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 https://opensource.org/licenses/CDDL-1.0. 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) 2012,2021 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 */ 104 105 #ifdef _KERNEL 106 /* 107 * Duplicate ereport Detection 108 * 109 * Some ereports are retained momentarily for detecting duplicates. These 110 * are kept in a recent_events_node_t in both a time-ordered list and an AVL 111 * tree of recent unique ereports. 112 * 113 * The lifespan of these recent ereports is bounded (15 mins) and a cleaner 114 * task is used to purge stale entries. 115 */ 116 static list_t recent_events_list; 117 static avl_tree_t recent_events_tree; 118 static kmutex_t recent_events_lock; 119 static taskqid_t recent_events_cleaner_tqid; 120 121 /* 122 * Each node is about 128 bytes so 2,000 would consume 1/4 MiB. 123 * 124 * This setting can be changed dynamically and setting it to zero 125 * disables duplicate detection. 126 */ 127 static unsigned int zfs_zevent_retain_max = 2000; 128 129 /* 130 * The lifespan for a recent ereport entry. The default of 15 minutes is 131 * intended to outlive the zfs diagnosis engine's threshold of 10 errors 132 * over a period of 10 minutes. 133 */ 134 static unsigned int zfs_zevent_retain_expire_secs = 900; 135 136 typedef enum zfs_subclass { 137 ZSC_IO, 138 ZSC_DATA, 139 ZSC_CHECKSUM 140 } zfs_subclass_t; 141 142 typedef struct { 143 /* common criteria */ 144 uint64_t re_pool_guid; 145 uint64_t re_vdev_guid; 146 int re_io_error; 147 uint64_t re_io_size; 148 uint64_t re_io_offset; 149 zfs_subclass_t re_subclass; 150 zio_priority_t re_io_priority; 151 152 /* logical zio criteria (optional) */ 153 zbookmark_phys_t re_io_bookmark; 154 155 /* internal state */ 156 avl_node_t re_tree_link; 157 list_node_t re_list_link; 158 uint64_t re_timestamp; 159 } recent_events_node_t; 160 161 static int 162 recent_events_compare(const void *a, const void *b) 163 { 164 const recent_events_node_t *node1 = a; 165 const recent_events_node_t *node2 = b; 166 int cmp; 167 168 /* 169 * The comparison order here is somewhat arbitrary. 170 * What's important is that if every criteria matches, then it 171 * is a duplicate (i.e. compare returns 0) 172 */ 173 if ((cmp = TREE_CMP(node1->re_subclass, node2->re_subclass)) != 0) 174 return (cmp); 175 if ((cmp = TREE_CMP(node1->re_pool_guid, node2->re_pool_guid)) != 0) 176 return (cmp); 177 if ((cmp = TREE_CMP(node1->re_vdev_guid, node2->re_vdev_guid)) != 0) 178 return (cmp); 179 if ((cmp = TREE_CMP(node1->re_io_error, node2->re_io_error)) != 0) 180 return (cmp); 181 if ((cmp = TREE_CMP(node1->re_io_priority, node2->re_io_priority)) != 0) 182 return (cmp); 183 if ((cmp = TREE_CMP(node1->re_io_size, node2->re_io_size)) != 0) 184 return (cmp); 185 if ((cmp = TREE_CMP(node1->re_io_offset, node2->re_io_offset)) != 0) 186 return (cmp); 187 188 const zbookmark_phys_t *zb1 = &node1->re_io_bookmark; 189 const zbookmark_phys_t *zb2 = &node2->re_io_bookmark; 190 191 if ((cmp = TREE_CMP(zb1->zb_objset, zb2->zb_objset)) != 0) 192 return (cmp); 193 if ((cmp = TREE_CMP(zb1->zb_object, zb2->zb_object)) != 0) 194 return (cmp); 195 if ((cmp = TREE_CMP(zb1->zb_level, zb2->zb_level)) != 0) 196 return (cmp); 197 if ((cmp = TREE_CMP(zb1->zb_blkid, zb2->zb_blkid)) != 0) 198 return (cmp); 199 200 return (0); 201 } 202 203 /* 204 * workaround: vdev properties don't have inheritance 205 */ 206 static uint64_t 207 vdev_prop_get_inherited(vdev_t *vd, vdev_prop_t prop) 208 { 209 uint64_t propdef, propval; 210 211 propdef = vdev_prop_default_numeric(prop); 212 switch (prop) { 213 case VDEV_PROP_CHECKSUM_N: 214 propval = vd->vdev_checksum_n; 215 break; 216 case VDEV_PROP_CHECKSUM_T: 217 propval = vd->vdev_checksum_t; 218 break; 219 case VDEV_PROP_IO_N: 220 propval = vd->vdev_io_n; 221 break; 222 case VDEV_PROP_IO_T: 223 propval = vd->vdev_io_t; 224 break; 225 default: 226 propval = propdef; 227 break; 228 } 229 230 if (propval != propdef) 231 return (propval); 232 233 if (vd->vdev_parent == NULL) 234 return (propdef); 235 236 return (vdev_prop_get_inherited(vd->vdev_parent, prop)); 237 } 238 239 static void zfs_ereport_schedule_cleaner(void); 240 241 /* 242 * background task to clean stale recent event nodes. 243 */ 244 static void 245 zfs_ereport_cleaner(void *arg) 246 { 247 recent_events_node_t *entry; 248 uint64_t now = gethrtime(); 249 250 /* 251 * purge expired entries 252 */ 253 mutex_enter(&recent_events_lock); 254 while ((entry = list_tail(&recent_events_list)) != NULL) { 255 uint64_t age = NSEC2SEC(now - entry->re_timestamp); 256 if (age <= zfs_zevent_retain_expire_secs) 257 break; 258 259 /* remove expired node */ 260 avl_remove(&recent_events_tree, entry); 261 list_remove(&recent_events_list, entry); 262 kmem_free(entry, sizeof (*entry)); 263 } 264 265 /* Restart the cleaner if more entries remain */ 266 recent_events_cleaner_tqid = 0; 267 if (!list_is_empty(&recent_events_list)) 268 zfs_ereport_schedule_cleaner(); 269 270 mutex_exit(&recent_events_lock); 271 } 272 273 static void 274 zfs_ereport_schedule_cleaner(void) 275 { 276 ASSERT(MUTEX_HELD(&recent_events_lock)); 277 278 uint64_t timeout = SEC2NSEC(zfs_zevent_retain_expire_secs + 1); 279 280 recent_events_cleaner_tqid = taskq_dispatch_delay( 281 system_delay_taskq, zfs_ereport_cleaner, NULL, TQ_SLEEP, 282 ddi_get_lbolt() + NSEC_TO_TICK(timeout)); 283 } 284 285 /* 286 * Clear entries for a given vdev or all vdevs in a pool when vdev == NULL 287 */ 288 void 289 zfs_ereport_clear(spa_t *spa, vdev_t *vd) 290 { 291 uint64_t vdev_guid, pool_guid; 292 293 ASSERT(vd != NULL || spa != NULL); 294 if (vd == NULL) { 295 vdev_guid = 0; 296 pool_guid = spa_guid(spa); 297 } else { 298 vdev_guid = vd->vdev_guid; 299 pool_guid = 0; 300 } 301 302 mutex_enter(&recent_events_lock); 303 304 recent_events_node_t *next = list_head(&recent_events_list); 305 while (next != NULL) { 306 recent_events_node_t *entry = next; 307 308 next = list_next(&recent_events_list, next); 309 310 if (entry->re_vdev_guid == vdev_guid || 311 entry->re_pool_guid == pool_guid) { 312 avl_remove(&recent_events_tree, entry); 313 list_remove(&recent_events_list, entry); 314 kmem_free(entry, sizeof (*entry)); 315 } 316 } 317 318 mutex_exit(&recent_events_lock); 319 } 320 321 /* 322 * Check if an ereport would be a duplicate of one recently posted. 