static int
vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift)
{
spa_t *spa = vd->vdev_spa;
vdev_disk_t *dvd = vd->vdev_tsd;
vnode_t *devvp = NULLVP;
vfs_context_t context = NULL;
uint64_t blkcnt;
uint32_t blksize;
int fmode = 0;
int error = 0;
int isssd;
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
if (dvd != NULL) {
if (dvd->vd_offline) {
/*
* If we are opening a device in its offline notify
* context, the LDI handle was just closed. Clean
* up the LDI event callbacks and free vd->vdev_tsd.
*/
vdev_disk_free(vd);
} else {
ASSERT(vd->vdev_reopening);
devvp = dvd->vd_devvp;
goto skip_open;
}
}
/*
* Create vd->vdev_tsd.
*/
vdev_disk_alloc(vd);
dvd = vd->vdev_tsd;
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the same device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*/
/* ### APPLE TODO ### */
#ifdef illumos
if (vd->vdev_devid != NULL) {
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
&dvd->vd_minor) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
}
#endif
error = EINVAL; /* presume failure */
if (vd->vdev_path != NULL) {
context = vfs_context_create( spl_vfs_context_kernel() );
/* Obtain an opened/referenced vnode for the device. */
if ((error = vnode_open(vd->vdev_path, spa_mode(spa), 0, 0,
&devvp, context))) {
goto out;
}
if (!vnode_isblk(devvp)) {
error = ENOTBLK;
goto out;
}
/*
* ### APPLE TODO ###
* vnode_authorize devvp for KAUTH_VNODE_READ_DATA and
* KAUTH_VNODE_WRITE_DATA
*/
/*
* Disallow opening of a device that is currently in use.
* Flush out any old buffers remaining from a previous use.
*/
if ((error = vfs_mountedon(devvp))) {
goto out;
}
if (VNOP_FSYNC(devvp, MNT_WAIT, context) != 0) {
error = ENOTBLK;
goto out;
}
if ((error = buf_invalidateblks(devvp, BUF_WRITE_DATA, 0, 0))) {
goto out;
}
} else {
goto out;
}
int len = MAXPATHLEN;
if (vn_getpath(devvp, dvd->vd_readlinkname, &len) == 0) {
dprintf("ZFS: '%s' resolved name is '%s'\n",
vd->vdev_path, dvd->vd_readlinkname);
} else {
dvd->vd_readlinkname[0] = 0;
}
skip_open:
/*
* Determine the actual size of the device.
*/
if (VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blksize, 0,
context) != 0 ||
VNOP_IOCTL(devvp, DKIOCGETBLOCKCOUNT, (caddr_t)&blkcnt, 0,
context) != 0) {
error = EINVAL;
goto out;
}
*psize = blkcnt * (uint64_t)blksize;
*max_psize = *psize;
dvd->vd_ashift = highbit(blksize) - 1;
dprintf("vdev_disk: Device %p ashift set to %d\n", devvp,
dvd->vd_ashift);
*ashift = highbit(MAX(blksize, SPA_MINBLOCKSIZE)) - 1;
/*
* ### APPLE TODO ###
*/
#ifdef illumos
if (vd->vdev_wholedisk == 1) {
int wce = 1;
if (error == 0) {
/*
* If we have the capability to expand, we'd have
* found out via success from DKIOCGMEDIAINFO{,EXT}.
* Adjust max_psize upward accordingly since we know
* we own the whole disk now.
*/
*max_psize = capacity * blksz;
}
/*
* Since we own the whole disk, try to enable disk write
* caching. We ignore errors because it's OK if we can't do it.
*/
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
FKIOCTL, kcred, NULL);
}
#endif
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
/* Inform the ZIO pipeline that we are non-rotational */
vd->vdev_nonrot = B_FALSE;
if (VNOP_IOCTL(devvp, DKIOCISSOLIDSTATE, (caddr_t)&isssd, 0,
context) == 0) {
if (isssd)
vd->vdev_nonrot = B_TRUE;
}
dprintf("ZFS: vdev_disk(%s) isSSD %d\n", vd->vdev_path ? vd->vdev_path : "",
isssd);
dvd->vd_devvp = devvp;
out:
if (error) {
if (devvp) {
vnode_close(devvp, fmode, context);
dvd->vd_devvp = NULL;
}
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
}
if (context)
(void) vfs_context_rele(context);
if (error) printf("ZFS: vdev_disk_open('%s') failed error %d\n",
vd->vdev_path ? vd->vdev_path : "", error);
return (error);
}
/*
* Device close routine
*/
int
spec_close(struct vnop_close_args *ap)
{
struct vnode *vp = ap->a_vp;
dev_t dev = vp->v_rdev;
int (*devclose)(dev_t, int, int, struct proc *);
int mode, error;
int flags = ap->a_fflag;
struct proc *p = vfs_context_proc(ap->a_context);
struct session *sessp;
switch (vp->v_type) {
case VCHR:
/*
* Hack: a tty device that is a controlling terminal
* has a reference from the session structure.
