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.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
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]
22 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
26 * Fault Management Architecture (FMA) Resource and Protocol Support
28 * The routines contained herein provide services to support kernel subsystems
29 * in publishing fault management telemetry (see PSARC 2002/412 and 2003/089).
31 * Name-Value Pair Lists
33 * The embodiment of an FMA protocol element (event, fmri or authority) is a
34 * name-value pair list (nvlist_t). FMA-specific nvlist constructor and
35 * destructor functions, fm_nvlist_create() and fm_nvlist_destroy(), are used
36 * to create an nvpair list using custom allocators. Callers may choose to
37 * allocate either from the kernel memory allocator, or from a preallocated
38 * buffer, useful in constrained contexts like high-level interrupt routines.
40 * Protocol Event and FMRI Construction
42 * Convenience routines are provided to construct nvlist events according to
43 * the FMA Event Protocol and Naming Schema specification for ereports and
44 * FMRIs for the dev, cpu, hc, mem, legacy hc and de schemes.
48 * Routines to generate ENA formats 0, 1 and 2 are available as well as
49 * routines to increment formats 1 and 2. Individual fields within the
50 * ENA are extractable via fm_ena_time_get(), fm_ena_id_get(),
51 * fm_ena_format_get() and fm_ena_gen_get().
54 #include <sys/types.h>
57 #include <sys/nvpair.h>
58 #include <sys/cmn_err.h>
59 #include <sys/sysmacros.h>
60 #include <sys/sunddi.h>
61 #include <sys/systeminfo.h>
62 #include <sys/fm/util.h>
63 #include <sys/fm/protocol.h>
64 #include <sys/kstat.h>
65 #include <sys/zfs_context.h>
67 #include <sys/atomic.h>
68 #include <sys/condvar.h>
69 #include <sys/zfs_ioctl.h>
71 static int zfs_zevent_len_max = 512;
73 static int zevent_len_cur = 0;
74 static int zevent_waiters = 0;
75 static int zevent_flags = 0;
77 /* Num events rate limited since the last time zfs_zevent_next() was called */
78 static uint64_t ratelimit_dropped = 0;
81 * The EID (Event IDentifier) is used to uniquely tag a zevent when it is
82 * posted. The posted EIDs are monotonically increasing but not persistent.
83 * They will be reset to the initial value (1) each time the kernel module is
86 static uint64_t zevent_eid = 0;
88 static kmutex_t zevent_lock;
89 static list_t zevent_list;
90 static kcondvar_t zevent_cv;
95 * Common fault management kstats to record event generation failures
99 kstat_named_t erpt_dropped; /* num erpts dropped on post */
100 kstat_named_t erpt_set_failed; /* num erpt set failures */
101 kstat_named_t fmri_set_failed; /* num fmri set failures */
102 kstat_named_t payload_set_failed; /* num payload set failures */
103 kstat_named_t erpt_duplicates; /* num duplicate erpts */
106 static struct erpt_kstat erpt_kstat_data = {
107 { "erpt-dropped", KSTAT_DATA_UINT64 },
108 { "erpt-set-failed", KSTAT_DATA_UINT64 },
109 { "fmri-set-failed", KSTAT_DATA_UINT64 },
110 { "payload-set-failed", KSTAT_DATA_UINT64 },
111 { "erpt-duplicates", KSTAT_DATA_UINT64 }
119 zfs_zevent_alloc(void)
123 ev = kmem_zalloc(sizeof (zevent_t), KM_SLEEP);
125 list_create(&ev->ev_ze_list, sizeof (zfs_zevent_t),
126 offsetof(zfs_zevent_t, ze_node));
127 list_link_init(&ev->ev_node);
133 zfs_zevent_free(zevent_t *ev)
135 /* Run provided cleanup callback */
136 ev->ev_cb(ev->ev_nvl, ev->ev_detector);
138 list_destroy(&ev->ev_ze_list);
139 kmem_free(ev, sizeof (zevent_t));
143 zfs_zevent_drain(zevent_t *ev)
147 ASSERT(MUTEX_HELD(&zevent_lock));
148 list_remove(&zevent_list, ev);
150 /* Remove references to this event in all private file data */
151 while ((ze = list_head(&ev->ev_ze_list)) != NULL) {
152 list_remove(&ev->ev_ze_list, ze);
153 ze->ze_zevent = NULL;
161 zfs_zevent_drain_all(int *count)
165 mutex_enter(&zevent_lock);
166 while ((ev = list_head(&zevent_list)) != NULL)
167 zfs_zevent_drain(ev);
169 *count = zevent_len_cur;
171 mutex_exit(&zevent_lock);
175 * New zevents are inserted at the head. If the maximum queue
176 * length is exceeded a zevent will be drained from the tail.
