13d7072f8Seschrock /* 23d7072f8Seschrock * CDDL HEADER START 33d7072f8Seschrock * 43d7072f8Seschrock * The contents of this file are subject to the terms of the 53d7072f8Seschrock * Common Development and Distribution License (the "License"). 63d7072f8Seschrock * You may not use this file except in compliance with the License. 73d7072f8Seschrock * 83d7072f8Seschrock * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 93d7072f8Seschrock * or http://www.opensolaris.org/os/licensing. 103d7072f8Seschrock * See the License for the specific language governing permissions 113d7072f8Seschrock * and limitations under the License. 123d7072f8Seschrock * 133d7072f8Seschrock * When distributing Covered Code, include this CDDL HEADER in each 143d7072f8Seschrock * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 153d7072f8Seschrock * If applicable, add the following below this CDDL HEADER, with the 163d7072f8Seschrock * fields enclosed by brackets "[]" replaced with your own identifying 173d7072f8Seschrock * information: Portions Copyright [yyyy] [name of copyright owner] 183d7072f8Seschrock * 193d7072f8Seschrock * CDDL HEADER END 203d7072f8Seschrock */ 213d7072f8Seschrock /* 22*b98131cfSEric Taylor * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. 233d7072f8Seschrock */ 243d7072f8Seschrock 253d7072f8Seschrock /* 263d7072f8Seschrock * ZFS syseventd module. 273d7072f8Seschrock * 283d7072f8Seschrock * The purpose of this module is to identify when devices are added to the 293d7072f8Seschrock * system, and appropriately online or replace the affected vdevs. 303d7072f8Seschrock * 313d7072f8Seschrock * When a device is added to the system: 323d7072f8Seschrock * 333d7072f8Seschrock * 1. Search for any vdevs whose devid matches that of the newly added 343d7072f8Seschrock * device. 353d7072f8Seschrock * 363d7072f8Seschrock * 2. If no vdevs are found, then search for any vdevs whose devfs path 373d7072f8Seschrock * matches that of the new device. 383d7072f8Seschrock * 393d7072f8Seschrock * 3. If no vdevs match by either method, then ignore the event. 403d7072f8Seschrock * 413d7072f8Seschrock * 4. Attempt to online the device with a flag to indicate that it should 423d7072f8Seschrock * be unspared when resilvering completes. If this succeeds, then the 433d7072f8Seschrock * same device was inserted and we should continue normally. 443d7072f8Seschrock * 453d7072f8Seschrock * 5. If the pool does not have the 'autoreplace' property set, attempt to 463d7072f8Seschrock * online the device again without the unspare flag, which will 473d7072f8Seschrock * generate a FMA fault. 483d7072f8Seschrock * 493d7072f8Seschrock * 6. If the pool has the 'autoreplace' property set, and the matching vdev 503d7072f8Seschrock * is a whole disk, then label the new disk and attempt a 'zpool 513d7072f8Seschrock * replace'. 523d7072f8Seschrock * 533d7072f8Seschrock * The module responds to EC_DEV_ADD events for both disks and lofi devices, 543d7072f8Seschrock * with the latter used for testing. The special ESC_ZFS_VDEV_CHECK event 553d7072f8Seschrock * indicates that a device failed to open during pool load, but the autoreplace 563d7072f8Seschrock * property was set. In this case, we deferred the associated FMA fault until 573d7072f8Seschrock * our module had a chance to process the autoreplace logic. If the device 583d7072f8Seschrock * could not be replaced, then the second online attempt will trigger the FMA 593d7072f8Seschrock * fault that we skipped earlier. 