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 nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
65 struct mtx nfsrc_udpmtx;
66 struct mtx nfs_v4root_mutex;
67 struct nfsrvfh nfs_rootfh, nfs_pubfh;
68 int nfs_pubfhset = 0, nfs_rootfhset = 0;
69 struct proc *nfsd_master_proc = NULL;
70 static pid_t nfsd_master_pid = (pid_t)-1;
71 static char nfsd_master_comm[MAXCOMLEN + 1];
72 static struct timeval nfsd_master_start;
73 static uint32_t nfsv4_sysid = 0;
75 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
78 int nfsrv_enable_crossmntpt = 1;
79 static int nfs_commit_blks;
80 static int nfs_commit_miss;
81 extern int nfsrv_issuedelegs;
82 extern int nfsrv_dolocallocks;
83 extern int nfsd_enable_stringtouid;
85 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
87 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
90 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
92 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
93 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
95 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
96 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
97 &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
99 #define MAX_REORDERED_RPC 16
100 #define NUM_HEURISTIC 1031
101 #define NHUSE_INIT 64
103 #define NHUSE_MAX 2048
105 static struct nfsheur {
106 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
107 off_t nh_nextoff; /* next offset for sequential detection */
108 int nh_use; /* use count for selection */
109 int nh_seqcount; /* heuristic */
110 } nfsheur[NUM_HEURISTIC];
114 * Heuristic to detect sequential operation.
116 static struct nfsheur *
117 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
122 /* Locate best candidate. */
124 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
127 if (nfsheur[hi].nh_vp == vp) {
131 if (nfsheur[hi].nh_use > 0)
132 --nfsheur[hi].nh_use;
133 hi = (hi + 1) % NUM_HEURISTIC;
134 if (nfsheur[hi].nh_use < nh->nh_use)
138 /* Initialize hint if this is a new file. */
139 if (nh->nh_vp != vp) {
141 nh->nh_nextoff = uio->uio_offset;
142 nh->nh_use = NHUSE_INIT;
143 if (uio->uio_offset == 0)
149 /* Calculate heuristic. */
150 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
151 uio->uio_offset == nh->nh_nextoff) {
152 /* See comments in vfs_vnops.c:sequential_heuristic(). */
153 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
154 if (nh->nh_seqcount > IO_SEQMAX)
155 nh->nh_seqcount = IO_SEQMAX;
156 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
157 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
158 /* Probably a reordered RPC, leave seqcount alone. */
159 } else if (nh->nh_seqcount > 1) {
160 nh->nh_seqcount /= 2;
164 nh->nh_use += NHUSE_INC;
165 if (nh->nh_use > NHUSE_MAX)
166 nh->nh_use = NHUSE_MAX;
171 * Get attributes into nfsvattr structure.
174 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
175 struct thread *p, int vpislocked)
177 int error, lockedit = 0;
179 if (vpislocked == 0) {
181 * When vpislocked == 0, the vnode is either exclusively
182 * locked by this thread or not locked by this thread.
183 * As such, shared lock it, if not exclusively locked.
185 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
187 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
190 error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
199 * Get a file handle for a vnode.
202 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
206 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
207 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
208 error = VOP_VPTOFH(vp, &fhp->fh_fid);
215 * Perform access checking for vnodes obtained from file handles that would
216 * refer to files already opened by a Unix client. You cannot just use
217 * vn_writechk() and VOP_ACCESSX() for two reasons.
218 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
220 * 2 - The owner is to be given access irrespective of mode bits for some
221 * operations, so that processes that chmod after opening a file don't
225 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
226 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
227 u_int32_t *supportedtypep)
230 int error = 0, getret = 0;
232 if (vpislocked == 0) {
233 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
238 if (accmode & VWRITE) {
239 /* Just vn_writechk() changed to check rdonly */
241 * Disallow write attempts on read-only file systems;
242 * unless the file is a socket or a block or character
243 * device resident on the file system.
245 if (NFSVNO_EXRDONLY(exp) ||
246 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
247 switch (vp->v_type) {
257 * If there's shared text associated with
258 * the inode, try to free it up once. If
259 * we fail, we can't allow writing.
261 if (VOP_IS_TEXT(vp) && error == 0)
271 * Should the override still be applied when ACLs are enabled?
273 error = VOP_ACCESSX(vp, accmode, cred, p);
274 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
276 * Try again with VEXPLICIT_DENY, to see if the test for
277 * deletion is supported.
279 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
281 if (vp->v_type == VDIR) {
282 accmode &= ~(VDELETE | VDELETE_CHILD);
284 error = VOP_ACCESSX(vp, accmode, cred, p);
285 } else if (supportedtypep != NULL) {
286 *supportedtypep &= ~NFSACCESS_DELETE;
292 * Allow certain operations for the owner (reads and writes
293 * on files that are already open).
295 if (override != NFSACCCHK_NOOVERRIDE &&
296 (error == EPERM || error == EACCES)) {
297 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
299 else if (override & NFSACCCHK_ALLOWOWNER) {
300 getret = VOP_GETATTR(vp, &vattr, cred);
301 if (getret == 0 && cred->cr_uid == vattr.va_uid)
314 * Set attribute(s) vnop.
317 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
318 struct thread *p, struct nfsexstuff *exp)
322 error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
328 * Set up nameidata for a lookup() call and do it.
331 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
332 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
333 struct vnode **retdirp)
335 struct componentname *cnp = &ndp->ni_cnd;
339 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
340 int error = 0, crossmnt;
344 cnp->cn_nameptr = cnp->cn_pnbuf;
345 ndp->ni_strictrelative = 0;
347 * Extract and set starting directory.
349 if (dp->v_type != VDIR) {
354 nfsvno_relpathbuf(ndp);
362 if (NFSVNO_EXRDONLY(exp))
363 cnp->cn_flags |= RDONLY;
364 ndp->ni_segflg = UIO_SYSSPACE;
367 if (nd->nd_flag & ND_PUBLOOKUP) {
369 if (cnp->cn_pnbuf[0] == '/') {
372 * Check for degenerate pathnames here, since lookup()
375 for (i = 1; i < ndp->ni_pathlen; i++)
376 if (cnp->cn_pnbuf[i] != '/')
378 if (i == ndp->ni_pathlen) {
379 error = NFSERR_ACCES;
385 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
386 (nd->nd_flag & ND_NFSV4) == 0) {
388 * Only cross mount points for NFSv4 when doing a
389 * mount while traversing the file system above
390 * the mount point, unless nfsrv_enable_crossmntpt is set.
392 cnp->cn_flags |= NOCROSSMOUNT;
397 * Initialize for scan, set ni_startdir and bump ref on dp again
398 * becuase lookup() will dereference ni_startdir.
402 ndp->ni_startdir = dp;
403 ndp->ni_rootdir = rootvnode;
404 ndp->ni_topdir = NULL;
407 cnp->cn_flags |= LOCKLEAF;
409 cnp->cn_nameptr = cnp->cn_pnbuf;
411 * Call lookup() to do the real work. If an error occurs,
412 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
413 * we do not have to dereference anything before returning.
414 * In either case ni_startdir will be dereferenced and NULLed
422 * Check for encountering a symbolic link. Trivial
423 * termination occurs if no symlink encountered.
425 if ((cnp->cn_flags & ISSYMLINK) == 0) {
426 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
427 nfsvno_relpathbuf(ndp);
428 if (ndp->ni_vp && !lockleaf)
429 NFSVOPUNLOCK(ndp->ni_vp, 0);
436 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
437 NFSVOPUNLOCK(ndp->ni_dvp, 0);
438 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
443 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
447 if (ndp->ni_pathlen > 1)
448 cp = uma_zalloc(namei_zone, M_WAITOK);
452 aiov.iov_len = MAXPATHLEN;
453 auio.uio_iov = &aiov;
456 auio.uio_rw = UIO_READ;
457 auio.uio_segflg = UIO_SYSSPACE;
459 auio.uio_resid = MAXPATHLEN;
460 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
463 if (ndp->ni_pathlen > 1)
464 uma_zfree(namei_zone, cp);
470 linklen = MAXPATHLEN - auio.uio_resid;
475 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
476 error = ENAMETOOLONG;
481 * Adjust or replace path
483 if (ndp->ni_pathlen > 1) {
484 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
485 uma_zfree(namei_zone, cnp->cn_pnbuf);
488 cnp->cn_pnbuf[linklen] = '\0';
489 ndp->ni_pathlen += linklen;
492 * Cleanup refs for next loop and check if root directory
493 * should replace current directory. Normally ni_dvp
494 * becomes the new base directory and is cleaned up when
495 * we loop. Explicitly null pointers after invalidation
496 * to clarify operation.
