2 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
3 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
4 * Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
6 * An implementation of the DCCP protocol
8 * This code has been developed by the University of Waikato WAND
9 * research group. For further information please see http://www.wand.net.nz/
11 * This code also uses code from Lulea University, rereleased as GPL by its
13 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
15 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
16 * and to make it work as a loadable module in the DCCP stack written by
17 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
19 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
21 * This program is free software; you can redistribute it and/or modify
22 * it under the terms of the GNU General Public License as published by
23 * the Free Software Foundation; either version 2 of the License, or
24 * (at your option) any later version.
26 * This program is distributed in the hope that it will be useful,
27 * but WITHOUT ANY WARRANTY; without even the implied warranty of
28 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
29 * GNU General Public License for more details.
31 * You should have received a copy of the GNU General Public License
32 * along with this program; if not, write to the Free Software
33 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
38 #include <asm/unaligned.h>
40 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
41 static int ccid3_debug
;
42 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
44 #define ccid3_pr_debug(format, a...)
48 * Transmitter Half-Connection Routines
50 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
51 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state
)
53 static const char *const ccid3_state_names
[] = {
54 [TFRC_SSTATE_NO_SENT
] = "NO_SENT",
55 [TFRC_SSTATE_NO_FBACK
] = "NO_FBACK",
56 [TFRC_SSTATE_FBACK
] = "FBACK",
57 [TFRC_SSTATE_TERM
] = "TERM",
60 return ccid3_state_names
[state
];
64 static void ccid3_hc_tx_set_state(struct sock
*sk
,
65 enum ccid3_hc_tx_states state
)
67 struct ccid3_hc_tx_sock
*hc
= ccid3_hc_tx_sk(sk
);
68 enum ccid3_hc_tx_states oldstate
= hc
->tx_state
;
70 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
71 dccp_role(sk
), sk
, ccid3_tx_state_name(oldstate
),
72 ccid3_tx_state_name(state
));
73 WARN_ON(state
== oldstate
);
78 * Compute the initial sending rate X_init in the manner of RFC 3390:
80 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
82 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
83 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
84 * For consistency with other parts of the code, X_init is scaled by 2^6.
86 static inline u64
rfc3390_initial_rate(struct sock
*sk
)
88 const struct ccid3_hc_tx_sock
*hc
= ccid3_hc_tx_sk(sk
);
89 const __u32 w_init
= clamp_t(__u32
, 4380U, 2 * hc
->tx_s
, 4 * hc
->tx_s
);
91 return scaled_div(w_init
<< 6, hc
->tx_rtt
);
95 * Recalculate t_ipi and delta (should be called whenever X changes)
97 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock
*hc
)
99 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
100 hc
->tx_t_ipi
= scaled_div32(((u64
)hc
->tx_s
) << 6, hc
->tx_x
);
102 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
103 hc
->tx_delta
= min_t(u32
, hc
->tx_t_ipi
/ 2, TFRC_OPSYS_HALF_TIME_GRAN
);
105 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n", hc
->tx_t_ipi
,
106 hc
->tx_delta
, hc
->tx_s
, (unsigned)(hc
->tx_x
>> 6));
109 static u32
ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock
*hc
, ktime_t now
)
111 u32 delta
= ktime_us_delta(now
, hc
->tx_t_last_win_count
);
113 return delta
/ hc
->tx_rtt
;
117 * ccid3_hc_tx_update_x - Update allowed sending rate X
118 * @stamp: most recent time if available - can be left NULL.
119 * This function tracks draft rfc3448bis, check there for latest details.
121 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
122 * fine-grained resolution of sending rates. This requires scaling by 2^6
123 * throughout the code. Only X_calc is unscaled (in bytes/second).
126 static void ccid3_hc_tx_update_x(struct sock
*sk
, ktime_t
*stamp
)
128 struct ccid3_hc_tx_sock
*hc
= ccid3_hc_tx_sk(sk
);
129 __u64 min_rate
= 2 * hc
->tx_x_recv
;
130 const __u64 old_x
= hc
->tx_x
;
131 ktime_t now
= stamp
? *stamp
: ktime_get_real();
134 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
135 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
136 * a sender is idle if it has not sent anything over a 2-RTT-period.
