2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include <sys/capability.h>
40 * Functions that perform the vfs operations required by the routines in
41 * nfsd_serv.c. It is hoped that this change will make the server more
45 #include <fs/nfs/nfsport.h>
47 #include <sys/sysctl.h>
48 #include <nlm/nlm_prot.h>
51 FEATURE(nfsd, "NFSv4 server");
53 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
54 extern int nfsrv_useacl;
55 extern int newnfs_numnfsd;
56 extern struct mount nfsv4root_mnt;
57 extern struct nfsrv_stablefirst nfsrv_stablefirst;
58 extern void (*nfsd_call_servertimer)(void);
59 extern SVCPOOL *nfsrvd_pool;
60 extern struct nfsv4lock nfsd_suspend_lock;
61 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
63 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
64 struct mtx nfsrc_udpmtx;
65 struct mtx nfs_v4root_mutex;
66 struct nfsrvfh nfs_rootfh, nfs_pubfh;
67 int nfs_pubfhset = 0, nfs_rootfhset = 0;
68 struct proc *nfsd_master_proc = NULL;
69 static pid_t nfsd_master_pid = (pid_t)-1;
70 static char nfsd_master_comm[MAXCOMLEN + 1];
71 static struct timeval nfsd_master_start;
72 static uint32_t nfsv4_sysid = 0;
74 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
77 int nfsrv_enable_crossmntpt = 1;
78 static int nfs_commit_blks;
79 static int nfs_commit_miss;
80 extern int nfsrv_issuedelegs;
81 extern int nfsrv_dolocallocks;
83 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
85 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
90 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
91 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
92 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
93 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
95 #define MAX_REORDERED_RPC 16
96 #define NUM_HEURISTIC 1031
99 #define NHUSE_MAX 2048
101 static struct nfsheur {
102 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
103 off_t nh_nextoff; /* next offset for sequential detection */
104 int nh_use; /* use count for selection */
105 int nh_seqcount; /* heuristic */
106 } nfsheur[NUM_HEURISTIC];
110 * Heuristic to detect sequential operation.
112 static struct nfsheur *
113 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
118 /* Locate best candidate. */
120 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
123 if (nfsheur[hi].nh_vp == vp) {
127 if (nfsheur[hi].nh_use > 0)
128 --nfsheur[hi].nh_use;
129 hi = (hi + 1) % NUM_HEURISTIC;
130 if (nfsheur[hi].nh_use < nh->nh_use)
134 /* Initialize hint if this is a new file. */
135 if (nh->nh_vp != vp) {
137 nh->nh_nextoff = uio->uio_offset;
138 nh->nh_use = NHUSE_INIT;
139 if (uio->uio_offset == 0)
145 /* Calculate heuristic. */
146 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
147 uio->uio_offset == nh->nh_nextoff) {
148 /* See comments in vfs_vnops.c:sequential_heuristic(). */
149 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
150 if (nh->nh_seqcount > IO_SEQMAX)
151 nh->nh_seqcount = IO_SEQMAX;
152 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
153 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
154 /* Probably a reordered RPC, leave seqcount alone. */
155 } else if (nh->nh_seqcount > 1) {
156 nh->nh_seqcount /= 2;
160 nh->nh_use += NHUSE_INC;
161 if (nh->nh_use > NHUSE_MAX)
162 nh->nh_use = NHUSE_MAX;
167 * Get attributes into nfsvattr structure.
170 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
171 struct thread *p, int vpislocked)
173 int error, lockedit = 0;
175 if (vpislocked == 0) {
177 * When vpislocked == 0, the vnode is either exclusively
178 * locked by this thread or not locked by this thread.
179 * As such, shared lock it, if not exclusively locked.
181 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
183 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
186 error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
195 * Get a file handle for a vnode.
198 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
202 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
203 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
204 error = VOP_VPTOFH(vp, &fhp->fh_fid);
211 * Perform access checking for vnodes obtained from file handles that would
212 * refer to files already opened by a Unix client. You cannot just use
213 * vn_writechk() and VOP_ACCESSX() for two reasons.
214 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
216 * 2 - The owner is to be given access irrespective of mode bits for some
217 * operations, so that processes that chmod after opening a file don't
221 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
222 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
223 u_int32_t *supportedtypep)
226 int error = 0, getret = 0;
228 if (vpislocked == 0) {
229 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
234 if (accmode & VWRITE) {
235 /* Just vn_writechk() changed to check rdonly */
237 * Disallow write attempts on read-only file systems;
238 * unless the file is a socket or a block or character
239 * device resident on the file system.
241 if (NFSVNO_EXRDONLY(exp) ||
242 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
243 switch (vp->v_type) {
253 * If there's shared text associated with
254 * the inode, try to free it up once. If
255 * we fail, we can't allow writing.
257 if (VOP_IS_TEXT(vp) && error == 0)
267 * Should the override still be applied when ACLs are enabled?
269 error = VOP_ACCESSX(vp, accmode, cred, p);
270 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
272 * Try again with VEXPLICIT_DENY, to see if the test for
273 * deletion is supported.
275 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
277 if (vp->v_type == VDIR) {
278 accmode &= ~(VDELETE | VDELETE_CHILD);
280 error = VOP_ACCESSX(vp, accmode, cred, p);
281 } else if (supportedtypep != NULL) {
282 *supportedtypep &= ~NFSACCESS_DELETE;
288 * Allow certain operations for the owner (reads and writes
289 * on files that are already open).
291 if (override != NFSACCCHK_NOOVERRIDE &&
292 (error == EPERM || error == EACCES)) {
293 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
295 else if (override & NFSACCCHK_ALLOWOWNER) {
296 getret = VOP_GETATTR(vp, &vattr, cred);
297 if (getret == 0 && cred->cr_uid == vattr.va_uid)
310 * Set attribute(s) vnop.
313 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
314 struct thread *p, struct nfsexstuff *exp)
318 error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
324 * Set up nameidata for a lookup() call and do it.
327 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
328 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
329 struct vnode **retdirp)
331 struct componentname *cnp = &ndp->ni_cnd;
335 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
336 int error = 0, crossmnt;
340 cnp->cn_nameptr = cnp->cn_pnbuf;
341 ndp->ni_strictrelative = 0;
343 * Extract and set starting directory.
345 if (dp->v_type != VDIR) {
350 nfsvno_relpathbuf(ndp);
358 if (NFSVNO_EXRDONLY(exp))
359 cnp->cn_flags |= RDONLY;
360 ndp->ni_segflg = UIO_SYSSPACE;
363 if (nd->nd_flag & ND_PUBLOOKUP) {
365 if (cnp->cn_pnbuf[0] == '/') {
368 * Check for degenerate pathnames here, since lookup()
371 for (i = 1; i < ndp->ni_pathlen; i++)
372 if (cnp->cn_pnbuf[i] != '/')
374 if (i == ndp->ni_pathlen) {
375 error = NFSERR_ACCES;
381 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
382 (nd->nd_flag & ND_NFSV4) == 0) {
384 * Only cross mount points for NFSv4 when doing a
385 * mount while traversing the file system above
386 * the mount point, unless nfsrv_enable_crossmntpt is set.
388 cnp->cn_flags |= NOCROSSMOUNT;
393 * Initialize for scan, set ni_startdir and bump ref on dp again
394 * because lookup() will dereference ni_startdir.
398 ndp->ni_startdir = dp;
399 ndp->ni_rootdir = rootvnode;
400 ndp->ni_topdir = NULL;
403 cnp->cn_flags |= LOCKLEAF;
405 cnp->cn_nameptr = cnp->cn_pnbuf;
407 * Call lookup() to do the real work. If an error occurs,
408 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
409 * we do not have to dereference anything before returning.
410 * In either case ni_startdir will be dereferenced and NULLed
418 * Check for encountering a symbolic link. Trivial
419 * termination occurs if no symlink encountered.
