[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/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());
	fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname()->nodename);

	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_SLEEP), ==, 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 = fnvlist_alloc();

	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_SLEEP));
	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;

	/*
	 * 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;
	}

	msg = kmem_vasprintf(fmt, adx);
	fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
	strfree(msg);

	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;

	/* 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(fnvlist_alloc(), 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 = fnvlist_alloc();

	ASSERT(tx != NULL);

	dsl_dataset_name(ds, namebuf);
	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 = fnvlist_alloc();

	ASSERT(tx != NULL);

	dsl_dir_name(dd, namebuf);
	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
	    dsl_dir_phys(dd)->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)
{
	utsname_t *u = utsname();

	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,
	    u->nodename, u->release, u->version, u->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/cmn_err.h>
34dc7c2f BB	36	#include <sys/sunddi.h>
6f1ffb06	37	#include <sys/cred.h>
428870ff	38	#include "zfs_comutil.h"
34dc7c2f BB	39	#ifdef _KERNEL
	40	#include <sys/zone.h>
	41	#endif
	42
	43	/*
	44	* Routines to manage the on-disk history log.
	45	*
	46	* The history log is stored as a dmu object containing
	47	* <packed record length, record nvlist> tuples.
	48	*
	49	* Where "record nvlist" is a nvlist containing uint64_ts and strings, and
	50	* "packed record length" is the packed length of the "record nvlist" stored
	51	* as a little endian uint64_t.
	52	*
	53	* The log is implemented as a ring buffer, though the original creation
	54	* of the pool ('zpool create') is never overwritten.
	55	*
	56	* The history log is tracked as object 'spa_t::spa_history'. The bonus buffer
	57	* of 'spa_history' stores the offsets for logging/retrieving history as
	58	* 'spa_history_phys_t'. 'sh_pool_create_len' is the ending offset in bytes of
	59	* where the 'zpool create' record is stored. This allows us to never
	60	* overwrite the original creation of the pool. 'sh_phys_max_off' is the
	61	* physical ending offset in bytes of the log. This tells you the length of
	62	* the buffer. 'sh_eof' is the logical EOF (in bytes). Whenever a record
	63	* is added, 'sh_eof' is incremented by the the size of the record.
	64	* 'sh_eof' is never decremented. 'sh_bof' is the logical BOF (in bytes).
	65	* This is where the consumer should start reading from after reading in
	66	* the 'zpool create' portion of the log.
	67	*
	68	* 'sh_records_lost' keeps track of how many records have been overwritten
	69	* and permanently lost.
	70	*/
	71
	72	/* convert a logical offset to physical */
	73	static uint64_t
	74	spa_history_log_to_phys(uint64_t log_off, spa_history_phys_t *shpp)
	75	{
	76	uint64_t phys_len;
	77
	78	phys_len = shpp->sh_phys_max_off - shpp->sh_pool_create_len;
	79	return ((log_off - shpp->sh_pool_create_len) % phys_len
	80	+ shpp->sh_pool_create_len);
	81	}
	82
	83	void
	84	spa_history_create_obj(spa_t spa, dmu_tx_t tx)
	85	{
	86	dmu_buf_t *dbp;
	87	spa_history_phys_t *shpp;
	88	objset_t *mos = spa->spa_meta_objset;
	89
	90	ASSERT(spa->spa_history == 0);
	91	spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY,
	92	SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS,
	93	sizeof (spa_history_phys_t), tx);
	94