177 * As part of this any user space processes which currently have
178 * a reference to this zevent_t in their private data will have
179 * this reference set to NULL.
182 zfs_zevent_insert(zevent_t *ev)
184 ASSERT(MUTEX_HELD(&zevent_lock));
185 list_insert_head(&zevent_list, ev);
187 if (zevent_len_cur >= zfs_zevent_len_max)
188 zfs_zevent_drain(list_tail(&zevent_list));
194 * Post a zevent. The cb will be called when nvl and detector are no longer
196 * - An error happened and a zevent can't be posted. In this case, cb is called
197 * before zfs_zevent_post() returns.
198 * - The event is being drained and freed.
201 zfs_zevent_post(nvlist_t *nvl, nvlist_t *detector, zevent_cb_t *cb)
213 tv_array[0] = tv.tv_sec;
214 tv_array[1] = tv.tv_nsec;
216 error = nvlist_add_int64_array(nvl, FM_EREPORT_TIME, tv_array, 2);
218 atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64);
222 eid = atomic_inc_64_nv(&zevent_eid);
223 error = nvlist_add_uint64(nvl, FM_EREPORT_EID, eid);
225 atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64);
229 error = nvlist_size(nvl, &nvl_size, NV_ENCODE_NATIVE);
231 atomic_inc_64(&erpt_kstat_data.erpt_dropped.value.ui64);
235 if (nvl_size > ERPT_DATA_SZ || nvl_size == 0) {
236 atomic_inc_64(&erpt_kstat_data.erpt_dropped.value.ui64);
241 ev = zfs_zevent_alloc();
243 atomic_inc_64(&erpt_kstat_data.erpt_dropped.value.ui64);
249 ev->ev_detector = detector;
253 mutex_enter(&zevent_lock);
254 zfs_zevent_insert(ev);
255 cv_broadcast(&zevent_cv);
256 mutex_exit(&zevent_lock);
266 zfs_zevent_track_duplicate(void)
268 atomic_inc_64(&erpt_kstat_data.erpt_duplicates.value.ui64);
272 zfs_zevent_minor_to_state(minor_t minor, zfs_zevent_t **ze)
274 *ze = zfsdev_get_state(minor, ZST_ZEVENT);
276 return (SET_ERROR(EBADF));
282 zfs_zevent_fd_hold(int fd, minor_t *minorp, zfs_zevent_t **ze)
284 zfs_file_t *fp = zfs_file_get(fd);
288 int error = zfsdev_getminor(fp, minorp);
290 error = zfs_zevent_minor_to_state(*minorp, ze);
293 zfs_zevent_fd_rele(fp);
301 zfs_zevent_fd_rele(zfs_file_t *fp)
307 * Get the next zevent in the stream and place a copy in 'event'. This
308 * may fail with ENOMEM if the encoded nvlist size exceeds the passed
309 * 'event_size'. In this case the stream pointer is not advanced and
310 * and 'event_size' is set to the minimum required buffer size.
313 zfs_zevent_next(zfs_zevent_t *ze, nvlist_t **event, uint64_t *event_size,
320 mutex_enter(&zevent_lock);
321 if (ze->ze_zevent == NULL) {
322 /* New stream start at the beginning/tail */
323 ev = list_tail(&zevent_list);
330 * Existing stream continue with the next element and remove
331 * ourselves from the wait queue for the previous element
333 ev = list_prev(&zevent_list, ze->ze_zevent);
340 VERIFY(nvlist_size(ev->ev_nvl, &size, NV_ENCODE_NATIVE) == 0);
341 if (size > *event_size) {
348 list_remove(&ze->ze_zevent->ev_ze_list, ze);
351 list_insert_head(&ev->ev_ze_list, ze);
352 (void) nvlist_dup(ev->ev_nvl, event, KM_SLEEP);
353 *dropped = ze->ze_dropped;
356 /* Include events dropped due to rate limiting */
357 *dropped += atomic_swap_64(&ratelimit_dropped, 0);
361 mutex_exit(&zevent_lock);
367 * Wait in an interruptible state for any new events.
