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_pkt_meta.h"
7 : #include "fd_quic_private.h"
8 : #include "fd_quic_conn.h"
9 : #include "fd_quic_conn_map.h"
10 : #include "fd_quic_proto.h"
11 : #include "fd_quic_proto.c"
12 : #include "fd_quic_retry.h"
13 : #include "fd_quic_svc_q.h"
14 :
15 : #define FD_TEMPL_FRAME_CTX fd_quic_frame_ctx_t
16 : #include "templ/fd_quic_frame_handler_decl.h"
17 : #include "templ/fd_quic_frames_templ.h"
18 : #include "templ/fd_quic_undefs.h"
19 :
20 : #include "fd_quic_pretty_print.c"
21 :
22 : #include "crypto/fd_quic_crypto_suites.h"
23 : #include "templ/fd_quic_transport_params.h"
24 : #include "templ/fd_quic_parse_util.h"
25 : #include "tls/fd_quic_tls.h"
26 :
27 : #include <fcntl.h> /* for keylog open(2) */
28 : #include <unistd.h> /* for keylog close(2) */
29 :
30 : #include "../../ballet/hex/fd_hex.h"
31 : #include "../../ballet/x509/fd_x509_mock.h"
32 : #include "../../tango/tempo/fd_tempo.h"
33 : #include "../../util/log/fd_dtrace.h"
34 :
35 : #include "../../disco/metrics/generated/fd_metrics_enums.h"
36 :
37 : /* Declare map type for stream_id -> stream* */
38 : #define MAP_NAME fd_quic_stream_map
39 0 : #define MAP_KEY stream_id
40 0 : #define MAP_T fd_quic_stream_map_t
41 0 : #define MAP_KEY_NULL FD_QUIC_STREAM_ID_UNUSED
42 0 : #define MAP_KEY_INVAL(key) ((key)==MAP_KEY_NULL)
43 : #define MAP_QUERY_OPT 1
44 : #include "../../util/tmpl/fd_map_dynamic.c"
45 :
46 :
47 : /* FD_QUIC_MAX_STREAMS_ALWAYS_UNLESS_ACKED */
48 : /* Defines whether a MAX_STREAMS frame is sent even if it was just */
49 : /* sent */
50 : /* They take very little space, and a dropped MAX_STREAMS frame can */
51 : /* be very consequential */
52 : /* Even when set, QUIC won't send this frame if the client has ackd */
53 : /* the most recent value */
54 0 : # define FD_QUIC_MAX_STREAMS_ALWAYS_UNLESS_ACKED 0
55 :
56 : /* Construction API ***************************************************/
57 :
58 : FD_QUIC_API FD_FN_CONST ulong
59 0 : fd_quic_align( void ) {
60 0 : return FD_QUIC_ALIGN;
61 0 : }
62 :
63 : /* fd_quic_footprint_ext returns footprint of QUIC memory region given
64 : limits. Also writes byte offsets to given layout struct. */
65 : static ulong
66 : fd_quic_footprint_ext( fd_quic_limits_t const * limits,
67 0 : fd_quic_layout_t * layout ) {
68 0 : memset( layout, 0, sizeof(fd_quic_layout_t) );
69 0 : if( FD_UNLIKELY( !limits ) ) return 0UL;
70 :
71 0 : ulong conn_cnt = limits->conn_cnt;
72 0 : ulong conn_id_cnt = limits->conn_id_cnt;
73 0 : ulong log_depth = limits->log_depth;
74 0 : ulong handshake_cnt = limits->handshake_cnt;
75 0 : ulong inflight_frame_cnt = limits->inflight_frame_cnt;
76 0 : ulong tx_buf_sz = limits->tx_buf_sz;
77 0 : ulong stream_pool_cnt = limits->stream_pool_cnt;
78 0 : ulong inflight_res_cnt = limits->min_inflight_frame_cnt_conn * conn_cnt;
79 0 : if( FD_UNLIKELY( conn_cnt ==0UL ) ) return 0UL;
80 0 : if( FD_UNLIKELY( handshake_cnt ==0UL ) ) return 0UL;
81 0 : if( FD_UNLIKELY( inflight_frame_cnt==0UL ) ) return 0UL;
82 :
83 0 : if( FD_UNLIKELY( inflight_res_cnt > inflight_frame_cnt ) ) return 0UL;
84 :
85 0 : if( FD_UNLIKELY( conn_id_cnt < FD_QUIC_MIN_CONN_ID_CNT ))
86 0 : return 0UL;
87 :
88 0 : layout->meta_sz = sizeof(fd_quic_layout_t);
89 :
90 0 : ulong offs = 0;
91 :
92 : /* allocate space for fd_quic_t */
93 0 : offs += sizeof(fd_quic_t);
94 :
95 : /* allocate space for state */
96 0 : offs = fd_ulong_align_up( offs, alignof(fd_quic_state_t) );
97 0 : offs += sizeof(fd_quic_state_t);
98 :
99 : /* allocate space for connections */
100 0 : offs = fd_ulong_align_up( offs, fd_quic_conn_align() );
101 0 : layout->conns_off = offs;
102 0 : ulong conn_footprint = fd_quic_conn_footprint( limits );
103 0 : if( FD_UNLIKELY( !conn_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_conn_footprint" )); return 0UL; }
104 0 : layout->conn_footprint = conn_footprint;
105 0 : ulong conn_foot_tot = conn_cnt * conn_footprint;
106 0 : offs += conn_foot_tot;
107 :
108 : /* allocate space for conn IDs */
109 0 : offs = fd_ulong_align_up( offs, fd_quic_conn_map_align() );
110 0 : layout->conn_map_off = offs;
111 0 : ulong slot_cnt_bound = (ulong)( FD_QUIC_DEFAULT_SPARSITY * (double)conn_cnt * (double)conn_id_cnt );
112 0 : int lg_slot_cnt = fd_ulong_find_msb( slot_cnt_bound - 1 ) + 1;
113 0 : layout->lg_slot_cnt = lg_slot_cnt;
114 0 : ulong conn_map_footprint = fd_quic_conn_map_footprint( lg_slot_cnt );
115 0 : if( FD_UNLIKELY( !conn_map_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_conn_map_footprint" )); return 0UL; }
116 0 : offs += conn_map_footprint;
117 :
118 : /* allocate space for handshake pool */
119 0 : offs = fd_ulong_align_up( offs, fd_quic_tls_hs_pool_align() );
120 0 : layout->hs_pool_off = offs;
121 0 : ulong hs_pool_fp = fd_quic_tls_hs_pool_footprint( limits->handshake_cnt );
122 0 : if( FD_UNLIKELY( !hs_pool_fp ) ) { FD_LOG_WARNING(( "invalid fd_quic_tls_hs_pool_footprint" )); return 0UL; }
123 0 : offs += hs_pool_fp;
124 :
125 : /* allocate space for stream pool */
126 0 : if( stream_pool_cnt && tx_buf_sz ) {
127 0 : offs = fd_ulong_align_up( offs, fd_quic_stream_pool_align() );
128 0 : layout->stream_pool_off = offs;
129 0 : ulong stream_pool_footprint = fd_quic_stream_pool_footprint( stream_pool_cnt, tx_buf_sz );
130 0 : if( FD_UNLIKELY( !stream_pool_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_stream_pool_footprint" )); return 0UL; }
131 0 : offs += stream_pool_footprint;
132 0 : } else {
133 0 : layout->stream_pool_off = 0UL;
134 0 : }
135 :
136 : /* allocate space for pkt_meta_pool */
137 0 : if( inflight_frame_cnt ) {
138 0 : offs = fd_ulong_align_up( offs, fd_quic_pkt_meta_pool_align() );
139 0 : layout->pkt_meta_pool_off = offs;
140 0 : ulong pkt_meta_footprint = fd_quic_pkt_meta_pool_footprint( inflight_frame_cnt );
141 0 : if( FD_UNLIKELY( !pkt_meta_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_pkt_meta_pool_footprint" )); return 0UL; }
142 0 : offs += pkt_meta_footprint;
143 0 : } else {
144 0 : layout->pkt_meta_pool_off = 0UL;
145 0 : }
146 :
147 : /* allocate space for quic_log_buf */
148 0 : offs = fd_ulong_align_up( offs, fd_quic_log_buf_align() );
149 0 : layout->log_off = offs;
150 0 : ulong log_footprint = fd_quic_log_buf_footprint( log_depth );
151 0 : if( FD_UNLIKELY( !log_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_log_buf_footprint for depth %lu", log_depth )); return 0UL; }
152 0 : offs += log_footprint;
153 :
154 : /* allocate space for service timers */
155 0 : offs = fd_ulong_align_up( offs, fd_quic_svc_timers_align() );
156 0 : layout->svc_timers_off = offs;
157 0 : ulong svc_timers_footprint = fd_quic_svc_timers_footprint( limits->conn_cnt );
158 0 : if( FD_UNLIKELY( !svc_timers_footprint ) ) { FD_LOG_WARNING(( "invalid fd_quic_svc_timers_footprint" )); return 0UL; }
159 0 : offs += svc_timers_footprint;
160 :
161 0 : return offs;
162 0 : }
163 :
164 : FD_QUIC_API ulong
165 0 : fd_quic_footprint( fd_quic_limits_t const * limits ) {
166 0 : fd_quic_layout_t layout;
167 0 : return fd_quic_footprint_ext( limits, &layout );
168 0 : }
169 :
170 : FD_QUIC_API void *
171 : fd_quic_new( void * mem,
172 0 : fd_quic_limits_t const * limits ) {
173 :
174 : /* Argument checks */
175 :
176 0 : if( FD_UNLIKELY( !mem ) ) {
177 0 : FD_LOG_WARNING(( "NULL mem" ));
178 0 : return NULL;
179 0 : }
180 :
181 0 : ulong align = fd_quic_align();
182 0 : 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 0 : if( FD_UNLIKELY( !limits ) ) {
188 0 : FD_LOG_WARNING(( "NULL limits" ));
189 0 : return NULL;
190 0 : }
191 :
192 0 : if( FD_UNLIKELY( ( limits->conn_cnt ==0UL )
193 0 : | ( limits->conn_cnt >=UINT_MAX )
194 0 : | ( limits->handshake_cnt ==0UL )
195 0 : | ( limits->inflight_frame_cnt==0UL ) ) ) {
196 0 : FD_LOG_WARNING(( "invalid limits" ));
197 0 : return NULL;
198 0 : }
199 :
200 0 : fd_quic_layout_t layout;
201 0 : ulong footprint = fd_quic_footprint_ext( limits, &layout );
202 0 : if( FD_UNLIKELY( !footprint ) ) {
203 0 : FD_LOG_WARNING(( "invalid footprint for config" ));
204 0 : return NULL;
205 0 : }
206 :
207 0 : fd_quic_t * quic = (fd_quic_t *)mem;
208 :
209 : /* Clear fd_quic_t memory region */
210 0 : fd_memset( quic, 0, footprint );
211 :
212 : /* Defaults */
213 0 : quic->config.idle_timeout = FD_QUIC_DEFAULT_IDLE_TIMEOUT;
214 0 : quic->config.ack_delay = FD_QUIC_DEFAULT_ACK_DELAY;
215 0 : quic->config.retry_ttl = FD_QUIC_DEFAULT_RETRY_TTL;
216 0 : quic->config.tls_hs_ttl = FD_QUIC_DEFAULT_TLS_HS_TTL;
217 :
218 : /* Copy layout descriptors */
219 0 : quic->limits = *limits;
220 0 : quic->layout = layout;
221 :
222 : /* Init log buffer (persists across init calls) */
223 0 : void * shmlog = (void *)( (ulong)quic + quic->layout.log_off );
224 0 : if( FD_UNLIKELY( !fd_quic_log_buf_new( shmlog, limits->log_depth ) ) ) {
225 0 : return NULL;
226 0 : }
227 :
228 0 : FD_COMPILER_MFENCE();
229 0 : quic->magic = FD_QUIC_MAGIC;
230 0 : FD_COMPILER_MFENCE();
231 :
232 0 : return quic;
233 0 : }
234 :
235 : FD_QUIC_API fd_quic_limits_t *
236 : fd_quic_limits_from_env( int * pargc,
237 : char *** pargv,
238 0 : fd_quic_limits_t * limits ) {
239 :
240 0 : if( FD_UNLIKELY( !limits ) ) return NULL;
241 :
242 0 : limits->conn_cnt = fd_env_strip_cmdline_ulong( pargc, pargv, "--quic-conns", "QUIC_CONN_CNT", 512UL );
243 0 : limits->conn_id_cnt = fd_env_strip_cmdline_ulong( pargc, pargv, "--quic-conn-ids", "QUIC_CONN_ID_CNT", 16UL );
244 0 : limits->stream_pool_cnt = fd_env_strip_cmdline_uint ( pargc, pargv, "--quic-streams", "QUIC_STREAM_CNT", 8UL );
245 0 : limits->handshake_cnt = fd_env_strip_cmdline_uint ( pargc, pargv, "--quic-handshakes", "QUIC_HANDSHAKE_CNT", 512UL );
246 0 : limits->inflight_frame_cnt = fd_env_strip_cmdline_ulong( pargc, pargv, "--quic-inflight-pkts", "QUIC_MAX_INFLIGHT_PKTS", 2500UL );
247 0 : limits->tx_buf_sz = fd_env_strip_cmdline_ulong( pargc, pargv, "--quic-tx-buf-sz", "QUIC_TX_BUF_SZ", 4096UL );
248 :
249 0 : return limits;
250 0 : }
251 :
252 : FD_QUIC_API fd_quic_config_t *
253 : fd_quic_config_from_env( int * pargc,
254 : char *** pargv,
255 0 : fd_quic_config_t * cfg ) {
256 :
257 0 : if( FD_UNLIKELY( !cfg ) ) return NULL;
258 :
259 0 : char const * keylog_file = fd_env_strip_cmdline_cstr( pargc, pargv, NULL, "SSLKEYLOGFILE", NULL );
260 0 : long idle_timeout_ms = fd_env_strip_cmdline_long( pargc, pargv, "--idle-timeout", NULL, 3000UL );
261 0 : ulong initial_rx_max_stream_data = fd_env_strip_cmdline_ulong(
262 0 : pargc,
263 0 : pargv,
264 0 : "--quic-initial-rx-max-stream-data",
265 0 : "QUIC_INITIAL_RX_MAX_STREAM_DATA",
266 0 : FD_QUIC_DEFAULT_INITIAL_RX_MAX_STREAM_DATA
267 0 : );
268 0 : cfg->retry = fd_env_strip_cmdline_contains( pargc, pargv, "--quic-retry" );
269 :
270 0 : if( keylog_file ) {
271 0 : fd_cstr_ncpy( cfg->keylog_file, keylog_file, sizeof(cfg->keylog_file) );
272 0 : } else {
273 0 : cfg->keylog_file[0]='\0';
274 0 : }
275 :
276 0 : cfg->idle_timeout = idle_timeout_ms * (long)1e6;
277 0 : cfg->initial_rx_max_stream_data = initial_rx_max_stream_data;
278 :
279 0 : return cfg;
280 0 : }
281 :
282 : FD_QUIC_API fd_aio_t const *
283 0 : fd_quic_get_aio_net_rx( fd_quic_t * quic ) {
284 0 : fd_aio_new( &quic->aio_rx, quic, fd_quic_aio_cb_receive );
285 0 : return &quic->aio_rx;
286 0 : }
287 :
288 : FD_QUIC_API void
289 : fd_quic_set_aio_net_tx( fd_quic_t * quic,
290 0 : fd_aio_t const * aio_tx ) {
291 :
292 0 : if( aio_tx ) {
293 0 : quic->aio_tx = *aio_tx;
294 0 : } else {
295 0 : memset( &quic->aio_tx, 0, sizeof(fd_aio_t) );
296 0 : }
297 0 : }
298 :
299 : /* initialize everything that mutates during runtime */
300 : static void
301 0 : fd_quic_stream_init( fd_quic_stream_t * stream ) {
302 0 : stream->context = NULL;
303 :
304 0 : stream->unacked_low = 0;
305 0 : stream->tx_buf.head = 0;
306 0 : stream->tx_sent = 0;
307 :
308 0 : stream->stream_flags = 0;
309 : /* don't update next here, since it's still in use */
310 :
311 0 : stream->state = 0;
312 :
313 0 : stream->tx_max_stream_data = 0;
314 0 : stream->tx_tot_data = 0;
315 :
316 0 : stream->rx_tot_data = 0;
317 :
318 0 : stream->upd_pkt_number = 0;
319 0 : }
320 :
321 : FD_QUIC_API fd_quic_t *
322 0 : fd_quic_join( void * shquic ) {
323 :
324 0 : if( FD_UNLIKELY( !shquic ) ) {
325 0 : FD_LOG_WARNING(( "null shquic" ));
326 0 : return NULL;
327 0 : }
328 0 : if( FD_UNLIKELY( !fd_ulong_is_aligned( (ulong)shquic, FD_QUIC_ALIGN ) ) ) {
329 0 : FD_LOG_WARNING(( "misaligned quic" ));
330 0 : return NULL;
331 0 : }
332 :
333 0 : fd_quic_t * quic = (fd_quic_t *)shquic;
334 0 : if( FD_UNLIKELY( quic->magic != FD_QUIC_MAGIC ) ) {
335 0 : FD_LOG_WARNING(( "bad magic" ));
336 0 : return NULL;
337 0 : }
338 :
339 0 : return quic;
340 0 : }
341 :
342 : FD_QUIC_API void *
343 0 : fd_quic_leave( fd_quic_t * quic ) {
344 0 : return (void *)quic;
345 0 : }
346 :
347 : FD_QUIC_API fd_quic_t *
348 0 : fd_quic_init( fd_quic_t * quic ) {
349 :
350 0 : fd_quic_limits_t const * limits = &quic->limits;
351 0 : fd_quic_config_t * config = &quic->config;
352 :
353 0 : if( FD_UNLIKELY( !config->role ) ) { FD_LOG_WARNING(( "cfg.role not set" )); return NULL; }
354 0 : if( FD_UNLIKELY( !config->idle_timeout ) ) { FD_LOG_WARNING(( "zero cfg.idle_timeout" )); return NULL; }
355 0 : if( FD_UNLIKELY( !config->ack_delay ) ) { FD_LOG_WARNING(( "zero cfg.ack_delay" )); return NULL; }
356 0 : if( FD_UNLIKELY( !config->retry_ttl ) ) { FD_LOG_WARNING(( "zero cfg.retry_ttl" )); return NULL; }
357 :
358 0 : do {
359 0 : ulong x = 0U;
360 0 : for( ulong i=0UL; i<32UL; i++ ) x |= quic->config.identity_public_key[i];
361 :
362 0 : if( FD_UNLIKELY( !x ) ) {
363 0 : FD_LOG_WARNING(( "cfg.identity_public_key not set" ));
364 0 : return NULL;
365 0 : }
366 0 : } while(0);
367 :
368 0 : switch( config->role ) {
369 0 : case FD_QUIC_ROLE_SERVER:
370 0 : if( FD_UNLIKELY( config->keep_alive ) ) { FD_LOG_WARNING(( "server keep-alive not supported" )); return NULL; }
371 0 : break;
372 0 : case FD_QUIC_ROLE_CLIENT:
373 0 : break;
374 0 : default:
375 0 : FD_LOG_WARNING(( "invalid cfg.role" ));
376 0 : return NULL;
377 0 : }
378 :
379 0 : if( FD_UNLIKELY( !config->ack_threshold ) ) {
380 0 : config->ack_threshold = FD_QUIC_DEFAULT_ACK_THRESHOLD;
381 0 : }
382 :
383 0 : fd_quic_layout_t layout = {0};
384 0 : if( FD_UNLIKELY( !fd_quic_footprint_ext( &quic->limits, &layout ) ) ) {
385 0 : FD_LOG_CRIT(( "fd_quic_footprint_ext failed" ));
386 0 : }
387 0 : if( FD_UNLIKELY( 0!=memcmp( &layout, &quic->layout, sizeof(fd_quic_layout_t) ) ) ) {
388 0 : FD_LOG_HEXDUMP_WARNING(( "saved layout", &quic->layout, sizeof(fd_quic_layout_t) ));
389 0 : FD_LOG_HEXDUMP_WARNING(( "derived layout", &layout, sizeof(fd_quic_layout_t) ));
390 0 : FD_LOG_CRIT(( "fd_quic_layout changed. Memory corruption?" ));
391 0 : }
392 :
393 : /* Reset state */
394 :
395 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
396 0 : memset( state, 0, sizeof(fd_quic_state_t) );
397 :
398 0 : void * shmlog = (void *)( (ulong)quic + layout.log_off );
399 0 : if( FD_UNLIKELY( !fd_quic_log_tx_join( state->log_tx, shmlog ) ) ) {
400 0 : FD_LOG_CRIT(( "fd_quic_log_tx_join failed, indicating memory corruption" ));
401 0 : }
402 :
403 : /* State: Initialize packet meta pool */
404 0 : if( layout.pkt_meta_pool_off ) {
405 0 : ulong pkt_meta_cnt = limits->inflight_frame_cnt;
406 0 : ulong pkt_meta_laddr = (ulong)quic + layout.pkt_meta_pool_off;
407 0 : fd_quic_pkt_meta_t * pkt_meta_pool = fd_quic_pkt_meta_pool_new( (void*)pkt_meta_laddr, pkt_meta_cnt );
408 0 : state->pkt_meta_pool = fd_quic_pkt_meta_pool_join( pkt_meta_pool );
409 0 : fd_quic_pkt_meta_ds_init_pool( pkt_meta_pool, pkt_meta_cnt );
410 0 : }
411 :
412 : /* State: initialize each connection, and add to free list */
413 :
414 0 : ulong conn_laddr = (ulong)quic + layout.conns_off;
415 :
416 : /* used for indexing */
417 0 : state->conn_base = conn_laddr;
418 0 : state->conn_sz = layout.conn_footprint;
419 :
420 : /* initialize free_conns */
421 0 : state->free_conn_list = 0;
422 :
423 0 : fd_quic_conn_t _catch[1] = {{.conn_idx = UINT_MAX }};
424 0 : fd_quic_conn_t * last = _catch;
425 0 : for( ulong j = 0; j < limits->conn_cnt; ++j ) {
426 0 : void * conn_mem = (void *)( conn_laddr );
427 0 : conn_laddr += layout.conn_footprint;
428 :
429 0 : fd_quic_conn_t * conn = fd_quic_conn_new( conn_mem, quic, limits );
430 0 : if( FD_UNLIKELY( !conn ) ) {
431 0 : FD_LOG_WARNING(( "NULL conn" ));
432 0 : return NULL;
433 0 : }
434 :
435 : /* used for indexing */
436 0 : conn->conn_idx = (uint)j;
437 0 : conn->free_conn_next = UINT_MAX;
438 :
439 : /* add to free list */
440 0 : last->free_conn_next = (uint)j;
441 :
442 0 : last = conn;
443 0 : }
444 :
445 : /* State: Initialize conn ID map */
446 :
447 0 : ulong conn_map_laddr = (ulong)quic + layout.conn_map_off;
448 0 : state->conn_map = fd_quic_conn_map_join( fd_quic_conn_map_new( (void *)conn_map_laddr, layout.lg_slot_cnt, (ulong)fd_tickcount() ) );
449 0 : if( FD_UNLIKELY( !state->conn_map ) ) {
450 0 : FD_LOG_WARNING(( "NULL conn_map" ));
451 0 : return NULL;
452 0 : }
453 :
454 : /* State: Initialize service queue */
455 0 : ulong svc_base = (ulong)quic + layout.svc_timers_off;
456 0 : state->svc_timers = fd_quic_svc_timers_init( (void *)svc_base, limits->conn_cnt, state );
457 :
458 : /* Check TX AIO */
459 :
460 0 : if( FD_UNLIKELY( !quic->aio_tx.send_func ) ) {
461 0 : FD_LOG_WARNING(( "NULL aio_tx" ));
462 0 : return NULL;
463 0 : }
464 :
465 : /* State: Initialize TLS */
466 :
467 0 : fd_quic_tls_cfg_t tls_cfg = {
468 0 : .max_concur_handshakes = limits->handshake_cnt,
469 :
470 : /* set up callbacks */
471 0 : .secret_cb = fd_quic_tls_cb_secret,
472 0 : .handshake_complete_cb = fd_quic_tls_cb_handshake_complete,
473 0 : .peer_params_cb = fd_quic_tls_cb_peer_params,
474 :
475 0 : .signer = {
476 0 : .ctx = config->sign_ctx,
477 0 : .sign_fn = config->sign,
478 0 : },
479 :
480 0 : .cert_public_key = quic->config.identity_public_key,
481 0 : };
482 :
483 : /* State: Initialize handshake pool */
484 :
485 0 : if( FD_UNLIKELY( !fd_quic_tls_new( state->tls, &tls_cfg ) ) ) {
486 0 : FD_DEBUG( FD_LOG_WARNING( ( "fd_quic_tls_new failed" ) ) );
487 0 : return NULL;
488 0 : }
489 :
490 0 : ulong hs_pool_laddr = (ulong)quic + layout.hs_pool_off;
491 0 : 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 ) );
492 0 : if( FD_UNLIKELY( !hs_pool ) ) {
493 0 : FD_LOG_WARNING(( "fd_quic_tls_hs_pool_new failed" ));
494 0 : return NULL;
495 0 : }
496 0 : state->hs_pool = hs_pool;
497 :
498 : /* State: Initialize TLS handshake cache */
499 0 : if( FD_UNLIKELY( !fd_quic_tls_hs_cache_join( fd_quic_tls_hs_cache_new( &state->hs_cache ) ) ) ) {
500 0 : FD_LOG_WARNING(( "fd_quic_tls_hs_cache_new failed" ));
501 0 : return NULL;
502 0 : }
503 :
504 :
505 0 : if( layout.stream_pool_off ) {
506 0 : ulong stream_pool_cnt = limits->stream_pool_cnt;
507 0 : ulong tx_buf_sz = limits->tx_buf_sz;
508 0 : ulong stream_pool_laddr = (ulong)quic + layout.stream_pool_off;
509 0 : state->stream_pool = fd_quic_stream_pool_new( (void*)stream_pool_laddr, stream_pool_cnt, tx_buf_sz );
510 0 : }
511 :
512 : /* generate a secure random number as seed for fd_rng */
513 0 : uint rng_seed = 0;
514 0 : int rng_seed_ok = !!fd_rng_secure( &rng_seed, sizeof(rng_seed) );
515 0 : if( FD_UNLIKELY( !rng_seed_ok ) ) {
516 0 : FD_LOG_ERR(( "fd_rng_secure failed" ));
517 0 : }
518 0 : fd_rng_new( state->_rng, rng_seed, 0UL );
519 :
520 : /* use rng to generate secret bytes for future RETRY token generation */
521 0 : int rng1_ok = !!fd_rng_secure( state->retry_secret, FD_QUIC_RETRY_SECRET_SZ );
522 0 : int rng2_ok = !!fd_rng_secure( state->retry_iv, FD_QUIC_RETRY_IV_SZ );
523 0 : if( FD_UNLIKELY( !rng1_ok || !rng2_ok ) ) {
524 0 : FD_LOG_ERR(( "fd_rng_secure failed" ));
525 0 : return NULL;
526 0 : }
527 :
528 : /* Initialize transport params */
529 :
530 0 : fd_quic_transport_params_t * tp = &state->transport_params;
531 :
532 : /* initial max streams is zero */
533 : /* we will send max_streams and max_data frames later to allow the peer to */
534 : /* send us data */
535 0 : ulong initial_max_streams_uni = quic->config.role==FD_QUIC_ROLE_SERVER ? 1UL<<60 : 0;
536 0 : ulong initial_max_stream_data = config->initial_rx_max_stream_data;
537 :
538 0 : long max_ack_delay_ns = config->ack_delay * 2L;
539 0 : long max_ack_delay_ms = max_ack_delay_ns / (long)1e6;
540 :
541 0 : long idle_timeout_ns = config->idle_timeout;
542 0 : long idle_timeout_ms = idle_timeout_ns / (long)1e6;
543 :
544 0 : memset( tp, 0, sizeof(fd_quic_transport_params_t) );
545 0 : FD_QUIC_TRANSPORT_PARAM_SET( tp, max_idle_timeout_ms, (ulong)idle_timeout_ms );
546 0 : FD_QUIC_TRANSPORT_PARAM_SET( tp, max_udp_payload_size, FD_QUIC_MAX_PAYLOAD_SZ ); /* TODO */
547 0 : FD_QUIC_TRANSPORT_PARAM_SET( tp, initial_max_data, (1UL<<62)-1UL );
548 0 : FD_QUIC_TRANSPORT_PARAM_SET( tp, initial_max_stream_data_uni, initial_max_stream_data );
549 0 : FD_QUIC_TRANSPORT_PARAM_SET( tp, initial_max_streams_bidi, 0 );
550 0 : FD_QUIC_TRANSPORT_PARAM_SET( tp, initial_max_streams_uni, initial_max_streams_uni );
551 0 : FD_QUIC_TRANSPORT_PARAM_SET( tp, ack_delay_exponent, 0 );
552 0 : FD_QUIC_TRANSPORT_PARAM_SET( tp, max_ack_delay, (ulong)max_ack_delay_ms );
553 0 : /* */tp->disable_active_migration_present = 1;
554 :
555 : /* Compute max inflight pkt cnt per conn */
556 0 : state->max_inflight_frame_cnt_conn = limits->inflight_frame_cnt - limits->min_inflight_frame_cnt_conn * (limits->conn_cnt-1);
557 :
558 0 : return quic;
559 0 : }
560 :
561 : FD_QUIC_API void
562 : fd_quic_set_identity_public_key( fd_quic_t * quic,
563 0 : uchar const public_key[ static 32 ] ) {
564 0 : memcpy( quic->config.identity_public_key, public_key, 32UL );
565 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
566 0 : fd_tls_t * tls = &state->tls->tls;
567 0 : memcpy( tls->cert_public_key, public_key, 32UL );
568 0 : fd_x509_mock_cert( tls->cert_x509, tls->cert_public_key );
569 0 : }
570 :
571 : /* fd_quic_enc_level_to_pn_space maps of encryption level in [0,4) to
572 : packet number space. */
573 : static uint
574 0 : fd_quic_enc_level_to_pn_space( uint enc_level ) {
575 : /* TODO improve this map */
576 0 : static uchar const el2pn_map[] = { 0, 2, 1, 2 };
577 :
578 0 : if( FD_UNLIKELY( enc_level >= 4U ) )
579 0 : FD_LOG_ERR(( "fd_quic_enc_level_to_pn_space called with invalid enc_level" ));
580 :
581 0 : return el2pn_map[ enc_level ];
582 0 : }
583 :
584 : /* This code is directly from rfc9000 A.3 */
585 : FD_FN_CONST ulong
586 : fd_quic_reconstruct_pkt_num( ulong pktnum_comp,
587 : ulong pktnum_sz,
588 0 : ulong exp_pkt_number ) {
589 0 : ulong pn_nbits = pktnum_sz << 3u;
590 0 : ulong pn_win = 1ul << pn_nbits;
591 0 : ulong pn_hwin = pn_win >> 1ul;
592 0 : ulong pn_mask = pn_win - 1ul;
593 : // The incoming packet number should be greater than
594 : // exp_pkt_number - pn_hwin and less than or equal to
595 : // exp_pkt_number + pn_hwin
596 : //
597 : // This means we cannot just strip the trailing bits from
598 : // exp_pkt_number and add the truncated_pn because that might
599 : // yield a value outside the window.
600 : //
601 : // The following code calculates a candidate value and
602 : // makes sure it's within the packet number window.
603 : // Note the extra checks to prevent overflow and underflow.
604 0 : ulong candidate_pn = ( exp_pkt_number & ~pn_mask ) | pktnum_comp;
605 0 : if( candidate_pn + pn_hwin <= exp_pkt_number &&
606 0 : candidate_pn + pn_win < ( 1ul << 62ul ) ) {
607 0 : return candidate_pn + pn_win;
608 0 : }
609 :
610 0 : if( candidate_pn > exp_pkt_number + pn_hwin &&
611 0 : candidate_pn >= pn_win ) {
612 0 : return candidate_pn - pn_win;
613 0 : }
614 :
615 0 : return candidate_pn;
616 0 : }
617 :
618 : /* fd_quic_svc_prep_schedule sets conn->svc_meta.next_timeout to
619 : min of current and provided expiry time. */
620 : static inline void
621 : fd_quic_svc_prep_schedule( fd_quic_conn_t * conn,
622 0 : long expiry ) {
623 0 : conn->svc_meta.next_timeout = fd_long_min( conn->svc_meta.next_timeout, expiry );
624 0 : }
625 :
626 : /* fd_quic_svc_prep_schedule_now sets conn->svc_meta.next_timeout to
627 : current time. For when state is not already available */
628 : static inline void
629 0 : fd_quic_svc_prep_schedule_now( fd_quic_conn_t * conn ) {
630 0 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
631 0 : fd_quic_svc_prep_schedule( conn, state->now );
632 0 : }
633 :
634 : /* Scheduling helper. Retrieves timers from conn to call schedule */
635 : static inline void
636 0 : fd_quic_svc_schedule1( fd_quic_conn_t * conn ) {
637 0 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
638 0 : fd_quic_svc_timers_schedule( state->svc_timers, conn, state->now );
639 0 : }
640 :
641 : /* Validation Helper */
642 : static inline void
643 0 : svc_cnt_eq_alloc_conn( fd_quic_svc_timers_t * timers, fd_quic_t * quic ) {
644 0 : ulong const event_cnt = fd_quic_svc_timers_cnt_events( timers );
645 0 : ulong const conn_cnt = quic->metrics.conn_alloc_cnt;
646 0 : if( FD_UNLIKELY( event_cnt != conn_cnt ) ) {
647 0 : FD_LOG_CRIT(( "only %lu out of %lu connections are in timer", event_cnt, conn_cnt ));
648 0 : }
649 0 : }
650 : /* validates the free conn list doesn't cycle, point nowhere, leak, or point to live conn */
651 : static void
652 0 : fd_quic_conn_free_validate( fd_quic_t * quic ) {
653 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
654 :
655 : /* initialize visited */
656 0 : fd_quic_conn_validate_init( quic );
657 :
658 0 : ulong cnt = 0UL;
659 0 : uint node = state->free_conn_list;
660 0 : while( node!=UINT_MAX ) {
661 0 : FD_TEST( node < quic->limits.conn_cnt );
662 0 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, node );
663 0 : FD_TEST( conn->state == FD_QUIC_CONN_STATE_INVALID );
664 0 : conn->visited = 1U;
665 0 : node = conn->free_conn_next;
666 0 : cnt++;
667 0 : FD_TEST( cnt <= quic->limits.conn_cnt );
668 0 : }
669 :
670 0 : for( ulong j=0UL; j < quic->limits.conn_cnt; j++ ) {
671 0 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, j );
672 0 : FD_TEST( conn->conn_idx==j );
673 0 : if( conn->state == FD_QUIC_CONN_STATE_INVALID ) {
674 0 : FD_TEST( conn->visited );
675 0 : } else {
676 0 : FD_TEST( !conn->visited );
677 0 : }
678 0 : }
679 0 : }
680 :
681 : void
682 0 : fd_quic_state_validate( fd_quic_t * quic ) {
683 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
684 :
685 : /* init visited for svc_timers_validate to use */
686 0 : fd_quic_conn_validate_init( quic );
687 0 : FD_TEST( fd_quic_svc_timers_validate( state->svc_timers, quic ) );
688 :
689 0 : fd_quic_conn_free_validate( quic );
690 0 : }
691 :
692 : fd_quic_conn_t *
693 : fd_quic_conn_query( fd_quic_conn_map_t * map,
694 0 : ulong conn_id ) {
695 0 : fd_quic_conn_map_t sentinel = {0};
696 0 : if( !conn_id ) return NULL;
697 0 : fd_quic_conn_map_t * entry = fd_quic_conn_map_query( map, conn_id, &sentinel );
698 0 : fd_quic_conn_t * conn = entry->conn;
699 0 : if( conn ) {
700 0 : if( FD_UNLIKELY( conn->state==FD_QUIC_CONN_STATE_INVALID ) ) {
701 0 : FD_LOG_CRIT(( "Conn ID %016lx at %p is in map but in free state", conn_id, (void *)conn ));
702 0 : }
703 0 : }
704 0 : return conn;
705 0 : }
706 :
707 : /* Helpers for generating fd_quic_log entries */
708 :
709 : static fd_quic_log_hdr_t
710 0 : fd_quic_log_conn_hdr( fd_quic_conn_t const * conn ) {
711 0 : fd_quic_log_hdr_t hdr = {
712 0 : .conn_id = conn->our_conn_id,
713 0 : .flags = 0
714 0 : };
715 0 : return hdr;
716 0 : }
717 :
718 : static fd_quic_log_hdr_t
719 : fd_quic_log_full_hdr( fd_quic_conn_t const * conn,
720 0 : fd_quic_pkt_t const * pkt ) {
721 0 : fd_quic_log_hdr_t hdr = {
722 0 : .conn_id = conn->our_conn_id,
723 0 : .pkt_num = pkt->pkt_number,
724 0 : .ip4_saddr = pkt->ip4->saddr,
725 0 : .udp_sport = pkt->udp->net_sport,
726 0 : .enc_level = (uchar)pkt->enc_level,
727 0 : .flags = 0
728 0 : };
729 0 : return hdr;
730 0 : }
731 :
732 : inline static void
733 : fd_quic_set_conn_state( fd_quic_conn_t * conn,
734 0 : uint state ) {
735 0 : FD_COMPILER_MFENCE();
736 0 : uint old_state = conn->state;
737 0 : conn->quic->metrics.conn_state_cnt[ old_state ]--;
738 0 : conn->quic->metrics.conn_state_cnt[ state ]++;
739 0 : conn->state = state;
740 0 : FD_COMPILER_MFENCE();
741 0 : }
742 :
743 :
744 : /* fd_quic_conn_error sets the connection state to aborted. This does
745 : not destroy the connection object. Rather, it will eventually cause
746 : the connection to be freed during a later fd_quic_service call.
