blob: 7b7eec4399069249ef949ef4b287e8cd0ef4df91 [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;
Daniel Lee2646c832015-04-06 14:37:27 -070031 u16 syn_loss:10, /* Recurring Fast Open SYN losses */
32 try_exp:2; /* Request w/ exp. option (once) */
Yuchung Chengaab48742012-07-19 06:43:10 +000033 unsigned long last_syn_loss; /* Last Fast Open SYN loss */
Yuchung Cheng1fe4c482012-07-19 06:43:06 +000034 struct tcp_fastopen_cookie cookie;
35};
36
Eric Dumazet740b0f12014-02-26 14:02:48 -080037/* 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
David S. Miller51c5d0c2012-07-10 00:49:14 -070042struct tcp_metrics_block {
43 struct tcp_metrics_block __rcu *tcpm_next;
Eric W. Biederman849e8a02015-03-13 00:05:52 -050044 possible_net_t tcpm_net;
Christoph Paascha5443022014-01-08 16:05:56 +010045 struct inetpeer_addr tcpm_saddr;
Christoph Paasch324fd552014-01-08 16:05:55 +010046 struct inetpeer_addr tcpm_daddr;
David S. Miller51c5d0c2012-07-10 00:49:14 -070047 unsigned long tcpm_stamp;
David S. Miller81166dd2012-07-10 03:14:24 -070048 u32 tcpm_ts;
49 u32 tcpm_ts_stamp;
David S. Miller51c5d0c2012-07-10 00:49:14 -070050 u32 tcpm_lock;
Eric Dumazet740b0f12014-02-26 14:02:48 -080051 u32 tcpm_vals[TCP_METRIC_MAX_KERNEL + 1];
Yuchung Cheng1fe4c482012-07-19 06:43:06 +000052 struct tcp_fastopen_metrics tcpm_fastopen;
Julian Anastasovd23ff702012-09-04 11:03:15 +000053
54 struct rcu_head rcu_head;
David S. Miller51c5d0c2012-07-10 00:49:14 -070055};
56
Eric W. Biederman849e8a02015-03-13 00:05:52 -050057static inline struct net *tm_net(struct tcp_metrics_block *tm)
58{
59 return read_pnet(&tm->tcpm_net);
60}
61
David S. Miller51c5d0c2012-07-10 00:49:14 -070062static 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
David S. Miller51c5d0c2012-07-10 00:49:14 -070074static 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
David S. Miller51c5d0c2012-07-10 00:49:14 -070081static bool addr_same(const struct inetpeer_addr *a,
82 const struct inetpeer_addr *b)
83{
David Ahernd39d14f2015-08-27 16:07:01 -070084 return inetpeer_addr_cmp(a, b) == 0;
David S. Miller51c5d0c2012-07-10 00:49:14 -070085}
86
87struct tcpm_hash_bucket {
88 struct tcp_metrics_block __rcu *chain;
89};
90
Eric W. Biederman098a6972015-03-13 00:07:44 -050091static struct tcpm_hash_bucket *tcp_metrics_hash __read_mostly;
92static unsigned int tcp_metrics_hash_log __read_mostly;
93
David S. Miller51c5d0c2012-07-10 00:49:14 -070094static 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;
Daniel Lee2646c832015-04-06 14:37:27 -0700131 tm->tcpm_fastopen.try_exp = 0;
132 tm->tcpm_fastopen.cookie.exp = false;
Eric Dumazetefeaa552013-05-03 19:12:45 +0000133 tm->tcpm_fastopen.cookie.len = 0;
134 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700135}
136
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100137#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
Eric Dumazet9f1ab182015-03-16 07:14:34 -0700148#define deref_locked(p) \
149 rcu_dereference_protected(p, lockdep_is_held(&tcp_metrics_lock))
150
David S. Miller51c5d0c2012-07-10 00:49:14 -0700151static struct tcp_metrics_block *tcpm_new(struct dst_entry *dst,
Christoph Paascha5443022014-01-08 16:05:56 +0100152 struct inetpeer_addr *saddr,
Christoph Paasch324fd552014-01-08 16:05:55 +0100153 struct inetpeer_addr *daddr,
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100154 unsigned int hash)
David S. Miller51c5d0c2012-07-10 00:49:14 -0700155{
156 struct tcp_metrics_block *tm;
157 struct net *net;
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100158 bool reclaim = false;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700159
160 spin_lock_bh(&tcp_metrics_lock);
161 net = dev_net(dst->dev);
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100162
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 */
David S. Miller41804422014-01-18 00:55:41 -0800166 tm = __tcp_get_metrics(saddr, daddr, net, hash);
Christoph Paasch77f99ad2014-01-16 20:01:21 +0100167 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
David S. Miller51c5d0c2012-07-10 00:49:14 -0700176 if (unlikely(reclaim)) {
177 struct tcp_metrics_block *oldest;
178
Eric Dumazet9f1ab182015-03-16 07:14:34 -0700179 oldest = deref_locked(tcp_metrics_hash[hash].chain);
180 for (tm = deref_locked(oldest->tcpm_next); tm;
181 tm = deref_locked(tm->tcpm_next)) {
David S. Miller51c5d0c2012-07-10 00:49:14 -0700182 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 }
Eric W. Biederman849e8a02015-03-13 00:05:52 -0500191 write_pnet(&tm->tcpm_net, net);
Christoph Paascha5443022014-01-08 16:05:56 +0100192 tm->tcpm_saddr = *saddr;
Christoph Paasch324fd552014-01-08 16:05:55 +0100193 tm->tcpm_daddr = *daddr;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700194
Eric Dumazetefeaa552013-05-03 19:12:45 +0000195 tcpm_suck_dst(tm, dst, true);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700196
197 if (likely(!reclaim)) {
Eric W. Biederman098a6972015-03-13 00:07:44 -0500198 tm->tcpm_next = tcp_metrics_hash[hash].chain;
199 rcu_assign_pointer(tcp_metrics_hash[hash].chain, tm);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700200 }
201
202out_unlock:
203 spin_unlock_bh(&tcp_metrics_lock);
204 return tm;
205}
206
David S. Miller51c5d0c2012-07-10 00:49:14 -0700207static 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
Christoph Paascha5443022014-01-08 16:05:56 +0100216static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr,
217 const struct inetpeer_addr *daddr,
David S. Miller51c5d0c2012-07-10 00:49:14 -0700218 struct net *net, unsigned int hash)
219{
220 struct tcp_metrics_block *tm;
221 int depth = 0;
222
Eric W. Biederman098a6972015-03-13 00:07:44 -0500223 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700224 tm = rcu_dereference(tm->tcpm_next)) {
Christoph Paascha5443022014-01-08 16:05:56 +0100225 if (addr_same(&tm->tcpm_saddr, saddr) &&
Eric W. Biederman849e8a02015-03-13 00:05:52 -0500226 addr_same(&tm->tcpm_daddr, daddr) &&
227 net_eq(tm_net(tm), net))
David S. Miller51c5d0c2012-07-10 00:49:14 -0700228 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;
Christoph Paascha5443022014-01-08 16:05:56 +0100238 struct inetpeer_addr saddr, daddr;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700239 unsigned int hash;
240 struct net *net;
241
Christoph Paascha5443022014-01-08 16:05:56 +0100242 saddr.family = req->rsk_ops->family;
Christoph Paasch324fd552014-01-08 16:05:55 +0100243 daddr.family = req->rsk_ops->family;
244 switch (daddr.family) {
David S. Miller51c5d0c2012-07-10 00:49:14 -0700245 case AF_INET:
David Ahern3abef282015-08-27 16:07:00 -0700246 inetpeer_set_addr_v4(&saddr, inet_rsk(req)->ir_loc_addr);
247 inetpeer_set_addr_v4(&daddr, inet_rsk(req)->ir_rmt_addr);
