/* $NetBSD: nfs_nfsdport.c,v 1.4 2021/03/29 02:13:38 simonb Exp $ */
/*-
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* Rick Macklem at The University of Guelph.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*/
/*
* Functions that perform the vfs operations required by the routines in
* nfsd_serv.c. It is hoped that this change will make the server more
* portable.
*/
int nfsrv_enable_crossmntpt = 1;
static int nfs_commit_blks;
static int nfs_commit_miss;
extern int nfsrv_issuedelegs;
extern int nfsrv_dolocallocks;
extern int nfsd_enable_stringtouid;
static struct nfsheur {
struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
off_t nh_nextoff; /* next offset for sequential detection */
int nh_use; /* use count for selection */
int nh_seqcount; /* heuristic */
} nfsheur[NUM_HEURISTIC];
/* Locate best candidate. */
try = 32;
hi = ((int)(vaddr_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
nh = &nfsheur[hi];
while (try--) {
if (nfsheur[hi].nh_vp == vp) {
nh = &nfsheur[hi];
break;
}
if (nfsheur[hi].nh_use > 0)
--nfsheur[hi].nh_use;
hi = (hi + 1) % NUM_HEURISTIC;
if (nfsheur[hi].nh_use < nh->nh_use)
nh = &nfsheur[hi];
}
/* Initialize hint if this is a new file. */
if (nh->nh_vp != vp) {
nh->nh_vp = vp;
nh->nh_nextoff = uio->uio_offset;
nh->nh_use = NHUSE_INIT;
if (uio->uio_offset == 0)
nh->nh_seqcount = 4;
else
nh->nh_seqcount = 1;
}
/*
* Get attributes into nfsvattr structure.
*/
int
nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
struct thread *p, int vpislocked)
{
int error, lockedit = 0;
if (vpislocked == 0) {
/*
* When vpislocked == 0, the vnode is either exclusively
* locked by this thread or not locked by this thread.
* As such, shared lock it, if not exclusively locked.
*/
if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
lockedit = 1;
NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
}
}
error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
if (lockedit != 0)
NFSVOPUNLOCK(vp, 0);
NFSEXITCODE(error);
return (error);
}
/*
* Get a file handle for a vnode.
*/
int
nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
{
int error;
/*
* Perform access checking for vnodes obtained from file handles that would
* refer to files already opened by a Unix client. You cannot just use
* vn_writechk() and VOP_ACCESSX() for two reasons.
* 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
* case.
* 2 - The owner is to be given access irrespective of mode bits for some
* operations, so that processes that chmod after opening a file don't
* break.
*/
int
nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
u_int32_t *supportedtypep)
{
struct vattr vattr;
int error = 0, getret = 0;
if (vpislocked == 0) {
if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
error = EPERM;
goto out;
}
}
if (accmode & VWRITE) {
/* Just vn_writechk() changed to check rdonly */
/*
* Disallow write attempts on read-only file systems;
* unless the file is a socket or a block or character
* device resident on the file system.
*/
if (NFSVNO_EXRDONLY(exp) ||
(vp->v_mount->mnt_flag & MNT_RDONLY)) {
switch (vp->v_type) {
case VREG:
case VDIR:
case VLNK:
error = EROFS;
default:
break;
}
}
/*
* If there's shared text associated with
* the inode, try to free it up once. If
* we fail, we can't allow writing.
*/
if (VOP_IS_TEXT(vp) && error == 0)
error = ETXTBSY;
}
if (error != 0) {
if (vpislocked == 0)
NFSVOPUNLOCK(vp, 0);
goto out;
}
/*
* Should the override still be applied when ACLs are enabled?
*/
error = VOP_ACCESSX(vp, accmode, cred, p);
if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
/*
* Try again with VEXPLICIT_DENY, to see if the test for
* deletion is supported.
*/
error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
if (error == 0) {
if (vp->v_type == VDIR) {
accmode &= ~(VDELETE | VDELETE_CHILD);
accmode |= VWRITE;
error = VOP_ACCESSX(vp, accmode, cred, p);
} else if (supportedtypep != NULL) {
*supportedtypep &= ~NFSACCESS_DELETE;
}
}
}
/*
* Allow certain operations for the owner (reads and writes
* on files that are already open).
*/
if (override != NFSACCCHK_NOOVERRIDE &&
(error == EPERM || error == EACCES)) {
if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
error = 0;
else if (override & NFSACCCHK_ALLOWOWNER) {
getret = VOP_GETATTR(vp, &vattr, cred);
if (getret == 0 && cred->cr_uid == vattr.va_uid)
error = 0;
}
}
if (vpislocked == 0)
NFSVOPUNLOCK(vp, 0);
out:
NFSEXITCODE(error);
return (error);
}
/*
* Set attribute(s) vnop.
*/
int
nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
struct thread *p, struct nfsexstuff *exp)
{
int error;
/*
* Set up nameidata for a lookup() call and do it.
*/
int
nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
struct vnode **retdirp)
{
struct componentname *cnp = &ndp->ni_cnd;
int i;
struct iovec aiov;
struct uio auio;
int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
int error = 0, crossmnt;
char *cp;
*retdirp = NULL;
cnp->cn_nameptr = cnp->cn_pnbuf;
ndp->ni_lcf = 0;
/*
* Extract and set starting directory.
*/
if (dp->v_type != VDIR) {
if (islocked)
vput(dp);
else
vrele(dp);
nfsvno_relpathbuf(ndp);
error = ENOTDIR;
goto out1;
}
if (islocked)
NFSVOPUNLOCK(dp, 0);
VREF(dp);
*retdirp = dp;
if (NFSVNO_EXRDONLY(exp))
cnp->cn_flags |= RDONLY;
ndp->ni_segflg = UIO_SYSSPACE;
crossmnt = 1;
if (nd->nd_flag & ND_PUBLOOKUP) {
ndp->ni_loopcnt = 0;
if (cnp->cn_pnbuf[0] == '/') {
vrele(dp);
/*
* Check for degenerate pathnames here, since lookup()
* panics on them.
