1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
4 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
5 * Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
6 *
7 * An implementation of the DCCP protocol
8 *
9 * This code has been developed by the University of Waikato WAND
10 * research group. For further information please see https://www.wand.net.nz/
11 *
12 * This code also uses code from Lulea University, rereleased as GPL by its
13 * authors:
14 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
15 *
16 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
17 * and to make it work as a loadable module in the DCCP stack written by
18 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
19 *
20 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
21 */
22 #include "../dccp.h"
23 #include "ccid3.h"
24
25 #include <asm/unaligned.h>
26
27 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
28 static bool ccid3_debug;
29 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
30 #else
31 #define ccid3_pr_debug(format, a...)
32 #endif
33
34 /*
35 * Transmitter Half-Connection Routines
36 */
37 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
ccid3_tx_state_name(enum ccid3_hc_tx_states state)38 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
39 {
40 static const char *const ccid3_state_names[] = {
41 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
42 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
43 [TFRC_SSTATE_FBACK] = "FBACK",
44 };
45
46 return ccid3_state_names[state];
47 }
48 #endif
49
ccid3_hc_tx_set_state(struct sock * sk,enum ccid3_hc_tx_states state)50 static void ccid3_hc_tx_set_state(struct sock *sk,
51 enum ccid3_hc_tx_states state)
52 {
53 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
54 enum ccid3_hc_tx_states oldstate = hc->tx_state;
55
56 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
57 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
58 ccid3_tx_state_name(state));
59 WARN_ON(state == oldstate);
60 hc->tx_state = state;
61 }
62
63 /*
64 * Compute the initial sending rate X_init in the manner of RFC 3390:
65 *
66 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
67 *
68 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
69 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
70 * For consistency with other parts of the code, X_init is scaled by 2^6.
71 */
rfc3390_initial_rate(struct sock * sk)72 static inline u64 rfc3390_initial_rate(struct sock *sk)
73 {
74 const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
75 const __u32 w_init = clamp_t(__u32, 4380U, 2 * hc->tx_s, 4 * hc->tx_s);
76
77 return scaled_div(w_init << 6, hc->tx_rtt);
78 }
79
80 /**
81 * ccid3_update_send_interval - Calculate new t_ipi = s / X_inst
82 * @hc: socket to have the send interval updated
83 *
84 * This respects the granularity of X_inst (64 * bytes/second).
85 */
ccid3_update_send_interval(struct ccid3_hc_tx_sock * hc)86 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hc)
87 {
88 hc->tx_t_ipi = scaled_div32(((u64)hc->tx_s) << 6, hc->tx_x);
89
90 DCCP_BUG_ON(hc->tx_t_ipi == 0);
91 ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hc->tx_t_ipi,
92 hc->tx_s, (unsigned int)(hc->tx_x >> 6));
93 }
94
ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock * hc,ktime_t now)95 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hc, ktime_t now)
96 {
97 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count);
98
99 return delta / hc->tx_rtt;
100 }
101
102 /**
103 * ccid3_hc_tx_update_x - Update allowed sending rate X
104 * @sk: socket to be updated
105 * @stamp: most recent time if available - can be left NULL.
106 *
107 * This function tracks draft rfc3448bis, check there for latest details.
108 *
109 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
110 * fine-grained resolution of sending rates. This requires scaling by 2^6
111 * throughout the code. Only X_calc is unscaled (in bytes/second).
112 *
113 */
ccid3_hc_tx_update_x(struct sock * sk,ktime_t * stamp)114 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
115 {
116 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
117 __u64 min_rate = 2 * hc->tx_x_recv;
118 const __u64 old_x = hc->tx_x;
119 ktime_t now = stamp ? *stamp : ktime_get_real();
120
121 /*
122 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
123 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
124 * a sender is idle if it has not sent anything over a 2-RTT-period.
