~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/dccp/ccids/ccid2.c

Version: ~ [ linux-5.7 ] ~ [ linux-5.6.15 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.43 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.125 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.182 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.225 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.225 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.84 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  Copyright (c) 2005, 2006 Andrea Bittau <a.bittau@cs.ucl.ac.uk>
  3  *
  4  *  Changes to meet Linux coding standards, and DCCP infrastructure fixes.
  5  *
  6  *  Copyright (c) 2006 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  7  *
  8  *  This program is free software; you can redistribute it and/or modify
  9  *  it under the terms of the GNU General Public License as published by
 10  *  the Free Software Foundation; either version 2 of the License, or
 11  *  (at your option) any later version.
 12  *
 13  *  This program is distributed in the hope that it will be useful,
 14  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 15  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 16  *  GNU General Public License for more details.
 17  *
 18  *  You should have received a copy of the GNU General Public License
 19  *  along with this program; if not, write to the Free Software
 20  *  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 21  */
 22 
 23 /*
 24  * This implementation should follow RFC 4341
 25  */
 26 #include <linux/slab.h>
 27 #include "../feat.h"
 28 #include "ccid2.h"
 29 
 30 
 31 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
 32 static bool ccid2_debug;
 33 #define ccid2_pr_debug(format, a...)    DCCP_PR_DEBUG(ccid2_debug, format, ##a)
 34 #else
 35 #define ccid2_pr_debug(format, a...)
 36 #endif
 37 
 38 static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hc)
 39 {
 40         struct ccid2_seq *seqp;
 41         int i;
 42 
 43         /* check if we have space to preserve the pointer to the buffer */
 44         if (hc->tx_seqbufc >= (sizeof(hc->tx_seqbuf) /
 45                                sizeof(struct ccid2_seq *)))
 46                 return -ENOMEM;
 47 
 48         /* allocate buffer and initialize linked list */
 49         seqp = kmalloc_array(CCID2_SEQBUF_LEN, sizeof(struct ccid2_seq),
 50                              gfp_any());
 51         if (seqp == NULL)
 52                 return -ENOMEM;
 53 
 54         for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) {
 55                 seqp[i].ccid2s_next = &seqp[i + 1];
 56                 seqp[i + 1].ccid2s_prev = &seqp[i];
 57         }
 58         seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp;
 59         seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
 60 
 61         /* This is the first allocation.  Initiate the head and tail.  */
 62         if (hc->tx_seqbufc == 0)
 63                 hc->tx_seqh = hc->tx_seqt = seqp;
 64         else {
 65                 /* link the existing list with the one we just created */
 66                 hc->tx_seqh->ccid2s_next = seqp;
 67                 seqp->ccid2s_prev = hc->tx_seqh;
 68 
 69                 hc->tx_seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
 70                 seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hc->tx_seqt;
 71         }
 72 
 73         /* store the original pointer to the buffer so we can free it */
 74         hc->tx_seqbuf[hc->tx_seqbufc] = seqp;
 75         hc->tx_seqbufc++;
 76 
 77         return 0;
 78 }
 79 
 80 static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
 81 {
 82         if (ccid2_cwnd_network_limited(ccid2_hc_tx_sk(sk)))
 83                 return CCID_PACKET_WILL_DEQUEUE_LATER;
 84         return CCID_PACKET_SEND_AT_ONCE;
 85 }
 86 
 87 static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
 88 {
 89         u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->tx_cwnd, 2);
 90 
 91         /*
 92          * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from
 93          * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always
 94          * acceptable since this causes starvation/deadlock whenever cwnd < 2.
 95          * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled).
