]>
Commit | Line | Data |
---|---|---|
2a91aa39 | 1 | /* |
2a91aa39 AB |
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 | /* | |
0e64e94e | 24 | * This implementation should follow RFC 4341 |
2a91aa39 | 25 | */ |
5a0e3ad6 | 26 | #include <linux/slab.h> |
e8ef967a | 27 | #include "../feat.h" |
2a91aa39 AB |
28 | #include "ccid2.h" |
29 | ||
2a91aa39 | 30 | |
8d424f6c | 31 | #ifdef CONFIG_IP_DCCP_CCID2_DEBUG |
84116716 GR |
32 | static int ccid2_debug; |
33 | #define ccid2_pr_debug(format, a...) DCCP_PR_DEBUG(ccid2_debug, format, ##a) | |
2a91aa39 | 34 | #else |
84116716 | 35 | #define ccid2_pr_debug(format, a...) |
2a91aa39 AB |
36 | #endif |
37 | ||
77d2dd93 | 38 | static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hc) |
07978aab AB |
39 | { |
40 | struct ccid2_seq *seqp; | |
41 | int i; | |
42 | ||
43 | /* check if we have space to preserve the pointer to the buffer */ | |
77d2dd93 GR |
44 | if (hc->tx_seqbufc >= (sizeof(hc->tx_seqbuf) / |
45 | sizeof(struct ccid2_seq *))) | |
07978aab AB |
46 | return -ENOMEM; |
47 | ||
48 | /* allocate buffer and initialize linked list */ | |
cd1f7d34 | 49 | seqp = kmalloc(CCID2_SEQBUF_LEN * sizeof(struct ccid2_seq), gfp_any()); |
07978aab AB |
50 | if (seqp == NULL) |
51 | return -ENOMEM; | |
52 | ||
cd1f7d34 | 53 | for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) { |
07978aab AB |
54 | seqp[i].ccid2s_next = &seqp[i + 1]; |
55 | seqp[i + 1].ccid2s_prev = &seqp[i]; | |
56 | } | |
cd1f7d34 GR |
57 | seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp; |
58 | seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1]; | |
07978aab AB |
59 | |
60 | /* This is the first allocation. Initiate the head and tail. */ | |
77d2dd93 GR |
61 | if (hc->tx_seqbufc == 0) |
62 | hc->tx_seqh = hc->tx_seqt = seqp; | |
07978aab AB |
63 | else { |
64 | /* link the existing list with the one we just created */ | |
77d2dd93 GR |
65 | hc->tx_seqh->ccid2s_next = seqp; |
66 | seqp->ccid2s_prev = hc->tx_seqh; | |
07978aab | 67 | |
77d2dd93 GR |
68 | hc->tx_seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1]; |
69 | seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hc->tx_seqt; | |
07978aab AB |
70 | } |
71 | ||
72 | /* store the original pointer to the buffer so we can free it */ | |
77d2dd93 GR |
73 | hc->tx_seqbuf[hc->tx_seqbufc] = seqp; |
74 | hc->tx_seqbufc++; | |
07978aab AB |
75 | |
76 | return 0; | |
77 | } | |
78 | ||
6b57c93d | 79 | static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb) |
2a91aa39 | 80 | { |
1c0e0a05 GR |
81 | if (ccid2_cwnd_network_limited(ccid2_hc_tx_sk(sk))) |
82 | return CCID_PACKET_WILL_DEQUEUE_LATER; | |
83 | return CCID_PACKET_SEND_AT_ONCE; | |
2a91aa39 AB |
84 | } |
85 | ||
df054e1d | 86 | static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val) |
2a91aa39 AB |
87 | { |
88 | struct dccp_sock *dp = dccp_sk(sk); | |
b1c00fe3 | 89 | u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->tx_cwnd, 2); |
d50ad163 | 90 | |
2a91aa39 | 91 | /* |
d50ad163 GR |
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). | |
2a91aa39 | 96 | */ |
d50ad163 GR |
97 | if (val == 0 || val > max_ratio) { |
98 | DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio); | |
99 | val = max_ratio; | |
2a91aa39 | 100 | } |
e8ef967a GR |
101 | if (val > DCCPF_ACK_RATIO_MAX) |
102 | val = DCCPF_ACK_RATIO_MAX; | |
2a91aa39 | 103 | |
d50ad163 GR |
104 | if (val == dp->dccps_l_ack_ratio) |
105 | return; | |
106 | ||
df054e1d | 107 | ccid2_pr_debug("changing local ack ratio to %u\n", val); |
2a91aa39 AB |
108 | dp->dccps_l_ack_ratio = val; |
109 | } | |
110 | ||
2a91aa39 AB |
111 | static void ccid2_hc_tx_rto_expire(unsigned long data) |
112 | { | |
113 | struct sock *sk = (struct sock *)data; | |