323 * 324 * An ereport is considered a duplicate if the set of criteria in 325 * recent_events_node_t all match. 326 * 327 * Only FM_EREPORT_ZFS_IO, FM_EREPORT_ZFS_DATA, and FM_EREPORT_ZFS_CHECKSUM 328 * are candidates for duplicate checking. 329 */ 330 static boolean_t 331 zfs_ereport_is_duplicate(const char *subclass, spa_t *spa, vdev_t *vd, 332 const zbookmark_phys_t *zb, zio_t *zio, uint64_t offset, uint64_t size) 333 { 334 recent_events_node_t search = {0}, *entry; 335 336 if (vd == NULL || zio == NULL) 337 return (B_FALSE); 338 339 if (zfs_zevent_retain_max == 0) 340 return (B_FALSE); 341 342 if (strcmp(subclass, FM_EREPORT_ZFS_IO) == 0) 343 search.re_subclass = ZSC_IO; 344 else if (strcmp(subclass, FM_EREPORT_ZFS_DATA) == 0) 345 search.re_subclass = ZSC_DATA; 346 else if (strcmp(subclass, FM_EREPORT_ZFS_CHECKSUM) == 0) 347 search.re_subclass = ZSC_CHECKSUM; 348 else 349 return (B_FALSE); 350 351 search.re_pool_guid = spa_guid(spa); 352 search.re_vdev_guid = vd->vdev_guid; 353 search.re_io_error = zio->io_error; 354 search.re_io_priority = zio->io_priority; 355 /* if size is supplied use it over what's in zio */ 356 if (size) { 357 search.re_io_size = size; 358 search.re_io_offset = offset; 359 } else { 360 search.re_io_size = zio->io_size; 361 search.re_io_offset = zio->io_offset; 362 } 363 364 /* grab optional logical zio criteria */ 365 if (zb != NULL) { 366 search.re_io_bookmark.zb_objset = zb->zb_objset; 367 search.re_io_bookmark.zb_object = zb->zb_object; 368 search.re_io_bookmark.zb_level = zb->zb_level; 369 search.re_io_bookmark.zb_blkid = zb->zb_blkid; 370 } 371 372 uint64_t now = gethrtime(); 373 374 mutex_enter(&recent_events_lock); 375 376 /* check if we have seen this one recently */ 377 entry = avl_find(&recent_events_tree, &search, NULL); 378 if (entry != NULL) { 379 uint64_t age = NSEC2SEC(now - entry->re_timestamp); 380 381 /* 382 * There is still an active cleaner (since we're here). 383 * Reset the last seen time for this duplicate entry 384 * so that its lifespand gets extended. 385 */ 386 list_remove(&recent_events_list, entry); 387 list_insert_head(&recent_events_list, entry); 388 entry->re_timestamp = now; 389 390 zfs_zevent_track_duplicate(); 391 mutex_exit(&recent_events_lock); 392 393 return (age <= zfs_zevent_retain_expire_secs); 394 } 395 396 if (avl_numnodes(&recent_events_tree) >= zfs_zevent_retain_max) { 397 /* recycle oldest node */ 398 entry = list_tail(&recent_events_list); 399 ASSERT(entry != NULL); 400 list_remove(&recent_events_list, entry); 401 avl_remove(&recent_events_tree, entry); 402 } else { 403 entry = kmem_alloc(sizeof (recent_events_node_t), KM_SLEEP); 404 } 405 406 /* record this as a recent ereport */ 407 *entry = search; 408 avl_add(&recent_events_tree, entry); 409 list_insert_head(&recent_events_list, entry); 410 entry->re_timestamp = now; 411 412 /* Start a cleaner if not already scheduled */ 413 if (recent_events_cleaner_tqid == 0) 414 zfs_ereport_schedule_cleaner(); 415 416 mutex_exit(&recent_events_lock); 417 return (B_FALSE); 418 } 419 420 void 421 zfs_zevent_post_cb(nvlist_t *nvl, nvlist_t *detector) 422 { 423 if (nvl) 424 fm_nvlist_destroy(nvl, FM_NVA_FREE); 425 426 if (detector) 427 fm_nvlist_destroy(detector, FM_NVA_FREE); 428 } 429 430 /* 431 * We want to rate limit ZIO delay, deadman, and checksum events so as to not 432 * flood zevent consumers when a disk is acting up. 433 * 434 * Returns 1 if we're ratelimiting, 0 if not. 435 */ 436 static int 437 zfs_is_ratelimiting_event(const char *subclass, vdev_t *vd) 438 { 439 int rc = 0; 440 /* 441 * zfs_ratelimit() returns 1 if we're *not* ratelimiting and 0 if we 442 * are. Invert it to get our return value. 443 */ 444 if (strcmp(subclass, FM_EREPORT_ZFS_DELAY) == 0) { 445 rc = !zfs_ratelimit(&vd->vdev_delay_rl); 446 } else if (strcmp(subclass, FM_EREPORT_ZFS_DEADMAN) == 0) { 447 rc = !zfs_ratelimit(&vd->vdev_deadman_rl); 448 } else if (strcmp(subclass, FM_EREPORT_ZFS_CHECKSUM) == 0) { 449 rc = !zfs_ratelimit(&vd->vdev_checksum_rl); 450 } 451 452 if (rc) { 453 /* We're rate limiting */ 454 fm_erpt_dropped_increment(); 455 } 456 457 return (rc); 458 } 459 460 /* 461 * Return B_TRUE if the event actually posted, B_FALSE if not. 462 */ 463 static boolean_t 464 zfs_ereport_start(nvlist_t **ereport_out, nvlist_t **detector_out, 465 const char *subclass, spa_t *spa, vdev_t *vd, const zbookmark_phys_t *zb, 466 zio_t *zio, uint64_t stateoroffset, uint64_t size) 467 { 468 nvlist_t *ereport, *detector; 469 470 uint64_t ena; 471 char class[64]; 472 473 if ((ereport = fm_nvlist_create(NULL)) == NULL) 474 return (B_FALSE); 475 476 if ((detector = fm_nvlist_create(NULL)) == NULL) { 477 fm_nvlist_destroy(ereport, FM_NVA_FREE); 478 return (B_FALSE); 479 } 480 481 /* 482 * Serialize ereport generation 483 */ 484 mutex_enter(&spa->spa_errlist_lock); 485 486 /* 487 * Determine the ENA to use for this event. If we are in a loading 488 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use 489 * a root zio-wide ENA. Otherwise, simply use a unique ENA. 490 */ 491 if (spa_load_state(spa) != SPA_LOAD_NONE) { 492 if (spa->spa_ena == 0) 493 spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1); 494 ena = spa->spa_ena; 495 } else if (zio != NULL && zio->io_logical != NULL) { 496 if (zio->io_logical->io_ena == 0) 497 zio->io_logical->io_ena = 498 fm_ena_generate(0, FM_ENA_FMT1); 499 ena = zio->io_logical->io_ena; 500 } else { 501 ena = fm_ena_generate(0, FM_ENA_FMT1); 502 } 503 504 /* 505 * Construct the full class, detector, and other standard FMA fields. 