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 * We want to rate limit ZIO delay and checksum events so as to not
117 * flood ZED when a disk is acting up.
119 * Returns 1 if we're ratelimiting, 0 if not.
122 zfs_is_ratelimiting_event(const char *subclass
, vdev_t
*vd
)
126 * __ratelimit() returns 1 if we're *not* ratelimiting and 0 if we
127 * are. Invert it to get our return value.
129 if (strcmp(subclass
, FM_EREPORT_ZFS_DELAY
) == 0) {
130 rc
= !zfs_ratelimit(&vd
->vdev_delay_rl
);
131 } else if (strcmp(subclass
, FM_EREPORT_ZFS_CHECKSUM
) == 0) {
132 rc
= !zfs_ratelimit(&vd
->vdev_checksum_rl
);
136 /* We're rate limiting */
137 fm_erpt_dropped_increment();
144 zfs_ereport_start(nvlist_t
**ereport_out
, nvlist_t
**detector_out
,
145 const char *subclass
, spa_t
*spa
, vdev_t
*vd
, zio_t
*zio
,
146 uint64_t stateoroffset
, uint64_t size
)
148 nvlist_t
*ereport
, *detector
;
154 * If we are doing a spa_tryimport() or in recovery mode,
157 if (spa_load_state(spa
) == SPA_LOAD_TRYIMPORT
||
158 spa_load_state(spa
) == SPA_LOAD_RECOVER
)
162 * If we are in the middle of opening a pool, and the previous attempt
163 * failed, don't bother logging any new ereports - we're just going to
164 * get the same diagnosis anyway.
166 if (spa_load_state(spa
) != SPA_LOAD_NONE
&&
167 spa
->spa_last_open_failed
)
172 * If this is not a read or write zio, ignore the error. This
173 * can occur if the DKIOCFLUSHWRITECACHE ioctl fails.
175 if (zio
->io_type
!= ZIO_TYPE_READ
&&
176 zio
->io_type
!= ZIO_TYPE_WRITE
)
181 * If the vdev has already been marked as failing due
182 * to a failed probe, then ignore any subsequent I/O
183 * errors, as the DE will automatically fault the vdev
184 * on the first such failure. This also catches cases
185 * where vdev_remove_wanted is set and the device has
186 * not yet been asynchronously placed into the REMOVED
189 if (zio
->io_vd
== vd
&& !vdev_accessible(vd
, zio
))
193 * Ignore checksum errors for reads from DTL regions of
196 if (zio
->io_type
== ZIO_TYPE_READ
&&
197 zio
->io_error
== ECKSUM
&&
198 vd
->vdev_ops
->vdev_op_leaf
&&
199 vdev_dtl_contains(vd
, DTL_MISSING
, zio
->io_txg
, 1))
205 * For probe failure, we want to avoid posting ereports if we've
206 * already removed the device in the meantime.
209 strcmp(subclass
, FM_EREPORT_ZFS_PROBE_FAILURE
) == 0 &&
210 (vd
->vdev_remove_wanted
|| vd
->vdev_state
== VDEV_STATE_REMOVED
))
213 if ((ereport
= fm_nvlist_create(NULL
)) == NULL
)
216 if ((detector
= fm_nvlist_create(NULL
)) == NULL
) {
217 fm_nvlist_destroy(ereport
, FM_NVA_FREE
);
221 if ((strcmp(subclass
, FM_EREPORT_ZFS_DELAY
) == 0) &&
222 (zio
!= NULL
) && (!zio
->io_timestamp
)) {
223 /* Ignore bogus delay events */
228 * Serialize ereport generation
230 mutex_enter(&spa
->spa_errlist_lock
);
233 * Determine the ENA to use for this event. If we are in a loading
234 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use
235 * a root zio-wide ENA. Otherwise, simply use a unique ENA.
