]> git.ipfire.org Git - thirdparty/openssl.git/blame - ssl/quic/quic_txp.c
QLOG: Wiring: QUIC FIFD
[thirdparty/openssl.git] / ssl / quic / quic_txp.c
CommitLineData
a73078b7 1/*
da1c088f 2 * Copyright 2022-2023 The OpenSSL Project Authors. All Rights Reserved.
a73078b7
HL
3 *
4 * Licensed under the Apache License 2.0 (the "License"). You may not use
5 * this file except in compliance with the License. You can obtain a copy
6 * in the file LICENSE in the source distribution or at
7 * https://www.openssl.org/source/license.html
8 */
9
10#include "internal/quic_txp.h"
11#include "internal/quic_fifd.h"
12#include "internal/quic_stream_map.h"
96fa10f3 13#include "internal/quic_error.h"
a73078b7
HL
14#include "internal/common.h"
15#include <openssl/err.h>
16
17#define MIN_CRYPTO_HDR_SIZE 3
18
19#define MIN_FRAME_SIZE_HANDSHAKE_DONE 1
20#define MIN_FRAME_SIZE_MAX_DATA 2
21#define MIN_FRAME_SIZE_ACK 5
22#define MIN_FRAME_SIZE_CRYPTO (MIN_CRYPTO_HDR_SIZE + 1)
23#define MIN_FRAME_SIZE_STREAM 3 /* minimum useful size (for non-FIN) */
24#define MIN_FRAME_SIZE_MAX_STREAMS_BIDI 2
25#define MIN_FRAME_SIZE_MAX_STREAMS_UNI 2
26
faebafda
HL
27/*
28 * Packet Archetypes
29 * =================
30 */
31
32/* Generate normal packets containing most frame types, subject to EL. */
33#define TX_PACKETISER_ARCHETYPE_NORMAL 0
34
35/*
36 * A probe packet is different in that:
37 * - It bypasses CC, but *is* counted as in flight for purposes of CC;
38 * - It must be ACK-eliciting.
39 */
40#define TX_PACKETISER_ARCHETYPE_PROBE 1
41
42/*
43 * An ACK-only packet is different in that:
44 * - It bypasses CC, and is considered a 'non-inflight' packet;
45 * - It may not contain anything other than an ACK frame, not even padding.
46 */
47#define TX_PACKETISER_ARCHETYPE_ACK_ONLY 2
48
49#define TX_PACKETISER_ARCHETYPE_NUM 3
50
a73078b7
HL
51struct ossl_quic_tx_packetiser_st {
52 OSSL_QUIC_TX_PACKETISER_ARGS args;
53
54 /*
55 * Opaque initial token blob provided by caller. TXP frees using the
56 * callback when it is no longer needed.
57 */
58 const unsigned char *initial_token;
59 size_t initial_token_len;
60 ossl_quic_initial_token_free_fn *initial_token_free_cb;
61 void *initial_token_free_cb_arg;
62
63 /* Subcomponents of the TXP that we own. */
64 QUIC_FIFD fifd; /* QUIC Frame-in-Flight Dispatcher */
65
66 /* Internal state. */
67 uint64_t next_pn[QUIC_PN_SPACE_NUM]; /* Next PN to use in given PN space. */
68 OSSL_TIME last_tx_time; /* Last time a packet was generated, or 0. */
69
70 /* Internal state - frame (re)generation flags. */
71 unsigned int want_handshake_done : 1;
72 unsigned int want_max_data : 1;
73 unsigned int want_max_streams_bidi : 1;
74 unsigned int want_max_streams_uni : 1;
75
76 /* Internal state - frame (re)generation flags - per PN space. */
77 unsigned int want_ack : QUIC_PN_SPACE_NUM;
78 unsigned int force_ack_eliciting : QUIC_PN_SPACE_NUM;
79
80 /*
81 * Internal state - connection close terminal state.
82 * Once this is set, it is not unset unlike other want_ flags - we keep
83 * sending it in every packet.
84 */
85 unsigned int want_conn_close : 1;
86
cda88baf
HL
87 /* Has the handshake been completed? */
88 unsigned int handshake_complete : 1;
89
a73078b7
HL
90 OSSL_QUIC_FRAME_CONN_CLOSE conn_close_frame;
91
50e76846
P
92 /*
93 * Counts of the number of bytes received and sent while in the closing
94 * state.
95 */
96 uint64_t closing_bytes_recv;
97 uint64_t closing_bytes_xmit;
98
a73078b7 99 /* Internal state - packet assembly. */
faebafda
HL
100 struct txp_el {
101 unsigned char *scratch; /* scratch buffer for packet assembly */
102 size_t scratch_len; /* number of bytes allocated for scratch */
103 OSSL_QTX_IOVEC *iovec; /* scratch iovec array for use with QTX */
104 size_t alloc_iovec; /* size of iovec array */
105 } el[QUIC_ENC_LEVEL_NUM];
5cf99b40
MC
106
107 /* Message callback related arguments */
108 ossl_msg_cb msg_callback;
109 void *msg_callback_arg;
c2786c8e 110 SSL *msg_callback_ssl;
5cf99b40 111
8f9c9213
HL
112 /* Callbacks. */
113 void (*ack_tx_cb)(const OSSL_QUIC_FRAME_ACK *ack,
114 uint32_t pn_space,
115 void *arg);
116 void *ack_tx_cb_arg;
a73078b7
HL
117};
118
119/*
120 * The TX helper records state used while generating frames into packets. It
121 * enables serialization into the packet to be done "transactionally" where
122 * serialization of a frame can be rolled back if it fails midway (e.g. if it
123 * does not fit).
124 */
125struct tx_helper {
126 OSSL_QUIC_TX_PACKETISER *txp;
127 /*
128 * The Maximum Packet Payload Length in bytes. This is the amount of
129 * space we have to generate frames into.
130 */
131 size_t max_ppl;
132 /*
133 * Number of bytes we have generated so far.
134 */
135 size_t bytes_appended;
136 /*
137 * Number of scratch bytes in txp->scratch we have used so far. Some iovecs
138 * will reference this scratch buffer. When we need to use more of it (e.g.
139 * when we need to put frame headers somewhere), we append to the scratch
140 * buffer, resizing if necessary, and increase this accordingly.
141 */
142 size_t scratch_bytes;
143 /*
144 * Bytes reserved in the MaxPPL budget. We keep this number of bytes spare
145 * until reserve_allowed is set to 1. Currently this is always at most 1, as
146 * a PING frame takes up one byte and this mechanism is only used to ensure
147 * we can encode a PING frame if we have been asked to ensure a packet is
148 * ACK-eliciting and we are unusure if we are going to add any other
149 * ACK-eliciting frames before we reach our MaxPPL budget.
150 */
151 size_t reserve;
152 /*
153 * Number of iovecs we have currently appended. This is the number of
154 * entries valid in txp->iovec.
155 */
156 size_t num_iovec;
faebafda
HL
157 /* The EL this TX helper is being used for. */
158 uint32_t enc_level;
a73078b7
HL
159 /*
160 * Whether we are allowed to make use of the reserve bytes in our MaxPPL
161 * budget. This is used to ensure we have room to append a PING frame later
162 * if we need to. Once we know we will not need to append a PING frame, this
163 * is set to 1.
164 */
165 unsigned int reserve_allowed : 1;
166 /*
167 * Set to 1 if we have appended a STREAM frame with an implicit length. If
168 * this happens we should never append another frame after that frame as it
169 * cannot be validly encoded. This is just a safety check.
170 */
171 unsigned int done_implicit : 1;
172 struct {
173 /*
174 * The fields in this structure are valid if active is set, which means
175 * that a serialization transaction is currently in progress.
176 */
177 unsigned char *data;
178 WPACKET wpkt;
179 unsigned int active : 1;
180 } txn;
181};
182
183static void tx_helper_rollback(struct tx_helper *h);
faebafda 184static int txp_el_ensure_iovec(struct txp_el *el, size_t num);
a73078b7
HL
185
186/* Initialises the TX helper. */
187static int tx_helper_init(struct tx_helper *h, OSSL_QUIC_TX_PACKETISER *txp,
faebafda 188 uint32_t enc_level, size_t max_ppl, size_t reserve)
a73078b7
HL
189{
190 if (reserve > max_ppl)
191 return 0;
192
193 h->txp = txp;
faebafda 194 h->enc_level = enc_level;
a73078b7
HL
195 h->max_ppl = max_ppl;
196 h->reserve = reserve;
197 h->num_iovec = 0;
198 h->bytes_appended = 0;
199 h->scratch_bytes = 0;
200 h->reserve_allowed = 0;
201 h->done_implicit = 0;
202 h->txn.data = NULL;
203 h->txn.active = 0;
204
faebafda 205 if (max_ppl > h->txp->el[enc_level].scratch_len) {
a73078b7
HL
206 unsigned char *scratch;
207
faebafda 208 scratch = OPENSSL_realloc(h->txp->el[enc_level].scratch, max_ppl);
a73078b7
HL
209 if (scratch == NULL)
210 return 0;
211
faebafda
HL
212 h->txp->el[enc_level].scratch = scratch;
213 h->txp->el[enc_level].scratch_len = max_ppl;
a73078b7
HL
214 }
215
216 return 1;
217}
218
219static void tx_helper_cleanup(struct tx_helper *h)
220{
221 if (h->txn.active)
222 tx_helper_rollback(h);
223
224 h->txp = NULL;
225}
226
227static void tx_helper_unrestrict(struct tx_helper *h)
228{
229 h->reserve_allowed = 1;
230}
231
232/*
233 * Append an extent of memory to the iovec list. The memory must remain
234 * allocated until we finish generating the packet and call the QTX.
235 *
236 * In general, the buffers passed to this function will be from one of two
237 * ranges:
238 *
239 * - Application data contained in stream buffers managed elsewhere
240 * in the QUIC stack; or
241 *
242 * - Control frame data appended into txp->scratch using tx_helper_begin and
243 * tx_helper_commit.
244 *
245 */
246static int tx_helper_append_iovec(struct tx_helper *h,
247 const unsigned char *buf,
248 size_t buf_len)
249{
faebafda
HL
250 struct txp_el *el = &h->txp->el[h->enc_level];
251
a73078b7
HL
252 if (buf_len == 0)
253 return 1;
254
255 if (!ossl_assert(!h->done_implicit))
256 return 0;
257
faebafda 258 if (!txp_el_ensure_iovec(el, h->num_iovec + 1))
a73078b7
HL
259 return 0;
260
faebafda
HL
261 el->iovec[h->num_iovec].buf = buf;
262 el->iovec[h->num_iovec].buf_len = buf_len;
a73078b7
HL
263
264 ++h->num_iovec;
265 h->bytes_appended += buf_len;
266 return 1;
267}
268
269/*
270 * How many more bytes of space do we have left in our plaintext packet payload?
271 */
272static size_t tx_helper_get_space_left(struct tx_helper *h)
273{
274 return h->max_ppl
275 - (h->reserve_allowed ? 0 : h->reserve) - h->bytes_appended;
276}
277
278/*
279 * Begin a control frame serialization transaction. This allows the
280 * serialization of the control frame to be backed out if it turns out it won't
281 * fit. Write the control frame to the returned WPACKET. Ensure you always
282 * call tx_helper_rollback or tx_helper_commit (or tx_helper_cleanup). Returns
283 * NULL on failure.
284 */
285static WPACKET *tx_helper_begin(struct tx_helper *h)
286{
287 size_t space_left, len;
288 unsigned char *data;
faebafda 289 struct txp_el *el = &h->txp->el[h->enc_level];
a73078b7
HL
290
291 if (!ossl_assert(!h->txn.active))
292 return NULL;
293
294 if (!ossl_assert(!h->done_implicit))
295 return NULL;
296
faebafda
HL
297 data = (unsigned char *)el->scratch + h->scratch_bytes;
298 len = el->scratch_len - h->scratch_bytes;
a73078b7
HL
299
300 space_left = tx_helper_get_space_left(h);
301 if (!ossl_assert(space_left <= len))
302 return NULL;
303
304 if (!WPACKET_init_static_len(&h->txn.wpkt, data, len, 0))
305 return NULL;
306
307 if (!WPACKET_set_max_size(&h->txn.wpkt, space_left)) {
308 WPACKET_cleanup(&h->txn.wpkt);
309 return NULL;
310 }
311
312 h->txn.data = data;
313 h->txn.active = 1;
314 return &h->txn.wpkt;
315}
316
317static void tx_helper_end(struct tx_helper *h, int success)
318{
319 if (success)
320 WPACKET_finish(&h->txn.wpkt);
321 else
322 WPACKET_cleanup(&h->txn.wpkt);
323
324 h->txn.active = 0;
325 h->txn.data = NULL;
326}
327
328/* Abort a control frame serialization transaction. */
329static void tx_helper_rollback(struct tx_helper *h)
330{
331 if (!h->txn.active)
332 return;
333
334 tx_helper_end(h, 0);
335}
336
337/* Commit a control frame. */
338static int tx_helper_commit(struct tx_helper *h)
339{
340 size_t l = 0;
341
342 if (!h->txn.active)
343 return 0;
344
345 if (!WPACKET_get_total_written(&h->txn.wpkt, &l)) {
346 tx_helper_end(h, 0);
347 return 0;
348 }
349
350 if (!tx_helper_append_iovec(h, h->txn.data, l)) {
351 tx_helper_end(h, 0);
352 return 0;
353 }
354
5cf99b40 355 if (h->txp->msg_callback != NULL && l > 0) {
45454ccc
MC
356 uint64_t ftype;
357 int ctype = SSL3_RT_QUIC_FRAME_FULL;
358 PACKET pkt;
359
360 if (!PACKET_buf_init(&pkt, h->txn.data, l)
6c1d0e28 361 || !ossl_quic_wire_peek_frame_header(&pkt, &ftype, NULL)) {
45454ccc
MC
362 tx_helper_end(h, 0);
363 return 0;
364 }
365
366 if (ftype == OSSL_QUIC_FRAME_TYPE_PADDING)
367 ctype = SSL3_RT_QUIC_FRAME_PADDING;
368 else if (OSSL_QUIC_FRAME_TYPE_IS_STREAM(ftype)
369 || ftype == OSSL_QUIC_FRAME_TYPE_CRYPTO)
370 ctype = SSL3_RT_QUIC_FRAME_HEADER;
371
5cf99b40 372 h->txp->msg_callback(1, OSSL_QUIC1_VERSION, ctype, h->txn.data, l,
c2786c8e 373 h->txp->msg_callback_ssl,
5cf99b40 374 h->txp->msg_callback_arg);
45454ccc
MC
375 }
376
a73078b7
HL
377 h->scratch_bytes += l;
378 tx_helper_end(h, 1);
379 return 1;
380}
381
faebafda
HL
382struct archetype_data {
383 unsigned int allow_ack : 1;
384 unsigned int allow_ping : 1;
385 unsigned int allow_crypto : 1;
386 unsigned int allow_handshake_done : 1;
387 unsigned int allow_path_challenge : 1;
388 unsigned int allow_path_response : 1;
389 unsigned int allow_new_conn_id : 1;
390 unsigned int allow_retire_conn_id : 1;
391 unsigned int allow_stream_rel : 1;
392 unsigned int allow_conn_fc : 1;
393 unsigned int allow_conn_close : 1;
394 unsigned int allow_cfq_other : 1;
395 unsigned int allow_new_token : 1;
396 unsigned int allow_force_ack_eliciting : 1;
397 unsigned int allow_padding : 1;
398 unsigned int require_ack_eliciting : 1;
399 unsigned int bypass_cc : 1;
400};
401
402struct txp_pkt_geom {
403 size_t cmpl, cmppl, hwm, pkt_overhead;
404 uint32_t archetype;
405 struct archetype_data adata;
406};
407
408struct txp_pkt {
409 struct tx_helper h;
410 int h_valid;
411 QUIC_TXPIM_PKT *tpkt;
412 QUIC_STREAM *stream_head;
413 QUIC_PKT_HDR phdr;
414 struct txp_pkt_geom geom;
c5cb85b6 415 int force_pad;
faebafda
HL
416};
417
a73078b7
HL
418static QUIC_SSTREAM *get_sstream_by_id(uint64_t stream_id, uint32_t pn_space,
419 void *arg);
420static void on_regen_notify(uint64_t frame_type, uint64_t stream_id,
421 QUIC_TXPIM_PKT *pkt, void *arg);
9cacba43
HL
422static void on_confirm_notify(uint64_t frame_type, uint64_t stream_id,
423 QUIC_TXPIM_PKT *pkt, void *arg);
5d27e7e9 424static void on_sstream_updated(uint64_t stream_id, void *arg);
a73078b7 425static int sstream_is_pending(QUIC_SSTREAM *sstream);
faebafda
HL
426static int txp_should_try_staging(OSSL_QUIC_TX_PACKETISER *txp,
427 uint32_t enc_level,
428 uint32_t archetype,
429 uint64_t cc_limit,
430 uint32_t *conn_close_enc_level);
a73078b7
HL
431static size_t txp_determine_pn_len(OSSL_QUIC_TX_PACKETISER *txp);
432static int txp_determine_ppl_from_pl(OSSL_QUIC_TX_PACKETISER *txp,
433 size_t pl,
434 uint32_t enc_level,
435 size_t hdr_len,
436 size_t *r);
437static size_t txp_get_mdpl(OSSL_QUIC_TX_PACKETISER *txp);
faebafda
HL
438static int txp_generate_for_el(OSSL_QUIC_TX_PACKETISER *txp,
439 struct txp_pkt *pkt,
440 int chosen_for_conn_close);
441static int txp_pkt_init(struct txp_pkt *pkt, OSSL_QUIC_TX_PACKETISER *txp,
442 uint32_t enc_level, uint32_t archetype,
443 size_t running_total);
444static void txp_pkt_cleanup(struct txp_pkt *pkt, OSSL_QUIC_TX_PACKETISER *txp);
445static int txp_pkt_postgen_update_pkt_overhead(struct txp_pkt *pkt,
446 OSSL_QUIC_TX_PACKETISER *txp);
447static int txp_pkt_append_padding(struct txp_pkt *pkt,
448 OSSL_QUIC_TX_PACKETISER *txp, size_t num_bytes);
449static int txp_pkt_commit(OSSL_QUIC_TX_PACKETISER *txp, struct txp_pkt *pkt,
6a2b70e2 450 uint32_t archetype, int *txpim_pkt_reffed);
c206f2aa
HL
451static uint32_t txp_determine_archetype(OSSL_QUIC_TX_PACKETISER *txp,
452 uint64_t cc_limit);
a73078b7
HL
453
454OSSL_QUIC_TX_PACKETISER *ossl_quic_tx_packetiser_new(const OSSL_QUIC_TX_PACKETISER_ARGS *args)
455{
456 OSSL_QUIC_TX_PACKETISER *txp;
457
458 if (args == NULL
459 || args->qtx == NULL
460 || args->txpim == NULL
461 || args->cfq == NULL
462 || args->ackm == NULL
463 || args->qsm == NULL
464 || args->conn_txfc == NULL
a6b6ea17
HL
465 || args->conn_rxfc == NULL
466 || args->max_streams_bidi_rxfc == NULL
467 || args->max_streams_uni_rxfc == NULL) {
a73078b7
HL
468 ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_NULL_PARAMETER);
469 return NULL;
470 }
471
472 txp = OPENSSL_zalloc(sizeof(*txp));
473 if (txp == NULL)
474 return NULL;
475
476 txp->args = *args;
477 txp->last_tx_time = ossl_time_zero();
478
479 if (!ossl_quic_fifd_init(&txp->fifd,
480 txp->args.cfq, txp->args.ackm, txp->args.txpim,
481 get_sstream_by_id, txp,
5d27e7e9 482 on_regen_notify, txp,
9cacba43 483 on_confirm_notify, txp,
3e528780
HL
484 on_sstream_updated, txp,
485 args->qlog)) {
a73078b7
HL
486 OPENSSL_free(txp);
487 return NULL;
488 }
489
490 return txp;
491}
492
493void ossl_quic_tx_packetiser_free(OSSL_QUIC_TX_PACKETISER *txp)
494{
faebafda
HL
495 uint32_t enc_level;
496
a73078b7
HL
497 if (txp == NULL)
498 return;
499
500 ossl_quic_tx_packetiser_set_initial_token(txp, NULL, 0, NULL, NULL);
501 ossl_quic_fifd_cleanup(&txp->fifd);
a73078b7 502 OPENSSL_free(txp->conn_close_frame.reason);
faebafda
HL
503
504 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
505 enc_level < QUIC_ENC_LEVEL_NUM;
506 ++enc_level) {
507 OPENSSL_free(txp->el[enc_level].iovec);
508 OPENSSL_free(txp->el[enc_level].scratch);
509 }
510
a73078b7
HL
511 OPENSSL_free(txp);
512}
513
461d4117
HL
514/*
515 * Determine if an Initial packet token length is reasonable based on the
516 * current MDPL, returning 1 if it is OK.