370 zfs_zevent_wait(zfs_zevent_t *ze)
374 mutex_enter(&zevent_lock);
377 while (error == EAGAIN) {
378 if (zevent_flags & ZEVENT_SHUTDOWN) {
379 error = SET_ERROR(ESHUTDOWN);
383 error = cv_wait_sig(&zevent_cv, &zevent_lock);
384 if (signal_pending(current)) {
385 error = SET_ERROR(EINTR);
387 } else if (!list_is_empty(&zevent_list)) {
396 mutex_exit(&zevent_lock);
402 * The caller may seek to a specific EID by passing that EID. If the EID
403 * is still available in the posted list of events the cursor is positioned
404 * there. Otherwise ENOENT is returned and the cursor is not moved.
406 * There are two reserved EIDs which may be passed and will never fail.
407 * ZEVENT_SEEK_START positions the cursor at the start of the list, and
408 * ZEVENT_SEEK_END positions the cursor at the end of the list.
411 zfs_zevent_seek(zfs_zevent_t *ze, uint64_t eid)
416 mutex_enter(&zevent_lock);
418 if (eid == ZEVENT_SEEK_START) {
420 list_remove(&ze->ze_zevent->ev_ze_list, ze);
422 ze->ze_zevent = NULL;
426 if (eid == ZEVENT_SEEK_END) {
428 list_remove(&ze->ze_zevent->ev_ze_list, ze);
430 ev = list_head(&zevent_list);
433 list_insert_head(&ev->ev_ze_list, ze);
435 ze->ze_zevent = NULL;
441 for (ev = list_tail(&zevent_list); ev != NULL;
442 ev = list_prev(&zevent_list, ev)) {
443 if (ev->ev_eid == eid) {
445 list_remove(&ze->ze_zevent->ev_ze_list, ze);
448 list_insert_head(&ev->ev_ze_list, ze);
457 mutex_exit(&zevent_lock);
463 zfs_zevent_init(zfs_zevent_t **zep)
467 ze = *zep = kmem_zalloc(sizeof (zfs_zevent_t), KM_SLEEP);
468 list_link_init(&ze->ze_node);
472 zfs_zevent_destroy(zfs_zevent_t *ze)
474 mutex_enter(&zevent_lock);
476 list_remove(&ze->ze_zevent->ev_ze_list, ze);
477 mutex_exit(&zevent_lock);
479 kmem_free(ze, sizeof (zfs_zevent_t));
484 * Wrappers for FM nvlist allocators
487 i_fm_alloc(nv_alloc_t *nva, size_t size)
490 return (kmem_zalloc(size, KM_SLEEP));
494 i_fm_free(nv_alloc_t *nva, void *buf, size_t size)
497 kmem_free(buf, size);
500 static const nv_alloc_ops_t fm_mem_alloc_ops = {
503 .nv_ao_alloc = i_fm_alloc,
504 .nv_ao_free = i_fm_free,
509 * Create and initialize a new nv_alloc_t for a fixed buffer, buf. A pointer
510 * to the newly allocated nv_alloc_t structure is returned upon success or NULL
511 * is returned to indicate that the nv_alloc structure could not be created.
514 fm_nva_xcreate(char *buf, size_t bufsz)
516 nv_alloc_t *nvhdl = kmem_zalloc(sizeof (nv_alloc_t), KM_SLEEP);
518 if (bufsz == 0 || nv_alloc_init(nvhdl, nv_fixed_ops, buf, bufsz) != 0) {
519 kmem_free(nvhdl, sizeof (nv_alloc_t));
527 * Destroy a previously allocated nv_alloc structure. The fixed buffer
528 * associated with nva must be freed by the caller.
531 fm_nva_xdestroy(nv_alloc_t *nva)
534 kmem_free(nva, sizeof (nv_alloc_t));
538 * Create a new nv list. A pointer to a new nv list structure is returned
539 * upon success or NULL is returned to indicate that the structure could
540 * not be created. The newly created nv list is created and managed by the
541 * operations installed in nva. If nva is NULL, the default FMA nva
542 * operations are installed and used.
544 * When called from the kernel and nva == NULL, this function must be called
545 * from passive kernel context with no locks held that can prevent a
546 * sleeping memory allocation from occurring. Otherwise, this function may
547 * be called from other kernel contexts as long a valid nva created via
548 * fm_nva_create() is supplied.