77d2dd93 | 114 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
1c0e0a05 | 115 | const bool sender_was_blocked = ccid2_cwnd_network_limited(hc); |
2a91aa39 | 116 | |
2a91aa39 AB |
117 | bh_lock_sock(sk); |
118 | if (sock_owned_by_user(sk)) { | |
77d2dd93 | 119 | sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5); |
2a91aa39 AB |
120 | goto out; |
121 | } | |
122 | ||
123 | ccid2_pr_debug("RTO_EXPIRE\n"); | |
124 | ||
2a91aa39 | 125 | /* back-off timer */ |
77d2dd93 | 126 | hc->tx_rto <<= 1; |
231cc2aa GR |
127 | if (hc->tx_rto > DCCP_RTO_MAX) |
128 | hc->tx_rto = DCCP_RTO_MAX; | |
410e27a4 | 129 | |
2a91aa39 | 130 | /* adjust pipe, cwnd etc */ |
77d2dd93 GR |
131 | hc->tx_ssthresh = hc->tx_cwnd / 2; |
132 | if (hc->tx_ssthresh < 2) | |
133 | hc->tx_ssthresh = 2; | |
67b67e36 GR |
134 | hc->tx_cwnd = 1; |
135 | hc->tx_pipe = 0; | |
2a91aa39 AB |
136 | |
137 | /* clear state about stuff we sent */ | |
77d2dd93 GR |
138 | hc->tx_seqt = hc->tx_seqh; |
139 | hc->tx_packets_acked = 0; | |
2a91aa39 AB |
140 | |
141 | /* clear ack ratio state. */ | |
77d2dd93 GR |
142 | hc->tx_rpseq = 0; |
143 | hc->tx_rpdupack = -1; | |
2a91aa39 | 144 | ccid2_change_l_ack_ratio(sk, 1); |
1c0e0a05 GR |
145 | |
146 | /* if we were blocked before, we may now send cwnd=1 packet */ | |
147 | if (sender_was_blocked) | |
148 | tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet); | |
149 | /* restart backed-off timer */ | |
150 | sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto); | |
2a91aa39 AB |
151 | out: |
152 | bh_unlock_sock(sk); | |
77ff72d5 | 153 | sock_put(sk); |
2a91aa39 AB |
154 | } |
155 | ||
113ced1f GR |
156 | /* |
157 | * Congestion window validation (RFC 2861). | |
158 | */ | |
159 | static int ccid2_do_cwv = 1; | |
160 | module_param(ccid2_do_cwv, bool, 0644); | |
161 | MODULE_PARM_DESC(ccid2_do_cwv, "Perform RFC2861 Congestion Window Validation"); | |
162 | ||
163 | /** | |
164 | * ccid2_update_used_window - Track how much of cwnd is actually used | |
165 | * This is done in addition to CWV. The sender needs to have an idea of how many | |
166 | * packets may be in flight, to set the local Sequence Window value accordingly | |
167 | * (RFC 4340, 7.5.2). The CWV mechanism is exploited to keep track of the | |
168 | * maximum-used window. We use an EWMA low-pass filter to filter out noise. | |
169 | */ | |
170 | static void ccid2_update_used_window(struct ccid2_hc_tx_sock *hc, u32 new_wnd) | |
171 | { | |
172 | hc->tx_expected_wnd = (3 * hc->tx_expected_wnd + new_wnd) / 4; | |
173 | } | |
174 | ||
175 | /* This borrows the code of tcp_cwnd_application_limited() */ | |
176 | static void ccid2_cwnd_application_limited(struct sock *sk, const u32 now) | |
177 | { | |
178 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); | |
179 | /* don't reduce cwnd below the initial window (IW) */ | |
180 | u32 init_win = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache), | |
181 | win_used = max(hc->tx_cwnd_used, init_win); | |
182 | ||
183 | if (win_used < hc->tx_cwnd) { | |
184 | hc->tx_ssthresh = max(hc->tx_ssthresh, | |
185 | (hc->tx_cwnd >> 1) + (hc->tx_cwnd >> 2)); | |
186 | hc->tx_cwnd = (hc->tx_cwnd + win_used) >> 1; | |
187 | } | |
188 | hc->tx_cwnd_used = 0; | |
189 | hc->tx_cwnd_stamp = now; | |
190 | } | |
191 | ||
192 | /* This borrows the code of tcp_cwnd_restart() */ | |
193 | static void ccid2_cwnd_restart(struct sock *sk, const u32 now) | |
194 | { | |
195 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); | |
196 | u32 cwnd = hc->tx_cwnd, restart_cwnd, | |
197 | iwnd = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache); | |
198 | ||
199 | hc->tx_ssthresh = max(hc->tx_ssthresh, (cwnd >> 1) + (cwnd >> 2)); | |
200 | ||
201 | /* don't reduce cwnd below the initial window (IW) */ | |