501 if (cnp->cn_pnbuf[0] == '/') {
503 ndp->ni_dvp = ndp->ni_rootdir;
506 ndp->ni_startdir = ndp->ni_dvp;
510 cnp->cn_flags &= ~LOCKLEAF;
514 uma_zfree(namei_zone, cnp->cn_pnbuf);
517 ndp->ni_startdir = NULL;
518 cnp->cn_flags &= ~HASBUF;
519 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
524 NFSEXITCODE2(error, nd);
529 * Set up a pathname buffer and return a pointer to it and, optionally
530 * set a hash pointer.
533 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
535 struct componentname *cnp = &ndp->ni_cnd;
537 cnp->cn_flags |= (NOMACCHECK | HASBUF);
538 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
541 *bufpp = cnp->cn_pnbuf;
545 * Release the above path buffer, if not released by nfsvno_namei().
548 nfsvno_relpathbuf(struct nameidata *ndp)
551 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
553 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
554 ndp->ni_cnd.cn_flags &= ~HASBUF;
558 * Readlink vnode op into an mbuf list.
561 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
562 struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
564 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
565 struct iovec *ivp = iv;
566 struct uio io, *uiop = &io;
567 struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
568 int i, len, tlen, error = 0;
572 while (len < NFS_MAXPATHLEN) {
575 mp->m_len = NFSMSIZ(mp);
582 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
583 mp->m_len = NFS_MAXPATHLEN - len;
584 len = NFS_MAXPATHLEN;
588 ivp->iov_base = mtod(mp, caddr_t);
589 ivp->iov_len = mp->m_len;
594 uiop->uio_iovcnt = i;
595 uiop->uio_offset = 0;
596 uiop->uio_resid = len;
597 uiop->uio_rw = UIO_READ;
598 uiop->uio_segflg = UIO_SYSSPACE;
600 error = VOP_READLINK(vp, uiop, cred);
606 if (uiop->uio_resid > 0) {
607 len -= uiop->uio_resid;
608 tlen = NFSM_RNDUP(len);
609 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
621 * Read vnode op call into mbuf list.
624 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
625 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
631 int error = 0, len, left, siz, tlen, ioflag = 0;
632 struct mbuf *m2 = NULL, *m3;
633 struct uio io, *uiop = &io;
636 len = left = NFSM_RNDUP(cnt);
639 * Generate the mbuf list with the uio_iov ref. to it.
646 siz = min(M_TRAILINGSPACE(m), left);
655 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
657 uiop->uio_iov = iv2 = iv;
663 panic("nfsvno_read iov");
664 siz = min(M_TRAILINGSPACE(m), left);
666 iv->iov_base = mtod(m, caddr_t) + m->m_len;
675 uiop->uio_iovcnt = i;
676 uiop->uio_offset = off;
677 uiop->uio_resid = len;
678 uiop->uio_rw = UIO_READ;
679 uiop->uio_segflg = UIO_SYSSPACE;
681 nh = nfsrv_sequential_heuristic(uiop, vp);
682 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
683 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
684 FREE((caddr_t)iv2, M_TEMP);
690 nh->nh_nextoff = uiop->uio_offset;
691 tlen = len - uiop->uio_resid;
692 cnt = cnt < tlen ? cnt : tlen;
693 tlen = NFSM_RNDUP(cnt);
697 } else if (len != tlen || tlen != cnt)
698 nfsrv_adj(m3, len - tlen, tlen - cnt);
708 * Write vnode op from an mbuf list.
711 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
712 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
718 struct uio io, *uiop = &io;
721 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
723 uiop->uio_iov = iv = ivp;
724 uiop->uio_iovcnt = cnt;
725 i = mtod(mp, caddr_t) + mp->m_len - cp;
729 panic("nfsvno_write");
740 cp = mtod(mp, caddr_t);
744 if (stable == NFSWRITE_UNSTABLE)
745 ioflags = IO_NODELOCKED;
747 ioflags = (IO_SYNC | IO_NODELOCKED);
748 uiop->uio_resid = retlen;
749 uiop->uio_rw = UIO_WRITE;
750 uiop->uio_segflg = UIO_SYSSPACE;
752 uiop->uio_offset = off;
753 nh = nfsrv_sequential_heuristic(uiop, vp);
754 ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
755 error = VOP_WRITE(vp, uiop, ioflags, cred);
757 nh->nh_nextoff = uiop->uio_offset;
758 FREE((caddr_t)iv, M_TEMP);
765 * Common code for creating a regular file (plus special files for V2).
768 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
769 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
770 int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
775 error = nd->nd_repstat;
776 if (!error && ndp->ni_vp == NULL) {
777 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
778 vrele(ndp->ni_startdir);
779 error = VOP_CREATE(ndp->ni_dvp,
780 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
782 nfsvno_relpathbuf(ndp);
784 if (*exclusive_flagp) {
785 *exclusive_flagp = 0;
786 NFSVNO_ATTRINIT(nvap);
787 nvap->na_atime.tv_sec = cverf[0];
788 nvap->na_atime.tv_nsec = cverf[1];
789 error = VOP_SETATTR(ndp->ni_vp,
790 &nvap->na_vattr, nd->nd_cred);
794 * NFS V2 Only. nfsrvd_mknod() does this for V3.
795 * (This implies, just get out on an error.)
797 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
798 nvap->na_type == VFIFO) {
799 if (nvap->na_type == VCHR && rdev == 0xffffffff)
800 nvap->na_type = VFIFO;
801 if (nvap->na_type != VFIFO &&
802 (error = priv_check_cred(nd->nd_cred,
803 PRIV_VFS_MKNOD_DEV, 0))) {
804 vrele(ndp->ni_startdir);
805 nfsvno_relpathbuf(ndp);
809 nvap->na_rdev = rdev;
810 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
811 &ndp->ni_cnd, &nvap->na_vattr);
813 nfsvno_relpathbuf(ndp);
814 vrele(ndp->ni_startdir);
818 vrele(ndp->ni_startdir);
819 nfsvno_relpathbuf(ndp);
827 * Handle cases where error is already set and/or
829 * 1 - clean up the lookup
830 * 2 - iff !error and na_size set, truncate it
832 vrele(ndp->ni_startdir);
833 nfsvno_relpathbuf(ndp);
835 if (ndp->ni_dvp == *vpp)
839 if (!error && nvap->na_size != VNOVAL) {
840 error = nfsvno_accchk(*vpp, VWRITE,
841 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
842 NFSACCCHK_VPISLOCKED, NULL);
844 tempsize = nvap->na_size;
845 NFSVNO_ATTRINIT(nvap);
846 nvap->na_size = tempsize;
847 error = VOP_SETATTR(*vpp,
848 &nvap->na_vattr, nd->nd_cred);
861 * Do a mknod vnode op.
864 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
870 vtyp = nvap->na_type;
872 * Iff doesn't exist, create it.
875 vrele(ndp->ni_startdir);
876 nfsvno_relpathbuf(ndp);
882 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
883 vrele(ndp->ni_startdir);
884 nfsvno_relpathbuf(ndp);
886 error = NFSERR_BADTYPE;
890 vrele(ndp->ni_startdir);
891 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
892 &ndp->ni_cnd, &nvap->na_vattr);
894 nfsvno_relpathbuf(ndp);
896 if (nvap->na_type != VFIFO &&
897 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
898 vrele(ndp->ni_startdir);
899 nfsvno_relpathbuf(ndp);
903 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
904 &ndp->ni_cnd, &nvap->na_vattr);
906 nfsvno_relpathbuf(ndp);
907 vrele(ndp->ni_startdir);
909 * Since VOP_MKNOD returns the ni_vp, I can't
910 * see any reason to do the lookup.
923 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
924 struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
928 if (ndp->ni_vp != NULL) {
929 if (ndp->ni_dvp == ndp->ni_vp)
934 nfsvno_relpathbuf(ndp);
938 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
941 nfsvno_relpathbuf(ndp);
952 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
953 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
954 struct nfsexstuff *exp)
959 vrele(ndp->ni_startdir);
960 nfsvno_relpathbuf(ndp);
961 if (ndp->ni_dvp == ndp->ni_vp)
970 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
971 &nvap->na_vattr, pathcp);
973 vrele(ndp->ni_startdir);
974 nfsvno_relpathbuf(ndp);
976 * Although FreeBSD still had the lookup code in
977 * it for 7/current, there doesn't seem to be any
978 * point, since VOP_SYMLINK() returns the ni_vp.
979 * Just vput it for v2.
981 if (!not_v2 && !error)
990 * Parse symbolic link arguments.
