[mirror_zfs.git] / include / sys / uberblock_impl.h

/*
 * 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 https://opensource.org/licenses/CDDL-1.0.
 * 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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2016, 2017 by Delphix. All rights reserved.
 */

#ifndef _SYS_UBERBLOCK_IMPL_H
#define	_SYS_UBERBLOCK_IMPL_H

#include <sys/uberblock.h>

#ifdef	__cplusplus
extern "C" {
#endif

/*
 * The uberblock version is incremented whenever an incompatible on-disk
 * format change is made to the SPA, DMU, or ZAP.
 *
 * Note: the first two fields should never be moved.  When a storage pool
 * is opened, the uberblock must be read off the disk before the version
 * can be checked.  If the ub_version field is moved, we may not detect
 * version mismatch.  If the ub_magic field is moved, applications that
 * expect the magic number in the first word won't work.
 */
#define	UBERBLOCK_MAGIC		0x00bab10c		/* oo-ba-bloc!	*/
#define	UBERBLOCK_SHIFT		10			/* up to 1K	*/
#define	MMP_MAGIC		0xa11cea11		/* all-see-all	*/

#define	MMP_INTERVAL_VALID_BIT	0x01
#define	MMP_SEQ_VALID_BIT	0x02
#define	MMP_FAIL_INT_VALID_BIT	0x04

#define	MMP_VALID(ubp)		((ubp)->ub_magic == UBERBLOCK_MAGIC && \
				    (ubp)->ub_mmp_magic == MMP_MAGIC)
#define	MMP_INTERVAL_VALID(ubp)	(MMP_VALID(ubp) && ((ubp)->ub_mmp_config & \
				    MMP_INTERVAL_VALID_BIT))
#define	MMP_SEQ_VALID(ubp)	(MMP_VALID(ubp) && ((ubp)->ub_mmp_config & \
				    MMP_SEQ_VALID_BIT))
#define	MMP_FAIL_INT_VALID(ubp)	(MMP_VALID(ubp) && ((ubp)->ub_mmp_config & \
				    MMP_FAIL_INT_VALID_BIT))

#define	MMP_INTERVAL(ubp)	(((ubp)->ub_mmp_config & 0x00000000FFFFFF00) \
				    >> 8)
#define	MMP_SEQ(ubp)		(((ubp)->ub_mmp_config & 0x0000FFFF00000000) \
				    >> 32)
#define	MMP_FAIL_INT(ubp)	(((ubp)->ub_mmp_config & 0xFFFF000000000000) \
				    >> 48)

#define	MMP_INTERVAL_SET(write) \
	    (((uint64_t)(write & 0xFFFFFF) << 8) | MMP_INTERVAL_VALID_BIT)

#define	MMP_SEQ_SET(seq) \
	    (((uint64_t)(seq & 0xFFFF) << 32) | MMP_SEQ_VALID_BIT)

#define	MMP_FAIL_INT_SET(fail) \
	    (((uint64_t)(fail & 0xFFFF) << 48) | MMP_FAIL_INT_VALID_BIT)

/*
 * RAIDZ expansion reflow information.
 *
 *	64      56      48      40      32      24      16      8       0
 *	+-------+-------+-------+-------+-------+-------+-------+-------+
 *	|Scratch |                    Reflow                            |
 *	| State  |                    Offset                            |
 *	+-------+-------+-------+-------+-------+-------+-------+-------+
 */
typedef enum raidz_reflow_scratch_state {
	RRSS_SCRATCH_NOT_IN_USE = 0,
	RRSS_SCRATCH_VALID,
	RRSS_SCRATCH_INVALID_SYNCED,
	RRSS_SCRATCH_INVALID_SYNCED_ON_IMPORT,
	RRSS_SCRATCH_INVALID_SYNCED_REFLOW
} raidz_reflow_scratch_state_t;

#define	RRSS_GET_OFFSET(ub) \
	BF64_GET_SB((ub)->ub_raidz_reflow_info, 0, 55, SPA_MINBLOCKSHIFT, 0)
#define	RRSS_SET_OFFSET(ub, x) \
	BF64_SET_SB((ub)->ub_raidz_reflow_info, 0, 55, SPA_MINBLOCKSHIFT, 0, x)

#define	RRSS_GET_STATE(ub) \
	BF64_GET((ub)->ub_raidz_reflow_info, 55, 9)
#define	RRSS_SET_STATE(ub, x) \
	BF64_SET((ub)->ub_raidz_reflow_info, 55, 9, x)

