[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.
 */

#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/utsname.h>
#include <sys/cmn_err.h>
#include <sys/sunddi.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
	 * 1% of pool size, with a max of 32MB and min of 128KB.
	 */
	shpp->sh_phys_max_off =
	    metaslab_class_get_dspace(spa_normal_class(spa)) / 100;
	shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 32<<20);
	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 ("global");
#endif
}

/*
 * Write out a history event.
 */
/*ARGSUSED*/
static void
spa_history_log_sync(void *arg1, void *arg2, dmu_tx_t *tx)
{
	spa_t		*spa = arg1;
	history_arg_t	*hap = arg2;
	const char	*history_str = hap->ha_history_str;
	objset_t	*mos = spa->spa_meta_objset;
	dmu_buf_t	*dbp;
	spa_history_phys_t *shpp;
	size_t		reclen;
	uint64_t	le_len;
	nvlist_t	*nvrecord;
	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.
	 */
	VERIFY(0 == 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

	VERIFY(nvlist_alloc(&nvrecord, NV_UNIQUE_NAME, KM_SLEEP) == 0);
	VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TIME,
	    gethrestime_sec()) == 0);
	VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_WHO, hap->ha_uid) == 0);
	if (hap->ha_zone != NULL)
		VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_ZONE,
		    hap->ha_zone) == 0);
#ifdef _KERNEL
	VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_HOST,
	    utsname.nodename) == 0);
#endif
	if (hap->ha_log_type == LOG_CMD_POOL_CREATE ||
	    hap->ha_log_type == LOG_CMD_NORMAL) {
		VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_CMD,
		    history_str) == 0);

		zfs_dbgmsg("command: %s", history_str);
	} else {
		VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_INT_EVENT,
		    hap->ha_event) == 0);
		VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TXG,
		    tx->tx_txg) == 0);
		VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_INT_STR,
		    history_str) == 0);

		zfs_dbgmsg("internal %s pool:%s txg:%llu %s",
		    zfs_history_event_names[hap->ha_event], spa_name(spa),
		    (longlong_t)tx->tx_txg, history_str);

	}

	VERIFY(nvlist_size(nvrecord, &reclen, NV_ENCODE_XDR) == 0);
	record_packed = kmem_alloc(reclen, KM_SLEEP);

	VERIFY(nvlist_pack(nvrecord, &record_packed, &reclen,
	    NV_ENCODE_XDR, KM_SLEEP) == 0);

	mutex_enter(&spa->spa_history_lock);
	if (hap->ha_log_type == LOG_CMD_POOL_CREATE)
		VERIFY(shpp->sh_eof == shpp->sh_pool_create_len);

	/* 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);

	if (!ret && hap->ha_log_type == LOG_CMD_POOL_CREATE) {
		shpp->sh_pool_create_len += sizeof (le_len) + reclen;
		shpp->sh_bof = shpp->sh_pool_create_len;
	}

	mutex_exit(&spa->spa_history_lock);
	nvlist_free(nvrecord);
	kmem_free(record_packed, reclen);
	dmu_buf_rele(dbp, FTAG);

	strfree(hap->ha_history_str);
	if (hap->ha_zone != NULL)
		strfree(hap->ha_zone);
	kmem_free(hap, sizeof (history_arg_t));
}

/*
 * Write out a history event.
 */
int
spa_history_log(spa_t *spa, const char *history_str, history_log_type_t what)
{
	history_arg_t *ha;
	int err = 0;
	dmu_tx_t *tx;

	ASSERT(what != LOG_INTERNAL);

	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);
	}

	ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
	ha->ha_history_str = strdup(history_str);
	ha->ha_zone = strdup(spa_history_zone());
	ha->ha_log_type = what;
	ha->ha_uid = crgetuid(CRED());

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

	/* spa_history_log_sync will free ha and strings */
	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 (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);
}

static void
log_internal(history_internal_events_t event, spa_t *spa,
    dmu_tx_t *tx, const char *fmt, va_list adx)
{
	history_arg_t *ha;
	va_list adx_copy;