237 if (spa_load_state(spa
) != SPA_LOAD_NONE
) {
238 if (spa
->spa_ena
== 0)
239 spa
->spa_ena
= fm_ena_generate(0, FM_ENA_FMT1
);
241 } else if (zio
!= NULL
&& zio
->io_logical
!= NULL
) {
242 if (zio
->io_logical
->io_ena
== 0)
243 zio
->io_logical
->io_ena
=
244 fm_ena_generate(0, FM_ENA_FMT1
);
245 ena
= zio
->io_logical
->io_ena
;
247 ena
= fm_ena_generate(0, FM_ENA_FMT1
);
251 * Construct the full class, detector, and other standard FMA fields.
253 (void) snprintf(class, sizeof (class), "%s.%s",
254 ZFS_ERROR_CLASS
, subclass
);
256 fm_fmri_zfs_set(detector
, FM_ZFS_SCHEME_VERSION
, spa_guid(spa
),
257 vd
!= NULL
? vd
->vdev_guid
: 0);
259 fm_ereport_set(ereport
, FM_EREPORT_VERSION
, class, ena
, detector
, NULL
);
262 * Construct the per-ereport payload, depending on which parameters are
267 * Generic payload members common to all ereports.
269 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_POOL
,
270 DATA_TYPE_STRING
, spa_name(spa
), FM_EREPORT_PAYLOAD_ZFS_POOL_GUID
,
271 DATA_TYPE_UINT64
, spa_guid(spa
),
272 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT
, DATA_TYPE_INT32
,
273 spa_load_state(spa
), NULL
);
276 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE
,
278 spa_get_failmode(spa
) == ZIO_FAILURE_MODE_WAIT
?
279 FM_EREPORT_FAILMODE_WAIT
:
280 spa_get_failmode(spa
) == ZIO_FAILURE_MODE_CONTINUE
?
281 FM_EREPORT_FAILMODE_CONTINUE
: FM_EREPORT_FAILMODE_PANIC
,
286 vdev_t
*pvd
= vd
->vdev_parent
;
287 vdev_queue_t
*vq
= &vd
->vdev_queue
;
288 vdev_stat_t
*vs
= &vd
->vdev_stat
;
290 uint64_t *spare_guids
;
294 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID
,
295 DATA_TYPE_UINT64
, vd
->vdev_guid
,
296 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE
,
297 DATA_TYPE_STRING
, vd
->vdev_ops
->vdev_op_type
, NULL
);
298 if (vd
->vdev_path
!= NULL
)
299 fm_payload_set(ereport
,
300 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH
,
301 DATA_TYPE_STRING
, vd
->vdev_path
, NULL
);
302 if (vd
->vdev_devid
!= NULL
)
303 fm_payload_set(ereport
,
304 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID
,
305 DATA_TYPE_STRING
, vd
->vdev_devid
, NULL
);
306 if (vd
->vdev_fru
!= NULL
)
307 fm_payload_set(ereport
,
308 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU
,
309 DATA_TYPE_STRING
, vd
->vdev_fru
, NULL
);
311 fm_payload_set(ereport
,
312 FM_EREPORT_PAYLOAD_ZFS_VDEV_ASHIFT
,
313 DATA_TYPE_UINT64
, vd
->vdev_ashift
, NULL
);
316 fm_payload_set(ereport
,
317 FM_EREPORT_PAYLOAD_ZFS_VDEV_COMP_TS
,
318 DATA_TYPE_UINT64
, vq
->vq_io_complete_ts
, NULL
);
319 fm_payload_set(ereport
,
320 FM_EREPORT_PAYLOAD_ZFS_VDEV_DELTA_TS
,
321 DATA_TYPE_UINT64
, vq
->vq_io_delta_ts
, NULL
);
325 fm_payload_set(ereport
,
326 FM_EREPORT_PAYLOAD_ZFS_VDEV_READ_ERRORS
,
327 DATA_TYPE_UINT64
, vs
->vs_read_errors
,
328 FM_EREPORT_PAYLOAD_ZFS_VDEV_WRITE_ERRORS
,
329 DATA_TYPE_UINT64
, vs
->vs_write_errors
,
330 FM_EREPORT_PAYLOAD_ZFS_VDEV_CKSUM_ERRORS
