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 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)nfs_serv.c 8.8 (Berkeley) 7/31/95
37 * $FreeBSD: src/sys/nfs/nfs_serv.c,v 1.93.2.6 2002/12/29 18:19:53 dillon Exp $
38 * $DragonFly: src/sys/vfs/nfs/nfs_serv.c,v 1.48 2008/09/17 21:44:24 dillon Exp $
42 * nfs version 2 and 3 server calls to vnode ops
43 * - these routines generally have 3 phases
44 * 1 - break down and validate rpc request in mbuf list
45 * 2 - do the vnode ops for the request
46 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
47 * 3 - build the rpc reply in an mbuf list
49 * - do not mix the phases, since the nfsm_?? macros can return failures
50 * on a bad rpc or similar and do not do any vrele() or vput()'s
52 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs
53 * error number iff error != 0 whereas
54 * returning an error from the server function implies a fatal error
55 * such as a badly constructed rpc request that should be dropped without
57 * For Version 3, nfsm_reply() does not return for the error case, since
58 * most version 3 rpcs return more than the status for error cases.
61 * Warning: always pay careful attention to resource cleanup on return
62 * and note that nfsm_*() macros can terminate a procedure on certain
66 #include <sys/param.h>
67 #include <sys/systm.h>
70 #include <sys/nlookup.h>
71 #include <sys/namei.h>
72 #include <sys/unistd.h>
73 #include <sys/vnode.h>
74 #include <sys/mount.h>
75 #include <sys/socket.h>
76 #include <sys/socketvar.h>
77 #include <sys/malloc.h>
79 #include <sys/dirent.h>
81 #include <sys/kernel.h>
82 #include <sys/sysctl.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_zone.h>
88 #include <vm/vm_object.h>
92 #include <sys/thread2.h>
98 #include "nfsm_subs.h"
101 #define nfsdbprintf(info) kprintf info
103 #define nfsdbprintf(info)
106 #define MAX_COMMIT_COUNT (1024 * 1024)
108 #define NUM_HEURISTIC 1017
109 #define NHUSE_INIT 64
111 #define NHUSE_MAX 2048
113 static struct nfsheur {
114 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
115 off_t nh_nextr; /* next offset for sequential detection */
116 int nh_use; /* use count for selection */
117 int nh_seqcount; /* heuristic */
118 } nfsheur[NUM_HEURISTIC];
120 nfstype nfsv3_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK,
123 nfstype nfsv2_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON,
126 extern u_int32_t nfs_xdrneg1;
127 extern u_int32_t nfs_false, nfs_true;
128 extern enum vtype nv3tov_type[8];
129 extern struct nfsstats nfsstats;
131 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
132 int nfsrvw_procrastinate_v3 = 0;
134 static struct timespec nfsver;
136 SYSCTL_DECL(_vfs_nfs);
139 SYSCTL_INT(_vfs_nfs, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0, "");
140 static int nfs_commit_blks;
141 static int nfs_commit_miss;
142 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, "");
143 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, "");
145 static int nfsrv_access (struct mount *, struct vnode *, int,
146 struct ucred *, int, struct thread *, int);
147 static void nfsrvw_coalesce (struct nfsrv_descript *,
148 struct nfsrv_descript *);
151 * nfs v3 access service
154 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
155 struct thread *td, struct mbuf **mrq)
157 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
158 struct sockaddr *nam = nfsd->nd_nam;
159 caddr_t dpos = nfsd->nd_dpos;
160 struct ucred *cred = &nfsd->nd_cr;
161 struct vnode *vp = NULL;
162 struct mount *mp = NULL;
168 int error = 0, rdonly, getret;
170 struct mbuf *mb, *mreq, *mb2;
171 struct vattr vattr, *vap = &vattr;
172 u_long testmode, nfsmode;
174 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
175 fhp = &nfh.fh_generic;
177 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
178 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam, &rdonly,
179 (nfsd->nd_flag & ND_KERBAUTH), TRUE);
181 nfsm_reply(NFSX_UNSIGNED);
182 nfsm_srvpostop_attr(1, NULL);
186 nfsmode = fxdr_unsigned(u_int32_t, *tl);
187 if ((nfsmode & NFSV3ACCESS_READ) &&
188 nfsrv_access(mp, vp, VREAD, cred, rdonly, td, 0))
189 nfsmode &= ~NFSV3ACCESS_READ;
190 if (vp->v_type == VDIR)
191 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
194 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
195 if ((nfsmode & testmode) &&
196 nfsrv_access(mp, vp, VWRITE, cred, rdonly, td, 0))
197 nfsmode &= ~testmode;
198 if (vp->v_type == VDIR)
199 testmode = NFSV3ACCESS_LOOKUP;
201 testmode = NFSV3ACCESS_EXECUTE;
202 if ((nfsmode & testmode) &&
203 nfsrv_access(mp, vp, VEXEC, cred, rdonly, td, 0))
204 nfsmode &= ~testmode;
205 getret = VOP_GETATTR(vp, vap);
208 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
209 nfsm_srvpostop_attr(getret, vap);
210 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
211 *tl = txdr_unsigned(nfsmode);
219 * nfs getattr service
222 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
223 struct thread *td, struct mbuf **mrq)
225 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
226 struct sockaddr *nam = nfsd->nd_nam;
227 caddr_t dpos = nfsd->nd_dpos;
228 struct ucred *cred = &nfsd->nd_cr;
229 struct nfs_fattr *fp;
231 struct vattr *vap = &va;
232 struct vnode *vp = NULL;
233 struct mount *mp = NULL;
239 int error = 0, rdonly;
241 struct mbuf *mb, *mb2, *mreq;
243 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
244 fhp = &nfh.fh_generic;
246 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
247 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
253 error = VOP_GETATTR(vp, vap);
256 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
261 nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
262 nfsm_srvfillattr(vap, fp);
272 * nfs setattr service
275 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
276 struct thread *td, struct mbuf **mrq)
278 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
279 struct sockaddr *nam = nfsd->nd_nam;
280 caddr_t dpos = nfsd->nd_dpos;
281 struct ucred *cred = &nfsd->nd_cr;
282 struct vattr va, preat;
283 struct vattr *vap = &va;
284 struct nfsv2_sattr *sp;
285 struct nfs_fattr *fp;
286 struct vnode *vp = NULL;
287 struct mount *mp = NULL;
293 int error = 0, rdonly, preat_ret = 1, postat_ret = 1;
294 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
296 struct mbuf *mb, *mb2, *mreq;
297 struct timespec guard;
299 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
300 fhp = &nfh.fh_generic;
305 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
306 gcheck = fxdr_unsigned(int, *tl);
308 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
309 fxdr_nfsv3time(tl, &guard);
312 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
314 * Nah nah nah nah na nah
315 * There is a bug in the Sun client that puts 0xffff in the mode
316 * field of sattr when it should put in 0xffffffff. The u_short
317 * doesn't sign extend.
