/*
* XXX: We cannot use this function as a cache constructor, because
* there is one global cache for all file systems and we need
* to pass vfsp here, which is not possible, because argument
* 'cdrarg' is defined at kmem_cache_create() time.
*/
static int
zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
{
znode_t *zp = buf;
vnode_t *vp;
vfs_t *vfsp = arg;
int error;
POINTER_INVALIDATE(&zp->z_zfsvfs);
ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
if (vfsp != NULL) {
error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp);
if (error != 0 && (kmflags & KM_NOSLEEP))
return (-1);
ASSERT(error == 0);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
zp->z_vnode = vp;
vp->v_data = (caddr_t)zp;
VN_LOCK_AREC(vp);
} else {
zp->z_vnode = NULL;
}
list_link_init(&zp->z_link_node);
mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
rw_init(&zp->z_map_lock, NULL, RW_DEFAULT, NULL);
rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
avl_create(&zp->z_range_avl, zfs_range_compare,
sizeof (rl_t), offsetof(rl_t, r_node));
zp->z_dbuf = NULL;
zp->z_dirlocks = NULL;
return (0);
}
/*
* Q_QUOTAON - set up a quota file for a particular filesystem.
*/
int
quotaon(struct thread *td, struct mount *mp, int type, void *fname)
{
struct ufsmount *ump;
struct vnode *vp, **vpp;
struct vnode *mvp;
struct dquot *dq;
int error, flags;
struct nameidata nd;
error = priv_check(td, PRIV_UFS_QUOTAON);
if (error != 0) {
vfs_unbusy(mp);
return (error);
}
if ((mp->mnt_flag & MNT_RDONLY) != 0) {
vfs_unbusy(mp);
return (EROFS);
}
ump = VFSTOUFS(mp);
dq = NODQUOT;
NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fname, td);
flags = FREAD | FWRITE;
vfs_ref(mp);
vfs_unbusy(mp);
error = vn_open(&nd, &flags, 0, NULL);
if (error != 0) {
vfs_rel(mp);
return (error);
}
NDFREE(&nd, NDF_ONLY_PNBUF);
vp = nd.ni_vp;
error = vfs_busy(mp, MBF_NOWAIT);
vfs_rel(mp);
if (error == 0) {
if (vp->v_type != VREG) {
error = EACCES;
vfs_unbusy(mp);
}
}
if (error != 0) {
VOP_UNLOCK(vp, 0);
(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
return (error);
}
UFS_LOCK(ump);
if ((ump->um_qflags[type] & (QTF_OPENING|QTF_CLOSING)) != 0) {
UFS_UNLOCK(ump);
VOP_UNLOCK(vp, 0);
(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
vfs_unbusy(mp);
return (EALREADY);
}
ump->um_qflags[type] |= QTF_OPENING|QTF_CLOSING;
UFS_UNLOCK(ump);
if ((error = dqopen(vp, ump, type)) != 0) {
VOP_UNLOCK(vp, 0);
UFS_LOCK(ump);
ump->um_qflags[type] &= ~(QTF_OPENING|QTF_CLOSING);
UFS_UNLOCK(ump);
(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
vfs_unbusy(mp);
return (error);
}
VOP_UNLOCK(vp, 0);
MNT_ILOCK(mp);
mp->mnt_flag |= MNT_QUOTA;
MNT_IUNLOCK(mp);
vpp = &ump->um_quotas[type];
if (*vpp != vp)
quotaoff1(td, mp, type);
/*
* When the directory vnode containing the quota file is
* inactivated, due to the shared lookup of the quota file
* vput()ing the dvp, the qsyncvp() call for the containing
* directory would try to acquire the quota lock exclusive.
* At the same time, lookup already locked the quota vnode
* shared. Mark the quota vnode lock as allowing recursion
* and automatically converting shared locks to exclusive.
*
* Also mark quota vnode as system.
