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
54 * Compute the initial sending rate X_init in the manner of RFC 3390:
56 * X_init = min(4 * MPS, max(2 * MPS, 4380 bytes)) / RTT
58 * For consistency with other parts of the code, X_init is scaled by 2^6.
60 static inline u64
rfc3390_initial_rate(struct sock
*sk
)
62 const u32 mps
= dccp_sk(sk
)->dccps_mss_cache
,
63 w_init
= clamp(4380U, 2 * mps
, 4 * mps
);
65 return scaled_div(w_init
<< 6, ccid3_hc_tx_sk(sk
)->rtt
);
69 * ccid3_update_send_interval - Calculate new t_ipi = s / X_inst
70 * This respects the granularity of X_inst (64 * bytes/second).
72 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock
*hctx
)
74 hctx
->t_ipi
= scaled_div32(((u64
)hctx
->s
) << 6, hctx
->x
);
76 ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hctx
->t_ipi
,
77 hctx
->s
, (unsigned)(hctx
->x
>> 6));
80 static u32
ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock
*hctx
, ktime_t now
)
82 u32 delta
= ktime_us_delta(now
, hctx
->t_last_win_count
);
84 return delta
/ hctx
->rtt
;
88 * ccid3_hc_tx_update_x - Update allowed sending rate X
89 * @stamp: most recent time if available - can be left NULL.
90 * This function tracks draft rfc3448bis, check there for latest details.
92 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
93 * fine-grained resolution of sending rates. This requires scaling by 2^6
94 * throughout the code. Only X_calc is unscaled (in bytes/second).
97 static void ccid3_hc_tx_update_x(struct sock
*sk
, ktime_t
*stamp
)
99 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
100 u64 min_rate
= 2 * hctx
->x_recv
;
101 const u64 old_x
= hctx
->x
;
102 ktime_t now
= stamp
? *stamp
: ktime_get_real();
105 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
106 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
107 * a sender is idle if it has not sent anything over a 2-RTT-period.
108 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
110 if (ccid3_hc_tx_idle_rtt(hctx
, now
) >= 2) {
111 min_rate
= rfc3390_initial_rate(sk
);
112 min_rate
= max(min_rate
, 2 * hctx
->x_recv
);
117 hctx
->x
= min(((u64
)hctx
->x_calc
) << 6, min_rate
);
118 hctx
->x
= max(hctx
->x
, (((u64
)hctx
->s
) << 6) / TFRC_T_MBI
);
120 } else if (ktime_us_delta(now
, hctx
->t_ld
) - (s64
)hctx
->rtt
>= 0) {
122 hctx
->x
= min(2 * hctx
->x
, min_rate
);
123 hctx
->x
= max(hctx
->x
,
124 scaled_div(((u64
)hctx
->s
) << 6, hctx
->rtt
));
128 if (hctx
->x
!= old_x
) {
129 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
130 "X_recv=%u\n", (unsigned)(old_x
>> 6),
131 (unsigned)(hctx
->x
>> 6), hctx
->x_calc
,
132 (unsigned)(hctx
->x_recv
>> 6));
134 ccid3_update_send_interval(hctx
);
139 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
140 * @len: DCCP packet payload size in bytes
142 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock
*hctx
, int len
)
144 const u16 old_s
= hctx
->s
;
146 hctx
->s
= tfrc_ewma(hctx
->s
, len
, 9);
148 if (hctx
->s
!= old_s
)
149 ccid3_update_send_interval(hctx
);
153 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
154 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
156 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock
*hctx
,
159 u32 delta
= ktime_us_delta(now
, hctx
->t_last_win_count
),
160 quarter_rtts
= (4 * delta
) / hctx
->rtt
;
162 if (quarter_rtts
> 0) {
163 hctx
->t_last_win_count
= now
;
164 hctx
->last_win_count
+= min(quarter_rtts
, 5U);
165 hctx
->last_win_count
&= 0xF; /* mod 16 */
169 static void ccid3_hc_tx_no_feedback_timer(unsigned long data
)
171 struct sock
*sk
= (struct sock
*)data
;
172 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
173 unsigned long t_nfb
= USEC_PER_SEC
/ 5;
176 if (sock_owned_by_user(sk
)) {
177 /* Try again later. */
178 /* XXX: set some sensible MIB */
182 ccid3_pr_debug("%s(%p) entry with%s feedback\n", dccp_role(sk
), sk
,
183 hctx
->feedback
? "" : "out");
185 /* Ignore and do not restart after leaving the established state */
186 if ((1 << sk
->sk_state
) & ~(DCCPF_OPEN
| DCCPF_PARTOPEN
))
189 /* Reset feedback state to "no feedback received" */
190 hctx
->feedback
= false;
193 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
194 * RTO is 0 if and only if no feedback has been received yet.