506 */ 507 (void) snprintf(class, sizeof (class), "%s.%s", 508 ZFS_ERROR_CLASS, subclass); 509 510 fm_fmri_zfs_set(detector, FM_ZFS_SCHEME_VERSION, spa_guid(spa), 511 vd != NULL ? vd->vdev_guid : 0); 512 513 fm_ereport_set(ereport, FM_EREPORT_VERSION, class, ena, detector, NULL); 514 515 /* 516 * Construct the per-ereport payload, depending on which parameters are 517 * passed in. 518 */ 519 520 /* 521 * Generic payload members common to all ereports. 522 */ 523 fm_payload_set(ereport, 524 FM_EREPORT_PAYLOAD_ZFS_POOL, DATA_TYPE_STRING, spa_name(spa), 525 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, DATA_TYPE_UINT64, spa_guid(spa), 526 FM_EREPORT_PAYLOAD_ZFS_POOL_STATE, DATA_TYPE_UINT64, 527 (uint64_t)spa_state(spa), 528 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, DATA_TYPE_INT32, 529 (int32_t)spa_load_state(spa), NULL); 530 531 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE, 532 DATA_TYPE_STRING, 533 spa_get_failmode(spa) == ZIO_FAILURE_MODE_WAIT ? 534 FM_EREPORT_FAILMODE_WAIT : 535 spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE ? 536 FM_EREPORT_FAILMODE_CONTINUE : FM_EREPORT_FAILMODE_PANIC, 537 NULL); 538 539 if (vd != NULL) { 540 vdev_t *pvd = vd->vdev_parent; 541 vdev_queue_t *vq = &vd->vdev_queue; 542 vdev_stat_t *vs = &vd->vdev_stat; 543 vdev_t *spare_vd; 544 uint64_t *spare_guids; 545 char **spare_paths; 546 int i, spare_count; 547 548 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, 549 DATA_TYPE_UINT64, vd->vdev_guid, 550 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, 551 DATA_TYPE_STRING, vd->vdev_ops->vdev_op_type, NULL); 552 if (vd->vdev_path != NULL) 553 fm_payload_set(ereport, 554 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH, 555 DATA_TYPE_STRING, vd->vdev_path, NULL); 556 if (vd->vdev_devid != NULL) 557 fm_payload_set(ereport, 558 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID, 559 DATA_TYPE_STRING, vd->vdev_devid, NULL); 560 if (vd->vdev_fru != NULL) 561 fm_payload_set(ereport, 562 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU, 563 DATA_TYPE_STRING, vd->vdev_fru, NULL); 564 if (vd->vdev_enc_sysfs_path != NULL) 565 fm_payload_set(ereport, 566 FM_EREPORT_PAYLOAD_ZFS_VDEV_ENC_SYSFS_PATH, 567 DATA_TYPE_STRING, vd->vdev_enc_sysfs_path, NULL); 568 if (vd->vdev_ashift) 569 fm_payload_set(ereport, 570 FM_EREPORT_PAYLOAD_ZFS_VDEV_ASHIFT, 571 DATA_TYPE_UINT64, vd->vdev_ashift, NULL); 572 573 if (vq != NULL) { 574 fm_payload_set(ereport, 575 FM_EREPORT_PAYLOAD_ZFS_VDEV_COMP_TS, 576 DATA_TYPE_UINT64, vq->vq_io_complete_ts, NULL); 577 fm_payload_set(ereport, 578 FM_EREPORT_PAYLOAD_ZFS_VDEV_DELTA_TS, 579 DATA_TYPE_UINT64, vq->vq_io_delta_ts, NULL); 580 } 581 582 if (vs != NULL) { 583 fm_payload_set(ereport, 584 FM_EREPORT_PAYLOAD_ZFS_VDEV_READ_ERRORS, 585 DATA_TYPE_UINT64, vs->vs_read_errors, 586 FM_EREPORT_PAYLOAD_ZFS_VDEV_WRITE_ERRORS, 587 DATA_TYPE_UINT64, vs->vs_write_errors, 588 FM_EREPORT_PAYLOAD_ZFS_VDEV_CKSUM_ERRORS, 589 DATA_TYPE_UINT64, vs->vs_checksum_errors, 590 FM_EREPORT_PAYLOAD_ZFS_VDEV_DELAYS, 591 DATA_TYPE_UINT64, vs->vs_slow_ios, 592 NULL); 593 } 594 595 if (pvd != NULL) { 596 fm_payload_set(ereport, 597 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID, 598 DATA_TYPE_UINT64, pvd->vdev_guid, 599 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE, 600 DATA_TYPE_STRING, pvd->vdev_ops->vdev_op_type, 601 NULL); 602 if (pvd->vdev_path) 603 fm_payload_set(ereport, 604 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH, 605 DATA_TYPE_STRING, pvd->vdev_path, NULL); 606 if (pvd->vdev_devid) 607 fm_payload_set(ereport, 608 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID, 609 DATA_TYPE_STRING, pvd->vdev_devid, NULL); 610 } 611 612 spare_count = spa->spa_spares.sav_count; 613 spare_paths = kmem_zalloc(sizeof (char *) * spare_count, 614 KM_SLEEP); 615 spare_guids = kmem_zalloc(sizeof (uint64_t) * spare_count, 616 KM_SLEEP); 617 618 for (i = 0; i < spare_count; i++) { 619 spare_vd = spa->spa_spares.sav_vdevs[i]; 620 if (spare_vd) { 621 spare_paths[i] = spare_vd->vdev_path; 622 spare_guids[i] = spare_vd->vdev_guid; 623 } 624 } 625 626 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_SPARE_PATHS, 627 DATA_TYPE_STRING_ARRAY, spare_count, spare_paths, 628 FM_EREPORT_PAYLOAD_ZFS_VDEV_SPARE_GUIDS, 629 DATA_TYPE_UINT64_ARRAY, spare_count, spare_guids, NULL); 630 631 kmem_free(spare_guids, sizeof (uint64_t) * spare_count); 632 kmem_free(spare_paths, sizeof (char *) * spare_count); 633 } 634 635 if (zio != NULL) { 636 /* 637 * Payload common to all I/Os. 638 */ 639 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR, 640 DATA_TYPE_INT32, zio->io_error, NULL); 641 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_FLAGS, 642 DATA_TYPE_INT32, zio->io_flags, NULL); 643 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_STAGE, 644 DATA_TYPE_UINT32, zio->io_stage, NULL); 645 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_PIPELINE, 646 DATA_TYPE_UINT32, zio->io_pipeline, NULL); 647 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELAY, 648 DATA_TYPE_UINT64, zio->io_delay, NULL); 649 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_TIMESTAMP, 650 DATA_TYPE_UINT64, zio->io_timestamp, NULL); 651 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_DELTA, 652 DATA_TYPE_UINT64, zio->io_delta, NULL); 653 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_PRIORITY, 654 DATA_TYPE_UINT32, zio->io_priority, NULL); 655 656 /* 657 * If the 'size' parameter is non-zero, it indicates this is a 658 * RAID-Z or other I/O where the physical offset