projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'char-misc-4.6-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
[linux-2.6-block.git] / net / ipv4 / tcp_metrics.c
CommitLineData
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
23int sysctl_tcp_nometrics_save __read_mostly;
24
41804420
DM		 25static 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		 29struct 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		 42struct 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		 57static inline struct net *tm_net(struct tcp_metrics_block *tm)
58{
59 return read_pnet(&tm->tcpm_net);
60}
61
51c5d0c4
DM		 62static 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
68static 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		 74static 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		 81static bool addr_same(const struct inetpeer_addr *a,
82 const struct inetpeer_addr *b)
83{
d39d14ff 84 return inetpeer_addr_cmp(a, b) == 0;
51c5d0c4
DM		 85}
86
87struct tcpm_hash_bucket {
88 struct tcp_metrics_block __rcu *chain;
89};
90
098a697b
EB		 91static struct tcpm_hash_bucket *tcp_metrics_hash __read_mostly;
92static unsigned int tcp_metrics_hash_log __read_mostly;
93
51c5d0c4
DM		 94static DEFINE_SPINLOCK(tcp_metrics_lock);
95
740b0f18
ED		 96static 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;
2646c831
DL		 131 tm->tcpm_fastopen.try_exp = 0;
132 tm->tcpm_fastopen.cookie.exp = false;
efeaa555
ED		 133 tm->tcpm_fastopen.cookie.len = 0;
134 }
51c5d0c4
DM		 135}
136
77f99ad1
CP		 137#define TCP_METRICS_TIMEOUT (60 * 60 * HZ)
138
139static void tcpm_check_stamp(struct tcp_metrics_block *tm, struct dst_entry *dst)
140{
141 if (tm && unlikely(time_after(jiffies, tm->tcpm_stamp + TCP_METRICS_TIMEOUT)))
142 tcpm_suck_dst(tm, dst, false);
143}
144
145#define TCP_METRICS_RECLAIM_DEPTH 5
146#define TCP_METRICS_RECLAIM_PTR (struct tcp_metrics_block *) 0x1UL
147
9f1ab186
ED		 148#define deref_locked(p) \
149 rcu_dereference_protected(p, lockdep_is_held(&tcp_metrics_lock))
150
51c5d0c4 151static struct tcp_metrics_block *tcpm_new(struct dst_entry *dst,
a5443028 152 struct inetpeer_addr *saddr,
324fd55a 153 struct inetpeer_addr *daddr,
77f99ad1 154 unsigned int hash)
51c5d0c4
DM		 155{
156 struct tcp_metrics_block *tm;
157 struct net *net;
77f99ad1 158 bool reclaim = false;
51c5d0c4
DM		 159
160 spin_lock_bh(&tcp_metrics_lock);
161 net = dev_net(dst->dev);
77f99ad1
CP		 162
163 /* While waiting for the spin-lock the cache might have been populated
164 * with this entry and so we have to check again.
165 */
41804420 166 tm = __tcp_get_metrics(saddr, daddr, net, hash);
77f99ad1
CP		 167 if (tm == TCP_METRICS_RECLAIM_PTR) {
168 reclaim = true;
169 tm = NULL;
170 }
171 if (tm) {
172 tcpm_check_stamp(tm, dst);
173 goto out_unlock;
174 }
175
51c5d0c4
DM		 176 if (unlikely(reclaim)) {
177 struct tcp_metrics_block *oldest;
178
9f1ab186
ED		 179 oldest = deref_locked(tcp_metrics_hash[hash].chain);
180 for (tm = deref_locked(oldest->tcpm_next); tm;
181 tm = deref_locked(tm->tcpm_next)) {
51c5d0c4
DM		 182 if (time_before(tm->tcpm_stamp, oldest->tcpm_stamp))
183 oldest = tm;
184 }
185 tm = oldest;
186 } else {
187 tm = kmalloc(sizeof(*tm), GFP_ATOMIC);
188 if (!tm)
189 goto out_unlock;
190 }
849e8a0c 191 write_pnet(&tm->tcpm_net, net);
a5443028 192 tm->tcpm_saddr = *saddr;
324fd55a 193 tm->tcpm_daddr = *daddr;
51c5d0c4 194
efeaa555 195 tcpm_suck_dst(tm, dst, true);
51c5d0c4
DM		 196
197 if (likely(!reclaim)) {
098a697b
EB		 198 tm->tcpm_next = tcp_metrics_hash[hash].chain;
199 rcu_assign_pointer(tcp_metrics_hash[hash].chain, tm);
51c5d0c4
DM		 200 }
201
202out_unlock:
203 spin_unlock_bh(&tcp_metrics_lock);
204 return tm;
205}
206
51c5d0c4
DM		 207static struct tcp_metrics_block *tcp_get_encode(struct tcp_metrics_block *tm, int depth)
208{
209 if (tm)
210 return tm;
211 if (depth > TCP_METRICS_RECLAIM_DEPTH)
212 return TCP_METRICS_RECLAIM_PTR;
213 return NULL;
214}
215
a5443028
CP		 216static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr,
217 const struct inetpeer_addr *daddr,
51c5d0c4
DM		 218 struct net *net, unsigned int hash)
219{
220 struct tcp_metrics_block *tm;
221 int depth = 0;
222
098a697b 223 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
51c5d0c4 224 tm = rcu_dereference(tm->tcpm_next)) {
a5443028 225 if (addr_same(&tm->tcpm_saddr, saddr) &&
849e8a0c
EB		 226 addr_same(&tm->tcpm_daddr, daddr) &&
227 net_eq(tm_net(tm), net))
51c5d0c4
DM		 228 break;
229 depth++;
230 }
231 return tcp_get_encode(tm, depth);
232}
233
234static struct tcp_metrics_block *__tcp_get_metrics_req(struct request_sock *req,
235 struct dst_entry *dst)
236{
237 struct tcp_metrics_block *tm;
