/*
* Create a reference to the root of a mounted file descriptor.
* This routine is called from lookupname() in the event a path
* is being searched that has a mounted file descriptor in it.
*/
static int
nm_root(vfs_t *vfsp, vnode_t **vpp)
{
struct namenode *nodep = (struct namenode *)vfsp->vfs_data;
struct vnode *vp = NMTOV(nodep);
VN_HOLD(vp);
*vpp = vp;
return (0);
}
/*
* Unmount a file descriptor from a node in the file system.
* If the user is not the owner of the file and is not privileged,
* the request is denied.
* Otherwise, remove the namenode from the hash list.
* If the mounted file descriptor was that of a stream and this
* was the last mount of the stream, turn off the STRMOUNT flag.
* If the rootvp is referenced other than through the mount,
* nm_inactive will clean up.
*/
static int
nm_unmount(vfs_t *vfsp, int flag, cred_t *crp)
{
struct namenode *nodep = (struct namenode *)vfsp->vfs_data;
vnode_t *vp, *thisvp;
struct file *fp = NULL;
ASSERT((nodep->nm_flag & NMNMNT) == 0);
/*
* forced unmount is not supported by this file system
* and thus, ENOTSUP, is being returned.
*/
if (flag & MS_FORCE) {
return (ENOTSUP);
}
vp = nodep->nm_filevp;
mutex_enter(&nodep->nm_lock);
if (secpolicy_vnode_owner(crp, nodep->nm_vattr.va_uid) != 0) {
mutex_exit(&nodep->nm_lock);
return (EPERM);
}
mutex_exit(&nodep->nm_lock);
mutex_enter(&ntable_lock);
nameremove(nodep);
thisvp = NMTOV(nodep);
mutex_enter(&thisvp->v_lock);
if (thisvp->v_count-- == 1) {
fp = nodep->nm_filep;
mutex_exit(&thisvp->v_lock);
vn_invalid(thisvp);
vn_free(thisvp);
VFS_RELE(vfsp);
namenodeno_free(nodep->nm_vattr.va_nodeid);
kmem_free(nodep, sizeof (struct namenode));
} else {
thisvp->v_flag &= ~VROOT;
mutex_exit(&thisvp->v_lock);
}
if (namefind(vp, NULLVP) == NULL && vp->v_stream) {
struct stdata *stp = vp->v_stream;
mutex_enter(&stp->sd_lock);
stp->sd_flag &= ~STRMOUNT;
mutex_exit(&stp->sd_lock);
}
mutex_exit(&ntable_lock);
if (fp != NULL)
(void) closef(fp);
return (0);
}
/*
* Standard access() like check. Figure out which mode bits apply
* to the caller then pass the missing mode bits to the secpolicy function.
*/
static int
nm_access_unlocked(void *vnp, int mode, cred_t *crp)
{
struct namenode *nodep = vnp;
int shift = 0;
if (crgetuid(crp) != nodep->nm_vattr.va_uid) {
shift += 3;
if (!groupmember(nodep->nm_vattr.va_gid, crp))
shift += 3;
}
return (secpolicy_vnode_access2(crp, NMTOV(nodep),
nodep->nm_vattr.va_uid, nodep->nm_vattr.va_mode << shift,
mode));
}
/*
* Mount a file descriptor onto the node in the file system.
* Create a new vnode, update the attributes with info from the
* file descriptor and the mount point. The mask, mode, uid, gid,
* atime, mtime and ctime are taken from the mountpt. Link count is
* set to one, the file system id is namedev and nodeid is unique
* for each mounted object. Other attributes are taken from mount point.
* Make sure user is owner (or root) with write permissions on mount point.
* Hash the new vnode and return 0.
* Upon entry to this routine, the file descriptor is in the
* fd field of a struct namefd. Copy that structure from user
* space and retrieve the file descriptor.
*/
static int
nm_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *crp)
{
struct namefd namefdp;
struct vnode *filevp; /* file descriptor vnode */
struct file *fp;
struct vnode *newvp; /* vnode representing this mount */
struct vnode *rvp; /* realvp (if any) for the mountpt */
struct namenode *nodep; /* namenode for this mount */
struct vattr filevattr; /* attributes of file dec. */
struct vattr *vattrp; /* attributes of this mount */
char *resource_name;
char *resource_nodetype;
statvfs64_t *svfsp;
int error = 0;
/*
* Get the file descriptor from user space.
* Make sure the file descriptor is valid and has an
* associated file pointer.
* If so, extract the vnode from the file pointer.
