zed: mark disks as REMOVED when they are removed

ZED does not take any action for disk removal events if there is no
spare VDEV available. Added zpool_vdev_remove_wanted() in libzfs
and vdev_remove_wanted() in vdev.c to remove the VDEV through ZED
on removal event.  This means that if you are running zed and
remove a disk, it will be properly marked as REMOVED.

Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Ameer Hamza <ahamza@ixsystems.com>
Closes #13797

cmd/zed/agents/zfs_agents.c

@@ -80,6 +80,7 @@ zfs_agent_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *arg)
 	char *path = NULL;
 	uint_t c, children;
 	nvlist_t **child;
+	uint64_t vdev_guid;
 
 	/*
 	 * First iterate over any children.
@@ -100,7 +101,7 @@ zfs_agent_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *arg)
 	    &child, &children) == 0) {
 		for (c = 0; c < children; c++) {
 			if (zfs_agent_iter_vdev(zhp, child[c], gsp)) {
-				gsp->gs_vdev_type = DEVICE_TYPE_L2ARC;
+				gsp->gs_vdev_type = DEVICE_TYPE_SPARE;
 				return (B_TRUE);
 			}
 		}
@@ -109,7 +110,7 @@ zfs_agent_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *arg)
 	    &child, &children) == 0) {
 		for (c = 0; c < children; c++) {
 			if (zfs_agent_iter_vdev(zhp, child[c], gsp)) {
-				gsp->gs_vdev_type = DEVICE_TYPE_SPARE;
+				gsp->gs_vdev_type = DEVICE_TYPE_L2ARC;
 				return (B_TRUE);
 			}
 		}
@@ -126,6 +127,21 @@ zfs_agent_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *arg)
 		    &gsp->gs_vdev_expandtime);
 		return (B_TRUE);
 	}
+	/*
+	 * Otherwise, on a vdev guid match, grab the devid and expansion
+	 * time. The devid might be missing on removal since its not part
+	 * of blkid cache and L2ARC VDEV does not contain pool guid in its
+	 * blkid, so this is a special case for L2ARC VDEV.
+	 */
+	else if (gsp->gs_vdev_guid != 0 && gsp->gs_devid == NULL &&
+	    nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_GUID, &vdev_guid) == 0 &&
+	    gsp->gs_vdev_guid == vdev_guid) {
+		(void) nvlist_lookup_string(nvl, ZPOOL_CONFIG_DEVID,
+		    &gsp->gs_devid);
+		(void) nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_EXPANSION_TIME,
+		    &gsp->gs_vdev_expandtime);
+		return (B_TRUE);
+	}
 
 	return (B_FALSE);
 }
@@ -148,7 +164,7 @@ zfs_agent_iter_pool(zpool_handle_t *zhp, void *arg)
 	/*
 	 * if a match was found then grab the pool guid
 	 */
-	if (gsp->gs_vdev_guid) {
+	if (gsp->gs_vdev_guid && gsp->gs_devid) {
 		(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 		    &gsp->gs_pool_guid);
 	}
@@ -195,11 +211,13 @@ zfs_agent_post_event(const char *class, const char *subclass, nvlist_t *nvl)
 		uint64_t pool_guid = 0, vdev_guid = 0;
 		guid_search_t search = { 0 };
 		device_type_t devtype = DEVICE_TYPE_PRIMARY;
+		char *devid = NULL;
 
 		class = "resource.fs.zfs.removed";
 		subclass = "";
 
 		(void) nvlist_add_string(payload, FM_CLASS, class);
+		(void) nvlist_lookup_string(nvl, DEV_IDENTIFIER, &devid);
 		(void) nvlist_lookup_uint64(nvl, ZFS_EV_POOL_GUID, &pool_guid);
 		(void) nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID, &vdev_guid);
 
@@ -209,21 +227,25 @@ zfs_agent_post_event(const char *class, const char *subclass, nvlist_t *nvl)
 		(void) nvlist_add_int64_array(payload, FM_EREPORT_TIME, tod, 2);
 
 		/*
+		 * If devid is missing but vdev_guid is available, find devid
+		 * and pool_guid from vdev_guid.
 		 * For multipath, spare and l2arc devices ZFS_EV_VDEV_GUID or
 		 * ZFS_EV_POOL_GUID may be missing so find them.
 		 */
-		if (pool_guid == 0 || vdev_guid == 0) {
+		if (devid == NULL || pool_guid == 0 || vdev_guid == 0) {
-			if ((nvlist_lookup_string(nvl, DEV_IDENTIFIER,
+			if (devid == NULL)
-			    &search.gs_devid) == 0) &&
+				search.gs_vdev_guid = vdev_guid;
-			    (zpool_iter(g_zfs_hdl, zfs_agent_iter_pool, &search)
+			else
-			    == 1)) {
+				search.gs_devid = devid;
+			zpool_iter(g_zfs_hdl, zfs_agent_iter_pool, &search);
+			if (devid == NULL)
+				devid = search.gs_devid;
 			if (pool_guid == 0)
 				pool_guid = search.gs_pool_guid;
 			if (vdev_guid == 0)
 				vdev_guid = search.gs_vdev_guid;
 			devtype = search.gs_vdev_type;
 		}
-		}
 
