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 http://www.opensolaris.org/os/licensing.
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 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.
106 zfs_zevent_post_cb(nvlist_t *nvl, nvlist_t *detector)
109 fm_nvlist_destroy(nvl, FM_NVA_FREE);
112 fm_nvlist_destroy(detector, FM_NVA_FREE);
116 zfs_ereport_start(nvlist_t **ereport_out, nvlist_t **detector_out,
117 const char *subclass, spa_t *spa, vdev_t *vd, zio_t *zio,
118 uint64_t stateoroffset, uint64_t size)
120 nvlist_t *ereport, *detector;
126 * If we are doing a spa_tryimport() or in recovery mode,
129 if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT ||
130 spa_load_state(spa) == SPA_LOAD_RECOVER)
134 * If we are in the middle of opening a pool, and the previous attempt
135 * failed, don't bother logging any new ereports - we're just going to
136 * get the same diagnosis anyway.
138 if (spa_load_state(spa) != SPA_LOAD_NONE &&
139 spa->spa_last_open_failed)
144 * If this is not a read or write zio, ignore the error. This
145 * can occur if the DKIOCFLUSHWRITECACHE ioctl fails.
147 if (zio->io_type != ZIO_TYPE_READ &&
148 zio->io_type != ZIO_TYPE_WRITE)
153 * If the vdev has already been marked as failing due
154 * to a failed probe, then ignore any subsequent I/O
155 * errors, as the DE will automatically fault the vdev
156 * on the first such failure. This also catches cases
157 * where vdev_remove_wanted is set and the device has
158 * not yet been asynchronously placed into the REMOVED
161 if (zio->io_vd == vd && !vdev_accessible(vd, zio))
165 * Ignore checksum errors for reads from DTL regions of
168 if (zio->io_type == ZIO_TYPE_READ &&
169 zio->io_error == ECKSUM &&
170 vd->vdev_ops->vdev_op_leaf &&
171 vdev_dtl_contains(vd, DTL_MISSING, zio->io_txg, 1))
177 * For probe failure, we want to avoid posting ereports if we've
178 * already removed the device in the meantime.
181 strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) == 0 &&
182 (vd->vdev_remove_wanted || vd->vdev_state == VDEV_STATE_REMOVED))
185 if ((ereport = fm_nvlist_create(NULL)) == NULL)
188 if ((detector = fm_nvlist_create(NULL)) == NULL) {
189 fm_nvlist_destroy(ereport, FM_NVA_FREE);
194 * Serialize ereport generation
196 mutex_enter(&spa->spa_errlist_lock);
199 * Determine the ENA to use for this event. If we are in a loading
200 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use
201 * a root zio-wide ENA. Otherwise, simply use a unique ENA.
203 if (spa_load_state(spa) != SPA_LOAD_NONE) {
204 if (spa->spa_ena == 0)
205 spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1);
207 } else if (zio != NULL && zio->io_logical != NULL) {
208 if (zio->io_logical->io_ena == 0)
209 zio->io_logical->io_ena =
210 fm_ena_generate(0, FM_ENA_FMT1);
211 ena = zio->io_logical->io_ena;
213 ena = fm_ena_generate(0, FM_ENA_FMT1);
217 * Construct the full class, detector, and other standard FMA fields.
219 (void) snprintf(class, sizeof (class), "%s.%s",
220 ZFS_ERROR_CLASS, subclass);
222 fm_fmri_zfs_set(detector, FM_ZFS_SCHEME_VERSION, spa_guid(spa),
223 vd != NULL ? vd->vdev_guid : 0);
225 fm_ereport_set(ereport, FM_EREPORT_VERSION, class, ena, detector, NULL);
228 * Construct the per-ereport payload, depending on which parameters are
233 * Generic payload members common to all ereports.
