2 * net/dccp/ccids/ccid3.c
4 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
5 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
6 * Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
8 * An implementation of the DCCP protocol
10 * This code has been developed by the University of Waikato WAND
11 * research group. For further information please see http://www.wand.net.nz/
13 * This code also uses code from Lulea University, rereleased as GPL by its
15 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
17 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
18 * and to make it work as a loadable module in the DCCP stack written by
19 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
21 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
23 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
33 * You should have received a copy of the GNU General Public License
34 * along with this program; if not, write to the Free Software
35 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
40 #include <asm/unaligned.h>
42 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
43 static int ccid3_debug
;
44 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
46 #define ccid3_pr_debug(format, a...)
50 * Transmitter Half-Connection Routines
52 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
53 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state
)
55 static char *ccid3_state_names
[] = {
56 [TFRC_SSTATE_NO_SENT
] = "NO_SENT",
57 [TFRC_SSTATE_NO_FBACK
] = "NO_FBACK",
58 [TFRC_SSTATE_FBACK
] = "FBACK",
59 [TFRC_SSTATE_TERM
] = "TERM",
62 return ccid3_state_names
[state
];
66 static void ccid3_hc_tx_set_state(struct sock
*sk
,
67 enum ccid3_hc_tx_states state
)
69 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
70 enum ccid3_hc_tx_states oldstate
= hctx
->state
;
72 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
73 dccp_role(sk
), sk
, ccid3_tx_state_name(oldstate
),
74 ccid3_tx_state_name(state
));
75 WARN_ON(state
== oldstate
);
80 * Compute the initial sending rate X_init in the manner of RFC 3390:
82 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
84 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
85 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
86 * For consistency with other parts of the code, X_init is scaled by 2^6.
88 static inline u64
rfc3390_initial_rate(struct sock
*sk
)
90 const struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
91 const __u32 w_init
= clamp_t(__u32
, 4380U, 2 * hctx
->s
, 4 * hctx
->s
);
93 return scaled_div(w_init
<< 6, hctx
->rtt
);
97 * Recalculate t_ipi and delta (should be called whenever X changes)
99 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock
*hctx
)
101 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
102 hctx
->t_ipi
= scaled_div32(((u64
)hctx
->s
) << 6, hctx
->x
);
104 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
105 hctx
->delta
= min_t(u32
, hctx
->t_ipi
/ 2, TFRC_OPSYS_HALF_TIME_GRAN
);
107 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n", hctx
->t_ipi
,
108 hctx
->delta
, hctx
->s
, (unsigned)(hctx
->x
>> 6));
111 static u32
ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock
*hctx
, ktime_t now
)
113 u32 delta
= ktime_us_delta(now
, hctx
->t_last_win_count
);
115 return delta
/ hctx
->rtt
;
119 * ccid3_hc_tx_update_x - Update allowed sending rate X
120 * @stamp: most recent time if available - can be left NULL.
121 * This function tracks draft rfc3448bis, check there for latest details.
123 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
124 * fine-grained resolution of sending rates. This requires scaling by 2^6
125 * throughout the code. Only X_calc is unscaled (in bytes/second).
128 static void ccid3_hc_tx_update_x(struct sock
*sk
, ktime_t
*stamp
)
130 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
131 u64 min_rate
= 2 * hctx
->x_recv
;
132 const u64 old_x
= hctx
->x
;
133 ktime_t now
= stamp
? *stamp
: ktime_get_real();
136 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
137 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
138 * a sender is idle if it has not sent anything over a 2-RTT-period.
