]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | /* | |
9babb374 | 22 | * Copyright 2009 Sun Microsystems, Inc. All rights reserved. |
34dc7c2f BB |
23 | * Use is subject to license terms. |
24 | */ | |
25 | ||
5ffb9d1d GW |
26 | /* |
27 | * Copyright (c) 2012 by Delphix. All rights reserved. | |
28 | */ | |
29 | ||
34dc7c2f BB |
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> | |
428870ff | 35 | #include <sys/zio_checksum.h> |
34dc7c2f BB |
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 | |
b128c09f | 54 | * SPA-wide ENA (Error Numeric Association). |
34dc7c2f BB |
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 | |
b128c09f BB |
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). | |
428870ff BB |
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. | |
34dc7c2f | 103 | */ |
428870ff | 104 | #ifdef _KERNEL |
26685276 BB |
105 | static void |
106 | zfs_zevent_post_cb(nvlist_t *nvl, nvlist_t *detector) | |
107 | { | |
108 | if (nvl) | |
109 | fm_nvlist_destroy(nvl, FM_NVA_FREE); | |
110 | ||
111 | if (detector) | |
112 | fm_nvlist_destroy(detector, FM_NVA_FREE); | |
113 | } | |
114 | ||
428870ff BB |
115 | static void |
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, | |
34dc7c2f BB |
118 | uint64_t stateoroffset, uint64_t size) |
119 | { | |
34dc7c2f | 120 | nvlist_t *ereport, *detector; |
428870ff | 121 | |
34dc7c2f BB |
122 | uint64_t ena; |
123 | char class[64]; | |
124 | ||
125 | /* | |
428870ff BB |
126 | * If we are doing a spa_tryimport() or in recovery mode, |
127 | * ignore errors. | |
34dc7c2f | 128 | */ |
428870ff BB |
129 | if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT || |
130 | spa_load_state(spa) == SPA_LOAD_RECOVER) | |
34dc7c2f BB |
131 | return; |
132 | ||
133 | /* | |
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. | |
137 | */ | |
428870ff | 138 | if (spa_load_state(spa) != SPA_LOAD_NONE && |
34dc7c2f BB |
139 | spa->spa_last_open_failed) |
140 | return; | |
141 | ||
b128c09f BB |
142 | if (zio != NULL) { |
143 | /* | |
144 | * If this is not a read or write zio, ignore the error. This | |
145 | * can occur if the DKIOCFLUSHWRITECACHE ioctl fails. | |
146 | */ | |
147 | if (zio->io_type != ZIO_TYPE_READ && | |
148 | zio->io_type != ZIO_TYPE_WRITE) | |
149 | return; | |
34dc7c2f | 150 | |
9babb374 BB |
151 | if (vd != NULL) { |
152 | /* | |
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 | |
159 | * state. | |
160 | */ | |
428870ff | 161 | if (zio->io_vd == vd && !vdev_accessible(vd, zio)) |
9babb374 BB |
162 | return; |
163 | ||
164 | /* | |
165 | * Ignore checksum errors for reads from DTL regions of | |
166 | * leaf vdevs. | |
167 | */ | |
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)) | |
172 | return; | |
173 | } | |
b128c09f | 174 | } |
34dc7c2f | 175 | |
428870ff BB |
176 | /* |
177 | * For probe failure, we want to avoid posting ereports if we've | |
178 | * already removed the device in the meantime. | |
179 | */ | |
180 | if (vd != NULL && | |
181 | strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) == 0 && | |
182 | (vd->vdev_remove_wanted || vd->vdev_state == VDEV_STATE_REMOVED)) | |
183 | return; | |
184 | ||
34dc7c2f BB |
185 | if ((ereport = fm_nvlist_create(NULL)) == NULL) |
186 | return; | |
187 | ||
188 | if ((detector = fm_nvlist_create(NULL)) == NULL) { | |
189 | fm_nvlist_destroy(ereport, FM_NVA_FREE); | |
190 | return; | |
191 | } | |
192 | ||
193 | /* | |
194 | * Serialize ereport generation | |
195 | */ | |
196 | mutex_enter(&spa->spa_errlist_lock); | |
197 | ||
198 | /* | |
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. | |
202 | */ | |
428870ff | 203 | if (spa_load_state(spa) != SPA_LOAD_NONE) { |
