static int
zvol_write(zvol_state_t *zv, uio_t *uio, boolean_t sync)
{
	uint64_t volsize = zv->zv_volsize;
	rl_t *rl;
	int error = 0;

	ASSERT(zv && zv->zv_open_count > 0);

	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
	    RL_WRITER);

	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
		uint64_t off = uio->uio_loffset;
		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);

		if (bytes > volsize - off)	/* don't write past the end */
			bytes = volsize - off;

		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);

		/* This will only fail for ENOSPC */
		error = dmu_tx_assign(tx, TXG_WAIT);
		if (error) {
			dmu_tx_abort(tx);
			break;
		}
		error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
		if (error == 0)
			zvol_log_write(zv, tx, off, bytes, sync);
		dmu_tx_commit(tx);

		if (error)
			break;
	}
	zfs_range_unlock(rl);
	if (sync)
		zil_commit(zv->zv_zilog, ZVOL_OBJ);
	return (error);
}

static uint64_t
zpios_dmu_object_create(run_args_t *run_args, objset_t *os)
{
	struct dmu_tx *tx;
	uint64_t obj = 0ULL;
	uint64_t blksize = run_args->block_size;
	int rc;

	if (blksize < SPA_MINBLOCKSIZE ||
	    blksize > spa_maxblocksize(dmu_objset_spa(os)) ||
	    !ISP2(blksize)) {
		zpios_print(run_args->file,
		    "invalid block size for pool: %d\n", (int)blksize);
		return (obj);
	}

	tx = dmu_tx_create(os);
	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, OBJ_SIZE);
	rc = dmu_tx_assign(tx, TXG_WAIT);
	if (rc) {
		zpios_print(run_args->file,
		    "dmu_tx_assign() failed: %d\n", rc);
		dmu_tx_abort(tx);
		return (obj);
	}

	obj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, DMU_OT_NONE, 0, tx);
	rc = dmu_object_set_blocksize(os, obj, blksize, 0, tx);
	if (rc) {
		zpios_print(run_args->file,
		    "dmu_object_set_blocksize to %d failed: %d\n",
		    (int)blksize, rc);
		dmu_tx_abort(tx);
		return (obj);
	}

	dmu_tx_commit(tx);

	return (obj);
}

/*
 * Set ZFS_PROP_VOLBLOCKSIZE set entry point.
 */
int
zvol_set_volblocksize(const char *name, uint64_t volblocksize)
{
	zvol_state_t *zv;
	dmu_tx_t *tx;
	int error;

	mutex_enter(&zvol_state_lock);

	zv = zvol_find_by_name(name);
	if (zv == NULL) {
		error = SET_ERROR(ENXIO);
		goto out;
	}

	if (zv->zv_flags & ZVOL_RDONLY) {
		error = SET_ERROR(EROFS);
		goto out;
	}

	tx = dmu_tx_create(zv->zv_objset);
	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
	error = dmu_tx_assign(tx, TXG_WAIT);
	if (error) {
		dmu_tx_abort(tx);
	} else {
		error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ,
		    volblocksize, 0, tx);
		if (error == ENOTSUP)
			error = SET_ERROR(EBUSY);
		dmu_tx_commit(tx);
		if (error == 0)
			zv->zv_volblocksize = volblocksize;
	}
out:
	mutex_exit(&zvol_state_lock);

	return (SET_ERROR(error));
}

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

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

	err = nvlist_lookup_nvlist(nvl, ZPOOL_HIST_INPUT_NVL, &in_nvl);
	if (err == 0) {
		(void) nvlist_remove_all(in_nvl, ZPOOL_HIDDEN_ARGS);
	}

	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);

}

static int
zpios_dmu_write(run_args_t *run_args, objset_t *os, uint64_t object,
		uint64_t offset, uint64_t size, const void *buf)
{
	struct dmu_tx *tx;
	int rc, how = TXG_WAIT;
//	int flags = 0;

	if (run_args->flags & DMU_WRITE_NOWAIT)
		how = TXG_NOWAIT;

	while (1) {
		tx = dmu_tx_create(os);
		dmu_tx_hold_write(tx, object, offset, size);
		rc = dmu_tx_assign(tx, how);

		if (rc) {
			if (rc == ERESTART && how == TXG_NOWAIT) {
				dmu_tx_wait(tx);
				dmu_tx_abort(tx);
				continue;
			}
			zpios_print(run_args->file,
				    "Error in dmu_tx_assign(), %d", rc);
			dmu_tx_abort(tx);
			return (rc);
		}
		break;
	}

