projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
vxlan: fix indentation
[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;
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		 41struct tcp_metrics_block {
42 struct tcp_metrics_block __rcu *tcpm_next;
849e8a0c 43 possible_net_t tcpm_net;
a5443028 44 struct inetpeer_addr tcpm_saddr;
324fd55a 45 struct inetpeer_addr tcpm_daddr;
51c5d0c4 46 unsigned long tcpm_stamp;
81166dd6
DM		 47 u32 tcpm_ts;
48 u32 tcpm_ts_stamp;
51c5d0c4 49 u32 tcpm_lock;
740b0f18 50 u32 tcpm_vals[TCP_METRIC_MAX_KERNEL + 1];
1fe4c481 51 struct tcp_fastopen_metrics tcpm_fastopen;
d23ff701
JA		 52
53 struct rcu_head rcu_head;
51c5d0c4
DM		 54};
55
849e8a0c
EB		 56static inline struct net *tm_net(struct tcp_metrics_block *tm)
57{
58 return read_pnet(&tm->tcpm_net);
59}
60
51c5d0c4
DM		 61static bool tcp_metric_locked(struct tcp_metrics_block *tm,
62 enum tcp_metric_index idx)
63{
64 return tm->tcpm_lock & (1 << idx);
65}
66
67static u32 tcp_metric_get(struct tcp_metrics_block *tm,
68 enum tcp_metric_index idx)
69{
70 return tm->tcpm_vals[idx];
71}
72
51c5d0c4
DM		 73static void tcp_metric_set(struct tcp_metrics_block *tm,
74 enum tcp_metric_index idx,
75 u32 val)
76{
77 tm->tcpm_vals[idx] = val;
78}
79
51c5d0c4
DM		 80static bool addr_same(const struct inetpeer_addr *a,
81 const struct inetpeer_addr *b)
82{
83 const struct in6_addr *a6, *b6;
84
85 if (a->family != b->family)
86 return false;
87 if (a->family == AF_INET)
88 return a->addr.a4 == b->addr.a4;
89
90 a6 = (const struct in6_addr *) &a->addr.a6[0];
91 b6 = (const struct in6_addr *) &b->addr.a6[0];
92
93 return ipv6_addr_equal(a6, b6);
94}
95
96struct tcpm_hash_bucket {
97 struct tcp_metrics_block __rcu *chain;
98};
99
098a697b
EB		 100static struct tcpm_hash_bucket *tcp_metrics_hash __read_mostly;
101static unsigned int tcp_metrics_hash_log __read_mostly;
102
51c5d0c4
DM		 103static DEFINE_SPINLOCK(tcp_metrics_lock);
104
740b0f18
ED		 105static void tcpm_suck_dst(struct tcp_metrics_block *tm,
106 const struct dst_entry *dst,
efeaa555 107 bool fastopen_clear)
51c5d0c4 108{
740b0f18 109 u32 msval;
51c5d0c4
DM		 110 u32 val;
111
9a0a9502
JA		 112 tm->tcpm_stamp = jiffies;
113
51c5d0c4
DM		 114 val = 0;
115 if (dst_metric_locked(dst, RTAX_RTT))
116 val |= 1 << TCP_METRIC_RTT;
117 if (dst_metric_locked(dst, RTAX_RTTVAR))
118 val |= 1 << TCP_METRIC_RTTVAR;
119 if (dst_metric_locked(dst, RTAX_SSTHRESH))
120 val |= 1 << TCP_METRIC_SSTHRESH;
121 if (dst_metric_locked(dst, RTAX_CWND))
122 val |= 1 << TCP_METRIC_CWND;
123 if (dst_metric_locked(dst, RTAX_REORDERING))
124 val |= 1 << TCP_METRIC_REORDERING;
125 tm->tcpm_lock = val;
126
740b0f18
ED		 127 msval = dst_metric_raw(dst, RTAX_RTT);
128 tm->tcpm_vals[TCP_METRIC_RTT] = msval * USEC_PER_MSEC;
129
130 msval = dst_metric_raw(dst, RTAX_RTTVAR);
131 tm->tcpm_vals[TCP_METRIC_RTTVAR] = msval * USEC_PER_MSEC;
51c5d0c4
DM		 132 tm->tcpm_vals[TCP_METRIC_SSTHRESH] = dst_metric_raw(dst, RTAX_SSTHRESH);
133 tm->tcpm_vals[TCP_METRIC_CWND] = dst_metric_raw(dst, RTAX_CWND);
134 tm->tcpm_vals[TCP_METRIC_REORDERING] = dst_metric_raw(dst, RTAX_REORDERING);
81166dd6
DM		 135 tm->tcpm_ts = 0;
136 tm->tcpm_ts_stamp = 0;
efeaa555
ED		 137 if (fastopen_clear) {
138 tm->tcpm_fastopen.mss = 0;
139 tm->tcpm_fastopen.syn_loss = 0;
140 tm->tcpm_fastopen.cookie.len = 0;
141 }
51c5d0c4
DM		 142}
143
77f99ad1
CP		 144#define TCP_METRICS_TIMEOUT (60 * 60 * HZ)
145
146static void tcpm_check_stamp(struct tcp_metrics_block *tm, struct dst_entry *dst)
147{
148 if (tm && unlikely(time_after(jiffies, tm->tcpm_stamp + TCP_METRICS_TIMEOUT)))
149 tcpm_suck_dst(tm, dst, false);
150}
151
152#define TCP_METRICS_RECLAIM_DEPTH 5
153#define TCP_METRICS_RECLAIM_PTR (struct tcp_metrics_block *) 0x1UL
154
9f1ab186
ED		 155#define deref_locked(p) \
156 rcu_dereference_protected(p, lockdep_is_held(&tcp_metrics_lock))
157
51c5d0c4 158static struct tcp_metrics_block *tcpm_new(struct dst_entry *dst,
a5443028 159 struct inetpeer_addr *saddr,
324fd55a 160 struct inetpeer_addr *daddr,
77f99ad1 161 unsigned int hash)
51c5d0c4
DM		 162{
163 struct tcp_metrics_block *tm;
164 struct net *net;
77f99ad1 165 bool reclaim = false;
51c5d0c4
DM		 166
167 spin_lock_bh(&tcp_metrics_lock);
168 net = dev_net(dst->dev);
77f99ad1
CP		 169
170 /* While waiting for the spin-lock the cache might have been populated
171 * with this entry and so we have to check again.