421 if ((cnp->cn_flags & ISSYMLINK) == 0) {
422 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
423 nfsvno_relpathbuf(ndp);
424 if (ndp->ni_vp && !lockleaf)
425 NFSVOPUNLOCK(ndp->ni_vp, 0);
432 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
433 NFSVOPUNLOCK(ndp->ni_dvp, 0);
434 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
439 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
443 if (ndp->ni_pathlen > 1)
444 cp = uma_zalloc(namei_zone, M_WAITOK);
448 aiov.iov_len = MAXPATHLEN;
449 auio.uio_iov = &aiov;
452 auio.uio_rw = UIO_READ;
453 auio.uio_segflg = UIO_SYSSPACE;
455 auio.uio_resid = MAXPATHLEN;
456 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
459 if (ndp->ni_pathlen > 1)
460 uma_zfree(namei_zone, cp);
466 linklen = MAXPATHLEN - auio.uio_resid;
471 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
472 error = ENAMETOOLONG;
477 * Adjust or replace path
479 if (ndp->ni_pathlen > 1) {
480 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
481 uma_zfree(namei_zone, cnp->cn_pnbuf);
484 cnp->cn_pnbuf[linklen] = '\0';
485 ndp->ni_pathlen += linklen;
488 * Cleanup refs for next loop and check if root directory
489 * should replace current directory. Normally ni_dvp
490 * becomes the new base directory and is cleaned up when
491 * we loop. Explicitly null pointers after invalidation
492 * to clarify operation.
497 if (cnp->cn_pnbuf[0] == '/') {
499 ndp->ni_dvp = ndp->ni_rootdir;
502 ndp->ni_startdir = ndp->ni_dvp;
506 cnp->cn_flags &= ~LOCKLEAF;
510 nfsvno_relpathbuf(ndp);
513 ndp->ni_startdir = NULL;
514 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
519 NFSEXITCODE2(error, nd);
524 * Set up a pathname buffer and return a pointer to it and, optionally
525 * set a hash pointer.
528 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
530 struct componentname *cnp = &ndp->ni_cnd;
532 cnp->cn_flags |= (NOMACCHECK | HASBUF);
533 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
536 *bufpp = cnp->cn_pnbuf;
540 * Release the above path buffer, if not released by nfsvno_namei().
543 nfsvno_relpathbuf(struct nameidata *ndp)
546 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
548 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
549 ndp->ni_cnd.cn_flags &= ~HASBUF;
553 * Readlink vnode op into an mbuf list.
556 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
557 struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
559 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
560 struct iovec *ivp = iv;
561 struct uio io, *uiop = &io;
562 struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
563 int i, len, tlen, error = 0;
567 while (len < NFS_MAXPATHLEN) {
569 MCLGET(mp, M_WAITOK);
570 mp->m_len = NFSMSIZ(mp);
577 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
578 mp->m_len = NFS_MAXPATHLEN - len;
579 len = NFS_MAXPATHLEN;
583 ivp->iov_base = mtod(mp, caddr_t);
584 ivp->iov_len = mp->m_len;
589 uiop->uio_iovcnt = i;
590 uiop->uio_offset = 0;
591 uiop->uio_resid = len;
592 uiop->uio_rw = UIO_READ;
593 uiop->uio_segflg = UIO_SYSSPACE;
595 error = VOP_READLINK(vp, uiop, cred);
601 if (uiop->uio_resid > 0) {
602 len -= uiop->uio_resid;
603 tlen = NFSM_RNDUP(len);
604 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
616 * Read vnode op call into mbuf list.
619 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
620 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
626 int error = 0, len, left, siz, tlen, ioflag = 0;
627 struct mbuf *m2 = NULL, *m3;
628 struct uio io, *uiop = &io;
631 len = left = NFSM_RNDUP(cnt);
634 * Generate the mbuf list with the uio_iov ref. to it.
641 siz = min(M_TRAILINGSPACE(m), left);
650 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
652 uiop->uio_iov = iv2 = iv;
658 panic("nfsvno_read iov");
659 siz = min(M_TRAILINGSPACE(m), left);
661 iv->iov_base = mtod(m, caddr_t) + m->m_len;
670 uiop->uio_iovcnt = i;
671 uiop->uio_offset = off;
672 uiop->uio_resid = len;
673 uiop->uio_rw = UIO_READ;
674 uiop->uio_segflg = UIO_SYSSPACE;
675 nh = nfsrv_sequential_heuristic(uiop, vp);
676 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
677 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
678 FREE((caddr_t)iv2, M_TEMP);
684 nh->nh_nextoff = uiop->uio_offset;
685 tlen = len - uiop->uio_resid;
686 cnt = cnt < tlen ? cnt : tlen;
687 tlen = NFSM_RNDUP(cnt);
691 } else if (len != tlen || tlen != cnt)
692 nfsrv_adj(m3, len - tlen, tlen - cnt);
702 * Write vnode op from an mbuf list.
705 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
706 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
712 struct uio io, *uiop = &io;
715 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
717 uiop->uio_iov = iv = ivp;
718 uiop->uio_iovcnt = cnt;
719 i = mtod(mp, caddr_t) + mp->m_len - cp;
723 panic("nfsvno_write");
734 cp = mtod(mp, caddr_t);
738 if (stable == NFSWRITE_UNSTABLE)
739 ioflags = IO_NODELOCKED;
741 ioflags = (IO_SYNC | IO_NODELOCKED);
742 uiop->uio_resid = retlen;
743 uiop->uio_rw = UIO_WRITE;
744 uiop->uio_segflg = UIO_SYSSPACE;
746 uiop->uio_offset = off;
747 nh = nfsrv_sequential_heuristic(uiop, vp);
748 ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
749 error = VOP_WRITE(vp, uiop, ioflags, cred);
751 nh->nh_nextoff = uiop->uio_offset;
752 FREE((caddr_t)iv, M_TEMP);
759 * Common code for creating a regular file (plus special files for V2).
762 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
763 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
764 int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
769 error = nd->nd_repstat;
770 if (!error && ndp->ni_vp == NULL) {
771 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
772 vrele(ndp->ni_startdir);
773 error = VOP_CREATE(ndp->ni_dvp,
774 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
776 nfsvno_relpathbuf(ndp);
778 if (*exclusive_flagp) {
779 *exclusive_flagp = 0;
780 NFSVNO_ATTRINIT(nvap);
781 nvap->na_atime.tv_sec = cverf[0];
782 nvap->na_atime.tv_nsec = cverf[1];
783 error = VOP_SETATTR(ndp->ni_vp,
784 &nvap->na_vattr, nd->nd_cred);
788 * NFS V2 Only. nfsrvd_mknod() does this for V3.
789 * (This implies, just get out on an error.)
791 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
792 nvap->na_type == VFIFO) {
793 if (nvap->na_type == VCHR && rdev == 0xffffffff)
794 nvap->na_type = VFIFO;
795 if (nvap->na_type != VFIFO &&
796 (error = priv_check_cred(nd->nd_cred,
797 PRIV_VFS_MKNOD_DEV, 0))) {
798 vrele(ndp->ni_startdir);
799 nfsvno_relpathbuf(ndp);
803 nvap->na_rdev = rdev;
804 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
805 &ndp->ni_cnd, &nvap->na_vattr);
807 nfsvno_relpathbuf(ndp);
808 vrele(ndp->ni_startdir);
812 vrele(ndp->ni_startdir);
813 nfsvno_relpathbuf(ndp);
821 * Handle cases where error is already set and/or
823 * 1 - clean up the lookup
824 * 2 - iff !error and na_size set, truncate it
826 vrele(ndp->ni_startdir);
827 nfsvno_relpathbuf(ndp);
829 if (ndp->ni_dvp == *vpp)
833 if (!error && nvap->na_size != VNOVAL) {
834 error = nfsvno_accchk(*vpp, VWRITE,
835 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
836 NFSACCCHK_VPISLOCKED, NULL);
838 tempsize = nvap->na_size;
839 NFSVNO_ATTRINIT(nvap);
840 nvap->na_size = tempsize;
841 error = VOP_SETATTR(*vpp,
842 &nvap->na_vattr, nd->nd_cred);
855 * Do a mknod vnode op.
858 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
864 vtyp = nvap->na_type;
866 * Iff doesn't exist, create it.
869 vrele(ndp->ni_startdir);
870 nfsvno_relpathbuf(ndp);
876 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
877 vrele(ndp->ni_startdir);
878 nfsvno_relpathbuf(ndp);
880 error = NFSERR_BADTYPE;
884 vrele(ndp->ni_startdir);
885 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
886 &ndp->ni_cnd, &nvap->na_vattr);
888 nfsvno_relpathbuf(ndp);
890 if (nvap->na_type != VFIFO &&
891 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
892 vrele(ndp->ni_startdir);
893 nfsvno_relpathbuf(ndp);
897 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
898 &ndp->ni_cnd, &nvap->na_vattr);
900 nfsvno_relpathbuf(ndp);
901 vrele(ndp->ni_startdir);
903 * Since VOP_MKNOD returns the ni_vp, I can't
904 * see any reason to do the lookup.
