Support forced export of ZFS pools

cmd/zpool/zpool_main.c

@@ -363,7 +363,7 @@ get_usage(zpool_help_t idx)
 	case HELP_DETACH:
 		return (gettext("\tdetach <pool> <device>\n"));
 	case HELP_EXPORT:
-		return (gettext("\texport [-af] <pool> ...\n"));
+		return (gettext("\texport [-afF] <pool> ...\n"));
 	case HELP_HISTORY:
 		return (gettext("\thistory [-il] [<pool>] ...\n"));
 	case HELP_IMPORT:
@@ -1889,7 +1889,7 @@ zpool_do_destroy(int argc, char **argv)
 		return (1);
 	}
 
-	if (zpool_disable_datasets(zhp, force) != 0) {
+	if (zpool_disable_datasets(zhp, force, FALSE) != 0) {
 		(void) fprintf(stderr, gettext("could not destroy '%s': "
 		    "could not unmount datasets\n"), zpool_get_name(zhp));
 		zpool_close(zhp);
@@ -1919,7 +1919,7 @@ zpool_export_one(zpool_handle_t *zhp, void *data)
 {
 	export_cbdata_t *cb = data;
 
-	if (zpool_disable_datasets(zhp, cb->force) != 0)
+	if (zpool_disable_datasets(zhp, cb->force, cb->hardforce) != 0)
 		return (1);
 
 	/* The history must be logged as part of the export */
@@ -1940,10 +1940,13 @@ zpool_export_one(zpool_handle_t *zhp, void *data)
  *
  *	-a	Export all pools
  *	-f	Forcefully unmount datasets
+ *	-F	Forcefully export, dropping all outstanding dirty data
  *
  * Export the given pools.  By default, the command will attempt to cleanly
  * unmount any active datasets within the pool.  If the '-f' flag is specified,
- * then the datasets will be forcefully unmounted.
+ * then the datasets will be forcefully unmounted.  If the '-F' flag is
+ * specified, the pool's dirty data, if any, will simply be dropped after a
+ * best-effort attempt to forcibly stop all activity.
  */
 int
 zpool_do_export(int argc, char **argv)

include/libzfs.h

@@ -424,6 +424,7 @@ typedef enum {
 	ZPOOL_STATUS_NON_NATIVE_ASHIFT,	/* (e.g. 512e dev with ashift of 9) */
 	ZPOOL_STATUS_COMPATIBILITY_ERR,	/* bad 'compatibility' property */
 	ZPOOL_STATUS_INCOMPATIBLE_FEAT,	/* feature set outside compatibility */
+	ZPOOL_STATUS_FORCE_EXPORTING,	/* pool is being force exported */
 
 	/*
 	 * Finally, the following indicates a healthy pool.
@@ -982,10 +983,16 @@ _LIBZFS_H int zfs_smb_acl_rename(libzfs_handle_t *, char *, char *, char *,
  * sharing/unsharing them.
  */
 _LIBZFS_H int zpool_enable_datasets(zpool_handle_t *, const char *, int);
-_LIBZFS_H int zpool_disable_datasets(zpool_handle_t *, boolean_t);
+_LIBZFS_H int zpool_disable_datasets(zpool_handle_t *, boolean_t, boolean_t);
 _LIBZFS_H void zpool_disable_datasets_os(zpool_handle_t *, boolean_t);
 _LIBZFS_H void zpool_disable_volume_os(const char *);
 
+/*
+ * Procedure to inform os that we have started force unmount (linux specific).
+ */
+_LIBZFS_H void zpool_unmount_mark_hard_force_begin(zpool_handle_t *zhp);
+_LIBZFS_H void zpool_unmount_mark_hard_force_end(zpool_handle_t *zhp);
+
 /*
  * Parse a features file for -o compatibility
  */

include/os/freebsd/spl/sys/thread.h

@@ -31,4 +31,7 @@
 
 #define	getcomm() curthread->td_name
 #define	getpid() curthread->td_tid
+#define	thread_signal spl_kthread_signal
+extern int spl_kthread_signal(kthread_t *tsk, int sig);
+
 #endif

include/os/freebsd/zfs/sys/zfs_znode_impl.h

@@ -135,6 +135,8 @@ zfs_enter(zfsvfs_t *zfsvfs, const char *tag)
 	return (0);
 }
 