137 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
139 if (ccid3_hc_tx_idle_rtt(hc
, now
) >= 2) {
140 min_rate
= rfc3390_initial_rate(sk
);
141 min_rate
= max(min_rate
, 2 * hc
->tx_x_recv
);
146 hc
->tx_x
= min(((__u64
)hc
->tx_x_calc
) << 6, min_rate
);
147 hc
->tx_x
= max(hc
->tx_x
, (((__u64
)hc
->tx_s
) << 6) / TFRC_T_MBI
);
149 } else if (ktime_us_delta(now
, hc
->tx_t_ld
) - (s64
)hc
->tx_rtt
>= 0) {
151 hc
->tx_x
= min(2 * hc
->tx_x
, min_rate
);
152 hc
->tx_x
= max(hc
->tx_x
,
153 scaled_div(((__u64
)hc
->tx_s
) << 6, hc
->tx_rtt
));
157 if (hc
->tx_x
!= old_x
) {
158 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
159 "X_recv=%u\n", (unsigned)(old_x
>> 6),
160 (unsigned)(hc
->tx_x
>> 6), hc
->tx_x_calc
,
161 (unsigned)(hc
->tx_x_recv
>> 6));
163 ccid3_update_send_interval(hc
);
168 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
169 * @len: DCCP packet payload size in bytes
171 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock
*hc
, int len
)
173 const u16 old_s
= hc
->tx_s
;
175 hc
->tx_s
= tfrc_ewma(hc
->tx_s
, len
, 9);
177 if (hc
->tx_s
!= old_s
)
178 ccid3_update_send_interval(hc
);
182 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
183 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
185 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock
*hc
,
188 u32 delta
= ktime_us_delta(now
, hc
->tx_t_last_win_count
),
189 quarter_rtts
= (4 * delta
) / hc
->tx_rtt
;
191 if (quarter_rtts
> 0) {
192 hc
->tx_t_last_win_count
= now
;
193 hc
->tx_last_win_count
+= min(quarter_rtts
, 5U);
194 hc
->tx_last_win_count
&= 0xF; /* mod 16 */
198 static void ccid3_hc_tx_no_feedback_timer(unsigned long data
)
200 struct sock
*sk
= (struct sock
*)data
;
201 struct ccid3_hc_tx_sock
*hc
= ccid3_hc_tx_sk(sk
);
202 unsigned long t_nfb
= USEC_PER_SEC
/ 5;
205 if (sock_owned_by_user(sk
)) {
206 /* Try again later. */
210 ccid3_pr_debug("%s(%p, state=%s) - entry\n", dccp_role(sk
), sk
,
211 ccid3_tx_state_name(hc
->tx_state
));
213 if (hc
->tx_state
== TFRC_SSTATE_FBACK
)
214 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_NO_FBACK
);
215 else if (hc
->tx_state
!= TFRC_SSTATE_NO_FBACK
)
219 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
221 if (hc
->tx_t_rto
== 0 || /* no feedback received yet */
224 /* halve send rate directly */
225 hc
->tx_x
= max(hc
->tx_x
/ 2,
226 (((__u64
)hc
->tx_s
) << 6) / TFRC_T_MBI
);
227 ccid3_update_send_interval(hc
);
230 * Modify the cached value of X_recv
232 * If (X_calc > 2 * X_recv)
233 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
235 * X_recv = X_calc / 4;
237 * Note that X_recv is scaled by 2^6 while X_calc is not
239 BUG_ON(hc
->tx_p
&& !hc
->tx_x_calc
);
241 if (hc
->tx_x_calc
> (hc
->tx_x_recv
>> 5))
243 max(hc
->tx_x_recv
/ 2,
244 (((__u64
)hc
->tx_s
) << 6) / (2*TFRC_T_MBI
));
246 hc
->tx_x_recv
= hc
->tx_x_calc
;
249 ccid3_hc_tx_update_x(sk
, NULL
);
251 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
252 (unsigned long long)hc
->tx_x
);
255 * Set new timeout for the nofeedback timer.
256 * See comments in packet_recv() regarding the value of t_RTO.
