1/*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26/*
27 * Copyright (c) 2012 by Delphix. All rights reserved.
28 */
29
30#include <sys/spa.h>
31#include <sys/spa_impl.h>
32#include <sys/vdev.h>
33#include <sys/vdev_impl.h>
34#include <sys/zio.h>
35#include <sys/zio_checksum.h>
36
37#include <sys/fm/fs/zfs.h>
38#include <sys/fm/protocol.h>
39#include <sys/fm/util.h>
40#include <sys/sysevent.h>
41
42/*
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:
46 *
47 *	EREPORT			POOL	VDEV	IO
48 *	block			X	X	X
49 *	data			X		X
50 *	device			X	X
51 *	pool			X
52 *
53 * If we are in a loading state, all errors are chained together by the same
54 * SPA-wide ENA (Error Numeric Association).
55 *
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
60 * layered diagram:
61 *
62 *      +---------------+
63 *	| Aggregate I/O |	No associated logical data or device
64 *	+---------------+
65 *              |
66 *              V
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.
70 *              |
71 *              V
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.
75 *
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).
80 *
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
86 * ENA.
87 *
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).
95 *
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.
103 */
104#ifdef _KERNEL
105static void
106zfs_ereport_start(nvlist_t **ereport_out, nvlist_t **detector_out,
107    const char *subclass, spa_t *spa, vdev_t *vd, const zbookmark_phys_t *zb,
108    zio_t *zio, uint64_t stateoroffset, uint64_t size)
109{
110	nvlist_t *ereport, *detector;
111
112	uint64_t ena;
113	char class[64];
114
115	/*
116	 * If we are doing a spa_tryimport() or in recovery mode,
117	 * ignore errors.
118	 */
119	if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT ||
120	    spa_load_state(spa) == SPA_LOAD_RECOVER)
121		return;
122
123	/*
124	 * If we are in the middle of opening a pool, and the previous attempt
125	 * failed, don't bother logging any new ereports - we're just going to
126	 * get the same diagnosis anyway.
127	 */
128	if (spa_load_state(spa) != SPA_LOAD_NONE &&
129	    spa->spa_last_open_failed)
130		return;
131
132	if (zio != NULL) {
133		/*
134		 * If this is not a read or write zio, ignore the error.  This
135		 * can occur if the DKIOCFLUSHWRITECACHE ioctl fails.
136		 */
137		if (zio->io_type != ZIO_TYPE_READ &&
138		    zio->io_type != ZIO_TYPE_WRITE)
139			return;
140
141		/*
142		 * Ignore any errors from speculative I/Os, as failure is an
143		 * expected result.
144		 */
145		if (zio->io_flags & ZIO_FLAG_SPECULATIVE)
146			return;
147
148		/*
149		 * If this I/O is not a retry I/O, don't post an ereport.
150		 * Otherwise, we risk making bad diagnoses based on B_FAILFAST
151		 * I/Os.
152		 */
153		if (zio->io_error == EIO &&
154		    !(zio->io_flags & ZIO_FLAG_IO_RETRY))
155			return;
156
157		if (vd != NULL) {
158			/*
159			 * If the vdev has already been marked as failing due
160			 * to a failed probe, then ignore any subsequent I/O
161			 * errors, as the DE will automatically fault the vdev
162			 * on the first such failure.  This also catches cases
163			 * where vdev_remove_wanted is set and the device has
164			 * not yet been asynchronously placed into the REMOVED
165			 * state.
166			 */
167			if (zio->io_vd == vd && !vdev_accessible(vd, zio))
168				return;
169
170			/*
171			 * Ignore checksum errors for reads from DTL regions of
172			 * leaf vdevs.
173			 */
174			if (zio->io_type == ZIO_TYPE_READ &&
175			    zio->io_error == ECKSUM &&
176			    vd->vdev_ops->vdev_op_leaf &&
177			    vdev_dtl_contains(vd, DTL_MISSING, zio->io_txg, 1))
178				return;
179		}
180	}
181
182	/*
183	 * For probe failure, we want to avoid posting ereports if we've
184	 * already removed the device in the meantime.
