Add libtpool (thread pools)

OpenZFS provides a library called tpool which implements thread
pools for user space applications.  Porting this library means
the zpool utility no longer needs to borrow the kernel mutex and
taskq interfaces from libzpool.  This code was updated to use
the tpool library which behaves in a very similar fashion.

Porting libtpool was relatively straight forward and minimal
modifications were needed.  The core changes were:

* Fully convert the library to use pthreads.
* Updated signal handling.
* lmalloc/lfree converted to calloc/free
* Implemented portable pthread_attr_clone() function.

Finally, update the build system such that libzpool.so is no
longer linked in to zfs(8), zpool(8), etc.  All that is required
is libzfs to which the zcommon soures were added (which is the way
it always should have been).  Removing the libzpool dependency
resulted in several build issues which needed to be resolved.

* Moved zfeature support to module/zcommon/zfeature_common.c
* Moved ratelimiting to to module/zfs/zfs_ratelimit.c
* Moved get_system_hostid() to lib/libspl/gethostid.c
* Removed use of cmn_err() in zcommon source
* Removed dprintf_setup() call from zpool_main.c and zfs_main.c
* Removed highbit() and lowbit()
* Removed unnecessary library dependencies from Makefiles
* Removed fletcher-4 kstat in user space
* Added sha2 support explicitly to libzfs
* Added highbit64() and lowbit64() to zpool_util.c

Reviewed-by: Tony Hutter <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Closes #6442 

15 files changed, 931 insertions, 227 deletions

diff --git a/lib/Makefile.am b/lib/Makefile.am
index 3aec922c9..e1833b842 100644
--- a/lib/Makefile.am
+++ b/lib/Makefile.am
@@ -1,6 +1,6 @@
 # NB: GNU Automake Manual, Chapter 8.3.5: Libtool Convenience Libraries
 # These six libraries are intermediary build components.
-SUBDIRS = libspl libavl libefi libshare libunicode libicp
+SUBDIRS = libavl libefi libicp libshare libspl libtpool libunicode
 
 # These four libraries, which are installed as the final build product,
 # incorporate the six convenience libraries given above.
diff --git a/lib/libnvpair/Makefile.am b/lib/libnvpair/Makefile.am
index 2833557a3..6da679fd5 100644
--- a/lib/libnvpair/Makefile.am
+++ b/lib/libnvpair/Makefile.am
@@ -25,9 +25,7 @@ nodist_libnvpair_la_SOURCES = \
 	$(USER_C) \
 	$(KERNEL_C)
 
-libnvpair_la_LIBADD = \
-	$(top_builddir)/lib/libuutil/libuutil.la \
-	$(LIBTIRPC)
+libnvpair_la_LIBADD = $(LIBTIRPC)
 libnvpair_la_LDFLAGS = -version-info 1:1:0
 
 EXTRA_DIST = $(USER_C)
diff --git a/lib/libspl/Makefile.am b/lib/libspl/Makefile.am
index 3c99529f1..28ae15351 100644
--- a/lib/libspl/Makefile.am
+++ b/lib/libspl/Makefile.am
@@ -19,6 +19,7 @@ noinst_LTLIBRARIES = libspl.la
 
 USER_C = \
 	getexecname.c \
+	gethostid.c \
 	gethrtime.c \
 	gethrestime.c \
 	getmntany.c \
diff --git a/lib/libspl/gethostid.c b/lib/libspl/gethostid.c
new file mode 100644
index 000000000..1eb93f441
--- /dev/null
+++ b/lib/libspl/gethostid.c
@@ -0,0 +1,86 @@
+/*
+ * 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) 2017, Lawrence Livermore National Security, LLC.
+ */
+
+#include <fcntl.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/systeminfo.h>
+
+static unsigned long
+get_spl_hostid(void)
+{
+	FILE *f;
+	unsigned long hostid;
+	char *env;
+
+	/*
+	 * Allow the hostid to be subverted for testing.
+	 */
+	env = getenv("ZFS_HOSTID");
+	if (env) {
+		hostid = strtoull(env, NULL, 0);
+		return (hostid & HOSTID_MASK);
+	}
+
+	f = fopen("/sys/module/spl/parameters/spl_hostid", "r");
+	if (!f)
+		return (0);
+
+	if (fscanf(f, "%lu", &hostid) != 1)
+		hostid = 0;
+
+	fclose(f);
+
+	return (hostid & HOSTID_MASK);
+}
+
+unsigned long
+get_system_hostid(void)
+{
+	unsigned long system_hostid = get_spl_hostid();
+	/*
+	 * We do not use the library call gethostid() because
+	 * it generates a hostid value that the kernel is
+	 * unaware of, if the spl_hostid module parameter has not
+	 * been set and there is no system hostid file (e.g.
+	 * /etc/hostid).  The kernel and userspace must agree.
+	 * See comments above hostid_read() in the SPL.
+	 */
+	if (system_hostid == 0) {
+		int fd, rc;
+		unsigned long hostid;
+		int hostid_size = 4;  /* 4 bytes regardless of arch */
+
+		fd = open("/etc/hostid", O_RDONLY);
+		if (fd >= 0) {
+			rc = read(fd, &hostid, hostid_size);
+			if (rc > 0)
+				system_hostid = (hostid & HOSTID_MASK);
+			close(fd);
+		}
+	}
+	return (system_hostid);
+}
diff --git a/lib/libspl/include/sys/systeminfo.h b/lib/libspl/include/sys/systeminfo.h
index 3f7cef5b9..cc6c1793c 100644
--- a/lib/libspl/include/sys/systeminfo.h
+++ b/lib/libspl/include/sys/systeminfo.h
@@ -27,11 +27,12 @@
 #ifndef _LIBSPL_SYS_SYSTEMINFO_H
 #define	_LIBSPL_SYS_SYSTEMINFO_H
 
+#define	HOSTID_MASK		0xFFFFFFFF
 #define	HW_INVALID_HOSTID	0xFFFFFFFF	/* an invalid hostid */
 #define	HW_HOSTID_LEN		11		/* minimum buffer size needed */
 						/* to hold a decimal or hex */
 						/* hostid string */
 
-#define	sysinfo(cmd, buf, cnt)		(-1)
+unsigned long get_system_hostid(void);
 
