xref: /freebsd/crypto/openssl/ssl/quic/quic_txp.c (revision f25b8c9fb4f58cf61adb47d7570abe7caa6d385d)
1 /*
2  * Copyright 2022-2025 The OpenSSL Project Authors. All Rights Reserved.
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"
13 #include "internal/quic_error.h"
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 
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 
51 struct 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     size_t unvalidated_credit; /* Limit of data we can send until validated */
71 
72     /* Internal state - frame (re)generation flags. */
73     unsigned int want_handshake_done : 1;
74     unsigned int want_max_data : 1;
75     unsigned int want_max_streams_bidi : 1;
76     unsigned int want_max_streams_uni : 1;
77 
78     /* Internal state - frame (re)generation flags - per PN space. */
79     unsigned int want_ack : QUIC_PN_SPACE_NUM;
80     unsigned int force_ack_eliciting : QUIC_PN_SPACE_NUM;
81 
82     /*
83      * Internal state - connection close terminal state.
84      * Once this is set, it is not unset unlike other want_ flags - we keep
85      * sending it in every packet.
86      */
87     unsigned int want_conn_close : 1;
88 
89     /* Has the handshake been completed? */
90     unsigned int handshake_complete : 1;
91 
92     OSSL_QUIC_FRAME_CONN_CLOSE conn_close_frame;
93 
94     /*
95      * Counts of the number of bytes received and sent while in the closing
96      * state.
97      */
98     uint64_t closing_bytes_recv;
99     uint64_t closing_bytes_xmit;
100 
101     /* Internal state - packet assembly. */
102     struct txp_el {
103         unsigned char *scratch; /* scratch buffer for packet assembly */
104         size_t scratch_len; /* number of bytes allocated for scratch */
105         OSSL_QTX_IOVEC *iovec; /* scratch iovec array for use with QTX */
106         size_t alloc_iovec; /* size of iovec array */
107     } el[QUIC_ENC_LEVEL_NUM];
108 
109     /* Message callback related arguments */
110     ossl_msg_cb msg_callback;
111     void *msg_callback_arg;
112     SSL *msg_callback_ssl;
113 
114     /* Callbacks. */
115     void (*ack_tx_cb)(const OSSL_QUIC_FRAME_ACK *ack,
116         uint32_t pn_space,
117         void *arg);
118     void *ack_tx_cb_arg;
119 };
120 
121 /*
122  * The TX helper records state used while generating frames into packets. It
123  * enables serialization into the packet to be done "transactionally" where
124  * serialization of a frame can be rolled back if it fails midway (e.g. if it
125  * does not fit).
126  */
127 struct tx_helper {
128     OSSL_QUIC_TX_PACKETISER *txp;
129     /*
130      * The Maximum Packet Payload Length in bytes. This is the amount of
131      * space we have to generate frames into.
132      */
133     size_t max_ppl;
134     /*
135      * Number of bytes we have generated so far.
136      */
137     size_t bytes_appended;
138     /*
139      * Number of scratch bytes in txp->scratch we have used so far. Some iovecs
140      * will reference this scratch buffer. When we need to use more of it (e.g.
141      * when we need to put frame headers somewhere), we append to the scratch
142      * buffer, resizing if necessary, and increase this accordingly.
143      */
144     size_t scratch_bytes;
145     /*
146      * Bytes reserved in the MaxPPL budget. We keep this number of bytes spare
147      * until reserve_allowed is set to 1. Currently this is always at most 1, as
148      * a PING frame takes up one byte and this mechanism is only used to ensure
149      * we can encode a PING frame if we have been asked to ensure a packet is
150      * ACK-eliciting and we are unusure if we are going to add any other
151      * ACK-eliciting frames before we reach our MaxPPL budget.
152      */
153     size_t reserve;
154     /*
155      * Number of iovecs we have currently appended. This is the number of
156      * entries valid in txp->iovec.
157      */
158     size_t num_iovec;
159     /* The EL this TX helper is being used for. */
160     uint32_t enc_level;
161     /*
162      * Whether we are allowed to make use of the reserve bytes in our MaxPPL
163      * budget. This is used to ensure we have room to append a PING frame later
164      * if we need to. Once we know we will not need to append a PING frame, this
165      * is set to 1.
166      */
167     unsigned int reserve_allowed : 1;
168     /*
169      * Set to 1 if we have appended a STREAM frame with an implicit length. If
170      * this happens we should never append another frame after that frame as it
171      * cannot be validly encoded. This is just a safety check.
172      */
173     unsigned int done_implicit : 1;
174     struct {
175         /*
176          * The fields in this structure are valid if active is set, which means
177          * that a serialization transaction is currently in progress.
178          */
179         unsigned char *data;
180         WPACKET wpkt;
181         unsigned int active : 1;
182     } txn;
183 };
184 
185 static void tx_helper_rollback(struct tx_helper *h);
186 static int txp_el_ensure_iovec(struct txp_el *el, size_t num);
187 
188 /* Initialises the TX helper. */
tx_helper_init(struct tx_helper * h,OSSL_QUIC_TX_PACKETISER * txp,uint32_t enc_level,size_t max_ppl,size_t reserve)189 static int tx_helper_init(struct tx_helper *h, OSSL_QUIC_TX_PACKETISER *txp,
190     uint32_t enc_level, size_t max_ppl, size_t reserve)
191 {
192     if (reserve > max_ppl)
193         return 0;
194 
195     h->txp = txp;
196     h->enc_level = enc_level;
197     h->max_ppl = max_ppl;
198     h->reserve = reserve;
199     h->num_iovec = 0;
200     h->bytes_appended = 0;
201     h->scratch_bytes = 0;
202     h->reserve_allowed = 0;
203     h->done_implicit = 0;
204     h->txn.data = NULL;
205     h->txn.active = 0;
206 
207     if (max_ppl > h->txp->el[enc_level].scratch_len) {
208         unsigned char *scratch;
209 
210         scratch = OPENSSL_realloc(h->txp->el[enc_level].scratch, max_ppl);
211         if (scratch == NULL)
212             return 0;
213 
214         h->txp->el[enc_level].scratch = scratch;
215         h->txp->el[enc_level].scratch_len = max_ppl;
216     }
217 
218     return 1;
219 }
220 
tx_helper_cleanup(struct tx_helper * h)221 static void tx_helper_cleanup(struct tx_helper *h)
222 {
223     if (h->txn.active)
224         tx_helper_rollback(h);
225 
226     h->txp = NULL;
227 }
228 
tx_helper_unrestrict(struct tx_helper * h)229 static void tx_helper_unrestrict(struct tx_helper *h)
230 {
231     h->reserve_allowed = 1;
232 }
233 
234 /*
235  * Append an extent of memory to the iovec list. The memory must remain
236  * allocated until we finish generating the packet and call the QTX.
237  *
238  * In general, the buffers passed to this function will be from one of two
239  * ranges:
240  *
241  *   - Application data contained in stream buffers managed elsewhere
242  *     in the QUIC stack; or
243  *
244  *   - Control frame data appended into txp->scratch using tx_helper_begin and
245  *     tx_helper_commit.
246  *
247  */
tx_helper_append_iovec(struct tx_helper * h,const unsigned char * buf,size_t buf_len)248 static int tx_helper_append_iovec(struct tx_helper *h,
249     const unsigned char *buf,
250     size_t buf_len)
251 {
252     struct txp_el *el = &h->txp->el[h->enc_level];
253 
254     if (buf_len == 0)
255         return 1;
256 
257     if (!ossl_assert(!h->done_implicit))
258         return 0;
259 
260     if (!txp_el_ensure_iovec(el, h->num_iovec + 1))
261         return 0;
262 
263     el->iovec[h->num_iovec].buf = buf;
264     el->iovec[h->num_iovec].buf_len = buf_len;
265 
266     ++h->num_iovec;
267     h->bytes_appended += buf_len;
268     return 1;
269 }
270 
271 /*
272  * How many more bytes of space do we have left in our plaintext packet payload?
273  */
tx_helper_get_space_left(struct tx_helper * h)274 static size_t tx_helper_get_space_left(struct tx_helper *h)
275 {
276     return h->max_ppl
277         - (h->reserve_allowed ? 0 : h->reserve) - h->bytes_appended;
278 }
279 
280 /*
281  * Begin a control frame serialization transaction. This allows the
282  * serialization of the control frame to be backed out if it turns out it won't
283  * fit. Write the control frame to the returned WPACKET. Ensure you always
284  * call tx_helper_rollback or tx_helper_commit (or tx_helper_cleanup). Returns
285  * NULL on failure.
286  */
tx_helper_begin(struct tx_helper * h)287 static WPACKET *tx_helper_begin(struct tx_helper *h)
288 {
289     size_t space_left, len;
290     unsigned char *data;
291     struct txp_el *el = &h->txp->el[h->enc_level];
292 
293     if (!ossl_assert(!h->txn.active))
294         return NULL;
295 
296     if (!ossl_assert(!h->done_implicit))
297         return NULL;
298 
299     data = (unsigned char *)el->scratch + h->scratch_bytes;
300     len = el->scratch_len - h->scratch_bytes;
301 
302     space_left = tx_helper_get_space_left(h);
303     if (!ossl_assert(space_left <= len))
304         return NULL;
305 
306     if (!WPACKET_init_static_len(&h->txn.wpkt, data, len, 0))
307         return NULL;
308 
309     if (!WPACKET_set_max_size(&h->txn.wpkt, space_left)) {
310         WPACKET_cleanup(&h->txn.wpkt);
311         return NULL;
312     }
313 
314     h->txn.data = data;
315     h->txn.active = 1;
316     return &h->txn.wpkt;
317 }
318 
tx_helper_end(struct tx_helper * h,int success)319 static void tx_helper_end(struct tx_helper *h, int success)
320 {
321     if (success)
322         WPACKET_finish(&h->txn.wpkt);
323     else
324         WPACKET_cleanup(&h->txn.wpkt);
325 
326     h->txn.active = 0;
327     h->txn.data = NULL;
328 }
329 
330 /* Abort a control frame serialization transaction. */
tx_helper_rollback(struct tx_helper * h)331 static void tx_helper_rollback(struct tx_helper *h)
332 {
333     if (!h->txn.active)
334         return;
335 
336     tx_helper_end(h, 0);
337 }
338 
339 /* Commit a control frame. */
tx_helper_commit(struct tx_helper * h)340 static int tx_helper_commit(struct tx_helper *h)
341 {
342     size_t l = 0;
343 
344     if (!h->txn.active)
345         return 0;
346 
347     if (!WPACKET_get_total_written(&h->txn.wpkt, &l)) {
348         tx_helper_end(h, 0);
349         return 0;
350     }
351 
352     if (!tx_helper_append_iovec(h, h->txn.data, l)) {
353         tx_helper_end(h, 0);
354         return 0;
355     }
356 
357     if (h->txp->msg_callback != NULL && l > 0) {
358         uint64_t ftype;
359         int ctype = SSL3_RT_QUIC_FRAME_FULL;
360         PACKET pkt;
361 
362         if (!PACKET_buf_init(&pkt, h->txn.data, l)
363             || !ossl_quic_wire_peek_frame_header(&pkt, &ftype, NULL)) {
364             tx_helper_end(h, 0);
365             return 0;
366         }
367 
368         if (ftype == OSSL_QUIC_FRAME_TYPE_PADDING)
369             ctype = SSL3_RT_QUIC_FRAME_PADDING;
370         else if (OSSL_QUIC_FRAME_TYPE_IS_STREAM(ftype)
371             || ftype == OSSL_QUIC_FRAME_TYPE_CRYPTO)
372             ctype = SSL3_RT_QUIC_FRAME_HEADER;
373 
374         h->txp->msg_callback(1, OSSL_QUIC1_VERSION, ctype, h->txn.data, l,
375             h->txp->msg_callback_ssl,
376             h->txp->msg_callback_arg);
377     }
378 
379     h->scratch_bytes += l;
380     tx_helper_end(h, 1);
381     return 1;
382 }
383 
384 struct archetype_data {
385     unsigned int allow_ack : 1;
386     unsigned int allow_ping : 1;
387     unsigned int allow_crypto : 1;
388     unsigned int allow_handshake_done : 1;
389     unsigned int allow_path_challenge : 1;
390     unsigned int allow_path_response : 1;
391     unsigned int allow_new_conn_id : 1;
392     unsigned int allow_retire_conn_id : 1;
393     unsigned int allow_stream_rel : 1;
394     unsigned int allow_conn_fc : 1;
395     unsigned int allow_conn_close : 1;
396     unsigned int allow_cfq_other : 1;
397     unsigned int allow_new_token : 1;
398     unsigned int allow_force_ack_eliciting : 1;
399     unsigned int allow_padding : 1;
400     unsigned int require_ack_eliciting : 1;
401     unsigned int bypass_cc : 1;
402 };
403 
404 struct txp_pkt_geom {
405     size_t cmpl, cmppl, hwm, pkt_overhead;
406     uint32_t archetype;
407     struct archetype_data adata;
408 };
409 
410 struct txp_pkt {
411     struct tx_helper h;
412     int h_valid;
413     QUIC_TXPIM_PKT *tpkt;
414     QUIC_STREAM *stream_head;
415     QUIC_PKT_HDR phdr;
416     struct txp_pkt_geom geom;
417     int force_pad;
418 };
419 
420 static QUIC_SSTREAM *get_sstream_by_id(uint64_t stream_id, uint32_t pn_space,
421     void *arg);
422 static void on_regen_notify(uint64_t frame_type, uint64_t stream_id,
423     QUIC_TXPIM_PKT *pkt, void *arg);
424 static void on_confirm_notify(uint64_t frame_type, uint64_t stream_id,
425     QUIC_TXPIM_PKT *pkt, void *arg);
426 static void on_sstream_updated(uint64_t stream_id, void *arg);
427 static int sstream_is_pending(QUIC_SSTREAM *sstream);
428 static int txp_should_try_staging(OSSL_QUIC_TX_PACKETISER *txp,
429     uint32_t enc_level,
430     uint32_t archetype,
431     uint64_t cc_limit,
432     uint32_t *conn_close_enc_level);
433 static size_t txp_determine_pn_len(OSSL_QUIC_TX_PACKETISER *txp);
434 static int txp_determine_ppl_from_pl(OSSL_QUIC_TX_PACKETISER *txp,
435     size_t pl,
436     uint32_t enc_level,
437     size_t hdr_len,
438     size_t *r);
439 static size_t txp_get_mdpl(OSSL_QUIC_TX_PACKETISER *txp);
440 static int txp_generate_for_el(OSSL_QUIC_TX_PACKETISER *txp,
441     struct txp_pkt *pkt,
442     int chosen_for_conn_close);
443 static int txp_pkt_init(struct txp_pkt *pkt, OSSL_QUIC_TX_PACKETISER *txp,
444     uint32_t enc_level, uint32_t archetype,
445     size_t running_total);
446 static void txp_pkt_cleanup(struct txp_pkt *pkt, OSSL_QUIC_TX_PACKETISER *txp);
447 static int txp_pkt_postgen_update_pkt_overhead(struct txp_pkt *pkt,
448     OSSL_QUIC_TX_PACKETISER *txp);
449 static int txp_pkt_append_padding(struct txp_pkt *pkt,
450     OSSL_QUIC_TX_PACKETISER *txp, size_t num_bytes);
451 static int txp_pkt_commit(OSSL_QUIC_TX_PACKETISER *txp, struct txp_pkt *pkt,
452     uint32_t archetype, int *txpim_pkt_reffed);
453 static uint32_t txp_determine_archetype(OSSL_QUIC_TX_PACKETISER *txp,
454     uint64_t cc_limit);
455 
456 /**
457  * Sets the validated state of a QUIC TX packetiser.
458  *
459  * This function marks the provided QUIC TX packetiser as having its credit
460  * fully validated by setting its `unvalidated_credit` field to `SIZE_MAX`.
461  *
462  * @param txp A pointer to the OSSL_QUIC_TX_PACKETISER structure to update.
463  */
ossl_quic_tx_packetiser_set_validated(OSSL_QUIC_TX_PACKETISER * txp)464 void ossl_quic_tx_packetiser_set_validated(OSSL_QUIC_TX_PACKETISER *txp)
465 {
466     txp->unvalidated_credit = SIZE_MAX;
467     return;
468 }
469 
470 /**
471  * Adds unvalidated credit to a QUIC TX packetiser.
472  *
473  * This function increases the unvalidated credit of the provided QUIC TX
474  * packetiser. If the current unvalidated credit is not `SIZE_MAX`, the
475  * function adds three times the specified `credit` value, ensuring it does
476  * not exceed the maximum allowable value (`SIZE_MAX - 1`). If the addition
477  * would cause an overflow, the unvalidated credit is capped at
478  * `SIZE_MAX - 1`. If the current unvalidated credit is already `SIZE_MAX`,
479  * the function does nothing.
480  *
481  * @param txp    A pointer to the OSSL_QUIC_TX_PACKETISER structure to update.
482  * @param credit The amount of credit to add, multiplied by 3.
483  */
ossl_quic_tx_packetiser_add_unvalidated_credit(OSSL_QUIC_TX_PACKETISER * txp,size_t credit)484 void ossl_quic_tx_packetiser_add_unvalidated_credit(OSSL_QUIC_TX_PACKETISER *txp,
485     size_t credit)
486 {
487     if (txp->unvalidated_credit != SIZE_MAX) {
488         if ((SIZE_MAX - txp->unvalidated_credit) > (credit * 3))
489             txp->unvalidated_credit += credit * 3;
490         else
491             txp->unvalidated_credit = SIZE_MAX - 1;
492     }
493 
494     return;
495 }
496 
497 /**
498  * Consumes unvalidated credit from a QUIC TX packetiser.