185	 */
186	if (vd != NULL &&
187	    strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) == 0 &&
188	    (vd->vdev_remove_wanted || vd->vdev_state == VDEV_STATE_REMOVED))
189		return;
190
191	if ((ereport = fm_nvlist_create(NULL)) == NULL)
192		return;
193
194	if ((detector = fm_nvlist_create(NULL)) == NULL) {
195		fm_nvlist_destroy(ereport, FM_NVA_FREE);
196		return;
197	}
198
199	/*
200	 * Serialize ereport generation
201	 */
202	mutex_enter(&spa->spa_errlist_lock);
203
204	/*
205	 * Determine the ENA to use for this event.  If we are in a loading
206	 * state, use a SPA-wide ENA.  Otherwise, if we are in an I/O state, use
207	 * a root zio-wide ENA.  Otherwise, simply use a unique ENA.
208	 */
209	if (spa_load_state(spa) != SPA_LOAD_NONE) {
210		if (spa->spa_ena == 0)
211			spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1);
212		ena = spa->spa_ena;
213	} else if (zio != NULL && zio->io_logical != NULL) {
214		if (zio->io_logical->io_ena == 0)
215			zio->io_logical->io_ena =
216			    fm_ena_generate(0, FM_ENA_FMT1);
217		ena = zio->io_logical->io_ena;
218	} else {
219		ena = fm_ena_generate(0, FM_ENA_FMT1);
220	}
221
222	/*
223	 * Construct the full class, detector, and other standard FMA fields.
224	 */
225	(void) snprintf(class, sizeof (class), "%s.%s",
226	    ZFS_ERROR_CLASS, subclass);
227
228	fm_fmri_zfs_set(detector, FM_ZFS_SCHEME_VERSION, spa_guid(spa),
229	    vd != NULL ? vd->vdev_guid : 0);
230
231	fm_ereport_set(ereport, FM_EREPORT_VERSION, class, ena, detector, NULL);
232
233	/*
234	 * Construct the per-ereport payload, depending on which parameters are
235	 * passed in.
236	 */
237
238	/*
239	 * Generic payload members common to all ereports.
240	 */
241	fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL,
242	    DATA_TYPE_STRING, spa_name(spa), FM_EREPORT_PAYLOAD_ZFS_POOL_GUID,
243	    DATA_TYPE_UINT64, spa_guid(spa),
244	    FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, DATA_TYPE_INT32,
245	    spa_load_state(spa), NULL);
246
247	if (spa != NULL) {
248		fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE,
249		    DATA_TYPE_STRING,
250		    spa_get_failmode(spa) == ZIO_FAILURE_MODE_WAIT ?
251		    FM_EREPORT_FAILMODE_WAIT :
252		    spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE ?
253		    FM_EREPORT_FAILMODE_CONTINUE : FM_EREPORT_FAILMODE_PANIC,
254		    NULL);
255	}
256
257	if (vd != NULL) {
258		vdev_t *pvd = vd->vdev_parent;
259
260		fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID,
261		    DATA_TYPE_UINT64, vd->vdev_guid,
262		    FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE,
263		    DATA_TYPE_STRING, vd->vdev_ops->vdev_op_type, NULL);
264		if (vd->vdev_path != NULL)
265			fm_payload_set(ereport,
266			    FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH,
267			    DATA_TYPE_STRING, vd->vdev_path, NULL);
268		if (vd->vdev_devid != NULL)
269			fm_payload_set(ereport,
270			    FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID,
271			    DATA_TYPE_STRING, vd->vdev_devid, NULL);
272		if (vd->vdev_fru != NULL)
273			fm_payload_set(ereport,
274			    FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU,
275			    DATA_TYPE_STRING, vd->vdev_fru, NULL);
276		if (vd->vdev_ashift)
277			fm_payload_set(ereport,
278			    FM_EREPORT_PAYLOAD_ZFS_VDEV_ASHIFT,
279			    DATA_TYPE_UINT64, vd->vdev_ashift, NULL);
280
281		if (pvd != NULL) {
282			fm_payload_set(ereport,
283			    FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID,
284			    DATA_TYPE_UINT64, pvd->vdev_guid,
285			    FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE,
286			    DATA_TYPE_STRING, pvd->vdev_ops->vdev_op_type,
287			    NULL);
288			if (pvd->vdev_path)
289				fm_payload_set(ereport,
290				    FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH,
291				    DATA_TYPE_STRING, pvd->vdev_path, NULL);
292			if (pvd->vdev_devid)
293				fm_payload_set(ereport,
294				    FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID,
295				    DATA_TYPE_STRING, pvd->vdev_devid, NULL);
296		}
297	}
298
299	if (zio != NULL) {
300		/*
301		 * Payload common to all I/Os.