603d7072f8Seschrock */ 613d7072f8Seschrock 623d7072f8Seschrock #include <alloca.h> 633d7072f8Seschrock #include <devid.h> 643d7072f8Seschrock #include <fcntl.h> 653d7072f8Seschrock #include <libnvpair.h> 663d7072f8Seschrock #include <libsysevent.h> 673d7072f8Seschrock #include <libzfs.h> 683d7072f8Seschrock #include <limits.h> 693d7072f8Seschrock #include <stdlib.h> 703d7072f8Seschrock #include <string.h> 713d7072f8Seschrock #include <syslog.h> 723d7072f8Seschrock #include <sys/sunddi.h> 733d7072f8Seschrock #include <sys/sysevent/eventdefs.h> 743d7072f8Seschrock #include <sys/sysevent/dev.h> 753d7072f8Seschrock #include <unistd.h> 76*b98131cfSEric Taylor #include "syseventd.h" 773d7072f8Seschrock 783d7072f8Seschrock #if defined(__i386) || defined(__amd64) 793d7072f8Seschrock #define PHYS_PATH ":q" 803d7072f8Seschrock #define RAW_SLICE "p0" 813d7072f8Seschrock #elif defined(__sparc) 823d7072f8Seschrock #define PHYS_PATH ":c" 833d7072f8Seschrock #define RAW_SLICE "s2" 843d7072f8Seschrock #else 853d7072f8Seschrock #error Unknown architecture 863d7072f8Seschrock #endif 873d7072f8Seschrock 883d7072f8Seschrock typedef void (*zfs_process_func_t)(zpool_handle_t *, nvlist_t *, boolean_t); 893d7072f8Seschrock 903d7072f8Seschrock libzfs_handle_t *g_zfshdl; 913d7072f8Seschrock 923d7072f8Seschrock /* 933d7072f8Seschrock * The device associated with the given vdev (either by devid or physical path) 943d7072f8Seschrock * has been added to the system. If 'isdisk' is set, then we only attempt a 953d7072f8Seschrock * replacement if it's a whole disk. This also implies that we should label the 963d7072f8Seschrock * disk first. 973d7072f8Seschrock * 983d7072f8Seschrock * First, we attempt to online the device (making sure to undo any spare 993d7072f8Seschrock * operation when finished). If this succeeds, then we're done. If it fails, 1003d7072f8Seschrock * and the new state is VDEV_CANT_OPEN, it indicates that the device was opened, 1013d7072f8Seschrock * but that the label was not what we expected. If the 'autoreplace' property 1023d7072f8Seschrock * is not set, then we relabel the disk (if specified), and attempt a 'zpool 1033d7072f8Seschrock * replace'. If the online is successful, but the new state is something else 1043d7072f8Seschrock * (REMOVED or FAULTED), it indicates that we're out of sync or in some sort of 1053d7072f8Seschrock * race, and we should avoid attempting to relabel the disk. 1063d7072f8Seschrock */ 1073d7072f8Seschrock static void 1083d7072f8Seschrock zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t isdisk) 1093d7072f8Seschrock { 1103d7072f8Seschrock char *path; 1113d7072f8Seschrock vdev_state_t newstate; 1123d7072f8Seschrock nvlist_t *nvroot, *newvd; 1133d7072f8Seschrock uint64_t wholedisk = 0ULL; 114bf82a41bSeschrock char *physpath = NULL; 1153d7072f8Seschrock char rawpath[PATH_MAX], fullpath[PATH_MAX]; 1163d7072f8Seschrock size_t len; 1173d7072f8Seschrock 1183d7072f8Seschrock if (nvlist_lookup_string(vdev, ZPOOL_CONFIG_PATH, &path) != 0) 1193d7072f8Seschrock return; 1203d7072f8Seschrock 121bf82a41bSeschrock (void) nvlist_lookup_string(vdev, ZPOOL_CONFIG_PHYS_PATH, &physpath); 1223d7072f8Seschrock (void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK, &wholedisk); 1233d7072f8Seschrock 1243d7072f8Seschrock /* 1253d7072f8Seschrock * We should have a way to online a device by guid. With the current 1263d7072f8Seschrock * interface, we are forced to chop off the 's0' for whole disks. 