747 : reason is an RFC 9000 QUIC error code. error_line is the source line
748 : of code in fd_quic.c */
749 :
750 : static void
751 : fd_quic_conn_error1( fd_quic_conn_t * conn,
752 0 : uint reason ) {
753 0 : if( FD_UNLIKELY( !conn || conn->state == FD_QUIC_CONN_STATE_DEAD ) ) return;
754 :
755 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_ABORT );
756 0 : conn->reason = reason;
757 :
758 : /* set connection to be serviced ASAP */
759 0 : fd_quic_svc_prep_schedule_now( conn );
760 0 : fd_quic_svc_schedule1( conn );
761 0 : }
762 :
763 : static void
764 : fd_quic_conn_error( fd_quic_conn_t * conn,
765 : uint reason,
766 0 : uint error_line ) {
767 0 : fd_quic_conn_error1( conn, reason );
768 :
769 0 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
770 :
771 0 : ulong sig = fd_quic_log_sig( FD_QUIC_EVENT_CONN_QUIC_CLOSE );
772 0 : fd_quic_log_error_t * frame = fd_quic_log_tx_prepare( state->log_tx );
773 0 : *frame = (fd_quic_log_error_t) {
774 0 : .hdr = fd_quic_log_conn_hdr( conn ),
775 0 : .code = { reason, 0UL },
776 0 : .src_file = "fd_quic.c",
777 0 : .src_line = error_line,
778 0 : };
779 0 : fd_quic_log_tx_submit( state->log_tx, sizeof(fd_quic_log_error_t), sig, (long)state->now );
780 0 : }
781 :
782 : static void
783 : fd_quic_frame_error( fd_quic_frame_ctx_t const * ctx,
784 : uint reason,
785 0 : uint error_line ) {
786 0 : fd_quic_t * quic = ctx->quic;
787 0 : fd_quic_conn_t * conn = ctx->conn;
788 0 : fd_quic_pkt_t const * pkt = ctx->pkt;
789 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
790 :
791 0 : fd_quic_conn_error1( conn, reason );
792 :
793 0 : uint tls_reason = 0U;
794 0 : if( conn->tls_hs ) tls_reason = conn->tls_hs->hs.base.reason;
795 :
796 0 : ulong sig = fd_quic_log_sig( FD_QUIC_EVENT_CONN_QUIC_CLOSE );
797 0 : fd_quic_log_error_t * frame = fd_quic_log_tx_prepare( state->log_tx );
798 0 : *frame = (fd_quic_log_error_t) {
799 0 : .hdr = fd_quic_log_full_hdr( conn, pkt ),
800 0 : .code = { reason, tls_reason },
801 0 : .src_file = "fd_quic.c",
802 0 : .src_line = error_line,
803 0 : };
804 0 : fd_quic_log_tx_submit( state->log_tx, sizeof(fd_quic_log_error_t), sig, state->now );
805 0 : }
806 :
807 : /* returns the encoding level we should use for the next tx quic packet
808 : or all 1's if nothing to tx */
809 : static uint
810 0 : fd_quic_tx_enc_level( fd_quic_conn_t * conn, int acks ) {
811 0 : uint app_pn_space = fd_quic_enc_level_to_pn_space( fd_quic_enc_level_appdata_id );
812 0 : ulong app_pkt_number = conn->pkt_number[app_pn_space];
813 :
814 : /* fd_quic_tx_enc_level( ... )
815 : check status - if closing, set based on handshake complete
816 : check for acks
817 : find lowest enc level
818 : check for hs_data
819 : find lowest enc level
820 : if any, use lowest
821 : else
822 : if stream data, use 1-rtt
823 : else
824 : nothing to do */
825 :
826 : /* check status */
827 0 : switch( conn->state ) {
828 0 : case FD_QUIC_CONN_STATE_DEAD:
829 : /* do not send on dead connection at all */
830 0 : return ~0u;
831 :
832 0 : case FD_QUIC_CONN_STATE_ABORT:
833 0 : case FD_QUIC_CONN_STATE_CLOSE_PENDING:
834 : /* use handshake or app enc level depending on handshake complete */
835 0 : if( !(conn->flags & FD_QUIC_CONN_FLAGS_CLOSE_SENT ) ) {
836 0 : if( conn->handshake_complete ) {
837 0 : return fd_quic_enc_level_appdata_id;
838 0 : } else if( fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_handshake_id ) ) {
839 0 : return fd_quic_enc_level_handshake_id;
840 0 : } else if( fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_initial_id ) ) {
841 0 : return fd_quic_enc_level_initial_id;
842 0 : }
843 0 : }
844 0 : return ~0u;
845 :
846 : /* TODO consider this optimization... but we want to ack all handshakes, even if there is stream_data */
847 0 : case FD_QUIC_CONN_STATE_ACTIVE:
848 0 : if( FD_LIKELY( !conn->tls_hs ) ) {
849 : /* optimization for case where we have stream data to send */
850 :
851 : /* find stream data to send */
852 0 : fd_quic_stream_t * sentinel = conn->send_streams;
853 0 : fd_quic_stream_t * stream = sentinel->next;
854 0 : if( !stream->sentinel && stream->upd_pkt_number >= app_pkt_number ) {
855 0 : return fd_quic_enc_level_appdata_id;
856 0 : }
857 0 : }
858 0 : }
859 :
860 : /* pick enc_level of oldest ACK not yet sent */
861 0 : fd_quic_ack_gen_t * ack_gen = conn->ack_gen;
862 0 : fd_quic_ack_t const * oldest_ack = fd_quic_ack_queue_ele( ack_gen, ack_gen->tail );
863 0 : uint ack_enc_level = oldest_ack->enc_level; /* speculative load (might be invalid) */
864 0 : if( (ack_gen->head != ack_gen->tail) & acks & fd_uint_extract_bit( conn->keys_avail, (int)ack_enc_level ) ) {
865 0 : return ack_enc_level;
866 0 : }
867 :
868 : /* Check for handshake data to send */
869 0 : uint min_keyed_enc_level = (uint)fd_uint_find_lsb_w_default( conn->keys_avail, (int)~0u );
870 0 : if( FD_UNLIKELY( conn->tls_hs ) ) {
871 0 : fd_quic_tls_hs_data_t * hs_data = NULL;
872 :
873 : /* Starting at min_keyed_enc_level guarantees keys are avail if we return in this loop
874 : - The discard order specified by RFC 9001 Section 4.9 prevents gaps in key availability
875 : - get_hs_data returning non-null means keys were avail at some point for this enc_level */
876 0 : for( uint i = min_keyed_enc_level; i < 4; ++i ) {
877 0 : hs_data = fd_quic_tls_get_hs_data( conn->tls_hs, i );
878 0 : if( hs_data ) {
879 : /* offset within stream */
880 0 : ulong offset = conn->hs_sent_bytes[i];
881 : /* skip packets we've sent */
882 0 : while( hs_data && hs_data->offset + hs_data->data_sz <= offset ) {
883 0 : hs_data = fd_quic_tls_get_next_hs_data( conn->tls_hs, hs_data );
884 0 : }
885 0 : if( hs_data ) {
886 0 : return i;
887 0 : }
888 0 : }
889 0 : }
890 0 : }
891 :
892 : /* handshake done? */
893 0 : if( FD_UNLIKELY( conn->handshake_done_send ) ) return fd_quic_enc_level_appdata_id;
894 :
895 : /* find stream data to send */
896 0 : fd_quic_stream_t * sentinel = conn->send_streams;
897 0 : fd_quic_stream_t * stream = sentinel->next;
898 0 : if( (!stream->sentinel) & (stream->upd_pkt_number >= app_pkt_number) & fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_appdata_id ) ) {
899 0 : return fd_quic_enc_level_appdata_id;
900 0 : }
901 :
902 : /* only allow 1-RTT "flag" frames when we have the keys, to prevent e.g. early 1-RTT PINGs */
903 0 : uint flags_pending = conn->flags & ~(FD_QUIC_CONN_FLAGS_CLOSE_SENT | FD_QUIC_CONN_FLAGS_PING_SENT);
904 0 : if( ( flags_pending != 0U )
905 0 : & ( conn->upd_pkt_number >= app_pkt_number )
906 0 : & fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_appdata_id ) ) {
907 0 : return fd_quic_enc_level_appdata_id;
908 0 : }
909 :
910 : /* nothing to send */
911 0 : return ~0u;
912 0 : }
913 :
914 : /* Include frame code generator */
915 :
916 : #include "templ/fd_quic_frame.c"
917 :
918 : /* handle single v1 frames */
919 : /* returns bytes consumed */
920 : ulong
921 : fd_quic_handle_v1_frame( fd_quic_t * quic,
922 : fd_quic_conn_t * conn,
923 : fd_quic_pkt_t * pkt,
924 : uint pkt_type,
925 : uchar const * buf,
926 0 : ulong buf_sz ) {
927 0 : if( conn->state == FD_QUIC_CONN_STATE_DEAD ) return FD_QUIC_PARSE_FAIL;
928 0 : if( FD_UNLIKELY( buf_sz<1UL ) ) return FD_QUIC_PARSE_FAIL;
929 :
930 : /* Frame ID is technically a varint but it's sufficient to look at the
931 : first byte. */
932 0 : uint id = buf[0];
933 :
934 0 : FD_DTRACE_PROBE_4( quic_handle_frame, id, conn->our_conn_id, pkt_type, pkt->pkt_number );
935 :
936 0 : fd_quic_frame_ctx_t frame_context[1] = {{ quic, conn, pkt }};
937 0 : if( FD_UNLIKELY( !fd_quic_frame_type_allowed( pkt_type, id ) ) ) {
938 0 : FD_DTRACE_PROBE_4( quic_err_frame_not_allowed, id, conn->our_conn_id, pkt_type, pkt->pkt_number );
939 0 : fd_quic_frame_error( frame_context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
940 0 : return FD_QUIC_PARSE_FAIL;
941 0 : }
942 0 : quic->metrics.frame_rx_cnt[ fd_quic_frame_metric_id[ id ] ]++;
943 :
944 0 : pkt->ack_flag |= fd_uint_if( fd_quic_frame_type_flags[ id ]&FD_QUIC_FRAME_FLAG_N, 0U, ACK_FLAG_RQD );
945 :
946 : /* tail call to frame handler */
947 0 : switch( id ) {
948 :
949 0 : # define F(T,MID,NAME,...) \
950 0 : case T: return fd_quic_interpret_##NAME##_frame( frame_context, buf, buf_sz );
951 0 : FD_QUIC_FRAME_TYPES(F)
952 0 : # undef F
953 :
954 0 : default:
955 : /* FIXME this should be unreachable, but gracefully handle this case as defense-in-depth */
956 : /* unknown frame types are PROTOCOL_VIOLATION errors */
957 0 : FD_DEBUG( FD_LOG_DEBUG(( "unexpected frame type: %u", id )); )
958 0 : fd_quic_frame_error( frame_context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
959 0 : return FD_QUIC_PARSE_FAIL;
960 0 : }
961 :
962 0 : }
963 :
964 : fd_quic_t *
965 0 : fd_quic_fini( fd_quic_t * quic ) {
966 :
967 0 : if( FD_UNLIKELY( !quic ) ) {
968 0 : FD_LOG_WARNING(("NULL quic"));
969 0 : return NULL;
970 0 : }
971 :
972 : /* Derive memory layout */
973 :
974 0 : fd_quic_layout_t layout = {0};
975 0 : fd_quic_footprint_ext( &quic->limits, &layout );
976 :
977 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
978 :
979 : /* Free conns */
980 :
981 0 : ulong conn_laddr = (ulong)quic + layout.conns_off;
982 0 : for( ulong i=0; i < quic->limits.conn_cnt; i++ ) {
983 0 : fd_quic_conn_t * conn = (fd_quic_conn_t *)( conn_laddr );
984 0 : conn_laddr += layout.conn_footprint;
985 :
986 0 : if( conn->state ) fd_quic_conn_free( quic, conn );
987 0 : }
988 :
989 : /* Deinit TLS */
990 :
991 0 : fd_quic_tls_hs_pool_delete( fd_quic_tls_hs_pool_leave( state->hs_pool ) ); state->hs_pool = NULL;
992 0 : fd_quic_tls_delete( state->tls );
993 0 : fd_quic_tls_hs_cache_delete( fd_quic_tls_hs_cache_leave( &state->hs_cache ) );
994 :
995 :
996 : /* Delete conn ID map */
997 :
998 0 : fd_quic_conn_map_delete( fd_quic_conn_map_leave( state->conn_map ) );
999 0 : state->conn_map = NULL;
1000 :
1001 : /* Clear join-lifetime memory regions */
1002 :
1003 0 : memset( state, 0, sizeof(fd_quic_state_t) );
1004 :
1005 0 : return quic;
1006 0 : }
1007 :
1008 : void *
1009 0 : fd_quic_delete( fd_quic_t * quic ) {
1010 :
1011 0 : if( FD_UNLIKELY( !quic ) ) {
1012 0 : FD_LOG_WARNING(( "NULL quic" ));
1013 0 : return NULL;
1014 0 : }
1015 :
1016 0 : if( FD_UNLIKELY( !fd_ulong_is_aligned( (ulong)quic, fd_quic_align() ) ) ) {
1017 0 : FD_LOG_WARNING(( "misaligned quic" ));
1018 0 : return NULL;
1019 0 : }
1020 :
1021 0 : if( FD_UNLIKELY( quic->magic!=FD_QUIC_MAGIC ) ) {
1022 0 : FD_LOG_WARNING(( "bad magic" ));
1023 0 : return NULL;
1024 0 : }
1025 :
1026 0 : void * shmlog = (void *)( (ulong)quic + quic->layout.log_off );
1027 0 : if( FD_UNLIKELY( !fd_quic_log_buf_delete( shmlog ) ) ) {
1028 0 : FD_LOG_WARNING(( "fd_quic_log_buf_delete failed" ));
1029 0 : return NULL;
1030 0 : }
1031 :
1032 0 : FD_COMPILER_MFENCE();
1033 0 : FD_VOLATILE( quic->magic ) = 0UL;
1034 0 : FD_COMPILER_MFENCE();
1035 :
1036 0 : return (void *)quic;
1037 0 : }
1038 :
1039 : fd_quic_stream_t *
1040 0 : fd_quic_conn_new_stream( fd_quic_conn_t * conn ) {
1041 0 : if( FD_UNLIKELY( !conn->stream_map ) ) {
1042 : /* QUIC config is receive-only */
1043 0 : return NULL;
1044 0 : }
1045 :
1046 0 : fd_quic_t * quic = conn->quic;
1047 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
1048 0 : if( FD_UNLIKELY( !state->stream_pool ) ) return NULL;
1049 :
1050 0 : ulong next_stream_id = conn->tx_next_stream_id;
1051 :
1052 : /* The user is responsible for calling this, for setting limits, */
1053 : /* and for setting stream_pool size */
1054 : /* Only current use cases for QUIC client is for testing */
1055 : /* So leaving this question unanswered for now */
1056 :
1057 : /* peer imposed limit on streams */
1058 0 : ulong peer_sup_stream_id = conn->tx_sup_stream_id;
1059 :
1060 : /* is connection inactive */
1061 0 : if( FD_UNLIKELY( conn->state != FD_QUIC_CONN_STATE_ACTIVE ||
1062 0 : next_stream_id >= peer_sup_stream_id ) ) {
1063 : /* this is a normal condition which occurs whenever we run up to
1064 : the peer advertised limit and represents one form of flow control */
1065 0 : return NULL;
1066 0 : }
1067 :
1068 : /* obtain a stream from stream_pool */
1069 0 : fd_quic_stream_t * stream = fd_quic_stream_pool_alloc( state->stream_pool );
1070 :
1071 0 : if( FD_UNLIKELY( !stream ) ) {
1072 : /* no streams available in the stream pool */
1073 0 : return NULL;
1074 0 : }
1075 :
1076 : /* add to map of stream ids */
1077 0 : fd_quic_stream_map_t * entry = fd_quic_stream_map_insert( conn->stream_map, next_stream_id );
1078 0 : if( FD_UNLIKELY( !entry ) ) {
1079 : /* return stream to pool */
1080 0 : fd_quic_stream_pool_free( state->stream_pool, stream );
1081 0 : FD_LOG_INFO(( "stream map insert failed" ));
1082 0 : return NULL;
1083 0 : }
1084 :
1085 0 : fd_quic_stream_init( stream );
1086 0 : FD_QUIC_STREAM_LIST_INIT_STREAM( stream );
1087 :
1088 : /* stream tx_buf already set */
1089 0 : stream->conn = conn;
1090 0 : stream->stream_id = next_stream_id;
1091 0 : stream->context = NULL;
1092 :
1093 : /* set the max stream data to the appropriate initial value */
1094 0 : stream->tx_max_stream_data = conn->tx_initial_max_stream_data_uni;
1095 :
1096 : /* set state depending on stream type */
1097 0 : stream->state = FD_QUIC_STREAM_STATE_RX_FIN;
1098 0 : stream->stream_flags = 0u;
1099 :
1100 0 : memset( stream->tx_ack, 0, stream->tx_buf.cap >> 3ul );
1101 :
1102 : /* insert into used streams */
1103 0 : FD_QUIC_STREAM_LIST_REMOVE( stream );
1104 0 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->used_streams, stream );
1105 :
1106 : /* generate a new stream id */
1107 0 : conn->tx_next_stream_id = next_stream_id + 4U;
1108 :
1109 : /* assign the stream to the entry */
1110 0 : entry->stream = stream;
1111 :
1112 : /* update metrics */
1113 0 : quic->metrics.stream_opened_cnt++;
1114 0 : quic->metrics.stream_active_cnt++;
1115 :
1116 0 : FD_DEBUG( FD_LOG_DEBUG(( "Created stream with ID %lu", next_stream_id )) );
1117 0 : return stream;
1118 0 : }
1119 :
1120 : int
1121 : fd_quic_stream_send( fd_quic_stream_t * stream,
1122 : void const * data,
1123 : ulong data_sz,
1124 0 : int fin ) {
1125 0 : if( FD_UNLIKELY( stream->state & FD_QUIC_STREAM_STATE_TX_FIN ) ) {
1126 0 : return FD_QUIC_SEND_ERR_FIN;
1127 0 : }
1128 :
1129 0 : fd_quic_conn_t * conn = stream->conn;
1130 :
1131 0 : fd_quic_buffer_t * tx_buf = &stream->tx_buf;
1132 :
1133 : /* are we allowed to send? */
1134 0 : ulong stream_id = stream->stream_id;
1135 :
1136 : /* stream_id & 2 == 0 is bidir
1137 : stream_id & 1 == 0 is client */
1138 0 : if( FD_UNLIKELY( ( ( (uint)stream_id & 2u ) == 2u ) &
1139 0 : ( ( (uint)stream_id & 1u ) != (uint)conn->server ) ) ) {
1140 0 : return FD_QUIC_SEND_ERR_INVAL_STREAM;
1141 0 : }
1142 :
1143 0 : if( FD_UNLIKELY( conn->state != FD_QUIC_CONN_STATE_ACTIVE ) ) {
1144 0 : if( conn->state == FD_QUIC_CONN_STATE_HANDSHAKE ||
1145 0 : conn->state == FD_QUIC_CONN_STATE_HANDSHAKE_COMPLETE ) {
1146 0 : return FD_QUIC_SEND_ERR_STREAM_STATE;
1147 0 : }
1148 0 : return FD_QUIC_SEND_ERR_INVAL_CONN;
1149 0 : }
1150 :
1151 : /* how many bytes are we allowed to send on the stream and on the connection? */
1152 0 : ulong allowed_stream = stream->tx_max_stream_data - stream->tx_tot_data;
1153 0 : ulong allowed_conn = conn->tx_max_data - conn->tx_tot_data;
1154 0 : ulong allowed = fd_ulong_min( allowed_conn, allowed_stream );
1155 :
1156 0 : if( data_sz > fd_quic_buffer_avail( tx_buf ) ) {
1157 0 : return FD_QUIC_SEND_ERR_FLOW;
1158 0 : }
1159 :
1160 0 : if( data_sz > allowed ) {
1161 0 : return FD_QUIC_SEND_ERR_FLOW;
1162 0 : }
1163 :
1164 : /* store data from data into tx_buf */
1165 0 : fd_quic_buffer_store( tx_buf, data, data_sz );
1166 :
1167 : /* adjust flow control limits on stream and connection */
1168 0 : stream->tx_tot_data += data_sz;
1169 0 : conn->tx_tot_data += data_sz;
1170 :
1171 : /* insert into send list */
1172 0 : if( !FD_QUIC_STREAM_ACTION( stream ) ) {
1173 0 : FD_QUIC_STREAM_LIST_REMOVE( stream );
1174 0 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->send_streams, stream );
1175 0 : }
1176 0 : stream->stream_flags |= FD_QUIC_STREAM_FLAGS_UNSENT; /* we have unsent data */
1177 0 : stream->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING; /* schedule tx */
1178 :
1179 : /* don't actually set fin flag if we didn't add the last
1180 : byte to the buffer */
1181 0 : if( fin ) {
1182 0 : fd_quic_stream_fin( stream );
1183 0 : }
1184 :
1185 : /* schedule send */
1186 0 : fd_quic_svc_prep_schedule_now( conn );
1187 0 : fd_quic_svc_schedule1( conn );
1188 :
1189 0 : return FD_QUIC_SUCCESS;
1190 0 : }
1191 :
1192 : void
1193 0 : fd_quic_stream_fin( fd_quic_stream_t * stream ) {
1194 0 : if( FD_UNLIKELY( stream->state & FD_QUIC_STREAM_STATE_TX_FIN ) ) {
1195 0 : return;
1196 0 : }
1197 :
1198 0 : fd_quic_conn_t * conn = stream->conn;
1199 :
1200 : /* insert into send list */
1201 0 : if( !FD_QUIC_STREAM_ACTION( stream ) ) {
1202 0 : FD_QUIC_STREAM_LIST_REMOVE( stream );
1203 0 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->send_streams, stream );
1204 0 : }
1205 0 : stream->stream_flags |= FD_QUIC_STREAM_FLAGS_TX_FIN; /* state immediately updated */
1206 0 : stream->state |= FD_QUIC_STREAM_STATE_TX_FIN; /* state immediately updated */
1207 0 : stream->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING; /* update to be sent in next packet */
1208 :
1209 : /* TODO update metrics */
1210 0 : }
1211 :
1212 : void
1213 0 : fd_quic_conn_set_rx_max_data( fd_quic_conn_t * conn, ulong rx_max_data ) {
1214 : /* cannot reduce max_data, and cannot increase beyond max varint */
1215 0 : if( rx_max_data > conn->srx->rx_max_data && rx_max_data < (1UL<<62)-1UL ) {
1216 0 : conn->srx->rx_max_data = rx_max_data;
1217 0 : conn->flags |= FD_QUIC_CONN_FLAGS_MAX_DATA;
1218 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
1219 0 : fd_quic_svc_prep_schedule_now( conn );
1220 0 : fd_quic_svc_schedule1( conn );
1221 0 : }
1222 0 : }
1223 :
1224 : /* packet processing */
1225 :
1226 : /* fd_quic_conn_free_pkt_meta frees all pkt_meta associated with
1227 : enc_level. Returns the number of freed pkt_meta. */
1228 : static ulong
1229 : fd_quic_conn_free_pkt_meta( fd_quic_conn_t * conn,
1230 0 : uint enc_level ) {
1231 0 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
1232 0 : fd_quic_pkt_meta_t * pool = tracker->pool;
1233 0 : fd_quic_pkt_meta_ds_t * sent = &tracker->sent_pkt_metas[enc_level];
1234 :
1235 0 : fd_quic_pkt_meta_t * prev = NULL;
1236 0 : for( fd_quic_pkt_meta_ds_fwd_iter_t iter = fd_quic_pkt_meta_ds_fwd_iter_init( sent, pool );
1237 0 : !fd_quic_pkt_meta_ds_fwd_iter_done( iter );
1238 0 : iter = fd_quic_pkt_meta_ds_fwd_iter_next( iter, pool ) ) {
1239 0 : fd_quic_pkt_meta_t * e = fd_quic_pkt_meta_ds_fwd_iter_ele( iter, pool );
1240 0 : if( FD_LIKELY( prev ) ) {
1241 0 : fd_quic_pkt_meta_pool_ele_release( pool, prev );
1242 0 : }
1243 0 : prev = e;
1244 0 : }
1245 0 : if( FD_LIKELY( prev ) ) {
1246 0 : fd_quic_pkt_meta_pool_ele_release( pool, prev );
1247 0 : }
1248 :
1249 :
1250 : /* Instead of paying log(n) to maintain the treap structure during ele_remove
1251 : on each iteration, we've returned all pkt_meta to the pool, but left them
1252 : in the treap. Then, we reinitialize the treap, in constant time deleting
1253 : all elements that were in the 'old treap'. This function now runs in linear
1254 : time, instead of O(n*lg(n)). */
1255 :
1256 0 : ulong pre_ele_cnt = fd_quic_pkt_meta_ds_ele_cnt( sent );
1257 :
1258 0 : fd_quic_pkt_meta_ds_clear( tracker, enc_level );
1259 0 : conn->used_pkt_meta -= pre_ele_cnt;
1260 :
1261 0 : return pre_ele_cnt;
1262 0 : }
1263 :
1264 : /* reclaim and free all pkt_meta for a given encryption level,
1265 : and return the number affected. */
1266 : static ulong
1267 0 : fd_quic_reclaim_pkt_meta_level( fd_quic_conn_t * conn, uint enc_level ) {
1268 0 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
1269 0 : fd_quic_pkt_meta_t * pool = tracker->pool;
1270 0 : fd_quic_pkt_meta_ds_t * sent = &tracker->sent_pkt_metas[enc_level];
1271 :
1272 0 : for( fd_quic_pkt_meta_ds_fwd_iter_t iter = fd_quic_pkt_meta_ds_fwd_iter_init( sent, pool );
1273 0 : !fd_quic_pkt_meta_ds_fwd_iter_done( iter );
1274 0 : iter = fd_quic_pkt_meta_ds_fwd_iter_next( iter, pool ) ) {
1275 0 : fd_quic_pkt_meta_t * e = fd_quic_pkt_meta_ds_fwd_iter_ele( iter, pool );
1276 0 : fd_quic_reclaim_pkt_meta( conn, e, enc_level );
1277 0 : }
1278 :
1279 0 : return fd_quic_conn_free_pkt_meta( conn, enc_level );
1280 0 : }
1281 :
1282 :
1283 : /* fd_quic_abandon_enc_level frees all resources associated encryption
1284 : levels less or equal to enc_level. Returns the number of freed
1285 : pkt_meta. */
1286 :
1287 : ulong
1288 : fd_quic_abandon_enc_level( fd_quic_conn_t * conn,
1289 0 : uint enc_level ) {
1290 0 : if( FD_LIKELY( !fd_uint_extract_bit( conn->keys_avail, (int)enc_level ) ) ) return 0UL;
1291 0 : FD_DEBUG( FD_LOG_DEBUG(( "conn=%p abandoning enc_level=%u", (void *)conn, enc_level )); )
1292 :
1293 0 : fd_quic_ack_gen_abandon_enc_level( conn->ack_gen, enc_level );
1294 :
1295 0 : ulong freed = 0UL;
1296 0 : for( uint j = 0; j <= enc_level; ++j ) {
1297 0 : conn->keys_avail = fd_uint_clear_bit( conn->keys_avail, (int)j );
1298 : /* treat all packets as ACKed (freeing handshake data, etc.) */
1299 0 : freed += fd_quic_reclaim_pkt_meta_level( conn, j );
1300 0 : }
1301 :
1302 0 : return freed;
1303 0 : }
1304 :
1305 : static void
1306 : fd_quic_gen_initial_secret_and_keys(
1307 : fd_quic_conn_t * conn,
1308 : fd_quic_conn_id_t const * dst_conn_id,
1309 0 : int is_server ) {
1310 :
1311 0 : fd_quic_gen_initial_secrets(
1312 0 : &conn->secrets,
1313 0 : dst_conn_id->conn_id, dst_conn_id->sz,
1314 0 : is_server );
1315 :
1316 0 : fd_quic_gen_keys(
1317 0 : &conn->keys[ fd_quic_enc_level_initial_id ][ 0 ],
1318 0 : conn->secrets.secret[ fd_quic_enc_level_initial_id ][ 0 ] );
1319 :
1320 0 : fd_quic_gen_keys(
1321 0 : &conn->keys[ fd_quic_enc_level_initial_id ][ 1 ],
1322 0 : conn->secrets.secret[ fd_quic_enc_level_initial_id ][ 1 ] );
1323 0 : }
1324 :
1325 : static ulong
1326 : fd_quic_send_retry( fd_quic_t * quic,
1327 : fd_quic_pkt_t * pkt,
1328 : fd_quic_conn_id_t const * odcid,
1329 : fd_quic_conn_id_t const * scid,
1330 0 : ulong new_conn_id ) {
1331 :
1332 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
1333 :
1334 0 : long expire_at = state->now + quic->config.retry_ttl;
1335 0 : uchar retry_pkt[ FD_QUIC_RETRY_LOCAL_SZ ];
1336 0 : 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 );
1337 :
1338 0 : quic->metrics.retry_tx_cnt++;
1339 :
1340 0 : uchar * tx_ptr = retry_pkt + retry_pkt_sz;
1341 0 : if( FD_UNLIKELY( fd_quic_tx_buffered_raw(
1342 0 : quic,
1343 : // these are state variable's normally updated on a conn, but irrelevant in retry so we
1344 : // just size it exactly as the encoded retry packet
1345 0 : &tx_ptr,
1346 0 : retry_pkt,
1347 : // encode buffer
1348 0 : &pkt->ip4->net_id,
1349 0 : pkt->ip4->saddr,
1350 0 : pkt->udp->net_sport,
1351 0 : pkt->ip4->daddr,
1352 0 : pkt->udp->net_dport ) == FD_QUIC_FAILED ) ) {
1353 0 : return FD_QUIC_PARSE_FAIL;
1354 0 : }
1355 0 : return 0UL;
1356 0 : }
1357 :
1358 : /* fd_quic_tls_hs_cache_evict evicts the oldest tls_hs if it's exceeded its ttl
1359 : Assumes cache is non-empty
1360 : and returns 1 if evicted, otherwise returns 0. */
1361 : static int
1362 : fd_quic_tls_hs_cache_evict( fd_quic_t * quic,
1363 0 : fd_quic_state_t * state ) {
1364 :
1365 0 : fd_quic_tls_hs_t* hs_to_free = fd_quic_tls_hs_cache_ele_peek_head( &state->hs_cache, state->hs_pool );
1366 :
1367 0 : if( state->now < hs_to_free->birthtime + quic->config.tls_hs_ttl ) {
1368 : /* oldest is too young to evict */
1369 0 : quic->metrics.hs_err_alloc_fail_cnt++;
1370 0 : return 0;
1371 0 : }
1372 :
1373 0 : fd_quic_conn_free( quic, hs_to_free->context );
1374 0 : quic->metrics.hs_evicted_cnt++;
1375 0 : return 1;
1376 0 : }
1377 :
1378 : /* fd_quic_handle_v1_initial handles an "Initial"-type packet.
1379 : Valid for both server and client. Initial packets are used to
1380 : establish QUIC conns and wrap the TLS handshake flow among other
1381 : things. */
1382 :
1383 : ulong
1384 : fd_quic_handle_v1_initial( fd_quic_t * quic,
1385 : fd_quic_conn_t ** p_conn,
1386 : fd_quic_pkt_t * pkt,
1387 : fd_quic_conn_id_t const * dcid,
1388 : fd_quic_conn_id_t const * peer_scid,
1389 : uchar * cur_ptr,
1390 0 : ulong cur_sz ) {
1391 0 : fd_quic_conn_t * conn = *p_conn;
1392 :
1393 0 : if( FD_UNLIKELY( conn && !fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_initial_id ) ) ) {
1394 0 : quic->metrics.pkt_no_key_cnt[ fd_quic_enc_level_initial_id ]++;
1395 0 : return FD_QUIC_PARSE_FAIL;
1396 0 : }
1397 :
1398 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
1399 0 : fd_quic_metrics_t * metrics = &quic->metrics;
1400 :
1401 : /* Initial packets are de-facto unencrypted. Packet protection is
1402 : still applied, albeit with publicly known encryption keys.