//	if (run_args->flags & DMU_WRITE_ZC)
//		flags |= DMU_WRITE_ZEROCOPY;

	dmu_write(os, object, offset, size, buf, tx);
	dmu_tx_commit(tx);

	return (0);
}

int
zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
{
	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
	znode_t *xzp;
	dmu_tx_t *tx;
	uint64_t xoid;
	int error;

	*xvpp = NULL;

	if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, cr))
		return (error);

	tx = dmu_tx_create(zfsvfs->z_os);
	dmu_tx_hold_bonus(tx, zp->z_id);
	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
	error = dmu_tx_assign(tx, zfsvfs->z_assign);
	if (error) {
		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT)
			dmu_tx_wait(tx);
		dmu_tx_abort(tx);
		return (error);
	}
	zfs_mknode(zp, vap, &xoid, tx, cr, IS_XATTR, &xzp, 0);
	ASSERT(xzp->z_id == xoid);
	ASSERT(xzp->z_phys->zp_parent == zp->z_id);
	dmu_buf_will_dirty(zp->z_dbuf, tx);
	zp->z_phys->zp_xattr = xoid;

	(void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp, xzp, "");
	dmu_tx_commit(tx);

	*xvpp = ZTOV(xzp);

	return (0);
}

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);
}

/*
 * Close an intent log.
 */
void
zil_close(zilog_t *zilog)
{
	/*
	 * If the log isn't already committed, mark the objset dirty
	 * (so zil_sync() will be called) and wait for that txg to sync.
	 */
	if (!zil_is_committed(zilog)) {
		uint64_t txg;
		dmu_tx_t *tx = dmu_tx_create(zilog->zl_os);
		(void) dmu_tx_assign(tx, TXG_WAIT);
		dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
		txg = dmu_tx_get_txg(tx);
		dmu_tx_commit(tx);
		txg_wait_synced(zilog->zl_dmu_pool, txg);
	}

	taskq_destroy(zilog->zl_clean_taskq);
	zilog->zl_clean_taskq = NULL;
	zilog->zl_get_data = NULL;

	zil_itx_clean(zilog);
	ASSERT(list_head(&zilog->zl_itx_list) == NULL);
}

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)
		return (EINVAL);

	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);
	}

	nvarg = fnvlist_dup(nvl);
	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);

}

int
zvol_set_volblocksize(zfs_cmd_t *zc)
{
	zvol_state_t *zv;
	dmu_tx_t *tx;
	int error;

	mutex_enter(&zvol_state_lock);

	if ((zv = zvol_minor_lookup(zc->zc_name)) == NULL) {
		mutex_exit(&zvol_state_lock);
		return (ENXIO);
	}

	if (zv->zv_readonly || (zv->zv_mode & DS_MODE_READONLY)) {
		mutex_exit(&zvol_state_lock);
		return (EROFS);
	}

	tx = dmu_tx_create(zv->zv_objset);
	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
	error = dmu_tx_assign(tx, TXG_WAIT);
	if (error) {
		dmu_tx_abort(tx);
	} else {
		error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ,
		    zc->zc_volblocksize, 0, tx);
		if (error == ENOTSUP)
			error = EBUSY;
		dmu_tx_commit(tx);
	}

	mutex_exit(&zvol_state_lock);

	return (error);
}

int
zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
{
	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
	znode_t *xzp;
	dmu_tx_t *tx;
	int error;
	zfs_acl_ids_t acl_ids;
	boolean_t fuid_dirtied;
	uint64_t parent;

	*xvpp = NULL;

