int
layerfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
{
struct vnode *vp;
int error;
error = VFS_VGET(MOUNTTOLAYERMOUNT(mp)->layerm_vfs, ino, &vp);
if (error) {
*vpp = NULL;
return error;
}
VOP_UNLOCK(vp);
error = layer_node_create(mp, vp, vpp);
if (error) {
vrele(vp);
*vpp = NULL;
return error;
}
error = vn_lock(*vpp, LK_EXCLUSIVE);
if (error) {
vrele(*vpp);
*vpp = NULL;
return error;
}
return 0;
}
int
layerfs_fhtovp(struct mount *mp, struct fid *fidp, struct vnode **vpp)
{
struct vnode *vp;
int error;
error = VFS_FHTOVP(MOUNTTOLAYERMOUNT(mp)->layerm_vfs, fidp, &vp);
if (error) {
*vpp = NULL;
return error;
}
VOP_UNLOCK(vp);
error = layer_node_create(mp, vp, vpp);
if (error) {
vput(vp);
*vpp = NULL;
return (error);
}
error = vn_lock(*vpp, LK_EXCLUSIVE);
if (error) {
vrele(*vpp);
*vpp = NULL;
return error;
}
return 0;
}
int
layerfs_statvfs(struct mount *mp, struct statvfs *sbp)
{
struct statvfs *sbuf;
int error;
sbuf = kmem_zalloc(sizeof(*sbuf), KM_SLEEP);
if (sbuf == NULL) {
return ENOMEM;
}
error = VFS_STATVFS(MOUNTTOLAYERMOUNT(mp)->layerm_vfs, sbuf);
if (error) {
goto done;
}
/* Copy across the relevant data and fake the rest. */
sbp->f_flag = sbuf->f_flag;
sbp->f_bsize = sbuf->f_bsize;
sbp->f_frsize = sbuf->f_frsize;
sbp->f_iosize = sbuf->f_iosize;
sbp->f_blocks = sbuf->f_blocks;
sbp->f_bfree = sbuf->f_bfree;
sbp->f_bavail = sbuf->f_bavail;
sbp->f_bresvd = sbuf->f_bresvd;
sbp->f_files = sbuf->f_files;
sbp->f_ffree = sbuf->f_ffree;
sbp->f_favail = sbuf->f_favail;
sbp->f_fresvd = sbuf->f_fresvd;
sbp->f_namemax = sbuf->f_namemax;
copy_statvfs_info(sbp, mp);
done:
kmem_free(sbuf, sizeof(*sbuf));
return error;
}
int
layerfs_loadvnode(struct mount *mp, struct vnode *vp,
const void *key, size_t key_len, const void **new_key)
{
struct layer_mount *lmp = MOUNTTOLAYERMOUNT(mp);
struct vnode *lowervp;
struct layer_node *xp;
KASSERT(key_len == sizeof(struct vnode *));
memcpy(&lowervp, key, key_len);
xp = kmem_alloc(lmp->layerm_size, KM_SLEEP);
if (xp == NULL)
return ENOMEM;
/* Share the interlock with the lower node. */
mutex_obj_hold(lowervp->v_interlock);
uvm_obj_setlock(&vp->v_uobj, lowervp->v_interlock);
vp->v_tag = lmp->layerm_tag;
vp->v_type = lowervp->v_type;
vp->v_op = lmp->layerm_vnodeop_p;
if (vp->v_type == VBLK || vp->v_type == VCHR)
spec_node_init(vp, lowervp->v_rdev);
vp->v_data = xp;
xp->layer_vnode = vp;
xp->layer_lowervp = lowervp;
xp->layer_flags = 0;
uvm_vnp_setsize(vp, 0);
/* Add a reference to the lower node. */
vref(lowervp);
*new_key = &xp->layer_lowervp;
return 0;
}
/*
* VFS start. Nothing needed here - the start routine on the underlying
* filesystem will have been called when that filesystem was mounted.
*/
int
layerfs_start(struct mount *mp, int flags)
{
#ifdef notyet
return VFS_START(MOUNTTOLAYERMOUNT(mp)->layerm_vfs, flags);
#else
return 0;
#endif
}
/*
* layer_node_create: try to find an existing layerfs vnode refering to it,
* otherwise make a new vnode which contains a reference to the lower vnode.
