833 }
6834 TRACE_DEVEL(" STREAM data length (hlen, stream.len, dlen)",
6835 QUIC_EV_CONN_BCFRMS, qc, &hlen, &cf->stream.len, &dlen);
6836 TRACE_DEVEL(" STREAM frame length (flen)",
6837 QUIC_EV_CONN_BCFRMS, qc, &flen);
6838 /* Add the STREAM data length and its encoded length to the packet
6839 * length and the length of this length.
6840 */
6841 *len += flen;
6842 room -= flen;
6843 if (dlen == cf->stream.len) {
6844 /* <cf> STREAM data have been consumed. */
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01006845 LIST_DEL_INIT(&cf->list);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006846 LIST_APPEND(outlist, &cf->list);
6847
6848 /* Do not notify MUX on retransmission. */
6849 if (qc->flags & QUIC_FL_CONN_TX_MUX_CONTEXT) {
6850 qcc_streams_sent_done(cf->stream.stream->ctx,
6851 cf->stream.len,
6852 cf->stream.offset.key);
6853 }
6854 }
6855 else {
6856 struct quic_frame *new_cf;
6857 struct buffer cf_buf;
6858
Amaury Denoyelle40c24f12023-01-27 17:47:49 +01006859 new_cf = qc_frm_alloc(cf->type);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006860 if (!new_cf) {
6861 TRACE_ERROR("No memory for new STREAM frame", QUIC_EV_CONN_BCFRMS, qc);
6862 continue;
6863 }
6864
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006865 new_cf->stream.stream = cf->stream.stream;
6866 new_cf->stream.buf = cf->stream.buf;
6867 new_cf->stream.id = cf->stream.id;
Amaury Denoyelle1dac0182023-02-02 16:45:07 +01006868 new_cf->stream.offset = cf->stream.offset;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006869 new_cf->stream.len = dlen;
6870 new_cf->type |= QUIC_STREAM_FRAME_TYPE_LEN_BIT;
6871 /* FIN bit reset */
6872 new_cf->type &= ~QUIC_STREAM_FRAME_TYPE_FIN_BIT;
6873 new_cf->stream.data = cf->stream.data;
Ilya Shipitsin4a689da2022-10-29 09:34:32 +05006874 TRACE_DEVEL("split frame", QUIC_EV_CONN_PRSAFRM, qc, new_cf);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006875 if (cf->origin) {
6876 TRACE_DEVEL("duplicated frame", QUIC_EV_CONN_PRSAFRM, qc);
6877 /* This <cf> frame was duplicated */
6878 LIST_APPEND(&cf->origin->reflist, &new_cf->ref);
6879 new_cf->origin = cf->origin;
Frédéric Lécailledd419462023-01-26 15:07:39 +01006880 /* Detach this STREAM frame from its origin */
6881 LIST_DEL_INIT(&cf->ref);
6882 cf->origin = NULL;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006883 }
6884 LIST_APPEND(outlist, &new_cf->list);
6885 cf->type |= QUIC_STREAM_FRAME_TYPE_OFF_BIT;
6886 /* Consume <dlen> bytes of the current frame. */
6887 cf_buf = b_make(b_orig(cf->stream.buf),
6888 b_size(cf->stream.buf),
6889 (char *)cf->stream.data - b_orig(cf->stream.buf), 0);
6890 cf->stream.len -= dlen;
6891 cf->stream.offset.key += dlen;
6892 cf->stream.data = (unsigned char *)b_peek(&cf_buf, dlen);
6893
6894 /* Do not notify MUX on retransmission. */
6895 if (qc->flags & QUIC_FL_CONN_TX_MUX_CONTEXT) {
6896 qcc_streams_sent_done(new_cf->stream.stream->ctx,
6897 new_cf->stream.len,
6898 new_cf->stream.offset.key);
6899 }
6900 }
6901
6902 /* TODO the MUX is notified about the frame sending via
6903 * previous qcc_streams_sent_done call. However, the
6904 * sending can fail later, for example if the sendto
6905 * system call returns an error. As the MUX has been
6906 * notified, the transport layer is responsible to
6907 * bufferize and resent the announced data later.
