module/zfs/zfs_fm.c

   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
 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
 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,
 118     uint64_t stateoroffset, uint64_t size)
 119 {
 120         nvlist_t *ereport, *detector;
 121
 122         uint64_t ena;
 123         char class[64];
 124
 125         /*
 126          * If we are doing a spa_tryimport() or in recovery mode,
 127          * ignore errors.
 128          */
 129         if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT ||
 130             spa_load_state(spa) == SPA_LOAD_RECOVER)
 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          */
 138         if (spa_load_state(spa) != SPA_LOAD_NONE &&
 139             spa->spa_last_open_failed)
 140                 return;
 141
 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;
 150
 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                          */
 161                         if (zio->io_vd == vd && !vdev_accessible(vd, zio))
 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                 }
 174         }
 175
 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
 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          */
 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);
 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.
 234          */
 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);
 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         }
 250
 251         if (vd != NULL) {
 252                 vdev_t *pvd = vd->vdev_parent;
 253                 vdev_queue_t *vq = &vd->vdev_queue;
 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);
 259                 if (vd->vdev_path != NULL)
 260                         fm_payload_set(ereport,
 261                             FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH,
 262                             DATA_TYPE_STRING, vd->vdev_path, NULL);
 263                 if (vd->vdev_devid != NULL)
 264                         fm_payload_set(ereport,
 265                             FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID,
 266                             DATA_TYPE_STRING, vd->vdev_devid, NULL);
 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);
 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);
 275
 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
 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);
 309                 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_FLAGS,
 310                     DATA_TYPE_INT32, zio->io_flags, NULL);
 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);
 315                 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELAY,
 316                     DATA_TYPE_UINT64, zio->io_delay, NULL);
 317                 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_TIMESTAMP,
 318                     DATA_TYPE_UINT64, zio->io_timestamp, NULL);
 319                 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_DEADLINE,
 320                     DATA_TYPE_UINT64, zio->io_deadline, NULL);
 321                 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELTA,
 322                     DATA_TYPE_UINT64, zio->io_delta, NULL);
 323
 324                 /*
 325                  * If the 'size' parameter is non-zero, it indicates this is a
 326                  * RAID-Z or other I/O where the physical offset and length are
 327                  * provided for us, instead of within the zio_t.
 328                  */
 329                 if (vd != NULL) {
 330                         if (size)
 331                                 fm_payload_set(ereport,
 332                                     FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET,
 333                                     DATA_TYPE_UINT64, stateoroffset,
 334                                     FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE,
 335                                     DATA_TYPE_UINT64, size, NULL);
 336                         else
 337                                 fm_payload_set(ereport,
 338                                     FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET,
 339                                     DATA_TYPE_UINT64, zio->io_offset,
 340                                     FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE,
 341                                     DATA_TYPE_UINT64, zio->io_size, NULL);
 342                 }
 343
 344                 /*
 345                  * Payload for I/Os with corresponding logical information.
 346                  */
 347                 if (zio->io_logical != NULL)
 348                         fm_payload_set(ereport,
 349                             FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET,
 350                             DATA_TYPE_UINT64,
 351                             zio->io_logical->io_bookmark.zb_objset,
 352                             FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT,
 353                             DATA_TYPE_UINT64,
 354                             zio->io_logical->io_bookmark.zb_object,
 355                             FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL,
 356                             DATA_TYPE_INT64,
 357                             zio->io_logical->io_bookmark.zb_level,
 358                             FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID,
 359                             DATA_TYPE_UINT64,
 360                             zio->io_logical->io_bookmark.zb_blkid, NULL);
 361         } else if (vd != NULL) {
 362                 /*
 363                  * If we have a vdev but no zio, this is a device fault, and the
 364                  * 'stateoroffset' parameter indicates the previous state of the
 365                  * vdev.
