]>
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 "../ccid.h" |
29 | #include "../dccp.h" | |
30 | #include "ccid2.h" | |
31 | ||
2a91aa39 | 32 | |
8d424f6c | 33 | #ifdef CONFIG_IP_DCCP_CCID2_DEBUG |
84116716 GR |
34 | static int ccid2_debug; |
35 | #define ccid2_pr_debug(format, a...) DCCP_PR_DEBUG(ccid2_debug, format, ##a) | |
2a91aa39 | 36 | #else |
84116716 | 37 | #define ccid2_pr_debug(format, a...) |
2a91aa39 AB |
38 | #endif |
39 | ||
77d2dd93 | 40 | static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hc) |
07978aab AB |
41 | { |
42 | struct ccid2_seq *seqp; | |
43 | int i; | |
44 | ||
45 | /* check if we have space to preserve the pointer to the buffer */ | |
77d2dd93 GR |
46 | if (hc->tx_seqbufc >= (sizeof(hc->tx_seqbuf) / |
47 | sizeof(struct ccid2_seq *))) | |
07978aab AB |
48 | return -ENOMEM; |
49 | ||
50 | /* allocate buffer and initialize linked list */ | |
cd1f7d34 | 51 | seqp = kmalloc(CCID2_SEQBUF_LEN * sizeof(struct ccid2_seq), gfp_any()); |
07978aab AB |
52 | if (seqp == NULL) |
53 | return -ENOMEM; | |
54 | ||
cd1f7d34 | 55 | for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) { |
07978aab AB |
56 | seqp[i].ccid2s_next = &seqp[i + 1]; |
57 | seqp[i + 1].ccid2s_prev = &seqp[i]; | |
58 | } | |
cd1f7d34 GR |
59 | seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp; |
60 | seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1]; | |
07978aab AB |
61 | |
62 | /* This is the first allocation. Initiate the head and tail. */ | |
77d2dd93 GR |
63 | if (hc->tx_seqbufc == 0) |
64 | hc->tx_seqh = hc->tx_seqt = seqp; | |
07978aab AB |
65 | else { |
66 | /* link the existing list with the one we just created */ | |
77d2dd93 GR |
67 | hc->tx_seqh->ccid2s_next = seqp; |
68 | seqp->ccid2s_prev = hc->tx_seqh; | |
07978aab | 69 | |
77d2dd93 GR |
70 | hc->tx_seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1]; |
71 | seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hc->tx_seqt; | |
07978aab AB |
72 | } |
73 | ||
74 | /* store the original pointer to the buffer so we can free it */ | |
77d2dd93 GR |
75 | hc->tx_seqbuf[hc->tx_seqbufc] = seqp; |
76 | hc->tx_seqbufc++; | |
07978aab AB |
77 | |
78 | return 0; | |
79 | } | |
80 | ||
6b57c93d | 81 | static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb) |
2a91aa39 | 82 | { |
77d2dd93 | 83 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
410e27a4 | 84 | |
77d2dd93 | 85 | if (hc->tx_pipe < hc->tx_cwnd) |
410e27a4 GR |
86 | return 0; |
87 | ||
88 | return 1; /* XXX CCID should dequeue when ready instead of polling */ | |
2a91aa39 AB |
89 | } |
90 | ||
df054e1d | 91 | static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val) |
2a91aa39 AB |
92 | { |
93 | struct dccp_sock *dp = dccp_sk(sk); | |
b1c00fe3 | 94 | u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->tx_cwnd, 2); |
d50ad163 | 95 | |
2a91aa39 | 96 | /* |
d50ad163 GR |
97 | * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from |
98 | * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always | |
99 | * acceptable since this causes starvation/deadlock whenever cwnd < 2. | |
100 | * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled). | |
2a91aa39 | 101 | */ |
d50ad163 GR |
102 | if (val == 0 || val > max_ratio) { |
103 | DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio); | |
104 | val = max_ratio; | |
2a91aa39 | 105 | } |
e8ef967a GR |
106 | if (val > DCCPF_ACK_RATIO_MAX) |
107 | val = DCCPF_ACK_RATIO_MAX; | |
2a91aa39 | 108 | |
d50ad163 GR |
109 | if (val == dp->dccps_l_ack_ratio) |
110 | return; | |
111 | ||