6908 */
6909
6910 break;
6911
6912 default:
6913 flen = qc_frm_len(cf);
6914 BUG_ON(!flen);
6915 if (flen > room)
6916 continue;
6917
6918 *len += flen;
6919 room -= flen;
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01006920 LIST_DEL_INIT(&cf->list);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02006921 LIST_APPEND(outlist, &cf->list);
6922 break;
6923 }
6924
6925 /* Successful status as soon as a frame could be built */
6926 ret = 1;
6927 }
6928
6929 leave:
6930 TRACE_LEAVE(QUIC_EV_CONN_BCFRMS, qc);
6931 return ret;
6932}
6933
6934/* Generate a CONNECTION_CLOSE frame for <qc> on <qel> encryption level. <out>
6935 * is used as return parameter and should be zero'ed by the caller.
6936 */
6937static void qc_build_cc_frm(struct quic_conn *qc, struct quic_enc_level *qel,
6938 struct quic_frame *out)
6939{
6940 /* TODO improve CONNECTION_CLOSE on Initial/Handshake encryption levels
6941 *
6942 * A CONNECTION_CLOSE frame should be sent in several packets with
6943 * different encryption levels depending on the client context. This is
6944 * to ensure that the client can decrypt it. See RFC 9000 10.2.3 for
6945 * more details on how to implement it.
6946 */
6947 TRACE_ENTER(QUIC_EV_CONN_BFRM, qc);
6948
6949
6950 if (qc->err.app) {
6951 if (unlikely(qel == &qc->els[QUIC_TLS_ENC_LEVEL_INITIAL] ||
6952 qel == &qc->els[QUIC_TLS_ENC_LEVEL_HANDSHAKE])) {
6953 /* RFC 9000 10.2.3. Immediate Close during the Handshake
6954 *
6955 * Sending a CONNECTION_CLOSE of type 0x1d in an Initial or Handshake
6956 * packet could expose application state or be used to alter application
6957 * state. A CONNECTION_CLOSE of type 0x1d MUST be replaced by a
6958 * CONNECTION_CLOSE of type 0x1c when sending the frame in Initial or
6959 * Handshake packets. Otherwise, information about the application
6960 * state might be revealed. Endpoints MUST clear the value of the
6961 * Reason Phrase field and SHOULD use the APPLICATION_ERROR code when
6962 * converting to a CONNECTION_CLOSE of type 0x1c.
6963 */
6964 out->type = QUIC_FT_CONNECTION_CLOSE;
6965 out->connection_close.error_code = QC_ERR_APPLICATION_ERROR;
6966 out->connection_close.reason_phrase_len = 0;
6967 }
6968 else {
6969 out->type = QUIC_FT_CONNECTION_CLOSE_APP;
6970 out->connection_close.error_code = qc->err.code;
6971 }
6972 }
6973 else {
6974 out->type = QUIC_FT_CONNECTION_CLOSE;
6975 out->connection_close.error_code = qc->err.code;
6976 }
6977 TRACE_LEAVE(QUIC_EV_CONN_BFRM, qc);
6978
6979}
6980
6981/* This function builds a clear packet from <pkt> information (its type)
6982 * into a buffer with <pos> as position pointer and <qel> as QUIC TLS encryption
6983 * level for <conn> QUIC connection and <qel> as QUIC TLS encryption level,
6984 * filling the buffer with as much frames as possible from <frms> list of
6985 * prebuilt frames.
6986 * The trailing QUIC_TLS_TAG_LEN bytes of this packet are not built. But they are
6987 * reserved so that to ensure there is enough room to build this AEAD TAG after
6988 * having returned from this function.
6989 * This function also updates the value of <buf_pn> pointer to point to the packet
6990 * number field in this packet. <pn_len> will also have the packet number
6991 * length as value.
6992 *
6993 * Return 1 if succeeded (enough room to buile this packet), O if not.
6994 */
6995static int qc_do_build_pkt(unsigned char *pos, const unsigned char *end,
6996 size_t dglen, struct quic_tx_packet *pkt,
6997 int64_t pn, size_t *pn_len, unsigned char **buf_pn,
6998 int force_ack, int padding, int cc, int probe,
6999 struct quic_enc_level *qel, struct quic_conn *qc,
7000 const struct quic_version *ver, struct list *frms)
7001{
7002 unsigned char *beg, *payload;
7003 size_t len, len_sz, len_frms, padding_len;
7004 struct quic_frame frm = { .type = QUIC_FT_CRYPTO, };
7005 struct quic_frame ack_frm = { .type = QUIC_FT_ACK, };
7006 struct quic_frame cc_frm = { };
7007 size_t ack_frm_len, head_len;
7008 int64_t rx_largest_acked_pn;
7009 int add_ping_frm;
7010 struct list frm_list = LIST_HEAD_INIT(frm_list);
7011 struct quic_frame *cf;
7012 int must_ack, ret = 0;
7013 int nb_aepkts_since_last_ack;
7014
7015 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc);
7016
7017 /* Length field value with CRYPTO frames if present. */
7018 len_frms = 0;
7019 beg = pos;
7020 /* When not probing, and no immediate close is required, reduce the size of this
7021 * buffer to respect the congestion controller window.