991 * This function has an ugly side effect. It will MALLOC() an area for
992 * the symlink and set iov_base to point to it, only if it succeeds.
993 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
997 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
998 struct thread *p, char **pathcpp, int *lenp)
1001 char *pathcp = NULL;
1003 struct nfsv2_sattr *sp;
1007 if ((nd->nd_flag & ND_NFSV3) &&
1008 (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1010 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1011 len = fxdr_unsigned(int, *tl);
1012 if (len > NFS_MAXPATHLEN || len <= 0) {
1016 MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1017 error = nfsrv_mtostr(nd, pathcp, len);
1020 if (nd->nd_flag & ND_NFSV2) {
1021 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1022 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1026 NFSEXITCODE2(0, nd);
1030 free(pathcp, M_TEMP);
1031 NFSEXITCODE2(error, nd);
1036 * Remove a non-directory object.
1039 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1040 struct thread *p, struct nfsexstuff *exp)
1046 if (vp->v_type == VDIR)
1047 error = NFSERR_ISDIR;
1049 error = nfsrv_checkremove(vp, 1, p);
1051 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1052 if (ndp->ni_dvp == vp)
1057 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1058 nfsvno_relpathbuf(ndp);
1064 * Remove a directory.
1067 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1068 struct thread *p, struct nfsexstuff *exp)
1074 if (vp->v_type != VDIR) {
1079 * No rmdir "." please.
1081 if (ndp->ni_dvp == vp) {
1086 * The root of a mounted filesystem cannot be deleted.
1088 if (vp->v_vflag & VV_ROOT)
1092 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1093 if (ndp->ni_dvp == vp)
1098 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1099 nfsvno_relpathbuf(ndp);
1108 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1109 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1111 struct vnode *fvp, *tvp, *tdvp;
1114 fvp = fromndp->ni_vp;
1116 vrele(fromndp->ni_dvp);
1121 tdvp = tondp->ni_dvp;
1124 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1125 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1127 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1128 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1131 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1132 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1137 * A rename to '.' or '..' results in a prematurely
1138 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1141 if ((tondp->ni_cnd.cn_namelen == 1 &&
1142 tondp->ni_cnd.cn_nameptr[0] == '.') ||
1143 (tondp->ni_cnd.cn_namelen == 2 &&
1144 tondp->ni_cnd.cn_nameptr[0] == '.' &&
1145 tondp->ni_cnd.cn_nameptr[1] == '.')) {
1150 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1151 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1154 if (fvp->v_mount != tdvp->v_mount) {
1155 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1159 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1164 * If source and destination are the same, there is nothing to
1165 * do. Set error to -1 to indicate this.
1170 if (ndflag & ND_NFSV4) {
1171 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1172 error = nfsrv_checkremove(fvp, 0, p);
1173 NFSVOPUNLOCK(fvp, 0);
1177 error = nfsrv_checkremove(tvp, 1, p);
1180 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1181 * that the NFSv4 client won't be confused by the rename.
1182 * Since nfsd_recalldelegation() can only be called on an
1183 * unlocked vnode at this point and fvp is the file that will
1184 * still exist after the rename, just do fvp.
1186 nfsd_recalldelegation(fvp, p);
1190 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1191 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1200 vrele(fromndp->ni_dvp);
1205 vrele(tondp->ni_startdir);
1206 nfsvno_relpathbuf(tondp);
1208 vrele(fromndp->ni_startdir);
1209 nfsvno_relpathbuf(fromndp);
1218 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1219 struct thread *p, struct nfsexstuff *exp)
1229 if (vp->v_mount != xp->v_mount)
1233 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1234 if ((vp->v_iflag & VI_DOOMED) == 0)
1235 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1238 if (ndp->ni_dvp == vp)
1242 NFSVOPUNLOCK(vp, 0);
1244 if (ndp->ni_dvp == ndp->ni_vp)
1251 nfsvno_relpathbuf(ndp);
1257 * Do the fsync() appropriate for the commit.
1260 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1266 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1267 * file is done. At this time VOP_FSYNC does not accept offset and
1268 * byte count parameters so call VOP_FSYNC the whole file for now.
1269 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1271 if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
1273 * Give up and do the whole thing
1276 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1277 VM_OBJECT_LOCK(vp->v_object);
1278 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1279 VM_OBJECT_UNLOCK(vp->v_object);
1281 error = VOP_FSYNC(vp, MNT_WAIT, td);
1284 * Locate and synchronously write any buffers that fall
1285 * into the requested range. Note: we are assuming that
1286 * f_iosize is a power of 2.
1288 int iosize = vp->v_mount->mnt_stat.f_iosize;
1289 int iomask = iosize - 1;
1294 * Align to iosize boundry, super-align to page boundry.
1297 cnt += off & iomask;
1298 off &= ~(u_quad_t)iomask;
1300 if (off & PAGE_MASK) {
1301 cnt += off & PAGE_MASK;
1302 off &= ~(u_quad_t)PAGE_MASK;
1304 lblkno = off / iosize;
1307 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1308 VM_OBJECT_LOCK(vp->v_object);
1309 vm_object_page_clean(vp->v_object, off, off + cnt,
1311 VM_OBJECT_UNLOCK(vp->v_object);
1320 * If we have a buffer and it is marked B_DELWRI we
1321 * have to lock and write it. Otherwise the prior
1322 * write is assumed to have already been committed.
1324 * gbincore() can return invalid buffers now so we
1325 * have to check that bit as well (though B_DELWRI
1326 * should not be set if B_INVAL is set there could be
1327 * a race here since we haven't locked the buffer).
1329 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1330 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1331 LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
1333 continue; /* retry */
1335 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1338 bp->b_flags &= ~B_ASYNC;
1361 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1365 error = VFS_STATFS(vp->v_mount, sf);
1368 * Since NFS handles these values as unsigned on the
1369 * wire, there is no way to represent negative values,
1370 * so set them to 0. Without this, they will appear
1371 * to be very large positive values for clients like
1374 if (sf->f_bavail < 0)
1376 if (sf->f_ffree < 0)
1384 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1385 * must handle nfsrv_opencheck() calls after any other access checks.
1388 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1389 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1390 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1391 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1392 struct nfsexstuff *exp, struct vnode **vpp)
1394 struct vnode *vp = NULL;
1396 struct nfsexstuff nes;
1398 if (ndp->ni_vp == NULL)
1399 nd->nd_repstat = nfsrv_opencheck(clientid,
1400 stateidp, stp, NULL, nd, p, nd->nd_repstat);
1401 if (!nd->nd_repstat) {
1402 if (ndp->ni_vp == NULL) {
1403 vrele(ndp->ni_startdir);
1404 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1405 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1407 nfsvno_relpathbuf(ndp);
1408 if (!nd->nd_repstat) {
1409 if (*exclusive_flagp) {
1410 *exclusive_flagp = 0;
1411 NFSVNO_ATTRINIT(nvap);
1412 nvap->na_atime.tv_sec = cverf[0];
1413 nvap->na_atime.tv_nsec = cverf[1];
1414 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1415 &nvap->na_vattr, cred);
1417 nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1418 aclp, p, attrbitp, exp);
1423 if (ndp->ni_startdir)
1424 vrele(ndp->ni_startdir);
1425 nfsvno_relpathbuf(ndp);
1427 if (create == NFSV4OPEN_CREATE) {
1428 if (ndp->ni_dvp == vp)
1433 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1434 if (ndp->ni_cnd.cn_flags & RDONLY)
1435 NFSVNO_SETEXRDONLY(&nes);
1437 NFSVNO_EXINIT(&nes);
1438 nd->nd_repstat = nfsvno_accchk(vp,
1439 VWRITE, cred, &nes, p,
1440 NFSACCCHK_NOOVERRIDE,
1441 NFSACCCHK_VPISLOCKED, NULL);
1442 nd->nd_repstat = nfsrv_opencheck(clientid,
1443 stateidp, stp, vp, nd, p, nd->nd_repstat);
1444 if (!nd->nd_repstat) {
1445 tempsize = nvap->na_size;
1446 NFSVNO_ATTRINIT(nvap);
1447 nvap->na_size = tempsize;
1448 nd->nd_repstat = VOP_SETATTR(vp,
1449 &nvap->na_vattr, cred);
1451 } else if (vp->v_type == VREG) {
1452 nd->nd_repstat = nfsrv_opencheck(clientid,
1453 stateidp, stp, vp, nd, p, nd->nd_repstat);
1457 if (ndp->ni_cnd.cn_flags & HASBUF)
1458 nfsvno_relpathbuf(ndp);
1459 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1460 vrele(ndp->ni_startdir);
1461 if (ndp->ni_dvp == ndp->ni_vp)
1471 NFSEXITCODE2(0, nd);
1475 * Updates the file rev and sets the mtime and ctime
1476 * to the current clock time, returning the va_filerev and va_Xtime
1478 * Return ESTALE to indicate the vnode is VI_DOOMED.
