xref: /titanic_50/usr/src/cmd/svc/startd/transition.c (revision c77a61a72b5ecdc507d6cf104142edd371a16c84)
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 (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * transition.c - Graph State Machine
30  *
31  * The graph state machine is implemented here, with a typical approach
32  * of a function per state.  Separating the implementation allows more
33  * clarity into the actions taken on notification of state change, as well
34  * as a place for future expansion including hooks for configurable actions.
35  * All functions are called with dgraph_lock held.
36  *
37  * The start action for this state machine is not explicit.  The states
38  * (ONLINE and DEGRADED) which need to know when they're entering the state
39  * due to a daemon restart implement this understanding by checking for
40  * transition from uninitialized.  In the future, this would likely be better
41  * as an explicit start action instead of relying on an overloaded transition.
42  *
43  * All gt_enter functions use the same set of return codes.
44  *    0              success
45  *    ECONNABORTED   repository connection aborted
46  */
47 
48 #include "startd.h"
49 
50 static int
51 gt_running(restarter_instance_state_t state)
52 {
53 	if (state == RESTARTER_STATE_ONLINE ||
54 	    state == RESTARTER_STATE_DEGRADED)
55 		return (1);
56 
57 	return (0);
58 }
59 
60 static int
61 gt_enter_uninit(scf_handle_t *h, graph_vertex_t *v,
62     restarter_instance_state_t old_state, restarter_error_t rerr)
63 {
64 	int err;
65 	scf_instance_t *inst;
66 
67 	/* Initialize instance by refreshing it. */
68 
69 	err = libscf_fmri_get_instance(h, v->gv_name, &inst);
70 	switch (err) {
71 	case 0:
72 		break;
73 
74 	case ECONNABORTED:
75 		return (ECONNABORTED);
76 
77 	case ENOENT:
78 		return (0);
79 
80 	case EINVAL:
81 	case ENOTSUP:
82 	default:
83 		bad_error("libscf_fmri_get_instance", err);
84 	}
85 
86 	err = refresh_vertex(v, inst);
87 	if (err == 0)
88 		graph_enable_by_vertex(v, v->gv_flags & GV_ENABLED, 0);
89 
90 	scf_instance_destroy(inst);
91 
92 	/* If the service was running, propagate a stop event. */
93 	if (gt_running(old_state)) {
94 		log_framework(LOG_DEBUG, "Propagating stop of %s.\n",
95 		    v->gv_name);
96 
97 		graph_transition_propagate(v, PROPAGATE_STOP, rerr);
98 	}
99 
100 	graph_transition_sulogin(RESTARTER_STATE_UNINIT, old_state);
101 	return (0);
102 }
103 
104 /* ARGSUSED */
105 static int
106 gt_enter_maint(scf_handle_t *h, graph_vertex_t *v,
107     restarter_instance_state_t old_state, restarter_error_t rerr)
108 {
109 	/*
110 	 * If the service was running, propagate a stop event.  If the
111 	 * service was not running the maintenance transition may satisfy
112 	 * optional dependencies and should be propagated to determine
113 	 * whether new dependents are satisfiable.
114 	 */
115 	if (gt_running(old_state)) {
116 		log_framework(LOG_DEBUG, "Propagating maintenance (stop) of "
117 		    "%s.\n", v->gv_name);
118 
119 		graph_transition_propagate(v, PROPAGATE_STOP, rerr);
120 	} else {
121 		log_framework(LOG_DEBUG, "Propagating maintenance of %s.\n",
122 		    v->gv_name);
123 
124 		graph_transition_propagate(v, PROPAGATE_SAT, rerr);
125 	}
126 
127 	graph_transition_sulogin(RESTARTER_STATE_MAINT, old_state);
128 	return (0);
129 }
130 
131 /* ARGSUSED */
132 static int
133 gt_enter_offline(scf_handle_t *h, graph_vertex_t *v,
134     restarter_instance_state_t old_state, restarter_error_t rerr)
135 {
136 	/*
137 	 * If the instance should be enabled, see if we can start it.
138 	 * Otherwise send a disable command.