235 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL,
236 DATA_TYPE_STRING, spa_name(spa), FM_EREPORT_PAYLOAD_ZFS_POOL_GUID,
237 DATA_TYPE_UINT64, spa_guid(spa),
238 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, DATA_TYPE_INT32,
239 spa_load_state(spa), NULL);
242 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE,
244 spa_get_failmode(spa) == ZIO_FAILURE_MODE_WAIT ?
245 FM_EREPORT_FAILMODE_WAIT :
246 spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE ?
247 FM_EREPORT_FAILMODE_CONTINUE : FM_EREPORT_FAILMODE_PANIC,
252 vdev_t *pvd = vd->vdev_parent;
254 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID,
255 DATA_TYPE_UINT64, vd->vdev_guid,
256 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE,
257 DATA_TYPE_STRING, vd->vdev_ops->vdev_op_type, NULL);
258 if (vd->vdev_path != NULL)
259 fm_payload_set(ereport,
260 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH,
261 DATA_TYPE_STRING, vd->vdev_path, NULL);
262 if (vd->vdev_devid != NULL)
263 fm_payload_set(ereport,
264 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID,
265 DATA_TYPE_STRING, vd->vdev_devid, NULL);
266 if (vd->vdev_fru != NULL)
267 fm_payload_set(ereport,
268 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU,
269 DATA_TYPE_STRING, vd->vdev_fru, NULL);
272 fm_payload_set(ereport,
273 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID,
274 DATA_TYPE_UINT64, pvd->vdev_guid,
275 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE,
276 DATA_TYPE_STRING, pvd->vdev_ops->vdev_op_type,
279 fm_payload_set(ereport,
280 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH,
281 DATA_TYPE_STRING, pvd->vdev_path, NULL);
283 fm_payload_set(ereport,
284 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID,
285 DATA_TYPE_STRING, pvd->vdev_devid, NULL);
291 * Payload common to all I/Os.
293 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR,
294 DATA_TYPE_INT32, zio->io_error, NULL);
295 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_FLAGS,
296 DATA_TYPE_INT32, zio->io_flags, NULL);
297 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_STAGE,
298 DATA_TYPE_UINT32, zio->io_stage, NULL);
299 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_PIPELINE,
300 DATA_TYPE_UINT32, zio->io_pipeline, NULL);
301 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELAY,
302 DATA_TYPE_UINT64, zio->io_delay, NULL);
305 * If the 'size' parameter is non-zero, it indicates this is a
306 * RAID-Z or other I/O where the physical offset and length are
307 * provided for us, instead of within the zio_t.
311 fm_payload_set(ereport,
312 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET,
313 DATA_TYPE_UINT64, stateoroffset,
314 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE,
315 DATA_TYPE_UINT64, size, NULL);
317 fm_payload_set(ereport,
318 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET,
319 DATA_TYPE_UINT64, zio->io_offset,
320 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE,
321 DATA_TYPE_UINT64, zio->io_size, NULL);
325 * Payload for I/Os with corresponding logical information.
327 if (zio->io_logical != NULL)
328 fm_payload_set(ereport,
329 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET,
331 zio->io_logical->io_bookmark.zb_objset,
332 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT,
334 zio->io_logical->io_bookmark.zb_object,
335 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL,
337 zio->io_logical->io_bookmark.zb_level,
338 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID,
340 zio->io_logical->io_bookmark.zb_blkid, NULL);
341 } else if (vd != NULL) {
343 * If we have a vdev but no zio, this is a device fault, and the
344 * 'stateoroffset' parameter indicates the previous state of the
347 fm_payload_set(ereport,
348 FM_EREPORT_PAYLOAD_ZFS_PREV_STATE,
349 DATA_TYPE_UINT64, stateoroffset, NULL);
352 mutex_exit(&spa->spa_errlist_lock);
354 *ereport_out = ereport;
355 *detector_out = detector;
358 /* if it's <= 128 bytes, save the corruption directly */
359 #define ZFM_MAX_INLINE (128 / sizeof (uint64_t))
361 #define MAX_RANGES 16
363 typedef struct zfs_ecksum_info {
364 /* histograms of set and cleared bits by bit number in a 64-bit word */
365 uint16_t zei_histogram_set[sizeof (uint64_t) * NBBY];
366 uint16_t zei_histogram_cleared[sizeof (uint64_t) * NBBY];
368 /* inline arrays of bits set and cleared. */
369 uint64_t zei_bits_set[ZFM_MAX_INLINE];
370 uint64_t zei_bits_cleared[ZFM_MAX_INLINE];
373 * for each range, the number of bits set and cleared. The Hamming
374 * distance between the good and bad buffers is the sum of them all.