+#define	zfs_enter_unmountok	zfs_enter
+
 /* Must be called before exiting the vop */
 static inline void
 zfs_exit(zfsvfs_t *zfsvfs, const char *tag)

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

@@ -53,6 +53,7 @@ typedef void (*thread_func_t)(void *);
 	__thread_create(stk, stksize, (thread_func_t)func, #func,	\
 	arg, len, pp, state, pri)
 
+#define	thread_signal(t, s)		spl_kthread_signal(t, s)
 #define	thread_exit()			spl_thread_exit()
 #define	thread_join(t)			VERIFY(0)
 #define	curthread			current
@@ -64,6 +65,7 @@ extern kthread_t *__thread_create(caddr_t stk, size_t  stksize,
     int state, pri_t pri);
 extern struct task_struct *spl_kthread_create(int (*func)(void *),
     void *data, const char namefmt[], ...);
+extern int spl_kthread_signal(kthread_t *tsk, int sig);
 
 static inline __attribute__((noreturn)) void
 spl_thread_exit(void)

include/os/linux/zfs/sys/zfs_vfsops_os.h

@@ -101,7 +101,8 @@ struct zfsvfs {
 	boolean_t	z_utf8;		/* utf8-only */
 	int		z_norm;		/* normalization flags */
 	boolean_t	z_relatime;	/* enable relatime mount option */
-	boolean_t	z_unmounted;	/* unmounted */
+	boolean_t	z_unmounted;	/* mount status */
+	boolean_t	z_force_unmounted; /* force-unmounted status */
 	rrmlock_t	z_teardown_lock;
 	krwlock_t	z_teardown_inactive_lock;
 	list_t		z_all_znodes;	/* all znodes in the fs */

include/os/linux/zfs/sys/zfs_znode_impl.h

@@ -98,24 +98,39 @@ extern "C" {
 #define	zhold(zp)	VERIFY3P(igrab(ZTOI((zp))), !=, NULL)
 #define	zrele(zp)	iput(ZTOI((zp)))
 
+#define	zfsvfs_is_unmounted(zfsvfs)				\
+	((zfsvfs)->z_unmounted || (zfsvfs)->z_force_unmounted)
+
+/* Must be called before exiting the operation. */
+static inline void
+zfs_exit(zfsvfs_t *zfsvfs, const char *tag)
+{
+	zfs_exit_fs(zfsvfs);
+	ZFS_TEARDOWN_EXIT_READ(zfsvfs, tag);
+}
+
 /* Called on entry to each ZFS inode and vfs operation. */
 static inline int
 zfs_enter(zfsvfs_t *zfsvfs, const char *tag)
 {
 	ZFS_TEARDOWN_ENTER_READ(zfsvfs, tag);
-	if (unlikely(zfsvfs->z_unmounted)) {
+	if (unlikely(zfsvfs_is_unmounted(zfsvfs))) {
 		ZFS_TEARDOWN_EXIT_READ(zfsvfs, tag);
 		return (SET_ERROR(EIO));
 	}
 	return (0);
 }
 
-/* Must be called before exiting the operation. */
-static inline void
-zfs_exit(zfsvfs_t *zfsvfs, const char *tag)
+/* zfs_enter() but ok with forced unmount having begun */
+static inline int
+zfs_enter_unmountok(zfsvfs_t *zfsvfs, const char *tag)
 {
-	zfs_exit_fs(zfsvfs);
-	ZFS_TEARDOWN_EXIT_READ(zfsvfs, tag);
+	ZFS_TEARDOWN_ENTER_READ(zfsvfs, tag);
+	if (unlikely((zfsvfs)->z_unmounted == B_TRUE)) {
+		zfs_exit(zfsvfs, tag);
+		return (SET_ERROR(EIO));
+	}
+	return (0);
 }
 
 static inline int

include/sys/arc.h

@@ -338,6 +338,7 @@ void l2arc_fini(void);
 void l2arc_start(void);
 void l2arc_stop(void);
 void l2arc_spa_rebuild_start(spa_t *spa);
+void l2arc_spa_rebuild_stop(spa_t *spa);
 
 #ifndef _KERNEL
 extern boolean_t arc_watch;

include/sys/dmu.h

@@ -283,6 +283,7 @@ typedef enum dmu_object_type {
 #define	TXG_NOWAIT	(0ULL)
 #define	TXG_WAIT	(1ULL<<0)
 #define	TXG_NOTHROTTLE	(1ULL<<1)
+#define	TXG_NOSUSPEND	(1ULL<<2)
 
 void byteswap_uint64_array(void *buf, size_t size);
 void byteswap_uint32_array(void *buf, size_t size);

include/sys/dmu_impl.h

@@ -241,6 +241,7 @@ typedef struct dmu_sendstatus {
 	list_node_t dss_link;
 	int dss_outfd;
 	proc_t *dss_proc;
+	kthread_t *dss_thread;
 	offset_t *dss_off;
 	uint64_t dss_blocks; /* blocks visited during the sending process */
 } dmu_sendstatus_t;

include/sys/dmu_objset.h

@@ -179,6 +179,7 @@ struct objset {
 
 	/* Protected by os_lock */
 	kmutex_t os_lock;
+	kthread_t *os_shutdown_initiator;
 	multilist_t os_dirty_dnodes[TXG_SIZE];
 	list_t os_dnodes;
 	list_t os_downgraded_dbufs;
@@ -266,6 +267,10 @@ int dmu_fsname(const char *snapname, char *buf);
 void dmu_objset_evict_done(objset_t *os);
 void dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx);
 
+int dmu_objset_shutdown_register(objset_t *os);
+boolean_t dmu_objset_exiting(objset_t *os);
+void dmu_objset_shutdown_unregister(objset_t *os);
+
 void dmu_objset_init(void);
 void dmu_objset_fini(void);
 

include/sys/dmu_recv.h

@@ -40,6 +40,7 @@ extern const char *const recv_clone_name;
 
 typedef struct dmu_recv_cookie {
 	struct dsl_dataset *drc_ds;
+	kthread_t *drc_initiator;
 	struct dmu_replay_record *drc_drr_begin;
 	struct drr_begin *drc_drrb;
 	const char *drc_tofs;
@@ -57,6 +58,8 @@ typedef struct dmu_recv_cookie {
 	nvlist_t *drc_keynvl;
 	uint64_t drc_fromsnapobj;
 	uint64_t drc_ivset_guid;
+	unsigned int drc_flags;
+	void *drc_rwa;
 	void *drc_owner;
 	cred_t *drc_cred;
 	proc_t *drc_proc;
@@ -83,6 +86,7 @@ int dmu_recv_begin(const char *, const char *, dmu_replay_record_t *,
     boolean_t, boolean_t, boolean_t, nvlist_t *, nvlist_t *, const char *,
     dmu_recv_cookie_t *, zfs_file_t *, offset_t *);
 int dmu_recv_stream(dmu_recv_cookie_t *, offset_t *);
+int dmu_recv_close(dsl_dataset_t *ds);
 int dmu_recv_end(dmu_recv_cookie_t *, void *);
 boolean_t dmu_objset_is_receiving(objset_t *);
 

include/sys/dmu_send.h

@@ -60,6 +60,7 @@ int dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
     boolean_t embedok, boolean_t large_block_ok, boolean_t compressok,
     boolean_t rawok, boolean_t savedok, int outfd, offset_t *off,
     struct dmu_send_outparams *dso);
+int dmu_send_close(struct dsl_dataset *ds);
 
 typedef int (*dmu_send_outfunc_t)(objset_t *os, void *buf, int len, void *arg);
 typedef struct dmu_send_outparams {

include/sys/dsl_dataset.h

@@ -242,6 +242,8 @@ typedef struct dsl_dataset {
 	kmutex_t ds_sendstream_lock;
 	list_t ds_sendstreams;
 