139 	 */
140 	if (v->gv_flags & GV_ENABLED) {
141 		graph_start_if_satisfied(v);
142 	} else {
143 		if (gt_running(old_state) && v->gv_post_disable_f)
144 			v->gv_post_disable_f();
145 		vertex_send_event(v, RESTARTER_EVENT_TYPE_DISABLE);
146 	}
147 
148 	if (gt_running(old_state)) {
149 		log_framework(LOG_DEBUG, "Propagating stop of %s.\n",
150 		    v->gv_name);
151 
152 		graph_transition_propagate(v, PROPAGATE_STOP, rerr);
153 	}
154 
155 	graph_transition_sulogin(RESTARTER_STATE_OFFLINE, old_state);
156 	return (0);
157 }
158 
159 /* ARGSUSED */
160 static int
161 gt_enter_disabled(scf_handle_t *h, graph_vertex_t *v,
162     restarter_instance_state_t old_state, restarter_error_t rerr)
163 {
164 	/*
165 	 * If the instance should be disabled, no problem.  Otherwise,
166 	 * send an enable command, which should result in the instance
167 	 * moving to OFFLINE.
168 	 */
169 	if (v->gv_flags & GV_ENABLED) {
170 		vertex_send_event(v, RESTARTER_EVENT_TYPE_ENABLE);
171 	} else if (gt_running(old_state) && v->gv_post_disable_f) {
172 		v->gv_post_disable_f();
173 	}
174 
175 	/*
176 	 * If the service was running, propagate this as a stop.  If the
177 	 * service was not running the disabled transition may satisfy
178 	 * optional dependencies and should be propagated to determine
179 	 * whether new dependents are satisfiable.
180 	 */
181 	if (gt_running(old_state)) {
182 		log_framework(LOG_DEBUG, "Propagating stop of %s.\n",
183 		    v->gv_name);
184 
185 		graph_transition_propagate(v, PROPAGATE_STOP, rerr);
186 
187 	} else {
188 		log_framework(LOG_DEBUG, "Propagating disable of %s.\n",
189 		    v->gv_name);
190 
191 		graph_transition_propagate(v, PROPAGATE_SAT, rerr);
192 	}
193 
194 	graph_transition_sulogin(RESTARTER_STATE_DISABLED, old_state);
195 	return (0);
196 }
197 
198 static int
199 gt_internal_online_or_degraded(scf_handle_t *h, graph_vertex_t *v,
200     restarter_instance_state_t old_state, restarter_error_t rerr)
201 {
202 	int r;
203 
204 	/*
205 	 * If the instance has just come up, update the start
206 	 * snapshot.
207 	 */
208 	if (gt_running(old_state) == 0) {
209 		/*
210 		 * Don't fire if we're just recovering state
211 		 * after a restart.
212 		 */
213 		if (old_state != RESTARTER_STATE_UNINIT &&
214 		    v->gv_post_online_f)
215 			v->gv_post_online_f();
216 
217 		r = libscf_snapshots_poststart(h, v->gv_name, B_TRUE);
218 		switch (r) {
219 		case 0:
220 		case ENOENT:
221 			/*
222 			 * If ENOENT, the instance must have been
223 			 * deleted.  Pretend we were successful since
224 			 * we should get a delete event later.
