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
->ccid3hctx_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
);
76 hctx
->ccid3hctx_state
= state
;
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
= min_t(__u32
, 4 * hctx
->ccid3hctx_s
,
92 max_t(__u32
, 2 * hctx
->ccid3hctx_s
, 4380));
94 return scaled_div(w_init
<< 6, hctx
->ccid3hctx_rtt
);
98 * Recalculate t_ipi and delta (should be called whenever X changes)
100 static inline void ccid3_update_send_interval(struct ccid3_hc_tx_sock
*hctx
)
102 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
103 hctx
->ccid3hctx_t_ipi
= scaled_div32(((u64
)hctx
->ccid3hctx_s
) << 6,
106 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
107 hctx
->ccid3hctx_delta
= min_t(u32
, hctx
->ccid3hctx_t_ipi
/ 2,
108 TFRC_OPSYS_HALF_TIME_GRAN
);
110 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n",
111 hctx
->ccid3hctx_t_ipi
, hctx
->ccid3hctx_delta
,
112 hctx
->ccid3hctx_s
, (unsigned)(hctx
->ccid3hctx_x
>> 6));
116 static u32
ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock
*hctx
, ktime_t now
)
118 u32 delta
= ktime_us_delta(now
, hctx
->ccid3hctx_t_last_win_count
);
120 return delta
/ hctx
->ccid3hctx_rtt
;
124 * ccid3_hc_tx_update_x - Update allowed sending rate X
125 * @stamp: most recent time if available - can be left NULL.
126 * This function tracks draft rfc3448bis, check there for latest details.
128 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
129 * fine-grained resolution of sending rates. This requires scaling by 2^6
130 * throughout the code. Only X_calc is unscaled (in bytes/second).
133 static void ccid3_hc_tx_update_x(struct sock
*sk
, ktime_t
*stamp
)
136 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
137 __u64 min_rate
= 2 * hctx
->ccid3hctx_x_recv
;
138 const __u64 old_x
= hctx
->ccid3hctx_x
;
139 ktime_t now
= stamp
? *stamp
: ktime_get_real();
142 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
143 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
144 * a sender is idle if it has not sent anything over a 2-RTT-period.
145 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
147 if (ccid3_hc_tx_idle_rtt(hctx
, now
) >= 2) {
148 min_rate
= rfc3390_initial_rate(sk
);
149 min_rate
= max(min_rate
, 2 * hctx
->ccid3hctx_x_recv
);
152 if (hctx
->ccid3hctx_p
> 0) {
154 hctx
->ccid3hctx_x
= min(((__u64
)hctx
->ccid3hctx_x_calc
) << 6,
156 hctx
->ccid3hctx_x
= max(hctx
->ccid3hctx_x
,
157 (((__u64
)hctx
->ccid3hctx_s
) << 6) /
160 } else if (ktime_us_delta(now
, hctx
->ccid3hctx_t_ld
)
161 - (s64
)hctx
->ccid3hctx_rtt
>= 0) {
164 max(min(2 * hctx
->ccid3hctx_x
, min_rate
),
165 scaled_div(((__u64
)hctx
->ccid3hctx_s
) << 6,
166 hctx
->ccid3hctx_rtt
));
167 hctx
->ccid3hctx_t_ld
= now
;
170 if (hctx
->ccid3hctx_x
!= old_x
) {
171 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
172 "X_recv=%u\n", (unsigned)(old_x
>> 6),
173 (unsigned)(hctx
->ccid3hctx_x
>> 6),
174 hctx
->ccid3hctx_x_calc
,
175 (unsigned)(hctx
->ccid3hctx_x_recv
>> 6));
177 ccid3_update_send_interval(hctx
);
182 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
183 * @len: DCCP packet payload size in bytes
185 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock
*hctx
, int len
)
187 const u16 old_s
= hctx
->ccid3hctx_s
;
189 hctx
->ccid3hctx_s
= tfrc_ewma(hctx
->ccid3hctx_s
, len
, 9);
191 if (hctx
->ccid3hctx_s
!= old_s
)
192 ccid3_update_send_interval(hctx
);
196 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
197 * The algorithm is not applicable if RTT < 4 microseconds.