David Ahern72afa352015-08-27 16:06:59 -0700248 hash = ipv4_addr_hash(inet_rsk(req)->ir_rmt_addr);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700249 break;
Eric Dumazet634fb9792013-10-09 15:21:29 -0700250#if IS_ENABLED(CONFIG_IPV6)
David S. Miller51c5d0c2012-07-10 00:49:14 -0700251 case AF_INET6:
David Ahern3abef282015-08-27 16:07:00 -0700252 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);
Eric Dumazet634fb9792013-10-09 15:21:29 -0700254 hash = ipv6_addr_hash(&inet_rsk(req)->ir_v6_rmt_addr);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700255 break;
Eric Dumazet634fb9792013-10-09 15:21:29 -0700256#endif
David S. Miller51c5d0c2012-07-10 00:49:14 -0700257 default:
258 return NULL;
259 }
260
David S. Miller51c5d0c2012-07-10 00:49:14 -0700261 net = dev_net(dst->dev);
Eric W. Biederman3e5da622015-03-13 00:05:24 -0500262 hash ^= net_hash_mix(net);
Eric W. Biederman098a6972015-03-13 00:07:44 -0500263 hash = hash_32(hash, tcp_metrics_hash_log);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700264
Eric W. Biederman098a6972015-03-13 00:07:44 -0500265 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700266 tm = rcu_dereference(tm->tcpm_next)) {
Christoph Paascha5443022014-01-08 16:05:56 +0100267 if (addr_same(&tm->tcpm_saddr, &saddr) &&
Eric W. Biederman849e8a02015-03-13 00:05:52 -0500268 addr_same(&tm->tcpm_daddr, &daddr) &&
269 net_eq(tm_net(tm), net))
David S. Miller51c5d0c2012-07-10 00:49:14 -0700270 break;
271 }
272 tcpm_check_stamp(tm, dst);
273 return tm;
274}
275
David S. Miller81166dd2012-07-10 03:14:24 -0700276static struct tcp_metrics_block *__tcp_get_metrics_tw(struct inet_timewait_sock *tw)
277{
David S. Miller81166dd2012-07-10 03:14:24 -0700278 struct tcp_metrics_block *tm;
Christoph Paascha5443022014-01-08 16:05:56 +0100279 struct inetpeer_addr saddr, daddr;
David S. Miller81166dd2012-07-10 03:14:24 -0700280 unsigned int hash;
281 struct net *net;
282
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100283 if (tw->tw_family == AF_INET) {
David Ahern3abef282015-08-27 16:07:00 -0700284 inetpeer_set_addr_v4(&saddr, tw->tw_rcv_saddr);
285 inetpeer_set_addr_v4(&daddr, tw->tw_daddr);
David Ahern72afa352015-08-27 16:06:59 -0700286 hash = ipv4_addr_hash(tw->tw_daddr);
David S. Miller81166dd2012-07-10 03:14:24 -0700287 }
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100288#if IS_ENABLED(CONFIG_IPV6)
289 else if (tw->tw_family == AF_INET6) {
290 if (ipv6_addr_v4mapped(&tw->tw_v6_daddr)) {
David Ahern3abef282015-08-27 16:07:00 -0700291 inetpeer_set_addr_v4(&saddr, tw->tw_rcv_saddr);
292 inetpeer_set_addr_v4(&daddr, tw->tw_daddr);
David Ahern72afa352015-08-27 16:06:59 -0700293 hash = ipv4_addr_hash(tw->tw_daddr);
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100294 } else {
David Ahern3abef282015-08-27 16:07:00 -0700295 inetpeer_set_addr_v6(&saddr, &tw->tw_v6_rcv_saddr);
296 inetpeer_set_addr_v6(&daddr, &tw->tw_v6_daddr);
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100297 hash = ipv6_addr_hash(&tw->tw_v6_daddr);
298 }
299 }
300#endif
301 else
302 return NULL;
David S. Miller81166dd2012-07-10 03:14:24 -0700303
David S. Miller81166dd2012-07-10 03:14:24 -0700304 net = twsk_net(tw);
Eric W. Biederman3e5da622015-03-13 00:05:24 -0500305 hash ^= net_hash_mix(net);
Eric W. Biederman098a6972015-03-13 00:07:44 -0500306 hash = hash_32(hash, tcp_metrics_hash_log);
David S. Miller81166dd2012-07-10 03:14:24 -0700307
Eric W. Biederman098a6972015-03-13 00:07:44 -0500308 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
David S. Miller81166dd2012-07-10 03:14:24 -0700309 tm = rcu_dereference(tm->tcpm_next)) {
Christoph Paascha5443022014-01-08 16:05:56 +0100310 if (addr_same(&tm->tcpm_saddr, &saddr) &&
Eric W. Biederman849e8a02015-03-13 00:05:52 -0500311 addr_same(&tm->tcpm_daddr, &daddr) &&
312 net_eq(tm_net(tm), net))
David S. Miller81166dd2012-07-10 03:14:24 -0700313 break;
314 }
315 return tm;
316}
317
David S. Miller51c5d0c2012-07-10 00:49:14 -0700318static 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;
Christoph Paascha5443022014-01-08 16:05:56 +0100323 struct inetpeer_addr saddr, daddr;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700324 unsigned int hash;
325 struct net *net;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700326
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100327 if (sk->sk_family == AF_INET) {
David Ahern3abef282015-08-27 16:07:00 -0700328 inetpeer_set_addr_v4(&saddr, inet_sk(sk)->inet_saddr);
329 inetpeer_set_addr_v4(&daddr, inet_sk(sk)->inet_daddr);
David Ahern72afa352015-08-27 16:06:59 -0700330 hash = ipv4_addr_hash(inet_sk(sk)->inet_daddr);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700331 }
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100332#if IS_ENABLED(CONFIG_IPV6)
333 else if (sk->sk_family == AF_INET6) {
334 if (ipv6_addr_v4mapped(&sk->sk_v6_daddr)) {
David Ahern3abef282015-08-27 16:07:00 -0700335 inetpeer_set_addr_v4(&saddr, inet_sk(sk)->inet_saddr);
336 inetpeer_set_addr_v4(&daddr, inet_sk(sk)->inet_daddr);
David Ahern72afa352015-08-27 16:06:59 -0700337 hash = ipv4_addr_hash(inet_sk(sk)->inet_daddr);
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100338 } else {
David Ahern3abef282015-08-27 16:07:00 -0700339 inetpeer_set_addr_v6(&saddr, &sk->sk_v6_rcv_saddr);
340 inetpeer_set_addr_v6(&daddr, &sk->sk_v6_daddr);
Christoph Paasch3ad88cf2014-01-22 13:58:44 +0100341 hash = ipv6_addr_hash(&sk->sk_v6_daddr);
342 }
343 }
344#endif
345 else
346 return NULL;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700347
David S. Miller51c5d0c2012-07-10 00:49:14 -0700348 net = dev_net(dst->dev);
Eric W. Biederman3e5da622015-03-13 00:05:24 -0500349 hash ^= net_hash_mix(net);
Eric W. Biederman098a6972015-03-13 00:07:44 -0500350 hash = hash_32(hash, 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);
Nikolay Borisov1043e252016-02-03 09:46:52 +0200372 struct net *net = sock_net(sk);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700373 struct tcp_metrics_block *tm;
374 unsigned long rtt;
375 u32 val;
376 int m;
David S. Miller4aabd8e2012-07-09 16:07:30 -0700377
David S. Miller51c5d0c2012-07-10 00:49:14 -0700378 if (sysctl_tcp_nometrics_save || !dst)
David S. Miller4aabd8e2012-07-09 16:07:30 -0700379 return;
380
David S. Miller51c5d0c2012-07-10 00:49:14 -0700381 if (dst->flags & DST_HOST)
David S. Miller4aabd8e2012-07-09 16:07:30 -0700382 dst_confirm(dst);
383
David S. Miller51c5d0c2012-07-10 00:49:14 -0700384 rcu_read_lock();
Eric Dumazet740b0f12014-02-26 14:02:48 -0800385 if (icsk->icsk_backoff || !tp->srtt_us) {
David S. Miller51c5d0c2012-07-10 00:49:14 -0700386 /* This session failed to estimate rtt. Why?
