1d9638e54Smws /* 2d9638e54Smws * CDDL HEADER START 3d9638e54Smws * 4d9638e54Smws * The contents of this file are subject to the terms of the 59f8ca725Sstephh * Common Development and Distribution License (the "License"). 69f8ca725Sstephh * You may not use this file except in compliance with the License. 7d9638e54Smws * 8d9638e54Smws * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9d9638e54Smws * or http://www.opensolaris.org/os/licensing. 10d9638e54Smws * See the License for the specific language governing permissions 11d9638e54Smws * and limitations under the License. 12d9638e54Smws * 13d9638e54Smws * When distributing Covered Code, include this CDDL HEADER in each 14d9638e54Smws * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15d9638e54Smws * If applicable, add the following below this CDDL HEADER, with the 16d9638e54Smws * fields enclosed by brackets "[]" replaced with your own identifying 17d9638e54Smws * information: Portions Copyright [yyyy] [name of copyright owner] 18d9638e54Smws * 19d9638e54Smws * CDDL HEADER END 20d9638e54Smws */ 21d9638e54Smws 22d9638e54Smws /* 23*cbf75e67SStephen Hanson * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 24d9638e54Smws * Use is subject to license terms. 25d9638e54Smws */ 26d9638e54Smws 27d9638e54Smws #include <sys/sysevent/eventdefs.h> 28d9638e54Smws #include <sys/sysevent.h> 29d9638e54Smws #include <sys/sysevent_impl.h> 30d9638e54Smws #include <sys/fm/protocol.h> 31d9638e54Smws #include <sys/sysmacros.h> 32d9638e54Smws #include <sys/dumphdr.h> 33d9638e54Smws #include <sys/dumpadm.h> 347aec1d6eScindi #include <sys/fm/util.h> 35d9638e54Smws 36d9638e54Smws #include <libsysevent.h> 37d9638e54Smws #include <libnvpair.h> 38d9638e54Smws #include <alloca.h> 39d9638e54Smws #include <limits.h> 40d9638e54Smws #include <strings.h> 41d9638e54Smws #include <unistd.h> 42d9638e54Smws #include <fcntl.h> 43d9638e54Smws #include <errno.h> 44d9638e54Smws 45d9638e54Smws #undef MUTEX_HELD 46d9638e54Smws #undef RW_READ_HELD 47d9638e54Smws #undef RW_WRITE_HELD 48d9638e54Smws 49d9638e54Smws #include <fmd_api.h> 50d9638e54Smws #include <fmd_log.h> 51d9638e54Smws #include <fmd_subr.h> 52d9638e54Smws #include <fmd_dispq.h> 539af3851aSeschrock #include <fmd_dr.h> 54d9638e54Smws #include <fmd_module.h> 559af3851aSeschrock #include <fmd_protocol.h> 56d9638e54Smws #include <fmd_scheme.h> 57d9638e54Smws #include <fmd_error.h> 58d9638e54Smws 59d9638e54Smws #include <fmd.h> 60d9638e54Smws 61d9638e54Smws static char *sysev_channel; /* event channel to which we are subscribed */ 62d9638e54Smws static char *sysev_class; /* event class to which we are subscribed */ 63d9638e54Smws static char *sysev_device; /* device path to use for replaying events */ 64d9638e54Smws static char *sysev_sid; /* event channel subscriber identifier */ 65d9638e54Smws static void *sysev_evc; /* event channel cookie from evc_bind */ 66d9638e54Smws 67d9638e54Smws static fmd_xprt_t *sysev_xprt; 6897c04605Scy152378 static int sysev_xprt_refcnt; 69d9638e54Smws static fmd_hdl_t *sysev_hdl; 70d9638e54Smws 71d9638e54Smws static struct sysev_stats { 72d9638e54Smws fmd_stat_t dump_replay; 73d9638e54Smws fmd_stat_t dump_lost; 74d9638e54Smws fmd_stat_t