Commit | Line | Data |
---|---|---|
51c5d0c4 DM |
1 | #include <linux/rcupdate.h> |
2 | #include <linux/spinlock.h> | |
3 | #include <linux/jiffies.h> | |
ab92bb2f | 4 | #include <linux/module.h> |
4aabd8ef | 5 | #include <linux/cache.h> |
51c5d0c4 DM |
6 | #include <linux/slab.h> |
7 | #include <linux/init.h> | |
4aabd8ef | 8 | #include <linux/tcp.h> |
5815d5e7 | 9 | #include <linux/hash.h> |
d23ff701 | 10 | #include <linux/tcp_metrics.h> |
976a702a | 11 | #include <linux/vmalloc.h> |
4aabd8ef DM |
12 | |
13 | #include <net/inet_connection_sock.h> | |
51c5d0c4 | 14 | #include <net/net_namespace.h> |
ab92bb2f | 15 | #include <net/request_sock.h> |
51c5d0c4 | 16 | #include <net/inetpeer.h> |
4aabd8ef | 17 | #include <net/sock.h> |
51c5d0c4 | 18 | #include <net/ipv6.h> |
4aabd8ef DM |
19 | #include <net/dst.h> |
20 | #include <net/tcp.h> | |
d23ff701 | 21 | #include <net/genetlink.h> |
4aabd8ef DM |
22 | |
23 | int sysctl_tcp_nometrics_save __read_mostly; | |
24 | ||
41804420 DM |
25 | static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr, |
26 | const struct inetpeer_addr *daddr, | |
77f99ad1 CP |
27 | struct net *net, unsigned int hash); |
28 | ||
1fe4c481 YC |
29 | struct tcp_fastopen_metrics { |
30 | u16 mss; | |
aab48743 YC |
31 | u16 syn_loss:10; /* Recurring Fast Open SYN losses */ |
32 | unsigned long last_syn_loss; /* Last Fast Open SYN loss */ | |
1fe4c481 YC |
33 | struct tcp_fastopen_cookie cookie; |
34 | }; | |
35 | ||
740b0f18 ED |
36 | /* TCP_METRIC_MAX includes 2 extra fields for userspace compatibility |
37 | * Kernel only stores RTT and RTTVAR in usec resolution | |
38 | */ | |
39 | #define TCP_METRIC_MAX_KERNEL (TCP_METRIC_MAX - 2) | |
40 | ||
51c5d0c4 DM |
41 | struct tcp_metrics_block { |
42 | struct tcp_metrics_block __rcu *tcpm_next; | |
a5443028 | 43 | struct inetpeer_addr tcpm_saddr; |
324fd55a | 44 | struct inetpeer_addr tcpm_daddr; |
51c5d0c4 | 45 | unsigned long tcpm_stamp; |
81166dd6 DM |
46 | u32 tcpm_ts; |
47 | u32 tcpm_ts_stamp; | |
51c5d0c4 | 48 | u32 tcpm_lock; |
740b0f18 | 49 | u32 tcpm_vals[TCP_METRIC_MAX_KERNEL + 1]; |
1fe4c481 | 50 | struct tcp_fastopen_metrics tcpm_fastopen; |
d23ff701 JA |
51 | |
52 | struct rcu_head rcu_head; | |
51c5d0c4 DM |
53 | }; |
54 | ||
55 | static bool tcp_metric_locked(struct tcp_metrics_block *tm, | |
56 | enum tcp_metric_index idx) | |
57 | { | |
58 | return tm->tcpm_lock & (1 << idx); | |
59 | } | |
60 | ||
61 | static u32 tcp_metric_get(struct tcp_metrics_block *tm, | |
62 | enum tcp_metric_index idx) | |
63 | { | |
64 | return tm->tcpm_vals[idx]; | |
65 | } | |
66 | ||
51c5d0c4 DM |
67 | static void tcp_metric_set(struct tcp_metrics_block *tm, |
68 | enum tcp_metric_index idx, | |
69 | u32 val) | |
70 | { | |
71 | tm->tcpm_vals[idx] = val; | |
72 | } | |
73 | ||
51c5d0c4 DM |
74 | static bool addr_same(const struct inetpeer_addr *a, |
75 | const struct inetpeer_addr *b) | |
76 | { | |
77 | const struct in6_addr *a6, *b6; | |
78 | ||
79 | if (a->family != b->family) | |
80 | return false; | |
81 | if (a->family == AF_INET) | |
82 | return a->addr.a4 == b->addr.a4; | |
83 | ||
84 | a6 = (const struct in6_addr *) &a->addr.a6[0]; | |
85 | b6 = (const struct in6_addr *) &b->addr.a6[0]; | |
86 | ||
87 | return ipv6_addr_equal(a6, b6); | |
88 | } | |
89 | ||
90 | struct tcpm_hash_bucket { | |
91 | struct tcp_metrics_block __rcu *chain; | |
92 | }; | |
93 | ||
94 | static DEFINE_SPINLOCK(tcp_metrics_lock); | |
95 | ||
740b0f18 ED |
96 | static void tcpm_suck_dst(struct tcp_metrics_block *tm, |
97 | const struct dst_entry *dst, | |
efeaa555 | 98 | bool fastopen_clear) |
51c5d0c4 | 99 | { |
740b0f18 | 100 | u32 msval; |
51c5d0c4 DM |
101 | u32 val; |
102 | ||
9a0a9502 JA |
103 | tm->tcpm_stamp = jiffies; |
104 | ||
51c5d0c4 DM |
105 | val = 0; |
106 | if (dst_metric_locked(dst, RTAX_RTT)) | |
107 | val |= 1 << TCP_METRIC_RTT; | |
108 | if (dst_metric_locked(dst, RTAX_RTTVAR)) | |
109 | val |= 1 << TCP_METRIC_RTTVAR; | |
110 | if (dst_metric_locked(dst, RTAX_SSTHRESH)) | |
111 | val |= 1 << TCP_METRIC_SSTHRESH; | |
112 | if (dst_metric_locked(dst, RTAX_CWND)) | |
113 | val |= 1 << TCP_METRIC_CWND; | |
114 | if (dst_metric_locked(dst, RTAX_REORDERING)) | |
115 | val |= 1 << TCP_METRIC_REORDERING; | |
116 | tm->tcpm_lock = val; | |
117 | ||
740b0f18 ED |
118 | msval = dst_metric_raw(dst, RTAX_RTT); |
119 | tm->tcpm_vals[TCP_METRIC_RTT] = msval * USEC_PER_MSEC; | |
120 | ||
121 | msval = dst_metric_raw(dst, RTAX_RTTVAR); | |
122 | tm->tcpm_vals[TCP_METRIC_RTTVAR] = msval * USEC_PER_MSEC; | |
51c5d0c4 DM |
123 | tm->tcpm_vals[TCP_METRIC_SSTHRESH] = dst_metric_raw(dst, RTAX_SSTHRESH); |
124 | tm->tcpm_vals[TCP_METRIC_CWND] = dst_metric_raw(dst, RTAX_CWND); | |
125 | tm->tcpm_vals[TCP_METRIC_REORDERING] = dst_metric_raw(dst, RTAX_REORDERING); | |
81166dd6 DM |
126 | tm->tcpm_ts = 0; |
127 | tm->tcpm_ts_stamp = 0; | |
efeaa555 ED |
128 | if (fastopen_clear) { |
129 | tm->tcpm_fastopen.mss = 0; | |
130 | tm->tcpm_fastopen.syn_loss = 0; | |
131 | tm->tcpm_fastopen.cookie.len = 0; | |
132 | } | |
51c5d0c4 DM |
133 | } |
134 | ||
77f99ad1 CP |
135 | #define TCP_METRICS_TIMEOUT (60 * 60 * HZ) |
136 | ||
137 | static void tcpm_check_stamp(struct tcp_metrics_block *tm, struct dst_entry *dst) | |
138 | { | |
139 | if (tm && unlikely(time_after(jiffies, tm->tcpm_stamp + TCP_METRICS_TIMEOUT))) | |
140 | tcpm_suck_dst(tm, dst, false); | |
141 | } | |
142 | ||
143 | #define TCP_METRICS_RECLAIM_DEPTH 5 | |
144 | #define TCP_METRICS_RECLAIM_PTR (struct tcp_metrics_block *) 0x1UL | |
145 | ||