551 fm_nvlist_create(nv_alloc_t *nva)
558 nvhdl = kmem_zalloc(sizeof (nv_alloc_t), KM_SLEEP);
560 if (nv_alloc_init(nvhdl, &fm_mem_alloc_ops, NULL, 0) != 0) {
561 kmem_free(nvhdl, sizeof (nv_alloc_t));
569 if (nvlist_xalloc(&nvl, NV_UNIQUE_NAME, nvhdl) != 0) {
571 nv_alloc_fini(nvhdl);
572 kmem_free(nvhdl, sizeof (nv_alloc_t));
581 * Destroy a previously allocated nvlist structure. flag indicates whether
582 * or not the associated nva structure should be freed (FM_NVA_FREE) or
583 * retained (FM_NVA_RETAIN). Retaining the nv alloc structure allows
584 * it to be re-used for future nvlist creation operations.
587 fm_nvlist_destroy(nvlist_t *nvl, int flag)
589 nv_alloc_t *nva = nvlist_lookup_nv_alloc(nvl);
594 if (flag == FM_NVA_FREE)
595 fm_nva_xdestroy(nva);
600 i_fm_payload_set(nvlist_t *payload, const char *name, va_list ap)
605 while (ret == 0 && name != NULL) {
606 type = va_arg(ap, data_type_t);
609 ret = nvlist_add_byte(payload, name,
612 case DATA_TYPE_BYTE_ARRAY:
613 nelem = va_arg(ap, int);
614 ret = nvlist_add_byte_array(payload, name,
615 va_arg(ap, uchar_t *), nelem);
617 case DATA_TYPE_BOOLEAN_VALUE:
618 ret = nvlist_add_boolean_value(payload, name,
619 va_arg(ap, boolean_t));
621 case DATA_TYPE_BOOLEAN_ARRAY:
622 nelem = va_arg(ap, int);
623 ret = nvlist_add_boolean_array(payload, name,
624 va_arg(ap, boolean_t *), nelem);
627 ret = nvlist_add_int8(payload, name,
630 case DATA_TYPE_INT8_ARRAY:
631 nelem = va_arg(ap, int);
632 ret = nvlist_add_int8_array(payload, name,
633 va_arg(ap, int8_t *), nelem);
635 case DATA_TYPE_UINT8:
636 ret = nvlist_add_uint8(payload, name,
639 case DATA_TYPE_UINT8_ARRAY:
640 nelem = va_arg(ap, int);
641 ret = nvlist_add_uint8_array(payload, name,
642 va_arg(ap, uint8_t *), nelem);
644 case DATA_TYPE_INT16:
645 ret = nvlist_add_int16(payload, name,
648 case DATA_TYPE_INT16_ARRAY:
649 nelem = va_arg(ap, int);
650 ret = nvlist_add_int16_array(payload, name,
651 va_arg(ap, int16_t *), nelem);
653 case DATA_TYPE_UINT16:
654 ret = nvlist_add_uint16(payload, name,
657 case DATA_TYPE_UINT16_ARRAY:
658 nelem = va_arg(ap, int);
659 ret = nvlist_add_uint16_array(payload, name,
660 va_arg(ap, uint16_t *), nelem);
662 case DATA_TYPE_INT32:
663 ret = nvlist_add_int32(payload, name,
664 va_arg(ap, int32_t));
666 case DATA_TYPE_INT32_ARRAY:
667 nelem = va_arg(ap, int);
668 ret = nvlist_add_int32_array(payload, name,
669 va_arg(ap, int32_t *), nelem);
671 case DATA_TYPE_UINT32:
672 ret = nvlist_add_uint32(payload, name,
673 va_arg(ap, uint32_t));
675 case DATA_TYPE_UINT32_ARRAY:
676 nelem = va_arg(ap, int);
677 ret = nvlist_add_uint32_array(payload, name,
678 va_arg(ap, uint32_t *), nelem);
680 case DATA_TYPE_INT64:
681 ret = nvlist_add_int64(payload, name,
682 va_arg(ap, int64_t));
684 case DATA_TYPE_INT64_ARRAY:
685 nelem = va_arg(ap, int);
686 ret = nvlist_add_int64_array(payload, name,
687 va_arg(ap, int64_t *), nelem);
689 case DATA_TYPE_UINT64:
690 ret = nvlist_add_uint64(payload, name,
691 va_arg(ap, uint64_t));
693 case DATA_TYPE_UINT64_ARRAY:
694 nelem = va_arg(ap, int);
695 ret = nvlist_add_uint64_array(payload, name,
696 va_arg(ap, uint64_t *), nelem);
698 case DATA_TYPE_STRING:
699 ret = nvlist_add_string(payload, name,
702 case DATA_TYPE_STRING_ARRAY:
703 nelem = va_arg(ap, int);
704 ret = nvlist_add_string_array(payload, name,
705 va_arg(ap, const char **), nelem);
707 case DATA_TYPE_NVLIST:
708 ret = nvlist_add_nvlist(payload, name,
709 va_arg(ap, nvlist_t *));
711 case DATA_TYPE_NVLIST_ARRAY:
712 nelem = va_arg(ap, int);
713 ret = nvlist_add_nvlist_array(payload, name,
714 va_arg(ap, const nvlist_t **), nelem);
720 name = va_arg(ap, char *);
726 fm_payload_set(nvlist_t *payload, ...)
