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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
27 * Copyright (c) 2012,2021 by Delphix. All rights reserved.
31 #include <sys/spa_impl.h>
33 #include <sys/vdev_impl.h>
35 #include <sys/zio_checksum.h>
37 #include <sys/fm/fs/zfs.h>
38 #include <sys/fm/protocol.h>
39 #include <sys/fm/util.h>
40 #include <sys/sysevent.h>
43 * This general routine is responsible for generating all the different ZFS
44 * ereports. The payload is dependent on the class, and which arguments are
45 * supplied to the function:
47 * EREPORT POOL VDEV IO
53 * If we are in a loading state, all errors are chained together by the same
54 * SPA-wide ENA (Error Numeric Association).
56 * For isolated I/O requests, we get the ENA from the zio_t. The propagation
57 * gets very complicated due to RAID-Z, gang blocks, and vdev caching. We want
58 * to chain together all ereports associated with a logical piece of data. For
59 * read I/Os, there are basically three 'types' of I/O, which form a roughly
63 * | Aggregate I/O | No associated logical data or device
67 * +---------------+ Reads associated with a piece of logical data.
68 * | Read I/O | This includes reads on behalf of RAID-Z,
69 * +---------------+ mirrors, gang blocks, retries, etc.
72 * +---------------+ Reads associated with a particular device, but
73 * | Physical I/O | no logical data. Issued as part of vdev caching
74 * +---------------+ and I/O aggregation.
76 * Note that 'physical I/O' here is not the same terminology as used in the rest
77 * of ZIO. Typically, 'physical I/O' simply means that there is no attached
78 * blockpointer. But I/O with no associated block pointer can still be related
79 * to a logical piece of data (i.e. RAID-Z requests).
81 * Purely physical I/O always have unique ENAs. They are not related to a
82 * particular piece of logical data, and therefore cannot be chained together.
83 * We still generate an ereport, but the DE doesn't correlate it with any
84 * logical piece of data. When such an I/O fails, the delegated I/O requests
85 * will issue a retry, which will trigger the 'real' ereport with the correct
88 * We keep track of the ENA for a ZIO chain through the 'io_logical' member.
89 * When a new logical I/O is issued, we set this to point to itself. Child I/Os
90 * then inherit this pointer, so that when it is first set subsequent failures
91 * will use the same ENA. For vdev cache fill and queue aggregation I/O,
92 * this pointer is set to NULL, and no ereport will be generated (since it
93 * doesn't actually correspond to any particular device or piece of data,
94 * and the caller will always retry without caching or queueing anyway).
96 * For checksum errors, we want to include more information about the actual
97 * error which occurs. Accordingly, we build an ereport when the error is
98 * noticed, but instead of sending it in immediately, we hang it off of the
99 * io_cksum_report field of the logical IO. When the logical IO completes
100 * (successfully or not), zfs_ereport_finish_checksum() is called with the
101 * good and bad versions of the buffer (if available), and we annotate the
102 * ereport with information about the differences.
107 * Duplicate ereport Detection
109 * Some ereports are retained momentarily for detecting duplicates. These
110 * are kept in a recent_events_node_t in both a time-ordered list and an AVL
111 * tree of recent unique ereports.
113 * The lifespan of these recent ereports is bounded (15 mins) and a cleaner
114 * task is used to purge stale entries.
116 static list_t recent_events_list;
117 static avl_tree_t recent_events_tree;
118 static kmutex_t recent_events_lock;
119 static taskqid_t recent_events_cleaner_tqid;
122 * Each node is about 128 bytes so 2,000 would consume 1/4 MiB.
124 * This setting can be changed dynamically and setting it to zero
125 * disables duplicate detection.
127 static unsigned int zfs_zevent_retain_max = 2000;
130 * The lifespan for a recent ereport entry. The default of 15 minutes is
131 * intended to outlive the zfs diagnosis engine's threshold of 10 errors
132 * over a period of 10 minutes.
134 static unsigned int zfs_zevent_retain_expire_secs = 900;
136 typedef enum zfs_subclass {
143 /* common criteria */
144 uint64_t re_pool_guid;
145 uint64_t re_vdev_guid;
148 uint64_t re_io_offset;
149 zfs_subclass_t re_subclass;
150 zio_priority_t re_io_priority;
152 /* logical zio criteria (optional) */
153 zbookmark_phys_t re_io_bookmark;
156 avl_node_t re_tree_link;
157 list_node_t re_list_link;
158 uint64_t re_timestamp;
159 } recent_events_node_t;
162 recent_events_compare(const void *a, const void *b)
164 const recent_events_node_t *node1 = a;
165 const recent_events_node_t *node2 = b;
169 * The comparison order here is somewhat arbitrary.
170 * What's important is that if every criteria matches, then it
171 * is a duplicate (i.e. compare returns 0)
173 if ((cmp = TREE_CMP(node1->re_subclass, node2->re_subclass)) != 0)
175 if ((cmp = TREE_CMP(node1->re_pool_guid, node2->re_pool_guid)) != 0)
177 if ((cmp = TREE_CMP(node1->re_vdev_guid, node2->re_vdev_guid)) != 0)
179 if ((cmp = TREE_CMP(node1->re_io_error, node2->re_io_error)) != 0)
181 if ((cmp = TREE_CMP(node1->re_io_priority, node2->re_io_priority)) != 0)
183 if ((cmp = TREE_CMP(node1->re_io_size, node2->re_io_size)) != 0)
185 if ((cmp = TREE_CMP(node1->re_io_offset, node2->re_io_offset)) != 0)
188 const zbookmark_phys_t *zb1 = &node1->re_io_bookmark;
189 const zbookmark_phys_t *zb2 = &node2->re_io_bookmark;
191 if ((cmp = TREE_CMP(zb1->zb_objset, zb2->zb_objset)) != 0)
193 if ((cmp = TREE_CMP(zb1->zb_object, zb2->zb_object)) != 0)
195 if ((cmp = TREE_CMP(zb1->zb_level, zb2->zb_level)) != 0)
197 if ((cmp = TREE_CMP(zb1->zb_blkid, zb2->zb_blkid)) != 0)
204 * workaround: vdev properties don't have inheritance
207 vdev_prop_get_inherited(vdev_t *vd, vdev_prop_t prop)
209 uint64_t propdef, propval;
211 propdef = vdev_prop_default_numeric(prop);
213 case VDEV_PROP_CHECKSUM_N:
214 propval = vd->vdev_checksum_n;
216 case VDEV_PROP_CHECKSUM_T:
217 propval = vd->vdev_checksum_t;
220 propval = vd->vdev_io_n;
223 propval = vd->vdev_io_t;
225 case VDEV_PROP_SLOW_IO_N:
226 propval = vd->vdev_slow_io_n;
228 case VDEV_PROP_SLOW_IO_T:
229 propval = vd->vdev_slow_io_t;
236 if (propval != propdef)
239 if (vd->vdev_parent == NULL)
242 return (vdev_prop_get_inherited(vd->vdev_parent, prop));
245 static void zfs_ereport_schedule_cleaner(void);
248 * background task to clean stale recent event nodes.