376 uint32_t zei_range_sets[MAX_RANGES];
377 uint32_t zei_range_clears[MAX_RANGES];
382 } zei_ranges[MAX_RANGES];
384 size_t zei_range_count;
386 uint32_t zei_allowed_mingap;
391 update_histogram(uint64_t value_arg, uint16_t *hist, uint32_t *count)
395 uint64_t value = BE_64(value_arg);
397 /* We store the bits in big-endian (largest-first) order */
398 for (i = 0; i < 64; i++) {
399 if (value & (1ull << i)) {
404 /* update the count of bits changed */
409 * We've now filled up the range array, and need to increase "mingap" and
410 * shrink the range list accordingly. zei_mingap is always the smallest
411 * distance between array entries, so we set the new_allowed_gap to be
412 * one greater than that. We then go through the list, joining together
413 * any ranges which are closer than the new_allowed_gap.
415 * By construction, there will be at least one. We also update zei_mingap
416 * to the new smallest gap, to prepare for our next invocation.
419 zei_shrink_ranges(zfs_ecksum_info_t *eip)
421 uint32_t mingap = UINT32_MAX;
422 uint32_t new_allowed_gap = eip->zei_mingap + 1;
425 size_t max = eip->zei_range_count;
427 struct zei_ranges *r = eip->zei_ranges;
429 ASSERT3U(eip->zei_range_count, >, 0);
430 ASSERT3U(eip->zei_range_count, <=, MAX_RANGES);
433 while (idx < max - 1) {
434 uint32_t start = r[idx].zr_start;
435 uint32_t end = r[idx].zr_end;
437 while (idx < max - 1) {
438 uint32_t nstart, nend, gap;
441 nstart = r[idx].zr_start;
442 nend = r[idx].zr_end;
445 if (gap < new_allowed_gap) {
453 r[output].zr_start = start;
454 r[output].zr_end = end;
457 ASSERT3U(output, <, eip->zei_range_count);
458 eip->zei_range_count = output;
459 eip->zei_mingap = mingap;
460 eip->zei_allowed_mingap = new_allowed_gap;
464 zei_add_range(zfs_ecksum_info_t *eip, int start, int end)
466 struct zei_ranges *r = eip->zei_ranges;
467 size_t count = eip->zei_range_count;
469 if (count >= MAX_RANGES) {
470 zei_shrink_ranges(eip);
471 count = eip->zei_range_count;
474 eip->zei_mingap = UINT32_MAX;
475 eip->zei_allowed_mingap = 1;
477 int gap = start - r[count - 1].zr_end;
479 if (gap < eip->zei_allowed_mingap) {
480 r[count - 1].zr_end = end;
483 if (gap < eip->zei_mingap)
484 eip->zei_mingap = gap;
486 r[count].zr_start = start;
487 r[count].zr_end = end;
488 eip->zei_range_count++;
492 zei_range_total_size(zfs_ecksum_info_t *eip)
494 struct zei_ranges *r = eip->zei_ranges;
495 size_t count = eip->zei_range_count;
499 for (idx = 0; idx < count; idx++)
500 result += (r[idx].zr_end - r[idx].zr_start);
505 static zfs_ecksum_info_t *
506 annotate_ecksum(nvlist_t *ereport, zio_bad_cksum_t *info,
507 const uint8_t *goodbuf, const uint8_t *badbuf, size_t size,
508 boolean_t drop_if_identical)
510 const uint64_t *good = (const uint64_t *)goodbuf;
511 const uint64_t *bad = (const uint64_t *)badbuf;
514 uint64_t allcleared = 0;
516 size_t nui64s = size / sizeof (uint64_t);
526 zfs_ecksum_info_t *eip = kmem_zalloc(sizeof (*eip), KM_PUSHPAGE);
