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 (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012 by Delphix. All rights reserved. 24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved. 25 * Copyright (c) 2016, 2017, Intel Corporation. 26 * Copyright (c) 2017 Open-E, Inc. All Rights Reserved. 27 */ 28 29 /* 30 * ZFS syseventd module. 31 * 32 * file origin: openzfs/usr/src/cmd/syseventd/modules/zfs_mod/zfs_mod.c 33 * 34 * The purpose of this module is to identify when devices are added to the 35 * system, and appropriately online or replace the affected vdevs. 36 * 37 * When a device is added to the system: 38 * 39 * 1. Search for any vdevs whose devid matches that of the newly added 40 * device. 41 * 42 * 2. If no vdevs are found, then search for any vdevs whose udev path 43 * matches that of the new device. 44 * 45 * 3. If no vdevs match by either method, then ignore the event. 46 * 47 * 4. Attempt to online the device with a flag to indicate that it should 48 * be unspared when resilvering completes. If this succeeds, then the 49 * same device was inserted and we should continue normally. 50 * 51 * 5. If the pool does not have the 'autoreplace' property set, attempt to 52 * online the device again without the unspare flag, which will 53 * generate a FMA fault. 54 * 55 * 6. If the pool has the 'autoreplace' property set, and the matching vdev 56 * is a whole disk, then label the new disk and attempt a 'zpool 57 * replace'. 58 * 59 * The module responds to EC_DEV_ADD events. The special ESC_ZFS_VDEV_CHECK 60 * event indicates that a device failed to open during pool load, but the 61 * autoreplace property was set. In this case, we deferred the associated 62 * FMA fault until our module had a chance to process the autoreplace logic. 63 * If the device could not be replaced, then the second online attempt will 64 * trigger the FMA fault that we skipped earlier. 65 * 66 * On Linux udev provides a disk insert for both the disk and the partition. 67 */ 68 69 #include <ctype.h> 70 #include <fcntl.h> 71 #include <libnvpair.h> 72 #include <libzfs.h> 73 #include <libzutil.h> 74 #include <limits.h> 75 #include <stddef.h> 76 #include <stdlib.h> 77 #include <string.h> 78 #include <syslog.h> 79 #include <sys/list.h> 80 #include <sys/sunddi.h> 81 #include <sys/sysevent/eventdefs.h> 82 #include <sys/sysevent/dev.h> 83 #include <thread_pool.h> 84 #include <pthread.h> 85 #include <unistd.h> 86 #include <errno.h> 87 #include "zfs_agents.h" 88 #include "../zed_log.h" 89 90 #define DEV_BYID_PATH "/dev/disk/by-id/" 91 #define DEV_BYPATH_PATH "/dev/disk/by-path/" 92 #define DEV_BYVDEV_PATH "/dev/disk/by-vdev/" 93 94 typedef void (*zfs_process_func_t)(zpool_handle_t *, nvlist_t *, boolean_t); 95 96 libzfs_handle_t *g_zfshdl; 97 list_t g_pool_list; /* list of unavailable pools at initialization */ 98 list_t g_device_list; /* list of disks with asynchronous label request */ 99 tpool_t *g_tpool; 100 boolean_t g_enumeration_done; 101 pthread_t g_zfs_tid; /* zfs_enum_pools() thread */ 102 103 typedef struct unavailpool { 104 zpool_handle_t *uap_zhp; 105 list_node_t uap_node; 106 } unavailpool_t; 107 108 typedef struct pendingdev { 109 char pd_physpath[128]; 110 list_node_t pd_node; 111 } pendingdev_t; 112 113 static int 114 zfs_toplevel_state(zpool_handle_t *zhp) 115 { 116 nvlist_t *nvroot; 117 vdev_stat_t *vs; 118 unsigned int c; 119 120 verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL), 121 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 122 verify(nvlist_lookup_uint64_array(nvroot, ZPOOL_CONFIG_VDEV_STATS, 123 (uint64_t **)&vs, &c) == 0); 124 return (vs->vs_state); 125 } 126 127 static int 128 zfs_unavail_pool(zpool_handle_t *zhp, void *data) 129 { 130 zed_log_msg(LOG_INFO, "zfs_unavail_pool: examining '%s' (state %d)", 131 zpool_get_name(zhp), (int)zfs_toplevel_state(zhp)); 132 133 if (zfs_toplevel_state(zhp) < VDEV_STATE_DEGRADED) { 134 unavailpool_t *uap; 135 uap = malloc(sizeof (unavailpool_t)); 136 if (uap == NULL) { 137 perror("malloc"); 138 exit(EXIT_FAILURE); 139 } 140 141 uap->uap_zhp = zhp; 142 list_insert_tail((list_t *)data, uap); 143 } else { 144 zpool_close(zhp); 145 } 146 return (0); 147 } 148 149 /* 150 * Two stage replace on Linux 151 * since we get disk notifications 152 * we can wait for partitioned disk slice to show up! 153 * 154 * First stage tags the disk, initiates async partitioning, and returns 155 * Second stage finds the tag and proceeds to ZFS labeling/replace 156 * 157 * disk-add --> label-disk + tag-disk --> partition-add --> zpool_vdev_attach 158 * 159 * 1. physical match with no fs, no partition 160 * tag it top, partition disk 161 * 162 * 2. physical match again, see partition and tag 163 * 164 */ 165 166 /* 167 * The device associated with the given vdev (either by devid or physical path) 168 * has been added to the system. If 'isdisk' is set, then we only attempt a 169 * replacement if it's a whole disk. This also implies that we should label the 170 * disk first. 