aboutsummaryrefslogtreecommitdiffstats
path: root/net/dccp/ccids/ccid3.c
diff options
context:
space:
mode:
Diffstat (limited to 'net/dccp/ccids/ccid3.c')
-rw-r--r--net/dccp/ccids/ccid3.c762
1 files changed, 461 insertions, 301 deletions
diff --git a/net/dccp/ccids/ccid3.c b/net/dccp/ccids/ccid3.c
index 06cfdad84a6a..3b8bd7ca6761 100644
--- a/net/dccp/ccids/ccid3.c
+++ b/net/dccp/ccids/ccid3.c
@@ -49,41 +49,75 @@ static int ccid3_debug;
49/* 49/*
50 * Transmitter Half-Connection Routines 50 * Transmitter Half-Connection Routines
51 */ 51 */
52/* Oscillation Prevention/Reduction: recommended by rfc3448bis, on by default */ 52#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
53static int do_osc_prev = true; 53static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
54{
55 static char *ccid3_state_names[] = {
56 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
57 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
58 [TFRC_SSTATE_FBACK] = "FBACK",
59 [TFRC_SSTATE_TERM] = "TERM",
60 };
61
62 return ccid3_state_names[state];
63}
64#endif
65
66static void ccid3_hc_tx_set_state(struct sock *sk,
67 enum ccid3_hc_tx_states state)
68{
69 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
70 enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
71
72 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
73 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
74 ccid3_tx_state_name(state));
75 WARN_ON(state == oldstate);
76 hctx->ccid3hctx_state = state;
77}
54 78
55/* 79/*
56 * Compute the initial sending rate X_init in the manner of RFC 3390: 80 * Compute the initial sending rate X_init in the manner of RFC 3390:
57 * 81 *
58 * X_init = min(4 * MPS, max(2 * MPS, 4380 bytes)) / RTT 82 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
59 * 83 *
84 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
85 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
60 * For consistency with other parts of the code, X_init is scaled by 2^6. 86 * For consistency with other parts of the code, X_init is scaled by 2^6.
61 */ 87 */
62static inline u64 rfc3390_initial_rate(struct sock *sk) 88static inline u64 rfc3390_initial_rate(struct sock *sk)
63{ 89{
64 const u32 mps = dccp_sk(sk)->dccps_mss_cache, 90 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
65 w_init = clamp(4380U, 2 * mps, 4 * mps); 91 const __u32 w_init = clamp_t(__u32, 4380U,
92 2 * hctx->ccid3hctx_s, 4 * hctx->ccid3hctx_s);
66 93
67 return scaled_div(w_init << 6, ccid3_hc_tx_sk(sk)->rtt); 94 return scaled_div(w_init << 6, hctx->ccid3hctx_rtt);
68} 95}
69 96
70/** 97/*
71 * ccid3_update_send_interval - Calculate new t_ipi = s / X 98 * Recalculate t_ipi and delta (should be called whenever X changes)
72 * This respects the granularity of X (64 * bytes/second) and enforces the
73 * scaled minimum of s * 64 / t_mbi = `s' bytes/second as per RFC 3448/4342.
74 */ 99 */
75static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx) 100static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
76{ 101{
77 if (unlikely(hctx->x <= hctx->s)) 102 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
78 hctx->x = hctx->s; 103 hctx->ccid3hctx_t_ipi = scaled_div32(((u64)hctx->ccid3hctx_s) << 6,
79 hctx->t_ipi = scaled_div32(((u64)hctx->s) << 6, hctx->x); 104 hctx->ccid3hctx_x);
105
106 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
107 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
108 TFRC_OPSYS_HALF_TIME_GRAN);
109
110 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n",
111 hctx->ccid3hctx_t_ipi, hctx->ccid3hctx_delta,
112 hctx->ccid3hctx_s, (unsigned)(hctx->ccid3hctx_x >> 6));
113
80} 114}
81 115
82static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now) 116static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
83{ 117{
84 u32 delta = ktime_us_delta(now, hctx->t_last_win_count); 118 u32 delta = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count);
85 119
86 return delta / hctx->rtt; 120 return delta / hctx->ccid3hctx_rtt;
87} 121}
88 122
89/** 123/**
@@ -99,8 +133,8 @@ static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
99static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp) 133static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
100{ 134{
101 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk); 135 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
102 u64 min_rate = 2 * hctx->x_recv; 136 __u64 min_rate = 2 * hctx->ccid3hctx_x_recv;
103 const u64 old_x = hctx->x; 137 const __u64 old_x = hctx->ccid3hctx_x;
104 ktime_t now = stamp ? *stamp : ktime_get_real(); 138 ktime_t now = stamp ? *stamp : ktime_get_real();
105 139
106 /* 140 /*
@@ -111,44 +145,50 @@ static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
111 */ 145 */
112 if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) { 146 if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) {
113 min_rate = rfc3390_initial_rate(sk); 147 min_rate = rfc3390_initial_rate(sk);
114 min_rate = max(min_rate, 2 * hctx->x_recv); 148 min_rate = max(min_rate, 2 * hctx->ccid3hctx_x_recv);
115 } 149 }
116 150
117 if (hctx->p > 0) { 151 if (hctx->ccid3hctx_p > 0) {
118 152
119 hctx->x = min(((u64)hctx->x_calc) << 6, min_rate); 153 hctx->ccid3hctx_x = min(((__u64)hctx->ccid3hctx_x_calc) << 6,
154 min_rate);
155 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
156 (((__u64)hctx->ccid3hctx_s) << 6) /
157 TFRC_T_MBI);
120 158
121 } else if (ktime_us_delta(now, hctx->t_ld) - (s64)hctx->rtt >= 0) { 159 } else if (ktime_us_delta(now, hctx->ccid3hctx_t_ld)
160 - (s64)hctx->ccid3hctx_rtt >= 0) {
122 161
123 hctx->x = min(2 * hctx->x, min_rate); 162 hctx->ccid3hctx_x = min(2 * hctx->ccid3hctx_x, min_rate);
