static int
null_vptocnp(struct vop_vptocnp_args *ap)
{
struct vnode *vp = ap->a_vp;
struct vnode **dvp = ap->a_vpp;
struct vnode *lvp, *ldvp;
struct ucred *cred = ap->a_cred;
int error, locked;
if (vp->v_type == VDIR)
return (vop_stdvptocnp(ap));
locked = VOP_ISLOCKED(vp);
lvp = NULLVPTOLOWERVP(vp);
vhold(lvp);
VOP_UNLOCK(vp, 0); /* vp is held by vn_vptocnp_locked that called us */
ldvp = lvp;
error = vn_vptocnp(&ldvp, cred, ap->a_buf, ap->a_buflen);
vdrop(lvp);
if (error != 0) {
vn_lock(vp, locked | LK_RETRY);
return (ENOENT);
}
/*
* Exclusive lock is required by insmntque1 call in
* null_nodeget()
*/
error = vn_lock(ldvp, LK_EXCLUSIVE);
if (error != 0) {
vn_lock(vp, locked | LK_RETRY);
vdrop(ldvp);
return (ENOENT);
}
vref(ldvp);
vdrop(ldvp);
error = null_nodeget(vp->v_mount, ldvp, dvp);
if (error == 0) {
#ifdef DIAGNOSTIC
NULLVPTOLOWERVP(*dvp);
#endif
vhold(*dvp);
vput(*dvp);
} else
vput(ldvp);
vn_lock(vp, locked | LK_RETRY);
return (error);
}
static int
nullfs_readlink(struct vnop_readlink_args * ap)
{
NULLFSDEBUG("%s %p\n", __FUNCTION__, ap->a_vp);
int error;
struct vnode *vp, *lvp;
if (nullfs_checkspecialvp(ap->a_vp)) {
return ENOTSUP; /* the special vnodes aren't links */
}
vp = ap->a_vp;
lvp = NULLVPTOLOWERVP(vp);
error = vnode_getwithref(lvp);
if (error == 0) {
error = VNOP_READLINK(lvp, ap->a_uio, ap->a_context);
vnode_put(lvp);
if (error) {
NULLFSDEBUG("readlink failed: %d\n", error);
}
}
return error;
}
static int
nullfs_readdir(struct vnop_readdir_args * ap)
{
struct vnode *vp, *lvp;
int error;
struct null_mount * null_mp = MOUNTTONULLMOUNT(vnode_mount(ap->a_vp));
NULLFSDEBUG("%s %p\n", __FUNCTION__, ap->a_vp);
/* assumption is that any vp that comes through here had to go through lookup
*/
lck_mtx_lock(&null_mp->nullm_lock);
if (nullfs_isspecialvp(ap->a_vp)) {
error = nullfs_special_readdir(ap);
lck_mtx_unlock(&null_mp->nullm_lock);
return error;
}
lck_mtx_unlock(&null_mp->nullm_lock);
vp = ap->a_vp;
lvp = NULLVPTOLOWERVP(vp);
error = vnode_getwithref(lvp);
if (error == 0) {
error = VNOP_READDIR(lvp, ap->a_uio, ap->a_flags, ap->a_eofflag, ap->a_numdirent, ap->a_context);
vnode_put(lvp);
}
return error;
}
static int
nullfs_getattr(struct vnop_getattr_args * args)
{
int error;
struct null_mount * null_mp = MOUNTTONULLMOUNT(vnode_mount(args->a_vp));
NULLFSDEBUG("%s %p\n", __FUNCTION__, args->a_vp);
lck_mtx_lock(&null_mp->nullm_lock);
if (nullfs_isspecialvp(args->a_vp)) {
error = nullfs_special_getattr(args);
lck_mtx_unlock(&null_mp->nullm_lock);
return error;
}
lck_mtx_unlock(&null_mp->nullm_lock);
/* this will return a different inode for third than read dir will */
struct vnode * lowervp = NULLVPTOLOWERVP(args->a_vp);
error = vnode_getwithref(lowervp);
if (error == 0) {
error = VNOP_GETATTR(lowervp, args->a_vap, args->a_context);
vnode_put(lowervp);
if (error == 0) {
/* fix up fsid so it doesn't say the underlying fs*/
VATTR_RETURN(args->a_vap, va_fsid, vfs_statfs(vnode_mount(args->a_vp))->f_fsid.val[0]);
}
}
return error;
}
/*
* We need to process our own vnode unlock and then clear the
* interlock flag as it applies only to our vnode, not the
* vnodes below us on the stack.