	/*
	 * In FreeBSD, access checking for creating an EA is being done
	 * in zfs_setextattr(),
	 */
#ifndef __FreeBSD_kernel__
	if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr))
		return (error);
#endif

	if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
	    &acl_ids)) != 0)
		return (error);
	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
		zfs_acl_ids_free(&acl_ids);
		return (SET_ERROR(EDQUOT));
	}

	getnewvnode_reserve(1);

	tx = dmu_tx_create(zfsvfs->z_os);
	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
	    ZFS_SA_BASE_ATTR_SIZE);
	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
	fuid_dirtied = zfsvfs->z_fuid_dirty;
	if (fuid_dirtied)
		zfs_fuid_txhold(zfsvfs, tx);
	error = dmu_tx_assign(tx, TXG_WAIT);
	if (error) {
		zfs_acl_ids_free(&acl_ids);
		dmu_tx_abort(tx);
		return (error);
	}
	zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);

	if (fuid_dirtied)
		zfs_fuid_sync(zfsvfs, tx);

#ifdef DEBUG
	error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
	    &parent, sizeof (parent));
	ASSERT(error == 0 && parent == zp->z_id);
#endif

	VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
	    sizeof (xzp->z_id), tx));

	(void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
	    xzp, "", NULL, acl_ids.z_fuidp, vap);

	zfs_acl_ids_free(&acl_ids);
	dmu_tx_commit(tx);

	getnewvnode_drop_reserve();

	*xvpp = ZTOV(xzp);

	return (0);
}

void
zfs_rmnode(znode_t *zp)
{
	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
	objset_t	*os = zfsvfs->z_os;
	znode_t		*xzp = NULL;
	char		obj_name[17];
	dmu_tx_t	*tx;
	uint64_t	acl_obj;
	int		error;

#ifndef __APPLE__
	ASSERT(ZTOV(zp)->v_count == 0);
#endif /*!__APPLE__*/
	ASSERT(zp->z_phys->zp_links == 0);

#ifdef ZFS_DEBUG
	znode_stalker(zp, N_zfs_rmnode);
#endif

	/*
	 * If this is an attribute directory, purge its contents.
	 */
#ifdef __APPLE__
	if (S_ISDIR(zp->z_phys->zp_mode) && (zp->z_phys->zp_flags & ZFS_XATTR)) 
#else
	if (ZTOV(zp)->v_type == VDIR && (zp->z_phys->zp_flags & ZFS_XATTR)) 
#endif
	{
		if (zfs_purgedir(zp) != 0) {
			/*
			 * Not enough space to delete some xattrs.
			 * Leave it on the unlinked set.
			 */
			return;
		}
	}

	/*
	 * If the file has extended attributes, we're going to unlink
	 * the xattr dir.
	 */
	if (zp->z_phys->zp_xattr) {
		error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
		ASSERT(error == 0);
	}

	acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj;

	/*
	 * Set up the transaction.
	 */
	tx = dmu_tx_create(os);
	dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
	if (xzp) {
		dmu_tx_hold_bonus(tx, xzp->z_id);
		dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
	}
	if (acl_obj)
		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
	error = dmu_tx_assign(tx, TXG_WAIT);
	if (error) {
		/*
		 * Not enough space to delete the file.  Leave it in the
		 * unlinked set, leaking it until the fs is remounted (at
		 * which point we'll call zfs_unlinked_drain() to process it).
		 */
		dmu_tx_abort(tx);
#ifdef __APPLE__
		/*XXX NOEL: double check this path logic. see radar 5182217.
		 * This may be disturbing some of the evict logic
		 * and hence causing the NULL ptr drefs seen every great while 
		 * in some of the test cases*/
		zp->z_dbuf_held = 0;
		ZFS_OBJ_HOLD_EXIT(zfsvfs, zp->z_id);
		dmu_buf_rele(zp->z_dbuf, NULL);
#endif /* __APPLE__ */
		return;
	}

	if (xzp) {
		dmu_buf_will_dirty(xzp->z_dbuf, tx);
		mutex_enter(&xzp->z_lock);
		xzp->z_unlinked = B_TRUE;	/* mark xzp for deletion */
		xzp->z_phys->zp_links = 0;	/* no more links to it */
		mutex_exit(&xzp->z_lock);
		zfs_unlinked_add(xzp, tx);
	}

	/* Remove this znode from the unlinked set */
	error = zap_remove(os, zfsvfs->z_unlinkedobj,
	    zfs_unlinked_hexname(obj_name, zp->z_id), tx);
	ASSERT3U(error, ==, 0);

	zfs_znode_delete(zp, tx);

	dmu_tx_commit(tx);

	if (xzp)
//.........这里部分代码省略.........

