vdev_file.c revision ea8dc4b6d2251b437950c0056bc626b311c73c27
fa9e4066f08beec538e775443c5be79dd423fcabahrens/*
fa9e4066f08beec538e775443c5be79dd423fcabahrens * CDDL HEADER START
fa9e4066f08beec538e775443c5be79dd423fcabahrens *
fa9e4066f08beec538e775443c5be79dd423fcabahrens * The contents of this file are subject to the terms of the
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * Common Development and Distribution License (the "License").
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * You may not use this file except in compliance with the License.
fa9e4066f08beec538e775443c5be79dd423fcabahrens *
fa9e4066f08beec538e775443c5be79dd423fcabahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
fa9e4066f08beec538e775443c5be79dd423fcabahrens * or http://www.opensolaris.org/os/licensing.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * See the License for the specific language governing permissions
fa9e4066f08beec538e775443c5be79dd423fcabahrens * and limitations under the License.
fa9e4066f08beec538e775443c5be79dd423fcabahrens *
fa9e4066f08beec538e775443c5be79dd423fcabahrens * When distributing Covered Code, include this CDDL HEADER in each
fa9e4066f08beec538e775443c5be79dd423fcabahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If applicable, add the following below this CDDL HEADER, with the
fa9e4066f08beec538e775443c5be79dd423fcabahrens * fields enclosed by brackets "[]" replaced with your own identifying
fa9e4066f08beec538e775443c5be79dd423fcabahrens * information: Portions Copyright [yyyy] [name of copyright owner]
fa9e4066f08beec538e775443c5be79dd423fcabahrens *
fa9e4066f08beec538e775443c5be79dd423fcabahrens * CDDL HEADER END
fa9e4066f08beec538e775443c5be79dd423fcabahrens */
fa9e4066f08beec538e775443c5be79dd423fcabahrens/*
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * Copyright 2006 Sun Microsystems, Inc. All rights reserved.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Use is subject to license terms.
fa9e4066f08beec538e775443c5be79dd423fcabahrens */
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens#pragma ident "%Z%%M% %I% %E% SMI"
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens#include <sys/zfs_context.h>
fa9e4066f08beec538e775443c5be79dd423fcabahrens#include <sys/spa.h>
fa9e4066f08beec538e775443c5be79dd423fcabahrens#include <sys/vdev_file.h>
fa9e4066f08beec538e775443c5be79dd423fcabahrens#include <sys/vdev_impl.h>
fa9e4066f08beec538e775443c5be79dd423fcabahrens#include <sys/zio.h>
fa9e4066f08beec538e775443c5be79dd423fcabahrens#include <sys/fs/zfs.h>
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens/*
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Virtual device vector for files.
fa9e4066f08beec538e775443c5be79dd423fcabahrens */
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrensstatic int
fa9e4066f08beec538e775443c5be79dd423fcabahrensvdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
fa9e4066f08beec538e775443c5be79dd423fcabahrens{
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_t *vf;
fa9e4066f08beec538e775443c5be79dd423fcabahrens vnode_t *vp;
fa9e4066f08beec538e775443c5be79dd423fcabahrens vattr_t vattr;
fa9e4066f08beec538e775443c5be79dd423fcabahrens int error;
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens /*
fa9e4066f08beec538e775443c5be79dd423fcabahrens * We must have a pathname, and it must be absolute.
fa9e4066f08beec538e775443c5be79dd423fcabahrens */
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
fa9e4066f08beec538e775443c5be79dd423fcabahrens vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (EINVAL);
fa9e4066f08beec538e775443c5be79dd423fcabahrens }
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens#ifdef _KERNEL
fa9e4066f08beec538e775443c5be79dd423fcabahrens /*
fa9e4066f08beec538e775443c5be79dd423fcabahrens * When using a file vdev in kernel context, the underlying filesystem
fa9e4066f08beec538e775443c5be79dd423fcabahrens * will already be caching the data. Don't cache it again here.