318 * --> check the low order 2 bytes for 0xffff
320 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
321 vap->va_mode = nfstov_mode(sp->sa_mode);
322 if (sp->sa_uid != nfs_xdrneg1)
323 vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
324 if (sp->sa_gid != nfs_xdrneg1)
325 vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
326 if (sp->sa_size != nfs_xdrneg1)
327 vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
328 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
330 fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
332 vap->va_atime.tv_sec =
333 fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
334 vap->va_atime.tv_nsec = 0;
337 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
338 fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
343 * Now that we have all the fields, lets do it.
345 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam, &rdonly,
346 (nfsd->nd_flag & ND_KERBAUTH), TRUE);
348 nfsm_reply(2 * NFSX_UNSIGNED);
349 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
355 * vp now an active resource, pay careful attention to cleanup
359 error = preat_ret = VOP_GETATTR(vp, &preat);
360 if (!error && gcheck &&
361 (preat.va_ctime.tv_sec != guard.tv_sec ||
362 preat.va_ctime.tv_nsec != guard.tv_nsec))
363 error = NFSERR_NOT_SYNC;
367 nfsm_reply(NFSX_WCCDATA(v3));
368 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
375 * If the size is being changed write acces is required, otherwise
376 * just check for a read only file system.
378 if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
379 if (rdonly || (mp->mnt_flag & MNT_RDONLY)) {
384 if (vp->v_type == VDIR) {
387 } else if ((error = nfsrv_access(mp, vp, VWRITE, cred, rdonly,
392 error = VOP_SETATTR(vp, vap, cred);
393 postat_ret = VOP_GETATTR(vp, vap);
399 nfsm_reply(NFSX_WCCORFATTR(v3));
401 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
405 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
406 nfsm_srvfillattr(vap, fp);
420 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
421 struct thread *td, struct mbuf **mrq)
423 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
424 struct sockaddr *nam = nfsd->nd_nam;
425 caddr_t dpos = nfsd->nd_dpos;
426 struct ucred *cred = &nfsd->nd_cr;
427 struct nfs_fattr *fp;
428 struct nlookupdata nd;
438 int error = 0, len, dirattr_ret = 1;
439 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
441 struct mbuf *mb, *mb2, *mreq;
442 struct vattr va, dirattr, *vap = &va;
444 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
449 fhp = &nfh.fh_generic;
451 nfsm_srvnamesiz(len);
453 pubflag = nfs_ispublicfh(fhp);
455 error = nfs_namei(&nd, cred, 0, NULL, &vp,
456 fhp, len, slp, nam, &md, &dpos,
457 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
460 * namei failure, only dirp to cleanup. Clear out garbarge from
461 * structure in case macros jump to nfsmout.
467 dirattr_ret = VOP_GETATTR(dirp, &dirattr);
471 nfsm_reply(NFSX_POSTOPATTR(v3));
472 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
478 * Locate index file for public filehandle
480 * error is 0 on entry and 0 on exit from this block.
484 if (vp->v_type == VDIR && nfs_pub.np_index != NULL) {
486 * Setup call to lookup() to see if we can find
487 * the index file. Arguably, this doesn't belong
488 * in a kernel.. Ugh. If an error occurs, do not
489 * try to install an index file and then clear the
492 * When we replace nd with ind and redirect ndp,
493 * maintenance of ni_startdir and ni_vp shift to
494 * ind and we have to clean them up in the old nd.
495 * However, the cnd resource continues to be maintained
496 * via the original nd. Confused? You aren't alone!
499 cache_copy(&nd.nl_nch, &nch);
501 error = nlookup_init_raw(&nd, nfs_pub.np_index,
502 UIO_SYSSPACE, 0, cred, &nch);
505 error = nlookup(&nd);
509 * Found an index file. Get rid of
510 * the old references. transfer vp and
511 * load up the new vp. Fortunately we do
512 * not have to deal with dvp, that would be
519 error = cache_vget(&nd.nl_nch, nd.nl_cred,
521 KKASSERT(error == 0);
526 * If the public filehandle was used, check that this lookup
527 * didn't result in a filehandle outside the publicly exported
528 * filesystem. We clear the poor vp here to avoid lockups due
532 if (vp->v_mount != nfs_pub.np_mount) {
541 dirattr_ret = VOP_GETATTR(dirp, &dirattr);
547 * Resources at this point:
548 * ndp->ni_vp may not be NULL
553 nfsm_reply(NFSX_POSTOPATTR(v3));
554 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
560 * Clear out some resources prior to potentially blocking. This
561 * is not as critical as ni_dvp resources in other routines, but
567 * Get underlying attribute, then release remaining resources ( for
568 * the same potential blocking reason ) and reply.