196 if (hctx
->t_rto
== 0 || hctx
->p
== 0) {
198 /* halve send rate directly */
199 hctx
->x
= max(hctx
->x
/ 2, (((u64
)hctx
->s
) << 6) / TFRC_T_MBI
);
200 ccid3_update_send_interval(hctx
);
203 * Modify the cached value of X_recv
205 * If (X_calc > 2 * X_recv)
206 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
208 * X_recv = X_calc / 4;
210 * Note that X_recv is scaled by 2^6 while X_calc is not
212 BUG_ON(hctx
->p
&& !hctx
->x_calc
);
214 if (hctx
->x_calc
> (hctx
->x_recv
>> 5))
216 max(hctx
->x_recv
/ 2,
217 (((__u64
)hctx
->s
) << 6) / (2 * TFRC_T_MBI
));
219 hctx
->x_recv
= hctx
->x_calc
;
222 ccid3_hc_tx_update_x(sk
, NULL
);
224 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
225 (unsigned long long)hctx
->x
);
228 * Set new timeout for the nofeedback timer.
229 * See comments in packet_recv() regarding the value of t_RTO.
231 if (unlikely(hctx
->t_rto
== 0)) /* no feedback received yet */
232 t_nfb
= TFRC_INITIAL_TIMEOUT
;
234 t_nfb
= max(hctx
->t_rto
, 2 * hctx
->t_ipi
);
237 sk_reset_timer(sk
, &hctx
->no_feedback_timer
,
238 jiffies
+ usecs_to_jiffies(t_nfb
));
245 * ccid3_hc_tx_send_packet - Delay-based dequeueing of TX packets
246 * @skb: next packet candidate to send on @sk
247 * This function uses the convention of ccid_packet_dequeue_eval() and
248 * returns a millisecond-delay value between 0 and t_mbi = 64000 msec.
250 static int ccid3_hc_tx_send_packet(struct sock
*sk
, struct sk_buff
*skb
)
252 struct dccp_sock
*dp
= dccp_sk(sk
);
253 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
254 ktime_t now
= ktime_get_real();
258 * This function is called only for Data and DataAck packets. Sending
259 * zero-sized Data(Ack)s is theoretically possible, but for congestion
260 * control this case is pathological - ignore it.
262 if (unlikely(skb
->len
== 0))
266 sk_reset_timer(sk
, &hctx
->no_feedback_timer
, (jiffies
+
267 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT
)));
268 hctx
->last_win_count
= 0;
269 hctx
->t_last_win_count
= now
;
271 /* Set t_0 for initial packet */
277 * Use initial RTT sample when available: recommended by erratum
278 * to RFC 4342. This implements the initialisation procedure of
279 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
281 if (dp
->dccps_syn_rtt
) {
282 ccid3_pr_debug("SYN RTT = %uus\n", dp
->dccps_syn_rtt
);
283 hctx
->rtt
= dp
->dccps_syn_rtt
;
284 hctx
->x
= rfc3390_initial_rate(sk
);
288 * Sender does not have RTT sample:
289 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
290 * is needed in several parts (e.g. window counter);
291 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
293 hctx
->rtt
= DCCP_FALLBACK_RTT
;
294 hctx
->x
= dp
->dccps_mss_cache
;
297 ccid3_update_send_interval(hctx
);
300 delay
= ktime_us_delta(hctx
->t_nom
, now
);
301 ccid3_pr_debug("delay=%ld\n", (long)delay
);
303 * Scheduling of packet transmissions [RFC 3448, 4.6]
305 * if (t_now > t_nom - delta)
306 * // send the packet now
308 * // send the packet in (t_nom - t_now) milliseconds.