 366                  */
 367                 fm_payload_set(ereport,
 368                     FM_EREPORT_PAYLOAD_ZFS_PREV_STATE,
 369                     DATA_TYPE_UINT64, stateoroffset, NULL);
 370         }
 371
 372         mutex_exit(&spa->spa_errlist_lock);
 373
 374         *ereport_out = ereport;
 375         *detector_out = detector;
 376 }
 377
 378 /* if it's <= 128 bytes, save the corruption directly */
 379 #define ZFM_MAX_INLINE          (128 / sizeof (uint64_t))
 380
 381 #define MAX_RANGES              16
 382
 383 typedef struct zfs_ecksum_info {
 384         /* histograms of set and cleared bits by bit number in a 64-bit word */
 385         uint16_t zei_histogram_set[sizeof (uint64_t) * NBBY];
 386         uint16_t zei_histogram_cleared[sizeof (uint64_t) * NBBY];
 387
 388         /* inline arrays of bits set and cleared. */
 389         uint64_t zei_bits_set[ZFM_MAX_INLINE];
 390         uint64_t zei_bits_cleared[ZFM_MAX_INLINE];
 391
 392         /*
 393          * for each range, the number of bits set and cleared.  The Hamming
 394          * distance between the good and bad buffers is the sum of them all.
 395          */
 396         uint32_t zei_range_sets[MAX_RANGES];
 397         uint32_t zei_range_clears[MAX_RANGES];
 398
 399         struct zei_ranges {
 400                 uint32_t        zr_start;
 401                 uint32_t        zr_end;
 402         } zei_ranges[MAX_RANGES];
 403
 404         size_t  zei_range_count;
 405         uint32_t zei_mingap;
 406         uint32_t zei_allowed_mingap;
 407
 408 } zfs_ecksum_info_t;
 409
 410 static void
 411 update_histogram(uint64_t value_arg, uint16_t *hist, uint32_t *count)
 412 {
 413         size_t i;
 414         size_t bits = 0;
 415         uint64_t value = BE_64(value_arg);
 416
 417         /* We store the bits in big-endian (largest-first) order */
 418         for (i = 0; i < 64; i++) {
 419                 if (value & (1ull << i)) {
 420                         hist[63 - i]++;
 421                         ++bits;
 422                 }
 423         }
 424         /* update the count of bits changed */
 425         *count += bits;
 426 }
 427
 428 /*
 429  * We've now filled up the range array, and need to increase "mingap" and
 430  * shrink the range list accordingly.  zei_mingap is always the smallest
 431  * distance between array entries, so we set the new_allowed_gap to be
 432  * one greater than that.  We then go through the list, joining together
 433  * any ranges which are closer than the new_allowed_gap.
 434  *
 435  * By construction, there will be at least one.  We also update zei_mingap
 436  * to the new smallest gap, to prepare for our next invocation.
 437  */
 438 static void
 439 zei_shrink_ranges(zfs_ecksum_info_t *eip)
 440 {
 441         uint32_t mingap = UINT32_MAX;
 442         uint32_t new_allowed_gap = eip->zei_mingap + 1;
 443
 444         size_t idx, output;
 445         size_t max = eip->zei_range_count;
 446
 447         struct zei_ranges *r = eip->zei_ranges;
 448
 449         ASSERT3U(eip->zei_range_count, >, 0);
 450         ASSERT3U(eip->zei_range_count, <=, MAX_RANGES);
 451
 452         output = idx = 0;
 453         while (idx < max - 1) {
 454                 uint32_t start = r[idx].zr_start;
 455                 uint32_t end = r[idx].zr_end;
 456
 457                 while (idx < max - 1) {
 458                         uint32_t nstart, nend, gap;
 459
 460                         idx++;
 461                         nstart = r[idx].zr_start;
 462                         nend = r[idx].zr_end;
 463
 464                         gap = nstart - end;
 465                         if (gap < new_allowed_gap) {
 466                                 end = nend;
 467                                 continue;
 468                         }
 469                         if (gap < mingap)
 470                                 mingap = gap;
 471                         break;
 472                 }
 473                 r[output].zr_start = start;
 474                 r[output].zr_end = end;
 475                 output++;
 476         }
 477         ASSERT3U(output, <, eip->zei_range_count);
 478         eip->zei_range_count = output;
 479         eip->zei_mingap = mingap;
 480         eip->zei_allowed_mingap = new_allowed_gap;
 481 }
 482
 483 static void
 484 zei_add_range(zfs_ecksum_info_t *eip, int start, int end)
 485 {
 486         struct zei_ranges *r = eip->zei_ranges;
 487         size_t count = eip->zei_range_count;