bad_class; 75d9638e54Smws fmd_stat_t bad_attr; 76d9638e54Smws fmd_stat_t eagain; 77d9638e54Smws } sysev_stats = { 78d9638e54Smws { "dump_replay", FMD_TYPE_UINT64, "events replayed from dump device" }, 79d9638e54Smws { "dump_lost", FMD_TYPE_UINT64, "events lost from dump device" }, 80d9638e54Smws { "bad_class", FMD_TYPE_UINT64, "events dropped due to invalid class" }, 81d9638e54Smws { "bad_attr", FMD_TYPE_UINT64, "events dropped due to invalid nvlist" }, 82d9638e54Smws { "eagain", FMD_TYPE_UINT64, "events retried due to low memory" }, 83d9638e54Smws }; 84d9638e54Smws 8597c04605Scy152378 static pthread_cond_t sysev_cv = PTHREAD_COND_INITIALIZER; 8697c04605Scy152378 static pthread_mutex_t sysev_mutex = PTHREAD_MUTEX_INITIALIZER; 879f8ca725Sstephh static int sysev_replay_wait = 1; 8897c04605Scy152378 static int sysev_exiting; 8997c04605Scy152378 909af3851aSeschrock /* 919af3851aSeschrock * Entry point for legacy sysevents. This function is responsible for two 929af3851aSeschrock * things: passing off interesting events to the DR handler, and converting 939af3851aSeschrock * sysevents into resource events that modules can then subscribe to. 949af3851aSeschrock */ 959af3851aSeschrock static void 969af3851aSeschrock sysev_legacy(sysevent_t *sep) 979af3851aSeschrock { 989af3851aSeschrock const char *class = sysevent_get_class_name(sep); 999af3851aSeschrock const char *subclass = sysevent_get_subclass_name(sep); 1009af3851aSeschrock char *fullclass; 1019af3851aSeschrock size_t len; 1029af3851aSeschrock nvlist_t *attr, *nvl; 1039af3851aSeschrock fmd_event_t *e; 1049af3851aSeschrock hrtime_t hrt; 1059af3851aSeschrock 1069af3851aSeschrock /* notify the DR subsystem of the event */ 1079af3851aSeschrock fmd_dr_event(sep); 1089af3851aSeschrock 1099af3851aSeschrock /* get the matching sysevent name */ 1109af3851aSeschrock len = snprintf(NULL, 0, "%s%s.%s", SYSEVENT_RSRC_CLASS, 1119af3851aSeschrock class, subclass); 1129af3851aSeschrock fullclass = alloca(len + 1); 1139af3851aSeschrock (void) snprintf(fullclass, len + 1, "%s%s.%s", 1149af3851aSeschrock SYSEVENT_RSRC_CLASS, class, subclass); 1159af3851aSeschrock 1169af3851aSeschrock /* construct the event payload */ 1179af3851aSeschrock (void) nvlist_xalloc(&nvl, NV_UNIQUE_NAME, &fmd.d_nva); 1189af3851aSeschrock (void) nvlist_add_string(nvl, FM_CLASS, fullclass); 1199af3851aSeschrock (void) nvlist_add_uint8(nvl, FM_VERSION, FM_RSRC_VERSION); 120825ba0f2Srobj if (sysevent_get_attr_list(sep, &attr) == 0) { 1219af3851aSeschrock (void) nvlist_merge(nvl, attr, 0); 122825ba0f2Srobj nvlist_free(attr); 123825ba0f2Srobj } 1249af3851aSeschrock 1259af3851aSeschrock /* 1269af3851aSeschrock * Dispatch the event. Ideally, we'd like to use the same transport 1279af3851aSeschrock * interface as sysev_recv(), but because the legacy sysevent mechanism 1289af3851aSeschrock * puts in a thread outside fmd's control, using the module APIs is 1299af3851aSeschrock * impossible. 1309af3851aSeschrock */ 1319af3851aSeschrock sysevent_get_time(sep, &hrt); 1329af3851aSeschrock (void) nvlist_lookup_string(nvl, FM_CLASS, &fullclass); 1339af3851aSeschrock e = fmd_event_create(FMD_EVT_PROTOCOL, hrt, nvl, fullclass); 1349af3851aSeschrock fmd_dispq_dispatch(fmd.d_disp, e, fullclass); 1359af3851aSeschrock } 1369f8ca725Sstephh 137d9638e54Smws /* 138d9638e54Smws * Receive an event from the SysEvent channel and post it to our transport. 