 96          */
 97         if (val == 0 || val > max_ratio) {
 98                 DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio);
 99                 val = max_ratio;
100         }
101         dccp_feat_signal_nn_change(sk, DCCPF_ACK_RATIO,
102                                    min_t(u32, val, DCCPF_ACK_RATIO_MAX));
103 }
104 
105 static void ccid2_check_l_ack_ratio(struct sock *sk)
106 {
107         struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
108 
109         /*
110          * After a loss, idle period, application limited period, or RTO we
111          * need to check that the ack ratio is still less than the congestion
112          * window. Otherwise, we will send an entire congestion window of
113          * packets and got no response because we haven't sent ack ratio
114          * packets yet.
115          * If the ack ratio does need to be reduced, we reduce it to half of
116          * the congestion window (or 1 if that's zero) instead of to the
117          * congestion window. This prevents problems if one ack is lost.
118          */
119         if (dccp_feat_nn_get(sk, DCCPF_ACK_RATIO) > hc->tx_cwnd)
120                 ccid2_change_l_ack_ratio(sk, hc->tx_cwnd/2 ? : 1U);
121 }
122 
123 static void ccid2_change_l_seq_window(struct sock *sk, u64 val)
124 {
125         dccp_feat_signal_nn_change(sk, DCCPF_SEQUENCE_WINDOW,
126                                    clamp_val(val, DCCPF_SEQ_WMIN,
127                                                   DCCPF_SEQ_WMAX));
128 }
129 
130 static void dccp_tasklet_schedule(struct sock *sk)
131 {
132         struct tasklet_struct *t = &dccp_sk(sk)->dccps_xmitlet;
133 
134         if (!test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
135                 sock_hold(sk);
136                 __tasklet_schedule(t);
137         }
138 }
139 
140 static void ccid2_hc_tx_rto_expire(struct timer_list *t)
141 {
142         struct ccid2_hc_tx_sock *hc = from_timer(hc, t, tx_rtotimer);
143         struct sock *sk = hc->sk;
144         const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
145 
146         bh_lock_sock(sk);
147         if (sock_owned_by_user(sk)) {
148                 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5);
149                 goto out;
150         }
151 
152         ccid2_pr_debug("RTO_EXPIRE\n");
153 
154         if (sk->sk_state == DCCP_CLOSED)
155                 goto out;
156 
157         /* back-off timer */
158         hc->tx_rto <<= 1;
159         if (hc->tx_rto > DCCP_RTO_MAX)
160                 hc->tx_rto = DCCP_RTO_MAX;
161 
162         /* adjust pipe, cwnd etc */
163         hc->tx_ssthresh = hc->tx_cwnd / 2;
164         if (hc->tx_ssthresh < 2)
165                 hc->tx_ssthresh = 2;
166         hc->tx_cwnd     = 1;
167         hc->tx_pipe     = 0;
168 
169         /* clear state about stuff we sent */
170         hc->tx_seqt = hc->tx_seqh;
171         hc->tx_packets_acked = 0;
172 
173         /* clear ack ratio state. */
174         hc->tx_rpseq    = 0;
175         hc->tx_rpdupack = -1;
176         ccid2_change_l_ack_ratio(sk, 1);
177 
178         /* if we were blocked before, we may now send cwnd=1 packet */
179         if (sender_was_blocked)
180                 dccp_tasklet_schedule(sk);
181         /* restart backed-off timer */
182         sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
183 out:
184         bh_unlock_sock(sk);
185         sock_put(sk);
186 }
187 
188 /*
189  *      Congestion window validation (RFC 2861).
190  */
191 static bool ccid2_do_cwv = true;
192 module_param(ccid2_do_cwv, bool, 0644);
193 MODULE_PARM_DESC(ccid2_do_cwv, "Perform RFC2861 Congestion Window Validation");
194 
195 /**
196  * ccid2_update_used_window  -  Track how much of cwnd is actually used
197  * This is done in addition to CWV. The sender needs to have an idea of how many
198  * packets may be in flight, to set the local Sequence Window value accordingly
199  * (RFC 4340, 7.5.2). The CWV mechanism is exploited to keep track of the
200  * maximum-used window. We use an EWMA low-pass filter to filter out noise.