570 bzero(&fhp->fh_fid, sizeof(fhp->fh_fid));
571 error = VFS_VPTOFH(vp, &fhp->fh_fid);
573 error = VOP_GETATTR(vp, vap);
577 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
579 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
583 nfsm_srvfhtom(fhp, v3);
585 nfsm_srvpostop_attr(0, vap);
586 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
588 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
589 nfsm_srvfillattr(vap, fp);
595 nlookup_done(&nd); /* may be called twice */
602 * nfs readlink service
605 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
606 struct thread *td, struct mbuf **mrq)
608 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
609 struct sockaddr *nam = nfsd->nd_nam;
610 caddr_t dpos = nfsd->nd_dpos;
611 struct ucred *cred = &nfsd->nd_cr;
612 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
613 struct iovec *ivp = iv;
617 int error = 0, rdonly, i, tlen, len, getret;
618 int v3 = (nfsd->nd_flag & ND_NFSV3);
620 struct mbuf *mb, *mb2, *mp1, *mp2, *mp3, *mreq;
621 struct vnode *vp = NULL;
622 struct mount *mp = NULL;
626 struct uio io, *uiop = &io;
628 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
633 fhp = &nfh.fh_generic;
637 while (len < NFS_MAXPATHLEN) {
638 mp1 = m_getcl(MB_WAIT, MT_DATA, 0);
639 mp1->m_len = MCLBYTES;
646 if ((len + mp1->m_len) > NFS_MAXPATHLEN) {
647 mp1->m_len = NFS_MAXPATHLEN-len;
648 len = NFS_MAXPATHLEN;
651 ivp->iov_base = mtod(mp1, caddr_t);
652 ivp->iov_len = mp1->m_len;
657 uiop->uio_iovcnt = i;
658 uiop->uio_offset = 0;
659 uiop->uio_resid = len;
660 uiop->uio_rw = UIO_READ;
661 uiop->uio_segflg = UIO_SYSSPACE;
663 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
664 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
666 nfsm_reply(2 * NFSX_UNSIGNED);
667 nfsm_srvpostop_attr(1, NULL);
671 if (vp->v_type != VLNK) {
678 error = VOP_READLINK(vp, uiop, cred);
680 getret = VOP_GETATTR(vp, &attr);
683 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
685 nfsm_srvpostop_attr(getret, &attr);
691 if (uiop->uio_resid > 0) {
692 len -= uiop->uio_resid;
693 tlen = nfsm_rndup(len);
694 nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
696 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
697 *tl = txdr_unsigned(len);
712 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
713 struct thread *td, struct mbuf **mrq)
715 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
716 struct sockaddr *nam = nfsd->nd_nam;
717 caddr_t dpos = nfsd->nd_dpos;
718 struct ucred *cred = &nfsd->nd_cr;
722 struct nfs_fattr *fp;
727 int error = 0, rdonly, cnt, len, left, siz, tlen, getret;
728 int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
730 struct mbuf *mb, *mb2, *mreq;
732 struct vnode *vp = NULL;
733 struct mount *mp = NULL;
736 struct uio io, *uiop = &io;
737 struct vattr va, *vap = &va;
742 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
743 fhp = &nfh.fh_generic;
746 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
747 off = fxdr_hyper(tl);
749 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
750 off = (off_t)fxdr_unsigned(u_int32_t, *tl);
752 nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
755 * Reference vp. If an error occurs, vp will be invalid, but we
756 * have to NULL it just in case. The macros might goto nfsmout
760 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
761 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
764 nfsm_reply(2 * NFSX_UNSIGNED);
765 nfsm_srvpostop_attr(1, NULL);
770 if (vp->v_type != VREG) {
774 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
777 if ((error = nfsrv_access(mp, vp, VREAD, cred, rdonly, td, 1)) != 0)
778 error = nfsrv_access(mp, vp, VEXEC, cred, rdonly, td, 1);
780 getret = VOP_GETATTR(vp, vap);
786 nfsm_reply(NFSX_POSTOPATTR(v3));
787 nfsm_srvpostop_attr(getret, vap);
793 * Calculate byte count to read
796 if (off >= vap->va_size)
798 else if ((off + reqlen) > vap->va_size)
799 cnt = vap->va_size - off;
804 * Calculate seqcount for heuristic
812 * Locate best candidate
815 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
819 if (nfsheur[hi].nh_vp == vp) {
823 if (nfsheur[hi].nh_use > 0)
824 --nfsheur[hi].nh_use;
825 hi = (hi + 1) % NUM_HEURISTIC;
826 if (nfsheur[hi].nh_use < nh->nh_use)
830 if (nh->nh_vp != vp) {
833 nh->nh_use = NHUSE_INIT;
841 * Calculate heuristic
844 if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
845 if (++nh->nh_seqcount > IO_SEQMAX)
846 nh->nh_seqcount = IO_SEQMAX;
847 } else if (nh->nh_seqcount > 1) {
852 nh->nh_use += NHUSE_INC;
853 if (nh->nh_use > NHUSE_MAX)
854 nh->nh_use = NHUSE_MAX;
855 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
858 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
860 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
862 fp = (struct nfs_fattr *)tl;
863 tl += (NFSX_V3FATTR / sizeof (u_int32_t));
865 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
866 fp = (struct nfs_fattr *)tl;
867 tl += (NFSX_V2FATTR / sizeof (u_int32_t));
869 len = left = nfsm_rndup(cnt);
872 * Generate the mbuf list with the uio_iov ref. to it.
877 siz = min(M_TRAILINGSPACE(m), left);
883 m = m_getcl(MB_WAIT, MT_DATA, 0);
889 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
891 uiop->uio_iov = iv2 = iv;
897 panic("nfsrv_read iov");
898 siz = min(M_TRAILINGSPACE(m), left);
900 iv->iov_base = mtod(m, caddr_t) + m->m_len;
909 uiop->uio_iovcnt = i;
910 uiop->uio_offset = off;
911 uiop->uio_resid = len;
912 uiop->uio_rw = UIO_READ;
913 uiop->uio_segflg = UIO_SYSSPACE;
914 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
915 off = uiop->uio_offset;
917 FREE((caddr_t)iv2, M_TEMP);
918 if (error || (getret = VOP_GETATTR(vp, vap))) {
924 nfsm_reply(NFSX_POSTOPATTR(v3));
925 nfsm_srvpostop_attr(getret, vap);
934 nfsm_srvfillattr(vap, fp);
935 tlen = len - uiop->uio_resid;
936 cnt = cnt < tlen ? cnt : tlen;
937 tlen = nfsm_rndup(cnt);
938 if (len != tlen || tlen != cnt)
939 nfsm_adj(mb, len - tlen, tlen - cnt);
941 *tl++ = txdr_unsigned(cnt);
947 *tl = txdr_unsigned(cnt);
958 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
959 struct thread *td, struct mbuf **mrq)
961 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
962 struct sockaddr *nam = nfsd->nd_nam;
963 caddr_t dpos = nfsd->nd_dpos;
964 struct ucred *cred = &nfsd->nd_cr;
968 struct nfs_fattr *fp;
970 struct vattr va, forat;
971 struct vattr *vap = &va;
975 int error = 0, rdonly, len, forat_ret = 1;
976 int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
977 int stable = NFSV3WRITE_FILESYNC;
978 int v3 = (nfsd->nd_flag & ND_NFSV3);
980 struct mbuf *mb, *mb2, *mreq;
981 struct vnode *vp = NULL;
982 struct mount *mp = NULL;
985 struct uio io, *uiop = &io;
988 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
994 fhp = &nfh.fh_generic;
997 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
998 off = fxdr_hyper(tl);
1000 stable = fxdr_unsigned(int, *tl++);
1002 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1003 off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1006 stable = NFSV3WRITE_UNSTABLE;
1008 retlen = len = fxdr_unsigned(int32_t, *tl);
1012 * For NFS Version 2, it is not obvious what a write of zero length
1013 * should do, but I might as well be consistent with Version 3,
1014 * which is to return ok so long as there are no permission problems.