*/
for (i = 1; i < ndp->ni_pathlen; i++)
if (cnp->cn_pnbuf[i] != '/')
break;
if (i == ndp->ni_pathlen) {
error = NFSERR_ACCES;
goto out;
}
dp = rootvnode;
VREF(dp);
}
} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
(nd->nd_flag & ND_NFSV4) == 0) {
/*
* Only cross mount points for NFSv4 when doing a
* mount while traversing the file system above
* the mount point, unless nfsrv_enable_crossmntpt is set.
*/
cnp->cn_flags |= NOCROSSMOUNT;
crossmnt = 0;
}
/*
* Initialize for scan, set ni_startdir and bump ref on dp again
* because lookup() will dereference ni_startdir.
*/
if (!lockleaf)
cnp->cn_flags |= LOCKLEAF;
for (;;) {
cnp->cn_nameptr = cnp->cn_pnbuf;
/*
* Call lookup() to do the real work. If an error occurs,
* ndp->ni_vp and ni_dvp are left uninitialized or NULL and
* we do not have to dereference anything before returning.
* In either case ni_startdir will be dereferenced and NULLed
* out.
*/
error = lookup(ndp);
if (error)
break;
/*
* Check for encountering a symbolic link. Trivial
* termination occurs if no symlink encountered.
*/
if ((cnp->cn_flags & ISSYMLINK) == 0) {
if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
nfsvno_relpathbuf(ndp);
if (ndp->ni_vp && !lockleaf)
NFSVOPUNLOCK(ndp->ni_vp, 0);
break;
}
/*
* Cleanup refs for next loop and check if root directory
* should replace current directory. Normally ni_dvp
* becomes the new base directory and is cleaned up when
* we loop. Explicitly null pointers after invalidation
* to clarify operation.
*/
vput(ndp->ni_vp);
ndp->ni_vp = NULL;
/*
* Set up a pathname buffer and return a pointer to it and, optionally
* set a hash pointer.
*/
void
nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
{
struct componentname *cnp = &ndp->ni_cnd;
/*
* symlink vnode op.
*/
int
nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
struct nfsexstuff *exp)
{
int error = 0;
if (ndp->ni_vp) {
vrele(ndp->ni_startdir);
nfsvno_relpathbuf(ndp);
if (ndp->ni_dvp == ndp->ni_vp)
vrele(ndp->ni_dvp);
else
vput(ndp->ni_dvp);
vrele(ndp->ni_vp);
error = EEXIST;
goto out;
}
error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
&nvap->na_vattr, pathcp);
vput(ndp->ni_dvp);
vrele(ndp->ni_startdir);
nfsvno_relpathbuf(ndp);
/*
* Although FreeBSD still had the lookup code in
* it for 7/current, there doesn't seem to be any
* point, since VOP_SYMLINK() returns the ni_vp.
* Just vput it for v2.
*/
if (!not_v2 && !error)
vput(ndp->ni_vp);
out:
NFSEXITCODE(error);
return (error);
}
/*
* Parse symbolic link arguments.
* This function has an ugly side effect. It will MALLOC() an area for
* the symlink and set iov_base to point to it, only if it succeeds.
* So, if it returns with uiop->uio_iov->iov_base != NULL, that must
* be FREE'd later.
*/
int
nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
struct thread *p, char **pathcpp, int *lenp)
{
u_int32_t *tl;
char *pathcp = NULL;
int error = 0, len;
struct nfsv2_sattr *sp;
/*
* A rename to '.' or '..' results in a prematurely
* unlocked vnode on FreeBSD5, so I'm just going to fail that
* here.
*/
if ((tondp->ni_cnd.cn_namelen == 1 &&
tondp->ni_cnd.cn_nameptr[0] == '.') ||
(tondp->ni_cnd.cn_namelen == 2 &&
tondp->ni_cnd.cn_nameptr[0] == '.' &&
tondp->ni_cnd.cn_nameptr[1] == '.')) {
error = EINVAL;
goto out;
}
}
if (fvp->v_type == VDIR && fvp->v_mountedhere) {
error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
goto out;
}
if (fvp->v_mount != tdvp->v_mount) {
error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
goto out;
}
if (fvp == tdvp) {
error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
goto out;
}
if (fvp == tvp) {
/*
* If source and destination are the same, there is nothing to
* do. Set error to -1 to indicate this.
*/
error = -1;
goto out;
}
if (ndflag & ND_NFSV4) {
if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
error = nfsrv_checkremove(fvp, 0, p);
NFSVOPUNLOCK(fvp, 0);
} else
error = EPERM;
if (tvp && !error)
error = nfsrv_checkremove(tvp, 1, p);
} else {
/*
* For NFSv2 and NFSv3, try to get rid of the delegation, so
* that the NFSv4 client won't be confused by the rename.
* Since nfsd_recalldelegation() can only be called on an
* unlocked vnode at this point and fvp is the file that will
* still exist after the rename, just do fvp.
*/
nfsd_recalldelegation(fvp, p);
}
out:
if (!error) {
error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
&fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
&tondp->ni_cnd);
} else {
if (tdvp == tvp)
vrele(tdvp);
else
vput(tdvp);
if (tvp)
vput(tvp);
vrele(fromndp->ni_dvp);
vrele(fvp);
if (error == -1)
error = 0;
}
vrele(tondp->ni_startdir);
nfsvno_relpathbuf(tondp);
out1:
vrele(fromndp->ni_startdir);
nfsvno_relpathbuf(fromndp);
NFSEXITCODE(error);
return (error);
}
/*
* Link vnode op.
*/
int
nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
struct thread *p, struct nfsexstuff *exp)
{
struct vnode *xp;
int error = 0;
xp = ndp->ni_vp;
if (xp != NULL) {
error = EEXIST;
} else {
xp = ndp->ni_dvp;
if (vp->v_mount != xp->v_mount)
error = EXDEV;
}
if (!error) {
NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
if ((vp->v_iflag & VI_DOOMED) == 0)
error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
else
error = EPERM;
if (ndp->ni_dvp == vp)
vrele(ndp->ni_dvp);
else
vput(ndp->ni_dvp);
NFSVOPUNLOCK(vp, 0);
} else {
if (ndp->ni_dvp == ndp->ni_vp)
vrele(ndp->ni_dvp);
else
vput(ndp->ni_dvp);
if (ndp->ni_vp)
vrele(ndp->ni_vp);
}
nfsvno_relpathbuf(ndp);
NFSEXITCODE(error);
return (error);
}
/*
* Do the fsync() appropriate for the commit.