	/*
	 * If this is part of creating a pool, not everything is
	 * initialized yet, so don't bother logging the internal events.
	 */
	if (tx->tx_txg == TXG_INITIAL)
		return;

	ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
	va_copy(adx_copy, adx);
	ha->ha_history_str = kmem_vasprintf(fmt, adx_copy);
	va_end(adx_copy);
	ha->ha_log_type = LOG_INTERNAL;
	ha->ha_event = event;
	ha->ha_zone = NULL;
	ha->ha_uid = 0;

	if (dmu_tx_is_syncing(tx)) {
		spa_history_log_sync(spa, ha, tx);
	} else {
		dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
		    spa_history_log_sync, spa, ha, 0, tx);
	}
	/* spa_history_log_sync() will free ha and strings */
}

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

	/* 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);
	log_internal(event, 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_version(spa_t *spa, history_internal_events_t event)
{
#ifdef _KERNEL
	uint64_t current_vers = spa_version(spa);

	if (current_vers >= SPA_VERSION_ZPOOL_HISTORY) {
		spa_history_log_internal(event, spa, NULL,
		    "pool spa %llu; zfs spa %llu; zpl %d; uts %s %s %s %s",
		    (u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
		    utsname.nodename, utsname.release, utsname.version,
		    utsname.machine);
	}
	cmn_err(CE_CONT, "!%s version %llu pool %s using %llu",
	    event == LOG_POOL_IMPORT ? "imported" :
	    event == LOG_POOL_CREATE ? "created" : "accessed",
	    (u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);
#endif
}