*/
static int
null_unlock(struct vop_unlock_args *ap)
{
struct vnode *vp = ap->a_vp;
int flags = ap->a_flags;
int mtxlkflag = 0;
struct null_node *nn;
struct vnode *lvp;
int error;
if ((flags & LK_INTERLOCK) != 0)
mtxlkflag = 1;
else if (mtx_owned(VI_MTX(vp)) == 0) {
VI_LOCK(vp);
mtxlkflag = 2;
}
nn = VTONULL(vp);
if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
VI_LOCK_FLAGS(lvp, MTX_DUPOK);
flags |= LK_INTERLOCK;
vholdl(lvp);
VI_UNLOCK(vp);
error = VOP_UNLOCK(lvp, flags);
vdrop(lvp);
if (mtxlkflag == 0)
VI_LOCK(vp);
} else {
if (mtxlkflag == 2)
VI_UNLOCK(vp);
error = vop_stdunlock(ap);
}
return (error);
}
static int
null_vptofh(struct vop_vptofh_args *ap)
{
struct vnode *lvp;
lvp = NULLVPTOLOWERVP(ap->a_vp);
return VOP_VPTOFH(lvp, ap->a_fhp);
}
static int
null_print(struct vop_print_args *ap)
{
struct vnode *vp = ap->a_vp;
printf("\tvp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp));
return (0);
}
static int
null_open(struct vop_open_args *ap)
{
int retval;
struct vnode *vp, *ldvp;
vp = ap->a_vp;
ldvp = NULLVPTOLOWERVP(vp);
retval = null_bypass(&ap->a_gen);
if (retval == 0)
vp->v_object = ldvp->v_object;
return (retval);
}
/*
* Increasing refcount of lower vnode is needed at least for the case
* when lower FS is NFS to do sillyrename if the file is in use.
* Unfortunately v_usecount is incremented in many places in
* the kernel and, as such, there may be races that result in
* the NFS client doing an extraneous silly rename, but that seems
* preferable to not doing a silly rename when it is needed.
*/
static int
null_remove(struct vop_remove_args *ap)
{
int retval, vreleit;
struct vnode *lvp;
if (vrefcnt(ap->a_vp) > 1) {
lvp = NULLVPTOLOWERVP(ap->a_vp);
VREF(lvp);
vreleit = 1;
} else
vreleit = 0;
retval = null_bypass(&ap->a_gen);
if (vreleit != 0)
vrele(lvp);
return (retval);
}
static int
null_add_writecount(struct vop_add_writecount_args *ap)
{
struct vnode *lvp, *vp;
int error;
vp = ap->a_vp;
lvp = NULLVPTOLOWERVP(vp);
KASSERT(vp->v_writecount + ap->a_inc >= 0, ("wrong writecount inc"));
if (vp->v_writecount > 0 && vp->v_writecount + ap->a_inc == 0)
error = VOP_ADD_WRITECOUNT(lvp, -1);
else if (vp->v_writecount == 0 && vp->v_writecount + ap->a_inc > 0)
error = VOP_ADD_WRITECOUNT(lvp, 1);
else
error = 0;
if (error == 0)
vp->v_writecount += ap->a_inc;
return (error);
}
static int
nullfs_read(struct vnop_read_args * ap)
{
int error = EIO;
struct vnode *vp, *lvp;
NULLFSDEBUG("%s %p\n", __FUNCTION__, ap->a_vp);
if (nullfs_checkspecialvp(ap->a_vp)) {
return ENOTSUP; /* the special vnodes can't be read */
}
vp = ap->a_vp;
lvp = NULLVPTOLOWERVP(vp);
/*
* First some house keeping
*/
if (vnode_getwithvid(lvp, NULLVPTOLOWERVID(vp)) == 0) {
if (!vnode_isreg(lvp) && !vnode_islnk(lvp)) {
error = EPERM;
goto end;
}
if (uio_resid(ap->a_uio) == 0) {
error = 0;
goto end;
}
/*
* Now ask VM/UBC/VFS to do our bidding
*/
error = VNOP_READ(lvp, ap->a_uio, ap->a_ioflag, ap->a_context);
if (error) {
NULLFSDEBUG("VNOP_READ failed: %d\n", error);
}
end:
vnode_put(lvp);
}
return error;
}
/*
* Increasing refcount of lower vnode is needed at least for the case
* when lower FS is NFS to do sillyrename if the file is in use.
* Unfortunately v_usecount is incremented in many places in
* the kernel and, as such, there may be races that result in
* the NFS client doing an extraneous silly rename, but that seems
* preferable to not doing a silly rename when it is needed.
