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; | |
2646c831 DL |
31 | u16 syn_loss:10, /* Recurring Fast Open SYN losses */ |
32 | try_exp:2; /* Request w/ exp. option (once) */ | |
aab48743 | 33 | unsigned long last_syn_loss; /* Last Fast Open SYN loss */ |
1fe4c481 YC |
34 | struct tcp_fastopen_cookie cookie; |
35 | }; | |
36 | ||
740b0f18 ED |
37 | /* TCP_METRIC_MAX includes 2 extra fields for userspace compatibility |
38 | * Kernel only stores RTT and RTTVAR in usec resolution | |
39 | */ | |
40 | #define TCP_METRIC_MAX_KERNEL (TCP_METRIC_MAX - 2) | |
41 | ||
51c5d0c4 DM |
42 | struct tcp_metrics_block { |
43 | struct tcp_metrics_block __rcu *tcpm_next; | |
849e8a0c | 44 | possible_net_t tcpm_net; |
a5443028 | 45 | struct inetpeer_addr tcpm_saddr; |
324fd55a | 46 | struct inetpeer_addr tcpm_daddr; |
51c5d0c4 | 47 | unsigned long tcpm_stamp; |
81166dd6 DM |
48 | u32 tcpm_ts; |
49 | u32 tcpm_ts_stamp; | |
51c5d0c4 | 50 | u32 tcpm_lock; |
740b0f18 | 51 | u32 tcpm_vals[TCP_METRIC_MAX_KERNEL + 1]; |
1fe4c481 | 52 | struct tcp_fastopen_metrics tcpm_fastopen; |
d23ff701 JA |
53 | |
54 | struct rcu_head rcu_head; | |
51c5d0c4 DM |
55 | }; |
56 | ||
849e8a0c EB |
57 | static inline struct net *tm_net(struct tcp_metrics_block *tm) |
58 | { | |
59 | return read_pnet(&tm->tcpm_net); | |
60 | } | |
61 | ||
51c5d0c4 DM |
62 | static bool tcp_metric_locked(struct tcp_metrics_block *tm, |
63 | enum tcp_metric_index idx) | |
64 | { | |
65 | return tm->tcpm_lock & (1 << idx); | |
66 | } | |
67 | ||
68 | static u32 tcp_metric_get(struct tcp_metrics_block *tm, | |
69 | enum tcp_metric_index idx) | |
70 | { | |
71 | return tm->tcpm_vals[idx]; | |
72 | } | |
73 | ||
51c5d0c4 DM |
74 | static void tcp_metric_set(struct tcp_metrics_block *tm, |
75 | enum tcp_metric_index idx, | |
76 | u32 val) | |
77 | { | |
78 | tm->tcpm_vals[idx] = val; | |
79 | } | |
80 | ||
51c5d0c4 DM |
81 | static bool addr_same(const struct inetpeer_addr *a, |
82 | const struct inetpeer_addr *b) | |
83 | { | |
51c5d0c4 DM |
84 | if (a->family != b->family) |
85 | return false; | |
86 | if (a->family == AF_INET) | |
87 | return a->addr.a4 == b->addr.a4; | |
8f55db48 | 88 | return ipv6_addr_equal(&a->addr.in6, &b->addr.in6); |
51c5d0c4 DM |
89 | } |
90 | ||
91 | struct tcpm_hash_bucket { | |
92 | struct tcp_metrics_block __rcu *chain; | |
93 | }; | |
94 | ||
098a697b EB |
95 | static struct tcpm_hash_bucket *tcp_metrics_hash __read_mostly; |
96 | static unsigned int tcp_metrics_hash_log __read_mostly; | |
97 | ||
51c5d0c4 DM |
98 | static DEFINE_SPINLOCK(tcp_metrics_lock); |
99 | ||
740b0f18 ED |
100 | static void tcpm_suck_dst(struct tcp_metrics_block *tm, |
101 | const struct dst_entry *dst, | |
efeaa555 | 102 | bool fastopen_clear) |
51c5d0c4 | 103 | { |
740b0f18 | 104 | u32 msval; |
51c5d0c4 DM |
105 | u32 val; |
106 | ||
9a0a9502 JA |
107 | tm->tcpm_stamp = jiffies; |
108 | ||
51c5d0c4 DM |
109 | val = 0; |
110 | if (dst_metric_locked(dst, RTAX_RTT)) | |
111 | val |= 1 << TCP_METRIC_RTT; | |
112 | if (dst_metric_locked(dst, RTAX_RTTVAR)) | |
113 | val |= 1 << TCP_METRIC_RTTVAR; | |
114 | if (dst_metric_locked(dst, RTAX_SSTHRESH)) | |
115 | val |= 1 << TCP_METRIC_SSTHRESH; | |
116 | if (dst_metric_locked(dst, RTAX_CWND)) | |
117 | val |= 1 << TCP_METRIC_CWND; | |
118 | if (dst_metric_locked(dst, RTAX_REORDERING)) | |
119 | val |= 1 << TCP_METRIC_REORDERING; | |
120 | tm->tcpm_lock = val; | |
121 | ||
740b0f18 ED |
122 | msval = dst_metric_raw(dst, RTAX_RTT); |
123 | tm->tcpm_vals[TCP_METRIC_RTT] = msval * USEC_PER_MSEC; | |
124 | ||
125 | msval = dst_metric_raw(dst, RTAX_RTTVAR); | |
126 | tm->tcpm_vals[TCP_METRIC_RTTVAR] = msval * USEC_PER_MSEC; | |
51c5d0c4 DM |
127 | tm->tcpm_vals[TCP_METRIC_SSTHRESH] = dst_metric_raw(dst, RTAX_SSTHRESH); |
128 | tm->tcpm_vals[TCP_METRIC_CWND] = dst_metric_raw(dst, RTAX_CWND); | |
129 | tm->tcpm_vals[TCP_METRIC_REORDERING] = dst_metric_raw(dst, RTAX_REORDERING); | |
81166dd6 DM |
130 | tm->tcpm_ts = 0; |
131 | tm->tcpm_ts_stamp = 0; | |
efeaa555 ED |
132 | if (fastopen_clear) { |
133 | tm->tcpm_fastopen.mss = 0; | |
134 | tm->tcpm_fastopen.syn_loss = 0; | |
2646c831 DL |
135 | tm->tcpm_fastopen.try_exp = 0; |
136 | tm->tcpm_fastopen.cookie.exp = false; | |
efeaa555 ED |
137 | tm->tcpm_fastopen.cookie.len = 0; |
138 | } | |
51c5d0c4 DM |
139 | } |
140 | ||
77f99ad1 CP |
141 | #define TCP_METRICS_TIMEOUT (60 * 60 * HZ) |
142 | ||
143 | static void tcpm_check_stamp(struct tcp_metrics_block *tm, struct dst_entry *dst) | |
144 | { | |
145 | if (tm && unlikely(time_after(jiffies, tm->tcpm_stamp + TCP_METRICS_TIMEOUT))) | |
146 | tcpm_suck_dst(tm, dst, false); | |
147 | } | |
148 | ||
149 | #define TCP_METRICS_RECLAIM_DEPTH 5 | |
150 | #define TCP_METRICS_RECLAIM_PTR (struct tcp_metrics_block *) 0x1UL | |
151 | ||
9f1ab186 ED |
152 | #define deref_locked(p) \ |
153 | rcu_dereference_protected(p, lockdep_is_held(&tcp_metrics_lock)) | |
154 | ||
51c5d0c4 | 155 | static struct tcp_metrics_block *tcpm_new(struct dst_entry *dst, |
a5443028 | 156 | struct inetpeer_addr *saddr, |
324fd55a | 157 | struct inetpeer_addr *daddr, |
77f99ad1 | 158 | unsigned int hash) |
51c5d0c4 DM |
159 | { |
160 | struct tcp_metrics_block *tm; | |
161 | struct net *net; | |
77f99ad1 | 162 | bool reclaim = false; |
51c5d0c4 DM |
163 | |
164 | spin_lock_bh(&tcp_metrics_lock); | |
165 | net = dev_net(dst->dev); | |
77f99ad1 CP |
166 | |
167 | /* While waiting for the spin-lock the cache might have been populated | |
168 | * with this entry and so we have to check again. | |
