Line data Source code
1 : #include "fd_quic.h"
2 : #include "fd_quic_ack_tx.h"
3 : #include "fd_quic_common.h"
4 : #include "fd_quic_conn_id.h"
5 : #include "fd_quic_enum.h"
6 : #include "fd_quic_private.h"
7 : #include "fd_quic_conn.h"
8 : #include "fd_quic_conn_map.h"
9 : #include "fd_quic_proto.h"
10 : #include "fd_quic_proto.c"
11 : #include "fd_quic_retry.h"
12 :
13 : #define FD_TEMPL_FRAME_CTX fd_quic_frame_ctx_t
14 : #include "templ/fd_quic_frame_handler_decl.h"
15 : #include "templ/fd_quic_frames_templ.h"
16 : #include "templ/fd_quic_undefs.h"
17 :
18 : #include "fd_quic_pretty_print.c"
19 :
20 : #include "crypto/fd_quic_crypto_suites.h"
21 : #include "templ/fd_quic_transport_params.h"
22 : #include "templ/fd_quic_parse_util.h"
23 : #include "tls/fd_quic_tls.h"
24 :
25 : #include <fcntl.h> /* for keylog open(2) */
26 : #include <unistd.h> /* for keylog close(2) */
27 :
28 : #include "../../ballet/hex/fd_hex.h"
29 : #include "../../tango/tempo/fd_tempo.h"
30 : #include "../../util/log/fd_dtrace.h"
31 :
32 : #include "../../disco/metrics/generated/fd_metrics_enums.h"
33 :
34 : /* Declare map type for stream_id -> stream* */
35 : #define MAP_NAME fd_quic_stream_map
36 72436425 : #define MAP_KEY stream_id
37 39276790 : #define MAP_T fd_quic_stream_map_t
38 46254334 : #define MAP_KEY_NULL FD_QUIC_STREAM_ID_UNUSED
39 31493347 : #define MAP_KEY_INVAL(key) ((key)==MAP_KEY_NULL)
40 : #define MAP_QUERY_OPT 1
41 : #include "../../util/tmpl/fd_map_dynamic.c"
42 :
43 :
44 : /* FD_QUIC_MAX_STREAMS_ALWAYS_UNLESS_ACKED */
45 : /* Defines whether a MAX_STREAMS frame is sent even if it was just */
46 : /* sent */
47 : /* They take very little space, and a dropped MAX_STREAMS frame can */
48 : /* be very consequential */
49 : /* Even when set, QUIC won't send this frame if the client has ackd */
50 : /* the most recent value */
51 12177 : # define FD_QUIC_MAX_STREAMS_ALWAYS_UNLESS_ACKED 0
52 :
53 : /* Construction API ***************************************************/
54 :
55 : FD_QUIC_API FD_FN_CONST ulong
56 4386 : fd_quic_align( void ) {
57 4386 : return FD_QUIC_ALIGN;
58 4386 : }
59 :
60 : /* fd_quic_footprint_ext returns footprint of QUIC memory region given
61 : limits. Also writes byte offsets to given layout struct. */
62 : static ulong
63 : fd_quic_footprint_ext( fd_quic_limits_t const * limits,
64 11022 : fd_quic_layout_t * layout ) {
65 11022 : memset( layout, 0, sizeof(fd_quic_layout_t) );
66 11022 : if( FD_UNLIKELY( !limits ) ) return 0UL;
67 :
68 11022 : ulong conn_cnt = limits->conn_cnt;
69 11022 : ulong conn_id_cnt = limits->conn_id_cnt;
70 11022 : ulong log_depth = limits->log_depth;
71 11022 : ulong handshake_cnt = limits->handshake_cnt;
72 11022 : ulong inflight_frame_cnt = limits->inflight_frame_cnt;
73 11022 : ulong tx_buf_sz = limits->tx_buf_sz;
74 11022 : ulong stream_pool_cnt = limits->stream_pool_cnt;
75 11022 : ulong inflight_res_cnt = limits->min_inflight_frame_cnt_conn * conn_cnt;
76 11022 : if( FD_UNLIKELY( conn_cnt ==0UL ) ) return 0UL;
77 11022 : if( FD_UNLIKELY( handshake_cnt ==0UL ) ) return 0UL;
78 11022 : if( FD_UNLIKELY( inflight_frame_cnt==0UL ) ) return 0UL;
79 :
80 11022 : if( FD_UNLIKELY( inflight_res_cnt > inflight_frame_cnt ) ) return 0UL;
81 :
82 11019 : if( FD_UNLIKELY( conn_id_cnt < FD_QUIC_MIN_CONN_ID_CNT ))
83 0 : return 0UL;
84 :
85 11019 : layout->meta_sz = sizeof(fd_quic_layout_t);
86 :
87 11019 : ulong offs = 0;
88 :
89 : /* allocate space for fd_quic_t */
90 11019 : offs += sizeof(fd_quic_t);
91 :
92 : /* allocate space for state */
93 11019 : offs = fd_ulong_align_up( offs, alignof(fd_quic_state_t) );
94 11019 : offs += sizeof(fd_quic_state_t);
95 :
96 : /* allocate space for connections */
97 11019 : offs = fd_ulong_align_up( offs, fd_quic_conn_align() );
98 11019 : layout->conns_off = offs;
99 11019 : ulong conn_footprint = fd_quic_conn_footprint( limits );
100 11019 : if( FD_UNLIKELY( !conn_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_conn_footprint" )); return 0UL; }
101 11019 : layout->conn_footprint = conn_footprint;
102 11019 : ulong conn_foot_tot = conn_cnt * conn_footprint;
103 11019 : offs += conn_foot_tot;
104 :
105 : /* allocate space for conn IDs */
106 11019 : offs = fd_ulong_align_up( offs, fd_quic_conn_map_align() );
107 11019 : layout->conn_map_off = offs;
108 11019 : ulong slot_cnt_bound = (ulong)( FD_QUIC_DEFAULT_SPARSITY * (double)conn_cnt * (double)conn_id_cnt );
109 11019 : int lg_slot_cnt = fd_ulong_find_msb( slot_cnt_bound - 1 ) + 1;
110 11019 : layout->lg_slot_cnt = lg_slot_cnt;
111 11019 : ulong conn_map_footprint = fd_quic_conn_map_footprint( lg_slot_cnt );
112 11019 : if( FD_UNLIKELY( !conn_map_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_conn_map_footprint" )); return 0UL; }
113 11019 : offs += conn_map_footprint;
114 :
115 : /* allocate space for handshake pool */
116 11019 : offs = fd_ulong_align_up( offs, fd_quic_tls_hs_pool_align() );
117 11019 : layout->hs_pool_off = offs;
118 11019 : ulong hs_pool_fp = fd_quic_tls_hs_pool_footprint( limits->handshake_cnt );
119 11019 : if( FD_UNLIKELY( !hs_pool_fp ) ) { FD_LOG_WARNING(( "invalid fd_quic_tls_hs_pool_footprint" )); return 0UL; }
120 11019 : offs += hs_pool_fp;
121 :
122 : /* allocate space for stream pool */
123 11019 : if( stream_pool_cnt && tx_buf_sz ) {
124 10974 : offs = fd_ulong_align_up( offs, fd_quic_stream_pool_align() );
125 10974 : layout->stream_pool_off = offs;
126 10974 : ulong stream_pool_footprint = fd_quic_stream_pool_footprint( stream_pool_cnt, tx_buf_sz );
127 10974 : if( FD_UNLIKELY( !stream_pool_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_stream_pool_footprint" )); return 0UL; }
128 10974 : offs += stream_pool_footprint;
129 10974 : } else {
130 45 : layout->stream_pool_off = 0UL;
131 45 : }
132 :
133 : /* allocate space for pkt_meta_pool */
134 11019 : if( inflight_frame_cnt ) {
135 11019 : offs = fd_ulong_align_up( offs, fd_quic_pkt_meta_pool_align() );
136 11019 : layout->pkt_meta_pool_off = offs;
137 11019 : ulong pkt_meta_footprint = fd_quic_pkt_meta_pool_footprint( inflight_frame_cnt );
138 11019 : if( FD_UNLIKELY( !pkt_meta_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_pkt_meta_pool_footprint" )); return 0UL; }
139 11019 : offs += pkt_meta_footprint;
140 11019 : } else {
141 0 : layout->pkt_meta_pool_off = 0UL;
142 0 : }
143 :
144 : /* allocate space for quic_log_buf */
145 11019 : offs = fd_ulong_align_up( offs, fd_quic_log_buf_align() );
146 11019 : layout->log_off = offs;
147 11019 : ulong log_footprint = fd_quic_log_buf_footprint( log_depth );
148 11019 : if( FD_UNLIKELY( !log_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_log_buf_footprint for depth %lu", log_depth )); return 0UL; }
149 11019 : offs += log_footprint;
150 :
151 11019 : return offs;
152 11019 : }
153 :
154 : FD_QUIC_API ulong
155 2181 : fd_quic_footprint( fd_quic_limits_t const * limits ) {
156 2181 : fd_quic_layout_t layout;
157 2181 : return fd_quic_footprint_ext( limits, &layout );
158 2181 : }
159 :
160 : static ulong
161 16920 : fd_quic_clock_wallclock( void * ctx FD_PARAM_UNUSED ) {
162 16920 : return (ulong)fd_log_wallclock();
163 16920 : }
164 :
165 : static ulong
166 39274792 : fd_quic_clock_tickcount( void * ctx FD_PARAM_UNUSED ) {
167 39274792 : return (ulong)fd_tickcount();
168 39274792 : }
169 :
170 : FD_QUIC_API void *
171 : fd_quic_new( void * mem,
172 2139 : fd_quic_limits_t const * limits ) {
173 :
174 : /* Argument checks */
175 :
176 2139 : if( FD_UNLIKELY( !mem ) ) {
177 0 : FD_LOG_WARNING(( "NULL mem" ));
178 0 : return NULL;
179 0 : }
180 :
181 2139 : ulong align = fd_quic_align();
182 2139 : if( FD_UNLIKELY( !fd_ulong_is_aligned( (ulong)mem, align ) ) ) {
183 0 : FD_LOG_WARNING(( "misaligned mem" ));
184 0 : return NULL;
185 0 : }
186 :
187 2139 : if( FD_UNLIKELY( !limits ) ) {
188 0 : FD_LOG_WARNING(( "NULL limits" ));
189 0 : return NULL;
190 0 : }
191 :
192 2139 : if( FD_UNLIKELY( ( limits->conn_cnt ==0UL )
193 2139 : | ( limits->conn_cnt >=UINT_MAX )
194 2139 : | ( limits->handshake_cnt ==0UL )
195 2139 : | ( limits->inflight_frame_cnt==0UL ) ) ) {
196 0 : FD_LOG_WARNING(( "invalid limits" ));
197 0 : return NULL;
198 0 : }
199 :
200 2139 : fd_quic_layout_t layout;
201 2139 : ulong footprint = fd_quic_footprint_ext( limits, &layout );
202 2139 : if( FD_UNLIKELY( !footprint ) ) {
203 0 : FD_LOG_WARNING(( "invalid footprint for config" ));
204 0 : return NULL;
205 0 : }
206 :
207 2139 : fd_quic_t * quic = (fd_quic_t *)mem;
208 :
209 : /* Clear fd_quic_t memory region */
210 2139 : fd_memset( quic, 0, footprint );
211 :
212 : /* Defaults */
213 2139 : quic->config.idle_timeout = FD_QUIC_DEFAULT_IDLE_TIMEOUT;
214 2139 : quic->config.ack_delay = FD_QUIC_DEFAULT_ACK_DELAY;
215 2139 : quic->config.retry_ttl = FD_QUIC_DEFAULT_RETRY_TTL;
216 2139 : quic->config.tls_hs_ttl = FD_QUIC_DEFAULT_TLS_HS_TTL;
217 :
218 : /* Default clock source */
219 2139 : quic->cb.now = fd_quic_clock_wallclock;
220 2139 : quic->cb.now_ctx = NULL;
221 2139 : quic->config.tick_per_us = 1000.0;
222 :
223 : /* Copy layout descriptors */
224 2139 : quic->limits = *limits;
225 2139 : quic->layout = layout;
226 :
227 : /* Init log buffer (persists across init calls) */
228 2139 : void * shmlog = (void *)( (ulong)quic + quic->layout.log_off );
229 2139 : if( FD_UNLIKELY( !fd_quic_log_buf_new( shmlog, limits->log_depth ) ) ) {
230 0 : return NULL;
231 0 : }
232 :
233 2139 : FD_COMPILER_MFENCE();
234 2139 : quic->magic = FD_QUIC_MAGIC;
235 2139 : FD_COMPILER_MFENCE();
236 :
237 2139 : return quic;
238 2139 : }
239 :
240 : FD_QUIC_API fd_quic_limits_t *
241 : fd_quic_limits_from_env( int * pargc,
242 : char *** pargv,
243 0 : fd_quic_limits_t * limits ) {
244 :
245 0 : if( FD_UNLIKELY( !limits ) ) return NULL;
246 :
247 0 : limits->conn_cnt = fd_env_strip_cmdline_ulong( pargc, pargv, "--quic-conns", "QUIC_CONN_CNT", 512UL );
248 0 : limits->conn_id_cnt = fd_env_strip_cmdline_ulong( pargc, pargv, "--quic-conn-ids", "QUIC_CONN_ID_CNT", 16UL );
249 0 : limits->stream_pool_cnt = fd_env_strip_cmdline_uint ( pargc, pargv, "--quic-streams", "QUIC_STREAM_CNT", 8UL );
250 0 : limits->handshake_cnt = fd_env_strip_cmdline_uint ( pargc, pargv, "--quic-handshakes", "QUIC_HANDSHAKE_CNT", 512UL );
251 0 : limits->inflight_frame_cnt = fd_env_strip_cmdline_ulong( pargc, pargv, "--quic-inflight-pkts", "QUIC_MAX_INFLIGHT_PKTS", 2500UL );
252 0 : limits->tx_buf_sz = fd_env_strip_cmdline_ulong( pargc, pargv, "--quic-tx-buf-sz", "QUIC_TX_BUF_SZ", 4096UL );
253 :
254 0 : return limits;
255 0 : }
256 :
257 : FD_QUIC_API fd_quic_config_t *
258 : fd_quic_config_from_env( int * pargc,
259 : char *** pargv,
260 0 : fd_quic_config_t * cfg ) {
261 :
262 0 : if( FD_UNLIKELY( !cfg ) ) return NULL;
263 :
264 0 : char const * keylog_file = fd_env_strip_cmdline_cstr ( pargc, pargv, NULL, "SSLKEYLOGFILE", NULL );
265 0 : ulong idle_timeout_ms = fd_env_strip_cmdline_ulong( pargc, pargv, "--idle-timeout", NULL, 3000UL );
266 0 : ulong initial_rx_max_stream_data = fd_env_strip_cmdline_ulong(
267 0 : pargc,
268 0 : pargv,
269 0 : "--quic-initial-rx-max-stream-data",
270 0 : "QUIC_INITIAL_RX_MAX_STREAM_DATA",
271 0 : FD_QUIC_DEFAULT_INITIAL_RX_MAX_STREAM_DATA
272 0 : );
273 0 : cfg->retry = fd_env_strip_cmdline_contains( pargc, pargv, "--quic-retry" );
274 :
275 0 : if( keylog_file ) {
276 0 : strncpy( cfg->keylog_file, keylog_file, FD_QUIC_PATH_LEN );
277 0 : } else {
278 0 : cfg->keylog_file[0]='\0';
279 0 : }
280 :
281 0 : cfg->idle_timeout = idle_timeout_ms * (ulong)1e6;
282 0 : cfg->initial_rx_max_stream_data = initial_rx_max_stream_data;
283 :
284 0 : return cfg;
285 0 : }
286 :
287 : FD_QUIC_API fd_aio_t const *
288 42 : fd_quic_get_aio_net_rx( fd_quic_t * quic ) {
289 42 : fd_aio_new( &quic->aio_rx, quic, fd_quic_aio_cb_receive );
290 42 : return &quic->aio_rx;
291 42 : }
292 :
293 : FD_QUIC_API void
294 : fd_quic_set_aio_net_tx( fd_quic_t * quic,
295 3414 : fd_aio_t const * aio_tx ) {
296 :
297 3414 : if( aio_tx ) {
298 3372 : quic->aio_tx = *aio_tx;
299 3372 : } else {
300 42 : memset( &quic->aio_tx, 0, sizeof(fd_aio_t) );
301 42 : }
302 3414 : }
303 :
304 : /* fd_quic_ticks_to_us converts ticks to microseconds
305 : fd_quic_us_to_ticks converts microseconds to ticks
306 : These should only be used after clock has been set
307 : Relies on conversion rate in config */
308 0 : FD_FN_UNUSED static ulong fd_quic_ticks_to_us( fd_quic_t * quic, ulong ticks ) {
309 0 : double ratio = quic->config.tick_per_us;
310 0 : return (ulong)( (double)ticks / ratio );
311 0 : }
312 :
313 12042 : static ulong fd_quic_us_to_ticks( fd_quic_t * quic, ulong us ) {
314 12042 : double ratio = quic->config.tick_per_us;
315 12042 : return (ulong)( (double)us * ratio );
316 12042 : }
317 :
318 : FD_QUIC_API void
319 : fd_quic_set_clock( fd_quic_t * quic,
320 : fd_quic_now_t now_fn,
321 : void * now_ctx,
322 6 : double tick_per_us ) {
323 6 : fd_quic_config_t * config = &quic->config;
324 6 : fd_quic_callbacks_t * cb = &quic->cb;
325 :
326 6 : double ratio = tick_per_us / config->tick_per_us;
327 :
328 6 : config->idle_timeout = (ulong)( ratio * (double)config->idle_timeout );
329 6 : config->ack_delay = (ulong)( ratio * (double)config->ack_delay );
330 6 : config->retry_ttl = (ulong)( ratio * (double)config->retry_ttl );
331 : /* Add more timing config here */
332 :
333 6 : config->tick_per_us = tick_per_us;
334 6 : cb->now = now_fn;
335 6 : cb->now_ctx = now_ctx;
336 6 : }
337 :
338 : FD_QUIC_API void
339 6 : fd_quic_set_clock_tickcount( fd_quic_t * quic ) {
340 : /* FIXME log warning and return error if tickcount ticks too slow or fluctuates too much */
341 6 : double tick_per_us = fd_tempo_tick_per_ns( NULL ) * 1000.0;
342 6 : fd_quic_set_clock( quic, fd_quic_clock_tickcount, NULL, tick_per_us );
343 6 : }
344 :
345 : /* initialize everything that mutates during runtime */
346 : static void
347 13088705 : fd_quic_stream_init( fd_quic_stream_t * stream ) {
348 13088705 : stream->context = NULL;
349 :
350 13088705 : stream->tx_buf.head = 0;
351 13088705 : stream->tx_buf.tail = 0;
352 13088705 : stream->tx_sent = 0;
353 :
354 13088705 : stream->stream_flags = 0;
355 : /* don't update next here, since it's still in use */
356 :
357 13088705 : stream->state = 0;
358 :
359 13088705 : stream->tx_max_stream_data = 0;
360 13088705 : stream->tx_tot_data = 0;
361 :
362 13088705 : stream->rx_tot_data = 0;
363 :
364 13088705 : stream->upd_pkt_number = 0;
365 13088705 : }
366 :
367 : FD_QUIC_API fd_quic_t *
368 2139 : fd_quic_join( void * shquic ) {
369 :
370 2139 : if( FD_UNLIKELY( !shquic ) ) {
371 0 : FD_LOG_WARNING(( "null shquic" ));
372 0 : return NULL;
373 0 : }
374 2139 : if( FD_UNLIKELY( !fd_ulong_is_aligned( (ulong)shquic, FD_QUIC_ALIGN ) ) ) {
375 0 : FD_LOG_WARNING(( "misaligned quic" ));
376 0 : return NULL;
377 0 : }
378 :
379 2139 : fd_quic_t * quic = (fd_quic_t *)shquic;
380 2139 : if( FD_UNLIKELY( quic->magic != FD_QUIC_MAGIC ) ) {
381 0 : FD_LOG_WARNING(( "bad magic" ));
382 0 : return NULL;
383 0 : }
384 :
385 2139 : return quic;
386 2139 : }
387 :
388 : FD_QUIC_API void *
389 2136 : fd_quic_leave( fd_quic_t * quic ) {
390 2136 : return (void *)quic;
391 2136 : }
392 :
393 : FD_QUIC_API fd_quic_t *
394 3378 : fd_quic_init( fd_quic_t * quic ) {
395 :
396 3378 : fd_quic_limits_t const * limits = &quic->limits;
397 3378 : fd_quic_config_t * config = &quic->config;
398 :
399 3378 : if( FD_UNLIKELY( !config->role ) ) { FD_LOG_WARNING(( "cfg.role not set" )); return NULL; }
400 3378 : if( FD_UNLIKELY( !config->idle_timeout ) ) { FD_LOG_WARNING(( "zero cfg.idle_timeout" )); return NULL; }
401 3378 : if( FD_UNLIKELY( !config->ack_delay ) ) { FD_LOG_WARNING(( "zero cfg.ack_delay" )); return NULL; }
402 3378 : if( FD_UNLIKELY( !config->retry_ttl ) ) { FD_LOG_WARNING(( "zero cfg.retry_ttl" )); return NULL; }
403 3378 : if( FD_UNLIKELY( !quic->cb.now ) ) { FD_LOG_WARNING(( "NULL cb.now" )); return NULL; }
404 3378 : if( FD_UNLIKELY( config->tick_per_us==0 ) ) { FD_LOG_WARNING(( "zero cfg.tick_per_us" )); return NULL; }
405 :
406 3378 : do {
407 3378 : ulong x = 0U;
408 111474 : for( ulong i=0UL; i<32UL; i++ ) x |= quic->config.identity_public_key[i];
409 :
410 3378 : if( FD_UNLIKELY( !x ) ) {
411 0 : FD_LOG_WARNING(( "cfg.identity_public_key not set" ));
412 0 : return NULL;
413 0 : }
414 3378 : } while(0);
415 :
416 3378 : switch( config->role ) {
417 2088 : case FD_QUIC_ROLE_SERVER:
418 3378 : case FD_QUIC_ROLE_CLIENT:
419 3378 : break;
420 0 : default:
421 0 : FD_LOG_WARNING(( "invalid cfg.role" ));
422 0 : return NULL;
423 3378 : }
424 :
425 3378 : if( FD_UNLIKELY( !config->ack_threshold ) ) {
426 6 : config->ack_threshold = FD_QUIC_DEFAULT_ACK_THRESHOLD;
427 6 : }
428 :
429 3378 : fd_quic_layout_t layout = {0};
430 3378 : if( FD_UNLIKELY( !fd_quic_footprint_ext( &quic->limits, &layout ) ) ) {
431 0 : FD_LOG_CRIT(( "fd_quic_footprint_ext failed" ));
432 0 : }
433 3378 : if( FD_UNLIKELY( 0!=memcmp( &layout, &quic->layout, sizeof(fd_quic_layout_t) ) ) ) {
434 0 : FD_LOG_HEXDUMP_WARNING(( "saved layout", &quic->layout, sizeof(fd_quic_layout_t) ));
435 0 : FD_LOG_HEXDUMP_WARNING(( "derived layout", &layout, sizeof(fd_quic_layout_t) ));
436 0 : FD_LOG_CRIT(( "fd_quic_layout changed. Memory corruption?" ));
437 0 : }
438 :
439 : /* Reset state */
440 :
441 3378 : fd_quic_state_t * state = fd_quic_get_state( quic );
442 3378 : memset( state, 0, sizeof(fd_quic_state_t) );
443 :
444 3378 : void * shmlog = (void *)( (ulong)quic + layout.log_off );
445 3378 : if( FD_UNLIKELY( !fd_quic_log_tx_join( state->log_tx, shmlog ) ) ) {
446 0 : FD_LOG_CRIT(( "fd_quic_log_tx_join failed, indicating memory corruption" ));
447 0 : }
448 :
449 : /* State: Initialize packet meta pool */
450 3378 : if( layout.pkt_meta_pool_off ) {
451 3378 : ulong pkt_meta_cnt = limits->inflight_frame_cnt;
452 3378 : ulong pkt_meta_laddr = (ulong)quic + layout.pkt_meta_pool_off;
453 3378 : fd_quic_pkt_meta_t * pkt_meta_pool = fd_quic_pkt_meta_pool_new( (void*)pkt_meta_laddr, pkt_meta_cnt );
454 3378 : state->pkt_meta_pool = fd_quic_pkt_meta_pool_join( pkt_meta_pool );
455 3378 : fd_quic_pkt_meta_ds_init_pool( pkt_meta_pool, pkt_meta_cnt );
456 3378 : }
457 :
458 : /* State: initialize each connection, and add to free list */
459 :
460 3378 : ulong conn_laddr = (ulong)quic + layout.conns_off;
461 :
462 : /* used for indexing */
463 3378 : state->conn_base = conn_laddr;
464 3378 : state->conn_sz = layout.conn_footprint;
465 :
466 : /* initialize free_conns */
467 3378 : state->free_conn_list = UINT_MAX;
468 :
469 3378 : fd_quic_conn_t * last = NULL;
470 320040 : for( ulong j = 0; j < limits->conn_cnt; ++j ) {
471 316662 : void * conn_mem = (void *)( conn_laddr );
472 316662 : conn_laddr += layout.conn_footprint;
473 :
474 316662 : fd_quic_conn_t * conn = fd_quic_conn_new( conn_mem, quic, limits );
475 316662 : if( FD_UNLIKELY( !conn ) ) {
476 0 : FD_LOG_WARNING(( "NULL conn" ));
477 0 : return NULL;
478 0 : }
479 :
480 : /* used for indexing */
481 316662 : conn->conn_idx = (uint)j;
482 :
483 316662 : conn->svc_type = UINT_MAX;
484 316662 : conn->svc_next = conn->svc_prev = UINT_MAX;
485 : /* start with minimum supported max datagram */
486 : /* peers may allow more */
487 316662 : conn->tx_max_datagram_sz = FD_QUIC_INITIAL_PAYLOAD_SZ_MAX;
488 :
489 : /* add to free list */
490 316662 : *fd_ptr_if( last!=NULL, &last->svc_next, &state->free_conn_list ) = (uint)j;
491 :
492 316662 : last = conn;
493 316662 : }
494 :
495 : /* State: Initialize conn ID map */
496 :
497 3378 : ulong conn_map_laddr = (ulong)quic + layout.conn_map_off;
498 3378 : state->conn_map = fd_quic_conn_map_join( fd_quic_conn_map_new( (void *)conn_map_laddr, layout.lg_slot_cnt ) );
499 3378 : if( FD_UNLIKELY( !state->conn_map ) ) {
500 0 : FD_LOG_WARNING(( "NULL conn_map" ));
501 0 : return NULL;
502 0 : }
503 :
504 : /* State: Initialize service queue */
505 :
506 13512 : for( uint j=0U; j<FD_QUIC_SVC_CNT; j++ ) {
507 10134 : state->svc_queue[j].head = UINT_MAX;
508 10134 : state->svc_queue[j].tail = UINT_MAX;
509 10134 : }
510 3378 : state->svc_delay[ FD_QUIC_SVC_INSTANT ] = 0UL;
511 3378 : state->svc_delay[ FD_QUIC_SVC_ACK_TX ] = quic->config.ack_delay;
512 3378 : state->svc_delay[ FD_QUIC_SVC_WAIT ] = (quic->config.idle_timeout)>>(quic->config.keep_alive);
513 :
514 : /* Check TX AIO */
515 :
516 3378 : if( FD_UNLIKELY( !quic->aio_tx.send_func ) ) {
517 0 : FD_LOG_WARNING(( "NULL aio_tx" ));
518 0 : return NULL;
519 0 : }
520 :
521 : /* State: Initialize TLS */
522 :
523 3378 : fd_quic_tls_cfg_t tls_cfg = {
524 3378 : .max_concur_handshakes = limits->handshake_cnt,
525 :
526 : /* set up callbacks */
527 3378 : .secret_cb = fd_quic_tls_cb_secret,
528 3378 : .handshake_complete_cb = fd_quic_tls_cb_handshake_complete,
529 3378 : .peer_params_cb = fd_quic_tls_cb_peer_params,
530 :
531 3378 : .signer = {
532 3378 : .ctx = config->sign_ctx,
533 3378 : .sign_fn = config->sign,
534 3378 : },
535 :
536 3378 : .cert_public_key = quic->config.identity_public_key,
537 3378 : };
538 :
539 : /* State: Initialize handshake pool */
540 :
541 3378 : if( FD_UNLIKELY( !fd_quic_tls_new( state->tls, &tls_cfg ) ) ) {
542 0 : FD_DEBUG( FD_LOG_WARNING( ( "fd_quic_tls_new failed" ) ) );
543 0 : return NULL;
544 0 : }
545 :
546 3378 : ulong hs_pool_laddr = (ulong)quic + layout.hs_pool_off;
547 3378 : fd_quic_tls_hs_t * hs_pool = fd_quic_tls_hs_pool_join( fd_quic_tls_hs_pool_new( (void *)hs_pool_laddr, limits->handshake_cnt ) );
548 3378 : if( FD_UNLIKELY( !hs_pool ) ) {
549 0 : FD_LOG_WARNING(( "fd_quic_tls_hs_pool_new failed" ));
550 0 : return NULL;
551 0 : }
552 3378 : state->hs_pool = hs_pool;
553 :
554 : /* State: Initialize TLS handshake cache */
555 3378 : if( FD_LIKELY( !fd_quic_tls_hs_cache_join(
556 3378 : fd_quic_tls_hs_cache_new( &state->hs_cache )
557 3378 : ))) {
558 0 : FD_LOG_WARNING(( "fd_quic_tls_hs_cache_new failed" ));
559 0 : return NULL;
560 0 : }
561 :
562 :
563 3378 : if( layout.stream_pool_off ) {
564 3369 : ulong stream_pool_cnt = limits->stream_pool_cnt;
565 3369 : ulong tx_buf_sz = limits->tx_buf_sz;
566 3369 : ulong stream_pool_laddr = (ulong)quic + layout.stream_pool_off;
567 3369 : state->stream_pool = fd_quic_stream_pool_new( (void*)stream_pool_laddr, stream_pool_cnt, tx_buf_sz );
568 3369 : }
569 :
570 : /* generate a secure random number as seed for fd_rng */
571 3378 : uint rng_seed = 0;
572 3378 : int rng_seed_ok = !!fd_rng_secure( &rng_seed, sizeof(rng_seed) );
573 3378 : if( FD_UNLIKELY( !rng_seed_ok ) ) {
574 0 : FD_LOG_ERR(( "fd_rng_secure failed" ));
575 0 : }
576 3378 : fd_rng_new( state->_rng, rng_seed, 0UL );
577 :
578 : /* use rng to generate secret bytes for future RETRY token generation */
579 3378 : int rng1_ok = !!fd_rng_secure( state->retry_secret, FD_QUIC_RETRY_SECRET_SZ );
580 3378 : int rng2_ok = !!fd_rng_secure( state->retry_iv, FD_QUIC_RETRY_IV_SZ );
581 3378 : if( FD_UNLIKELY( !rng1_ok || !rng2_ok ) ) {
582 0 : FD_LOG_ERR(( "fd_rng_secure failed" ));
583 0 : return NULL;
584 0 : }
585 :
586 : /* Initialize transport params */
587 :
588 3378 : fd_quic_transport_params_t * tp = &state->transport_params;
589 :
590 : /* initial max streams is zero */
591 : /* we will send max_streams and max_data frames later to allow the peer to */
592 : /* send us data */
593 3378 : ulong initial_max_streams_uni = quic->config.role==FD_QUIC_ROLE_SERVER ? 1UL<<60 : 0;
594 3378 : ulong initial_max_stream_data = config->initial_rx_max_stream_data;
595 :
596 3378 : double tick_per_ns = (double)quic->config.tick_per_us / 1e3;
597 :
598 3378 : double max_ack_delay_ticks = (double)(config->ack_delay * 2UL);
599 3378 : double max_ack_delay_ns = max_ack_delay_ticks / tick_per_ns;
600 3378 : double max_ack_delay_ms = max_ack_delay_ns / 1e6;
601 3378 : ulong max_ack_delay_ms_u = (ulong)round( max_ack_delay_ms );
602 :
603 3378 : double idle_timeout_ns = (double)config->idle_timeout / tick_per_ns;
604 3378 : double idle_timeout_ms = idle_timeout_ns / 1e6;
605 3378 : ulong idle_timeout_ms_u = (ulong)round( idle_timeout_ms );
606 :
607 3378 : memset( tp, 0, sizeof(fd_quic_transport_params_t) );
608 3378 : FD_QUIC_TRANSPORT_PARAM_SET( tp, max_idle_timeout_ms, idle_timeout_ms_u );
609 3378 : FD_QUIC_TRANSPORT_PARAM_SET( tp, max_udp_payload_size, FD_QUIC_MAX_PAYLOAD_SZ ); /* TODO */
610 3378 : FD_QUIC_TRANSPORT_PARAM_SET( tp, initial_max_data, (1UL<<62)-1UL );
611 3378 : FD_QUIC_TRANSPORT_PARAM_SET( tp, initial_max_stream_data_uni, initial_max_stream_data );
612 3378 : FD_QUIC_TRANSPORT_PARAM_SET( tp, initial_max_streams_bidi, 0 );
613 3378 : FD_QUIC_TRANSPORT_PARAM_SET( tp, initial_max_streams_uni, initial_max_streams_uni );
614 3378 : FD_QUIC_TRANSPORT_PARAM_SET( tp, ack_delay_exponent, 0 );
615 3378 : FD_QUIC_TRANSPORT_PARAM_SET( tp, max_ack_delay, max_ack_delay_ms_u );
616 3378 : /* */tp->disable_active_migration_present = 1;
617 :
618 : /* Compute max inflight pkt cnt per conn */
619 3378 : state->max_inflight_frame_cnt_conn = limits->inflight_frame_cnt - limits->min_inflight_frame_cnt_conn * (limits->conn_cnt-1);
620 :
621 3378 : return quic;
622 3378 : }
623 :
624 : /* fd_quic_enc_level_to_pn_space maps of encryption level in [0,4) to
625 : packet number space. */
626 : static uint
627 40115970 : fd_quic_enc_level_to_pn_space( uint enc_level ) {
628 : /* TODO improve this map */
629 40115970 : static uchar const el2pn_map[] = { 0, 2, 1, 2 };
630 :
631 40115970 : if( FD_UNLIKELY( enc_level >= 4U ) )
632 0 : FD_LOG_ERR(( "fd_quic_enc_level_to_pn_space called with invalid enc_level" ));
633 :
634 40115970 : return el2pn_map[ enc_level ];
635 40115970 : }
636 :
637 : /* This code is directly from rfc9000 A.3 */
638 : FD_FN_CONST ulong
639 : fd_quic_reconstruct_pkt_num( ulong pktnum_comp,
640 : ulong pktnum_sz,
641 13371846 : ulong exp_pkt_number ) {
642 13371846 : ulong pn_nbits = pktnum_sz << 3u;
643 13371846 : ulong pn_win = 1ul << pn_nbits;
644 13371846 : ulong pn_hwin = pn_win >> 1ul;
645 13371846 : ulong pn_mask = pn_win - 1ul;
646 : // The incoming packet number should be greater than
647 : // exp_pkt_number - pn_hwin and less than or equal to
648 : // exp_pkt_number + pn_hwin
649 : //
650 : // This means we cannot just strip the trailing bits from
651 : // exp_pkt_number and add the truncated_pn because that might
652 : // yield a value outside the window.
653 : //
654 : // The following code calculates a candidate value and
655 : // makes sure it's within the packet number window.
656 : // Note the extra checks to prevent overflow and underflow.