1273d7072f8Seschrock */ 1283d7072f8Seschrock (void) strlcpy(fullpath, path, sizeof (fullpath)); 1293d7072f8Seschrock if (wholedisk) 1303d7072f8Seschrock fullpath[strlen(fullpath) - 2] = '\0'; 1313d7072f8Seschrock 1323d7072f8Seschrock /* 1333d7072f8Seschrock * Attempt to online the device. It would be nice to online this by 1343d7072f8Seschrock * GUID, but the current interface only supports lookup by path. 1353d7072f8Seschrock */ 1363d7072f8Seschrock if (zpool_vdev_online(zhp, fullpath, 1373d7072f8Seschrock ZFS_ONLINE_CHECKREMOVE | ZFS_ONLINE_UNSPARE, &newstate) == 0 && 138bf82a41bSeschrock (newstate == VDEV_STATE_HEALTHY || newstate == VDEV_STATE_DEGRADED)) 1393d7072f8Seschrock return; 1403d7072f8Seschrock 1413d7072f8Seschrock /* 1423d7072f8Seschrock * If the pool doesn't have the autoreplace property set, then attempt a 1433d7072f8Seschrock * true online (without the unspare flag), which will trigger a FMA 1443d7072f8Seschrock * fault. 1453d7072f8Seschrock */ 146990b4856Slling if (!zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOREPLACE, NULL) || 1473d7072f8Seschrock (isdisk && !wholedisk)) { 1483d7072f8Seschrock (void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT, 1493d7072f8Seschrock &newstate); 1503d7072f8Seschrock return; 1513d7072f8Seschrock } 1523d7072f8Seschrock 1533d7072f8Seschrock if (isdisk) { 1543d7072f8Seschrock /* 1553d7072f8Seschrock * If this is a request to label a whole disk, then attempt to 1563d7072f8Seschrock * write out the label. Before we can label the disk, we need 1573d7072f8Seschrock * access to a raw node. Ideally, we'd like to walk the devinfo 1583d7072f8Seschrock * tree and find a raw node from the corresponding parent node. 1593d7072f8Seschrock * This is overly complicated, and since we know how we labeled 1603d7072f8Seschrock * this device in the first place, we know it's save to switch 1613d7072f8Seschrock * from /dev/dsk to /dev/rdsk and append the backup slice. 162c5904d13Seschrock * 163c5904d13Seschrock * If any part of this process fails, then do a force online to 164c5904d13Seschrock * trigger a ZFS fault for the device (and any hot spare 165c5904d13Seschrock * replacement). 1663d7072f8Seschrock */ 167c5904d13Seschrock if (strncmp(path, "/dev/dsk/", 9) != 0) { 168c5904d13Seschrock (void) zpool_vdev_online(zhp, fullpath, 169c5904d13Seschrock ZFS_ONLINE_FORCEFAULT, &newstate); 1703d7072f8Seschrock return; 171c5904d13Seschrock } 1723d7072f8Seschrock 1733d7072f8Seschrock (void) strlcpy(rawpath, path + 9, sizeof (rawpath)); 1743d7072f8Seschrock len = strlen(rawpath); 1753d7072f8Seschrock rawpath[len - 2] = '\0'; 1763d7072f8Seschrock 177c5904d13Seschrock if (zpool_label_disk(g_zfshdl, zhp, rawpath) != 0) { 178c5904d13Seschrock (void) zpool_vdev_online(zhp, fullpath, 179c5904d13Seschrock ZFS_ONLINE_FORCEFAULT, &newstate); 1803d7072f8Seschrock return; 181c5904d13Seschrock } 1823d7072f8Seschrock } 1833d7072f8Seschrock 1843d7072f8Seschrock /* 1853d7072f8Seschrock * Cosntruct the root vdev to pass to zpool_vdev_attach(). While adding 1863d7072f8Seschrock * the entire vdev structure is harmless, we construct a reduced set of 187bf82a41bSeschrock * path/physpath/wholedisk to keep it simple. 1883d7072f8Seschrock */ 1893d7072f8Seschrock if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0) 1903d7072f8Seschrock return; 1913d7072f8Seschrock 1923d7072f8Seschrock if (nvlist_alloc(&newvd, NV_UNIQUE_NAME, 0) != 0) { 1933d7072f8Seschrock nvlist_free(nvroot); 1943d7072f8Seschrock return; 1953d7072f8Seschrock } 1963d7072f8Seschrock 1973d7072f8Seschrock if (nvlist_add_string(newvd, ZPOOL_CONFIG_TYPE, VDEV_TYPE_DISK) != 0 || 1983d7072f8Seschrock nvlist_add_string(newvd, ZPOOL_CONFIG_PATH, path) != 0 || 199bf82a41bSeschrock (physpath != NULL && nvlist_add_string(newvd, 200bf82a41bSeschrock ZPOOL_CONFIG_PHYS_PATH, physpath) != 0) || 2013d7072f8Seschrock nvlist_add_uint64(newvd, ZPOOL_CONFIG_WHOLE_DISK, wholedisk) != 0 || 2023d7072f8Seschrock nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) != 0 || 2033d7072f8Seschrock nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &newvd, 2043d7072f8Seschrock 1) != 0) { 2053d7072f8Seschrock nvlist_free(newvd); 2063d7072f8Seschrock nvlist_free(nvroot); 2073d7072f8Seschrock return; 2083d7072f8Seschrock } 2093d7072f8Seschrock 2103d7072f8Seschrock nvlist_free(newvd); 2113d7072f8Seschrock 2123d7072f8Seschrock (void) zpool_vdev_attach(zhp, fullpath, path, nvroot, B_TRUE); 2133d7072f8Seschrock 2143d7072f8Seschrock nvlist_free(nvroot); 2153d7072f8Seschrock 2163d7072f8Seschrock } 2173d7072f8Seschrock 2183d7072f8Seschrock /* 2193d7072f8Seschrock * Utility functions to find a vdev matching given criteria. 2203d7072f8Seschrock */ 2213d7072f8Seschrock typedef struct dev_data { 2223d7072f8Seschrock const char *dd_compare; 2233d7072f8Seschrock const char *dd_prop; 2243d7072f8Seschrock zfs_process_func_t dd_func; 2253d7072f8Seschrock boolean_t dd_found; 2263d7072f8Seschrock boolean_t dd_isdisk; 2273d7072f8Seschrock uint64_t dd_pool_guid; 2283d7072f8Seschrock uint64_t dd_vdev_guid; 2293d7072f8Seschrock } dev_data_t; 2303d7072f8Seschrock 2313d7072f8Seschrock static void 2323d7072f8Seschrock zfs_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *data) 2333d7072f8Seschrock { 2343d7072f8Seschrock dev_data_t *dp = data; 2353d7072f8Seschrock char *path; 2363d7072f8Seschrock uint_t c, children; 2373d7072f8Seschrock nvlist_t **child; 238b01c3b58Seschrock size_t len; 2393d7072f8Seschrock uint64_t guid; 2403d7072f8Seschrock 2413d7072f8Seschrock /* 2423d7072f8Seschrock * First iterate over any children. 2433d7072f8Seschrock */ 2443d7072f8Seschrock if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, 2453d7072f8Seschrock &child, &children) == 0) { 2463d7072f8Seschrock for (c = 0; c < children; c++) 2473d7072f8Seschrock zfs_iter_vdev(zhp, child[c], data); 2483d7072f8Seschrock return; 2493d7072f8Seschrock } 2503d7072f8Seschrock 2513d7072f8Seschrock if (dp->dd_vdev_guid != 0) { 2523d7072f8Seschrock if (nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_GUID, 2533d7072f8Seschrock &guid) != 0 || guid != dp->dd_vdev_guid) 2543d7072f8Seschrock return; 2553d7072f8Seschrock } else { 256b01c3b58Seschrock len = strlen(dp->dd_compare); 257b01c3b58Seschrock 2583d7072f8Seschrock if (nvlist_lookup_string(nvl, dp->dd_prop, &path) != 0 || 2593d7072f8Seschrock strncmp(dp->dd_compare, path, len) != 0) 2603d7072f8Seschrock return; 2613d7072f8Seschrock 2623d7072f8Seschrock /* 2633d7072f8Seschrock * Normally, we want to have an exact match for the comparison 2643d7072f8Seschrock * string. However, we allow substring matches in the following 2653d7072f8Seschrock * cases: 2663d7072f8Seschrock * 2673d7072f8Seschrock * <path>: This is a devpath, and the target is one 2683d7072f8Seschrock * of its children. 2693d7072f8Seschrock * 2703d7072f8Seschrock * <path/> This is a devid for a whole disk, and 2713d7072f8Seschrock * the target is one of its children. 2723d7072f8Seschrock */ 2733d7072f8Seschrock if (path[len] != '\0' && path[len] != ':' && 2743d7072f8Seschrock path[len - 1] != '/') 2753d7072f8Seschrock return; 2763d7072f8Seschrock } 2773d7072f8Seschrock 2783d7072f8Seschrock (dp->dd_func)(zhp, nvl, dp->dd_isdisk); 2793d7072f8Seschrock } 2803d7072f8Seschrock 2813d7072f8Seschrock static int 2823d7072f8Seschrock zfs_iter_pool(zpool_handle_t *zhp, void *data) 2833d7072f8Seschrock { 2843d7072f8Seschrock nvlist_t *config, *nvl; 2853d7072f8Seschrock dev_data_t *dp = data; 2863d7072f8Seschrock uint64_t pool_guid; 2873d7072f8Seschrock 2883d7072f8Seschrock if ((config = zpool_get_config(zhp, NULL)) != NULL) { 2893d7072f8Seschrock if (dp->dd_pool_guid == 0 || 2903d7072f8Seschrock (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 