1403 :
1404 : RFC 9001 specifies use of the TLS_AES_128_GCM_SHA256_ID suite for
1405 : initial secrets and keys. */
1406 :
1407 : /* Parse initial packet */
1408 :
1409 0 : fd_quic_initial_t initial[1] = {0};
1410 0 : ulong rc = fd_quic_decode_initial( initial, cur_ptr, cur_sz );
1411 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
1412 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_decode_initial failed" )) );
1413 0 : return FD_QUIC_PARSE_FAIL;
1414 0 : }
1415 :
1416 : /* Check bounds on initial */
1417 :
1418 : /* len indicated the number of bytes after the packet number offset
1419 : so verify this value is within the packet */
1420 0 : ulong pn_offset = initial->pkt_num_pnoff;
1421 0 : ulong body_sz = initial->len; /* length of packet number, frames, and auth tag */
1422 0 : ulong tot_sz = pn_offset + body_sz;
1423 0 : if( FD_UNLIKELY( tot_sz > cur_sz ) ) {
1424 0 : FD_DEBUG( FD_LOG_DEBUG(( "Bogus initial packet length" )) );
1425 0 : return FD_QUIC_PARSE_FAIL;
1426 0 : }
1427 :
1428 : /* count received token len */
1429 0 : int const token_len_match = initial->token_len == sizeof(fd_quic_retry_token_t);
1430 0 : ulong const token_len_idx = fd_ulong_if( !!initial->token_len,
1431 0 : fd_ulong_if( token_len_match, 1, 2 ),
1432 0 : 0 );
1433 0 : metrics->initial_token_len_cnt[ token_len_idx ]++;
1434 :
1435 : /* Check it is valid for a token to be present in an initial packet in the current context.
1436 :
1437 : quic->config.role == FD_QUIC_ROLE_CLIENT
1438 : - Indicates the client received an initial packet with a token from a server. "Initial packets
1439 : sent by the server MUST set the Token Length field to 0; clients that receive an Initial packet
1440 : with a non-zero Token Length field MUST either discard the packet or generate a connection
1441 : error of type PROTOCOL_VIOLATION (RFC 9000, Section 17.2.2)"
1442 :
1443 : quic->config.retry == false
1444 : - Indicates the server is not configured to retry, but a client attached a token to this
1445 : initial packet. NEW_TOKEN frames are not supported, so this implementation treats the presence
1446 : of a token when retry is disabled as an error. */
1447 0 : if( FD_UNLIKELY( initial->token_len > 0 &&
1448 0 : ( quic->config.role == FD_QUIC_ROLE_CLIENT || !quic->config.retry ) ) ) {
1449 0 : FD_DEBUG( FD_LOG_DEBUG(( "Rejecting initial with token" )); )
1450 0 : return FD_QUIC_PARSE_FAIL;
1451 0 : }
1452 :
1453 :
1454 0 : ulong scid; /* outgoing scid */
1455 0 : fd_quic_conn_id_t odcid; /* dst conn id from client's original Initial */
1456 :
1457 : /* Do we have a conn object for this dest conn ID?
1458 : If not, sanity check, send/verify retry if needed */
1459 0 : if( FD_UNLIKELY( !conn ) ) {
1460 : /* if we're a client, and no conn, discard */
1461 0 : if( quic->config.role == FD_QUIC_ROLE_CLIENT ) {
1462 : /* connection may have been torn down */
1463 0 : FD_DEBUG( FD_LOG_DEBUG(( "unknown connection ID" )); )
1464 0 : metrics->pkt_no_conn_cnt[ fd_quic_enc_level_initial_id ]++;
1465 0 : FD_DTRACE_PROBE_2( fd_quic_handle_v1_initial_no_conn , state->now, pkt->pkt_number );
1466 0 : return FD_QUIC_PARSE_FAIL;
1467 0 : }
1468 :
1469 : /* According to RFC 9000 Section 14.1, INITIAL packets less than a
1470 : certain length must be discarded, and the connection may be closed.
1471 : (Mitigates UDP amplification) */
1472 0 : if( pkt->datagram_sz < FD_QUIC_INITIAL_PAYLOAD_SZ_MIN ) {
1473 : /* can't trust the included values, so can't reply */
1474 0 : return FD_QUIC_PARSE_FAIL;
1475 0 : }
1476 :
1477 : /* Early check: Is conn free? */
1478 0 : if( FD_UNLIKELY( state->free_conn_list==UINT_MAX ) ) {
1479 0 : FD_DEBUG( FD_LOG_DEBUG(( "ignoring conn request: no free conn slots" )) );
1480 0 : metrics->conn_err_no_slots_cnt++;
1481 0 : return FD_QUIC_PARSE_FAIL; /* FIXME better error code? */
1482 0 : }
1483 :
1484 :
1485 : /* Primary objective is to send or verify retry.
1486 : We'll also select the scid we'll use from now on.
1487 :
1488 : Rules for selecting the SCID:
1489 : - No retry token, accepted: generate new random ID
1490 : - No retry token, retry request: generate new random ID
1491 : - Retry token, accepted: reuse SCID from retry token */
1492 0 : if( !quic->config.retry ) {
1493 0 : scid = fd_rng_ulong( state->_rng );
1494 0 : } else { /* retry configured */
1495 :
1496 : /* Need to send retry? Do so before more work */
1497 0 : if( initial->token_len != sizeof(fd_quic_retry_token_t) ) {
1498 :
1499 0 : ulong new_conn_id_u64 = fd_rng_ulong( state->_rng );
1500 0 : if( FD_UNLIKELY( fd_quic_send_retry(
1501 0 : quic, pkt,
1502 0 : dcid, peer_scid, new_conn_id_u64 ) ) ) {
1503 0 : return FD_QUIC_FAILED;
1504 0 : }
1505 0 : return (initial->pkt_num_pnoff + initial->len);
1506 0 : } else {
1507 : /* This Initial packet is in response to our Retry.
1508 : Validate the relevant fields of this post-retry INITIAL packet,
1509 : i.e. retry src conn id, ip, port
1510 : Also populate odcid and scid from the retry data */
1511 0 : int retry_ok = fd_quic_retry_server_verify( pkt, initial, &odcid, &scid, state->retry_secret, state->retry_iv, state->now, quic->config.retry_ttl );
1512 0 : if( FD_UNLIKELY( retry_ok!=FD_QUIC_SUCCESS ) ) {
1513 0 : metrics->conn_err_retry_fail_cnt++;
1514 : /* No need to set conn error, no conn object exists */
1515 0 : return FD_QUIC_PARSE_FAIL;
1516 0 : };
1517 0 : }
1518 0 : }
1519 0 : }
1520 :
1521 : /* Determine decryption keys, related data */
1522 :
1523 : /* Placeholder for generated crypto material before allocating conn */
1524 0 : fd_quic_crypto_keys_t _rx_keys[1];
1525 0 : fd_quic_crypto_secrets_t _secrets[1];
1526 :
1527 : /* Conditional inputs to decryption stage */
1528 0 : fd_quic_crypto_keys_t * rx_keys = NULL;
1529 0 : fd_quic_crypto_secrets_t * secrets = NULL;
1530 0 : ulong exp_pkt_num;
1531 :
1532 0 : if( !conn ) {
1533 : /* no conn, generate secret and rx keys */
1534 0 : rx_keys = _rx_keys;
1535 0 : secrets = _secrets;
1536 0 : exp_pkt_num = 0;
1537 :
1538 0 : fd_quic_gen_initial_secrets(
1539 0 : secrets,
1540 0 : dcid->conn_id, dcid->sz,
1541 0 : /* is_server */ 1 );
1542 0 : fd_quic_gen_keys(
1543 0 : rx_keys,
1544 0 : secrets->secret[ fd_quic_enc_level_initial_id ][ 0 ] );
1545 0 : } else {
1546 : /* conn, use existing keys/secrets */
1547 0 : rx_keys = &conn->keys[ fd_quic_enc_level_initial_id ][0];
1548 0 : secrets = &conn->secrets;
1549 0 : exp_pkt_num = conn->exp_pkt_number[0];
1550 0 : }
1551 :
1552 : /* Decrypt incoming packet */
1553 :
1554 : /* header protection needs the offset to the packet number */
1555 :
1556 0 : # if !FD_QUIC_DISABLE_CRYPTO
1557 : /* this decrypts the header */
1558 0 : if( FD_UNLIKELY(
1559 0 : fd_quic_crypto_decrypt_hdr( cur_ptr, cur_sz,
1560 0 : pn_offset,
1561 0 : rx_keys ) != FD_QUIC_SUCCESS ) ) {
1562 : /* As this is an INITIAL packet, change the status to DEAD, and allow
1563 : it to be reaped */
1564 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_crypto_decrypt_hdr failed" )) );
1565 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_initial_id ]++;
1566 0 : return FD_QUIC_PARSE_FAIL;
1567 0 : }
1568 0 : # endif /* !FD_QUIC_DISABLE_CRYPTO */
1569 :
1570 0 : ulong pkt_number_sz = fd_quic_h0_pkt_num_len( cur_ptr[0] ) + 1u;
1571 0 : ulong pktnum_comp = fd_quic_pktnum_decode( cur_ptr+pn_offset, pkt_number_sz );
1572 :
1573 : /* reconstruct packet number */
1574 0 : ulong pkt_number = fd_quic_reconstruct_pkt_num( pktnum_comp, pkt_number_sz, exp_pkt_num );
1575 :
1576 0 : # if !FD_QUIC_DISABLE_CRYPTO
1577 : /* NOTE from rfc9002 s3
1578 : It is permitted for some packet numbers to never be used, leaving intentional gaps. */
1579 : /* this decrypts the header and payload */
1580 0 : if( FD_UNLIKELY(
1581 0 : fd_quic_crypto_decrypt( cur_ptr, tot_sz,
1582 0 : pn_offset,
1583 0 : pkt_number,
1584 0 : rx_keys ) != FD_QUIC_SUCCESS ) ) {
1585 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_crypto_decrypt failed" )) );
1586 0 : FD_DTRACE_PROBE_2( quic_err_decrypt_initial_pkt, pkt->ip4, pkt->pkt_number );
1587 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_initial_id ]++;
1588 0 : return FD_QUIC_PARSE_FAIL;
1589 0 : }
1590 0 : # endif /* FD_QUIC_DISABLE_CRYPTO */
1591 :
1592 : /* set packet number on the context */
1593 0 : pkt->pkt_number = pkt_number;
1594 :
1595 0 : if( FD_UNLIKELY( body_sz < pkt_number_sz + FD_QUIC_CRYPTO_TAG_SZ ) ) {
1596 0 : return FD_QUIC_PARSE_FAIL;
1597 0 : }
1598 :
1599 : /* If no conn, create one. Due to previous checks, role must be server
1600 : and this must be response to Retry (if needed). */
1601 0 : if( FD_UNLIKELY( !conn ) ) {
1602 :
1603 : /* Save peer's conn ID, which we will use to address peer with. */
1604 0 : fd_quic_conn_id_t peer_conn_id = {0};
1605 0 : fd_memcpy( peer_conn_id.conn_id, initial->src_conn_id, FD_QUIC_MAX_CONN_ID_SZ );
1606 0 : peer_conn_id.sz = initial->src_conn_id_len;
1607 :
1608 : /* Prepare QUIC-TLS transport params object (sent as a TLS extension).
1609 : Take template from state and mutate certain params in-place.
1610 :
1611 : See RFC 9000 Section 18 */
1612 :
1613 : /* TODO Each transport param is a TLV tuple. This allows serializing
1614 : most transport params ahead of time. Only the conn-specific
1615 : differences will have to be appended here. */
1616 :
1617 0 : fd_quic_transport_params_t tp[1] = { state->transport_params };
1618 :
1619 0 : if( !quic->config.retry ) {
1620 : /* assume no retry */
1621 0 : tp->retry_source_connection_id_present = 0;
1622 :
1623 : /* Send orig conn ID back to client (server only) */
1624 :
1625 0 : tp->original_destination_connection_id_present = 1;
1626 0 : tp->original_destination_connection_id_len = dcid->sz;
1627 0 : fd_memcpy( tp->original_destination_connection_id,
1628 0 : dcid->conn_id,
1629 0 : dcid->sz );
1630 0 : } else { /* retry configured */
1631 :
1632 : /* From rfc 9000:
1633 :
1634 : Figure 8 shows a similar handshake that includes a Retry packet.
1635 :
1636 : Client Server
1637 : Initial: DCID=S1, SCID=C1 ->
1638 : <- Retry: DCID=C1, SCID=S2
1639 : Initial: DCID=S2, SCID=C1 ->
1640 : <- Initial: DCID=C1, SCID=S3
1641 : ...
1642 : 1-RTT: DCID=S3 ->
1643 : <- 1-RTT: DCID=C1
1644 :
1645 : Figure 8: Use of Connection IDs in a Handshake with Retry
1646 : In both cases (Figures 7 and 8), the client sets the value of the
1647 : initial_source_connection_id transport parameter to C1.
1648 :
1649 : When the handshake does not include a Retry (Figure 7), the server
1650 : sets original_destination_connection_id to S1 (note that this value
1651 : is chosen by the client) and initial_source_connection_id to S3. In
1652 : this case, the server does not include a retry_source_connection_id
1653 : transport parameter.
1654 :
1655 : When the handshake includes a Retry (Figure 8), the server sets
1656 : original_destination_connection_id to S1, retry_source_connection_id
1657 : to S2, and initial_source_connection_id to S3. */
1658 0 : tp->original_destination_connection_id_present = 1;
1659 0 : tp->original_destination_connection_id_len = odcid.sz;
1660 0 : memcpy( tp->original_destination_connection_id,
1661 0 : odcid.conn_id,
1662 0 : odcid.sz );
1663 :
1664 : /* Client echoes back the SCID we sent via Retry. Safe to trust
1665 : because we signed the Retry Token. (Length and content validated
1666 : in fd_quic_retry_server_verify) */
1667 0 : tp->retry_source_connection_id_present = 1;
1668 0 : tp->retry_source_connection_id_len = FD_QUIC_CONN_ID_SZ;
1669 0 : FD_STORE( ulong, tp->retry_source_connection_id, scid );
1670 :
1671 0 : metrics->conn_retry_cnt++;
1672 0 : }
1673 :
1674 : /* Repeat the conn ID we picked in transport params (this is done
1675 : to authenticate conn IDs via TLS by including them in TLS-
1676 : protected data).
1677 :
1678 : Per spec, this field should be the source conn ID field we've set
1679 : on the first Initial packet we've sent. At this point, we might
1680 : not have sent an Initial packet yet -- so this field should hold
1681 : a value we are about to pick.
1682 :
1683 : fd_quic_conn_create will set conn->initial_source_conn_id to
1684 : the random new_conn_id we've created earlier. */
1685 :
1686 0 : tp->initial_source_connection_id_present = 1;
1687 0 : tp->initial_source_connection_id_len = FD_QUIC_CONN_ID_SZ;
1688 0 : FD_STORE( ulong, tp->initial_source_connection_id, scid );
1689 :
1690 : /* tls hs available? After decrypting because might evict another hs */
1691 0 : if( FD_UNLIKELY( !fd_quic_tls_hs_pool_free( state->hs_pool ) ) ) {
1692 : /* try evicting, 0 if oldest is too young so fail */
1693 0 : if( !fd_quic_tls_hs_cache_evict( quic, state )) {
1694 0 : return FD_QUIC_PARSE_FAIL;
1695 0 : }
1696 0 : }
1697 :
1698 : /* Allocate new conn */
1699 0 : conn = fd_quic_conn_create( quic,
1700 0 : scid,
1701 0 : &peer_conn_id,
1702 0 : pkt->ip4->saddr,
1703 0 : pkt->udp->net_sport,
1704 0 : pkt->ip4->daddr,
1705 0 : pkt->udp->net_dport,
1706 0 : 1 /* server */ );
1707 :
1708 0 : if( FD_UNLIKELY( !conn ) ) { /* no free connections */
1709 : /* TODO send failure back to origin? */
1710 : /* FIXME unreachable? conn_cnt already checked above */
1711 0 : FD_DEBUG( FD_LOG_WARNING( ( "failed to allocate QUIC conn" ) ) );
1712 0 : return FD_QUIC_PARSE_FAIL;
1713 0 : }
1714 0 : FD_DEBUG( FD_LOG_DEBUG(( "new connection allocated" )) );
1715 :
1716 : /* set the value for the caller */
1717 0 : *p_conn = conn;
1718 :
1719 : /* Create a TLS handshake */
1720 0 : fd_quic_tls_hs_t * tls_hs = fd_quic_tls_hs_new(
1721 0 : fd_quic_tls_hs_pool_ele_acquire( state->hs_pool ),
1722 0 : state->tls,
1723 0 : (void*)conn,
1724 0 : 1 /*is_server*/,
1725 0 : tp,
1726 0 : state->now );
1727 0 : fd_quic_tls_hs_cache_ele_push_tail( &state->hs_cache, tls_hs, state->hs_pool );
1728 :
1729 0 : conn->tls_hs = tls_hs;
1730 0 : quic->metrics.hs_created_cnt++;
1731 :
1732 : /* copy secrets and rx keys */
1733 0 : conn->secrets = *secrets;
1734 0 : conn->keys[ fd_quic_enc_level_initial_id ][0] = *rx_keys;
1735 :
1736 : /* generate tx keys */
1737 0 : fd_quic_gen_keys(
1738 0 : &conn->keys[ fd_quic_enc_level_initial_id ][ 1 ],
1739 0 : secrets->secret[ fd_quic_enc_level_initial_id ][ 1 ] );
1740 0 : }
1741 :
1742 0 : if( FD_UNLIKELY( !conn->host.ip_addr ) ) {
1743 : /* Lock src IP address in place (previously chosen by layer-4 based
1744 : on the route table) */
1745 0 : conn->host.ip_addr = pkt->ip4->daddr;
1746 0 : }
1747 :
1748 : /* check if reply conn id needs to change */
1749 0 : if( FD_UNLIKELY( !( conn->server | conn->established ) ) ) {
1750 : /* switch to the source connection id for future replies */
1751 :
1752 : /* replace peer 0 connection id */
1753 0 : conn->peer_cids[0].sz = initial->src_conn_id_len;
1754 0 : fd_memcpy( conn->peer_cids[0].conn_id, initial->src_conn_id, FD_QUIC_MAX_CONN_ID_SZ );
1755 :
1756 : /* don't repeat this procedure */
1757 0 : conn->established = 1;
1758 0 : }
1759 :
1760 : /* handle frames */
1761 0 : ulong payload_off = pn_offset + pkt_number_sz;
1762 0 : uchar const * frame_ptr = cur_ptr + payload_off;
1763 0 : ulong frame_sz = body_sz - pkt_number_sz - FD_QUIC_CRYPTO_TAG_SZ; /* total size of all frames in packet */
1764 0 : while( frame_sz != 0UL ) {
1765 0 : rc = fd_quic_handle_v1_frame( quic,
1766 0 : conn,
1767 0 : pkt,
1768 0 : FD_QUIC_PKT_TYPE_INITIAL,
1769 0 : frame_ptr,
1770 0 : frame_sz );
1771 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
1772 0 : FD_DEBUG( FD_LOG_DEBUG(( "Failed to handle frame (Initial, frame=0x%02x)", frame_ptr[0] )) );
1773 0 : quic->metrics.frame_rx_err_cnt++;
1774 0 : return FD_QUIC_PARSE_FAIL;
1775 0 : }
1776 :
1777 0 : if( FD_UNLIKELY( rc==0UL || rc>frame_sz ) ) {
1778 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
1779 0 : return FD_QUIC_PARSE_FAIL;
1780 0 : }
1781 :
1782 : /* next frame, and remaining size */
1783 0 : frame_ptr += rc;
1784 0 : frame_sz -= rc;
1785 0 : }
1786 :
1787 : /* update last activity */
1788 0 : conn->last_activity = state->now;
1789 0 : conn->flags &= ~( FD_QUIC_CONN_FLAGS_PING_SENT | FD_QUIC_CONN_FLAGS_PING );
1790 :
1791 : /* update expected packet number */
1792 0 : conn->exp_pkt_number[0] = fd_ulong_max( conn->exp_pkt_number[0], pkt_number+1UL );
1793 :
1794 : /* insert into service queue */
1795 0 : fd_quic_svc_prep_schedule( conn, state->now );
1796 :
1797 : /* return number of bytes consumed */
1798 0 : return tot_sz;
1799 0 : }
1800 :
1801 : ulong
1802 : fd_quic_handle_v1_handshake(
1803 : fd_quic_t * quic,
1804 : fd_quic_conn_t * conn,
1805 : fd_quic_pkt_t * pkt,
1806 : uchar * cur_ptr,
1807 : ulong cur_sz
1808 0 : ) {
1809 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
1810 0 : if( FD_UNLIKELY( !conn ) ) {
1811 0 : quic->metrics.pkt_no_conn_cnt[ fd_quic_enc_level_handshake_id ]++;
1812 0 : FD_DTRACE_PROBE_2( fd_quic_handle_v1_handshake_no_conn , state->now, pkt->pkt_number );
1813 0 : return FD_QUIC_PARSE_FAIL;
1814 0 : }
1815 :
1816 0 : if( FD_UNLIKELY( !fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_handshake_id ) ) ) {
1817 0 : quic->metrics.pkt_no_key_cnt[ fd_quic_enc_level_handshake_id ]++;
1818 0 : return FD_QUIC_PARSE_FAIL;
1819 0 : }
1820 :
1821 : /* do parse here */
1822 0 : fd_quic_handshake_t handshake[1];
1823 0 : ulong rc = fd_quic_decode_handshake( handshake, cur_ptr, cur_sz );
1824 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
1825 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_decode_handshake failed" )) );
1826 0 : return FD_QUIC_PARSE_FAIL;
1827 0 : }
1828 :
1829 : /* check bounds on handshake */
1830 :
1831 : /* len indicated the number of bytes after the packet number offset
1832 : so verify this value is within the packet */
1833 0 : ulong len = (ulong)( handshake->pkt_num_pnoff + handshake->len );
1834 0 : if( FD_UNLIKELY( len > cur_sz ) ) {
1835 0 : FD_DEBUG( FD_LOG_DEBUG(( "Handshake packet bounds check failed" )); )
1836 0 : return FD_QUIC_PARSE_FAIL;
1837 0 : }
1838 :
1839 : /* connection ids should already be in the relevant structures */
1840 :
1841 : /* TODO prepare most of the transport parameters, and only append the
1842 : necessary differences */
1843 :
1844 : /* fetch TLS handshake */
1845 0 : fd_quic_tls_hs_t * tls_hs = conn->tls_hs;
1846 0 : if( FD_UNLIKELY( !tls_hs ) ) {
1847 0 : FD_DEBUG( FD_LOG_DEBUG(( "no tls handshake" )) );
1848 0 : return FD_QUIC_PARSE_FAIL;
1849 0 : }
1850 :
1851 : /* decryption */
1852 :
1853 : /* header protection needs the offset to the packet number */
1854 0 : ulong pn_offset = handshake->pkt_num_pnoff;
1855 :
1856 0 : ulong body_sz = handshake->len; /* not a protected field */
1857 : /* length of payload + num packet bytes */
1858 :
1859 0 : # if !FD_QUIC_DISABLE_CRYPTO
1860 : /* this decrypts the header */
1861 0 : if( FD_UNLIKELY(
1862 0 : fd_quic_crypto_decrypt_hdr( cur_ptr, cur_sz,
1863 0 : pn_offset,
1864 0 : &conn->keys[2][0] ) != FD_QUIC_SUCCESS ) ) {
1865 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_crypto_decrypt_hdr failed" )) );
1866 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_handshake_id ]++;
1867 0 : return FD_QUIC_PARSE_FAIL;
1868 0 : }
1869 0 : # endif /* !FD_QUIC_DISABLE_CRYPTO */
1870 :
1871 : /* number of bytes in the packet header */
1872 0 : ulong pkt_number_sz = fd_quic_h0_pkt_num_len( cur_ptr[0] ) + 1u;
1873 0 : ulong tot_sz = pn_offset + body_sz; /* total including header and payload */
1874 :
1875 : /* now we have decrypted packet number */
1876 0 : ulong pktnum_comp = fd_quic_pktnum_decode( cur_ptr+pn_offset, pkt_number_sz );
1877 :
1878 : /* reconstruct packet number */
1879 0 : ulong pkt_number = fd_quic_reconstruct_pkt_num( pktnum_comp, pkt_number_sz, conn->exp_pkt_number[1] );
1880 :
1881 : /* NOTE from rfc9002 s3
1882 : It is permitted for some packet numbers to never be used, leaving intentional gaps. */
1883 :
1884 0 : # if !FD_QUIC_DISABLE_CRYPTO
1885 : /* this decrypts the header and payload */
1886 0 : if( FD_UNLIKELY(
1887 0 : fd_quic_crypto_decrypt( cur_ptr, tot_sz,
1888 0 : pn_offset,
1889 0 : pkt_number,
1890 0 : &conn->keys[2][0] ) != FD_QUIC_SUCCESS ) ) {
1891 : /* remove connection from map, and insert into free list */
1892 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_crypto_decrypt failed" )) );
1893 0 : FD_DTRACE_PROBE_3( quic_err_decrypt_handshake_pkt, pkt->ip4, conn->our_conn_id, pkt->pkt_number );
1894 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_handshake_id ]++;
1895 0 : return FD_QUIC_PARSE_FAIL;
1896 0 : }
1897 0 : # endif /* FD_QUIC_DISABLE_CRYPTO */
1898 :
1899 : /* set packet number on the context */
1900 0 : pkt->pkt_number = pkt_number;
1901 :
1902 : /* check body size large enough for required elements */
1903 0 : if( FD_UNLIKELY( body_sz < pkt_number_sz + FD_QUIC_CRYPTO_TAG_SZ ) ) {
1904 0 : return FD_QUIC_PARSE_FAIL;
1905 0 : }
1906 :
1907 : /* handle frames */
1908 0 : ulong payload_off = pn_offset + pkt_number_sz;
1909 0 : uchar const * frame_ptr = cur_ptr + payload_off;
1910 0 : ulong frame_sz = body_sz - pkt_number_sz - FD_QUIC_CRYPTO_TAG_SZ; /* total size of all frames in packet */
1911 0 : while( frame_sz != 0UL ) {
1912 0 : rc = fd_quic_handle_v1_frame( quic,
1913 0 : conn,
1914 0 : pkt,
1915 0 : FD_QUIC_PKT_TYPE_HANDSHAKE,
1916 0 : frame_ptr,
1917 0 : frame_sz );
1918 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
1919 0 : FD_DEBUG( FD_LOG_DEBUG(( "Failed to handle frame (Handshake, frame=0x%02x)", frame_ptr[0] )) );
1920 0 : quic->metrics.frame_rx_err_cnt++;
1921 0 : return FD_QUIC_PARSE_FAIL;
1922 0 : }
1923 :
1924 0 : if( FD_UNLIKELY( rc == 0UL || rc > frame_sz ) ) {
1925 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
1926 0 : return FD_QUIC_PARSE_FAIL;
1927 0 : }
1928 :
1929 : /* next frame and remaining size */
1930 0 : frame_ptr += rc;
1931 0 : frame_sz -= rc;
1932 0 : }
1933 :
1934 : /* RFC 9000 Section 17.2.2.1. Abandoning Initial Packets
1935 : > A server stops sending and processing Initial packets when it
1936 : > receives its first Handshake packet.