657 13371846 : ulong candidate_pn = ( exp_pkt_number & ~pn_mask ) | pktnum_comp;
658 13371846 : if( candidate_pn + pn_hwin <= exp_pkt_number &&
659 13371846 : candidate_pn + pn_win < ( 1ul << 62ul ) ) {
660 0 : return candidate_pn + pn_win;
661 0 : }
662 :
663 13371846 : if( candidate_pn > exp_pkt_number + pn_hwin &&
664 13371846 : candidate_pn >= pn_win ) {
665 0 : return candidate_pn - pn_win;
666 0 : }
667 :
668 13371846 : return candidate_pn;
669 13371846 : }
670 :
671 : static void
672 : fd_quic_svc_unqueue( fd_quic_state_t * state,
673 14143127 : fd_quic_conn_t * conn ) {
674 :
675 14143127 : fd_quic_svc_queue_t * queue = &state->svc_queue[ conn->svc_type ];
676 14143127 : uint prev_idx = conn->svc_prev;
677 14143127 : uint next_idx = conn->svc_next;
678 14143127 : fd_quic_conn_t * prev_ele = fd_quic_conn_at_idx( state, prev_idx );
679 14143127 : fd_quic_conn_t * next_ele = fd_quic_conn_at_idx( state, next_idx );
680 :
681 14143127 : *fd_ptr_if( next_idx!=UINT_MAX, &next_ele->svc_prev, &queue->head ) = prev_idx;
682 14143127 : *fd_ptr_if( prev_idx!=UINT_MAX, &prev_ele->svc_next, &queue->tail ) = next_idx;
683 :
684 14143127 : }
685 :
686 : void
687 : fd_quic_svc_schedule( fd_quic_state_t * state,
688 : fd_quic_conn_t * conn,
689 136456338 : uint svc_type ) {
690 136456338 : if( FD_UNLIKELY( svc_type >= FD_QUIC_SVC_CNT ) ) {
691 0 : FD_LOG_ERR(( "fd_quic_svc_schedule called with invalid svc_type (%u)", svc_type ));
692 0 : }
693 136456338 : if( FD_UNLIKELY( conn->state == FD_QUIC_CONN_STATE_INVALID ) ) {
694 0 : FD_LOG_ERR(( "fd_quic_svc_schedule called with invalid conn" ));
695 0 : }
696 :
697 136456338 : int is_queued = conn->svc_type < FD_QUIC_SVC_CNT;
698 136456338 : long cur_delay = (long)conn->svc_time - (long)state->now;
699 136456338 : long tgt_delay = (long)state->svc_delay[ svc_type ];
700 :
701 : /* Don't reschedule if already scheduled sooner */
702 136456338 : if( is_queued && cur_delay<=tgt_delay ) return;
703 :
704 : /* Remove entry from current queue */
705 27800042 : if( is_queued ) {
706 14130968 : fd_quic_svc_unqueue( state, conn );
707 14130968 : is_queued = 0;
708 14130968 : }
709 :
710 : /* Add into new queue */
711 27800042 : fd_quic_svc_queue_t * queue = &state->svc_queue[ svc_type ];
712 27800042 : uint old_tail_idx = queue->tail;
713 27800042 : fd_quic_conn_t * old_tail_ele = fd_quic_conn_at_idx( state, old_tail_idx );
714 27800042 : conn->svc_type = svc_type;
715 27800042 : conn->svc_time = state->now + (ulong)tgt_delay;
716 27800042 : conn->svc_prev = UINT_MAX;
717 27800042 : conn->svc_next = old_tail_idx;
718 27800042 : *fd_ptr_if( old_tail_idx!=UINT_MAX, &old_tail_ele->svc_prev, &queue->head ) = (uint)conn->conn_idx;
719 27800042 : queue->tail = (uint)conn->conn_idx;
720 :
721 27800042 : }
722 :
723 : /* fd_quic_svc_queue_validate checks the following:
724 : - dlist prev and next chains are in agreement
725 : - all nodes belong to the same list
726 : - no cycles in list
727 : - no excessive delays (assumes no monotonically increasing timestamp) */
728 :
729 : static void
730 : fd_quic_svc_queue_validate( fd_quic_t * quic,
731 135 : uint svc_type ) {
732 135 : FD_TEST( svc_type < FD_QUIC_SVC_CNT );
733 135 : fd_quic_state_t * state = fd_quic_get_state( quic );
734 135 : ulong now = state->now;
735 :
736 135 : ulong cnt = 0UL;
737 135 : uint prev = UINT_MAX;
738 135 : uint node = state->svc_queue[ svc_type ].tail;
739 300165 : while( node!=UINT_MAX ) {
740 300030 : FD_TEST( node <= quic->limits.conn_cnt );
741 300030 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, node );
742 300030 : FD_TEST( conn->state != FD_QUIC_CONN_STATE_INVALID );
743 300030 : FD_TEST( conn->svc_type == svc_type );
744 300030 : FD_TEST( conn->svc_time <= now + state->svc_delay[ svc_type ] );
745 300030 : FD_TEST( conn->svc_prev == prev );
746 300030 : conn->visited = 1U;
747 :
748 300030 : prev = node;
749 300030 : node = conn->svc_next;
750 300030 : cnt++;
751 300030 : FD_TEST( cnt <= quic->limits.conn_cnt );
752 :
753 300030 : }
754 135 : FD_TEST( prev == state->svc_queue[ svc_type ].head );
755 135 : }
756 :
757 : /* validates the free conn list doesn't cycle, point nowhere, leak, or point to live conn */
758 : static void
759 45 : fd_quic_conn_free_validate( fd_quic_t * quic ) {
760 45 : fd_quic_state_t * state = fd_quic_get_state( quic );
761 45 : ulong cnt = 0UL;
762 45 : uint node = state->free_conn_list;
763 363 : while( node!=UINT_MAX ) {
764 318 : FD_TEST( node <= quic->limits.conn_cnt );
765 318 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, node );
766 318 : FD_TEST( conn->state == FD_QUIC_CONN_STATE_INVALID );
767 318 : FD_TEST( conn->svc_prev == UINT_MAX );
768 318 : FD_TEST( conn->svc_type == UINT_MAX );
769 318 : conn->visited = 1U;
770 318 : node = conn->svc_next;
771 318 : cnt++;
772 318 : FD_TEST( cnt <= quic->limits.conn_cnt );
773 318 : }
774 45 : }
775 :
776 : void
777 45 : fd_quic_svc_validate( fd_quic_t * quic ) {
778 45 : fd_quic_state_t * state = fd_quic_get_state( quic );
779 300393 : for( ulong j=0UL; j < quic->limits.conn_cnt; j++ ) {
780 300348 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, j );
781 300348 : FD_TEST( conn->conn_idx==j );
782 300348 : conn->visited = 0U;
783 300348 : if( conn->state == FD_QUIC_CONN_STATE_INVALID ) {
784 318 : FD_TEST( conn->svc_type==UINT_MAX );
785 318 : FD_TEST( conn->svc_prev==UINT_MAX );
786 318 : continue;
787 318 : }
788 300348 : }
789 :
790 45 : fd_quic_svc_queue_validate( quic, FD_QUIC_SVC_INSTANT );
791 45 : fd_quic_svc_queue_validate( quic, FD_QUIC_SVC_ACK_TX );
792 45 : fd_quic_svc_queue_validate( quic, FD_QUIC_SVC_WAIT );
793 :
794 300393 : for( ulong j=0UL; j < quic->limits.conn_cnt; j++ ) {
795 300348 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, j );
796 300348 : FD_TEST( conn->conn_idx==j );
797 300348 : if( conn->state == FD_QUIC_CONN_STATE_INVALID ) {
798 318 : FD_TEST( conn->svc_type==UINT_MAX );
799 318 : FD_TEST( conn->svc_prev==UINT_MAX );
800 318 : FD_TEST( !conn->visited );
801 318 : continue;
802 318 : }
803 300030 : FD_TEST( conn->visited ); /* if assertion fails, the conn was leaked */
804 300030 : }
805 :
806 45 : fd_quic_conn_free_validate( quic );
807 300393 : for( ulong j=0UL; j < quic->limits.conn_cnt; j++ ) {
808 300348 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, j );
809 300348 : FD_TEST( conn->conn_idx==j );
810 300348 : FD_TEST( conn->visited );
811 300348 : }
812 45 : }
813 :
814 : /* Helpers for generating fd_quic_log entries */
815 :
816 : static fd_quic_log_hdr_t
817 0 : fd_quic_log_conn_hdr( fd_quic_conn_t const * conn ) {
818 0 : fd_quic_log_hdr_t hdr = {
819 0 : .conn_id = conn->our_conn_id,
820 0 : .flags = 0
821 0 : };
822 0 : return hdr;
823 0 : }
824 :
825 : static fd_quic_log_hdr_t
826 : fd_quic_log_full_hdr( fd_quic_conn_t const * conn,
827 90 : fd_quic_pkt_t const * pkt ) {
828 90 : fd_quic_log_hdr_t hdr = {
829 90 : .conn_id = conn->our_conn_id,
830 90 : .pkt_num = pkt->pkt_number,
831 90 : .ip4_saddr = pkt->ip4->saddr,
832 90 : .udp_sport = pkt->udp->net_sport,
833 90 : .enc_level = (uchar)pkt->enc_level,
834 90 : .flags = 0
835 90 : };
836 90 : return hdr;
837 90 : }
838 :
839 : /* fd_quic_conn_error sets the connection state to aborted. This does
840 : not destroy the connection object. Rather, it will eventually cause
841 : the connection to be freed during a later fd_quic_service call.
842 : reason is an RFC 9000 QUIC error code. error_line is the source line
843 : of code in fd_quic.c */
844 :
845 : static void
846 : fd_quic_conn_error1( fd_quic_conn_t * conn,
847 90 : uint reason ) {
848 90 : if( FD_UNLIKELY( !conn || conn->state == FD_QUIC_CONN_STATE_DEAD ) ) return;
849 :
850 90 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_ABORT );
851 90 : conn->reason = reason;
852 :
853 : /* set connection to be serviced ASAP */
854 90 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
855 90 : }
856 :
857 : static void
858 : fd_quic_conn_error( fd_quic_conn_t * conn,
859 : uint reason,
860 0 : uint error_line ) {
861 0 : fd_quic_conn_error1( conn, reason );
862 :
863 0 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
864 :
865 0 : ulong sig = fd_quic_log_sig( FD_QUIC_EVENT_CONN_QUIC_CLOSE );
866 0 : fd_quic_log_error_t * frame = fd_quic_log_tx_prepare( state->log_tx );
867 0 : *frame = (fd_quic_log_error_t) {
868 0 : .hdr = fd_quic_log_conn_hdr( conn ),
869 0 : .code = { reason, 0UL },
870 0 : .src_file = "fd_quic.c",
871 0 : .src_line = error_line,
872 0 : };
873 0 : fd_quic_log_tx_submit( state->log_tx, sizeof(fd_quic_log_error_t), sig, (long)state->now );
874 0 : }
875 :
876 : static void
877 : fd_quic_frame_error( fd_quic_frame_ctx_t const * ctx,
878 : uint reason,
879 90 : uint error_line ) {
880 90 : fd_quic_t * quic = ctx->quic;
881 90 : fd_quic_conn_t * conn = ctx->conn;
882 90 : fd_quic_pkt_t const * pkt = ctx->pkt;
883 90 : fd_quic_state_t * state = fd_quic_get_state( quic );
884 :
885 90 : fd_quic_conn_error1( conn, reason );
886 :
887 90 : uint tls_reason = 0U;
888 90 : if( conn->tls_hs ) tls_reason = conn->tls_hs->hs.base.reason;
889 :
890 90 : ulong sig = fd_quic_log_sig( FD_QUIC_EVENT_CONN_QUIC_CLOSE );
891 90 : fd_quic_log_error_t * frame = fd_quic_log_tx_prepare( state->log_tx );
892 90 : *frame = (fd_quic_log_error_t) {
893 90 : .hdr = fd_quic_log_full_hdr( conn, pkt ),
894 90 : .code = { reason, tls_reason },
895 90 : .src_file = "fd_quic.c",
896 90 : .src_line = error_line,
897 90 : };
898 90 : fd_quic_log_tx_submit( state->log_tx, sizeof(fd_quic_log_error_t), sig, (long)state->now );
899 90 : }
900 :
901 : /* returns the encoding level we should use for the next tx quic packet
902 : or all 1's if nothing to tx */
903 : static uint
904 26732358 : fd_quic_tx_enc_level( fd_quic_conn_t * conn, int acks ) {
905 26732358 : uint enc_level = ~0u;
906 :
907 26732358 : uint app_pn_space = fd_quic_enc_level_to_pn_space( fd_quic_enc_level_appdata_id );
908 26732358 : ulong app_pkt_number = conn->pkt_number[app_pn_space];
909 :
910 : /* fd_quic_tx_enc_level( ... )
911 : check status - if closing, set based on handshake complete
912 : check for acks
913 : find lowest enc level
914 : check for hs_data
915 : find lowest enc level
916 : if any, use lowest
917 : else
918 : if stream data, use 1-rtt
919 : else
920 : nothing to do */
921 :
922 : /* check status */
923 26732358 : switch( conn->state ) {
924 0 : case FD_QUIC_CONN_STATE_DEAD:
925 : /* do not send on dead connection at all */
926 0 : return ~0u;
927 :
928 168 : case FD_QUIC_CONN_STATE_ABORT:
929 12216 : case FD_QUIC_CONN_STATE_CLOSE_PENDING:
930 : /* use handshake or app enc level depending on handshake complete */
931 12216 : if( !(conn->flags & FD_QUIC_CONN_FLAGS_CLOSE_SENT ) ) {
932 6108 : if( conn->handshake_complete ) {
933 6024 : return fd_quic_enc_level_appdata_id;
934 6024 : } else if( fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_handshake_id ) ) {
935 0 : return fd_quic_enc_level_handshake_id;
936 84 : } else if( fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_initial_id ) ) {
937 84 : return fd_quic_enc_level_initial_id;
938 84 : }
939 6108 : }
940 6108 : return ~0u;
941 :
942 : /* TODO consider this optimization... but we want to ack all handshakes, even if there is stream_data */
943 26665989 : case FD_QUIC_CONN_STATE_ACTIVE:
944 26665989 : if( FD_LIKELY( !conn->tls_hs ) ) {
945 : /* optimization for case where we have stream data to send */
946 :
947 : /* find stream data to send */
948 26653947 : fd_quic_stream_t * sentinel = conn->send_streams;
949 26653947 : fd_quic_stream_t * stream = sentinel->next;
950 26653947 : if( !stream->sentinel && stream->upd_pkt_number >= app_pkt_number ) {
951 13099583 : return fd_quic_enc_level_appdata_id;
952 13099583 : }
953 26653947 : }
954 26732358 : }
955 :
956 : /* pick enc_level of oldest ACK not yet sent */
957 13620559 : fd_quic_ack_gen_t * ack_gen = conn->ack_gen;
958 13620559 : fd_quic_ack_t const * oldest_ack = fd_quic_ack_queue_ele( ack_gen, ack_gen->tail );
959 13620559 : uint ack_enc_level = oldest_ack->enc_level; /* speculative load (might be invalid) */
960 13620559 : if( ack_gen->head != ack_gen->tail && acks ) {
961 253774 : return ack_enc_level;
962 253774 : }
963 :
964 : /* Check for handshake data to send */
965 13366785 : uint peer_enc_level = conn->peer_enc_level;
966 13366785 : if( FD_UNLIKELY( conn->tls_hs ) ) {
967 42153 : fd_quic_tls_hs_data_t * hs_data = NULL;
968 :
969 138501 : for( uint i = peer_enc_level; i < 4 && i < enc_level; ++i ) {
970 108393 : if( enc_level == ~0u || enc_level == i ) {
971 108393 : hs_data = fd_quic_tls_get_hs_data( conn->tls_hs, i );
972 108393 : if( hs_data ) {
973 : /* offset within stream */
974 42153 : ulong offset = conn->hs_sent_bytes[i];
975 : /* skip packets we've sent */
976 90324 : while( hs_data && hs_data->offset + hs_data->data_sz <= offset ) {
977 48171 : hs_data = fd_quic_tls_get_next_hs_data( conn->tls_hs, hs_data );
978 48171 : }
979 42153 : if( hs_data ) {
980 12045 : enc_level = i;
981 12045 : return enc_level;
982 12045 : }
983 42153 : }
984 108393 : }
985 108393 : }
986 42153 : }
987 :
988 : /* if we have acks to send or handshake data, then use that enc_level */
989 13354740 : if( enc_level != ~0u ) return enc_level;
990 :
991 : /* handshake done? */
992 13354740 : if( FD_UNLIKELY( conn->handshake_done_send ) ) return fd_quic_enc_level_appdata_id;
993 :
994 : /* find stream data to send */
995 13348719 : fd_quic_stream_t * sentinel = conn->send_streams;
996 13348719 : fd_quic_stream_t * stream = sentinel->next;
997 13348719 : if( !stream->sentinel && stream->upd_pkt_number >= app_pkt_number ) {
998 0 : return fd_quic_enc_level_appdata_id;
999 0 : }
1000 :
1001 13348719 : if( conn->flags && conn->upd_pkt_number >= app_pkt_number ) {
1002 6081 : return fd_quic_enc_level_appdata_id;
1003 6081 : }
1004 :
1005 : /* nothing to send */
1006 13342638 : return ~0u;
1007 13348719 : }
1008 :
1009 : /* Include frame code generator */
1010 :
1011 : #include "templ/fd_quic_frame.c"
1012 :
1013 : /* handle single v1 frames */
1014 : /* returns bytes consumed */
1015 : ulong
1016 : fd_quic_handle_v1_frame( fd_quic_t * quic,
1017 : fd_quic_conn_t * conn,
1018 : fd_quic_pkt_t * pkt,
1019 : uint pkt_type,
1020 : uchar const * buf,
1021 83408883 : ulong buf_sz ) {
1022 83408883 : if( conn->state == FD_QUIC_CONN_STATE_DEAD ) return FD_QUIC_PARSE_FAIL;
1023 83408883 : if( FD_UNLIKELY( buf_sz<1UL ) ) return FD_QUIC_PARSE_FAIL;
1024 :
1025 : /* Frame ID is technically a varint but it's sufficient to look at the
1026 : first byte. */
1027 83408883 : uint id = buf[0];
1028 :
1029 83408883 : FD_DTRACE_PROBE_4( quic_handle_frame, id, conn->our_conn_id, pkt_type, pkt->pkt_number );
1030 :
1031 83408883 : fd_quic_frame_ctx_t frame_context[1] = {{ quic, conn, pkt }};
1032 83408883 : if( FD_UNLIKELY( !fd_quic_frame_type_allowed( pkt_type, id ) ) ) {
1033 51 : FD_DTRACE_PROBE_4( quic_err_frame_not_allowed, id, conn->our_conn_id, pkt_type, pkt->pkt_number );
1034 51 : fd_quic_frame_error( frame_context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
1035 51 : return FD_QUIC_PARSE_FAIL;
1036 51 : }
1037 83408832 : quic->metrics.frame_rx_cnt[ fd_quic_frame_metric_id[ id ] ]++;
1038 :
1039 83408832 : pkt->ack_flag |= fd_uint_if( fd_quic_frame_type_flags[ id ]&FD_QUIC_FRAME_FLAG_N, 0U, ACK_FLAG_RQD );
1040 :
1041 : /* tail call to frame handler */
1042 83408832 : switch( id ) {
1043 :
1044 0 : # define F(T,MID,NAME,...) \
1045 83408832 : case T: return fd_quic_interpret_##NAME##_frame( frame_context, buf, buf_sz );
1046 83408832 : FD_QUIC_FRAME_TYPES(F)
1047 0 : # undef F
1048 :
1049 0 : default:
1050 : /* FIXME this should be unreachable, but gracefully handle this case as defense-in-depth */
1051 : /* unknown frame types are PROTOCOL_VIOLATION errors */
1052 0 : FD_DEBUG( FD_LOG_DEBUG(( "unexpected frame type: %u", id )); )
1053 0 : fd_quic_frame_error( frame_context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
1054 0 : return FD_QUIC_PARSE_FAIL;
1055 83408832 : }
1056 :
1057 83408832 : }
1058 :
1059 : fd_quic_t *
1060 3324 : fd_quic_fini( fd_quic_t * quic ) {
1061 :
1062 3324 : if( FD_UNLIKELY( !quic ) ) {
1063 0 : FD_LOG_WARNING(("NULL quic"));
1064 0 : return NULL;
1065 0 : }
1066 :
1067 : /* Derive memory layout */
1068 :
1069 3324 : fd_quic_layout_t layout = {0};
1070 3324 : fd_quic_footprint_ext( &quic->limits, &layout );
1071 :
1072 3324 : fd_quic_state_t * state = fd_quic_get_state( quic );
1073 :
1074 : /* Free conns */
1075 :
1076 3324 : ulong conn_laddr = (ulong)quic + layout.conns_off;
1077 19758 : for( ulong i=0; i < quic->limits.conn_cnt; i++ ) {
1078 16434 : fd_quic_conn_t * conn = (fd_quic_conn_t *)( conn_laddr );
1079 16434 : conn_laddr += layout.conn_footprint;
1080 :
1081 16434 : if( conn->state ) fd_quic_conn_free( quic, conn );
1082 16434 : }
1083 :
1084 : /* Deinit TLS */
1085 :
1086 3324 : fd_quic_tls_hs_pool_delete( fd_quic_tls_hs_pool_leave( state->hs_pool ) ); state->hs_pool = NULL;
1087 3324 : fd_quic_tls_delete( state->tls );
1088 3324 : fd_quic_tls_hs_cache_delete( fd_quic_tls_hs_cache_leave( &state->hs_cache ) );
1089 :
1090 :
1091 : /* Delete conn ID map */
1092 :
1093 3324 : fd_quic_conn_map_delete( fd_quic_conn_map_leave( state->conn_map ) );
1094 3324 : state->conn_map = NULL;
1095 :
1096 : /* Clear join-lifetime memory regions */
1097 :
1098 3324 : memset( state, 0, sizeof(fd_quic_state_t) );
1099 :
1100 3324 : return quic;
1101 3324 : }
1102 :
1103 : void *
1104 2136 : fd_quic_delete( fd_quic_t * quic ) {
1105 :
1106 2136 : if( FD_UNLIKELY( !quic ) ) {
1107 0 : FD_LOG_WARNING(( "NULL quic" ));
1108 0 : return NULL;
1109 0 : }
1110 :
1111 2136 : if( FD_UNLIKELY( !fd_ulong_is_aligned( (ulong)quic, fd_quic_align() ) ) ) {
1112 0 : FD_LOG_WARNING(( "misaligned quic" ));
1113 0 : return NULL;
1114 0 : }
1115 :
1116 2136 : if( FD_UNLIKELY( quic->magic!=FD_QUIC_MAGIC ) ) {
1117 0 : FD_LOG_WARNING(( "bad magic" ));
1118 0 : return NULL;
1119 0 : }
1120 :
1121 2136 : void * shmlog = (void *)( (ulong)quic + quic->layout.log_off );
1122 2136 : if( FD_UNLIKELY( !fd_quic_log_buf_delete( shmlog ) ) ) {
1123 0 : FD_LOG_WARNING(( "fd_quic_log_buf_delete failed" ));
1124 0 : return NULL;
1125 0 : }
1126 :
1127 2136 : FD_COMPILER_MFENCE();
1128 2136 : FD_VOLATILE( quic->magic ) = 0UL;
1129 2136 : FD_COMPILER_MFENCE();
1130 :
1131 2136 : return (void *)quic;
1132 2136 : }
1133 :
1134 : fd_quic_stream_t *
1135 13088705 : fd_quic_conn_new_stream( fd_quic_conn_t * conn ) {
1136 13088705 : if( FD_UNLIKELY( !conn->stream_map ) ) {
1137 : /* QUIC config is receive-only */
1138 0 : return NULL;
1139 0 : }
1140 :
1141 13088705 : fd_quic_t * quic = conn->quic;
1142 13088705 : fd_quic_state_t * state = fd_quic_get_state( quic );
1143 13088705 : if( FD_UNLIKELY( !state->stream_pool ) ) return NULL;
1144 :
1145 13088705 : ulong next_stream_id = conn->tx_next_stream_id;
1146 :
1147 : /* The user is responsible for calling this, for setting limits, */
1148 : /* and for setting stream_pool size */
1149 : /* Only current use cases for QUIC client is for testing */
1150 : /* So leaving this question unanswered for now */
1151 :
1152 : /* peer imposed limit on streams */
1153 13088705 : ulong peer_sup_stream_id = conn->tx_sup_stream_id;
1154 :
1155 : /* is connection inactive */
1156 13088705 : if( FD_UNLIKELY( conn->state != FD_QUIC_CONN_STATE_ACTIVE ||
1157 13088705 : next_stream_id >= peer_sup_stream_id ) ) {
1158 : /* this is a normal condition which occurs whenever we run up to
1159 : the peer advertised limit and represents one form of flow control */
1160 0 : return NULL;
1161 0 : }
1162 :
1163 : /* obtain a stream from stream_pool */
1164 13088705 : fd_quic_stream_t * stream = fd_quic_stream_pool_alloc( state->stream_pool );
1165 :
1166 13088705 : if( FD_UNLIKELY( !stream ) ) {
1167 : /* no streams available in the stream pool */
1168 0 : return NULL;
1169 0 : }
1170 :
1171 : /* add to map of stream ids */
1172 13088705 : fd_quic_stream_map_t * entry = fd_quic_stream_map_insert( conn->stream_map, next_stream_id );
1173 13088705 : if( FD_UNLIKELY( !entry ) ) {
1174 : /* return stream to pool */
1175 0 : fd_quic_stream_pool_free( state->stream_pool, stream );
1176 0 : return NULL;
1177 0 : }
1178 :
1179 13088705 : fd_quic_stream_init( stream );
1180 13088705 : FD_QUIC_STREAM_LIST_INIT_STREAM( stream );
1181 :
1182 : /* stream tx_buf already set */
1183 13088705 : stream->conn = conn;
1184 13088705 : stream->stream_id = next_stream_id;
1185 13088705 : stream->context = NULL;
1186 :
1187 : /* set the max stream data to the appropriate initial value */
1188 13088705 : stream->tx_max_stream_data = conn->tx_initial_max_stream_data_uni;
1189 :
1190 : /* set state depending on stream type */
1191 13088705 : stream->state = FD_QUIC_STREAM_STATE_RX_FIN;
1192 13088705 : stream->stream_flags = 0u;
1193 :
1194 13088705 : memset( stream->tx_ack, 0, stream->tx_buf.cap >> 3ul );
1195 :
1196 : /* insert into used streams */
1197 13088705 : FD_QUIC_STREAM_LIST_REMOVE( stream );
1198 13088705 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->used_streams, stream );
1199 :
1200 : /* generate a new stream id */
1201 13088705 : conn->tx_next_stream_id = next_stream_id + 4U;
1202 :
1203 : /* assign the stream to the entry */
1204 13088705 : entry->stream = stream;
1205 :
1206 : /* update metrics */
1207 13088705 : quic->metrics.stream_opened_cnt++;
1208 13088705 : quic->metrics.stream_active_cnt++;
1209 :
1210 13088705 : FD_DEBUG( FD_LOG_DEBUG(( "Created stream with ID %lu", next_stream_id )) );
1211 13088705 : return stream;
1212 13088705 : }
1213 :
1214 : int
1215 : fd_quic_stream_send( fd_quic_stream_t * stream,
1216 : void const * data,
1217 : ulong data_sz,
1218 13088780 : int fin ) {
1219 13088780 : if( FD_UNLIKELY( stream->state & FD_QUIC_STREAM_STATE_TX_FIN ) ) {
1220 0 : return FD_QUIC_SEND_ERR_FIN;
1221 0 : }
1222 :
1223 13088780 : fd_quic_conn_t * conn = stream->conn;
1224 :
1225 13088780 : fd_quic_buffer_t * tx_buf = &stream->tx_buf;
1226 :
1227 : /* are we allowed to send? */
1228 13088780 : ulong stream_id = stream->stream_id;
1229 :
1230 : /* stream_id & 2 == 0 is bidir
1231 : stream_id & 1 == 0 is client */
1232 13088780 : if( FD_UNLIKELY( ( ( (uint)stream_id & 2u ) == 2u ) &
1233 13088780 : ( ( (uint)stream_id & 1u ) != (uint)conn->server ) ) ) {
1234 0 : return FD_QUIC_SEND_ERR_INVAL_STREAM;
1235 0 : }
1236 :
1237 13088780 : if( FD_UNLIKELY( conn->state != FD_QUIC_CONN_STATE_ACTIVE ) ) {
1238 0 : if( conn->state == FD_QUIC_CONN_STATE_HANDSHAKE ||
1239 0 : conn->state == FD_QUIC_CONN_STATE_HANDSHAKE_COMPLETE ) {
1240 0 : return FD_QUIC_SEND_ERR_STREAM_STATE;
1241 0 : }
1242 0 : return FD_QUIC_SEND_ERR_INVAL_CONN;
1243 0 : }
1244 :
1245 : /* how many bytes are we allowed to send on the stream and on the connection? */
1246 13088780 : ulong allowed_stream = stream->tx_max_stream_data - stream->tx_tot_data;
1247 13088780 : ulong allowed_conn = conn->tx_max_data - conn->tx_tot_data;
1248 13088780 : ulong allowed = fd_ulong_min( allowed_conn, allowed_stream );
1249 :
1250 13088780 : if( data_sz > fd_quic_buffer_avail( tx_buf ) ) {
1251 0 : return FD_QUIC_SEND_ERR_FLOW;
1252 0 : }
1253 :
1254 13088780 : if( data_sz > allowed ) {
1255 0 : return FD_QUIC_SEND_ERR_FLOW;
1256 0 : }
1257 :
1258 : /* store data from data into tx_buf
1259 : this stores, but does not move the head offset */
1260 13088780 : fd_quic_buffer_store( tx_buf, data, data_sz );
1261 :
1262 : /* advance head */
1263 13088780 : tx_buf->head += data_sz;
1264 :
1265 : /* adjust flow control limits on stream and connection */
1266 13088780 : stream->tx_tot_data += data_sz;
1267 13088780 : conn->tx_tot_data += data_sz;
1268 :
1269 : /* insert into send list */
1270 13088780 : if( !FD_QUIC_STREAM_ACTION( stream ) ) {
1271 13088780 : FD_QUIC_STREAM_LIST_REMOVE( stream );
1272 13088780 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->send_streams, stream );
1273 13088780 : }
1274 13088780 : stream->stream_flags |= FD_QUIC_STREAM_FLAGS_UNSENT; /* we have unsent data */
1275 13088780 : stream->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING; /* schedule tx */
1276 :
1277 : /* don't actually set fin flag if we didn't add the last
1278 : byte to the buffer */
1279 13088780 : if( fin ) {
1280 13088681 : fd_quic_stream_fin( stream );
1281 13088681 : }
1282 :
1283 : /* schedule send */
1284 13088780 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
1285 :
1286 13088780 : return FD_QUIC_SUCCESS;
1287 13088780 : }
1288 :
1289 : void
1290 13088681 : fd_quic_stream_fin( fd_quic_stream_t * stream ) {
1291 13088681 : if( FD_UNLIKELY( stream->state & FD_QUIC_STREAM_STATE_TX_FIN ) ) {
1292 0 : return;
1293 0 : }
1294 :
1295 13088681 : fd_quic_conn_t * conn = stream->conn;
1296 :
1297 : /* insert into send list */
1298 13088681 : if( !FD_QUIC_STREAM_ACTION( stream ) ) {
1299 0 : FD_QUIC_STREAM_LIST_REMOVE( stream );
1300 0 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->send_streams, stream );
1301 0 : }
1302 13088681 : stream->stream_flags |= FD_QUIC_STREAM_FLAGS_TX_FIN; /* state immediately updated */
1303 13088681 : stream->state |= FD_QUIC_STREAM_STATE_TX_FIN; /* state immediately updated */
1304 13088681 : stream->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING; /* update to be sent in next packet */
1305 :
1306 : /* TODO update metrics */
1307 13088681 : }
1308 :
1309 : void
1310 0 : fd_quic_conn_set_rx_max_data( fd_quic_conn_t * conn, ulong rx_max_data ) {
1311 : /* cannot reduce max_data, and cannot increase beyond max varint */
1312 0 : if( rx_max_data > conn->srx->rx_max_data && rx_max_data < (1UL<<62)-1UL ) {
1313 0 : conn->srx->rx_max_data = rx_max_data;
1314 0 : conn->flags |= FD_QUIC_CONN_FLAGS_MAX_DATA;
1315 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
1316 0 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
1317 0 : }
1318 0 : }
1319 :
1320 : /* packet processing */
1321 :
1322 : /* fd_quic_abandon_enc_level frees all resources associated encryption
1323 : levels less or equal to enc_level. Returns the number of freed
1324 : pkt_meta. */
1325 :
1326 : ulong
1327 : fd_quic_abandon_enc_level( fd_quic_conn_t * conn,
1328 30105 : uint enc_level ) {
1329 30105 : if( FD_LIKELY( !fd_uint_extract_bit( conn->keys_avail, (int)enc_level ) ) ) return 0UL;
1330 24084 : FD_DEBUG( FD_LOG_DEBUG(( "conn=%p abandoning enc_level=%u", (void *)conn, enc_level )); )
1331 :
1332 24084 : ulong freed = 0UL;
1333 :
1334 24084 : fd_quic_ack_gen_abandon_enc_level( conn->ack_gen, enc_level );
1335 :
1336 24084 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
1337 24084 : fd_quic_pkt_meta_t * pool = tracker->pool;
1338 :
1339 72252 : for( uint j = 0; j <= enc_level; ++j ) {
1340 48168 : conn->keys_avail = fd_uint_clear_bit( conn->keys_avail, (int)j );
1341 : /* treat all packets as ACKed (freeing handshake data, etc.) */
1342 48168 : fd_quic_pkt_meta_ds_t * sent = &tracker->sent_pkt_metas[j];
1343 :
1344 48168 : fd_quic_pkt_meta_t * prev = NULL;
1345 48168 : for( fd_quic_pkt_meta_ds_fwd_iter_t iter = fd_quic_pkt_meta_treap_fwd_iter_init( sent, pool );
1346 60213 : !fd_quic_pkt_meta_ds_fwd_iter_done( iter );
1347 48168 : iter = fd_quic_pkt_meta_ds_fwd_iter_next( iter, pool ) ) {
1348 12045 : fd_quic_pkt_meta_t * e = fd_quic_pkt_meta_ds_fwd_iter_ele( iter, pool );
1349 12045 : if( FD_LIKELY( prev ) ) {
1350 0 : fd_quic_pkt_meta_pool_ele_release( pool, prev );
1351 0 : }
1352 12045 : fd_quic_reclaim_pkt_meta( conn, e, j );
1353 12045 : prev = e;
1354 12045 : }
1355 48168 : if( FD_LIKELY( prev ) ) {
1356 12045 : fd_quic_pkt_meta_pool_ele_release( pool, prev );
1357 12045 : }
1358 :
1359 48168 : freed += fd_quic_pkt_meta_ds_ele_cnt( sent );
1360 48168 : conn->used_pkt_meta -= fd_quic_pkt_meta_ds_ele_cnt( sent );
1361 48168 : fd_quic_pkt_meta_ds_clear( tracker, j );
1362 48168 : }
1363 :
1364 24084 : return freed;
1365 30105 : }
1366 :
1367 : static void
1368 : fd_quic_gen_initial_secret_and_keys(
1369 : fd_quic_conn_t * conn,
1370 : fd_quic_conn_id_t const * dst_conn_id,
1371 6057 : int is_server ) {
1372 :
1373 6057 : fd_quic_gen_initial_secrets(
1374 6057 : &conn->secrets,
1375 6057 : dst_conn_id->conn_id, dst_conn_id->sz,
1376 6057 : is_server );
1377 :
1378 6057 : fd_quic_gen_keys(
1379 6057 : &conn->keys[ fd_quic_enc_level_initial_id ][ 0 ],
1380 6057 : conn->secrets.secret[ fd_quic_enc_level_initial_id ][ 0 ] );
1381 :
1382 6057 : fd_quic_gen_keys(
1383 6057 : &conn->keys[ fd_quic_enc_level_initial_id ][ 1 ],
1384 6057 : conn->secrets.secret[ fd_quic_enc_level_initial_id ][ 1 ] );
1385 6057 : }
1386 :
1387 : static ulong
1388 : fd_quic_send_retry( fd_quic_t * quic,
1389 : fd_quic_pkt_t * pkt,
1390 : fd_quic_conn_id_t const * odcid,
1391 : fd_quic_conn_id_t const * scid,
1392 105 : ulong new_conn_id ) {
1393 :
1394 105 : fd_quic_state_t * state = fd_quic_get_state( quic );
1395 :
1396 105 : ulong expire_at = state->now + quic->config.retry_ttl;
1397 105 : uchar retry_pkt[ FD_QUIC_RETRY_LOCAL_SZ ];
1398 105 : ulong retry_pkt_sz = fd_quic_retry_create( retry_pkt, pkt, state->_rng, state->retry_secret, state->retry_iv, odcid, scid, new_conn_id, expire_at );
1399 :
1400 105 : quic->metrics.retry_tx_cnt++;
1401 :
1402 105 : uchar * tx_ptr = retry_pkt + retry_pkt_sz;
1403 105 : if( FD_UNLIKELY( fd_quic_tx_buffered_raw(
1404 105 : quic,
1405 : // these are state variable's normally updated on a conn, but irrelevant in retry so we
1406 : // just size it exactly as the encoded retry packet
1407 105 : &tx_ptr,
1408 105 : retry_pkt,
1409 : // encode buffer
1410 105 : &pkt->ip4->net_id,
1411 105 : pkt->ip4->saddr,
1412 105 : pkt->udp->net_sport,
1413 105 : pkt->ip4->daddr,
1414 105 : pkt->udp->net_dport ) == FD_QUIC_FAILED ) ) {
1415 0 : return FD_QUIC_PARSE_FAIL;
1416 0 : }
1417 105 : return 0UL;
1418 105 : }
1419 :
1420 : /* fd_quic_tls_hs_cache_evict evicts the oldest tls_hs if it's exceeded its ttl
1421 : Assumes cache is non-empty
1422 : and returns 1 if evicted, otherwise returns 0. */
1423 : static int
1424 : fd_quic_tls_hs_cache_evict( fd_quic_t * quic,
1425 6 : fd_quic_state_t * state ) {
1426 :
1427 6 : fd_quic_tls_hs_t* hs_to_free = fd_quic_tls_hs_cache_ele_peek_head( &state->hs_cache, state->hs_pool );
1428 :
1429 6 : if( state->now < hs_to_free->birthtime + quic->config.tls_hs_ttl ) {
1430 : /* oldest is too young to evict */
1431 3 : quic->metrics.hs_err_alloc_fail_cnt++;
1432 3 : return 0;
1433 3 : }
1434 :
1435 3 : fd_quic_conn_free( quic, hs_to_free->context );
1436 3 : quic->metrics.hs_evicted_cnt++;
1437 3 : return 1;
1438 6 : }
1439 :
1440 : /* fd_quic_handle_v1_initial handles an "Initial"-type packet.
