]> git.ipfire.org Git - thirdparty/openssl.git/blame - ssl/quic/quic_txp.c
Fix SHA3_squeeze on armv4.
[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,
5d27e7e9 484 on_sstream_updated, txp)) {
a73078b7
HL
485 OPENSSL_free(txp);
486 return NULL;
487 }
488
489 return txp;
490}
491
492void ossl_quic_tx_packetiser_free(OSSL_QUIC_TX_PACKETISER *txp)
493{
faebafda
HL
494 uint32_t enc_level;
495
a73078b7
HL
496 if (txp == NULL)
497 return;
498
499 ossl_quic_tx_packetiser_set_initial_token(txp, NULL, 0, NULL, NULL);
500 ossl_quic_fifd_cleanup(&txp->fifd);
a73078b7 501 OPENSSL_free(txp->conn_close_frame.reason);
faebafda
HL
502
503 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
504 enc_level < QUIC_ENC_LEVEL_NUM;
505 ++enc_level) {
506 OPENSSL_free(txp->el[enc_level].iovec);
507 OPENSSL_free(txp->el[enc_level].scratch);
508 }
509
a73078b7
HL
510 OPENSSL_free(txp);
511}
512
461d4117
HL
513/*
514 * Determine if an Initial packet token length is reasonable based on the
515 * current MDPL, returning 1 if it is OK.
516 *
517 * The real PMTU to the peer could differ from our (pessimistic) understanding
518 * of the PMTU, therefore it is possible we could receive an Initial token from
519 * a server in a Retry packet which is bigger than the MDPL. In this case it is
520 * impossible for us ever to make forward progress and we need to error out
521 * and fail the connection attempt.
522 *
523 * The specific boundary condition is complex: for example, after the size of
524 * the Initial token, there are the Initial packet header overheads and then
525 * encryption/AEAD tag overheads. After that, the minimum room for frame data in
526 * order to guarantee forward progress must be guaranteed. For example, a crypto
527 * stream needs to always be able to serialize at least one byte in a CRYPTO
528 * frame in order to make forward progress. Because the offset field of a CRYPTO
529 * frame uses a variable-length integer, the number of bytes needed to ensure
530 * this also varies.
531 *
532 * Rather than trying to get this boundary condition check actually right,
533 * require a reasonable amount of slack to avoid pathological behaviours. (After
534 * all, transmitting a CRYPTO stream one byte at a time is probably not
535 * desirable anyway.)
536 *
537 * We choose 160 bytes as the required margin, which is double the rough
538 * estimation of the minimum we would require to guarantee forward progress
539 * under worst case packet overheads.
540 */
541#define TXP_REQUIRED_TOKEN_MARGIN 160
542
543static int txp_check_token_len(size_t token_len, size_t mdpl)
544{
545 if (token_len == 0)
546 return 1;
547
548 if (token_len >= mdpl)
549 return 0;
550
551 if (TXP_REQUIRED_TOKEN_MARGIN >= mdpl)
552 /* (should not be possible because MDPL must be at least 1200) */
553 return 0;
554
555 if (token_len > mdpl - TXP_REQUIRED_TOKEN_MARGIN)
556 return 0;
557
558 return 1;
559}
560
561int ossl_quic_tx_packetiser_set_initial_token(OSSL_QUIC_TX_PACKETISER *txp,
562 const unsigned char *token,
563 size_t token_len,
564 ossl_quic_initial_token_free_fn *free_cb,
565 void *free_cb_arg)
a73078b7 566{
461d4117
HL
567 if (!txp_check_token_len(token_len, txp_get_mdpl(txp)))
568 return 0;
569
a73078b7
HL
570 if (txp->initial_token != NULL && txp->initial_token_free_cb != NULL)
571 txp->initial_token_free_cb(txp->initial_token, txp->initial_token_len,
572 txp->initial_token_free_cb_arg);
573
574 txp->initial_token = token;
575 txp->initial_token_len = token_len;
576 txp->initial_token_free_cb = free_cb;
577 txp->initial_token_free_cb_arg = free_cb_arg;
461d4117 578 return 1;
a73078b7
HL
579}
580
581int ossl_quic_tx_packetiser_set_cur_dcid(OSSL_QUIC_TX_PACKETISER *txp,
582 const QUIC_CONN_ID *dcid)
583{
584 if (dcid == NULL) {
585 ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_NULL_PARAMETER);
586 return 0;
587 }
588
589 txp->args.cur_dcid = *dcid;
590 return 1;
591}
592
593int ossl_quic_tx_packetiser_set_cur_scid(OSSL_QUIC_TX_PACKETISER *txp,
594 const QUIC_CONN_ID *scid)
595{
596 if (scid == NULL) {
597 ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_NULL_PARAMETER);
598 return 0;
599 }
600
601 txp->args.cur_scid = *scid;
602 return 1;
603}
604
605/* Change the destination L4 address the TXP uses to send datagrams. */
606int ossl_quic_tx_packetiser_set_peer(OSSL_QUIC_TX_PACKETISER *txp,
607 const BIO_ADDR *peer)
608{
609 if (peer == NULL) {
617b459d
HL
610 BIO_ADDR_clear(&txp->args.peer);
611 return 1;
a73078b7
HL
612 }
613
614 txp->args.peer = *peer;
615 return 1;
616}
617
8f9c9213
HL
618void ossl_quic_tx_packetiser_set_ack_tx_cb(OSSL_QUIC_TX_PACKETISER *txp,
619 void (*cb)(const OSSL_QUIC_FRAME_ACK *ack,
620 uint32_t pn_space,
621 void *arg),
622 void *cb_arg)
623{
624 txp->ack_tx_cb = cb;
625 txp->ack_tx_cb_arg = cb_arg;
626}
627
a73078b7
HL
628int ossl_quic_tx_packetiser_discard_enc_level(OSSL_QUIC_TX_PACKETISER *txp,
629 uint32_t enc_level)
630{
631 if (enc_level >= QUIC_ENC_LEVEL_NUM) {
632 ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_INVALID_ARGUMENT);
633 return 0;
634 }
635
636 if (enc_level != QUIC_ENC_LEVEL_0RTT)
637 txp->args.crypto[ossl_quic_enc_level_to_pn_space(enc_level)] = NULL;
638
a73078b7
HL
639 return 1;
640}
641
cda88baf
HL
642void ossl_quic_tx_packetiser_notify_handshake_complete(OSSL_QUIC_TX_PACKETISER *txp)
643{
644 txp->handshake_complete = 1;
645}
646
a73078b7
HL
647void ossl_quic_tx_packetiser_schedule_handshake_done(OSSL_QUIC_TX_PACKETISER *txp)
648{
649 txp->want_handshake_done = 1;
650}
651
652void ossl_quic_tx_packetiser_schedule_ack_eliciting(OSSL_QUIC_TX_PACKETISER *txp,
653 uint32_t pn_space)
654{
655 txp->force_ack_eliciting |= (1UL << pn_space);
656}
657
37ba2bc7
HL
658void ossl_quic_tx_packetiser_schedule_ack(OSSL_QUIC_TX_PACKETISER *txp,
659 uint32_t pn_space)
660{
661 txp->want_ack |= (1UL << pn_space);
662}
663
a73078b7
HL
664#define TXP_ERR_INTERNAL 0 /* Internal (e.g. alloc) error */
665#define TXP_ERR_SUCCESS 1 /* Success */
666#define TXP_ERR_SPACE 2 /* Not enough room for another packet */
667#define TXP_ERR_INPUT 3 /* Invalid/malformed input */
668
669/*
670 * Generates a datagram by polling the various ELs to determine if they want to
671 * generate any frames, and generating a datagram which coalesces packets for
672 * any ELs which do.
673 */
674int ossl_quic_tx_packetiser_generate(OSSL_QUIC_TX_PACKETISER *txp,
a3a51d6e 675 QUIC_TXP_STATUS *status)
a73078b7 676{
fee8f48e 677 /*
faebafda
HL
678 * Called to generate one or more datagrams, each containing one or more
679 * packets.
680 *
681 * There are some tricky things to note here:
682 *
683 * - The TXP is only concerned with generating encrypted packets;
684 * other packets use a different path.
685 *
686 * - Any datagram containing an Initial packet must have a payload length
687 * (DPL) of at least 1200 bytes. This padding need not necessarily be
688 * found in the Initial packet.
689 *
690 * - It is desirable to be able to coalesce an Initial packet
691 * with a Handshake packet. Since, before generating the Handshake
692 * packet, we do not know how long it will be, we cannot know the
693 * correct amount of padding to ensure a DPL of at least 1200 bytes.
694 * Thus this padding must added to the Handshake packet (or whatever
695 * packet is the last in the datagram).
696 *
697 * - However, at the time that we generate the Initial packet,
698 * we do not actually know for sure that we will be followed
699 * in the datagram by another packet. For example, suppose we have
700 * some queued data (e.g. crypto stream data for the HANDSHAKE EL)
701 * it looks like we will want to send on the HANDSHAKE EL.
702 * We could assume padding will be placed in the Handshake packet
703 * subsequently and avoid adding any padding to the Initial packet
704 * (which would leave no room for the Handshake packet in the
705 * datagram).
706 *
707 * However, this is not actually a safe assumption. Suppose that we
708 * are using a link with a MDPL of 1200 bytes, the minimum allowed by
709 * QUIC. Suppose that the Initial packet consumes 1195 bytes in total.
710 * Since it is not possible to fit a Handshake packet in just 5 bytes,
711 * upon trying to add a Handshake packet after generating the Initial
712 * packet, we will discover we have no room to fit it! This is not a
713 * problem in itself as another datagram can be sent subsequently, but
714 * it is a problem because we were counting to use that packet to hold
715 * the essential padding. But if we have already finished encrypting
716 * the Initial packet, we cannot go and add padding to it anymore.
717 * This leaves us stuck.
718 *
719 * Because of this, we have to plan multiple packets simultaneously, such
720 * that we can start generating a Handshake (or 0-RTT or 1-RTT, or so on)
721 * packet while still having the option to go back and add padding to the
722 * Initial packet if it turns out to be needed.
723 *
724 * Trying to predict ahead of time (e.g. during Initial packet generation)
725 * whether we will successfully generate a subsequent packet is fraught with
726 * error as it relies on a large number of variables:
727 *
728 * - Do we have room to fit a packet header? (Consider that due to
729 * variable-length integer encoding this is highly variable and can even
730 * depend on payload length due to a variable-length Length field.)
731 *
732 * - Can we fit even a single one of the frames we want to put in this
733 * packet in the packet? (Each frame type has a bespoke encoding. While
734 * our encodings of some frame types are adaptive based on the available
735 * room - e.g. STREAM frames - ultimately all frame types have some
736 * absolute minimum number of bytes to be successfully encoded. For
737 * example, if after an Initial packet there is enough room to encode
738 * only one byte of frame data, it is quite likely we can't send any of
739 * the frames we wanted to send.) While this is not strictly a problem
740 * because we could just fill the packet with padding frames, this is a
741 * pointless packet and is wasteful.
742 *
743 * Thus we adopt a multi-phase architecture:
744 *
745 * 1. Archetype Selection: Determine desired packet archetype.
746 *
747 * 2. Packet Staging: Generation of packet information and packet payload
748 * data (frame data) into staging areas.
749 *
750 * 3. Packet Adjustment: Adjustment of staged packets, adding padding to
751 * the staged packets if needed.
752 *
753 * 4. Commit: The packets are sent to the QTX and recorded as having been
754 * sent to the FIFM.
755 *
fee8f48e 756 */
64fd6991 757 int res = 0, rc;
faebafda
HL
758 uint32_t archetype, enc_level;
759 uint32_t conn_close_enc_level = QUIC_ENC_LEVEL_NUM;
760 struct txp_pkt pkt[QUIC_ENC_LEVEL_NUM];
761 size_t pkts_done = 0;
762 uint64_t cc_limit = txp->args.cc_method->get_tx_allowance(txp->args.cc_data);
6a2b70e2 763 int need_padding = 0, txpim_pkt_reffed;
a73078b7
HL
764
765 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
766 enc_level < QUIC_ENC_LEVEL_NUM;
faebafda
HL
767 ++enc_level)
768 pkt[enc_level].h_valid = 0;
a73078b7 769
64fd6991
TM
770 memset(status, 0, sizeof(*status));
771
a73078b7
HL
772 /*
773 * Should not be needed, but a sanity check in case anyone else has been
774 * using the QTX.
775 */
776 ossl_qtx_finish_dgram(txp->args.qtx);
777
faebafda 778 /* 1. Archetype Selection */
c206f2aa 779 archetype = txp_determine_archetype(txp, cc_limit);
faebafda
HL
780
781 /* 2. Packet Staging */
a73078b7
HL
782 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
783 enc_level < QUIC_ENC_LEVEL_NUM;
784 ++enc_level) {
faebafda
HL
785 size_t running_total = (enc_level > QUIC_ENC_LEVEL_INITIAL)
786 ? pkt[enc_level - 1].geom.hwm : 0;
a73078b7 787
faebafda 788 pkt[enc_level].geom.hwm = running_total;
a73078b7 789
faebafda
HL
790 if (!txp_should_try_staging(txp, enc_level, archetype, cc_limit,
791 &conn_close_enc_level))
792 continue;
793
794 if (!txp_pkt_init(&pkt[enc_level], txp, enc_level, archetype,
795 running_total))
a73078b7 796 /*
faebafda
HL
797 * If this fails this is not a fatal error - it means the geometry
798 * planning determined there was not enough space for another
799 * packet. So just proceed with what we've already planned for.
a73078b7 800 */
faebafda
HL
801 break;
802
803 rc = txp_generate_for_el(txp, &pkt[enc_level],
804 conn_close_enc_level == enc_level);
805 if (rc != TXP_ERR_SUCCESS)
806 goto out;
807
c5cb85b6
HL
808 if (pkt[enc_level].force_pad)
809 /*
810 * txp_generate_for_el emitted a frame which forces packet padding.
811 */
812 need_padding = 1;
813
faebafda
HL
814 pkt[enc_level].geom.hwm = running_total
815 + pkt[enc_level].h.bytes_appended
816 + pkt[enc_level].geom.pkt_overhead;
817 }
818
819 /* 3. Packet Adjustment */
820 if (pkt[QUIC_ENC_LEVEL_INITIAL].h_valid
c5cb85b6 821 && pkt[QUIC_ENC_LEVEL_INITIAL].h.bytes_appended > 0)
faebafda
HL
822 /*
823 * We have an Initial packet in this datagram, so we need to make sure
824 * the total size of the datagram is adequate.