202 | restart_cwnd = min(cwnd, iwnd); | |
203 | cwnd >>= (now - hc->tx_lsndtime) / hc->tx_rto; | |
204 | hc->tx_cwnd = max(cwnd, restart_cwnd); | |
205 | ||
206 | hc->tx_cwnd_stamp = now; | |
207 | hc->tx_cwnd_used = 0; | |
208 | } | |
209 | ||
baf9e782 | 210 | static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len) |
2a91aa39 AB |
211 | { |
212 | struct dccp_sock *dp = dccp_sk(sk); | |
77d2dd93 | 213 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
113ced1f | 214 | const u32 now = ccid2_time_stamp; |
07978aab | 215 | struct ccid2_seq *next; |
2a91aa39 | 216 | |
113ced1f GR |
217 | /* slow-start after idle periods (RFC 2581, RFC 2861) */ |
218 | if (ccid2_do_cwv && !hc->tx_pipe && | |
219 | (s32)(now - hc->tx_lsndtime) >= hc->tx_rto) | |
220 | ccid2_cwnd_restart(sk, now); | |
221 | ||
222 | hc->tx_lsndtime = now; | |
223 | hc->tx_pipe += 1; | |
224 | ||
225 | /* see whether cwnd was fully used (RFC 2861), update expected window */ | |
226 | if (ccid2_cwnd_network_limited(hc)) { | |
227 | ccid2_update_used_window(hc, hc->tx_cwnd); | |
228 | hc->tx_cwnd_used = 0; | |
229 | hc->tx_cwnd_stamp = now; | |
230 | } else { | |
231 | if (hc->tx_pipe > hc->tx_cwnd_used) | |
232 | hc->tx_cwnd_used = hc->tx_pipe; | |
233 | ||
234 | ccid2_update_used_window(hc, hc->tx_cwnd_used); | |
235 | ||
236 | if (ccid2_do_cwv && (s32)(now - hc->tx_cwnd_stamp) >= hc->tx_rto) | |
237 | ccid2_cwnd_application_limited(sk, now); | |
238 | } | |
2a91aa39 | 239 | |
77d2dd93 GR |
240 | hc->tx_seqh->ccid2s_seq = dp->dccps_gss; |
241 | hc->tx_seqh->ccid2s_acked = 0; | |
113ced1f | 242 | hc->tx_seqh->ccid2s_sent = now; |
2a91aa39 | 243 | |
77d2dd93 | 244 | next = hc->tx_seqh->ccid2s_next; |
07978aab | 245 | /* check if we need to alloc more space */ |
77d2dd93 GR |
246 | if (next == hc->tx_seqt) { |
247 | if (ccid2_hc_tx_alloc_seq(hc)) { | |
7d9e8931 GR |
248 | DCCP_CRIT("packet history - out of memory!"); |
249 | /* FIXME: find a more graceful way to bail out */ | |
250 | return; | |
251 | } | |
77d2dd93 GR |
252 | next = hc->tx_seqh->ccid2s_next; |
253 | BUG_ON(next == hc->tx_seqt); | |
2a91aa39 | 254 | } |
77d2dd93 | 255 | hc->tx_seqh = next; |
07978aab | 256 | |
77d2dd93 | 257 | ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe); |
2a91aa39 | 258 | |
900bfed4 GR |
259 | /* |
260 | * FIXME: The code below is broken and the variables have been removed | |
261 | * from the socket struct. The `ackloss' variable was always set to 0, | |
262 | * and with arsent there are several problems: | |
263 | * (i) it doesn't just count the number of Acks, but all sent packets; | |
264 | * (ii) it is expressed in # of packets, not # of windows, so the | |
265 | * comparison below uses the wrong formula: Appendix A of RFC 4341 | |
266 | * comes up with the number K = cwnd / (R^2 - R) of consecutive windows | |
267 | * of data with no lost or marked Ack packets. If arsent were the # of | |
268 | * consecutive Acks received without loss, then Ack Ratio needs to be | |
269 | * decreased by 1 when | |
270 | * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2) | |
271 | * where cwnd / R is the number of Acks received per window of data | |
272 | * (cf. RFC 4341, App. A). The problems are that | |
273 | * - arsent counts other packets as well; | |
274 | * - the comparison uses a formula different from RFC 4341; | |
275 | * - computing a cubic/quadratic equation each time is too complicated. | |
276 | * Hence a different algorithm is needed. | |
277 | */ | |
278 | #if 0 | |
2a91aa39 | 279 | /* Ack Ratio. Need to maintain a concept of how many windows we sent */ |
77d2dd93 | 280 | hc->tx_arsent++; |
2a91aa39 | 281 | /* We had an ack loss in this window... */ |
77d2dd93 GR |