,
331 DATA_TYPE_UINT64
, vs
->vs_checksum_errors
, NULL
);
335 fm_payload_set(ereport
,
336 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID
,
337 DATA_TYPE_UINT64
, pvd
->vdev_guid
,
338 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE
,
339 DATA_TYPE_STRING
, pvd
->vdev_ops
->vdev_op_type
,
342 fm_payload_set(ereport
,
343 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH
,
344 DATA_TYPE_STRING
, pvd
->vdev_path
, NULL
);
346 fm_payload_set(ereport
,
347 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID
,
348 DATA_TYPE_STRING
, pvd
->vdev_devid
, NULL
);
351 spare_count
= spa
->spa_spares
.sav_count
;
352 spare_paths
= kmem_zalloc(sizeof (char *) * spare_count
,
354 spare_guids
= kmem_zalloc(sizeof (uint64_t) * spare_count
,
357 for (i
= 0; i
< spare_count
; i
++) {
358 spare_vd
= spa
->spa_spares
.sav_vdevs
[i
];
360 spare_paths
[i
] = spare_vd
->vdev_path
;
361 spare_guids
[i
] = spare_vd
->vdev_guid
;
365 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_VDEV_SPARE_PATHS
,
366 DATA_TYPE_STRING_ARRAY
, spare_count
, spare_paths
,
367 FM_EREPORT_PAYLOAD_ZFS_VDEV_SPARE_GUIDS
,
368 DATA_TYPE_UINT64_ARRAY
, spare_count
, spare_guids
, NULL
);
370 kmem_free(spare_guids
, sizeof (uint64_t) * spare_count
);
371 kmem_free(spare_paths
, sizeof (char *) * spare_count
);
376 * Payload common to all I/Os.
378 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR
,
379 DATA_TYPE_INT32
, zio
->io_error
, NULL
);
380 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_ZIO_FLAGS
,
381 DATA_TYPE_INT32
, zio
->io_flags
, NULL
);
382 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_ZIO_STAGE
,
383 DATA_TYPE_UINT32
, zio
->io_stage
, NULL
);
384 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_ZIO_PIPELINE
,
385 DATA_TYPE_UINT32
, zio
->io_pipeline
, NULL
);
386 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELAY
,
387 DATA_TYPE_UINT64
, zio
->io_delay
, NULL
);
388 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_ZIO_TIMESTAMP
,
389 DATA_TYPE_UINT64
, zio
->io_timestamp
, NULL
);
390 fm_payload_set(ereport
, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELTA
,
391 DATA_TYPE_UINT64
, zio
->io_delta
, NULL
);
394 * If the 'size' parameter is non-zero, it indicates this is a
395 * RAID-Z or other I/O where the physical offset and length are
396 * provided for us, instead of within the zio_t.
400 fm_payload_set(ereport
,
401 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET
,
402 DATA_TYPE_UINT64
, stateoroffset
,
403 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE
,
404 DATA_TYPE_UINT64
, size
, NULL
);
406 fm_payload_set(ereport
,
407 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET
,
408 DATA_TYPE_UINT64
, zio
->io_offset
,
409 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE
,
410 DATA_TYPE_UINT64
, zio
->io_size
, NULL
);
414 * Payload for I/Os with corresponding logical information.