258 if (unlikely(hc
->tx_t_rto
== 0)) /* no feedback yet */
259 t_nfb
= TFRC_INITIAL_TIMEOUT
;
261 t_nfb
= max(hc
->tx_t_rto
, 2 * hc
->tx_t_ipi
);
264 sk_reset_timer(sk
, &hc
->tx_no_feedback_timer
,
265 jiffies
+ usecs_to_jiffies(t_nfb
));
273 * > 0: delay (in msecs) that should pass before actually sending
274 * = 0: can send immediately
275 * < 0: error condition; do not send packet
277 static int ccid3_hc_tx_send_packet(struct sock
*sk
, struct sk_buff
*skb
)
279 struct dccp_sock
*dp
= dccp_sk(sk
);
280 struct ccid3_hc_tx_sock
*hc
= ccid3_hc_tx_sk(sk
);
281 ktime_t now
= ktime_get_real();
285 * This function is called only for Data and DataAck packets. Sending
286 * zero-sized Data(Ack)s is theoretically possible, but for congestion
287 * control this case is pathological - ignore it.
289 if (unlikely(skb
->len
== 0))
292 switch (hc
->tx_state
) {
293 case TFRC_SSTATE_NO_SENT
:
294 sk_reset_timer(sk
, &hc
->tx_no_feedback_timer
, (jiffies
+
295 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT
)));
296 hc
->tx_last_win_count
= 0;
297 hc
->tx_t_last_win_count
= now
;
299 /* Set t_0 for initial packet */
305 * Use initial RTT sample when available: recommended by erratum
306 * to RFC 4342. This implements the initialisation procedure of
307 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
309 if (dp
->dccps_syn_rtt
) {
310 ccid3_pr_debug("SYN RTT = %uus\n", dp
->dccps_syn_rtt
);
311 hc
->tx_rtt
= dp
->dccps_syn_rtt
;
312 hc
->tx_x
= rfc3390_initial_rate(sk
);
316 * Sender does not have RTT sample:
317 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
318 * is needed in several parts (e.g. window counter);
319 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
321 hc
->tx_rtt
= DCCP_FALLBACK_RTT
;
325 ccid3_update_send_interval(hc
);
327 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_NO_FBACK
);
329 case TFRC_SSTATE_NO_FBACK
:
330 case TFRC_SSTATE_FBACK
:
331 delay
= ktime_us_delta(hc
->tx_t_nom
, now
);
332 ccid3_pr_debug("delay=%ld\n", (long)delay
);
334 * Scheduling of packet transmissions [RFC 3448, 4.6]
336 * if (t_now > t_nom - delta)
337 * // send the packet now
339 * // send the packet in (t_nom - t_now) milliseconds.
341 if (delay
- (s64
)hc
->tx_delta
>= 1000)
342 return (u32
)delay
/ 1000L;
344 ccid3_hc_tx_update_win_count(hc
, now
);
346 case TFRC_SSTATE_TERM
:
347 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk
), sk
);
351 /* prepare to send now (add options etc.) */
352 dp
->dccps_hc_tx_insert_options
= 1;
353 DCCP_SKB_CB(skb
)->dccpd_ccval
= hc
->tx_last_win_count
;
355 /* set the nominal send time for the next following packet */
356 hc
->tx_t_nom
= ktime_add_us(hc
->tx_t_nom
, hc
->tx_t_ipi
);
360 static void ccid3_hc_tx_packet_sent(struct sock
*sk
, int more
,
363 struct ccid3_hc_tx_sock
*hc
= ccid3_hc_tx_sk(sk
);
365 ccid3_hc_tx_update_s(hc
, len
);
367 if (tfrc_tx_hist_add(&hc
->tx_hist