282 | if (hc->tx_ackloss) { |
283 | if (hc->tx_arsent >= hc->tx_cwnd) { | |
284 | hc->tx_arsent = 0; | |
285 | hc->tx_ackloss = 0; | |
2a91aa39 | 286 | } |
c0c736db ACM |
287 | } else { |
288 | /* No acks lost up to now... */ | |
2a91aa39 AB |
289 | /* decrease ack ratio if enough packets were sent */ |
290 | if (dp->dccps_l_ack_ratio > 1) { | |
291 | /* XXX don't calculate denominator each time */ | |
c0c736db ACM |
292 | int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio - |
293 | dp->dccps_l_ack_ratio; | |
2a91aa39 | 294 | |
77d2dd93 | 295 | denom = hc->tx_cwnd * hc->tx_cwnd / denom; |
2a91aa39 | 296 | |
77d2dd93 | 297 | if (hc->tx_arsent >= denom) { |
2a91aa39 | 298 | ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1); |
77d2dd93 | 299 | hc->tx_arsent = 0; |
2a91aa39 | 300 | } |
c0c736db ACM |
301 | } else { |
302 | /* we can't increase ack ratio further [1] */ | |
77d2dd93 | 303 | hc->tx_arsent = 0; /* or maybe set it to cwnd*/ |
2a91aa39 AB |
304 | } |
305 | } | |
900bfed4 | 306 | #endif |
2a91aa39 | 307 | |
d26eeb07 | 308 | sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto); |
c0c736db | 309 | |
8d424f6c | 310 | #ifdef CONFIG_IP_DCCP_CCID2_DEBUG |
2a91aa39 | 311 | do { |
77d2dd93 | 312 | struct ccid2_seq *seqp = hc->tx_seqt; |
2a91aa39 | 313 | |
77d2dd93 | 314 | while (seqp != hc->tx_seqh) { |
d82b6f85 | 315 | ccid2_pr_debug("out seq=%llu acked=%d time=%u\n", |
8109b02b | 316 | (unsigned long long)seqp->ccid2s_seq, |
234af484 | 317 | seqp->ccid2s_acked, seqp->ccid2s_sent); |
2a91aa39 AB |
318 | seqp = seqp->ccid2s_next; |
319 | } | |
c0c736db | 320 | } while (0); |
2a91aa39 | 321 | ccid2_pr_debug("=========\n"); |
2a91aa39 AB |
322 | #endif |
323 | } | |
324 | ||
231cc2aa GR |
325 | /** |
326 | * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm | |
327 | * This code is almost identical with TCP's tcp_rtt_estimator(), since | |
328 | * - it has a higher sampling frequency (recommended by RFC 1323), | |
329 | * - the RTO does not collapse into RTT due to RTTVAR going towards zero, | |
330 | * - it is simple (cf. more complex proposals such as Eifel timer or research | |
331 | * which suggests that the gain should be set according to window size), | |
332 | * - in tests it was found to work well with CCID2 [gerrit]. | |
333 | */ | |
334 | static void ccid2_rtt_estimator(struct sock *sk, const long mrtt) | |
335 | { | |
336 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); | |
337 | long m = mrtt ? : 1; | |
338 | ||
339 | if (hc->tx_srtt == 0) { | |
340 | /* First measurement m */ | |
341 | hc->tx_srtt = m << 3; | |
342 | hc->tx_mdev = m << 1; | |
343 | ||
4886fcad | 344 | hc->tx_mdev_max = max(hc->tx_mdev, tcp_rto_min(sk)); |
231cc2aa | 345 | hc->tx_rttvar = hc->tx_mdev_max; |
4886fcad | 346 | |
231cc2aa GR |
347 | hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss; |
348 | } else { | |
349 | /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */ | |
350 | m -= (hc->tx_srtt >> 3); | |
351 | hc->tx_srtt += m; | |
352 | ||
353 | /* Similarly, update scaled mdev with regard to |m| */ | |
354 | if (m < 0) { | |
355 | m = -m; | |
356 | m -= (hc->tx_mdev >> 2); | |
357 | /* | |
358 | * This neutralises RTO increase when RTT < SRTT - mdev | |
359 | * (see P. Sarolahti, A. Kuznetsov,"Congestion Control | |
360 | * in Linux TCP", USENIX 2002, pp. 49-62). | |
361 | */ | |
362 | if (m > 0) | |
363 | m >>= 3; | |
364 | } else { | |
365 | m -= (hc->tx_mdev >> 2); | |
366 | } | |
367 | hc->tx_mdev += m; | |
368 | ||
369 | if (hc->tx_mdev > hc->tx_mdev_max) { | |
370 | hc->tx_mdev_max = hc->tx_mdev; | |
371 | if (hc->tx_mdev_max > hc->tx_rttvar) | |
372 | hc->tx_rttvar = hc->tx_mdev_max; | |
373 | } | |
374 | ||
375 | /* | |
376 | * Decay RTTVAR at most once per flight, exploiting that | |
