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 /*
23 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
25 */
26
27 #include <sys/fm/protocol.h>
28 #include <limits.h>
29
30 #include <fmd_alloc.h>
31 #include <fmd_subr.h>
32 #include <fmd_event.h>
33 #include <fmd_string.h>
34 #include <fmd_module.h>
35 #include <fmd_case.h>
36 #include <fmd_log.h>
37 #include <fmd_time.h>
38 #include <fmd_topo.h>
39 #include <fmd_ctl.h>
40
41 #include <fmd.h>
42
43 static void
fmd_event_nvwrap(fmd_event_impl_t * ep)44 fmd_event_nvwrap(fmd_event_impl_t *ep)
45 {
46 (void) nvlist_remove_all(ep->ev_nvl, FMD_EVN_TTL);
47 (void) nvlist_remove_all(ep->ev_nvl, FMD_EVN_TOD);
48
49 (void) nvlist_add_uint8(ep->ev_nvl,
50 FMD_EVN_TTL, ep->ev_ttl);
51 (void) nvlist_add_uint64_array(ep->ev_nvl,
52 FMD_EVN_TOD, (uint64_t *)&ep->ev_time, 2);
53 }
54
55 static void
fmd_event_nvunwrap(fmd_event_impl_t * ep,const fmd_timeval_t * tp)56 fmd_event_nvunwrap(fmd_event_impl_t *ep, const fmd_timeval_t *tp)
57 {
58 uint64_t *tod;
59 uint_t n;
60
61 if (nvlist_lookup_uint8(ep->ev_nvl, FMD_EVN_TTL, &ep->ev_ttl) != 0) {
62 ep->ev_flags |= FMD_EVF_LOCAL;
63 ep->ev_ttl = (uint8_t)fmd.d_xprt_ttl;
64 }
65
66 if (tp != NULL)
67 ep->ev_time = *tp;
68 else if (nvlist_lookup_uint64_array(ep->ev_nvl,
69 FMD_EVN_TOD, &tod, &n) == 0 && n >= 2)
70 ep->ev_time = *(const fmd_timeval_t *)tod;
71 else
72 fmd_time_sync(&ep->ev_time, &ep->ev_hrt, 1);
73 }
74
75 fmd_event_t *
fmd_event_recreate(uint_t type,const fmd_timeval_t * tp,nvlist_t * nvl,void * data,fmd_log_t * lp,off64_t off,size_t len)76 fmd_event_recreate(uint_t type, const fmd_timeval_t *tp,
77 nvlist_t *nvl, void *data, fmd_log_t *lp, off64_t off, size_t len)
78 {
79 fmd_event_impl_t *ep = fmd_alloc(sizeof (fmd_event_impl_t), FMD_SLEEP);
80
81 fmd_timeval_t tod;
82 hrtime_t hr0;
83
84 (void) pthread_mutex_init(&ep->ev_lock, NULL);
85 ep->ev_refs = 0;
86 ASSERT(type < FMD_EVT_NTYPES);
87 ep->ev_type = (uint8_t)type;
88 ep->ev_state = FMD_EVS_RECEIVED;
89 ep->ev_flags = FMD_EVF_REPLAY;
90 ep->ev_nvl = nvl;
91 ep->ev_data = data;
92 ep->ev_log = lp;
93 ep->ev_off = off;
94 ep->ev_len = len;
95
96 fmd_event_nvunwrap(ep, tp);
97
98 /*
99 * If we're not restoring from a log, the event is marked volatile. If
100 * we are restoring from a log, then hold the log pointer and increment
101 * the pending count. If we're using a log but no offset and data len
102 * are specified, it's a checkpoint event: don't replay or set pending.