*/
static int
crypto_remove(struct vop_remove_args *ap)
{
int retval, vreleit;
struct vnode *lvp, *vp;
vp = ap->a_vp;
if (vrefcnt(vp) > 1) {
lvp = NULLVPTOLOWERVP(vp);
VREF(lvp);
vreleit = 1;
} else
vreleit = 0;
VTONULL(vp)->crypto_flags |= NULLV_DROP;
retval = crypto_bypass(&ap->a_gen);
if (vreleit != 0)
vrele(lvp);
return (retval);
}
static int
nullfs_listxattr(struct vnop_listxattr_args * args)
{
int error;
struct vnode *vp, *lvp;
NULLFSDEBUG("%s %p\n", __FUNCTION__, args->a_vp);
if (nullfs_checkspecialvp(args->a_vp)) {
return 0; /* nothing extra needed */
}
vp = args->a_vp;
lvp = NULLVPTOLOWERVP(vp);
error = vnode_getwithref(lvp);
if (error == 0) {
error = VNOP_LISTXATTR(lvp, args->a_uio, args->a_size, args->a_options, args->a_context);
vnode_put(lvp);
}
return error;
}
/*
* We have to carry on the locking protocol on the null layer vnodes
* as we progress through the tree. We also have to enforce read-only
* if this layer is mounted read-only.
*/
static int
null_lookup(struct vop_lookup_args *ap)
{
struct componentname *cnp = ap->a_cnp;
struct vnode *dvp = ap->a_dvp;
int flags = cnp->cn_flags;
struct vnode *vp, *ldvp, *lvp;
int error;
if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
return (EROFS);
/*
* Although it is possible to call null_bypass(), we'll do
* a direct call to reduce overhead
*/
ldvp = NULLVPTOLOWERVP(dvp);
vp = lvp = NULL;
error = VOP_LOOKUP(ldvp, &lvp, cnp);
if (error == EJUSTRETURN && (flags & ISLASTCN) &&
(dvp->v_mount->mnt_flag & MNT_RDONLY) &&
(cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
error = EROFS;
if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
if (ldvp == lvp) {
*ap->a_vpp = dvp;
VREF(dvp);
vrele(lvp);
} else {
error = null_nodeget(dvp->v_mount, lvp, &vp);
if (error)
vput(lvp);
else
*ap->a_vpp = vp;
}
}
return (error);
}
static int
nullfs_mnomap(struct vnop_mnomap_args * args)
{
int error;
struct vnode *vp, *lvp;
NULLFSDEBUG("%s %p\n", __FUNCTION__, args->a_vp);
if (nullfs_checkspecialvp(args->a_vp)) {
return 0; /* nothing extra needed */
}
vp = args->a_vp;
lvp = NULLVPTOLOWERVP(vp);
error = vnode_getwithref(lvp);
if (error == 0) {
error = VNOP_MNOMAP(lvp, args->a_context);
vnode_put(lvp);
}
return error;
}
/*
* Mount null layer
*/
static int
nullfs_mount(struct mount *mp)
{
int error = 0;
struct vnode *lowerrootvp, *vp;
struct vnode *nullm_rootvp;
struct null_mount *xmp;
struct thread *td = curthread;
char *target;
int isvnunlocked = 0, len;
struct nameidata nd, *ndp = &nd;
NULLFSDEBUG("nullfs_mount(mp = %p)\n", (void *)mp);
if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_NULLFS))
return (EPERM);
if (mp->mnt_flag & MNT_ROOTFS)
return (EOPNOTSUPP);
/*
* Update is a no-op
*/
if (mp->mnt_flag & MNT_UPDATE) {
/*
* Only support update mounts for NFS export.
*/
if (vfs_flagopt(mp->mnt_optnew, "export", NULL, 0))
return (0);
else
return (EOPNOTSUPP);
}
/*
* Get argument
*/
error = vfs_getopt(mp->mnt_optnew, "target", (void **)&target, &len);
if (error || target[len - 1] != '\0')
return (EINVAL);
/*
* Unlock lower node to avoid possible deadlock.
*/
if ((mp->mnt_vnodecovered->v_op == &null_vnodeops) &&
VOP_ISLOCKED(mp->mnt_vnodecovered) == LK_EXCLUSIVE) {
VOP_UNLOCK(mp->mnt_vnodecovered, 0);
isvnunlocked = 1;
}
/*
* Find lower node
*/
NDINIT(ndp, LOOKUP, FOLLOW|LOCKLEAF, UIO_SYSSPACE, target, curthread);
error = namei(ndp);
/*
* Re-lock vnode.