387 * Probably, no packets returned in time. Reset our
388 * results.
David S. Miller4aabd8e2012-07-09 16:07:30 -0700389 */
David S. Miller51c5d0c2012-07-10 00:49:14 -0700390 tm = tcp_get_metrics(sk, dst, false);
391 if (tm && !tcp_metric_locked(tm, TCP_METRIC_RTT))
392 tcp_metric_set(tm, TCP_METRIC_RTT, 0);
393 goto out_unlock;
394 } else
395 tm = tcp_get_metrics(sk, dst, true);
396
397 if (!tm)
398 goto out_unlock;
399
Eric Dumazet740b0f12014-02-26 14:02:48 -0800400 rtt = tcp_metric_get(tm, TCP_METRIC_RTT);
401 m = rtt - tp->srtt_us;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700402
403 /* If newly calculated rtt larger than stored one, store new
404 * one. Otherwise, use EWMA. Remember, rtt overestimation is
405 * always better than underestimation.
406 */
407 if (!tcp_metric_locked(tm, TCP_METRIC_RTT)) {
408 if (m <= 0)
Eric Dumazet740b0f12014-02-26 14:02:48 -0800409 rtt = tp->srtt_us;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700410 else
411 rtt -= (m >> 3);
Eric Dumazet740b0f12014-02-26 14:02:48 -0800412 tcp_metric_set(tm, TCP_METRIC_RTT, rtt);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700413 }
414
415 if (!tcp_metric_locked(tm, TCP_METRIC_RTTVAR)) {
416 unsigned long var;
417
418 if (m < 0)
419 m = -m;
420
421 /* Scale deviation to rttvar fixed point */
422 m >>= 1;
Eric Dumazet740b0f12014-02-26 14:02:48 -0800423 if (m < tp->mdev_us)
424 m = tp->mdev_us;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700425
Eric Dumazet740b0f12014-02-26 14:02:48 -0800426 var = tcp_metric_get(tm, TCP_METRIC_RTTVAR);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700427 if (m >= var)
428 var = m;
429 else
430 var -= (var - m) >> 2;
431
Eric Dumazet740b0f12014-02-26 14:02:48 -0800432 tcp_metric_set(tm, TCP_METRIC_RTTVAR, var);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700433 }
434
435 if (tcp_in_initial_slowstart(tp)) {
436 /* Slow start still did not finish. */
437 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) {
438 val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
439 if (val && (tp->snd_cwnd >> 1) > val)
440 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
441 tp->snd_cwnd >> 1);
David S. Miller4aabd8e2012-07-09 16:07:30 -0700442 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700443 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
444 val = tcp_metric_get(tm, TCP_METRIC_CWND);
445 if (tp->snd_cwnd > val)
446 tcp_metric_set(tm, TCP_METRIC_CWND,
447 tp->snd_cwnd);
David S. Miller4aabd8e2012-07-09 16:07:30 -0700448 }
Yuchung Cheng071d5082015-07-09 13:16:29 -0700449 } else if (!tcp_in_slow_start(tp) &&
David S. Miller51c5d0c2012-07-10 00:49:14 -0700450 icsk->icsk_ca_state == TCP_CA_Open) {
451 /* Cong. avoidance phase, cwnd is reliable. */
452 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH))
453 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
454 max(tp->snd_cwnd >> 1, tp->snd_ssthresh));
455 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
456 val = tcp_metric_get(tm, TCP_METRIC_CWND);
Alexander Duyck21008442012-07-11 17:18:04 -0700457 tcp_metric_set(tm, TCP_METRIC_CWND, (val + tp->snd_cwnd) >> 1);
David S. Miller4aabd8e2012-07-09 16:07:30 -0700458 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700459 } else {
460 /* Else slow start did not finish, cwnd is non-sense,
461 * ssthresh may be also invalid.
462 */
463 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
464 val = tcp_metric_get(tm, TCP_METRIC_CWND);
465 tcp_metric_set(tm, TCP_METRIC_CWND,
466 (val + tp->snd_ssthresh) >> 1);
467 }
468 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) {
469 val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
470 if (val && tp->snd_ssthresh > val)
471 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
472 tp->snd_ssthresh);
473 }
474 if (!tcp_metric_locked(tm, TCP_METRIC_REORDERING)) {
475 val = tcp_metric_get(tm, TCP_METRIC_REORDERING);
476 if (val < tp->reordering &&
Nikolay Borisov1043e252016-02-03 09:46:52 +0200477 tp->reordering != net->ipv4.sysctl_tcp_reordering)
David S. Miller51c5d0c2012-07-10 00:49:14 -0700478 tcp_metric_set(tm, TCP_METRIC_REORDERING,
479 tp->reordering);
David S. Miller4aabd8e2012-07-09 16:07:30 -0700480 }
481 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700482 tm->tcpm_stamp = jiffies;
483out_unlock:
484 rcu_read_unlock();
David S. Miller4aabd8e2012-07-09 16:07:30 -0700485}
486
487/* Initialize metrics on socket. */
488
489void tcp_init_metrics(struct sock *sk)
490{
David S. Miller4aabd8e2012-07-09 16:07:30 -0700491 struct dst_entry *dst = __sk_dst_get(sk);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700492 struct tcp_sock *tp = tcp_sk(sk);
493 struct tcp_metrics_block *tm;
Yuchung Cheng1b7fdd22013-08-30 08:35:53 -0700494 u32 val, crtt = 0; /* cached RTT scaled by 8 */
David S. Miller4aabd8e2012-07-09 16:07:30 -0700495
Ian Morris51456b22015-04-03 09:17:26 +0100496 if (!dst)
David S. Miller4aabd8e2012-07-09 16:07:30 -0700497 goto reset;
498
499 dst_confirm(dst);
500
David S. Miller51c5d0c2012-07-10 00:49:14 -0700501 rcu_read_lock();
502 tm = tcp_get_metrics(sk, dst, true);
503 if (!tm) {
504 rcu_read_unlock();
505 goto reset;
506 }
507
508 if (tcp_metric_locked(tm, TCP_METRIC_CWND))
509 tp->snd_cwnd_clamp = tcp_metric_get(tm, TCP_METRIC_CWND);
510
511 val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
512 if (val) {
513 tp->snd_ssthresh = val;
David S. Miller4aabd8e2012-07-09 16:07:30 -0700514 if (tp->snd_ssthresh > tp->snd_cwnd_clamp)
515 tp->snd_ssthresh = tp->snd_cwnd_clamp;
516 } else {
517 /* ssthresh may have been reduced unnecessarily during.