, dccp_sk(sk
)->dccps_gss
))
368 DCCP_CRIT("packet history - out of memory!");
371 static void ccid3_hc_tx_packet_recv(struct sock
*sk
, struct sk_buff
*skb
)
373 struct ccid3_hc_tx_sock
*hc
= ccid3_hc_tx_sk(sk
);
374 struct ccid3_options_received
*opt_recv
;
379 /* we are only interested in ACKs */
380 if (!(DCCP_SKB_CB(skb
)->dccpd_type
== DCCP_PKT_ACK
||
381 DCCP_SKB_CB(skb
)->dccpd_type
== DCCP_PKT_DATAACK
))
383 /* ... and only in the established state */
384 if (hc
->tx_state
!= TFRC_SSTATE_FBACK
&&
385 hc
->tx_state
!= TFRC_SSTATE_NO_FBACK
)
388 opt_recv
= &hc
->tx_options_received
;
389 now
= ktime_get_real();
391 /* Estimate RTT from history if ACK number is valid */
392 r_sample
= tfrc_tx_hist_rtt(hc
->tx_hist
,
393 DCCP_SKB_CB(skb
)->dccpd_ack_seq
, now
);
395 DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk
), sk
,
396 dccp_packet_name(DCCP_SKB_CB(skb
)->dccpd_type
),
397 (unsigned long long)DCCP_SKB_CB(skb
)->dccpd_ack_seq
);
401 /* Update receive rate in units of 64 * bytes/second */
402 hc
->tx_x_recv
= opt_recv
->ccid3or_receive_rate
;
405 /* Update loss event rate (which is scaled by 1e6) */
406 pinv
= opt_recv
->ccid3or_loss_event_rate
;
407 if (pinv
== ~0U || pinv
== 0) /* see RFC 4342, 8.5 */
409 else /* can not exceed 100% */
410 hc
->tx_p
= scaled_div(1, pinv
);
412 * Validate new RTT sample and update moving average
414 r_sample
= dccp_sample_rtt(sk
, r_sample
);
415 hc
->tx_rtt
= tfrc_ewma(hc
->tx_rtt
, r_sample
, 9);
417 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
419 if (hc
->tx_state
== TFRC_SSTATE_NO_FBACK
) {
420 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_FBACK
);
422 if (hc
->tx_t_rto
== 0) {
424 * Initial feedback packet: Larger Initial Windows (4.2)
426 hc
->tx_x
= rfc3390_initial_rate(sk
);
429 ccid3_update_send_interval(hc
);
431 goto done_computing_x
;
432 } else if (hc
->tx_p
== 0) {
434 * First feedback after nofeedback timer expiry (4.3)
436 goto done_computing_x
;
440 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
442 hc
->tx_x_calc
= tfrc_calc_x(hc
->tx_s
, hc
->tx_rtt
, hc
->tx_p
);
443 ccid3_hc_tx_update_x(sk
, &now
);
446 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
447 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
448 dccp_role(sk
), sk
, hc
->tx_rtt
, r_sample
,
449 hc
->tx_s
, hc
->tx_p
, hc
->tx_x_calc
,
450 (unsigned)(hc
->tx_x_recv
>> 6),
451 (unsigned)(hc
->tx_x
>> 6));
453 /* unschedule no feedback timer */
454 sk_stop_timer(sk
, &hc
->tx_no_feedback_timer
);
457 * As we have calculated new ipi, delta, t_nom it is possible
458 * that we now can send a packet, so wake up dccp_wait_for_ccid
460 sk
->sk_write_space(sk
);
463 * Update timeout interval for the nofeedback timer.
464 * We use a configuration option to increase the lower bound.
465 * This can help avoid triggering the nofeedback timer too
466 * often ('spinning') on LANs with small RTTs.