377 | * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2) | |
378 | * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1) | |
379 | * GAR is a useful bound for FlightSize = pipe. | |
380 | * AWL is probably too low here, as it over-estimates pipe. | |
381 | */ | |
382 | if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) { | |
383 | if (hc->tx_mdev_max < hc->tx_rttvar) | |
384 | hc->tx_rttvar -= (hc->tx_rttvar - | |
385 | hc->tx_mdev_max) >> 2; | |
386 | hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss; | |
4886fcad | 387 | hc->tx_mdev_max = tcp_rto_min(sk); |
231cc2aa GR |
388 | } |
389 | } | |
390 | ||
391 | /* | |
392 | * Set RTO from SRTT and RTTVAR | |
393 | * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms. | |
394 | * This agrees with RFC 4341, 5: | |
395 | * "Because DCCP does not retransmit data, DCCP does not require | |
396 | * TCP's recommended minimum timeout of one second". | |
397 | */ | |
398 | hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar; | |
399 | ||
400 | if (hc->tx_rto > DCCP_RTO_MAX) | |
401 | hc->tx_rto = DCCP_RTO_MAX; | |
402 | } | |
403 | ||
404 | static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp, | |
405 | unsigned int *maxincr) | |
374bcf32 | 406 | { |
77d2dd93 | 407 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
d50ad163 | 408 | |
77d2dd93 GR |
409 | if (hc->tx_cwnd < hc->tx_ssthresh) { |
410 | if (*maxincr > 0 && ++hc->tx_packets_acked == 2) { | |
411 | hc->tx_cwnd += 1; | |
412 | *maxincr -= 1; | |
413 | hc->tx_packets_acked = 0; | |
410e27a4 | 414 | } |
77d2dd93 GR |
415 | } else if (++hc->tx_packets_acked >= hc->tx_cwnd) { |
416 | hc->tx_cwnd += 1; | |
417 | hc->tx_packets_acked = 0; | |
374bcf32 | 418 | } |
231cc2aa GR |
419 | /* |
420 | * FIXME: RTT is sampled several times per acknowledgment (for each | |
421 | * entry in the Ack Vector), instead of once per Ack (as in TCP SACK). | |
422 | * This causes the RTT to be over-estimated, since the older entries | |
423 | * in the Ack Vector have earlier sending times. | |
424 | * The cleanest solution is to not use the ccid2s_sent field at all | |
425 | * and instead use DCCP timestamps: requires changes in other places. | |
426 | */ | |
d82b6f85 | 427 | ccid2_rtt_estimator(sk, ccid2_time_stamp - seqp->ccid2s_sent); |
410e27a4 GR |
428 | } |
429 | ||
430 | static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp) | |
431 | { | |
77d2dd93 | 432 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
410e27a4 | 433 | |
d82b6f85 | 434 | if ((s32)(seqp->ccid2s_sent - hc->tx_last_cong) < 0) { |
410e27a4 GR |
435 | ccid2_pr_debug("Multiple losses in an RTT---treating as one\n"); |
436 | return; | |
c8bf462b | 437 | } |
410e27a4 | 438 | |
d82b6f85 | 439 | hc->tx_last_cong = ccid2_time_stamp; |
410e27a4 | 440 | |
77d2dd93 GR |
441 | hc->tx_cwnd = hc->tx_cwnd / 2 ? : 1U; |
442 | hc->tx_ssthresh = max(hc->tx_cwnd, 2U); | |
410e27a4 GR |
443 | |
444 | /* Avoid spurious timeouts resulting from Ack Ratio > cwnd */ | |
77d2dd93 GR |
445 | if (dccp_sk(sk)->dccps_l_ack_ratio > hc->tx_cwnd) |
446 | ccid2_change_l_ack_ratio(sk, hc->tx_cwnd); | |
c8bf462b GR |
447 | } |
448 | ||
7e87fe84 GR |
449 | static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type, |
450 | u8 option, u8 *optval, u8 optlen) | |
451 | { | |
452 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); | |
453 | ||
454 | switch (option) { | |
455 | case DCCPO_ACK_VECTOR_0: | |
456 | case DCCPO_ACK_VECTOR_1: | |
457 | return dccp_ackvec_parsed_add(&hc->tx_av_chunks, optval, optlen, | |
458 | option - DCCPO_ACK_VECTOR_0); | |
459 | } | |
460 | return 0; | |
461 | } | |
462 | ||
2a91aa39 AB |
463 | static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb) |