517 *
518 * The real PMTU to the peer could differ from our (pessimistic) understanding
519 * of the PMTU, therefore it is possible we could receive an Initial token from
520 * a server in a Retry packet which is bigger than the MDPL. In this case it is
521 * impossible for us ever to make forward progress and we need to error out
522 * and fail the connection attempt.
523 *
524 * The specific boundary condition is complex: for example, after the size of
525 * the Initial token, there are the Initial packet header overheads and then
526 * encryption/AEAD tag overheads. After that, the minimum room for frame data in
527 * order to guarantee forward progress must be guaranteed. For example, a crypto
528 * stream needs to always be able to serialize at least one byte in a CRYPTO
529 * frame in order to make forward progress. Because the offset field of a CRYPTO
530 * frame uses a variable-length integer, the number of bytes needed to ensure
531 * this also varies.
532 *
533 * Rather than trying to get this boundary condition check actually right,
534 * require a reasonable amount of slack to avoid pathological behaviours. (After
535 * all, transmitting a CRYPTO stream one byte at a time is probably not
536 * desirable anyway.)
537 *
538 * We choose 160 bytes as the required margin, which is double the rough
539 * estimation of the minimum we would require to guarantee forward progress
540 * under worst case packet overheads.
541 */
542#define TXP_REQUIRED_TOKEN_MARGIN 160
543
544static int txp_check_token_len(size_t token_len, size_t mdpl)
545{
546 if (token_len == 0)
547 return 1;
548
549 if (token_len >= mdpl)
550 return 0;
551
552 if (TXP_REQUIRED_TOKEN_MARGIN >= mdpl)
553 /* (should not be possible because MDPL must be at least 1200) */
554 return 0;
555
556 if (token_len > mdpl - TXP_REQUIRED_TOKEN_MARGIN)
557 return 0;
558
559 return 1;
560}
561
562int ossl_quic_tx_packetiser_set_initial_token(OSSL_QUIC_TX_PACKETISER *txp,
563 const unsigned char *token,
564 size_t token_len,
565 ossl_quic_initial_token_free_fn *free_cb,
566 void *free_cb_arg)
a73078b7 567{
461d4117
HL
568 if (!txp_check_token_len(token_len, txp_get_mdpl(txp)))
569 return 0;
570
a73078b7
HL
571 if (txp->initial_token != NULL && txp->initial_token_free_cb != NULL)
572 txp->initial_token_free_cb(txp->initial_token, txp->initial_token_len,
573 txp->initial_token_free_cb_arg);
574
575 txp->initial_token = token;
576 txp->initial_token_len = token_len;
577 txp->initial_token_free_cb = free_cb;
578 txp->initial_token_free_cb_arg = free_cb_arg;
461d4117 579 return 1;
a73078b7
HL
580}
581
582int ossl_quic_tx_packetiser_set_cur_dcid(OSSL_QUIC_TX_PACKETISER *txp,
583 const QUIC_CONN_ID *dcid)
584{
585 if (dcid == NULL) {
586 ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_NULL_PARAMETER);
587 return 0;
588 }
589
590 txp->args.cur_dcid = *dcid;
591 return 1;
592}
593
594int ossl_quic_tx_packetiser_set_cur_scid(OSSL_QUIC_TX_PACKETISER *txp,
595 const QUIC_CONN_ID *scid)
596{
597 if (scid == NULL) {
598 ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_NULL_PARAMETER);
599 return 0;
600 }
601
602 txp->args.cur_scid = *scid;
603 return 1;
604}
605
606/* Change the destination L4 address the TXP uses to send datagrams. */
607int ossl_quic_tx_packetiser_set_peer(OSSL_QUIC_TX_PACKETISER *txp,
608 const BIO_ADDR *peer)
609{
610 if (peer == NULL) {
617b459d
HL
611 BIO_ADDR_clear(&txp->args.peer);
612 return 1;
a73078b7
HL
613 }
614
615 txp->args.peer = *peer;
616 return 1;
617}
618
8f9c9213
HL
619void ossl_quic_tx_packetiser_set_ack_tx_cb(OSSL_QUIC_TX_PACKETISER *txp,
620 void (*cb)(const OSSL_QUIC_FRAME_ACK *ack,
621 uint32_t pn_space,
622 void *arg),
623 void *cb_arg)
624{
625 txp->ack_tx_cb = cb;
626 txp->ack_tx_cb_arg = cb_arg;
627}
628
a73078b7
HL
629int ossl_quic_tx_packetiser_discard_enc_level(OSSL_QUIC_TX_PACKETISER *txp,
630 uint32_t enc_level)
631{
632 if (enc_level >= QUIC_ENC_LEVEL_NUM) {
633 ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_INVALID_ARGUMENT);
634 return 0;
635 }
636
637 if (enc_level != QUIC_ENC_LEVEL_0RTT)
638 txp->args.crypto[ossl_quic_enc_level_to_pn_space(enc_level)] = NULL;
639
a73078b7
HL
640 return 1;
641}
642
cda88baf
HL
643void ossl_quic_tx_packetiser_notify_handshake_complete(OSSL_QUIC_TX_PACKETISER *txp)
644{
645 txp->handshake_complete = 1;
646}
647
a73078b7
HL
648void ossl_quic_tx_packetiser_schedule_handshake_done(OSSL_QUIC_TX_PACKETISER *txp)
649{
650 txp->want_handshake_done = 1;
651}
652
653void ossl_quic_tx_packetiser_schedule_ack_eliciting(OSSL_QUIC_TX_PACKETISER *txp,
654 uint32_t pn_space)
655{
656 txp->force_ack_eliciting |= (1UL << pn_space);
657}
658
37ba2bc7
HL
659void ossl_quic_tx_packetiser_schedule_ack(OSSL_QUIC_TX_PACKETISER *txp,
660 uint32_t pn_space)
661{
662 txp->want_ack |= (1UL << pn_space);
663}
664
a73078b7
HL
665#define TXP_ERR_INTERNAL 0 /* Internal (e.g. alloc) error */
666#define TXP_ERR_SUCCESS 1 /* Success */
667#define TXP_ERR_SPACE 2 /* Not enough room for another packet */
668#define TXP_ERR_INPUT 3 /* Invalid/malformed input */
669
670/*
671 * Generates a datagram by polling the various ELs to determine if they want to
672 * generate any frames, and generating a datagram which coalesces packets for
673 * any ELs which do.
674 */
675int ossl_quic_tx_packetiser_generate(OSSL_QUIC_TX_PACKETISER *txp,
a3a51d6e 676 QUIC_TXP_STATUS *status)
a73078b7 677{
fee8f48e 678 /*
faebafda
HL
679 * Called to generate one or more datagrams, each containing one or more
680 * packets.
681 *
682 * There are some tricky things to note here:
683 *
684 * - The TXP is only concerned with generating encrypted packets;
685 * other packets use a different path.
686 *
687 * - Any datagram containing an Initial packet must have a payload length
688 * (DPL) of at least 1200 bytes. This padding need not necessarily be
689 * found in the Initial packet.
690 *
691 * - It is desirable to be able to coalesce an Initial packet
692 * with a Handshake packet. Since, before generating the Handshake
693 * packet, we do not know how long it will be, we cannot know the
694 * correct amount of padding to ensure a DPL of at least 1200 bytes.
695 * Thus this padding must added to the Handshake packet (or whatever
696 * packet is the last in the datagram).
697 *
698 * - However, at the time that we generate the Initial packet,
699 * we do not actually know for sure that we will be followed
700 * in the datagram by another packet. For example, suppose we have
701 * some queued data (e.g. crypto stream data for the HANDSHAKE EL)
702 * it looks like we will want to send on the HANDSHAKE EL.
703 * We could assume padding will be placed in the Handshake packet
704 * subsequently and avoid adding any padding to the Initial packet
705 * (which would leave no room for the Handshake packet in the
706 * datagram).
707 *
708 * However, this is not actually a safe assumption. Suppose that we
709 * are using a link with a MDPL of 1200 bytes, the minimum allowed by
710 * QUIC. Suppose that the Initial packet consumes 1195 bytes in total.
711 * Since it is not possible to fit a Handshake packet in just 5 bytes,
712 * upon trying to add a Handshake packet after generating the Initial
713 * packet, we will discover we have no room to fit it! This is not a
714 * problem in itself as another datagram can be sent subsequently, but
715 * it is a problem because we were counting to use that packet to hold
716 * the essential padding. But if we have already finished encrypting
717 * the Initial packet, we cannot go and add padding to it anymore.
718 * This leaves us stuck.
719 *
720 * Because of this, we have to plan multiple packets simultaneously, such
721 * that we can start generating a Handshake (or 0-RTT or 1-RTT, or so on)
722 * packet while still having the option to go back and add padding to the
723 * Initial packet if it turns out to be needed.
724 *
725 * Trying to predict ahead of time (e.g. during Initial packet generation)
726 * whether we will successfully generate a subsequent packet is fraught with
727 * error as it relies on a large number of variables:
728 *
729 * - Do we have room to fit a packet header? (Consider that due to
730 * variable-length integer encoding this is highly variable and can even
731 * depend on payload length due to a variable-length Length field.)
732 *
733 * - Can we fit even a single one of the frames we want to put in this
734 * packet in the packet? (Each frame type has a bespoke encoding. While
735 * our encodings of some frame types are adaptive based on the available
736 * room - e.g. STREAM frames - ultimately all frame types have some
737 * absolute minimum number of bytes to be successfully encoded. For
738 * example, if after an Initial packet there is enough room to encode
739 * only one byte of frame data, it is quite likely we can't send any of
740 * the frames we wanted to send.) While this is not strictly a problem
741 * because we could just fill the packet with padding frames, this is a
742 * pointless packet and is wasteful.
743 *
744 * Thus we adopt a multi-phase architecture:
745 *
746 * 1. Archetype Selection: Determine desired packet archetype.
747 *
748 * 2. Packet Staging: Generation of packet information and packet payload
749 * data (frame data) into staging areas.
750 *
751 * 3. Packet Adjustment: Adjustment of staged packets, adding padding to
752 * the staged packets if needed.
753 *
754 * 4. Commit: The packets are sent to the QTX and recorded as having been
755 * sent to the FIFM.
756 *
fee8f48e 757 */
64fd6991 758 int res = 0, rc;
faebafda
HL
759 uint32_t archetype, enc_level;
760 uint32_t conn_close_enc_level = QUIC_ENC_LEVEL_NUM;
761 struct txp_pkt pkt[QUIC_ENC_LEVEL_NUM];
762 size_t pkts_done = 0;
763 uint64_t cc_limit = txp->args.cc_method->get_tx_allowance(txp->args.cc_data);
6a2b70e2 764 int need_padding = 0, txpim_pkt_reffed;
a73078b7
HL
765
766 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
767 enc_level < QUIC_ENC_LEVEL_NUM;
faebafda
HL
768 ++enc_level)
769 pkt[enc_level].h_valid = 0;
a73078b7 770
64fd6991
TM
771 memset(status, 0, sizeof(*status));
772
a73078b7
HL
773 /*
774 * Should not be needed, but a sanity check in case anyone else has been
775 * using the QTX.
776 */
777 ossl_qtx_finish_dgram(txp->args.qtx);
778
faebafda 779 /* 1. Archetype Selection */
c206f2aa 780 archetype = txp_determine_archetype(txp, cc_limit);
faebafda
HL
781
782 /* 2. Packet Staging */
a73078b7
HL
783 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
784 enc_level < QUIC_ENC_LEVEL_NUM;
785 ++enc_level) {
faebafda
HL
786 size_t running_total = (enc_level > QUIC_ENC_LEVEL_INITIAL)
787 ? pkt[enc_level - 1].geom.hwm : 0;
a73078b7 788
faebafda 789 pkt[enc_level].geom.hwm = running_total;
a73078b7 790
faebafda
HL
791 if (!txp_should_try_staging(txp, enc_level, archetype, cc_limit,
792 &conn_close_enc_level))
793 continue;
794
795 if (!txp_pkt_init(&pkt[enc_level], txp, enc_level, archetype,
796 running_total))
a73078b7 797 /*
faebafda
HL
798 * If this fails this is not a fatal error - it means the geometry
799 * planning determined there was not enough space for another
800 * packet. So just proceed with what we've already planned for.
a73078b7 801 */
faebafda
HL
802 break;
803
804 rc = txp_generate_for_el(txp, &pkt[enc_level],
805 conn_close_enc_level == enc_level);
806 if (rc != TXP_ERR_SUCCESS)
807 goto out;
808
c5cb85b6
HL
809 if (pkt[enc_level].force_pad)
810 /*
811 * txp_generate_for_el emitted a frame which forces packet padding.