917 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
918 struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
922 if (ndp->ni_vp != NULL) {
923 if (ndp->ni_dvp == ndp->ni_vp)
928 nfsvno_relpathbuf(ndp);
932 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
935 nfsvno_relpathbuf(ndp);
946 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
947 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
948 struct nfsexstuff *exp)
953 vrele(ndp->ni_startdir);
954 nfsvno_relpathbuf(ndp);
955 if (ndp->ni_dvp == ndp->ni_vp)
964 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
965 &nvap->na_vattr, pathcp);
967 vrele(ndp->ni_startdir);
968 nfsvno_relpathbuf(ndp);
970 * Although FreeBSD still had the lookup code in
971 * it for 7/current, there doesn't seem to be any
972 * point, since VOP_SYMLINK() returns the ni_vp.
973 * Just vput it for v2.
975 if (!not_v2 && !error)
984 * Parse symbolic link arguments.
985 * This function has an ugly side effect. It will MALLOC() an area for
986 * the symlink and set iov_base to point to it, only if it succeeds.
987 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
991 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
992 struct thread *p, char **pathcpp, int *lenp)
997 struct nfsv2_sattr *sp;
1001 if ((nd->nd_flag & ND_NFSV3) &&
1002 (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1004 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1005 len = fxdr_unsigned(int, *tl);
1006 if (len > NFS_MAXPATHLEN || len <= 0) {
1010 MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1011 error = nfsrv_mtostr(nd, pathcp, len);
1014 if (nd->nd_flag & ND_NFSV2) {
1015 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1016 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1020 NFSEXITCODE2(0, nd);
1024 free(pathcp, M_TEMP);
1025 NFSEXITCODE2(error, nd);
1030 * Remove a non-directory object.
1033 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1034 struct thread *p, struct nfsexstuff *exp)
1040 if (vp->v_type == VDIR)
1041 error = NFSERR_ISDIR;
1043 error = nfsrv_checkremove(vp, 1, p);
1045 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1046 if (ndp->ni_dvp == vp)
1051 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1052 nfsvno_relpathbuf(ndp);
1058 * Remove a directory.
1061 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1062 struct thread *p, struct nfsexstuff *exp)
1068 if (vp->v_type != VDIR) {
1073 * No rmdir "." please.
1075 if (ndp->ni_dvp == vp) {
1080 * The root of a mounted filesystem cannot be deleted.
1082 if (vp->v_vflag & VV_ROOT)
1086 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1087 if (ndp->ni_dvp == vp)
1092 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1093 nfsvno_relpathbuf(ndp);
1102 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1103 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1105 struct vnode *fvp, *tvp, *tdvp;
1108 fvp = fromndp->ni_vp;
1110 vrele(fromndp->ni_dvp);
1115 tdvp = tondp->ni_dvp;
1118 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1119 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1121 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1122 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1125 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1126 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1131 * A rename to '.' or '..' results in a prematurely
1132 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1135 if ((tondp->ni_cnd.cn_namelen == 1 &&
1136 tondp->ni_cnd.cn_nameptr[0] == '.') ||
1137 (tondp->ni_cnd.cn_namelen == 2 &&
1138 tondp->ni_cnd.cn_nameptr[0] == '.' &&
1139 tondp->ni_cnd.cn_nameptr[1] == '.')) {
1144 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1145 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1148 if (fvp->v_mount != tdvp->v_mount) {
1149 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1153 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1158 * If source and destination are the same, there is nothing to
1159 * do. Set error to -1 to indicate this.
1164 if (ndflag & ND_NFSV4) {
1165 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1166 error = nfsrv_checkremove(fvp, 0, p);
1167 NFSVOPUNLOCK(fvp, 0);
1171 error = nfsrv_checkremove(tvp, 1, p);
1174 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1175 * that the NFSv4 client won't be confused by the rename.
1176 * Since nfsd_recalldelegation() can only be called on an
1177 * unlocked vnode at this point and fvp is the file that will
1178 * still exist after the rename, just do fvp.
1180 nfsd_recalldelegation(fvp, p);
1184 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1185 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1194 vrele(fromndp->ni_dvp);
1199 vrele(tondp->ni_startdir);
1200 nfsvno_relpathbuf(tondp);
1202 vrele(fromndp->ni_startdir);
1203 nfsvno_relpathbuf(fromndp);
1212 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1213 struct thread *p, struct nfsexstuff *exp)
1223 if (vp->v_mount != xp->v_mount)
1227 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1228 if ((vp->v_iflag & VI_DOOMED) == 0)
1229 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1232 if (ndp->ni_dvp == vp)
1236 NFSVOPUNLOCK(vp, 0);
1238 if (ndp->ni_dvp == ndp->ni_vp)
1245 nfsvno_relpathbuf(ndp);
1251 * Do the fsync() appropriate for the commit.
1254 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1260 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1261 * file is done. At this time VOP_FSYNC does not accept offset and
1262 * byte count parameters so call VOP_FSYNC the whole file for now.
1263 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1265 if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
1267 * Give up and do the whole thing
1270 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1271 VM_OBJECT_WLOCK(vp->v_object);
1272 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1273 VM_OBJECT_WUNLOCK(vp->v_object);
1275 error = VOP_FSYNC(vp, MNT_WAIT, td);
1278 * Locate and synchronously write any buffers that fall
1279 * into the requested range. Note: we are assuming that
1280 * f_iosize is a power of 2.
1282 int iosize = vp->v_mount->mnt_stat.f_iosize;
1283 int iomask = iosize - 1;
1288 * Align to iosize boundry, super-align to page boundry.
1291 cnt += off & iomask;
1292 off &= ~(u_quad_t)iomask;
1294 if (off & PAGE_MASK) {
1295 cnt += off & PAGE_MASK;
1296 off &= ~(u_quad_t)PAGE_MASK;
1298 lblkno = off / iosize;
1301 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1302 VM_OBJECT_WLOCK(vp->v_object);
1303 vm_object_page_clean(vp->v_object, off, off + cnt,
1305 VM_OBJECT_WUNLOCK(vp->v_object);
1314 * If we have a buffer and it is marked B_DELWRI we
1315 * have to lock and write it. Otherwise the prior
1316 * write is assumed to have already been committed.
1318 * gbincore() can return invalid buffers now so we
1319 * have to check that bit as well (though B_DELWRI
1320 * should not be set if B_INVAL is set there could be
1321 * a race here since we haven't locked the buffer).
1323 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1324 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1325 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1327 continue; /* retry */
1329 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1332 bp->b_flags &= ~B_ASYNC;
1355 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1359 error = VFS_STATFS(vp->v_mount, sf);
1362 * Since NFS handles these values as unsigned on the
1363 * wire, there is no way to represent negative values,
1364 * so set them to 0. Without this, they will appear
1365 * to be very large positive values for clients like
1368 if (sf->f_bavail < 0)
1370 if (sf->f_ffree < 0)
1378 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1379 * must handle nfsrv_opencheck() calls after any other access checks.
1382 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1383 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1384 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1385 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1386 struct nfsexstuff *exp, struct vnode **vpp)
1388 struct vnode *vp = NULL;
1390 struct nfsexstuff nes;
1392 if (ndp->ni_vp == NULL)
1393 nd->nd_repstat = nfsrv_opencheck(clientid,
1394 stateidp, stp, NULL, nd, p, nd->nd_repstat);
1395 if (!nd->nd_repstat) {
1396 if (ndp->ni_vp == NULL) {
1397 vrele(ndp->ni_startdir);
1398 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1399 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1401 nfsvno_relpathbuf(ndp);
1402 if (!nd->nd_repstat) {
1403 if (*exclusive_flagp) {
1404 *exclusive_flagp = 0;
1405 NFSVNO_ATTRINIT(nvap);
1406 nvap->na_atime.tv_sec = cverf[0];
1407 nvap->na_atime.tv_nsec = cverf[1];
1408 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1409 &nvap->na_vattr, cred);
1411 nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1412 aclp, p, attrbitp, exp);
1417 if (ndp->ni_startdir)
1418 vrele(ndp->ni_startdir);
1419 nfsvno_relpathbuf(ndp);
1421 if (create == NFSV4OPEN_CREATE) {
1422 if (ndp->ni_dvp == vp)
1427 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1428 if (ndp->ni_cnd.cn_flags & RDONLY)
1429 NFSVNO_SETEXRDONLY(&nes);
1431 NFSVNO_EXINIT(&nes);
1432 nd->nd_repstat = nfsvno_accchk(vp,
1433 VWRITE, cred, &nes, p,
1434 NFSACCCHK_NOOVERRIDE,
1435 NFSACCCHK_VPISLOCKED, NULL);
1436 nd->nd_repstat = nfsrv_opencheck(clientid,
1437 stateidp, stp, vp, nd, p, nd->nd_repstat);
1438 if (!nd->nd_repstat) {
1439 tempsize = nvap->na_size;
1440 NFSVNO_ATTRINIT(nvap);
1441 nvap->na_size = tempsize;
1442 nd->nd_repstat = VOP_SETATTR(vp,
1443 &nvap->na_vattr, cred);
1445 } else if (vp->v_type == VREG) {
1446 nd->nd_repstat = nfsrv_opencheck(clientid,
1447 stateidp, stp, vp, nd, p, nd->nd_repstat);
1451 if (ndp->ni_cnd.cn_flags & HASBUF)
1452 nfsvno_relpathbuf(ndp);
1453 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1454 vrele(ndp->ni_startdir);
1455 if (ndp->ni_dvp == ndp->ni_vp)
1465 NFSEXITCODE2(0, nd);
1469 * Updates the file rev and sets the mtime and ctime
1470 * to the current clock time, returning the va_filerev and va_Xtime
1474 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1475 struct ucred *cred, struct thread *p)
1480 vfs_timestamp(&va.va_mtime);
1481 (void) VOP_SETATTR(vp, &va, cred);
1482 (void) nfsvno_getattr(vp, nvap, cred, p, 1);
1486 * Glue routine to nfsv4_fillattr().