464 | { | |
465 | struct dccp_sock *dp = dccp_sk(sk); | |
77d2dd93 | 466 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
1c0e0a05 | 467 | const bool sender_was_blocked = ccid2_cwnd_network_limited(hc); |
7e87fe84 | 468 | struct dccp_ackvec_parsed *avp; |
2a91aa39 AB |
469 | u64 ackno, seqno; |
470 | struct ccid2_seq *seqp; | |
2a91aa39 | 471 | int done = 0; |
2a91aa39 AB |
472 | unsigned int maxincr = 0; |
473 | ||
2a91aa39 AB |
474 | /* check reverse path congestion */ |
475 | seqno = DCCP_SKB_CB(skb)->dccpd_seq; | |
476 | ||
477 | /* XXX this whole "algorithm" is broken. Need to fix it to keep track | |
478 | * of the seqnos of the dupacks so that rpseq and rpdupack are correct | |
479 | * -sorbo. | |
480 | */ | |
481 | /* need to bootstrap */ | |
77d2dd93 GR |
482 | if (hc->tx_rpdupack == -1) { |
483 | hc->tx_rpdupack = 0; | |
484 | hc->tx_rpseq = seqno; | |
c0c736db | 485 | } else { |
2a91aa39 | 486 | /* check if packet is consecutive */ |
77d2dd93 GR |
487 | if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1) |
488 | hc->tx_rpseq = seqno; | |
2a91aa39 | 489 | /* it's a later packet */ |
77d2dd93 GR |
490 | else if (after48(seqno, hc->tx_rpseq)) { |
491 | hc->tx_rpdupack++; | |
2a91aa39 AB |
492 | |
493 | /* check if we got enough dupacks */ | |
77d2dd93 GR |
494 | if (hc->tx_rpdupack >= NUMDUPACK) { |
495 | hc->tx_rpdupack = -1; /* XXX lame */ | |
496 | hc->tx_rpseq = 0; | |
31daf039 GR |
497 | #ifdef __CCID2_COPES_GRACEFULLY_WITH_ACK_CONGESTION_CONTROL__ |
498 | /* | |
499 | * FIXME: Ack Congestion Control is broken; in | |
500 | * the current state instabilities occurred with | |
501 | * Ack Ratios greater than 1; causing hang-ups | |
502 | * and long RTO timeouts. This needs to be fixed | |
503 | * before opening up dynamic changes. -- gerrit | |
504 | */ | |
df054e1d | 505 | ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio); |
31daf039 | 506 | #endif |
2a91aa39 AB |
507 | } |
508 | } | |
509 | } | |
510 | ||
511 | /* check forward path congestion */ | |
7e87fe84 | 512 | if (dccp_packet_without_ack(skb)) |
2a91aa39 AB |
513 | return; |
514 | ||
7e87fe84 GR |
515 | /* still didn't send out new data packets */ |
516 | if (hc->tx_seqh == hc->tx_seqt) | |
517 | goto done; | |
2a91aa39 AB |
518 | |
519 | ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq; | |
77d2dd93 GR |
520 | if (after48(ackno, hc->tx_high_ack)) |
521 | hc->tx_high_ack = ackno; | |
32aac18d | 522 | |
77d2dd93 | 523 | seqp = hc->tx_seqt; |
32aac18d AB |
524 | while (before48(seqp->ccid2s_seq, ackno)) { |
525 | seqp = seqp->ccid2s_next; | |
77d2dd93 GR |
526 | if (seqp == hc->tx_seqh) { |
527 | seqp = hc->tx_seqh->ccid2s_prev; | |
32aac18d AB |
528 | break; |
529 | } | |
530 | } | |
2a91aa39 | 531 | |
a3020025 GR |
532 | /* |
533 | * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2 | |
534 | * packets per acknowledgement. Rounding up avoids that cwnd is not | |
535 | * advanced when Ack Ratio is 1 and gives a slight edge otherwise. | |
2a91aa39 | 536 | */ |
77d2dd93 | 537 | if (hc->tx_cwnd < hc->tx_ssthresh) |
a3020025 | 538 | maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2); |
2a91aa39 AB |
539 | |
540 | /* go through all ack vectors */ | |
7e87fe84 | 541 | list_for_each_entry(avp, &hc->tx_av_chunks, node) { |
2a91aa39 | 542 | /* go through this ack vector */ |
7e87fe84 GR |
543 | for (; avp->len--; avp->vec++) { |
544 | u64 ackno_end_rl = SUB48(ackno, | |
545 | dccp_ackvec_runlen(avp->vec)); | |
2a91aa39 | 546 | |
7e87fe84 | 547 | ccid2_pr_debug("ackvec %llu |%u,%u|\n", |
234af484 | 548 | (unsigned long long)ackno, |
7e87fe84 GR |
549 | dccp_ackvec_state(avp->vec) >> 6, |
550 | dccp_ackvec_runlen(avp->vec)); | |
2a91aa39 AB |
551 | /* if the seqno we are analyzing is larger than the |