812 */
813 need_padding = 1;
814
faebafda
HL
815 pkt[enc_level].geom.hwm = running_total
816 + pkt[enc_level].h.bytes_appended
817 + pkt[enc_level].geom.pkt_overhead;
818 }
819
820 /* 3. Packet Adjustment */
821 if (pkt[QUIC_ENC_LEVEL_INITIAL].h_valid
c5cb85b6 822 && pkt[QUIC_ENC_LEVEL_INITIAL].h.bytes_appended > 0)
faebafda
HL
823 /*
824 * We have an Initial packet in this datagram, so we need to make sure
825 * the total size of the datagram is adequate.
826 */
c5cb85b6
HL
827 need_padding = 1;
828
829 if (need_padding) {
faebafda
HL
830 size_t total_dgram_size = 0;
831 const size_t min_dpl = QUIC_MIN_INITIAL_DGRAM_LEN;
e1c15a8a 832 uint32_t pad_el = QUIC_ENC_LEVEL_NUM;
faebafda
HL
833
834 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
835 enc_level < QUIC_ENC_LEVEL_NUM;
836 ++enc_level)
837 if (pkt[enc_level].h_valid && pkt[enc_level].h.bytes_appended > 0) {
e1c15a8a
HL
838 if (pad_el == QUIC_ENC_LEVEL_NUM
839 /*
840 * We might not be able to add padding, for example if we
841 * are using the ACK_ONLY archetype.
842 */
843 && pkt[enc_level].geom.adata.allow_padding
844 && !pkt[enc_level].h.done_implicit)
845 pad_el = enc_level;
c5cb85b6 846
faebafda
HL
847 txp_pkt_postgen_update_pkt_overhead(&pkt[enc_level], txp);
848 total_dgram_size += pkt[enc_level].geom.pkt_overhead
849 + pkt[enc_level].h.bytes_appended;
850 }
851
e1c15a8a 852 if (pad_el != QUIC_ENC_LEVEL_NUM && total_dgram_size < min_dpl) {
faebafda
HL
853 size_t deficit = min_dpl - total_dgram_size;
854
e1c15a8a 855 if (!txp_pkt_append_padding(&pkt[pad_el], txp, deficit))
3bef14c5
HL
856 goto out;
857
e1c15a8a 858 total_dgram_size += deficit;
3bef14c5
HL
859
860 /*
861 * Padding frames make a packet ineligible for being a non-inflight
862 * packet.
863 */
e1c15a8a
HL
864 pkt[pad_el].tpkt->ackm_pkt.is_inflight = 1;
865 }
866
867 /*
868 * If we have failed to make a datagram of adequate size, for example
869 * because we have a padding requirement but are using the ACK_ONLY
870 * archetype (because we are CC limited), which precludes us from
871 * sending padding, give up on generating the datagram - there is
872 * nothing we can do.
873 */
874 if (total_dgram_size < min_dpl) {
875 res = 1;
876 goto out;
a73078b7 877 }
faebafda
HL
878 }
879
880 /* 4. Commit */
faebafda
HL
881 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
882 enc_level < QUIC_ENC_LEVEL_NUM;
883 ++enc_level) {
884
885 if (!pkt[enc_level].h_valid)
886 /* Did not attempt to generate a packet for this EL. */
887 continue;
888
889 if (pkt[enc_level].h.bytes_appended == 0)
890 /* Nothing was generated for this EL, so skip. */
891 continue;
892
6a2b70e2
HL
893 rc = txp_pkt_commit(txp, &pkt[enc_level], archetype,
894 &txpim_pkt_reffed);
e1c15a8a 895 if (rc) {
6a2b70e2
HL
896 status->sent_ack_eliciting
897 = status->sent_ack_eliciting
898 || pkt[enc_level].tpkt->ackm_pkt.is_ack_eliciting;
899
e1c15a8a
HL
900 if (enc_level == QUIC_ENC_LEVEL_HANDSHAKE)
901 status->sent_handshake
902 = (pkt[enc_level].h_valid
903 && pkt[enc_level].h.bytes_appended > 0);
904 }
905
6a2b70e2
HL
906 if (txpim_pkt_reffed)
907 pkt[enc_level].tpkt = NULL; /* don't free */
faebafda 908
6a2b70e2
HL
909 if (!rc)
910 goto out;
faebafda 911
a73078b7
HL
912 ++pkts_done;
913 }
914
faebafda 915 /* Flush & Cleanup */
64fd6991 916 res = 1;
faebafda 917out:
a73078b7 918 ossl_qtx_finish_dgram(txp->args.qtx);
a73078b7 919
faebafda
HL
920 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
921 enc_level < QUIC_ENC_LEVEL_NUM;
922 ++enc_level)
923 txp_pkt_cleanup(&pkt[enc_level], txp);
924
64fd6991
TM
925 status->sent_pkt = pkts_done;
926
96014840 927 return res;
faebafda 928}
a73078b7
HL
929
930static const struct archetype_data archetypes[QUIC_ENC_LEVEL_NUM][TX_PACKETISER_ARCHETYPE_NUM] = {
931 /* EL 0(INITIAL) */
932 {
933 /* EL 0(INITIAL) - Archetype 0(NORMAL) */
934 {
935 /*allow_ack =*/ 1,
936 /*allow_ping =*/ 1,
937 /*allow_crypto =*/ 1,
938 /*allow_handshake_done =*/ 0,
939 /*allow_path_challenge =*/ 0,
940 /*allow_path_response =*/ 0,
941 /*allow_new_conn_id =*/ 0,
942 /*allow_retire_conn_id =*/ 0,
943 /*allow_stream_rel =*/ 0,
944 /*allow_conn_fc =*/ 0,
945 /*allow_conn_close =*/ 1,
8d2e353d 946 /*allow_cfq_other =*/ 0,
a73078b7
HL
947 /*allow_new_token =*/ 0,
948 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
949 /*allow_padding =*/ 1,
950 /*require_ack_eliciting =*/ 0,
951 /*bypass_cc =*/ 0,
a73078b7 952 },
faebafda 953 /* EL 0(INITIAL) - Archetype 1(PROBE) */
a73078b7
HL
954 {
955 /*allow_ack =*/ 1,
d56b564b 956 /*allow_ping =*/ 1,
faebafda
HL
957 /*allow_crypto =*/ 1,
958 /*allow_handshake_done =*/ 0,
959 /*allow_path_challenge =*/ 0,
960 /*allow_path_response =*/ 0,
961 /*allow_new_conn_id =*/ 0,
962 /*allow_retire_conn_id =*/ 0,
963 /*allow_stream_rel =*/ 0,
964 /*allow_conn_fc =*/ 0,
965 /*allow_conn_close =*/ 1,
966 /*allow_cfq_other =*/ 0,
967 /*allow_new_token =*/ 0,
968 /*allow_force_ack_eliciting =*/ 1,
969 /*allow_padding =*/ 1,
970 /*require_ack_eliciting =*/ 1,
971 /*bypass_cc =*/ 1,
972 },
973 /* EL 0(INITIAL) - Archetype 2(ACK_ONLY) */
974 {
975 /*allow_ack =*/ 1,
976 /*allow_ping =*/ 0,
a73078b7
HL
977 /*allow_crypto =*/ 0,
978 /*allow_handshake_done =*/ 0,
979 /*allow_path_challenge =*/ 0,
980 /*allow_path_response =*/ 0,
981 /*allow_new_conn_id =*/ 0,
982 /*allow_retire_conn_id =*/ 0,
983 /*allow_stream_rel =*/ 0,
984 /*allow_conn_fc =*/ 0,
985 /*allow_conn_close =*/ 0,
986 /*allow_cfq_other =*/ 0,
987 /*allow_new_token =*/ 0,
988 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
989 /*allow_padding =*/ 0,
990 /*require_ack_eliciting =*/ 0,
991 /*bypass_cc =*/ 1,
a73078b7
HL
992 },
993 },
994 /* EL 1(HANDSHAKE) */
995 {
996 /* EL 1(HANDSHAKE) - Archetype 0(NORMAL) */
997 {
998 /*allow_ack =*/ 1,
999 /*allow_ping =*/ 1,
1000 /*allow_crypto =*/ 1,
1001 /*allow_handshake_done =*/ 0,
1002 /*allow_path_challenge =*/ 0,
1003 /*allow_path_response =*/ 0,
1004 /*allow_new_conn_id =*/ 0,
1005 /*allow_retire_conn_id =*/ 0,
1006 /*allow_stream_rel =*/ 0,
1007 /*allow_conn_fc =*/ 0,
1008 /*allow_conn_close =*/ 1,
8d2e353d 1009 /*allow_cfq_other =*/ 0,
a73078b7
HL
1010 /*allow_new_token =*/ 0,
1011 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
1012 /*allow_padding =*/ 1,
1013 /*require_ack_eliciting =*/ 0,
1014 /*bypass_cc =*/ 0,
a73078b7 1015 },
faebafda 1016 /* EL 1(HANDSHAKE) - Archetype 1(PROBE) */
a73078b7
HL
1017 {
1018 /*allow_ack =*/ 1,
d56b564b 1019 /*allow_ping =*/ 1,
faebafda
HL
1020 /*allow_crypto =*/ 1,
1021 /*allow_handshake_done =*/ 0,
1022 /*allow_path_challenge =*/ 0,
1023 /*allow_path_response =*/ 0,
1024 /*allow_new_conn_id =*/ 0,
1025 /*allow_retire_conn_id =*/ 0,
1026 /*allow_stream_rel =*/ 0,
1027 /*allow_conn_fc =*/ 0,
1028 /*allow_conn_close =*/ 1,
1029 /*allow_cfq_other =*/ 0,
1030 /*allow_new_token =*/ 0,
1031 /*allow_force_ack_eliciting =*/ 1,
1032 /*allow_padding =*/ 1,
1033 /*require_ack_eliciting =*/ 1,
1034 /*bypass_cc =*/ 1,
1035 },
1036 /* EL 1(HANDSHAKE) - Archetype 2(ACK_ONLY) */
1037 {
1038 /*allow_ack =*/ 1,
1039 /*allow_ping =*/ 0,
a73078b7
HL
1040 /*allow_crypto =*/ 0,
1041 /*allow_handshake_done =*/ 0,
1042 /*allow_path_challenge =*/ 0,
1043 /*allow_path_response =*/ 0,
1044 /*allow_new_conn_id =*/ 0,
1045 /*allow_retire_conn_id =*/ 0,
1046 /*allow_stream_rel =*/ 0,
1047 /*allow_conn_fc =*/ 0,
1048 /*allow_conn_close =*/ 0,
1049 /*allow_cfq_other =*/ 0,
1050 /*allow_new_token =*/ 0,
1051 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
1052 /*allow_padding =*/ 0,
1053 /*require_ack_eliciting =*/ 0,
1054 /*bypass_cc =*/ 1,
a73078b7
HL
1055 },
1056 },
1057 /* EL 2(0RTT) */
1058 {
1059 /* EL 2(0RTT) - Archetype 0(NORMAL) */
1060 {
1061 /*allow_ack =*/ 0,
1062 /*allow_ping =*/ 1,
1063 /*allow_crypto =*/ 0,
1064 /*allow_handshake_done =*/ 0,
1065 /*allow_path_challenge =*/ 0,
1066 /*allow_path_response =*/ 0,
1067 /*allow_new_conn_id =*/ 1,
1068 /*allow_retire_conn_id =*/ 1,
1069 /*allow_stream_rel =*/ 1,
1070 /*allow_conn_fc =*/ 1,
1071 /*allow_conn_close =*/ 1,
1072 /*allow_cfq_other =*/ 0,
1073 /*allow_new_token =*/ 0,
1074 /*allow_force_ack_eliciting =*/ 0,
faebafda
HL
1075 /*allow_padding =*/ 1,
1076 /*require_ack_eliciting =*/ 0,
1077 /*bypass_cc =*/ 0,
a73078b7 1078 },
faebafda 1079 /* EL 2(0RTT) - Archetype 1(PROBE) */
a73078b7
HL
1080 {
1081 /*allow_ack =*/ 0,
d56b564b 1082 /*allow_ping =*/ 1,
a73078b7
HL
1083 /*allow_crypto =*/ 0,
1084 /*allow_handshake_done =*/ 0,
1085 /*allow_path_challenge =*/ 0,
1086 /*allow_path_response =*/ 0,
faebafda
HL
1087 /*allow_new_conn_id =*/ 1,
1088 /*allow_retire_conn_id =*/ 1,
1089 /*allow_stream_rel =*/ 1,
1090 /*allow_conn_fc =*/ 1,
1091 /*allow_conn_close =*/ 1,
1092 /*allow_cfq_other =*/ 0,
1093 /*allow_new_token =*/ 0,
1094 /*allow_force_ack_eliciting =*/ 0,
1095 /*allow_padding =*/ 1,
1096 /*require_ack_eliciting =*/ 1,
1097 /*bypass_cc =*/ 1,
1098 },
1099 /* EL 2(0RTT) - Archetype 2(ACK_ONLY) */
1100 {
1101 /*allow_ack =*/ 0,
1102 /*allow_ping =*/ 0,
1103 /*allow_crypto =*/ 0,
1104 /*allow_handshake_done =*/ 0,
1105 /*allow_path_challenge =*/ 0,
1106 /*allow_path_response =*/ 0,
a73078b7
HL
1107 /*allow_new_conn_id =*/ 0,
1108 /*allow_retire_conn_id =*/ 0,
1109 /*allow_stream_rel =*/ 0,
1110 /*allow_conn_fc =*/ 0,
1111 /*allow_conn_close =*/ 0,
1112 /*allow_cfq_other =*/ 0,
1113 /*allow_new_token =*/ 0,
1114 /*allow_force_ack_eliciting =*/ 0,
faebafda
HL
1115 /*allow_padding =*/ 0,
1116 /*require_ack_eliciting =*/ 0,
1117 /*bypass_cc =*/ 1,
a73078b7
HL
1118 },
1119 },
1120 /* EL 3(1RTT) */
1121 {
1122 /* EL 3(1RTT) - Archetype 0(NORMAL) */
1123 {
1124 /*allow_ack =*/ 1,
1125 /*allow_ping =*/ 1,
1126 /*allow_crypto =*/ 1,
1127 /*allow_handshake_done =*/ 1,
1128 /*allow_path_challenge =*/ 0,
7eb330ff 1129 /*allow_path_response =*/ 1,
a73078b7
HL
1130 /*allow_new_conn_id =*/ 1,
1131 /*allow_retire_conn_id =*/ 1,
1132 /*allow_stream_rel =*/ 1,
1133 /*allow_conn_fc =*/ 1,
1134 /*allow_conn_close =*/ 1,
1135 /*allow_cfq_other =*/ 1,
1136 /*allow_new_token =*/ 1,
1137 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
1138 /*allow_padding =*/ 1,
1139 /*require_ack_eliciting =*/ 0,
1140 /*bypass_cc =*/ 0,
a73078b7 1141 },
faebafda 1142 /* EL 3(1RTT) - Archetype 1(PROBE) */
a73078b7
HL
1143 {
1144 /*allow_ack =*/ 1,
d56b564b 1145 /*allow_ping =*/ 1,
faebafda
HL
1146 /*allow_crypto =*/ 1,
1147 /*allow_handshake_done =*/ 1,
1148 /*allow_path_challenge =*/ 0,
7eb330ff 1149 /*allow_path_response =*/ 1,
faebafda
HL
1150 /*allow_new_conn_id =*/ 1,
1151 /*allow_retire_conn_id =*/ 1,
1152 /*allow_stream_rel =*/ 1,
1153 /*allow_conn_fc =*/ 1,
1154 /*allow_conn_close =*/ 1,
1155 /*allow_cfq_other =*/ 1,
1156 /*allow_new_token =*/ 1,
1157 /*allow_force_ack_eliciting =*/ 1,
1158 /*allow_padding =*/ 1,
1159 /*require_ack_eliciting =*/ 1,
1160 /*bypass_cc =*/ 1,
1161 },
1162 /* EL 3(1RTT) - Archetype 2(ACK_ONLY) */
1163 {
1164 /*allow_ack =*/ 1,
1165 /*allow_ping =*/ 0,
a73078b7
HL
1166 /*allow_crypto =*/ 0,
1167 /*allow_handshake_done =*/ 0,
1168 /*allow_path_challenge =*/ 0,
1169 /*allow_path_response =*/ 0,
1170 /*allow_new_conn_id =*/ 0,
1171 /*allow_retire_conn_id =*/ 0,
1172 /*allow_stream_rel =*/ 0,
1173 /*allow_conn_fc =*/ 0,
1174 /*allow_conn_close =*/ 0,
1175 /*allow_cfq_other =*/ 0,
1176 /*allow_new_token =*/ 0,
1177 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
1178 /*allow_padding =*/ 0,
1179 /*require_ack_eliciting =*/ 0,
1180 /*bypass_cc =*/ 1,
a73078b7
HL
1181 }
1182 }
1183};
1184
1185static int txp_get_archetype_data(uint32_t enc_level,
1186 uint32_t archetype,
1187 struct archetype_data *a)
1188{
1189 if (enc_level >= QUIC_ENC_LEVEL_NUM
1190 || archetype >= TX_PACKETISER_ARCHETYPE_NUM)
1191 return 0;
1192
1193 /* No need to avoid copying this as it should not exceed one int in size. */
1194 *a = archetypes[enc_level][archetype];
1195 return 1;
1196}
1197
faebafda
HL
1198static int txp_determine_geometry(OSSL_QUIC_TX_PACKETISER *txp,
1199 uint32_t archetype,
1200 uint32_t enc_level,
1201 size_t running_total,
1202 QUIC_PKT_HDR *phdr,
1203 struct txp_pkt_geom *geom)
1204{
1205 size_t mdpl, cmpl, hdr_len;
1206
1207 /* Get information about packet archetype. */
1208 if (!txp_get_archetype_data(enc_level, archetype, &geom->adata))
1209 return 0;
1210
1211 /* Assemble packet header. */
1212 phdr->type = ossl_quic_enc_level_to_pkt_type(enc_level);
1213 phdr->spin_bit = 0;
1214 phdr->pn_len = txp_determine_pn_len(txp);
1215 phdr->partial = 0;
1216 phdr->fixed = 1;
1217 phdr->reserved = 0;
1218 phdr->version = QUIC_VERSION_1;
1219 phdr->dst_conn_id = txp->args.cur_dcid;
1220 phdr->src_conn_id = txp->args.cur_scid;
1221
1222 /*
1223 * We need to know the length of the payload to get an accurate header
1224 * length for non-1RTT packets, because the Length field found in
1225 * Initial/Handshake/0-RTT packets uses a variable-length encoding. However,
1226 * we don't have a good idea of the length of our payload, because the
1227 * length of the payload depends on the room in the datagram after fitting
1228 * the header, which depends on the size of the header.