139 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
141 if (ccid3_hc_tx_idle_rtt(hctx
, now
) >= 2) {
142 min_rate
= rfc3390_initial_rate(sk
);
143 min_rate
= max(min_rate
, 2 * hctx
->x_recv
);
148 hctx
->x
= min(((u64
)hctx
->x_calc
) << 6, min_rate
);
149 hctx
->x
= max(hctx
->x
, (((u64
)hctx
->s
) << 6) / TFRC_T_MBI
);
151 } else if (ktime_us_delta(now
, hctx
->t_ld
) - (s64
)hctx
->rtt
>= 0) {
153 hctx
->x
= min(2 * hctx
->x
, min_rate
);
154 hctx
->x
= max(hctx
->x
,
155 scaled_div(((u64
)hctx
->s
) << 6, hctx
->rtt
));
159 if (hctx
->x
!= old_x
) {
160 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
161 "X_recv=%u\n", (unsigned)(old_x
>> 6),
162 (unsigned)(hctx
->x
>> 6), hctx
->x_calc
,
163 (unsigned)(hctx
->x_recv
>> 6));
165 ccid3_update_send_interval(hctx
);
170 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
171 * @len: DCCP packet payload size in bytes
173 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock
*hctx
, int len
)
175 const u16 old_s
= hctx
->s
;
177 hctx
->s
= tfrc_ewma(hctx
->s
, len
, 9);
179 if (hctx
->s
!= old_s
)
180 ccid3_update_send_interval(hctx
);
184 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
185 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
187 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock
*hctx
,
190 u32 delta
= ktime_us_delta(now
, hctx
->t_last_win_count
),
191 quarter_rtts
= (4 * delta
) / hctx
->rtt
;
193 if (quarter_rtts
> 0) {
194 hctx
->t_last_win_count
= now
;
195 hctx
->last_win_count
+= min(quarter_rtts
, 5U);
196 hctx
->last_win_count
&= 0xF; /* mod 16 */
200 static void ccid3_hc_tx_no_feedback_timer(unsigned long data
)
202 struct sock
*sk
= (struct sock
*)data
;
203 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
204 unsigned long t_nfb
= USEC_PER_SEC
/ 5;
207 if (sock_owned_by_user(sk
)) {
208 /* Try again later. */
209 /* XXX: set some sensible MIB */
213 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk
), sk
,
214 ccid3_tx_state_name(hctx
->state
));
216 if (hctx
->state
== TFRC_SSTATE_FBACK
)
217 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_NO_FBACK
);
218 else if (hctx
->state
!= TFRC_SSTATE_NO_FBACK
)
222 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
223 * RTO is 0 if and only if no feedback has been received yet.
225 if (hctx
->t_rto
== 0 || hctx
->p
== 0) {
227 /* halve send rate directly */
228 hctx
->x
= max(hctx
->x
/ 2, (((u64
)hctx
->s
) << 6) / TFRC_T_MBI
);
229 ccid3_update_send_interval(hctx
);
232 * Modify the cached value of X_recv
234 * If (X_calc > 2 * X_recv)
235 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
237 * X_recv = X_calc / 4;
239 * Note that X_recv is scaled by 2^6 while X_calc is not
241 BUG_ON(hctx
->p
&& !hctx
->x_calc
);
243 if (hctx
->x_calc
> (hctx
->x_recv
>> 5))
245 max(hctx
->x_recv
/ 2,
246 (((__u64
)hctx
->s
) << 6) / (2 * TFRC_T_MBI
));
248 hctx
->x_recv
= hctx
->x_calc
;
251 ccid3_hc_tx_update_x(sk
, NULL
);
253 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
254 (unsigned long long)hctx
->x
);
257 * Set new timeout for the nofeedback timer.
258 * See comments in packet_recv() regarding the value of t_RTO.
260 if (unlikely(hctx
->t_rto
== 0)) /* no feedback received yet */
261 t_nfb
= TFRC_INITIAL_TIMEOUT
;
263 t_nfb
= max(hctx
->t_rto
, 2 * hctx
->t_ipi
);
266 sk_reset_timer(sk
, &hctx
->no_feedback_timer
,
267 jiffies
+ usecs_to_jiffies(t_nfb
));
275 * > 0: delay (in msecs) that should pass before actually sending
276 * = 0: can send immediately
277 * < 0: error condition; do not send packet
279 static int ccid3_hc_tx_send_packet(struct sock
*sk
, struct sk_buff
*skb
)
281 struct dccp_sock
*dp
= dccp_sk(sk
);
282 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
283 ktime_t now
= ktime_get_real();
287 * This function is called only for Data and DataAck packets. Sending
288 * zero-sized Data(Ack)s is theoretically possible, but for congestion
289 * control this case is pathological - ignore it.