225 			 */
226 			break;
227 
228 		case ECONNABORTED:
229 			return (ECONNABORTED);
230 
231 		case EACCES:
232 		case ENOTSUP:
233 		default:
234 			bad_error("libscf_snapshots_poststart", r);
235 		}
236 	}
237 	if (!(v->gv_flags & GV_ENABLED))
238 		vertex_send_event(v, RESTARTER_EVENT_TYPE_DISABLE);
239 
240 	if (gt_running(old_state) == 0) {
241 		log_framework(LOG_DEBUG, "Propagating start of %s.\n",
242 		    v->gv_name);
243 
244 		graph_transition_propagate(v, PROPAGATE_START, rerr);
245 	} else if (rerr == RERR_REFRESH) {
246 		/* For refresh we'll get a message sans state change */
247 
248 		log_framework(LOG_DEBUG, "Propagating refresh of %s.\n",
249 		    v->gv_name);
250 
251 		graph_transition_propagate(v, PROPAGATE_STOP, rerr);
252 	}
253 
254 	return (0);
255 }
256 
257 static int
258 gt_enter_online(scf_handle_t *h, graph_vertex_t *v,
259     restarter_instance_state_t old_state, restarter_error_t rerr)
260 {
261 	int r;
262 
263 	r = gt_internal_online_or_degraded(h, v, old_state, rerr);
264 	if (r != 0)
265 		return (r);
266 
267 	graph_transition_sulogin(RESTARTER_STATE_ONLINE, old_state);
268 	return (0);
269 }
270 
271 static int
272 gt_enter_degraded(scf_handle_t *h, graph_vertex_t *v,
273     restarter_instance_state_t old_state, restarter_error_t rerr)
274 {
275 	int r;
276 
277 	r = gt_internal_online_or_degraded(h, v, old_state, rerr);
278 	if (r != 0)
279 		return (r);
280 
281 	graph_transition_sulogin(RESTARTER_STATE_DEGRADED, old_state);
282 	return (0);
283 }
284 
285 /*
286  * gt_transition() implements the state transition for the graph
287  * state machine.  It can return:
288  *    0              success
289  *    ECONNABORTED   repository connection aborted
290  *
291  * v->gv_state should be set to the state we're transitioning to before
292  * calling this function.
293  */
294 int
295 gt_transition(scf_handle_t *h, graph_vertex_t *v, restarter_error_t rerr,
296     restarter_instance_state_t old_state)
297 {
298 	int err;
299 	int lost_repository = 0;
300 
301 	/*
302 	 * If there's a common set of work to be done on exit from the
303 	 * old_state, include it as a separate set of functions here.  For
304 	 * now there's no such work, so there are no gt_exit functions.
305 	 */
306 
307 	err = vertex_subgraph_dependencies_shutdown(h, v, old_state);
308 	switch (err) {
309 	case 0:
310 		break;
311 
312 	case ECONNABORTED:
313 		lost_repository = 1;
314 		break;
315 
316 	default:
317 		bad_error("vertex_subgraph_dependencies_shutdown", err);
318 	}
319 
320 	/*
321 	 * Now call the appropriate gt_enter function for the new state.
322 	 */
323 	switch (v->gv_state) {
324 	case RESTARTER_STATE_UNINIT:
325 		err = gt_enter_uninit(h, v, old_state, rerr);
326 		break;
327 
328 	case RESTARTER_STATE_DISABLED:
329 		err = gt_enter_disabled(h, v, old_state, rerr);
330 		break;
331 
332 	case RESTARTER_STATE_OFFLINE:
333 		err = gt_enter_offline(h, v, old_state, rerr);
334 		break;
335 
336 	case RESTARTER_STATE_ONLINE:
337 		err = gt_enter_online(h, v, old_state, rerr);
338 		break;
339 
340 	case RESTARTER_STATE_DEGRADED:
341 		err = gt_enter_degraded(h, v, old_state, rerr);
342 		break;
343 
344 	case RESTARTER_STATE_MAINT:
345 		err = gt_enter_maint(h, v, old_state, rerr);
346 		break;
347 
348 	default:
349 		/* Shouldn't be in an invalid state. */
350 #ifndef NDEBUG
351 		uu_warn("%s:%d: Invalid state %d.\n", __FILE__, __LINE__,
352 		    v->gv_state);
353 #endif
354 		abort();
355 	}
356 
357 	switch (err) {
358 	case 0:
359 		break;
360 
361 	case ECONNABORTED:
362 		lost_repository = 1;
363 		break;
364 
365 	default:
366 #ifndef NDEBUG
367 		uu_warn("%s:%d: "
368 		    "gt_enter_%s() failed with unexpected error %d.\n",
369 		    __FILE__, __LINE__, instance_state_str[v->gv_state], err);
370 #endif
371 		abort();
372 	}
373 
374 	return (lost_repository ? ECONNABORTED : 0);
375 }
376