*/
if (uap->datalen != sizeof (struct namefd))
return (EINVAL);
if (copyin(uap->dataptr, &namefdp, uap->datalen))
return (EFAULT);
if ((fp = getf(namefdp.fd)) == NULL)
return (EBADF);
/*
* If the mount point already has something mounted
* on it, disallow this mount. (This restriction may
* be removed in a later release).
* Or unmount has completed but the namefs ROOT vnode
* count has not decremented to zero, disallow this mount.
*/
mutex_enter(&mvp->v_lock);
if ((mvp->v_flag & VROOT) ||
vfs_matchops(mvp->v_vfsp, namefs_vfsops)) {
mutex_exit(&mvp->v_lock);
releasef(namefdp.fd);
return (EBUSY);
}
mutex_exit(&mvp->v_lock);
/*
* Cannot allow users to fattach() in /dev/pts.
* First, there is no need for doing so and secondly
* we cannot allow arbitrary users to park on a node in
* /dev/pts or /dev/vt.
*/
rvp = NULLVP;
if (vn_matchops(mvp, spec_getvnodeops()) &&
VOP_REALVP(mvp, &rvp, NULL) == 0 && rvp &&
(vn_matchops(rvp, devpts_getvnodeops()) ||
vn_matchops(rvp, devvt_getvnodeops()))) {
releasef(namefdp.fd);
return (ENOTSUP);
}
filevp = fp->f_vnode;
if (filevp->v_type == VDIR || filevp->v_type == VPORT) {
releasef(namefdp.fd);
return (EINVAL);
}
/*
* If the fd being mounted refers to neither a door nor a stream,
* make sure the caller is privileged.
*/
if (filevp->v_type != VDOOR && filevp->v_stream == NULL) {
if (secpolicy_fs_mount(crp, filevp, vfsp) != 0) {
/* fd is neither a stream nor a door */
releasef(namefdp.fd);
return (EINVAL);
}
}
/*
* Make sure the file descriptor is not the root of some
* file system.
* If it's not, create a reference and allocate a namenode
* to represent this mount request.
*/
if (filevp->v_flag & VROOT) {
releasef(namefdp.fd);
return (EBUSY);
}
nodep = kmem_zalloc(sizeof (struct namenode), KM_SLEEP);
mutex_init(&nodep->nm_lock, NULL, MUTEX_DEFAULT, NULL);
vattrp = &nodep->nm_vattr;
vattrp->va_mask = AT_ALL;
if (error = VOP_GETATTR(mvp, vattrp, 0, crp, NULL))
goto out;
filevattr.va_mask = AT_ALL;
if (error = VOP_GETATTR(filevp, &filevattr, 0, crp, NULL))
goto out;
/*
* Make sure the user is the owner of the mount point
* or has sufficient privileges.
*/
if (error = secpolicy_vnode_owner(crp, vattrp->va_uid))
goto out;
/*
* Make sure the user has write permissions on the
* mount point (or has sufficient privileges).
*/
if (!(vattrp->va_mode & VWRITE) &&
secpolicy_vnode_access(crp, mvp, vattrp->va_uid, VWRITE) != 0) {
error = EACCES;
goto out;
}
/*
* If the file descriptor has file/record locking, don't
* allow the mount to succeed.
*/
if (vn_has_flocks(filevp)) {
error = EACCES;
goto out;
}
/*
* Initialize the namenode.
*/
if (filevp->v_stream) {
struct stdata *stp = filevp->v_stream;
mutex_enter(&stp->sd_lock);
stp->sd_flag |= STRMOUNT;
mutex_exit(&stp->sd_lock);
}
nodep->nm_filevp = filevp;
mutex_enter(&fp->f_tlock);
fp->f_count++;
mutex_exit(&fp->f_tlock);
releasef(namefdp.fd);
nodep->nm_filep = fp;
nodep->nm_mountpt = mvp;
/*
* The attributes for the mounted file descriptor were initialized
* above by applying VOP_GETATTR to the mount point. Some of
* the fields of the attributes structure will be overwritten
* by the attributes from the file descriptor.
*/
vattrp->va_type = filevattr.va_type;
vattrp->va_fsid = namedev;
vattrp->va_nodeid = namenodeno_alloc();
vattrp->va_nlink = 1;
vattrp->va_size = filevattr.va_size;
vattrp->va_rdev = filevattr.va_rdev;
vattrp->va_blksize = filevattr.va_blksize;
vattrp->va_nblocks = filevattr.va_nblocks;
vattrp->va_seq = 0;
/*
* Initialize new vnode structure for the mounted file descriptor.