51c5d0c4 | 146 | static struct tcp_metrics_block *tcpm_new(struct dst_entry *dst, |
a5443028 | 147 | struct inetpeer_addr *saddr, |
324fd55a | 148 | struct inetpeer_addr *daddr, |
77f99ad1 | 149 | unsigned int hash) |
51c5d0c4 DM |
150 | { |
151 | struct tcp_metrics_block *tm; | |
152 | struct net *net; | |
77f99ad1 | 153 | bool reclaim = false; |
51c5d0c4 DM |
154 | |
155 | spin_lock_bh(&tcp_metrics_lock); | |
156 | net = dev_net(dst->dev); | |
77f99ad1 CP |
157 | |
158 | /* While waiting for the spin-lock the cache might have been populated | |
159 | * with this entry and so we have to check again. | |
160 | */ | |
41804420 | 161 | tm = __tcp_get_metrics(saddr, daddr, net, hash); |
77f99ad1 CP |
162 | if (tm == TCP_METRICS_RECLAIM_PTR) { |
163 | reclaim = true; | |
164 | tm = NULL; | |
165 | } | |
166 | if (tm) { | |
167 | tcpm_check_stamp(tm, dst); | |
168 | goto out_unlock; | |
169 | } | |
170 | ||
51c5d0c4 DM |
171 | if (unlikely(reclaim)) { |
172 | struct tcp_metrics_block *oldest; | |
173 | ||
174 | oldest = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); | |
175 | for (tm = rcu_dereference(oldest->tcpm_next); tm; | |
176 | tm = rcu_dereference(tm->tcpm_next)) { | |
177 | if (time_before(tm->tcpm_stamp, oldest->tcpm_stamp)) | |
178 | oldest = tm; | |
179 | } | |
180 | tm = oldest; | |
181 | } else { | |
182 | tm = kmalloc(sizeof(*tm), GFP_ATOMIC); | |
183 | if (!tm) | |
184 | goto out_unlock; | |
185 | } | |
a5443028 | 186 | tm->tcpm_saddr = *saddr; |
324fd55a | 187 | tm->tcpm_daddr = *daddr; |
51c5d0c4 | 188 | |
efeaa555 | 189 | tcpm_suck_dst(tm, dst, true); |
51c5d0c4 DM |
190 | |
191 | if (likely(!reclaim)) { | |
192 | tm->tcpm_next = net->ipv4.tcp_metrics_hash[hash].chain; | |
193 | rcu_assign_pointer(net->ipv4.tcp_metrics_hash[hash].chain, tm); | |
194 | } | |
195 | ||
196 | out_unlock: | |
197 | spin_unlock_bh(&tcp_metrics_lock); | |
198 | return tm; | |
199 | } | |
200 | ||
51c5d0c4 DM |
201 | static struct tcp_metrics_block *tcp_get_encode(struct tcp_metrics_block *tm, int depth) |
202 | { | |
203 | if (tm) | |
204 | return tm; | |
205 | if (depth > TCP_METRICS_RECLAIM_DEPTH) | |
206 | return TCP_METRICS_RECLAIM_PTR; | |
207 | return NULL; | |
208 | } | |
209 | ||
a5443028 CP |
210 | static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr, |
211 | const struct inetpeer_addr *daddr, | |
51c5d0c4 DM |
212 | struct net *net, unsigned int hash) |
213 | { | |
214 | struct tcp_metrics_block *tm; | |
215 | int depth = 0; | |
216 | ||
217 | for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm; | |
218 | tm = rcu_dereference(tm->tcpm_next)) { | |
a5443028 CP |
219 | if (addr_same(&tm->tcpm_saddr, saddr) && |
220 | addr_same(&tm->tcpm_daddr, daddr)) | |
51c5d0c4 DM |
221 | break; |
222 | depth++; | |
223 | } | |
224 | return tcp_get_encode(tm, depth); | |
225 | } | |
226 | ||
227 | static struct tcp_metrics_block *__tcp_get_metrics_req(struct request_sock *req, | |
228 | struct dst_entry *dst) | |
229 | { | |
230 | struct tcp_metrics_block *tm; | |
a5443028 | 231 | struct inetpeer_addr saddr, daddr; |
51c5d0c4 DM |
232 | unsigned int hash; |
233 | struct net *net; | |
234 | ||
a5443028 | 235 | saddr.family = req->rsk_ops->family; |
324fd55a CP |
236 | daddr.family = req->rsk_ops->family; |
237 | switch (daddr.family) { | |
51c5d0c4 | 238 | case AF_INET: |
a5443028 | 239 | saddr.addr.a4 = inet_rsk(req)->ir_loc_addr; |
324fd55a CP |
240 | daddr.addr.a4 = inet_rsk(req)->ir_rmt_addr; |
241 | hash = (__force unsigned int) daddr.addr.a4; | |
51c5d0c4 | 242 | break; |
634fb979 | 243 | #if IS_ENABLED(CONFIG_IPV6) |
51c5d0c4 | 244 | case AF_INET6: |
a5443028 | 245 | *(struct in6_addr *)saddr.addr.a6 = inet_rsk(req)->ir_v6_loc_addr; |
324fd55a | 246 | *(struct in6_addr *)daddr.addr.a6 = inet_rsk(req)->ir_v6_rmt_addr; |
634fb979 | 247 | hash = ipv6_addr_hash(&inet_rsk(req)->ir_v6_rmt_addr); |
51c5d0c4 | 248 | break; |
634fb979 | 249 | #endif |
51c5d0c4 DM |
250 | default: |
251 | return NULL; | |
252 | } | |
253 | ||
51c5d0c4 | 254 | net = dev_net(dst->dev); |
3e5da62d | 255 | hash ^= net_hash_mix(net); |
5815d5e7 | 256 | hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log); |
51c5d0c4 DM |
257 | |
258 | for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm; | |
259 | tm = rcu_dereference(tm->tcpm_next)) { | |
a5443028 CP |
260 | if (addr_same(&tm->tcpm_saddr, &saddr) && |
261 | addr_same(&tm->tcpm_daddr, &daddr)) | |
51c5d0c4 DM |
262 | break; |
263 | } | |
264 | tcpm_check_stamp(tm, dst); | |
265 | return tm; | |
266 | } | |
267 | ||
81166dd6 DM |
268 | static struct tcp_metrics_block *__tcp_get_metrics_tw(struct inet_timewait_sock *tw) |
269 | { | |
81166dd6 | 270 | struct tcp_metrics_block *tm; |
a5443028 | 271 | struct inetpeer_addr saddr, daddr; |
81166dd6 DM |
272 | unsigned int hash; |
273 | struct net *net; | |
274 | ||
3ad88cf7 CP |
275 | if (tw->tw_family == AF_INET) { |
276 | saddr.family = AF_INET; | |
a5443028 | 277 | saddr.addr.a4 = tw->tw_rcv_saddr; |
3ad88cf7 | 278 | daddr.family = AF_INET; |
324fd55a CP |
279 | daddr.addr.a4 = tw->tw_daddr; |
280 | hash = (__force unsigned int) daddr.addr.a4; | |
3ad88cf7 | 281 | } |
c2bb06db | 282 | #if IS_ENABLED(CONFIG_IPV6) |
3ad88cf7 CP |
283 | else if (tw->tw_family == AF_INET6) { |
284 | if (ipv6_addr_v4mapped(&tw->tw_v6_daddr)) { | |
285 | saddr.family = AF_INET; | |
286 | saddr.addr.a4 = tw->tw_rcv_saddr; | |
287 | daddr.family = AF_INET; | |
288 | daddr.addr.a4 = tw->tw_daddr; | |
289 | hash = (__force unsigned int) daddr.addr.a4; | |
290 | } else { | |
291 | saddr.family = AF_INET6; | |