199 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock
*hctx
,
204 if (unlikely(hctx
->ccid3hctx_rtt
< 4)) /* avoid divide-by-zero */
207 quarter_rtts
= ktime_us_delta(now
, hctx
->ccid3hctx_t_last_win_count
);
208 quarter_rtts
/= hctx
->ccid3hctx_rtt
/ 4;
210 if (quarter_rtts
> 0) {
211 hctx
->ccid3hctx_t_last_win_count
= now
;
212 hctx
->ccid3hctx_last_win_count
+= min_t(u32
, quarter_rtts
, 5);
213 hctx
->ccid3hctx_last_win_count
&= 0xF; /* mod 16 */
217 static void ccid3_hc_tx_no_feedback_timer(unsigned long data
)
219 struct sock
*sk
= (struct sock
*)data
;
220 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
221 unsigned long t_nfb
= USEC_PER_SEC
/ 5;
224 if (sock_owned_by_user(sk
)) {
225 /* Try again later. */
226 /* XXX: set some sensible MIB */
230 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk
), sk
,
231 ccid3_tx_state_name(hctx
->ccid3hctx_state
));
233 switch (hctx
->ccid3hctx_state
) {
234 case TFRC_SSTATE_NO_FBACK
:
235 /* RFC 3448, 4.4: Halve send rate directly */
236 hctx
->ccid3hctx_x
= max(hctx
->ccid3hctx_x
/ 2,
237 (((__u64
)hctx
->ccid3hctx_s
) << 6) /
240 ccid3_pr_debug("%s(%p, state=%s), updated tx rate to %u "
241 "bytes/s\n", dccp_role(sk
), sk
,
242 ccid3_tx_state_name(hctx
->ccid3hctx_state
),
243 (unsigned)(hctx
->ccid3hctx_x
>> 6));
244 /* The value of R is still undefined and so we can not recompute
245 * the timeout value. Keep initial value as per [RFC 4342, 5]. */
246 t_nfb
= TFRC_INITIAL_TIMEOUT
;
247 ccid3_update_send_interval(hctx
);
249 case TFRC_SSTATE_FBACK
:
251 * Modify the cached value of X_recv [RFC 3448, 4.4]
253 * If (p == 0 || X_calc > 2 * X_recv)
254 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
256 * X_recv = X_calc / 4;
258 * Note that X_recv is scaled by 2^6 while X_calc is not
260 BUG_ON(hctx
->ccid3hctx_p
&& !hctx
->ccid3hctx_x_calc
);
262 if (hctx
->ccid3hctx_p
== 0 ||
263 (hctx
->ccid3hctx_x_calc
> (hctx
->ccid3hctx_x_recv
>> 5))) {
265 hctx
->ccid3hctx_x_recv
=
266 max(hctx
->ccid3hctx_x_recv
/ 2,
267 (((__u64
)hctx
->ccid3hctx_s
) << 6) /
270 hctx
->ccid3hctx_x_recv
= hctx
->ccid3hctx_x_calc
;
271 hctx
->ccid3hctx_x_recv
<<= 4;
273 /* Now recalculate X [RFC 3448, 4.3, step (4)] */
274 ccid3_hc_tx_update_x(sk
, NULL
);
276 * Schedule no feedback timer to expire in
277 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
278 * See comments in packet_recv() regarding the value of t_RTO.
280 t_nfb
= max(hctx
->ccid3hctx_t_rto
, 2 * hctx
->ccid3hctx_t_ipi
);
282 case TFRC_SSTATE_NO_SENT
:
283 DCCP_BUG("%s(%p) - Illegal state NO_SENT", dccp_role(sk
), sk
);
285 case TFRC_SSTATE_TERM
:
290 sk_reset_timer(sk
, &hctx
->ccid3hctx_no_feedback_timer
,
291 jiffies
+ usecs_to_jiffies(t_nfb
));
299 * > 0: delay (in msecs) that should pass before actually sending
300 * = 0: can send immediately
301 * < 0: error condition; do not send packet
303 static int ccid3_hc_tx_send_packet(struct sock
*sk
, struct sk_buff
*skb
)
305 struct dccp_sock
*dp
= dccp_sk(sk
);
306 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
307 ktime_t now
= ktime_get_real();
311 * This function is called only for Data and DataAck packets. Sending
312 * zero-sized Data(Ack)s is theoretically possible, but for congestion
313 * control this case is pathological - ignore it.