 488
 489         if (count >= MAX_RANGES) {
 490                 zei_shrink_ranges(eip);
 491                 count = eip->zei_range_count;
 492         }
 493         if (count == 0) {
 494                 eip->zei_mingap = UINT32_MAX;
 495                 eip->zei_allowed_mingap = 1;
 496         } else {
 497                 int gap = start - r[count - 1].zr_end;
 498
 499                 if (gap < eip->zei_allowed_mingap) {
 500                         r[count - 1].zr_end = end;
 501                         return;
 502                 }
 503                 if (gap < eip->zei_mingap)
 504                         eip->zei_mingap = gap;
 505         }
 506         r[count].zr_start = start;
 507         r[count].zr_end = end;
 508         eip->zei_range_count++;
 509 }
 510
 511 static size_t
 512 zei_range_total_size(zfs_ecksum_info_t *eip)
 513 {
 514         struct zei_ranges *r = eip->zei_ranges;
 515         size_t count = eip->zei_range_count;
 516         size_t result = 0;
 517         size_t idx;
 518
 519         for (idx = 0; idx < count; idx++)
 520                 result += (r[idx].zr_end - r[idx].zr_start);
 521
 522         return (result);
 523 }
 524
 525 static zfs_ecksum_info_t *
 526 annotate_ecksum(nvlist_t *ereport, zio_bad_cksum_t *info,
 527     const uint8_t *goodbuf, const uint8_t *badbuf, size_t size,
 528     boolean_t drop_if_identical)
 529 {
 530         const uint64_t *good = (const uint64_t *)goodbuf;
 531         const uint64_t *bad = (const uint64_t *)badbuf;
 532
 533         uint64_t allset = 0;
 534         uint64_t allcleared = 0;
 535
 536         size_t nui64s = size / sizeof (uint64_t);
 537
 538         size_t inline_size;
 539         int no_inline = 0;
 540         size_t idx;
 541         size_t range;
 542
 543         size_t offset = 0;
 544         ssize_t start = -1;
 545
 546         zfs_ecksum_info_t *eip = kmem_zalloc(sizeof (*eip), KM_PUSHPAGE);
 547
 548         /* don't do any annotation for injected checksum errors */
 549         if (info != NULL && info->zbc_injected)
 550                 return (eip);
 551
 552         if (info != NULL && info->zbc_has_cksum) {
 553                 fm_payload_set(ereport,
 554                     FM_EREPORT_PAYLOAD_ZFS_CKSUM_EXPECTED,
 555                     DATA_TYPE_UINT64_ARRAY,
 556                     sizeof (info->zbc_expected) / sizeof (uint64_t),
 557                     (uint64_t *)&info->zbc_expected,
 558                     FM_EREPORT_PAYLOAD_ZFS_CKSUM_ACTUAL,
 559                     DATA_TYPE_UINT64_ARRAY,
 560                     sizeof (info->zbc_actual) / sizeof (uint64_t),
 561                     (uint64_t *)&info->zbc_actual,
 562                     FM_EREPORT_PAYLOAD_ZFS_CKSUM_ALGO,
 563                     DATA_TYPE_STRING,
 564                     info->zbc_checksum_name,
 565                     NULL);
 566
 567                 if (info->zbc_byteswapped) {
 568                         fm_payload_set(ereport,
 569                             FM_EREPORT_PAYLOAD_ZFS_CKSUM_BYTESWAP,
 570                             DATA_TYPE_BOOLEAN, 1,
 571                             NULL);
 572                 }
 573         }
 574
 575         if (badbuf == NULL || goodbuf == NULL)
 576                 return (eip);
 577
 578         ASSERT3U(nui64s, <=, UINT16_MAX);
 579         ASSERT3U(size, ==, nui64s * sizeof (uint64_t));
 580         ASSERT3U(size, <=, SPA_MAXBLOCKSIZE);
 581         ASSERT3U(size, <=, UINT32_MAX);
 582
 583         /* build up the range list by comparing the two buffers. */
 584         for (idx = 0; idx < nui64s; idx++) {
 585                 if (good[idx] == bad[idx]) {
 586                         if (start == -1)
 587                                 continue;
 588
 589                         zei_add_range(eip, start, idx);
 590                         start = -1;
 591                 } else {
 592                         if (start != -1)
 593                                 continue;
 594
 595                         start = idx;
 596                 }
 597         }
 598         if (start != -1)
 599                 zei_add_range(eip, start, idx);
 600
 601         /* See if it will fit in our inline buffers */
 602         inline_size = zei_range_total_size(eip);
 603         if (inline_size > ZFM_MAX_INLINE)
 604                 no_inline = 1;
 605
 606         /*
 607          * If there is no change and we want to drop if the buffers are
 608          * identical, do so.
