[mirror_zfs.git] / module / zfs / spa_history.c

/*
 * 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) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2013 by Delphix. All rights reserved.
 */

#include <sys/spa.h>
#include <sys/spa_impl.h>
#include <sys/zap.h>
#include <sys/dsl_synctask.h>
#include <sys/dmu_tx.h>
#include <sys/dmu_objset.h>
#include <sys/dsl_dataset.h>
#include <sys/dsl_dir.h>
#include <sys/utsname.h>
#include <sys/cmn_err.h>
#include <sys/sunddi.h>
#include <sys/cred.h>
#include "zfs_comutil.h"
#ifdef _KERNEL
#include <sys/zone.h>
#endif

/*
 * Routines to manage the on-disk history log.
 *
 * The history log is stored as a dmu object containing
 * <packed record length, record nvlist> tuples.
 *
 * Where "record nvlist" is a nvlist containing uint64_ts and strings, and
 * "packed record length" is the packed length of the "record nvlist" stored
 * as a little endian uint64_t.
 *
 * The log is implemented as a ring buffer, though the original creation
 * of the pool ('zpool create') is never overwritten.
 *
 * The history log is tracked as object 'spa_t::spa_history'.  The bonus buffer
 * of 'spa_history' stores the offsets for logging/retrieving history as
 * 'spa_history_phys_t'.  'sh_pool_create_len' is the ending offset in bytes of
 * where the 'zpool create' record is stored.  This allows us to never
 * overwrite the original creation of the pool.  'sh_phys_max_off' is the
 * physical ending offset in bytes of the log.  This tells you the length of
 * the buffer. 'sh_eof' is the logical EOF (in bytes).  Whenever a record
 * is added, 'sh_eof' is incremented by the the size of the record.
 * 'sh_eof' is never decremented.  'sh_bof' is the logical BOF (in bytes).
 * This is where the consumer should start reading from after reading in
 * the 'zpool create' portion of the log.
 *
 * 'sh_records_lost' keeps track of how many records have been overwritten
 * and permanently lost.
 */

/* convert a logical offset to physical */
static uint64_t
spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
{
	uint64_t phys_len;

	phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len;
	return ((log_off - shpp->sh_pool_create_len) % phys_len
	    + shpp->sh_pool_create_len);
}

void
spa_history_create_obj(spa_t *spa, dmu_tx_t *tx)
{
	dmu_buf_t *dbp;
	spa_history_phys_t *shpp;
	objset_t *mos = spa->spa_meta_objset;

	ASSERT(spa->spa_history == 0);
	spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY,
	    SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS,
	    sizeof (spa_history_phys_t), tx);

	VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
	    DMU_POOL_HISTORY, sizeof (uint64_t), 1,
	    &spa->spa_history, tx) == 0);

	VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
	ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));

	shpp = dbp->db_data;
	dmu_buf_will_dirty(dbp, tx);

	/*
	 * Figure out maximum size of history log.  We set it at
	 * 0.1% of pool size, with a max of 1G and min of 128KB.
	 */
	shpp->sh_phys_max_off =
	    metaslab_class_get_dspace(spa_normal_class(spa)) / 1000;
	shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 1<<30);
	shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10);

	dmu_buf_rele(dbp, FTAG);
}

/*
 * Change 'sh_bof' to the beginning of the next record.
 */
static int
spa_history_advance_bof(spa_t *spa, spa_history_phys_t *shpp)
{
	objset_t *mos = spa->spa_meta_objset;
	uint64_t firstread, reclen, phys_bof;
	char buf[sizeof (reclen)];
	int err;

	phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
	firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);

	if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
	    buf, DMU_READ_PREFETCH)) != 0)
		return (err);
	if (firstread != sizeof (reclen)) {
		if ((err = dmu_read(mos, spa->spa_history,
		    shpp->sh_pool_create_len, sizeof (reclen) - firstread,
		    buf + firstread, DMU_READ_PREFETCH)) != 0)
			return (err);
	}

	reclen = LE_64(*((uint64_t *)buf));
	shpp->sh_bof += reclen + sizeof (reclen);
	shpp->sh_records_lost++;
	return (0);
}

static int
spa_history_write(spa_t *spa, void *buf, uint64_t len, spa_history_phys_t *shpp,
    dmu_tx_t *tx)
{
	uint64_t firstwrite, phys_eof;
	objset_t *mos = spa->spa_meta_objset;
	int err;

	ASSERT(MUTEX_HELD(&spa->spa_history_lock));