315 if (unlikely(skb
->len
== 0))
318 switch (hctx
->ccid3hctx_state
) {
319 case TFRC_SSTATE_NO_SENT
:
320 sk_reset_timer(sk
, &hctx
->ccid3hctx_no_feedback_timer
,
322 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT
)));
323 hctx
->ccid3hctx_last_win_count
= 0;
324 hctx
->ccid3hctx_t_last_win_count
= now
;
326 /* Set t_0 for initial packet */
327 hctx
->ccid3hctx_t_nom
= now
;
329 hctx
->ccid3hctx_s
= skb
->len
;
332 * Use initial RTT sample when available: recommended by erratum
333 * to RFC 4342. This implements the initialisation procedure of
334 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
336 if (dp
->dccps_syn_rtt
) {
337 ccid3_pr_debug("SYN RTT = %uus\n", dp
->dccps_syn_rtt
);
338 hctx
->ccid3hctx_rtt
= dp
->dccps_syn_rtt
;
339 hctx
->ccid3hctx_x
= rfc3390_initial_rate(sk
);
340 hctx
->ccid3hctx_t_ld
= now
;
342 /* Sender does not have RTT sample: X_pps = 1 pkt/sec */
343 hctx
->ccid3hctx_x
= hctx
->ccid3hctx_s
;
344 hctx
->ccid3hctx_x
<<= 6;
346 ccid3_update_send_interval(hctx
);
348 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_NO_FBACK
);
350 case TFRC_SSTATE_NO_FBACK
:
351 case TFRC_SSTATE_FBACK
:
352 delay
= ktime_us_delta(hctx
->ccid3hctx_t_nom
, now
);
353 ccid3_pr_debug("delay=%ld\n", (long)delay
);
355 * Scheduling of packet transmissions [RFC 3448, 4.6]
357 * if (t_now > t_nom - delta)
358 * // send the packet now
360 * // send the packet in (t_nom - t_now) milliseconds.
362 if (delay
- (s64
)hctx
->ccid3hctx_delta
>= 1000)
363 return (u32
)delay
/ 1000L;
365 ccid3_hc_tx_update_win_count(hctx
, now
);
367 case TFRC_SSTATE_TERM
:
368 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk
), sk
);
372 /* prepare to send now (add options etc.) */
373 dp
->dccps_hc_tx_insert_options
= 1;
374 DCCP_SKB_CB(skb
)->dccpd_ccval
= hctx
->ccid3hctx_last_win_count
;
376 /* set the nominal send time for the next following packet */
377 hctx
->ccid3hctx_t_nom
= ktime_add_us(hctx
->ccid3hctx_t_nom
,
378 hctx
->ccid3hctx_t_ipi
);
382 static void ccid3_hc_tx_packet_sent(struct sock
*sk
, int more
,
385 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
387 ccid3_hc_tx_update_s(hctx
, len
);
389 if (tfrc_tx_hist_add(&hctx
->ccid3hctx_hist
, dccp_sk(sk
)->dccps_gss
))
390 DCCP_CRIT("packet history - out of memory!");
393 static void ccid3_hc_tx_packet_recv(struct sock
*sk
, struct sk_buff
*skb
)
395 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
396 struct ccid3_options_received
*opt_recv
;
401 /* we are only interested in ACKs */
402 if (!(DCCP_SKB_CB(skb
)->dccpd_type
== DCCP_PKT_ACK
||
403 DCCP_SKB_CB(skb
)->dccpd_type
== DCCP_PKT_DATAACK
))
406 opt_recv
= &hctx
->ccid3hctx_options_received
;
408 switch (hctx
->ccid3hctx_state
) {
409 case TFRC_SSTATE_NO_FBACK
:
410 case TFRC_SSTATE_FBACK
:
411 now
= ktime_get_real();
413 /* estimate RTT from history if ACK number is valid */
414 r_sample
= tfrc_tx_hist_rtt(hctx
->ccid3hctx_hist
,
415 DCCP_SKB_CB(skb
)->dccpd_ack_seq
, now
);
417 DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk
), sk
,
418 dccp_packet_name(DCCP_SKB_CB(skb