*/
int
nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
struct thread *td)
{
int error = 0;
/*
* RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
* file is done. At this time VOP_FSYNC does not accept offset and
* byte count parameters so call VOP_FSYNC the whole file for now.
* The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
* File systems that do not use the buffer cache (as indicated
* by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
*/
if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
(vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
/*
* Give up and do the whole thing
*/
if (vp->v_object &&
(vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
VM_OBJECT_WLOCK(vp->v_object);
vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
VM_OBJECT_WUNLOCK(vp->v_object);
}
error = VOP_FSYNC(vp, MNT_WAIT, td);
} else {
/*
* Locate and synchronously write any buffers that fall
* into the requested range. Note: we are assuming that
* f_iosize is a power of 2.
*/
int iosize = vp->v_mount->mnt_stat.f_iosize;
int iomask = iosize - 1;
struct bufobj *bo;
daddr_t lblkno;
/*
* Align to iosize boundary, super-align to page boundary.
*/
if (off & iomask) {
cnt += off & iomask;
off &= ~(u_quad_t)iomask;
}
if (off & PAGE_MASK) {
cnt += off & PAGE_MASK;
off &= ~(u_quad_t)PAGE_MASK;
}
lblkno = off / iosize;
if (vp->v_object &&
(vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
VM_OBJECT_WLOCK(vp->v_object);
vm_object_page_clean(vp->v_object, off, off + cnt,
OBJPC_SYNC);
VM_OBJECT_WUNLOCK(vp->v_object);
}
bo = &vp->v_bufobj;
BO_LOCK(bo);
while (cnt > 0) {
struct buf *bp;
/*
* If we have a buffer and it is marked B_DELWRI we
* have to lock and write it. Otherwise the prior
* write is assumed to have already been committed.
*
* gbincore() can return invalid buffers now so we
* have to check that bit as well (though B_DELWRI
* should not be set if B_INVAL is set there could be
* a race here since we haven't locked the buffer).
*/
if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
BO_LOCK(bo);
continue; /* retry */
}
if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
B_DELWRI) {
bremfree(bp);
bp->b_flags &= ~B_ASYNC;
bwrite(bp);
++nfs_commit_miss;
} else
BUF_UNLOCK(bp);
BO_LOCK(bo);
}
++nfs_commit_blks;
if (cnt < iosize)
break;
cnt -= iosize;
++lblkno;
}
BO_UNLOCK(bo);
}
NFSEXITCODE(error);
return (error);
}
/*
* Statfs vnode op.
*/
int
nfsvno_statfs(struct vnode *vp, struct statfs *sf)
{
int error;
error = VFS_STATFS(vp->v_mount, sf);
if (error == 0) {
/*
* Since NFS handles these values as unsigned on the
* wire, there is no way to represent negative values,
* so set them to 0. Without this, they will appear
* to be very large positive values for clients like
* Solaris10.
*/
if (sf->f_bavail < 0)
sf->f_bavail = 0;
if (sf->f_ffree < 0)
sf->f_ffree = 0;
}
NFSEXITCODE(error);
return (error);
}
/*
* Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
* must handle nfsrv_opencheck() calls after any other access checks.
*/
void
nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
struct nfsexstuff *exp, struct vnode **vpp)
{
struct vnode *vp = NULL;
u_quad_t tempsize;
struct nfsexstuff nes;
/*
* Updates the file rev and sets the mtime and ctime
* to the current clock time, returning the va_filerev and va_Xtime
* values.
* Return ESTALE to indicate the vnode is VI_DOOMED.
*/
int
nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
struct ucred *cred, struct thread *p)
{
struct vattr va;
/*
* Glue routine to nfsv4_fillattr().
*/
int
nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
struct ucred *cred, struct thread *p, int isdgram, int reterr,
int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
{
int error;
/* Since the Readdir vnode ops vary, put the entire functions in here. */
/*
* nfs readdir service
* - mallocs what it thinks is enough to read
* count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
* - calls VOP_READDIR()
* - loops around building the reply
* if the output generated exceeds count break out of loop
* The NFSM_CLGET macro is used here so that the reply will be packed
* tightly in mbuf clusters.
* - it trims out records with d_fileno == 0
* this doesn't matter for Unix clients, but they might confuse clients
* for other os'.
* - it trims out records with d_type == DT_WHT
* these cannot be seen through NFS (unless we extend the protocol)
* The alternate call nfsrvd_readdirplus() does lookups as well.
* PS: The NFS protocol spec. does not clarify what the "count" byte
* argument is a count of.. just name strings and file id's or the
* entire reply rpc or ...
* I tried just file name and id sizes and it confused the Sun client,
* so I am using the full rpc size now. The "paranoia.." comment refers
* to including the status longwords that are not a part of the dir.
* "entry" structures, but are in the rpc.
*/
int
nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
{
struct dirent *dp;
u_int32_t *tl;
int dirlen;
char *cpos, *cend, *rbuf;
struct nfsvattr at;
int nlen, error = 0, getret = 1;
int siz, cnt, fullsiz, eofflag, ncookies;
u_int64_t off, toff, verf;
u_long *cookies = NULL, *cookiep;
struct uio io;
struct iovec iv;
int is_ufs;
if (nd->nd_repstat) {
nfsrv_postopattr(nd, getret, &at);
goto out;
}
if (nd->nd_flag & ND_NFSV2) {
NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
off = fxdr_unsigned(u_quad_t, *tl++);
} else {
NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
off = fxdr_hyper(tl);
tl += 2;
verf = fxdr_hyper(tl);
tl += 2;
}
toff = off;
cnt = fxdr_unsigned(int, *tl);
if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
cnt = NFS_SRVMAXDATA(nd);
siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
fullsiz = siz;
if (nd->nd_flag & ND_NFSV3) {
nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
p, 1);
#if 0
/*
* va_filerev is not sufficient as a cookie verifier,
* since it is not supposed to change when entries are
* removed/added unless that offset cookies returned to
* the client are no longer valid.