172 */
41804420 173 tm = __tcp_get_metrics(saddr, daddr, net, hash);
77f99ad1
CP		 174 if (tm == TCP_METRICS_RECLAIM_PTR) {
175 reclaim = true;
176 tm = NULL;
177 }
178 if (tm) {
179 tcpm_check_stamp(tm, dst);
180 goto out_unlock;
181 }
182
51c5d0c4
DM		 183 if (unlikely(reclaim)) {
184 struct tcp_metrics_block *oldest;
185
9f1ab186
ED		 186 oldest = deref_locked(tcp_metrics_hash[hash].chain);
187 for (tm = deref_locked(oldest->tcpm_next); tm;
188 tm = deref_locked(tm->tcpm_next)) {
51c5d0c4
DM		 189 if (time_before(tm->tcpm_stamp, oldest->tcpm_stamp))
190 oldest = tm;
191 }
192 tm = oldest;
193 } else {
194 tm = kmalloc(sizeof(*tm), GFP_ATOMIC);
195 if (!tm)
196 goto out_unlock;
197 }
849e8a0c 198 write_pnet(&tm->tcpm_net, net);
a5443028 199 tm->tcpm_saddr = *saddr;
324fd55a 200 tm->tcpm_daddr = *daddr;
51c5d0c4 201
efeaa555 202 tcpm_suck_dst(tm, dst, true);
51c5d0c4
DM		 203
204 if (likely(!reclaim)) {
098a697b
EB		 205 tm->tcpm_next = tcp_metrics_hash[hash].chain;
206 rcu_assign_pointer(tcp_metrics_hash[hash].chain, tm);
51c5d0c4
DM		 207 }
208
209out_unlock:
210 spin_unlock_bh(&tcp_metrics_lock);
211 return tm;
212}
213
51c5d0c4
DM		 214static struct tcp_metrics_block *tcp_get_encode(struct tcp_metrics_block *tm, int depth)
215{
216 if (tm)
217 return tm;
218 if (depth > TCP_METRICS_RECLAIM_DEPTH)
219 return TCP_METRICS_RECLAIM_PTR;
220 return NULL;
221}
222
a5443028
CP		 223static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr,
224 const struct inetpeer_addr *daddr,
51c5d0c4
DM		 225 struct net *net, unsigned int hash)
226{
227 struct tcp_metrics_block *tm;
228 int depth = 0;
229
098a697b 230 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
51c5d0c4 231 tm = rcu_dereference(tm->tcpm_next)) {
a5443028 232 if (addr_same(&tm->tcpm_saddr, saddr) &&
849e8a0c
EB		 233 addr_same(&tm->tcpm_daddr, daddr) &&
234 net_eq(tm_net(tm), net))
51c5d0c4
DM		 235 break;
236 depth++;
237 }
238 return tcp_get_encode(tm, depth);
239}
240
241static struct tcp_metrics_block *__tcp_get_metrics_req(struct request_sock *req,
242 struct dst_entry *dst)
243{
244 struct tcp_metrics_block *tm;
a5443028 245 struct inetpeer_addr saddr, daddr;
51c5d0c4
DM		 246 unsigned int hash;
247 struct net *net;
248
a5443028 249 saddr.family = req->rsk_ops->family;
324fd55a
CP		 250 daddr.family = req->rsk_ops->family;
251 switch (daddr.family) {
51c5d0c4 252 case AF_INET:
a5443028 253 saddr.addr.a4 = inet_rsk(req)->ir_loc_addr;
324fd55a
CP		 254 daddr.addr.a4 = inet_rsk(req)->ir_rmt_addr;
255 hash = (__force unsigned int) daddr.addr.a4;
51c5d0c4 256 break;
634fb979 257#if IS_ENABLED(CONFIG_IPV6)
51c5d0c4 258 case AF_INET6:
a5443028 259 *(struct in6_addr *)saddr.addr.a6 = inet_rsk(req)->ir_v6_loc_addr;
324fd55a 260 *(struct in6_addr *)daddr.addr.a6 = inet_rsk(req)->ir_v6_rmt_addr;
634fb979 261 hash = ipv6_addr_hash(&inet_rsk(req)->ir_v6_rmt_addr);
51c5d0c4 262 break;
634fb979 263#endif
51c5d0c4
DM		 264 default:
265 return NULL;
266 }
267
51c5d0c4 268 net = dev_net(dst->dev);
3e5da62d 269 hash ^= net_hash_mix(net);
098a697b 270 hash = hash_32(hash, tcp_metrics_hash_log);
51c5d0c4 271
098a697b 272 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
51c5d0c4 273 tm = rcu_dereference(tm->tcpm_next)) {
a5443028 274 if (addr_same(&tm->tcpm_saddr, &saddr) &&
849e8a0c
EB		 275 addr_same(&tm->tcpm_daddr, &daddr) &&
276 net_eq(tm_net(tm), net))
51c5d0c4
DM		 277 break;
278 }
279 tcpm_check_stamp(tm, dst);
280 return tm;
281}
282
81166dd6
DM		 283static struct tcp_metrics_block *__tcp_get_metrics_tw(struct inet_timewait_sock *tw)
284{
81166dd6 285 struct tcp_metrics_block *tm;
a5443028 286 struct inetpeer_addr saddr, daddr;
81166dd6
DM		 287 unsigned int hash;
288 struct net *net;
289
3ad88cf7
CP		 290 if (tw->tw_family == AF_INET) {
291 saddr.family = AF_INET;
a5443028 292 saddr.addr.a4 = tw->tw_rcv_saddr;
3ad88cf7 293 daddr.family = AF_INET;
324fd55a
CP		 294 daddr.addr.a4 = tw->tw_daddr;
295 hash = (__force unsigned int) daddr.addr.a4;
3ad88cf7 296 }
c2bb06db 297#if IS_ENABLED(CONFIG_IPV6)
3ad88cf7
CP		 298 else if (tw->tw_family == AF_INET6) {
299 if (ipv6_addr_v4mapped(&tw->tw_v6_daddr)) {
