blob: ed9c9a91851ced8e719aeec58530fb326ec8a2db [file] [log] [blame]
David S. Miller51c5d0c2012-07-10 00:49:14 -07001#include <linux/rcupdate.h>
2#include <linux/spinlock.h>
3#include <linux/jiffies.h>
David S. Millerab92bb22012-07-09 16:19:30 -07004#include <linux/module.h>
David S. Miller4aabd8e2012-07-09 16:07:30 -07005#include <linux/cache.h>
David S. Miller51c5d0c2012-07-10 00:49:14 -07006#include <linux/slab.h>
7#include <linux/init.h>
David S. Miller4aabd8e2012-07-09 16:07:30 -07008#include <linux/tcp.h>
Eric Dumazet5815d5e2012-07-19 23:02:34 +00009#include <linux/hash.h>
Julian Anastasovd23ff702012-09-04 11:03:15 +000010#include <linux/tcp_metrics.h>
Eric Dumazet976a7022012-11-16 05:31:53 +000011#include <linux/vmalloc.h>
David S. Miller4aabd8e2012-07-09 16:07:30 -070012
13#include <net/inet_connection_sock.h>
David S. Miller51c5d0c2012-07-10 00:49:14 -070014#include <net/net_namespace.h>
David S. Millerab92bb22012-07-09 16:19:30 -070015#include <net/request_sock.h>
David S. Miller51c5d0c2012-07-10 00:49:14 -070016#include <net/inetpeer.h>
David S. Miller4aabd8e2012-07-09 16:07:30 -070017#include <net/sock.h>
David S. Miller51c5d0c2012-07-10 00:49:14 -070018#include <net/ipv6.h>
David S. Miller4aabd8e2012-07-09 16:07:30 -070019#include <net/dst.h>
20#include <net/tcp.h>
Julian Anastasovd23ff702012-09-04 11:03:15 +000021#include <net/genetlink.h>
David S. Miller4aabd8e2012-07-09 16:07:30 -070022
23int sysctl_tcp_nometrics_save __read_mostly;
24
David S. Miller41804422014-01-18 00:55:41 -080025static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr,
26 const struct inetpeer_addr *daddr,
Christoph Paasch77f99ad2014-01-16 20:01:21 +010027 struct net *net, unsigned int hash);
28
Yuchung Cheng1fe4c482012-07-19 06:43:06 +000029struct tcp_fastopen_metrics {
30 u16 mss;
Yuchung Chengaab48742012-07-19 06:43:10 +000031 u16 syn_loss:10; /* Recurring Fast Open SYN losses */
32 unsigned long last_syn_loss; /* Last Fast Open SYN loss */
Yuchung Cheng1fe4c482012-07-19 06:43:06 +000033 struct tcp_fastopen_cookie cookie;
34};
35
Eric Dumazet740b0f12014-02-26 14:02:48 -080036/* 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
David S. Miller51c5d0c2012-07-10 00:49:14 -070041struct tcp_metrics_block {
42 struct tcp_metrics_block __rcu *tcpm_next;
Christoph Paascha5443022014-01-08 16:05:56 +010043 struct inetpeer_addr tcpm_saddr;
Christoph Paasch324fd552014-01-08 16:05:55 +010044 struct inetpeer_addr tcpm_daddr;
David S. Miller51c5d0c2012-07-10 00:49:14 -070045 unsigned long tcpm_stamp;
David S. Miller81166dd2012-07-10 03:14:24 -070046 u32 tcpm_ts;
47 u32 tcpm_ts_stamp;
David S. Miller51c5d0c2012-07-10 00:49:14 -070048 u32 tcpm_lock;
Eric Dumazet740b0f12014-02-26 14:02:48 -080049 u32 tcpm_vals[TCP_METRIC_MAX_KERNEL + 1];
Yuchung Cheng1fe4c482012-07-19 06:43:06 +000050 struct tcp_fastopen_metrics tcpm_fastopen;
Julian Anastasovd23ff702012-09-04 11:03:15 +000051
52 struct rcu_head rcu_head;
David S. Miller51c5d0c2012-07-10 00:49:14 -070053};
54
55static bool tcp_metric_locked(struct tcp_metrics_block *tm,
56 enum tcp_metric_index idx)
57{
58 return tm->tcpm_lock & (1 << idx);
59}
60
61static u32 tcp_metric_get(struct tcp_metrics_block *tm,
62 enum tcp_metric_index idx)
63{
64 return tm->tcpm_vals[idx];
65}
66
David S. Miller51c5d0c2012-07-10 00:49:14 -070067static void tcp_metric_set(struct tcp_metrics_block *tm,
68 enum tcp_metric_index idx,
69 u32 val)
70{
71 tm->tcpm_vals[idx] = val;
72}
73
David S. Miller51c5d0c2012-07-10 00:49:14 -070074static bool addr_same(const struct inetpeer_addr *a,
75 const struct inetpeer_addr *b)
76{
77 const struct in6_addr *a6, *b6;
78
79 if (a->family != b->family)
80 return false;
81 if (a->family == AF_INET)
82 return a->addr.a4 == b->addr.a4;
83
84 a6 = (const struct in6_addr *) &a->addr.a6[0];
85 b6 = (const struct in6_addr *) &b->addr.a6[0];
86
87 return ipv6_addr_equal(a6, b6);
88}
89
90struct tcpm_hash_bucket {
91 struct tcp_metrics_block __rcu *chain;
92};
93
94static DEFINE_SPINLOCK(tcp_metrics_lock);
95
Eric Dumazet740b0f12014-02-26 14:02:48 -080096static void tcpm_suck_dst(struct tcp_metrics_block *tm,
97 const struct dst_entry *dst,
Eric Dumazetefeaa552013-05-03 19:12:45 +000098 bool fastopen_clear)
David S. Miller51c5d0c2012-07-10 00:49:14 -070099{
Eric Dumazet740b0f12014-02-26 14:02:48 -0800100 u32 msval;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700101 u32 val;
102
Julian Anastasov9a0a9502012-07-23 10:46:38 +0300103 tm->tcpm_stamp = jiffies;
104
David S. Miller51c5d0c2012-07-10 00:49:14 -0700105 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
Eric Dumazet740b0f12014-02-26 14:02:48 -0800118 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;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700123 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);
David S. Miller81166dd2012-07-10 03:14:24 -0700126 tm->tcpm_ts = 0;
127 tm->tcpm_ts_stamp = 0;
Eric Dumazetefeaa552013-05-03 19:12:45 +0000128 if (fastopen_clear) {
129 tm->tcpm_fastopen.mss = 0;
130 tm->tcpm_fastopen.syn_loss = 0;
131 tm->tcpm_fastopen.cookie.len = 0;
132 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700133}
134
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100135#define TCP_METRICS_TIMEOUT (60 * 60 * HZ)
136
137static void tcpm_check_stamp(struct tcp_metrics_block *tm, struct dst_entry *dst)
138{
139 if (tm && unlikely(time_after(jiffies, tm->tcpm_stamp + TCP_METRICS_TIMEOUT)))
140 tcpm_suck_dst(tm, dst, false);
141}
142
143#define TCP_METRICS_RECLAIM_DEPTH 5
144#define TCP_METRICS_RECLAIM_PTR (struct tcp_metrics_block *) 0x1UL
145
David S. Miller51c5d0c2012-07-10 00:49:14 -0700146static struct tcp_metrics_block *tcpm_new(struct dst_entry *dst,
Christoph Paascha5443022014-01-08 16:05:56 +0100147 struct inetpeer_addr *saddr,
Christoph Paasch324fd552014-01-08 16:05:55 +0100148 struct inetpeer_addr *daddr,
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100149 unsigned int hash)
David S. Miller51c5d0c2012-07-10 00:49:14 -0700150{
151 struct tcp_metrics_block *tm;
152 struct net *net;
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100153 bool reclaim = false;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700154
155 spin_lock_bh(&tcp_metrics_lock);
156 net = dev_net(dst->dev);
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100157
158 /* While waiting for the spin-lock the cache might have been populated
159 * with this entry and so we have to check again.