416 if (zio
->io_logical
!= NULL
)
417 fm_payload_set(ereport
,
418 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET
,
420 zio
->io_logical
->io_bookmark
.zb_objset
,
421 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT
,
423 zio
->io_logical
->io_bookmark
.zb_object
,
424 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL
,
426 zio
->io_logical
->io_bookmark
.zb_level
,
427 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID
,
429 zio
->io_logical
->io_bookmark
.zb_blkid
, NULL
);
430 } else if (vd
!= NULL
) {
432 * If we have a vdev but no zio, this is a device fault, and the
433 * 'stateoroffset' parameter indicates the previous state of the
436 fm_payload_set(ereport
,
437 FM_EREPORT_PAYLOAD_ZFS_PREV_STATE
,
438 DATA_TYPE_UINT64
, stateoroffset
, NULL
);
441 mutex_exit(&spa
->spa_errlist_lock
);
443 *ereport_out
= ereport
;
444 *detector_out
= detector
;
447 /* if it's <= 128 bytes, save the corruption directly */
448 #define ZFM_MAX_INLINE (128 / sizeof (uint64_t))
450 #define MAX_RANGES 16
452 typedef struct zfs_ecksum_info
{
453 /* histograms of set and cleared bits by bit number in a 64-bit word */
454 uint16_t zei_histogram_set
[sizeof (uint64_t) * NBBY
];
455 uint16_t zei_histogram_cleared
[sizeof (uint64_t) * NBBY
];
457 /* inline arrays of bits set and cleared. */
458 uint64_t zei_bits_set
[ZFM_MAX_INLINE
];
459 uint64_t zei_bits_cleared
[ZFM_MAX_INLINE
];
462 * for each range, the number of bits set and cleared. The Hamming
463 * distance between the good and bad buffers is the sum of them all.
465 uint32_t zei_range_sets
[MAX_RANGES
];
466 uint32_t zei_range_clears
[MAX_RANGES
];
471 } zei_ranges
[MAX_RANGES
];
473 size_t zei_range_count
;
475 uint32_t zei_allowed_mingap
;
480 update_histogram(uint64_t value_arg
, uint16_t *hist
, uint32_t *count
)
484 uint64_t value
= BE_64(value_arg
);
486 /* We store the bits in big-endian (largest-first) order */
487 for (i
= 0; i
< 64; i
++) {
488 if (value
& (1ull << i
)) {
489 if (hist
[63 - i
] < UINT16_MAX
)
494 /* update the count of bits changed */
499 * We've now filled up the range array, and need to increase "mingap" and
500 * shrink the range list accordingly. zei_mingap is always the smallest
501 * distance between array entries, so we set the new_allowed_gap to be
502 * one greater than that. We then go through the list, joining together
503 * any ranges which are closer than the new_allowed_gap.
505 * By construction, there will be at least one. We also update zei_mingap
506 * to the new smallest gap, to prepare for our next invocation.
509 zei_shrink_ranges(zfs_ecksum_info_t
*eip
)
511 uint32_t mingap
= UINT32_MAX
;
512 uint32_t new_allowed_gap
= eip
->zei_mingap
+ 1;
515 size_t max
= eip
->zei_range_count
;
517 struct zei_ranges
*r
= eip
->zei_ranges
;
519 ASSERT3U(eip
->zei_range_count
, >, 0);
520 ASSERT3U(eip
->zei_range_count
, <=, MAX_RANGES
);
523 while (idx
< max
- 1) {
524 uint32_t start
= r
[idx
].zr_start
;
525 uint32_t end
= r
[idx
].zr_end
;
527 while (idx
< max
- 1) {
528 uint32_t nstart
, nend
, gap
;
531 nstart
= r
[idx
].zr_start
;
532 nend
= r