139d9638e54Smws * Under extreme low-memory situations where we cannot event unpack the event, 140d9638e54Smws * we can request that SysEvent redeliver the event later by returning EAGAIN. 141d9638e54Smws * If we do this too many times, the kernel will drop the event. Rather than 142d9638e54Smws * keeping state per-event, we simply attempt a garbage-collect, hoping that 143d9638e54Smws * enough free memory will be available by the time the event is redelivered. 144d9638e54Smws */ 145d9638e54Smws static int 146d9638e54Smws sysev_recv(sysevent_t *sep, void *arg) 147d9638e54Smws { 148d9638e54Smws uint64_t seq = sysevent_get_seq(sep); 149d9638e54Smws fmd_xprt_t *xp = arg; 150d9638e54Smws nvlist_t *nvl; 151d9638e54Smws hrtime_t hrt; 15297c04605Scy152378 int rc = 0; 153d9638e54Smws 15497c04605Scy152378 (void) pthread_mutex_lock(&sysev_mutex); 15597c04605Scy152378 if (sysev_exiting == 1) { 15697c04605Scy152378 while (sysev_xprt_refcnt > 0) 15797c04605Scy152378 (void) pthread_cond_wait(&sysev_cv, &sysev_mutex); 15897c04605Scy152378 (void) pthread_mutex_unlock(&sysev_mutex); 15997c04605Scy152378 return (EAGAIN); 16097c04605Scy152378 } 16197c04605Scy152378 sysev_xprt_refcnt++; 1629f8ca725Sstephh while (sysev_replay_wait) 16397c04605Scy152378 (void) pthread_cond_wait(&sysev_cv, &sysev_mutex); 16497c04605Scy152378 (void) pthread_mutex_unlock(&sysev_mutex); 1659f8ca725Sstephh 166d9638e54Smws if (strcmp(sysevent_get_class_name(sep), EC_FM) != 0) { 167d9638e54Smws fmd_hdl_error(sysev_hdl, "discarding event 0x%llx: unexpected" 168d9638e54Smws " transport class %s\n", seq, sysevent_get_class_name(sep)); 169d9638e54Smws sysev_stats.bad_class.fmds_value.ui64++; 17097c04605Scy152378 } else if (sysevent_get_attr_list(sep, &nvl) != 0) { 171d9638e54Smws if (errno == EAGAIN || errno == ENOMEM) { 172d9638e54Smws fmd_modhash_tryapply(fmd.d_mod_hash, fmd_module_trygc); 173d9638e54Smws fmd_scheme_hash_trygc(fmd.d_schemes); 174d9638e54Smws sysev_stats.eagain.fmds_value.ui64++; 17597c04605Scy152378 rc = EAGAIN; 17697c04605Scy152378 } else { 17797c04605Scy152378 fmd_hdl_error(sysev_hdl, "discarding event 0x%llx: " 17897c04605Scy152378 "missing or invalid payload", seq); 179d9638e54Smws sysev_stats.bad_attr.fmds_value.ui64++; 180d9638e54Smws } 18197c04605Scy152378 } else { 182d9638e54Smws sysevent_get_time(sep, &hrt); 183d9638e54Smws fmd_xprt_post(sysev_hdl, xp, nvl, hrt); 18497c04605Scy152378 } 18597c04605Scy152378 18697c04605Scy152378 (void) pthread_mutex_lock(&sysev_mutex); 18797c04605Scy152378 if (--sysev_xprt_refcnt == 0 && sysev_exiting == 1) 18897c04605Scy152378 (void) pthread_cond_broadcast(&sysev_cv); 18997c04605Scy152378 (void) pthread_mutex_unlock(&sysev_mutex); 19097c04605Scy152378 19197c04605Scy152378 return (rc); 192d9638e54Smws } 193d9638e54Smws 194d9638e54Smws /* 195d9638e54Smws * Checksum algorithm used by the dump transport for verifying the content of 196d9638e54Smws * error reports saved on the dump device (copy of the kernel's checksum32()). 