xref: /freebsd/sys/netinet/tcp_stats.c (revision e0c4386e7e71d93b0edc0c8fa156263fc4a8b0b6)
1 /*-
2  * Copyright (c) 2016-2018 Netflix, Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 /*
28  * Author: Lawrence Stewart <lstewart@netflix.com>
29  */
30 
31 #include <sys/param.h>
32 #include <sys/arb.h>
33 #include <sys/errno.h>
34 #include <sys/malloc.h>
35 #include <sys/qmath.h>
36 #include <sys/queue.h>
37 #include <sys/socket.h>
38 #include <sys/socketvar.h>
39 #include <sys/sysctl.h>
40 #ifdef _KERNEL
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/rmlock.h>
44 #include <sys/systm.h>
45 #endif
46 #include <sys/stats.h>
47 
48 #include <net/vnet.h>
49 
50 #include <netinet/in.h>
51 #include <netinet/in_pcb.h>
52 #include <netinet/tcp.h>
53 #include <netinet/tcp_var.h>
54 
55 #include <netinet/cc/cc.h>
56 
57 VNET_DEFINE(int, tcp_perconn_stats_dflt_tpl) = -1;
58 
59 #ifndef _KERNEL
60 #define	V_tcp_perconn_stats_enable	VNET(tcp_perconn_stats_enable)
61 #define	V_tcp_perconn_stats_dflt_tpl	VNET(tcp_perconn_stats_dflt_tpl)
62 #else /* _KERNEL */
63 
64 VNET_DEFINE(int, tcp_perconn_stats_enable) = 2;
65 VNET_DEFINE_STATIC(struct stats_tpl_sample_rate *, tcp_perconn_stats_sample_rates);
66 VNET_DEFINE_STATIC(int, tcp_stats_nrates) = 0;
67 #define	V_tcp_perconn_stats_sample_rates VNET(tcp_perconn_stats_sample_rates)
68 #define	V_tcp_stats_nrates		VNET(tcp_stats_nrates)
69 
70 static struct rmlock tcp_stats_tpl_sampling_lock;
71 static int tcp_stats_tpl_sr_cb(enum stats_tpl_sr_cb_action action,
72     struct stats_tpl_sample_rate **rates, int *nrates, void *ctx);
73 
74 SYSCTL_INT(_net_inet_tcp, OID_AUTO, perconn_stats_enable,
75     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_perconn_stats_enable), 0,
76     "Enable per-connection TCP stats gathering; 1 enables for all connections, "
77     "2 enables random sampling across log id connection groups");
78 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, perconn_stats_sample_rates,
79     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, tcp_stats_tpl_sr_cb,
80     sizeof(struct rm_priotracker), stats_tpl_sample_rates, "A",
81     "TCP stats per template random sampling rates, in CSV tpl_spec=percent "
82     "key-value pairs (see stats(9) for template spec details)");
83 #endif /* _KERNEL */
84 
85 #ifdef _KERNEL
86 int
87 #else
88 static int
89 /* Ensure all templates are also added to the userland template list. */
90 __attribute__ ((constructor))
91 #endif
92 tcp_stats_init(void)
93 {
94 	int err, lasterr;
95 
96 	err = lasterr = 0;
97 
98 	V_tcp_perconn_stats_dflt_tpl = stats_tpl_alloc("TCP_DEFAULT", 0);
99 	if (V_tcp_perconn_stats_dflt_tpl < 0)
100 		return (-V_tcp_perconn_stats_dflt_tpl);
101 
102 	struct voistatspec vss_sum[] = {
103 		STATS_VSS_SUM(),
104 	};