1441 : Valid for both server and client. Initial packets are used to
1442 : establish QUIC conns and wrap the TLS handshake flow among other
1443 : things. */
1444 :
1445 : ulong
1446 : fd_quic_handle_v1_initial( fd_quic_t * quic,
1447 : fd_quic_conn_t ** p_conn,
1448 : fd_quic_pkt_t * pkt,
1449 : fd_quic_conn_id_t const * dcid,
1450 : fd_quic_conn_id_t const * peer_scid,
1451 : uchar * cur_ptr,
1452 15525 : ulong cur_sz ) {
1453 15525 : fd_quic_conn_t * conn = *p_conn;
1454 15525 : if( FD_UNLIKELY( conn &&
1455 15525 : ( conn->state==FD_QUIC_CONN_STATE_INVALID ||
1456 15525 : !fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_initial_id ) ) ) ) {
1457 3 : quic->metrics.pkt_no_key_cnt[ fd_quic_enc_level_initial_id ]++;
1458 3 : return FD_QUIC_PARSE_FAIL;
1459 3 : }
1460 :
1461 15522 : fd_quic_state_t * state = fd_quic_get_state( quic );
1462 15522 : fd_quic_metrics_t * metrics = &quic->metrics;
1463 :
1464 : /* Initial packets are de-facto unencrypted. Packet protection is
1465 : still applied, albeit with publicly known encryption keys.
1466 :
1467 : RFC 9001 specifies use of the TLS_AES_128_GCM_SHA256_ID suite for
1468 : initial secrets and keys. */
1469 :
1470 : /* Parse initial packet */
1471 :
1472 15522 : fd_quic_initial_t initial[1] = {0};
1473 15522 : ulong rc = fd_quic_decode_initial( initial, cur_ptr, cur_sz );
1474 15522 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
1475 1494 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_decode_initial failed" )) );
1476 1494 : return FD_QUIC_PARSE_FAIL;
1477 1494 : }
1478 :
1479 : /* Check bounds on initial */
1480 :
1481 : /* len indicated the number of bytes after the packet number offset
1482 : so verify this value is within the packet */
1483 14028 : ulong pn_offset = initial->pkt_num_pnoff;
1484 14028 : ulong body_sz = initial->len; /* length of packet number, frames, and auth tag */
1485 14028 : ulong tot_sz = pn_offset + body_sz;
1486 14028 : if( FD_UNLIKELY( tot_sz > cur_sz ) ) {
1487 66 : FD_DEBUG( FD_LOG_DEBUG(( "Bogus initial packet length" )) );
1488 66 : return FD_QUIC_PARSE_FAIL;
1489 66 : }
1490 :
1491 : /* Check it is valid for a token to be present in an initial packet in the current context.
1492 :
1493 : quic->config.role == FD_QUIC_ROLE_CLIENT
1494 : - Indicates the client received an initial packet with a token from a server. "Initial packets
1495 : sent by the server MUST set the Token Length field to 0; clients that receive an Initial packet
1496 : with a non-zero Token Length field MUST either discard the packet or generate a connection
1497 : error of type PROTOCOL_VIOLATION (RFC 9000, Section 17.2.2)"
1498 :
1499 : quic->config.retry == false
1500 : - Indicates the server is not configured to retry, but a client attached a token to this
1501 : initial packet. NEW_TOKEN frames are not supported, so this implementation treats the presence
1502 : of a token when retry is disabled as an error. */
1503 13962 : if( FD_UNLIKELY( initial->token_len > 0 &&
1504 13962 : ( quic->config.role == FD_QUIC_ROLE_CLIENT || !quic->config.retry ) ) ) {
1505 21 : FD_DEBUG( FD_LOG_DEBUG(( "Rejecting initial with token" )); )
1506 21 : return FD_QUIC_PARSE_FAIL;
1507 21 : }
1508 :
1509 :
1510 13941 : ulong scid; /* outgoing scid */
1511 13941 : fd_quic_conn_id_t odcid; /* dst conn id from client's original Initial */
1512 :
1513 : /* Do we have a conn object for this dest conn ID?
1514 : If not, sanity check, send/verify retry if needed */
1515 13941 : if( FD_UNLIKELY( !conn ) ) {
1516 : /* if we're a client, and no conn, discard */
1517 7920 : if( quic->config.role == FD_QUIC_ROLE_CLIENT ) {
1518 : /* connection may have been torn down */
1519 774 : FD_DEBUG( FD_LOG_DEBUG(( "unknown connection ID" )); )
1520 774 : metrics->pkt_no_conn_cnt++;
1521 774 : return FD_QUIC_PARSE_FAIL;
1522 774 : }
1523 :
1524 : /* According to RFC 9000 Section 14.1, INITIAL packets less than a
1525 : certain length must be discarded, and the connection may be closed.
1526 : (Mitigates UDP amplification) */
1527 7146 : if( pkt->datagram_sz < FD_QUIC_INITIAL_PAYLOAD_SZ_MIN ) {
1528 : /* can't trust the included values, so can't reply */
1529 735 : return FD_QUIC_PARSE_FAIL;
1530 735 : }
1531 :
1532 : /* Early check: Is conn free? */
1533 6411 : if( FD_UNLIKELY( state->free_conn_list==UINT_MAX ) ) {
1534 0 : FD_DEBUG( FD_LOG_DEBUG(( "ignoring conn request: no free conn slots" )) );
1535 0 : metrics->conn_err_no_slots_cnt++;
1536 0 : return FD_QUIC_PARSE_FAIL; /* FIXME better error code? */
1537 0 : }
1538 :
1539 :
1540 : /* Primary objective is to send or verify retry.
1541 : We'll also select the scid we'll use from now on.
1542 :
1543 : Rules for selecting the SCID:
1544 : - No retry token, accepted: generate new random ID
1545 : - No retry token, retry request: generate new random ID
1546 : - Retry token, accepted: reuse SCID from retry token */
1547 6411 : if( !quic->config.retry ) {
1548 6276 : scid = fd_rng_ulong( state->_rng );
1549 6276 : } else { /* retry configured */
1550 :
1551 : /* Need to send retry? Do so before more work */
1552 135 : if( initial->token_len == 0 ) {
1553 105 : ulong new_conn_id_u64 = fd_rng_ulong( state->_rng );
1554 105 : if( FD_UNLIKELY( fd_quic_send_retry(
1555 105 : quic, pkt,
1556 105 : dcid, peer_scid, new_conn_id_u64 ) ) ) {
1557 0 : return FD_QUIC_FAILED;
1558 0 : }
1559 105 : return (initial->pkt_num_pnoff + initial->len);
1560 :
1561 105 : } else {
1562 : /* This Initial packet is in response to our Retry.
1563 : Validate the relevant fields of this post-retry INITIAL packet,
1564 : i.e. retry src conn id, ip, port
1565 : Also populate odcid and scid from the retry data */
1566 30 : int retry_ok = fd_quic_retry_server_verify( pkt, initial, &odcid, &scid, state->retry_secret, state->retry_iv, state->now, quic->config.retry_ttl );
1567 30 : if( FD_UNLIKELY( retry_ok!=FD_QUIC_SUCCESS ) ) {
1568 27 : metrics->conn_err_retry_fail_cnt++;
1569 : /* No need to set conn error, no conn object exists */
1570 27 : return FD_QUIC_PARSE_FAIL;
1571 27 : };
1572 3 : }
1573 135 : }
1574 6411 : }
1575 :
1576 : /* Determine decryption keys, related data */
1577 :
1578 : /* Placeholder for generated crypto material before allocating conn */
1579 12300 : fd_quic_crypto_keys_t _rx_keys[1];
1580 12300 : fd_quic_crypto_secrets_t _secrets[1];
1581 :
1582 : /* Conditional inputs to decryption stage */
1583 12300 : fd_quic_crypto_keys_t * rx_keys = NULL;
1584 12300 : fd_quic_crypto_secrets_t * secrets = NULL;
1585 12300 : ulong exp_pkt_num;
1586 :
1587 12300 : if( !conn ) {
1588 : /* no conn, generate secret and rx keys */
1589 6279 : rx_keys = _rx_keys;
1590 6279 : secrets = _secrets;
1591 6279 : exp_pkt_num = 0;
1592 :
1593 6279 : fd_quic_gen_initial_secrets(
1594 6279 : secrets,
1595 6279 : dcid->conn_id, dcid->sz,
1596 6279 : /* is_server */ 1 );
1597 6279 : fd_quic_gen_keys(
1598 6279 : rx_keys,
1599 6279 : secrets->secret[ fd_quic_enc_level_initial_id ][ 0 ] );
1600 6279 : } else {
1601 : /* conn, use existing keys/secrets */
1602 6021 : rx_keys = &conn->keys[ fd_quic_enc_level_initial_id ][0];
1603 6021 : secrets = &conn->secrets;
1604 6021 : exp_pkt_num = conn->exp_pkt_number[0];
1605 6021 : }
1606 :
1607 : /* Decrypt incoming packet */
1608 :
1609 : /* header protection needs the offset to the packet number */
1610 :
1611 12300 : # if !FD_QUIC_DISABLE_CRYPTO
1612 : /* this decrypts the header */
1613 12300 : if( FD_UNLIKELY(
1614 12300 : fd_quic_crypto_decrypt_hdr( cur_ptr, cur_sz,
1615 12300 : pn_offset,
1616 12300 : rx_keys ) != FD_QUIC_SUCCESS ) ) {
1617 : /* As this is an INITIAL packet, change the status to DEAD, and allow
1618 : it to be reaped */
1619 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_crypto_decrypt_hdr failed" )) );
1620 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_initial_id ]++;
1621 0 : return FD_QUIC_PARSE_FAIL;
1622 0 : }
1623 12300 : # endif /* !FD_QUIC_DISABLE_CRYPTO */
1624 :
1625 12300 : ulong pkt_number_sz = fd_quic_h0_pkt_num_len( cur_ptr[0] ) + 1u;
1626 12300 : ulong pktnum_comp = fd_quic_pktnum_decode( cur_ptr+pn_offset, pkt_number_sz );
1627 :
1628 : /* reconstruct packet number */
1629 12300 : ulong pkt_number = fd_quic_reconstruct_pkt_num( pktnum_comp, pkt_number_sz, exp_pkt_num );
1630 :
1631 12300 : # if !FD_QUIC_DISABLE_CRYPTO
1632 : /* NOTE from rfc9002 s3
1633 : It is permitted for some packet numbers to never be used, leaving intentional gaps. */
1634 : /* this decrypts the header and payload */
1635 12300 : if( FD_UNLIKELY(
1636 12300 : fd_quic_crypto_decrypt( cur_ptr, tot_sz,
1637 12300 : pn_offset,
1638 12300 : pkt_number,
1639 12300 : rx_keys ) != FD_QUIC_SUCCESS ) ) {
1640 177 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_crypto_decrypt failed" )) );
1641 177 : FD_DTRACE_PROBE_2( quic_err_decrypt_initial_pkt, pkt->ip4, pkt->pkt_number );
1642 177 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_initial_id ]++;
1643 177 : return FD_QUIC_PARSE_FAIL;
1644 177 : }
1645 12123 : # endif /* FD_QUIC_DISABLE_CRYPTO */
1646 :
1647 : /* set packet number on the context */
1648 12123 : pkt->pkt_number = pkt_number;
1649 :
1650 12123 : if( FD_UNLIKELY( body_sz < pkt_number_sz + FD_QUIC_CRYPTO_TAG_SZ ) ) {
1651 0 : return FD_QUIC_PARSE_FAIL;
1652 0 : }
1653 :
1654 : /* If no conn, create one. Due to previous checks, role must be server
1655 : and this must be response to Retry (if needed). */
1656 12123 : if( FD_UNLIKELY( !conn ) ) {
1657 :
1658 : /* Save peer's conn ID, which we will use to address peer with. */
1659 6102 : fd_quic_conn_id_t peer_conn_id = {0};
1660 6102 : fd_memcpy( peer_conn_id.conn_id, initial->src_conn_id, FD_QUIC_MAX_CONN_ID_SZ );
1661 6102 : peer_conn_id.sz = initial->src_conn_id_len;
1662 :
1663 : /* Prepare QUIC-TLS transport params object (sent as a TLS extension).
1664 : Take template from state and mutate certain params in-place.
1665 :
1666 : See RFC 9000 Section 18 */
1667 :
1668 : /* TODO Each transport param is a TLV tuple. This allows serializing
1669 : most transport params ahead of time. Only the conn-specific
1670 : differences will have to be appended here. */
1671 :
1672 6102 : fd_quic_transport_params_t tp[1] = { state->transport_params };
1673 :
1674 6102 : if( !quic->config.retry ) {
1675 : /* assume no retry */
1676 6099 : tp->retry_source_connection_id_present = 0;
1677 :
1678 : /* Send orig conn ID back to client (server only) */
1679 :
1680 6099 : tp->original_destination_connection_id_present = 1;
1681 6099 : tp->original_destination_connection_id_len = dcid->sz;
1682 6099 : fd_memcpy( tp->original_destination_connection_id,
1683 6099 : dcid->conn_id,
1684 6099 : dcid->sz );
1685 6099 : } else { /* retry configured */
1686 :
1687 : /* From rfc 9000:
1688 :
1689 : Figure 8 shows a similar handshake that includes a Retry packet.
1690 :
1691 : Client Server
1692 : Initial: DCID=S1, SCID=C1 ->
1693 : <- Retry: DCID=C1, SCID=S2
1694 : Initial: DCID=S2, SCID=C1 ->
1695 : <- Initial: DCID=C1, SCID=S3
1696 : ...
1697 : 1-RTT: DCID=S3 ->
1698 : <- 1-RTT: DCID=C1
1699 :
1700 : Figure 8: Use of Connection IDs in a Handshake with Retry
1701 : In both cases (Figures 7 and 8), the client sets the value of the
1702 : initial_source_connection_id transport parameter to C1.
1703 :
1704 : When the handshake does not include a Retry (Figure 7), the server
1705 : sets original_destination_connection_id to S1 (note that this value
1706 : is chosen by the client) and initial_source_connection_id to S3. In
1707 : this case, the server does not include a retry_source_connection_id
1708 : transport parameter.
1709 :
1710 : When the handshake includes a Retry (Figure 8), the server sets
1711 : original_destination_connection_id to S1, retry_source_connection_id
1712 : to S2, and initial_source_connection_id to S3. */
1713 3 : tp->original_destination_connection_id_present = 1;
1714 3 : tp->original_destination_connection_id_len = odcid.sz;
1715 3 : memcpy( tp->original_destination_connection_id,
1716 3 : odcid.conn_id,
1717 3 : odcid.sz );
1718 :
1719 : /* Client echoes back the SCID we sent via Retry. Safe to trust
1720 : because we signed the Retry Token. (Length and content validated
1721 : in fd_quic_retry_server_verify) */
1722 3 : tp->retry_source_connection_id_present = 1;
1723 3 : tp->retry_source_connection_id_len = FD_QUIC_CONN_ID_SZ;
1724 3 : FD_STORE( ulong, tp->retry_source_connection_id, scid );
1725 :
1726 3 : metrics->conn_retry_cnt++;
1727 3 : }
1728 :
1729 : /* Repeat the conn ID we picked in transport params (this is done
1730 : to authenticate conn IDs via TLS by including them in TLS-
1731 : protected data).
1732 :
1733 : Per spec, this field should be the source conn ID field we've set
1734 : on the first Initial packet we've sent. At this point, we might
1735 : not have sent an Initial packet yet -- so this field should hold
1736 : a value we are about to pick.
1737 :
1738 : fd_quic_conn_create will set conn->initial_source_conn_id to
1739 : the random new_conn_id we've created earlier. */
1740 :
1741 6102 : tp->initial_source_connection_id_present = 1;
1742 6102 : tp->initial_source_connection_id_len = FD_QUIC_CONN_ID_SZ;
1743 6102 : FD_STORE( ulong, tp->initial_source_connection_id, scid );
1744 :
1745 : /* tls hs available? After decrypting because might evict another hs */
1746 6102 : if( FD_UNLIKELY( !fd_quic_tls_hs_pool_free( state->hs_pool ) ) ) {
1747 : /* try evicting, 0 if oldest is too young so fail */
1748 0 : if( !fd_quic_tls_hs_cache_evict( quic, state )) {
1749 0 : return FD_QUIC_PARSE_FAIL;
1750 0 : }
1751 0 : }
1752 :
1753 : /* Allocate new conn */
1754 6102 : conn = fd_quic_conn_create( quic,
1755 6102 : scid,
1756 6102 : &peer_conn_id,
1757 6102 : pkt->ip4->saddr,
1758 6102 : pkt->udp->net_sport,
1759 6102 : pkt->ip4->daddr,
1760 6102 : pkt->udp->net_dport,
1761 6102 : 1 /* server */ );
1762 :
1763 6102 : if( FD_UNLIKELY( !conn ) ) { /* no free connections */
1764 : /* TODO send failure back to origin? */
1765 : /* FIXME unreachable? conn_cnt already checked above */
1766 0 : FD_DEBUG( FD_LOG_WARNING( ( "failed to allocate QUIC conn" ) ) );
1767 0 : return FD_QUIC_PARSE_FAIL;
1768 0 : }
1769 6102 : FD_DEBUG( FD_LOG_DEBUG(( "new connection allocated" )) );
1770 :
1771 : /* set the value for the caller */
1772 6102 : *p_conn = conn;
1773 :
1774 : /* Create a TLS handshake */
1775 6102 : fd_quic_tls_hs_t * tls_hs = fd_quic_tls_hs_new(
1776 6102 : fd_quic_tls_hs_pool_ele_acquire( state->hs_pool ),
1777 6102 : state->tls,
1778 6102 : (void*)conn,
1779 6102 : 1 /*is_server*/,
1780 6102 : tp,
1781 6102 : state->now );
1782 6102 : fd_quic_tls_hs_cache_ele_push_tail( &state->hs_cache, tls_hs, state->hs_pool );
1783 :
1784 6102 : conn->tls_hs = tls_hs;
1785 6102 : quic->metrics.hs_created_cnt++;
1786 :
1787 : /* copy secrets and rx keys */
1788 6102 : conn->secrets = *secrets;
1789 6102 : conn->keys[ fd_quic_enc_level_initial_id ][0] = *rx_keys;
1790 :
1791 : /* generate tx keys */
1792 6102 : fd_quic_gen_keys(
1793 6102 : &conn->keys[ fd_quic_enc_level_initial_id ][ 1 ],
1794 6102 : secrets->secret[ fd_quic_enc_level_initial_id ][ 1 ] );
1795 6102 : }
1796 :
1797 12123 : if( FD_UNLIKELY( !conn->host.ip_addr ) ) {
1798 : /* Lock src IP address in place (previously chosen by layer-4 based
1799 : on the route table) */
1800 12123 : conn->host.ip_addr = pkt->ip4->daddr;
1801 12123 : }
1802 :
1803 : /* check if reply conn id needs to change */
1804 12123 : if( FD_UNLIKELY( !( conn->server | conn->established ) ) ) {
1805 : /* switch to the source connection id for future replies */
1806 :
1807 : /* replace peer 0 connection id */
1808 6021 : conn->peer_cids[0].sz = initial->src_conn_id_len;
1809 6021 : fd_memcpy( conn->peer_cids[0].conn_id, initial->src_conn_id, FD_QUIC_MAX_CONN_ID_SZ );
1810 :
1811 : /* don't repeat this procedure */
1812 6021 : conn->established = 1;
1813 6021 : }
1814 :
1815 : /* handle frames */
1816 12123 : ulong payload_off = pn_offset + pkt_number_sz;
1817 12123 : uchar const * frame_ptr = cur_ptr + payload_off;
1818 12123 : ulong frame_sz = body_sz - pkt_number_sz - FD_QUIC_CRYPTO_TAG_SZ; /* total size of all frames in packet */
1819 36219 : while( frame_sz != 0UL ) {
1820 24177 : rc = fd_quic_handle_v1_frame( quic,
1821 24177 : conn,
1822 24177 : pkt,
1823 24177 : FD_QUIC_PKT_TYPE_INITIAL,
1824 24177 : frame_ptr,
1825 24177 : frame_sz );
1826 24177 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
1827 81 : FD_DEBUG( FD_LOG_DEBUG(( "Failed to handle frame (Initial, frame=0x%02x)", frame_ptr[0] )) );
1828 81 : quic->metrics.frame_rx_err_cnt++;
1829 81 : return FD_QUIC_PARSE_FAIL;
1830 81 : }
1831 :
1832 24096 : if( FD_UNLIKELY( rc==0UL || rc>frame_sz ) ) {
1833 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
1834 0 : return FD_QUIC_PARSE_FAIL;
1835 0 : }
1836 :
1837 : /* next frame, and remaining size */
1838 24096 : frame_ptr += rc;
1839 24096 : frame_sz -= rc;
1840 24096 : }
1841 :
1842 : /* update last activity */
1843 12042 : conn->last_activity = state->now;
1844 12042 : conn->flags &= ~( FD_QUIC_CONN_FLAGS_PING_SENT | FD_QUIC_CONN_FLAGS_PING );
1845 :
1846 : /* update expected packet number */
1847 12042 : conn->exp_pkt_number[0] = fd_ulong_max( conn->exp_pkt_number[0], pkt_number+1UL );
1848 :
1849 : /* insert into service queue */
1850 12042 : fd_quic_svc_schedule( state, conn, FD_QUIC_SVC_INSTANT );
1851 :
1852 : /* return number of bytes consumed */
1853 12042 : return tot_sz;
1854 12123 : }
1855 :
1856 : ulong
1857 : fd_quic_handle_v1_handshake(
1858 : fd_quic_t * quic,
1859 : fd_quic_conn_t * conn,
1860 : fd_quic_pkt_t * pkt,
1861 : uchar * cur_ptr,
1862 : ulong cur_sz
1863 12147 : ) {
1864 12147 : if( FD_UNLIKELY( !conn ) ) {
1865 102 : quic->metrics.pkt_no_conn_cnt++;
1866 102 : return FD_QUIC_PARSE_FAIL;
1867 102 : }
1868 :
1869 12045 : if( FD_UNLIKELY( conn->state==FD_QUIC_CONN_STATE_INVALID ||
1870 12045 : !fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_handshake_id ) ) ) {
1871 3 : quic->metrics.pkt_no_key_cnt[ fd_quic_enc_level_handshake_id ]++;
1872 3 : return FD_QUIC_PARSE_FAIL;
1873 3 : }
1874 :
1875 : /* do parse here */
1876 12042 : fd_quic_handshake_t handshake[1];
1877 12042 : ulong rc = fd_quic_decode_handshake( handshake, cur_ptr, cur_sz );
1878 12042 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
1879 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_decode_handshake failed" )) );
1880 0 : return FD_QUIC_PARSE_FAIL;
1881 0 : }
1882 :
1883 : /* check bounds on handshake */
1884 :
1885 : /* len indicated the number of bytes after the packet number offset
1886 : so verify this value is within the packet */
1887 12042 : ulong len = (ulong)( handshake->pkt_num_pnoff + handshake->len );
1888 12042 : if( FD_UNLIKELY( len > cur_sz ) ) {
1889 0 : FD_DEBUG( FD_LOG_DEBUG(( "Handshake packet bounds check failed" )); )
1890 0 : return FD_QUIC_PARSE_FAIL;
1891 0 : }
1892 :
1893 : /* connection ids should already be in the relevant structures */
1894 :
1895 : /* TODO prepare most of the transport parameters, and only append the
1896 : necessary differences */
1897 :
1898 : /* fetch TLS handshake */
1899 12042 : fd_quic_tls_hs_t * tls_hs = conn->tls_hs;
1900 12042 : if( FD_UNLIKELY( !tls_hs ) ) {
1901 0 : FD_DEBUG( FD_LOG_DEBUG(( "no tls handshake" )) );
1902 0 : return FD_QUIC_PARSE_FAIL;
1903 0 : }
1904 :
1905 : /* decryption */
1906 :
1907 : /* header protection needs the offset to the packet number */
1908 12042 : ulong pn_offset = handshake->pkt_num_pnoff;
1909 :
1910 12042 : ulong body_sz = handshake->len; /* not a protected field */
1911 : /* length of payload + num packet bytes */
1912 :
1913 12042 : # if !FD_QUIC_DISABLE_CRYPTO
1914 : /* this decrypts the header */
1915 12042 : if( FD_UNLIKELY(
1916 12042 : fd_quic_crypto_decrypt_hdr( cur_ptr, cur_sz,
1917 12042 : pn_offset,
1918 12042 : &conn->keys[2][0] ) != FD_QUIC_SUCCESS ) ) {
1919 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_crypto_decrypt_hdr failed" )) );
1920 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_handshake_id ]++;
1921 0 : return FD_QUIC_PARSE_FAIL;
1922 0 : }
1923 12042 : # endif /* !FD_QUIC_DISABLE_CRYPTO */
1924 :
1925 : /* number of bytes in the packet header */
1926 12042 : ulong pkt_number_sz = fd_quic_h0_pkt_num_len( cur_ptr[0] ) + 1u;
1927 12042 : ulong tot_sz = pn_offset + body_sz; /* total including header and payload */
1928 :
1929 : /* now we have decrypted packet number */
1930 12042 : ulong pktnum_comp = fd_quic_pktnum_decode( cur_ptr+pn_offset, pkt_number_sz );
1931 :
1932 : /* reconstruct packet number */
1933 12042 : ulong pkt_number = fd_quic_reconstruct_pkt_num( pktnum_comp, pkt_number_sz, conn->exp_pkt_number[1] );
1934 :
1935 : /* NOTE from rfc9002 s3
1936 : It is permitted for some packet numbers to never be used, leaving intentional gaps. */
1937 :
1938 12042 : # if !FD_QUIC_DISABLE_CRYPTO
1939 : /* this decrypts the header and payload */
1940 12042 : if( FD_UNLIKELY(
1941 12042 : fd_quic_crypto_decrypt( cur_ptr, tot_sz,
1942 12042 : pn_offset,
1943 12042 : pkt_number,
1944 12042 : &conn->keys[2][0] ) != FD_QUIC_SUCCESS ) ) {
1945 : /* remove connection from map, and insert into free list */
1946 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_crypto_decrypt failed" )) );
1947 0 : FD_DTRACE_PROBE_3( quic_err_decrypt_handshake_pkt, pkt->ip4, conn->our_conn_id, pkt->pkt_number );
1948 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_handshake_id ]++;
1949 0 : return FD_QUIC_PARSE_FAIL;
1950 0 : }
1951 12042 : # endif /* FD_QUIC_DISABLE_CRYPTO */
1952 :
1953 : /* set packet number on the context */
1954 12042 : pkt->pkt_number = pkt_number;
1955 :
1956 : /* check body size large enough for required elements */
1957 12042 : if( FD_UNLIKELY( body_sz < pkt_number_sz + FD_QUIC_CRYPTO_TAG_SZ ) ) {
1958 0 : return FD_QUIC_PARSE_FAIL;
1959 0 : }
1960 :
1961 : /* RFC 9000 Section 17.2.2.1. Abandoning Initial Packets
1962 : > A server stops sending and processing Initial packets when it
1963 : > receives its first Handshake packet. */
1964 12042 : fd_quic_abandon_enc_level( conn, fd_quic_enc_level_initial_id );
1965 12042 : conn->peer_enc_level = (uchar)fd_uchar_max( conn->peer_enc_level, fd_quic_enc_level_handshake_id );
1966 :
1967 : /* handle frames */
1968 12042 : ulong payload_off = pn_offset + pkt_number_sz;
1969 12042 : uchar const * frame_ptr = cur_ptr + payload_off;
1970 12042 : ulong frame_sz = body_sz - pkt_number_sz - FD_QUIC_CRYPTO_TAG_SZ; /* total size of all frames in packet */
1971 48168 : while( frame_sz != 0UL ) {
1972 36126 : rc = fd_quic_handle_v1_frame( quic,
1973 36126 : conn,
1974 36126 : pkt,
1975 36126 : FD_QUIC_PKT_TYPE_HANDSHAKE,
1976 36126 : frame_ptr,
1977 36126 : frame_sz );
1978 36126 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
1979 0 : FD_DEBUG( FD_LOG_DEBUG(( "Failed to handle frame (Handshake, frame=0x%02x)", frame_ptr[0] )) );
1980 0 : quic->metrics.frame_rx_err_cnt++;
1981 0 : return FD_QUIC_PARSE_FAIL;
1982 0 : }
1983 :
1984 36126 : if( FD_UNLIKELY( rc == 0UL || rc > frame_sz ) ) {
1985 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
1986 0 : return FD_QUIC_PARSE_FAIL;
1987 0 : }
1988 :
1989 : /* next frame and remaining size */
1990 36126 : frame_ptr += rc;
1991 36126 : frame_sz -= rc;
1992 36126 : }
1993 :
1994 : /* update last activity */
1995 12042 : conn->last_activity = fd_quic_get_state( quic )->now;
1996 12042 : conn->flags &= ~( FD_QUIC_CONN_FLAGS_PING_SENT | FD_QUIC_CONN_FLAGS_PING );
1997 :
1998 : /* update expected packet number */
1999 12042 : conn->exp_pkt_number[1] = fd_ulong_max( conn->exp_pkt_number[1], pkt_number+1UL );
2000 :
2001 : /* return number of bytes consumed */
2002 12042 : return tot_sz;
2003 12042 : }
2004 :
2005 : ulong
2006 : fd_quic_handle_v1_retry(
2007 : fd_quic_t * quic,
2008 : fd_quic_conn_t * conn,
2009 : fd_quic_pkt_t const * pkt,
2010 : uchar const * cur_ptr,
2011 : ulong cur_sz
2012 147 : ) {
2013 147 : (void)pkt;
2014 :
2015 147 : if( FD_UNLIKELY( quic->config.role == FD_QUIC_ROLE_SERVER ) ) {
2016 36 : if( FD_UNLIKELY( conn ) ) { /* likely a misbehaving client w/o a conn */
2017 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
2018 0 : }
2019 36 : return FD_QUIC_PARSE_FAIL;
2020 36 : }
2021 :
2022 111 : if( FD_UNLIKELY( !conn ) ) {
2023 108 : quic->metrics.pkt_no_conn_cnt++;
2024 108 : return FD_QUIC_PARSE_FAIL;
2025 108 : }
2026 :
2027 3 : fd_quic_conn_id_t const * orig_dst_conn_id = &conn->peer_cids[0];
2028 3 : uchar const * retry_token = NULL;
2029 3 : ulong retry_token_sz = 0UL;
2030 :
2031 3 : int rc = fd_quic_retry_client_verify(
2032 3 : cur_ptr, cur_sz,
2033 3 : orig_dst_conn_id,
2034 3 : &conn->retry_src_conn_id,
2035 3 : &retry_token, &retry_token_sz
2036 3 : );
2037 3 : if( FD_UNLIKELY( rc!=FD_QUIC_SUCCESS ) ) {
2038 0 : quic->metrics.conn_err_retry_fail_cnt++;
2039 0 : return FD_QUIC_PARSE_FAIL;
2040 0 : }
2041 :
2042 : /* Update the peer using the retry src conn id */
2043 3 : conn->peer_cids[0] = conn->retry_src_conn_id;
2044 :
2045 : /* Re-send the ClientHello */
2046 3 : conn->hs_sent_bytes[fd_quic_enc_level_initial_id] = 0;
2047 :
2048 : /* Need to regenerate keys using the retry source connection id */
2049 3 : fd_quic_gen_initial_secret_and_keys( conn, &conn->retry_src_conn_id, /* is_server */ 0 );
2050 :
2051 : /* The token length is the remaining bytes in the retry packet after subtracting known fields. */
2052 3 : conn->token_len = retry_token_sz;
2053 3 : fd_memcpy( &conn->token, retry_token, conn->token_len );
2054 :
2055 : /* have to rewind the handshake data */
2056 3 : uint enc_level = fd_quic_enc_level_initial_id;
2057 3 : conn->hs_sent_bytes[enc_level] = 0;
2058 :
2059 : /* send the INITIAL */
2060 3 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
2061 :
2062 3 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
2063 :
2064 3 : return cur_sz;
2065 3 : }
2066 :
2067 : ulong
2068 78 : fd_quic_handle_v1_zero_rtt( fd_quic_t * quic, fd_quic_conn_t * conn, fd_quic_pkt_t const * pkt, uchar const * cur_ptr, ulong cur_sz ) {
2069 78 : (void)pkt;
2070 78 : (void)quic;
2071 78 : (void)cur_ptr;
2072 78 : (void)cur_sz;
2073 : /* since we do not support zero-rtt, simply fail the packet */
2074 78 : if( conn ) {
2075 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_INTERNAL_ERROR, __LINE__ );
2076 0 : }
2077 78 : return FD_QUIC_PARSE_FAIL;
2078 78 : }
2079 :
2080 : int
2081 : fd_quic_lazy_ack_pkt( fd_quic_t * quic,
2082 : fd_quic_conn_t * conn,
2083 83372658 : fd_quic_pkt_t const * pkt ) {
2084 83372658 : if( pkt->ack_flag & ACK_FLAG_CANCEL ) {
2085 0 : return FD_QUIC_ACK_TX_CANCEL;
2086 0 : }
2087 :
2088 83372658 : fd_quic_state_t * state = fd_quic_get_state( quic );
2089 83372658 : int res = fd_quic_ack_pkt( conn->ack_gen, pkt->pkt_number, pkt->enc_level, state->now );
2090 83372658 : conn->ack_gen->is_elicited |= fd_uchar_if( pkt->ack_flag & ACK_FLAG_RQD, 1, 0 );
2091 :
2092 : /* Trigger immediate ACK send? */
2093 83372658 : int ack_sz_threshold_hit = conn->unacked_sz > quic->config.ack_threshold;
2094 83372658 : int force_instant_ack =
2095 83372658 : ( !!(pkt->ack_flag & ACK_FLAG_RQD) ) &
2096 83372658 : ( ( pkt->enc_level == fd_quic_enc_level_initial_id ) |
2097 83372658 : ( pkt->enc_level == fd_quic_enc_level_handshake_id ) );
2098 83372658 : uint svc_type;
2099 83372658 : if( ack_sz_threshold_hit | force_instant_ack ) {
2100 252823 : conn->unacked_sz = 0UL;
2101 252823 : svc_type = FD_QUIC_SVC_INSTANT;
2102 83119835 : } else {
2103 83119835 : svc_type = FD_QUIC_SVC_ACK_TX;
2104 83119835 : }
2105 83372658 : fd_quic_svc_schedule( state, conn, svc_type );
2106 :
2107 83372658 : return res;
2108 83372658 : }
2109 :
2110 : /* This thunk works around a compiler bug (bogus stringop-overflow warning) in GCC 11 */
2111 : __attribute__((noinline)) static void
2112 96 : fd_quic_key_update_derive1( fd_quic_conn_t * conn ) {
2113 96 : fd_quic_key_update_derive( &conn->secrets, conn->new_keys );
2114 96 : }
2115 :
2116 : static void
2117 96 : fd_quic_key_update_complete( fd_quic_conn_t * conn ) {
2118 : /* Key updates are only possible for 1-RTT packets, which are appdata */
2119 96 : ulong const enc_level = fd_quic_enc_level_appdata_id;
2120 :
2121 : /* Update payload keys */
2122 96 : memcpy( conn->keys[enc_level][0].pkt_key, conn->new_keys[0].pkt_key, FD_AES_128_KEY_SZ );
2123 96 : memcpy( conn->keys[enc_level][0].iv, conn->new_keys[0].iv, FD_AES_GCM_IV_SZ );
2124 96 : memcpy( conn->keys[enc_level][1].pkt_key, conn->new_keys[1].pkt_key, FD_AES_128_KEY_SZ );
2125 96 : memcpy( conn->keys[enc_level][1].iv, conn->new_keys[1].iv, FD_AES_GCM_IV_SZ );
2126 :
2127 : /* Update IVs */
2128 96 : memcpy( conn->secrets.secret[enc_level][0], conn->secrets.new_secret[0], FD_QUIC_SECRET_SZ );
2129 96 : memcpy( conn->secrets.secret[enc_level][1], conn->secrets.new_secret[1], FD_QUIC_SECRET_SZ );
2130 :
2131 : /* Packet header encryption keys are not updated */
2132 :
2133 : /* Wind up for next key phase update */
2134 96 : conn->key_phase = !conn->key_phase;
2135 96 : conn->key_update = 0;
2136 96 : fd_quic_key_update_derive1( conn );
2137 :
2138 96 : FD_DEBUG( FD_LOG_DEBUG(( "key update completed" )); )
2139 96 : }
2140 :
2141 : ulong
2142 : fd_quic_handle_v1_one_rtt( fd_quic_t * quic,
2143 : fd_quic_conn_t * conn,
2144 : fd_quic_pkt_t * pkt,
2145 : uchar * const cur_ptr,
2146 13360398 : ulong const tot_sz ) {
2147 13360398 : if( !conn ) {
2148 6882 : quic->metrics.pkt_no_conn_cnt++;
2149 6882 : return FD_QUIC_PARSE_FAIL;
2150 6882 : }
2151 13353516 : if( FD_UNLIKELY( conn->state==FD_QUIC_CONN_STATE_INVALID ||
2152 13353516 : !fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_appdata_id ) ) ) {
2153 6012 : quic->metrics.pkt_no_key_cnt[ fd_quic_enc_level_appdata_id ]++;
2154 6012 : return FD_QUIC_PARSE_FAIL;
2155 6012 : }
2156 :
2157 13347504 : if( FD_UNLIKELY( tot_sz < (1+FD_QUIC_CONN_ID_SZ+1) ) ) {
2158 : /* One-RTT header: 1 byte
2159 : DCID: FD_QUIC_CONN_ID_SZ
2160 : Pkt number: 1-4 bytes */
2161 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_appdata_id ]++;
2162 0 : return FD_QUIC_PARSE_FAIL;
2163 0 : }
2164 13347504 : ulong pn_offset = 1UL + FD_QUIC_CONN_ID_SZ;
2165 :
2166 13347504 : pkt->enc_level = fd_quic_enc_level_appdata_id;
2167 :
2168 13347504 : # if !FD_QUIC_DISABLE_CRYPTO
2169 13347504 : if( FD_UNLIKELY(
2170 13347504 : fd_quic_crypto_decrypt_hdr( cur_ptr, tot_sz,
2171 13347504 : pn_offset,
2172 13347504 : &conn->keys[3][0] ) != FD_QUIC_SUCCESS ) ) {
2173 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_crypto_decrypt_hdr failed" )) );
2174 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_appdata_id ]++;
2175 0 : return FD_QUIC_PARSE_FAIL;
2176 0 : }
2177 13347504 : # endif /* !FD_QUIC_DISABLE_CRYPTO */
2178 :
2179 13347504 : uint pkt_number_sz = fd_quic_h0_pkt_num_len( cur_ptr[0] ) + 1u;
2180 13347504 : uint key_phase = fd_quic_one_rtt_key_phase( cur_ptr[0] );
2181 :
2182 : /* reconstruct packet number */
2183 13347504 : ulong pktnum_comp = fd_quic_pktnum_decode( cur_ptr+pn_offset, pkt_number_sz );
2184 13347504 : ulong pkt_number = fd_quic_reconstruct_pkt_num( pktnum_comp, pkt_number_sz, conn->exp_pkt_number[2] );
2185 :
2186 : /* NOTE from rfc9002 s3
2187 : It is permitted for some packet numbers to never be used, leaving intentional gaps. */
2188 :
2189 : /* is current packet in the current key phase? */
2190 13347504 : int current_key_phase = conn->key_phase == key_phase;
2191 :
2192 13347504 : # if !FD_QUIC_DISABLE_CRYPTO
2193 : /* If the key phase bit flips, decrypt with the new pair of keys
2194 : instead. Note that the key phase bit is untrusted at this point. */
2195 13347504 : fd_quic_crypto_keys_t * keys = current_key_phase ? &conn->keys[3][0] : &conn->new_keys[0];
2196 :
2197 : /* this decrypts the header and payload */
2198 13347504 : if( FD_UNLIKELY(
2199 13347504 : fd_quic_crypto_decrypt( cur_ptr, tot_sz,
2200 13347504 : pn_offset,
2201 13347504 : pkt_number,
2202 13347504 : keys ) != FD_QUIC_SUCCESS ) ) {
2203 : /* remove connection from map, and insert into free list */
2204 0 : FD_DTRACE_PROBE_3( quic_err_decrypt_1rtt_pkt, pkt->ip4, conn->our_conn_id, pkt->pkt_number );
2205 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_appdata_id ]++;
2206 0 : return FD_QUIC_PARSE_FAIL;
2207 0 : }
2208 13347504 : # endif /* !FD_QUIC_DISABLE_CRYPTO */
2209 :
2210 : /* set packet number on the context */
2211 13347504 : pkt->pkt_number = pkt_number;
2212 :
2213 13347504 : if( !current_key_phase ) {
2214 : /* Decryption succeeded. Commit the key phase update and throw
2215 : away the old keys. (May cause a few decryption failures if old
2216 : packets get reordered past the current incoming packet) */
2217 96 : fd_quic_key_update_complete( conn );
2218 96 : }
2219 :
2220 13347504 : conn->peer_enc_level = (uchar)fd_uchar_max( conn->peer_enc_level, fd_quic_enc_level_appdata_id );
2221 :
2222 : /* handle frames */
2223 13347504 : ulong payload_off = pn_offset + pkt_number_sz;
2224 13347504 : uchar const * frame_ptr = cur_ptr + payload_off;
2225 13347504 : ulong payload_sz = tot_sz - pn_offset - pkt_number_sz; /* includes auth tag */
2226 13347504 : if( FD_UNLIKELY( payload_sz<FD_QUIC_CRYPTO_TAG_SZ ) ) return FD_QUIC_PARSE_FAIL;
2227 13347504 : ulong frame_sz = payload_sz - FD_QUIC_CRYPTO_TAG_SZ; /* total size of all frames in packet */
2228 26695014 : while( frame_sz != 0UL ) {
2229 13347510 : ulong rc = fd_quic_handle_v1_frame(
2230 13347510 : quic, conn, pkt, FD_QUIC_PKT_TYPE_ONE_RTT,
2231 13347510 : frame_ptr, frame_sz );
2232 13347510 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2233 0 : FD_DEBUG( FD_LOG_DEBUG(( "Failed to handle frame (1-RTT, frame=0x%02x)", frame_ptr[0] )) );
2234 0 : quic->metrics.frame_rx_err_cnt++;
2235 0 : return FD_QUIC_PARSE_FAIL;
2236 0 : }
2237 :
2238 13347510 : if( FD_UNLIKELY( rc == 0UL || rc > frame_sz ) ) {
2239 0 : FD_LOG_WARNING(( "fd_quic_handle_v1_frame returned invalid size" ));
2240 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
2241 0 : return FD_QUIC_PARSE_FAIL;
2242 0 : }
2243 :
2244 : /* next frame, and remaining size */
2245 13347510 : frame_ptr += rc;
2246 13347510 : frame_sz -= rc;
2247 13347510 : }
2248 :
2249 : /* update last activity */
2250 13347504 : conn->last_activity = fd_quic_get_state( quic )->now;
2251 :
2252 : /* update expected packet number */
2253 13347504 : conn->exp_pkt_number[2] = fd_ulong_max( conn->exp_pkt_number[2], pkt_number+1UL );
2254 :
2255 13347504 : return tot_sz;
2256 13347504 : }
2257 :
2258 :
2259 : /* process v1 quic packets
2260 : returns number of bytes consumed, or FD_QUIC_PARSE_FAIL upon error */
2261 : ulong
2262 : fd_quic_process_quic_packet_v1( fd_quic_t * quic,
2263 : fd_quic_pkt_t * pkt,
2264 : uchar * cur_ptr,
2265 13389327 : ulong cur_sz ) {
2266 :
2267 : /* bounds check packet size */
2268 13389327 : if( FD_UNLIKELY( cur_sz < FD_QUIC_SHORTEST_PKT ) ) {
2269 0 : quic->metrics.pkt_undersz_cnt++;
2270 0 : return FD_QUIC_PARSE_FAIL;
2271 0 : }
2272 13389327 : if( FD_UNLIKELY( cur_sz > 1500 ) ) {
2273 681 : quic->metrics.pkt_oversz_cnt++;
2274 681 : return FD_QUIC_PARSE_FAIL;
2275 681 : }
2276 :
2277 13388646 : fd_quic_state_t * state = fd_quic_get_state( quic );
2278 13388646 : fd_quic_conn_t * conn = NULL;
2279 :
2280 :
2281 : /* keep end */
2282 13388646 : uchar * orig_ptr = cur_ptr;
2283 :
2284 : /* No need for cur_sz check, since we are safe from the above check.