a5443028 238 struct inetpeer_addr saddr, daddr;
51c5d0c4
DM		 239 unsigned int hash;
240 struct net *net;
241
a5443028 242 saddr.family = req->rsk_ops->family;
324fd55a
CP		 243 daddr.family = req->rsk_ops->family;
244 switch (daddr.family) {
51c5d0c4 245 case AF_INET:
3abef286
DA		 246 inetpeer_set_addr_v4(&saddr, inet_rsk(req)->ir_loc_addr);
247 inetpeer_set_addr_v4(&daddr, inet_rsk(req)->ir_rmt_addr);
72afa352 248 hash = ipv4_addr_hash(inet_rsk(req)->ir_rmt_addr);
51c5d0c4 249 break;
634fb979 250#if IS_ENABLED(CONFIG_IPV6)
51c5d0c4 251 case AF_INET6:
3abef286
DA		 252 inetpeer_set_addr_v6(&saddr, &inet_rsk(req)->ir_v6_loc_addr);
253 inetpeer_set_addr_v6(&daddr, &inet_rsk(req)->ir_v6_rmt_addr);
634fb979 254 hash = ipv6_addr_hash(&inet_rsk(req)->ir_v6_rmt_addr);
51c5d0c4 255 break;
634fb979 256#endif
51c5d0c4
DM		 257 default:
258 return NULL;
259 }
260
51c5d0c4 261 net = dev_net(dst->dev);
3e5da62d 262 hash ^= net_hash_mix(net);
098a697b 263 hash = hash_32(hash, tcp_metrics_hash_log);
51c5d0c4 264
098a697b 265 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
51c5d0c4 266 tm = rcu_dereference(tm->tcpm_next)) {
a5443028 267 if (addr_same(&tm->tcpm_saddr, &saddr) &&
849e8a0c
EB		 268 addr_same(&tm->tcpm_daddr, &daddr) &&
269 net_eq(tm_net(tm), net))
51c5d0c4
DM		 270 break;
271 }
272 tcpm_check_stamp(tm, dst);
273 return tm;
274}
275
81166dd6
DM		 276static struct tcp_metrics_block *__tcp_get_metrics_tw(struct inet_timewait_sock *tw)
277{
81166dd6 278 struct tcp_metrics_block *tm;
a5443028 279 struct inetpeer_addr saddr, daddr;
81166dd6
DM		 280 unsigned int hash;
281 struct net *net;
282
3ad88cf7 283 if (tw->tw_family == AF_INET) {
3abef286
DA		 284 inetpeer_set_addr_v4(&saddr, tw->tw_rcv_saddr);
285 inetpeer_set_addr_v4(&daddr, tw->tw_daddr);
72afa352 286 hash = ipv4_addr_hash(tw->tw_daddr);
3ad88cf7 287 }
c2bb06db 288#if IS_ENABLED(CONFIG_IPV6)
3ad88cf7
CP		 289 else if (tw->tw_family == AF_INET6) {
290 if (ipv6_addr_v4mapped(&tw->tw_v6_daddr)) {
3abef286
DA		 291 inetpeer_set_addr_v4(&saddr, tw->tw_rcv_saddr);
292 inetpeer_set_addr_v4(&daddr, tw->tw_daddr);
72afa352 293 hash = ipv4_addr_hash(tw->tw_daddr);
3ad88cf7 294 } else {
3abef286
DA		 295 inetpeer_set_addr_v6(&saddr, &tw->tw_v6_rcv_saddr);
296 inetpeer_set_addr_v6(&daddr, &tw->tw_v6_daddr);
3ad88cf7
CP		 297 hash = ipv6_addr_hash(&tw->tw_v6_daddr);
298 }
299 }
c2bb06db 300#endif
3ad88cf7 301 else
81166dd6 302 return NULL;
81166dd6 303
81166dd6 304 net = twsk_net(tw);
3e5da62d 305 hash ^= net_hash_mix(net);
098a697b 306 hash = hash_32(hash, tcp_metrics_hash_log);
81166dd6 307
098a697b 308 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
81166dd6 309 tm = rcu_dereference(tm->tcpm_next)) {
a5443028 310 if (addr_same(&tm->tcpm_saddr, &saddr) &&
849e8a0c
EB		 311 addr_same(&tm->tcpm_daddr, &daddr) &&
312 net_eq(tm_net(tm), net))
81166dd6
DM		 313 break;
314 }
315 return tm;
316}
317
51c5d0c4
DM		 318static struct tcp_metrics_block *tcp_get_metrics(struct sock *sk,
319 struct dst_entry *dst,
320 bool create)
321{
322 struct tcp_metrics_block *tm;
a5443028 323 struct inetpeer_addr saddr, daddr;
51c5d0c4
DM		 324 unsigned int hash;
325 struct net *net;
51c5d0c4 326
3ad88cf7 327 if (sk->sk_family == AF_INET) {
3abef286
DA		 328 inetpeer_set_addr_v4(&saddr, inet_sk(sk)->inet_saddr);
329 inetpeer_set_addr_v4(&daddr, inet_sk(sk)->inet_daddr);
72afa352 330 hash = ipv4_addr_hash(inet_sk(sk)->inet_daddr);
3ad88cf7 331 }
c2bb06db 332#if IS_ENABLED(CONFIG_IPV6)
3ad88cf7
CP		 333 else if (sk->sk_family == AF_INET6) {
334 if (ipv6_addr_v4mapped(&sk->sk_v6_daddr)) {
3abef286
DA		 335 inetpeer_set_addr_v4(&saddr, inet_sk(sk)->inet_saddr);
336 inetpeer_set_addr_v4(&daddr, inet_sk(sk)->inet_daddr);
72afa352 337 hash = ipv4_addr_hash(inet_sk(sk)->inet_daddr);
3ad88cf7 338 } else {
3abef286
DA		 339 inetpeer_set_addr_v6(&saddr, &sk->sk_v6_rcv_saddr);
340 inetpeer_set_addr_v6(&daddr, &sk->sk_v6_daddr);
3ad88cf7
CP		 341 hash = ipv6_addr_hash(&sk->sk_v6_daddr);
342 }
343 }
c2bb06db 344#endif
3ad88cf7 345 else
51c5d0c4 346 return NULL;
51c5d0c4 347
51c5d0c4 348 net = dev_net(dst->dev);
3e5da62d 349 hash ^= net_hash_mix(net);
098a697b 350 hash = hash_32(hash, tcp_metrics_hash_log);
51c5d0c4 351
a5443028 352 tm = __tcp_get_metrics(&saddr, &daddr, net, hash);
77f99ad1 353 if (tm == TCP_METRICS_RECLAIM_PTR)
51c5d0c4 354 tm = NULL;
51c5d0c4 355 if (!tm && create)
41804420 356 tm = tcpm_new(dst, &saddr, &daddr, hash);
51c5d0c4
DM		 357 else
358 tcpm_check_stamp(tm, dst);
359
360 return tm;
361}
362
4aabd8ef
DM		 363/* Save metrics learned by this TCP session. This function is called
364 * only, when TCP finishes successfully i.e. when it enters TIME-WAIT
365 * or goes from LAST-ACK to CLOSE.