251 zfs_ereport_cleaner(void *arg)
253 recent_events_node_t *entry;
254 uint64_t now = gethrtime();
257 * purge expired entries
259 mutex_enter(&recent_events_lock);
260 while ((entry = list_tail(&recent_events_list)) != NULL) {
261 uint64_t age = NSEC2SEC(now - entry->re_timestamp);
262 if (age <= zfs_zevent_retain_expire_secs)
265 /* remove expired node */
266 avl_remove(&recent_events_tree, entry);
267 list_remove(&recent_events_list, entry);
268 kmem_free(entry, sizeof (*entry));
271 /* Restart the cleaner if more entries remain */
272 recent_events_cleaner_tqid = 0;
273 if (!list_is_empty(&recent_events_list))
274 zfs_ereport_schedule_cleaner();
276 mutex_exit(&recent_events_lock);
280 zfs_ereport_schedule_cleaner(void)
282 ASSERT(MUTEX_HELD(&recent_events_lock));
284 uint64_t timeout = SEC2NSEC(zfs_zevent_retain_expire_secs + 1);
286 recent_events_cleaner_tqid = taskq_dispatch_delay(
287 system_delay_taskq, zfs_ereport_cleaner, NULL, TQ_SLEEP,
288 ddi_get_lbolt() + NSEC_TO_TICK(timeout));
292 * Clear entries for a given vdev or all vdevs in a pool when vdev == NULL
295 zfs_ereport_clear(spa_t *spa, vdev_t *vd)
297 uint64_t vdev_guid, pool_guid;
299 ASSERT(vd != NULL || spa != NULL);
302 pool_guid = spa_guid(spa);
304 vdev_guid = vd->vdev_guid;
308 mutex_enter(&recent_events_lock);
310 recent_events_node_t *next = list_head(&recent_events_list);
311 while (next != NULL) {
312 recent_events_node_t *entry = next;
314 next = list_next(&recent_events_list, next);
316 if (entry->re_vdev_guid == vdev_guid ||
317 entry->re_pool_guid == pool_guid) {
318 avl_remove(&recent_events_tree, entry);
319 list_remove(&recent_events_list, entry);
320 kmem_free(entry, sizeof (*entry));
324 mutex_exit(&recent_events_lock);
328 * Check if an ereport would be a duplicate of one recently posted.
330 * An ereport is considered a duplicate if the set of criteria in
331 * recent_events_node_t all match.
333 * Only FM_EREPORT_ZFS_IO, FM_EREPORT_ZFS_DATA, and FM_EREPORT_ZFS_CHECKSUM
334 * are candidates for duplicate checking.
337 zfs_ereport_is_duplicate(const char *subclass, spa_t *spa, vdev_t *vd,
338 const zbookmark_phys_t *zb, zio_t *zio, uint64_t offset, uint64_t size)
340 recent_events_node_t search = {0}, *entry;
342 if (vd == NULL || zio == NULL)
345 if (zfs_zevent_retain_max == 0)
348 if (strcmp(subclass, FM_EREPORT_ZFS_IO) == 0)
349 search.re_subclass = ZSC_IO;
350 else if (strcmp(subclass, FM_EREPORT_ZFS_DATA) == 0)
351 search.re_subclass = ZSC_DATA;
352 else if (strcmp(subclass, FM_EREPORT_ZFS_CHECKSUM) == 0)
353 search.re_subclass = ZSC_CHECKSUM;
357 search.re_pool_guid = spa_guid(spa);
358 search.re_vdev_guid = vd->vdev_guid;
359 search.re_io_error = zio->io_error;
360 search.re_io_priority = zio->io_priority;
361 /* if size is supplied use it over what's in zio */
363 search.re_io_size = size;
364 search.re_io_offset = offset;
366 search.re_io_size = zio->io_size;
367 search.re_io_offset = zio->io_offset;
370 /* grab optional logical zio criteria */
372 search.re_io_bookmark.zb_objset = zb->zb_objset;
373 search.re_io_bookmark.zb_object = zb->zb_object;
374 search.re_io_bookmark.zb_level = zb->zb_level;
375 search.re_io_bookmark.zb_blkid = zb->zb_blkid;
378 uint64_t now = gethrtime();
380 mutex_enter(&recent_events_lock);
382 /* check if we have seen this one recently */
383 entry = avl_find(&recent_events_tree, &search, NULL);
385 uint64_t age = NSEC2SEC(now - entry->re_timestamp);
388 * There is still an active cleaner (since we're here).
389 * Reset the last seen time for this duplicate entry
390 * so that its lifespand gets extended.
392 list_remove(&recent_events_list, entry);
393 list_insert_head(&recent_events_list, entry);
394 entry->re_timestamp = now;
396 zfs_zevent_track_duplicate();
397 mutex_exit(&recent_events_lock);
399 return (age <= zfs_zevent_retain_expire_secs);
402 if (avl_numnodes(&recent_events_tree) >= zfs_zevent_retain_max) {
403 /* recycle oldest node */
404 entry = list_tail(&recent_events_list);
405 ASSERT(entry != NULL);
406 list_remove(&recent_events_list, entry);
407 avl_remove(&recent_events_tree, entry);
409 entry = kmem_alloc(sizeof (recent_events_node_t), KM_SLEEP);
412 /* record this as a recent ereport */
414 avl_add(&recent_events_tree, entry);
415 list_insert_head(&recent_events_list, entry);
416 entry->re_timestamp = now;
418 /* Start a cleaner if not already scheduled */
419 if (recent_events_cleaner_tqid == 0)
420 zfs_ereport_schedule_cleaner();
422 mutex_exit(&recent_events_lock);
427 zfs_zevent_post_cb(nvlist_t *nvl, nvlist_t *detector)
430 fm_nvlist_destroy(nvl, FM_NVA_FREE);
433 fm_nvlist_destroy(detector, FM_NVA_FREE);
437 * We want to rate limit ZIO delay, deadman, and checksum events so as to not
438 * flood zevent consumers when a disk is acting up.