300 saddr.family = AF_INET;
301 saddr.addr.a4 = tw->tw_rcv_saddr;
302 daddr.family = AF_INET;
303 daddr.addr.a4 = tw->tw_daddr;
304 hash = (__force unsigned int) daddr.addr.a4;
305 } else {
306 saddr.family = AF_INET6;
307 *(struct in6_addr *)saddr.addr.a6 = tw->tw_v6_rcv_saddr;
308 daddr.family = AF_INET6;
309 *(struct in6_addr *)daddr.addr.a6 = tw->tw_v6_daddr;
310 hash = ipv6_addr_hash(&tw->tw_v6_daddr);
311 }
312 }
c2bb06db 313#endif
3ad88cf7 314 else
81166dd6 315 return NULL;
81166dd6 316
81166dd6 317 net = twsk_net(tw);
3e5da62d 318 hash ^= net_hash_mix(net);
098a697b 319 hash = hash_32(hash, tcp_metrics_hash_log);
81166dd6 320
098a697b 321 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
81166dd6 322 tm = rcu_dereference(tm->tcpm_next)) {
a5443028 323 if (addr_same(&tm->tcpm_saddr, &saddr) &&
849e8a0c
EB		 324 addr_same(&tm->tcpm_daddr, &daddr) &&
325 net_eq(tm_net(tm), net))
81166dd6
DM		 326 break;
327 }
328 return tm;
329}
330
51c5d0c4
DM		 331static struct tcp_metrics_block *tcp_get_metrics(struct sock *sk,
332 struct dst_entry *dst,
333 bool create)
334{
335 struct tcp_metrics_block *tm;
a5443028 336 struct inetpeer_addr saddr, daddr;
51c5d0c4
DM		 337 unsigned int hash;
338 struct net *net;
51c5d0c4 339
3ad88cf7
CP		 340 if (sk->sk_family == AF_INET) {
341 saddr.family = AF_INET;
a5443028 342 saddr.addr.a4 = inet_sk(sk)->inet_saddr;
3ad88cf7 343 daddr.family = AF_INET;
324fd55a
CP		 344 daddr.addr.a4 = inet_sk(sk)->inet_daddr;
345 hash = (__force unsigned int) daddr.addr.a4;
3ad88cf7 346 }
c2bb06db 347#if IS_ENABLED(CONFIG_IPV6)
3ad88cf7
CP		 348 else if (sk->sk_family == AF_INET6) {
349 if (ipv6_addr_v4mapped(&sk->sk_v6_daddr)) {
350 saddr.family = AF_INET;
351 saddr.addr.a4 = inet_sk(sk)->inet_saddr;
352 daddr.family = AF_INET;
353 daddr.addr.a4 = inet_sk(sk)->inet_daddr;
354 hash = (__force unsigned int) daddr.addr.a4;
355 } else {
356 saddr.family = AF_INET6;
357 *(struct in6_addr *)saddr.addr.a6 = sk->sk_v6_rcv_saddr;
358 daddr.family = AF_INET6;
359 *(struct in6_addr *)daddr.addr.a6 = sk->sk_v6_daddr;
360 hash = ipv6_addr_hash(&sk->sk_v6_daddr);
361 }
362 }
c2bb06db 363#endif
3ad88cf7 364 else
51c5d0c4 365 return NULL;
51c5d0c4 366
51c5d0c4 367 net = dev_net(dst->dev);
3e5da62d 368 hash ^= net_hash_mix(net);
098a697b 369 hash = hash_32(hash, tcp_metrics_hash_log);
51c5d0c4 370
a5443028 371 tm = __tcp_get_metrics(&saddr, &daddr, net, hash);
77f99ad1 372 if (tm == TCP_METRICS_RECLAIM_PTR)
51c5d0c4 373 tm = NULL;
51c5d0c4 374 if (!tm && create)
41804420 375 tm = tcpm_new(dst, &saddr, &daddr, hash);
51c5d0c4
DM		 376 else
377 tcpm_check_stamp(tm, dst);
378
379 return tm;
380}
381
4aabd8ef
DM		 382/* Save metrics learned by this TCP session. This function is called
383 * only, when TCP finishes successfully i.e. when it enters TIME-WAIT
384 * or goes from LAST-ACK to CLOSE.
385 */
386void tcp_update_metrics(struct sock *sk)
387{
51c5d0c4 388 const struct inet_connection_sock *icsk = inet_csk(sk);
4aabd8ef 389 struct dst_entry *dst = __sk_dst_get(sk);
51c5d0c4
DM		 390 struct tcp_sock *tp = tcp_sk(sk);
391 struct tcp_metrics_block *tm;
392 unsigned long rtt;
393 u32 val;
394 int m;
4aabd8ef 395
51c5d0c4 396 if (sysctl_tcp_nometrics_save || !dst)
4aabd8ef
DM		 397 return;
398
51c5d0c4 399 if (dst->flags & DST_HOST)
4aabd8ef
DM		 400 dst_confirm(dst);
401
51c5d0c4 402 rcu_read_lock();
740b0f18 403 if (icsk->icsk_backoff || !tp->srtt_us) {
51c5d0c4
DM		 404 /* This session failed to estimate rtt. Why?
405 * Probably, no packets returned in time. Reset our
406 * results.
407 */
408 tm = tcp_get_metrics(sk, dst, false);
409 if (tm && !tcp_metric_locked(tm, TCP_METRIC_RTT))
410 tcp_metric_set(tm, TCP_METRIC_RTT, 0);
411 goto out_unlock;
412 } else
413 tm = tcp_get_metrics(sk, dst, true);
4aabd8ef 414
51c5d0c4
DM		 415 if (!tm)
416 goto out_unlock;
4aabd8ef 417
740b0f18
ED		 418 rtt = tcp_metric_get(tm, TCP_METRIC_RTT);
419 m = rtt - tp->srtt_us;
4aabd8ef 420
51c5d0c4
DM		 421 /* If newly calculated rtt larger than stored one, store new
422 * one. Otherwise, use EWMA. Remember, rtt overestimation is
423 * always better than underestimation.