201  */
202 static void ccid2_update_used_window(struct ccid2_hc_tx_sock *hc, u32 new_wnd)
203 {
204         hc->tx_expected_wnd = (3 * hc->tx_expected_wnd + new_wnd) / 4;
205 }
206 
207 /* This borrows the code of tcp_cwnd_application_limited() */
208 static void ccid2_cwnd_application_limited(struct sock *sk, const u32 now)
209 {
210         struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
211         /* don't reduce cwnd below the initial window (IW) */
212         u32 init_win = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache),
213             win_used = max(hc->tx_cwnd_used, init_win);
214 
215         if (win_used < hc->tx_cwnd) {
216                 hc->tx_ssthresh = max(hc->tx_ssthresh,
217                                      (hc->tx_cwnd >> 1) + (hc->tx_cwnd >> 2));
218                 hc->tx_cwnd = (hc->tx_cwnd + win_used) >> 1;
219         }
220         hc->tx_cwnd_used  = 0;
221         hc->tx_cwnd_stamp = now;
222 
223         ccid2_check_l_ack_ratio(sk);
224 }
225 
226 /* This borrows the code of tcp_cwnd_restart() */
227 static void ccid2_cwnd_restart(struct sock *sk, const u32 now)
228 {
229         struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
230         u32 cwnd = hc->tx_cwnd, restart_cwnd,
231             iwnd = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache);
232         s32 delta = now - hc->tx_lsndtime;
233 
234         hc->tx_ssthresh = max(hc->tx_ssthresh, (cwnd >> 1) + (cwnd >> 2));
235 
236         /* don't reduce cwnd below the initial window (IW) */
237         restart_cwnd = min(cwnd, iwnd);
238 
239         while ((delta -= hc->tx_rto) >= 0 && cwnd > restart_cwnd)
240                 cwnd >>= 1;
241         hc->tx_cwnd = max(cwnd, restart_cwnd);
242         hc->tx_cwnd_stamp = now;
243         hc->tx_cwnd_used  = 0;
244 
245         ccid2_check_l_ack_ratio(sk);
246 }
247 
248 static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
249 {
250         struct dccp_sock *dp = dccp_sk(sk);
251         struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
252         const u32 now = ccid2_jiffies32;
253         struct ccid2_seq *next;
254 
255         /* slow-start after idle periods (RFC 2581, RFC 2861) */
256         if (ccid2_do_cwv && !hc->tx_pipe &&
257             (s32)(now - hc->tx_lsndtime) >= hc->tx_rto)
258                 ccid2_cwnd_restart(sk, now);
259 
260         hc->tx_lsndtime = now;
261         hc->tx_pipe    += 1;
262 
263         /* see whether cwnd was fully used (RFC 2861), update expected window */
264         if (ccid2_cwnd_network_limited(hc)) {
265                 ccid2_update_used_window(hc, hc->tx_cwnd);
266                 hc->tx_cwnd_used  = 0;
267                 hc->tx_cwnd_stamp = now;
268         } else {
269                 if (hc->tx_pipe > hc->tx_cwnd_used)
270                         hc->tx_cwnd_used = hc->tx_pipe;
271 
272                 ccid2_update_used_window(hc, hc->tx_cwnd_used);
273 
274                 if (ccid2_do_cwv && (s32)(now - hc->tx_cwnd_stamp) >= hc->tx_rto)
275                         ccid2_cwnd_application_limited(sk, now);
276         }
277 
278         hc->tx_seqh->ccid2s_seq   = dp->dccps_gss;
279         hc->tx_seqh->ccid2s_acked = 0;
280         hc->tx_seqh->ccid2s_sent  = now;
281 
282         next = hc->tx_seqh->ccid2s_next;
283         /* check if we need to alloc more space */
284         if (next == hc->tx_seqt) {
285                 if (ccid2_hc_tx_alloc_seq(hc)) {
286                         DCCP_CRIT("packet history - out of memory!");
287                         /* FIXME: find a more graceful way to bail out */
288                         return;
289                 }
290                 next = hc->tx_seqh->ccid2s_next;
291                 BUG_ON(next == hc->tx_seqt);
292         }
293         hc->tx_seqh = next;
294 