void
zfs_rmnode(znode_t *zp)
{
	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
	objset_t	*os = zfsvfs->z_os;
	znode_t		*xzp = NULL;
	dmu_tx_t	*tx;
	uint64_t	acl_obj;
	uint64_t	xattr_obj;
	uint64_t	links;
	int		error;

	ASSERT(ZTOI(zp)->i_nlink == 0);
	ASSERT(atomic_read(&ZTOI(zp)->i_count) == 0);

	/*
	 * If this is an attribute directory, purge its contents.
	 */
	if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) {
		if (zfs_purgedir(zp) != 0) {
			/*
			 * Not enough space to delete some xattrs.
			 * Leave it in the unlinked set.
			 */
			zfs_znode_dmu_fini(zp);

			return;
		}
	}

	/*
	 * Free up all the data in the file.  We don't do this for directories
	 * because we need truncate and remove to be in the same tx, like in
	 * zfs_znode_delete(). Otherwise, if we crash here we'll end up with
	 * an inconsistent truncated zap object in the delete queue.  Note a
	 * truncated file is harmless since it only contains user data.
	 */
	if (S_ISREG(ZTOI(zp)->i_mode)) {
		error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
		if (error) {
			/*
			 * Not enough space or we were interrupted by unmount.
			 * Leave the file in the unlinked set.
			 */
			zfs_znode_dmu_fini(zp);
			return;
		}
	}

	/*
	 * If the file has extended attributes, we're going to unlink
	 * the xattr dir.
	 */
	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
	    &xattr_obj, sizeof (xattr_obj));
	if (error == 0 && xattr_obj) {
		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
		ASSERT(error == 0);
	}

	acl_obj = zfs_external_acl(zp);

	/*
	 * Set up the final transaction.
	 */
	tx = dmu_tx_create(os);
	dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
	if (xzp) {
		dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
	}
	if (acl_obj)
		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);

	zfs_sa_upgrade_txholds(tx, zp);
	error = dmu_tx_assign(tx, TXG_WAIT);
	if (error) {
		/*
		 * Not enough space to delete the file.  Leave it in the
		 * unlinked set, leaking it until the fs is remounted (at
		 * which point we'll call zfs_unlinked_drain() to process it).
		 */
		dmu_tx_abort(tx);
		zfs_znode_dmu_fini(zp);
		goto out;
	}

	if (xzp) {
		ASSERT(error == 0);
		mutex_enter(&xzp->z_lock);
		xzp->z_unlinked = B_TRUE;	/* mark xzp for deletion */
		clear_nlink(ZTOI(xzp));		/* no more links to it */
		links = 0;
		VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
		    &links, sizeof (links), tx));
		mutex_exit(&xzp->z_lock);
		zfs_unlinked_add(xzp, tx);
	}

//.........这里部分代码省略.........

/*
 * Free space in a file
 *
 *	IN:	zp	- znode of file to free data in.
 *		off	- start of range
 *		len	- end of range (0 => EOF)
 *		flag	- current file open mode flags.
 *		log	- TRUE if this action should be logged
 *
 * 	RETURN:	0 on success, error code on failure
 */
int
zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
{
	struct inode *ip = ZTOI(zp);
	dmu_tx_t *tx;
	zfs_sb_t *zsb = ZTOZSB(zp);
	zilog_t *zilog = zsb->z_log;
	uint64_t mode;
	uint64_t mtime[2], ctime[2];
	sa_bulk_attr_t bulk[3];
	int count = 0;
	int error;

	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zsb), &mode,
	    sizeof (mode))) != 0)
		return (error);

	if (off > zp->z_size) {
		error =  zfs_extend(zp, off+len);
		if (error == 0 && log)
			goto log;
		else
			return (error);
	}

	/*
	 * Check for any locks in the region to be freed.
	 */
	if (ip->i_flock && mandatory_lock(ip)) {
		uint64_t length = (len ? len : zp->z_size - off);
		if (!lock_may_write(ip, off, length))
			return (SET_ERROR(EAGAIN));
	}

	if (len == 0) {
		error = zfs_trunc(zp, off);
	} else {
		if ((error = zfs_free_range(zp, off, len)) == 0 &&
		    off + len > zp->z_size)
			error = zfs_extend(zp, off+len);
	}
	if (error || !log)
		return (error);
log:
	tx = dmu_tx_create(zsb->z_os);
	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
	zfs_sa_upgrade_txholds(tx, zp);
	error = dmu_tx_assign(tx, TXG_WAIT);
	if (error) {
		dmu_tx_abort(tx);
		return (error);
	}