424 */
425 if (!tcp_metric_locked(tm, TCP_METRIC_RTT)) {
426 if (m <= 0)
740b0f18 427 rtt = tp->srtt_us;
51c5d0c4
DM		 428 else
429 rtt -= (m >> 3);
740b0f18 430 tcp_metric_set(tm, TCP_METRIC_RTT, rtt);
51c5d0c4 431 }
4aabd8ef 432
51c5d0c4
DM		 433 if (!tcp_metric_locked(tm, TCP_METRIC_RTTVAR)) {
434 unsigned long var;
4aabd8ef 435
51c5d0c4
DM		 436 if (m < 0)
437 m = -m;
4aabd8ef 438
51c5d0c4
DM		 439 /* Scale deviation to rttvar fixed point */
440 m >>= 1;
740b0f18
ED		 441 if (m < tp->mdev_us)
442 m = tp->mdev_us;
4aabd8ef 443
740b0f18 444 var = tcp_metric_get(tm, TCP_METRIC_RTTVAR);
51c5d0c4
DM		 445 if (m >= var)
446 var = m;
447 else
448 var -= (var - m) >> 2;
4aabd8ef 449
740b0f18 450 tcp_metric_set(tm, TCP_METRIC_RTTVAR, var);
51c5d0c4
DM		 451 }
452
453 if (tcp_in_initial_slowstart(tp)) {
454 /* Slow start still did not finish. */
455 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) {
456 val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
457 if (val && (tp->snd_cwnd >> 1) > val)
458 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
459 tp->snd_cwnd >> 1);
460 }
461 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
462 val = tcp_metric_get(tm, TCP_METRIC_CWND);
463 if (tp->snd_cwnd > val)
464 tcp_metric_set(tm, TCP_METRIC_CWND,
465 tp->snd_cwnd);
466 }
467 } else if (tp->snd_cwnd > tp->snd_ssthresh &&
468 icsk->icsk_ca_state == TCP_CA_Open) {
469 /* Cong. avoidance phase, cwnd is reliable. */
470 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH))
471 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
472 max(tp->snd_cwnd >> 1, tp->snd_ssthresh));
473 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
474 val = tcp_metric_get(tm, TCP_METRIC_CWND);
2100844c 475 tcp_metric_set(tm, TCP_METRIC_CWND, (val + tp->snd_cwnd) >> 1);
51c5d0c4
DM		 476 }
477 } else {
478 /* Else slow start did not finish, cwnd is non-sense,
479 * ssthresh may be also invalid.
480 */
481 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
482 val = tcp_metric_get(tm, TCP_METRIC_CWND);
483 tcp_metric_set(tm, TCP_METRIC_CWND,
484 (val + tp->snd_ssthresh) >> 1);
485 }
486 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) {
487 val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
488 if (val && tp->snd_ssthresh > val)
489 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
490 tp->snd_ssthresh);
491 }
492 if (!tcp_metric_locked(tm, TCP_METRIC_REORDERING)) {
493 val = tcp_metric_get(tm, TCP_METRIC_REORDERING);
494 if (val < tp->reordering &&
4aabd8ef 495 tp->reordering != sysctl_tcp_reordering)
51c5d0c4
DM		 496 tcp_metric_set(tm, TCP_METRIC_REORDERING,
497 tp->reordering);
4aabd8ef
DM		 498 }
499 }
51c5d0c4
DM		 500 tm->tcpm_stamp = jiffies;
501out_unlock:
502 rcu_read_unlock();
4aabd8ef
DM		 503}
504
505/* Initialize metrics on socket. */
506
507void tcp_init_metrics(struct sock *sk)
508{
4aabd8ef 509 struct dst_entry *dst = __sk_dst_get(sk);
51c5d0c4
DM		 510 struct tcp_sock *tp = tcp_sk(sk);
511 struct tcp_metrics_block *tm;
1b7fdd2a 512 u32 val, crtt = 0; /* cached RTT scaled by 8 */
4aabd8ef
DM		 513
514 if (dst == NULL)
515 goto reset;
516
517 dst_confirm(dst);
518
51c5d0c4
DM		 519 rcu_read_lock();
520 tm = tcp_get_metrics(sk, dst, true);
521 if (!tm) {
522 rcu_read_unlock();
523 goto reset;
524 }
525
526 if (tcp_metric_locked(tm, TCP_METRIC_CWND))
527 tp->snd_cwnd_clamp = tcp_metric_get(tm, TCP_METRIC_CWND);
528
529 val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
530 if (val) {
531 tp->snd_ssthresh = val;
4aabd8ef
DM		 532 if (tp->snd_ssthresh > tp->snd_cwnd_clamp)
533 tp->snd_ssthresh = tp->snd_cwnd_clamp;
534 } else {
535 /* ssthresh may have been reduced unnecessarily during.
536 * 3WHS. Restore it back to its initial default.
537 */
538 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
539 }
51c5d0c4
DM		 540 val = tcp_metric_get(tm, TCP_METRIC_REORDERING);
541 if (val && tp->reordering != val) {
4aabd8ef
DM		 542 tcp_disable_fack(tp);
543 tcp_disable_early_retrans(tp);
51c5d0c4 544 tp->reordering = val;
4aabd8ef
DM		 545 }
546
740b0f18 547 crtt = tcp_metric_get(tm, TCP_METRIC_RTT);
51c5d0c4 548 rcu_read_unlock();
4aabd8ef 549reset:
52f20e65
YC		 550 /* The initial RTT measurement from the SYN/SYN-ACK is not ideal
551 * to seed the RTO for later data packets because SYN packets are
552 * small. Use the per-dst cached values to seed the RTO but keep
553 * the RTT estimator variables intact (e.g., srtt, mdev, rttvar).
554 * Later the RTO will be updated immediately upon obtaining the first
555 * data RTT sample (tcp_rtt_estimator()). Hence the cached RTT only
556 * influences the first RTO but not later RTT estimation.
557 *
558 * But if RTT is not available from the SYN (due to retransmits or
559 * syn cookies) or the cache, force a conservative 3secs timeout.
560 *
561 * A bit of theory. RTT is time passed after "normal" sized packet
562 * is sent until it is ACKed. In normal circumstances sending small
563 * packets force peer to delay ACKs and calculation is correct too.
564 * The algorithm is adaptive and, provided we follow specs, it
565 * NEVER underestimate RTT. BUT! If peer tries to make some clever
566 * tricks sort of "quick acks" for time long enough to decrease RTT
567 * to low value, and then abruptly stops to do it and starts to delay
568 * ACKs, wait for troubles.