292 | *(struct in6_addr *)saddr.addr.a6 = tw->tw_v6_rcv_saddr; | |
293 | daddr.family = AF_INET6; | |
294 | *(struct in6_addr *)daddr.addr.a6 = tw->tw_v6_daddr; | |
295 | hash = ipv6_addr_hash(&tw->tw_v6_daddr); | |
296 | } | |
297 | } | |
c2bb06db | 298 | #endif |
3ad88cf7 | 299 | else |
81166dd6 | 300 | return NULL; |
81166dd6 | 301 | |
81166dd6 | 302 | net = twsk_net(tw); |
3e5da62d | 303 | hash ^= net_hash_mix(net); |
5815d5e7 | 304 | hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log); |
81166dd6 DM |
305 | |
306 | for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm; | |
307 | tm = rcu_dereference(tm->tcpm_next)) { | |
a5443028 CP |
308 | if (addr_same(&tm->tcpm_saddr, &saddr) && |
309 | addr_same(&tm->tcpm_daddr, &daddr)) | |
81166dd6 DM |
310 | break; |
311 | } | |
312 | return tm; | |
313 | } | |
314 | ||
51c5d0c4 DM |
315 | static struct tcp_metrics_block *tcp_get_metrics(struct sock *sk, |
316 | struct dst_entry *dst, | |
317 | bool create) | |
318 | { | |
319 | struct tcp_metrics_block *tm; | |
a5443028 | 320 | struct inetpeer_addr saddr, daddr; |
51c5d0c4 DM |
321 | unsigned int hash; |
322 | struct net *net; | |
51c5d0c4 | 323 | |
3ad88cf7 CP |
324 | if (sk->sk_family == AF_INET) { |
325 | saddr.family = AF_INET; | |
a5443028 | 326 | saddr.addr.a4 = inet_sk(sk)->inet_saddr; |
3ad88cf7 | 327 | daddr.family = AF_INET; |
324fd55a CP |
328 | daddr.addr.a4 = inet_sk(sk)->inet_daddr; |
329 | hash = (__force unsigned int) daddr.addr.a4; | |
3ad88cf7 | 330 | } |
c2bb06db | 331 | #if IS_ENABLED(CONFIG_IPV6) |
3ad88cf7 CP |
332 | else if (sk->sk_family == AF_INET6) { |
333 | if (ipv6_addr_v4mapped(&sk->sk_v6_daddr)) { | |
334 | saddr.family = AF_INET; | |
335 | saddr.addr.a4 = inet_sk(sk)->inet_saddr; | |
336 | daddr.family = AF_INET; | |
337 | daddr.addr.a4 = inet_sk(sk)->inet_daddr; | |
338 | hash = (__force unsigned int) daddr.addr.a4; | |
339 | } else { | |
340 | saddr.family = AF_INET6; | |
341 | *(struct in6_addr *)saddr.addr.a6 = sk->sk_v6_rcv_saddr; | |
342 | daddr.family = AF_INET6; | |
343 | *(struct in6_addr *)daddr.addr.a6 = sk->sk_v6_daddr; | |
344 | hash = ipv6_addr_hash(&sk->sk_v6_daddr); | |
345 | } | |
346 | } | |
c2bb06db | 347 | #endif |
3ad88cf7 | 348 | else |
51c5d0c4 | 349 | return NULL; |
51c5d0c4 | 350 | |
51c5d0c4 | 351 | net = dev_net(dst->dev); |
3e5da62d | 352 | hash ^= net_hash_mix(net); |
5815d5e7 | 353 | hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log); |
51c5d0c4 | 354 | |
a5443028 | 355 | tm = __tcp_get_metrics(&saddr, &daddr, net, hash); |
77f99ad1 | 356 | if (tm == TCP_METRICS_RECLAIM_PTR) |
51c5d0c4 | 357 | tm = NULL; |
51c5d0c4 | 358 | if (!tm && create) |
41804420 | 359 | tm = tcpm_new(dst, &saddr, &daddr, hash); |
51c5d0c4 DM |
360 | else |
361 | tcpm_check_stamp(tm, dst); | |
362 | ||
363 | return tm; | |
364 | } | |
365 | ||
4aabd8ef DM |
366 | /* Save metrics learned by this TCP session. This function is called |
367 | * only, when TCP finishes successfully i.e. when it enters TIME-WAIT | |
368 | * or goes from LAST-ACK to CLOSE. | |
369 | */ | |
370 | void tcp_update_metrics(struct sock *sk) | |
371 | { | |
51c5d0c4 | 372 | const struct inet_connection_sock *icsk = inet_csk(sk); |
4aabd8ef | 373 | struct dst_entry *dst = __sk_dst_get(sk); |
51c5d0c4 DM |
374 | struct tcp_sock *tp = tcp_sk(sk); |
375 | struct tcp_metrics_block *tm; | |
376 | unsigned long rtt; | |
377 | u32 val; | |
378 | int m; | |
4aabd8ef | 379 | |
51c5d0c4 | 380 | if (sysctl_tcp_nometrics_save || !dst) |
4aabd8ef DM |
381 | return; |
382 | ||
51c5d0c4 | 383 | if (dst->flags & DST_HOST) |
4aabd8ef DM |
384 | dst_confirm(dst); |
385 | ||
51c5d0c4 | 386 | rcu_read_lock(); |
740b0f18 | 387 | if (icsk->icsk_backoff || !tp->srtt_us) { |
51c5d0c4 DM |
388 | /* This session failed to estimate rtt. Why? |
389 | * Probably, no packets returned in time. Reset our | |
390 | * results. | |
391 | */ | |
392 | tm = tcp_get_metrics(sk, dst, false); | |
393 | if (tm && !tcp_metric_locked(tm, TCP_METRIC_RTT)) | |
394 | tcp_metric_set(tm, TCP_METRIC_RTT, 0); | |
395 | goto out_unlock; | |
396 | } else | |
397 | tm = tcp_get_metrics(sk, dst, true); | |
4aabd8ef | 398 | |
51c5d0c4 DM |
399 | if (!tm) |
400 | goto out_unlock; | |
4aabd8ef | 401 | |
740b0f18 ED |
402 | rtt = tcp_metric_get(tm, TCP_METRIC_RTT); |
403 | m = rtt - tp->srtt_us; | |
4aabd8ef | 404 | |
51c5d0c4 DM |
405 | /* If newly calculated rtt larger than stored one, store new |
406 | * one. Otherwise, use EWMA. Remember, rtt overestimation is | |
407 | * always better than underestimation. | |
408 | */ | |
409 | if (!tcp_metric_locked(tm, TCP_METRIC_RTT)) { | |
410 | if (m <= 0) | |
740b0f18 | 411 | rtt = tp->srtt_us; |
51c5d0c4 DM |
412 | else |
413 | rtt -= (m >> 3); | |
740b0f18 | 414 | tcp_metric_set(tm, TCP_METRIC_RTT, rtt); |
51c5d0c4 | 415 | } |
4aabd8ef | 416 | |
51c5d0c4 DM |
417 | if (!tcp_metric_locked(tm, TCP_METRIC_RTTVAR)) { |
418 | unsigned long var; | |
4aabd8ef | 419 | |
51c5d0c4 DM |
420 | if (m < 0) |
421 | m = -m; | |
4aabd8ef | 422 | |
51c5d0c4 DM |
423 | /* Scale deviation to rttvar fixed point */ |
424 | m >>= 1; | |
740b0f18 ED |
425 | if (m < tp->mdev_us) |
426 | m = tp->mdev_us; | |
4aabd8ef | 427 | |
740b0f18 | 428 | var = tcp_metric_get(tm, TCP_METRIC_RTTVAR); |
51c5d0c4 DM |
429 | if (m >= var) |
430 | var = m; | |
431 | else | |
432 | var -= (var - m) >> 2; | |
4aabd8ef | 433 | |
740b0f18 | 434 | tcp_metric_set(tm, TCP_METRIC_RTTVAR, var); |
51c5d0c4 DM |
435 | } |
436 | ||
437 | if (tcp_in_initial_slowstart(tp)) { | |
438 | /* Slow start still did not finish. */ | |