	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, mtime, 16);
	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, ctime, 16);
	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb),
	    NULL, &zp->z_pflags, 8);
	zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
	ASSERT(error == 0);

	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);

	dmu_tx_commit(tx);
	zfs_inode_update(zp);
	return (0);
}

/*
 * Lookup/Create an extended attribute entry.
 *
 * Input arguments:
 *	dzp	- znode for hidden attribute directory
 *	name	- name of attribute
 *	flag	- ZNEW: if the entry already exists, fail with EEXIST.
 *		  ZEXISTS: if the entry does not exist, fail with ENOENT.
 *
 * Output arguments:
 *	vpp	- pointer to the vnode for the entry (NULL if there isn't one)
 *
 * Return value: 0 on success or errno value on failure.
 */
int
zfs_obtain_xattr(znode_t *dzp, const char *name, mode_t mode, cred_t *cr,
                 vnode_t **vpp, int flag)
{
	znode_t  *xzp = NULL;
	zfsvfs_t  *zfsvfs = dzp->z_zfsvfs;
	zilog_t  *zilog;
	zfs_dirlock_t  *dl;
	dmu_tx_t  *tx;
	struct vnode_attr  vattr;
	int error;
	struct componentname cn;
	zfs_acl_ids_t	acl_ids;

	/* zfs_dirent_lock() expects a component name */
	bzero(&cn, sizeof (cn));
	cn.cn_nameiop = LOOKUP;
	cn.cn_flags = ISLASTCN;
	cn.cn_nameptr = (char *)name;
	cn.cn_namelen = strlen(name);

    ZFS_ENTER(zfsvfs);
    ZFS_VERIFY_ZP(dzp);
    zilog = zfsvfs->z_log;

	VATTR_INIT(&vattr);
	VATTR_SET(&vattr, va_type, VREG);
	VATTR_SET(&vattr, va_mode, mode & ~S_IFMT);

	if ((error = zfs_acl_ids_create(dzp, 0,
                                    &vattr, cr, NULL, &acl_ids)) != 0) {
		ZFS_EXIT(zfsvfs);
		return (error);
	}
 top:
	/* Lock the attribute entry name. */
	if ( (error = zfs_dirent_lock(&dl, dzp, (char *)name, &xzp, flag,
                                  NULL, &cn)) ) {
		goto out;
	}
	/* If the name already exists, we're done. */
	if (xzp != NULL) {
		zfs_dirent_unlock(dl);
		goto out;
	}
	tx = dmu_tx_create(zfsvfs->z_os);
	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
	//dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
	//dmu_tx_hold_bonus(tx, dzp->z_id);
	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, (char *)name);

#if 1 // FIXME
	if (dzp->z_pflags & ZFS_INHERIT_ACE) {
		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
	}
#endif
    zfs_sa_upgrade_txholds(tx, dzp);
	error = dmu_tx_assign(tx, TXG_NOWAIT);
	if (error) {
		zfs_dirent_unlock(dl);
		if (error == ERESTART) {
			dmu_tx_wait(tx);
			dmu_tx_abort(tx);
			goto top;
		}
		dmu_tx_abort(tx);
		goto out;
	}

	zfs_mknode(dzp, &vattr, tx, cr, 0, &xzp, &acl_ids);

    /*
      ASSERT(xzp->z_id == zoid);
    */
	(void) zfs_link_create(dl, xzp, tx, ZNEW);
	zfs_log_create(zilog, tx, TX_CREATE, dzp, xzp, (char *)name,
                   NULL /* vsecp */, 0 /*acl_ids.z_fuidp*/, &vattr);
    zfs_acl_ids_free(&acl_ids);
	dmu_tx_commit(tx);
    zfs_znode_wait_vnode(xzp);

	zfs_dirent_unlock(dl);
 out:
	if (error == EEXIST)
		error = ENOATTR;
//.........这里部分代码省略.........