499  *
500  * This function decreases the unvalidated credit of the specified
501  * QUIC TX packetiser by the given `credit` value. If the unvalidated credit
502  * is set to `SIZE_MAX`, the function does nothing, as `SIZE_MAX` represents
503  * an unlimited credit state.
504  *
505  * @param txp    A pointer to the OSSL_QUIC_TX_PACKETISER structure to update.
506  * @param credit The amount of credit to consume.
507  */
ossl_quic_tx_packetiser_consume_unvalidated_credit(OSSL_QUIC_TX_PACKETISER * txp,size_t credit)508 void ossl_quic_tx_packetiser_consume_unvalidated_credit(OSSL_QUIC_TX_PACKETISER *txp,
509     size_t credit)
510 {
511     if (txp->unvalidated_credit != SIZE_MAX) {
512         if (txp->unvalidated_credit < credit)
513             txp->unvalidated_credit = 0;
514         else
515             txp->unvalidated_credit -= credit;
516     }
517 }
518 
519 /**
520  * Checks if the QUIC TX packetiser has sufficient unvalidated credit.
521  *
522  * This function determines whether the unvalidated credit of the specified
523  * QUIC TX packetiser exceeds the required credit value (`req_credit`).
524  * If the unvalidated credit is greater than `req_credit`, the function
525  * returns 1 (true); otherwise, it returns 0 (false).
526  *
527  * @param txp        A pointer to the OSSL_QUIC_TX_PACKETISER structure to check.
528  * @param req_credit The required credit value to compare against.
529  *
530  * @return 1 if the unvalidated credit exceeds `req_credit`, 0 otherwise.
531  */
ossl_quic_tx_packetiser_check_unvalidated_credit(OSSL_QUIC_TX_PACKETISER * txp,size_t req_credit)532 int ossl_quic_tx_packetiser_check_unvalidated_credit(OSSL_QUIC_TX_PACKETISER *txp,
533     size_t req_credit)
534 {
535     return (txp->unvalidated_credit > req_credit);
536 }
537 
ossl_quic_tx_packetiser_new(const OSSL_QUIC_TX_PACKETISER_ARGS * args)538 OSSL_QUIC_TX_PACKETISER *ossl_quic_tx_packetiser_new(const OSSL_QUIC_TX_PACKETISER_ARGS *args)
539 {
540     OSSL_QUIC_TX_PACKETISER *txp;
541 
542     if (args == NULL
543         || args->qtx == NULL
544         || args->txpim == NULL
545         || args->cfq == NULL
546         || args->ackm == NULL
547         || args->qsm == NULL
548         || args->conn_txfc == NULL
549         || args->conn_rxfc == NULL
550         || args->max_streams_bidi_rxfc == NULL
551         || args->max_streams_uni_rxfc == NULL
552         || args->protocol_version == 0) {
553         ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_NULL_PARAMETER);
554         return NULL;
555     }
556 
557     txp = OPENSSL_zalloc(sizeof(*txp));
558     if (txp == NULL)
559         return NULL;
560 
561     txp->args = *args;
562     txp->last_tx_time = ossl_time_zero();
563 
564     if (!ossl_quic_fifd_init(&txp->fifd,
565             txp->args.cfq, txp->args.ackm, txp->args.txpim,
566             get_sstream_by_id, txp,
567             on_regen_notify, txp,
568             on_confirm_notify, txp,
569             on_sstream_updated, txp,
570             args->get_qlog_cb,
571             args->get_qlog_cb_arg)) {
572         OPENSSL_free(txp);
573         return NULL;
574     }
575 
576     return txp;
577 }
578 
ossl_quic_tx_packetiser_free(OSSL_QUIC_TX_PACKETISER * txp)579 void ossl_quic_tx_packetiser_free(OSSL_QUIC_TX_PACKETISER *txp)
580 {
581     uint32_t enc_level;
582 
583     if (txp == NULL)
584         return;
585 
586     ossl_quic_tx_packetiser_set_initial_token(txp, NULL, 0, NULL, NULL);
587     ossl_quic_fifd_cleanup(&txp->fifd);
588     OPENSSL_free(txp->conn_close_frame.reason);
589 
590     for (enc_level = QUIC_ENC_LEVEL_INITIAL;
591         enc_level < QUIC_ENC_LEVEL_NUM;
592         ++enc_level) {
593         OPENSSL_free(txp->el[enc_level].iovec);
594         OPENSSL_free(txp->el[enc_level].scratch);
595     }
596 
597     OPENSSL_free(txp);
598 }
599 
600 /*
601  * Determine if an Initial packet token length is reasonable based on the
602  * current MDPL, returning 1 if it is OK.
603  *
604  * The real PMTU to the peer could differ from our (pessimistic) understanding
605  * of the PMTU, therefore it is possible we could receive an Initial token from
606  * a server in a Retry packet which is bigger than the MDPL. In this case it is
607  * impossible for us ever to make forward progress and we need to error out
608  * and fail the connection attempt.
609  *
610  * The specific boundary condition is complex: for example, after the size of
611  * the Initial token, there are the Initial packet header overheads and then
612  * encryption/AEAD tag overheads. After that, the minimum room for frame data in
613  * order to guarantee forward progress must be guaranteed. For example, a crypto
614  * stream needs to always be able to serialize at least one byte in a CRYPTO
615  * frame in order to make forward progress. Because the offset field of a CRYPTO
616  * frame uses a variable-length integer, the number of bytes needed to ensure
617  * this also varies.
618  *
619  * Rather than trying to get this boundary condition check actually right,
620  * require a reasonable amount of slack to avoid pathological behaviours. (After
621  * all, transmitting a CRYPTO stream one byte at a time is probably not
622  * desirable anyway.)
623  *
624  * We choose 160 bytes as the required margin, which is double the rough
625  * estimation of the minimum we would require to guarantee forward progress
626  * under worst case packet overheads.
627  */
628 #define TXP_REQUIRED_TOKEN_MARGIN 160
629 
txp_check_token_len(size_t token_len,size_t mdpl)630 static int txp_check_token_len(size_t token_len, size_t mdpl)
631 {
632     if (token_len == 0)
633         return 1;
634 
635     if (token_len >= mdpl)
636         return 0;
637 
638     if (TXP_REQUIRED_TOKEN_MARGIN >= mdpl)
639         /* (should not be possible because MDPL must be at least 1200) */
640         return 0;
641 
642     if (token_len > mdpl - TXP_REQUIRED_TOKEN_MARGIN)
643         return 0;
644 
645     return 1;
646 }
647 
ossl_quic_tx_packetiser_set_initial_token(OSSL_QUIC_TX_PACKETISER * txp,const unsigned char * token,size_t token_len,ossl_quic_initial_token_free_fn * free_cb,void * free_cb_arg)648 int ossl_quic_tx_packetiser_set_initial_token(OSSL_QUIC_TX_PACKETISER *txp,
649     const unsigned char *token,
650     size_t token_len,
651     ossl_quic_initial_token_free_fn *free_cb,
652     void *free_cb_arg)
653 {
654     if (!txp_check_token_len(token_len, txp_get_mdpl(txp)))
655         return 0;
656 
657     if (txp->initial_token != NULL && txp->initial_token_free_cb != NULL)
658         txp->initial_token_free_cb(txp->initial_token, txp->initial_token_len,
659             txp->initial_token_free_cb_arg);
660 
661     txp->initial_token = token;
662     txp->initial_token_len = token_len;
663     txp->initial_token_free_cb = free_cb;
664     txp->initial_token_free_cb_arg = free_cb_arg;
665     return 1;
666 }
667 
ossl_quic_tx_packetiser_set_protocol_version(OSSL_QUIC_TX_PACKETISER * txp,uint32_t protocol_version)668 int ossl_quic_tx_packetiser_set_protocol_version(OSSL_QUIC_TX_PACKETISER *txp,
669     uint32_t protocol_version)
670 {
671     txp->args.protocol_version = protocol_version;
672     return 1;
673 }
674 
ossl_quic_tx_packetiser_set_cur_dcid(OSSL_QUIC_TX_PACKETISER * txp,const QUIC_CONN_ID * dcid)675 int ossl_quic_tx_packetiser_set_cur_dcid(OSSL_QUIC_TX_PACKETISER *txp,
676     const QUIC_CONN_ID *dcid)
677 {
678     if (dcid == NULL) {
679         ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_NULL_PARAMETER);
680         return 0;
681     }
682 
683     txp->args.cur_dcid = *dcid;
684     return 1;
685 }
686 
ossl_quic_tx_packetiser_set_cur_scid(OSSL_QUIC_TX_PACKETISER * txp,const QUIC_CONN_ID * scid)687 int ossl_quic_tx_packetiser_set_cur_scid(OSSL_QUIC_TX_PACKETISER *txp,
688     const QUIC_CONN_ID *scid)
689 {
690     if (scid == NULL) {
691         ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_NULL_PARAMETER);
692         return 0;
693     }
694 
695     txp->args.cur_scid = *scid;
696     return 1;
697 }
698 
699 /* Change the destination L4 address the TXP uses to send datagrams. */
ossl_quic_tx_packetiser_set_peer(OSSL_QUIC_TX_PACKETISER * txp,const BIO_ADDR * peer)700 int ossl_quic_tx_packetiser_set_peer(OSSL_QUIC_TX_PACKETISER *txp,
701     const BIO_ADDR *peer)
702 {
703     if (peer == NULL) {
704         BIO_ADDR_clear(&txp->args.peer);
705         return 1;
706     }
707 
708     return BIO_ADDR_copy(&txp->args.peer, peer);
709 }
710 
ossl_quic_tx_packetiser_set_ack_tx_cb(OSSL_QUIC_TX_PACKETISER * txp,void (* cb)(const OSSL_QUIC_FRAME_ACK * ack,uint32_t pn_space,void * arg),void * cb_arg)711 void ossl_quic_tx_packetiser_set_ack_tx_cb(OSSL_QUIC_TX_PACKETISER *txp,
712     void (*cb)(const OSSL_QUIC_FRAME_ACK *ack,
713         uint32_t pn_space,
714         void *arg),
715     void *cb_arg)
716 {
717     txp->ack_tx_cb = cb;
718     txp->ack_tx_cb_arg = cb_arg;
719 }
720 
ossl_quic_tx_packetiser_set_qlog_cb(OSSL_QUIC_TX_PACKETISER * txp,QLOG * (* get_qlog_cb)(void * arg),void * get_qlog_cb_arg)721 void ossl_quic_tx_packetiser_set_qlog_cb(OSSL_QUIC_TX_PACKETISER *txp,
722     QLOG *(*get_qlog_cb)(void *arg),
723     void *get_qlog_cb_arg)
724 {
725     ossl_quic_fifd_set_qlog_cb(&txp->fifd, get_qlog_cb, get_qlog_cb_arg);
726 }
727 
ossl_quic_tx_packetiser_discard_enc_level(OSSL_QUIC_TX_PACKETISER * txp,uint32_t enc_level)728 int ossl_quic_tx_packetiser_discard_enc_level(OSSL_QUIC_TX_PACKETISER *txp,
729     uint32_t enc_level)
730 {
731     if (enc_level >= QUIC_ENC_LEVEL_NUM) {
732         ERR_raise(ERR_LIB_SSL, ERR_R_PASSED_INVALID_ARGUMENT);
733         return 0;
734     }
735 
736     if (enc_level != QUIC_ENC_LEVEL_0RTT)
737         txp->args.crypto[ossl_quic_enc_level_to_pn_space(enc_level)] = NULL;
738 
739     return 1;
740 }
741 
ossl_quic_tx_packetiser_notify_handshake_complete(OSSL_QUIC_TX_PACKETISER * txp)742 void ossl_quic_tx_packetiser_notify_handshake_complete(OSSL_QUIC_TX_PACKETISER *txp)
743 {
744     txp->handshake_complete = 1;
745 }
746 
ossl_quic_tx_packetiser_schedule_handshake_done(OSSL_QUIC_TX_PACKETISER * txp)747 void ossl_quic_tx_packetiser_schedule_handshake_done(OSSL_QUIC_TX_PACKETISER *txp)
748 {
749     txp->want_handshake_done = 1;
750 }
751 
ossl_quic_tx_packetiser_schedule_ack_eliciting(OSSL_QUIC_TX_PACKETISER * txp,uint32_t pn_space)752 void ossl_quic_tx_packetiser_schedule_ack_eliciting(OSSL_QUIC_TX_PACKETISER *txp,
753     uint32_t pn_space)
754 {
755     txp->force_ack_eliciting |= (1UL << pn_space);
756 }
757 
ossl_quic_tx_packetiser_schedule_ack(OSSL_QUIC_TX_PACKETISER * txp,uint32_t pn_space)758 void ossl_quic_tx_packetiser_schedule_ack(OSSL_QUIC_TX_PACKETISER *txp,
759     uint32_t pn_space)
760 {
761     txp->want_ack |= (1UL << pn_space);
762 }
763 
764 #define TXP_ERR_INTERNAL 0 /* Internal (e.g. alloc) error */
765 #define TXP_ERR_SUCCESS 1 /* Success */
766 #define TXP_ERR_SPACE 2 /* Not enough room for another packet */
767 #define TXP_ERR_INPUT 3 /* Invalid/malformed input */
768 
769 /*
770  * Generates a datagram by polling the various ELs to determine if they want to
771  * generate any frames, and generating a datagram which coalesces packets for
772  * any ELs which do.
773  */
ossl_quic_tx_packetiser_generate(OSSL_QUIC_TX_PACKETISER * txp,QUIC_TXP_STATUS * status)774 int ossl_quic_tx_packetiser_generate(OSSL_QUIC_TX_PACKETISER *txp,
775     QUIC_TXP_STATUS *status)
776 {
777     /*
778      * Called to generate one or more datagrams, each containing one or more
779      * packets.
780      *
781      * There are some tricky things to note here:
782      *
783      *   - The TXP is only concerned with generating encrypted packets;
784      *     other packets use a different path.
785      *
786      *   - Any datagram containing an Initial packet must have a payload length
787      *     (DPL) of at least 1200 bytes. This padding need not necessarily be
788      *     found in the Initial packet.
789      *
790      *     - It is desirable to be able to coalesce an Initial packet
791      *       with a Handshake packet. Since, before generating the Handshake
792      *       packet, we do not know how long it will be, we cannot know the
793      *       correct amount of padding to ensure a DPL of at least 1200 bytes.
794      *       Thus this padding must added to the Handshake packet (or whatever
795      *       packet is the last in the datagram).
796      *
797      *     - However, at the time that we generate the Initial packet,
798      *       we do not actually know for sure that we will be followed
799      *       in the datagram by another packet. For example, suppose we have
800      *       some queued data (e.g. crypto stream data for the HANDSHAKE EL)
801      *       it looks like we will want to send on the HANDSHAKE EL.
802      *       We could assume padding will be placed in the Handshake packet
803      *       subsequently and avoid adding any padding to the Initial packet
804      *       (which would leave no room for the Handshake packet in the
805      *       datagram).
806      *
807      *       However, this is not actually a safe assumption. Suppose that we
808      *       are using a link with a MDPL of 1200 bytes, the minimum allowed by
809      *       QUIC. Suppose that the Initial packet consumes 1195 bytes in total.
810      *       Since it is not possible to fit a Handshake packet in just 5 bytes,
811      *       upon trying to add a Handshake packet after generating the Initial
812      *       packet, we will discover we have no room to fit it! This is not a
813      *       problem in itself as another datagram can be sent subsequently, but
814      *       it is a problem because we were counting to use that packet to hold
815      *       the essential padding. But if we have already finished encrypting
816      *       the Initial packet, we cannot go and add padding to it anymore.
817      *       This leaves us stuck.
818      *
819      * Because of this, we have to plan multiple packets simultaneously, such
820      * that we can start generating a Handshake (or 0-RTT or 1-RTT, or so on)
821      * packet while still having the option to go back and add padding to the
822      * Initial packet if it turns out to be needed.
823      *
824      * Trying to predict ahead of time (e.g. during Initial packet generation)
825      * whether we will successfully generate a subsequent packet is fraught with
826      * error as it relies on a large number of variables:
827      *
828      *   - Do we have room to fit a packet header? (Consider that due to
829      *     variable-length integer encoding this is highly variable and can even
830      *     depend on payload length due to a variable-length Length field.)
831      *
832      *   - Can we fit even a single one of the frames we want to put in this
833      *     packet in the packet? (Each frame type has a bespoke encoding. While
834      *     our encodings of some frame types are adaptive based on the available
835      *     room - e.g. STREAM frames - ultimately all frame types have some
836      *     absolute minimum number of bytes to be successfully encoded. For
837      *     example, if after an Initial packet there is enough room to encode
838      *     only one byte of frame data, it is quite likely we can't send any of
839      *     the frames we wanted to send.) While this is not strictly a problem
840      *     because we could just fill the packet with padding frames, this is a
841      *     pointless packet and is wasteful.
842      *
843      * Thus we adopt a multi-phase architecture:
844      *
845      *   1. Archetype Selection: Determine desired packet archetype.
846      *
847      *   2. Packet Staging: Generation of packet information and packet payload
848      *      data (frame data) into staging areas.
