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 * ccid3_hc_tx_measure_packet_size - Measuring the packet size `s' (sec 4.1)
140 * @new_len: DCCP payload size in bytes (not used by all methods)
142 static u32
ccid3_hc_tx_measure_packet_size(struct sock
*sk
, const u16 new_len
)
144 #if defined(CONFIG_IP_DCCP_CCID3_MEASURE_S_AS_AVG)
145 return tfrc_ewma(ccid3_hc_tx_sk(sk
)->s
, new_len
, 9);
146 #elif defined(CONFIG_IP_DCCP_CCID3_MEASURE_S_AS_MAX)
147 return max(ccid3_hc_tx_sk(sk
)->s
, new_len
);
148 #else /* CONFIG_IP_DCCP_CCID3_MEASURE_S_AS_MPS */
149 return dccp_sk(sk
)->dccps_mss_cache
;
154 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
155 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
157 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock
*hctx
,
160 u32 delta
= ktime_us_delta(now
, hctx
->t_last_win_count
),
161 quarter_rtts
= (4 * delta
) / hctx
->rtt
;
163 if (quarter_rtts
> 0) {
164 hctx
->t_last_win_count
= now
;
165 hctx
->last_win_count
+= min(quarter_rtts
, 5U);
166 hctx
->last_win_count
&= 0xF; /* mod 16 */
170 static void ccid3_hc_tx_no_feedback_timer(unsigned long data
)
172 struct sock
*sk
= (struct sock
*)data
;
173 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
174 unsigned long t_nfb
= USEC_PER_SEC
/ 5;
177 if (sock_owned_by_user(sk
)) {
178 /* Try again later. */
179 /* XXX: set some sensible MIB */
183 ccid3_pr_debug("%s(%p) entry with%s feedback\n", dccp_role(sk
), sk
,
184 hctx
->feedback
? "" : "out");
186 /* Ignore and do not restart after leaving the established state */
187 if ((1 << sk
->sk_state
) & ~(DCCPF_OPEN
| DCCPF_PARTOPEN
))
190 /* Reset feedback state to "no feedback received" */
191 hctx
->feedback
= false;
194 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
195 * RTO is 0 if and only if no feedback has been received yet.
197 if (hctx
->t_rto
== 0 || hctx
->p
== 0) {
199 /* halve send rate directly */
200 hctx
->x
= max(hctx
->x
/ 2, (((u64
)hctx
->s
) << 6) / TFRC_T_MBI
);
201 ccid3_update_send_interval(hctx
);
204 * Modify the cached value of X_recv
206 * If (X_calc > 2 * X_recv)
207 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
209 * X_recv = X_calc / 4;
211 * Note that X_recv is scaled by 2^6 while X_calc is not
213 BUG_ON(hctx
->p
&& !hctx
->x_calc
);
215 if (hctx
->x_calc
> (hctx
->x_recv
>> 5))
217 max(hctx
->x_recv
/ 2,
218 (((__u64
)hctx
->s
) << 6) / (2 * TFRC_T_MBI
));
220 hctx
->x_recv
= hctx
->x_calc
;
223 ccid3_hc_tx_update_x(sk
, NULL
);
225 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
226 (unsigned long long)hctx
->x
);
229 * Set new timeout for the nofeedback timer.
230 * See comments in packet_recv() regarding the value of t_RTO.
232 if (unlikely(hctx
->t_rto
== 0)) /* no feedback received yet */
233 t_nfb
= TFRC_INITIAL_TIMEOUT
;
235 t_nfb
= max(hctx
->t_rto
, 2 * hctx
->t_ipi
);
238 sk_reset_timer(sk
, &hctx
->no_feedback_timer
,
239 jiffies
+ usecs_to_jiffies(t_nfb
));
246 * ccid3_hc_tx_send_packet - Delay-based dequeueing of TX packets
247 * @skb: next packet candidate to send on @sk
248 * This function uses the convention of ccid_packet_dequeue_eval() and
249 * returns a millisecond-delay value between 0 and t_mbi = 64000 msec.