518 * 3WHS. Restore it back to its initial default.
519 */
520 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
521 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700522 val = tcp_metric_get(tm, TCP_METRIC_REORDERING);
523 if (val && tp->reordering != val) {
David S. Miller4aabd8e2012-07-09 16:07:30 -0700524 tcp_disable_fack(tp);
525 tcp_disable_early_retrans(tp);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700526 tp->reordering = val;
David S. Miller4aabd8e2012-07-09 16:07:30 -0700527 }
528
Eric Dumazet740b0f12014-02-26 14:02:48 -0800529 crtt = tcp_metric_get(tm, TCP_METRIC_RTT);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700530 rcu_read_unlock();
David S. Miller4aabd8e2012-07-09 16:07:30 -0700531reset:
Yuchung Cheng52f20e62013-09-03 14:14:35 -0700532 /* The initial RTT measurement from the SYN/SYN-ACK is not ideal
533 * to seed the RTO for later data packets because SYN packets are
534 * small. Use the per-dst cached values to seed the RTO but keep
535 * the RTT estimator variables intact (e.g., srtt, mdev, rttvar).
536 * Later the RTO will be updated immediately upon obtaining the first
537 * data RTT sample (tcp_rtt_estimator()). Hence the cached RTT only
538 * influences the first RTO but not later RTT estimation.
539 *
540 * But if RTT is not available from the SYN (due to retransmits or
541 * syn cookies) or the cache, force a conservative 3secs timeout.
542 *
543 * A bit of theory. RTT is time passed after "normal" sized packet
544 * is sent until it is ACKed. In normal circumstances sending small
545 * packets force peer to delay ACKs and calculation is correct too.
546 * The algorithm is adaptive and, provided we follow specs, it
547 * NEVER underestimate RTT. BUT! If peer tries to make some clever
548 * tricks sort of "quick acks" for time long enough to decrease RTT
549 * to low value, and then abruptly stops to do it and starts to delay
550 * ACKs, wait for troubles.
551 */
Eric Dumazet740b0f12014-02-26 14:02:48 -0800552 if (crtt > tp->srtt_us) {
Neal Cardwell269aa752013-09-16 21:44:20 -0400553 /* Set RTO like tcp_rtt_estimator(), but from cached RTT. */
Konstantin Khlebnikov9bdfb3b2016-02-21 10:12:39 +0300554 crtt /= 8 * USEC_PER_SEC / HZ;
Neal Cardwell269aa752013-09-16 21:44:20 -0400555 inet_csk(sk)->icsk_rto = crtt + max(2 * crtt, tcp_rto_min(sk));
Eric Dumazet740b0f12014-02-26 14:02:48 -0800556 } else if (tp->srtt_us == 0) {
David S. Miller4aabd8e2012-07-09 16:07:30 -0700557 /* RFC6298: 5.7 We've failed to get a valid RTT sample from
558 * 3WHS. This is most likely due to retransmission,
559 * including spurious one. Reset the RTO back to 3secs
560 * from the more aggressive 1sec to avoid more spurious
561 * retransmission.
562 */
Eric Dumazet740b0f12014-02-26 14:02:48 -0800563 tp->rttvar_us = jiffies_to_usecs(TCP_TIMEOUT_FALLBACK);
564 tp->mdev_us = tp->mdev_max_us = tp->rttvar_us;
565
David S. Miller4aabd8e2012-07-09 16:07:30 -0700566 inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK;
567 }
568 /* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
569 * retransmitted. In light of RFC6298 more aggressive 1sec
570 * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
571 * retransmission has occurred.
572 */
573 if (tp->total_retrans > 1)
574 tp->snd_cwnd = 1;
575 else
576 tp->snd_cwnd = tcp_init_cwnd(tp, dst);
577 tp->snd_cwnd_stamp = tcp_time_stamp;
578}
David S. Millerab92bb22012-07-09 16:19:30 -0700579
Hannes Frederic Sowaa26552a2014-08-14 22:06:12 +0200580bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
581 bool paws_check, bool timestamps)
David S. Millerab92bb22012-07-09 16:19:30 -0700582{
David S. Miller51c5d0c2012-07-10 00:49:14 -0700583 struct tcp_metrics_block *tm;
584 bool ret;
585
David S. Millerab92bb22012-07-09 16:19:30 -0700586 if (!dst)
587 return false;
David S. Miller51c5d0c2012-07-10 00:49:14 -0700588
589 rcu_read_lock();
590 tm = __tcp_get_metrics_req(req, dst);
David S. Miller81166dd2012-07-10 03:14:24 -0700591 if (paws_check) {
592 if (tm &&
593 (u32)get_seconds() - tm->tcpm_ts_stamp < TCP_PAWS_MSL &&
Hannes Frederic Sowaa26552a2014-08-14 22:06:12 +0200594 ((s32)(tm->tcpm_ts - req->ts_recent) > TCP_PAWS_WINDOW ||
595 !timestamps))
David S. Miller81166dd2012-07-10 03:14:24 -0700596 ret = false;
597 else
598 ret = true;
599 } else {
600 if (tm && tcp_metric_get(tm, TCP_METRIC_RTT) && tm->tcpm_ts_stamp)
601 ret = true;
602 else
603 ret = false;
604 }
David S. Miller51c5d0c2012-07-10 00:49:14 -0700605 rcu_read_unlock();
606
607 return ret;
David S. Millerab92bb22012-07-09 16:19:30 -0700608}
609EXPORT_SYMBOL_GPL(tcp_peer_is_proven);
David S. Miller51c5d0c2012-07-10 00:49:14 -0700610
David S. Miller81166dd2012-07-10 03:14:24 -0700611void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst)
612{
613 struct tcp_metrics_block *tm;
614
615 rcu_read_lock();
616 tm = tcp_get_metrics(sk, dst, true);
617 if (tm) {
618 struct tcp_sock *tp = tcp_sk(sk);
619
620 if ((u32)get_seconds() - tm->tcpm_ts_stamp <= TCP_PAWS_MSL) {
621 tp->rx_opt.ts_recent_stamp = tm->tcpm_ts_stamp;
622 tp->rx_opt.ts_recent = tm->tcpm_ts;
623 }
624 }
625 rcu_read_unlock();
626}
627EXPORT_SYMBOL_GPL(tcp_fetch_timewait_stamp);
628
629/* VJ's idea. Save last timestamp seen from this destination and hold
630 * it at least for normal timewait interval to use for duplicate
631 * segment detection in subsequent connections, before they enter
632 * synchronized state.