295         ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe);
296 
297         /*
298          * FIXME: The code below is broken and the variables have been removed
299          * from the socket struct. The `ackloss' variable was always set to 0,
300          * and with arsent there are several problems:
301          *  (i) it doesn't just count the number of Acks, but all sent packets;
302          *  (ii) it is expressed in # of packets, not # of windows, so the
303          *  comparison below uses the wrong formula: Appendix A of RFC 4341
304          *  comes up with the number K = cwnd / (R^2 - R) of consecutive windows
305          *  of data with no lost or marked Ack packets. If arsent were the # of
306          *  consecutive Acks received without loss, then Ack Ratio needs to be
307          *  decreased by 1 when
308          *            arsent >=  K * cwnd / R  =  cwnd^2 / (R^3 - R^2)
309          *  where cwnd / R is the number of Acks received per window of data
310          *  (cf. RFC 4341, App. A). The problems are that
311          *  - arsent counts other packets as well;
312          *  - the comparison uses a formula different from RFC 4341;
313          *  - computing a cubic/quadratic equation each time is too complicated.
314          *  Hence a different algorithm is needed.
315          */
316 #if 0
317         /* Ack Ratio.  Need to maintain a concept of how many windows we sent */
318         hc->tx_arsent++;
319         /* We had an ack loss in this window... */
320         if (hc->tx_ackloss) {
321                 if (hc->tx_arsent >= hc->tx_cwnd) {
322                         hc->tx_arsent  = 0;
323                         hc->tx_ackloss = 0;
324                 }
325         } else {
326                 /* No acks lost up to now... */
327                 /* decrease ack ratio if enough packets were sent */
328                 if (dp->dccps_l_ack_ratio > 1) {
329                         /* XXX don't calculate denominator each time */
330                         int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
331                                     dp->dccps_l_ack_ratio;
332 
333                         denom = hc->tx_cwnd * hc->tx_cwnd / denom;
334 
335                         if (hc->tx_arsent >= denom) {
336                                 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
337                                 hc->tx_arsent = 0;
338                         }
339                 } else {
340                         /* we can't increase ack ratio further [1] */
341                         hc->tx_arsent = 0; /* or maybe set it to cwnd*/
342                 }
343         }
344 #endif
345 
346         sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
347 
348 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
349         do {
350                 struct ccid2_seq *seqp = hc->tx_seqt;
351 
352                 while (seqp != hc->tx_seqh) {
353                         ccid2_pr_debug("out seq=%llu acked=%d time=%u\n",
354                                        (unsigned long long)seqp->ccid2s_seq,
355                                        seqp->ccid2s_acked, seqp->ccid2s_sent);
356                         seqp = seqp->ccid2s_next;
357                 }
358         } while (0);
359         ccid2_pr_debug("=========\n");
360 #endif
361 }
362 
363 /**
364  * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
365  * This code is almost identical with TCP's tcp_rtt_estimator(), since
366  * - it has a higher sampling frequency (recommended by RFC 1323),
367  * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
368  * - it is simple (cf. more complex proposals such as Eifel timer or research
369  *   which suggests that the gain should be set according to window size),
370  * - in tests it was found to work well with CCID2 [gerrit].