251 static int ccid3_hc_tx_send_packet(struct sock
*sk
, struct sk_buff
*skb
)
253 struct dccp_sock
*dp
= dccp_sk(sk
);
254 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
255 ktime_t now
= ktime_get_real();
259 * This function is called only for Data and DataAck packets. Sending
260 * zero-sized Data(Ack)s is theoretically possible, but for congestion
261 * control this case is pathological - ignore it.
263 if (unlikely(skb
->len
== 0))
267 sk_reset_timer(sk
, &hctx
->no_feedback_timer
, (jiffies
+
268 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT
)));
269 hctx
->last_win_count
= 0;
270 hctx
->t_last_win_count
= now
;
272 /* Set t_0 for initial packet */
276 * Use initial RTT sample when available: recommended by erratum
277 * to RFC 4342. This implements the initialisation procedure of
278 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
280 if (dp
->dccps_syn_rtt
) {
281 ccid3_pr_debug("SYN RTT = %uus\n", dp
->dccps_syn_rtt
);
282 hctx
->rtt
= dp
->dccps_syn_rtt
;
283 hctx
->x
= rfc3390_initial_rate(sk
);
287 * Sender does not have RTT sample:
288 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
289 * is needed in several parts (e.g. window counter);
290 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
292 hctx
->rtt
= DCCP_FALLBACK_RTT
;
293 hctx
->x
= dp
->dccps_mss_cache
;
297 /* Compute t_ipi = s / X */
298 hctx
->s
= ccid3_hc_tx_measure_packet_size(sk
, skb
->len
);
299 ccid3_update_send_interval(hctx
);
302 delay
= ktime_us_delta(hctx
->t_nom
, now
);
303 ccid3_pr_debug("delay=%ld\n", (long)delay
);
305 * Scheduling of packet transmissions [RFC 3448, 4.6]
307 * if (t_now > t_nom - delta)
308 * // send the packet now
310 * // send the packet in (t_nom - t_now) milliseconds.
312 if (delay
>= TFRC_T_DELTA
)
313 return (u32
)delay
/ USEC_PER_MSEC
;
315 ccid3_hc_tx_update_win_count(hctx
, now
);
318 /* prepare to send now (add options etc.) */
319 dp
->dccps_hc_tx_insert_options
= 1;
320 DCCP_SKB_CB(skb
)->dccpd_ccval
= hctx
->last_win_count
;
322 /* set the nominal send time for the next following packet */
323 hctx
->t_nom
= ktime_add_us(hctx
->t_nom
, hctx
->t_ipi
);
324 return CCID_PACKET_SEND_AT_ONCE
;
327 static void ccid3_hc_tx_packet_sent(struct sock
*sk
, unsigned int len
)
329 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
331 /* Changes to s will become effective the next time X is computed */
332 hctx
->s
= ccid3_hc_tx_measure_packet_size(sk
, len
);
334 if (tfrc_tx_hist_add(&hctx
->hist
, dccp_sk(sk
)->dccps_gss
))
335 DCCP_CRIT("packet history - out of memory!");
338 static void ccid3_hc_tx_packet_recv(struct sock
*sk
, struct sk_buff
*skb
)
340 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
341 struct tfrc_tx_hist_entry
*acked
;
346 /* we are only interested in ACKs */
347 if (!(DCCP_SKB_CB(skb
)->dccpd_type
== DCCP_PKT_ACK
||
348 DCCP_SKB_CB(skb
)->dccpd_type
== DCCP_PKT_DATAACK
))
351 * Locate the acknowledged packet in the TX history.
353 * Returning "entry not found" here can for instance happen when
354 * - the host has not sent out anything (e.g. a passive server),
355 * - the Ack is outdated (packet with higher Ack number was received),
356 * - it is a bogus Ack (for a packet not sent on this connection).