34dc7c2f BB |
204 | if (spa->spa_ena == 0) |
205 | spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1); | |
206 | ena = spa->spa_ena; | |
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; | |
212 | } else { | |
213 | ena = fm_ena_generate(0, FM_ENA_FMT1); | |
214 | } | |
215 | ||
216 | /* | |
217 | * Construct the full class, detector, and other standard FMA fields. | |
218 | */ | |
219 | (void) snprintf(class, sizeof (class), "%s.%s", | |
220 | ZFS_ERROR_CLASS, subclass); | |
221 | ||
222 | fm_fmri_zfs_set(detector, FM_ZFS_SCHEME_VERSION, spa_guid(spa), | |
223 | vd != NULL ? vd->vdev_guid : 0); | |
224 | ||
225 | fm_ereport_set(ereport, FM_EREPORT_VERSION, class, ena, detector, NULL); | |
226 | ||
227 | /* | |
228 | * Construct the per-ereport payload, depending on which parameters are | |
229 | * passed in. | |
230 | */ | |
231 | ||
232 | /* | |
233 | * Generic payload members common to all ereports. | |
34dc7c2f BB |
234 | */ |
235 | fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL, | |
b128c09f | 236 | DATA_TYPE_STRING, spa_name(spa), FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, |
34dc7c2f BB |
237 | DATA_TYPE_UINT64, spa_guid(spa), |
238 | FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, DATA_TYPE_INT32, | |
428870ff | 239 | spa_load_state(spa), NULL); |
b128c09f BB |
240 | |
241 | if (spa != NULL) { | |
242 | fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE, | |
243 | DATA_TYPE_STRING, | |
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, | |
248 | NULL); | |
249 | } | |
34dc7c2f BB |
250 | |
251 | if (vd != NULL) { | |
252 | vdev_t *pvd = vd->vdev_parent; | |
cc92e9d0 | 253 | vdev_queue_t *vq = &vd->vdev_queue; |
34dc7c2f BB |
254 | |
255 | fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, | |
256 | DATA_TYPE_UINT64, vd->vdev_guid, | |
257 | FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, | |
258 | DATA_TYPE_STRING, vd->vdev_ops->vdev_op_type, NULL); | |
9babb374 | 259 | if (vd->vdev_path != NULL) |
34dc7c2f BB |
260 | fm_payload_set(ereport, |
261 | FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH, | |
262 | DATA_TYPE_STRING, vd->vdev_path, NULL); | |
9babb374 | 263 | if (vd->vdev_devid != NULL) |
34dc7c2f BB |
264 | fm_payload_set(ereport, |
265 | FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID, | |
266 | DATA_TYPE_STRING, vd->vdev_devid, NULL); | |
9babb374 BB |
267 | if (vd->vdev_fru != NULL) |
268 | fm_payload_set(ereport, | |
269 | FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU, | |
270 | DATA_TYPE_STRING, vd->vdev_fru, NULL); | |
32a9872b GW |
271 | if (vd->vdev_ashift) |
272 | fm_payload_set(ereport, | |
273 | FM_EREPORT_PAYLOAD_ZFS_VDEV_ASHIFT, | |
274 | DATA_TYPE_UINT64, vd->vdev_ashift, NULL); | |
34dc7c2f | 275 | |
cc92e9d0 GW |
276 | if (vq != NULL) { |
277 | fm_payload_set(ereport, | |
278 | FM_EREPORT_PAYLOAD_ZFS_VDEV_COMP_TS, | |
279 | DATA_TYPE_UINT64, vq->vq_io_complete_ts, NULL); | |
280 | fm_payload_set(ereport, | |
281 | FM_EREPORT_PAYLOAD_ZFS_VDEV_DELTA_TS, | |
282 | DATA_TYPE_UINT64, vq->vq_io_delta_ts, NULL); | |
283 | } | |
284 | ||
34dc7c2f BB |
285 | if (pvd != NULL) { |
286 | fm_payload_set(ereport, | |
287 | FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID, | |
288 | DATA_TYPE_UINT64, pvd->vdev_guid, | |
289 | FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE, | |
290 | DATA_TYPE_STRING, pvd->vdev_ops->vdev_op_type, | |
291 | NULL); | |
292 | if (pvd->vdev_path) | |
293 | fm_payload_set(ereport, | |
294 | FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH, | |
295 | DATA_TYPE_STRING, pvd->vdev_path, NULL); | |
296 | if (pvd->vdev_devid) | |
297 | fm_payload_set(ereport, | |