 #endif
diff --git a/lib/libtpool/Makefile.am b/lib/libtpool/Makefile.am
new file mode 100644
index 000000000..adbaee6c3
--- /dev/null
+++ b/lib/libtpool/Makefile.am
@@ -0,0 +1,26 @@
+include $(top_srcdir)/config/Rules.am
+
+AM_CFLAGS += $(DEBUG_STACKFLAGS)
+
+DEFAULT_INCLUDES += \
+	-I$(top_srcdir)/include \
+	-I$(top_srcdir)/lib/libspl/include
+
+noinst_LTLIBRARIES = libtpool.la
+
+USER_C = \
+	thread_pool.c \
+	thread_pool_impl.h
+
+KERNEL_C =
+
+nodist_libtpool_la_SOURCES = \
+	$(USER_C) \
+	$(KERNEL_C)
+
+libtpool_la_LIBADD = \
+	$(top_builddir)/lib/libspl/libspl.la
+
+libtpool_la_LDFLAGS = -pthread
+
+EXTRA_DIST = $(USER_C)
diff --git a/lib/libtpool/thread_pool.c b/lib/libtpool/thread_pool.c
new file mode 100644
index 000000000..166ce3e67
--- /dev/null
+++ b/lib/libtpool/thread_pool.c
@@ -0,0 +1,650 @@
+/*
+ * 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 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#include <stdlib.h>
+#include <signal.h>
+#include <errno.h>
+#include <assert.h>
+#include "thread_pool_impl.h"
+
+static pthread_mutex_t thread_pool_lock = PTHREAD_MUTEX_INITIALIZER;
+static tpool_t *thread_pools = NULL;
+
+static void
+delete_pool(tpool_t *tpool)
+{
+	tpool_job_t *job;
+
+	ASSERT(tpool->tp_current == 0 && tpool->tp_active == NULL);
+
+	/*
+	 * Unlink the pool from the global list of all pools.
+	 */
+	(void) pthread_mutex_lock(&thread_pool_lock);
+	if (thread_pools == tpool)
+		thread_pools = tpool->tp_forw;
+	if (thread_pools == tpool)
+		thread_pools = NULL;
+	else {
+		tpool->tp_back->tp_forw = tpool->tp_forw;
+		tpool->tp_forw->tp_back = tpool->tp_back;
+	}
+	pthread_mutex_unlock(&thread_pool_lock);
+
+	/*
+	 * There should be no pending jobs, but just in case...
+	 */
+	for (job = tpool->tp_head; job != NULL; job = tpool->tp_head) {
+		tpool->tp_head = job->tpj_next;
+		free(job);
+	}
+	(void) pthread_attr_destroy(&tpool->tp_attr);
+	free(tpool);
+}
+
+/*
+ * Worker thread is terminating.
+ */
+static void
+worker_cleanup(void *arg)
+{
+	tpool_t *tpool = (tpool_t *)arg;
+
+	if (--tpool->tp_current == 0 &&
+	    (tpool->tp_flags & (TP_DESTROY | TP_ABANDON))) {
+		if (tpool->tp_flags & TP_ABANDON) {
+			pthread_mutex_unlock(&tpool->tp_mutex);
+			delete_pool(tpool);
+			return;
+		}
+		if (tpool->tp_flags & TP_DESTROY)
+			(void) pthread_cond_broadcast(&tpool->tp_busycv);
+	}
+	pthread_mutex_unlock(&tpool->tp_mutex);
+}
+
+static void
+notify_waiters(tpool_t *tpool)
+{
+	if (tpool->tp_head == NULL && tpool->tp_active == NULL) {
+		tpool->tp_flags &= ~TP_WAIT;
+		(void) pthread_cond_broadcast(&tpool->tp_waitcv);
+	}
+}
+
+/*
+ * Called by a worker thread on return from a tpool_dispatch()d job.
+ */
+static void
+job_cleanup(void *arg)
+{
+	tpool_t *tpool = (tpool_t *)arg;
+
+	pthread_t my_tid = pthread_self();
+	tpool_active_t *activep;
+	tpool_active_t **activepp;
+
+	pthread_mutex_lock(&tpool->tp_mutex);
+	/* CSTYLED */
+	for (activepp = &tpool->tp_active;; activepp = &activep->tpa_next) {
+		activep = *activepp;
+		if (activep->tpa_tid == my_tid) {
+			*activepp = activep->tpa_next;
+			break;
+		}
+	}
+	if (tpool->tp_flags & TP_WAIT)
+		notify_waiters(tpool);
+}
+
+static void *
+tpool_worker(void *arg)
+{
+	tpool_t *tpool = (tpool_t *)arg;
+	int elapsed;
+	tpool_job_t *job;
+	void (*func)(void *);
+	tpool_active_t active;
+
+	pthread_mutex_lock(&tpool->tp_mutex);
+	pthread_cleanup_push(worker_cleanup, tpool);
+
+	/*
+	 * This is the worker's main loop.
+	 * It will only be left if a timeout or an error has occured.
+	 */
+	active.tpa_tid = pthread_self();
+	for (;;) {
+		elapsed = 0;
+		tpool->tp_idle++;
+		if (tpool->tp_flags & TP_WAIT)
+			notify_waiters(tpool);
+		while ((tpool->tp_head == NULL ||
+		    (tpool->tp_flags & TP_SUSPEND)) &&
+		    !(tpool->tp_flags & (TP_DESTROY | TP_ABANDON))) {
+			if (tpool->tp_current <= tpool->tp_minimum ||
+			    tpool->tp_linger == 0) {
+				(void) pthread_cond_wait(&tpool->tp_workcv,
+				    &tpool->tp_mutex);
+			} else {
+				struct timespec ts;
+
+				clock_gettime(CLOCK_REALTIME, &ts);
+				ts.tv_sec += tpool->tp_linger;
+
+				if (pthread_cond_timedwait(&tpool->tp_workcv,
+				    &tpool->tp_mutex, &ts) != 0) {
+					elapsed = 1;
+					break;
+				}
+			}
+		}
+		tpool->tp_idle--;
+		if (tpool->tp_flags & TP_DESTROY)
+			break;
+		if (tpool->tp_flags & TP_ABANDON) {
+			/* can't abandon a suspended pool */
+			if (tpool->tp_flags & TP_SUSPEND) {
+				tpool->tp_flags &= ~TP_SUSPEND;
+				(void) pthread_cond_broadcast(
+				    &tpool->tp_workcv);
+			}
+			if (tpool->tp_head == NULL)
+				break;
+		}
+		if ((job = tpool->tp_head) != NULL &&
+		    !(tpool->tp_flags & TP_SUSPEND)) {
+			elapsed = 0;
+			func = job->tpj_func;
+			arg = job->tpj_arg;
+			tpool->tp_head = job->tpj_next;
+			if (job == tpool->tp_tail)
+				tpool->tp_tail = NULL;
+			tpool->tp_njobs--;
+			active.tpa_next = tpool->tp_active;
+			tpool->tp_active = &active;