528 /* don't do any annotation for injected checksum errors */
529 if (info != NULL && info->zbc_injected)
532 if (info != NULL && info->zbc_has_cksum) {
533 fm_payload_set(ereport,
534 FM_EREPORT_PAYLOAD_ZFS_CKSUM_EXPECTED,
535 DATA_TYPE_UINT64_ARRAY,
536 sizeof (info->zbc_expected) / sizeof (uint64_t),
537 (uint64_t *)&info->zbc_expected,
538 FM_EREPORT_PAYLOAD_ZFS_CKSUM_ACTUAL,
539 DATA_TYPE_UINT64_ARRAY,
540 sizeof (info->zbc_actual) / sizeof (uint64_t),
541 (uint64_t *)&info->zbc_actual,
542 FM_EREPORT_PAYLOAD_ZFS_CKSUM_ALGO,
544 info->zbc_checksum_name,
547 if (info->zbc_byteswapped) {
548 fm_payload_set(ereport,
549 FM_EREPORT_PAYLOAD_ZFS_CKSUM_BYTESWAP,
550 DATA_TYPE_BOOLEAN, 1,
555 if (badbuf == NULL || goodbuf == NULL)
558 ASSERT3U(nui64s, <=, UINT16_MAX);
559 ASSERT3U(size, ==, nui64s * sizeof (uint64_t));
560 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE);
561 ASSERT3U(size, <=, UINT32_MAX);
563 /* build up the range list by comparing the two buffers. */
564 for (idx = 0; idx < nui64s; idx++) {
565 if (good[idx] == bad[idx]) {
569 zei_add_range(eip, start, idx);
579 zei_add_range(eip, start, idx);
581 /* See if it will fit in our inline buffers */
582 inline_size = zei_range_total_size(eip);
583 if (inline_size > ZFM_MAX_INLINE)
587 * If there is no change and we want to drop if the buffers are
590 if (inline_size == 0 && drop_if_identical) {
591 kmem_free(eip, sizeof (*eip));
596 * Now walk through the ranges, filling in the details of the
597 * differences. Also convert our uint64_t-array offsets to byte
600 for (range = 0; range < eip->zei_range_count; range++) {
601 size_t start = eip->zei_ranges[range].zr_start;
602 size_t end = eip->zei_ranges[range].zr_end;
604 for (idx = start; idx < end; idx++) {
605 uint64_t set, cleared;
607 // bits set in bad, but not in good
608 set = ((~good[idx]) & bad[idx]);
609 // bits set in good, but not in bad
610 cleared = (good[idx] & (~bad[idx]));
613 allcleared |= cleared;
616 ASSERT3U(offset, <, inline_size);
617 eip->zei_bits_set[offset] = set;
618 eip->zei_bits_cleared[offset] = cleared;
622 update_histogram(set, eip->zei_histogram_set,
623 &eip->zei_range_sets[range]);
624 update_histogram(cleared, eip->zei_histogram_cleared,
625 &eip->zei_range_clears[range]);
628 /* convert to byte offsets */
629 eip->zei_ranges[range].zr_start *= sizeof (uint64_t);
630 eip->zei_ranges[range].zr_end *= sizeof (uint64_t);
632 eip->zei_allowed_mingap *= sizeof (uint64_t);
633 inline_size *= sizeof (uint64_t);
635 /* fill in ereport */
636 fm_payload_set(ereport,
637 FM_EREPORT_PAYLOAD_ZFS_BAD_OFFSET_RANGES,
638 DATA_TYPE_UINT32_ARRAY, 2 * eip->zei_range_count,
639 (uint32_t *)eip->zei_ranges,
640 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_MIN_GAP,
641 DATA_TYPE_UINT32, eip->zei_allowed_mingap,
642 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_SETS,
643 DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_sets,
644 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_CLEARS,