* XXXKIB This is deadlock-prone as well.
*/
if (isvnunlocked)
vn_lock(mp->mnt_vnodecovered, LK_EXCLUSIVE | LK_RETRY);
if (error)
return (error);
NDFREE(ndp, NDF_ONLY_PNBUF);
/*
* Sanity check on lower vnode
*/
lowerrootvp = ndp->ni_vp;
/*
* Check multi null mount to avoid `lock against myself' panic.
*/
if (lowerrootvp == VTONULL(mp->mnt_vnodecovered)->null_lowervp) {
NULLFSDEBUG("nullfs_mount: multi null mount?\n");
vput(lowerrootvp);
return (EDEADLK);
}
xmp = (struct null_mount *) malloc(sizeof(struct null_mount),
M_NULLFSMNT, M_WAITOK | M_ZERO);
/*
* Save reference to underlying FS
*/
xmp->nullm_vfs = lowerrootvp->v_mount;
/*
* Save reference. Each mount also holds
* a reference on the root vnode.
*/
error = null_nodeget(mp, lowerrootvp, &vp);
/*
* Make sure the node alias worked
*/
if (error) {
free(xmp, M_NULLFSMNT);
return (error);
}
/*
* Keep a held reference to the root vnode.
* It is vrele'd in nullfs_unmount.
*/
nullm_rootvp = vp;
nullm_rootvp->v_vflag |= VV_ROOT;
xmp->nullm_rootvp = nullm_rootvp;
/*
* Unlock the node (either the lower or the alias)
*/
VOP_UNLOCK(vp, 0);
if (NULLVPTOLOWERVP(nullm_rootvp)->v_mount->mnt_flag & MNT_LOCAL) {
MNT_ILOCK(mp);
mp->mnt_flag |= MNT_LOCAL;
MNT_IUNLOCK(mp);
}
xmp->nullm_flags |= NULLM_CACHE;
if (vfs_getopt(mp->mnt_optnew, "nocache", NULL, NULL) == 0)
xmp->nullm_flags &= ~NULLM_CACHE;
MNT_ILOCK(mp);
if ((xmp->nullm_flags & NULLM_CACHE) != 0) {
mp->mnt_kern_flag |= lowerrootvp->v_mount->mnt_kern_flag &
(MNTK_SHARED_WRITES | MNTK_LOOKUP_SHARED |
MNTK_EXTENDED_SHARED);
}
mp->mnt_kern_flag |= MNTK_LOOKUP_EXCL_DOTDOT;
MNT_IUNLOCK(mp);
mp->mnt_data = xmp;
vfs_getnewfsid(mp);
if ((xmp->nullm_flags & NULLM_CACHE) != 0) {
MNT_ILOCK(xmp->nullm_vfs);
TAILQ_INSERT_TAIL(&xmp->nullm_vfs->mnt_uppers, mp,
mnt_upper_link);
MNT_IUNLOCK(xmp->nullm_vfs);
}
vfs_mountedfrom(mp, target);
NULLFSDEBUG("nullfs_mount: lower %s, alias at %s\n",
mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
return (0);
}
/*
* We need to process our own vnode lock and then clear the
* interlock flag as it applies only to our vnode, not the
* vnodes below us on the stack.
*/
static int
null_lock(struct vop_lock1_args *ap)
{
struct vnode *vp = ap->a_vp;
int flags = ap->a_flags;
struct null_node *nn;
struct vnode *lvp;
int error;
if ((flags & LK_INTERLOCK) == 0) {
VI_LOCK(vp);
ap->a_flags = flags |= LK_INTERLOCK;
}
nn = VTONULL(vp);
/*
* If we're still active we must ask the lower layer to
* lock as ffs has special lock considerations in it's
* vop lock.
*/
if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
VI_LOCK_FLAGS(lvp, MTX_DUPOK);
VI_UNLOCK(vp);
/*
* We have to hold the vnode here to solve a potential
* reclaim race. If we're forcibly vgone'd while we
* still have refs, a thread could be sleeping inside
* the lowervp's vop_lock routine. When we vgone we will
* drop our last ref to the lowervp, which would allow it
* to be reclaimed. The lowervp could then be recycled,
* in which case it is not legal to be sleeping in it's VOP.
* We prevent it from being recycled by holding the vnode
* here.
*/
vholdl(lvp);
error = VOP_LOCK(lvp, flags);
/*
* We might have slept to get the lock and someone might have
* clean our vnode already, switching vnode lock from one in
* lowervp to v_lock in our own vnode structure. Handle this
* case by reacquiring correct lock in requested mode.