	95	VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
	96	DMU_POOL_HISTORY, sizeof (uint64_t), 1,
	97	&spa->spa_history, tx) == 0);
	98
	99	VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
	100	ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));
	101
	102	shpp = dbp->db_data;
103	dmu_buf_will_dirty(dbp, tx);
104
105	/*
106	* Figure out maximum size of history log. We set it at
330d06f9	107	* 0.1% of pool size, with a max of 1G and min of 128KB.
34dc7c2f	108	*/
428870ff	109	shpp->sh_phys_max_off =
330d06f9 MA	110	metaslab_class_get_dspace(spa_normal_class(spa)) / 1000;
330d06f9 MA	111	shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 1<<30);
34dc7c2f BB	112	shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10);
	113
	114	dmu_buf_rele(dbp, FTAG);
	115	}
	116
	117	/*
	118	* Change 'sh_bof' to the beginning of the next record.
	119	*/
	120	static int
	121	spa_history_advance_bof(spa_t spa, spa_history_phys_t shpp)
	122	{
	123	objset_t *mos = spa->spa_meta_objset;
	124	uint64_t firstread, reclen, phys_bof;
	125	char buf[sizeof (reclen)];
	126	int err;
	127
	128	phys_bof = spa_history_log_to_phys(shpp->sh_bof, shpp);
	129	firstread = MIN(sizeof (reclen), shpp->sh_phys_max_off - phys_bof);
	130
	131	if ((err = dmu_read(mos, spa->spa_history, phys_bof, firstread,
9babb374	132	buf, DMU_READ_PREFETCH)) != 0)
34dc7c2f BB	133	return (err);
	134	if (firstread != sizeof (reclen)) {
	135	if ((err = dmu_read(mos, spa->spa_history,
	136	shpp->sh_pool_create_len, sizeof (reclen) - firstread,
9babb374	137	buf + firstread, DMU_READ_PREFETCH)) != 0)
34dc7c2f BB	138	return (err);
	139	}
	140
	141	reclen = LE_64(((uint64_t )buf));
	142	shpp->sh_bof += reclen + sizeof (reclen);
	143	shpp->sh_records_lost++;
	144	return (0);
	145	}
	146
	147	static int
	148	spa_history_write(spa_t spa, void buf, uint64_t len, spa_history_phys_t *shpp,
	149	dmu_tx_t *tx)
	150	{
	151	uint64_t firstwrite, phys_eof;
	152	objset_t *mos = spa->spa_meta_objset;
	153	int err;
	154
	155	ASSERT(MUTEX_HELD(&spa->spa_history_lock));
	156
	157	/* see if we need to reset logical BOF */
	158	while (shpp->sh_phys_max_off - shpp->sh_pool_create_len -
	159	(shpp->sh_eof - shpp->sh_bof) <= len) {
	160	if ((err = spa_history_advance_bof(spa, shpp)) != 0) {
	161	return (err);
	162	}
	163	}
	164
	165	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
	166	firstwrite = MIN(len, shpp->sh_phys_max_off - phys_eof);
	167	shpp->sh_eof += len;
	168	dmu_write(mos, spa->spa_history, phys_eof, firstwrite, buf, tx);
	169
	170	len -= firstwrite;
	171	if (len > 0) {
	172	/* write out the rest at the beginning of physical file */
	173	dmu_write(mos, spa->spa_history, shpp->sh_pool_create_len,
	174	len, (char *)buf + firstwrite, tx);
	175	}
	176
	177	return (0);
	178	}
	179
	180	static char *
0bc8fd78	181	spa_history_zone(void)
34dc7c2f BB	182	{
34dc7c2f BB	183	#ifdef _KERNEL
c28b2279 BB	184	#ifdef HAVE_SPL
	185	return ("linux");
	186	#else
34dc7c2f	187	return (curproc->p_zone->zone_name);
c28b2279	188	#endif
34dc7c2f	189	#else
6f1ffb06	190	return (NULL);
34dc7c2f BB	191	#endif
	192	}
	193
	194	/*
	195	* Write out a history event.
	196	*/
428870ff	197	/ARGSUSED/
34dc7c2f	198	static void
13fe0198	199	spa_history_log_sync(void arg, dmu_tx_t tx)
34dc7c2f	200	{
13fe0198 MA	201	nvlist_t *nvl = arg;
13fe0198 MA	202	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
34dc7c2f BB	203	objset_t *mos = spa->spa_meta_objset;
	204	dmu_buf_t *dbp;
	205	spa_history_phys_t *shpp;
	206	size_t reclen;
	207	uint64_t le_len;
34dc7c2f BB	208	char *record_packed = NULL;
	209	int ret;
	210
	211	/*
	212	* If we have an older pool that doesn't have a command
	213	* history object, create it now.
	214	*/