+			pthread_mutex_unlock(&tpool->tp_mutex);
+			pthread_cleanup_push(job_cleanup, tpool);
+			free(job);
+
+			sigset_t maskset;
+			(void) pthread_sigmask(SIG_SETMASK, NULL, &maskset);
+
+			/*
+			 * Call the specified function.
+			 */
+			func(arg);
+			/*
+			 * We don't know what this thread has been doing,
+			 * so we reset its signal mask and cancellation
+			 * state back to the values prior to calling func().
+			 */
+			(void) pthread_sigmask(SIG_SETMASK, &maskset, NULL);
+			(void) pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED,
+			    NULL);
+			(void) pthread_setcancelstate(PTHREAD_CANCEL_ENABLE,
+			    NULL);
+			pthread_cleanup_pop(1);
+		}
+		if (elapsed && tpool->tp_current > tpool->tp_minimum) {
+			/*
+			 * We timed out and there is no work to be done
+			 * and the number of workers exceeds the minimum.
+			 * Exit now to reduce the size of the pool.
+			 */
+			break;
+		}
+	}
+	pthread_cleanup_pop(1);
+	return (arg);
+}
+
+/*
+ * Create a worker thread, with default signals blocked.
+ */
+static int
+create_worker(tpool_t *tpool)
+{
+	pthread_t thread;
+	sigset_t oset;
+	int error;
+
+	(void) pthread_sigmask(SIG_SETMASK, NULL, &oset);
+	error = pthread_create(&thread, &tpool->tp_attr, tpool_worker, tpool);
+	(void) pthread_sigmask(SIG_SETMASK, &oset, NULL);
+	return (error);
+}
+
+
+/*
+ * pthread_attr_clone: make a copy of a pthread_attr_t.  When old_attr
+ * is NULL initialize the cloned attr using default values.
+ */
+static int
+pthread_attr_clone(pthread_attr_t *attr, const pthread_attr_t *old_attr)
+{
+	int error;
+
+	error = pthread_attr_init(attr);
+	if (error || (old_attr == NULL))
+		return (error);
+
+	cpu_set_t cpuset;
+	size_t cpusetsize = sizeof (cpuset);
+	error = pthread_attr_getaffinity_np(old_attr, cpusetsize, &cpuset);
+	if (error == 0)
+		error = pthread_attr_setaffinity_np(attr, cpusetsize, &cpuset);
+	if (error)
+		goto error;
+
+	int detachstate;
+	error = pthread_attr_getdetachstate(old_attr, &detachstate);
+	if (error == 0)
+		error = pthread_attr_setdetachstate(attr, detachstate);
+	if (error)
+		goto error;
+
+	size_t guardsize;
+	error = pthread_attr_getguardsize(old_attr, &guardsize);
+	if (error == 0)
+		error = pthread_attr_setguardsize(attr, guardsize);
+	if (error)
+		goto error;
+
+	int inheritsched;
+	error = pthread_attr_getinheritsched(old_attr, &inheritsched);
+	if (error == 0)
+		error = pthread_attr_setinheritsched(attr, inheritsched);
+	if (error)
+		goto error;
+
+	struct sched_param param;
+	error = pthread_attr_getschedparam(old_attr, &param);
+	if (error == 0)
+		error = pthread_attr_setschedparam(attr, &param);
+	if (error)
+		goto error;
+
+	int policy;
+	error = pthread_attr_getschedpolicy(old_attr, &policy);
+	if (error == 0)
+		error = pthread_attr_setschedpolicy(attr, policy);
+	if (error)
+		goto error;
+
+	int scope;
+	error = pthread_attr_getscope(old_attr, &scope);
+	if (error == 0)
+		error = pthread_attr_setscope(attr, scope);
+	if (error)
+		goto error;
+
+	void *stackaddr;
+	size_t stacksize;
+	error = pthread_attr_getstack(old_attr, &stackaddr, &stacksize);
+	if (error == 0)
+		error = pthread_attr_setstack(attr, stackaddr, stacksize);
+	if (error)
+		goto error;
+
+	return (0);
+error:
+	pthread_attr_destroy(attr);
+	return (error);
+}
+
+tpool_t	*
+tpool_create(uint_t min_threads, uint_t max_threads, uint_t linger,
+    pthread_attr_t *attr)
+{
+	tpool_t	*tpool;
+	void *stackaddr;
+	size_t stacksize;
+	size_t minstack;
+	int error;
+
+	if (min_threads > max_threads || max_threads < 1) {
+		errno = EINVAL;
+		return (NULL);
+	}
+	if (attr != NULL) {
+		if (pthread_attr_getstack(attr, &stackaddr, &stacksize) != 0) {
+			errno = EINVAL;
+			return (NULL);
+		}
+		/*
+		 * Allow only one thread in the pool with a specified stack.
+		 * Require threads to have at least the minimum stack size.
+		 */
+		minstack = PTHREAD_STACK_MIN;
+		if (stackaddr != NULL) {
+			if (stacksize < minstack || max_threads != 1) {
+				errno = EINVAL;
+				return (NULL);
+			}
+		} else if (stacksize != 0 && stacksize < minstack) {
+			errno = EINVAL;
+			return (NULL);
+		}
+	}
+
+	tpool = calloc(1, sizeof (*tpool));
+	if (tpool == NULL) {
+		errno = ENOMEM;
+		return (NULL);
+	}
+	(void) pthread_mutex_init(&tpool->tp_mutex, NULL);
+	(void) pthread_cond_init(&tpool->tp_busycv, NULL);
+	(void) pthread_cond_init(&tpool->tp_workcv, NULL);
+	(void) pthread_cond_init(&tpool->tp_waitcv, NULL);
+	tpool->tp_minimum = min_threads;
+	tpool->tp_maximum = max_threads;
+	tpool->tp_linger = linger;
+
+	/*
+	 * We cannot just copy the attribute pointer.
+	 * We need to initialize a new pthread_attr_t structure
+	 * with the values from the user-supplied pthread_attr_t.
+	 * If the attribute pointer is NULL, we need to initialize
+	 * the new pthread_attr_t structure with default values.
+	 */
+	error = pthread_attr_clone(&tpool->tp_attr, attr);
+	if (error) {
+		free(tpool);