1489 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1490 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1491 struct ucred *cred, struct thread *p, int isdgram, int reterr,
1492 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1496 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1497 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1499 NFSEXITCODE2(0, nd);
1503 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1505 * nfs readdir service
1506 * - mallocs what it thinks is enough to read
1507 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1508 * - calls VOP_READDIR()
1509 * - loops around building the reply
1510 * if the output generated exceeds count break out of loop
1511 * The NFSM_CLGET macro is used here so that the reply will be packed
1512 * tightly in mbuf clusters.
1513 * - it trims out records with d_fileno == 0
1514 * this doesn't matter for Unix clients, but they might confuse clients
1516 * - it trims out records with d_type == DT_WHT
1517 * these cannot be seen through NFS (unless we extend the protocol)
1518 * The alternate call nfsrvd_readdirplus() does lookups as well.
1519 * PS: The NFS protocol spec. does not clarify what the "count" byte
1520 * argument is a count of.. just name strings and file id's or the
1521 * entire reply rpc or ...
1522 * I tried just file name and id sizes and it confused the Sun client,
1523 * so I am using the full rpc size now. The "paranoia.." comment refers
1524 * to including the status longwords that are not a part of the dir.
1525 * "entry" structures, but are in the rpc.
1528 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1529 struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1534 char *cpos, *cend, *rbuf;
1536 int nlen, error = 0, getret = 1;
1537 int siz, cnt, fullsiz, eofflag, ncookies;
1538 u_int64_t off, toff, verf;
1539 u_long *cookies = NULL, *cookiep;
1544 if (nd->nd_repstat) {
1545 nfsrv_postopattr(nd, getret, &at);
1548 if (nd->nd_flag & ND_NFSV2) {
1549 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1550 off = fxdr_unsigned(u_quad_t, *tl++);
1552 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1553 off = fxdr_hyper(tl);
1555 verf = fxdr_hyper(tl);
1559 cnt = fxdr_unsigned(int, *tl);
1560 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1561 cnt = NFS_SRVMAXDATA(nd);
1562 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1564 if (nd->nd_flag & ND_NFSV3) {
1565 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1569 * va_filerev is not sufficient as a cookie verifier,
1570 * since it is not supposed to change when entries are
1571 * removed/added unless that offset cookies returned to
1572 * the client are no longer valid.
1574 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1575 nd->nd_repstat = NFSERR_BAD_COOKIE;
1578 if (!nd->nd_repstat && vp->v_type != VDIR)
1579 nd->nd_repstat = NFSERR_NOTDIR;
1580 if (nd->nd_repstat == 0 && cnt == 0) {
1581 if (nd->nd_flag & ND_NFSV2)
1582 /* NFSv2 does not have NFSERR_TOOSMALL */
1583 nd->nd_repstat = EPERM;
1585 nd->nd_repstat = NFSERR_TOOSMALL;
1587 if (!nd->nd_repstat)
1588 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1589 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1590 NFSACCCHK_VPISLOCKED, NULL);
1591 if (nd->nd_repstat) {
1593 if (nd->nd_flag & ND_NFSV3)
1594 nfsrv_postopattr(nd, getret, &at);
1597 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1598 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1602 free((caddr_t)cookies, M_TEMP);
1610 io.uio_offset = (off_t)off;
1612 io.uio_segflg = UIO_SYSSPACE;
1613 io.uio_rw = UIO_READ;
1615 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1617 off = (u_int64_t)io.uio_offset;
1619 siz -= io.uio_resid;
1621 if (!cookies && !nd->nd_repstat)
1622 nd->nd_repstat = NFSERR_PERM;
1623 if (nd->nd_flag & ND_NFSV3) {
1624 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1625 if (!nd->nd_repstat)
1626 nd->nd_repstat = getret;
1630 * Handles the failed cases. nd->nd_repstat == 0 past here.
1632 if (nd->nd_repstat) {
1634 free((caddr_t)rbuf, M_TEMP);
1636 free((caddr_t)cookies, M_TEMP);
1637 if (nd->nd_flag & ND_NFSV3)
1638 nfsrv_postopattr(nd, getret, &at);
1642 * If nothing read, return eof
1647 if (nd->nd_flag & ND_NFSV2) {
1648 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1650 nfsrv_postopattr(nd, getret, &at);
1651 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1652 txdr_hyper(at.na_filerev, tl);
1655 *tl++ = newnfs_false;
1657 FREE((caddr_t)rbuf, M_TEMP);
1658 FREE((caddr_t)cookies, M_TEMP);
1663 * Check for degenerate cases of nothing useful read.
1664 * If so go try again
1668 dp = (struct dirent *)cpos;
1672 * For some reason FreeBSD's ufs_readdir() chooses to back the
1673 * directory offset up to a block boundary, so it is necessary to
1674 * skip over the records that precede the requested offset. This
1675 * requires the assumption that file offset cookies monotonically
1677 * Since the offset cookies don't monotonically increase for ZFS,
1678 * this is not done when ZFS is the file system.
1680 while (cpos < cend && ncookies > 0 &&
1681 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1682 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1683 cpos += dp->d_reclen;
1684 dp = (struct dirent *)cpos;
1688 if (cpos >= cend || ncookies == 0) {
1696 * dirlen is the size of the reply, including all XDR and must
1697 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1698 * if the XDR should be included in "count", but to be safe, we do.
1699 * (Include the two booleans at the end of the reply in dirlen now.)
1701 if (nd->nd_flag & ND_NFSV3) {
1702 nfsrv_postopattr(nd, getret, &at);
1703 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1704 txdr_hyper(at.na_filerev, tl);
1705 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1707 dirlen = 2 * NFSX_UNSIGNED;
1710 /* Loop through the records and build reply */
1711 while (cpos < cend && ncookies > 0) {
1712 nlen = dp->d_namlen;
1713 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1714 nlen <= NFS_MAXNAMLEN) {
1715 if (nd->nd_flag & ND_NFSV3)
1716 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1718 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1725 * Build the directory record xdr from
1728 if (nd->nd_flag & ND_NFSV3) {
1729 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1730 *tl++ = newnfs_true;
1733 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1734 *tl++ = newnfs_true;
1736 *tl = txdr_unsigned(dp->d_fileno);
1737 (void) nfsm_strtom(nd, dp->d_name, nlen);
1738 if (nd->nd_flag & ND_NFSV3) {
1739 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1742 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1743 *tl = txdr_unsigned(*cookiep);
1745 cpos += dp->d_reclen;
1746 dp = (struct dirent *)cpos;
1752 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1753 *tl++ = newnfs_false;
1758 FREE((caddr_t)rbuf, M_TEMP);
1759 FREE((caddr_t)cookies, M_TEMP);
1762 NFSEXITCODE2(0, nd);
1766 NFSEXITCODE2(error, nd);
1771 * Readdirplus for V3 and Readdir for V4.
