void
fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags)
{
	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
	zap->zap_ismicro = FALSE;

	zap->zap_dbu.dbu_evict_func_sync = zap_evict_sync;
	zap->zap_dbu.dbu_evict_func_async = NULL;

	mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
	zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1;

	zap_phys_t *zp = zap_f_phys(zap);
	/*
	 * explicitly zero it since it might be coming from an
	 * initialized microzap
	 */
	bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size);
	zp->zap_block_type = ZBT_HEADER;
	zp->zap_magic = ZAP_MAGIC;

	zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap);

	zp->zap_freeblk = 2;		/* block 1 will be the first leaf */
	zp->zap_num_leafs = 1;
	zp->zap_num_entries = 0;
	zp->zap_salt = zap->zap_salt;
	zp->zap_normflags = zap->zap_normflags;
	zp->zap_flags = flags;

	/* block 1 will be the first leaf */
	for (int i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++)
		ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1;

	/*
	 * set up block 1 - the first leaf
	 */
	dmu_buf_t *db;
	VERIFY0(dmu_buf_hold(zap->zap_objset, zap->zap_object,
	    1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH));
	dmu_buf_will_dirty(db, tx);

	zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
	l->l_dbuf = db;

	zap_leaf_init(l, zp->zap_normflags != 0);

	kmem_free(l, sizeof (zap_leaf_t));
	dmu_buf_rele(db, FTAG);
}

/*
 * load initial fuid domain and idx trees.  This function is used by
 * both the kernel and zdb.
 */
uint64_t
zfs_fuid_table_load(objset_t *os, uint64_t fuid_obj, avl_tree_t *idx_tree,
    avl_tree_t *domain_tree)
{
	dmu_buf_t *db;
	uint64_t fuid_size;

	ASSERT(fuid_obj != 0);
	VERIFY(0 == dmu_bonus_hold(os, fuid_obj,
	    FTAG, &db));
	fuid_size = *(uint64_t *)db->db_data;
	dmu_buf_rele(db, FTAG);

	if (fuid_size)  {
		nvlist_t **fuidnvp;
		nvlist_t *nvp = NULL;
		uint_t count;
		char *packed;
		int i;

		packed = kmem_alloc(fuid_size, KM_SLEEP);
		VERIFY(dmu_read(os, fuid_obj, 0,
		    fuid_size, packed, DMU_READ_PREFETCH) == 0);
		VERIFY(nvlist_unpack(packed, fuid_size,
		    &nvp, 0) == 0);
		VERIFY(nvlist_lookup_nvlist_array(nvp, FUID_NVP_ARRAY,
		    &fuidnvp, &count) == 0);

		for (i = 0; i != count; i++) {
			fuid_domain_t *domnode;
			char *domain;
			uint64_t idx;

			VERIFY(nvlist_lookup_string(fuidnvp[i], FUID_DOMAIN,
			    &domain) == 0);
			VERIFY(nvlist_lookup_uint64(fuidnvp[i], FUID_IDX,
			    &idx) == 0);

			domnode = kmem_alloc(sizeof (fuid_domain_t), KM_SLEEP);

			domnode->f_idx = idx;
			domnode->f_ksid = ksid_lookupdomain(domain);
			avl_add(idx_tree, domnode);
			avl_add(domain_tree, domnode);
		}
		nvlist_free(nvp);
		kmem_free(packed, fuid_size);
	}
	return (fuid_size);
}

void
dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
{
	ASSERT(dmu_tx_is_syncing(tx));

	mutex_enter(&dd->dd_lock);
	ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
	dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
	    dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
	dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
	mutex_exit(&dd->dd_lock);