 		/*
 		 * We want to avoid reporting "remove" events coming from
@@ -235,7 +257,9 @@ zfs_agent_post_event(const char *class, const char *subclass, nvlist_t *nvl)
 		    search.gs_vdev_expandtime + 10 > tv.tv_sec) {
 			zed_log_msg(LOG_INFO, "agent post event: ignoring '%s' "
 			    "for recently expanded device '%s'", EC_DEV_REMOVE,
-			    search.gs_devid);
+			    devid);
+			fnvlist_free(payload);
+			free(event);
 			goto out;
 		}
 

cmd/zed/agents/zfs_retire.c

@@ -323,6 +323,9 @@ zfs_retire_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl,
 	boolean_t is_disk;
 	vdev_aux_t aux;
 	uint64_t state = 0;
+	int l2arc;
+	vdev_stat_t *vs;
+	unsigned int c;
 
 	fmd_hdl_debug(hdl, "zfs_retire_recv: '%s'", class);
 
@@ -352,13 +355,32 @@ zfs_retire_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl,
 
 		devname = zpool_vdev_name(NULL, zhp, vdev, B_FALSE);
 
-		/* Can't replace l2arc with a spare: offline the device */
+		nvlist_lookup_uint64_array(vdev, ZPOOL_CONFIG_VDEV_STATS,
-		if (nvlist_lookup_string(nvl, FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE,
+		    (uint64_t **)&vs, &c);
-		    &devtype) == 0 && strcmp(devtype, VDEV_TYPE_L2CACHE) == 0) {
+
-			fmd_hdl_debug(hdl, "zpool_vdev_offline '%s'", devname);
+		/*
-			zpool_vdev_offline(zhp, devname, B_TRUE);
+		 * If state removed is requested for already removed vdev,
-		} else if (!fmd_prop_get_int32(hdl, "spare_on_remove") ||
+		 * its a loopback event from spa_async_remove(). Just
-		    replace_with_spare(hdl, zhp, vdev) == B_FALSE) {
+		 * ignore it.
+		 */
+		if (vs->vs_state == VDEV_STATE_REMOVED &&
+		    state == VDEV_STATE_REMOVED)
+			return;
+
+		l2arc = (nvlist_lookup_string(nvl,
+		    FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, &devtype) == 0 &&
+		    strcmp(devtype, VDEV_TYPE_L2CACHE) == 0);
+
+		/* Remove the vdev since device is unplugged */
+		if (l2arc || (strcmp(class, "resource.fs.zfs.removed") == 0)) {
+			int status = zpool_vdev_remove_wanted(zhp, devname);
+			fmd_hdl_debug(hdl, "zpool_vdev_remove_wanted '%s'"
+			    ", ret:%d", devname, status);
+		}
+
+		/* Replace the vdev with a spare if its not a l2arc */
+		if (!l2arc && (!fmd_prop_get_int32(hdl, "spare_on_remove") ||
+		    replace_with_spare(hdl, zhp, vdev) == B_FALSE)) {
 			/* Could not handle with spare */
 			fmd_hdl_debug(hdl, "no spare for '%s'", devname);
 		}

cmd/ztest.c

@@ -1166,7 +1166,7 @@ ztest_kill(ztest_shared_t *zs)
 	 * See comment above spa_write_cachefile().
 	 */
 	mutex_enter(&spa_namespace_lock);
-	spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE);
+	spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE, B_FALSE);
 	mutex_exit(&spa_namespace_lock);
 
 	(void) raise(SIGKILL);

config/kernel-blkdev.m4

@@ -103,6 +103,57 @@ AC_DEFUN([ZFS_AC_KERNEL_BLKDEV_CHECK_DISK_CHANGE], [
 	])
 ])
 