1937 :
1938 : RFC 9001 Section 4.9.1 Discarding Initial Keys
1939 : > a server MUST discard Initial keys when it first successfully processes a Handshake packet */
1940 0 : if( FD_LIKELY( quic->config.role==FD_QUIC_ROLE_SERVER ) ) fd_quic_abandon_enc_level( conn, fd_quic_enc_level_initial_id );
1941 :
1942 : /* update last activity */
1943 0 : conn->last_activity = fd_quic_get_state( quic )->now;
1944 0 : conn->flags &= ~( FD_QUIC_CONN_FLAGS_PING_SENT | FD_QUIC_CONN_FLAGS_PING );
1945 :
1946 : /* update expected packet number */
1947 0 : conn->exp_pkt_number[1] = fd_ulong_max( conn->exp_pkt_number[1], pkt_number+1UL );
1948 :
1949 : /* return number of bytes consumed */
1950 0 : return tot_sz;
1951 0 : }
1952 :
1953 : ulong
1954 : fd_quic_handle_v1_retry(
1955 : fd_quic_t * quic,
1956 : fd_quic_conn_t * conn,
1957 : fd_quic_pkt_t const * pkt,
1958 : uchar const * cur_ptr,
1959 : ulong cur_sz
1960 0 : ) {
1961 0 : (void)pkt;
1962 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
1963 :
1964 0 : if( FD_UNLIKELY( quic->config.role == FD_QUIC_ROLE_SERVER ) ) {
1965 0 : if( FD_UNLIKELY( conn ) ) { /* likely a misbehaving client w/o a conn */
1966 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
1967 0 : }
1968 0 : return FD_QUIC_PARSE_FAIL;
1969 0 : }
1970 :
1971 0 : if( FD_UNLIKELY( !conn ) ) {
1972 0 : FD_DTRACE_PROBE_2( fd_quic_handle_v1_retry_no_conn , state->now, pkt->pkt_number );
1973 0 : quic->metrics.pkt_no_conn_cnt[1]++;
1974 0 : return FD_QUIC_PARSE_FAIL;
1975 0 : }
1976 :
1977 0 : fd_quic_conn_id_t const * orig_dst_conn_id = &conn->peer_cids[0];
1978 0 : uchar const * retry_token = NULL;
1979 0 : ulong retry_token_sz = 0UL;
1980 :
1981 0 : int rc = fd_quic_retry_client_verify(
1982 0 : cur_ptr, cur_sz,
1983 0 : orig_dst_conn_id,
1984 0 : &conn->retry_src_conn_id,
1985 0 : &retry_token, &retry_token_sz
1986 0 : );
1987 0 : if( FD_UNLIKELY( rc!=FD_QUIC_SUCCESS ) ) {
1988 0 : quic->metrics.conn_err_retry_fail_cnt++;
1989 0 : return FD_QUIC_PARSE_FAIL;
1990 0 : }
1991 :
1992 : /* Update the peer using the retry src conn id */
1993 0 : conn->peer_cids[0] = conn->retry_src_conn_id;
1994 :
1995 : /* Re-send the ClientHello */
1996 0 : conn->hs_sent_bytes[fd_quic_enc_level_initial_id] = 0;
1997 :
1998 : /* Need to regenerate keys using the retry source connection id */
1999 0 : fd_quic_gen_initial_secret_and_keys( conn, &conn->retry_src_conn_id, /* is_server */ 0 );
2000 :
2001 : /* The token length is the remaining bytes in the retry packet after subtracting known fields. */
2002 0 : conn->token_len = retry_token_sz;
2003 0 : fd_memcpy( &conn->token, retry_token, conn->token_len );
2004 :
2005 : /* have to rewind the handshake data */
2006 0 : uint enc_level = fd_quic_enc_level_initial_id;
2007 0 : conn->hs_sent_bytes[enc_level] = 0;
2008 :
2009 : /* send the INITIAL */
2010 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
2011 :
2012 0 : fd_quic_svc_prep_schedule_now( conn );
2013 :
2014 0 : return cur_sz;
2015 0 : }
2016 :
2017 : ulong
2018 0 : 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 ) {
2019 0 : (void)pkt;
2020 0 : (void)quic;
2021 0 : (void)cur_ptr;
2022 0 : (void)cur_sz;
2023 : /* since we do not support zero-rtt, simply fail the packet */
2024 0 : if( conn ) {
2025 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_INTERNAL_ERROR, __LINE__ );
2026 0 : }
2027 0 : return FD_QUIC_PARSE_FAIL;
2028 0 : }
2029 :
2030 : int
2031 : fd_quic_lazy_ack_pkt( fd_quic_t * quic,
2032 : fd_quic_conn_t * conn,
2033 0 : fd_quic_pkt_t const * pkt ) {
2034 0 : if( pkt->ack_flag & ACK_FLAG_CANCEL ) {
2035 0 : return FD_QUIC_ACK_TX_CANCEL;
2036 0 : }
2037 :
2038 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
2039 0 : fd_quic_ack_gen_t * ack_gen = conn->ack_gen;
2040 0 : int res = fd_quic_ack_pkt( conn->ack_gen, pkt->pkt_number, pkt->enc_level, state->now );
2041 0 : conn->ack_gen->is_elicited |= fd_uchar_if( pkt->ack_flag & ACK_FLAG_RQD, 1, 0 );
2042 :
2043 : /* Trigger immediate ACK send? */
2044 0 : int ack_sz_threshold_hit = conn->unacked_sz > quic->config.ack_threshold;
2045 0 : int force_instant_ack =
2046 0 : ( !!(pkt->ack_flag & ACK_FLAG_RQD) ) &
2047 0 : ( ( pkt->enc_level == fd_quic_enc_level_initial_id ) |
2048 0 : ( pkt->enc_level == fd_quic_enc_level_handshake_id ) );
2049 0 : if( ack_sz_threshold_hit | force_instant_ack ) {
2050 0 : conn->unacked_sz = 0UL;
2051 0 : fd_quic_svc_prep_schedule( conn, state->now );
2052 0 : } else if( ack_gen->is_elicited ) {
2053 0 : fd_quic_svc_prep_schedule( conn, state->now + quic->config.ack_delay );
2054 0 : }
2055 :
2056 0 : return res;
2057 0 : }
2058 :
2059 : /* This thunk works around a compiler bug (bogus stringop-overflow warning) in GCC 11 */
2060 : __attribute__((noinline)) static void
2061 0 : fd_quic_key_update_derive1( fd_quic_conn_t * conn ) {
2062 0 : fd_quic_key_update_derive( &conn->secrets, conn->new_keys );
2063 0 : }
2064 :
2065 : static void
2066 0 : fd_quic_key_update_complete( fd_quic_conn_t * conn ) {
2067 : /* Key updates are only possible for 1-RTT packets, which are appdata */
2068 0 : ulong const enc_level = fd_quic_enc_level_appdata_id;
2069 :
2070 : /* Update payload keys */
2071 0 : memcpy( conn->keys[enc_level][0].pkt_key, conn->new_keys[0].pkt_key, FD_AES_128_KEY_SZ );
2072 0 : memcpy( conn->keys[enc_level][0].iv, conn->new_keys[0].iv, FD_AES_GCM_IV_SZ );
2073 0 : memcpy( conn->keys[enc_level][1].pkt_key, conn->new_keys[1].pkt_key, FD_AES_128_KEY_SZ );
2074 0 : memcpy( conn->keys[enc_level][1].iv, conn->new_keys[1].iv, FD_AES_GCM_IV_SZ );
2075 :
2076 : /* Update IVs */
2077 0 : memcpy( conn->secrets.secret[enc_level][0], conn->secrets.new_secret[0], FD_QUIC_SECRET_SZ );
2078 0 : memcpy( conn->secrets.secret[enc_level][1], conn->secrets.new_secret[1], FD_QUIC_SECRET_SZ );
2079 :
2080 : /* Packet header encryption keys are not updated */
2081 :
2082 : /* Wind up for next key phase update */
2083 0 : conn->key_phase = !conn->key_phase;
2084 0 : conn->key_update = 0;
2085 0 : fd_quic_key_update_derive1( conn );
2086 :
2087 0 : FD_DEBUG( FD_LOG_DEBUG(( "key update completed" )); )
2088 0 : }
2089 :
2090 : ulong
2091 : fd_quic_handle_v1_one_rtt( fd_quic_t * quic,
2092 : fd_quic_conn_t * conn,
2093 : fd_quic_pkt_t * pkt,
2094 : uchar * const cur_ptr,
2095 0 : ulong const tot_sz ) {
2096 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
2097 :
2098 0 : if( !conn ) {
2099 0 : quic->metrics.pkt_no_conn_cnt[ fd_quic_enc_level_appdata_id ]++;
2100 0 : FD_DTRACE_PROBE_2( fd_quic_handle_v1_one_rtt_no_conn , state->now, pkt->pkt_number );
2101 0 : FD_DEBUG( FD_LOG_DEBUG(( "one_rtt failed: no connection found" )) );
2102 0 : return FD_QUIC_PARSE_FAIL;
2103 0 : }
2104 :
2105 0 : if( FD_UNLIKELY( !fd_uint_extract_bit( conn->keys_avail, fd_quic_enc_level_appdata_id ) ) ) {
2106 0 : quic->metrics.pkt_no_key_cnt[ fd_quic_enc_level_appdata_id ]++;
2107 0 : return FD_QUIC_PARSE_FAIL;
2108 0 : }
2109 :
2110 0 : if( FD_UNLIKELY( tot_sz < (1+FD_QUIC_CONN_ID_SZ+1) ) ) {
2111 : /* One-RTT header: 1 byte
2112 : DCID: FD_QUIC_CONN_ID_SZ
2113 : Pkt number: 1-4 bytes */
2114 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_appdata_id ]++;
2115 0 : return FD_QUIC_PARSE_FAIL;
2116 0 : }
2117 0 : ulong pn_offset = 1UL + FD_QUIC_CONN_ID_SZ;
2118 :
2119 0 : pkt->enc_level = fd_quic_enc_level_appdata_id;
2120 :
2121 0 : # if !FD_QUIC_DISABLE_CRYPTO
2122 0 : if( FD_UNLIKELY(
2123 0 : fd_quic_crypto_decrypt_hdr( cur_ptr, tot_sz,
2124 0 : pn_offset,
2125 0 : &conn->keys[3][0] ) != FD_QUIC_SUCCESS ) ) {
2126 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_crypto_decrypt_hdr failed" )) );
2127 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_appdata_id ]++;
2128 0 : return FD_QUIC_PARSE_FAIL;
2129 0 : }
2130 0 : # endif /* !FD_QUIC_DISABLE_CRYPTO */
2131 :
2132 0 : uint pkt_number_sz = fd_quic_h0_pkt_num_len( cur_ptr[0] ) + 1u;
2133 0 : uint key_phase = fd_quic_one_rtt_key_phase( cur_ptr[0] );
2134 :
2135 : /* reconstruct packet number */
2136 0 : ulong pktnum_comp = fd_quic_pktnum_decode( cur_ptr+pn_offset, pkt_number_sz );
2137 0 : ulong pkt_number = fd_quic_reconstruct_pkt_num( pktnum_comp, pkt_number_sz, conn->exp_pkt_number[2] );
2138 :
2139 : /* NOTE from rfc9002 s3
2140 : It is permitted for some packet numbers to never be used, leaving intentional gaps. */
2141 :
2142 : /* is current packet in the current key phase? */
2143 0 : int current_key_phase = conn->key_phase == key_phase;
2144 :
2145 0 : # if !FD_QUIC_DISABLE_CRYPTO
2146 : /* If the key phase bit flips, decrypt with the new pair of keys
2147 : instead. Note that the key phase bit is untrusted at this point. */
2148 0 : fd_quic_crypto_keys_t * keys = current_key_phase ? &conn->keys[3][0] : &conn->new_keys[0];
2149 :
2150 : /* this decrypts the header and payload */
2151 0 : if( FD_UNLIKELY(
2152 0 : fd_quic_crypto_decrypt( cur_ptr, tot_sz,
2153 0 : pn_offset,
2154 0 : pkt_number,
2155 0 : keys ) != FD_QUIC_SUCCESS ) ) {
2156 : /* remove connection from map, and insert into free list */
2157 0 : FD_DTRACE_PROBE_3( quic_err_decrypt_1rtt_pkt, pkt->ip4, conn->our_conn_id, pkt->pkt_number );
2158 0 : quic->metrics.pkt_decrypt_fail_cnt[ fd_quic_enc_level_appdata_id ]++;
2159 0 : return FD_QUIC_PARSE_FAIL;
2160 0 : }
2161 0 : # endif /* !FD_QUIC_DISABLE_CRYPTO */
2162 :
2163 : /* set packet number on the context */
2164 0 : pkt->pkt_number = pkt_number;
2165 :
2166 0 : if( !current_key_phase ) {
2167 : /* Decryption succeeded. Commit the key phase update and throw
2168 : away the old keys. (May cause a few decryption failures if old
2169 : packets get reordered past the current incoming packet) */
2170 0 : fd_quic_key_update_complete( conn );
2171 0 : }
2172 :
2173 : /* handle frames */
2174 0 : ulong payload_off = pn_offset + pkt_number_sz;
2175 0 : uchar const * frame_ptr = cur_ptr + payload_off;
2176 0 : ulong payload_sz = tot_sz - pn_offset - pkt_number_sz; /* includes auth tag */
2177 0 : if( FD_UNLIKELY( payload_sz<FD_QUIC_CRYPTO_TAG_SZ ) ) return FD_QUIC_PARSE_FAIL;
2178 0 : ulong frame_sz = payload_sz - FD_QUIC_CRYPTO_TAG_SZ; /* total size of all frames in packet */
2179 0 : while( frame_sz != 0UL ) {
2180 0 : ulong rc = fd_quic_handle_v1_frame(
2181 0 : quic, conn, pkt, FD_QUIC_PKT_TYPE_ONE_RTT,
2182 0 : frame_ptr, frame_sz );
2183 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2184 0 : FD_DEBUG( FD_LOG_DEBUG(( "Failed to handle frame (1-RTT, frame=0x%02x)", frame_ptr[0] )) );
2185 0 : quic->metrics.frame_rx_err_cnt++;
2186 0 : return FD_QUIC_PARSE_FAIL;
2187 0 : }
2188 :
2189 0 : if( FD_UNLIKELY( rc == 0UL || rc > frame_sz ) ) {
2190 0 : FD_LOG_WARNING(( "fd_quic_handle_v1_frame returned invalid size" ));
2191 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
2192 0 : return FD_QUIC_PARSE_FAIL;
2193 0 : }
2194 :
2195 : /* next frame, and remaining size */
2196 0 : frame_ptr += rc;
2197 0 : frame_sz -= rc;
2198 0 : }
2199 :
2200 : /* update last activity */
2201 0 : conn->last_activity = state->now;
2202 :
2203 : /* update expected packet number */
2204 0 : conn->exp_pkt_number[2] = fd_ulong_max( conn->exp_pkt_number[2], pkt_number+1UL );
2205 :
2206 0 : return tot_sz;
2207 0 : }
2208 :
2209 :
2210 : /* process v1 quic packets
2211 : returns number of bytes consumed, or FD_QUIC_PARSE_FAIL upon error */
2212 : ulong
2213 : fd_quic_process_quic_packet_v1( fd_quic_t * quic,
2214 : fd_quic_pkt_t * pkt,
2215 : uchar * cur_ptr,
2216 0 : ulong cur_sz ) {
2217 :
2218 : /* bounds check packet size */
2219 0 : if( FD_UNLIKELY( cur_sz < FD_QUIC_SHORTEST_PKT ) ) {
2220 0 : quic->metrics.pkt_undersz_cnt++;
2221 0 : return FD_QUIC_PARSE_FAIL;
2222 0 : }
2223 0 : if( FD_UNLIKELY( cur_sz > 1500 ) ) {
2224 0 : quic->metrics.pkt_oversz_cnt++;
2225 0 : return FD_QUIC_PARSE_FAIL;
2226 0 : }
2227 :
2228 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
2229 0 : fd_quic_conn_t * conn = NULL;
2230 :
2231 :
2232 : /* keep end */
2233 0 : uchar * orig_ptr = cur_ptr;
2234 :
2235 : /* No need for cur_sz check, since we are safe from the above check.
2236 : Decrementing cur_sz is done in the long header branch, the short header
2237 : branch parses the first byte again using the parser generator.
2238 : */
2239 0 : uchar hdr_form = fd_quic_h0_hdr_form( *cur_ptr );
2240 0 : ulong rc;
2241 :
2242 0 : if( hdr_form ) { /* long header */
2243 0 : fd_quic_long_hdr_t long_hdr[1];
2244 0 : rc = fd_quic_decode_long_hdr( long_hdr, cur_ptr+1, cur_sz-1 );
2245 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2246 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_decode_long_hdr failed" )); )
2247 0 : quic->metrics.pkt_quic_hdr_err_cnt++;
2248 0 : return FD_QUIC_PARSE_FAIL;
2249 0 : }
2250 :
2251 0 : fd_quic_conn_id_t dcid = fd_quic_conn_id_new( long_hdr->dst_conn_id, long_hdr->dst_conn_id_len );
2252 0 : if( dcid.sz == FD_QUIC_CONN_ID_SZ ) {
2253 0 : conn = fd_quic_conn_query( state->conn_map, fd_ulong_load_8( dcid.conn_id ) );
2254 0 : }
2255 0 : fd_quic_conn_id_t scid = fd_quic_conn_id_new( long_hdr->src_conn_id, long_hdr->src_conn_id_len );
2256 :
2257 0 : uchar long_packet_type = fd_quic_h0_long_packet_type( *cur_ptr );
2258 :
2259 : /* encryption level matches that of TLS */
2260 0 : pkt->enc_level = long_packet_type; /* V2 uses an indirect mapping */
2261 :
2262 : /* initialize packet number to unused value */
2263 0 : pkt->pkt_number = FD_QUIC_PKT_NUM_UNUSED;
2264 :
2265 0 : switch( long_packet_type ) {
2266 0 : case FD_QUIC_PKT_TYPE_INITIAL:
2267 0 : rc = fd_quic_handle_v1_initial( quic, &conn, pkt, &dcid, &scid, cur_ptr, cur_sz );
2268 0 : if( FD_UNLIKELY( !conn ) ) {
2269 : /* FIXME not really a fail - Could be a retry */
2270 0 : return FD_QUIC_PARSE_FAIL;
2271 0 : }
2272 0 : break;
2273 0 : case FD_QUIC_PKT_TYPE_HANDSHAKE:
2274 0 : rc = fd_quic_handle_v1_handshake( quic, conn, pkt, cur_ptr, cur_sz );
2275 0 : break;
2276 0 : case FD_QUIC_PKT_TYPE_RETRY:
2277 0 : rc = fd_quic_handle_v1_retry( quic, conn, pkt, cur_ptr, cur_sz );
2278 0 : break;
2279 0 : case FD_QUIC_PKT_TYPE_ZERO_RTT:
2280 0 : rc = fd_quic_handle_v1_zero_rtt( quic, conn, pkt, cur_ptr, cur_sz );
2281 0 : break;
2282 0 : }
2283 :
2284 0 : fd_quic_svc_timers_schedule( state->svc_timers, conn, state->now );
2285 :
2286 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2287 0 : FD_DEBUG( FD_LOG_DEBUG(( "Rejected packet (type=%d)", long_packet_type )); )
2288 0 : return FD_QUIC_PARSE_FAIL;
2289 0 : }
2290 :
2291 0 : } else { /* short header */
2292 : /* encryption level of short header packets is fd_quic_enc_level_appdata_id */
2293 0 : pkt->enc_level = fd_quic_enc_level_appdata_id;
2294 :
2295 : /* initialize packet number to unused value */
2296 0 : pkt->pkt_number = FD_QUIC_PKT_NUM_UNUSED;
2297 :
2298 : /* find connection id */
2299 0 : ulong dst_conn_id = fd_ulong_load_8( cur_ptr+1 );
2300 0 : conn = fd_quic_conn_query( state->conn_map, dst_conn_id );
2301 0 : rc = fd_quic_handle_v1_one_rtt( quic, conn, pkt, cur_ptr, cur_sz );
2302 :
2303 0 : fd_quic_svc_timers_schedule( state->svc_timers, conn, state->now );
2304 :
2305 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2306 0 : return FD_QUIC_PARSE_FAIL;
2307 0 : }
2308 0 : }
2309 :
2310 0 : if( FD_UNLIKELY( rc == 0UL ) ) {
2311 : /* this is an error because it causes infinite looping */
2312 0 : return FD_QUIC_PARSE_FAIL;
2313 0 : }
2314 0 : cur_ptr += rc;
2315 :
2316 : /* if we get here we parsed all the frames, so ack the packet */
2317 0 : int ack_type = fd_quic_lazy_ack_pkt( quic, conn, pkt );
2318 0 : quic->metrics.ack_tx[ ack_type ]++;
2319 :
2320 : /* fd_quic_lazy_ack_pkt may have prepped schedule */
2321 0 : fd_quic_svc_timers_schedule( state->svc_timers, conn, state->now );
2322 :
2323 0 : if( pkt->rtt_ack_time ) {
2324 0 : fd_quic_sample_rtt( conn, (long)pkt->rtt_ack_time, (long)pkt->rtt_ack_delay );
2325 0 : }
2326 :
2327 : /* return bytes consumed */
2328 0 : return (ulong)( cur_ptr - orig_ptr );
2329 0 : }
2330 :
2331 :
2332 : /* version negotiation packet has version 0 */
2333 : static inline int
2334 0 : is_version_invalid( fd_quic_t * quic, uint version ) {
2335 0 : if( version == 0 ) {
2336 : /* TODO implement version negotiation */
2337 0 : quic->metrics.pkt_verneg_cnt++;
2338 0 : FD_DEBUG( FD_LOG_DEBUG(( "Got version negotiation packet" )) );
2339 0 : return 1;
2340 0 : }
2341 :
2342 : /* 0x?a?a?a?au is intended to force version negotiation
2343 : TODO implement */
2344 0 : if( ( version & 0x0a0a0a0au ) == 0x0a0a0a0au ) {
2345 : /* at present, ignore */
2346 0 : quic->metrics.pkt_verneg_cnt++;
2347 0 : FD_DEBUG( FD_LOG_DEBUG(( "Got version negotiation packet (forced)" )) );
2348 0 : return 1;
2349 0 : }
2350 :
2351 0 : if( version != 1 ) {
2352 : /* cannot interpret length, so discard entire packet */
2353 : /* TODO send version negotiation */
2354 0 : quic->metrics.pkt_verneg_cnt++;
2355 0 : FD_DEBUG( FD_LOG_DEBUG(( "Got unknown version QUIC packet" )) );
2356 0 : return 1;
2357 0 : }
2358 0 : return 0;
2359 0 : }
2360 :
2361 : static inline void
2362 : fd_quic_process_packet_impl( fd_quic_t * quic,
2363 : uchar * data,
2364 : ulong data_sz,
2365 0 : long now ) {
2366 0 : fd_quic_get_state( quic )->now = now;
2367 0 : quic->metrics.net_rx_byte_cnt += data_sz;
2368 0 : quic->metrics.net_rx_pkt_cnt++;
2369 :
2370 0 : ulong rc = 0;
2371 :
2372 : /* holds the remainder of the packet*/
2373 0 : uchar * cur_ptr = data;
2374 0 : ulong cur_sz = data_sz;
2375 :
2376 0 : if( FD_UNLIKELY( data_sz > 0xffffu ) ) {
2377 0 : FD_DTRACE_PROBE( quic_err_rx_oversz );
2378 0 : quic->metrics.pkt_oversz_cnt++;
2379 0 : return;
2380 0 : }
2381 :
2382 0 : fd_quic_pkt_t pkt = { .datagram_sz = (uint)data_sz };
2383 :
2384 0 : pkt.rcv_time = now;
2385 0 : pkt.rtt_pkt_number = 0;
2386 0 : pkt.rtt_ack_time = 0;
2387 :
2388 : /* parse ip, udp */
2389 :
2390 0 : rc = fd_quic_decode_ip4( pkt.ip4, cur_ptr, cur_sz );
2391 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2392 : /* TODO count failure */
2393 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2394 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2395 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_decode_ip4 failed" )) );
2396 0 : return;
2397 0 : }
2398 :
2399 : /* check version, tot_len, protocol, checksum? */
2400 0 : if( FD_UNLIKELY( pkt.ip4->protocol != FD_IP4_HDR_PROTOCOL_UDP ) ) {
2401 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2402 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2403 0 : FD_DEBUG( FD_LOG_DEBUG(( "Packet is not UDP" )) );
2404 0 : return;
2405 0 : }
2406 :
2407 : /* verify ip4 packet isn't truncated
2408 : * AF_XDP can silently do this */
2409 0 : if( FD_UNLIKELY( pkt.ip4->net_tot_len > cur_sz ) ) {
2410 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2411 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2412 0 : FD_DEBUG( FD_LOG_DEBUG(( "IPv4 header indicates truncation" )) );
2413 0 : return;
2414 0 : }
2415 :
2416 : /* update pointer + size */
2417 0 : cur_ptr += rc;
2418 0 : cur_sz -= rc;
2419 :
2420 0 : rc = fd_quic_decode_udp( pkt.udp, cur_ptr, cur_sz );
2421 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
2422 : /* TODO count failure */
2423 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2424 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2425 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_decode_udp failed" )) );
2426 0 : return;
2427 0 : }
2428 :
2429 : /* sanity check udp length */
2430 0 : if( FD_UNLIKELY( pkt.udp->net_len < sizeof(fd_udp_hdr_t) ||
2431 0 : pkt.udp->net_len > cur_sz ) ) {
2432 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2433 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2434 0 : FD_DEBUG( FD_LOG_DEBUG(( "UDP header indicates truncation" )) );
2435 0 : return;
2436 0 : }
2437 :
2438 : /* update pointer + size */
2439 0 : cur_ptr += rc;
2440 0 : cur_sz = pkt.udp->net_len - rc; /* replace with udp length */
2441 :
2442 : /* cur_ptr[0..cur_sz-1] should be payload */
2443 :
2444 : /* filter */
2445 : /* check dst eth address, ip address? probably not necessary */
2446 : /* usually look up port here, but let's jump straight into decoding as-if
2447 : quic */
2448 :
2449 : /* update counters */
2450 :
2451 : /* shortest valid quic payload? */
2452 0 : if( FD_UNLIKELY( cur_sz < FD_QUIC_SHORTEST_PKT ) ) {
2453 0 : FD_DTRACE_PROBE( quic_err_rx_net_hdr );
2454 0 : quic->metrics.pkt_net_hdr_err_cnt++;
2455 0 : FD_DEBUG( FD_LOG_DEBUG(( "Undersize QUIC packet" )) );
2456 0 : return;
2457 0 : }
2458 :
2459 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
2460 :
2461 : /* short packets don't have version */
2462 0 : int long_pkt = !!( (uint)cur_ptr[0] & 0x80u );
2463 :
2464 0 : if( long_pkt ) {
2465 : /* version at offset 1..4 */
2466 0 : uint version = fd_uint_bswap( FD_LOAD( uint, cur_ptr + 1 ) );
2467 : /* we only support version 1 */
2468 0 : if( FD_UNLIKELY( is_version_invalid( quic, version ) ) ) {
2469 0 : return;
2470 0 : }
2471 :
2472 : /* multiple QUIC packets in a UDP packet */
2473 : /* shortest valid quic payload? */
2474 0 : ulong pkt_idx;
2475 0 : for( pkt_idx=0UL; pkt_idx<FD_QUIC_PKT_COALESCE_LIMIT; pkt_idx++ ) {
2476 : /* Are we done? Omit short packet handling that follows */
2477 0 : if( FD_UNLIKELY( cur_sz < FD_QUIC_SHORTEST_PKT ) ) return;
2478 :
2479 : /* short packet requires different handling */
2480 0 : int short_pkt = !( (uint)cur_ptr[0] & 0x80u );
2481 :
2482 0 : if( FD_UNLIKELY( short_pkt ) ) break;
2483 :
2484 : /* check version */
2485 0 : uint cur_version = fd_uint_bswap( FD_LOAD( uint, cur_ptr + 1 ) );
2486 :
2487 0 : if( cur_version != version ) {
2488 : /* multiple versions in a single connection is a violation, and by
2489 : extension so is multiple versions in a single udp datagram
2490 : these are silently ignored
2491 :
2492 : for reference
2493 : all quic packets in a udp datagram must be for the same connection id
2494 : (section 12.2) and therefore the same connection
2495 : all packets on a connection must be of the same version (5.2) */
2496 0 : quic->metrics.pkt_quic_hdr_err_cnt++;
2497 0 : FD_DEBUG( FD_LOG_DEBUG(( "Mixed QUIC versions in packet" )) );
2498 0 : return;
2499 0 : }
2500 :
2501 0 : rc = fd_quic_process_quic_packet_v1( quic, &pkt, cur_ptr, cur_sz );
2502 0 : svc_cnt_eq_alloc_conn( state->svc_timers, quic );
2503 :
2504 : /* 0UL means no progress, so fail */
2505 0 : if( FD_UNLIKELY( ( rc == FD_QUIC_PARSE_FAIL ) |
2506 0 : ( rc == 0UL ) ) ) {
2507 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_process_quic_packet_v1 failed (stuck=%d)", rc==0UL )) );
2508 0 : return;
2509 0 : }
2510 :
2511 0 : if( FD_UNLIKELY( rc > cur_sz ) ) {
2512 0 : FD_DEBUG( FD_LOG_WARNING(( "fd_quic_process_quic_packet_v1 read too much" )) );
2513 0 : return;
2514 0 : }
2515 :
2516 : /* return code (rc) is the number of bytes consumed */
2517 0 : cur_sz -= rc;
2518 0 : cur_ptr += rc;
2519 0 : }
2520 0 : if( pkt_idx==FD_QUIC_PKT_COALESCE_LIMIT ) {
2521 : /* too many packets in a single udp datagram */
2522 0 : return;
2523 0 : }
2524 0 : }
2525 :
2526 : /* above can drop out of loop if a short packet is detected */
2527 0 : if( FD_UNLIKELY( cur_sz < FD_QUIC_SHORTEST_PKT ) ) return;
2528 :
2529 : /* short header packet
2530 : only one_rtt packets currently have short headers */
2531 0 : fd_quic_process_quic_packet_v1( quic, &pkt, cur_ptr, cur_sz );
2532 0 : svc_cnt_eq_alloc_conn( state->svc_timers, quic );
2533 0 : }
2534 :
2535 : void
2536 : fd_quic_process_packet( fd_quic_t * quic,
2537 : uchar * data,
2538 : ulong data_sz,
2539 0 : long now ) {
2540 0 : long now_ticks = fd_tickcount();
2541 0 : fd_quic_process_packet_impl( quic, data, data_sz, now );
2542 0 : long delta_ticks = fd_tickcount() - now_ticks;
2543 0 : fd_histf_sample( quic->metrics.receive_duration, (ulong)delta_ticks );
2544 0 : }
2545 :
2546 : /* main receive-side entry point */
2547 : int
2548 : fd_quic_aio_cb_receive( void * context,
2549 : fd_aio_pkt_info_t const * batch,
2550 : ulong batch_cnt,
2551 : ulong * opt_batch_idx,
2552 0 : int flush ) {
2553 0 : (void)flush;
2554 :
2555 0 : fd_quic_t * quic = context;
2556 0 : long now = fd_quic_get_state( quic )->now;
2557 :
2558 : /* this aio interface is configured as one-packet per buffer
2559 : so batch[0] refers to one buffer
2560 : as such, we simply forward each individual packet to a handling function */
2561 0 : for( ulong j = 0; j < batch_cnt; ++j ) {
2562 0 : fd_quic_process_packet( quic, batch[ j ].buf, batch[ j ].buf_sz, now );
2563 0 : }
2564 :
2565 : /* the assumption here at present is that any packet that could not be processed
2566 : is simply dropped
2567 : hence, all packets were consumed */
2568 0 : if( FD_LIKELY( opt_batch_idx ) ) {
2569 0 : *opt_batch_idx = batch_cnt;
2570 0 : }
2571 :
2572 0 : return FD_AIO_SUCCESS;
2573 0 : }
2574 :
2575 : void
2576 : fd_quic_tls_cb_alert( fd_quic_tls_hs_t * hs,
2577 : void * context,
2578 0 : int alert ) {
2579 0 : (void)hs;
2580 0 : fd_quic_conn_t * conn = (fd_quic_conn_t *)context;
2581 0 : (void)conn;
2582 0 : (void)alert;
2583 0 : FD_DEBUG( FD_LOG_DEBUG(( "TLS callback: %s", conn->server ? "SERVER" : "CLIENT" ));
2584 0 : FD_LOG_DEBUG(( "TLS alert: (%d-%s)", alert, fd_tls_alert_cstr( (uint)alert ) )); );
2585 :
2586 : /* TODO store alert to reply to peer */
2587 0 : }
2588 :
2589 : void
2590 : fd_quic_tls_cb_secret( fd_quic_tls_hs_t * hs,
2591 : void * context,
2592 0 : fd_quic_tls_secret_t const * secret ) {
2593 :
2594 0 : fd_quic_conn_t * conn = (fd_quic_conn_t*)context;
2595 0 : fd_quic_t * quic = conn->quic;
2596 :
2597 : /* look up suite */
2598 : /* set secrets */
2599 0 : FD_TEST( secret->enc_level < FD_QUIC_NUM_ENC_LEVELS );
2600 :
2601 0 : uint enc_level = secret->enc_level;
2602 :
2603 0 : fd_quic_crypto_secrets_t * crypto_secret = &conn->secrets;
2604 :
2605 0 : memcpy( crypto_secret->secret[enc_level][0], secret->read_secret, FD_QUIC_SECRET_SZ );
2606 0 : memcpy( crypto_secret->secret[enc_level][1], secret->write_secret, FD_QUIC_SECRET_SZ );
2607 :
2608 0 : conn->keys_avail = fd_uint_set_bit( conn->keys_avail, (int)enc_level );
2609 :
2610 : /* gen local keys */
2611 0 : fd_quic_gen_keys(
2612 0 : &conn->keys[enc_level][0],
2613 0 : conn->secrets.secret[enc_level][0] );
2614 :
2615 : /* gen peer keys */
2616 0 : fd_quic_gen_keys(
2617 0 : &conn->keys[enc_level][1],
2618 0 : conn->secrets.secret[enc_level][1] );
2619 :
2620 0 : if( enc_level==fd_quic_enc_level_appdata_id ) {
2621 0 : fd_quic_key_update_derive( &conn->secrets, conn->new_keys );
2622 0 : }
2623 :
2624 : /* Key logging */
2625 :
2626 0 : void * keylog_ctx = quic->cb.quic_ctx;
2627 0 : fd_quic_cb_tls_keylog_t keylog_fn = quic->cb.tls_keylog;
2628 0 : if( FD_UNLIKELY( keylog_fn ) ) {
2629 : /* Ignore stdout, stderr, stdin */
2630 :
2631 0 : uchar const * recv_secret = secret->read_secret;
2632 0 : uchar const * send_secret = secret->write_secret;
2633 :
2634 0 : uchar const * client_secret = hs->is_server ? recv_secret : send_secret;
2635 0 : uchar const * server_secret = hs->is_server ? send_secret : recv_secret;
2636 :
2637 0 : char buf[256];
2638 0 : char * s;
2639 0 : switch( enc_level ) {
2640 0 : case FD_TLS_LEVEL_HANDSHAKE:
2641 0 : /* 0 chars */ s = fd_cstr_init( buf );
2642 0 : /* 0+32 chars */ s = fd_cstr_append_cstr( s, "CLIENT_HANDSHAKE_TRAFFIC_SECRET " );
2643 0 : /* 32+64 chars */ s = fd_hex_encode( s, hs->hs.base.client_random, 32UL );
2644 0 : /* 96+ 1 chars */ s = fd_cstr_append_char( s, ' ' );
2645 0 : /* 97+64 chars */ s = fd_hex_encode( s, client_secret, 32UL );
2646 0 : /* 161 chars */ fd_cstr_fini( s );
2647 0 : keylog_fn( keylog_ctx, buf );
2648 0 : /* 0 chars */ s = fd_cstr_init( buf );
2649 0 : /* 0+32 chars */ s = fd_cstr_append_cstr( s, "SERVER_HANDSHAKE_TRAFFIC_SECRET " );
2650 0 : /* 32+64 chars */ s = fd_hex_encode( s, hs->hs.base.client_random, 32UL );
2651 0 : /* 96+ 1 chars */ s = fd_cstr_append_char( s, ' ' );
2652 0 : /* 97+64 chars */ s = fd_hex_encode( s, server_secret, 32UL );
2653 0 : /* 161 chars */ fd_cstr_fini( s );
2654 0 : keylog_fn( keylog_ctx, buf );
2655 0 : break;
2656 0 : case FD_TLS_LEVEL_APPLICATION:
2657 0 : /* 0 chars */ s = fd_cstr_init( buf );
2658 0 : /* 0+24 chars */ s = fd_cstr_append_cstr( s, "CLIENT_TRAFFIC_SECRET_0 " );
2659 0 : /* 24+64 chars */ s = fd_hex_encode( s, hs->hs.base.client_random, 32UL );
2660 0 : /* 88+ 1 chars */ s = fd_cstr_append_char( s, ' ' );
2661 0 : /* 89+64 chars */ s = fd_hex_encode( s, client_secret, 32UL );
2662 0 : /* 153 chars */ fd_cstr_fini( s );
2663 0 : keylog_fn( keylog_ctx, buf );
2664 0 : /* 0 chars */ s = fd_cstr_init( buf );
2665 0 : /* 0+24 chars */ s = fd_cstr_append_cstr( s, "SERVER_TRAFFIC_SECRET_0 " );
2666 0 : /* 24+64 chars */ s = fd_hex_encode( s, hs->hs.base.client_random, 32UL );
2667 0 : /* 88+ 1 chars */ s = fd_cstr_append_char( s, ' ' );
2668 0 : /* 89+64 chars */ s = fd_hex_encode( s, server_secret, 32UL );
2669 0 : /* 153 chars */ fd_cstr_fini( s );
2670 0 : keylog_fn( keylog_ctx, buf );
2671 0 : break;
2672 0 : }
2673 0 : }
2674 :
2675 0 : }
2676 :
2677 : void
2678 : fd_quic_apply_peer_params( fd_quic_conn_t * conn,
2679 0 : fd_quic_transport_params_t const * peer_tp ) {
2680 : /* flow control parameters */
2681 0 : conn->tx_max_data = peer_tp->initial_max_data;
2682 0 : conn->tx_initial_max_stream_data_uni= peer_tp->initial_max_stream_data_uni;
2683 :
2684 0 : if( !conn->server ) {
2685 : /* verify retry_src_conn_id */
2686 0 : uint retry_src_conn_id_sz = conn->retry_src_conn_id.sz;
2687 0 : if( retry_src_conn_id_sz ) {
2688 0 : if( FD_UNLIKELY( !peer_tp->retry_source_connection_id_present
2689 0 : || peer_tp->retry_source_connection_id_len != retry_src_conn_id_sz
2690 0 : || 0 != memcmp( peer_tp->retry_source_connection_id,
2691 0 : conn->retry_src_conn_id.conn_id,
2692 0 : retry_src_conn_id_sz ) ) ) {
2693 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_TRANSPORT_PARAMETER_ERROR, __LINE__ );
2694 0 : return;
2695 0 : }
2696 0 : } else {
2697 0 : if( FD_UNLIKELY( peer_tp->retry_source_connection_id_present ) ) {
2698 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_TRANSPORT_PARAMETER_ERROR, __LINE__ );
2699 0 : return;
2700 0 : }
2701 0 : }
2702 0 : }
2703 :
2704 : /* max datagram size */
2705 0 : ulong tx_max_datagram_sz = peer_tp->max_udp_payload_size;
2706 0 : if( tx_max_datagram_sz < FD_QUIC_INITIAL_PAYLOAD_SZ_MAX ) {
2707 0 : tx_max_datagram_sz = FD_QUIC_INITIAL_PAYLOAD_SZ_MAX;
2708 0 : }
2709 0 : if( tx_max_datagram_sz > FD_QUIC_INITIAL_PAYLOAD_SZ_MAX ) {
2710 0 : tx_max_datagram_sz = FD_QUIC_INITIAL_PAYLOAD_SZ_MAX;
2711 0 : }
2712 0 : conn->tx_max_datagram_sz = (uint)tx_max_datagram_sz;
2713 :
2714 : /* initial max_streams */
2715 :
2716 0 : if( conn->server ) {
2717 0 : conn->tx_sup_stream_id = ( (ulong)peer_tp->initial_max_streams_uni << 2UL ) + FD_QUIC_STREAM_TYPE_UNI_SERVER;
2718 0 : } else {
2719 0 : conn->tx_sup_stream_id = ( (ulong)peer_tp->initial_max_streams_uni << 2UL ) + FD_QUIC_STREAM_TYPE_UNI_CLIENT;
2720 0 : }
2721 :
2722 : /* set the max_idle_timeout to the min of our and peer max_idle_timeout */
2723 0 : if( peer_tp->max_idle_timeout_ms ) {
2724 0 : long peer_max_idle_timeout_ns = fd_long_sat_mul( (long)peer_tp->max_idle_timeout_ms, (long)1e6);
2725 0 : conn->idle_timeout_ns = fd_long_min( peer_max_idle_timeout_ns, conn->idle_timeout_ns );
2726 0 : }
2727 :
2728 : /* set ack_delay_exponent so we can properly interpret peer's ack_delays
2729 : if unspecified, the value is 3 */
2730 0 : ulong peer_ack_delay_exponent = fd_ulong_if(
2731 0 : peer_tp->ack_delay_exponent_present,
2732 0 : peer_tp->ack_delay_exponent,
2733 0 : 3UL );
2734 :
2735 0 : conn->peer_ack_delay_scale = (float)( 1UL << peer_ack_delay_exponent ) * 1e3f;
2736 :
2737 : /* peer max ack delay in microseconds
2738 : peer_tp->max_ack_delay is milliseconds */
2739 0 : float peer_max_ack_delay_us = (float)fd_ulong_if(
2740 0 : peer_tp->max_ack_delay_present,
2741 0 : peer_tp->max_ack_delay * 1000UL,
2742 0 : 25000UL );
2743 0 : conn->peer_max_ack_delay_ns = peer_max_ack_delay_us * 1e3f;
2744 :
2745 0 : conn->transport_params_set = 1;
2746 0 : }
2747 :
2748 : void
2749 : fd_quic_tls_cb_peer_params( void * context,
2750 : uchar const * peer_tp_enc,
2751 0 : ulong peer_tp_enc_sz ) {
2752 0 : fd_quic_conn_t * conn = (fd_quic_conn_t*)context;
2753 :
2754 : /* decode peer transport parameters */
2755 0 : fd_quic_transport_params_t peer_tp[1] = {0};
2756 0 : int rc = fd_quic_decode_transport_params( peer_tp, peer_tp_enc, peer_tp_enc_sz );
2757 0 : if( FD_UNLIKELY( rc != 0 ) ) {
2758 0 : FD_DEBUG( FD_LOG_NOTICE(( "fd_quic_decode_transport_params failed" )); )
2759 : /* failed to parse transport params */
2760 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_TRANSPORT_PARAMETER_ERROR, __LINE__ );
2761 0 : return;
2762 0 : }
2763 :
2764 0 : fd_quic_apply_peer_params( conn, peer_tp );
2765 0 : }
2766 :
2767 : void
2768 : fd_quic_tls_cb_handshake_complete( fd_quic_tls_hs_t * hs,
2769 0 : void * context ) {
2770 0 : (void)hs;
2771 0 : fd_quic_conn_t * conn = (fd_quic_conn_t *)context;
2772 :
2773 : /* need to send quic handshake completion */
2774 0 : switch( conn->state ) {
2775 0 : case FD_QUIC_CONN_STATE_ABORT:
2776 0 : case FD_QUIC_CONN_STATE_CLOSE_PENDING:
2777 0 : case FD_QUIC_CONN_STATE_DEAD:
2778 : /* ignore */
2779 0 : return;
2780 :
2781 0 : case FD_QUIC_CONN_STATE_HANDSHAKE:
2782 0 : if( FD_UNLIKELY( !conn->transport_params_set ) ) { /* unreachable */
2783 0 : FD_LOG_WARNING(( "Handshake marked as completed but transport params are not set. This is a bug!" ));
2784 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_INTERNAL_ERROR, __LINE__ );
2785 0 : return;
2786 0 : }
2787 0 : conn->handshake_complete = 1;
2788 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_HANDSHAKE_COMPLETE );
2789 0 : return;
2790 :
2791 0 : default:
2792 0 : FD_LOG_WARNING(( "handshake in unexpected state: %u", conn->state ));
2793 0 : }
2794 0 : }
2795 :
2796 : static ulong
2797 : fd_quic_handle_crypto_frame( fd_quic_frame_ctx_t * context,
2798 : fd_quic_crypto_frame_t * crypto,
2799 : uchar const * p,
2800 0 : ulong p_sz ) {
2801 : /* determine whether any of the data was already provided */
2802 0 : fd_quic_conn_t * conn = context->conn;
2803 0 : fd_quic_tls_hs_t * tls_hs = conn->tls_hs;
2804 0 : uint enc_level = context->pkt->enc_level;
2805 :
2806 : /* offset expected */
2807 0 : ulong rcv_off = crypto->offset; /* in [0,2^62-1] */
2808 0 : ulong rcv_sz = crypto->length; /* in [0,2^62-1] */
2809 0 : ulong rcv_hi = rcv_off + rcv_sz; /* in [0,2^63-1] */
2810 :
2811 0 : if( FD_UNLIKELY( rcv_sz > p_sz ) ) {
2812 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
2813 0 : return FD_QUIC_PARSE_FAIL;
2814 0 : }
2815 :
2816 0 : if( !tls_hs ) {
2817 : /* Handshake already completed. Ignore frame */
2818 : /* TODO consider aborting conn if too many unsolicited crypto frames arrive */
2819 0 : return rcv_sz;
2820 0 : }
2821 :
2822 0 : if( enc_level < tls_hs->rx_enc_level ) {
2823 0 : return rcv_sz;
2824 0 : }
2825 :
2826 0 : if( enc_level > tls_hs->rx_enc_level ) {
2827 : /* Discard data from any previous handshake level. Currently only
2828 : happens at the Initial->Handshake encryption level change. */
2829 0 : tls_hs->rx_enc_level = (uchar)enc_level;
2830 0 : tls_hs->rx_off = 0;
2831 0 : tls_hs->rx_sz = 0;
2832 0 : }
2833 :
2834 0 : if( rcv_off > tls_hs->rx_sz ) {
2835 0 : context->pkt->ack_flag |= ACK_FLAG_CANCEL;
2836 0 : return rcv_sz;
2837 0 : }
2838 :
2839 0 : if( rcv_hi < tls_hs->rx_off ) {
2840 0 : return rcv_sz;
2841 0 : }
2842 :
2843 0 : if( rcv_hi > FD_QUIC_TLS_RX_DATA_SZ ) {
2844 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_CRYPTO_BUFFER_EXCEEDED, __LINE__ );
2845 0 : return FD_QUIC_PARSE_FAIL;
2846 0 : }
2847 :
2848 0 : tls_hs->rx_sz = (ushort)rcv_hi;
2849 0 : fd_memcpy( tls_hs->rx_hs_buf + rcv_off, p, rcv_sz );
2850 :
2851 0 : int provide_rc = fd_quic_tls_process( conn->tls_hs );
2852 0 : if( provide_rc == FD_QUIC_FAILED ) {
2853 : /* if TLS fails, ABORT connection */
2854 :
2855 : /* if TLS returns an error, we present that as reason:
2856 : FD_QUIC_CONN_REASON_CRYPTO_BASE + tls-alert
2857 : otherwise, send INTERNAL_ERROR */
2858 0 : uint alert = conn->tls_hs->alert;
2859 0 : uint reason = conn->tls_hs->hs.base.reason;
2860 0 : FD_DTRACE_PROBE_3( quic_handle_crypto_frame, conn->our_conn_id, alert, reason );
2861 0 : if( alert == 0u ) {
2862 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_INTERNAL_ERROR, __LINE__ );
2863 0 : } else {
2864 0 : FD_DEBUG(
2865 0 : FD_LOG_DEBUG(( "QUIC TLS handshake failed (alert %u-%s; reason %u-%s)",
2866 0 : alert, fd_tls_alert_cstr( alert ),
2867 0 : reason, fd_tls_reason_cstr( reason ) ));
2868 0 : )
2869 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_CRYPTO_BASE + alert, __LINE__ );
2870 0 : }
2871 0 : return FD_QUIC_PARSE_FAIL;
2872 0 : }
2873 :
2874 0 : return rcv_sz;
2875 0 : }
2876 :
2877 : static int
2878 : fd_quic_svc_poll( fd_quic_t * quic,
2879 : fd_quic_conn_t * conn,
2880 0 : long now ) {
2881 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
2882 0 : if( FD_UNLIKELY( conn->state == FD_QUIC_CONN_STATE_INVALID ) ) {
2883 : /* connection shouldn't have been scheduled,
2884 : and is now removed, so just continue */
2885 0 : FD_LOG_CRIT(( "Invalid conn in schedule" ));
2886 0 : return 1;
2887 0 : }
2888 :
2889 0 : if( FD_UNLIKELY( now >= conn->last_activity + ( conn->idle_timeout_ns / 2 ) ) ) {
2890 0 : if( FD_UNLIKELY( now >= conn->last_activity + conn->idle_timeout_ns ) ) {
2891 0 : if( FD_LIKELY( conn->state != FD_QUIC_CONN_STATE_DEAD ) ) {
2892 : /* rfc9000 10.1 Idle Timeout
2893 : "... the connection is silently closed and its state is discarded
2894 : when it remains idle for longer than the minimum of the
2895 : max_idle_timeout value advertised by both endpoints." */
2896 0 : FD_DEBUG( FD_LOG_WARNING(("%s conn %p conn_idx: %u closing due to idle timeout=%gms last_activity=%ld now=%ld",
2897 0 : conn->server?"SERVER":"CLIENT",
2898 0 : (void *)conn, conn->conn_idx,
2899 0 : (double)conn->idle_timeout_ns / 1e6,
2900 0 : conn->last_activity,
2901 0 : now
2902 0 : )); )
2903 :
2904 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_DEAD );
2905 0 : quic->metrics.conn_timeout_cnt++;
2906 0 : }
2907 0 : } else if( quic->config.keep_alive & !!(conn->let_die_time_ns > now) ) {
2908 : /* send PING */
2909 0 : if( !( conn->flags & ( FD_QUIC_CONN_FLAGS_PING | FD_QUIC_CONN_FLAGS_PING_SENT ) ) ) {
2910 0 : conn->flags |= FD_QUIC_CONN_FLAGS_PING;
2911 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING; /* update to be sent in next packet */
2912 0 : }
2913 0 : }
2914 0 : }
2915 :
2916 0 : if( FD_UNLIKELY( conn->state == FD_QUIC_CONN_STATE_DEAD ) ) {
2917 0 : fd_quic_cb_conn_final( quic, conn ); /* inform user before freeing */
2918 0 : fd_quic_conn_free( quic, conn );
2919 0 : return 1; /* do NOT reschedule freed connection */
2920 0 : }
2921 :
2922 : /* state cannot be DEAD here */
2923 0 : fd_quic_conn_service( quic, conn, now );
2924 :
2925 : /* dead? don't reinsert, just clean up */
2926 0 : switch( conn->state ) {
2927 0 : case FD_QUIC_CONN_STATE_INVALID:
2928 : /* skip entirely */
2929 0 : break;
2930 0 : case FD_QUIC_CONN_STATE_DEAD:
2931 0 : fd_quic_cb_conn_final( quic, conn ); /* inform user before freeing */
2932 0 : fd_quic_conn_free( quic, conn );
2933 0 : break;
2934 0 : default: {
2935 : /* Should we schedule for keep-alive or timeout?