df054e1d | 112 | ccid2_pr_debug("changing local ack ratio to %u\n", val); |
2a91aa39 AB |
113 | dp->dccps_l_ack_ratio = val; |
114 | } | |
115 | ||
410e27a4 GR |
116 | static void ccid2_start_rto_timer(struct sock *sk); |
117 | ||
2a91aa39 AB |
118 | static void ccid2_hc_tx_rto_expire(unsigned long data) |
119 | { | |
120 | struct sock *sk = (struct sock *)data; | |
77d2dd93 | 121 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
2a91aa39 | 122 | |
2a91aa39 AB |
123 | bh_lock_sock(sk); |
124 | if (sock_owned_by_user(sk)) { | |
77d2dd93 | 125 | sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5); |
2a91aa39 AB |
126 | goto out; |
127 | } | |
128 | ||
129 | ccid2_pr_debug("RTO_EXPIRE\n"); | |
130 | ||
2a91aa39 | 131 | /* back-off timer */ |
77d2dd93 | 132 | hc->tx_rto <<= 1; |
231cc2aa GR |
133 | if (hc->tx_rto > DCCP_RTO_MAX) |
134 | hc->tx_rto = DCCP_RTO_MAX; | |
410e27a4 GR |
135 | |
136 | ccid2_start_rto_timer(sk); | |
2a91aa39 | 137 | |
2a91aa39 | 138 | /* adjust pipe, cwnd etc */ |
77d2dd93 GR |
139 | hc->tx_ssthresh = hc->tx_cwnd / 2; |
140 | if (hc->tx_ssthresh < 2) | |
141 | hc->tx_ssthresh = 2; | |
67b67e36 GR |
142 | hc->tx_cwnd = 1; |
143 | hc->tx_pipe = 0; | |
2a91aa39 AB |
144 | |
145 | /* clear state about stuff we sent */ | |
77d2dd93 GR |
146 | hc->tx_seqt = hc->tx_seqh; |
147 | hc->tx_packets_acked = 0; | |
2a91aa39 AB |
148 | |
149 | /* clear ack ratio state. */ | |
77d2dd93 GR |
150 | hc->tx_rpseq = 0; |
151 | hc->tx_rpdupack = -1; | |
2a91aa39 | 152 | ccid2_change_l_ack_ratio(sk, 1); |
2a91aa39 AB |
153 | out: |
154 | bh_unlock_sock(sk); | |
77ff72d5 | 155 | sock_put(sk); |
2a91aa39 AB |
156 | } |
157 | ||
410e27a4 GR |
158 | static void ccid2_start_rto_timer(struct sock *sk) |
159 | { | |
77d2dd93 | 160 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
410e27a4 | 161 | |
231cc2aa | 162 | ccid2_pr_debug("setting RTO timeout=%u\n", hc->tx_rto); |
410e27a4 | 163 | |
77d2dd93 GR |
164 | BUG_ON(timer_pending(&hc->tx_rtotimer)); |
165 | sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto); | |
410e27a4 GR |
166 | } |
167 | ||
168 | static void ccid2_hc_tx_packet_sent(struct sock *sk, int more, unsigned int len) | |
2a91aa39 AB |
169 | { |
170 | struct dccp_sock *dp = dccp_sk(sk); | |
77d2dd93 | 171 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
07978aab | 172 | struct ccid2_seq *next; |
2a91aa39 | 173 | |
77d2dd93 | 174 | hc->tx_pipe++; |
2a91aa39 | 175 | |
77d2dd93 GR |
176 | hc->tx_seqh->ccid2s_seq = dp->dccps_gss; |
177 | hc->tx_seqh->ccid2s_acked = 0; | |
178 | hc->tx_seqh->ccid2s_sent = jiffies; | |
2a91aa39 | 179 | |
77d2dd93 | 180 | next = hc->tx_seqh->ccid2s_next; |
07978aab | 181 | /* check if we need to alloc more space */ |
77d2dd93 GR |
182 | if (next == hc->tx_seqt) { |
183 | if (ccid2_hc_tx_alloc_seq(hc)) { | |
7d9e8931 GR |
184 | DCCP_CRIT("packet history - out of memory!"); |
185 | /* FIXME: find a more graceful way to bail out */ | |
186 | return; | |
187 | } | |
77d2dd93 GR |
188 | next = hc->tx_seqh->ccid2s_next; |
189 | BUG_ON(next == hc->tx_seqt); | |
2a91aa39 | 190 | } |
77d2dd93 | 191 | hc->tx_seqh = next; |
07978aab | 192 | |
77d2dd93 | 193 | ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe); |
2a91aa39 | 194 | |
900bfed4 GR |
195 | /* |
196 | * FIXME: The code below is broken and the variables have been removed | |
197 | * from the socket struct. The `ackloss' variable was always set to 0, | |
198 | * and with arsent there are several problems: | |
199 | * (i) it doesn't just count the number of Acks, but all sent packets; | |
