Rebase master to b105

10 files changed, 68 insertions, 31 deletions

diff --git a/module/zfs/include/sys/spa.h b/module/zfs/include/sys/spa.h
index 24b3ca447..519b1d0c0 100644
--- a/module/zfs/include/sys/spa.h
+++ b/module/zfs/include/sys/spa.h
@@ -332,7 +332,8 @@ extern int spa_import(const char *pool, nvlist_t *config, nvlist_t *props);
 extern int spa_import_faulted(const char *, nvlist_t *, nvlist_t *);
 extern nvlist_t *spa_tryimport(nvlist_t *tryconfig);
 extern int spa_destroy(char *pool);
-extern int spa_export(char *pool, nvlist_t **oldconfig, boolean_t force);
+extern int spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
+    boolean_t hardforce);
 extern int spa_reset(char *pool);
 extern void spa_async_request(spa_t *spa, int flag);
 extern void spa_async_unrequest(spa_t *spa, int flag);
@@ -351,7 +352,8 @@ extern void spa_inject_delref(spa_t *spa);
 extern int spa_vdev_add(spa_t *spa, nvlist_t *nvroot);
 extern int spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot,
     int replacing);
-extern int spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done);
+extern int spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid,
+    int replace_done);
 extern int spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare);
 extern int spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath);
 
@@ -475,6 +477,8 @@ extern boolean_t spa_has_spare(spa_t *, uint64_t guid);
 extern uint64_t bp_get_dasize(spa_t *spa, const blkptr_t *bp);
 extern boolean_t spa_has_slogs(spa_t *spa);
 extern boolean_t spa_is_root(spa_t *spa);
+extern boolean_t spa_writeable(spa_t *spa);
+extern int spa_mode(spa_t *spa);
 
 /* history logging */
 typedef enum history_log_type {
@@ -545,7 +549,7 @@ _NOTE(CONSTCOND) } while (0)
 #define	dprintf_bp(bp, fmt, ...)
 #endif
 
-extern int spa_mode;			/* mode, e.g. FREAD | FWRITE */
+extern int spa_mode_global;			/* mode, e.g. FREAD | FWRITE */
 
 #ifdef	__cplusplus
 }
diff --git a/module/zfs/include/sys/spa_impl.h b/module/zfs/include/sys/spa_impl.h
index 8aeb414fe..588b4f5a9 100644
--- a/module/zfs/include/sys/spa_impl.h
+++ b/module/zfs/include/sys/spa_impl.h
@@ -170,6 +170,7 @@ struct spa {
 	boolean_t	spa_import_faulted;	/* allow faulted vdevs */
 	boolean_t	spa_is_root;		/* pool is root */
 	int		spa_minref;		/* num refs when first opened */
+	int		spa_mode;		/* FREAD | FWRITE */
 	spa_log_state_t spa_log_state;		/* log state */
 	/*
 	 * spa_refcnt & spa_config_lock must be the last elements
diff --git a/module/zfs/include/sys/space_map.h b/module/zfs/include/sys/space_map.h
index db9daef1f..8d7860660 100644
--- a/module/zfs/include/sys/space_map.h
+++ b/module/zfs/include/sys/space_map.h
@@ -19,15 +19,13 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
 #ifndef _SYS_SPACE_MAP_H
 #define	_SYS_SPACE_MAP_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/avl.h>
 #include <sys/dmu.h>
 
@@ -58,6 +56,12 @@ typedef struct space_seg {
 	uint64_t	ss_end;		/* ending offset (non-inclusive) */
 } space_seg_t;
 
+typedef struct space_ref {
+	avl_node_t	sr_node;	/* AVL node */
+	uint64_t	sr_offset;	/* offset (start or end) */
+	int64_t		sr_refcnt;	/* associated reference count */
+} space_ref_t;
+
 typedef struct space_map_obj {
 	uint64_t	smo_object;	/* on-disk space map object */
 	uint64_t	smo_objsize;	/* size of the object */
@@ -133,13 +137,12 @@ extern void space_map_create(space_map_t *sm, uint64_t start, uint64_t size,
 extern void space_map_destroy(space_map_t *sm);
 extern void space_map_add(space_map_t *sm, uint64_t start, uint64_t size);
 extern void space_map_remove(space_map_t *sm, uint64_t start, uint64_t size);
-extern int space_map_contains(space_map_t *sm, uint64_t start, uint64_t size);
+extern boolean_t space_map_contains(space_map_t *sm,
+    uint64_t start, uint64_t size);
 extern void space_map_vacate(space_map_t *sm,
     space_map_func_t *func, space_map_t *mdest);
 extern void space_map_walk(space_map_t *sm,
     space_map_func_t *func, space_map_t *mdest);
-extern void space_map_excise(space_map_t *sm, uint64_t start, uint64_t size);
-extern void space_map_union(space_map_t *smd, space_map_t *sms);
 
