Add TRIM support for disk vdevs.

This adds TRIM (a.k.a UNMAP, DISCARD, hole punching) support for disk
vdevs. The original patch is from Pawel Jakub Dawidek
<[email protected]> who wrote it for FreeBSD. Etienne Dechamps
<[email protected]> ported it to ZFS On Linux.

The code builds a map of regions that were freed. On every write the
code consults the map and eventually removes ranges that were freed
before, but are now overwritten.

Freed blocks are not TRIMed immediately. There is a tunable that defines
how many txg we should wait with TRIMming freed blocks (64 by default).

There is a low priority thread that TRIMs ranges when the time comes.
During TRIM we keep in-flight ranges on a list to detect colliding
writes - we have to delay writes that collide with in-flight TRIMs in
case something will be reordered and write will reached the disk before
the TRIM. We don't have to do the same for in-flight writes, as
colliding writes just remove ranges to TRIM.

Most of the code stayed unchanged during the porting to Linux. The only
big change is in the vdev disk module, since the FreeBSD and Linux
interfaces for issuing discards to block devices is obviously different.
On FreeBSD it seems that issuing a DELETE request of any size is
sufficient; on Linux we have to be careful not to exceed maximum discard
limits. That's why we introduce a new vdev_disk_io_trim() function
inspired from the Linux blkdev_issue_discard() function and the
pre-existing vdev_disk_physio() function. The new function takes care of
splitting discard requests into smaller ones if necessary.

In theory, the code should work for main pool disk vdevs, slog disk
vdevs, L2ARC disk vdevs, and supports mirror and raidz. File vdevs are
not supported yet.

Note that the new feature is disabled by default (zfs_notrim=1). To use
it, you have to explictly set the module parameter "zfs_notrim" to "0".
Be aware that this code is largely untested and brings huge risks of
potential data corruption. Use at your own risk and expect data loss.

include/sys/spa_impl.h

@@ -217,6 +217,9 @@ struct spa {
 	spa_proc_state_t spa_proc_state;	/* see definition */
 	proc_t		*spa_proc;		/* "zpool-poolname" process */
 	uint64_t	spa_did;		/* if procp != p0, did of t1 */
+	kthread_t	*spa_trim_thread;	/* thread sending TRIM I/Os */
+	kmutex_t	spa_trim_lock;		/* protects spa_trim_cv */
+	kcondvar_t	spa_trim_cv;		/* used to notify TRIM thread */
 	boolean_t	spa_autoreplace;	/* autoreplace set in open */
 	int		spa_vdev_locks;		/* locks grabbed */
 	uint64_t	spa_creation_version;	/* version at pool creation */

include/sys/trim_map.h

@@ -0,0 +1,51 @@
+/*
+ * 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) 2012 Pawel Jakub Dawidek <[email protected]>.
+ * All rights reserved.
+ */
+
+#ifndef _SYS_TRIM_MAP_H
+#define	_SYS_TRIM_MAP_H
+
+#include <sys/avl.h>
+#include <sys/list.h>
+#include <sys/spa.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+extern void trim_map_create(vdev_t *vd);
+extern void trim_map_destroy(vdev_t *vd);
+extern void trim_map_free(zio_t *zio);
+extern boolean_t trim_map_write_start(zio_t *zio);
+extern void trim_map_write_done(zio_t *zio);
+
+extern void trim_thread_create(spa_t *spa);
+extern void trim_thread_destroy(spa_t *spa);
+extern void trim_thread_wakeup(spa_t *spa);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_TRIM_MAP_H */

include/sys/vdev.h

@@ -45,6 +45,7 @@ typedef enum vdev_dtl_type {
 } vdev_dtl_type_t;
 
 extern int zfs_nocacheflush;
+extern int zfs_notrim;
 
 extern int vdev_open(vdev_t *);
 extern void vdev_open_children(vdev_t *);

include/sys/vdev_impl.h

@@ -186,6 +186,7 @@ struct vdev {
 	uint64_t	vdev_unspare;	/* unspare when resilvering done */
 	hrtime_t	vdev_last_try;	/* last reopen time		*/
 	boolean_t	vdev_nowritecache; /* true if flushwritecache failed */
+	boolean_t	vdev_notrim;	/* true if trim failed 		*/
 	boolean_t	vdev_checkremove; /* temporary online test	*/
 	boolean_t	vdev_forcefault; /* force online fault		*/
 	boolean_t	vdev_splitting;	/* split or repair in progress  */
@@ -201,6 +202,7 @@ struct vdev {
 	spa_aux_vdev_t	*vdev_aux;	/* for l2cache vdevs		*/
 	zio_t		*vdev_probe_zio; /* root of current probe	*/
 	vdev_aux_t	vdev_label_aux;	/* on-disk aux state		*/
+	struct trim_map *vdev_trimmap;
 
 	/*
 	 * For DTrace to work in userland (libzpool) context, these fields must

include/sys/zio.h

@@ -140,7 +140,8 @@ enum zio_compress {
 #define	ZIO_PRIORITY_RESILVER		(zio_priority_table[9])
 #define	ZIO_PRIORITY_SCRUB		(zio_priority_table[10])
 #define	ZIO_PRIORITY_DDT_PREFETCH	(zio_priority_table[11])
-#define	ZIO_PRIORITY_TABLE_SIZE		12
+#define	ZIO_PRIORITY_TRIM		(zio_priority_table[12])
+#define	ZIO_PRIORITY_TABLE_SIZE		13
 
 #define	ZIO_PIPELINE_CONTINUE		0x100
 #define	ZIO_PIPELINE_STOP		0x101
@@ -429,6 +430,9 @@ struct zio {
 