358 acked
= tfrc_tx_hist_find_entry(hctx
->hist
, dccp_hdr_ack_seq(skb
));
361 /* For the sake of RTT sampling, ignore/remove all older entries */
362 tfrc_tx_hist_purge(&acked
->next
);
364 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
365 now
= ktime_get_real();
366 r_sample
= dccp_sample_rtt(sk
, ktime_us_delta(now
, acked
->stamp
));
367 hctx
->rtt
= tfrc_ewma(hctx
->rtt
, r_sample
, 9);
370 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
372 if (!hctx
->feedback
) {
373 hctx
->feedback
= true;
375 if (hctx
->t_rto
== 0) {
377 * Initial feedback packet: Larger Initial Windows (4.2)
379 hctx
->x
= rfc3390_initial_rate(sk
);
382 ccid3_update_send_interval(hctx
);
384 goto done_computing_x
;
385 } else if (hctx
->p
== 0) {
387 * First feedback after nofeedback timer expiry (4.3)
389 goto done_computing_x
;
393 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
395 hctx
->x_calc
= tfrc_calc_x(hctx
->s
, hctx
->rtt
, hctx
->p
);
396 ccid3_hc_tx_update_x(sk
, &now
);
399 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
400 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
401 dccp_role(sk
), sk
, hctx
->rtt
, r_sample
,
402 hctx
->s
, hctx
->p
, hctx
->x_calc
,
403 (unsigned)(hctx
->x_recv
>> 6),
404 (unsigned)(hctx
->x
>> 6));
406 /* unschedule no feedback timer */
407 sk_stop_timer(sk
, &hctx
->no_feedback_timer
);
410 * As we have calculated new ipi, delta, t_nom it is possible
411 * that we now can send a packet, so wake up dccp_wait_for_ccid
413 sk
->sk_write_space(sk
);
416 * Update timeout interval for the nofeedback timer.
417 * We use a configuration option to increase the lower bound.
418 * This can help avoid triggering the nofeedback timer too
419 * often ('spinning') on LANs with small RTTs.
421 hctx
->t_rto
= max_t(u32
, 4 * hctx
->rtt
, (CONFIG_IP_DCCP_CCID3_RTO
*
422 (USEC_PER_SEC
/ 1000)));
424 * Schedule no feedback timer to expire in
425 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
427 t_nfb
= max(hctx
->t_rto
, 2 * hctx
->t_ipi
);
429 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
430 "expire in %lu jiffies (%luus)\n",
431 dccp_role(sk
), sk
, usecs_to_jiffies(t_nfb
), t_nfb
);
433 sk_reset_timer(sk
, &hctx
->no_feedback_timer
,
434 jiffies
+ usecs_to_jiffies(t_nfb
));
437 static int ccid3_hc_tx_parse_options(struct sock
*sk
, u8 packet_type
,
438 u8 option
, u8
*optval
, u8 optlen
)
440 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
444 case TFRC_OPT_RECEIVE_RATE
:
445 case TFRC_OPT_LOSS_EVENT_RATE
:
446 /* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
447 if (packet_type
== DCCP_PKT_DATA
)
449 if (unlikely(optlen
!= 4)) {
450 DCCP_WARN("%s(%p), invalid len %d for %u\n",
451 dccp_role(sk
), sk
, optlen
, option
);
454 opt_val
= ntohl(get_unaligned((__be32
*)optval
));
456 if (option
== TFRC_OPT_RECEIVE_RATE
) {
457 /* Receive Rate is kept in units of 64 bytes/second */
458 hctx
->x_recv
= opt_val
;
461 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
462 dccp_role(sk
), sk
, opt_val
);
464 /* Update the fixpoint Loss Event Rate fraction */
465 hctx
->p
= tfrc_invert_loss_event_rate(opt_val
);
467 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
468 dccp_role(sk
), sk
, opt_val
);
474 static int ccid3_hc_tx_init(struct ccid
*ccid
, struct sock
*sk
)
476 struct ccid3_hc_tx_sock
*hctx
= ccid_priv(ccid
);
479 setup_timer(&hctx
->no_feedback_timer
,
480 ccid3_hc_tx_no_feedback_timer
, (unsigned long)sk
);
484 static void ccid3_hc_tx_exit(struct sock
*sk
)