125 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
126 */
127 if (ccid3_hc_tx_idle_rtt(hc, now) >= 2) {
128 min_rate = rfc3390_initial_rate(sk);
129 min_rate = max(min_rate, 2 * hc->tx_x_recv);
130 }
131
132 if (hc->tx_p > 0) {
133
134 hc->tx_x = min(((__u64)hc->tx_x_calc) << 6, min_rate);
135 hc->tx_x = max(hc->tx_x, (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
136
137 } else if (ktime_us_delta(now, hc->tx_t_ld) - (s64)hc->tx_rtt >= 0) {
138
139 hc->tx_x = min(2 * hc->tx_x, min_rate);
140 hc->tx_x = max(hc->tx_x,
141 scaled_div(((__u64)hc->tx_s) << 6, hc->tx_rtt));
142 hc->tx_t_ld = now;
143 }
144
145 if (hc->tx_x != old_x) {
146 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
147 "X_recv=%u\n", (unsigned int)(old_x >> 6),
148 (unsigned int)(hc->tx_x >> 6), hc->tx_x_calc,
149 (unsigned int)(hc->tx_x_recv >> 6));
150
151 ccid3_update_send_interval(hc);
152 }
153 }
154
155 /**
156 * ccid3_hc_tx_update_s - Track the mean packet size `s'
157 * @hc: socket to be updated
158 * @len: DCCP packet payload size in bytes
159 *
160 * cf. RFC 4342, 5.3 and RFC 3448, 4.1
161 */
ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock * hc,int len)162 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hc, int len)
163 {
164 const u16 old_s = hc->tx_s;
165
166 hc->tx_s = tfrc_ewma(hc->tx_s, len, 9);
167
168 if (hc->tx_s != old_s)
169 ccid3_update_send_interval(hc);
170 }
171
172 /*
173 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
174 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
175 */
ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock * hc,ktime_t now)176 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hc,
177 ktime_t now)
178 {
179 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count),
180 quarter_rtts = (4 * delta) / hc->tx_rtt;
181
182 if (quarter_rtts > 0) {
183 hc->tx_t_last_win_count = now;
184 hc->tx_last_win_count += min(quarter_rtts, 5U);
185 hc->tx_last_win_count &= 0xF; /* mod 16 */
186 }
187 }
188
ccid3_hc_tx_no_feedback_timer(struct timer_list * t)189 static void ccid3_hc_tx_no_feedback_timer(struct timer_list *t)
190 {
191 struct ccid3_hc_tx_sock *hc = from_timer(hc, t, tx_no_feedback_timer);
192 struct sock *sk = hc->sk;
193 unsigned long t_nfb = USEC_PER_SEC / 5;
194
195 bh_lock_sock(sk);
196 if (sock_owned_by_user(sk)) {
197 /* Try again later. */
198 /* XXX: set some sensible MIB */
199 goto restart_timer;
200 }
201
202 ccid3_pr_debug("%s(%p, state=%s) - entry\n", dccp_role(sk), sk,
203 ccid3_tx_state_name(hc->tx_state));
204
205 /* Ignore and do not restart after leaving the established state */
206 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
207 goto out;
208
209 /* Reset feedback state to "no feedback received" */
210 if (hc->tx_state == TFRC_SSTATE_FBACK)
211 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
212
213 /*
214 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
215 * RTO is 0 if and only if no feedback has been received yet.
216 */
217 if (hc->tx_t_rto == 0 || hc->tx_p == 0) {
218
219 /* halve send rate directly */
220 hc->tx_x = max(hc->tx_x / 2,
221 (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
222 ccid3_update_send_interval(hc);
223 } else {
224 /*
225 * Modify the cached value of X_recv
226 *
227 * If (X_calc > 2 * X_recv)
228 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
229 * Else
230 * X_recv = X_calc / 4;
231 *
232 * Note that X_recv is scaled by 2^6 while X_calc is not
233 */
234 if (hc->tx_x_calc > (hc->tx_x_recv >> 5))
235 hc->tx_x_recv =
236 max(hc->tx_x_recv / 2,
237 (((__u64)hc->tx_s) << 6) / (2*TFRC_T_MBI));
238 else {
239 hc->tx_x_recv = hc->tx_x_calc;
240 hc->tx_x_recv <<= 4;
241 }
242 ccid3_hc_tx_update_x(sk, NULL);
243 }
244 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
245 (unsigned long long)hc->tx_x);
246
247 /*
248 * Set new timeout for the nofeedback timer.
249 * See comments in packet_recv() regarding the value of t_RTO.