and length are 659 * provided for us, instead of within the zio_t. 660 */ 661 if (vd != NULL) { 662 if (size) 663 fm_payload_set(ereport, 664 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 665 DATA_TYPE_UINT64, stateoroffset, 666 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 667 DATA_TYPE_UINT64, size, NULL); 668 else 669 fm_payload_set(ereport, 670 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 671 DATA_TYPE_UINT64, zio->io_offset, 672 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 673 DATA_TYPE_UINT64, zio->io_size, NULL); 674 } 675 } else if (vd != NULL) { 676 /* 677 * If we have a vdev but no zio, this is a device fault, and the 678 * 'stateoroffset' parameter indicates the previous state of the 679 * vdev. 680 */ 681 fm_payload_set(ereport, 682 FM_EREPORT_PAYLOAD_ZFS_PREV_STATE, 683 DATA_TYPE_UINT64, stateoroffset, NULL); 684 } 685 686 /* 687 * Payload for I/Os with corresponding logical information. 688 */ 689 if (zb != NULL && (zio == NULL || zio->io_logical != NULL)) { 690 fm_payload_set(ereport, 691 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET, 692 DATA_TYPE_UINT64, zb->zb_objset, 693 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT, 694 DATA_TYPE_UINT64, zb->zb_object, 695 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL, 696 DATA_TYPE_INT64, zb->zb_level, 697 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID, 698 DATA_TYPE_UINT64, zb->zb_blkid, NULL); 699 } 700 701 /* 702 * Payload for tuning the zed 703 */ 704 if (vd != NULL && strcmp(subclass, FM_EREPORT_ZFS_CHECKSUM) == 0) { 705 uint64_t cksum_n, cksum_t; 706 707 cksum_n = vdev_prop_get_inherited(vd, VDEV_PROP_CHECKSUM_N); 708 if (cksum_n != vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_N)) 709 fm_payload_set(ereport, 710 FM_EREPORT_PAYLOAD_ZFS_VDEV_CKSUM_N, 711 DATA_TYPE_UINT64, 712 cksum_n, 713 NULL); 714 715 cksum_t = vdev_prop_get_inherited(vd, VDEV_PROP_CHECKSUM_T); 716 if (cksum_t != vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_T)) 717 fm_payload_set(ereport, 718 FM_EREPORT_PAYLOAD_ZFS_VDEV_CKSUM_T, 719 DATA_TYPE_UINT64, 720 cksum_t, 721 NULL); 722 } 723 724 if (vd != NULL && strcmp(subclass, FM_EREPORT_ZFS_IO) == 0) { 725 uint64_t io_n, io_t; 726 727 io_n = vdev_prop_get_inherited(vd, VDEV_PROP_IO_N); 728 if (io_n != vdev_prop_default_numeric(VDEV_PROP_IO_N)) 729 fm_payload_set(ereport, 730 FM_EREPORT_PAYLOAD_ZFS_VDEV_IO_N, 731 DATA_TYPE_UINT64, 732 io_n, 733 NULL); 734 735 io_t = vdev_prop_get_inherited(vd, VDEV_PROP_IO_T); 736 if (io_t != vdev_prop_default_numeric(VDEV_PROP_IO_T)) 737 fm_payload_set(ereport, 738 FM_EREPORT_PAYLOAD_ZFS_VDEV_IO_T, 739 DATA_TYPE_UINT64, 740 io_t, 741 NULL); 742 } 743 744 mutex_exit(&spa->spa_errlist_lock); 745 746 *ereport_out = ereport; 747 *detector_out = detector; 748 return (B_TRUE); 749 } 750 751 /* if it's <= 128 bytes, save the corruption directly */ 752 #define ZFM_MAX_INLINE (128 / sizeof (uint64_t)) 753 754 #define MAX_RANGES 16 755 756 typedef struct zfs_ecksum_info { 757 /* histograms of set and cleared bits by bit number in a 64-bit word */ 758 uint8_t zei_histogram_set[sizeof (uint64_t) * NBBY]; 759 uint8_t zei_histogram_cleared[sizeof (uint64_t) * NBBY]; 760 761 /* inline arrays of bits set and cleared. */ 762 uint64_t zei_bits_set[ZFM_MAX_INLINE]; 763 uint64_t zei_bits_cleared[ZFM_MAX_INLINE]; 764 765 /* 766 * for each range, the number of bits set and cleared. The Hamming 767 * distance between the good and bad buffers is the sum of them all. 768 */ 769 uint32_t zei_range_sets[MAX_RANGES]; 770 uint32_t zei_range_clears[MAX_RANGES]; 771 772 struct zei_ranges { 773 uint32_t zr_start; 774 uint32_t zr_end; 775 } zei_ranges[MAX_RANGES]; 776 777 size_t zei_range_count; 778 uint32_t zei_mingap; 779 uint32_t zei_allowed_mingap; 780 781 } zfs_ecksum_info_t; 782 783 static void 784 update_histogram(uint64_t value_arg, uint8_t *hist, uint32_t *count) 785 { 786 size_t i; 787 size_t bits = 0; 788 uint64_t value = BE_64(value_arg); 789 790 /* We store the bits in big-endian (largest-first) order */ 791 for (i = 0; i < 64; i++) { 792 if (value & (1ull << i)) { 793 hist[63 - i]++; 794 ++bits; 795 } 796 } 797 /* update the count of bits changed */ 798 *count += bits; 799 } 800 801 /* 802 * We've now filled up the range array, and need to increase "mingap" and 803 * shrink the range list accordingly. zei_mingap is always the smallest 804 * distance between array entries, so we set the new_allowed_gap to be 805 * one greater than that. We then go through the list, joining together 806 * any ranges which are closer than the new_allowed_gap. 807 * 808 * By construction, there will be at least one. We also update zei_mingap 809 * to the new smallest gap, to prepare for our next invocation. 810 */ 811 static void 812 zei_shrink_ranges(zfs_ecksum_info_t *eip) 813 { 814 uint32_t mingap = UINT32_MAX; 815 uint32_t new_allowed_gap = eip->zei_mingap + 1; 816 817 size_t idx, output; 818 size_t max = eip->zei_range_count; 819 820 struct zei_ranges *r = eip->zei_ranges; 821 822 ASSERT3U(eip->zei_range_count, >, 0); 823 ASSERT3U(eip->zei_range_count, <=, MAX_RANGES); 824 825 output = idx = 0; 826 while (idx < max - 1) { 827 uint32_t start = r[idx].zr_start; 828 uint32_t end = r[idx].zr_end; 829 830 while (idx < max - 1) { 831 idx++; 832 833 uint32_t nstart = r[idx].zr_start; 834 uint32_t nend = r[idx].zr_end; 835 836 uint32_t gap = nstart - end; 837 if (gap < new_allowed_gap) { 838 end = nend; 839 continue; 840 } 841 if (gap < mingap) 842 mingap = gap; 843 break; 844 } 845 r[output].zr_start = start; 846 r[output].zr_end = end; 847 output++; 848 } 849 ASSERT3U(output, <, eip->zei_range_count); 850 eip->zei_range_count = output; 851 eip->zei_mingap = mingap; 852 eip->zei_allowed_mingap = new_allowed_gap; 853 } 854 855 static void 856 