2285 : Decrementing cur_sz is done in the long header branch, the short header
2286 : branch parses the first byte again using the parser generator.
2287 : */
2288 13388646 : uchar hdr_form = fd_quic_h0_hdr_form( *cur_ptr );
2289 13388646 : ulong rc;
2290 :
2291 13388646 : if( hdr_form ) { /* long header */
2292 28251 : fd_quic_long_hdr_t long_hdr[1];
2293 28251 : rc = fd_quic_decode_long_hdr( long_hdr, cur_ptr+1, cur_sz-1 );
2294 28251 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2295 360 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_decode_long_hdr failed" )); )
2296 360 : quic->metrics.pkt_quic_hdr_err_cnt++;
2297 360 : return FD_QUIC_PARSE_FAIL;
2298 360 : }
2299 :
2300 27891 : fd_quic_conn_id_t dcid = fd_quic_conn_id_new( long_hdr->dst_conn_id, long_hdr->dst_conn_id_len );
2301 27891 : if( dcid.sz == FD_QUIC_CONN_ID_SZ ) {
2302 27030 : conn = fd_quic_conn_query( state->conn_map, fd_ulong_load_8( dcid.conn_id ) );
2303 27030 : }
2304 27891 : fd_quic_conn_id_t scid = fd_quic_conn_id_new( long_hdr->src_conn_id, long_hdr->src_conn_id_len );
2305 :
2306 27891 : uchar long_packet_type = fd_quic_h0_long_packet_type( *cur_ptr );
2307 :
2308 : /* encryption level matches that of TLS */
2309 27891 : pkt->enc_level = long_packet_type; /* V2 uses an indirect mapping */
2310 :
2311 : /* initialize packet number to unused value */
2312 27891 : pkt->pkt_number = FD_QUIC_PKT_NUM_UNUSED;
2313 :
2314 27891 : switch( long_packet_type ) {
2315 15522 : case FD_QUIC_PKT_TYPE_INITIAL:
2316 15522 : rc = fd_quic_handle_v1_initial( quic, &conn, pkt, &dcid, &scid, cur_ptr, cur_sz );
2317 15522 : if( FD_UNLIKELY( !conn ) ) {
2318 : /* FIXME not really a fail - Could be a retry */
2319 3399 : return FD_QUIC_PARSE_FAIL;
2320 3399 : }
2321 12123 : break;
2322 12144 : case FD_QUIC_PKT_TYPE_HANDSHAKE:
2323 12144 : rc = fd_quic_handle_v1_handshake( quic, conn, pkt, cur_ptr, cur_sz );
2324 12144 : break;
2325 147 : case FD_QUIC_PKT_TYPE_RETRY:
2326 147 : rc = fd_quic_handle_v1_retry( quic, conn, pkt, cur_ptr, cur_sz );
2327 147 : break;
2328 78 : case FD_QUIC_PKT_TYPE_ZERO_RTT:
2329 78 : rc = fd_quic_handle_v1_zero_rtt( quic, conn, pkt, cur_ptr, cur_sz );
2330 78 : break;
2331 27891 : }
2332 :
2333 24492 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2334 405 : FD_DEBUG( FD_LOG_DEBUG(( "Rejected packet (type=%d)", long_packet_type )); )
2335 405 : return FD_QUIC_PARSE_FAIL;
2336 405 : }
2337 :
2338 13360395 : } else { /* short header */
2339 : /* encryption level of short header packets is fd_quic_enc_level_appdata_id */
2340 13360395 : pkt->enc_level = fd_quic_enc_level_appdata_id;
2341 :
2342 : /* initialize packet number to unused value */
2343 13360395 : pkt->pkt_number = FD_QUIC_PKT_NUM_UNUSED;
2344 :
2345 : /* find connection id */
2346 13360395 : ulong dst_conn_id = fd_ulong_load_8( cur_ptr+1 );
2347 13360395 : conn = fd_quic_conn_query( state->conn_map, dst_conn_id );
2348 13360395 : rc = fd_quic_handle_v1_one_rtt( quic, conn, pkt, cur_ptr, cur_sz );
2349 13360395 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2350 12891 : return FD_QUIC_PARSE_FAIL;
2351 12891 : }
2352 13360395 : }
2353 :
2354 13371591 : if( FD_UNLIKELY( rc == 0UL ) ) {
2355 : /* this is an error because it causes infinite looping */
2356 0 : return FD_QUIC_PARSE_FAIL;
2357 0 : }
2358 13371591 : cur_ptr += rc;
2359 :
2360 : /* if we get here we parsed all the frames, so ack the packet */
2361 13371591 : int ack_type = fd_quic_lazy_ack_pkt( quic, conn, pkt );
2362 13371591 : quic->metrics.ack_tx[ ack_type ]++;
2363 :
2364 13371591 : if( pkt->rtt_ack_time ) {
2365 240379 : fd_quic_sample_rtt( conn, (long)pkt->rtt_ack_time, (long)pkt->rtt_ack_delay );
2366 240379 : }
2367 :
2368 : /* return bytes consumed */
2369 13371591 : return (ulong)( cur_ptr - orig_ptr );
2370 13371591 : }
2371 :
2372 :
2373 : /* version negotiation packet has version 0 */
2374 : static inline int
2375 23556 : is_version_invalid( fd_quic_t * quic, uint version ) {
2376 23556 : if( version == 0 ) {
2377 : /* TODO implement version negotiation */
2378 423 : quic->metrics.pkt_verneg_cnt++;
2379 423 : FD_DEBUG( FD_LOG_DEBUG(( "Got version negotiation packet" )) );
2380 423 : return 1;
2381 423 : }
2382 :
2383 : /* 0x?a?a?a?au is intended to force version negotiation
2384 : TODO implement */
2385 23133 : if( ( version & 0x0a0a0a0au ) == 0x0a0a0a0au ) {
2386 : /* at present, ignore */
2387 3 : quic->metrics.pkt_verneg_cnt++;
2388 3 : FD_DEBUG( FD_LOG_DEBUG(( "Got version negotiation packet (forced)" )) );
2389 3 : return 1;
2390 3 : }
2391 :
2392 23130 : if( version != 1 ) {
2393 : /* cannot interpret length, so discard entire packet */
2394 : /* TODO send version negotiation */
2395 219 : quic->metrics.pkt_verneg_cnt++;
2396 219 : FD_DEBUG( FD_LOG_DEBUG(( "Got unknown version QUIC packet" )) );
2397 219 : return 1;
2398 219 : }
2399 22911 : return 0;
2400 23130 : }
2401 :
2402 : static inline void
2403 : fd_quic_process_packet_impl( fd_quic_t * quic,
2404 : uchar * data,
2405 13390380 : ulong data_sz ) {
2406 :
2407 13390380 : fd_quic_state_t * state = fd_quic_get_state( quic );
2408 13390380 : state->now = fd_quic_now( quic );
2409 :
2410 13390380 : ulong rc = 0;
2411 :
2412 : /* holds the remainder of the packet*/
2413 13390380 : uchar * cur_ptr = data;
2414 13390380 : ulong cur_sz = data_sz;
2415 :
2416 13390380 : if( FD_UNLIKELY( data_sz > 0xffffu ) ) {
2417 0 : FD_DTRACE_PROBE( quic_err_rx_oversz );
2418 0 : quic->metrics.pkt_oversz_cnt++;
2419 0 : return;
2420 0 : }
2421 :
2422 13390380 : fd_quic_pkt_t pkt = { .datagram_sz = (uint)data_sz };
2423 :
2424 13390380 : pkt.rcv_time = state->now;
2425 13390380 : pkt.rtt_pkt_number = 0;
2426 13390380 : pkt.rtt_ack_time = 0;
2427 :
2428 : /* parse ip, udp */
2429 :
2430 13390380 : rc = fd_quic_decode_ip4( pkt.ip4, cur_ptr, cur_sz );
2431 13390380 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2432 : /* TODO count failure */
2433 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2434 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2435 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_decode_ip4 failed" )) );
2436 0 : return;
2437 0 : }
2438 :
2439 : /* check version, tot_len, protocol, checksum? */
2440 13390380 : if( FD_UNLIKELY( pkt.ip4->protocol != FD_IP4_HDR_PROTOCOL_UDP ) ) {
2441 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2442 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2443 0 : FD_DEBUG( FD_LOG_DEBUG(( "Packet is not UDP" )) );
2444 0 : return;
2445 0 : }
2446 :
2447 : /* verify ip4 packet isn't truncated
2448 : * AF_XDP can silently do this */
2449 13390380 : if( FD_UNLIKELY( pkt.ip4->net_tot_len > cur_sz ) ) {
2450 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2451 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2452 0 : FD_DEBUG( FD_LOG_DEBUG(( "IPv4 header indicates truncation" )) );
2453 0 : return;
2454 0 : }
2455 :
2456 : /* update pointer + size */
2457 13390380 : cur_ptr += rc;
2458 13390380 : cur_sz -= rc;
2459 :
2460 13390380 : rc = fd_quic_decode_udp( pkt.udp, cur_ptr, cur_sz );
2461 13390380 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2462 : /* TODO count failure */
2463 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2464 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2465 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_decode_udp failed" )) );
2466 0 : return;
2467 0 : }
2468 :
2469 : /* sanity check udp length */
2470 13390380 : if( FD_UNLIKELY( pkt.udp->net_len < sizeof(fd_udp_hdr_t) ||
2471 13390380 : pkt.udp->net_len > cur_sz ) ) {
2472 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2473 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2474 0 : FD_DEBUG( FD_LOG_DEBUG(( "UDP header indicates truncation" )) );
2475 0 : return;
2476 0 : }
2477 :
2478 : /* update pointer + size */
2479 13390380 : cur_ptr += rc;
2480 13390380 : cur_sz = pkt.udp->net_len - rc; /* replace with udp length */
2481 :
2482 : /* cur_ptr[0..cur_sz-1] should be payload */
2483 :
2484 : /* filter */
2485 : /* check dst eth address, ip address? probably not necessary */
2486 : /* usually look up port here, but let's jump straight into decoding as-if
2487 : quic */
2488 :
2489 : /* update counters */
2490 :
2491 : /* shortest valid quic payload? */
2492 13390380 : if( FD_UNLIKELY( cur_sz < FD_QUIC_SHORTEST_PKT ) ) {
2493 6639 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2494 6639 : quic->metrics.pkt_net_hdr_err_cnt++;
2495 6639 : FD_DEBUG( FD_LOG_DEBUG(( "Undersize QUIC packet" )) );
2496 6639 : return;
2497 6639 : }
2498 :
2499 : /* short packets don't have version */
2500 13383741 : int long_pkt = !!( (uint)cur_ptr[0] & 0x80u );
2501 :
2502 :
2503 13383741 : if( long_pkt ) {
2504 : /* version at offset 1..4 */
2505 23556 : uint version = fd_uint_bswap( FD_LOAD( uint, cur_ptr + 1 ) );
2506 : /* we only support version 1 */
2507 23556 : if( FD_UNLIKELY( is_version_invalid( quic, version ) ) ) {
2508 645 : return;
2509 645 : }
2510 :
2511 : /* multiple QUIC packets in a UDP packet */
2512 : /* shortest valid quic payload? */
2513 22911 : ulong pkt_idx;
2514 46998 : for( pkt_idx=0UL; pkt_idx<FD_QUIC_PKT_COALESCE_LIMIT; pkt_idx++ ) {
2515 : /* Are we done? Omit short packet handling that follows */
2516 46998 : if( FD_UNLIKELY( cur_sz < FD_QUIC_SHORTEST_PKT ) ) return;
2517 :
2518 : /* short packet requires different handling */
2519 28932 : int short_pkt = !( (uint)cur_ptr[0] & 0x80u );
2520 :
2521 28932 : if( FD_UNLIKELY( short_pkt ) ) break;
2522 :
2523 : /* check version */
2524 28932 : uint cur_version = fd_uint_bswap( FD_LOAD( uint, cur_ptr + 1 ) );
2525 :
2526 28932 : if( cur_version != version ) {
2527 : /* multiple versions in a single connection is a violation, and by
2528 : extension so is multiple versions in a single udp datagram
2529 : these are silently ignored
2530 :
2531 : for reference
2532 : all quic packets in a udp datagram must be for the same connection id
2533 : (section 12.2) and therefore the same connection
2534 : all packets on a connection must be of the same version (5.2) */
2535 0 : quic->metrics.pkt_quic_hdr_err_cnt++;
2536 0 : FD_DEBUG( FD_LOG_DEBUG(( "Mixed QUIC versions in packet" )) );
2537 0 : return;
2538 0 : }
2539 :
2540 28932 : rc = fd_quic_process_quic_packet_v1( quic, &pkt, cur_ptr, cur_sz );
2541 :
2542 : /* 0UL means no progress, so fail */
2543 28932 : if( FD_UNLIKELY( ( rc == FD_QUIC_PARSE_FAIL ) |
2544 28932 : ( rc == 0UL ) ) ) {
2545 4845 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_process_quic_packet_v1 failed (stuck=%d)", rc==0UL )) );
2546 4845 : return;
2547 4845 : }
2548 :
2549 24087 : if( FD_UNLIKELY( rc > cur_sz ) ) {
2550 0 : FD_DEBUG( FD_LOG_WARNING(( "fd_quic_process_quic_packet_v1 read too much" )) );
2551 0 : return;
2552 0 : }
2553 :
2554 : /* return code (rc) is the number of bytes consumed */
2555 24087 : cur_sz -= rc;
2556 24087 : cur_ptr += rc;
2557 24087 : }
2558 0 : if( pkt_idx==FD_QUIC_PKT_COALESCE_LIMIT ) {
2559 : /* too many packets in a single udp datagram */
2560 0 : return;
2561 0 : }
2562 0 : }
2563 :
2564 : /* above can drop out of loop if a short packet is detected */
2565 13360185 : if( FD_UNLIKELY( cur_sz < FD_QUIC_SHORTEST_PKT ) ) return;
2566 :
2567 : /* short header packet
2568 : only one_rtt packets currently have short headers */
2569 13360185 : fd_quic_process_quic_packet_v1( quic, &pkt, cur_ptr, cur_sz );
2570 13360185 : }
2571 :
2572 : void
2573 : fd_quic_process_packet( fd_quic_t * quic,
2574 : uchar * data,
2575 13390380 : ulong data_sz ) {
2576 13390380 : long dt = -fd_tickcount();
2577 13390380 : fd_quic_process_packet_impl( quic, data, data_sz );
2578 13390380 : dt += fd_tickcount();
2579 13390380 : fd_histf_sample( quic->metrics.receive_duration, (ulong)dt );
2580 :
2581 13390380 : quic->metrics.net_rx_byte_cnt += data_sz;
2582 13390380 : quic->metrics.net_rx_pkt_cnt++;
2583 13390380 : }
2584 :
2585 : /* main receive-side entry point */
2586 : int
2587 : fd_quic_aio_cb_receive( void * context,
2588 : fd_aio_pkt_info_t const * batch,
2589 : ulong batch_cnt,
2590 : ulong * opt_batch_idx,
2591 13371369 : int flush ) {
2592 13371369 : (void)flush;
2593 :
2594 13371369 : fd_quic_t * quic = context;
2595 :
2596 13371369 : FD_DEBUG(
2597 13371369 : fd_quic_state_t * state = fd_quic_get_state( quic );
2598 13371369 : static ulong t0 = 0;
2599 13371369 : static ulong t1 = 0;
2600 13371369 : t0 = state->now;
2601 13371369 : )
2602 :
2603 : /* this aio interface is configured as one-packet per buffer
2604 : so batch[0] refers to one buffer
2605 : as such, we simply forward each individual packet to a handling function */
2606 26742738 : for( ulong j = 0; j < batch_cnt; ++j ) {
2607 13371369 : fd_quic_process_packet( quic, batch[ j ].buf, batch[ j ].buf_sz );
2608 13371369 : }
2609 :
2610 : /* the assumption here at present is that any packet that could not be processed
2611 : is simply dropped
2612 : hence, all packets were consumed */
2613 13371369 : if( FD_LIKELY( opt_batch_idx ) ) {
2614 0 : *opt_batch_idx = batch_cnt;
2615 0 : }
2616 :
2617 13371369 : FD_DEBUG(
2618 13371369 : t1 = fd_quic_now( quic );
2619 13371369 : ulong delta = t1 - t0;
2620 13371369 : if( delta > (ulong)500e3 ) {
2621 13371369 : FD_LOG_WARNING(( "CALLBACK - took %lu t0: %lu t1: %lu batch_cnt: %lu", delta, t0, t1, (ulong)batch_cnt ));
2622 13371369 : }
2623 13371369 : )
2624 :
2625 13371369 : return FD_AIO_SUCCESS;
2626 13371369 : }
2627 :
2628 : void
2629 : fd_quic_tls_cb_alert( fd_quic_tls_hs_t * hs,
2630 : void * context,
2631 0 : int alert ) {
2632 0 : (void)hs;
2633 0 : fd_quic_conn_t * conn = (fd_quic_conn_t *)context;
2634 0 : (void)conn;
2635 0 : (void)alert;
2636 0 : FD_DEBUG( FD_LOG_DEBUG(( "TLS callback: %s", conn->server ? "SERVER" : "CLIENT" ));
2637 0 : FD_LOG_DEBUG(( "TLS alert: (%d-%s)", alert, fd_tls_alert_cstr( (uint)alert ) )); );
2638 :
2639 : /* TODO store alert to reply to peer */
2640 0 : }
2641 :
2642 : void
2643 : fd_quic_tls_cb_secret( fd_quic_tls_hs_t * hs,
2644 : void * context,
2645 24084 : fd_quic_tls_secret_t const * secret ) {
2646 :
2647 24084 : fd_quic_conn_t * conn = (fd_quic_conn_t*)context;
2648 24084 : fd_quic_t * quic = conn->quic;
2649 :
2650 : /* look up suite */
2651 : /* set secrets */
2652 24084 : FD_TEST( secret->enc_level < FD_QUIC_NUM_ENC_LEVELS );
2653 :
2654 24084 : uint enc_level = secret->enc_level;
2655 :
2656 24084 : fd_quic_crypto_secrets_t * crypto_secret = &conn->secrets;
2657 :
2658 24084 : memcpy( crypto_secret->secret[enc_level][0], secret->read_secret, FD_QUIC_SECRET_SZ );
2659 24084 : memcpy( crypto_secret->secret[enc_level][1], secret->write_secret, FD_QUIC_SECRET_SZ );
2660 :
2661 24084 : conn->keys_avail = fd_uint_set_bit( conn->keys_avail, (int)enc_level );
2662 :
2663 : /* gen local keys */
2664 24084 : fd_quic_gen_keys(
2665 24084 : &conn->keys[enc_level][0],
2666 24084 : conn->secrets.secret[enc_level][0] );
2667 :
2668 : /* gen peer keys */
2669 24084 : fd_quic_gen_keys(
2670 24084 : &conn->keys[enc_level][1],
2671 24084 : conn->secrets.secret[enc_level][1] );
2672 :
2673 24084 : if( enc_level==fd_quic_enc_level_appdata_id ) {
2674 12042 : fd_quic_key_update_derive( &conn->secrets, conn->new_keys );
2675 12042 : }
2676 :
2677 : /* Key logging */
2678 :
2679 24084 : void * keylog_ctx = quic->cb.quic_ctx;
2680 24084 : fd_quic_cb_tls_keylog_t keylog_fn = quic->cb.tls_keylog;
2681 24084 : if( FD_UNLIKELY( keylog_fn ) ) {
2682 : /* Ignore stdout, stderr, stdin */
2683 :
2684 24084 : uchar const * recv_secret = secret->read_secret;
2685 24084 : uchar const * send_secret = secret->write_secret;
2686 :
2687 24084 : uchar const * client_secret = hs->is_server ? recv_secret : send_secret;
2688 24084 : uchar const * server_secret = hs->is_server ? send_secret : recv_secret;
2689 :
2690 24084 : char buf[256];
2691 24084 : char * s;
2692 24084 : switch( enc_level ) {
2693 12042 : case FD_TLS_LEVEL_HANDSHAKE:
2694 12042 : /* 0 chars */ s = fd_cstr_init( buf );
2695 12042 : /* 0+32 chars */ s = fd_cstr_append_cstr( s, "CLIENT_HANDSHAKE_TRAFFIC_SECRET " );
2696 12042 : /* 32+64 chars */ s = fd_hex_encode( s, hs->hs.base.client_random, 32UL );
2697 12042 : /* 96+ 1 chars */ s = fd_cstr_append_char( s, ' ' );
2698 12042 : /* 97+64 chars */ s = fd_hex_encode( s, client_secret, 32UL );
2699 12042 : /* 161 chars */ fd_cstr_fini( s );
2700 12042 : keylog_fn( keylog_ctx, buf );
2701 12042 : /* 0 chars */ s = fd_cstr_init( buf );
2702 12042 : /* 0+32 chars */ s = fd_cstr_append_cstr( s, "SERVER_HANDSHAKE_TRAFFIC_SECRET " );
2703 12042 : /* 32+64 chars */ s = fd_hex_encode( s, hs->hs.base.client_random, 32UL );
2704 12042 : /* 96+ 1 chars */ s = fd_cstr_append_char( s, ' ' );
2705 12042 : /* 97+64 chars */ s = fd_hex_encode( s, server_secret, 32UL );
2706 12042 : /* 161 chars */ fd_cstr_fini( s );
2707 12042 : keylog_fn( keylog_ctx, buf );
2708 12042 : break;
2709 12042 : case FD_TLS_LEVEL_APPLICATION:
2710 12042 : /* 0 chars */ s = fd_cstr_init( buf );
2711 12042 : /* 0+24 chars */ s = fd_cstr_append_cstr( s, "CLIENT_TRAFFIC_SECRET_0 " );
2712 12042 : /* 24+64 chars */ s = fd_hex_encode( s, hs->hs.base.client_random, 32UL );
2713 12042 : /* 88+ 1 chars */ s = fd_cstr_append_char( s, ' ' );
2714 12042 : /* 89+64 chars */ s = fd_hex_encode( s, client_secret, 32UL );
2715 12042 : /* 153 chars */ fd_cstr_fini( s );
2716 12042 : keylog_fn( keylog_ctx, buf );
2717 12042 : /* 0 chars */ s = fd_cstr_init( buf );
2718 12042 : /* 0+24 chars */ s = fd_cstr_append_cstr( s, "SERVER_TRAFFIC_SECRET_0 " );
2719 12042 : /* 24+64 chars */ s = fd_hex_encode( s, hs->hs.base.client_random, 32UL );
2720 12042 : /* 88+ 1 chars */ s = fd_cstr_append_char( s, ' ' );
2721 12042 : /* 89+64 chars */ s = fd_hex_encode( s, server_secret, 32UL );
2722 12042 : /* 153 chars */ fd_cstr_fini( s );
2723 12042 : keylog_fn( keylog_ctx, buf );
2724 12042 : break;
2725 24084 : }
2726 24084 : }
2727 :
2728 24084 : }
2729 :
2730 : void
2731 : fd_quic_apply_peer_params( fd_quic_conn_t * conn,
2732 12048 : fd_quic_transport_params_t const * peer_tp ) {
2733 : /* flow control parameters */
2734 12048 : conn->tx_max_data = peer_tp->initial_max_data;
2735 12048 : conn->tx_initial_max_stream_data_uni= peer_tp->initial_max_stream_data_uni;
2736 :
2737 12048 : if( !conn->server ) {
2738 : /* verify retry_src_conn_id */
2739 6024 : uint retry_src_conn_id_sz = conn->retry_src_conn_id.sz;
2740 6024 : if( retry_src_conn_id_sz ) {
2741 3 : if( FD_UNLIKELY( !peer_tp->retry_source_connection_id_present
2742 3 : || peer_tp->retry_source_connection_id_len != retry_src_conn_id_sz
2743 3 : || 0 != memcmp( peer_tp->retry_source_connection_id,
2744 3 : conn->retry_src_conn_id.conn_id,
2745 3 : retry_src_conn_id_sz ) ) ) {
2746 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_TRANSPORT_PARAMETER_ERROR, __LINE__ );
2747 0 : return;
2748 0 : }
2749 6021 : } else {
2750 6021 : if( FD_UNLIKELY( peer_tp->retry_source_connection_id_present ) ) {
2751 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_TRANSPORT_PARAMETER_ERROR, __LINE__ );
2752 0 : return;
2753 0 : }
2754 6021 : }
2755 6024 : }
2756 :
2757 : /* max datagram size */
2758 12048 : ulong tx_max_datagram_sz = peer_tp->max_udp_payload_size;
2759 12048 : if( tx_max_datagram_sz < FD_QUIC_INITIAL_PAYLOAD_SZ_MAX ) {
2760 3 : tx_max_datagram_sz = FD_QUIC_INITIAL_PAYLOAD_SZ_MAX;
2761 3 : }
2762 12048 : if( tx_max_datagram_sz > FD_QUIC_INITIAL_PAYLOAD_SZ_MAX ) {
2763 12045 : tx_max_datagram_sz = FD_QUIC_INITIAL_PAYLOAD_SZ_MAX;
2764 12045 : }
2765 12048 : conn->tx_max_datagram_sz = (uint)tx_max_datagram_sz;
2766 :
2767 : /* initial max_streams */
2768 :
2769 12048 : if( conn->server ) {
2770 6024 : conn->tx_sup_stream_id = ( (ulong)peer_tp->initial_max_streams_uni << 2UL ) + FD_QUIC_STREAM_TYPE_UNI_SERVER;
2771 6024 : } else {
2772 6024 : conn->tx_sup_stream_id = ( (ulong)peer_tp->initial_max_streams_uni << 2UL ) + FD_QUIC_STREAM_TYPE_UNI_CLIENT;
2773 6024 : }
2774 :
2775 : /* set the max_idle_timeout to the min of our and peer max_idle_timeout */
2776 12048 : if( peer_tp->max_idle_timeout_ms ) {
2777 12042 : double peer_max_idle_timeout_us = (double)peer_tp->max_idle_timeout_ms * 1e3;
2778 12042 : ulong peer_max_idle_timeout_ticks = fd_quic_us_to_ticks( conn->quic, (ulong)peer_max_idle_timeout_us );
2779 12042 : conn->idle_timeout_ticks = fd_ulong_min( peer_max_idle_timeout_ticks, conn->idle_timeout_ticks );
2780 12042 : }
2781 :
2782 : /* set ack_delay_exponent so we can properly interpret peer's ack_delays
2783 : if unspecified, the value is 3 */
2784 12048 : ulong peer_ack_delay_exponent = fd_ulong_if(
2785 12048 : peer_tp->ack_delay_exponent_present,
2786 12048 : peer_tp->ack_delay_exponent,
2787 12048 : 3UL );
2788 :
2789 12048 : float tick_per_us = (float)conn->quic->config.tick_per_us;
2790 12048 : conn->peer_ack_delay_scale = (float)( 1UL << peer_ack_delay_exponent ) * tick_per_us;
2791 :
2792 : /* peer max ack delay in microseconds
2793 : peer_tp->max_ack_delay is milliseconds */
2794 12048 : float peer_max_ack_delay_us = (float)fd_ulong_if(
2795 12048 : peer_tp->max_ack_delay_present,
2796 12048 : peer_tp->max_ack_delay * 1000UL,
2797 12048 : 25000UL );
2798 12048 : conn->peer_max_ack_delay_ticks = peer_max_ack_delay_us * tick_per_us;
2799 :
2800 12048 : conn->transport_params_set = 1;
2801 12048 : }
2802 :
2803 : void
2804 : fd_quic_tls_cb_peer_params( void * context,
2805 : uchar const * peer_tp_enc,
2806 12045 : ulong peer_tp_enc_sz ) {
2807 12045 : fd_quic_conn_t * conn = (fd_quic_conn_t*)context;
2808 :
2809 : /* decode peer transport parameters */
2810 12045 : fd_quic_transport_params_t peer_tp[1] = {0};
2811 12045 : int rc = fd_quic_decode_transport_params( peer_tp, peer_tp_enc, peer_tp_enc_sz );
2812 12045 : if( FD_UNLIKELY( rc != 0 ) ) {
2813 0 : FD_DEBUG( FD_LOG_NOTICE(( "fd_quic_decode_transport_params failed" )); )
2814 :
2815 : /* failed to parse transport params */
2816 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_TRANSPORT_PARAMETER_ERROR, __LINE__ );
2817 0 : return;
2818 0 : }
2819 :
2820 12045 : fd_quic_apply_peer_params( conn, peer_tp );
2821 12045 : }
2822 :
2823 : void
2824 : fd_quic_tls_cb_handshake_complete( fd_quic_tls_hs_t * hs,
2825 12042 : void * context ) {
2826 12042 : (void)hs;
2827 12042 : fd_quic_conn_t * conn = (fd_quic_conn_t *)context;
2828 :
2829 : /* need to send quic handshake completion */
2830 12042 : switch( conn->state ) {
2831 0 : case FD_QUIC_CONN_STATE_ABORT:
2832 0 : case FD_QUIC_CONN_STATE_CLOSE_PENDING:
2833 0 : case FD_QUIC_CONN_STATE_DEAD:
2834 : /* ignore */
2835 0 : return;
2836 :
2837 12042 : case FD_QUIC_CONN_STATE_HANDSHAKE:
2838 12042 : if( FD_UNLIKELY( !conn->transport_params_set ) ) { /* unreachable */
2839 0 : FD_LOG_WARNING(( "Handshake marked as completed but transport params are not set. This is a bug!" ));
2840 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_INTERNAL_ERROR, __LINE__ );
2841 0 : return;
2842 0 : }
2843 12042 : conn->handshake_complete = 1;
2844 12042 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_HANDSHAKE_COMPLETE );
2845 12042 : return;
2846 :
2847 0 : default:
2848 0 : FD_LOG_WARNING(( "handshake in unexpected state: %u", conn->state ));
2849 12042 : }
2850 12042 : }
2851 :
2852 : static ulong
2853 : fd_quic_handle_crypto_frame( fd_quic_frame_ctx_t * context,
2854 : fd_quic_crypto_frame_t * crypto,
2855 : uchar const * p,
2856 42156 : ulong p_sz ) {
2857 : /* determine whether any of the data was already provided */
2858 42156 : fd_quic_conn_t * conn = context->conn;
2859 42156 : fd_quic_tls_hs_t * tls_hs = conn->tls_hs;
2860 42156 : uint enc_level = context->pkt->enc_level;
2861 :
2862 : /* offset expected */
2863 42156 : ulong rcv_off = crypto->offset; /* in [0,2^62-1] */
2864 42156 : ulong rcv_sz = crypto->length; /* in [0,2^62-1] */
2865 42156 : ulong rcv_hi = rcv_off + rcv_sz; /* in [0,2^63-1] */
2866 :
2867 42156 : if( FD_UNLIKELY( rcv_sz > p_sz ) ) {
2868 6 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
2869 6 : return FD_QUIC_PARSE_FAIL;
2870 6 : }
2871 :
2872 42150 : if( !tls_hs ) {
2873 : /* Handshake already completed. Ignore frame */
2874 : /* TODO consider aborting conn if too many unsolicited crypto frames arrive */
2875 0 : return rcv_sz;
2876 0 : }
2877 :
2878 42150 : if( enc_level < tls_hs->rx_enc_level ) {
2879 0 : return rcv_sz;
2880 0 : }
2881 :
2882 42150 : if( enc_level > tls_hs->rx_enc_level ) {
2883 : /* Discard data from any previous handshake level. Currently only
2884 : happens at the Initial->Handshake encryption level change. */
2885 12042 : tls_hs->rx_enc_level = (uchar)enc_level;
2886 12042 : tls_hs->rx_off = 0;
2887 12042 : tls_hs->rx_sz = 0;
2888 12042 : }
2889 :
2890 42150 : if( rcv_off > tls_hs->rx_sz ) {
2891 0 : context->pkt->ack_flag |= ACK_FLAG_CANCEL;
2892 0 : return rcv_sz;
2893 0 : }
2894 :
2895 42150 : if( rcv_hi < tls_hs->rx_off ) {
2896 0 : return rcv_sz;
2897 0 : }
2898 :
2899 42150 : if( rcv_hi > FD_QUIC_TLS_RX_DATA_SZ ) {
2900 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_CRYPTO_BUFFER_EXCEEDED, __LINE__ );
2901 0 : return FD_QUIC_PARSE_FAIL;
2902 0 : }
2903 :
2904 42150 : tls_hs->rx_sz = (ushort)rcv_hi;