1022 adjust = dpos - mtod(mp1, caddr_t);
1023 mp1->m_len -= adjust;
1024 if (mp1->m_len > 0 && adjust > 0)
1025 NFSMADV(mp1, adjust);
1029 else if (mp1->m_len > 0) {
1032 mp1->m_len -= (i - len);
1041 if (len > NFS_MAXDATA || len < 0 || i < len) {
1043 nfsm_reply(2 * NFSX_UNSIGNED);
1044 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1048 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
1049 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
1052 nfsm_reply(2 * NFSX_UNSIGNED);
1053 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1058 forat_ret = VOP_GETATTR(vp, &forat);
1059 if (vp->v_type != VREG) {
1063 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1066 error = nfsrv_access(mp, vp, VWRITE, cred, rdonly, td, 1);
1071 nfsm_reply(NFSX_WCCDATA(v3));
1072 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1078 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
1080 uiop->uio_iov = iv = ivp;
1081 uiop->uio_iovcnt = cnt;
1084 if (mp1->m_len > 0) {
1085 ivp->iov_base = mtod(mp1, caddr_t);
1086 ivp->iov_len = mp1->m_len;
1094 * The IO_METASYNC flag indicates that all metadata (and not just
1095 * enough to ensure data integrity) mus be written to stable storage
1097 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1099 if (stable == NFSV3WRITE_UNSTABLE)
1100 ioflags = IO_NODELOCKED;
1101 else if (stable == NFSV3WRITE_DATASYNC)
1102 ioflags = (IO_SYNC | IO_NODELOCKED);
1104 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1105 uiop->uio_resid = len;
1106 uiop->uio_rw = UIO_WRITE;
1107 uiop->uio_segflg = UIO_SYSSPACE;
1108 uiop->uio_td = NULL;
1109 uiop->uio_offset = off;
1110 error = VOP_WRITE(vp, uiop, ioflags, cred);
1111 nfsstats.srvvop_writes++;
1112 FREE((caddr_t)iv, M_TEMP);
1114 aftat_ret = VOP_GETATTR(vp, vap);
1119 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1120 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1122 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1127 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1128 *tl++ = txdr_unsigned(retlen);
1130 * If nfs_async is set, then pretend the write was FILESYNC.
1132 if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1133 *tl++ = txdr_unsigned(stable);
1135 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1137 * Actually, there is no need to txdr these fields,
1138 * but it may make the values more human readable,
1139 * for debugging purposes.
1141 if (nfsver.tv_sec == 0)
1143 *tl++ = txdr_unsigned(nfsver.tv_sec);
1144 *tl = txdr_unsigned(nfsver.tv_nsec / 1000);
1146 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1147 nfsm_srvfillattr(vap, fp);
1156 * NFS write service with write gathering support. Called when
1157 * nfsrvw_procrastinate > 0.
1158 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1159 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1163 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp,
1164 struct thread *td, struct mbuf **mrq)
1167 struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1168 struct nfs_fattr *fp;
1171 struct nfsrvw_delayhash *wpp;
1173 struct vattr va, forat;
1177 int error = 0, rdonly, len, forat_ret = 1;
1178 int ioflags, aftat_ret = 1, adjust, v3, zeroing;
1180 struct mbuf *mb, *mb2, *mreq, *mrep, *md, *mp1;
1181 struct vnode *vp = NULL;
1182 struct mount *mp = NULL;
1183 struct uio io, *uiop = &io;
1186 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1195 mrep = nfsd->nd_mrep;
1197 dpos = nfsd->nd_dpos;
1198 cred = &nfsd->nd_cr;
1199 v3 = (nfsd->nd_flag & ND_NFSV3);
1200 LIST_INIT(&nfsd->nd_coalesce);
1201 nfsd->nd_mreq = NULL;
1202 nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1203 cur_usec = nfs_curusec();
1204 nfsd->nd_time = cur_usec +
1205 (v3 ? nfsrvw_procrastinate_v3 : nfsrvw_procrastinate);
1208 * Now, get the write header..
1210 nfsm_srvmtofh(&nfsd->nd_fh);
1212 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1213 nfsd->nd_off = fxdr_hyper(tl);
1215 nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1217 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1218 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1221 nfsd->nd_stable = NFSV3WRITE_UNSTABLE;
1223 len = fxdr_unsigned(int32_t, *tl);
1225 nfsd->nd_eoff = nfsd->nd_off + len;
1228 * Trim the header out of the mbuf list and trim off any trailing
1229 * junk so that the mbuf list has only the write data.
1237 adjust = dpos - mtod(mp1, caddr_t);
1238 mp1->m_len -= adjust;
1239 if (mp1->m_len > 0 && adjust > 0)
1240 NFSMADV(mp1, adjust);
1247 mp1->m_len -= (i - len);
1253 if (len > NFS_MAXDATA || len < 0 || i < len) {
1257 nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1259 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1260 nfsd->nd_mreq = mreq;
1261 nfsd->nd_mrep = NULL;
1266 * Add this entry to the hash and time queues.
1270 wp = slp->ns_tq.lh_first;
1271 while (wp && wp->nd_time < nfsd->nd_time) {
1273 wp = wp->nd_tq.le_next;
1275 NFS_DPF(WG, ("Q%03x", nfsd->nd_retxid & 0xfff));
1277 LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1279 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1281 if (nfsd->nd_mrep) {
1282 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1286 bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
1288 wp = wp->nd_hash.le_next;
1290 while (wp && wp->nd_off < nfsd->nd_off &&
1291 !bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
1293 wp = wp->nd_hash.le_next;
1296 LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1299 * Search the hash list for overlapping entries and
1302 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1303 wp = nfsd->nd_hash.le_next;
1304 if (NFSW_SAMECRED(owp, nfsd))
1305 nfsrvw_coalesce(owp, nfsd);
1308 LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1315 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1316 * and generate the associated reply mbuf list(s).