1774 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1775 struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1780 char *cpos, *cend, *rbuf;
1783 struct nfsvattr nva, at, *nvap = &nva;
1784 struct mbuf *mb0, *mb1;
1785 struct nfsreferral *refp;
1786 int nlen, r, error = 0, getret = 1, usevget = 1;
1787 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1788 caddr_t bpos0, bpos1;
1789 u_int64_t off, toff, verf;
1790 u_long *cookies = NULL, *cookiep;
1791 nfsattrbit_t attrbits, rderrbits, savbits;
1794 struct componentname cn;
1795 int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1796 struct mount *mp, *new_mp;
1797 uint64_t mounted_on_fileno;
1799 if (nd->nd_repstat) {
1800 nfsrv_postopattr(nd, getret, &at);
1803 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1804 off = fxdr_hyper(tl);
1807 verf = fxdr_hyper(tl);
1809 siz = fxdr_unsigned(int, *tl++);
1810 cnt = fxdr_unsigned(int, *tl);
1813 * Use the server's maximum data transfer size as the upper bound
1816 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1817 cnt = NFS_SRVMAXDATA(nd);
1820 * siz is a "hint" of how much directory information (name, fileid,
1821 * cookie) should be in the reply. At least one client "hints" 0,
1822 * so I set it to cnt for that case. I also round it up to the
1823 * next multiple of DIRBLKSIZ.
1827 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1829 if (nd->nd_flag & ND_NFSV4) {
1830 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1833 NFSSET_ATTRBIT(&savbits, &attrbits);
1834 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1835 NFSZERO_ATTRBIT(&rderrbits);
1836 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1838 NFSZERO_ATTRBIT(&attrbits);
1841 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1842 if (!nd->nd_repstat) {
1843 if (off && verf != at.na_filerev) {
1845 * va_filerev is not sufficient as a cookie verifier,
1846 * since it is not supposed to change when entries are
1847 * removed/added unless that offset cookies returned to
1848 * the client are no longer valid.
1851 if (nd->nd_flag & ND_NFSV4) {
1852 nd->nd_repstat = NFSERR_NOTSAME;
1854 nd->nd_repstat = NFSERR_BAD_COOKIE;
1857 } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1858 nd->nd_repstat = NFSERR_BAD_COOKIE;
1861 if (!nd->nd_repstat && vp->v_type != VDIR)
1862 nd->nd_repstat = NFSERR_NOTDIR;
1863 if (!nd->nd_repstat && cnt == 0)
1864 nd->nd_repstat = NFSERR_TOOSMALL;
1865 if (!nd->nd_repstat)
1866 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1867 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1868 NFSACCCHK_VPISLOCKED, NULL);
1869 if (nd->nd_repstat) {
1871 if (nd->nd_flag & ND_NFSV3)
1872 nfsrv_postopattr(nd, getret, &at);
1875 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1877 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1881 free((caddr_t)cookies, M_TEMP);
1889 io.uio_offset = (off_t)off;
1891 io.uio_segflg = UIO_SYSSPACE;
1892 io.uio_rw = UIO_READ;
1894 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1896 off = (u_int64_t)io.uio_offset;
1898 siz -= io.uio_resid;
1900 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1902 if (!cookies && !nd->nd_repstat)
1903 nd->nd_repstat = NFSERR_PERM;
1904 if (!nd->nd_repstat)
1905 nd->nd_repstat = getret;
1906 if (nd->nd_repstat) {
1909 free((caddr_t)cookies, M_TEMP);
1910 free((caddr_t)rbuf, M_TEMP);
1911 if (nd->nd_flag & ND_NFSV3)
1912 nfsrv_postopattr(nd, getret, &at);
1916 * If nothing read, return eof
1921 if (nd->nd_flag & ND_NFSV3)
1922 nfsrv_postopattr(nd, getret, &at);
1923 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1924 txdr_hyper(at.na_filerev, tl);
1926 *tl++ = newnfs_false;
1928 free((caddr_t)cookies, M_TEMP);
1929 free((caddr_t)rbuf, M_TEMP);
1934 * Check for degenerate cases of nothing useful read.
1935 * If so go try again
1939 dp = (struct dirent *)cpos;
1943 * For some reason FreeBSD's ufs_readdir() chooses to back the
1944 * directory offset up to a block boundary, so it is necessary to
1945 * skip over the records that precede the requested offset. This
1946 * requires the assumption that file offset cookies monotonically
1948 * Since the offset cookies don't monotonically increase for ZFS,
1949 * this is not done when ZFS is the file system.
1951 while (cpos < cend && ncookies > 0 &&
1952 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1953 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1954 ((nd->nd_flag & ND_NFSV4) &&
1955 ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1956 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1957 cpos += dp->d_reclen;
1958 dp = (struct dirent *)cpos;
1962 if (cpos >= cend || ncookies == 0) {
1969 * Busy the file system so that the mount point won't go away
1970 * and, as such, VFS_VGET() can be used safely.
1974 NFSVOPUNLOCK(vp, 0);
1975 nd->nd_repstat = vfs_busy(mp, 0);
1977 if (nd->nd_repstat != 0) {
1979 free(cookies, M_TEMP);
1981 if (nd->nd_flag & ND_NFSV3)
1982 nfsrv_postopattr(nd, getret, &at);
1987 * Save this position, in case there is an error before one entry
1991 bpos0 = nd->nd_bpos;
1994 * Fill in the first part of the reply.
1995 * dirlen is the reply length in bytes and cannot exceed cnt.
1996 * (Include the two booleans at the end of the reply in dirlen now,
1997 * so we recognize when we have exceeded cnt.)
1999 if (nd->nd_flag & ND_NFSV3) {
2000 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2001 nfsrv_postopattr(nd, getret, &at);
2003 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2005 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2006 txdr_hyper(at.na_filerev, tl);
2009 * Save this position, in case there is an empty reply needed.
2012 bpos1 = nd->nd_bpos;
2014 /* Loop through the records and build reply */
2016 while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2017 nlen = dp->d_namlen;
2018 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2019 nlen <= NFS_MAXNAMLEN &&
2020 ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2021 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2022 || (nlen == 1 && dp->d_name[0] != '.'))) {
2024 * Save the current position in the reply, in case
2025 * this entry exceeds cnt.
2028 bpos1 = nd->nd_bpos;
2031 * For readdir_and_lookup get the vnode using
2040 mounted_on_fileno = (uint64_t)dp->d_fileno;
2041 if ((nd->nd_flag & ND_NFSV3) ||
2042 NFSNONZERO_ATTRBIT(&savbits)) {
2043 if (nd->nd_flag & ND_NFSV4)
2044 refp = nfsv4root_getreferral(NULL,
2048 r = VFS_VGET(mp, dp->d_fileno,
2052 if (r == EOPNOTSUPP) {
2055 cn.cn_nameiop = LOOKUP;
2063 cn.cn_nameptr = dp->d_name;
2064 cn.cn_namelen = nlen;
2065 cn.cn_flags = ISLASTCN |
2066 NOFOLLOW | LOCKLEAF;
2068 dp->d_name[0] == '.' &&
2069 dp->d_name[1] == '.')
2072 if (NFSVOPLOCK(vp, LK_SHARED)
2074 nd->nd_repstat = EPERM;
2077 if ((vp->v_vflag & VV_ROOT) != 0
2078 && (cn.cn_flags & ISDOTDOT)
2084 r = VOP_LOOKUP(vp, &nvp,
2093 * For NFSv4, check to see if nvp is
2094 * a mount point and get the mount
2095 * point vnode, as required.
2098 nfsrv_enable_crossmntpt != 0 &&
2099 (nd->nd_flag & ND_NFSV4) != 0 &&
2100 nvp->v_type == VDIR &&
2101 nvp->v_mountedhere != NULL) {
2102 new_mp = nvp->v_mountedhere;
2103 r = vfs_busy(new_mp, 0);
2107 r = VFS_ROOT(new_mp,
2117 ((nd->nd_flag & ND_NFSV3) ||
2118 NFSNONZERO_ATTRBIT(&attrbits))) {
2119 r = nfsvno_getfh(nvp, &nfh, p);
2121 r = nfsvno_getattr(nvp, nvap,
2128 if (!NFSISSET_ATTRBIT(&attrbits,
2129 NFSATTRBIT_RDATTRERROR)) {
2132 if (needs_unbusy != 0)
2141 * Build the directory record xdr
2143 if (nd->nd_flag & ND_NFSV3) {
2144 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2145 *tl++ = newnfs_true;
2147 *tl = txdr_unsigned(dp->d_fileno);
2148 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2149 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2151 *tl = txdr_unsigned(*cookiep);
2152 nfsrv_postopattr(nd, 0, nvap);
2153 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2154 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2158 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2159 *tl++ = newnfs_true;
2161 *tl = txdr_unsigned(*cookiep);
2162 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2164 supports_nfsv4acls =
2165 nfs_supportsnfsv4acls(nvp);
2166 NFSVOPUNLOCK(nvp, 0);
2168 supports_nfsv4acls = 0;
2170 dirlen += nfsrv_putreferralattr(nd,
2173 if (nd->nd_repstat) {
2176 if (needs_unbusy != 0)
2181 dirlen += nfsvno_fillattr(nd, new_mp,
2182 nvp, nvap, &nfh, r, &rderrbits,
2183 nd->nd_cred, p, isdgram, 0,
2184 supports_nfsv4acls, at_root,
2187 dirlen += nfsvno_fillattr(nd, new_mp,
2188 nvp, nvap, &nfh, r, &attrbits,
2189 nd->nd_cred, p, isdgram, 0,
2190 supports_nfsv4acls, at_root,
2195 dirlen += (3 * NFSX_UNSIGNED);
2197 if (needs_unbusy != 0)
2202 cpos += dp->d_reclen;
2203 dp = (struct dirent *)cpos;
2211 * If dirlen > cnt, we must strip off the last entry. If that
2212 * results in an empty reply, report NFSERR_TOOSMALL.