197d9638e54Smws */ 198d9638e54Smws static uint32_t 199d9638e54Smws sysev_checksum(void *cp_arg, size_t length) 200d9638e54Smws { 201d9638e54Smws uchar_t *cp, *ep; 202d9638e54Smws uint32_t sum = 0; 203d9638e54Smws 204d9638e54Smws for (cp = cp_arg, ep = cp + length; cp < ep; cp++) 205d9638e54Smws sum = ((sum >> 1) | (sum << 31)) + *cp; 206d9638e54Smws 207d9638e54Smws return (sum); 208d9638e54Smws } 209d9638e54Smws 210d9638e54Smws /* 211d9638e54Smws * Replay saved events from the dump transport. This function is installed as 212d9638e54Smws * the timer callback and is called only once during the module's lifetime. 213d9638e54Smws */ 214d9638e54Smws /*ARGSUSED*/ 215d9638e54Smws static void 216d9638e54Smws sysev_replay(fmd_hdl_t *hdl, id_t id, void *arg) 217d9638e54Smws { 218d9638e54Smws char *dumpdev; 219d9638e54Smws off64_t off, off0; 220d9638e54Smws int fd, err; 221d9638e54Smws 222d9638e54Smws /* 223d9638e54Smws * Determine the appropriate dump device to use for replaying pending 224d9638e54Smws * error reports. If the device property is NULL (default), we 225d9638e54Smws * open and query /dev/dump to determine the current dump device. 226d9638e54Smws */ 227d9638e54Smws if ((dumpdev = sysev_device) == NULL) { 228d9638e54Smws if ((fd = open("/dev/dump", O_RDONLY)) == -1) { 229d9638e54Smws fmd_hdl_error(hdl, "failed to open /dev/dump " 230d9638e54Smws "to locate dump device for event replay"); 2319f8ca725Sstephh goto done; 232d9638e54Smws } 233d9638e54Smws 234d9638e54Smws dumpdev = alloca(PATH_MAX); 235d9638e54Smws err = ioctl(fd, DIOCGETDEV, dumpdev); 236d9638e54Smws (void) close(fd); 237d9638e54Smws 238d9638e54Smws if (err == -1) { 239d9638e54Smws if (errno != ENODEV) { 240d9638e54Smws fmd_hdl_error(hdl, "failed to obtain " 241d9638e54Smws "path to dump device for event replay"); 242d9638e54Smws } 2439f8ca725Sstephh goto done; 244d9638e54Smws } 245d9638e54Smws } 246d9638e54Smws 247d9638e54Smws if (strcmp(dumpdev, "/dev/null") == 0) 2489f8ca725Sstephh goto done; /* return silently and skip replay for /dev/null */ 249d9638e54Smws 250d9638e54Smws /* 251d9638e54Smws * Open the appropriate device and then determine the offset of the 252d9638e54Smws * start of the ereport dump region located at the end of the device. 253d9638e54Smws */ 254d9638e54Smws if ((fd = open64(dumpdev, O_RDWR | O_DSYNC)) == -1) { 255d9638e54Smws fmd_hdl_error(hdl, "failed to open dump transport %s " 256d9638e54Smws "(pending events will not be replayed)", dumpdev); 2579f8ca725Sstephh goto done; 258d9638e54Smws } 259d9638e54Smws 260d9638e54Smws off = DUMP_OFFSET + DUMP_LOGSIZE + DUMP_ERPTSIZE; 261d9638e54Smws off = off0 = lseek64(fd, -off, SEEK_END) & -DUMP_OFFSET; 262d9638e54Smws 263d9638e54Smws if (off == (off64_t)-1LL) { 264d9638e54Smws fmd_hdl_error(hdl, "failed to seek dump transport %s " 265d9638e54Smws "(pending events will not be replayed)", dumpdev); 266d9638e54Smws (void) close(fd); 2679f8ca725Sstephh goto done; 268d9638e54Smws } 269d9638e54Smws 270d9638e54Smws /* 271d9638e54Smws * The ereport dump region is a sequence of erpt_dump_t headers each of 272d9638e54Smws * which is followed by packed nvlist data. We iterate over them in 273d9638e54Smws * order, unpacking and dispatching each one to our dispatch queue. 