645 DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_clears,
649 fm_payload_set(ereport,
650 FM_EREPORT_PAYLOAD_ZFS_BAD_SET_BITS,
651 DATA_TYPE_UINT8_ARRAY,
652 inline_size, (uint8_t *)eip->zei_bits_set,
653 FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_BITS,
654 DATA_TYPE_UINT8_ARRAY,
655 inline_size, (uint8_t *)eip->zei_bits_cleared,
658 fm_payload_set(ereport,
659 FM_EREPORT_PAYLOAD_ZFS_BAD_SET_HISTOGRAM,
660 DATA_TYPE_UINT16_ARRAY,
661 NBBY * sizeof (uint64_t), eip->zei_histogram_set,
662 FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_HISTOGRAM,
663 DATA_TYPE_UINT16_ARRAY,
664 NBBY * sizeof (uint64_t), eip->zei_histogram_cleared,
672 zfs_ereport_post(const char *subclass, spa_t *spa, vdev_t *vd, zio_t *zio,
673 uint64_t stateoroffset, uint64_t size)
676 nvlist_t *ereport = NULL;
677 nvlist_t *detector = NULL;
679 zfs_ereport_start(&ereport, &detector,
680 subclass, spa, vd, zio, stateoroffset, size);
685 /* Cleanup is handled by the callback function */
686 zfs_zevent_post(ereport, detector, zfs_zevent_post_cb);
691 zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd,
692 struct zio *zio, uint64_t offset, uint64_t length, void *arg,
693 zio_bad_cksum_t *info)
695 zio_cksum_report_t *report = kmem_zalloc(sizeof (*report), KM_PUSHPAGE);
697 if (zio->io_vsd != NULL)
698 zio->io_vsd_ops->vsd_cksum_report(zio, report, arg);
700 zio_vsd_default_cksum_report(zio, report, arg);
702 /* copy the checksum failure information if it was provided */
704 report->zcr_ckinfo = kmem_zalloc(sizeof (*info), KM_PUSHPAGE);
705 bcopy(info, report->zcr_ckinfo, sizeof (*info));
708 report->zcr_align = 1ULL << vd->vdev_top->vdev_ashift;
709 report->zcr_length = length;
712 zfs_ereport_start(&report->zcr_ereport, &report->zcr_detector,
713 FM_EREPORT_ZFS_CHECKSUM, spa, vd, zio, offset, length);
715 if (report->zcr_ereport == NULL) {
716 report->zcr_free(report->zcr_cbdata, report->zcr_cbinfo);
717 if (report->zcr_ckinfo != NULL) {
718 kmem_free(report->zcr_ckinfo,
719 sizeof (*report->zcr_ckinfo));
721 kmem_free(report, sizeof (*report));
726 mutex_enter(&spa->spa_errlist_lock);
727 report->zcr_next = zio->io_logical->io_cksum_report;
728 zio->io_logical->io_cksum_report = report;
729 mutex_exit(&spa->spa_errlist_lock);
733 zfs_ereport_finish_checksum(zio_cksum_report_t *report,
734 const void *good_data, const void *bad_data, boolean_t drop_if_identical)
737 zfs_ecksum_info_t *info = NULL;
738 info = annotate_ecksum(report->zcr_ereport, report->zcr_ckinfo,
739 good_data, bad_data, report->zcr_length, drop_if_identical);
742 zfs_zevent_post(report->zcr_ereport,
743 report->zcr_detector, zfs_zevent_post_cb);
745 report->zcr_ereport = report->zcr_detector = NULL;
747 kmem_free(info, sizeof (*info));
752 zfs_ereport_free_checksum(zio_cksum_report_t *rpt)
755 if (rpt->zcr_ereport != NULL) {
756 fm_nvlist_destroy(rpt->zcr_ereport,
758 fm_nvlist_destroy(rpt->zcr_detector,
