zfsd: add support for hotplugging spares

If you remove an unused spare and then reinsert it, zfsd will now online
it in all pools.

Signed-off-by: Ameer Hamza <ahamza@ixsystems.com>

cmd/zfsd/case_file.cc

@@ -116,6 +116,26 @@ CaseFile::Find(Guid poolGUID, Guid vdevGUID)
 	return (NULL);
 }
 
+void
+CaseFile::Find(Guid poolGUID, Guid vdevGUID, CaseFileList &cases)
+{
+	for (CaseFileList::iterator curCase = s_activeCases.begin();
+	    curCase != s_activeCases.end(); curCase++) {
+		if (((*curCase)->PoolGUID() != poolGUID &&
+		    Guid::InvalidGuid() != poolGUID) ||
+		    (*curCase)->VdevGUID() != vdevGUID)
+			continue;
+
+		/*
+		 * We can have multiple cases for spare vdevs
+		 */
+		cases.push_back(*curCase);
+		if (!(*curCase)->IsSpare()) {
+			return;
+		}
+	}
+}
+
 CaseFile *
 CaseFile::Find(const string &physPath)
 {
@@ -217,6 +237,12 @@ CaseFile::PurgeAll()
 
 }
 
+int
+CaseFile::IsSpare()
+{
+	return (m_is_spare);
+}
+
 //- CaseFile Public Methods ----------------------------------------------------
 bool
 CaseFile::RefreshVdevState()
@@ -240,6 +266,7 @@ CaseFile::ReEvaluate(const string &devPath, const string &physPath, Vdev *vdev)
 {
 	ZpoolList zpl(ZpoolList::ZpoolByGUID, &m_poolGUID);
 	zpool_handle_t *pool(zpl.empty() ? NULL : zpl.front());
+	int flags = ZFS_ONLINE_CHECKREMOVE | ZFS_ONLINE_UNSPARE;
 
 	if (pool == NULL || !RefreshVdevState()) {
 		/*
@@ -280,9 +307,10 @@ CaseFile::ReEvaluate(const string &devPath, const string &physPath, Vdev *vdev)
 	   || vdev->PoolGUID() == Guid::InvalidGuid())
 	 && vdev->GUID() == m_vdevGUID) {
 
+		if (IsSpare())
+			flags |= ZFS_ONLINE_SPARE;
 		zpool_vdev_online(pool, vdev->GUIDString().c_str(),
-				  ZFS_ONLINE_CHECKREMOVE | ZFS_ONLINE_UNSPARE,
+		    flags, &m_vdevState);
-				  &m_vdevState);
 		syslog(LOG_INFO, "Onlined vdev(%s/%s:%s).  State now %s.\n",
 		       zpool_get_name(pool), vdev->GUIDString().c_str(),
 		       devPath.c_str(),
@@ -783,7 +811,8 @@ CaseFile::CaseFile(const Vdev &vdev)
  : m_poolGUID(vdev.PoolGUID()),
    m_vdevGUID(vdev.GUID()),
    m_vdevState(vdev.State()),
-   m_vdevPhysPath(vdev.PhysicalPath())
+   m_vdevPhysPath(vdev.PhysicalPath()),
+   m_is_spare(vdev.IsSpare())
 {
 	stringstream guidString;
 

cmd/zfsd/case_file.h

@@ -98,6 +98,19 @@ public:
 	 */
 	static CaseFile *Find(DevdCtl::Guid poolGUID, DevdCtl::Guid vdevGUID);
 
+	/**
+	 * \brief Find multiple CaseFile objects by a vdev's pool/vdev
+	 *        GUID tuple (special case for spare vdevs)
+	 *
+	 * \param poolGUID  Pool GUID for the vdev of the CaseFile to find.
+	 * 		    If InvalidGuid, then only match the vdev GUID
+	 * 		    instead of both pool and vdev GUIDs.
+	 * \param vdevGUID  Vdev GUID for the vdev of the CaseFile to find.
+	 * \param caseList  List of cases associated with the vdev.
+	 */
+	static void  Find(DevdCtl::Guid poolGUID, DevdCtl::Guid vdevGUID,
+				     CaseFileList &caseList);
+
 	/**
 	 * \brief Find a CaseFile object by a vdev's current/last known
 	 *        physical path.
@@ -219,6 +232,11 @@ public:
 	 */
 	bool ShouldFault() const;
 
+	/**
+	 * \brief If this vdev is spare
+	 */
+	int IsSpare();
+
 protected:
 	enum {
 		/**
@@ -384,6 +402,7 @@ protected:
 	string		   m_poolGUIDString;
 	string		   m_vdevGUIDString;
 	string		   m_vdevPhysPath;
+	int		   m_is_spare;
 
 	/**
 	 * \brief Callout activated when a grace period

cmd/zfsd/vdev_iterator.cc

@@ -78,8 +78,10 @@ VdevIterator::Reset()
 {
 	nvlist_t  *rootVdev;
 	nvlist	  **cache_child;
+	nvlist	  **spare_child;
 	int	   result;
 	uint_t   cache_children;
+	uint_t	 spare_children;
 
 	result = nvlist_lookup_nvlist(m_poolConfig,
 				      ZPOOL_CONFIG_VDEV_TREE,
@@ -95,6 +97,13 @@ VdevIterator::Reset()
 	if (result == 0)
 		for (uint_t c = 0; c < cache_children; c++)
 			m_vdevQueue.push_back(cache_child[c]);
+	result = nvlist_lookup_nvlist_array(rootVdev,
+					    ZPOOL_CONFIG_SPARES,
+					    &spare_child,
+					    &spare_children);
+	if (result == 0)
+		for (uint_t c = 0; c < spare_children; c++)
+			m_vdevQueue.push_back(spare_child[c]);
 }
 
 nvlist_t *

cmd/zfsd/zfsd_event.cc

@@ -231,7 +231,9 @@ bool
 GeomEvent::OnlineByLabel(const string &devPath, const string& physPath,
 			      nvlist_t *devConfig)
 {
+	bool ret = false;
 	try {
+		CaseFileList case_list;
 		/*
 		 * A device with ZFS label information has been
 		 * inserted.  If it matches a device for which we
@@ -240,10 +242,12 @@ GeomEvent::OnlineByLabel(const string &devPath, const string& physPath,
 		syslog(LOG_INFO, "Interrogating VDEV label for %s\n",
 		       devPath.c_str());
 		Vdev vdev(devConfig);
-		CaseFile *caseFile(CaseFile::Find(vdev.PoolGUID(),
+		CaseFile::Find(vdev.PoolGUID(),vdev.GUID(), case_list);
-						  vdev.GUID()));
+		for (CaseFileList::iterator curr = case_list.begin();
-		if (caseFile != NULL)
+		    curr != case_list.end(); curr++) {
-			return (caseFile->ReEvaluate(devPath, physPath, &vdev));
+			ret |= (*curr)->ReEvaluate(devPath, physPath, &vdev);
+		}
+		return (ret);
 
 	} catch (ZfsdException &exp) {
 		string context("GeomEvent::OnlineByLabel: " + devPath + ": ");
@@ -251,7 +255,7 @@ GeomEvent::OnlineByLabel(const string &devPath, const string& physPath,
 		exp.GetString().insert(0, context);
 		exp.Log();
 	}
-	return (false);
+	return (ret);
 }