274d9638e54Smws */ 275d9638e54Smws for (;;) { 276d9638e54Smws char nvbuf[ERPT_DATA_SZ]; 277d9638e54Smws uint32_t chksum; 278d9638e54Smws erpt_dump_t ed; 279d9638e54Smws nvlist_t *nvl; 280d9638e54Smws 281d9638e54Smws fmd_timeval_t ftv, tod; 282d9638e54Smws hrtime_t hrt; 283d9638e54Smws uint64_t ena; 284d9638e54Smws 285d9638e54Smws if (pread64(fd, &ed, sizeof (ed), off) != sizeof (ed)) { 286d9638e54Smws fmd_hdl_error(hdl, "failed to read from dump " 287d9638e54Smws "transport %s (pending events lost)", dumpdev); 288d9638e54Smws break; 289d9638e54Smws } 290d9638e54Smws 291d9638e54Smws if (ed.ed_magic == 0 && ed.ed_size == 0) 292d9638e54Smws break; /* end of list: all zero */ 293d9638e54Smws 294d9638e54Smws if (ed.ed_magic == 0) { 295d9638e54Smws off += sizeof (ed) + ed.ed_size; 296d9638e54Smws continue; /* continue searching */ 297d9638e54Smws } 298d9638e54Smws 299d9638e54Smws if (ed.ed_magic != ERPT_MAGIC) { 300d9638e54Smws /* 301d9638e54Smws * Stop reading silently if the first record has the 302d9638e54Smws * wrong magic number; this likely indicates that we 303d9638e54Smws * rebooted from non-FMA bits or paged over the dump. 304d9638e54Smws */ 305d9638e54Smws if (off == off0) 306d9638e54Smws break; 307d9638e54Smws 308d9638e54Smws fmd_hdl_error(hdl, "invalid dump transport " 309d9638e54Smws "record at %llx (magic number %x, expected %x)\n", 310d9638e54Smws (u_longlong_t)off, ed.ed_magic, ERPT_MAGIC); 311d9638e54Smws break; 312d9638e54Smws } 313d9638e54Smws 314d9638e54Smws if (ed.ed_size > ERPT_DATA_SZ) { 315d9638e54Smws fmd_hdl_error(hdl, "invalid dump transport " 316d9638e54Smws "record at %llx size (%u exceeds limit)\n", 317d9638e54Smws (u_longlong_t)off, ed.ed_size); 318d9638e54Smws break; 319d9638e54Smws } 320d9638e54Smws 321d9638e54Smws if (pread64(fd, nvbuf, ed.ed_size, 322d9638e54Smws off + sizeof (ed)) != ed.ed_size) { 323d9638e54Smws fmd_hdl_error(hdl, "failed to read dump " 324d9638e54Smws "transport event (offset %llx)", (u_longlong_t)off); 325d9638e54Smws 326d9638e54Smws sysev_stats.dump_lost.fmds_value.ui64++; 327d9638e54Smws goto next; 328d9638e54Smws } 329d9638e54Smws 330d9638e54Smws if ((chksum = sysev_checksum(nvbuf, 331d9638e54Smws ed.ed_size)) != ed.ed_chksum) { 332d9638e54Smws fmd_hdl_error(hdl, "dump transport event at " 333d9638e54Smws "offset %llx is corrupt (checksum %x != %x)\n", 334d9638e54Smws (u_longlong_t)off, chksum, ed.ed_chksum); 335d9638e54Smws 336d9638e54Smws sysev_stats.dump_lost.fmds_value.ui64++; 337d9638e54Smws goto next; 338d9638e54Smws } 339d9638e54Smws 340d9638e54Smws if ((err = nvlist_xunpack(nvbuf, 341d9638e54Smws ed.ed_size, &nvl, &fmd.d_nva)) != 0) { 342d9638e54Smws fmd_hdl_error(hdl, "failed to unpack dump " 343d9638e54Smws "transport event at offset %llx: %s\n", 344d9638e54Smws (u_longlong_t)off, fmd_strerror(err)); 345d9638e54Smws 346d9638e54Smws sysev_stats.dump_lost.fmds_value.ui64++; 347d9638e54Smws goto next; 348d9638e54Smws } 349d9638e54Smws 350d9638e54Smws /* 351d9638e54Smws * If ed_hrt_nsec is set it contains the gethrtime() value from 352d9638e54Smws * when the event was originally enqueued for the transport. 