171 * 172 * First, we attempt to online the device (making sure to undo any spare 173 * operation when finished). If this succeeds, then we're done. If it fails, 174 * and the new state is VDEV_CANT_OPEN, it indicates that the device was opened, 175 * but that the label was not what we expected. If the 'autoreplace' property 176 * is enabled, then we relabel the disk (if specified), and attempt a 'zpool 177 * replace'. If the online is successful, but the new state is something else 178 * (REMOVED or FAULTED), it indicates that we're out of sync or in some sort of 179 * race, and we should avoid attempting to relabel the disk. 180 * 181 * Also can arrive here from a ESC_ZFS_VDEV_CHECK event 182 */ 183 static void 184 zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t labeled) 185 { 186 char *path; 187 vdev_state_t newstate; 188 nvlist_t *nvroot, *newvd; 189 pendingdev_t *device; 190 uint64_t wholedisk = 0ULL; 191 uint64_t offline = 0ULL, faulted = 0ULL; 192 uint64_t guid = 0ULL; 193 char *physpath = NULL, *new_devid = NULL, *enc_sysfs_path = NULL; 194 char rawpath[PATH_MAX], fullpath[PATH_MAX]; 195 char devpath[PATH_MAX]; 196 int ret; 197 boolean_t is_sd = B_FALSE; 198 boolean_t is_mpath_wholedisk = B_FALSE; 199 uint_t c; 200 vdev_stat_t *vs; 201 202 if (nvlist_lookup_string(vdev, ZPOOL_CONFIG_PATH, &path) != 0) 203 return; 204 205 /* Skip healthy disks */ 206 verify(nvlist_lookup_uint64_array(vdev, ZPOOL_CONFIG_VDEV_STATS, 207 (uint64_t **)&vs, &c) == 0); 208 if (vs->vs_state == VDEV_STATE_HEALTHY) { 209 zed_log_msg(LOG_INFO, "%s: %s is already healthy, skip it.", 210 __func__, path); 211 return; 212 } 213 214 (void) nvlist_lookup_string(vdev, ZPOOL_CONFIG_PHYS_PATH, &physpath); 215 (void) nvlist_lookup_string(vdev, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, 216 &enc_sysfs_path); 217 (void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK, &wholedisk); 218 (void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_OFFLINE, &offline); 219 (void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_FAULTED, &faulted); 220 221 (void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_GUID, &guid); 222 223 /* 224 * Special case: 225 * 226 * We've seen times where a disk won't have a ZPOOL_CONFIG_PHYS_PATH 227 * entry in their config. For example, on this force-faulted disk: 228 * 229 * children[0]: 230 * type: 'disk' 231 * id: 0 232 * guid: 14309659774640089719 233 * path: '/dev/disk/by-vdev/L28' 234 * whole_disk: 0 235 * DTL: 654 236 * create_txg: 4 237 * com.delphix:vdev_zap_leaf: 1161 238 * faulted: 1 239 * aux_state: 'external' 240 * children[1]: 241 * type: 'disk' 242 * id: 1 243 * guid: 16002508084177980912 244 * path: '/dev/disk/by-vdev/L29' 245 * devid: 'dm-uuid-mpath-35000c500a61d68a3' 246 * phys_path: 'L29' 247 * vdev_enc_sysfs_path: '/sys/class/enclosure/0:0:1:0/SLOT 30 32' 248 * whole_disk: 0 249 * DTL: 1028 250 * create_txg: 4 251 * com.delphix:vdev_zap_leaf: 131 252 * 253 * If the disk's path is a /dev/disk/by-vdev/ path, then we can infer 254 * the ZPOOL_CONFIG_PHYS_PATH from the by-vdev disk name. 255 */ 256 if (physpath == NULL && path != NULL) { 257 /* If path begins with "/dev/disk/by-vdev/" ... */ 258 if (strncmp(path, DEV_BYVDEV_PATH, 259 strlen(DEV_BYVDEV_PATH)) == 0) { 260 /* Set physpath to the char after "/dev/disk/by-vdev" */ 261 physpath = &path[strlen(DEV_BYVDEV_PATH)]; 262 } 263 } 264 265 /* 266 * We don't want to autoreplace offlined disks. However, we do want to 267 * replace force-faulted disks (`zpool offline -f`). Force-faulted 268 * disks have both offline=1 and faulted=1 in the nvlist. 269 */ 270 if (offline && !faulted) { 271 zed_log_msg(LOG_INFO, "%s: %s is offline, skip autoreplace", 272 __func__, path); 273 return; 274 } 275 276 is_mpath_wholedisk = is_mpath_whole_disk(path); 277 zed_log_msg(LOG_INFO, "zfs_process_add: pool '%s' vdev '%s', phys '%s'" 278 " %s blank disk, %s mpath blank disk, %s labeled, enc sysfs '%s', " 279 "(guid %llu)", 280 zpool_get_name(zhp), path, 281 physpath ? physpath : "NULL", 282 wholedisk ? "is" : "not", 283 is_mpath_wholedisk? "is" : "not", 284 labeled ? "is" : "not", 285 enc_sysfs_path, 286 (long long unsigned int)guid); 287 288 /* 289 * The VDEV guid is preferred for identification (gets passed in path) 290 */ 291 if (guid != 0) { 292 (void) snprintf(fullpath, sizeof (fullpath), "%llu", 293 (long long unsigned int)guid); 294 } else { 295 /* 296 * otherwise use path sans partition suffix for whole disks 297 */ 298 (void) strlcpy(fullpath, path, sizeof (fullpath)); 299 if (wholedisk) { 300 char *spath = zfs_strip_partition(fullpath); 301 if (!spath) { 302 zed_log_msg(LOG_INFO, "%s: Can't alloc", 303 __func__); 304 return; 305 } 306 307 (void) strlcpy(fullpath, spath, sizeof (fullpath)); 308 free(spath); 309 } 310 } 311 312 /* 313 * Attempt to online the device. 