 	/* Taskq dispatching state */
 	taskq_ent_t	io_tqent;
+
+	avl_node_t	io_trim_node;
+	list_node_t	io_trim_link;
 };
 
 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
@@ -459,7 +463,8 @@ extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
     zio_done_func_t *done, void *private, enum zio_flag flags);
 
 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
-    zio_done_func_t *done, void *private, int priority, enum zio_flag flags);
+    uint64_t offset, uint64_t size, zio_done_func_t *done, 
+    void *private, int priority, enum zio_flag flags);
 
 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
     uint64_t size, void *data, int checksum,
@@ -478,6 +483,7 @@ extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
     blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
 extern void zio_flush(zio_t *zio, vdev_t *vd);
+extern void zio_trim(zio_t *zio, vdev_t *vd, uint64_t offset, uint64_t size);
 extern void zio_shrink(zio_t *zio, uint64_t size);
 
 extern int zio_wait(zio_t *zio);

include/sys/zio_impl.h

@@ -60,9 +60,9 @@ enum zio_stage {
 
 	ZIO_STAGE_READY			= 1 << 15,	/* RWFCI */
 
-	ZIO_STAGE_VDEV_IO_START		= 1 << 16,	/* RW--I */
-	ZIO_STAGE_VDEV_IO_DONE		= 1 << 17,	/* RW--I */
-	ZIO_STAGE_VDEV_IO_ASSESS	= 1 << 18,	/* RW--I */
+	ZIO_STAGE_VDEV_IO_START		= 1 << 16,	/* RWF-I */
+	ZIO_STAGE_VDEV_IO_DONE		= 1 << 17,	/* RWF-- */
+	ZIO_STAGE_VDEV_IO_ASSESS	= 1 << 18,	/* RWF-I */
 
 	ZIO_STAGE_CHECKSUM_VERIFY	= 1 << 19,	/* R---- */
 
@@ -143,7 +143,9 @@ enum zio_stage {
 #define	ZIO_FREE_PIPELINE			\
 	(ZIO_INTERLOCK_STAGES |			\
 	ZIO_STAGE_FREE_BP_INIT |		\
-	ZIO_STAGE_DVA_FREE)
+	ZIO_STAGE_DVA_FREE |			\
+	ZIO_STAGE_VDEV_IO_START |		\
+	ZIO_STAGE_VDEV_IO_ASSESS)
 
 #define	ZIO_DDT_FREE_PIPELINE			\
 	(ZIO_INTERLOCK_STAGES |			\

lib/libspl/include/sys/dkio.h

@@ -188,6 +188,8 @@ struct dk_geom {
  */
 #define	DKIOCFLUSHWRITECACHE	(DKIOC|34)	/* flush cache to phys medium */
 
+#define DKIOCTRIM		(DKIOC|35)	/* TRIM a block */
+
 struct dk_callback {
 	void (*dkc_callback)(void *dkc_cookie, int error);
 	void *dkc_cookie;

lib/libzpool/Makefile.am

@@ -59,6 +59,7 @@ libzpool_la_SOURCES = \
 	$(top_srcdir)/module/zfs/spa_history.c \
 	$(top_srcdir)/module/zfs/spa_misc.c \
 	$(top_srcdir)/module/zfs/space_map.c \
+	$(top_srcdir)/module/zfs/trim_map.c \
 	$(top_srcdir)/module/zfs/txg.c \
 	$(top_srcdir)/module/zfs/uberblock.c \
 	$(top_srcdir)/module/zfs/unique.c \

module/avl/avl.c

@@ -1052,6 +1052,7 @@ EXPORT_SYMBOL(avl_nearest);
 EXPORT_SYMBOL(avl_add);
 EXPORT_SYMBOL(avl_remove);
 EXPORT_SYMBOL(avl_numnodes);
+EXPORT_SYMBOL(avl_is_empty);
 EXPORT_SYMBOL(avl_destroy_nodes);
 EXPORT_SYMBOL(avl_destroy);
 #endif

module/zfs/Makefile.in

@@ -43,6 +43,7 @@ $(MODULE)-objs += @top_srcdir@/module/zfs/spa_errlog.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/spa_history.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/spa_misc.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/space_map.o
+$(MODULE)-objs += @top_srcdir@/module/zfs/trim_map.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/txg.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/uberblock.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/unique.o

module/zfs/spa.c

@@ -63,6 +63,7 @@
 #include <sys/spa_boot.h>
 #include <sys/zfs_ioctl.h>
 #include <sys/dsl_scan.h>
+#include <sys/trim_map.h>
 
 #ifdef	_KERNEL
 #include <sys/bootprops.h>
@@ -854,6 +855,11 @@ spa_activate(spa_t *spa, int mode)
 		spa_create_zio_taskqs(spa);
 	}
 
+	/*
+	 * Start TRIM thread.
+	 */
+	trim_thread_create(spa);
+
 	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
 	    offsetof(vdev_t, vdev_config_dirty_node));
 	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
@@ -884,6 +890,12 @@ spa_deactivate(spa_t *spa)
 	ASSERT(spa->spa_async_zio_root == NULL);
 	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
 
+	/*
+	 * Stop TRIM thread in case spa_unload() wasn't called before
+	 * spa_deactivate().
+	 */
+	trim_thread_destroy(spa);
+
 	txg_list_destroy(&spa->spa_vdev_txg_list);
 
 	list_destroy(&spa->spa_config_dirty_list);
@@ -998,6 +1010,11 @@ spa_unload(spa_t *spa)
 
 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
 
+	/*
+	 * Stop TRIM thread.
+	 */
+	trim_thread_destroy(spa);
+
 	/*
 	 * Stop async tasks.
 	 */