	/* see if we need to reset logical BOF */
	while (shpp->sh_phys_max_off - shpp->sh_pool_create_len -
	    (shpp->sh_eof - shpp->sh_bof) <= len) {
		if ((err = spa_history_advance_bof(spa, shpp)) != 0) {
			return (err);
		}
	}

	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
	firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
	shpp->sh_eof += len;
	dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);

	len -= firstwrite;
	if (len > 0) {
		/* write out the rest at the beginning of physical file */
		dmu_write(mos, spa->spa_history, shpp->sh_pool_create_len,
		    len, (char *)buf + firstwrite, tx);
	}

	return (0);
}

static char *
spa_history_zone(void)
{
#ifdef _KERNEL
#ifdef HAVE_SPL
	return ("linux");
#else
	return (curproc->p_zone->zone_name);
#endif
#else
	return (NULL);
#endif
}

/*
 * Write out a history event.
 */
/*ARGSUSED*/
static void
spa_history_log_sync(void *arg, dmu_tx_t *tx)
{
	nvlist_t	*nvl = arg;
	spa_t		*spa = dmu_tx_pool(tx)->dp_spa;
	objset_t	*mos = spa->spa_meta_objset;
	dmu_buf_t	*dbp;
	spa_history_phys_t *shpp;
	size_t		reclen;
	uint64_t	le_len;
	char		*record_packed = NULL;
	int		ret;

	/*
	 * If we have an older pool that doesn't have a command
	 * history object, create it now.
	 */
	mutex_enter(&spa->spa_history_lock);
	if (!spa->spa_history)
		spa_history_create_obj(spa, tx);
	mutex_exit(&spa->spa_history_lock);

	/*
	 * Get the offset of where we need to write via the bonus buffer.
	 * Update the offset when the write completes.
	 */
	VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
	shpp = dbp->db_data;

	dmu_buf_will_dirty(dbp, tx);

#ifdef ZFS_DEBUG
	{
		dmu_object_info_t doi;
		dmu_object_info_from_db(dbp, &doi);
		ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
	}
#endif

	fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
#ifdef _KERNEL
	fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname.nodename);
#endif
	if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
		zfs_dbgmsg("command: %s",
		    fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
	} else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
		if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
			zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
			    fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
			    fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
			    fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
		} else {
			zfs_dbgmsg("txg %lld %s %s",
			    fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
			    fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
		}
	} else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
		zfs_dbgmsg("ioctl %s",
		    fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
	}

	VERIFY3U(nvlist_pack(nvl, &record_packed, &reclen, NV_ENCODE_NATIVE,
	    KM_PUSHPAGE), ==, 0);

	mutex_enter(&spa->spa_history_lock);

	/* write out the packed length as little endian */
	le_len = LE_64((uint64_t)reclen);
	ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
	if (!ret)
		ret = spa_history_write(spa, record_packed, reclen, shpp, tx);

	/* The first command is the create, which we keep forever */
	if (ret == 0 && shpp->sh_pool_create_len == 0 &&
	    nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
		shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
	}

	mutex_exit(&spa->spa_history_lock);
	fnvlist_pack_free(record_packed, reclen);
	dmu_buf_rele(dbp, FTAG);
	fnvlist_free(nvl);
}

/*
 * Write out a history event.
 */
int
spa_history_log(spa_t *spa, const char *msg)
{
	int err;
	nvlist_t *nvl;

        VERIFY0(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE));

	fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg);
	err = spa_history_log_nvl(spa, nvl);
	fnvlist_free(nvl);
	return (err);
}

int
spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
{
	int err = 0;
	dmu_tx_t *tx;
	nvlist_t *nvarg;

	if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY || !spa_writeable(spa))
		return (SET_ERROR(EINVAL));

	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
	err = dmu_tx_assign(tx, TXG_WAIT);
	if (err) {
		dmu_tx_abort(tx);
		return (err);
	}

	VERIFY0(nvlist_dup(nvl, &nvarg, KM_PUSHPAGE));
	if (spa_history_zone() != NULL) {
		fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
		    spa_history_zone());
	}
	fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));

	/* Kick this off asynchronously; errors are ignored. */
	dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
	    nvarg, 0, tx);
	dmu_tx_commit(tx);

	/* spa_history_log_sync will free nvl */
	return (err);

}

/*
 * Read out the command history.
 */
int
spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
{
	objset_t *mos = spa->spa_meta_objset;
	dmu_buf_t *dbp;
	uint64_t read_len, phys_read_off, phys_eof;
	uint64_t leftover = 0;
	spa_history_phys_t *shpp;
	int err;