468 hc
->tx_t_rto
= max_t(u32
, 4 * hc
->tx_rtt
, (CONFIG_IP_DCCP_CCID3_RTO
*
469 (USEC_PER_SEC
/ 1000)));
471 * Schedule no feedback timer to expire in
472 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
474 t_nfb
= max(hc
->tx_t_rto
, 2 * hc
->tx_t_ipi
);
476 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
477 "expire in %lu jiffies (%luus)\n",
478 dccp_role(sk
), sk
, usecs_to_jiffies(t_nfb
), t_nfb
);
480 sk_reset_timer(sk
, &hc
->tx_no_feedback_timer
,
481 jiffies
+ usecs_to_jiffies(t_nfb
));
484 static int ccid3_hc_tx_parse_options(struct sock
*sk
, unsigned char option
,
485 unsigned char len
, u16 idx
,
486 unsigned char *value
)
489 const struct dccp_sock
*dp
= dccp_sk(sk
);
490 struct ccid3_hc_tx_sock
*hc
= ccid3_hc_tx_sk(sk
);
491 struct ccid3_options_received
*opt_recv
;
494 opt_recv
= &hc
->tx_options_received
;
496 if (opt_recv
->ccid3or_seqno
!= dp
->dccps_gsr
) {
497 opt_recv
->ccid3or_seqno
= dp
->dccps_gsr
;
498 opt_recv
->ccid3or_loss_event_rate
= ~0;
499 opt_recv
->ccid3or_loss_intervals_idx
= 0;
500 opt_recv
->ccid3or_loss_intervals_len
= 0;
501 opt_recv
->ccid3or_receive_rate
= 0;
505 case TFRC_OPT_LOSS_EVENT_RATE
:
506 if (unlikely(len
!= 4)) {
507 DCCP_WARN("%s(%p), invalid len %d "
508 "for TFRC_OPT_LOSS_EVENT_RATE\n",
509 dccp_role(sk
), sk
, len
);
512 opt_val
= get_unaligned((__be32
*)value
);
513 opt_recv
->ccid3or_loss_event_rate
= ntohl(opt_val
);
514 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
516 opt_recv
->ccid3or_loss_event_rate
);
519 case TFRC_OPT_LOSS_INTERVALS
:
520 opt_recv
->ccid3or_loss_intervals_idx
= idx
;
521 opt_recv
->ccid3or_loss_intervals_len
= len
;
522 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
524 opt_recv
->ccid3or_loss_intervals_idx
,
525 opt_recv
->ccid3or_loss_intervals_len
);
527 case TFRC_OPT_RECEIVE_RATE
:
528 if (unlikely(len
!= 4)) {
529 DCCP_WARN("%s(%p), invalid len %d "
530 "for TFRC_OPT_RECEIVE_RATE\n",
531 dccp_role(sk
), sk
, len
);
534 opt_val
= get_unaligned((__be32
*)value
);
535 opt_recv
->ccid3or_receive_rate
= ntohl(opt_val
);
536 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
538 opt_recv
->ccid3or_receive_rate
);
546 static int ccid3_hc_tx_init(struct ccid
*ccid
, struct sock
*sk
)
548 struct ccid3_hc_tx_sock
*hc
= ccid_priv(ccid
);
550 hc
->tx_state
= TFRC_SSTATE_NO_SENT
;
552 setup_timer(&hc
->tx_no_feedback_timer
,
553 ccid3_hc_tx_no_feedback_timer
, (unsigned long)sk
);
557 static void ccid3_hc_tx_exit(struct sock
*sk
)
559 struct ccid3_hc_tx_sock
*hc
= ccid3_hc_tx_sk(sk
);
561 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_TERM
);
562 sk_stop_timer(sk
, &hc
->tx_no_feedback_timer
);
564 tfrc_tx_hist_purge(&hc
->tx_hist
);
567 static void ccid3_hc_tx_get_info(struct sock
*sk
, struct tcp_info
*info
)
569 struct ccid3_hc_tx_sock
*hc
;
571 /* Listen socks doesn't have a private CCID block */
572 if (sk
->sk_state
== DCCP_LISTEN
)
575 hc
= ccid3_hc_tx_sk(sk
);
576 info
->tcpi_rto
= hc
->tx_t_rto
;
577 info
->tcpi_rtt
= hc
->tx_rtt
;
580 static int ccid3_hc_tx_getsockopt(struct sock
*sk
, const int optname
, int len
,
581 u32 __user
*optval
, int __user
*optlen
)
583 const struct ccid3_hc_tx_sock
*hc
;
586 /* Listen socks doesn't have a private CCID block */
587 if (sk
->sk_state
== DCCP_LISTEN
)
590 hc