1319 cur_usec = nfs_curusec();
1321 for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = owp) {
1322 owp = nfsd->nd_tq.le_next;
1323 if (nfsd->nd_time > cur_usec)
1327 NFS_DPF(WG, ("P%03x", nfsd->nd_retxid & 0xfff));
1328 LIST_REMOVE(nfsd, nd_tq);
1329 LIST_REMOVE(nfsd, nd_hash);
1331 mrep = nfsd->nd_mrep;
1332 nfsd->nd_mrep = NULL;
1333 cred = &nfsd->nd_cr;
1334 v3 = (nfsd->nd_flag & ND_NFSV3);
1335 forat_ret = aftat_ret = 1;
1336 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &mp, &vp, cred, slp,
1337 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
1340 forat_ret = VOP_GETATTR(vp, &forat);
1341 if (vp->v_type != VREG) {
1345 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1351 error = nfsrv_access(mp, vp, VWRITE, cred, rdonly, td, 1);
1354 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1355 ioflags = IO_NODELOCKED;
1356 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1357 ioflags = (IO_SYNC | IO_NODELOCKED);
1359 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1360 uiop->uio_rw = UIO_WRITE;
1361 uiop->uio_segflg = UIO_SYSSPACE;
1362 uiop->uio_td = NULL;
1363 uiop->uio_offset = nfsd->nd_off;
1364 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1365 if (uiop->uio_resid > 0) {
1373 uiop->uio_iovcnt = i;
1374 MALLOC(iov, struct iovec *, i * sizeof (struct iovec),
1376 uiop->uio_iov = ivp = iov;
1379 if (mp1->m_len > 0) {
1380 ivp->iov_base = mtod(mp1, caddr_t);
1381 ivp->iov_len = mp1->m_len;
1387 error = VOP_WRITE(vp, uiop, ioflags, cred);
1388 nfsstats.srvvop_writes++;
1390 FREE((caddr_t)iov, M_TEMP);
1394 aftat_ret = VOP_GETATTR(vp, &va);
1400 * Loop around generating replies for all write rpcs that have
1401 * now been completed.
1405 NFS_DPF(WG, ("R%03x", nfsd->nd_retxid & 0xfff));
1407 nfsm_writereply(NFSX_WCCDATA(v3), v3);
1409 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1412 nfsm_writereply(NFSX_PREOPATTR(v3) +
1413 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1414 NFSX_WRITEVERF(v3), v3);
1416 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1417 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1418 *tl++ = txdr_unsigned(nfsd->nd_len);
1419 *tl++ = txdr_unsigned(swp->nd_stable);
1421 * Actually, there is no need to txdr these fields,
1422 * but it may make the values more human readable,
1423 * for debugging purposes.
1425 if (nfsver.tv_sec == 0)
1427 *tl++ = txdr_unsigned(nfsver.tv_sec);
1428 *tl = txdr_unsigned(nfsver.tv_nsec / 1000);
1430 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1431 nfsm_srvfillattr(&va, fp);
1434 nfsd->nd_mreq = mreq;
1436 panic("nfsrv_write: nd_mrep not free");
1439 * Done. Put it at the head of the timer queue so that
1440 * the final phase can return the reply.
1445 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1447 nfsd = swp->nd_coalesce.lh_first;
1449 LIST_REMOVE(nfsd, nd_tq);
1455 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1462 * Search for a reply to return.
1465 for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = nfsd->nd_tq.le_next)
1466 if (nfsd->nd_mreq) {
1467 NFS_DPF(WG, ("X%03x", nfsd->nd_retxid & 0xfff));
1468 LIST_REMOVE(nfsd, nd_tq);
1469 *mrq = nfsd->nd_mreq;
1478 * Coalesce the write request nfsd into owp. To do this we must:
1479 * - remove nfsd from the queues
1480 * - merge nfsd->nd_mrep into owp->nd_mrep
1481 * - update the nd_eoff and nd_stable for owp
1482 * - put nfsd on owp's nd_coalesce list
1483 * NB: Must be called at splsoftclock().
1486 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1490 struct nfsrv_descript *p;
1492 NFS_DPF(WG, ("C%03x-%03x",
1493 nfsd->nd_retxid & 0xfff, owp->nd_retxid & 0xfff));
1494 LIST_REMOVE(nfsd, nd_hash);
1495 LIST_REMOVE(nfsd, nd_tq);
1496 if (owp->nd_eoff < nfsd->nd_eoff) {
1497 overlap = owp->nd_eoff - nfsd->nd_off;
1499 panic("nfsrv_coalesce: bad off");
1501 m_adj(nfsd->nd_mrep, overlap);
1505 mp1->m_next = nfsd->nd_mrep;
1506 owp->nd_eoff = nfsd->nd_eoff;
1508 m_freem(nfsd->nd_mrep);
1509 nfsd->nd_mrep = NULL;
1510 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1511 owp->nd_stable = NFSV3WRITE_FILESYNC;
1512 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1513 owp->nd_stable == NFSV3WRITE_UNSTABLE)
1514 owp->nd_stable = NFSV3WRITE_DATASYNC;
1515 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1518 * If nfsd had anything else coalesced into it, transfer them
1519 * to owp, otherwise their replies will never get sent.
1521 for (p = nfsd->nd_coalesce.lh_first; p;
1522 p = nfsd->nd_coalesce.lh_first) {
1523 LIST_REMOVE(p, nd_tq);
1524 LIST_INSERT_HEAD(&owp->nd_coalesce, p, nd_tq);
1529 * nfs create service
1530 * now does a truncate to 0 length via. setattr if it already exists
1533 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1534 struct thread *td, struct mbuf **mrq)
1536 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1537 struct sockaddr *nam = nfsd->nd_nam;
1538 caddr_t dpos = nfsd->nd_dpos;
1539 struct ucred *cred = &nfsd->nd_cr;
1540 struct nfs_fattr *fp;
1541 struct vattr va, dirfor, diraft;
1542 struct vattr *vap = &va;
1543 struct nfsv2_sattr *sp;
1545 struct nlookupdata nd;
1548 int error = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1549 udev_t rdev = NOUDEV;
1550 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1553 struct mbuf *mb, *mb2, *mreq;
1561 u_char cverf[NFSX_V3CREATEVERF];
1563 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1569 fhp = &nfh.fh_generic;
1571 nfsm_srvnamesiz(len);
1574 * Call namei and do initial cleanup to get a few things
1575 * out of the way. If we get an initial error we cleanup
1576 * and return here to avoid special-casing the invalid nd
1577 * structure through the rest of the case. dirp may be
1578 * set even if an error occurs, but the nd structure will not
1579 * be valid at all if an error occurs so we have to invalidate it
1580 * prior to calling nfsm_reply ( which might goto nfsmout ).