440 * Returns 1 if we're ratelimiting, 0 if not.
443 zfs_is_ratelimiting_event(const char *subclass, vdev_t *vd)
447 * zfs_ratelimit() returns 1 if we're *not* ratelimiting and 0 if we
448 * are. Invert it to get our return value.
450 if (strcmp(subclass, FM_EREPORT_ZFS_DELAY) == 0) {
451 rc = !zfs_ratelimit(&vd->vdev_delay_rl);
452 } else if (strcmp(subclass, FM_EREPORT_ZFS_DEADMAN) == 0) {
453 rc = !zfs_ratelimit(&vd->vdev_deadman_rl);
454 } else if (strcmp(subclass, FM_EREPORT_ZFS_CHECKSUM) == 0) {
455 rc = !zfs_ratelimit(&vd->vdev_checksum_rl);
459 /* We're rate limiting */
460 fm_erpt_dropped_increment();
467 * Return B_TRUE if the event actually posted, B_FALSE if not.
470 zfs_ereport_start(nvlist_t **ereport_out, nvlist_t **detector_out,
471 const char *subclass, spa_t *spa, vdev_t *vd, const zbookmark_phys_t *zb,
472 zio_t *zio, uint64_t stateoroffset, uint64_t size)
474 nvlist_t *ereport, *detector;
479 if ((ereport = fm_nvlist_create(NULL)) == NULL)
482 if ((detector = fm_nvlist_create(NULL)) == NULL) {
483 fm_nvlist_destroy(ereport, FM_NVA_FREE);
488 * Serialize ereport generation
490 mutex_enter(&spa->spa_errlist_lock);
493 * Determine the ENA to use for this event. If we are in a loading
494 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use
495 * a root zio-wide ENA. Otherwise, simply use a unique ENA.
497 if (spa_load_state(spa) != SPA_LOAD_NONE) {
498 if (spa->spa_ena == 0)
499 spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1);
501 } else if (zio != NULL && zio->io_logical != NULL) {
502 if (zio->io_logical->io_ena == 0)
503 zio->io_logical->io_ena =
504 fm_ena_generate(0, FM_ENA_FMT1);
505 ena = zio->io_logical->io_ena;
507 ena = fm_ena_generate(0, FM_ENA_FMT1);
511 * Construct the full class, detector, and other standard FMA fields.
513 (void) snprintf(class, sizeof (class), "%s.%s",
514 ZFS_ERROR_CLASS, subclass);
516 fm_fmri_zfs_set(detector, FM_ZFS_SCHEME_VERSION, spa_guid(spa),
517 vd != NULL ? vd->vdev_guid : 0);
519 fm_ereport_set(ereport, FM_EREPORT_VERSION, class, ena, detector, NULL);
522 * Construct the per-ereport payload, depending on which parameters are
527 * Generic payload members common to all ereports.
529 fm_payload_set(ereport,
530 FM_EREPORT_PAYLOAD_ZFS_POOL, DATA_TYPE_STRING, spa_name(spa),
531 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, DATA_TYPE_UINT64, spa_guid(spa),
532 FM_EREPORT_PAYLOAD_ZFS_POOL_STATE, DATA_TYPE_UINT64,
533 (uint64_t)spa_state(spa),
534 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, DATA_TYPE_INT32,
535 (int32_t)spa_load_state(spa), NULL);
537 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE,
539 spa_get_failmode(spa) == ZIO_FAILURE_MODE_WAIT ?
540 FM_EREPORT_FAILMODE_WAIT :
541 spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE ?
542 FM_EREPORT_FAILMODE_CONTINUE : FM_EREPORT_FAILMODE_PANIC,
546 vdev_t *pvd = vd->vdev_parent;
547 vdev_queue_t *vq = &vd->vdev_queue;
548 vdev_stat_t *vs = &vd->vdev_stat;
550 uint64_t *spare_guids;
554 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID,
555 DATA_TYPE_UINT64, vd->vdev_guid,
556 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE,
557 DATA_TYPE_STRING, vd->vdev_ops->vdev_op_type, NULL);
558 if (vd->vdev_path != NULL)
559 fm_payload_set(ereport,
560 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH,
561 DATA_TYPE_STRING, vd->vdev_path, NULL);
562 if (vd->vdev_devid != NULL)
563 fm_payload_set(ereport,
564 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID,
565 DATA_TYPE_STRING, vd->vdev_devid, NULL);
566 if (vd->vdev_fru != NULL)
567 fm_payload_set(ereport,
568 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU,
569 DATA_TYPE_STRING, vd->vdev_fru, NULL);
570 if (vd->vdev_enc_sysfs_path != NULL)
571 fm_payload_set(ereport,
572 FM_EREPORT_PAYLOAD_ZFS_VDEV_ENC_SYSFS_PATH,
573 DATA_TYPE_STRING, vd->vdev_enc_sysfs_path, NULL);
575 fm_payload_set(ereport,
576 FM_EREPORT_PAYLOAD_ZFS_VDEV_ASHIFT,
577 DATA_TYPE_UINT64, vd->vdev_ashift, NULL);
580 fm_payload_set(ereport,
581 FM_EREPORT_PAYLOAD_ZFS_VDEV_COMP_TS,
582 DATA_TYPE_UINT64, vq->vq_io_complete_ts, NULL);
583 fm_payload_set(ereport,
584 FM_EREPORT_PAYLOAD_ZFS_VDEV_DELTA_TS,
585 DATA_TYPE_UINT64, vq->vq_io_delta_ts, NULL);
589 fm_payload_set(ereport,
590 FM_EREPORT_PAYLOAD_ZFS_VDEV_READ_ERRORS,
591 DATA_TYPE_UINT64, vs->vs_read_errors,
592 FM_EREPORT_PAYLOAD_ZFS_VDEV_WRITE_ERRORS,
593 DATA_TYPE_UINT64, vs->vs_write_errors,
594 FM_EREPORT_PAYLOAD_ZFS_VDEV_CKSUM_ERRORS,
595 DATA_TYPE_UINT64, vs->vs_checksum_errors,
596 FM_EREPORT_PAYLOAD_ZFS_VDEV_DELAYS,
597 DATA_TYPE_UINT64, vs->vs_slow_ios,
602 fm_payload_set(ereport,
603 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID,
604 DATA_TYPE_UINT64, pvd->vdev_guid,
605 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE,
606 DATA_TYPE_STRING, pvd->vdev_ops->vdev_op_type,
609 fm_payload_set(ereport,
610 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH,
611 DATA_TYPE_STRING, pvd->vdev_path, NULL);
613 fm_payload_set(ereport,
614 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID,
615 DATA_TYPE_STRING, pvd->vdev_devid, NULL);
618 spare_count = spa->spa_spares.sav_count;
619 spare_paths = kmem_zalloc(sizeof (char *) * spare_count,
621 spare_guids = kmem_zalloc(sizeof (uint64_t) * spare_count,
624 for (i = 0; i < spare_count; i++) {
625 spare_vd = spa->spa_spares.sav_vdevs[i];
627 spare_paths[i] = spare_vd->vdev_path;
628 spare_guids[i] = spare_vd->vdev_guid;
632 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_SPARE_PATHS,
633 DATA_TYPE_STRING_ARRAY, spare_count, spare_paths,
634 FM_EREPORT_PAYLOAD_ZFS_VDEV_SPARE_GUIDS,
635 DATA_TYPE_UINT64_ARRAY, spare_count, spare_guids, NULL);
637 kmem_free(spare_guids, sizeof (uint64_t) * spare_count);
638 kmem_free(spare_paths, sizeof (char *) * spare_count);
643 * Payload common to all I/Os.