+		errno = error;
+		return (NULL);
+	}
+
+	/* make all pool threads be detached daemon threads */
+	(void) pthread_attr_setdetachstate(&tpool->tp_attr,
+	    PTHREAD_CREATE_DETACHED);
+
+	/* insert into the global list of all thread pools */
+	pthread_mutex_lock(&thread_pool_lock);
+	if (thread_pools == NULL) {
+		tpool->tp_forw = tpool;
+		tpool->tp_back = tpool;
+		thread_pools = tpool;
+	} else {
+		thread_pools->tp_back->tp_forw = tpool;
+		tpool->tp_forw = thread_pools;
+		tpool->tp_back = thread_pools->tp_back;
+		thread_pools->tp_back = tpool;
+	}
+	pthread_mutex_unlock(&thread_pool_lock);
+
+	return (tpool);
+}
+
+/*
+ * Dispatch a work request to the thread pool.
+ * If there are idle workers, awaken one.
+ * Else, if the maximum number of workers has
+ * not been reached, spawn a new worker thread.
+ * Else just return with the job added to the queue.
+ */
+int
+tpool_dispatch(tpool_t *tpool, void (*func)(void *), void *arg)
+{
+	tpool_job_t *job;
+
+	ASSERT(!(tpool->tp_flags & (TP_DESTROY | TP_ABANDON)));
+
+	if ((job = calloc(1, sizeof (*job))) == NULL)
+		return (-1);
+	job->tpj_next = NULL;
+	job->tpj_func = func;
+	job->tpj_arg = arg;
+
+	pthread_mutex_lock(&tpool->tp_mutex);
+
+	if (tpool->tp_head == NULL)
+		tpool->tp_head = job;
+	else
+		tpool->tp_tail->tpj_next = job;
+	tpool->tp_tail = job;
+	tpool->tp_njobs++;
+
+	if (!(tpool->tp_flags & TP_SUSPEND)) {
+		if (tpool->tp_idle > 0)
+			(void) pthread_cond_signal(&tpool->tp_workcv);
+		else if (tpool->tp_current < tpool->tp_maximum &&
+		    create_worker(tpool) == 0)
+			tpool->tp_current++;
+	}
+
+	pthread_mutex_unlock(&tpool->tp_mutex);
+	return (0);
+}
+
+static void
+tpool_cleanup(void *arg)
+{
+	tpool_t *tpool = (tpool_t *)arg;
+
+	pthread_mutex_unlock(&tpool->tp_mutex);
+}
+
+/*
+ * Assumes: by the time tpool_destroy() is called no one will use this
+ * thread pool in any way and no one will try to dispatch entries to it.
+ * Calling tpool_destroy() from a job in the pool will cause deadlock.
+ */
+void
+tpool_destroy(tpool_t *tpool)
+{
+	tpool_active_t *activep;
+
+	ASSERT(!tpool_member(tpool));
+	ASSERT(!(tpool->tp_flags & (TP_DESTROY | TP_ABANDON)));
+
+	pthread_mutex_lock(&tpool->tp_mutex);
+	pthread_cleanup_push(tpool_cleanup, tpool);
+
+	/* mark the pool as being destroyed; wakeup idle workers */
+	tpool->tp_flags |= TP_DESTROY;
+	tpool->tp_flags &= ~TP_SUSPEND;
+	(void) pthread_cond_broadcast(&tpool->tp_workcv);
+
+	/* cancel all active workers */
+	for (activep = tpool->tp_active; activep; activep = activep->tpa_next)
+		(void) pthread_cancel(activep->tpa_tid);
+
+	/* wait for all active workers to finish */
+	while (tpool->tp_active != NULL) {
+		tpool->tp_flags |= TP_WAIT;
+		(void) pthread_cond_wait(&tpool->tp_waitcv, &tpool->tp_mutex);
+	}
+
+	/* the last worker to terminate will wake us up */
+	while (tpool->tp_current != 0)
+		(void) pthread_cond_wait(&tpool->tp_busycv, &tpool->tp_mutex);
+
+	pthread_cleanup_pop(1);	/* pthread_mutex_unlock(&tpool->tp_mutex); */
+	delete_pool(tpool);
+}
+
+/*
+ * Like tpool_destroy(), but don't cancel workers or wait for them to finish.
+ * The last worker to terminate will delete the pool.
+ */
+void
+tpool_abandon(tpool_t *tpool)
+{
+	ASSERT(!(tpool->tp_flags & (TP_DESTROY | TP_ABANDON)));
+
+	pthread_mutex_lock(&tpool->tp_mutex);
+	if (tpool->tp_current == 0) {
+		/* no workers, just delete the pool */
+		pthread_mutex_unlock(&tpool->tp_mutex);
+		delete_pool(tpool);
+	} else {
+		/* wake up all workers, last one will delete the pool */
+		tpool->tp_flags |= TP_ABANDON;
+		tpool->tp_flags &= ~TP_SUSPEND;
+		(void) pthread_cond_broadcast(&tpool->tp_workcv);
+		pthread_mutex_unlock(&tpool->tp_mutex);
+	}
+}
+
+/*
+ * Wait for all jobs to complete.
+ * Calling tpool_wait() from a job in the pool will cause deadlock.
+ */
+void
+tpool_wait(tpool_t *tpool)
+{
+	ASSERT(!tpool_member(tpool));
+	ASSERT(!(tpool->tp_flags & (TP_DESTROY | TP_ABANDON)));
+
+	pthread_mutex_lock(&tpool->tp_mutex);
+	pthread_cleanup_push(tpool_cleanup, tpool);
+	while (tpool->tp_head != NULL || tpool->tp_active != NULL) {
+		tpool->tp_flags |= TP_WAIT;
+		(void) pthread_cond_wait(&tpool->tp_waitcv, &tpool->tp_mutex);
+		ASSERT(!(tpool->tp_flags & (TP_DESTROY | TP_ABANDON)));
+	}
+	pthread_cleanup_pop(1);	/* pthread_mutex_unlock(&tpool->tp_mutex); */
+}
+
+void
+tpool_suspend(tpool_t *tpool)
+{
+	ASSERT(!(tpool->tp_flags & (TP_DESTROY | TP_ABANDON)));
+
+	pthread_mutex_lock(&tpool->tp_mutex);
+	tpool->tp_flags |= TP_SUSPEND;
+	pthread_mutex_unlock(&tpool->tp_mutex);
+}
+
+int
+tpool_suspended(tpool_t *tpool)
+{
+	int suspended;
+
+	ASSERT(!(tpool->tp_flags & (TP_DESTROY | TP_ABANDON)));
+
+	pthread_mutex_lock(&tpool->tp_mutex);
+	suspended = (tpool->tp_flags & TP_SUSPEND) != 0;
+	pthread_mutex_unlock(&tpool->tp_mutex);
+
+	return (suspended);
+}
+
+void
+tpool_resume(tpool_t *tpool)
+{
+	int excess;
+
+	ASSERT(!(tpool->tp_flags & (TP_DESTROY | TP_ABANDON)));