2913d7072f8Seschrock &pool_guid) == 0 && pool_guid == dp->dd_pool_guid)) { 2923d7072f8Seschrock (void) nvlist_lookup_nvlist(config, 2933d7072f8Seschrock ZPOOL_CONFIG_VDEV_TREE, &nvl); 2943d7072f8Seschrock zfs_iter_vdev(zhp, nvl, data); 2953d7072f8Seschrock } 2963d7072f8Seschrock } 2973d7072f8Seschrock 2983d7072f8Seschrock zpool_close(zhp); 2993d7072f8Seschrock return (0); 3003d7072f8Seschrock } 3013d7072f8Seschrock 3023d7072f8Seschrock /* 3033d7072f8Seschrock * Given a physical device path, iterate over all (pool, vdev) pairs which 3043d7072f8Seschrock * correspond to the given path. 3053d7072f8Seschrock */ 3063d7072f8Seschrock static boolean_t 3073d7072f8Seschrock devpath_iter(const char *devpath, zfs_process_func_t func, boolean_t wholedisk) 3083d7072f8Seschrock { 3093d7072f8Seschrock dev_data_t data = { 0 }; 3103d7072f8Seschrock 3113d7072f8Seschrock data.dd_compare = devpath; 3123d7072f8Seschrock data.dd_func = func; 3133d7072f8Seschrock data.dd_prop = ZPOOL_CONFIG_PHYS_PATH; 3143d7072f8Seschrock data.dd_found = B_FALSE; 3153d7072f8Seschrock data.dd_isdisk = wholedisk; 3163d7072f8Seschrock 3173d7072f8Seschrock (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data); 3183d7072f8Seschrock 3193d7072f8Seschrock return (data.dd_found); 3203d7072f8Seschrock } 3213d7072f8Seschrock 3223d7072f8Seschrock /* 3233d7072f8Seschrock * Given a /devices path, lookup the corresponding devid for each minor node, 3243d7072f8Seschrock * and find any vdevs with matching devids. Doing this straight up would be 3253d7072f8Seschrock * rather inefficient, O(minor nodes * vdevs in system), so we take advantage of 3263d7072f8Seschrock * the fact that each devid ends with "/<minornode>". Once we find any valid 3273d7072f8Seschrock * minor node, we chop off the portion after the last slash, and then search for 3283d7072f8Seschrock * matching vdevs, which is O(vdevs in system). 3293d7072f8Seschrock */ 3303d7072f8Seschrock static boolean_t 3313d7072f8Seschrock devid_iter(const char *devpath, zfs_process_func_t func, boolean_t wholedisk) 3323d7072f8Seschrock { 3333d7072f8Seschrock size_t len = strlen(devpath) + sizeof ("/devices") + 3343d7072f8Seschrock sizeof (PHYS_PATH) - 1; 3353d7072f8Seschrock char *fullpath; 3363d7072f8Seschrock int fd; 3373d7072f8Seschrock ddi_devid_t devid; 3383d7072f8Seschrock char *devidstr, *fulldevid; 3393d7072f8Seschrock dev_data_t data = { 0 }; 3403d7072f8Seschrock 3413d7072f8Seschrock /* 3423d7072f8Seschrock * Try to open a known minor node. 3433d7072f8Seschrock */ 3443d7072f8Seschrock fullpath = alloca(len); 3453d7072f8Seschrock (void) snprintf(fullpath, len, "/devices%s%s", devpath, PHYS_PATH); 3463d7072f8Seschrock if ((fd = open(fullpath, O_RDONLY)) < 0) 3473d7072f8Seschrock return (B_FALSE); 3483d7072f8Seschrock 3493d7072f8Seschrock /* 3503d7072f8Seschrock * Determine the devid as a string, with no trailing slash for the minor 3513d7072f8Seschrock * node. 3523d7072f8Seschrock */ 3533d7072f8Seschrock if (devid_get(fd, &devid) != 0) { 3543d7072f8Seschrock (void) close(fd); 3553d7072f8Seschrock return (B_FALSE); 3563d7072f8Seschrock } 3573d7072f8Seschrock (void) close(fd); 3583d7072f8Seschrock 3593d7072f8Seschrock if ((devidstr = devid_str_encode(devid, NULL)) == NULL) { 3603d7072f8Seschrock devid_free(devid); 3613d7072f8Seschrock return (B_FALSE); 3623d7072f8Seschrock } 3633d7072f8Seschrock 3643d7072f8Seschrock len = strlen(devidstr) + 2; 3653d7072f8Seschrock fulldevid = alloca(len); 3663d7072f8Seschrock (void) snprintf(fulldevid, len, "%s/", devidstr); 3673d7072f8Seschrock 3683d7072f8Seschrock data.dd_compare = fulldevid; 3693d7072f8Seschrock data.dd_func = func; 