314 */ 315 if (zpool_vdev_online(zhp, fullpath, 316 ZFS_ONLINE_CHECKREMOVE | ZFS_ONLINE_UNSPARE, &newstate) == 0 && 317 (newstate == VDEV_STATE_HEALTHY || 318 newstate == VDEV_STATE_DEGRADED)) { 319 zed_log_msg(LOG_INFO, 320 " zpool_vdev_online: vdev '%s' ('%s') is " 321 "%s", fullpath, physpath, (newstate == VDEV_STATE_HEALTHY) ? 322 "HEALTHY" : "DEGRADED"); 323 return; 324 } 325 326 /* 327 * vdev_id alias rule for using scsi_debug devices (FMA automated 328 * testing) 329 */ 330 if (physpath != NULL && strcmp("scsidebug", physpath) == 0) 331 is_sd = B_TRUE; 332 333 /* 334 * If the pool doesn't have the autoreplace property set, then use 335 * vdev online to trigger a FMA fault by posting an ereport. 336 */ 337 if (!zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOREPLACE, NULL) || 338 !(wholedisk || is_mpath_wholedisk) || (physpath == NULL)) { 339 (void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT, 340 &newstate); 341 zed_log_msg(LOG_INFO, "Pool's autoreplace is not enabled or " 342 "not a blank disk for '%s' ('%s')", fullpath, 343 physpath); 344 return; 345 } 346 347 /* 348 * Convert physical path into its current device node. Rawpath 349 * needs to be /dev/disk/by-vdev for a scsi_debug device since 350 * /dev/disk/by-path will not be present. 351 */ 352 (void) snprintf(rawpath, sizeof (rawpath), "%s%s", 353 is_sd ? DEV_BYVDEV_PATH : DEV_BYPATH_PATH, physpath); 354 355 if (realpath(rawpath, devpath) == NULL && !is_mpath_wholedisk) { 356 zed_log_msg(LOG_INFO, " realpath: %s failed (%s)", 357 rawpath, strerror(errno)); 358 359 (void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT, 360 &newstate); 361 362 zed_log_msg(LOG_INFO, " zpool_vdev_online: %s FORCEFAULT (%s)", 363 fullpath, libzfs_error_description(g_zfshdl)); 364 return; 365 } 366 367 /* Only autoreplace bad disks */ 368 if ((vs->vs_state != VDEV_STATE_DEGRADED) && 369 (vs->vs_state != VDEV_STATE_FAULTED) && 370 (vs->vs_state != VDEV_STATE_CANT_OPEN)) { 371 zed_log_msg(LOG_INFO, " not autoreplacing since disk isn't in " 372 "a bad state (currently %llu)", vs->vs_state); 373 return; 374 } 375 376 nvlist_lookup_string(vdev, "new_devid", &new_devid); 377 378 if (is_mpath_wholedisk) { 379 /* Don't label device mapper or multipath disks. */ 380 } else if (!labeled) { 381 /* 382 * we're auto-replacing a raw disk, so label it first 383 */ 384 char *leafname; 385 386 /* 387 * If this is a request to label a whole disk, then attempt to 388 * write out the label. Before we can label the disk, we need 389 * to map the physical string that was matched on to the under 390 * lying device node. 391 * 392 * If any part of this process fails, then do a force online 393 * to trigger a ZFS fault for the device (and any hot spare 394 * replacement). 395 */ 396 leafname = strrchr(devpath, '/') + 1; 397 398 /* 399 * If this is a request to label a whole disk, then attempt to 400 * write out the label. 401 */ 402 if (zpool_label_disk(g_zfshdl, zhp, leafname) != 0) { 403 zed_log_msg(LOG_INFO, " zpool_label_disk: could not " 404 "label '%s' (%s)", leafname, 405 libzfs_error_description(g_zfshdl)); 406 407 (void) zpool_vdev_online(zhp, fullpath, 408 ZFS_ONLINE_FORCEFAULT, &newstate); 409 return; 410 } 411 412 /* 413 * The disk labeling is asynchronous on Linux. Just record 414 * this label request and return as there will be another 415 * disk add event for the partition after the labeling is 416 * completed. 417 */ 418 device = malloc(sizeof (pendingdev_t)); 419 if (device == NULL) { 420 perror("malloc"); 421 exit(EXIT_FAILURE); 422 } 423 424 (void) strlcpy(device->pd_physpath, physpath, 425 sizeof (device->pd_physpath)); 426 list_insert_tail(&g_device_list, device); 427 428 zed_log_msg(LOG_INFO, " zpool_label_disk: async '%s' (%llu)", 429 leafname, (u_longlong_t)guid); 430 431 return; /* resumes at EC_DEV_ADD.ESC_DISK for partition */ 432 433 } else /* labeled */ { 434 boolean_t found = B_FALSE; 435 /* 436 * match up with request above to label the disk 437 */ 438 for (device = list_head(&g_device_list); device != NULL; 439 device = list_next(&g_device_list, device)) { 440 if (strcmp(physpath, device->pd_physpath) == 0) { 441 list_remove(&g_device_list, device); 442 free(device); 443 found = B_TRUE; 444 break; 445 } 446 zed_log_msg(LOG_INFO, "zpool_label_disk: %s != %s", 447 physpath, device->pd_physpath); 448 } 449 if (!found) { 450 /* unexpected partition slice encountered */ 451 zed_log_msg(LOG_INFO, "labeled disk %s unexpected here", 452 fullpath); 453 (void) zpool_vdev_online(zhp, fullpath, 454 ZFS_ONLINE_FORCEFAULT, &newstate); 455 return; 456 } 457 458 zed_log_msg(LOG_INFO, " zpool_label_disk: resume '%s' (%llu)", 459 physpath, (u_longlong_t)guid); 460 461 (void) snprintf(devpath, sizeof (devpath), "%s%s", 462 DEV_BYID_PATH, new_devid); 463 } 464 465 /* 466 * Construct the root vdev to pass to zpool_vdev_attach(). While adding 467 * the entire vdev structure is harmless, we construct a reduced set of 468 * path/physpath/wholedisk to keep it simple. 