825 */
c5cb85b6
HL
826 need_padding = 1;
827
828 if (need_padding) {
faebafda
HL
829 size_t total_dgram_size = 0;
830 const size_t min_dpl = QUIC_MIN_INITIAL_DGRAM_LEN;
e1c15a8a 831 uint32_t pad_el = QUIC_ENC_LEVEL_NUM;
faebafda
HL
832
833 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
834 enc_level < QUIC_ENC_LEVEL_NUM;
835 ++enc_level)
836 if (pkt[enc_level].h_valid && pkt[enc_level].h.bytes_appended > 0) {
e1c15a8a
HL
837 if (pad_el == QUIC_ENC_LEVEL_NUM
838 /*
839 * We might not be able to add padding, for example if we
840 * are using the ACK_ONLY archetype.
841 */
842 && pkt[enc_level].geom.adata.allow_padding
843 && !pkt[enc_level].h.done_implicit)
844 pad_el = enc_level;
c5cb85b6 845
faebafda
HL
846 txp_pkt_postgen_update_pkt_overhead(&pkt[enc_level], txp);
847 total_dgram_size += pkt[enc_level].geom.pkt_overhead
848 + pkt[enc_level].h.bytes_appended;
849 }
850
e1c15a8a 851 if (pad_el != QUIC_ENC_LEVEL_NUM && total_dgram_size < min_dpl) {
faebafda
HL
852 size_t deficit = min_dpl - total_dgram_size;
853
e1c15a8a 854 if (!txp_pkt_append_padding(&pkt[pad_el], txp, deficit))
3bef14c5
HL
855 goto out;
856
e1c15a8a 857 total_dgram_size += deficit;
3bef14c5
HL
858
859 /*
860 * Padding frames make a packet ineligible for being a non-inflight
861 * packet.
862 */
e1c15a8a
HL
863 pkt[pad_el].tpkt->ackm_pkt.is_inflight = 1;
864 }
865
866 /*
867 * If we have failed to make a datagram of adequate size, for example
868 * because we have a padding requirement but are using the ACK_ONLY
869 * archetype (because we are CC limited), which precludes us from
870 * sending padding, give up on generating the datagram - there is
871 * nothing we can do.
872 */
873 if (total_dgram_size < min_dpl) {
874 res = 1;
875 goto out;
a73078b7 876 }
faebafda
HL
877 }
878
879 /* 4. Commit */
faebafda
HL
880 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
881 enc_level < QUIC_ENC_LEVEL_NUM;
882 ++enc_level) {
883
884 if (!pkt[enc_level].h_valid)
885 /* Did not attempt to generate a packet for this EL. */
886 continue;
887
888 if (pkt[enc_level].h.bytes_appended == 0)
889 /* Nothing was generated for this EL, so skip. */
890 continue;
891
6a2b70e2
HL
892 rc = txp_pkt_commit(txp, &pkt[enc_level], archetype,
893 &txpim_pkt_reffed);
e1c15a8a 894 if (rc) {
6a2b70e2
HL
895 status->sent_ack_eliciting
896 = status->sent_ack_eliciting
897 || pkt[enc_level].tpkt->ackm_pkt.is_ack_eliciting;
898
e1c15a8a
HL
899 if (enc_level == QUIC_ENC_LEVEL_HANDSHAKE)
900 status->sent_handshake
901 = (pkt[enc_level].h_valid
902 && pkt[enc_level].h.bytes_appended > 0);
903 }
904
6a2b70e2
HL
905 if (txpim_pkt_reffed)
906 pkt[enc_level].tpkt = NULL; /* don't free */
faebafda 907
6a2b70e2
HL
908 if (!rc)
909 goto out;
faebafda 910
a73078b7
HL
911 ++pkts_done;
912 }
913
faebafda 914 /* Flush & Cleanup */
64fd6991 915 res = 1;
faebafda 916out:
a73078b7 917 ossl_qtx_finish_dgram(txp->args.qtx);
a73078b7 918
faebafda
HL
919 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
920 enc_level < QUIC_ENC_LEVEL_NUM;
921 ++enc_level)
922 txp_pkt_cleanup(&pkt[enc_level], txp);
923
64fd6991
TM
924 status->sent_pkt = pkts_done;
925
96014840 926 return res;
faebafda 927}
a73078b7
HL
928
929static const struct archetype_data archetypes[QUIC_ENC_LEVEL_NUM][TX_PACKETISER_ARCHETYPE_NUM] = {
930 /* EL 0(INITIAL) */
931 {
932 /* EL 0(INITIAL) - Archetype 0(NORMAL) */
933 {
934 /*allow_ack =*/ 1,
935 /*allow_ping =*/ 1,
936 /*allow_crypto =*/ 1,
937 /*allow_handshake_done =*/ 0,
938 /*allow_path_challenge =*/ 0,
939 /*allow_path_response =*/ 0,
940 /*allow_new_conn_id =*/ 0,
941 /*allow_retire_conn_id =*/ 0,
942 /*allow_stream_rel =*/ 0,
943 /*allow_conn_fc =*/ 0,
944 /*allow_conn_close =*/ 1,
8d2e353d 945 /*allow_cfq_other =*/ 0,
a73078b7
HL
946 /*allow_new_token =*/ 0,
947 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
948 /*allow_padding =*/ 1,
949 /*require_ack_eliciting =*/ 0,
950 /*bypass_cc =*/ 0,
a73078b7 951 },
faebafda 952 /* EL 0(INITIAL) - Archetype 1(PROBE) */
a73078b7
HL
953 {
954 /*allow_ack =*/ 1,
d56b564b 955 /*allow_ping =*/ 1,
faebafda
HL
956 /*allow_crypto =*/ 1,
957 /*allow_handshake_done =*/ 0,
958 /*allow_path_challenge =*/ 0,
959 /*allow_path_response =*/ 0,
960 /*allow_new_conn_id =*/ 0,
961 /*allow_retire_conn_id =*/ 0,
962 /*allow_stream_rel =*/ 0,
963 /*allow_conn_fc =*/ 0,
964 /*allow_conn_close =*/ 1,
965 /*allow_cfq_other =*/ 0,
966 /*allow_new_token =*/ 0,
967 /*allow_force_ack_eliciting =*/ 1,
968 /*allow_padding =*/ 1,
969 /*require_ack_eliciting =*/ 1,
970 /*bypass_cc =*/ 1,
971 },
972 /* EL 0(INITIAL) - Archetype 2(ACK_ONLY) */
973 {
974 /*allow_ack =*/ 1,
975 /*allow_ping =*/ 0,
a73078b7
HL
976 /*allow_crypto =*/ 0,
977 /*allow_handshake_done =*/ 0,
978 /*allow_path_challenge =*/ 0,
979 /*allow_path_response =*/ 0,
980 /*allow_new_conn_id =*/ 0,
981 /*allow_retire_conn_id =*/ 0,
982 /*allow_stream_rel =*/ 0,
983 /*allow_conn_fc =*/ 0,
984 /*allow_conn_close =*/ 0,
985 /*allow_cfq_other =*/ 0,
986 /*allow_new_token =*/ 0,
987 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
988 /*allow_padding =*/ 0,
989 /*require_ack_eliciting =*/ 0,
990 /*bypass_cc =*/ 1,
a73078b7
HL
991 },
992 },
993 /* EL 1(HANDSHAKE) */
994 {
995 /* EL 1(HANDSHAKE) - Archetype 0(NORMAL) */
996 {
997 /*allow_ack =*/ 1,
998 /*allow_ping =*/ 1,
999 /*allow_crypto =*/ 1,
1000 /*allow_handshake_done =*/ 0,
1001 /*allow_path_challenge =*/ 0,
1002 /*allow_path_response =*/ 0,
1003 /*allow_new_conn_id =*/ 0,
1004 /*allow_retire_conn_id =*/ 0,
1005 /*allow_stream_rel =*/ 0,
1006 /*allow_conn_fc =*/ 0,
1007 /*allow_conn_close =*/ 1,
8d2e353d 1008 /*allow_cfq_other =*/ 0,
a73078b7
HL
1009 /*allow_new_token =*/ 0,
1010 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
1011 /*allow_padding =*/ 1,
1012 /*require_ack_eliciting =*/ 0,
1013 /*bypass_cc =*/ 0,
a73078b7 1014 },
faebafda 1015 /* EL 1(HANDSHAKE) - Archetype 1(PROBE) */
a73078b7
HL
1016 {
1017 /*allow_ack =*/ 1,
d56b564b 1018 /*allow_ping =*/ 1,
faebafda
HL
1019 /*allow_crypto =*/ 1,
1020 /*allow_handshake_done =*/ 0,
1021 /*allow_path_challenge =*/ 0,
1022 /*allow_path_response =*/ 0,
1023 /*allow_new_conn_id =*/ 0,
1024 /*allow_retire_conn_id =*/ 0,
1025 /*allow_stream_rel =*/ 0,
1026 /*allow_conn_fc =*/ 0,
1027 /*allow_conn_close =*/ 1,
1028 /*allow_cfq_other =*/ 0,
1029 /*allow_new_token =*/ 0,
1030 /*allow_force_ack_eliciting =*/ 1,
1031 /*allow_padding =*/ 1,
1032 /*require_ack_eliciting =*/ 1,
1033 /*bypass_cc =*/ 1,
1034 },
1035 /* EL 1(HANDSHAKE) - Archetype 2(ACK_ONLY) */
1036 {
1037 /*allow_ack =*/ 1,
1038 /*allow_ping =*/ 0,
a73078b7
HL
1039 /*allow_crypto =*/ 0,
1040 /*allow_handshake_done =*/ 0,
1041 /*allow_path_challenge =*/ 0,
1042 /*allow_path_response =*/ 0,
1043 /*allow_new_conn_id =*/ 0,
1044 /*allow_retire_conn_id =*/ 0,
1045 /*allow_stream_rel =*/ 0,
1046 /*allow_conn_fc =*/ 0,
1047 /*allow_conn_close =*/ 0,
1048 /*allow_cfq_other =*/ 0,
1049 /*allow_new_token =*/ 0,
1050 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
1051 /*allow_padding =*/ 0,
1052 /*require_ack_eliciting =*/ 0,
1053 /*bypass_cc =*/ 1,
a73078b7
HL
1054 },
1055 },
1056 /* EL 2(0RTT) */
1057 {
1058 /* EL 2(0RTT) - Archetype 0(NORMAL) */
1059 {
1060 /*allow_ack =*/ 0,
1061 /*allow_ping =*/ 1,
1062 /*allow_crypto =*/ 0,
1063 /*allow_handshake_done =*/ 0,
1064 /*allow_path_challenge =*/ 0,
1065 /*allow_path_response =*/ 0,
1066 /*allow_new_conn_id =*/ 1,
1067 /*allow_retire_conn_id =*/ 1,
1068 /*allow_stream_rel =*/ 1,
1069 /*allow_conn_fc =*/ 1,
1070 /*allow_conn_close =*/ 1,
1071 /*allow_cfq_other =*/ 0,
1072 /*allow_new_token =*/ 0,
1073 /*allow_force_ack_eliciting =*/ 0,
faebafda
HL
1074 /*allow_padding =*/ 1,
1075 /*require_ack_eliciting =*/ 0,
1076 /*bypass_cc =*/ 0,
a73078b7 1077 },
faebafda 1078 /* EL 2(0RTT) - Archetype 1(PROBE) */
a73078b7
HL
1079 {
1080 /*allow_ack =*/ 0,
d56b564b 1081 /*allow_ping =*/ 1,
a73078b7
HL
1082 /*allow_crypto =*/ 0,
1083 /*allow_handshake_done =*/ 0,
1084 /*allow_path_challenge =*/ 0,
1085 /*allow_path_response =*/ 0,
faebafda
HL
1086 /*allow_new_conn_id =*/ 1,
1087 /*allow_retire_conn_id =*/ 1,
1088 /*allow_stream_rel =*/ 1,
1089 /*allow_conn_fc =*/ 1,
1090 /*allow_conn_close =*/ 1,
1091 /*allow_cfq_other =*/ 0,
1092 /*allow_new_token =*/ 0,
1093 /*allow_force_ack_eliciting =*/ 0,
1094 /*allow_padding =*/ 1,
1095 /*require_ack_eliciting =*/ 1,
1096 /*bypass_cc =*/ 1,
1097 },
1098 /* EL 2(0RTT) - Archetype 2(ACK_ONLY) */
1099 {
1100 /*allow_ack =*/ 0,
1101 /*allow_ping =*/ 0,
1102 /*allow_crypto =*/ 0,
1103 /*allow_handshake_done =*/ 0,
1104 /*allow_path_challenge =*/ 0,
1105 /*allow_path_response =*/ 0,
a73078b7
HL
1106 /*allow_new_conn_id =*/ 0,
1107 /*allow_retire_conn_id =*/ 0,
1108 /*allow_stream_rel =*/ 0,
1109 /*allow_conn_fc =*/ 0,
1110 /*allow_conn_close =*/ 0,
1111 /*allow_cfq_other =*/ 0,
1112 /*allow_new_token =*/ 0,
1113 /*allow_force_ack_eliciting =*/ 0,
faebafda
HL
1114 /*allow_padding =*/ 0,
1115 /*require_ack_eliciting =*/ 0,
1116 /*bypass_cc =*/ 1,
a73078b7
HL
1117 },
1118 },
1119 /* EL 3(1RTT) */
1120 {
1121 /* EL 3(1RTT) - Archetype 0(NORMAL) */
1122 {
1123 /*allow_ack =*/ 1,
1124 /*allow_ping =*/ 1,
1125 /*allow_crypto =*/ 1,
1126 /*allow_handshake_done =*/ 1,
1127 /*allow_path_challenge =*/ 0,
7eb330ff 1128 /*allow_path_response =*/ 1,
a73078b7
HL
1129 /*allow_new_conn_id =*/ 1,
1130 /*allow_retire_conn_id =*/ 1,
1131 /*allow_stream_rel =*/ 1,
1132 /*allow_conn_fc =*/ 1,
1133 /*allow_conn_close =*/ 1,
1134 /*allow_cfq_other =*/ 1,
1135 /*allow_new_token =*/ 1,
1136 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
1137 /*allow_padding =*/ 1,
1138 /*require_ack_eliciting =*/ 0,
1139 /*bypass_cc =*/ 0,
a73078b7 1140 },
faebafda 1141 /* EL 3(1RTT) - Archetype 1(PROBE) */
a73078b7
HL
1142 {
1143 /*allow_ack =*/ 1,
d56b564b 1144 /*allow_ping =*/ 1,
faebafda
HL
1145 /*allow_crypto =*/ 1,
1146 /*allow_handshake_done =*/ 1,
1147 /*allow_path_challenge =*/ 0,
7eb330ff 1148 /*allow_path_response =*/ 1,
faebafda
HL
1149 /*allow_new_conn_id =*/ 1,
1150 /*allow_retire_conn_id =*/ 1,
1151 /*allow_stream_rel =*/ 1,
1152 /*allow_conn_fc =*/ 1,
1153 /*allow_conn_close =*/ 1,
1154 /*allow_cfq_other =*/ 1,
1155 /*allow_new_token =*/ 1,
1156 /*allow_force_ack_eliciting =*/ 1,
1157 /*allow_padding =*/ 1,
1158 /*require_ack_eliciting =*/ 1,
1159 /*bypass_cc =*/ 1,
1160 },
1161 /* EL 3(1RTT) - Archetype 2(ACK_ONLY) */
1162 {
1163 /*allow_ack =*/ 1,
1164 /*allow_ping =*/ 0,
a73078b7
HL
1165 /*allow_crypto =*/ 0,
1166 /*allow_handshake_done =*/ 0,
1167 /*allow_path_challenge =*/ 0,
1168 /*allow_path_response =*/ 0,
1169 /*allow_new_conn_id =*/ 0,
1170 /*allow_retire_conn_id =*/ 0,
1171 /*allow_stream_rel =*/ 0,
1172 /*allow_conn_fc =*/ 0,
1173 /*allow_conn_close =*/ 0,
1174 /*allow_cfq_other =*/ 0,
1175 /*allow_new_token =*/ 0,
1176 /*allow_force_ack_eliciting =*/ 1,
faebafda
HL
1177 /*allow_padding =*/ 0,
1178 /*require_ack_eliciting =*/ 0,
1179 /*bypass_cc =*/ 1,
a73078b7
HL
1180 }
1181 }
1182};
1183
1184static int txp_get_archetype_data(uint32_t enc_level,
1185 uint32_t archetype,
1186 struct archetype_data *a)
1187{
1188 if (enc_level >= QUIC_ENC_LEVEL_NUM
1189 || archetype >= TX_PACKETISER_ARCHETYPE_NUM)
1190 return 0;
1191
1192 /* No need to avoid copying this as it should not exceed one int in size. */
1193 *a = archetypes[enc_level][archetype];
1194 return 1;
1195}
1196
faebafda
HL
1197static int txp_determine_geometry(OSSL_QUIC_TX_PACKETISER *txp,
1198 uint32_t archetype,
1199 uint32_t enc_level,
1200 size_t running_total,
1201 QUIC_PKT_HDR *phdr,
1202 struct txp_pkt_geom *geom)
1203{
1204 size_t mdpl, cmpl, hdr_len;
1205
1206 /* Get information about packet archetype. */
1207 if (!txp_get_archetype_data(enc_level, archetype, &geom->adata))
1208 return 0;
1209
1210 /* Assemble packet header. */
1211 phdr->type = ossl_quic_enc_level_to_pkt_type(enc_level);
1212 phdr->spin_bit = 0;
1213 phdr->pn_len = txp_determine_pn_len(txp);
1214 phdr->partial = 0;
1215 phdr->fixed = 1;
1216 phdr->reserved = 0;
1217 phdr->version = QUIC_VERSION_1;
1218 phdr->dst_conn_id = txp->args.cur_dcid;
1219 phdr->src_conn_id = txp->args.cur_scid;
1220
1221 /*
1222 * We need to know the length of the payload to get an accurate header
1223 * length for non-1RTT packets, because the Length field found in
1224 * Initial/Handshake/0-RTT packets uses a variable-length encoding. However,
1225 * we don't have a good idea of the length of our payload, because the
1226 * length of the payload depends on the room in the datagram after fitting
1227 * the header, which depends on the size of the header.