*/
if (VTONULL(vp) == NULL && error == 0) {
ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
switch (flags & LK_TYPE_MASK) {
case LK_SHARED:
ap->a_flags |= LK_SHARED;
break;
case LK_UPGRADE:
case LK_EXCLUSIVE:
ap->a_flags |= LK_EXCLUSIVE;
break;
default:
panic("Unsupported lock request %d\n",
ap->a_flags);
}
VOP_UNLOCK(lvp, 0);
error = vop_stdlock(ap);
}
vdrop(lvp);
} else
error = vop_stdlock(ap);
return (error);
}
/*
* Mount null layer
*/
static int
nullfs_mount(struct mount *mp)
{
int error = 0;
struct vnode *lowerrootvp, *vp;
struct vnode *nullm_rootvp;
struct null_mount *xmp;
char *target;
int isvnunlocked = 0, len;
struct nameidata nd, *ndp = &nd;
NULLFSDEBUG("nullfs_mount(mp = %p)\n", (void *)mp);
if (mp->mnt_flag & MNT_ROOTFS)
return (EOPNOTSUPP);
/*
* Update is a no-op
*/
if (mp->mnt_flag & MNT_UPDATE) {
/*
* Only support update mounts for NFS export.
*/
if (vfs_flagopt(mp->mnt_optnew, "export", NULL, 0))
return (0);
else
return (EOPNOTSUPP);
}
/*
* Get argument
*/
error = vfs_getopt(mp->mnt_optnew, "target", (void **)&target, &len);
if (error || target[len - 1] != '\0')
return (EINVAL);
/*
* Unlock lower node to avoid deadlock.
* (XXX) VOP_ISLOCKED is needed?
*/
if ((mp->mnt_vnodecovered->v_op == &null_vnodeops) &&
VOP_ISLOCKED(mp->mnt_vnodecovered)) {
VOP_UNLOCK(mp->mnt_vnodecovered, 0);
isvnunlocked = 1;
}
/*
* Find lower node
*/
NDINIT(ndp, LOOKUP, FOLLOW|LOCKLEAF, UIO_SYSSPACE, target, curthread);
error = namei(ndp);
/*
* Re-lock vnode.
*/
if (isvnunlocked && !VOP_ISLOCKED(mp->mnt_vnodecovered))
vn_lock(mp->mnt_vnodecovered, LK_EXCLUSIVE | LK_RETRY);
if (error)
return (error);
NDFREE(ndp, NDF_ONLY_PNBUF);
/*
* Sanity check on lower vnode
*/
lowerrootvp = ndp->ni_vp;
/*
* Check multi null mount to avoid `lock against myself' panic.
*/
if (lowerrootvp == VTONULL(mp->mnt_vnodecovered)->null_lowervp) {
NULLFSDEBUG("nullfs_mount: multi null mount?\n");
vput(lowerrootvp);
return (EDEADLK);
}
xmp = (struct null_mount *) malloc(sizeof(struct null_mount),
M_NULLFSMNT, M_WAITOK); /* XXX */
/*
* Save reference to underlying FS
*/
xmp->nullm_vfs = lowerrootvp->v_mount;
/*
* Save reference. Each mount also holds
* a reference on the root vnode.
*/
error = null_nodeget(mp, lowerrootvp, &vp);
/*
* Make sure the node alias worked
*/
if (error) {
VOP_UNLOCK(vp, 0);
vrele(lowerrootvp);
free(xmp, M_NULLFSMNT); /* XXX */
return (error);
}
/*
* Keep a held reference to the root vnode.
* It is vrele'd in nullfs_unmount.
*/
nullm_rootvp = vp;
nullm_rootvp->v_vflag |= VV_ROOT;
xmp->nullm_rootvp = nullm_rootvp;
/*
* Unlock the node (either the lower or the alias)
*/
VOP_UNLOCK(vp, 0);
if (NULLVPTOLOWERVP(nullm_rootvp)->v_mount->mnt_flag & MNT_LOCAL) {
MNT_ILOCK(mp);
mp->mnt_flag |= MNT_LOCAL;
MNT_IUNLOCK(mp);
}
MNT_ILOCK(mp);
mp->mnt_kern_flag |= lowerrootvp->v_mount->mnt_kern_flag & MNTK_MPSAFE;
MNT_IUNLOCK(mp);
mp->mnt_data = xmp;
vfs_getnewfsid(mp);
vfs_mountedfrom(mp, target);
NULLFSDEBUG("nullfs_mount: lower %s, alias at %s\n",
mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
return (0);
}
/*
* We have to carry on the locking protocol on the null layer vnodes
* as we progress through the tree. We also have to enforce read-only
* if this layer is mounted read-only.