645 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR,
646 DATA_TYPE_INT32, zio->io_error, NULL);
647 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_FLAGS,
648 DATA_TYPE_INT32, zio->io_flags, NULL);
649 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_STAGE,
650 DATA_TYPE_UINT32, zio->io_stage, NULL);
651 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_PIPELINE,
652 DATA_TYPE_UINT32, zio->io_pipeline, NULL);
653 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELAY,
654 DATA_TYPE_UINT64, zio->io_delay, NULL);
655 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_TIMESTAMP,
656 DATA_TYPE_UINT64, zio->io_timestamp, NULL);
657 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELTA,
658 DATA_TYPE_UINT64, zio->io_delta, NULL);
659 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_PRIORITY,
660 DATA_TYPE_UINT32, zio->io_priority, NULL);
663 * If the 'size' parameter is non-zero, it indicates this is a
664 * RAID-Z or other I/O where the physical offset and length are
665 * provided for us, instead of within the zio_t.
669 fm_payload_set(ereport,
670 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET,
671 DATA_TYPE_UINT64, stateoroffset,
672 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE,
673 DATA_TYPE_UINT64, size, NULL);
675 fm_payload_set(ereport,
676 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET,
677 DATA_TYPE_UINT64, zio->io_offset,
678 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE,
679 DATA_TYPE_UINT64, zio->io_size, NULL);
681 } else if (vd != NULL) {
683 * If we have a vdev but no zio, this is a device fault, and the
684 * 'stateoroffset' parameter indicates the previous state of the
687 fm_payload_set(ereport,
688 FM_EREPORT_PAYLOAD_ZFS_PREV_STATE,
689 DATA_TYPE_UINT64, stateoroffset, NULL);
693 * Payload for I/Os with corresponding logical information.
695 if (zb != NULL && (zio == NULL || zio->io_logical != NULL)) {
696 fm_payload_set(ereport,
697 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET,
698 DATA_TYPE_UINT64, zb->zb_objset,
699 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT,
700 DATA_TYPE_UINT64, zb->zb_object,
701 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL,
702 DATA_TYPE_INT64, zb->zb_level,
703 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID,
704 DATA_TYPE_UINT64, zb->zb_blkid, NULL);
708 * Payload for tuning the zed
710 if (vd != NULL && strcmp(subclass, FM_EREPORT_ZFS_CHECKSUM) == 0) {
711 uint64_t cksum_n, cksum_t;
713 cksum_n = vdev_prop_get_inherited(vd, VDEV_PROP_CHECKSUM_N);
714 if (cksum_n != vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_N))
715 fm_payload_set(ereport,
716 FM_EREPORT_PAYLOAD_ZFS_VDEV_CKSUM_N,
721 cksum_t = vdev_prop_get_inherited(vd, VDEV_PROP_CHECKSUM_T);
722 if (cksum_t != vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_T))
723 fm_payload_set(ereport,
724 FM_EREPORT_PAYLOAD_ZFS_VDEV_CKSUM_T,
730 if (vd != NULL && strcmp(subclass, FM_EREPORT_ZFS_IO) == 0) {
733 io_n = vdev_prop_get_inherited(vd, VDEV_PROP_IO_N);
734 if (io_n != vdev_prop_default_numeric(VDEV_PROP_IO_N))
735 fm_payload_set(ereport,
736 FM_EREPORT_PAYLOAD_ZFS_VDEV_IO_N,
741 io_t = vdev_prop_get_inherited(vd, VDEV_PROP_IO_T);
742 if (io_t != vdev_prop_default_numeric(VDEV_PROP_IO_T))
743 fm_payload_set(ereport,
744 FM_EREPORT_PAYLOAD_ZFS_VDEV_IO_T,
750 if (vd != NULL && strcmp(subclass, FM_EREPORT_ZFS_DELAY) == 0) {
751 uint64_t slow_io_n, slow_io_t;
753 slow_io_n = vdev_prop_get_inherited(vd, VDEV_PROP_SLOW_IO_N);
754 if (slow_io_n != vdev_prop_default_numeric(VDEV_PROP_SLOW_IO_N))
755 fm_payload_set(ereport,
756 FM_EREPORT_PAYLOAD_ZFS_VDEV_SLOW_IO_N,
761 slow_io_t = vdev_prop_get_inherited(vd, VDEV_PROP_SLOW_IO_T);
762 if (slow_io_t != vdev_prop_default_numeric(VDEV_PROP_SLOW_IO_T))
763 fm_payload_set(ereport,
764 FM_EREPORT_PAYLOAD_ZFS_VDEV_SLOW_IO_T,
770 mutex_exit(&spa->spa_errlist_lock);
772 *ereport_out = ereport;
773 *detector_out = detector;
777 /* if it's <= 128 bytes, save the corruption directly */
778 #define ZFM_MAX_INLINE (128 / sizeof (uint64_t))
780 #define MAX_RANGES 16
782 typedef struct zfs_ecksum_info {
783 /* inline arrays of bits set and cleared. */
784 uint64_t zei_bits_set[ZFM_MAX_INLINE];
785 uint64_t zei_bits_cleared[ZFM_MAX_INLINE];
788 * for each range, the number of bits set and cleared. The Hamming
789 * distance between the good and bad buffers is the sum of them all.