*/
if (!nd->nd_repstat && toff && verf != at.na_filerev)
nd->nd_repstat = NFSERR_BAD_COOKIE;
#endif
}
if (!nd->nd_repstat && vp->v_type != VDIR)
nd->nd_repstat = NFSERR_NOTDIR;
if (nd->nd_repstat == 0 && cnt == 0) {
if (nd->nd_flag & ND_NFSV2)
/* NFSv2 does not have NFSERR_TOOSMALL */
nd->nd_repstat = EPERM;
else
nd->nd_repstat = NFSERR_TOOSMALL;
}
if (!nd->nd_repstat)
nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
NFSACCCHK_VPISLOCKED, NULL);
if (nd->nd_repstat) {
vput(vp);
if (nd->nd_flag & ND_NFSV3)
nfsrv_postopattr(nd, getret, &at);
goto out;
}
is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
again:
eofflag = 0;
if (cookies) {
free((caddr_t)cookies, M_TEMP);
cookies = NULL;
}
/*
* Check for degenerate cases of nothing useful read.
* If so go try again
*/
cpos = rbuf;
cend = rbuf + siz;
dp = (struct dirent *)cpos;
cookiep = cookies;
/*
* For some reason FreeBSD's ufs_readdir() chooses to back the
* directory offset up to a block boundary, so it is necessary to
* skip over the records that precede the requested offset. This
* requires the assumption that file offset cookies monotonically
* increase.
*/
while (cpos < cend && ncookies > 0 &&
(dp->d_fileno == 0 || dp->d_type == DT_WHT ||
(is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
cpos += dp->d_reclen;
dp = (struct dirent *)cpos;
cookiep++;
ncookies--;
}
if (cpos >= cend || ncookies == 0) {
siz = fullsiz;
toff = off;
goto again;
}
vput(vp);
/*
* dirlen is the size of the reply, including all XDR and must
* not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
* if the XDR should be included in "count", but to be safe, we do.
* (Include the two booleans at the end of the reply in dirlen now.)
*/
if (nd->nd_flag & ND_NFSV3) {
nfsrv_postopattr(nd, getret, &at);
NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
txdr_hyper(at.na_filerev, tl);
dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
} else {
dirlen = 2 * NFSX_UNSIGNED;
}
/* Loop through the records and build reply */
while (cpos < cend && ncookies > 0) {
nlen = dp->d_namlen;
if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
nlen <= NFS_MAXNAMLEN) {
if (nd->nd_flag & ND_NFSV3)
dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
else
dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
if (dirlen > cnt) {
eofflag = 0;
break;
}
/*
* Use the server's maximum data transfer size as the upper bound
* on reply datalen.
*/
if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
cnt = NFS_SRVMAXDATA(nd);
/*
* siz is a "hint" of how much directory information (name, fileid,
* cookie) should be in the reply. At least one client "hints" 0,
* so I set it to cnt for that case. I also round it up to the
* next multiple of DIRBLKSIZ.
*/
if (siz <= 0)
siz = cnt;
siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
if (nd->nd_flag & ND_NFSV4) {
error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
if (error)
goto nfsmout;
NFSSET_ATTRBIT(&savbits, &attrbits);
NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
NFSZERO_ATTRBIT(&rderrbits);
NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
} else {
NFSZERO_ATTRBIT(&attrbits);
}
fullsiz = siz;
nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
if (!nd->nd_repstat) {
if (off && verf != at.na_filerev) {
/*
* va_filerev is not sufficient as a cookie verifier,
* since it is not supposed to change when entries are
* removed/added unless that offset cookies returned to
* the client are no longer valid.
*/
#if 0
if (nd->nd_flag & ND_NFSV4) {
nd->nd_repstat = NFSERR_NOTSAME;
} else {
nd->nd_repstat = NFSERR_BAD_COOKIE;
}
#endif
} else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
nd->nd_repstat = NFSERR_BAD_COOKIE;
}
}
if (!nd->nd_repstat && vp->v_type != VDIR)
nd->nd_repstat = NFSERR_NOTDIR;
if (!nd->nd_repstat && cnt == 0)
nd->nd_repstat = NFSERR_TOOSMALL;
if (!nd->nd_repstat)
nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
NFSACCCHK_VPISLOCKED, NULL);
if (nd->nd_repstat) {
vput(vp);
if (nd->nd_flag & ND_NFSV3)
nfsrv_postopattr(nd, getret, &at);
goto out;
}
is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
again:
eofflag = 0;
if (cookies) {
free((caddr_t)cookies, M_TEMP);
cookies = NULL;
}
if (!cookies && !nd->nd_repstat)
nd->nd_repstat = NFSERR_PERM;
if (!nd->nd_repstat)
nd->nd_repstat = getret;
if (nd->nd_repstat) {
vput(vp);
if (cookies)
free((caddr_t)cookies, M_TEMP);
free((caddr_t)rbuf, M_TEMP);
if (nd->nd_flag & ND_NFSV3)
nfsrv_postopattr(nd, getret, &at);
goto out;
}
/*
* If nothing read, return eof
* rpc reply
*/
if (siz == 0) {
vput(vp);
if (nd->nd_flag & ND_NFSV3)
nfsrv_postopattr(nd, getret, &at);
NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
txdr_hyper(at.na_filerev, tl);
tl += 2;
*tl++ = newnfs_false;
*tl = newnfs_true;
free((caddr_t)cookies, M_TEMP);
free((caddr_t)rbuf, M_TEMP);
goto out;
}
/*
* Check for degenerate cases of nothing useful read.
* If so go try again
*/
cpos = rbuf;
cend = rbuf + siz;
dp = (struct dirent *)cpos;
cookiep = cookies;
/*
* For some reason FreeBSD's ufs_readdir() chooses to back the
* directory offset up to a block boundary, so it is necessary to
* skip over the records that precede the requested offset. This
* requires the assumption that file offset cookies monotonically
* increase.