469 */ 470 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0) { 471 zed_log_msg(LOG_WARNING, "zfs_mod: nvlist_alloc out of memory"); 472 return; 473 } 474 if (nvlist_alloc(&newvd, NV_UNIQUE_NAME, 0) != 0) { 475 zed_log_msg(LOG_WARNING, "zfs_mod: nvlist_alloc out of memory"); 476 nvlist_free(nvroot); 477 return; 478 } 479 480 if (nvlist_add_string(newvd, ZPOOL_CONFIG_TYPE, VDEV_TYPE_DISK) != 0 || 481 nvlist_add_string(newvd, ZPOOL_CONFIG_PATH, path) != 0 || 482 nvlist_add_string(newvd, ZPOOL_CONFIG_DEVID, new_devid) != 0 || 483 (physpath != NULL && nvlist_add_string(newvd, 484 ZPOOL_CONFIG_PHYS_PATH, physpath) != 0) || 485 (enc_sysfs_path != NULL && nvlist_add_string(newvd, 486 ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, enc_sysfs_path) != 0) || 487 nvlist_add_uint64(newvd, ZPOOL_CONFIG_WHOLE_DISK, wholedisk) != 0 || 488 nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) != 0 || 489 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 490 (const nvlist_t **)&newvd, 1) != 0) { 491 zed_log_msg(LOG_WARNING, "zfs_mod: unable to add nvlist pairs"); 492 nvlist_free(newvd); 493 nvlist_free(nvroot); 494 return; 495 } 496 497 nvlist_free(newvd); 498 499 /* 500 * Wait for udev to verify the links exist, then auto-replace 501 * the leaf disk at same physical location. 502 */ 503 if (zpool_label_disk_wait(path, 3000) != 0) { 504 zed_log_msg(LOG_WARNING, "zfs_mod: expected replacement " 505 "disk %s is missing", path); 506 nvlist_free(nvroot); 507 return; 508 } 509 510 /* 511 * Prefer sequential resilvering when supported (mirrors and dRAID), 512 * otherwise fallback to a traditional healing resilver. 513 */ 514 ret = zpool_vdev_attach(zhp, fullpath, path, nvroot, B_TRUE, B_TRUE); 515 if (ret != 0) { 516 ret = zpool_vdev_attach(zhp, fullpath, path, nvroot, 517 B_TRUE, B_FALSE); 518 } 519 520 zed_log_msg(LOG_INFO, " zpool_vdev_replace: %s with %s (%s)", 521 fullpath, path, (ret == 0) ? "no errors" : 522 libzfs_error_description(g_zfshdl)); 523 524 nvlist_free(nvroot); 525 } 526 527 /* 528 * Utility functions to find a vdev matching given criteria. 529 */ 530 typedef struct dev_data { 531 const char *dd_compare; 532 const char *dd_prop; 533 zfs_process_func_t dd_func; 534 boolean_t dd_found; 535 boolean_t dd_islabeled; 536 uint64_t dd_pool_guid; 537 uint64_t dd_vdev_guid; 538 uint64_t dd_new_vdev_guid; 539 const char *dd_new_devid; 540 } dev_data_t; 541 542 static void 543 zfs_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *data) 544 { 545 dev_data_t *dp = data; 546 char *path = NULL; 547 uint_t c, children; 548 nvlist_t **child; 549 uint64_t guid = 0; 550 551 /* 552 * First iterate over any children. 553 */ 554 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, 555 &child, &children) == 0) { 556 for (c = 0; c < children; c++) 557 zfs_iter_vdev(zhp, child[c], data); 558 } 559 560 /* 561 * Iterate over any spares and cache devices 562 */ 563 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_SPARES, 564 &child, &children) == 0) { 565 for (c = 0; c < children; c++) 566 zfs_iter_vdev(zhp, child[c], data); 567 } 568 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_L2CACHE, 569 &child, &children) == 0) { 570 for (c = 0; c < children; c++) 571 zfs_iter_vdev(zhp, child[c], data); 572 } 573 574 /* once a vdev was matched and processed there is nothing left to do */ 575 if (dp->dd_found) 576 return; 577 (void) nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_GUID, &guid); 578 579 /* 580 * Match by GUID if available otherwise fallback to devid or physical 581 */ 582 if (dp->dd_vdev_guid != 0) { 583 if (guid != dp->dd_vdev_guid) 584 return; 585 zed_log_msg(LOG_INFO, " zfs_iter_vdev: matched on %llu", guid); 586 dp->dd_found = B_TRUE; 587 588 } else if (dp->dd_compare != NULL) { 589 /* 590 * NOTE: On Linux there is an event for partition, so unlike 591 * illumos, substring matching is not required to accommodate 592 * the partition suffix. An exact match will be present in 593 * the dp->dd_compare value. 594 * If the attached disk already contains a vdev GUID, it means 595 * the disk is not clean. In such a scenario, the physical path 596 * would be a match that makes the disk faulted when trying to 597 * online it. So, we would only want to proceed if either GUID 598 * matches with the last attached disk or the disk is in clean 599 * state. 600 */ 601 if (nvlist_lookup_string(nvl, dp->dd_prop, &path) != 0 || 602 strcmp(dp->dd_compare, path) != 0) { 603 zed_log_msg(LOG_INFO, " %s: no match (%s != vdev %s)", 604 __func__, dp->dd_compare, path); 605 return; 606 } 607 if (dp->dd_new_vdev_guid != 0 && dp->dd_new_vdev_guid != guid) { 608 zed_log_msg(LOG_INFO, " %s: no match (GUID:%llu" 609 " != vdev GUID:%llu)", __func__, 610 dp->dd_new_vdev_guid, guid); 611 return; 612 } 613 614 zed_log_msg(LOG_INFO, " zfs_iter_vdev: matched %s on %s", 615 dp->dd_prop, path); 616 dp->dd_found = B_TRUE; 617 618 /* pass the new devid for use by replacing code */ 619 if (dp->dd_new_devid != NULL) { 620 (void) nvlist_add_string(nvl, "new_devid", 621 dp->dd_new_devid); 622 } 623 } 624 625 (dp->dd_func)(zhp, nvl, dp->dd_islabeled); 626 } 627 628 static void 629 