732 va_start(ap, payload);
733 name = va_arg(ap, char *);
734 ret = i_fm_payload_set(payload, name, ap);
738 atomic_inc_64(&erpt_kstat_data.payload_set_failed.value.ui64);
742 * Set-up and validate the members of an ereport event according to:
744 * Member name Type Value
745 * ====================================================
746 * class string ereport
749 * detector nvlist_t <detector>
750 * ereport-payload nvlist_t <var args>
752 * We don't actually add a 'version' member to the payload. Really,
753 * the version quoted to us by our caller is that of the category 1
754 * "ereport" event class (and we require FM_EREPORT_VERS0) but
755 * the payload version of the actual leaf class event under construction
756 * may be something else. Callers should supply a version in the varargs,
757 * or (better) we could take two version arguments - one for the
758 * ereport category 1 classification (expect FM_EREPORT_VERS0) and one
759 * for the leaf class.
762 fm_ereport_set(nvlist_t *ereport, int version, const char *erpt_class,
763 uint64_t ena, const nvlist_t *detector, ...)
765 char ereport_class[FM_MAX_CLASS];
770 if (version != FM_EREPORT_VERS0) {
771 atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64);
775 (void) snprintf(ereport_class, FM_MAX_CLASS, "%s.%s",
776 FM_EREPORT_CLASS, erpt_class);
777 if (nvlist_add_string(ereport, FM_CLASS, ereport_class) != 0) {
778 atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64);
782 if (nvlist_add_uint64(ereport, FM_EREPORT_ENA, ena)) {
783 atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64);
786 if (nvlist_add_nvlist(ereport, FM_EREPORT_DETECTOR,
787 (nvlist_t *)detector) != 0) {
788 atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64);
791 va_start(ap, detector);
792 name = va_arg(ap, const char *);
793 ret = i_fm_payload_set(ereport, name, ap);
797 atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64);
801 * Set-up and validate the members of an hc fmri according to;
803 * Member name Type Value
804 * ===================================================
806 * auth nvlist_t <auth>
807 * hc-name string <name>
810 * Note that auth and hc-id are optional members.
813 #define HC_MAXPAIRS 20
814 #define HC_MAXNAMELEN 50
817 fm_fmri_hc_set_common(nvlist_t *fmri, int version, const nvlist_t *auth)
819 if (version != FM_HC_SCHEME_VERSION) {
820 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
824 if (nvlist_add_uint8(fmri, FM_VERSION, version) != 0 ||
825 nvlist_add_string(fmri, FM_FMRI_SCHEME, FM_FMRI_SCHEME_HC) != 0) {
826 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
830 if (auth != NULL && nvlist_add_nvlist(fmri, FM_FMRI_AUTHORITY,
831 (nvlist_t *)auth) != 0) {
832 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
840 fm_fmri_hc_set(nvlist_t *fmri, int version, const nvlist_t *auth,
841 nvlist_t *snvl, int npairs, ...)
843 nv_alloc_t *nva = nvlist_lookup_nv_alloc(fmri);
844 nvlist_t *pairs[HC_MAXPAIRS];
848 if (!fm_fmri_hc_set_common(fmri, version, auth))
851 npairs = MIN(npairs, HC_MAXPAIRS);
853 va_start(ap, npairs);
854 for (i = 0; i < npairs; i++) {
855 const char *name = va_arg(ap, const char *);
856 uint32_t id = va_arg(ap, uint32_t);
859 (void) snprintf(idstr, sizeof (idstr), "%u", id);
861 pairs[i] = fm_nvlist_create(nva);
862 if (nvlist_add_string(pairs[i], FM_FMRI_HC_NAME, name) != 0 ||
863 nvlist_add_string(pairs[i], FM_FMRI_HC_ID, idstr) != 0) {
865 &erpt_kstat_data.fmri_set_failed.value.ui64);
870 if (nvlist_add_nvlist_array(fmri, FM_FMRI_HC_LIST,
871 (const nvlist_t **)pairs, npairs) != 0) {
872 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
875 for (i = 0; i < npairs; i++)
876 fm_nvlist_destroy(pairs[i], FM_NVA_RETAIN);
879 if (nvlist_add_nvlist(fmri, FM_FMRI_HC_SPECIFIC, snvl) != 0) {
881 &erpt_kstat_data.fmri_set_failed.value.ui64);
887 fm_fmri_hc_create(nvlist_t *fmri, int version, const nvlist_t *auth,
888 nvlist_t *snvl, nvlist_t *bboard, int npairs, ...)