2905 42150 : fd_memcpy( tls_hs->rx_hs_buf + rcv_off, p, rcv_sz );
2906 :
2907 42150 : int provide_rc = fd_quic_tls_process( conn->tls_hs );
2908 42150 : if( provide_rc == FD_QUIC_FAILED ) {
2909 : /* if TLS fails, ABORT connection */
2910 :
2911 : /* if TLS returns an error, we present that as reason:
2912 : FD_QUIC_CONN_REASON_CRYPTO_BASE + tls-alert
2913 : otherwise, send INTERNAL_ERROR */
2914 3 : uint alert = conn->tls_hs->alert;
2915 3 : uint reason = conn->tls_hs->hs.base.reason;
2916 3 : FD_DTRACE_PROBE_3( quic_handle_crypto_frame, conn->our_conn_id, alert, reason );
2917 3 : if( alert == 0u ) {
2918 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_INTERNAL_ERROR, __LINE__ );
2919 3 : } else {
2920 3 : FD_DEBUG(
2921 3 : FD_LOG_DEBUG(( "QUIC TLS handshake failed (alert %u-%s; reason %u-%s)",
2922 3 : alert, fd_tls_alert_cstr( alert ),
2923 3 : reason, fd_tls_reason_cstr( reason ) ));
2924 3 : )
2925 3 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_CRYPTO_BASE + alert, __LINE__ );
2926 3 : }
2927 3 : return FD_QUIC_PARSE_FAIL;
2928 3 : }
2929 :
2930 42147 : return rcv_sz;
2931 42150 : }
2932 :
2933 : static int
2934 : fd_quic_svc_poll( fd_quic_t * quic,
2935 : fd_quic_conn_t * conn,
2936 13356867 : ulong now ) {
2937 13356867 : fd_quic_state_t * state = fd_quic_get_state( quic );
2938 13356867 : if( FD_UNLIKELY( conn->state == FD_QUIC_CONN_STATE_INVALID ) ) {
2939 : /* connection shouldn't have been scheduled,
2940 : and is now removed, so just continue */
2941 0 : FD_LOG_ERR(( "Invalid conn in schedule (svc_type=%u)", conn->svc_type ));
2942 0 : return 1;
2943 0 : }
2944 :
2945 : //FD_DEBUG( FD_LOG_DEBUG(( "svc_poll conn=%p svc_type=%u", (void *)conn, conn->svc_type )); )
2946 13356867 : conn->svc_type = UINT_MAX;
2947 13356867 : conn->svc_time = LONG_MAX;
2948 :
2949 13356867 : if( FD_UNLIKELY( now >= conn->last_activity + ( conn->idle_timeout_ticks / 2 ) ) ) {
2950 2115 : if( FD_UNLIKELY( now >= conn->last_activity + conn->idle_timeout_ticks ) ) {
2951 2100 : if( FD_LIKELY( conn->state != FD_QUIC_CONN_STATE_DEAD ) ) {
2952 : /* rfc9000 10.1 Idle Timeout
2953 : "... the connection is silently closed and its state is discarded
2954 : when it remains idle for longer than the minimum of the
2955 : max_idle_timeout value advertised by both endpoints." */
2956 2100 : FD_DEBUG( FD_LOG_WARNING(("%s conn %p conn_idx: %u closing due to idle timeout (%g ms)",
2957 2100 : conn->server?"SERVER":"CLIENT",
2958 2100 : (void *)conn, conn->conn_idx, (double)fd_quic_ticks_to_us(conn->idle_timeout_ticks) / 1e3 )); )
2959 :
2960 2100 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_DEAD );
2961 2100 : quic->metrics.conn_timeout_cnt++;
2962 2100 : }
2963 2100 : } else if( quic->config.keep_alive & !!(conn->let_die_ticks > now) ) {
2964 : /* send PING */
2965 3 : if( !( conn->flags & FD_QUIC_CONN_FLAGS_PING ) ) {
2966 3 : conn->flags |= FD_QUIC_CONN_FLAGS_PING;
2967 3 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING; /* update to be sent in next packet */
2968 3 : }
2969 3 : }
2970 2115 : }
2971 :
2972 13356867 : if( FD_UNLIKELY( conn->state == FD_QUIC_CONN_STATE_DEAD ) ) {
2973 2100 : fd_quic_cb_conn_final( quic, conn ); /* inform user before freeing */
2974 2100 : fd_quic_conn_free( quic, conn );
2975 2100 : return 1; /* do NOT reschedule freed connection */
2976 2100 : }
2977 :
2978 : /* state cannot be DEAD here */
2979 13354767 : fd_quic_conn_service( quic, conn, now );
2980 :
2981 : /* dead? don't reinsert, just clean up */
2982 13354767 : switch( conn->state ) {
2983 0 : case FD_QUIC_CONN_STATE_INVALID:
2984 : /* skip entirely */
2985 0 : break;
2986 12108 : case FD_QUIC_CONN_STATE_DEAD:
2987 12108 : fd_quic_cb_conn_final( quic, conn ); /* inform user before freeing */
2988 12108 : fd_quic_conn_free( quic, conn );
2989 12108 : break;
2990 13342659 : default:
2991 13342659 : fd_quic_svc_schedule( state, conn, FD_QUIC_SVC_WAIT );
2992 13342659 : break;
2993 13354767 : }
2994 :
2995 13354767 : return 1;
2996 13354767 : }
2997 :
2998 : static int
2999 : fd_quic_svc_poll_head( fd_quic_t * quic,
3000 : uint svc_type,
3001 192546080 : ulong now ) {
3002 192546080 : fd_quic_state_t * state = fd_quic_get_state( quic );
3003 :
3004 : /* Peek head of queue */
3005 192546080 : fd_quic_svc_queue_t * queue = &state->svc_queue[ svc_type ];
3006 192546080 : if( queue->head==UINT_MAX ) return 0;
3007 158548572 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, queue->head );
3008 158548572 : if( conn->svc_time > now ) return 0;
3009 :
3010 : /* Remove head of queue */
3011 3075 : uint prev_idx = conn->svc_prev;
3012 3075 : fd_quic_conn_t * prev_ele = fd_quic_conn_at_idx( state, prev_idx );
3013 3075 : *fd_ptr_if( prev_idx!=UINT_MAX, &prev_ele->svc_next, &queue->tail ) = UINT_MAX;
3014 3075 : queue->head = prev_idx;
3015 :
3016 3075 : return fd_quic_svc_poll( quic, conn, now );
3017 158548572 : }
3018 :
3019 : static int
3020 : fd_quic_svc_poll_tail( fd_quic_t * quic,
3021 : uint svc_type,
3022 96273040 : ulong now ) {
3023 96273040 : fd_quic_state_t * state = fd_quic_get_state( quic );
3024 :
3025 : /* Peek tail of queue */
3026 96273040 : fd_quic_svc_queue_t * queue = &state->svc_queue[ svc_type ];
3027 96273040 : if( queue->tail==UINT_MAX ) return 0;
3028 13353792 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, queue->tail );
3029 13353792 : if( conn->svc_time > now ) return 0;
3030 :
3031 : /* Remove tail of queue */
3032 13353792 : uint next_idx = conn->svc_next;
3033 13353792 : fd_quic_conn_t * next_ele = fd_quic_conn_at_idx( state, next_idx );
3034 13353792 : *fd_ptr_if( next_idx!=UINT_MAX, &next_ele->svc_prev, &queue->head ) = UINT_MAX;
3035 13353792 : queue->tail = next_idx;
3036 :
3037 13353792 : return fd_quic_svc_poll( quic, conn, now );
3038 13353792 : }
3039 :
3040 : int
3041 96273040 : fd_quic_service( fd_quic_t * quic ) {
3042 96273040 : fd_quic_state_t * state = fd_quic_get_state( quic );
3043 :
3044 96273040 : ulong now = fd_quic_now( quic );
3045 96273040 : state->now = now;
3046 :
3047 96273040 : long now_ticks = fd_tickcount();
3048 :
3049 96273040 : int cnt = 0;
3050 96273040 : cnt += fd_quic_svc_poll_tail( quic, FD_QUIC_SVC_INSTANT, now );
3051 96273040 : cnt += fd_quic_svc_poll_head( quic, FD_QUIC_SVC_ACK_TX, now );
3052 96273040 : cnt += fd_quic_svc_poll_head( quic, FD_QUIC_SVC_WAIT, now );
3053 :
3054 96273040 : long delta_ticks = fd_tickcount() - now_ticks;
3055 :
3056 96273040 : fd_histf_sample( quic->metrics.service_duration, (ulong)delta_ticks );
3057 :
3058 96273040 : return cnt;
3059 96273040 : }
3060 :
3061 : static inline ulong
3062 13377552 : fd_quic_conn_tx_buf_remaining( fd_quic_conn_t * conn ) {
3063 13377552 : return (ulong)( sizeof( conn->tx_buf_conn ) - (ulong)( conn->tx_ptr - conn->tx_buf_conn ) );
3064 13377552 : }
3065 :
3066 : /* attempt to transmit buffered data
3067 :
3068 : prior to call, conn->tx_ptr points to the first free byte in tx_buf
3069 : the data in tx_buf..tx_ptr is prepended by networking headers
3070 : and put on the wire
3071 :
3072 : returns 0 if successful, or 1 otherwise */
3073 : uint
3074 : fd_quic_tx_buffered_raw(
3075 : fd_quic_t * quic,
3076 : uchar ** tx_ptr_ptr,
3077 : uchar * tx_buf,
3078 : ushort * ipv4_id,
3079 : uint dst_ipv4_addr,
3080 : ushort dst_udp_port,
3081 : uint src_ipv4_addr,
3082 : ushort src_udp_port
3083 26708292 : ) {
3084 :
3085 : /* TODO leave space at front of tx_buf for header
3086 : then encode directly into it to avoid 1 copy */
3087 26708292 : uchar *tx_ptr = *tx_ptr_ptr;
3088 26708292 : long payload_sz = tx_ptr - tx_buf;
3089 :
3090 : /* nothing to do */
3091 26708292 : if( FD_UNLIKELY( payload_sz<=0L ) ) {
3092 13336656 : return 0u;
3093 13336656 : }
3094 :
3095 13371636 : fd_quic_config_t * config = &quic->config;
3096 13371636 : fd_quic_state_t * state = fd_quic_get_state( quic );
3097 :
3098 13371636 : uchar * const crypt_scratch = state->crypt_scratch;
3099 :
3100 13371636 : uchar * cur_ptr = state->crypt_scratch;
3101 13371636 : ulong cur_sz = sizeof( state->crypt_scratch );
3102 :
3103 : /* TODO much of this may be prepared ahead of time */
3104 13371636 : fd_quic_pkt_t pkt;
3105 :
3106 13371636 : pkt.ip4->verihl = FD_IP4_VERIHL(4,5);
3107 13371636 : pkt.ip4->tos = (uchar)(config->net.dscp << 2); /* could make this per-connection or per-stream */
3108 13371636 : pkt.ip4->net_tot_len = (ushort)( 20 + 8 + payload_sz );
3109 13371636 : pkt.ip4->net_id = *ipv4_id;
3110 13371636 : pkt.ip4->net_frag_off = 0x4000u; /* don't fragment */
3111 13371636 : pkt.ip4->ttl = 64; /* TODO make configurable */
3112 13371636 : pkt.ip4->protocol = FD_IP4_HDR_PROTOCOL_UDP;
3113 13371636 : pkt.ip4->check = 0;
3114 13371636 : pkt.ip4->saddr = src_ipv4_addr;
3115 13371636 : pkt.ip4->daddr = dst_ipv4_addr;
3116 13371636 : pkt.udp->net_sport = src_udp_port;
3117 13371636 : pkt.udp->net_dport = dst_udp_port;
3118 13371636 : pkt.udp->net_len = (ushort)( 8 + payload_sz );
3119 13371636 : pkt.udp->check = 0x0000;
3120 13371636 : *ipv4_id = (ushort)( *ipv4_id + 1 );
3121 :
3122 13371636 : ulong rc = fd_quic_encode_ip4( cur_ptr, cur_sz, pkt.ip4 );
3123 13371636 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
3124 0 : FD_LOG_ERR(( "fd_quic_encode_ip4 failed with buffer overrun" ));
3125 0 : }
3126 :
3127 : /* Compute checksum over network byte order header */
3128 13371636 : fd_ip4_hdr_t * ip4_encoded = (fd_ip4_hdr_t *)fd_type_pun( cur_ptr );
3129 13371636 : ip4_encoded->check = (ushort)fd_ip4_hdr_check_fast( ip4_encoded );
3130 :
3131 13371636 : cur_ptr += rc;
3132 13371636 : cur_sz -= rc;
3133 :
3134 13371636 : rc = fd_quic_encode_udp( cur_ptr, cur_sz, pkt.udp );
3135 13371636 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
3136 0 : FD_LOG_ERR(( "fd_quic_encode_udp failed with buffer overrun" ));
3137 0 : }
3138 :
3139 13371636 : cur_ptr += rc;
3140 13371636 : cur_sz -= rc;
3141 :
3142 : /* need enough space for payload */
3143 13371636 : if( FD_UNLIKELY( (ulong)payload_sz > cur_sz ) ) {
3144 0 : FD_LOG_WARNING(( "%s : payload too big for buffer", __func__ ));
3145 :
3146 : /* reset buffer, since we can't use its contents */
3147 0 : *tx_ptr_ptr = tx_buf;
3148 0 : return FD_QUIC_FAILED;
3149 0 : }
3150 13371636 : fd_memcpy( cur_ptr, tx_buf, (ulong)payload_sz );
3151 :
3152 13371636 : cur_ptr += (ulong)payload_sz;
3153 13371636 : cur_sz -= (ulong)payload_sz;
3154 :
3155 13371636 : fd_aio_pkt_info_t aio_buf = { .buf = crypt_scratch, .buf_sz = (ushort)( cur_ptr - crypt_scratch ) };
3156 13371636 : int aio_rc = fd_aio_send( &quic->aio_tx, &aio_buf, 1, NULL, 1 );
3157 13371636 : if( aio_rc == FD_AIO_ERR_AGAIN ) {
3158 : /* transient condition - try later */
3159 0 : return FD_QUIC_FAILED;
3160 13371636 : } else if( aio_rc != FD_AIO_SUCCESS ) {
3161 0 : FD_LOG_WARNING(( "Fatal error reported by aio peer" ));
3162 : /* fallthrough to reset buffer */
3163 0 : }
3164 :
3165 : /* after send, reset tx_ptr and tx_sz */
3166 13371636 : *tx_ptr_ptr = tx_buf;
3167 :
3168 13371636 : quic->metrics.net_tx_pkt_cnt += aio_rc==FD_AIO_SUCCESS;
3169 13371636 : if( FD_LIKELY( aio_rc==FD_AIO_SUCCESS ) ) {
3170 13371636 : quic->metrics.net_tx_byte_cnt += aio_buf.buf_sz;
3171 13371636 : }
3172 :
3173 13371636 : return FD_QUIC_SUCCESS; /* success */
3174 13371636 : }
3175 :
3176 : uint
3177 : fd_quic_tx_buffered( fd_quic_t * quic,
3178 26708187 : fd_quic_conn_t * conn ) {
3179 26708187 : fd_quic_net_endpoint_t const * endpoint = conn->peer;
3180 26708187 : return fd_quic_tx_buffered_raw(
3181 26708187 : quic,
3182 26708187 : &conn->tx_ptr,
3183 26708187 : conn->tx_buf_conn,
3184 26708187 : &conn->ipv4_id,
3185 26708187 : endpoint->ip_addr,
3186 26708187 : endpoint->udp_port,
3187 26708187 : conn->host.ip_addr,
3188 26708187 : conn->host.udp_port);
3189 26708187 : }
3190 :
3191 : static inline int
3192 : fd_quic_conn_can_acquire_pkt_meta( fd_quic_conn_t * conn,
3193 26265556 : fd_quic_pkt_meta_tracker_t * tracker ) {
3194 26265556 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
3195 26265556 : fd_quic_metrics_t * metrics = &conn->quic->metrics;
3196 :
3197 26265556 : ulong pool_free = fd_quic_pkt_meta_pool_free( tracker->pool );
3198 26265556 : if( !pool_free || conn->used_pkt_meta >= state->max_inflight_frame_cnt_conn ) {
3199 48 : if( !pool_free ) {
3200 45 : metrics->frame_tx_alloc_cnt[FD_METRICS_ENUM_FRAME_TX_ALLOC_RESULT_V_FAIL_EMPTY_POOL_IDX]++;
3201 45 : } else {
3202 3 : metrics->frame_tx_alloc_cnt[FD_METRICS_ENUM_FRAME_TX_ALLOC_RESULT_V_FAIL_CONN_MAX_IDX]++;
3203 3 : }
3204 48 : return 0;
3205 48 : }
3206 26265508 : metrics->frame_tx_alloc_cnt[FD_METRICS_ENUM_FRAME_TX_ALLOC_RESULT_V_SUCCESS_IDX]++;
3207 :
3208 26265508 : return 1;
3209 26265556 : }
3210 :
3211 : /* fd_quic_gen_frame_store_pkt_meta stores a pkt_meta into tracker.
3212 : Value and type take the passed args; all other fields are copied
3213 : from pkt_meta_tmpl. Returns 1 if successful, 0 if not.
3214 : Failure reasons include empty pkt_meta pool, or this conn reached
3215 : its pkt_meta limit. Theoretically only need latter, but let's be safe! */
3216 : static inline int
3217 : fd_quic_gen_frame_store_pkt_meta( const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3218 : uchar type,
3219 : fd_quic_pkt_meta_value_t value,
3220 : fd_quic_pkt_meta_tracker_t * tracker,
3221 13141883 : fd_quic_conn_t * conn ) {
3222 13141883 : if( !fd_quic_conn_can_acquire_pkt_meta( conn, tracker ) ) return 0;
3223 :
3224 13141838 : conn->used_pkt_meta++;
3225 13141838 : fd_quic_pkt_meta_t * pkt_meta = fd_quic_pkt_meta_pool_ele_acquire( tracker->pool );
3226 13141838 : *pkt_meta = *pkt_meta_tmpl;
3227 13141838 : FD_QUIC_PKT_META_SET_TYPE( pkt_meta, type );
3228 13141838 : pkt_meta->val = value;
3229 13141838 : fd_quic_pkt_meta_insert( &tracker->sent_pkt_metas[pkt_meta->enc_level], pkt_meta, tracker->pool );
3230 13141838 : return 1;
3231 13141883 : }
3232 :
3233 : static ulong
3234 : fd_quic_gen_close_frame( fd_quic_conn_t * conn,
3235 : uchar * payload_ptr,
3236 : uchar * payload_end,
3237 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3238 18108 : fd_quic_pkt_meta_tracker_t * tracker ) {
3239 :
3240 18108 : if( conn->flags & FD_QUIC_CONN_FLAGS_CLOSE_SENT ) return 0UL;
3241 12108 : conn->flags |= FD_QUIC_CONN_FLAGS_CLOSE_SENT;
3242 :
3243 12108 : ulong frame_sz;
3244 12108 : if( conn->reason != 0u || conn->state == FD_QUIC_CONN_STATE_PEER_CLOSE ) {
3245 6084 : fd_quic_conn_close_0_frame_t frame = {
3246 6084 : .error_code = conn->reason,
3247 6084 : .frame_type = 0u, /* we do not know the frame in question */
3248 6084 : .reason_phrase_length = 0u /* no reason phrase */
3249 6084 : };
3250 6084 : frame_sz = fd_quic_encode_conn_close_0_frame( payload_ptr,
3251 6084 : (ulong)( payload_end - payload_ptr ),
3252 6084 : &frame );
3253 6084 : } else {
3254 6024 : fd_quic_conn_close_1_frame_t frame = {
3255 6024 : .error_code = conn->app_reason,
3256 6024 : .reason_phrase_length = 0u /* no reason phrase */
3257 6024 : };
3258 6024 : frame_sz = fd_quic_encode_conn_close_1_frame( payload_ptr,
3259 6024 : (ulong)( payload_end - payload_ptr ),
3260 6024 : &frame );
3261 6024 : }
3262 :
3263 12108 : if( FD_UNLIKELY( frame_sz == FD_QUIC_PARSE_FAIL ) ) {
3264 0 : FD_LOG_WARNING(( "fd_quic_encode_conn_close_frame failed, but space should have been available" ));
3265 0 : return 0UL;
3266 0 : }
3267 :
3268 : /* create and save pkt_meta, return 0 if fail */
3269 12108 : if( !fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3270 12108 : FD_QUIC_PKT_META_TYPE_CLOSE,
3271 12108 : (fd_quic_pkt_meta_value_t){0}, /* value doesn't matter */
3272 12108 : tracker,
3273 12108 : conn )) return 0UL;
3274 :
3275 12108 : return frame_sz;
3276 12108 : }
3277 :
3278 : static uchar *
3279 : fd_quic_gen_handshake_frames( fd_quic_conn_t * conn,
3280 : uchar * payload_ptr,
3281 : uchar * payload_end,
3282 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3283 13365504 : fd_quic_pkt_meta_tracker_t * tracker ) {
3284 13365504 : uint enc_level = pkt_meta_tmpl->enc_level;
3285 13365504 : fd_quic_tls_hs_data_t * hs_data = fd_quic_tls_get_hs_data( conn->tls_hs, enc_level );
3286 13365504 : if( !hs_data ) return payload_ptr;
3287 :
3288 : /* confirm we have pkt_meta space */
3289 24087 : if( !fd_quic_conn_can_acquire_pkt_meta( conn, tracker ) ) return payload_ptr;
3290 :
3291 24087 : ulong hs_offset = 0; /* offset within the current hs_data */
3292 24087 : ulong sent_offset = conn->hs_sent_bytes[enc_level];
3293 24087 : ulong ackd_offset = conn->hs_ackd_bytes[enc_level];
3294 : /* offset within stream */
3295 24087 : ulong offset = fd_ulong_max( sent_offset, ackd_offset );
3296 :
3297 : /* track pkt_meta values */
3298 24087 : ulong offset_lo = offset;
3299 24087 : ulong offset_hi = offset;
3300 :
3301 108387 : while( hs_data ) {
3302 : /* skip data we've sent */
3303 84300 : if( hs_data->offset + hs_data->data_sz <= offset ) {
3304 42150 : hs_data = fd_quic_tls_get_next_hs_data( conn->tls_hs, hs_data );
3305 42150 : continue;
3306 42150 : }
3307 :
3308 42150 : if( FD_UNLIKELY( hs_data->offset > offset ) ) {
3309 : /* we have a gap - this shouldn't happen */
3310 0 : FD_LOG_WARNING(( "%s - gap in TLS handshake data", __func__ ));
3311 : /* TODO should probably tear down connection */
3312 0 : break;
3313 0 : }
3314 :
3315 : /* encode hs_data into frame */
3316 42150 : hs_offset = offset - hs_data->offset;
3317 :
3318 : /* handshake data to send */
3319 42150 : uchar const * cur_data = hs_data->data + hs_offset;
3320 42150 : ulong cur_data_sz = hs_data->data_sz - hs_offset;
3321 :
3322 : /* 9 bytes header + cur_data_sz */
3323 42150 : if( payload_ptr + 9UL + cur_data_sz > payload_end ) break;
3324 : /* FIXME reduce cur_data_sz if it doesn't fit in frame
3325 : Practically don't need to, because fd_tls generates a small amount of data */
3326 :
3327 42150 : payload_ptr[0] = 0x06; /* CRYPTO frame */
3328 42150 : uint offset_varint = 0x80U | ( fd_uint_bswap( (uint)offset & 0x3fffffffU ) );
3329 42150 : uint length_varint = 0x80U | ( fd_uint_bswap( (uint)cur_data_sz & 0x3fffffffU ) );
3330 42150 : FD_STORE( uint, payload_ptr+1, offset_varint );
3331 42150 : FD_STORE( uint, payload_ptr+5, length_varint );
3332 42150 : payload_ptr += 9;
3333 :
3334 42150 : fd_memcpy( payload_ptr, cur_data, cur_data_sz );
3335 42150 : payload_ptr += cur_data_sz;
3336 :
3337 : /* update pkt_meta values */
3338 42150 : offset_hi += cur_data_sz;
3339 :
3340 : /* move to next hs_data */
3341 42150 : offset += cur_data_sz;
3342 42150 : conn->hs_sent_bytes[enc_level] += cur_data_sz;
3343 :
3344 : /* TODO load more hs_data into a crypto frame, if available
3345 : currently tricky, because encode_crypto_frame copies payload */
3346 42150 : }
3347 :
3348 : /* update packet meta */
3349 24087 : if( offset_hi > offset_lo ) {
3350 24087 : fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3351 24087 : FD_QUIC_PKT_META_TYPE_HS_DATA,
3352 24087 : (fd_quic_pkt_meta_value_t){
3353 24087 : .range = {
3354 24087 : .offset_lo = offset_lo,
3355 24087 : .offset_hi = offset_hi
3356 24087 : }
3357 24087 : },
3358 24087 : tracker,
3359 24087 : conn );
3360 24087 : }
3361 :
3362 24087 : return payload_ptr;
3363 24087 : }
3364 :
3365 : static ulong
3366 : fd_quic_gen_handshake_done_frame( fd_quic_conn_t * conn,
3367 : uchar * payload_ptr,
3368 : uchar * payload_end,
3369 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3370 13341417 : fd_quic_pkt_meta_tracker_t * tracker ) {
3371 13341417 : FD_DTRACE_PROBE_1( quic_gen_handshake_done_frame, conn->our_conn_id );
3372 13341417 : if( conn->handshake_done_send==0 ) return 0UL;
3373 6021 : conn->handshake_done_send = 0;
3374 6021 : if( FD_UNLIKELY( conn->handshake_done_ackd ) ) return 0UL;
3375 6021 : if( FD_UNLIKELY( payload_ptr >= payload_end ) ) return 0UL;
3376 : /* send handshake done frame */
3377 6021 : payload_ptr[0] = 0x1E;
3378 :
3379 : /* record the send for retx */
3380 6021 : if( !fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3381 6021 : FD_QUIC_PKT_META_TYPE_HS_DONE,
3382 6021 : (fd_quic_pkt_meta_value_t){0}, /* value doesn't matter */
3383 6021 : tracker,
3384 6021 : conn) ) return 0UL;
3385 :
3386 6021 : return 1UL;
3387 6021 : }
3388 :
3389 : static ulong
3390 : fd_quic_gen_max_data_frame( fd_quic_conn_t * conn,
3391 : uchar * payload_ptr,
3392 : uchar * payload_end,
3393 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3394 12177 : fd_quic_pkt_meta_tracker_t * tracker ) {
3395 12177 : fd_quic_conn_stream_rx_t * srx = conn->srx;
3396 :
3397 12177 : if( !( conn->flags & FD_QUIC_CONN_FLAGS_MAX_DATA ) ) return 0UL;
3398 0 : if( srx->rx_max_data <= srx->rx_max_data_ackd ) return 0UL; /* peer would ignore anyway */
3399 :
3400 : /* send max_data frame */
3401 0 : fd_quic_max_data_frame_t frame = { .max_data = srx->rx_max_data };
3402 :
3403 : /* attempt to write into buffer */
3404 0 : ulong frame_sz = fd_quic_encode_max_data_frame( payload_ptr,
3405 0 : (ulong)( payload_end - payload_ptr ),
3406 0 : &frame );
3407 0 : if( FD_UNLIKELY( frame_sz==FD_QUIC_ENCODE_FAIL ) ) return 0UL;
3408 :
3409 : /* acquire and set a pkt_meta, return 0 if not successful */
3410 0 : if( !fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3411 0 : FD_QUIC_PKT_META_TYPE_MAX_DATA,
3412 0 : (fd_quic_pkt_meta_value_t){
3413 0 : .scalar = srx->rx_max_data
3414 0 : },
3415 0 : tracker,
3416 0 : conn ) ) return 0UL;
3417 :
3418 0 : conn->upd_pkt_number = pkt_meta_tmpl->key.pkt_num;
3419 0 : return frame_sz;
3420 0 : }
3421 :
3422 : static ulong
3423 : fd_quic_gen_max_streams_frame( fd_quic_conn_t * conn,
3424 : uchar * payload_ptr,
3425 : uchar * payload_end,
3426 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3427 12177 : fd_quic_pkt_meta_tracker_t * tracker ) {
3428 12177 : fd_quic_conn_stream_rx_t * srx = conn->srx;
3429 :
3430 : /* 0x02 Client-Initiated, Unidirectional
3431 : 0x03 Server-Initiated, Unidirectional */
3432 12177 : ulong max_streams_unidir = srx->rx_sup_stream_id >> 2;
3433 :
3434 12177 : uint flags = conn->flags;
3435 12177 : if( !FD_QUIC_MAX_STREAMS_ALWAYS_UNLESS_ACKED ) {
3436 12177 : if( !( flags & FD_QUIC_CONN_FLAGS_MAX_STREAMS_UNIDIR ) ) return 0UL;
3437 0 : if( max_streams_unidir <= srx->rx_max_streams_unidir_ackd ) return 0UL;
3438 0 : }
3439 :
3440 0 : fd_quic_max_streams_frame_t max_streams = {
3441 0 : .type = 0x13, /* unidirectional */
3442 0 : .max_streams = max_streams_unidir
3443 0 : };
3444 0 : ulong frame_sz = fd_quic_encode_max_streams_frame( payload_ptr,
3445 0 : (ulong)( payload_end - payload_ptr ),
3446 0 : &max_streams );
3447 0 : if( FD_UNLIKELY( frame_sz==FD_QUIC_ENCODE_FAIL ) ) return 0UL;
3448 :
3449 0 : if( !fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3450 0 : FD_QUIC_PKT_META_TYPE_MAX_STREAMS_UNIDIR,
3451 0 : (fd_quic_pkt_meta_value_t){0}, /* value doesn't matter */
3452 0 : tracker,
3453 0 : conn ) ) return 0UL;
3454 :
3455 0 : conn->flags = flags & (~FD_QUIC_CONN_FLAGS_MAX_STREAMS_UNIDIR);
3456 0 : conn->upd_pkt_number = pkt_meta_tmpl->key.pkt_num;
3457 0 : return frame_sz;
3458 0 : }
3459 :
3460 : static ulong
3461 : fd_quic_gen_ping_frame( fd_quic_conn_t * conn,
3462 : uchar * payload_ptr,
3463 : uchar * payload_end,
3464 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3465 12177 : fd_quic_pkt_meta_tracker_t * tracker ) {
3466 :
3467 12177 : if( ~conn->flags & FD_QUIC_CONN_FLAGS_PING ) return 0UL;
3468 84 : if( conn->flags & FD_QUIC_CONN_FLAGS_PING_SENT ) return 0UL;
3469 :
3470 84 : fd_quic_ping_frame_t ping = {0};
3471 84 : ulong frame_sz = fd_quic_encode_ping_frame( payload_ptr,
3472 84 : (ulong)( payload_end - payload_ptr ),
3473 84 : &ping );
3474 84 : if( FD_UNLIKELY( frame_sz==FD_QUIC_ENCODE_FAIL ) ) return 0UL;
3475 84 : conn->flags |= FD_QUIC_CONN_FLAGS_PING_SENT;
3476 84 : conn->flags &= ~FD_QUIC_CONN_FLAGS_PING;
3477 :
3478 84 : conn->upd_pkt_number = pkt_meta_tmpl->key.pkt_num;
3479 : /* record the send for retx, 0 if fail */
3480 84 : if( !fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3481 84 : FD_QUIC_PKT_META_TYPE_PING,
3482 84 : (fd_quic_pkt_meta_value_t){0}, /* value doesn't matter */
3483 84 : tracker,
3484 84 : conn ) ) return 0UL;
3485 :
3486 39 : return frame_sz;
3487 84 : }
3488 :
3489 : uchar *
3490 : fd_quic_gen_stream_frames( fd_quic_conn_t * conn,
3491 : uchar * payload_ptr,
3492 : uchar * payload_end,
3493 : fd_quic_pkt_meta_t * pkt_meta_tmpl,
3494 13335405 : fd_quic_pkt_meta_tracker_t * tracker ) {
3495 :
3496 : /* loop serves two purposes:
3497 : 1. finds a stream with data to send
3498 : 2. appends max_stream_data frames as necessary */
3499 13335405 : fd_quic_stream_t * sentinel = conn->send_streams;
3500 13335405 : fd_quic_stream_t * cur_stream = sentinel->next;
3501 13335405 : ulong pkt_num = pkt_meta_tmpl->key.pkt_num;
3502 26434991 : while( !cur_stream->sentinel ) {
3503 : /* required, since cur_stream may get removed from list */
3504 13099589 : fd_quic_stream_t * nxt_stream = cur_stream->next;
3505 13099589 : _Bool sent_all_data = 1u;
3506 :
3507 13099589 : if( cur_stream->upd_pkt_number >= pkt_num ) {
3508 :
3509 : /* any stream data? */
3510 13099589 : if( FD_LIKELY( FD_QUIC_STREAM_ACTION( cur_stream ) ) ) {
3511 :
3512 : /* data_avail is the number of stream bytes available for sending.