302		 */
303		fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR,
304		    DATA_TYPE_INT32, zio->io_error, NULL);
305
306		/*
307		 * If the 'size' parameter is non-zero, it indicates this is a
308		 * RAID-Z or other I/O where the physical offset and length are
309		 * provided for us, instead of within the zio_t.
310		 */
311		if (vd != NULL) {
312			if (size)
313				fm_payload_set(ereport,
314				    FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET,
315				    DATA_TYPE_UINT64, stateoroffset,
316				    FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE,
317				    DATA_TYPE_UINT64, size, NULL);
318			else
319				fm_payload_set(ereport,
320				    FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET,
321				    DATA_TYPE_UINT64, zio->io_offset,
322				    FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE,
323				    DATA_TYPE_UINT64, zio->io_size, NULL);
324		}
325	} else if (vd != NULL) {
326		/*
327		 * If we have a vdev but no zio, this is a device fault, and the
328		 * 'stateoroffset' parameter indicates the previous state of the
329		 * vdev.
330		 */
331		fm_payload_set(ereport,
332		    FM_EREPORT_PAYLOAD_ZFS_PREV_STATE,
333		    DATA_TYPE_UINT64, stateoroffset, NULL);
334	}
335
336	/*
337	 * Payload for I/Os with corresponding logical information.
338	 */
339	if (zb != NULL && (zio == NULL || zio->io_logical != NULL))
340		fm_payload_set(ereport,
341		    FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET,
342		    DATA_TYPE_UINT64, zb->zb_objset,
343		    FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT,
344		    DATA_TYPE_UINT64, zb->zb_object,
345		    FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL,
346		    DATA_TYPE_INT64, zb->zb_level,
347		    FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID,
348		    DATA_TYPE_UINT64, zb->zb_blkid, NULL);
349
350	mutex_exit(&spa->spa_errlist_lock);
351
352	*ereport_out = ereport;
353	*detector_out = detector;
354}
355
356/* if it's <= 128 bytes, save the corruption directly */
357#define	ZFM_MAX_INLINE		(128 / sizeof (uint64_t))
358
359#define	MAX_RANGES		16
360
361typedef struct zfs_ecksum_info {
362	/* histograms of set and cleared bits by bit number in a 64-bit word */
363	uint32_t zei_histogram_set[sizeof (uint64_t) * NBBY];
364	uint32_t zei_histogram_cleared[sizeof (uint64_t) * NBBY];
365
366	/* inline arrays of bits set and cleared. */
367	uint64_t zei_bits_set[ZFM_MAX_INLINE];
368	uint64_t zei_bits_cleared[ZFM_MAX_INLINE];
369
370	/*
371	 * for each range, the number of bits set and cleared.  The Hamming
372	 * distance between the good and bad buffers is the sum of them all.
373	 */
374	uint32_t zei_range_sets[MAX_RANGES];
375	uint32_t zei_range_clears[MAX_RANGES];
376
377	struct zei_ranges {
378		uint32_t	zr_start;
379		uint32_t	zr_end;
380	} zei_ranges[MAX_RANGES];
381
382	size_t	zei_range_count;
383	uint32_t zei_mingap;
384	uint32_t zei_allowed_mingap;
385
386} zfs_ecksum_info_t;
387
388static void
389update_histogram(uint64_t value_arg, uint32_t *hist, uint32_t *count)
390{
391	size_t i;
392	size_t bits = 0;
393	uint64_t value = BE_64(value_arg);
394
395	/* We store the bits in big-endian (largest-first) order */
396	for (i = 0; i < 64; i++) {
397		if (value & (1ull << i)) {
398			hist[63 - i]++;
399			++bits;
400		}
401	}
402	/* update the count of bits changed */
403	*count += bits;
404}
405
406/*
407 * We've now filled up the range array, and need to increase "mingap" and
408 * shrink the range list accordingly.  zei_mingap is always the smallest
409 * distance between array entries, so we set the new_allowed_gap to be
410 * one greater than that.  We then go through the list, joining together
411 * any ranges which are closer than the new_allowed_gap.
412 *
413 * By construction, there will be at least one.  We also update zei_mingap
414 * to the new smallest gap, to prepare for our next invocation.