250 */
251 if (unlikely(hc->tx_t_rto == 0)) /* no feedback received yet */
252 t_nfb = TFRC_INITIAL_TIMEOUT;
253 else
254 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
255
256 restart_timer:
257 sk_reset_timer(sk, &hc->tx_no_feedback_timer,
258 jiffies + usecs_to_jiffies(t_nfb));
259 out:
260 bh_unlock_sock(sk);
261 sock_put(sk);
262 }
263
264 /**
265 * ccid3_hc_tx_send_packet - Delay-based dequeueing of TX packets
266 * @sk: socket to send packet from
267 * @skb: next packet candidate to send on @sk
268 *
269 * This function uses the convention of ccid_packet_dequeue_eval() and
270 * returns a millisecond-delay value between 0 and t_mbi = 64000 msec.
271 */
ccid3_hc_tx_send_packet(struct sock * sk,struct sk_buff * skb)272 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
273 {
274 struct dccp_sock *dp = dccp_sk(sk);
275 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
276 ktime_t now = ktime_get_real();
277 s64 delay;
278
279 /*
280 * This function is called only for Data and DataAck packets. Sending
281 * zero-sized Data(Ack)s is theoretically possible, but for congestion
282 * control this case is pathological - ignore it.
283 */
284 if (unlikely(skb->len == 0))
285 return -EBADMSG;
286
287 if (hc->tx_state == TFRC_SSTATE_NO_SENT) {
288 sk_reset_timer(sk, &hc->tx_no_feedback_timer, (jiffies +
289 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
290 hc->tx_last_win_count = 0;
291 hc->tx_t_last_win_count = now;
292
293 /* Set t_0 for initial packet */
294 hc->tx_t_nom = now;
295
296 hc->tx_s = skb->len;
297
298 /*
299 * Use initial RTT sample when available: recommended by erratum
300 * to RFC 4342. This implements the initialisation procedure of
301 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
302 */
303 if (dp->dccps_syn_rtt) {
304 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
305 hc->tx_rtt = dp->dccps_syn_rtt;
306 hc->tx_x = rfc3390_initial_rate(sk);
307 hc->tx_t_ld = now;
308 } else {
309 /*
310 * Sender does not have RTT sample:
311 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
312 * is needed in several parts (e.g. window counter);
313 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
314 */
315 hc->tx_rtt = DCCP_FALLBACK_RTT;
316 hc->tx_x = hc->tx_s;
317 hc->tx_x <<= 6;
318 }
319 ccid3_update_send_interval(hc);
320
321 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
322
323 } else {
324 delay = ktime_us_delta(hc->tx_t_nom, now);
325 ccid3_pr_debug("delay=%ld\n", (long)delay);
326 /*
327 * Scheduling of packet transmissions (RFC 5348, 8.3)
328 *
329 * if (t_now > t_nom - delta)
330 * // send the packet now
331 * else
332 * // send the packet in (t_nom - t_now) milliseconds.
333 */
334 if (delay >= TFRC_T_DELTA)
335 return (u32)delay / USEC_PER_MSEC;
336
337 ccid3_hc_tx_update_win_count(hc, now);
338 }
339
340 /* prepare to send now (add options etc.) */
341 dp->dccps_hc_tx_insert_options = 1;
342 DCCP_SKB_CB(skb)->dccpd_ccval = hc->tx_last_win_count;
343
344 /* set the nominal send time for the next following packet */
345 hc->tx_t_nom = ktime_add_us(hc->tx_t_nom, hc->tx_t_ipi);
346 return CCID_PACKET_SEND_AT_ONCE;
347 }
348
ccid3_hc_tx_packet_sent(struct sock * sk,unsigned int len)349 static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len)
350 {
351 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
352
353 ccid3_hc_tx_update_s(hc, len);
354
355 if (tfrc_tx_hist_add(&hc->tx_hist, dccp_sk(sk)->dccps_gss))
356 DCCP_CRIT("packet history - out of memory!");
357 }
358
ccid3_hc_tx_packet_recv(struct sock * sk,struct sk_buff * skb)359 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
360 {
361 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
362 struct tfrc_tx_hist_entry *acked;
363 ktime_t now;
364 unsigned long t_nfb;
365 u32 r_sample;
366
367 /* we are only interested in ACKs */
368 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
369 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
370 return;
371 /*
372 * Locate the acknowledged packet in the TX history.
373 *
374 * Returning "entry not found" here can for instance happen when
375 * - the host has not sent out anything (e.g. a passive server),
376 * - the Ack is outdated (packet with higher Ack number was received),
377 * - it is a bogus Ack (for a packet not sent on this connection).