fa9e4066f08beec538e775443c5be79dd423fcabahrens */
fa9e4066f08beec538e775443c5be79dd423fcabahrens vd->vdev_cache.vc_size = 0;
fa9e4066f08beec538e775443c5be79dd423fcabahrens#endif
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens /*
fa9e4066f08beec538e775443c5be79dd423fcabahrens * We always open the files from the root of the global zone, even if
fa9e4066f08beec538e775443c5be79dd423fcabahrens * we're in a local zone. If the user has gotten to this point, the
fa9e4066f08beec538e775443c5be79dd423fcabahrens * administrator has already decided that the pool should be available
fa9e4066f08beec538e775443c5be79dd423fcabahrens * to local zone users, so the underlying devices should be as well.
fa9e4066f08beec538e775443c5be79dd423fcabahrens */
fa9e4066f08beec538e775443c5be79dd423fcabahrens ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
fa9e4066f08beec538e775443c5be79dd423fcabahrens error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE, spa_mode | FOFFMAX,
fa9e4066f08beec538e775443c5be79dd423fcabahrens 0, &vp, 0, 0, rootdir);
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (error) {
fa9e4066f08beec538e775443c5be79dd423fcabahrens vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (error);
fa9e4066f08beec538e775443c5be79dd423fcabahrens }
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens vf->vf_vnode = vp;
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens#ifdef _KERNEL
fa9e4066f08beec538e775443c5be79dd423fcabahrens /*
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Make sure it's a regular file.
fa9e4066f08beec538e775443c5be79dd423fcabahrens */
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (vp->v_type != VREG) {
fa9e4066f08beec538e775443c5be79dd423fcabahrens vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (ENODEV);
fa9e4066f08beec538e775443c5be79dd423fcabahrens }
fa9e4066f08beec538e775443c5be79dd423fcabahrens#endif
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens /*
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Determine the physical size of the file.
fa9e4066f08beec538e775443c5be79dd423fcabahrens */
fa9e4066f08beec538e775443c5be79dd423fcabahrens vattr.va_mask = AT_SIZE;
fa9e4066f08beec538e775443c5be79dd423fcabahrens error = VOP_GETATTR(vp, &vattr, 0, kcred);
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (error) {
fa9e4066f08beec538e775443c5be79dd423fcabahrens vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (error);
fa9e4066f08beec538e775443c5be79dd423fcabahrens }
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens *psize = vattr.va_size;
fa9e4066f08beec538e775443c5be79dd423fcabahrens *ashift = SPA_MINBLOCKSHIFT;
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (0);
fa9e4066f08beec538e775443c5be79dd423fcabahrens}
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrensstatic void
fa9e4066f08beec538e775443c5be79dd423fcabahrensvdev_file_close(vdev_t *vd)
fa9e4066f08beec538e775443c5be79dd423fcabahrens{
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_t *vf = vd->vdev_tsd;
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (vf == NULL)
fa9e4066f08beec538e775443c5be79dd423fcabahrens return;
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (vf->vf_vnode != NULL) {
fa9e4066f08beec538e775443c5be79dd423fcabahrens (void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred);
fa9e4066f08beec538e775443c5be79dd423fcabahrens (void) VOP_CLOSE(vf->vf_vnode, spa_mode, 1, 0, kcred);
fa9e4066f08beec538e775443c5be79dd423fcabahrens VN_RELE(vf->vf_vnode);
fa9e4066f08beec538e775443c5be79dd423fcabahrens }
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens kmem_free(vf, sizeof (vdev_file_t));
fa9e4066f08beec538e775443c5be79dd423fcabahrens vd->vdev_tsd = NULL;
fa9e4066f08beec538e775443c5be79dd423fcabahrens}
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrensstatic void
fa9e4066f08beec538e775443c5be79dd423fcabahrensvdev_file_io_start(zio_t *zio)
fa9e4066f08beec538e775443c5be79dd423fcabahrens{
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_t *vd = zio->io_vd;
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_t *vf = vd->vdev_tsd;
fa9e4066f08beec538e775443c5be79dd423fcabahrens ssize_t resid;
fa9e4066f08beec538e775443c5be79dd423fcabahrens int error;
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (zio->io_type == ZIO_TYPE_IOCTL) {
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio_vdev_io_bypass(zio);
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens /* XXPOLICY */
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (vdev_is_dead(vd)) {
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio->io_error = ENXIO;
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio_next_stage_async(zio);
fa9e4066f08beec538e775443c5be79dd423fcabahrens return;
fa9e4066f08beec538e775443c5be79dd423fcabahrens }
