OpenZFS 9079 - race condition in starting and ending condensing thread for indirect vdevs

The timeline of the race condition is the following:

[1] Thread A is about to finish condesing the first vdev in
    spa_condense_indirect_thread(), so it calls the
    spa_condense_indirect_complete_sync() sync task which sets
    the spa_condensing_indirect field to NULL. Waiting for the
    sync task to finish, thread A sleeps until the txg is done.
    When this happens, thread A will acquire spa_async_lock and
    set spa_condense_thread to NULL.

[2] While thread A waits for the txg to finish, thread B which is
    running spa_sync() checks whether it should condense the
    second vdev in vdev_indirect_should_condense() by checking the
    spa_condensing_indirect field which was set to NULL by
    spa_condense_indirect_thread() from thread A. So it goes on
    and tries to spawn a new condensing thread in
    spa_condense_indirect_start_sync() and the aforementioned
    assertions fails because thread A has not set spa_condense_thread
    to NULL (which is basically the last thing it does before returning).

The main issue here is that we rely on both spa_condensing_indirect
and spa_condense_thread to signify whether a condensing thread is
running. Ideally we would only use one throughout the codebase. In
addition, for managing spa_condense_thread we currently use
spa_async_lock which basically tights condensing to scrubing when
it comes to pausing and resuming those actions during spa export.

This commit introduces the ZTHR infrastructure, which is basically
threads created during spa_load()/spa_create() and exist until we
export or destroy the pool. ZTHRs sleep the majority of the time,
until they are notified to wake up and do some predefined type of work.

In the context of the current bug, a zthr to does the condensing of
indirect mappings replacing the older code that used bare kthreads.
When a pool is created, the condensing zthr is spawned but sleeps
right away, until it is awaken by a signal from spa_sync(). If an
existing pool is loaded, the condensing zthr looks if there is
anything to condense before going to sleep, in case we were condensing
mappings in the pool before it got exported.

The benefits of this solution are the following:
- The current bug is fixed
- spa_condensing_indirect is the sole indicator of whether we are
  currently condensing or not
- condensing is more decoupled from the spa_async_thread related
  functionality.

As a final note, this commit also sets up the path on upstreaming
other features that use the ZTHR code like zpool checkpoint and
fast clone deletion.

Authored by: Serapheim Dimitropoulos <[email protected]>
Reviewed by: Matt Ahrens <[email protected]>
Reviewed by: Pavel Zakharov <[email protected]>
Approved by: Hans Rosenfeld <[email protected]>
Ported-by: Tim Chase <[email protected]>

OpenZFS-issue: https://illumos.org/issues/9079
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3dc606ee
Closes #6900

4 files changed, 57 insertions, 3 deletions

diff --git a/include/sys/Makefile.am b/include/sys/Makefile.am
index f30c9427e..e73be52f3 100644
--- a/include/sys/Makefile.am
+++ b/include/sys/Makefile.am
@@ -125,7 +125,8 @@ COMMON_H = \
 	$(top_srcdir)/include/sys/zio.h \
 	$(top_srcdir)/include/sys/zio_impl.h \
 	$(top_srcdir)/include/sys/zio_priority.h \
-	$(top_srcdir)/include/sys/zrlock.h
+	$(top_srcdir)/include/sys/zrlock.h \
+	$(top_srcdir)/include/sys/zthr.h
 
 KERNEL_H = \
 	$(top_srcdir)/include/sys/zfs_ioctl.h \
diff --git a/include/sys/spa_impl.h b/include/sys/spa_impl.h
index 1741eb9e5..f49138d0f 100644
--- a/include/sys/spa_impl.h
+++ b/include/sys/spa_impl.h
@@ -45,6 +45,7 @@
 #include <sys/bpobj.h>
 #include <sys/dsl_crypt.h>
 #include <sys/zfeature.h>
+#include <sys/zthr.h>
 #include <zfeature_common.h>
 
 #ifdef	__cplusplus
@@ -268,7 +269,7 @@ struct spa {
 
 	spa_condensing_indirect_phys_t	spa_condensing_indirect_phys;
 	spa_condensing_indirect_t	*spa_condensing_indirect;
-	kthread_t	*spa_condense_thread;	/* thread doing condense. */
+	zthr_t		*spa_condense_zthr;	/* zthr doing condense. */
 
 	char		*spa_root;		/* alternate root directory */
 	uint64_t	spa_ena;		/* spa-wide ereport ENA */
diff --git a/include/sys/vdev_removal.h b/include/sys/vdev_removal.h
index 80ee09456..fb6c8c0ce 100644
--- a/include/sys/vdev_removal.h
+++ b/include/sys/vdev_removal.h
@@ -76,7 +76,7 @@ extern int spa_remove_init(spa_t *);
 extern void spa_restart_removal(spa_t *);
 extern int spa_condense_init(spa_t *);
 extern void spa_condense_fini(spa_t *);
-extern void spa_condense_indirect_restart(spa_t *);
+extern void spa_start_indirect_condensing_thread(spa_t *);
 extern void spa_vdev_condense_suspend(spa_t *);
 extern int spa_vdev_remove(spa_t *, uint64_t, boolean_t);
 extern void free_from_removing_vdev(vdev_t *, uint64_t, uint64_t, uint64_t);
diff --git a/include/sys/zthr.h b/include/sys/zthr.h
new file mode 100644
index 000000000..6bfb6b6c0
--- /dev/null
+++ b/include/sys/zthr.h
@@ -0,0 +1,52 @@
+/*
+ * CDDL HEADER START
+ *
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source. A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ *
+ * CDDL HEADER END
+ */
+
+
+/*
+ * Copyright (c) 2017 by Delphix. All rights reserved.
+ */
+
+#ifndef _SYS_ZTHR_H
+#define	_SYS_ZTHR_H
+
+typedef struct zthr zthr_t;
+typedef int (zthr_func_t)(void *, zthr_t *);
+typedef boolean_t (zthr_checkfunc_t)(void *, zthr_t *);
+
+struct zthr {
+	kthread_t	*zthr_thread;
+	kmutex_t	zthr_lock;
+	kcondvar_t	zthr_cv;
+	boolean_t	zthr_cancel;
+
+	zthr_checkfunc_t	*zthr_checkfunc;
+	zthr_func_t	*zthr_func;
+	void		*zthr_arg;
+	int		zthr_rc;
+};
+
+extern zthr_t *zthr_create(zthr_checkfunc_t checkfunc,
+    zthr_func_t *func, void *arg);
+extern void zthr_exit(zthr_t *t, int rc);
+extern void zthr_destroy(zthr_t *t);
+
+extern void zthr_wakeup(zthr_t *t);
+extern int zthr_cancel(zthr_t *t);
+extern void zthr_resume(zthr_t *t);
+
+extern boolean_t zthr_iscancelled(zthr_t *t);
+extern boolean_t zthr_isrunning(zthr_t *t);
+
+#endif /* _SYS_ZTHR_H */