415 */
416static void
417shrink_ranges(zfs_ecksum_info_t *eip)
418{
419	uint32_t mingap = UINT32_MAX;
420	uint32_t new_allowed_gap = eip->zei_mingap + 1;
421
422	size_t idx, output;
423	size_t max = eip->zei_range_count;
424
425	struct zei_ranges *r = eip->zei_ranges;
426
427	ASSERT3U(eip->zei_range_count, >, 0);
428	ASSERT3U(eip->zei_range_count, <=, MAX_RANGES);
429
430	output = idx = 0;
431	while (idx < max - 1) {
432		uint32_t start = r[idx].zr_start;
433		uint32_t end = r[idx].zr_end;
434
435		while (idx < max - 1) {
436			idx++;
437
438			uint32_t nstart = r[idx].zr_start;
439			uint32_t nend = r[idx].zr_end;
440
441			uint32_t gap = nstart - end;
442			if (gap < new_allowed_gap) {
443				end = nend;
444				continue;
445			}
446			if (gap < mingap)
447				mingap = gap;
448			break;
449		}
450		r[output].zr_start = start;
451		r[output].zr_end = end;
452		output++;
453	}
454	ASSERT3U(output, <, eip->zei_range_count);
455	eip->zei_range_count = output;
456	eip->zei_mingap = mingap;
457	eip->zei_allowed_mingap = new_allowed_gap;
458}
459
460static void
461add_range(zfs_ecksum_info_t *eip, int start, int end)
462{
463	struct zei_ranges *r = eip->zei_ranges;
464	size_t count = eip->zei_range_count;
465
466	if (count >= MAX_RANGES) {
467		shrink_ranges(eip);
468		count = eip->zei_range_count;
469	}
470	if (count == 0) {
471		eip->zei_mingap = UINT32_MAX;
472		eip->zei_allowed_mingap = 1;
473	} else {
474		int gap = start - r[count - 1].zr_end;
475
476		if (gap < eip->zei_allowed_mingap) {
477			r[count - 1].zr_end = end;
478			return;
479		}
480		if (gap < eip->zei_mingap)
481			eip->zei_mingap = gap;
482	}
483	r[count].zr_start = start;
484	r[count].zr_end = end;
485	eip->zei_range_count++;
486}
487
488static size_t
489range_total_size(zfs_ecksum_info_t *eip)
490{
491	struct zei_ranges *r = eip->zei_ranges;
492	size_t count = eip->zei_range_count;
493	size_t result = 0;
494	size_t idx;
495
496	for (idx = 0; idx < count; idx++)
497		result += (r[idx].zr_end - r[idx].zr_start);
498
499	return (result);
500}
501
502static zfs_ecksum_info_t *
503annotate_ecksum(nvlist_t *ereport, zio_bad_cksum_t *info,
504    const abd_t *goodabd, const abd_t *badabd, size_t size,
505    boolean_t drop_if_identical)
506{
507	const uint64_t *good;
508	const uint64_t *bad;
509
510	uint64_t allset = 0;
511	uint64_t allcleared = 0;
512
513	size_t nui64s = size / sizeof (uint64_t);
514
515	size_t inline_size;
516	int no_inline = 0;
517	size_t idx;
518	size_t range;
519
520	size_t offset = 0;
521	ssize_t start = -1;
522
523	zfs_ecksum_info_t *eip = kmem_zalloc(sizeof (*eip), KM_SLEEP);
524
525	/* don't do any annotation for injected checksum errors */
526	if (info != NULL && info->zbc_injected)
527		return (eip);
528
529	if (info != NULL && info->zbc_has_cksum) {
530		fm_payload_set(ereport,
531		    FM_EREPORT_PAYLOAD_ZFS_CKSUM_EXPECTED,
532		    DATA_TYPE_UINT64_ARRAY,
533		    sizeof (info->zbc_expected) / sizeof (uint64_t),
534		    (uint64_t *)&info->zbc_expected,
535		    FM_EREPORT_PAYLOAD_ZFS_CKSUM_ACTUAL,
536		    DATA_TYPE_UINT64_ARRAY,
537		    sizeof (info->zbc_actual) / sizeof (uint64_t),
538		    (uint64_t *)&info->zbc_actual,
539		    FM_EREPORT_PAYLOAD_ZFS_CKSUM_ALGO,
540		    DATA_TYPE_STRING,
541		    info->zbc_checksum_name,
542		    NULL);
543
544		if (info->zbc_byteswapped) {
545			fm_payload_set(ereport,
546			    FM_EREPORT_PAYLOAD_ZFS_CKSUM_BYTESWAP,
547			    DATA_TYPE_BOOLEAN, 1,