#define	RAIDZ_REFLOW_SET(ub, state, offset) do { \
	(ub)->ub_raidz_reflow_info = 0; \
	RRSS_SET_OFFSET(ub, offset); \
	RRSS_SET_STATE(ub, state); \
} while (0)

struct uberblock {
	uint64_t	ub_magic;	/* UBERBLOCK_MAGIC		*/
	uint64_t	ub_version;	/* SPA_VERSION			*/
	uint64_t	ub_txg;		/* txg of last sync		*/
	uint64_t	ub_guid_sum;	/* sum of all vdev guids	*/
	uint64_t	ub_timestamp;	/* UTC time of last sync	*/
	blkptr_t	ub_rootbp;	/* MOS objset_phys_t		*/

	/* highest SPA_VERSION supported by software that wrote this txg */
	uint64_t	ub_software_version;

	/* Maybe missing in uberblocks we read, but always written */
	uint64_t	ub_mmp_magic;	/* MMP_MAGIC			*/
	/*
	 * If ub_mmp_delay == 0 and ub_mmp_magic is valid, MMP is off.
	 * Otherwise, nanosec since last MMP write.
	 */
	uint64_t	ub_mmp_delay;

	/*
	 * The ub_mmp_config contains the multihost write interval, multihost
	 * fail intervals, sequence number for sub-second granularity, and
	 * valid bit mask.  This layout is as follows:
	 *
	 *   64      56      48      40      32      24      16      8       0
	 *   +-------+-------+-------+-------+-------+-------+-------+-------+
	 * 0 | Fail Intervals|      Seq      |   Write Interval (ms) | VALID |
	 *   +-------+-------+-------+-------+-------+-------+-------+-------+
	 *
	 * This allows a write_interval of (2^24/1000)s, over 4.5 hours
	 *
	 * VALID Bits:
	 * - 0x01 - Write Interval (ms)
	 * - 0x02 - Sequence number exists
	 * - 0x04 - Fail Intervals
	 * - 0xf8 - Reserved
	 */
	uint64_t	ub_mmp_config;

	/*
	 * ub_checkpoint_txg indicates two things about the current uberblock:
	 *
	 * 1] If it is not zero then this uberblock is a checkpoint. If it is
	 *    zero, then this uberblock is not a checkpoint.
	 *
	 * 2] On checkpointed uberblocks, the value of ub_checkpoint_txg is
	 *    the ub_txg that the uberblock had at the time we moved it to
	 *    the MOS config.
	 *
	 * The field is set when we checkpoint the uberblock and continues to
	 * hold that value even after we've rewound (unlike the ub_txg that
	 * is reset to a higher value).
	 *
	 * Besides checks used to determine whether we are reopening the
	 * pool from a checkpointed uberblock [see spa_ld_select_uberblock()],
	 * the value of the field is used to determine which ZIL blocks have
	 * been allocated according to the ms_sm when we are rewinding to a
	 * checkpoint. Specifically, if logical birth > ub_checkpoint_txg,then
	 * the ZIL block is not allocated [see uses of spa_min_claim_txg()].
	 */
	uint64_t	ub_checkpoint_txg;

	uint64_t	ub_raidz_reflow_info;
};