2214 if (dirlen > cnt || nd->nd_repstat) {
2215 if (!nd->nd_repstat && entrycnt == 0)
2216 nd->nd_repstat = NFSERR_TOOSMALL;
2218 newnfs_trimtrailing(nd, mb0, bpos0);
2220 newnfs_trimtrailing(nd, mb1, bpos1);
2222 } else if (cpos < cend)
2224 if (!nd->nd_repstat) {
2225 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2226 *tl++ = newnfs_false;
2232 FREE((caddr_t)cookies, M_TEMP);
2233 FREE((caddr_t)rbuf, M_TEMP);
2236 NFSEXITCODE2(0, nd);
2240 NFSEXITCODE2(error, nd);
2245 * Get the settable attributes out of the mbuf list.
2246 * (Return 0 or EBADRPC)
2249 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2250 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2253 struct nfsv2_sattr *sp;
2254 int error = 0, toclient = 0;
2256 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2258 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2260 * Some old clients didn't fill in the high order 16bits.
2261 * --> check the low order 2 bytes for 0xffff
2263 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2264 nvap->na_mode = nfstov_mode(sp->sa_mode);
2265 if (sp->sa_uid != newnfs_xdrneg1)
2266 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2267 if (sp->sa_gid != newnfs_xdrneg1)
2268 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2269 if (sp->sa_size != newnfs_xdrneg1)
2270 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2271 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2273 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2275 nvap->na_atime.tv_sec =
2276 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2277 nvap->na_atime.tv_nsec = 0;
2280 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2281 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2284 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2285 if (*tl == newnfs_true) {
2286 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2287 nvap->na_mode = nfstov_mode(*tl);
2289 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2290 if (*tl == newnfs_true) {
2291 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2292 nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2294 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2295 if (*tl == newnfs_true) {
2296 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2297 nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2299 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2300 if (*tl == newnfs_true) {
2301 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2302 nvap->na_size = fxdr_hyper(tl);
2304 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2305 switch (fxdr_unsigned(int, *tl)) {
2306 case NFSV3SATTRTIME_TOCLIENT:
2307 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2308 fxdr_nfsv3time(tl, &nvap->na_atime);
2311 case NFSV3SATTRTIME_TOSERVER:
2312 vfs_timestamp(&nvap->na_atime);
2313 nvap->na_vaflags |= VA_UTIMES_NULL;
2316 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2317 switch (fxdr_unsigned(int, *tl)) {
2318 case NFSV3SATTRTIME_TOCLIENT:
2319 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2320 fxdr_nfsv3time(tl, &nvap->na_mtime);
2321 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2323 case NFSV3SATTRTIME_TOSERVER:
2324 vfs_timestamp(&nvap->na_mtime);
2326 nvap->na_vaflags |= VA_UTIMES_NULL;
2331 error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2334 NFSEXITCODE2(error, nd);
2339 * Handle the setable attributes for V4.
2340 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2343 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2344 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2349 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2351 u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2355 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2358 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2359 attrsize = fxdr_unsigned(int, *tl);
2362 * Loop around getting the setable attributes. If an unsupported
2363 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2366 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2367 bitpos = NFSATTRBIT_MAX;
2371 for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2372 if (attrsum > attrsize) {
2373 error = NFSERR_BADXDR;
2376 if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2378 case NFSATTRBIT_SIZE:
2379 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2380 nvap->na_size = fxdr_hyper(tl);
2381 attrsum += NFSX_HYPER;
2383 case NFSATTRBIT_ACL:
2384 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2388 if (aceerr && !nd->nd_repstat)
2389 nd->nd_repstat = aceerr;
2392 case NFSATTRBIT_ARCHIVE:
2393 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2394 if (!nd->nd_repstat)
2395 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2396 attrsum += NFSX_UNSIGNED;
2398 case NFSATTRBIT_HIDDEN:
2399 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2400 if (!nd->nd_repstat)
2401 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2402 attrsum += NFSX_UNSIGNED;
2404 case NFSATTRBIT_MIMETYPE:
2405 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2406 i = fxdr_unsigned(int, *tl);
2407 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2410 if (!nd->nd_repstat)
2411 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2412 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2414 case NFSATTRBIT_MODE:
2415 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2416 nvap->na_mode = nfstov_mode(*tl);
2417 attrsum += NFSX_UNSIGNED;
2419 case NFSATTRBIT_OWNER:
2420 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2421 j = fxdr_unsigned(int, *tl);
2423 error = NFSERR_BADXDR;
2426 if (j > NFSV4_SMALLSTR)
2427 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2430 error = nfsrv_mtostr(nd, cp, j);
2432 if (j > NFSV4_SMALLSTR)
2433 free(cp, M_NFSSTRING);
2436 if (!nd->nd_repstat) {
2437 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2439 if (!nd->nd_repstat)
2442 if (j > NFSV4_SMALLSTR)
2443 free(cp, M_NFSSTRING);
2444 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2446 case NFSATTRBIT_OWNERGROUP:
2447 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2448 j = fxdr_unsigned(int, *tl);
2450 error = NFSERR_BADXDR;
2453 if (j > NFSV4_SMALLSTR)
2454 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2457 error = nfsrv_mtostr(nd, cp, j);
2459 if (j > NFSV4_SMALLSTR)
2460 free(cp, M_NFSSTRING);
2463 if (!nd->nd_repstat) {
2464 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2466 if (!nd->nd_repstat)
2469 if (j > NFSV4_SMALLSTR)
2470 free(cp, M_NFSSTRING);
2471 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2473 case NFSATTRBIT_SYSTEM:
2474 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2475 if (!nd->nd_repstat)
2476 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2477 attrsum += NFSX_UNSIGNED;
2479 case NFSATTRBIT_TIMEACCESSSET:
2480 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2481 attrsum += NFSX_UNSIGNED;
2482 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2483 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2484 fxdr_nfsv4time(tl, &nvap->na_atime);
2486 attrsum += NFSX_V4TIME;
2488 vfs_timestamp(&nvap->na_atime);
2489 nvap->na_vaflags |= VA_UTIMES_NULL;
2492 case NFSATTRBIT_TIMEBACKUP:
2493 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2494 if (!nd->nd_repstat)
2495 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2496 attrsum += NFSX_V4TIME;
2498 case NFSATTRBIT_TIMECREATE:
2499 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2500 if (!nd->nd_repstat)
2501 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2502 attrsum += NFSX_V4TIME;
2504 case NFSATTRBIT_TIMEMODIFYSET:
2505 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2506 attrsum += NFSX_UNSIGNED;
2507 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2508 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2509 fxdr_nfsv4time(tl, &nvap->na_mtime);
2510 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2511 attrsum += NFSX_V4TIME;
2513 vfs_timestamp(&nvap->na_mtime);
2515 nvap->na_vaflags |= VA_UTIMES_NULL;
2519 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2521 * set bitpos so we drop out of the loop.
2523 bitpos = NFSATTRBIT_MAX;
2529 * some clients pad the attrlist, so we need to skip over the
2532 if (attrsum > attrsize) {
2533 error = NFSERR_BADXDR;
2535 attrsize = NFSM_RNDUP(attrsize);
2536 if (attrsum < attrsize)
2537 error = nfsm_advance(nd, attrsize - attrsum, -1);
2540 NFSEXITCODE2(error, nd);
2545 * Check/setup export credentials.
2548 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2549 struct ucred *credanon)
2554 * Check/setup credentials.
2556 if (nd->nd_flag & ND_GSS)
2557 exp->nes_exflag &= ~MNT_EXPORTANON;
2560 * Check to see if the operation is allowed for this security flavor.