103 */
104 if (lp == NULL)
105 ep->ev_flags |= FMD_EVF_VOLATILE;
106 else if (off != 0 && len != 0)
107 fmd_log_hold_pending(lp);
108 else {
109 ep->ev_flags &= ~FMD_EVF_REPLAY;
110 fmd_log_hold(lp);
111 }
112
113 /*
114 * Sample a (TOD, hrtime) pair from the current system clocks and then
115 * compute ev_hrt by taking the delta between this TOD and ev_time.
116 */
117 fmd_time_sync(&tod, &hr0, 1);
118 fmd_time_tod2hrt(hr0, &tod, &ep->ev_time, &ep->ev_hrt);
119
120 fmd_event_nvwrap(ep);
121 return ((fmd_event_t *)ep);
122 }
123
124 fmd_event_t *
fmd_event_create(uint_t type,hrtime_t hrt,nvlist_t * nvl,void * data)125 fmd_event_create(uint_t type, hrtime_t hrt, nvlist_t *nvl, void *data)
126 {
127 fmd_event_impl_t *ep = fmd_alloc(sizeof (fmd_event_impl_t), FMD_SLEEP);
128
129 fmd_timeval_t tod;
130 hrtime_t hr0;
131 const char *p;
132 uint64_t ena;
133
134 (void) pthread_mutex_init(&ep->ev_lock, NULL);
135 ep->ev_refs = 0;
136 ASSERT(type < FMD_EVT_NTYPES);
137 ep->ev_type = (uint8_t)type;
138 ep->ev_state = FMD_EVS_RECEIVED;
139 ep->ev_flags = FMD_EVF_VOLATILE | FMD_EVF_REPLAY | FMD_EVF_LOCAL;
140 ep->ev_ttl = (uint8_t)fmd.d_xprt_ttl;
141 ep->ev_nvl = nvl;
142 ep->ev_data = data;
143 ep->ev_log = NULL;
144 ep->ev_off = 0;
145 ep->ev_len = 0;
146
147 /*
148 * Sample TOD and then set ev_time to the earlier TOD corresponding to
149 * the input hrtime value. This needs to be improved later: hrestime
150 * should be sampled by the transport and passed as an input parameter.
151 */
152 fmd_time_sync(&tod, &hr0, 1);
153
154 if (hrt == FMD_HRT_NOW)
155 hrt = hr0; /* use hrtime sampled by fmd_time_sync() */
156
157 /*
158 * If this is an FMA protocol event of class "ereport.*" that contains
159 * valid ENA, we can compute a more precise bound on the event time.
160 */
161 if (type == FMD_EVT_PROTOCOL && (p = strchr(data, '.')) != NULL &&
162 strncmp(data, FM_EREPORT_CLASS, (size_t)(p - (char *)data)) == 0 &&
163 nvlist_lookup_uint64(nvl, FM_EREPORT_ENA, &ena) == 0 &&
164 fmd.d_clockops == &fmd_timeops_native)
165 hrt = fmd_time_ena2hrt(hrt, ena);
166
167 fmd_time_hrt2tod(hr0, &tod, hrt, &ep->ev_time);
168 ep->ev_hrt = hrt;
169
170 fmd_event_nvwrap(ep);
171 return ((fmd_event_t *)ep);
172 }
173
174 void
fmd_event_destroy(fmd_event_t * e)175 fmd_event_destroy(fmd_event_t *e)
176 {
177 fmd_event_impl_t *ep = (fmd_event_impl_t *)e;
178
179 ASSERT(MUTEX_HELD(&ep->ev_lock));
180 ASSERT(ep->ev_refs == 0);
181
182 /*
183 * If the current state is RECEIVED (i.e. no module has accepted the
184 * event) and the event was logged, then change the state to DISCARDED.
185 */
186 if (ep->ev_state == FMD_EVS_RECEIVED)
187 ep->ev_state = FMD_EVS_DISCARDED;
188
189 /*
190 * If the current state is DISCARDED, ACCEPTED, or DIAGNOSED and the
191 * event has not yet been commited, then attempt to commit it now.