zei_add_range(zfs_ecksum_info_t *eip, int start, int end) 857 { 858 struct zei_ranges *r = eip->zei_ranges; 859 size_t count = eip->zei_range_count; 860 861 if (count >= MAX_RANGES) { 862 zei_shrink_ranges(eip); 863 count = eip->zei_range_count; 864 } 865 if (count == 0) { 866 eip->zei_mingap = UINT32_MAX; 867 eip->zei_allowed_mingap = 1; 868 } else { 869 int gap = start - r[count - 1].zr_end; 870 871 if (gap < eip->zei_allowed_mingap) { 872 r[count - 1].zr_end = end; 873 return; 874 } 875 if (gap < eip->zei_mingap) 876 eip->zei_mingap = gap; 877 } 878 r[count].zr_start = start; 879 r[count].zr_end = end; 880 eip->zei_range_count++; 881 } 882 883 static size_t 884 zei_range_total_size(zfs_ecksum_info_t *eip) 885 { 886 struct zei_ranges *r = eip->zei_ranges; 887 size_t count = eip->zei_range_count; 888 size_t result = 0; 889 size_t idx; 890 891 for (idx = 0; idx < count; idx++) 892 result += (r[idx].zr_end - r[idx].zr_start); 893 894 return (result); 895 } 896 897 static zfs_ecksum_info_t * 898 annotate_ecksum(nvlist_t *ereport, zio_bad_cksum_t *info, 899 const abd_t *goodabd, const abd_t *badabd, size_t size, 900 boolean_t drop_if_identical) 901 { 902 const uint64_t *good; 903 const uint64_t *bad; 904 905 size_t nui64s = size / sizeof (uint64_t); 906 907 size_t inline_size; 908 int no_inline = 0; 909 size_t idx; 910 size_t range; 911 912 size_t offset = 0; 913 ssize_t start = -1; 914 915 zfs_ecksum_info_t *eip = kmem_zalloc(sizeof (*eip), KM_SLEEP); 916 917 /* don't do any annotation for injected checksum errors */ 918 if (info != NULL && info->zbc_injected) 919 return (eip); 920 921 if (info != NULL && info->zbc_has_cksum) { 922 fm_payload_set(ereport, 923 FM_EREPORT_PAYLOAD_ZFS_CKSUM_EXPECTED, 924 DATA_TYPE_UINT64_ARRAY, 925 sizeof (info->zbc_expected) / sizeof (uint64_t), 926 (uint64_t *)&info->zbc_expected, 927 FM_EREPORT_PAYLOAD_ZFS_CKSUM_ACTUAL, 928 DATA_TYPE_UINT64_ARRAY, 929 sizeof (info->zbc_actual) / sizeof (uint64_t), 930 (uint64_t *)&info->zbc_actual, 931 FM_EREPORT_PAYLOAD_ZFS_CKSUM_ALGO, 932 DATA_TYPE_STRING, 933 info->zbc_checksum_name, 934 NULL); 935 936 if (info->zbc_byteswapped) { 937 fm_payload_set(ereport, 938 FM_EREPORT_PAYLOAD_ZFS_CKSUM_BYTESWAP, 939 DATA_TYPE_BOOLEAN, 1, 940 NULL); 941 } 942 } 943 944 if (badabd == NULL || goodabd == NULL) 945 return (eip); 946 947 ASSERT3U(nui64s, <=, UINT32_MAX); 948 ASSERT3U(size, ==, nui64s * sizeof (uint64_t)); 949 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); 950 ASSERT3U(size, <=, UINT32_MAX); 951 952 good = (const uint64_t *) abd_borrow_buf_copy((abd_t *)goodabd, size); 953 bad = (const uint64_t *) abd_borrow_buf_copy((abd_t *)badabd, size); 954 955 /* build up the range list by comparing the two buffers. */ 956 for (idx = 0; idx < nui64s; idx++) { 957 if (good[idx] == bad[idx]) { 958 if (start == -1) 959 continue; 960 961 zei_add_range(eip, start, idx); 962 start = -1; 963 } else { 964 if (start != -1) 965 continue; 966 967 start = idx; 968 } 969 } 970 if (start != -1) 971 zei_add_range(eip, start, idx); 972 973 /* See if it will fit in our inline buffers */ 974 inline_size = zei_range_total_size(eip); 975 if (inline_size > ZFM_MAX_INLINE) 976 no_inline = 1; 977 978 /* 979 * If there is no change and we want to drop if the buffers are 980 * identical, do so. 981 */ 982 if (inline_size == 0 && drop_if_identical) { 983 kmem_free(eip, sizeof (*eip)); 984 abd_return_buf((abd_t *)goodabd, (void *)good, size); 985 abd_return_buf((abd_t *)badabd, (void *)bad, size); 986 return (NULL); 987 } 988 989 /* 990 * Now walk through the ranges, filling in the details of the 991 * differences. Also convert our uint64_t-array offsets to byte 992 * offsets. 993 */ 994 for (range = 0; range < eip->zei_range_count; range++) { 995 size_t start = eip->zei_ranges[range].zr_start; 996 size_t end = eip->zei_ranges[range].zr_end; 997 998 for (idx = start; idx < end; idx++) { 999 uint64_t set, cleared; 1000 1001 // bits set in bad, but not in good 1002 set = ((~good[idx]) & bad[idx]); 1003 // bits set in good, but not in bad 1004 cleared = (good[idx] & (~bad[idx])); 1005 1006 if (!no_inline) { 1007 ASSERT3U(offset, <, inline_size); 1008 eip->zei_bits_set[offset] = set; 1009 eip->zei_bits_cleared[offset] = cleared; 1010 offset++; 1011 } 1012 1013 update_histogram(set, eip->zei_histogram_set, 1014 &eip->zei_range_sets[range]); 1015 update_histogram(cleared, eip->zei_histogram_cleared, 1016 &eip->zei_range_clears[range]); 1017 } 1018 1019 /* convert to byte offsets */ 1020 eip->zei_ranges[range].zr_start *= sizeof (uint64_t); 1021 eip->zei_ranges[range].zr_end *= sizeof (uint64_t); 1022 } 1023 1024 abd_return_buf((abd_t *)goodabd, (void *)good, size); 1025 abd_return_buf((abd_t *)badabd, (void *)bad, size); 1026 1027 eip->zei_allowed_mingap *= sizeof (uint64_t); 1028 inline_size *= sizeof (uint64_t); 1029 1030 /* fill in ereport */ 1031 fm_payload_set(ereport, 1032 FM_EREPORT_PAYLOAD_ZFS_BAD_OFFSET_RANGES, 1033 DATA_TYPE_UINT32_ARRAY, 2 * eip->zei_range_count, 1034 (uint32_t *)eip->zei_ranges, 1035 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_MIN_GAP, 1036 DATA_TYPE_UINT32, eip->zei_allowed_mingap, 1037 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_SETS, 1038 DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_sets, 1039 FM_EREPORT_PAYLOAD_ZFS_BAD_RANGE_CLEARS, 1040 DATA_TYPE_UINT32_ARRAY, eip->zei_range_count, eip->zei_range_clears, 1041 NULL); 1042 1043 if (!no_inline) { 1044 fm_payload_set(ereport, 1045 FM_EREPORT_PAYLOAD_ZFS_BAD_SET_BITS, 1046 DATA_TYPE_UINT8_ARRAY, 1047 inline_size, (uint8_t *)eip->zei_bits_set, 1048 FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_BITS, 1049 DATA_TYPE_UINT8_ARRAY, 1050 inline_size, (uint8_t *)eip->zei_bits_cleared, 1051 NULL); 1052 } else { 1053 fm_payload_set(ereport, 1054 FM_EREPORT_PAYLOAD_ZFS_BAD_SET_HISTOGRAM, 1055 DATA_TYPE_UINT8_ARRAY, 1056 NBBY * sizeof (uint64_t), eip->zei_histogram_set, 1057 FM_EREPORT_PAYLOAD_ZFS_BAD_CLEARED_HISTOGRAM, 1058 DATA_TYPE_UINT8_ARRAY, 1059 NBBY * sizeof (uint64_t), eip->zei_histogram_cleared, 1060 NULL); 1061 } 1062 return (eip); 1063 } 1064 #else 1065 void 1066 zfs_ereport_clear(spa_t *spa, vdev_t *vd) 1067 { 1068 (void) spa, (void) vd; 1069 } 1070 #endif 1071 1072 /* 1073 * Make sure our event is still valid for the given zio/vdev/pool. For example, 1074 * we don't want to keep logging events for a faulted or missing vdev. 1075 */ 1076 boolean_t 1077 zfs_ereport_is_valid(const char *subclass, spa_t *spa, vdev_t *vd, zio_t *zio) 1078 { 1079 #ifdef _KERNEL 1080 /* 1081 * If we are doing a spa_tryimport() or in recovery mode, 1082 * ignore errors. 