void
zfs_rmnode(znode_t *zp)
{
	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
	objset_t	*os = zfsvfs->z_os;
	znode_t		*xzp = NULL;
	dmu_tx_t	*tx;
	uint64_t	acl_obj;
	int		error;
	int		vfslocked;

	vfslocked = VFS_LOCK_GIANT(zfsvfs->z_vfs);

	ASSERT(zp->z_phys->zp_links == 0);

	/*
	 * If this is a ZIL replay then leave the object in the unlinked set.
	 * Otherwise we can get a deadlock, because the delete can be
	 * quite large and span multiple tx's and txgs, but each replay
	 * creates a tx to atomically run the replay function and mark the
	 * replay record as complete. We deadlock trying to start a tx in
	 * a new txg to further the deletion but can't because the replay
	 * tx hasn't finished.
	 *
	 * We actually delete the object if we get a failure to create an
	 * object in zil_replay_log_record(), or after calling zil_replay().
	 */
	if (zfsvfs->z_assign >= TXG_INITIAL) {
		zfs_znode_dmu_fini(zp);
		zfs_znode_free(zp);
		return;
	}

	/*
	 * If this is an attribute directory, purge its contents.
	 */
	if (ZTOV(zp) != NULL && ZTOV(zp)->v_type == VDIR &&
	    (zp->z_phys->zp_flags & ZFS_XATTR)) {
		if (zfs_purgedir(zp) != 0) {
			/*
			 * Not enough space to delete some xattrs.
			 * Leave it in the unlinked set.
			 */
			zfs_znode_dmu_fini(zp);
			zfs_znode_free(zp);
			VFS_UNLOCK_GIANT(vfslocked);
			return;
		}
	}

	/*
	 * Free up all the data in the file.
	 */
	error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
	if (error) {
		/*
		 * Not enough space.  Leave the file in the unlinked set.
		 */
		zfs_znode_dmu_fini(zp);
		zfs_znode_free(zp);
		return;
	}

	/*
	 * If the file has extended attributes, we're going to unlink
	 * the xattr dir.
	 */
	if (zp->z_phys->zp_xattr) {
		error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
		ASSERT(error == 0);
	}

	acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj;

	/*
	 * Set up the final transaction.
	 */
	tx = dmu_tx_create(os);
	dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
	if (xzp) {
		dmu_tx_hold_bonus(tx, xzp->z_id);
		dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
	}
	if (acl_obj)
		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
	error = dmu_tx_assign(tx, TXG_WAIT);
	if (error) {
		/*
		 * Not enough space to delete the file.  Leave it in the
		 * unlinked set, leaking it until the fs is remounted (at
		 * which point we'll call zfs_unlinked_drain() to process it).
		 */
		dmu_tx_abort(tx);
		zfs_znode_dmu_fini(zp);
		zfs_znode_free(zp);
		goto out;
	}

	if (xzp) {
//.........这里部分代码省略.........

int
zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
{
    zfsvfs_t *zfsvfs = zp->z_zfsvfs;
    znode_t *xzp;
    dmu_tx_t *tx;
    uint64_t xoid;
    int error;
    zfs_acl_ids_t acl_ids;
    boolean_t fuid_dirtied;

    *xvpp = NULL;

#ifndef __APPLE__
    /* In Mac OS X access preflighting is done above the file system. */
    if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, cr))
        return (error);
#endif /*!__APPLE__*/

    if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
                                    &acl_ids)) != 0)
        return (error);
    if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
        zfs_acl_ids_free(&acl_ids);
        return (EDQUOT);
    }
top:
    tx = dmu_tx_create(zfsvfs->z_os);
    dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
                          ZFS_SA_BASE_ATTR_SIZE);
    dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
    dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
    fuid_dirtied = zfsvfs->z_fuid_dirty;
    if (fuid_dirtied)
        zfs_fuid_txhold(zfsvfs, tx);
    error = dmu_tx_assign(tx, TXG_NOWAIT);
    if (error) {
        if (error == ERESTART) {
            dmu_tx_wait(tx);
            dmu_tx_abort(tx);
            goto top;
        }
        zfs_acl_ids_free(&acl_ids);
        dmu_tx_abort(tx);
        return (error);
    }
    zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);

    if (fuid_dirtied)
        zfs_fuid_sync(zfsvfs, tx);
    VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
                          sizeof (xzp->z_id), tx));