849      *
850      *   3. Packet Adjustment: Adjustment of staged packets, adding padding to
851      *      the staged packets if needed.
852      *
853      *   4. Commit: The packets are sent to the QTX and recorded as having been
854      *      sent to the FIFM.
855      *
856      */
857     int res = 0, rc;
858     uint32_t archetype, enc_level;
859     uint32_t conn_close_enc_level = QUIC_ENC_LEVEL_NUM;
860     struct txp_pkt pkt[QUIC_ENC_LEVEL_NUM];
861     size_t pkts_done = 0;
862     uint64_t cc_limit = txp->args.cc_method->get_tx_allowance(txp->args.cc_data);
863     int need_padding = 0, txpim_pkt_reffed;
864 
865     memset(status, 0, sizeof(*status));
866 
867     for (enc_level = QUIC_ENC_LEVEL_INITIAL;
868         enc_level < QUIC_ENC_LEVEL_NUM;
869         ++enc_level)
870         pkt[enc_level].h_valid = 0;
871 
872     /*
873      * Should not be needed, but a sanity check in case anyone else has been
874      * using the QTX.
875      */
876     ossl_qtx_finish_dgram(txp->args.qtx);
877 
878     /* 1. Archetype Selection */
879     archetype = txp_determine_archetype(txp, cc_limit);
880 
881     /* 2. Packet Staging */
882     for (enc_level = QUIC_ENC_LEVEL_INITIAL;
883         enc_level < QUIC_ENC_LEVEL_NUM;
884         ++enc_level) {
885         size_t running_total = (enc_level > QUIC_ENC_LEVEL_INITIAL)
886             ? pkt[enc_level - 1].geom.hwm
887             : 0;
888 
889         pkt[enc_level].geom.hwm = running_total;
890 
891         if (!txp_should_try_staging(txp, enc_level, archetype, cc_limit,
892                 &conn_close_enc_level))
893             continue;
894 
895         if (!txp_pkt_init(&pkt[enc_level], txp, enc_level, archetype,
896                 running_total))
897             /*
898              * If this fails this is not a fatal error - it means the geometry
899              * planning determined there was not enough space for another
900              * packet. So just proceed with what we've already planned for.
901              */
902             break;
903 
904         rc = txp_generate_for_el(txp, &pkt[enc_level],
905             conn_close_enc_level == enc_level);
906         if (rc != TXP_ERR_SUCCESS)
907             goto out;
908 
909         if (pkt[enc_level].force_pad)
910             /*
911              * txp_generate_for_el emitted a frame which forces packet padding.
912              */
913             need_padding = 1;
914 
915         pkt[enc_level].geom.hwm = running_total
916             + pkt[enc_level].h.bytes_appended
917             + pkt[enc_level].geom.pkt_overhead;
918     }
919 
920     /* 3. Packet Adjustment */
921     if (pkt[QUIC_ENC_LEVEL_INITIAL].h_valid
922         && pkt[QUIC_ENC_LEVEL_INITIAL].h.bytes_appended > 0)
923         /*
924          * We have an Initial packet in this datagram, so we need to make sure
925          * the total size of the datagram is adequate.
926          */
927         need_padding = 1;
928 
929     if (need_padding) {
930         size_t total_dgram_size = 0;
931         const size_t min_dpl = QUIC_MIN_INITIAL_DGRAM_LEN;
932         uint32_t pad_el = QUIC_ENC_LEVEL_NUM;
933 
934         for (enc_level = QUIC_ENC_LEVEL_INITIAL;
935             enc_level < QUIC_ENC_LEVEL_NUM;
936             ++enc_level)
937             if (pkt[enc_level].h_valid && pkt[enc_level].h.bytes_appended > 0) {
938                 if (pad_el == QUIC_ENC_LEVEL_NUM
939                     /*
940                      * We might not be able to add padding, for example if we
941                      * are using the ACK_ONLY archetype.
942                      */
943                     && pkt[enc_level].geom.adata.allow_padding
944                     && !pkt[enc_level].h.done_implicit)
945                     pad_el = enc_level;
946 
947                 txp_pkt_postgen_update_pkt_overhead(&pkt[enc_level], txp);
948                 total_dgram_size += pkt[enc_level].geom.pkt_overhead
949                     + pkt[enc_level].h.bytes_appended;
950             }
951 
952         if (pad_el != QUIC_ENC_LEVEL_NUM && total_dgram_size < min_dpl) {
953             size_t deficit = min_dpl - total_dgram_size;
954 
955             if (!txp_pkt_append_padding(&pkt[pad_el], txp, deficit))
956                 goto out;
957 
958             total_dgram_size += deficit;
959 
960             /*
961              * Padding frames make a packet ineligible for being a non-inflight
962              * packet.
963              */
964             pkt[pad_el].tpkt->ackm_pkt.is_inflight = 1;
965         }
966 
967         /*
968          * If we have failed to make a datagram of adequate size, for example
969          * because we have a padding requirement but are using the ACK_ONLY
970          * archetype (because we are CC limited), which precludes us from
971          * sending padding, give up on generating the datagram - there is
972          * nothing we can do.
973          */
974         if (total_dgram_size < min_dpl) {
975             res = 1;
976             goto out;
977         }
978     }
979 
980     /* 4. Commit */
981     for (enc_level = QUIC_ENC_LEVEL_INITIAL;
982         enc_level < QUIC_ENC_LEVEL_NUM;
983         ++enc_level) {
984 
985         if (!pkt[enc_level].h_valid)
986             /* Did not attempt to generate a packet for this EL. */
987             continue;
988 
989         if (pkt[enc_level].h.bytes_appended == 0)
990             /* Nothing was generated for this EL, so skip. */
991             continue;
992 
993         if (!ossl_quic_tx_packetiser_check_unvalidated_credit(txp,
994                 pkt[enc_level].h.bytes_appended)) {
995             res = TXP_ERR_SPACE;
996             goto out;
997         }
998         ossl_quic_tx_packetiser_consume_unvalidated_credit(txp, pkt[enc_level].h.bytes_appended);
999 
1000         rc = txp_pkt_commit(txp, &pkt[enc_level], archetype,
1001             &txpim_pkt_reffed);
1002         if (rc) {
1003             status->sent_ack_eliciting
1004                 = status->sent_ack_eliciting
1005                 || pkt[enc_level].tpkt->ackm_pkt.is_ack_eliciting;
1006 
1007             if (enc_level == QUIC_ENC_LEVEL_HANDSHAKE)
1008                 status->sent_handshake
1009                     = (pkt[enc_level].h_valid
1010                         && pkt[enc_level].h.bytes_appended > 0);
1011         }
1012 
1013         if (txpim_pkt_reffed)
1014             pkt[enc_level].tpkt = NULL; /* don't free */
1015 
1016         if (!rc)
1017             goto out;
1018 
1019         ++pkts_done;
1020     }
1021 
1022     /* Flush & Cleanup */
1023     res = 1;
1024 out:
1025     ossl_qtx_finish_dgram(txp->args.qtx);
1026 
1027     for (enc_level = QUIC_ENC_LEVEL_INITIAL;
1028         enc_level < QUIC_ENC_LEVEL_NUM;
1029         ++enc_level)
1030         txp_pkt_cleanup(&pkt[enc_level], txp);
1031 
1032     status->sent_pkt = pkts_done;
1033 
1034     return res;
1035 }
1036 
1037 static const struct archetype_data archetypes[QUIC_ENC_LEVEL_NUM][TX_PACKETISER_ARCHETYPE_NUM] = {
1038     /* EL 0(INITIAL) */
1039     {
1040         /* EL 0(INITIAL) - Archetype 0(NORMAL) */
1041         {
1042             /*allow_ack                       =*/1,
1043             /*allow_ping                      =*/1,
1044             /*allow_crypto                    =*/1,
1045             /*allow_handshake_done            =*/0,
1046             /*allow_path_challenge            =*/0,
1047             /*allow_path_response             =*/0,
1048             /*allow_new_conn_id               =*/0,
1049             /*allow_retire_conn_id            =*/0,
1050             /*allow_stream_rel                =*/0,
1051             /*allow_conn_fc                   =*/0,
1052             /*allow_conn_close                =*/1,
1053             /*allow_cfq_other                 =*/0,
1054             /*allow_new_token                 =*/0,
1055             /*allow_force_ack_eliciting       =*/1,
1056             /*allow_padding                   =*/1,
1057             /*require_ack_eliciting           =*/0,
1058             /*bypass_cc                       =*/0,
1059         },
1060         /* EL 0(INITIAL) - Archetype 1(PROBE) */
1061         {
1062             /*allow_ack                       =*/1,
1063             /*allow_ping                      =*/1,
1064             /*allow_crypto                    =*/1,
1065             /*allow_handshake_done            =*/0,
1066             /*allow_path_challenge            =*/0,
1067             /*allow_path_response             =*/0,
1068             /*allow_new_conn_id               =*/0,
1069             /*allow_retire_conn_id            =*/0,
1070             /*allow_stream_rel                =*/0,
1071             /*allow_conn_fc                   =*/0,
1072             /*allow_conn_close                =*/1,
1073             /*allow_cfq_other                 =*/0,
1074             /*allow_new_token                 =*/0,
1075             /*allow_force_ack_eliciting       =*/1,
1076             /*allow_padding                   =*/1,
1077             /*require_ack_eliciting           =*/1,
1078             /*bypass_cc                       =*/1,
1079         },
1080         /* EL 0(INITIAL) - Archetype 2(ACK_ONLY) */
1081         {
1082             /*allow_ack                       =*/1,
1083             /*allow_ping                      =*/0,
1084             /*allow_crypto                    =*/0,
1085             /*allow_handshake_done            =*/0,
1086             /*allow_path_challenge            =*/0,
1087             /*allow_path_response             =*/0,
1088             /*allow_new_conn_id               =*/0,
1089             /*allow_retire_conn_id            =*/0,
1090             /*allow_stream_rel                =*/0,
1091             /*allow_conn_fc                   =*/0,
1092             /*allow_conn_close                =*/0,
1093             /*allow_cfq_other                 =*/0,
1094             /*allow_new_token                 =*/0,
1095             /*allow_force_ack_eliciting       =*/1,
1096             /*allow_padding                   =*/0,
1097             /*require_ack_eliciting           =*/0,
1098             /*bypass_cc                       =*/1,
1099         },
1100     },
1101     /* EL 1(0RTT) */
1102     {
1103         /* EL 1(0RTT) - Archetype 0(NORMAL) */
1104         {
1105             /*allow_ack                       =*/0,
1106             /*allow_ping                      =*/1,
1107             /*allow_crypto                    =*/0,
1108             /*allow_handshake_done            =*/0,
1109             /*allow_path_challenge            =*/0,
1110             /*allow_path_response             =*/0,
1111             /*allow_new_conn_id               =*/1,
1112             /*allow_retire_conn_id            =*/1,
1113             /*allow_stream_rel                =*/1,
1114             /*allow_conn_fc                   =*/1,
1115             /*allow_conn_close                =*/1,
1116             /*allow_cfq_other                 =*/0,
1117             /*allow_new_token                 =*/0,
1118             /*allow_force_ack_eliciting       =*/0,
1119             /*allow_padding                   =*/1,
1120             /*require_ack_eliciting           =*/0,
1121             /*bypass_cc                       =*/0,
1122         },
1123         /* EL 1(0RTT) - Archetype 1(PROBE) */
1124         {
1125             /*allow_ack                       =*/0,
1126             /*allow_ping                      =*/1,
1127             /*allow_crypto                    =*/0,
1128             /*allow_handshake_done            =*/0,
1129             /*allow_path_challenge            =*/0,
1130             /*allow_path_response             =*/0,
1131             /*allow_new_conn_id               =*/1,
1132             /*allow_retire_conn_id            =*/1,
1133             /*allow_stream_rel                =*/1,
1134             /*allow_conn_fc                   =*/1,
1135             /*allow_conn_close                =*/1,
1136             /*allow_cfq_other                 =*/0,
1137             /*allow_new_token                 =*/0,
1138             /*allow_force_ack_eliciting       =*/0,
1139             /*allow_padding                   =*/1,
1140             /*require_ack_eliciting           =*/1,
1141             /*bypass_cc                       =*/1,
1142         },
1143         /* EL 1(0RTT) - Archetype 2(ACK_ONLY) */
1144         {
1145             /*allow_ack                       =*/0,
1146             /*allow_ping                      =*/0,
1147             /*allow_crypto                    =*/0,
1148             /*allow_handshake_done            =*/0,
1149             /*allow_path_challenge            =*/0,
1150             /*allow_path_response             =*/0,
1151             /*allow_new_conn_id               =*/0,
1152             /*allow_retire_conn_id            =*/0,
1153             /*allow_stream_rel                =*/0,
1154             /*allow_conn_fc                   =*/0,
1155             /*allow_conn_close                =*/0,
1156             /*allow_cfq_other                 =*/0,
1157             /*allow_new_token                 =*/0,
1158             /*allow_force_ack_eliciting       =*/0,
1159             /*allow_padding                   =*/0,
1160             /*require_ack_eliciting           =*/0,
1161             /*bypass_cc                       =*/1,
1162         },
1163     },
1164     /* EL (HANDSHAKE) */
1165     {
1166         /* EL 2(HANDSHAKE) - Archetype 0(NORMAL) */
1167         {
1168             /*allow_ack                       =*/1,
1169             /*allow_ping                      =*/1,
1170             /*allow_crypto                    =*/1,
1171             /*allow_handshake_done            =*/0,
1172             /*allow_path_challenge            =*/0,
1173             /*allow_path_response             =*/0,
1174             /*allow_new_conn_id               =*/0,
1175             /*allow_retire_conn_id            =*/0,
1176             /*allow_stream_rel                =*/0,
1177             /*allow_conn_fc                   =*/0,
1178             /*allow_conn_close                =*/1,
1179             /*allow_cfq_other                 =*/0,
1180             /*allow_new_token                 =*/0,
1181             /*allow_force_ack_eliciting       =*/1,
1182             /*allow_padding                   =*/1,
1183             /*require_ack_eliciting           =*/0,
1184             /*bypass_cc                       =*/0,
1185         },
1186         /* EL 2(HANDSHAKE) - Archetype 1(PROBE) */
1187         {
1188             /*allow_ack                       =*/1,
1189             /*allow_ping                      =*/1,
1190             /*allow_crypto                    =*/1,
1191             /*allow_handshake_done            =*/0,
1192             /*allow_path_challenge            =*/0,
1193             /*allow_path_response             =*/0,
1194             /*allow_new_conn_id               =*/0,
1195             /*allow_retire_conn_id            =*/0,
1196             /*allow_stream_rel                =*/0,
1197             /*allow_conn_fc                   =*/0,
1198             /*allow_conn_close                =*/1,
1199             /*allow_cfq_other                 =*/0,
1200             /*allow_new_token                 =*/0,
1201             /*allow_force_ack_eliciting       =*/1,
1202             /*allow_padding                   =*/1,
1203             /*require_ack_eliciting           =*/1,
1204             /*bypass_cc                       =*/1,
1205         },
1206         /* EL 2(HANDSHAKE) - Archetype 2(ACK_ONLY) */
1207         {
1208             /*allow_ack                       =*/1,
1209             /*allow_ping                      =*/0,
1210             /*allow_crypto                    =*/0,
1211             /*allow_handshake_done            =*/0,
1212             /*allow_path_challenge            =*/0,
1213             /*allow_path_response             =*/0,
1214             /*allow_new_conn_id               =*/0,
1215             /*allow_retire_conn_id            =*/0,
1216             /*allow_stream_rel                =*/0,
1217             /*allow_conn_fc                   =*/0,
1218             /*allow_conn_close                =*/0,
1219             /*allow_cfq_other                 =*/0,
1220             /*allow_new_token                 =*/0,
1221             /*allow_force_ack_eliciting       =*/1,
1222             /*allow_padding                   =*/0,
1223             /*require_ack_eliciting           =*/0,
1224             /*bypass_cc                       =*/1,
1225         },
1226     },
1227     /* EL 3(1RTT) */
1228     {
1229         /* EL 3(1RTT) - Archetype 0(NORMAL) */
1230         {
1231             /*allow_ack                       =*/1,
1232             /*allow_ping                      =*/1,
1233             /*allow_crypto                    =*/1,
1234             /*allow_handshake_done            =*/1,
1235             /*allow_path_challenge            =*/0,
1236             /*allow_path_response             =*/1,
1237             /*allow_new_conn_id               =*/1,
1238             /*allow_retire_conn_id            =*/1,
1239             /*allow_stream_rel                =*/1,
1240             /*allow_conn_fc                   =*/1,
1241             /*allow_conn_close                =*/1,
1242             /*allow_cfq_other                 =*/1,
1243             /*allow_new_token                 =*/1,
1244             /*allow_force_ack_eliciting       =*/1,
1245             /*allow_padding                   =*/1,
1246             /*require_ack_eliciting           =*/0,
1247             /*bypass_cc                       =*/0,
1248         },
1249         /* EL 3(1RTT) - Archetype 1(PROBE) */
1250         {
1251             /*allow_ack                       =*/1,
1252             /*allow_ping                      =*/1,
1253             /*allow_crypto                    =*/1,
1254             /*allow_handshake_done            =*/1,
1255             /*allow_path_challenge            =*/0,
1256             /*allow_path_response             =*/1,
1257             /*allow_new_conn_id               =*/1,
1258             /*allow_retire_conn_id            =*/1,
1259             /*allow_stream_rel                =*/1,
1260             /*allow_conn_fc                   =*/1,
1261             /*allow_conn_close                =*/1,
1262             /*allow_cfq_other                 =*/1,
1263             /*allow_new_token                 =*/1,
1264             /*allow_force_ack_eliciting       =*/1,
1265             /*allow_padding                   =*/1,
1266             /*require_ack_eliciting           =*/1,
1267             /*bypass_cc                       =*/1,
1268         },
1269         /* EL 3(1RTT) - Archetype 2(ACK_ONLY) */
1270         {
1271             /*allow_ack                       =*/1,
1272             /*allow_ping                      =*/0,
1273             /*allow_crypto                    =*/0,
1274             /*allow_handshake_done            =*/0,
1275             /*allow_path_challenge            =*/0,
1276             /*allow_path_response             =*/0,
1277             /*allow_new_conn_id               =*/0,
1278             /*allow_retire_conn_id            =*/0,
1279             /*allow_stream_rel                =*/0,
1280             /*allow_conn_fc                   =*/0,
1281             /*allow_conn_close                =*/0,
1282             /*allow_cfq_other                 =*/0,
1283             /*allow_new_token                 =*/0,
1284             /*allow_force_ack_eliciting       =*/1,
1285             /*allow_padding                   =*/0,
1286             /*require_ack_eliciting           =*/0,
1287             /*bypass_cc                       =*/1,
1288         } }
1289 };
1290 
txp_get_archetype_data(uint32_t enc_level,uint32_t archetype,struct archetype_data * a)1291 static int txp_get_archetype_data(uint32_t enc_level,
1292     uint32_t archetype,
1293     struct archetype_data *a)
1294 {
1295     if (enc_level >= QUIC_ENC_LEVEL_NUM
1296         || archetype >= TX_PACKETISER_ARCHETYPE_NUM)
1297         return 0;
1298 
1299     /* No need to avoid copying this as it should not exceed one int in size. */
1300     *a = archetypes[enc_level][archetype];
1301     return 1;
1302 }
1303 
txp_determine_geometry(OSSL_QUIC_TX_PACKETISER * txp,uint32_t archetype,uint32_t enc_level,size_t running_total,QUIC_PKT_HDR * phdr,struct txp_pkt_geom * geom)1304 static int txp_determine_geometry(OSSL_QUIC_TX_PACKETISER *txp,
1305     uint32_t archetype,
1306     uint32_t enc_level,
1307     size_t running_total,
1308     QUIC_PKT_HDR *phdr,
1309     struct txp_pkt_geom *geom)
1310 {
1311     size_t mdpl, cmpl, hdr_len;
1312 
1313     /* Get information about packet archetype. */
1314     if (!txp_get_archetype_data(enc_level, archetype, &geom->adata))
1315         return 0;
1316 
1317     /* Assemble packet header. */
1318     phdr->type = ossl_quic_enc_level_to_pkt_type(enc_level);
1319     phdr->spin_bit = 0;
1320     phdr->pn_len = txp_determine_pn_len(txp);
1321     phdr->partial = 0;
1322     phdr->fixed = 1;
1323     phdr->reserved = 0;
1324     phdr->version = txp->args.protocol_version;
1325     phdr->dst_conn_id = txp->args.cur_dcid;
1326     phdr->src_conn_id = txp->args.cur_scid;
1327 
1328     /*
1329      * We need to know the length of the payload to get an accurate header
1330      * length for non-1RTT packets, because the Length field found in
1331      * Initial/Handshake/0-RTT packets uses a variable-length encoding. However,
1332      * we don't have a good idea of the length of our payload, because the
1333      * length of the payload depends on the room in the datagram after fitting
1334      * the header, which depends on the size of the header.