569 */
740b0f18 570 if (crtt > tp->srtt_us) {
269aa759 571 /* Set RTO like tcp_rtt_estimator(), but from cached RTT. */
740b0f18 572 crtt /= 8 * USEC_PER_MSEC;
269aa759 573 inet_csk(sk)->icsk_rto = crtt + max(2 * crtt, tcp_rto_min(sk));
740b0f18 574 } else if (tp->srtt_us == 0) {
4aabd8ef
DM		 575 /* RFC6298: 5.7 We've failed to get a valid RTT sample from
576 * 3WHS. This is most likely due to retransmission,
577 * including spurious one. Reset the RTO back to 3secs
578 * from the more aggressive 1sec to avoid more spurious
579 * retransmission.
580 */
740b0f18
ED		 581 tp->rttvar_us = jiffies_to_usecs(TCP_TIMEOUT_FALLBACK);
582 tp->mdev_us = tp->mdev_max_us = tp->rttvar_us;
583
4aabd8ef
DM		 584 inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK;
585 }
586 /* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
587 * retransmitted. In light of RFC6298 more aggressive 1sec
588 * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
589 * retransmission has occurred.
590 */
591 if (tp->total_retrans > 1)
592 tp->snd_cwnd = 1;
593 else
594 tp->snd_cwnd = tcp_init_cwnd(tp, dst);
595 tp->snd_cwnd_stamp = tcp_time_stamp;
596}
ab92bb2f 597
a26552af
HFS		 598bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
599 bool paws_check, bool timestamps)
ab92bb2f 600{
51c5d0c4
DM		 601 struct tcp_metrics_block *tm;
602 bool ret;
603
ab92bb2f
DM		 604 if (!dst)
605 return false;
51c5d0c4
DM		 606
607 rcu_read_lock();
608 tm = __tcp_get_metrics_req(req, dst);
81166dd6
DM		 609 if (paws_check) {
610 if (tm &&
611 (u32)get_seconds() - tm->tcpm_ts_stamp < TCP_PAWS_MSL &&
a26552af
HFS		 612 ((s32)(tm->tcpm_ts - req->ts_recent) > TCP_PAWS_WINDOW ||
613 !timestamps))
81166dd6
DM		 614 ret = false;
615 else
616 ret = true;
617 } else {
618 if (tm && tcp_metric_get(tm, TCP_METRIC_RTT) && tm->tcpm_ts_stamp)
619 ret = true;
620 else
621 ret = false;
622 }
51c5d0c4
DM		 623 rcu_read_unlock();
624
625 return ret;
ab92bb2f
DM		 626}
627EXPORT_SYMBOL_GPL(tcp_peer_is_proven);
51c5d0c4 628
81166dd6
DM		 629void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst)
630{
631 struct tcp_metrics_block *tm;
632
633 rcu_read_lock();
634 tm = tcp_get_metrics(sk, dst, true);
635 if (tm) {
636 struct tcp_sock *tp = tcp_sk(sk);
637
638 if ((u32)get_seconds() - tm->tcpm_ts_stamp <= TCP_PAWS_MSL) {
639 tp->rx_opt.ts_recent_stamp = tm->tcpm_ts_stamp;
640 tp->rx_opt.ts_recent = tm->tcpm_ts;
641 }
642 }
643 rcu_read_unlock();
644}
645EXPORT_SYMBOL_GPL(tcp_fetch_timewait_stamp);
646
647/* VJ's idea. Save last timestamp seen from this destination and hold
648 * it at least for normal timewait interval to use for duplicate
649 * segment detection in subsequent connections, before they enter
650 * synchronized state.
651 */
652bool tcp_remember_stamp(struct sock *sk)
653{
654 struct dst_entry *dst = __sk_dst_get(sk);
655 bool ret = false;
656
657 if (dst) {
658 struct tcp_metrics_block *tm;
659
660 rcu_read_lock();
661 tm = tcp_get_metrics(sk, dst, true);
662 if (tm) {
663 struct tcp_sock *tp = tcp_sk(sk);
664
665 if ((s32)(tm->tcpm_ts - tp->rx_opt.ts_recent) <= 0 ||
666 ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL &&
667 tm->tcpm_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) {
668 tm->tcpm_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp;
669 tm->tcpm_ts = tp->rx_opt.ts_recent;
670 }
671 ret = true;
672 }
673 rcu_read_unlock();
674 }
675 return ret;
676}
677
678bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw)
679{
680 struct tcp_metrics_block *tm;
681 bool ret = false;
682
683 rcu_read_lock();
684 tm = __tcp_get_metrics_tw(tw);
9a0a9502 685 if (tm) {
81166dd6
DM		 686 const struct tcp_timewait_sock *tcptw;
687 struct sock *sk = (struct sock *) tw;
688
689 tcptw = tcp_twsk(sk);
690 if ((s32)(tm->tcpm_ts - tcptw->tw_ts_recent) <= 0 ||
691 ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL &&
692 tm->tcpm_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) {
693 tm->tcpm_ts_stamp = (u32)tcptw->tw_ts_recent_stamp;
694 tm->tcpm_ts = tcptw->tw_ts_recent;
695 }
696 ret = true;
697 }
698 rcu_read_unlock();
699
700 return ret;
701}
702
1fe4c481
YC		 703static DEFINE_SEQLOCK(fastopen_seqlock);
704
705void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
aab48743
YC		 706 struct tcp_fastopen_cookie *cookie,
707 int *syn_loss, unsigned long *last_syn_loss)
1fe4c481
YC		 708{
709 struct tcp_metrics_block *tm;
710
711 rcu_read_lock();
712 tm = tcp_get_metrics(sk, __sk_dst_get(sk), false);
713 if (tm) {
714 struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen;
715 unsigned int seq;
716
717 do {
718 seq = read_seqbegin(&fastopen_seqlock);