2561 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2562 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2563 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2565 if (nfsvno_testexp(nd, exp) &&
2566 nd->nd_procnum != NFSV4OP_SECINFO &&
2567 nd->nd_procnum != NFSPROC_FSINFO) {
2568 if (nd->nd_flag & ND_NFSV4)
2569 error = NFSERR_WRONGSEC;
2571 error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2576 * Check to see if the file system is exported V4 only.
2578 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2579 error = NFSERR_PROGNOTV4;
2584 * Now, map the user credentials.
2585 * (Note that ND_AUTHNONE will only be set for an NFSv3
2586 * Fsinfo RPC. If set for anything else, this code might need
2589 if (NFSVNO_EXPORTED(exp) &&
2590 ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2591 NFSVNO_EXPORTANON(exp) ||
2592 (nd->nd_flag & ND_AUTHNONE))) {
2593 nd->nd_cred->cr_uid = credanon->cr_uid;
2594 nd->nd_cred->cr_gid = credanon->cr_gid;
2595 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2596 credanon->cr_groups);
2600 NFSEXITCODE2(error, nd);
2608 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2609 struct ucred **credp)
2611 int i, error, *secflavors;
2613 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2614 &exp->nes_numsecflavor, &secflavors);
2616 if (nfs_rootfhset) {
2617 exp->nes_exflag = 0;
2618 exp->nes_numsecflavor = 0;
2622 /* Copy the security flavors. */
2623 for (i = 0; i < exp->nes_numsecflavor; i++)
2624 exp->nes_secflavors[i] = secflavors[i];
2631 * Get a vnode for a file handle and export stuff.
2634 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2635 int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2636 struct ucred **credp)
2638 int i, error, *secflavors;
2641 exp->nes_numsecflavor = 0;
2642 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2644 /* Make sure the server replies ESTALE to the client. */
2646 if (nam && !error) {
2647 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2648 &exp->nes_numsecflavor, &secflavors);
2650 if (nfs_rootfhset) {
2651 exp->nes_exflag = 0;
2652 exp->nes_numsecflavor = 0;
2658 /* Copy the security flavors. */
2659 for (i = 0; i < exp->nes_numsecflavor; i++)
2660 exp->nes_secflavors[i] = secflavors[i];
2668 * nfsd_fhtovp() - convert a fh to a vnode ptr
2669 * - look up fsid in mount list (if not found ret error)
2670 * - get vp and export rights by calling nfsvno_fhtovp()
2671 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2673 * - if mpp != NULL, return the mount point so that it can
2674 * be used for vn_finished_write() by the caller
2677 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2678 struct vnode **vpp, struct nfsexstuff *exp,
2679 struct mount **mpp, int startwrite, struct thread *p)
2682 struct ucred *credanon;
2685 fhp = (fhandle_t *)nfp->nfsrvfh_data;
2687 * Check for the special case of the nfsv4root_fh.
2689 mp = vfs_busyfs(&fhp->fh_fsid);
2694 nd->nd_repstat = ESTALE;
2699 vn_start_write(NULL, mpp, V_WAIT);
2700 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2701 lktype = LK_EXCLUSIVE;
2703 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2708 * For NFSv4 without a pseudo root fs, unexported file handles
2709 * can be returned, so that Lookup works everywhere.
2711 if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2712 !(nd->nd_flag & ND_NFSV4)) {
2714 nd->nd_repstat = EACCES;
2718 * Personally, I've never seen any point in requiring a
2719 * reserved port#, since only in the rare case where the
2720 * clients are all boxes with secure system priviledges,
2721 * does it provide any enhanced security, but... some people
2722 * believe it to be useful and keep putting this code back in.
2723 * (There is also some "security checker" out there that
2724 * complains if the nfs server doesn't enforce this.)
2725 * However, note the following:
2726 * RFC3530 (NFSv4) specifies that a reserved port# not be
2728 * RFC2623 recommends that, if a reserved port# is checked for,
2729 * that there be a way to turn that off--> ifdef'd.
2731 #ifdef NFS_REQRSVPORT
2732 if (!nd->nd_repstat) {
2733 struct sockaddr_in *saddr;
2734 struct sockaddr_in6 *saddr6;
2736 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2737 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2738 if (!(nd->nd_flag & ND_NFSV4) &&
2739 ((saddr->sin_family == AF_INET &&
2740 ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2741 (saddr6->sin6_family == AF_INET6 &&
2742 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2744 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2747 #endif /* NFS_REQRSVPORT */
2750 * Check/setup credentials.
2752 if (!nd->nd_repstat) {
2753 nd->nd_saveduid = nd->nd_cred->cr_uid;
2754 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2758 if (credanon != NULL)
2760 if (nd->nd_repstat) {
2762 vn_finished_write(mp);
2769 NFSEXITCODE2(0, nd);
2776 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2778 struct filedesc *fdp;
2782 fdp = p->td_proc->p_fd;
2783 if (fd < 0 || fd >= fdp->fd_nfiles ||
2784 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2796 * Called from nfssvc() to update the exports list. Just call
2797 * vfs_export(). This has to be done, since the v4 root fake fs isn't
2798 * in the mount list.
2801 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2803 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2805 struct nameidata nd;
2808 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2809 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2811 else if (error == 0) {
2812 if (nfsexargp->fspec == NULL) {
2817 * If fspec != NULL, this is the v4root path.
2819 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2820 nfsexargp->fspec, p);
2821 if ((error = namei(&nd)) != 0)
2823 error = nfsvno_getfh(nd.ni_vp, &fh, p);
2826 nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2827 NFSBCOPY((caddr_t)&fh,
2828 nfs_rootfh.nfsrvfh_data,
2829 sizeof (fhandle_t));
2840 * Get the tcp socket sequence numbers we need.
2841 * (Maybe this should be moved to the tcp sources?)
2844 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2850 inp = sotoinpcb(so);
2851 KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2853 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2858 tp = intotcpcb(inp);
2859 if (tp->t_state != TCPS_ESTABLISHED) {
2864 *maxp = tp->snd_max;
2865 *unap = tp->snd_una;
2874 * This function needs to test to see if the system is near its limit
2875 * for memory allocation via malloc() or mget() and return True iff
2876 * either of these resources are near their limit.
2877 * XXX (For now, this is just a stub.)
2879 int nfsrv_testmalloclimit = 0;
2881 nfsrv_mallocmget_limit(void)
2883 static int printmesg = 0;
2884 static int testval = 1;
2886 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2887 if ((printmesg++ % 100) == 0)
2888 printf("nfsd: malloc/mget near limit\n");
2895 * BSD specific initialization of a mount point.
2900 static int inited = 0;
2905 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2906 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2907 TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2908 nfsv4root_mnt.mnt_export = NULL;
2909 TAILQ_INIT(&nfsv4root_opt);
2910 TAILQ_INIT(&nfsv4root_newopt);
2911 nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2912 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2913 nfsv4root_mnt.mnt_nvnodelistsize = 0;
2914 nfsv4root_mnt.mnt_activevnodelistsize = 0;
2918 * Get a vnode for a file handle, without checking exports, etc.
2921 nfsvno_getvp(fhandle_t *fhp)
2927 mp = vfs_busyfs(&fhp->fh_fsid);
2930 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2938 * Do a local VOP_ADVLOCK().
2941 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2942 u_int64_t end, struct thread *td)
2948 if (nfsrv_dolocallocks == 0)
2951 /* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2952 if ((vp->v_iflag & VI_DOOMED) != 0) {
2957 fl.l_whence = SEEK_SET;
2959 fl.l_start = (off_t)first;
2960 if (end == NFS64BITSSET) {
2964 fl.l_len = (off_t)tlen;
2967 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2968 * values for all calls, so that all locks will be held by the
2969 * nfsd server. (The nfsd server handles conflicts between the
2971 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2972 * bytes, so it can't be put in l_sysid.
2974 if (nfsv4_sysid == 0)
2975 nfsv4_sysid = nlm_acquire_next_sysid();
2976 fl.l_pid = (pid_t)0;
2977 fl.l_sysid = (int)nfsv4_sysid;
2979 NFSVOPUNLOCK(vp, 0);
2980 if (ftype == F_UNLCK)
2981 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2982 (F_POSIX | F_REMOTE));
2984 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2985 (F_POSIX | F_REMOTE));
2986 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2994 * Check the nfsv4 root exports.