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens switch (zio->io_cmd) {
fa9e4066f08beec538e775443c5be79dd423fcabahrens case DKIOCFLUSHWRITECACHE:
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
fa9e4066f08beec538e775443c5be79dd423fcabahrens kcred);
fa9e4066f08beec538e775443c5be79dd423fcabahrens dprintf("fsync(%s) = %d\n", vdev_description(vd),
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio->io_error);
fa9e4066f08beec538e775443c5be79dd423fcabahrens break;
fa9e4066f08beec538e775443c5be79dd423fcabahrens default:
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio->io_error = ENOTSUP;
fa9e4066f08beec538e775443c5be79dd423fcabahrens }
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio_next_stage_async(zio);
fa9e4066f08beec538e775443c5be79dd423fcabahrens return;
fa9e4066f08beec538e775443c5be79dd423fcabahrens }
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0)
fa9e4066f08beec538e775443c5be79dd423fcabahrens return;
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens if ((zio = vdev_queue_io(zio)) == NULL)
fa9e4066f08beec538e775443c5be79dd423fcabahrens return;
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens /* XXPOLICY */
fa9e4066f08beec538e775443c5be79dd423fcabahrens error = vdev_is_dead(vd) ? ENXIO : vdev_error_inject(vd, zio);
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (error) {
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio->io_error = error;
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio_next_stage_async(zio);
fa9e4066f08beec538e775443c5be79dd423fcabahrens return;
fa9e4066f08beec538e775443c5be79dd423fcabahrens }
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
fa9e4066f08beec538e775443c5be79dd423fcabahrens UIO_READ : UIO_WRITE, vf->vf_vnode, zio->io_data,
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio->io_size, zio->io_offset, UIO_SYSSPACE,
fa9e4066f08beec538e775443c5be79dd423fcabahrens 0, RLIM64_INFINITY, kcred, &resid);
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (resid != 0 && zio->io_error == 0)
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio->io_error = ENOSPC;
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio_next_stage_async(zio);
fa9e4066f08beec538e775443c5be79dd423fcabahrens}
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrensstatic void
fa9e4066f08beec538e775443c5be79dd423fcabahrensvdev_file_io_done(zio_t *zio)
fa9e4066f08beec538e775443c5be79dd423fcabahrens{
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_queue_io_done(zio);
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (zio->io_type == ZIO_TYPE_WRITE)
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_cache_write(zio);
fa9e4066f08beec538e775443c5be79dd423fcabahrens
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock if (zio_injection_enabled && zio->io_error == 0)
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock zio->io_error = zio_handle_device_injection(zio->io_vd, EIO);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock
fa9e4066f08beec538e775443c5be79dd423fcabahrens zio_next_stage(zio);
fa9e4066f08beec538e775443c5be79dd423fcabahrens}
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrensvdev_ops_t vdev_file_ops = {
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_open,
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_close,
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_default_asize,
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_io_start,
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_io_done,
fa9e4066f08beec538e775443c5be79dd423fcabahrens NULL,
fa9e4066f08beec538e775443c5be79dd423fcabahrens VDEV_TYPE_FILE, /* name of this vdev type */
fa9e4066f08beec538e775443c5be79dd423fcabahrens B_TRUE /* leaf vdev */
fa9e4066f08beec538e775443c5be79dd423fcabahrens};
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens/*
fa9e4066f08beec538e775443c5be79dd423fcabahrens * From userland we access disks just like files.
fa9e4066f08beec538e775443c5be79dd423fcabahrens */
fa9e4066f08beec538e775443c5be79dd423fcabahrens#ifndef _KERNEL
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrensvdev_ops_t vdev_disk_ops = {
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_open,
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_close,
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_default_asize,
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_io_start,
fa9e4066f08beec538e775443c5be79dd423fcabahrens vdev_file_io_done,
fa9e4066f08beec538e775443c5be79dd423fcabahrens NULL,
fa9e4066f08beec538e775443c5be79dd423fcabahrens VDEV_TYPE_DISK, /* name of this vdev type */
fa9e4066f08beec538e775443c5be79dd423fcabahrens B_TRUE /* leaf vdev */
fa9e4066f08beec538e775443c5be79dd423fcabahrens};
fa9e4066f08beec538e775443c5be79dd423fcabahrens
fa9e4066f08beec538e775443c5be79dd423fcabahrens#endif