169 | */ | |
41804420 | 170 | tm = __tcp_get_metrics(saddr, daddr, net, hash); |
77f99ad1 CP |
171 | if (tm == TCP_METRICS_RECLAIM_PTR) { |
172 | reclaim = true; | |
173 | tm = NULL; | |
174 | } | |
175 | if (tm) { | |
176 | tcpm_check_stamp(tm, dst); | |
177 | goto out_unlock; | |
178 | } | |
179 | ||
51c5d0c4 DM |
180 | if (unlikely(reclaim)) { |
181 | struct tcp_metrics_block *oldest; | |
182 | ||
9f1ab186 ED |
183 | oldest = deref_locked(tcp_metrics_hash[hash].chain); |
184 | for (tm = deref_locked(oldest->tcpm_next); tm; | |
185 | tm = deref_locked(tm->tcpm_next)) { | |
51c5d0c4 DM |
186 | if (time_before(tm->tcpm_stamp, oldest->tcpm_stamp)) |
187 | oldest = tm; | |
188 | } | |
189 | tm = oldest; | |
190 | } else { | |
191 | tm = kmalloc(sizeof(*tm), GFP_ATOMIC); | |
192 | if (!tm) | |
193 | goto out_unlock; | |
194 | } | |
849e8a0c | 195 | write_pnet(&tm->tcpm_net, net); |
a5443028 | 196 | tm->tcpm_saddr = *saddr; |
324fd55a | 197 | tm->tcpm_daddr = *daddr; |
51c5d0c4 | 198 | |
efeaa555 | 199 | tcpm_suck_dst(tm, dst, true); |
51c5d0c4 DM |
200 | |
201 | if (likely(!reclaim)) { | |
098a697b EB |
202 | tm->tcpm_next = tcp_metrics_hash[hash].chain; |
203 | rcu_assign_pointer(tcp_metrics_hash[hash].chain, tm); | |
51c5d0c4 DM |
204 | } |
205 | ||
206 | out_unlock: | |
207 | spin_unlock_bh(&tcp_metrics_lock); | |
208 | return tm; | |
209 | } | |
210 | ||
51c5d0c4 DM |
211 | static struct tcp_metrics_block *tcp_get_encode(struct tcp_metrics_block *tm, int depth) |
212 | { | |
213 | if (tm) | |
214 | return tm; | |
215 | if (depth > TCP_METRICS_RECLAIM_DEPTH) | |
216 | return TCP_METRICS_RECLAIM_PTR; | |
217 | return NULL; | |
218 | } | |
219 | ||
a5443028 CP |
220 | static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr, |
221 | const struct inetpeer_addr *daddr, | |
51c5d0c4 DM |
222 | struct net *net, unsigned int hash) |
223 | { | |
224 | struct tcp_metrics_block *tm; | |
225 | int depth = 0; | |
226 | ||
098a697b | 227 | for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm; |
51c5d0c4 | 228 | tm = rcu_dereference(tm->tcpm_next)) { |
a5443028 | 229 | if (addr_same(&tm->tcpm_saddr, saddr) && |
849e8a0c EB |
230 | addr_same(&tm->tcpm_daddr, daddr) && |
231 | net_eq(tm_net(tm), net)) | |
51c5d0c4 DM |
232 | break; |
233 | depth++; | |
234 | } | |
235 | return tcp_get_encode(tm, depth); | |
236 | } | |
237 | ||
238 | static struct tcp_metrics_block *__tcp_get_metrics_req(struct request_sock *req, | |
239 | struct dst_entry *dst) | |
240 | { | |
241 | struct tcp_metrics_block *tm; | |
a5443028 | 242 | struct inetpeer_addr saddr, daddr; |
51c5d0c4 DM |
243 | unsigned int hash; |
244 | struct net *net; | |
245 | ||
a5443028 | 246 | saddr.family = req->rsk_ops->family; |
324fd55a CP |
247 | daddr.family = req->rsk_ops->family; |
248 | switch (daddr.family) { | |
51c5d0c4 | 249 | case AF_INET: |
a5443028 | 250 | saddr.addr.a4 = inet_rsk(req)->ir_loc_addr; |
324fd55a CP |
251 | daddr.addr.a4 = inet_rsk(req)->ir_rmt_addr; |
252 | hash = (__force unsigned int) daddr.addr.a4; | |
51c5d0c4 | 253 | break; |
634fb979 | 254 | #if IS_ENABLED(CONFIG_IPV6) |
51c5d0c4 | 255 | case AF_INET6: |
8f55db48 JB |
256 | saddr.addr.in6 = inet_rsk(req)->ir_v6_loc_addr; |
257 | daddr.addr.in6 = inet_rsk(req)->ir_v6_rmt_addr; | |
634fb979 | 258 | hash = ipv6_addr_hash(&inet_rsk(req)->ir_v6_rmt_addr); |
51c5d0c4 | 259 | break; |
634fb979 | 260 | #endif |
51c5d0c4 DM |
261 | default: |
262 | return NULL; | |
263 | } | |
264 | ||
51c5d0c4 | 265 | net = dev_net(dst->dev); |
3e5da62d | 266 | hash ^= net_hash_mix(net); |
098a697b | 267 | hash = hash_32(hash, tcp_metrics_hash_log); |
51c5d0c4 | 268 | |
098a697b | 269 | for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm; |
51c5d0c4 | 270 | tm = rcu_dereference(tm->tcpm_next)) { |
a5443028 | 271 | if (addr_same(&tm->tcpm_saddr, &saddr) && |
849e8a0c EB |
272 | addr_same(&tm->tcpm_daddr, &daddr) && |
273 | net_eq(tm_net(tm), net)) | |
51c5d0c4 DM |
274 | break; |
275 | } | |
276 | tcpm_check_stamp(tm, dst); | |
277 | return tm; | |
278 | } | |
279 | ||
81166dd6 DM |
280 | static struct tcp_metrics_block *__tcp_get_metrics_tw(struct inet_timewait_sock *tw) |
281 | { | |
81166dd6 | 282 | struct tcp_metrics_block *tm; |
a5443028 | 283 | struct inetpeer_addr saddr, daddr; |
81166dd6 DM |
284 | unsigned int hash; |
285 | struct net *net; | |
286 | ||
3ad88cf7 CP |
287 | if (tw->tw_family == AF_INET) { |
288 | saddr.family = AF_INET; | |
a5443028 | 289 | saddr.addr.a4 = tw->tw_rcv_saddr; |
3ad88cf7 | 290 | daddr.family = AF_INET; |
324fd55a CP |
291 | daddr.addr.a4 = tw->tw_daddr; |
292 | hash = (__force unsigned int) daddr.addr.a4; | |
3ad88cf7 | 293 | } |
c2bb06db | 294 | #if IS_ENABLED(CONFIG_IPV6) |
3ad88cf7 CP |
295 | else if (tw->tw_family == AF_INET6) { |
296 | if (ipv6_addr_v4mapped(&tw->tw_v6_daddr)) { | |
297 | saddr.family = AF_INET; | |
298 | saddr.addr.a4 = tw->tw_rcv_saddr; | |
299 | daddr.family = AF_INET; | |
300 | daddr.addr.a4 = tw->tw_daddr; | |
301 | hash = (__force unsigned int) daddr.addr.a4; | |
302 | } else { | |
303 | saddr.family = AF_INET6; | |
8f55db48 | 304 | saddr.addr.in6 = tw->tw_v6_rcv_saddr; |
3ad88cf7 | 305 | daddr.family = AF_INET6; |
8f55db48 | 306 | daddr.addr.in6 = tw->tw_v6_daddr; |
3ad88cf7 CP |
307 | hash = ipv6_addr_hash(&tw->tw_v6_daddr); |
308 | } | |
309 | } | |
c2bb06db | 310 | #endif |
3ad88cf7 | 311 | else |
81166dd6 | 312 | return NULL; |
81166dd6 | 313 | |
81166dd6 | 314 | net = twsk_net(tw); |
3e5da62d | 315 | hash ^= net_hash_mix(net); |