160 */
David S. Miller41804422014-01-18 00:55:41 -0800161 tm = __tcp_get_metrics(saddr, daddr, net, hash);
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100162 if (tm == TCP_METRICS_RECLAIM_PTR) {
163 reclaim = true;
164 tm = NULL;
165 }
166 if (tm) {
167 tcpm_check_stamp(tm, dst);
168 goto out_unlock;
169 }
170
David S. Miller51c5d0c2012-07-10 00:49:14 -0700171 if (unlikely(reclaim)) {
172 struct tcp_metrics_block *oldest;
173
174 oldest = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain);
175 for (tm = rcu_dereference(oldest->tcpm_next); tm;
176 tm = rcu_dereference(tm->tcpm_next)) {
177 if (time_before(tm->tcpm_stamp, oldest->tcpm_stamp))
178 oldest = tm;
179 }
180 tm = oldest;
181 } else {
182 tm = kmalloc(sizeof(*tm), GFP_ATOMIC);
183 if (!tm)
184 goto out_unlock;
185 }
Christoph Paascha5443022014-01-08 16:05:56 +0100186 tm->tcpm_saddr = *saddr;
Christoph Paasch324fd552014-01-08 16:05:55 +0100187 tm->tcpm_daddr = *daddr;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700188
Eric Dumazetefeaa552013-05-03 19:12:45 +0000189 tcpm_suck_dst(tm, dst, true);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700190
191 if (likely(!reclaim)) {
192 tm->tcpm_next = net->ipv4.tcp_metrics_hash[hash].chain;
193 rcu_assign_pointer(net->ipv4.tcp_metrics_hash[hash].chain, tm);
194 }
195
196out_unlock:
197 spin_unlock_bh(&tcp_metrics_lock);
198 return tm;
199}
200
David S. Miller51c5d0c2012-07-10 00:49:14 -0700201static struct tcp_metrics_block *tcp_get_encode(struct tcp_metrics_block *tm, int depth)
202{
203 if (tm)
204 return tm;
205 if (depth > TCP_METRICS_RECLAIM_DEPTH)
206 return TCP_METRICS_RECLAIM_PTR;
207 return NULL;
208}
209
Christoph Paascha5443022014-01-08 16:05:56 +0100210static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr,
211 const struct inetpeer_addr *daddr,
David S. Miller51c5d0c2012-07-10 00:49:14 -0700212 struct net *net, unsigned int hash)
213{
214 struct tcp_metrics_block *tm;
215 int depth = 0;
216
217 for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
218 tm = rcu_dereference(tm->tcpm_next)) {
Christoph Paascha5443022014-01-08 16:05:56 +0100219 if (addr_same(&tm->tcpm_saddr, saddr) &&
220 addr_same(&tm->tcpm_daddr, daddr))
David S. Miller51c5d0c2012-07-10 00:49:14 -0700221 break;
222 depth++;
223 }
224 return tcp_get_encode(tm, depth);
225}
226
227static struct tcp_metrics_block *__tcp_get_metrics_req(struct request_sock *req,
228 struct dst_entry *dst)
229{
230 struct tcp_metrics_block *tm;
Christoph Paascha5443022014-01-08 16:05:56 +0100231 struct inetpeer_addr saddr, daddr;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700232 unsigned int hash;
233 struct net *net;
234
Christoph Paascha5443022014-01-08 16:05:56 +0100235 saddr.family = req->rsk_ops->family;
Christoph Paasch324fd552014-01-08 16:05:55 +0100236 daddr.family = req->rsk_ops->family;
237 switch (daddr.family) {
David S. Miller51c5d0c2012-07-10 00:49:14 -0700238 case AF_INET:
Christoph Paascha5443022014-01-08 16:05:56 +0100239 saddr.addr.a4 = inet_rsk(req)->ir_loc_addr;
Christoph Paasch324fd552014-01-08 16:05:55 +0100240 daddr.addr.a4 = inet_rsk(req)->ir_rmt_addr;
241 hash = (__force unsigned int) daddr.addr.a4;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700242 break;
Eric Dumazet634fb9792013-10-09 15:21:29 -0700243#if IS_ENABLED(CONFIG_IPV6)
David S. Miller51c5d0c2012-07-10 00:49:14 -0700244 case AF_INET6:
Christoph Paascha5443022014-01-08 16:05:56 +0100245 *(struct in6_addr *)saddr.addr.a6 = inet_rsk(req)->ir_v6_loc_addr;
Christoph Paasch324fd552014-01-08 16:05:55 +0100246 *(struct in6_addr *)daddr.addr.a6 = inet_rsk(req)->ir_v6_rmt_addr;
Eric Dumazet634fb9792013-10-09 15:21:29 -0700247 hash = ipv6_addr_hash(&inet_rsk(req)->ir_v6_rmt_addr);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700248 break;
Eric Dumazet634fb9792013-10-09 15:21:29 -0700249#endif
David S. Miller51c5d0c2012-07-10 00:49:14 -0700250 default:
251 return NULL;
252 }
253
David S. Miller51c5d0c2012-07-10 00:49:14 -0700254 net = dev_net(dst->dev);
Eric Dumazet5815d5e2012-07-19 23:02:34 +0000255 hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700256
257 for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
258 tm = rcu_dereference(tm->tcpm_next)) {
Christoph Paascha5443022014-01-08 16:05:56 +0100259 if (addr_same(&tm->tcpm_saddr, &saddr) &&
260 addr_same(&tm->tcpm_daddr, &daddr))
David S. Miller51c5d0c2012-07-10 00:49:14 -0700261 break;
262 }
263 tcpm_check_stamp(tm, dst);
264 return tm;
265}
266
David S. Miller81166dd2012-07-10 03:14:24 -0700267static struct tcp_metrics_block *__tcp_get_metrics_tw(struct inet_timewait_sock *tw)
268{
David S. Miller81166dd2012-07-10 03:14:24 -0700269 struct tcp_metrics_block *tm;
Christoph Paascha5443022014-01-08 16:05:56 +0100270 struct inetpeer_addr saddr, daddr;
David S. Miller81166dd2012-07-10 03:14:24 -0700271 unsigned int hash;
272 struct net *net;
273
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100274 if (tw->tw_family == AF_INET) {
275 saddr.family = AF_INET;