200 | * (ii) it is expressed in # of packets, not # of windows, so the | |
201 | * comparison below uses the wrong formula: Appendix A of RFC 4341 | |
202 | * comes up with the number K = cwnd / (R^2 - R) of consecutive windows | |
203 | * of data with no lost or marked Ack packets. If arsent were the # of | |
204 | * consecutive Acks received without loss, then Ack Ratio needs to be | |
205 | * decreased by 1 when | |
206 | * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2) | |
207 | * where cwnd / R is the number of Acks received per window of data | |
208 | * (cf. RFC 4341, App. A). The problems are that | |
209 | * - arsent counts other packets as well; | |
210 | * - the comparison uses a formula different from RFC 4341; | |
211 | * - computing a cubic/quadratic equation each time is too complicated. | |
212 | * Hence a different algorithm is needed. | |
213 | */ | |
214 | #if 0 | |
2a91aa39 | 215 | /* Ack Ratio. Need to maintain a concept of how many windows we sent */ |
77d2dd93 | 216 | hc->tx_arsent++; |
2a91aa39 | 217 | /* We had an ack loss in this window... */ |
77d2dd93 GR |
218 | if (hc->tx_ackloss) { |
219 | if (hc->tx_arsent >= hc->tx_cwnd) { | |
220 | hc->tx_arsent = 0; | |
221 | hc->tx_ackloss = 0; | |
2a91aa39 | 222 | } |
c0c736db ACM |
223 | } else { |
224 | /* No acks lost up to now... */ | |
2a91aa39 AB |
225 | /* decrease ack ratio if enough packets were sent */ |
226 | if (dp->dccps_l_ack_ratio > 1) { | |
227 | /* XXX don't calculate denominator each time */ | |
c0c736db ACM |
228 | int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio - |
229 | dp->dccps_l_ack_ratio; | |
2a91aa39 | 230 | |
77d2dd93 | 231 | denom = hc->tx_cwnd * hc->tx_cwnd / denom; |
2a91aa39 | 232 | |
77d2dd93 | 233 | if (hc->tx_arsent >= denom) { |
2a91aa39 | 234 | ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1); |
77d2dd93 | 235 | hc->tx_arsent = 0; |
2a91aa39 | 236 | } |
c0c736db ACM |
237 | } else { |
238 | /* we can't increase ack ratio further [1] */ | |
77d2dd93 | 239 | hc->tx_arsent = 0; /* or maybe set it to cwnd*/ |
2a91aa39 AB |
240 | } |
241 | } | |
900bfed4 | 242 | #endif |
2a91aa39 AB |
243 | |
244 | /* setup RTO timer */ | |
77d2dd93 | 245 | if (!timer_pending(&hc->tx_rtotimer)) |
410e27a4 | 246 | ccid2_start_rto_timer(sk); |
c0c736db | 247 | |
8d424f6c | 248 | #ifdef CONFIG_IP_DCCP_CCID2_DEBUG |
2a91aa39 | 249 | do { |
77d2dd93 | 250 | struct ccid2_seq *seqp = hc->tx_seqt; |
2a91aa39 | 251 | |
77d2dd93 | 252 | while (seqp != hc->tx_seqh) { |
2a91aa39 | 253 | ccid2_pr_debug("out seq=%llu acked=%d time=%lu\n", |
8109b02b | 254 | (unsigned long long)seqp->ccid2s_seq, |
234af484 | 255 | seqp->ccid2s_acked, seqp->ccid2s_sent); |
2a91aa39 AB |
256 | seqp = seqp->ccid2s_next; |
257 | } | |
c0c736db | 258 | } while (0); |
2a91aa39 | 259 | ccid2_pr_debug("=========\n"); |
2a91aa39 AB |
260 | #endif |
261 | } | |
262 | ||
410e27a4 GR |
263 | /* XXX Lame code duplication! |
264 | * returns -1 if none was found. | |
265 | * else returns the next offset to use in the function call. | |
1435562d | 266 | */ |
410e27a4 GR |
267 | static int ccid2_ackvector(struct sock *sk, struct sk_buff *skb, int offset, |
268 | unsigned char **vec, unsigned char *veclen) | |
1435562d | 269 | { |
410e27a4 GR |
270 | const struct dccp_hdr *dh = dccp_hdr(skb); |
271 | unsigned char *options = (unsigned char *)dh + dccp_hdr_len(skb); | |
272 | unsigned char *opt_ptr; | |
273 | const unsigned char *opt_end = (unsigned char *)dh + | |
274 | (dh->dccph_doff * 4); | |