+dnl #
+dnl # bdev_kobj() is introduced from 5.12
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_SRC_BLKDEV_BDEV_KOBJ], [
+	ZFS_LINUX_TEST_SRC([bdev_kobj], [
+		#include <linux/fs.h>
+		#include <linux/blkdev.h>
+		#include <linux/kobject.h>
+	], [
+		struct block_device *bdev = NULL;
+		struct kobject *disk_kobj;
+		disk_kobj = bdev_kobj(bdev);
+	])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_BLKDEV_BDEV_KOBJ], [
+	AC_MSG_CHECKING([whether bdev_kobj() exists])
+	ZFS_LINUX_TEST_RESULT([bdev_kobj], [
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_BDEV_KOBJ, 1,
+		    [bdev_kobj() exists])
+	], [
+		AC_MSG_RESULT(no)
+	])
+])
+
+dnl #
+dnl # part_to_dev() was removed in 5.12
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_SRC_BLKDEV_PART_TO_DEV], [
+	ZFS_LINUX_TEST_SRC([part_to_dev], [
+		#include <linux/fs.h>
+		#include <linux/blkdev.h>
+	], [
+		struct hd_struct *p = NULL;
+		struct device *pdev;
+		pdev = part_to_dev(p);
+	])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_BLKDEV_PART_TO_DEV], [
+	AC_MSG_CHECKING([whether part_to_dev() exists])
+	ZFS_LINUX_TEST_RESULT([part_to_dev], [
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_PART_TO_DEV, 1,
+		    [part_to_dev() exists])
+	], [
+		AC_MSG_RESULT(no)
+	])
+])
+
 dnl #
 dnl # 5.10 API, check_disk_change() is removed, in favor of
 dnl # bdev_check_media_change(), which doesn't force revalidation
@@ -405,6 +456,8 @@ AC_DEFUN([ZFS_AC_KERNEL_SRC_BLKDEV], [
 	ZFS_AC_KERNEL_SRC_BLKDEV_BDEV_WHOLE
 	ZFS_AC_KERNEL_SRC_BLKDEV_BDEVNAME
 	ZFS_AC_KERNEL_SRC_BLKDEV_ISSUE_SECURE_ERASE
+	ZFS_AC_KERNEL_SRC_BLKDEV_BDEV_KOBJ
+	ZFS_AC_KERNEL_SRC_BLKDEV_PART_TO_DEV
 ])
 
 AC_DEFUN([ZFS_AC_KERNEL_BLKDEV], [
@@ -421,4 +474,6 @@ AC_DEFUN([ZFS_AC_KERNEL_BLKDEV], [
 	ZFS_AC_KERNEL_BLKDEV_BDEVNAME
 	ZFS_AC_KERNEL_BLKDEV_GET_ERESTARTSYS
 	ZFS_AC_KERNEL_BLKDEV_ISSUE_SECURE_ERASE
+	ZFS_AC_KERNEL_BLKDEV_BDEV_KOBJ
+	ZFS_AC_KERNEL_BLKDEV_PART_TO_DEV
 ])

include/libzfs.h

@@ -310,6 +310,7 @@ _LIBZFS_H int zpool_vdev_indirect_size(zpool_handle_t *, const char *,
     uint64_t *);
 _LIBZFS_H int zpool_vdev_split(zpool_handle_t *, char *, nvlist_t **,
     nvlist_t *, splitflags_t);
+_LIBZFS_H int zpool_vdev_remove_wanted(zpool_handle_t *, const char *);
 
 _LIBZFS_H int zpool_vdev_fault(zpool_handle_t *, uint64_t, vdev_aux_t);
 _LIBZFS_H int zpool_vdev_degrade(zpool_handle_t *, uint64_t, vdev_aux_t);

include/os/linux/Makefile.am

@@ -71,6 +71,7 @@ kernel_spl_sys_HEADERS = \
 	%D%/spl/sys/kmem_cache.h \
 	%D%/spl/sys/kstat.h \
 	%D%/spl/sys/list.h \
+	%D%/spl/sys/misc.h \
 	%D%/spl/sys/mod_os.h \
 	%D%/spl/sys/mutex.h \
 	%D%/spl/sys/param.h \

include/os/linux/kernel/linux/blkdev_compat.h

@@ -261,6 +261,32 @@ bio_set_bi_error(struct bio *bio, int error)
 #define	BIO_END_IO(bio, error)		bio_endio(bio, error);
 #endif /* HAVE_1ARG_BIO_END_IO_T */
 