124 hctx->x = max(hctx->x, 163 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
125 scaled_div(((u64)hctx->s) << 6, hctx->rtt)); 164 scaled_div(((__u64)hctx->ccid3hctx_s) << 6,
126 hctx->t_ld = now; 165 hctx->ccid3hctx_rtt));
166 hctx->ccid3hctx_t_ld = now;
127 } 167 }
128 168
129 if (hctx->x != old_x) { 169 if (hctx->ccid3hctx_x != old_x) {
130 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, " 170 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
131 "X_recv=%u\n", (unsigned)(old_x >> 6), 171 "X_recv=%u\n", (unsigned)(old_x >> 6),
132 (unsigned)(hctx->x >> 6), hctx->x_calc, 172 (unsigned)(hctx->ccid3hctx_x >> 6),
133 (unsigned)(hctx->x_recv >> 6)); 173 hctx->ccid3hctx_x_calc,
174 (unsigned)(hctx->ccid3hctx_x_recv >> 6));
134 175
135 ccid3_update_send_interval(hctx); 176 ccid3_update_send_interval(hctx);
136 } 177 }
137} 178}
138 179
139/* 180/*
140 * ccid3_hc_tx_measure_packet_size - Measuring the packet size `s' (sec 4.1) 181 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
141 * @new_len: DCCP payload size in bytes (not used by all methods) 182 * @len: DCCP packet payload size in bytes
142 */ 183 */
143static u32 ccid3_hc_tx_measure_packet_size(struct sock *sk, const u16 new_len) 184static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
144{ 185{
145#if defined(CONFIG_IP_DCCP_CCID3_MEASURE_S_AS_AVG) 186 const u16 old_s = hctx->ccid3hctx_s;
146 return tfrc_ewma(ccid3_hc_tx_sk(sk)->s, new_len, 9); 187
147#elif defined(CONFIG_IP_DCCP_CCID3_MEASURE_S_AS_MAX) 188 hctx->ccid3hctx_s = tfrc_ewma(hctx->ccid3hctx_s, len, 9);
148 return max(ccid3_hc_tx_sk(sk)->s, new_len); 189
149#else /* CONFIG_IP_DCCP_CCID3_MEASURE_S_AS_MPS */ 190 if (hctx->ccid3hctx_s != old_s)
150 return dccp_sk(sk)->dccps_mss_cache; 191 ccid3_update_send_interval(hctx);
151#endif
152} 192}
153 193
154/* 194/*
@@ -158,13 +198,13 @@ static u32 ccid3_hc_tx_measure_packet_size(struct sock *sk, const u16 new_len)
158static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx, 198static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
159 ktime_t now) 199 ktime_t now)
160{ 200{
161 u32 delta = ktime_us_delta(now, hctx->t_last_win_count), 201 u32 delta = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count),
162 quarter_rtts = (4 * delta) / hctx->rtt; 202 quarter_rtts = (4 * delta) / hctx->ccid3hctx_rtt;
163 203
164 if (quarter_rtts > 0) { 204 if (quarter_rtts > 0) {
165 hctx->t_last_win_count = now; 205 hctx->ccid3hctx_t_last_win_count = now;
166 hctx->last_win_count += min(quarter_rtts, 5U); 206 hctx->ccid3hctx_last_win_count += min(quarter_rtts, 5U);
167 hctx->last_win_count &= 0xF; /* mod 16 */ 207 hctx->ccid3hctx_last_win_count &= 0xF; /* mod 16 */
168 } 208 }
169} 209}
170 210
@@ -181,26 +221,25 @@ static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
181 goto restart_timer; 221 goto restart_timer;
182 } 222 }
183 223
184 ccid3_pr_debug("%s(%p) entry with%s feedback\n", dccp_role(sk), sk, 224 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
185 hctx->feedback ? "" : "out"); 225 ccid3_tx_state_name(hctx->ccid3hctx_state));
186 226
187 /* Ignore and do not restart after leaving the established state */ 227 if (hctx->ccid3hctx_state == TFRC_SSTATE_FBACK)
188 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN)) 228 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
229 else if (hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
189 goto out; 230 goto out;
190 231
191 /* Reset feedback state to "no feedback received" */
192 hctx->feedback = false;
193
194 /* 232 /*
195 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4 233 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
196 * RTO is 0 if and only if no feedback has been received yet.
197 */ 234 */
198 if (hctx->t_rto == 0 || hctx->p == 0) { 235 if (hctx->ccid3hctx_t_rto == 0 || /* no feedback received yet */
236 hctx->ccid3hctx_p == 0) {
199 237
200 /* halve send rate directly */ 238 /* halve send rate directly */
201 hctx->x /= 2; 239 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x / 2,
240 (((__u64)hctx->ccid3hctx_s) << 6) /
241 TFRC_T_MBI);
202 ccid3_update_send_interval(hctx); 242 ccid3_update_send_interval(hctx);
203
204 } else { 243 } else {
205 /* 244 /*
206 * Modify the cached value of X_recv 245 * Modify the cached value of X_recv
@@ -212,41 +251,44 @@ static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
212 * 251 *
213 * Note that X_recv is scaled by 2^6 while X_calc is not 252 * Note that X_recv is scaled by 2^6 while X_calc is not
214 */ 253 */
215 BUG_ON(hctx->p && !hctx->x_calc); 254 BUG_ON(hctx->ccid3hctx_p && !hctx->ccid3hctx_x_calc);
216 255
217 if (hctx->x_calc > (hctx->x_recv >> 5)) 256 if (hctx->ccid3hctx_x_calc > (hctx->ccid3hctx_x_recv >> 5))
218 hctx->x_recv /= 2; 257 hctx->ccid3hctx_x_recv =
258 max(hctx->ccid3hctx_x_recv / 2,
259 (((__u64)hctx->ccid3hctx_s) << 6) /
260 (2 * TFRC_T_MBI));
219 else { 261 else {
220 hctx->x_recv = hctx->x_calc; 262 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc;
221 hctx->x_recv <<= 4; 263 hctx->ccid3hctx_x_recv <<= 4;
222 } 264 }
223 ccid3_hc_tx_update_x(sk, NULL); 265 ccid3_hc_tx_update_x(sk, NULL);
224 } 266 }
225 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n", 267 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
226 (unsigned long long)hctx->x); 268 (unsigned long long)hctx->ccid3hctx_x);
227 269
228 /* 270 /*
229 * Set new timeout for the nofeedback timer. 271 * Set new timeout for the nofeedback timer.
230 * See comments in packet_recv() regarding the value of t_RTO. 272 * See comments in packet_recv() regarding the value of t_RTO.