1481 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1482 struct ucred *cred, struct thread *p)
1487 vfs_timestamp(&va.va_mtime);
1488 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1489 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1490 if ((vp->v_iflag & VI_DOOMED) != 0)
1493 (void) VOP_SETATTR(vp, &va, cred);
1494 (void) nfsvno_getattr(vp, nvap, cred, p, 1);
1499 * Glue routine to nfsv4_fillattr().
1502 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1503 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1504 struct ucred *cred, struct thread *p, int isdgram, int reterr,
1505 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1509 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1510 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1512 NFSEXITCODE2(0, nd);
1516 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1518 * nfs readdir service
1519 * - mallocs what it thinks is enough to read
1520 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1521 * - calls VOP_READDIR()
1522 * - loops around building the reply
1523 * if the output generated exceeds count break out of loop
1524 * The NFSM_CLGET macro is used here so that the reply will be packed
1525 * tightly in mbuf clusters.
1526 * - it trims out records with d_fileno == 0
1527 * this doesn't matter for Unix clients, but they might confuse clients
1529 * - it trims out records with d_type == DT_WHT
1530 * these cannot be seen through NFS (unless we extend the protocol)
1531 * The alternate call nfsrvd_readdirplus() does lookups as well.
1532 * PS: The NFS protocol spec. does not clarify what the "count" byte
1533 * argument is a count of.. just name strings and file id's or the
1534 * entire reply rpc or ...
1535 * I tried just file name and id sizes and it confused the Sun client,
1536 * so I am using the full rpc size now. The "paranoia.." comment refers
1537 * to including the status longwords that are not a part of the dir.
1538 * "entry" structures, but are in the rpc.
1541 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1542 struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1547 char *cpos, *cend, *rbuf;
1549 int nlen, error = 0, getret = 1;
1550 int siz, cnt, fullsiz, eofflag, ncookies;
1551 u_int64_t off, toff, verf;
1552 u_long *cookies = NULL, *cookiep;
1557 if (nd->nd_repstat) {
1558 nfsrv_postopattr(nd, getret, &at);
1561 if (nd->nd_flag & ND_NFSV2) {
1562 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1563 off = fxdr_unsigned(u_quad_t, *tl++);
1565 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1566 off = fxdr_hyper(tl);
1568 verf = fxdr_hyper(tl);
1572 cnt = fxdr_unsigned(int, *tl);
1573 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1574 cnt = NFS_SRVMAXDATA(nd);
1575 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1577 if (nd->nd_flag & ND_NFSV3) {
1578 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1582 * va_filerev is not sufficient as a cookie verifier,
1583 * since it is not supposed to change when entries are
1584 * removed/added unless that offset cookies returned to
1585 * the client are no longer valid.
1587 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1588 nd->nd_repstat = NFSERR_BAD_COOKIE;
1591 if (!nd->nd_repstat && vp->v_type != VDIR)
1592 nd->nd_repstat = NFSERR_NOTDIR;
1593 if (nd->nd_repstat == 0 && cnt == 0) {
1594 if (nd->nd_flag & ND_NFSV2)
1595 /* NFSv2 does not have NFSERR_TOOSMALL */
1596 nd->nd_repstat = EPERM;
1598 nd->nd_repstat = NFSERR_TOOSMALL;
1600 if (!nd->nd_repstat)
1601 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1602 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1603 NFSACCCHK_VPISLOCKED, NULL);
1604 if (nd->nd_repstat) {
1606 if (nd->nd_flag & ND_NFSV3)
1607 nfsrv_postopattr(nd, getret, &at);
1610 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1611 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1615 free((caddr_t)cookies, M_TEMP);
1623 io.uio_offset = (off_t)off;
1625 io.uio_segflg = UIO_SYSSPACE;
1626 io.uio_rw = UIO_READ;
1628 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1630 off = (u_int64_t)io.uio_offset;
1632 siz -= io.uio_resid;
1634 if (!cookies && !nd->nd_repstat)
1635 nd->nd_repstat = NFSERR_PERM;
1636 if (nd->nd_flag & ND_NFSV3) {
1637 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1638 if (!nd->nd_repstat)
1639 nd->nd_repstat = getret;
1643 * Handles the failed cases. nd->nd_repstat == 0 past here.
1645 if (nd->nd_repstat) {
1647 free((caddr_t)rbuf, M_TEMP);
1649 free((caddr_t)cookies, M_TEMP);
1650 if (nd->nd_flag & ND_NFSV3)
1651 nfsrv_postopattr(nd, getret, &at);
1655 * If nothing read, return eof
1660 if (nd->nd_flag & ND_NFSV2) {
1661 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1663 nfsrv_postopattr(nd, getret, &at);
1664 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1665 txdr_hyper(at.na_filerev, tl);
1668 *tl++ = newnfs_false;
1670 FREE((caddr_t)rbuf, M_TEMP);
1671 FREE((caddr_t)cookies, M_TEMP);
1676 * Check for degenerate cases of nothing useful read.
1677 * If so go try again
1681 dp = (struct dirent *)cpos;
1685 * For some reason FreeBSD's ufs_readdir() chooses to back the
1686 * directory offset up to a block boundary, so it is necessary to
1687 * skip over the records that precede the requested offset. This
1688 * requires the assumption that file offset cookies monotonically
1690 * Since the offset cookies don't monotonically increase for ZFS,
1691 * this is not done when ZFS is the file system.
1693 while (cpos < cend && ncookies > 0 &&
1694 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1695 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1696 cpos += dp->d_reclen;
1697 dp = (struct dirent *)cpos;
1701 if (cpos >= cend || ncookies == 0) {
1709 * dirlen is the size of the reply, including all XDR and must
1710 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1711 * if the XDR should be included in "count", but to be safe, we do.
1712 * (Include the two booleans at the end of the reply in dirlen now.)
1714 if (nd->nd_flag & ND_NFSV3) {
1715 nfsrv_postopattr(nd, getret, &at);
1716 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1717 txdr_hyper(at.na_filerev, tl);
1718 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1720 dirlen = 2 * NFSX_UNSIGNED;
1723 /* Loop through the records and build reply */
1724 while (cpos < cend && ncookies > 0) {
1725 nlen = dp->d_namlen;
1726 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1727 nlen <= NFS_MAXNAMLEN) {
1728 if (nd->nd_flag & ND_NFSV3)
1729 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1731 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1738 * Build the directory record xdr from
1741 if (nd->nd_flag & ND_NFSV3) {
1742 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1743 *tl++ = newnfs_true;
1746 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1747 *tl++ = newnfs_true;
1749 *tl = txdr_unsigned(dp->d_fileno);
1750 (void) nfsm_strtom(nd, dp->d_name, nlen);
1751 if (nd->nd_flag & ND_NFSV3) {
1752 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1755 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1756 *tl = txdr_unsigned(*cookiep);
1758 cpos += dp->d_reclen;
1759 dp = (struct dirent *)cpos;
1765 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1766 *tl++ = newnfs_false;
1771 FREE((caddr_t)rbuf, M_TEMP);
1772 FREE((caddr_t)cookies, M_TEMP);
1775 NFSEXITCODE2(0, nd);
1779 NFSEXITCODE2(error, nd);
1784 * Readdirplus for V3 and Readdir for V4.
1787 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1788 struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1793 char *cpos, *cend, *rbuf;
1796 struct nfsvattr nva, at, *nvap = &nva;
1797 struct mbuf *mb0, *mb1;
1798 struct nfsreferral *refp;
1799 int nlen, r, error = 0, getret = 1, usevget = 1;
1800 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1801 caddr_t bpos0, bpos1;
1802 u_int64_t off, toff, verf;
1803 u_long *cookies = NULL, *cookiep;
1804 nfsattrbit_t attrbits, rderrbits, savbits;
1807 struct componentname cn;
1808 int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1809 struct mount *mp, *new_mp;
1810 uint64_t mounted_on_fileno;
1812 if (nd->nd_repstat) {
1813 nfsrv_postopattr(nd, getret, &at);
1816 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1817 off = fxdr_hyper(tl);
1820 verf = fxdr_hyper(tl);
1822 siz = fxdr_unsigned(int, *tl++);
1823 cnt = fxdr_unsigned(int, *tl);
1826 * Use the server's maximum data transfer size as the upper bound
1829 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1830 cnt = NFS_SRVMAXDATA(nd);
1833 * siz is a "hint" of how much directory information (name, fileid,
1834 * cookie) should be in the reply. At least one client "hints" 0,
1835 * so I set it to cnt for that case. I also round it up to the
1836 * next multiple of DIRBLKSIZ.