    (void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
                          xzp, "", NULL, acl_ids.z_fuidp, vap);

    zfs_acl_ids_free(&acl_ids);
    dmu_tx_commit(tx);

    /* Cleanup any znode we consumed during zfs_mknode() */
    printf("ZFS_POSTPROCESS_ZP(xzp);\n");
    printf("zfs_dir attach 2\n");
    zfs_attach_vnode(xzp);

    *xvpp = ZTOV(xzp);

    return (0);
}

/**
 * update/insert/delete the specified OI mapping (@fid @id) according to the ops
 *
 * \retval   1, changed nothing
 * \retval   0, changed successfully
 * \retval -ve, on error
 */
static int osd_scrub_refresh_mapping(const struct lu_env *env,
				     struct osd_device *dev,
				     const struct lu_fid *fid,
				     uint64_t oid, int ops,
				     bool force, const char *name)
{
	struct osd_thread_info *info = osd_oti_get(env);
	struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
	char *buf = info->oti_str;
	dmu_tx_t *tx = NULL;
	dnode_t *dn = NULL;
	uint64_t zapid;
	int rc;
	ENTRY;

	if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
		GOTO(log, rc = 0);

	tx = dmu_tx_create(dev->od_os);
	if (!tx)
		GOTO(log, rc = -ENOMEM);

	zapid = osd_get_name_n_idx(env, dev, fid, buf,
				   sizeof(info->oti_str), &dn);
	osd_tx_hold_zap(tx, zapid, dn,
			ops == DTO_INDEX_INSERT ? TRUE : FALSE, NULL);
	rc = -dmu_tx_assign(tx, TXG_WAIT);
	if (rc) {
		dmu_tx_abort(tx);
		GOTO(log, rc);
	}

	switch (ops) {
	case DTO_INDEX_UPDATE:
		zde->zde_pad = 0;
		zde->zde_dnode = oid;
		zde->zde_type = 0; /* The type in OI mapping is useless. */
		rc = -zap_update(dev->od_os, zapid, buf, 8, sizeof(*zde) / 8,
				 zde, tx);
		if (unlikely(rc == -ENOENT)) {
			/* Some unlink thread may removed the OI mapping. */
			rc = 1;
		}
		break;
	case DTO_INDEX_INSERT:
		zde->zde_pad = 0;
		zde->zde_dnode = oid;
		zde->zde_type = 0; /* The type in OI mapping is useless. */
		rc = osd_zap_add(dev, zapid, dn, buf, 8, sizeof(*zde) / 8,
				 zde, tx);
		if (unlikely(rc == -EEXIST))
			rc = 1;
		break;
	case DTO_INDEX_DELETE:
		rc = osd_zap_remove(dev, zapid, dn, buf, tx);
		if (rc == -ENOENT) {
			/* It is normal that the unlink thread has removed the
			 * OI mapping already. */
			rc = 1;
		}
		break;
	default:
		LASSERTF(0, "Unexpected ops %d\n", ops);
		rc = -EINVAL;
		break;
	}

	dmu_tx_commit(tx);
	GOTO(log, rc);

log:
	CDEBUG(D_LFSCK, "%s: refresh OI map for scrub, op %d, force %s, "
	       DFID" => %llu (%s): rc = %d\n", osd_name(dev), ops,
	       force ? "yes" : "no", PFID(fid), oid, name ? name : "null", rc);

	return rc;
}

/*
 * Delete the entire contents of a directory.  Return a count
 * of the number of entries that could not be deleted.
 *
 * NOTE: this function assumes that the directory is inactive,
 *	so there is no need to lock its entries before deletion.
 *	Also, it assumes the directory contents is *only* regular
 *	files.
 */
static int
zfs_purgedir(znode_t *dzp)
{
    zap_cursor_t	zc;
    zap_attribute_t	zap;
    znode_t		*xzp;
    dmu_tx_t	*tx;
    zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
    zfs_dirlock_t	dl;
    int skipped = 0;
    int error;

    for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
            (error = zap_cursor_retrieve(&zc, &zap)) == 0;
            zap_cursor_advance(&zc)) {
#ifdef __APPLE__
        error = zfs_zget_sans_vnode(zfsvfs,  ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
        ASSERT3U(error, ==, 0);

#else
        error = zfs_zget(zfsvfs,
                         ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
        ASSERT3U(error, ==, 0);