552 | * current ackno, then move towards the tail of our | |
553 | * seqnos. | |
554 | */ | |
555 | while (after48(seqp->ccid2s_seq, ackno)) { | |
77d2dd93 | 556 | if (seqp == hc->tx_seqt) { |
2a91aa39 AB |
557 | done = 1; |
558 | break; | |
559 | } | |
560 | seqp = seqp->ccid2s_prev; | |
561 | } | |
562 | if (done) | |
563 | break; | |
564 | ||
565 | /* check all seqnos in the range of the vector | |
566 | * run length | |
567 | */ | |
568 | while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) { | |
7e87fe84 | 569 | const u8 state = dccp_ackvec_state(avp->vec); |
2a91aa39 AB |
570 | |
571 | /* new packet received or marked */ | |
f17a37c9 | 572 | if (state != DCCPAV_NOT_RECEIVED && |
2a91aa39 | 573 | !seqp->ccid2s_acked) { |
f17a37c9 | 574 | if (state == DCCPAV_ECN_MARKED) |
d50ad163 | 575 | ccid2_congestion_event(sk, |
374bcf32 | 576 | seqp); |
f17a37c9 | 577 | else |
2a91aa39 AB |
578 | ccid2_new_ack(sk, seqp, |
579 | &maxincr); | |
2a91aa39 AB |
580 | |
581 | seqp->ccid2s_acked = 1; | |
582 | ccid2_pr_debug("Got ack for %llu\n", | |
234af484 | 583 | (unsigned long long)seqp->ccid2s_seq); |
c38c92a8 | 584 | hc->tx_pipe--; |
2a91aa39 | 585 | } |
77d2dd93 | 586 | if (seqp == hc->tx_seqt) { |
2a91aa39 AB |
587 | done = 1; |
588 | break; | |
589 | } | |
3de5489f | 590 | seqp = seqp->ccid2s_prev; |
2a91aa39 AB |
591 | } |
592 | if (done) | |
593 | break; | |
594 | ||
cfbbeabc | 595 | ackno = SUB48(ackno_end_rl, 1); |
2a91aa39 AB |
596 | } |
597 | if (done) | |
598 | break; | |
599 | } | |
600 | ||
601 | /* The state about what is acked should be correct now | |
602 | * Check for NUMDUPACK | |
603 | */ | |
77d2dd93 GR |
604 | seqp = hc->tx_seqt; |
605 | while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) { | |
32aac18d | 606 | seqp = seqp->ccid2s_next; |
77d2dd93 GR |
607 | if (seqp == hc->tx_seqh) { |
608 | seqp = hc->tx_seqh->ccid2s_prev; | |
32aac18d AB |
609 | break; |
610 | } | |
611 | } | |
2a91aa39 AB |
612 | done = 0; |
613 | while (1) { | |
614 | if (seqp->ccid2s_acked) { | |
615 | done++; | |
63df18ad | 616 | if (done == NUMDUPACK) |
2a91aa39 | 617 | break; |
2a91aa39 | 618 | } |
77d2dd93 | 619 | if (seqp == hc->tx_seqt) |
2a91aa39 | 620 | break; |
2a91aa39 AB |
621 | seqp = seqp->ccid2s_prev; |
622 | } | |
623 | ||
624 | /* If there are at least 3 acknowledgements, anything unacknowledged | |
625 | * below the last sequence number is considered lost | |
626 | */ | |
63df18ad | 627 | if (done == NUMDUPACK) { |
2a91aa39 AB |
628 | struct ccid2_seq *last_acked = seqp; |
629 | ||
630 | /* check for lost packets */ | |
631 | while (1) { | |
632 | if (!seqp->ccid2s_acked) { | |
374bcf32 | 633 | ccid2_pr_debug("Packet lost: %llu\n", |
234af484 | 634 | (unsigned long long)seqp->ccid2s_seq); |
374bcf32 AB |
635 | /* XXX need to traverse from tail -> head in |
636 | * order to detect multiple congestion events in | |
637 | * one ack vector. | |
638 | */ | |
d50ad163 | 639 | ccid2_congestion_event(sk, seqp); |
c38c92a8 | 640 | hc->tx_pipe--; |
2a91aa39 | 641 | } |
77d2dd93 | 642 | if (seqp == hc->tx_seqt) |
2a91aa39 AB |
643 | break; |
644 | seqp = seqp->ccid2s_prev; | |
645 | } | |
646 | ||
77d2dd93 | 647 | hc->tx_seqt = last_acked; |
2a91aa39 AB |
648 | } |
649 | ||
650 | /* trim acked packets in tail */ | |
77d2dd93 GR |
651 | while (hc->tx_seqt != hc->tx_seqh) { |
652 | if (!hc->tx_seqt->ccid2s_acked) | |
2a91aa39 AB |
653 | break; |
654 | ||
77d2dd93 | 655 | hc->tx_seqt = hc->tx_seqt->ccid2s_next; |
2a91aa39 | 656 | } |
c38c92a8 GR |
657 | |
658 | /* restart RTO timer if not all outstanding data has been acked */ | |
659 | if (hc->tx_pipe == 0) | |
660 | sk_stop_timer(sk, &hc->tx_rtotimer); | |
661 | else | |