486 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
488 sk_stop_timer(sk
, &hctx
->no_feedback_timer
);
489 tfrc_tx_hist_purge(&hctx
->hist
);
492 static void ccid3_hc_tx_get_info(struct sock
*sk
, struct tcp_info
*info
)
494 info
->tcpi_rto
= ccid3_hc_tx_sk(sk
)->t_rto
;
495 info
->tcpi_rtt
= ccid3_hc_tx_sk(sk
)->rtt
;
498 static int ccid3_hc_tx_getsockopt(struct sock
*sk
, const int optname
, int len
,
499 u32 __user
*optval
, int __user
*optlen
)
501 const struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
502 struct tfrc_tx_info tfrc
;
506 case DCCP_SOCKOPT_CCID_TX_INFO
:
507 if (len
< sizeof(tfrc
))
509 tfrc
.tfrctx_x
= hctx
->x
;
510 tfrc
.tfrctx_x_recv
= hctx
->x_recv
;
511 tfrc
.tfrctx_x_calc
= hctx
->x_calc
;
512 tfrc
.tfrctx_rtt
= hctx
->rtt
;
513 tfrc
.tfrctx_p
= hctx
->p
;
514 tfrc
.tfrctx_rto
= hctx
->t_rto
;
515 tfrc
.tfrctx_ipi
= hctx
->t_ipi
;
523 if (put_user(len
, optlen
) || copy_to_user(optval
, val
, len
))
530 * Receiver Half-Connection Routines
532 static void ccid3_hc_rx_send_feedback(struct sock
*sk
,
533 const struct sk_buff
*skb
,
534 enum ccid3_fback_type fbtype
)
536 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
539 case CCID3_FBACK_INITIAL
:
541 hcrx
->p_inverse
= ~0U; /* see RFC 4342, 8.5 */
543 case CCID3_FBACK_PARAM_CHANGE
:
544 if (unlikely(hcrx
->feedback
== CCID3_FBACK_NONE
)) {
546 * rfc3448bis-06, 6.3.1: First packet(s) lost or marked
547 * FIXME: in rfc3448bis the receiver returns X_recv=0
548 * here as it normally would in the first feedback packet.
549 * However this is not possible yet, since the code still
550 * uses RFC 3448, i.e.
552 * Calculate X_calc using the TCP throughput equation.
553 * X = max(min(X_calc, 2*X_recv), s/t_mbi);
554 * would bring X down to s/t_mbi. That is why we return
555 * X_recv according to rfc3448bis-06 for the moment.
557 u32 s
= tfrc_rx_hist_packet_size(&hcrx
->hist
),
558 rtt
= tfrc_rx_hist_rtt(&hcrx
->hist
);
560 hcrx
->x_recv
= scaled_div32(s
, 2 * rtt
);
564 * When parameters change (new loss or p > p_prev), we do not
565 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
566 * always check whether at least RTT time units were covered.
568 hcrx
->x_recv
= tfrc_rx_hist_x_recv(&hcrx
->hist
, hcrx
->x_recv
);
570 case CCID3_FBACK_PERIODIC
:
572 * Step (2) of rfc3448bis-06, 6.2:
573 * - if no data packets have been received, just restart timer
574 * - if data packets have been received, re-compute X_recv
576 if (hcrx
->hist
.bytes_recvd
== 0)
577 goto prepare_for_next_time
;
578 hcrx
->x_recv
= tfrc_rx_hist_x_recv(&hcrx
->hist
, hcrx
->x_recv
);
584 ccid3_pr_debug("X_recv=%u, 1/p=%u\n", hcrx
->x_recv
, hcrx
->p_inverse
);
586 dccp_sk(sk
)->dccps_hc_rx_insert_options
= 1;
589 prepare_for_next_time
:
590 tfrc_rx_hist_restart_byte_counter(&hcrx
->hist
);
591 hcrx
->last_counter
= dccp_hdr(skb
)->dccph_ccval
;
592 hcrx
->feedback
= fbtype
;
595 static int ccid3_hc_rx_insert_options(struct sock
*sk
, struct sk_buff
*skb
)
597 const struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
600 if (!(sk
->sk_state
== DCCP_OPEN
|| sk
->sk_state
== DCCP_PARTOPEN
))
603 if (dccp_packet_without_ack(skb
))
606 x_recv
= htonl(hcrx
->x_recv
);
607 pinv
= htonl(hcrx
->p_inverse
);
609 if (dccp_insert_option(sk
, skb
, TFRC_OPT_LOSS_EVENT_RATE
,
610 &pinv
, sizeof(pinv
)) ||
611 dccp_insert_option(sk
, skb
, TFRC_OPT_RECEIVE_RATE
,
612 &x_recv
, sizeof(x_recv
)))
618 /** ccid3_first_li - Implements [RFC 3448, 6.3.1]
620 * Determine the length of the first loss interval via inverse lookup.