890 nv_alloc_t *nva = nvlist_lookup_nv_alloc(fmri);
891 nvlist_t *pairs[HC_MAXPAIRS];
898 if (!fm_fmri_hc_set_common(fmri, version, auth))
902 * copy the bboard nvpairs to the pairs array
904 if (nvlist_lookup_nvlist_array(bboard, FM_FMRI_HC_LIST, &hcl, &n)
906 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
910 for (i = 0; i < n; i++) {
911 if (nvlist_lookup_string(hcl[i], FM_FMRI_HC_NAME,
914 &erpt_kstat_data.fmri_set_failed.value.ui64);
917 if (nvlist_lookup_string(hcl[i], FM_FMRI_HC_ID, &hcid) != 0) {
919 &erpt_kstat_data.fmri_set_failed.value.ui64);
923 pairs[i] = fm_nvlist_create(nva);
924 if (nvlist_add_string(pairs[i], FM_FMRI_HC_NAME, hcname) != 0 ||
925 nvlist_add_string(pairs[i], FM_FMRI_HC_ID, hcid) != 0) {
926 for (j = 0; j <= i; j++) {
927 if (pairs[j] != NULL)
928 fm_nvlist_destroy(pairs[j],
932 &erpt_kstat_data.fmri_set_failed.value.ui64);
938 * create the pairs from passed in pairs
940 npairs = MIN(npairs, HC_MAXPAIRS);
942 va_start(ap, npairs);
943 for (i = n; i < npairs + n; i++) {
944 const char *name = va_arg(ap, const char *);
945 uint32_t id = va_arg(ap, uint32_t);
947 (void) snprintf(idstr, sizeof (idstr), "%u", id);
948 pairs[i] = fm_nvlist_create(nva);
949 if (nvlist_add_string(pairs[i], FM_FMRI_HC_NAME, name) != 0 ||
950 nvlist_add_string(pairs[i], FM_FMRI_HC_ID, idstr) != 0) {
951 for (j = 0; j <= i; j++) {
952 if (pairs[j] != NULL)
953 fm_nvlist_destroy(pairs[j],
957 &erpt_kstat_data.fmri_set_failed.value.ui64);
964 * Create the fmri hc list
966 if (nvlist_add_nvlist_array(fmri, FM_FMRI_HC_LIST,
967 (const nvlist_t **)pairs, npairs + n) != 0) {
968 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
972 for (i = 0; i < npairs + n; i++) {
973 fm_nvlist_destroy(pairs[i], FM_NVA_RETAIN);
977 if (nvlist_add_nvlist(fmri, FM_FMRI_HC_SPECIFIC, snvl) != 0) {
979 &erpt_kstat_data.fmri_set_failed.value.ui64);
986 * Set-up and validate the members of an dev fmri according to:
988 * Member name Type Value
989 * ====================================================
991 * auth nvlist_t <auth>
992 * devpath string <devpath>
993 * [devid] string <devid>
994 * [target-port-l0id] string <target-port-lun0-id>
996 * Note that auth and devid are optional members.
999 fm_fmri_dev_set(nvlist_t *fmri_dev, int version, const nvlist_t *auth,
1000 const char *devpath, const char *devid, const char *tpl0)
1004 if (version != DEV_SCHEME_VERSION0) {
1005 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1009 err |= nvlist_add_uint8(fmri_dev, FM_VERSION, version);
1010 err |= nvlist_add_string(fmri_dev, FM_FMRI_SCHEME, FM_FMRI_SCHEME_DEV);
1013 err |= nvlist_add_nvlist(fmri_dev, FM_FMRI_AUTHORITY,
1017 err |= nvlist_add_string(fmri_dev, FM_FMRI_DEV_PATH, devpath);
1020 err |= nvlist_add_string(fmri_dev, FM_FMRI_DEV_ID, devid);
1023 err |= nvlist_add_string(fmri_dev, FM_FMRI_DEV_TGTPTLUN0, tpl0);
1026 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1031 * Set-up and validate the members of an cpu fmri according to:
1033 * Member name Type Value
1034 * ====================================================
1036 * auth nvlist_t <auth>
1037 * cpuid uint32_t <cpu_id>
1038 * cpumask uint8_t <cpu_mask>
1039 * serial uint64_t <serial_id>
1041 * Note that auth, cpumask, serial are optional members.