+/*
+ * 5.15 MACRO,
+ *   GD_DEAD
+ *
+ * 2.6.36 - 5.14 MACRO,
+ *   GENHD_FL_UP
+ *
+ * Check the disk status and return B_TRUE if alive
+ * otherwise B_FALSE
+ */
+static inline boolean_t
+zfs_check_disk_status(struct block_device *bdev)
+{
+#if defined(GENHD_FL_UP)
+	return (!!(bdev->bd_disk->flags & GENHD_FL_UP));
+#elif defined(GD_DEAD)
+	return (!test_bit(GD_DEAD, &bdev->bd_disk->state));
+#else
+/*
+ * This is encountered if neither GENHD_FL_UP nor GD_DEAD is available in
+ * the kernel - likely due to an MACRO change that needs to be chased down.
+ */
+#error "Unsupported kernel: no usable disk status check"
+#endif
+}
+
 /*
  * 4.1 API,
  * 3.10.0 CentOS 7.x API,

include/os/linux/spl/sys/misc.h

@@ -0,0 +1,29 @@
+/*
+ * 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 https://opensource.org/licenses/CDDL-1.0.
+ * 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
+ */
+
+#ifndef _OS_LINUX_SPL_MISC_H
+#define	_OS_LINUX_SPL_MISC_H
+
+#include <linux/kobject.h>
+
+extern void spl_signal_kobj_evt(struct block_device *bdev);
+
+#endif

include/sys/spa.h

@@ -838,7 +838,7 @@ extern kmutex_t spa_namespace_lock;
 #define	SPA_CONFIG_UPDATE_POOL	0
 #define	SPA_CONFIG_UPDATE_VDEVS	1
 
-extern void spa_write_cachefile(spa_t *, boolean_t, boolean_t);
+extern void spa_write_cachefile(spa_t *, boolean_t, boolean_t, boolean_t);
 extern void spa_config_load(void);
 extern nvlist_t *spa_all_configs(uint64_t *);
 extern void spa_config_set(spa_t *spa, nvlist_t *config);

include/sys/vdev.h

@@ -148,6 +148,7 @@ extern int vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux);
 extern int vdev_online(spa_t *spa, uint64_t guid, uint64_t flags,
     vdev_state_t *);
 extern int vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags);
+extern int vdev_remove_wanted(spa_t *spa, uint64_t guid);
 extern void vdev_clear(spa_t *spa, vdev_t *vd);
 
 extern boolean_t vdev_is_dead(vdev_t *vd);
@@ -190,6 +191,8 @@ typedef enum vdev_config_flag {
 	VDEV_CONFIG_MISSING = 1 << 4
 } vdev_config_flag_t;
 
+extern void vdev_post_kobj_evt(vdev_t *vd);
+extern void vdev_clear_kobj_evt(vdev_t *vd);
 extern void vdev_top_config_generate(spa_t *spa, nvlist_t *config);
 extern nvlist_t *vdev_config_generate(spa_t *spa, vdev_t *vd,
     boolean_t getstats, vdev_config_flag_t flags);

include/sys/vdev_impl.h

@@ -69,6 +69,7 @@ extern uint_t zfs_vdev_async_write_max_active;
  * Virtual device operations
  */
 typedef int	vdev_init_func_t(spa_t *spa, nvlist_t *nv, void **tsd);
+typedef void	vdev_kobj_post_evt_func_t(vdev_t *vd);
 typedef void	vdev_fini_func_t(vdev_t *vd);
 typedef int	vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size,
     uint64_t *ashift, uint64_t *pshift);
@@ -123,6 +124,7 @@ typedef const struct vdev_ops {
 	vdev_config_generate_func_t	*vdev_op_config_generate;
 	vdev_nparity_func_t		*vdev_op_nparity;
 	vdev_ndisks_func_t		*vdev_op_ndisks;
+	vdev_kobj_post_evt_func_t	*vdev_op_kobj_evt_post;
 	char				vdev_op_type[16];
 	boolean_t			vdev_op_leaf;
 } vdev_ops_t;
@@ -436,6 +438,7 @@ struct vdev {
 	boolean_t	vdev_isl2cache;	/* was a l2cache device		*/
 	boolean_t	vdev_copy_uberblocks;  /* post expand copy uberblocks */
 	boolean_t	vdev_resilver_deferred;  /* resilver deferred */
+	boolean_t	vdev_kobj_flag; /* kobj event record */
 	vdev_queue_t	vdev_queue;	/* I/O deadline schedule queue	*/
 	vdev_cache_t	vdev_cache;	/* physical block cache		*/
 	spa_aux_vdev_t	*vdev_aux;	/* for l2cache and spares vdevs	*/

include/sys/zfs_context.h

@@ -50,6 +50,7 @@ extern "C" {
 #include <sys/kmem.h>
 #include <sys/kmem_cache.h>
 #include <sys/vmem.h>
+#include <sys/misc.h>
 #include <sys/taskq.h>
 #include <sys/param.h>
 #include <sys/disp.h>