	215	mutex_enter(&spa->spa_history_lock);
	216	if (!spa->spa_history)
	217	spa_history_create_obj(spa, tx);
	218	mutex_exit(&spa->spa_history_lock);
	219
	220	/*
	221	* Get the offset of where we need to write via the bonus buffer.
	222	* Update the offset when the write completes.
	223	*/
13fe0198	224	VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
34dc7c2f BB	225	shpp = dbp->db_data;
	226
	227	dmu_buf_will_dirty(dbp, tx);
	228
	229	#ifdef ZFS_DEBUG
	230	{
	231	dmu_object_info_t doi;
	232	dmu_object_info_from_db(dbp, &doi);
	233	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
	234	}
	235	#endif
	236
6f1ffb06	237	fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
f0e324f2 BB	238	fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname()->nodename);
f0e324f2 BB	239
6f1ffb06 MA	240	if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
	241	zfs_dbgmsg("command: %s",
	242	fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
	243	} else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
	244	if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
	245	zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
	246	fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
	247	fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
	248	fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
	249	fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
	250	fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
	251	} else {
	252	zfs_dbgmsg("txg %lld %s %s",
	253	fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
	254	fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
	255	fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
	256	}
	257	} else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
	258	zfs_dbgmsg("ioctl %s",
	259	fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
34dc7c2f BB	260	}
34dc7c2f BB	261
6f1ffb06	262	VERIFY3U(nvlist_pack(nvl, &record_packed, &reclen, NV_ENCODE_NATIVE,
79c76d5b	263	KM_SLEEP), ==, 0);
34dc7c2f BB	264
34dc7c2f BB	265	mutex_enter(&spa->spa_history_lock);
34dc7c2f BB	266
	267	/* write out the packed length as little endian */
	268	le_len = LE_64((uint64_t)reclen);
	269	ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
	270	if (!ret)
	271	ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
	272
6f1ffb06 MA	273	/* The first command is the create, which we keep forever */
	274	if (ret == 0 && shpp->sh_pool_create_len == 0 &&
	275	nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
	276	shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
34dc7c2f BB	277	}
	278
	279	mutex_exit(&spa->spa_history_lock);
6f1ffb06	280	fnvlist_pack_free(record_packed, reclen);
34dc7c2f	281	dmu_buf_rele(dbp, FTAG);
6f1ffb06	282	fnvlist_free(nvl);
34dc7c2f BB	283	}
	284
	285	/*
	286	* Write out a history event.
	287	*/
	288	int
6f1ffb06 MA	289	spa_history_log(spa_t spa, const char msg)
	290	{
	291	int err;
79c76d5b	292	nvlist_t *nvl = fnvlist_alloc();
6f1ffb06 MA	293
	294	fnvlist_add_string(nvl, ZPOOL_HIST_CMD, msg);
	295	err = spa_history_log_nvl(spa, nvl);
	296	fnvlist_free(nvl);
	297	return (err);
	298	}
	299
	300	int
	301	spa_history_log_nvl(spa_t spa, nvlist_t nvl)
34dc7c2f	302	{
428870ff BB	303	int err = 0;
428870ff BB	304	dmu_tx_t *tx;
6f1ffb06	305	nvlist_t *nvarg;
34dc7c2f	306
6f1ffb06	307	if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY \|\| !spa_writeable(spa))
2e528b49	308	return (SET_ERROR(EINVAL));
34dc7c2f	309
428870ff BB	310	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
	311	err = dmu_tx_assign(tx, TXG_WAIT);
	312	if (err) {
	313	dmu_tx_abort(tx);
	314	return (err);
	315	}
	316
79c76d5b	317	VERIFY0(nvlist_dup(nvl, &nvarg, KM_SLEEP));