	/* release the hold from dsl_dir_dirty */
	dmu_buf_rele(dd->dd_dbuf, dd);
}

void
sa_handle_destroy(sa_handle_t *hdl)
{
	mutex_enter(&hdl->sa_lock);
	(void) dmu_buf_update_user((dmu_buf_t *)hdl->sa_bonus, hdl,
	    NULL, NULL, NULL);

	if (hdl->sa_bonus_tab) {
		sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
		hdl->sa_bonus_tab = NULL;
	}
	if (hdl->sa_spill_tab) {
		sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
		hdl->sa_spill_tab = NULL;
	}

	dmu_buf_rele(hdl->sa_bonus, NULL);

	if (hdl->sa_spill)
		dmu_buf_rele((dmu_buf_t *)hdl->sa_spill, NULL);
	mutex_exit(&hdl->sa_lock);

	kmem_cache_free(sa_cache, hdl);
}

int
dmu_objset_userspace_upgrade(objset_t *os)
{
	uint64_t obj;
	int err = 0;

	if (dmu_objset_userspace_present(os))
		return (0);
	if (!dmu_objset_userused_enabled(os->os))
		return (ENOTSUP);
	if (dmu_objset_is_snapshot(os))
		return (EINVAL);

	/*
	 * We simply need to mark every object dirty, so that it will be
	 * synced out and now accounted.  If this is called
	 * concurrently, or if we already did some work before crashing,
	 * that's fine, since we track each object's accounted state
	 * independently.
	 */

	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
		dmu_tx_t *tx;
		dmu_buf_t *db;
		int objerr;

		if (issig(JUSTLOOKING) && issig(FORREAL))
			return (EINTR);

		objerr = dmu_bonus_hold(os, obj, FTAG, &db);
		if (objerr)
			continue;
		tx = dmu_tx_create(os);
		dmu_tx_hold_bonus(tx, obj);
		objerr = dmu_tx_assign(tx, TXG_WAIT);
		if (objerr) {
			dmu_tx_abort(tx);
			continue;
		}
		dmu_buf_will_dirty(db, tx);
		dmu_buf_rele(db, FTAG);
		dmu_tx_commit(tx);
	}

	os->os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
	txg_wait_synced(dmu_objset_pool(os), 0);
	return (0);
}

static int
bplist_cache(bplist_t *bpl, uint64_t blkid)
{
	int err = 0;

	if (bpl->bpl_cached_dbuf == NULL ||
	    bpl->bpl_cached_dbuf->db_offset != (blkid << bpl->bpl_blockshift)) {
		if (bpl->bpl_cached_dbuf != NULL)
			dmu_buf_rele(bpl->bpl_cached_dbuf, bpl);
		err = dmu_buf_hold(bpl->bpl_mos,
		    bpl->bpl_object, blkid << bpl->bpl_blockshift,
		    bpl, &bpl->bpl_cached_dbuf);
		ASSERT(err || bpl->bpl_cached_dbuf->db_size ==
		    1ULL << bpl->bpl_blockshift);
	}
	return (err);
}

int
zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
{
	zfs_acl_ids_t acl_ids;
	vattr_t vattr;
	znode_t *sharezp;
	vnode_t *vp;
	znode_t *zp;
	int error;

	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
	vattr.va_type = VDIR;
	vattr.va_mode = S_IFDIR|0555;
	vattr.va_uid = crgetuid(kcred);
	vattr.va_gid = crgetgid(kcred);

	sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
	sharezp->z_unlinked = 0;
	sharezp->z_atime_dirty = 0;
	sharezp->z_zfsvfs = zfsvfs;

	vp = ZTOV(sharezp);
	vn_reinit(vp);
	vp->v_type = VDIR;

	VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
	    kcred, NULL, &acl_ids));
	zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE,
	    &zp, 0, &acl_ids);
	ASSERT3P(zp, ==, sharezp);
	ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */
	POINTER_INVALIDATE(&sharezp->z_zfsvfs);
	error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
	    ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
	zfsvfs->z_shares_dir = sharezp->z_id;

	zfs_acl_ids_free(&acl_ids);
	ZTOV(sharezp)->v_count = 0;
	dmu_buf_rele(sharezp->z_dbuf, NULL);
	sharezp->z_dbuf = NULL;
	kmem_cache_free(znode_cache, sharezp);

	return (error);
}

static void
zvol_last_close(zvol_state_t *zv)
{
	zil_close(zv->zv_zilog);
	zv->zv_zilog = NULL;

	dmu_buf_rele(zv->zv_dbuf, zvol_tag);
	zv->zv_dbuf = NULL;