1335      *
1336      * In general, it does not matter if a packet is slightly shorter (because
1337      * e.g. we predicted use of a 2-byte length field, but ended up only needing
1338      * a 1-byte length field). However this does matter for Initial packets
1339      * which must be at least 1200 bytes, which is also the assumed default MTU;
1340      * therefore in many cases Initial packets will be padded to 1200 bytes,
1341      * which means if we overestimated the header size, we will be short by a
1342      * few bytes and the server will ignore the packet for being too short. In
1343      * this case, however, such packets always *will* be padded to meet 1200
1344      * bytes, which requires a 2-byte length field, so we don't actually need to
1345      * worry about this. Thus we estimate the header length assuming a 2-byte
1346      * length field here, which should in practice work well in all cases.
1347      */
1348     phdr->len = OSSL_QUIC_VLINT_2B_MAX - phdr->pn_len;
1349 
1350     if (enc_level == QUIC_ENC_LEVEL_INITIAL) {
1351         phdr->token = txp->initial_token;
1352         phdr->token_len = txp->initial_token_len;
1353     } else {
1354         phdr->token = NULL;
1355         phdr->token_len = 0;
1356     }
1357 
1358     hdr_len = ossl_quic_wire_get_encoded_pkt_hdr_len(phdr->dst_conn_id.id_len,
1359         phdr);
1360     if (hdr_len == 0)
1361         return 0;
1362 
1363     /* MDPL: Maximum datagram payload length. */
1364     mdpl = txp_get_mdpl(txp);
1365 
1366     /*
1367      * CMPL: Maximum encoded packet size we can put into this datagram given any
1368      * previous packets coalesced into it.
1369      */
1370     if (running_total > mdpl)
1371         /* Should not be possible, but if it happens: */
1372         cmpl = 0;
1373     else
1374         cmpl = mdpl - running_total;
1375 
1376     /* CMPPL: Maximum amount we can put into the current packet payload */
1377     if (!txp_determine_ppl_from_pl(txp, cmpl, enc_level, hdr_len, &geom->cmppl))
1378         return 0;
1379 
1380     geom->cmpl = cmpl;
1381     geom->pkt_overhead = cmpl - geom->cmppl;
1382     geom->archetype = archetype;
1383     return 1;
1384 }
1385 
txp_determine_archetype(OSSL_QUIC_TX_PACKETISER * txp,uint64_t cc_limit)1386 static uint32_t txp_determine_archetype(OSSL_QUIC_TX_PACKETISER *txp,
1387     uint64_t cc_limit)
1388 {
1389     OSSL_ACKM_PROBE_INFO *probe_info
1390         = ossl_ackm_get0_probe_request(txp->args.ackm);
1391     uint32_t pn_space;
1392 
1393     /*
1394      * If ACKM has requested probe generation (e.g. due to PTO), we generate a
1395      * Probe-archetype packet. Actually, we determine archetype on a
1396      * per-datagram basis, so if any EL wants a probe, do a pass in which
1397      * we try and generate a probe (if needed) for all ELs.
1398      */
1399     if (probe_info->anti_deadlock_initial > 0
1400         || probe_info->anti_deadlock_handshake > 0)
1401         return TX_PACKETISER_ARCHETYPE_PROBE;
1402 
1403     for (pn_space = QUIC_PN_SPACE_INITIAL;
1404         pn_space < QUIC_PN_SPACE_NUM;
1405         ++pn_space)
1406         if (probe_info->pto[pn_space] > 0)
1407             return TX_PACKETISER_ARCHETYPE_PROBE;
1408 
1409     /*
1410      * If we are out of CC budget, we cannot send a normal packet,
1411      * but we can do an ACK-only packet (potentially, if we
1412      * want to send an ACK).
1413      */
1414     if (cc_limit == 0)
1415         return TX_PACKETISER_ARCHETYPE_ACK_ONLY;
1416 
1417     /* All other packets. */
1418     return TX_PACKETISER_ARCHETYPE_NORMAL;
1419 }
1420 
txp_should_try_staging(OSSL_QUIC_TX_PACKETISER * txp,uint32_t enc_level,uint32_t archetype,uint64_t cc_limit,uint32_t * conn_close_enc_level)1421 static int txp_should_try_staging(OSSL_QUIC_TX_PACKETISER *txp,
1422     uint32_t enc_level,
1423     uint32_t archetype,
1424     uint64_t cc_limit,
1425     uint32_t *conn_close_enc_level)
1426 {
1427     struct archetype_data a;
1428     uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
1429     QUIC_CFQ_ITEM *cfq_item;
1430 
1431     if (!ossl_qtx_is_enc_level_provisioned(txp->args.qtx, enc_level))
1432         return 0;
1433 
1434     if (!txp_get_archetype_data(enc_level, archetype, &a))
1435         return 0;
1436 
1437     if (!a.bypass_cc && cc_limit == 0)
1438         /* CC not allowing us to send. */
1439         return 0;
1440 
1441     /*
1442      * We can produce CONNECTION_CLOSE frames on any EL in principle, which
1443      * means we need to choose which EL we would prefer to use. After a
1444      * connection is fully established we have only one provisioned EL and this
1445      * is a non-issue. Where multiple ELs are provisioned, it is possible the
1446      * peer does not have the keys for the EL yet, which suggests in general it
1447      * is preferable to use the lowest EL which is still provisioned.
1448      *
1449      * However (RFC 9000 s. 10.2.3 & 12.5) we are also required to not send
1450      * application CONNECTION_CLOSE frames in non-1-RTT ELs, so as to not
1451      * potentially leak application data on a connection which has yet to be
1452      * authenticated. Thus when we have an application CONNECTION_CLOSE frame
1453      * queued and need to send it on a non-1-RTT EL, we have to convert it
1454      * into a transport CONNECTION_CLOSE frame which contains no application
1455      * data. Since this loses information, it suggests we should use the 1-RTT
1456      * EL to avoid this if possible, even if a lower EL is also available.
1457      *
1458      * At the same time, just because we have the 1-RTT EL provisioned locally
1459      * does not necessarily mean the peer does, for example if a handshake
1460      * CRYPTO frame has been lost. It is fairly important that CONNECTION_CLOSE
1461      * is signalled in a way we know our peer can decrypt, as we stop processing
1462      * connection retransmission logic for real after connection close and
1463      * simply 'blindly' retransmit the same CONNECTION_CLOSE frame.
1464      *
1465      * This is not a major concern for clients, since if a client has a 1-RTT EL
1466      * provisioned the server is guaranteed to also have a 1-RTT EL provisioned.
1467      *
1468      * TODO(QUIC FUTURE): Revisit this when when have reached a decision on how
1469      * best to implement this
1470      */
1471     if (*conn_close_enc_level > enc_level
1472         && *conn_close_enc_level != QUIC_ENC_LEVEL_1RTT)
1473         *conn_close_enc_level = enc_level;
1474 
1475     /* Do we need to send a PTO probe? */
1476     if (a.allow_force_ack_eliciting) {
1477         OSSL_ACKM_PROBE_INFO *probe_info
1478             = ossl_ackm_get0_probe_request(txp->args.ackm);
1479 
1480         if ((enc_level == QUIC_ENC_LEVEL_INITIAL
1481                 && probe_info->anti_deadlock_initial > 0)
1482             || (enc_level == QUIC_ENC_LEVEL_HANDSHAKE
1483                 && probe_info->anti_deadlock_handshake > 0)
1484             || probe_info->pto[pn_space] > 0)
1485             return 1;
1486     }
1487 
1488     /* Does the crypto stream for this EL want to produce anything? */
1489     if (a.allow_crypto && sstream_is_pending(txp->args.crypto[pn_space]))
1490         return 1;
1491 
1492     /* Does the ACKM for this PN space want to produce anything? */
1493     if (a.allow_ack && (ossl_ackm_is_ack_desired(txp->args.ackm, pn_space) || (txp->want_ack & (1UL << pn_space)) != 0))
1494         return 1;
1495 
1496     /* Do we need to force emission of an ACK-eliciting packet? */
1497     if (a.allow_force_ack_eliciting
1498         && (txp->force_ack_eliciting & (1UL << pn_space)) != 0)
1499         return 1;
1500 
1501     /* Does the connection-level RXFC want to produce a frame? */
1502     if (a.allow_conn_fc && (txp->want_max_data || ossl_quic_rxfc_has_cwm_changed(txp->args.conn_rxfc, 0)))
1503         return 1;
1504 
1505     /* Do we want to produce a MAX_STREAMS frame? */
1506     if (a.allow_conn_fc
1507         && (txp->want_max_streams_bidi
1508             || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_bidi_rxfc,
1509                 0)
1510             || txp->want_max_streams_uni
1511             || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_uni_rxfc,
1512                 0)))
1513         return 1;
1514 
1515     /* Do we want to produce a HANDSHAKE_DONE frame? */
1516     if (a.allow_handshake_done && txp->want_handshake_done)
1517         return 1;
1518 
1519     /* Do we want to produce a CONNECTION_CLOSE frame? */
1520     if (a.allow_conn_close && txp->want_conn_close && *conn_close_enc_level == enc_level)
1521         /*
1522          * This is a bit of a special case since CONNECTION_CLOSE can appear in
1523          * most packet types, and when we decide we want to send it this status
1524          * isn't tied to a specific EL. So if we want to send it, we send it
1525          * only on the lowest non-dropped EL.
1526          */
1527         return 1;
1528 
1529     /* Does the CFQ have any frames queued for this PN space? */
1530     if (enc_level != QUIC_ENC_LEVEL_0RTT)
1531         for (cfq_item = ossl_quic_cfq_get_priority_head(txp->args.cfq, pn_space);
1532             cfq_item != NULL;
1533             cfq_item = ossl_quic_cfq_item_get_priority_next(cfq_item, pn_space)) {
1534             uint64_t frame_type = ossl_quic_cfq_item_get_frame_type(cfq_item);
1535 
1536             switch (frame_type) {
1537             case OSSL_QUIC_FRAME_TYPE_NEW_CONN_ID:
1538                 if (a.allow_new_conn_id)
1539                     return 1;
1540                 break;
1541             case OSSL_QUIC_FRAME_TYPE_RETIRE_CONN_ID:
1542                 if (a.allow_retire_conn_id)
1543                     return 1;
1544                 break;
1545             case OSSL_QUIC_FRAME_TYPE_NEW_TOKEN:
1546                 if (a.allow_new_token)
1547                     return 1;
1548                 break;
1549             case OSSL_QUIC_FRAME_TYPE_PATH_RESPONSE:
1550                 if (a.allow_path_response)
1551                     return 1;
1552                 break;
1553             default:
1554                 if (a.allow_cfq_other)
1555                     return 1;
1556                 break;
1557             }
1558         }
1559 
1560     if (a.allow_stream_rel && txp->handshake_complete) {
1561         QUIC_STREAM_ITER it;
1562 
1563         /* If there are any active streams, 0/1-RTT wants to produce a packet.
1564          * Whether a stream is on the active list is required to be precise
1565          * (i.e., a stream is never on the active list if we cannot produce a
1566          * frame for it), and all stream-related frames are governed by
1567          * a.allow_stream_rel (i.e., if we can send one type of stream-related
1568          * frame, we can send any of them), so we don't need to inspect
1569          * individual streams on the active list, just confirm that the active
1570          * list is non-empty.
