1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 
23 /*
24  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
25  * Use is subject to license terms.
26  */
27 
28 #pragma ident	"%Z%%M%	%I%	%E% SMI"
29 
30 #include <dt_impl.h>
31 #include <stddef.h>
32 #include <errno.h>
33 #include <assert.h>
34 #include <time.h>
35 
36 static const struct {
37 	int dtslt_option;
38 	size_t dtslt_offs;
39 } _dtrace_sleeptab[] = {
40 	{ DTRACEOPT_STATUSRATE, offsetof(dtrace_hdl_t, dt_laststatus) },
41 	{ DTRACEOPT_AGGRATE, offsetof(dtrace_hdl_t, dt_lastagg) },
42 	{ DTRACEOPT_SWITCHRATE, offsetof(dtrace_hdl_t, dt_lastswitch) },
43 	{ DTRACEOPT_MAX, 0 }
44 };
45 
46 void
47 dtrace_sleep(dtrace_hdl_t *dtp)
48 {
49 	dt_proc_hash_t *dph = dtp->dt_procs;
50 	dtrace_optval_t policy = dtp->dt_options[DTRACEOPT_BUFPOLICY];
51 	dt_proc_notify_t *dprn;
52 
53 	hrtime_t earliest = INT64_MAX;
54 	struct timespec tv;
55 	hrtime_t now;
56 	int i;
57 
58 	for (i = 0; _dtrace_sleeptab[i].dtslt_option < DTRACEOPT_MAX; i++) {
59 		uintptr_t a = (uintptr_t)dtp + _dtrace_sleeptab[i].dtslt_offs;
60 		int opt = _dtrace_sleeptab[i].dtslt_option;
61 		dtrace_optval_t interval = dtp->dt_options[opt];
62 
63 		/*
64 		 * If the buffering policy is set to anything other than
65 		 * "switch", we ignore the aggrate and switchrate -- they're
66 		 * meaningless.
67 		 */
68 		if (policy != DTRACEOPT_BUFPOLICY_SWITCH &&
69 		    _dtrace_sleeptab[i].dtslt_option != DTRACEOPT_STATUSRATE)
70 			continue;
71 
72 		if (*((hrtime_t *)a) + interval < earliest)
73 			earliest = *((hrtime_t *)a) + interval;
74 	}
75 
76 	(void) pthread_mutex_lock(&dph->dph_lock);
77 
78 	now = gethrtime();
79 
80 	if (earliest < now) {
81 		(void) pthread_mutex_unlock(&dph->dph_lock);
82 		return; /* sleep duration has already past */
83 	}
84 
85 #if defined(sun)
86 	tv.tv_sec = (earliest - now) / NANOSEC;
87 	tv.tv_nsec = (earliest - now) % NANOSEC;
88 
89 	/*
90 	 * Wait for either 'tv' nanoseconds to pass or to receive notification
91 	 * that a process is in an interesting state.  Regardless of why we
92 	 * awaken, iterate over any pending notifications and process them.
93 	 */
94 	(void) pthread_cond_reltimedwait_np(&dph->dph_cv, &dph->dph_lock, &tv);
95 #else
96 	earliest -= now;
97 	clock_gettime(CLOCK_REALTIME,&tv);
98 	tv.tv_sec += earliest / NANOSEC;
99 	tv.tv_nsec += earliest % NANOSEC;
100 	while (tv.tv_nsec > NANOSEC) {
101 		tv.tv_sec += 1;
102 		tv.tv_nsec -= NANOSEC;
103 	}
104 
105 	/*
106 	 * Wait for either 'tv' nanoseconds to pass or to receive notification
107 	 * that a process is in an interesting state.  Regardless of why we
108 	 * awaken, iterate over any pending notifications and process them.