+
+	pthread_mutex_lock(&tpool->tp_mutex);
+	if (!(tpool->tp_flags & TP_SUSPEND)) {
+		pthread_mutex_unlock(&tpool->tp_mutex);
+		return;
+	}
+	tpool->tp_flags &= ~TP_SUSPEND;
+	(void) pthread_cond_broadcast(&tpool->tp_workcv);
+	excess = tpool->tp_njobs - tpool->tp_idle;
+	while (excess-- > 0 && tpool->tp_current < tpool->tp_maximum) {
+		if (create_worker(tpool) != 0)
+			break;		/* pthread_create() failed */
+		tpool->tp_current++;
+	}
+	pthread_mutex_unlock(&tpool->tp_mutex);
+}
+
+int
+tpool_member(tpool_t *tpool)
+{
+	pthread_t my_tid = pthread_self();
+	tpool_active_t *activep;
+
+	ASSERT(!(tpool->tp_flags & (TP_DESTROY | TP_ABANDON)));
+
+	pthread_mutex_lock(&tpool->tp_mutex);
+	for (activep = tpool->tp_active; activep; activep = activep->tpa_next) {
+		if (activep->tpa_tid == my_tid) {
+			pthread_mutex_unlock(&tpool->tp_mutex);
+			return (1);
+		}
+	}
+	pthread_mutex_unlock(&tpool->tp_mutex);
+	return (0);
+}
+
+void
+postfork1_child_tpool(void)
+{
+	pthread_t my_tid = pthread_self();
+	tpool_t *tpool;
+	tpool_job_t *job;
+
+	/*
+	 * All of the thread pool workers are gone, except possibly
+	 * for the current thread, if it is a thread pool worker thread.
+	 * Retain the thread pools, but make them all empty.  Whatever
+	 * jobs were queued or running belong to the parent process.
+	 */
+top:
+	if ((tpool = thread_pools) == NULL)
+		return;
+
+	do {
+		tpool_active_t *activep;
+
+		(void) pthread_mutex_init(&tpool->tp_mutex, NULL);
+		(void) pthread_cond_init(&tpool->tp_busycv, NULL);
+		(void) pthread_cond_init(&tpool->tp_workcv, NULL);
+		(void) pthread_cond_init(&tpool->tp_waitcv, NULL);
+		for (job = tpool->tp_head; job; job = tpool->tp_head) {
+			tpool->tp_head = job->tpj_next;
+			free(job);
+		}
+		tpool->tp_tail = NULL;
+		tpool->tp_njobs = 0;
+		for (activep = tpool->tp_active; activep;
+		    activep = activep->tpa_next) {
+			if (activep->tpa_tid == my_tid) {
+				activep->tpa_next = NULL;
+				break;
+			}
+		}
+		tpool->tp_idle = 0;
+		tpool->tp_current = 0;
+		if ((tpool->tp_active = activep) != NULL)
+			tpool->tp_current = 1;
+		tpool->tp_flags &= ~TP_WAIT;
+		if (tpool->tp_flags & (TP_DESTROY | TP_ABANDON)) {
+			tpool->tp_flags &= ~TP_DESTROY;
+			tpool->tp_flags |= TP_ABANDON;
+			if (tpool->tp_current == 0) {
+				delete_pool(tpool);
+				goto top;	/* start over */
+			}
+		}
+	} while ((tpool = tpool->tp_forw) != thread_pools);
+}
diff --git a/lib/libtpool/thread_pool_impl.h b/lib/libtpool/thread_pool_impl.h
new file mode 100644
index 000000000..5349c2ade
--- /dev/null
+++ b/lib/libtpool/thread_pool_impl.h
@@ -0,0 +1,93 @@
+/*
+ * 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 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#ifndef _THREAD_POOL_IMPL_H
+#define	_THREAD_POOL_IMPL_H
+
+#include <thread_pool.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * Thread pool implementation definitions.
+ * See <thread_pool.h> for interface declarations.
+ */
+
+/*
+ * FIFO queued job
+ */
+typedef struct tpool_job tpool_job_t;
+struct tpool_job {
+	tpool_job_t	*tpj_next;		/* list of jobs */
+	void		(*tpj_func)(void *);	/* function to call */
+	void		*tpj_arg;		/* its argument */
+};
+
+/*
+ * List of active threads, linked through their stacks.
+ */
+typedef struct tpool_active tpool_active_t;
+struct tpool_active {
+	tpool_active_t	*tpa_next;	/* list of active threads */
+	pthread_t	tpa_tid;	/* active thread id */
+};
+
+/*
+ * The thread pool.
+ */
+struct tpool {
+	tpool_t		*tp_forw;	/* circular list of all thread pools */
+	tpool_t		*tp_back;
+	pthread_mutex_t	tp_mutex;	/* protects the pool data */
+	pthread_cond_t	tp_busycv;	/* synchronization in tpool_dispatch */
+	pthread_cond_t	tp_workcv;	/* synchronization with workers */
+	pthread_cond_t	tp_waitcv;	/* synchronization in tpool_wait() */
+	tpool_active_t	*tp_active;	/* threads performing work */
+	tpool_job_t	*tp_head;	/* FIFO job queue */
+	tpool_job_t	*tp_tail;
+	pthread_attr_t	tp_attr;	/* attributes of the workers */
+	int		tp_flags;	/* see below */
+	uint_t		tp_linger;	/* seconds before idle workers exit */
+	int		tp_njobs;	/* number of jobs in job queue */
+	int		tp_minimum;	/* minimum number of worker threads */
+	int		tp_maximum;	/* maximum number of worker threads */
+	int		tp_current;	/* current number of worker threads */
+	int		tp_idle;	/* number of idle workers */
+};
+
+/* tp_flags */
+#define	TP_WAIT		0x01		/* waiting in tpool_wait() */
+#define	TP_SUSPEND	0x02		/* pool is being suspended */
+#define	TP_DESTROY	0x04		/* pool is being destroyed */
+#define	TP_ABANDON	0x08		/* pool is abandoned (auto-destroy) */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* _THREAD_POOL_IMPL_H */
diff --git a/lib/libuutil/Makefile.am b/lib/libuutil/Makefile.am
index 4d54485b7..c2659b619 100644
--- a/lib/libuutil/Makefile.am
+++ b/lib/libuutil/Makefile.am
@@ -28,8 +28,7 @@ nodist_libuutil_la_SOURCES = \
 