366 */
367void tcp_update_metrics(struct sock *sk)
368{
51c5d0c4 369 const struct inet_connection_sock *icsk = inet_csk(sk);
4aabd8ef 370 struct dst_entry *dst = __sk_dst_get(sk);
51c5d0c4
DM		 371 struct tcp_sock *tp = tcp_sk(sk);
372 struct tcp_metrics_block *tm;
373 unsigned long rtt;
374 u32 val;
375 int m;
4aabd8ef 376
51c5d0c4 377 if (sysctl_tcp_nometrics_save || !dst)
4aabd8ef
DM		 378 return;
379
51c5d0c4 380 if (dst->flags & DST_HOST)
4aabd8ef
DM		 381 dst_confirm(dst);
382
51c5d0c4 383 rcu_read_lock();
740b0f18 384 if (icsk->icsk_backoff || !tp->srtt_us) {
51c5d0c4
DM		 385 /* This session failed to estimate rtt. Why?
386 * Probably, no packets returned in time. Reset our
387 * results.
388 */
389 tm = tcp_get_metrics(sk, dst, false);
390 if (tm && !tcp_metric_locked(tm, TCP_METRIC_RTT))
391 tcp_metric_set(tm, TCP_METRIC_RTT, 0);
392 goto out_unlock;
393 } else
394 tm = tcp_get_metrics(sk, dst, true);
4aabd8ef 395
51c5d0c4
DM		 396 if (!tm)
397 goto out_unlock;
4aabd8ef 398
740b0f18
ED		 399 rtt = tcp_metric_get(tm, TCP_METRIC_RTT);
400 m = rtt - tp->srtt_us;
4aabd8ef 401
51c5d0c4
DM		 402 /* If newly calculated rtt larger than stored one, store new
403 * one. Otherwise, use EWMA. Remember, rtt overestimation is
404 * always better than underestimation.
405 */
406 if (!tcp_metric_locked(tm, TCP_METRIC_RTT)) {
407 if (m <= 0)
740b0f18 408 rtt = tp->srtt_us;
51c5d0c4
DM		 409 else
410 rtt -= (m >> 3);
740b0f18 411 tcp_metric_set(tm, TCP_METRIC_RTT, rtt);
51c5d0c4 412 }
4aabd8ef 413
51c5d0c4
DM		 414 if (!tcp_metric_locked(tm, TCP_METRIC_RTTVAR)) {
415 unsigned long var;
4aabd8ef 416
51c5d0c4
DM		 417 if (m < 0)
418 m = -m;
4aabd8ef 419
51c5d0c4
DM		 420 /* Scale deviation to rttvar fixed point */
421 m >>= 1;
740b0f18
ED		 422 if (m < tp->mdev_us)
423 m = tp->mdev_us;
4aabd8ef 424
740b0f18 425 var = tcp_metric_get(tm, TCP_METRIC_RTTVAR);
51c5d0c4
DM		 426 if (m >= var)
427 var = m;
428 else
429 var -= (var - m) >> 2;
4aabd8ef 430
740b0f18 431 tcp_metric_set(tm, TCP_METRIC_RTTVAR, var);
51c5d0c4
DM		 432 }
433
434 if (tcp_in_initial_slowstart(tp)) {
435 /* Slow start still did not finish. */
436 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) {
437 val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
438 if (val && (tp->snd_cwnd >> 1) > val)
439 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
440 tp->snd_cwnd >> 1);
441 }
442 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
443 val = tcp_metric_get(tm, TCP_METRIC_CWND);
444 if (tp->snd_cwnd > val)
445 tcp_metric_set(tm, TCP_METRIC_CWND,
446 tp->snd_cwnd);
447 }
071d5080 448 } else if (!tcp_in_slow_start(tp) &&
51c5d0c4
DM		 449 icsk->icsk_ca_state == TCP_CA_Open) {
450 /* Cong. avoidance phase, cwnd is reliable. */
451 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH))
452 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
453 max(tp->snd_cwnd >> 1, tp->snd_ssthresh));
454 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
455 val = tcp_metric_get(tm, TCP_METRIC_CWND);
2100844c 456 tcp_metric_set(tm, TCP_METRIC_CWND, (val + tp->snd_cwnd) >> 1);
51c5d0c4
DM		 457 }
458 } else {
459 /* Else slow start did not finish, cwnd is non-sense,
460 * ssthresh may be also invalid.