*/
static int
null_lookup(struct vnop_lookup_args * ap)
{
struct componentname * cnp = ap->a_cnp;
struct vnode * dvp = ap->a_dvp;
struct vnode *vp, *ldvp, *lvp;
struct mount * mp;
struct null_mount * null_mp;
int error;
NULLFSDEBUG("%s parent: %p component: %.*s\n", __FUNCTION__, ap->a_dvp, cnp->cn_namelen, cnp->cn_nameptr);
mp = vnode_mount(dvp);
/* rename and delete are not allowed. this is a read only file system */
if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME || cnp->cn_nameiop == CREATE) {
return (EROFS);
}
null_mp = MOUNTTONULLMOUNT(mp);
lck_mtx_lock(&null_mp->nullm_lock);
if (nullfs_isspecialvp(dvp)) {
error = null_special_lookup(ap);
lck_mtx_unlock(&null_mp->nullm_lock);
return error;
}
lck_mtx_unlock(&null_mp->nullm_lock);
// . and .. handling
if (cnp->cn_nameptr[0] == '.') {
if (cnp->cn_namelen == 1) {
vp = dvp;
} else if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
/* mount point crossing is handled in null_special_lookup */
vp = vnode_parent(dvp);
} else {
goto notdot;
}
error = vp ? vnode_get(vp) : ENOENT;
if (error == 0) {
*ap->a_vpp = vp;
}
return error;
}
notdot:
ldvp = NULLVPTOLOWERVP(dvp);
vp = lvp = NULL;
/*
* Hold ldvp. The reference on it, owned by dvp, is lost in
* case of dvp reclamation.
*/
error = vnode_getwithref(ldvp);
if (error) {
return error;
}
error = VNOP_LOOKUP(ldvp, &lvp, cnp, ap->a_context);
vnode_put(ldvp);
if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
if (ldvp == lvp) {
vp = dvp;
error = vnode_get(vp);
} else {
error = null_nodeget(mp, lvp, dvp, &vp, cnp, 0);
}
if (error == 0) {
*ap->a_vpp = vp;
}
}
/* if we got lvp, drop the iocount from VNOP_LOOKUP */
if (lvp != NULL) {
vnode_put(lvp);
}
return (error);
}
/*
* We have to carry on the locking protocol on the null layer vnodes
* as we progress through the tree. We also have to enforce read-only
* if this layer is mounted read-only.
*/
static int
null_lookup(struct vop_lookup_args *ap)
{
struct componentname *cnp = ap->a_cnp;
struct vnode *dvp = ap->a_dvp;
int flags = cnp->cn_flags;
struct vnode *vp, *ldvp, *lvp;
struct mount *mp;
int error;
mp = dvp->v_mount;
if ((flags & ISLASTCN) != 0 && (mp->mnt_flag & MNT_RDONLY) != 0 &&
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
return (EROFS);
/*
* Although it is possible to call null_bypass(), we'll do
* a direct call to reduce overhead
*/
ldvp = NULLVPTOLOWERVP(dvp);
vp = lvp = NULL;
KASSERT((ldvp->v_vflag & VV_ROOT) == 0 ||
((dvp->v_vflag & VV_ROOT) != 0 && (flags & ISDOTDOT) == 0),
("ldvp %p fl %#x dvp %p fl %#x flags %#x", ldvp, ldvp->v_vflag,
dvp, dvp->v_vflag, flags));
/*
* Hold ldvp. The reference on it, owned by dvp, is lost in
* case of dvp reclamation, and we need ldvp to move our lock
* from ldvp to dvp.
*/
vhold(ldvp);
error = VOP_LOOKUP(ldvp, &lvp, cnp);
/*
* VOP_LOOKUP() on lower vnode may unlock ldvp, which allows
* dvp to be reclaimed due to shared v_vnlock. Check for the
* doomed state and return error.
*/
if ((error == 0 || error == EJUSTRETURN) &&
(dvp->v_iflag & VI_DOOMED) != 0) {
error = ENOENT;
if (lvp != NULL)
vput(lvp);
/*
* If vgone() did reclaimed dvp before curthread
* relocked ldvp, the locks of dvp and ldpv are no
* longer shared. In this case, relock of ldvp in
* lower fs VOP_LOOKUP() does not restore the locking
* state of dvp. Compensate for this by unlocking
* ldvp and locking dvp, which is also correct if the
* locks are still shared.