633 */
634bool tcp_remember_stamp(struct sock *sk)
635{
636 struct dst_entry *dst = __sk_dst_get(sk);
637 bool ret = false;
638
639 if (dst) {
640 struct tcp_metrics_block *tm;
641
642 rcu_read_lock();
643 tm = tcp_get_metrics(sk, dst, true);
644 if (tm) {
645 struct tcp_sock *tp = tcp_sk(sk);
646
647 if ((s32)(tm->tcpm_ts - tp->rx_opt.ts_recent) <= 0 ||
648 ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL &&
649 tm->tcpm_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) {
650 tm->tcpm_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp;
651 tm->tcpm_ts = tp->rx_opt.ts_recent;
652 }
653 ret = true;
654 }
655 rcu_read_unlock();
656 }
657 return ret;
658}
659
660bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw)
661{
662 struct tcp_metrics_block *tm;
663 bool ret = false;
664
665 rcu_read_lock();
666 tm = __tcp_get_metrics_tw(tw);
Julian Anastasov9a0a9502012-07-23 10:46:38 +0300667 if (tm) {
David S. Miller81166dd2012-07-10 03:14:24 -0700668 const struct tcp_timewait_sock *tcptw;
669 struct sock *sk = (struct sock *) tw;
670
671 tcptw = tcp_twsk(sk);
672 if ((s32)(tm->tcpm_ts - tcptw->tw_ts_recent) <= 0 ||
673 ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL &&
674 tm->tcpm_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) {
675 tm->tcpm_ts_stamp = (u32)tcptw->tw_ts_recent_stamp;
676 tm->tcpm_ts = tcptw->tw_ts_recent;
677 }
678 ret = true;
679 }
680 rcu_read_unlock();
681
682 return ret;
683}
684
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000685static DEFINE_SEQLOCK(fastopen_seqlock);
686
687void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
Yuchung Chengaab48742012-07-19 06:43:10 +0000688 struct tcp_fastopen_cookie *cookie,
689 int *syn_loss, unsigned long *last_syn_loss)
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000690{
691 struct tcp_metrics_block *tm;
692
693 rcu_read_lock();
694 tm = tcp_get_metrics(sk, __sk_dst_get(sk), false);
695 if (tm) {
696 struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen;
697 unsigned int seq;
698
699 do {
700 seq = read_seqbegin(&fastopen_seqlock);
701 if (tfom->mss)
702 *mss = tfom->mss;
703 *cookie = tfom->cookie;
Daniel Lee2646c832015-04-06 14:37:27 -0700704 if (cookie->len <= 0 && tfom->try_exp == 1)
705 cookie->exp = true;
Yuchung Chengaab48742012-07-19 06:43:10 +0000706 *syn_loss = tfom->syn_loss;
707 *last_syn_loss = *syn_loss ? tfom->last_syn_loss : 0;
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000708 } while (read_seqretry(&fastopen_seqlock, seq));
709 }
710 rcu_read_unlock();
711}
712
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000713void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
Daniel Lee2646c832015-04-06 14:37:27 -0700714 struct tcp_fastopen_cookie *cookie, bool syn_lost,
715 u16 try_exp)
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000716{
Eric Dumazetdccf76c2013-11-13 15:00:46 -0800717 struct dst_entry *dst = __sk_dst_get(sk);
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000718 struct tcp_metrics_block *tm;
719
Eric Dumazetdccf76c2013-11-13 15:00:46 -0800720 if (!dst)
721 return;
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000722 rcu_read_lock();
Eric Dumazetdccf76c2013-11-13 15:00:46 -0800723 tm = tcp_get_metrics(sk, dst, true);
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000724 if (tm) {
725 struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen;
726
727 write_seqlock_bh(&fastopen_seqlock);
Yuchung Chengc9686012013-10-29 10:09:05 -0700728 if (mss)
729 tfom->mss = mss;
730 if (cookie && cookie->len > 0)
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000731 tfom->cookie = *cookie;
Daniel Lee2646c832015-04-06 14:37:27 -0700732 else if (try_exp > tfom->try_exp &&
733 tfom->cookie.len <= 0 && !tfom->cookie.exp)
734 tfom->try_exp = try_exp;
Yuchung Chengaab48742012-07-19 06:43:10 +0000735 if (syn_lost) {
736 ++tfom->syn_loss;
737 tfom->last_syn_loss = jiffies;
738 } else
739 tfom->syn_loss = 0;
Yuchung Cheng1fe4c482012-07-19 06:43:06 +0000740 write_sequnlock_bh(&fastopen_seqlock);
741 }
742 rcu_read_unlock();
743}
744
Julian Anastasovd23ff702012-09-04 11:03:15 +0000745static struct genl_family tcp_metrics_nl_family = {
746 .id = GENL_ID_GENERATE,
747 .hdrsize = 0,
748 .name = TCP_METRICS_GENL_NAME,
749 .version = TCP_METRICS_GENL_VERSION,
750 .maxattr = TCP_METRICS_ATTR_MAX,
751 .netnsok = true,
752};
753
754static struct nla_policy tcp_metrics_nl_policy[TCP_METRICS_ATTR_MAX + 1] = {
755 [TCP_METRICS_ATTR_ADDR_IPV4] = { .type = NLA_U32, },
756 [TCP_METRICS_ATTR_ADDR_IPV6] = { .type = NLA_BINARY,
757 .len = sizeof(struct in6_addr), },
758 /* Following attributes are not received for GET/DEL,
759 * we keep them for reference
760 */
761#if 0
762 [TCP_METRICS_ATTR_AGE] = { .type = NLA_MSECS, },
763 [TCP_METRICS_ATTR_TW_TSVAL] = { .type = NLA_U32, },
764 [TCP_METRICS_ATTR_TW_TS_STAMP] = { .type = NLA_S32, },
765 [TCP_METRICS_ATTR_VALS] = { .type = NLA_NESTED, },
766 [TCP_METRICS_ATTR_FOPEN_MSS] = { .type = NLA_U16, },
767 [TCP_METRICS_ATTR_FOPEN_SYN_DROPS] = { .type = NLA_U16, },
768 [TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS] = { .type = NLA_MSECS, },
769 [TCP_METRICS_ATTR_FOPEN_COOKIE] = { .type = NLA_BINARY,
770 .len = TCP_FASTOPEN_COOKIE_MAX, },
771#endif
772};
773
774/* Add attributes, caller cancels its header on failure */
775static int tcp_metrics_fill_info(struct sk_buff *msg,
776 struct tcp_metrics_block *tm)
777{
778 struct nlattr *nest;
779 int i;
780
Christoph Paasch324fd552014-01-08 16:05:55 +0100781 switch (tm->tcpm_daddr.family) {