6f1ffb06 MA	318	if (spa_history_zone() != NULL) {
	319	fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
	320	spa_history_zone());
	321	}
	322	fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
428870ff BB	323
428870ff BB	324	/* Kick this off asynchronously; errors are ignored. */
13fe0198 MA	325	dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
13fe0198 MA	326	nvarg, 0, tx);
428870ff BB	327	dmu_tx_commit(tx);
428870ff BB	328
6f1ffb06	329	/* spa_history_log_sync will free nvl */
428870ff	330	return (err);
6f1ffb06	331
34dc7c2f BB	332	}
	333
	334	/*
	335	* Read out the command history.
	336	*/
	337	int
	338	spa_history_get(spa_t spa, uint64_t offp, uint64_t len, char buf)
	339	{
	340	objset_t *mos = spa->spa_meta_objset;
	341	dmu_buf_t *dbp;
	342	uint64_t read_len, phys_read_off, phys_eof;
	343	uint64_t leftover = 0;
	344	spa_history_phys_t *shpp;
	345	int err;
	346
	347	/*
6f1ffb06	348	* If the command history doesn't exist (older pool),
34dc7c2f BB	349	* that's ok, just return ENOENT.
	350	*/
	351	if (!spa->spa_history)
2e528b49	352	return (SET_ERROR(ENOENT));
34dc7c2f	353
428870ff BB	354	/*
	355	* The history is logged asynchronously, so when they request
	356	* the first chunk of history, make sure everything has been
	357	* synced to disk so that we get it.
	358	*/
	359	if (*offp == 0 && spa_writeable(spa))
	360	txg_wait_synced(spa_get_dsl(spa), 0);
	361
34dc7c2f BB	362	if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
	363	return (err);
	364	shpp = dbp->db_data;
	365
	366	#ifdef ZFS_DEBUG
	367	{
	368	dmu_object_info_t doi;
	369	dmu_object_info_from_db(dbp, &doi);
	370	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
	371	}
	372	#endif
	373
	374	mutex_enter(&spa->spa_history_lock);
	375	phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
	376
	377	if (*offp < shpp->sh_pool_create_len) {
	378	/* read in just the zpool create history */
	379	phys_read_off = *offp;
	380	read_len = MIN(*len, shpp->sh_pool_create_len -
	381	phys_read_off);
	382	} else {
	383	/*
	384	* Need to reset passed in offset to BOF if the passed in
	385	* offset has since been overwritten.
	386	*/
	387	offp = MAX(offp, shpp->sh_bof);
	388	phys_read_off = spa_history_log_to_phys(*offp, shpp);
	389
	390	/*
	391	* Read up to the minimum of what the user passed down or
	392	* the EOF (physical or logical). If we hit physical EOF,
	393	* use 'leftover' to read from the physical BOF.
	394	*/
	395	if (phys_read_off <= phys_eof) {
	396	read_len = MIN(*len, phys_eof - phys_read_off);
	397	} else {
	398	read_len = MIN(*len,
	399	shpp->sh_phys_max_off - phys_read_off);
	400	if (phys_read_off + *len > shpp->sh_phys_max_off) {
	401	leftover = MIN(*len - read_len,
	402	phys_eof - shpp->sh_pool_create_len);
	403	}
	404	}
	405	}
	406
	407	/* offset for consumer to use next */
	408	*offp += read_len + leftover;
	409
	410	/* tell the consumer how much you actually read */
	411	*len = read_len + leftover;
	412
	413	if (read_len == 0) {
	414	mutex_exit(&spa->spa_history_lock);
	415	dmu_buf_rele(dbp, FTAG);
	416	return (0);
	417	}
	418
9babb374 BB	419	err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
9babb374 BB	420	DMU_READ_PREFETCH);
34dc7c2f BB	421	if (leftover && err == 0) {
34dc7c2f BB	422	err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
9babb374	423	leftover, buf + read_len, DMU_READ_PREFETCH);
34dc7c2f BB	424	}
	425	mutex_exit(&spa->spa_history_lock);
	426
	427	dmu_buf_rele(dbp, FTAG);
	428	return (err);
	429	}
	430
6f1ffb06 MA	431	/*
	432	* The nvlist will be consumed by this call.
	433	*/
45d1cae3	434	static void
6f1ffb06	435	log_internal(nvlist_t nvl, const char operation, spa_t *spa,
428870ff	436	dmu_tx_t tx, const char fmt, va_list adx)