= ccid3_hc_tx_sk(sk
);
592 case DCCP_SOCKOPT_CCID_TX_INFO
:
593 if (len
< sizeof(hc
->tx_tfrc
))
595 len
= sizeof(hc
->tx_tfrc
);
602 if (put_user(len
, optlen
) || copy_to_user(optval
, val
, len
))
609 * Receiver Half-Connection Routines
612 /* CCID3 feedback types */
613 enum ccid3_fback_type
{
614 CCID3_FBACK_NONE
= 0,
616 CCID3_FBACK_PERIODIC
,
617 CCID3_FBACK_PARAM_CHANGE
620 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
621 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state
)
623 static const char *const ccid3_rx_state_names
[] = {
624 [TFRC_RSTATE_NO_DATA
] = "NO_DATA",
625 [TFRC_RSTATE_DATA
] = "DATA",
626 [TFRC_RSTATE_TERM
] = "TERM",
629 return ccid3_rx_state_names
[state
];
633 static void ccid3_hc_rx_set_state(struct sock
*sk
,
634 enum ccid3_hc_rx_states state
)
636 struct ccid3_hc_rx_sock
*hc
= ccid3_hc_rx_sk(sk
);
637 enum ccid3_hc_rx_states oldstate
= hc
->rx_state
;
639 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
640 dccp_role(sk
), sk
, ccid3_rx_state_name(oldstate
),
641 ccid3_rx_state_name(state
));
642 WARN_ON(state
== oldstate
);
643 hc
->rx_state
= state
;
646 static void ccid3_hc_rx_send_feedback(struct sock
*sk
,
647 const struct sk_buff
*skb
,
648 enum ccid3_fback_type fbtype
)
650 struct ccid3_hc_rx_sock
*hc
= ccid3_hc_rx_sk(sk
);
651 struct dccp_sock
*dp
= dccp_sk(sk
);
655 if (unlikely(hc
->rx_state
== TFRC_RSTATE_TERM
))
658 now
= ktime_get_real();
661 case CCID3_FBACK_INITIAL
:
663 hc
->rx_pinv
= ~0U; /* see RFC 4342, 8.5 */
665 case CCID3_FBACK_PARAM_CHANGE
:
667 * When parameters change (new loss or p > p_prev), we do not
668 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
669 * need to reuse the previous value of X_recv. However, when
670 * X_recv was 0 (due to early loss), this would kill X down to
671 * s/t_mbi (i.e. one packet in 64 seconds).
672 * To avoid such drastic reduction, we approximate X_recv as
673 * the number of bytes since last feedback.
674 * This is a safe fallback, since X is bounded above by X_calc.
676 if (hc
->rx_x_recv
> 0)
679 case CCID3_FBACK_PERIODIC
:
680 delta
= ktime_us_delta(now
, hc
->rx_tstamp_last_feedback
);
682 DCCP_BUG("delta (%ld) <= 0", (long)delta
);
684 hc
->rx_x_recv
= scaled_div32(hc
->rx_bytes_recv
, delta
);
690 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta
,
691 hc
->rx_x_recv
, hc
->rx_pinv
);
693 hc
->rx_tstamp_last_feedback
= now
;
694 hc
->rx_last_counter
= dccp_hdr(skb
)->dccph_ccval
;
695 hc
->rx_bytes_recv
= 0;
697 dp
->dccps_hc_rx_insert_options
= 1;
701 static int ccid3_hc_rx_insert_options(struct sock
*sk
, struct sk_buff
*skb
)
703 const struct ccid3_hc_rx_sock
*hc
;
706 if (!(sk
->sk_state
== DCCP_OPEN
|| sk
->sk_state
== DCCP_PARTOPEN
))
709 hc
= ccid3_hc_rx_sk(sk
);
711 if (dccp_packet_without_ack(skb
))
714 x_recv
= htonl(hc
->rx_x_recv
);
715 pinv
= htonl(hc
->rx_pinv
);
717 if (dccp_insert_option(skb
, TFRC_OPT_LOSS_EVENT_RATE
,
718 &pinv
, sizeof(pinv
)) ||
719 dccp_insert_option(skb
, TFRC_OPT_RECEIVE_RATE
,
720 &x_recv
, sizeof(x_recv
)))
727 * ccid3_first_li - Implements [RFC 5348, 6.3.1]
729 * Determine the length of the first loss interval via inverse lookup.
730 * Assume that X_recv can be computed by the throughput equation
734 * Find some p such that f(p) = fval; return 1/p (scaled).