*/
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
vp->v_vflag |= VV_SYSTEM;
VN_LOCK_AREC(vp);
VN_LOCK_DSHARE(vp);
VOP_UNLOCK(vp, 0);
*vpp = vp;
/*
* Save the credential of the process that turned on quotas.
* Set up the time limits for this quota.
*/
ump->um_cred[type] = crhold(td->td_ucred);
ump->um_btime[type] = MAX_DQ_TIME;
ump->um_itime[type] = MAX_IQ_TIME;
if (dqget(NULLVP, 0, ump, type, &dq) == 0) {
if (dq->dq_btime > 0)
ump->um_btime[type] = dq->dq_btime;
if (dq->dq_itime > 0)
ump->um_itime[type] = dq->dq_itime;
dqrele(NULLVP, dq);
}
/*
* Allow the getdq from getinoquota below to read the quota
* from file.
*/
UFS_LOCK(ump);
ump->um_qflags[type] &= ~QTF_CLOSING;
UFS_UNLOCK(ump);
/*
* Search vnodes associated with this mount point,
* adding references to quota file being opened.
* NB: only need to add dquot's for inodes being modified.
*/
again:
MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
goto again;
}
if (vp->v_type == VNON || vp->v_writecount == 0) {
VOP_UNLOCK(vp, 0);
vrele(vp);
continue;
}
error = getinoquota(VTOI(vp));
VOP_UNLOCK(vp, 0);
vrele(vp);
if (error) {
MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
break;
}
}
if (error)
quotaoff_inchange(td, mp, type);
UFS_LOCK(ump);
ump->um_qflags[type] &= ~QTF_OPENING;
KASSERT((ump->um_qflags[type] & QTF_CLOSING) == 0,
("quotaon: leaking flags"));
UFS_UNLOCK(ump);
vfs_unbusy(mp);
return (error);
}
/*
* ONLY USED FOR THE ROOT DIRECTORY. nfscl_nget() does the rest. If this
* function is going to be used to get Regular Files, code must be added
* to fill in the "struct nfsv4node".
* Look up a vnode/nfsnode by file handle.
* Callers must check for mount points!!
* In all cases, a pointer to a
* nfsnode structure is returned.
*/
int
ncl_nget(struct mount *mntp, u_int8_t *fhp, int fhsize, struct nfsnode **npp,
int lkflags)
{
struct thread *td = curthread; /* XXX */
struct nfsnode *np;
struct vnode *vp;
struct vnode *nvp;
int error;
u_int hash;
struct nfsmount *nmp;
struct nfsfh *nfhp;
nmp = VFSTONFS(mntp);
*npp = NULL;
hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
M_NFSFH, M_WAITOK);
bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
nfhp->nfh_len = fhsize;
error = vfs_hash_get(mntp, hash, lkflags,
td, &nvp, newnfs_vncmpf, nfhp);
FREE(nfhp, M_NFSFH);
if (error)
return (error);
if (nvp != NULL) {
*npp = VTONFS(nvp);
return (0);
}
np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
if (error) {
uma_zfree(newnfsnode_zone, np);
return (error);
}
vp = nvp;
KASSERT(vp->v_bufobj.bo_bsize != 0, ("ncl_nget: bo_bsize == 0"));
vp->v_bufobj.bo_ops = &buf_ops_newnfs;
vp->v_data = np;
np->n_vnode = vp;
/*
* Initialize the mutex even if the vnode is going to be a loser.
* This simplifies the logic in reclaim, which can then unconditionally
* destroy the mutex (in the case of the loser, or if hash_insert
* happened to return an error no special casing is needed).
*/
mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
/*
* NFS supports recursive and shared locking.