461 */
462 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
463 val = tcp_metric_get(tm, TCP_METRIC_CWND);
464 tcp_metric_set(tm, TCP_METRIC_CWND,
465 (val + tp->snd_ssthresh) >> 1);
466 }
467 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) {
468 val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
469 if (val && tp->snd_ssthresh > val)
470 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
471 tp->snd_ssthresh);
472 }
473 if (!tcp_metric_locked(tm, TCP_METRIC_REORDERING)) {
474 val = tcp_metric_get(tm, TCP_METRIC_REORDERING);
475 if (val < tp->reordering &&
4aabd8ef 476 tp->reordering != sysctl_tcp_reordering)
51c5d0c4
DM		 477 tcp_metric_set(tm, TCP_METRIC_REORDERING,
478 tp->reordering);
4aabd8ef
DM		 479 }
480 }
51c5d0c4
DM		 481 tm->tcpm_stamp = jiffies;
482out_unlock:
483 rcu_read_unlock();
4aabd8ef
DM		 484}
485
486/* Initialize metrics on socket. */
487
488void tcp_init_metrics(struct sock *sk)
489{
4aabd8ef 490 struct dst_entry *dst = __sk_dst_get(sk);
51c5d0c4
DM		 491 struct tcp_sock *tp = tcp_sk(sk);
492 struct tcp_metrics_block *tm;
1b7fdd2a 493 u32 val, crtt = 0; /* cached RTT scaled by 8 */
4aabd8ef 494
51456b29 495 if (!dst)
4aabd8ef
DM		 496 goto reset;
497
498 dst_confirm(dst);
499
51c5d0c4
DM		 500 rcu_read_lock();
501 tm = tcp_get_metrics(sk, dst, true);
502 if (!tm) {
503 rcu_read_unlock();
504 goto reset;
505 }
506
507 if (tcp_metric_locked(tm, TCP_METRIC_CWND))
508 tp->snd_cwnd_clamp = tcp_metric_get(tm, TCP_METRIC_CWND);
509
510 val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
511 if (val) {
512 tp->snd_ssthresh = val;
4aabd8ef
DM		 513 if (tp->snd_ssthresh > tp->snd_cwnd_clamp)
514 tp->snd_ssthresh = tp->snd_cwnd_clamp;
515 } else {
516 /* ssthresh may have been reduced unnecessarily during.
517 * 3WHS. Restore it back to its initial default.
518 */
519 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
520 }
51c5d0c4
DM		 521 val = tcp_metric_get(tm, TCP_METRIC_REORDERING);
522 if (val && tp->reordering != val) {
4aabd8ef
DM		 523 tcp_disable_fack(tp);
524 tcp_disable_early_retrans(tp);
51c5d0c4 525 tp->reordering = val;
4aabd8ef
DM		 526 }
527
740b0f18 528 crtt = tcp_metric_get(tm, TCP_METRIC_RTT);
51c5d0c4 529 rcu_read_unlock();
4aabd8ef 530reset:
52f20e65
YC		 531 /* The initial RTT measurement from the SYN/SYN-ACK is not ideal
532 * to seed the RTO for later data packets because SYN packets are
533 * small. Use the per-dst cached values to seed the RTO but keep
534 * the RTT estimator variables intact (e.g., srtt, mdev, rttvar).
535 * Later the RTO will be updated immediately upon obtaining the first
536 * data RTT sample (tcp_rtt_estimator()). Hence the cached RTT only
537 * influences the first RTO but not later RTT estimation.
538 *
539 * But if RTT is not available from the SYN (due to retransmits or
540 * syn cookies) or the cache, force a conservative 3secs timeout.
541 *
542 * A bit of theory. RTT is time passed after "normal" sized packet
543 * is sent until it is ACKed. In normal circumstances sending small
544 * packets force peer to delay ACKs and calculation is correct too.
545 * The algorithm is adaptive and, provided we follow specs, it
546 * NEVER underestimate RTT. BUT! If peer tries to make some clever
547 * tricks sort of "quick acks" for time long enough to decrease RTT
548 * to low value, and then abruptly stops to do it and starts to delay
549 * ACKs, wait for troubles.
550 */
740b0f18 551 if (crtt > tp->srtt_us) {
269aa759 552 /* Set RTO like tcp_rtt_estimator(), but from cached RTT. */
9bdfb3b7 553 crtt /= 8 * USEC_PER_SEC / HZ;
269aa759 554 inet_csk(sk)->icsk_rto = crtt + max(2 * crtt, tcp_rto_min(sk));
740b0f18 555 } else if (tp->srtt_us == 0) {
4aabd8ef
DM		 556 /* RFC6298: 5.7 We've failed to get a valid RTT sample from
557 * 3WHS. This is most likely due to retransmission,
558 * including spurious one. Reset the RTO back to 3secs
559 * from the more aggressive 1sec to avoid more spurious
560 * retransmission.
561 */
740b0f18
ED		 562 tp->rttvar_us = jiffies_to_usecs(TCP_TIMEOUT_FALLBACK);
563 tp->mdev_us = tp->mdev_max_us = tp->rttvar_us;
564
4aabd8ef
DM		 565 inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK;
566 }
567 /* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
568 * retransmitted. In light of RFC6298 more aggressive 1sec
569 * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
570 * retransmission has occurred.