7022 * This size will be limited if we have ack-eliciting frames to send from <frms>.
7023 */
7024 if (!probe && !LIST_ISEMPTY(frms) && !cc) {
7025 size_t path_room;
7026
7027 path_room = quic_path_prep_data(qc->path);
7028 if (end - beg > path_room)
7029 end = beg + path_room;
7030 }
7031
7032 /* Ensure there is enough room for the TLS encryption tag and a zero token
7033 * length field if any.
7034 */
7035 if (end - pos < QUIC_TLS_TAG_LEN +
7036 (pkt->type == QUIC_PACKET_TYPE_INITIAL ? 1 : 0))
7037 goto no_room;
7038
7039 end -= QUIC_TLS_TAG_LEN;
7040 rx_largest_acked_pn = qel->pktns->rx.largest_acked_pn;
7041 /* packet number length */
7042 *pn_len = quic_packet_number_length(pn, rx_largest_acked_pn);
7043 /* Build the header */
7044 if ((pkt->type == QUIC_PACKET_TYPE_SHORT &&
7045 !quic_build_packet_short_header(&pos, end, *pn_len, qc, qel->tls_ctx.flags)) ||
7046 (pkt->type != QUIC_PACKET_TYPE_SHORT &&
7047 !quic_build_packet_long_header(&pos, end, pkt->type, *pn_len, qc, ver)))
7048 goto no_room;
7049
7050 /* Encode the token length (0) for an Initial packet. */
7051 if (pkt->type == QUIC_PACKET_TYPE_INITIAL)
7052 *pos++ = 0;
7053 head_len = pos - beg;
7054 /* Build an ACK frame if required. */
7055 ack_frm_len = 0;
7056 nb_aepkts_since_last_ack = qel->pktns->rx.nb_aepkts_since_last_ack;
7057 must_ack = !qel->pktns->tx.pto_probe &&
7058 (force_ack || ((qel->pktns->flags & QUIC_FL_PKTNS_ACK_REQUIRED) &&
7059 (LIST_ISEMPTY(frms) || nb_aepkts_since_last_ack >= QUIC_MAX_RX_AEPKTS_SINCE_LAST_ACK)));
7060 if (must_ack) {
7061 struct quic_arngs *arngs = &qel->pktns->rx.arngs;
7062 BUG_ON(eb_is_empty(&qel->pktns->rx.arngs.root));
7063 ack_frm.tx_ack.arngs = arngs;
7064 if (qel->pktns->flags & QUIC_FL_PKTNS_NEW_LARGEST_PN) {
7065 qel->pktns->tx.ack_delay =
7066 quic_compute_ack_delay_us(qel->pktns->rx.largest_time_received, qc);
7067 qel->pktns->flags &= ~QUIC_FL_PKTNS_NEW_LARGEST_PN;
7068 }
7069 ack_frm.tx_ack.ack_delay = qel->pktns->tx.ack_delay;
7070 /* XXX BE CAREFUL XXX : here we reserved at least one byte for the
7071 * smallest frame (PING) and <*pn_len> more for the packet number. Note
7072 * that from here, we do not know if we will have to send a PING frame.
7073 * This will be decided after having computed the ack-eliciting frames
7074 * to be added to this packet.
7075 */
7076 ack_frm_len = quic_ack_frm_reduce_sz(qc, &ack_frm, end - 1 - *pn_len - pos);
7077 if (!ack_frm_len)
7078 goto no_room;
7079 }
7080
7081 /* Length field value without the ack-eliciting frames. */
7082 len = ack_frm_len + *pn_len;
7083 len_frms = 0;
7084 if (!cc && !LIST_ISEMPTY(frms)) {
7085 ssize_t room = end - pos;
7086
7087 TRACE_DEVEL("Avail. ack eliciting frames", QUIC_EV_CONN_FRMLIST, qc, frms);
7088 /* Initialize the length of the frames built below to <len>.
7089 * If any frame could be successfully built by qc_build_frms(),
7090 * we will have len_frms > len.