*/
lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
VN_LOCK_AREC(vp);
VN_LOCK_ASHARE(vp);
/*
* Are we getting the root? If so, make sure the vnode flags
* are correct
*/
if ((fhsize == nmp->nm_fhsize) &&
!bcmp(fhp, nmp->nm_fh, fhsize)) {
if (vp->v_type == VNON)
vp->v_type = VDIR;
vp->v_vflag |= VV_ROOT;
}
MALLOC(np->n_fhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
M_NFSFH, M_WAITOK);
bcopy(fhp, np->n_fhp->nfh_fh, fhsize);
np->n_fhp->nfh_len = fhsize;
error = insmntque(vp, mntp);
if (error != 0) {
*npp = NULL;
FREE((caddr_t)np->n_fhp, M_NFSFH);
mtx_destroy(&np->n_mtx);
uma_zfree(newnfsnode_zone, np);
return (error);
}
error = vfs_hash_insert(vp, hash, lkflags,
td, &nvp, newnfs_vncmpf, np->n_fhp);
if (error)
return (error);
if (nvp != NULL) {
*npp = VTONFS(nvp);
/* vfs_hash_insert() vput()'s the losing vnode */
return (0);
}
*npp = np;
return (0);
}
/*
* Look up a vnode/nfsnode by file handle.
* Callers must check for mount points!!
* In all cases, a pointer to a
* nfsnode structure is returned.
*/
int
nfs_nget(struct mount *mntp, nfsfh_t *fhp, int fhsize, struct nfsnode **npp, int flags)
{
struct thread *td = curthread; /* XXX */
struct nfsnode *np;
struct vnode *vp;
struct vnode *nvp;
int error;
u_int hash;
struct nfsmount *nmp;
struct nfs_vncmp ncmp;
nmp = VFSTONFS(mntp);
*npp = NULL;
hash = fnv_32_buf(fhp->fh_bytes, fhsize, FNV1_32_INIT);
ncmp.fhsize = fhsize;
ncmp.fh = fhp;
error = vfs_hash_get(mntp, hash, flags,
td, &nvp, nfs_vncmpf, &ncmp);
if (error)
return (error);
if (nvp != NULL) {
*npp = VTONFS(nvp);
return (0);
}
/*
* Allocate before getnewvnode since doing so afterward
* might cause a bogus v_data pointer to get dereferenced
* elsewhere if zalloc should block.
*/
np = uma_zalloc(nfsnode_zone, M_WAITOK | M_ZERO);
error = getnewvnode("nfs", mntp, &nfs_vnodeops, &nvp);
if (error) {
uma_zfree(nfsnode_zone, np);
return (error);
}
vp = nvp;
vp->v_bufobj.bo_ops = &buf_ops_nfs;
vp->v_data = np;
np->n_vnode = vp;
/*
* Initialize the mutex even if the vnode is going to be a loser.
* This simplifies the logic in reclaim, which can then unconditionally
* destroy the mutex (in the case of the loser, or if hash_insert happened
* to return an error no special casing is needed).
*/
mtx_init(&np->n_mtx, "NFSnode lock", NULL, MTX_DEF);
/*
* NFS supports recursive and shared locking.
*/
lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
VN_LOCK_AREC(vp);
VN_LOCK_ASHARE(vp);
if (fhsize > NFS_SMALLFH) {
np->n_fhp = malloc(fhsize, M_NFSBIGFH, M_WAITOK);
} else
np->n_fhp = &np->n_fh;
bcopy((caddr_t)fhp, (caddr_t)np->n_fhp, fhsize);
np->n_fhsize = fhsize;
error = insmntque(vp, mntp);
if (error != 0) {
*npp = NULL;
if (np->n_fhsize > NFS_SMALLFH) {
free((caddr_t)np->n_fhp, M_NFSBIGFH);
}
mtx_destroy(&np->n_mtx);
uma_zfree(nfsnode_zone, np);
return (error);
}
error = vfs_hash_insert(vp, hash, flags,
td, &nvp, nfs_vncmpf, &ncmp);
if (error)
return (error);
if (nvp != NULL) {
*npp = VTONFS(nvp);
/* vfs_hash_insert() vput()'s the losing vnode */
return (0);
}
*npp = np;
return (0);
}
/*
* Allocate a vnode
*/
int
pfs_vncache_alloc(struct mount *mp, struct vnode **vpp,
struct pfs_node *pn, pid_t pid)
{
struct pfs_vdata *pvd, *pvd2;
struct vnode *vp;
int error;
/*
* See if the vnode is in the cache.