571 */
572 if (tp->total_retrans > 1)
573 tp->snd_cwnd = 1;
574 else
575 tp->snd_cwnd = tcp_init_cwnd(tp, dst);
576 tp->snd_cwnd_stamp = tcp_time_stamp;
577}
ab92bb2f 578
a26552af
HFS		 579bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
580 bool paws_check, bool timestamps)
ab92bb2f 581{
51c5d0c4
DM		 582 struct tcp_metrics_block *tm;
583 bool ret;
584
ab92bb2f
DM		 585 if (!dst)
586 return false;
51c5d0c4
DM		 587
588 rcu_read_lock();
589 tm = __tcp_get_metrics_req(req, dst);
81166dd6
DM		 590 if (paws_check) {
591 if (tm &&
592 (u32)get_seconds() - tm->tcpm_ts_stamp < TCP_PAWS_MSL &&
a26552af
HFS		 593 ((s32)(tm->tcpm_ts - req->ts_recent) > TCP_PAWS_WINDOW ||
594 !timestamps))
81166dd6
DM		 595 ret = false;
596 else
597 ret = true;
598 } else {
599 if (tm && tcp_metric_get(tm, TCP_METRIC_RTT) && tm->tcpm_ts_stamp)
600 ret = true;
601 else
602 ret = false;
603 }
51c5d0c4
DM		 604 rcu_read_unlock();
605
606 return ret;
ab92bb2f
DM		 607}
608EXPORT_SYMBOL_GPL(tcp_peer_is_proven);
51c5d0c4 609
81166dd6
DM		 610void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst)
611{
612 struct tcp_metrics_block *tm;
613
614 rcu_read_lock();
615 tm = tcp_get_metrics(sk, dst, true);
616 if (tm) {
617 struct tcp_sock *tp = tcp_sk(sk);
618
619 if ((u32)get_seconds() - tm->tcpm_ts_stamp <= TCP_PAWS_MSL) {
620 tp->rx_opt.ts_recent_stamp = tm->tcpm_ts_stamp;
621 tp->rx_opt.ts_recent = tm->tcpm_ts;
622 }
623 }
624 rcu_read_unlock();
625}
626EXPORT_SYMBOL_GPL(tcp_fetch_timewait_stamp);
627
628/* VJ's idea. Save last timestamp seen from this destination and hold
629 * it at least for normal timewait interval to use for duplicate
630 * segment detection in subsequent connections, before they enter
631 * synchronized state.
632 */
633bool tcp_remember_stamp(struct sock *sk)
634{
635 struct dst_entry *dst = __sk_dst_get(sk);
636 bool ret = false;
637
638 if (dst) {
639 struct tcp_metrics_block *tm;
640
641 rcu_read_lock();
642 tm = tcp_get_metrics(sk, dst, true);
643 if (tm) {
644 struct tcp_sock *tp = tcp_sk(sk);
645
646 if ((s32)(tm->tcpm_ts - tp->rx_opt.ts_recent) <= 0 ||
647 ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL &&
648 tm->tcpm_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) {
649 tm->tcpm_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp;
650 tm->tcpm_ts = tp->rx_opt.ts_recent;
651 }
652 ret = true;
653 }
654 rcu_read_unlock();
655 }
656 return ret;
657}
658
659bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw)
660{
661 struct tcp_metrics_block *tm;
662 bool ret = false;
663
664 rcu_read_lock();
665 tm = __tcp_get_metrics_tw(tw);
9a0a9502 666 if (tm) {
81166dd6
DM		 667 const struct tcp_timewait_sock *tcptw;
668 struct sock *sk = (struct sock *) tw;
669
670 tcptw = tcp_twsk(sk);
671 if ((s32)(tm->tcpm_ts - tcptw->tw_ts_recent) <= 0 ||
672 ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL &&
673 tm->tcpm_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) {
674 tm->tcpm_ts_stamp = (u32)tcptw->tw_ts_recent_stamp;
675 tm->tcpm_ts = tcptw->tw_ts_recent;
676 }
677 ret = true;
678 }
679 rcu_read_unlock();
680
681 return ret;
682}
683
1fe4c481
YC		 684static DEFINE_SEQLOCK(fastopen_seqlock);
685
686void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
aab48743
YC		 687 struct tcp_fastopen_cookie *cookie,
688 int *syn_loss, unsigned long *last_syn_loss)
1fe4c481
YC		 689{
690 struct tcp_metrics_block *tm;
691
692 rcu_read_lock();
693 tm = tcp_get_metrics(sk, __sk_dst_get(sk), false);
694 if (tm) {
695 struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen;
696 unsigned int seq;
697
698 do {
699 seq = read_seqbegin(&fastopen_seqlock);
700 if (tfom->mss)
701 *mss = tfom->mss;
702 *cookie = tfom->cookie;
2646c831
DL		 703 if (cookie->len <= 0 && tfom->try_exp == 1)
704 cookie->exp = true;
aab48743
YC		 705 *syn_loss = tfom->syn_loss;
706 *last_syn_loss = *syn_loss ? tfom->last_syn_loss : 0;
1fe4c481
YC		 707 } while (read_seqretry(&fastopen_seqlock, seq));
708 }
709 rcu_read_unlock();
710}
711
1fe4c481 712void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
2646c831
DL		 713 struct tcp_fastopen_cookie *cookie, bool syn_lost,
714 u16 try_exp)
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;
2646c831
DL		 731 else if (try_exp > tfom->try_exp &&
732 tfom->cookie.len <= 0 && !tfom->cookie.exp)
733 tfom->try_exp = try_exp;
aab48743