791 uint32_t zei_range_sets[MAX_RANGES];
792 uint32_t zei_range_clears[MAX_RANGES];
797 } zei_ranges[MAX_RANGES];
799 size_t zei_range_count;
801 uint32_t zei_allowed_mingap;
806 update_bad_bits(uint64_t value_arg, uint32_t *count)
810 uint64_t value = BE_64(value_arg);
812 /* We store the bits in big-endian (largest-first) order */
813 for (i = 0; i < 64; i++) {
814 if (value & (1ull << i))
817 /* update the count of bits changed */
822 * We've now filled up the range array, and need to increase "mingap" and
823 * shrink the range list accordingly. zei_mingap is always the smallest
824 * distance between array entries, so we set the new_allowed_gap to be
825 * one greater than that. We then go through the list, joining together
826 * any ranges which are closer than the new_allowed_gap.
828 * By construction, there will be at least one. We also update zei_mingap
829 * to the new smallest gap, to prepare for our next invocation.
832 zei_shrink_ranges(zfs_ecksum_info_t *eip)
834 uint32_t mingap = UINT32_MAX;
835 uint32_t new_allowed_gap = eip->zei_mingap + 1;
838 size_t max = eip->zei_range_count;
840 struct zei_ranges *r = eip->zei_ranges;
842 ASSERT3U(eip->zei_range_count, >, 0);
843 ASSERT3U(eip->zei_range_count, <=, MAX_RANGES);
846 while (idx < max - 1) {
847 uint32_t start = r[idx].zr_start;
848 uint32_t end = r[idx].zr_end;
850 while (idx < max - 1) {
853 uint32_t nstart = r[idx].zr_start;
854 uint32_t nend = r[idx].zr_end;
856 uint32_t gap = nstart - end;
857 if (gap < new_allowed_gap) {
865 r[output].zr_start = start;
866 r[output].zr_end = end;
869 ASSERT3U(output, <, eip->zei_range_count);
870 eip->zei_range_count = output;
871 eip->zei_mingap = mingap;
872 eip->zei_allowed_mingap = new_allowed_gap;
876 zei_add_range(zfs_ecksum_info_t *eip, int start, int end)
878 struct zei_ranges *r = eip->zei_ranges;
879 size_t count = eip->zei_range_count;
881 if (count >= MAX_RANGES) {
882 zei_shrink_ranges(eip);
883 count = eip->zei_range_count;
886 eip->zei_mingap = UINT32_MAX;
887 eip->zei_allowed_mingap = 1;
889 int gap = start - r[count - 1].zr_end;
891 if (gap < eip->zei_allowed_mingap) {
892 r[count - 1].zr_end = end;
895 if (gap < eip->zei_mingap)
896 eip->zei_mingap = gap;
898 r[count].zr_start = start;
899 r[count].zr_end = end;
900 eip->zei_range_count++;
904 zei_range_total_size(zfs_ecksum_info_t *eip)
906 struct zei_ranges *r = eip->zei_ranges;
907 size_t count = eip->zei_range_count;
911 for (idx = 0; idx < count; idx++)
912 result += (r[idx].zr_end - r[idx].zr_start);
917 static zfs_ecksum_info_t *
918 annotate_ecksum(nvlist_t *ereport, zio_bad_cksum_t *info,
919 const abd_t *goodabd, const abd_t *badabd, size_t size,
920 boolean_t drop_if_identical)
922 const uint64_t *good;
925 size_t nui64s = size / sizeof (uint64_t);
935 zfs_ecksum_info_t *eip = kmem_zalloc(sizeof (*eip), KM_SLEEP);
937 /* don't do any annotation for injected checksum errors */
938 if (info != NULL && info->zbc_injected)
941 if (info != NULL && info->zbc_has_cksum) {
942 fm_payload_set(ereport,
943 FM_EREPORT_PAYLOAD_ZFS_CKSUM_ALGO,
945 info->zbc_checksum_name,
948 if (info->zbc_byteswapped) {
949 fm_payload_set(ereport,
950 FM_EREPORT_PAYLOAD_ZFS_CKSUM_BYTESWAP,
951 DATA_TYPE_BOOLEAN, 1,
956 if (badabd == NULL || goodabd == NULL)
959 ASSERT3U(nui64s, <=, UINT32_MAX);
960 ASSERT3U(size, ==, nui64s * sizeof (uint64_t));
961 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE);
962 ASSERT3U(size, <=, UINT32_MAX);
964 good = (const uint64_t *) abd_borrow_buf_copy((abd_t *)goodabd, size);
965 bad = (const uint64_t *) abd_borrow_buf_copy((abd_t *)badabd, size);
967 /* build up the range list by comparing the two buffers. */
968 for (idx = 0; idx < nui64s; idx++) {
969 if (good[idx] == bad[idx]) {
973 zei_add_range(eip, start, idx);
983 zei_add_range(eip, start, idx);
985 /* See if it will fit in our inline buffers */
986 inline_size = zei_range_total_size(eip);
987 if (inline_size > ZFM_MAX_INLINE)
991 * If there is no change and we want to drop if the buffers are
994 if (inline_size == 0 && drop_if_identical) {
995 kmem_free(eip, sizeof (*eip));
996 abd_return_buf((abd_t *)goodabd, (void *)good, size);
997 abd_return_buf((abd_t *)badabd, (void *)bad, size);
1002 * Now walk through the ranges, filling in the details of the
1003 * differences. Also convert our uint64_t-array offsets to byte
1006 for (range = 0; range < eip->zei_range_count; range++) {
1007 size_t start = eip->zei_ranges[range].zr_start;
1008 size_t end = eip->zei_ranges[range].zr_end;
1010 for (idx = start; idx < end; idx++) {
1011 uint64_t set, cleared;
1013 // bits set in bad, but not in good
1014 set = ((~good[idx]) & bad[idx]);
1015 // bits set in good, but not in bad
1016 cleared = (good[idx] & (~bad[idx]));
1019 ASSERT3U(offset, <, inline_size);
1020 eip->zei_bits_set[offset] = set;
1021 eip->zei_bits_cleared[offset] = cleared;
1025 update_bad_bits(set, &eip->zei_range_sets[range]);
1026 update_bad_bits(cleared, &eip->zei_range_clears[range]);
1029 /* convert to byte offsets */
1030 eip->zei_ranges[range].zr_start *= sizeof (uint64_t);
1031 eip->zei_ranges[range].zr_end *= sizeof (uint64_t);
1034 abd_return_buf((abd_t *)goodabd, (void *)good, size);
1035 abd_return_buf((abd_t *)badabd, (void *)bad, size);
1037 eip->zei_allowed_mingap *= sizeof (uint64_t);
1038 inline_size *= sizeof (uint64_t);
1040 /* fill in ereport */
1041 fm_payload_set(ereport,
1042 FM_EREPORT_PAYLOAD_ZFS_BAD_OFFSET_RANGES,
1043 DATA_TYPE_UINT32_ARRAY, 2 * eip->zei_range_count,
1044 (uint32_t *)eip->zei_ranges,
1045 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_MIN_GAP,
1046 DATA_TYPE_UINT32, eip->zei_allowed_mingap,
1047 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_SETS,
1048 DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_sets,
1049 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_CLEARS,
1050 DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_clears,
1054 fm_payload_set(ereport,
1055 FM_EREPORT_PAYLOAD_ZFS_BAD_SET_BITS,
1056 DATA_TYPE_UINT8_ARRAY,
1057 inline_size, (uint8_t *)eip->zei_bits_set,
1058 FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_BITS,
1059 DATA_TYPE_UINT8_ARRAY,
1060 inline_size, (uint8_t *)eip->zei_bits_cleared,
1067 zfs_ereport_clear(spa_t *spa, vdev_t *vd)
1069 (void) spa, (void) vd;
1074 * Make sure our event is still valid for the given zio/vdev/pool. For example,
1075 * we don't want to keep logging events for a faulted or missing vdev.