[idx
].zr_end
;
535 if (gap
< new_allowed_gap
) {
543 r
[output
].zr_start
= start
;
544 r
[output
].zr_end
= end
;
547 ASSERT3U(output
, <, eip
->zei_range_count
);
548 eip
->zei_range_count
= output
;
549 eip
->zei_mingap
= mingap
;
550 eip
->zei_allowed_mingap
= new_allowed_gap
;
554 zei_add_range(zfs_ecksum_info_t
*eip
, int start
, int end
)
556 struct zei_ranges
*r
= eip
->zei_ranges
;
557 size_t count
= eip
->zei_range_count
;
559 if (count
>= MAX_RANGES
) {
560 zei_shrink_ranges(eip
);
561 count
= eip
->zei_range_count
;
564 eip
->zei_mingap
= UINT32_MAX
;
565 eip
->zei_allowed_mingap
= 1;
567 int gap
= start
- r
[count
- 1].zr_end
;
569 if (gap
< eip
->zei_allowed_mingap
) {
570 r
[count
- 1].zr_end
= end
;
573 if (gap
< eip
->zei_mingap
)
574 eip
->zei_mingap
= gap
;
576 r
[count
].zr_start
= start
;
577 r
[count
].zr_end
= end
;
578 eip
->zei_range_count
++;
582 zei_range_total_size(zfs_ecksum_info_t
*eip
)
584 struct zei_ranges
*r
= eip
->zei_ranges
;
585 size_t count
= eip
->zei_range_count
;
589 for (idx
= 0; idx
< count
; idx
++)
590 result
+= (r
[idx
].zr_end
- r
[idx
].zr_start
);
595 static zfs_ecksum_info_t
*
596 annotate_ecksum(nvlist_t
*ereport
, zio_bad_cksum_t
*info
,
597 const uint8_t *goodbuf
, const uint8_t *badbuf
, size_t size
,
598 boolean_t drop_if_identical
)
600 const uint64_t *good
= (const uint64_t *)goodbuf
;
601 const uint64_t *bad
= (const uint64_t *)badbuf
;
604 uint64_t allcleared
= 0;
606 size_t nui64s
= size
/ sizeof (uint64_t);
616 zfs_ecksum_info_t
*eip
= kmem_zalloc(sizeof (*eip
), KM_SLEEP
);
618 /* don't do any annotation for injected checksum errors */
619 if (info
!= NULL
&& info
->zbc_injected
)
622 if (info
!= NULL
&& info
->zbc_has_cksum
) {
623 fm_payload_set(ereport
,
624 FM_EREPORT_PAYLOAD_ZFS_CKSUM_EXPECTED
,
625 DATA_TYPE_UINT64_ARRAY
,
626 sizeof (info
->zbc_expected
) / sizeof (uint64_t),
627 (uint64_t *)&info
->zbc_expected
,
628 FM_EREPORT_PAYLOAD_ZFS_CKSUM_ACTUAL
,
629 DATA_TYPE_UINT64_ARRAY
,
630 sizeof (info
->zbc_actual
) / sizeof (uint64_t),
631 (uint64_t *)&info
->zbc_actual
,
632 FM_EREPORT_PAYLOAD_ZFS_CKSUM_ALGO
,
634 info
->zbc_checksum_name
,
637 if (info
->zbc_byteswapped
) {
638 fm_payload_set(ereport
,
639 FM_EREPORT_PAYLOAD_ZFS_CKSUM_BYTESWAP
,
640 DATA_TYPE_BOOLEAN
, 1,
645 if (badbuf
== NULL
|| goodbuf
== NULL
)
648 ASSERT3U(size
, ==, nui64s
* sizeof (uint64_t));
649 ASSERT3U(size
, <=, SPA_MAXBLOCKSIZE
);
650 ASSERT3U(size
, <=, UINT32_MAX
);
652 /* build up the range list by comparing the two buffers. */
653 for (idx
= 0; idx
< nui64s
; idx
++) {
654 if (good
[idx
] == bad
[idx
]) {
658 zei_add_range(eip
, start
, idx
);
668 zei_add_range(eip
, start
, idx
);
670 /* See if it will fit in our inline buffers */
671 inline_size
= zei_range_total_size(eip
);
672 if (inline_size
> ZFM_MAX_INLINE
)
676 * If there is no change and we want to drop if the buffers are
679 if (inline_size
== 0 && drop_if_identical
) {
680 kmem_free(eip
, sizeof (*eip
));
685 * Now walk through the ranges, filling in the details of the
686 * differences. Also convert our uint64_t-array offsets to byte