762 rpt->zcr_free(rpt->zcr_cbdata, rpt->zcr_cbinfo);
764 if (rpt->zcr_ckinfo != NULL)
765 kmem_free(rpt->zcr_ckinfo, sizeof (*rpt->zcr_ckinfo));
767 kmem_free(rpt, sizeof (*rpt));
771 zfs_ereport_send_interim_checksum(zio_cksum_report_t *report)
774 zfs_zevent_post(report->zcr_ereport, report->zcr_detector, NULL);
779 zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
780 struct zio *zio, uint64_t offset, uint64_t length,
781 const void *good_data, const void *bad_data, zio_bad_cksum_t *zbc)
784 nvlist_t *ereport = NULL;
785 nvlist_t *detector = NULL;
786 zfs_ecksum_info_t *info;
788 zfs_ereport_start(&ereport, &detector,
789 FM_EREPORT_ZFS_CHECKSUM, spa, vd, zio, offset, length);
794 info = annotate_ecksum(ereport, zbc, good_data, bad_data, length,
798 zfs_zevent_post(ereport, detector, zfs_zevent_post_cb);
799 kmem_free(info, sizeof (*info));
805 zfs_post_common(spa_t *spa, vdev_t *vd, const char *name)
811 if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT)
814 if ((resource = fm_nvlist_create(NULL)) == NULL)
817 (void) snprintf(class, sizeof (class), "%s.%s.%s", FM_RSRC_RESOURCE,
818 ZFS_ERROR_CLASS, name);
819 VERIFY(nvlist_add_uint8(resource, FM_VERSION, FM_RSRC_VERSION) == 0);
820 VERIFY(nvlist_add_string(resource, FM_CLASS, class) == 0);
821 VERIFY(nvlist_add_uint64(resource,
822 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, spa_guid(spa)) == 0);
824 VERIFY(nvlist_add_uint64(resource,
825 FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vd->vdev_guid) == 0);
826 VERIFY(nvlist_add_uint64(resource,
827 FM_EREPORT_PAYLOAD_ZFS_VDEV_STATE, vd->vdev_state) == 0);
830 zfs_zevent_post(resource, NULL, zfs_zevent_post_cb);
835 * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev
836 * has been removed from the system. This will cause the DE to ignore any
837 * recent I/O errors, inferring that they are due to the asynchronous device
841 zfs_post_remove(spa_t *spa, vdev_t *vd)
843 zfs_post_common(spa, vd, FM_EREPORT_RESOURCE_REMOVED);
847 * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool
848 * has the 'autoreplace' property set, and therefore any broken vdevs will be
849 * handled by higher level logic, and no vdev fault should be generated.
852 zfs_post_autoreplace(spa_t *spa, vdev_t *vd)
854 zfs_post_common(spa, vd, FM_EREPORT_RESOURCE_AUTOREPLACE);
858 * The 'resource.fs.zfs.statechange' event is an internal signal that the
859 * given vdev has transitioned its state to DEGRADED or HEALTHY. This will
860 * cause the retire agent to repair any outstanding fault management cases
861 * open because the device was not found (fault.fs.zfs.device).
864 zfs_post_state_change(spa_t *spa, vdev_t *vd)
866 zfs_post_common(spa, vd, FM_EREPORT_RESOURCE_STATECHANGE);
869 #if defined(_KERNEL) && defined(HAVE_SPL)
870 EXPORT_SYMBOL(zfs_ereport_post);
871 EXPORT_SYMBOL(zfs_ereport_post_checksum);
872 EXPORT_SYMBOL(zfs_post_remove);
873 EXPORT_SYMBOL(zfs_post_autoreplace);
874 EXPORT_SYMBOL(zfs_post_state_change);