1582 error = nfs_namei(&nd, cred, NLC_CREATE, &dvp, &vp,
1583 fhp, len, slp, nam, &md, &dpos, &dirp,
1584 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1585 mp = vfs_getvfs(&fhp->fh_fsid);
1589 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
1596 nfsm_reply(NFSX_WCCDATA(v3));
1597 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1603 * No error. Continue. State:
1606 * vp may be valid or NULL if the target does not
1610 * The error state is set through the code and we may also do some
1611 * opportunistic releasing of vnodes to avoid holding locks through
1612 * NFS I/O. The cleanup at the end is a catch-all
1617 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1618 how = fxdr_unsigned(int, *tl);
1620 case NFSV3CREATE_GUARDED:
1626 case NFSV3CREATE_UNCHECKED:
1629 case NFSV3CREATE_EXCLUSIVE:
1630 nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1631 bcopy(cp, cverf, NFSX_V3CREATEVERF);
1635 vap->va_type = VREG;
1637 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1638 vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1639 if (vap->va_type == VNON)
1640 vap->va_type = VREG;
1641 vap->va_mode = nfstov_mode(sp->sa_mode);
1642 switch (vap->va_type) {
1644 tsize = fxdr_unsigned(int32_t, sp->sa_size);
1646 vap->va_size = (u_quad_t)tsize;
1651 rdev = fxdr_unsigned(long, sp->sa_size);
1659 * Iff doesn't exist, create it
1660 * otherwise just truncate to 0 length
1661 * should I set the mode too ?
1663 * The only possible error we can have at this point is EEXIST.
1664 * nd.ni_vp will also be non-NULL in that case.
1667 if (vap->va_mode == (mode_t)VNOVAL)
1669 if (vap->va_type == VREG || vap->va_type == VSOCK) {
1671 error = VOP_NCREATE(&nd.nl_nch, dvp, &vp,
1676 if (exclusive_flag) {
1679 bcopy(cverf, (caddr_t)&vap->va_atime,
1681 error = VOP_SETATTR(vp, vap, cred);
1685 vap->va_type == VCHR ||
1686 vap->va_type == VBLK ||
1687 vap->va_type == VFIFO
1690 * Handle SysV FIFO node special cases. All other
1691 * devices require super user to access.
1693 if (vap->va_type == VCHR && rdev == 0xffffffff)
1694 vap->va_type = VFIFO;
1695 if (vap->va_type != VFIFO &&
1696 (error = priv_check_cred(cred, PRIV_ROOT, 0))) {
1699 vap->va_rmajor = umajor(rdev);
1700 vap->va_rminor = uminor(rdev);
1703 error = VOP_NMKNOD(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
1710 * XXX what is this junk supposed to do ?
1717 * release dvp prior to lookup
1725 * Even though LOCKPARENT was cleared, ni_dvp may
1728 nd.ni_cnd.cn_nameiop = NAMEI_LOOKUP;
1729 nd.ni_cnd.cn_flags &= ~(CNP_LOCKPARENT);
1730 nd.ni_cnd.cn_td = td;
1731 nd.ni_cnd.cn_cred = cred;
1733 error = lookup(&nd);
1738 /* fall through on certain errors */
1740 nfsrv_object_create(nd.ni_vp);
1741 if (nd.ni_cnd.cn_flags & CNP_ISSYMLINK) {
1750 if (vap->va_size != -1) {
1751 error = nfsrv_access(mp, vp, VWRITE, cred,
1752 (nd.nl_flags & NLC_NFS_RDONLY), td, 0);
1754 tempsize = vap->va_size;
1756 vap->va_size = tempsize;
1757 error = VOP_SETATTR(vp, vap, cred);
1763 bzero(&fhp->fh_fid, sizeof(fhp->fh_fid));
1764 error = VFS_VPTOFH(vp, &fhp->fh_fid);
1766 error = VOP_GETATTR(vp, vap);
1769 if (exclusive_flag && !error &&
1770 bcmp(cverf, (caddr_t)&vap->va_atime, NFSX_V3CREATEVERF))
1772 diraft_ret = VOP_GETATTR(dirp, &diraft);
1776 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1779 nfsm_srvpostop_fh(fhp);
1780 nfsm_srvpostop_attr(0, vap);
1782 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1785 nfsm_srvfhtom(fhp, v3);
1786 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1787 nfsm_srvfillattr(vap, fp);
1812 * nfs v3 mknod service
1815 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1816 struct thread *td, struct mbuf **mrq)
1818 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1819 struct sockaddr *nam = nfsd->nd_nam;
1820 caddr_t dpos = nfsd->nd_dpos;
1821 struct ucred *cred = &nfsd->nd_cr;
1822 struct vattr va, dirfor, diraft;
1823 struct vattr *vap = &va;
1825 struct nlookupdata nd;
1828 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1831 struct mbuf *mb, *mb2, *mreq;
1838 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1844 fhp = &nfh.fh_generic;
1846 nfsm_srvnamesiz(len);
1849 * Handle nfs_namei() call. If an error occurs, the nd structure
1850 * is not valid. However, nfsm_*() routines may still jump to
1854 error = nfs_namei(&nd, cred, NLC_CREATE, &dvp, &vp,
1855 fhp, len, slp, nam, &md, &dpos, &dirp,
1856 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1858 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
1860 nfsm_reply(NFSX_WCCDATA(1));
1861 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1865 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1866 vtyp = nfsv3tov_type(*tl);
1867 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1868 error = NFSERR_BADTYPE;
1873 if (vtyp == VCHR || vtyp == VBLK) {
1874 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1875 vap->va_rmajor = fxdr_unsigned(u_int32_t, *tl++);
1876 vap->va_rminor = fxdr_unsigned(u_int32_t, *tl);
1880 * Iff doesn't exist, create it.