1045 fm_fmri_cpu_set(nvlist_t *fmri_cpu, int version, const nvlist_t *auth,
1046 uint32_t cpu_id, uint8_t *cpu_maskp, const char *serial_idp)
1048 uint64_t *failedp = &erpt_kstat_data.fmri_set_failed.value.ui64;
1050 if (version < CPU_SCHEME_VERSION1) {
1051 atomic_inc_64(failedp);
1055 if (nvlist_add_uint8(fmri_cpu, FM_VERSION, version) != 0) {
1056 atomic_inc_64(failedp);
1060 if (nvlist_add_string(fmri_cpu, FM_FMRI_SCHEME,
1061 FM_FMRI_SCHEME_CPU) != 0) {
1062 atomic_inc_64(failedp);
1066 if (auth != NULL && nvlist_add_nvlist(fmri_cpu, FM_FMRI_AUTHORITY,
1067 (nvlist_t *)auth) != 0)
1068 atomic_inc_64(failedp);
1070 if (nvlist_add_uint32(fmri_cpu, FM_FMRI_CPU_ID, cpu_id) != 0)
1071 atomic_inc_64(failedp);
1073 if (cpu_maskp != NULL && nvlist_add_uint8(fmri_cpu, FM_FMRI_CPU_MASK,
1075 atomic_inc_64(failedp);
1077 if (serial_idp == NULL || nvlist_add_string(fmri_cpu,
1078 FM_FMRI_CPU_SERIAL_ID, (char *)serial_idp) != 0)
1079 atomic_inc_64(failedp);
1083 * Set-up and validate the members of a mem according to:
1085 * Member name Type Value
1086 * ====================================================
1088 * auth nvlist_t <auth> [optional]
1089 * unum string <unum>
1090 * serial string <serial> [optional*]
1091 * offset uint64_t <offset> [optional]
1093 * * serial is required if offset is present
1096 fm_fmri_mem_set(nvlist_t *fmri, int version, const nvlist_t *auth,
1097 const char *unum, const char *serial, uint64_t offset)
1099 if (version != MEM_SCHEME_VERSION0) {
1100 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1104 if (!serial && (offset != (uint64_t)-1)) {
1105 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1109 if (nvlist_add_uint8(fmri, FM_VERSION, version) != 0) {
1110 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1114 if (nvlist_add_string(fmri, FM_FMRI_SCHEME, FM_FMRI_SCHEME_MEM) != 0) {
1115 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1120 if (nvlist_add_nvlist(fmri, FM_FMRI_AUTHORITY,
1121 (nvlist_t *)auth) != 0) {
1123 &erpt_kstat_data.fmri_set_failed.value.ui64);
1127 if (nvlist_add_string(fmri, FM_FMRI_MEM_UNUM, unum) != 0) {
1128 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1131 if (serial != NULL) {
1132 if (nvlist_add_string_array(fmri, FM_FMRI_MEM_SERIAL_ID,
1133 (const char **)&serial, 1) != 0) {
1135 &erpt_kstat_data.fmri_set_failed.value.ui64);
1137 if (offset != (uint64_t)-1 && nvlist_add_uint64(fmri,
1138 FM_FMRI_MEM_OFFSET, offset) != 0) {
1140 &erpt_kstat_data.fmri_set_failed.value.ui64);
1146 fm_fmri_zfs_set(nvlist_t *fmri, int version, uint64_t pool_guid,
1149 if (version != ZFS_SCHEME_VERSION0) {
1150 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1154 if (nvlist_add_uint8(fmri, FM_VERSION, version) != 0) {
1155 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1159 if (nvlist_add_string(fmri, FM_FMRI_SCHEME, FM_FMRI_SCHEME_ZFS) != 0) {
1160 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1164 if (nvlist_add_uint64(fmri, FM_FMRI_ZFS_POOL, pool_guid) != 0) {
1165 atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64);