105 	err |= stats_tpl_add_voistats(V_tcp_perconn_stats_dflt_tpl,
106 	    VOI_TCP_TXPB, "TCP_TXPB", VSD_DTYPE_INT_U64,
107 	    NVSS(vss_sum), vss_sum, 0);
108 	lasterr = err ? err : lasterr;
109 	err |= stats_tpl_add_voistats(V_tcp_perconn_stats_dflt_tpl,
110 	    VOI_TCP_RETXPB, "TCP_RETXPB", VSD_DTYPE_INT_U32,
111 	    NVSS(vss_sum), vss_sum, 0);
112 	lasterr = err ? err : lasterr;
113 
114 	struct voistatspec vss_max[] = {
115 		STATS_VSS_MAX(),
116 	};
117 	err |= stats_tpl_add_voistats(V_tcp_perconn_stats_dflt_tpl,
118 	    VOI_TCP_FRWIN, "TCP_FRWIN", VSD_DTYPE_INT_ULONG,
119 	    NVSS(vss_max), vss_max, 0);
120 	lasterr = err ? err : lasterr;
121 	err |= stats_tpl_add_voistats(V_tcp_perconn_stats_dflt_tpl,
122 	    VOI_TCP_LCWIN, "TCP_LCWIN", VSD_DTYPE_INT_ULONG,
123 	    NVSS(vss_max), vss_max, 0);
124 	lasterr = err ? err : lasterr;
125 
126 	struct voistatspec vss_rtt[] = {
127 		STATS_VSS_MAX(),
128 		STATS_VSS_MIN(),
129 		STATS_VSS_TDGSTCLUST32(20, 4),
130 	};
131 	err |= stats_tpl_add_voistats(V_tcp_perconn_stats_dflt_tpl,
132 	    VOI_TCP_RTT, "TCP_RTT", VSD_DTYPE_INT_U32,
133 	    NVSS(vss_rtt), vss_rtt, 0);
134 	lasterr = err ? err : lasterr;
135 
136 	struct voistatspec vss_congsig[] = {
137 		STATS_VSS_DVHIST32_USR(HBKTS(DVBKT(CC_ECN), DVBKT(CC_RTO),
138 		    DVBKT(CC_RTO_ERR), DVBKT(CC_NDUPACK)), 0)
139 	};
140 	err |= stats_tpl_add_voistats(V_tcp_perconn_stats_dflt_tpl,
141 	    VOI_TCP_CSIG, "TCP_CSIG", VSD_DTYPE_INT_U32,
142 	    NVSS(vss_congsig), vss_congsig, 0);
143 	lasterr = err ? err : lasterr;
144 
145 	struct voistatspec vss_gput[] = {
146 		STATS_VSS_MAX(),
147 		STATS_VSS_TDGSTCLUST32(20, 4),
148 	};
149 	err |= stats_tpl_add_voistats(V_tcp_perconn_stats_dflt_tpl,
150 	    VOI_TCP_GPUT, "TCP_GPUT", VSD_DTYPE_INT_U32,
151 	    NVSS(vss_gput), vss_gput, 0);
152 	lasterr = err ? err : lasterr;
153 
154 	struct voistatspec vss_gput_nd[] = {
155 		STATS_VSS_TDGSTCLUST32(10, 4),
156 	};
157 	err |= stats_tpl_add_voistats(V_tcp_perconn_stats_dflt_tpl,
158 	    VOI_TCP_GPUT_ND, "TCP_GPUT_ND", VSD_DTYPE_INT_S32,
159 	    NVSS(vss_gput_nd), vss_gput_nd, 0);
160 	lasterr = err ? err : lasterr;
161 
162 	struct voistatspec vss_windiff[] = {
163 		STATS_VSS_CRHIST32_USR(HBKTS(CRBKT(0)), VSD_HIST_LBOUND_INF)
164 	};
165 	err |= stats_tpl_add_voistats(V_tcp_perconn_stats_dflt_tpl,
166 	    VOI_TCP_CALCFRWINDIFF, "TCP_CALCFRWINDIFF", VSD_DTYPE_INT_S32,
167 	    NVSS(vss_windiff), vss_windiff, 0);
168 	lasterr = err ? err : lasterr;
169 
170 	struct voistatspec vss_acklen[] = {
171 		STATS_VSS_MAX(),
172 		STATS_VSS_CRHIST32_LIN(0, 9, 1, VSD_HIST_UBOUND_INF)
173 	};
174 	err |= stats_tpl_add_voistats(V_tcp_perconn_stats_dflt_tpl,
175 	    VOI_TCP_ACKLEN, "TCP_ACKLEN", VSD_DTYPE_INT_U32,
176 	    NVSS(vss_acklen), vss_acklen, 0);