689 for (range
= 0; range
< eip
->zei_range_count
; range
++) {
690 size_t start
= eip
->zei_ranges
[range
].zr_start
;
691 size_t end
= eip
->zei_ranges
[range
].zr_end
;
693 for (idx
= start
; idx
< end
; idx
++) {
694 uint64_t set
, cleared
;
696 // bits set in bad, but not in good
697 set
= ((~good
[idx
]) & bad
[idx
]);
698 // bits set in good, but not in bad
699 cleared
= (good
[idx
] & (~bad
[idx
]));
702 allcleared
|= cleared
;
705 ASSERT3U(offset
, <, inline_size
);
706 eip
->zei_bits_set
[offset
] = set
;
707 eip
->zei_bits_cleared
[offset
] = cleared
;
711 update_histogram(set
, eip
->zei_histogram_set
,
712 &eip
->zei_range_sets
[range
]);
713 update_histogram(cleared
, eip
->zei_histogram_cleared
,
714 &eip
->zei_range_clears
[range
]);
717 /* convert to byte offsets */
718 eip
->zei_ranges
[range
].zr_start
*= sizeof (uint64_t);
719 eip
->zei_ranges
[range
].zr_end
*= sizeof (uint64_t);
721 eip
->zei_allowed_mingap
*= sizeof (uint64_t);
722 inline_size
*= sizeof (uint64_t);
724 /* fill in ereport */
725 fm_payload_set(ereport
,
726 FM_EREPORT_PAYLOAD_ZFS_BAD_OFFSET_RANGES
,
727 DATA_TYPE_UINT32_ARRAY
, 2 * eip
->zei_range_count
,
728 (uint32_t *)eip
->zei_ranges
,
729 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_MIN_GAP
,
730 DATA_TYPE_UINT32
, eip
->zei_allowed_mingap
,
731 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_SETS
,
732 DATA_TYPE_UINT32_ARRAY
, eip
->zei_range_count
, eip
->zei_range_sets
,
733 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_CLEARS
,
734 DATA_TYPE_UINT32_ARRAY
, eip
->zei_range_count
, eip
->zei_range_clears
,
738 fm_payload_set(ereport
,
739 FM_EREPORT_PAYLOAD_ZFS_BAD_SET_BITS
,
740 DATA_TYPE_UINT8_ARRAY
,
741 inline_size
, (uint8_t *)eip
->zei_bits_set
,
742 FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_BITS
,
743 DATA_TYPE_UINT8_ARRAY
,
744 inline_size
, (uint8_t *)eip
->zei_bits_cleared
,
747 fm_payload_set(ereport
,
748 FM_EREPORT_PAYLOAD_ZFS_BAD_SET_HISTOGRAM
,
749 DATA_TYPE_UINT16_ARRAY
,
750 NBBY
* sizeof (uint64_t), eip
->zei_histogram_set
,
751 FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_HISTOGRAM
,
752 DATA_TYPE_UINT16_ARRAY
,
753 NBBY
* sizeof (uint64_t), eip
->zei_histogram_cleared
,
761 zfs_ereport_post(const char *subclass
, spa_t
*spa
, vdev_t
*vd
, zio_t
*zio
,
762 uint64_t stateoroffset
, uint64_t size
)
765 nvlist_t
*ereport
= NULL
;
766 nvlist_t
*detector
= NULL
;
768 zfs_ereport_start(&ereport
, &detector
,
769 subclass
, spa
, vd
, zio
, stateoroffset
, size
);
774 if (zfs_is_ratelimiting_event(subclass
, vd
))
777 /* Cleanup is handled by the callback function */
778 zfs_zevent_post(ereport
, detector
, zfs_zevent_post_cb
);
783 zfs_ereport_start_checksum(spa_t
*spa
, vdev_t
*vd
,
784 struct zio
*zio
, uint64_t offset
, uint64_t length
, void *arg
,
785 zio_bad_cksum_t
*info
)
787 zio_cksum_report_t
*report
;
791 if (zfs_is_ratelimiting_event(FM_EREPORT_ZFS_CHECKSUM
, vd
))
795 report
= kmem_zalloc(sizeof (*report
), KM_SLEEP
);
797 if (zio
->io_vsd
!= NULL
)
798 zio
->io_vsd_ops
->vsd_cksum_report(zio
, report
, arg
);
800 zio_vsd_default_cksum_report(zio