1228 *
1229 * In general, it does not matter if a packet is slightly shorter (because
1230 * e.g. we predicted use of a 2-byte length field, but ended up only needing
1231 * a 1-byte length field). However this does matter for Initial packets
1232 * which must be at least 1200 bytes, which is also the assumed default MTU;
1233 * therefore in many cases Initial packets will be padded to 1200 bytes,
1234 * which means if we overestimated the header size, we will be short by a
1235 * few bytes and the server will ignore the packet for being too short. In
1236 * this case, however, such packets always *will* be padded to meet 1200
1237 * bytes, which requires a 2-byte length field, so we don't actually need to
1238 * worry about this. Thus we estimate the header length assuming a 2-byte
1239 * length field here, which should in practice work well in all cases.
1240 */
1241 phdr->len = OSSL_QUIC_VLINT_2B_MAX - phdr->pn_len;
1242
1243 if (enc_level == QUIC_ENC_LEVEL_INITIAL) {
1244 phdr->token = txp->initial_token;
1245 phdr->token_len = txp->initial_token_len;
1246 } else {
1247 phdr->token = NULL;
1248 phdr->token_len = 0;
1249 }
1250
1251 hdr_len = ossl_quic_wire_get_encoded_pkt_hdr_len(phdr->dst_conn_id.id_len,
1252 phdr);
1253 if (hdr_len == 0)
1254 return 0;
1255
1256 /* MDPL: Maximum datagram payload length. */
1257 mdpl = txp_get_mdpl(txp);
1258
1259 /*
1260 * CMPL: Maximum encoded packet size we can put into this datagram given any
1261 * previous packets coalesced into it.
1262 */
1263 if (running_total > mdpl)
1264 /* Should not be possible, but if it happens: */
1265 cmpl = 0;
1266 else
1267 cmpl = mdpl - running_total;
1268
1269 /* CMPPL: Maximum amount we can put into the current packet payload */
1270 if (!txp_determine_ppl_from_pl(txp, cmpl, enc_level, hdr_len, &geom->cmppl))
1271 return 0;
1272
1273 geom->cmpl = cmpl;
1274 geom->pkt_overhead = cmpl - geom->cmppl;
1275 geom->archetype = archetype;
1276 return 1;
1277}
1278
c206f2aa
HL
1279static uint32_t txp_determine_archetype(OSSL_QUIC_TX_PACKETISER *txp,
1280 uint64_t cc_limit)
faebafda
HL
1281{
1282 OSSL_ACKM_PROBE_INFO *probe_info
1283 = ossl_ackm_get0_probe_request(txp->args.ackm);
faebafda
HL
1284 uint32_t pn_space;
1285
1286 /*
1287 * If ACKM has requested probe generation (e.g. due to PTO), we generate a
1288 * Probe-archetype packet. Actually, we determine archetype on a
1289 * per-datagram basis, so if any EL wants a probe, do a pass in which
1290 * we try and generate a probe (if needed) for all ELs.
1291 */
1292 if (probe_info->anti_deadlock_initial > 0
1293 || probe_info->anti_deadlock_handshake > 0)
1294 return TX_PACKETISER_ARCHETYPE_PROBE;
1295
1296 for (pn_space = QUIC_PN_SPACE_INITIAL;
1297 pn_space < QUIC_PN_SPACE_NUM;
1298 ++pn_space)
1299 if (probe_info->pto[pn_space] > 0)
1300 return TX_PACKETISER_ARCHETYPE_PROBE;
1301
1302 /*
1303 * If we are out of CC budget, we cannot send a normal packet,
1304 * but we can do an ACK-only packet (potentially, if we
1305 * want to send an ACK).
1306 */
1307 if (cc_limit == 0)
1308 return TX_PACKETISER_ARCHETYPE_ACK_ONLY;
1309
1310 /* All other packets. */
1311 return TX_PACKETISER_ARCHETYPE_NORMAL;
1312}
1313
1314static int txp_should_try_staging(OSSL_QUIC_TX_PACKETISER *txp,
1315 uint32_t enc_level,
1316 uint32_t archetype,
1317 uint64_t cc_limit,
1318 uint32_t *conn_close_enc_level)
a73078b7
HL
1319{
1320 struct archetype_data a;
1321 uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
1322 QUIC_CFQ_ITEM *cfq_item;
1323
1324 if (!ossl_qtx_is_enc_level_provisioned(txp->args.qtx, enc_level))
1325 return 0;
1326
faebafda
HL
1327 if (!txp_get_archetype_data(enc_level, archetype, &a))
1328 return 0;
1329
1330 if (!a.bypass_cc && cc_limit == 0)
1331 /* CC not allowing us to send. */
1332 return 0;
1333
96fa10f3
HL
1334 /*
1335 * We can produce CONNECTION_CLOSE frames on any EL in principle, which
1336 * means we need to choose which EL we would prefer to use. After a
1337 * connection is fully established we have only one provisioned EL and this
1338 * is a non-issue. Where multiple ELs are provisioned, it is possible the
1339 * peer does not have the keys for the EL yet, which suggests in general it
1340 * is preferable to use the lowest EL which is still provisioned.
1341 *
d15d5ea6
P
1342 * However (RFC 9000 s. 10.2.3 & 12.5) we are also required to not send
1343 * application CONNECTION_CLOSE frames in non-1-RTT ELs, so as to not
1344 * potentially leak application data on a connection which has yet to be
1345 * authenticated. Thus when we have an application CONNECTION_CLOSE frame
1346 * queued and need to send it on a non-1-RTT EL, we have to convert it
1347 * into a transport CONNECTION_CLOSE frame which contains no application
1348 * data. Since this loses information, it suggests we should use the 1-RTT
1349 * EL to avoid this if possible, even if a lower EL is also available.
96fa10f3
HL
1350 *
1351 * At the same time, just because we have the 1-RTT EL provisioned locally
1352 * does not necessarily mean the peer does, for example if a handshake
1353 * CRYPTO frame has been lost. It is fairly important that CONNECTION_CLOSE
1354 * is signalled in a way we know our peer can decrypt, as we stop processing
1355 * connection retransmission logic for real after connection close and
1356 * simply 'blindly' retransmit the same CONNECTION_CLOSE frame.
1357 *
1358 * This is not a major concern for clients, since if a client has a 1-RTT EL
1359 * provisioned the server is guaranteed to also have a 1-RTT EL provisioned.
1360 *
44cb36d0 1361 * TODO(QUIC SERVER): Revisit this when server support is added.
96fa10f3
HL
1362 */
1363 if (*conn_close_enc_level > enc_level
1364 && *conn_close_enc_level != QUIC_ENC_LEVEL_1RTT)
7f9d1249
HL
1365 *conn_close_enc_level = enc_level;
1366
fee8f48e
HL
1367 /* Do we need to send a PTO probe? */
1368 if (a.allow_force_ack_eliciting) {
1369 OSSL_ACKM_PROBE_INFO *probe_info
2477e99f 1370 = ossl_ackm_get0_probe_request(txp->args.ackm);
fee8f48e
HL
1371
1372 if ((enc_level == QUIC_ENC_LEVEL_INITIAL
1373 && probe_info->anti_deadlock_initial > 0)
1374 || (enc_level == QUIC_ENC_LEVEL_HANDSHAKE
1375 && probe_info->anti_deadlock_handshake > 0)
1376 || probe_info->pto[pn_space] > 0)
1377 return 1;
1378 }
1379
a73078b7
HL
1380 /* Does the crypto stream for this EL want to produce anything? */
1381 if (a.allow_crypto && sstream_is_pending(txp->args.crypto[pn_space]))
1382 return 1;
1383
1384 /* Does the ACKM for this PN space want to produce anything? */
1385 if (a.allow_ack && (ossl_ackm_is_ack_desired(txp->args.ackm, pn_space)
1386 || (txp->want_ack & (1UL << pn_space)) != 0))
1387 return 1;
1388
1389 /* Do we need to force emission of an ACK-eliciting packet? */
1390 if (a.allow_force_ack_eliciting
1391 && (txp->force_ack_eliciting & (1UL << pn_space)) != 0)
1392 return 1;
1393
1394 /* Does the connection-level RXFC want to produce a frame? */
1395 if (a.allow_conn_fc && (txp->want_max_data
1396 || ossl_quic_rxfc_has_cwm_changed(txp->args.conn_rxfc, 0)))
1397 return 1;
1398
1399 /* Do we want to produce a MAX_STREAMS frame? */
a6b6ea17
HL
1400 if (a.allow_conn_fc
1401 && (txp->want_max_streams_bidi
1402 || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_bidi_rxfc,
1403 0)
1404 || txp->want_max_streams_uni
1405 || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_uni_rxfc,
1406 0)))
a73078b7
HL
1407 return 1;
1408
1409 /* Do we want to produce a HANDSHAKE_DONE frame? */
1410 if (a.allow_handshake_done && txp->want_handshake_done)
1411 return 1;
1412
1413 /* Do we want to produce a CONNECTION_CLOSE frame? */
7f9d1249
HL
1414 if (a.allow_conn_close && txp->want_conn_close &&
1415 *conn_close_enc_level == enc_level)
1416 /*
1417 * This is a bit of a special case since CONNECTION_CLOSE can appear in
1418 * most packet types, and when we decide we want to send it this status
1419 * isn't tied to a specific EL. So if we want to send it, we send it
1420 * only on the lowest non-dropped EL.
1421 */
a73078b7
HL
1422 return 1;
1423
1424 /* Does the CFQ have any frames queued for this PN space? */
1425 if (enc_level != QUIC_ENC_LEVEL_0RTT)
1426 for (cfq_item = ossl_quic_cfq_get_priority_head(txp->args.cfq, pn_space);
1427 cfq_item != NULL;
1428 cfq_item = ossl_quic_cfq_item_get_priority_next(cfq_item, pn_space)) {
1429 uint64_t frame_type = ossl_quic_cfq_item_get_frame_type(cfq_item);
1430
1431 switch (frame_type) {
1432 case OSSL_QUIC_FRAME_TYPE_NEW_CONN_ID:
1433 if (a.allow_new_conn_id)
1434 return 1;
1435 break;
1436 case OSSL_QUIC_FRAME_TYPE_RETIRE_CONN_ID:
1437 if (a.allow_retire_conn_id)
1438 return 1;
1439 break;
1440 case OSSL_QUIC_FRAME_TYPE_NEW_TOKEN:
1441 if (a.allow_new_token)
1442 return 1;
1443 break;
7eb330ff
HL
1444 case OSSL_QUIC_FRAME_TYPE_PATH_RESPONSE:
1445 if (a.allow_path_response)
1446 return 1;
1447 break;
a73078b7
HL
1448 default:
1449 if (a.allow_cfq_other)
1450 return 1;
1451 break;
1452 }
1453 }
1454
cda88baf 1455 if (a.allow_stream_rel && txp->handshake_complete) {
a73078b7
HL
1456 QUIC_STREAM_ITER it;
1457
1458 /* If there are any active streams, 0/1-RTT wants to produce a packet.