#ifdef	__cplusplus
}
#endif

#endif	/* _SYS_UBERBLOCK_IMPL_H */
Commit	Line	Data
34dc7c2f BB	1	/*
	2	* CDDL HEADER START
	3	*
	4	* The contents of this file are subject to the terms of the
	5	* Common Development and Distribution License (the "License").
	6	* You may not use this file except in compliance with the License.
	7	*
	8	* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
1d3ba0bf	9	* or https://opensource.org/licenses/CDDL-1.0.
34dc7c2f BB	10	* See the License for the specific language governing permissions
	11	* and limitations under the License.
	12	*
	13	* When distributing Covered Code, include this CDDL HEADER in each
	14	* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
	15	* If applicable, add the following below this CDDL HEADER, with the
	16	* fields enclosed by brackets "[]" replaced with your own identifying
	17	* information: Portions Copyright [yyyy] [name of copyright owner]
	18	*
	19	* CDDL HEADER END
	20	*/
	21	/*
428870ff	22	* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
d2734cce	23	* Copyright (c) 2016, 2017 by Delphix. All rights reserved.
34dc7c2f BB	24	*/
	25
	26	#ifndef _SYS_UBERBLOCK_IMPL_H
	27	#define _SYS_UBERBLOCK_IMPL_H
	28
34dc7c2f BB	29	#include <sys/uberblock.h>
	30
	31	#ifdef __cplusplus
	32	extern "C" {
	33	#endif
	34
	35	/*
	36	* The uberblock version is incremented whenever an incompatible on-disk
	37	* format change is made to the SPA, DMU, or ZAP.
	38	*
	39	* Note: the first two fields should never be moved. When a storage pool
	40	* is opened, the uberblock must be read off the disk before the version
	41	* can be checked. If the ub_version field is moved, we may not detect
	42	* version mismatch. If the ub_magic field is moved, applications that
	43	* expect the magic number in the first word won't work.
	44	*/
	45	#define UBERBLOCK_MAGIC 0x00bab10c /* oo-ba-bloc! */
	46	#define UBERBLOCK_SHIFT 10 /* up to 1K */
060f0226 OF	47	#define MMP_MAGIC 0xa11cea11 /* all-see-all */
	48
	49	#define MMP_INTERVAL_VALID_BIT 0x01
	50	#define MMP_SEQ_VALID_BIT 0x02
	51	#define MMP_FAIL_INT_VALID_BIT 0x04
	52
c3f2f1aa DB	53	#define MMP_VALID(ubp) ((ubp)->ub_magic == UBERBLOCK_MAGIC && \
	54	(ubp)->ub_mmp_magic == MMP_MAGIC)
	55	#define MMP_INTERVAL_VALID(ubp) (MMP_VALID(ubp) && ((ubp)->ub_mmp_config & \
060f0226	56	MMP_INTERVAL_VALID_BIT))
c3f2f1aa	57	#define MMP_SEQ_VALID(ubp) (MMP_VALID(ubp) && ((ubp)->ub_mmp_config & \
060f0226	58	MMP_SEQ_VALID_BIT))
c3f2f1aa	59	#define MMP_FAIL_INT_VALID(ubp) (MMP_VALID(ubp) && ((ubp)->ub_mmp_config & \
060f0226 OF	60	MMP_FAIL_INT_VALID_BIT))
060f0226 OF	61
c3f2f1aa	62	#define MMP_INTERVAL(ubp) (((ubp)->ub_mmp_config & 0x00000000FFFFFF00) \
060f0226	63	>> 8)
c3f2f1aa	64	#define MMP_SEQ(ubp) (((ubp)->ub_mmp_config & 0x0000FFFF00000000) \
060f0226	65	>> 32)
c3f2f1aa	66	#define MMP_FAIL_INT(ubp) (((ubp)->ub_mmp_config & 0xFFFF000000000000) \
060f0226 OF	67	>> 48)
	68
	69	#define MMP_INTERVAL_SET(write) \
	70	(((uint64_t)(write & 0xFFFFFF) << 8) \| MMP_INTERVAL_VALID_BIT)
	71
	72	#define MMP_SEQ_SET(seq) \
	73	(((uint64_t)(seq & 0xFFFF) << 32) \| MMP_SEQ_VALID_BIT)
	74
	75	#define MMP_FAIL_INT_SET(fail) \
	76	(((uint64_t)(fail & 0xFFFF) << 48) \| MMP_FAIL_INT_VALID_BIT)
34dc7c2f	77
5caeef02 DB	78	/*
	79	* RAIDZ expansion reflow information.
	80	*
	81	* 64 56 48 40 32 24 16 8 0
	82	* +-------+-------+-------+-------+-------+-------+-------+-------+
	83	* \|Scratch \| Reflow \|
	84	* \| State \| Offset \|
	85	* +-------+-------+-------+-------+-------+-------+-------+-------+
	86	*/
	87	typedef enum raidz_reflow_scratch_state {