291 if (unlikely(skb
->len
== 0))
294 switch (hctx
->state
) {
295 case TFRC_SSTATE_NO_SENT
:
296 sk_reset_timer(sk
, &hctx
->no_feedback_timer
, (jiffies
+
297 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT
)));
298 hctx
->last_win_count
= 0;
299 hctx
->t_last_win_count
= now
;
301 /* Set t_0 for initial packet */
307 * Use initial RTT sample when available: recommended by erratum
308 * to RFC 4342. This implements the initialisation procedure of
309 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
311 if (dp
->dccps_syn_rtt
) {
312 ccid3_pr_debug("SYN RTT = %uus\n", dp
->dccps_syn_rtt
);
313 hctx
->rtt
= dp
->dccps_syn_rtt
;
314 hctx
->x
= rfc3390_initial_rate(sk
);
318 * Sender does not have RTT sample:
319 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
320 * is needed in several parts (e.g. window counter);
321 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
323 hctx
->rtt
= DCCP_FALLBACK_RTT
;
327 ccid3_update_send_interval(hctx
);
329 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_NO_FBACK
);
331 case TFRC_SSTATE_NO_FBACK
:
332 case TFRC_SSTATE_FBACK
:
333 delay
= ktime_us_delta(hctx
->t_nom
, now
);
334 ccid3_pr_debug("delay=%ld\n", (long)delay
);
336 * Scheduling of packet transmissions [RFC 3448, 4.6]
338 * if (t_now > t_nom - delta)
339 * // send the packet now
341 * // send the packet in (t_nom - t_now) milliseconds.
343 if (delay
- (s64
)hctx
->delta
>= 1000)
344 return (u32
)delay
/ 1000L;
346 ccid3_hc_tx_update_win_count(hctx
, now
);
348 case TFRC_SSTATE_TERM
:
349 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk
), sk
);
353 /* prepare to send now (add options etc.) */
354 dp
->dccps_hc_tx_insert_options
= 1;
355 DCCP_SKB_CB(skb
)->dccpd_ccval
= hctx
->last_win_count
;
357 /* set the nominal send time for the next following packet */
358 hctx
->t_nom
= ktime_add_us(hctx
->t_nom
, hctx
->t_ipi
);
362 static void ccid3_hc_tx_packet_sent(struct sock
*sk
, int more
,
365 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
367 ccid3_hc_tx_update_s(hctx
, len
);
369 if (tfrc_tx_hist_add(&hctx
->hist
, dccp_sk(sk
)->dccps_gss
))
370 DCCP_CRIT("packet history - out of memory!");
373 static void ccid3_hc_tx_packet_recv(struct sock
*sk
, struct sk_buff
*skb
)
375 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
376 struct ccid3_options_received
*opt_recv
= &hctx
->options_received
;
381 /* we are only interested in ACKs */
382 if (!(DCCP_SKB_CB(skb
)->dccpd_type
== DCCP_PKT_ACK
||
383 DCCP_SKB_CB(skb
)->dccpd_type
== DCCP_PKT_DATAACK
))
385 /* ... and only in the established state */
386 if (hctx
->state
!= TFRC_SSTATE_FBACK
&&
387 hctx
->state
!= TFRC_SSTATE_NO_FBACK
)
390 now
= ktime_get_real();
392 /* Estimate RTT from history if ACK number is valid */
393 r_sample
= tfrc_tx_hist_rtt(hctx
->hist
,
394 DCCP_SKB_CB(skb
)->dccpd_ack_seq
, now
);
396 DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk
), sk
,
397 dccp_packet_name(DCCP_SKB_CB(skb