*/
VOP_UNLOCK(ldvp, 0);
vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
}
vdrop(ldvp);
if (error == EJUSTRETURN && (flags & ISLASTCN) != 0 &&
(mp->mnt_flag & MNT_RDONLY) != 0 &&
(cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
error = EROFS;
if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
if (ldvp == lvp) {
*ap->a_vpp = dvp;
VREF(dvp);
vrele(lvp);
} else {
error = null_nodeget(mp, lvp, &vp);
if (error == 0)
*ap->a_vpp = vp;
}
}
return (error);
}
/* relies on v1 paging */
static int
nullfs_pagein(struct vnop_pagein_args * ap)
{
int error = EIO;
struct vnode *vp, *lvp;
NULLFSDEBUG("%s %p\n", __FUNCTION__, ap->a_vp);
vp = ap->a_vp;
lvp = NULLVPTOLOWERVP(vp);
if (vnode_vtype(vp) != VREG) {
return ENOTSUP;
}
/*
* Ask VM/UBC/VFS to do our bidding
*/
if (vnode_getwithvid(lvp, NULLVPTOLOWERVID(vp)) == 0) {
vm_offset_t ioaddr;
uio_t auio;
kern_return_t kret;
off_t bytes_to_commit;
off_t lowersize;
upl_t upl = ap->a_pl;
user_ssize_t bytes_remaining = 0;
auio = uio_create(1, ap->a_f_offset, UIO_SYSSPACE, UIO_READ);
if (auio == NULL) {
error = EIO;
goto exit_no_unmap;
}
kret = ubc_upl_map(upl, &ioaddr);
if (KERN_SUCCESS != kret) {
panic("nullfs_pagein: ubc_upl_map() failed with (%d)", kret);
}
ioaddr += ap->a_pl_offset;
error = uio_addiov(auio, (user_addr_t)ioaddr, ap->a_size);
if (error) {
goto exit;
}
lowersize = ubc_getsize(lvp);
if (lowersize != ubc_getsize(vp)) {
(void)ubc_setsize(vp, lowersize); /* ignore failures, nothing can be done */
}
error = VNOP_READ(lvp, auio, ((ap->a_flags & UPL_IOSYNC) ? IO_SYNC : 0), ap->a_context);
bytes_remaining = uio_resid(auio);
if (bytes_remaining > 0 && bytes_remaining <= (user_ssize_t)ap->a_size)
{
/* zero bytes that weren't read in to the upl */
bzero((void*)((uintptr_t)(ioaddr + ap->a_size - bytes_remaining)), (size_t) bytes_remaining);
}
exit:
kret = ubc_upl_unmap(upl);
if (KERN_SUCCESS != kret) {
panic("nullfs_pagein: ubc_upl_unmap() failed with (%d)", kret);
}
if (auio != NULL) {
uio_free(auio);
}
exit_no_unmap:
if ((ap->a_flags & UPL_NOCOMMIT) == 0) {
if (!error && (bytes_remaining >= 0) && (bytes_remaining <= (user_ssize_t)ap->a_size)) {
/* only commit what was read in (page aligned)*/
bytes_to_commit = ap->a_size - bytes_remaining;
if (bytes_to_commit)
{
/* need to make sure bytes_to_commit and byte_remaining are page aligned before calling ubc_upl_commit_range*/
if (bytes_to_commit & PAGE_MASK)
{
bytes_to_commit = (bytes_to_commit & (~PAGE_MASK)) + (PAGE_MASK + 1);
assert(bytes_to_commit <= (off_t)ap->a_size);
bytes_remaining = ap->a_size - bytes_to_commit;
}
ubc_upl_commit_range(upl, ap->a_pl_offset, (upl_size_t)bytes_to_commit, UPL_COMMIT_FREE_ON_EMPTY);
}
/* abort anything thats left */
if (bytes_remaining) {
ubc_upl_abort_range(upl, ap->a_pl_offset + bytes_to_commit, (upl_size_t)bytes_remaining, UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY);
}
} else {
ubc_upl_abort_range(upl, ap->a_pl_offset, (upl_size_t)ap->a_size, UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY);
}
}
vnode_put(lvp);
} else if((ap->a_flags & UPL_NOCOMMIT) == 0) {
ubc_upl_abort_range(ap->a_pl, ap->a_pl_offset, (upl_size_t)ap->a_size, UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY);
}
return error;
}
/*
* This is the 10-Apr-92 bypass routine.
* This version has been optimized for speed, throwing away some
* safety checks. It should still always work, but it's not as
* robust to programmer errors.
*
* In general, we map all vnodes going down and unmap them on the way back.