2936 : 1. If keep_alive not configured, for timeout
2937 : 2. Else, if we've crossed the halfway point, we must have just pinged,
2938 : and should therefore schedule timeout. Otherwise, we should schedule
2939 : for the keep-alive time. */
2940 0 : long const timeout_ns = conn->idle_timeout_ns;
2941 0 : long const last_activity = conn->last_activity;
2942 0 : int const keep_alive = quic->config.keep_alive & (now < last_activity+timeout_ns/2L);
2943 0 : fd_quic_svc_prep_schedule( conn, last_activity + (timeout_ns>>keep_alive) );
2944 0 : fd_quic_svc_timers_schedule( state->svc_timers, conn, state->now );
2945 0 : break;
2946 0 : }
2947 0 : }
2948 :
2949 0 : return 1;
2950 0 : }
2951 :
2952 : int
2953 : fd_quic_service( fd_quic_t * quic,
2954 0 : long now ) {
2955 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
2956 :
2957 0 : state->now = now;
2958 0 : long now_ticks = fd_tickcount();
2959 :
2960 0 : fd_quic_svc_timers_t * timers = state->svc_timers;
2961 0 : svc_cnt_eq_alloc_conn( timers, quic );
2962 0 : fd_quic_svc_event_t next = fd_quic_svc_timers_next( timers, now, 1 /* pop */);
2963 0 : if( FD_UNLIKELY( next.conn == NULL ) ) {
2964 0 : return 0;
2965 0 : }
2966 :
2967 0 : int cnt = fd_quic_svc_poll( quic, next.conn, now );
2968 :
2969 0 : long delta_ticks = fd_tickcount() - now_ticks;
2970 :
2971 0 : fd_histf_sample( quic->metrics.service_duration, (ulong)delta_ticks );
2972 :
2973 0 : return cnt;
2974 0 : }
2975 :
2976 : static inline ulong FD_FN_UNUSED
2977 0 : fd_quic_conn_tx_buf_remaining( fd_quic_conn_t * conn ) {
2978 0 : return (ulong)( sizeof( conn->tx_buf_conn ) - (ulong)( conn->tx_ptr - conn->tx_buf_conn ) );
2979 0 : }
2980 :
2981 : /* attempt to transmit buffered data
2982 :
2983 : prior to call, conn->tx_ptr points to the first free byte in tx_buf
2984 : the data in tx_buf..tx_ptr is prepended by networking headers
2985 : and put on the wire
2986 :
2987 : returns 0 if successful, or 1 otherwise */
2988 : uint
2989 : fd_quic_tx_buffered_raw(
2990 : fd_quic_t * quic,
2991 : uchar ** tx_ptr_ptr,
2992 : uchar * tx_buf,
2993 : ushort * ipv4_id,
2994 : uint dst_ipv4_addr,
2995 : ushort dst_udp_port,
2996 : uint src_ipv4_addr,
2997 : ushort src_udp_port
2998 0 : ) {
2999 :
3000 : /* TODO leave space at front of tx_buf for header
3001 : then encode directly into it to avoid 1 copy */
3002 0 : uchar *tx_ptr = *tx_ptr_ptr;
3003 0 : long payload_sz = tx_ptr - tx_buf;
3004 :
3005 : /* nothing to do */
3006 0 : if( FD_UNLIKELY( payload_sz<=0L ) ) {
3007 0 : return 0u;
3008 0 : }
3009 :
3010 0 : fd_quic_config_t * config = &quic->config;
3011 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
3012 :
3013 0 : uchar * const crypt_scratch = state->crypt_scratch;
3014 :
3015 0 : uchar * cur_ptr = state->crypt_scratch;
3016 0 : ulong cur_sz = sizeof( state->crypt_scratch );
3017 :
3018 : /* TODO much of this may be prepared ahead of time */
3019 0 : fd_quic_pkt_t pkt;
3020 :
3021 0 : pkt.ip4->verihl = FD_IP4_VERIHL(4,5);
3022 0 : pkt.ip4->tos = (uchar)(config->net.dscp << 2); /* could make this per-connection or per-stream */
3023 0 : pkt.ip4->net_tot_len = (ushort)( 20 + 8 + payload_sz );
3024 0 : pkt.ip4->net_id = *ipv4_id;
3025 0 : pkt.ip4->net_frag_off = 0x4000u; /* don't fragment */
3026 0 : pkt.ip4->ttl = 64; /* TODO make configurable */
3027 0 : pkt.ip4->protocol = FD_IP4_HDR_PROTOCOL_UDP;
3028 0 : pkt.ip4->check = 0;
3029 0 : pkt.ip4->saddr = src_ipv4_addr;
3030 0 : pkt.ip4->daddr = dst_ipv4_addr;
3031 0 : pkt.udp->net_sport = src_udp_port;
3032 0 : pkt.udp->net_dport = dst_udp_port;
3033 0 : pkt.udp->net_len = (ushort)( 8 + payload_sz );
3034 0 : pkt.udp->check = 0x0000;
3035 0 : *ipv4_id = (ushort)( *ipv4_id + 1 );
3036 :
3037 0 : ulong rc = fd_quic_encode_ip4( cur_ptr, cur_sz, pkt.ip4 );
3038 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
3039 0 : FD_LOG_ERR(( "fd_quic_encode_ip4 failed with buffer overrun" ));
3040 0 : }
3041 :
3042 : /* Compute checksum over network byte order header */
3043 0 : fd_ip4_hdr_t * ip4_encoded = (fd_ip4_hdr_t *)fd_type_pun( cur_ptr );
3044 0 : ip4_encoded->check = (ushort)fd_ip4_hdr_check_fast( ip4_encoded );
3045 :
3046 0 : cur_ptr += rc;
3047 0 : cur_sz -= rc;
3048 :
3049 0 : rc = fd_quic_encode_udp( cur_ptr, cur_sz, pkt.udp );
3050 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
3051 0 : FD_LOG_ERR(( "fd_quic_encode_udp failed with buffer overrun" ));
3052 0 : }
3053 :
3054 0 : cur_ptr += rc;
3055 0 : cur_sz -= rc;
3056 :
3057 : /* need enough space for payload */
3058 0 : if( FD_UNLIKELY( (ulong)payload_sz > cur_sz ) ) {
3059 0 : FD_LOG_WARNING(( "%s : payload too big for buffer", __func__ ));
3060 :
3061 : /* reset buffer, since we can't use its contents */
3062 0 : *tx_ptr_ptr = tx_buf;
3063 0 : return FD_QUIC_FAILED;
3064 0 : }
3065 0 : fd_memcpy( cur_ptr, tx_buf, (ulong)payload_sz );
3066 :
3067 0 : cur_ptr += (ulong)payload_sz;
3068 0 : cur_sz -= (ulong)payload_sz;
3069 :
3070 0 : fd_aio_pkt_info_t aio_buf = { .buf = crypt_scratch, .buf_sz = (ushort)( cur_ptr - crypt_scratch ) };
3071 0 : int aio_rc = fd_aio_send( &quic->aio_tx, &aio_buf, 1, NULL, 1 );
3072 0 : if( aio_rc == FD_AIO_ERR_AGAIN ) {
3073 : /* transient condition - try later */
3074 0 : return FD_QUIC_FAILED;
3075 0 : } else if( aio_rc != FD_AIO_SUCCESS ) {
3076 0 : FD_LOG_WARNING(( "Fatal error reported by aio peer" ));
3077 : /* fallthrough to reset buffer */
3078 0 : }
3079 :
3080 : /* after send, reset tx_ptr and tx_sz */
3081 0 : *tx_ptr_ptr = tx_buf;
3082 :
3083 0 : quic->metrics.net_tx_pkt_cnt += aio_rc==FD_AIO_SUCCESS;
3084 0 : if( FD_LIKELY( aio_rc==FD_AIO_SUCCESS ) ) {
3085 0 : quic->metrics.net_tx_byte_cnt += aio_buf.buf_sz;
3086 0 : }
3087 :
3088 0 : return FD_QUIC_SUCCESS; /* success */
3089 0 : }
3090 :
3091 : uint
3092 : fd_quic_tx_buffered( fd_quic_t * quic,
3093 0 : fd_quic_conn_t * conn ) {
3094 0 : fd_quic_net_endpoint_t const * endpoint = conn->peer;
3095 0 : return fd_quic_tx_buffered_raw(
3096 0 : quic,
3097 0 : &conn->tx_ptr,
3098 0 : conn->tx_buf_conn,
3099 0 : &conn->ipv4_id,
3100 0 : endpoint->ip_addr,
3101 0 : endpoint->udp_port,
3102 0 : conn->host.ip_addr,
3103 0 : conn->host.udp_port);
3104 0 : }
3105 :
3106 : static inline int
3107 : fd_quic_conn_can_acquire_pkt_meta( fd_quic_conn_t * conn,
3108 0 : fd_quic_pkt_meta_tracker_t * tracker ) {
3109 0 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
3110 0 : fd_quic_metrics_t * metrics = &conn->quic->metrics;
3111 :
3112 0 : ulong pool_free = fd_quic_pkt_meta_pool_free( tracker->pool );
3113 0 : if( !pool_free || conn->used_pkt_meta >= state->max_inflight_frame_cnt_conn ) {
3114 0 : if( !pool_free ) {
3115 0 : metrics->frame_tx_alloc_cnt[FD_METRICS_ENUM_FRAME_TX_ALLOC_RESULT_V_FAIL_EMPTY_POOL_IDX]++;
3116 0 : } else {
3117 0 : metrics->frame_tx_alloc_cnt[FD_METRICS_ENUM_FRAME_TX_ALLOC_RESULT_V_FAIL_CONN_MAX_IDX]++;
3118 0 : }
3119 0 : return 0;
3120 0 : }
3121 0 : metrics->frame_tx_alloc_cnt[FD_METRICS_ENUM_FRAME_TX_ALLOC_RESULT_V_SUCCESS_IDX]++;
3122 :
3123 0 : return 1;
3124 0 : }
3125 :
3126 : /* fd_quic_gen_frame_store_pkt_meta stores a pkt_meta into tracker.
3127 : Value and type take the passed args; all other fields are copied
3128 : from pkt_meta_tmpl. Returns 1 if successful, 0 if not.
3129 : Failure reasons include empty pkt_meta pool, or this conn reached
3130 : its pkt_meta limit. Theoretically only need latter, but let's be safe! */
3131 : static inline int
3132 : fd_quic_gen_frame_store_pkt_meta( const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3133 : uchar type,
3134 : fd_quic_pkt_meta_value_t value,
3135 : fd_quic_pkt_meta_tracker_t * tracker,
3136 0 : fd_quic_conn_t * conn ) {
3137 0 : if( !fd_quic_conn_can_acquire_pkt_meta( conn, tracker ) ) return 0;
3138 :
3139 0 : conn->used_pkt_meta++;
3140 0 : fd_quic_pkt_meta_t * pkt_meta = fd_quic_pkt_meta_pool_ele_acquire( tracker->pool );
3141 0 : *pkt_meta = *pkt_meta_tmpl;
3142 0 : FD_QUIC_PKT_META_SET_TYPE( pkt_meta, type );
3143 0 : pkt_meta->val = value;
3144 0 : fd_quic_pkt_meta_insert( &tracker->sent_pkt_metas[pkt_meta->enc_level], pkt_meta, tracker->pool );
3145 0 : return 1;
3146 0 : }
3147 :
3148 : static ulong
3149 : fd_quic_gen_close_frame( fd_quic_conn_t * conn,
3150 : uchar * payload_ptr,
3151 : uchar * payload_end,
3152 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3153 0 : fd_quic_pkt_meta_tracker_t * tracker ) {
3154 :
3155 0 : if( conn->flags & FD_QUIC_CONN_FLAGS_CLOSE_SENT ) return 0UL;
3156 0 : conn->flags |= FD_QUIC_CONN_FLAGS_CLOSE_SENT;
3157 :
3158 0 : ulong frame_sz;
3159 0 : if( conn->reason != 0u || conn->state == FD_QUIC_CONN_STATE_PEER_CLOSE ) {
3160 0 : fd_quic_conn_close_0_frame_t frame = {
3161 0 : .error_code = conn->reason,
3162 0 : .frame_type = 0u, /* we do not know the frame in question */
3163 0 : .reason_phrase_length = 0u /* no reason phrase */
3164 0 : };
3165 0 : frame_sz = fd_quic_encode_conn_close_0_frame( payload_ptr,
3166 0 : (ulong)( payload_end - payload_ptr ),
3167 0 : &frame );
3168 0 : } else {
3169 0 : fd_quic_conn_close_1_frame_t frame = {
3170 0 : .error_code = conn->app_reason,
3171 0 : .reason_phrase_length = 0u /* no reason phrase */
3172 0 : };
3173 0 : frame_sz = fd_quic_encode_conn_close_1_frame( payload_ptr,
3174 0 : (ulong)( payload_end - payload_ptr ),
3175 0 : &frame );
3176 0 : }
3177 :
3178 0 : if( FD_UNLIKELY( frame_sz == FD_QUIC_PARSE_FAIL ) ) {
3179 0 : FD_LOG_WARNING(( "fd_quic_encode_conn_close_frame failed, but space should have been available" ));
3180 0 : return 0UL;
3181 0 : }
3182 :
3183 : /* create and save pkt_meta, return 0 if fail */
3184 0 : if( !fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3185 0 : FD_QUIC_PKT_META_TYPE_CLOSE,
3186 0 : (fd_quic_pkt_meta_value_t){0}, /* value doesn't matter */
3187 0 : tracker,
3188 0 : conn )) return 0UL;
3189 :
3190 0 : return frame_sz;
3191 0 : }
3192 :
3193 : static uchar *
3194 : fd_quic_gen_handshake_frames( fd_quic_conn_t * conn,
3195 : uchar * payload_ptr,
3196 : uchar * payload_end,
3197 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3198 0 : fd_quic_pkt_meta_tracker_t * tracker ) {
3199 0 : uint enc_level = pkt_meta_tmpl->enc_level;
3200 0 : fd_quic_tls_hs_data_t * hs_data = fd_quic_tls_get_hs_data( conn->tls_hs, enc_level );
3201 0 : if( !hs_data ) return payload_ptr;
3202 :
3203 : /* confirm we have pkt_meta space */
3204 0 : if( !fd_quic_conn_can_acquire_pkt_meta( conn, tracker ) ) return payload_ptr;
3205 :
3206 0 : ulong hs_offset = 0; /* offset within the current hs_data */
3207 0 : ulong sent_offset = conn->hs_sent_bytes[enc_level];
3208 0 : ulong ackd_offset = conn->hs_ackd_bytes[enc_level];
3209 : /* offset within stream */
3210 0 : ulong offset = fd_ulong_max( sent_offset, ackd_offset );
3211 :
3212 : /* track pkt_meta values */
3213 0 : ulong offset_lo = offset;
3214 0 : ulong offset_hi = offset;
3215 :
3216 0 : while( hs_data ) {
3217 : /* skip data we've sent */
3218 0 : if( hs_data->offset + hs_data->data_sz <= offset ) {
3219 0 : hs_data = fd_quic_tls_get_next_hs_data( conn->tls_hs, hs_data );
3220 0 : continue;
3221 0 : }
3222 :
3223 0 : if( FD_UNLIKELY( hs_data->offset > offset ) ) {
3224 : /* we have a gap - this shouldn't happen */
3225 0 : FD_LOG_WARNING(( "%s - gap in TLS handshake data", __func__ ));
3226 : /* TODO should probably tear down connection */
3227 0 : break;
3228 0 : }
3229 :
3230 : /* encode hs_data into frame */
3231 0 : hs_offset = offset - hs_data->offset;
3232 :
3233 : /* handshake data to send */
3234 0 : uchar const * cur_data = hs_data->data + hs_offset;
3235 0 : ulong cur_data_sz = hs_data->data_sz - hs_offset;
3236 :
3237 : /* 9 bytes header + cur_data_sz */
3238 0 : if( payload_ptr + 9UL + cur_data_sz > payload_end ) break;
3239 : /* FIXME reduce cur_data_sz if it doesn't fit in frame
3240 : Practically don't need to, because fd_tls generates a small amount of data */
3241 :
3242 0 : payload_ptr[0] = 0x06; /* CRYPTO frame */
3243 0 : uint offset_varint = 0x80U | ( fd_uint_bswap( (uint)offset & 0x3fffffffU ) );
3244 0 : uint length_varint = 0x80U | ( fd_uint_bswap( (uint)cur_data_sz & 0x3fffffffU ) );
3245 0 : FD_STORE( uint, payload_ptr+1, offset_varint );
3246 0 : FD_STORE( uint, payload_ptr+5, length_varint );
3247 0 : payload_ptr += 9;
3248 :
3249 0 : fd_memcpy( payload_ptr, cur_data, cur_data_sz );
3250 0 : payload_ptr += cur_data_sz;
3251 :
3252 : /* update pkt_meta values */
3253 0 : offset_hi += cur_data_sz;
3254 :
3255 : /* move to next hs_data */
3256 0 : offset += cur_data_sz;
3257 0 : conn->hs_sent_bytes[enc_level] += cur_data_sz;
3258 :
3259 : /* TODO load more hs_data into a crypto frame, if available
3260 : currently tricky, because encode_crypto_frame copies payload */
3261 0 : }
3262 :
3263 : /* update packet meta */
3264 0 : if( offset_hi > offset_lo ) {
3265 0 : fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3266 0 : FD_QUIC_PKT_META_TYPE_HS_DATA,
3267 0 : (fd_quic_pkt_meta_value_t){
3268 0 : .range = {
3269 0 : .offset_lo = offset_lo,
3270 0 : .offset_hi = offset_hi
3271 0 : }
3272 0 : },
3273 0 : tracker,
3274 0 : conn );
3275 0 : }
3276 :
3277 0 : return payload_ptr;
3278 0 : }
3279 :
3280 : static ulong
3281 : fd_quic_gen_handshake_done_frame( fd_quic_conn_t * conn,
3282 : uchar * payload_ptr,
3283 : uchar * payload_end,
3284 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3285 0 : fd_quic_pkt_meta_tracker_t * tracker ) {
3286 0 : FD_DTRACE_PROBE_1( quic_gen_handshake_done_frame, conn->our_conn_id );
3287 0 : if( conn->handshake_done_send==0 ) return 0UL;
3288 0 : conn->handshake_done_send = 0;
3289 0 : if( FD_UNLIKELY( conn->handshake_done_ackd ) ) return 0UL;
3290 0 : if( FD_UNLIKELY( payload_ptr >= payload_end ) ) return 0UL;
3291 : /* send handshake done frame */
3292 0 : payload_ptr[0] = 0x1E;
3293 :
3294 : /* record the send for retx */
3295 0 : if( !fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3296 0 : FD_QUIC_PKT_META_TYPE_HS_DONE,
3297 0 : (fd_quic_pkt_meta_value_t){0}, /* value doesn't matter */
3298 0 : tracker,
3299 0 : conn) ) return 0UL;
3300 :
3301 0 : return 1UL;
3302 0 : }
3303 :
3304 : static ulong
3305 : fd_quic_gen_max_data_frame( fd_quic_conn_t * conn,
3306 : uchar * payload_ptr,
3307 : uchar * payload_end,
3308 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3309 0 : fd_quic_pkt_meta_tracker_t * tracker ) {
3310 0 : fd_quic_conn_stream_rx_t * srx = conn->srx;
3311 :
3312 0 : if( !( conn->flags & FD_QUIC_CONN_FLAGS_MAX_DATA ) ) return 0UL;
3313 0 : if( srx->rx_max_data <= srx->rx_max_data_ackd ) return 0UL; /* peer would ignore anyway */
3314 :
3315 : /* send max_data frame */
3316 0 : fd_quic_max_data_frame_t frame = { .max_data = srx->rx_max_data };
3317 :
3318 : /* attempt to write into buffer */
3319 0 : ulong frame_sz = fd_quic_encode_max_data_frame( payload_ptr,
3320 0 : (ulong)( payload_end - payload_ptr ),
3321 0 : &frame );
3322 0 : if( FD_UNLIKELY( frame_sz==FD_QUIC_ENCODE_FAIL ) ) return 0UL;
3323 :
3324 : /* acquire and set a pkt_meta, return 0 if not successful */
3325 0 : if( !fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3326 0 : FD_QUIC_PKT_META_TYPE_MAX_DATA,
3327 0 : (fd_quic_pkt_meta_value_t){
3328 0 : .scalar = srx->rx_max_data
3329 0 : },
3330 0 : tracker,
3331 0 : conn ) ) return 0UL;
3332 :
3333 0 : conn->upd_pkt_number = pkt_meta_tmpl->key.pkt_num;
3334 0 : return frame_sz;
3335 0 : }
3336 :
3337 : static ulong
3338 : fd_quic_gen_max_streams_frame( fd_quic_conn_t * conn,
3339 : uchar * payload_ptr,
3340 : uchar * payload_end,
3341 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3342 0 : fd_quic_pkt_meta_tracker_t * tracker ) {
3343 0 : fd_quic_conn_stream_rx_t * srx = conn->srx;
3344 :
3345 : /* 0x02 Client-Initiated, Unidirectional
3346 : 0x03 Server-Initiated, Unidirectional */
3347 0 : ulong max_streams_unidir = srx->rx_sup_stream_id >> 2;
3348 :
3349 0 : uint flags = conn->flags;
3350 0 : if( !FD_QUIC_MAX_STREAMS_ALWAYS_UNLESS_ACKED ) {
3351 0 : if( !( flags & FD_QUIC_CONN_FLAGS_MAX_STREAMS_UNIDIR ) ) return 0UL;
3352 0 : if( max_streams_unidir <= srx->rx_max_streams_unidir_ackd ) return 0UL;
3353 0 : }
3354 :
3355 0 : fd_quic_max_streams_frame_t max_streams = {
3356 0 : .type = 0x13, /* unidirectional */
3357 0 : .max_streams = max_streams_unidir
3358 0 : };
3359 0 : ulong frame_sz = fd_quic_encode_max_streams_frame( payload_ptr,
3360 0 : (ulong)( payload_end - payload_ptr ),
3361 0 : &max_streams );
3362 0 : if( FD_UNLIKELY( frame_sz==FD_QUIC_ENCODE_FAIL ) ) return 0UL;
3363 :
3364 0 : if( !fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3365 0 : FD_QUIC_PKT_META_TYPE_MAX_STREAMS_UNIDIR,
3366 0 : (fd_quic_pkt_meta_value_t){0}, /* value doesn't matter */
3367 0 : tracker,
3368 0 : conn ) ) return 0UL;
3369 :
3370 0 : conn->flags = flags & (~FD_QUIC_CONN_FLAGS_MAX_STREAMS_UNIDIR);
3371 0 : conn->upd_pkt_number = pkt_meta_tmpl->key.pkt_num;
3372 0 : return frame_sz;
3373 0 : }
3374 :
3375 : static ulong
3376 : fd_quic_gen_ping_frame( fd_quic_conn_t * conn,
3377 : uchar * payload_ptr,
3378 : uchar * payload_end,
3379 : const fd_quic_pkt_meta_t * pkt_meta_tmpl,
3380 0 : fd_quic_pkt_meta_tracker_t * tracker ) {
3381 :
3382 0 : if( ~conn->flags & FD_QUIC_CONN_FLAGS_PING ) return 0UL;
3383 0 : if( conn->flags & FD_QUIC_CONN_FLAGS_PING_SENT ) return 0UL;
3384 :
3385 0 : fd_quic_ping_frame_t ping = {0};
3386 0 : ulong frame_sz = fd_quic_encode_ping_frame( payload_ptr,
3387 0 : (ulong)( payload_end - payload_ptr ),
3388 0 : &ping );
3389 0 : if( FD_UNLIKELY( frame_sz==FD_QUIC_ENCODE_FAIL ) ) return 0UL;
3390 0 : conn->flags |= FD_QUIC_CONN_FLAGS_PING_SENT;
3391 0 : conn->flags &= ~FD_QUIC_CONN_FLAGS_PING;
3392 :
3393 0 : conn->upd_pkt_number = pkt_meta_tmpl->key.pkt_num;
3394 : /* record the send for retx, 0 if fail */
3395 0 : if( !fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3396 0 : FD_QUIC_PKT_META_TYPE_PING,
3397 0 : (fd_quic_pkt_meta_value_t){0}, /* value doesn't matter */
3398 0 : tracker,
3399 0 : conn ) ) return 0UL;
3400 :
3401 0 : return frame_sz;
3402 0 : }
3403 :
3404 : uchar *
3405 : fd_quic_gen_stream_frames( fd_quic_conn_t * conn,
3406 : uchar * payload_ptr,
3407 : uchar * payload_end,
3408 : fd_quic_pkt_meta_t * pkt_meta_tmpl,
3409 0 : fd_quic_pkt_meta_tracker_t * tracker ) {
3410 :
3411 : /* loop serves two purposes:
3412 : 1. finds a stream with data to send
3413 : 2. appends max_stream_data frames as necessary */
3414 0 : fd_quic_stream_t * sentinel = conn->send_streams;
3415 0 : fd_quic_stream_t * cur_stream = sentinel->next;
3416 0 : ulong pkt_num = pkt_meta_tmpl->key.pkt_num;
3417 0 : while( !cur_stream->sentinel ) {
3418 : /* required, since cur_stream may get removed from list */
3419 0 : fd_quic_stream_t * nxt_stream = cur_stream->next;
3420 0 : _Bool sent_all_data = 1u;
3421 :
3422 0 : if( cur_stream->upd_pkt_number >= pkt_num ) {
3423 :
3424 : /* any stream data? */
3425 0 : if( FD_LIKELY( FD_QUIC_STREAM_ACTION( cur_stream ) ) ) {
3426 :
3427 : /* data_avail is the number of stream bytes available for sending.