)->dccpd_type
),
398 (unsigned long long)DCCP_SKB_CB(skb
)->dccpd_ack_seq
);
402 /* Update receive rate in units of 64 * bytes/second */
403 hctx
->x_recv
= opt_recv
->ccid3or_receive_rate
;
406 /* Update loss event rate (which is scaled by 1e6) */
407 pinv
= opt_recv
->ccid3or_loss_event_rate
;
408 if (pinv
== ~0U || pinv
== 0) /* see RFC 4342, 8.5 */
410 else /* can not exceed 100% */
411 hctx
->p
= scaled_div(1, pinv
);
413 * Validate new RTT sample and update moving average
415 r_sample
= dccp_sample_rtt(sk
, r_sample
);
416 hctx
->rtt
= tfrc_ewma(hctx
->rtt
, r_sample
, 9);
418 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
420 if (hctx
->state
== TFRC_SSTATE_NO_FBACK
) {
421 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_FBACK
);
423 if (hctx
->t_rto
== 0) {
425 * Initial feedback packet: Larger Initial Windows (4.2)
427 hctx
->x
= rfc3390_initial_rate(sk
);
430 ccid3_update_send_interval(hctx
);
432 goto done_computing_x
;
433 } else if (hctx
->p
== 0) {
435 * First feedback after nofeedback timer expiry (4.3)
437 goto done_computing_x
;
441 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
443 hctx
->x_calc
= tfrc_calc_x(hctx
->s
, hctx
->rtt
, hctx
->p
);
444 ccid3_hc_tx_update_x(sk
, &now
);
447 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
448 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
449 dccp_role(sk
), sk
, hctx
->rtt
, r_sample
,
450 hctx
->s
, hctx
->p
, hctx
->x_calc
,
451 (unsigned)(hctx
->x_recv
>> 6),
452 (unsigned)(hctx
->x
>> 6));
454 /* unschedule no feedback timer */
455 sk_stop_timer(sk
, &hctx
->no_feedback_timer
);
458 * As we have calculated new ipi, delta, t_nom it is possible
459 * that we now can send a packet, so wake up dccp_wait_for_ccid
461 sk
->sk_write_space(sk
);
464 * Update timeout interval for the nofeedback timer.
465 * We use a configuration option to increase the lower bound.
466 * This can help avoid triggering the nofeedback timer too
467 * often ('spinning') on LANs with small RTTs.
469 hctx
->t_rto
= max_t(u32
, 4 * hctx
->rtt
, (CONFIG_IP_DCCP_CCID3_RTO
*
470 (USEC_PER_SEC
/ 1000)));
472 * Schedule no feedback timer to expire in
473 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
475 t_nfb
= max(hctx
->t_rto
, 2 * hctx
->t_ipi
);
477 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
478 "expire in %lu jiffies (%luus)\n",
479 dccp_role(sk
), sk
, usecs_to_jiffies(t_nfb
), t_nfb
);
481 sk_reset_timer(sk
, &hctx
->no_feedback_timer
,
482 jiffies
+ usecs_to_jiffies(t_nfb
));
485 static int ccid3_hc_tx_parse_options(struct sock
*sk
, unsigned char option
,
486 unsigned char len
, u16 idx
,
487 unsigned char *value
)
490 const struct dccp_sock
*dp
= dccp_sk(sk
);
491 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
492 struct ccid3_options_received
*opt_recv
= &hctx
->options_received
;
495 if (opt_recv
->ccid3or_seqno
!= dp
->dccps_gsr
) {
496 opt_recv
->ccid3or_seqno
= dp
->dccps_gsr