7091 */
7092 len_frms = len;
7093 if (!qc_build_frms(&frm_list, frms,
7094 end - pos, &len_frms, pos - beg, qel, qc)) {
7095 TRACE_DEVEL("Not enough room", QUIC_EV_CONN_TXPKT,
7096 qc, NULL, NULL, &room);
7097 if (!ack_frm_len && !qel->pktns->tx.pto_probe)
7098 goto no_room;
7099 }
7100 }
7101
7102 /* Length (of the remaining data). Must not fail because, the buffer size
7103 * has been checked above. Note that we have reserved QUIC_TLS_TAG_LEN bytes
7104 * for the encryption tag. It must be taken into an account for the length
7105 * of this packet.
7106 */
7107 if (len_frms)
7108 len = len_frms + QUIC_TLS_TAG_LEN;
7109 else
7110 len += QUIC_TLS_TAG_LEN;
7111 /* CONNECTION_CLOSE frame */
7112 if (cc) {
7113 qc_build_cc_frm(qc, qel, &cc_frm);
7114 len += qc_frm_len(&cc_frm);
7115 }
7116 add_ping_frm = 0;
7117 padding_len = 0;
7118 len_sz = quic_int_getsize(len);
7119 /* Add this packet size to <dglen> */
7120 dglen += head_len + len_sz + len;
7121 if (padding && dglen < QUIC_INITIAL_PACKET_MINLEN) {
7122 /* This is a maximum padding size */
7123 padding_len = QUIC_INITIAL_PACKET_MINLEN - dglen;
7124 /* The length field value is of this packet is <len> + <padding_len>
7125 * the size of which may be greater than the initial computed size
7126 * <len_sz>. So, let's deduce the difference between these to packet
7127 * sizes from <padding_len>.
7128 */
7129 padding_len -= quic_int_getsize(len + padding_len) - len_sz;
7130 len += padding_len;
7131 }
7132 else if (LIST_ISEMPTY(&frm_list) || len_frms == len) {
7133 if (qel->pktns->tx.pto_probe) {
7134 /* If we cannot send a frame, we send a PING frame. */
7135 add_ping_frm = 1;
7136 len += 1;
7137 }
7138 /* If there is no frame at all to follow, add at least a PADDING frame. */
7139 if (!ack_frm_len && !cc)
7140 len += padding_len = QUIC_PACKET_PN_MAXLEN - *pn_len;
7141 }
7142
7143 if (pkt->type != QUIC_PACKET_TYPE_SHORT && !quic_enc_int(&pos, end, len))
7144 goto no_room;
7145
7146 /* Packet number field address. */
7147 *buf_pn = pos;
7148
7149 /* Packet number encoding. */
7150 if (!quic_packet_number_encode(&pos, end, pn, *pn_len))
7151 goto no_room;
7152
7153 /* payload building (ack-eliciting or not frames) */
7154 payload = pos;
7155 if (ack_frm_len) {
7156 if (!qc_build_frm(&pos, end, &ack_frm, pkt, qc))
7157 goto no_room;
7158
7159 pkt->largest_acked_pn = quic_pktns_get_largest_acked_pn(qel->pktns);
7160 pkt->flags |= QUIC_FL_TX_PACKET_ACK;
7161 }
7162
7163 /* Ack-eliciting frames */
7164 if (!LIST_ISEMPTY(&frm_list)) {
7165 struct quic_frame *tmp_cf;
7166 list_for_each_entry_safe(cf, tmp_cf, &frm_list, list) {
7167 if (!qc_build_frm(&pos, end, cf, pkt, qc)) {
7168 ssize_t room = end - pos;
7169 TRACE_DEVEL("Not enough room", QUIC_EV_CONN_TXPKT,
7170 qc, NULL, NULL, &room);
7171 /* Note that <cf> was added from <frms> to <frm_list> list by
7172 * qc_build_frms().