 609          */
 610         if (inline_size == 0 && drop_if_identical) {
 611                 kmem_free(eip, sizeof (*eip));
 612                 return (NULL);
 613         }
 614
 615         /*
 616          * Now walk through the ranges, filling in the details of the
 617          * differences.  Also convert our uint64_t-array offsets to byte
 618          * offsets.
 619          */
 620         for (range = 0; range < eip->zei_range_count; range++) {
 621                 size_t start = eip->zei_ranges[range].zr_start;
 622                 size_t end = eip->zei_ranges[range].zr_end;
 623
 624                 for (idx = start; idx < end; idx++) {
 625                         uint64_t set, cleared;
 626
 627                         // bits set in bad, but not in good
 628                         set = ((~good[idx]) & bad[idx]);
 629                         // bits set in good, but not in bad
 630                         cleared = (good[idx] & (~bad[idx]));
 631
 632                         allset |= set;
 633                         allcleared |= cleared;
 634
 635                         if (!no_inline) {
 636                                 ASSERT3U(offset, <, inline_size);
 637                                 eip->zei_bits_set[offset] = set;
 638                                 eip->zei_bits_cleared[offset] = cleared;
 639                                 offset++;
 640                         }
 641
 642                         update_histogram(set, eip->zei_histogram_set,
 643                             &eip->zei_range_sets[range]);
 644                         update_histogram(cleared, eip->zei_histogram_cleared,
 645                             &eip->zei_range_clears[range]);
 646                 }
 647
 648                 /* convert to byte offsets */
 649                 eip->zei_ranges[range].zr_start *= sizeof (uint64_t);
 650                 eip->zei_ranges[range].zr_end   *= sizeof (uint64_t);
 651         }
 652         eip->zei_allowed_mingap *= sizeof (uint64_t);
 653         inline_size             *= sizeof (uint64_t);
 654
 655         /* fill in ereport */
 656         fm_payload_set(ereport,
 657             FM_EREPORT_PAYLOAD_ZFS_BAD_OFFSET_RANGES,
 658             DATA_TYPE_UINT32_ARRAY, 2 * eip->zei_range_count,
 659             (uint32_t *)eip->zei_ranges,
 660             FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_MIN_GAP,
 661             DATA_TYPE_UINT32, eip->zei_allowed_mingap,
 662             FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_SETS,
 663             DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_sets,
 664             FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_CLEARS,
 665             DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_clears,
 666             NULL);
 667
 668         if (!no_inline) {
 669                 fm_payload_set(ereport,
 670                     FM_EREPORT_PAYLOAD_ZFS_BAD_SET_BITS,
 671                     DATA_TYPE_UINT8_ARRAY,
 672                     inline_size, (uint8_t *)eip->zei_bits_set,
 673                     FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_BITS,
 674                     DATA_TYPE_UINT8_ARRAY,
 675                     inline_size, (uint8_t *)eip->zei_bits_cleared,
 676                     NULL);
 677         } else {
 678                 fm_payload_set(ereport,
 679                     FM_EREPORT_PAYLOAD_ZFS_BAD_SET_HISTOGRAM,
 680                     DATA_TYPE_UINT16_ARRAY,
 681                     NBBY * sizeof (uint64_t), eip->zei_histogram_set,
 682                     FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_HISTOGRAM,
 683                     DATA_TYPE_UINT16_ARRAY,
 684                     NBBY * sizeof (uint64_t), eip->zei_histogram_cleared,