;
497 opt_recv
->ccid3or_loss_event_rate
= ~0;
498 opt_recv
->ccid3or_loss_intervals_idx
= 0;
499 opt_recv
->ccid3or_loss_intervals_len
= 0;
500 opt_recv
->ccid3or_receive_rate
= 0;
504 case TFRC_OPT_LOSS_EVENT_RATE
:
505 if (unlikely(len
!= 4)) {
506 DCCP_WARN("%s(%p), invalid len %d "
507 "for TFRC_OPT_LOSS_EVENT_RATE\n",
508 dccp_role(sk
), sk
, len
);
511 opt_val
= get_unaligned((__be32
*)value
);
512 opt_recv
->ccid3or_loss_event_rate
= ntohl(opt_val
);
513 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
515 opt_recv
->ccid3or_loss_event_rate
);
518 case TFRC_OPT_LOSS_INTERVALS
:
519 opt_recv
->ccid3or_loss_intervals_idx
= idx
;
520 opt_recv
->ccid3or_loss_intervals_len
= len
;
521 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
523 opt_recv
->ccid3or_loss_intervals_idx
,
524 opt_recv
->ccid3or_loss_intervals_len
);
526 case TFRC_OPT_RECEIVE_RATE
:
527 if (unlikely(len
!= 4)) {
528 DCCP_WARN("%s(%p), invalid len %d "
529 "for TFRC_OPT_RECEIVE_RATE\n",
530 dccp_role(sk
), sk
, len
);
533 opt_val
= get_unaligned((__be32
*)value
);
534 opt_recv
->ccid3or_receive_rate
= ntohl(opt_val
);
535 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
537 opt_recv
->ccid3or_receive_rate
);
545 static int ccid3_hc_tx_init(struct ccid
*ccid
, struct sock
*sk
)
547 struct ccid3_hc_tx_sock
*hctx
= ccid_priv(ccid
);
549 hctx
->state
= TFRC_SSTATE_NO_SENT
;
551 setup_timer(&hctx
->no_feedback_timer
,
552 ccid3_hc_tx_no_feedback_timer
, (unsigned long)sk
);
556 static void ccid3_hc_tx_exit(struct sock
*sk
)
558 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
560 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_TERM
);
561 sk_stop_timer(sk
, &hctx
->no_feedback_timer
);
563 tfrc_tx_hist_purge(&hctx
->hist
);
566 static void ccid3_hc_tx_get_info(struct sock
*sk
, struct tcp_info
*info
)
568 struct ccid3_hc_tx_sock
*hctx
;
570 /* Listen socks doesn't have a private CCID block */
571 if (sk
->sk_state
== DCCP_LISTEN
)
574 hctx
= ccid3_hc_tx_sk(sk
);
575 info
->tcpi_rto
= hctx
->t_rto
;
576 info
->tcpi_rtt
= hctx
->rtt
;
579 static int ccid3_hc_tx_getsockopt(struct sock
*sk
, const int optname
, int len
,
580 u32 __user
*optval
, int __user
*optlen
)
582 const struct ccid3_hc_tx_sock
*hctx
;
583 struct tfrc_tx_info tfrc
;
586 /* Listen socks doesn't have a private CCID block */
587 if (sk
->sk_state
== DCCP_LISTEN
)
590 hctx
= ccid3_hc_tx_sk(sk
);
592 case DCCP_SOCKOPT_CCID_TX_INFO
:
593 if (len
< sizeof(tfrc
))
595 tfrc
.tfrctx_x
= hctx
->x
;
596 tfrc
.tfrctx_x_recv
= hctx
->x_recv
;
597 tfrc
.tfrctx_x_calc
= hctx
->x_calc
;
598 tfrc
.tfrctx_rtt
= hctx
->rtt
;
599 tfrc
.tfrctx_p
= hctx
->p
;
600 tfrc
.tfrctx_rto
= hctx
->t_rto
;
601 tfrc
.tfrctx_ipi
= hctx
->t_ipi
;
609 if (put_user(len
, optlen
) || copy_to_user(optval
, val
, len
))
616 * Receiver Half-Connection Routines
619 /* CCID3 feedback types */
620 enum ccid3_fback_type
{
621 CCID3_FBACK_NONE
= 0,
623 CCID3_FBACK_PERIODIC