439 | if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) { | |
440 | val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH); | |
441 | if (val && (tp->snd_cwnd >> 1) > val) | |
442 | tcp_metric_set(tm, TCP_METRIC_SSTHRESH, | |
443 | tp->snd_cwnd >> 1); | |
444 | } | |
445 | if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) { | |
446 | val = tcp_metric_get(tm, TCP_METRIC_CWND); | |
447 | if (tp->snd_cwnd > val) | |
448 | tcp_metric_set(tm, TCP_METRIC_CWND, | |
449 | tp->snd_cwnd); | |
450 | } | |
451 | } else if (tp->snd_cwnd > tp->snd_ssthresh && | |
452 | icsk->icsk_ca_state == TCP_CA_Open) { | |
453 | /* Cong. avoidance phase, cwnd is reliable. */ | |
454 | if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) | |
455 | tcp_metric_set(tm, TCP_METRIC_SSTHRESH, | |
456 | max(tp->snd_cwnd >> 1, tp->snd_ssthresh)); | |
457 | if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) { | |
458 | val = tcp_metric_get(tm, TCP_METRIC_CWND); | |
2100844c | 459 | tcp_metric_set(tm, TCP_METRIC_CWND, (val + tp->snd_cwnd) >> 1); |
51c5d0c4 DM |
460 | } |
461 | } else { | |
462 | /* Else slow start did not finish, cwnd is non-sense, | |
463 | * ssthresh may be also invalid. | |
464 | */ | |
465 | if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) { | |
466 | val = tcp_metric_get(tm, TCP_METRIC_CWND); | |
467 | tcp_metric_set(tm, TCP_METRIC_CWND, | |
468 | (val + tp->snd_ssthresh) >> 1); | |
469 | } | |
470 | if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) { | |
471 | val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH); | |
472 | if (val && tp->snd_ssthresh > val) | |
473 | tcp_metric_set(tm, TCP_METRIC_SSTHRESH, | |
474 | tp->snd_ssthresh); | |
475 | } | |
476 | if (!tcp_metric_locked(tm, TCP_METRIC_REORDERING)) { | |
477 | val = tcp_metric_get(tm, TCP_METRIC_REORDERING); | |
478 | if (val < tp->reordering && | |
4aabd8ef | 479 | tp->reordering != sysctl_tcp_reordering) |
51c5d0c4 DM |
480 | tcp_metric_set(tm, TCP_METRIC_REORDERING, |
481 | tp->reordering); | |
4aabd8ef DM |
482 | } |
483 | } | |
51c5d0c4 DM |
484 | tm->tcpm_stamp = jiffies; |
485 | out_unlock: | |
486 | rcu_read_unlock(); | |
4aabd8ef DM |
487 | } |
488 | ||
489 | /* Initialize metrics on socket. */ | |
490 | ||
491 | void tcp_init_metrics(struct sock *sk) | |
492 | { | |
4aabd8ef | 493 | struct dst_entry *dst = __sk_dst_get(sk); |
51c5d0c4 DM |
494 | struct tcp_sock *tp = tcp_sk(sk); |
495 | struct tcp_metrics_block *tm; | |
1b7fdd2a | 496 | u32 val, crtt = 0; /* cached RTT scaled by 8 */ |
4aabd8ef DM |
497 | |
498 | if (dst == NULL) | |
499 | goto reset; | |
500 | ||
501 | dst_confirm(dst); | |
502 | ||
51c5d0c4 DM |
503 | rcu_read_lock(); |
504 | tm = tcp_get_metrics(sk, dst, true); | |
505 | if (!tm) { | |
506 | rcu_read_unlock(); | |
507 | goto reset; | |
508 | } | |
509 | ||
510 | if (tcp_metric_locked(tm, TCP_METRIC_CWND)) | |
511 | tp->snd_cwnd_clamp = tcp_metric_get(tm, TCP_METRIC_CWND); | |
512 | ||
513 | val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH); | |
514 | if (val) { | |
515 | tp->snd_ssthresh = val; | |
4aabd8ef DM |
516 | if (tp->snd_ssthresh > tp->snd_cwnd_clamp) |
517 | tp->snd_ssthresh = tp->snd_cwnd_clamp; | |
518 | } else { | |
519 | /* ssthresh may have been reduced unnecessarily during. | |
520 | * 3WHS. Restore it back to its initial default. | |
521 | */ | |
522 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; | |
523 | } | |
51c5d0c4 DM |
524 | val = tcp_metric_get(tm, TCP_METRIC_REORDERING); |
525 | if (val && tp->reordering != val) { | |
4aabd8ef DM |
526 | tcp_disable_fack(tp); |
527 | tcp_disable_early_retrans(tp); | |
51c5d0c4 | 528 | tp->reordering = val; |
4aabd8ef DM |
529 | } |
530 | ||
740b0f18 | 531 | crtt = tcp_metric_get(tm, TCP_METRIC_RTT); |
51c5d0c4 | 532 | rcu_read_unlock(); |
4aabd8ef | 533 | reset: |
52f20e65 YC |
534 | /* The initial RTT measurement from the SYN/SYN-ACK is not ideal |
535 | * to seed the RTO for later data packets because SYN packets are | |
536 | * small. Use the per-dst cached values to seed the RTO but keep | |
537 | * the RTT estimator variables intact (e.g., srtt, mdev, rttvar). | |
538 | * Later the RTO will be updated immediately upon obtaining the first | |
539 | * data RTT sample (tcp_rtt_estimator()). Hence the cached RTT only | |
540 | * influences the first RTO but not later RTT estimation. | |
541 | * | |
542 | * But if RTT is not available from the SYN (due to retransmits or | |
543 | * syn cookies) or the cache, force a conservative 3secs timeout. | |
544 | * | |
545 | * A bit of theory. RTT is time passed after "normal" sized packet | |
546 | * is sent until it is ACKed. In normal circumstances sending small | |
547 | * packets force peer to delay ACKs and calculation is correct too. | |
548 | * The algorithm is adaptive and, provided we follow specs, it | |
549 | * NEVER underestimate RTT. BUT! If peer tries to make some clever | |
550 | * tricks sort of "quick acks" for time long enough to decrease RTT | |
551 | * to low value, and then abruptly stops to do it and starts to delay | |
552 | * ACKs, wait for troubles. | |
553 | */ | |
740b0f18 | 554 | if (crtt > tp->srtt_us) { |
269aa759 | 555 | /* Set RTO like tcp_rtt_estimator(), but from cached RTT. */ |
740b0f18 | 556 | crtt /= 8 * USEC_PER_MSEC; |
269aa759 | 557 | inet_csk(sk)->icsk_rto = crtt + max(2 * crtt, tcp_rto_min(sk)); |
740b0f18 | 558 | } else if (tp->srtt_us == 0) { |
4aabd8ef DM |
559 | /* RFC6298: 5.7 We've failed to get a valid RTT sample from |
560 | * 3WHS. This is most likely due to retransmission, | |