lib/libzfs/libzfs.abi

@@ -550,6 +550,7 @@
     <elf-symbol name='zpool_vdev_path_to_guid' type='func-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
     <elf-symbol name='zpool_vdev_remove' type='func-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
     <elf-symbol name='zpool_vdev_remove_cancel' type='func-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
+    <elf-symbol name='zpool_vdev_remove_wanted' type='func-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
     <elf-symbol name='zpool_vdev_split' type='func-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
     <elf-symbol name='zpool_wait' type='func-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
     <elf-symbol name='zpool_wait_status' type='func-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
@@ -3505,6 +3506,11 @@
       <parameter type-id='c19b74c3' name='istmp'/>
       <return type-id='95e97e5e'/>
     </function-decl>
+    <function-decl name='zpool_vdev_remove_wanted' mangled-name='zpool_vdev_remove_wanted' visibility='default' binding='global' size-in-bits='64' elf-symbol-id='zpool_vdev_remove_wanted'>
+      <parameter type-id='4c81de99' name='zhp'/>
+      <parameter type-id='80f4b756' name='path'/>
+      <return type-id='95e97e5e'/>
+    </function-decl>
     <function-decl name='zpool_vdev_fault' mangled-name='zpool_vdev_fault' visibility='default' binding='global' size-in-bits='64' elf-symbol-id='zpool_vdev_fault'>
       <parameter type-id='4c81de99' name='zhp'/>
       <parameter type-id='9c313c2d' name='guid'/>

lib/libzfs/libzfs_pool.c

@@ -3073,6 +3073,43 @@ zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
 	}
 }
 
+/*
+ * Remove the specified vdev asynchronously from the configuration, so
+ * that it may come ONLINE if reinserted. This is called from zed on
+ * Udev remove event.
+ * Note: We also have a similar function zpool_vdev_remove() that
+ * removes the vdev from the pool.
+ */
+int
+zpool_vdev_remove_wanted(zpool_handle_t *zhp, const char *path)
+{
+	zfs_cmd_t zc = {"\0"};
+	char errbuf[ERRBUFLEN];
+	nvlist_t *tgt;
+	boolean_t avail_spare, l2cache;
+	libzfs_handle_t *hdl = zhp->zpool_hdl;
+
+	(void) snprintf(errbuf, sizeof (errbuf),
+	    dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
+
+	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
+	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
+	    NULL)) == NULL)
+		return (zfs_error(hdl, EZFS_NODEVICE, errbuf));
+
+	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
+
+	if (avail_spare)
+		return (zfs_error(hdl, EZFS_ISSPARE, errbuf));
+
+	zc.zc_cookie = VDEV_STATE_REMOVED;
+
+	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
+		return (0);
+
+	return (zpool_standard_error(hdl, errno, errbuf));
+}
+
 /*
  * Mark the given vdev faulted.
  */

module/os/linux/spl/spl-generic.c

@@ -47,6 +47,7 @@
 #include <linux/mod_compat.h>
 #include <sys/cred.h>
 #include <sys/vnode.h>
+#include <sys/misc.h>
 
 unsigned long spl_hostid = 0;
 EXPORT_SYMBOL(spl_hostid);
@@ -517,6 +518,38 @@ ddi_copyin(const void *from, void *to, size_t len, int flags)
 }
 EXPORT_SYMBOL(ddi_copyin);
 
+/*
+ * Post a uevent to userspace whenever a new vdev adds to the pool. It is
+ * necessary to sync blkid information with udev, which zed daemon uses
+ * during device hotplug to identify the vdev.
+ */
+void
+spl_signal_kobj_evt(struct block_device *bdev)
+{
+#if defined(HAVE_BDEV_KOBJ) || defined(HAVE_PART_TO_DEV)
+#ifdef HAVE_BDEV_KOBJ
+	struct kobject *disk_kobj = bdev_kobj(bdev);
+#else
+	struct kobject *disk_kobj = &part_to_dev(bdev->bd_part)->kobj;
+#endif
+	if (disk_kobj) {
+		int ret = kobject_uevent(disk_kobj, KOBJ_CHANGE);
+		if (ret) {
+			pr_warn("ZFS: Sending event '%d' to kobject: '%s'"
+			    " (%p): failed(ret:%d)\n", KOBJ_CHANGE,
+			    kobject_name(disk_kobj), disk_kobj, ret);
+		}
+	}
+#else
+/*
+ * This is encountered if neither bdev_kobj() nor part_to_dev() is available
+ * in the kernel - likely due to an API change that needs to be chased down.
+ */
+#error "Unsupported kernel: unable to get struct kobj from bdev"
+#endif
+}
+EXPORT_SYMBOL(spl_signal_kobj_evt);
+
 int
 ddi_copyout(const void *from, void *to, size_t len, int flags)
 {