1083 */ 1084 if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT || 1085 spa_load_state(spa) == SPA_LOAD_RECOVER) 1086 return (B_FALSE); 1087 1088 /* 1089 * If we are in the middle of opening a pool, and the previous attempt 1090 * failed, don't bother logging any new ereports - we're just going to 1091 * get the same diagnosis anyway. 1092 */ 1093 if (spa_load_state(spa) != SPA_LOAD_NONE && 1094 spa->spa_last_open_failed) 1095 return (B_FALSE); 1096 1097 if (zio != NULL) { 1098 /* 1099 * If this is not a read or write zio, ignore the error. This 1100 * can occur if the DKIOCFLUSHWRITECACHE ioctl fails. 1101 */ 1102 if (zio->io_type != ZIO_TYPE_READ && 1103 zio->io_type != ZIO_TYPE_WRITE) 1104 return (B_FALSE); 1105 1106 if (vd != NULL) { 1107 /* 1108 * If the vdev has already been marked as failing due 1109 * to a failed probe, then ignore any subsequent I/O 1110 * errors, as the DE will automatically fault the vdev 1111 * on the first such failure. This also catches cases 1112 * where vdev_remove_wanted is set and the device has 1113 * not yet been asynchronously placed into the REMOVED 1114 * state. 1115 */ 1116 if (zio->io_vd == vd && !vdev_accessible(vd, zio)) 1117 return (B_FALSE); 1118 1119 /* 1120 * Ignore checksum errors for reads from DTL regions of 1121 * leaf vdevs. 1122 */ 1123 if (zio->io_type == ZIO_TYPE_READ && 1124 zio->io_error == ECKSUM && 1125 vd->vdev_ops->vdev_op_leaf && 1126 vdev_dtl_contains(vd, DTL_MISSING, zio->io_txg, 1)) 1127 return (B_FALSE); 1128 } 1129 } 1130 1131 /* 1132 * For probe failure, we want to avoid posting ereports if we've 1133 * already removed the device in the meantime. 1134 */ 1135 if (vd != NULL && 1136 strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) == 0 && 1137 (vd->vdev_remove_wanted || vd->vdev_state == VDEV_STATE_REMOVED)) 1138 return (B_FALSE); 1139 1140 /* Ignore bogus delay events (like from ioctls or unqueued IOs) */ 1141 if ((strcmp(subclass, FM_EREPORT_ZFS_DELAY) == 0) && 1142 (zio != NULL) && (!zio->io_timestamp)) { 1143 return (B_FALSE); 1144 } 1145 #else 1146 (void) subclass, (void) spa, (void) vd, (void) zio; 1147 #endif 1148 return (B_TRUE); 1149 } 1150 1151 /* 1152 * Post an ereport for the given subclass 1153 * 1154 * Returns 1155 * - 0 if an event was posted 1156 * - EINVAL if there was a problem posting event 1157 * - EBUSY if the event was rate limited 1158 * - EALREADY if the event was already posted (duplicate) 1159 */ 1160 int 1161 zfs_ereport_post(const char *subclass, spa_t *spa, vdev_t *vd, 1162 const zbookmark_phys_t *zb, zio_t *zio, uint64_t state) 1163 { 1164 int rc = 0; 1165 #ifdef _KERNEL 1166 nvlist_t *ereport = NULL; 1167 nvlist_t *detector = NULL; 1168 1169 if (!zfs_ereport_is_valid(subclass, spa, vd, zio)) 1170 return (EINVAL); 1171 1172 if (zfs_ereport_is_duplicate(subclass, spa, vd, zb, zio, 0, 0)) 1173 return (SET_ERROR(EALREADY)); 1174 1175 if (zfs_is_ratelimiting_event(subclass, vd)) 1176 return (SET_ERROR(EBUSY)); 1177 1178 if (!zfs_ereport_start(&ereport, &detector, subclass, spa, vd, 1179 zb, zio, state, 0)) 1180 return (SET_ERROR(EINVAL)); /* couldn't post event */ 1181 1182 if (ereport == NULL) 1183 return (SET_ERROR(EINVAL)); 1184 1185 /* Cleanup is handled by the callback function */ 1186 rc = zfs_zevent_post(ereport, detector, zfs_zevent_post_cb); 1187 #else 1188 (void) subclass, (void) spa, (void) vd, (void) zb, (void) zio, 1189 (void) state; 1190 #endif 1191 return (rc); 1192 } 1193 1194 /* 1195 * Prepare a checksum ereport 1196 * 1197 * Returns 1198 * - 0 if an event was posted 1199 * - EINVAL if there was a problem posting event 1200 * - EBUSY if the event was rate limited 1201 * - EALREADY if the event was already posted (duplicate) 1202 */ 1203 int 1204 zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, const zbookmark_phys_t *zb, 1205 struct zio *zio, uint64_t offset, uint64_t length, zio_bad_cksum_t *info) 1206 { 1207 zio_cksum_report_t *report; 1208 1209 #ifdef _KERNEL 1210 if (!zfs_ereport_is_valid(FM_EREPORT_ZFS_CHECKSUM, spa, vd, zio)) 1211 return (SET_ERROR(EINVAL)); 1212 1213 if (zfs_ereport_is_duplicate(FM_EREPORT_ZFS_CHECKSUM, spa, vd, zb, zio, 1214 offset, length)) 1215 return (SET_ERROR(EALREADY)); 1216 1217 if (zfs_is_ratelimiting_event(FM_EREPORT_ZFS_CHECKSUM, vd)) 1218 return (SET_ERROR(EBUSY)); 1219 #else 1220 (void) zb, (void) offset; 1221 #endif 1222 1223 report = kmem_zalloc(sizeof (*report), KM_SLEEP); 1224 1225 zio_vsd_default_cksum_report(zio, report); 1226 1227 /* copy the checksum failure information if it was provided */ 1228 if (info != NULL) { 1229 report->zcr_ckinfo = kmem_zalloc(sizeof (*info), KM_SLEEP); 1230 memcpy(report->zcr_ckinfo, info, sizeof (*info)); 1231 } 1232 1233 report->zcr_sector = 1ULL << vd->vdev_top->vdev_ashift; 1234 report->zcr_align = 1235 vdev_psize_to_asize(vd->vdev_top, report->zcr_sector); 1236 report->zcr_length = length; 1237 1238 #ifdef _KERNEL 1239 (void) zfs_ereport_start(&report->zcr_ereport, &report->zcr_detector, 1240 FM_EREPORT_ZFS_CHECKSUM, spa, vd, zb, zio, offset, length); 