109 	 */
110 	(void) pthread_cond_timedwait(&dph->dph_cv, &dph->dph_lock, &tv);
111 #endif
112 
113 	while ((dprn = dph->dph_notify) != NULL) {
114 		if (dtp->dt_prochdlr != NULL) {
115 			char *err = dprn->dprn_errmsg;
116 			if (*err == '\0')
117 				err = NULL;
118 
119 			dtp->dt_prochdlr(dprn->dprn_dpr->dpr_proc, err,
120 			    dtp->dt_procarg);
121 		}
122 
123 		dph->dph_notify = dprn->dprn_next;
124 		dt_free(dtp, dprn);
125 	}
126 
127 	(void) pthread_mutex_unlock(&dph->dph_lock);
128 }
129 
130 int
131 dtrace_status(dtrace_hdl_t *dtp)
132 {
133 	int gen = dtp->dt_statusgen;
134 	dtrace_optval_t interval = dtp->dt_options[DTRACEOPT_STATUSRATE];
135 	hrtime_t now = gethrtime();
136 
137 	if (!dtp->dt_active)
138 		return (DTRACE_STATUS_NONE);
139 
140 	if (dtp->dt_stopped)
141 		return (DTRACE_STATUS_STOPPED);
142 
143 	if (dtp->dt_laststatus != 0) {
144 		if (now - dtp->dt_laststatus < interval)
145 			return (DTRACE_STATUS_NONE);
146 
147 		dtp->dt_laststatus += interval;
148 	} else {
149 		dtp->dt_laststatus = now;
150 	}
151 
152 	if (dt_ioctl(dtp, DTRACEIOC_STATUS, &dtp->dt_status[gen]) == -1)
153 		return (dt_set_errno(dtp, errno));
154 
155 	dtp->dt_statusgen ^= 1;
156 
157 	if (dt_handle_status(dtp, &dtp->dt_status[dtp->dt_statusgen],
158 	    &dtp->dt_status[gen]) == -1)
159 		return (-1);
160 
161 	if (dtp->dt_status[gen].dtst_exiting) {
162 		if (!dtp->dt_stopped)
163 			(void) dtrace_stop(dtp);
164 
165 		return (DTRACE_STATUS_EXITED);
166 	}
167 
168 	if (dtp->dt_status[gen].dtst_filled == 0)
169 		return (DTRACE_STATUS_OKAY);
170 
171 	if (dtp->dt_options[DTRACEOPT_BUFPOLICY] != DTRACEOPT_BUFPOLICY_FILL)
172 		return (DTRACE_STATUS_OKAY);
173 
174 	if (!dtp->dt_stopped) {
175 		if (dtrace_stop(dtp) == -1)
176 			return (-1);
177 	}
178 
179 	return (DTRACE_STATUS_FILLED);
180 }
181 
182 int
183 dtrace_go(dtrace_hdl_t *dtp)
184 {
185 	dtrace_enable_io_t args;
186 	void *dof;
187 	int err;
188 
189 	if (dtp->dt_active)
190 		return (dt_set_errno(dtp, EINVAL));
191 
192 	/*
193 	 * If a dtrace:::ERROR program and callback are registered, enable the
194 	 * program before we start tracing.  If this fails for a vector open
195 	 * with ENOTTY, we permit dtrace_go() to succeed so that vector clients
196 	 * such as mdb's dtrace module can execute the rest of dtrace_go() even
197 	 * though they do not provide support for the DTRACEIOC_ENABLE ioctl.