YC		 734 if (syn_lost) {
735 ++tfom->syn_loss;
736 tfom->last_syn_loss = jiffies;
737 } else
738 tfom->syn_loss = 0;
1fe4c481
YC		 739 write_sequnlock_bh(&fastopen_seqlock);
740 }
741 rcu_read_unlock();
742}
743
d23ff701
JA		 744static struct genl_family tcp_metrics_nl_family = {
745 .id = GENL_ID_GENERATE,
746 .hdrsize = 0,
747 .name = TCP_METRICS_GENL_NAME,
748 .version = TCP_METRICS_GENL_VERSION,
749 .maxattr = TCP_METRICS_ATTR_MAX,
750 .netnsok = true,
751};
752
753static struct nla_policy tcp_metrics_nl_policy[TCP_METRICS_ATTR_MAX + 1] = {
754 [TCP_METRICS_ATTR_ADDR_IPV4] = { .type = NLA_U32, },
755 [TCP_METRICS_ATTR_ADDR_IPV6] = { .type = NLA_BINARY,
756 .len = sizeof(struct in6_addr), },
757 /* Following attributes are not received for GET/DEL,
758 * we keep them for reference
759 */
760#if 0
761 [TCP_METRICS_ATTR_AGE] = { .type = NLA_MSECS, },
762 [TCP_METRICS_ATTR_TW_TSVAL] = { .type = NLA_U32, },
763 [TCP_METRICS_ATTR_TW_TS_STAMP] = { .type = NLA_S32, },
764 [TCP_METRICS_ATTR_VALS] = { .type = NLA_NESTED, },
765 [TCP_METRICS_ATTR_FOPEN_MSS] = { .type = NLA_U16, },
766 [TCP_METRICS_ATTR_FOPEN_SYN_DROPS] = { .type = NLA_U16, },
767 [TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS] = { .type = NLA_MSECS, },
768 [TCP_METRICS_ATTR_FOPEN_COOKIE] = { .type = NLA_BINARY,
769 .len = TCP_FASTOPEN_COOKIE_MAX, },
770#endif
771};
772
773/* Add attributes, caller cancels its header on failure */
774static int tcp_metrics_fill_info(struct sk_buff *msg,
775 struct tcp_metrics_block *tm)
776{
777 struct nlattr *nest;
778 int i;
779
324fd55a 780 switch (tm->tcpm_daddr.family) {
d23ff701 781 case AF_INET:
930345ea 782 if (nla_put_in_addr(msg, TCP_METRICS_ATTR_ADDR_IPV4,
3abef286 783 inetpeer_get_addr_v4(&tm->tcpm_daddr)) < 0)
d23ff701 784 goto nla_put_failure;
930345ea 785 if (nla_put_in_addr(msg, TCP_METRICS_ATTR_SADDR_IPV4,
3abef286 786 inetpeer_get_addr_v4(&tm->tcpm_saddr)) < 0)
8a59359c 787 goto nla_put_failure;
d23ff701
JA		 788 break;
789 case AF_INET6:
930345ea 790 if (nla_put_in6_addr(msg, TCP_METRICS_ATTR_ADDR_IPV6,
3abef286 791 inetpeer_get_addr_v6(&tm->tcpm_daddr)) < 0)
d23ff701 792 goto nla_put_failure;
930345ea 793 if (nla_put_in6_addr(msg, TCP_METRICS_ATTR_SADDR_IPV6,
3abef286 794 inetpeer_get_addr_v6(&tm->tcpm_saddr)) < 0)
8a59359c 795 goto nla_put_failure;
d23ff701
JA		 796 break;
797 default:
798 return -EAFNOSUPPORT;
799 }
800
801 if (nla_put_msecs(msg, TCP_METRICS_ATTR_AGE,
802 jiffies - tm->tcpm_stamp) < 0)
803 goto nla_put_failure;
804 if (tm->tcpm_ts_stamp) {
805 if (nla_put_s32(msg, TCP_METRICS_ATTR_TW_TS_STAMP,
806 (s32) (get_seconds() - tm->tcpm_ts_stamp)) < 0)
807 goto nla_put_failure;
808 if (nla_put_u32(msg, TCP_METRICS_ATTR_TW_TSVAL,
809 tm->tcpm_ts) < 0)
810 goto nla_put_failure;
811 }
812
813 {
814 int n = 0;
815
816 nest = nla_nest_start(msg, TCP_METRICS_ATTR_VALS);
817 if (!nest)
818 goto nla_put_failure;
740b0f18
ED		 819 for (i = 0; i < TCP_METRIC_MAX_KERNEL + 1; i++) {
820 u32 val = tm->tcpm_vals[i];
821
822 if (!val)
d23ff701 823 continue;
740b0f18
ED		 824 if (i == TCP_METRIC_RTT) {
825 if (nla_put_u32(msg, TCP_METRIC_RTT_US + 1,
826 val) < 0)
827 goto nla_put_failure;
828 n++;
829 val = max(val / 1000, 1U);
830 }
831 if (i == TCP_METRIC_RTTVAR) {
832 if (nla_put_u32(msg, TCP_METRIC_RTTVAR_US + 1,
833 val) < 0)
834 goto nla_put_failure;
835 n++;
836 val = max(val / 1000, 1U);
837 }
838 if (nla_put_u32(msg, i + 1, val) < 0)
d23ff701
JA		 839 goto nla_put_failure;
840 n++;
841 }
842 if (n)
843 nla_nest_end(msg, nest);
844 else
845 nla_nest_cancel(msg, nest);
846 }
847
848 {
849 struct tcp_fastopen_metrics tfom_copy[1], *tfom;
850 unsigned int seq;
851
852 do {
853 seq = read_seqbegin(&fastopen_seqlock);
854 tfom_copy[0] = tm->tcpm_fastopen;
855 } while (read_seqretry(&fastopen_seqlock, seq));
856
857 tfom = tfom_copy;
858 if (tfom->mss &&
859 nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_MSS,
860 tfom->mss) < 0)
861 goto nla_put_failure;
862 if (tfom->syn_loss &&
863 (nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROPS,
864 tfom->syn_loss) < 0 ||
865 nla_put_msecs(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS,
866 jiffies - tfom->last_syn_loss) < 0))
867 goto nla_put_failure;
868 if (tfom->cookie.len > 0 &&
869 nla_put(msg, TCP_METRICS_ATTR_FOPEN_COOKIE,
870 tfom->cookie.len, tfom->cookie.val) < 0)
871 goto nla_put_failure;