353d9638e54Smws * If it is zero, we use the weaker bound ed_hrt_base instead. 354d9638e54Smws */ 355d9638e54Smws if (ed.ed_hrt_nsec != 0) 356d9638e54Smws hrt = ed.ed_hrt_nsec; 357d9638e54Smws else 358d9638e54Smws hrt = ed.ed_hrt_base; 359d9638e54Smws 360d9638e54Smws /* 361d9638e54Smws * If this is an FMA protocol event of class "ereport.*" that 362d9638e54Smws * contains valid ENA, we can improve the precision of 'hrt'. 363d9638e54Smws */ 364d9638e54Smws if (nvlist_lookup_uint64(nvl, FM_EREPORT_ENA, &ena) == 0) 365d9638e54Smws hrt = fmd_time_ena2hrt(hrt, ena); 366d9638e54Smws 367d9638e54Smws /* 368d9638e54Smws * Now convert 'hrt' to an adjustable TOD based on the values 369d9638e54Smws * in ed_tod_base which correspond to one another and are 370d9638e54Smws * sampled before reboot using the old gethrtime() clock. 371d9638e54Smws * fmd_event_recreate() will use this TOD value to re-assign 372d9638e54Smws * the event an updated gethrtime() value based on the current 373d9638e54Smws * value of the non-adjustable gethrtime() clock. Phew. 374d9638e54Smws */ 375d9638e54Smws tod.ftv_sec = ed.ed_tod_base.sec; 376d9638e54Smws tod.ftv_nsec = ed.ed_tod_base.nsec; 377d9638e54Smws fmd_time_hrt2tod(ed.ed_hrt_base, &tod, hrt, &ftv); 378d9638e54Smws 379d9638e54Smws (void) nvlist_remove_all(nvl, FMD_EVN_TOD); 380d9638e54Smws (void) nvlist_add_uint64_array(nvl, 381d9638e54Smws FMD_EVN_TOD, (uint64_t *)&ftv, 2); 382d9638e54Smws 383d9638e54Smws fmd_xprt_post(hdl, sysev_xprt, nvl, 0); 384d9638e54Smws sysev_stats.dump_replay.fmds_value.ui64++; 385d9638e54Smws 386d9638e54Smws next: 387d9638e54Smws /* 388d9638e54Smws * Reset the magic number for the event record to zero so that 389d9638e54Smws * we do not replay the same event multiple times. 390d9638e54Smws */ 391d9638e54Smws ed.ed_magic = 0; 392d9638e54Smws 393d9638e54Smws if (pwrite64(fd, &ed, sizeof (ed), off) != sizeof (ed)) { 394d9638e54Smws fmd_hdl_error(hdl, "failed to mark dump " 395d9638e54Smws "transport event (offset %llx)", (u_longlong_t)off); 396d9638e54Smws } 397d9638e54Smws 398d9638e54Smws off += sizeof (ed) + ed.ed_size; 399d9638e54Smws } 400d9638e54Smws 401d9638e54Smws (void) close(fd); 4029f8ca725Sstephh done: 40397c04605Scy152378 (void) pthread_mutex_lock(&sysev_mutex); 4049f8ca725Sstephh sysev_replay_wait = 0; 40597c04605Scy152378 (void) pthread_cond_broadcast(&sysev_cv); 40697c04605Scy152378 (void) pthread_mutex_unlock(&sysev_mutex); 407d9638e54Smws } 408d9638e54Smws 409d9638e54Smws static const fmd_prop_t sysev_props[] = { 410d9638e54Smws { "class", FMD_TYPE_STRING, EC_ALL }, /* event class */ 411d9638e54Smws { "device", FMD_TYPE_STRING, NULL }, /* replay device */ 412d9638e54Smws { "channel", FMD_TYPE_STRING, FM_ERROR_CHAN }, /* channel name */ 413d9638e54Smws { "sid", FMD_TYPE_STRING, "fmd" }, /* subscriber id */ 414d9638e54Smws { NULL, 0, NULL } 415d9638e54Smws }; 416d9638e54Smws 417d9638e54Smws static const fmd_hdl_ops_t sysev_ops = { 418d9638e54Smws NULL, /* fmdo_recv */ 419d9638e54Smws sysev_replay, /* fmdo_timeout */ 420d9638e54Smws NULL, /* fmdo_close */ 421d9638e54Smws NULL, /* fmdo_stats */ 422d9638e54Smws NULL, /* fmdo_gc */ 423d9638e54Smws NULL, /* fmdo_send */ 424d9638e54Smws }; 425d9638e54Smws 