371  */
372 static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
373 {
374         struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
375         long m = mrtt ? : 1;
376 
377         if (hc->tx_srtt == 0) {
378                 /* First measurement m */
379                 hc->tx_srtt = m << 3;
380                 hc->tx_mdev = m << 1;
381 
382                 hc->tx_mdev_max = max(hc->tx_mdev, tcp_rto_min(sk));
383                 hc->tx_rttvar   = hc->tx_mdev_max;
384 
385                 hc->tx_rtt_seq  = dccp_sk(sk)->dccps_gss;
386         } else {
387                 /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
388                 m -= (hc->tx_srtt >> 3);
389                 hc->tx_srtt += m;
390 
391                 /* Similarly, update scaled mdev with regard to |m| */
392                 if (m < 0) {
393                         m = -m;
394                         m -= (hc->tx_mdev >> 2);
395                         /*
396                          * This neutralises RTO increase when RTT < SRTT - mdev
397                          * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
398                          * in Linux TCP", USENIX 2002, pp. 49-62).
399                          */
400                         if (m > 0)
401                                 m >>= 3;
402                 } else {
403                         m -= (hc->tx_mdev >> 2);
404                 }
405                 hc->tx_mdev += m;
406 
407                 if (hc->tx_mdev > hc->tx_mdev_max) {
408                         hc->tx_mdev_max = hc->tx_mdev;
409                         if (hc->tx_mdev_max > hc->tx_rttvar)
410                                 hc->tx_rttvar = hc->tx_mdev_max;
411                 }
412 
413                 /*
414                  * Decay RTTVAR at most once per flight, exploiting that
415                  *  1) pipe <= cwnd <= Sequence_Window = W  (RFC 4340, 7.5.2)
416                  *  2) AWL = GSS-W+1 <= GAR <= GSS          (RFC 4340, 7.5.1)
417                  * GAR is a useful bound for FlightSize = pipe.
418                  * AWL is probably too low here, as it over-estimates pipe.
419                  */
420                 if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) {
421                         if (hc->tx_mdev_max < hc->tx_rttvar)
422                                 hc->tx_rttvar -= (hc->tx_rttvar -
423                                                   hc->tx_mdev_max) >> 2;
424                         hc->tx_rtt_seq  = dccp_sk(sk)->dccps_gss;
425                         hc->tx_mdev_max = tcp_rto_min(sk);
426                 }
427         }
428 
429         /*
430          * Set RTO from SRTT and RTTVAR
431          * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms.
432          * This agrees with RFC 4341, 5:
433          *      "Because DCCP does not retransmit data, DCCP does not require
434          *       TCP's recommended minimum timeout of one second".
435          */
436         hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar;
437 
438         if (hc->tx_rto > DCCP_RTO_MAX)
439                 hc->tx_rto = DCCP_RTO_MAX;
440 }
441 
442 static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
443                           unsigned int *maxincr)
444 {
445         struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
446         struct dccp_sock *dp = dccp_sk(sk);
447         int r_seq_used = hc->tx_cwnd / dp->dccps_l_ack_ratio;
448 
449         if (hc->tx_cwnd < dp->dccps_l_seq_win &&
450             r_seq_used < dp->dccps_r_seq_win) {
451                 if (hc->tx_cwnd < hc->tx_ssthresh) {
452                         if (*maxincr > 0 && ++hc->tx_packets_acked >= 2) {
453                                 hc->tx_cwnd += 1;
454                                 *maxincr    -= 1;
455                                 hc->tx_packets_acked = 0;
456                         }
457                 } else if (++hc->tx_packets_acked >= hc->tx_cwnd) {
458                         hc->tx_cwnd += 1;
459                         hc->tx_packets_acked = 0;
460                 }
461         }
462 
463         /*
464          * Adjust the local sequence window and the ack ratio to allow about
465          * 5 times the number of packets in the network (RFC 4340 7.5.2)
466          */
467         if (r_seq_used * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_r_seq_win)
468                 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio * 2);
469         else if (r_seq_used * CCID2_WIN_CHANGE_FACTOR < dp->dccps_r_seq_win/2)
470                 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio / 2 ? : 1U);
471 
472         if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_l_seq_win)
473                 ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win * 2);
474         else if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR < dp->dccps_l_seq_win/2)
475                 ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win / 2);
476 
477         /*
478          * FIXME: RTT is sampled several times per acknowledgment (for each
479          * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
480          * This causes the RTT to be over-estimated, since the older entries
481          * in the Ack Vector have earlier sending times.