* We cannot easily tell that a character device is
* a controlling terminal, unless it is the closing
* process' controlling terminal. In that case,
* if the reference count is 1 (this is the very
* last close)
*/
sessp = proc_session(p);
if (sessp != SESSION_NULL) {
if ((vcount(vp) == 1) &&
(vp == sessp->s_ttyvp)) {
session_lock(sessp);
sessp->s_ttyvp = NULL;
sessp->s_ttyvid = 0;
sessp->s_ttyp = TTY_NULL;
sessp->s_ttypgrpid = NO_PID;
session_unlock(sessp);
vnode_rele(vp);
}
session_rele(sessp);
}
devclose = cdevsw[major(dev)].d_close;
mode = S_IFCHR;
/*
* close on last reference or on vnode revoke call
*/
if ((flags & IO_REVOKE) != 0)
break;
if (vcount(vp) > 0)
return (0);
break;
case VBLK:
/*
* Since every use (buffer, vnode, swap, blockmap)
* holds a reference to the vnode, and because we mark
* any other vnodes that alias this device, when the
* sum of the reference counts on all the aliased
* vnodes descends to zero, we are on last close.
*/
if (vcount(vp) > 0)
return (0);
/*
* On last close of a block device (that isn't mounted)
* we must invalidate any in core blocks, so that
* we can, for instance, change floppy disks.
*/
if ((error = spec_fsync_internal(vp, MNT_WAIT, ap->a_context)))
return (error);
error = buf_invalidateblks(vp, BUF_WRITE_DATA, 0, 0);
if (error)
return (error);
devclose = bdevsw[major(dev)].d_close;
mode = S_IFBLK;
break;
default:
panic("spec_close: not special");
return(EBADF);
}
return ((*devclose)(dev, flags, mode, p));
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *size, uint64_t *max_size, uint64_t *ashift)
{
vdev_disk_t *dvd = NULL;
vnode_t *devvp = NULLVP;
vfs_context_t context = NULL;
uint64_t blkcnt;
uint32_t blksize;
int fmode = 0;
int error = 0;
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (EINVAL);
}
dvd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
if (dvd == NULL)
return ENOMEM;
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the save device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*
*/
/* ### APPLE TODO ### */
/* ddi_devid_str_decode */
context = vfs_context_create((vfs_context_t)0);
/* Obtain an opened/referenced vnode for the device. */
error = vnode_open(vd->vdev_path, spa_mode(vd->vdev_spa), 0, 0, &devvp, context);
if (error) {
goto out;
}
if (!vnode_isblk(devvp)) {
error = ENOTBLK;
goto out;
}
/* ### APPLE TODO ### */
/* vnode_authorize devvp for KAUTH_VNODE_READ_DATA and
* KAUTH_VNODE_WRITE_DATA
*/
/*
* Disallow opening of a device that is currently in use.
* Flush out any old buffers remaining from a previous use.
*/
if ((error = vfs_mountedon(devvp))) {
goto out;
}
if (VNOP_FSYNC(devvp, MNT_WAIT, context) != 0) {
error = ENOTBLK;
goto out;
}
if ((error = buf_invalidateblks(devvp, BUF_WRITE_DATA, 0, 0))) {
goto out;
}
/*
* Determine the actual size of the device.
*/
if (VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blksize, 0, context)
!= 0 || VNOP_IOCTL(devvp, DKIOCGETBLOCKCOUNT, (caddr_t)&blkcnt,
0, context) != 0) {
error = EINVAL;
goto out;
}
*size = blkcnt * (uint64_t)blksize;
/*
* ### APPLE TODO ###
* If we own the whole disk, try to enable disk write caching.
*/
/*
* Take the device's minimum transfer size into account.