1459 * Whether a stream is on the active list is required to be precise
1460 * (i.e., a stream is never on the active list if we cannot produce a
1461 * frame for it), and all stream-related frames are governed by
1462 * a.allow_stream_rel (i.e., if we can send one type of stream-related
1463 * frame, we can send any of them), so we don't need to inspect
1464 * individual streams on the active list, just confirm that the active
1465 * list is non-empty.
1466 */
1467 ossl_quic_stream_iter_init(&it, txp->args.qsm, 0);
1468 if (it.stream != NULL)
1469 return 1;
1470 }
1471
1472 return 0;
1473}
1474
1475static int sstream_is_pending(QUIC_SSTREAM *sstream)
1476{
1477 OSSL_QUIC_FRAME_STREAM hdr;
1478 OSSL_QTX_IOVEC iov[2];
1479 size_t num_iov = OSSL_NELEM(iov);
1480
1481 return ossl_quic_sstream_get_stream_frame(sstream, 0, &hdr, iov, &num_iov);
1482}
1483
a73078b7
HL
1484/* Determine how many bytes we should use for the encoded PN. */
1485static size_t txp_determine_pn_len(OSSL_QUIC_TX_PACKETISER *txp)
1486{
44cb36d0 1487 return 4; /* TODO(QUIC FUTURE) */
a73078b7
HL
1488}
1489
1490/* Determine plaintext packet payload length from payload length. */
1491static int txp_determine_ppl_from_pl(OSSL_QUIC_TX_PACKETISER *txp,
1492 size_t pl,
1493 uint32_t enc_level,
1494 size_t hdr_len,
1495 size_t *r)
1496{
1497 if (pl < hdr_len)
1498 return 0;
1499
1500 pl -= hdr_len;
1501
1502 if (!ossl_qtx_calculate_plaintext_payload_len(txp->args.qtx, enc_level,
1503 pl, &pl))
1504 return 0;
1505
1506 *r = pl;
1507 return 1;
1508}
1509
1510static size_t txp_get_mdpl(OSSL_QUIC_TX_PACKETISER *txp)
1511{
1512 return ossl_qtx_get_mdpl(txp->args.qtx);
1513}
1514
1515static QUIC_SSTREAM *get_sstream_by_id(uint64_t stream_id, uint32_t pn_space,
1516 void *arg)
1517{
1518 OSSL_QUIC_TX_PACKETISER *txp = arg;
1519 QUIC_STREAM *s;
1520
1521 if (stream_id == UINT64_MAX)
1522 return txp->args.crypto[pn_space];
1523
1524 s = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1525 if (s == NULL)
1526 return NULL;
1527
1528 return s->sstream;
1529}
1530
1531static void on_regen_notify(uint64_t frame_type, uint64_t stream_id,
1532 QUIC_TXPIM_PKT *pkt, void *arg)
1533{
1534 OSSL_QUIC_TX_PACKETISER *txp = arg;
1535
1536 switch (frame_type) {
1537 case OSSL_QUIC_FRAME_TYPE_HANDSHAKE_DONE:
1538 txp->want_handshake_done = 1;
1539 break;
1540 case OSSL_QUIC_FRAME_TYPE_MAX_DATA:
1541 txp->want_max_data = 1;
1542 break;
1543 case OSSL_QUIC_FRAME_TYPE_MAX_STREAMS_BIDI:
1544 txp->want_max_streams_bidi = 1;
1545 break;
1546 case OSSL_QUIC_FRAME_TYPE_MAX_STREAMS_UNI:
1547 txp->want_max_streams_uni = 1;
1548 break;
1549 case OSSL_QUIC_FRAME_TYPE_ACK_WITH_ECN:
1550 txp->want_ack |= (1UL << pkt->ackm_pkt.pkt_space);
1551 break;
1552 case OSSL_QUIC_FRAME_TYPE_MAX_STREAM_DATA:
1553 {
1554 QUIC_STREAM *s
1555 = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1556
1557 if (s == NULL)
1558 return;
1559
1560 s->want_max_stream_data = 1;
1561 ossl_quic_stream_map_update_state(txp->args.qsm, s);
1562 }
1563 break;
1564 case OSSL_QUIC_FRAME_TYPE_STOP_SENDING:
1565 {
1566 QUIC_STREAM *s
1567 = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1568
1569 if (s == NULL)
1570 return;
1571
418e122c 1572 ossl_quic_stream_map_schedule_stop_sending(txp->args.qsm, s);
a73078b7
HL
1573 }
1574 break;
1575 case OSSL_QUIC_FRAME_TYPE_RESET_STREAM:
1576 {
1577 QUIC_STREAM *s
1578 = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1579
1580 if (s == NULL)
1581 return;
1582
1583 s->want_reset_stream = 1;
1584 ossl_quic_stream_map_update_state(txp->args.qsm, s);
1585 }
1586 break;
1587 default:
1588 assert(0);
1589 break;
1590 }
1591}
1592
faebafda
HL
1593static int txp_pkt_init(struct txp_pkt *pkt, OSSL_QUIC_TX_PACKETISER *txp,
1594 uint32_t enc_level, uint32_t archetype,
1595 size_t running_total)
1596{
1597 if (!txp_determine_geometry(txp, archetype, enc_level,
1598 running_total, &pkt->phdr, &pkt->geom))
1599 return 0;
1600
1601 /*
1602 * Initialise TX helper. If we must be ACK eliciting, reserve 1 byte for
1603 * PING.
1604 */
1605 if (!tx_helper_init(&pkt->h, txp, enc_level,
1606 pkt->geom.cmppl,
1607 pkt->geom.adata.require_ack_eliciting ? 1 : 0))
1608 return 0;
1609
1610 pkt->h_valid = 1;
1611 pkt->tpkt = NULL;
1612 pkt->stream_head = NULL;
c5cb85b6 1613 pkt->force_pad = 0;
faebafda
HL
1614 return 1;
1615}
1616
1617static void txp_pkt_cleanup(struct txp_pkt *pkt, OSSL_QUIC_TX_PACKETISER *txp)
1618{
1619 if (!pkt->h_valid)
1620 return;
1621
1622 tx_helper_cleanup(&pkt->h);
1623 pkt->h_valid = 0;
1624
1625 if (pkt->tpkt != NULL) {
1626 ossl_quic_txpim_pkt_release(txp->args.txpim, pkt->tpkt);
1627 pkt->tpkt = NULL;
1628 }
1629}
1630
1631static int txp_pkt_postgen_update_pkt_overhead(struct txp_pkt *pkt,
1632 OSSL_QUIC_TX_PACKETISER *txp)
1633{
1634 /*
1635 * After we have staged and generated our packets, but before we commit
1636 * them, it is possible for the estimated packet overhead (packet header +
1637 * AEAD tag size) to shrink slightly because we generated a short packet
1638 * whose which can be represented in fewer bytes as a variable-length
1639 * integer than we were (pessimistically) budgeting for. We need to account
1640 * for this to ensure that we get our padding calculation exactly right.
1641 *
1642 * Update pkt_overhead to be accurate now that we know how much data is
1643 * going in a packet.
1644 */
1645 size_t hdr_len, ciphertext_len;
1646
1647 if (pkt->h.enc_level == QUIC_ENC_LEVEL_INITIAL)
1648 /*
1649 * Don't update overheads for the INITIAL EL - we have not finished
1650 * appending padding to it and would potentially miscalculate the
1651 * correct padding if we now update the pkt_overhead field to switch to
1652 * e.g. a 1-byte length field in the packet header. Since we are padding
1653 * to QUIC_MIN_INITIAL_DGRAM_LEN which requires a 2-byte length field,
1654 * this is guaranteed to be moot anyway. See comment in
1655 * txp_determine_geometry for more information.
1656 */
1657 return 1;
1658
1659 if (!ossl_qtx_calculate_ciphertext_payload_len(txp->args.qtx, pkt->h.enc_level,
1660 pkt->h.bytes_appended,
1661 &ciphertext_len))
1662 return 0;
1663
1664 pkt->phdr.len = ciphertext_len;
1665
1666 hdr_len = ossl_quic_wire_get_encoded_pkt_hdr_len(pkt->phdr.dst_conn_id.id_len,
1667 &pkt->phdr);
1668
1669 pkt->geom.pkt_overhead = hdr_len + ciphertext_len - pkt->h.bytes_appended;
1670 return 1;
1671}
1672
9cacba43
HL
1673static void on_confirm_notify(uint64_t frame_type, uint64_t stream_id,
1674 QUIC_TXPIM_PKT *pkt, void *arg)
1675{
1676 OSSL_QUIC_TX_PACKETISER *txp = arg;
1677
1678 switch (frame_type) {
1679 case OSSL_QUIC_FRAME_TYPE_STOP_SENDING:
1680 {
1681 QUIC_STREAM *s
1682 = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1683
1684 if (s == NULL)
1685 return;
1686
1687 s->acked_stop_sending = 1;
1688 ossl_quic_stream_map_update_state(txp->args.qsm, s);
1689 }
1690 break;
1691 case OSSL_QUIC_FRAME_TYPE_RESET_STREAM:
1692 {
1693 QUIC_STREAM *s
1694 = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1695
1696 if (s == NULL)
1697 return;
1698
2f018d14
HL
1699 /*
1700 * We must already be in RESET_SENT or RESET_RECVD if we are
1701 * here, so we don't need to check state here.
1702 */
1703 ossl_quic_stream_map_notify_reset_stream_acked(txp->args.qsm, s);
9cacba43
HL
1704 ossl_quic_stream_map_update_state(txp->args.qsm, s);
1705 }
1706 break;
1707 default:
1708 assert(0);
1709 break;
1710 }
1711}
1712
faebafda
HL
1713static int txp_pkt_append_padding(struct txp_pkt *pkt,
1714 OSSL_QUIC_TX_PACKETISER *txp, size_t num_bytes)
1715{
1716 WPACKET *wpkt;
1717
1718 if (num_bytes == 0)
1719 return 1;
1720
1721 if (!ossl_assert(pkt->h_valid))
1722 return 0;
1723
1724 if (!ossl_assert(pkt->tpkt != NULL))
1725 return 0;
1726
1727 wpkt = tx_helper_begin(&pkt->h);
1728 if (wpkt == NULL)
1729 return 0;
1730
1731 if (!ossl_quic_wire_encode_padding(wpkt, num_bytes)) {
1732 tx_helper_rollback(&pkt->h);
1733 return 0;
1734 }
1735
1736 if (!tx_helper_commit(&pkt->h))
1737 return 0;
1738
1739 pkt->tpkt->ackm_pkt.num_bytes += num_bytes;
1740 /* Cannot be non-inflight if we have a PADDING frame */
1741 pkt->tpkt->ackm_pkt.is_inflight = 1;
1742 return 1;
1743}
1744
5d27e7e9
HL
1745static void on_sstream_updated(uint64_t stream_id, void *arg)
1746{
1747 OSSL_QUIC_TX_PACKETISER *txp = arg;
1748 QUIC_STREAM *s;
1749
1750 s = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1751 if (s == NULL)
1752 return;
1753
1754 ossl_quic_stream_map_update_state(txp->args.qsm, s);
1755}
1756
50e76846
P
1757/*
1758 * Returns 1 if we can send that many bytes in closing state, 0 otherwise.
1759 * Also maintains the bytes sent state if it returns a success.
1760 */
1761static int try_commit_conn_close(OSSL_QUIC_TX_PACKETISER *txp, size_t n)
1762{
1763 int res;
1764
1765 /* We can always send the first connection close frame */
1766 if (txp->closing_bytes_recv == 0)
1767 return 1;
1768
1769 /*
1770 * RFC 9000 s. 10.2.1 Closing Connection State:
1771 * To avoid being used for an amplification attack, such
1772 * endpoints MUST limit the cumulative size of packets it sends
1773 * to three times the cumulative size of the packets that are
1774 * received and attributed to the connection.
1775 * and:
1776 * An endpoint in the closing state MUST either discard packets
1777 * received from an unvalidated address or limit the cumulative
1778 * size of packets it sends to an unvalidated address to three
1779 * times the size of packets it receives from that address.
1780 */
1781 res = txp->closing_bytes_xmit + n <= txp->closing_bytes_recv * 3;
1782
1783 /*
1784 * Attribute the bytes to the connection, if we are allowed to send them
1785 * and this isn't the first closing frame.