561 | * including spurious one. Reset the RTO back to 3secs | |
562 | * from the more aggressive 1sec to avoid more spurious | |
563 | * retransmission. | |
564 | */ | |
740b0f18 ED |
565 | tp->rttvar_us = jiffies_to_usecs(TCP_TIMEOUT_FALLBACK); |
566 | tp->mdev_us = tp->mdev_max_us = tp->rttvar_us; | |
567 | ||
4aabd8ef DM |
568 | inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK; |
569 | } | |
570 | /* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been | |
571 | * retransmitted. In light of RFC6298 more aggressive 1sec | |
572 | * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK | |
573 | * retransmission has occurred. | |
574 | */ | |
575 | if (tp->total_retrans > 1) | |
576 | tp->snd_cwnd = 1; | |
577 | else | |
578 | tp->snd_cwnd = tcp_init_cwnd(tp, dst); | |
579 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
580 | } | |
ab92bb2f | 581 | |
a26552af HFS |
582 | bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, |
583 | bool paws_check, bool timestamps) | |
ab92bb2f | 584 | { |
51c5d0c4 DM |
585 | struct tcp_metrics_block *tm; |
586 | bool ret; | |
587 | ||
ab92bb2f DM |
588 | if (!dst) |
589 | return false; | |
51c5d0c4 DM |
590 | |
591 | rcu_read_lock(); | |
592 | tm = __tcp_get_metrics_req(req, dst); | |
81166dd6 DM |
593 | if (paws_check) { |
594 | if (tm && | |
595 | (u32)get_seconds() - tm->tcpm_ts_stamp < TCP_PAWS_MSL && | |
a26552af HFS |
596 | ((s32)(tm->tcpm_ts - req->ts_recent) > TCP_PAWS_WINDOW || |
597 | !timestamps)) | |
81166dd6 DM |
598 | ret = false; |
599 | else | |
600 | ret = true; | |
601 | } else { | |
602 | if (tm && tcp_metric_get(tm, TCP_METRIC_RTT) && tm->tcpm_ts_stamp) | |
603 | ret = true; | |
604 | else | |
605 | ret = false; | |
606 | } | |
51c5d0c4 DM |
607 | rcu_read_unlock(); |
608 | ||
609 | return ret; | |
ab92bb2f DM |
610 | } |
611 | EXPORT_SYMBOL_GPL(tcp_peer_is_proven); | |
51c5d0c4 | 612 | |
81166dd6 DM |
613 | void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst) |
614 | { | |
615 | struct tcp_metrics_block *tm; | |
616 | ||
617 | rcu_read_lock(); | |
618 | tm = tcp_get_metrics(sk, dst, true); | |
619 | if (tm) { | |
620 | struct tcp_sock *tp = tcp_sk(sk); | |
621 | ||
622 | if ((u32)get_seconds() - tm->tcpm_ts_stamp <= TCP_PAWS_MSL) { | |
623 | tp->rx_opt.ts_recent_stamp = tm->tcpm_ts_stamp; | |
624 | tp->rx_opt.ts_recent = tm->tcpm_ts; | |
625 | } | |
626 | } | |
627 | rcu_read_unlock(); | |
628 | } | |
629 | EXPORT_SYMBOL_GPL(tcp_fetch_timewait_stamp); | |
630 | ||
631 | /* VJ's idea. Save last timestamp seen from this destination and hold | |
632 | * it at least for normal timewait interval to use for duplicate | |
633 | * segment detection in subsequent connections, before they enter | |
634 | * synchronized state. | |
635 | */ | |
636 | bool tcp_remember_stamp(struct sock *sk) | |
637 | { | |
638 | struct dst_entry *dst = __sk_dst_get(sk); | |
639 | bool ret = false; | |
640 | ||
641 | if (dst) { | |
642 | struct tcp_metrics_block *tm; | |
643 | ||
644 | rcu_read_lock(); | |
645 | tm = tcp_get_metrics(sk, dst, true); | |
646 | if (tm) { | |
647 | struct tcp_sock *tp = tcp_sk(sk); | |
648 | ||
649 | if ((s32)(tm->tcpm_ts - tp->rx_opt.ts_recent) <= 0 || | |
650 | ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL && | |
651 | tm->tcpm_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) { | |
652 | tm->tcpm_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp; | |
653 | tm->tcpm_ts = tp->rx_opt.ts_recent; | |
654 | } | |
655 | ret = true; | |
656 | } | |
657 | rcu_read_unlock(); | |
658 | } | |
659 | return ret; | |
660 | } | |
661 | ||
662 | bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw) | |
663 | { | |
664 | struct tcp_metrics_block *tm; | |
665 | bool ret = false; | |
666 | ||
667 | rcu_read_lock(); | |
668 | tm = __tcp_get_metrics_tw(tw); | |
9a0a9502 | 669 | if (tm) { |
81166dd6 DM |
670 | const struct tcp_timewait_sock *tcptw; |
671 | struct sock *sk = (struct sock *) tw; | |
672 | ||
673 | tcptw = tcp_twsk(sk); | |
674 | if ((s32)(tm->tcpm_ts - tcptw->tw_ts_recent) <= 0 || | |
675 | ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL && | |
676 | tm->tcpm_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) { | |
677 | tm->tcpm_ts_stamp = (u32)tcptw->tw_ts_recent_stamp; | |
678 | tm->tcpm_ts = tcptw->tw_ts_recent; | |
679 | } | |
680 | ret = true; | |
681 | } | |
682 | rcu_read_unlock(); | |
683 | ||
684 | return ret; | |
685 | } | |
686 | ||
1fe4c481 YC |
687 | static DEFINE_SEQLOCK(fastopen_seqlock); |
688 | ||
689 | void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, | |
aab48743 YC |
690 | struct tcp_fastopen_cookie *cookie, |
691 | int *syn_loss, unsigned long *last_syn_loss) | |
1fe4c481 YC |
692 | { |
693 | struct tcp_metrics_block *tm; | |
694 | ||
695 | rcu_read_lock(); | |
696 | tm = tcp_get_metrics(sk, __sk_dst_get(sk), false); | |
697 | if (tm) { | |
698 | struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen; | |
699 | unsigned int seq; | |
700 | ||
701 | do { | |
702 | seq = read_seqbegin(&fastopen_seqlock); | |
703 | if (tfom->mss) | |
704 | *mss = tfom->mss; | |
705 | *cookie = tfom->cookie; | |
aab48743 YC |
706 | *syn_loss = tfom->syn_loss; |
707 | *last_syn_loss = *syn_loss ? tfom->last_syn_loss : 0; | |
1fe4c481 YC |
708 | } while (read_seqretry(&fastopen_seqlock, seq)); |
709 | } | |
710 | rcu_read_unlock(); | |
711 | } | |
712 | ||
1fe4c481 | 713 | void tcp_fastopen_cache_set(struct sock *sk, u16 mss, |
aab48743 | 714 | struct tcp_fastopen_cookie *cookie, bool syn_lost) |
1fe4c481 | 715 | { |
dccf76ca | 716 | struct dst_entry *dst = __sk_dst_get(sk); |