098a697b | 316 | hash = hash_32(hash, tcp_metrics_hash_log); |
81166dd6 | 317 | |
098a697b | 318 | for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm; |
81166dd6 | 319 | tm = rcu_dereference(tm->tcpm_next)) { |
a5443028 | 320 | if (addr_same(&tm->tcpm_saddr, &saddr) && |
849e8a0c EB |
321 | addr_same(&tm->tcpm_daddr, &daddr) && |
322 | net_eq(tm_net(tm), net)) | |
81166dd6 DM |
323 | break; |
324 | } | |
325 | return tm; | |
326 | } | |
327 | ||
51c5d0c4 DM |
328 | static struct tcp_metrics_block *tcp_get_metrics(struct sock *sk, |
329 | struct dst_entry *dst, | |
330 | bool create) | |
331 | { | |
332 | struct tcp_metrics_block *tm; | |
a5443028 | 333 | struct inetpeer_addr saddr, daddr; |
51c5d0c4 DM |
334 | unsigned int hash; |
335 | struct net *net; | |
51c5d0c4 | 336 | |
3ad88cf7 CP |
337 | if (sk->sk_family == AF_INET) { |
338 | saddr.family = AF_INET; | |
a5443028 | 339 | saddr.addr.a4 = inet_sk(sk)->inet_saddr; |
3ad88cf7 | 340 | daddr.family = AF_INET; |
324fd55a CP |
341 | daddr.addr.a4 = inet_sk(sk)->inet_daddr; |
342 | hash = (__force unsigned int) daddr.addr.a4; | |
3ad88cf7 | 343 | } |
c2bb06db | 344 | #if IS_ENABLED(CONFIG_IPV6) |
3ad88cf7 CP |
345 | else if (sk->sk_family == AF_INET6) { |
346 | if (ipv6_addr_v4mapped(&sk->sk_v6_daddr)) { | |
347 | saddr.family = AF_INET; | |
348 | saddr.addr.a4 = inet_sk(sk)->inet_saddr; | |
349 | daddr.family = AF_INET; | |
350 | daddr.addr.a4 = inet_sk(sk)->inet_daddr; | |
351 | hash = (__force unsigned int) daddr.addr.a4; | |
352 | } else { | |
353 | saddr.family = AF_INET6; | |
8f55db48 | 354 | saddr.addr.in6 = sk->sk_v6_rcv_saddr; |
3ad88cf7 | 355 | daddr.family = AF_INET6; |
8f55db48 | 356 | daddr.addr.in6 = sk->sk_v6_daddr; |
3ad88cf7 CP |
357 | hash = ipv6_addr_hash(&sk->sk_v6_daddr); |
358 | } | |
359 | } | |
c2bb06db | 360 | #endif |
3ad88cf7 | 361 | else |
51c5d0c4 | 362 | return NULL; |
51c5d0c4 | 363 | |
51c5d0c4 | 364 | net = dev_net(dst->dev); |
3e5da62d | 365 | hash ^= net_hash_mix(net); |
098a697b | 366 | hash = hash_32(hash, tcp_metrics_hash_log); |
51c5d0c4 | 367 | |
a5443028 | 368 | tm = __tcp_get_metrics(&saddr, &daddr, net, hash); |
77f99ad1 | 369 | if (tm == TCP_METRICS_RECLAIM_PTR) |
51c5d0c4 | 370 | tm = NULL; |
51c5d0c4 | 371 | if (!tm && create) |
41804420 | 372 | tm = tcpm_new(dst, &saddr, &daddr, hash); |
51c5d0c4 DM |
373 | else |
374 | tcpm_check_stamp(tm, dst); | |
375 | ||
376 | return tm; | |
377 | } | |
378 | ||
4aabd8ef DM |
379 | /* Save metrics learned by this TCP session. This function is called |
380 | * only, when TCP finishes successfully i.e. when it enters TIME-WAIT | |
381 | * or goes from LAST-ACK to CLOSE. | |
382 | */ | |
383 | void tcp_update_metrics(struct sock *sk) | |
384 | { | |
51c5d0c4 | 385 | const struct inet_connection_sock *icsk = inet_csk(sk); |
4aabd8ef | 386 | struct dst_entry *dst = __sk_dst_get(sk); |
51c5d0c4 DM |
387 | struct tcp_sock *tp = tcp_sk(sk); |
388 | struct tcp_metrics_block *tm; | |
389 | unsigned long rtt; | |
390 | u32 val; | |
391 | int m; | |
4aabd8ef | 392 | |
51c5d0c4 | 393 | if (sysctl_tcp_nometrics_save || !dst) |
4aabd8ef DM |
394 | return; |
395 | ||
51c5d0c4 | 396 | if (dst->flags & DST_HOST) |
4aabd8ef DM |
397 | dst_confirm(dst); |
398 | ||
51c5d0c4 | 399 | rcu_read_lock(); |
740b0f18 | 400 | if (icsk->icsk_backoff || !tp->srtt_us) { |
51c5d0c4 DM |
401 | /* This session failed to estimate rtt. Why? |
402 | * Probably, no packets returned in time. Reset our | |
403 | * results. | |
404 | */ | |
405 | tm = tcp_get_metrics(sk, dst, false); | |
406 | if (tm && !tcp_metric_locked(tm, TCP_METRIC_RTT)) | |
407 | tcp_metric_set(tm, TCP_METRIC_RTT, 0); | |
408 | goto out_unlock; | |
409 | } else | |
410 | tm = tcp_get_metrics(sk, dst, true); | |
4aabd8ef | 411 | |
51c5d0c4 DM |
412 | if (!tm) |
413 | goto out_unlock; | |
4aabd8ef | 414 | |
740b0f18 ED |
415 | rtt = tcp_metric_get(tm, TCP_METRIC_RTT); |
416 | m = rtt - tp->srtt_us; | |
4aabd8ef | 417 | |
51c5d0c4 DM |
418 | /* If newly calculated rtt larger than stored one, store new |
419 | * one. Otherwise, use EWMA. Remember, rtt overestimation is | |
420 | * always better than underestimation. | |
421 | */ | |
422 | if (!tcp_metric_locked(tm, TCP_METRIC_RTT)) { | |
423 | if (m <= 0) | |
740b0f18 | 424 | rtt = tp->srtt_us; |
51c5d0c4 DM |
425 | else |
426 | rtt -= (m >> 3); | |
740b0f18 | 427 | tcp_metric_set(tm, TCP_METRIC_RTT, rtt); |
51c5d0c4 | 428 | } |
4aabd8ef | 429 | |
51c5d0c4 DM |
430 | if (!tcp_metric_locked(tm, TCP_METRIC_RTTVAR)) { |
431 | unsigned long var; | |
4aabd8ef | 432 | |
51c5d0c4 DM |
433 | if (m < 0) |
434 | m = -m; | |
4aabd8ef | 435 | |
51c5d0c4 DM |
436 | /* Scale deviation to rttvar fixed point */ |
437 | m >>= 1; | |
740b0f18 ED |
438 | if (m < tp->mdev_us) |
439 | m = tp->mdev_us; | |
4aabd8ef | 440 | |
740b0f18 | 441 | var = tcp_metric_get(tm, TCP_METRIC_RTTVAR); |
51c5d0c4 DM |
442 | if (m >= var) |
443 | var = m; | |
444 | else | |
445 | var -= (var - m) >> 2; | |
4aabd8ef | 446 | |
740b0f18 | 447 | tcp_metric_set(tm, TCP_METRIC_RTTVAR, var); |
51c5d0c4 DM |
448 | } |
449 | ||
450 | if (tcp_in_initial_slowstart(tp)) { | |
451 | /* Slow start still did not finish. */ | |
452 | if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) { | |
453 | val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH); | |
454 | if (val && (tp->snd_cwnd >> 1) > val) | |
455 | tcp_metric_set(tm, TCP_METRIC_SSTHRESH, | |
456 | tp->snd_cwnd >> 1); | |
457 | } | |
458 | if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) { | |