719 if (tfom->mss)
720 *mss = tfom->mss;
721 *cookie = tfom->cookie;
aab48743
YC		 722 *syn_loss = tfom->syn_loss;
723 *last_syn_loss = *syn_loss ? tfom->last_syn_loss : 0;
1fe4c481
YC		 724 } while (read_seqretry(&fastopen_seqlock, seq));
725 }
726 rcu_read_unlock();
727}
728
1fe4c481 729void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
aab48743 730 struct tcp_fastopen_cookie *cookie, bool syn_lost)
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;
aab48743
YC		 747 if (syn_lost) {
748 ++tfom->syn_loss;
749 tfom->last_syn_loss = jiffies;
750 } else
751 tfom->syn_loss = 0;
1fe4c481
YC		 752 write_sequnlock_bh(&fastopen_seqlock);
753 }
754 rcu_read_unlock();
755}
756
d23ff701
JA		 757static struct genl_family tcp_metrics_nl_family = {
758 .id = GENL_ID_GENERATE,
759 .hdrsize = 0,
760 .name = TCP_METRICS_GENL_NAME,
761 .version = TCP_METRICS_GENL_VERSION,
762 .maxattr = TCP_METRICS_ATTR_MAX,
763 .netnsok = true,
764};
765
766static struct nla_policy tcp_metrics_nl_policy[TCP_METRICS_ATTR_MAX + 1] = {
767 [TCP_METRICS_ATTR_ADDR_IPV4] = { .type = NLA_U32, },
768 [TCP_METRICS_ATTR_ADDR_IPV6] = { .type = NLA_BINARY,
769 .len = sizeof(struct in6_addr), },
770 /* Following attributes are not received for GET/DEL,
771 * we keep them for reference
772 */
773#if 0
774 [TCP_METRICS_ATTR_AGE] = { .type = NLA_MSECS, },
775 [TCP_METRICS_ATTR_TW_TSVAL] = { .type = NLA_U32, },
776 [TCP_METRICS_ATTR_TW_TS_STAMP] = { .type = NLA_S32, },
777 [TCP_METRICS_ATTR_VALS] = { .type = NLA_NESTED, },
778 [TCP_METRICS_ATTR_FOPEN_MSS] = { .type = NLA_U16, },
779 [TCP_METRICS_ATTR_FOPEN_SYN_DROPS] = { .type = NLA_U16, },
780 [TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS] = { .type = NLA_MSECS, },
781 [TCP_METRICS_ATTR_FOPEN_COOKIE] = { .type = NLA_BINARY,
782 .len = TCP_FASTOPEN_COOKIE_MAX, },
783#endif
784};
785
786/* Add attributes, caller cancels its header on failure */
787static int tcp_metrics_fill_info(struct sk_buff *msg,
788 struct tcp_metrics_block *tm)
789{
790 struct nlattr *nest;
791 int i;
792
324fd55a 793 switch (tm->tcpm_daddr.family) {
d23ff701
JA		 794 case AF_INET:
795 if (nla_put_be32(msg, TCP_METRICS_ATTR_ADDR_IPV4,
324fd55a 796 tm->tcpm_daddr.addr.a4) < 0)
d23ff701 797 goto nla_put_failure;
8a59359c
CP		 798 if (nla_put_be32(msg, TCP_METRICS_ATTR_SADDR_IPV4,
799 tm->tcpm_saddr.addr.a4) < 0)
800 goto nla_put_failure;
d23ff701
JA		 801 break;
802 case AF_INET6:
803 if (nla_put(msg, TCP_METRICS_ATTR_ADDR_IPV6, 16,
324fd55a 804 tm->tcpm_daddr.addr.a6) < 0)
d23ff701 805 goto nla_put_failure;
8a59359c
CP		 806 if (nla_put(msg, TCP_METRICS_ATTR_SADDR_IPV6, 16,
807 tm->tcpm_saddr.addr.a6) < 0)
808 goto nla_put_failure;
d23ff701
JA		 809 break;
810 default:
811 return -EAFNOSUPPORT;
812 }
813
814 if (nla_put_msecs(msg, TCP_METRICS_ATTR_AGE,
815 jiffies - tm->tcpm_stamp) < 0)
816 goto nla_put_failure;
817 if (tm->tcpm_ts_stamp) {
818 if (nla_put_s32(msg, TCP_METRICS_ATTR_TW_TS_STAMP,
819 (s32) (get_seconds() - tm->tcpm_ts_stamp)) < 0)
820 goto nla_put_failure;
821 if (nla_put_u32(msg, TCP_METRICS_ATTR_TW_TSVAL,
822 tm->tcpm_ts) < 0)
823 goto nla_put_failure;
824 }
825
826 {
827 int n = 0;
828
829 nest = nla_nest_start(msg, TCP_METRICS_ATTR_VALS);
830 if (!nest)
831 goto nla_put_failure;
740b0f18
ED		 832 for (i = 0; i < TCP_METRIC_MAX_KERNEL + 1; i++) {
833 u32 val = tm->tcpm_vals[i];
834
835 if (!val)
d23ff701 836 continue;
740b0f18
ED		 837 if (i == TCP_METRIC_RTT) {
838 if (nla_put_u32(msg, TCP_METRIC_RTT_US + 1,
839 val) < 0)
840 goto nla_put_failure;
841 n++;
842 val = max(val / 1000, 1U);
843 }
844 if (i == TCP_METRIC_RTTVAR) {
845 if (nla_put_u32(msg, TCP_METRIC_RTTVAR_US + 1,
846 val) < 0)
847 goto nla_put_failure;
848 n++;
849 val = max(val / 1000, 1U);
850 }
851 if (nla_put_u32(msg, i + 1, val) < 0)
d23ff701
JA		 852 goto nla_put_failure;
853 n++;
854 }
855 if (n)
856 nla_nest_end(msg, nest);
857 else
858 nla_nest_cancel(msg, nest);
859 }
860
861 {
862 struct tcp_fastopen_metrics tfom_copy[1], *tfom;
863 unsigned int seq;
864
865 do {
866 seq = read_seqbegin(&fastopen_seqlock);
867 tfom_copy[0] = tm->tcpm_fastopen;
868 } while (read_seqretry(&fastopen_seqlock, seq));
869
870 tfom = tfom_copy;
871 if (tfom->mss &&
872 nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_MSS,
873 tfom->mss) < 0)
874 goto nla_put_failure;
875 if (tfom->syn_loss &&
876 (nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROPS,
877 tfom->syn_loss) < 0 ||
878 nla_put_msecs(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS,
879 jiffies - tfom->last_syn_loss) < 0))
880 goto nla_put_failure;
881 if (tfom->cookie.len > 0 &&