310 if (delay
>= TFRC_T_DELTA
)
311 return (u32
)delay
/ USEC_PER_MSEC
;
313 ccid3_hc_tx_update_win_count(hctx
, now
);
316 /* prepare to send now (add options etc.) */
317 dp
->dccps_hc_tx_insert_options
= 1;
318 DCCP_SKB_CB(skb
)->dccpd_ccval
= hctx
->last_win_count
;
320 /* set the nominal send time for the next following packet */
321 hctx
->t_nom
= ktime_add_us(hctx
->t_nom
, hctx
->t_ipi
);
322 return CCID_PACKET_SEND_AT_ONCE
;
325 static void ccid3_hc_tx_packet_sent(struct sock
*sk
, unsigned int len
)
327 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
329 ccid3_hc_tx_update_s(hctx
, len
);
331 if (tfrc_tx_hist_add(&hctx
->hist
, dccp_sk(sk
)->dccps_gss
))
332 DCCP_CRIT("packet history - out of memory!");
335 static void ccid3_hc_tx_packet_recv(struct sock
*sk
, struct sk_buff
*skb
)
337 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
338 struct tfrc_tx_hist_entry
*acked
;
343 /* we are only interested in ACKs */
344 if (!(DCCP_SKB_CB(skb
)->dccpd_type
== DCCP_PKT_ACK
||
345 DCCP_SKB_CB(skb
)->dccpd_type
== DCCP_PKT_DATAACK
))
348 * Locate the acknowledged packet in the TX history.
350 * Returning "entry not found" here can for instance happen when
351 * - the host has not sent out anything (e.g. a passive server),
352 * - the Ack is outdated (packet with higher Ack number was received),
353 * - it is a bogus Ack (for a packet not sent on this connection).
355 acked
= tfrc_tx_hist_find_entry(hctx
->hist
, dccp_hdr_ack_seq(skb
));
358 /* For the sake of RTT sampling, ignore/remove all older entries */
359 tfrc_tx_hist_purge(&acked
->next
);
361 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
362 now
= ktime_get_real();
363 r_sample
= dccp_sample_rtt(sk
, ktime_us_delta(now
, acked
->stamp
));
364 hctx
->rtt
= tfrc_ewma(hctx
->rtt
, r_sample
, 9);
367 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
369 if (!hctx
->feedback
) {
370 hctx
->feedback
= true;
372 if (hctx
->t_rto
== 0) {
374 * Initial feedback packet: Larger Initial Windows (4.2)
376 hctx
->x
= rfc3390_initial_rate(sk
);
379 ccid3_update_send_interval(hctx
);
381 goto done_computing_x
;
382 } else if (hctx
->p
== 0) {
384 * First feedback after nofeedback timer expiry (4.3)
386 goto done_computing_x
;
390 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
392 hctx
->x_calc
= tfrc_calc_x(hctx
->s
, hctx
->rtt
, hctx
->p
);
393 ccid3_hc_tx_update_x(sk
, &now
);
396 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
397 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
398 dccp_role(sk
), sk
, hctx
->rtt
, r_sample
,
399 hctx
->s
, hctx
->p
, hctx
->x_calc
,
400 (unsigned)(hctx
->x_recv
>> 6),
401 (unsigned)(hctx
->x
>> 6));
403 /* unschedule no feedback timer */
404 sk_stop_timer(sk
, &hctx
->no_feedback_timer
);
407 * As we have calculated new ipi, delta, t_nom it is possible
408 * that we now can send a packet, so wake up dccp_wait_for_ccid
410 sk
->sk_write_space(sk
);
413 * Update timeout interval for the nofeedback timer.
414 * We use a configuration option to increase the lower bound.
415 * This can help avoid triggering the nofeedback timer too
416 * often ('spinning') on LANs with small RTTs.