872 }
873
874 return 0;
875
876nla_put_failure:
877 return -EMSGSIZE;
878}
879
880static int tcp_metrics_dump_info(struct sk_buff *skb,
881 struct netlink_callback *cb,
882 struct tcp_metrics_block *tm)
883{
884 void *hdr;
885
15e47304 886 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
d23ff701
JA		 887 &tcp_metrics_nl_family, NLM_F_MULTI,
888 TCP_METRICS_CMD_GET);
889 if (!hdr)
890 return -EMSGSIZE;
891
892 if (tcp_metrics_fill_info(skb, tm) < 0)
893 goto nla_put_failure;
894
053c095a
JB		 895 genlmsg_end(skb, hdr);
896 return 0;
d23ff701
JA		 897
898nla_put_failure:
899 genlmsg_cancel(skb, hdr);
900 return -EMSGSIZE;
901}
902
903static int tcp_metrics_nl_dump(struct sk_buff *skb,
904 struct netlink_callback *cb)
905{
906 struct net *net = sock_net(skb->sk);
098a697b 907 unsigned int max_rows = 1U << tcp_metrics_hash_log;
d23ff701
JA		 908 unsigned int row, s_row = cb->args[0];
909 int s_col = cb->args[1], col = s_col;
910
911 for (row = s_row; row < max_rows; row++, s_col = 0) {
912 struct tcp_metrics_block *tm;
098a697b 913 struct tcpm_hash_bucket *hb = tcp_metrics_hash + row;
d23ff701
JA		 914
915 rcu_read_lock();
916 for (col = 0, tm = rcu_dereference(hb->chain); tm;
917 tm = rcu_dereference(tm->tcpm_next), col++) {
849e8a0c
EB		 918 if (!net_eq(tm_net(tm), net))
919 continue;
d23ff701
JA		 920 if (col < s_col)
921 continue;
922 if (tcp_metrics_dump_info(skb, cb, tm) < 0) {
923 rcu_read_unlock();
924 goto done;
925 }
926 }
927 rcu_read_unlock();
928 }
929
930done:
931 cb->args[0] = row;
932 cb->args[1] = col;
933 return skb->len;
934}
935
3e7013dd
CP		 936static int __parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
937 unsigned int *hash, int optional, int v4, int v6)
d23ff701
JA		 938{
939 struct nlattr *a;
940
3e7013dd 941 a = info->attrs[v4];
d23ff701 942 if (a) {
3abef286 943 inetpeer_set_addr_v4(addr, nla_get_in_addr(a));
3e7013dd 944 if (hash)
3abef286 945 *hash = ipv4_addr_hash(inetpeer_get_addr_v4(addr));
d23ff701
JA		 946 return 0;
947 }
3e7013dd 948 a = info->attrs[v6];
d23ff701 949 if (a) {
3abef286
DA		 950 struct in6_addr in6;
951
2c42a3fb 952 if (nla_len(a) != sizeof(struct in6_addr))
d23ff701 953 return -EINVAL;
3abef286
DA		 954 in6 = nla_get_in6_addr(a);
955 inetpeer_set_addr_v6(addr, &in6);
3e7013dd 956 if (hash)
3abef286 957 *hash = ipv6_addr_hash(inetpeer_get_addr_v6(addr));
d23ff701
JA		 958 return 0;
959 }
960 return optional ? 1 : -EAFNOSUPPORT;
961}
962
3e7013dd
CP		 963static int parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
964 unsigned int *hash, int optional)
965{
966 return __parse_nl_addr(info, addr, hash, optional,
967 TCP_METRICS_ATTR_ADDR_IPV4,
968 TCP_METRICS_ATTR_ADDR_IPV6);
969}
970
971static int parse_nl_saddr(struct genl_info *info, struct inetpeer_addr *addr)
972{
973 return __parse_nl_addr(info, addr, NULL, 0,
974 TCP_METRICS_ATTR_SADDR_IPV4,
975 TCP_METRICS_ATTR_SADDR_IPV6);
976}
977
d23ff701
JA		 978static int tcp_metrics_nl_cmd_get(struct sk_buff *skb, struct genl_info *info)
979{
980 struct tcp_metrics_block *tm;
3e7013dd 981 struct inetpeer_addr saddr, daddr;
d23ff701
JA		 982 unsigned int hash;
983 struct sk_buff *msg;
984 struct net *net = genl_info_net(info);
985 void *reply;
986 int ret;
3e7013dd 987 bool src = true;
d23ff701 988
324fd55a 989 ret = parse_nl_addr(info, &daddr, &hash, 0);
d23ff701
JA		 990 if (ret < 0)
991 return ret;
992
3e7013dd
CP		 993 ret = parse_nl_saddr(info, &saddr);
994 if (ret < 0)
995 src = false;
996
d23ff701
JA		 997 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
998 if (!msg)
999 return -ENOMEM;
1000
1001 reply = genlmsg_put_reply(msg, info, &tcp_metrics_nl_family, 0,
1002 info->genlhdr->cmd);
1003 if (!reply)
1004 goto nla_put_failure;
1005
3e5da62d 1006 hash ^= net_hash_mix(net);
098a697b 1007 hash = hash_32(hash, tcp_metrics_hash_log);
d23ff701
JA		 1008 ret = -ESRCH;
1009 rcu_read_lock();
098a697b 1010 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
d23ff701 1011 tm = rcu_dereference(tm->tcpm_next)) {
3e7013dd 1012 if (addr_same(&tm->tcpm_daddr, &daddr) &&
849e8a0c
EB		 1013 (!src || addr_same(&tm->tcpm_saddr, &saddr)) &&
1014 net_eq(tm_net(tm), net)) {
d23ff701
JA		 1015 ret = tcp_metrics_fill_info(msg, tm);
1016 break;
1017 }
1018 }
1019 rcu_read_unlock();
1020 if (ret < 0)
1021 goto out_free;
1022
1023 genlmsg_end(msg, reply);