Christoph Paascha5443022014-01-08 16:05:56 +0100276 saddr.addr.a4 = tw->tw_rcv_saddr;
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100277 daddr.family = AF_INET;
Christoph Paasch324fd552014-01-08 16:05:55 +0100278 daddr.addr.a4 = tw->tw_daddr;
279 hash = (__force unsigned int) daddr.addr.a4;
David S. Miller81166dd2012-07-10 03:14:24 -0700280 }
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100281#if IS_ENABLED(CONFIG_IPV6)
282 else if (tw->tw_family == AF_INET6) {
283 if (ipv6_addr_v4mapped(&tw->tw_v6_daddr)) {
284 saddr.family = AF_INET;
285 saddr.addr.a4 = tw->tw_rcv_saddr;
286 daddr.family = AF_INET;
287 daddr.addr.a4 = tw->tw_daddr;
288 hash = (__force unsigned int) daddr.addr.a4;
289 } else {
290 saddr.family = AF_INET6;
291 *(struct in6_addr *)saddr.addr.a6 = tw->tw_v6_rcv_saddr;
292 daddr.family = AF_INET6;
293 *(struct in6_addr *)daddr.addr.a6 = tw->tw_v6_daddr;
294 hash = ipv6_addr_hash(&tw->tw_v6_daddr);
295 }
296 }
297#endif
298 else
299 return NULL;
David S. Miller81166dd2012-07-10 03:14:24 -0700300
David S. Miller81166dd2012-07-10 03:14:24 -0700301 net = twsk_net(tw);
Eric Dumazet5815d5e2012-07-19 23:02:34 +0000302 hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
David S. Miller81166dd2012-07-10 03:14:24 -0700303
304 for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
305 tm = rcu_dereference(tm->tcpm_next)) {
Christoph Paascha5443022014-01-08 16:05:56 +0100306 if (addr_same(&tm->tcpm_saddr, &saddr) &&
307 addr_same(&tm->tcpm_daddr, &daddr))
David S. Miller81166dd2012-07-10 03:14:24 -0700308 break;
309 }
310 return tm;
311}
312
David S. Miller51c5d0c2012-07-10 00:49:14 -0700313static struct tcp_metrics_block *tcp_get_metrics(struct sock *sk,
314 struct dst_entry *dst,
315 bool create)
316{
317 struct tcp_metrics_block *tm;
Christoph Paascha5443022014-01-08 16:05:56 +0100318 struct inetpeer_addr saddr, daddr;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700319 unsigned int hash;
320 struct net *net;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700321
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100322 if (sk->sk_family == AF_INET) {
323 saddr.family = AF_INET;
Christoph Paascha5443022014-01-08 16:05:56 +0100324 saddr.addr.a4 = inet_sk(sk)->inet_saddr;
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100325 daddr.family = AF_INET;
Christoph Paasch324fd552014-01-08 16:05:55 +0100326 daddr.addr.a4 = inet_sk(sk)->inet_daddr;
327 hash = (__force unsigned int) daddr.addr.a4;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700328 }
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100329#if IS_ENABLED(CONFIG_IPV6)
330 else if (sk->sk_family == AF_INET6) {
331 if (ipv6_addr_v4mapped(&sk->sk_v6_daddr)) {
332 saddr.family = AF_INET;
333 saddr.addr.a4 = inet_sk(sk)->inet_saddr;
334 daddr.family = AF_INET;
335 daddr.addr.a4 = inet_sk(sk)->inet_daddr;
336 hash = (__force unsigned int) daddr.addr.a4;
337 } else {
338 saddr.family = AF_INET6;
339 *(struct in6_addr *)saddr.addr.a6 = sk->sk_v6_rcv_saddr;
340 daddr.family = AF_INET6;
341 *(struct in6_addr *)daddr.addr.a6 = sk->sk_v6_daddr;
342 hash = ipv6_addr_hash(&sk->sk_v6_daddr);
343 }
344 }
345#endif
346 else
347 return NULL;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700348
David S. Miller51c5d0c2012-07-10 00:49:14 -0700349 net = dev_net(dst->dev);
Eric Dumazet5815d5e2012-07-19 23:02:34 +0000350 hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700351
Christoph Paascha5443022014-01-08 16:05:56 +0100352 tm = __tcp_get_metrics(&saddr, &daddr, net, hash);
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100353 if (tm == TCP_METRICS_RECLAIM_PTR)
David S. Miller51c5d0c2012-07-10 00:49:14 -0700354 tm = NULL;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700355 if (!tm && create)
David S. Miller41804422014-01-18 00:55:41 -0800356 tm = tcpm_new(dst, &saddr, &daddr, hash);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700357 else
358 tcpm_check_stamp(tm, dst);
359
360 return tm;
361}
362
David S. Miller4aabd8e2012-07-09 16:07:30 -0700363/* 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{
David S. Miller51c5d0c2012-07-10 00:49:14 -0700369 const struct inet_connection_sock *icsk = inet_csk(sk);
David S. Miller4aabd8e2012-07-09 16:07:30 -0700370 struct dst_entry *dst = __sk_dst_get(sk);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700371 struct tcp_sock *tp = tcp_sk(sk);
372 struct tcp_metrics_block *tm;
373 unsigned long rtt;
374 u32 val;
375 int m;
David S. Miller4aabd8e2012-07-09 16:07:30 -0700376
David S. Miller51c5d0c2012-07-10 00:49:14 -0700377 if (sysctl_tcp_nometrics_save || !dst)
David S. Miller4aabd8e2012-07-09 16:07:30 -0700378 return;
379
David S. Miller51c5d0c2012-07-10 00:49:14 -0700380 if (dst->flags & DST_HOST)
David S. Miller4aabd8e2012-07-09 16:07:30 -0700381 dst_confirm(dst);
382
David S. Miller51c5d0c2012-07-10 00:49:14 -0700383 rcu_read_lock();
Eric Dumazet740b0f12014-02-26 14:02:48 -0800384 if (icsk->icsk_backoff || !tp->srtt_us) {
David S. Miller51c5d0c2012-07-10 00:49:14 -0700385 /* This session failed to estimate rtt. Why?