1886 vap->va_type = vtyp;
1887 if (vap->va_mode == (mode_t)VNOVAL)
1889 if (vtyp == VSOCK) {
1891 error = VOP_NCREATE(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
1895 if (vtyp != VFIFO && (error = priv_check_cred(cred, PRIV_ROOT, 0)))
1899 error = VOP_NMKNOD(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
1907 * send response, cleanup, return.
1919 bzero(&fhp->fh_fid, sizeof(fhp->fh_fid));
1920 error = VFS_VPTOFH(vp, &fhp->fh_fid);
1922 error = VOP_GETATTR(vp, vap);
1928 diraft_ret = VOP_GETATTR(dirp, &diraft);
1933 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1935 nfsm_srvpostop_fh(fhp);
1936 nfsm_srvpostop_attr(0, vap);
1938 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1956 * nfs remove service
1959 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1960 struct thread *td, struct mbuf **mrq)
1962 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1963 struct sockaddr *nam = nfsd->nd_nam;
1964 caddr_t dpos = nfsd->nd_dpos;
1965 struct ucred *cred = &nfsd->nd_cr;
1966 struct nlookupdata nd;
1970 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1971 int v3 = (nfsd->nd_flag & ND_NFSV3);
1973 struct mbuf *mb, *mreq;
1977 struct vattr dirfor, diraft;
1981 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1987 fhp = &nfh.fh_generic;
1989 nfsm_srvnamesiz(len);
1991 error = nfs_namei(&nd, cred, NLC_DELETE, &dvp, &vp,
1992 fhp, len, slp, nam, &md, &dpos, &dirp,
1993 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1996 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
1999 if (vp->v_type == VDIR) {
2000 error = EPERM; /* POSIX */
2004 * The root of a mounted filesystem cannot be deleted.
2006 if (vp->v_flag & VROOT) {
2018 error = VOP_NREMOVE(&nd.nl_nch, dvp, nd.nl_cred);
2024 diraft_ret = VOP_GETATTR(dirp, &diraft);
2025 nfsm_reply(NFSX_WCCDATA(v3));
2027 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2046 * nfs rename service
2049 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2050 struct thread *td, struct mbuf **mrq)
2052 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2053 struct sockaddr *nam = nfsd->nd_nam;
2054 caddr_t dpos = nfsd->nd_dpos;
2055 struct ucred *cred = &nfsd->nd_cr;
2059 int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
2060 int tdirfor_ret = 1, tdiraft_ret = 1;
2061 int v3 = (nfsd->nd_flag & ND_NFSV3);
2063 struct mbuf *mb, *mreq;
2064 struct nlookupdata fromnd, tond;
2065 struct vnode *fvp, *fdirp, *fdvp;
2066 struct vnode *tvp, *tdirp, *tdvp;
2067 struct namecache *ncp;
2068 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
2070 fhandle_t *ffhp, *tfhp;
2073 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2077 ffhp = &fnfh.fh_generic;
2078 tfhp = &tnfh.fh_generic;
2081 * Clear fields incase goto nfsmout occurs from macro.
2084 nlookup_zero(&fromnd);
2085 nlookup_zero(&tond);
2089 nfsm_srvmtofh(ffhp);
2090 nfsm_srvnamesiz(len);
2092 * Remember our original uid so that we can reset cr_uid before
2093 * the second nfs_namei() call, in case it is remapped.
2095 saved_uid = cred->cr_uid;
2096 error = nfs_namei(&fromnd, cred, NLC_RENAME_SRC,
2098 ffhp, len, slp, nam, &md, &dpos, &fdirp,
2099 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2102 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor);
2105 nfsm_reply(2 * NFSX_WCCDATA(v3));
2106 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2107 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2113 * We have to unlock the from ncp before we can safely lookup
2116 KKASSERT(fromnd.nl_flags & NLC_NCPISLOCKED);
2117 cache_unlock(&fromnd.nl_nch);
2118 fromnd.nl_flags &= ~NLC_NCPISLOCKED;
2119 nfsm_srvmtofh(tfhp);
2120 nfsm_strsiz(len2, NFS_MAXNAMLEN);
2121 cred->cr_uid = saved_uid;
2123 error = nfs_namei(&tond, cred, NLC_RENAME_DST, NULL, NULL,
2124 tfhp, len2, slp, nam, &md, &dpos, &tdirp,
2125 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2128 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor);
2136 if (cache_lock_nonblock(&fromnd.nl_nch) == 0) {
2137 cache_resolve(&fromnd.nl_nch, fromnd.nl_cred);
2138 } else if (fromnd.nl_nch.ncp > tond.nl_nch.ncp) {
2139 cache_lock(&fromnd.nl_nch);
2140 cache_resolve(&fromnd.nl_nch, fromnd.nl_cred);
2142 cache_unlock(&tond.nl_nch);
2143 cache_lock(&fromnd.nl_nch);
2144 cache_resolve(&fromnd.nl_nch, fromnd.nl_cred);
2145 cache_lock(&tond.nl_nch);
2146 cache_resolve(&tond.nl_nch, tond.nl_cred);
2148 fromnd.nl_flags |= NLC_NCPISLOCKED;
2150 fvp = fromnd.nl_nch.ncp->nc_vp;
2151 tvp = tond.nl_nch.ncp->nc_vp;
2154 * Set fdvp and tdvp. We haven't done all the topology checks
2155 * so these can wind up NULL (e.g. if either fvp or tvp is a mount
2156 * point). If we get through the checks these will be guarenteed
2159 * Holding the children ncp's should be sufficient to prevent
2160 * fdvp and tdvp ripouts.
2162 if (fromnd.nl_nch.ncp->nc_parent)
2163 fdvp = fromnd.nl_nch.ncp->nc_parent->nc_vp;
2166 if (tond.nl_nch.ncp->nc_parent)
2167 tdvp = tond.nl_nch.ncp->nc_parent->nc_vp;
2172 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2178 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2185 if (tvp->v_type == VDIR && (tond.nl_nch.ncp->nc_flag & NCF_ISMOUNTPT)) {
2193 if (fvp->v_type == VDIR && (fromnd.nl_nch.ncp->nc_flag & NCF_ISMOUNTPT)) {
2200 if (fromnd.nl_nch.mount != tond.nl_nch.mount) {
2207 if (fromnd.nl_nch.ncp == tond.nl_nch.ncp->nc_parent) {
2215 * You cannot rename a source into itself or a subdirectory of itself.