1fe4c481 YC |
717 | struct tcp_metrics_block *tm; |
718 | ||
dccf76ca ED |
719 | if (!dst) |
720 | return; | |
1fe4c481 | 721 | rcu_read_lock(); |
dccf76ca | 722 | tm = tcp_get_metrics(sk, dst, true); |
1fe4c481 YC |
723 | if (tm) { |
724 | struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen; | |
725 | ||
726 | write_seqlock_bh(&fastopen_seqlock); | |
c968601d YC |
727 | if (mss) |
728 | tfom->mss = mss; | |
729 | if (cookie && cookie->len > 0) | |
1fe4c481 | 730 | tfom->cookie = *cookie; |
aab48743 YC |
731 | if (syn_lost) { |
732 | ++tfom->syn_loss; | |
733 | tfom->last_syn_loss = jiffies; | |
734 | } else | |
735 | tfom->syn_loss = 0; | |
1fe4c481 YC |
736 | write_sequnlock_bh(&fastopen_seqlock); |
737 | } | |
738 | rcu_read_unlock(); | |
739 | } | |
740 | ||
d23ff701 JA |
741 | static struct genl_family tcp_metrics_nl_family = { |
742 | .id = GENL_ID_GENERATE, | |
743 | .hdrsize = 0, | |
744 | .name = TCP_METRICS_GENL_NAME, | |
745 | .version = TCP_METRICS_GENL_VERSION, | |
746 | .maxattr = TCP_METRICS_ATTR_MAX, | |
747 | .netnsok = true, | |
748 | }; | |
749 | ||
750 | static struct nla_policy tcp_metrics_nl_policy[TCP_METRICS_ATTR_MAX + 1] = { | |
751 | [TCP_METRICS_ATTR_ADDR_IPV4] = { .type = NLA_U32, }, | |
752 | [TCP_METRICS_ATTR_ADDR_IPV6] = { .type = NLA_BINARY, | |
753 | .len = sizeof(struct in6_addr), }, | |
754 | /* Following attributes are not received for GET/DEL, | |
755 | * we keep them for reference | |
756 | */ | |
757 | #if 0 | |
758 | [TCP_METRICS_ATTR_AGE] = { .type = NLA_MSECS, }, | |
759 | [TCP_METRICS_ATTR_TW_TSVAL] = { .type = NLA_U32, }, | |
760 | [TCP_METRICS_ATTR_TW_TS_STAMP] = { .type = NLA_S32, }, | |
761 | [TCP_METRICS_ATTR_VALS] = { .type = NLA_NESTED, }, | |
762 | [TCP_METRICS_ATTR_FOPEN_MSS] = { .type = NLA_U16, }, | |
763 | [TCP_METRICS_ATTR_FOPEN_SYN_DROPS] = { .type = NLA_U16, }, | |
764 | [TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS] = { .type = NLA_MSECS, }, | |
765 | [TCP_METRICS_ATTR_FOPEN_COOKIE] = { .type = NLA_BINARY, | |
766 | .len = TCP_FASTOPEN_COOKIE_MAX, }, | |
767 | #endif | |
768 | }; | |
769 | ||
770 | /* Add attributes, caller cancels its header on failure */ | |
771 | static int tcp_metrics_fill_info(struct sk_buff *msg, | |
772 | struct tcp_metrics_block *tm) | |
773 | { | |
774 | struct nlattr *nest; | |
775 | int i; | |
776 | ||
324fd55a | 777 | switch (tm->tcpm_daddr.family) { |
d23ff701 JA |
778 | case AF_INET: |
779 | if (nla_put_be32(msg, TCP_METRICS_ATTR_ADDR_IPV4, | |
324fd55a | 780 | tm->tcpm_daddr.addr.a4) < 0) |
d23ff701 | 781 | goto nla_put_failure; |
8a59359c CP |
782 | if (nla_put_be32(msg, TCP_METRICS_ATTR_SADDR_IPV4, |
783 | tm->tcpm_saddr.addr.a4) < 0) | |
784 | goto nla_put_failure; | |
d23ff701 JA |
785 | break; |
786 | case AF_INET6: | |
787 | if (nla_put(msg, TCP_METRICS_ATTR_ADDR_IPV6, 16, | |
324fd55a | 788 | tm->tcpm_daddr.addr.a6) < 0) |
d23ff701 | 789 | goto nla_put_failure; |
8a59359c CP |
790 | if (nla_put(msg, TCP_METRICS_ATTR_SADDR_IPV6, 16, |
791 | tm->tcpm_saddr.addr.a6) < 0) | |
792 | goto nla_put_failure; | |
d23ff701 JA |
793 | break; |
794 | default: | |
795 | return -EAFNOSUPPORT; | |
796 | } | |
797 | ||
798 | if (nla_put_msecs(msg, TCP_METRICS_ATTR_AGE, | |
799 | jiffies - tm->tcpm_stamp) < 0) | |
800 | goto nla_put_failure; | |
801 | if (tm->tcpm_ts_stamp) { | |
802 | if (nla_put_s32(msg, TCP_METRICS_ATTR_TW_TS_STAMP, | |
803 | (s32) (get_seconds() - tm->tcpm_ts_stamp)) < 0) | |
804 | goto nla_put_failure; | |
805 | if (nla_put_u32(msg, TCP_METRICS_ATTR_TW_TSVAL, | |
806 | tm->tcpm_ts) < 0) | |
807 | goto nla_put_failure; | |
808 | } | |
809 | ||
810 | { | |
811 | int n = 0; | |
812 | ||
813 | nest = nla_nest_start(msg, TCP_METRICS_ATTR_VALS); | |
814 | if (!nest) | |
815 | goto nla_put_failure; | |
740b0f18 ED |
816 | for (i = 0; i < TCP_METRIC_MAX_KERNEL + 1; i++) { |
817 | u32 val = tm->tcpm_vals[i]; | |
818 | ||
819 | if (!val) | |
d23ff701 | 820 | continue; |
740b0f18 ED |
821 | if (i == TCP_METRIC_RTT) { |
822 | if (nla_put_u32(msg, TCP_METRIC_RTT_US + 1, | |
823 | val) < 0) | |
824 | goto nla_put_failure; | |
825 | n++; | |
826 | val = max(val / 1000, 1U); | |
827 | } | |
828 | if (i == TCP_METRIC_RTTVAR) { | |
829 | if (nla_put_u32(msg, TCP_METRIC_RTTVAR_US + 1, | |
830 | val) < 0) | |
831 | goto nla_put_failure; | |
832 | n++; | |
833 | val = max(val / 1000, 1U); | |
834 | } | |
835 | if (nla_put_u32(msg, i + 1, val) < 0) | |
d23ff701 JA |
836 | goto nla_put_failure; |
837 | n++; | |
838 | } | |
839 | if (n) | |
840 | nla_nest_end(msg, nest); | |
841 | else | |
842 | nla_nest_cancel(msg, nest); | |
843 | } | |
844 | ||
845 | { | |
846 | struct tcp_fastopen_metrics tfom_copy[1], *tfom; | |
847 | unsigned int seq; | |
848 | ||
849 | do { | |
850 | seq = read_seqbegin(&fastopen_seqlock); | |
851 | tfom_copy[0] = tm->tcpm_fastopen; | |
852 | } while (read_seqretry(&fastopen_seqlock, seq)); | |
853 | ||
854 | tfom = tfom_copy; | |
855 | if (tfom->mss && | |
856 | nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_MSS, | |
857 | tfom->mss) < 0) | |
858 | goto nla_put_failure; | |
859 | if (tfom->syn_loss && | |
860 | (nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROPS, | |
861 | tfom->syn_loss) < 0 || | |
862 | nla_put_msecs(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS, | |
863 | jiffies - tfom->last_syn_loss) < 0)) | |
864 | goto nla_put_failure; | |
865 | if (tfom->cookie.len > 0 && | |
866 | nla_put(msg, TCP_METRICS_ATTR_FOPEN_COOKIE, | |
867 | tfom->cookie.len, tfom->cookie.val) < 0) | |
868 | goto nla_put_failure; | |