426d9638e54Smws static const fmd_hdl_info_t sysev_info = { 427d9638e54Smws "SysEvent Transport Agent", "1.0", &sysev_ops, sysev_props 428d9638e54Smws }; 429d9638e54Smws 430d9638e54Smws /* 431d9638e54Smws * Bind to the sysevent channel we use for listening for error events and then 4329af3851aSeschrock * subscribe to appropriate events received over this channel. Setup the 4339af3851aSeschrock * legacy sysevent handler for creating sysevent resources and forwarding DR 4349af3851aSeschrock * events. 435d9638e54Smws */ 436d9638e54Smws void 437d9638e54Smws sysev_init(fmd_hdl_t *hdl) 438d9638e54Smws { 439d9638e54Smws uint_t flags; 4409af3851aSeschrock const char *subclasses[] = { EC_SUB_ALL }; 441d9638e54Smws 442d9638e54Smws if (fmd_hdl_register(hdl, FMD_API_VERSION, &sysev_info) != 0) 443d9638e54Smws return; /* invalid property settings */ 444d9638e54Smws 445d9638e54Smws (void) fmd_stat_create(hdl, FMD_STAT_NOALLOC, sizeof (sysev_stats) / 446d9638e54Smws sizeof (fmd_stat_t), (fmd_stat_t *)&sysev_stats); 447d9638e54Smws 448d9638e54Smws sysev_channel = fmd_prop_get_string(hdl, "channel"); 449d9638e54Smws sysev_class = fmd_prop_get_string(hdl, "class"); 450d9638e54Smws sysev_device = fmd_prop_get_string(hdl, "device"); 451d9638e54Smws sysev_sid = fmd_prop_get_string(hdl, "sid"); 452d9638e54Smws 453d9638e54Smws if (sysev_channel == NULL) 454d9638e54Smws fmd_hdl_abort(hdl, "channel property must be defined\n"); 455d9638e54Smws 456d9638e54Smws if (sysev_sid == NULL) 457d9638e54Smws fmd_hdl_abort(hdl, "sid property must be defined\n"); 458d9638e54Smws 459d9638e54Smws if ((errno = sysevent_evc_bind(sysev_channel, &sysev_evc, 460d9638e54Smws EVCH_CREAT | EVCH_HOLD_PEND)) != 0) { 461d9638e54Smws fmd_hdl_abort(hdl, "failed to bind to event transport " 462d9638e54Smws "channel %s", sysev_channel); 463d9638e54Smws } 464d9638e54Smws 465*cbf75e67SStephen Hanson sysev_xprt = fmd_xprt_open(hdl, FMD_XPRT_RDONLY | 466*cbf75e67SStephen Hanson FMD_XPRT_CACHE_AS_LOCAL, NULL, NULL); 467d9638e54Smws sysev_hdl = hdl; 468d9638e54Smws 469d9638e54Smws /* 470d9638e54Smws * If we're subscribing to the default channel, keep our subscription 471d9638e54Smws * active even if we die unexpectedly so we continue queuing events. 472d9638e54Smws * If we're not (e.g. running under fmsim), do not specify SUB_KEEP so 473d9638e54Smws * that our event channel will be destroyed if we die unpleasantly. 474d9638e54Smws */ 475d9638e54Smws if (strcmp(sysev_channel, FM_ERROR_CHAN) == 0) 476d9638e54Smws flags = EVCH_SUB_KEEP | EVCH_SUB_DUMP; 477d9638e54Smws else 478d9638e54Smws flags = EVCH_SUB_DUMP; 479d9638e54Smws 480d9638e54Smws errno = sysevent_evc_subscribe(sysev_evc, 481d9638e54Smws sysev_sid, sysev_class, sysev_recv, sysev_xprt, flags); 482d9638e54Smws 483d9638e54Smws if (errno != 0) { 484d9638e54Smws if (errno == EEXIST) { 485d9638e54Smws fmd_hdl_abort(hdl, "another fault management daemon is " 486d9638e54Smws "active on transport channel %s\n", sysev_channel); 487d9638e54Smws } else { 488d9638e54Smws fmd_hdl_abort(hdl, "failed to subscribe to %s on " 489d9638e54Smws "transport channel %s", sysev_class, sysev_channel); 490d9638e54Smws } 491d9638e54Smws } 492d9638e54Smws 493d9638e54Smws /* 494d9638e54Smws * Once the transport is open, install a single timer to fire at once 495d9638e54Smws * in the context of the module's thread to run sysev_replay(). This 496d9638e54Smws * thread will block in its first fmd_xprt_post() until fmd is ready. 