386 * Probably, no packets returned in time. Reset our
387 * results.
David S. Miller4aabd8e2012-07-09 16:07:30 -0700388 */
David S. Miller51c5d0c2012-07-10 00:49:14 -0700389 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);
395
396 if (!tm)
397 goto out_unlock;
398
Eric Dumazet740b0f12014-02-26 14:02:48 -0800399 rtt = tcp_metric_get(tm, TCP_METRIC_RTT);
400 m = rtt - tp->srtt_us;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700401
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)
Eric Dumazet740b0f12014-02-26 14:02:48 -0800408 rtt = tp->srtt_us;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700409 else
410 rtt -= (m >> 3);
Eric Dumazet740b0f12014-02-26 14:02:48 -0800411 tcp_metric_set(tm, TCP_METRIC_RTT, rtt);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700412 }
413
414 if (!tcp_metric_locked(tm, TCP_METRIC_RTTVAR)) {
415 unsigned long var;
416
417 if (m < 0)
418 m = -m;
419
420 /* Scale deviation to rttvar fixed point */
421 m >>= 1;
Eric Dumazet740b0f12014-02-26 14:02:48 -0800422 if (m < tp->mdev_us)
423 m = tp->mdev_us;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700424
Eric Dumazet740b0f12014-02-26 14:02:48 -0800425 var = tcp_metric_get(tm, TCP_METRIC_RTTVAR);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700426 if (m >= var)
427 var = m;
428 else
429 var -= (var - m) >> 2;
430
Eric Dumazet740b0f12014-02-26 14:02:48 -0800431 tcp_metric_set(tm, TCP_METRIC_RTTVAR, var);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700432 }
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);
David S. Miller4aabd8e2012-07-09 16:07:30 -0700441 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700442 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);
David S. Miller4aabd8e2012-07-09 16:07:30 -0700447 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700448 } else if (tp->snd_cwnd > tp->snd_ssthresh &&
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);
Alexander Duyck21008442012-07-11 17:18:04 -0700456 tcp_metric_set(tm, TCP_METRIC_CWND, (val + tp->snd_cwnd) >> 1);
David S. Miller4aabd8e2012-07-09 16:07:30 -0700457 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700458 } 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 &&
David S. Miller4aabd8e2012-07-09 16:07:30 -0700476 tp->reordering != sysctl_tcp_reordering)
David S. Miller51c5d0c2012-07-10 00:49:14 -0700477 tcp_metric_set(tm, TCP_METRIC_REORDERING,
478 tp->reordering);
David S. Miller4aabd8e2012-07-09 16:07:30 -0700479 }
480 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700481 tm->tcpm_stamp = jiffies;
482out_unlock:
483 rcu_read_unlock();
David S. Miller4aabd8e2012-07-09 16:07:30 -0700484}
485
486/* Initialize metrics on socket. */
487
488void tcp_init_metrics(struct sock *sk)
489{
David S. Miller4aabd8e2012-07-09 16:07:30 -0700490 struct dst_entry *dst = __sk_dst_get(sk);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700491 struct tcp_sock *tp = tcp_sk(sk);
492 struct tcp_metrics_block *tm;
Yuchung Cheng1b7fdd22013-08-30 08:35:53 -0700493 u32 val, crtt = 0; /* cached RTT scaled by 8 */
David S. Miller4aabd8e2012-07-09 16:07:30 -0700494
495 if (dst == NULL)
496 goto reset;
497
498 dst_confirm(dst);
499
David S. Miller51c5d0c2012-07-10 00:49:14 -0700500 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;
David S. Miller4aabd8e2012-07-09 16:07:30 -0700513 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 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700521 val = tcp_metric_get(tm, TCP_METRIC_REORDERING);
522 if (val && tp->reordering != val) {
David S. Miller4aabd8e2012-07-09 16:07:30 -0700523 tcp_disable_fack(tp);
524 tcp_disable_early_retrans(tp);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700525 tp->reordering = val;
David S. Miller4aabd8e2012-07-09 16:07:30 -0700526 }
527
Eric Dumazet740b0f12014-02-26 14:02:48 -0800528 crtt = tcp_metric_get(tm, TCP_METRIC_RTT);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700529 rcu_read_unlock();
David S. Miller4aabd8e2012-07-09 16:07:30 -0700530reset:
Yuchung Cheng52f20e62013-09-03 14:14:35 -0700531 /* 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 */
Eric Dumazet740b0f12014-02-26 14:02:48 -0800551 if (crtt > tp->srtt_us) {
Neal Cardwell269aa752013-09-16 21:44:20 -0400552 /* Set RTO like tcp_rtt_estimator(), but from cached RTT. */
Eric Dumazet740b0f12014-02-26 14:02:48 -0800553 crtt /= 8 * USEC_PER_MSEC;
Neal Cardwell269aa752013-09-16 21:44:20 -0400554 inet_csk(sk)->icsk_rto = crtt + max(2 * crtt, tcp_rto_min(sk));
Eric Dumazet740b0f12014-02-26 14:02:48 -0800555 } else if (tp->srtt_us == 0) {
David S. Miller4aabd8e2012-07-09 16:07:30 -0700556 /* 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 */
Eric Dumazet740b0f12014-02-26 14:02:48 -0800562 tp->rttvar_us = jiffies_to_usecs(TCP_TIMEOUT_FALLBACK);
563 tp->mdev_us = tp->mdev_max_us = tp->rttvar_us;
564