275 | unsigned char opt, len; | |
276 | unsigned char *value; | |
277 | ||
278 | BUG_ON(offset < 0); | |
279 | options += offset; | |
280 | opt_ptr = options; | |
281 | if (opt_ptr >= opt_end) | |
282 | return -1; | |
283 | ||
284 | while (opt_ptr != opt_end) { | |
285 | opt = *opt_ptr++; | |
286 | len = 0; | |
287 | value = NULL; | |
288 | ||
289 | /* Check if this isn't a single byte option */ | |
290 | if (opt > DCCPO_MAX_RESERVED) { | |
291 | if (opt_ptr == opt_end) | |
292 | goto out_invalid_option; | |
293 | ||
294 | len = *opt_ptr++; | |
295 | if (len < 3) | |
296 | goto out_invalid_option; | |
1435562d | 297 | /* |
410e27a4 GR |
298 | * Remove the type and len fields, leaving |
299 | * just the value size | |
1435562d | 300 | */ |
410e27a4 GR |
301 | len -= 2; |
302 | value = opt_ptr; | |
303 | opt_ptr += len; | |
1435562d | 304 | |
410e27a4 GR |
305 | if (opt_ptr > opt_end) |
306 | goto out_invalid_option; | |
1435562d GR |
307 | } |
308 | ||
410e27a4 GR |
309 | switch (opt) { |
310 | case DCCPO_ACK_VECTOR_0: | |
311 | case DCCPO_ACK_VECTOR_1: | |
312 | *vec = value; | |
313 | *veclen = len; | |
314 | return offset + (opt_ptr - options); | |
1435562d GR |
315 | } |
316 | } | |
317 | ||
410e27a4 | 318 | return -1; |
1435562d | 319 | |
410e27a4 GR |
320 | out_invalid_option: |
321 | DCCP_BUG("Invalid option - this should not happen (previous parsing)!"); | |
322 | return -1; | |
1435562d GR |
323 | } |
324 | ||
410e27a4 | 325 | static void ccid2_hc_tx_kill_rto_timer(struct sock *sk) |
2a91aa39 | 326 | { |
77d2dd93 | 327 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
2a91aa39 | 328 | |
77d2dd93 | 329 | sk_stop_timer(sk, &hc->tx_rtotimer); |
410e27a4 | 330 | ccid2_pr_debug("deleted RTO timer\n"); |
2a91aa39 AB |
331 | } |
332 | ||
231cc2aa GR |
333 | /** |
334 | * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm | |
335 | * This code is almost identical with TCP's tcp_rtt_estimator(), since | |
336 | * - it has a higher sampling frequency (recommended by RFC 1323), | |
337 | * - the RTO does not collapse into RTT due to RTTVAR going towards zero, | |
338 | * - it is simple (cf. more complex proposals such as Eifel timer or research | |
339 | * which suggests that the gain should be set according to window size), | |
340 | * - in tests it was found to work well with CCID2 [gerrit]. | |
341 | */ | |
342 | static void ccid2_rtt_estimator(struct sock *sk, const long mrtt) | |
343 | { | |
344 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); | |
345 | long m = mrtt ? : 1; | |
346 | ||
347 | if (hc->tx_srtt == 0) { | |
348 | /* First measurement m */ | |
349 | hc->tx_srtt = m << 3; | |
350 | hc->tx_mdev = m << 1; | |
351 | ||
352 | hc->tx_mdev_max = max(TCP_RTO_MIN, hc->tx_mdev); | |
353 | hc->tx_rttvar = hc->tx_mdev_max; | |
354 | hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss; | |
355 | } else { | |
356 | /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */ | |
357 | m -= (hc->tx_srtt >> 3); | |
358 | hc->tx_srtt += m; | |
359 | ||
360 | /* Similarly, update scaled mdev with regard to |m| */ | |
361 | if (m < 0) { | |
362 | m = -m; | |
363 | m -= (hc->tx_mdev >> 2); | |
364 | /* | |
365 | * This neutralises RTO increase when RTT < SRTT - mdev | |
366 | * (see P. Sarolahti, A. Kuznetsov,"Congestion Control | |
367 | * in Linux TCP", USENIX 2002, pp. 49-62). | |
368 | */ | |
369 | if (m > 0) | |
370 | m >>= 3; | |
371 | } else { | |
372 | m -= (hc->tx_mdev >> 2); | |
373 | } | |
374 | hc->tx_mdev += m; | |
375 | ||
376 | if (hc->tx_mdev > hc->tx_mdev_max) { | |