882 nla_put(msg, TCP_METRICS_ATTR_FOPEN_COOKIE,
883 tfom->cookie.len, tfom->cookie.val) < 0)
884 goto nla_put_failure;
885 }
886
887 return 0;
888
889nla_put_failure:
890 return -EMSGSIZE;
891}
892
893static int tcp_metrics_dump_info(struct sk_buff *skb,
894 struct netlink_callback *cb,
895 struct tcp_metrics_block *tm)
896{
897 void *hdr;
898
15e47304 899 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
d23ff701
JA		 900 &tcp_metrics_nl_family, NLM_F_MULTI,
901 TCP_METRICS_CMD_GET);
902 if (!hdr)
903 return -EMSGSIZE;
904
905 if (tcp_metrics_fill_info(skb, tm) < 0)
906 goto nla_put_failure;
907
053c095a
JB		 908 genlmsg_end(skb, hdr);
909 return 0;
d23ff701
JA		 910
911nla_put_failure:
912 genlmsg_cancel(skb, hdr);
913 return -EMSGSIZE;
914}
915
916static int tcp_metrics_nl_dump(struct sk_buff *skb,
917 struct netlink_callback *cb)
918{
919 struct net *net = sock_net(skb->sk);
098a697b 920 unsigned int max_rows = 1U << tcp_metrics_hash_log;
d23ff701
JA		 921 unsigned int row, s_row = cb->args[0];
922 int s_col = cb->args[1], col = s_col;
923
924 for (row = s_row; row < max_rows; row++, s_col = 0) {
925 struct tcp_metrics_block *tm;
098a697b 926 struct tcpm_hash_bucket *hb = tcp_metrics_hash + row;
d23ff701
JA		 927
928 rcu_read_lock();
929 for (col = 0, tm = rcu_dereference(hb->chain); tm;
930 tm = rcu_dereference(tm->tcpm_next), col++) {
849e8a0c
EB		 931 if (!net_eq(tm_net(tm), net))
932 continue;
d23ff701
JA		 933 if (col < s_col)
934 continue;
935 if (tcp_metrics_dump_info(skb, cb, tm) < 0) {
936 rcu_read_unlock();
937 goto done;
938 }
939 }
940 rcu_read_unlock();
941 }
942
943done:
944 cb->args[0] = row;
945 cb->args[1] = col;
946 return skb->len;
947}
948
3e7013dd
CP		 949static int __parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
950 unsigned int *hash, int optional, int v4, int v6)
d23ff701
JA		 951{
952 struct nlattr *a;
953
3e7013dd 954 a = info->attrs[v4];
d23ff701
JA		 955 if (a) {
956 addr->family = AF_INET;
957 addr->addr.a4 = nla_get_be32(a);
3e7013dd
CP		 958 if (hash)
959 *hash = (__force unsigned int) addr->addr.a4;
d23ff701
JA		 960 return 0;
961 }
3e7013dd 962 a = info->attrs[v6];
d23ff701 963 if (a) {
2c42a3fb 964 if (nla_len(a) != sizeof(struct in6_addr))
d23ff701
JA		 965 return -EINVAL;
966 addr->family = AF_INET6;
967 memcpy(addr->addr.a6, nla_data(a), sizeof(addr->addr.a6));
3e7013dd
CP		 968 if (hash)
969 *hash = ipv6_addr_hash((struct in6_addr *) addr->addr.a6);
d23ff701
JA		 970 return 0;
971 }
972 return optional ? 1 : -EAFNOSUPPORT;
973}
974
3e7013dd
CP		 975static int parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
976 unsigned int *hash, int optional)
977{
978 return __parse_nl_addr(info, addr, hash, optional,
979 TCP_METRICS_ATTR_ADDR_IPV4,
980 TCP_METRICS_ATTR_ADDR_IPV6);
981}
982
983static int parse_nl_saddr(struct genl_info *info, struct inetpeer_addr *addr)
984{
985 return __parse_nl_addr(info, addr, NULL, 0,
986 TCP_METRICS_ATTR_SADDR_IPV4,
987 TCP_METRICS_ATTR_SADDR_IPV6);
988}
989
d23ff701
JA		 990static int tcp_metrics_nl_cmd_get(struct sk_buff *skb, struct genl_info *info)
991{
992 struct tcp_metrics_block *tm;
3e7013dd 993 struct inetpeer_addr saddr, daddr;
d23ff701
JA		 994 unsigned int hash;
995 struct sk_buff *msg;
996 struct net *net = genl_info_net(info);
997 void *reply;
998 int ret;
3e7013dd 999 bool src = true;
d23ff701 1000
324fd55a 1001 ret = parse_nl_addr(info, &daddr, &hash, 0);
d23ff701
JA		 1002 if (ret < 0)
1003 return ret;
1004
3e7013dd
CP		 1005 ret = parse_nl_saddr(info, &saddr);
1006 if (ret < 0)
1007 src = false;
1008
d23ff701
JA		 1009 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1010 if (!msg)
1011 return -ENOMEM;
1012
1013 reply = genlmsg_put_reply(msg, info, &tcp_metrics_nl_family, 0,
1014 info->genlhdr->cmd);
1015 if (!reply)
1016 goto nla_put_failure;
1017
3e5da62d 1018 hash ^= net_hash_mix(net);
098a697b 1019 hash = hash_32(hash, tcp_metrics_hash_log);
d23ff701
JA		 1020 ret = -ESRCH;
1021 rcu_read_lock();
098a697b 1022 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
d23ff701 1023 tm = rcu_dereference(tm->tcpm_next)) {
3e7013dd 1024 if (addr_same(&tm->tcpm_daddr, &daddr) &&
849e8a0c
EB		 1025 (!src || addr_same(&tm->tcpm_saddr, &saddr)) &&
1026 net_eq(tm_net(tm), net)) {
d23ff701
JA		 1027 ret = tcp_metrics_fill_info(msg, tm);
1028 break;
1029 }
1030 }
1031 rcu_read_unlock();
1032 if (ret < 0)
1033 goto out_free;
1034
1035 genlmsg_end(msg, reply);
1036 return genlmsg_reply(msg, info);
1037
1038nla_put_failure:
1039 ret = -EMSGSIZE;
1040
1041out_free:
1042 nlmsg_free(msg);
1043 return ret;
1044}
1045
8a4bff71 1046static void tcp_metrics_flush_all(struct net *net)