David S. Miller4aabd8e2012-07-09 16:07:30 -0700565 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}
David S. Millerab92bb22012-07-09 16:19:30 -0700578
Hannes Frederic Sowaa26552a2014-08-14 22:06:12 +0200579bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
580 bool paws_check, bool timestamps)
David S. Millerab92bb22012-07-09 16:19:30 -0700581{
David S. Miller51c5d0c2012-07-10 00:49:14 -0700582 struct tcp_metrics_block *tm;
583 bool ret;
584
David S. Millerab92bb22012-07-09 16:19:30 -0700585 if (!dst)
586 return false;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700587
588 rcu_read_lock();
589 tm = __tcp_get_metrics_req(req, dst);
David S. Miller81166dd2012-07-10 03:14:24 -0700590 if (paws_check) {
591 if (tm &&
592 (u32)get_seconds() - tm->tcpm_ts_stamp < TCP_PAWS_MSL &&
Hannes Frederic Sowaa26552a2014-08-14 22:06:12 +0200593 ((s32)(tm->tcpm_ts - req->ts_recent) > TCP_PAWS_WINDOW ||
594 !timestamps))
David S. Miller81166dd2012-07-10 03:14:24 -0700595 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 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700604 rcu_read_unlock();
605
606 return ret;
David S. Millerab92bb22012-07-09 16:19:30 -0700607}
608EXPORT_SYMBOL_GPL(tcp_peer_is_proven);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700609
David S. Miller81166dd2012-07-10 03:14:24 -0700610void 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);
Julian Anastasov9a0a9502012-07-23 10:46:38 +0300666 if (tm) {
David S. Miller81166dd2012-07-10 03:14:24 -0700667 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
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000684static DEFINE_SEQLOCK(fastopen_seqlock);
685
686void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
Yuchung Chengaab48742012-07-19 06:43:10 +0000687 struct tcp_fastopen_cookie *cookie,
688 int *syn_loss, unsigned long *last_syn_loss)
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000689{
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;
Yuchung Chengaab48742012-07-19 06:43:10 +0000703 *syn_loss = tfom->syn_loss;
704 *last_syn_loss = *syn_loss ? tfom->last_syn_loss : 0;
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000705 } while (read_seqretry(&fastopen_seqlock, seq));
706 }
707 rcu_read_unlock();
708}
709
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000710void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
Yuchung Chengaab48742012-07-19 06:43:10 +0000711 struct tcp_fastopen_cookie *cookie, bool syn_lost)
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000712{
Eric Dumazetdccf76c2013-11-13 15:00:46 -0800713 struct dst_entry *dst = __sk_dst_get(sk);
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000714 struct tcp_metrics_block *tm;
715
Eric Dumazetdccf76c2013-11-13 15:00:46 -0800716 if (!dst)
717 return;
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000718 rcu_read_lock();
Eric Dumazetdccf76c2013-11-13 15:00:46 -0800719 tm = tcp_get_metrics(sk, dst, true);
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000720 if (tm) {
721 struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen;
722
723 write_seqlock_bh(&fastopen_seqlock);
Yuchung Chengc9686012013-10-29 10:09:05 -0700724 if (mss)
725 tfom->mss = mss;
726 if (cookie && cookie->len > 0)
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000727 tfom->cookie = *cookie;
Yuchung Chengaab48742012-07-19 06:43:10 +0000728 if (syn_lost) {
729 ++tfom->syn_loss;
730 tfom->last_syn_loss = jiffies;
731 } else
732 tfom->syn_loss = 0;
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000733 write_sequnlock_bh(&fastopen_seqlock);
734 }
735 rcu_read_unlock();
736}
737
Julian Anastasovd23ff702012-09-04 11:03:15 +0000738static struct genl_family tcp_metrics_nl_family = {
739 .id = GENL_ID_GENERATE,
740 .hdrsize = 0,
741 .name = TCP_METRICS_GENL_NAME,
742 .version = TCP_METRICS_GENL_VERSION,
743 .maxattr = TCP_METRICS_ATTR_MAX,
744 .netnsok = true,
745};
746
747static struct nla_policy tcp_metrics_nl_policy[TCP_METRICS_ATTR_MAX + 1] = {
748 [TCP_METRICS_ATTR_ADDR_IPV4] = { .type = NLA_U32, },
749 [TCP_METRICS_ATTR_ADDR_IPV6] = { .type = NLA_BINARY,
750 .len = sizeof(struct in6_addr), },
751 /* Following attributes are not received for GET/DEL,
752 * we keep them for reference
753 */
754#if 0
755 [TCP_METRICS_ATTR_AGE] = { .type = NLA_MSECS, },
756 [TCP_METRICS_ATTR_TW_TSVAL] = { .type = NLA_U32, },
757 [TCP_METRICS_ATTR_TW_TS_STAMP] = { .type = NLA_S32, },
758 [TCP_METRICS_ATTR_VALS] = { .type = NLA_NESTED, },
759 [TCP_METRICS_ATTR_FOPEN_MSS] = { .type = NLA_U16, },
760 [TCP_METRICS_ATTR_FOPEN_SYN_DROPS] = { .type = NLA_U16, },
761 [TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS] = { .type = NLA_MSECS, },
762 [TCP_METRICS_ATTR_FOPEN_COOKIE] = { .type = NLA_BINARY,
763 .len = TCP_FASTOPEN_COOKIE_MAX, },
764#endif
765};
766
767/* Add attributes, caller cancels its header on failure */
768static int tcp_metrics_fill_info(struct sk_buff *msg,
769 struct tcp_metrics_block *tm)