1241 1242 if (report->zcr_ereport == NULL) { 1243 zfs_ereport_free_checksum(report); 1244 return (0); 1245 } 1246 #endif 1247 1248 mutex_enter(&spa->spa_errlist_lock); 1249 report->zcr_next = zio->io_logical->io_cksum_report; 1250 zio->io_logical->io_cksum_report = report; 1251 mutex_exit(&spa->spa_errlist_lock); 1252 return (0); 1253 } 1254 1255 void 1256 zfs_ereport_finish_checksum(zio_cksum_report_t *report, const abd_t *good_data, 1257 const abd_t *bad_data, boolean_t drop_if_identical) 1258 { 1259 #ifdef _KERNEL 1260 zfs_ecksum_info_t *info; 1261 1262 info = annotate_ecksum(report->zcr_ereport, report->zcr_ckinfo, 1263 good_data, bad_data, report->zcr_length, drop_if_identical); 1264 if (info != NULL) 1265 zfs_zevent_post(report->zcr_ereport, 1266 report->zcr_detector, zfs_zevent_post_cb); 1267 else 1268 zfs_zevent_post_cb(report->zcr_ereport, report->zcr_detector); 1269 1270 report->zcr_ereport = report->zcr_detector = NULL; 1271 if (info != NULL) 1272 kmem_free(info, sizeof (*info)); 1273 #else 1274 (void) report, (void) good_data, (void) bad_data, 1275 (void) drop_if_identical; 1276 #endif 1277 } 1278 1279 void 1280 zfs_ereport_free_checksum(zio_cksum_report_t *rpt) 1281 { 1282 #ifdef _KERNEL 1283 if (rpt->zcr_ereport != NULL) { 1284 fm_nvlist_destroy(rpt->zcr_ereport, 1285 FM_NVA_FREE); 1286 fm_nvlist_destroy(rpt->zcr_detector, 1287 FM_NVA_FREE); 1288 } 1289 #endif 1290 rpt->zcr_free(rpt->zcr_cbdata, rpt->zcr_cbinfo); 1291 1292 if (rpt->zcr_ckinfo != NULL) 1293 kmem_free(rpt->zcr_ckinfo, sizeof (*rpt->zcr_ckinfo)); 1294 1295 kmem_free(rpt, sizeof (*rpt)); 1296 } 1297 1298 /* 1299 * Post a checksum ereport 1300 * 1301 * Returns 1302 * - 0 if an event was posted 1303 * - EINVAL if there was a problem posting event 1304 * - EBUSY if the event was rate limited 1305 * - EALREADY if the event was already posted (duplicate) 1306 */ 1307 int 1308 zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd, const zbookmark_phys_t *zb, 1309 struct zio *zio, uint64_t offset, uint64_t length, 1310 const abd_t *good_data, const abd_t *bad_data, zio_bad_cksum_t *zbc) 1311 { 1312 int rc = 0; 1313 #ifdef _KERNEL 1314 nvlist_t *ereport = NULL; 1315 nvlist_t *detector = NULL; 1316 zfs_ecksum_info_t *info; 1317 1318 if (!zfs_ereport_is_valid(FM_EREPORT_ZFS_CHECKSUM, spa, vd, zio)) 1319 return (SET_ERROR(EINVAL)); 1320 1321 if (zfs_ereport_is_duplicate(FM_EREPORT_ZFS_CHECKSUM, spa, vd, zb, zio, 1322 offset, length)) 1323 return (SET_ERROR(EALREADY)); 1324 1325 if (zfs_is_ratelimiting_event(FM_EREPORT_ZFS_CHECKSUM, vd)) 1326 return (SET_ERROR(EBUSY)); 1327 1328 if (!zfs_ereport_start(&ereport, &detector, FM_EREPORT_ZFS_CHECKSUM, 1329 spa, vd, zb, zio, offset, length) || (ereport == NULL)) { 1330 return (SET_ERROR(EINVAL)); 1331 } 1332 1333 info = annotate_ecksum(ereport, zbc, good_data, bad_data, length, 1334 B_FALSE); 1335 1336 if (info != NULL) { 1337 rc = zfs_zevent_post(ereport, detector, zfs_zevent_post_cb); 1338 kmem_free(info, sizeof (*info)); 1339 } 1340 #else 1341 (void) spa, (void) vd, (void) zb, (void) zio, (void) offset, 1342 (void) length, (void) good_data, (void) bad_data, (void) zbc; 1343 #endif 1344 return (rc); 1345 } 1346 1347 /* 1348 * The 'sysevent.fs.zfs.*' events are signals posted to notify user space of 1349 * change in the pool. All sysevents are listed in sys/sysevent/eventdefs.h 1350 * and are designed to be consumed by the ZFS Event Daemon (ZED). For 1351 * additional details refer to the zed(8) man page. 1352 */ 1353 nvlist_t * 1354 zfs_event_create(spa_t *spa, vdev_t *vd, const char *type, const char *name, 1355 nvlist_t *aux) 1356 { 1357 nvlist_t *resource = NULL; 1358 #ifdef _KERNEL 1359 char class[64]; 1360 1361 if (spa_load_state(spa) == SPA_LOAD_TRYIMPORT) 1362 return (NULL); 1363 1364 if ((resource = fm_nvlist_create(NULL)) == NULL) 1365 return (NULL); 1366 1367 (void) snprintf(class, sizeof (class), "%s.%s.%s", type, 1368 ZFS_ERROR_CLASS, name); 1369 VERIFY0(nvlist_add_uint8(resource, FM_VERSION, FM_RSRC_VERSION)); 1370 VERIFY0(nvlist_add_string(resource, FM_CLASS, class)); 1371 VERIFY0(nvlist_add_string(resource, 1372 FM_EREPORT_PAYLOAD_ZFS_POOL, spa_name(spa))); 1373 VERIFY0(nvlist_add_uint64(resource, 1374 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, spa_guid(spa))); 1375 VERIFY0(nvlist_add_uint64(resource, 1376 FM_EREPORT_PAYLOAD_ZFS_POOL_STATE, spa_state(spa))); 1377 VERIFY0(nvlist_add_int32(resource, 1378 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, spa_load_state(spa))); 1379 1380 if (vd) { 1381 VERIFY0(nvlist_add_uint64(resource, 1382 FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vd->vdev_guid)); 1383 VERIFY0(nvlist_add_uint64(resource, 1384 FM_EREPORT_PAYLOAD_ZFS_VDEV_STATE, vd->vdev_state)); 1385 if (vd->vdev_path != NULL) 1386 VERIFY0(nvlist_add_string(resource, 1387 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH, vd->vdev_path)); 1388 if (vd->vdev_devid != NULL) 1389 VERIFY0(nvlist_add_string(resource, 1390 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID, vd->vdev_devid)); 1391 if (vd->vdev_fru != NULL) 1392 VERIFY0(nvlist_add_string(resource, 1393 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU, vd->vdev_fru)); 1394 if (vd->vdev_enc_sysfs_path != NULL) 1395 VERIFY0(nvlist_add_string(resource, 1396 FM_EREPORT_PAYLOAD_ZFS_VDEV_ENC_SYSFS_PATH, 1397 vd->vdev_enc_sysfs_path)); 1398 } 1399 1400 /* also copy any optional payload data */ 1401 if (aux) { 1402 nvpair_t *elem = NULL; 1403 1404 while ((elem = nvlist_next_nvpair(aux, elem)) != NULL) 1405 (void) nvlist_add_nvpair(resource, elem); 1406 } 1407 #else 1408 (void) spa, (void) vd, (void) type, (void) name, (void) aux; 1409 #endif 1410 return (resource); 1411 } 1412 1413 static void 1414 zfs_post_common(spa_t *spa, vdev_t *vd, const char *type, const char *name, 1415 nvlist_t *aux) 1416 { 1417 #ifdef _KERNEL 1418 nvlist_t *resource; 1419 1420 resource = zfs_event_create(spa, vd, type, name, aux); 1421 if (resource) 1422 zfs_zevent_post(resource, NULL, zfs_zevent_post_cb); 1423 #else 1424 (void) spa, (void) vd, (void) type, (void) name, (void) aux; 1425 #endif 1426 } 1427 1428 /* 1429 * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev 1430 * has been removed from the system. This will cause the DE to ignore any 1431 * recent I/O errors, inferring that they are due to the asynchronous device 1432 * removal. 