298 | FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID, | |
299 | DATA_TYPE_STRING, pvd->vdev_devid, NULL); | |
300 | } | |
301 | } | |
302 | ||
303 | if (zio != NULL) { | |
304 | /* | |
305 | * Payload common to all I/Os. | |
306 | */ | |
307 | fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR, | |
308 | DATA_TYPE_INT32, zio->io_error, NULL); | |
312c07ed BB |
309 | fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_FLAGS, |
310 | DATA_TYPE_INT32, zio->io_flags, NULL); | |
9dcb9719 BB |
311 | fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_STAGE, |
312 | DATA_TYPE_UINT32, zio->io_stage, NULL); | |
313 | fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_PIPELINE, | |
314 | DATA_TYPE_UINT32, zio->io_pipeline, NULL); | |
a69052be BB |
315 | fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELAY, |
316 | DATA_TYPE_UINT64, zio->io_delay, NULL); | |
cc92e9d0 GW |
317 | fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_TIMESTAMP, |
318 | DATA_TYPE_UINT64, zio->io_timestamp, NULL); | |
cc92e9d0 GW |
319 | fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELTA, |
320 | DATA_TYPE_UINT64, zio->io_delta, NULL); | |
34dc7c2f BB |
321 | |
322 | /* | |
323 | * If the 'size' parameter is non-zero, it indicates this is a | |
324 | * RAID-Z or other I/O where the physical offset and length are | |
325 | * provided for us, instead of within the zio_t. | |
326 | */ | |
327 | if (vd != NULL) { | |
328 | if (size) | |
329 | fm_payload_set(ereport, | |
330 | FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, | |
331 | DATA_TYPE_UINT64, stateoroffset, | |
332 | FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, | |
333 | DATA_TYPE_UINT64, size, NULL); | |
334 | else | |
335 | fm_payload_set(ereport, | |
336 | FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, | |
337 | DATA_TYPE_UINT64, zio->io_offset, | |
338 | FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, | |
339 | DATA_TYPE_UINT64, zio->io_size, NULL); | |
340 | } | |
341 | ||
342 | /* | |
343 | * Payload for I/Os with corresponding logical information. | |
344 | */ | |
345 | if (zio->io_logical != NULL) | |
346 | fm_payload_set(ereport, | |
347 | FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET, | |
348 | DATA_TYPE_UINT64, | |
349 | zio->io_logical->io_bookmark.zb_objset, | |
350 | FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT, | |
351 | DATA_TYPE_UINT64, | |
352 | zio->io_logical->io_bookmark.zb_object, | |
353 | FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL, | |
354 | DATA_TYPE_INT64, | |
355 | zio->io_logical->io_bookmark.zb_level, | |
356 | FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID, | |
357 | DATA_TYPE_UINT64, | |
358 | zio->io_logical->io_bookmark.zb_blkid, NULL); | |
359 | } else if (vd != NULL) { | |
360 | /* | |
361 | * If we have a vdev but no zio, this is a device fault, and the | |
362 | * 'stateoroffset' parameter indicates the previous state of the | |
363 | * vdev. | |
364 | */ | |
365 | fm_payload_set(ereport, | |
366 | FM_EREPORT_PAYLOAD_ZFS_PREV_STATE, | |
367 | DATA_TYPE_UINT64, stateoroffset, NULL); | |
368 | } | |
428870ff | 369 | |
34dc7c2f BB |
370 | mutex_exit(&spa->spa_errlist_lock); |
371 | ||
428870ff BB |
372 | *ereport_out = ereport; |
373 | *detector_out = detector; | |
374 | } | |
375 | ||
376 | /* if it's <= 128 bytes, save the corruption directly */ | |
377 | #define ZFM_MAX_INLINE (128 / sizeof (uint64_t)) | |
378 | ||
379 | #define MAX_RANGES 16 | |
380 | ||
381 | typedef struct zfs_ecksum_info { | |
382 | /* histograms of set and cleared bits by bit number in a 64-bit word */ | |
383 | uint16_t zei_histogram_set[sizeof (uint64_t) * NBBY]; | |