378 */
379 acked = tfrc_tx_hist_find_entry(hc->tx_hist, dccp_hdr_ack_seq(skb));
380 if (acked == NULL)
381 return;
382 /* For the sake of RTT sampling, ignore/remove all older entries */
383 tfrc_tx_hist_purge(&acked->next);
384
385 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
386 now = ktime_get_real();
387 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
388 hc->tx_rtt = tfrc_ewma(hc->tx_rtt, r_sample, 9);
389
390 /*
391 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
392 */
393 if (hc->tx_state == TFRC_SSTATE_NO_FBACK) {
394 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
395
396 if (hc->tx_t_rto == 0) {
397 /*
398 * Initial feedback packet: Larger Initial Windows (4.2)
399 */
400 hc->tx_x = rfc3390_initial_rate(sk);
401 hc->tx_t_ld = now;
402
403 ccid3_update_send_interval(hc);
404
405 goto done_computing_x;
406 } else if (hc->tx_p == 0) {
407 /*
408 * First feedback after nofeedback timer expiry (4.3)
409 */
410 goto done_computing_x;
411 }
412 }
413
414 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
415 if (hc->tx_p > 0)
416 hc->tx_x_calc = tfrc_calc_x(hc->tx_s, hc->tx_rtt, hc->tx_p);
417 ccid3_hc_tx_update_x(sk, &now);
418
419 done_computing_x:
420 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
421 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
422 dccp_role(sk), sk, hc->tx_rtt, r_sample,
423 hc->tx_s, hc->tx_p, hc->tx_x_calc,
424 (unsigned int)(hc->tx_x_recv >> 6),
425 (unsigned int)(hc->tx_x >> 6));
426
427 /* unschedule no feedback timer */
428 sk_stop_timer(sk, &hc->tx_no_feedback_timer);
429
430 /*
431 * As we have calculated new ipi, delta, t_nom it is possible
432 * that we now can send a packet, so wake up dccp_wait_for_ccid
433 */
434 sk->sk_write_space(sk);
435
436 /*
437 * Update timeout interval for the nofeedback timer. In order to control
438 * rate halving on networks with very low RTTs (<= 1 ms), use per-route
439 * tunable RTAX_RTO_MIN value as the lower bound.
440 */
441 hc->tx_t_rto = max_t(u32, 4 * hc->tx_rtt,
442 USEC_PER_SEC/HZ * tcp_rto_min(sk));
443 /*
444 * Schedule no feedback timer to expire in
445 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
446 */
447 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
448
449 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
450 "expire in %lu jiffies (%luus)\n",
451 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
452
453 sk_reset_timer(sk, &hc->tx_no_feedback_timer,
454 jiffies + usecs_to_jiffies(t_nfb));
455 }
456
ccid3_hc_tx_parse_options(struct sock * sk,u8 packet_type,u8 option,u8 * optval,u8 optlen)457 static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type,
458 u8 option, u8 *optval, u8 optlen)
459 {
460 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
461 __be32 opt_val;
462
463 switch (option) {
464 case TFRC_OPT_RECEIVE_RATE:
465 case TFRC_OPT_LOSS_EVENT_RATE:
466 /* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
467 if (packet_type == DCCP_PKT_DATA)
468 break;
469 if (unlikely(optlen != 4)) {
470 DCCP_WARN("%s(%p), invalid len %d for %u\n",
471 dccp_role(sk), sk, optlen, option);
472 return -EINVAL;
473 }
474 opt_val = ntohl(get_unaligned((__be32 *)optval));
475
476 if (option == TFRC_OPT_RECEIVE_RATE) {
477 /* Receive Rate is kept in units of 64 bytes/second */
478 hc->tx_x_recv = opt_val;
479 hc->tx_x_recv <<= 6;
480
481 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
482 dccp_role(sk), sk, opt_val);
483 } else {
484 /* Update the fixpoint Loss Event Rate fraction */
485 hc->tx_p = tfrc_invert_loss_event_rate(opt_val);
486
487 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
488 dccp_role(sk), sk, opt_val);
489 }
490 }
491 return 0;
492 }
493
ccid3_hc_tx_init(struct ccid * ccid,struct sock * sk)494 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