1786 */
1787 if (res && txp->closing_bytes_recv != 0)
1788 txp->closing_bytes_xmit += n;
1789 return res;
1790}
1791
1792void ossl_quic_tx_packetiser_record_received_closing_bytes(
1793 OSSL_QUIC_TX_PACKETISER *txp, size_t n)
1794{
1795 txp->closing_bytes_recv += n;
1796}
1797
a73078b7 1798static int txp_generate_pre_token(OSSL_QUIC_TX_PACKETISER *txp,
faebafda 1799 struct txp_pkt *pkt,
178c104d
HL
1800 int chosen_for_conn_close,
1801 int *can_be_non_inflight)
a73078b7 1802{
faebafda
HL
1803 const uint32_t enc_level = pkt->h.enc_level;
1804 const uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
1805 const struct archetype_data *a = &pkt->geom.adata;
1806 QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
1807 struct tx_helper *h = &pkt->h;
a73078b7
HL
1808 const OSSL_QUIC_FRAME_ACK *ack;
1809 OSSL_QUIC_FRAME_ACK ack2;
1810
1811 tpkt->ackm_pkt.largest_acked = QUIC_PN_INVALID;
1812
1813 /* ACK Frames (Regenerate) */
1814 if (a->allow_ack
1815 && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_ACK
d13488b9 1816 && (((txp->want_ack & (1UL << pn_space)) != 0)
a73078b7
HL
1817 || ossl_ackm_is_ack_desired(txp->args.ackm, pn_space))
1818 && (ack = ossl_ackm_get_ack_frame(txp->args.ackm, pn_space)) != NULL) {
1819 WPACKET *wpkt = tx_helper_begin(h);
1820
1821 if (wpkt == NULL)
1822 return 0;
1823
1824 /* We do not currently support ECN */
1825 ack2 = *ack;
1826 ack2.ecn_present = 0;
1827
1828 if (ossl_quic_wire_encode_frame_ack(wpkt,
1829 txp->args.ack_delay_exponent,
1830 &ack2)) {
1831 if (!tx_helper_commit(h))
1832 return 0;
1833
1834 tpkt->had_ack_frame = 1;
1835
1836 if (ack->num_ack_ranges > 0)
1837 tpkt->ackm_pkt.largest_acked = ack->ack_ranges[0].end;
8f9c9213
HL
1838
1839 if (txp->ack_tx_cb != NULL)
1840 txp->ack_tx_cb(&ack2, pn_space, txp->ack_tx_cb_arg);
a73078b7
HL
1841 } else {
1842 tx_helper_rollback(h);
1843 }
1844 }
1845
1846 /* CONNECTION_CLOSE Frames (Regenerate) */
7f9d1249 1847 if (a->allow_conn_close && txp->want_conn_close && chosen_for_conn_close) {
a73078b7 1848 WPACKET *wpkt = tx_helper_begin(h);
96fa10f3 1849 OSSL_QUIC_FRAME_CONN_CLOSE f, *pf = &txp->conn_close_frame;
50e76846 1850 size_t l;
a73078b7
HL
1851
1852 if (wpkt == NULL)
1853 return 0;
1854
96fa10f3
HL
1855 /*
1856 * Application CONNECTION_CLOSE frames may only be sent in the
1857 * Application PN space, as otherwise they may be sent before a
1858 * connection is authenticated and leak application data. Therefore, if
1859 * we need to send a CONNECTION_CLOSE frame in another PN space and were
1860 * given an application CONNECTION_CLOSE frame, convert it into a
1861 * transport CONNECTION_CLOSE frame, removing any sensitive application
1862 * data.
1863 *
1864 * RFC 9000 s. 10.2.3: "A CONNECTION_CLOSE of type 0x1d MUST be replaced
1865 * by a CONNECTION_CLOSE of type 0x1c when sending the frame in Initial
1866 * or Handshake packets. Otherwise, information about the application
1867 * state might be revealed. Endpoints MUST clear the value of the Reason
1868 * Phrase field and SHOULD use the APPLICATION_ERROR code when
1869 * converting to a CONNECTION_CLOSE of type 0x1c."
1870 */
1871 if (pn_space != QUIC_PN_SPACE_APP && pf->is_app) {
1872 pf = &f;
1873 pf->is_app = 0;
1874 pf->frame_type = 0;
1875 pf->error_code = QUIC_ERR_APPLICATION_ERROR;
1876 pf->reason = NULL;
1877 pf->reason_len = 0;
1878 }
1879
50e76846
P
1880 if (ossl_quic_wire_encode_frame_conn_close(wpkt, pf)
1881 && WPACKET_get_total_written(wpkt, &l)
1882 && try_commit_conn_close(txp, l)) {
a73078b7
HL
1883 if (!tx_helper_commit(h))
1884 return 0;
178c104d 1885
0b31072e 1886 tpkt->had_conn_close = 1;
178c104d 1887 *can_be_non_inflight = 0;
a73078b7
HL
1888 } else {
1889 tx_helper_rollback(h);
1890 }
1891 }
1892
1893 return 1;
1894}
1895
1896static int try_len(size_t space_left, size_t orig_len,
1897 size_t base_hdr_len, size_t lenbytes,
1898 uint64_t maxn, size_t *hdr_len, size_t *payload_len)
1899{
1900 size_t n;
1901 size_t maxn_ = maxn > SIZE_MAX ? SIZE_MAX : (size_t)maxn;
1902
1903 *hdr_len = base_hdr_len + lenbytes;
1904
cf06f347
HL
1905 if (orig_len == 0 && space_left >= *hdr_len) {
1906 *payload_len = 0;
1907 return 1;
1908 }
1909
a73078b7
HL
1910 n = orig_len;
1911 if (n > maxn_)
1912 n = maxn_;
1913 if (n + *hdr_len > space_left)
1914 n = (space_left >= *hdr_len) ? space_left - *hdr_len : 0;
1915
1916 *payload_len = n;
1917 return n > 0;
1918}
1919
cf06f347
HL
1920static int determine_len(size_t space_left, size_t orig_len,
1921 size_t base_hdr_len,
1922 uint64_t *hlen, uint64_t *len)
a73078b7 1923{
cf06f347 1924 int ok = 0;
a73078b7
HL
1925 size_t chosen_payload_len = 0;
1926 size_t chosen_hdr_len = 0;
1927 size_t payload_len[4], hdr_len[4];
1928 int i, valid[4] = {0};
1929
1930 valid[0] = try_len(space_left, orig_len, base_hdr_len,
1931 1, OSSL_QUIC_VLINT_1B_MAX,
1932 &hdr_len[0], &payload_len[0]);
1933 valid[1] = try_len(space_left, orig_len, base_hdr_len,
1934 2, OSSL_QUIC_VLINT_2B_MAX,
1935 &hdr_len[1], &payload_len[1]);
1936 valid[2] = try_len(space_left, orig_len, base_hdr_len,
1937 4, OSSL_QUIC_VLINT_4B_MAX,
1938 &hdr_len[2], &payload_len[2]);
1939 valid[3] = try_len(space_left, orig_len, base_hdr_len,
1940 8, OSSL_QUIC_VLINT_8B_MAX,
1941 &hdr_len[3], &payload_len[3]);
1942
1943 for (i = OSSL_NELEM(valid) - 1; i >= 0; --i)
1944 if (valid[i] && payload_len[i] >= chosen_payload_len) {
1945 chosen_payload_len = payload_len[i];
1946 chosen_hdr_len = hdr_len[i];
cf06f347 1947 ok = 1;
a73078b7
HL
1948 }
1949
1950 *hlen = chosen_hdr_len;
1951 *len = chosen_payload_len;
cf06f347 1952 return ok;
a73078b7
HL
1953}
1954
1955/*
1956 * Given a CRYPTO frame header with accurate chdr->len and a budget
1957 * (space_left), try to find the optimal value of chdr->len to fill as much of
1958 * the budget as possible. This is slightly hairy because larger values of
1959 * chdr->len cause larger encoded sizes of the length field of the frame, which
1960 * in turn mean less space available for payload data. We check all possible
1961 * encodings and choose the optimal encoding.
1962 */
1963static int determine_crypto_len(struct tx_helper *h,
1964 OSSL_QUIC_FRAME_CRYPTO *chdr,
1965 size_t space_left,
1966 uint64_t *hlen,
1967 uint64_t *len)
1968{
1969 size_t orig_len;
1970 size_t base_hdr_len; /* CRYPTO header length without length field */
1971
1972 if (chdr->len > SIZE_MAX)
1973 return 0;
1974
1975 orig_len = (size_t)chdr->len;
1976
1977 chdr->len = 0;
1978 base_hdr_len = ossl_quic_wire_get_encoded_frame_len_crypto_hdr(chdr);
1979 chdr->len = orig_len;
1980 if (base_hdr_len == 0)
1981 return 0;
1982
1983 --base_hdr_len;
1984
cf06f347 1985 return determine_len(space_left, orig_len, base_hdr_len, hlen, len);
a73078b7
HL
1986}
1987
1988static int determine_stream_len(struct tx_helper *h,
1989 OSSL_QUIC_FRAME_STREAM *shdr,
1990 size_t space_left,
1991 uint64_t *hlen,
1992 uint64_t *len)
1993{
1994 size_t orig_len;
1995 size_t base_hdr_len; /* STREAM header length without length field */
1996
1997 if (shdr->len > SIZE_MAX)
1998 return 0;
1999
2000 orig_len = (size_t)shdr->len;
2001
2002 shdr->len = 0;
2003 base_hdr_len = ossl_quic_wire_get_encoded_frame_len_stream_hdr(shdr);
2004 shdr->len = orig_len;
2005 if (base_hdr_len == 0)
2006 return 0;
2007
2008 if (shdr->has_explicit_len)
2009 --base_hdr_len;
2010
cf06f347 2011 return determine_len(space_left, orig_len, base_hdr_len, hlen, len);
a73078b7
HL
2012}
2013
2014static int txp_generate_crypto_frames(OSSL_QUIC_TX_PACKETISER *txp,
faebafda 2015 struct txp_pkt *pkt,
091f532e 2016 int *have_ack_eliciting)
a73078b7 2017{
faebafda
HL
2018 const uint32_t enc_level = pkt->h.enc_level;
2019 const uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
2020 QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
2021 struct tx_helper *h = &pkt->h;
a73078b7
HL
2022 size_t num_stream_iovec;
2023 OSSL_QUIC_FRAME_STREAM shdr = {0};
2024 OSSL_QUIC_FRAME_CRYPTO chdr = {0};
2025 OSSL_QTX_IOVEC iov[2];
2026 uint64_t hdr_bytes;
2027 WPACKET *wpkt;
24c1be5c 2028 QUIC_TXPIM_CHUNK chunk = {0};
a73078b7
HL
2029 size_t i, space_left;
2030
2031 for (i = 0;; ++i) {
2032 space_left = tx_helper_get_space_left(h);
2033
2034 if (space_left < MIN_FRAME_SIZE_CRYPTO)
2035 return 1; /* no point trying */
2036
2037 /* Do we have any CRYPTO data waiting? */
2038 num_stream_iovec = OSSL_NELEM(iov);
2039 if (!ossl_quic_sstream_get_stream_frame(txp->args.crypto[pn_space],
2040 i, &shdr, iov,
2041 &num_stream_iovec))
2042 return 1; /* nothing to do */
2043
2044 /* Convert STREAM frame header to CRYPTO frame header */
2045 chdr.offset = shdr.offset;
2046 chdr.len = shdr.len;
2047
2048 if (chdr.len == 0)
2049 return 1; /* nothing to do */
2050
2051 /* Find best fit (header length, payload length) combination. */
2052 if (!determine_crypto_len(h, &chdr, space_left, &hdr_bytes,
cf06f347 2053 &chdr.len))
a73078b7 2054 return 1; /* can't fit anything */
a73078b7
HL
2055
2056 /*
2057 * Truncate IOVs to match our chosen length.
2058 *
2059 * The length cannot be more than SIZE_MAX because this length comes
2060 * from our send stream buffer.
2061 */
2062 ossl_quic_sstream_adjust_iov((size_t)chdr.len, iov, num_stream_iovec);
2063
2064 /*
2065 * Ensure we have enough iovecs allocated (1 for the header, up to 2 for
ad31628c 2066 * the stream data.)
a73078b7 2067 */
faebafda 2068 if (!txp_el_ensure_iovec(&txp->el[enc_level], h->num_iovec + 3))
a73078b7
HL
2069 return 0; /* alloc error */
2070
2071 /* Encode the header. */
2072 wpkt = tx_helper_begin(h);
2073 if (wpkt == NULL)
2074 return 0; /* alloc error */
2075
2076 if (!ossl_quic_wire_encode_frame_crypto_hdr(wpkt, &chdr)) {
2077 tx_helper_rollback(h);
2078 return 1; /* can't fit */
2079 }
2080
2081 if (!tx_helper_commit(h))
2082 return 0; /* alloc error */
2083
2084 /* Add payload iovecs to the helper (infallible). */
2085 for (i = 0; i < num_stream_iovec; ++i)
2086 tx_helper_append_iovec(h, iov[i].buf, iov[i].buf_len);
2087
2088 *have_ack_eliciting = 1;
2089 tx_helper_unrestrict(h); /* no longer need PING */
2090
2091 /* Log chunk to TXPIM. */
2092 chunk.stream_id = UINT64_MAX; /* crypto stream */
2093 chunk.start = chdr.offset;
2094 chunk.end = chdr.offset + chdr.len - 1;
2095 chunk.has_fin = 0; /* Crypto stream never ends */
2096 if (!ossl_quic_txpim_pkt_append_chunk(tpkt, &chunk))
2097 return 0; /* alloc error */
2098 }
2099}
2100
2101struct chunk_info {
2102 OSSL_QUIC_FRAME_STREAM shdr;
e718b248 2103 uint64_t orig_len;
a73078b7
HL
2104 OSSL_QTX_IOVEC iov[2];
2105 size_t num_stream_iovec;
091f532e 2106 int valid;
a73078b7
HL
2107};
2108
2109static int txp_plan_stream_chunk(OSSL_QUIC_TX_PACKETISER *txp,
2110 struct tx_helper *h,
2111 QUIC_SSTREAM *sstream,
2112 QUIC_TXFC *stream_txfc,
2113 size_t skip,
2114 struct chunk_info *chunk)
2115{
2116 uint64_t fc_credit, fc_swm, fc_limit;
2117
2118 chunk->num_stream_iovec = OSSL_NELEM(chunk->iov);
2119 chunk->valid = ossl_quic_sstream_get_stream_frame(sstream, skip,
2120 &chunk->shdr,
2121 chunk->iov,
2122 &chunk->num_stream_iovec);
2123 if (!chunk->valid)
2124 return 1;
2125
2126 if (!ossl_assert(chunk->shdr.len > 0 || chunk->shdr.is_fin))
2127 /* Should only have 0-length chunk if FIN */
2128 return 0;
2129
e718b248
MC
2130 chunk->orig_len = chunk->shdr.len;
2131
a73078b7
HL
2132 /* Clamp according to connection and stream-level TXFC. */
2133 fc_credit = ossl_quic_txfc_get_credit(stream_txfc);
2134 fc_swm = ossl_quic_txfc_get_swm(stream_txfc);
2135 fc_limit = fc_swm + fc_credit;
2136
2137 if (chunk->shdr.len > 0 && chunk->shdr.offset + chunk->shdr.len > fc_limit) {
2138 chunk->shdr.len = (fc_limit <= chunk->shdr.offset)
2139 ? 0 : fc_limit - chunk->shdr.offset;
2140 chunk->shdr.is_fin = 0;
2141 }
2142
2143 if (chunk->shdr.len == 0 && !chunk->shdr.is_fin) {
2144 /*
2145 * Nothing to do due to TXFC. Since SSTREAM returns chunks in ascending
2146 * order of offset we don't need to check any later chunks, so stop
2147 * iterating here.