770{
771 struct nlattr *nest;
772 int i;
773
Christoph Paasch324fd552014-01-08 16:05:55 +0100774 switch (tm->tcpm_daddr.family) {
Julian Anastasovd23ff702012-09-04 11:03:15 +0000775 case AF_INET:
776 if (nla_put_be32(msg, TCP_METRICS_ATTR_ADDR_IPV4,
Christoph Paasch324fd552014-01-08 16:05:55 +0100777 tm->tcpm_daddr.addr.a4) < 0)
Julian Anastasovd23ff702012-09-04 11:03:15 +0000778 goto nla_put_failure;
Christoph Paasch8a593592014-01-08 16:05:57 +0100779 if (nla_put_be32(msg, TCP_METRICS_ATTR_SADDR_IPV4,
780 tm->tcpm_saddr.addr.a4) < 0)
781 goto nla_put_failure;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000782 break;
783 case AF_INET6:
784 if (nla_put(msg, TCP_METRICS_ATTR_ADDR_IPV6, 16,
Christoph Paasch324fd552014-01-08 16:05:55 +0100785 tm->tcpm_daddr.addr.a6) < 0)
Julian Anastasovd23ff702012-09-04 11:03:15 +0000786 goto nla_put_failure;
Christoph Paasch8a593592014-01-08 16:05:57 +0100787 if (nla_put(msg, TCP_METRICS_ATTR_SADDR_IPV6, 16,
788 tm->tcpm_saddr.addr.a6) < 0)
789 goto nla_put_failure;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000790 break;
791 default:
792 return -EAFNOSUPPORT;
793 }
794
795 if (nla_put_msecs(msg, TCP_METRICS_ATTR_AGE,
796 jiffies - tm->tcpm_stamp) < 0)
797 goto nla_put_failure;
798 if (tm->tcpm_ts_stamp) {
799 if (nla_put_s32(msg, TCP_METRICS_ATTR_TW_TS_STAMP,
800 (s32) (get_seconds() - tm->tcpm_ts_stamp)) < 0)
801 goto nla_put_failure;
802 if (nla_put_u32(msg, TCP_METRICS_ATTR_TW_TSVAL,
803 tm->tcpm_ts) < 0)
804 goto nla_put_failure;
805 }
806
807 {
808 int n = 0;
809
810 nest = nla_nest_start(msg, TCP_METRICS_ATTR_VALS);
811 if (!nest)
812 goto nla_put_failure;
Eric Dumazet740b0f12014-02-26 14:02:48 -0800813 for (i = 0; i < TCP_METRIC_MAX_KERNEL + 1; i++) {
814 u32 val = tm->tcpm_vals[i];
815
816 if (!val)
Julian Anastasovd23ff702012-09-04 11:03:15 +0000817 continue;
Eric Dumazet740b0f12014-02-26 14:02:48 -0800818 if (i == TCP_METRIC_RTT) {
819 if (nla_put_u32(msg, TCP_METRIC_RTT_US + 1,
820 val) < 0)
821 goto nla_put_failure;
822 n++;
823 val = max(val / 1000, 1U);
824 }
825 if (i == TCP_METRIC_RTTVAR) {
826 if (nla_put_u32(msg, TCP_METRIC_RTTVAR_US + 1,
827 val) < 0)
828 goto nla_put_failure;
829 n++;
830 val = max(val / 1000, 1U);
831 }
832 if (nla_put_u32(msg, i + 1, val) < 0)
Julian Anastasovd23ff702012-09-04 11:03:15 +0000833 goto nla_put_failure;
834 n++;
835 }
836 if (n)
837 nla_nest_end(msg, nest);
838 else
839 nla_nest_cancel(msg, nest);
840 }
841
842 {
843 struct tcp_fastopen_metrics tfom_copy[1], *tfom;
844 unsigned int seq;
845
846 do {
847 seq = read_seqbegin(&fastopen_seqlock);
848 tfom_copy[0] = tm->tcpm_fastopen;
849 } while (read_seqretry(&fastopen_seqlock, seq));
850
851 tfom = tfom_copy;
852 if (tfom->mss &&
853 nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_MSS,
854 tfom->mss) < 0)
855 goto nla_put_failure;
856 if (tfom->syn_loss &&
857 (nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROPS,
858 tfom->syn_loss) < 0 ||
859 nla_put_msecs(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS,
860 jiffies - tfom->last_syn_loss) < 0))
861 goto nla_put_failure;
862 if (tfom->cookie.len > 0 &&
863 nla_put(msg, TCP_METRICS_ATTR_FOPEN_COOKIE,
864 tfom->cookie.len, tfom->cookie.val) < 0)
865 goto nla_put_failure;
866 }
867
868 return 0;
869
870nla_put_failure:
871 return -EMSGSIZE;
872}
873
874static int tcp_metrics_dump_info(struct sk_buff *skb,
875 struct netlink_callback *cb,
876 struct tcp_metrics_block *tm)
877{
878 void *hdr;
879
Eric W. Biederman15e47302012-09-07 20:12:54 +0000880 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
Julian Anastasovd23ff702012-09-04 11:03:15 +0000881 &tcp_metrics_nl_family, NLM_F_MULTI,
882 TCP_METRICS_CMD_GET);
883 if (!hdr)
884 return -EMSGSIZE;
885
886 if (tcp_metrics_fill_info(skb, tm) < 0)
887 goto nla_put_failure;
888
889 return genlmsg_end(skb, hdr);
890
891nla_put_failure:
892 genlmsg_cancel(skb, hdr);
893 return -EMSGSIZE;
894}
895
896static int tcp_metrics_nl_dump(struct sk_buff *skb,
897 struct netlink_callback *cb)
898{
899 struct net *net = sock_net(skb->sk);
900 unsigned int max_rows = 1U << net->ipv4.tcp_metrics_hash_log;
901 unsigned int row, s_row = cb->args[0];
902 int s_col = cb->args[1], col = s_col;
903
904 for (row = s_row; row < max_rows; row++, s_col = 0) {
905 struct tcp_metrics_block *tm;
906 struct tcpm_hash_bucket *hb = net->ipv4.tcp_metrics_hash + row;
907
908 rcu_read_lock();
909 for (col = 0, tm = rcu_dereference(hb->chain); tm;
910 tm = rcu_dereference(tm->tcpm_next), col++) {
911 if (col < s_col)
912 continue;
913 if (tcp_metrics_dump_info(skb, cb, tm) < 0) {
914 rcu_read_unlock();
915 goto done;
916 }
917 }
918 rcu_read_unlock();
919 }
920
921done:
922 cb->args[0] = row;
923 cb->args[1] = col;
924 return skb->len;
925}
926
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100927static int __parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