548			    NULL);
549		}
550	}
551
552	if (badabd == NULL || goodabd == NULL)
553		return (eip);
554
555	ASSERT3U(nui64s, <=, UINT32_MAX);
556	ASSERT3U(size, ==, nui64s * sizeof (uint64_t));
557	ASSERT3U(size, <=, SPA_MAXBLOCKSIZE);
558	ASSERT3U(size, <=, UINT32_MAX);
559
560	good = (const uint64_t *) abd_borrow_buf_copy((abd_t *)goodabd, size);
561	bad = (const uint64_t *) abd_borrow_buf_copy((abd_t *)badabd, size);
562
563	/* build up the range list by comparing the two buffers. */
564	for (idx = 0; idx < nui64s; idx++) {
565		if (good[idx] == bad[idx]) {
566			if (start == -1)
567				continue;
568
569			add_range(eip, start, idx);
570			start = -1;
571		} else {
572			if (start != -1)
573				continue;
574
575			start = idx;
576		}
577	}
578	if (start != -1)
579		add_range(eip, start, idx);
580
581	/* See if it will fit in our inline buffers */
582	inline_size = range_total_size(eip);
583	if (inline_size > ZFM_MAX_INLINE)
584		no_inline = 1;
585
586	/*
587	 * If there is no change and we want to drop if the buffers are
588	 * identical, do so.
589	 */
590	if (inline_size == 0 && drop_if_identical) {
591		kmem_free(eip, sizeof (*eip));
592		abd_return_buf((abd_t *)goodabd, (void *)good, size);
593		abd_return_buf((abd_t *)badabd, (void *)bad, size);
594		return (NULL);
595	}
596
597	/*
598	 * Now walk through the ranges, filling in the details of the
599	 * differences.  Also convert our uint64_t-array offsets to byte
600	 * offsets.
601	 */
602	for (range = 0; range < eip->zei_range_count; range++) {
603		size_t start = eip->zei_ranges[range].zr_start;
604		size_t end = eip->zei_ranges[range].zr_end;
605
606		for (idx = start; idx < end; idx++) {
607			uint64_t set, cleared;
608
609			// bits set in bad, but not in good
610			set = ((~good[idx]) & bad[idx]);
611			// bits set in good, but not in bad
612			cleared = (good[idx] & (~bad[idx]));
613
614			allset |= set;
615			allcleared |= cleared;
616
617			if (!no_inline) {
618				ASSERT3U(offset, <, inline_size);
619				eip->zei_bits_set[offset] = set;
620				eip->zei_bits_cleared[offset] = cleared;
621				offset++;
622			}
623
624			update_histogram(set, eip->zei_histogram_set,
625			    &eip->zei_range_sets[range]);
626			update_histogram(cleared, eip->zei_histogram_cleared,
627			    &eip->zei_range_clears[range]);
628		}
629
630		/* convert to byte offsets */
631		eip->zei_ranges[range].zr_start	*= sizeof (uint64_t);
632		eip->zei_ranges[range].zr_end	*= sizeof (uint64_t);
633	}
634
635	abd_return_buf((abd_t *)goodabd, (void *)good, size);
636	abd_return_buf((abd_t *)badabd, (void *)bad, size);
637
638	eip->zei_allowed_mingap	*= sizeof (uint64_t);
639	inline_size		*= sizeof (uint64_t);
640
641	/* fill in ereport */
642	fm_payload_set(ereport,
643	    FM_EREPORT_PAYLOAD_ZFS_BAD_OFFSET_RANGES,
644	    DATA_TYPE_UINT32_ARRAY, 2 * eip->zei_range_count,
645	    (uint32_t *)eip->zei_ranges,
646	    FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_MIN_GAP,
647	    DATA_TYPE_UINT32, eip->zei_allowed_mingap,
648	    FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_SETS,
649	    DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_sets,
650	    FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_CLEARS,
651	    DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_clears,
652	    NULL);
653
654	if (!no_inline) {
655		fm_payload_set(ereport,
656		    FM_EREPORT_PAYLOAD_ZFS_BAD_SET_BITS,