231 */ 273 */
232 if (unlikely(hctx->t_rto == 0)) /* no feedback received yet */ 274 if (unlikely(hctx->ccid3hctx_t_rto == 0)) /* no feedback yet */
233 t_nfb = TFRC_INITIAL_TIMEOUT; 275 t_nfb = TFRC_INITIAL_TIMEOUT;
234 else 276 else
235 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi); 277 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
236 278
237restart_timer: 279restart_timer:
238 sk_reset_timer(sk, &hctx->no_feedback_timer, 280 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
239 jiffies + usecs_to_jiffies(t_nfb)); 281 jiffies + usecs_to_jiffies(t_nfb));
240out: 282out:
241 bh_unlock_sock(sk); 283 bh_unlock_sock(sk);
242 sock_put(sk); 284 sock_put(sk);
243} 285}
244 286
245/** 287/*
246 * ccid3_hc_tx_send_packet - Delay-based dequeueing of TX packets 288 * returns
247 * @skb: next packet candidate to send on @sk 289 * > 0: delay (in msecs) that should pass before actually sending
248 * This function uses the convention of ccid_packet_dequeue_eval() and 290 * = 0: can send immediately
249 * returns a millisecond-delay value between 0 and t_mbi = 64000 msec. 291 * < 0: error condition; do not send packet
250 */ 292 */
251static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb) 293static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
252{ 294{
@@ -263,14 +305,18 @@ static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
263 if (unlikely(skb->len == 0)) 305 if (unlikely(skb->len == 0))
264 return -EBADMSG; 306 return -EBADMSG;
265 307
266 if (hctx->s == 0) { 308 switch (hctx->ccid3hctx_state) {
267 sk_reset_timer(sk, &hctx->no_feedback_timer, (jiffies + 309 case TFRC_SSTATE_NO_SENT:
310 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
311 (jiffies +
268 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT))); 312 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
269 hctx->last_win_count = 0; 313 hctx->ccid3hctx_last_win_count = 0;
270 hctx->t_last_win_count = now; 314 hctx->ccid3hctx_t_last_win_count = now;
271 315
272 /* Set t_0 for initial packet */ 316 /* Set t_0 for initial packet */
273 hctx->t_nom = now; 317 hctx->ccid3hctx_t_nom = now;
318
319 hctx->ccid3hctx_s = skb->len;
274 320
275 /* 321 /*
276 * Use initial RTT sample when available: recommended by erratum 322 * Use initial RTT sample when available: recommended by erratum
@@ -279,9 +325,9 @@ static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
279 */ 325 */
280 if (dp->dccps_syn_rtt) { 326 if (dp->dccps_syn_rtt) {
281 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt); 327 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
282 hctx->rtt = dp->dccps_syn_rtt; 328 hctx->ccid3hctx_rtt = dp->dccps_syn_rtt;
283 hctx->x = rfc3390_initial_rate(sk); 329 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
284 hctx->t_ld = now; 330 hctx->ccid3hctx_t_ld = now;
285 } else { 331 } else {
286 /* 332 /*
287 * Sender does not have RTT sample: 333 * Sender does not have RTT sample:
@@ -289,20 +335,17 @@ static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
289 * is needed in several parts (e.g. window counter); 335 * is needed in several parts (e.g. window counter);
290 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2. 336 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
291 */ 337 */
292 hctx->rtt = DCCP_FALLBACK_RTT; 338 hctx->ccid3hctx_rtt = DCCP_FALLBACK_RTT;
293 hctx->x = dp->dccps_mss_cache; 339 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
294 hctx->x <<= 6; 340 hctx->ccid3hctx_x <<= 6;
295 } 341 }
296
297 /* Compute t_ipi = s / X */
298 hctx->s = ccid3_hc_tx_measure_packet_size(sk, skb->len);
299 ccid3_update_send_interval(hctx); 342 ccid3_update_send_interval(hctx);
300 343
301 /* Seed value for Oscillation Prevention (sec. 4.5) */ 344 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
302 hctx->r_sqmean = tfrc_scaled_sqrt(hctx->rtt); 345 break;
303 346 case TFRC_SSTATE_NO_FBACK:
304 } else { 347 case TFRC_SSTATE_FBACK:
305 delay = ktime_us_delta(hctx->t_nom, now); 348 delay = ktime_us_delta(hctx->ccid3hctx_t_nom, now);
306 ccid3_pr_debug("delay=%ld\n", (long)delay); 349 ccid3_pr_debug("delay=%ld\n", (long)delay);
307 /* 350 /*
308 * Scheduling of packet transmissions [RFC 3448, 4.6] 351 * Scheduling of packet transmissions [RFC 3448, 4.6]
@@ -312,80 +355,99 @@ static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
312 * else 355 * else
313 * // send the packet in (t_nom - t_now) milliseconds. 356 * // send the packet in (t_nom - t_now) milliseconds.
314 */ 357 */
315 if (delay >= TFRC_T_DELTA) 358 if (delay - (s64)hctx->ccid3hctx_delta >= 1000)
316 return (u32)delay / USEC_PER_MSEC; 359 return (u32)delay / 1000L;
317 360
318 ccid3_hc_tx_update_win_count(hctx, now); 361 ccid3_hc_tx_update_win_count(hctx, now);
362 break;
363 case TFRC_SSTATE_TERM:
364 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
365 return -EINVAL;
319 } 366 }
320 367
321 /* prepare to send now (add options etc.) */ 368 /* prepare to send now (add options etc.) */
322 dp->dccps_hc_tx_insert_options = 1; 369 dp->dccps_hc_tx_insert_options = 1;
323 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->last_win_count; 370 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
324 371
325 /* set the nominal send time for the next following packet */ 372 /* set the nominal send time for the next following packet */
326 hctx->t_nom = ktime_add_us(hctx->t_nom, hctx->t_ipi); 373 hctx->ccid3hctx_t_nom = ktime_add_us(hctx->ccid3hctx_t_nom,
327 return CCID_PACKET_SEND_AT_ONCE; 374 hctx->ccid3hctx_t_ipi);
375 return 0;
328} 376}
329 377
330static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len) 378static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
379 unsigned int len)
331{ 380{
332 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk); 381 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
333 382
334 /* Changes to s will become effective the next time X is computed */ 383 ccid3_hc_tx_update_s(hctx, len);
335 hctx->s = ccid3_hc_tx_measure_packet_size(sk, len);
336 384
337 if (tfrc_tx_hist_add(&hctx->hist, dccp_sk(sk)->dccps_gss)) 385 if (tfrc_tx_hist_add(&hctx->ccid3hctx_hist, dccp_sk(sk)->dccps_gss))
338 DCCP_CRIT("packet history - out of memory!"); 386 DCCP_CRIT("packet history - out of memory!");
339} 387}
340 388