1840 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1842 if (nd->nd_flag & ND_NFSV4) {
1843 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1846 NFSSET_ATTRBIT(&savbits, &attrbits);
1847 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1848 NFSZERO_ATTRBIT(&rderrbits);
1849 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1851 NFSZERO_ATTRBIT(&attrbits);
1854 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1855 if (!nd->nd_repstat) {
1856 if (off && verf != at.na_filerev) {
1858 * va_filerev is not sufficient as a cookie verifier,
1859 * since it is not supposed to change when entries are
1860 * removed/added unless that offset cookies returned to
1861 * the client are no longer valid.
1864 if (nd->nd_flag & ND_NFSV4) {
1865 nd->nd_repstat = NFSERR_NOTSAME;
1867 nd->nd_repstat = NFSERR_BAD_COOKIE;
1870 } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1871 nd->nd_repstat = NFSERR_BAD_COOKIE;
1874 if (!nd->nd_repstat && vp->v_type != VDIR)
1875 nd->nd_repstat = NFSERR_NOTDIR;
1876 if (!nd->nd_repstat && cnt == 0)
1877 nd->nd_repstat = NFSERR_TOOSMALL;
1878 if (!nd->nd_repstat)
1879 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1880 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1881 NFSACCCHK_VPISLOCKED, NULL);
1882 if (nd->nd_repstat) {
1884 if (nd->nd_flag & ND_NFSV3)
1885 nfsrv_postopattr(nd, getret, &at);
1888 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1890 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1894 free((caddr_t)cookies, M_TEMP);
1902 io.uio_offset = (off_t)off;
1904 io.uio_segflg = UIO_SYSSPACE;
1905 io.uio_rw = UIO_READ;
1907 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1909 off = (u_int64_t)io.uio_offset;
1911 siz -= io.uio_resid;
1913 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1915 if (!cookies && !nd->nd_repstat)
1916 nd->nd_repstat = NFSERR_PERM;
1917 if (!nd->nd_repstat)
1918 nd->nd_repstat = getret;
1919 if (nd->nd_repstat) {
1922 free((caddr_t)cookies, M_TEMP);
1923 free((caddr_t)rbuf, M_TEMP);
1924 if (nd->nd_flag & ND_NFSV3)
1925 nfsrv_postopattr(nd, getret, &at);
1929 * If nothing read, return eof
1934 if (nd->nd_flag & ND_NFSV3)
1935 nfsrv_postopattr(nd, getret, &at);
1936 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1937 txdr_hyper(at.na_filerev, tl);
1939 *tl++ = newnfs_false;
1941 free((caddr_t)cookies, M_TEMP);
1942 free((caddr_t)rbuf, M_TEMP);
1947 * Check for degenerate cases of nothing useful read.
1948 * If so go try again
1952 dp = (struct dirent *)cpos;
1956 * For some reason FreeBSD's ufs_readdir() chooses to back the
1957 * directory offset up to a block boundary, so it is necessary to
1958 * skip over the records that precede the requested offset. This
1959 * requires the assumption that file offset cookies monotonically
1961 * Since the offset cookies don't monotonically increase for ZFS,
1962 * this is not done when ZFS is the file system.
1964 while (cpos < cend && ncookies > 0 &&
1965 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1966 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1967 ((nd->nd_flag & ND_NFSV4) &&
1968 ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1969 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1970 cpos += dp->d_reclen;
1971 dp = (struct dirent *)cpos;
1975 if (cpos >= cend || ncookies == 0) {
1982 * Busy the file system so that the mount point won't go away
1983 * and, as such, VFS_VGET() can be used safely.
1987 NFSVOPUNLOCK(vp, 0);
1988 nd->nd_repstat = vfs_busy(mp, 0);
1990 if (nd->nd_repstat != 0) {
1992 free(cookies, M_TEMP);
1994 if (nd->nd_flag & ND_NFSV3)
1995 nfsrv_postopattr(nd, getret, &at);
2000 * Check to see if entries in this directory can be safely acquired
2001 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2002 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2003 * automount of the snapshot directory that is required will
2005 * This needs to be done here for NFSv4, since NFSv4 never does
2006 * a VFS_VGET() for "." or "..".
2009 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2010 if (r == EOPNOTSUPP) {
2012 cn.cn_nameiop = LOOKUP;
2013 cn.cn_lkflags = LK_SHARED | LK_RETRY;
2014 cn.cn_cred = nd->nd_cred;
2021 * Save this position, in case there is an error before one entry
2025 bpos0 = nd->nd_bpos;
2028 * Fill in the first part of the reply.
2029 * dirlen is the reply length in bytes and cannot exceed cnt.
2030 * (Include the two booleans at the end of the reply in dirlen now,
2031 * so we recognize when we have exceeded cnt.)
2033 if (nd->nd_flag & ND_NFSV3) {
2034 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2035 nfsrv_postopattr(nd, getret, &at);
2037 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2039 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2040 txdr_hyper(at.na_filerev, tl);
2043 * Save this position, in case there is an empty reply needed.
2046 bpos1 = nd->nd_bpos;
2048 /* Loop through the records and build reply */
2050 while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2051 nlen = dp->d_namlen;
2052 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2053 nlen <= NFS_MAXNAMLEN &&
2054 ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2055 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2056 || (nlen == 1 && dp->d_name[0] != '.'))) {
2058 * Save the current position in the reply, in case
2059 * this entry exceeds cnt.
2062 bpos1 = nd->nd_bpos;
2065 * For readdir_and_lookup get the vnode using
2074 mounted_on_fileno = (uint64_t)dp->d_fileno;
2075 if ((nd->nd_flag & ND_NFSV3) ||
2076 NFSNONZERO_ATTRBIT(&savbits)) {
2077 if (nd->nd_flag & ND_NFSV4)
2078 refp = nfsv4root_getreferral(NULL,
2082 r = VFS_VGET(mp, dp->d_fileno,
2086 if (r == EOPNOTSUPP) {
2089 cn.cn_nameiop = LOOKUP;
2097 cn.cn_nameptr = dp->d_name;
2098 cn.cn_namelen = nlen;
2099 cn.cn_flags = ISLASTCN |
2100 NOFOLLOW | LOCKLEAF |
2103 dp->d_name[0] == '.' &&
2104 dp->d_name[1] == '.')
2107 if (NFSVOPLOCK(vp, LK_SHARED)
2109 nd->nd_repstat = EPERM;
2112 if ((vp->v_vflag & VV_ROOT) != 0
2113 && (cn.cn_flags & ISDOTDOT)
2119 r = VOP_LOOKUP(vp, &nvp,
2128 * For NFSv4, check to see if nvp is
2129 * a mount point and get the mount
2130 * point vnode, as required.
2133 nfsrv_enable_crossmntpt != 0 &&
2134 (nd->nd_flag & ND_NFSV4) != 0 &&
2135 nvp->v_type == VDIR &&
2136 nvp->v_mountedhere != NULL) {
2137 new_mp = nvp->v_mountedhere;
2138 r = vfs_busy(new_mp, 0);
2142 r = VFS_ROOT(new_mp,
2152 ((nd->nd_flag & ND_NFSV3) ||
2153 NFSNONZERO_ATTRBIT(&attrbits))) {
2154 r = nfsvno_getfh(nvp, &nfh, p);
2156 r = nfsvno_getattr(nvp, nvap,
2158 if (r == 0 && not_zfs == 0 &&
2159 nfsrv_enable_crossmntpt != 0 &&
2160 (nd->nd_flag & ND_NFSV4) != 0 &&
2161 nvp->v_type == VDIR &&
2162 vp->v_mount != nvp->v_mount) {
2164 * For a ZFS snapshot, there is a
2165 * pseudo mount that does not set
2166 * v_mountedhere, so it needs to
2167 * be detected via a different
2172 new_mp = nvp->v_mount;
2179 if (!NFSISSET_ATTRBIT(&attrbits,
2180 NFSATTRBIT_RDATTRERROR)) {
2183 if (needs_unbusy != 0)
2192 * Build the directory record xdr
2194 if (nd->nd_flag & ND_NFSV3) {
2195 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2196 *tl++ = newnfs_true;
2198 *tl = txdr_unsigned(dp->d_fileno);
2199 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2200 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2202 *tl = txdr_unsigned(*cookiep);
2203 nfsrv_postopattr(nd, 0, nvap);
2204 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2205 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2209 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2210 *tl++ = newnfs_true;
2212 *tl = txdr_unsigned(*cookiep);
2213 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2215 supports_nfsv4acls =
2216 nfs_supportsnfsv4acls(nvp);
2217 NFSVOPUNLOCK(nvp, 0);
2219 supports_nfsv4acls = 0;
2221 dirlen += nfsrv_putreferralattr(nd,
2224 if (nd->nd_repstat) {
2227 if (needs_unbusy != 0)
2232 dirlen += nfsvno_fillattr(nd, new_mp,
2233 nvp, nvap, &nfh, r, &rderrbits,
2234 nd->nd_cred, p, isdgram, 0,
2235 supports_nfsv4acls, at_root,
2238 dirlen += nfsvno_fillattr(nd, new_mp,
2239 nvp, nvap, &nfh, r, &attrbits,
2240 nd->nd_cred, p, isdgram, 0,
2241 supports_nfsv4acls, at_root,
2246 dirlen += (3 * NFSX_UNSIGNED);
2248 if (needs_unbusy != 0)
2253 cpos += dp->d_reclen;
2254 dp = (struct dirent *)cpos;
2262 * If dirlen > cnt, we must strip off the last entry. If that
2263 * results in an empty reply, report NFSERR_TOOSMALL.