736 static u32
ccid3_first_li(struct sock
*sk
)
738 struct ccid3_hc_rx_sock
*hc
= ccid3_hc_rx_sk(sk
);
739 u32 x_recv
, p
, delta
;
742 if (hc
->rx_rtt
== 0) {
743 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
744 hc
->rx_rtt
= DCCP_FALLBACK_RTT
;
747 delta
= ktime_to_us(net_timedelta(hc
->rx_tstamp_last_feedback
));
748 x_recv
= scaled_div32(hc
->rx_bytes_recv
, delta
);
749 if (x_recv
== 0) { /* would also trigger divide-by-zero */
750 DCCP_WARN("X_recv==0\n");
751 if ((x_recv
= hc
->rx_x_recv
) == 0) {
752 DCCP_BUG("stored value of X_recv is zero");
757 fval
= scaled_div(hc
->rx_s
, hc
->rx_rtt
);
758 fval
= scaled_div32(fval
, x_recv
);
759 p
= tfrc_calc_x_reverse_lookup(fval
);
761 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
762 "loss rate=%u\n", dccp_role(sk
), sk
, x_recv
, p
);
764 return p
== 0 ? ~0U : scaled_div(1, p
);
767 static void ccid3_hc_rx_packet_recv(struct sock
*sk
, struct sk_buff
*skb
)
769 struct ccid3_hc_rx_sock
*hc
= ccid3_hc_rx_sk(sk
);
770 enum ccid3_fback_type do_feedback
= CCID3_FBACK_NONE
;
771 const u64 ndp
= dccp_sk(sk
)->dccps_options_received
.dccpor_ndp
;
772 const bool is_data_packet
= dccp_data_packet(skb
);
774 if (unlikely(hc
->rx_state
== TFRC_RSTATE_NO_DATA
)) {
775 if (is_data_packet
) {
776 const u32 payload
= skb
->len
- dccp_hdr(skb
)->dccph_doff
* 4;
777 do_feedback
= CCID3_FBACK_INITIAL
;
778 ccid3_hc_rx_set_state(sk
, TFRC_RSTATE_DATA
);
781 * Not necessary to update rx_bytes_recv here,
782 * since X_recv = 0 for the first feedback packet (cf.
783 * RFC 3448, 6.3) -- gerrit
789 if (tfrc_rx_hist_duplicate(&hc
->rx_hist
, skb
))
790 return; /* done receiving */
792 if (is_data_packet
) {
793 const u32 payload
= skb
->len
- dccp_hdr(skb
)->dccph_doff
* 4;
795 * Update moving-average of s and the sum of received payload bytes
797 hc
->rx_s
= tfrc_ewma(hc
->rx_s
, payload
, 9);
798 hc
->rx_bytes_recv
+= payload
;
802 * Perform loss detection and handle pending losses
804 if (tfrc_rx_handle_loss(&hc
->rx_hist
, &hc
->rx_li_hist
,
805 skb
, ndp
, ccid3_first_li
, sk
)) {
806 do_feedback
= CCID3_FBACK_PARAM_CHANGE
;
810 if (tfrc_rx_hist_loss_pending(&hc
->rx_hist
))
811 return; /* done receiving */
814 * Handle data packets: RTT sampling and monitoring p
816 if (unlikely(!is_data_packet
))
819 if (!tfrc_lh_is_initialised(&hc
->rx_li_hist
)) {
820 const u32 sample
= tfrc_rx_hist_sample_rtt(&hc
->rx_hist
, skb
);
822 * Empty loss history: no loss so far, hence p stays 0.
823 * Sample RTT values, since an RTT estimate is required for the
824 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
827 hc
->rx_rtt
= tfrc_ewma(hc
->rx_rtt
, sample
, 9);
829 } else if (tfrc_lh_update_i_mean(&hc
->rx_li_hist
, skb
)) {
831 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
832 * has decreased (resp. p has increased), send feedback now.