Julian Anastasovd23ff702012-09-04 11:03:15 +0000782 case AF_INET:
Jiri Benc930345e2015-03-29 16:59:25 +0200783 if (nla_put_in_addr(msg, TCP_METRICS_ATTR_ADDR_IPV4,
David Ahern3abef282015-08-27 16:07:00 -0700784 inetpeer_get_addr_v4(&tm->tcpm_daddr)) < 0)
Julian Anastasovd23ff702012-09-04 11:03:15 +0000785 goto nla_put_failure;
Jiri Benc930345e2015-03-29 16:59:25 +0200786 if (nla_put_in_addr(msg, TCP_METRICS_ATTR_SADDR_IPV4,
David Ahern3abef282015-08-27 16:07:00 -0700787 inetpeer_get_addr_v4(&tm->tcpm_saddr)) < 0)
Christoph Paasch8a593592014-01-08 16:05:57 +0100788 goto nla_put_failure;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000789 break;
790 case AF_INET6:
Jiri Benc930345e2015-03-29 16:59:25 +0200791 if (nla_put_in6_addr(msg, TCP_METRICS_ATTR_ADDR_IPV6,
David Ahern3abef282015-08-27 16:07:00 -0700792 inetpeer_get_addr_v6(&tm->tcpm_daddr)) < 0)
Julian Anastasovd23ff702012-09-04 11:03:15 +0000793 goto nla_put_failure;
Jiri Benc930345e2015-03-29 16:59:25 +0200794 if (nla_put_in6_addr(msg, TCP_METRICS_ATTR_SADDR_IPV6,
David Ahern3abef282015-08-27 16:07:00 -0700795 inetpeer_get_addr_v6(&tm->tcpm_saddr)) < 0)
Christoph Paasch8a593592014-01-08 16:05:57 +0100796 goto nla_put_failure;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000797 break;
798 default:
799 return -EAFNOSUPPORT;
800 }
801
802 if (nla_put_msecs(msg, TCP_METRICS_ATTR_AGE,
803 jiffies - tm->tcpm_stamp) < 0)
804 goto nla_put_failure;
805 if (tm->tcpm_ts_stamp) {
806 if (nla_put_s32(msg, TCP_METRICS_ATTR_TW_TS_STAMP,
807 (s32) (get_seconds() - tm->tcpm_ts_stamp)) < 0)
808 goto nla_put_failure;
809 if (nla_put_u32(msg, TCP_METRICS_ATTR_TW_TSVAL,
810 tm->tcpm_ts) < 0)
811 goto nla_put_failure;
812 }
813
814 {
815 int n = 0;
816
817 nest = nla_nest_start(msg, TCP_METRICS_ATTR_VALS);
818 if (!nest)
819 goto nla_put_failure;
Eric Dumazet740b0f12014-02-26 14:02:48 -0800820 for (i = 0; i < TCP_METRIC_MAX_KERNEL + 1; i++) {
821 u32 val = tm->tcpm_vals[i];
822
823 if (!val)
Julian Anastasovd23ff702012-09-04 11:03:15 +0000824 continue;
Eric Dumazet740b0f12014-02-26 14:02:48 -0800825 if (i == TCP_METRIC_RTT) {
826 if (nla_put_u32(msg, TCP_METRIC_RTT_US + 1,
827 val) < 0)
828 goto nla_put_failure;
829 n++;
830 val = max(val / 1000, 1U);
831 }
832 if (i == TCP_METRIC_RTTVAR) {
833 if (nla_put_u32(msg, TCP_METRIC_RTTVAR_US + 1,
834 val) < 0)
835 goto nla_put_failure;
836 n++;
837 val = max(val / 1000, 1U);
838 }
839 if (nla_put_u32(msg, i + 1, val) < 0)
Julian Anastasovd23ff702012-09-04 11:03:15 +0000840 goto nla_put_failure;
841 n++;
842 }
843 if (n)
844 nla_nest_end(msg, nest);
845 else
846 nla_nest_cancel(msg, nest);
847 }
848
849 {
850 struct tcp_fastopen_metrics tfom_copy[1], *tfom;
851 unsigned int seq;
852
853 do {
854 seq = read_seqbegin(&fastopen_seqlock);
855 tfom_copy[0] = tm->tcpm_fastopen;
856 } while (read_seqretry(&fastopen_seqlock, seq));
857
858 tfom = tfom_copy;
859 if (tfom->mss &&
860 nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_MSS,
861 tfom->mss) < 0)
862 goto nla_put_failure;
863 if (tfom->syn_loss &&
864 (nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROPS,
865 tfom->syn_loss) < 0 ||
866 nla_put_msecs(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS,
867 jiffies - tfom->last_syn_loss) < 0))
868 goto nla_put_failure;
869 if (tfom->cookie.len > 0 &&
870 nla_put(msg, TCP_METRICS_ATTR_FOPEN_COOKIE,
871 tfom->cookie.len, tfom->cookie.val) < 0)
872 goto nla_put_failure;
873 }
874
875 return 0;
876
877nla_put_failure:
878 return -EMSGSIZE;
879}
880
881static int tcp_metrics_dump_info(struct sk_buff *skb,
882 struct netlink_callback *cb,
883 struct tcp_metrics_block *tm)
884{
885 void *hdr;
886
Eric W. Biederman15e47302012-09-07 20:12:54 +0000887 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
Julian Anastasovd23ff702012-09-04 11:03:15 +0000888 &tcp_metrics_nl_family, NLM_F_MULTI,
889 TCP_METRICS_CMD_GET);
890 if (!hdr)
891 return -EMSGSIZE;
892
893 if (tcp_metrics_fill_info(skb, tm) < 0)
894 goto nla_put_failure;
895
Johannes Berg053c0952015-01-16 22:09:00 +0100896 genlmsg_end(skb, hdr);
897 return 0;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000898
899nla_put_failure:
900 genlmsg_cancel(skb, hdr);
901 return -EMSGSIZE;
902}
903
904static int tcp_metrics_nl_dump(struct sk_buff *skb,
905 struct netlink_callback *cb)
906{
907 struct net *net = sock_net(skb->sk);
Eric W. Biederman098a6972015-03-13 00:07:44 -0500908 unsigned int max_rows = 1U << tcp_metrics_hash_log;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000909 unsigned int row, s_row = cb->args[0];
910 int s_col = cb->args[1], col = s_col;
911
912 for (row = s_row; row < max_rows; row++, s_col = 0) {
913 struct tcp_metrics_block *tm;
Eric W. Biederman098a6972015-03-13 00:07:44 -0500914 struct tcpm_hash_bucket *hb = tcp_metrics_hash + row;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000915
916 rcu_read_lock();
917 for (col = 0, tm = rcu_dereference(hb->chain); tm;
918 tm = rcu_dereference(tm->tcpm_next), col++) {
Eric W. Biederman849e8a02015-03-13 00:05:52 -0500919 if (!net_eq(tm_net(tm), net))
920 continue;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000921 if (col < s_col)
922 continue;
923 if (tcp_metrics_dump_info(skb, cb, tm) < 0) {
924 rcu_read_unlock();
925 goto done;
926 }
927 }
928 rcu_read_unlock();
929 }
930
931done:
932 cb->args[0] = row;
933 cb->args[1] = col;
934 return skb->len;
935}
936
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100937static int __parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
938 unsigned int *hash, int optional, int v4, int v6)