3428 : fin_flag_set is 1 if no more bytes will get added to the stream. */
3429 0 : ulong const data_avail = cur_stream->tx_buf.head - cur_stream->tx_sent;
3430 0 : int const fin_flag_set = !!(cur_stream->state & FD_QUIC_STREAM_STATE_TX_FIN);
3431 0 : ulong const stream_id = cur_stream->stream_id;
3432 0 : ulong const stream_off = cur_stream->tx_sent;
3433 :
3434 : /* No information to send? */
3435 0 : if( data_avail==0u && !fin_flag_set ) break;
3436 :
3437 : /* No space to write frame?
3438 : (Buffer should fit max stream header size and at least 1 byte of data) */
3439 0 : if( payload_ptr+FD_QUIC_MAX_FOOTPRINT( stream_e_frame )+1 > payload_end ) break;
3440 :
3441 : /* check pkt_meta availability */
3442 0 : if( !fd_quic_conn_can_acquire_pkt_meta( conn, tracker ) ) break;
3443 :
3444 : /* Leave placeholder for frame/stream type */
3445 0 : uchar * const frame_type_p = payload_ptr++;
3446 0 : uint frame_type = 0x0a; /* stream frame with length */
3447 :
3448 : /* Encode stream ID */
3449 0 : payload_ptr += fd_quic_varint_encode( payload_ptr, stream_id );
3450 :
3451 : /* Optionally encode offset */
3452 0 : if( stream_off>0 ) {
3453 0 : frame_type |= 0x04; /* with offset field */
3454 0 : payload_ptr += fd_quic_varint_encode( payload_ptr, stream_off );
3455 0 : }
3456 :
3457 : /* Leave placeholder for length length */
3458 0 : uchar * data_sz_p = payload_ptr;
3459 0 : payload_ptr += 2;
3460 :
3461 : /* Stream metadata */
3462 0 : ulong data_max = (ulong)payload_end - (ulong)payload_ptr; /* assume no underflow */
3463 0 : ulong data_sz = fd_ulong_min( data_avail, data_max );
3464 0 : /* */ data_sz = fd_ulong_min( data_sz, 0x3fffUL ); /* max 2 byte varint */
3465 0 : /* */ sent_all_data = data_sz == data_avail;
3466 0 : _Bool fin = fin_flag_set && sent_all_data;
3467 :
3468 : /* Finish encoding stream header */
3469 0 : ushort data_sz_varint = fd_ushort_bswap( (ushort)( 0x4000u | (uint)data_sz ) );
3470 0 : FD_STORE( ushort, data_sz_p, data_sz_varint );
3471 0 : frame_type |= fin;
3472 0 : *frame_type_p = (uchar)frame_type;
3473 :
3474 : /* Write stream payload */
3475 0 : fd_quic_buffer_t * tx_buf = &cur_stream->tx_buf;
3476 0 : fd_quic_buffer_load( tx_buf, stream_off, payload_ptr, data_sz );
3477 0 : payload_ptr += data_sz;
3478 :
3479 : /* Update stream metadata */
3480 0 : cur_stream->tx_sent += data_sz;
3481 0 : cur_stream->upd_pkt_number = fd_ulong_if( fin, pkt_num, FD_QUIC_PKT_NUM_PENDING );
3482 0 : cur_stream->stream_flags &= fd_uint_if( fin, ~FD_QUIC_STREAM_FLAGS_ACTION, UINT_MAX );
3483 :
3484 : /* Packet metadata for potential retransmits */
3485 0 : pkt_meta_tmpl->key.stream_id = cur_stream->stream_id;
3486 0 : fd_quic_gen_frame_store_pkt_meta( pkt_meta_tmpl,
3487 0 : FD_QUIC_PKT_META_TYPE_STREAM,
3488 0 : (fd_quic_pkt_meta_value_t){
3489 0 : .range = {
3490 0 : .offset_lo = stream_off,
3491 0 : .offset_hi = stream_off + data_sz
3492 0 : }
3493 0 : },
3494 0 : tracker,
3495 0 : conn );
3496 0 : }
3497 0 : }
3498 :
3499 0 : if( sent_all_data ) {
3500 0 : cur_stream->stream_flags &= ~FD_QUIC_STREAM_FLAGS_ACTION;
3501 0 : FD_QUIC_STREAM_LIST_REMOVE( cur_stream );
3502 0 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->used_streams, cur_stream );
3503 0 : }
3504 :
3505 0 : cur_stream = nxt_stream;
3506 0 : }
3507 :
3508 0 : return payload_ptr;
3509 0 : }
3510 :
3511 : uchar *
3512 : fd_quic_gen_frames( fd_quic_conn_t * conn,
3513 : uchar * payload_ptr,
3514 : uchar * payload_end,
3515 : fd_quic_pkt_meta_t * pkt_meta_tmpl,
3516 0 : long now ) {
3517 :
3518 0 : uint closing = 0U;
3519 0 : switch( conn->state ) {
3520 0 : case FD_QUIC_CONN_STATE_PEER_CLOSE:
3521 0 : case FD_QUIC_CONN_STATE_ABORT:
3522 0 : case FD_QUIC_CONN_STATE_CLOSE_PENDING:
3523 0 : closing = 1u;
3524 0 : }
3525 :
3526 0 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
3527 :
3528 0 : payload_ptr = fd_quic_gen_ack_frames( conn->ack_gen, payload_ptr, payload_end, pkt_meta_tmpl->enc_level, now );
3529 0 : if( conn->ack_gen->head == conn->ack_gen->tail ) conn->unacked_sz = 0UL;
3530 :
3531 0 : if( FD_UNLIKELY( closing ) ) {
3532 0 : payload_ptr += fd_quic_gen_close_frame( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3533 0 : } else {
3534 0 : payload_ptr = fd_quic_gen_handshake_frames( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3535 0 : if( pkt_meta_tmpl->enc_level == fd_quic_enc_level_appdata_id ) {
3536 0 : payload_ptr += fd_quic_gen_handshake_done_frame( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3537 0 : if( conn->upd_pkt_number >= pkt_meta_tmpl->key.pkt_num ) {
3538 0 : payload_ptr += fd_quic_gen_max_data_frame ( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3539 0 : payload_ptr += fd_quic_gen_max_streams_frame( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3540 0 : payload_ptr += fd_quic_gen_ping_frame ( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3541 0 : }
3542 0 : if( FD_LIKELY( !conn->tls_hs ) ) {
3543 0 : payload_ptr = fd_quic_gen_stream_frames( conn, payload_ptr, payload_end, pkt_meta_tmpl, tracker );
3544 0 : }
3545 0 : }
3546 0 : }
3547 :
3548 0 : fd_quic_svc_prep_schedule( conn, pkt_meta_tmpl->expiry );
3549 :
3550 0 : return payload_ptr;
3551 0 : }
3552 :
3553 : /* transmit
3554 : looks at each of the following dependent on state, and creates
3555 : a packet to transmit:
3556 : acks
3557 : handshake data (tls)
3558 : handshake done
3559 : ping
3560 : stream data */
3561 : static void
3562 : fd_quic_conn_tx( fd_quic_t * quic,
3563 0 : fd_quic_conn_t * conn ) {
3564 :
3565 0 : if( FD_UNLIKELY( conn->state == FD_QUIC_CONN_STATE_DEAD ) ) return;
3566 :
3567 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
3568 :
3569 : /* used for encoding frames into before encrypting */
3570 0 : uchar * crypt_scratch = state->crypt_scratch;
3571 0 : ulong crypt_scratch_sz = sizeof( state->crypt_scratch );
3572 :
3573 : /* max packet size */
3574 : /* TODO probably should be called tx_max_udp_payload_sz */
3575 0 : ulong tx_max_datagram_sz = conn->tx_max_datagram_sz;
3576 :
3577 0 : if( conn->tx_ptr != conn->tx_buf_conn ) {
3578 0 : fd_quic_tx_buffered( quic, conn );
3579 0 : fd_quic_svc_prep_schedule( conn, state->now );
3580 0 : return;
3581 0 : }
3582 :
3583 : /* choose enc_level to tx at */
3584 : /* this function accepts an argument "acks"
3585 : * We want to minimize the number of packets that carry only acks.
3586 : * fd_quic_tx_enc_level determines whether a packet needs sending,
3587 : * and when encryption level should be used.
3588 : * If "acks" is set to 1 (true), fd_quic_tx_enc_level checks for acks.
3589 : * Otherwise, it does not check for acks
3590 : * We set "acks" only on the first call in this function. All subsequent
3591 : * calls do not set it.
3592 : * This ensures that ack-only packets only occur when nothing else needs
3593 : * to be sent */
3594 0 : uint enc_level = fd_quic_tx_enc_level( conn, 1 /* acks */ );
3595 :
3596 : /* nothing to send / bad state? */
3597 0 : if( enc_level == ~0u ) return;
3598 :
3599 0 : int key_phase_upd = (int)conn->key_update;
3600 0 : uint key_phase = conn->key_phase;
3601 0 : int key_phase_tx = (int)key_phase ^ key_phase_upd;
3602 :
3603 : /* get time */
3604 0 : long now = state->now;
3605 :
3606 : /* initialize expiry and tx_time */
3607 0 : long expiry = now + fd_quic_calc_expiry_duration( conn, 0 /* use PTO */, conn->server );
3608 0 : fd_quic_pkt_meta_t pkt_meta_tmpl[1] = {{.expiry = expiry, .tx_time = now}};
3609 :
3610 0 : while( enc_level != ~0u ) {
3611 : /* RFC 9000 Section 17.2.2.1. Abandoning Initial Packets
3612 : > A client stops both sending and processing Initial packets when
3613 : > it sends its first Handshake packet.
3614 :
3615 : RFC 9001 Section 4.9.1 Discarding Initial Keys
3616 : > a client MUST discard Initial keys when it first sends a Handshake packet */
3617 0 : if( FD_UNLIKELY( (quic->config.role==FD_QUIC_ROLE_CLIENT) & (enc_level==fd_quic_enc_level_handshake_id) ) ) {
3618 0 : fd_quic_abandon_enc_level( conn, fd_quic_enc_level_initial_id );
3619 0 : }
3620 :
3621 0 : uint initial_pkt = 0; /* is this the first initial packet? */
3622 :
3623 : /* remaining in datagram */
3624 : /* invariant: tx_ptr >= tx_buf */
3625 0 : ulong datagram_rem = tx_max_datagram_sz - (ulong)( conn->tx_ptr - conn->tx_buf_conn );
3626 :
3627 : /* encode into here */
3628 : /* this is the start of a new quic packet
3629 : cur_ptr points at the next byte to fill with a quic pkt */
3630 : /* currently, cur_ptr just points at the start of crypt_scratch
3631 : each quic packet gets encrypted into tx_buf, and the space in
3632 : crypt_scratch is reused */
3633 0 : uchar * cur_ptr = crypt_scratch;
3634 0 : ulong cur_sz = crypt_scratch_sz;
3635 :
3636 : /* TODO determine actual datagrams size to use */
3637 0 : cur_sz = fd_ulong_min( cur_sz, datagram_rem );
3638 :
3639 : /* determine pn_space */
3640 0 : uint pn_space = fd_quic_enc_level_to_pn_space( enc_level );
3641 0 : pkt_meta_tmpl->pn_space = (uchar)pn_space;
3642 0 : pkt_meta_tmpl->enc_level = (uchar)(enc_level&0x3);
3643 :
3644 : /* get next packet number
3645 : Returned to pool if not sent as gaps are harmful for ACK frame
3646 : compression. */
3647 0 : ulong pkt_number = conn->pkt_number[pn_space];
3648 0 : FD_QUIC_PKT_META_SET_PKT_NUM( pkt_meta_tmpl, pkt_number );
3649 :
3650 : /* are we the client initial packet? */
3651 0 : ulong hs_data_offset = conn->hs_sent_bytes[enc_level];
3652 0 : initial_pkt = (uint)( hs_data_offset == 0 ) & (uint)( !conn->server ) & (uint)( enc_level == fd_quic_enc_level_initial_id );
3653 :
3654 : /* current peer endpoint */
3655 0 : fd_quic_conn_id_t const * peer_conn_id = &conn->peer_cids[0];
3656 :
3657 : /* our current conn_id */
3658 0 : ulong conn_id = conn->our_conn_id;
3659 0 : uint const pkt_num_len = 4u; /* 4-byte packet number */
3660 0 : uint const pkt_num_len_enc = pkt_num_len - 1; /* -1 offset for protocol */
3661 :
3662 :
3663 : /* encode packet header (including packet number)
3664 : While encoding, remember where the 'length' field is, if one
3665 : exists. We'll have to update it later. */
3666 0 : uchar * hdr_ptr = cur_ptr;
3667 0 : ulong hdr_sz = 0UL;
3668 0 : uchar _hdr_len_field[2]; /* if no len field exists, catch the write here */
3669 0 : uchar * hdr_len_field = _hdr_len_field;
3670 0 : switch( enc_level ) {
3671 0 : case fd_quic_enc_level_initial_id: {
3672 0 : fd_quic_initial_t initial = {0};
3673 0 : initial.h0 = fd_quic_initial_h0( pkt_num_len_enc );
3674 0 : initial.version = 1;
3675 0 : initial.dst_conn_id_len = peer_conn_id->sz;
3676 : // .dst_conn_id
3677 0 : initial.src_conn_id_len = FD_QUIC_CONN_ID_SZ;
3678 : // .src_conn_id
3679 : // .token - below
3680 0 : initial.len = 0x3fff; /* use 2 byte varint encoding */
3681 0 : initial.pkt_num = pkt_number;
3682 :
3683 0 : fd_memcpy( initial.dst_conn_id, peer_conn_id->conn_id, peer_conn_id->sz );
3684 0 : memcpy( initial.src_conn_id, &conn_id, FD_QUIC_CONN_ID_SZ );
3685 :
3686 : /* Initial packets sent by the server MUST set the Token Length field to 0. */
3687 0 : initial.token = conn->token;
3688 0 : if( conn->quic->config.role == FD_QUIC_ROLE_CLIENT && conn->token_len ) {
3689 0 : initial.token_len = conn->token_len;
3690 0 : } else {
3691 0 : initial.token_len = 0;
3692 0 : }
3693 :
3694 0 : hdr_sz = fd_quic_encode_initial( cur_ptr, cur_sz, &initial );
3695 0 : hdr_len_field = cur_ptr + hdr_sz - 6; /* 2 byte len, 4 byte packet number */
3696 0 : FD_DTRACE_PROBE_2( quic_encode_initial, initial.src_conn_id, initial.dst_conn_id );
3697 0 : break;
3698 0 : }
3699 :
3700 0 : case fd_quic_enc_level_handshake_id: {
3701 0 : fd_quic_handshake_t handshake = {0};
3702 0 : handshake.h0 = fd_quic_handshake_h0( pkt_num_len_enc );
3703 0 : handshake.version = 1;
3704 :
3705 : /* destination */
3706 0 : fd_memcpy( handshake.dst_conn_id, peer_conn_id->conn_id, peer_conn_id->sz );
3707 0 : handshake.dst_conn_id_len = peer_conn_id->sz;
3708 :
3709 : /* source */
3710 0 : FD_STORE( ulong, handshake.src_conn_id, conn_id );
3711 0 : handshake.src_conn_id_len = sizeof(ulong);
3712 :
3713 0 : handshake.len = 0x3fff; /* use 2 byte varint encoding */
3714 0 : handshake.pkt_num = pkt_number;
3715 :
3716 0 : hdr_sz = fd_quic_encode_handshake( cur_ptr, cur_sz, &handshake );
3717 0 : hdr_len_field = cur_ptr + hdr_sz - 6; /* 2 byte len, 4 byte packet number */
3718 0 : FD_DTRACE_PROBE_2( quic_encode_handshake, handshake.src_conn_id, handshake.dst_conn_id );
3719 0 : break;
3720 0 : }
3721 :
3722 0 : case fd_quic_enc_level_appdata_id:
3723 0 : {
3724 0 : fd_quic_one_rtt_t one_rtt = {0};
3725 0 : one_rtt.h0 = fd_quic_one_rtt_h0( /* spin */ 0, !!key_phase_tx, pkt_num_len_enc );
3726 :
3727 : /* destination */
3728 0 : fd_memcpy( one_rtt.dst_conn_id, peer_conn_id->conn_id, peer_conn_id->sz );
3729 0 : one_rtt.dst_conn_id_len = peer_conn_id->sz;
3730 :
3731 0 : one_rtt.pkt_num = pkt_number;
3732 :
3733 0 : hdr_sz = fd_quic_encode_one_rtt( cur_ptr, cur_sz, &one_rtt );
3734 0 : FD_DTRACE_PROBE_2( quic_encode_one_rtt, one_rtt.dst_conn_id, one_rtt.pkt_num );
3735 0 : break;
3736 0 : }
3737 :
3738 0 : default:
3739 0 : FD_LOG_ERR(( "%s - logic error: unexpected enc_level", __func__ ));
3740 0 : }
3741 :
3742 : /* if we don't have reasonable amt of space for a new packet, tx to free space */
3743 0 : const ulong min_rqd = 64;
3744 0 : if( FD_UNLIKELY( hdr_sz==FD_QUIC_ENCODE_FAIL || hdr_sz + min_rqd > cur_sz ) ) {
3745 : /* try to free space */
3746 0 : fd_quic_tx_buffered( quic, conn );
3747 :
3748 : /* we have lots of space, so try again */
3749 0 : if( conn->tx_buf_conn == conn->tx_ptr ) {
3750 0 : enc_level = fd_quic_tx_enc_level( conn, 0 /* acks */ );
3751 0 : continue;
3752 0 : }
3753 :
3754 : /* reschedule, since some data was unable to be sent */
3755 : /* TODO might want to add a backoff here */
3756 0 : fd_quic_svc_prep_schedule( conn, now );
3757 :
3758 0 : break;
3759 0 : }
3760 :
3761 0 : cur_ptr += hdr_sz;
3762 0 : cur_sz -= hdr_sz;
3763 :
3764 : /* start writing payload, leaving room for header and expansion
3765 : due to varint coding */
3766 :
3767 0 : uchar * payload_ptr = cur_ptr;
3768 0 : ulong payload_sz = cur_sz;
3769 : /* payload_end leaves room for TAG */
3770 0 : uchar * payload_end = payload_ptr + payload_sz - FD_QUIC_CRYPTO_TAG_SZ;
3771 :
3772 0 : uchar * const frame_start = payload_ptr;
3773 0 : payload_ptr = fd_quic_gen_frames( conn, frame_start, payload_end, pkt_meta_tmpl, now );
3774 0 : if( FD_UNLIKELY( payload_ptr < frame_start ) ) FD_LOG_CRIT(( "fd_quic_gen_frames failed" ));
3775 :
3776 : /* did we add any frames? */
3777 :
3778 0 : if( payload_ptr==frame_start ) {
3779 : /* we have data to add, but none was added, presumably due
3780 : so space in the datagram */
3781 0 : ulong free_bytes = (ulong)( payload_end - payload_ptr );
3782 : /* sanity check */
3783 0 : if( free_bytes > 64 ) {
3784 : /* we should have been able to fit data into 64 bytes
3785 : so stop trying here */
3786 0 : break;
3787 0 : }
3788 :
3789 : /* try to free space */
3790 0 : fd_quic_tx_buffered( quic, conn );
3791 :
3792 : /* we have lots of space, so try again */
3793 0 : if( conn->tx_buf_conn == conn->tx_ptr ) {
3794 0 : enc_level = fd_quic_tx_enc_level( conn, 0 /* acks */ );
3795 0 : continue;
3796 0 : }
3797 0 : }
3798 :
3799 : /* first initial frame is padded to FD_QUIC_INITIAL_PAYLOAD_SZ_MIN
3800 : all short quic packets are padded so 16 bytes of sample are available */
3801 0 : uint tot_frame_sz = (uint)( payload_ptr - frame_start );
3802 0 : uint base_pkt_len = (uint)tot_frame_sz + pkt_num_len + FD_QUIC_CRYPTO_TAG_SZ;
3803 0 : uint padding = ( initial_pkt && (base_pkt_len < FD_QUIC_INITIAL_PAYLOAD_SZ_MIN) )
3804 0 : ? FD_QUIC_INITIAL_PAYLOAD_SZ_MIN - base_pkt_len : 0u;
3805 :
3806 0 : if( base_pkt_len + padding < FD_QUIC_CRYPTO_SAMPLE_OFFSET_FROM_PKT_NUM_START + FD_QUIC_CRYPTO_SAMPLE_SZ ) {
3807 0 : padding = FD_QUIC_CRYPTO_SAMPLE_SZ + FD_QUIC_CRYPTO_SAMPLE_OFFSET_FROM_PKT_NUM_START - base_pkt_len;
3808 0 : }
3809 :
3810 : /* this length includes the packet number length (pkt_number_len_enc+1),
3811 : padding and the final TAG */
3812 0 : uint quic_pkt_len = base_pkt_len + padding;
3813 :
3814 : /* set the length on the packet header */
3815 0 : uint quic_pkt_len_varint = 0x4000u | fd_uint_min( quic_pkt_len, 0x3fff );
3816 0 : FD_STORE( ushort, hdr_len_field, fd_ushort_bswap( (ushort)quic_pkt_len_varint ) );
3817 :
3818 : /* add padding */
3819 0 : if( padding ) {
3820 0 : fd_memset( payload_ptr, 0, padding );
3821 0 : payload_ptr += padding;
3822 0 : }
3823 :
3824 : /* everything successful up to here
3825 : encrypt into tx_ptr,tx_ptr+tx_sz */
3826 :
3827 : #if FD_QUIC_DISABLE_CRYPTO
3828 : ulong quic_pkt_sz = hdr_sz + tot_frame_sz + padding;
3829 : fd_memcpy( conn->tx_ptr, hdr_ptr, quic_pkt_sz );
3830 : conn->tx_ptr += quic_pkt_sz;
3831 :
3832 : /* append MAC tag */
3833 : memset( conn->tx_ptr, 0, FD_QUIC_CRYPTO_TAG_SZ );
3834 : conn->tx_ptr += FD_QUIC_CRYPTO_TAG_SZ;
3835 : #else
3836 0 : ulong cipher_text_sz = fd_quic_conn_tx_buf_remaining( conn );
3837 0 : ulong frames_sz = (ulong)( payload_ptr - frame_start ); /* including padding */
3838 :
3839 0 : fd_quic_crypto_keys_t * hp_keys = &conn->keys[enc_level][1];
3840 0 : fd_quic_crypto_keys_t * pkt_keys = key_phase_upd ? &conn->new_keys[1] : &conn->keys[enc_level][1];
3841 :
3842 0 : if( FD_UNLIKELY( fd_quic_crypto_encrypt( conn->tx_ptr, &cipher_text_sz, hdr_ptr, hdr_sz,
3843 0 : frame_start, frames_sz, pkt_keys, hp_keys, pkt_number ) != FD_QUIC_SUCCESS ) ) {
3844 0 : FD_LOG_WARNING(( "fd_quic_crypto_encrypt failed" ));
3845 :
3846 : /* this situation is unlikely to improve, so kill the connection */
3847 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_DEAD );
3848 0 : fd_quic_svc_prep_schedule_now( conn );
3849 0 : quic->metrics.conn_aborted_cnt++;
3850 0 : break;
3851 0 : }
3852 :
3853 0 : conn->tx_ptr += cipher_text_sz;
3854 0 : #endif
3855 :
3856 : /* we have committed the packet into the buffer, so inc pkt_number */
3857 0 : conn->pkt_number[pn_space]++;
3858 :
3859 0 : if( enc_level == fd_quic_enc_level_appdata_id ) {
3860 : /* short header must be last in datagram
3861 : so send in packet immediately */
3862 0 : fd_quic_tx_buffered( quic, conn );
3863 :
3864 0 : if( conn->tx_ptr == conn->tx_buf_conn ) {
3865 0 : enc_level = fd_quic_tx_enc_level( conn, 0 /* acks */ );
3866 0 : continue;
3867 0 : }
3868 :
3869 : /* TODO count here */
3870 :
3871 : /* drop packet */
3872 : /* this is a workaround for leaving a short=header-packet in the buffer
3873 : for the next tx_conn call. Next time around the tx_conn call will
3874 : not be aware that the buffer cannot be added to */
3875 0 : conn->tx_ptr = conn->tx_buf_conn;
3876 :
3877 0 : break;
3878 0 : }
3879 :
3880 : /* Refresh enc_level in case we can coalesce another packet */
3881 0 : enc_level = fd_quic_tx_enc_level( conn, 0 /* acks */ );
3882 0 : FD_DEBUG( if( enc_level!=~0u) FD_LOG_DEBUG(( "Attempting to append enc_level=%u packet", enc_level )); )
3883 0 : }
3884 :
3885 : /* try to send? */
3886 0 : fd_quic_tx_buffered( quic, conn );
3887 0 : }
3888 :
3889 : void
3890 0 : fd_quic_conn_service( fd_quic_t * quic, fd_quic_conn_t * conn, long now ) {
3891 :
3892 : /* Send new rtt measurement probe? */
3893 0 : if( FD_UNLIKELY( now > conn->last_ack + (long)conn->rtt_period_ns ) ) {
3894 : /* send PING */
3895 0 : if( !( conn->flags & ( FD_QUIC_CONN_FLAGS_PING | FD_QUIC_CONN_FLAGS_PING_SENT ) ) ) {
3896 0 : conn->flags |= FD_QUIC_CONN_FLAGS_PING;
3897 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING; /* update to be sent in next packet */
3898 0 : }
3899 0 : }
3900 :
3901 : /* handle expiry on pkt_meta */
3902 0 : fd_quic_pkt_meta_retry( quic, conn, 0 /* don't force */, ~0u /* all enc_levels */ );
3903 :
3904 : /* check state
3905 : need reset?
3906 : need close?
3907 : need acks?
3908 : replies?
3909 : data to send?
3910 : dead */
3911 0 : switch( conn->state ) {
3912 0 : case FD_QUIC_CONN_STATE_HANDSHAKE:
3913 0 : case FD_QUIC_CONN_STATE_HANDSHAKE_COMPLETE:
3914 0 : {
3915 0 : if( conn->tls_hs ) {
3916 : /* if we're the server, we send "handshake-done" frame */
3917 0 : if( conn->state == FD_QUIC_CONN_STATE_HANDSHAKE_COMPLETE && conn->server ) {
3918 0 : conn->handshake_done_send = 1;
3919 :
3920 : /* move straight to ACTIVE */
3921 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_ACTIVE );
3922 :
3923 : /* RFC 9001 4.9.2. Discarding Handshake Keys
3924 : > An endpoint MUST discard its Handshake keys when the
3925 : > TLS handshake is confirmed
3926 : RFC 9001 4.1.2. Handshake Confirmed
3927 : > [...] the TLS handshake is considered confirmed at the
3928 : > server when the handshake completes */
3929 0 : fd_quic_abandon_enc_level( conn, fd_quic_enc_level_handshake_id );
3930 :
3931 : /* user callback */
3932 0 : fd_quic_cb_conn_new( quic, conn );
3933 :
3934 : /* clear out hs_data here, as we don't need it anymore */
3935 0 : fd_quic_tls_clear_hs_data( conn->tls_hs, fd_quic_enc_level_appdata_id );
3936 0 : }
3937 :
3938 : /* if we're the client, fd_quic_conn_tx will flush the hs
3939 : buffer so we can receive the HANDSHAKE_DONE frame, and
3940 : transition from CONN_STATE HANDSHAKE_COMPLETE to ACTIVE. */
3941 0 : }
3942 :
3943 : /* do we have data to transmit? */
3944 0 : fd_quic_conn_tx( quic, conn );
3945 :
3946 0 : break;
3947 0 : }
3948 :
3949 0 : case FD_QUIC_CONN_STATE_CLOSE_PENDING:
3950 0 : case FD_QUIC_CONN_STATE_PEER_CLOSE:
3951 : /* user requested close, and may have set a reason code */
3952 : /* transmit the failure reason */
3953 0 : fd_quic_conn_tx( quic, conn );
3954 :
3955 : /* schedule another fd_quic_conn_service to free the conn */
3956 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_DEAD ); /* TODO need draining state wait for 3 * TPO */
3957 0 : quic->metrics.conn_closed_cnt++;
3958 :
3959 0 : break;
3960 :
3961 0 : case FD_QUIC_CONN_STATE_ABORT:
3962 : /* transmit the failure reason */
3963 0 : fd_quic_conn_tx( quic, conn );
3964 :
3965 : /* schedule another fd_quic_conn_service to free the conn */
3966 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_DEAD );
3967 0 : quic->metrics.conn_aborted_cnt++;
3968 :
3969 0 : break;
3970 :
3971 0 : case FD_QUIC_CONN_STATE_ACTIVE:
3972 : /* do we have data to transmit? */
3973 0 : fd_quic_conn_tx( quic, conn );
3974 :
3975 0 : break;
3976 :
3977 0 : case FD_QUIC_CONN_STATE_DEAD:
3978 0 : case FD_QUIC_CONN_STATE_INVALID:
3979 : /* fall thru */
3980 0 : default:
3981 0 : FD_LOG_CRIT(( "invalid conn state %u", conn->state ));
3982 0 : return;
3983 0 : }
3984 :
3985 : /* check routing and arp for this connection */
3986 :
3987 0 : }
3988 :
3989 : void
3990 : fd_quic_conn_free( fd_quic_t * quic,
3991 0 : fd_quic_conn_t * conn ) {
3992 0 : if( FD_UNLIKELY( !conn ) ) {
3993 0 : FD_LOG_WARNING(( "NULL conn" ));
3994 0 : return;
3995 0 : }
3996 0 : if( FD_UNLIKELY( conn->state == FD_QUIC_CONN_STATE_INVALID ) ) {
3997 0 : FD_LOG_CRIT(( "double free detected" ));
3998 0 : return;
3999 0 : }
4000 :
4001 0 : FD_COMPILER_MFENCE();
4002 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_INVALID );
4003 0 : FD_COMPILER_MFENCE();
4004 :
4005 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
4006 :
4007 : /* remove connection id from conn_map */
4008 :
4009 0 : fd_quic_conn_map_t * entry = fd_quic_conn_map_query( state->conn_map, conn->our_conn_id, NULL );
4010 0 : if( FD_LIKELY( entry ) ) fd_quic_conn_map_remove( state->conn_map, entry );
4011 :
4012 : /* no need to remove this connection from the events queue
4013 : free is called from two places:
4014 : fini - service will never be called again. All events are destroyed
4015 : service - removes event before calling free. Event only allowed to be
4016 : enqueued once */
4017 :
4018 : /* free pkt_meta */
4019 0 : for( uint j=0; j<=fd_quic_enc_level_appdata_id; ++j ) fd_quic_conn_free_pkt_meta( conn, j );
4020 :
4021 : /* remove all stream ids from map, and free stream */
4022 :
4023 : /* remove used streams */
4024 0 : fd_quic_stream_t * used_sentinel = conn->used_streams;
4025 0 : while( 1 ) {
4026 0 : fd_quic_stream_t * stream = used_sentinel->next;
4027 :
4028 0 : if( FD_UNLIKELY( stream == used_sentinel ) ) break;
4029 :
4030 0 : fd_quic_tx_stream_free( quic, conn, stream, FD_QUIC_STREAM_NOTIFY_CONN );
4031 0 : }
4032 :
4033 : /* remove send streams */
4034 0 : fd_quic_stream_t * send_sentinel = conn->send_streams;
4035 0 : while( 1 ) {
4036 0 : fd_quic_stream_t * stream = send_sentinel->next;
4037 :
4038 0 : if( FD_UNLIKELY( stream == send_sentinel ) ) break;
4039 :
4040 0 : fd_quic_tx_stream_free( quic, conn, stream, FD_QUIC_STREAM_NOTIFY_CONN );
4041 0 : }
4042 :
4043 : /* if any stream map entries are left over, remove them
4044 : this should not occur, so this branch should not execute
4045 : but if a stream doesn't get cleaned up properly, this fixes
4046 : the stream map */
4047 0 : if( FD_UNLIKELY( conn->stream_map && fd_quic_stream_map_key_cnt( conn->stream_map ) > 0 ) ) {
4048 0 : FD_LOG_WARNING(( "stream_map not empty. cnt: %lu",
4049 0 : (ulong)fd_quic_stream_map_key_cnt( conn->stream_map ) ));
4050 0 : while( fd_quic_stream_map_key_cnt( conn->stream_map ) > 0 ) {
4051 0 : int removed = 0;
4052 0 : for( ulong j = 0; j < fd_quic_stream_map_slot_cnt( conn->stream_map ); ++j ) {
4053 0 : if( conn->stream_map[j].stream_id != FD_QUIC_STREAM_ID_UNUSED ) {
4054 0 : fd_quic_stream_map_remove( conn->stream_map, &conn->stream_map[j] );
4055 0 : removed = 1;
4056 0 : j--; /* retry this entry */
4057 0 : }
4058 0 : }
4059 0 : if( !removed ) {
4060 0 : FD_LOG_WARNING(( "None removed. Remain: %lu",
4061 0 : (ulong)fd_quic_stream_map_key_cnt( conn->stream_map ) ));
4062 0 : break;
4063 0 : }
4064 0 : }
4065 0 : }
4066 :
4067 0 : if( conn->tls_hs ) {
4068 : /* free tls-hs */
4069 0 : fd_quic_tls_hs_delete( conn->tls_hs );
4070 :
4071 : /* Remove the handshake from the cache before releasing it */
4072 0 : fd_quic_tls_hs_cache_ele_remove( &state->hs_cache, conn->tls_hs, state->hs_pool);
4073 0 : fd_quic_tls_hs_pool_ele_release( state->hs_pool, conn->tls_hs );
4074 0 : }
4075 0 : conn->tls_hs = NULL;
4076 :
4077 : /* remove from service queue */
4078 0 : fd_quic_svc_timers_cancel( state->svc_timers, conn );
4079 :
4080 : /* put connection back in free list */
4081 0 : conn->free_conn_next = state->free_conn_list;
4082 0 : state->free_conn_list = conn->conn_idx;
4083 :
4084 0 : quic->metrics.conn_alloc_cnt--;
4085 :
4086 : /* clear keys */
4087 0 : memset( &conn->secrets, 0, sizeof(fd_quic_crypto_secrets_t) );
4088 0 : memset( conn->keys, 0, sizeof( conn->keys ) );
4089 0 : memset( conn->new_keys, 0, sizeof( conn->new_keys ) );
4090 0 : }
4091 :
4092 : fd_quic_conn_t *
4093 : fd_quic_connect( fd_quic_t * quic,
4094 : uint dst_ip_addr,
4095 : ushort dst_udp_port,
4096 : uint src_ip_addr,
4097 : ushort src_udp_port,
4098 0 : long now ) {
4099 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
4100 0 : state->now = now;
4101 :
4102 0 : if( FD_UNLIKELY( !fd_quic_tls_hs_pool_free( state->hs_pool ) ) ) {
4103 : /* try evicting, 0 if oldest is too young so fail */
4104 0 : if( !fd_quic_tls_hs_cache_evict( quic, state ) ) {
4105 0 : return NULL;
4106 0 : }
4107 0 : }
4108 :
4109 0 : fd_rng_t * rng = state->_rng;
4110 :
4111 : /* create conn ids for us and them
4112 : client creates connection id for the peer, peer immediately replaces it */
4113 0 : ulong our_conn_id_u64 = fd_rng_ulong( rng );
4114 0 : fd_quic_conn_id_t peer_conn_id; fd_quic_conn_id_rand( &peer_conn_id, rng );
4115 :
4116 0 : fd_quic_conn_t * conn = fd_quic_conn_create(
4117 0 : quic,
4118 0 : our_conn_id_u64,
4119 0 : &peer_conn_id,
4120 0 : dst_ip_addr,
4121 0 : dst_udp_port,
4122 0 : src_ip_addr,
4123 0 : src_udp_port,
4124 0 : 0 /* client */ );
4125 :
4126 0 : if( FD_UNLIKELY( !conn ) ) {
4127 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_conn_create failed" )) );
4128 0 : return NULL;
4129 0 : }
4130 :
4131 : /* Prepare QUIC-TLS transport params object (sent as a TLS extension).