482          * The cleanest solution is to not use the ccid2s_sent field at all
483          * and instead use DCCP timestamps: requires changes in other places.
484          */
485         ccid2_rtt_estimator(sk, ccid2_jiffies32 - seqp->ccid2s_sent);
486 }
487 
488 static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
489 {
490         struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
491 
492         if ((s32)(seqp->ccid2s_sent - hc->tx_last_cong) < 0) {
493                 ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
494                 return;
495         }
496 
497         hc->tx_last_cong = ccid2_jiffies32;
498 
499         hc->tx_cwnd      = hc->tx_cwnd / 2 ? : 1U;
500         hc->tx_ssthresh  = max(hc->tx_cwnd, 2U);
501 
502         ccid2_check_l_ack_ratio(sk);
503 }
504 
505 static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
506                                      u8 option, u8 *optval, u8 optlen)
507 {
508         struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
509 
510         switch (option) {
511         case DCCPO_ACK_VECTOR_0:
512         case DCCPO_ACK_VECTOR_1:
513                 return dccp_ackvec_parsed_add(&hc->tx_av_chunks, optval, optlen,
514                                               option - DCCPO_ACK_VECTOR_0);
515         }
516         return 0;
517 }
518 
519 static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
520 {
521         struct dccp_sock *dp = dccp_sk(sk);
522         struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
523         const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
524         struct dccp_ackvec_parsed *avp;
525         u64 ackno, seqno;
526         struct ccid2_seq *seqp;
527         int done = 0;
528         unsigned int maxincr = 0;
529 
530         /* check reverse path congestion */
531         seqno = DCCP_SKB_CB(skb)->dccpd_seq;
532 
533         /* XXX this whole "algorithm" is broken.  Need to fix it to keep track
534          * of the seqnos of the dupacks so that rpseq and rpdupack are correct
535          * -sorbo.
536          */
537         /* need to bootstrap */
538         if (hc->tx_rpdupack == -1) {
539                 hc->tx_rpdupack = 0;
540                 hc->tx_rpseq    = seqno;
541         } else {
542                 /* check if packet is consecutive */
543                 if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1)
544                         hc->tx_rpseq = seqno;
545                 /* it's a later packet */
546                 else if (after48(seqno, hc->tx_rpseq)) {
547                         hc->tx_rpdupack++;
548 
549                         /* check if we got enough dupacks */
550                         if (hc->tx_rpdupack >= NUMDUPACK) {
551                                 hc->tx_rpdupack = -1; /* XXX lame */
552                                 hc->tx_rpseq    = 0;
553 #ifdef __CCID2_COPES_GRACEFULLY_WITH_ACK_CONGESTION_CONTROL__
554                                 /*
555                                  * FIXME: Ack Congestion Control is broken; in
556                                  * the current state instabilities occurred with
557                                  * Ack Ratios greater than 1; causing hang-ups
558                                  * and long RTO timeouts. This needs to be fixed
559                                  * before opening up dynamic changes. -- gerrit
560                                  */
561                                 ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
562 #endif
563                         }
564                 }
565         }
566 
567         /* check forward path congestion */
568         if (dccp_packet_without_ack(skb))
569                 return;
570 
571         /* still didn't send out new data packets */
572         if (hc->tx_seqh == hc->tx_seqt)
573                 goto done;
574 
575         ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
576         if (after48(ackno, hc->tx_high_ack))
577                 hc->tx_high_ack = ackno;
578 
579         seqp = hc->tx_seqt;
580         while (before48(seqp->ccid2s_seq, ackno)) {
581                 seqp = seqp->ccid2s_next;
582                 if (seqp == hc->tx_seqh) {
583                         seqp = hc->tx_seqh->ccid2s_prev;
584                         break;
585                 }
586         }
587 
588         /*
589          * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
590          * packets per acknowledgement. Rounding up avoids that cwnd is not
591          * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
592          */
593         if (hc->tx_cwnd < hc->tx_ssthresh)
594                 maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
595 
596         /* go through all ack vectors */
597         list_for_each_entry(avp, &hc->tx_av_chunks, node) {
598                 /* go through this ack vector */
599                 for (; avp->len--; avp->vec++) {
600                         u64 ackno_end_rl = SUB48(ackno,
601                                                  dccp_ackvec_runlen(avp->vec));
602 
603                         ccid2_pr_debug("ackvec %llu |%u,%u|\n",
604                                        (unsigned long long)ackno,
605                                        dccp_ackvec_state(avp->vec) >> 6,
606                                        dccp_ackvec_runlen(avp->vec));
607                         /* if the seqno we are analyzing is larger than the
608                          * current ackno, then move towards the tail of our
609                          * seqnos.