d23ff701 1047{
098a697b
EB		 1048 unsigned int max_rows = 1U << tcp_metrics_hash_log;
1049 struct tcpm_hash_bucket *hb = tcp_metrics_hash;
d23ff701
JA		 1050 struct tcp_metrics_block *tm;
1051 unsigned int row;
1052
1053 for (row = 0; row < max_rows; row++, hb++) {
04f721c6 1054 struct tcp_metrics_block __rcu **pp;
d23ff701 1055 spin_lock_bh(&tcp_metrics_lock);
04f721c6 1056 pp = &hb->chain;
9f1ab186 1057 for (tm = deref_locked(*pp); tm; tm = deref_locked(*pp)) {
04f721c6
EB		 1058 if (net_eq(tm_net(tm), net)) {
1059 *pp = tm->tcpm_next;
1060 kfree_rcu(tm, rcu_head);
1061 } else {
1062 pp = &tm->tcpm_next;
1063 }
d23ff701 1064 }
04f721c6 1065 spin_unlock_bh(&tcp_metrics_lock);
d23ff701 1066 }
d23ff701
JA		 1067}
1068
1069static int tcp_metrics_nl_cmd_del(struct sk_buff *skb, struct genl_info *info)
1070{
1071 struct tcpm_hash_bucket *hb;
00ca9c5b 1072 struct tcp_metrics_block *tm;
d23ff701 1073 struct tcp_metrics_block __rcu **pp;
3e7013dd 1074 struct inetpeer_addr saddr, daddr;
d23ff701
JA		 1075 unsigned int hash;
1076 struct net *net = genl_info_net(info);
1077 int ret;
00ca9c5b 1078 bool src = true, found = false;
d23ff701 1079
324fd55a 1080 ret = parse_nl_addr(info, &daddr, &hash, 1);
d23ff701
JA		 1081 if (ret < 0)
1082 return ret;
8a4bff71
EB		 1083 if (ret > 0) {
1084 tcp_metrics_flush_all(net);
1085 return 0;
1086 }
3e7013dd
CP		 1087 ret = parse_nl_saddr(info, &saddr);
1088 if (ret < 0)
1089 src = false;
d23ff701 1090
3e5da62d 1091 hash ^= net_hash_mix(net);
098a697b
EB		 1092 hash = hash_32(hash, tcp_metrics_hash_log);
1093 hb = tcp_metrics_hash + hash;
d23ff701
JA		 1094 pp = &hb->chain;
1095 spin_lock_bh(&tcp_metrics_lock);
9f1ab186 1096 for (tm = deref_locked(*pp); tm; tm = deref_locked(*pp)) {
3e7013dd 1097 if (addr_same(&tm->tcpm_daddr, &daddr) &&
849e8a0c
EB		 1098 (!src || addr_same(&tm->tcpm_saddr, &saddr)) &&
1099 net_eq(tm_net(tm), net)) {
d23ff701 1100 *pp = tm->tcpm_next;
00ca9c5b
CP		 1101 kfree_rcu(tm, rcu_head);
1102 found = true;
bbf852b9
CP		 1103 } else {
1104 pp = &tm->tcpm_next;
d23ff701
JA		 1105 }
1106 }
1107 spin_unlock_bh(&tcp_metrics_lock);
00ca9c5b 1108 if (!found)
d23ff701 1109 return -ESRCH;
d23ff701
JA		 1110 return 0;
1111}
1112
4534de83 1113static const struct genl_ops tcp_metrics_nl_ops[] = {
d23ff701
JA		 1114 {
1115 .cmd = TCP_METRICS_CMD_GET,
1116 .doit = tcp_metrics_nl_cmd_get,
1117 .dumpit = tcp_metrics_nl_dump,
1118 .policy = tcp_metrics_nl_policy,
d23ff701
JA		 1119 },
1120 {
1121 .cmd = TCP_METRICS_CMD_DEL,
1122 .doit = tcp_metrics_nl_cmd_del,
1123 .policy = tcp_metrics_nl_policy,
1124 .flags = GENL_ADMIN_PERM,
1125 },
1126};
1127
5815d5e7 1128static unsigned int tcpmhash_entries;
51c5d0c4
DM		 1129static int __init set_tcpmhash_entries(char *str)
1130{
1131 ssize_t ret;
1132
1133 if (!str)
1134 return 0;
1135
5815d5e7 1136 ret = kstrtouint(str, 0, &tcpmhash_entries);
51c5d0c4
DM		 1137 if (ret)
1138 return 0;
1139
1140 return 1;
1141}
1142__setup("tcpmhash_entries=", set_tcpmhash_entries);
1143
1144static int __net_init tcp_net_metrics_init(struct net *net)
1145{
5815d5e7
ED		 1146 size_t size;
1147 unsigned int slots;
51c5d0c4 1148
098a697b
EB		 1149 if (!net_eq(net, &init_net))
1150 return 0;
1151
51c5d0c4
DM		 1152 slots = tcpmhash_entries;
1153 if (!slots) {
1154 if (totalram_pages >= 128 * 1024)
1155 slots = 16 * 1024;
1156 else
1157 slots = 8 * 1024;
1158 }
1159
098a697b
EB		 1160 tcp_metrics_hash_log = order_base_2(slots);
1161 size = sizeof(struct tcpm_hash_bucket) << tcp_metrics_hash_log;
51c5d0c4 1162
098a697b
EB		 1163 tcp_metrics_hash = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
1164 if (!tcp_metrics_hash)
1165 tcp_metrics_hash = vzalloc(size);
976a702a 1166
098a697b 1167 if (!tcp_metrics_hash)
51c5d0c4
DM		 1168 return -ENOMEM;
1169
51c5d0c4
DM		 1170 return 0;
1171}
1172
1173static void __net_exit tcp_net_metrics_exit(struct net *net)
1174{
098a697b 1175 tcp_metrics_flush_all(net);
51c5d0c4
DM		 1176}
1177
1178static __net_initdata struct pernet_operations tcp_net_metrics_ops = {
1179 .init = tcp_net_metrics_init,
1180 .exit = tcp_net_metrics_exit,
1181};
1182
1183void __init tcp_metrics_init(void)
1184{
d23ff701
JA		 1185 int ret;
1186
1187 ret = register_pernet_subsys(&tcp_net_metrics_ops);
1188 if (ret < 0)
6493517e
EB		 1189 panic("Could not allocate the tcp_metrics hash table\n");
1190
d23ff701 1191 ret = genl_register_family_with_ops(&tcp_metrics_nl_family,
c53ed742 1192 tcp_metrics_nl_ops);
d23ff701 1193 if (ret < 0)
6493517e 1194 panic("Could not register tcp_metrics generic netlink\n");
51c5d0c4 1195}
Linux 6.x block layer and io_uring tree(s)