1229 *
1230 * In general, it does not matter if a packet is slightly shorter (because
1231 * e.g. we predicted use of a 2-byte length field, but ended up only needing
1232 * a 1-byte length field). However this does matter for Initial packets
1233 * which must be at least 1200 bytes, which is also the assumed default MTU;
1234 * therefore in many cases Initial packets will be padded to 1200 bytes,
1235 * which means if we overestimated the header size, we will be short by a
1236 * few bytes and the server will ignore the packet for being too short. In
1237 * this case, however, such packets always *will* be padded to meet 1200
1238 * bytes, which requires a 2-byte length field, so we don't actually need to
1239 * worry about this. Thus we estimate the header length assuming a 2-byte
1240 * length field here, which should in practice work well in all cases.
1241 */
1242 phdr->len = OSSL_QUIC_VLINT_2B_MAX - phdr->pn_len;
1243
1244 if (enc_level == QUIC_ENC_LEVEL_INITIAL) {
1245 phdr->token = txp->initial_token;
1246 phdr->token_len = txp->initial_token_len;
1247 } else {
1248 phdr->token = NULL;
1249 phdr->token_len = 0;
1250 }
1251
1252 hdr_len = ossl_quic_wire_get_encoded_pkt_hdr_len(phdr->dst_conn_id.id_len,
1253 phdr);
1254 if (hdr_len == 0)
1255 return 0;
1256
1257 /* MDPL: Maximum datagram payload length. */
1258 mdpl = txp_get_mdpl(txp);
1259
1260 /*
1261 * CMPL: Maximum encoded packet size we can put into this datagram given any
1262 * previous packets coalesced into it.
1263 */
1264 if (running_total > mdpl)
1265 /* Should not be possible, but if it happens: */
1266 cmpl = 0;
1267 else
1268 cmpl = mdpl - running_total;
1269
1270 /* CMPPL: Maximum amount we can put into the current packet payload */
1271 if (!txp_determine_ppl_from_pl(txp, cmpl, enc_level, hdr_len, &geom->cmppl))
1272 return 0;
1273
1274 geom->cmpl = cmpl;
1275 geom->pkt_overhead = cmpl - geom->cmppl;
1276 geom->archetype = archetype;
1277 return 1;
1278}
1279
c206f2aa
HL
1280static uint32_t txp_determine_archetype(OSSL_QUIC_TX_PACKETISER *txp,
1281 uint64_t cc_limit)
faebafda
HL
1282{
1283 OSSL_ACKM_PROBE_INFO *probe_info
1284 = ossl_ackm_get0_probe_request(txp->args.ackm);
faebafda
HL
1285 uint32_t pn_space;
1286
1287 /*
1288 * If ACKM has requested probe generation (e.g. due to PTO), we generate a
1289 * Probe-archetype packet. Actually, we determine archetype on a
1290 * per-datagram basis, so if any EL wants a probe, do a pass in which
1291 * we try and generate a probe (if needed) for all ELs.
1292 */
1293 if (probe_info->anti_deadlock_initial > 0
1294 || probe_info->anti_deadlock_handshake > 0)
1295 return TX_PACKETISER_ARCHETYPE_PROBE;
1296
1297 for (pn_space = QUIC_PN_SPACE_INITIAL;
1298 pn_space < QUIC_PN_SPACE_NUM;
1299 ++pn_space)
1300 if (probe_info->pto[pn_space] > 0)
1301 return TX_PACKETISER_ARCHETYPE_PROBE;
1302
1303 /*
1304 * If we are out of CC budget, we cannot send a normal packet,
1305 * but we can do an ACK-only packet (potentially, if we
1306 * want to send an ACK).
1307 */
1308 if (cc_limit == 0)
1309 return TX_PACKETISER_ARCHETYPE_ACK_ONLY;
1310
1311 /* All other packets. */
1312 return TX_PACKETISER_ARCHETYPE_NORMAL;
1313}
1314
1315static int txp_should_try_staging(OSSL_QUIC_TX_PACKETISER *txp,
1316 uint32_t enc_level,
1317 uint32_t archetype,
1318 uint64_t cc_limit,
1319 uint32_t *conn_close_enc_level)
a73078b7
HL
1320{
1321 struct archetype_data a;
1322 uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
1323 QUIC_CFQ_ITEM *cfq_item;
1324
1325 if (!ossl_qtx_is_enc_level_provisioned(txp->args.qtx, enc_level))
1326 return 0;
1327
faebafda
HL
1328 if (!txp_get_archetype_data(enc_level, archetype, &a))
1329 return 0;
1330
1331 if (!a.bypass_cc && cc_limit == 0)
1332 /* CC not allowing us to send. */
1333 return 0;
1334
96fa10f3
HL
1335 /*
1336 * We can produce CONNECTION_CLOSE frames on any EL in principle, which
1337 * means we need to choose which EL we would prefer to use. After a
1338 * connection is fully established we have only one provisioned EL and this
1339 * is a non-issue. Where multiple ELs are provisioned, it is possible the
1340 * peer does not have the keys for the EL yet, which suggests in general it
1341 * is preferable to use the lowest EL which is still provisioned.
1342 *
d15d5ea6
P
1343 * However (RFC 9000 s. 10.2.3 & 12.5) we are also required to not send
1344 * application CONNECTION_CLOSE frames in non-1-RTT ELs, so as to not
1345 * potentially leak application data on a connection which has yet to be
1346 * authenticated. Thus when we have an application CONNECTION_CLOSE frame
1347 * queued and need to send it on a non-1-RTT EL, we have to convert it
1348 * into a transport CONNECTION_CLOSE frame which contains no application
1349 * data. Since this loses information, it suggests we should use the 1-RTT
1350 * EL to avoid this if possible, even if a lower EL is also available.
96fa10f3
HL
1351 *
1352 * At the same time, just because we have the 1-RTT EL provisioned locally
1353 * does not necessarily mean the peer does, for example if a handshake
1354 * CRYPTO frame has been lost. It is fairly important that CONNECTION_CLOSE
1355 * is signalled in a way we know our peer can decrypt, as we stop processing
1356 * connection retransmission logic for real after connection close and
1357 * simply 'blindly' retransmit the same CONNECTION_CLOSE frame.
1358 *
1359 * This is not a major concern for clients, since if a client has a 1-RTT EL
1360 * provisioned the server is guaranteed to also have a 1-RTT EL provisioned.
1361 *
44cb36d0 1362 * TODO(QUIC SERVER): Revisit this when server support is added.
96fa10f3
HL
1363 */
1364 if (*conn_close_enc_level > enc_level
1365 && *conn_close_enc_level != QUIC_ENC_LEVEL_1RTT)
7f9d1249
HL
1366 *conn_close_enc_level = enc_level;
1367
fee8f48e
HL
1368 /* Do we need to send a PTO probe? */
1369 if (a.allow_force_ack_eliciting) {
1370 OSSL_ACKM_PROBE_INFO *probe_info
2477e99f 1371 = ossl_ackm_get0_probe_request(txp->args.ackm);
fee8f48e
HL
1372
1373 if ((enc_level == QUIC_ENC_LEVEL_INITIAL
1374 && probe_info->anti_deadlock_initial > 0)
1375 || (enc_level == QUIC_ENC_LEVEL_HANDSHAKE
1376 && probe_info->anti_deadlock_handshake > 0)
1377 || probe_info->pto[pn_space] > 0)
1378 return 1;
1379 }
1380
a73078b7
HL
1381 /* Does the crypto stream for this EL want to produce anything? */
1382 if (a.allow_crypto && sstream_is_pending(txp->args.crypto[pn_space]))
1383 return 1;
1384
1385 /* Does the ACKM for this PN space want to produce anything? */
1386 if (a.allow_ack && (ossl_ackm_is_ack_desired(txp->args.ackm, pn_space)
1387 || (txp->want_ack & (1UL << pn_space)) != 0))
1388 return 1;
1389
1390 /* Do we need to force emission of an ACK-eliciting packet? */
1391 if (a.allow_force_ack_eliciting
1392 && (txp->force_ack_eliciting & (1UL << pn_space)) != 0)
1393 return 1;
1394
1395 /* Does the connection-level RXFC want to produce a frame? */
1396 if (a.allow_conn_fc && (txp->want_max_data
1397 || ossl_quic_rxfc_has_cwm_changed(txp->args.conn_rxfc, 0)))
1398 return 1;
1399
1400 /* Do we want to produce a MAX_STREAMS frame? */
a6b6ea17
HL
1401 if (a.allow_conn_fc
1402 && (txp->want_max_streams_bidi
1403 || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_bidi_rxfc,
1404 0)
1405 || txp->want_max_streams_uni
1406 || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_uni_rxfc,
1407 0)))
a73078b7
HL
1408 return 1;
1409
1410 /* Do we want to produce a HANDSHAKE_DONE frame? */
1411 if (a.allow_handshake_done && txp->want_handshake_done)
1412 return 1;
1413
1414 /* Do we want to produce a CONNECTION_CLOSE frame? */
7f9d1249
HL
1415 if (a.allow_conn_close && txp->want_conn_close &&
1416 *conn_close_enc_level == enc_level)
1417 /*
1418 * This is a bit of a special case since CONNECTION_CLOSE can appear in
1419 * most packet types, and when we decide we want to send it this status
1420 * isn't tied to a specific EL. So if we want to send it, we send it
1421 * only on the lowest non-dropped EL.
1422 */
a73078b7
HL
1423 return 1;
1424
1425 /* Does the CFQ have any frames queued for this PN space? */
1426 if (enc_level != QUIC_ENC_LEVEL_0RTT)
1427 for (cfq_item = ossl_quic_cfq_get_priority_head(txp->args.cfq, pn_space);
1428 cfq_item != NULL;
1429 cfq_item = ossl_quic_cfq_item_get_priority_next(cfq_item, pn_space)) {
1430 uint64_t frame_type = ossl_quic_cfq_item_get_frame_type(cfq_item);
1431
1432 switch (frame_type) {
1433 case OSSL_QUIC_FRAME_TYPE_NEW_CONN_ID:
1434 if (a.allow_new_conn_id)
1435 return 1;
1436 break;
1437 case OSSL_QUIC_FRAME_TYPE_RETIRE_CONN_ID:
1438 if (a.allow_retire_conn_id)
1439 return 1;
1440 break;
1441 case OSSL_QUIC_FRAME_TYPE_NEW_TOKEN:
1442 if (a.allow_new_token)
1443 return 1;
1444 break;
7eb330ff
HL
1445 case OSSL_QUIC_FRAME_TYPE_PATH_RESPONSE:
1446 if (a.allow_path_response)
1447 return 1;
1448 break;
a73078b7
HL
1449 default:
1450 if (a.allow_cfq_other)
1451 return 1;
1452 break;
1453 }
1454 }
1455
cda88baf 1456 if (a.allow_stream_rel && txp->handshake_complete) {
a73078b7
HL
1457 QUIC_STREAM_ITER it;
1458
1459 /* If there are any active streams, 0/1-RTT wants to produce a packet.
1460 * Whether a stream is on the active list is required to be precise
1461 * (i.e., a stream is never on the active list if we cannot produce a
1462 * frame for it), and all stream-related frames are governed by
1463 * a.allow_stream_rel (i.e., if we can send one type of stream-related
1464 * frame, we can send any of them), so we don't need to inspect
1465 * individual streams on the active list, just confirm that the active
1466 * list is non-empty.
1467 */
1468 ossl_quic_stream_iter_init(&it, txp->args.qsm, 0);
1469 if (it.stream != NULL)
1470 return 1;
1471 }
1472
1473 return 0;
1474}
1475
1476static int sstream_is_pending(QUIC_SSTREAM *sstream)
1477{
1478 OSSL_QUIC_FRAME_STREAM hdr;
1479 OSSL_QTX_IOVEC iov[2];
1480 size_t num_iov = OSSL_NELEM(iov);
1481
1482 return ossl_quic_sstream_get_stream_frame(sstream, 0, &hdr, iov, &num_iov);
1483}
1484
a73078b7
HL
1485/* Determine how many bytes we should use for the encoded PN. */
1486static size_t txp_determine_pn_len(OSSL_QUIC_TX_PACKETISER *txp)
1487{
44cb36d0 1488 return 4; /* TODO(QUIC FUTURE) */
a73078b7
HL
1489}
1490
1491/* Determine plaintext packet payload length from payload length. */
1492static int txp_determine_ppl_from_pl(OSSL_QUIC_TX_PACKETISER *txp,
1493 size_t pl,
1494 uint32_t enc_level,
1495 size_t hdr_len,
1496 size_t *r)
1497{
1498 if (pl < hdr_len)
1499 return 0;
1500
1501 pl -= hdr_len;
1502
1503 if (!ossl_qtx_calculate_plaintext_payload_len(txp->args.qtx, enc_level,
1504 pl, &pl))
1505 return 0;
1506
1507 *r = pl;
1508 return 1;
1509}
1510
1511static size_t txp_get_mdpl(OSSL_QUIC_TX_PACKETISER *txp)
1512{
1513 return ossl_qtx_get_mdpl(txp->args.qtx);
1514}
1515
1516static QUIC_SSTREAM *get_sstream_by_id(uint64_t stream_id, uint32_t pn_space,
1517 void *arg)
1518{
1519 OSSL_QUIC_TX_PACKETISER *txp = arg;
1520 QUIC_STREAM *s;
1521
1522 if (stream_id == UINT64_MAX)
1523 return txp->args.crypto[pn_space];
1524
1525 s = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1526 if (s == NULL)
1527 return NULL;
1528
1529 return s->sstream;
1530}
1531
1532static void on_regen_notify(uint64_t frame_type, uint64_t stream_id,
1533 QUIC_TXPIM_PKT *pkt, void *arg)
1534{
1535 OSSL_QUIC_TX_PACKETISER *txp = arg;
1536
1537 switch (frame_type) {
1538 case OSSL_QUIC_FRAME_TYPE_HANDSHAKE_DONE:
1539 txp->want_handshake_done = 1;
1540 break;
1541 case OSSL_QUIC_FRAME_TYPE_MAX_DATA:
1542 txp->want_max_data = 1;
1543 break;
1544 case OSSL_QUIC_FRAME_TYPE_MAX_STREAMS_BIDI:
1545 txp->want_max_streams_bidi = 1;
1546 break;
1547 case OSSL_QUIC_FRAME_TYPE_MAX_STREAMS_UNI:
1548 txp->want_max_streams_uni = 1;
1549 break;
1550 case OSSL_QUIC_FRAME_TYPE_ACK_WITH_ECN:
1551 txp->want_ack |= (1UL << pkt->ackm_pkt.pkt_space);
1552 break;
1553 case OSSL_QUIC_FRAME_TYPE_MAX_STREAM_DATA:
1554 {
1555 QUIC_STREAM *s
1556 = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1557
1558 if (s == NULL)
1559 return;
1560
1561 s->want_max_stream_data = 1;
1562 ossl_quic_stream_map_update_state(txp->args.qsm, s);
1563 }
1564 break;
1565 case OSSL_QUIC_FRAME_TYPE_STOP_SENDING:
1566 {
1567 QUIC_STREAM *s
1568 = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1569
1570 if (s == NULL)
1571 return;
1572
418e122c 1573 ossl_quic_stream_map_schedule_stop_sending(txp->args.qsm, s);
a73078b7
HL
1574 }
1575 break;
1576 case OSSL_QUIC_FRAME_TYPE_RESET_STREAM:
1577 {
1578 QUIC_STREAM *s
1579 = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1580
1581 if (s == NULL)
1582 return;
1583
1584 s->want_reset_stream = 1;
1585 ossl_quic_stream_map_update_state(txp->args.qsm, s);
1586 }
1587 break;
1588 default:
1589 assert(0);
1590 break;
1591 }
1592}
1593
faebafda
HL
1594static int txp_pkt_init(struct txp_pkt *pkt, OSSL_QUIC_TX_PACKETISER *txp,
1595 uint32_t enc_level, uint32_t archetype,
1596 size_t running_total)
1597{
1598 if (!txp_determine_geometry(txp, archetype, enc_level,
1599 running_total, &pkt->phdr, &pkt->geom))
1600 return 0;
1601
1602 /*
1603 * Initialise TX helper. If we must be ACK eliciting, reserve 1 byte for
1604 * PING.