1571          */
1572         ossl_quic_stream_iter_init(&it, txp->args.qsm, 0);
1573         if (it.stream != NULL)
1574             return 1;
1575     }
1576 
1577     return 0;
1578 }
1579 
sstream_is_pending(QUIC_SSTREAM * sstream)1580 static int sstream_is_pending(QUIC_SSTREAM *sstream)
1581 {
1582     OSSL_QUIC_FRAME_STREAM hdr;
1583     OSSL_QTX_IOVEC iov[2];
1584     size_t num_iov = OSSL_NELEM(iov);
1585 
1586     return ossl_quic_sstream_get_stream_frame(sstream, 0, &hdr, iov, &num_iov);
1587 }
1588 
1589 /* Determine how many bytes we should use for the encoded PN. */
txp_determine_pn_len(OSSL_QUIC_TX_PACKETISER * txp)1590 static size_t txp_determine_pn_len(OSSL_QUIC_TX_PACKETISER *txp)
1591 {
1592     return 4; /* TODO(QUIC FUTURE) */
1593 }
1594 
1595 /* Determine plaintext packet payload length from payload length. */
txp_determine_ppl_from_pl(OSSL_QUIC_TX_PACKETISER * txp,size_t pl,uint32_t enc_level,size_t hdr_len,size_t * r)1596 static int txp_determine_ppl_from_pl(OSSL_QUIC_TX_PACKETISER *txp,
1597     size_t pl,
1598     uint32_t enc_level,
1599     size_t hdr_len,
1600     size_t *r)
1601 {
1602     if (pl < hdr_len)
1603         return 0;
1604 
1605     pl -= hdr_len;
1606 
1607     if (!ossl_qtx_calculate_plaintext_payload_len(txp->args.qtx, enc_level,
1608             pl, &pl))
1609         return 0;
1610 
1611     *r = pl;
1612     return 1;
1613 }
1614 
txp_get_mdpl(OSSL_QUIC_TX_PACKETISER * txp)1615 static size_t txp_get_mdpl(OSSL_QUIC_TX_PACKETISER *txp)
1616 {
1617     return ossl_qtx_get_mdpl(txp->args.qtx);
1618 }
1619 
get_sstream_by_id(uint64_t stream_id,uint32_t pn_space,void * arg)1620 static QUIC_SSTREAM *get_sstream_by_id(uint64_t stream_id, uint32_t pn_space,
1621     void *arg)
1622 {
1623     OSSL_QUIC_TX_PACKETISER *txp = arg;
1624     QUIC_STREAM *s;
1625 
1626     if (stream_id == UINT64_MAX)
1627         return txp->args.crypto[pn_space];
1628 
1629     s = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1630     if (s == NULL)
1631         return NULL;
1632 
1633     return s->sstream;
1634 }
1635 
on_regen_notify(uint64_t frame_type,uint64_t stream_id,QUIC_TXPIM_PKT * pkt,void * arg)1636 static void on_regen_notify(uint64_t frame_type, uint64_t stream_id,
1637     QUIC_TXPIM_PKT *pkt, void *arg)
1638 {
1639     OSSL_QUIC_TX_PACKETISER *txp = arg;
1640 
1641     switch (frame_type) {
1642     case OSSL_QUIC_FRAME_TYPE_HANDSHAKE_DONE:
1643         txp->want_handshake_done = 1;
1644         break;
1645     case OSSL_QUIC_FRAME_TYPE_MAX_DATA:
1646         txp->want_max_data = 1;
1647         break;
1648     case OSSL_QUIC_FRAME_TYPE_MAX_STREAMS_BIDI:
1649         txp->want_max_streams_bidi = 1;
1650         break;
1651     case OSSL_QUIC_FRAME_TYPE_MAX_STREAMS_UNI:
1652         txp->want_max_streams_uni = 1;
1653         break;
1654     case OSSL_QUIC_FRAME_TYPE_ACK_WITH_ECN:
1655         txp->want_ack |= (1UL << pkt->ackm_pkt.pkt_space);
1656         break;
1657     case OSSL_QUIC_FRAME_TYPE_MAX_STREAM_DATA: {
1658         QUIC_STREAM *s
1659             = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1660 
1661         if (s == NULL)
1662             return;
1663 
1664         s->want_max_stream_data = 1;
1665         ossl_quic_stream_map_update_state(txp->args.qsm, s);
1666     } break;
1667     case OSSL_QUIC_FRAME_TYPE_STOP_SENDING: {
1668         QUIC_STREAM *s
1669             = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1670 
1671         if (s == NULL)
1672             return;
1673 
1674         ossl_quic_stream_map_schedule_stop_sending(txp->args.qsm, s);
1675     } break;
1676     case OSSL_QUIC_FRAME_TYPE_RESET_STREAM: {
1677         QUIC_STREAM *s
1678             = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1679 
1680         if (s == NULL)
1681             return;
1682 
1683         s->want_reset_stream = 1;
1684         ossl_quic_stream_map_update_state(txp->args.qsm, s);
1685     } break;
1686     default:
1687         assert(0);
1688         break;
1689     }
1690 }
1691 
txp_need_ping(OSSL_QUIC_TX_PACKETISER * txp,uint32_t pn_space,const struct archetype_data * adata)1692 static int txp_need_ping(OSSL_QUIC_TX_PACKETISER *txp,
1693     uint32_t pn_space,
1694     const struct archetype_data *adata)
1695 {
1696     return adata->allow_ping
1697         && (adata->require_ack_eliciting
1698             || (txp->force_ack_eliciting & (1UL << pn_space)) != 0);
1699 }
1700 
txp_pkt_init(struct txp_pkt * pkt,OSSL_QUIC_TX_PACKETISER * txp,uint32_t enc_level,uint32_t archetype,size_t running_total)1701 static int txp_pkt_init(struct txp_pkt *pkt, OSSL_QUIC_TX_PACKETISER *txp,
1702     uint32_t enc_level, uint32_t archetype,
1703     size_t running_total)
1704 {
1705     uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
1706 
1707     if (!txp_determine_geometry(txp, archetype, enc_level,
1708             running_total, &pkt->phdr, &pkt->geom))
1709         return 0;
1710 
1711     /*
1712      * Initialise TX helper. If we must be ACK eliciting, reserve 1 byte for
1713      * PING.
1714      */
1715     if (!tx_helper_init(&pkt->h, txp, enc_level,
1716             pkt->geom.cmppl,
1717             txp_need_ping(txp, pn_space, &pkt->geom.adata) ? 1 : 0))
1718         return 0;
1719 
1720     pkt->h_valid = 1;
1721     pkt->tpkt = NULL;
1722     pkt->stream_head = NULL;
1723     pkt->force_pad = 0;
1724     return 1;
1725 }
1726 
txp_pkt_cleanup(struct txp_pkt * pkt,OSSL_QUIC_TX_PACKETISER * txp)1727 static void txp_pkt_cleanup(struct txp_pkt *pkt, OSSL_QUIC_TX_PACKETISER *txp)
1728 {
1729     if (!pkt->h_valid)
1730         return;
1731 
1732     tx_helper_cleanup(&pkt->h);
1733     pkt->h_valid = 0;
1734 
1735     if (pkt->tpkt != NULL) {
1736         ossl_quic_txpim_pkt_release(txp->args.txpim, pkt->tpkt);
1737         pkt->tpkt = NULL;
1738     }
1739 }
1740 
txp_pkt_postgen_update_pkt_overhead(struct txp_pkt * pkt,OSSL_QUIC_TX_PACKETISER * txp)1741 static int txp_pkt_postgen_update_pkt_overhead(struct txp_pkt *pkt,
1742     OSSL_QUIC_TX_PACKETISER *txp)
1743 {
1744     /*
1745      * After we have staged and generated our packets, but before we commit
1746      * them, it is possible for the estimated packet overhead (packet header +
1747      * AEAD tag size) to shrink slightly because we generated a short packet
1748      * whose which can be represented in fewer bytes as a variable-length
1749      * integer than we were (pessimistically) budgeting for. We need to account
1750      * for this to ensure that we get our padding calculation exactly right.
1751      *
1752      * Update pkt_overhead to be accurate now that we know how much data is
1753      * going in a packet.
1754      */
1755     size_t hdr_len, ciphertext_len;
1756 
1757     if (pkt->h.enc_level == QUIC_ENC_LEVEL_INITIAL)
1758         /*
1759          * Don't update overheads for the INITIAL EL - we have not finished
1760          * appending padding to it and would potentially miscalculate the
1761          * correct padding if we now update the pkt_overhead field to switch to
1762          * e.g. a 1-byte length field in the packet header. Since we are padding
1763          * to QUIC_MIN_INITIAL_DGRAM_LEN which requires a 2-byte length field,
1764          * this is guaranteed to be moot anyway. See comment in
1765          * txp_determine_geometry for more information.
1766          */
1767         return 1;
1768 
1769     if (!ossl_qtx_calculate_ciphertext_payload_len(txp->args.qtx, pkt->h.enc_level,
1770             pkt->h.bytes_appended,
1771             &ciphertext_len))
1772         return 0;
1773 
1774     pkt->phdr.len = ciphertext_len;
1775 
1776     hdr_len = ossl_quic_wire_get_encoded_pkt_hdr_len(pkt->phdr.dst_conn_id.id_len,
1777         &pkt->phdr);
1778 
1779     pkt->geom.pkt_overhead = hdr_len + ciphertext_len - pkt->h.bytes_appended;
1780     return 1;
1781 }
1782 
on_confirm_notify(uint64_t frame_type,uint64_t stream_id,QUIC_TXPIM_PKT * pkt,void * arg)1783 static void on_confirm_notify(uint64_t frame_type, uint64_t stream_id,
1784     QUIC_TXPIM_PKT *pkt, void *arg)
1785 {
1786     OSSL_QUIC_TX_PACKETISER *txp = arg;
1787 
1788     switch (frame_type) {
1789     case OSSL_QUIC_FRAME_TYPE_STOP_SENDING: {
1790         QUIC_STREAM *s
1791             = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1792 
1793         if (s == NULL)
1794             return;
1795 
1796         s->acked_stop_sending = 1;
1797         ossl_quic_stream_map_update_state(txp->args.qsm, s);
1798     } break;
1799     case OSSL_QUIC_FRAME_TYPE_RESET_STREAM: {
1800         QUIC_STREAM *s
1801             = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1802 
1803         if (s == NULL)
1804             return;
1805 
1806         /*
1807          * We must already be in RESET_SENT or RESET_RECVD if we are
1808          * here, so we don't need to check state here.
1809          */
1810         ossl_quic_stream_map_notify_reset_stream_acked(txp->args.qsm, s);
1811         ossl_quic_stream_map_update_state(txp->args.qsm, s);
1812     } break;
1813     default:
1814         assert(0);
1815         break;
1816     }
1817 }
1818 
txp_pkt_append_padding(struct txp_pkt * pkt,OSSL_QUIC_TX_PACKETISER * txp,size_t num_bytes)1819 static int txp_pkt_append_padding(struct txp_pkt *pkt,
1820     OSSL_QUIC_TX_PACKETISER *txp, size_t num_bytes)
1821 {
1822     WPACKET *wpkt;
1823 
1824     if (num_bytes == 0)
1825         return 1;
1826 
1827     if (!ossl_assert(pkt->h_valid))
1828         return 0;
1829 
1830     if (!ossl_assert(pkt->tpkt != NULL))
1831         return 0;
1832 
1833     wpkt = tx_helper_begin(&pkt->h);
1834     if (wpkt == NULL)
1835         return 0;
1836 
1837     if (!ossl_quic_wire_encode_padding(wpkt, num_bytes)) {
1838         tx_helper_rollback(&pkt->h);
1839         return 0;
1840     }
1841 
1842     if (!tx_helper_commit(&pkt->h))
1843         return 0;
1844 
1845     pkt->tpkt->ackm_pkt.num_bytes += num_bytes;
1846     /* Cannot be non-inflight if we have a PADDING frame */
1847     pkt->tpkt->ackm_pkt.is_inflight = 1;
1848     return 1;
1849 }
1850 
on_sstream_updated(uint64_t stream_id,void * arg)1851 static void on_sstream_updated(uint64_t stream_id, void *arg)
1852 {
1853     OSSL_QUIC_TX_PACKETISER *txp = arg;
1854     QUIC_STREAM *s;
1855 
1856     s = ossl_quic_stream_map_get_by_id(txp->args.qsm, stream_id);
1857     if (s == NULL)
1858         return;
1859 
1860     ossl_quic_stream_map_update_state(txp->args.qsm, s);
1861 }
1862 
1863 /*
1864  * Returns 1 if we can send that many bytes in closing state, 0 otherwise.
1865  * Also maintains the bytes sent state if it returns a success.
1866  */
try_commit_conn_close(OSSL_QUIC_TX_PACKETISER * txp,size_t n)1867 static int try_commit_conn_close(OSSL_QUIC_TX_PACKETISER *txp, size_t n)
1868 {
1869     int res;
1870 
1871     /* We can always send the first connection close frame */
1872     if (txp->closing_bytes_recv == 0)
1873         return 1;
1874 
1875     /*
1876      * RFC 9000 s. 10.2.1 Closing Connection State:
1877      *      To avoid being used for an amplification attack, such
1878      *      endpoints MUST limit the cumulative size of packets it sends
1879      *      to three times the cumulative size of the packets that are
1880      *      received and attributed to the connection.
1881      * and:
1882      *      An endpoint in the closing state MUST either discard packets
1883      *      received from an unvalidated address or limit the cumulative
1884      *      size of packets it sends to an unvalidated address to three
1885      *      times the size of packets it receives from that address.
1886      */
1887     res = txp->closing_bytes_xmit + n <= txp->closing_bytes_recv * 3;
1888 
1889     /*
1890      * Attribute the bytes to the connection, if we are allowed to send them
1891      * and this isn't the first closing frame.
1892      */
1893     if (res && txp->closing_bytes_recv != 0)
1894         txp->closing_bytes_xmit += n;
1895     return res;
1896 }
1897 
ossl_quic_tx_packetiser_record_received_closing_bytes(OSSL_QUIC_TX_PACKETISER * txp,size_t n)1898 void ossl_quic_tx_packetiser_record_received_closing_bytes(
1899     OSSL_QUIC_TX_PACKETISER *txp, size_t n)
1900 {
1901     txp->closing_bytes_recv += n;
1902 }
1903 
txp_generate_pre_token(OSSL_QUIC_TX_PACKETISER * txp,struct txp_pkt * pkt,int chosen_for_conn_close,int * can_be_non_inflight)1904 static int txp_generate_pre_token(OSSL_QUIC_TX_PACKETISER *txp,
1905     struct txp_pkt *pkt,
1906     int chosen_for_conn_close,
1907     int *can_be_non_inflight)
1908 {
1909     const uint32_t enc_level = pkt->h.enc_level;
1910     const uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
1911     const struct archetype_data *a = &pkt->geom.adata;
1912     QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
1913     struct tx_helper *h = &pkt->h;
1914     const OSSL_QUIC_FRAME_ACK *ack;
1915     OSSL_QUIC_FRAME_ACK ack2;
1916 
1917     tpkt->ackm_pkt.largest_acked = QUIC_PN_INVALID;
1918 
1919     /* ACK Frames (Regenerate) */
1920     if (a->allow_ack
1921         && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_ACK
1922         && (((txp->want_ack & (1UL << pn_space)) != 0)
1923             || ossl_ackm_is_ack_desired(txp->args.ackm, pn_space))
1924         && (ack = ossl_ackm_get_ack_frame(txp->args.ackm, pn_space)) != NULL) {
1925         WPACKET *wpkt = tx_helper_begin(h);
1926 
1927         if (wpkt == NULL)
1928             return 0;
1929 
1930         /* We do not currently support ECN */
1931         ack2 = *ack;
1932         ack2.ecn_present = 0;
1933 
1934         if (ossl_quic_wire_encode_frame_ack(wpkt,
1935                 txp->args.ack_delay_exponent,
1936                 &ack2)) {
1937             if (!tx_helper_commit(h))
1938                 return 0;
1939 
1940             tpkt->had_ack_frame = 1;
1941 
1942             if (ack->num_ack_ranges > 0)
1943                 tpkt->ackm_pkt.largest_acked = ack->ack_ranges[0].end;
1944 
1945             if (txp->ack_tx_cb != NULL)
1946                 txp->ack_tx_cb(&ack2, pn_space, txp->ack_tx_cb_arg);
1947         } else {
1948             tx_helper_rollback(h);
1949         }
1950     }
1951 
1952     /* CONNECTION_CLOSE Frames (Regenerate) */
1953     if (a->allow_conn_close && txp->want_conn_close && chosen_for_conn_close) {
1954         WPACKET *wpkt = tx_helper_begin(h);
1955         OSSL_QUIC_FRAME_CONN_CLOSE f, *pf = &txp->conn_close_frame;
1956         size_t l;
1957 
1958         if (wpkt == NULL)
1959             return 0;
1960 
1961         /*
1962          * Application CONNECTION_CLOSE frames may only be sent in the
1963          * Application PN space, as otherwise they may be sent before a
1964          * connection is authenticated and leak application data. Therefore, if
1965          * we need to send a CONNECTION_CLOSE frame in another PN space and were
1966          * given an application CONNECTION_CLOSE frame, convert it into a
1967          * transport CONNECTION_CLOSE frame, removing any sensitive application
1968          * data.
1969          *
1970          * RFC 9000 s. 10.2.3: "A CONNECTION_CLOSE of type 0x1d MUST be replaced
1971          * by a CONNECTION_CLOSE of type 0x1c when sending the frame in Initial
1972          * or Handshake packets. Otherwise, information about the application
1973          * state might be revealed. Endpoints MUST clear the value of the Reason
1974          * Phrase field and SHOULD use the APPLICATION_ERROR code when
1975          * converting to a CONNECTION_CLOSE of type 0x1c."
1976          */
1977         if (pn_space != QUIC_PN_SPACE_APP && pf->is_app) {
1978             pf = &f;
1979             pf->is_app = 0;
1980             pf->frame_type = 0;
1981             pf->error_code = OSSL_QUIC_ERR_APPLICATION_ERROR;
1982             pf->reason = NULL;
1983             pf->reason_len = 0;
1984         }
1985 
1986         if (ossl_quic_wire_encode_frame_conn_close(wpkt, pf)
1987             && WPACKET_get_total_written(wpkt, &l)
1988             && try_commit_conn_close(txp, l)) {
1989             if (!tx_helper_commit(h))
1990                 return 0;
1991 
1992             tpkt->had_conn_close = 1;
1993             *can_be_non_inflight = 0;
1994         } else {
1995             tx_helper_rollback(h);
1996         }
1997     }
1998 
1999     return 1;
2000 }
2001 
try_len(size_t space_left,size_t orig_len,size_t base_hdr_len,size_t lenbytes,uint64_t maxn,size_t * hdr_len,size_t * payload_len)2002 static int try_len(size_t space_left, size_t orig_len,
2003     size_t base_hdr_len, size_t lenbytes,
2004     uint64_t maxn, size_t *hdr_len, size_t *payload_len)
2005 {
2006     size_t n;
2007     size_t maxn_ = maxn > SIZE_MAX ? SIZE_MAX : (size_t)maxn;
2008 
2009     *hdr_len = base_hdr_len + lenbytes;
2010 
2011     if (orig_len == 0 && space_left >= *hdr_len) {
2012         *payload_len = 0;
2013         return 1;
2014     }
2015 
2016     n = orig_len;
2017     if (n > maxn_)
2018         n = maxn_;
2019     if (n + *hdr_len > space_left)
2020         n = (space_left >= *hdr_len) ? space_left - *hdr_len : 0;
2021 
2022     *payload_len = n;
2023     return n > 0;
2024 }
2025 
determine_len(size_t space_left,size_t orig_len,size_t base_hdr_len,uint64_t * hlen,uint64_t * len)2026 static int determine_len(size_t space_left, size_t orig_len,
2027     size_t base_hdr_len,
2028     uint64_t *hlen, uint64_t *len)
2029 {
2030     int ok = 0;
2031     size_t chosen_payload_len = 0;
2032     size_t chosen_hdr_len = 0;
2033     size_t payload_len[4], hdr_len[4];
2034     int i, valid[4] = { 0 };
2035 
2036     valid[0] = try_len(space_left, orig_len, base_hdr_len,
2037         1, OSSL_QUIC_VLINT_1B_MAX,
2038         &hdr_len[0], &payload_len[0]);
2039     valid[1] = try_len(space_left, orig_len, base_hdr_len,
2040         2, OSSL_QUIC_VLINT_2B_MAX,
2041         &hdr_len[1], &payload_len[1]);
2042     valid[2] = try_len(space_left, orig_len, base_hdr_len,
2043         4, OSSL_QUIC_VLINT_4B_MAX,
2044         &hdr_len[2], &payload_len[2]);
2045     valid[3] = try_len(space_left, orig_len, base_hdr_len,
2046         8, OSSL_QUIC_VLINT_8B_MAX,
2047         &hdr_len[3], &payload_len[3]);
2048 
2049     for (i = OSSL_NELEM(valid) - 1; i >= 0; --i)
2050         if (valid[i] && payload_len[i] >= chosen_payload_len) {
2051             chosen_payload_len = payload_len[i];
2052             chosen_hdr_len = hdr_len[i];
2053             ok = 1;
2054         }
2055 
2056     *hlen = chosen_hdr_len;
2057     *len = chosen_payload_len;
2058     return ok;
2059 }
2060 
2061 /*
2062  * Given a CRYPTO frame header with accurate chdr->len and a budget
2063  * (space_left), try to find the optimal value of chdr->len to fill as much of
2064  * the budget as possible. This is slightly hairy because larger values of
2065  * chdr->len cause larger encoded sizes of the length field of the frame, which
2066  * in turn mean less space available for payload data. We check all possible
2067  * encodings and choose the optimal encoding.