377 | hc->tx_mdev_max = hc->tx_mdev; | |
378 | if (hc->tx_mdev_max > hc->tx_rttvar) | |
379 | hc->tx_rttvar = hc->tx_mdev_max; | |
380 | } | |
381 | ||
382 | /* | |
383 | * Decay RTTVAR at most once per flight, exploiting that | |
384 | * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2) | |
385 | * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1) | |
386 | * GAR is a useful bound for FlightSize = pipe. | |
387 | * AWL is probably too low here, as it over-estimates pipe. | |
388 | */ | |
389 | if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) { | |
390 | if (hc->tx_mdev_max < hc->tx_rttvar) | |
391 | hc->tx_rttvar -= (hc->tx_rttvar - | |
392 | hc->tx_mdev_max) >> 2; | |
393 | hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss; | |
394 | hc->tx_mdev_max = TCP_RTO_MIN; | |
395 | } | |
396 | } | |
397 | ||
398 | /* | |
399 | * Set RTO from SRTT and RTTVAR | |
400 | * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms. | |
401 | * This agrees with RFC 4341, 5: | |
402 | * "Because DCCP does not retransmit data, DCCP does not require | |
403 | * TCP's recommended minimum timeout of one second". | |
404 | */ | |
405 | hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar; | |
406 | ||
407 | if (hc->tx_rto > DCCP_RTO_MAX) | |
408 | hc->tx_rto = DCCP_RTO_MAX; | |
409 | } | |
410 | ||
411 | static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp, | |
412 | unsigned int *maxincr) | |
374bcf32 | 413 | { |
77d2dd93 | 414 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
d50ad163 | 415 | |
77d2dd93 GR |
416 | if (hc->tx_cwnd < hc->tx_ssthresh) { |
417 | if (*maxincr > 0 && ++hc->tx_packets_acked == 2) { | |
418 | hc->tx_cwnd += 1; | |
419 | *maxincr -= 1; | |
420 | hc->tx_packets_acked = 0; | |
410e27a4 | 421 | } |
77d2dd93 GR |
422 | } else if (++hc->tx_packets_acked >= hc->tx_cwnd) { |
423 | hc->tx_cwnd += 1; | |
424 | hc->tx_packets_acked = 0; | |
374bcf32 | 425 | } |
231cc2aa GR |
426 | /* |
427 | * FIXME: RTT is sampled several times per acknowledgment (for each | |
428 | * entry in the Ack Vector), instead of once per Ack (as in TCP SACK). | |
429 | * This causes the RTT to be over-estimated, since the older entries | |
430 | * in the Ack Vector have earlier sending times. | |
431 | * The cleanest solution is to not use the ccid2s_sent field at all | |
432 | * and instead use DCCP timestamps: requires changes in other places. | |
433 | */ | |
434 | ccid2_rtt_estimator(sk, jiffies - seqp->ccid2s_sent); | |
410e27a4 GR |
435 | } |
436 | ||
437 | static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp) | |
438 | { | |
77d2dd93 | 439 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
410e27a4 | 440 | |
77d2dd93 | 441 | if (time_before(seqp->ccid2s_sent, hc->tx_last_cong)) { |
410e27a4 GR |
442 | ccid2_pr_debug("Multiple losses in an RTT---treating as one\n"); |
443 | return; | |
c8bf462b | 444 | } |
410e27a4 | 445 | |
77d2dd93 | 446 | hc->tx_last_cong = jiffies; |
410e27a4 | 447 | |
77d2dd93 GR |
448 | hc->tx_cwnd = hc->tx_cwnd / 2 ? : 1U; |
449 | hc->tx_ssthresh = max(hc->tx_cwnd, 2U); | |
410e27a4 GR |
450 | |
451 | /* Avoid spurious timeouts resulting from Ack Ratio > cwnd */ | |
77d2dd93 GR |
452 | if (dccp_sk(sk)->dccps_l_ack_ratio > hc->tx_cwnd) |
453 | ccid2_change_l_ack_ratio(sk, hc->tx_cwnd); | |
c8bf462b GR |
454 | } |
455 | ||
2a91aa39 AB |
456 | static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb) |
457 | { | |
458 | struct dccp_sock *dp = dccp_sk(sk); | |
77d2dd93 | 459 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
2a91aa39 AB |