	/*
	 * Evict cached data
	 */
	if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
	    !(zv->zv_flags & ZVOL_RDONLY))
		txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
	(void) dmu_objset_evict_dbufs(zv->zv_objset);

	dmu_objset_disown(zv->zv_objset, zvol_tag);
	zv->zv_objset = NULL;
}

void
vdev_indirect_births_close(vdev_indirect_births_t *vib)
{
	ASSERT(vdev_indirect_births_verify(vib));

	if (vib->vib_phys->vib_count > 0) {
		uint64_t births_size = vdev_indirect_births_size_impl(vib);

		kmem_free(vib->vib_entries, births_size);
		vib->vib_entries = NULL;
	}

	dmu_buf_rele(vib->vib_dbuf, vib);

	vib->vib_objset = NULL;
	vib->vib_object = 0;
	vib->vib_dbuf = NULL;
	vib->vib_phys = NULL;

	kmem_free(vib, sizeof (*vib));
}

int
zfs_sa_readlink(znode_t *zp, uio_t *uio)
{
	dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
	size_t bufsz;
	int error;

	bufsz = zp->z_size;
	if (bufsz + ZFS_OLD_ZNODE_PHYS_SIZE <= db->db_size) {
		error = uiomove((caddr_t)db->db_data +
		    ZFS_OLD_ZNODE_PHYS_SIZE,
                        MIN((size_t)bufsz, uio_resid(uio)), UIO_READ, uio);
	} else {
		dmu_buf_t *dbp;
		if ((error = dmu_buf_hold(zp->z_zfsvfs->z_os, zp->z_id,
		    0, FTAG, &dbp, DMU_READ_NO_PREFETCH)) == 0) {
			error = uiomove(dbp->db_data,
                            MIN((size_t)bufsz, uio_resid(uio)), UIO_READ, uio);
			dmu_buf_rele(dbp, FTAG);
		}
	}
	return (error);
}

/*
 * XXX: for the moment I don't want to use lnb_flags for osd-internal
 *      purposes as it's not very well defined ...
 *      instead I use the lowest bit of the address so that:
 *        arc buffer:  .lnb_data = abuf          (arc we loan for write)
 *        dbuf buffer: .lnb_data = dbuf | 1      (dbuf we get for read)
 *        copy buffer: .lnb_page->mapping = obj (page we allocate for write)
 *
 *      bzzz, to blame
 */
static int osd_bufs_put(const struct lu_env *env, struct dt_object *dt,
			struct niobuf_local *lnb, int npages)
{
	struct osd_object *obj  = osd_dt_obj(dt);
	struct osd_device *osd = osd_obj2dev(obj);
	unsigned long      ptr;
	int                i;

	LASSERT(dt_object_exists(dt));
	LASSERT(obj->oo_db);

	for (i = 0; i < npages; i++) {
		if (lnb[i].lnb_page == NULL)
			continue;
		if (lnb[i].lnb_page->mapping == (void *)obj) {
			/* this is anonymous page allocated for copy-write */
			lnb[i].lnb_page->mapping = NULL;
			__free_page(lnb[i].lnb_page);
			atomic_dec(&osd->od_zerocopy_alloc);
		} else {
			/* see comment in osd_bufs_get_read() */
			ptr = (unsigned long)lnb[i].lnb_data;
			if (ptr & 1UL) {
				ptr &= ~1UL;
				dmu_buf_rele((void *)ptr, osd_zerocopy_tag);
				atomic_dec(&osd->od_zerocopy_pin);
			} else if (lnb[i].lnb_data != NULL) {
				dmu_return_arcbuf(lnb[i].lnb_data);
				atomic_dec(&osd->od_zerocopy_loan);
			}
		}
		lnb[i].lnb_page = NULL;
		lnb[i].lnb_data = NULL;
	}

	return 0;
}

uint64_t
dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
    dmu_tx_t *tx)
{
	objset_t *mos = dp->dp_meta_objset;
	uint64_t ddobj;
	dsl_dir_phys_t *ddphys;
	dmu_buf_t *dbuf;

	ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
	    DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
	if (pds) {
		VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
		    name, sizeof (uint64_t), 1, &ddobj, tx));
	} else {
		/* it's the root dir */
		VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
		    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
	}
	VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
	dmu_buf_will_dirty(dbuf, tx);
	ddphys = dbuf->db_data;

	ddphys->dd_creation_time = gethrestime_sec();
	if (pds)
		ddphys->dd_parent_obj = pds->dd_object;
	ddphys->dd_props_zapobj = zap_create(mos,
	    DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
	ddphys->dd_child_dir_zapobj = zap_create(mos,
	    DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
	if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
		ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
	dmu_buf_rele(dbuf, FTAG);

	return (ddobj);
}

int
dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
    const char *tail, void *tag, dsl_dir_t **ddp)
{
	dmu_buf_t *dbuf;
	dsl_dir_t *dd;
	int err;

	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
	    dsl_pool_sync_context(dp));

	err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
	if (err)
		return (err);
	dd = dmu_buf_get_user(dbuf);
#ifdef ZFS_DEBUG
	{
		dmu_object_info_t doi;
		dmu_object_info_from_db(dbuf, &doi);
		ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
	}
#endif
	/* XXX assert bonus buffer size is correct */
	if (dd == NULL) {
		dsl_dir_t *winner;
#ifndef __APPLE__
		int err;
#endif

		dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
		dd->dd_object = ddobj;
		dd->dd_dbuf = dbuf;
		dd->dd_pool = dp;
		dd->dd_phys = dbuf->db_data;
		dd->dd_used_bytes = dd->dd_phys->dd_used_bytes;
		mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);

		list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
		    offsetof(dsl_prop_cb_record_t, cbr_node));

		if (dd->dd_phys->dd_parent_obj) {
			err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj,
			    NULL, dd, &dd->dd_parent);
			if (err) {
				mutex_destroy(&dd->dd_lock);
				kmem_free(dd, sizeof (dsl_dir_t));
				dmu_buf_rele(dbuf, tag);
				return (err);
			}
			if (tail) {
#ifdef ZFS_DEBUG
				uint64_t foundobj;

				err = zap_lookup(dp->dp_meta_objset,
				    dd->dd_parent->dd_phys->dd_child_dir_zapobj,
				    tail, sizeof (foundobj), 1, &foundobj);
				ASSERT(err || foundobj == ddobj);
#endif
				(void) strcpy(dd->dd_myname, tail);
			} else {
				err = zap_value_search(dp->dp_meta_objset,
				    dd->dd_parent->dd_phys->dd_child_dir_zapobj,
				    ddobj, 0, dd->dd_myname);
			}
			if (err) {
				dsl_dir_close(dd->dd_parent, dd);
				mutex_destroy(&dd->dd_lock);
				kmem_free(dd, sizeof (dsl_dir_t));
				dmu_buf_rele(dbuf, tag);
				return (err);
			}
		} else {
			(void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
		}

		winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
		    dsl_dir_evict);
		if (winner) {
			if (dd->dd_parent)
				dsl_dir_close(dd->dd_parent, dd);
			mutex_destroy(&dd->dd_lock);
			kmem_free(dd, sizeof (dsl_dir_t));
			dd = winner;
		} else {
			spa_open_ref(dp->dp_spa, dd);
		}
	}

	/*
	 * The dsl_dir_t has both open-to-close and instantiate-to-evict
	 * holds on the spa.  We need the open-to-close holds because
	 * otherwise the spa_refcnt wouldn't change when we open a
	 * dir which the spa also has open, so we could incorrectly
	 * think it was OK to unload/export/destroy the pool.  We need
	 * the instantiate-to-evict hold because the dsl_dir_t has a
	 * pointer to the dd_pool, which has a pointer to the spa_t.
	 */
	spa_open_ref(dp->dp_spa, tag);
	ASSERT3P(dd->dd_pool, ==, dp);
	ASSERT3U(dd->dd_object, ==, ddobj);
//.........这里部分代码省略.........