2997 nfsvno_v4rootexport(struct nfsrv_descript *nd)
2999 struct ucred *credanon;
3000 int exflags, error = 0, numsecflavor, *secflavors, i;
3002 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3003 &credanon, &numsecflavor, &secflavors);
3005 error = NFSERR_PROGUNAVAIL;
3008 if (credanon != NULL)
3010 for (i = 0; i < numsecflavor; i++) {
3011 if (secflavors[i] == AUTH_SYS)
3012 nd->nd_flag |= ND_EXAUTHSYS;
3013 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3014 nd->nd_flag |= ND_EXGSS;
3015 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3016 nd->nd_flag |= ND_EXGSSINTEGRITY;
3017 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3018 nd->nd_flag |= ND_EXGSSPRIVACY;
3027 * Nfs server psuedo system call for the nfsd's
3033 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3036 struct nfsd_addsock_args sockarg;
3037 struct nfsd_nfsd_args nfsdarg;
3038 cap_rights_t rights;
3041 if (uap->flag & NFSSVC_NFSDADDSOCK) {
3042 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3046 * Since we don't know what rights might be required,
3047 * pretend that we need them all. It is better to be too
3048 * careful than too reckless.
3050 error = fget(td, sockarg.sock,
3051 cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3054 if (fp->f_type != DTYPE_SOCKET) {
3059 error = nfsrvd_addsock(fp);
3061 } else if (uap->flag & NFSSVC_NFSDNFSD) {
3062 if (uap->argp == NULL) {
3066 error = copyin(uap->argp, (caddr_t)&nfsdarg,
3070 error = nfsrvd_nfsd(td, &nfsdarg);
3072 error = nfssvc_srvcall(td, uap, td->td_ucred);
3081 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3083 struct nfsex_args export;
3084 struct file *fp = NULL;
3086 struct nfsd_clid adminrevoke;
3087 struct nfsd_dumplist dumplist;
3088 struct nfsd_dumpclients *dumpclients;
3089 struct nfsd_dumplocklist dumplocklist;
3090 struct nfsd_dumplocks *dumplocks;
3091 struct nameidata nd;
3093 int error = EINVAL, igotlock;
3095 static int suspend_nfsd = 0;
3097 if (uap->flag & NFSSVC_PUBLICFH) {
3098 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3099 sizeof (fhandle_t));
3100 error = copyin(uap->argp,
3101 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3104 } else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3105 error = copyin(uap->argp,(caddr_t)&export,
3106 sizeof (struct nfsex_args));
3108 error = nfsrv_v4rootexport(&export, cred, p);
3109 } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3112 } else if (uap->flag & NFSSVC_STABLERESTART) {
3113 error = copyin(uap->argp, (caddr_t)&stablefd,
3116 error = fp_getfvp(p, stablefd, &fp, &vp);
3117 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3119 if (!error && newnfs_numnfsd != 0)
3122 nfsrv_stablefirst.nsf_fp = fp;
3123 nfsrv_setupstable(p);
3125 } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3126 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3127 sizeof (struct nfsd_clid));
3129 error = nfsrv_adminrevoke(&adminrevoke, p);
3130 } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3131 error = copyin(uap->argp, (caddr_t)&dumplist,
3132 sizeof (struct nfsd_dumplist));
3133 if (!error && (dumplist.ndl_size < 1 ||
3134 dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3137 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3138 dumpclients = (struct nfsd_dumpclients *)malloc(len,
3140 nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3141 error = copyout(dumpclients,
3142 CAST_USER_ADDR_T(dumplist.ndl_list), len);
3143 free((caddr_t)dumpclients, M_TEMP);
3145 } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3146 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3147 sizeof (struct nfsd_dumplocklist));
3148 if (!error && (dumplocklist.ndllck_size < 1 ||
3149 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3152 error = nfsrv_lookupfilename(&nd,
3153 dumplocklist.ndllck_fname, p);
3155 len = sizeof (struct nfsd_dumplocks) *
3156 dumplocklist.ndllck_size;
3157 dumplocks = (struct nfsd_dumplocks *)malloc(len,
3159 nfsrv_dumplocks(nd.ni_vp, dumplocks,
3160 dumplocklist.ndllck_size, p);
3162 error = copyout(dumplocks,
3163 CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3164 free((caddr_t)dumplocks, M_TEMP);
3166 } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3169 nfsd_master_pid = procp->p_pid;
3170 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3171 nfsd_master_start = procp->p_stats->p_start;
3172 nfsd_master_proc = procp;
3174 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3175 NFSLOCKV4ROOTMUTEX();
3176 if (suspend_nfsd == 0) {
3177 /* Lock out all nfsd threads */
3179 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3180 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3181 } while (igotlock == 0 && suspend_nfsd == 0);
3184 NFSUNLOCKV4ROOTMUTEX();
3186 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3187 NFSLOCKV4ROOTMUTEX();
3188 if (suspend_nfsd != 0) {
3189 nfsv4_unlock(&nfsd_suspend_lock, 0);
3192 NFSUNLOCKV4ROOTMUTEX();
3202 * Returns 0 if ok, 1 otherwise.
3205 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3210 * This seems odd, but allow the case where the security flavor
3211 * list is empty. This happens when NFSv4 is traversing non-exported
3212 * file systems. Exported file systems should always have a non-empty
3213 * security flavor list.
3215 if (exp->nes_numsecflavor == 0)
3218 for (i = 0; i < exp->nes_numsecflavor; i++) {
3220 * The tests for privacy and integrity must be first,
3221 * since ND_GSS is set for everything but AUTH_SYS.
3223 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3224 (nd->nd_flag & ND_GSSPRIVACY))
3226 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3227 (nd->nd_flag & ND_GSSINTEGRITY))
3229 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3230 (nd->nd_flag & ND_GSS))
3232 if (exp->nes_secflavors[i] == AUTH_SYS &&
3233 (nd->nd_flag & ND_GSS) == 0)
3240 * Calculate a hash value for the fid in a file handle.
3243 nfsrv_hashfh(fhandle_t *fhp)
3247 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3252 * Signal the userland master nfsd to backup the stable restart file.
3255 nfsrv_backupstable(void)
3259 if (nfsd_master_proc != NULL) {
3260 procp = pfind(nfsd_master_pid);
3261 /* Try to make sure it is the correct process. */
3262 if (procp == nfsd_master_proc &&
3263 procp->p_stats->p_start.tv_sec ==
3264 nfsd_master_start.tv_sec &&
3265 procp->p_stats->p_start.tv_usec ==
3266 nfsd_master_start.tv_usec &&
3267 strcmp(procp->p_comm, nfsd_master_comm) == 0)
3268 kern_psignal(procp, SIGUSR2);
3270 nfsd_master_proc = NULL;
3277 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3280 * Called once to initialize data structures...
3283 nfsd_modevent(module_t mod, int type, void *data)
3286 static int loaded = 0;
3293 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3294 snprintf(nfsrchash_table[i].lock_name,
3295 sizeof(nfsrchash_table[i].lock_name), "nfsrc_tcp%d",
3297 mtx_init(&nfsrchash_table[i].mtx,
3298 nfsrchash_table[i].lock_name, NULL, MTX_DEF);
3300 mtx_init(&nfsrc_udpmtx, "nfs_udpcache_mutex", NULL, MTX_DEF);
3301 mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3302 mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3304 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3311 #ifdef VV_DISABLEDELEG
3312 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3313 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3315 nfsd_call_servertimer = nfsrv_servertimer;
3316 nfsd_call_nfsd = nfssvc_nfsd;
3321 if (newnfs_numnfsd != 0) {
3326 #ifdef VV_DISABLEDELEG
3327 vn_deleg_ops.vndeleg_recall = NULL;
3328 vn_deleg_ops.vndeleg_disable = NULL;
3330 nfsd_call_servertimer = NULL;
3331 nfsd_call_nfsd = NULL;
3333 /* Clean out all NFSv4 state. */
3334 nfsrv_throwawayallstate(curthread);
3336 /* Clean the NFS server reply cache */
3337 nfsrvd_cleancache();
3339 /* Free up the krpc server pool. */
3340 if (nfsrvd_pool != NULL)
3341 svcpool_destroy(nfsrvd_pool);
3343 /* and get rid of the locks */
3344 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++)
3345 mtx_destroy(&nfsrchash_table[i].mtx);
3346 mtx_destroy(&nfsrc_udpmtx);
3347 mtx_destroy(&nfs_v4root_mutex);
3348 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3349 lockdestroy(&nfsv4root_mnt.mnt_explock);
3361 static moduledata_t nfsd_mod = {
3366 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3368 /* So that loader and kldload(2) can find us, wherever we are.. */
3369 MODULE_VERSION(nfsd, 1);
3370 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3371 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3372 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3373 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3374 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);