341static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb) 389static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
342{ 390{
343 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk); 391 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
344 struct tfrc_tx_hist_entry *acked; 392 struct ccid3_options_received *opt_recv;
345 ktime_t now; 393 ktime_t now;
346 unsigned long t_nfb; 394 unsigned long t_nfb;
347 u32 r_sample; 395 u32 pinv, r_sample;
348 396
349 /* we are only interested in ACKs */ 397 /* we are only interested in ACKs */
350 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK || 398 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
351 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK)) 399 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
352 return; 400 return;
353 /* 401 /* ... and only in the established state */
354 * Locate the acknowledged packet in the TX history. 402 if (hctx->ccid3hctx_state != TFRC_SSTATE_FBACK &&
355 * 403 hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
356 * Returning "entry not found" here can for instance happen when 404 return;
357 * - the host has not sent out anything (e.g. a passive server), 405
358 * - the Ack is outdated (packet with higher Ack number was received), 406 opt_recv = &hctx->ccid3hctx_options_received;
359 * - it is a bogus Ack (for a packet not sent on this connection). 407 now = ktime_get_real();
360 */ 408
361 acked = tfrc_tx_hist_find_entry(hctx->hist, dccp_hdr_ack_seq(skb)); 409 /* Estimate RTT from history if ACK number is valid */
362 if (acked == NULL) 410 r_sample = tfrc_tx_hist_rtt(hctx->ccid3hctx_hist,
411 DCCP_SKB_CB(skb)->dccpd_ack_seq, now);
412 if (r_sample == 0) {
413 DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk), sk,
414 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type),
415 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq);
363 return; 416 return;
364 /* For the sake of RTT sampling, ignore/remove all older entries */ 417 }
365 tfrc_tx_hist_purge(&acked->next);
366 418
367 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */ 419 /* Update receive rate in units of 64 * bytes/second */
368 now = ktime_get_real(); 420 hctx->ccid3hctx_x_recv = opt_recv->ccid3or_receive_rate;
369 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp)); 421 hctx->ccid3hctx_x_recv <<= 6;
370 hctx->rtt = tfrc_ewma(hctx->rtt, r_sample, 9);
371 422
423 /* Update loss event rate (which is scaled by 1e6) */
424 pinv = opt_recv->ccid3or_loss_event_rate;
425 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
426 hctx->ccid3hctx_p = 0;
427 else /* can not exceed 100% */
428 hctx->ccid3hctx_p = scaled_div(1, pinv);
429 /*
430 * Validate new RTT sample and update moving average
431 */
432 r_sample = dccp_sample_rtt(sk, r_sample);
433 hctx->ccid3hctx_rtt = tfrc_ewma(hctx->ccid3hctx_rtt, r_sample, 9);
372 /* 434 /*
373 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3 435 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
374 */ 436 */
375 if (!hctx->feedback) { 437 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
376 hctx->feedback = true; 438 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
377 439
378 if (hctx->t_rto == 0) { 440 if (hctx->ccid3hctx_t_rto == 0) {
379 /* 441 /*
380 * Initial feedback packet: Larger Initial Windows (4.2) 442 * Initial feedback packet: Larger Initial Windows (4.2)
381 */ 443 */
382 hctx->x = rfc3390_initial_rate(sk); 444 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
383 hctx->t_ld = now; 445 hctx->ccid3hctx_t_ld = now;
384 446
385 ccid3_update_send_interval(hctx); 447 ccid3_update_send_interval(hctx);
386 448
387 goto done_computing_x; 449 goto done_computing_x;
388 } else if (hctx->p == 0) { 450 } else if (hctx->ccid3hctx_p == 0) {
389 /* 451 /*
390 * First feedback after nofeedback timer expiry (4.3) 452 * First feedback after nofeedback timer expiry (4.3)
391 */ 453 */
@@ -394,52 +456,25 @@ static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
394 } 456 }
395 457
396 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */ 458 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
397 if (hctx->p > 0) 459 if (hctx->ccid3hctx_p > 0)
398 hctx->x_calc = tfrc_calc_x(hctx->s, hctx->rtt, hctx->p); 460 hctx->ccid3hctx_x_calc =
461 tfrc_calc_x(hctx->ccid3hctx_s,
462 hctx->ccid3hctx_rtt,
463 hctx->ccid3hctx_p);
399 ccid3_hc_tx_update_x(sk, &now); 464 ccid3_hc_tx_update_x(sk, &now);
400 465
401done_computing_x: 466done_computing_x:
402 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, " 467 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
403 "p=%u, X_calc=%u, X_recv=%u, X=%u\n", 468 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
404 dccp_role(sk), sk, hctx->rtt, r_sample, 469 dccp_role(sk),
405 hctx->s, hctx->p, hctx->x_calc, 470 sk, hctx->ccid3hctx_rtt, r_sample,
406 (unsigned)(hctx->x_recv >> 6), 471 hctx->ccid3hctx_s, hctx->ccid3hctx_p,
407 (unsigned)(hctx->x >> 6)); 472 hctx->ccid3hctx_x_calc,
408 /* 473 (unsigned)(hctx->ccid3hctx_x_recv >> 6),
409 * Oscillation Reduction (RFC 3448, 4.5) - modifying t_ipi according to 474 (unsigned)(hctx->ccid3hctx_x >> 6));
410 * RTT changes, multiplying by X/X_inst = sqrt(R_sample)/R_sqmean. This
411 * can be useful if few connections share a link, avoiding that buffer
412 * fill levels (RTT) oscillate as a result of frequent adjustments to X.
413 * A useful presentation with background information is in
414 * Joerg Widmer, "Equation-Based Congestion Control",
415 * MSc Thesis, University of Mannheim, Germany, 2000
416 * (sec. 3.6.4), who calls this ISM ("Inter-packet Space Modulation").
417 */
418 if (do_osc_prev) {
419 r_sample = tfrc_scaled_sqrt(r_sample);
420 /*
421 * The modulation can work in both ways: increase/decrease t_ipi
422 * according to long-term increases/decreases of the RTT. The
423 * former is a useful measure, since it works against queue
424 * build-up. The latter temporarily increases the sending rate,
425 * so that buffers fill up more quickly. This in turn causes
426 * the RTT to increase, so that either later reduction becomes
427 * necessary or the RTT stays at a very high level. Decreasing
428 * t_ipi is therefore not supported.
429 * Furthermore, during the initial slow-start phase the RTT
430 * naturally increases, where using the algorithm would cause
431 * delays. Hence it is disabled during the initial slow-start.