460 | u64 ackno, seqno; |
461 | struct ccid2_seq *seqp; | |
410e27a4 GR |
462 | unsigned char *vector; |
463 | unsigned char veclen; | |
464 | int offset = 0; | |
2a91aa39 | 465 | int done = 0; |
2a91aa39 AB |
466 | unsigned int maxincr = 0; |
467 | ||
2a91aa39 AB |
468 | /* check reverse path congestion */ |
469 | seqno = DCCP_SKB_CB(skb)->dccpd_seq; | |
470 | ||
471 | /* XXX this whole "algorithm" is broken. Need to fix it to keep track | |
472 | * of the seqnos of the dupacks so that rpseq and rpdupack are correct | |
473 | * -sorbo. | |
474 | */ | |
475 | /* need to bootstrap */ | |
77d2dd93 GR |
476 | if (hc->tx_rpdupack == -1) { |
477 | hc->tx_rpdupack = 0; | |
478 | hc->tx_rpseq = seqno; | |
c0c736db | 479 | } else { |
2a91aa39 | 480 | /* check if packet is consecutive */ |
77d2dd93 GR |
481 | if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1) |
482 | hc->tx_rpseq = seqno; | |
2a91aa39 | 483 | /* it's a later packet */ |
77d2dd93 GR |
484 | else if (after48(seqno, hc->tx_rpseq)) { |
485 | hc->tx_rpdupack++; | |
2a91aa39 AB |
486 | |
487 | /* check if we got enough dupacks */ | |
77d2dd93 GR |
488 | if (hc->tx_rpdupack >= NUMDUPACK) { |
489 | hc->tx_rpdupack = -1; /* XXX lame */ | |
490 | hc->tx_rpseq = 0; | |
2a91aa39 | 491 | |
df054e1d | 492 | ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio); |
2a91aa39 AB |
493 | } |
494 | } | |
495 | } | |
496 | ||
497 | /* check forward path congestion */ | |
410e27a4 | 498 | /* still didn't send out new data packets */ |
77d2dd93 | 499 | if (hc->tx_seqh == hc->tx_seqt) |
2a91aa39 AB |
500 | return; |
501 | ||
410e27a4 GR |
502 | switch (DCCP_SKB_CB(skb)->dccpd_type) { |
503 | case DCCP_PKT_ACK: | |
504 | case DCCP_PKT_DATAACK: | |
505 | break; | |
506 | default: | |
507 | return; | |
508 | } | |
2a91aa39 AB |
509 | |
510 | ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq; | |
77d2dd93 GR |
511 | if (after48(ackno, hc->tx_high_ack)) |
512 | hc->tx_high_ack = ackno; | |
32aac18d | 513 | |
77d2dd93 | 514 | seqp = hc->tx_seqt; |
32aac18d AB |
515 | while (before48(seqp->ccid2s_seq, ackno)) { |
516 | seqp = seqp->ccid2s_next; | |
77d2dd93 GR |
517 | if (seqp == hc->tx_seqh) { |
518 | seqp = hc->tx_seqh->ccid2s_prev; | |
32aac18d AB |
519 | break; |
520 | } | |
521 | } | |
2a91aa39 | 522 | |
a3020025 GR |
523 | /* |
524 | * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2 | |
525 | * packets per acknowledgement. Rounding up avoids that cwnd is not | |
526 | * advanced when Ack Ratio is 1 and gives a slight edge otherwise. | |
2a91aa39 | 527 | */ |
77d2dd93 | 528 | if (hc->tx_cwnd < hc->tx_ssthresh) |
a3020025 | 529 | maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2); |
2a91aa39 AB |
530 | |
531 | /* go through all ack vectors */ | |
410e27a4 GR |
532 | while ((offset = ccid2_ackvector(sk, skb, offset, |
533 | &vector, &veclen)) != -1) { | |
2a91aa39 | 534 | /* go through this ack vector */ |
410e27a4 GR |
535 | while (veclen--) { |
536 | const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK; | |
537 | u64 ackno_end_rl = SUB48(ackno, rl); | |
2a91aa39 | 538 | |
410e27a4 | 539 | ccid2_pr_debug("ackvec start:%llu end:%llu\n", |
234af484 | 540 | (unsigned long long)ackno, |
410e27a4 | 541 | (unsigned long long)ackno_end_rl); |
2a91aa39 AB |
542 | /* if the seqno we are analyzing is larger than the |
543 | * current ackno, then move towards the tail of our | |
544 | * seqnos. | |
545 | */ | |
546 | while (after48(seqp->ccid2s_seq, ackno)) { | |
77d2dd93 | 547 | if (seqp == hc->tx_seqt) { |
2a91aa39 AB |