7173 */
Amaury Denoyelle57b3eaa2023-02-02 16:12:09 +01007174 LIST_DEL_INIT(&cf->list);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007175 LIST_INSERT(frms, &cf->list);
7176 continue;
7177 }
7178
7179 quic_tx_packet_refinc(pkt);
7180 cf->pkt = pkt;
7181 }
7182 }
7183
7184 /* Build a PING frame if needed. */
7185 if (add_ping_frm) {
7186 frm.type = QUIC_FT_PING;
7187 if (!qc_build_frm(&pos, end, &frm, pkt, qc))
7188 goto no_room;
7189 }
7190
7191 /* Build a CONNECTION_CLOSE frame if needed. */
7192 if (cc) {
7193 if (!qc_build_frm(&pos, end, &cc_frm, pkt, qc))
7194 goto no_room;
7195
7196 pkt->flags |= QUIC_FL_TX_PACKET_CC;
7197 }
7198
7199 /* Build a PADDING frame if needed. */
7200 if (padding_len) {
7201 frm.type = QUIC_FT_PADDING;
7202 frm.padding.len = padding_len;
7203 if (!qc_build_frm(&pos, end, &frm, pkt, qc))
7204 goto no_room;
7205 }
7206
7207 if (pos == payload) {
7208 /* No payload was built because of congestion control */
7209 TRACE_DEVEL("limited by congestion control", QUIC_EV_CONN_TXPKT, qc);
7210 goto no_room;
7211 }
7212
7213 /* If this packet is ack-eliciting and we are probing let's
7214 * decrement the PTO probe counter.
7215 */
7216 if (pkt->flags & QUIC_FL_TX_PACKET_ACK_ELICITING &&
7217 qel->pktns->tx.pto_probe)
7218 qel->pktns->tx.pto_probe--;
7219
7220 pkt->len = pos - beg;
7221 LIST_SPLICE(&pkt->frms, &frm_list);
7222
7223 ret = 1;
7224 TRACE_DEVEL("Packet ack-eliciting frames", QUIC_EV_CONN_TXPKT, qc, pkt);
7225 leave:
7226 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc);
7227 return ret;
7228
7229 no_room:
7230 /* Replace the pre-built frames which could not be add to this packet */
7231 LIST_SPLICE(frms, &frm_list);
7232 TRACE_DEVEL("Remaining ack-eliciting frames", QUIC_EV_CONN_FRMLIST, qc, frms);
7233 goto leave;
7234}
7235
7236static inline void quic_tx_packet_init(struct quic_tx_packet *pkt, int type)
7237{
7238 pkt->type = type;
7239 pkt->len = 0;
7240 pkt->in_flight_len = 0;
7241 pkt->pn_node.key = (uint64_t)-1;
7242 LIST_INIT(&pkt->frms);
7243 pkt->time_sent = TICK_ETERNITY;
7244 pkt->next = NULL;
Frédéric Lécaille814645f2022-11-18 18:15:28 +01007245 pkt->prev = NULL;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007246 pkt->largest_acked_pn = -1;
7247 pkt->flags = 0;
7248 pkt->refcnt = 0;
7249}
7250
7251/* Build a packet into <buf> packet buffer with <pkt_type> as packet
7252 * type for <qc> QUIC connection from <qel> encryption level from <frms> list
7253 * of prebuilt frames.
7254 *
7255 * Return -2 if the packet could not be allocated or encrypted for any reason,
7256 * -1 if there was not enough room to build a packet.
7257 * XXX NOTE XXX
7258 * If you provide provide qc_build_pkt() with a big enough buffer to build a packet as big as
7259 * possible (to fill an MTU), the unique reason why this function may fail is the congestion
7260 * control window limitation.