432 */
433 if (r_sample > hctx->r_sqmean && hctx->p > 0)
434 hctx->t_ipi = div_u64((u64)hctx->t_ipi * (u64)r_sample,
435 hctx->r_sqmean);
436 hctx->t_ipi = min_t(u32, hctx->t_ipi, TFRC_T_MBI);
437 /* update R_sqmean _after_ computing the modulation factor */
438 hctx->r_sqmean = tfrc_ewma(hctx->r_sqmean, r_sample, 9);
439 }
440 475
441 /* unschedule no feedback timer */ 476 /* unschedule no feedback timer */
442 sk_stop_timer(sk, &hctx->no_feedback_timer); 477 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
443 478
444 /* 479 /*
445 * As we have calculated new ipi, delta, t_nom it is possible 480 * As we have calculated new ipi, delta, t_nom it is possible
@@ -453,66 +488,95 @@ done_computing_x:
453 * This can help avoid triggering the nofeedback timer too 488 * This can help avoid triggering the nofeedback timer too
454 * often ('spinning') on LANs with small RTTs. 489 * often ('spinning') on LANs with small RTTs.
455 */ 490 */
456 hctx->t_rto = max_t(u32, 4 * hctx->rtt, (CONFIG_IP_DCCP_CCID3_RTO * 491 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
457 (USEC_PER_SEC / 1000))); 492 (CONFIG_IP_DCCP_CCID3_RTO *
493 (USEC_PER_SEC / 1000)));
458 /* 494 /*
459 * Schedule no feedback timer to expire in 495 * Schedule no feedback timer to expire in
460 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi) 496 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
461 */ 497 */
462 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi); 498 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
463 499
464 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to " 500 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
465 "expire in %lu jiffies (%luus)\n", 501 "expire in %lu jiffies (%luus)\n",
466 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb); 502 dccp_role(sk),
503 sk, usecs_to_jiffies(t_nfb), t_nfb);
467 504
468 sk_reset_timer(sk, &hctx->no_feedback_timer, 505 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
469 jiffies + usecs_to_jiffies(t_nfb)); 506 jiffies + usecs_to_jiffies(t_nfb));
470} 507}
471 508
472static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type, 509static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
473 u8 option, u8 *optval, u8 optlen) 510 unsigned char len, u16 idx,
511 unsigned char *value)
474{ 512{
513 int rc = 0;
514 const struct dccp_sock *dp = dccp_sk(sk);
475 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk); 515 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
516 struct ccid3_options_received *opt_recv;
476 __be32 opt_val; 517 __be32 opt_val;
477 518
478 switch (option) { 519 opt_recv = &hctx->ccid3hctx_options_received;
479 case TFRC_OPT_RECEIVE_RATE:
480 case TFRC_OPT_LOSS_EVENT_RATE:
481 /* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
482 if (packet_type == DCCP_PKT_DATA)
483 break;
484 if (unlikely(optlen != 4)) {
485 DCCP_WARN("%s(%p), invalid len %d for %u\n",
486 dccp_role(sk), sk, optlen, option);
487 return -EINVAL;
488 }
489 opt_val = ntohl(get_unaligned((__be32 *)optval));
490 520
491 if (option == TFRC_OPT_RECEIVE_RATE) { 521 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
492 /* Receive Rate is kept in units of 64 bytes/second */ 522 opt_recv->ccid3or_seqno = dp->dccps_gsr;
493 hctx->x_recv = opt_val; 523 opt_recv->ccid3or_loss_event_rate = ~0;
494 hctx->x_recv <<= 6; 524 opt_recv->ccid3or_loss_intervals_idx = 0;
525 opt_recv->ccid3or_loss_intervals_len = 0;
526 opt_recv->ccid3or_receive_rate = 0;
527 }
495 528
496 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n", 529 switch (option) {
497 dccp_role(sk), sk, opt_val); 530 case TFRC_OPT_LOSS_EVENT_RATE:
531 if (unlikely(len != 4)) {
532 DCCP_WARN("%s(%p), invalid len %d "
533 "for TFRC_OPT_LOSS_EVENT_RATE\n",
534 dccp_role(sk), sk, len);
535 rc = -EINVAL;
498 } else { 536 } else {
499 /* Update the fixpoint Loss Event Rate fraction */ 537 opt_val = get_unaligned((__be32 *)value);
500 hctx->p = tfrc_invert_loss_event_rate(opt_val); 538 opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
501
502 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n", 539 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
503 dccp_role(sk), sk, opt_val); 540 dccp_role(sk), sk,
541 opt_recv->ccid3or_loss_event_rate);
504 } 542 }
543 break;
544 case TFRC_OPT_LOSS_INTERVALS:
545 opt_recv->ccid3or_loss_intervals_idx = idx;
546 opt_recv->ccid3or_loss_intervals_len = len;
547 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
548 dccp_role(sk), sk,
549 opt_recv->ccid3or_loss_intervals_idx,
550 opt_recv->ccid3or_loss_intervals_len);
551 break;
552 case TFRC_OPT_RECEIVE_RATE:
553 if (unlikely(len != 4)) {
554 DCCP_WARN("%s(%p), invalid len %d "
555 "for TFRC_OPT_RECEIVE_RATE\n",
556 dccp_role(sk), sk, len);
557 rc = -EINVAL;
558 } else {
559 opt_val = get_unaligned((__be32 *)value);
560 opt_recv->ccid3or_receive_rate = ntohl(opt_val);
561 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
562 dccp_role(sk), sk,
563 opt_recv->ccid3or_receive_rate);
564 }
565 break;
505 } 566 }
506 return 0; 567
568 return rc;
507} 569}
508 570
509static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk) 571static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
510{ 572{
511 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid); 573 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
512 574
513 hctx->hist = NULL; 575 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
514 setup_timer(&hctx->no_feedback_timer, 576 hctx->ccid3hctx_hist = NULL;
515 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk); 577 setup_timer(&hctx->ccid3hctx_no_feedback_timer,
578 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
579
516 return 0; 580 return 0;
517} 581}
518 582