621 * Assume that X_recv can be computed by the throughput equation
625 * Find some p such that f(p) = fval; return 1/p (scaled).
627 static u32
ccid3_first_li(struct sock
*sk
)
629 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
630 u32 s
= tfrc_rx_hist_packet_size(&hcrx
->hist
),
631 rtt
= tfrc_rx_hist_rtt(&hcrx
->hist
), x_recv
, p
;
635 * rfc3448bis-06, 6.3.1: First data packet(s) are marked or lost. Set p
636 * to give the equivalent of X_target = s/(2*R). Thus fval = 2 and so p
637 * is about 20.64%. This yields an interval length of 4.84 (rounded up).
639 if (unlikely(hcrx
->feedback
== CCID3_FBACK_NONE
))
642 x_recv
= tfrc_rx_hist_x_recv(&hcrx
->hist
, hcrx
->x_recv
);
646 fval
= scaled_div32(scaled_div(s
, rtt
), x_recv
);
647 p
= tfrc_calc_x_reverse_lookup(fval
);
649 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
650 "loss rate=%u\n", dccp_role(sk
), sk
, x_recv
, p
);
653 return scaled_div(1, p
);
658 static void ccid3_hc_rx_packet_recv(struct sock
*sk
, struct sk_buff
*skb
)
660 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
661 const u64 ndp
= dccp_sk(sk
)->dccps_options_received
.dccpor_ndp
;
662 const bool is_data_packet
= dccp_data_packet(skb
);
665 * Perform loss detection and handle pending losses
667 if (tfrc_rx_congestion_event(&hcrx
->hist
, &hcrx
->li_hist
,
668 skb
, ndp
, ccid3_first_li
, sk
))
669 ccid3_hc_rx_send_feedback(sk
, skb
, CCID3_FBACK_PARAM_CHANGE
);
671 * Feedback for first non-empty data packet (RFC 3448, 6.3)
673 else if (unlikely(hcrx
->feedback
== CCID3_FBACK_NONE
&& is_data_packet
))
674 ccid3_hc_rx_send_feedback(sk
, skb
, CCID3_FBACK_INITIAL
);
676 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
678 else if (!tfrc_rx_hist_loss_pending(&hcrx
->hist
) && is_data_packet
&&
679 SUB16(dccp_hdr(skb
)->dccph_ccval
, hcrx
->last_counter
) > 3)
680 ccid3_hc_rx_send_feedback(sk
, skb
, CCID3_FBACK_PERIODIC
);
683 static int ccid3_hc_rx_init(struct ccid
*ccid
, struct sock
*sk
)
685 struct ccid3_hc_rx_sock
*hcrx
= ccid_priv(ccid
);
687 tfrc_lh_init(&hcrx
->li_hist
);
688 return tfrc_rx_hist_init(&hcrx
->hist
, sk
);
691 static void ccid3_hc_rx_exit(struct sock
*sk
)
693 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
695 tfrc_rx_hist_purge(&hcrx
->hist
);
696 tfrc_lh_cleanup(&hcrx
->li_hist
);
699 static void ccid3_hc_rx_get_info(struct sock
*sk
, struct tcp_info
*info
)
701 info
->tcpi_options
|= TCPI_OPT_TIMESTAMPS
;
702 info
->tcpi_rcv_rtt
= tfrc_rx_hist_rtt(&ccid3_hc_rx_sk(sk
)->hist
);
705 static int ccid3_hc_rx_getsockopt(struct sock
*sk
, const int optname
, int len
,