7261 */
7262static struct quic_tx_packet *qc_build_pkt(unsigned char **pos,
7263 const unsigned char *buf_end,
7264 struct quic_enc_level *qel,
7265 struct quic_tls_ctx *tls_ctx, struct list *frms,
7266 struct quic_conn *qc, const struct quic_version *ver,
7267 size_t dglen, int pkt_type, int force_ack,
7268 int padding, int probe, int cc, int *err)
7269{
7270 struct quic_tx_packet *ret_pkt = NULL;
7271 /* The pointer to the packet number field. */
7272 unsigned char *buf_pn;
7273 unsigned char *beg, *end, *payload;
7274 int64_t pn;
7275 size_t pn_len, payload_len, aad_len;
7276 struct quic_tx_packet *pkt;
7277
7278 TRACE_ENTER(QUIC_EV_CONN_TXPKT, qc, NULL, qel);
7279 *err = 0;
7280 pkt = pool_alloc(pool_head_quic_tx_packet);
7281 if (!pkt) {
7282 TRACE_DEVEL("Not enough memory for a new packet", QUIC_EV_CONN_TXPKT, qc);
7283 *err = -2;
7284 goto err;
7285 }
7286
7287 quic_tx_packet_init(pkt, pkt_type);
7288 beg = *pos;
7289 pn_len = 0;
7290 buf_pn = NULL;
7291
7292 pn = qel->pktns->tx.next_pn + 1;
7293 if (!qc_do_build_pkt(*pos, buf_end, dglen, pkt, pn, &pn_len, &buf_pn,
7294 force_ack, padding, cc, probe, qel, qc, ver, frms)) {
7295 // trace already emitted by function above
7296 *err = -1;
7297 goto err;
7298 }
7299
7300 end = beg + pkt->len;
7301 payload = buf_pn + pn_len;
7302 payload_len = end - payload;
7303 aad_len = payload - beg;
7304
7305 if (!quic_packet_encrypt(payload, payload_len, beg, aad_len, pn, tls_ctx, qc)) {
7306 // trace already emitted by function above
7307 *err = -2;
7308 goto err;
7309 }
7310
7311 end += QUIC_TLS_TAG_LEN;
7312 pkt->len += QUIC_TLS_TAG_LEN;
7313 if (!quic_apply_header_protection(qc, beg, buf_pn, pn_len, tls_ctx)) {
7314 // trace already emitted by function above
7315 *err = -2;
7316 goto err;
7317 }
7318
7319 /* Consume a packet number */
7320 qel->pktns->tx.next_pn++;
7321 qc->tx.prep_bytes += pkt->len;
7322 if (qc->tx.prep_bytes >= 3 * qc->rx.bytes && !quic_peer_validated_addr(qc)) {
7323 qc->flags |= QUIC_FL_CONN_ANTI_AMPLIFICATION_REACHED;
7324 TRACE_PROTO("anti-amplification limit reached", QUIC_EV_CONN_TXPKT, qc);
7325 }
7326 /* Now that a correct packet is built, let us consume <*pos> buffer. */
7327 *pos = end;
7328 /* Attach the built packet to its tree. */
7329 pkt->pn_node.key = pn;
7330 /* Set the packet in fligth length for in flight packet only. */
7331 if (pkt->flags & QUIC_FL_TX_PACKET_IN_FLIGHT) {
7332 pkt->in_flight_len = pkt->len;
7333 qc->path->prep_in_flight += pkt->len;
7334 }
7335 /* Always reset this flags */
7336 qc->flags &= ~QUIC_FL_CONN_IMMEDIATE_CLOSE;
7337 if (pkt->flags & QUIC_FL_TX_PACKET_ACK) {
7338 qel->pktns->flags &= ~QUIC_FL_PKTNS_ACK_REQUIRED;
7339 qel->pktns->rx.nb_aepkts_since_last_ack = 0;
7340 }
7341
7342 pkt->pktns = qel->pktns;
7343
7344 ret_pkt = pkt;
7345 leave:
7346 TRACE_LEAVE(QUIC_EV_CONN_TXPKT, qc, ret_pkt);
7347 return ret_pkt;
7348
7349 err:
7350 /* TODO: what about the frames which have been built
7351 * for this packet.
7352 */
7353 free_quic_tx_packet(qc, pkt);
7354 goto leave;
7355}
7356
7357
7358static void __quic_conn_init(void)
7359{
7360 ha_quic_meth = BIO_meth_new(0x666, "ha QUIC methods");
7361}
7362INITCALL0(STG_REGISTER, __quic_conn_init);
7363
7364static void __quic_conn_deinit(void)
7365{
7366 BIO_meth_free(ha_quic_meth);
7367}
7368REGISTER_POST_DEINIT(__quic_conn_deinit);
7369
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007370/* Handle a new <dgram> received. Parse each QUIC packets and copied their
7371 * content to a quic-conn instance. The datagram content can be released after
7372 * this function.
7373 *
7374 * If datagram has been received on a quic-conn owned FD, <from_qc> must be set
7375 * to the connection instance. <li> is the attached listener. The caller is
7376 * responsible to ensure that the first packet is destined to this connection
7377 * by comparing CIDs.
7378 *
7379 * If datagram has been received on a receiver FD, <from_qc> will be NULL. This
7380 * function will thus retrieve the connection from the CID tree or allocate a
7381 * new one if possible. <li> is the listener attached to the receiver.
7382 *
7383 * Returns 0 on success else non-zero. If an error happens, some packets from
7384 * the datagram may not have been parsed.