4132 : Take template from state and mutate certain params in-place.
4133 :
4134 : See RFC 9000 Section 18 */
4135 :
4136 0 : fd_quic_transport_params_t tp[1] = { state->transport_params };
4137 :
4138 : /* The original_destination_connection_id is omitted by clients.
4139 : Since this is a mutable field, explicitly clear it here. */
4140 :
4141 0 : tp->original_destination_connection_id_present = 0;
4142 0 : tp->original_destination_connection_id_len = 0;
4143 :
4144 : /* Similarly, explicitly zero out retry fields. */
4145 0 : tp->retry_source_connection_id_present = 0;
4146 0 : tp->retry_source_connection_id_len = 0;
4147 :
4148 : /* Repeat source conn ID -- rationale see fd_quic_handle_v1_initial */
4149 :
4150 0 : FD_STORE( ulong, tp->initial_source_connection_id, conn->initial_source_conn_id );
4151 0 : tp->initial_source_connection_id_present = 1;
4152 0 : tp->initial_source_connection_id_len = FD_QUIC_CONN_ID_SZ;
4153 :
4154 : /* Create a TLS handshake (free>0 validated above) */
4155 :
4156 0 : fd_quic_tls_hs_t * tls_hs = fd_quic_tls_hs_new(
4157 0 : fd_quic_tls_hs_pool_ele_acquire( state->hs_pool ),
4158 0 : state->tls,
4159 0 : (void*)conn,
4160 0 : 0 /*is_server*/,
4161 0 : tp,
4162 0 : now );
4163 0 : if( FD_UNLIKELY( tls_hs->alert ) ) {
4164 0 : FD_LOG_WARNING(( "fd_quic_tls_hs_client_new failed" ));
4165 : /* free hs and the conn */
4166 0 : fd_quic_tls_hs_delete( tls_hs );
4167 0 : fd_quic_tls_hs_pool_ele_release( state->hs_pool, tls_hs );
4168 0 : fd_quic_conn_free( quic, conn );
4169 0 : return NULL;
4170 0 : }
4171 0 : fd_quic_tls_hs_cache_ele_push_tail( &state->hs_cache, tls_hs, state->hs_pool );
4172 :
4173 0 : quic->metrics.hs_created_cnt++;
4174 0 : conn->tls_hs = tls_hs;
4175 :
4176 0 : fd_quic_gen_initial_secret_and_keys( conn, &peer_conn_id, /* is_server */ 0 );
4177 :
4178 : /* set "called_conn_new" to indicate we should call conn_final
4179 : upon teardown */
4180 0 : conn->called_conn_new = 1;
4181 :
4182 0 : fd_quic_svc_prep_schedule( conn, now );
4183 0 : fd_quic_svc_timers_schedule( state->svc_timers, conn, now );
4184 :
4185 : /* everything initialized */
4186 0 : return conn;
4187 :
4188 0 : }
4189 :
4190 : fd_quic_conn_t *
4191 : fd_quic_conn_create( fd_quic_t * quic,
4192 : ulong our_conn_id,
4193 : fd_quic_conn_id_t const * peer_conn_id,
4194 : uint peer_ip_addr,
4195 : ushort peer_udp_port,
4196 : uint self_ip_addr,
4197 : ushort self_udp_port,
4198 0 : int server ) {
4199 0 : if( FD_UNLIKELY( !our_conn_id ) ) return NULL;
4200 :
4201 0 : fd_quic_config_t * config = &quic->config;
4202 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
4203 :
4204 : /* fetch top of connection free list */
4205 0 : uint conn_idx = state->free_conn_list;
4206 0 : if( FD_UNLIKELY( conn_idx==UINT_MAX ) ) {
4207 0 : FD_DEBUG( FD_LOG_DEBUG(( "fd_quic_conn_create failed: no free conn slots" )) );
4208 0 : quic->metrics.conn_err_no_slots_cnt++;
4209 0 : return NULL;
4210 0 : }
4211 0 : if( FD_UNLIKELY( conn_idx >= quic->limits.conn_cnt ) ) {
4212 0 : FD_LOG_CRIT(( "Conn free list corruption detected" ));
4213 0 : return NULL;
4214 0 : }
4215 0 : fd_quic_conn_t * conn = fd_quic_conn_at_idx( state, conn_idx );
4216 0 : if( FD_UNLIKELY( conn->state != FD_QUIC_CONN_STATE_INVALID ) ) {
4217 0 : FD_LOG_CRIT(( "conn %p not free, this is a bug", (void *)conn ));
4218 0 : return NULL;
4219 0 : }
4220 :
4221 : /* insert into conn map */
4222 0 : fd_quic_conn_map_t * insert_entry = fd_quic_conn_map_insert( state->conn_map, our_conn_id );
4223 :
4224 : /* if insert failed (should be impossible) fail, and do not remove connection
4225 : from free list */
4226 0 : if( FD_UNLIKELY( insert_entry == NULL ) ) {
4227 : /* FIXME This has ~1e-6 probability of happening with 10M conns
4228 : Retry generating our_conn_id instead of logging a warning */
4229 0 : FD_LOG_WARNING(( "fd_quic_conn_create failed: failed to register new conn ID %lu with map size %lu", our_conn_id, fd_quic_conn_map_key_cnt( state->conn_map ) ));
4230 0 : return NULL;
4231 0 : }
4232 :
4233 : /* set connection map insert_entry to new connection */
4234 0 : insert_entry->conn = conn;
4235 :
4236 : /* remove from free list */
4237 0 : state->free_conn_list = conn->free_conn_next;
4238 0 : conn->free_conn_next = UINT_MAX;
4239 :
4240 : /* if conn not marked free, skip */
4241 0 : if( FD_UNLIKELY( conn->state != FD_QUIC_CONN_STATE_INVALID ) ) {
4242 0 : FD_LOG_CRIT(( "conn %p not free, this is a bug", (void *)conn ));
4243 0 : return NULL;
4244 0 : }
4245 :
4246 : /* initialize connection members */
4247 0 : fd_quic_conn_clear( conn );
4248 :
4249 0 : conn->server = !!server;
4250 0 : conn->our_conn_id = our_conn_id;
4251 0 : conn->host = (fd_quic_net_endpoint_t){
4252 0 : .ip_addr = self_ip_addr, /* may be 0, if outgoing */
4253 0 : .udp_port = self_udp_port,
4254 0 : };
4255 0 : conn->conn_gen++;
4256 :
4257 :
4258 : /* pkt_meta */
4259 0 : fd_quic_pkt_meta_tracker_init( &conn->pkt_meta_tracker,
4260 0 : quic->limits.inflight_frame_cnt,
4261 0 : state->pkt_meta_pool );
4262 :
4263 : /* Initialize streams */
4264 0 : FD_QUIC_STREAM_LIST_SENTINEL( conn->send_streams );
4265 0 : FD_QUIC_STREAM_LIST_SENTINEL( conn->used_streams );
4266 :
4267 : /* initialize stream_id members */
4268 0 : fd_quic_conn_stream_rx_t * srx = conn->srx;
4269 0 : fd_quic_transport_params_t * our_tp = &state->transport_params;
4270 0 : srx->rx_hi_stream_id = server ? FD_QUIC_STREAM_TYPE_UNI_CLIENT : FD_QUIC_STREAM_TYPE_UNI_SERVER;
4271 0 : srx->rx_sup_stream_id = server ? FD_QUIC_STREAM_TYPE_UNI_CLIENT : FD_QUIC_STREAM_TYPE_UNI_SERVER;
4272 0 : conn->tx_next_stream_id = server ? FD_QUIC_STREAM_TYPE_UNI_SERVER : FD_QUIC_STREAM_TYPE_UNI_CLIENT;
4273 0 : conn->tx_sup_stream_id = server ? FD_QUIC_STREAM_TYPE_UNI_SERVER : FD_QUIC_STREAM_TYPE_UNI_CLIENT;
4274 :
4275 0 : srx->rx_max_data = our_tp->initial_max_data;
4276 :
4277 0 : if( state->transport_params.initial_max_streams_uni_present ) {
4278 0 : srx->rx_sup_stream_id = (state->transport_params.initial_max_streams_uni<<2) + FD_QUIC_STREAM_TYPE_UNI_CLIENT;
4279 0 : }
4280 0 : if( state->transport_params.initial_max_data ) {
4281 0 : srx->rx_max_data = state->transport_params.initial_max_data;
4282 0 : }
4283 :
4284 : /* points to free tx space */
4285 0 : conn->tx_ptr = conn->tx_buf_conn;
4286 :
4287 0 : conn->keys_avail = fd_uint_set_bit( 0U, fd_quic_enc_level_initial_id );
4288 :
4289 : /* Packet numbers left as 0
4290 : rfc9000: s12.3:
4291 : Packet numbers in each packet space start at 0.
4292 : Subsequent packets sent in the same packet number space
4293 : MUST increase the packet number by at least 1
4294 : rfc9002: s3
4295 : It is permitted for some packet numbers to never be used, leaving intentional gaps. */
4296 :
4297 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_HANDSHAKE );
4298 :
4299 : /* start with minimum supported max datagram */
4300 : /* peers may allow more */
4301 0 : conn->tx_max_datagram_sz = FD_QUIC_INITIAL_PAYLOAD_SZ_MAX;
4302 :
4303 : /* initial source connection id */
4304 0 : conn->initial_source_conn_id = our_conn_id;
4305 :
4306 : /* peer connection id */
4307 0 : conn->peer_cids[0] = *peer_conn_id;
4308 0 : conn->peer[0].ip_addr = peer_ip_addr;
4309 0 : conn->peer[0].udp_port = peer_udp_port;
4310 :
4311 0 : fd_quic_ack_gen_init( conn->ack_gen );
4312 :
4313 : /* initial rtt */
4314 : /* overridden when acks start returning */
4315 0 : fd_rtt_estimate_t * rtt = conn->rtt;
4316 :
4317 0 : ulong peer_ack_delay_exponent = 3UL; /* by spec, default is 3 */
4318 0 : conn->peer_ack_delay_scale = (float)( 1UL << peer_ack_delay_exponent ) * 1e3f;
4319 0 : conn->peer_max_ack_delay_ns = 0.0f; /* starts at zero, since peers respond immediately to */
4320 : /* INITIAL and HANDSHAKE */
4321 : /* updated when we get transport parameters */
4322 0 : rtt->smoothed_rtt = FD_QUIC_INITIAL_RTT_US * 1e3f;
4323 0 : rtt->latest_rtt = FD_QUIC_INITIAL_RTT_US * 1e3f;
4324 0 : rtt->min_rtt = FD_QUIC_INITIAL_RTT_US * 1e3f;
4325 0 : rtt->var_rtt = FD_QUIC_INITIAL_RTT_US * 1e3f * 0.5f;
4326 0 : conn->rtt_period_ns = FD_QUIC_RTT_PERIOD_US * 1e3f;
4327 :
4328 : /* idle timeout */
4329 0 : conn->idle_timeout_ns = config->idle_timeout;
4330 0 : conn->last_activity = state->now;
4331 0 : if( !conn->last_activity ) FD_LOG_CRIT(( "last activity: %ld", conn->last_activity ));
4332 0 : conn->let_die_time_ns = LONG_MAX;
4333 :
4334 : /* update metrics */
4335 0 : quic->metrics.conn_alloc_cnt++;
4336 0 : quic->metrics.conn_created_cnt++;
4337 :
4338 0 : fd_quic_svc_timers_init_conn( conn );
4339 :
4340 : /* prep idle timeout or keep alive at idle timeout/2 */
4341 0 : long delay = quic->config.idle_timeout>>(quic->config.keep_alive);
4342 0 : fd_quic_svc_prep_schedule( conn, state->now+delay );
4343 :
4344 : /* return connection */
4345 0 : return conn;
4346 0 : }
4347 :
4348 : long
4349 0 : fd_quic_get_next_wakeup( fd_quic_t * quic ) {
4350 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
4351 0 : long now = state->now;
4352 0 : fd_quic_svc_event_t next = fd_quic_svc_timers_next( state->svc_timers, now, 0 );
4353 0 : return next.timeout;
4354 0 : }
4355 :
4356 : /* frame handling function default definitions */
4357 : static ulong
4358 : fd_quic_handle_padding_frame(
4359 : fd_quic_frame_ctx_t * ctx FD_PARAM_UNUSED,
4360 : fd_quic_padding_frame_t * data FD_PARAM_UNUSED,
4361 : uchar const * const p0,
4362 0 : ulong p_sz ) {
4363 0 : uchar const * p = p0;
4364 0 : uchar const * const p_end = p + p_sz;
4365 0 : while( p<p_end && p[0]==0 ) p++;
4366 0 : return (ulong)( p - p0 );
4367 0 : }
4368 :
4369 : static ulong
4370 : fd_quic_handle_ping_frame(
4371 : fd_quic_frame_ctx_t * ctx,
4372 : fd_quic_ping_frame_t * data FD_PARAM_UNUSED,
4373 : uchar const * p0,
4374 0 : ulong p_sz ) {
4375 0 : FD_DTRACE_PROBE_1( quic_handle_ping_frame, ctx->conn->our_conn_id );
4376 : /* skip pings and pads */
4377 0 : uchar const * p = p0;
4378 0 : uchar const * const p_end = p + p_sz;
4379 0 : while( p < p_end && ((uint)p[0] & 0xfeu) == 0 ) p++;
4380 0 : return (ulong)( p - p0 );
4381 0 : }
4382 :
4383 : /* Retry packet metadata
4384 : This will force pkt_meta to be returned to the free list
4385 : for use. It does so by finding unack'ed packet metadata
4386 : and setting the data up for retransmission.
4387 : force_below_pkt_num will force retry of all frames
4388 : with pkt_num < force_below_pkt_num.
4389 : 'arg_enc_level' is the enc_level to retry, or can be
4390 : set to ~0u for all enc_levels.
4391 : */
4392 : void
4393 : fd_quic_pkt_meta_retry( fd_quic_t * quic,
4394 : fd_quic_conn_t * conn,
4395 : ulong force_below_pkt_num,
4396 0 : uint arg_enc_level ) {
4397 0 : fd_quic_conn_stream_rx_t * srx = conn->srx;
4398 :
4399 0 : long now = fd_quic_get_state( quic )->now;
4400 :
4401 0 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
4402 0 : fd_quic_pkt_meta_t * pool = tracker->pool;
4403 :
4404 : /* used for metric tracking */
4405 0 : ulong prev_retx_pkt_num[FD_QUIC_NUM_ENC_LEVELS] = { ~0ul, ~0ul, ~0ul, ~0ul };
4406 :
4407 0 : long const pto_duration = fd_quic_calc_expiry_duration( conn, 0, conn->server );
4408 0 : long const loss_duration = fd_quic_calc_expiry_duration( conn, 1, conn->server );
4409 :
4410 0 : while(1) {
4411 : /* find earliest expiring pkt_meta, over smallest pkt number at each enc_level */
4412 0 : fd_quic_pkt_meta_t * pkt_meta = NULL;
4413 0 : long expiry = LONG_MAX;
4414 0 : for( uint j=0u; j<4u; ++j ) {
4415 : /* if arg_enc_level set, only consider that enc_level */
4416 0 : if( !(arg_enc_level==~0u) & !(j==arg_enc_level) ) continue;
4417 :
4418 0 : fd_quic_pkt_meta_t * pkt_meta_cand = fd_quic_pkt_meta_min( &tracker->sent_pkt_metas[j], pool );
4419 0 : if( !pkt_meta_cand ) continue;
4420 :
4421 0 : uint const pn_space = fd_quic_enc_level_to_pn_space( j );
4422 0 : ulong const highest_acked = conn->highest_acked[pn_space];
4423 0 : long const cand_duration = fd_long_if( pkt_meta_cand->key.pkt_num < highest_acked, loss_duration, pto_duration );
4424 :
4425 0 : pkt_meta_cand->expiry = fd_long_min( pkt_meta_cand->tx_time + cand_duration, pkt_meta_cand->expiry );
4426 :
4427 0 : if( !pkt_meta || pkt_meta_cand->expiry < expiry ) {
4428 0 : pkt_meta = pkt_meta_cand;
4429 0 : expiry = pkt_meta_cand->expiry;
4430 0 : }
4431 0 : }
4432 :
4433 0 : if( !pkt_meta ) return;
4434 :
4435 0 : uint const enc_level = pkt_meta->enc_level;
4436 0 : ulong const pkt_num = pkt_meta->key.pkt_num;
4437 :
4438 : /* Continue until nothing expired nor to be skipped */
4439 0 : if( !!(pkt_num >= force_below_pkt_num) & !!(expiry > now) ) {
4440 : /* safe even when expiry is LONG_MAX, because prep_schedule takes min */
4441 0 : fd_quic_svc_prep_schedule( conn, expiry );
4442 0 : return;
4443 0 : };
4444 :
4445 0 : quic->metrics.pkt_retransmissions_cnt[enc_level] += !(pkt_meta->key.pkt_num == prev_retx_pkt_num[enc_level]);
4446 0 : prev_retx_pkt_num[enc_level] = pkt_num;
4447 :
4448 0 : uint type = pkt_meta->key.type;
4449 0 : FD_DTRACE_PROBE_4( quic_pkt_meta_retry, conn->our_conn_id, pkt_num, expiry, type);
4450 : /* set the data to retry */
4451 0 : switch( type ) {
4452 0 : case FD_QUIC_PKT_META_TYPE_HS_DATA:
4453 0 : do {
4454 0 : ulong offset = fd_ulong_max( conn->hs_ackd_bytes[enc_level], pkt_meta->val.range.offset_lo );
4455 0 : if( offset < conn->hs_sent_bytes[enc_level] ) {
4456 0 : conn->hs_sent_bytes[enc_level] = offset;
4457 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4458 0 : }
4459 0 : } while(0);
4460 0 : break;
4461 :
4462 0 : case FD_QUIC_PKT_META_TYPE_STREAM:
4463 0 : do {
4464 0 : ulong stream_id = pkt_meta->key.stream_id;
4465 :
4466 : /* find the stream */
4467 0 : fd_quic_stream_t * stream = NULL;
4468 0 : fd_quic_stream_map_t * stream_entry = fd_quic_stream_map_query( conn->stream_map, stream_id, NULL );
4469 0 : if( FD_LIKELY( stream_entry && stream_entry->stream &&
4470 0 : ( stream_entry->stream->stream_flags & FD_QUIC_STREAM_FLAGS_DEAD ) == 0 ) ) {
4471 0 : stream = stream_entry->stream;
4472 :
4473 : /* do not try sending data that has been acked */
4474 0 : ulong offset = fd_ulong_max( pkt_meta->val.range.offset_lo, stream->unacked_low );
4475 :
4476 : /* This pkt_meta may be stale: when ACK-driven loss detection
4477 : force-retries an earlier pkt_meta for the same stream, the
4478 : retransmitted data can be ACKed (advancing unacked_low)
4479 : before this pkt_meta expires. Skip the retry if the
4480 : stream has nothing left to send. */
4481 0 : if( FD_UNLIKELY( offset>=stream->tx_buf.head &&
4482 0 : !( stream->state & FD_QUIC_STREAM_STATE_TX_FIN ) ) ) {
4483 0 : break;
4484 0 : }
4485 :
4486 : /* any data left to retry? */
4487 0 : stream->tx_sent = fd_ulong_min( stream->tx_sent, offset );
4488 :
4489 : /* We must have something to send if the stream was found */
4490 0 : if( !( (stream->tx_sent < stream->tx_buf.head) | (stream->state & FD_QUIC_STREAM_STATE_TX_FIN) ) ) {
4491 0 : FD_LOG_CRIT(( "unexpected retry for completed stream %lu", stream_id ));
4492 0 : }
4493 :
4494 : /* insert into send list */
4495 0 : FD_QUIC_STREAM_LIST_REMOVE( stream );
4496 0 : FD_QUIC_STREAM_LIST_INSERT_BEFORE( conn->send_streams, stream );
4497 :
4498 : /* set the data to go out on the next packet */
4499 0 : stream->stream_flags |= FD_QUIC_STREAM_FLAGS_UNSENT; /* we have unsent data */
4500 0 : stream->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4501 0 : }
4502 0 : } while(0);
4503 0 : break;
4504 :
4505 0 : case FD_QUIC_PKT_META_TYPE_HS_DONE:
4506 0 : if( FD_LIKELY( !conn->handshake_done_ackd ) ) {
4507 0 : conn->handshake_done_send = 1;
4508 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4509 0 : }
4510 0 : break;
4511 :
4512 0 : case FD_QUIC_PKT_META_TYPE_MAX_DATA:
4513 0 : if( srx->rx_max_data_ackd < srx->rx_max_data ) {
4514 0 : conn->flags |= FD_QUIC_CONN_FLAGS_MAX_DATA;
4515 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4516 0 : }
4517 0 : break;
4518 :
4519 0 : case FD_QUIC_PKT_META_TYPE_MAX_STREAMS_UNIDIR:
4520 0 : do {
4521 : /* do we still need to send? */
4522 : /* get required value */
4523 0 : ulong max_streams_unidir = srx->rx_sup_stream_id >> 2;
4524 :
4525 0 : if( max_streams_unidir > srx->rx_max_streams_unidir_ackd ) {
4526 : /* set the data to go out on the next packet */
4527 0 : conn->flags |= FD_QUIC_CONN_FLAGS_MAX_STREAMS_UNIDIR;
4528 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4529 0 : }
4530 0 : } while(0);
4531 0 : break;
4532 :
4533 0 : case FD_QUIC_PKT_META_TYPE_CLOSE:
4534 0 : conn->flags &= ~FD_QUIC_CONN_FLAGS_CLOSE_SENT;
4535 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4536 0 : break;
4537 :
4538 0 : case FD_QUIC_PKT_META_TYPE_PING:
4539 0 : conn->flags = ( conn->flags & ~FD_QUIC_CONN_FLAGS_PING_SENT )
4540 0 : | FD_QUIC_CONN_FLAGS_PING;
4541 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4542 0 : break;
4543 0 : }
4544 :
4545 : /* reschedule to ensure the data gets processed */
4546 0 : fd_quic_svc_prep_schedule_now( conn );
4547 :
4548 0 : fd_quic_pkt_meta_remove( &tracker->sent_pkt_metas[enc_level], pool, pkt_meta );
4549 0 : conn->used_pkt_meta--;
4550 0 : }
4551 0 : }
4552 :
4553 : /* reclaim resources associated with packet metadata
4554 : this is called in response to received acks */
4555 : void
4556 : fd_quic_reclaim_pkt_meta( fd_quic_conn_t * conn,
4557 : fd_quic_pkt_meta_t * pkt_meta,
4558 0 : uint enc_level ) {
4559 0 : fd_quic_conn_stream_rx_t * srx = conn->srx;
4560 :
4561 0 : uint type = pkt_meta->key.type;
4562 0 : fd_quic_range_t range = pkt_meta->val.range;
4563 :
4564 0 : switch( type ) {
4565 :
4566 0 : case FD_QUIC_PKT_META_TYPE_PING:
4567 0 : do {
4568 0 : conn->flags &= ~( FD_QUIC_CONN_FLAGS_PING | FD_QUIC_CONN_FLAGS_PING_SENT );
4569 0 : } while(0);
4570 0 : break;
4571 :
4572 0 : case FD_QUIC_PKT_META_TYPE_HS_DATA:
4573 0 : do {
4574 : /* Note that tls_hs could already be freed */
4575 : /* is this ack'ing the next consecutive bytes?
4576 : if so, we can increase the ack'd bytes
4577 : if not, we retransmit the bytes expected to be ack'd
4578 : we assume a gap means a dropped packet, and
4579 : this policy allows us to free up the pkt_meta here */
4580 0 : ulong hs_ackd_bytes = conn->hs_ackd_bytes[enc_level];
4581 0 : if( range.offset_lo <= hs_ackd_bytes ) {
4582 0 : hs_ackd_bytes = conn->hs_ackd_bytes[enc_level]
4583 0 : = fd_ulong_max( hs_ackd_bytes, range.offset_hi );
4584 :
4585 : /* remove any unused hs_data */
4586 0 : fd_quic_tls_hs_data_t * hs_data = NULL;
4587 :
4588 0 : hs_data = fd_quic_tls_get_hs_data( conn->tls_hs, enc_level );
4589 0 : while( hs_data && hs_data->offset + hs_data->data_sz <= hs_ackd_bytes ) {
4590 0 : fd_quic_tls_pop_hs_data( conn->tls_hs, enc_level );
4591 0 : hs_data = fd_quic_tls_get_hs_data( conn->tls_hs, enc_level );
4592 0 : }
4593 0 : } else {
4594 0 : conn->hs_sent_bytes[enc_level] =
4595 0 : fd_ulong_min( conn->hs_sent_bytes[enc_level], hs_ackd_bytes );
4596 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
4597 0 : }
4598 0 : } while(0);
4599 0 : break;
4600 :
4601 0 : case FD_QUIC_PKT_META_TYPE_HS_DONE:
4602 0 : do {
4603 0 : conn->handshake_done_ackd = 1;
4604 0 : conn->handshake_done_send = 0;
4605 0 : if( FD_LIKELY( conn->tls_hs ) ) {
4606 0 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
4607 0 : fd_quic_tls_hs_delete( conn->tls_hs );
4608 0 : fd_quic_tls_hs_cache_ele_remove( &state->hs_cache, conn->tls_hs, state->hs_pool );
4609 0 : fd_quic_tls_hs_pool_ele_release( state->hs_pool, conn->tls_hs );
4610 0 : conn->tls_hs = NULL;
4611 0 : }
4612 0 : } while(0);
4613 0 : break;
4614 :
4615 0 : case FD_QUIC_PKT_META_TYPE_MAX_DATA:
4616 0 : do {
4617 0 : ulong max_data_ackd = pkt_meta->val.scalar;
4618 :
4619 : /* ack can only increase max_data_ackd */
4620 0 : max_data_ackd = fd_ulong_max( max_data_ackd, srx->rx_max_data_ackd );
4621 :
4622 : /* max_data_ackd > rx_max_data is a protocol violation */
4623 0 : if( FD_UNLIKELY( max_data_ackd > srx->rx_max_data ) ) {
4624 : /* this is a protocol violation, so inform the peer */
4625 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
4626 0 : return;
4627 0 : }
4628 :
4629 : /* clear flag only if acked value == current value */
4630 0 : if( FD_LIKELY( max_data_ackd == srx->rx_max_data ) ) {
4631 0 : conn->flags &= ~FD_QUIC_CONN_FLAGS_MAX_DATA;
4632 0 : }
4633 :
4634 : /* set the ackd value */
4635 0 : srx->rx_max_data_ackd = max_data_ackd;
4636 0 : } while(0);
4637 0 : break;
4638 :
4639 0 : case FD_QUIC_PKT_META_TYPE_MAX_STREAMS_UNIDIR:
4640 0 : do {
4641 0 : ulong max_streams_unidir_ackd = pkt_meta->val.scalar;
4642 :
4643 : /* ack can only increase max_streams_unidir_ackd */
4644 0 : max_streams_unidir_ackd = fd_ulong_max( max_streams_unidir_ackd, srx->rx_max_streams_unidir_ackd );
4645 :
4646 : /* get required value */
4647 0 : ulong max_streams_unidir = srx->rx_sup_stream_id >> 2;
4648 :
4649 : /* clear flag only if acked value == current value */
4650 0 : if( FD_LIKELY( max_streams_unidir_ackd == max_streams_unidir ) ) {
4651 0 : conn->flags &= ~FD_QUIC_CONN_FLAGS_MAX_STREAMS_UNIDIR;
4652 0 : }
4653 :
4654 : /* set the ackd value */
4655 0 : srx->rx_max_streams_unidir_ackd = max_streams_unidir_ackd;
4656 0 : } while(0);
4657 0 : break;
4658 :
4659 0 : case FD_QUIC_PKT_META_TYPE_STREAM:
4660 0 : do {
4661 0 : ulong stream_id = pkt_meta->key.stream_id;
4662 0 : fd_quic_range_t range = pkt_meta->val.range;
4663 :
4664 : /* find the stream */
4665 0 : fd_quic_stream_t * stream = NULL;
4666 0 : fd_quic_stream_map_t * stream_entry = fd_quic_stream_map_query( conn->stream_map, stream_id, NULL );
4667 0 : if( FD_LIKELY( stream_entry && stream_entry->stream &&
4668 0 : ( stream_entry->stream->stream_flags & FD_QUIC_STREAM_FLAGS_DEAD ) == 0 ) ) {
4669 0 : stream = stream_entry->stream;
4670 :
4671 0 : ulong tx_tail = stream->unacked_low;
4672 0 : ulong tx_sent = stream->tx_sent;
4673 :
4674 : /* ignore bytes which were already acked */
4675 0 : range.offset_lo = fd_ulong_max( range.offset_lo, tx_tail );
4676 :
4677 : /* verify offset_hi */
4678 0 : if( FD_UNLIKELY( range.offset_hi > stream->tx_buf.head ) ) {
4679 : /* offset_hi in the pkt_meta (the highest byte offset in the packet */
4680 : /* should never exceed tx_buf.head - the highest byte offset in the */
4681 : /* stream */
4682 0 : fd_quic_conn_error( conn, FD_QUIC_CONN_REASON_INTERNAL_ERROR, __LINE__ );
4683 0 : return;
4684 0 : }
4685 :
4686 0 : uchar * tx_ack = stream->tx_ack;
4687 : /* did they ack the first unacked byte? */
4688 0 : if( FD_LIKELY( range.offset_lo == tx_tail ) ) {
4689 : /* move tail to first unacked byte */
4690 0 : tx_tail = range.offset_hi;
4691 0 : while( tx_tail < tx_sent ) {
4692 : /* TODO - optimize this for larger strides */
4693 : /* can we skip a whole byte? */
4694 0 : if( ( tx_tail & 7ul ) == 0ul && tx_tail + 8ul <= tx_sent && tx_ack[tx_tail>>3ul] == 0xffu ) {
4695 0 : tx_tail += 8ul;
4696 0 : } else {
4697 0 : if( tx_ack[tx_tail>>3ul] & ( 1u << ( tx_tail & 7u ) ) ) {
4698 0 : tx_tail++;
4699 0 : } else {
4700 0 : break;
4701 0 : }
4702 0 : }
4703 0 : }
4704 0 : stream->unacked_low = tx_tail;
4705 0 : } else {
4706 : /* mark this range as acked in tx_ack, so we can skip it later */
4707 : /* TODO optimize this for larger strides */
4708 0 : for( ulong k = range.offset_lo; k < range.offset_hi; ) {
4709 0 : if( ( k & 7ul ) == 0ul && k + 8ul <= range.offset_hi ) {
4710 : /* set whole byte */
4711 0 : tx_ack[k>>3ul] = 0xffu;
4712 0 : k += 8ul;
4713 0 : } else {
4714 : /* compiler is not smart enough to know ( 1u << ( k & 7u ) ) fits in a uchar */
4715 0 : tx_ack[k>>3ul] |= (uchar)( 1ul << ( k & 7ul ) );
4716 0 : k++;
4717 0 : }
4718 0 : }
4719 0 : }
4720 :
4721 : /* For convenience */
4722 0 : uint fin_state_mask = FD_QUIC_STREAM_STATE_TX_FIN | FD_QUIC_STREAM_STATE_RX_FIN;
4723 :
4724 : /* Free stream if it's done:
4725 : 1. peer has acked every data byte user has tried to send
4726 : 2. peer has acked the fin --> user has fin'd */
4727 0 : if( tx_tail == stream->tx_buf.head &&
4728 0 : ( stream->state & fin_state_mask ) == fin_state_mask ) {
4729 : /* fd_quic_tx_stream_free also notifies the user */
4730 0 : fd_quic_tx_stream_free( conn->quic, conn, stream, FD_QUIC_STREAM_NOTIFY_END );
4731 0 : }
4732 0 : }
4733 0 : } while(0);
4734 0 : break;
4735 0 : }
4736 0 : }
4737 :
4738 : /* process ack range
4739 : applies to pkt_number in [largest_ack - ack_range, largest_ack] */
4740 : void
4741 : fd_quic_process_ack_range( fd_quic_conn_t * conn,
4742 : fd_quic_frame_ctx_t * context,
4743 : uint enc_level,
4744 : ulong largest_ack,
4745 : ulong ack_range,
4746 : int is_largest,
4747 : long now,
4748 0 : ulong ack_delay ) {
4749 : /* FIXME: Close connection if peer ACKed a higher packet number than we sent */
4750 :
4751 0 : fd_quic_pkt_t * pkt = context->pkt;
4752 :
4753 : /* inclusive range */
4754 0 : ulong hi = largest_ack;
4755 0 : ulong lo = largest_ack - ack_range;
4756 0 : FD_DTRACE_PROBE_4( quic_process_ack_range, conn->our_conn_id, enc_level, lo, hi );
4757 :
4758 0 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
4759 0 : fd_quic_pkt_meta_t * pool = tracker->pool;
4760 0 : fd_quic_pkt_meta_ds_t * sent = &tracker->sent_pkt_metas[enc_level];
4761 :
4762 : /* start at oldest sent */
4763 0 : for( fd_quic_pkt_meta_ds_fwd_iter_t iter = fd_quic_pkt_meta_ds_idx_ge( sent, lo, pool );
4764 0 : !fd_quic_pkt_meta_ds_fwd_iter_done( iter );
4765 0 : iter = fd_quic_pkt_meta_ds_fwd_iter_next( iter, pool ) ) {
4766 0 : fd_quic_pkt_meta_t * e = fd_quic_pkt_meta_ds_fwd_iter_ele( iter, pool );
4767 0 : if( FD_UNLIKELY( e->key.pkt_num > hi ) ) break;
4768 0 : if( is_largest && e->key.pkt_num == hi && hi >= pkt->rtt_pkt_number ) {
4769 0 : pkt->rtt_pkt_number = hi;
4770 0 : pkt->rtt_ack_time = now - e->tx_time; /* in ns */
4771 0 : pkt->rtt_ack_delay = ack_delay; /* in peer units */
4772 0 : }
4773 0 : fd_quic_reclaim_pkt_meta( conn, e, enc_level );
4774 0 : }
4775 :
4776 0 : conn->used_pkt_meta -= fd_quic_pkt_meta_remove_range( sent, pool, lo, hi );
4777 0 : }
4778 :
4779 : static ulong
4780 : fd_quic_handle_ack_frame( fd_quic_frame_ctx_t * context,
4781 : fd_quic_ack_frame_t * data,
4782 : uchar const * p,
4783 0 : ulong p_sz ) {
4784 0 : fd_quic_conn_t * conn = context->conn;
4785 0 : uint enc_level = context->pkt->enc_level;
4786 0 : uint pn_space = fd_quic_enc_level_to_pn_space( enc_level );
4787 0 : ulong const largest_ack = data->largest_ack;
4788 :
4789 0 : if( FD_UNLIKELY( data->first_ack_range > largest_ack ) ) {
4790 : /* this is a protocol violation, so inform the peer */
4791 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
4792 0 : return FD_QUIC_PARSE_FAIL;
4793 0 : }
4794 :
4795 : /* update highest_acked */
4796 0 : conn->highest_acked[pn_space] = fd_ulong_max( conn->highest_acked[pn_space], largest_ack );
4797 :
4798 0 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
4799 0 : long const now = state->now;
4800 0 : /******/ conn->last_ack = now;
4801 :
4802 : /* RFC 9002, Section 6.1: We declare a packet p lost if either:
4803 : 1. It was 'skipped': at least three packets were sent after p but before
4804 : the highest ack'ed packet in this ack frame, e.g. (p x y z highest_acked).
