xref: /illumos-gate/usr/src/cmd/syseventd/modules/zfs_mod/zfs_mod.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2012 by Delphix. All rights reserved.
 * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
 */

/*
 * ZFS syseventd module.
 *
 * The purpose of this module is to identify when devices are added to the
 * system, and appropriately online or replace the affected vdevs.
 *
 * When a device is added to the system:
 *
 * 	1. Search for any vdevs whose devid matches that of the newly added
 *	   device.
 *
 * 	2. If no vdevs are found, then search for any vdevs whose devfs path
 *	   matches that of the new device.
 *
 *	3. If no vdevs match by either method, then ignore the event.
 *
 * 	4. Attempt to online the device with a flag to indicate that it should
 *	   be unspared when resilvering completes.  If this succeeds, then the
 *	   same device was inserted and we should continue normally.
 *
 *	5. If the pool does not have the 'autoreplace' property set, attempt to
 *	   online the device again without the unspare flag, which will
 *	   generate a FMA fault.
 *
 *	6. If the pool has the 'autoreplace' property set, and the matching vdev
 *	   is a whole disk, then label the new disk and attempt a 'zpool
 *	   replace'.
 *
 * The module responds to EC_DEV_ADD events for both disks and lofi devices,
 * with the latter used for testing.  The special ESC_ZFS_VDEV_CHECK event
 * indicates that a device failed to open during pool load, but the autoreplace
 * property was set.  In this case, we deferred the associated FMA fault until
 * our module had a chance to process the autoreplace logic.  If the device
 * could not be replaced, then the second online attempt will trigger the FMA
 * fault that we skipped earlier.
 */

#include <alloca.h>
#include <devid.h>
#include <fcntl.h>
#include <libnvpair.h>
#include <libsysevent.h>
#include <libzfs.h>
#include <limits.h>
#include <stdlib.h>
#include <string.h>
#include <syslog.h>
#include <sys/list.h>
#include <sys/sunddi.h>
#include <sys/sysevent/eventdefs.h>
#include <sys/sysevent/dev.h>
#include <thread_pool.h>
#include <unistd.h>
#include "syseventd.h"

#if defined(__i386) || defined(__amd64)
#define	PHYS_PATH	":q"
#define	RAW_SLICE	"p0"
#elif defined(__sparc)
#define	PHYS_PATH	":c"
#define	RAW_SLICE	"s2"
#else
#error Unknown architecture
#endif

typedef void (*zfs_process_func_t)(zpool_handle_t *, nvlist_t *, boolean_t);

libzfs_handle_t *g_zfshdl;
list_t g_pool_list;
tpool_t *g_tpool;
boolean_t g_enumeration_done;
thread_t g_zfs_tid;

typedef struct unavailpool {
	zpool_handle_t	*uap_zhp;
	list_node_t	uap_node;
} unavailpool_t;

int
zfs_toplevel_state(zpool_handle_t *zhp)
{
	nvlist_t *nvroot;
	vdev_stat_t *vs;
	unsigned int c;

	verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
	verify(nvlist_lookup_uint64_array(nvroot, ZPOOL_CONFIG_VDEV_STATS,
	    (uint64_t **)&vs, &c) == 0);
	return (vs->vs_state);
}

static int
zfs_unavail_pool(zpool_handle_t *zhp, void *data)
{
	if (zfs_toplevel_state(zhp) < VDEV_STATE_DEGRADED) {
		unavailpool_t *uap;
		uap = malloc(sizeof (unavailpool_t));
		uap->uap_zhp = zhp;
		list_insert_tail((list_t *)data, uap);
	} else {
		zpool_close(zhp);
	}
	return (0);
}

/*
 * The device associated with the given vdev (either by devid or physical path)
 * has been added to the system.  If 'isdisk' is set, then we only attempt a
 * replacement if it's a whole disk.  This also implies that we should label the
 * disk first.
 *
 * First, we attempt to online the device (making sure to undo any spare
 * operation when finished).  If this succeeds, then we're done.  If it fails,
 * and the new state is VDEV_CANT_OPEN, it indicates that the device was opened,
 * but that the label was not what we expected.  If the 'autoreplace' property
 * is not set, then we relabel the disk (if specified), and attempt a 'zpool
 * replace'.  If the online is successful, but the new state is something else
 * (REMOVED or FAULTED), it indicates that we're out of sync or in some sort of
 * race, and we should avoid attempting to relabel the disk.
 */
static void
zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t isdisk)
{
	char *path;
	vdev_state_t newstate;
	nvlist_t *nvroot, *newvd;
	uint64_t wholedisk = 0ULL;
	uint64_t offline = 0ULL;
	char *physpath = NULL;
	char rawpath[PATH_MAX], fullpath[PATH_MAX];
	zpool_boot_label_t boot_type;
	uint64_t boot_size;
	size_t len;

	if (nvlist_lookup_string(vdev, ZPOOL_CONFIG_PATH, &path) != 0)
		return;

	(void) nvlist_lookup_string(vdev, ZPOOL_CONFIG_PHYS_PATH, &physpath);
	(void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK, &wholedisk);
	(void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_OFFLINE, &offline);

	/*
	 * We should have a way to online a device by guid.  With the current
	 * interface, we are forced to chop off the 's0' for whole disks.
	 */
	(void) strlcpy(fullpath, path, sizeof (fullpath));
	if (wholedisk)
		fullpath[strlen(fullpath) - 2] = '\0';