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007385 */
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007386int quic_dgram_parse(struct quic_dgram *dgram, struct quic_conn *from_qc,
7387 struct listener *li)
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007388{
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007389 struct quic_rx_packet *pkt;
7390 struct quic_conn *qc = NULL;
7391 unsigned char *pos, *end;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007392 struct list *tasklist_head = NULL;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007393
7394 TRACE_ENTER(QUIC_EV_CONN_LPKT);
7395
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007396 pos = dgram->buf;
7397 end = pos + dgram->len;
7398 do {
7399 /* TODO replace zalloc -> alloc. */
7400 pkt = pool_zalloc(pool_head_quic_rx_packet);
7401 if (!pkt) {
7402 TRACE_ERROR("RX packet allocation failed", QUIC_EV_CONN_LPKT);
7403 goto err;
7404 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007405
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007406 pkt->version = NULL;
7407 pkt->pn_offset = 0;
Amaury Denoyelle98289692022-10-19 15:37:44 +02007408
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007409 /* Set flag if pkt is the first one in dgram. */
7410 if (pos == dgram->buf)
7411 pkt->flags |= QUIC_FL_RX_PACKET_DGRAM_FIRST;
Amaury Denoyelle98289692022-10-19 15:37:44 +02007412
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007413 LIST_INIT(&pkt->qc_rx_pkt_list);
7414 pkt->time_received = now_ms;
7415 quic_rx_packet_refinc(pkt);
7416 if (quic_rx_pkt_parse(pkt, pos, end, dgram, li))
7417 goto next;
Amaury Denoyelle98289692022-10-19 15:37:44 +02007418
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007419 /* Search quic-conn instance for first packet of the datagram.
7420 * quic_rx_packet_parse() is responsible to discard packets
7421 * with different DCID as the first one in the same datagram.
7422 */
7423 if (!qc) {
7424 qc = from_qc ? from_qc : quic_rx_pkt_retrieve_conn(pkt, dgram, li);
7425 /* qc is NULL if receiving a non Initial packet for an
7426 * unknown connection.
7427 */
7428 if (!qc) {
Amaury Denoyelle98289692022-10-19 15:37:44 +02007429 /* Skip the entire datagram. */
7430 pkt->len = end - pos;
7431 goto next;
7432 }
7433
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007434 dgram->qc = qc;
7435 }
Amaury Denoyelle98289692022-10-19 15:37:44 +02007436
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007437 if (qc_rx_check_closing(qc, pkt)) {
7438 /* Skip the entire datagram. */
7439 pkt->len = end - pos;
7440 goto next;
7441 }
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007442
Amaury Denoyelleeec0b3c2022-12-02 09:57:32 +01007443 /* Detect QUIC connection migration. */
Frédéric Lécaillef6769542023-01-12 10:36:26 +01007444 if (ipcmp(&qc->peer_addr, &dgram->saddr, 1)) {
Amaury Denoyelleeec0b3c2022-12-02 09:57:32 +01007445 if (qc_handle_conn_migration(qc, &dgram->saddr, &dgram->daddr)) {
7446 /* Skip the entire datagram. */
7447 TRACE_ERROR("error during connection migration, datagram dropped", QUIC_EV_CONN_LPKT, qc);
7448 pkt->len = end - pos;
7449 goto next;
7450 }
7451 }
7452
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007453 qc_rx_pkt_handle(qc, pkt, dgram, pos, &tasklist_head);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007454
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007455 next:
7456 pos += pkt->len;
7457 quic_rx_packet_refdec(pkt);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007458
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007459 /* Free rejected packets */
7460 if (!pkt->refcnt) {
7461 BUG_ON(LIST_INLIST(&pkt->qc_rx_pkt_list));
7462 pool_free(pool_head_quic_rx_packet, pkt);
7463 }
7464 } while (pos < end);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007465
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007466 /* Increasing the received bytes counter by the UDP datagram length
7467 * if this datagram could be associated to a connection.
7468 */
7469 if (dgram->qc)
7470 dgram->qc->rx.bytes += dgram->len;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007471
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007472 /* This must never happen. */
7473 BUG_ON(pos > end);
7474 BUG_ON(pos < end || pos > dgram->buf + dgram->len);
7475 /* Mark this datagram as consumed */
7476 HA_ATOMIC_STORE(&dgram->buf, NULL);
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007477
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007478 TRACE_LEAVE(QUIC_EV_CONN_LPKT);
7479 return 0;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007480
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007481 err:
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007482 TRACE_LEAVE(QUIC_EV_CONN_LPKT);
Amaury Denoyelle8687b632022-09-27 14:22:09 +02007483 return -1;
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007484}
7485
Amaury Denoyelle7c9fdd92022-11-16 11:01:02 +01007486/* Check if connection ID <dcid> of length <dcid_len> belongs to <qc> local
7487 * CIDs. This can be used to determine if a datagram is addressed to the right
7488 * connection instance.