 extern void space_map_load_wait(space_map_t *sm);
 extern int space_map_load(space_map_t *sm, space_map_ops_t *ops,
@@ -155,6 +158,15 @@ extern void space_map_sync(space_map_t *sm, uint8_t maptype,
 extern void space_map_truncate(space_map_obj_t *smo,
     objset_t *os, dmu_tx_t *tx);
 
+extern void space_map_ref_create(avl_tree_t *t);
+extern void space_map_ref_destroy(avl_tree_t *t);
+extern void space_map_ref_add_seg(avl_tree_t *t,
+    uint64_t start, uint64_t end, int64_t refcnt);
+extern void space_map_ref_add_map(avl_tree_t *t,
+    space_map_t *sm, int64_t refcnt);
+extern void space_map_ref_generate_map(avl_tree_t *t,
+    space_map_t *sm, int64_t minref);
+
 #ifdef	__cplusplus
 }
 #endif
diff --git a/module/zfs/include/sys/uberblock_impl.h b/module/zfs/include/sys/uberblock_impl.h
index 55a0dd5ae..b49df8ae0 100644
--- a/module/zfs/include/sys/uberblock_impl.h
+++ b/module/zfs/include/sys/uberblock_impl.h
@@ -19,15 +19,13 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
 #ifndef _SYS_UBERBLOCK_IMPL_H
 #define	_SYS_UBERBLOCK_IMPL_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/uberblock.h>
 
 #ifdef	__cplusplus
@@ -35,6 +33,11 @@ extern "C" {
 #endif
 
 /*
+ * For zdb use and debugging purposes only
+ */
+extern uint64_t ub_max_txg;
+
+/*
  * The uberblock version is incremented whenever an incompatible on-disk
  * format change is made to the SPA, DMU, or ZAP.
  *
diff --git a/module/zfs/include/sys/vdev.h b/module/zfs/include/sys/vdev.h
index c070d6f3d..b8313a920 100644
--- a/module/zfs/include/sys/vdev.h
+++ b/module/zfs/include/sys/vdev.h
@@ -36,6 +36,14 @@
 extern "C" {
 #endif
 
+typedef enum vdev_dtl_type {
+	DTL_MISSING,	/* 0% replication: no copies of the data */
+	DTL_PARTIAL,	/* less than 100% replication: some copies missing */
+	DTL_SCRUB,	/* unable to fully repair during scrub/resilver */
+	DTL_OUTAGE,	/* temporarily missing (used to attempt detach) */
+	DTL_TYPES
+} vdev_dtl_type_t;
+
 extern boolean_t zfs_nocacheflush;
 
 extern int vdev_open(vdev_t *);
@@ -50,10 +58,14 @@ extern zio_t *vdev_probe(vdev_t *vd, zio_t *pio);
 extern boolean_t vdev_is_bootable(vdev_t *vd);
 extern vdev_t *vdev_lookup_top(spa_t *spa, uint64_t vdev);
 extern vdev_t *vdev_lookup_by_guid(vdev_t *vd, uint64_t guid);
-extern void vdev_dtl_dirty(space_map_t *sm, uint64_t txg, uint64_t size);
-extern int vdev_dtl_contains(space_map_t *sm, uint64_t txg, uint64_t size);
+extern void vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t d,
+    uint64_t txg, uint64_t size);
+extern boolean_t vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t d,
+    uint64_t txg, uint64_t size);
+extern boolean_t vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t d);
 extern void vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg,
     int scrub_done);
+extern boolean_t vdev_dtl_required(vdev_t *vd);
 extern boolean_t vdev_resilver_needed(vdev_t *vd,
     uint64_t *minp, uint64_t *maxp);
 