3513 : fin_flag_set is 1 if no more bytes will get added to the stream. */
3514 13099586 : ulong const data_avail = cur_stream->tx_buf.head - cur_stream->tx_sent;
3515 13099586 : int const fin_flag_set = !!(cur_stream->state & FD_QUIC_STREAM_STATE_TX_FIN);
3516 13099586 : ulong const stream_id = cur_stream->stream_id;
3517 13099586 : ulong const stream_off = cur_stream->tx_sent;
3518 :
3519 : /* No information to send? */
3520 13099586 : if( data_avail==0u && !fin_flag_set ) break;
3521 :
3522 : /* No space to write frame?
3523 : (Buffer should fit max stream header size and at least 1 byte of data) */
3524 13099586 : if( payload_ptr+FD_QUIC_MAX_FOOTPRINT( stream_e_frame )+1 > payload_end ) break;
3525 :
3526 : /* check pkt_meta availability */
3527 13099586 : if( !fd_quic_conn_can_acquire_pkt_meta( conn, tracker ) ) break;
3528 :
3529 : /* Leave placeholder for frame/stream type */
3530 13099583 : uchar * const frame_type_p = payload_ptr++;
3531 13099583 : uint frame_type = 0x0a; /* stream frame with length */
3532 :
3533 : /* Encode stream ID */
3534 13099583 : payload_ptr += fd_quic_varint_encode( payload_ptr, stream_id );
3535 :
3536 : /* Optionally encode offset */
3537 13099583 : if( stream_off>0 ) {
3538 16863 : frame_type |= 0x04; /* with offset field */
3539 16863 : payload_ptr += fd_quic_varint_encode( payload_ptr, stream_off );
3540 16863 : }
3541 :
3542 : /* Leave placeholder for length length */
3543 13099583 : uchar * data_sz_p = payload_ptr;
3544 13099583 : payload_ptr += 2;
3545 :
3546 : /* Stream metadata */
3547 13099583 : ulong data_max = (ulong)payload_end - (ulong)payload_ptr; /* assume no underflow */
3548 13099583 : ulong data_sz = fd_ulong_min( data_avail, data_max );
3549 13099583 : /* */ data_sz = fd_ulong_min( data_sz, 0x3fffUL ); /* max 2 byte varint */
3550 13099583 : /* */ sent_all_data = data_sz == data_avail;
3551 13099583 : _Bool fin = fin_flag_set && sent_all_data;
3552 :
3553 : /* Finish encoding stream header */
3554 13099583 : ushort data_sz_varint = fd_ushort_bswap( (ushort)( 0x4000u | (uint)data_sz ) );
3555 13099583 : FD_STORE( ushort, data_sz_p, data_sz_varint );
3556 13099583 : frame_type |= fin;
3557 13099583 : *frame_type_p = (uchar)frame_type;
3558 :
3559 : /* Write stream payload */
3560 13099583 : fd_quic_buffer_t * tx_buf = &cur_stream->tx_buf;
3561 13099583 : fd_quic_buffer_load( tx_buf, stream_off, payload_ptr, data_sz );
3562 13099583 : payload_ptr += data_sz;
3563 :
3564 : /* Update stream metadata */
3565 13099583 : cur_stream->tx_sent += data_sz;
3566 13099583 : cur_stream->upd_pkt_number = fd_ulong_if( fin, pkt_num, FD_QUIC_PKT_NUM_PENDING );
3567 13099583 : cur_stream->stream_flags &= fd_uint_if( fin, ~FD_QUIC_STREAM_FLAGS_ACTION, UINT_MAX );
3568 :
3569 : /* Packet metadata for potential retransmits */
3570 13099583 : pkt_meta_tmpl->key.stream_id = cur_stream->stream_id;
3571 13099583 : fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3572 13099583 : FD_QUIC_PKT_META_TYPE_STREAM,
3573 13099583 : (fd_quic_pkt_meta_value_t){
3574 13099583 : .range = {
3575 13099583 : .offset_lo = stream_off,
3576 13099583 : .offset_hi = stream_off + data_sz
3577 13099583 : }
3578 13099583 : },
3579 13099583 : tracker,
3580 13099583 : conn );
3581 13099583 : }
3582 13099589 : }
3583 :
3584 13099586 : if( sent_all_data ) {
3585 13082798 : cur_stream->stream_flags &= ~FD_QUIC_STREAM_FLAGS_ACTION;
3586 13082798 : FD_QUIC_STREAM_LIST_REMOVE( cur_stream );
3587 13082798 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->used_streams, cur_stream );
3588 13082798 : }
3589 :
3590 13099586 : cur_stream = nxt_stream;
3591 13099586 : }
3592 :
3593 13335405 : return payload_ptr;
3594 13335405 : }
3595 :
3596 : uchar *
3597 : fd_quic_gen_frames( fd_quic_conn_t * conn,
3598 : uchar * payload_ptr,
3599 : uchar * payload_end,
3600 : fd_quic_pkt_meta_t * pkt_meta_tmpl,
3601 13383612 : ulong now ) {
3602 :
3603 13383612 : uint closing = 0U;
3604 13383612 : switch( conn->state ) {
3605 12000 : case FD_QUIC_CONN_STATE_PEER_CLOSE:
3606 12084 : case FD_QUIC_CONN_STATE_ABORT:
3607 18108 : case FD_QUIC_CONN_STATE_CLOSE_PENDING:
3608 18108 : closing = 1u;
3609 13383612 : }
3610 :
3611 13383612 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
3612 :
3613 13383612 : payload_ptr = fd_quic_gen_ack_frames( conn->ack_gen, payload_ptr, payload_end, pkt_meta_tmpl->enc_level, now, (float)conn->quic->config.tick_per_us );
3614 13383612 : if( conn->ack_gen->head == conn->ack_gen->tail ) conn->unacked_sz = 0UL;
3615 :
3616 13383612 : if( FD_UNLIKELY( closing ) ) {
3617 18108 : payload_ptr += fd_quic_gen_close_frame( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3618 13365504 : } else {
3619 13365504 : payload_ptr = fd_quic_gen_handshake_frames( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3620 13365504 : if( pkt_meta_tmpl->enc_level == fd_quic_enc_level_appdata_id ) {
3621 13341417 : payload_ptr += fd_quic_gen_handshake_done_frame( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3622 13341417 : if( conn->upd_pkt_number >= pkt_meta_tmpl->key.pkt_num ) {
3623 12177 : payload_ptr += fd_quic_gen_max_data_frame ( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3624 12177 : payload_ptr += fd_quic_gen_max_streams_frame( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3625 12177 : payload_ptr += fd_quic_gen_ping_frame ( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3626 12177 : }
3627 13341417 : if( FD_LIKELY( !conn->tls_hs ) ) {
3628 13335396 : payload_ptr = fd_quic_gen_stream_frames( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3629 13335396 : }
3630 13341417 : }
3631 13365504 : }
3632 :
3633 13383612 : return payload_ptr;
3634 13383612 : }
3635 :
3636 : /* transmit
3637 : looks at each of the following dependent on state, and creates
3638 : a packet to transmit:
3639 : acks
3640 : handshake data (tls)
3641 : handshake done
3642 : ping
3643 : stream data */
3644 : static void
3645 : fd_quic_conn_tx( fd_quic_t * quic,
3646 13354806 : fd_quic_conn_t * conn ) {
3647 :
3648 13354806 : if( FD_UNLIKELY( conn->state == FD_QUIC_CONN_STATE_DEAD ) ) return;
3649 :
3650 13354806 : fd_quic_state_t * state = fd_quic_get_state( quic );
3651 :
3652 : /* used for encoding frames into before encrypting */
3653 13354806 : uchar * crypt_scratch = state->crypt_scratch;
3654 13354806 : ulong crypt_scratch_sz = sizeof( state->crypt_scratch );
3655 :
3656 : /* max packet size */
3657 : /* TODO probably should be called tx_max_udp_payload_sz */
3658 13354806 : ulong tx_max_datagram_sz = conn->tx_max_datagram_sz;
3659 :
3660 13354806 : if( conn->tx_ptr != conn->tx_buf_conn ) {
3661 0 : fd_quic_tx_buffered( quic, conn );
3662 0 : fd_quic_svc_schedule( state, conn, FD_QUIC_SVC_INSTANT );
3663 0 : return;
3664 0 : }
3665 :
3666 : /* choose enc_level to tx at */
3667 : /* this function accepts an argument "acks"
3668 : * We want to minimize the number of packets that carry only acks.
3669 : * fd_quic_tx_enc_level determines whether a packet needs sending,
3670 : * and when encryption level should be used.
3671 : * If "acks" is set to 1 (true), fd_quic_tx_enc_level checks for acks.
3672 : * Otherwise, it does not check for acks
3673 : * We set "acks" only on the first call in this function. All subsequent
3674 : * calls do not set it.
3675 : * This ensures that ack-only packets only occur when nothing else needs
3676 : * to be sent */
3677 13354806 : uint enc_level = fd_quic_tx_enc_level( conn, 1 /* acks */ );
3678 : /* RFC 9000 Section 17.2.2.1. Abandoning Initial Packets
3679 : > A client stops both sending and processing Initial packets when
3680 : > it sends its first Handshake packet. */
3681 13354806 : if( quic->config.role==FD_QUIC_ROLE_CLIENT && enc_level==fd_quic_enc_level_handshake_id ) {
3682 6021 : fd_quic_abandon_enc_level( conn, fd_quic_enc_level_initial_id );
3683 6021 : }
3684 :
3685 : /* nothing to send / bad state? */
3686 13354806 : if( enc_level == ~0u ) return;
3687 :
3688 13354806 : int key_phase_upd = (int)conn->key_update;
3689 13354806 : uint key_phase = conn->key_phase;
3690 13354806 : int key_phase_tx = (int)key_phase ^ key_phase_upd;
3691 :
3692 : /* get time, and set reschedule time for at most the idle timeout */
3693 13354806 : ulong now = fd_quic_get_state( quic )->now;
3694 :
3695 : /* initialize expiry and tx_time */
3696 13354806 : fd_quic_pkt_meta_t pkt_meta_tmpl[1] = {{.expiry = now+500000000UL, .tx_time = now}};
3697 : // pkt_meta_tmpl->expiry = fd_quic_calc_expiry( conn, now );
3698 : //ulong margin = (ulong)(conn->rtt->smoothed_rtt) + (ulong)(3 * conn->rtt->var_rtt);
3699 : //if( margin < pkt_meta->expiry ) {
3700 : // pkt_meta->expiry -= margin;
3701 : //}
3702 :
3703 26732358 : while( enc_level != ~0u ) {
3704 13383612 : uint initial_pkt = 0; /* is this the first initial packet? */
3705 :
3706 :
3707 : /* remaining in datagram */
3708 : /* invariant: tx_ptr >= tx_buf */
3709 13383612 : ulong datagram_rem = tx_max_datagram_sz - (ulong)( conn->tx_ptr - conn->tx_buf_conn );
3710 :
3711 : /* encode into here */
3712 : /* this is the start of a new quic packet
3713 : cur_ptr points at the next byte to fill with a quic pkt */
3714 : /* currently, cur_ptr just points at the start of crypt_scratch
3715 : each quic packet gets encrypted into tx_buf, and the space in
3716 : crypt_scratch is reused */
3717 13383612 : uchar * cur_ptr = crypt_scratch;
3718 13383612 : ulong cur_sz = crypt_scratch_sz;
3719 :
3720 : /* TODO determine actual datagrams size to use */
3721 13383612 : cur_sz = fd_ulong_min( cur_sz, datagram_rem );
3722 :
3723 : /* determine pn_space */
3724 13383612 : uint pn_space = fd_quic_enc_level_to_pn_space( enc_level );
3725 13383612 : pkt_meta_tmpl->pn_space = (uchar)pn_space;
3726 13383612 : pkt_meta_tmpl->enc_level = (uchar)(enc_level&0x3);
3727 :
3728 : /* get next packet number
3729 : Returned to pool if not sent as gaps are harmful for ACK frame
3730 : compression. */
3731 13383612 : ulong pkt_number = conn->pkt_number[pn_space];
3732 13383612 : FD_QUIC_PKT_META_SET_PKT_NUM( pkt_meta_tmpl, pkt_number );
3733 :
3734 : /* are we the client initial packet? */
3735 13383612 : ulong hs_data_offset = conn->hs_sent_bytes[enc_level];
3736 13383612 : initial_pkt = (uint)( hs_data_offset == 0 ) & (uint)( !conn->server ) & (uint)( enc_level == fd_quic_enc_level_initial_id );
3737 :
3738 : /* current peer endpoint */
3739 13383612 : fd_quic_conn_id_t const * peer_conn_id = &conn->peer_cids[0];
3740 :
3741 : /* our current conn_id */
3742 13383612 : ulong conn_id = conn->our_conn_id;
3743 13383612 : uint const pkt_num_len = 4u; /* 4-byte packet number */
3744 13383612 : uint const pkt_num_len_enc = pkt_num_len - 1; /* -1 offset for protocol */
3745 :
3746 :
3747 : /* encode packet header (including packet number)
3748 : While encoding, remember where the 'length' field is, if one
3749 : exists. We'll have to update it later. */
3750 13383612 : uchar * hdr_ptr = cur_ptr;
3751 13383612 : ulong hdr_sz = 0UL;
3752 13383612 : uchar _hdr_len_field[2]; /* if no len field exists, catch the write here */
3753 13383612 : uchar * hdr_len_field = _hdr_len_field;
3754 13383612 : switch( enc_level ) {
3755 12129 : case fd_quic_enc_level_initial_id: {
3756 12129 : fd_quic_initial_t initial = {0};
3757 12129 : initial.h0 = fd_quic_initial_h0( pkt_num_len_enc );
3758 12129 : initial.version = 1;
3759 12129 : initial.dst_conn_id_len = peer_conn_id->sz;
3760 : // .dst_conn_id
3761 12129 : initial.src_conn_id_len = FD_QUIC_CONN_ID_SZ;
3762 : // .src_conn_id
3763 : // .token - below
3764 12129 : initial.len = 0x3fff; /* use 2 byte varint encoding */
3765 12129 : initial.pkt_num = pkt_number;
3766 :
3767 12129 : fd_memcpy( initial.dst_conn_id, peer_conn_id->conn_id, peer_conn_id->sz );
3768 12129 : memcpy( initial.src_conn_id, &conn_id, FD_QUIC_CONN_ID_SZ );
3769 :
3770 : /* Initial packets sent by the server MUST set the Token Length field to 0. */
3771 12129 : initial.token = conn->token;
3772 12129 : if( conn->quic->config.role == FD_QUIC_ROLE_CLIENT && conn->token_len ) {
3773 3 : initial.token_len = conn->token_len;
3774 12126 : } else {
3775 12126 : initial.token_len = 0;
3776 12126 : }
3777 :
3778 12129 : hdr_sz = fd_quic_encode_initial( cur_ptr, cur_sz, &initial );
3779 12129 : hdr_len_field = cur_ptr + hdr_sz - 6; /* 2 byte len, 4 byte packet number */
3780 12129 : FD_DTRACE_PROBE_2( quic_encode_initial, initial.src_conn_id, initial.dst_conn_id );
3781 12129 : break;
3782 0 : }
3783 :
3784 12042 : case fd_quic_enc_level_handshake_id: {
3785 12042 : fd_quic_handshake_t handshake = {0};
3786 12042 : handshake.h0 = fd_quic_handshake_h0( pkt_num_len_enc );
3787 12042 : handshake.version = 1;
3788 :
3789 : /* destination */
3790 12042 : fd_memcpy( handshake.dst_conn_id, peer_conn_id->conn_id, peer_conn_id->sz );
3791 12042 : handshake.dst_conn_id_len = peer_conn_id->sz;
3792 :
3793 : /* source */
3794 12042 : FD_STORE( ulong, handshake.src_conn_id, conn_id );
3795 12042 : handshake.src_conn_id_len = sizeof(ulong);
3796 :
3797 12042 : handshake.len = 0x3fff; /* use 2 byte varint encoding */
3798 12042 : handshake.pkt_num = pkt_number;
3799 :
3800 12042 : hdr_sz = fd_quic_encode_handshake( cur_ptr, cur_sz, &handshake );
3801 12042 : hdr_len_field = cur_ptr + hdr_sz - 6; /* 2 byte len, 4 byte packet number */
3802 12042 : FD_DTRACE_PROBE_2( quic_encode_handshake, handshake.src_conn_id, handshake.dst_conn_id );
3803 12042 : break;
3804 0 : }
3805 :
3806 13359441 : case fd_quic_enc_level_appdata_id:
3807 13359441 : {
3808 13359441 : fd_quic_one_rtt_t one_rtt = {0};
3809 13359441 : one_rtt.h0 = fd_quic_one_rtt_h0( /* spin */ 0, !!key_phase_tx, pkt_num_len_enc );
3810 :
3811 : /* destination */
3812 13359441 : fd_memcpy( one_rtt.dst_conn_id, peer_conn_id->conn_id, peer_conn_id->sz );
3813 13359441 : one_rtt.dst_conn_id_len = peer_conn_id->sz;
3814 :
3815 13359441 : one_rtt.pkt_num = pkt_number;
3816 :
3817 13359441 : hdr_sz = fd_quic_encode_one_rtt( cur_ptr, cur_sz, &one_rtt );
3818 13359441 : FD_DTRACE_PROBE_2( quic_encode_one_rtt, one_rtt.dst_conn_id, one_rtt.pkt_num );
3819 13359441 : break;
3820 0 : }
3821 :
3822 0 : default:
3823 0 : FD_LOG_ERR(( "%s - logic error: unexpected enc_level", __func__ ));
3824 13383612 : }
3825 :
3826 : /* if we don't have reasonable amt of space for a new packet, tx to free space */
3827 13383612 : const ulong min_rqd = 64;
3828 13383612 : if( FD_UNLIKELY( hdr_sz==FD_QUIC_ENCODE_FAIL || hdr_sz + min_rqd > cur_sz ) ) {
3829 : /* try to free space */
3830 0 : fd_quic_tx_buffered( quic, conn );
3831 :
3832 : /* we have lots of space, so try again */
3833 0 : if( conn->tx_buf_conn == conn->tx_ptr ) {
3834 0 : enc_level = fd_quic_tx_enc_level( conn, 0 /* acks */ );
3835 0 : continue;
3836 0 : }
3837 :
3838 : /* reschedule, since some data was unable to be sent */
3839 : /* TODO might want to add a backoff here */
3840 0 : fd_quic_svc_schedule( state, conn, FD_QUIC_SVC_INSTANT );
3841 :
3842 0 : break;
3843 0 : }
3844 :
3845 13383612 : cur_ptr += hdr_sz;
3846 13383612 : cur_sz -= hdr_sz;
3847 :
3848 : /* start writing payload, leaving room for header and expansion
3849 : due to varint coding */
3850 :
3851 13383612 : uchar * payload_ptr = cur_ptr;
3852 13383612 : ulong payload_sz = cur_sz;
3853 : /* payload_end leaves room for TAG */
3854 13383612 : uchar * payload_end = payload_ptr + payload_sz - FD_QUIC_CRYPTO_TAG_SZ;
3855 :
3856 13383612 : uchar * const frame_start = payload_ptr;
3857 13383612 : payload_ptr = fd_quic_gen_frames( conn, frame_start, payload_end, pkt_meta_tmpl, now );
3858 13383612 : if( FD_UNLIKELY( payload_ptr < frame_start ) ) FD_LOG_CRIT(( "fd_quic_gen_frames failed" ));
3859 :
3860 : /* did we add any frames? */
3861 :
3862 13383612 : if( payload_ptr==frame_start ) {
3863 : /* we have data to add, but none was added, presumably due
3864 : so space in the datagram */
3865 6060 : ulong free_bytes = (ulong)( payload_ptr - payload_end );
3866 : /* sanity check */
3867 6060 : if( free_bytes > 64 ) {
3868 : /* we should have been able to fit data into 64 bytes
3869 : so stop trying here */
3870 6060 : break;
3871 6060 : }
3872 :
3873 : /* try to free space */
3874 0 : fd_quic_tx_buffered( quic, conn );
3875 :
3876 : /* we have lots of space, so try again */
3877 0 : if( conn->tx_buf_conn == conn->tx_ptr ) {
3878 0 : enc_level = fd_quic_tx_enc_level( conn, 0 /* acks */ );
3879 0 : continue;
3880 0 : }
3881 0 : }
3882 :
3883 : /* first initial frame is padded to FD_QUIC_INITIAL_PAYLOAD_SZ_MIN
3884 : all short quic packets are padded so 16 bytes of sample are available */
3885 13377552 : uint tot_frame_sz = (uint)( payload_ptr - frame_start );
3886 13377552 : uint base_pkt_len = (uint)tot_frame_sz + pkt_num_len + FD_QUIC_CRYPTO_TAG_SZ;
3887 13377552 : uint padding = initial_pkt ? FD_QUIC_INITIAL_PAYLOAD_SZ_MIN - base_pkt_len : 0u;
3888 :
3889 13377552 : if( base_pkt_len + padding < FD_QUIC_CRYPTO_SAMPLE_OFFSET_FROM_PKT_NUM_START + FD_QUIC_CRYPTO_SAMPLE_SZ ) {
3890 0 : padding = FD_QUIC_CRYPTO_SAMPLE_SZ + FD_QUIC_CRYPTO_SAMPLE_OFFSET_FROM_PKT_NUM_START - base_pkt_len;
3891 0 : }
3892 :
3893 : /* this length includes the packet number length (pkt_number_len_enc+1),
3894 : padding and the final TAG */
3895 13377552 : uint quic_pkt_len = base_pkt_len + padding;
3896 :
3897 : /* set the length on the packet header */
3898 13377552 : uint quic_pkt_len_varint = 0x4000u | fd_uint_min( quic_pkt_len, 0x3fff );
3899 13377552 : FD_STORE( ushort, hdr_len_field, fd_ushort_bswap( (ushort)quic_pkt_len_varint ) );
3900 :
3901 : /* add padding */
3902 13377552 : if( padding ) {
3903 6024 : fd_memset( payload_ptr, 0, padding );
3904 6024 : payload_ptr += padding;
3905 6024 : }
3906 :
3907 : /* everything successful up to here
3908 : encrypt into tx_ptr,tx_ptr+tx_sz */
3909 :
3910 : #if FD_QUIC_DISABLE_CRYPTO
3911 : ulong quic_pkt_sz = hdr_sz + tot_frame_sz + padding;
3912 : fd_memcpy( conn->tx_ptr, hdr_ptr, quic_pkt_sz );
3913 : conn->tx_ptr += quic_pkt_sz;
3914 :
3915 : /* append MAC tag */
3916 : memset( conn->tx_ptr, 0, FD_QUIC_CRYPTO_TAG_SZ );
3917 : conn->tx_ptr += FD_QUIC_CRYPTO_TAG_SZ;
3918 : #else
3919 13377552 : ulong cipher_text_sz = fd_quic_conn_tx_buf_remaining( conn );
3920 13377552 : ulong frames_sz = (ulong)( payload_ptr - frame_start ); /* including padding */
3921 :
3922 13377552 : fd_quic_crypto_keys_t * hp_keys = &conn->keys[enc_level][1];
3923 13377552 : fd_quic_crypto_keys_t * pkt_keys = key_phase_upd ? &conn->new_keys[1] : &conn->keys[enc_level][1];
3924 :
3925 13377552 : if( FD_UNLIKELY( fd_quic_crypto_encrypt( conn->tx_ptr, &cipher_text_sz, hdr_ptr, hdr_sz,
3926 13377552 : frame_start, frames_sz, pkt_keys, hp_keys, pkt_number ) != FD_QUIC_SUCCESS ) ) {
3927 0 : FD_LOG_WARNING(( "fd_quic_crypto_encrypt failed" ));
3928 :
3929 : /* this situation is unlikely to improve, so kill the connection */
3930 0 : conn->state = FD_QUIC_CONN_STATE_DEAD;
3931 0 : fd_quic_svc_schedule( state, conn, FD_QUIC_SVC_INSTANT );
3932 0 : quic->metrics.conn_aborted_cnt++;
3933 0 : break;
3934 0 : }
3935 :
3936 13377552 : conn->tx_ptr += cipher_text_sz;
3937 13377552 : #endif
3938 :
3939 : /* we have committed the packet into the buffer, so inc pkt_number */
3940 13377552 : conn->pkt_number[pn_space]++;
3941 :
3942 13377552 : fd_quic_svc_schedule( state, conn, FD_QUIC_SVC_WAIT );
3943 :
3944 13377552 : if( enc_level == fd_quic_enc_level_appdata_id ) {
3945 : /* short header must be last in datagram
3946 : so send in packet immediately */
3947 13353381 : fd_quic_tx_buffered( quic, conn );
3948 :
3949 13353381 : if( conn->tx_ptr == conn->tx_buf_conn ) {
3950 13353381 : enc_level = fd_quic_tx_enc_level( conn, 0 /* acks */ );
3951 13353381 : continue;
3952 13353381 : }
3953 :
3954 : /* TODO count here */
3955 :
3956 : /* drop packet */
3957 : /* this is a workaround for leaving a short=header-packet in the buffer
3958 : for the next tx_conn call. Next time around the tx_conn call will
3959 : not be aware that the buffer cannot be added to */
3960 0 : conn->tx_ptr = conn->tx_buf_conn;
3961 :
3962 0 : break;
3963 13353381 : }
3964 :
3965 : /* Refresh enc_level in case we can coalesce another packet */
3966 24171 : enc_level = fd_quic_tx_enc_level( conn, 0 /* acks */ );
3967 24171 : FD_DEBUG( if( enc_level!=~0u) FD_LOG_DEBUG(( "Attempting to append enc_level=%u packet", enc_level )); )
3968 24171 : }
3969 :
3970 : /* try to send? */
3971 13354806 : fd_quic_tx_buffered( quic, conn );
3972 13354806 : }
3973 :
3974 : void
3975 13354806 : fd_quic_conn_service( fd_quic_t * quic, fd_quic_conn_t * conn, ulong now ) {
3976 13354806 : (void)now;
3977 :
3978 : /* Send new rtt measurement probe? */
3979 13354806 : if( FD_UNLIKELY(now > conn->last_ack + (ulong)conn->rtt_period_ticks) ) {
3980 : /* send PING */
3981 18 : if( !( conn->flags & ( FD_QUIC_CONN_FLAGS_PING | FD_QUIC_CONN_FLAGS_PING_SENT ) )
3982 18 : && conn->state == FD_QUIC_CONN_STATE_ACTIVE ) {
3983 0 : conn->flags |= FD_QUIC_CONN_FLAGS_PING;
3984 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING; /* update to be sent in next packet */
3985 0 : }
3986 18 : }
3987 :
3988 : /* handle expiry on pkt_meta */
3989 13354806 : fd_quic_pkt_meta_retry( quic, conn, 0 /* don't force */, ~0u /* enc_level */ );
3990 :
3991 : /* check state
3992 : need reset?
3993 : need close?
3994 : need acks?
3995 : replies?
3996 : data to send?
3997 : dead */
3998 13354806 : switch( conn->state ) {
3999 12045 : case FD_QUIC_CONN_STATE_HANDSHAKE:
4000 24087 : case FD_QUIC_CONN_STATE_HANDSHAKE_COMPLETE:
4001 24087 : {
4002 24087 : if( conn->tls_hs ) {
4003 : /* if we're the server, we send "handshake-done" frame */
4004 24087 : if( conn->state == FD_QUIC_CONN_STATE_HANDSHAKE_COMPLETE && conn->server ) {
4005 6021 : conn->handshake_done_send = 1;
4006 :
4007 : /* move straight to ACTIVE */
4008 6021 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_ACTIVE );
4009 :
4010 : /* RFC 9001 4.9.2. Discarding Handshake Keys
4011 : > An endpoint MUST discard its Handshake keys when the
4012 : > TLS handshake is confirmed
4013 : RFC 9001 4.1.2. Handshake Confirmed
4014 : > [...] the TLS handshake is considered confirmed at the
4015 : > server when the handshake completes */
4016 6021 : fd_quic_abandon_enc_level( conn, fd_quic_enc_level_handshake_id );
4017 :
4018 : /* user callback */
4019 6021 : fd_quic_cb_conn_new( quic, conn );
4020 :
4021 : /* clear out hs_data here, as we don't need it anymore */
4022 6021 : fd_quic_tls_clear_hs_data( conn->tls_hs, fd_quic_enc_level_appdata_id );
4023 6021 : }
4024 :
4025 : /* if we're the client, fd_quic_conn_tx will flush the hs
4026 : buffer so we can receive the HANDSHAKE_DONE frame, and
4027 : transition from CONN_STATE HANDSHAKE_COMPLETE to ACTIVE. */
4028 24087 : }
4029 :
4030 : /* do we have data to transmit? */
4031 24087 : fd_quic_conn_tx( quic, conn );
4032 :
4033 24087 : break;
4034 12045 : }
4035 :
4036 6024 : case FD_QUIC_CONN_STATE_CLOSE_PENDING:
4037 12024 : case FD_QUIC_CONN_STATE_PEER_CLOSE:
4038 : /* user requested close, and may have set a reason code */
4039 : /* transmit the failure reason */
4040 12024 : fd_quic_conn_tx( quic, conn );
4041 :
4042 : /* schedule another fd_quic_conn_service to free the conn */
4043 12024 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_DEAD ); /* TODO need draining state wait for 3 * TPO */
4044 12024 : quic->metrics.conn_closed_cnt++;
4045 12024 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
4046 :
4047 12024 : break;
4048 :
4049 84 : case FD_QUIC_CONN_STATE_ABORT:
4050 : /* transmit the failure reason */
4051 84 : fd_quic_conn_tx( quic, conn );
4052 :
4053 : /* schedule another fd_quic_conn_service to free the conn */
4054 84 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_DEAD );
4055 84 : quic->metrics.conn_aborted_cnt++;
4056 84 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
4057 :
4058 84 : break;
4059 :
4060 13318611 : case FD_QUIC_CONN_STATE_ACTIVE:
4061 : /* do we have data to transmit? */
4062 13318611 : fd_quic_conn_tx( quic, conn );
4063 :
4064 13318611 : break;
4065 :
4066 0 : case FD_QUIC_CONN_STATE_DEAD:
4067 0 : case FD_QUIC_CONN_STATE_INVALID:
4068 : /* fall thru */
4069 0 : default:
4070 0 : return;
4071 13354806 : }
4072 :
4073 : /* check routing and arp for this connection */
4074 :
4075 13354806 : }
4076 :
4077 : void
4078 : fd_quic_conn_free( fd_quic_t * quic,
4079 14259 : fd_quic_conn_t * conn ) {
4080 14259 : if( FD_UNLIKELY( !conn ) ) {
4081 0 : FD_LOG_WARNING(( "NULL conn" ));
4082 0 : return;
4083 0 : }
4084 14259 : if( FD_UNLIKELY( conn->state == FD_QUIC_CONN_STATE_INVALID ) ) {
4085 0 : FD_LOG_CRIT(( "double free detected" ));
4086 0 : return;
4087 0 : }
4088 :
4089 14259 : FD_COMPILER_MFENCE();
4090 14259 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_INVALID );
4091 14259 : FD_COMPILER_MFENCE();
4092 :
4093 14259 : fd_quic_state_t * state = fd_quic_get_state( quic );
4094 :
4095 : /* no need to remove this connection from the events queue
4096 : free is called from two places:
4097 : fini - service will never be called again. All events are destroyed
4098 : service - removes event before calling free. Event only allowed to be
4099 : enqueued once */
4100 :
4101 : /* remove all stream ids from map, and free stream */
4102 :
4103 : /* remove used streams */
4104 14259 : fd_quic_stream_t * used_sentinel = conn->used_streams;
4105 20309 : while( 1 ) {
4106 20309 : fd_quic_stream_t * stream = used_sentinel->next;
4107 :
4108 20309 : if( FD_UNLIKELY( stream == used_sentinel ) ) break;
4109 :
4110 6050 : fd_quic_tx_stream_free( quic, conn, stream, FD_QUIC_STREAM_NOTIFY_CONN );
4111 6050 : }
4112 :
4113 : /* remove send streams */
4114 14259 : fd_quic_stream_t * send_sentinel = conn->send_streams;
4115 20271 : while( 1 ) {
4116 20271 : fd_quic_stream_t * stream = send_sentinel->next;
4117 :
4118 20271 : if( FD_UNLIKELY( stream == send_sentinel ) ) break;
4119 :
4120 6012 : fd_quic_tx_stream_free( quic, conn, stream, FD_QUIC_STREAM_NOTIFY_CONN );
4121 6012 : }
4122 :
4123 : /* if any stream map entries are left over, remove them
4124 : this should not occur, so this branch should not execute
4125 : but if a stream doesn't get cleaned up properly, this fixes
4126 : the stream map */
4127 14259 : if( FD_UNLIKELY( conn->stream_map && fd_quic_stream_map_key_cnt( conn->stream_map ) > 0 ) ) {
4128 0 : FD_LOG_WARNING(( "stream_map not empty. cnt: %lu",
4129 0 : (ulong)fd_quic_stream_map_key_cnt( conn->stream_map ) ));
4130 0 : while( fd_quic_stream_map_key_cnt( conn->stream_map ) > 0 ) {
4131 0 : int removed = 0;
4132 0 : for( ulong j = 0; j < fd_quic_stream_map_slot_cnt( conn->stream_map ); ++j ) {
4133 0 : if( conn->stream_map[j].stream_id != FD_QUIC_STREAM_ID_UNUSED ) {
4134 0 : fd_quic_stream_map_remove( conn->stream_map, &conn->stream_map[j] );
4135 0 : removed = 1;
4136 0 : j--; /* retry this entry */
4137 0 : }
4138 0 : }
4139 0 : if( !removed ) {
4140 0 : FD_LOG_WARNING(( "None removed. Remain: %lu",
4141 0 : (ulong)fd_quic_stream_map_key_cnt( conn->stream_map ) ));
4142 0 : break;
4143 0 : }
4144 0 : }
4145 0 : }
4146 :
4147 14259 : if( conn->tls_hs ) {
4148 : /* free tls-hs */
4149 117 : fd_quic_tls_hs_delete( conn->tls_hs );
4150 :
4151 : /* Remove the handshake from the cache before releasing it */
4152 117 : fd_quic_tls_hs_cache_ele_remove( &state->hs_cache, conn->tls_hs, state->hs_pool);
4153 117 : fd_quic_tls_hs_pool_ele_release( state->hs_pool, conn->tls_hs );
4154 117 : }
4155 14259 : conn->tls_hs = NULL;
4156 :
4157 : /* remove connection from service queue */
4158 14259 : if( FD_LIKELY( conn->svc_type != UINT_MAX ) ) {
4159 12159 : fd_quic_svc_unqueue( state, conn );
4160 12159 : }
4161 :
4162 : /* put connection back in free list */
4163 14259 : conn->svc_type = UINT_MAX;
4164 14259 : conn->svc_prev = UINT_MAX;
4165 14259 : conn->svc_next = state->free_conn_list;
4166 14259 : state->free_conn_list = conn->conn_idx;
4167 14259 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_INVALID );
4168 :
4169 14259 : quic->metrics.conn_active_cnt--;
4170 :
4171 : /* clear keys */
4172 14259 : memset( &conn->secrets, 0, sizeof(fd_quic_crypto_secrets_t) );
4173 14259 : memset( conn->keys, 0, sizeof( conn->keys ) );
4174 14259 : memset( conn->new_keys, 0, sizeof( conn->new_keys ) );
4175 14259 : }
4176 :
4177 : fd_quic_conn_t *
4178 : fd_quic_connect( fd_quic_t * quic,
4179 : uint dst_ip_addr,
4180 : ushort dst_udp_port,
4181 : uint src_ip_addr,
4182 6057 : ushort src_udp_port ) {
4183 :
4184 6057 : fd_quic_state_t * state = fd_quic_get_state( quic );
4185 6057 : state->now = fd_quic_now( quic );
4186 :
4187 6057 : if( FD_UNLIKELY( !fd_quic_tls_hs_pool_free( state->hs_pool ) ) ) {
4188 : /* try evicting, 0 if oldest is too young so fail */
4189 6 : if( !fd_quic_tls_hs_cache_evict( quic, state ) ) {
4190 3 : return NULL;
4191 3 : }
4192 6 : }
4193 :
4194 :
4195 6054 : fd_rng_t * rng = state->_rng;
4196 :
4197 : /* create conn ids for us and them
4198 : client creates connection id for the peer, peer immediately replaces it */
4199 6054 : ulong our_conn_id_u64 = fd_rng_ulong( rng );
4200 6054 : fd_quic_conn_id_t peer_conn_id; fd_quic_conn_id_rand( &peer_conn_id, rng );
4201 :
4202 6054 : fd_quic_conn_t * conn = fd_quic_conn_create(
4203 6054 : quic,
4204 6054 : our_conn_id_u64,
4205 6054 : &peer_conn_id,
4206 6054 : dst_ip_addr,
4207 6054 : dst_udp_port,
4208 6054 : src_ip_addr,
4209 6054 : src_udp_port,
4210 6054 : 0 /* client */ );
4211 :
4212 6054 : if( FD_UNLIKELY( !conn ) ) {
4213 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_conn_create failed" )) );
4214 0 : return NULL;
4215 0 : }
4216 :
4217 : /* Prepare QUIC-TLS transport params object (sent as a TLS extension).