zfs_enable_ds(void *arg) 630 { 631 unavailpool_t *pool = (unavailpool_t *)arg; 632 633 (void) zpool_enable_datasets(pool->uap_zhp, NULL, 0); 634 zpool_close(pool->uap_zhp); 635 free(pool); 636 } 637 638 static int 639 zfs_iter_pool(zpool_handle_t *zhp, void *data) 640 { 641 nvlist_t *config, *nvl; 642 dev_data_t *dp = data; 643 uint64_t pool_guid; 644 unavailpool_t *pool; 645 646 zed_log_msg(LOG_INFO, "zfs_iter_pool: evaluating vdevs on %s (by %s)", 647 zpool_get_name(zhp), dp->dd_vdev_guid ? "GUID" : dp->dd_prop); 648 649 /* 650 * For each vdev in this pool, look for a match to apply dd_func 651 */ 652 if ((config = zpool_get_config(zhp, NULL)) != NULL) { 653 if (dp->dd_pool_guid == 0 || 654 (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 655 &pool_guid) == 0 && pool_guid == dp->dd_pool_guid)) { 656 (void) nvlist_lookup_nvlist(config, 657 ZPOOL_CONFIG_VDEV_TREE, &nvl); 658 zfs_iter_vdev(zhp, nvl, data); 659 } 660 } else { 661 zed_log_msg(LOG_INFO, "%s: no config\n", __func__); 662 } 663 664 /* 665 * if this pool was originally unavailable, 666 * then enable its datasets asynchronously 667 */ 668 if (g_enumeration_done) { 669 for (pool = list_head(&g_pool_list); pool != NULL; 670 pool = list_next(&g_pool_list, pool)) { 671 672 if (strcmp(zpool_get_name(zhp), 673 zpool_get_name(pool->uap_zhp))) 674 continue; 675 if (zfs_toplevel_state(zhp) >= VDEV_STATE_DEGRADED) { 676 list_remove(&g_pool_list, pool); 677 (void) tpool_dispatch(g_tpool, zfs_enable_ds, 678 pool); 679 break; 680 } 681 } 682 } 683 684 zpool_close(zhp); 685 return (dp->dd_found); /* cease iteration after a match */ 686 } 687 688 /* 689 * Given a physical device location, iterate over all 690 * (pool, vdev) pairs which correspond to that location. 691 */ 692 static boolean_t 693 devphys_iter(const char *physical, const char *devid, zfs_process_func_t func, 694 boolean_t is_slice, uint64_t new_vdev_guid) 695 { 696 dev_data_t data = { 0 }; 697 698 data.dd_compare = physical; 699 data.dd_func = func; 700 data.dd_prop = ZPOOL_CONFIG_PHYS_PATH; 701 data.dd_found = B_FALSE; 702 data.dd_islabeled = is_slice; 703 data.dd_new_devid = devid; /* used by auto replace code */ 704 data.dd_new_vdev_guid = new_vdev_guid; 705 706 (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data); 707 708 return (data.dd_found); 709 } 710 711 /* 712 * Given a device identifier, find any vdevs with a matching by-vdev 713 * path. Normally we shouldn't need this as the comparison would be 714 * made earlier in the devphys_iter(). For example, if we were replacing 715 * /dev/disk/by-vdev/L28, normally devphys_iter() would match the 716 * ZPOOL_CONFIG_PHYS_PATH of "L28" from the old disk config to "L28" 717 * of the new disk config. However, we've seen cases where 718 * ZPOOL_CONFIG_PHYS_PATH was not in the config for the old disk. Here's 719 * an example of a real 2-disk mirror pool where one disk was force 720 * faulted: 721 * 722 * com.delphix:vdev_zap_top: 129 723 * children[0]: 724 * type: 'disk' 725 * id: 0 726 * guid: 14309659774640089719 727 * path: '/dev/disk/by-vdev/L28' 728 * whole_disk: 0 729 * DTL: 654 730 * create_txg: 4 731 * com.delphix:vdev_zap_leaf: 1161 732 * faulted: 1 733 * aux_state: 'external' 734 * children[1]: 735 * type: 'disk' 736 * id: 1 737 * guid: 16002508084177980912 738 * path: '/dev/disk/by-vdev/L29' 739 * devid: 'dm-uuid-mpath-35000c500a61d68a3' 740 * phys_path: 'L29' 741 * vdev_enc_sysfs_path: '/sys/class/enclosure/0:0:1:0/SLOT 30 32' 742 * whole_disk: 0 743 * DTL: 1028 744 * create_txg: 4 745 * com.delphix:vdev_zap_leaf: 131 746 * 747 * So in the case above, the only thing we could compare is the path. 748 * 749 * We can do this because we assume by-vdev paths are authoritative as physical 750 * paths. We could not assume this for normal paths like /dev/sda since the 751 * physical location /dev/sda points to could change over time. 752 */ 753 static boolean_t 754 by_vdev_path_iter(const char *by_vdev_path, const char *devid, 755 zfs_process_func_t func, boolean_t is_slice) 756 { 757 dev_data_t data = { 0 }; 758 759 data.dd_compare = by_vdev_path; 760 data.dd_func = func; 761 data.dd_prop = ZPOOL_CONFIG_PATH; 762 data.dd_found = B_FALSE; 763 data.dd_islabeled = is_slice; 764 data.dd_new_devid = devid; 765 766 if (strncmp(by_vdev_path, DEV_BYVDEV_PATH, 767 strlen(DEV_BYVDEV_PATH)) != 0) { 768 /* by_vdev_path doesn't start with "/dev/disk/by-vdev/" */ 769 return (B_FALSE); 770 } 771 772 (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data); 773 774 return (data.dd_found); 775 } 776 777 /* 778 * Given a device identifier, find any vdevs with a matching devid. 