#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.
34dc7c2f BB	24	*/
34dc7c2f BB	25
34dc7c2f BB	26	#include <sys/spa.h>
	27	#include <sys/spa_impl.h>
	28	#include <sys/zap.h>
	29	#include <sys/dsl_synctask.h>
	30	#include <sys/dmu_tx.h>
	31	#include <sys/dmu_objset.h>
	32	#include <sys/utsname.h>
	33	#include <sys/cmn_err.h>
	34	#include <sys/sunddi.h>
428870ff	35	#include "zfs_comutil.h"
34dc7c2f BB	36	#ifdef _KERNEL
	37	#include <sys/zone.h>
	38	#endif
	39
	40	/*
	41	* Routines to manage the on-disk history log.
	42	*
	43	* The history log is stored as a dmu object containing
	44	* <packed record length, record nvlist> tuples.
	45	*
	46	* Where "record nvlist" is a nvlist containing uint64_ts and strings, and
	47	* "packed record length" is the packed length of the "record nvlist" stored
	48	* as a little endian uint64_t.
	49	*
	50	* The log is implemented as a ring buffer, though the original creation
	51	* of the pool ('zpool create') is never overwritten.
	52	*
	53	* The history log is tracked as object 'spa_t::spa_history'. The bonus buffer
	54	* of 'spa_history' stores the offsets for logging/retrieving history as
	55	* 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of
	56	* where the 'zpool create' record is stored. This allows us to never
	57	* overwrite the original creation of the pool. 'sh_phys_max_off' is the
	58	* physical ending offset in bytes of the log. This tells you the length of
	59	* the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record
	60	* is added, 'sh_eof' is incremented by the the size of the record.
	61	* 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes).
	62	* This is where the consumer should start reading from after reading in
	63	* the 'zpool create' portion of the log.
	64	*
	65	* 'sh_records_lost' keeps track of how many records have been overwritten
	66	* and permanently lost.
	67	*/
	68
	69	/* convert a logical offset to physical */
	70	static uint64_t
	71	spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
	72	{
	73	uint64_t phys_len;
	74
	75	phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len;
	76	return ((log_off - shpp->sh_pool_create_len) % phys_len
	77	+ shpp->sh_pool_create_len);
	78	}
	79
	80	void
	81	spa_history_create_obj(spa_t spa, dmu_tx_t tx)
	82	{
	83	dmu_buf_t *dbp;
	84	spa_history_phys_t *shpp;
	85	objset_t *mos = spa->spa_meta_objset;
	86
	87	ASSERT(spa->spa_history == 0);
	88	spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY,
	89	SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS,
	90	sizeof (spa_history_phys_t), tx);
	91
	92	VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
	93	DMU_POOL_HISTORY, sizeof (uint64_t), 1,
	94	&spa->spa_history, tx) == 0);
	95
	96	VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
	97	ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));
	98
	99	shpp = dbp->db_data;
100	dmu_buf_will_dirty(dbp, tx);
101
102	/*
103	* Figure out maximum size of history log. We set it at
104	* 1% of pool size, with a max of 32MB and min of 128KB.
105	*/
428870ff BB	106	shpp->sh_phys_max_off =
428870ff BB	107	metaslab_class_get_dspace(spa_normal_class(spa)) / 100;
34dc7c2f BB	108	shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 32<<20);
	109	shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10);
	110
	111	dmu_buf_rele(dbp, FTAG);
	112	}
	113
	114	/*
	115	* Change 'sh_bof' to the beginning of the next record.
	116	*/
	117	static int
	118	spa_history_advance_bof(spa_t spa, spa_history_phys_t shpp)
	119	{
	120	objset_t *mos = spa->spa_meta_objset;
	121	uint64_t firstread, reclen, phys_bof;
	122	char buf[sizeof (reclen)];
	123	int err;
	124
	125	phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
	126	firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);
	127
	128	if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
9babb374	129	buf, DMU_READ_PREFETCH)) != 0)
34dc7c2f BB	130	return (err);
	131	if (firstread != sizeof (reclen)) {
	132	if ((err = dmu_read(mos, spa->spa_history,