	88	RRSS_SCRATCH_NOT_IN_USE = 0,
	89	RRSS_SCRATCH_VALID,
	90	RRSS_SCRATCH_INVALID_SYNCED,
	91	RRSS_SCRATCH_INVALID_SYNCED_ON_IMPORT,
	92	RRSS_SCRATCH_INVALID_SYNCED_REFLOW
	93	} raidz_reflow_scratch_state_t;
	94
	95	#define RRSS_GET_OFFSET(ub) \
	96	BF64_GET_SB((ub)->ub_raidz_reflow_info, 0, 55, SPA_MINBLOCKSHIFT, 0)
	97	#define RRSS_SET_OFFSET(ub, x) \
	98	BF64_SET_SB((ub)->ub_raidz_reflow_info, 0, 55, SPA_MINBLOCKSHIFT, 0, x)
	99
	100	#define RRSS_GET_STATE(ub) \
	101	BF64_GET((ub)->ub_raidz_reflow_info, 55, 9)
	102	#define RRSS_SET_STATE(ub, x) \
	103	BF64_SET((ub)->ub_raidz_reflow_info, 55, 9, x)
	104
	105	#define RAIDZ_REFLOW_SET(ub, state, offset) do { \
	106	(ub)->ub_raidz_reflow_info = 0; \
	107	RRSS_SET_OFFSET(ub, offset); \
	108	RRSS_SET_STATE(ub, state); \
	109	} while (0)
	110
34dc7c2f BB	111	struct uberblock {
	112	uint64_t ub_magic; /* UBERBLOCK_MAGIC */
	113	uint64_t ub_version; /* SPA_VERSION */
	114	uint64_t ub_txg; /* txg of last sync */
	115	uint64_t ub_guid_sum; /* sum of all vdev guids */
	116	uint64_t ub_timestamp; /* UTC time of last sync */
	117	blkptr_t ub_rootbp; /* MOS objset_phys_t */
428870ff BB	118
	119	/* highest SPA_VERSION supported by software that wrote this txg */
	120	uint64_t ub_software_version;
379ca9cf OF	121
	122	/* Maybe missing in uberblocks we read, but always written */
	123	uint64_t ub_mmp_magic; /* MMP_MAGIC */
060f0226 OF	124	/*
	125	* If ub_mmp_delay == 0 and ub_mmp_magic is valid, MMP is off.
	126	* Otherwise, nanosec since last MMP write.
	127	*/
	128	uint64_t ub_mmp_delay;
	129
	130	/*
	131	* The ub_mmp_config contains the multihost write interval, multihost
	132	* fail intervals, sequence number for sub-second granularity, and
	133	* valid bit mask. This layout is as follows:
	134	*
	135	* 64 56 48 40 32 24 16 8 0
	136	* +-------+-------+-------+-------+-------+-------+-------+-------+
	137	* 0 \| Fail Intervals\| Seq \| Write Interval (ms) \| VALID \|
	138	* +-------+-------+-------+-------+-------+-------+-------+-------+
	139	*
	140	* This allows a write_interval of (2^24/1000)s, over 4.5 hours
	141	*
	142	* VALID Bits:
	143	* - 0x01 - Write Interval (ms)
	144	* - 0x02 - Sequence number exists
	145	* - 0x04 - Fail Intervals
	146	* - 0xf8 - Reserved
	147	*/
	148	uint64_t ub_mmp_config;
d2734cce SD	149
	150	/*
	151	* ub_checkpoint_txg indicates two things about the current uberblock:
	152	*
	153	* 1] If it is not zero then this uberblock is a checkpoint. If it is
	154	* zero, then this uberblock is not a checkpoint.
	155	*
	156	* 2] On checkpointed uberblocks, the value of ub_checkpoint_txg is
	157	* the ub_txg that the uberblock had at the time we moved it to
	158	* the MOS config.
	159	*
	160	* The field is set when we checkpoint the uberblock and continues to
	161	* hold that value even after we've rewound (unlike the ub_txg that
	162	* is reset to a higher value).
	163	*
	164	* Besides checks used to determine whether we are reopening the
	165	* pool from a checkpointed uberblock [see spa_ld_select_uberblock()],
	166	* the value of the field is used to determine which ZIL blocks have
	167	* been allocated according to the ms_sm when we are rewinding to a
493fcce9	168	* checkpoint. Specifically, if logical birth > ub_checkpoint_txg,then
d2734cce SD	169	* the ZIL block is not allocated [see uses of spa_min_claim_txg()].
d2734cce SD	170	*/
829f9251	171	uint64_t ub_checkpoint_txg;
5caeef02 DB	172
5caeef02 DB	173	uint64_t ub_raidz_reflow_info;
34dc7c2f BB	174	};
	175
	176	#ifdef __cplusplus
	177	}
	178	#endif
	179
	180	#endif /* _SYS_UBERBLOCK_IMPL_H */