* XXX linear search is not very efficient.
*/
retry:
mtx_lock(&pfs_vncache_mutex);
for (pvd = pfs_vncache; pvd; pvd = pvd->pvd_next) {
if (pvd->pvd_pn == pn && pvd->pvd_pid == pid &&
pvd->pvd_vnode->v_mount == mp) {
vp = pvd->pvd_vnode;
VI_LOCK(vp);
mtx_unlock(&pfs_vncache_mutex);
if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) {
++pfs_vncache_hits;
*vpp = vp;
/*
* Some callers cache_enter(vp) later, so
* we have to make sure it's not in the
* VFS cache so it doesn't get entered
* twice. A better solution would be to
* make pfs_vncache_alloc() responsible
* for entering the vnode in the VFS
* cache.
*/
cache_purge(vp);
return (0);
}
goto retry;
}
}
mtx_unlock(&pfs_vncache_mutex);
/* nope, get a new one */
pvd = malloc(sizeof *pvd, M_PFSVNCACHE, M_WAITOK);
pvd->pvd_next = pvd->pvd_prev = NULL;
error = getnewvnode("pseudofs", mp, &pfs_vnodeops, vpp);
if (error) {
free(pvd, M_PFSVNCACHE);
return (error);
}
pvd->pvd_pn = pn;
pvd->pvd_pid = pid;
(*vpp)->v_data = pvd;
switch (pn->pn_type) {
case pfstype_root:
(*vpp)->v_vflag = VV_ROOT;
#if 0
printf("root vnode allocated\n");
#endif
/* fall through */
case pfstype_dir:
case pfstype_this:
case pfstype_parent:
case pfstype_procdir:
(*vpp)->v_type = VDIR;
break;
case pfstype_file:
(*vpp)->v_type = VREG;
break;
case pfstype_symlink:
(*vpp)->v_type = VLNK;
break;
case pfstype_none:
KASSERT(0, ("pfs_vncache_alloc called for null node\n"));
default:
panic("%s has unexpected type: %d", pn->pn_name, pn->pn_type);
}
/*
* Propagate flag through to vnode so users know it can change
* if the process changes (i.e. execve)
*/
if ((pn->pn_flags & PFS_PROCDEP) != 0)
(*vpp)->v_vflag |= VV_PROCDEP;
pvd->pvd_vnode = *vpp;
vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
VN_LOCK_AREC(*vpp);
error = insmntque(*vpp, mp);
if (error != 0) {
free(pvd, M_PFSVNCACHE);
*vpp = NULLVP;
return (error);
}
retry2:
mtx_lock(&pfs_vncache_mutex);
/*
* Other thread may race with us, creating the entry we are
* going to insert into the cache. Recheck after
* pfs_vncache_mutex is reacquired.
*/
for (pvd2 = pfs_vncache; pvd2; pvd2 = pvd2->pvd_next) {
if (pvd2->pvd_pn == pn && pvd2->pvd_pid == pid &&
pvd2->pvd_vnode->v_mount == mp) {
vp = pvd2->pvd_vnode;
VI_LOCK(vp);
mtx_unlock(&pfs_vncache_mutex);
if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) {
++pfs_vncache_hits;
vgone(*vpp);
vput(*vpp);
*vpp = vp;
cache_purge(vp);
return (0);
}
goto retry2;
}
}
++pfs_vncache_misses;
if (++pfs_vncache_entries > pfs_vncache_maxentries)
pfs_vncache_maxentries = pfs_vncache_entries;
pvd->pvd_prev = NULL;
pvd->pvd_next = pfs_vncache;
if (pvd->pvd_next)
pvd->pvd_next->pvd_prev = pvd;
pfs_vncache = pvd;
mtx_unlock(&pfs_vncache_mutex);
return (0);
}