2068  */
determine_crypto_len(struct tx_helper * h,OSSL_QUIC_FRAME_CRYPTO * chdr,size_t space_left,uint64_t * hlen,uint64_t * len)2069 static int determine_crypto_len(struct tx_helper *h,
2070     OSSL_QUIC_FRAME_CRYPTO *chdr,
2071     size_t space_left,
2072     uint64_t *hlen,
2073     uint64_t *len)
2074 {
2075     size_t orig_len;
2076     size_t base_hdr_len; /* CRYPTO header length without length field */
2077 
2078     if (chdr->len > SIZE_MAX)
2079         return 0;
2080 
2081     orig_len = (size_t)chdr->len;
2082 
2083     chdr->len = 0;
2084     base_hdr_len = ossl_quic_wire_get_encoded_frame_len_crypto_hdr(chdr);
2085     chdr->len = orig_len;
2086     if (base_hdr_len == 0)
2087         return 0;
2088 
2089     --base_hdr_len;
2090 
2091     return determine_len(space_left, orig_len, base_hdr_len, hlen, len);
2092 }
2093 
determine_stream_len(struct tx_helper * h,OSSL_QUIC_FRAME_STREAM * shdr,size_t space_left,uint64_t * hlen,uint64_t * len)2094 static int determine_stream_len(struct tx_helper *h,
2095     OSSL_QUIC_FRAME_STREAM *shdr,
2096     size_t space_left,
2097     uint64_t *hlen,
2098     uint64_t *len)
2099 {
2100     size_t orig_len;
2101     size_t base_hdr_len; /* STREAM header length without length field */
2102 
2103     if (shdr->len > SIZE_MAX)
2104         return 0;
2105 
2106     orig_len = (size_t)shdr->len;
2107 
2108     shdr->len = 0;
2109     base_hdr_len = ossl_quic_wire_get_encoded_frame_len_stream_hdr(shdr);
2110     shdr->len = orig_len;
2111     if (base_hdr_len == 0)
2112         return 0;
2113 
2114     if (shdr->has_explicit_len)
2115         --base_hdr_len;
2116 
2117     return determine_len(space_left, orig_len, base_hdr_len, hlen, len);
2118 }
2119 
txp_generate_crypto_frames(OSSL_QUIC_TX_PACKETISER * txp,struct txp_pkt * pkt,int * have_ack_eliciting)2120 static int txp_generate_crypto_frames(OSSL_QUIC_TX_PACKETISER *txp,
2121     struct txp_pkt *pkt,
2122     int *have_ack_eliciting)
2123 {
2124     const uint32_t enc_level = pkt->h.enc_level;
2125     const uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
2126     QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
2127     struct tx_helper *h = &pkt->h;
2128     size_t num_stream_iovec;
2129     OSSL_QUIC_FRAME_STREAM shdr = { 0 };
2130     OSSL_QUIC_FRAME_CRYPTO chdr = { 0 };
2131     OSSL_QTX_IOVEC iov[2];
2132     uint64_t hdr_bytes;
2133     WPACKET *wpkt;
2134     QUIC_TXPIM_CHUNK chunk = { 0 };
2135     size_t i, space_left;
2136 
2137     for (i = 0;; ++i) {
2138         space_left = tx_helper_get_space_left(h);
2139 
2140         if (space_left < MIN_FRAME_SIZE_CRYPTO)
2141             return 1; /* no point trying */
2142 
2143         /* Do we have any CRYPTO data waiting? */
2144         num_stream_iovec = OSSL_NELEM(iov);
2145         if (!ossl_quic_sstream_get_stream_frame(txp->args.crypto[pn_space],
2146                 i, &shdr, iov,
2147                 &num_stream_iovec))
2148             return 1; /* nothing to do */
2149 
2150         /* Convert STREAM frame header to CRYPTO frame header */
2151         chdr.offset = shdr.offset;
2152         chdr.len = shdr.len;
2153 
2154         if (chdr.len == 0)
2155             return 1; /* nothing to do */
2156 
2157         /* Find best fit (header length, payload length) combination. */
2158         if (!determine_crypto_len(h, &chdr, space_left, &hdr_bytes,
2159                 &chdr.len))
2160             return 1; /* can't fit anything */
2161 
2162         /*
2163          * Truncate IOVs to match our chosen length.
2164          *
2165          * The length cannot be more than SIZE_MAX because this length comes
2166          * from our send stream buffer.
2167          */
2168         ossl_quic_sstream_adjust_iov((size_t)chdr.len, iov, num_stream_iovec);
2169 
2170         /*
2171          * Ensure we have enough iovecs allocated (1 for the header, up to 2 for
2172          * the stream data.)
2173          */
2174         if (!txp_el_ensure_iovec(&txp->el[enc_level], h->num_iovec + 3))
2175             return 0; /* alloc error */
2176 
2177         /* Encode the header. */
2178         wpkt = tx_helper_begin(h);
2179         if (wpkt == NULL)
2180             return 0; /* alloc error */
2181 
2182         if (!ossl_quic_wire_encode_frame_crypto_hdr(wpkt, &chdr)) {
2183             tx_helper_rollback(h);
2184             return 1; /* can't fit */
2185         }
2186 
2187         if (!tx_helper_commit(h))
2188             return 0; /* alloc error */
2189 
2190         /* Add payload iovecs to the helper (infallible). */
2191         for (i = 0; i < num_stream_iovec; ++i)
2192             tx_helper_append_iovec(h, iov[i].buf, iov[i].buf_len);
2193 
2194         *have_ack_eliciting = 1;
2195         tx_helper_unrestrict(h); /* no longer need PING */
2196 
2197         /* Log chunk to TXPIM. */
2198         chunk.stream_id = UINT64_MAX; /* crypto stream */
2199         chunk.start = chdr.offset;
2200         chunk.end = chdr.offset + chdr.len - 1;
2201         chunk.has_fin = 0; /* Crypto stream never ends */
2202         if (!ossl_quic_txpim_pkt_append_chunk(tpkt, &chunk))
2203             return 0; /* alloc error */
2204     }
2205 }
2206 
2207 struct chunk_info {
2208     OSSL_QUIC_FRAME_STREAM shdr;
2209     uint64_t orig_len;
2210     OSSL_QTX_IOVEC iov[2];
2211     size_t num_stream_iovec;
2212     int valid;
2213 };
2214 
txp_plan_stream_chunk(OSSL_QUIC_TX_PACKETISER * txp,struct tx_helper * h,QUIC_SSTREAM * sstream,QUIC_TXFC * stream_txfc,size_t skip,struct chunk_info * chunk,uint64_t consumed)2215 static int txp_plan_stream_chunk(OSSL_QUIC_TX_PACKETISER *txp,
2216     struct tx_helper *h,
2217     QUIC_SSTREAM *sstream,
2218     QUIC_TXFC *stream_txfc,
2219     size_t skip,
2220     struct chunk_info *chunk,
2221     uint64_t consumed)
2222 {
2223     uint64_t fc_credit, fc_swm, fc_limit;
2224 
2225     chunk->num_stream_iovec = OSSL_NELEM(chunk->iov);
2226     chunk->valid = ossl_quic_sstream_get_stream_frame(sstream, skip,
2227         &chunk->shdr,
2228         chunk->iov,
2229         &chunk->num_stream_iovec);
2230     if (!chunk->valid)
2231         return 1;
2232 
2233     if (!ossl_assert(chunk->shdr.len > 0 || chunk->shdr.is_fin))
2234         /* Should only have 0-length chunk if FIN */
2235         return 0;
2236 
2237     chunk->orig_len = chunk->shdr.len;
2238 
2239     /* Clamp according to connection and stream-level TXFC. */
2240     fc_credit = ossl_quic_txfc_get_credit(stream_txfc, consumed);
2241     fc_swm = ossl_quic_txfc_get_swm(stream_txfc);
2242     fc_limit = fc_swm + fc_credit;
2243 
2244     if (chunk->shdr.len > 0 && chunk->shdr.offset + chunk->shdr.len > fc_limit) {
2245         chunk->shdr.len = (fc_limit <= chunk->shdr.offset)
2246             ? 0
2247             : fc_limit - chunk->shdr.offset;
2248         chunk->shdr.is_fin = 0;
2249     }
2250 
2251     if (chunk->shdr.len == 0 && !chunk->shdr.is_fin) {
2252         /*
2253          * Nothing to do due to TXFC. Since SSTREAM returns chunks in ascending
2254          * order of offset we don't need to check any later chunks, so stop
2255          * iterating here.
2256          */
2257         chunk->valid = 0;
2258         return 1;
2259     }
2260 
2261     return 1;
2262 }
2263 
2264 /*
2265  * Returns 0 on fatal error (e.g. allocation failure), 1 on success.
2266  * *packet_full is set to 1 if there is no longer enough room for another STREAM
2267  * frame.
2268  */
txp_generate_stream_frames(OSSL_QUIC_TX_PACKETISER * txp,struct txp_pkt * pkt,uint64_t id,QUIC_SSTREAM * sstream,QUIC_TXFC * stream_txfc,QUIC_STREAM * next_stream,int * have_ack_eliciting,int * packet_full,uint64_t * new_credit_consumed,uint64_t conn_consumed)2269 static int txp_generate_stream_frames(OSSL_QUIC_TX_PACKETISER *txp,
2270     struct txp_pkt *pkt,
2271     uint64_t id,
2272     QUIC_SSTREAM *sstream,
2273     QUIC_TXFC *stream_txfc,
2274     QUIC_STREAM *next_stream,
2275     int *have_ack_eliciting,
2276     int *packet_full,
2277     uint64_t *new_credit_consumed,
2278     uint64_t conn_consumed)
2279 {
2280     int rc = 0;
2281     struct chunk_info chunks[2] = { 0 };
2282     const uint32_t enc_level = pkt->h.enc_level;
2283     QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
2284     struct tx_helper *h = &pkt->h;
2285     OSSL_QUIC_FRAME_STREAM *shdr;
2286     WPACKET *wpkt;
2287     QUIC_TXPIM_CHUNK chunk;
2288     size_t i, j, space_left;
2289     int can_fill_payload, use_explicit_len;
2290     int could_have_following_chunk;
2291     uint64_t orig_len;
2292     uint64_t hdr_len_implicit, payload_len_implicit;
2293     uint64_t hdr_len_explicit, payload_len_explicit;
2294     uint64_t fc_swm, fc_new_hwm;
2295 
2296     fc_swm = ossl_quic_txfc_get_swm(stream_txfc);
2297     fc_new_hwm = fc_swm;
2298 
2299     /*
2300      * Load the first two chunks if any offered by the send stream. We retrieve
2301      * the next chunk in advance so we can determine if we need to send any more
2302      * chunks from the same stream after this one, which is needed when
2303      * determining when we can use an implicit length in a STREAM frame.
2304      */
2305     for (i = 0; i < 2; ++i) {
2306         if (!txp_plan_stream_chunk(txp, h, sstream, stream_txfc, i, &chunks[i],
2307                 conn_consumed))
2308             goto err;
2309 
2310         if (i == 0 && !chunks[i].valid) {
2311             /* No chunks, nothing to do. */
2312             rc = 1;
2313             goto err;
2314         }
2315         chunks[i].shdr.stream_id = id;
2316     }
2317 
2318     for (i = 0;; ++i) {
2319         space_left = tx_helper_get_space_left(h);
2320 
2321         if (!chunks[i % 2].valid) {
2322             /* Out of chunks; we're done. */
2323             rc = 1;
2324             goto err;
2325         }
2326 
2327         if (space_left < MIN_FRAME_SIZE_STREAM) {
2328             *packet_full = 1;
2329             rc = 1;
2330             goto err;
2331         }
2332 
2333         if (!ossl_assert(!h->done_implicit))
2334             /*
2335              * Logic below should have ensured we didn't append an
2336              * implicit-length unless we filled the packet or didn't have
2337              * another stream to handle, so this should not be possible.
2338              */
2339             goto err;
2340 
2341         shdr = &chunks[i % 2].shdr;
2342         orig_len = chunks[i % 2].orig_len;
2343         if (i > 0)
2344             /* Load next chunk for lookahead. */
2345             if (!txp_plan_stream_chunk(txp, h, sstream, stream_txfc, i + 1,
2346                     &chunks[(i + 1) % 2], conn_consumed))
2347                 goto err;
2348 
2349         /*
2350          * Find best fit (header length, payload length) combination for if we
2351          * use an implicit length.
2352          */
2353         shdr->has_explicit_len = 0;
2354         hdr_len_implicit = payload_len_implicit = 0;
2355         if (!determine_stream_len(h, shdr, space_left,
2356                 &hdr_len_implicit, &payload_len_implicit)) {
2357             *packet_full = 1;
2358             rc = 1;
2359             goto err; /* can't fit anything */
2360         }
2361 
2362         /*
2363          * If there is a next stream, we don't use the implicit length so we can
2364          * add more STREAM frames after this one, unless there is enough data
2365          * for this STREAM frame to fill the packet.
2366          */
2367         can_fill_payload = (hdr_len_implicit + payload_len_implicit
2368             >= space_left);
2369 
2370         /*
2371          * Is there is a stream after this one, or another chunk pending
2372          * transmission in this stream?
2373          */
2374         could_have_following_chunk
2375             = (next_stream != NULL || chunks[(i + 1) % 2].valid);
2376 
2377         /* Choose between explicit or implicit length representations. */
2378         use_explicit_len = !((can_fill_payload || !could_have_following_chunk)
2379             && !pkt->force_pad);
2380 
2381         if (use_explicit_len) {
2382             /*
2383              * Find best fit (header length, payload length) combination for if
2384              * we use an explicit length.
2385              */
2386             shdr->has_explicit_len = 1;
2387             hdr_len_explicit = payload_len_explicit = 0;
2388             if (!determine_stream_len(h, shdr, space_left,
2389                     &hdr_len_explicit, &payload_len_explicit)) {
2390                 *packet_full = 1;
2391                 rc = 1;
2392                 goto err; /* can't fit anything */
2393             }
2394 
2395             shdr->len = payload_len_explicit;
2396         } else {
2397             *packet_full = 1;
2398             shdr->has_explicit_len = 0;
2399             shdr->len = payload_len_implicit;
2400         }
2401 
2402         /* If this is a FIN, don't keep filling the packet with more FINs. */
2403         if (shdr->is_fin)
2404             chunks[(i + 1) % 2].valid = 0;
2405 
2406         /*
2407          * We are now committed to our length (shdr->len can't change).
2408          * If we truncated the chunk, clear the FIN bit.
2409          */
2410         if (shdr->len < orig_len)
2411             shdr->is_fin = 0;
2412 
2413         /* Truncate IOVs to match our chosen length. */
2414         ossl_quic_sstream_adjust_iov((size_t)shdr->len, chunks[i % 2].iov,
2415             chunks[i % 2].num_stream_iovec);
2416 
2417         /*
2418          * Ensure we have enough iovecs allocated (1 for the header, up to 2 for
2419          * the stream data.)