1078 zfs_ereport_is_valid(const char *subclass, spa_t *spa, vdev_t *vd, zio_t *zio)
1082 * If we are doing a spa_tryimport() or in recovery mode,
1085 if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT ||
1086 spa_load_state(spa) == SPA_LOAD_RECOVER)
1090 * If we are in the middle of opening a pool, and the previous attempt
1091 * failed, don't bother logging any new ereports - we're just going to
1092 * get the same diagnosis anyway.
1094 if (spa_load_state(spa) != SPA_LOAD_NONE &&
1095 spa->spa_last_open_failed)
1100 * If this is not a read or write zio, ignore the error. This
1101 * can occur if the DKIOCFLUSHWRITECACHE ioctl fails.
1103 if (zio->io_type != ZIO_TYPE_READ &&
1104 zio->io_type != ZIO_TYPE_WRITE)
1109 * If the vdev has already been marked as failing due
1110 * to a failed probe, then ignore any subsequent I/O
1111 * errors, as the DE will automatically fault the vdev
1112 * on the first such failure. This also catches cases
1113 * where vdev_remove_wanted is set and the device has
1114 * not yet been asynchronously placed into the REMOVED
1117 if (zio->io_vd == vd && !vdev_accessible(vd, zio))
1121 * Ignore checksum errors for reads from DTL regions of
1124 if (zio->io_type == ZIO_TYPE_READ &&
1125 zio->io_error == ECKSUM &&
1126 vd->vdev_ops->vdev_op_leaf &&
1127 vdev_dtl_contains(vd, DTL_MISSING, zio->io_txg, 1))
1133 * For probe failure, we want to avoid posting ereports if we've
1134 * already removed the device in the meantime.
1137 strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) == 0 &&
1138 (vd->vdev_remove_wanted || vd->vdev_state == VDEV_STATE_REMOVED))
1141 /* Ignore bogus delay events (like from ioctls or unqueued IOs) */
1142 if ((strcmp(subclass, FM_EREPORT_ZFS_DELAY) == 0) &&
1143 (zio != NULL) && (!zio->io_timestamp)) {
1147 (void) subclass, (void) spa, (void) vd, (void) zio;
1153 * Post an ereport for the given subclass
1156 * - 0 if an event was posted
1157 * - EINVAL if there was a problem posting event
1158 * - EBUSY if the event was rate limited
1159 * - EALREADY if the event was already posted (duplicate)
1162 zfs_ereport_post(const char *subclass, spa_t *spa, vdev_t *vd,
1163 const zbookmark_phys_t *zb, zio_t *zio, uint64_t state)
1167 nvlist_t *ereport = NULL;
1168 nvlist_t *detector = NULL;
1170 if (!zfs_ereport_is_valid(subclass, spa, vd, zio))
1173 if (zfs_ereport_is_duplicate(subclass, spa, vd, zb, zio, 0, 0))
1174 return (SET_ERROR(EALREADY));
1176 if (zfs_is_ratelimiting_event(subclass, vd))
1177 return (SET_ERROR(EBUSY));
1179 if (!zfs_ereport_start(&ereport, &detector, subclass, spa, vd,
1181 return (SET_ERROR(EINVAL)); /* couldn't post event */
1183 if (ereport == NULL)
1184 return (SET_ERROR(EINVAL));
1186 /* Cleanup is handled by the callback function */
1187 rc = zfs_zevent_post(ereport, detector, zfs_zevent_post_cb);
1189 (void) subclass, (void) spa, (void) vd, (void) zb, (void) zio,
1196 * Prepare a checksum ereport
1199 * - 0 if an event was posted
1200 * - EINVAL if there was a problem posting event
1201 * - EBUSY if the event was rate limited
1202 * - EALREADY if the event was already posted (duplicate)
1205 zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, const zbookmark_phys_t *zb,
1206 struct zio *zio, uint64_t offset, uint64_t length, zio_bad_cksum_t *info)
1208 zio_cksum_report_t *report;
1211 if (!zfs_ereport_is_valid(FM_EREPORT_ZFS_CHECKSUM, spa, vd, zio))
1212 return (SET_ERROR(EINVAL));
1214 if (zfs_ereport_is_duplicate(FM_EREPORT_ZFS_CHECKSUM, spa, vd, zb, zio,
1216 return (SET_ERROR(EALREADY));
1218 if (zfs_is_ratelimiting_event(FM_EREPORT_ZFS_CHECKSUM, vd))
1219 return (SET_ERROR(EBUSY));
1221 (void) zb, (void) offset;
1224 report = kmem_zalloc(sizeof (*report), KM_SLEEP);
1226 zio_vsd_default_cksum_report(zio, report);
1228 /* copy the checksum failure information if it was provided */
1230 report->zcr_ckinfo = kmem_zalloc(sizeof (*info), KM_SLEEP);
1231 memcpy(report->zcr_ckinfo, info, sizeof (*info));
1234 report->zcr_sector = 1ULL << vd->vdev_top->vdev_ashift;
1236 vdev_psize_to_asize(vd->vdev_top, report->zcr_sector);
1237 report->zcr_length = length;
1240 (void) zfs_ereport_start(&report->zcr_ereport, &report->zcr_detector,
1241 FM_EREPORT_ZFS_CHECKSUM, spa, vd, zb, zio, offset, length);
1243 if (report->zcr_ereport == NULL) {
1244 zfs_ereport_free_checksum(report);
1249 mutex_enter(&spa->spa_errlist_lock);
1250 report->zcr_next = zio->io_logical->io_cksum_report;
1251 zio->io_logical->io_cksum_report = report;
1252 mutex_exit(&spa->spa_errlist_lock);
1257 zfs_ereport_finish_checksum(zio_cksum_report_t *report, const abd_t *good_data,
1258 const abd_t *bad_data, boolean_t drop_if_identical)
1261 zfs_ecksum_info_t *info;
1263 info = annotate_ecksum(report->zcr_ereport, report->zcr_ckinfo,
1264 good_data, bad_data, report->zcr_length, drop_if_identical);
1266 zfs_zevent_post(report->zcr_ereport,
1267 report->zcr_detector, zfs_zevent_post_cb);
1269 zfs_zevent_post_cb(report->zcr_ereport, report->zcr_detector);