	133	shpp->sh_pool_create_len, sizeof (reclen) - firstread,
9babb374	134	buf + firstread, DMU_READ_PREFETCH)) != 0)
34dc7c2f BB	135	return (err);
	136	}
	137
	138	reclen = LE_64(((uint64_t )buf));
	139	shpp->sh_bof += reclen + sizeof (reclen);
	140	shpp->sh_records_lost++;
	141	return (0);
	142	}
	143
	144	static int
	145	spa_history_write(spa_t spa, void buf, uint64_t len, spa_history_phys_t *shpp,
	146	dmu_tx_t *tx)
	147	{
	148	uint64_t firstwrite, phys_eof;
	149	objset_t *mos = spa->spa_meta_objset;
	150	int err;
	151
	152	ASSERT(MUTEX_HELD(&spa->spa_history_lock));
	153
	154	/* see if we need to reset logical BOF */
	155	while (shpp->sh_phys_max_off - shpp->sh_pool_create_len -
	156	(shpp->sh_eof - shpp->sh_bof) <= len) {
	157	if ((err = spa_history_advance_bof(spa, shpp)) != 0) {
	158	return (err);
	159	}
	160	}
	161
	162	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
	163	firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
	164	shpp->sh_eof += len;
	165	dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);
	166
	167	len -= firstwrite;
	168	if (len > 0) {
	169	/* write out the rest at the beginning of physical file */
	170	dmu_write(mos, spa->spa_history, shpp->sh_pool_create_len,
	171	len, (char *)buf + firstwrite, tx);
	172	}
	173
	174	return (0);
	175	}
	176
	177	static char *
0bc8fd78	178	spa_history_zone(void)
34dc7c2f BB	179	{
34dc7c2f BB	180	#ifdef _KERNEL
c28b2279 BB	181	#ifdef HAVE_SPL
	182	return ("linux");
	183	#else
34dc7c2f	184	return (curproc->p_zone->zone_name);
c28b2279	185	#endif
34dc7c2f BB	186	#else
	187	return ("global");
	188	#endif
	189	}
	190
	191	/*
	192	* Write out a history event.
	193	*/
428870ff	194	/ARGSUSED/
34dc7c2f	195	static void
428870ff	196	spa_history_log_sync(void arg1, void arg2, dmu_tx_t *tx)
34dc7c2f BB	197	{
	198	spa_t *spa = arg1;
	199	history_arg_t *hap = arg2;
	200	const char *history_str = hap->ha_history_str;
	201	objset_t *mos = spa->spa_meta_objset;
	202	dmu_buf_t *dbp;
	203	spa_history_phys_t *shpp;
	204	size_t reclen;
	205	uint64_t le_len;
	206	nvlist_t *nvrecord;
	207	char *record_packed = NULL;
	208	int ret;
	209
	210	/*
	211	* If we have an older pool that doesn't have a command
	212	* history object, create it now.
	213	*/
	214	mutex_enter(&spa->spa_history_lock);
	215	if (!spa->spa_history)
	216	spa_history_create_obj(spa, tx);
	217	mutex_exit(&spa->spa_history_lock);
	218
	219	/*
	220	* Get the offset of where we need to write via the bonus buffer.
	221	* Update the offset when the write completes.
	222	*/
	223	VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
	224	shpp = dbp->db_data;
	225
	226	dmu_buf_will_dirty(dbp, tx);
	227
	228	#ifdef ZFS_DEBUG
	229	{
	230	dmu_object_info_t doi;
	231	dmu_object_info_from_db(dbp, &doi);
	232	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
	233	}
	234	#endif
	235
	236	VERIFY(nvlist_alloc(&nvrecord, NV_UNIQUE_NAME, KM_SLEEP) == 0);
	237	VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TIME,
	238	gethrestime_sec()) == 0);
428870ff BB	239	VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_WHO, hap->ha_uid) == 0);
428870ff BB	240	if (hap->ha_zone != NULL)
34dc7c2f BB	241	VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_ZONE,
	242	hap->ha_zone) == 0);
	243	#ifdef _KERNEL
	244	VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_HOST,
	245	utsname.nodename) == 0);
	246	#endif
	247	if (hap->ha_log_type == LOG_CMD_POOL_CREATE \|\|
	248	hap->ha_log_type == LOG_CMD_NORMAL) {
	249	VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_CMD,
	250	history_str) == 0);
428870ff BB	251
428870ff BB	252	zfs_dbgmsg("command: %s", history_str);
34dc7c2f BB	253	} else {
	254	VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_INT_EVENT,
	255	hap->ha_event) == 0);
	256	VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TXG,
	257	tx->tx_txg) == 0);
	258	VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_INT_STR,
	259	history_str) == 0);
428870ff BB	260
	261	zfs_dbgmsg("internal %s pool:%s txg:%llu %s",