*
* => Caller should lock the lower node.
*/
int
layer_node_create(struct mount *mp, struct vnode *lowervp, struct vnode **nvpp)
{
struct vnode *aliasvp;
struct layer_mount *lmp = MOUNTTOLAYERMOUNT(mp);
mutex_enter(&lmp->layerm_hashlock);
aliasvp = layer_node_find(mp, lowervp);
if (aliasvp != NULL) {
/*
* Note: layer_node_find() has taken another reference to
* the alias vnode and moved the lock holding to aliasvp.
*/
#ifdef LAYERFS_DIAGNOSTIC
if (layerfs_debug)
vprint("layer_node_create: exists", aliasvp);
#endif
} else {
int error;
mutex_exit(&lmp->layerm_hashlock);
/*
* Get a new vnode. Make it to reference the layer_node.
* Note: aliasvp will be return with the reference held.
*/
error = (lmp->layerm_alloc)(mp, lowervp, &aliasvp);
if (error)
return error;
#ifdef LAYERFS_DIAGNOSTIC
if (layerfs_debug)
printf("layer_node_create: create new alias vnode\n");
#endif
}
/*
* Now that we acquired a reference on the upper vnode, release one
* on the lower node. The existence of the layer_node retains one
* reference to the lower node.
*/
vrele(lowervp);
KASSERT(lowervp->v_usecount > 0);
#ifdef LAYERFS_DIAGNOSTIC
if (layerfs_debug)
vprint("layer_node_create: alias", aliasvp);
#endif
*nvpp = aliasvp;
return 0;
}
int
layerfs_root(struct mount *mp, struct vnode **vpp)
{
struct vnode *vp;
vp = MOUNTTOLAYERMOUNT(mp)->layerm_rootvp;
if (vp == NULL) {
*vpp = NULL;
return EINVAL;
}
/*
* Return root vnode with locked and with a reference held.
*/
vref(vp);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
*vpp = vp;
return 0;
}
/*
* layer_node_find: find and return alias for lower vnode or NULL.
*
* => Return alias vnode referenced. if already exists.
* => The layermp's hashlock must be held on entry, we will unlock on success.
*/
struct vnode *
layer_node_find(struct mount *mp, struct vnode *lowervp)
{
struct layer_mount *lmp = MOUNTTOLAYERMOUNT(mp);
struct layer_node_hashhead *hd;
struct layer_node *a;
struct vnode *vp;
int error;
/*
* Find hash bucket and search the (two-way) linked list looking
* for a layerfs node structure which is referencing the lower vnode.
* If found, the increment the layer_node reference count, but NOT
* the lower vnode's reference counter.
*/
KASSERT(mutex_owned(&lmp->layerm_hashlock));
hd = LAYER_NHASH(lmp, lowervp);
loop:
LIST_FOREACH(a, hd, layer_hash) {
if (a->layer_lowervp != lowervp) {
continue;
}
vp = LAYERTOV(a);
if (vp->v_mount != mp) {
continue;
}
mutex_enter(vp->v_interlock);
mutex_exit(&lmp->layerm_hashlock);
error = vget(vp, 0);
if (error) {
mutex_enter(&lmp->layerm_hashlock);
goto loop;
}
return vp;
}
return NULL;
}
int
layerfs_quotactl(struct mount *mp, struct quotactl_args *args)
{
return VFS_QUOTACTL(MOUNTTOLAYERMOUNT(mp)->layerm_vfs, args);
}
/*
* layer_node_alloc: make a new layerfs vnode.
*
* => vp is the alias vnode, lowervp is the lower vnode.
* => We will hold a reference to lowervp.