	/*
	 * If the command history doesn't exist (older pool),
	 * that's ok, just return ENOENT.
	 */
	if (!spa->spa_history)
		return (SET_ERROR(ENOENT));

	/*
	 * The history is logged asynchronously, so when they request
	 * the first chunk of history, make sure everything has been
	 * synced to disk so that we get it.
	 */
	if (*offp == 0 && spa_writeable(spa))
		txg_wait_synced(spa_get_dsl(spa), 0);

	if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
		return (err);
	shpp = dbp->db_data;

#ifdef ZFS_DEBUG
	{
		dmu_object_info_t doi;
		dmu_object_info_from_db(dbp, &doi);
		ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
	}
#endif

	mutex_enter(&spa->spa_history_lock);
	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);

	if (*offp < shpp->sh_pool_create_len) {
		/* read in just the zpool create history */
		phys_read_off = *offp;
		read_len = MIN(*len, shpp->sh_pool_create_len -
		    phys_read_off);
	} else {
		/*
		 * Need to reset passed in offset to BOF if the passed in
		 * offset has since been overwritten.
		 */
		*offp = MAX(*offp, shpp->sh_bof);
		phys_read_off = spa_history_log_to_phys(*offp, shpp);

		/*
		 * Read up to the minimum of what the user passed down or
		 * the EOF (physical or logical).  If we hit physical EOF,
		 * use 'leftover' to read from the physical BOF.
		 */
		if (phys_read_off <= phys_eof) {
			read_len = MIN(*len, phys_eof - phys_read_off);
		} else {
			read_len = MIN(*len,
			    shpp->sh_phys_max_off - phys_read_off);
			if (phys_read_off + *len > shpp->sh_phys_max_off) {
				leftover = MIN(*len - read_len,
				    phys_eof - shpp->sh_pool_create_len);
			}
		}
	}

	/* offset for consumer to use next */
	*offp += read_len + leftover;

	/* tell the consumer how much you actually read */
	*len = read_len + leftover;

	if (read_len == 0) {
		mutex_exit(&spa->spa_history_lock);
		dmu_buf_rele(dbp, FTAG);
		return (0);
	}

	err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
	    DMU_READ_PREFETCH);
	if (leftover && err == 0) {
		err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
		    leftover, buf + read_len, DMU_READ_PREFETCH);
	}
	mutex_exit(&spa->spa_history_lock);

	dmu_buf_rele(dbp, FTAG);
	return (err);
}

/*
 * The nvlist will be consumed by this call.
 */
static void
log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
    dmu_tx_t *tx, const char *fmt, va_list adx)
{
	char *msg;
	va_list adx1;
	int size;

	/*
	 * If this is part of creating a pool, not everything is
	 * initialized yet, so don't bother logging the internal events.
	 * Likewise if the pool is not writeable.
	 */
	if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
		fnvlist_free(nvl);
		return;
	}

	va_copy(adx1, adx);
	size = vsnprintf(NULL, 0, fmt, adx1) + 1;
	msg = kmem_alloc(size, KM_PUSHPAGE);
	va_end(adx1);
	va_copy(adx1, adx);
	(void) vsprintf(msg, fmt, adx1);
	va_end(adx1);
	fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
	kmem_free(msg, size);

	fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
	fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);

	if (dmu_tx_is_syncing(tx)) {
		spa_history_log_sync(nvl, tx);
	} else {
		dsl_sync_task_nowait(spa_get_dsl(spa),
		    spa_history_log_sync, nvl, 0, tx);
	}
	/* spa_history_log_sync() will free nvl */
}

void
spa_history_log_internal(spa_t *spa, const char *operation,
    dmu_tx_t *tx, const char *fmt, ...)
{
	dmu_tx_t *htx = tx;
	va_list adx;
	nvlist_t *nvl;

	/* create a tx if we didn't get one */
	if (tx == NULL) {
		htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
		if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
			dmu_tx_abort(htx);
			return;
		}
	}

	va_start(adx, fmt);
	VERIFY0(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE));
	log_internal(nvl, operation, spa, htx, fmt, adx);
	va_end(adx);

	/* if we didn't get a tx from the caller, commit the one we made */
	if (tx == NULL)
		dmu_tx_commit(htx);
}

void
spa_history_log_internal_ds(dsl_dataset_t *ds, const char *operation,
    dmu_tx_t *tx, const char *fmt, ...)
{
	va_list adx;
	char namebuf[MAXNAMELEN];
	nvlist_t *nvl;

