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