)->dccpd_type
),
419 (unsigned long long)DCCP_SKB_CB(skb
)->dccpd_ack_seq
);
423 /* Update receive rate in units of 64 * bytes/second */
424 hctx
->ccid3hctx_x_recv
= opt_recv
->ccid3or_receive_rate
;
425 hctx
->ccid3hctx_x_recv
<<= 6;
427 /* Update loss event rate */
428 pinv
= opt_recv
->ccid3or_loss_event_rate
;
429 if (pinv
== ~0U || pinv
== 0) /* see RFC 4342, 8.5 */
430 hctx
->ccid3hctx_p
= 0;
431 else /* can not exceed 100% */
432 hctx
->ccid3hctx_p
= 1000000 / pinv
;
434 * Validate new RTT sample and update moving average
436 r_sample
= dccp_sample_rtt(sk
, r_sample
);
437 hctx
->ccid3hctx_rtt
= tfrc_ewma(hctx
->ccid3hctx_rtt
, r_sample
, 9);
439 if (hctx
->ccid3hctx_state
== TFRC_SSTATE_NO_FBACK
) {
441 * Larger Initial Windows [RFC 4342, sec. 5]
443 hctx
->ccid3hctx_x
= rfc3390_initial_rate(sk
);
444 hctx
->ccid3hctx_t_ld
= now
;
446 ccid3_update_send_interval(hctx
);
448 ccid3_pr_debug("%s(%p), s=%u, MSS=%u, "
449 "R_sample=%uus, X=%u\n", dccp_role(sk
),
450 sk
, hctx
->ccid3hctx_s
,
451 dccp_sk(sk
)->dccps_mss_cache
, r_sample
,
452 (unsigned)(hctx
->ccid3hctx_x
>> 6));
454 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_FBACK
);
457 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
458 if (hctx
->ccid3hctx_p
> 0)
459 hctx
->ccid3hctx_x_calc
=
460 tfrc_calc_x(hctx
->ccid3hctx_s
,
463 ccid3_hc_tx_update_x(sk
, &now
);
465 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
466 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
468 sk
, hctx
->ccid3hctx_rtt
, r_sample
,
469 hctx
->ccid3hctx_s
, hctx
->ccid3hctx_p
,
470 hctx
->ccid3hctx_x_calc
,
471 (unsigned)(hctx
->ccid3hctx_x_recv
>> 6),
472 (unsigned)(hctx
->ccid3hctx_x
>> 6));
475 /* unschedule no feedback timer */
476 sk_stop_timer(sk
, &hctx
->ccid3hctx_no_feedback_timer
);
479 * As we have calculated new ipi, delta, t_nom it is possible
480 * that we now can send a packet, so wake up dccp_wait_for_ccid
482 sk
->sk_write_space(sk
);
485 * Update timeout interval for the nofeedback timer.
486 * We use a configuration option to increase the lower bound.
487 * This can help avoid triggering the nofeedback timer too
488 * often ('spinning') on LANs with small RTTs.
490 hctx
->ccid3hctx_t_rto
= max_t(u32
, 4 * hctx
->ccid3hctx_rtt
,
491 CONFIG_IP_DCCP_CCID3_RTO
*
492 (USEC_PER_SEC
/1000));
494 * Schedule no feedback timer to expire in
495 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
497 t_nfb
= max(hctx
->ccid3hctx_t_rto
, 2 * hctx
->ccid3hctx_t_ipi
);
499 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
500 "expire in %lu jiffies (%luus)\n",
502 sk
, usecs_to_jiffies(t_nfb
), t_nfb
);
504 sk_reset_timer(sk
, &hctx
->ccid3hctx_no_feedback_timer
,
505 jiffies
+ usecs_to_jiffies(t_nfb
));
507 case TFRC_SSTATE_NO_SENT
: /* fall through */
508 case TFRC_SSTATE_TERM
: /* ignore feedback when closing */
513 static int ccid3_hc_tx_parse_options(struct sock
*sk
, unsigned char option
,
514 unsigned char len
, u16 idx
,
515 unsigned char *value
)
518 const struct dccp_sock
*dp
= dccp_sk(sk
);
519 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