*/
while (cpos < cend && ncookies > 0 &&
(dp->d_fileno == 0 || dp->d_type == DT_WHT ||
(is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
((nd->nd_flag & ND_NFSV4) &&
((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
(dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
cpos += dp->d_reclen;
dp = (struct dirent *)cpos;
cookiep++;
ncookies--;
}
if (cpos >= cend || ncookies == 0) {
siz = fullsiz;
toff = off;
goto again;
}
/*
* Busy the file system so that the mount point won't go away
* and, as such, VFS_VGET() can be used safely.
*/
mp = vp->v_mount;
vfs_ref(mp);
NFSVOPUNLOCK(vp, 0);
nd->nd_repstat = vfs_busy(mp, 0);
vfs_rel(mp);
if (nd->nd_repstat != 0) {
vrele(vp);
free(cookies, M_TEMP);
free(rbuf, M_TEMP);
if (nd->nd_flag & ND_NFSV3)
nfsrv_postopattr(nd, getret, &at);
goto out;
}
/*
* Check to see if entries in this directory can be safely acquired
* via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
* ZFS snapshot directories need VOP_LOOKUP(), so that any
* automount of the snapshot directory that is required will
* be done.
* This needs to be done here for NFSv4, since NFSv4 never does
* a VFS_VGET() for "." or "..".
*/
if (is_zfs == 1) {
r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
if (r == EOPNOTSUPP) {
usevget = 0;
cn.cn_nameiop = LOOKUP;
cn.cn_lkflags = LK_SHARED | LK_RETRY;
cn.cn_cred = nd->nd_cred;
cn.cn_thread = p;
} else if (r == 0)
vput(nvp);
}
/*
* Save this position, in case there is an error before one entry
* is created.
*/
mb0 = nd->nd_mb;
bpos0 = nd->nd_bpos;
/*
* Fill in the first part of the reply.
* dirlen is the reply length in bytes and cannot exceed cnt.
* (Include the two booleans at the end of the reply in dirlen now,
* so we recognize when we have exceeded cnt.)
*/
if (nd->nd_flag & ND_NFSV3) {
dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
nfsrv_postopattr(nd, getret, &at);
} else {
dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
}
NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
txdr_hyper(at.na_filerev, tl);
/*
* Save this position, in case there is an empty reply needed.
*/
mb1 = nd->nd_mb;
bpos1 = nd->nd_bpos;
/* Loop through the records and build reply */
entrycnt = 0;
while (cpos < cend && ncookies > 0 && dirlen < cnt) {
nlen = dp->d_namlen;
if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
nlen <= NFS_MAXNAMLEN &&
((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
(nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
|| (nlen == 1 && dp->d_name[0] != '.'))) {
/*
* Save the current position in the reply, in case
* this entry exceeds cnt.
*/
mb1 = nd->nd_mb;
bpos1 = nd->nd_bpos;
/*
* For readdir_and_lookup get the vnode using
* the file number.
*/
nvp = NULL;
refp = NULL;
r = 0;
at_root = 0;
needs_unbusy = 0;
new_mp = mp;
mounted_on_fileno = (uint64_t)dp->d_fileno;
if ((nd->nd_flag & ND_NFSV3) ||
NFSNONZERO_ATTRBIT(&savbits)) {
if (nd->nd_flag & ND_NFSV4)
refp = nfsv4root_getreferral(NULL,
vp, dp->d_fileno);
if (refp == NULL) {
if (usevget)
r = VFS_VGET(mp, dp->d_fileno,
LK_SHARED, &nvp);
else
r = EOPNOTSUPP;
if (r == EOPNOTSUPP) {
if (usevget) {
usevget = 0;
cn.cn_nameiop = LOOKUP;
cn.cn_lkflags =
LK_SHARED |
LK_RETRY;
cn.cn_cred =
nd->nd_cred;
cn.cn_thread = p;
}
cn.cn_nameptr = dp->d_name;
cn.cn_namelen = nlen;
cn.cn_flags = ISLASTCN |
NOFOLLOW | LOCKLEAF;
if (nlen == 2 &&
dp->d_name[0] == '.' &&
dp->d_name[1] == '.')
cn.cn_flags |=
ISDOTDOT;
if (NFSVOPLOCK(vp, LK_SHARED)
!= 0) {
nd->nd_repstat = EPERM;
break;
}
if ((vp->v_vflag & VV_ROOT) != 0
&& (cn.cn_flags & ISDOTDOT)
!= 0) {
vref(vp);
nvp = vp;
r = 0;
} else {
r = VOP_LOOKUP(vp, &nvp,
&cn);
if (vp != nvp)
NFSVOPUNLOCK(vp,
0);
}
}
/*
* For NFSv4, check to see if nvp is
* a mount point and get the mount
* point vnode, as required.
*/
if (r == 0 &&
nfsrv_enable_crossmntpt != 0 &&
(nd->nd_flag & ND_NFSV4) != 0 &&
nvp->v_type == VDIR &&
nvp->v_mountedhere != NULL) {
new_mp = nvp->v_mountedhere;
r = vfs_busy(new_mp, 0);
vput(nvp);
nvp = NULL;
if (r == 0) {
r = VFS_ROOT(new_mp,
LK_SHARED, &nvp);
needs_unbusy = 1;
if (r == 0)
at_root = 1;
}
}
}
if (!r) {
if (refp == NULL &&
((nd->nd_flag & ND_NFSV3) ||
NFSNONZERO_ATTRBIT(&attrbits))) {
r = nfsvno_getfh(nvp, &nfh, p);
if (!r)
r = nfsvno_getattr(nvp, nvap,
nd->nd_cred, p, 1);
if (r == 0 && is_zfs == 1 &&
nfsrv_enable_crossmntpt != 0 &&
(nd->nd_flag & ND_NFSV4) != 0 &&
nvp->v_type == VDIR &&
vp->v_mount != nvp->v_mount) {
/*
* For a ZFS snapshot, there is a
* pseudo mount that does not set
* v_mountedhere, so it needs to
* be detected via a different
* mount structure.