459 | val = tcp_metric_get(tm, TCP_METRIC_CWND); | |
460 | if (tp->snd_cwnd > val) | |
461 | tcp_metric_set(tm, TCP_METRIC_CWND, | |
462 | tp->snd_cwnd); | |
463 | } | |
464 | } else if (tp->snd_cwnd > tp->snd_ssthresh && | |
465 | icsk->icsk_ca_state == TCP_CA_Open) { | |
466 | /* Cong. avoidance phase, cwnd is reliable. */ | |
467 | if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) | |
468 | tcp_metric_set(tm, TCP_METRIC_SSTHRESH, | |
469 | max(tp->snd_cwnd >> 1, tp->snd_ssthresh)); | |
470 | if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) { | |
471 | val = tcp_metric_get(tm, TCP_METRIC_CWND); | |
2100844c | 472 | tcp_metric_set(tm, TCP_METRIC_CWND, (val + tp->snd_cwnd) >> 1); |
51c5d0c4 DM |
473 | } |
474 | } else { | |
475 | /* Else slow start did not finish, cwnd is non-sense, | |
476 | * ssthresh may be also invalid. | |
477 | */ | |
478 | if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) { | |
479 | val = tcp_metric_get(tm, TCP_METRIC_CWND); | |
480 | tcp_metric_set(tm, TCP_METRIC_CWND, | |
481 | (val + tp->snd_ssthresh) >> 1); | |
482 | } | |
483 | if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) { | |
484 | val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH); | |
485 | if (val && tp->snd_ssthresh > val) | |
486 | tcp_metric_set(tm, TCP_METRIC_SSTHRESH, | |
487 | tp->snd_ssthresh); | |
488 | } | |
489 | if (!tcp_metric_locked(tm, TCP_METRIC_REORDERING)) { | |
490 | val = tcp_metric_get(tm, TCP_METRIC_REORDERING); | |
491 | if (val < tp->reordering && | |
4aabd8ef | 492 | tp->reordering != sysctl_tcp_reordering) |
51c5d0c4 DM |
493 | tcp_metric_set(tm, TCP_METRIC_REORDERING, |
494 | tp->reordering); | |
4aabd8ef DM |
495 | } |
496 | } | |
51c5d0c4 DM |
497 | tm->tcpm_stamp = jiffies; |
498 | out_unlock: | |
499 | rcu_read_unlock(); | |
4aabd8ef DM |
500 | } |
501 | ||
502 | /* Initialize metrics on socket. */ | |
503 | ||
504 | void tcp_init_metrics(struct sock *sk) | |
505 | { | |
4aabd8ef | 506 | struct dst_entry *dst = __sk_dst_get(sk); |
51c5d0c4 DM |
507 | struct tcp_sock *tp = tcp_sk(sk); |
508 | struct tcp_metrics_block *tm; | |
1b7fdd2a | 509 | u32 val, crtt = 0; /* cached RTT scaled by 8 */ |
4aabd8ef | 510 | |
51456b29 | 511 | if (!dst) |
4aabd8ef DM |
512 | goto reset; |
513 | ||
514 | dst_confirm(dst); | |
515 | ||
51c5d0c4 DM |
516 | rcu_read_lock(); |
517 | tm = tcp_get_metrics(sk, dst, true); | |
518 | if (!tm) { | |
519 | rcu_read_unlock(); | |
520 | goto reset; | |
521 | } | |
522 | ||
523 | if (tcp_metric_locked(tm, TCP_METRIC_CWND)) | |
524 | tp->snd_cwnd_clamp = tcp_metric_get(tm, TCP_METRIC_CWND); | |
525 | ||
526 | val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH); | |
527 | if (val) { | |
528 | tp->snd_ssthresh = val; | |
4aabd8ef DM |
529 | if (tp->snd_ssthresh > tp->snd_cwnd_clamp) |
530 | tp->snd_ssthresh = tp->snd_cwnd_clamp; | |
531 | } else { | |
532 | /* ssthresh may have been reduced unnecessarily during. | |
533 | * 3WHS. Restore it back to its initial default. | |
534 | */ | |
535 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; | |
536 | } | |
51c5d0c4 DM |
537 | val = tcp_metric_get(tm, TCP_METRIC_REORDERING); |
538 | if (val && tp->reordering != val) { | |
4aabd8ef DM |
539 | tcp_disable_fack(tp); |
540 | tcp_disable_early_retrans(tp); | |
51c5d0c4 | 541 | tp->reordering = val; |
4aabd8ef DM |
542 | } |
543 | ||
740b0f18 | 544 | crtt = tcp_metric_get(tm, TCP_METRIC_RTT); |
51c5d0c4 | 545 | rcu_read_unlock(); |
4aabd8ef | 546 | reset: |
52f20e65 YC |
547 | /* The initial RTT measurement from the SYN/SYN-ACK is not ideal |
548 | * to seed the RTO for later data packets because SYN packets are | |
549 | * small. Use the per-dst cached values to seed the RTO but keep | |
550 | * the RTT estimator variables intact (e.g., srtt, mdev, rttvar). | |
551 | * Later the RTO will be updated immediately upon obtaining the first | |
552 | * data RTT sample (tcp_rtt_estimator()). Hence the cached RTT only | |
553 | * influences the first RTO but not later RTT estimation. | |
554 | * | |
555 | * But if RTT is not available from the SYN (due to retransmits or | |
556 | * syn cookies) or the cache, force a conservative 3secs timeout. | |
557 | * | |
558 | * A bit of theory. RTT is time passed after "normal" sized packet | |
559 | * is sent until it is ACKed. In normal circumstances sending small | |
560 | * packets force peer to delay ACKs and calculation is correct too. | |
561 | * The algorithm is adaptive and, provided we follow specs, it | |
562 | * NEVER underestimate RTT. BUT! If peer tries to make some clever | |
563 | * tricks sort of "quick acks" for time long enough to decrease RTT | |
564 | * to low value, and then abruptly stops to do it and starts to delay | |
565 | * ACKs, wait for troubles. | |
566 | */ | |
740b0f18 | 567 | if (crtt > tp->srtt_us) { |
269aa759 | 568 | /* Set RTO like tcp_rtt_estimator(), but from cached RTT. */ |
740b0f18 | 569 | crtt /= 8 * USEC_PER_MSEC; |
269aa759 | 570 | inet_csk(sk)->icsk_rto = crtt + max(2 * crtt, tcp_rto_min(sk)); |
740b0f18 | 571 | } else if (tp->srtt_us == 0) { |
4aabd8ef DM |
572 | /* RFC6298: 5.7 We've failed to get a valid RTT sample from |
573 | * 3WHS. This is most likely due to retransmission, | |
574 | * including spurious one. Reset the RTO back to 3secs | |
575 | * from the more aggressive 1sec to avoid more spurious | |
576 | * retransmission. | |
577 | */ | |
740b0f18 ED |
578 | tp->rttvar_us = jiffies_to_usecs(TCP_TIMEOUT_FALLBACK); |