7489 *
7490 * Returns a boolean value.
7491 */
7492int qc_check_dcid(struct quic_conn *qc, unsigned char *dcid, size_t dcid_len)
7493{
7494 struct ebmb_node *node;
7495 struct quic_connection_id *id;
7496
7497 /* For ODCID, address is concatenated to it after qc.odcid.len so this
7498 * comparison is safe.
7499 */
7500 if ((qc->scid.len == dcid_len &&
7501 memcmp(qc->scid.data, dcid, dcid_len) == 0) ||
7502 (qc->odcid.len == dcid_len &&
7503 memcmp(qc->odcid.data, dcid, dcid_len)) == 0) {
7504 return 1;
7505 }
7506
7507 node = ebmb_lookup(&quic_dghdlrs[tid].cids, dcid, dcid_len);
7508 if (node) {
7509 id = ebmb_entry(node, struct quic_connection_id, node);
7510 if (qc == id->qc)
7511 return 1;
7512 }
7513
7514 return 0;
7515}
7516
Amaury Denoyelle92fa63f2022-09-30 18:11:13 +02007517/* Retrieve the DCID from a QUIC datagram or packet with <buf> as first octet.
7518 * Returns 1 if succeeded, 0 if not.
7519 */
7520int quic_get_dgram_dcid(unsigned char *buf, const unsigned char *end,
7521 unsigned char **dcid, size_t *dcid_len)
7522{
7523 int ret = 0, long_header;
7524 size_t minlen, skip;
7525
7526 TRACE_ENTER(QUIC_EV_CONN_RXPKT);
7527
7528 if (!(*buf & QUIC_PACKET_FIXED_BIT)) {
7529 TRACE_PROTO("fixed bit not set", QUIC_EV_CONN_RXPKT);
7530 goto err;
7531 }
7532
7533 long_header = *buf & QUIC_PACKET_LONG_HEADER_BIT;
7534 minlen = long_header ? QUIC_LONG_PACKET_MINLEN :
7535 QUIC_SHORT_PACKET_MINLEN + QUIC_HAP_CID_LEN + QUIC_TLS_TAG_LEN;
7536 skip = long_header ? QUIC_LONG_PACKET_DCID_OFF : QUIC_SHORT_PACKET_DCID_OFF;
7537 if (end - buf < minlen)
7538 goto err;
7539
7540 buf += skip;
7541 *dcid_len = long_header ? *buf++ : QUIC_HAP_CID_LEN;
7542 if (*dcid_len > QUIC_CID_MAXLEN || end - buf <= *dcid_len)
7543 goto err;
7544
7545 *dcid = buf;
7546
7547 ret = 1;
7548 leave:
7549 TRACE_LEAVE(QUIC_EV_CONN_RXPKT);
7550 return ret;
7551
7552 err:
7553 TRACE_PROTO("wrong datagram", QUIC_EV_CONN_RXPKT);
7554 goto leave;
7555}
7556
7557/* Notify the MUX layer if alive about an imminent close of <qc>. */
7558void qc_notify_close(struct quic_conn *qc)
7559{
7560 TRACE_ENTER(QUIC_EV_CONN_CLOSE, qc);
7561
7562 if (qc->flags & QUIC_FL_CONN_NOTIFY_CLOSE)
7563 goto leave;
7564
7565 qc->flags |= QUIC_FL_CONN_NOTIFY_CLOSE;
7566 /* wake up the MUX */
7567 if (qc->mux_state == QC_MUX_READY && qc->conn->mux->wake) {
7568 TRACE_STATE("connection closure notidfied to mux",
7569 QUIC_FL_CONN_NOTIFY_CLOSE, qc);
7570 qc->conn->mux->wake(qc->conn);
7571 }
7572 else
7573 TRACE_STATE("connection closure not notidfied to mux",
7574 QUIC_FL_CONN_NOTIFY_CLOSE, qc);
7575 leave:
7576 TRACE_LEAVE(QUIC_EV_CONN_CLOSE, qc);
7577}
7578
7579/*
7580 * Local variables:
7581 * c-indent-level: 8
7582 * c-basic-offset: 8
7583 * End:
7584 */