Julian Anastasovd23ff702012-09-04 11:03:15 +0000939{
940 struct nlattr *a;
941
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100942 a = info->attrs[v4];
Julian Anastasovd23ff702012-09-04 11:03:15 +0000943 if (a) {
David Ahern3abef282015-08-27 16:07:00 -0700944 inetpeer_set_addr_v4(addr, nla_get_in_addr(a));
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100945 if (hash)
David Ahern3abef282015-08-27 16:07:00 -0700946 *hash = ipv4_addr_hash(inetpeer_get_addr_v4(addr));
Julian Anastasovd23ff702012-09-04 11:03:15 +0000947 return 0;
948 }
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100949 a = info->attrs[v6];
Julian Anastasovd23ff702012-09-04 11:03:15 +0000950 if (a) {
David Ahern3abef282015-08-27 16:07:00 -0700951 struct in6_addr in6;
952
Julian Anastasov2c42a3f2012-10-30 12:03:09 +0000953 if (nla_len(a) != sizeof(struct in6_addr))
Julian Anastasovd23ff702012-09-04 11:03:15 +0000954 return -EINVAL;
David Ahern3abef282015-08-27 16:07:00 -0700955 in6 = nla_get_in6_addr(a);
956 inetpeer_set_addr_v6(addr, &in6);
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100957 if (hash)
David Ahern3abef282015-08-27 16:07:00 -0700958 *hash = ipv6_addr_hash(inetpeer_get_addr_v6(addr));
Julian Anastasovd23ff702012-09-04 11:03:15 +0000959 return 0;
960 }
961 return optional ? 1 : -EAFNOSUPPORT;
962}
963
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100964static int parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
965 unsigned int *hash, int optional)
966{
967 return __parse_nl_addr(info, addr, hash, optional,
968 TCP_METRICS_ATTR_ADDR_IPV4,
969 TCP_METRICS_ATTR_ADDR_IPV6);
970}
971
972static int parse_nl_saddr(struct genl_info *info, struct inetpeer_addr *addr)
973{
974 return __parse_nl_addr(info, addr, NULL, 0,
975 TCP_METRICS_ATTR_SADDR_IPV4,
976 TCP_METRICS_ATTR_SADDR_IPV6);
977}
978
Julian Anastasovd23ff702012-09-04 11:03:15 +0000979static int tcp_metrics_nl_cmd_get(struct sk_buff *skb, struct genl_info *info)
980{
981 struct tcp_metrics_block *tm;
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100982 struct inetpeer_addr saddr, daddr;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000983 unsigned int hash;
984 struct sk_buff *msg;
985 struct net *net = genl_info_net(info);
986 void *reply;
987 int ret;
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100988 bool src = true;
Julian Anastasovd23ff702012-09-04 11:03:15 +0000989
Christoph Paasch324fd552014-01-08 16:05:55 +0100990 ret = parse_nl_addr(info, &daddr, &hash, 0);
Julian Anastasovd23ff702012-09-04 11:03:15 +0000991 if (ret < 0)
992 return ret;
993
Christoph Paasch3e7013d2014-01-08 16:05:59 +0100994 ret = parse_nl_saddr(info, &saddr);
995 if (ret < 0)
996 src = false;
997
Julian Anastasovd23ff702012-09-04 11:03:15 +0000998 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
999 if (!msg)
1000 return -ENOMEM;
1001
1002 reply = genlmsg_put_reply(msg, info, &tcp_metrics_nl_family, 0,
1003 info->genlhdr->cmd);
1004 if (!reply)
1005 goto nla_put_failure;
1006
Eric W. Biederman3e5da622015-03-13 00:05:24 -05001007 hash ^= net_hash_mix(net);
Eric W. Biederman098a6972015-03-13 00:07:44 -05001008 hash = hash_32(hash, tcp_metrics_hash_log);
Julian Anastasovd23ff702012-09-04 11:03:15 +00001009 ret = -ESRCH;
1010 rcu_read_lock();
Eric W. Biederman098a6972015-03-13 00:07:44 -05001011 for (tm = rcu_dereference(tcp_metrics_hash[hash].chain); tm;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001012 tm = rcu_dereference(tm->tcpm_next)) {
Christoph Paasch3e7013d2014-01-08 16:05:59 +01001013 if (addr_same(&tm->tcpm_daddr, &daddr) &&
Eric W. Biederman849e8a02015-03-13 00:05:52 -05001014 (!src || addr_same(&tm->tcpm_saddr, &saddr)) &&
1015 net_eq(tm_net(tm), net)) {
Julian Anastasovd23ff702012-09-04 11:03:15 +00001016 ret = tcp_metrics_fill_info(msg, tm);
1017 break;
1018 }
1019 }
1020 rcu_read_unlock();
1021 if (ret < 0)
1022 goto out_free;
1023
1024 genlmsg_end(msg, reply);
1025 return genlmsg_reply(msg, info);
1026
1027nla_put_failure:
1028 ret = -EMSGSIZE;
1029
1030out_free:
1031 nlmsg_free(msg);
1032 return ret;
1033}
1034
Eric W. Biederman8a4bff72015-03-13 00:06:43 -05001035static void tcp_metrics_flush_all(struct net *net)
Julian Anastasovd23ff702012-09-04 11:03:15 +00001036{
Eric W. Biederman098a6972015-03-13 00:07:44 -05001037 unsigned int max_rows = 1U << tcp_metrics_hash_log;
1038 struct tcpm_hash_bucket *hb = tcp_metrics_hash;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001039 struct tcp_metrics_block *tm;
1040 unsigned int row;
1041
1042 for (row = 0; row < max_rows; row++, hb++) {
Eric W. Biederman04f721c2015-03-13 00:07:10 -05001043 struct tcp_metrics_block __rcu **pp;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001044 spin_lock_bh(&tcp_metrics_lock);
Eric W. Biederman04f721c2015-03-13 00:07:10 -05001045 pp = &hb->chain;
Eric Dumazet9f1ab182015-03-16 07:14:34 -07001046 for (tm = deref_locked(*pp); tm; tm = deref_locked(*pp)) {
Eric W. Biederman04f721c2015-03-13 00:07:10 -05001047 if (net_eq(tm_net(tm), net)) {
1048 *pp = tm->tcpm_next;
1049 kfree_rcu(tm, rcu_head);
1050 } else {
1051 pp = &tm->tcpm_next;
1052 }
Julian Anastasovd23ff702012-09-04 11:03:15 +00001053 }
Eric W. Biederman04f721c2015-03-13 00:07:10 -05001054 spin_unlock_bh(&tcp_metrics_lock);
Julian Anastasovd23ff702012-09-04 11:03:15 +00001055 }
Julian Anastasovd23ff702012-09-04 11:03:15 +00001056}
1057
1058static int tcp_metrics_nl_cmd_del(struct sk_buff *skb, struct genl_info *info)
1059{
1060 struct tcpm_hash_bucket *hb;