*/
nodep->nm_vnode = vn_alloc(KM_SLEEP);
newvp = NMTOV(nodep);
newvp->v_flag = filevp->v_flag | VROOT | VNOMAP | VNOSWAP;
vn_setops(newvp, nm_vnodeops);
newvp->v_vfsp = vfsp;
newvp->v_stream = filevp->v_stream;
newvp->v_type = filevp->v_type;
newvp->v_rdev = filevp->v_rdev;
newvp->v_data = (caddr_t)nodep;
VFS_HOLD(vfsp);
vn_exists(newvp);
/*
* Initialize the vfs structure.
*/
vfsp->vfs_vnodecovered = NULL;
vfsp->vfs_flag |= VFS_UNLINKABLE;
vfsp->vfs_bsize = 1024;
vfsp->vfs_fstype = namefstype;
vfs_make_fsid(&vfsp->vfs_fsid, namedev, namefstype);
vfsp->vfs_data = (caddr_t)nodep;
vfsp->vfs_dev = namedev;
vfsp->vfs_bcount = 0;
/*
* Set the name we mounted from.
*/
switch (filevp->v_type) {
case VPROC: /* VOP_GETATTR() translates this to VREG */
case VREG: resource_nodetype = "file"; break;
case VDIR: resource_nodetype = "directory"; break;
case VBLK: resource_nodetype = "device"; break;
case VCHR: resource_nodetype = "device"; break;
case VLNK: resource_nodetype = "link"; break;
case VFIFO: resource_nodetype = "fifo"; break;
case VDOOR: resource_nodetype = "door"; break;
case VSOCK: resource_nodetype = "socket"; break;
default: resource_nodetype = "resource"; break;
}
#define RESOURCE_NAME_SZ 128 /* Maximum length of the resource name */
resource_name = kmem_alloc(RESOURCE_NAME_SZ, KM_SLEEP);
svfsp = kmem_alloc(sizeof (statvfs64_t), KM_SLEEP);
error = VFS_STATVFS(filevp->v_vfsp, svfsp);
if (error == 0) {
(void) snprintf(resource_name, RESOURCE_NAME_SZ,
"unspecified_%s_%s", svfsp->f_basetype, resource_nodetype);
} else {
(void) snprintf(resource_name, RESOURCE_NAME_SZ,
"unspecified_%s", resource_nodetype);
}
vfs_setresource(vfsp, resource_name);
kmem_free(svfsp, sizeof (statvfs64_t));
kmem_free(resource_name, RESOURCE_NAME_SZ);
#undef RESOURCE_NAME_SZ
/*
* Insert the namenode.
*/
mutex_enter(&ntable_lock);
nameinsert(nodep);
mutex_exit(&ntable_lock);
return (0);
out:
releasef(namefdp.fd);
kmem_free(nodep, sizeof (struct namenode));
return (error);
}
/*
* Force the unmouting of a file descriptor from ALL of the nodes
* that it was mounted to.
* At the present time, the only usage for this routine is in the
* event one end of a pipe was mounted. At the time the unmounted
* end gets closed down, the mounted end is forced to be unmounted.
*
* This routine searches the namenode hash list for all namenodes
* that have a nm_filevp field equal to vp. Each time one is found,
* the dounmount() routine is called. This causes the nm_unmount()
* routine to be called and thus, the file descriptor is unmounted
* from the node.
*
* At the start of this routine, the reference count for vp is
* incremented to protect the vnode from being released in the
* event the mount was the only thing keeping the vnode active.
* If that is the case, the VOP_CLOSE operation is applied to
* the vnode, prior to it being released.
*/
static int
nm_umountall(vnode_t *vp, cred_t *crp)
{
vfs_t *vfsp;
struct namenode *nodep;
int error = 0;
int realerr = 0;
/*
* For each namenode that is associated with the file:
* If the v_vfsp field is not namevfs, dounmount it. Otherwise,
* it was created in nm_open() and will be released in time.
* The following loop replicates some code from nm_find. That
* routine can't be used as is since the list isn't strictly
* consumed as it is traversed.
*/
mutex_enter(&ntable_lock);
nodep = *NM_FILEVP_HASH(vp);
while (nodep) {
if (nodep->nm_filevp == vp &&
(vfsp = NMTOV(nodep)->v_vfsp) != NULL &&
vfsp != &namevfs && (NMTOV(nodep)->v_flag & VROOT)) {
/*
* If the vn_vfswlock fails, skip the vfs since
* somebody else may be unmounting it.