177 	lasterr = err ? err : lasterr;
178 
179 	return (lasterr);
180 }
181 
182 #ifdef _KERNEL
183 int
184 tcp_stats_sample_rollthedice(struct tcpcb *tp, void *seed_bytes,
185     size_t seed_len)
186 {
187 	struct rm_priotracker tracker;
188 	int tpl;
189 
190 	tpl = -1;
191 
192 	if (V_tcp_stats_nrates > 0) {
193 		rm_rlock(&tcp_stats_tpl_sampling_lock, &tracker);
194 		tpl = stats_tpl_sample_rollthedice(V_tcp_perconn_stats_sample_rates,
195 		    V_tcp_stats_nrates, seed_bytes, seed_len);
196 		rm_runlock(&tcp_stats_tpl_sampling_lock, &tracker);
197 
198 		if (tpl >= 0) {
199 			INP_WLOCK_ASSERT(tptoinpcb(tp));
200 			if (tp->t_stats != NULL)
201 				stats_blob_destroy(tp->t_stats);
202 			tp->t_stats = stats_blob_alloc(tpl, 0);
203 			if (tp->t_stats == NULL)
204 				tpl = -ENOMEM;
205 		}
206 	}
207 
208 	return (tpl);
209 }
210 
211 /*
212  * Callback function for stats_tpl_sample_rates() to interact with the TCP
213  * subsystem's stats template sample rates list.
214  */
215 int
216 tcp_stats_tpl_sr_cb(enum stats_tpl_sr_cb_action action,
217     struct stats_tpl_sample_rate **rates, int *nrates, void *ctx)
218 {
219 	struct stats_tpl_sample_rate *old_rates;
220 	int old_nrates;
221 
222 	if (ctx == NULL)
223 		return (ENOMEM);
224 
225 	switch (action) {
226 	case TPL_SR_RLOCKED_GET:
227 		/*
228 		 * Return with rlock held i.e. this call must be paired with a
229 		 * "action == TPL_SR_RUNLOCK" call.
230 		 */
231 		rm_assert(&tcp_stats_tpl_sampling_lock, RA_UNLOCKED);
232 		rm_rlock(&tcp_stats_tpl_sampling_lock,
233 		    (struct rm_priotracker *)ctx);
234 		/* FALLTHROUGH */
235 	case TPL_SR_UNLOCKED_GET:
236 		if (rates != NULL)
237 			*rates = V_tcp_perconn_stats_sample_rates;
238 		if (nrates != NULL)
239 			*nrates = V_tcp_stats_nrates;
240 		break;
241 	case TPL_SR_RUNLOCK:
242 		rm_assert(&tcp_stats_tpl_sampling_lock, RA_RLOCKED);
243 		rm_runlock(&tcp_stats_tpl_sampling_lock,
244 		    (struct rm_priotracker *)ctx);
245 		break;
246 	case TPL_SR_PUT:
247 		KASSERT(rates != NULL && nrates != NULL,
248 		    ("%s: PUT without new rates", __func__));
249 		rm_assert(&tcp_stats_tpl_sampling_lock, RA_UNLOCKED);
250 		if (rates == NULL || nrates == NULL)
251 			return (EINVAL);
252 		rm_wlock(&tcp_stats_tpl_sampling_lock);
253 		old_rates = V_tcp_perconn_stats_sample_rates;
254 		old_nrates = V_tcp_stats_nrates;
255 		V_tcp_perconn_stats_sample_rates = *rates;
256 		V_tcp_stats_nrates = *nrates;
257 		rm_wunlock(&tcp_stats_tpl_sampling_lock);
258 		*rates = old_rates;
259 		*nrates = old_nrates;
260 		break;
261 	default:
262 		return (EINVAL);
263 		break;
264 	}
265 
266 	return (0);
267 }
268 
269 RM_SYSINIT(tcp_stats_tpl_sampling_lock, &tcp_stats_tpl_sampling_lock,
270     "tcp_stats_tpl_sampling_lock");
271 #endif /* _KERNEL */
272