1024 return genlmsg_reply(msg, info);
1025
1026nla_put_failure:
1027 ret = -EMSGSIZE;
1028
1029out_free:
1030 nlmsg_free(msg);
1031 return ret;
1032}
1033
8a4bff71 1034static void tcp_metrics_flush_all(struct net *net)
d23ff701 1035{
098a697b
EB		 1036 unsigned int max_rows = 1U << tcp_metrics_hash_log;
1037 struct tcpm_hash_bucket *hb = tcp_metrics_hash;
d23ff701
JA		 1038 struct tcp_metrics_block *tm;
1039 unsigned int row;
1040
1041 for (row = 0; row < max_rows; row++, hb++) {
04f721c6 1042 struct tcp_metrics_block __rcu **pp;
d23ff701 1043 spin_lock_bh(&tcp_metrics_lock);
04f721c6 1044 pp = &hb->chain;
9f1ab186 1045 for (tm = deref_locked(*pp); tm; tm = deref_locked(*pp)) {
04f721c6
EB		 1046 if (net_eq(tm_net(tm), net)) {
1047 *pp = tm->tcpm_next;
1048 kfree_rcu(tm, rcu_head);
1049 } else {
1050 pp = &tm->tcpm_next;
1051 }
d23ff701 1052 }
04f721c6 1053 spin_unlock_bh(&tcp_metrics_lock);
d23ff701 1054 }
d23ff701
JA		 1055}
1056
1057static 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;
8a4bff71
EB		 1071 if (ret > 0) {
1072 tcp_metrics_flush_all(net);
1073 return 0;
1074 }
3e7013dd
CP		 1075 ret = parse_nl_saddr(info, &saddr);
1076 if (ret < 0)
1077 src = false;
d23ff701 1078
3e5da62d 1079 hash ^= net_hash_mix(net);
098a697b
EB		 1080 hash = hash_32(hash, tcp_metrics_hash_log);
1081 hb = tcp_metrics_hash + hash;
d23ff701
JA		 1082 pp = &hb->chain;
1083 spin_lock_bh(&tcp_metrics_lock);
9f1ab186 1084 for (tm = deref_locked(*pp); tm; tm = deref_locked(*pp)) {
3e7013dd 1085 if (addr_same(&tm->tcpm_daddr, &daddr) &&
849e8a0c
EB		 1086 (!src || addr_same(&tm->tcpm_saddr, &saddr)) &&
1087 net_eq(tm_net(tm), net)) {
d23ff701 1088 *pp = tm->tcpm_next;
00ca9c5b
CP		 1089 kfree_rcu(tm, rcu_head);
1090 found = true;
bbf852b9
CP		 1091 } else {
1092 pp = &tm->tcpm_next;
d23ff701
JA		 1093 }
1094 }
1095 spin_unlock_bh(&tcp_metrics_lock);
00ca9c5b 1096 if (!found)
d23ff701 1097 return -ESRCH;
d23ff701
JA		 1098 return 0;
1099}
1100
4534de83 1101static const struct genl_ops tcp_metrics_nl_ops[] = {
d23ff701
JA		 1102 {
1103 .cmd = TCP_METRICS_CMD_GET,
1104 .doit = tcp_metrics_nl_cmd_get,
1105 .dumpit = tcp_metrics_nl_dump,
1106 .policy = tcp_metrics_nl_policy,
d23ff701
JA		 1107 },
1108 {
1109 .cmd = TCP_METRICS_CMD_DEL,
1110 .doit = tcp_metrics_nl_cmd_del,
1111 .policy = tcp_metrics_nl_policy,
1112 .flags = GENL_ADMIN_PERM,
1113 },
1114};
1115
5815d5e7 1116static unsigned int tcpmhash_entries;
51c5d0c4
DM		 1117static int __init set_tcpmhash_entries(char *str)
1118{
1119 ssize_t ret;
1120
1121 if (!str)
1122 return 0;
1123
5815d5e7 1124 ret = kstrtouint(str, 0, &tcpmhash_entries);
51c5d0c4
DM		 1125 if (ret)
1126 return 0;
1127
1128 return 1;
1129}
1130__setup("tcpmhash_entries=", set_tcpmhash_entries);
1131
1132static int __net_init tcp_net_metrics_init(struct net *net)
1133{
5815d5e7
ED		 1134 size_t size;
1135 unsigned int slots;
51c5d0c4 1136
098a697b
EB		 1137 if (!net_eq(net, &init_net))
1138 return 0;
1139
51c5d0c4
DM		 1140 slots = tcpmhash_entries;
1141 if (!slots) {
1142 if (totalram_pages >= 128 * 1024)
1143 slots = 16 * 1024;
1144 else
1145 slots = 8 * 1024;
1146 }
1147
098a697b
EB		 1148 tcp_metrics_hash_log = order_base_2(slots);
1149 size = sizeof(struct tcpm_hash_bucket) << tcp_metrics_hash_log;
51c5d0c4 1150
098a697b
EB		 1151 tcp_metrics_hash = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
1152 if (!tcp_metrics_hash)
1153 tcp_metrics_hash = vzalloc(size);
976a702a 1154
098a697b 1155 if (!tcp_metrics_hash)
51c5d0c4
DM		 1156 return -ENOMEM;
1157
51c5d0c4
DM		 1158 return 0;
1159}
1160
1161static void __net_exit tcp_net_metrics_exit(struct net *net)
1162{
098a697b 1163 tcp_metrics_flush_all(net);
51c5d0c4
DM		 1164}
1165
1166static __net_initdata struct pernet_operations tcp_net_metrics_ops = {
1167 .init = tcp_net_metrics_init,
1168 .exit = tcp_net_metrics_exit,
1169};
1170
1171void __init tcp_metrics_init(void)
1172{
d23ff701
JA		 1173 int ret;
1174
1175 ret = register_pernet_subsys(&tcp_net_metrics_ops);
1176 if (ret < 0)
6493517e
EB		 1177 panic("Could not allocate the tcp_metrics hash table\n");
1178
d23ff701 1179 ret = genl_register_family_with_ops(&tcp_metrics_nl_family,
c53ed742 1180 tcp_metrics_nl_ops);
d23ff701 1181 if (ret < 0)
6493517e 1182 panic("Could not register tcp_metrics generic netlink\n");
51c5d0c4 1183}
Linux 6.x block layer and io_uring tree(s)