548 | done = 1; |
549 | break; | |
550 | } | |
551 | seqp = seqp->ccid2s_prev; | |
552 | } | |
553 | if (done) | |
554 | break; | |
555 | ||
556 | /* check all seqnos in the range of the vector | |
557 | * run length | |
558 | */ | |
559 | while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) { | |
410e27a4 GR |
560 | const u8 state = *vector & |
561 | DCCP_ACKVEC_STATE_MASK; | |
2a91aa39 AB |
562 | |
563 | /* new packet received or marked */ | |
410e27a4 | 564 | if (state != DCCP_ACKVEC_STATE_NOT_RECEIVED && |
2a91aa39 | 565 | !seqp->ccid2s_acked) { |
410e27a4 GR |
566 | if (state == |
567 | DCCP_ACKVEC_STATE_ECN_MARKED) { | |
d50ad163 | 568 | ccid2_congestion_event(sk, |
374bcf32 | 569 | seqp); |
410e27a4 | 570 | } else |
2a91aa39 AB |
571 | ccid2_new_ack(sk, seqp, |
572 | &maxincr); | |
2a91aa39 AB |
573 | |
574 | seqp->ccid2s_acked = 1; | |
575 | ccid2_pr_debug("Got ack for %llu\n", | |
234af484 | 576 | (unsigned long long)seqp->ccid2s_seq); |
c38c92a8 | 577 | hc->tx_pipe--; |
2a91aa39 | 578 | } |
77d2dd93 | 579 | if (seqp == hc->tx_seqt) { |
2a91aa39 AB |
580 | done = 1; |
581 | break; | |
582 | } | |
3de5489f | 583 | seqp = seqp->ccid2s_prev; |
2a91aa39 AB |
584 | } |
585 | if (done) | |
586 | break; | |
587 | ||
cfbbeabc | 588 | ackno = SUB48(ackno_end_rl, 1); |
410e27a4 | 589 | vector++; |
2a91aa39 AB |
590 | } |
591 | if (done) | |
592 | break; | |
593 | } | |
594 | ||
595 | /* The state about what is acked should be correct now | |
596 | * Check for NUMDUPACK | |
597 | */ | |
77d2dd93 GR |
598 | seqp = hc->tx_seqt; |
599 | while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) { | |
32aac18d | 600 | seqp = seqp->ccid2s_next; |
77d2dd93 GR |
601 | if (seqp == hc->tx_seqh) { |
602 | seqp = hc->tx_seqh->ccid2s_prev; | |
32aac18d AB |
603 | break; |
604 | } | |
605 | } | |
2a91aa39 AB |
606 | done = 0; |
607 | while (1) { | |
608 | if (seqp->ccid2s_acked) { | |
609 | done++; | |
63df18ad | 610 | if (done == NUMDUPACK) |
2a91aa39 | 611 | break; |
2a91aa39 | 612 | } |
77d2dd93 | 613 | if (seqp == hc->tx_seqt) |
2a91aa39 | 614 | break; |
2a91aa39 AB |
615 | seqp = seqp->ccid2s_prev; |
616 | } | |
617 | ||
618 | /* If there are at least 3 acknowledgements, anything unacknowledged | |
619 | * below the last sequence number is considered lost | |
620 | */ | |
63df18ad | 621 | if (done == NUMDUPACK) { |
2a91aa39 AB |
622 | struct ccid2_seq *last_acked = seqp; |
623 | ||
624 | /* check for lost packets */ | |
625 | while (1) { | |
626 | if (!seqp->ccid2s_acked) { | |
374bcf32 | 627 | ccid2_pr_debug("Packet lost: %llu\n", |
234af484 | 628 | (unsigned long long)seqp->ccid2s_seq); |
374bcf32 AB |
629 | /* XXX need to traverse from tail -> head in |
630 | * order to detect multiple congestion events in | |
631 | * one ack vector. | |
632 | */ | |
d50ad163 | 633 | ccid2_congestion_event(sk, seqp); |
c38c92a8 | 634 | hc->tx_pipe--; |
2a91aa39 | 635 | } |
77d2dd93 | 636 | if (seqp == hc->tx_seqt) |
2a91aa39 AB |
637 | break; |
638 | seqp = seqp->ccid2s_prev; | |
639 | } | |
640 | ||
77d2dd93 | 641 | hc->tx_seqt = last_acked; |
2a91aa39 AB |
642 | } |
643 | ||
644 | /* trim acked packets in tail */ | |
77d2dd93 GR |
645 | while (hc->tx_seqt != hc->tx_seqh) { |
646 | if (!hc->tx_seqt->ccid2s_acked) | |
2a91aa39 AB |
647 | break; |
648 | ||
77d2dd93 | 649 | hc->tx_seqt = hc->tx_seqt->ccid2s_next; |
2a91aa39 | 650 | } |
c38c92a8 GR |
651 | |
652 | /* restart RTO timer if not all outstanding data has been acked */ | |
653 | if (hc->tx_pipe == 0) | |
654 | sk_stop_timer(sk, &hc->tx_rtotimer); | |