1605 */
1606 if (!tx_helper_init(&pkt->h, txp, enc_level,
1607 pkt->geom.cmppl,
1608 pkt->geom.adata.require_ack_eliciting ? 1 : 0))
1609 return 0;
1610
1611 pkt->h_valid = 1;
1612 pkt->tpkt = NULL;
1613 pkt->stream_head = NULL;
c5cb85b6 1614 pkt->force_pad = 0;
faebafda
HL
1615 return 1;
1616}
1617
1618static void txp_pkt_cleanup(struct txp_pkt *pkt, OSSL_QUIC_TX_PACKETISER *txp)
1619{
1620 if (!pkt->h_valid)
1621 return;
1622
1623 tx_helper_cleanup(&pkt->h);
1624 pkt->h_valid = 0;
1625
1626 if (pkt->tpkt != NULL) {
1627 ossl_quic_txpim_pkt_release(txp->args.txpim, pkt->tpkt);
1628 pkt->tpkt = NULL;
1629 }
1630}
1631
1632static int txp_pkt_postgen_update_pkt_overhead(struct txp_pkt *pkt,
1633 OSSL_QUIC_TX_PACKETISER *txp)
1634{
1635 /*
1636 * After we have staged and generated our packets, but before we commit
1637 * them, it is possible for the estimated packet overhead (packet header +
1638 * AEAD tag size) to shrink slightly because we generated a short packet
1639 * whose which can be represented in fewer bytes as a variable-length
1640 * integer than we were (pessimistically) budgeting for. We need to account
1641 * for this to ensure that we get our padding calculation exactly right.
1642 *
1643 * Update pkt_overhead to be accurate now that we know how much data is
1644 * going in a packet.
1645 */
1646 size_t hdr_len, ciphertext_len;
1647
1648 if (pkt->h.enc_level == QUIC_ENC_LEVEL_INITIAL)
1649 /*
1650 * Don't update overheads for the INITIAL EL - we have not finished
1651 * appending padding to it and would potentially miscalculate the
1652 * correct padding if we now update the pkt_overhead field to switch to
1653 * e.g. a 1-byte length field in the packet header. Since we are padding
1654 * to QUIC_MIN_INITIAL_DGRAM_LEN which requires a 2-byte length field,
1655 * this is guaranteed to be moot anyway. See comment in
1656 * txp_determine_geometry for more information.
1657 */
1658 return 1;
1659
1660 if (!ossl_qtx_calculate_ciphertext_payload_len(txp->args.qtx, pkt->h.enc_level,
1661 pkt->h.bytes_appended,
1662 &ciphertext_len))
1663 return 0;
1664
1665 pkt->phdr.len = ciphertext_len;
1666
1667 hdr_len = ossl_quic_wire_get_encoded_pkt_hdr_len(pkt->phdr.dst_conn_id.id_len,
1668 &pkt->phdr);
1669
1670 pkt->geom.pkt_overhead = hdr_len + ciphertext_len - pkt->h.bytes_appended;
1671 return 1;
1672}
1673
9cacba43
HL
1674static void on_confirm_notify(uint64_t frame_type, uint64_t stream_id,
1675 QUIC_TXPIM_PKT *pkt, void *arg)
1676{
1677 OSSL_QUIC_TX_PACKETISER *txp = arg;
1678
1679 switch (frame_type) {
1680 case OSSL_QUIC_FRAME_TYPE_STOP_SENDING:
1681 {
1682 QUIC_STREAM *s
1683 = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1684
1685 if (s == NULL)
1686 return;
1687
1688 s->acked_stop_sending = 1;
1689 ossl_quic_stream_map_update_state(txp->args.qsm, s);
1690 }
1691 break;
1692 case OSSL_QUIC_FRAME_TYPE_RESET_STREAM:
1693 {
1694 QUIC_STREAM *s
1695 = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1696
1697 if (s == NULL)
1698 return;
1699
2f018d14
HL
1700 /*
1701 * We must already be in RESET_SENT or RESET_RECVD if we are
1702 * here, so we don't need to check state here.
1703 */
1704 ossl_quic_stream_map_notify_reset_stream_acked(txp->args.qsm, s);
9cacba43
HL
1705 ossl_quic_stream_map_update_state(txp->args.qsm, s);
1706 }
1707 break;
1708 default:
1709 assert(0);
1710 break;
1711 }
1712}
1713
faebafda
HL
1714static int txp_pkt_append_padding(struct txp_pkt *pkt,
1715 OSSL_QUIC_TX_PACKETISER *txp, size_t num_bytes)
1716{
1717 WPACKET *wpkt;
1718
1719 if (num_bytes == 0)
1720 return 1;
1721
1722 if (!ossl_assert(pkt->h_valid))
1723 return 0;
1724
1725 if (!ossl_assert(pkt->tpkt != NULL))
1726 return 0;
1727
1728 wpkt = tx_helper_begin(&pkt->h);
1729 if (wpkt == NULL)
1730 return 0;
1731
1732 if (!ossl_quic_wire_encode_padding(wpkt, num_bytes)) {
1733 tx_helper_rollback(&pkt->h);
1734 return 0;
1735 }
1736
1737 if (!tx_helper_commit(&pkt->h))
1738 return 0;
1739
1740 pkt->tpkt->ackm_pkt.num_bytes += num_bytes;
1741 /* Cannot be non-inflight if we have a PADDING frame */
1742 pkt->tpkt->ackm_pkt.is_inflight = 1;
1743 return 1;
1744}
1745
5d27e7e9
HL
1746static void on_sstream_updated(uint64_t stream_id, void *arg)
1747{
1748 OSSL_QUIC_TX_PACKETISER *txp = arg;
1749 QUIC_STREAM *s;
1750
1751 s = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1752 if (s == NULL)
1753 return;
1754
1755 ossl_quic_stream_map_update_state(txp->args.qsm, s);
1756}
1757
50e76846
P
1758/*
1759 * Returns 1 if we can send that many bytes in closing state, 0 otherwise.
1760 * Also maintains the bytes sent state if it returns a success.
1761 */
1762static int try_commit_conn_close(OSSL_QUIC_TX_PACKETISER *txp, size_t n)
1763{
1764 int res;
1765
1766 /* We can always send the first connection close frame */
1767 if (txp->closing_bytes_recv == 0)
1768 return 1;
1769
1770 /*
1771 * RFC 9000 s. 10.2.1 Closing Connection State:
1772 * To avoid being used for an amplification attack, such
1773 * endpoints MUST limit the cumulative size of packets it sends
1774 * to three times the cumulative size of the packets that are
1775 * received and attributed to the connection.
1776 * and:
1777 * An endpoint in the closing state MUST either discard packets
1778 * received from an unvalidated address or limit the cumulative
1779 * size of packets it sends to an unvalidated address to three
1780 * times the size of packets it receives from that address.
1781 */
1782 res = txp->closing_bytes_xmit + n <= txp->closing_bytes_recv * 3;
1783
1784 /*
1785 * Attribute the bytes to the connection, if we are allowed to send them
1786 * and this isn't the first closing frame.
1787 */
1788 if (res && txp->closing_bytes_recv != 0)
1789 txp->closing_bytes_xmit += n;
1790 return res;
1791}
1792
1793void ossl_quic_tx_packetiser_record_received_closing_bytes(
1794 OSSL_QUIC_TX_PACKETISER *txp, size_t n)
1795{
1796 txp->closing_bytes_recv += n;
1797}
1798
a73078b7 1799static int txp_generate_pre_token(OSSL_QUIC_TX_PACKETISER *txp,
faebafda 1800 struct txp_pkt *pkt,
178c104d
HL
1801 int chosen_for_conn_close,
1802 int *can_be_non_inflight)
a73078b7 1803{
faebafda
HL
1804 const uint32_t enc_level = pkt->h.enc_level;
1805 const uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
1806 const struct archetype_data *a = &pkt->geom.adata;
1807 QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
1808 struct tx_helper *h = &pkt->h;
a73078b7
HL
1809 const OSSL_QUIC_FRAME_ACK *ack;
1810 OSSL_QUIC_FRAME_ACK ack2;
1811
1812 tpkt->ackm_pkt.largest_acked = QUIC_PN_INVALID;
1813
1814 /* ACK Frames (Regenerate) */
1815 if (a->allow_ack
1816 && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_ACK
d13488b9 1817 && (((txp->want_ack & (1UL << pn_space)) != 0)
a73078b7
HL
1818 || ossl_ackm_is_ack_desired(txp->args.ackm, pn_space))
1819 && (ack = ossl_ackm_get_ack_frame(txp->args.ackm, pn_space)) != NULL) {
1820 WPACKET *wpkt = tx_helper_begin(h);
1821
1822 if (wpkt == NULL)
1823 return 0;
1824
1825 /* We do not currently support ECN */
1826 ack2 = *ack;
1827 ack2.ecn_present = 0;
1828
1829 if (ossl_quic_wire_encode_frame_ack(wpkt,
1830 txp->args.ack_delay_exponent,
1831 &ack2)) {
1832 if (!tx_helper_commit(h))
1833 return 0;
1834
1835 tpkt->had_ack_frame = 1;
1836
1837 if (ack->num_ack_ranges > 0)
1838 tpkt->ackm_pkt.largest_acked = ack->ack_ranges[0].end;
8f9c9213
HL
1839
1840 if (txp->ack_tx_cb != NULL)
1841 txp->ack_tx_cb(&ack2, pn_space, txp->ack_tx_cb_arg);
a73078b7
HL
1842 } else {
1843 tx_helper_rollback(h);
1844 }
1845 }
1846
1847 /* CONNECTION_CLOSE Frames (Regenerate) */
7f9d1249 1848 if (a->allow_conn_close && txp->want_conn_close && chosen_for_conn_close) {
a73078b7 1849 WPACKET *wpkt = tx_helper_begin(h);
96fa10f3 1850 OSSL_QUIC_FRAME_CONN_CLOSE f, *pf = &txp->conn_close_frame;
50e76846 1851 size_t l;
a73078b7
HL
1852
1853 if (wpkt == NULL)
1854 return 0;
1855
96fa10f3
HL
1856 /*
1857 * Application CONNECTION_CLOSE frames may only be sent in the
1858 * Application PN space, as otherwise they may be sent before a
1859 * connection is authenticated and leak application data. Therefore, if
1860 * we need to send a CONNECTION_CLOSE frame in another PN space and were
1861 * given an application CONNECTION_CLOSE frame, convert it into a
1862 * transport CONNECTION_CLOSE frame, removing any sensitive application
1863 * data.
1864 *
1865 * RFC 9000 s. 10.2.3: "A CONNECTION_CLOSE of type 0x1d MUST be replaced
1866 * by a CONNECTION_CLOSE of type 0x1c when sending the frame in Initial
1867 * or Handshake packets. Otherwise, information about the application
1868 * state might be revealed. Endpoints MUST clear the value of the Reason
1869 * Phrase field and SHOULD use the APPLICATION_ERROR code when
1870 * converting to a CONNECTION_CLOSE of type 0x1c."
1871 */
1872 if (pn_space != QUIC_PN_SPACE_APP && pf->is_app) {
1873 pf = &f;
1874 pf->is_app = 0;
1875 pf->frame_type = 0;
1876 pf->error_code = QUIC_ERR_APPLICATION_ERROR;
1877 pf->reason = NULL;
1878 pf->reason_len = 0;
1879 }
1880
50e76846
P
1881 if (ossl_quic_wire_encode_frame_conn_close(wpkt, pf)
1882 && WPACKET_get_total_written(wpkt, &l)
1883 && try_commit_conn_close(txp, l)) {
a73078b7
HL
1884 if (!tx_helper_commit(h))
1885 return 0;
178c104d 1886
0b31072e 1887 tpkt->had_conn_close = 1;
178c104d 1888 *can_be_non_inflight = 0;
a73078b7
HL
1889 } else {
1890 tx_helper_rollback(h);
1891 }
1892 }
1893
1894 return 1;
1895}
1896
1897static int try_len(size_t space_left, size_t orig_len,
1898 size_t base_hdr_len, size_t lenbytes,
1899 uint64_t maxn, size_t *hdr_len, size_t *payload_len)
1900{
1901 size_t n;
1902 size_t maxn_ = maxn > SIZE_MAX ? SIZE_MAX : (size_t)maxn;
1903
1904 *hdr_len = base_hdr_len + lenbytes;
1905
cf06f347
HL
1906 if (orig_len == 0 && space_left >= *hdr_len) {
1907 *payload_len = 0;
1908 return 1;
1909 }
1910
a73078b7
HL
1911 n = orig_len;
1912 if (n > maxn_)
1913 n = maxn_;
1914 if (n + *hdr_len > space_left)
1915 n = (space_left >= *hdr_len) ? space_left - *hdr_len : 0;
1916
1917 *payload_len = n;
1918 return n > 0;
1919}
1920
cf06f347
HL
1921static int determine_len(size_t space_left, size_t orig_len,
1922 size_t base_hdr_len,
1923 uint64_t *hlen, uint64_t *len)
a73078b7 1924{
cf06f347 1925 int ok = 0;
a73078b7
HL
1926 size_t chosen_payload_len = 0;
1927 size_t chosen_hdr_len = 0;
1928 size_t payload_len[4], hdr_len[4];
1929 int i, valid[4] = {0};
1930
1931 valid[0] = try_len(space_left, orig_len, base_hdr_len,
1932 1, OSSL_QUIC_VLINT_1B_MAX,
1933 &hdr_len[0], &payload_len[0]);
1934 valid[1] = try_len(space_left, orig_len, base_hdr_len,
1935 2, OSSL_QUIC_VLINT_2B_MAX,
1936 &hdr_len[1], &payload_len[1]);
1937 valid[2] = try_len(space_left, orig_len, base_hdr_len,
1938 4, OSSL_QUIC_VLINT_4B_MAX,
1939 &hdr_len[2], &payload_len[2]);
1940 valid[3] = try_len(space_left, orig_len, base_hdr_len,
1941 8, OSSL_QUIC_VLINT_8B_MAX,
1942 &hdr_len[3], &payload_len[3]);
1943
1944 for (i = OSSL_NELEM(valid) - 1; i >= 0; --i)
1945 if (valid[i] && payload_len[i] >= chosen_payload_len) {
1946 chosen_payload_len = payload_len[i];
1947 chosen_hdr_len = hdr_len[i];
cf06f347 1948 ok = 1;
a73078b7
HL
1949 }
1950
1951 *hlen = chosen_hdr_len;
1952 *len = chosen_payload_len;
cf06f347 1953 return ok;
a73078b7
HL
1954}
1955
1956/*
1957 * Given a CRYPTO frame header with accurate chdr->len and a budget
1958 * (space_left), try to find the optimal value of chdr->len to fill as much of
1959 * the budget as possible. This is slightly hairy because larger values of
1960 * chdr->len cause larger encoded sizes of the length field of the frame, which
1961 * in turn mean less space available for payload data. We check all possible
1962 * encodings and choose the optimal encoding.