module/os/linux/zfs/vdev_disk.c

@@ -179,6 +179,18 @@ vdev_disk_error(zio_t *zio)
 	    zio->io_flags);
 }
 
+static void
+vdev_disk_kobj_evt_post(vdev_t *v)
+{
+	vdev_disk_t *vd = v->vdev_tsd;
+	if (vd && vd->vd_bdev) {
+		spl_signal_kobj_evt(vd->vd_bdev);
+	} else {
+		vdev_dbgmsg(v, "vdev_disk_t is NULL for VDEV:%s\n",
+		    v->vdev_path);
+	}
+}
+
 static int
 vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
     uint64_t *logical_ashift, uint64_t *physical_ashift)
@@ -290,6 +302,13 @@ vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
 		bdev = blkdev_get_by_path(v->vdev_path, mode | FMODE_EXCL,
 		    zfs_vdev_holder);
 		if (unlikely(PTR_ERR(bdev) == -ENOENT)) {
+			/*
+			 * There is no point of waiting since device is removed
+			 * explicitly
+			 */
+			if (v->vdev_removed)
+				break;
+
 			schedule_timeout(MSEC_TO_TICK(10));
 		} else if (unlikely(PTR_ERR(bdev) == -ERESTARTSYS)) {
 			timeout = MSEC2NSEC(zfs_vdev_open_timeout_ms * 10);
@@ -901,7 +920,7 @@ vdev_disk_io_done(zio_t *zio)
 		vdev_t *v = zio->io_vd;
 		vdev_disk_t *vd = v->vdev_tsd;
 
-		if (zfs_check_media_change(vd->vd_bdev)) {
+		if (!zfs_check_disk_status(vd->vd_bdev)) {
 			invalidate_bdev(vd->vd_bdev);
 			v->vdev_remove_wanted = B_TRUE;
 			spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
@@ -957,7 +976,8 @@ vdev_ops_t vdev_disk_ops = {
 	.vdev_op_nparity = NULL,
 	.vdev_op_ndisks = NULL,
 	.vdev_op_type = VDEV_TYPE_DISK,		/* name of this vdev type */
-	.vdev_op_leaf = B_TRUE			/* leaf vdev */
+	.vdev_op_leaf = B_TRUE,			/* leaf vdev */
+	.vdev_op_kobj_evt_post = vdev_disk_kobj_evt_post
 };
 
 /*

module/zfs/spa.c

@@ -909,7 +909,16 @@ spa_change_guid(spa_t *spa)
 	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
 
 	if (error == 0) {
-		spa_write_cachefile(spa, B_FALSE, B_TRUE);
+		/*
+		 * Clear the kobj flag from all the vdevs to allow
+		 * vdev_cache_process_kobj_evt() to post events to all the
+		 * vdevs since GUID is updated.
+		 */
+		vdev_clear_kobj_evt(spa->spa_root_vdev);
+		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
+			vdev_clear_kobj_evt(spa->spa_l2cache.sav_vdevs[i]);
+
+		spa_write_cachefile(spa, B_FALSE, B_TRUE, B_TRUE);
 		spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
 	}
 
@@ -5220,7 +5229,7 @@ spa_open_common(const char *pool, spa_t **spapp, const void *tag,
 			 */
 			spa_unload(spa);
 			spa_deactivate(spa);
-			spa_write_cachefile(spa, B_TRUE, B_TRUE);
+			spa_write_cachefile(spa, B_TRUE, B_TRUE, B_FALSE);
 			spa_remove(spa);
 			if (locked)
 				mutex_exit(&spa_namespace_lock);
@@ -6044,7 +6053,7 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 
 	spa_spawn_aux_threads(spa);
 
-	spa_write_cachefile(spa, B_FALSE, B_TRUE);
+	spa_write_cachefile(spa, B_FALSE, B_TRUE, B_TRUE);
 
 	/*
 	 * Don't count references from objsets that are already closed
@@ -6107,7 +6116,7 @@ spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
 		if (props != NULL)
 			spa_configfile_set(spa, props, B_FALSE);
 
-		spa_write_cachefile(spa, B_FALSE, B_TRUE);
+		spa_write_cachefile(spa, B_FALSE, B_TRUE, B_FALSE);
 		spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
 		zfs_dbgmsg("spa_import: verbatim import of %s", pool);
 		mutex_exit(&spa_namespace_lock);
@@ -6503,7 +6512,7 @@ export_spa:
 
 	if (new_state != POOL_STATE_UNINITIALIZED) {
 		if (!hardforce)
-			spa_write_cachefile(spa, B_TRUE, B_TRUE);
+			spa_write_cachefile(spa, B_TRUE, B_TRUE, B_FALSE);
 		spa_remove(spa);
 	} else {
 		/*