384 | uint16_t zei_histogram_cleared[sizeof (uint64_t) * NBBY]; | |
385 | ||
386 | /* inline arrays of bits set and cleared. */ | |
387 | uint64_t zei_bits_set[ZFM_MAX_INLINE]; | |
388 | uint64_t zei_bits_cleared[ZFM_MAX_INLINE]; | |
389 | ||
390 | /* | |
391 | * for each range, the number of bits set and cleared. The Hamming | |
392 | * distance between the good and bad buffers is the sum of them all. | |
393 | */ | |
394 | uint32_t zei_range_sets[MAX_RANGES]; | |
395 | uint32_t zei_range_clears[MAX_RANGES]; | |
396 | ||
397 | struct zei_ranges { | |
398 | uint32_t zr_start; | |
399 | uint32_t zr_end; | |
400 | } zei_ranges[MAX_RANGES]; | |
401 | ||
402 | size_t zei_range_count; | |
403 | uint32_t zei_mingap; | |
404 | uint32_t zei_allowed_mingap; | |
405 | ||
406 | } zfs_ecksum_info_t; | |
407 | ||
408 | static void | |
409 | update_histogram(uint64_t value_arg, uint16_t *hist, uint32_t *count) | |
410 | { | |
411 | size_t i; | |
412 | size_t bits = 0; | |
413 | uint64_t value = BE_64(value_arg); | |
414 | ||
415 | /* We store the bits in big-endian (largest-first) order */ | |
416 | for (i = 0; i < 64; i++) { | |
417 | if (value & (1ull << i)) { | |
418 | hist[63 - i]++; | |
419 | ++bits; | |
420 | } | |
421 | } | |
422 | /* update the count of bits changed */ | |
423 | *count += bits; | |
424 | } | |
425 | ||
426 | /* | |
427 | * We've now filled up the range array, and need to increase "mingap" and | |
428 | * shrink the range list accordingly. zei_mingap is always the smallest | |
429 | * distance between array entries, so we set the new_allowed_gap to be | |
430 | * one greater than that. We then go through the list, joining together | |
431 | * any ranges which are closer than the new_allowed_gap. | |
432 | * | |
433 | * By construction, there will be at least one. We also update zei_mingap | |
434 | * to the new smallest gap, to prepare for our next invocation. | |
435 | */ | |
436 | static void | |
26685276 | 437 | zei_shrink_ranges(zfs_ecksum_info_t *eip) |
428870ff BB |
438 | { |
439 | uint32_t mingap = UINT32_MAX; | |
440 | uint32_t new_allowed_gap = eip->zei_mingap + 1; | |
441 | ||
442 | size_t idx, output; | |
443 | size_t max = eip->zei_range_count; | |
444 | ||
445 | struct zei_ranges *r = eip->zei_ranges; | |
446 | ||
447 | ASSERT3U(eip->zei_range_count, >, 0); | |
448 | ASSERT3U(eip->zei_range_count, <=, MAX_RANGES); | |
449 | ||
450 | output = idx = 0; | |
451 | while (idx < max - 1) { | |
452 | uint32_t start = r[idx].zr_start; | |
453 | uint32_t end = r[idx].zr_end; | |
454 | ||
455 | while (idx < max - 1) { | |
26685276 | 456 | uint32_t nstart, nend, gap; |
428870ff | 457 | |
26685276 BB |
458 | idx++; |
459 | nstart = r[idx].zr_start; | |
460 | nend = r[idx].zr_end; | |
428870ff | 461 | |
26685276 | 462 | gap = nstart - end; |
428870ff BB |
463 | if (gap < new_allowed_gap) { |
464 | end = nend; | |
465 | continue; | |
466 | } | |
467 | if (gap < mingap) | |
468 | mingap = gap; | |
469 | break; | |
470 | } | |
471 | r[output].zr_start = start; | |
472 | r[output].zr_end = end; | |
473 | output++; | |
474 | } | |
475 | ASSERT3U(output, <, eip->zei_range_count); | |
476 | eip->zei_range_count = output; | |
477 | eip->zei_mingap = mingap; | |
478 | eip->zei_allowed_mingap = new_allowed_gap; | |
479 | } | |
480 | ||
481 | static void | |
26685276 | 482 | zei_add_range(zfs_ecksum_info_t *eip, int start, int end) |
428870ff BB |
483 | { |
484 | struct zei_ranges *r = eip->zei_ranges; | |
485 | size_t count = eip->zei_range_count; | |
486 | ||
487 | if (count >= MAX_RANGES) { | |
26685276 | 488 | zei_shrink_ranges(eip); |
428870ff BB |
489 | count = eip->zei_range_count; |
490 | } | |