3703d7072f8Seschrock data.dd_prop = ZPOOL_CONFIG_DEVID; 3713d7072f8Seschrock data.dd_found = B_FALSE; 3723d7072f8Seschrock data.dd_isdisk = wholedisk; 3733d7072f8Seschrock 3743d7072f8Seschrock (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data); 3753d7072f8Seschrock 3763d7072f8Seschrock devid_str_free(devidstr); 3773d7072f8Seschrock 3783d7072f8Seschrock return (data.dd_found); 3793d7072f8Seschrock } 3803d7072f8Seschrock 3813d7072f8Seschrock /* 3823d7072f8Seschrock * This function is called when we receive a devfs add event. This can be 3833d7072f8Seschrock * either a disk event or a lofi event, and the behavior is slightly different 3843d7072f8Seschrock * depending on which it is. 3853d7072f8Seschrock */ 3863d7072f8Seschrock static int 3873d7072f8Seschrock zfs_deliver_add(nvlist_t *nvl, boolean_t is_lofi) 3883d7072f8Seschrock { 3893d7072f8Seschrock char *devpath, *devname; 3903d7072f8Seschrock char path[PATH_MAX], realpath[PATH_MAX]; 3913d7072f8Seschrock char *colon, *raw; 3923d7072f8Seschrock int ret; 3933d7072f8Seschrock 3943d7072f8Seschrock /* 3953d7072f8Seschrock * The main unit of operation is the physical device path. For disks, 3963d7072f8Seschrock * this is the device node, as all minor nodes are affected. For lofi 3973d7072f8Seschrock * devices, this includes the minor path. Unfortunately, this isn't 3983d7072f8Seschrock * represented in the DEV_PHYS_PATH for various reasons. 3993d7072f8Seschrock */ 4003d7072f8Seschrock if (nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devpath) != 0) 4013d7072f8Seschrock return (-1); 4023d7072f8Seschrock 4033d7072f8Seschrock /* 4043d7072f8Seschrock * If this is a lofi device, then also get the minor instance name. 4053d7072f8Seschrock * Unfortunately, the current payload doesn't include an easy way to get 4063d7072f8Seschrock * this information. So we cheat by resolving the 'dev_name' (which 4073d7072f8Seschrock * refers to the raw device) and taking the portion between ':(*),raw'. 4083d7072f8Seschrock */ 4093d7072f8Seschrock (void) strlcpy(realpath, devpath, sizeof (realpath)); 4103d7072f8Seschrock if (is_lofi) { 4113d7072f8Seschrock if (nvlist_lookup_string(nvl, DEV_NAME, 4123d7072f8Seschrock &devname) == 0 && 4133d7072f8Seschrock (ret = resolvepath(devname, path, 4143d7072f8Seschrock sizeof (path))) > 0) { 4153d7072f8Seschrock path[ret] = '\0'; 4163d7072f8Seschrock colon = strchr(path, ':'); 4173d7072f8Seschrock if (colon != NULL) 4183d7072f8Seschrock raw = strstr(colon + 1, ",raw"); 4193d7072f8Seschrock if (colon != NULL && raw != NULL) { 4203d7072f8Seschrock *raw = '\0'; 4213d7072f8Seschrock (void) snprintf(realpath, 4223d7072f8Seschrock sizeof (realpath), "%s%s", 4233d7072f8Seschrock devpath, colon); 4243d7072f8Seschrock *raw = ','; 4253d7072f8Seschrock } 4263d7072f8Seschrock } 4273d7072f8Seschrock } 4283d7072f8Seschrock 4293d7072f8Seschrock /* 4303d7072f8Seschrock * Iterate over all vdevs with a matching devid, and then those with a 4313d7072f8Seschrock * matching /devices path. For disks, we only want to pay attention to 4323d7072f8Seschrock * vdevs marked as whole disks. For lofi, we don't care (because we're 4333d7072f8Seschrock * matching an exact minor name). 4343d7072f8Seschrock */ 4353d7072f8Seschrock if (!devid_iter(realpath, zfs_process_add, !is_lofi)) 4363d7072f8Seschrock (void) devpath_iter(realpath, zfs_process_add, !is_lofi); 4373d7072f8Seschrock 4383d7072f8Seschrock return (0); 4393d7072f8Seschrock } 4403d7072f8Seschrock 4413d7072f8Seschrock /* 4423d7072f8Seschrock * Called when we receive a VDEV_CHECK event, which indicates a device could not 4433d7072f8Seschrock * be opened during initial pool open, but the autoreplace property was set on 4443d7072f8Seschrock * the pool. In this case, we treat it as if it were an add event. 