610                          */
611                         while (after48(seqp->ccid2s_seq, ackno)) {
612                                 if (seqp == hc->tx_seqt) {
613                                         done = 1;
614                                         break;
615                                 }
616                                 seqp = seqp->ccid2s_prev;
617                         }
618                         if (done)
619                                 break;
620 
621                         /* check all seqnos in the range of the vector
622                          * run length
623                          */
624                         while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
625                                 const u8 state = dccp_ackvec_state(avp->vec);
626 
627                                 /* new packet received or marked */
628                                 if (state != DCCPAV_NOT_RECEIVED &&
629                                     !seqp->ccid2s_acked) {
630                                         if (state == DCCPAV_ECN_MARKED)
631                                                 ccid2_congestion_event(sk,
632                                                                        seqp);
633                                         else
634                                                 ccid2_new_ack(sk, seqp,
635                                                               &maxincr);
636 
637                                         seqp->ccid2s_acked = 1;
638                                         ccid2_pr_debug("Got ack for %llu\n",
639                                                        (unsigned long long)seqp->ccid2s_seq);
640                                         hc->tx_pipe--;
641                                 }
642                                 if (seqp == hc->tx_seqt) {
643                                         done = 1;
644                                         break;
645                                 }
646                                 seqp = seqp->ccid2s_prev;
647                         }
648                         if (done)
649                                 break;
650 
651                         ackno = SUB48(ackno_end_rl, 1);
652                 }
653                 if (done)
654                         break;
655         }
656 
657         /* The state about what is acked should be correct now
658          * Check for NUMDUPACK
659          */
660         seqp = hc->tx_seqt;
661         while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) {
662                 seqp = seqp->ccid2s_next;
663                 if (seqp == hc->tx_seqh) {
664                         seqp = hc->tx_seqh->ccid2s_prev;
665                         break;
666                 }
667         }
668         done = 0;
669         while (1) {
670                 if (seqp->ccid2s_acked) {
671                         done++;
672                         if (done == NUMDUPACK)
673                                 break;
674                 }
675                 if (seqp == hc->tx_seqt)
676                         break;
677                 seqp = seqp->ccid2s_prev;
678         }
679 
680         /* If there are at least 3 acknowledgements, anything unacknowledged
681          * below the last sequence number is considered lost
682          */
683         if (done == NUMDUPACK) {
684                 struct ccid2_seq *last_acked = seqp;
685 
686                 /* check for lost packets */
687                 while (1) {
688                         if (!seqp->ccid2s_acked) {
689                                 ccid2_pr_debug("Packet lost: %llu\n",
690                                                (unsigned long long)seqp->ccid2s_seq);
691                                 /* XXX need to traverse from tail -> head in
692                                  * order to detect multiple congestion events in
693                                  * one ack vector.