779 * On Linux we can match devid directly which is always a whole disk. 780 */ 781 static boolean_t 782 devid_iter(const char *devid, zfs_process_func_t func, boolean_t is_slice) 783 { 784 dev_data_t data = { 0 }; 785 786 data.dd_compare = devid; 787 data.dd_func = func; 788 data.dd_prop = ZPOOL_CONFIG_DEVID; 789 data.dd_found = B_FALSE; 790 data.dd_islabeled = is_slice; 791 data.dd_new_devid = devid; 792 793 (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data); 794 795 return (data.dd_found); 796 } 797 798 /* 799 * Given a device guid, find any vdevs with a matching guid. 800 */ 801 static boolean_t 802 guid_iter(uint64_t pool_guid, uint64_t vdev_guid, const char *devid, 803 zfs_process_func_t func, boolean_t is_slice) 804 { 805 dev_data_t data = { 0 }; 806 807 data.dd_func = func; 808 data.dd_found = B_FALSE; 809 data.dd_pool_guid = pool_guid; 810 data.dd_vdev_guid = vdev_guid; 811 data.dd_islabeled = is_slice; 812 data.dd_new_devid = devid; 813 814 (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data); 815 816 return (data.dd_found); 817 } 818 819 /* 820 * Handle a EC_DEV_ADD.ESC_DISK event. 821 * 822 * illumos 823 * Expects: DEV_PHYS_PATH string in schema 824 * Matches: vdev's ZPOOL_CONFIG_PHYS_PATH or ZPOOL_CONFIG_DEVID 825 * 826 * path: '/dev/dsk/c0t1d0s0' (persistent) 827 * devid: 'id1,sd@SATA_____Hitachi_HDS72101______JP2940HZ3H74MC/a' 828 * phys_path: '/pci@0,0/pci103c,1609@11/disk@1,0:a' 829 * 830 * linux 831 * provides: DEV_PHYS_PATH and DEV_IDENTIFIER strings in schema 832 * Matches: vdev's ZPOOL_CONFIG_PHYS_PATH or ZPOOL_CONFIG_DEVID 833 * 834 * path: '/dev/sdc1' (not persistent) 835 * devid: 'ata-SAMSUNG_HD204UI_S2HGJD2Z805891-part1' 836 * phys_path: 'pci-0000:04:00.0-sas-0x4433221106000000-lun-0' 837 */ 838 static int 839 zfs_deliver_add(nvlist_t *nvl) 840 { 841 char *devpath = NULL, *devid = NULL; 842 uint64_t pool_guid = 0, vdev_guid = 0; 843 boolean_t is_slice; 844 845 /* 846 * Expecting a devid string and an optional physical location and guid 847 */ 848 if (nvlist_lookup_string(nvl, DEV_IDENTIFIER, &devid) != 0) { 849 zed_log_msg(LOG_INFO, "%s: no dev identifier\n", __func__); 850 return (-1); 851 } 852 853 (void) nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devpath); 854 (void) nvlist_lookup_uint64(nvl, ZFS_EV_POOL_GUID, &pool_guid); 855 (void) nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID, &vdev_guid); 856 857 is_slice = (nvlist_lookup_boolean(nvl, DEV_IS_PART) == 0); 858 859 zed_log_msg(LOG_INFO, "zfs_deliver_add: adding %s (%s) (is_slice %d)", 860 devid, devpath ? devpath : "NULL", is_slice); 861 862 /* 863 * Iterate over all vdevs looking for a match in the following order: 864 * 1. ZPOOL_CONFIG_DEVID (identifies the unique disk) 865 * 2. ZPOOL_CONFIG_PHYS_PATH (identifies disk physical location). 866 * 3. ZPOOL_CONFIG_GUID (identifies unique vdev). 867 * 4. ZPOOL_CONFIG_PATH for /dev/disk/by-vdev devices only (since 868 * by-vdev paths represent physical paths). 869 */ 870 if (devid_iter(devid, zfs_process_add, is_slice)) 871 return (0); 872 if (devpath != NULL && devphys_iter(devpath, devid, zfs_process_add, 873 is_slice, vdev_guid)) 874 return (0); 875 if (vdev_guid != 0) 876 (void) guid_iter(pool_guid, vdev_guid, devid, zfs_process_add, 877 is_slice); 878 879 if (devpath != NULL) { 880 /* Can we match a /dev/disk/by-vdev/ path? */ 881 char by_vdev_path[MAXPATHLEN]; 882 snprintf(by_vdev_path, sizeof (by_vdev_path), 883 "/dev/disk/by-vdev/%s", devpath); 884 if (by_vdev_path_iter(by_vdev_path, devid, zfs_process_add, 885 is_slice)) 886 return (0); 887 } 888 889 return (0); 890 } 891 892 /* 893 * Called when we receive a VDEV_CHECK event, which indicates a device could not 894 * be opened during initial pool open, but the autoreplace property was set on 895 * the pool. In this case, we treat it as if it were an add event. 896 */ 897 static int 898 zfs_deliver_check(nvlist_t *nvl) 899 { 900 dev_data_t data = { 0 }; 901 902 if (nvlist_lookup_uint64(nvl, ZFS_EV_POOL_GUID, 903 &data.dd_pool_guid) != 0 || 904 nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID, 905 &data.dd_vdev_guid) != 0 || 906 data.dd_vdev_guid == 0) 907 return (0); 908 909 zed_log_msg(LOG_INFO, "zfs_deliver_check: pool '%llu', vdev %llu", 910 data.dd_pool_guid, data.dd_vdev_guid); 911 912 data.dd_func = zfs_process_add; 913 914 (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data); 915 916 return (0); 917 } 918 919 /* 920 * Given a path to a vdev, lookup the vdev's physical size from its 921 * config nvlist. 922 * 923 * Returns the vdev's physical size in bytes on success, 0 on error. 924 */ 925 static uint64_t 926 vdev_size_from_config(zpool_handle_t *zhp, const char *vdev_path) 927 { 928 nvlist_t *nvl = NULL; 929 boolean_t avail_spare, l2cache, log; 930 vdev_stat_t *vs = NULL; 931 uint_t c; 932 933 nvl = zpool_find_vdev(zhp, vdev_path, &avail_spare, &l2cache, &log); 934 if (!nvl) 935 return (0); 936 937 verify(nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_VDEV_STATS, 938 (uint64_t **)&vs, &c) == 0); 939 if (!vs) { 940 zed_log_msg(LOG_INFO, "%s: no nvlist for '%s'", __func__, 941 vdev_path); 942 return (0); 943 } 944 945 return (vs->vs_pspace); 946 } 947 948 /* 949 * Given a path to a vdev, lookup if the vdev is a "whole disk" in the 950 * config nvlist. "whole disk" means that ZFS was passed a whole disk 951 * at pool creation time, which it partitioned up and has full control over. 952 * Thus a partition with wholedisk=1 set tells us that zfs created the 953 * partition at creation time. A partition without whole disk set would have 954 * been created by externally (like with fdisk) and passed to ZFS. 955 * 956 * Returns the whole disk value (either 0 or 1). 