	262	zfs_history_event_names[hap->ha_event], spa_name(spa),
	263	(longlong_t)tx->tx_txg, history_str);
	264
34dc7c2f BB	265	}
	266
	267	VERIFY(nvlist_size(nvrecord, &reclen, NV_ENCODE_XDR) == 0);
	268	record_packed = kmem_alloc(reclen, KM_SLEEP);
	269
	270	VERIFY(nvlist_pack(nvrecord, &record_packed, &reclen,
	271	NV_ENCODE_XDR, KM_SLEEP) == 0);
	272
	273	mutex_enter(&spa->spa_history_lock);
	274	if (hap->ha_log_type == LOG_CMD_POOL_CREATE)
	275	VERIFY(shpp->sh_eof == shpp->sh_pool_create_len);
	276
	277	/* write out the packed length as little endian */
	278	le_len = LE_64((uint64_t)reclen);
	279	ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
	280	if (!ret)
	281	ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
	282
	283	if (!ret && hap->ha_log_type == LOG_CMD_POOL_CREATE) {
	284	shpp->sh_pool_create_len += sizeof (le_len) + reclen;
	285	shpp->sh_bof = shpp->sh_pool_create_len;
	286	}
	287
	288	mutex_exit(&spa->spa_history_lock);
	289	nvlist_free(nvrecord);
	290	kmem_free(record_packed, reclen);
	291	dmu_buf_rele(dbp, FTAG);
	292
428870ff BB	293	strfree(hap->ha_history_str);
	294	if (hap->ha_zone != NULL)
	295	strfree(hap->ha_zone);
	296	kmem_free(hap, sizeof (history_arg_t));
34dc7c2f BB	297	}
	298
	299	/*
	300	* Write out a history event.
	301	*/
	302	int
	303	spa_history_log(spa_t spa, const char history_str, history_log_type_t what)
	304	{
428870ff BB	305	history_arg_t *ha;
	306	int err = 0;
	307	dmu_tx_t *tx;
34dc7c2f BB	308
	309	ASSERT(what != LOG_INTERNAL);
	310
428870ff BB	311	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
	312	err = dmu_tx_assign(tx, TXG_WAIT);
	313	if (err) {
	314	dmu_tx_abort(tx);
	315	return (err);
	316	}
	317
	318	ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
	319	ha->ha_history_str = strdup(history_str);
	320	ha->ha_zone = strdup(spa_history_zone());
	321	ha->ha_log_type = what;
	322	ha->ha_uid = crgetuid(CRED());
	323
	324	/* Kick this off asynchronously; errors are ignored. */
	325	dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
	326	spa_history_log_sync, spa, ha, 0, tx);
	327	dmu_tx_commit(tx);
	328
	329	/* spa_history_log_sync will free ha and strings */
	330	return (err);
34dc7c2f BB	331	}
	332
	333	/*
	334	* Read out the command history.
	335	*/
	336	int
	337	spa_history_get(spa_t spa, uint64_t offp, uint64_t len, char buf)
	338	{
	339	objset_t *mos = spa->spa_meta_objset;
	340	dmu_buf_t *dbp;
	341	uint64_t read_len, phys_read_off, phys_eof;
	342	uint64_t leftover = 0;
	343	spa_history_phys_t *shpp;
	344	int err;
	345
	346	/*
	347	* If the command history doesn't exist (older pool),
	348	* that's ok, just return ENOENT.
	349	*/
	350	if (!spa->spa_history)
	351	return (ENOENT);
	352
428870ff BB	353	/*
	354	* The history is logged asynchronously, so when they request
	355	* the first chunk of history, make sure everything has been
	356	* synced to disk so that we get it.
	357	*/
	358	if (*offp == 0 && spa_writeable(spa))
	359	txg_wait_synced(spa_get_dsl(spa), 0);
	360
34dc7c2f BB	361	if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
	362	return (err);
	363	shpp = dbp->db_data;
	364
	365	#ifdef ZFS_DEBUG
	366	{
	367	dmu_object_info_t doi;
	368	dmu_object_info_from_db(dbp, &doi);
	369	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
	370	}
	371	#endif
	372
	373	mutex_enter(&spa->spa_history_lock);
	374	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
	375
	376	if (*offp < shpp->sh_pool_create_len) {
	377	/* read in just the zpool create history */
	378	phys_read_off = *offp;
	379	read_len = MIN(*len, shpp->sh_pool_create_len -
	380	phys_read_off);
	381	} else {
	382	/*
	383	* Need to reset passed in offset to BOF if the passed in
	384	* offset has since been overwritten.
	385	*/
	386	offp = MAX(offp, shpp->sh_bof);
	387	phys_read_off = spa_history_log_to_phys(*offp, shpp);
	388
	389	/*
	390	* Read up to the minimum of what the user passed down or
	391	* the EOF (physical or logical). If we hit physical EOF,