,
624 CCID3_FBACK_PARAM_CHANGE
627 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
628 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state
)
630 static char *ccid3_rx_state_names
[] = {
631 [TFRC_RSTATE_NO_DATA
] = "NO_DATA",
632 [TFRC_RSTATE_DATA
] = "DATA",
633 [TFRC_RSTATE_TERM
] = "TERM",
636 return ccid3_rx_state_names
[state
];
640 static void ccid3_hc_rx_set_state(struct sock
*sk
,
641 enum ccid3_hc_rx_states state
)
643 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
644 enum ccid3_hc_rx_states oldstate
= hcrx
->state
;
646 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
647 dccp_role(sk
), sk
, ccid3_rx_state_name(oldstate
),
648 ccid3_rx_state_name(state
));
649 WARN_ON(state
== oldstate
);
653 static void ccid3_hc_rx_send_feedback(struct sock
*sk
,
654 const struct sk_buff
*skb
,
655 enum ccid3_fback_type fbtype
)
657 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
658 struct dccp_sock
*dp
= dccp_sk(sk
);
662 if (unlikely(hcrx
->state
== TFRC_RSTATE_TERM
))
665 now
= ktime_get_real();
668 case CCID3_FBACK_INITIAL
:
670 hcrx
->p_inverse
= ~0U; /* see RFC 4342, 8.5 */
672 case CCID3_FBACK_PARAM_CHANGE
:
674 * When parameters change (new loss or p > p_prev), we do not
675 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
676 * need to reuse the previous value of X_recv. However, when
677 * X_recv was 0 (due to early loss), this would kill X down to
678 * s/t_mbi (i.e. one packet in 64 seconds).
679 * To avoid such drastic reduction, we approximate X_recv as
680 * the number of bytes since last feedback.
681 * This is a safe fallback, since X is bounded above by X_calc.
683 if (hcrx
->x_recv
> 0)
686 case CCID3_FBACK_PERIODIC
:
687 delta
= ktime_us_delta(now
, hcrx
->tstamp_last_feedback
);
689 DCCP_BUG("delta (%ld) <= 0", (long)delta
);
691 hcrx
->x_recv
= scaled_div32(hcrx
->bytes_recv
, delta
);
697 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n",
698 (long)delta
, hcrx
->x_recv
, hcrx
->p_inverse
);
700 hcrx
->tstamp_last_feedback
= now
;
701 hcrx
->last_counter
= dccp_hdr(skb
)->dccph_ccval
;
702 hcrx
->bytes_recv
= 0;
704 dp
->dccps_hc_rx_insert_options
= 1;
708 static int ccid3_hc_rx_insert_options(struct sock
*sk
, struct sk_buff
*skb
)
710 const struct ccid3_hc_rx_sock
*hcrx
;
713 if (!(sk
->sk_state
== DCCP_OPEN
|| sk
->sk_state
== DCCP_PARTOPEN
))
716 hcrx
= ccid3_hc_rx_sk(sk
);
718 if (dccp_packet_without_ack(skb
))
721 x_recv
= htonl(hcrx
->x_recv
);
722 pinv
= htonl(hcrx
->p_inverse
);
724 if (dccp_insert_option(sk
, skb
, TFRC_OPT_LOSS_EVENT_RATE
,
725 &pinv
, sizeof(pinv
)) ||
726 dccp_insert_option(sk
, skb
, TFRC_OPT_RECEIVE_RATE
,
727 &x_recv
, sizeof(x_recv
)))
733 /** ccid3_first_li - Implements [RFC 3448, 6.3.1]
735 * Determine the length of the first loss interval via inverse lookup.
736 * Assume that X_recv can be computed by the throughput equation
740 * Find some p such that f(p) = fval; return 1/p (scaled).