491 | if (count == 0) { | |
492 | eip->zei_mingap = UINT32_MAX; | |
493 | eip->zei_allowed_mingap = 1; | |
494 | } else { | |
495 | int gap = start - r[count - 1].zr_end; | |
496 | ||
497 | if (gap < eip->zei_allowed_mingap) { | |
498 | r[count - 1].zr_end = end; | |
499 | return; | |
500 | } | |
501 | if (gap < eip->zei_mingap) | |
502 | eip->zei_mingap = gap; | |
503 | } | |
504 | r[count].zr_start = start; | |
505 | r[count].zr_end = end; | |
506 | eip->zei_range_count++; | |
507 | } | |
508 | ||
509 | static size_t | |
26685276 | 510 | zei_range_total_size(zfs_ecksum_info_t *eip) |
428870ff BB |
511 | { |
512 | struct zei_ranges *r = eip->zei_ranges; | |
513 | size_t count = eip->zei_range_count; | |
514 | size_t result = 0; | |
515 | size_t idx; | |
516 | ||
517 | for (idx = 0; idx < count; idx++) | |
518 | result += (r[idx].zr_end - r[idx].zr_start); | |
519 | ||
520 | return (result); | |
521 | } | |
522 | ||
523 | static zfs_ecksum_info_t * | |
524 | annotate_ecksum(nvlist_t *ereport, zio_bad_cksum_t *info, | |
525 | const uint8_t *goodbuf, const uint8_t *badbuf, size_t size, | |
526 | boolean_t drop_if_identical) | |
527 | { | |
528 | const uint64_t *good = (const uint64_t *)goodbuf; | |
529 | const uint64_t *bad = (const uint64_t *)badbuf; | |
530 | ||
531 | uint64_t allset = 0; | |
532 | uint64_t allcleared = 0; | |
533 | ||
534 | size_t nui64s = size / sizeof (uint64_t); | |
535 | ||
536 | size_t inline_size; | |
537 | int no_inline = 0; | |
538 | size_t idx; | |
539 | size_t range; | |
540 | ||
541 | size_t offset = 0; | |
542 | ssize_t start = -1; | |
543 | ||
b8d06fca | 544 | zfs_ecksum_info_t *eip = kmem_zalloc(sizeof (*eip), KM_PUSHPAGE); |
428870ff BB |
545 | |
546 | /* don't do any annotation for injected checksum errors */ | |
547 | if (info != NULL && info->zbc_injected) | |
548 | return (eip); | |
549 | ||
550 | if (info != NULL && info->zbc_has_cksum) { | |
551 | fm_payload_set(ereport, | |
552 | FM_EREPORT_PAYLOAD_ZFS_CKSUM_EXPECTED, | |
553 | DATA_TYPE_UINT64_ARRAY, | |
554 | sizeof (info->zbc_expected) / sizeof (uint64_t), | |
555 | (uint64_t *)&info->zbc_expected, | |
556 | FM_EREPORT_PAYLOAD_ZFS_CKSUM_ACTUAL, | |
557 | DATA_TYPE_UINT64_ARRAY, | |
558 | sizeof (info->zbc_actual) / sizeof (uint64_t), | |
559 | (uint64_t *)&info->zbc_actual, | |
560 | FM_EREPORT_PAYLOAD_ZFS_CKSUM_ALGO, | |
561 | DATA_TYPE_STRING, | |
562 | info->zbc_checksum_name, | |
563 | NULL); | |
564 | ||
565 | if (info->zbc_byteswapped) { | |
566 | fm_payload_set(ereport, | |
567 | FM_EREPORT_PAYLOAD_ZFS_CKSUM_BYTESWAP, | |
568 | DATA_TYPE_BOOLEAN, 1, | |
569 | NULL); | |
570 | } | |
571 | } | |
572 | ||
573 | if (badbuf == NULL || goodbuf == NULL) | |
574 | return (eip); | |
575 | ||
576 | ASSERT3U(nui64s, <=, UINT16_MAX); | |
577 | ASSERT3U(size, ==, nui64s * sizeof (uint64_t)); | |
578 | ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); | |
579 | ASSERT3U(size, <=, UINT32_MAX); | |
580 | ||
581 | /* build up the range list by comparing the two buffers. */ | |
582 | for (idx = 0; idx < nui64s; idx++) { | |
583 | if (good[idx] == bad[idx]) { | |
584 | if (start == -1) | |
585 | continue; | |
586 | ||
26685276 | 587 | zei_add_range(eip, start, idx); |
428870ff BB |
588 | start = -1; |
589 | } else { | |
590 | if (start != -1) | |
591 | continue; | |
592 | ||
593 | start = idx; | |
594 | } | |
595 | } | |
596 | if (start != -1) | |
26685276 | 597 | zei_add_range(eip, start, idx); |
428870ff BB |
598 | |
599 | /* See if it will fit in our inline buffers */ | |
26685276 | 600 | inline_size = zei_range_total_size(eip); |
428870ff BB |
601 | if (inline_size > ZFM_MAX_INLINE) |
602 | no_inline = 1; | |
603 | ||