928 unsigned int *hash, int optional, int v4, int v6)
Julian Anastasovd23ff702012-09-04 11:03:15 +0000929{
930 struct nlattr *a;
931
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100932 a = info->attrs[v4];
Julian Anastasovd23ff702012-09-04 11:03:15 +0000933 if (a) {
934 addr->family = AF_INET;
935 addr->addr.a4 = nla_get_be32(a);
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100936 if (hash)
937 *hash = (__force unsigned int) addr->addr.a4;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000938 return 0;
939 }
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100940 a = info->attrs[v6];
Julian Anastasovd23ff702012-09-04 11:03:15 +0000941 if (a) {
Julian Anastasov2c42a3f2012-10-30 12:03:09 +0000942 if (nla_len(a) != sizeof(struct in6_addr))
Julian Anastasovd23ff702012-09-04 11:03:15 +0000943 return -EINVAL;
944 addr->family = AF_INET6;
945 memcpy(addr->addr.a6, nla_data(a), sizeof(addr->addr.a6));
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100946 if (hash)
947 *hash = ipv6_addr_hash((struct in6_addr *) addr->addr.a6);
Julian Anastasovd23ff702012-09-04 11:03:15 +0000948 return 0;
949 }
950 return optional ? 1 : -EAFNOSUPPORT;
951}
952
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100953static int parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
954 unsigned int *hash, int optional)
955{
956 return __parse_nl_addr(info, addr, hash, optional,
957 TCP_METRICS_ATTR_ADDR_IPV4,
958 TCP_METRICS_ATTR_ADDR_IPV6);
959}
960
961static int parse_nl_saddr(struct genl_info *info, struct inetpeer_addr *addr)
962{
963 return __parse_nl_addr(info, addr, NULL, 0,
964 TCP_METRICS_ATTR_SADDR_IPV4,
965 TCP_METRICS_ATTR_SADDR_IPV6);
966}
967
Julian Anastasovd23ff702012-09-04 11:03:15 +0000968static int tcp_metrics_nl_cmd_get(struct sk_buff *skb, struct genl_info *info)
969{
970 struct tcp_metrics_block *tm;
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100971 struct inetpeer_addr saddr, daddr;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000972 unsigned int hash;
973 struct sk_buff *msg;
974 struct net *net = genl_info_net(info);
975 void *reply;
976 int ret;
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100977 bool src = true;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000978
Christoph Paasch324fd552014-01-08 16:05:55 +0100979 ret = parse_nl_addr(info, &daddr, &hash, 0);
Julian Anastasovd23ff702012-09-04 11:03:15 +0000980 if (ret < 0)
981 return ret;
982
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100983 ret = parse_nl_saddr(info, &saddr);
984 if (ret < 0)
985 src = false;
986
Julian Anastasovd23ff702012-09-04 11:03:15 +0000987 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
988 if (!msg)
989 return -ENOMEM;
990
991 reply = genlmsg_put_reply(msg, info, &tcp_metrics_nl_family, 0,
992 info->genlhdr->cmd);
993 if (!reply)
994 goto nla_put_failure;
995
996 hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
997 ret = -ESRCH;
998 rcu_read_lock();
999 for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
1000 tm = rcu_dereference(tm->tcpm_next)) {
Christoph Paasch3e7013d2014-01-08 16:05:59 +01001001 if (addr_same(&tm->tcpm_daddr, &daddr) &&
1002 (!src || addr_same(&tm->tcpm_saddr, &saddr))) {
Julian Anastasovd23ff702012-09-04 11:03:15 +00001003 ret = tcp_metrics_fill_info(msg, tm);
1004 break;
1005 }
1006 }
1007 rcu_read_unlock();
1008 if (ret < 0)
1009 goto out_free;
1010
1011 genlmsg_end(msg, reply);
1012 return genlmsg_reply(msg, info);
1013
1014nla_put_failure:
1015 ret = -EMSGSIZE;
1016
1017out_free:
1018 nlmsg_free(msg);
1019 return ret;
1020}
1021
1022#define deref_locked_genl(p) \
1023 rcu_dereference_protected(p, lockdep_genl_is_held() && \
1024 lockdep_is_held(&tcp_metrics_lock))
1025
1026#define deref_genl(p) rcu_dereference_protected(p, lockdep_genl_is_held())
1027
1028static int tcp_metrics_flush_all(struct net *net)
1029{
1030 unsigned int max_rows = 1U << net->ipv4.tcp_metrics_hash_log;
1031 struct tcpm_hash_bucket *hb = net->ipv4.tcp_metrics_hash;
1032 struct tcp_metrics_block *tm;
1033 unsigned int row;
1034
1035 for (row = 0; row < max_rows; row++, hb++) {
1036 spin_lock_bh(&tcp_metrics_lock);
1037 tm = deref_locked_genl(hb->chain);
1038 if (tm)
1039 hb->chain = NULL;
1040 spin_unlock_bh(&tcp_metrics_lock);
1041 while (tm) {
1042 struct tcp_metrics_block *next;
1043
1044 next = deref_genl(tm->tcpm_next);
1045 kfree_rcu(tm, rcu_head);
1046 tm = next;
1047 }
1048 }
1049 return 0;
1050}
1051
1052static int tcp_metrics_nl_cmd_del(struct sk_buff *skb, struct genl_info *info)
1053{
1054 struct tcpm_hash_bucket *hb;
Christoph Paasch00ca9c52014-01-21 13:30:26 +01001055 struct tcp_metrics_block *tm;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001056 struct tcp_metrics_block __rcu **pp;
Christoph Paasch3e7013d2014-01-08 16:05:59 +01001057 struct inetpeer_addr saddr, daddr;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001058 unsigned int hash;