4453d7072f8Seschrock */ 4463d7072f8Seschrock static int 4473d7072f8Seschrock zfs_deliver_check(nvlist_t *nvl) 4483d7072f8Seschrock { 4493d7072f8Seschrock dev_data_t data = { 0 }; 4503d7072f8Seschrock 4513d7072f8Seschrock if (nvlist_lookup_uint64(nvl, ZFS_EV_POOL_GUID, 4523d7072f8Seschrock &data.dd_pool_guid) != 0 || 4533d7072f8Seschrock nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID, 4543d7072f8Seschrock &data.dd_vdev_guid) != 0) 4553d7072f8Seschrock return (0); 4563d7072f8Seschrock 4573d7072f8Seschrock data.dd_isdisk = B_TRUE; 4583d7072f8Seschrock data.dd_func = zfs_process_add; 4593d7072f8Seschrock 4603d7072f8Seschrock (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data); 4613d7072f8Seschrock 4623d7072f8Seschrock return (0); 4633d7072f8Seschrock } 4643d7072f8Seschrock 465*b98131cfSEric Taylor #define DEVICE_PREFIX "/devices" 466*b98131cfSEric Taylor 467*b98131cfSEric Taylor static int 468*b98131cfSEric Taylor zfsdle_vdev_online(zpool_handle_t *zhp, void *data) 469*b98131cfSEric Taylor { 470*b98131cfSEric Taylor char *devname = data; 471*b98131cfSEric Taylor boolean_t avail_spare, l2cache; 472*b98131cfSEric Taylor vdev_state_t newstate; 473*b98131cfSEric Taylor nvlist_t *tgt; 474*b98131cfSEric Taylor 475*b98131cfSEric Taylor syseventd_print(9, "zfsdle_vdev_online: searching for %s in pool %s\n", 476*b98131cfSEric Taylor devname, zpool_get_name(zhp)); 477*b98131cfSEric Taylor 478*b98131cfSEric Taylor if ((tgt = zpool_find_vdev_by_physpath(zhp, devname, 479*b98131cfSEric Taylor &avail_spare, &l2cache, NULL)) != NULL) { 480*b98131cfSEric Taylor char *path, fullpath[MAXPATHLEN]; 481*b98131cfSEric Taylor uint64_t wholedisk = 0ULL; 482*b98131cfSEric Taylor 483*b98131cfSEric Taylor verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH, 484*b98131cfSEric Taylor &path) == 0); 485*b98131cfSEric Taylor verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK, 486*b98131cfSEric Taylor &wholedisk) == 0); 487*b98131cfSEric Taylor 488*b98131cfSEric Taylor (void) strlcpy(fullpath, path, sizeof (fullpath)); 489*b98131cfSEric Taylor if (wholedisk) 490*b98131cfSEric Taylor fullpath[strlen(fullpath) - 2] = '\0'; 491*b98131cfSEric Taylor 492*b98131cfSEric Taylor if (zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) { 493*b98131cfSEric Taylor syseventd_print(9, "zfsdle_vdev_online: setting device" 494*b98131cfSEric Taylor " device %s to ONLINE state in pool %s.\n", 495*b98131cfSEric Taylor fullpath, zpool_get_name(zhp)); 496*b98131cfSEric Taylor if (zpool_get_state(zhp) != POOL_STATE_UNAVAIL) 497*b98131cfSEric Taylor (void) zpool_vdev_online(zhp, fullpath, 0, 498*b98131cfSEric Taylor &newstate); 499*b98131cfSEric Taylor } 500*b98131cfSEric Taylor return (1); 501*b98131cfSEric Taylor } 502*b98131cfSEric Taylor return (0); 503*b98131cfSEric Taylor } 504*b98131cfSEric Taylor 505*b98131cfSEric Taylor int 506*b98131cfSEric Taylor zfs_deliver_dle(nvlist_t *nvl) 507*b98131cfSEric Taylor { 508*b98131cfSEric Taylor char *devname; 509*b98131cfSEric Taylor if (nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devname) != 0) { 510*b98131cfSEric Taylor syseventd_print(9, "zfs_deliver_event: no physpath\n"); 511*b98131cfSEric Taylor return (-1); 512*b98131cfSEric Taylor } 513*b98131cfSEric Taylor if (strncmp(devname, DEVICE_PREFIX, strlen(DEVICE_PREFIX)) != 0) { 514*b98131cfSEric Taylor syseventd_print(9, "zfs_deliver_event: invalid " 515*b98131cfSEric Taylor "device '%s'", devname); 516*b98131cfSEric Taylor return (-1); 517*b98131cfSEric Taylor } 518*b98131cfSEric Taylor 519*b98131cfSEric Taylor /* 520*b98131cfSEric Taylor * We try to find the device using the physical 521*b98131cfSEric Taylor * path that has been supplied. We need to strip off 522*b98131cfSEric Taylor * the /devices prefix before starting our search. 