        ASSERT((ZTOV(xzp)->v_type == VREG) ||
               (ZTOV(xzp)->v_type == VLNK));
#endif /* __APPLE__ */
        tx = dmu_tx_create(zfsvfs->z_os);
        dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
        dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
        dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
        dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
        /* Is this really needed ? */
        zfs_sa_upgrade_txholds(tx, xzp);

        error = dmu_tx_assign(tx, TXG_WAIT);
        if (error) {
            dmu_tx_abort(tx);
#ifdef __APPLE__
            if (ZTOV(xzp) == NULL) {
                zfs_zinactive(xzp);
            } else {
                VN_RELE(ZTOV(xzp));
            }
#else
            VN_RELE(ZTOV(xzp));
#endif /* __APPLE__ */
            skipped += 1;
            continue;
        }
        bzero(&dl, sizeof (dl));
        dl.dl_dzp = dzp;
        dl.dl_name = zap.za_name;

        error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
        ASSERT3U(error, ==, 0);
        dmu_tx_commit(tx);

#ifdef __APPLE__
        if (ZTOV(xzp) == NULL) {
            zfs_zinactive(xzp);
        } else {
            VN_RELE(ZTOV(xzp));
        }
#else
        VN_RELE(ZTOV(xzp));
#endif /* __APPLE__ */
    }
    zap_cursor_fini(&zc);
    ASSERT(error == ENOENT);
    return (skipped);
}

int
zvol_strategy(buf_t *bp)
{
	zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(bp->b_edev));
	uint64_t off, volsize;
	size_t size, resid;
	char *addr;
	objset_t *os;
	int error = 0;
	int sync;
	int reading;
	int txg_sync_needed = B_FALSE;

	if (zv == NULL) {
		bioerror(bp, ENXIO);
		biodone(bp);
		return (0);
	}

	if (getminor(bp->b_edev) == 0) {
		bioerror(bp, EINVAL);
		biodone(bp);
		return (0);
	}

	if (zv->zv_readonly && !(bp->b_flags & B_READ)) {
		bioerror(bp, EROFS);
		biodone(bp);
		return (0);
	}

	off = ldbtob(bp->b_blkno);
	volsize = zv->zv_volsize;

	os = zv->zv_objset;
	ASSERT(os != NULL);
	sync = !(bp->b_flags & B_ASYNC) && !(zil_disable);

	bp_mapin(bp);
	addr = bp->b_un.b_addr;
	resid = bp->b_bcount;

	/*
	 * There must be no buffer changes when doing a dmu_sync() because
	 * we can't change the data whilst calculating the checksum.
	 * A better approach than a per zvol rwlock would be to lock ranges.
	 */
	reading = bp->b_flags & B_READ;
	if (reading || resid <= zvol_immediate_write_sz)
		rw_enter(&zv->zv_dslock, RW_READER);
	else
		rw_enter(&zv->zv_dslock, RW_WRITER);

	while (resid != 0 && off < volsize) {

		size = MIN(resid, 1UL << 20);	/* cap at 1MB per tx */

		if (size > volsize - off)	/* don't write past the end */
			size = volsize - off;

		if (reading) {
			error = dmu_read(os, ZVOL_OBJ, off, size, addr);
		} else {
			dmu_tx_t *tx = dmu_tx_create(os);
			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
			error = dmu_tx_assign(tx, TXG_WAIT);
			if (error) {
				dmu_tx_abort(tx);
			} else {
				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
				if (sync) {
					/* use the ZIL to commit this write */
					if (zvol_log_write(zv, tx, off, size,
					    addr) != 0) {
						txg_sync_needed = B_TRUE;
					}
				}
				dmu_tx_commit(tx);
			}
		}
		if (error)
			break;
		off += size;
		addr += size;
		resid -= size;
	}
	rw_exit(&zv->zv_dslock);

	if ((bp->b_resid = resid) == bp->b_bcount)
		bioerror(bp, off > volsize ? EINVAL : error);

	biodone(bp);

	if (sync) {
		if (txg_sync_needed)
			txg_wait_synced(dmu_objset_pool(os), 0);
		else
			zil_commit(zv->zv_zilog, UINT64_MAX, 0);
	}

//.........这里部分代码省略.........