1059 struct net *net = genl_info_net(info);
1060 int ret;
Christoph Paasch00ca9c52014-01-21 13:30:26 +01001061 bool src = true, found = false;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001062
Christoph Paasch324fd552014-01-08 16:05:55 +01001063 ret = parse_nl_addr(info, &daddr, &hash, 1);
Julian Anastasovd23ff702012-09-04 11:03:15 +00001064 if (ret < 0)
1065 return ret;
1066 if (ret > 0)
1067 return tcp_metrics_flush_all(net);
Christoph Paasch3e7013d2014-01-08 16:05:59 +01001068 ret = parse_nl_saddr(info, &saddr);
1069 if (ret < 0)
1070 src = false;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001071
1072 hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
1073 hb = net->ipv4.tcp_metrics_hash + hash;
1074 pp = &hb->chain;
1075 spin_lock_bh(&tcp_metrics_lock);
Christoph Paaschbbf852b2014-01-08 16:05:58 +01001076 for (tm = deref_locked_genl(*pp); tm; tm = deref_locked_genl(*pp)) {
Christoph Paasch3e7013d2014-01-08 16:05:59 +01001077 if (addr_same(&tm->tcpm_daddr, &daddr) &&
1078 (!src || addr_same(&tm->tcpm_saddr, &saddr))) {
Julian Anastasovd23ff702012-09-04 11:03:15 +00001079 *pp = tm->tcpm_next;
Christoph Paasch00ca9c52014-01-21 13:30:26 +01001080 kfree_rcu(tm, rcu_head);
1081 found = true;
Christoph Paaschbbf852b2014-01-08 16:05:58 +01001082 } else {
1083 pp = &tm->tcpm_next;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001084 }
1085 }
1086 spin_unlock_bh(&tcp_metrics_lock);
Christoph Paasch00ca9c52014-01-21 13:30:26 +01001087 if (!found)
Julian Anastasovd23ff702012-09-04 11:03:15 +00001088 return -ESRCH;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001089 return 0;
1090}
1091
Johannes Berg4534de82013-11-14 17:14:46 +01001092static const struct genl_ops tcp_metrics_nl_ops[] = {
Julian Anastasovd23ff702012-09-04 11:03:15 +00001093 {
1094 .cmd = TCP_METRICS_CMD_GET,
1095 .doit = tcp_metrics_nl_cmd_get,
1096 .dumpit = tcp_metrics_nl_dump,
1097 .policy = tcp_metrics_nl_policy,
Julian Anastasovd23ff702012-09-04 11:03:15 +00001098 },
1099 {
1100 .cmd = TCP_METRICS_CMD_DEL,
1101 .doit = tcp_metrics_nl_cmd_del,
1102 .policy = tcp_metrics_nl_policy,
1103 .flags = GENL_ADMIN_PERM,
1104 },
1105};
1106
Eric Dumazet5815d5e2012-07-19 23:02:34 +00001107static unsigned int tcpmhash_entries;
David S. Miller51c5d0c2012-07-10 00:49:14 -07001108static int __init set_tcpmhash_entries(char *str)
1109{
1110 ssize_t ret;
1111
1112 if (!str)
1113 return 0;
1114
Eric Dumazet5815d5e2012-07-19 23:02:34 +00001115 ret = kstrtouint(str, 0, &tcpmhash_entries);
David S. Miller51c5d0c2012-07-10 00:49:14 -07001116 if (ret)
1117 return 0;
1118
1119 return 1;
1120}
1121__setup("tcpmhash_entries=", set_tcpmhash_entries);
1122
1123static int __net_init tcp_net_metrics_init(struct net *net)
1124{
Eric Dumazet5815d5e2012-07-19 23:02:34 +00001125 size_t size;
1126 unsigned int slots;
David S. Miller51c5d0c2012-07-10 00:49:14 -07001127
1128 slots = tcpmhash_entries;
1129 if (!slots) {
1130 if (totalram_pages >= 128 * 1024)
1131 slots = 16 * 1024;
1132 else
1133 slots = 8 * 1024;
1134 }
1135
Eric Dumazet5815d5e2012-07-19 23:02:34 +00001136 net->ipv4.tcp_metrics_hash_log = order_base_2(slots);
1137 size = sizeof(struct tcpm_hash_bucket) << net->ipv4.tcp_metrics_hash_log;
David S. Miller51c5d0c2012-07-10 00:49:14 -07001138
Eric Dumazet976a7022012-11-16 05:31:53 +00001139 net->ipv4.tcp_metrics_hash = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
1140 if (!net->ipv4.tcp_metrics_hash)
1141 net->ipv4.tcp_metrics_hash = vzalloc(size);
1142
David S. Miller51c5d0c2012-07-10 00:49:14 -07001143 if (!net->ipv4.tcp_metrics_hash)
1144 return -ENOMEM;
1145
David S. Miller51c5d0c2012-07-10 00:49:14 -07001146 return 0;
1147}
1148
1149static void __net_exit tcp_net_metrics_exit(struct net *net)
1150{
Eric Dumazet36471012012-08-09 11:19:13 +02001151 unsigned int i;
1152
1153 for (i = 0; i < (1U << net->ipv4.tcp_metrics_hash_log) ; i++) {
1154 struct tcp_metrics_block *tm, *next;
1155
1156 tm = rcu_dereference_protected(net->ipv4.tcp_metrics_hash[i].chain, 1);
1157 while (tm) {
1158 next = rcu_dereference_protected(tm->tcpm_next, 1);
1159 kfree(tm);
1160 tm = next;
1161 }
1162 }
WANG Cong4cb28972014-06-02 15:55:22 -07001163 kvfree(net->ipv4.tcp_metrics_hash);
David S. Miller51c5d0c2012-07-10 00:49:14 -07001164}
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{
Julian Anastasovd23ff702012-09-04 11:03:15 +00001173 int ret;
1174
1175 ret = register_pernet_subsys(&tcp_net_metrics_ops);
1176 if (ret < 0)
1177 goto cleanup;
1178 ret = genl_register_family_with_ops(&tcp_metrics_nl_family,
Johannes Bergc53ed742013-11-19 15:19:31 +01001179 tcp_metrics_nl_ops);
Julian Anastasovd23ff702012-09-04 11:03:15 +00001180 if (ret < 0)
1181 goto cleanup_subsys;
1182 return;
1183
1184cleanup_subsys:
1185 unregister_pernet_subsys(&tcp_net_metrics_ops);
1186
1187cleanup:
1188 return;
David S. Miller51c5d0c2012-07-10 00:49:14 -07001189}