198 	 */
199 	if (dtp->dt_errprog != NULL &&
200 	    dtrace_program_exec(dtp, dtp->dt_errprog, NULL) == -1 && (
201 	    dtp->dt_errno != ENOTTY || dtp->dt_vector == NULL))
202 		return (-1); /* dt_errno has been set for us */
203 
204 	if ((dof = dtrace_getopt_dof(dtp)) == NULL)
205 		return (-1); /* dt_errno has been set for us */
206 
207 	args.dof = dof;
208 	args.n_matched = 0;
209 	err = dt_ioctl(dtp, DTRACEIOC_ENABLE, &args);
210 	dtrace_dof_destroy(dtp, dof);
211 
212 	if (err == -1 && (errno != ENOTTY || dtp->dt_vector == NULL))
213 		return (dt_set_errno(dtp, errno));
214 
215 	if (dt_ioctl(dtp, DTRACEIOC_GO, &dtp->dt_beganon) == -1) {
216 		if (errno == EACCES)
217 			return (dt_set_errno(dtp, EDT_DESTRUCTIVE));
218 
219 		if (errno == EALREADY)
220 			return (dt_set_errno(dtp, EDT_ISANON));
221 
222 		if (errno == ENOENT)
223 			return (dt_set_errno(dtp, EDT_NOANON));
224 
225 		if (errno == E2BIG)
226 			return (dt_set_errno(dtp, EDT_ENDTOOBIG));
227 
228 		if (errno == ENOSPC)
229 			return (dt_set_errno(dtp, EDT_BUFTOOSMALL));
230 
231 		return (dt_set_errno(dtp, errno));
232 	}
233 
234 	dtp->dt_active = 1;
235 
236 	if (dt_options_load(dtp) == -1)
237 		return (dt_set_errno(dtp, errno));
238 
239 	return (dt_aggregate_go(dtp));
240 }
241 
242 int
243 dtrace_stop(dtrace_hdl_t *dtp)
244 {
245 	int gen = dtp->dt_statusgen;
246 
247 	if (dtp->dt_stopped)
248 		return (0);
249 
250 	if (dt_ioctl(dtp, DTRACEIOC_STOP, &dtp->dt_endedon) == -1)
251 		return (dt_set_errno(dtp, errno));
252 
253 	dtp->dt_stopped = 1;
254 
255 	/*
256 	 * Now that we're stopped, we're going to get status one final time.
257 	 */
258 	if (dt_ioctl(dtp, DTRACEIOC_STATUS, &dtp->dt_status[gen]) == -1)
259 		return (dt_set_errno(dtp, errno));
260 
261 	if (dt_handle_status(dtp, &dtp->dt_status[gen ^ 1],
262 	    &dtp->dt_status[gen]) == -1)
263 		return (-1);
264 
265 	return (0);
266 }
267 
268 
269 dtrace_workstatus_t
270 dtrace_work(dtrace_hdl_t *dtp, FILE *fp,
271     dtrace_consume_probe_f *pfunc, dtrace_consume_rec_f *rfunc, void *arg)
272 {
273 	int status = dtrace_status(dtp);
274 	dtrace_optval_t policy = dtp->dt_options[DTRACEOPT_BUFPOLICY];
275 	dtrace_workstatus_t rval;
276 
277 	switch (status) {
278 	case DTRACE_STATUS_EXITED:
279 	case DTRACE_STATUS_FILLED:
280 	case DTRACE_STATUS_STOPPED:
281 		/*
282 		 * Tracing is stopped.  We now want to force dtrace_consume()
283 		 * and dtrace_aggregate_snap() to proceed, regardless of
284 		 * switchrate and aggrate.  We do this by clearing the times.
285 		 */
286 		dtp->dt_lastswitch = 0;
287 		dtp->dt_lastagg = 0;
288 		rval = DTRACE_WORKSTATUS_DONE;
289 		break;
290 
291 	case DTRACE_STATUS_NONE:
292 	case DTRACE_STATUS_OKAY:
293 		rval = DTRACE_WORKSTATUS_OKAY;
294 		break;
295 
296 	case -1:
297 		return (DTRACE_WORKSTATUS_ERROR);
298 	}
299 
300 	if ((status == DTRACE_STATUS_NONE || status == DTRACE_STATUS_OKAY) &&
301 	    policy != DTRACEOPT_BUFPOLICY_SWITCH) {
302 		/*
303 		 * There either isn't any status or things are fine -- and
304 		 * this is a "ring" or "fill" buffer.  We don't want to consume
305 		 * any of the trace data or snapshot the aggregations; we just
306 		 * return.
307 		 */
308 		assert(rval == DTRACE_WORKSTATUS_OKAY);
309 		return (rval);
310 	}
311 
312 	if (dtrace_aggregate_snap(dtp) == -1)
313 		return (DTRACE_WORKSTATUS_ERROR);
314 
315 	if (dtrace_consume(dtp, fp, pfunc, rfunc, arg) == -1)
316 		return (DTRACE_WORKSTATUS_ERROR);
317 
318 	return (rval);
319 }
320