*/
int
layer_node_alloc(struct mount *mp, struct vnode *lowervp, struct vnode **vpp)
{
struct layer_mount *lmp = MOUNTTOLAYERMOUNT(mp);
struct layer_node_hashhead *hd;
struct layer_node *xp;
struct vnode *vp, *nvp;
int error;
/* Get a new vnode and share its interlock with underlying vnode. */
error = getnewvnode(lmp->layerm_tag, mp, lmp->layerm_vnodeop_p,
lowervp->v_interlock, &vp);
if (error) {
return error;
}
vp->v_type = lowervp->v_type;
mutex_enter(vp->v_interlock);
vp->v_iflag |= VI_LAYER;
mutex_exit(vp->v_interlock);
xp = kmem_alloc(lmp->layerm_size, KM_SLEEP);
if (xp == NULL) {
ungetnewvnode(vp);
return ENOMEM;
}
if (vp->v_type == VBLK || vp->v_type == VCHR) {
spec_node_init(vp, lowervp->v_rdev);
}
/*
* Before inserting the node into the hash, check if other thread
* did not race with us. If so - return that node, destroy ours.
*/
mutex_enter(&lmp->layerm_hashlock);
if ((nvp = layer_node_find(mp, lowervp)) != NULL) {
ungetnewvnode(vp);
kmem_free(xp, lmp->layerm_size);
*vpp = nvp;
return 0;
}
vp->v_data = xp;
vp->v_vflag = (vp->v_vflag & ~VV_MPSAFE) |
(lowervp->v_vflag & VV_MPSAFE);
xp->layer_vnode = vp;
xp->layer_lowervp = lowervp;
xp->layer_flags = 0;
/*
* Insert the new node into the hash.
* Add a reference to the lower node.
*/
vref(lowervp);
hd = LAYER_NHASH(lmp, lowervp);
LIST_INSERT_HEAD(hd, xp, layer_hash);
uvm_vnp_setsize(vp, 0);
mutex_exit(&lmp->layerm_hashlock);
*vpp = vp;
return 0;
}
/*
* Mount overlay layer
*/
int
ov_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
{
struct lwp *l = curlwp;
int error = 0;
struct overlay_args *args = data;
struct vnode *lowerrootvp, *vp;
struct overlay_mount *nmp;
struct layer_mount *lmp;
#ifdef OVERLAYFS_DIAGNOSTIC
printf("ov_mount(mp = %p)\n", mp);
#endif
if (args == NULL)
return EINVAL;
if (*data_len < sizeof *args)
return EINVAL;
if (mp->mnt_flag & MNT_GETARGS) {
lmp = MOUNTTOLAYERMOUNT(mp);
if (lmp == NULL)
return EIO;
args->la.target = NULL;
*data_len = sizeof *args;
return 0;
}
/*
* Update is not supported
*/
if (mp->mnt_flag & MNT_UPDATE)
return EOPNOTSUPP;
/*
* Find lower node
*/
lowerrootvp = mp->mnt_vnodecovered;
vref(lowerrootvp);
if ((error = vn_lock(lowerrootvp, LK_EXCLUSIVE | LK_RETRY))) {
vrele(lowerrootvp);
return (error);
}
/*
* First cut at fixing up upper mount point
*/
nmp = kmem_zalloc(sizeof(struct overlay_mount), KM_SLEEP);
mp->mnt_data = nmp;
nmp->ovm_vfs = lowerrootvp->v_mount;
if (nmp->ovm_vfs->mnt_flag & MNT_LOCAL)
mp->mnt_flag |= MNT_LOCAL;
/*
* Make sure that the mount point is sufficiently initialized
* that the node create call will work.
*/
vfs_getnewfsid(mp);
nmp->ovm_size = sizeof (struct overlay_node);
nmp->ovm_tag = VT_OVERLAY;
nmp->ovm_bypass = layer_bypass;
nmp->ovm_vnodeop_p = overlay_vnodeop_p;
/*
* Fix up overlay node for root vnode
*/
VOP_UNLOCK(lowerrootvp);
error = layer_node_create(mp, lowerrootvp, &vp);
/*
* Make sure the fixup worked
*/
if (error) {
vrele(lowerrootvp);
kmem_free(nmp, sizeof(struct overlay_mount));
return error;
}
/*
* Keep a held reference to the root vnode.
* It is vrele'd in ov_unmount.
*/
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
vp->v_vflag |= VV_ROOT;
nmp->ovm_rootvp = vp;
VOP_UNLOCK(vp);
error = set_statvfs_info(path, UIO_USERSPACE, args->la.target,
UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l);
#ifdef OVERLAYFS_DIAGNOSTIC
printf("ov_mount: lower %s, alias at %s\n",
mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
#endif
return error;
}