module/zfs/spa_config.c

@@ -240,7 +240,8 @@ spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
  * would be required.
  */
 void
-spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent)
+spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent,
+    boolean_t postblkidevent)
 {
 	spa_config_dirent_t *dp, *tdp;
 	nvlist_t *nvl;
@@ -346,6 +347,16 @@ spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent)
 
 	if (postsysevent)
 		spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
+
+	/*
+	 * Post udev event to sync blkid information if the pool is created
+	 * or a new vdev is added to the pool.
+	 */
+	if ((target->spa_root_vdev) && postblkidevent) {
+		vdev_post_kobj_evt(target->spa_root_vdev);
+		for (int i = 0; i < target->spa_l2cache.sav_count; i++)
+			vdev_post_kobj_evt(target->spa_l2cache.sav_vdevs[i]);
+	}
 }
 
 /*
@@ -600,6 +611,7 @@ spa_config_update(spa_t *spa, int what)
 	 */
 	if (!spa->spa_is_root) {
 		spa_write_cachefile(spa, B_FALSE,
+		    what != SPA_CONFIG_UPDATE_POOL,
 		    what != SPA_CONFIG_UPDATE_POOL);
 	}
 

module/zfs/spa_misc.c

@@ -1290,7 +1290,7 @@ spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error,
 	 * If the config changed, update the config cache.
 	 */
 	if (config_changed)
-		spa_write_cachefile(spa, B_FALSE, B_TRUE);
+		spa_write_cachefile(spa, B_FALSE, B_TRUE, B_FALSE);
 }
 
 /*
@@ -1385,7 +1385,7 @@ spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
 	 */
 	if (config_changed) {
 		mutex_enter(&spa_namespace_lock);
-		spa_write_cachefile(spa, B_FALSE, B_TRUE);
+		spa_write_cachefile(spa, B_FALSE, B_TRUE, B_FALSE);
 		mutex_exit(&spa_namespace_lock);
 	}
 

module/zfs/vdev.c

@@ -1948,6 +1948,14 @@ vdev_open(vdev_t *vd)
 
 	error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
 	    &logical_ashift, &physical_ashift);
+
+	/* Keep the device in removed state if unplugged */
+	if (error == ENOENT && vd->vdev_removed) {
+		vdev_set_state(vd, B_TRUE, VDEV_STATE_REMOVED,
+		    VDEV_AUX_NONE);
+		return (error);
+	}
+
 	/*
 	 * Physical volume size should never be larger than its max size, unless
 	 * the disk has shrunk while we were reading it or the device is buggy
@@ -3166,6 +3174,34 @@ vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg,
 	mutex_exit(&vd->vdev_dtl_lock);
 }
 
+/*
+ * Iterate over all the vdevs except spare, and post kobj events
+ */
+void
+vdev_post_kobj_evt(vdev_t *vd)
+{
+	if (vd->vdev_ops->vdev_op_kobj_evt_post &&
+	    vd->vdev_kobj_flag == B_FALSE) {
+		vd->vdev_kobj_flag = B_TRUE;
+		vd->vdev_ops->vdev_op_kobj_evt_post(vd);
+	}
+
+	for (int c = 0; c < vd->vdev_children; c++)
+		vdev_post_kobj_evt(vd->vdev_child[c]);
+}
+
+/*
+ * Iterate over all the vdevs except spare, and clear kobj events
+ */
+void
+vdev_clear_kobj_evt(vdev_t *vd)
+{
+	vd->vdev_kobj_flag = B_FALSE;
+
+	for (int c = 0; c < vd->vdev_children; c++)
+		vdev_clear_kobj_evt(vd->vdev_child[c]);
+}
+
 int
 vdev_dtl_load(vdev_t *vd)
 {
@@ -3947,6 +3983,29 @@ vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
 	return (spa_vdev_state_exit(spa, vd, 0));
 }
 
+int
+vdev_remove_wanted(spa_t *spa, uint64_t guid)
+{
+	vdev_t *vd;
+
+	spa_vdev_state_enter(spa, SCL_NONE);
+
+	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
+		return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV)));
+
+	/*
+	 * If the vdev is already removed, then don't do anything.
+	 */
+	if (vd->vdev_removed)
+		return (spa_vdev_state_exit(spa, NULL, 0));
+
+	vd->vdev_remove_wanted = B_TRUE;
+	spa_async_request(spa, SPA_ASYNC_REMOVE);
+
+	return (spa_vdev_state_exit(spa, vd, 0));
+}
+
+
 /*
  * Online the given vdev.
  *

module/zfs/zfs_ioctl.c

@@ -1912,6 +1912,10 @@ zfs_ioc_vdev_set_state(zfs_cmd_t *zc)
 		error = vdev_degrade(spa, zc->zc_guid, zc->zc_obj);
 		break;
 