706 u32 __user
*optval
, int __user
*optlen
)
708 const struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
709 struct tfrc_rx_info rx_info
;
713 case DCCP_SOCKOPT_CCID_RX_INFO
:
714 if (len
< sizeof(rx_info
))
716 rx_info
.tfrcrx_x_recv
= hcrx
->x_recv
;
717 rx_info
.tfrcrx_rtt
= tfrc_rx_hist_rtt(&hcrx
->hist
);
718 rx_info
.tfrcrx_p
= tfrc_invert_loss_event_rate(hcrx
->p_inverse
);
719 len
= sizeof(rx_info
);
726 if (put_user(len
, optlen
) || copy_to_user(optval
, val
, len
))
732 static struct ccid_operations ccid3
= {
733 .ccid_id
= DCCPC_CCID3
,
734 .ccid_name
= "TCP-Friendly Rate Control",
735 .ccid_owner
= THIS_MODULE
,
736 .ccid_hc_tx_obj_size
= sizeof(struct ccid3_hc_tx_sock
),
737 .ccid_hc_tx_init
= ccid3_hc_tx_init
,
738 .ccid_hc_tx_exit
= ccid3_hc_tx_exit
,
739 .ccid_hc_tx_send_packet
= ccid3_hc_tx_send_packet
,
740 .ccid_hc_tx_packet_sent
= ccid3_hc_tx_packet_sent
,
741 .ccid_hc_tx_packet_recv
= ccid3_hc_tx_packet_recv
,
742 .ccid_hc_tx_parse_options
= ccid3_hc_tx_parse_options
,
743 .ccid_hc_rx_obj_size
= sizeof(struct ccid3_hc_rx_sock
),
744 .ccid_hc_rx_init
= ccid3_hc_rx_init
,
745 .ccid_hc_rx_exit
= ccid3_hc_rx_exit
,
746 .ccid_hc_rx_insert_options
= ccid3_hc_rx_insert_options
,
747 .ccid_hc_rx_packet_recv
= ccid3_hc_rx_packet_recv
,
748 .ccid_hc_rx_get_info
= ccid3_hc_rx_get_info
,
749 .ccid_hc_tx_get_info
= ccid3_hc_tx_get_info
,
750 .ccid_hc_rx_getsockopt
= ccid3_hc_rx_getsockopt
,
751 .ccid_hc_tx_getsockopt
= ccid3_hc_tx_getsockopt
,
754 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
755 module_param(ccid3_debug
, bool, 0644);
756 MODULE_PARM_DESC(ccid3_debug
, "Enable debug messages");
759 static __init
int ccid3_module_init(void)
764 * Without a fine-grained clock resolution, RTTs/X_recv are not sampled
765 * correctly and feedback is sent either too early or too late.
767 hrtimer_get_res(CLOCK_MONOTONIC
, &tp
);
768 if (tp
.tv_sec
|| tp
.tv_nsec
> DCCP_TIME_RESOLUTION
* NSEC_PER_USEC
) {
769 printk(KERN_ERR
"%s: Timer too coarse (%ld usec), need %u-usec"
770 " resolution - check your clocksource.\n", __func__
,
771 tp
.tv_nsec
/NSEC_PER_USEC
, DCCP_TIME_RESOLUTION
);
772 return -ESOCKTNOSUPPORT
;
774 return ccid_register(&ccid3
);
776 module_init(ccid3_module_init
);
778 static __exit
void ccid3_module_exit(void)
780 ccid_unregister(&ccid3
);
782 module_exit(ccid3_module_exit
);
784 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
785 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
786 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
787 MODULE_LICENSE("GPL");
788 MODULE_ALIAS("net-dccp-ccid-3");