	392	* use 'leftover' to read from the physical BOF.
	393	*/
	394	if (phys_read_off <= phys_eof) {
	395	read_len = MIN(*len, phys_eof - phys_read_off);
	396	} else {
	397	read_len = MIN(*len,
	398	shpp->sh_phys_max_off - phys_read_off);
	399	if (phys_read_off + *len > shpp->sh_phys_max_off) {
	400	leftover = MIN(*len - read_len,
	401	phys_eof - shpp->sh_pool_create_len);
	402	}
	403	}
	404	}
	405
	406	/* offset for consumer to use next */
	407	*offp += read_len + leftover;
	408
	409	/* tell the consumer how much you actually read */
	410	*len = read_len + leftover;
	411
	412	if (read_len == 0) {
	413	mutex_exit(&spa->spa_history_lock);
	414	dmu_buf_rele(dbp, FTAG);
	415	return (0);
	416	}
	417
9babb374 BB	418	err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
9babb374 BB	419	DMU_READ_PREFETCH);
34dc7c2f BB	420	if (leftover && err == 0) {
34dc7c2f BB	421	err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
9babb374	422	leftover, buf + read_len, DMU_READ_PREFETCH);
34dc7c2f BB	423	}
	424	mutex_exit(&spa->spa_history_lock);
	425
	426	dmu_buf_rele(dbp, FTAG);
	427	return (err);
	428	}
	429
45d1cae3 BB	430	static void
45d1cae3 BB	431	log_internal(history_internal_events_t event, spa_t *spa,
428870ff	432	dmu_tx_t tx, const char fmt, va_list adx)
34dc7c2f	433	{
428870ff	434	history_arg_t *ha;
00b46022	435	va_list adx_copy;
34dc7c2f	436
b128c09f BB	437	/*
	438	* If this is part of creating a pool, not everything is
	439	* initialized yet, so don't bother logging the internal events.
	440	*/
	441	if (tx->tx_txg == TXG_INITIAL)
	442	return;
	443
428870ff	444	ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
00b46022 BB	445	va_copy(adx_copy, adx);
	446	ha->ha_history_str = kmem_vasprintf(fmt, adx_copy);
	447	va_end(adx_copy);
428870ff BB	448	ha->ha_log_type = LOG_INTERNAL;
	449	ha->ha_event = event;
	450	ha->ha_zone = NULL;
	451	ha->ha_uid = 0;
34dc7c2f BB	452
34dc7c2f BB	453	if (dmu_tx_is_syncing(tx)) {
428870ff	454	spa_history_log_sync(spa, ha, tx);
34dc7c2f BB	455	} else {
34dc7c2f BB	456	dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
428870ff	457	spa_history_log_sync, spa, ha, 0, tx);
34dc7c2f	458	}
428870ff	459	/* spa_history_log_sync() will free ha and strings */
34dc7c2f	460	}
45d1cae3 BB	461
45d1cae3 BB	462	void
428870ff BB	463	spa_history_log_internal(history_internal_events_t event, spa_t *spa,
428870ff BB	464	dmu_tx_t tx, const char fmt, ...)
45d1cae3 BB	465	{
	466	dmu_tx_t *htx = tx;
	467	va_list adx;
	468
	469	/* create a tx if we didn't get one */
	470	if (tx == NULL) {
	471	htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
	472	if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
	473	dmu_tx_abort(htx);
	474	return;
	475	}
	476	}
	477
	478	va_start(adx, fmt);
428870ff	479	log_internal(event, spa, htx, fmt, adx);
45d1cae3 BB	480	va_end(adx);
	481
	482	/* if we didn't get a tx from the caller, commit the one we made */
	483	if (tx == NULL)
	484	dmu_tx_commit(htx);
	485	}
	486
	487	void
	488	spa_history_log_version(spa_t *spa, history_internal_events_t event)
	489	{
	490	#ifdef _KERNEL
	491	uint64_t current_vers = spa_version(spa);
	492
	493	if (current_vers >= SPA_VERSION_ZPOOL_HISTORY) {
428870ff	494	spa_history_log_internal(event, spa, NULL,
45d1cae3 BB	495	"pool spa %llu; zfs spa %llu; zpl %d; uts %s %s %s %s",
	496	(u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
	497	utsname.nodename, utsname.release, utsname.version,
	498	utsname.machine);
	499	}
	500	cmn_err(CE_CONT, "!%s version %llu pool %s using %llu",
	501	event == LOG_POOL_IMPORT ? "imported" :
	502	event == LOG_POOL_CREATE ? "created" : "accessed",
	503	(u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);
	504	#endif
	505	}
c28b2279 BB	506
	507	#if defined(_KERNEL) && defined(HAVE_SPL)
	508	EXPORT_SYMBOL(spa_history_create_obj);
	509	EXPORT_SYMBOL(spa_history_get);
	510	EXPORT_SYMBOL(spa_history_log);
	511	EXPORT_SYMBOL(spa_history_log_internal);
	512	EXPORT_SYMBOL(spa_history_log_version);
	513	#endif