957 */ 958 static uint64_t 959 vdev_whole_disk_from_config(zpool_handle_t *zhp, const char *vdev_path) 960 { 961 nvlist_t *nvl = NULL; 962 boolean_t avail_spare, l2cache, log; 963 uint64_t wholedisk = 0; 964 965 nvl = zpool_find_vdev(zhp, vdev_path, &avail_spare, &l2cache, &log); 966 if (!nvl) 967 return (0); 968 969 (void) nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_WHOLE_DISK, &wholedisk); 970 971 return (wholedisk); 972 } 973 974 /* 975 * If the device size grew more than 1% then return true. 976 */ 977 #define DEVICE_GREW(oldsize, newsize) \ 978 ((newsize > oldsize) && \ 979 ((newsize / (newsize - oldsize)) <= 100)) 980 981 static int 982 zfsdle_vdev_online(zpool_handle_t *zhp, void *data) 983 { 984 boolean_t avail_spare, l2cache; 985 nvlist_t *udev_nvl = data; 986 nvlist_t *tgt; 987 int error; 988 989 char *tmp_devname, devname[MAXPATHLEN] = ""; 990 uint64_t guid; 991 992 if (nvlist_lookup_uint64(udev_nvl, ZFS_EV_VDEV_GUID, &guid) == 0) { 993 sprintf(devname, "%llu", (u_longlong_t)guid); 994 } else if (nvlist_lookup_string(udev_nvl, DEV_PHYS_PATH, 995 &tmp_devname) == 0) { 996 strlcpy(devname, tmp_devname, MAXPATHLEN); 997 zfs_append_partition(devname, MAXPATHLEN); 998 } else { 999 zed_log_msg(LOG_INFO, "%s: no guid or physpath", __func__); 1000 } 1001 1002 zed_log_msg(LOG_INFO, "zfsdle_vdev_online: searching for '%s' in '%s'", 1003 devname, zpool_get_name(zhp)); 1004 1005 if ((tgt = zpool_find_vdev_by_physpath(zhp, devname, 1006 &avail_spare, &l2cache, NULL)) != NULL) { 1007 char *path, fullpath[MAXPATHLEN]; 1008 uint64_t wholedisk = 0; 1009 1010 error = nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH, &path); 1011 if (error) { 1012 zpool_close(zhp); 1013 return (0); 1014 } 1015 1016 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK, 1017 &wholedisk); 1018 1019 if (wholedisk) { 1020 path = strrchr(path, '/'); 1021 if (path != NULL) { 1022 path = zfs_strip_partition(path + 1); 1023 if (path == NULL) { 1024 zpool_close(zhp); 1025 return (0); 1026 } 1027 } else { 1028 zpool_close(zhp); 1029 return (0); 1030 } 1031 1032 (void) strlcpy(fullpath, path, sizeof (fullpath)); 1033 free(path); 1034 1035 /* 1036 * We need to reopen the pool associated with this 1037 * device so that the kernel can update the size of 1038 * the expanded device. When expanding there is no 1039 * need to restart the scrub from the beginning. 1040 */ 1041 boolean_t scrub_restart = B_FALSE; 1042 (void) zpool_reopen_one(zhp, &scrub_restart); 1043 } else { 1044 (void) strlcpy(fullpath, path, sizeof (fullpath)); 1045 } 1046 1047 if (zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) { 1048 vdev_state_t newstate; 1049 1050 if (zpool_get_state(zhp) != POOL_STATE_UNAVAIL) { 1051 /* 1052 * If this disk size has not changed, then 1053 * there's no need to do an autoexpand. To 1054 * check we look at the disk's size in its 1055 * config, and compare it to the disk size 1056 * that udev is reporting. 1057 */ 1058 uint64_t udev_size = 0, conf_size = 0, 1059 wholedisk = 0, udev_parent_size = 0; 1060 1061 /* 1062 * Get the size of our disk that udev is 1063 * reporting. 1064 */ 1065 if (nvlist_lookup_uint64(udev_nvl, DEV_SIZE, 1066 &udev_size) != 0) { 1067 udev_size = 0; 1068 } 1069 1070 /* 1071 * Get the size of our disk's parent device 1072 * from udev (where sda1's parent is sda). 1073 */ 1074 if (nvlist_lookup_uint64(udev_nvl, 1075 DEV_PARENT_SIZE, &udev_parent_size) != 0) { 1076 udev_parent_size = 0; 1077 } 1078 1079 conf_size = vdev_size_from_config(zhp, 1080 fullpath); 1081 1082 wholedisk = vdev_whole_disk_from_config(zhp, 1083 fullpath); 1084 1085 /* 1086 * Only attempt an autoexpand if the vdev size 1087 * changed. There are two different cases 1088 * to consider. 1089 * 1090 * 1. wholedisk=1 1091 * If you do a 'zpool create' on a whole disk 1092 * (like /dev/sda), then zfs will create 1093 * partitions on the disk (like /dev/sda1). In 1094 * that case, wholedisk=1 will be set in the 1095 * partition's nvlist config. So zed will need 1096 * to see if your parent device (/dev/sda) 1097 * expanded in size, and if so, then attempt 1098 * the autoexpand. 