* As an exception to this, vnodes can be marked "unmapped" by setting
* the Nth bit in operation's vdesc_flags.
*
* Also, some BSD vnode operations have the side effect of vrele'ing
* their arguments. With stacking, the reference counts are held
* by the upper node, not the lower one, so we must handle these
* side-effects here. This is not of concern in Sun-derived systems
* since there are no such side-effects.
*
* This makes the following assumptions:
* - only one returned vpp
* - no INOUT vpp's (Sun's vop_open has one of these)
* - the vnode operation vector of the first vnode should be used
* to determine what implementation of the op should be invoked
* - all mapped vnodes are of our vnode-type (NEEDSWORK:
* problems on rmdir'ing mount points and renaming?)
*/
int
null_bypass(struct vop_generic_args *ap)
{
struct vnode **this_vp_p;
int error;
struct vnode *old_vps[VDESC_MAX_VPS];
struct vnode **vps_p[VDESC_MAX_VPS];
struct vnode ***vppp;
struct vnodeop_desc *descp = ap->a_desc;
int reles, i;
if (null_bug_bypass)
printf ("null_bypass: %s\n", descp->vdesc_name);
#ifdef DIAGNOSTIC
/*
* We require at least one vp.
*/
if (descp->vdesc_vp_offsets == NULL ||
descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
panic ("null_bypass: no vp's in map");
#endif
/*
* Map the vnodes going in.
* Later, we'll invoke the operation based on
* the first mapped vnode's operation vector.
*/
reles = descp->vdesc_flags;
for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
break; /* bail out at end of list */
vps_p[i] = this_vp_p =
VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap);
/*
* We're not guaranteed that any but the first vnode
* are of our type. Check for and don't map any
* that aren't. (We must always map first vp or vclean fails.)
*/
if (i && (*this_vp_p == NULLVP ||
(*this_vp_p)->v_op != &null_vnodeops)) {
old_vps[i] = NULLVP;
} else {
old_vps[i] = *this_vp_p;
*(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
/*
* XXX - Several operations have the side effect
* of vrele'ing their vp's. We must account for
* that. (This should go away in the future.)
*/
if (reles & VDESC_VP0_WILLRELE)
VREF(*this_vp_p);
}
}
/*
* Call the operation on the lower layer
* with the modified argument structure.
*/
if (vps_p[0] && *vps_p[0])
error = VCALL(ap);
else {
printf("null_bypass: no map for %s\n", descp->vdesc_name);
error = EINVAL;
}
/*
* Maintain the illusion of call-by-value
* by restoring vnodes in the argument structure
* to their original value.
*/
reles = descp->vdesc_flags;
for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
break; /* bail out at end of list */
if (old_vps[i]) {
*(vps_p[i]) = old_vps[i];
#if 0
if (reles & VDESC_VP0_WILLUNLOCK)
VOP_UNLOCK(*(vps_p[i]), 0);
#endif
if (reles & VDESC_VP0_WILLRELE)
vrele(*(vps_p[i]));
}
}
/*
* Map the possible out-going vpp
* (Assumes that the lower layer always returns
* a VREF'ed vpp unless it gets an error.)
*/
if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
!(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
!error) {
/*
* XXX - even though some ops have vpp returned vp's,
* several ops actually vrele this before returning.
* We must avoid these ops.
* (This should go away when these ops are regularized.)
*/
if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
goto out;
vppp = VOPARG_OFFSETTO(struct vnode***,
descp->vdesc_vpp_offset,ap);
if (*vppp)
error = null_nodeget(old_vps[0]->v_mount, **vppp, *vppp);
}
out:
return (error);
}
/*
* Do not allow the VOP_INACTIVE to be passed to the lower layer,
* since the reference count on the lower vnode is not related to
* ours.
*/
static int
null_inactive(struct vop_inactive_args *ap __unused)
{
struct vnode *vp, *lvp;
struct null_node *xp;
struct mount *mp;
struct null_mount *xmp;
vp = ap->a_vp;
xp = VTONULL(vp);
lvp = NULLVPTOLOWERVP(vp);
mp = vp->v_mount;
xmp = MOUNTTONULLMOUNT(mp);
if ((xmp->nullm_flags & NULLM_CACHE) == 0 ||
(xp->null_flags & NULLV_DROP) != 0 ||
(lvp->v_vflag & VV_NOSYNC) != 0) {
/*
* If this is the last reference and caching of the
* nullfs vnodes is not enabled, or the lower vnode is
* deleted, then free up the vnode so as not to tie up
* the lower vnodes.
*/
vp->v_object = NULL;
vrecycle(vp, curthread);
}
return (0);