520 struct ccid3_options_received
*opt_recv
;
523 opt_recv
= &hctx
->ccid3hctx_options_received
;
525 if (opt_recv
->ccid3or_seqno
!= dp
->dccps_gsr
) {
526 opt_recv
->ccid3or_seqno
= dp
->dccps_gsr
;
527 opt_recv
->ccid3or_loss_event_rate
= ~0;
528 opt_recv
->ccid3or_loss_intervals_idx
= 0;
529 opt_recv
->ccid3or_loss_intervals_len
= 0;
530 opt_recv
->ccid3or_receive_rate
= 0;
534 case TFRC_OPT_LOSS_EVENT_RATE
:
535 if (unlikely(len
!= 4)) {
536 DCCP_WARN("%s(%p), invalid len %d "
537 "for TFRC_OPT_LOSS_EVENT_RATE\n",
538 dccp_role(sk
), sk
, len
);
541 opt_val
= get_unaligned((__be32
*)value
);
542 opt_recv
->ccid3or_loss_event_rate
= ntohl(opt_val
);
543 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
545 opt_recv
->ccid3or_loss_event_rate
);
548 case TFRC_OPT_LOSS_INTERVALS
:
549 opt_recv
->ccid3or_loss_intervals_idx
= idx
;
550 opt_recv
->ccid3or_loss_intervals_len
= len
;
551 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
553 opt_recv
->ccid3or_loss_intervals_idx
,
554 opt_recv
->ccid3or_loss_intervals_len
);
556 case TFRC_OPT_RECEIVE_RATE
:
557 if (unlikely(len
!= 4)) {
558 DCCP_WARN("%s(%p), invalid len %d "
559 "for TFRC_OPT_RECEIVE_RATE\n",
560 dccp_role(sk
), sk
, len
);
563 opt_val
= get_unaligned((__be32
*)value
);
564 opt_recv
->ccid3or_receive_rate
= ntohl(opt_val
);
565 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
567 opt_recv
->ccid3or_receive_rate
);
575 static int ccid3_hc_tx_init(struct ccid
*ccid
, struct sock
*sk
)
577 struct ccid3_hc_tx_sock
*hctx
= ccid_priv(ccid
);
579 hctx
->ccid3hctx_state
= TFRC_SSTATE_NO_SENT
;
580 hctx
->ccid3hctx_hist
= NULL
;
581 setup_timer(&hctx
->ccid3hctx_no_feedback_timer
,
582 ccid3_hc_tx_no_feedback_timer
, (unsigned long)sk
);
587 static void ccid3_hc_tx_exit(struct sock
*sk
)
589 struct ccid3_hc_tx_sock
*hctx
= ccid3_hc_tx_sk(sk
);
591 ccid3_hc_tx_set_state(sk
, TFRC_SSTATE_TERM
);
592 sk_stop_timer(sk
, &hctx
->ccid3hctx_no_feedback_timer
);
594 tfrc_tx_hist_purge(&hctx
->ccid3hctx_hist
);
597 static void ccid3_hc_tx_get_info(struct sock
*sk
, struct tcp_info
*info
)
599 struct ccid3_hc_tx_sock
*hctx
;
601 /* Listen socks doesn't have a private CCID block */
602 if (sk
->sk_state
== DCCP_LISTEN
)
605 hctx
= ccid3_hc_tx_sk(sk
);
606 info
->tcpi_rto
= hctx
->ccid3hctx_t_rto
;
607 info
->tcpi_rtt
= hctx
->ccid3hctx_rtt
;
610 static int ccid3_hc_tx_getsockopt(struct sock
*sk
, const int optname
, int len
,
611 u32 __user
*optval
, int __user
*optlen
)
613 const struct ccid3_hc_tx_sock
*hctx
;
616 /* Listen socks doesn't have a private CCID block */
617 if (sk
->sk_state
== DCCP_LISTEN
)
620 hctx
= ccid3_hc_tx_sk(sk
);
622 case DCCP_SOCKOPT_CCID_TX_INFO
:
623 if (len
< sizeof(hctx
->ccid3hctx_tfrc
))
625 len
= sizeof(hctx
->ccid3hctx_tfrc
);
626 val
= &hctx
->ccid3hctx_tfrc
;
632 if (put_user(len
, optlen
) || copy_to_user(optval
, val
, len
))
639 * Receiver Half-Connection Routines
642 /* CCID3 feedback types */
643 enum ccid3_fback_type
{
644 CCID3_FBACK_NONE
= 0,
646 CCID3_FBACK_PERIODIC
,
647 CCID3_FBACK_PARAM_CHANGE