 685                     NULL);
 686         }
 687         return (eip);
 688 }
 689 #endif
 690
 691 void
 692 zfs_ereport_post(const char *subclass, spa_t *spa, vdev_t *vd, zio_t *zio,
 693     uint64_t stateoroffset, uint64_t size)
 694 {
 695 #ifdef _KERNEL
 696         nvlist_t *ereport = NULL;
 697         nvlist_t *detector = NULL;
 698
 699         zfs_ereport_start(&ereport, &detector,
 700             subclass, spa, vd, zio, stateoroffset, size);
 701
 702         if (ereport == NULL)
 703                 return;
 704
 705         /* Cleanup is handled by the callback function */
 706         zfs_zevent_post(ereport, detector, zfs_zevent_post_cb);
 707 #endif
 708 }
 709
 710 void
 711 zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd,
 712     struct zio *zio, uint64_t offset, uint64_t length, void *arg,
 713     zio_bad_cksum_t *info)
 714 {
 715         zio_cksum_report_t *report = kmem_zalloc(sizeof (*report), KM_PUSHPAGE);
 716
 717         if (zio->io_vsd != NULL)
 718                 zio->io_vsd_ops->vsd_cksum_report(zio, report, arg);
 719         else
 720                 zio_vsd_default_cksum_report(zio, report, arg);
 721
 722         /* copy the checksum failure information if it was provided */
 723         if (info != NULL) {
 724                 report->zcr_ckinfo = kmem_zalloc(sizeof (*info), KM_PUSHPAGE);
 725                 bcopy(info, report->zcr_ckinfo, sizeof (*info));
 726         }
 727
 728         report->zcr_align = 1ULL << vd->vdev_top->vdev_ashift;
 729         report->zcr_length = length;
 730
 731 #ifdef _KERNEL
 732         zfs_ereport_start(&report->zcr_ereport, &report->zcr_detector,
 733             FM_EREPORT_ZFS_CHECKSUM, spa, vd, zio, offset, length);
 734
 735         if (report->zcr_ereport == NULL) {
 736                 report->zcr_free(report->zcr_cbdata, report->zcr_cbinfo);
 737                 if (report->zcr_ckinfo != NULL) {
 738                         kmem_free(report->zcr_ckinfo,
 739                             sizeof (*report->zcr_ckinfo));
 740                 }
 741                 kmem_free(report, sizeof (*report));
 742                 return;
 743         }
 744 #endif
 745
 746         mutex_enter(&spa->spa_errlist_lock);
 747         report->zcr_next = zio->io_logical->io_cksum_report;
 748         zio->io_logical->io_cksum_report = report;
 749         mutex_exit(&spa->spa_errlist_lock);
 750 }
 751
 752 void
 753 zfs_ereport_finish_checksum(zio_cksum_report_t *report,
 754     const void *good_data, const void *bad_data, boolean_t drop_if_identical)
 755 {
 756 #ifdef _KERNEL
 757         zfs_ecksum_info_t *info = NULL;
 758         info = annotate_ecksum(report->zcr_ereport, report->zcr_ckinfo,
 759             good_data, bad_data, report->zcr_length, drop_if_identical);
 760
 761         if (info != NULL)
 762                 zfs_zevent_post(report->zcr_ereport,
 763                     report->zcr_detector, zfs_zevent_post_cb);
 764
 765         report->zcr_ereport = report->zcr_detector = NULL;
 766         if (info != NULL)
 767                 kmem_free(info, sizeof (*info));
 768 #endif
 769 }
 770
 771 void
 772 zfs_ereport_free_checksum(zio_cksum_report_t *rpt)
 773 {
 774 #ifdef _KERNEL
 775         if (rpt->zcr_ereport != NULL) {
 776                 fm_nvlist_destroy(rpt->zcr_ereport,
 777                     FM_NVA_FREE);
 778                 fm_nvlist_destroy(rpt->zcr_detector,
 779                     FM_NVA_FREE);
 780         }
 781 #endif
 782         rpt->zcr_free(rpt->zcr_cbdata, rpt->zcr_cbinfo);
 783
 784         if (rpt->zcr_ckinfo != NULL)
 785                 kmem_free(rpt->zcr_ckinfo, sizeof (*rpt->zcr_ckinfo));
 786
 787         kmem_free(rpt, sizeof (*rpt));
 788 }
 789
 790 void
 791 zfs_ereport_send_interim_checksum(zio_cksum_report_t *report)
