xref: /freebsd/sys/kern/kern_tslog.c (revision 1323ec57)
1 /*-
2  * Copyright (c) 2017 Colin Percival
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 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/kernel.h>
32 #include <sys/linker.h>
33 #include <sys/malloc.h>
34 #include <sys/proc.h>
35 #include <sys/sbuf.h>
36 #include <sys/sysctl.h>
37 #include <sys/systm.h>
38 #include <sys/tslog.h>
39 
40 #include <machine/atomic.h>
41 #include <machine/cpu.h>
42 
43 #ifndef TSLOGSIZE
44 #define TSLOGSIZE 262144
45 #endif
46 
47 static volatile long nrecs = 0;
48 static struct timestamp {
49 	void * td;
50 	int type;
51 	const char * f;
52 	const char * s;
53 	uint64_t tsc;
54 } timestamps[TSLOGSIZE];
55 
56 void
57 tslog(void * td, int type, const char * f, const char * s)
58 {
59 	uint64_t tsc = get_cyclecount();
60 	long pos;
61 
62 	/* Grab a slot. */
63 	pos = atomic_fetchadd_long(&nrecs, 1);
64 
65 	/* Store record. */
66 	if (pos < nitems(timestamps)) {
67 		timestamps[pos].td = td;
68 		timestamps[pos].type = type;
69 		timestamps[pos].f = f;
70 		timestamps[pos].s = s;
71 		timestamps[pos].tsc = tsc;
72 	}
73 }
74 
75 static int
76 sysctl_debug_tslog(SYSCTL_HANDLER_ARGS)
77 {
78 	int error;
79 	struct sbuf *sb;
80 	size_t i, limit;
81 	caddr_t loader_tslog;
82 	void * loader_tslog_buf;
83 	size_t loader_tslog_len;
84 
85 	/*
86 	 * This code can race against the code in tslog() which stores
87 	 * records: Theoretically we could end up reading a record after
88 	 * its slots have been reserved but before it has been written.
89 	 * Since this code takes orders of magnitude longer to run than
90 	 * tslog() takes to write a record, it is highly unlikely that
91 	 * anyone will ever experience this race.
92 	 */
93 	sb = sbuf_new_for_sysctl(NULL, NULL, 1024, req);
94 
95 	/* Get data from the boot loader, if it provided any. */
96 	loader_tslog = preload_search_by_type("TSLOG data");
97 	if (loader_tslog != NULL) {
98 		loader_tslog_buf = preload_fetch_addr(loader_tslog);
99 		loader_tslog_len = preload_fetch_size(loader_tslog);
100 		sbuf_bcat(sb, loader_tslog_buf, loader_tslog_len);
101 	}
102 
103 	/* Add data logged within the kernel. */
104 	limit = MIN(nrecs, nitems(timestamps));
105 	for (i = 0; i < limit; i++) {
106 		sbuf_printf(sb, "%p", timestamps[i].td);
107 		sbuf_printf(sb, " %llu",
108 		    (unsigned long long)timestamps[i].tsc);
109 		switch (timestamps[i].type) {
110 		case TS_ENTER:
111 			sbuf_printf(sb, " ENTER");
112 			break;
113 		case TS_EXIT:
114 			sbuf_printf(sb, " EXIT");
115 			break;
116 		case TS_THREAD:
117 			sbuf_printf(sb, " THREAD");
118 			break;
119 		case TS_EVENT:
120 			sbuf_printf(sb, " EVENT");
121 			break;
122 		}
123 		sbuf_printf(sb, " %s", timestamps[i].f ? timestamps[i].f : "(null)");
124 		if (timestamps[i].s)
125 			sbuf_printf(sb, " %s\n", timestamps[i].s);
126 		else
127 			sbuf_printf(sb, "\n");
128 	}
129 	error = sbuf_finish(sb);
130 	sbuf_delete(sb);
131 	return (error);
132 }
133 
134 SYSCTL_PROC(_debug, OID_AUTO, tslog,
135     CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_SKIP,
136     0, 0, sysctl_debug_tslog, "", "Dump recorded event timestamps");
137 
138 MALLOC_DEFINE(M_TSLOGUSER, "tsloguser", "Strings used by userland tslog");
139 static struct procdata {
140 	pid_t ppid;
141 	uint64_t tsc_forked;
142 	uint64_t tsc_exited;
143 	char * execname;
144 	char * namei;
145 	int reused;
146 } procs[PID_MAX + 1];
147 
148 void
149 tslog_user(pid_t pid, pid_t ppid, const char * execname, const char * namei)
150 {
151 	uint64_t tsc = get_cyclecount();
152 
153 	/* If we wrapped, do nothing. */
154 	if (procs[pid].reused)
155 		return;
156 
157 	/* If we have a ppid, we're recording a fork. */
158 	if (ppid != (pid_t)(-1)) {
159 		/* If we have a ppid already, we wrapped. */
160 		if (procs[pid].ppid) {
161 			procs[pid].reused = 1;
162 			return;
163 		}
164 
165 		/* Fill in some fields. */
166 		procs[pid].ppid = ppid;
167 		procs[pid].tsc_forked = tsc;
168 		return;
169 	}
170 
171 	/* If we have an execname, record it. */
172 	if (execname != NULL) {
173 		if (procs[pid].execname != NULL)
174 			free(procs[pid].execname, M_TSLOGUSER);
175 		procs[pid].execname = strdup(execname, M_TSLOGUSER);
176 		return;
177 	}
178 
179 	/* Record the first namei for the process. */
180 	if (namei != NULL) {
181 		if (procs[pid].namei == NULL)
182 			procs[pid].namei = strdup(namei, M_TSLOGUSER);
183 		return;
184 	}
185 
186 	/* Otherwise we're recording an exit. */
187 	procs[pid].tsc_exited = tsc;
188 }
189 
190 static int
191 sysctl_debug_tslog_user(SYSCTL_HANDLER_ARGS)
192 {
193 	int error;
194 	struct sbuf *sb;
195 	pid_t pid;
196 
197 	sb = sbuf_new_for_sysctl(NULL, NULL, 1024, req);
198 
199 	/* Export the data we logged. */
200 	for (pid = 0; pid <= PID_MAX; pid++) {
201 		sbuf_printf(sb, "%zu", (size_t)pid);
202 		sbuf_printf(sb, " %zu", (size_t)procs[pid].ppid);
203 		sbuf_printf(sb, " %llu",
204 		    (unsigned long long)procs[pid].tsc_forked);
205 		sbuf_printf(sb, " %llu",
206 		    (unsigned long long)procs[pid].tsc_exited);
207 		sbuf_printf(sb, " \"%s\"", procs[pid].execname ?
208 		    procs[pid].execname : "");
209 		sbuf_printf(sb, " \"%s\"", procs[pid].namei ?
210 		    procs[pid].namei : "");
211 		sbuf_printf(sb, "\n");
212 	}
213 	error = sbuf_finish(sb);
214 	sbuf_delete(sb);
215 	return (error);
216 }
217 
218 SYSCTL_PROC(_debug, OID_AUTO, tslog_user,
219     CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_SKIP,
220     0, 0, sysctl_debug_tslog_user,
221     "", "Dump recorded userland event timestamps");
222