/*
 * 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);
}

/*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);
//.........这里部分代码省略.........

//.........这里部分代码省略.........

	LASSERT(obj->oo_sa_hdl);

	if (obj->oo_xattr == ZFS_NO_OBJECT) {
		struct lu_attr *la = &osd_oti_get(env)->oti_la;

		la->la_valid = LA_MODE;
		la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
		rc = __osd_zap_create(env, osd, &xa_zap_db, tx, la,
				      obj->oo_db->db_object, 0);
		if (rc)
			return rc;

		obj->oo_xattr = xa_zap_db->db_object;
		rc = osd_object_sa_update(obj, SA_ZPL_XATTR(osd),
				&obj->oo_xattr, 8, oh);
		if (rc)
			goto out;
	}

	rc = -zap_lookup(osd->od_os, obj->oo_xattr, name, sizeof(uint64_t), 1,
			 &xa_data_obj);
	if (rc == 0) {
		if (fl & LU_XATTR_CREATE) {
			rc = -EEXIST;
			goto out;
		}
		/*
		 * Entry already exists.
		 * We'll truncate the existing object.
		 */
		rc = __osd_obj2dbuf(env, osd->od_os, xa_data_obj,
					&xa_data_db);
		if (rc)
			goto out;

		rc = -sa_handle_get(osd->od_os, xa_data_obj, NULL,
					SA_HDL_PRIVATE, &sa_hdl);
		if (rc)
			goto out;

		rc = -sa_lookup(sa_hdl, SA_ZPL_SIZE(osd), &size, 8);
		if (rc)
			goto out_sa;

		rc = -dmu_free_range(osd->od_os, xa_data_db->db_object,
				     0, DMU_OBJECT_END, tx);
		if (rc)
			goto out_sa;
	} else if (rc == -ENOENT) {
		struct lu_attr *la = &osd_oti_get(env)->oti_la;
		/*
		 * Entry doesn't exist, we need to create a new one and a new
		 * object to store the value.
		 */
		if (fl & LU_XATTR_REPLACE) {
			/* should be ENOATTR according to the
			 * man, but that is undefined here */
			rc = -ENODATA;
			goto out;
		}

		la->la_valid = LA_MODE;
		la->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
		rc = __osd_object_create(env, obj, &xa_data_db, tx, la,
					 obj->oo_xattr);
		if (rc)
			goto out;
		xa_data_obj = xa_data_db->db_object;

		rc = -sa_handle_get(osd->od_os, xa_data_obj, NULL,
					SA_HDL_PRIVATE, &sa_hdl);
		if (rc)
			goto out;

		rc = -zap_add(osd->od_os, obj->oo_xattr, name, sizeof(uint64_t),
				1, &xa_data_obj, tx);
		if (rc)
			goto out_sa;
	} else {
		/* There was an error looking up the xattr name */
		goto out;
	}

	/* Finally write the xattr value */
	dmu_write(osd->od_os, xa_data_obj, 0, buf->lb_len, buf->lb_buf, tx);

	size = buf->lb_len;
	rc = -sa_update(sa_hdl, SA_ZPL_SIZE(osd), &size, 8, tx);

out_sa:
	sa_handle_destroy(sa_hdl);
out:
	if (xa_data_db != NULL)
		dmu_buf_rele(xa_data_db, FTAG);
	if (xa_zap_db != NULL)
		dmu_buf_rele(xa_zap_db, FTAG);

	return rc;
}

//.........这里部分代码省略.........
			VERIFY(0 == sa_resize_spill(hdl,
			    BUF_SPACE_NEEDED(spill_used, spillhdrsize), tx));
	}