*/
if (vn_vfswlock(vfsp->vfs_vnodecovered)) {
realerr = EBUSY;
nodep = nodep->nm_nextp;
continue;
}
/*
* Can't hold ntable_lock across call to do_unmount
* because nm_unmount tries to acquire it. This means
* there is a window where another mount of vp can
* happen so it is possible that after nm_unmountall
* there are still some mounts. This situation existed
* without MT locking because dounmount can sleep
* so another mount could happen during that time.
* This situation is unlikely and doesn't really cause
* any problems.
*/
mutex_exit(&ntable_lock);
if ((error = dounmount(vfsp, 0, crp)) != 0)
realerr = error;
mutex_enter(&ntable_lock);
/*
* Since we dropped the ntable_lock, we
* have to start over from the beginning.
* If for some reasons dounmount() fails,
* start from beginning means that we will keep on
* trying unless another thread unmounts it for us.
*/
nodep = *NM_FILEVP_HASH(vp);
} else
nodep = nodep->nm_nextp;
}
mutex_exit(&ntable_lock);
return (realerr);
}
/*
* Create a reference to the vnode representing the file descriptor.
* Then, apply the VOP_OPEN operation to that vnode.
*
* The vnode for the file descriptor may be switched under you.
* If it is, search the hash list for an nodep - nodep->nm_filevp
* pair. If it exists, return that nodep to the user.
* If it does not exist, create a new namenode to attach
* to the nodep->nm_filevp then place the pair on the hash list.
*
* Newly created objects are like children/nodes in the mounted
* file system, with the parent being the initial mount.
*/
int
nm_open(vnode_t **vpp, int flag, cred_t *crp, caller_context_t *ct)
{
struct namenode *nodep = VTONM(*vpp);
int error = 0;
struct namenode *newnamep;
struct vnode *newvp;
struct vnode *infilevp;
struct vnode *outfilevp;
/*
* If the vnode is switched under us, the corresponding
* VN_RELE for this VN_HOLD will be done by the file system
* performing the switch. Otherwise, the corresponding
* VN_RELE will be done by nm_close().
*/
infilevp = outfilevp = nodep->nm_filevp;
VN_HOLD(outfilevp);
if ((error = VOP_OPEN(&outfilevp, flag, crp, ct)) != 0) {
VN_RELE(outfilevp);
return (error);
}
if (infilevp != outfilevp) {
/*
* See if the new filevp (outfilevp) is already associated
* with the mount point. If it is, then it already has a
* namenode associated with it.
*/
mutex_enter(&ntable_lock);
if ((newnamep =
namefind(outfilevp, nodep->nm_mountpt)) != NULL) {
struct vnode *vp = NMTOV(newnamep);
VN_HOLD(vp);
goto gotit;
}
newnamep = kmem_zalloc(sizeof (struct namenode), KM_SLEEP);
newvp = vn_alloc(KM_SLEEP);
newnamep->nm_vnode = newvp;
mutex_init(&newnamep->nm_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_enter(&nodep->nm_lock);
newvp->v_flag = ((*vpp)->v_flag | VNOMAP | VNOSWAP) & ~VROOT;
vn_setops(newvp, vn_getops(*vpp));
newvp->v_vfsp = &namevfs;
newvp->v_stream = outfilevp->v_stream;
newvp->v_type = outfilevp->v_type;
newvp->v_rdev = outfilevp->v_rdev;
newvp->v_data = (caddr_t)newnamep;
vn_exists(newvp);
bcopy(&nodep->nm_vattr, &newnamep->nm_vattr, sizeof (vattr_t));
newnamep->nm_vattr.va_type = outfilevp->v_type;
newnamep->nm_vattr.va_nodeid = namenodeno_alloc();
newnamep->nm_vattr.va_size = (u_offset_t)0;
newnamep->nm_vattr.va_rdev = outfilevp->v_rdev;
newnamep->nm_flag = NMNMNT;
newnamep->nm_filevp = outfilevp;
newnamep->nm_filep = nodep->nm_filep;
newnamep->nm_mountpt = nodep->nm_mountpt;
mutex_exit(&nodep->nm_lock);
/*
* Insert the new namenode into the hash list.
*/
nameinsert(newnamep);
gotit:
mutex_exit(&ntable_lock);
/*
* Release the above reference to the infilevp, the reference
* to the NAMEFS vnode, create a reference to the new vnode
* and return the new vnode to the user.
*/
VN_RELE(*vpp);
*vpp = NMTOV(newnamep);
}
return (0);
}