2265 if (dirlen > cnt || nd->nd_repstat) {
2266 if (!nd->nd_repstat && entrycnt == 0)
2267 nd->nd_repstat = NFSERR_TOOSMALL;
2269 newnfs_trimtrailing(nd, mb0, bpos0);
2271 newnfs_trimtrailing(nd, mb1, bpos1);
2273 } else if (cpos < cend)
2275 if (!nd->nd_repstat) {
2276 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2277 *tl++ = newnfs_false;
2283 FREE((caddr_t)cookies, M_TEMP);
2284 FREE((caddr_t)rbuf, M_TEMP);
2287 NFSEXITCODE2(0, nd);
2291 NFSEXITCODE2(error, nd);
2296 * Get the settable attributes out of the mbuf list.
2297 * (Return 0 or EBADRPC)
2300 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2301 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2304 struct nfsv2_sattr *sp;
2305 int error = 0, toclient = 0;
2307 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2309 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2311 * Some old clients didn't fill in the high order 16bits.
2312 * --> check the low order 2 bytes for 0xffff
2314 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2315 nvap->na_mode = nfstov_mode(sp->sa_mode);
2316 if (sp->sa_uid != newnfs_xdrneg1)
2317 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2318 if (sp->sa_gid != newnfs_xdrneg1)
2319 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2320 if (sp->sa_size != newnfs_xdrneg1)
2321 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2322 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2324 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2326 nvap->na_atime.tv_sec =
2327 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2328 nvap->na_atime.tv_nsec = 0;
2331 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2332 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2335 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2336 if (*tl == newnfs_true) {
2337 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2338 nvap->na_mode = nfstov_mode(*tl);
2340 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2341 if (*tl == newnfs_true) {
2342 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2343 nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2345 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2346 if (*tl == newnfs_true) {
2347 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2348 nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2350 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2351 if (*tl == newnfs_true) {
2352 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2353 nvap->na_size = fxdr_hyper(tl);
2355 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2356 switch (fxdr_unsigned(int, *tl)) {
2357 case NFSV3SATTRTIME_TOCLIENT:
2358 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2359 fxdr_nfsv3time(tl, &nvap->na_atime);
2362 case NFSV3SATTRTIME_TOSERVER:
2363 vfs_timestamp(&nvap->na_atime);
2364 nvap->na_vaflags |= VA_UTIMES_NULL;
2367 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2368 switch (fxdr_unsigned(int, *tl)) {
2369 case NFSV3SATTRTIME_TOCLIENT:
2370 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2371 fxdr_nfsv3time(tl, &nvap->na_mtime);
2372 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2374 case NFSV3SATTRTIME_TOSERVER:
2375 vfs_timestamp(&nvap->na_mtime);
2377 nvap->na_vaflags |= VA_UTIMES_NULL;
2382 error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2385 NFSEXITCODE2(error, nd);
2390 * Handle the setable attributes for V4.
2391 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2394 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2395 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2400 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2402 u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2406 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2409 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2410 attrsize = fxdr_unsigned(int, *tl);
2413 * Loop around getting the setable attributes. If an unsupported
2414 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2417 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2418 bitpos = NFSATTRBIT_MAX;
2422 for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2423 if (attrsum > attrsize) {
2424 error = NFSERR_BADXDR;
2427 if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2429 case NFSATTRBIT_SIZE:
2430 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2431 nvap->na_size = fxdr_hyper(tl);
2432 attrsum += NFSX_HYPER;
2434 case NFSATTRBIT_ACL:
2435 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2439 if (aceerr && !nd->nd_repstat)
2440 nd->nd_repstat = aceerr;
2443 case NFSATTRBIT_ARCHIVE:
2444 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2445 if (!nd->nd_repstat)
2446 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2447 attrsum += NFSX_UNSIGNED;
2449 case NFSATTRBIT_HIDDEN:
2450 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2451 if (!nd->nd_repstat)
2452 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2453 attrsum += NFSX_UNSIGNED;
2455 case NFSATTRBIT_MIMETYPE:
2456 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2457 i = fxdr_unsigned(int, *tl);
2458 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2461 if (!nd->nd_repstat)
2462 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2463 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2465 case NFSATTRBIT_MODE:
2466 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2467 nvap->na_mode = nfstov_mode(*tl);
2468 attrsum += NFSX_UNSIGNED;
2470 case NFSATTRBIT_OWNER:
2471 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2472 j = fxdr_unsigned(int, *tl);
2474 error = NFSERR_BADXDR;
2477 if (j > NFSV4_SMALLSTR)
2478 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2481 error = nfsrv_mtostr(nd, cp, j);
2483 if (j > NFSV4_SMALLSTR)
2484 free(cp, M_NFSSTRING);
2487 if (!nd->nd_repstat) {
2488 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2490 if (!nd->nd_repstat)
2493 if (j > NFSV4_SMALLSTR)
2494 free(cp, M_NFSSTRING);
2495 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2497 case NFSATTRBIT_OWNERGROUP:
2498 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2499 j = fxdr_unsigned(int, *tl);
2501 error = NFSERR_BADXDR;
2504 if (j > NFSV4_SMALLSTR)
2505 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2508 error = nfsrv_mtostr(nd, cp, j);
2510 if (j > NFSV4_SMALLSTR)
2511 free(cp, M_NFSSTRING);
2514 if (!nd->nd_repstat) {
2515 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2517 if (!nd->nd_repstat)
2520 if (j > NFSV4_SMALLSTR)
2521 free(cp, M_NFSSTRING);
2522 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2524 case NFSATTRBIT_SYSTEM:
2525 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2526 if (!nd->nd_repstat)
2527 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2528 attrsum += NFSX_UNSIGNED;
2530 case NFSATTRBIT_TIMEACCESSSET:
2531 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2532 attrsum += NFSX_UNSIGNED;
2533 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2534 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2535 fxdr_nfsv4time(tl, &nvap->na_atime);
2537 attrsum += NFSX_V4TIME;
2539 vfs_timestamp(&nvap->na_atime);
2540 nvap->na_vaflags |= VA_UTIMES_NULL;
2543 case NFSATTRBIT_TIMEBACKUP:
2544 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2545 if (!nd->nd_repstat)
2546 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2547 attrsum += NFSX_V4TIME;
2549 case NFSATTRBIT_TIMECREATE:
2550 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2551 if (!nd->nd_repstat)
2552 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2553 attrsum += NFSX_V4TIME;
2555 case NFSATTRBIT_TIMEMODIFYSET:
2556 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2557 attrsum += NFSX_UNSIGNED;
2558 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2559 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2560 fxdr_nfsv4time(tl, &nvap->na_mtime);
2561 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2562 attrsum += NFSX_V4TIME;
2564 vfs_timestamp(&nvap->na_mtime);
2566 nvap->na_vaflags |= VA_UTIMES_NULL;
2570 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2572 * set bitpos so we drop out of the loop.
2574 bitpos = NFSATTRBIT_MAX;
2580 * some clients pad the attrlist, so we need to skip over the
2583 if (attrsum > attrsize) {
2584 error = NFSERR_BADXDR;
2586 attrsize = NFSM_RNDUP(attrsize);
2587 if (attrsum < attrsize)
2588 error = nfsm_advance(nd, attrsize - attrsum, -1);
2591 NFSEXITCODE2(error, nd);
2596 * Check/setup export credentials.
2599 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2600 struct ucred *credanon)
2605 * Check/setup credentials.