1168 if (vdev_guid != 0) {
1169 if (nvlist_add_uint64(fmri, FM_FMRI_ZFS_VDEV, vdev_guid) != 0) {
1171 &erpt_kstat_data.fmri_set_failed.value.ui64);
1177 fm_ena_increment(uint64_t ena)
1181 switch (ENA_FORMAT(ena)) {
1183 new_ena = ena + (1 << ENA_FMT1_GEN_SHFT);
1186 new_ena = ena + (1 << ENA_FMT2_GEN_SHFT);
1196 fm_ena_generate_cpu(uint64_t timestamp, processorid_t cpuid, uchar_t format)
1203 ena = (uint64_t)((format & ENA_FORMAT_MASK) |
1204 ((cpuid << ENA_FMT1_CPUID_SHFT) &
1205 ENA_FMT1_CPUID_MASK) |
1206 ((timestamp << ENA_FMT1_TIME_SHFT) &
1207 ENA_FMT1_TIME_MASK));
1209 ena = (uint64_t)((format & ENA_FORMAT_MASK) |
1210 ((cpuid << ENA_FMT1_CPUID_SHFT) &
1211 ENA_FMT1_CPUID_MASK) |
1212 ((gethrtime() << ENA_FMT1_TIME_SHFT) &
1213 ENA_FMT1_TIME_MASK));
1217 ena = (uint64_t)((format & ENA_FORMAT_MASK) |
1218 ((timestamp << ENA_FMT2_TIME_SHFT) & ENA_FMT2_TIME_MASK));
1228 fm_ena_generate(uint64_t timestamp, uchar_t format)
1233 ena = fm_ena_generate_cpu(timestamp, getcpuid(), format);
1240 fm_ena_generation_get(uint64_t ena)
1244 switch (ENA_FORMAT(ena)) {
1246 gen = (ena & ENA_FMT1_GEN_MASK) >> ENA_FMT1_GEN_SHFT;
1249 gen = (ena & ENA_FMT2_GEN_MASK) >> ENA_FMT2_GEN_SHFT;
1260 fm_ena_format_get(uint64_t ena)
1263 return (ENA_FORMAT(ena));
1267 fm_ena_id_get(uint64_t ena)
1271 switch (ENA_FORMAT(ena)) {
1273 id = (ena & ENA_FMT1_ID_MASK) >> ENA_FMT1_ID_SHFT;
1276 id = (ena & ENA_FMT2_ID_MASK) >> ENA_FMT2_ID_SHFT;
1286 fm_ena_time_get(uint64_t ena)
1290 switch (ENA_FORMAT(ena)) {
1292 time = (ena & ENA_FMT1_TIME_MASK) >> ENA_FMT1_TIME_SHFT;
1295 time = (ena & ENA_FMT2_TIME_MASK) >> ENA_FMT2_TIME_SHFT;
1306 * Helper function to increment ereport dropped count. Used by the event
1307 * rate limiting code to give feedback to the user about how many events were
1308 * rate limited by including them in the 'dropped' count.
1311 fm_erpt_dropped_increment(void)
1313 atomic_inc_64(&ratelimit_dropped);
1322 /* Initialize zevent allocation and generation kstats */
1323 fm_ksp = kstat_create("zfs", 0, "fm", "misc", KSTAT_TYPE_NAMED,
1324 sizeof (struct erpt_kstat) / sizeof (kstat_named_t),
1325 KSTAT_FLAG_VIRTUAL);
1327 if (fm_ksp != NULL) {
1328 fm_ksp->ks_data = &erpt_kstat_data;
1329 kstat_install(fm_ksp);
1331 cmn_err(CE_NOTE, "failed to create fm/misc kstat\n");
1334 mutex_init(&zevent_lock, NULL, MUTEX_DEFAULT, NULL);
1335 list_create(&zevent_list, sizeof (zevent_t),
1336 offsetof(zevent_t, ev_node));
1337 cv_init(&zevent_cv, NULL, CV_DEFAULT, NULL);
1349 zfs_zevent_drain_all(&count);
1351 mutex_enter(&zevent_lock);
1352 cv_broadcast(&zevent_cv);
1354 zevent_flags |= ZEVENT_SHUTDOWN;
1355 while (zevent_waiters > 0) {
1356 mutex_exit(&zevent_lock);
1358 mutex_enter(&zevent_lock);
1360 mutex_exit(&zevent_lock);
1362 cv_destroy(&zevent_cv);
1363 list_destroy(&zevent_list);
1364 mutex_destroy(&zevent_lock);
1366 if (fm_ksp != NULL) {
1367 kstat_delete(fm_ksp);
1371 #endif /* _KERNEL */
1373 ZFS_MODULE_PARAM(zfs_zevent, zfs_zevent_, len_max, INT, ZMOD_RW,
1374 "Max event queue length");