742 static u32
ccid3_first_li(struct sock
*sk
)
744 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
745 u32 x_recv
, p
, delta
;
748 if (hcrx
->rtt
== 0) {
749 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
750 hcrx
->rtt
= DCCP_FALLBACK_RTT
;
753 delta
= ktime_to_us(net_timedelta(hcrx
->tstamp_last_feedback
));
754 x_recv
= scaled_div32(hcrx
->bytes_recv
, delta
);
755 if (x_recv
== 0) { /* would also trigger divide-by-zero */
756 DCCP_WARN("X_recv==0\n");
757 if (hcrx
->x_recv
== 0) {
758 DCCP_BUG("stored value of X_recv is zero");
761 x_recv
= hcrx
->x_recv
;
764 fval
= scaled_div(hcrx
->s
, hcrx
->rtt
);
765 fval
= scaled_div32(fval
, x_recv
);
766 p
= tfrc_calc_x_reverse_lookup(fval
);
768 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
769 "loss rate=%u\n", dccp_role(sk
), sk
, x_recv
, p
);
771 return p
== 0 ? ~0U : scaled_div(1, p
);
774 static void ccid3_hc_rx_packet_recv(struct sock
*sk
, struct sk_buff
*skb
)
776 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
777 enum ccid3_fback_type do_feedback
= CCID3_FBACK_NONE
;
778 const u64 ndp
= dccp_sk(sk
)->dccps_options_received
.dccpor_ndp
;
779 const bool is_data_packet
= dccp_data_packet(skb
);
781 if (unlikely(hcrx
->state
== TFRC_RSTATE_NO_DATA
)) {
782 if (is_data_packet
) {
783 const u32 payload
= skb
->len
- dccp_hdr(skb
)->dccph_doff
* 4;
784 do_feedback
= CCID3_FBACK_INITIAL
;
785 ccid3_hc_rx_set_state(sk
, TFRC_RSTATE_DATA
);
788 * Not necessary to update bytes_recv here,
789 * since X_recv = 0 for the first feedback packet (cf.
790 * RFC 3448, 6.3) -- gerrit
796 if (tfrc_rx_hist_duplicate(&hcrx
->hist
, skb
))
797 return; /* done receiving */
799 if (is_data_packet
) {
800 const u32 payload
= skb
->len
- dccp_hdr(skb
)->dccph_doff
* 4;
802 * Update moving-average of s and the sum of received payload bytes
804 hcrx
->s
= tfrc_ewma(hcrx
->s
, payload
, 9);
805 hcrx
->bytes_recv
+= payload
;
809 * Perform loss detection and handle pending losses
811 if (tfrc_rx_handle_loss(&hcrx
->hist
, &hcrx
->li_hist
,
812 skb
, ndp
, ccid3_first_li
, sk
)) {
813 do_feedback
= CCID3_FBACK_PARAM_CHANGE
;
817 if (tfrc_rx_hist_loss_pending(&hcrx
->hist
))
818 return; /* done receiving */
821 * Handle data packets: RTT sampling and monitoring p
823 if (unlikely(!is_data_packet
))
826 if (!tfrc_lh_is_initialised(&hcrx
->li_hist
)) {
827 const u32 sample
= tfrc_rx_hist_sample_rtt(&hcrx
->hist
, skb
);
829 * Empty loss history: no loss so far, hence p stays 0.
830 * Sample RTT values, since an RTT estimate is required for the
831 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
834 hcrx
->rtt
= tfrc_ewma(hcrx
->rtt
, sample
, 9);
836 } else if (tfrc_lh_update_i_mean(&hcrx
->li_hist
, skb
)) {
838 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
839 * has decreased (resp. p has increased), send feedback now.