495 {
496 struct ccid3_hc_tx_sock *hc = ccid_priv(ccid);
497
498 hc->tx_state = TFRC_SSTATE_NO_SENT;
499 hc->tx_hist = NULL;
500 hc->sk = sk;
501 timer_setup(&hc->tx_no_feedback_timer,
502 ccid3_hc_tx_no_feedback_timer, 0);
503 return 0;
504 }
505
ccid3_hc_tx_exit(struct sock * sk)506 static void ccid3_hc_tx_exit(struct sock *sk)
507 {
508 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
509
510 sk_stop_timer(sk, &hc->tx_no_feedback_timer);
511 tfrc_tx_hist_purge(&hc->tx_hist);
512 }
513
ccid3_hc_tx_get_info(struct sock * sk,struct tcp_info * info)514 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
515 {
516 info->tcpi_rto = ccid3_hc_tx_sk(sk)->tx_t_rto;
517 info->tcpi_rtt = ccid3_hc_tx_sk(sk)->tx_rtt;
518 }
519
ccid3_hc_tx_getsockopt(struct sock * sk,const int optname,int len,u32 __user * optval,int __user * optlen)520 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
521 u32 __user *optval, int __user *optlen)
522 {
523 const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
524 struct tfrc_tx_info tfrc;
525 const void *val;
526
527 switch (optname) {
528 case DCCP_SOCKOPT_CCID_TX_INFO:
529 if (len < sizeof(tfrc))
530 return -EINVAL;
531 memset(&tfrc, 0, sizeof(tfrc));
532 tfrc.tfrctx_x = hc->tx_x;
533 tfrc.tfrctx_x_recv = hc->tx_x_recv;
534 tfrc.tfrctx_x_calc = hc->tx_x_calc;
535 tfrc.tfrctx_rtt = hc->tx_rtt;
536 tfrc.tfrctx_p = hc->tx_p;
537 tfrc.tfrctx_rto = hc->tx_t_rto;
538 tfrc.tfrctx_ipi = hc->tx_t_ipi;
539 len = sizeof(tfrc);
540 val = &tfrc;
541 break;
542 default:
543 return -ENOPROTOOPT;
544 }
545
546 if (put_user(len, optlen) || copy_to_user(optval, val, len))
547 return -EFAULT;
548
549 return 0;
550 }
551
552 /*
553 * Receiver Half-Connection Routines
554 */
555
556 /* CCID3 feedback types */
557 enum ccid3_fback_type {
558 CCID3_FBACK_NONE = 0,
559 CCID3_FBACK_INITIAL,
560 CCID3_FBACK_PERIODIC,
561 CCID3_FBACK_PARAM_CHANGE
562 };
563
564 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
ccid3_rx_state_name(enum ccid3_hc_rx_states state)565 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
566 {
567 static const char *const ccid3_rx_state_names[] = {
568 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
569 [TFRC_RSTATE_DATA] = "DATA",
570 };
571
572 return ccid3_rx_state_names[state];
573 }
574 #endif
575
ccid3_hc_rx_set_state(struct sock * sk,enum ccid3_hc_rx_states state)576 static void ccid3_hc_rx_set_state(struct sock *sk,
577 enum ccid3_hc_rx_states state)
578 {
579 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
580 enum ccid3_hc_rx_states oldstate = hc->rx_state;
581
582 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
583 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
584 ccid3_rx_state_name(state));
585 WARN_ON(state == oldstate);
586 hc->rx_state = state;
587 }
588
ccid3_hc_rx_send_feedback(struct sock * sk,const struct sk_buff * skb,enum ccid3_fback_type fbtype)589 static void ccid3_hc_rx_send_feedback(struct sock *sk,
590 const struct sk_buff *skb,
591 enum ccid3_fback_type fbtype)
592 {
593 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
594 struct dccp_sock *dp = dccp_sk(sk);
595 ktime_t now = ktime_get();
596 s64 delta = 0;
597
598 switch (fbtype) {
599 case CCID3_FBACK_INITIAL:
600 hc->rx_x_recv = 0;
601 hc->rx_pinv = ~0U; /* see RFC 4342, 8.5 */
602 break;
603 case CCID3_FBACK_PARAM_CHANGE:
604 /*
605 * When parameters change (new loss or p > p_prev), we do not
606 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
607 * need to reuse the previous value of X_recv. However, when
608 * X_recv was 0 (due to early loss), this would kill X down to
609 * s/t_mbi (i.e. one packet in 64 seconds).
610 * To avoid such drastic reduction, we approximate X_recv as
611 * the number of bytes since last feedback.