869 | } | |
870 | ||
871 | return 0; | |
872 | ||
873 | nla_put_failure: | |
874 | return -EMSGSIZE; | |
875 | } | |
876 | ||
877 | static int tcp_metrics_dump_info(struct sk_buff *skb, | |
878 | struct netlink_callback *cb, | |
879 | struct tcp_metrics_block *tm) | |
880 | { | |
881 | void *hdr; | |
882 | ||
15e47304 | 883 | hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, |
d23ff701 JA |
884 | &tcp_metrics_nl_family, NLM_F_MULTI, |
885 | TCP_METRICS_CMD_GET); | |
886 | if (!hdr) | |
887 | return -EMSGSIZE; | |
888 | ||
889 | if (tcp_metrics_fill_info(skb, tm) < 0) | |
890 | goto nla_put_failure; | |
891 | ||
053c095a JB |
892 | genlmsg_end(skb, hdr); |
893 | return 0; | |
d23ff701 JA |
894 | |
895 | nla_put_failure: | |
896 | genlmsg_cancel(skb, hdr); | |
897 | return -EMSGSIZE; | |
898 | } | |
899 | ||
900 | static int tcp_metrics_nl_dump(struct sk_buff *skb, | |
901 | struct netlink_callback *cb) | |
902 | { | |
903 | struct net *net = sock_net(skb->sk); | |
904 | unsigned int max_rows = 1U << net->ipv4.tcp_metrics_hash_log; | |
905 | unsigned int row, s_row = cb->args[0]; | |
906 | int s_col = cb->args[1], col = s_col; | |
907 | ||
908 | for (row = s_row; row < max_rows; row++, s_col = 0) { | |
909 | struct tcp_metrics_block *tm; | |
910 | struct tcpm_hash_bucket *hb = net->ipv4.tcp_metrics_hash + row; | |
911 | ||
912 | rcu_read_lock(); | |
913 | for (col = 0, tm = rcu_dereference(hb->chain); tm; | |
914 | tm = rcu_dereference(tm->tcpm_next), col++) { | |
915 | if (col < s_col) | |
916 | continue; | |
917 | if (tcp_metrics_dump_info(skb, cb, tm) < 0) { | |
918 | rcu_read_unlock(); | |
919 | goto done; | |
920 | } | |
921 | } | |
922 | rcu_read_unlock(); | |
923 | } | |
924 | ||
925 | done: | |
926 | cb->args[0] = row; | |
927 | cb->args[1] = col; | |
928 | return skb->len; | |
929 | } | |
930 | ||
3e7013dd CP |
931 | static int __parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr, |
932 | unsigned int *hash, int optional, int v4, int v6) | |
d23ff701 JA |
933 | { |
934 | struct nlattr *a; | |
935 | ||
3e7013dd | 936 | a = info->attrs[v4]; |
d23ff701 JA |
937 | if (a) { |
938 | addr->family = AF_INET; | |
939 | addr->addr.a4 = nla_get_be32(a); | |
3e7013dd CP |
940 | if (hash) |
941 | *hash = (__force unsigned int) addr->addr.a4; | |
d23ff701 JA |
942 | return 0; |
943 | } | |
3e7013dd | 944 | a = info->attrs[v6]; |
d23ff701 | 945 | if (a) { |
2c42a3fb | 946 | if (nla_len(a) != sizeof(struct in6_addr)) |
d23ff701 JA |
947 | return -EINVAL; |
948 | addr->family = AF_INET6; | |
949 | memcpy(addr->addr.a6, nla_data(a), sizeof(addr->addr.a6)); | |
3e7013dd CP |
950 | if (hash) |
951 | *hash = ipv6_addr_hash((struct in6_addr *) addr->addr.a6); | |
d23ff701 JA |
952 | return 0; |
953 | } | |
954 | return optional ? 1 : -EAFNOSUPPORT; | |
955 | } | |
956 | ||
3e7013dd CP |
957 | static int parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr, |
958 | unsigned int *hash, int optional) | |
959 | { | |
960 | return __parse_nl_addr(info, addr, hash, optional, | |
961 | TCP_METRICS_ATTR_ADDR_IPV4, | |
962 | TCP_METRICS_ATTR_ADDR_IPV6); | |
963 | } | |
964 | ||
965 | static int parse_nl_saddr(struct genl_info *info, struct inetpeer_addr *addr) | |
966 | { | |
967 | return __parse_nl_addr(info, addr, NULL, 0, | |
968 | TCP_METRICS_ATTR_SADDR_IPV4, | |
969 | TCP_METRICS_ATTR_SADDR_IPV6); | |
970 | } | |
971 | ||
d23ff701 JA |
972 | static int tcp_metrics_nl_cmd_get(struct sk_buff *skb, struct genl_info *info) |
973 | { | |
974 | struct tcp_metrics_block *tm; | |
3e7013dd | 975 | struct inetpeer_addr saddr, daddr; |
d23ff701 JA |
976 | unsigned int hash; |
977 | struct sk_buff *msg; | |
978 | struct net *net = genl_info_net(info); | |
979 | void *reply; | |
980 | int ret; | |
3e7013dd | 981 | bool src = true; |
d23ff701 | 982 | |
324fd55a | 983 | ret = parse_nl_addr(info, &daddr, &hash, 0); |
d23ff701 JA |
984 | if (ret < 0) |
985 | return ret; | |
986 | ||
3e7013dd CP |
987 | ret = parse_nl_saddr(info, &saddr); |
988 | if (ret < 0) | |
989 | src = false; | |
990 | ||
d23ff701 JA |
991 | msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); |
992 | if (!msg) | |
993 | return -ENOMEM; | |
994 | ||
995 | reply = genlmsg_put_reply(msg, info, &tcp_metrics_nl_family, 0, | |
996 | info->genlhdr->cmd); | |
997 | if (!reply) | |
998 | goto nla_put_failure; | |
999 | ||
3e5da62d | 1000 | hash ^= net_hash_mix(net); |
d23ff701 JA |
1001 | hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log); |
1002 | ret = -ESRCH; | |
1003 | rcu_read_lock(); | |
1004 | for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm; | |
1005 | tm = rcu_dereference(tm->tcpm_next)) { | |
3e7013dd CP |
1006 | if (addr_same(&tm->tcpm_daddr, &daddr) && |
1007 | (!src || addr_same(&tm->tcpm_saddr, &saddr))) { | |
d23ff701 JA |
1008 | ret = tcp_metrics_fill_info(msg, tm); |
1009 | break; | |
1010 | } | |
1011 | } | |
1012 | rcu_read_unlock(); | |
1013 | if (ret < 0) | |
1014 | goto out_free; | |
1015 | ||
1016 | genlmsg_end(msg, reply); | |
1017 | return genlmsg_reply(msg, info); | |
1018 | ||
1019 | nla_put_failure: | |
1020 | ret = -EMSGSIZE; | |
1021 | ||
1022 | out_free: | |
1023 | nlmsg_free(msg); | |
1024 | return ret; | |
1025 | } | |
1026 | ||
1027 | #define deref_locked_genl(p) \ | |
1028 | rcu_dereference_protected(p, lockdep_genl_is_held() && \ | |
1029 | lockdep_is_held(&tcp_metrics_lock)) | |
1030 | ||
1031 | #define deref_genl(p) rcu_dereference_protected(p, lockdep_genl_is_held()) | |
1032 | ||
1033 | static int tcp_metrics_flush_all(struct net *net) | |
1034 | { | |