	/* setup starting pointers to lay down data */
	data_start = (void *)((uintptr_t)hdl->sa_bonus->db_data + hdrsize);
	sahdr = (sa_hdr_phys_t *)hdl->sa_bonus->db_data;
	buftype = SA_BONUS;

	if (spilling)
		buf_space = (sa->sa_force_spill) ?
		    0 : SA_BLKPTR_SPACE - hdrsize;
	else
		buf_space = hdl->sa_bonus->db_size - hdrsize;

	attrs_start = attrs = kmem_alloc(sizeof (sa_attr_type_t) * attr_count,
	    KM_SLEEP);
	lot_count = 0;

	for (i = 0, len_idx = 0, hash = -1ULL; i != attr_count; i++) {
		uint16_t length;

		attrs[i] = attr_desc[i].sa_attr;
		length = SA_REGISTERED_LEN(sa, attrs[i]);
		if (length == 0)
			length = attr_desc[i].sa_length;

		if (buf_space < length) {  /* switch to spill buffer */
			VERIFY(bonustype == DMU_OT_SA);
			if (buftype == SA_BONUS && !sa->sa_force_spill) {
				sa_find_layout(hdl->sa_os, hash, attrs_start,
				    lot_count, tx, &lot);
				SA_SET_HDR(sahdr, lot->lot_num, hdrsize);
			}

			buftype = SA_SPILL;
			hash = -1ULL;
			len_idx = 0;

			sahdr = (sa_hdr_phys_t *)hdl->sa_spill->db_data;
			sahdr->sa_magic = SA_MAGIC;
			data_start = (void *)((uintptr_t)sahdr +
			    spillhdrsize);
			attrs_start = &attrs[i];
			buf_space = hdl->sa_spill->db_size - spillhdrsize;
			lot_count = 0;
		}
		hash ^= SA_ATTR_HASH(attrs[i]);
		attr_desc[i].sa_addr = data_start;
		attr_desc[i].sa_size = length;
		SA_COPY_DATA(attr_desc[i].sa_data_func, attr_desc[i].sa_data,
		    data_start, length);
		if (sa->sa_attr_table[attrs[i]].sa_length == 0) {
			sahdr->sa_lengths[len_idx++] = length;
		}
		data_start = (void *)P2ROUNDUP(((uintptr_t)data_start +
		    length), 8);
		buf_space -= P2ROUNDUP(length, 8);
		lot_count++;
	}

	sa_find_layout(hdl->sa_os, hash, attrs_start, lot_count, tx, &lot);

	/*
	 * Verify that old znodes always have layout number 0.
	 * Must be DMU_OT_SA for arbitrary layouts
	 */
	VERIFY((bonustype == DMU_OT_ZNODE && lot->lot_num == 0) ||
	    (bonustype == DMU_OT_SA && lot->lot_num > 1));

	if (bonustype == DMU_OT_SA) {
		SA_SET_HDR(sahdr, lot->lot_num,
		    buftype == SA_BONUS ? hdrsize : spillhdrsize);
	}

	kmem_free(attrs, sizeof (sa_attr_type_t) * attr_count);
	if (hdl->sa_bonus_tab) {
		sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
		hdl->sa_bonus_tab = NULL;
	}
	if (!sa->sa_force_spill)
		VERIFY(0 == sa_build_index(hdl, SA_BONUS));
	if (hdl->sa_spill) {
		sa_idx_tab_rele(hdl->sa_os, hdl->sa_spill_tab);
		if (!spilling) {
			/*
			 * remove spill block that is no longer needed.
			 */
			dmu_buf_rele(hdl->sa_spill, NULL);
			hdl->sa_spill = NULL;
			hdl->sa_spill_tab = NULL;
			VERIFY(0 == dmu_rm_spill(hdl->sa_os,
			    sa_handle_object(hdl), tx));
		} else {
			VERIFY(0 == sa_build_index(hdl, SA_SPILL));
		}
	}

	return (0);
}

int
dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
    const char *tail, void *tag, dsl_dir_t **ddp)
{
	dmu_buf_t *dbuf;
	dsl_dir_t *dd;
	int err;