612 * This is a safe fallback, since X is bounded above by X_calc.
613 */
614 if (hc->rx_x_recv > 0)
615 break;
616 fallthrough;
617 case CCID3_FBACK_PERIODIC:
618 delta = ktime_us_delta(now, hc->rx_tstamp_last_feedback);
619 if (delta <= 0)
620 delta = 1;
621 hc->rx_x_recv = scaled_div32(hc->rx_bytes_recv, delta);
622 break;
623 default:
624 return;
625 }
626
627 ccid3_pr_debug("Interval %lldusec, X_recv=%u, 1/p=%u\n", delta,
628 hc->rx_x_recv, hc->rx_pinv);
629
630 hc->rx_tstamp_last_feedback = now;
631 hc->rx_last_counter = dccp_hdr(skb)->dccph_ccval;
632 hc->rx_bytes_recv = 0;
633
634 dp->dccps_hc_rx_insert_options = 1;
635 dccp_send_ack(sk);
636 }
637
ccid3_hc_rx_insert_options(struct sock * sk,struct sk_buff * skb)638 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
639 {
640 const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
641 __be32 x_recv, pinv;
642
643 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
644 return 0;
645
646 if (dccp_packet_without_ack(skb))
647 return 0;
648
649 x_recv = htonl(hc->rx_x_recv);
650 pinv = htonl(hc->rx_pinv);
651
652 if (dccp_insert_option(skb, TFRC_OPT_LOSS_EVENT_RATE,
653 &pinv, sizeof(pinv)) ||
654 dccp_insert_option(skb, TFRC_OPT_RECEIVE_RATE,
655 &x_recv, sizeof(x_recv)))
656 return -1;
657
658 return 0;
659 }
660
661 /**
662 * ccid3_first_li - Implements [RFC 5348, 6.3.1]
663 * @sk: socket to calculate loss interval for
664 *
665 * Determine the length of the first loss interval via inverse lookup.
666 * Assume that X_recv can be computed by the throughput equation
667 * s
668 * X_recv = --------
669 * R * fval
670 * Find some p such that f(p) = fval; return 1/p (scaled).
671 */
ccid3_first_li(struct sock * sk)672 static u32 ccid3_first_li(struct sock *sk)
673 {
674 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
675 u32 x_recv, p;
676 s64 delta;
677 u64 fval;
678
679 if (hc->rx_rtt == 0) {
680 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
681 hc->rx_rtt = DCCP_FALLBACK_RTT;
682 }
683
684 delta = ktime_us_delta(ktime_get(), hc->rx_tstamp_last_feedback);
685 if (delta <= 0)
686 delta = 1;
687 x_recv = scaled_div32(hc->rx_bytes_recv, delta);
688 if (x_recv == 0) { /* would also trigger divide-by-zero */
689 DCCP_WARN("X_recv==0\n");
690 if (hc->rx_x_recv == 0) {
691 DCCP_BUG("stored value of X_recv is zero");
692 return ~0U;
693 }
694 x_recv = hc->rx_x_recv;
695 }
696
697 fval = scaled_div(hc->rx_s, hc->rx_rtt);
698 fval = scaled_div32(fval, x_recv);
699 p = tfrc_calc_x_reverse_lookup(fval);
700
701 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
702 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
703
704 return p == 0 ? ~0U : scaled_div(1, p);
705 }
706
ccid3_hc_rx_packet_recv(struct sock * sk,struct sk_buff * skb)707 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
708 {
709 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
710 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
711 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
712 const bool is_data_packet = dccp_data_packet(skb);
713
714 if (unlikely(hc->rx_state == TFRC_RSTATE_NO_DATA)) {
715 if (is_data_packet) {
716 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
717 do_feedback = CCID3_FBACK_INITIAL;
718 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
719 hc->rx_s = payload;
720 /*
721 * Not necessary to update rx_bytes_recv here,
722 * since X_recv = 0 for the first feedback packet (cf.