418 hctx
->t_rto
= max_t(u32
, 4 * hctx
->rtt
, (CONFIG_IP_DCCP_CCID3_RTO
*
419 (USEC_PER_SEC
/ 1000)));
421 * Schedule no feedback timer to expire in
422 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
424 t_nfb
= max(hctx
->t_rto
, 2 * hctx
->t_ipi
);
426 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
427 "expire in %lu jiffies (%luus)\n",
428 dccp_role(sk
), sk
, usecs_to_jiffies(t_nfb
), t_nfb
);
430 sk_reset_timer(sk
, &hctx
->no_feedback_timer
,
431 jiffies
+ usecs_to_jiffies(t_nfb
));
434 static int ccid3_hc_tx_parse_options(struct sock
*sk
, u8 packet_type
,
435 u8 option
, u8
*optval
, u8 optlen
)
437 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
441 case TFRC_OPT_RECEIVE_RATE
:
442 case TFRC_OPT_LOSS_EVENT_RATE
:
443 /* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
444 if (packet_type
== DCCP_PKT_DATA
)
446 if (unlikely(optlen
!= 4)) {
447 DCCP_WARN("%s(%p), invalid len %d for %u\n",
448 dccp_role(sk
), sk
, optlen
, option
);
451 opt_val
= ntohl(get_unaligned((__be32
*)optval
));
453 if (option
== TFRC_OPT_RECEIVE_RATE
) {
454 /* Receive Rate is kept in units of 64 bytes/second */
455 hctx
->x_recv
= opt_val
;
458 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
459 dccp_role(sk
), sk
, opt_val
);
461 /* Update the fixpoint Loss Event Rate fraction */
462 hctx
->p
= tfrc_invert_loss_event_rate(opt_val
);
464 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
465 dccp_role(sk
), sk
, opt_val
);
471 static int ccid3_hc_tx_init(struct ccid
*ccid
, struct sock
*sk
)
473 struct ccid3_hc_tx_sock
*hctx
= ccid_priv(ccid
);
476 setup_timer(&hctx
->no_feedback_timer
,
477 ccid3_hc_tx_no_feedback_timer
, (unsigned long)sk
);
481 static void ccid3_hc_tx_exit(struct sock
*sk
)
483 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
485 sk_stop_timer(sk
, &hctx
->no_feedback_timer
);
486 tfrc_tx_hist_purge(&hctx
->hist
);
489 static void ccid3_hc_tx_get_info(struct sock
*sk
, struct tcp_info
*info
)
491 info
->tcpi_rto
= ccid3_hc_tx_sk(sk
)->t_rto
;
492 info
->tcpi_rtt
= ccid3_hc_tx_sk(sk
)->rtt
;
495 static int ccid3_hc_tx_getsockopt(struct sock
*sk
, const int optname
, int len
,
496 u32 __user
*optval
, int __user
*optlen
)
498 const struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
499 struct tfrc_tx_info tfrc
;
503 case DCCP_SOCKOPT_CCID_TX_INFO
:
504 if (len
< sizeof(tfrc
))
506 tfrc
.tfrctx_x
= hctx
->x
;
507 tfrc
.tfrctx_x_recv
= hctx
->x_recv
;
508 tfrc
.tfrctx_x_calc
= hctx
->x_calc
;
509 tfrc
.tfrctx_rtt
= hctx
->rtt
;
510 tfrc
.tfrctx_p
= hctx
->p
;
511 tfrc
.tfrctx_rto
= hctx
->t_rto
;
512 tfrc
.tfrctx_ipi
= hctx
->t_ipi
;
520 if (put_user(len
, optlen
) || copy_to_user(optval
, val
, len
))
527 * Receiver Half-Connection Routines
529 static void ccid3_hc_rx_send_feedback(struct sock
*sk
,
530 const struct sk_buff
*skb
,
531 enum ccid3_fback_type fbtype
)
533 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
536 case CCID3_FBACK_INITIAL
:
538 hcrx
->p_inverse
= ~0U; /* see RFC 4342, 8.5 */
540 case CCID3_FBACK_PARAM_CHANGE
:
541 if (unlikely(hcrx
->feedback
== CCID3_FBACK_NONE
)) {
543 * rfc3448bis-06, 6.3.1: First packet(s) lost or marked
544 * FIXME: in rfc3448bis the receiver returns X_recv=0
545 * here as it normally would in the first feedback packet.
546 * However this is not possible yet, since the code still
547 * uses RFC 3448, i.e.
549 * Calculate X_calc using the TCP throughput equation.
550 * X = max(min(X_calc, 2*X_recv), s/t_mbi);
551 * would bring X down to s/t_mbi. That is why we return
552 * X_recv according to rfc3448bis-06 for the moment.
554 u32 s
= tfrc_rx_hist_packet_size(&hcrx
->hist
),
555 rtt
= tfrc_rx_hist_rtt(&hcrx
->hist
);
557 hcrx
->x_recv
= scaled_div32(s
, 2 * rtt
);
561 * When parameters change (new loss or p > p_prev), we do not
562 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
563 * always check whether at least RTT time units were covered.