662 | sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto); | |
7e87fe84 | 663 | done: |
1c0e0a05 GR |
664 | /* check if incoming Acks allow pending packets to be sent */ |
665 | if (sender_was_blocked && !ccid2_cwnd_network_limited(hc)) | |
666 | tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet); | |
7e87fe84 | 667 | dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks); |
2a91aa39 AB |
668 | } |
669 | ||
91f0ebf7 | 670 | static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk) |
2a91aa39 | 671 | { |
77d2dd93 | 672 | struct ccid2_hc_tx_sock *hc = ccid_priv(ccid); |
b00d2bbc GR |
673 | struct dccp_sock *dp = dccp_sk(sk); |
674 | u32 max_ratio; | |
2a91aa39 | 675 | |
b00d2bbc | 676 | /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */ |
77d2dd93 | 677 | hc->tx_ssthresh = ~0U; |
2a91aa39 | 678 | |
22b71c8f GR |
679 | /* Use larger initial windows (RFC 4341, section 5). */ |
680 | hc->tx_cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache); | |
113ced1f | 681 | hc->tx_expected_wnd = hc->tx_cwnd; |
b00d2bbc GR |
682 | |
683 | /* Make sure that Ack Ratio is enabled and within bounds. */ | |
77d2dd93 | 684 | max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2); |
b00d2bbc GR |
685 | if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio) |
686 | dp->dccps_l_ack_ratio = max_ratio; | |
687 | ||
2a91aa39 | 688 | /* XXX init ~ to window size... */ |
77d2dd93 | 689 | if (ccid2_hc_tx_alloc_seq(hc)) |
2a91aa39 | 690 | return -ENOMEM; |
91f0ebf7 | 691 | |
231cc2aa | 692 | hc->tx_rto = DCCP_TIMEOUT_INIT; |
77d2dd93 | 693 | hc->tx_rpdupack = -1; |
113ced1f GR |
694 | hc->tx_last_cong = hc->tx_lsndtime = hc->tx_cwnd_stamp = ccid2_time_stamp; |
695 | hc->tx_cwnd_used = 0; | |
77d2dd93 | 696 | setup_timer(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire, |
410e27a4 | 697 | (unsigned long)sk); |
7e87fe84 | 698 | INIT_LIST_HEAD(&hc->tx_av_chunks); |
2a91aa39 AB |
699 | return 0; |
700 | } | |
701 | ||
702 | static void ccid2_hc_tx_exit(struct sock *sk) | |
703 | { | |
77d2dd93 | 704 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
07978aab | 705 | int i; |
2a91aa39 | 706 | |
d26eeb07 | 707 | sk_stop_timer(sk, &hc->tx_rtotimer); |
07978aab | 708 | |
77d2dd93 GR |
709 | for (i = 0; i < hc->tx_seqbufc; i++) |
710 | kfree(hc->tx_seqbuf[i]); | |
711 | hc->tx_seqbufc = 0; | |
2a91aa39 AB |
712 | } |
713 | ||
714 | static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb) | |
715 | { | |
77d2dd93 | 716 | struct ccid2_hc_rx_sock *hc = ccid2_hc_rx_sk(sk); |
2a91aa39 | 717 | |
58fdea0f GR |
718 | if (!dccp_data_packet(skb)) |
719 | return; | |
720 | ||
721 | if (++hc->rx_num_data_pkts >= dccp_sk(sk)->dccps_r_ack_ratio) { | |
722 | dccp_send_ack(sk); | |
723 | hc->rx_num_data_pkts = 0; | |
2a91aa39 AB |
724 | } |
725 | } | |
726 | ||
ddebc973 | 727 | struct ccid_operations ccid2_ops = { |
7e87fe84 GR |
728 | .ccid_id = DCCPC_CCID2, |
729 | .ccid_name = "TCP-like", | |
730 | .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock), | |
731 | .ccid_hc_tx_init = ccid2_hc_tx_init, | |
732 | .ccid_hc_tx_exit = ccid2_hc_tx_exit, | |
733 | .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet, | |
734 | .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent, | |
735 | .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options, | |
736 | .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv, | |
737 | .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock), | |
738 | .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv, | |
2a91aa39 AB |
739 | }; |
740 | ||
84116716 | 741 | #ifdef CONFIG_IP_DCCP_CCID2_DEBUG |
43264991 | 742 | module_param(ccid2_debug, bool, 0644); |
ddebc973 | 743 | MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages"); |
84116716 | 744 | #endif |