	ASSERT(tx != NULL);

	dsl_dataset_name(ds, namebuf);
	VERIFY0(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE));
	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID, ds->ds_object);

	va_start(adx, fmt);
	log_internal(nvl, operation, dsl_dataset_get_spa(ds), tx, fmt, adx);
	va_end(adx);
}

void
spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
    dmu_tx_t *tx, const char *fmt, ...)
{
	va_list adx;
	char namebuf[MAXNAMELEN];
	nvlist_t *nvl;

	ASSERT(tx != NULL);

	dsl_dir_name(dd, namebuf);
	VERIFY0(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE));
	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
	    dd->dd_phys->dd_head_dataset_obj);

	va_start(adx, fmt);
	log_internal(nvl, operation, dd->dd_pool->dp_spa, tx, fmt, adx);
	va_end(adx);
}

void
spa_history_log_version(spa_t *spa, const char *operation)
{
	spa_history_log_internal(spa, operation, NULL,
	    "pool version %llu; software version %llu/%d; uts %s %s %s %s",
	    (u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
	    utsname.nodename, utsname.release, utsname.version,
	    utsname.machine);
}

#if defined(_KERNEL) && defined(HAVE_SPL)
EXPORT_SYMBOL(spa_history_create_obj);
EXPORT_SYMBOL(spa_history_get);
EXPORT_SYMBOL(spa_history_log);
EXPORT_SYMBOL(spa_history_log_internal);
EXPORT_SYMBOL(spa_history_log_version);
#endif
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
	9	* or http://www.opensolaris.org/os/licensing.
	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
	22	/*
428870ff	23	* Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
2e528b49	24	* Copyright (c) 2013 by Delphix. All rights reserved.
34dc7c2f BB	25	*/
34dc7c2f BB	26
34dc7c2f BB	27	#include <sys/spa.h>
	28	#include <sys/spa_impl.h>
	29	#include <sys/zap.h>
	30	#include <sys/dsl_synctask.h>
	31	#include <sys/dmu_tx.h>
	32	#include <sys/dmu_objset.h>
6f1ffb06 MA	33	#include <sys/dsl_dataset.h>
6f1ffb06 MA	34	#include <sys/dsl_dir.h>
34dc7c2f BB	35	#include <sys/utsname.h>
	36	#include <sys/cmn_err.h>
	37	#include <sys/sunddi.h>
6f1ffb06	38	#include <sys/cred.h>
428870ff	39	#include "zfs_comutil.h"
34dc7c2f BB	40	#ifdef _KERNEL
	41	#include <sys/zone.h>
	42	#endif
	43
	44	/*
	45	* Routines to manage the on-disk history log.
	46	*
	47	* The history log is stored as a dmu object containing
	48	* <packed record length, record nvlist> tuples.
	49	*
	50	* Where "record nvlist" is a nvlist containing uint64_ts and strings, and
	51	* "packed record length" is the packed length of the "record nvlist" stored
	52	* as a little endian uint64_t.
	53	*
	54	* The log is implemented as a ring buffer, though the original creation
	55	* of the pool ('zpool create') is never overwritten.
	56	*
	57	* The history log is tracked as object 'spa_t::spa_history'. The bonus buffer
	58	* of 'spa_history' stores the offsets for logging/retrieving history as
	59	* 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of
	60	* where the 'zpool create' record is stored. This allows us to never
	61	* overwrite the original creation of the pool. 'sh_phys_max_off' is the
	62	* physical ending offset in bytes of the log. This tells you the length of
	63	* the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record
	64	* is added, 'sh_eof' is incremented by the the size of the record.
	65	* 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes).
	66	* This is where the consumer should start reading from after reading in
	67	* the 'zpool create' portion of the log.
	68	*
	69	* 'sh_records_lost' keeps track of how many records have been overwritten
	70	* and permanently lost.
	71	*/
	72
	73	/* convert a logical offset to physical */
	74	static uint64_t
	75	spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
	76	{
	77	uint64_t phys_len;
	78
	79	phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len;
	80	return ((log_off - shpp->sh_pool_create_len) % phys_len
	81	+ shpp->sh_pool_create_len);