657		    DATA_TYPE_UINT8_ARRAY,
658		    inline_size, (uint8_t *)eip->zei_bits_set,
659		    FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_BITS,
660		    DATA_TYPE_UINT8_ARRAY,
661		    inline_size, (uint8_t *)eip->zei_bits_cleared,
662		    NULL);
663	} else {
664		fm_payload_set(ereport,
665		    FM_EREPORT_PAYLOAD_ZFS_BAD_SET_HISTOGRAM,
666		    DATA_TYPE_UINT32_ARRAY,
667		    NBBY * sizeof (uint64_t), eip->zei_histogram_set,
668		    FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_HISTOGRAM,
669		    DATA_TYPE_UINT32_ARRAY,
670		    NBBY * sizeof (uint64_t), eip->zei_histogram_cleared,
671		    NULL);
672	}
673	return (eip);
674}
675#endif
676
677void
678zfs_ereport_post(const char *subclass, spa_t *spa, vdev_t *vd,
679    const struct zbookmark_phys *zb, zio_t *zio, uint64_t stateoroffset,
680    uint64_t size)
681{
682#ifdef _KERNEL
683	nvlist_t *ereport = NULL;
684	nvlist_t *detector = NULL;
685
686	zfs_ereport_start(&ereport, &detector, subclass, spa, vd,
687	    zb, zio, stateoroffset, size);
688
689	if (ereport == NULL)
690		return;
691
692	fm_ereport_post(ereport, EVCH_SLEEP);
693
694	fm_nvlist_destroy(ereport, FM_NVA_FREE);
695	fm_nvlist_destroy(detector, FM_NVA_FREE);
696#endif
697}
698
699void
700zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, const zbookmark_phys_t *zb,
701    struct zio *zio, uint64_t offset, uint64_t length, void *arg,
702    zio_bad_cksum_t *info)
703{
704	zio_cksum_report_t *report = kmem_zalloc(sizeof (*report), KM_SLEEP);
705
706	if (zio->io_vsd != NULL)
707		zio->io_vsd_ops->vsd_cksum_report(zio, report, arg);
708	else
709		zio_vsd_default_cksum_report(zio, report, arg);
710
711	/* copy the checksum failure information if it was provided */
712	if (info != NULL) {
713		report->zcr_ckinfo = kmem_zalloc(sizeof (*info), KM_SLEEP);
714		bcopy(info, report->zcr_ckinfo, sizeof (*info));
715	}
716
717	report->zcr_align = 1ULL << vd->vdev_top->vdev_ashift;
718	report->zcr_length = length;
719
720#ifdef _KERNEL
721	zfs_ereport_start(&report->zcr_ereport, &report->zcr_detector,
722	    FM_EREPORT_ZFS_CHECKSUM, spa, vd, zb, zio, offset, length);
723
724	if (report->zcr_ereport == NULL) {
725		report->zcr_free(report->zcr_cbdata, report->zcr_cbinfo);
726		if (report->zcr_ckinfo != NULL) {
727			kmem_free(report->zcr_ckinfo,
728			    sizeof (*report->zcr_ckinfo));
729		}
730		kmem_free(report, sizeof (*report));
731		return;
732	}
733#endif
734
735	mutex_enter(&spa->spa_errlist_lock);
736	report->zcr_next = zio->io_logical->io_cksum_report;
737	zio->io_logical->io_cksum_report = report;
738	mutex_exit(&spa->spa_errlist_lock);
739}
740
741void
742zfs_ereport_finish_checksum(zio_cksum_report_t *report, const abd_t *good_data,
743    const abd_t *bad_data, boolean_t drop_if_identical)
744{
745#ifdef _KERNEL
746	zfs_ecksum_info_t *info = NULL;
747	info = annotate_ecksum(report->zcr_ereport, report->zcr_ckinfo,
748	    good_data, bad_data, report->zcr_length, drop_if_identical);
749
750	if (info != NULL)
751		fm_ereport_post(report->zcr_ereport, EVCH_SLEEP);
752
753	fm_nvlist_destroy(report->zcr_ereport, FM_NVA_FREE);
754	fm_nvlist_destroy(report->zcr_detector, FM_NVA_FREE);
755	report->zcr_ereport = report->zcr_detector = NULL;
756
757	if (info != NULL)
758		kmem_free(info, sizeof (*info));
759#endif
760}
761
762void
763zfs_ereport_free_checksum(zio_cksum_report_t *rpt)
764{
765#ifdef _KERNEL
766	if (rpt->zcr_ereport != NULL) {