192 */
193 if (ep->ev_state != FMD_EVS_RECEIVED && (ep->ev_flags & (
194 FMD_EVF_VOLATILE | FMD_EVF_REPLAY)) == FMD_EVF_REPLAY)
195 fmd_log_commit(ep->ev_log, e);
196
197 if (ep->ev_log != NULL) {
198 if (ep->ev_flags & FMD_EVF_REPLAY)
199 fmd_log_decommit(ep->ev_log, e);
200 fmd_log_rele(ep->ev_log);
201 }
202
203 /*
204 * Perform any event type-specific cleanup activities, and then free
205 * the name-value pair list and underlying event data structure.
206 */
207 switch (ep->ev_type) {
208 case FMD_EVT_TIMEOUT:
209 fmd_free(ep->ev_data, sizeof (fmd_modtimer_t));
210 break;
211 case FMD_EVT_CLOSE:
212 case FMD_EVT_PUBLISH:
213 fmd_case_rele(ep->ev_data);
214 break;
215 case FMD_EVT_CTL:
216 fmd_ctl_fini(ep->ev_data);
217 break;
218 case FMD_EVT_TOPO:
219 fmd_topo_rele(ep->ev_data);
220 break;
221 }
222
223 nvlist_free(ep->ev_nvl);
224
225 fmd_free(ep, sizeof (fmd_event_impl_t));
226 }
227
228 void
fmd_event_hold(fmd_event_t * e)229 fmd_event_hold(fmd_event_t *e)
230 {
231 fmd_event_impl_t *ep = (fmd_event_impl_t *)e;
232
233 (void) pthread_mutex_lock(&ep->ev_lock);
234 ep->ev_refs++;
235 ASSERT(ep->ev_refs != 0);
236 (void) pthread_mutex_unlock(&ep->ev_lock);
237
238 if (ep->ev_type == FMD_EVT_CTL)
239 fmd_ctl_hold(ep->ev_data);
240 }
241
242 void
fmd_event_rele(fmd_event_t * e)243 fmd_event_rele(fmd_event_t *e)
244 {
245 fmd_event_impl_t *ep = (fmd_event_impl_t *)e;
246
247 if (ep->ev_type == FMD_EVT_CTL)
248 fmd_ctl_rele(ep->ev_data);
249
250 (void) pthread_mutex_lock(&ep->ev_lock);
251 ASSERT(ep->ev_refs != 0);
252
253 if (--ep->ev_refs == 0)
254 fmd_event_destroy(e);
255 else
256 (void) pthread_mutex_unlock(&ep->ev_lock);
257 }
258
259 /*
260 * Transition event from its current state to the specified state. The states
261 * for events are defined in fmd_event.h and work according to the diagram:
262 *
263 * ------------- ------------- State Description
264 * ( RECEIVED =1 )-->( ACCEPTED =2 ) ---------- ---------------------------
265 * -----+-------\ ------+------ DISCARDED No active references in fmd
266 * | \ | RECEIVED Active refs in fmd, no case
267 * -----v------- \ ------v------ ACCEPTED Active refs, case assigned
268 * ( DISCARDED=0 ) v( DIAGNOSED=3 ) DIAGNOSED Active refs, case solved
269 * ------------- -------------
270 *
271 * Since events are reference counted on behalf of multiple subscribers, any
272 * attempt to transition an event to an "earlier" or "equal" state (as defined
273 * by the numeric state values shown in the diagram) is silently ignored.
274 * An event begins life in the RECEIVED state, so the RECEIVED -> DISCARDED
275 * transition is handled by fmd_event_destroy() when no references remain.