+	case VDEV_STATE_REMOVED:
+		error = vdev_remove_wanted(spa, zc->zc_guid);
+		break;
+
 	default:
 		error = SET_ERROR(EINVAL);
 	}
@@ -2928,7 +2932,7 @@ zfs_ioc_pool_set_props(zfs_cmd_t *zc)
 		mutex_enter(&spa_namespace_lock);
 		if ((spa = spa_lookup(zc->zc_name)) != NULL) {
 			spa_configfile_set(spa, props, B_FALSE);
-			spa_write_cachefile(spa, B_FALSE, B_TRUE);
+			spa_write_cachefile(spa, B_FALSE, B_TRUE, B_FALSE);
 		}
 		mutex_exit(&spa_namespace_lock);
 		if (spa != NULL) {

module/zfs/zio.c

@@ -3923,7 +3923,7 @@ zio_vdev_io_done(zio_t *zio)
 
 	ops->vdev_op_io_done(zio);
 
-	if (unexpected_error)
+	if (unexpected_error && vd->vdev_remove_wanted == B_FALSE)
 		VERIFY(vdev_probe(vd, zio) == NULL);
 
 	return (zio);

tests/zfs-tests/include/libtest.shlib

@@ -1884,7 +1884,7 @@ function wait_hotspare_state # pool disk state timeout
 #
 # Return 0 is pool/disk matches expected state, 1 otherwise
 #
-function check_vdev_state # pool disk state{online,offline,unavail}
+function check_vdev_state # pool disk state{online,offline,unavail,removed}
 {
 	typeset pool=$1
 	typeset disk=${2#*$DEV_DSKDIR/}

tests/zfs-tests/tests/functional/fault/auto_offline_001_pos.ksh

@@ -24,29 +24,28 @@
 
 #
 # DESCRIPTION:
-# Testing Fault Management Agent ZED Logic - Physically removed device is
+# Testing Fault Management Agent ZED Logic - Physically detached device is
-# made unavail and onlined when reattached
+# made removed and onlined when reattached
 #
 # STRATEGY:
 # 1. Create a pool
 # 2. Simulate physical removal of one device
-# 3. Verify the device is unavailable
+# 3. Verify the device is removed when detached
 # 4. Reattach the device
 # 5. Verify the device is onlined
 # 6. Repeat the same tests with a spare device:
 #    zed will use the spare to handle the removed data device
 # 7. Repeat the same tests again with a faulted spare device:
-#    the removed data device should be unavailable
+#    the removed data device should be removed
 #
 # NOTE: the use of 'block_device_wait' throughout the test helps avoid race
 # conditions caused by mixing creation/removal events from partitioning the
 # disk (zpool create) and events from physically removing it (remove_disk).
 #
-# NOTE: the test relies on 'zpool sync' to prompt the kmods to transition a
+# NOTE: the test relies on ZED to transit state to removed on device removed
-# vdev to the unavailable state.  The ZED does receive a removal notification
+# event.  The ZED does receive a removal notification but only relies on it to
-# but only relies on it to activate a hot spare.  Additional work is planned
+# activate a hot spare.  Additional work is planned to extend an existing ioctl
-# to extend an existing ioctl interface to allow the ZED to transition the
+# interface to allow the ZED to transition the vdev in to a removed state.
-# vdev in to a removed state.
 #
 verify_runnable "both"
 
@@ -103,8 +102,8 @@ do
 	log_must mkfile 1m $mntpnt/file
 	sync_pool $TESTPOOL
 
-	# 3. Verify the device is unavailable.
+	# 3. Verify the device is removed.
-	log_must wait_vdev_state $TESTPOOL $removedev "UNAVAIL"
+	log_must wait_vdev_state $TESTPOOL $removedev "REMOVED"
 
 	# 4. Reattach the device
 	insert_disk $removedev
@@ -136,7 +135,7 @@ do
 
 	# 3. Verify the device is handled by the spare.
 	log_must wait_hotspare_state $TESTPOOL $sparedev "INUSE"
-	log_must wait_vdev_state $TESTPOOL $removedev "UNAVAIL"
+	log_must wait_vdev_state $TESTPOOL $removedev "REMOVED"
 
 	# 4. Reattach the device
 	insert_disk $removedev
@@ -170,8 +169,8 @@ do
 	log_must mkfile 1m $mntpnt/file
 	sync_pool $TESTPOOL
 
-	# 4. Verify the device is unavailable
+	# 4. Verify the device is removed
-	log_must wait_vdev_state $TESTPOOL $removedev "UNAVAIL"
+	log_must wait_vdev_state $TESTPOOL $removedev "REMOVED"
 
 	# 5. Reattach the device
 	insert_disk $removedev