579 | tp->mdev_us = tp->mdev_max_us = tp->rttvar_us; | |
580 | ||
4aabd8ef DM |
581 | inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK; |
582 | } | |
583 | /* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been | |
584 | * retransmitted. In light of RFC6298 more aggressive 1sec | |
585 | * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK | |
586 | * retransmission has occurred. | |
587 | */ | |
588 | if (tp->total_retrans > 1) | |
589 | tp->snd_cwnd = 1; | |
590 | else | |
591 | tp->snd_cwnd = tcp_init_cwnd(tp, dst); | |
592 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
593 | } | |
ab92bb2f | 594 | |
a26552af HFS |
595 | bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, |
596 | bool paws_check, bool timestamps) | |
ab92bb2f | 597 | { |
51c5d0c4 DM |
598 | struct tcp_metrics_block *tm; |
599 | bool ret; | |
600 | ||
ab92bb2f DM |
601 | if (!dst) |
602 | return false; | |
51c5d0c4 DM |
603 | |
604 | rcu_read_lock(); | |
605 | tm = __tcp_get_metrics_req(req, dst); | |
81166dd6 DM |
606 | if (paws_check) { |
607 | if (tm && | |
608 | (u32)get_seconds() - tm->tcpm_ts_stamp < TCP_PAWS_MSL && | |
a26552af HFS |
609 | ((s32)(tm->tcpm_ts - req->ts_recent) > TCP_PAWS_WINDOW || |
610 | !timestamps)) | |
81166dd6 DM |
611 | ret = false; |
612 | else | |
613 | ret = true; | |
614 | } else { | |
615 | if (tm && tcp_metric_get(tm, TCP_METRIC_RTT) && tm->tcpm_ts_stamp) | |
616 | ret = true; | |
617 | else | |
618 | ret = false; | |
619 | } | |
51c5d0c4 DM |
620 | rcu_read_unlock(); |
621 | ||
622 | return ret; | |
ab92bb2f DM |
623 | } |
624 | EXPORT_SYMBOL_GPL(tcp_peer_is_proven); | |
51c5d0c4 | 625 | |
81166dd6 DM |
626 | void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst) |
627 | { | |
628 | struct tcp_metrics_block *tm; | |
629 | ||
630 | rcu_read_lock(); | |
631 | tm = tcp_get_metrics(sk, dst, true); | |
632 | if (tm) { | |
633 | struct tcp_sock *tp = tcp_sk(sk); | |
634 | ||
635 | if ((u32)get_seconds() - tm->tcpm_ts_stamp <= TCP_PAWS_MSL) { | |
636 | tp->rx_opt.ts_recent_stamp = tm->tcpm_ts_stamp; | |
637 | tp->rx_opt.ts_recent = tm->tcpm_ts; | |
638 | } | |
639 | } | |
640 | rcu_read_unlock(); | |
641 | } | |
642 | EXPORT_SYMBOL_GPL(tcp_fetch_timewait_stamp); | |
643 | ||
644 | /* VJ's idea. Save last timestamp seen from this destination and hold | |
645 | * it at least for normal timewait interval to use for duplicate | |
646 | * segment detection in subsequent connections, before they enter | |
647 | * synchronized state. | |
648 | */ | |
649 | bool tcp_remember_stamp(struct sock *sk) | |
650 | { | |
651 | struct dst_entry *dst = __sk_dst_get(sk); | |
652 | bool ret = false; | |
653 | ||
654 | if (dst) { | |
655 | struct tcp_metrics_block *tm; | |
656 | ||
657 | rcu_read_lock(); | |
658 | tm = tcp_get_metrics(sk, dst, true); | |
659 | if (tm) { | |
660 | struct tcp_sock *tp = tcp_sk(sk); | |
661 | ||
662 | if ((s32)(tm->tcpm_ts - tp->rx_opt.ts_recent) <= 0 || | |
663 | ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL && | |
664 | tm->tcpm_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) { | |
665 | tm->tcpm_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp; | |
666 | tm->tcpm_ts = tp->rx_opt.ts_recent; | |
667 | } | |
668 | ret = true; | |
669 | } | |
670 | rcu_read_unlock(); | |
671 | } | |
672 | return ret; | |
673 | } | |
674 | ||
675 | bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw) | |
676 | { | |
677 | struct tcp_metrics_block *tm; | |
678 | bool ret = false; | |
679 | ||
680 | rcu_read_lock(); | |
681 | tm = __tcp_get_metrics_tw(tw); | |
9a0a9502 | 682 | if (tm) { |
81166dd6 DM |
683 | const struct tcp_timewait_sock *tcptw; |
684 | struct sock *sk = (struct sock *) tw; | |
685 | ||
686 | tcptw = tcp_twsk(sk); | |
687 | if ((s32)(tm->tcpm_ts - tcptw->tw_ts_recent) <= 0 || | |
688 | ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL && | |
689 | tm->tcpm_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) { | |
690 | tm->tcpm_ts_stamp = (u32)tcptw->tw_ts_recent_stamp; | |
691 | tm->tcpm_ts = tcptw->tw_ts_recent; | |
692 | } | |
693 | ret = true; | |
694 | } | |
695 | rcu_read_unlock(); | |
696 | ||
697 | return ret; | |
698 | } | |
699 | ||
1fe4c481 YC |
700 | static DEFINE_SEQLOCK(fastopen_seqlock); |
701 | ||
702 | void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, | |
aab48743 YC |
703 | struct tcp_fastopen_cookie *cookie, |
704 | int *syn_loss, unsigned long *last_syn_loss) | |
1fe4c481 YC |
705 | { |
706 | struct tcp_metrics_block *tm; | |
707 | ||
708 | rcu_read_lock(); | |
709 | tm = tcp_get_metrics(sk, __sk_dst_get(sk), false); | |
710 | if (tm) { | |
711 | struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen; | |
712 | unsigned int seq; | |
713 | ||
714 | do { | |
715 | seq = read_seqbegin(&fastopen_seqlock); | |
716 | if (tfom->mss) | |
717 | *mss = tfom->mss; | |
718 | *cookie = tfom->cookie; | |
2646c831 DL |
719 | if (cookie->len <= 0 && tfom->try_exp == 1) |
720 | cookie->exp = true; | |
aab48743 YC |
721 | *syn_loss = tfom->syn_loss; |
722 | *last_syn_loss = *syn_loss ? tfom->last_syn_loss : 0; | |
1fe4c481 YC |
723 | } while (read_seqretry(&fastopen_seqlock, seq)); |
724 | } | |
725 | rcu_read_unlock(); | |
726 | } | |
727 | ||
1fe4c481 | 728 | void tcp_fastopen_cache_set(struct sock *sk, u16 mss, |
2646c831 DL |
729 | struct tcp_fastopen_cookie *cookie, bool syn_lost, |
730 | u16 try_exp) | |
1fe4c481 | 731 | { |
dccf76ca | 732 | struct dst_entry *dst = __sk_dst_get(sk); |
1fe4c481 YC |
733 | struct tcp_metrics_block *tm; |
734 | ||