650 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
651 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state
)
653 static char *ccid3_rx_state_names
[] = {
654 [TFRC_RSTATE_NO_DATA
] = "NO_DATA",
655 [TFRC_RSTATE_DATA
] = "DATA",
656 [TFRC_RSTATE_TERM
] = "TERM",
659 return ccid3_rx_state_names
[state
];
663 static void ccid3_hc_rx_set_state(struct sock
*sk
,
664 enum ccid3_hc_rx_states state
)
666 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
667 enum ccid3_hc_rx_states oldstate
= hcrx
->ccid3hcrx_state
;
669 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
670 dccp_role(sk
), sk
, ccid3_rx_state_name(oldstate
),
671 ccid3_rx_state_name(state
));
672 WARN_ON(state
== oldstate
);
673 hcrx
->ccid3hcrx_state
= state
;
676 static void ccid3_hc_rx_send_feedback(struct sock
*sk
,
677 const struct sk_buff
*skb
,
678 enum ccid3_fback_type fbtype
)
680 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
681 struct dccp_sock
*dp
= dccp_sk(sk
);
685 if (unlikely(hcrx
->ccid3hcrx_state
== TFRC_RSTATE_TERM
))
688 now
= ktime_get_real();
691 case CCID3_FBACK_INITIAL
:
692 hcrx
->ccid3hcrx_x_recv
= 0;
693 hcrx
->ccid3hcrx_pinv
= ~0U; /* see RFC 4342, 8.5 */
695 case CCID3_FBACK_PARAM_CHANGE
:
697 * When parameters change (new loss or p > p_prev), we do not
698 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
699 * need to reuse the previous value of X_recv. However, when
700 * X_recv was 0 (due to early loss), this would kill X down to
701 * s/t_mbi (i.e. one packet in 64 seconds).
702 * To avoid such drastic reduction, we approximate X_recv as
703 * the number of bytes since last feedback.
704 * This is a safe fallback, since X is bounded above by X_calc.
706 if (hcrx
->ccid3hcrx_x_recv
> 0)
709 case CCID3_FBACK_PERIODIC
:
710 delta
= ktime_us_delta(now
, hcrx
->ccid3hcrx_tstamp_last_feedback
);
712 DCCP_BUG("delta (%ld) <= 0", (long)delta
);
714 hcrx
->ccid3hcrx_x_recv
=
715 scaled_div32(hcrx
->ccid3hcrx_bytes_recv
, delta
);
721 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta
,
722 hcrx
->ccid3hcrx_x_recv
, hcrx
->ccid3hcrx_pinv
);
724 hcrx
->ccid3hcrx_tstamp_last_feedback
= now
;
725 hcrx
->ccid3hcrx_last_counter
= dccp_hdr(skb
)->dccph_ccval
;
726 hcrx
->ccid3hcrx_bytes_recv
= 0;
728 dp
->dccps_hc_rx_insert_options
= 1;
732 static int ccid3_hc_rx_insert_options(struct sock
*sk
, struct sk_buff
*skb
)
734 const struct ccid3_hc_rx_sock
*hcrx
;
737 if (!(sk
->sk_state
== DCCP_OPEN
|| sk
->sk_state
== DCCP_PARTOPEN
))
740 hcrx
= ccid3_hc_rx_sk(sk
);
742 if (dccp_packet_without_ack(skb
))
745 x_recv
= htonl(hcrx
->ccid3hcrx_x_recv
);
746 pinv
= htonl(hcrx
->ccid3hcrx_pinv
);
748 if (dccp_insert_option(sk
, skb
, TFRC_OPT_LOSS_EVENT_RATE
,
749 &pinv
, sizeof(pinv
)) ||
750 dccp_insert_option(sk
, skb
, TFRC_OPT_RECEIVE_RATE
,
751 &x_recv
, sizeof(x_recv
)))
757 /** ccid3_first_li - Implements [RFC 3448, 6.3.1]
759 * Determine the length of the first loss interval via inverse lookup.
760 * Assume that X_recv can be computed by the throughput equation
764 * Find some p such that f(p) = fval; return 1/p (scaled).