523*b98131cfSEric Taylor */ 524*b98131cfSEric Taylor devname += strlen(DEVICE_PREFIX); 525*b98131cfSEric Taylor if (zpool_iter(g_zfshdl, zfsdle_vdev_online, devname) != 1) { 526*b98131cfSEric Taylor syseventd_print(9, "zfs_deliver_event: device '%s' not" 527*b98131cfSEric Taylor " found\n", devname); 528*b98131cfSEric Taylor return (1); 529*b98131cfSEric Taylor } 530*b98131cfSEric Taylor nvlist_free(nvl); 531*b98131cfSEric Taylor return (0); 532*b98131cfSEric Taylor } 533*b98131cfSEric Taylor 534*b98131cfSEric Taylor 5353d7072f8Seschrock /*ARGSUSED*/ 5363d7072f8Seschrock static int 5373d7072f8Seschrock zfs_deliver_event(sysevent_t *ev, int unused) 5383d7072f8Seschrock { 5393d7072f8Seschrock const char *class = sysevent_get_class_name(ev); 5403d7072f8Seschrock const char *subclass = sysevent_get_subclass_name(ev); 5413d7072f8Seschrock nvlist_t *nvl; 5423d7072f8Seschrock int ret; 543*b98131cfSEric Taylor boolean_t is_lofi, is_check, is_dle = B_FALSE; 5443d7072f8Seschrock 5453d7072f8Seschrock if (strcmp(class, EC_DEV_ADD) == 0) { 5463d7072f8Seschrock /* 5473d7072f8Seschrock * We're mainly interested in disk additions, but we also listen 5483d7072f8Seschrock * for new lofi devices, to allow for simplified testing. 5493d7072f8Seschrock */ 5503d7072f8Seschrock if (strcmp(subclass, ESC_DISK) == 0) 5513d7072f8Seschrock is_lofi = B_FALSE; 5523d7072f8Seschrock else if (strcmp(subclass, ESC_LOFI) == 0) 5533d7072f8Seschrock is_lofi = B_TRUE; 5543d7072f8Seschrock else 5553d7072f8Seschrock return (0); 5563d7072f8Seschrock 5573d7072f8Seschrock is_check = B_FALSE; 5583d7072f8Seschrock } else if (strcmp(class, EC_ZFS) == 0 && 5593d7072f8Seschrock strcmp(subclass, ESC_ZFS_VDEV_CHECK) == 0) { 5603d7072f8Seschrock /* 5613d7072f8Seschrock * This event signifies that a device failed to open during pool 5623d7072f8Seschrock * load, but the 'autoreplace' property was set, so we should 5633d7072f8Seschrock * pretend it's just been added. 5643d7072f8Seschrock */ 5653d7072f8Seschrock is_check = B_TRUE; 566*b98131cfSEric Taylor } else if (strcmp(class, EC_DEV_STATUS) == 0 && 567*b98131cfSEric Taylor strcmp(subclass, ESC_DEV_DLE) == 0) { 568*b98131cfSEric Taylor is_dle = B_TRUE; 5693d7072f8Seschrock } else { 5703d7072f8Seschrock return (0); 5713d7072f8Seschrock } 5723d7072f8Seschrock 5733d7072f8Seschrock if (sysevent_get_attr_list(ev, &nvl) != 0) 5743d7072f8Seschrock return (-1); 5753d7072f8Seschrock 576*b98131cfSEric Taylor if (is_dle) 577*b98131cfSEric Taylor ret = zfs_deliver_dle(nvl); 578*b98131cfSEric Taylor else if (is_check) 5793d7072f8Seschrock ret = zfs_deliver_check(nvl); 5803d7072f8Seschrock else 5813d7072f8Seschrock ret = zfs_deliver_add(nvl, is_lofi); 5823d7072f8Seschrock 5833d7072f8Seschrock nvlist_free(nvl); 5843d7072f8Seschrock return (ret); 5853d7072f8Seschrock } 5863d7072f8Seschrock 5873d7072f8Seschrock static struct slm_mod_ops zfs_mod_ops = { 5883d7072f8Seschrock SE_MAJOR_VERSION, SE_MINOR_VERSION, 10, zfs_deliver_event 5893d7072f8Seschrock }; 5903d7072f8Seschrock 5913d7072f8Seschrock struct slm_mod_ops * 5923d7072f8Seschrock slm_init() 5933d7072f8Seschrock { 5943d7072f8Seschrock if ((g_zfshdl = libzfs_init()) == NULL) 5953d7072f8Seschrock return (NULL); 5963d7072f8Seschrock 5973d7072f8Seschrock return (&zfs_mod_ops); 5983d7072f8Seschrock } 5993d7072f8Seschrock 6003d7072f8Seschrock void 6013d7072f8Seschrock slm_fini() 6023d7072f8Seschrock { 6033d7072f8Seschrock } 604