2148 */
2149 chunk->valid = 0;
2150 return 1;
2151 }
2152
2153 return 1;
2154}
2155
2156/*
2157 * Returns 0 on fatal error (e.g. allocation failure), 1 on success.
2158 * *packet_full is set to 1 if there is no longer enough room for another STREAM
26652896 2159 * frame.
a73078b7
HL
2160 */
2161static int txp_generate_stream_frames(OSSL_QUIC_TX_PACKETISER *txp,
faebafda 2162 struct txp_pkt *pkt,
a73078b7
HL
2163 uint64_t id,
2164 QUIC_SSTREAM *sstream,
2165 QUIC_TXFC *stream_txfc,
2166 QUIC_STREAM *next_stream,
091f532e
HL
2167 int *have_ack_eliciting,
2168 int *packet_full,
a73078b7
HL
2169 uint64_t *new_credit_consumed)
2170{
2171 int rc = 0;
2172 struct chunk_info chunks[2] = {0};
faebafda
HL
2173 const uint32_t enc_level = pkt->h.enc_level;
2174 QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
2175 struct tx_helper *h = &pkt->h;
a73078b7
HL
2176 OSSL_QUIC_FRAME_STREAM *shdr;
2177 WPACKET *wpkt;
2178 QUIC_TXPIM_CHUNK chunk;
2179 size_t i, j, space_left;
3bef14c5 2180 int can_fill_payload, use_explicit_len;
a73078b7 2181 int could_have_following_chunk;
05f97354 2182 uint64_t orig_len;
a73078b7
HL
2183 uint64_t hdr_len_implicit, payload_len_implicit;
2184 uint64_t hdr_len_explicit, payload_len_explicit;
2185 uint64_t fc_swm, fc_new_hwm;
2186
2187 fc_swm = ossl_quic_txfc_get_swm(stream_txfc);
2188 fc_new_hwm = fc_swm;
2189
2190 /*
2191 * Load the first two chunks if any offered by the send stream. We retrieve
2192 * the next chunk in advance so we can determine if we need to send any more
2193 * chunks from the same stream after this one, which is needed when
2194 * determining when we can use an implicit length in a STREAM frame.
2195 */
2196 for (i = 0; i < 2; ++i) {
2197 if (!txp_plan_stream_chunk(txp, h, sstream, stream_txfc, i, &chunks[i]))
2198 goto err;
2199
2200 if (i == 0 && !chunks[i].valid) {
2201 /* No chunks, nothing to do. */
a73078b7
HL
2202 rc = 1;
2203 goto err;
2204 }
2205 }
2206
2207 for (i = 0;; ++i) {
2208 space_left = tx_helper_get_space_left(h);
2209
cf06f347
HL
2210 if (!chunks[i % 2].valid) {
2211 /* Out of chunks; we're done. */
a73078b7
HL
2212 rc = 1;
2213 goto err;
2214 }
2215
cf06f347
HL
2216 if (space_left < MIN_FRAME_SIZE_STREAM) {
2217 *packet_full = 1;
a73078b7
HL
2218 rc = 1;
2219 goto err;
2220 }
2221
2222 if (!ossl_assert(!h->done_implicit))
2223 /*
2224 * Logic below should have ensured we didn't append an
2225 * implicit-length unless we filled the packet or didn't have
2226 * another stream to handle, so this should not be possible.
2227 */
2228 goto err;
2229
2230 shdr = &chunks[i % 2].shdr;
e718b248 2231 orig_len = chunks[i % 2].orig_len;
a73078b7
HL
2232 if (i > 0)
2233 /* Load next chunk for lookahead. */
2234 if (!txp_plan_stream_chunk(txp, h, sstream, stream_txfc, i + 1,
2235 &chunks[(i + 1) % 2]))
2236 goto err;
2237
2238 /*
2239 * Find best fit (header length, payload length) combination for if we
2240 * use an implicit length.
2241 */
2242 shdr->has_explicit_len = 0;
2243 hdr_len_implicit = payload_len_implicit = 0;
2244 if (!determine_stream_len(h, shdr, space_left,
cf06f347 2245 &hdr_len_implicit, &payload_len_implicit)) {
a73078b7
HL
2246 *packet_full = 1;
2247 rc = 1;
2248 goto err; /* can't fit anything */
2249 }
2250
a73078b7
HL
2251 /*
2252 * If there is a next stream, we don't use the implicit length so we can
2253 * add more STREAM frames after this one, unless there is enough data
2254 * for this STREAM frame to fill the packet.
2255 */
2256 can_fill_payload = (hdr_len_implicit + payload_len_implicit
2257 >= space_left);
2258
2259 /*
2260 * Is there is a stream after this one, or another chunk pending
2261 * transmission in this stream?
2262 */
2263 could_have_following_chunk
2264 = (next_stream != NULL || chunks[(i + 1) % 2].valid);
2265
2266 /* Choose between explicit or implicit length representations. */
2267 use_explicit_len = !((can_fill_payload || !could_have_following_chunk)
3bef14c5 2268 && !pkt->force_pad);
a73078b7
HL
2269
2270 if (use_explicit_len) {
2271 /*
2272 * Find best fit (header length, payload length) combination for if
2273 * we use an explicit length.
2274 */
2275 shdr->has_explicit_len = 1;
2276 hdr_len_explicit = payload_len_explicit = 0;
2277 if (!determine_stream_len(h, shdr, space_left,
cf06f347 2278 &hdr_len_explicit, &payload_len_explicit)) {
a73078b7
HL
2279 *packet_full = 1;
2280 rc = 1;
2281 goto err; /* can't fit anything */
2282 }
2283
2284 shdr->len = payload_len_explicit;
2285 } else {
7fe30104 2286 *packet_full = 1;
a73078b7
HL
2287 shdr->has_explicit_len = 0;
2288 shdr->len = payload_len_implicit;
2289 }
2290
cf06f347
HL
2291 /* If this is a FIN, don't keep filling the packet with more FINs. */
2292 if (shdr->is_fin)
2293 chunks[(i + 1) % 2].valid = 0;
2294
553122cd
HL
2295 /*
2296 * We are now committed to our length (shdr->len can't change).
2297 * If we truncated the chunk, clear the FIN bit.
2298 */
2299 if (shdr->len < orig_len)
2300 shdr->is_fin = 0;
2301
a73078b7
HL
2302 /* Truncate IOVs to match our chosen length. */
2303 ossl_quic_sstream_adjust_iov((size_t)shdr->len, chunks[i % 2].iov,
2304 chunks[i % 2].num_stream_iovec);
2305
2306 /*
2307 * Ensure we have enough iovecs allocated (1 for the header, up to 2 for
ad31628c 2308 * the stream data.)
a73078b7 2309 */
faebafda 2310 if (!txp_el_ensure_iovec(&txp->el[enc_level], h->num_iovec + 3))
a73078b7
HL
2311 goto err; /* alloc error */
2312
2313 /* Encode the header. */
2314 wpkt = tx_helper_begin(h);
2315 if (wpkt == NULL)
2316 goto err; /* alloc error */
2317
2318 shdr->stream_id = id;
2319 if (!ossl_assert(ossl_quic_wire_encode_frame_stream_hdr(wpkt, shdr))) {
2320 /* (Should not be possible.) */
2321 tx_helper_rollback(h);
2322 *packet_full = 1;
2323 rc = 1;
2324 goto err; /* can't fit */
2325 }
2326
2327 if (!tx_helper_commit(h))
2328 goto err; /* alloc error */
2329
2330 /* Add payload iovecs to the helper (infallible). */
2331 for (j = 0; j < chunks[i % 2].num_stream_iovec; ++j)
2332 tx_helper_append_iovec(h, chunks[i % 2].iov[j].buf,
2333 chunks[i % 2].iov[j].buf_len);
2334
2335 *have_ack_eliciting = 1;
2336 tx_helper_unrestrict(h); /* no longer need PING */
2337 if (!shdr->has_explicit_len)
2338 h->done_implicit = 1;
2339
2340 /* Log new TXFC credit which was consumed. */
2341 if (shdr->len > 0 && shdr->offset + shdr->len > fc_new_hwm)
2342 fc_new_hwm = shdr->offset + shdr->len;
2343
2344 /* Log chunk to TXPIM. */
2345 chunk.stream_id = shdr->stream_id;
2346 chunk.start = shdr->offset;
2347 chunk.end = shdr->offset + shdr->len - 1;
2348 chunk.has_fin = shdr->is_fin;
2349 chunk.has_stop_sending = 0;
2350 chunk.has_reset_stream = 0;
2351 if (!ossl_quic_txpim_pkt_append_chunk(tpkt, &chunk))
2352 goto err; /* alloc error */
05f97354
HL
2353
2354 if (shdr->len < orig_len) {
2355 /*
2356 * If we did not serialize all of this chunk we definitely do not
e718b248 2357 * want to try the next chunk
05f97354
HL
2358 */
2359 rc = 1;
2360 goto err;
2361 }
a73078b7
HL
2362 }
2363
2364err:
2365 *new_credit_consumed = fc_new_hwm - fc_swm;
2366 return rc;
2367}
2368
2369static void txp_enlink_tmp(QUIC_STREAM **tmp_head, QUIC_STREAM *stream)
2370{
2371 stream->txp_next = *tmp_head;
2372 *tmp_head = stream;
2373}
2374
2375static int txp_generate_stream_related(OSSL_QUIC_TX_PACKETISER *txp,
faebafda 2376 struct txp_pkt *pkt,
091f532e 2377 int *have_ack_eliciting,
a73078b7
HL
2378 QUIC_STREAM **tmp_head)
2379{
2380 QUIC_STREAM_ITER it;
a73078b7
HL
2381 WPACKET *wpkt;
2382 uint64_t cwm;
2383 QUIC_STREAM *stream, *snext;
faebafda 2384 struct tx_helper *h = &pkt->h;
a73078b7
HL
2385
2386 for (ossl_quic_stream_iter_init(&it, txp->args.qsm, 1);
2387 it.stream != NULL;) {
2388
2389 stream = it.stream;
2390 ossl_quic_stream_iter_next(&it);
2391 snext = it.stream;
2392
2393 stream->txp_sent_fc = 0;
2394 stream->txp_sent_stop_sending = 0;
2395 stream->txp_sent_reset_stream = 0;
a73078b7
HL
2396 stream->txp_blocked = 0;
2397 stream->txp_txfc_new_credit_consumed = 0;
2398
a73078b7
HL
2399 /* Stream Abort Frames (STOP_SENDING, RESET_STREAM) */
2400 if (stream->want_stop_sending) {
2401 OSSL_QUIC_FRAME_STOP_SENDING f;
2402
2403 wpkt = tx_helper_begin(h);
2404 if (wpkt == NULL)
2405 return 0; /* alloc error */
2406
2407 f.stream_id = stream->id;
2408 f.app_error_code = stream->stop_sending_aec;
2409 if (!ossl_quic_wire_encode_frame_stop_sending(wpkt, &f)) {
2410 tx_helper_rollback(h); /* can't fit */
2411 txp_enlink_tmp(tmp_head, stream);
2412 break;
2413 }
2414
2415 if (!tx_helper_commit(h))
2416 return 0; /* alloc error */
2417
2418 *have_ack_eliciting = 1;
2419 tx_helper_unrestrict(h); /* no longer need PING */
2420 stream->txp_sent_stop_sending = 1;
2421 }
2422
2423 if (stream->want_reset_stream) {
2424 OSSL_QUIC_FRAME_RESET_STREAM f;
2425
96b7df60
HL
2426 if (!ossl_assert(stream->send_state == QUIC_SSTREAM_STATE_RESET_SENT))
2427 return 0;
01715f2b 2428
a73078b7
HL
2429 wpkt = tx_helper_begin(h);
2430 if (wpkt == NULL)
2431 return 0; /* alloc error */
2432
2433 f.stream_id = stream->id;
2434 f.app_error_code = stream->reset_stream_aec;
28d0e35c
HL
2435 if (!ossl_quic_stream_send_get_final_size(stream, &f.final_size))
2436 return 0; /* should not be possible */
2437
a73078b7
HL
2438 if (!ossl_quic_wire_encode_frame_reset_stream(wpkt, &f)) {
2439 tx_helper_rollback(h); /* can't fit */
2440 txp_enlink_tmp(tmp_head, stream);
2441 break;
2442 }
2443
2444 if (!tx_helper_commit(h))
2445 return 0; /* alloc error */
2446
2447 *have_ack_eliciting = 1;
2448 tx_helper_unrestrict(h); /* no longer need PING */
2449 stream->txp_sent_reset_stream = 1;
7e3fa44f
HL
2450
2451 /*
2452 * The final size of the stream as indicated by RESET_STREAM is used
2453 * to ensure a consistent view of flow control state by both
2454 * parties; if we happen to send a RESET_STREAM that consumes more
2455 * flow control credit, make sure we account for that.
2456 */
96b7df60
HL
2457 if (!ossl_assert(f.final_size <= ossl_quic_txfc_get_swm(&stream->txfc)))
2458 return 0;
7e3fa44f
HL
2459
2460 stream->txp_txfc_new_credit_consumed
2461 = f.final_size - ossl_quic_txfc_get_swm(&stream->txfc);
a73078b7
HL
2462 }
2463
22f21fbd
HL
2464 /*
2465 * Stream Flow Control Frames (MAX_STREAM_DATA)
2466 *
2467 * RFC 9000 s. 13.3: "An endpoint SHOULD stop sending MAX_STREAM_DATA
2468 * frames when the receiving part of the stream enters a "Size Known" or
2469 * "Reset Recvd" state." -- In practice, RECV is the only state
2470 * in which it makes sense to generate more MAX_STREAM_DATA frames.
2471 */
2472 if (stream->recv_state == QUIC_RSTREAM_STATE_RECV
a73078b7
HL
2473 && (stream->want_max_stream_data
2474 || ossl_quic_rxfc_has_cwm_changed(&stream->rxfc, 0))) {
2475
2476 wpkt = tx_helper_begin(h);
2477 if (wpkt == NULL)
2478 return 0; /* alloc error */
2479
2480 cwm = ossl_quic_rxfc_get_cwm(&stream->rxfc);
2481
2482 if (!ossl_quic_wire_encode_frame_max_stream_data(wpkt, stream->id,
2483 cwm)) {
2484 tx_helper_rollback(h); /* can't fit */
2485 txp_enlink_tmp(tmp_head, stream);
2486 break;
2487 }
2488
2489 if (!tx_helper_commit(h))
2490 return 0; /* alloc error */
2491
2492 *have_ack_eliciting = 1;
2493 tx_helper_unrestrict(h); /* no longer need PING */
2494 stream->txp_sent_fc = 1;
2495 }
2496
01715f2b
HL
2497 /*
2498 * Stream Data Frames (STREAM)
2499 *
2500 * RFC 9000 s. 3.3: A sender MUST NOT send a STREAM [...] frame for a
2501 * stream in the "Reset Sent" state [or any terminal state]. We don't
1d547f8f 2502 * send any more STREAM frames if we are sending, have sent, or are
01715f2b
HL
2503 * planning to send, RESET_STREAM. The other terminal state is Data
2504 * Recvd, but txp_generate_stream_frames() is guaranteed to generate
2505 * nothing in this case.