 libuutil_la_LIBADD = \
 	$(top_builddir)/lib/libavl/libavl.la \
-	$(top_builddir)/lib/libspl/libspl.la \
-	$(top_builddir)/lib/libefi/libefi.la
+	$(top_builddir)/lib/libspl/libspl.la
 
 libuutil_la_LDFLAGS = -pthread -version-info 1:1:0
 
diff --git a/lib/libzfs/Makefile.am b/lib/libzfs/Makefile.am
index f1260ea71..7ab8658e4 100644
--- a/lib/libzfs/Makefile.am
+++ b/lib/libzfs/Makefile.am
@@ -1,10 +1,16 @@
 include $(top_srcdir)/config/Rules.am
 
+VPATH = \
+	$(top_srcdir)/module/icp \
+	$(top_srcdir)/module/zcommon \
+	$(top_srcdir)/lib/libzfs
+
 libzfs_pcdir = $(datarootdir)/pkgconfig
 libzfs_pc_DATA = libzfs.pc libzfs_core.pc
 
 DEFAULT_INCLUDES += \
 	-I$(top_srcdir)/include \
+	-I$(top_srcdir)/module/icp/include \
 	-I$(top_srcdir)/lib/libspl/include
 
 lib_LTLIBRARIES = libzfs.la
@@ -23,17 +29,35 @@ USER_C = \
 	libzfs_status.c \
 	libzfs_util.c
 
-KERNEL_C =
+KERNEL_C = \
+	algs/sha2/sha2.c \
+	zfeature_common.c \
+	zfs_comutil.c \
+	zfs_deleg.c \
+	zfs_fletcher.c \
+	zfs_fletcher_aarch64_neon.c \
+	zfs_fletcher_avx512.c \
+	zfs_fletcher_intel.c \
+	zfs_fletcher_sse.c \
+	zfs_fletcher_superscalar.c \
+	zfs_fletcher_superscalar4.c \
+	zfs_namecheck.c \
+	zfs_prop.c \
+	zfs_uio.c \
+	zpool_prop.c \
+	zprop_common.c
 
 nodist_libzfs_la_SOURCES = \
 	$(USER_C) \
 	$(KERNEL_C)
 
 libzfs_la_LIBADD = \
-	$(top_builddir)/lib/libzfs_core/libzfs_core.la \
-	$(top_builddir)/lib/libshare/libshare.la \
+        $(top_builddir)/lib/libefi/libefi.la \
 	$(top_builddir)/lib/libnvpair/libnvpair.la \
-	$(top_builddir)/lib/libzpool/libzpool.la
+	$(top_builddir)/lib/libshare/libshare.la \
+	$(top_builddir)/lib/libtpool/libtpool.la \
+	$(top_builddir)/lib/libuutil/libuutil.la \
+	$(top_builddir)/lib/libzfs_core/libzfs_core.la
 
 libzfs_la_LIBADD += -lm $(LIBBLKID) $(LIBUDEV)
 libzfs_la_LDFLAGS = -version-info 2:0:0
diff --git a/lib/libzfs/libzfs_import.c b/lib/libzfs/libzfs_import.c
index f371d925b..833568356 100644
--- a/lib/libzfs/libzfs_import.c
+++ b/lib/libzfs/libzfs_import.c
@@ -60,6 +60,7 @@
 #include <sys/vtoc.h>
 #include <sys/dktp/fdisk.h>
 #include <sys/efi_partition.h>
+#include <thread_pool.h>
 #include <sys/vdev_impl.h>
 #include <blkid/blkid.h>
 #include "libzfs.h"
@@ -1388,7 +1389,7 @@ typedef struct rdsk_node {
 	nvlist_t *rn_config;		/* Label config */
 	avl_tree_t *rn_avl;
 	avl_node_t rn_node;
-	kmutex_t *rn_lock;
+	pthread_mutex_t *rn_lock;
 	boolean_t rn_labelpaths;
 } rdsk_node_t;
 
@@ -1603,14 +1604,14 @@ zpool_open_func(void *arg)
 			slice->rn_hdl = hdl;
 			slice->rn_order = IMPORT_ORDER_PREFERRED_1;
 			slice->rn_labelpaths = B_FALSE;
-			mutex_enter(rn->rn_lock);
+			pthread_mutex_lock(rn->rn_lock);
 			if (avl_find(rn->rn_avl, slice, &where)) {
-			mutex_exit(rn->rn_lock);
+			pthread_mutex_unlock(rn->rn_lock);
 				free(slice->rn_name);
 				free(slice);
 			} else {
 				avl_insert(rn->rn_avl, slice, where);
-				mutex_exit(rn->rn_lock);
+				pthread_mutex_unlock(rn->rn_lock);
 				zpool_open_func(slice);
 			}
 		}
@@ -1629,14 +1630,14 @@ zpool_open_func(void *arg)
 			slice->rn_hdl = hdl;
 			slice->rn_order = IMPORT_ORDER_PREFERRED_2;
 			slice->rn_labelpaths = B_FALSE;
-			mutex_enter(rn->rn_lock);
+			pthread_mutex_lock(rn->rn_lock);
 			if (avl_find(rn->rn_avl, slice, &where)) {
-				mutex_exit(rn->rn_lock);
+				pthread_mutex_unlock(rn->rn_lock);
 				free(slice->rn_name);
 				free(slice);
 			} else {
 				avl_insert(rn->rn_avl, slice, where);
-				mutex_exit(rn->rn_lock);
+				pthread_mutex_unlock(rn->rn_lock);
 				zpool_open_func(slice);
 			}
 		}
@@ -1679,7 +1680,7 @@ zpool_clear_label(int fd)
  * Scan a list of directories for zfs devices.
  */
 static int
-zpool_find_import_scan(libzfs_handle_t *hdl, kmutex_t *lock,
+zpool_find_import_scan(libzfs_handle_t *hdl, pthread_mutex_t *lock,
     avl_tree_t **slice_cache, char **dir, int dirs)
 {
 	avl_tree_t *cache;
@@ -1735,9 +1736,9 @@ zpool_find_import_scan(libzfs_handle_t *hdl, kmutex_t *lock,
 			slice->rn_hdl = hdl;
 			slice->rn_order = i + IMPORT_ORDER_SCAN_OFFSET;
 			slice->rn_labelpaths = B_FALSE;
-			mutex_enter(lock);
+			pthread_mutex_lock(lock);
 			avl_add(cache, slice);
-			mutex_exit(lock);
+			pthread_mutex_unlock(lock);
 		}
 