1963 */
1964static int determine_crypto_len(struct tx_helper *h,
1965 OSSL_QUIC_FRAME_CRYPTO *chdr,
1966 size_t space_left,
1967 uint64_t *hlen,
1968 uint64_t *len)
1969{
1970 size_t orig_len;
1971 size_t base_hdr_len; /* CRYPTO header length without length field */
1972
1973 if (chdr->len > SIZE_MAX)
1974 return 0;
1975
1976 orig_len = (size_t)chdr->len;
1977
1978 chdr->len = 0;
1979 base_hdr_len = ossl_quic_wire_get_encoded_frame_len_crypto_hdr(chdr);
1980 chdr->len = orig_len;
1981 if (base_hdr_len == 0)
1982 return 0;
1983
1984 --base_hdr_len;
1985
cf06f347 1986 return determine_len(space_left, orig_len, base_hdr_len, hlen, len);
a73078b7
HL
1987}
1988
1989static int determine_stream_len(struct tx_helper *h,
1990 OSSL_QUIC_FRAME_STREAM *shdr,
1991 size_t space_left,
1992 uint64_t *hlen,
1993 uint64_t *len)
1994{
1995 size_t orig_len;
1996 size_t base_hdr_len; /* STREAM header length without length field */
1997
1998 if (shdr->len > SIZE_MAX)
1999 return 0;
2000
2001 orig_len = (size_t)shdr->len;
2002
2003 shdr->len = 0;
2004 base_hdr_len = ossl_quic_wire_get_encoded_frame_len_stream_hdr(shdr);
2005 shdr->len = orig_len;
2006 if (base_hdr_len == 0)
2007 return 0;
2008
2009 if (shdr->has_explicit_len)
2010 --base_hdr_len;
2011
cf06f347 2012 return determine_len(space_left, orig_len, base_hdr_len, hlen, len);
a73078b7
HL
2013}
2014
2015static int txp_generate_crypto_frames(OSSL_QUIC_TX_PACKETISER *txp,
faebafda 2016 struct txp_pkt *pkt,
091f532e 2017 int *have_ack_eliciting)
a73078b7 2018{
faebafda
HL
2019 const uint32_t enc_level = pkt->h.enc_level;
2020 const uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
2021 QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
2022 struct tx_helper *h = &pkt->h;
a73078b7
HL
2023 size_t num_stream_iovec;
2024 OSSL_QUIC_FRAME_STREAM shdr = {0};
2025 OSSL_QUIC_FRAME_CRYPTO chdr = {0};
2026 OSSL_QTX_IOVEC iov[2];
2027 uint64_t hdr_bytes;
2028 WPACKET *wpkt;
24c1be5c 2029 QUIC_TXPIM_CHUNK chunk = {0};
a73078b7
HL
2030 size_t i, space_left;
2031
2032 for (i = 0;; ++i) {
2033 space_left = tx_helper_get_space_left(h);
2034
2035 if (space_left < MIN_FRAME_SIZE_CRYPTO)
2036 return 1; /* no point trying */
2037
2038 /* Do we have any CRYPTO data waiting? */
2039 num_stream_iovec = OSSL_NELEM(iov);
2040 if (!ossl_quic_sstream_get_stream_frame(txp->args.crypto[pn_space],
2041 i, &shdr, iov,
2042 &num_stream_iovec))
2043 return 1; /* nothing to do */
2044
2045 /* Convert STREAM frame header to CRYPTO frame header */
2046 chdr.offset = shdr.offset;
2047 chdr.len = shdr.len;
2048
2049 if (chdr.len == 0)
2050 return 1; /* nothing to do */
2051
2052 /* Find best fit (header length, payload length) combination. */
2053 if (!determine_crypto_len(h, &chdr, space_left, &hdr_bytes,
cf06f347 2054 &chdr.len))
a73078b7 2055 return 1; /* can't fit anything */
a73078b7
HL
2056
2057 /*
2058 * Truncate IOVs to match our chosen length.
2059 *
2060 * The length cannot be more than SIZE_MAX because this length comes
2061 * from our send stream buffer.
2062 */
2063 ossl_quic_sstream_adjust_iov((size_t)chdr.len, iov, num_stream_iovec);
2064
2065 /*
2066 * Ensure we have enough iovecs allocated (1 for the header, up to 2 for
ad31628c 2067 * the stream data.)
a73078b7 2068 */
faebafda 2069 if (!txp_el_ensure_iovec(&txp->el[enc_level], h->num_iovec + 3))
a73078b7
HL
2070 return 0; /* alloc error */
2071
2072 /* Encode the header. */
2073 wpkt = tx_helper_begin(h);
2074 if (wpkt == NULL)
2075 return 0; /* alloc error */
2076
2077 if (!ossl_quic_wire_encode_frame_crypto_hdr(wpkt, &chdr)) {
2078 tx_helper_rollback(h);
2079 return 1; /* can't fit */
2080 }
2081
2082 if (!tx_helper_commit(h))
2083 return 0; /* alloc error */
2084
2085 /* Add payload iovecs to the helper (infallible). */
2086 for (i = 0; i < num_stream_iovec; ++i)
2087 tx_helper_append_iovec(h, iov[i].buf, iov[i].buf_len);
2088
2089 *have_ack_eliciting = 1;
2090 tx_helper_unrestrict(h); /* no longer need PING */
2091
2092 /* Log chunk to TXPIM. */
2093 chunk.stream_id = UINT64_MAX; /* crypto stream */
2094 chunk.start = chdr.offset;
2095 chunk.end = chdr.offset + chdr.len - 1;
2096 chunk.has_fin = 0; /* Crypto stream never ends */
2097 if (!ossl_quic_txpim_pkt_append_chunk(tpkt, &chunk))
2098 return 0; /* alloc error */
2099 }
2100}
2101
2102struct chunk_info {
2103 OSSL_QUIC_FRAME_STREAM shdr;
e718b248 2104 uint64_t orig_len;
a73078b7
HL
2105 OSSL_QTX_IOVEC iov[2];
2106 size_t num_stream_iovec;
091f532e 2107 int valid;
a73078b7
HL
2108};
2109
2110static int txp_plan_stream_chunk(OSSL_QUIC_TX_PACKETISER *txp,
2111 struct tx_helper *h,
2112 QUIC_SSTREAM *sstream,
2113 QUIC_TXFC *stream_txfc,
2114 size_t skip,
e57bf6b3
MC
2115 struct chunk_info *chunk,
2116 uint64_t consumed)
a73078b7
HL
2117{
2118 uint64_t fc_credit, fc_swm, fc_limit;
2119
2120 chunk->num_stream_iovec = OSSL_NELEM(chunk->iov);
2121 chunk->valid = ossl_quic_sstream_get_stream_frame(sstream, skip,
2122 &chunk->shdr,
2123 chunk->iov,
2124 &chunk->num_stream_iovec);
2125 if (!chunk->valid)
2126 return 1;
2127
2128 if (!ossl_assert(chunk->shdr.len > 0 || chunk->shdr.is_fin))
2129 /* Should only have 0-length chunk if FIN */
2130 return 0;
2131
e718b248
MC
2132 chunk->orig_len = chunk->shdr.len;
2133
a73078b7 2134 /* Clamp according to connection and stream-level TXFC. */
e57bf6b3 2135 fc_credit = ossl_quic_txfc_get_credit(stream_txfc, consumed);
a73078b7
HL
2136 fc_swm = ossl_quic_txfc_get_swm(stream_txfc);
2137 fc_limit = fc_swm + fc_credit;
2138
2139 if (chunk->shdr.len > 0 && chunk->shdr.offset + chunk->shdr.len > fc_limit) {
2140 chunk->shdr.len = (fc_limit <= chunk->shdr.offset)
2141 ? 0 : fc_limit - chunk->shdr.offset;
2142 chunk->shdr.is_fin = 0;
2143 }
2144
2145 if (chunk->shdr.len == 0 && !chunk->shdr.is_fin) {
2146 /*
2147 * Nothing to do due to TXFC. Since SSTREAM returns chunks in ascending
2148 * order of offset we don't need to check any later chunks, so stop
2149 * iterating here.
2150 */
2151 chunk->valid = 0;
2152 return 1;
2153 }
2154
2155 return 1;
2156}
2157
2158/*
2159 * Returns 0 on fatal error (e.g. allocation failure), 1 on success.
2160 * *packet_full is set to 1 if there is no longer enough room for another STREAM
26652896 2161 * frame.
a73078b7
HL
2162 */
2163static int txp_generate_stream_frames(OSSL_QUIC_TX_PACKETISER *txp,
faebafda 2164 struct txp_pkt *pkt,
a73078b7
HL
2165 uint64_t id,
2166 QUIC_SSTREAM *sstream,
2167 QUIC_TXFC *stream_txfc,
2168 QUIC_STREAM *next_stream,
091f532e
HL
2169 int *have_ack_eliciting,
2170 int *packet_full,
e57bf6b3
MC
2171 uint64_t *new_credit_consumed,
2172 uint64_t conn_consumed)
a73078b7
HL
2173{
2174 int rc = 0;
2175 struct chunk_info chunks[2] = {0};
faebafda
HL
2176 const uint32_t enc_level = pkt->h.enc_level;
2177 QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
2178 struct tx_helper *h = &pkt->h;
a73078b7
HL
2179 OSSL_QUIC_FRAME_STREAM *shdr;
2180 WPACKET *wpkt;
2181 QUIC_TXPIM_CHUNK chunk;
2182 size_t i, j, space_left;
3bef14c5 2183 int can_fill_payload, use_explicit_len;
a73078b7 2184 int could_have_following_chunk;
05f97354 2185 uint64_t orig_len;
a73078b7
HL
2186 uint64_t hdr_len_implicit, payload_len_implicit;
2187 uint64_t hdr_len_explicit, payload_len_explicit;
2188 uint64_t fc_swm, fc_new_hwm;
2189
2190 fc_swm = ossl_quic_txfc_get_swm(stream_txfc);
2191 fc_new_hwm = fc_swm;
2192
2193 /*
2194 * Load the first two chunks if any offered by the send stream. We retrieve
2195 * the next chunk in advance so we can determine if we need to send any more
2196 * chunks from the same stream after this one, which is needed when
2197 * determining when we can use an implicit length in a STREAM frame.
2198 */
2199 for (i = 0; i < 2; ++i) {
e57bf6b3
MC
2200 if (!txp_plan_stream_chunk(txp, h, sstream, stream_txfc, i, &chunks[i],
2201 conn_consumed))
a73078b7
HL
2202 goto err;
2203
2204 if (i == 0 && !chunks[i].valid) {
2205 /* No chunks, nothing to do. */
a73078b7
HL
2206 rc = 1;
2207 goto err;
2208 }
2209 }
2210
2211 for (i = 0;; ++i) {
2212 space_left = tx_helper_get_space_left(h);
2213
cf06f347
HL
2214 if (!chunks[i % 2].valid) {
2215 /* Out of chunks; we're done. */
a73078b7
HL
2216 rc = 1;
2217 goto err;
2218 }
2219
cf06f347
HL
2220 if (space_left < MIN_FRAME_SIZE_STREAM) {
2221 *packet_full = 1;
a73078b7
HL
2222 rc = 1;
2223 goto err;
2224 }
2225
2226 if (!ossl_assert(!h->done_implicit))
2227 /*
2228 * Logic below should have ensured we didn't append an
2229 * implicit-length unless we filled the packet or didn't have
2230 * another stream to handle, so this should not be possible.
2231 */
2232 goto err;
2233
2234 shdr = &chunks[i % 2].shdr;
e718b248 2235 orig_len = chunks[i % 2].orig_len;
a73078b7
HL
2236 if (i > 0)
2237 /* Load next chunk for lookahead. */
2238 if (!txp_plan_stream_chunk(txp, h, sstream, stream_txfc, i + 1,
e57bf6b3 2239 &chunks[(i + 1) % 2], conn_consumed))
a73078b7
HL
2240 goto err;
2241
2242 /*
2243 * Find best fit (header length, payload length) combination for if we
2244 * use an implicit length.
2245 */
2246 shdr->has_explicit_len = 0;
2247 hdr_len_implicit = payload_len_implicit = 0;
2248 if (!determine_stream_len(h, shdr, space_left,
cf06f347 2249 &hdr_len_implicit, &payload_len_implicit)) {
a73078b7
HL
2250 *packet_full = 1;
2251 rc = 1;
2252 goto err; /* can't fit anything */
2253 }
2254
a73078b7
HL
2255 /*
2256 * If there is a next stream, we don't use the implicit length so we can
2257 * add more STREAM frames after this one, unless there is enough data
2258 * for this STREAM frame to fill the packet.
2259 */
2260 can_fill_payload = (hdr_len_implicit + payload_len_implicit
2261 >= space_left);
2262
2263 /*
2264 * Is there is a stream after this one, or another chunk pending
2265 * transmission in this stream?
2266 */
2267 could_have_following_chunk
2268 = (next_stream != NULL || chunks[(i + 1) % 2].valid);
2269
2270 /* Choose between explicit or implicit length representations. */
2271 use_explicit_len = !((can_fill_payload || !could_have_following_chunk)
3bef14c5 2272 && !pkt->force_pad);
a73078b7
HL
2273
2274 if (use_explicit_len) {
2275 /*
2276 * Find best fit (header length, payload length) combination for if
2277 * we use an explicit length.
2278 */
2279 shdr->has_explicit_len = 1;
2280 hdr_len_explicit = payload_len_explicit = 0;
2281 if (!determine_stream_len(h, shdr, space_left,
cf06f347 2282 &hdr_len_explicit, &payload_len_explicit)) {
a73078b7
HL
2283 *packet_full = 1;
2284 rc = 1;
2285 goto err; /* can't fit anything */
2286 }
2287
2288 shdr->len = payload_len_explicit;
2289 } else {
7fe30104 2290 *packet_full = 1;
a73078b7
HL
2291 shdr->has_explicit_len = 0;
2292 shdr->len = payload_len_implicit;
2293 }
2294
cf06f347
HL
2295 /* If this is a FIN, don't keep filling the packet with more FINs. */
2296 if (shdr->is_fin)
2297 chunks[(i + 1) % 2].valid = 0;
2298
553122cd
HL
2299 /*
2300 * We are now committed to our length (shdr->len can't change).
2301 * If we truncated the chunk, clear the FIN bit.
2302 */
2303 if (shdr->len < orig_len)
2304 shdr->is_fin = 0;
2305
a73078b7
HL
2306 /* Truncate IOVs to match our chosen length. */
2307 ossl_quic_sstream_adjust_iov((size_t)shdr->len, chunks[i % 2].iov,
2308 chunks[i % 2].num_stream_iovec);
2309
2310 /*
2311 * Ensure we have enough iovecs allocated (1 for the header, up to 2 for
ad31628c 2312 * the stream data.)