655 | else | |
656 | sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto); | |
2a91aa39 AB |
657 | } |
658 | ||
91f0ebf7 | 659 | static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk) |
2a91aa39 | 660 | { |
77d2dd93 | 661 | struct ccid2_hc_tx_sock *hc = ccid_priv(ccid); |
b00d2bbc GR |
662 | struct dccp_sock *dp = dccp_sk(sk); |
663 | u32 max_ratio; | |
2a91aa39 | 664 | |
b00d2bbc | 665 | /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */ |
77d2dd93 | 666 | hc->tx_ssthresh = ~0U; |
2a91aa39 | 667 | |
410e27a4 GR |
668 | /* |
669 | * RFC 4341, 5: "The cwnd parameter is initialized to at most four | |
670 | * packets for new connections, following the rules from [RFC3390]". | |
671 | * We need to convert the bytes of RFC3390 into the packets of RFC 4341. | |
672 | */ | |
77d2dd93 | 673 | hc->tx_cwnd = clamp(4380U / dp->dccps_mss_cache, 2U, 4U); |
b00d2bbc GR |
674 | |
675 | /* Make sure that Ack Ratio is enabled and within bounds. */ | |
77d2dd93 | 676 | max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2); |
b00d2bbc GR |
677 | if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio) |
678 | dp->dccps_l_ack_ratio = max_ratio; | |
679 | ||
2a91aa39 | 680 | /* XXX init ~ to window size... */ |
77d2dd93 | 681 | if (ccid2_hc_tx_alloc_seq(hc)) |
2a91aa39 | 682 | return -ENOMEM; |
91f0ebf7 | 683 | |
231cc2aa | 684 | hc->tx_rto = DCCP_TIMEOUT_INIT; |
77d2dd93 GR |
685 | hc->tx_rpdupack = -1; |
686 | hc->tx_last_cong = jiffies; | |
687 | setup_timer(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire, | |
410e27a4 | 688 | (unsigned long)sk); |
2a91aa39 AB |
689 | return 0; |
690 | } | |
691 | ||
692 | static void ccid2_hc_tx_exit(struct sock *sk) | |
693 | { | |
77d2dd93 | 694 | struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); |
07978aab | 695 | int i; |
2a91aa39 | 696 | |
410e27a4 | 697 | ccid2_hc_tx_kill_rto_timer(sk); |
07978aab | 698 | |
77d2dd93 GR |
699 | for (i = 0; i < hc->tx_seqbufc; i++) |
700 | kfree(hc->tx_seqbuf[i]); | |
701 | hc->tx_seqbufc = 0; | |
2a91aa39 AB |
702 | } |
703 | ||
704 | static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb) | |
705 | { | |
706 | const struct dccp_sock *dp = dccp_sk(sk); | |
77d2dd93 | 707 | struct ccid2_hc_rx_sock *hc = ccid2_hc_rx_sk(sk); |
2a91aa39 AB |
708 | |
709 | switch (DCCP_SKB_CB(skb)->dccpd_type) { | |
710 | case DCCP_PKT_DATA: | |
711 | case DCCP_PKT_DATAACK: | |
77d2dd93 GR |
712 | hc->rx_data++; |
713 | if (hc->rx_data >= dp->dccps_r_ack_ratio) { | |
2a91aa39 | 714 | dccp_send_ack(sk); |
77d2dd93 | 715 | hc->rx_data = 0; |
2a91aa39 AB |
716 | } |
717 | break; | |
718 | } | |
719 | } | |
720 | ||
ddebc973 | 721 | struct ccid_operations ccid2_ops = { |
410e27a4 GR |
722 | .ccid_id = DCCPC_CCID2, |
723 | .ccid_name = "TCP-like", | |
410e27a4 GR |
724 | .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock), |
725 | .ccid_hc_tx_init = ccid2_hc_tx_init, | |
726 | .ccid_hc_tx_exit = ccid2_hc_tx_exit, | |
727 | .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet, | |
728 | .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent, | |
729 | .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv, | |
730 | .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock), | |
731 | .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv, | |
2a91aa39 AB |
732 | }; |
733 | ||
84116716 | 734 | #ifdef CONFIG_IP_DCCP_CCID2_DEBUG |
43264991 | 735 | module_param(ccid2_debug, bool, 0644); |
ddebc973 | 736 | MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages"); |
84116716 | 737 | #endif |