	82	}
	83
	84	void
	85	spa_history_create_obj(spa_t spa, dmu_tx_t tx)
	86	{
	87	dmu_buf_t *dbp;
	88	spa_history_phys_t *shpp;
	89	objset_t *mos = spa->spa_meta_objset;
	90
	91	ASSERT(spa->spa_history == 0);
	92	spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY,
	93	SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS,
	94	sizeof (spa_history_phys_t), tx);
	95
	96	VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
	97	DMU_POOL_HISTORY, sizeof (uint64_t), 1,
	98	&spa->spa_history, tx) == 0);
	99
	100	VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
	101	ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));
	102
	103	shpp = dbp->db_data;
104	dmu_buf_will_dirty(dbp, tx);
105
106	/*
107	* Figure out maximum size of history log. We set it at
330d06f9	108	* 0.1% of pool size, with a max of 1G and min of 128KB.
34dc7c2f	109	*/
428870ff	110	shpp->sh_phys_max_off =
330d06f9 MA	111	metaslab_class_get_dspace(spa_normal_class(spa)) / 1000;
330d06f9 MA	112	shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 1<<30);
34dc7c2f BB	113	shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10);
	114
	115	dmu_buf_rele(dbp, FTAG);
	116	}
	117
	118	/*
	119	* Change 'sh_bof' to the beginning of the next record.
	120	*/
	121	static int
	122	spa_history_advance_bof(spa_t spa, spa_history_phys_t shpp)
	123	{
	124	objset_t *mos = spa->spa_meta_objset;
	125	uint64_t firstread, reclen, phys_bof;
	126	char buf[sizeof (reclen)];
	127	int err;
	128
	129	phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
	130	firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);
	131
	132	if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
9babb374	133	buf, DMU_READ_PREFETCH)) != 0)
34dc7c2f BB	134	return (err);
	135	if (firstread != sizeof (reclen)) {
	136	if ((err = dmu_read(mos, spa->spa_history,
	137	shpp->sh_pool_create_len, sizeof (reclen) - firstread,
9babb374	138	buf + firstread, DMU_READ_PREFETCH)) != 0)
34dc7c2f BB	139	return (err);
	140	}
	141
	142	reclen = LE_64(((uint64_t )buf));
	143	shpp->sh_bof += reclen + sizeof (reclen);
	144	shpp->sh_records_lost++;
	145	return (0);
	146	}
	147
	148	static int
	149	spa_history_write(spa_t spa, void buf, uint64_t len, spa_history_phys_t *shpp,
	150	dmu_tx_t *tx)
	151	{
	152	uint64_t firstwrite, phys_eof;
	153	objset_t *mos = spa->spa_meta_objset;
	154	int err;
	155
	156	ASSERT(MUTEX_HELD(&spa->spa_history_lock));
	157
	158	/* see if we need to reset logical BOF */
	159	while (shpp->sh_phys_max_off - shpp->sh_pool_create_len -
	160	(shpp->sh_eof - shpp->sh_bof) <= len) {
	161	if ((err = spa_history_advance_bof(spa, shpp)) != 0) {
	162	return (err);
	163	}
	164	}
	165
	166	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
	167	firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
	168	shpp->sh_eof += len;
	169	dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);
	170
	171	len -= firstwrite;
	172	if (len > 0) {
	173	/* write out the rest at the beginning of physical file */
	174	dmu_write(mos, spa->spa_history, shpp->sh_pool_create_len,
	175	len, (char *)buf + firstwrite, tx);
	176	}
	177
	178	return (0);
	179	}
	180
	181	static char *
0bc8fd78	182	spa_history_zone(void)
34dc7c2f BB	183	{
34dc7c2f BB	184	#ifdef _KERNEL
c28b2279 BB	185	#ifdef HAVE_SPL
	186	return ("linux");
	187	#else
34dc7c2f	188	return (curproc->p_zone->zone_name);
c28b2279	189	#endif
34dc7c2f	190	#else
6f1ffb06	191	return (NULL);
34dc7c2f BB	192	#endif
	193	}
	194
	195	/*
	196	* Write out a history event.
	197	*/
428870ff	198	/ARGSUSED/
34dc7c2f	199	static void
13fe0198	200	spa_history_log_sync(void arg, dmu_tx_t tx)
34dc7c2f	201	{
13fe0198 MA	202	nvlist_t *nvl = arg;
13fe0198 MA	203	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
34dc7c2f BB	204	objset_t *mos = spa->spa_meta_objset;
	205	dmu_buf_t *dbp;
	206	spa_history_phys_t *shpp;
	207	size_t reclen;
	208	uint64_t le_len;