2607 if (nd->nd_flag & ND_GSS)
2608 exp->nes_exflag &= ~MNT_EXPORTANON;
2611 * Check to see if the operation is allowed for this security flavor.
2612 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2613 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2614 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2616 if (nfsvno_testexp(nd, exp) &&
2617 nd->nd_procnum != NFSV4OP_SECINFO &&
2618 nd->nd_procnum != NFSPROC_FSINFO) {
2619 if (nd->nd_flag & ND_NFSV4)
2620 error = NFSERR_WRONGSEC;
2622 error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2627 * Check to see if the file system is exported V4 only.
2629 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2630 error = NFSERR_PROGNOTV4;
2635 * Now, map the user credentials.
2636 * (Note that ND_AUTHNONE will only be set for an NFSv3
2637 * Fsinfo RPC. If set for anything else, this code might need
2640 if (NFSVNO_EXPORTED(exp) &&
2641 ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2642 NFSVNO_EXPORTANON(exp) ||
2643 (nd->nd_flag & ND_AUTHNONE))) {
2644 nd->nd_cred->cr_uid = credanon->cr_uid;
2645 nd->nd_cred->cr_gid = credanon->cr_gid;
2646 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2647 credanon->cr_groups);
2651 NFSEXITCODE2(error, nd);
2659 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2660 struct ucred **credp)
2662 int i, error, *secflavors;
2664 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2665 &exp->nes_numsecflavor, &secflavors);
2667 if (nfs_rootfhset) {
2668 exp->nes_exflag = 0;
2669 exp->nes_numsecflavor = 0;
2673 /* Copy the security flavors. */
2674 for (i = 0; i < exp->nes_numsecflavor; i++)
2675 exp->nes_secflavors[i] = secflavors[i];
2682 * Get a vnode for a file handle and export stuff.
2685 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2686 int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2687 struct ucred **credp)
2689 int i, error, *secflavors;
2692 exp->nes_numsecflavor = 0;
2693 if (VFS_NEEDSGIANT(mp))
2696 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2698 /* Make sure the server replies ESTALE to the client. */
2700 if (nam && !error) {
2701 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2702 &exp->nes_numsecflavor, &secflavors);
2704 if (nfs_rootfhset) {
2705 exp->nes_exflag = 0;
2706 exp->nes_numsecflavor = 0;
2712 /* Copy the security flavors. */
2713 for (i = 0; i < exp->nes_numsecflavor; i++)
2714 exp->nes_secflavors[i] = secflavors[i];
2722 * nfsd_fhtovp() - convert a fh to a vnode ptr
2723 * - look up fsid in mount list (if not found ret error)
2724 * - get vp and export rights by calling nfsvno_fhtovp()
2725 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2727 * - if mpp != NULL, return the mount point so that it can
2728 * be used for vn_finished_write() by the caller
2731 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2732 struct vnode **vpp, struct nfsexstuff *exp,
2733 struct mount **mpp, int startwrite, struct thread *p)
2736 struct ucred *credanon;
2739 fhp = (fhandle_t *)nfp->nfsrvfh_data;
2741 * Check for the special case of the nfsv4root_fh.
2743 mp = vfs_busyfs(&fhp->fh_fsid);
2748 nd->nd_repstat = ESTALE;
2753 vn_start_write(NULL, mpp, V_WAIT);
2754 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2755 lktype = LK_EXCLUSIVE;
2757 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2762 * For NFSv4 without a pseudo root fs, unexported file handles
2763 * can be returned, so that Lookup works everywhere.
2765 if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2766 !(nd->nd_flag & ND_NFSV4)) {
2768 nd->nd_repstat = EACCES;
2772 * Personally, I've never seen any point in requiring a
2773 * reserved port#, since only in the rare case where the
2774 * clients are all boxes with secure system priviledges,
2775 * does it provide any enhanced security, but... some people
2776 * believe it to be useful and keep putting this code back in.
2777 * (There is also some "security checker" out there that
2778 * complains if the nfs server doesn't enforce this.)
2779 * However, note the following:
2780 * RFC3530 (NFSv4) specifies that a reserved port# not be
2782 * RFC2623 recommends that, if a reserved port# is checked for,
2783 * that there be a way to turn that off--> ifdef'd.
2785 #ifdef NFS_REQRSVPORT
2786 if (!nd->nd_repstat) {
2787 struct sockaddr_in *saddr;
2788 struct sockaddr_in6 *saddr6;
2790 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2791 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2792 if (!(nd->nd_flag & ND_NFSV4) &&
2793 ((saddr->sin_family == AF_INET &&
2794 ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2795 (saddr6->sin6_family == AF_INET6 &&
2796 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2798 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2801 #endif /* NFS_REQRSVPORT */
2804 * Check/setup credentials.
2806 if (!nd->nd_repstat) {
2807 nd->nd_saveduid = nd->nd_cred->cr_uid;
2808 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2812 if (credanon != NULL)
2814 if (nd->nd_repstat) {
2816 vn_finished_write(mp);
2823 NFSEXITCODE2(0, nd);
2830 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2832 struct filedesc *fdp;
2836 fdp = p->td_proc->p_fd;
2837 if (fd >= fdp->fd_nfiles ||
2838 (fp = fdp->fd_ofiles[fd]) == NULL) {
2850 * Called from nfssvc() to update the exports list. Just call
2851 * vfs_export(). This has to be done, since the v4 root fake fs isn't
2852 * in the mount list.
2855 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2857 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2859 struct nameidata nd;
2862 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2863 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2865 else if (error == 0) {
2866 if (nfsexargp->fspec == NULL) {
2871 * If fspec != NULL, this is the v4root path.
2873 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
2874 nfsexargp->fspec, p);
2875 if ((error = namei(&nd)) != 0)
2877 error = nfsvno_getfh(nd.ni_vp, &fh, p);
2880 nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2881 NFSBCOPY((caddr_t)&fh,
2882 nfs_rootfh.nfsrvfh_data,
2883 sizeof (fhandle_t));
2894 * This function needs to test to see if the system is near its limit
2895 * for memory allocation via malloc() or mget() and return True iff
2896 * either of these resources are near their limit.
2897 * XXX (For now, this is just a stub.)
2899 int nfsrv_testmalloclimit = 0;
2901 nfsrv_mallocmget_limit(void)
2903 static int printmesg = 0;
2904 static int testval = 1;
2906 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2907 if ((printmesg++ % 100) == 0)
2908 printf("nfsd: malloc/mget near limit\n");
2915 * BSD specific initialization of a mount point.
2920 static int inited = 0;
2925 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2926 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2927 TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2928 nfsv4root_mnt.mnt_export = NULL;
2929 TAILQ_INIT(&nfsv4root_opt);
2930 TAILQ_INIT(&nfsv4root_newopt);
2931 nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2932 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2933 nfsv4root_mnt.mnt_nvnodelistsize = 0;
2934 nfsv4root_mnt.mnt_activevnodelistsize = 0;
2938 * Get a vnode for a file handle, without checking exports, etc.
2941 nfsvno_getvp(fhandle_t *fhp)
2947 mp = vfs_busyfs(&fhp->fh_fsid);
2950 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2958 * Do a local VOP_ADVLOCK().
2961 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2962 u_int64_t end, struct thread *td)
2968 if (nfsrv_dolocallocks == 0)
2971 /* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2972 if ((vp->v_iflag & VI_DOOMED) != 0) {
2977 fl.l_whence = SEEK_SET;
2979 fl.l_start = (off_t)first;
2980 if (end == NFS64BITSSET) {
2984 fl.l_len = (off_t)tlen;
2987 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2988 * values for all calls, so that all locks will be held by the
2989 * nfsd server. (The nfsd server handles conflicts between the
2991 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2992 * bytes, so it can't be put in l_sysid.
2994 if (nfsv4_sysid == 0)
2995 nfsv4_sysid = nlm_acquire_next_sysid();
2996 fl.l_pid = (pid_t)0;
2997 fl.l_sysid = (int)nfsv4_sysid;
2999 NFSVOPUNLOCK(vp, 0);
3000 if (ftype == F_UNLCK)
3001 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3002 (F_POSIX | F_REMOTE));
3004 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3005 (F_POSIX | F_REMOTE));
3006 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3014 * Check the nfsv4 root exports.