diff --git a/module/zfs/include/sys/vdev_impl.h b/module/zfs/include/sys/vdev_impl.h
index 26904d089..652349341 100644
--- a/module/zfs/include/sys/vdev_impl.h
+++ b/module/zfs/include/sys/vdev_impl.h
@@ -123,8 +123,7 @@ struct vdev {
 	vdev_t		*vdev_parent;	/* parent vdev			*/
 	vdev_t		**vdev_child;	/* array of children		*/
 	uint64_t	vdev_children;	/* number of children		*/
-	space_map_t	vdev_dtl_map;	/* dirty time log in-core state	*/
-	space_map_t	vdev_dtl_scrub;	/* DTL for scrub repair writes	*/
+	space_map_t	vdev_dtl[DTL_TYPES]; /* in-core dirty time logs	*/
 	vdev_stat_t	vdev_stat;	/* virtual device statistics	*/
 
 	/*
@@ -149,7 +148,7 @@ struct vdev {
 	 * Leaf vdev state.
 	 */
 	uint64_t	vdev_psize;	/* physical device capacity	*/
-	space_map_obj_t	vdev_dtl;	/* dirty time log on-disk state	*/
+	space_map_obj_t	vdev_dtl_smo;	/* dirty time log space map obj	*/
 	txg_node_t	vdev_dtl_node;	/* per-txg dirty DTL linkage	*/
 	uint64_t	vdev_wholedisk;	/* true if this is a whole disk */
 	uint64_t	vdev_offline;	/* persistent offline state	*/
diff --git a/module/zfs/include/sys/zfs_vfsops.h b/module/zfs/include/sys/zfs_vfsops.h
index 87b75e6e7..7e0440be4 100644
--- a/module/zfs/include/sys/zfs_vfsops.h
+++ b/module/zfs/include/sys/zfs_vfsops.h
@@ -26,8 +26,6 @@
 #ifndef	_SYS_FS_ZFS_VFSOPS_H
 #define	_SYS_FS_ZFS_VFSOPS_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/isa_defs.h>
 #include <sys/types32.h>
 #include <sys/list.h>
@@ -49,7 +47,6 @@ struct zfsvfs {
 	uint64_t	z_root;		/* id of root znode */
 	uint64_t	z_unlinkedobj;	/* id of unlinked zapobj */
 	uint64_t	z_max_blksz;	/* maximum block size for files */
-	uint64_t	z_assign;	/* TXG_NOWAIT or set by zil_replay() */
 	uint64_t	z_fuid_obj;	/* fuid table object number */
 	uint64_t	z_fuid_size;	/* fuid table size */
 	avl_tree_t	z_fuid_idx;	/* fuid tree keyed by index */
@@ -74,6 +71,7 @@ struct zfsvfs {
 	boolean_t	z_issnap;	/* true if this is a snapshot */
 	boolean_t	z_vscan;	/* virus scan on/off */
 	boolean_t	z_use_fuids;	/* version allows fuids */
+	boolean_t	z_replay;	/* set during ZIL replay */
 	kmutex_t	z_online_recv_lock; /* recv in prog grabs as WRITER */
 	uint64_t	z_version;	/* ZPL version */
 #define	ZFS_OBJ_MTX_SZ	64
diff --git a/module/zfs/include/sys/zil.h b/module/zfs/include/sys/zil.h
index 4d02d14f7..b69323cfa 100644
--- a/module/zfs/include/sys/zil.h
+++ b/module/zfs/include/sys/zil.h
@@ -335,7 +335,6 @@ typedef void zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg,
 typedef void zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg,
     uint64_t txg);
 typedef int zil_replay_func_t();
-typedef void zil_replay_cleaner_t();
 typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio);
 
 extern uint64_t	zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
@@ -350,9 +349,8 @@ extern void	zil_free(zilog_t *zilog);
 extern zilog_t	*zil_open(objset_t *os, zil_get_data_t *get_data);
 extern void	zil_close(zilog_t *zilog);
 