766 static u32
ccid3_first_li(struct sock
*sk
)
768 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
769 u32 x_recv
, p
, delta
;
772 if (hcrx
->ccid3hcrx_rtt
== 0) {
773 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
774 hcrx
->ccid3hcrx_rtt
= DCCP_FALLBACK_RTT
;
777 delta
= ktime_to_us(net_timedelta(hcrx
->ccid3hcrx_tstamp_last_feedback
));
778 x_recv
= scaled_div32(hcrx
->ccid3hcrx_bytes_recv
, delta
);
779 if (x_recv
== 0) { /* would also trigger divide-by-zero */
780 DCCP_WARN("X_recv==0\n");
781 if ((x_recv
= hcrx
->ccid3hcrx_x_recv
) == 0) {
782 DCCP_BUG("stored value of X_recv is zero");
787 fval
= scaled_div(hcrx
->ccid3hcrx_s
, hcrx
->ccid3hcrx_rtt
);
788 fval
= scaled_div32(fval
, x_recv
);
789 p
= tfrc_calc_x_reverse_lookup(fval
);
791 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
792 "loss rate=%u\n", dccp_role(sk
), sk
, x_recv
, p
);
794 return p
== 0 ? ~0U : scaled_div(1, p
);
797 static void ccid3_hc_rx_packet_recv(struct sock
*sk
, struct sk_buff
*skb
)
799 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
800 enum ccid3_fback_type do_feedback
= CCID3_FBACK_NONE
;
801 const u32 ndp
= dccp_sk(sk
)->dccps_options_received
.dccpor_ndp
;
802 const bool is_data_packet
= dccp_data_packet(skb
);
804 if (unlikely(hcrx
->ccid3hcrx_state
== TFRC_RSTATE_NO_DATA
)) {
805 if (is_data_packet
) {
806 const u32 payload
= skb
->len
- dccp_hdr(skb
)->dccph_doff
* 4;
807 do_feedback
= CCID3_FBACK_INITIAL
;
808 ccid3_hc_rx_set_state(sk
, TFRC_RSTATE_DATA
);
809 hcrx
->ccid3hcrx_s
= payload
;
811 * Not necessary to update ccid3hcrx_bytes_recv here,
812 * since X_recv = 0 for the first feedback packet (cf.
813 * RFC 3448, 6.3) -- gerrit
819 if (tfrc_rx_hist_duplicate(&hcrx
->ccid3hcrx_hist
, skb
))
820 return; /* done receiving */
822 if (is_data_packet
) {
823 const u32 payload
= skb
->len
- dccp_hdr(skb
)->dccph_doff
* 4;
825 * Update moving-average of s and the sum of received payload bytes
827 hcrx
->ccid3hcrx_s
= tfrc_ewma(hcrx
->ccid3hcrx_s
, payload
, 9);
828 hcrx
->ccid3hcrx_bytes_recv
+= payload
;
832 * Handle pending losses and otherwise check for new loss
834 if (tfrc_rx_hist_loss_pending(&hcrx
->ccid3hcrx_hist
) &&
835 tfrc_rx_handle_loss(&hcrx
->ccid3hcrx_hist
,
836 &hcrx
->ccid3hcrx_li_hist
,
837 skb
, ndp
, ccid3_first_li
, sk
) ) {
838 do_feedback
= CCID3_FBACK_PARAM_CHANGE
;
842 if (tfrc_rx_hist_new_loss_indicated(&hcrx
->ccid3hcrx_hist
, skb
, ndp
))
846 * Handle data packets: RTT sampling and monitoring p
848 if (unlikely(!is_data_packet
))
851 if (!tfrc_lh_is_initialised(&hcrx
->ccid3hcrx_li_hist
)) {
852 const u32 sample
= tfrc_rx_hist_sample_rtt(&hcrx
->ccid3hcrx_hist
, skb
);
854 * Empty loss history: no loss so far, hence p stays 0.
855 * Sample RTT values, since an RTT estimate is required for the
856 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
859 hcrx
->ccid3hcrx_rtt
= tfrc_ewma(hcrx
->ccid3hcrx_rtt
, sample
, 9);
861 } else if (tfrc_lh_update_i_mean(&hcrx
->ccid3hcrx_li_hist
, skb
)) {
863 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
864 * has decreased (resp. p has increased), send feedback now.