767		fm_nvlist_destroy(rpt->zcr_ereport,
768		    FM_NVA_FREE);
769		fm_nvlist_destroy(rpt->zcr_detector,
770		    FM_NVA_FREE);
771	}
772#endif
773	rpt->zcr_free(rpt->zcr_cbdata, rpt->zcr_cbinfo);
774
775	if (rpt->zcr_ckinfo != NULL)
776		kmem_free(rpt->zcr_ckinfo, sizeof (*rpt->zcr_ckinfo));
777
778	kmem_free(rpt, sizeof (*rpt));
779}
780
781void
782zfs_ereport_send_interim_checksum(zio_cksum_report_t *report)
783{
784#ifdef _KERNEL
785	fm_ereport_post(report->zcr_ereport, EVCH_SLEEP);
786#endif
787}
788
789void
790zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd, const zbookmark_phys_t *zb,
791    struct zio *zio, uint64_t offset, uint64_t length,
792    const abd_t *good_data, const abd_t *bad_data, zio_bad_cksum_t *zbc)
793{
794#ifdef _KERNEL
795	nvlist_t *ereport = NULL;
796	nvlist_t *detector = NULL;
797	zfs_ecksum_info_t *info;
798
799	zfs_ereport_start(&ereport, &detector, FM_EREPORT_ZFS_CHECKSUM,
800	    spa, vd, zb, zio, offset, length);
801
802	if (ereport == NULL)
803		return;
804
805	info = annotate_ecksum(ereport, zbc, good_data, bad_data, length,
806	    B_FALSE);
807
808	if (info != NULL)
809		fm_ereport_post(ereport, EVCH_SLEEP);
810
811	fm_nvlist_destroy(ereport, FM_NVA_FREE);
812	fm_nvlist_destroy(detector, FM_NVA_FREE);
813
814	if (info != NULL)
815		kmem_free(info, sizeof (*info));
816#endif
817}
818
819static void
820zfs_post_common(spa_t *spa, vdev_t *vd, const char *name)
821{
822#ifdef _KERNEL
823	nvlist_t *resource;
824	char class[64];
825
826	if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT)
827		return;
828
829	if ((resource = fm_nvlist_create(NULL)) == NULL)
830		return;
831
832	(void) snprintf(class, sizeof (class), "%s.%s.%s", FM_RSRC_RESOURCE,
833	    ZFS_ERROR_CLASS, name);
834	VERIFY(nvlist_add_uint8(resource, FM_VERSION, FM_RSRC_VERSION) == 0);
835	VERIFY(nvlist_add_string(resource, FM_CLASS, class) == 0);
836	VERIFY(nvlist_add_uint64(resource,
837	    FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, spa_guid(spa)) == 0);
838	if (vd)
839		VERIFY(nvlist_add_uint64(resource,
840		    FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vd->vdev_guid) == 0);
841
842	fm_ereport_post(resource, EVCH_SLEEP);
843
844	fm_nvlist_destroy(resource, FM_NVA_FREE);
845#endif
846}
847
848/*
849 * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev
850 * has been removed from the system.  This will cause the DE to ignore any
851 * recent I/O errors, inferring that they are due to the asynchronous device
852 * removal.
853 */
854void
855zfs_post_remove(spa_t *spa, vdev_t *vd)
856{
857	zfs_post_common(spa, vd, FM_RESOURCE_REMOVED);
858}
859
860/*
861 * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool
862 * has the 'autoreplace' property set, and therefore any broken vdevs will be
863 * handled by higher level logic, and no vdev fault should be generated.
864 */
865void
866zfs_post_autoreplace(spa_t *spa, vdev_t *vd)
867{
868	zfs_post_common(spa, vd, FM_RESOURCE_AUTOREPLACE);
869}
870
871/*
872 * The 'resource.fs.zfs.statechange' event is an internal signal that the
873 * given vdev has transitioned its state to DEGRADED or HEALTHY.  This will
874 * cause the retire agent to repair any outstanding fault management cases
875 * open because the device was not found (fault.fs.zfs.device).
876 */
877void
878zfs_post_state_change(spa_t *spa, vdev_t *vd)
879{
880	zfs_post_common(spa, vd, FM_RESOURCE_STATECHANGE);
881}
882