4805 : 2. It 'expired': p was sent at least time-threshold time ago.
4806 : Time-threshold is computed in calc_expiry, and typically differs from
4807 : pkt_meta->expiry.
4808 :
4809 : Let skip_ceil be the smallest pkt_num such that any unacked pkt_meta
4810 : with pkt_num < skip_ceil is 'skipped' as defined above. We compute this
4811 : before removing acked packets from the tracker.
4812 :
4813 : We unfortunately can't just use 'largest_ack-3' because 1) largest_ack'd
4814 : may have been previously acknowledged and 2) we may have skipped pkt_nums.
4815 : */
4816 0 : ulong skip_ceil = 0UL;
4817 0 : if( FD_LIKELY( largest_ack > 0UL ) ) {
4818 0 : #define FD_QUIC_K_PACKET_THRESHOLD 3
4819 0 : fd_quic_pkt_meta_tracker_t * tracker = &conn->pkt_meta_tracker;
4820 0 : fd_quic_pkt_meta_t * pool = tracker->pool;
4821 0 : fd_quic_pkt_meta_ds_t * sent = &tracker->sent_pkt_metas[enc_level];
4822 :
4823 0 : uint pkt_num_changes = 0;
4824 0 : ulong prev_pkt_num = ~0UL;
4825 0 : fd_quic_pkt_meta_ds_fwd_iter_t start = fd_quic_pkt_meta_ds_idx_le( sent, pool, largest_ack-1 );
4826 0 : for( fd_quic_pkt_meta_ds_rev_iter_t iter = start; /* rev<>fwd iter */
4827 0 : !!(pkt_num_changes<FD_QUIC_K_PACKET_THRESHOLD) & !fd_quic_pkt_meta_ds_rev_iter_done(iter);
4828 0 : iter=fd_quic_pkt_meta_ds_rev_iter_next( iter, pool ) ) {
4829 0 : fd_quic_pkt_meta_t * e = fd_quic_pkt_meta_ds_rev_iter_ele( iter, pool );
4830 0 : pkt_num_changes += !(e->key.pkt_num==prev_pkt_num);
4831 0 : prev_pkt_num = e->key.pkt_num;
4832 0 : }
4833 0 : skip_ceil = fd_ulong_if( pkt_num_changes==FD_QUIC_K_PACKET_THRESHOLD, prev_pkt_num, 0UL );
4834 0 : }
4835 :
4836 : /* track lowest packet acked */
4837 0 : ulong low_ack_pkt_number = largest_ack - data->first_ack_range;
4838 :
4839 : /* process ack range
4840 : applies to pkt_number in [largest_ack - first_ack_range, largest_ack] */
4841 0 : fd_quic_process_ack_range( conn,
4842 0 : context,
4843 0 : enc_level,
4844 0 : largest_ack,
4845 0 : data->first_ack_range,
4846 0 : 1 /* is_largest */,
4847 0 : now,
4848 0 : data->ack_delay );
4849 :
4850 0 : uchar const * p_str = p;
4851 0 : uchar const * p_end = p + p_sz;
4852 :
4853 0 : ulong ack_range_count = data->ack_range_count;
4854 :
4855 : /* cur_pkt_number holds the packet number of the lowest processed
4856 : and acknowledged packet
4857 : This should always be a valid packet number >= 0 */
4858 0 : ulong cur_pkt_number = largest_ack - data->first_ack_range;
4859 :
4860 : /* walk thru ack ranges */
4861 0 : for( ulong j = 0UL; j < ack_range_count; ++j ) {
4862 0 : if( FD_UNLIKELY( p_end <= p ) ) {
4863 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
4864 0 : return FD_QUIC_PARSE_FAIL;
4865 0 : }
4866 :
4867 0 : fd_quic_ack_range_frag_t ack_range[1];
4868 0 : ulong rc = fd_quic_decode_ack_range_frag( ack_range, p, (ulong)( p_end - p ) );
4869 0 : if( FD_UNLIKELY( rc == FD_QUIC_PARSE_FAIL ) ) {
4870 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
4871 0 : return FD_QUIC_PARSE_FAIL;
4872 0 : }
4873 :
4874 : /* ensure we have ulong local vars, regardless of ack_range definition */
4875 0 : ulong gap = (ulong)ack_range->gap;
4876 0 : ulong length = (ulong)ack_range->length;
4877 :
4878 : /* sanity check before unsigned arithmetic */
4879 0 : if( FD_UNLIKELY( ( gap > ( ~0x3UL ) ) |
4880 0 : ( length > ( ~0x3UL ) ) ) ) {
4881 : /* This is an unreasonably large value, so fail with protocol violation
4882 : It's also likely impossible due to the encoding method */
4883 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
4884 0 : return FD_QUIC_PARSE_FAIL;
4885 0 : }
4886 :
4887 : /* The number of packet numbers to skip (they are not being acked) is
4888 : ack_range->gap + 2
4889 : This is +1 to get from the lowest acked packet to the highest unacked packet
4890 : and +1 because the count of packets in the gap is (ack_range->gap+1) */
4891 0 : ulong skip = gap + 2UL;
4892 :
4893 : /* verify the skip and length values are valid */
4894 0 : if( FD_UNLIKELY( skip + length > cur_pkt_number ) ) {
4895 : /* this is a protocol violation, so inform the peer */
4896 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
4897 0 : return FD_QUIC_PARSE_FAIL;
4898 0 : }
4899 :
4900 : /* track lowest */
4901 0 : ulong lo_pkt_number = cur_pkt_number - skip - length;
4902 0 : low_ack_pkt_number = fd_ulong_min( low_ack_pkt_number, lo_pkt_number );
4903 :
4904 : /* process ack range */
4905 0 : fd_quic_process_ack_range( conn,
4906 0 : context,
4907 0 : enc_level,
4908 0 : cur_pkt_number - skip,
4909 0 : length,
4910 0 : 0 /* is_largest */,
4911 0 : now,
4912 0 : 0 /* ack_delay not used here */ );
4913 :
4914 : /* Find the next lowest processed and acknowledged packet number
4915 : This should get us to the next lowest processed and acknowledged packet
4916 : number */
4917 0 : cur_pkt_number -= skip + length;
4918 :
4919 0 : p += rc;
4920 0 : }
4921 :
4922 : /* Process packets declared lost for either:
4923 : 1. 'skipped': see skip_ceil computation above
4924 : 2. 'expired': checked inside pkt_meta_retry */
4925 0 : fd_quic_pkt_meta_retry( conn->quic, conn, skip_ceil, enc_level );
4926 :
4927 : /* ECN counts
4928 : we currently ignore them, but we must process them to get to the following bytes */
4929 0 : if( data->type & 1U ) {
4930 0 : if( FD_UNLIKELY( p_end <= p ) ) {
4931 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
4932 0 : return FD_QUIC_PARSE_FAIL;
4933 0 : }
4934 :
4935 0 : fd_quic_ecn_counts_frag_t ecn_counts[1];
4936 0 : ulong rc = fd_quic_decode_ecn_counts_frag( ecn_counts, p, (ulong)( p_end - p ) );
4937 0 : if( rc == FD_QUIC_PARSE_FAIL ) {
4938 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
4939 0 : return FD_QUIC_PARSE_FAIL;
4940 0 : }
4941 :
4942 0 : p += rc;
4943 0 : }
4944 :
4945 0 : return (ulong)( p - p_str );
4946 0 : }
4947 :
4948 : static ulong
4949 : fd_quic_handle_reset_stream_frame(
4950 : fd_quic_frame_ctx_t * context,
4951 : fd_quic_reset_stream_frame_t * data,
4952 : uchar const * p FD_PARAM_UNUSED,
4953 0 : ulong p_sz FD_PARAM_UNUSED ) {
4954 : /* TODO implement */
4955 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 );
4956 0 : return 0UL;
4957 0 : }
4958 :
4959 : static ulong
4960 : fd_quic_handle_stop_sending_frame(
4961 : fd_quic_frame_ctx_t * context,
4962 : fd_quic_stop_sending_frame_t * data,
4963 : uchar const * p FD_PARAM_UNUSED,
4964 0 : ulong p_sz FD_PARAM_UNUSED ) {
4965 0 : FD_DTRACE_PROBE_3( quic_handle_stop_sending_frame, context->conn->our_conn_id, data->stream_id, data->app_proto_err_code );
4966 0 : return 0UL;
4967 0 : }
4968 :
4969 : static ulong
4970 : fd_quic_handle_new_token_frame(
4971 : fd_quic_frame_ctx_t * context,
4972 : fd_quic_new_token_frame_t * data,
4973 : uchar const * p FD_PARAM_UNUSED,
4974 0 : ulong p_sz FD_PARAM_UNUSED ) {
4975 : /* FIXME A server MUST treat receipt of a NEW_TOKEN frame as a connection error of type PROTOCOL_VIOLATION. */
4976 0 : (void)data;
4977 0 : FD_DTRACE_PROBE_1( quic_handle_new_token_frame, context->conn->our_conn_id );
4978 0 : return 0UL;
4979 0 : }
4980 :
4981 : void
4982 : fd_quic_tx_stream_free( fd_quic_t * quic,
4983 : fd_quic_conn_t * conn,
4984 : fd_quic_stream_t * stream,
4985 0 : int code ) {
4986 :
4987 : /* TODO rename FD_QUIC_NOTIFY_END to FD_QUIC_STREAM_NOTIFY_END et al */
4988 0 : if( FD_LIKELY( stream->state != FD_QUIC_STREAM_STATE_DEAD ) ) {
4989 0 : fd_quic_cb_stream_notify( quic, stream, stream->context, code );
4990 0 : stream->state = FD_QUIC_STREAM_STATE_DEAD;
4991 0 : }
4992 :
4993 0 : ulong stream_id = stream->stream_id;
4994 :
4995 : /* remove from stream map */
4996 0 : fd_quic_stream_map_t * stream_map = conn->stream_map;
4997 0 : fd_quic_stream_map_t * stream_entry = fd_quic_stream_map_query( stream_map, stream_id, NULL );
4998 0 : if( FD_LIKELY( stream_entry ) ) {
4999 0 : if( FD_LIKELY( stream_entry->stream ) ) {
5000 0 : stream_entry->stream->stream_flags = FD_QUIC_STREAM_FLAGS_DEAD;
5001 0 : }
5002 0 : fd_quic_stream_map_remove( stream_map, stream_entry );
5003 0 : }
5004 :
5005 : /* remove from list - idempotent */
5006 0 : FD_QUIC_STREAM_LIST_REMOVE( stream );
5007 0 : stream->stream_flags = FD_QUIC_STREAM_FLAGS_DEAD;
5008 0 : stream->stream_id = ~0UL;
5009 :
5010 : /* add to stream_pool */
5011 0 : fd_quic_state_t * state = fd_quic_get_state( quic );
5012 0 : fd_quic_stream_pool_free( state->stream_pool, stream );
5013 :
5014 0 : }
5015 :
5016 :
5017 : static inline __attribute__((always_inline)) ulong
5018 : fd_quic_handle_stream_frame(
5019 : fd_quic_frame_ctx_t * context,
5020 : uchar const * p,
5021 : ulong p_sz,
5022 : ulong stream_id,
5023 : ulong offset,
5024 : ulong data_sz,
5025 0 : int fin ) {
5026 0 : fd_quic_t * quic = context->quic;
5027 0 : fd_quic_conn_t * conn = context->conn;
5028 0 : fd_quic_pkt_t * pkt = context->pkt;
5029 :
5030 0 : FD_DTRACE_PROBE_5( quic_handle_stream_frame, conn->our_conn_id, stream_id, offset, data_sz, fin );
5031 :
5032 : /* stream_id type check */
5033 0 : ulong stream_type = stream_id & 3UL;
5034 0 : if( FD_UNLIKELY( stream_type != ( conn->server ? FD_QUIC_STREAM_TYPE_UNI_CLIENT : FD_QUIC_STREAM_TYPE_UNI_SERVER ) ) ) {
5035 0 : FD_DEBUG( FD_LOG_DEBUG(( "Received forbidden stream type" )); )
5036 : /* Technically should switch between STREAM_LIMIT_ERROR and STREAM_STATE_ERROR here */
5037 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_STREAM_LIMIT_ERROR, __LINE__ );
5038 0 : return FD_QUIC_PARSE_FAIL;
5039 0 : }
5040 :
5041 : /* length check */
5042 0 : if( FD_UNLIKELY( data_sz > p_sz ) ) {
5043 0 : FD_DEBUG( FD_LOG_DEBUG(( "Stream header indicates %lu bytes length, but only have %lu", data_sz, p_sz )); )
5044 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5045 0 : return FD_QUIC_PARSE_FAIL;
5046 0 : }
5047 :
5048 0 : conn->unacked_sz += data_sz;
5049 :
5050 : /* stream_id outside allowed range - protocol error */
5051 0 : if( FD_UNLIKELY( stream_id >= conn->srx->rx_sup_stream_id ) ) {
5052 0 : FD_DEBUG( FD_LOG_DEBUG(( "Stream ID violation detected" )); )
5053 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_STREAM_LIMIT_ERROR, __LINE__ );
5054 0 : return FD_QUIC_PARSE_FAIL;
5055 0 : }
5056 :
5057 : /* A receiver MUST close the connection with an error of type FLOW_CONTROL_ERROR if the sender
5058 : violates the advertised connection or stream data limits */
5059 0 : if( FD_UNLIKELY( quic->config.initial_rx_max_stream_data < offset + data_sz ) ) {
5060 0 : FD_DEBUG( FD_LOG_DEBUG(( "Stream data limit exceeded" )); )
5061 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FLOW_CONTROL_ERROR, __LINE__ );
5062 0 : return FD_QUIC_PARSE_FAIL;
5063 0 : }
5064 :
5065 0 : int rx_res = fd_quic_cb_stream_rx( quic, conn, stream_id, offset, p, data_sz, fin );
5066 0 : pkt->ack_flag |= fd_uint_if( rx_res==FD_QUIC_SUCCESS, 0U, ACK_FLAG_CANCEL );
5067 :
5068 : /* packet bytes consumed */
5069 0 : return data_sz;
5070 0 : }
5071 :
5072 : static ulong
5073 : fd_quic_handle_stream_8_frame(
5074 : fd_quic_frame_ctx_t * context,
5075 : fd_quic_stream_8_frame_t * data,
5076 : uchar const * p,
5077 0 : ulong p_sz ) {
5078 0 : return fd_quic_handle_stream_frame( context, p, p_sz, data->stream_id, 0UL, p_sz, data->type&1 );
5079 0 : }
5080 :
5081 : static ulong
5082 : fd_quic_handle_stream_a_frame(
5083 : fd_quic_frame_ctx_t * context,
5084 : fd_quic_stream_a_frame_t * data,
5085 : uchar const * p,
5086 0 : ulong p_sz ) {
5087 0 : return fd_quic_handle_stream_frame( context, p, p_sz, data->stream_id, 0UL, data->length, data->type&1 );
5088 0 : }
5089 :
5090 : static ulong
5091 : fd_quic_handle_stream_c_frame(
5092 : fd_quic_frame_ctx_t * context,
5093 : fd_quic_stream_c_frame_t * data,
5094 : uchar const * p,
5095 0 : ulong p_sz ) {
5096 0 : return fd_quic_handle_stream_frame( context, p, p_sz, data->stream_id, data->offset, p_sz, data->type&1 );
5097 0 : }
5098 :
5099 : static ulong
5100 : fd_quic_handle_stream_e_frame(
5101 : fd_quic_frame_ctx_t * context,
5102 : fd_quic_stream_e_frame_t * data,
5103 : uchar const * p,
5104 0 : ulong p_sz ) {
5105 0 : return fd_quic_handle_stream_frame( context, p, p_sz, data->stream_id, data->offset, data->length, data->type&1 );
5106 0 : }
5107 :
5108 : static ulong
5109 : fd_quic_handle_max_data_frame(
5110 : fd_quic_frame_ctx_t * context,
5111 : fd_quic_max_data_frame_t * data,
5112 : uchar const * p FD_PARAM_UNUSED,
5113 0 : ulong p_sz FD_PARAM_UNUSED ) {
5114 0 : fd_quic_conn_t * conn = context->conn;
5115 :
5116 0 : ulong max_data_old = conn->tx_max_data;
5117 0 : ulong max_data_new = data->max_data;
5118 0 : FD_DTRACE_PROBE_3( quic_handle_max_data_frame, conn->our_conn_id, max_data_new, max_data_old );
5119 :
5120 : /* max data is only allowed to increase the limit. Transgressing frames
5121 : are silently ignored */
5122 0 : conn->tx_max_data = fd_ulong_max( max_data_old, max_data_new );
5123 0 : return 0; /* no additional bytes consumed from buffer */
5124 0 : }
5125 :
5126 : static ulong
5127 : fd_quic_handle_max_stream_data_frame(
5128 : fd_quic_frame_ctx_t * context,
5129 : fd_quic_max_stream_data_frame_t * data,
5130 : uchar const * p FD_PARAM_UNUSED,
5131 0 : ulong p_sz FD_PARAM_UNUSED ) {
5132 : /* FIXME unsupported for now */
5133 0 : FD_DTRACE_PROBE_3( quic_handle_max_stream_data_frame, context->conn->our_conn_id, data->stream_id, data->max_stream_data );
5134 0 : return 0;
5135 0 : }
5136 :
5137 : static ulong
5138 : fd_quic_handle_max_streams_frame(
5139 : fd_quic_frame_ctx_t * context,
5140 : fd_quic_max_streams_frame_t * data,
5141 : uchar const * p FD_PARAM_UNUSED,
5142 0 : ulong p_sz FD_PARAM_UNUSED ) {
5143 0 : fd_quic_conn_t * conn = context->conn;
5144 0 : FD_DTRACE_PROBE_3( quic_handle_max_streams_frame, conn->our_conn_id, data->type, data->max_streams );
5145 :
5146 0 : if( data->type == 0x13 ) {
5147 : /* Only handle unidirectional streams */
5148 0 : ulong type = (ulong)conn->server | 2UL;
5149 : /* Receipt of a frame that permits opening of a stream larger than this limit (2^60)
5150 : MUST be treated as a connection error of type FRAME_ENCODING_ERROR. */
5151 0 : if( FD_UNLIKELY( data->max_streams > FD_QUIC_STREAM_COUNT_MAX ) ) {
5152 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5153 0 : return FD_QUIC_PARSE_FAIL;
5154 0 : }
5155 0 : ulong peer_sup_stream_id = data->max_streams * 4UL + type;
5156 0 : conn->tx_sup_stream_id = fd_ulong_max( peer_sup_stream_id, conn->tx_sup_stream_id );
5157 0 : }
5158 :
5159 0 : return 0;
5160 0 : }
5161 :
5162 : static ulong
5163 : fd_quic_handle_data_blocked_frame(
5164 : fd_quic_frame_ctx_t * context,
5165 : fd_quic_data_blocked_frame_t * data,
5166 : uchar const * p FD_PARAM_UNUSED,
5167 0 : ulong p_sz FD_PARAM_UNUSED ) {
5168 0 : FD_DTRACE_PROBE_2( quic_handle_data_blocked, context->conn->our_conn_id, data->max_data );
5169 :
5170 : /* Since we do not do runtime allocations, we will not attempt
5171 : to find more memory in the case of DATA_BLOCKED. */
5172 0 : return 0;
5173 0 : }
5174 :
5175 : static ulong
5176 : fd_quic_handle_stream_data_blocked_frame(
5177 : fd_quic_frame_ctx_t * context,
5178 : fd_quic_stream_data_blocked_frame_t * data,
5179 : uchar const * p FD_PARAM_UNUSED,
5180 0 : ulong p_sz FD_PARAM_UNUSED ) {
5181 0 : FD_DTRACE_PROBE_3( quic_handle_stream_data_blocked, context->conn->our_conn_id, data->stream_id, data->max_stream_data );
5182 :
5183 : /* Since we do not do runtime allocations, we will not attempt
5184 : to find more memory in the case of STREAM_DATA_BLOCKED.*/
5185 0 : (void)data;
5186 0 : return 0;
5187 0 : }
5188 :
5189 : static ulong
5190 : fd_quic_handle_streams_blocked_frame(
5191 : fd_quic_frame_ctx_t * context,
5192 : fd_quic_streams_blocked_frame_t * data,
5193 : uchar const * p FD_PARAM_UNUSED,
5194 0 : ulong p_sz FD_PARAM_UNUSED ) {
5195 0 : FD_DTRACE_PROBE_2( quic_handle_streams_blocked_frame, context->conn->our_conn_id, data->max_streams );
5196 :
5197 : /* STREAMS_BLOCKED should be sent by client when it wants
5198 : to use a new stream, but is unable to due to the max_streams
5199 : value
5200 : We can support this in the future, but as of 2024-Dec, the
5201 : Agave TPU client does not currently use it */
5202 0 : return 0;
5203 0 : }
5204 :
5205 : static ulong
5206 : fd_quic_handle_new_conn_id_frame(
5207 : fd_quic_frame_ctx_t * context,
5208 : fd_quic_new_conn_id_frame_t * data,
5209 : uchar const * p FD_PARAM_UNUSED,
5210 0 : ulong p_sz FD_PARAM_UNUSED ) {
5211 : /* FIXME This is a mandatory feature but we don't support it yet */
5212 0 : FD_DTRACE_PROBE_1( quic_handle_new_conn_id_frame, context->conn->our_conn_id );
5213 0 : (void)data;
5214 0 : return 0;
5215 0 : }
5216 :
5217 : static ulong
5218 : fd_quic_handle_retire_conn_id_frame(
5219 : fd_quic_frame_ctx_t * context,
5220 : fd_quic_retire_conn_id_frame_t * data,
5221 : uchar const * p FD_PARAM_UNUSED,
5222 0 : ulong p_sz FD_PARAM_UNUSED ) {
5223 : /* FIXME This is a mandatory feature but we don't support it yet */
5224 0 : FD_DTRACE_PROBE_1( quic_handle_retire_conn_id_frame, context->conn->our_conn_id );
5225 0 : (void)data;
5226 0 : FD_DEBUG( FD_LOG_DEBUG(( "retire_conn_id requested" )); )
5227 0 : return 0;
5228 0 : }
5229 :
5230 : static ulong
5231 : fd_quic_handle_path_challenge_frame(
5232 : fd_quic_frame_ctx_t * context,
5233 : fd_quic_path_challenge_frame_t * data,
5234 : uchar const * p FD_PARAM_UNUSED,
5235 0 : ulong p_sz FD_PARAM_UNUSED ) {
5236 : /* FIXME The recipient of this frame MUST generate a PATH_RESPONSE frame (Section 19.18) containing the same Data value. */
5237 0 : FD_DTRACE_PROBE_1( quic_handle_path_challenge_frame, context->conn->our_conn_id );
5238 0 : (void)data;
5239 0 : return 0UL;
5240 0 : }
5241 :
5242 : static ulong
5243 : fd_quic_handle_path_response_frame(
5244 : fd_quic_frame_ctx_t * context,
5245 : fd_quic_path_response_frame_t * data,
5246 : uchar const * p FD_PARAM_UNUSED,
5247 0 : ulong p_sz FD_PARAM_UNUSED ) {
5248 : /* We don't generate PATH_CHALLENGE frames, so this frame should never arrive */
5249 0 : FD_DTRACE_PROBE_1( quic_handle_path_response_frame, context->conn->our_conn_id );
5250 0 : (void)data;
5251 0 : return 0UL;
5252 0 : }
5253 :
5254 : static void
5255 0 : fd_quic_handle_conn_close_frame( fd_quic_conn_t * conn ) {
5256 : /* frame type 0x1c means no error, or only error at quic level
5257 : frame type 0x1d means error at application layer
5258 : TODO provide APP with this info */
5259 0 : FD_DEBUG( FD_LOG_DEBUG(( "peer requested close" )) );
5260 :
5261 0 : switch( conn->state ) {
5262 0 : case FD_QUIC_CONN_STATE_PEER_CLOSE:
5263 0 : case FD_QUIC_CONN_STATE_ABORT:
5264 0 : case FD_QUIC_CONN_STATE_CLOSE_PENDING:
5265 0 : return;
5266 :
5267 0 : default:
5268 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_PEER_CLOSE );
5269 0 : }
5270 :
5271 0 : conn->upd_pkt_number = FD_QUIC_PKT_NUM_PENDING;
5272 0 : fd_quic_svc_prep_schedule_now( conn );
5273 0 : }
5274 :
5275 : static ulong
5276 : fd_quic_handle_conn_close_0_frame(
5277 : fd_quic_frame_ctx_t * context,
5278 : fd_quic_conn_close_0_frame_t * data,
5279 : uchar const * p,
5280 0 : ulong p_sz ) {
5281 0 : (void)p;
5282 :
5283 0 : ulong reason_phrase_length = data->reason_phrase_length;
5284 0 : if( FD_UNLIKELY( reason_phrase_length > p_sz ) ) {
5285 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5286 0 : return FD_QUIC_PARSE_FAIL;
5287 0 : }
5288 :
5289 : /* the information here can be invaluable for debugging */
5290 0 : FD_DEBUG(
5291 0 : char reason_buf[256] = {0};
5292 0 : ulong reason_len = fd_ulong_min( sizeof(reason_buf)-1, reason_phrase_length );
5293 0 : memcpy( reason_buf, p, reason_len );
5294 :
5295 0 : FD_LOG_WARNING(( "fd_quic_handle_conn_close_frame - "
5296 0 : "error_code: %lu "
5297 0 : "frame_type: %lx "
5298 0 : "reason: %s "
5299 0 : "peer " FD_IP4_ADDR_FMT ":%u "
5300 0 : "conn_id %lu",
5301 0 : data->error_code,
5302 0 : data->frame_type,
5303 0 : reason_buf,
5304 0 : FD_IP4_ADDR_FMT_ARGS(context->conn->peer->ip_addr),
5305 0 : context->conn->peer->udp_port,
5306 0 : context->conn->our_conn_id ));
5307 0 : );
5308 :
5309 0 : fd_quic_handle_conn_close_frame( context->conn );
5310 :
5311 0 : return reason_phrase_length;
5312 0 : }
5313 :
5314 : static ulong
5315 : fd_quic_handle_conn_close_1_frame(
5316 : fd_quic_frame_ctx_t * context,
5317 : fd_quic_conn_close_1_frame_t * data,
5318 : uchar const * p,
5319 0 : ulong p_sz ) {
5320 0 : (void)p;
5321 :
5322 0 : ulong reason_phrase_length = data->reason_phrase_length;
5323 0 : if( FD_UNLIKELY( reason_phrase_length > p_sz ) ) {
5324 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_FRAME_ENCODING_ERROR, __LINE__ );
5325 0 : return FD_QUIC_PARSE_FAIL;
5326 0 : }
5327 :
5328 : /* the information here can be invaluable for debugging */
5329 0 : FD_DEBUG(
5330 0 : char reason_buf[256] = {0};
5331 0 : ulong reason_len = fd_ulong_min( sizeof(reason_buf)-1, reason_phrase_length );
5332 0 : memcpy( reason_buf, p, reason_len );
5333 :
5334 0 : FD_LOG_WARNING(( "fd_quic_handle_conn_close_frame - "
5335 0 : "error_code: %lu "
5336 0 : "reason: %s "
5337 0 : "peer " FD_IP4_ADDR_FMT ":%u "
5338 0 : "conn_id %lu",
5339 0 : data->error_code,
5340 0 : reason_buf,
5341 0 : FD_IP4_ADDR_FMT_ARGS(context->conn->peer->ip_addr),
5342 0 : context->conn->peer->udp_port,
5343 0 : context->conn->our_conn_id ));
5344 0 : );
5345 :
5346 0 : fd_quic_handle_conn_close_frame( context->conn );
5347 :
5348 0 : return reason_phrase_length;
5349 0 : }
5350 :
5351 : static ulong
5352 : fd_quic_handle_handshake_done_frame(
5353 : fd_quic_frame_ctx_t * context,
5354 : fd_quic_handshake_done_frame_t * data,
5355 : uchar const * p FD_PARAM_UNUSED,
5356 0 : ulong p_sz FD_PARAM_UNUSED ) {
5357 0 : fd_quic_conn_t * conn = context->conn;
5358 0 : (void)data;
5359 :
5360 : /* servers must treat receipt of HANDSHAKE_DONE as a protocol violation */
5361 0 : if( FD_UNLIKELY( conn->server ) ) {
5362 0 : fd_quic_frame_error( context, FD_QUIC_CONN_REASON_PROTOCOL_VIOLATION, __LINE__ );
5363 0 : return FD_QUIC_PARSE_FAIL;
5364 0 : }
5365 :
5366 0 : if( conn->state == FD_QUIC_CONN_STATE_HANDSHAKE ) {
5367 : /* still handshaking... assume packet was reordered */
5368 0 : context->pkt->ack_flag |= ACK_FLAG_CANCEL;
5369 0 : return 0UL;
5370 0 : } else if( conn->state != FD_QUIC_CONN_STATE_HANDSHAKE_COMPLETE ) {
5371 : /* duplicate frame or conn closing? */
5372 0 : return 0UL;
5373 0 : }
5374 :
5375 : /* Instantly acknowledge the first HANDSHAKE_DONE frame */
5376 0 : fd_quic_svc_prep_schedule_now( conn );
5377 :
5378 : /* RFC 9001 4.9.2. Discarding Handshake Keys
5379 : > An endpoint MUST discard its Handshake keys when the
5380 : > TLS handshake is confirmed
5381 : RFC 9001 4.1.2. Handshake Confirmed
5382 : > At the client, the handshake is considered confirmed when a
5383 : > HANDSHAKE_DONE frame is received. */
5384 0 : fd_quic_abandon_enc_level( conn, fd_quic_enc_level_handshake_id );
5385 :
5386 0 : if( FD_UNLIKELY( !conn->tls_hs ) ) {
5387 : /* sanity check */
5388 0 : return 0;
5389 0 : }
5390 :
5391 : /* eliminate any remaining hs_data at application level */
5392 0 : fd_quic_tls_clear_hs_data( conn->tls_hs, fd_quic_enc_level_appdata_id );
5393 :
5394 : /* we shouldn't be receiving this unless handshake is complete */
5395 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_ACTIVE );
5396 :
5397 : /* user callback */
5398 0 : fd_quic_cb_conn_hs_complete( conn->quic, conn );
5399 :
5400 : /* Deallocate tls_hs once completed */
5401 0 : if( FD_LIKELY( conn->tls_hs ) ) {
5402 0 : fd_quic_state_t * state = fd_quic_get_state( conn->quic );
5403 0 : fd_quic_tls_hs_delete( conn->tls_hs );
5404 0 : fd_quic_tls_hs_cache_ele_remove( &state->hs_cache, conn->tls_hs, state->hs_pool );
5405 0 : fd_quic_tls_hs_pool_ele_release( state->hs_pool, conn->tls_hs );
5406 0 : conn->tls_hs = NULL;
5407 0 : }
5408 :
5409 0 : return 0;
5410 0 : }
5411 :
5412 : /* initiate the shutdown of a connection
5413 : may select a reason code */
5414 : void
5415 : fd_quic_conn_close( fd_quic_conn_t * conn,
5416 0 : uint app_reason ) {
5417 0 : if( FD_UNLIKELY( !conn ) ) return;
5418 :
5419 0 : switch( conn->state ) {
5420 0 : case FD_QUIC_CONN_STATE_INVALID:
5421 0 : case FD_QUIC_CONN_STATE_DEAD:
5422 0 : case FD_QUIC_CONN_STATE_ABORT:
5423 0 : return; /* close has no effect in these states */
5424 :
5425 0 : default:
5426 0 : {
5427 0 : fd_quic_set_conn_state( conn, FD_QUIC_CONN_STATE_CLOSE_PENDING );
5428 0 : conn->app_reason = app_reason;
5429 0 : }
5430 0 : }
5431 :
5432 : /* set connection to be serviced ASAP */
5433 0 : fd_quic_svc_prep_schedule_now( conn );
5434 0 : fd_quic_svc_schedule1( conn );
5435 0 : }
5436 :
5437 : void
5438 : fd_quic_conn_let_die( fd_quic_conn_t * conn,
5439 : long keep_alive_duration,
5440 0 : long now ) {
5441 0 : fd_quic_get_state( conn->quic )->now = now;
5442 0 : conn->let_die_time_ns = fd_long_sat_add( now, keep_alive_duration );
5443 : /* If we've missed the keep-alive time, schedule for immediate servicing */
5444 0 : if( FD_UNLIKELY( conn->last_activity+conn->idle_timeout_ns/2L < now ) ) {
5445 0 : fd_quic_svc_prep_schedule( conn, now );
5446 0 : fd_quic_svc_schedule1( conn );
5447 0 : }
5448 0 : }
|