	ASSERT(dsl_pool_config_held(dp));

	err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
	if (err != 0)
		return (err);
	dd = dmu_buf_get_user(dbuf);
#ifdef ZFS_DEBUG
	{
		dmu_object_info_t doi;
		dmu_object_info_from_db(dbuf, &doi);
		ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
		ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
	}
#endif
	if (dd == NULL) {
		dsl_dir_t *winner;

		dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
		dd->dd_object = ddobj;
		dd->dd_dbuf = dbuf;
		dd->dd_pool = dp;
		dd->dd_phys = dbuf->db_data;
		mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);

		list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
		    offsetof(dsl_prop_cb_record_t, cbr_node));

		dsl_dir_snap_cmtime_update(dd);

		if (dd->dd_phys->dd_parent_obj) {
			err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
			    NULL, dd, &dd->dd_parent);
			if (err != 0)
				goto errout;
			if (tail) {
#ifdef ZFS_DEBUG
				uint64_t foundobj;

				err = zap_lookup(dp->dp_meta_objset,
				    dd->dd_parent->dd_phys->dd_child_dir_zapobj,
				    tail, sizeof (foundobj), 1, &foundobj);
				ASSERT(err || foundobj == ddobj);
#endif
				(void) strcpy(dd->dd_myname, tail);
			} else {
				err = zap_value_search(dp->dp_meta_objset,
				    dd->dd_parent->dd_phys->dd_child_dir_zapobj,
				    ddobj, 0, dd->dd_myname);
			}
			if (err != 0)
				goto errout;
		} else {
			(void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
		}

		if (dsl_dir_is_clone(dd)) {
			dmu_buf_t *origin_bonus;
			dsl_dataset_phys_t *origin_phys;

			/*
			 * We can't open the origin dataset, because
			 * that would require opening this dsl_dir.
			 * Just look at its phys directly instead.
			 */
			err = dmu_bonus_hold(dp->dp_meta_objset,
			    dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
			if (err != 0)
				goto errout;
			origin_phys = origin_bonus->db_data;
			dd->dd_origin_txg =
			    origin_phys->ds_creation_txg;
			dmu_buf_rele(origin_bonus, FTAG);
		}

		winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
		    dsl_dir_evict);
		if (winner) {
			if (dd->dd_parent)
				dsl_dir_rele(dd->dd_parent, dd);
			mutex_destroy(&dd->dd_lock);
			kmem_free(dd, sizeof (dsl_dir_t));
			dd = winner;
		} else {
			spa_open_ref(dp->dp_spa, dd);
		}
	}

	/*
	 * The dsl_dir_t has both open-to-close and instantiate-to-evict
	 * holds on the spa.  We need the open-to-close holds because
	 * otherwise the spa_refcnt wouldn't change when we open a
	 * dir which the spa also has open, so we could incorrectly
//.........这里部分代码省略.........

int
zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t len,
    char *addr)
{
	dmu_object_info_t doi;
	ssize_t nbytes;
	itx_t *itx;
	lr_write_t *lr;
	objset_t *os;
	dmu_buf_t *db;
	uint64_t txg;
	uint64_t boff;
	int error;
	uint32_t blocksize;

	/* handle common case */
	if (len <= zvol_immediate_write_sz) {
		itx = zvol_immediate_itx(off, len, addr);
		(void) zil_itx_assign(zv->zv_zilog, itx, tx);
		return (0);
	}

	txg = dmu_tx_get_txg(tx);
	os = zv->zv_objset;

	/*
	 * We need to dmu_sync() each block in the range.
	 * For this we need the blocksize.
	 */
	error = dmu_object_info(os, ZVOL_OBJ, &doi);
	if (error)
		return (error);
	blocksize = doi.doi_data_block_size;