3017 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3019 struct ucred *credanon;
3020 int exflags, error = 0, numsecflavor, *secflavors, i;
3022 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3023 &credanon, &numsecflavor, &secflavors);
3025 error = NFSERR_PROGUNAVAIL;
3028 if (credanon != NULL)
3030 for (i = 0; i < numsecflavor; i++) {
3031 if (secflavors[i] == AUTH_SYS)
3032 nd->nd_flag |= ND_EXAUTHSYS;
3033 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3034 nd->nd_flag |= ND_EXGSS;
3035 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3036 nd->nd_flag |= ND_EXGSSINTEGRITY;
3037 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3038 nd->nd_flag |= ND_EXGSSPRIVACY;
3047 * Nfs server psuedo system call for the nfsd's
3053 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3056 struct nfsd_addsock_args sockarg;
3057 struct nfsd_nfsd_args nfsdarg;
3060 if (uap->flag & NFSSVC_NFSDADDSOCK) {
3061 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3065 * Since we don't know what rights might be required,
3066 * pretend that we need them all. It is better to be too
3067 * careful than too reckless.
3069 if ((error = fget(td, sockarg.sock, CAP_SOCK_ALL, &fp)) != 0)
3071 if (fp->f_type != DTYPE_SOCKET) {
3076 error = nfsrvd_addsock(fp);
3078 } else if (uap->flag & NFSSVC_NFSDNFSD) {
3079 if (uap->argp == NULL) {
3083 error = copyin(uap->argp, (caddr_t)&nfsdarg,
3087 error = nfsrvd_nfsd(td, &nfsdarg);
3089 error = nfssvc_srvcall(td, uap, td->td_ucred);
3098 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3100 struct nfsex_args export;
3101 struct file *fp = NULL;
3103 struct nfsd_clid adminrevoke;
3104 struct nfsd_dumplist dumplist;
3105 struct nfsd_dumpclients *dumpclients;
3106 struct nfsd_dumplocklist dumplocklist;
3107 struct nfsd_dumplocks *dumplocks;
3108 struct nameidata nd;
3110 int error = EINVAL, igotlock;
3112 static int suspend_nfsd = 0;
3114 if (uap->flag & NFSSVC_PUBLICFH) {
3115 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3116 sizeof (fhandle_t));
3117 error = copyin(uap->argp,
3118 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3121 } else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3122 error = copyin(uap->argp,(caddr_t)&export,
3123 sizeof (struct nfsex_args));
3125 error = nfsrv_v4rootexport(&export, cred, p);
3126 } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3129 } else if (uap->flag & NFSSVC_STABLERESTART) {
3130 error = copyin(uap->argp, (caddr_t)&stablefd,
3133 error = fp_getfvp(p, stablefd, &fp, &vp);
3134 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3136 if (!error && newnfs_numnfsd != 0)
3139 nfsrv_stablefirst.nsf_fp = fp;
3140 nfsrv_setupstable(p);
3142 } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3143 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3144 sizeof (struct nfsd_clid));
3146 error = nfsrv_adminrevoke(&adminrevoke, p);
3147 } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3148 error = copyin(uap->argp, (caddr_t)&dumplist,
3149 sizeof (struct nfsd_dumplist));
3150 if (!error && (dumplist.ndl_size < 1 ||
3151 dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3154 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3155 dumpclients = (struct nfsd_dumpclients *)malloc(len,
3157 nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3158 error = copyout(dumpclients,
3159 CAST_USER_ADDR_T(dumplist.ndl_list), len);
3160 free((caddr_t)dumpclients, M_TEMP);
3162 } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3163 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3164 sizeof (struct nfsd_dumplocklist));
3165 if (!error && (dumplocklist.ndllck_size < 1 ||
3166 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3169 error = nfsrv_lookupfilename(&nd,
3170 dumplocklist.ndllck_fname, p);
3172 len = sizeof (struct nfsd_dumplocks) *
3173 dumplocklist.ndllck_size;
3174 dumplocks = (struct nfsd_dumplocks *)malloc(len,
3176 nfsrv_dumplocks(nd.ni_vp, dumplocks,
3177 dumplocklist.ndllck_size, p);
3179 error = copyout(dumplocks,
3180 CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3181 free((caddr_t)dumplocks, M_TEMP);
3183 } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3186 nfsd_master_pid = procp->p_pid;
3187 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3188 nfsd_master_start = procp->p_stats->p_start;
3189 nfsd_master_proc = procp;
3191 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3192 NFSLOCKV4ROOTMUTEX();
3193 if (suspend_nfsd == 0) {
3194 /* Lock out all nfsd threads */
3196 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3197 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3198 } while (igotlock == 0 && suspend_nfsd == 0);
3201 NFSUNLOCKV4ROOTMUTEX();
3203 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3204 NFSLOCKV4ROOTMUTEX();
3205 if (suspend_nfsd != 0) {
3206 nfsv4_unlock(&nfsd_suspend_lock, 0);
3209 NFSUNLOCKV4ROOTMUTEX();
3219 * Returns 0 if ok, 1 otherwise.
3222 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3227 * This seems odd, but allow the case where the security flavor
3228 * list is empty. This happens when NFSv4 is traversing non-exported
3229 * file systems. Exported file systems should always have a non-empty
3230 * security flavor list.
3232 if (exp->nes_numsecflavor == 0)
3235 for (i = 0; i < exp->nes_numsecflavor; i++) {
3237 * The tests for privacy and integrity must be first,
3238 * since ND_GSS is set for everything but AUTH_SYS.
3240 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3241 (nd->nd_flag & ND_GSSPRIVACY))
3243 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3244 (nd->nd_flag & ND_GSSINTEGRITY))
3246 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3247 (nd->nd_flag & ND_GSS))
3249 if (exp->nes_secflavors[i] == AUTH_SYS &&
3250 (nd->nd_flag & ND_GSS) == 0)
3257 * Calculate a hash value for the fid in a file handle.
3260 nfsrv_hashfh(fhandle_t *fhp)
3264 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3269 * Signal the userland master nfsd to backup the stable restart file.
3272 nfsrv_backupstable(void)
3276 if (nfsd_master_proc != NULL) {
3277 procp = pfind(nfsd_master_pid);
3278 /* Try to make sure it is the correct process. */
3279 if (procp == nfsd_master_proc &&
3280 procp->p_stats->p_start.tv_sec ==
3281 nfsd_master_start.tv_sec &&
3282 procp->p_stats->p_start.tv_usec ==
3283 nfsd_master_start.tv_usec &&
3284 strcmp(procp->p_comm, nfsd_master_comm) == 0)
3285 kern_psignal(procp, SIGUSR2);
3287 nfsd_master_proc = NULL;
3294 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3297 * Called once to initialize data structures...
3300 nfsd_modevent(module_t mod, int type, void *data)
3303 static int loaded = 0;
3310 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3311 snprintf(nfsrchash_table[i].lock_name,
3312 sizeof(nfsrchash_table[i].lock_name), "nfsrc_tcp%d",
3314 mtx_init(&nfsrchash_table[i].mtx,
3315 nfsrchash_table[i].lock_name, NULL, MTX_DEF);
3316 snprintf(nfsrcahash_table[i].lock_name,
3317 sizeof(nfsrcahash_table[i].lock_name), "nfsrc_tcpa%d",
3319 mtx_init(&nfsrcahash_table[i].mtx,
3320 nfsrcahash_table[i].lock_name, NULL, MTX_DEF);
3322 mtx_init(&nfsrc_udpmtx, "nfs_udpcache_mutex", NULL, MTX_DEF);
3323 mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3324 mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3326 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3333 #ifdef VV_DISABLEDELEG
3334 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3335 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3337 nfsd_call_servertimer = nfsrv_servertimer;
3338 nfsd_call_nfsd = nfssvc_nfsd;
3343 if (newnfs_numnfsd != 0) {
3348 #ifdef VV_DISABLEDELEG
3349 vn_deleg_ops.vndeleg_recall = NULL;
3350 vn_deleg_ops.vndeleg_disable = NULL;
3352 nfsd_call_servertimer = NULL;
3353 nfsd_call_nfsd = NULL;
3355 /* Clean out all NFSv4 state. */
3356 nfsrv_throwawayallstate(curthread);
3358 /* Clean the NFS server reply cache */
3359 nfsrvd_cleancache();
3361 /* Free up the krpc server pool. */
3362 if (nfsrvd_pool != NULL)
3363 svcpool_destroy(nfsrvd_pool);
3365 /* and get rid of the locks */
3366 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3367 mtx_destroy(&nfsrchash_table[i].mtx);
3368 mtx_destroy(&nfsrcahash_table[i].mtx);
3370 mtx_destroy(&nfsrc_udpmtx);
3371 mtx_destroy(&nfs_v4root_mutex);
3372 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3373 lockdestroy(&nfsv4root_mnt.mnt_explock);
3385 static moduledata_t nfsd_mod = {
3390 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3392 /* So that loader and kldload(2) can find us, wherever we are.. */
3393 MODULE_VERSION(nfsd, 1);
3394 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3395 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3396 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3397 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3398 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);