+	struct dmu_recv_cookie *ds_receiver;
+
 	/*
 	 * When in the middle of a resumable receive, tracks how much
 	 * progress we have made.
@@ -324,7 +326,8 @@ typedef struct dsl_dataset_rename_snapshot_arg {
 /* flags for holding the dataset */
 typedef enum ds_hold_flags {
 	DS_HOLD_FLAG_NONE	= 0 << 0,
-	DS_HOLD_FLAG_DECRYPT	= 1 << 0 /* needs access to encrypted data */
+	DS_HOLD_FLAG_DECRYPT	= 1 << 0, /* needs access to encrypted data */
+	DS_HOLD_FLAG_MUST_BE_OPEN = 1 << 1, /* dataset must already be open */
 } ds_hold_flags_t;
 
 int dsl_dataset_hold(struct dsl_pool *dp, const char *name, const void *tag,
@@ -453,6 +456,8 @@ void dsl_dataset_long_hold(dsl_dataset_t *ds, const void *tag);
 void dsl_dataset_long_rele(dsl_dataset_t *ds, const void *tag);
 boolean_t dsl_dataset_long_held(dsl_dataset_t *ds);
 
+int dsl_dataset_sendrecv_cancel_all(spa_t *spa);
+
 int dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx);
 void dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,

include/sys/dsl_scan.h

@@ -193,7 +193,7 @@ void dsl_scan_assess_vdev(struct dsl_pool *dp, vdev_t *vd);
 boolean_t dsl_scan_scrubbing(const struct dsl_pool *dp);
 boolean_t dsl_errorscrubbing(const struct dsl_pool *dp);
 boolean_t dsl_errorscrub_active(dsl_scan_t *scn);
-void dsl_scan_restart_resilver(struct dsl_pool *, uint64_t txg);
+int dsl_scan_restart_resilver(struct dsl_pool *dp, uint64_t txg);
 int dsl_scrub_set_pause_resume(const struct dsl_pool *dp,
     pool_scrub_cmd_t cmd);
 void dsl_errorscrub_sync(struct dsl_pool *, dmu_tx_t *);

include/sys/fs/zfs.h

@@ -1487,6 +1487,8 @@ typedef enum zfs_ioc {
 	ZFS_IOC_USERNS_DETACH = ZFS_IOC_UNJAIL,	/* 0x86 (Linux) */
 	ZFS_IOC_SET_BOOTENV,			/* 0x87 */
 	ZFS_IOC_GET_BOOTENV,			/* 0x88 */
+	ZFS_IOC_HARD_FORCE_UNMOUNT_BEGIN,	/* 0x89 (Linux) */
+	ZFS_IOC_HARD_FORCE_UNMOUNT_END,		/* 0x8a (Linux) */
 	ZFS_IOC_LAST
 } zfs_ioc_t;
 

include/sys/metaslab.h

@@ -114,6 +114,7 @@ boolean_t metaslab_class_throttle_reserve(metaslab_class_t *, int, int,
     zio_t *, int);
 void metaslab_class_throttle_unreserve(metaslab_class_t *, int, int, zio_t *);
 void metaslab_class_evict_old(metaslab_class_t *, uint64_t);
+void metaslab_class_force_discard(metaslab_class_t *);
 uint64_t metaslab_class_get_alloc(metaslab_class_t *);
 uint64_t metaslab_class_get_space(metaslab_class_t *);
 uint64_t metaslab_class_get_dspace(metaslab_class_t *);

include/sys/spa.h

@@ -755,6 +755,7 @@ extern int spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 extern int spa_import(char *pool, nvlist_t *config, nvlist_t *props,
     uint64_t flags);
 extern nvlist_t *spa_tryimport(nvlist_t *tryconfig);
+extern int spa_set_pre_export_status(const char *pool, boolean_t status);
 extern int spa_destroy(const char *pool);
 extern int spa_checkpoint(const char *pool);
 extern int spa_checkpoint_discard(const char *pool);
@@ -838,16 +839,13 @@ extern kmutex_t spa_namespace_lock;
  * SPA configuration functions in spa_config.c
  */
 