2506 */
2507 if (ossl_quic_stream_has_send_buffer(stream)
2508 && !ossl_quic_stream_send_is_reset(stream)) {
26652896 2509 int packet_full = 0;
a73078b7 2510
96b7df60
HL
2511 if (!ossl_assert(!stream->want_reset_stream))
2512 return 0;
7e3fa44f 2513
faebafda 2514 if (!txp_generate_stream_frames(txp, pkt,
a73078b7
HL
2515 stream->id, stream->sstream,
2516 &stream->txfc,
3bef14c5 2517 snext,
a73078b7
HL
2518 have_ack_eliciting,
2519 &packet_full,
a73078b7
HL
2520 &stream->txp_txfc_new_credit_consumed)) {
2521 /* Fatal error (allocation, etc.) */
2522 txp_enlink_tmp(tmp_head, stream);
2523 return 0;
2524 }
2525
a73078b7
HL
2526 if (packet_full) {
2527 txp_enlink_tmp(tmp_head, stream);
2528 break;
2529 }
2530 }
2531
2532 txp_enlink_tmp(tmp_head, stream);
2533 }
2534
2535 return 1;
2536}
2537
faebafda
HL
2538static int txp_generate_for_el(OSSL_QUIC_TX_PACKETISER *txp,
2539 struct txp_pkt *pkt,
2540 int chosen_for_conn_close)
a73078b7
HL
2541{
2542 int rc = TXP_ERR_SUCCESS;
faebafda
HL
2543 const uint32_t enc_level = pkt->h.enc_level;
2544 const uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
2545 int have_ack_eliciting = 0, done_pre_token = 0;
2546 const struct archetype_data a = pkt->geom.adata;
178c104d
HL
2547 /*
2548 * Cleared if we encode any non-ACK-eliciting frame type which rules out the
2549 * packet being a non-inflight frame. This means any non-ACK ACK-eliciting
2550 * frame, even PADDING frames. ACK eliciting frames always cause a packet to
2551 * become ineligible for non-inflight treatment so it is not necessary to
2552 * clear this in cases where have_ack_eliciting is set, as it is ignored in
2553 * that case.
2554 */
2555 int can_be_non_inflight = 1;
a73078b7
HL
2556 QUIC_CFQ_ITEM *cfq_item;
2557 QUIC_TXPIM_PKT *tpkt = NULL;
faebafda 2558 struct tx_helper *h = &pkt->h;
a73078b7
HL
2559
2560 /* Maximum PN reached? */
b65b0d4e 2561 if (!ossl_quic_pn_valid(txp->next_pn[pn_space]))
a73078b7
HL
2562 goto fatal_err;
2563
faebafda 2564 if (!ossl_assert(pkt->tpkt == NULL))
a73078b7
HL
2565 goto fatal_err;
2566
faebafda 2567 if ((pkt->tpkt = tpkt = ossl_quic_txpim_pkt_alloc(txp->args.txpim)) == NULL)
a73078b7
HL
2568 goto fatal_err;
2569
a73078b7
HL
2570 /*
2571 * Frame Serialization
2572 * ===================
2573 *
2574 * We now serialize frames into the packet in descending order of priority.
2575 */
2576
2577 /* HANDSHAKE_DONE (Regenerate) */
2578 if (a.allow_handshake_done && txp->want_handshake_done
faebafda
HL
2579 && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_HANDSHAKE_DONE) {
2580 WPACKET *wpkt = tx_helper_begin(h);
a73078b7
HL
2581
2582 if (wpkt == NULL)
2583 goto fatal_err;
2584
2585 if (ossl_quic_wire_encode_frame_handshake_done(wpkt)) {
2586 tpkt->had_handshake_done_frame = 1;
2587 have_ack_eliciting = 1;
2588
faebafda 2589 if (!tx_helper_commit(h))
a73078b7
HL
2590 goto fatal_err;
2591
faebafda 2592 tx_helper_unrestrict(h); /* no longer need PING */
a73078b7 2593 } else {
faebafda 2594 tx_helper_rollback(h);
a73078b7
HL
2595 }
2596 }
2597
2598 /* MAX_DATA (Regenerate) */
2599 if (a.allow_conn_fc
2600 && (txp->want_max_data
2601 || ossl_quic_rxfc_has_cwm_changed(txp->args.conn_rxfc, 0))
faebafda
HL
2602 && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_MAX_DATA) {
2603 WPACKET *wpkt = tx_helper_begin(h);
a73078b7
HL
2604 uint64_t cwm = ossl_quic_rxfc_get_cwm(txp->args.conn_rxfc);
2605
2606 if (wpkt == NULL)
2607 goto fatal_err;
2608
2609 if (ossl_quic_wire_encode_frame_max_data(wpkt, cwm)) {
2610 tpkt->had_max_data_frame = 1;
2611 have_ack_eliciting = 1;
2612
faebafda 2613 if (!tx_helper_commit(h))
a73078b7
HL
2614 goto fatal_err;
2615
faebafda 2616 tx_helper_unrestrict(h); /* no longer need PING */
a73078b7 2617 } else {
faebafda 2618 tx_helper_rollback(h);
a73078b7
HL
2619 }
2620 }
2621
2622 /* MAX_STREAMS_BIDI (Regenerate) */
a73078b7 2623 if (a.allow_conn_fc
a6b6ea17
HL
2624 && (txp->want_max_streams_bidi
2625 || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_bidi_rxfc, 0))
faebafda
HL
2626 && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_MAX_STREAMS_BIDI) {
2627 WPACKET *wpkt = tx_helper_begin(h);
a6b6ea17
HL
2628 uint64_t max_streams
2629 = ossl_quic_rxfc_get_cwm(txp->args.max_streams_bidi_rxfc);
a73078b7
HL
2630
2631 if (wpkt == NULL)
2632 goto fatal_err;
2633
2634 if (ossl_quic_wire_encode_frame_max_streams(wpkt, /*is_uni=*/0,
2635 max_streams)) {
2636 tpkt->had_max_streams_bidi_frame = 1;
2637 have_ack_eliciting = 1;
2638
faebafda 2639 if (!tx_helper_commit(h))
a73078b7
HL
2640 goto fatal_err;
2641
faebafda 2642 tx_helper_unrestrict(h); /* no longer need PING */
a73078b7 2643 } else {
faebafda 2644 tx_helper_rollback(h);
a73078b7
HL
2645 }
2646 }
2647
2648 /* MAX_STREAMS_UNI (Regenerate) */
2649 if (a.allow_conn_fc
a6b6ea17
HL
2650 && (txp->want_max_streams_uni
2651 || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_uni_rxfc, 0))
faebafda
HL
2652 && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_MAX_STREAMS_UNI) {
2653 WPACKET *wpkt = tx_helper_begin(h);
a6b6ea17
HL
2654 uint64_t max_streams
2655 = ossl_quic_rxfc_get_cwm(txp->args.max_streams_uni_rxfc);
a73078b7
HL
2656
2657 if (wpkt == NULL)
2658 goto fatal_err;
2659
2660 if (ossl_quic_wire_encode_frame_max_streams(wpkt, /*is_uni=*/1,
2661 max_streams)) {
2662 tpkt->had_max_streams_uni_frame = 1;
2663 have_ack_eliciting = 1;
2664
faebafda 2665 if (!tx_helper_commit(h))
a73078b7
HL
2666 goto fatal_err;
2667
faebafda 2668 tx_helper_unrestrict(h); /* no longer need PING */
a73078b7 2669 } else {
faebafda 2670 tx_helper_rollback(h);
a73078b7
HL
2671 }
2672 }
2673
2674 /* GCR Frames */
2675 for (cfq_item = ossl_quic_cfq_get_priority_head(txp->args.cfq, pn_space);
2676 cfq_item != NULL;
2677 cfq_item = ossl_quic_cfq_item_get_priority_next(cfq_item, pn_space)) {
2678 uint64_t frame_type = ossl_quic_cfq_item_get_frame_type(cfq_item);
2679 const unsigned char *encoded = ossl_quic_cfq_item_get_encoded(cfq_item);
2680 size_t encoded_len = ossl_quic_cfq_item_get_encoded_len(cfq_item);
2681
2682 switch (frame_type) {
2683 case OSSL_QUIC_FRAME_TYPE_NEW_CONN_ID:
2684 if (!a.allow_new_conn_id)
2685 continue;
2686 break;
2687 case OSSL_QUIC_FRAME_TYPE_RETIRE_CONN_ID:
2688 if (!a.allow_retire_conn_id)
2689 continue;
2690 break;
2691 case OSSL_QUIC_FRAME_TYPE_NEW_TOKEN:
2692 if (!a.allow_new_token)
2693 continue;
2694
2695 /*
2696 * NEW_TOKEN frames are handled via GCR, but some
2697 * Regenerate-strategy frames should come before them (namely
2698 * ACK, CONNECTION_CLOSE, PATH_CHALLENGE and PATH_RESPONSE). If
2699 * we find a NEW_TOKEN frame, do these now. If there are no
2700 * NEW_TOKEN frames in the GCR queue we will handle these below.
2701 */
2702 if (!done_pre_token)
faebafda 2703 if (txp_generate_pre_token(txp, pkt,
178c104d
HL
2704 chosen_for_conn_close,
2705 &can_be_non_inflight))
a73078b7
HL
2706 done_pre_token = 1;
2707
c5cb85b6
HL
2708 break;
2709 case OSSL_QUIC_FRAME_TYPE_PATH_RESPONSE:
2710 if (!a.allow_path_response)
2711 continue;
2712
2713 /*
2714 * RFC 9000 s. 8.2.2: An endpoint MUST expand datagrams that
2715 * contain a PATH_RESPONSE frame to at least the smallest
2716 * allowed maximum datagram size of 1200 bytes.
2717 */
2718 pkt->force_pad = 1;
a73078b7
HL
2719 break;
2720 default:
2721 if (!a.allow_cfq_other)
2722 continue;
2723 break;
2724 }
2725
2726 /*
2727 * If the frame is too big, don't try to schedule any more GCR frames in
2728 * this packet rather than sending subsequent ones out of order.
2729 */
faebafda 2730 if (encoded_len > tx_helper_get_space_left(h))
a73078b7
HL
2731 break;
2732
faebafda 2733 if (!tx_helper_append_iovec(h, encoded, encoded_len))
a73078b7
HL
2734 goto fatal_err;
2735
2736 ossl_quic_txpim_pkt_add_cfq_item(tpkt, cfq_item);
2737
2738 if (ossl_quic_frame_type_is_ack_eliciting(frame_type)) {
2739 have_ack_eliciting = 1;
faebafda 2740 tx_helper_unrestrict(h); /* no longer need PING */
a73078b7
HL
2741 }
2742 }
2743
2744 /*
2745 * If we didn't generate ACK, CONNECTION_CLOSE, PATH_CHALLENGE or
2746 * PATH_RESPONSE (as desired) before, do so now.
2747 */
2748 if (!done_pre_token)
faebafda 2749 if (txp_generate_pre_token(txp, pkt,
178c104d
HL
2750 chosen_for_conn_close,
2751 &can_be_non_inflight))
a73078b7
HL
2752 done_pre_token = 1;
2753
2754 /* CRYPTO Frames */
2755 if (a.allow_crypto)
faebafda 2756 if (!txp_generate_crypto_frames(txp, pkt, &have_ack_eliciting))
a73078b7
HL
2757 goto fatal_err;
2758
2759 /* Stream-specific frames */
cda88baf 2760 if (a.allow_stream_rel && txp->handshake_complete)
3bef14c5 2761 if (!txp_generate_stream_related(txp, pkt,
a73078b7 2762 &have_ack_eliciting,
faebafda 2763 &pkt->stream_head))
a73078b7
HL
2764 goto fatal_err;
2765
2766 /* PING */
faebafda 2767 tx_helper_unrestrict(h);
a73078b7 2768
1e2e683a
HL
2769 if ((a.require_ack_eliciting
2770 || (txp->force_ack_eliciting & (1UL << pn_space)) != 0)
2771 && !have_ack_eliciting && a.allow_ping) {
a73078b7
HL
2772 WPACKET *wpkt;
2773
faebafda 2774 wpkt = tx_helper_begin(h);
a73078b7
HL
2775 if (wpkt == NULL)
2776 goto fatal_err;
2777
2778 if (!ossl_quic_wire_encode_frame_ping(wpkt)
faebafda 2779 || !tx_helper_commit(h))
a73078b7
HL
2780 /*
2781 * We treat a request to be ACK-eliciting as a requirement, so this
2782 * is an error.
2783 */
2784 goto fatal_err;
2785
2786 have_ack_eliciting = 1;
2787 }
2788
3bef14c5 2789 /* PADDING is added by ossl_quic_tx_packetiser_generate(). */
a73078b7
HL
2790
2791 /*
faebafda
HL
2792 * ACKM Data
2793 * =========
a73078b7 2794 */
178c104d
HL
2795 if (have_ack_eliciting)
2796 can_be_non_inflight = 0;
2797
a73078b7 2798 /* ACKM Data */
faebafda 2799 tpkt->ackm_pkt.num_bytes = h->bytes_appended + pkt->geom.pkt_overhead;
a73078b7
HL
2800 tpkt->ackm_pkt.pkt_num = txp->next_pn[pn_space];
2801 /* largest_acked is set in txp_generate_pre_token */
2802 tpkt->ackm_pkt.pkt_space = pn_space;
178c104d 2803 tpkt->ackm_pkt.is_inflight = !can_be_non_inflight;
a73078b7
HL
2804 tpkt->ackm_pkt.is_ack_eliciting = have_ack_eliciting;
2805 tpkt->ackm_pkt.is_pto_probe = 0;
2806 tpkt->ackm_pkt.is_mtu_probe = 0;
b98c38d4 2807 tpkt->ackm_pkt.time = txp->args.now(txp->args.now_arg);
a73078b7 2808
faebafda
HL
2809 /* Done. */
2810 return rc;
2811
2812fatal_err:
2813 /*
2814 * Handler for fatal errors, i.e. errors causing us to abort the entire
2815 * packet rather than just one frame. Examples of such errors include
2816 * allocation errors.