 		(void) closedir(dirp);
@@ -1761,7 +1762,7 @@ error:
  * Use libblkid to quickly enumerate all known zfs devices.
  */
 static int
-zpool_find_import_blkid(libzfs_handle_t *hdl, kmutex_t *lock,
+zpool_find_import_blkid(libzfs_handle_t *hdl, pthread_mutex_t *lock,
     avl_tree_t **slice_cache)
 {
 	rdsk_node_t *slice;
@@ -1815,14 +1816,14 @@ zpool_find_import_blkid(libzfs_handle_t *hdl, kmutex_t *lock,
 		else
 			slice->rn_order = IMPORT_ORDER_DEFAULT;
 
-		mutex_enter(lock);
+		pthread_mutex_lock(lock);
 		if (avl_find(*slice_cache, slice, &where)) {
 			free(slice->rn_name);
 			free(slice);
 		} else {
 			avl_insert(*slice_cache, slice, where);
 		}
-		mutex_exit(lock);
+		pthread_mutex_unlock(lock);
 	}
 
 	blkid_dev_iterate_end(iter);
@@ -1860,14 +1861,14 @@ zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
 	vdev_entry_t *ve, *venext;
 	config_entry_t *ce, *cenext;
 	name_entry_t *ne, *nenext;
-	kmutex_t lock;
+	pthread_mutex_t lock;
 	avl_tree_t *cache;
 	rdsk_node_t *slice;
 	void *cookie;
-	taskq_t *t;
+	tpool_t *t;
 
 	verify(iarg->poolname == NULL || iarg->guid == 0);
-	mutex_init(&lock, NULL, MUTEX_DEFAULT, NULL);
+	pthread_mutex_init(&lock, NULL);
 
 	/*
 	 * Locate pool member vdevs using libblkid or by directory scanning.
@@ -1896,15 +1897,13 @@ zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
 	 * validating labels, a large number of threads can be used due to
 	 * minimal contention.
 	 */
-	t = taskq_create("z_import", 2 * boot_ncpus, defclsyspri,
-	    2 * boot_ncpus, INT_MAX, TASKQ_PREPOPULATE);
-
+	t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN), 0, NULL);
 	for (slice = avl_first(cache); slice;
 	    (slice = avl_walk(cache, slice, AVL_AFTER)))
-		(void) taskq_dispatch(t, zpool_open_func, slice, TQ_SLEEP);
+		(void) tpool_dispatch(t, zpool_open_func, slice);
 
-	taskq_wait(t);
-	taskq_destroy(t);
+	tpool_wait(t);
+	tpool_destroy(t);
 
 	/*
 	 * Process the cache filtering out any entries which are not
@@ -1974,7 +1973,7 @@ zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
 	}
 	avl_destroy(cache);
 	free(cache);
-	mutex_destroy(&lock);
+	pthread_mutex_destroy(&lock);
 
 	ret = get_configs(hdl, &pools, iarg->can_be_active);
 
diff --git a/lib/libzfs/libzfs_pool.c b/lib/libzfs/libzfs_pool.c
index 73af75f9d..d3363809d 100644
--- a/lib/libzfs/libzfs_pool.c
+++ b/lib/libzfs/libzfs_pool.c
@@ -40,6 +40,7 @@
 #include <zone.h>
 #include <sys/stat.h>
 #include <sys/efi_partition.h>
+#include <sys/systeminfo.h>
 #include <sys/vtoc.h>
 #include <sys/zfs_ioctl.h>
 #include <dlfcn.h>
diff --git a/lib/libzfs/libzfs_status.c b/lib/libzfs/libzfs_status.c
index 05a9afce8..ccc472153 100644
--- a/lib/libzfs/libzfs_status.c
+++ b/lib/libzfs/libzfs_status.c
@@ -44,6 +44,7 @@
 #include <libzfs.h>
 #include <string.h>
 #include <unistd.h>
+#include <sys/systeminfo.h>
 #include "libzfs_impl.h"
 #include "zfeature_common.h"
 
diff --git a/lib/libzpool/Makefile.am b/lib/libzpool/Makefile.am
index a8c72c98d..aed8868ae 100644
--- a/lib/libzpool/Makefile.am
+++ b/lib/libzpool/Makefile.am
@@ -19,13 +19,14 @@ USER_C = \
 	util.c
 
 KERNEL_C = \
+	zfeature_common.c \
 	zfs_comutil.c \
 	zfs_deleg.c \
 	zfs_fletcher.c \
+	zfs_fletcher_aarch64_neon.c \
+	zfs_fletcher_avx512.c \
 	zfs_fletcher_intel.c \
 	zfs_fletcher_sse.c \
-	zfs_fletcher_avx512.c \
-	zfs_fletcher_aarch64_neon.c \
 	zfs_fletcher_superscalar.c \
 	zfs_fletcher_superscalar4.c \
 	zfs_namecheck.c \
@@ -115,13 +116,13 @@ KERNEL_C = \
 	zap_leaf.c \
 	zap_micro.c \
 	zfeature.c \
-	zfeature_common.c \
 	zfs_byteswap.c \
 	zfs_debug.c \
 	zfs_fm.c \
 	zfs_fuid.c \
 	zfs_sa.c \
 	zfs_znode.c \
+	zfs_ratelimit.c \
 	zfs_rlock.c \
 	zil.c \
 	zio.c \
@@ -136,12 +137,13 @@ nodist_libzpool_la_SOURCES = \
 	$(KERNEL_C)
 
 libzpool_la_LIBADD = \
-	$(top_builddir)/lib/libunicode/libunicode.la \
-	$(top_builddir)/lib/libuutil/libuutil.la \
+	$(top_builddir)/lib/libavl/libavl.la \
+	$(top_builddir)/lib/libicp/libicp.la \
 	$(top_builddir)/lib/libnvpair/libnvpair.la \
-	$(top_builddir)/lib/libicp/libicp.la
+	$(top_builddir)/lib/libspl/libspl.la \
+	$(top_builddir)/lib/libunicode/libunicode.la
 
 libzpool_la_LIBADD += $(ZLIB) -ldl
-libzpool_la_LDFLAGS = -version-info 2:0:0
+libzpool_la_LDFLAGS = -pthread -version-info 2:0:0
 