723 * RFC 3448, 6.3) -- gerrit
724 */
725 }
726 goto update_records;
727 }
728
729 if (tfrc_rx_hist_duplicate(&hc->rx_hist, skb))
730 return; /* done receiving */
731
732 if (is_data_packet) {
733 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
734 /*
735 * Update moving-average of s and the sum of received payload bytes
736 */
737 hc->rx_s = tfrc_ewma(hc->rx_s, payload, 9);
738 hc->rx_bytes_recv += payload;
739 }
740
741 /*
742 * Perform loss detection and handle pending losses
743 */
744 if (tfrc_rx_handle_loss(&hc->rx_hist, &hc->rx_li_hist,
745 skb, ndp, ccid3_first_li, sk)) {
746 do_feedback = CCID3_FBACK_PARAM_CHANGE;
747 goto done_receiving;
748 }
749
750 if (tfrc_rx_hist_loss_pending(&hc->rx_hist))
751 return; /* done receiving */
752
753 /*
754 * Handle data packets: RTT sampling and monitoring p
755 */
756 if (unlikely(!is_data_packet))
757 goto update_records;
758
759 if (!tfrc_lh_is_initialised(&hc->rx_li_hist)) {
760 const u32 sample = tfrc_rx_hist_sample_rtt(&hc->rx_hist, skb);
761 /*
762 * Empty loss history: no loss so far, hence p stays 0.
763 * Sample RTT values, since an RTT estimate is required for the
764 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
765 */
766 if (sample != 0)
767 hc->rx_rtt = tfrc_ewma(hc->rx_rtt, sample, 9);
768
769 } else if (tfrc_lh_update_i_mean(&hc->rx_li_hist, skb)) {
770 /*
771 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
772 * has decreased (resp. p has increased), send feedback now.
773 */
774 do_feedback = CCID3_FBACK_PARAM_CHANGE;
775 }
776
777 /*
778 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
779 */
780 if (SUB16(dccp_hdr(skb)->dccph_ccval, hc->rx_last_counter) > 3)
781 do_feedback = CCID3_FBACK_PERIODIC;
782
783 update_records:
784 tfrc_rx_hist_add_packet(&hc->rx_hist, skb, ndp);
785
786 done_receiving:
787 if (do_feedback)
788 ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
789 }
790
ccid3_hc_rx_init(struct ccid * ccid,struct sock * sk)791 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
792 {
793 struct ccid3_hc_rx_sock *hc = ccid_priv(ccid);
794
795 hc->rx_state = TFRC_RSTATE_NO_DATA;
796 tfrc_lh_init(&hc->rx_li_hist);
797 return tfrc_rx_hist_alloc(&hc->rx_hist);
798 }
799
ccid3_hc_rx_exit(struct sock * sk)800 static void ccid3_hc_rx_exit(struct sock *sk)
801 {
802 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
803
804 tfrc_rx_hist_purge(&hc->rx_hist);
805 tfrc_lh_cleanup(&hc->rx_li_hist);
806 }
807
ccid3_hc_rx_get_info(struct sock * sk,struct tcp_info * info)808 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
809 {
810 info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->rx_state;
811 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
812 info->tcpi_rcv_rtt = ccid3_hc_rx_sk(sk)->rx_rtt;
813 }
814
ccid3_hc_rx_getsockopt(struct sock * sk,const int optname,int len,u32 __user * optval,int __user * optlen)815 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
816 u32 __user *optval, int __user *optlen)
817 {
818 const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
819 struct tfrc_rx_info rx_info;
820 const void *val;
821
822 switch (optname) {
823 case DCCP_SOCKOPT_CCID_RX_INFO:
824 if (len < sizeof(rx_info))
825 return -EINVAL;
826 rx_info.tfrcrx_x_recv = hc->rx_x_recv;
827 rx_info.tfrcrx_rtt = hc->rx_rtt;
828 rx_info.tfrcrx_p = tfrc_invert_loss_event_rate(hc->rx_pinv);
829 len = sizeof(rx_info);
830 val = &rx_info;
831 break;
832 default:
833 return -ENOPROTOOPT;
834 }
835
836 if (put_user(len, optlen) || copy_to_user(optval, val, len))
837 return -EFAULT;
838
839 return 0;
840 }
841
842 struct ccid_operations ccid3_ops = {
843 .ccid_id = DCCPC_CCID3,
844 .ccid_name = "TCP-Friendly Rate Control",
845 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
846 .ccid_hc_tx_init = ccid3_hc_tx_init,
847 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
848 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
849 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
850 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
851 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
852 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
853 .ccid_hc_rx_init = ccid3_hc_rx_init,
854 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
855 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
856 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
857 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
858 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
859 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
860 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
861 };
862
863 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
864 module_param(ccid3_debug, bool, 0644);
865 MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages");
866 #endif
867