, report
, arg
);
802 /* copy the checksum failure information if it was provided */
804 report
->zcr_ckinfo
= kmem_zalloc(sizeof (*info
), KM_SLEEP
);
805 bcopy(info
, report
->zcr_ckinfo
, sizeof (*info
));
808 report
->zcr_align
= 1ULL << vd
->vdev_top
->vdev_ashift
;
809 report
->zcr_length
= length
;
812 zfs_ereport_start(&report
->zcr_ereport
, &report
->zcr_detector
,
813 FM_EREPORT_ZFS_CHECKSUM
, spa
, vd
, zio
, offset
, length
);
815 if (report
->zcr_ereport
== NULL
) {
816 zfs_ereport_free_checksum(report
);
821 mutex_enter(&spa
->spa_errlist_lock
);
822 report
->zcr_next
= zio
->io_logical
->io_cksum_report
;
823 zio
->io_logical
->io_cksum_report
= report
;
824 mutex_exit(&spa
->spa_errlist_lock
);
828 zfs_ereport_finish_checksum(zio_cksum_report_t
*report
,
829 const void *good_data
, const void *bad_data
, boolean_t drop_if_identical
)
832 zfs_ecksum_info_t
*info
;
834 info
= annotate_ecksum(report
->zcr_ereport
, report
->zcr_ckinfo
,
835 good_data
, bad_data
, report
->zcr_length
, drop_if_identical
);
837 zfs_zevent_post(report
->zcr_ereport
,
838 report
->zcr_detector
, zfs_zevent_post_cb
);
840 zfs_zevent_post_cb(report
->zcr_ereport
, report
->zcr_detector
);
842 report
->zcr_ereport
= report
->zcr_detector
= NULL
;
844 kmem_free(info
, sizeof (*info
));
849 zfs_ereport_free_checksum(zio_cksum_report_t
*rpt
)
852 if (rpt
->zcr_ereport
!= NULL
) {
853 fm_nvlist_destroy(rpt
->zcr_ereport
,
855 fm_nvlist_destroy(rpt
->zcr_detector
,
859 rpt
->zcr_free(rpt
->zcr_cbdata
, rpt
->zcr_cbinfo
);
861 if (rpt
->zcr_ckinfo
!= NULL
)
862 kmem_free(rpt
->zcr_ckinfo
, sizeof (*rpt
->zcr_ckinfo
));
864 kmem_free(rpt
, sizeof (*rpt
));
869 zfs_ereport_post_checksum(spa_t
*spa
, vdev_t
*vd
,
870 struct zio
*zio
, uint64_t offset
, uint64_t length
,
871 const void *good_data
, const void *bad_data
, zio_bad_cksum_t
*zbc
)
874 nvlist_t
*ereport
= NULL
;
875 nvlist_t
*detector
= NULL
;
876 zfs_ecksum_info_t
*info
;
878 zfs_ereport_start(&ereport
, &detector
,
879 FM_EREPORT_ZFS_CHECKSUM
, spa
, vd
, zio
, offset
, length
);
884 info
= annotate_ecksum(ereport
, zbc
, good_data
, bad_data
, length
,
888 zfs_zevent_post(ereport
, detector
, zfs_zevent_post_cb
);
889 kmem_free(info
, sizeof (*info
));
895 zfs_post_common(spa_t
*spa
, vdev_t
*vd
, const char *type
, const char *name
,
902 if (spa_load_state(spa
) == SPA_LOAD_TRYIMPORT
)
905 if ((resource
= fm_nvlist_create(NULL
)) == NULL
)
908 (void) snprintf(class, sizeof (class), "%s.%s.%s", type
,
909 ZFS_ERROR_CLASS
, name
);
910 VERIFY0(nvlist_add_uint8(resource
, FM_VERSION
, FM_RSRC_VERSION
));
911 VERIFY0(nvlist_add_string(resource
, FM_CLASS
, class));
912 VERIFY0(nvlist_add_uint64(resource
,
913 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID
, spa_guid(spa
)));
914 VERIFY0(nvlist_add_int32(resource
,
915 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT
, spa_load_state(spa
)));
918 VERIFY0(nvlist_add_uint64(resource
,
919 FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID
, vd
->vdev_guid
));
920 VERIFY0(nvlist_add_uint64(resource
,
921 FM_EREPORT_PAYLOAD_ZFS_VDEV_STATE