dccf76ca ED |
735 | if (!dst) |
736 | return; | |
1fe4c481 | 737 | rcu_read_lock(); |
dccf76ca | 738 | tm = tcp_get_metrics(sk, dst, true); |
1fe4c481 YC |
739 | if (tm) { |
740 | struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen; | |
741 | ||
742 | write_seqlock_bh(&fastopen_seqlock); | |
c968601d YC |
743 | if (mss) |
744 | tfom->mss = mss; | |
745 | if (cookie && cookie->len > 0) | |
1fe4c481 | 746 | tfom->cookie = *cookie; |
2646c831 DL |
747 | else if (try_exp > tfom->try_exp && |
748 | tfom->cookie.len <= 0 && !tfom->cookie.exp) | |
749 | tfom->try_exp = try_exp; | |
aab48743 YC |
750 | if (syn_lost) { |
751 | ++tfom->syn_loss; | |
752 | tfom->last_syn_loss = jiffies; | |
753 | } else | |
754 | tfom->syn_loss = 0; | |
1fe4c481 YC |
755 | write_sequnlock_bh(&fastopen_seqlock); |
756 | } | |
757 | rcu_read_unlock(); | |
758 | } | |
759 | ||
d23ff701 JA |
760 | static struct genl_family tcp_metrics_nl_family = { |
761 | .id = GENL_ID_GENERATE, | |
762 | .hdrsize = 0, | |
763 | .name = TCP_METRICS_GENL_NAME, | |
764 | .version = TCP_METRICS_GENL_VERSION, | |
765 | .maxattr = TCP_METRICS_ATTR_MAX, | |
766 | .netnsok = true, | |
767 | }; | |
768 | ||
769 | static struct nla_policy tcp_metrics_nl_policy[TCP_METRICS_ATTR_MAX + 1] = { | |
770 | [TCP_METRICS_ATTR_ADDR_IPV4] = { .type = NLA_U32, }, | |
771 | [TCP_METRICS_ATTR_ADDR_IPV6] = { .type = NLA_BINARY, | |
772 | .len = sizeof(struct in6_addr), }, | |
773 | /* Following attributes are not received for GET/DEL, | |
774 | * we keep them for reference | |
775 | */ | |
776 | #if 0 | |
777 | [TCP_METRICS_ATTR_AGE] = { .type = NLA_MSECS, }, | |
778 | [TCP_METRICS_ATTR_TW_TSVAL] = { .type = NLA_U32, }, | |
779 | [TCP_METRICS_ATTR_TW_TS_STAMP] = { .type = NLA_S32, }, | |
780 | [TCP_METRICS_ATTR_VALS] = { .type = NLA_NESTED, }, | |
781 | [TCP_METRICS_ATTR_FOPEN_MSS] = { .type = NLA_U16, }, | |
782 | [TCP_METRICS_ATTR_FOPEN_SYN_DROPS] = { .type = NLA_U16, }, | |
783 | [TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS] = { .type = NLA_MSECS, }, | |
784 | [TCP_METRICS_ATTR_FOPEN_COOKIE] = { .type = NLA_BINARY, | |
785 | .len = TCP_FASTOPEN_COOKIE_MAX, }, | |
786 | #endif | |
787 | }; | |
788 | ||
789 | /* Add attributes, caller cancels its header on failure */ | |
790 | static int tcp_metrics_fill_info(struct sk_buff *msg, | |
791 | struct tcp_metrics_block *tm) | |
792 | { | |
793 | struct nlattr *nest; | |
794 | int i; | |
795 | ||
324fd55a | 796 | switch (tm->tcpm_daddr.family) { |
d23ff701 | 797 | case AF_INET: |
930345ea JB |
798 | if (nla_put_in_addr(msg, TCP_METRICS_ATTR_ADDR_IPV4, |
799 | tm->tcpm_daddr.addr.a4) < 0) | |
d23ff701 | 800 | goto nla_put_failure; |
930345ea JB |
801 | if (nla_put_in_addr(msg, TCP_METRICS_ATTR_SADDR_IPV4, |
802 | tm->tcpm_saddr.addr.a4) < 0) | |
8a59359c | 803 | goto nla_put_failure; |
d23ff701 JA |
804 | break; |
805 | case AF_INET6: | |
930345ea JB |
806 | if (nla_put_in6_addr(msg, TCP_METRICS_ATTR_ADDR_IPV6, |
807 | &tm->tcpm_daddr.addr.in6) < 0) | |
d23ff701 | 808 | goto nla_put_failure; |
930345ea JB |
809 | if (nla_put_in6_addr(msg, TCP_METRICS_ATTR_SADDR_IPV6, |
810 | &tm->tcpm_saddr.addr.in6) < 0) | |
8a59359c | 811 | goto nla_put_failure; |
d23ff701 JA |
812 | break; |
813 | default: | |
814 | return -EAFNOSUPPORT; | |
815 | } | |
816 | ||
817 | if (nla_put_msecs(msg, TCP_METRICS_ATTR_AGE, | |
818 | jiffies - tm->tcpm_stamp) < 0) | |
819 | goto nla_put_failure; | |
820 | if (tm->tcpm_ts_stamp) { | |
821 | if (nla_put_s32(msg, TCP_METRICS_ATTR_TW_TS_STAMP, | |
822 | (s32) (get_seconds() - tm->tcpm_ts_stamp)) < 0) | |
823 | goto nla_put_failure; | |
824 | if (nla_put_u32(msg, TCP_METRICS_ATTR_TW_TSVAL, | |
825 | tm->tcpm_ts) < 0) | |
826 | goto nla_put_failure; | |
827 | } | |
828 | ||
829 | { | |
830 | int n = 0; | |
831 | ||
832 | nest = nla_nest_start(msg, TCP_METRICS_ATTR_VALS); | |
833 | if (!nest) | |
834 | goto nla_put_failure; | |
740b0f18 ED |
835 | for (i = 0; i < TCP_METRIC_MAX_KERNEL + 1; i++) { |
836 | u32 val = tm->tcpm_vals[i]; | |
837 | ||
838 | if (!val) | |
d23ff701 | 839 | continue; |
740b0f18 ED |
840 | if (i == TCP_METRIC_RTT) { |
841 | if (nla_put_u32(msg, TCP_METRIC_RTT_US + 1, | |
842 | val) < 0) | |
843 | goto nla_put_failure; | |
844 | n++; | |
845 | val = max(val / 1000, 1U); | |
846 | } | |
847 | if (i == TCP_METRIC_RTTVAR) { | |
848 | if (nla_put_u32(msg, TCP_METRIC_RTTVAR_US + 1, | |
849 | val) < 0) | |
850 | goto nla_put_failure; | |
851 | n++; | |
852 | val = max(val / 1000, 1U); | |
853 | } | |
854 | if (nla_put_u32(msg, i + 1, val) < 0) | |
d23ff701 JA |
855 | goto nla_put_failure; |
856 | n++; | |
857 | } | |
858 | if (n) | |
859 | nla_nest_end(msg, nest); | |
860 | else | |
861 | nla_nest_cancel(msg, nest); | |
862 | } | |
863 | ||
864 | { | |
865 | struct tcp_fastopen_metrics tfom_copy[1], *tfom; | |
866 | unsigned int seq; | |
867 | ||
868 | do { | |
869 | seq = read_seqbegin(&fastopen_seqlock); | |
870 | tfom_copy[0] = tm->tcpm_fastopen; | |
871 | } while (read_seqretry(&fastopen_seqlock, seq)); | |
872 | ||
873 | tfom = tfom_copy; | |
874 | if (tfom->mss && | |
875 | nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_MSS, | |
876 | tfom->mss) < 0) | |
877 | goto nla_put_failure; | |
878 | if (tfom->syn_loss && | |
879 | (nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROPS, | |
880 | tfom->syn_loss) < 0 || | |
881 | nla_put_msecs(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS, | |
882 | jiffies - tfom->last_syn_loss) < 0)) | |
883 | goto nla_put_failure; | |