2420          */
2421         if (!txp_el_ensure_iovec(&txp->el[enc_level], h->num_iovec + 3))
2422             goto err; /* alloc error */
2423 
2424         /* Encode the header. */
2425         wpkt = tx_helper_begin(h);
2426         if (wpkt == NULL)
2427             goto err; /* alloc error */
2428 
2429         if (!ossl_assert(ossl_quic_wire_encode_frame_stream_hdr(wpkt, shdr))) {
2430             /* (Should not be possible.) */
2431             tx_helper_rollback(h);
2432             *packet_full = 1;
2433             rc = 1;
2434             goto err; /* can't fit */
2435         }
2436 
2437         if (!tx_helper_commit(h))
2438             goto err; /* alloc error */
2439 
2440         /* Add payload iovecs to the helper (infallible). */
2441         for (j = 0; j < chunks[i % 2].num_stream_iovec; ++j)
2442             tx_helper_append_iovec(h, chunks[i % 2].iov[j].buf,
2443                 chunks[i % 2].iov[j].buf_len);
2444 
2445         *have_ack_eliciting = 1;
2446         tx_helper_unrestrict(h); /* no longer need PING */
2447         if (!shdr->has_explicit_len)
2448             h->done_implicit = 1;
2449 
2450         /* Log new TXFC credit which was consumed. */
2451         if (shdr->len > 0 && shdr->offset + shdr->len > fc_new_hwm)
2452             fc_new_hwm = shdr->offset + shdr->len;
2453 
2454         /* Log chunk to TXPIM. */
2455         chunk.stream_id = shdr->stream_id;
2456         chunk.start = shdr->offset;
2457         chunk.end = shdr->offset + shdr->len - 1;
2458         chunk.has_fin = shdr->is_fin;
2459         chunk.has_stop_sending = 0;
2460         chunk.has_reset_stream = 0;
2461         if (!ossl_quic_txpim_pkt_append_chunk(tpkt, &chunk))
2462             goto err; /* alloc error */
2463 
2464         if (shdr->len < orig_len) {
2465             /*
2466              * If we did not serialize all of this chunk we definitely do not
2467              * want to try the next chunk
2468              */
2469             rc = 1;
2470             goto err;
2471         }
2472     }
2473 
2474 err:
2475     *new_credit_consumed = fc_new_hwm - fc_swm;
2476     return rc;
2477 }
2478 
txp_enlink_tmp(QUIC_STREAM ** tmp_head,QUIC_STREAM * stream)2479 static void txp_enlink_tmp(QUIC_STREAM **tmp_head, QUIC_STREAM *stream)
2480 {
2481     stream->txp_next = *tmp_head;
2482     *tmp_head = stream;
2483 }
2484 
txp_generate_stream_related(OSSL_QUIC_TX_PACKETISER * txp,struct txp_pkt * pkt,int * have_ack_eliciting,QUIC_STREAM ** tmp_head)2485 static int txp_generate_stream_related(OSSL_QUIC_TX_PACKETISER *txp,
2486     struct txp_pkt *pkt,
2487     int *have_ack_eliciting,
2488     QUIC_STREAM **tmp_head)
2489 {
2490     QUIC_STREAM_ITER it;
2491     WPACKET *wpkt;
2492     uint64_t cwm;
2493     QUIC_STREAM *stream, *snext;
2494     struct tx_helper *h = &pkt->h;
2495     uint64_t conn_consumed = 0;
2496 
2497     for (ossl_quic_stream_iter_init(&it, txp->args.qsm, 1);
2498         it.stream != NULL;) {
2499 
2500         stream = it.stream;
2501         ossl_quic_stream_iter_next(&it);
2502         snext = it.stream;
2503 
2504         stream->txp_sent_fc = 0;
2505         stream->txp_sent_stop_sending = 0;
2506         stream->txp_sent_reset_stream = 0;
2507         stream->txp_blocked = 0;
2508         stream->txp_txfc_new_credit_consumed = 0;
2509 
2510         /* Stream Abort Frames (STOP_SENDING, RESET_STREAM) */
2511         if (stream->want_stop_sending) {
2512             OSSL_QUIC_FRAME_STOP_SENDING f;
2513 
2514             wpkt = tx_helper_begin(h);
2515             if (wpkt == NULL)
2516                 return 0; /* alloc error */
2517 
2518             f.stream_id = stream->id;
2519             f.app_error_code = stream->stop_sending_aec;
2520             if (!ossl_quic_wire_encode_frame_stop_sending(wpkt, &f)) {
2521                 tx_helper_rollback(h); /* can't fit */
2522                 txp_enlink_tmp(tmp_head, stream);
2523                 break;
2524             }
2525 
2526             if (!tx_helper_commit(h))
2527                 return 0; /* alloc error */
2528 
2529             *have_ack_eliciting = 1;
2530             tx_helper_unrestrict(h); /* no longer need PING */
2531             stream->txp_sent_stop_sending = 1;
2532         }
2533 
2534         if (stream->want_reset_stream) {
2535             OSSL_QUIC_FRAME_RESET_STREAM f;
2536 
2537             if (!ossl_assert(stream->send_state == QUIC_SSTREAM_STATE_RESET_SENT))
2538                 return 0;
2539 
2540             wpkt = tx_helper_begin(h);
2541             if (wpkt == NULL)
2542                 return 0; /* alloc error */
2543 
2544             f.stream_id = stream->id;
2545             f.app_error_code = stream->reset_stream_aec;
2546             if (!ossl_quic_stream_send_get_final_size(stream, &f.final_size))
2547                 return 0; /* should not be possible */
2548 
2549             if (!ossl_quic_wire_encode_frame_reset_stream(wpkt, &f)) {
2550                 tx_helper_rollback(h); /* can't fit */
2551                 txp_enlink_tmp(tmp_head, stream);
2552                 break;
2553             }
2554 
2555             if (!tx_helper_commit(h))
2556                 return 0; /* alloc error */
2557 
2558             *have_ack_eliciting = 1;
2559             tx_helper_unrestrict(h); /* no longer need PING */
2560             stream->txp_sent_reset_stream = 1;
2561 
2562             /*
2563              * The final size of the stream as indicated by RESET_STREAM is used
2564              * to ensure a consistent view of flow control state by both
2565              * parties; if we happen to send a RESET_STREAM that consumes more
2566              * flow control credit, make sure we account for that.
2567              */
2568             if (!ossl_assert(f.final_size <= ossl_quic_txfc_get_swm(&stream->txfc)))
2569                 return 0;
2570 
2571             stream->txp_txfc_new_credit_consumed
2572                 = f.final_size - ossl_quic_txfc_get_swm(&stream->txfc);
2573         }
2574 
2575         /*
2576          * Stream Flow Control Frames (MAX_STREAM_DATA)
2577          *
2578          * RFC 9000 s. 13.3: "An endpoint SHOULD stop sending MAX_STREAM_DATA
2579          * frames when the receiving part of the stream enters a "Size Known" or
2580          * "Reset Recvd" state." -- In practice, RECV is the only state
2581          * in which it makes sense to generate more MAX_STREAM_DATA frames.
2582          */
2583         if (stream->recv_state == QUIC_RSTREAM_STATE_RECV
2584             && (stream->want_max_stream_data
2585                 || ossl_quic_rxfc_has_cwm_changed(&stream->rxfc, 0))) {
2586 
2587             wpkt = tx_helper_begin(h);
2588             if (wpkt == NULL)
2589                 return 0; /* alloc error */
2590 
2591             cwm = ossl_quic_rxfc_get_cwm(&stream->rxfc);
2592 
2593             if (!ossl_quic_wire_encode_frame_max_stream_data(wpkt, stream->id,
2594                     cwm)) {
2595                 tx_helper_rollback(h); /* can't fit */
2596                 txp_enlink_tmp(tmp_head, stream);
2597                 break;
2598             }
2599 
2600             if (!tx_helper_commit(h))
2601                 return 0; /* alloc error */
2602 
2603             *have_ack_eliciting = 1;
2604             tx_helper_unrestrict(h); /* no longer need PING */
2605             stream->txp_sent_fc = 1;
2606         }
2607 
2608         /*
2609          * Stream Data Frames (STREAM)
2610          *
2611          * RFC 9000 s. 3.3: A sender MUST NOT send a STREAM [...] frame for a
2612          * stream in the "Reset Sent" state [or any terminal state]. We don't
2613          * send any more STREAM frames if we are sending, have sent, or are
2614          * planning to send, RESET_STREAM. The other terminal state is Data
2615          * Recvd, but txp_generate_stream_frames() is guaranteed to generate
2616          * nothing in this case.
2617          */
2618         if (ossl_quic_stream_has_send_buffer(stream)
2619             && !ossl_quic_stream_send_is_reset(stream)) {
2620             int packet_full = 0;
2621 
2622             if (!ossl_assert(!stream->want_reset_stream))
2623                 return 0;
2624 
2625             if (!txp_generate_stream_frames(txp, pkt,
2626                     stream->id, stream->sstream,
2627                     &stream->txfc,
2628                     snext,
2629                     have_ack_eliciting,
2630                     &packet_full,
2631                     &stream->txp_txfc_new_credit_consumed,
2632                     conn_consumed)) {
2633                 /* Fatal error (allocation, etc.) */
2634                 txp_enlink_tmp(tmp_head, stream);
2635                 return 0;
2636             }
2637             conn_consumed += stream->txp_txfc_new_credit_consumed;
2638 
2639             if (packet_full) {
2640                 txp_enlink_tmp(tmp_head, stream);
2641                 break;
2642             }
2643         }
2644 
2645         txp_enlink_tmp(tmp_head, stream);
2646     }
2647 
2648     return 1;
2649 }
2650 
txp_generate_for_el(OSSL_QUIC_TX_PACKETISER * txp,struct txp_pkt * pkt,int chosen_for_conn_close)2651 static int txp_generate_for_el(OSSL_QUIC_TX_PACKETISER *txp,
2652     struct txp_pkt *pkt,
2653     int chosen_for_conn_close)
2654 {
2655     int rc = TXP_ERR_SUCCESS;
2656     const uint32_t enc_level = pkt->h.enc_level;
2657     const uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
2658     int have_ack_eliciting = 0, done_pre_token = 0;
2659     const struct archetype_data a = pkt->geom.adata;
2660     /*
2661      * Cleared if we encode any non-ACK-eliciting frame type which rules out the
2662      * packet being a non-inflight frame. This means any non-ACK ACK-eliciting
2663      * frame, even PADDING frames. ACK eliciting frames always cause a packet to
2664      * become ineligible for non-inflight treatment so it is not necessary to
2665      * clear this in cases where have_ack_eliciting is set, as it is ignored in
2666      * that case.
2667      */
2668     int can_be_non_inflight = 1;
2669     QUIC_CFQ_ITEM *cfq_item;
2670     QUIC_TXPIM_PKT *tpkt = NULL;
2671     struct tx_helper *h = &pkt->h;
2672 
2673     /* Maximum PN reached? */
2674     if (!ossl_quic_pn_valid(txp->next_pn[pn_space]))
2675         goto fatal_err;
2676 
2677     if (!ossl_assert(pkt->tpkt == NULL))
2678         goto fatal_err;
2679 
2680     if ((pkt->tpkt = tpkt = ossl_quic_txpim_pkt_alloc(txp->args.txpim)) == NULL)
2681         goto fatal_err;
2682 
2683     /*
2684      * Frame Serialization
2685      * ===================
2686      *
2687      * We now serialize frames into the packet in descending order of priority.
2688      */
2689 
2690     /* HANDSHAKE_DONE (Regenerate) */
2691     if (a.allow_handshake_done && txp->want_handshake_done
2692         && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_HANDSHAKE_DONE) {
2693         WPACKET *wpkt = tx_helper_begin(h);
2694 
2695         if (wpkt == NULL)
2696             goto fatal_err;
2697 
2698         if (ossl_quic_wire_encode_frame_handshake_done(wpkt)) {
2699             tpkt->had_handshake_done_frame = 1;
2700             have_ack_eliciting = 1;
2701 
2702             if (!tx_helper_commit(h))
2703                 goto fatal_err;
2704 
2705             tx_helper_unrestrict(h); /* no longer need PING */
2706         } else {
2707             tx_helper_rollback(h);
2708         }
2709     }
2710 
2711     /* MAX_DATA (Regenerate) */
2712     if (a.allow_conn_fc
2713         && (txp->want_max_data
2714             || ossl_quic_rxfc_has_cwm_changed(txp->args.conn_rxfc, 0))
2715         && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_MAX_DATA) {
2716         WPACKET *wpkt = tx_helper_begin(h);
2717         uint64_t cwm = ossl_quic_rxfc_get_cwm(txp->args.conn_rxfc);
2718 
2719         if (wpkt == NULL)
2720             goto fatal_err;
2721 
2722         if (ossl_quic_wire_encode_frame_max_data(wpkt, cwm)) {
2723             tpkt->had_max_data_frame = 1;
2724             have_ack_eliciting = 1;
2725 
2726             if (!tx_helper_commit(h))
2727                 goto fatal_err;
2728 
2729             tx_helper_unrestrict(h); /* no longer need PING */
2730         } else {
2731             tx_helper_rollback(h);
2732         }
2733     }
2734 
2735     /* MAX_STREAMS_BIDI (Regenerate) */
2736     if (a.allow_conn_fc
2737         && (txp->want_max_streams_bidi
2738             || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_bidi_rxfc, 0))
2739         && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_MAX_STREAMS_BIDI) {
2740         WPACKET *wpkt = tx_helper_begin(h);
2741         uint64_t max_streams
2742             = ossl_quic_rxfc_get_cwm(txp->args.max_streams_bidi_rxfc);
2743 
2744         if (wpkt == NULL)
2745             goto fatal_err;
2746 
2747         if (ossl_quic_wire_encode_frame_max_streams(wpkt, /*is_uni=*/0,
2748                 max_streams)) {
2749             tpkt->had_max_streams_bidi_frame = 1;
2750             have_ack_eliciting = 1;
2751 
2752             if (!tx_helper_commit(h))
2753                 goto fatal_err;
2754 
2755             tx_helper_unrestrict(h); /* no longer need PING */
2756         } else {
2757             tx_helper_rollback(h);
2758         }
2759     }
2760 
2761     /* MAX_STREAMS_UNI (Regenerate) */
2762     if (a.allow_conn_fc
2763         && (txp->want_max_streams_uni
2764             || ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_uni_rxfc, 0))
2765         && tx_helper_get_space_left(h) >= MIN_FRAME_SIZE_MAX_STREAMS_UNI) {
2766         WPACKET *wpkt = tx_helper_begin(h);
2767         uint64_t max_streams
2768             = ossl_quic_rxfc_get_cwm(txp->args.max_streams_uni_rxfc);
2769 
2770         if (wpkt == NULL)
2771             goto fatal_err;
2772 
2773         if (ossl_quic_wire_encode_frame_max_streams(wpkt, /*is_uni=*/1,
2774                 max_streams)) {
2775             tpkt->had_max_streams_uni_frame = 1;
2776             have_ack_eliciting = 1;
2777 
2778             if (!tx_helper_commit(h))
2779                 goto fatal_err;
2780 
2781             tx_helper_unrestrict(h); /* no longer need PING */
2782         } else {
2783             tx_helper_rollback(h);
2784         }
2785     }
2786 
2787     /* GCR Frames */
2788     for (cfq_item = ossl_quic_cfq_get_priority_head(txp->args.cfq, pn_space);
2789         cfq_item != NULL;
2790         cfq_item = ossl_quic_cfq_item_get_priority_next(cfq_item, pn_space)) {
2791         uint64_t frame_type = ossl_quic_cfq_item_get_frame_type(cfq_item);
2792         const unsigned char *encoded = ossl_quic_cfq_item_get_encoded(cfq_item);
2793         size_t encoded_len = ossl_quic_cfq_item_get_encoded_len(cfq_item);
2794 
2795         switch (frame_type) {
2796         case OSSL_QUIC_FRAME_TYPE_NEW_CONN_ID:
2797             if (!a.allow_new_conn_id)
2798                 continue;
2799             break;
2800         case OSSL_QUIC_FRAME_TYPE_RETIRE_CONN_ID:
2801             if (!a.allow_retire_conn_id)
2802                 continue;
2803             break;
2804         case OSSL_QUIC_FRAME_TYPE_NEW_TOKEN:
2805             if (!a.allow_new_token)
2806                 continue;
2807 
2808             /*
2809              * NEW_TOKEN frames are handled via GCR, but some
2810              * Regenerate-strategy frames should come before them (namely
2811              * ACK, CONNECTION_CLOSE, PATH_CHALLENGE and PATH_RESPONSE). If
2812              * we find a NEW_TOKEN frame, do these now. If there are no
2813              * NEW_TOKEN frames in the GCR queue we will handle these below.
2814              */
2815             if (!done_pre_token)
2816                 if (txp_generate_pre_token(txp, pkt,
2817                         chosen_for_conn_close,
2818                         &can_be_non_inflight))
2819                     done_pre_token = 1;
2820 
2821             break;
2822         case OSSL_QUIC_FRAME_TYPE_PATH_RESPONSE:
2823             if (!a.allow_path_response)
2824                 continue;
2825 
2826             /*
2827              * RFC 9000 s. 8.2.2: An endpoint MUST expand datagrams that
2828              * contain a PATH_RESPONSE frame to at least the smallest
2829              * allowed maximum datagram size of 1200 bytes.
2830              */
2831             pkt->force_pad = 1;
2832             break;
2833         default:
2834             if (!a.allow_cfq_other)
2835                 continue;
2836             break;
2837         }
2838 
2839         /*
2840          * If the frame is too big, don't try to schedule any more GCR frames in
2841          * this packet rather than sending subsequent ones out of order.