276 */
277 void
fmd_event_transition(fmd_event_t * e,uint_t state)278 fmd_event_transition(fmd_event_t *e, uint_t state)
279 {
280 fmd_event_impl_t *ep = (fmd_event_impl_t *)e;
281
282 (void) pthread_mutex_lock(&ep->ev_lock);
283
284 TRACE((FMD_DBG_EVT, "event %p transition %u -> %u",
285 (void *)ep, ep->ev_state, state));
286
287 if (state <= ep->ev_state) {
288 (void) pthread_mutex_unlock(&ep->ev_lock);
289 return; /* no state change necessary */
290 }
291
292 if (ep->ev_state < FMD_EVS_RECEIVED || ep->ev_state > FMD_EVS_DIAGNOSED)
293 fmd_panic("illegal transition %u -> %u\n", ep->ev_state, state);
294
295 ep->ev_state = state;
296 (void) pthread_mutex_unlock(&ep->ev_lock);
297 }
298
299 /*
300 * If the specified event is DISCARDED, ACCEPTED, OR DIAGNOSED and it has been
301 * written to a log but is still marked for replay, attempt to commit it to the
302 * log so that it will not be replayed. If fmd_log_commit() is successful, it
303 * will clear the FMD_EVF_REPLAY flag on the event for us.
304 */
305 void
fmd_event_commit(fmd_event_t * e)306 fmd_event_commit(fmd_event_t *e)
307 {
308 fmd_event_impl_t *ep = (fmd_event_impl_t *)e;
309
310 (void) pthread_mutex_lock(&ep->ev_lock);
311
312 if (ep->ev_state != FMD_EVS_RECEIVED && (ep->ev_flags & (
313 FMD_EVF_VOLATILE | FMD_EVF_REPLAY)) == FMD_EVF_REPLAY)
314 fmd_log_commit(ep->ev_log, e);
315
316 (void) pthread_mutex_unlock(&ep->ev_lock);
317 }
318
319 /*
320 * Compute the delta between events in nanoseconds. To account for very old
321 * events which are replayed, we must handle the case where ev_hrt is negative.
322 * We convert the hrtime_t's to unsigned 64-bit integers and then handle the
323 * case where 'old' is greater than 'new' (i.e. high-res time has wrapped).
324 */
325 hrtime_t
fmd_event_delta(fmd_event_t * e1,fmd_event_t * e2)326 fmd_event_delta(fmd_event_t *e1, fmd_event_t *e2)
327 {
328 uint64_t old = ((fmd_event_impl_t *)e1)->ev_hrt;
329 uint64_t new = ((fmd_event_impl_t *)e2)->ev_hrt;
330
331 return (new >= old ? new - old : (UINT64_MAX - old) + new + 1);
332 }
333
334 hrtime_t
fmd_event_hrtime(fmd_event_t * ep)335 fmd_event_hrtime(fmd_event_t *ep)
336 {
337 return (((fmd_event_impl_t *)ep)->ev_hrt);
338 }
339
340 int
fmd_event_match(fmd_event_t * e,uint_t type,const void * data)341 fmd_event_match(fmd_event_t *e, uint_t type, const void *data)
342 {
343 fmd_event_impl_t *ep = (fmd_event_impl_t *)e;
344
345 if (ep->ev_type != type)
346 return (0);
347
348 if (type == FMD_EVT_PROTOCOL)
349 return (fmd_strmatch(ep->ev_data, data));
350 else if (type == FMD_EVT_TIMEOUT)
351 return ((id_t)data == ((fmd_modtimer_t *)ep->ev_data)->mt_id);
352 else
353 return (ep->ev_data == data);
354 }
355
356 int
fmd_event_equal(fmd_event_t * e1,fmd_event_t * e2)357 fmd_event_equal(fmd_event_t *e1, fmd_event_t *e2)
358 {
359 fmd_event_impl_t *ep1 = (fmd_event_impl_t *)e1;
360 fmd_event_impl_t *ep2 = (fmd_event_impl_t *)e2;
361
362 return (ep1->ev_log != NULL &&
363 ep1->ev_log == ep2->ev_log && ep1->ev_off == ep2->ev_off);
364 }
365