497d9638e54Smws */ 498d9638e54Smws fmd_hdl_debug(hdl, "transport '%s' open\n", sysev_channel); 499d9638e54Smws (void) fmd_timer_install(hdl, NULL, NULL, 0); 5009af3851aSeschrock 5019af3851aSeschrock /* 5029af3851aSeschrock * Open the legacy sysevent handle and subscribe to all events. These 5039af3851aSeschrock * are automatically converted to "resource.sysevent.*" events so that 5049af3851aSeschrock * modules can manage these events without additional infrastructure. 5059af3851aSeschrock */ 5069af3851aSeschrock if (geteuid() != 0) 5079af3851aSeschrock return; 5089af3851aSeschrock 5099af3851aSeschrock if ((fmd.d_sysev_hdl = 5109af3851aSeschrock sysevent_bind_handle(sysev_legacy)) == NULL) 5119af3851aSeschrock fmd_hdl_abort(hdl, "failed to bind to legacy sysevent channel"); 5129af3851aSeschrock 5139af3851aSeschrock if (sysevent_subscribe_event(fmd.d_sysev_hdl, EC_ALL, 5149af3851aSeschrock subclasses, 1) != 0) 5159af3851aSeschrock fmd_hdl_abort(hdl, "failed to subscribe to legacy sysevents"); 516d9638e54Smws } 517d9638e54Smws 518d9638e54Smws /* 519d9638e54Smws * Close the channel by unsubscribing and unbinding. We only do this when a 520d9638e54Smws * a non-default channel has been selected. If we're using FM_ERROR_CHAN, 521d9638e54Smws * the system default, we do *not* want to unsubscribe because the kernel will 522d9638e54Smws * remove the subscriber queue and any events published in our absence will 523d9638e54Smws * therefore be lost. This scenario may occur when, for example, fmd is sent 524d9638e54Smws * a SIGTERM by init(1M) during reboot but an error is detected and makes it 525d9638e54Smws * into the sysevent channel queue before init(1M) manages to call uadmin(2). 526d9638e54Smws */ 527d9638e54Smws void 528d9638e54Smws sysev_fini(fmd_hdl_t *hdl) 529d9638e54Smws { 530d9638e54Smws if (strcmp(sysev_channel, FM_ERROR_CHAN) != 0) { 531d9638e54Smws sysevent_evc_unsubscribe(sysev_evc, sysev_sid); 532d9638e54Smws sysevent_evc_unbind(sysev_evc); 533d9638e54Smws } 534d9638e54Smws 5359af3851aSeschrock if (fmd.d_sysev_hdl != NULL) 5369af3851aSeschrock sysevent_unbind_handle(fmd.d_sysev_hdl); 5379af3851aSeschrock 53897c04605Scy152378 if (sysev_xprt != NULL) { 53997c04605Scy152378 /* 54097c04605Scy152378 * Wait callback returns before destroy the transport. 54197c04605Scy152378 */ 54297c04605Scy152378 (void) pthread_mutex_lock(&sysev_mutex); 54397c04605Scy152378 sysev_exiting = 1; 54497c04605Scy152378 while (sysev_xprt_refcnt > 0) 54597c04605Scy152378 (void) pthread_cond_wait(&sysev_cv, &sysev_mutex); 54697c04605Scy152378 (void) pthread_mutex_unlock(&sysev_mutex); 547d9638e54Smws fmd_xprt_close(hdl, sysev_xprt); 54897c04605Scy152378 } 549d9638e54Smws 550d9638e54Smws fmd_prop_free_string(hdl, sysev_class); 551d9638e54Smws fmd_prop_free_string(hdl, sysev_channel); 552d9638e54Smws fmd_prop_free_string(hdl, sysev_device); 553d9638e54Smws fmd_prop_free_string(hdl, sysev_sid); 554d9638e54Smws } 555