2842          */
2843         if (encoded_len > tx_helper_get_space_left(h))
2844             break;
2845 
2846         if (!tx_helper_append_iovec(h, encoded, encoded_len))
2847             goto fatal_err;
2848 
2849         ossl_quic_txpim_pkt_add_cfq_item(tpkt, cfq_item);
2850 
2851         if (ossl_quic_frame_type_is_ack_eliciting(frame_type)) {
2852             have_ack_eliciting = 1;
2853             tx_helper_unrestrict(h); /* no longer need PING */
2854         }
2855     }
2856 
2857     /*
2858      * If we didn't generate ACK, CONNECTION_CLOSE, PATH_CHALLENGE or
2859      * PATH_RESPONSE (as desired) before, do so now.
2860      */
2861     if (!done_pre_token)
2862         if (txp_generate_pre_token(txp, pkt,
2863                 chosen_for_conn_close,
2864                 &can_be_non_inflight))
2865             done_pre_token = 1;
2866 
2867     /* CRYPTO Frames */
2868     if (a.allow_crypto)
2869         if (!txp_generate_crypto_frames(txp, pkt, &have_ack_eliciting))
2870             goto fatal_err;
2871 
2872     /* Stream-specific frames */
2873     if (a.allow_stream_rel && txp->handshake_complete)
2874         if (!txp_generate_stream_related(txp, pkt,
2875                 &have_ack_eliciting,
2876                 &pkt->stream_head))
2877             goto fatal_err;
2878 
2879     /* PING */
2880     tx_helper_unrestrict(h);
2881 
2882     if (!have_ack_eliciting && txp_need_ping(txp, pn_space, &a)) {
2883         WPACKET *wpkt;
2884 
2885         assert(h->reserve > 0);
2886         wpkt = tx_helper_begin(h);
2887         if (wpkt == NULL)
2888             goto fatal_err;
2889 
2890         if (!ossl_quic_wire_encode_frame_ping(wpkt)
2891             || !tx_helper_commit(h))
2892             /*
2893              * We treat a request to be ACK-eliciting as a requirement, so this
2894              * is an error.
2895              */
2896             goto fatal_err;
2897 
2898         have_ack_eliciting = 1;
2899     }
2900 
2901     /* PADDING is added by ossl_quic_tx_packetiser_generate(). */
2902 
2903     /*
2904      * ACKM Data
2905      * =========
2906      */
2907     if (have_ack_eliciting)
2908         can_be_non_inflight = 0;
2909 
2910     /* ACKM Data */
2911     tpkt->ackm_pkt.num_bytes = h->bytes_appended + pkt->geom.pkt_overhead;
2912     tpkt->ackm_pkt.pkt_num = txp->next_pn[pn_space];
2913     /* largest_acked is set in txp_generate_pre_token */
2914     tpkt->ackm_pkt.pkt_space = pn_space;
2915     tpkt->ackm_pkt.is_inflight = !can_be_non_inflight;
2916     tpkt->ackm_pkt.is_ack_eliciting = have_ack_eliciting;
2917     tpkt->ackm_pkt.is_pto_probe = 0;
2918     tpkt->ackm_pkt.is_mtu_probe = 0;
2919     tpkt->ackm_pkt.time = txp->args.now(txp->args.now_arg);
2920     tpkt->pkt_type = pkt->phdr.type;
2921 
2922     /* Done. */
2923     return rc;
2924 
2925 fatal_err:
2926     /*
2927      * Handler for fatal errors, i.e. errors causing us to abort the entire
2928      * packet rather than just one frame. Examples of such errors include
2929      * allocation errors.
2930      */
2931     if (tpkt != NULL) {
2932         ossl_quic_txpim_pkt_release(txp->args.txpim, tpkt);
2933         pkt->tpkt = NULL;
2934     }
2935     return TXP_ERR_INTERNAL;
2936 }
2937 
2938 /*
2939  * Commits and queues a packet for transmission. There is no backing out after
2940  * this.
2941  *
2942  * This:
2943  *
2944  *   - Sends the packet to the QTX for encryption and transmission;
2945  *
2946  *   - Records the packet as having been transmitted in FIFM. ACKM is informed,
2947  *     etc. and the TXPIM record is filed.
2948  *
2949  *   - Informs various subsystems of frames that were sent and clears frame
2950  *     wanted flags so that we do not generate the same frames again.
2951  *
2952  * Assumptions:
2953  *
2954  *   - pkt is a txp_pkt for the correct EL;
2955  *
2956  *   - pkt->tpkt is valid;
2957  *
2958  *   - pkt->tpkt->ackm_pkt has been fully filled in;
2959  *
2960  *   - Stream chunk records have been appended to pkt->tpkt for STREAM and
2961  *     CRYPTO frames, but not for RESET_STREAM or STOP_SENDING frames;
2962  *
2963  *   - The chosen stream list for the packet can be fully walked from
2964  *     pkt->stream_head using stream->txp_next;
2965  *
2966  *   - pkt->has_ack_eliciting is set correctly.
2967  *
2968  */
txp_pkt_commit(OSSL_QUIC_TX_PACKETISER * txp,struct txp_pkt * pkt,uint32_t archetype,int * txpim_pkt_reffed)2969 static int txp_pkt_commit(OSSL_QUIC_TX_PACKETISER *txp,
2970     struct txp_pkt *pkt,
2971     uint32_t archetype,
2972     int *txpim_pkt_reffed)
2973 {
2974     int rc = 1;
2975     uint32_t enc_level = pkt->h.enc_level;
2976     uint32_t pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
2977     QUIC_TXPIM_PKT *tpkt = pkt->tpkt;
2978     QUIC_STREAM *stream;
2979     OSSL_QTX_PKT txpkt;
2980     struct archetype_data a;
2981 
2982     *txpim_pkt_reffed = 0;
2983 
2984     /* Cannot send a packet with an empty payload. */
2985     if (pkt->h.bytes_appended == 0)
2986         return 0;
2987 
2988     if (!txp_get_archetype_data(enc_level, archetype, &a))
2989         return 0;
2990 
2991     /* Packet Information for QTX */
2992     txpkt.hdr = &pkt->phdr;
2993     txpkt.iovec = txp->el[enc_level].iovec;
2994     txpkt.num_iovec = pkt->h.num_iovec;
2995     txpkt.local = NULL;
2996     txpkt.peer = BIO_ADDR_family(&txp->args.peer) == AF_UNSPEC
2997         ? NULL
2998         : &txp->args.peer;
2999     txpkt.pn = txp->next_pn[pn_space];
3000     txpkt.flags = OSSL_QTX_PKT_FLAG_COALESCE; /* always try to coalesce */
3001 
3002     /* Generate TXPIM chunks representing STOP_SENDING and RESET_STREAM frames. */
3003     for (stream = pkt->stream_head; stream != NULL; stream = stream->txp_next)
3004         if (stream->txp_sent_stop_sending || stream->txp_sent_reset_stream) {
3005             /* Log STOP_SENDING/RESET_STREAM chunk to TXPIM. */
3006             QUIC_TXPIM_CHUNK chunk;
3007 
3008             chunk.stream_id = stream->id;
3009             chunk.start = UINT64_MAX;
3010             chunk.end = 0;
3011             chunk.has_fin = 0;
3012             chunk.has_stop_sending = stream->txp_sent_stop_sending;
3013             chunk.has_reset_stream = stream->txp_sent_reset_stream;
3014             if (!ossl_quic_txpim_pkt_append_chunk(tpkt, &chunk))
3015                 return 0; /* alloc error */
3016         }
3017 
3018     /* Dispatch to FIFD. */
3019     if (!ossl_quic_fifd_pkt_commit(&txp->fifd, tpkt))
3020         return 0;
3021 
3022     /*
3023      * Transmission and Post-Packet Generation Bookkeeping
3024      * ===================================================
3025      *
3026      * No backing out anymore - at this point the ACKM has recorded the packet
3027      * as having been sent, so we need to increment our next PN counter, or
3028      * the ACKM will complain when we try to record a duplicate packet with
3029      * the same PN later. At this point actually sending the packet may still
3030      * fail. In this unlikely event it will simply be handled as though it
3031      * were a lost packet.
3032      */
3033     ++txp->next_pn[pn_space];
3034     *txpim_pkt_reffed = 1;
3035 
3036     /* Send the packet. */
3037     if (!ossl_qtx_write_pkt(txp->args.qtx, &txpkt))
3038         return 0;
3039 
3040     /*
3041      * Record FC and stream abort frames as sent; deactivate streams which no
3042      * longer have anything to do.
3043      */
3044     for (stream = pkt->stream_head; stream != NULL; stream = stream->txp_next) {
3045         if (stream->txp_sent_fc) {
3046             stream->want_max_stream_data = 0;
3047             ossl_quic_rxfc_has_cwm_changed(&stream->rxfc, 1);
3048         }
3049 
3050         if (stream->txp_sent_stop_sending)
3051             stream->want_stop_sending = 0;
3052 
3053         if (stream->txp_sent_reset_stream)
3054             stream->want_reset_stream = 0;
3055 
3056         if (stream->txp_txfc_new_credit_consumed > 0) {
3057             if (!ossl_assert(ossl_quic_txfc_consume_credit(&stream->txfc,
3058                     stream->txp_txfc_new_credit_consumed)))
3059                 /*
3060                  * Should not be possible, but we should continue with our
3061                  * bookkeeping as we have already committed the packet to the
3062                  * FIFD. Just change the value we return.
3063                  */
3064                 rc = 0;
3065 
3066             stream->txp_txfc_new_credit_consumed = 0;
3067         }
3068 
3069         /*
3070          * If we no longer need to generate any flow control (MAX_STREAM_DATA),
3071          * STOP_SENDING or RESET_STREAM frames, nor any STREAM frames (because
3072          * the stream is drained of data or TXFC-blocked), we can mark the
3073          * stream as inactive.
3074          */
3075         ossl_quic_stream_map_update_state(txp->args.qsm, stream);
3076 
3077         if (ossl_quic_stream_has_send_buffer(stream)
3078             && !ossl_quic_sstream_has_pending(stream->sstream)
3079             && ossl_quic_sstream_get_final_size(stream->sstream, NULL))
3080             /*
3081              * Transition to DATA_SENT if stream has a final size and we have
3082              * sent all data.
3083              */
3084             ossl_quic_stream_map_notify_all_data_sent(txp->args.qsm, stream);
3085     }
3086 
3087     /* We have now sent the packet, so update state accordingly. */
3088     if (tpkt->ackm_pkt.is_ack_eliciting)
3089         txp->force_ack_eliciting &= ~(1UL << pn_space);
3090 
3091     if (tpkt->had_handshake_done_frame)
3092         txp->want_handshake_done = 0;
3093 
3094     if (tpkt->had_max_data_frame) {
3095         txp->want_max_data = 0;
3096         ossl_quic_rxfc_has_cwm_changed(txp->args.conn_rxfc, 1);
3097     }
3098 
3099     if (tpkt->had_max_streams_bidi_frame) {
3100         txp->want_max_streams_bidi = 0;
3101         ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_bidi_rxfc, 1);
3102     }
3103 
3104     if (tpkt->had_max_streams_uni_frame) {
3105         txp->want_max_streams_uni = 0;
3106         ossl_quic_rxfc_has_cwm_changed(txp->args.max_streams_uni_rxfc, 1);
3107     }
3108 
3109     if (tpkt->had_ack_frame)
3110         txp->want_ack &= ~(1UL << pn_space);
3111 
3112     if (tpkt->had_conn_close)
3113         txp->want_conn_close = 0;
3114 
3115     /*
3116      * Decrement probe request counts if we have sent a packet that meets
3117      * the requirement of a probe, namely being ACK-eliciting.
3118      */
3119     if (tpkt->ackm_pkt.is_ack_eliciting) {
3120         OSSL_ACKM_PROBE_INFO *probe_info
3121             = ossl_ackm_get0_probe_request(txp->args.ackm);
3122 
3123         if (enc_level == QUIC_ENC_LEVEL_INITIAL
3124             && probe_info->anti_deadlock_initial > 0)
3125             --probe_info->anti_deadlock_initial;
3126 
3127         if (enc_level == QUIC_ENC_LEVEL_HANDSHAKE
3128             && probe_info->anti_deadlock_handshake > 0)
3129             --probe_info->anti_deadlock_handshake;
3130 
3131         if (a.allow_force_ack_eliciting /* (i.e., not for 0-RTT) */
3132             && probe_info->pto[pn_space] > 0)
3133             --probe_info->pto[pn_space];
3134     }
3135 
3136     return rc;
3137 }
3138 
3139 /* Ensure the iovec array is at least num elements long. */
txp_el_ensure_iovec(struct txp_el * el,size_t num)3140 static int txp_el_ensure_iovec(struct txp_el *el, size_t num)
3141 {
3142     OSSL_QTX_IOVEC *iovec;
3143 
3144     if (el->alloc_iovec >= num)
3145         return 1;
3146 
3147     num = el->alloc_iovec != 0 ? el->alloc_iovec * 2 : 8;
3148 
3149     iovec = OPENSSL_realloc(el->iovec, sizeof(OSSL_QTX_IOVEC) * num);
3150     if (iovec == NULL)
3151         return 0;
3152 
3153     el->iovec = iovec;
3154     el->alloc_iovec = num;
3155     return 1;
3156 }
3157 
ossl_quic_tx_packetiser_schedule_conn_close(OSSL_QUIC_TX_PACKETISER * txp,const OSSL_QUIC_FRAME_CONN_CLOSE * f)3158 int ossl_quic_tx_packetiser_schedule_conn_close(OSSL_QUIC_TX_PACKETISER *txp,
3159     const OSSL_QUIC_FRAME_CONN_CLOSE *f)
3160 {
3161     char *reason = NULL;
3162     size_t reason_len = f->reason_len;
3163     size_t max_reason_len = txp_get_mdpl(txp) / 2;
3164 
3165     if (txp->want_conn_close)
3166         return 0;
3167 
3168     /*
3169      * Arbitrarily limit the length of the reason length string to half of the
3170      * MDPL.
3171      */
3172     if (reason_len > max_reason_len)
3173         reason_len = max_reason_len;
3174 
3175     if (reason_len > 0) {
3176         reason = OPENSSL_memdup(f->reason, reason_len);
3177         if (reason == NULL)
3178             return 0;
3179     }
3180 
3181     txp->conn_close_frame = *f;
3182     txp->conn_close_frame.reason = reason;
3183     txp->conn_close_frame.reason_len = reason_len;
3184     txp->want_conn_close = 1;
3185     return 1;
3186 }
3187 
ossl_quic_tx_packetiser_set_msg_callback(OSSL_QUIC_TX_PACKETISER * txp,ossl_msg_cb msg_callback,SSL * msg_callback_ssl)3188 void ossl_quic_tx_packetiser_set_msg_callback(OSSL_QUIC_TX_PACKETISER *txp,
3189     ossl_msg_cb msg_callback,
3190     SSL *msg_callback_ssl)
3191 {
3192     txp->msg_callback = msg_callback;
3193     txp->msg_callback_ssl = msg_callback_ssl;
3194 }
3195 
ossl_quic_tx_packetiser_set_msg_callback_arg(OSSL_QUIC_TX_PACKETISER * txp,void * msg_callback_arg)3196 void ossl_quic_tx_packetiser_set_msg_callback_arg(OSSL_QUIC_TX_PACKETISER *txp,
3197     void *msg_callback_arg)
3198 {
3199     txp->msg_callback_arg = msg_callback_arg;
3200 }
3201 
ossl_quic_tx_packetiser_get_next_pn(OSSL_QUIC_TX_PACKETISER * txp,uint32_t pn_space)3202 QUIC_PN ossl_quic_tx_packetiser_get_next_pn(OSSL_QUIC_TX_PACKETISER *txp,
3203     uint32_t pn_space)
3204 {
3205     if (pn_space >= QUIC_PN_SPACE_NUM)
3206         return UINT64_MAX;
3207 
3208     return txp->next_pn[pn_space];
3209 }
3210 
ossl_quic_tx_packetiser_get_deadline(OSSL_QUIC_TX_PACKETISER * txp)3211 OSSL_TIME ossl_quic_tx_packetiser_get_deadline(OSSL_QUIC_TX_PACKETISER *txp)
3212 {
3213     /*
3214      * TXP-specific deadline computations which rely on TXP innards. This is in
3215      * turn relied on by the QUIC_CHANNEL code to determine the channel event
3216      * handling deadline.
3217      */
3218     OSSL_TIME deadline = ossl_time_infinite();
3219     uint32_t enc_level, pn_space;
3220 
3221     /*
3222      * ACK generation is not CC-gated - packets containing only ACKs are allowed
3223      * to bypass CC. We want to generate ACK frames even if we are currently
3224      * restricted by CC so the peer knows we have received data. The generate
3225      * call will take care of selecting the correct packet archetype.
3226      */
3227     for (enc_level = QUIC_ENC_LEVEL_INITIAL;
3228         enc_level < QUIC_ENC_LEVEL_NUM;
3229         ++enc_level)
3230         if (ossl_qtx_is_enc_level_provisioned(txp->args.qtx, enc_level)) {
3231             pn_space = ossl_quic_enc_level_to_pn_space(enc_level);
3232             deadline = ossl_time_min(deadline,
3233                 ossl_ackm_get_ack_deadline(txp->args.ackm, pn_space));
3234         }
3235 
3236     /* When will CC let us send more? */
3237     if (txp->args.cc_method->get_tx_allowance(txp->args.cc_data) == 0)
3238         deadline = ossl_time_min(deadline,
3239             txp->args.cc_method->get_wakeup_deadline(txp->args.cc_data));
3240 
3241     return deadline;
3242 }
3243