884 | if (tfom->cookie.len > 0 && | |
885 | nla_put(msg, TCP_METRICS_ATTR_FOPEN_COOKIE, | |
886 | tfom->cookie.len, tfom->cookie.val) < 0) | |
887 | goto nla_put_failure; | |
888 | } | |
889 | ||
890 | return 0; | |
891 | ||
892 | nla_put_failure: | |
893 | return -EMSGSIZE; | |
894 | } | |
895 | ||
896 | static int tcp_metrics_dump_info(struct sk_buff *skb, | |
897 | struct netlink_callback *cb, | |
898 | struct tcp_metrics_block *tm) | |
899 | { | |
900 | void *hdr; | |
901 | ||
15e47304 | 902 | hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, |
d23ff701 JA |
903 | &tcp_metrics_nl_family, NLM_F_MULTI, |
904 | TCP_METRICS_CMD_GET); | |
905 | if (!hdr) | |
906 | return -EMSGSIZE; | |
907 | ||
908 | if (tcp_metrics_fill_info(skb, tm) < 0) | |
909 | goto nla_put_failure; | |
910 | ||
053c095a JB |
911 | genlmsg_end(skb, hdr); |
912 | return 0; | |
d23ff701 JA |
913 | |
914 | nla_put_failure: | |
915 | genlmsg_cancel(skb, hdr); | |
916 | return -EMSGSIZE; | |
917 | } | |
918 | ||
919 | static int tcp_metrics_nl_dump(struct sk_buff *skb, | |
920 | struct netlink_callback *cb) | |
921 | { | |
922 | struct net *net = sock_net(skb->sk); | |
098a697b | 923 | unsigned int max_rows = 1U << tcp_metrics_hash_log; |
d23ff701 JA |
924 | unsigned int row, s_row = cb->args[0]; |
925 | int s_col = cb->args[1], col = s_col; | |
926 | ||
927 | for (row = s_row; row < max_rows; row++, s_col = 0) { | |
928 | struct tcp_metrics_block *tm; | |
098a697b | 929 | struct tcpm_hash_bucket *hb = tcp_metrics_hash + row; |
d23ff701 JA |
930 | |
931 | rcu_read_lock(); | |
932 | for (col = 0, tm = rcu_dereference(hb->chain); tm; | |
933 | tm = rcu_dereference(tm->tcpm_next), col++) { | |
849e8a0c EB |
934 | if (!net_eq(tm_net(tm), net)) |
935 | continue; | |
d23ff701 JA |
936 | if (col < s_col) |
937 | continue; | |
938 | if (tcp_metrics_dump_info(skb, cb, tm) < 0) { | |
939 | rcu_read_unlock(); | |
940 | goto done; | |
941 | } | |
942 | } | |
943 | rcu_read_unlock(); | |
944 | } | |
945 | ||
946 | done: | |
947 | cb->args[0] = row; | |
948 | cb->args[1] = col; | |
949 | return skb->len; | |
950 | } | |
951 | ||
3e7013dd CP |
952 | static int __parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr, |
953 | unsigned int *hash, int optional, int v4, int v6) | |
d23ff701 JA |
954 | { |
955 | struct nlattr *a; | |
956 | ||
3e7013dd | 957 | a = info->attrs[v4]; |
d23ff701 JA |
958 | if (a) { |
959 | addr->family = AF_INET; | |
67b61f6c | 960 | addr->addr.a4 = nla_get_in_addr(a); |
3e7013dd CP |
961 | if (hash) |
962 | *hash = (__force unsigned int) addr->addr.a4; | |
d23ff701 JA |
963 | return 0; |
964 | } | |
3e7013dd | 965 | a = info->attrs[v6]; |
d23ff701 | 966 | if (a) { |
2c42a3fb | 967 | if (nla_len(a) != sizeof(struct in6_addr)) |
d23ff701 JA |
968 | return -EINVAL; |
969 | addr->family = AF_INET6; | |
67b61f6c | 970 | addr->addr.in6 = nla_get_in6_addr(a); |
3e7013dd | 971 | if (hash) |
8f55db48 | 972 | *hash = ipv6_addr_hash(&addr->addr.in6); |
d23ff701 JA |
973 | return 0; |
974 | } | |
975 | return optional ? 1 : -EAFNOSUPPORT; | |
976 | } | |
977 | ||
3e7013dd CP |
978 | static int parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr, |
979 | unsigned int *hash, int optional) | |
980 | { | |
981 | return __parse_nl_addr(info, addr, hash, optional, | |
982 | TCP_METRICS_ATTR_ADDR_IPV4, | |
983 | TCP_METRICS_ATTR_ADDR_IPV6); | |
984 | } | |
985 | ||
986 | static int parse_nl_saddr(struct genl_info *info, struct inetpeer_addr *addr) | |
987 | { | |
988 | return __parse_nl_addr(info, addr, NULL, 0, | |
989 | TCP_METRICS_ATTR_SADDR_IPV4, | |
990 | TCP_METRICS_ATTR_SADDR_IPV6); | |
991 | } | |
992 | ||
d23ff701 JA |
993 | static int tcp_metrics_nl_cmd_get(struct sk_buff *skb, struct genl_info *info) |
994 | { | |
995 | struct tcp_metrics_block *tm; | |
3e7013dd | 996 | struct inetpeer_addr saddr, daddr; |
d23ff701 JA |
997 | unsigned int hash; |
998 | struct sk_buff *msg; | |
999 | struct net *net = genl_info_net(info); | |
1000 | void *reply; | |
1001 | int ret; | |
3e7013dd | 1002 | bool src = true; |
d23ff701 | 1003 | |
324fd55a | 1004 | ret = parse_nl_addr(info, &daddr, &hash, 0); |
d23ff701 JA |
1005 | if (ret < 0) |
1006 | return ret; | |
1007 | ||
3e7013dd CP |
1008 | ret = parse_nl_saddr(info, &saddr); |
1009 | if (ret < 0) | |
1010 | src = false; | |
1011 | ||
d23ff701 JA |
1012 | msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); |
1013 | if (!msg) | |
1014 | return -ENOMEM; | |
1015 | ||
1016 | reply = genlmsg_put_reply(msg, info, &tcp_metrics_nl_family, 0, | |
1017 | info->genlhdr->cmd); | |
1018 | if (!reply) | |
1019 | goto nla_put_failure; | |
1020 | ||
3e5da62d | 1021 | hash ^= net_hash_mix(net); |
098a697b | 1022 | hash = hash_32(hash, tcp_metrics_hash_log); |
d23ff701 JA |
1023 | ret = -ESRCH; |
1024 | rcu_read_lock(); | |
098a697b | 1025 | for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm; |
d23ff701 | 1026 | tm = rcu_dereference(tm->tcpm_next)) { |
3e7013dd | 1027 | if (addr_same(&tm->tcpm_daddr, &daddr) && |
849e8a0c EB |
1028 | (!src || addr_same(&tm->tcpm_saddr, &saddr)) && |
1029 | net_eq(tm_net(tm), net)) { | |
d23ff701 JA |
1030 | ret = tcp_metrics_fill_info(msg, tm); |
1031 | break; | |
1032 | } | |
1033 | } | |
1034 | rcu_read_unlock(); | |
1035 | if (ret < 0) | |
1036 | goto out_free; | |
1037 | ||
1038 | genlmsg_end(msg, reply); | |
1039 | return genlmsg_reply(msg, info); | |
1040 | ||
1041 | nla_put_failure: | |
1042 | ret = -EMSGSIZE; | |
1043 | ||
1044 | out_free: | |
1045 | nlmsg_free(msg); | |