 EXTRA_DIST = $(USER_C)
diff --git a/lib/libzpool/kernel.c b/lib/libzpool/kernel.c
index e67d13c92..3fc7337e8 100644
--- a/lib/libzpool/kernel.c
+++ b/lib/libzpool/kernel.c
@@ -71,7 +71,7 @@ pthread_mutex_t kthread_lock = PTHREAD_MUTEX_INITIALIZER;
 pthread_key_t kthread_key;
 int kthread_nr = 0;
 
-void
+static void
 thread_init(void)
 {
 	kthread_t *kt;
@@ -90,7 +90,7 @@ thread_init(void)
 	kthread_nr = 1;
 }
 
-void
+static void
 thread_fini(void)
 {
 	kthread_t *kt = curthread;
@@ -1051,149 +1051,30 @@ delay(clock_t ticks)
 
 /*
  * Find highest one bit set.
- *	Returns bit number + 1 of highest bit that is set, otherwise returns 0.
- * High order bit is 31 (or 63 in _LP64 kernel).
+ * Returns bit number + 1 of highest bit that is set, otherwise returns 0.
+ * The __builtin_clzll() function is supported by both GCC and Clang.
  */
 int
 highbit64(uint64_t i)
 {
-	register int h = 1;
-
 	if (i == 0)
-		return (0);
-	if (i & 0xffffffff00000000ULL) {
-		h += 32; i >>= 32;
-	}
-	if (i & 0xffff0000) {
-		h += 16; i >>= 16;
-	}
-	if (i & 0xff00) {
-		h += 8; i >>= 8;
-	}
-	if (i & 0xf0) {
-		h += 4; i >>= 4;
-	}
-	if (i & 0xc) {
-		h += 2; i >>= 2;
-	}
-	if (i & 0x2) {
-		h += 1;
-	}
-	return (h);
+	return (0);
+
+	return (NBBY * sizeof (uint64_t) - __builtin_clzll(i));
 }
 
 /*
  * Find lowest one bit set.
  * Returns bit number + 1 of lowest bit that is set, otherwise returns 0.
- * This is basically a reimplementation of ffsll(), which is GNU specific.
+ * The __builtin_ffsll() function is supported by both GCC and Clang.
  */
 int
 lowbit64(uint64_t i)
 {
-	register int h = 64;
 	if (i == 0)
 		return (0);
 
-	if (i & 0x00000000ffffffffULL)
-		h -= 32;
-	else
-		i >>= 32;
-
-	if (i & 0x0000ffff)
-		h -= 16;
-	else
-		i >>= 16;
-
-	if (i & 0x00ff)
-		h -= 8;
-	else
-		i >>= 8;
-
-	if (i & 0x0f)
-		h -= 4;
-	else
-		i >>= 4;
-
-	if (i & 0x3)
-		h -= 2;
-	else
-		i >>= 2;
-
-	if (i & 0x1)
-		h -= 1;
-
-	return (h);
-}
-
-/*
- * Find highest one bit set.
- * Returns bit number + 1 of highest bit that is set, otherwise returns 0.
- * High order bit is 31 (or 63 in _LP64 kernel).
- */
-int
-highbit(ulong_t i)
-{
-register int h = 1;
-
-	if (i == 0)
-		return (0);
-#ifdef _LP64
-	if (i & 0xffffffff00000000ul) {
-		h += 32; i >>= 32;
-	}
-#endif
-	if (i & 0xffff0000) {
-		h += 16; i >>= 16;
-	}
-	if (i & 0xff00) {
-		h += 8; i >>= 8;
-	}
-	if (i & 0xf0) {
-		h += 4; i >>= 4;
-	}
-	if (i & 0xc) {
-		h += 2; i >>= 2;
-	}
-	if (i & 0x2) {
-		h += 1;
-	}
-	return (h);
-}
-
-/*
- * Find lowest one bit set.
- *     Returns bit number + 1 of lowest bit that is set, otherwise returns 0.
- * Low order bit is 0.
- */
-int
-lowbit(ulong_t i)
-{
-	register int h = 1;
-
-	if (i == 0)
-		return (0);
-
-#ifdef _LP64
-	if (!(i & 0xffffffff)) {
-		h += 32; i >>= 32;
-	}
-#endif
-	if (!(i & 0xffff)) {
-		h += 16; i >>= 16;
-	}
-	if (!(i & 0xff)) {
-		h += 8; i >>= 8;
-	}
-	if (!(i & 0xf)) {
-		h += 4; i >>= 4;
-	}
-	if (!(i & 0x3)) {
-		h += 2; i >>= 2;
-	}
-	if (!(i & 0x1)) {
-		h += 1;
-	}
-	return (h);
+	return (__builtin_ffsll(i));
 }
 
 static int random_fd = -1, urandom_fd = -1;
@@ -1288,64 +1169,6 @@ umem_out_of_memory(void)
 	return (0);
 }
 
-#define	HOSTID_MASK 0xffffffff
-
-static unsigned long
-get_spl_hostid(void)
-{
-	FILE *f;
-	unsigned long hostid;
-	char *env;
-
-	/*
-	 * Allow the hostid to be subverted for testing.
-	 */
-	env = getenv("ZFS_HOSTID");
-	if (env) {
-		hostid = strtoull(env, NULL, 0);
-		return (hostid & HOSTID_MASK);
-	}
-
-	f = fopen("/sys/module/spl/parameters/spl_hostid", "r");
-	if (!f)
-		return (0);
-
-	if (fscanf(f, "%lu", &hostid) != 1)
-		hostid = 0;
-
-	fclose(f);
-
-	return (hostid & HOSTID_MASK);
-}
-
-unsigned long
-get_system_hostid(void)
-{
-	unsigned long system_hostid = get_spl_hostid();
-	/*
-	 * We do not use the library call gethostid() because
-	 * it generates a hostid value that the kernel is
-	 * unaware of, if the spl_hostid module parameter has not
-	 * been set and there is no system hostid file (e.g.
-	 * /etc/hostid).  The kernel and userspace must agree.
-	 * See comments above hostid_read() in the SPL.
-	 */
-	if (system_hostid == 0) {
-		int fd, rc;
-		unsigned long hostid;
-		int hostid_size = 4;  /* 4 bytes regardless of arch */
-
-		fd = open("/etc/hostid", O_RDONLY);
-		if (fd >= 0) {
-			rc = read(fd, &hostid, hostid_size);
-			if (rc > 0)
-				system_hostid = (hostid & HOSTID_MASK);
-			close(fd);
-		}
-	}
-	return (system_hostid);
-}
-
 void
 kernel_init(int mode)
 {