34dc7c2f BB	209	char *record_packed = NULL;
	210	int ret;
	211
	212	/*
	213	* If we have an older pool that doesn't have a command
	214	* history object, create it now.
	215	*/
	216	mutex_enter(&spa->spa_history_lock);
	217	if (!spa->spa_history)
	218	spa_history_create_obj(spa, tx);
	219	mutex_exit(&spa->spa_history_lock);
	220
	221	/*
	222	* Get the offset of where we need to write via the bonus buffer.
	223	* Update the offset when the write completes.
	224	*/
13fe0198	225	VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
34dc7c2f BB	226	shpp = dbp->db_data;
	227
	228	dmu_buf_will_dirty(dbp, tx);
	229
	230	#ifdef ZFS_DEBUG
	231	{
	232	dmu_object_info_t doi;
	233	dmu_object_info_from_db(dbp, &doi);
	234	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
	235	}
	236	#endif
	237
6f1ffb06	238	fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
34dc7c2f	239	#ifdef _KERNEL
6f1ffb06	240	fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname.nodename);
34dc7c2f	241	#endif
6f1ffb06 MA	242	if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
	243	zfs_dbgmsg("command: %s",
	244	fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
	245	} else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
	246	if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
	247	zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
	248	fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
	249	fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
	250	fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
	251	fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
	252	fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
	253	} else {
	254	zfs_dbgmsg("txg %lld %s %s",
	255	fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
	256	fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
	257	fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
	258	}
	259	} else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
	260	zfs_dbgmsg("ioctl %s",
	261	fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
34dc7c2f BB	262	}
34dc7c2f BB	263
6f1ffb06 MA	264	VERIFY3U(nvlist_pack(nvl, &record_packed, &reclen, NV_ENCODE_NATIVE,
6f1ffb06 MA	265	KM_PUSHPAGE), ==, 0);
34dc7c2f BB	266
34dc7c2f BB	267	mutex_enter(&spa->spa_history_lock);
34dc7c2f BB	268
	269	/* write out the packed length as little endian */
	270	le_len = LE_64((uint64_t)reclen);
	271	ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
	272	if (!ret)
	273	ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
	274
6f1ffb06 MA	275	/* The first command is the create, which we keep forever */
	276	if (ret == 0 && shpp->sh_pool_create_len == 0 &&
	277	nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
	278	shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
34dc7c2f BB	279	}
	280
	281	mutex_exit(&spa->spa_history_lock);
6f1ffb06	282	fnvlist_pack_free(record_packed, reclen);
34dc7c2f	283	dmu_buf_rele(dbp, FTAG);
6f1ffb06	284	fnvlist_free(nvl);
34dc7c2f BB	285	}
	286
	287	/*
	288	* Write out a history event.
	289	*/
	290	int
6f1ffb06 MA	291	spa_history_log(spa_t spa, const char msg)
	292	{
	293	int err;
	294	nvlist_t *nvl;
	295
	296	VERIFY0(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE));
	297
	298	fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg);
	299	err = spa_history_log_nvl(spa, nvl);
	300	fnvlist_free(nvl);
	301	return (err);
	302	}
	303
	304	int
	305	spa_history_log_nvl(spa_t spa, nvlist_t nvl)
34dc7c2f	306	{
428870ff BB	307	int err = 0;
428870ff BB	308	dmu_tx_t *tx;
6f1ffb06	309	nvlist_t *nvarg;
34dc7c2f	310
6f1ffb06	311	if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY \|\| !spa_writeable(spa))
2e528b49	312	return (SET_ERROR(EINVAL));
34dc7c2f	313
428870ff BB	314	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
	315	err = dmu_tx_assign(tx, TXG_WAIT);
	316	if (err) {
	317	dmu_tx_abort(tx);
	318	return (err);
	319	}
	320
6f1ffb06 MA	321	VERIFY0(nvlist_dup(nvl, &nvarg, KM_PUSHPAGE));