@@ -520,36 +584,42 @@ static void ccid3_hc_tx_exit(struct sock *sk)
520{ 584{
521 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk); 585 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
522 586
523 sk_stop_timer(sk, &hctx->no_feedback_timer); 587 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
524 tfrc_tx_hist_purge(&hctx->hist); 588 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
589
590 tfrc_tx_hist_purge(&hctx->ccid3hctx_hist);
525} 591}
526 592
527static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info) 593static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
528{ 594{
529 info->tcpi_rto = ccid3_hc_tx_sk(sk)->t_rto; 595 struct ccid3_hc_tx_sock *hctx;
530 info->tcpi_rtt = ccid3_hc_tx_sk(sk)->rtt; 596
597 /* Listen socks doesn't have a private CCID block */
598 if (sk->sk_state == DCCP_LISTEN)
599 return;
600
601 hctx = ccid3_hc_tx_sk(sk);
602 info->tcpi_rto = hctx->ccid3hctx_t_rto;
603 info->tcpi_rtt = hctx->ccid3hctx_rtt;
531} 604}
532 605
533static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len, 606static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
534 u32 __user *optval, int __user *optlen) 607 u32 __user *optval, int __user *optlen)
535{ 608{
536 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk); 609 const struct ccid3_hc_tx_sock *hctx;
537 struct tfrc_tx_info tfrc;
538 const void *val; 610 const void *val;
539 611
612 /* Listen socks doesn't have a private CCID block */
613 if (sk->sk_state == DCCP_LISTEN)
614 return -EINVAL;
615
616 hctx = ccid3_hc_tx_sk(sk);
540 switch (optname) { 617 switch (optname) {
541 case DCCP_SOCKOPT_CCID_TX_INFO: 618 case DCCP_SOCKOPT_CCID_TX_INFO:
542 if (len < sizeof(tfrc)) 619 if (len < sizeof(hctx->ccid3hctx_tfrc))
543 return -EINVAL; 620 return -EINVAL;
544 tfrc.tfrctx_x = hctx->x; 621 len = sizeof(hctx->ccid3hctx_tfrc);
545 tfrc.tfrctx_x_recv = hctx->x_recv; 622 val = &hctx->ccid3hctx_tfrc;
546 tfrc.tfrctx_x_calc = hctx->x_calc;
547 tfrc.tfrctx_rtt = hctx->rtt;
548 tfrc.tfrctx_p = hctx->p;
549 tfrc.tfrctx_rto = hctx->t_rto;
550 tfrc.tfrctx_ipi = hctx->t_ipi;
551 len = sizeof(tfrc);
552 val = &tfrc;
553 break; 623 break;
554 default: 624 default:
555 return -ENOPROTOOPT; 625 return -ENOPROTOOPT;
@@ -564,82 +634,112 @@ static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
564/* 634/*
565 * Receiver Half-Connection Routines 635 * Receiver Half-Connection Routines
566 */ 636 */
637
638/* CCID3 feedback types */
639enum ccid3_fback_type {
640 CCID3_FBACK_NONE = 0,
641 CCID3_FBACK_INITIAL,
642 CCID3_FBACK_PERIODIC,
643 CCID3_FBACK_PARAM_CHANGE
644};
645
646#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
647static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
648{
649 static char *ccid3_rx_state_names[] = {
650 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
651 [TFRC_RSTATE_DATA] = "DATA",
652 [TFRC_RSTATE_TERM] = "TERM",
653 };
654
655 return ccid3_rx_state_names[state];
656}
657#endif
658
659static void ccid3_hc_rx_set_state(struct sock *sk,
660 enum ccid3_hc_rx_states state)
661{
662 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
663 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
664
665 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
666 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
667 ccid3_rx_state_name(state));
668 WARN_ON(state == oldstate);
669 hcrx->ccid3hcrx_state = state;
670}
671
567static void ccid3_hc_rx_send_feedback(struct sock *sk, 672static void ccid3_hc_rx_send_feedback(struct sock *sk,
568 const struct sk_buff *skb, 673 const struct sk_buff *skb,
569 enum ccid3_fback_type fbtype) 674 enum ccid3_fback_type fbtype)
570{ 675{
571 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk); 676 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
677 struct dccp_sock *dp = dccp_sk(sk);
678 ktime_t now;
679 s64 delta = 0;
680
681 if (unlikely(hcrx->ccid3hcrx_state == TFRC_RSTATE_TERM))
682 return;
683
684 now = ktime_get_real();
572 685
573 switch (fbtype) { 686 switch (fbtype) {
574 case CCID3_FBACK_INITIAL: 687 case CCID3_FBACK_INITIAL:
575 hcrx->x_recv = 0; 688 hcrx->ccid3hcrx_x_recv = 0;
576 hcrx->p_inverse = ~0U; /* see RFC 4342, 8.5 */ 689 hcrx->ccid3hcrx_pinv = ~0U; /* see RFC 4342, 8.5 */
577 break; 690 break;
578 case CCID3_FBACK_PARAM_CHANGE: 691 case CCID3_FBACK_PARAM_CHANGE:
579 if (unlikely(hcrx->feedback == CCID3_FBACK_NONE)) {
580 /*
581 * rfc3448bis-06, 6.3.1: First packet(s) lost or marked
582 * FIXME: in rfc3448bis the receiver returns X_recv=0
583 * here as it normally would in the first feedback packet.
584 * However this is not possible yet, since the code still
585 * uses RFC 3448, i.e.
586 * If (p > 0)
587 * Calculate X_calc using the TCP throughput equation.
588 * X = max(min(X_calc, 2*X_recv), s/t_mbi);
589 * would bring X down to s/t_mbi. That is why we return
590 * X_recv according to rfc3448bis-06 for the moment.
591 */
592 u32 s = tfrc_rx_hist_packet_size(&hcrx->hist),
593 rtt = tfrc_rx_hist_rtt(&hcrx->hist);
594
595 hcrx->x_recv = scaled_div32(s, 2 * rtt);
596 break;
597 }
598 /* 692 /*
599 * When parameters change (new loss or p > p_prev), we do not 693 * When parameters change (new loss or p > p_prev), we do not
600 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so 694 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
601 * always check whether at least RTT time units were covered. 695 * need to reuse the previous value of X_recv. However, when
696 * X_recv was 0 (due to early loss), this would kill X down to
697 * s/t_mbi (i.e. one packet in 64 seconds).
698 * To avoid such drastic reduction, we approximate X_recv as
699 * the number of bytes since last feedback.
700 * This is a safe fallback, since X is bounded above by X_calc.