, vd
->vdev_state
));
922 if (vd
->vdev_path
!= NULL
)
923 VERIFY0(nvlist_add_string(resource
,
924 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH
, vd
->vdev_path
));
925 if (vd
->vdev_devid
!= NULL
)
926 VERIFY0(nvlist_add_string(resource
,
927 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID
, vd
->vdev_devid
));
928 if (vd
->vdev_fru
!= NULL
)
929 VERIFY0(nvlist_add_string(resource
,
930 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU
, vd
->vdev_fru
));
931 /* also copy any optional payload data */
933 nvpair_t
*elem
= NULL
;
935 while ((elem
= nvlist_next_nvpair(aux
, elem
)) != NULL
)
936 (void) nvlist_add_nvpair(resource
, elem
);
940 zfs_zevent_post(resource
, NULL
, zfs_zevent_post_cb
);
945 * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev
946 * has been removed from the system. This will cause the DE to ignore any
947 * recent I/O errors, inferring that they are due to the asynchronous device
951 zfs_post_remove(spa_t
*spa
, vdev_t
*vd
)
953 zfs_post_common(spa
, vd
, FM_RSRC_CLASS
, FM_RESOURCE_REMOVED
, NULL
);
957 * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool
958 * has the 'autoreplace' property set, and therefore any broken vdevs will be
959 * handled by higher level logic, and no vdev fault should be generated.
962 zfs_post_autoreplace(spa_t
*spa
, vdev_t
*vd
)
964 zfs_post_common(spa
, vd
, FM_RSRC_CLASS
, FM_RESOURCE_AUTOREPLACE
, NULL
);
968 * The 'resource.fs.zfs.statechange' event is an internal signal that the
969 * given vdev has transitioned its state to DEGRADED or HEALTHY. This will
970 * cause the retire agent to repair any outstanding fault management cases
971 * open because the device was not found (fault.fs.zfs.device).
974 zfs_post_state_change(spa_t
*spa
, vdev_t
*vd
, uint64_t laststate
)
980 * Add optional supplemental keys to payload
982 aux
= fm_nvlist_create(NULL
);
984 if (vd
->vdev_physpath
) {
985 (void) nvlist_add_string(aux
,
986 FM_EREPORT_PAYLOAD_ZFS_VDEV_PHYSPATH
,
989 if (vd
->vdev_enc_sysfs_path
) {
990 (void) nvlist_add_string(aux
,
991 FM_EREPORT_PAYLOAD_ZFS_VDEV_ENC_SYSFS_PATH
,
992 vd
->vdev_enc_sysfs_path
);
995 (void) nvlist_add_uint64(aux
,
996 FM_EREPORT_PAYLOAD_ZFS_VDEV_LASTSTATE
, laststate
);
999 zfs_post_common(spa
, vd
, FM_RSRC_CLASS
, FM_RESOURCE_STATECHANGE
,
1003 fm_nvlist_destroy(aux
, FM_NVA_FREE
);
1008 * The 'sysevent.fs.zfs.*' events are signals posted to notify user space of
1009 * change in the pool. All sysevents are listed in sys/sysevent/eventdefs.h
1010 * and are designed to be consumed by the ZFS Event Daemon (ZED). For
1011 * additional details refer to the zed(8) man page.
1014 zfs_post_sysevent(spa_t
*spa
, vdev_t
*vd
, const char *name
)
1016 zfs_post_common(spa
, vd
, FM_SYSEVENT_CLASS
, name
, NULL
);
1019 #if defined(_KERNEL) && defined(HAVE_SPL)
1020 EXPORT_SYMBOL(zfs_ereport_post
);
1021 EXPORT_SYMBOL(zfs_ereport_post_checksum
);
1022 EXPORT_SYMBOL(zfs_post_remove
);
1023 EXPORT_SYMBOL(zfs_post_autoreplace
);
1024 EXPORT_SYMBOL(zfs_post_state_change
);
1025 EXPORT_SYMBOL(zfs_post_sysevent
);
1026 #endif /* _KERNEL */