*/
at_root = 1;
if (new_mp == mp)
new_mp = nvp->v_mount;
}
}
} else {
nvp = NULL;
}
if (r) {
if (!NFSISSET_ATTRBIT(&attrbits,
NFSATTRBIT_RDATTRERROR)) {
if (nvp != NULL)
vput(nvp);
if (needs_unbusy != 0)
vfs_unbusy(new_mp);
nd->nd_repstat = r;
break;
}
}
}
/*
* Loop around getting the setable attributes. If an unsupported
* one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
*/
if (retnotsup) {
nd->nd_repstat = NFSERR_ATTRNOTSUPP;
bitpos = NFSATTRBIT_MAX;
} else {
bitpos = 0;
}
for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
if (attrsum > attrsize) {
error = NFSERR_BADXDR;
goto nfsmout;
}
if (NFSISSET_ATTRBIT(attrbitp, bitpos))
switch (bitpos) {
case NFSATTRBIT_SIZE:
NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
if (vp != NULL && vp->v_type != VREG) {
error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
NFSERR_INVAL;
goto nfsmout;
}
nvap->na_size = fxdr_hyper(tl);
attrsum += NFSX_HYPER;
break;
case NFSATTRBIT_ACL:
error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
p);
if (error)
goto nfsmout;
if (aceerr && !nd->nd_repstat)
nd->nd_repstat = aceerr;
attrsum += aclsize;
break;
case NFSATTRBIT_ARCHIVE:
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
if (!nd->nd_repstat)
nd->nd_repstat = NFSERR_ATTRNOTSUPP;
attrsum += NFSX_UNSIGNED;
break;
case NFSATTRBIT_HIDDEN:
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
if (!nd->nd_repstat)
nd->nd_repstat = NFSERR_ATTRNOTSUPP;
attrsum += NFSX_UNSIGNED;
break;
case NFSATTRBIT_MIMETYPE:
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
i = fxdr_unsigned(int, *tl);
error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
if (error)
goto nfsmout;
if (!nd->nd_repstat)
nd->nd_repstat = NFSERR_ATTRNOTSUPP;
attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
break;
case NFSATTRBIT_MODE:
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
nvap->na_mode = nfstov_mode(*tl);
attrsum += NFSX_UNSIGNED;
break;
case NFSATTRBIT_OWNER:
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
j = fxdr_unsigned(int, *tl);
if (j < 0) {
error = NFSERR_BADXDR;
goto nfsmout;
}
if (j > NFSV4_SMALLSTR)
cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
else
cp = namestr;
error = nfsrv_mtostr(nd, cp, j);
if (error) {
if (j > NFSV4_SMALLSTR)
free(cp, M_NFSSTRING);
goto nfsmout;
}
if (!nd->nd_repstat) {
nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
p);
if (!nd->nd_repstat)
nvap->na_uid = uid;
}
if (j > NFSV4_SMALLSTR)
free(cp, M_NFSSTRING);
attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
break;
case NFSATTRBIT_OWNERGROUP:
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
j = fxdr_unsigned(int, *tl);
if (j < 0) {
error = NFSERR_BADXDR;
goto nfsmout;
}
if (j > NFSV4_SMALLSTR)
cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
else
cp = namestr;
error = nfsrv_mtostr(nd, cp, j);
if (error) {
if (j > NFSV4_SMALLSTR)
free(cp, M_NFSSTRING);
goto nfsmout;
}
if (!nd->nd_repstat) {
nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
p);
if (!nd->nd_repstat)
nvap->na_gid = gid;
}
if (j > NFSV4_SMALLSTR)
free(cp, M_NFSSTRING);
attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
break;
case NFSATTRBIT_SYSTEM:
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
if (!nd->nd_repstat)
nd->nd_repstat = NFSERR_ATTRNOTSUPP;
attrsum += NFSX_UNSIGNED;
break;
case NFSATTRBIT_TIMEACCESSSET:
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
attrsum += NFSX_UNSIGNED;
if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
fxdr_nfsv4time(tl, &nvap->na_atime);
toclient = 1;
attrsum += NFSX_V4TIME;
} else {
vfs_timestamp(&nvap->na_atime);
nvap->na_vaflags |= VA_UTIMES_NULL;
}
break;
case NFSATTRBIT_TIMEBACKUP:
NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
if (!nd->nd_repstat)
nd->nd_repstat = NFSERR_ATTRNOTSUPP;
attrsum += NFSX_V4TIME;
break;
case NFSATTRBIT_TIMECREATE:
NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
if (!nd->nd_repstat)
nd->nd_repstat = NFSERR_ATTRNOTSUPP;
attrsum += NFSX_V4TIME;
break;
case NFSATTRBIT_TIMEMODIFYSET:
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
attrsum += NFSX_UNSIGNED;
if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
fxdr_nfsv4time(tl, &nvap->na_mtime);
nvap->na_vaflags &= ~VA_UTIMES_NULL;
attrsum += NFSX_V4TIME;
} else {
vfs_timestamp(&nvap->na_mtime);
if (!toclient)
nvap->na_vaflags |= VA_UTIMES_NULL;
}
break;
default:
nd->nd_repstat = NFSERR_ATTRNOTSUPP;
/*
* set bitpos so we drop out of the loop.
*/
bitpos = NFSATTRBIT_MAX;
break;
}
}
/*
* some clients pad the attrlist, so we need to skip over the
* padding.
*/
if (attrsum > attrsize) {
error = NFSERR_BADXDR;
} else {
attrsize = NFSM_RNDUP(attrsize);
if (attrsum < attrsize)
error = nfsm_advance(nd, attrsize - attrsum, -1);
}
nfsmout:
NFSEXITCODE2(error, nd);
return (error);
}
/*
* Check to see if the operation is allowed for this security flavor.
* RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
* AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
* Also, allow Secinfo, so that it can acquire the correct flavor(s).
*/
if (nfsvno_testexp(nd, exp) &&
nd->nd_procnum != NFSV4OP_SECINFO &&
nd->nd_procnum != NFSPROC_FSINFO) {
if (nd->nd_flag & ND_NFSV4)
error = NFSERR_WRONGSEC;
else
error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
goto out;
}
/*
* Check to see if the file system is exported V4 only.
*/
if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
error = NFSERR_PROGNOTV4;
goto out;
}
/*
* Now, map the user credentials.