	/*
	 * We need to immediate write or dmu_sync() each block in the range.
	 */
	while (len) {
		nbytes = MIN(len, blocksize - P2PHASE(off, blocksize));
		if (nbytes <= zvol_immediate_write_sz) {
			itx = zvol_immediate_itx(off, nbytes, addr);
		} else {
			boff =  P2ALIGN_TYPED(off, blocksize, uint64_t);
			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
			lr = (lr_write_t *)&itx->itx_lr;
			lr->lr_foid = ZVOL_OBJ;
			lr->lr_offset = off;
			lr->lr_length = nbytes;
			lr->lr_blkoff = off - boff;
			BP_ZERO(&lr->lr_blkptr);

			/* XXX - we should do these IOs in parallel */
			VERIFY(0 == dmu_buf_hold(os, ZVOL_OBJ, boff,
			    FTAG, &db));
			ASSERT(boff == db->db_offset);
			error = dmu_sync(NULL, db, &lr->lr_blkptr,
			    txg, NULL, NULL);
			dmu_buf_rele(db, FTAG);
			if (error) {
				kmem_free(itx, offsetof(itx_t, itx_lr));
				return (error);
			}
			itx->itx_wr_state = WR_COPIED;
		}
		(void) zil_itx_assign(zv->zv_zilog, itx, tx);
		len -= nbytes;
		off += nbytes;
	}
	return (0);
}

//.........这里部分代码省略.........
			if (vfsp->vfs_count > 1)
				return (EBUSY);
		} else {
			if (vfsp->vfs_count > 2 ||
			    zfsvfs->z_ctldir->v_count > 1) {
				return (EBUSY);
			}
		}
#endif
	}

	rw_enter(&zfsvfs->z_unmount_lock, RW_WRITER);
	rw_enter(&zfsvfs->z_unmount_inactive_lock, RW_WRITER);

	/*
	 * At this point there are no vops active, and any new vops will
	 * fail with EIO since we have z_unmount_lock for writer (only
	 * relavent for forced unmount).
	 *
	 * Release all holds on dbufs.
	 * Note, the dmu can still callback via znode_pageout_func()
	 * which can zfs_znode_free() the znode.  So we lock
	 * z_all_znodes; search the list for a held dbuf; drop the lock
	 * (we know zp can't disappear if we hold a dbuf lock) then
	 * regrab the lock and restart.
	 */
	mutex_enter(&zfsvfs->z_znodes_lock);
	for (zp = list_head(&zfsvfs->z_all_znodes); zp; zp = nextzp) {
		nextzp = list_next(&zfsvfs->z_all_znodes, zp);
		if (zp->z_dbuf_held) {
			/* dbufs should only be held when force unmounting */
			zp->z_dbuf_held = 0;
			mutex_exit(&zfsvfs->z_znodes_lock);
			dmu_buf_rele(zp->z_dbuf, NULL);
			/* Start again */
			mutex_enter(&zfsvfs->z_znodes_lock);
			nextzp = list_head(&zfsvfs->z_all_znodes);
		}
	}
	mutex_exit(&zfsvfs->z_znodes_lock);

	/*
	 * Set the unmounted flag and let new vops unblock.
	 * zfs_inactive will have the unmounted behavior, and all other
	 * vops will fail with EIO.
	 */
	zfsvfs->z_unmounted = B_TRUE;
	rw_exit(&zfsvfs->z_unmount_lock);
	rw_exit(&zfsvfs->z_unmount_inactive_lock);

	/*
	 * Unregister properties.
	 */
#ifndef __APPLE__
	if (!dmu_objset_is_snapshot(os))
		zfs_unregister_callbacks(zfsvfs);
#endif
	/*
	 * Close the zil. NB: Can't close the zil while zfs_inactive
	 * threads are blocked as zil_close can call zfs_inactive.
	 */
	if (zfsvfs->z_log) {
		zil_close(zfsvfs->z_log);
		zfsvfs->z_log = NULL;
	}