604 | /* | |
605 | * If there is no change and we want to drop if the buffers are | |
606 | * identical, do so. | |
607 | */ | |
608 | if (inline_size == 0 && drop_if_identical) { | |
609 | kmem_free(eip, sizeof (*eip)); | |
610 | return (NULL); | |
611 | } | |
612 | ||
613 | /* | |
614 | * Now walk through the ranges, filling in the details of the | |
615 | * differences. Also convert our uint64_t-array offsets to byte | |
616 | * offsets. | |
617 | */ | |
618 | for (range = 0; range < eip->zei_range_count; range++) { | |
619 | size_t start = eip->zei_ranges[range].zr_start; | |
620 | size_t end = eip->zei_ranges[range].zr_end; | |
621 | ||
622 | for (idx = start; idx < end; idx++) { | |
623 | uint64_t set, cleared; | |
624 | ||
625 | // bits set in bad, but not in good | |
626 | set = ((~good[idx]) & bad[idx]); | |
627 | // bits set in good, but not in bad | |
628 | cleared = (good[idx] & (~bad[idx])); | |
629 | ||
630 | allset |= set; | |
631 | allcleared |= cleared; | |
632 | ||
633 | if (!no_inline) { | |
634 | ASSERT3U(offset, <, inline_size); | |
635 | eip->zei_bits_set[offset] = set; | |
636 | eip->zei_bits_cleared[offset] = cleared; | |
637 | offset++; | |
638 | } | |
639 | ||
640 | update_histogram(set, eip->zei_histogram_set, | |
641 | &eip->zei_range_sets[range]); | |
642 | update_histogram(cleared, eip->zei_histogram_cleared, | |
643 | &eip->zei_range_clears[range]); | |
644 | } | |
645 | ||
646 | /* convert to byte offsets */ | |
647 | eip->zei_ranges[range].zr_start *= sizeof (uint64_t); | |
648 | eip->zei_ranges[range].zr_end *= sizeof (uint64_t); | |
649 | } | |
650 | eip->zei_allowed_mingap *= sizeof (uint64_t); | |
651 | inline_size *= sizeof (uint64_t); | |
652 | ||
653 | /* fill in ereport */ | |
654 | fm_payload_set(ereport, | |
655 | FM_EREPORT_PAYLOAD_ZFS_BAD_OFFSET_RANGES, | |
656 | DATA_TYPE_UINT32_ARRAY, 2 * eip->zei_range_count, | |
657 | (uint32_t *)eip->zei_ranges, | |
658 | FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_MIN_GAP, | |
659 | DATA_TYPE_UINT32, eip->zei_allowed_mingap, | |
660 | FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_SETS, | |
661 | DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_sets, | |
662 | FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_CLEARS, | |
663 | DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_clears, | |
664 | NULL); | |
665 | ||
666 | if (!no_inline) { | |
667 | fm_payload_set(ereport, | |
668 | FM_EREPORT_PAYLOAD_ZFS_BAD_SET_BITS, | |
669 | DATA_TYPE_UINT8_ARRAY, | |
670 | inline_size, (uint8_t *)eip->zei_bits_set, | |
671 | FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_BITS, | |
672 | DATA_TYPE_UINT8_ARRAY, | |
673 | inline_size, (uint8_t *)eip->zei_bits_cleared, | |
674 | NULL); | |
675 | } else { | |
676 | fm_payload_set(ereport, | |
677 | FM_EREPORT_PAYLOAD_ZFS_BAD_SET_HISTOGRAM, | |
678 | DATA_TYPE_UINT16_ARRAY, | |
679 | NBBY * sizeof (uint64_t), eip->zei_histogram_set, | |
680 | FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_HISTOGRAM, | |
681 | DATA_TYPE_UINT16_ARRAY, | |
682 | NBBY * sizeof (uint64_t), eip->zei_histogram_cleared, | |
683 | NULL); | |
684 | } | |
685 | return (eip); | |
686 | } | |
687 | #endif | |
688 | ||
689 | void | |
690 | zfs_ereport_post(const char *subclass, spa_t *spa, vdev_t *vd, zio_t *zio, | |
691 | uint64_t stateoroffset, uint64_t size) | |
692 | { | |
693 | #ifdef _KERNEL | |
694 | nvlist_t *ereport = NULL; | |
695 | nvlist_t *detector = NULL; | |
696 | ||
697 | zfs_ereport_start(&ereport, &detector, | |
698 | subclass, spa, vd, zio, stateoroffset, size); | |
699 | ||
700 | if (ereport == NULL) | |
701 | return; | |
702 | ||
26685276 BB |