Christoph Paasch00ca9c52014-01-21 13:30:26 +01001061 struct tcp_metrics_block *tm;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001062 struct tcp_metrics_block __rcu **pp;
Christoph Paasch3e7013d2014-01-08 16:05:59 +01001063 struct inetpeer_addr saddr, daddr;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001064 unsigned int hash;
1065 struct net *net = genl_info_net(info);
1066 int ret;
Christoph Paasch00ca9c52014-01-21 13:30:26 +01001067 bool src = true, found = false;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001068
Christoph Paasch324fd552014-01-08 16:05:55 +01001069 ret = parse_nl_addr(info, &daddr, &hash, 1);
Julian Anastasovd23ff702012-09-04 11:03:15 +00001070 if (ret < 0)
1071 return ret;
Eric W. Biederman8a4bff72015-03-13 00:06:43 -05001072 if (ret > 0) {
1073 tcp_metrics_flush_all(net);
1074 return 0;
1075 }
Christoph Paasch3e7013d2014-01-08 16:05:59 +01001076 ret = parse_nl_saddr(info, &saddr);
1077 if (ret < 0)
1078 src = false;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001079
Eric W. Biederman3e5da622015-03-13 00:05:24 -05001080 hash ^= net_hash_mix(net);
Eric W. Biederman098a6972015-03-13 00:07:44 -05001081 hash = hash_32(hash, tcp_metrics_hash_log);
1082 hb = tcp_metrics_hash + hash;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001083 pp = &hb->chain;
1084 spin_lock_bh(&tcp_metrics_lock);
Eric Dumazet9f1ab182015-03-16 07:14:34 -07001085 for (tm = deref_locked(*pp); tm; tm = deref_locked(*pp)) {
Christoph Paasch3e7013d2014-01-08 16:05:59 +01001086 if (addr_same(&tm->tcpm_daddr, &daddr) &&
Eric W. Biederman849e8a02015-03-13 00:05:52 -05001087 (!src || addr_same(&tm->tcpm_saddr, &saddr)) &&
1088 net_eq(tm_net(tm), net)) {
Julian Anastasovd23ff702012-09-04 11:03:15 +00001089 *pp = tm->tcpm_next;
Christoph Paasch00ca9c52014-01-21 13:30:26 +01001090 kfree_rcu(tm, rcu_head);
1091 found = true;
Christoph Paaschbbf852b2014-01-08 16:05:58 +01001092 } else {
1093 pp = &tm->tcpm_next;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001094 }
1095 }
1096 spin_unlock_bh(&tcp_metrics_lock);
Christoph Paasch00ca9c52014-01-21 13:30:26 +01001097 if (!found)
Julian Anastasovd23ff702012-09-04 11:03:15 +00001098 return -ESRCH;
Julian Anastasovd23ff702012-09-04 11:03:15 +00001099 return 0;
1100}
1101
Johannes Berg4534de82013-11-14 17:14:46 +01001102static const struct genl_ops tcp_metrics_nl_ops[] = {
Julian Anastasovd23ff702012-09-04 11:03:15 +00001103 {
1104 .cmd = TCP_METRICS_CMD_GET,
1105 .doit = tcp_metrics_nl_cmd_get,
1106 .dumpit = tcp_metrics_nl_dump,
1107 .policy = tcp_metrics_nl_policy,
Julian Anastasovd23ff702012-09-04 11:03:15 +00001108 },
1109 {
1110 .cmd = TCP_METRICS_CMD_DEL,
1111 .doit = tcp_metrics_nl_cmd_del,
1112 .policy = tcp_metrics_nl_policy,
1113 .flags = GENL_ADMIN_PERM,
1114 },
1115};
1116
Eric Dumazet5815d5e2012-07-19 23:02:34 +00001117static unsigned int tcpmhash_entries;
David S. Miller51c5d0c2012-07-10 00:49:14 -07001118static int __init set_tcpmhash_entries(char *str)
1119{
1120 ssize_t ret;
1121
1122 if (!str)
1123 return 0;
1124
Eric Dumazet5815d5e2012-07-19 23:02:34 +00001125 ret = kstrtouint(str, 0, &tcpmhash_entries);
David S. Miller51c5d0c2012-07-10 00:49:14 -07001126 if (ret)
1127 return 0;
1128
1129 return 1;
1130}
1131__setup("tcpmhash_entries=", set_tcpmhash_entries);
1132
1133static int __net_init tcp_net_metrics_init(struct net *net)
1134{
Eric Dumazet5815d5e2012-07-19 23:02:34 +00001135 size_t size;
1136 unsigned int slots;
David S. Miller51c5d0c2012-07-10 00:49:14 -07001137
Eric W. Biederman098a6972015-03-13 00:07:44 -05001138 if (!net_eq(net, &init_net))
1139 return 0;
1140
David S. Miller51c5d0c2012-07-10 00:49:14 -07001141 slots = tcpmhash_entries;
1142 if (!slots) {
1143 if (totalram_pages >= 128 * 1024)
1144 slots = 16 * 1024;
1145 else
1146 slots = 8 * 1024;
1147 }
1148
Eric W. Biederman098a6972015-03-13 00:07:44 -05001149 tcp_metrics_hash_log = order_base_2(slots);
1150 size = sizeof(struct tcpm_hash_bucket) << tcp_metrics_hash_log;
David S. Miller51c5d0c2012-07-10 00:49:14 -07001151
Eric W. Biederman098a6972015-03-13 00:07:44 -05001152 tcp_metrics_hash = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
1153 if (!tcp_metrics_hash)
1154 tcp_metrics_hash = vzalloc(size);
Eric Dumazet976a7022012-11-16 05:31:53 +00001155
Eric W. Biederman098a6972015-03-13 00:07:44 -05001156 if (!tcp_metrics_hash)
David S. Miller51c5d0c2012-07-10 00:49:14 -07001157 return -ENOMEM;
1158
David S. Miller51c5d0c2012-07-10 00:49:14 -07001159 return 0;
1160}
1161
1162static void __net_exit tcp_net_metrics_exit(struct net *net)
1163{
Eric W. Biederman098a6972015-03-13 00:07:44 -05001164 tcp_metrics_flush_all(net);
David S. Miller51c5d0c2012-07-10 00:49:14 -07001165}
1166
1167static __net_initdata struct pernet_operations tcp_net_metrics_ops = {
1168 .init = tcp_net_metrics_init,
1169 .exit = tcp_net_metrics_exit,
1170};
1171
1172void __init tcp_metrics_init(void)
1173{
Julian Anastasovd23ff702012-09-04 11:03:15 +00001174 int ret;
1175
1176 ret = register_pernet_subsys(&tcp_net_metrics_ops);
1177 if (ret < 0)
Eric W. Biederman64935172015-03-13 00:04:51 -05001178 panic("Could not allocate the tcp_metrics hash table\n");
1179
Julian Anastasovd23ff702012-09-04 11:03:15 +00001180 ret = genl_register_family_with_ops(&tcp_metrics_nl_family,
Johannes Bergc53ed742013-11-19 15:19:31 +01001181 tcp_metrics_nl_ops);
Julian Anastasovd23ff702012-09-04 11:03:15 +00001182 if (ret < 0)
Eric W. Biederman64935172015-03-13 00:04:51 -05001183 panic("Could not register tcp_metrics generic netlink\n");
David S. Miller51c5d0c2012-07-10 00:49:14 -07001184}