*/
*ashift = highbit(MAX(blksize, SPA_MINBLOCKSIZE)) - 1;
/*
* Setting the vdev_ashift did in fact break the pool for import
* on ZEVO. This puts the logic into question. It appears that vdev_top
* will also then change. It then panics in space_map from metaslab_alloc
*/
//vd->vdev_ashift = *ashift;
dvd->vd_ashift = *ashift;
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
vd->vdev_tsd = dvd;
dvd->vd_devvp = devvp;
out:
if (error) {
if (devvp)
vnode_close(devvp, fmode, context);
if (dvd)
kmem_free(dvd, sizeof (vdev_disk_t));
/*
* Since the open has failed, vd->vdev_tsd should
* be NULL when we get here, signaling to the
* rest of the spa not to try and reopen or close this device
*/
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
}
if (context) {
(void) vfs_context_rele(context);
}
return (error);
}
/*
* Function: devfs_kernel_mount
* Purpose:
* Mount devfs at the given mount point from within the kernel.
*/
int
devfs_kernel_mount(char * mntname)
{
struct mount *mp;
int error;
struct nameidata nd;
struct vnode * vp;
vfs_context_t ctx = vfs_context_kernel();
struct vfstable *vfsp;
/* Find our vfstable entry */
for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
if (!strncmp(vfsp->vfc_name, "devfs", sizeof(vfsp->vfc_name)))
break;
if (!vfsp) {
panic("Could not find entry in vfsconf for devfs.\n");
}
/*
* Get vnode to be covered
*/
NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
CAST_USER_ADDR_T(mntname), ctx);
if ((error = namei(&nd))) {
printf("devfs_kernel_mount: failed to find directory '%s', %d",
mntname, error);
return (error);
}
nameidone(&nd);
vp = nd.ni_vp;
if ((error = VNOP_FSYNC(vp, MNT_WAIT, ctx))) {
printf("devfs_kernel_mount: vnop_fsync failed: %d\n", error);
vnode_put(vp);
return (error);
}
if ((error = buf_invalidateblks(vp, BUF_WRITE_DATA, 0, 0))) {
printf("devfs_kernel_mount: buf_invalidateblks failed: %d\n", error);
vnode_put(vp);
return (error);
}
if (vnode_isdir(vp) == 0) {
printf("devfs_kernel_mount: '%s' is not a directory\n", mntname);
vnode_put(vp);
return (ENOTDIR);
}
if ((vnode_mountedhere(vp))) {
vnode_put(vp);
return (EBUSY);
}
/*
* Allocate and initialize the filesystem.
*/
MALLOC_ZONE(mp, struct mount *, sizeof(struct mount),
M_MOUNT, M_WAITOK);
bzero((char *)mp, sizeof(struct mount));
/* Initialize the default IO constraints */
mp->mnt_maxreadcnt = mp->mnt_maxwritecnt = MAXPHYS;
mp->mnt_segreadcnt = mp->mnt_segwritecnt = 32;
mp->mnt_ioflags = 0;
mp->mnt_realrootvp = NULLVP;
mp->mnt_authcache_ttl = CACHED_LOOKUP_RIGHT_TTL;
mount_lock_init(mp);
TAILQ_INIT(&mp->mnt_vnodelist);
TAILQ_INIT(&mp->mnt_workerqueue);
TAILQ_INIT(&mp->mnt_newvnodes);
(void)vfs_busy(mp, LK_NOWAIT);
mp->mnt_op = &devfs_vfsops;
mp->mnt_vtable = vfsp;
mp->mnt_flag = 0;
mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
strlcpy(mp->mnt_vfsstat.f_fstypename, vfsp->vfc_name, MFSTYPENAMELEN);
vp->v_mountedhere = mp;
mp->mnt_vnodecovered = vp;
mp->mnt_vfsstat.f_owner = kauth_cred_getuid(kauth_cred_get());
(void) copystr(mntname, mp->mnt_vfsstat.f_mntonname, MAXPATHLEN - 1, 0);
#if CONFIG_MACF
mac_mount_label_init(mp);
mac_mount_label_associate(ctx, mp);
#endif
error = devfs_mount(mp, NULL, USER_ADDR_NULL, ctx);
if (error) {
printf("devfs_kernel_mount: mount %s failed: %d", mntname, error);
mp->mnt_vtable->vfc_refcount--;
vfs_unbusy(mp);
mount_lock_destroy(mp);
#if CONFIG_MACF
mac_mount_label_destroy(mp);
#endif
FREE_ZONE(mp, sizeof (struct mount), M_MOUNT);
vnode_put(vp);
return (error);
}
vnode_ref(vp);
vnode_put(vp);
vfs_unbusy(mp);
mount_list_add(mp);
return (0);
}