1046 | return ret; | |
1047 | } | |
1048 | ||
8a4bff71 | 1049 | static void tcp_metrics_flush_all(struct net *net) |
d23ff701 | 1050 | { |
098a697b EB |
1051 | unsigned int max_rows = 1U << tcp_metrics_hash_log; |
1052 | struct tcpm_hash_bucket *hb = tcp_metrics_hash; | |
d23ff701 JA |
1053 | struct tcp_metrics_block *tm; |
1054 | unsigned int row; | |
1055 | ||
1056 | for (row = 0; row < max_rows; row++, hb++) { | |
04f721c6 | 1057 | struct tcp_metrics_block __rcu **pp; |
d23ff701 | 1058 | spin_lock_bh(&tcp_metrics_lock); |
04f721c6 | 1059 | pp = &hb->chain; |
9f1ab186 | 1060 | for (tm = deref_locked(*pp); tm; tm = deref_locked(*pp)) { |
04f721c6 EB |
1061 | if (net_eq(tm_net(tm), net)) { |
1062 | *pp = tm->tcpm_next; | |
1063 | kfree_rcu(tm, rcu_head); | |
1064 | } else { | |
1065 | pp = &tm->tcpm_next; | |
1066 | } | |
d23ff701 | 1067 | } |
04f721c6 | 1068 | spin_unlock_bh(&tcp_metrics_lock); |
d23ff701 | 1069 | } |
d23ff701 JA |
1070 | } |
1071 | ||
1072 | static int tcp_metrics_nl_cmd_del(struct sk_buff *skb, struct genl_info *info) | |
1073 | { | |
1074 | struct tcpm_hash_bucket *hb; | |
00ca9c5b | 1075 | struct tcp_metrics_block *tm; |
d23ff701 | 1076 | struct tcp_metrics_block __rcu **pp; |
3e7013dd | 1077 | struct inetpeer_addr saddr, daddr; |
d23ff701 JA |
1078 | unsigned int hash; |
1079 | struct net *net = genl_info_net(info); | |
1080 | int ret; | |
00ca9c5b | 1081 | bool src = true, found = false; |
d23ff701 | 1082 | |
324fd55a | 1083 | ret = parse_nl_addr(info, &daddr, &hash, 1); |
d23ff701 JA |
1084 | if (ret < 0) |
1085 | return ret; | |
8a4bff71 EB |
1086 | if (ret > 0) { |
1087 | tcp_metrics_flush_all(net); | |
1088 | return 0; | |
1089 | } | |
3e7013dd CP |
1090 | ret = parse_nl_saddr(info, &saddr); |
1091 | if (ret < 0) | |
1092 | src = false; | |
d23ff701 | 1093 | |
3e5da62d | 1094 | hash ^= net_hash_mix(net); |
098a697b EB |
1095 | hash = hash_32(hash, tcp_metrics_hash_log); |
1096 | hb = tcp_metrics_hash + hash; | |
d23ff701 JA |
1097 | pp = &hb->chain; |
1098 | spin_lock_bh(&tcp_metrics_lock); | |
9f1ab186 | 1099 | for (tm = deref_locked(*pp); tm; tm = deref_locked(*pp)) { |
3e7013dd | 1100 | if (addr_same(&tm->tcpm_daddr, &daddr) && |
849e8a0c EB |
1101 | (!src || addr_same(&tm->tcpm_saddr, &saddr)) && |
1102 | net_eq(tm_net(tm), net)) { | |
d23ff701 | 1103 | *pp = tm->tcpm_next; |
00ca9c5b CP |
1104 | kfree_rcu(tm, rcu_head); |
1105 | found = true; | |
bbf852b9 CP |
1106 | } else { |
1107 | pp = &tm->tcpm_next; | |
d23ff701 JA |
1108 | } |
1109 | } | |
1110 | spin_unlock_bh(&tcp_metrics_lock); | |
00ca9c5b | 1111 | if (!found) |
d23ff701 | 1112 | return -ESRCH; |
d23ff701 JA |
1113 | return 0; |
1114 | } | |
1115 | ||
4534de83 | 1116 | static const struct genl_ops tcp_metrics_nl_ops[] = { |
d23ff701 JA |
1117 | { |
1118 | .cmd = TCP_METRICS_CMD_GET, | |
1119 | .doit = tcp_metrics_nl_cmd_get, | |
1120 | .dumpit = tcp_metrics_nl_dump, | |
1121 | .policy = tcp_metrics_nl_policy, | |
d23ff701 JA |
1122 | }, |
1123 | { | |
1124 | .cmd = TCP_METRICS_CMD_DEL, | |
1125 | .doit = tcp_metrics_nl_cmd_del, | |
1126 | .policy = tcp_metrics_nl_policy, | |
1127 | .flags = GENL_ADMIN_PERM, | |
1128 | }, | |
1129 | }; | |
1130 | ||
5815d5e7 | 1131 | static unsigned int tcpmhash_entries; |
51c5d0c4 DM |
1132 | static int __init set_tcpmhash_entries(char *str) |
1133 | { | |
1134 | ssize_t ret; | |
1135 | ||
1136 | if (!str) | |
1137 | return 0; | |
1138 | ||
5815d5e7 | 1139 | ret = kstrtouint(str, 0, &tcpmhash_entries); |
51c5d0c4 DM |
1140 | if (ret) |
1141 | return 0; | |
1142 | ||
1143 | return 1; | |
1144 | } | |
1145 | __setup("tcpmhash_entries=", set_tcpmhash_entries); | |
1146 | ||
1147 | static int __net_init tcp_net_metrics_init(struct net *net) | |
1148 | { | |
5815d5e7 ED |
1149 | size_t size; |
1150 | unsigned int slots; | |
51c5d0c4 | 1151 | |
098a697b EB |
1152 | if (!net_eq(net, &init_net)) |
1153 | return 0; | |
1154 | ||
51c5d0c4 DM |
1155 | slots = tcpmhash_entries; |
1156 | if (!slots) { | |
1157 | if (totalram_pages >= 128 * 1024) | |
1158 | slots = 16 * 1024; | |
1159 | else | |
1160 | slots = 8 * 1024; | |
1161 | } | |
1162 | ||
098a697b EB |
1163 | tcp_metrics_hash_log = order_base_2(slots); |
1164 | size = sizeof(struct tcpm_hash_bucket) << tcp_metrics_hash_log; | |
51c5d0c4 | 1165 | |
098a697b EB |
1166 | tcp_metrics_hash = kzalloc(size, GFP_KERNEL | __GFP_NOWARN); |
1167 | if (!tcp_metrics_hash) | |
1168 | tcp_metrics_hash = vzalloc(size); | |
976a702a | 1169 | |
098a697b | 1170 | if (!tcp_metrics_hash) |
51c5d0c4 DM |
1171 | return -ENOMEM; |
1172 | ||
51c5d0c4 DM |
1173 | return 0; |
1174 | } | |
1175 | ||
1176 | static void __net_exit tcp_net_metrics_exit(struct net *net) | |
1177 | { | |
098a697b | 1178 | tcp_metrics_flush_all(net); |
51c5d0c4 DM |
1179 | } |
1180 | ||
1181 | static __net_initdata struct pernet_operations tcp_net_metrics_ops = { | |
1182 | .init = tcp_net_metrics_init, | |
1183 | .exit = tcp_net_metrics_exit, | |
1184 | }; | |
1185 | ||
1186 | void __init tcp_metrics_init(void) | |
1187 | { | |
d23ff701 JA |
1188 | int ret; |
1189 | ||
1190 | ret = register_pernet_subsys(&tcp_net_metrics_ops); | |
1191 | if (ret < 0) | |
6493517e EB |
1192 | panic("Could not allocate the tcp_metrics hash table\n"); |
1193 | ||
d23ff701 | 1194 | ret = genl_register_family_with_ops(&tcp_metrics_nl_family, |
c53ed742 | 1195 | tcp_metrics_nl_ops); |
d23ff701 | 1196 | if (ret < 0) |
6493517e | 1197 | panic("Could not register tcp_metrics generic netlink\n"); |
51c5d0c4 | 1198 | } |