694                                  */
695                                 ccid2_congestion_event(sk, seqp);
696                                 hc->tx_pipe--;
697                         }
698                         if (seqp == hc->tx_seqt)
699                                 break;
700                         seqp = seqp->ccid2s_prev;
701                 }
702 
703                 hc->tx_seqt = last_acked;
704         }
705 
706         /* trim acked packets in tail */
707         while (hc->tx_seqt != hc->tx_seqh) {
708                 if (!hc->tx_seqt->ccid2s_acked)
709                         break;
710 
711                 hc->tx_seqt = hc->tx_seqt->ccid2s_next;
712         }
713 
714         /* restart RTO timer if not all outstanding data has been acked */
715         if (hc->tx_pipe == 0)
716                 sk_stop_timer(sk, &hc->tx_rtotimer);
717         else
718                 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
719 done:
720         /* check if incoming Acks allow pending packets to be sent */
721         if (sender_was_blocked && !ccid2_cwnd_network_limited(hc))
722                 dccp_tasklet_schedule(sk);
723         dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
724 }
725 
726 static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
727 {
728         struct ccid2_hc_tx_sock *hc = ccid_priv(ccid);
729         struct dccp_sock *dp = dccp_sk(sk);
730         u32 max_ratio;
731 
732         /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
733         hc->tx_ssthresh = ~0U;
734 
735         /* Use larger initial windows (RFC 4341, section 5). */
736         hc->tx_cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache);
737         hc->tx_expected_wnd = hc->tx_cwnd;
738 
739         /* Make sure that Ack Ratio is enabled and within bounds. */
740         max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2);
741         if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
742                 dp->dccps_l_ack_ratio = max_ratio;
743 
744         /* XXX init ~ to window size... */
745         if (ccid2_hc_tx_alloc_seq(hc))
746                 return -ENOMEM;
747 
748         hc->tx_rto       = DCCP_TIMEOUT_INIT;
749         hc->tx_rpdupack  = -1;
750         hc->tx_last_cong = hc->tx_lsndtime = hc->tx_cwnd_stamp = ccid2_jiffies32;
751         hc->tx_cwnd_used = 0;
752         hc->sk           = sk;
753         timer_setup(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire, 0);
754         INIT_LIST_HEAD(&hc->tx_av_chunks);
755         return 0;
756 }
757 
758 static void ccid2_hc_tx_exit(struct sock *sk)
759 {
760         struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
761         int i;
762 
763         sk_stop_timer(sk, &hc->tx_rtotimer);
764 
765         for (i = 0; i < hc->tx_seqbufc; i++)
766                 kfree(hc->tx_seqbuf[i]);
767         hc->tx_seqbufc = 0;
768         dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
769 }
770 
771 static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
772 {
773         struct ccid2_hc_rx_sock *hc = ccid2_hc_rx_sk(sk);
774 
775         if (!dccp_data_packet(skb))
776                 return;
777 
778         if (++hc->rx_num_data_pkts >= dccp_sk(sk)->dccps_r_ack_ratio) {
779                 dccp_send_ack(sk);
780                 hc->rx_num_data_pkts = 0;
781         }
782 }
783 
784 struct ccid_operations ccid2_ops = {
785         .ccid_id                  = DCCPC_CCID2,
786         .ccid_name                = "TCP-like",
787         .ccid_hc_tx_obj_size      = sizeof(struct ccid2_hc_tx_sock),
788         .ccid_hc_tx_init          = ccid2_hc_tx_init,
789         .ccid_hc_tx_exit          = ccid2_hc_tx_exit,
790         .ccid_hc_tx_send_packet   = ccid2_hc_tx_send_packet,
791         .ccid_hc_tx_packet_sent   = ccid2_hc_tx_packet_sent,
792         .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
793         .ccid_hc_tx_packet_recv   = ccid2_hc_tx_packet_recv,
794         .ccid_hc_rx_obj_size      = sizeof(struct ccid2_hc_rx_sock),
795         .ccid_hc_rx_packet_recv   = ccid2_hc_rx_packet_recv,
796 };
797 
798 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
799 module_param(ccid2_debug, bool, 0644);
800 MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages");
801 #endif
802 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp