Add zpool status --lockless|--trylock

cmd/zdb/zdb.c

@@ -7723,7 +7723,8 @@ import_checkpointed_state(char *target, nvlist_t *cfg, char **new_path)
 
 	if (cfg == NULL) {
 		zdb_set_skip_mmp(poolname);
-		error = spa_get_stats(poolname, &cfg, NULL, 0);
+		error = spa_get_stats(poolname, &cfg, NULL, 0,
+		    ZPOOL_LOCK_BEHAVIOR_DEFAULT);
 		if (error != 0) {
 			fatal("Tried to read config of pool \"%s\" but "
 			    "spa_get_stats() failed with error %d\n",

cmd/zpool/zpool_iter.c

@@ -118,6 +118,7 @@ pool_list_get(int argc, char **argv, zprop_list_t **proplist, zfs_type_t type,
     boolean_t literal, int *err)
 {
 	zpool_list_t *zlp;
+	int rc;
 
 	zlp = safe_malloc(sizeof (zpool_list_t));
 
@@ -137,7 +138,11 @@ pool_list_get(int argc, char **argv, zprop_list_t **proplist, zfs_type_t type,
 	zlp->zl_literal = literal;
 
 	if (argc == 0) {
-		(void) zpool_iter(g_zfs, add_pool, zlp);
+		rc = zpool_iter(g_zfs, add_pool, zlp);
+		if (rc != 0) {
+			free(zlp);
+			return (NULL);
+		}
 		zlp->zl_findall = B_TRUE;
 	} else {
 		int i;

cmd/zpool/zpool_main.c

@@ -146,7 +146,9 @@ enum zpool_options {
 	ZPOOL_OPTION_ALLOW_ASHIFT_MISMATCH,
 	ZPOOL_OPTION_POOL_KEY_GUID,
 	ZPOOL_OPTION_JSON_NUMS_AS_INT,
-	ZPOOL_OPTION_JSON_FLAT_VDEVS
+	ZPOOL_OPTION_JSON_FLAT_VDEVS,
+	ZPOOL_OPTION_LOCKLESS,
+	ZPOOL_OPTION_TRYLOCK,
 };
 
 /*
@@ -490,8 +492,8 @@ get_usage(zpool_help_t idx)
 	case HELP_LABELCLEAR:
 		return (gettext("\tlabelclear [-f] <vdev>\n"));
 	case HELP_LIST:
-		return (gettext("\tlist [-gHLpPv] [-o property[,...]] [-j "
-		    "[--json-int, --json-pool-key-guid]] ...\n"
+		return (gettext("\tlist [-gHLpPv] [-o property[,...]] "
+		    "[-j [--json-int, --json-pool-key-guid]] ...\n"
 		    "\t    [-T d|u] [pool] [interval [count]]\n"));
 	case HELP_PREFETCH:
 		return (gettext("\tprefetch -t <type> [<type opts>] <pool>\n"
@@ -521,8 +523,8 @@ get_usage(zpool_help_t idx)
 		return (gettext("\ttrim [-dw] [-r <rate>] [-c | -s] <pool> "
 		    "[<device> ...]\n"));
 	case HELP_STATUS:
-		return (gettext("\tstatus [--power] [-j [--json-int, "
-		    "--json-flat-vdevs, ...\n"
+		return (gettext("\tstatus [--power] [--lockless|--trylock] "
+		    "[-j [--json-int, --json-flat-vdevs, ...\n"
 		    "\t    --json-pool-key-guid]] [-c [script1,script2,...]] "
 		    "[-dDegiLpPstvx] ...\n"
 		    "\t    [-T d|u] [pool] [interval [count]]\n"));
@@ -2614,6 +2616,9 @@ typedef struct status_cbdata {
 	nvlist_t	*cb_jsobj;
 	boolean_t	cb_json_as_int;
 	boolean_t	cb_json_pool_key_guid;
+	boolean_t	cb_lockless;
+	boolean_t	cb_trylock;
+
 } status_cbdata_t;
 
 /* Return 1 if string is NULL, empty, or whitespace; return 0 otherwise. */
@@ -11002,6 +11007,8 @@ status_callback(zpool_handle_t *zhp, void *data)
  *	--json-int Display numbers in inteeger format instead of string
  *	--json-flat-vdevs Display vdevs in flat hierarchy
  *	--json-pool-key-guid Use pool GUID as key for pool objects
+ *	--lockless	No locks
+ *	--trylock	Try to get namespace lock, but abort if not
  *
  * Describes the health status of all pools or some subset.
  */
@@ -11024,6 +11031,10 @@ zpool_do_status(int argc, char **argv)
 		    ZPOOL_OPTION_JSON_FLAT_VDEVS},
 		{"json-pool-key-guid", no_argument, NULL,
 		    ZPOOL_OPTION_POOL_KEY_GUID},
+		{"lockless", no_argument, NULL,
+		    ZPOOL_OPTION_LOCKLESS},
+		{"trylock", no_argument, NULL,
+		    ZPOOL_OPTION_TRYLOCK},
 		{0, 0, 0, 0}
 	};
 
@@ -11111,6 +11122,12 @@ zpool_do_status(int argc, char **argv)
 		case ZPOOL_OPTION_POOL_KEY_GUID:
 			cb.cb_json_pool_key_guid = B_TRUE;
 			break;
+		case ZPOOL_OPTION_LOCKLESS:
+			cb.cb_lockless = B_TRUE;
+			break;
+		case ZPOOL_OPTION_TRYLOCK:
+			cb.cb_trylock = B_TRUE;
+			break;
 		case '?':
 			if (optopt == 'c') {
 				print_zpool_script_list("status");
@@ -11152,6 +11169,16 @@ zpool_do_status(int argc, char **argv)
 		usage(B_FALSE);
 	}
 
+	if (cb.cb_lockless && cb.cb_trylock) {
+		(void) fprintf(stderr, gettext("cannot pass both --lockless and"
+		    " --trylock\n"));
+		usage(B_FALSE);
+	} else if (cb.cb_lockless) {
+		libzfs_set_lock_behavior(g_zfs, ZPOOL_LOCK_BEHAVIOR_LOCKLESS);
+	} else if (cb.cb_trylock) {
+		libzfs_set_lock_behavior(g_zfs, ZPOOL_LOCK_BEHAVIOR_TRYLOCK);
+	}
+
 	for (;;) {
 		if (cb.cb_json) {
 			cb.cb_jsobj = zpool_json_schema(0, 1);

include/libzfs.h

@@ -264,6 +264,8 @@ _LIBZFS_H int zpool_create(libzfs_handle_t *, const char *, nvlist_t *,
     nvlist_t *, nvlist_t *);
 _LIBZFS_H int zpool_destroy(zpool_handle_t *, const char *);
 _LIBZFS_H int zpool_add(zpool_handle_t *, nvlist_t *, boolean_t check_ashift);
+_LIBZFS_H void libzfs_set_lock_behavior(libzfs_handle_t *,
+    zpool_lock_behavior_t);
 
 typedef struct splitflags {
 	/* do not split, but return the config that would be split off */

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

@@ -72,4 +72,19 @@ typedef enum {
 #define	mutex_owned(lock)	sx_xlocked(lock)
 #define	mutex_owner(lock)	sx_xholder(lock)
 
+/*
+ * Poor-man's version of Linux kernel's down_timeout(). Try to acquire a mutex
+ * for 'ns' number of nanoseconds.  Returns 0 if mutex was acquired or ETIME
+ * if timeout occurred.
+ */
+static inline int mutex_enter_timeout(kmutex_t *mutex, uint64_t ns)
+{
+	hrtime_t end = gethrtime() + ns;
+	while (gethrtime() < end) {
+		if (mutex_tryenter(mutex))
+			return (0);	/* success */
+	}
+	return (ETIME);
+}
+
 #endif	/* _OPENSOLARIS_SYS_MUTEX_H_ */

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

@@ -26,6 +26,7 @@
 #define	_SPL_MUTEX_H
 
 #include <sys/types.h>
+#include <sys/time.h>
 #include <linux/sched.h>
 #include <linux/mutex.h>
 #include <linux/lockdep.h>
@@ -187,4 +188,19 @@ spl_mutex_lockdep_on_maybe(kmutex_t *mp)			\
 	/* NOTE: do not dereference mp after this point */	\
 }
 
+/*
+ * Poor-man's version of Linux kernel's down_timeout(). Try to acquire a mutex
+ * for 'ns' number of nanoseconds.  Returns 0 if mutex was acquired or ETIME
+ * if timeout occurred.
+ */
+static inline int mutex_enter_timeout(kmutex_t *mutex, uint64_t ns)
+{
+	hrtime_t end = gethrtime() + ns;
+	while (gethrtime() < end) {
+		if (mutex_tryenter(mutex))
+			return (0);	/* success */
+	}
+	return (ETIME);
+}
+
 #endif /* _SPL_MUTEX_H */

include/sys/fs/zfs.h

@@ -866,6 +866,9 @@ typedef struct zpool_load_policy {
 #define	ZPOOL_CONFIG_REBUILD_STATS	"org.openzfs:rebuild_stats"
 #define	ZPOOL_CONFIG_COMPATIBILITY	"compatibility"
 
+/* ZFS_IOC_POOL_STATS argument to for spa_namespace locking behavior */
+#define	ZPOOL_CONFIG_LOCK_BEHAVIOR	"lock_behavior"	/* not stored on disk */
+
 /*
  * The persistent vdev state is stored as separate values rather than a single
  * 'vdev_state' entry.  This is because a device can be in multiple states, such
@@ -1978,6 +1981,30 @@ enum zio_encrypt {
 	    ZFS_XA_NS_PREFIX_MATCH(LINUX_TRUSTED, name) || \
 	    ZFS_XA_NS_PREFIX_MATCH(LINUX_USER, name))
 
+/*
+ * Set locking behavior for zpool commands.
+ */
+typedef enum {
+	/* Wait to acquire the lock on the zpool config */
+	ZPOOL_LOCK_BEHAVIOR_WAIT = 0,
+	ZPOOL_LOCK_BEHAVIOR_DEFAULT = ZPOOL_LOCK_BEHAVIOR_WAIT,
+	/*
+	 * Return an error if it's taking an unnecessarily long time to
+	 * acquire the lock on the pool config (default 100ms)
+	 */
+	ZPOOL_LOCK_BEHAVIOR_TRYLOCK = 1,
+
+	/*
+	 * DANGER: THIS CAN CRASH YOUR SYSTEM
+	 *
+	 * If you can't acquire the pool config lock after 100ms then do a
+	 * a lockless lookup.  This should only be done in emergencies, as it
+	 * can crash the kernel module!
+	 */
+	ZPOOL_LOCK_BEHAVIOR_LOCKLESS = 2,
+	ZPOOL_LOCK_BEHAVIOR_END = 3	/* last entry marker */
+} zpool_lock_behavior_t;
+
 #ifdef	__cplusplus
 }
 #endif

include/sys/spa.h

@@ -742,10 +742,13 @@ typedef enum trim_type {
 
 /* state manipulation functions */
 extern int spa_open(const char *pool, spa_t **, const void *tag);
+extern int spa_open_common_lock_behavior(const char *pool, spa_t **spapp,
+    const void *tag, nvlist_t *nvpolicy, nvlist_t **config,
+    zpool_lock_behavior_t zpool_lock_behavior);
 extern int spa_open_rewind(const char *pool, spa_t **, const void *tag,
     nvlist_t *policy, nvlist_t **config);
 extern int spa_get_stats(const char *pool, nvlist_t **config, char *altroot,
-    size_t buflen);
+    size_t buflen, zpool_lock_behavior_t zpool_lock_behavior);
 extern int spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
     nvlist_t *zplprops, struct dsl_crypto_params *dcp);
 extern int spa_import(char *pool, nvlist_t *config, nvlist_t *props,
@@ -850,10 +853,13 @@ extern kcondvar_t spa_namespace_cv;
 
 extern void spa_write_cachefile(spa_t *, boolean_t, boolean_t, boolean_t);
 extern void spa_config_load(void);
-extern int spa_all_configs(uint64_t *generation, nvlist_t **pools);
+extern int spa_all_configs(uint64_t *generation, nvlist_t **pools,
+    zpool_lock_behavior_t zpool_lock_behavior);
 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 nvlist_t *spa_config_generate_lock_behavior(spa_t *spa, vdev_t *vd,
+    uint64_t txg, int getstats, zpool_lock_behavior_t zpool_lock_behavior);
 extern void spa_config_update(spa_t *spa, int what);
 extern int spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv,
     vdev_t *parent, uint_t id, int atype);
@@ -865,9 +871,11 @@ extern int spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv,
 
 /* Namespace manipulation */
 extern spa_t *spa_lookup(const char *name);
+extern spa_t *spa_lookup_lockless(const char *name);
 extern spa_t *spa_add(const char *name, nvlist_t *config, const char *altroot);
 extern void spa_remove(spa_t *spa);
 extern spa_t *spa_next(spa_t *prev);
+extern spa_t *spa_next_lockless(spa_t *prev);
 
 /* Refcount functions */
 extern void spa_open_ref(spa_t *spa, const void *tag);

include/sys/spa_impl.h

@@ -491,6 +491,8 @@ extern void spa_set_deadman_synctime(hrtime_t ns);
 extern void spa_set_deadman_ziotime(hrtime_t ns);
 extern const char *spa_history_zone(void);
 extern const char *zfs_active_allocator;
+extern int zfs_allow_lockless_zpool_status;
+extern unsigned int spa_namespace_trylock_ms;
 extern int param_set_active_allocator_common(const char *val);
 
 #ifdef	__cplusplus

include/sys/zfs_context.h

@@ -277,6 +277,7 @@ extern void mutex_enter(kmutex_t *mp);
 extern int mutex_enter_check_return(kmutex_t *mp);
 extern void mutex_exit(kmutex_t *mp);
 extern int mutex_tryenter(kmutex_t *mp);
+extern int mutex_enter_timeout(kmutex_t *mp, uint64_t ns);
 
 #define	NESTED_SINGLE 1
 #define	mutex_enter_nested(mp, class) mutex_enter(mp)

lib/libnvpair/libnvpair.abi

@@ -2194,6 +2194,7 @@
       </data-member>
     </class-decl>
     <typedef-decl name='stack_t' type-id='380f9954' id='ac5e685f'/>
+    <typedef-decl name='unw_regnum_t' type-id='95e97e5e' id='c53620f0'/>
     <class-decl name='unw_cursor' size-in-bits='8128' is-struct='yes' visibility='default' id='384a1f22'>
       <data-member access='public' layout-offset-in-bits='0'>
         <var-decl name='opaque' type-id='dc70ec0b' visibility='default'/>
@@ -2306,6 +2307,10 @@
       <parameter type-id='b59d7dce'/>
       <return type-id='79a0948f'/>
     </function-decl>
+    <function-decl name='_Ux86_64_regname' visibility='default' binding='global' size-in-bits='64'>
+      <parameter type-id='c53620f0'/>
+      <return type-id='80f4b756'/>
+    </function-decl>
     <function-decl name='_ULx86_64_init_local' visibility='default' binding='global' size-in-bits='64'>
       <parameter type-id='3946e4d1'/>
       <parameter type-id='2e408b96'/>