34dc7c2f	437	{
6f1ffb06	438	char *msg;
34dc7c2f	439
b128c09f BB	440	/*
	441	* If this is part of creating a pool, not everything is
	442	* initialized yet, so don't bother logging the internal events.
6f1ffb06	443	* Likewise if the pool is not writeable.
b128c09f	444	*/
6f1ffb06 MA	445	if (tx->tx_txg == TXG_INITIAL \|\| !spa_writeable(spa)) {
6f1ffb06 MA	446	fnvlist_free(nvl);
b128c09f	447	return;
6f1ffb06	448	}
b128c09f	449
841c9d43	450	msg = kmem_vasprintf(fmt, adx);
6f1ffb06	451	fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
841c9d43	452	strfree(msg);
6f1ffb06 MA	453
	454	fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
	455	fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
34dc7c2f BB	456
34dc7c2f BB	457	if (dmu_tx_is_syncing(tx)) {
13fe0198	458	spa_history_log_sync(nvl, tx);
34dc7c2f	459	} else {
13fe0198 MA	460	dsl_sync_task_nowait(spa_get_dsl(spa),
13fe0198 MA	461	spa_history_log_sync, nvl, 0, tx);
34dc7c2f	462	}
6f1ffb06	463	/* spa_history_log_sync() will free nvl */
34dc7c2f	464	}
45d1cae3 BB	465
45d1cae3 BB	466	void
6f1ffb06	467	spa_history_log_internal(spa_t spa, const char operation,
428870ff	468	dmu_tx_t tx, const char fmt, ...)
45d1cae3 BB	469	{
	470	dmu_tx_t *htx = tx;
	471	va_list adx;
	472
	473	/* create a tx if we didn't get one */
	474	if (tx == NULL) {
	475	htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
	476	if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
	477	dmu_tx_abort(htx);
	478	return;
	479	}
	480	}
	481
	482	va_start(adx, fmt);
79c76d5b	483	log_internal(fnvlist_alloc(), operation, spa, htx, fmt, adx);
45d1cae3 BB	484	va_end(adx);
	485
	486	/* if we didn't get a tx from the caller, commit the one we made */
	487	if (tx == NULL)
	488	dmu_tx_commit(htx);
	489	}
	490
	491	void
6f1ffb06 MA	492	spa_history_log_internal_ds(dsl_dataset_t ds, const char operation,
	493	dmu_tx_t tx, const char fmt, ...)
	494	{
	495	va_list adx;
	496	char namebuf[MAXNAMELEN];
79c76d5b	497	nvlist_t *nvl = fnvlist_alloc();
6f1ffb06 MA	498
	499	ASSERT(tx != NULL);
	500
	501	dsl_dataset_name(ds, namebuf);
6f1ffb06 MA	502	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
	503	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID, ds->ds_object);
	504
	505	va_start(adx, fmt);
	506	log_internal(nvl, operation, dsl_dataset_get_spa(ds), tx, fmt, adx);
	507	va_end(adx);
	508	}
	509
	510	void
	511	spa_history_log_internal_dd(dsl_dir_t dd, const char operation,
	512	dmu_tx_t tx, const char fmt, ...)
	513	{
	514	va_list adx;
	515	char namebuf[MAXNAMELEN];
79c76d5b	516	nvlist_t *nvl = fnvlist_alloc();
6f1ffb06 MA	517
	518	ASSERT(tx != NULL);
	519
	520	dsl_dir_name(dd, namebuf);
6f1ffb06 MA	521	fnvlist_add_string(nvl, ZPOOL_HIST_DSNAME, namebuf);
6f1ffb06 MA	522	fnvlist_add_uint64(nvl, ZPOOL_HIST_DSID,
d683ddbb	523	dsl_dir_phys(dd)->dd_head_dataset_obj);
6f1ffb06 MA	524
	525	va_start(adx, fmt);
	526	log_internal(nvl, operation, dd->dd_pool->dp_spa, tx, fmt, adx);
	527	va_end(adx);
	528	}
	529
	530	void
	531	spa_history_log_version(spa_t spa, const char operation)
45d1cae3	532	{
f0e324f2 BB	533	utsname_t *u = utsname();
f0e324f2 BB	534
6f1ffb06 MA	535	spa_history_log_internal(spa, operation, NULL,
6f1ffb06 MA	536	"pool version %llu; software version %llu/%d; uts %s %s %s %s",
13fe0198	537	(u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
f0e324f2	538	u->nodename, u->release, u->version, u->machine);
45d1cae3	539	}
c28b2279 BB	540
	541	#if defined(_KERNEL) && defined(HAVE_SPL)
	542	EXPORT_SYMBOL(spa_history_create_obj);
	543	EXPORT_SYMBOL(spa_history_get);
	544	EXPORT_SYMBOL(spa_history_log);
	545	EXPORT_SYMBOL(spa_history_log_internal);
	546	EXPORT_SYMBOL(spa_history_log_version);
	547	#endif