834 do_feedback
= CCID3_FBACK_PARAM_CHANGE
;
838 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
840 if (SUB16(dccp_hdr(skb
)->dccph_ccval
, hc
->rx_last_counter
) > 3)
841 do_feedback
= CCID3_FBACK_PERIODIC
;
844 tfrc_rx_hist_add_packet(&hc
->rx_hist
, skb
, ndp
);
848 ccid3_hc_rx_send_feedback(sk
, skb
, do_feedback
);
851 static int ccid3_hc_rx_init(struct ccid
*ccid
, struct sock
*sk
)
853 struct ccid3_hc_rx_sock
*hc
= ccid_priv(ccid
);
855 hc
->rx_state
= TFRC_RSTATE_NO_DATA
;
856 tfrc_lh_init(&hc
->rx_li_hist
);
857 return tfrc_rx_hist_alloc(&hc
->rx_hist
);
860 static void ccid3_hc_rx_exit(struct sock
*sk
)
862 struct ccid3_hc_rx_sock
*hc
= ccid3_hc_rx_sk(sk
);
864 ccid3_hc_rx_set_state(sk
, TFRC_RSTATE_TERM
);
866 tfrc_rx_hist_purge(&hc
->rx_hist
);
867 tfrc_lh_cleanup(&hc
->rx_li_hist
);
870 static void ccid3_hc_rx_get_info(struct sock
*sk
, struct tcp_info
*info
)
872 const struct ccid3_hc_rx_sock
*hc
;
874 /* Listen socks doesn't have a private CCID block */
875 if (sk
->sk_state
== DCCP_LISTEN
)
878 hc
= ccid3_hc_rx_sk(sk
);
879 info
->tcpi_ca_state
= hc
->rx_state
;
880 info
->tcpi_options
|= TCPI_OPT_TIMESTAMPS
;
881 info
->tcpi_rcv_rtt
= hc
->rx_rtt
;
884 static int ccid3_hc_rx_getsockopt(struct sock
*sk
, const int optname
, int len
,
885 u32 __user
*optval
, int __user
*optlen
)
887 const struct ccid3_hc_rx_sock
*hc
;
888 struct tfrc_rx_info rx_info
;
891 /* Listen socks doesn't have a private CCID block */
892 if (sk
->sk_state
== DCCP_LISTEN
)
895 hc
= ccid3_hc_rx_sk(sk
);
897 case DCCP_SOCKOPT_CCID_RX_INFO
:
898 if (len
< sizeof(rx_info
))
900 rx_info
.tfrcrx_x_recv
= hc
->rx_x_recv
;
901 rx_info
.tfrcrx_rtt
= hc
->rx_rtt
;
902 rx_info
.tfrcrx_p
= hc
->rx_pinv
== 0 ? ~0U :
903 scaled_div(1, hc
->rx_pinv
);
904 len
= sizeof(rx_info
);
911 if (put_user(len
, optlen
) || copy_to_user(optval
, val
, len
))
917 struct ccid_operations ccid3_ops
= {
918 .ccid_id
= DCCPC_CCID3
,
919 .ccid_name
= "TCP-Friendly Rate Control",
920 .ccid_hc_tx_obj_size
= sizeof(struct ccid3_hc_tx_sock
),
921 .ccid_hc_tx_init
= ccid3_hc_tx_init
,
922 .ccid_hc_tx_exit
= ccid3_hc_tx_exit
,
923 .ccid_hc_tx_send_packet
= ccid3_hc_tx_send_packet
,
924 .ccid_hc_tx_packet_sent
= ccid3_hc_tx_packet_sent
,
925 .ccid_hc_tx_packet_recv
= ccid3_hc_tx_packet_recv
,
926 .ccid_hc_tx_parse_options
= ccid3_hc_tx_parse_options
,
927 .ccid_hc_rx_obj_size
= sizeof(struct ccid3_hc_rx_sock
),
928 .ccid_hc_rx_init
= ccid3_hc_rx_init
,
929 .ccid_hc_rx_exit
= ccid3_hc_rx_exit
,
930 .ccid_hc_rx_insert_options
= ccid3_hc_rx_insert_options
,
931 .ccid_hc_rx_packet_recv
= ccid3_hc_rx_packet_recv
,
932 .ccid_hc_rx_get_info
= ccid3_hc_rx_get_info
,
933 .ccid_hc_tx_get_info
= ccid3_hc_tx_get_info
,
934 .ccid_hc_rx_getsockopt
= ccid3_hc_rx_getsockopt
,
935 .ccid_hc_tx_getsockopt
= ccid3_hc_tx_getsockopt
,
938 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
939 module_param(ccid3_debug
, bool, 0644);
940 MODULE_PARM_DESC(ccid3_debug
, "Enable CCID-3 debug messages");