1035 | unsigned int max_rows = 1U << net->ipv4.tcp_metrics_hash_log; | |
1036 | struct tcpm_hash_bucket *hb = net->ipv4.tcp_metrics_hash; | |
1037 | struct tcp_metrics_block *tm; | |
1038 | unsigned int row; | |
1039 | ||
1040 | for (row = 0; row < max_rows; row++, hb++) { | |
1041 | spin_lock_bh(&tcp_metrics_lock); | |
1042 | tm = deref_locked_genl(hb->chain); | |
1043 | if (tm) | |
1044 | hb->chain = NULL; | |
1045 | spin_unlock_bh(&tcp_metrics_lock); | |
1046 | while (tm) { | |
1047 | struct tcp_metrics_block *next; | |
1048 | ||
1049 | next = deref_genl(tm->tcpm_next); | |
1050 | kfree_rcu(tm, rcu_head); | |
1051 | tm = next; | |
1052 | } | |
1053 | } | |
1054 | return 0; | |
1055 | } | |
1056 | ||
1057 | static int tcp_metrics_nl_cmd_del(struct sk_buff *skb, struct genl_info *info) | |
1058 | { | |
1059 | struct tcpm_hash_bucket *hb; | |
00ca9c5b | 1060 | struct tcp_metrics_block *tm; |
d23ff701 | 1061 | struct tcp_metrics_block __rcu **pp; |
3e7013dd | 1062 | struct inetpeer_addr saddr, daddr; |
d23ff701 JA |
1063 | unsigned int hash; |
1064 | struct net *net = genl_info_net(info); | |
1065 | int ret; | |
00ca9c5b | 1066 | bool src = true, found = false; |
d23ff701 | 1067 | |
324fd55a | 1068 | ret = parse_nl_addr(info, &daddr, &hash, 1); |
d23ff701 JA |
1069 | if (ret < 0) |
1070 | return ret; | |
1071 | if (ret > 0) | |
1072 | return tcp_metrics_flush_all(net); | |
3e7013dd CP |
1073 | ret = parse_nl_saddr(info, &saddr); |
1074 | if (ret < 0) | |
1075 | src = false; | |
d23ff701 | 1076 | |
3e5da62d | 1077 | hash ^= net_hash_mix(net); |
d23ff701 JA |
1078 | hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log); |
1079 | hb = net->ipv4.tcp_metrics_hash + hash; | |
1080 | pp = &hb->chain; | |
1081 | spin_lock_bh(&tcp_metrics_lock); | |
bbf852b9 | 1082 | for (tm = deref_locked_genl(*pp); tm; tm = deref_locked_genl(*pp)) { |
3e7013dd CP |
1083 | if (addr_same(&tm->tcpm_daddr, &daddr) && |
1084 | (!src || addr_same(&tm->tcpm_saddr, &saddr))) { | |
d23ff701 | 1085 | *pp = tm->tcpm_next; |
00ca9c5b CP |
1086 | kfree_rcu(tm, rcu_head); |
1087 | found = true; | |
bbf852b9 CP |
1088 | } else { |
1089 | pp = &tm->tcpm_next; | |
d23ff701 JA |
1090 | } |
1091 | } | |
1092 | spin_unlock_bh(&tcp_metrics_lock); | |
00ca9c5b | 1093 | if (!found) |
d23ff701 | 1094 | return -ESRCH; |
d23ff701 JA |
1095 | return 0; |
1096 | } | |
1097 | ||
4534de83 | 1098 | static const struct genl_ops tcp_metrics_nl_ops[] = { |
d23ff701 JA |
1099 | { |
1100 | .cmd = TCP_METRICS_CMD_GET, | |
1101 | .doit = tcp_metrics_nl_cmd_get, | |
1102 | .dumpit = tcp_metrics_nl_dump, | |
1103 | .policy = tcp_metrics_nl_policy, | |
d23ff701 JA |
1104 | }, |
1105 | { | |
1106 | .cmd = TCP_METRICS_CMD_DEL, | |
1107 | .doit = tcp_metrics_nl_cmd_del, | |
1108 | .policy = tcp_metrics_nl_policy, | |
1109 | .flags = GENL_ADMIN_PERM, | |
1110 | }, | |
1111 | }; | |
1112 | ||
5815d5e7 | 1113 | static unsigned int tcpmhash_entries; |
51c5d0c4 DM |
1114 | static int __init set_tcpmhash_entries(char *str) |
1115 | { | |
1116 | ssize_t ret; | |
1117 | ||
1118 | if (!str) | |
1119 | return 0; | |
1120 | ||
5815d5e7 | 1121 | ret = kstrtouint(str, 0, &tcpmhash_entries); |
51c5d0c4 DM |
1122 | if (ret) |
1123 | return 0; | |
1124 | ||
1125 | return 1; | |
1126 | } | |
1127 | __setup("tcpmhash_entries=", set_tcpmhash_entries); | |
1128 | ||
1129 | static int __net_init tcp_net_metrics_init(struct net *net) | |
1130 | { | |
5815d5e7 ED |
1131 | size_t size; |
1132 | unsigned int slots; | |
51c5d0c4 DM |
1133 | |
1134 | slots = tcpmhash_entries; | |
1135 | if (!slots) { | |
1136 | if (totalram_pages >= 128 * 1024) | |
1137 | slots = 16 * 1024; | |
1138 | else | |
1139 | slots = 8 * 1024; | |
1140 | } | |
1141 | ||
5815d5e7 ED |
1142 | net->ipv4.tcp_metrics_hash_log = order_base_2(slots); |
1143 | size = sizeof(struct tcpm_hash_bucket) << net->ipv4.tcp_metrics_hash_log; | |
51c5d0c4 | 1144 | |
976a702a ED |
1145 | net->ipv4.tcp_metrics_hash = kzalloc(size, GFP_KERNEL | __GFP_NOWARN); |
1146 | if (!net->ipv4.tcp_metrics_hash) | |
1147 | net->ipv4.tcp_metrics_hash = vzalloc(size); | |
1148 | ||
51c5d0c4 DM |
1149 | if (!net->ipv4.tcp_metrics_hash) |
1150 | return -ENOMEM; | |
1151 | ||
51c5d0c4 DM |
1152 | return 0; |
1153 | } | |
1154 | ||
1155 | static void __net_exit tcp_net_metrics_exit(struct net *net) | |
1156 | { | |
36471012 ED |
1157 | unsigned int i; |
1158 | ||
1159 | for (i = 0; i < (1U << net->ipv4.tcp_metrics_hash_log) ; i++) { | |
1160 | struct tcp_metrics_block *tm, *next; | |
1161 | ||
1162 | tm = rcu_dereference_protected(net->ipv4.tcp_metrics_hash[i].chain, 1); | |
1163 | while (tm) { | |
1164 | next = rcu_dereference_protected(tm->tcpm_next, 1); | |
1165 | kfree(tm); | |
1166 | tm = next; | |
1167 | } | |
1168 | } | |
4cb28970 | 1169 | kvfree(net->ipv4.tcp_metrics_hash); |
51c5d0c4 DM |
1170 | } |
1171 | ||
1172 | static __net_initdata struct pernet_operations tcp_net_metrics_ops = { | |
1173 | .init = tcp_net_metrics_init, | |
1174 | .exit = tcp_net_metrics_exit, | |
1175 | }; | |
1176 | ||
1177 | void __init tcp_metrics_init(void) | |
1178 | { | |
d23ff701 JA |
1179 | int ret; |
1180 | ||
1181 | ret = register_pernet_subsys(&tcp_net_metrics_ops); | |
1182 | if (ret < 0) | |
6493517e EB |
1183 | panic("Could not allocate the tcp_metrics hash table\n"); |
1184 | ||
d23ff701 | 1185 | ret = genl_register_family_with_ops(&tcp_metrics_nl_family, |
c53ed742 | 1186 | tcp_metrics_nl_ops); |
d23ff701 | 1187 | if (ret < 0) |
6493517e | 1188 | panic("Could not register tcp_metrics generic netlink\n"); |
51c5d0c4 | 1189 | } |