C++ spa_get_dsl函数代码示例

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, ZFS_SPACE_CHECK_NONE, tx);
    dmu_tx_commit(tx);

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

}

/*
 * 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);
}

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) {
        if (!spa_get_dsl(spa)) return;
		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);
}

/* Takes care of physical writing and limiting # of concurrent ZIOs. */
static int
vdev_initialize_write(vdev_t *vd, uint64_t start, uint64_t size, abd_t *data)
{
	spa_t *spa = vd->vdev_spa;

	/* Limit inflight initializing I/Os */
	mutex_enter(&vd->vdev_initialize_io_lock);
	while (vd->vdev_initialize_inflight >= zfs_initialize_limit) {
		cv_wait(&vd->vdev_initialize_io_cv,
		    &vd->vdev_initialize_io_lock);
	}
	vd->vdev_initialize_inflight++;
	mutex_exit(&vd->vdev_initialize_io_lock);

	dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
	uint64_t txg = dmu_tx_get_txg(tx);

	spa_config_enter(spa, SCL_STATE_ALL, vd, RW_READER);
	mutex_enter(&vd->vdev_initialize_lock);

	if (vd->vdev_initialize_offset[txg & TXG_MASK] == 0) {
		uint64_t *guid = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
		*guid = vd->vdev_guid;

		/* This is the first write of this txg. */
		dsl_sync_task_nowait(spa_get_dsl(spa),
		    vdev_initialize_zap_update_sync, guid, 2,
		    ZFS_SPACE_CHECK_RESERVED, tx);
	}

	/*
	 * We know the vdev struct will still be around since all
	 * consumers of vdev_free must stop the initialization first.
	 */
	if (vdev_initialize_should_stop(vd)) {
		mutex_enter(&vd->vdev_initialize_io_lock);
		ASSERT3U(vd->vdev_initialize_inflight, >, 0);
		vd->vdev_initialize_inflight--;
		mutex_exit(&vd->vdev_initialize_io_lock);
		spa_config_exit(vd->vdev_spa, SCL_STATE_ALL, vd);
		mutex_exit(&vd->vdev_initialize_lock);
		dmu_tx_commit(tx);
		return (SET_ERROR(EINTR));
	}
	mutex_exit(&vd->vdev_initialize_lock);

	vd->vdev_initialize_offset[txg & TXG_MASK] = start + size;
	zio_nowait(zio_write_phys(spa->spa_txg_zio[txg & TXG_MASK], vd, start,
	    size, data, ZIO_CHECKSUM_OFF, vdev_initialize_cb, NULL,
	    ZIO_PRIORITY_INITIALIZING, ZIO_FLAG_CANFAIL, B_FALSE));
	/* vdev_initialize_cb releases SCL_STATE_ALL */

	dmu_tx_commit(tx);

	return (0);
}

static void
vdev_initialize_change_state(vdev_t *vd, vdev_initializing_state_t new_state)
{
	ASSERT(MUTEX_HELD(&vd->vdev_initialize_lock));
	spa_t *spa = vd->vdev_spa;

	if (new_state == vd->vdev_initialize_state)
		return;

	/*
	 * Copy the vd's guid, this will be freed by the sync task.
	 */
	uint64_t *guid = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
	*guid = vd->vdev_guid;

	/*
	 * If we're suspending, then preserving the original start time.
	 */
	if (vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED) {
		vd->vdev_initialize_action_time = gethrestime_sec();
	}
	vd->vdev_initialize_state = new_state;

	dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
	dsl_sync_task_nowait(spa_get_dsl(spa), vdev_initialize_zap_update_sync,
	    guid, 2, ZFS_SPACE_CHECK_RESERVED, tx);

	switch (new_state) {
	case VDEV_INITIALIZE_ACTIVE:
		spa_history_log_internal(spa, "initialize", tx,
		    "vdev=%s activated", vd->vdev_path);
		break;
	case VDEV_INITIALIZE_SUSPENDED:
		spa_history_log_internal(spa, "initialize", tx,
		    "vdev=%s suspended", vd->vdev_path);
		break;
	case VDEV_INITIALIZE_CANCELED:
		spa_history_log_internal(spa, "initialize", tx,
		    "vdev=%s canceled", vd->vdev_path);
		break;
	case VDEV_INITIALIZE_COMPLETE:
		spa_history_log_internal(spa, "initialize", tx,
		    "vdev=%s complete", vd->vdev_path);
		break;
	default:
		panic("invalid state %llu", (unsigned long long)new_state);
	}

	dmu_tx_commit(tx);
}

void
spa_history_internal_log(history_internal_events_t event, spa_t *spa,
    dmu_tx_t *tx, cred_t *cr, const char *fmt, ...)
{
	history_arg_t *hap;
	char *str;
	va_list adx;

	hap = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
	str = kmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP);

	va_start(adx, fmt);
	(void) vsnprintf(str, HIS_MAX_RECORD_LEN, fmt, adx);
	va_end(adx);

	hap->ha_log_type = LOG_INTERNAL;
	hap->ha_history_str = str;
	hap->ha_event = event;
	hap->ha_zone[0] = '\0';