2817 */
2818 if (tpkt != NULL) {
2819 ossl_quic_txpim_pkt_release(txp->args.txpim, tpkt);
2820 pkt->tpkt = NULL;
2821 }
2822 return TXP_ERR_INTERNAL;
2823}
2824
2825/*
2826 * Commits and queues a packet for transmission. There is no backing out after
2827 * this.
2828 *
2829 * This:
2830 *
2831 * - Sends the packet to the QTX for encryption and transmission;
2832 *
2833 * - Records the packet as having been transmitted in FIFM. ACKM is informed,
2834 * etc. and the TXPIM record is filed.
2835 *
2836 * - Informs various subsystems of frames that were sent and clears frame
2837 * wanted flags so that we do not generate the same frames again.
2838 *
2839 * Assumptions:
2840 *
2841 * - pkt is a txp_pkt for the correct EL;
2842 *
2843 * - pkt->tpkt is valid;
2844 *
2845 * - pkt->tpkt->ackm_pkt has been fully filled in;
2846 *
2847 * - Stream chunk records have been appended to pkt->tpkt for STREAM and
2848 * CRYPTO frames, but not for RESET_STREAM or STOP_SENDING frames;
2849 *
2850 * - The chosen stream list for the packet can be fully walked from
2851 * pkt->stream_head using stream->txp_next;
2852 *
2853 * - pkt->has_ack_eliciting is set correctly.
2854 *
2855 */
2856static int txp_pkt_commit(OSSL_QUIC_TX_PACKETISER *txp,
2857 struct txp_pkt *pkt,
6a2b70e2
HL
2858 uint32_t archetype,
2859 int *txpim_pkt_reffed)
faebafda
HL
2860{
2861 int rc = 1;
2862 uint32_t enc_level = pkt->h.enc_level;
2863 uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
2864 QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
2865 QUIC_STREAM *stream;
2866 OSSL_QTX_PKT txpkt;
2867 struct archetype_data a;
2868
6a2b70e2
HL
2869 *txpim_pkt_reffed = 0;
2870
faebafda
HL
2871 /* Cannot send a packet with an empty payload. */
2872 if (pkt->h.bytes_appended == 0)
2873 return 0;
2874
2875 if (!txp_get_archetype_data(enc_level, archetype, &a))
2876 return 0;
2877
a73078b7 2878 /* Packet Information for QTX */
faebafda
HL
2879 txpkt.hdr = &pkt->phdr;
2880 txpkt.iovec = txp->el[enc_level].iovec;
2881 txpkt.num_iovec = pkt->h.num_iovec;
2882 txpkt.local = NULL;
2883 txpkt.peer = BIO_ADDR_family(&txp->args.peer) == AF_UNSPEC
a73078b7 2884 ? NULL : &txp->args.peer;
faebafda
HL
2885 txpkt.pn = txp->next_pn[pn_space];
2886 txpkt.flags = OSSL_QTX_PKT_FLAG_COALESCE; /* always try to coalesce */
a73078b7
HL
2887
2888 /* Generate TXPIM chunks representing STOP_SENDING and RESET_STREAM frames. */
faebafda 2889 for (stream = pkt->stream_head; stream != NULL; stream = stream->txp_next)
a73078b7 2890 if (stream->txp_sent_stop_sending || stream->txp_sent_reset_stream) {
faebafda 2891 /* Log STOP_SENDING/RESET_STREAM chunk to TXPIM. */
a73078b7
HL
2892 QUIC_TXPIM_CHUNK chunk;
2893
2894 chunk.stream_id = stream->id;
2895 chunk.start = UINT64_MAX;
2896 chunk.end = 0;
2897 chunk.has_fin = 0;
2898 chunk.has_stop_sending = stream->txp_sent_stop_sending;
2899 chunk.has_reset_stream = stream->txp_sent_reset_stream;
2900 if (!ossl_quic_txpim_pkt_append_chunk(tpkt, &chunk))
2901 return 0; /* alloc error */
2902 }
2903
2904 /* Dispatch to FIFD. */
2905 if (!ossl_quic_fifd_pkt_commit(&txp->fifd, tpkt))
faebafda 2906 return 0;
a73078b7 2907
faebafda 2908 /*
ed75eb32
HL
2909 * Transmission and Post-Packet Generation Bookkeeping
2910 * ===================================================
faebafda 2911 *
ed75eb32
HL
2912 * No backing out anymore - at this point the ACKM has recorded the packet
2913 * as having been sent, so we need to increment our next PN counter, or
2914 * the ACKM will complain when we try to record a duplicate packet with
2915 * the same PN later. At this point actually sending the packet may still
2916 * fail. In this unlikely event it will simply be handled as though it
2917 * were a lost packet.
faebafda 2918 */
a73078b7 2919 ++txp->next_pn[pn_space];
6a2b70e2 2920 *txpim_pkt_reffed = 1;
a73078b7 2921
ed75eb32
HL
2922 /* Send the packet. */
2923 if (!ossl_qtx_write_pkt(txp->args.qtx, &txpkt))
2924 return 0;
2925
a73078b7
HL
2926 /*
2927 * Record FC and stream abort frames as sent; deactivate streams which no
2928 * longer have anything to do.
2929 */
faebafda 2930 for (stream = pkt->stream_head; stream != NULL; stream = stream->txp_next) {
a73078b7
HL
2931 if (stream->txp_sent_fc) {
2932 stream->want_max_stream_data = 0;
2933 ossl_quic_rxfc_has_cwm_changed(&stream->rxfc, 1);
2934 }
2935
2936 if (stream->txp_sent_stop_sending)
2937 stream->want_stop_sending = 0;
2938
2939 if (stream->txp_sent_reset_stream)
2940 stream->want_reset_stream = 0;
2941
2942 if (stream->txp_txfc_new_credit_consumed > 0) {
2943 if (!ossl_assert(ossl_quic_txfc_consume_credit(&stream->txfc,
2944 stream->txp_txfc_new_credit_consumed)))
2945 /*
2946 * Should not be possible, but we should continue with our
2947 * bookkeeping as we have already committed the packet to the
2948 * FIFD. Just change the value we return.
2949 */
faebafda 2950 rc = 0;
a73078b7
HL
2951
2952 stream->txp_txfc_new_credit_consumed = 0;
2953 }
2954
2955 /*
2956 * If we no longer need to generate any flow control (MAX_STREAM_DATA),
2957 * STOP_SENDING or RESET_STREAM frames, nor any STREAM frames (because
2958 * the stream is drained of data or TXFC-blocked), we can mark the
2959 * stream as inactive.
2960 */
2961 ossl_quic_stream_map_update_state(txp->args.qsm, stream);
2962
26652896
HL
2963 if (ossl_quic_stream_has_send_buffer(stream)
2964 && !ossl_quic_sstream_has_pending(stream->sstream)
2965 && ossl_quic_sstream_get_final_size(stream->sstream, NULL))
c068f4d1
HL
2966 /*
2967 * Transition to DATA_SENT if stream has a final size and we have
2968 * sent all data.
2969 */
26652896 2970 ossl_quic_stream_map_notify_all_data_sent(txp->args.qsm, stream);
a73078b7
HL
2971 }
2972
2973 /* We have now sent the packet, so update state accordingly. */
faebafda 2974 if (tpkt->ackm_pkt.is_ack_eliciting)
a73078b7
HL
2975 txp->force_ack_eliciting &= ~(1UL << pn_space);
2976
2977 if (tpkt->had_handshake_done_frame)
2978 txp->want_handshake_done = 0;
2979
2980 if (tpkt->had_max_data_frame) {
2981 txp->want_max_data = 0;
2982 ossl_quic_rxfc_has_cwm_changed(txp->args.conn_rxfc, 1);
2983 }
2984
a6b6ea17 2985 if (tpkt->had_max_streams_bidi_frame) {
a73078b7 2986 txp->want_max_streams_bidi = 0;
a6b6ea17
HL
2987 ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_bidi_rxfc, 1);
2988 }
a73078b7 2989
a6b6ea17 2990 if (tpkt->had_max_streams_uni_frame) {
a73078b7 2991 txp->want_max_streams_uni = 0;
a6b6ea17
HL
2992 ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_uni_rxfc, 1);
2993 }
a73078b7
HL
2994
2995 if (tpkt->had_ack_frame)
2996 txp->want_ack &= ~(1UL << pn_space);
2997
0b31072e
MC
2998 if (tpkt->had_conn_close)
2999 txp->want_conn_close = 0;
3000
fee8f48e
HL
3001 /*
3002 * Decrement probe request counts if we have sent a packet that meets
3003 * the requirement of a probe, namely being ACK-eliciting.
3004 */
faebafda
HL
3005 if (tpkt->ackm_pkt.is_ack_eliciting) {
3006 OSSL_ACKM_PROBE_INFO *probe_info
3007 = ossl_ackm_get0_probe_request(txp->args.ackm);
3008
fee8f48e
HL
3009 if (enc_level == QUIC_ENC_LEVEL_INITIAL
3010 && probe_info->anti_deadlock_initial > 0)
3011 --probe_info->anti_deadlock_initial;
3012
3013 if (enc_level == QUIC_ENC_LEVEL_HANDSHAKE
3014 && probe_info->anti_deadlock_handshake > 0)
3015 --probe_info->anti_deadlock_handshake;
3016
3017 if (a.allow_force_ack_eliciting /* (i.e., not for 0-RTT) */
3018 && probe_info->pto[pn_space] > 0)
3019 --probe_info->pto[pn_space];
3020 }
3021
a73078b7 3022 return rc;
a73078b7
HL
3023}
3024
3025/* Ensure the iovec array is at least num elements long. */
faebafda 3026static int txp_el_ensure_iovec(struct txp_el *el, size_t num)
a73078b7
HL
3027{
3028 OSSL_QTX_IOVEC *iovec;
3029
faebafda 3030 if (el->alloc_iovec >= num)
a73078b7
HL
3031 return 1;
3032
faebafda 3033 num = el->alloc_iovec != 0 ? el->alloc_iovec * 2 : 8;
a73078b7 3034
faebafda 3035 iovec = OPENSSL_realloc(el->iovec, sizeof(OSSL_QTX_IOVEC) * num);
a73078b7
HL
3036 if (iovec == NULL)
3037 return 0;
3038
faebafda
HL
3039 el->iovec = iovec;
3040 el->alloc_iovec = num;
a73078b7
HL
3041 return 1;
3042}
3043
3044int ossl_quic_tx_packetiser_schedule_conn_close(OSSL_QUIC_TX_PACKETISER *txp,
3045 const OSSL_QUIC_FRAME_CONN_CLOSE *f)
3046{
3047 char *reason = NULL;
3048 size_t reason_len = f->reason_len;
3049 size_t max_reason_len = txp_get_mdpl(txp) / 2;
3050
3051 if (txp->want_conn_close)
3052 return 0;
3053
3054 /*
3055 * Arbitrarily limit the length of the reason length string to half of the
3056 * MDPL.
3057 */
3058 if (reason_len > max_reason_len)
3059 reason_len = max_reason_len;
3060
3061 if (reason_len > 0) {
3062 reason = OPENSSL_memdup(f->reason, reason_len);
3063 if (reason == NULL)
3064 return 0;
3065 }
3066
3067 txp->conn_close_frame = *f;
3068 txp->conn_close_frame.reason = reason;
3069 txp->conn_close_frame.reason_len = reason_len;
3070 txp->want_conn_close = 1;
3071 return 1;
3072}
5cf99b40
MC
3073
3074void ossl_quic_tx_packetiser_set_msg_callback(OSSL_QUIC_TX_PACKETISER *txp,
3075 ossl_msg_cb msg_callback,
c2786c8e 3076 SSL *msg_callback_ssl)
5cf99b40
MC
3077{
3078 txp->msg_callback = msg_callback;
c2786c8e 3079 txp->msg_callback_ssl = msg_callback_ssl;
5cf99b40
MC
3080}
3081
3082void ossl_quic_tx_packetiser_set_msg_callback_arg(OSSL_QUIC_TX_PACKETISER *txp,
3083 void *msg_callback_arg)
3084{
3085 txp->msg_callback_arg = msg_callback_arg;
3086}
007f9e99
HL
3087
3088QUIC_PN ossl_quic_tx_packetiser_get_next_pn(OSSL_QUIC_TX_PACKETISER *txp,
3089 uint32_t pn_space)
3090{
3091 if (pn_space >= QUIC_PN_SPACE_NUM)
3092 return UINT64_MAX;
3093
3094 return txp->next_pn[pn_space];
3095}
c206f2aa
HL
3096
3097OSSL_TIME ossl_quic_tx_packetiser_get_deadline(OSSL_QUIC_TX_PACKETISER *txp)
3098{
3099 /*
3100 * TXP-specific deadline computations which rely on TXP innards. This is in
3101 * turn relied on by the QUIC_CHANNEL code to determine the channel event
3102 * handling deadline.
3103 */
3104 OSSL_TIME deadline = ossl_time_infinite();
3105 uint32_t enc_level, pn_space;
3106
3107 /*
3108 * ACK generation is not CC-gated - packets containing only ACKs are allowed
3109 * to bypass CC. We want to generate ACK frames even if we are currently
3110 * restricted by CC so the peer knows we have received data. The generate
3111 * call will take care of selecting the correct packet archetype.
3112 */
3113 for (enc_level = QUIC_ENC_LEVEL_INITIAL;
3114 enc_level < QUIC_ENC_LEVEL_NUM;
3115 ++enc_level)
3116 if (ossl_qtx_is_enc_level_provisioned(txp->args.qtx, enc_level)) {
3117 pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
3118 deadline = ossl_time_min(deadline,
3119 ossl_ackm_get_ack_deadline(txp->args.ackm, pn_space));
3120 }
3121
3122 /* When will CC let us send more? */
63fac76c
HL
3123 if (txp->args.cc_method->get_tx_allowance(txp->args.cc_data) == 0)
3124 deadline = ossl_time_min(deadline,
3125 txp->args.cc_method->get_wakeup_deadline(txp->args.cc_data));
c206f2aa
HL
3126
3127 return deadline;
3128}