1271 report->zcr_ereport = report->zcr_detector = NULL;
1273 kmem_free(info, sizeof (*info));
1275 (void) report, (void) good_data, (void) bad_data,
1276 (void) drop_if_identical;
1281 zfs_ereport_free_checksum(zio_cksum_report_t *rpt)
1284 if (rpt->zcr_ereport != NULL) {
1285 fm_nvlist_destroy(rpt->zcr_ereport,
1287 fm_nvlist_destroy(rpt->zcr_detector,
1291 rpt->zcr_free(rpt->zcr_cbdata, rpt->zcr_cbinfo);
1293 if (rpt->zcr_ckinfo != NULL)
1294 kmem_free(rpt->zcr_ckinfo, sizeof (*rpt->zcr_ckinfo));
1296 kmem_free(rpt, sizeof (*rpt));
1300 * Post a checksum ereport
1303 * - 0 if an event was posted
1304 * - EINVAL if there was a problem posting event
1305 * - EBUSY if the event was rate limited
1306 * - EALREADY if the event was already posted (duplicate)
1309 zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd, const zbookmark_phys_t *zb,
1310 struct zio *zio, uint64_t offset, uint64_t length,
1311 const abd_t *good_data, const abd_t *bad_data, zio_bad_cksum_t *zbc)
1315 nvlist_t *ereport = NULL;
1316 nvlist_t *detector = NULL;
1317 zfs_ecksum_info_t *info;
1319 if (!zfs_ereport_is_valid(FM_EREPORT_ZFS_CHECKSUM, spa, vd, zio))
1320 return (SET_ERROR(EINVAL));
1322 if (zfs_ereport_is_duplicate(FM_EREPORT_ZFS_CHECKSUM, spa, vd, zb, zio,
1324 return (SET_ERROR(EALREADY));
1326 if (zfs_is_ratelimiting_event(FM_EREPORT_ZFS_CHECKSUM, vd))
1327 return (SET_ERROR(EBUSY));
1329 if (!zfs_ereport_start(&ereport, &detector, FM_EREPORT_ZFS_CHECKSUM,
1330 spa, vd, zb, zio, offset, length) || (ereport == NULL)) {
1331 return (SET_ERROR(EINVAL));
1334 info = annotate_ecksum(ereport, zbc, good_data, bad_data, length,
1338 rc = zfs_zevent_post(ereport, detector, zfs_zevent_post_cb);
1339 kmem_free(info, sizeof (*info));
1342 (void) spa, (void) vd, (void) zb, (void) zio, (void) offset,
1343 (void) length, (void) good_data, (void) bad_data, (void) zbc;
1349 * The 'sysevent.fs.zfs.*' events are signals posted to notify user space of
1350 * change in the pool. All sysevents are listed in sys/sysevent/eventdefs.h
1351 * and are designed to be consumed by the ZFS Event Daemon (ZED). For
1352 * additional details refer to the zed(8) man page.
1355 zfs_event_create(spa_t *spa, vdev_t *vd, const char *type, const char *name,
1358 nvlist_t *resource = NULL;
1362 if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT)
1365 if ((resource = fm_nvlist_create(NULL)) == NULL)
1368 (void) snprintf(class, sizeof (class), "%s.%s.%s", type,
1369 ZFS_ERROR_CLASS, name);
1370 VERIFY0(nvlist_add_uint8(resource, FM_VERSION, FM_RSRC_VERSION));
1371 VERIFY0(nvlist_add_string(resource, FM_CLASS, class));
1372 VERIFY0(nvlist_add_string(resource,
1373 FM_EREPORT_PAYLOAD_ZFS_POOL, spa_name(spa)));
1374 VERIFY0(nvlist_add_uint64(resource,
1375 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, spa_guid(spa)));
1376 VERIFY0(nvlist_add_uint64(resource,
1377 FM_EREPORT_PAYLOAD_ZFS_POOL_STATE, spa_state(spa)));
1378 VERIFY0(nvlist_add_int32(resource,
1379 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, spa_load_state(spa)));
1382 VERIFY0(nvlist_add_uint64(resource,
1383 FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vd->vdev_guid));
1384 VERIFY0(nvlist_add_uint64(resource,
1385 FM_EREPORT_PAYLOAD_ZFS_VDEV_STATE, vd->vdev_state));
1386 if (vd->vdev_path != NULL)
1387 VERIFY0(nvlist_add_string(resource,
1388 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH, vd->vdev_path));
1389 if (vd->vdev_devid != NULL)
1390 VERIFY0(nvlist_add_string(resource,
1391 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID, vd->vdev_devid));
1392 if (vd->vdev_fru != NULL)
1393 VERIFY0(nvlist_add_string(resource,
1394 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU, vd->vdev_fru));
1395 if (vd->vdev_enc_sysfs_path != NULL)
1396 VERIFY0(nvlist_add_string(resource,
1397 FM_EREPORT_PAYLOAD_ZFS_VDEV_ENC_SYSFS_PATH,
1398 vd->vdev_enc_sysfs_path));
1401 /* also copy any optional payload data */
1403 nvpair_t *elem = NULL;
1405 while ((elem = nvlist_next_nvpair(aux, elem)) != NULL)
1406 (void) nvlist_add_nvpair(resource, elem);
1409 (void) spa, (void) vd, (void) type, (void) name, (void) aux;
1415 zfs_post_common(spa_t *spa, vdev_t *vd, const char *type, const char *name,
1421 resource = zfs_event_create(spa, vd, type, name, aux);
1423 zfs_zevent_post(resource, NULL, zfs_zevent_post_cb);
1425 (void) spa, (void) vd, (void) type, (void) name, (void) aux;
1430 * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev
1431 * has been removed from the system. This will cause the DE to ignore any
1432 * recent I/O errors, inferring that they are due to the asynchronous device
1436 zfs_post_remove(spa_t *spa, vdev_t *vd)
1438 zfs_post_common(spa, vd, FM_RSRC_CLASS, FM_RESOURCE_REMOVED, NULL);
1442 * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool
1443 * has the 'autoreplace' property set, and therefore any broken vdevs will be
1444 * handled by higher level logic, and no vdev fault should be generated.