4218 : Take template from state and mutate certain params in-place.
4219 :
4220 : See RFC 9000 Section 18 */
4221 :
4222 6054 : fd_quic_transport_params_t tp[1] = { state->transport_params };
4223 :
4224 : /* The original_destination_connection_id is omitted by clients.
4225 : Since this is a mutable field, explicitly clear it here. */
4226 :
4227 6054 : tp->original_destination_connection_id_present = 0;
4228 6054 : tp->original_destination_connection_id_len = 0;
4229 :
4230 : /* Similarly, explicitly zero out retry fields. */
4231 6054 : tp->retry_source_connection_id_present = 0;
4232 6054 : tp->retry_source_connection_id_len = 0;
4233 :
4234 : /* Repeat source conn ID -- rationale see fd_quic_handle_v1_initial */
4235 :
4236 6054 : FD_STORE( ulong, tp->initial_source_connection_id, conn->initial_source_conn_id );
4237 6054 : tp->initial_source_connection_id_present = 1;
4238 6054 : tp->initial_source_connection_id_len = FD_QUIC_CONN_ID_SZ;
4239 :
4240 : /* Create a TLS handshake (free>0 validated above) */
4241 :
4242 6054 : fd_quic_tls_hs_t * tls_hs = fd_quic_tls_hs_new(
4243 6054 : fd_quic_tls_hs_pool_ele_acquire( state->hs_pool ),
4244 6054 : state->tls,
4245 6054 : (void*)conn,
4246 6054 : 0 /*is_server*/,
4247 6054 : tp,
4248 6054 : state->now );
4249 6054 : if( FD_UNLIKELY( tls_hs->alert ) ) {
4250 0 : FD_LOG_WARNING(( "fd_quic_tls_hs_client_new failed" ));
4251 : /* shut down tls_hs */
4252 0 : fd_quic_conn_free( quic, conn );
4253 0 : return NULL;
4254 0 : }
4255 6054 : fd_quic_tls_hs_cache_ele_push_tail( &state->hs_cache, tls_hs, state->hs_pool );
4256 :
4257 6054 : quic->metrics.hs_created_cnt++;
4258 6054 : conn->tls_hs = tls_hs;
4259 :
4260 6054 : fd_quic_gen_initial_secret_and_keys( conn, &peer_conn_id, /* is_server */ 0 );
4261 :
4262 6054 : fd_quic_svc_schedule( state, conn, FD_QUIC_SVC_INSTANT );
4263 :
4264 : /* set "called_conn_new" to indicate we should call conn_final
4265 : upon teardown */
4266 6054 : conn->called_conn_new = 1;
4267 :
4268 : /* everything initialized */
4269 6054 : return conn;
4270 :
4271 6054 : }
4272 :
4273 : fd_quic_conn_t *
4274 : fd_quic_conn_create( fd_quic_t * quic,
4275 : ulong our_conn_id,
4276 : fd_quic_conn_id_t const * peer_conn_id,
4277 : uint peer_ip_addr,
4278 : ushort peer_udp_port,
4279 : uint self_ip_addr,
4280 : ushort self_udp_port,
4281 314307 : int server ) {
4282 314307 : if( FD_UNLIKELY( !our_conn_id ) ) return NULL;
4283 :
4284 314307 : fd_quic_config_t * config = &quic->config;
4285 314307 : fd_quic_state_t * state = fd_quic_get_state( quic );
4286 :
4287 : /* fetch top of connection free list */
4288 314307 : uint conn_idx = state->free_conn_list;
4289 314307 : if( FD_UNLIKELY( conn_idx==UINT_MAX ) ) {
4290 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_conn_create failed: no free conn slots" )) );
4291 0 : quic->metrics.conn_err_no_slots_cnt++;
4292 0 : return NULL;
4293 0 : }
4294 314307 : if( FD_UNLIKELY( conn_idx >= quic->limits.conn_cnt ) ) {
4295 0 : FD_LOG_ERR(( "Conn free list corruption detected" ));
4296 0 : return NULL;
4297 0 : }
4298 314307 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, conn_idx );
4299 314307 : if( FD_UNLIKELY( conn->state != FD_QUIC_CONN_STATE_INVALID ) ) {
4300 0 : FD_LOG_ERR(( "conn %p not free, this is a bug", (void *)conn ));
4301 0 : return NULL;
4302 0 : }
4303 :
4304 : /* prune previous conn map entry */
4305 314307 : fd_quic_conn_map_t * entry = fd_quic_conn_query1( state->conn_map, conn->our_conn_id, NULL );
4306 314307 : if( entry ) fd_quic_conn_map_remove( state->conn_map, entry );
4307 :
4308 : /* insert into conn map */
4309 314307 : fd_quic_conn_map_t * insert_entry = fd_quic_conn_map_insert( state->conn_map, our_conn_id );
4310 :
4311 : /* if insert failed (should be impossible) fail, and do not remove connection
4312 : from free list */
4313 314307 : if( FD_UNLIKELY( insert_entry == NULL ) ) {
4314 : /* FIXME This has ~1e-6 probability of happening with 10M conns
4315 : Retry generating our_conn_id instead of logging a warning */
4316 0 : FD_LOG_WARNING(( "fd_quic_conn_create failed: failed to register new conn ID" ));
4317 0 : return NULL;
4318 0 : }
4319 :
4320 : /* set connection map insert_entry to new connection */
4321 314307 : insert_entry->conn = conn;
4322 :
4323 : /* remove from free list */
4324 314307 : state->free_conn_list = conn->svc_next;
4325 314307 : conn->svc_next = UINT_MAX;
4326 :
4327 : /* initialize connection members */
4328 314307 : conn->quic = quic;
4329 314307 : conn->server = !!server;
4330 314307 : conn->established = 0;
4331 314307 : conn->called_conn_new = 0;
4332 314307 : conn->svc_type = UINT_MAX;
4333 314307 : conn->svc_time = LONG_MAX;
4334 314307 : conn->our_conn_id = our_conn_id;
4335 314307 : conn->host = (fd_quic_net_endpoint_t){
4336 314307 : .ip_addr = self_ip_addr, /* may be 0, if outgoing */
4337 314307 : .udp_port = self_udp_port,
4338 314307 : };
4339 314307 : memset( &conn->peer[0], 0, sizeof( conn->peer ) );
4340 314307 : conn->conn_gen++;
4341 314307 : conn->token_len = 0;
4342 :
4343 : /* start with smallest value we allow, then allow peer to increase */
4344 314307 : conn->tx_max_datagram_sz = FD_QUIC_INITIAL_PAYLOAD_SZ_MAX;
4345 314307 : conn->handshake_complete = 0;
4346 314307 : conn->handshake_done_send = 0;
4347 314307 : conn->handshake_done_ackd = 0;
4348 314307 : conn->tls_hs = NULL; /* created later */
4349 :
4350 : /* initialize stream_id members */
4351 314307 : fd_quic_conn_stream_rx_t * srx = conn->srx;
4352 314307 : fd_quic_transport_params_t * our_tp = &state->transport_params;
4353 314307 : srx->rx_hi_stream_id = server ? FD_QUIC_STREAM_TYPE_UNI_CLIENT : FD_QUIC_STREAM_TYPE_UNI_SERVER;
4354 314307 : srx->rx_sup_stream_id = server ? FD_QUIC_STREAM_TYPE_UNI_CLIENT : FD_QUIC_STREAM_TYPE_UNI_SERVER;
4355 314307 : conn->tx_next_stream_id = server ? FD_QUIC_STREAM_TYPE_UNI_SERVER : FD_QUIC_STREAM_TYPE_UNI_CLIENT;
4356 314307 : conn->tx_sup_stream_id = server ? FD_QUIC_STREAM_TYPE_UNI_SERVER : FD_QUIC_STREAM_TYPE_UNI_CLIENT;
4357 :
4358 314307 : srx->rx_max_streams_unidir_ackd = 0;
4359 314307 : srx->rx_max_data = our_tp->initial_max_data;
4360 314307 : srx->rx_tot_data = 0;
4361 314307 : srx->rx_streams_active = 0L;
4362 :
4363 314307 : if( state->transport_params.initial_max_streams_uni_present ) {
4364 314307 : srx->rx_sup_stream_id = (state->transport_params.initial_max_streams_uni<<2) + FD_QUIC_STREAM_TYPE_UNI_CLIENT;
4365 314307 : }
4366 314307 : if( state->transport_params.initial_max_data ) {
4367 314307 : srx->rx_max_data = state->transport_params.initial_max_data;
4368 314307 : }
4369 :
4370 : /* points to free tx space */
4371 314307 : conn->tx_ptr = conn->tx_buf_conn;
4372 :
4373 314307 : conn->keys_avail = fd_uint_set_bit( 0U, fd_quic_enc_level_initial_id );
4374 :
4375 : /* rfc9000: s12.3:
4376 : Packet numbers in each packet space start at 0.
4377 : Subsequent packets sent in the same packet number space
4378 : MUST increase the packet number by at least 1
4379 : rfc9002: s3
4380 : It is permitted for some packet numbers to never be used, leaving intentional gaps. */
4381 314307 : memset( conn->exp_pkt_number, 0, sizeof( conn->exp_pkt_number ) );
4382 314307 : memset( conn->last_pkt_number, 0, sizeof( conn->last_pkt_number ) );
4383 314307 : memset( conn->pkt_number, 0, sizeof( conn->pkt_number ) );
4384 :
4385 314307 : memset( conn->hs_sent_bytes, 0, sizeof( conn->hs_sent_bytes ) );
4386 314307 : memset( conn->hs_ackd_bytes, 0, sizeof( conn->hs_ackd_bytes ) );
4387 :
4388 314307 : memset( &conn->secrets, 0, sizeof( conn->secrets ) );
4389 314307 : memset( &conn->keys, 0, sizeof( conn->keys ) );
4390 314307 : memset( &conn->new_keys, 0, sizeof( conn->new_keys ) );
4391 : /* suites initialized above */
4392 :
4393 314307 : conn->key_phase = 0;
4394 314307 : conn->key_update = 0;
4395 :
4396 314307 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_HANDSHAKE );
4397 314307 : conn->reason = 0;
4398 314307 : conn->app_reason = 0;
4399 314307 : conn->flags = 0;
4400 314307 : conn->upd_pkt_number = 0;
4401 :
4402 : /* start with minimum supported max datagram */
4403 : /* peers may allow more */
4404 314307 : conn->tx_max_datagram_sz = FD_QUIC_INITIAL_PAYLOAD_SZ_MAX;
4405 :
4406 : /* initial source connection id */
4407 314307 : conn->initial_source_conn_id = our_conn_id;
4408 :
4409 : /* peer connection id */
4410 314307 : conn->peer_cids[0] = *peer_conn_id;
4411 314307 : conn->peer[0].ip_addr = peer_ip_addr;
4412 314307 : conn->peer[0].udp_port = peer_udp_port;
4413 :
4414 314307 : fd_quic_ack_gen_init( conn->ack_gen );
4415 314307 : conn->unacked_sz = 0UL;
4416 :
4417 : /* flow control params */
4418 314307 : conn->tx_max_data = 0;
4419 :
4420 : /* no stream bytes sent or received yet */
4421 314307 : conn->tx_tot_data = 0;
4422 :
4423 : /* initial rtt */
4424 : /* overridden when acks start returning */
4425 314307 : fd_rtt_estimate_t * rtt = conn->rtt;
4426 :
4427 314307 : ulong peer_ack_delay_exponent = 3UL; /* by spec, default is 3 */
4428 314307 : conn->peer_ack_delay_scale = (float)( 1UL << peer_ack_delay_exponent )
4429 314307 : * (float)quic->config.tick_per_us;
4430 314307 : conn->peer_max_ack_delay_ticks = 0.0f; /* starts at zero, since peers respond immediately to */
4431 : /* INITIAL and HANDSHAKE */
4432 : /* updated when we get transport parameters */
4433 314307 : rtt->smoothed_rtt = FD_QUIC_INITIAL_RTT_US * (float)quic->config.tick_per_us;
4434 314307 : rtt->latest_rtt = FD_QUIC_INITIAL_RTT_US * (float)quic->config.tick_per_us;
4435 314307 : rtt->min_rtt = FD_QUIC_INITIAL_RTT_US * (float)quic->config.tick_per_us;
4436 314307 : rtt->var_rtt = FD_QUIC_INITIAL_RTT_US * (float)quic->config.tick_per_us * 0.5f;
4437 314307 : conn->rtt_period_ticks = FD_QUIC_RTT_PERIOD_US * (float)quic->config.tick_per_us;
4438 :
4439 : /* highest peer encryption level */
4440 314307 : conn->peer_enc_level = 0;
4441 :
4442 : /* idle timeout */
4443 314307 : conn->idle_timeout_ticks = config->idle_timeout;
4444 314307 : conn->last_activity = state->now;
4445 314307 : conn->let_die_ticks = ULONG_MAX;
4446 :
4447 : /* update metrics */
4448 314307 : quic->metrics.conn_active_cnt++;
4449 314307 : quic->metrics.conn_created_cnt++;
4450 :
4451 : /* immediately schedule it */
4452 314307 : fd_quic_svc_schedule( state, conn, FD_QUIC_SVC_WAIT );
4453 :
4454 : /* return connection */
4455 314307 : return conn;
4456 314307 : }
4457 :
4458 : ulong
4459 150147 : fd_quic_get_next_wakeup( fd_quic_t * quic ) {
4460 : /* FIXME not optimized for performance */
4461 150147 : fd_quic_state_t * state = fd_quic_get_state( quic );
4462 150147 : if( state->svc_queue[ FD_QUIC_SVC_INSTANT ].tail != UINT_MAX ) return 0UL;
4463 :
4464 102099 : long ack_wakeup = LONG_MAX;
4465 102099 : long wait_wakeup = LONG_MAX;
4466 102099 : if( state->svc_queue[ FD_QUIC_SVC_ACK_TX ].head != UINT_MAX ) {
4467 47100 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, state->svc_queue[ FD_QUIC_SVC_ACK_TX ].head );
4468 47100 : ack_wakeup = (long)conn->svc_time;
4469 47100 : }
4470 102099 : if( state->svc_queue[ FD_QUIC_SVC_WAIT ].head != UINT_MAX ) {
4471 54996 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, state->svc_queue[ FD_QUIC_SVC_WAIT ].head );
4472 54996 : wait_wakeup = (long)conn->svc_time;
4473 54996 : }
4474 :
4475 102099 : return (ulong)fd_long_max( fd_long_min( ack_wakeup, wait_wakeup ), 0L );
4476 150147 : }
4477 :
4478 : /* frame handling function default definitions */
4479 : static ulong
4480 : fd_quic_handle_padding_frame(
4481 : fd_quic_frame_ctx_t * ctx FD_PARAM_UNUSED,
4482 : fd_quic_padding_frame_t * data FD_PARAM_UNUSED,
4483 : uchar const * const p0,
4484 6021 : ulong p_sz ) {
4485 6021 : uchar const * p = p0;
4486 6021 : uchar const * const p_end = p + p_sz;
4487 5954775 : while( p<p_end && p[0]==0 ) p++;
4488 6021 : return (ulong)( p - p0 );
4489 6021 : }
4490 :
4491 : static ulong
4492 : fd_quic_handle_ping_frame(
4493 : fd_quic_frame_ctx_t * ctx,
4494 : fd_quic_ping_frame_t * data FD_PARAM_UNUSED,
4495 : uchar const * p0,
4496 231 : ulong p_sz ) {
4497 231 : FD_DTRACE_PROBE_1( quic_handle_ping_frame, ctx->conn->our_conn_id );
4498 : /* skip pings and pads */
4499 231 : uchar const * p = p0;
4500 231 : uchar const * const p_end = p + p_sz;
4501 4011 : while( p < p_end && ((uint)p[0] & 0xfeu) == 0 ) p++;
4502 231 : return (ulong)( p - p0 );
4503 231 : }
4504 :
4505 : /* Retry packet metadata
4506 : This will force pkt_meta to be returned to the free list
4507 : for use. It does so by finding unack'ed packet metadata
4508 : and setting the data up for retransmission.
4509 : Set force to 1 to force pkt_meta to be reclaimed even if
4510 : the ack timer hasn't expired. This is used when pkt_meta
4511 : is required immediately and none is available */
4512 : void
4513 : fd_quic_pkt_meta_retry( fd_quic_t * quic,
4514 : fd_quic_conn_t * conn,
4515 : int force,
4516 13354806 : uint arg_enc_level ) {
4517 13354806 : fd_quic_conn_stream_rx_t * srx = conn->srx;
4518 :
4519 13354806 : ulong now = fd_quic_get_state( quic )->now;
4520 :
4521 : /* minimum pkt_meta required to be freed
4522 : If not forcing, 0 is applicable
4523 : Otherwise, we should allow for a normal packet, which
4524 : will likely consist of the following:
4525 : 1 ack
4526 : 1 max streams
4527 : 1 max data
4528 : 1 stream data */
4529 13354806 : ulong min_freed = force ? 4U : 0U;
4530 :
4531 : /* count of freed pkt_meta */
4532 13354806 : ulong cnt_freed = 0u;
4533 :
4534 13354806 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
4535 13354806 : fd_quic_pkt_meta_t * pool = tracker->pool;
4536 :
4537 : /* used for metric tracking */
4538 13354806 : ulong prev_retx_pkt_num[FD_QUIC_NUM_ENC_LEVELS] = { ~0ul, ~0ul, ~0ul, ~0ul };
4539 :
4540 13354821 : while(1) {
4541 : /* find earliest expiring pkt_meta, over smallest pkt number at each enc_level */
4542 13354821 : uint enc_level = arg_enc_level;
4543 13354821 : uint peer_enc_level = conn->peer_enc_level;
4544 13354821 : ulong expiry = ~0ul;
4545 13354821 : if( arg_enc_level == ~0u ) {
4546 66774105 : for( uint j = 0u; j < 4u; ++j ) {
4547 : /* TODO this only checks smallest pkt number,
4548 : assuming that pkt numbers are monotonically increasing
4549 : over time. So it checks in 'sent' time order, but not expiry time. */
4550 53419284 : #if 1
4551 53419284 : fd_quic_pkt_meta_t * pkt_meta = fd_quic_pkt_meta_min( &tracker->sent_pkt_metas[j], pool );
4552 53419284 : if( !pkt_meta ) continue;
4553 :
4554 12895189 : if( enc_level == ~0u || pkt_meta->expiry < expiry ) {
4555 12895189 : enc_level = j;
4556 12895189 : expiry = pkt_meta->expiry;
4557 12895189 : }
4558 : #else
4559 : fd_quic_pkt_meta_t * pkt_meta = pool->sent_pkt_meta[j].head;
4560 : while( pkt_meta ) {
4561 : if( enc_level == ~0u || pkt_meta->expiry < expiry ) {
4562 : enc_level = j;
4563 : expiry = pkt_meta->expiry;
4564 : }
4565 : if( enc_level < peer_enc_level ) break;
4566 : pkt_meta = pkt_meta->next;
4567 : }
4568 : if( enc_level != ~0u ) break;
4569 : #endif
4570 12895189 : }
4571 13354821 : } else {
4572 0 : fd_quic_pkt_meta_t * pkt_meta = fd_quic_pkt_meta_min( &tracker->sent_pkt_metas[enc_level], pool );
4573 0 : if( !pkt_meta ) {
4574 0 : return;
4575 0 : }
4576 :
4577 0 : expiry = pkt_meta->expiry;
4578 0 : }
4579 :
4580 13354821 : if( enc_level == ~0u ) return;
4581 :
4582 12895189 : int exit = 0;
4583 12895189 : if( force ) {
4584 : /* we're forcing, quit when we've freed enough */
4585 0 : if( cnt_freed >= min_freed ) exit = 1;
4586 12895189 : } else {
4587 : /* not forcing, so quit if nothing has expired */
4588 12895189 : if( expiry > now ) {
4589 12895174 : exit = 1;
4590 12895174 : }
4591 12895189 : }
4592 :
4593 12895189 : if( exit ) {
4594 12895174 : if( expiry != ~0ul ) fd_quic_svc_schedule1( conn, FD_QUIC_SVC_WAIT );
4595 12895174 : return;
4596 12895174 : };
4597 :
4598 15 : fd_quic_pkt_meta_t * pkt_meta = fd_quic_pkt_meta_min( &tracker->sent_pkt_metas[enc_level], pool );
4599 :
4600 : /* already moved to another enc_level */
4601 15 : if( enc_level < peer_enc_level ) {
4602 0 : cnt_freed += fd_quic_abandon_enc_level( conn, peer_enc_level );
4603 0 : continue;
4604 0 : }
4605 :
4606 15 : quic->metrics.pkt_retransmissions_cnt += !(pkt_meta->key.pkt_num == prev_retx_pkt_num[enc_level]);
4607 15 : prev_retx_pkt_num[enc_level] = pkt_meta->key.pkt_num;
4608 :
4609 15 : FD_DTRACE_PROBE_4( quic_pkt_meta_retry, conn->our_conn_id, (ulong)pkt_meta->key.pkt_num, pkt_meta->expiry, (uchar)pkt_meta->key.type);
4610 :
4611 : /* set the data to retry */
4612 15 : uint type = pkt_meta->key.type;
4613 15 : switch( type ) {
4614 0 : case FD_QUIC_PKT_META_TYPE_HS_DATA:
4615 0 : do {
4616 0 : ulong offset = fd_ulong_max( conn->hs_ackd_bytes[enc_level], pkt_meta->val.range.offset_lo );
4617 0 : if( offset < conn->hs_sent_bytes[enc_level] ) {
4618 0 : conn->hs_sent_bytes[enc_level] = offset;
4619 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4620 0 : }
4621 0 : } while(0);
4622 0 : break;
4623 :
4624 0 : case FD_QUIC_PKT_META_TYPE_STREAM:
4625 0 : do {
4626 0 : ulong stream_id = pkt_meta->key.stream_id;
4627 :
4628 : /* find the stream */
4629 0 : fd_quic_stream_t * stream = NULL;
4630 0 : fd_quic_stream_map_t * stream_entry = fd_quic_stream_map_query( conn->stream_map, stream_id, NULL );
4631 0 : if( FD_LIKELY( stream_entry && stream_entry->stream &&
4632 0 : ( stream_entry->stream->stream_flags & FD_QUIC_STREAM_FLAGS_DEAD ) == 0 ) ) {
4633 0 : stream = stream_entry->stream;
4634 :
4635 : /* do not try sending data that has been acked */
4636 0 : ulong offset = fd_ulong_max( pkt_meta->val.range.offset_lo, stream->tx_buf.tail );
4637 :
4638 : /* any data left to retry? */
4639 0 : stream->tx_sent = fd_ulong_min( stream->tx_sent, offset );
4640 :
4641 : /* do we have anything to send? */
4642 : /* TODO may need to send fin, also */
4643 0 : if( FD_LIKELY( stream->tx_sent < stream->tx_buf.head ) ) {
4644 :
4645 : /* insert into send list */
4646 0 : FD_QUIC_STREAM_LIST_REMOVE( stream );
4647 0 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->send_streams, stream );
4648 :
4649 : /* set the data to go out on the next packet */
4650 0 : stream->stream_flags |= FD_QUIC_STREAM_FLAGS_UNSENT; /* we have unsent data */
4651 0 : stream->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4652 0 : } else {
4653 : /* fd_quic_tx_stream_free also notifies the user */
4654 0 : fd_quic_tx_stream_free( conn->quic, conn, stream, FD_QUIC_STREAM_NOTIFY_END );
4655 0 : }
4656 0 : }
4657 0 : } while(0);
4658 0 : break;
4659 :
4660 0 : case FD_QUIC_PKT_META_TYPE_HS_DONE:
4661 0 : if( FD_LIKELY( !conn->handshake_done_ackd ) ) {
4662 0 : conn->handshake_done_send = 1;
4663 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4664 0 : }
4665 0 : break;
4666 :
4667 0 : case FD_QUIC_PKT_META_TYPE_MAX_DATA:
4668 0 : if( srx->rx_max_data_ackd < srx->rx_max_data ) {
4669 0 : conn->flags |= FD_QUIC_CONN_FLAGS_MAX_DATA;
4670 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4671 0 : }
4672 0 : break;
4673 :
4674 0 : case FD_QUIC_PKT_META_TYPE_MAX_STREAMS_UNIDIR:
4675 0 : do {
4676 : /* do we still need to send? */
4677 : /* get required value */
4678 0 : ulong max_streams_unidir = srx->rx_sup_stream_id >> 2;
4679 :
4680 0 : if( max_streams_unidir > srx->rx_max_streams_unidir_ackd ) {
4681 : /* set the data to go out on the next packet */
4682 0 : conn->flags |= FD_QUIC_CONN_FLAGS_MAX_STREAMS_UNIDIR;
4683 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4684 0 : }
4685 0 : } while(0);
4686 0 : break;
4687 :
4688 0 : case FD_QUIC_PKT_META_TYPE_CLOSE:
4689 0 : conn->flags &= ~FD_QUIC_CONN_FLAGS_CLOSE_SENT;
4690 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4691 0 : break;
4692 :
4693 15 : case FD_QUIC_PKT_META_TYPE_PING:
4694 15 : conn->flags = ( conn->flags & ~FD_QUIC_CONN_FLAGS_PING_SENT )
4695 15 : | FD_QUIC_CONN_FLAGS_PING;
4696 15 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4697 15 : break;
4698 15 : }
4699 :
4700 : /* reschedule to ensure the data gets processed */
4701 15 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
4702 :
4703 : /* free pkt_meta */
4704 15 : fd_quic_pkt_meta_remove_range( &tracker->sent_pkt_metas[enc_level],
4705 15 : pool,
4706 15 : pkt_meta->key.pkt_num,
4707 15 : pkt_meta->key.pkt_num );
4708 :
4709 15 : conn->used_pkt_meta -= 1;
4710 15 : cnt_freed++;
4711 15 : }
4712 13354806 : }
4713 :
4714 : /* reclaim resources associated with packet metadata
4715 : this is called in response to received acks */
4716 : void
4717 : fd_quic_reclaim_pkt_meta( fd_quic_conn_t * conn,
4718 : fd_quic_pkt_meta_t * pkt_meta,
4719 13130406 : uint enc_level ) {
4720 13130406 : fd_quic_conn_stream_rx_t * srx = conn->srx;
4721 :
4722 13130406 : uint type = pkt_meta->key.type;
4723 13130406 : fd_quic_range_t range = pkt_meta->val.range;
4724 :
4725 13130406 : switch( type ) {
4726 :
4727 3 : case FD_QUIC_PKT_META_TYPE_PING:
4728 3 : do {
4729 3 : conn->flags &= ~( FD_QUIC_CONN_FLAGS_PING | FD_QUIC_CONN_FLAGS_PING_SENT );
4730 3 : } while(0);
4731 3 : break;
4732 :
4733 24987 : case FD_QUIC_PKT_META_TYPE_HS_DATA:
4734 24987 : do {
4735 : /* Note that tls_hs could already be freed */
4736 : /* is this ack'ing the next consecutive bytes?