1099 * 1100 * 2. wholedisk=0 1101 * If you do a 'zpool create' on an existing 1102 * partition, or a device that doesn't allow 1103 * partitions, then wholedisk=0, and you will 1104 * simply need to check if the device itself 1105 * expanded in size. 1106 */ 1107 if (DEVICE_GREW(conf_size, udev_size) || 1108 (wholedisk && DEVICE_GREW(conf_size, 1109 udev_parent_size))) { 1110 error = zpool_vdev_online(zhp, fullpath, 1111 0, &newstate); 1112 1113 zed_log_msg(LOG_INFO, 1114 "%s: autoexpanding '%s' from %llu" 1115 " to %llu bytes in pool '%s': %d", 1116 __func__, fullpath, conf_size, 1117 MAX(udev_size, udev_parent_size), 1118 zpool_get_name(zhp), error); 1119 } 1120 } 1121 } 1122 zpool_close(zhp); 1123 return (1); 1124 } 1125 zpool_close(zhp); 1126 return (0); 1127 } 1128 1129 /* 1130 * This function handles the ESC_DEV_DLE device change event. Use the 1131 * provided vdev guid when looking up a disk or partition, when the guid 1132 * is not present assume the entire disk is owned by ZFS and append the 1133 * expected -part1 partition information then lookup by physical path. 1134 */ 1135 static int 1136 zfs_deliver_dle(nvlist_t *nvl) 1137 { 1138 char *devname, name[MAXPATHLEN]; 1139 uint64_t guid; 1140 1141 if (nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID, &guid) == 0) { 1142 sprintf(name, "%llu", (u_longlong_t)guid); 1143 } else if (nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devname) == 0) { 1144 strlcpy(name, devname, MAXPATHLEN); 1145 zfs_append_partition(name, MAXPATHLEN); 1146 } else { 1147 sprintf(name, "unknown"); 1148 zed_log_msg(LOG_INFO, "zfs_deliver_dle: no guid or physpath"); 1149 } 1150 1151 if (zpool_iter(g_zfshdl, zfsdle_vdev_online, nvl) != 1) { 1152 zed_log_msg(LOG_INFO, "zfs_deliver_dle: device '%s' not " 1153 "found", name); 1154 return (1); 1155 } 1156 1157 return (0); 1158 } 1159 1160 /* 1161 * syseventd daemon module event handler 1162 * 1163 * Handles syseventd daemon zfs device related events: 1164 * 1165 * EC_DEV_ADD.ESC_DISK 1166 * EC_DEV_STATUS.ESC_DEV_DLE 1167 * EC_ZFS.ESC_ZFS_VDEV_CHECK 1168 * 1169 * Note: assumes only one thread active at a time (not thread safe) 1170 */ 1171 static int 1172 zfs_slm_deliver_event(const char *class, const char *subclass, nvlist_t *nvl) 1173 { 1174 int ret; 1175 boolean_t is_check = B_FALSE, is_dle = B_FALSE; 1176 1177 if (strcmp(class, EC_DEV_ADD) == 0) { 1178 /* 1179 * We're mainly interested in disk additions, but we also listen 1180 * for new loop devices, to allow for simplified testing. 1181 */ 1182 if (strcmp(subclass, ESC_DISK) != 0 && 1183 strcmp(subclass, ESC_LOFI) != 0) 1184 return (0); 1185 1186 is_check = B_FALSE; 1187 } else if (strcmp(class, EC_ZFS) == 0 && 1188 strcmp(subclass, ESC_ZFS_VDEV_CHECK) == 0) { 1189 /* 1190 * This event signifies that a device failed to open 1191 * during pool load, but the 'autoreplace' property was 1192 * set, so we should pretend it's just been added. 1193 */ 1194 is_check = B_TRUE; 1195 } else if (strcmp(class, EC_DEV_STATUS) == 0 && 1196 strcmp(subclass, ESC_DEV_DLE) == 0) { 1197 is_dle = B_TRUE; 1198 } else { 1199 return (0); 1200 } 1201 1202 if (is_dle) 1203 ret = zfs_deliver_dle(nvl); 1204 else if (is_check) 1205 ret = zfs_deliver_check(nvl); 1206 else 1207 ret = zfs_deliver_add(nvl); 1208 1209 return (ret); 1210 } 1211 1212 static void * 1213 zfs_enum_pools(void *arg) 1214 { 1215 (void) arg; 1216 1217 (void) zpool_iter(g_zfshdl, zfs_unavail_pool, (void *)&g_pool_list); 1218 /* 1219 * Linux - instead of using a thread pool, each list entry 1220 * will spawn a thread when an unavailable pool transitions 1221 * to available. zfs_slm_fini will wait for these threads. 1222 */ 1223 g_enumeration_done = B_TRUE; 1224 return (NULL); 1225 } 1226 1227 /* 1228 * called from zed daemon at startup 1229 * 1230 * sent messages from zevents or udev monitor 1231 * 1232 * For now, each agent has its own libzfs instance 1233 */ 1234 int 1235 zfs_slm_init(void) 1236 { 1237 if ((g_zfshdl = libzfs_init()) == NULL) 1238 return (-1); 1239 1240 /* 1241 * collect a list of unavailable pools (asynchronously, 1242 * since this can take a while) 1243 */ 1244 list_create(&g_pool_list, sizeof (struct unavailpool), 1245 offsetof(struct unavailpool, uap_node)); 1246 1247 if (pthread_create(&g_zfs_tid, NULL, zfs_enum_pools, NULL) != 0) { 1248 list_destroy(&g_pool_list); 1249 libzfs_fini(g_zfshdl); 1250 return (-1); 1251 } 1252 1253 pthread_setname_np(g_zfs_tid, "enum-pools"); 1254 list_create(&g_device_list, sizeof (struct pendingdev), 1255 offsetof(struct pendingdev, pd_node)); 1256 1257 return (0); 1258 } 1259 1260 void 1261 zfs_slm_fini(void) 1262 { 1263 unavailpool_t *pool; 1264 pendingdev_t *device; 1265 1266 /* wait for zfs_enum_pools thread to complete */ 1267 (void) pthread_join(g_zfs_tid, NULL); 1268 /* destroy the thread pool */ 1269 if (g_tpool != NULL) { 1270 tpool_wait(g_tpool); 1271 tpool_destroy(g_tpool); 1272 } 1273 1274 while ((pool = (list_head(&g_pool_list))) != NULL) { 1275 list_remove(&g_pool_list, pool); 1276 zpool_close(pool->uap_zhp); 1277 free(pool); 1278 } 1279 list_destroy(&g_pool_list); 1280 1281 while ((device = (list_head(&g_device_list))) != NULL) { 1282 list_remove(&g_device_list, device); 1283 free(device); 1284 } 1285 list_destroy(&g_device_list); 1286 1287 libzfs_fini(g_zfshdl); 1288 } 1289 1290 void 1291 zfs_slm_event(const char *class, const char *subclass, nvlist_t *nvl) 1292 { 1293 zed_log_msg(LOG_INFO, "zfs_slm_event: %s.%s", class, subclass); 1294 (void) zfs_slm_deliver_event(class, subclass, nvl); 1295 } 1296