866 do_feedback
= CCID3_FBACK_PARAM_CHANGE
;
870 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
872 if (SUB16(dccp_hdr(skb
)->dccph_ccval
, hcrx
->ccid3hcrx_last_counter
) > 3)
873 do_feedback
= CCID3_FBACK_PERIODIC
;
876 tfrc_rx_hist_add_packet(&hcrx
->ccid3hcrx_hist
, skb
, ndp
);
880 ccid3_hc_rx_send_feedback(sk
, skb
, do_feedback
);
883 static int ccid3_hc_rx_init(struct ccid
*ccid
, struct sock
*sk
)
885 struct ccid3_hc_rx_sock
*hcrx
= ccid_priv(ccid
);
887 hcrx
->ccid3hcrx_state
= TFRC_RSTATE_NO_DATA
;
888 tfrc_lh_init(&hcrx
->ccid3hcrx_li_hist
);
889 return tfrc_rx_hist_alloc(&hcrx
->ccid3hcrx_hist
);
892 static void ccid3_hc_rx_exit(struct sock
*sk
)
894 struct ccid3_hc_rx_sock
*hcrx
= ccid3_hc_rx_sk(sk
);
896 ccid3_hc_rx_set_state(sk
, TFRC_RSTATE_TERM
);
898 tfrc_rx_hist_purge(&hcrx
->ccid3hcrx_hist
);
899 tfrc_lh_cleanup(&hcrx
->ccid3hcrx_li_hist
);
902 static void ccid3_hc_rx_get_info(struct sock
*sk
, struct tcp_info
*info
)
904 const struct ccid3_hc_rx_sock
*hcrx
;
906 /* Listen socks doesn't have a private CCID block */
907 if (sk
->sk_state
== DCCP_LISTEN
)
910 hcrx
= ccid3_hc_rx_sk(sk
);
911 info
->tcpi_ca_state
= hcrx
->ccid3hcrx_state
;
912 info
->tcpi_options
|= TCPI_OPT_TIMESTAMPS
;
913 info
->tcpi_rcv_rtt
= hcrx
->ccid3hcrx_rtt
;
916 static int ccid3_hc_rx_getsockopt(struct sock
*sk
, const int optname
, int len
,
917 u32 __user
*optval
, int __user
*optlen
)
919 const struct ccid3_hc_rx_sock
*hcrx
;
922 /* Listen socks doesn't have a private CCID block */
923 if (sk
->sk_state
== DCCP_LISTEN
)
926 hcrx
= ccid3_hc_rx_sk(sk
);
928 case DCCP_SOCKOPT_CCID_RX_INFO
:
929 if (len
< sizeof(hcrx
->ccid3hcrx_tfrc
))
931 len
= sizeof(hcrx
->ccid3hcrx_tfrc
);
932 val
= &hcrx
->ccid3hcrx_tfrc
;
938 if (put_user(len
, optlen
) || copy_to_user(optval
, val
, len
))
944 static struct ccid_operations ccid3
= {
945 .ccid_id
= DCCPC_CCID3
,
946 .ccid_name
= "TCP-Friendly Rate Control",
947 .ccid_owner
= THIS_MODULE
,
948 .ccid_hc_tx_obj_size
= sizeof(struct ccid3_hc_tx_sock
),
949 .ccid_hc_tx_init
= ccid3_hc_tx_init
,
950 .ccid_hc_tx_exit
= ccid3_hc_tx_exit
,
951 .ccid_hc_tx_send_packet
= ccid3_hc_tx_send_packet
,
952 .ccid_hc_tx_packet_sent
= ccid3_hc_tx_packet_sent
,
953 .ccid_hc_tx_packet_recv
= ccid3_hc_tx_packet_recv
,
954 .ccid_hc_tx_parse_options
= ccid3_hc_tx_parse_options
,
955 .ccid_hc_rx_obj_size
= sizeof(struct ccid3_hc_rx_sock
),
956 .ccid_hc_rx_init
= ccid3_hc_rx_init
,
957 .ccid_hc_rx_exit
= ccid3_hc_rx_exit
,
958 .ccid_hc_rx_insert_options
= ccid3_hc_rx_insert_options
,
959 .ccid_hc_rx_packet_recv
= ccid3_hc_rx_packet_recv
,
960 .ccid_hc_rx_get_info
= ccid3_hc_rx_get_info
,
961 .ccid_hc_tx_get_info
= ccid3_hc_tx_get_info
,
962 .ccid_hc_rx_getsockopt
= ccid3_hc_rx_getsockopt
,
963 .ccid_hc_tx_getsockopt
= ccid3_hc_tx_getsockopt
,
966 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
967 module_param(ccid3_debug
, bool, 0444);
968 MODULE_PARM_DESC(ccid3_debug
, "Enable debug messages");
971 static __init
int ccid3_module_init(void)
973 return ccid_register(&ccid3
);
975 module_init(ccid3_module_init
);
977 static __exit
void ccid3_module_exit(void)
979 ccid_unregister(&ccid3
);
981 module_exit(ccid3_module_exit
);
983 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
984 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
985 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
986 MODULE_LICENSE("GPL");
987 MODULE_ALIAS("net-dccp-ccid-3");