565 hcrx
->x_recv
= tfrc_rx_hist_x_recv(&hcrx
->hist
, hcrx
->x_recv
);
567 case CCID3_FBACK_PERIODIC
:
569 * Step (2) of rfc3448bis-06, 6.2:
570 * - if no data packets have been received, just restart timer
571 * - if data packets have been received, re-compute X_recv
573 if (hcrx
->hist
.bytes_recvd
== 0)
574 goto prepare_for_next_time
;
575 hcrx
->x_recv
= tfrc_rx_hist_x_recv(&hcrx
->hist
, hcrx
->x_recv
);
581 ccid3_pr_debug("X_recv=%u, 1/p=%u\n", hcrx
->x_recv
, hcrx
->p_inverse
);
583 dccp_sk(sk
)->dccps_hc_rx_insert_options
= 1;
586 prepare_for_next_time
:
587 tfrc_rx_hist_restart_byte_counter(&hcrx
->hist
);
588 hcrx
->last_counter
= dccp_hdr(skb
)->dccph_ccval
;
589 hcrx
->feedback
= fbtype
;
592 static int ccid3_hc_rx_insert_options(struct sock
*sk
, struct sk_buff
*skb
)
594 const struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
597 if (!(sk
->sk_state
== DCCP_OPEN
|| sk
->sk_state
== DCCP_PARTOPEN
))
600 if (dccp_packet_without_ack(skb
))
603 x_recv
= htonl(hcrx
->x_recv
);
604 pinv
= htonl(hcrx
->p_inverse
);
606 if (dccp_insert_option(sk
, skb
, TFRC_OPT_LOSS_EVENT_RATE
,
607 &pinv
, sizeof(pinv
)) ||
608 dccp_insert_option(sk
, skb
, TFRC_OPT_RECEIVE_RATE
,
609 &x_recv
, sizeof(x_recv
)))
615 /** ccid3_first_li - Implements [RFC 3448, 6.3.1]
617 * Determine the length of the first loss interval via inverse lookup.
618 * Assume that X_recv can be computed by the throughput equation
622 * Find some p such that f(p) = fval; return 1/p (scaled).
624 static u32
ccid3_first_li(struct sock
*sk
)
626 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
627 u32 s
= tfrc_rx_hist_packet_size(&hcrx
->hist
),
628 rtt
= tfrc_rx_hist_rtt(&hcrx
->hist
), x_recv
, p
;
632 * rfc3448bis-06, 6.3.1: First data packet(s) are marked or lost. Set p
633 * to give the equivalent of X_target = s/(2*R). Thus fval = 2 and so p
634 * is about 20.64%. This yields an interval length of 4.84 (rounded up).
636 if (unlikely(hcrx
->feedback
== CCID3_FBACK_NONE
))
639 x_recv
= tfrc_rx_hist_x_recv(&hcrx
->hist
, hcrx
->x_recv
);
643 fval
= scaled_div32(scaled_div(s
, rtt
), x_recv
);
644 p
= tfrc_calc_x_reverse_lookup(fval
);
646 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
647 "loss rate=%u\n", dccp_role(sk
), sk
, x_recv
, p
);
650 return scaled_div(1, p
);
655 static void ccid3_hc_rx_packet_recv(struct sock
*sk
, struct sk_buff
*skb
)
657 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
658 const u64 ndp
= dccp_sk(sk
)->dccps_options_received
.dccpor_ndp
;
659 const bool is_data_packet
= dccp_data_packet(skb
);
662 * Perform loss detection and handle pending losses
664 if (tfrc_rx_congestion_event(&hcrx
->hist
, &hcrx
->li_hist
,
665 skb
, ndp
, ccid3_first_li
, sk
))
666 ccid3_hc_rx_send_feedback(sk
, skb
, CCID3_FBACK_PARAM_CHANGE
);
668 * Feedback for first non-empty data packet (RFC 3448, 6.3)
670 else if (unlikely(hcrx
->feedback
== CCID3_FBACK_NONE
&& is_data_packet
))
671 ccid3_hc_rx_send_feedback(sk
, skb
, CCID3_FBACK_INITIAL
);
673 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
675 else if (!tfrc_rx_hist_loss_pending(&hcrx
->hist
) && is_data_packet
&&
676 SUB16(dccp_hdr(skb
)->dccph_ccval
, hcrx
->last_counter
) > 3)
677 ccid3_hc_rx_send_feedback(sk
, skb
, CCID3_FBACK_PERIODIC
);