	322	if (spa_history_zone() != NULL) {
	323	fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
	324	spa_history_zone());
	325	}
	326	fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
428870ff BB	327
428870ff BB	328	/* Kick this off asynchronously; errors are ignored. */
13fe0198 MA	329	dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
13fe0198 MA	330	nvarg, 0, tx);
428870ff BB	331	dmu_tx_commit(tx);
428870ff BB	332
6f1ffb06	333	/* spa_history_log_sync will free nvl */
428870ff	334	return (err);
6f1ffb06	335
34dc7c2f BB	336	}
	337
	338	/*
	339	* Read out the command history.
	340	*/
	341	int
	342	spa_history_get(spa_t spa, uint64_t offp, uint64_t len, char buf)
	343	{
	344	objset_t *mos = spa->spa_meta_objset;
	345	dmu_buf_t *dbp;
	346	uint64_t read_len, phys_read_off, phys_eof;
	347	uint64_t leftover = 0;
	348	spa_history_phys_t *shpp;
	349	int err;
	350
	351	/*
6f1ffb06	352	* If the command history doesn't exist (older pool),
34dc7c2f BB	353	* that's ok, just return ENOENT.
	354	*/
	355	if (!spa->spa_history)
2e528b49	356	return (SET_ERROR(ENOENT));
34dc7c2f	357
428870ff BB	358	/*
	359	* The history is logged asynchronously, so when they request
	360	* the first chunk of history, make sure everything has been
	361	* synced to disk so that we get it.
	362	*/
	363	if (*offp == 0 && spa_writeable(spa))
	364	txg_wait_synced(spa_get_dsl(spa), 0);
	365
34dc7c2f BB	366	if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
	367	return (err);
	368	shpp = dbp->db_data;
	369
	370	#ifdef ZFS_DEBUG
	371	{
	372	dmu_object_info_t doi;
	373	dmu_object_info_from_db(dbp, &doi);
	374	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
	375	}
	376	#endif
	377
	378	mutex_enter(&spa->spa_history_lock);
	379	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
	380
	381	if (*offp < shpp->sh_pool_create_len) {
	382	/* read in just the zpool create history */
	383	phys_read_off = *offp;
	384	read_len = MIN(*len, shpp->sh_pool_create_len -
	385	phys_read_off);
	386	} else {
	387	/*
	388	* Need to reset passed in offset to BOF if the passed in
	389	* offset has since been overwritten.
	390	*/
	391	offp = MAX(offp, shpp->sh_bof);
	392	phys_read_off = spa_history_log_to_phys(*offp, shpp);
	393
	394	/*
	395	* Read up to the minimum of what the user passed down or
	396	* the EOF (physical or logical). If we hit physical EOF,
	397	* use 'leftover' to read from the physical BOF.
	398	*/
	399	if (phys_read_off <= phys_eof) {
	400	read_len = MIN(*len, phys_eof - phys_read_off);
	401	} else {
	402	read_len = MIN(*len,
	403	shpp->sh_phys_max_off - phys_read_off);
	404	if (phys_read_off + *len > shpp->sh_phys_max_off) {
	405	leftover = MIN(*len - read_len,
	406	phys_eof - shpp->sh_pool_create_len);
	407	}
	408	}
	409	}
	410
	411	/* offset for consumer to use next */
	412	*offp += read_len + leftover;
	413
	414	/* tell the consumer how much you actually read */
	415	*len = read_len + leftover;
	416
	417	if (read_len == 0) {
	418	mutex_exit(&spa->spa_history_lock);
	419	dmu_buf_rele(dbp, FTAG);
	420	return (0);
	421	}
	422
9babb374 BB	423	err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
9babb374 BB	424	DMU_READ_PREFETCH);
34dc7c2f BB	425	if (leftover && err == 0) {
34dc7c2f BB	426	err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
9babb374	427	leftover, buf + read_len, DMU_READ_PREFETCH);
34dc7c2f BB	428	}
	429	mutex_exit(&spa->spa_history_lock);
	430
	431	dmu_buf_rele(dbp, FTAG);
	432	return (err);
	433	}
	434
6f1ffb06 MA	435	/*
	436	* The nvlist will be consumed by this call.
	437	*/
45d1cae3	438	static void
6f1ffb06	439	log_internal(nvlist_t nvl, const char operation, spa_t *spa,
428870ff	440	dmu_tx_t tx, const char fmt, va_list adx)
34dc7c2f	441	{
6f1ffb06 MA	442	char *msg;
6f1ffb06 MA	443	va_list adx1;
222b9480	444	int size;
34dc7c2f	445
b128c09f BB	446	/*
	447	* If this is part of creating a pool, not everything is