	/*
	 * Attempt to online the device.  It would be nice to online this by
	 * GUID, but the current interface only supports lookup by path.
	 */
	if (offline ||
	    (zpool_vdev_online(zhp, fullpath,
	    ZFS_ONLINE_CHECKREMOVE | ZFS_ONLINE_UNSPARE, &newstate) == 0 &&
	    (newstate == VDEV_STATE_HEALTHY ||
	    newstate == VDEV_STATE_DEGRADED)))
		return;

	/*
	 * If the pool doesn't have the autoreplace property set, then attempt a
	 * true online (without the unspare flag), which will trigger a FMA
	 * fault.
	 */
	if (!zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOREPLACE, NULL) ||
	    (isdisk && !wholedisk)) {
		(void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT,
		    &newstate);
		return;
	}

	if (isdisk) {
		/*
		 * If this is a request to label a whole disk, then attempt to
		 * write out the label.  Before we can label the disk, we need
		 * access to a raw node.  Ideally, we'd like to walk the devinfo
		 * tree and find a raw node from the corresponding parent node.
		 * This is overly complicated, and since we know how we labeled
		 * this device in the first place, we know it's save to switch
		 * from /dev/dsk to /dev/rdsk and append the backup slice.
		 *
		 * If any part of this process fails, then do a force online to
		 * trigger a ZFS fault for the device (and any hot spare
		 * replacement).
		 */
		if (strncmp(path, ZFS_DISK_ROOTD,
		    strlen(ZFS_DISK_ROOTD)) != 0) {
			(void) zpool_vdev_online(zhp, fullpath,
			    ZFS_ONLINE_FORCEFAULT, &newstate);
			return;
		}

		(void) strlcpy(rawpath, path + 9, sizeof (rawpath));
		len = strlen(rawpath);
		rawpath[len - 2] = '\0';

		if (zpool_is_bootable(zhp))
			boot_type = ZPOOL_COPY_BOOT_LABEL;
		else
			boot_type = ZPOOL_NO_BOOT_LABEL;

		boot_size = zpool_get_prop_int(zhp, ZPOOL_PROP_BOOTSIZE, NULL);
		if (zpool_label_disk(g_zfshdl, zhp, rawpath,
		    boot_type, boot_size, NULL) != 0) {
			(void) zpool_vdev_online(zhp, fullpath,
			    ZFS_ONLINE_FORCEFAULT, &newstate);
			return;
		}
	}

	/*
	 * Cosntruct the root vdev to pass to zpool_vdev_attach().  While adding
	 * the entire vdev structure is harmless, we construct a reduced set of
	 * path/physpath/wholedisk to keep it simple.
	 */
	if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
		return;

	if (nvlist_alloc(&newvd, NV_UNIQUE_NAME, 0) != 0) {
		nvlist_free(nvroot);
		return;
	}

	if (nvlist_add_string(newvd, ZPOOL_CONFIG_TYPE, VDEV_TYPE_DISK) != 0 ||
	    nvlist_add_string(newvd, ZPOOL_CONFIG_PATH, path) != 0 ||
	    (physpath != NULL && nvlist_add_string(newvd,
	    ZPOOL_CONFIG_PHYS_PATH, physpath) != 0) ||
	    nvlist_add_uint64(newvd, ZPOOL_CONFIG_WHOLE_DISK, wholedisk) != 0 ||
	    nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) != 0 ||
	    nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &newvd,
	    1) != 0) {
		nvlist_free(newvd);
		nvlist_free(nvroot);
		return;
	}

	nvlist_free(newvd);

	(void) zpool_vdev_attach(zhp, fullpath, path, nvroot, B_TRUE);

	nvlist_free(nvroot);

}

/*
 * Utility functions to find a vdev matching given criteria.
 */
typedef struct dev_data {
	const char		*dd_compare;
	const char		*dd_prop;
	zfs_process_func_t	dd_func;
	boolean_t		dd_found;
	boolean_t		dd_isdisk;
	uint64_t		dd_pool_guid;
	uint64_t		dd_vdev_guid;
} dev_data_t;

static void
zfs_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *data)
{
	dev_data_t *dp = data;
	char *path;
	uint_t c, children;
	nvlist_t **child;
	size_t len;
	uint64_t guid;

	/*
	 * First iterate over any children.
	 */
	if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN,
	    &child, &children) == 0) {
		for (c = 0; c < children; c++)
			zfs_iter_vdev(zhp, child[c], data);
		return;
	}

	if (dp->dd_vdev_guid != 0) {
		if (nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_GUID,
		    &guid) != 0 || guid != dp->dd_vdev_guid)
			return;
	} else if (dp->dd_compare != NULL) {
		len = strlen(dp->dd_compare);

		if (nvlist_lookup_string(nvl, dp->dd_prop, &path) != 0 ||
		    strncmp(dp->dd_compare, path, len) != 0)
			return;

		/*
		 * Normally, we want to have an exact match for the comparison
		 * string.  However, we allow substring matches in the following
		 * cases:
		 *
		 * 	<path>:		This is a devpath, and the target is one
		 * 			of its children.
		 *
		 * 	<path/>		This is a devid for a whole disk, and