lib/libuutil/libuutil.abi

@@ -652,6 +652,7 @@
       </data-member>
     </class-decl>
     <typedef-decl name='stack_t' type-id='380f9954' id='ac5e685f'/>
+    <typedef-decl name='unw_regnum_t' type-id='95e97e5e' id='c53620f0'/>
     <class-decl name='unw_cursor' size-in-bits='8128' is-struct='yes' visibility='default' id='384a1f22'>
       <data-member access='public' layout-offset-in-bits='0'>
         <var-decl name='opaque' type-id='dc70ec0b' visibility='default'/>
@@ -763,6 +764,10 @@
       <parameter type-id='b59d7dce'/>
       <return type-id='79a0948f'/>
     </function-decl>
+    <function-decl name='_Ux86_64_regname' visibility='default' binding='global' size-in-bits='64'>
+      <parameter type-id='c53620f0'/>
+      <return type-id='80f4b756'/>
+    </function-decl>
     <function-decl name='_ULx86_64_init_local' visibility='default' binding='global' size-in-bits='64'>
       <parameter type-id='3946e4d1'/>
       <parameter type-id='2e408b96'/>
@@ -1011,16 +1016,9 @@
     </function-decl>
   </abi-instr>
   <abi-instr address-size='64' path='lib/libspl/os/linux/getmntany.c' language='LANG_C99'>
-    <array-type-def dimensions='1' type-id='38b51b3c' size-in-bits='832' id='02b72c00'>
-      <subrange length='13' type-id='7359adad' id='487fded1'/>
-    </array-type-def>
     <array-type-def dimensions='1' type-id='03085adc' size-in-bits='192' id='083f8d58'>
       <subrange length='3' type-id='7359adad' id='56f209d2'/>
     </array-type-def>
-    <class-decl name='__locale_data' is-struct='yes' visibility='default' is-declaration-only='yes' id='23de8b96'/>
-    <array-type-def dimensions='1' type-id='80f4b756' size-in-bits='832' id='39e6f84a'>
-      <subrange length='13' type-id='7359adad' id='487fded1'/>
-    </array-type-def>
     <class-decl name='mnttab' size-in-bits='256' is-struct='yes' visibility='default' id='1b055409'>
       <data-member access='public' layout-offset-in-bits='0'>
         <var-decl name='mnt_special' type-id='26a90f95' visibility='default'/>
@@ -1130,25 +1128,6 @@
     <typedef-decl name='__blksize_t' type-id='bd54fe1a' id='d3f10a7f'/>
     <typedef-decl name='__blkcnt64_t' type-id='bd54fe1a' id='4e711bf1'/>
     <typedef-decl name='__syscall_slong_t' type-id='bd54fe1a' id='03085adc'/>
-    <class-decl name='__locale_struct' size-in-bits='1856' is-struct='yes' visibility='default' id='90cc1ce3'>
-      <data-member access='public' layout-offset-in-bits='0'>
-        <var-decl name='__locales' type-id='02b72c00' visibility='default'/>
-      </data-member>
-      <data-member access='public' layout-offset-in-bits='832'>
-        <var-decl name='__ctype_b' type-id='31347b7a' visibility='default'/>
-      </data-member>
-      <data-member access='public' layout-offset-in-bits='896'>
-        <var-decl name='__ctype_tolower' type-id='6d60f45d' visibility='default'/>
-      </data-member>
-      <data-member access='public' layout-offset-in-bits='960'>
-        <var-decl name='__ctype_toupper' type-id='6d60f45d' visibility='default'/>
-      </data-member>
-      <data-member access='public' layout-offset-in-bits='1024'>
-        <var-decl name='__names' type-id='39e6f84a' visibility='default'/>
-      </data-member>
-    </class-decl>
-    <typedef-decl name='__locale_t' type-id='f01e1813' id='b7ac9b5f'/>
-    <typedef-decl name='locale_t' type-id='b7ac9b5f' id='973a4f8d'/>
     <class-decl name='timespec' size-in-bits='128' is-struct='yes' visibility='default' id='a9c79a1f'>
       <data-member access='public' layout-offset-in-bits='0'>
         <var-decl name='tv_sec' type-id='65eda9c0' visibility='default'/>
@@ -1157,23 +1136,12 @@
         <var-decl name='tv_nsec' type-id='03085adc' visibility='default'/>
       </data-member>
     </class-decl>
-    <pointer-type-def type-id='23de8b96' size-in-bits='64' id='38b51b3c'/>
-    <pointer-type-def type-id='90cc1ce3' size-in-bits='64' id='f01e1813'/>
-    <qualified-type-def type-id='95e97e5e' const='yes' id='2448a865'/>
-    <pointer-type-def type-id='2448a865' size-in-bits='64' id='6d60f45d'/>
-    <qualified-type-def type-id='8efea9e5' const='yes' id='3beb2af4'/>
-    <pointer-type-def type-id='3beb2af4' size-in-bits='64' id='31347b7a'/>
     <pointer-type-def type-id='0c544dc0' size-in-bits='64' id='394fc496'/>
     <pointer-type-def type-id='56fe4a37' size-in-bits='64' id='b6b61d2f'/>
     <qualified-type-def type-id='b6b61d2f' restrict='yes' id='3cad23cd'/>
     <pointer-type-def type-id='1b055409' size-in-bits='64' id='9d424d31'/>
     <pointer-type-def type-id='0bbec9cd' size-in-bits='64' id='62f7a03d'/>
     <qualified-type-def type-id='62f7a03d' restrict='yes' id='f1cadedf'/>
-    <class-decl name='__locale_data' is-struct='yes' visibility='default' is-declaration-only='yes' id='23de8b96'/>
-    <function-decl name='uselocale' visibility='default' binding='global' size-in-bits='64'>
-      <parameter type-id='973a4f8d'/>
-      <return type-id='973a4f8d'/>
-    </function-decl>
     <function-decl name='getmntent_r' visibility='default' binding='global' size-in-bits='64'>
       <parameter type-id='e75a27e9'/>
       <parameter type-id='3cad23cd'/>
@@ -1185,9 +1153,8 @@
       <parameter type-id='822cd80b'/>
       <return type-id='95e97e5e'/>
     </function-decl>
-    <function-decl name='strerror_l' visibility='default' binding='global' size-in-bits='64'>
+    <function-decl name='strerror' visibility='default' binding='global' size-in-bits='64'>
       <parameter type-id='95e97e5e'/>
-      <parameter type-id='973a4f8d'/>
       <return type-id='26a90f95'/>
     </function-decl>
     <function-decl name='__fprintf_chk' visibility='default' binding='global' size-in-bits='64'>