1447 zfs_post_autoreplace(spa_t *spa, vdev_t *vd)
1449 zfs_post_common(spa, vd, FM_RSRC_CLASS, FM_RESOURCE_AUTOREPLACE, NULL);
1453 * The 'resource.fs.zfs.statechange' event is an internal signal that the
1454 * given vdev has transitioned its state to DEGRADED or HEALTHY. This will
1455 * cause the retire agent to repair any outstanding fault management cases
1456 * open because the device was not found (fault.fs.zfs.device).
1459 zfs_post_state_change(spa_t *spa, vdev_t *vd, uint64_t laststate)
1465 * Add optional supplemental keys to payload
1467 aux = fm_nvlist_create(NULL);
1469 if (vd->vdev_physpath) {
1470 fnvlist_add_string(aux,
1471 FM_EREPORT_PAYLOAD_ZFS_VDEV_PHYSPATH,
1474 if (vd->vdev_enc_sysfs_path) {
1475 fnvlist_add_string(aux,
1476 FM_EREPORT_PAYLOAD_ZFS_VDEV_ENC_SYSFS_PATH,
1477 vd->vdev_enc_sysfs_path);
1480 fnvlist_add_uint64(aux,
1481 FM_EREPORT_PAYLOAD_ZFS_VDEV_LASTSTATE, laststate);
1484 zfs_post_common(spa, vd, FM_RSRC_CLASS, FM_RESOURCE_STATECHANGE,
1488 fm_nvlist_destroy(aux, FM_NVA_FREE);
1490 (void) spa, (void) vd, (void) laststate;
1496 zfs_ereport_init(void)
1498 mutex_init(&recent_events_lock, NULL, MUTEX_DEFAULT, NULL);
1499 list_create(&recent_events_list, sizeof (recent_events_node_t),
1500 offsetof(recent_events_node_t, re_list_link));
1501 avl_create(&recent_events_tree, recent_events_compare,
1502 sizeof (recent_events_node_t), offsetof(recent_events_node_t,
1507 * This 'early' fini needs to run before zfs_fini() which on Linux waits
1508 * for the system_delay_taskq to drain.
1511 zfs_ereport_taskq_fini(void)
1513 mutex_enter(&recent_events_lock);
1514 if (recent_events_cleaner_tqid != 0) {
1515 taskq_cancel_id(system_delay_taskq, recent_events_cleaner_tqid);
1516 recent_events_cleaner_tqid = 0;
1518 mutex_exit(&recent_events_lock);
1522 zfs_ereport_fini(void)
1524 recent_events_node_t *entry;
1526 while ((entry = list_remove_head(&recent_events_list)) != NULL) {
1527 avl_remove(&recent_events_tree, entry);
1528 kmem_free(entry, sizeof (*entry));
1530 avl_destroy(&recent_events_tree);
1531 list_destroy(&recent_events_list);
1532 mutex_destroy(&recent_events_lock);
1536 zfs_ereport_snapshot_post(const char *subclass, spa_t *spa, const char *name)
1540 aux = fm_nvlist_create(NULL);
1541 fnvlist_add_string(aux, FM_EREPORT_PAYLOAD_ZFS_SNAPSHOT_NAME, name);
1543 zfs_post_common(spa, NULL, FM_RSRC_CLASS, subclass, aux);
1544 fm_nvlist_destroy(aux, FM_NVA_FREE);
1548 * Post when a event when a zvol is created or removed
1550 * This is currently only used by macOS, since it uses the event to create
1551 * symlinks between the volume name (mypool/myvol) and the actual /dev
1552 * device (/dev/disk3). For example:
1554 * /var/run/zfs/dsk/mypool/myvol -> /dev/disk3
1556 * name: The full name of the zvol ("mypool/myvol")
1557 * dev_name: The full /dev name for the zvol ("/dev/disk3")
1558 * raw_name: The raw /dev name for the zvol ("/dev/rdisk3")
1561 zfs_ereport_zvol_post(const char *subclass, const char *name,
1562 const char *dev_name, const char *raw_name)
1567 boolean_t locked = mutex_owned(&spa_namespace_lock);
1568 if (!locked) mutex_enter(&spa_namespace_lock);
1569 spa_t *spa = spa_lookup(name);
1570 if (!locked) mutex_exit(&spa_namespace_lock);
1575 aux = fm_nvlist_create(NULL);
1576 fnvlist_add_string(aux, FM_EREPORT_PAYLOAD_ZFS_DEVICE_NAME, dev_name);
1577 fnvlist_add_string(aux, FM_EREPORT_PAYLOAD_ZFS_RAW_DEVICE_NAME,
1579 r = strchr(name, '/');
1581 fnvlist_add_string(aux, FM_EREPORT_PAYLOAD_ZFS_VOLUME, &r[1]);
1583 zfs_post_common(spa, NULL, FM_RSRC_CLASS, subclass, aux);
1584 fm_nvlist_destroy(aux, FM_NVA_FREE);
1587 EXPORT_SYMBOL(zfs_ereport_post);
1588 EXPORT_SYMBOL(zfs_ereport_is_valid);
1589 EXPORT_SYMBOL(zfs_ereport_post_checksum);
1590 EXPORT_SYMBOL(zfs_post_remove);
1591 EXPORT_SYMBOL(zfs_post_autoreplace);
1592 EXPORT_SYMBOL(zfs_post_state_change);
1594 ZFS_MODULE_PARAM(zfs_zevent, zfs_zevent_, retain_max, UINT, ZMOD_RW,
1595 "Maximum recent zevents records to retain for duplicate checking");
1596 ZFS_MODULE_PARAM(zfs_zevent, zfs_zevent_, retain_expire_secs, UINT, ZMOD_RW,
1597 "Expiration time for recent zevents records");
1598 #endif /* _KERNEL */