	448	* initialized yet, so don't bother logging the internal events.
6f1ffb06	449	* Likewise if the pool is not writeable.
b128c09f	450	*/
6f1ffb06 MA	451	if (tx->tx_txg == TXG_INITIAL \|\| !spa_writeable(spa)) {
6f1ffb06 MA	452	fnvlist_free(nvl);
b128c09f	453	return;
6f1ffb06	454	}
b128c09f	455
6f1ffb06	456	va_copy(adx1, adx);
222b9480 BB	457	size = vsnprintf(NULL, 0, fmt, adx1) + 1;
222b9480 BB	458	msg = kmem_alloc(size, KM_PUSHPAGE);
6f1ffb06 MA	459	va_end(adx1);
	460	va_copy(adx1, adx);
	461	(void) vsprintf(msg, fmt, adx1);
	462	va_end(adx1);
	463	fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
222b9480	464	kmem_free(msg, size);
6f1ffb06 MA	465
	466	fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
	467	fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
34dc7c2f BB	468
34dc7c2f BB	469	if (dmu_tx_is_syncing(tx)) {
13fe0198	470	spa_history_log_sync(nvl, tx);
34dc7c2f	471	} else {
13fe0198 MA	472	dsl_sync_task_nowait(spa_get_dsl(spa),
13fe0198 MA	473	spa_history_log_sync, nvl, 0, tx);
34dc7c2f	474	}
6f1ffb06	475	/* spa_history_log_sync() will free nvl */
34dc7c2f	476	}
45d1cae3 BB	477
45d1cae3 BB	478	void
6f1ffb06	479	spa_history_log_internal(spa_t spa, const char operation,
428870ff	480	dmu_tx_t tx, const char fmt, ...)
45d1cae3 BB	481	{
	482	dmu_tx_t *htx = tx;
	483	va_list adx;
6f1ffb06	484	nvlist_t *nvl;
45d1cae3 BB	485
	486	/* create a tx if we didn't get one */
	487	if (tx == NULL) {
	488	htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
	489	if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
	490	dmu_tx_abort(htx);
	491	return;
	492	}
	493	}
	494
	495	va_start(adx, fmt);
6f1ffb06 MA	496	VERIFY0(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE));
6f1ffb06 MA	497	log_internal(nvl, operation, spa, htx, fmt, adx);
45d1cae3 BB	498	va_end(adx);
	499
	500	/* if we didn't get a tx from the caller, commit the one we made */
	501	if (tx == NULL)
	502	dmu_tx_commit(htx);
	503	}
	504
	505	void
6f1ffb06 MA	506	spa_history_log_internal_ds(dsl_dataset_t ds, const char operation,
	507	dmu_tx_t tx, const char fmt, ...)
	508	{
	509	va_list adx;
	510	char namebuf[MAXNAMELEN];
	511	nvlist_t *nvl;
	512
	513	ASSERT(tx != NULL);
	514
	515	dsl_dataset_name(ds, namebuf);
	516	VERIFY0(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE));
	517	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
	518	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID, ds->ds_object);
	519
	520	va_start(adx, fmt);
	521	log_internal(nvl, operation, dsl_dataset_get_spa(ds), tx, fmt, adx);
	522	va_end(adx);
	523	}
	524
	525	void
	526	spa_history_log_internal_dd(dsl_dir_t dd, const char operation,
	527	dmu_tx_t tx, const char fmt, ...)
	528	{
	529	va_list adx;
	530	char namebuf[MAXNAMELEN];
	531	nvlist_t *nvl;
	532
	533	ASSERT(tx != NULL);
	534
	535	dsl_dir_name(dd, namebuf);
	536	VERIFY0(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE));
	537	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
	538	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
	539	dd->dd_phys->dd_head_dataset_obj);
	540
	541	va_start(adx, fmt);
	542	log_internal(nvl, operation, dd->dd_pool->dp_spa, tx, fmt, adx);
	543	va_end(adx);
	544	}
	545
	546	void
	547	spa_history_log_version(spa_t spa, const char operation)
45d1cae3	548	{
6f1ffb06 MA	549	spa_history_log_internal(spa, operation, NULL,
6f1ffb06 MA	550	"pool version %llu; software version %llu/%d; uts %s %s %s %s",
13fe0198	551	(u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
6f1ffb06 MA	552	utsname.nodename, utsname.release, utsname.version,
6f1ffb06 MA	553	utsname.machine);
45d1cae3	554	}
c28b2279 BB	555
	556	#if defined(_KERNEL) && defined(HAVE_SPL)
	557	EXPORT_SYMBOL(spa_history_create_obj);
	558	EXPORT_SYMBOL(spa_history_get);
	559	EXPORT_SYMBOL(spa_history_log);
	560	EXPORT_SYMBOL(spa_history_log_internal);
	561	EXPORT_SYMBOL(spa_history_log_version);
	562	#endif