 792 {
 793 #ifdef _KERNEL
 794         zfs_zevent_post(report->zcr_ereport, report->zcr_detector, NULL);
 795 #endif
 796 }
 797
 798 void
 799 zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
 800     struct zio *zio, uint64_t offset, uint64_t length,
 801     const void *good_data, const void *bad_data, zio_bad_cksum_t *zbc)
 802 {
 803 #ifdef _KERNEL
 804         nvlist_t *ereport = NULL;
 805         nvlist_t *detector = NULL;
 806         zfs_ecksum_info_t *info;
 807
 808         zfs_ereport_start(&ereport, &detector,
 809             FM_EREPORT_ZFS_CHECKSUM, spa, vd, zio, offset, length);
 810
 811         if (ereport == NULL)
 812                 return;
 813
 814         info = annotate_ecksum(ereport, zbc, good_data, bad_data, length,
 815             B_FALSE);
 816
 817         if (info != NULL) {
 818                 zfs_zevent_post(ereport, detector, zfs_zevent_post_cb);
 819                 kmem_free(info, sizeof (*info));
 820         }
 821 #endif
 822 }
 823
 824 static void
 825 zfs_post_common(spa_t *spa, vdev_t *vd, const char *name)
 826 {
 827 #ifdef _KERNEL
 828         nvlist_t *resource;
 829         char class[64];
 830
 831         if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT)
 832                 return;
 833
 834         if ((resource = fm_nvlist_create(NULL)) == NULL)
 835                 return;
 836
 837         (void) snprintf(class, sizeof (class), "%s.%s.%s", FM_RSRC_RESOURCE,
 838             ZFS_ERROR_CLASS, name);
 839         VERIFY(nvlist_add_uint8(resource, FM_VERSION, FM_RSRC_VERSION) == 0);
 840         VERIFY(nvlist_add_string(resource, FM_CLASS, class) == 0);
 841         VERIFY(nvlist_add_uint64(resource,
 842             FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, spa_guid(spa)) == 0);
 843         if (vd) {
 844                 VERIFY(nvlist_add_uint64(resource,
 845                     FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vd->vdev_guid) == 0);
 846                 VERIFY(nvlist_add_uint64(resource,
 847                     FM_EREPORT_PAYLOAD_ZFS_VDEV_STATE, vd->vdev_state) == 0);
 848         }
 849
 850         zfs_zevent_post(resource, NULL, zfs_zevent_post_cb);
 851 #endif
 852 }
 853
 854 /*
 855  * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev
 856  * has been removed from the system.  This will cause the DE to ignore any
 857  * recent I/O errors, inferring that they are due to the asynchronous device
 858  * removal.
 859  */
 860 void
 861 zfs_post_remove(spa_t *spa, vdev_t *vd)
 862 {
 863         zfs_post_common(spa, vd, FM_EREPORT_RESOURCE_REMOVED);
 864 }
 865
 866 /*
 867  * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool
 868  * has the 'autoreplace' property set, and therefore any broken vdevs will be
 869  * handled by higher level logic, and no vdev fault should be generated.
 870  */
 871 void
 872 zfs_post_autoreplace(spa_t *spa, vdev_t *vd)
 873 {
 874         zfs_post_common(spa, vd, FM_EREPORT_RESOURCE_AUTOREPLACE);
 875 }
 876
 877 /*
 878  * The 'resource.fs.zfs.statechange' event is an internal signal that the
 879  * given vdev has transitioned its state to DEGRADED or HEALTHY.  This will
 880  * cause the retire agent to repair any outstanding fault management cases
 881  * open because the device was not found (fault.fs.zfs.device).
 882  */
 883 void
 884 zfs_post_state_change(spa_t *spa, vdev_t *vd)
 885 {
 886         zfs_post_common(spa, vd, FM_EREPORT_RESOURCE_STATECHANGE);
 887 }
 888
 889 #if defined(_KERNEL) && defined(HAVE_SPL)
 890 EXPORT_SYMBOL(zfs_ereport_post);
 891 EXPORT_SYMBOL(zfs_ereport_post_checksum);
 892 EXPORT_SYMBOL(zfs_post_remove);
 893 EXPORT_SYMBOL(zfs_post_autoreplace);
 894 EXPORT_SYMBOL(zfs_post_state_change);
 895 #endif /* _KERNEL */