703 | /* Cleanup is handled by the callback function */ |
704 | zfs_zevent_post(ereport, detector, zfs_zevent_post_cb); | |
34dc7c2f BB |
705 | #endif |
706 | } | |
707 | ||
428870ff BB |
708 | void |
709 | zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, | |
710 | struct zio *zio, uint64_t offset, uint64_t length, void *arg, | |
711 | zio_bad_cksum_t *info) | |
712 | { | |
b8d06fca | 713 | zio_cksum_report_t *report = kmem_zalloc(sizeof (*report), KM_PUSHPAGE); |
428870ff BB |
714 | |
715 | if (zio->io_vsd != NULL) | |
716 | zio->io_vsd_ops->vsd_cksum_report(zio, report, arg); | |
717 | else | |
718 | zio_vsd_default_cksum_report(zio, report, arg); | |
719 | ||
720 | /* copy the checksum failure information if it was provided */ | |
721 | if (info != NULL) { | |
b8d06fca | 722 | report->zcr_ckinfo = kmem_zalloc(sizeof (*info), KM_PUSHPAGE); |
428870ff BB |
723 | bcopy(info, report->zcr_ckinfo, sizeof (*info)); |
724 | } | |
725 | ||
726 | report->zcr_align = 1ULL << vd->vdev_top->vdev_ashift; | |
727 | report->zcr_length = length; | |
728 | ||
729 | #ifdef _KERNEL | |
730 | zfs_ereport_start(&report->zcr_ereport, &report->zcr_detector, | |
731 | FM_EREPORT_ZFS_CHECKSUM, spa, vd, zio, offset, length); | |
732 | ||
733 | if (report->zcr_ereport == NULL) { | |
734 | report->zcr_free(report->zcr_cbdata, report->zcr_cbinfo); | |
5ffb9d1d GW |
735 | if (report->zcr_ckinfo != NULL) { |
736 | kmem_free(report->zcr_ckinfo, | |
737 | sizeof (*report->zcr_ckinfo)); | |
738 | } | |
428870ff BB |
739 | kmem_free(report, sizeof (*report)); |
740 | return; | |
741 | } | |
742 | #endif | |
743 | ||
744 | mutex_enter(&spa->spa_errlist_lock); | |
745 | report->zcr_next = zio->io_logical->io_cksum_report; | |
746 | zio->io_logical->io_cksum_report = report; | |
747 | mutex_exit(&spa->spa_errlist_lock); | |
748 | } | |
749 | ||
750 | void | |
751 | zfs_ereport_finish_checksum(zio_cksum_report_t *report, | |
752 | const void *good_data, const void *bad_data, boolean_t drop_if_identical) | |
753 | { | |
754 | #ifdef _KERNEL | |
755 | zfs_ecksum_info_t *info = NULL; | |
756 | info = annotate_ecksum(report->zcr_ereport, report->zcr_ckinfo, | |
757 | good_data, bad_data, report->zcr_length, drop_if_identical); | |
758 | ||
759 | if (info != NULL) | |
26685276 BB |
760 | zfs_zevent_post(report->zcr_ereport, |
761 | report->zcr_detector, zfs_zevent_post_cb); | |
428870ff | 762 | |
428870ff | 763 | report->zcr_ereport = report->zcr_detector = NULL; |
428870ff BB |
764 | if (info != NULL) |
765 | kmem_free(info, sizeof (*info)); | |
766 | #endif | |
767 | } | |
768 | ||
769 | void | |
770 | zfs_ereport_free_checksum(zio_cksum_report_t *rpt) | |
771 | { | |
772 | #ifdef _KERNEL | |
773 | if (rpt->zcr_ereport != NULL) { | |
774 | fm_nvlist_destroy(rpt->zcr_ereport, | |
775 | FM_NVA_FREE); | |
776 | fm_nvlist_destroy(rpt->zcr_detector, | |
777 | FM_NVA_FREE); | |
778 | } | |
779 | #endif | |
780 | rpt->zcr_free(rpt->zcr_cbdata, rpt->zcr_cbinfo); | |
781 | ||
782 | if (rpt->zcr_ckinfo != NULL) | |
783 | kmem_free(rpt->zcr_ckinfo, sizeof (*rpt->zcr_ckinfo)); | |
784 | ||
785 | kmem_free(rpt, sizeof (*rpt)); | |
786 | } | |
787 | ||
788 | void | |
789 | zfs_ereport_send_interim_checksum(zio_cksum_report_t *report) | |
790 | { | |
791 | #ifdef _KERNEL | |
26685276 | 792 | zfs_zevent_post(report->zcr_ereport, report->zcr_detector, NULL); |
428870ff BB |
793 | #endif |
794 | } | |
795 | ||
796 | void | |
797 | zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd, | |
798 | struct zio *zio, uint64_t offset, uint64_t length, | |
799 | const void *good_data, const void *bad_data, zio_bad_cksum_t *zbc) | |
800 | { | |
801 | #ifdef _KERNEL | |
802 | nvlist_t *ereport = NULL; | |