* (Note that ND_AUTHNONE will only be set for an NFSv3
* Fsinfo RPC. If set for anything else, this code might need
* to change.)
*/
if (NFSVNO_EXPORTED(exp)) {
if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
NFSVNO_EXPORTANON(exp) ||
(nd->nd_flag & ND_AUTHNONE) != 0) {
nd->nd_cred->cr_uid = credanon->cr_uid;
nd->nd_cred->cr_gid = credanon->cr_gid;
crsetgroups(nd->nd_cred, credanon->cr_ngroups,
credanon->cr_groups);
} else if ((nd->nd_flag & ND_GSS) == 0) {
/*
* If using AUTH_SYS, call nfsrv_getgrpscred() to see
* if there is a replacement credential with a group
* list set up by "nfsuserd -manage-gids".
* If there is no replacement, nfsrv_getgrpscred()
* simply returns its argument.
*/
nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
}
}
out:
NFSEXITCODE2(error, nd);
return (error);
}
/*
* Check exports.
*/
int
nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
struct ucred **credp)
{
int i, error, *secflavors;
error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
&exp->nes_numsecflavor, &secflavors);
if (error) {
if (nfs_rootfhset) {
exp->nes_exflag = 0;
exp->nes_numsecflavor = 0;
error = 0;
}
} else {
/* Copy the security flavors. */
for (i = 0; i < exp->nes_numsecflavor; i++)
exp->nes_secflavors[i] = secflavors[i];
}
NFSEXITCODE(error);
return (error);
}
/*
* Get a vnode for a file handle and export stuff.
*/
int
nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
int lktype, struct vnode **vpp, struct nfsexstuff *exp,
struct ucred **credp)
{
int i, error, *secflavors;
*credp = NULL;
exp->nes_numsecflavor = 0;
error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
if (error != 0)
/* Make sure the server replies ESTALE to the client. */
error = ESTALE;
if (nam && !error) {
error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
&exp->nes_numsecflavor, &secflavors);
if (error) {
if (nfs_rootfhset) {
exp->nes_exflag = 0;
exp->nes_numsecflavor = 0;
error = 0;
} else {
vput(*vpp);
}
} else {
/* Copy the security flavors. */
for (i = 0; i < exp->nes_numsecflavor; i++)
exp->nes_secflavors[i] = secflavors[i];
}
}
NFSEXITCODE(error);
return (error);
}
/*
* nfsd_fhtovp() - convert a fh to a vnode ptr
* - look up fsid in mount list (if not found ret error)
* - get vp and export rights by calling nfsvno_fhtovp()
* - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
* for AUTH_SYS
* - if mpp != NULL, return the mount point so that it can
* be used for vn_finished_write() by the caller
*/
void
nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
struct vnode **vpp, struct nfsexstuff *exp,
struct mount **mpp, int startwrite, struct thread *p)
{
struct mount *mp;
struct ucred *credanon;
fhandle_t *fhp;
fhp = (fhandle_t *)nfp->nfsrvfh_data;
/*
* Check for the special case of the nfsv4root_fh.
*/
mp = vfs_busyfs(&fhp->fh_fsid);
if (mpp != NULL)
*mpp = mp;
if (mp == NULL) {
*vpp = NULL;
nd->nd_repstat = ESTALE;
goto out;
}
/*
* For NFSv4 without a pseudo root fs, unexported file handles
* can be returned, so that Lookup works everywhere.
*/
if (!nd->nd_repstat && exp->nes_exflag == 0 &&
!(nd->nd_flag & ND_NFSV4)) {
vput(*vpp);
nd->nd_repstat = EACCES;
}
/*
* Personally, I've never seen any point in requiring a
* reserved port#, since only in the rare case where the
* clients are all boxes with secure system privileges,
* does it provide any enhanced security, but... some people
* believe it to be useful and keep putting this code back in.
* (There is also some "security checker" out there that
* complains if the nfs server doesn't enforce this.)
* However, note the following:
* RFC3530 (NFSv4) specifies that a reserved port# not be
* required.
* RFC2623 recommends that, if a reserved port# is checked for,
* that there be a way to turn that off--> ifdef'd.
*/
#ifdef NFS_REQRSVPORT
if (!nd->nd_repstat) {
struct sockaddr_in *saddr;
struct sockaddr_in6 *saddr6;
/*
* Called from nfssvc() to update the exports list. Just call
* vfs_export(). This has to be done, since the v4 root fake fs isn't
* in the mount list.
*/
int
nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
{
struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
int error = 0;
struct nameidata nd;
fhandle_t fh;
error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
nfs_rootfhset = 0;
else if (error == 0) {
if (nfsexargp->fspec == NULL) {
error = EPERM;
goto out;
}
/*
* If fspec != NULL, this is the v4root path.
*/
NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
nfsexargp->fspec, p);
if ((error = namei(&nd)) != 0)
goto out;
error = nfsvno_getfh(nd.ni_vp, &fh, p);
vrele(nd.ni_vp);
if (!error) {
nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
NFSBCOPY((caddr_t)&fh,
nfs_rootfh.nfsrvfh_data,
sizeof (fhandle_t));
nfs_rootfhset = 1;
}
}
out:
NFSEXITCODE(error);
return (error);
}
/*
* This function needs to test to see if the system is near its limit
* for memory allocation via malloc() or mget() and return True iff
* either of these resources are near their limit.
* XXX (For now, this is just a stub.)
*/
int nfsrv_testmalloclimit = 0;
int
nfsrv_mallocmget_limit(void)
{
static int printmesg = 0;
static int testval = 1;
if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
if ((printmesg++ % 100) == 0)
printf("nfsd: malloc/mget near limit\n");
return (1);
}
return (0);
}
/*
* BSD specific initialization of a mount point.
*/
void
nfsd_mntinit(void)
{
static int inited = 0;
/*
* Get a vnode for a file handle, without checking exports, etc.
*/
struct vnode *
nfsvno_getvp(fhandle_t *fhp)
{
struct mount *mp;
struct vnode *vp;
int error;
/*
* Do a local VOP_ADVLOCK().