602 */ 701 */
603 hcrx->x_recv = tfrc_rx_hist_x_recv(&hcrx->hist, hcrx->x_recv); 702 if (hcrx->ccid3hcrx_x_recv > 0)
604 break; 703 break;
704 /* fall through */
605 case CCID3_FBACK_PERIODIC: 705 case CCID3_FBACK_PERIODIC:
606 /* 706 delta = ktime_us_delta(now, hcrx->ccid3hcrx_tstamp_last_feedback);
607 * Step (2) of rfc3448bis-06, 6.2: 707 if (delta <= 0)
608 * - if no data packets have been received, just restart timer 708 DCCP_BUG("delta (%ld) <= 0", (long)delta);
609 * - if data packets have been received, re-compute X_recv 709 else
610 */ 710 hcrx->ccid3hcrx_x_recv =
611 if (hcrx->hist.bytes_recvd == 0) 711 scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
612 goto prepare_for_next_time;
613 hcrx->x_recv = tfrc_rx_hist_x_recv(&hcrx->hist, hcrx->x_recv);
614 break; 712 break;
615 default: 713 default:
616 return; 714 return;
617 } 715 }
618 716
619 ccid3_pr_debug("X_recv=%u, 1/p=%u\n", hcrx->x_recv, hcrx->p_inverse); 717 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta,
718 hcrx->ccid3hcrx_x_recv, hcrx->ccid3hcrx_pinv);
620 719
621 dccp_sk(sk)->dccps_hc_rx_insert_options = 1; 720 hcrx->ccid3hcrx_tstamp_last_feedback = now;
622 dccp_send_ack(sk); 721 hcrx->ccid3hcrx_last_counter = dccp_hdr(skb)->dccph_ccval;
722 hcrx->ccid3hcrx_bytes_recv = 0;
623 723
624prepare_for_next_time: 724 dp->dccps_hc_rx_insert_options = 1;
625 tfrc_rx_hist_restart_byte_counter(&hcrx->hist); 725 dccp_send_ack(sk);
626 hcrx->last_counter = dccp_hdr(skb)->dccph_ccval;
627 hcrx->feedback = fbtype;
628} 726}
629 727
630static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb) 728static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
631{ 729{
632 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk); 730 const struct ccid3_hc_rx_sock *hcrx;
633 __be32 x_recv, pinv; 731 __be32 x_recv, pinv;
634 732
635 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)) 733 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
636 return 0; 734 return 0;
637 735
736 hcrx = ccid3_hc_rx_sk(sk);
737
638 if (dccp_packet_without_ack(skb)) 738 if (dccp_packet_without_ack(skb))
639 return 0; 739 return 0;
640 740
641 x_recv = htonl(hcrx->x_recv); 741 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
642 pinv = htonl(hcrx->p_inverse); 742 pinv = htonl(hcrx->ccid3hcrx_pinv);
643 743
644 if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE, 744 if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
645 &pinv, sizeof(pinv)) || 745 &pinv, sizeof(pinv)) ||
@@ -662,95 +762,171 @@ static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
662static u32 ccid3_first_li(struct sock *sk) 762static u32 ccid3_first_li(struct sock *sk)
663{ 763{
664 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk); 764 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
665 u32 s = tfrc_rx_hist_packet_size(&hcrx->hist), 765 u32 x_recv, p, delta;
666 rtt = tfrc_rx_hist_rtt(&hcrx->hist), x_recv, p;
667 u64 fval; 766 u64 fval;
668 767
669 /* 768 if (hcrx->ccid3hcrx_rtt == 0) {
670 * rfc3448bis-06, 6.3.1: First data packet(s) are marked or lost. Set p 769 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
671 * to give the equivalent of X_target = s/(2*R). Thus fval = 2 and so p 770 hcrx->ccid3hcrx_rtt = DCCP_FALLBACK_RTT;
672 * is about 20.64%. This yields an interval length of 4.84 (rounded up). 771 }
673 */
674 if (unlikely(hcrx->feedback == CCID3_FBACK_NONE))
675 return 5;
676 772
677 x_recv = tfrc_rx_hist_x_recv(&hcrx->hist, hcrx->x_recv); 773 delta = ktime_to_us(net_timedelta(hcrx->ccid3hcrx_tstamp_last_feedback));
678 if (x_recv == 0) 774 x_recv = scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
679 goto failed; 775 if (x_recv == 0) { /* would also trigger divide-by-zero */
776 DCCP_WARN("X_recv==0\n");
777 if ((x_recv = hcrx->ccid3hcrx_x_recv) == 0) {
778 DCCP_BUG("stored value of X_recv is zero");
779 return ~0U;
780 }
781 }
680 782
681 fval = scaled_div32(scaled_div(s, rtt), x_recv); 783 fval = scaled_div(hcrx->ccid3hcrx_s, hcrx->ccid3hcrx_rtt);
784 fval = scaled_div32(fval, x_recv);
682 p = tfrc_calc_x_reverse_lookup(fval); 785 p = tfrc_calc_x_reverse_lookup(fval);
683 786
684 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied " 787 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
685 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p); 788 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
686 789
687 if (p > 0) 790 return p == 0 ? ~0U : scaled_div(1, p);
688 return scaled_div(1, p);
689failed:
690 return UINT_MAX;
691} 791}
692 792
693static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb) 793static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
694{ 794{
695 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk); 795 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
796 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
696 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp; 797 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
697 const bool is_data_packet = dccp_data_packet(skb); 798 const bool is_data_packet = dccp_data_packet(skb);
698 799
800 if (unlikely(hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)) {
801 if (is_data_packet) {
802 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
803 do_feedback = CCID3_FBACK_INITIAL;
804 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
805 hcrx->ccid3hcrx_s = payload;
806 /*
807 * Not necessary to update ccid3hcrx_bytes_recv here,
808 * since X_recv = 0 for the first feedback packet (cf.
809 * RFC 3448, 6.3) -- gerrit
810 */
811 }
812 goto update_records;
813 }
814
815 if (tfrc_rx_hist_duplicate(&hcrx->ccid3hcrx_hist, skb))
816 return; /* done receiving */
817
818 if (is_data_packet) {
819 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
820 /*
821 * Update moving-average of s and the sum of received payload bytes
822 */
823 hcrx->ccid3hcrx_s = tfrc_ewma(hcrx->ccid3hcrx_s, payload, 9);
824 hcrx->ccid3hcrx_bytes_recv += payload;
825 }
826
699 /* 827 /*
700 * Perform loss detection and handle pending losses 828 * Perform loss detection and handle pending losses
701 */ 829 */
702 if (tfrc_rx_congestion_event(&hcrx->hist, &hcrx->li_hist, 830 if (tfrc_rx_handle_loss(&hcrx->ccid3hcrx_hist, &hcrx->ccid3hcrx_li_hist,
703 skb, ndp, ccid3_first_li, sk)) 831 skb, ndp, ccid3_first_li, sk)) {
704 ccid3_hc_rx_send_feedback(sk, skb, CCID3_FBACK_PARAM_CHANGE); 832 do_feedback = CCID3_FBACK_PARAM_CHANGE;
833 goto done_receiving;
834 }
835
836 if (tfrc_rx_hist_loss_pending(&hcrx->ccid3hcrx_hist))
837 return; /* done receiving */
838
705 /* 839 /*
706 * Feedback for first non-empty data packet (RFC 3448, 6.3) 840 * Handle data packets: RTT sampling and monitoring p
707 */ 841 */
708 else if (unlikely(hcrx->feedback == CCID3_FBACK_NONE && is_data_packet)) 842 if (unlikely(!is_data_packet))
709 ccid3_hc_rx_send_feedback(sk, skb, CCID3_FBACK_INITIAL); 843 goto update_records;
844
845 if (!tfrc_lh_is_initialised(&hcrx->ccid3hcrx_li_hist)) {
846 const u32 sample = tfrc_rx_hist_sample_rtt(&hcrx->ccid3hcrx_hist, skb);
847 /*
848 * Empty loss history: no loss so far, hence p stays 0.