4737 : if so, we can increase the ack'd bytes
4738 : if not, we retransmit the bytes expected to be ack'd
4739 : we assume a gap means a dropped packet, and
4740 : this policy allows us to free up the pkt_meta here */
4741 24987 : ulong hs_ackd_bytes = conn->hs_ackd_bytes[enc_level];
4742 24987 : if( range.offset_lo <= hs_ackd_bytes ) {
4743 24987 : hs_ackd_bytes = conn->hs_ackd_bytes[enc_level]
4744 24987 : = fd_ulong_max( hs_ackd_bytes, range.offset_hi );
4745 :
4746 : /* remove any unused hs_data */
4747 24987 : fd_quic_tls_hs_data_t * hs_data = NULL;
4748 :
4749 24987 : hs_data = fd_quic_tls_get_hs_data( conn->tls_hs, enc_level );
4750 67134 : while( hs_data && hs_data->offset + hs_data->data_sz <= hs_ackd_bytes ) {
4751 42147 : fd_quic_tls_pop_hs_data( conn->tls_hs, enc_level );
4752 42147 : hs_data = fd_quic_tls_get_hs_data( conn->tls_hs, enc_level );
4753 42147 : }
4754 24987 : } else {
4755 0 : conn->hs_sent_bytes[enc_level] =
4756 0 : fd_ulong_min( conn->hs_sent_bytes[enc_level], hs_ackd_bytes );
4757 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4758 0 : }
4759 24987 : } while(0);
4760 24987 : break;
4761 :
4762 6021 : case FD_QUIC_PKT_META_TYPE_HS_DONE:
4763 6021 : do {
4764 6021 : conn->handshake_done_ackd = 1;
4765 6021 : conn->handshake_done_send = 0;
4766 6021 : if( FD_LIKELY( conn->tls_hs ) ) {
4767 6021 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
4768 6021 : fd_quic_tls_hs_delete( conn->tls_hs );
4769 6021 : fd_quic_tls_hs_cache_ele_remove( &state->hs_cache, conn->tls_hs, state->hs_pool );
4770 6021 : fd_quic_tls_hs_pool_ele_release( state->hs_pool, conn->tls_hs );
4771 6021 : conn->tls_hs = NULL;
4772 6021 : }
4773 6021 : } while(0);
4774 6021 : break;
4775 :
4776 0 : case FD_QUIC_PKT_META_TYPE_MAX_DATA:
4777 0 : do {
4778 0 : ulong max_data_ackd = pkt_meta->val.scalar;
4779 :
4780 : /* ack can only increase max_data_ackd */
4781 0 : max_data_ackd = fd_ulong_max( max_data_ackd, srx->rx_max_data_ackd );
4782 :
4783 : /* max_data_ackd > rx_max_data is a protocol violation */
4784 0 : if( FD_UNLIKELY( max_data_ackd > srx->rx_max_data ) ) {
4785 : /* this is a protocol violation, so inform the peer */
4786 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
4787 0 : return;
4788 0 : }
4789 :
4790 : /* clear flag only if acked value == current value */
4791 0 : if( FD_LIKELY( max_data_ackd == srx->rx_max_data ) ) {
4792 0 : conn->flags &= ~FD_QUIC_CONN_FLAGS_MAX_DATA;
4793 0 : }
4794 :
4795 : /* set the ackd value */
4796 0 : srx->rx_max_data_ackd = max_data_ackd;
4797 0 : } while(0);
4798 0 : break;
4799 :
4800 0 : case FD_QUIC_PKT_META_TYPE_MAX_STREAMS_UNIDIR:
4801 0 : do {
4802 0 : ulong max_streams_unidir_ackd = pkt_meta->val.scalar;
4803 :
4804 : /* ack can only increase max_streams_unidir_ackd */
4805 0 : max_streams_unidir_ackd = fd_ulong_max( max_streams_unidir_ackd, srx->rx_max_streams_unidir_ackd );
4806 :
4807 : /* get required value */
4808 0 : ulong max_streams_unidir = srx->rx_sup_stream_id >> 2;
4809 :
4810 : /* clear flag only if acked value == current value */
4811 0 : if( FD_LIKELY( max_streams_unidir_ackd == max_streams_unidir ) ) {
4812 0 : conn->flags &= ~FD_QUIC_CONN_FLAGS_MAX_STREAMS_UNIDIR;
4813 0 : }
4814 :
4815 : /* set the ackd value */
4816 0 : srx->rx_max_streams_unidir_ackd = max_streams_unidir_ackd;
4817 0 : } while(0);
4818 0 : break;
4819 :
4820 13099392 : case FD_QUIC_PKT_META_TYPE_STREAM:
4821 13099392 : do {
4822 13099392 : ulong stream_id = pkt_meta->key.stream_id;
4823 13099392 : fd_quic_range_t range = pkt_meta->val.range;
4824 :
4825 : /* find the stream */
4826 13099392 : fd_quic_stream_t * stream = NULL;
4827 13099392 : fd_quic_stream_map_t * stream_entry = fd_quic_stream_map_query( conn->stream_map, stream_id, NULL );
4828 13099392 : if( FD_LIKELY( stream_entry && stream_entry->stream &&
4829 13099392 : ( stream_entry->stream->stream_flags & FD_QUIC_STREAM_FLAGS_DEAD ) == 0 ) ) {
4830 13093398 : stream = stream_entry->stream;
4831 :
4832 : /* do not try sending data that has been acked */
4833 :
4834 13093398 : ulong tx_tail = stream->tx_buf.tail;
4835 13093398 : ulong tx_sent = stream->tx_sent;
4836 :
4837 : /* ignore bytes which were already acked */
4838 13093398 : if( range.offset_lo < tx_tail ) range.offset_lo = tx_tail;
4839 :
4840 : /* verify offset_hi */
4841 13093398 : if( FD_UNLIKELY( range.offset_hi > stream->tx_buf.head ) ) {
4842 : /* offset_hi in the pkt_meta (the highest byte offset in the packet */
4843 : /* should never exceed tx_buf.head - the highest byte offset in the */
4844 : /* stream */
4845 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_INTERNAL_ERROR, __LINE__ );
4846 0 : return;
4847 13093398 : } else {
4848 : /* did they ack the first byte in the range? */
4849 13093398 : if( FD_LIKELY( range.offset_lo == tx_tail ) ) {
4850 :
4851 : /* then simply move the tail up */
4852 13093398 : tx_tail = range.offset_hi;
4853 :
4854 : /* need to clear the acks */
4855 13093398 : ulong tx_mask = stream->tx_buf.cap - 1ul;
4856 13093398 : uchar * tx_ack = stream->tx_ack;
4857 1943146029 : for( ulong j = range.offset_lo; j < range.offset_hi; ) {
4858 1930052631 : ulong k = j & tx_mask;
4859 1930052631 : if( ( k & 7ul ) == 0ul && j + 8ul <= range.offset_hi ) {
4860 : /* process 8 bits */
4861 1890587547 : tx_ack[k>>3ul] = 0;
4862 1890587547 : j+=8;
4863 1890587547 : } else {
4864 : /* process 1 bit */
4865 39465084 : tx_ack[k>>3ul] &= (uchar)(0xff ^ ( 1ul << ( k & 7ul ) ) );
4866 39465084 : j++;
4867 39465084 : }
4868 1930052631 : }
4869 13093398 : } else {
4870 : /* set appropriate bits in tx_ack */
4871 : /* TODO optimize this */
4872 0 : ulong tx_mask = stream->tx_buf.cap - 1ul;
4873 0 : ulong cnt = range.offset_hi - range.offset_lo;
4874 0 : uchar * tx_ack = stream->tx_ack;
4875 0 : for( ulong j = 0ul; j < cnt; ) {
4876 0 : ulong k = ( j + range.offset_lo ) & tx_mask;
4877 0 : if( ( k & 7ul ) == 0ul && j + 8ul <= cnt ) {
4878 : /* set whole byte */
4879 0 : tx_ack[k>>3ul] = 0xffu;
4880 :
4881 0 : j += 8ul;
4882 0 : } else {
4883 : /* compiler is not smart enough to know ( 1u << ( k & 7u ) ) fits in a uchar */
4884 0 : tx_ack[k>>3ul] |= (uchar)( 1ul << ( k & 7ul ) );
4885 0 : j++;
4886 0 : }
4887 0 : }
4888 :
4889 : /* determine whether tx_tail may be moved up */
4890 0 : for( ulong j = tx_tail; j < tx_sent; ) {
4891 0 : ulong k = j & tx_mask;
4892 :
4893 : /* can we skip a whole byte? */
4894 0 : if( ( k & 7ul ) == 0ul && j + 8ul <= tx_sent && tx_ack[k>>3ul] == 0xffu ) {
4895 0 : tx_ack[k>>3ul] = 0u;
4896 0 : tx_tail += 8ul;
4897 :
4898 0 : j += 8ul;
4899 0 : } else {
4900 0 : if( tx_ack[k>>3ul] & ( 1u << ( k & 7u ) ) ) {
4901 0 : tx_ack[k>>3ul] = (uchar)( tx_ack[k>>3ul] & ~( 1u << ( k & 7u ) ) );
4902 0 : tx_tail++;
4903 0 : j++;
4904 0 : } else {
4905 0 : break;
4906 0 : }
4907 0 : }
4908 0 : }
4909 0 : }
4910 :
4911 : /* For convenience */
4912 13093398 : uint fin_state_mask = FD_QUIC_STREAM_STATE_TX_FIN | FD_QUIC_STREAM_STATE_RX_FIN;
4913 :
4914 : /* move up tail, and adjust to maintain circular queue invariants, and send
4915 : max_data and max_stream_data, if necessary */
4916 13093398 : if( tx_tail > stream->tx_buf.tail ) {
4917 13093392 : stream->tx_buf.tail = tx_tail;
4918 :
4919 : /* if we have data to send, reschedule */
4920 13093392 : if( fd_quic_buffer_used( &stream->tx_buf ) ) {
4921 16767 : stream->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4922 16767 : if( !FD_QUIC_STREAM_ACTION( stream ) ) {
4923 : /* going from 0 to nonzero, so insert into action list */
4924 9345 : FD_QUIC_STREAM_LIST_REMOVE( stream );
4925 9345 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->send_streams, stream );
4926 9345 : }
4927 :
4928 16767 : stream->stream_flags |= FD_QUIC_STREAM_FLAGS_UNSENT;
4929 :
4930 16767 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
4931 13076625 : } else {
4932 : /* if no data to send, check whether fin bits are set */
4933 13076625 : if( ( stream->state & fin_state_mask ) == fin_state_mask ) {
4934 : /* fd_quic_tx_stream_free also notifies the user */
4935 13076625 : fd_quic_tx_stream_free( conn->quic, conn, stream, FD_QUIC_STREAM_NOTIFY_END );
4936 13076625 : }
4937 13076625 : }
4938 13093392 : } else if( tx_tail == stream->tx_buf.tail &&
4939 6 : ( stream->state & fin_state_mask ) == fin_state_mask ) {
4940 : /* fd_quic_tx_stream_free also notifies the user */
4941 6 : fd_quic_tx_stream_free( conn->quic, conn, stream, FD_QUIC_STREAM_NOTIFY_END );
4942 6 : }
4943 :
4944 : /* we could retransmit (timeout) the bytes which have not been acked (by implication) */
4945 13093398 : }
4946 13093398 : }
4947 13099392 : } while(0);
4948 13099392 : break;
4949 13130406 : }
4950 13130406 : }
4951 : /* process lost packets
4952 : * These packets will be declared lost and relevant data potentially resent */
4953 : void
4954 0 : fd_quic_process_lost( fd_quic_conn_t * conn, uint enc_level, ulong cnt ) {
4955 : /* start at oldest sent */
4956 0 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
4957 0 : fd_quic_pkt_meta_t * pool = tracker->pool;
4958 0 : fd_quic_pkt_meta_ds_t * sent = &tracker->sent_pkt_metas[enc_level];
4959 0 : ulong j = 0;
4960 :
4961 0 : for( fd_quic_pkt_meta_ds_fwd_iter_t iter = fd_quic_pkt_meta_ds_fwd_iter_init( sent, pool );
4962 0 : !fd_quic_pkt_meta_ds_fwd_iter_done( iter );
4963 0 : iter = fd_quic_pkt_meta_ds_fwd_iter_next( iter, pool ) ) {
4964 0 : fd_quic_pkt_meta_t * pkt_meta = fd_quic_pkt_meta_ds_fwd_iter_ele( iter, pool );
4965 0 : if( FD_LIKELY( j < cnt ) ) {
4966 0 : pkt_meta->expiry = 0; /* force expiry */
4967 0 : } else {
4968 0 : break;
4969 0 : }
4970 0 : j++;
4971 0 : }
4972 :
4973 : /* trigger the retries */
4974 0 : fd_quic_pkt_meta_retry( conn->quic, conn, 0 /* don't force */, enc_level );
4975 0 : }
4976 :
4977 : /* process ack range
4978 : applies to pkt_number in [largest_ack - ack_range, largest_ack] */
4979 : void
4980 : fd_quic_process_ack_range( fd_quic_conn_t * conn,
4981 : fd_quic_frame_ctx_t * context,
4982 : uint enc_level,
4983 : ulong largest_ack,
4984 : ulong ack_range,
4985 : int is_largest,
4986 : ulong now,
4987 247939 : ulong ack_delay ) {
4988 : /* FIXME: Close connection if peer ACKed a higher packet number than we sent */
4989 :
4990 247939 : fd_quic_pkt_t * pkt = context->pkt;
4991 :
4992 : /* inclusive range */
4993 247939 : ulong hi = largest_ack;
4994 247939 : ulong lo = largest_ack - ack_range;
4995 247939 : FD_DTRACE_PROBE_4( quic_process_ack_range, conn->our_conn_id, enc_level, lo, hi );
4996 :
4997 247939 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
4998 247939 : fd_quic_pkt_meta_t * pool = tracker->pool;
4999 247939 : fd_quic_pkt_meta_ds_t * sent = &tracker->sent_pkt_metas[enc_level];
5000 :
5001 : /* start at oldest sent */
5002 247939 : for( fd_quic_pkt_meta_ds_fwd_iter_t iter = fd_quic_pkt_meta_ds_idx_ge( sent, lo, pool );
5003 13366300 : !fd_quic_pkt_meta_ds_fwd_iter_done( iter );
5004 13124427 : iter = fd_quic_pkt_meta_ds_fwd_iter_next( iter, pool ) ) {
5005 13124427 : fd_quic_pkt_meta_t * e = fd_quic_pkt_meta_ds_fwd_iter_ele( iter, pool );
5006 13124427 : if( FD_UNLIKELY( e->key.pkt_num > hi ) ) break;
5007 13118361 : if( is_largest && e->key.pkt_num == hi && hi >= pkt->rtt_pkt_number ) {
5008 253750 : pkt->rtt_pkt_number = hi;
5009 253750 : pkt->rtt_ack_time = now - e->tx_time; /* in ticks */
5010 253750 : pkt->rtt_ack_delay = ack_delay; /* in peer units */
5011 253750 : }
5012 13118361 : fd_quic_reclaim_pkt_meta( conn, e, enc_level );
5013 13118361 : }
5014 :
5015 247939 : conn->used_pkt_meta -= fd_quic_pkt_meta_remove_range( sent, pool, lo, hi );
5016 247939 : }
5017 :
5018 : static ulong
5019 : fd_quic_handle_ack_frame( fd_quic_frame_ctx_t * context,
5020 : fd_quic_ack_frame_t * data,
5021 : uchar const * p,
5022 247777 : ulong p_sz ) {
5023 247777 : fd_quic_conn_t * conn = context->conn;
5024 247777 : uint enc_level = context->pkt->enc_level;
5025 :
5026 247777 : if( FD_UNLIKELY( data->first_ack_range > data->largest_ack ) ) {
5027 : /* this is a protocol violation, so inform the peer */
5028 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
5029 0 : return FD_QUIC_PARSE_FAIL;
5030 0 : }
5031 :
5032 247777 : fd_quic_state_t * state = fd_quic_get_state( context->quic );
5033 247777 : conn->last_ack = state->now;
5034 :
5035 : /* track lowest packet acked */
5036 247777 : ulong low_ack_pkt_number = data->largest_ack - data->first_ack_range;
5037 :
5038 : /* process ack range
5039 : applies to pkt_number in [largest_ack - first_ack_range, largest_ack] */
5040 247777 : fd_quic_process_ack_range( conn,
5041 247777 : context,
5042 247777 : enc_level,
5043 247777 : data->largest_ack,
5044 247777 : data->first_ack_range,
5045 247777 : 1 /* is_largest */,
5046 247777 : state->now,
5047 247777 : data->ack_delay );
5048 :
5049 247777 : uchar const * p_str = p;
5050 247777 : uchar const * p_end = p + p_sz;
5051 :
5052 247777 : ulong ack_range_count = data->ack_range_count;
5053 :
5054 : /* cur_pkt_number holds the packet number of the lowest processed
5055 : and acknowledged packet
5056 : This should always be a valid packet number >= 0 */
5057 247777 : ulong cur_pkt_number = data->largest_ack - data->first_ack_range;
5058 :
5059 : /* walk thru ack ranges */
5060 247807 : for( ulong j = 0UL; j < ack_range_count; ++j ) {
5061 54 : if( FD_UNLIKELY( p_end <= p ) ) {
5062 6 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5063 6 : return FD_QUIC_PARSE_FAIL;
5064 6 : }
5065 :
5066 48 : fd_quic_ack_range_frag_t ack_range[1];
5067 48 : ulong rc = fd_quic_decode_ack_range_frag( ack_range, p, (ulong)( p_end - p ) );
5068 48 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
5069 6 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5070 6 : return FD_QUIC_PARSE_FAIL;
5071 6 : }
5072 :
5073 : /* ensure we have ulong local vars, regardless of ack_range definition */
5074 42 : ulong gap = (ulong)ack_range->gap;
5075 42 : ulong length = (ulong)ack_range->length;
5076 :
5077 : /* sanity check before unsigned arithmetic */
5078 42 : if( FD_UNLIKELY( ( gap > ( ~0x3UL ) ) |
5079 42 : ( length > ( ~0x3UL ) ) ) ) {
5080 : /* This is an unreasonably large value, so fail with protocol violation
5081 : It's also likely impossible due to the encoding method */
5082 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
5083 0 : return FD_QUIC_PARSE_FAIL;
5084 0 : }
5085 :
5086 : /* The number of packet numbers to skip (they are not being acked) is
5087 : ack_range->gap + 2
5088 : This is +1 to get from the lowest acked packet to the highest unacked packet
5089 : and +1 because the count of packets in the gap is (ack_range->gap+1) */
5090 42 : ulong skip = gap + 2UL;
5091 :
5092 : /* verify the skip and length values are valid */
5093 42 : if( FD_UNLIKELY( skip + length > cur_pkt_number ) ) {
5094 : /* this is a protocol violation, so inform the peer */
5095 12 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
5096 12 : return FD_QUIC_PARSE_FAIL;
5097 12 : }
5098 :
5099 : /* track lowest */
5100 30 : ulong lo_pkt_number = cur_pkt_number - skip - length;
5101 30 : low_ack_pkt_number = fd_ulong_min( low_ack_pkt_number, lo_pkt_number );
5102 :
5103 : /* process ack range */
5104 30 : fd_quic_process_ack_range( conn,
5105 30 : context,
5106 30 : enc_level,
5107 30 : cur_pkt_number - skip,
5108 30 : length,
5109 30 : 0 /* is_largest */,
5110 30 : state->now,
5111 30 : 0 /* ack_delay not used here */ );
5112 :
5113 : /* Find the next lowest processed and acknowledged packet number
5114 : This should get us to the next lowest processed and acknowledged packet
5115 : number */
5116 30 : cur_pkt_number -= skip + length;
5117 :
5118 30 : p += rc;
5119 30 : }
5120 :
5121 : /* process lost packets */
5122 247753 : {
5123 247753 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
5124 247753 : fd_quic_pkt_meta_t * pool = tracker->pool;
5125 247753 : fd_quic_pkt_meta_ds_t * sent = &tracker->sent_pkt_metas[enc_level];
5126 247753 : fd_quic_pkt_meta_t * min_meta = fd_quic_pkt_meta_min( sent, pool );
5127 :
5128 247753 : if( FD_UNLIKELY( min_meta && min_meta->key.pkt_num < low_ack_pkt_number ) ) {
5129 3 : ulong skipped = 0;
5130 3 : for( fd_quic_pkt_meta_ds_fwd_iter_t iter = fd_quic_pkt_meta_ds_fwd_iter_init( sent, pool );
5131 6 : !fd_quic_pkt_meta_ds_fwd_iter_done( iter );
5132 3 : iter = fd_quic_pkt_meta_ds_fwd_iter_next( iter, pool ) ) {
5133 3 : fd_quic_pkt_meta_t * e = fd_quic_pkt_meta_ds_fwd_iter_ele( iter, pool );
5134 3 : if( FD_UNLIKELY( e->key.pkt_num >= low_ack_pkt_number ) ) break;
5135 3 : skipped++;
5136 3 : }
5137 :
5138 3 : if( FD_UNLIKELY( skipped > 3 ) ) {
5139 0 : fd_quic_process_lost( conn, enc_level, skipped - 3 );
5140 0 : }
5141 3 : }
5142 247753 : }
5143 :
5144 : /* ECN counts
5145 : we currently ignore them, but we must process them to get to the following bytes */
5146 247753 : if( data->type & 1U ) {
5147 0 : if( FD_UNLIKELY( p_end <= p ) ) {
5148 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5149 0 : return FD_QUIC_PARSE_FAIL;
5150 0 : }
5151 :
5152 0 : fd_quic_ecn_counts_frag_t ecn_counts[1];
5153 0 : ulong rc = fd_quic_decode_ecn_counts_frag( ecn_counts, p, (ulong)( p_end - p ) );
5154 0 : if( rc == FD_QUIC_PARSE_FAIL ) {
5155 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5156 0 : return FD_QUIC_PARSE_FAIL;
5157 0 : }
5158 :
5159 0 : p += rc;
5160 0 : }
5161 :
5162 247753 : return (ulong)( p - p_str );
5163 247753 : }
5164 :
5165 : static ulong
5166 : fd_quic_handle_reset_stream_frame(
5167 : fd_quic_frame_ctx_t * context,
5168 : fd_quic_reset_stream_frame_t * data,
5169 : uchar const * p FD_PARAM_UNUSED,
5170 0 : ulong p_sz FD_PARAM_UNUSED ) {
5171 : /* TODO implement */
5172 0 : FD_DTRACE_PROBE_4( quic_handle_reset_stream_frame, context->conn->our_conn_id, data->stream_id, data->app_proto_err_code, data->final_size );
5173 0 : return 0UL;
5174 0 : }
5175 :
5176 : static ulong
5177 : fd_quic_handle_stop_sending_frame(
5178 : fd_quic_frame_ctx_t * context,
5179 : fd_quic_stop_sending_frame_t * data,
5180 : uchar const * p FD_PARAM_UNUSED,
5181 0 : ulong p_sz FD_PARAM_UNUSED ) {
5182 0 : FD_DTRACE_PROBE_3( quic_handle_stop_sending_frame, context->conn->our_conn_id, data->stream_id, data->app_proto_err_code );
5183 0 : return 0UL;
5184 0 : }
5185 :
5186 : static ulong
5187 : fd_quic_handle_new_token_frame(
5188 : fd_quic_frame_ctx_t * context,
5189 : fd_quic_new_token_frame_t * data,
5190 : uchar const * p FD_PARAM_UNUSED,
5191 0 : ulong p_sz FD_PARAM_UNUSED ) {
5192 : /* FIXME A server MUST treat receipt of a NEW_TOKEN frame as a connection error of type PROTOCOL_VIOLATION. */
5193 0 : (void)data;
5194 0 : FD_DTRACE_PROBE_1( quic_handle_new_token_frame, context->conn->our_conn_id );
5195 0 : return 0UL;
5196 0 : }
5197 :
5198 : void
5199 : fd_quic_tx_stream_free( fd_quic_t * quic,
5200 : fd_quic_conn_t * conn,
5201 : fd_quic_stream_t * stream,
5202 13088693 : int code ) {
5203 :
5204 : /* TODO rename FD_QUIC_NOTIFY_END to FD_QUIC_STREAM_NOTIFY_END et al */
5205 13088693 : if( FD_LIKELY( stream->state != FD_QUIC_STREAM_STATE_UNUSED ) ) {
5206 13088693 : fd_quic_cb_stream_notify( quic, stream, stream->context, code );
5207 13088693 : stream->state = FD_QUIC_STREAM_STATE_UNUSED;
5208 13088693 : }
5209 :
5210 13088693 : ulong stream_id = stream->stream_id;
5211 :
5212 : /* remove from stream map */
5213 13088693 : fd_quic_stream_map_t * stream_map = conn->stream_map;
5214 13088693 : fd_quic_stream_map_t * stream_entry = fd_quic_stream_map_query( stream_map, stream_id, NULL );
5215 13088693 : if( FD_LIKELY( stream_entry ) ) {
5216 13088693 : if( FD_LIKELY( stream_entry->stream ) ) {
5217 13088693 : stream_entry->stream->stream_flags = FD_QUIC_STREAM_FLAGS_DEAD;
5218 13088693 : }
5219 13088693 : fd_quic_stream_map_remove( stream_map, stream_entry );
5220 13088693 : }
5221 :
5222 : /* remove from list - idempotent */
5223 13088693 : FD_QUIC_STREAM_LIST_REMOVE( stream );
5224 13088693 : stream->stream_flags = FD_QUIC_STREAM_FLAGS_DEAD;
5225 13088693 : stream->stream_id = ~0UL;
5226 :
5227 : /* add to stream_pool */
5228 13088693 : fd_quic_state_t * state = fd_quic_get_state( quic );
5229 13088693 : fd_quic_stream_pool_free( state->stream_pool, stream );
5230 :
5231 13088693 : }
5232 :
5233 :
5234 : static inline __attribute__((always_inline)) ulong
5235 : fd_quic_handle_stream_frame(
5236 : fd_quic_frame_ctx_t * context,
5237 : uchar const * p,
5238 : ulong p_sz,
5239 : ulong stream_id,
5240 : ulong offset,
5241 : ulong data_sz,
5242 83100596 : int fin ) {
5243 83100596 : fd_quic_t * quic = context->quic;
5244 83100596 : fd_quic_conn_t * conn = context->conn;
5245 83100596 : fd_quic_pkt_t * pkt = context->pkt;
5246 :
5247 83100596 : FD_DTRACE_PROBE_5( quic_handle_stream_frame, conn->our_conn_id, stream_id, offset, data_sz, fin );
5248 :
5249 : /* stream_id type check */
5250 83100596 : ulong stream_type = stream_id & 3UL;
5251 83100596 : if( FD_UNLIKELY( stream_type != ( conn->server ? FD_QUIC_STREAM_TYPE_UNI_CLIENT : FD_QUIC_STREAM_TYPE_UNI_SERVER ) ) ) {
5252 0 : FD_DEBUG( FD_LOG_DEBUG(( "Received forbidden stream type" )); )
5253 : /* Technically should switch between STREAM_LIMIT_ERROR and STREAM_STATE_ERROR here */
5254 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_STREAM_LIMIT_ERROR, __LINE__ );
5255 0 : return FD_QUIC_PARSE_FAIL;
5256 0 : }
5257 :
5258 : /* length check */
5259 83100596 : if( FD_UNLIKELY( data_sz > p_sz ) ) {
5260 0 : FD_DEBUG( FD_LOG_DEBUG(( "Stream header indicates %lu bytes length, but only have %lu", data_sz, p_sz )); )
5261 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5262 0 : return FD_QUIC_PARSE_FAIL;
5263 0 : }
5264 :
5265 83100596 : conn->unacked_sz += data_sz;
5266 :
5267 : /* stream_id outside allowed range - protocol error */
5268 83100596 : if( FD_UNLIKELY( stream_id >= conn->srx->rx_sup_stream_id ) ) {
5269 3 : FD_DEBUG( FD_LOG_DEBUG(( "Stream ID violation detected" )); )
5270 3 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_STREAM_LIMIT_ERROR, __LINE__ );
5271 3 : return FD_QUIC_PARSE_FAIL;
5272 3 : }
5273 :
5274 : /* A receiver MUST close the connection with an error of type FLOW_CONTROL_ERROR if the sender
5275 : violates the advertised connection or stream data limits */
5276 83100593 : if( FD_UNLIKELY( quic->config.initial_rx_max_stream_data < offset + data_sz ) ) {
5277 3 : FD_DEBUG( FD_LOG_DEBUG(( "Stream data limit exceeded" )); )
5278 3 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FLOW_CONTROL_ERROR, __LINE__ );
5279 3 : return FD_QUIC_PARSE_FAIL;
5280 3 : }
5281 :
5282 83100590 : int rx_res = fd_quic_cb_stream_rx( quic, conn, stream_id, offset, p, data_sz, fin );
5283 83100590 : pkt->ack_flag |= fd_uint_if( rx_res==FD_QUIC_SUCCESS, 0U, ACK_FLAG_CANCEL );
5284 :
5285 : /* packet bytes consumed */
5286 83100590 : return data_sz;
5287 83100593 : }
5288 :
5289 : static ulong
5290 : fd_quic_handle_stream_8_frame(
5291 : fd_quic_frame_ctx_t * context,
5292 : fd_quic_stream_8_frame_t * data,
5293 : uchar const * p,
5294 0 : ulong p_sz ) {
5295 0 : return fd_quic_handle_stream_frame( context, p, p_sz, data->stream_id, 0UL, p_sz, data->type&1 );
5296 0 : }
5297 :
5298 : static ulong
5299 : fd_quic_handle_stream_a_frame(
5300 : fd_quic_frame_ctx_t * context,
5301 : fd_quic_stream_a_frame_t * data,
5302 : uchar const * p,
5303 83083733 : ulong p_sz ) {
5304 83083733 : return fd_quic_handle_stream_frame( context, p, p_sz, data->stream_id, 0UL, data->length, data->type&1 );
5305 83083733 : }
5306 :
5307 : static ulong
5308 : fd_quic_handle_stream_c_frame(
5309 : fd_quic_frame_ctx_t * context,
5310 : fd_quic_stream_c_frame_t * data,
5311 : uchar const * p,
5312 0 : ulong p_sz ) {
5313 0 : return fd_quic_handle_stream_frame( context, p, p_sz, data->stream_id, data->offset, p_sz, data->type&1 );
5314 0 : }
5315 :
5316 : static ulong
5317 : fd_quic_handle_stream_e_frame(
5318 : fd_quic_frame_ctx_t * context,
5319 : fd_quic_stream_e_frame_t * data,
5320 : uchar const * p,
5321 16863 : ulong p_sz ) {
5322 16863 : return fd_quic_handle_stream_frame( context, p, p_sz, data->stream_id, data->offset, data->length, data->type&1 );
5323 16863 : }
5324 :
5325 : static ulong
5326 : fd_quic_handle_max_data_frame(
5327 : fd_quic_frame_ctx_t * context,
5328 : fd_quic_max_data_frame_t * data,
5329 : uchar const * p FD_PARAM_UNUSED,
5330 6 : ulong p_sz FD_PARAM_UNUSED ) {
5331 6 : fd_quic_conn_t * conn = context->conn;
5332 :
5333 6 : ulong max_data_old = conn->tx_max_data;
5334 6 : ulong max_data_new = data->max_data;
5335 6 : FD_DTRACE_PROBE_3( quic_handle_max_data_frame, conn->our_conn_id, max_data_new, max_data_old );
5336 :
5337 : /* max data is only allowed to increase the limit. Transgressing frames
5338 : are silently ignored */
5339 6 : conn->tx_max_data = fd_ulong_max( max_data_old, max_data_new );
5340 6 : return 0; /* no additional bytes consumed from buffer */
5341 6 : }
5342 :
5343 : static ulong
5344 : fd_quic_handle_max_stream_data_frame(
5345 : fd_quic_frame_ctx_t * context,
5346 : fd_quic_max_stream_data_frame_t * data,
5347 : uchar const * p FD_PARAM_UNUSED,
5348 0 : ulong p_sz FD_PARAM_UNUSED ) {
5349 : /* FIXME unsupported for now */
5350 0 : FD_DTRACE_PROBE_3( quic_handle_max_stream_data_frame, context->conn->our_conn_id, data->stream_id, data->max_stream_data );
5351 0 : return 0;
5352 0 : }
5353 :
5354 : static ulong
5355 : fd_quic_handle_max_streams_frame(
5356 : fd_quic_frame_ctx_t * context,
5357 : fd_quic_max_streams_frame_t * data,
5358 : uchar const * p FD_PARAM_UNUSED,
5359 9 : ulong p_sz FD_PARAM_UNUSED ) {
5360 9 : fd_quic_conn_t * conn = context->conn;
5361 9 : FD_DTRACE_PROBE_3( quic_handle_max_streams_frame, conn->our_conn_id, data->type, data->max_streams );
5362 :
5363 9 : if( data->type == 0x13 ) {
5364 : /* Only handle unidirectional streams */
5365 6 : ulong type = (ulong)conn->server | 2UL;
5366 6 : ulong peer_sup_stream_id = data->max_streams * 4UL + type;
5367 6 : conn->tx_sup_stream_id = fd_ulong_max( peer_sup_stream_id, conn->tx_sup_stream_id );
5368 6 : }
5369 :
5370 9 : return 0;
5371 9 : }
5372 :
5373 : static ulong
5374 : fd_quic_handle_data_blocked_frame(
5375 : fd_quic_frame_ctx_t * context,
5376 : fd_quic_data_blocked_frame_t * data,
5377 : uchar const * p FD_PARAM_UNUSED,
5378 0 : ulong p_sz FD_PARAM_UNUSED ) {
5379 0 : FD_DTRACE_PROBE_2( quic_handle_data_blocked, context->conn->our_conn_id, data->max_data );
5380 :
5381 : /* Since we do not do runtime allocations, we will not attempt
5382 : to find more memory in the case of DATA_BLOCKED. */
5383 0 : return 0;
5384 0 : }
5385 :
5386 : static ulong
5387 : fd_quic_handle_stream_data_blocked_frame(
5388 : fd_quic_frame_ctx_t * context,
5389 : fd_quic_stream_data_blocked_frame_t * data,
5390 : uchar const * p FD_PARAM_UNUSED,
5391 0 : ulong p_sz FD_PARAM_UNUSED ) {
5392 0 : FD_DTRACE_PROBE_3( quic_handle_stream_data_blocked, context->conn->our_conn_id, data->stream_id, data->max_stream_data );
5393 :
5394 : /* Since we do not do runtime allocations, we will not attempt
5395 : to find more memory in the case of STREAM_DATA_BLOCKED.*/
5396 0 : (void)data;
5397 0 : return 0;
5398 0 : }
5399 :
5400 : static ulong
5401 : fd_quic_handle_streams_blocked_frame(
5402 : fd_quic_frame_ctx_t * context,
5403 : fd_quic_streams_blocked_frame_t * data,
5404 : uchar const * p FD_PARAM_UNUSED,
5405 0 : ulong p_sz FD_PARAM_UNUSED ) {
5406 0 : FD_DTRACE_PROBE_2( quic_handle_streams_blocked_frame, context->conn->our_conn_id, data->max_streams );
5407 :
5408 : /* STREAMS_BLOCKED should be sent by client when it wants
5409 : to use a new stream, but is unable to due to the max_streams
5410 : value
5411 : We can support this in the future, but as of 2024-Dec, the
5412 : Agave TPU client does not currently use it */
5413 0 : return 0;
5414 0 : }
5415 :
5416 : static ulong
5417 : fd_quic_handle_new_conn_id_frame(
5418 : fd_quic_frame_ctx_t * context,
5419 : fd_quic_new_conn_id_frame_t * data,
5420 : uchar const * p FD_PARAM_UNUSED,
5421 0 : ulong p_sz FD_PARAM_UNUSED ) {
5422 : /* FIXME This is a mandatory feature but we don't support it yet */
5423 0 : FD_DTRACE_PROBE_1( quic_handle_new_conn_id_frame, context->conn->our_conn_id );
5424 0 : (void)data;
5425 0 : return 0;
5426 0 : }
5427 :
5428 : static ulong
5429 : fd_quic_handle_retire_conn_id_frame(
5430 : fd_quic_frame_ctx_t * context,
5431 : fd_quic_retire_conn_id_frame_t * data,
5432 : uchar const * p FD_PARAM_UNUSED,
5433 0 : ulong p_sz FD_PARAM_UNUSED ) {
5434 : /* FIXME This is a mandatory feature but we don't support it yet */
5435 0 : FD_DTRACE_PROBE_1( quic_handle_retire_conn_id_frame, context->conn->our_conn_id );
5436 0 : (void)data;
5437 0 : FD_DEBUG( FD_LOG_DEBUG(( "retire_conn_id requested" )); )
5438 0 : return 0;
5439 0 : }
5440 :
5441 : static ulong
5442 : fd_quic_handle_path_challenge_frame(
5443 : fd_quic_frame_ctx_t * context,
5444 : fd_quic_path_challenge_frame_t * data,
5445 : uchar const * p FD_PARAM_UNUSED,
5446 0 : ulong p_sz FD_PARAM_UNUSED ) {
5447 : /* FIXME The recipient of this frame MUST generate a PATH_RESPONSE frame (Section 19.18) containing the same Data value. */
5448 0 : FD_DTRACE_PROBE_1( quic_handle_path_challenge_frame, context->conn->our_conn_id );
5449 0 : (void)data;
5450 0 : return 0UL;
5451 0 : }
5452 :
5453 : static ulong
5454 : fd_quic_handle_path_response_frame(
5455 : fd_quic_frame_ctx_t * context,
5456 : fd_quic_path_response_frame_t * data,
5457 : uchar const * p FD_PARAM_UNUSED,
5458 0 : ulong p_sz FD_PARAM_UNUSED ) {
5459 : /* We don't generate PATH_CHALLENGE frames, so this frame should never arrive */
5460 0 : FD_DTRACE_PROBE_1( quic_handle_path_response_frame, context->conn->our_conn_id );
5461 0 : (void)data;
5462 0 : return 0UL;
5463 0 : }
5464 :
5465 : static void
5466 6015 : fd_quic_handle_conn_close_frame( fd_quic_conn_t * conn ) {
5467 : /* frame type 0x1c means no error, or only error at quic level
5468 : frame type 0x1d means error at application layer
5469 : TODO provide APP with this info */
5470 6015 : FD_DEBUG( FD_LOG_DEBUG(( "peer requested close" )) );
5471 :
5472 6015 : switch( conn->state ) {
5473 0 : case FD_QUIC_CONN_STATE_PEER_CLOSE:
5474 0 : case FD_QUIC_CONN_STATE_ABORT:
5475 12 : case FD_QUIC_CONN_STATE_CLOSE_PENDING:
5476 12 : return;
5477 :
5478 6003 : default:
5479 6003 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_PEER_CLOSE );
5480 6015 : }
5481 :
5482 6003 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
5483 6003 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
5484 6003 : }
5485 :
5486 : static ulong
5487 : fd_quic_handle_conn_close_0_frame(
5488 : fd_quic_frame_ctx_t * context,
5489 : fd_quic_conn_close_0_frame_t * data,
5490 : uchar const * p,
5491 0 : ulong p_sz ) {
5492 0 : (void)p;
5493 :
5494 0 : ulong reason_phrase_length = data->reason_phrase_length;
5495 0 : if( FD_UNLIKELY( reason_phrase_length > p_sz ) ) {
5496 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5497 0 : return FD_QUIC_PARSE_FAIL;
5498 0 : }
5499 :
5500 : /* the information here can be invaluable for debugging */
5501 0 : FD_DEBUG(
5502 0 : char reason_buf[256] = {0};
5503 0 : ulong reason_len = fd_ulong_min( sizeof(reason_buf)-1, reason_phrase_length );
5504 0 : memcpy( reason_buf, p, reason_len );
5505 :
5506 0 : FD_LOG_WARNING(( "fd_quic_handle_conn_close_frame - "
5507 0 : "error_code: %lu "
5508 0 : "frame_type: %lx "
5509 0 : "reason: %s",
5510 0 : data->error_code,
5511 0 : data->frame_type,
5512 0 : reason_buf ));
5513 0 : );
5514 :
5515 0 : fd_quic_handle_conn_close_frame( context->conn );
5516 :
5517 0 : return reason_phrase_length;
5518 0 : }
5519 :
5520 : static ulong
5521 : fd_quic_handle_conn_close_1_frame(
5522 : fd_quic_frame_ctx_t * context,
5523 : fd_quic_conn_close_1_frame_t * data,
5524 : uchar const * p,
5525 6015 : ulong p_sz ) {
5526 6015 : (void)p;
5527 :
5528 6015 : ulong reason_phrase_length = data->reason_phrase_length;
5529 6015 : if( FD_UNLIKELY( reason_phrase_length > p_sz ) ) {
5530 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5531 0 : return FD_QUIC_PARSE_FAIL;
5532 0 : }
5533 :
5534 : /* the information here can be invaluable for debugging */
5535 6015 : FD_DEBUG(
5536 6015 : char reason_buf[256] = {0};
5537 6015 : ulong reason_len = fd_ulong_min( sizeof(reason_buf)-1, reason_phrase_length );
5538 6015 : memcpy( reason_buf, p, reason_len );
5539 :
5540 6015 : FD_LOG_WARNING(( "fd_quic_handle_conn_close_frame - "
5541 6015 : "error_code: %lu "
5542 6015 : "reason: %s",
5543 6015 : data->error_code,
5544 6015 : reason_buf ));
5545 6015 : );
5546 :
5547 6015 : fd_quic_handle_conn_close_frame( context->conn );
5548 :
5549 6015 : return reason_phrase_length;
5550 6015 : }
5551 :
5552 : static ulong
5553 : fd_quic_handle_handshake_done_frame(
5554 : fd_quic_frame_ctx_t * context,
5555 : fd_quic_handshake_done_frame_t * data,
5556 : uchar const * p FD_PARAM_UNUSED,
5557 6021 : ulong p_sz FD_PARAM_UNUSED ) {
5558 6021 : fd_quic_conn_t * conn = context->conn;
5559 6021 : (void)data;
5560 :
5561 : /* servers must treat receipt of HANDSHAKE_DONE as a protocol violation */
5562 6021 : if( FD_UNLIKELY( conn->server ) ) {
5563 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
5564 0 : return FD_QUIC_PARSE_FAIL;
5565 0 : }
5566 :
5567 6021 : if( conn->state == FD_QUIC_CONN_STATE_HANDSHAKE ) {
5568 : /* still handshaking... assume packet was reordered */
5569 0 : context->pkt->ack_flag |= ACK_FLAG_CANCEL;
5570 0 : return 0UL;
5571 6021 : } else if( conn->state != FD_QUIC_CONN_STATE_HANDSHAKE_COMPLETE ) {
5572 : /* duplicate frame or conn closing? */
5573 0 : return 0UL;
5574 0 : }
5575 :
5576 : /* Instantly acknowledge the first HANDSHAKE_DONE frame */
5577 6021 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
5578 :
5579 : /* RFC 9001 4.9.2. Discarding Handshake Keys
5580 : > An endpoint MUST discard its Handshake keys when the
5581 : > TLS handshake is confirmed
5582 : RFC 9001 4.1.2. Handshake Confirmed
5583 : > At the client, the handshake is considered confirmed when a
5584 : > HANDSHAKE_DONE frame is received. */
5585 6021 : fd_quic_abandon_enc_level( conn, fd_quic_enc_level_handshake_id );
5586 :
5587 6021 : if( FD_UNLIKELY( !conn->tls_hs ) ) {
5588 : /* sanity check */
5589 0 : return 0;
5590 0 : }
5591 :
5592 : /* eliminate any remaining hs_data at application level */
5593 6021 : fd_quic_tls_clear_hs_data( conn->tls_hs, fd_quic_enc_level_appdata_id );
5594 :
5595 : /* we shouldn't be receiving this unless handshake is complete */
5596 6021 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_ACTIVE );
5597 :
5598 : /* user callback */
5599 6021 : fd_quic_cb_conn_hs_complete( conn->quic, conn );
5600 :
5601 : /* Deallocate tls_hs once completed */
5602 6021 : if( FD_LIKELY( conn->tls_hs ) ) {
5603 6021 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
5604 6021 : fd_quic_tls_hs_delete( conn->tls_hs );
5605 6021 : fd_quic_tls_hs_cache_ele_remove( &state->hs_cache, conn->tls_hs, state->hs_pool );
5606 6021 : fd_quic_tls_hs_pool_ele_release( state->hs_pool, conn->tls_hs );
5607 6021 : conn->tls_hs = NULL;
5608 6021 : }
5609 :
5610 6021 : return 0;
5611 6021 : }
5612 :
5613 : /* initiate the shutdown of a connection
5614 : may select a reason code */
5615 : void
5616 : fd_quic_conn_close( fd_quic_conn_t * conn,
5617 6030 : uint app_reason ) {
5618 6030 : if( FD_UNLIKELY( !conn ) ) return;
5619 :
5620 6030 : switch( conn->state ) {
5621 0 : case FD_QUIC_CONN_STATE_INVALID:
5622 0 : case FD_QUIC_CONN_STATE_DEAD:
5623 0 : case FD_QUIC_CONN_STATE_ABORT:
5624 0 : return; /* close has no effect in these states */
5625 :
5626 6030 : default:
5627 6030 : {
5628 6030 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_CLOSE_PENDING );
5629 6030 : conn->app_reason = app_reason;
5630 6030 : }
5631 6030 : }
5632 :
5633 : /* set connection to be serviced ASAP */
5634 6030 : fd_quic_svc_schedule1( conn, FD_QUIC_SVC_INSTANT );
5635 6030 : }
5636 :
5637 : void
5638 : fd_quic_conn_let_die( fd_quic_conn_t * conn,
5639 12 : ulong keep_alive_duration_ticks ) {
5640 12 : ulong const now = fd_quic_get_state( conn->quic )->now;
5641 12 : conn->let_die_ticks = fd_ulong_sat_add( now, keep_alive_duration_ticks );
5642 12 : }
|