a73078b7 2313 */
faebafda 2314 if (!txp_el_ensure_iovec(&txp->el[enc_level], h->num_iovec + 3))
a73078b7
HL
2315 goto err; /* alloc error */
2316
2317 /* Encode the header. */
2318 wpkt = tx_helper_begin(h);
2319 if (wpkt == NULL)
2320 goto err; /* alloc error */
2321
2322 shdr->stream_id = id;
2323 if (!ossl_assert(ossl_quic_wire_encode_frame_stream_hdr(wpkt, shdr))) {
2324 /* (Should not be possible.) */
2325 tx_helper_rollback(h);
2326 *packet_full = 1;
2327 rc = 1;
2328 goto err; /* can't fit */
2329 }
2330
2331 if (!tx_helper_commit(h))
2332 goto err; /* alloc error */
2333
2334 /* Add payload iovecs to the helper (infallible). */
2335 for (j = 0; j < chunks[i % 2].num_stream_iovec; ++j)
2336 tx_helper_append_iovec(h, chunks[i % 2].iov[j].buf,
2337 chunks[i % 2].iov[j].buf_len);
2338
2339 *have_ack_eliciting = 1;
2340 tx_helper_unrestrict(h); /* no longer need PING */
2341 if (!shdr->has_explicit_len)
2342 h->done_implicit = 1;
2343
2344 /* Log new TXFC credit which was consumed. */
2345 if (shdr->len > 0 && shdr->offset + shdr->len > fc_new_hwm)
2346 fc_new_hwm = shdr->offset + shdr->len;
2347
2348 /* Log chunk to TXPIM. */
2349 chunk.stream_id = shdr->stream_id;
2350 chunk.start = shdr->offset;
2351 chunk.end = shdr->offset + shdr->len - 1;
2352 chunk.has_fin = shdr->is_fin;
2353 chunk.has_stop_sending = 0;
2354 chunk.has_reset_stream = 0;
2355 if (!ossl_quic_txpim_pkt_append_chunk(tpkt, &chunk))
2356 goto err; /* alloc error */
05f97354
HL
2357
2358 if (shdr->len < orig_len) {
2359 /*
2360 * If we did not serialize all of this chunk we definitely do not
e718b248 2361 * want to try the next chunk
05f97354
HL
2362 */
2363 rc = 1;
2364 goto err;
2365 }
a73078b7
HL
2366 }
2367
2368err:
2369 *new_credit_consumed = fc_new_hwm - fc_swm;
2370 return rc;
2371}
2372
2373static void txp_enlink_tmp(QUIC_STREAM **tmp_head, QUIC_STREAM *stream)
2374{
2375 stream->txp_next = *tmp_head;
2376 *tmp_head = stream;
2377}
2378
2379static int txp_generate_stream_related(OSSL_QUIC_TX_PACKETISER *txp,
faebafda 2380 struct txp_pkt *pkt,
091f532e 2381 int *have_ack_eliciting,
a73078b7
HL
2382 QUIC_STREAM **tmp_head)
2383{
2384 QUIC_STREAM_ITER it;
a73078b7
HL
2385 WPACKET *wpkt;
2386 uint64_t cwm;
2387 QUIC_STREAM *stream, *snext;
faebafda 2388 struct tx_helper *h = &pkt->h;
e57bf6b3 2389 uint64_t conn_consumed = 0;
a73078b7
HL
2390
2391 for (ossl_quic_stream_iter_init(&it, txp->args.qsm, 1);
2392 it.stream != NULL;) {
2393
2394 stream = it.stream;
2395 ossl_quic_stream_iter_next(&it);
2396 snext = it.stream;
2397
2398 stream->txp_sent_fc = 0;
2399 stream->txp_sent_stop_sending = 0;
2400 stream->txp_sent_reset_stream = 0;
a73078b7
HL
2401 stream->txp_blocked = 0;
2402 stream->txp_txfc_new_credit_consumed = 0;
2403
a73078b7
HL
2404 /* Stream Abort Frames (STOP_SENDING, RESET_STREAM) */
2405 if (stream->want_stop_sending) {
2406 OSSL_QUIC_FRAME_STOP_SENDING f;
2407
2408 wpkt = tx_helper_begin(h);
2409 if (wpkt == NULL)
2410 return 0; /* alloc error */
2411
2412 f.stream_id = stream->id;
2413 f.app_error_code = stream->stop_sending_aec;
2414 if (!ossl_quic_wire_encode_frame_stop_sending(wpkt, &f)) {
2415 tx_helper_rollback(h); /* can't fit */
2416 txp_enlink_tmp(tmp_head, stream);
2417 break;
2418 }
2419
2420 if (!tx_helper_commit(h))
2421 return 0; /* alloc error */
2422
2423 *have_ack_eliciting = 1;
2424 tx_helper_unrestrict(h); /* no longer need PING */
2425 stream->txp_sent_stop_sending = 1;
2426 }
2427
2428 if (stream->want_reset_stream) {
2429 OSSL_QUIC_FRAME_RESET_STREAM f;
2430
96b7df60
HL
2431 if (!ossl_assert(stream->send_state == QUIC_SSTREAM_STATE_RESET_SENT))
2432 return 0;
01715f2b 2433
a73078b7
HL
2434 wpkt = tx_helper_begin(h);
2435 if (wpkt == NULL)
2436 return 0; /* alloc error */
2437
2438 f.stream_id = stream->id;
2439 f.app_error_code = stream->reset_stream_aec;
28d0e35c
HL
2440 if (!ossl_quic_stream_send_get_final_size(stream, &f.final_size))
2441 return 0; /* should not be possible */
2442
a73078b7
HL
2443 if (!ossl_quic_wire_encode_frame_reset_stream(wpkt, &f)) {
2444 tx_helper_rollback(h); /* can't fit */
2445 txp_enlink_tmp(tmp_head, stream);
2446 break;
2447 }
2448
2449 if (!tx_helper_commit(h))
2450 return 0; /* alloc error */
2451
2452 *have_ack_eliciting = 1;
2453 tx_helper_unrestrict(h); /* no longer need PING */
2454 stream->txp_sent_reset_stream = 1;
7e3fa44f
HL
2455
2456 /*
2457 * The final size of the stream as indicated by RESET_STREAM is used
2458 * to ensure a consistent view of flow control state by both
2459 * parties; if we happen to send a RESET_STREAM that consumes more
2460 * flow control credit, make sure we account for that.
2461 */
96b7df60
HL
2462 if (!ossl_assert(f.final_size <= ossl_quic_txfc_get_swm(&stream->txfc)))
2463 return 0;
7e3fa44f
HL
2464
2465 stream->txp_txfc_new_credit_consumed
2466 = f.final_size - ossl_quic_txfc_get_swm(&stream->txfc);
a73078b7
HL
2467 }
2468
22f21fbd
HL
2469 /*
2470 * Stream Flow Control Frames (MAX_STREAM_DATA)
2471 *
2472 * RFC 9000 s. 13.3: "An endpoint SHOULD stop sending MAX_STREAM_DATA
2473 * frames when the receiving part of the stream enters a "Size Known" or
2474 * "Reset Recvd" state." -- In practice, RECV is the only state
2475 * in which it makes sense to generate more MAX_STREAM_DATA frames.
2476 */
2477 if (stream->recv_state == QUIC_RSTREAM_STATE_RECV
a73078b7
HL
2478 && (stream->want_max_stream_data
2479 || ossl_quic_rxfc_has_cwm_changed(&stream->rxfc, 0))) {
2480
2481 wpkt = tx_helper_begin(h);
2482 if (wpkt == NULL)
2483 return 0; /* alloc error */
2484
2485 cwm = ossl_quic_rxfc_get_cwm(&stream->rxfc);
2486
2487 if (!ossl_quic_wire_encode_frame_max_stream_data(wpkt, stream->id,
2488 cwm)) {
2489 tx_helper_rollback(h); /* can't fit */
2490 txp_enlink_tmp(tmp_head, stream);
2491 break;
2492 }
2493
2494 if (!tx_helper_commit(h))
2495 return 0; /* alloc error */
2496
2497 *have_ack_eliciting = 1;
2498 tx_helper_unrestrict(h); /* no longer need PING */
2499 stream->txp_sent_fc = 1;
2500 }
2501
01715f2b
HL
2502 /*
2503 * Stream Data Frames (STREAM)
2504 *
2505 * RFC 9000 s. 3.3: A sender MUST NOT send a STREAM [...] frame for a
2506 * stream in the "Reset Sent" state [or any terminal state]. We don't
1d547f8f 2507 * send any more STREAM frames if we are sending, have sent, or are
01715f2b
HL
2508 * planning to send, RESET_STREAM. The other terminal state is Data
2509 * Recvd, but txp_generate_stream_frames() is guaranteed to generate
2510 * nothing in this case.
2511 */
2512 if (ossl_quic_stream_has_send_buffer(stream)
2513 && !ossl_quic_stream_send_is_reset(stream)) {
26652896 2514 int packet_full = 0;
a73078b7 2515
96b7df60
HL
2516 if (!ossl_assert(!stream->want_reset_stream))
2517 return 0;
7e3fa44f 2518
faebafda 2519 if (!txp_generate_stream_frames(txp, pkt,
a73078b7
HL
2520 stream->id, stream->sstream,
2521 &stream->txfc,
3bef14c5 2522 snext,
a73078b7
HL
2523 have_ack_eliciting,
2524 &packet_full,
e57bf6b3
MC
2525 &stream->txp_txfc_new_credit_consumed,
2526 conn_consumed)) {
a73078b7
HL
2527 /* Fatal error (allocation, etc.) */
2528 txp_enlink_tmp(tmp_head, stream);
2529 return 0;
2530 }
e57bf6b3 2531 conn_consumed += stream->txp_txfc_new_credit_consumed;
a73078b7 2532
a73078b7
HL
2533 if (packet_full) {
2534 txp_enlink_tmp(tmp_head, stream);
2535 break;
2536 }
2537 }
2538
2539 txp_enlink_tmp(tmp_head, stream);
2540 }
2541
2542 return 1;
2543}
2544
faebafda
HL
2545static int txp_generate_for_el(OSSL_QUIC_TX_PACKETISER *txp,
2546 struct txp_pkt *pkt,
2547 int chosen_for_conn_close)
a73078b7
HL
2548{
2549 int rc = TXP_ERR_SUCCESS;
faebafda
HL
2550 const uint32_t enc_level = pkt->h.enc_level;
2551 const uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
2552 int have_ack_eliciting = 0, done_pre_token = 0;
2553 const struct archetype_data a = pkt->geom.adata;
178c104d
HL
2554 /*
2555 * Cleared if we encode any non-ACK-eliciting frame type which rules out the
2556 * packet being a non-inflight frame. This means any non-ACK ACK-eliciting
2557 * frame, even PADDING frames. ACK eliciting frames always cause a packet to
2558 * become ineligible for non-inflight treatment so it is not necessary to
2559 * clear this in cases where have_ack_eliciting is set, as it is ignored in
2560 * that case.
2561 */
2562 int can_be_non_inflight = 1;
a73078b7
HL
2563 QUIC_CFQ_ITEM *cfq_item;
2564 QUIC_TXPIM_PKT *tpkt = NULL;
faebafda 2565 struct tx_helper *h = &pkt->h;
a73078b7
HL
2566
2567 /* Maximum PN reached? */
b65b0d4e 2568 if (!ossl_quic_pn_valid(txp->next_pn[pn_space]))
a73078b7
HL
2569 goto fatal_err;
2570
faebafda 2571 if (!ossl_assert(pkt->tpkt == NULL))
a73078b7
HL
2572 goto fatal_err;
2573
faebafda 2574 if ((pkt->tpkt = tpkt = ossl_quic_txpim_pkt_alloc(txp->args.txpim)) == NULL)
a73078b7
HL
2575 goto fatal_err;
2576
a73078b7
HL
2577 /*
2578 * Frame Serialization
2579 * ===================
2580 *
2581 * We now serialize frames into the packet in descending order of priority.
2582 */
2583
2584 /* HANDSHAKE_DONE (Regenerate) */
2585 if (a.allow_handshake_done && txp->want_handshake_done
faebafda
HL
2586 && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_HANDSHAKE_DONE) {
2587 WPACKET *wpkt = tx_helper_begin(h);
a73078b7
HL
2588
2589 if (wpkt == NULL)
2590 goto fatal_err;
2591
2592 if (ossl_quic_wire_encode_frame_handshake_done(wpkt)) {
2593 tpkt->had_handshake_done_frame = 1;
2594 have_ack_eliciting = 1;
2595
faebafda 2596 if (!tx_helper_commit(h))
a73078b7
HL
2597 goto fatal_err;
2598
faebafda 2599 tx_helper_unrestrict(h); /* no longer need PING */
a73078b7 2600 } else {
faebafda 2601 tx_helper_rollback(h);
a73078b7
HL
2602 }
2603 }
2604
2605 /* MAX_DATA (Regenerate) */
2606 if (a.allow_conn_fc
2607 && (txp->want_max_data
2608 || ossl_quic_rxfc_has_cwm_changed(txp->args.conn_rxfc, 0))
faebafda
HL
2609 && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_MAX_DATA) {
2610 WPACKET *wpkt = tx_helper_begin(h);
a73078b7
HL
2611 uint64_t cwm = ossl_quic_rxfc_get_cwm(txp->args.conn_rxfc);
2612
2613 if (wpkt == NULL)
2614 goto fatal_err;
2615
2616 if (ossl_quic_wire_encode_frame_max_data(wpkt, cwm)) {
2617 tpkt->had_max_data_frame = 1;
2618 have_ack_eliciting = 1;
2619
faebafda 2620 if (!tx_helper_commit(h))
a73078b7
HL
2621 goto fatal_err;
2622
faebafda 2623 tx_helper_unrestrict(h); /* no longer need PING */
a73078b7 2624 } else {
faebafda 2625 tx_helper_rollback(h);
a73078b7
HL
2626 }
2627 }
2628
2629 /* MAX_STREAMS_BIDI (Regenerate) */
a73078b7 2630 if (a.allow_conn_fc
a6b6ea17
HL
2631 && (txp->want_max_streams_bidi
2632 || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_bidi_rxfc, 0))
faebafda
HL
2633 && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_MAX_STREAMS_BIDI) {
2634 WPACKET *wpkt = tx_helper_begin(h);
a6b6ea17
HL
2635 uint64_t max_streams
2636 = ossl_quic_rxfc_get_cwm(txp->args.max_streams_bidi_rxfc);
a73078b7
HL
2637
2638 if (wpkt == NULL)
2639 goto fatal_err;
2640
2641 if (ossl_quic_wire_encode_frame_max_streams(wpkt, /*is_uni=*/0,
2642 max_streams)) {
2643 tpkt->had_max_streams_bidi_frame = 1;
2644 have_ack_eliciting = 1;
2645
faebafda 2646 if (!tx_helper_commit(h))
a73078b7
HL
2647 goto fatal_err;
2648
faebafda 2649 tx_helper_unrestrict(h); /* no longer need PING */
a73078b7 2650 } else {
faebafda 2651 tx_helper_rollback(h);
a73078b7
HL
2652 }
2653 }
2654
2655 /* MAX_STREAMS_UNI (Regenerate) */
2656 if (a.allow_conn_fc
a6b6ea17
HL
2657 && (txp->want_max_streams_uni
2658 || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_uni_rxfc, 0))
faebafda
HL
2659 && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_MAX_STREAMS_UNI) {
2660 WPACKET *wpkt = tx_helper_begin(h);
a6b6ea17
HL
2661 uint64_t max_streams
2662 = ossl_quic_rxfc_get_cwm(txp->args.max_streams_uni_rxfc);
a73078b7
HL
2663
2664 if (wpkt == NULL)
2665 goto fatal_err;
2666
2667 if (ossl_quic_wire_encode_frame_max_streams(wpkt, /*is_uni=*/1,
2668 max_streams)) {
2669 tpkt->had_max_streams_uni_frame = 1;
2670 have_ack_eliciting = 1;
2671
faebafda 2672 if (!tx_helper_commit(h))
a73078b7
HL
2673 goto fatal_err;
2674
faebafda 2675 tx_helper_unrestrict(h); /* no longer need PING */
a73078b7 2676 } else {
faebafda 2677 tx_helper_rollback(h);
a73078b7
HL
2678 }
2679 }
2680
2681 /* GCR Frames */
2682 for (cfq_item = ossl_quic_cfq_get_priority_head(txp->args.cfq, pn_space);
2683 cfq_item != NULL;
2684 cfq_item = ossl_quic_cfq_item_get_priority_next(cfq_item, pn_space)) {
2685 uint64_t frame_type = ossl_quic_cfq_item_get_frame_type(cfq_item);
2686 const unsigned char *encoded = ossl_quic_cfq_item_get_encoded(cfq_item);
2687 size_t encoded_len = ossl_quic_cfq_item_get_encoded_len(cfq_item);
2688
2689 switch (frame_type) {
2690 case OSSL_QUIC_FRAME_TYPE_NEW_CONN_ID:
2691 if (!a.allow_new_conn_id)
2692 continue;
2693 break;
2694 case OSSL_QUIC_FRAME_TYPE_RETIRE_CONN_ID:
2695 if (!a.allow_retire_conn_id)
2696 continue;
2697 break;
2698 case OSSL_QUIC_FRAME_TYPE_NEW_TOKEN:
2699 if (!a.allow_new_token)
2700 continue;
2701
2702 /*
2703 * NEW_TOKEN frames are handled via GCR, but some
2704 * Regenerate-strategy frames should come before them (namely
2705 * ACK, CONNECTION_CLOSE, PATH_CHALLENGE and PATH_RESPONSE). If
2706 * we find a NEW_TOKEN frame, do these now. If there are no
2707 * NEW_TOKEN frames in the GCR queue we will handle these below.
2708 */
2709 if (!done_pre_token)
faebafda 2710 if (txp_generate_pre_token(txp, pkt,
178c104d
HL
2711 chosen_for_conn_close,
2712 &can_be_non_inflight))
a73078b7
HL
2713 done_pre_token = 1;
2714
c5cb85b6
HL
2715 break;
2716 case OSSL_QUIC_FRAME_TYPE_PATH_RESPONSE:
2717 if (!a.allow_path_response)
2718 continue;
2719
2720 /*
2721 * RFC 9000 s. 8.2.2: An endpoint MUST expand datagrams that
2722 * contain a PATH_RESPONSE frame to at least the smallest
2723 * allowed maximum datagram size of 1200 bytes.
2724 */
2725 pkt->force_pad = 1;
a73078b7
HL
2726 break;
2727 default:
2728 if (!a.allow_cfq_other)
2729 continue;
2730 break;
2731 }
2732
2733 /*
2734 * If the frame is too big, don't try to schedule any more GCR frames in
2735 * this packet rather than sending subsequent ones out of order.
2736 */
faebafda 2737 if (encoded_len > tx_helper_get_space_left(h))
a73078b7
HL
2738 break;
2739
faebafda 2740 if (!tx_helper_append_iovec(h, encoded, encoded_len))
a73078b7
HL
2741 goto fatal_err;
2742
2743 ossl_quic_txpim_pkt_add_cfq_item(tpkt, cfq_item);
2744
2745 if (ossl_quic_frame_type_is_ack_eliciting(frame_type)) {
2746 have_ack_eliciting = 1;
faebafda 2747 tx_helper_unrestrict(h); /* no longer need PING */
a73078b7
HL
2748 }
2749 }
2750
2751 /*
2752 * If we didn't generate ACK, CONNECTION_CLOSE, PATH_CHALLENGE or
2753 * PATH_RESPONSE (as desired) before, do so now.