803 | nvlist_t *detector = NULL; | |
804 | zfs_ecksum_info_t *info; | |
805 | ||
806 | zfs_ereport_start(&ereport, &detector, | |
807 | FM_EREPORT_ZFS_CHECKSUM, spa, vd, zio, offset, length); | |
808 | ||
809 | if (ereport == NULL) | |
810 | return; | |
811 | ||
812 | info = annotate_ecksum(ereport, zbc, good_data, bad_data, length, | |
813 | B_FALSE); | |
814 | ||
26685276 BB |
815 | if (info != NULL) { |
816 | zfs_zevent_post(ereport, detector, zfs_zevent_post_cb); | |
428870ff | 817 | kmem_free(info, sizeof (*info)); |
26685276 | 818 | } |
428870ff BB |
819 | #endif |
820 | } | |
821 | ||
34dc7c2f BB |
822 | static void |
823 | zfs_post_common(spa_t *spa, vdev_t *vd, const char *name) | |
824 | { | |
825 | #ifdef _KERNEL | |
826 | nvlist_t *resource; | |
827 | char class[64]; | |
828 | ||
428870ff BB |
829 | if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT) |
830 | return; | |
831 | ||
34dc7c2f BB |
832 | if ((resource = fm_nvlist_create(NULL)) == NULL) |
833 | return; | |
834 | ||
835 | (void) snprintf(class, sizeof (class), "%s.%s.%s", FM_RSRC_RESOURCE, | |
836 | ZFS_ERROR_CLASS, name); | |
837 | VERIFY(nvlist_add_uint8(resource, FM_VERSION, FM_RSRC_VERSION) == 0); | |
838 | VERIFY(nvlist_add_string(resource, FM_CLASS, class) == 0); | |
839 | VERIFY(nvlist_add_uint64(resource, | |
840 | FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, spa_guid(spa)) == 0); | |
26685276 | 841 | if (vd) { |
34dc7c2f BB |
842 | VERIFY(nvlist_add_uint64(resource, |
843 | FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vd->vdev_guid) == 0); | |
26685276 BB |
844 | VERIFY(nvlist_add_uint64(resource, |
845 | FM_EREPORT_PAYLOAD_ZFS_VDEV_STATE, vd->vdev_state) == 0); | |
846 | } | |
34dc7c2f | 847 | |
26685276 | 848 | zfs_zevent_post(resource, NULL, zfs_zevent_post_cb); |
34dc7c2f BB |
849 | #endif |
850 | } | |
851 | ||
34dc7c2f BB |
852 | /* |
853 | * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev | |
854 | * has been removed from the system. This will cause the DE to ignore any | |
855 | * recent I/O errors, inferring that they are due to the asynchronous device | |
856 | * removal. | |
857 | */ | |
858 | void | |
859 | zfs_post_remove(spa_t *spa, vdev_t *vd) | |
860 | { | |
26685276 | 861 | zfs_post_common(spa, vd, FM_EREPORT_RESOURCE_REMOVED); |
34dc7c2f BB |
862 | } |
863 | ||
864 | /* | |
865 | * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool | |
866 | * has the 'autoreplace' property set, and therefore any broken vdevs will be | |
867 | * handled by higher level logic, and no vdev fault should be generated. | |
868 | */ | |
869 | void | |
870 | zfs_post_autoreplace(spa_t *spa, vdev_t *vd) | |
871 | { | |
26685276 | 872 | zfs_post_common(spa, vd, FM_EREPORT_RESOURCE_AUTOREPLACE); |
34dc7c2f | 873 | } |
428870ff BB |
874 | |
875 | /* | |
876 | * The 'resource.fs.zfs.statechange' event is an internal signal that the | |
877 | * given vdev has transitioned its state to DEGRADED or HEALTHY. This will | |
878 | * cause the retire agent to repair any outstanding fault management cases | |
879 | * open because the device was not found (fault.fs.zfs.device). | |
880 | */ | |
881 | void | |
882 | zfs_post_state_change(spa_t *spa, vdev_t *vd) | |
883 | { | |
26685276 | 884 | zfs_post_common(spa, vd, FM_EREPORT_RESOURCE_STATECHANGE); |
428870ff | 885 | } |
26685276 BB |
886 | |
887 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
888 | EXPORT_SYMBOL(zfs_ereport_post); | |
889 | EXPORT_SYMBOL(zfs_ereport_post_checksum); | |
890 | EXPORT_SYMBOL(zfs_post_remove); | |
891 | EXPORT_SYMBOL(zfs_post_autoreplace); | |
892 | EXPORT_SYMBOL(zfs_post_state_change); | |
893 | #endif /* _KERNEL */ |