*/
int
nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
u_int64_t end, struct thread *td)
{
int error = 0;
struct flock fl;
u_int64_t tlen;
if (nfsrv_dolocallocks == 0)
goto out;
ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
fl.l_whence = SEEK_SET;
fl.l_type = ftype;
fl.l_start = (off_t)first;
if (end == NFS64BITSSET) {
fl.l_len = 0;
} else {
tlen = end - first;
fl.l_len = (off_t)tlen;
}
/*
* For FreeBSD8, the l_pid and l_sysid must be set to the same
* values for all calls, so that all locks will be held by the
* nfsd server. (The nfsd server handles conflicts between the
* various clients.)
* Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
* bytes, so it can't be put in l_sysid.
*/
if (nfsv4_sysid == 0)
nfsv4_sysid = nlm_acquire_next_sysid();
fl.l_pid = (pid_t)0;
fl.l_sysid = (int)nfsv4_sysid;
/*
* Check the nfsv4 root exports.
*/
int
nfsvno_v4rootexport(struct nfsrv_descript *nd)
{
struct ucred *credanon;
int exflags, error = 0, numsecflavor, *secflavors, i;
error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
&credanon, &numsecflavor, &secflavors);
if (error) {
error = NFSERR_PROGUNAVAIL;
goto out;
}
if (credanon != NULL)
crfree(credanon);
for (i = 0; i < numsecflavor; i++) {
if (secflavors[i] == AUTH_SYS)
nd->nd_flag |= ND_EXAUTHSYS;
else if (secflavors[i] == RPCSEC_GSS_KRB5)
nd->nd_flag |= ND_EXGSS;
else if (secflavors[i] == RPCSEC_GSS_KRB5I)
nd->nd_flag |= ND_EXGSSINTEGRITY;
else if (secflavors[i] == RPCSEC_GSS_KRB5P)
nd->nd_flag |= ND_EXGSSPRIVACY;
}
out:
NFSEXITCODE(error);
return (error);
}
/*
* Nfs server pseudo system call for the nfsd's
*/
/*
* MPSAFE
*/
static int
nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
{
struct file *fp;
struct nfsd_addsock_args sockarg;
struct nfsd_nfsd_args nfsdarg;
cap_rights_t rights;
int error;
if (uap->flag & NFSSVC_NFSDADDSOCK) {
error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
if (error)
goto out;
/*
* Since we don't know what rights might be required,
* pretend that we need them all. It is better to be too
* careful than too reckless.
*/
error = fget(td, sockarg.sock,
cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
if (error != 0)
goto out;
if (fp->f_type != DTYPE_SOCKET) {
fdrop(fp, td);
error = EPERM;
goto out;
}
error = nfsrvd_addsock(fp);
fdrop(fp, td);
} else if (uap->flag & NFSSVC_NFSDNFSD) {
if (uap->argp == NULL) {
error = EINVAL;
goto out;
}
error = copyin(uap->argp, (caddr_t)&nfsdarg,
sizeof (nfsdarg));
if (error)
goto out;
error = nfsrvd_nfsd(td, &nfsdarg);
} else {
error = nfssvc_srvcall(td, uap, td->td_ucred);
}
/*
* Check exports.
* Returns 0 if ok, 1 otherwise.
*/
int
nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
{
int i;
/*
* This seems odd, but allow the case where the security flavor
* list is empty. This happens when NFSv4 is traversing non-exported
* file systems. Exported file systems should always have a non-empty
* security flavor list.
*/
if (exp->nes_numsecflavor == 0)
return (0);
for (i = 0; i < exp->nes_numsecflavor; i++) {
/*
* The tests for privacy and integrity must be first,
* since ND_GSS is set for everything but AUTH_SYS.
*/
if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
(nd->nd_flag & ND_GSSPRIVACY))
return (0);
if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
(nd->nd_flag & ND_GSSINTEGRITY))
return (0);
if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
(nd->nd_flag & ND_GSS))
return (0);
if (exp->nes_secflavors[i] == AUTH_SYS &&
(nd->nd_flag & ND_GSS) == 0)
return (0);
}
return (1);
}
/*
* Calculate a hash value for the fid in a file handle.
*/
uint32_t
nfsrv_hashfh(fhandle_t *fhp)
{
uint32_t hashval;
/*
* Signal the userland master nfsd to backup the stable restart file.
*/
void
nfsrv_backupstable(void)
{
struct proc *procp;
if (nfsd_master_proc != NULL) {
procp = pfind(nfsd_master_pid);
/* Try to make sure it is the correct process. */
if (procp == nfsd_master_proc &&
procp->p_stats->p_start.tv_sec ==
nfsd_master_start.tv_sec &&
procp->p_stats->p_start.tv_usec ==
nfsd_master_start.tv_usec &&
strcmp(procp->p_comm, nfsd_master_comm) == 0)
kern_psignal(procp, SIGUSR2);
else
nfsd_master_proc = NULL;
if (procp != NULL)
PROC_UNLOCK(procp);
}
}
extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
/*
* Called once to initialize data structures...
*/
static int
nfsd_modevent(module_t mod, int type, void *data)
{
int error = 0, i;
static int loaded = 0;
/* Clean out all NFSv4 state. */
nfsrv_throwawayallstate(curthread);
/* Clean the NFS server reply cache */
nfsrvd_cleancache();
/* Free up the krpc server pool. */
if (nfsrvd_pool != NULL)
svcpool_destroy(nfsrvd_pool);
/* and get rid of the locks */
for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
mtx_destroy(&nfsrchash_table[i].mtx);
mtx_destroy(&nfsrcahash_table[i].mtx);
}
mtx_destroy(&nfsrc_udpmtx);
mtx_destroy(&nfs_v4root_mutex);
mtx_destroy(&nfsv4root_mnt.mnt_mtx);
for (i = 0; i < nfsrv_sessionhashsize; i++)
mtx_destroy(&nfssessionhash[i].mtx);
lockdestroy(&nfsv4root_mnt.mnt_explock);
free(nfsclienthash, M_NFSDCLIENT);
free(nfslockhash, M_NFSDLOCKFILE);
free(nfssessionhash, M_NFSDSESSION);
loaded = 0;
break;
default:
error = EOPNOTSUPP;
break;
}