2754 */
2755 if (!done_pre_token)
faebafda 2756 if (txp_generate_pre_token(txp, pkt,
178c104d
HL
2757 chosen_for_conn_close,
2758 &can_be_non_inflight))
a73078b7
HL
2759 done_pre_token = 1;
2760
2761 /* CRYPTO Frames */
2762 if (a.allow_crypto)
faebafda 2763 if (!txp_generate_crypto_frames(txp, pkt, &have_ack_eliciting))
a73078b7
HL
2764 goto fatal_err;
2765
2766 /* Stream-specific frames */
cda88baf 2767 if (a.allow_stream_rel && txp->handshake_complete)
3bef14c5 2768 if (!txp_generate_stream_related(txp, pkt,
a73078b7 2769 &have_ack_eliciting,
faebafda 2770 &pkt->stream_head))
a73078b7
HL
2771 goto fatal_err;
2772
2773 /* PING */
faebafda 2774 tx_helper_unrestrict(h);
a73078b7 2775
1e2e683a
HL
2776 if ((a.require_ack_eliciting
2777 || (txp->force_ack_eliciting & (1UL << pn_space)) != 0)
2778 && !have_ack_eliciting && a.allow_ping) {
a73078b7
HL
2779 WPACKET *wpkt;
2780
faebafda 2781 wpkt = tx_helper_begin(h);
a73078b7
HL
2782 if (wpkt == NULL)
2783 goto fatal_err;
2784
2785 if (!ossl_quic_wire_encode_frame_ping(wpkt)
faebafda 2786 || !tx_helper_commit(h))
a73078b7
HL
2787 /*
2788 * We treat a request to be ACK-eliciting as a requirement, so this
2789 * is an error.
2790 */
2791 goto fatal_err;
2792
2793 have_ack_eliciting = 1;
2794 }
2795
3bef14c5 2796 /* PADDING is added by ossl_quic_tx_packetiser_generate(). */
a73078b7
HL
2797
2798 /*
faebafda
HL
2799 * ACKM Data
2800 * =========
a73078b7 2801 */
178c104d
HL
2802 if (have_ack_eliciting)
2803 can_be_non_inflight = 0;
2804
a73078b7 2805 /* ACKM Data */
faebafda 2806 tpkt->ackm_pkt.num_bytes = h->bytes_appended + pkt->geom.pkt_overhead;
a73078b7
HL
2807 tpkt->ackm_pkt.pkt_num = txp->next_pn[pn_space];
2808 /* largest_acked is set in txp_generate_pre_token */
2809 tpkt->ackm_pkt.pkt_space = pn_space;
178c104d 2810 tpkt->ackm_pkt.is_inflight = !can_be_non_inflight;
a73078b7
HL
2811 tpkt->ackm_pkt.is_ack_eliciting = have_ack_eliciting;
2812 tpkt->ackm_pkt.is_pto_probe = 0;
2813 tpkt->ackm_pkt.is_mtu_probe = 0;
b98c38d4 2814 tpkt->ackm_pkt.time = txp->args.now(txp->args.now_arg);
a73078b7 2815
faebafda
HL
2816 /* Done. */
2817 return rc;
2818
2819fatal_err:
2820 /*
2821 * Handler for fatal errors, i.e. errors causing us to abort the entire
2822 * packet rather than just one frame. Examples of such errors include
2823 * allocation errors.
2824 */
2825 if (tpkt != NULL) {
2826 ossl_quic_txpim_pkt_release(txp->args.txpim, tpkt);
2827 pkt->tpkt = NULL;
2828 }
2829 return TXP_ERR_INTERNAL;
2830}
2831
2832/*
2833 * Commits and queues a packet for transmission. There is no backing out after
2834 * this.
2835 *
2836 * This:
2837 *
2838 * - Sends the packet to the QTX for encryption and transmission;
2839 *
2840 * - Records the packet as having been transmitted in FIFM. ACKM is informed,
2841 * etc. and the TXPIM record is filed.
2842 *
2843 * - Informs various subsystems of frames that were sent and clears frame
2844 * wanted flags so that we do not generate the same frames again.
2845 *
2846 * Assumptions:
2847 *
2848 * - pkt is a txp_pkt for the correct EL;
2849 *
2850 * - pkt->tpkt is valid;
2851 *
2852 * - pkt->tpkt->ackm_pkt has been fully filled in;
2853 *
2854 * - Stream chunk records have been appended to pkt->tpkt for STREAM and
2855 * CRYPTO frames, but not for RESET_STREAM or STOP_SENDING frames;
2856 *
2857 * - The chosen stream list for the packet can be fully walked from
2858 * pkt->stream_head using stream->txp_next;
2859 *
2860 * - pkt->has_ack_eliciting is set correctly.
2861 *
2862 */
2863static int txp_pkt_commit(OSSL_QUIC_TX_PACKETISER *txp,
2864 struct txp_pkt *pkt,
6a2b70e2
HL
2865 uint32_t archetype,
2866 int *txpim_pkt_reffed)
faebafda
HL
2867{
2868 int rc = 1;
2869 uint32_t enc_level = pkt->h.enc_level;
2870 uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
2871 QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
2872 QUIC_STREAM *stream;
2873 OSSL_QTX_PKT txpkt;
2874 struct archetype_data a;
2875
6a2b70e2
HL
2876 *txpim_pkt_reffed = 0;
2877
faebafda
HL
2878 /* Cannot send a packet with an empty payload. */
2879 if (pkt->h.bytes_appended == 0)
2880 return 0;
2881
2882 if (!txp_get_archetype_data(enc_level, archetype, &a))
2883 return 0;
2884
a73078b7 2885 /* Packet Information for QTX */
faebafda
HL
2886 txpkt.hdr = &pkt->phdr;
2887 txpkt.iovec = txp->el[enc_level].iovec;
2888 txpkt.num_iovec = pkt->h.num_iovec;
2889 txpkt.local = NULL;
2890 txpkt.peer = BIO_ADDR_family(&txp->args.peer) == AF_UNSPEC
a73078b7 2891 ? NULL : &txp->args.peer;
faebafda
HL
2892 txpkt.pn = txp->next_pn[pn_space];
2893 txpkt.flags = OSSL_QTX_PKT_FLAG_COALESCE; /* always try to coalesce */
a73078b7
HL
2894
2895 /* Generate TXPIM chunks representing STOP_SENDING and RESET_STREAM frames. */
faebafda 2896 for (stream = pkt->stream_head; stream != NULL; stream = stream->txp_next)
a73078b7 2897 if (stream->txp_sent_stop_sending || stream->txp_sent_reset_stream) {
faebafda 2898 /* Log STOP_SENDING/RESET_STREAM chunk to TXPIM. */
a73078b7
HL
2899 QUIC_TXPIM_CHUNK chunk;
2900
2901 chunk.stream_id = stream->id;
2902 chunk.start = UINT64_MAX;
2903 chunk.end = 0;
2904 chunk.has_fin = 0;
2905 chunk.has_stop_sending = stream->txp_sent_stop_sending;
2906 chunk.has_reset_stream = stream->txp_sent_reset_stream;
2907 if (!ossl_quic_txpim_pkt_append_chunk(tpkt, &chunk))
2908 return 0; /* alloc error */
2909 }
2910
2911 /* Dispatch to FIFD. */
2912 if (!ossl_quic_fifd_pkt_commit(&txp->fifd, tpkt))
faebafda 2913 return 0;
a73078b7 2914
faebafda 2915 /*
ed75eb32
HL
2916 * Transmission and Post-Packet Generation Bookkeeping
2917 * ===================================================
faebafda 2918 *
ed75eb32
HL
2919 * No backing out anymore - at this point the ACKM has recorded the packet
2920 * as having been sent, so we need to increment our next PN counter, or
2921 * the ACKM will complain when we try to record a duplicate packet with
2922 * the same PN later. At this point actually sending the packet may still
2923 * fail. In this unlikely event it will simply be handled as though it
2924 * were a lost packet.
faebafda 2925 */
a73078b7 2926 ++txp->next_pn[pn_space];
6a2b70e2 2927 *txpim_pkt_reffed = 1;
a73078b7 2928
ed75eb32
HL
2929 /* Send the packet. */
2930 if (!ossl_qtx_write_pkt(txp->args.qtx, &txpkt))
2931 return 0;
2932
a73078b7
HL
2933 /*
2934 * Record FC and stream abort frames as sent; deactivate streams which no
2935 * longer have anything to do.
2936 */
faebafda 2937 for (stream = pkt->stream_head; stream != NULL; stream = stream->txp_next) {
a73078b7
HL
2938 if (stream->txp_sent_fc) {
2939 stream->want_max_stream_data = 0;
2940 ossl_quic_rxfc_has_cwm_changed(&stream->rxfc, 1);
2941 }
2942
2943 if (stream->txp_sent_stop_sending)
2944 stream->want_stop_sending = 0;
2945
2946 if (stream->txp_sent_reset_stream)
2947 stream->want_reset_stream = 0;
2948
2949 if (stream->txp_txfc_new_credit_consumed > 0) {
2950 if (!ossl_assert(ossl_quic_txfc_consume_credit(&stream->txfc,
2951 stream->txp_txfc_new_credit_consumed)))
2952 /*
2953 * Should not be possible, but we should continue with our
2954 * bookkeeping as we have already committed the packet to the
2955 * FIFD. Just change the value we return.
2956 */
faebafda 2957 rc = 0;
a73078b7
HL
2958
2959 stream->txp_txfc_new_credit_consumed = 0;
2960 }
2961
2962 /*
2963 * If we no longer need to generate any flow control (MAX_STREAM_DATA),
2964 * STOP_SENDING or RESET_STREAM frames, nor any STREAM frames (because
2965 * the stream is drained of data or TXFC-blocked), we can mark the
2966 * stream as inactive.
2967 */
2968 ossl_quic_stream_map_update_state(txp->args.qsm, stream);
2969
26652896
HL
2970 if (ossl_quic_stream_has_send_buffer(stream)
2971 && !ossl_quic_sstream_has_pending(stream->sstream)
2972 && ossl_quic_sstream_get_final_size(stream->sstream, NULL))
c068f4d1
HL
2973 /*
2974 * Transition to DATA_SENT if stream has a final size and we have
2975 * sent all data.
2976 */
26652896 2977 ossl_quic_stream_map_notify_all_data_sent(txp->args.qsm, stream);
a73078b7
HL
2978 }
2979
2980 /* We have now sent the packet, so update state accordingly. */
faebafda 2981 if (tpkt->ackm_pkt.is_ack_eliciting)
a73078b7
HL
2982 txp->force_ack_eliciting &= ~(1UL << pn_space);
2983
2984 if (tpkt->had_handshake_done_frame)
2985 txp->want_handshake_done = 0;
2986
2987 if (tpkt->had_max_data_frame) {
2988 txp->want_max_data = 0;
2989 ossl_quic_rxfc_has_cwm_changed(txp->args.conn_rxfc, 1);
2990 }
2991
a6b6ea17 2992 if (tpkt->had_max_streams_bidi_frame) {
a73078b7 2993 txp->want_max_streams_bidi = 0;
a6b6ea17
HL
2994 ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_bidi_rxfc, 1);
2995 }
a73078b7 2996
a6b6ea17 2997 if (tpkt->had_max_streams_uni_frame) {
a73078b7 2998 txp->want_max_streams_uni = 0;
a6b6ea17
HL
2999 ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_uni_rxfc, 1);
3000 }
a73078b7
HL
3001
3002 if (tpkt->had_ack_frame)
3003 txp->want_ack &= ~(1UL << pn_space);
3004
0b31072e
MC
3005 if (tpkt->had_conn_close)
3006 txp->want_conn_close = 0;
3007
fee8f48e
HL
3008 /*
3009 * Decrement probe request counts if we have sent a packet that meets
3010 * the requirement of a probe, namely being ACK-eliciting.
3011 */
faebafda
HL
3012 if (tpkt->ackm_pkt.is_ack_eliciting) {
3013 OSSL_ACKM_PROBE_INFO *probe_info
3014 = ossl_ackm_get0_probe_request(txp->args.ackm);
3015
fee8f48e
HL
3016 if (enc_level == QUIC_ENC_LEVEL_INITIAL
3017 && probe_info->anti_deadlock_initial > 0)
3018 --probe_info->anti_deadlock_initial;
3019
3020 if (enc_level == QUIC_ENC_LEVEL_HANDSHAKE
3021 && probe_info->anti_deadlock_handshake > 0)
3022 --probe_info->anti_deadlock_handshake;
3023
3024 if (a.allow_force_ack_eliciting /* (i.e., not for 0-RTT) */
3025 && probe_info->pto[pn_space] > 0)
3026 --probe_info->pto[pn_space];
3027 }
3028
a73078b7 3029 return rc;
a73078b7
HL
3030}
3031
3032/* Ensure the iovec array is at least num elements long. */
faebafda 3033static int txp_el_ensure_iovec(struct txp_el *el, size_t num)
a73078b7
HL
3034{
3035 OSSL_QTX_IOVEC *iovec;
3036
faebafda 3037 if (el->alloc_iovec >= num)
a73078b7
HL
3038 return 1;
3039
faebafda 3040 num = el->alloc_iovec != 0 ? el->alloc_iovec * 2 : 8;
a73078b7 3041
faebafda 3042 iovec = OPENSSL_realloc(el->iovec, sizeof(OSSL_QTX_IOVEC) * num);
a73078b7
HL
3043 if (iovec == NULL)
3044 return 0;
3045
faebafda
HL
3046 el->iovec = iovec;
3047 el->alloc_iovec = num;
a73078b7
HL
3048 return 1;
3049}
3050
3051int ossl_quic_tx_packetiser_schedule_conn_close(OSSL_QUIC_TX_PACKETISER *txp,
3052 const OSSL_QUIC_FRAME_CONN_CLOSE *f)
3053{
3054 char *reason = NULL;
3055 size_t reason_len = f->reason_len;
3056 size_t max_reason_len = txp_get_mdpl(txp) / 2;
3057
3058 if (txp->want_conn_close)
3059 return 0;
3060
3061 /*
3062 * Arbitrarily limit the length of the reason length string to half of the
3063 * MDPL.
3064 */
3065 if (reason_len > max_reason_len)
3066 reason_len = max_reason_len;
3067
3068 if (reason_len > 0) {
3069 reason = OPENSSL_memdup(f->reason, reason_len);
3070 if (reason == NULL)
3071 return 0;
3072 }
3073
3074 txp->conn_close_frame = *f;
3075 txp->conn_close_frame.reason = reason;
3076 txp->conn_close_frame.reason_len = reason_len;
3077 txp->want_conn_close = 1;
3078 return 1;
3079}
5cf99b40
MC
3080
3081void ossl_quic_tx_packetiser_set_msg_callback(OSSL_QUIC_TX_PACKETISER *txp,
3082 ossl_msg_cb msg_callback,
c2786c8e 3083 SSL *msg_callback_ssl)
5cf99b40
MC
3084{
3085 txp->msg_callback = msg_callback;
c2786c8e 3086 txp->msg_callback_ssl = msg_callback_ssl;
5cf99b40
MC
3087}
3088
3089void ossl_quic_tx_packetiser_set_msg_callback_arg(OSSL_QUIC_TX_PACKETISER *txp,
3090 void *msg_callback_arg)
3091{
3092 txp->msg_callback_arg = msg_callback_arg;
3093}
007f9e99
HL
3094
3095QUIC_PN ossl_quic_tx_packetiser_get_next_pn(OSSL_QUIC_TX_PACKETISER *txp,
3096 uint32_t pn_space)
3097{
3098 if (pn_space >= QUIC_PN_SPACE_NUM)
3099 return UINT64_MAX;
3100
3101 return txp->next_pn[pn_space];
3102}
c206f2aa
HL
3103
3104OSSL_TIME ossl_quic_tx_packetiser_get_deadline(OSSL_QUIC_TX_PACKETISER *txp)
3105{
3106 /*
3107 * TXP-specific deadline computations which rely on TXP innards. This is in
3108 * turn relied on by the QUIC_CHANNEL code to determine the channel event
3109 * handling deadline.
3110 */
3111 OSSL_TIME deadline = ossl_time_infinite();
3112 uint32_t enc_level, pn_space;
3113
3114 /*
3115 * ACK generation is not CC-gated - packets containing only ACKs are allowed
3116 * to bypass CC. We want to generate ACK frames even if we are currently
3117 * restricted by CC so the peer knows we have received data. The generate
3118 * call will take care of selecting the correct packet archetype.
3119 */
3120 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
3121 enc_level < QUIC_ENC_LEVEL_NUM;
3122 ++enc_level)
3123 if (ossl_qtx_is_enc_level_provisioned(txp->args.qtx, enc_level)) {
3124 pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
3125 deadline = ossl_time_min(deadline,
3126 ossl_ackm_get_ack_deadline(txp->args.ackm, pn_space));
3127 }
3128
3129 /* When will CC let us send more? */
63fac76c
HL
3130 if (txp->args.cc_method->get_tx_allowance(txp->args.cc_data) == 0)
3131 deadline = ossl_time_min(deadline,
3132 txp->args.cc_method->get_wakeup_deadline(txp->args.cc_data));
c206f2aa
HL
3133
3134 return deadline;
3135}