680 static int ccid3_hc_rx_init(struct ccid
*ccid
, struct sock
*sk
)
682 struct ccid3_hc_rx_sock
*hcrx
= ccid_priv(ccid
);
684 tfrc_lh_init(&hcrx
->li_hist
);
685 return tfrc_rx_hist_init(&hcrx
->hist
, sk
);
688 static void ccid3_hc_rx_exit(struct sock
*sk
)
690 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
692 tfrc_rx_hist_purge(&hcrx
->hist
);
693 tfrc_lh_cleanup(&hcrx
->li_hist
);
696 static void ccid3_hc_rx_get_info(struct sock
*sk
, struct tcp_info
*info
)
698 info
->tcpi_options
|= TCPI_OPT_TIMESTAMPS
;
699 info
->tcpi_rcv_rtt
= tfrc_rx_hist_rtt(&ccid3_hc_rx_sk(sk
)->hist
);
702 static int ccid3_hc_rx_getsockopt(struct sock
*sk
, const int optname
, int len
,
703 u32 __user
*optval
, int __user
*optlen
)
705 const struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
706 struct tfrc_rx_info rx_info
;
710 case DCCP_SOCKOPT_CCID_RX_INFO
:
711 if (len
< sizeof(rx_info
))
713 rx_info
.tfrcrx_x_recv
= hcrx
->x_recv
;
714 rx_info
.tfrcrx_rtt
= tfrc_rx_hist_rtt(&hcrx
->hist
);
715 rx_info
.tfrcrx_p
= tfrc_invert_loss_event_rate(hcrx
->p_inverse
);
716 len
= sizeof(rx_info
);
723 if (put_user(len
, optlen
) || copy_to_user(optval
, val
, len
))
729 static struct ccid_operations ccid3
= {
730 .ccid_id
= DCCPC_CCID3
,
731 .ccid_name
= "TCP-Friendly Rate Control",
732 .ccid_owner
= THIS_MODULE
,
733 .ccid_hc_tx_obj_size
= sizeof(struct ccid3_hc_tx_sock
),
734 .ccid_hc_tx_init
= ccid3_hc_tx_init
,
735 .ccid_hc_tx_exit
= ccid3_hc_tx_exit
,
736 .ccid_hc_tx_send_packet
= ccid3_hc_tx_send_packet
,
737 .ccid_hc_tx_packet_sent
= ccid3_hc_tx_packet_sent
,
738 .ccid_hc_tx_packet_recv
= ccid3_hc_tx_packet_recv
,
739 .ccid_hc_tx_parse_options
= ccid3_hc_tx_parse_options
,
740 .ccid_hc_rx_obj_size
= sizeof(struct ccid3_hc_rx_sock
),
741 .ccid_hc_rx_init
= ccid3_hc_rx_init
,
742 .ccid_hc_rx_exit
= ccid3_hc_rx_exit
,
743 .ccid_hc_rx_insert_options
= ccid3_hc_rx_insert_options
,
744 .ccid_hc_rx_packet_recv
= ccid3_hc_rx_packet_recv
,
745 .ccid_hc_rx_get_info
= ccid3_hc_rx_get_info
,
746 .ccid_hc_tx_get_info
= ccid3_hc_tx_get_info
,
747 .ccid_hc_rx_getsockopt
= ccid3_hc_rx_getsockopt
,
748 .ccid_hc_tx_getsockopt
= ccid3_hc_tx_getsockopt
,
751 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
752 module_param(ccid3_debug
, bool, 0644);
753 MODULE_PARM_DESC(ccid3_debug
, "Enable debug messages");
756 static __init
int ccid3_module_init(void)
761 * Without a fine-grained clock resolution, RTTs/X_recv are not sampled
762 * correctly and feedback is sent either too early or too late.
764 hrtimer_get_res(CLOCK_MONOTONIC
, &tp
);
765 if (tp
.tv_sec
|| tp
.tv_nsec
> DCCP_TIME_RESOLUTION
* NSEC_PER_USEC
) {
766 printk(KERN_ERR
"%s: Timer too coarse (%ld usec), need %u-usec"
767 " resolution - check your clocksource.\n", __func__
,
768 tp
.tv_nsec
/NSEC_PER_USEC
, DCCP_TIME_RESOLUTION
);
769 return -ESOCKTNOSUPPORT
;
771 return ccid_register(&ccid3
);
773 module_init(ccid3_module_init
);
775 static __exit
void ccid3_module_exit(void)
777 ccid_unregister(&ccid3
);
779 module_exit(ccid3_module_exit
);
781 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
782 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
783 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
784 MODULE_LICENSE("GPL");
785 MODULE_ALIAS("net-dccp-ccid-3");