1433 */ 1434 void 1435 zfs_post_remove(spa_t *spa, vdev_t *vd) 1436 { 1437 zfs_post_common(spa, vd, FM_RSRC_CLASS, FM_RESOURCE_REMOVED, NULL); 1438 } 1439 1440 /* 1441 * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool 1442 * has the 'autoreplace' property set, and therefore any broken vdevs will be 1443 * handled by higher level logic, and no vdev fault should be generated. 1444 */ 1445 void 1446 zfs_post_autoreplace(spa_t *spa, vdev_t *vd) 1447 { 1448 zfs_post_common(spa, vd, FM_RSRC_CLASS, FM_RESOURCE_AUTOREPLACE, NULL); 1449 } 1450 1451 /* 1452 * The 'resource.fs.zfs.statechange' event is an internal signal that the 1453 * given vdev has transitioned its state to DEGRADED or HEALTHY. This will 1454 * cause the retire agent to repair any outstanding fault management cases 1455 * open because the device was not found (fault.fs.zfs.device). 1456 */ 1457 void 1458 zfs_post_state_change(spa_t *spa, vdev_t *vd, uint64_t laststate) 1459 { 1460 #ifdef _KERNEL 1461 nvlist_t *aux; 1462 1463 /* 1464 * Add optional supplemental keys to payload 1465 */ 1466 aux = fm_nvlist_create(NULL); 1467 if (vd && aux) { 1468 if (vd->vdev_physpath) { 1469 fnvlist_add_string(aux, 1470 FM_EREPORT_PAYLOAD_ZFS_VDEV_PHYSPATH, 1471 vd->vdev_physpath); 1472 } 1473 if (vd->vdev_enc_sysfs_path) { 1474 fnvlist_add_string(aux, 1475 FM_EREPORT_PAYLOAD_ZFS_VDEV_ENC_SYSFS_PATH, 1476 vd->vdev_enc_sysfs_path); 1477 } 1478 1479 fnvlist_add_uint64(aux, 1480 FM_EREPORT_PAYLOAD_ZFS_VDEV_LASTSTATE, laststate); 1481 } 1482 1483 zfs_post_common(spa, vd, FM_RSRC_CLASS, FM_RESOURCE_STATECHANGE, 1484 aux); 1485 1486 if (aux) 1487 fm_nvlist_destroy(aux, FM_NVA_FREE); 1488 #else 1489 (void) spa, (void) vd, (void) laststate; 1490 #endif 1491 } 1492 1493 #ifdef _KERNEL 1494 void 1495 zfs_ereport_init(void) 1496 { 1497 mutex_init(&recent_events_lock, NULL, MUTEX_DEFAULT, NULL); 1498 list_create(&recent_events_list, sizeof (recent_events_node_t), 1499 offsetof(recent_events_node_t, re_list_link)); 1500 avl_create(&recent_events_tree, recent_events_compare, 1501 sizeof (recent_events_node_t), offsetof(recent_events_node_t, 1502 re_tree_link)); 1503 } 1504 1505 /* 1506 * This 'early' fini needs to run before zfs_fini() which on Linux waits 1507 * for the system_delay_taskq to drain. 1508 */ 1509 void 1510 zfs_ereport_taskq_fini(void) 1511 { 1512 mutex_enter(&recent_events_lock); 1513 if (recent_events_cleaner_tqid != 0) { 1514 taskq_cancel_id(system_delay_taskq, recent_events_cleaner_tqid); 1515 recent_events_cleaner_tqid = 0; 1516 } 1517 mutex_exit(&recent_events_lock); 1518 } 1519 1520 void 1521 zfs_ereport_fini(void) 1522 { 1523 recent_events_node_t *entry; 1524 1525 while ((entry = list_remove_head(&recent_events_list)) != NULL) { 1526 avl_remove(&recent_events_tree, entry); 1527 kmem_free(entry, sizeof (*entry)); 1528 } 1529 avl_destroy(&recent_events_tree); 1530 list_destroy(&recent_events_list); 1531 mutex_destroy(&recent_events_lock); 1532 } 1533 1534 void 1535 zfs_ereport_snapshot_post(const char *subclass, spa_t *spa, const char *name) 1536 { 1537 nvlist_t *aux; 1538 1539 aux = fm_nvlist_create(NULL); 1540 fnvlist_add_string(aux, FM_EREPORT_PAYLOAD_ZFS_SNAPSHOT_NAME, name); 1541 1542 zfs_post_common(spa, NULL, FM_RSRC_CLASS, subclass, aux); 1543 fm_nvlist_destroy(aux, FM_NVA_FREE); 1544 } 1545 1546 /* 1547 * Post when a event when a zvol is created or removed 1548 * 1549 * This is currently only used by macOS, since it uses the event to create 1550 * symlinks between the volume name (mypool/myvol) and the actual /dev 1551 * device (/dev/disk3). For example: 1552 * 1553 * /var/run/zfs/dsk/mypool/myvol -> /dev/disk3 1554 * 1555 * name: The full name of the zvol ("mypool/myvol") 1556 * dev_name: The full /dev name for the zvol ("/dev/disk3") 1557 * raw_name: The raw /dev name for the zvol ("/dev/rdisk3") 1558 */ 1559 void 1560 zfs_ereport_zvol_post(const char *subclass, const char *name, 1561 const char *dev_name, const char *raw_name) 1562 { 1563 nvlist_t *aux; 1564 char *r; 1565 1566 boolean_t locked = mutex_owned(&spa_namespace_lock); 1567 if (!locked) mutex_enter(&spa_namespace_lock); 1568 spa_t *spa = spa_lookup(name); 1569 if (!locked) mutex_exit(&spa_namespace_lock); 1570 1571 if (spa == NULL) 1572 return; 1573 1574 aux = fm_nvlist_create(NULL); 1575 fnvlist_add_string(aux, FM_EREPORT_PAYLOAD_ZFS_DEVICE_NAME, dev_name); 1576 fnvlist_add_string(aux, FM_EREPORT_PAYLOAD_ZFS_RAW_DEVICE_NAME, 1577 raw_name); 1578 r = strchr(name, '/'); 1579 if (r && r[1]) 1580 fnvlist_add_string(aux, FM_EREPORT_PAYLOAD_ZFS_VOLUME, &r[1]); 1581 1582 zfs_post_common(spa, NULL, FM_RSRC_CLASS, subclass, aux); 1583 fm_nvlist_destroy(aux, FM_NVA_FREE); 1584 } 1585 1586 EXPORT_SYMBOL(zfs_ereport_post); 1587 EXPORT_SYMBOL(zfs_ereport_is_valid); 1588 EXPORT_SYMBOL(zfs_ereport_post_checksum); 1589 EXPORT_SYMBOL(zfs_post_remove); 1590 EXPORT_SYMBOL(zfs_post_autoreplace); 1591 EXPORT_SYMBOL(zfs_post_state_change); 1592 1593 ZFS_MODULE_PARAM(zfs_zevent, zfs_zevent_, retain_max, UINT, ZMOD_RW, 1594 "Maximum recent zevents records to retain for duplicate checking"); 1595 ZFS_MODULE_PARAM(zfs_zevent, zfs_zevent_, retain_expire_secs, UINT, ZMOD_RW, 1596 "Expiration time for recent zevents records"); 1597 #endif /* _KERNEL */ 1598