-#define	SPA_CONFIG_UPDATE_POOL	0
-#define	SPA_CONFIG_UPDATE_VDEVS	1
-
 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);
 extern nvlist_t *spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg,
     int getstats);
-extern void spa_config_update(spa_t *spa, int what);
+extern int spa_config_update_pool(spa_t *spa);
 extern int spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv,
     vdev_t *parent, uint_t id, int atype);
 
@@ -964,6 +962,15 @@ extern void spa_iostats_trim_add(spa_t *spa, trim_type_t type,
     uint64_t extents_written, uint64_t bytes_written,
     uint64_t extents_skipped, uint64_t bytes_skipped,
     uint64_t extents_failed, uint64_t bytes_failed);
+
+typedef enum {
+	/* Config lock handling flags */
+	SCL_FLAG_TRYENTER	= 1U << 0,
+	SCL_FLAG_NOSUSPEND	= 1U << 1,
+	/* MMP flag */
+	SCL_FLAG_MMP		= 1U << 2,
+} spa_config_flag_t;
+
 extern void spa_import_progress_add(spa_t *spa);
 extern void spa_import_progress_remove(uint64_t spa_guid);
 extern int spa_import_progress_set_mmp_check(uint64_t pool_guid,
@@ -976,6 +983,8 @@ extern int spa_import_progress_set_state(uint64_t pool_guid,
 /* Pool configuration locks */
 extern int spa_config_tryenter(spa_t *spa, int locks, const void *tag,
     krw_t rw);
+extern int spa_config_enter_flags(spa_t *spa, int locks, const void *tag,
+    krw_t rw, spa_config_flag_t flags);
 extern void spa_config_enter(spa_t *spa, int locks, const void *tag, krw_t rw);
 extern void spa_config_enter_mmp(spa_t *spa, int locks, const void *tag,
     krw_t rw);
@@ -1026,6 +1035,7 @@ extern uint64_t spa_last_synced_txg(spa_t *spa);
 extern uint64_t spa_first_txg(spa_t *spa);
 extern uint64_t spa_syncing_txg(spa_t *spa);
 extern uint64_t spa_final_dirty_txg(spa_t *spa);
+extern void spa_verify_dirty_txg(spa_t *spa, uint64_t txg);
 extern uint64_t spa_version(spa_t *spa);
 extern pool_state_t spa_state(spa_t *spa);
 extern spa_load_state_t spa_load_state(spa_t *spa);
@@ -1045,6 +1055,8 @@ extern metaslab_class_t *spa_dedup_class(spa_t *spa);
 extern metaslab_class_t *spa_preferred_class(spa_t *spa, uint64_t size,
     dmu_object_type_t objtype, uint_t level, uint_t special_smallblk);
 
+extern void spa_evicting_os_lock(spa_t *);
+extern void spa_evicting_os_unlock(spa_t *);
 extern void spa_evicting_os_register(spa_t *, objset_t *os);
 extern void spa_evicting_os_deregister(spa_t *, objset_t *os);
 extern void spa_evicting_os_wait(spa_t *spa);
@@ -1136,6 +1148,10 @@ extern void spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
 
 extern const char *spa_state_to_name(spa_t *spa);
 
+extern boolean_t spa_exiting_any(spa_t *spa);
+extern boolean_t spa_exiting(spa_t *spa);
+extern int spa_operation_interrupted(spa_t *spa);
+
 /* error handling */
 struct zbookmark_phys;
 extern void spa_log_error(spa_t *spa, const zbookmark_phys_t *zb,

include/sys/spa_impl.h

@@ -245,6 +245,8 @@ struct spa {
 	kmutex_t	spa_evicting_os_lock;	/* Evicting objset list lock */
 	list_t		spa_evicting_os_list;	/* Objsets being evicted. */
 	kcondvar_t	spa_evicting_os_cv;	/* Objset Eviction Completion */
+	kthread_t	*spa_export_initiator;	/* thread exporting the pool */
+	boolean_t	spa_pre_exporting;	/* allow fails before export */
 	txg_list_t	spa_vdev_txg_list;	/* per-txg dirty vdev list */
 	vdev_t		*spa_root_vdev;		/* top-level vdev container */
 	uint64_t	spa_min_ashift;		/* of vdevs in normal class */