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

/*
 * Open-context function to add one entry to the new mapping.  The new
 * entry will be remembered and written from syncing context.
 */
static void
spa_condense_indirect_commit_entry(spa_t *spa,
    vdev_indirect_mapping_entry_phys_t *vimep, uint32_t count)
{
	spa_condensing_indirect_t *sci = spa->spa_condensing_indirect;

	ASSERT3U(count, <, DVA_GET_ASIZE(&vimep->vimep_dst));

	dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
	dmu_tx_hold_space(tx, sizeof (*vimep) + sizeof (count));
	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
	int txgoff = dmu_tx_get_txg(tx) & TXG_MASK;

	/*
	 * If we are the first entry committed this txg, kick off the sync
	 * task to write to the MOS on our behalf.
	 */
	if (list_is_empty(&sci->sci_new_mapping_entries[txgoff])) {
		dsl_sync_task_nowait(dmu_tx_pool(tx),
		    spa_condense_indirect_commit_sync, sci,
		    0, ZFS_SPACE_CHECK_NONE, tx);
	}

	vdev_indirect_mapping_entry_t *vime =
	    kmem_alloc(sizeof (*vime), KM_SLEEP);
	vime->vime_mapping = *vimep;
	vime->vime_obsolete_count = count;
	list_insert_tail(&sci->sci_new_mapping_entries[txgoff], vime);

	dmu_tx_commit(tx);
}

static void
dsl_dataset_user_release_onexit(void *arg)
{
	zfs_hold_cleanup_arg_t *ca = arg;
	spa_t *spa;
	int error;

	error = spa_open(ca->zhca_spaname, &spa, FTAG);
	if (error != 0) {
		zfs_dbgmsg("couldn't release holds on pool=%s "
		    "because pool is no longer loaded",
		    ca->zhca_spaname);
		return;
	}
	if (spa_load_guid(spa) != ca->zhca_spa_load_guid) {
		zfs_dbgmsg("couldn't release holds on pool=%s "
		    "because pool is no longer loaded (guid doesn't match)",
		    ca->zhca_spaname);
		spa_close(spa, FTAG);
		return;
	}

	(void) dsl_dataset_user_release_tmp(spa_get_dsl(spa), ca->zhca_holds);
	fnvlist_free(ca->zhca_holds);
	kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
	spa_close(spa, FTAG);
}

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 */
}

/*
 * 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, ZFS_SPACE_CHECK_NONE, tx);
    }
    /* spa_history_log_sync() will free nvl */
}

int
dmu_objset_snapshot(char *fsname, char *snapname, boolean_t recursive)
{
	dsl_sync_task_t *dst;
	struct osnode *osn;
	struct snaparg sn = { 0 };
	spa_t *spa;
	int err;

	(void) strcpy(sn.failed, fsname);

	err = spa_open(fsname, &spa, FTAG);
	if (err)
		return (err);

	sn.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
	sn.snapname = snapname;
	list_create(&sn.objsets, sizeof (struct osnode),
	    offsetof(struct osnode, node));

	if (recursive) {
		sn.checkperms = B_TRUE;
		err = dmu_objset_find(fsname,
		    dmu_objset_snapshot_one, &sn, DS_FIND_CHILDREN);
	} else {
		sn.checkperms = B_FALSE;
		err = dmu_objset_snapshot_one(fsname, &sn);
	}

	if (err)
		goto out;

	err = dsl_sync_task_group_wait(sn.dstg);

	for (dst = list_head(&sn.dstg->dstg_tasks); dst;
	    dst = list_next(&sn.dstg->dstg_tasks, dst)) {
		dsl_dataset_t *ds = dst->dst_arg1;
		if (dst->dst_err)
			dsl_dataset_name(ds, sn.failed);
	}

out:
	while (osn = list_head(&sn.objsets)) {
		list_remove(&sn.objsets, osn);
		zil_resume(dmu_objset_zil(osn->os));
		dmu_objset_close(osn->os);
		kmem_free(osn, sizeof (struct osnode));
	}
	list_destroy(&sn.objsets);

	if (err)
		(void) strcpy(fsname, sn.failed);
	dsl_sync_task_group_destroy(sn.dstg);
	spa_close(spa, FTAG);
	return (err);
}

dsl_pool_t *
dmu_objset_pool(objset_t *os)
{
	dsl_dataset_t *ds;

	if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
		return (ds->ds_dir->dd_pool);
	else
		return (spa_get_dsl(os->os_spa));
}