-extern void	zil_replay(objset_t *os, void *arg, uint64_t *txgp,
-    zil_replay_func_t *replay_func[TX_MAX_TYPE],
-    zil_replay_cleaner_t *replay_cleaner);
+extern void	zil_replay(objset_t *os, void *arg,
+    zil_replay_func_t *replay_func[TX_MAX_TYPE]);
 extern void	zil_destroy(zilog_t *zilog, boolean_t keep_first);
 extern void	zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx);
 
diff --git a/module/zfs/include/sys/zil_impl.h b/module/zfs/include/sys/zil_impl.h
index 0fc800b96..3f2582931 100644
--- a/module/zfs/include/sys/zil_impl.h
+++ b/module/zfs/include/sys/zil_impl.h
@@ -19,15 +19,13 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
 #ifndef	_SYS_ZIL_IMPL_H
 #define	_SYS_ZIL_IMPL_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/zil.h>
 #include <sys/dmu_objset.h>
 
@@ -74,13 +72,14 @@ struct zilog {
 	uint64_t	zl_commit_seq;	/* committed upto this number */
 	uint64_t	zl_lr_seq;	/* log record sequence number */
 	uint64_t	zl_destroy_txg;	/* txg of last zil_destroy() */
-	uint64_t	zl_replay_seq[TXG_SIZE]; /* seq of last replayed rec */
+	uint64_t	zl_replayed_seq[TXG_SIZE]; /* last replayed rec seq */
+	uint64_t	zl_replaying_seq; /* current replay seq number */
 	uint32_t	zl_suspend;	/* log suspend count */
 	kcondvar_t	zl_cv_writer;	/* log writer thread completion */
 	kcondvar_t	zl_cv_suspend;	/* log suspend completion */
 	uint8_t		zl_suspending;	/* log is currently suspending */
 	uint8_t		zl_keep_first;	/* keep first log block in destroy */
-	uint8_t		zl_stop_replay;	/* don't replay any further */
+	uint8_t		zl_replay;	/* replaying records while set */
 	uint8_t		zl_stop_sync;	/* for debugging */
 	uint8_t		zl_writer;	/* boolean: write setup in progress */
 	uint8_t		zl_log_error;	/* boolean: log write error */
diff --git a/module/zfs/include/sys/zio.h b/module/zfs/include/sys/zio.h
index 4de78dfee..21b0fbc6b 100644
--- a/module/zfs/include/sys/zio.h
+++ b/module/zfs/include/sys/zio.h
@@ -132,12 +132,14 @@ enum zio_compress {
 #define	ZIO_FLAG_IO_RETRY		0x00400
 #define	ZIO_FLAG_IO_REWRITE		0x00800
 
-#define	ZIO_FLAG_PROBE			0x01000
+#define	ZIO_FLAG_SELF_HEAL		0x01000
 #define	ZIO_FLAG_RESILVER		0x02000
 #define	ZIO_FLAG_SCRUB			0x04000
 #define	ZIO_FLAG_SCRUB_THREAD		0x08000
 
-#define	ZIO_FLAG_GANG_CHILD		0x10000
+#define	ZIO_FLAG_PROBE			0x10000
+#define	ZIO_FLAG_GANG_CHILD		0x20000
+#define	ZIO_FLAG_RAW			0x40000
 
 #define	ZIO_FLAG_GANG_INHERIT		\
 	(ZIO_FLAG_CANFAIL |		\
@@ -146,6 +148,7 @@ enum zio_compress {
 	ZIO_FLAG_DONT_RETRY |		\
 	ZIO_FLAG_DONT_CACHE |		\
 	ZIO_FLAG_DONT_AGGREGATE |	\
+	ZIO_FLAG_SELF_HEAL |		\
 	ZIO_FLAG_RESILVER |		\
 	ZIO_FLAG_SCRUB |		\
 	ZIO_FLAG_SCRUB_THREAD)
@@ -156,6 +159,14 @@ enum zio_compress {
 	ZIO_FLAG_IO_RETRY |		\
 	ZIO_FLAG_PROBE)
 
+#define	ZIO_FLAG_AGG_INHERIT		\
+	(ZIO_FLAG_DONT_AGGREGATE |	\
+	ZIO_FLAG_IO_REPAIR |		\
+	ZIO_FLAG_SELF_HEAL |		\
+	ZIO_FLAG_RESILVER |		\
+	ZIO_FLAG_SCRUB |		\
+	ZIO_FLAG_SCRUB_THREAD)
+
 #define	ZIO_PIPELINE_CONTINUE		0x100
 #define	ZIO_PIPELINE_STOP		0x101