841 do_feedback
= CCID3_FBACK_PARAM_CHANGE
;
845 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
847 if (SUB16(dccp_hdr(skb
)->dccph_ccval
, hcrx
->last_counter
) > 3)
848 do_feedback
= CCID3_FBACK_PERIODIC
;
851 tfrc_rx_hist_add_packet(&hcrx
->hist
, skb
, ndp
);
855 ccid3_hc_rx_send_feedback(sk
, skb
, do_feedback
);
858 static int ccid3_hc_rx_init(struct ccid
*ccid
, struct sock
*sk
)
860 struct ccid3_hc_rx_sock
*hcrx
= ccid_priv(ccid
);
862 hcrx
->state
= TFRC_RSTATE_NO_DATA
;
863 tfrc_lh_init(&hcrx
->li_hist
);
864 return tfrc_rx_hist_alloc(&hcrx
->hist
);
867 static void ccid3_hc_rx_exit(struct sock
*sk
)
869 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
871 ccid3_hc_rx_set_state(sk
, TFRC_RSTATE_TERM
);
873 tfrc_rx_hist_purge(&hcrx
->hist
);
874 tfrc_lh_cleanup(&hcrx
->li_hist
);
877 static void ccid3_hc_rx_get_info(struct sock
*sk
, struct tcp_info
*info
)
879 const struct ccid3_hc_rx_sock
*hcrx
;
881 /* Listen socks doesn't have a private CCID block */
882 if (sk
->sk_state
== DCCP_LISTEN
)
885 hcrx
= ccid3_hc_rx_sk(sk
);
886 info
->tcpi_ca_state
= hcrx
->state
;
887 info
->tcpi_options
|= TCPI_OPT_TIMESTAMPS
;
888 info
->tcpi_rcv_rtt
= hcrx
->rtt
;
891 static int ccid3_hc_rx_getsockopt(struct sock
*sk
, const int optname
, int len
,
892 u32 __user
*optval
, int __user
*optlen
)
894 const struct ccid3_hc_rx_sock
*hcrx
;
895 struct tfrc_rx_info rx_info
;
898 /* Listen socks doesn't have a private CCID block */
899 if (sk
->sk_state
== DCCP_LISTEN
)
902 hcrx
= ccid3_hc_rx_sk(sk
);
904 case DCCP_SOCKOPT_CCID_RX_INFO
:
905 if (len
< sizeof(rx_info
))
907 rx_info
.tfrcrx_x_recv
= hcrx
->x_recv
;
908 rx_info
.tfrcrx_rtt
= hcrx
->rtt
;
909 rx_info
.tfrcrx_p
= hcrx
->p_inverse
== 0 ? ~0U :
910 scaled_div(1, hcrx
->p_inverse
);
911 len
= sizeof(rx_info
);
918 if (put_user(len
, optlen
) || copy_to_user(optval
, val
, len
))
924 static struct ccid_operations ccid3
= {
925 .ccid_id
= DCCPC_CCID3
,
926 .ccid_name
= "TCP-Friendly Rate Control",
927 .ccid_owner
= THIS_MODULE
,
928 .ccid_hc_tx_obj_size
= sizeof(struct ccid3_hc_tx_sock
),
929 .ccid_hc_tx_init
= ccid3_hc_tx_init
,
930 .ccid_hc_tx_exit
= ccid3_hc_tx_exit
,
931 .ccid_hc_tx_send_packet
= ccid3_hc_tx_send_packet
,
932 .ccid_hc_tx_packet_sent
= ccid3_hc_tx_packet_sent
,
933 .ccid_hc_tx_packet_recv
= ccid3_hc_tx_packet_recv
,
934 .ccid_hc_tx_parse_options
= ccid3_hc_tx_parse_options
,
935 .ccid_hc_rx_obj_size
= sizeof(struct ccid3_hc_rx_sock
),
936 .ccid_hc_rx_init
= ccid3_hc_rx_init
,
937 .ccid_hc_rx_exit
= ccid3_hc_rx_exit
,
938 .ccid_hc_rx_insert_options
= ccid3_hc_rx_insert_options
,
939 .ccid_hc_rx_packet_recv
= ccid3_hc_rx_packet_recv
,
940 .ccid_hc_rx_get_info
= ccid3_hc_rx_get_info
,
941 .ccid_hc_tx_get_info
= ccid3_hc_tx_get_info
,
942 .ccid_hc_rx_getsockopt
= ccid3_hc_rx_getsockopt
,
943 .ccid_hc_tx_getsockopt
= ccid3_hc_tx_getsockopt
,
946 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
947 module_param(ccid3_debug
, bool, 0644);
948 MODULE_PARM_DESC(ccid3_debug
, "Enable debug messages");
951 static __init
int ccid3_module_init(void)
953 return ccid_register(&ccid3
);
955 module_init(ccid3_module_init
);
957 static __exit
void ccid3_module_exit(void)
959 ccid_unregister(&ccid3
);
961 module_exit(ccid3_module_exit
);
963 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
964 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
965 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
966 MODULE_LICENSE("GPL");
967 MODULE_ALIAS("net-dccp-ccid-3");