/*
 * Called from open context to perform a callback in syncing context.  Waits
 * for the operation to complete.
 *
 * The checkfunc will be called from open context as a preliminary check
 * which can quickly fail.  If it succeeds, it will be called again from
 * syncing context.  The checkfunc should generally be designed to work
 * properly in either context, but if necessary it can check
 * dmu_tx_is_syncing(tx).
 *
 * The synctask infrastructure enforces proper locking strategy with respect
 * to the dp_config_rwlock -- the lock will always be held when the callbacks
 * are called.  It will be held for read during the open-context (preliminary)
 * call to the checkfunc, and then held for write from syncing context during
 * the calls to the check and sync funcs.
 *
 * A dataset or pool name can be passed as the first argument.  Typically,
 * the check func will hold, check the return value of the hold, and then
 * release the dataset.  The sync func will VERIFYO(hold()) the dataset.
 * This is safe because no changes can be made between the check and sync funcs,
 * and the sync func will only be called if the check func successfully opened
 * the dataset.
 */
int
dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
    dsl_syncfunc_t *syncfunc, void *arg,
    int blocks_modified, zfs_space_check_t space_check)
{
	spa_t *spa;
	dmu_tx_t *tx;
	int err;
	dsl_sync_task_t dst = { { { NULL } } };
	dsl_pool_t *dp;

	err = spa_open(pool, &spa, FTAG);
	if (err != 0)
		return (err);
	dp = spa_get_dsl(spa);

top:
	tx = dmu_tx_create_dd(dp->dp_mos_dir);
	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));

	dst.dst_pool = dp;
	dst.dst_txg = dmu_tx_get_txg(tx);
	dst.dst_space = blocks_modified << DST_AVG_BLKSHIFT;
	dst.dst_space_check = space_check;
	dst.dst_checkfunc = checkfunc != NULL ? checkfunc : dsl_null_checkfunc;
	dst.dst_syncfunc = syncfunc;
	dst.dst_arg = arg;
	dst.dst_error = 0;
	dst.dst_nowaiter = B_FALSE;

	dsl_pool_config_enter(dp, FTAG);
	err = dst.dst_checkfunc(arg, tx);
	dsl_pool_config_exit(dp, FTAG);

	if (err != 0) {
		dmu_tx_commit(tx);
		spa_close(spa, FTAG);
		return (err);
	}

	VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, &dst, dst.dst_txg));

	dmu_tx_commit(tx);

	txg_wait_synced(dp, dst.dst_txg);

	if (dst.dst_error == EAGAIN) {
		txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE);
		goto top;
	}

	spa_close(spa, FTAG);
	return (dst.dst_error);
}

/*
 * 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;

	ASSERT(what != LOG_INTERNAL);

	ha.ha_history_str = history_str;
	ha.ha_log_type = what;
	(void) strlcpy(ha.ha_zone, spa_history_zone(), sizeof (ha.ha_zone));
	return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_history_log_sync,
	    spa, &ha, 0));
}

int
dmu_objset_snapshot(char *fsname, char *snapname,
    nvlist_t *props, boolean_t recursive)
{
	dsl_sync_task_t *dst;
	struct snaparg *sn;
	spa_t *spa;
	int err;

	sn = kmem_alloc(sizeof (struct snaparg), KM_SLEEP);
	(void) strcpy(sn->failed, fsname);

	err = spa_open(fsname, &spa, FTAG);
	if (err) {
		kmem_free(sn, sizeof (struct snaparg));
		return (err);
	}

	sn->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
	sn->snapname = snapname;
	sn->props = props;

	if (recursive) {
		sn->checkperms = B_TRUE;
		err = dmu_objset_find(fsname,
		    dmu_objset_snapshot_one, sn, DS_FIND_CHILDREN);
	} else {
		sn->checkperms = B_FALSE;
		err = dmu_objset_snapshot_one(fsname, sn);
	}

	if (err == 0)
		err = dsl_sync_task_group_wait(sn->dstg);

	for (dst = list_head(&sn->dstg->dstg_tasks); dst;
	    dst = list_next(&sn->dstg->dstg_tasks, dst)) {
		objset_t *os = dst->dst_arg1;
		dsl_dataset_t *ds = os->os->os_dsl_dataset;
		if (dst->dst_err)
			dsl_dataset_name(ds, sn->failed);
		zil_resume(dmu_objset_zil(os));
		dmu_objset_close(os);
	}

	if (err)
		(void) strcpy(fsname, sn->failed);
	dsl_sync_task_group_destroy(sn->dstg);
	spa_close(spa, FTAG);
	kmem_free(sn, sizeof (struct snaparg));
	return (err);
}