849 * Sample RTT values, since an RTT estimate is required for the
850 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
851 */
852 if (sample != 0)
853 hcrx->ccid3hcrx_rtt = tfrc_ewma(hcrx->ccid3hcrx_rtt, sample, 9);
854
855 } else if (tfrc_lh_update_i_mean(&hcrx->ccid3hcrx_li_hist, skb)) {
856 /*
857 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
858 * has decreased (resp. p has increased), send feedback now.
859 */
860 do_feedback = CCID3_FBACK_PARAM_CHANGE;
861 }
862
710 /* 863 /*
711 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3 864 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
712 */ 865 */
713 else if (!tfrc_rx_hist_loss_pending(&hcrx->hist) && is_data_packet && 866 if (SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->ccid3hcrx_last_counter) > 3)
714 SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->last_counter) > 3) 867 do_feedback = CCID3_FBACK_PERIODIC;
715 ccid3_hc_rx_send_feedback(sk, skb, CCID3_FBACK_PERIODIC); 868
869update_records:
870 tfrc_rx_hist_add_packet(&hcrx->ccid3hcrx_hist, skb, ndp);
871
872done_receiving:
873 if (do_feedback)
874 ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
716} 875}
717 876
718static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk) 877static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
719{ 878{
720 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid); 879 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
721 880
722 tfrc_lh_init(&hcrx->li_hist); 881 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
723 return tfrc_rx_hist_init(&hcrx->hist, sk); 882 tfrc_lh_init(&hcrx->ccid3hcrx_li_hist);
883 return tfrc_rx_hist_alloc(&hcrx->ccid3hcrx_hist);
724} 884}
725 885
726static void ccid3_hc_rx_exit(struct sock *sk) 886static void ccid3_hc_rx_exit(struct sock *sk)
727{ 887{
728 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk); 888 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
729 889
730 tfrc_rx_hist_purge(&hcrx->hist); 890 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
731 tfrc_lh_cleanup(&hcrx->li_hist); 891
892 tfrc_rx_hist_purge(&hcrx->ccid3hcrx_hist);
893 tfrc_lh_cleanup(&hcrx->ccid3hcrx_li_hist);
732} 894}
733 895
734static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info) 896static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
735{ 897{
898 const struct ccid3_hc_rx_sock *hcrx;
899
900 /* Listen socks doesn't have a private CCID block */
901 if (sk->sk_state == DCCP_LISTEN)
902 return;
903
904 hcrx = ccid3_hc_rx_sk(sk);
905 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
736 info->tcpi_options |= TCPI_OPT_TIMESTAMPS; 906 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
737 info->tcpi_rcv_rtt = tfrc_rx_hist_rtt(&ccid3_hc_rx_sk(sk)->hist); 907 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
738} 908}
739 909
740static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len, 910static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
741 u32 __user *optval, int __user *optlen) 911 u32 __user *optval, int __user *optlen)
742{ 912{
743 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk); 913 const struct ccid3_hc_rx_sock *hcrx;
744 struct tfrc_rx_info rx_info; 914 struct tfrc_rx_info rx_info;
745 const void *val; 915 const void *val;
746 916
917 /* Listen socks doesn't have a private CCID block */
918 if (sk->sk_state == DCCP_LISTEN)
919 return -EINVAL;
920
921 hcrx = ccid3_hc_rx_sk(sk);
747 switch (optname) { 922 switch (optname) {
748 case DCCP_SOCKOPT_CCID_RX_INFO: 923 case DCCP_SOCKOPT_CCID_RX_INFO:
749 if (len < sizeof(rx_info)) 924 if (len < sizeof(rx_info))
750 return -EINVAL; 925 return -EINVAL;
751 rx_info.tfrcrx_x_recv = hcrx->x_recv; 926 rx_info.tfrcrx_x_recv = hcrx->ccid3hcrx_x_recv;
752 rx_info.tfrcrx_rtt = tfrc_rx_hist_rtt(&hcrx->hist); 927 rx_info.tfrcrx_rtt = hcrx->ccid3hcrx_rtt;
753 rx_info.tfrcrx_p = tfrc_invert_loss_event_rate(hcrx->p_inverse); 928 rx_info.tfrcrx_p = hcrx->ccid3hcrx_pinv == 0 ? ~0U :
929 scaled_div(1, hcrx->ccid3hcrx_pinv);
754 len = sizeof(rx_info); 930 len = sizeof(rx_info);
755 val = &rx_info; 931 val = &rx_info;
756 break; 932 break;
@@ -786,9 +962,6 @@ static struct ccid_operations ccid3 = {
786 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt, 962 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
787}; 963};
788 964
789module_param(do_osc_prev, bool, 0644);
790MODULE_PARM_DESC(do_osc_prev, "Use Oscillation Prevention (RFC 3448, 4.5)");
791
792#ifdef CONFIG_IP_DCCP_CCID3_DEBUG 965#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
793module_param(ccid3_debug, bool, 0644); 966module_param(ccid3_debug, bool, 0644);
794MODULE_PARM_DESC(ccid3_debug, "Enable debug messages"); 967MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
@@ -796,19 +969,6 @@ MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
796 969
797static __init int ccid3_module_init(void) 970static __init int ccid3_module_init(void)
798{ 971{
799 struct timespec tp;
800
801 /*
802 * Without a fine-grained clock resolution, RTTs/X_recv are not sampled
803 * correctly and feedback is sent either too early or too late.
804 */
805 hrtimer_get_res(CLOCK_MONOTONIC, &tp);
806 if (tp.tv_sec || tp.tv_nsec > DCCP_TIME_RESOLUTION * NSEC_PER_USEC) {
807 printk(KERN_ERR "%s: Timer too coarse (%ld usec), need %u-usec"
808 " resolution - check your clocksource.\n", __func__,
809 tp.tv_nsec/NSEC_PER_USEC, DCCP_TIME_RESOLUTION);
810 return -ESOCKTNOSUPPORT;
811 }
812 return ccid_register(&ccid3); 972 return ccid_register(&ccid3);
813} 973}
814module_init(ccid3_module_init); 974module_init(ccid3_module_init);