2216 * We check this by travsering the target directory upwards looking
2217 * for a match against the source.
2220 for (ncp = tond.nl_nch.ncp; ncp; ncp = ncp->nc_parent) {
2221 if (fromnd.nl_nch.ncp == ncp) {
2229 * If source is the same as the destination (that is the
2230 * same vnode with the same name in the same directory),
2231 * then there is nothing to do.
2233 if (fromnd.nl_nch.ncp == tond.nl_nch.ncp)
2238 * The VOP_NRENAME function releases all vnode references &
2239 * locks prior to returning so we need to clear the pointers
2240 * to bypass cleanup code later on.
2242 error = VOP_NRENAME(&fromnd.nl_nch, &tond.nl_nch,
2243 fdvp, tdvp, tond.nl_cred);
2252 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft);
2254 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft);
2255 nfsm_reply(2 * NFSX_WCCDATA(v3));
2257 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2258 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2266 nlookup_done(&tond);
2269 nlookup_done(&fromnd);
2277 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2278 struct thread *td, struct mbuf **mrq)
2280 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2281 struct sockaddr *nam = nfsd->nd_nam;
2282 caddr_t dpos = nfsd->nd_dpos;
2283 struct ucred *cred = &nfsd->nd_cr;
2284 struct nlookupdata nd;
2288 int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1;
2289 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2291 struct mbuf *mb, *mreq;
2298 struct vattr dirfor, diraft, at;
2300 fhandle_t *fhp, *dfhp;
2302 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2304 dirp = dvp = vp = xp = NULL;
2307 fhp = &nfh.fh_generic;
2308 dfhp = &dnfh.fh_generic;
2310 nfsm_srvmtofh(dfhp);
2311 nfsm_srvnamesiz(len);
2313 error = nfsrv_fhtovp(fhp, FALSE, &xmp, &xp, cred, slp, nam,
2314 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
2316 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2317 nfsm_srvpostop_attr(getret, &at);
2318 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2323 if (xp->v_type == VDIR) {
2324 error = EPERM; /* POSIX */
2328 error = nfs_namei(&nd, cred, NLC_CREATE, &dvp, &vp,
2329 dfhp, len, slp, nam, &md, &dpos, &dirp,
2330 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2333 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2342 if (xp->v_mount != dvp->v_mount)
2347 error = VOP_NLINK(&nd.nl_nch, dvp, xp, nd.nl_cred);
2355 getret = VOP_GETATTR(xp, &at);
2357 diraft_ret = VOP_GETATTR(dirp, &diraft);
2358 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2360 nfsm_srvpostop_attr(getret, &at);
2361 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2384 * nfs symbolic link service
2387 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2388 struct thread *td, struct mbuf **mrq)
2390 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2391 struct sockaddr *nam = nfsd->nd_nam;
2392 caddr_t dpos = nfsd->nd_dpos;
2393 struct ucred *cred = &nfsd->nd_cr;
2394 struct vattr va, dirfor, diraft;
2395 struct nlookupdata nd;
2396 struct vattr *vap = &va;
2399 struct nfsv2_sattr *sp;
2400 char *bpos, *pathcp = NULL, *cp2;
2403 int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2404 int v3 = (nfsd->nd_flag & ND_NFSV3);
2405 struct mbuf *mb, *mreq, *mb2;
2412 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2418 fhp = &nfh.fh_generic;
2420 nfsm_srvnamesiz(len);
2422 error = nfs_namei(&nd, cred, NLC_CREATE, &dvp, &vp,
2423 fhp, len, slp, nam, &md, &dpos, &dirp,
2424 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2427 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2435 nfsm_strsiz(len2, NFS_MAXPATHLEN);
2436 MALLOC(pathcp, caddr_t, len2 + 1, M_TEMP, M_WAITOK);
2437 iv.iov_base = pathcp;
2439 io.uio_resid = len2;
2443 io.uio_segflg = UIO_SYSSPACE;
2444 io.uio_rw = UIO_READ;
2446 nfsm_mtouio(&io, len2);
2448 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2449 vap->va_mode = nfstov_mode(sp->sa_mode);
2451 *(pathcp + len2) = '\0';
2457 if (vap->va_mode == (mode_t)VNOVAL)
2461 error = VOP_NSYMLINK(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap, pathcp);
2465 bzero(&fhp->fh_fid, sizeof(fhp->fh_fid));
2466 error = VFS_VPTOFH(vp, &fhp->fh_fid);
2468 error = VOP_GETATTR(vp, vap);
2483 FREE(pathcp, M_TEMP);
2487 diraft_ret = VOP_GETATTR(dirp, &diraft);
2491 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2494 nfsm_srvpostop_fh(fhp);
2495 nfsm_srvpostop_attr(0, vap);
2497 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2509 FREE(pathcp, M_TEMP);
2517 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2518 struct thread *td, struct mbuf **mrq)
2520 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2521 struct sockaddr *nam = nfsd->nd_nam;
2522 caddr_t dpos = nfsd->nd_dpos;
2523 struct ucred *cred = &nfsd->nd_cr;
2524 struct vattr va, dirfor, diraft;
2525 struct vattr *vap = &va;
2526 struct nfs_fattr *fp;
2527 struct nlookupdata nd;
2532 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2533 int v3 = (nfsd->nd_flag & ND_NFSV3);
2535 struct mbuf *mb, *mb2, *mreq;
2542 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2548 fhp = &nfh.fh_generic;
2550 nfsm_srvnamesiz(len);
2552 error = nfs_namei(&nd, cred, NLC_CREATE, &dvp, &vp,
2553 fhp, len, slp, nam, &md, &dpos, &dirp,
2554 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2557 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2560 nfsm_reply(NFSX_WCCDATA(v3));
2561 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2569 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2570 vap->va_mode = nfstov_mode(*tl++);
2574 * At this point nd.ni_dvp is referenced and exclusively locked and
2575 * nd.ni_vp, if it exists, is referenced but not locked.
2578 vap->va_type = VDIR;
2585 * Issue mkdir op. Since SAVESTART is not set, the pathname
2586 * component is freed by the VOP call. This will fill-in
2587 * nd.ni_vp, reference, and exclusively lock it.
2589 if (vap->va_mode == (mode_t)VNOVAL)
2592 error = VOP_NMKDIR(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
2597 bzero(&fhp->fh_fid, sizeof(fhp->fh_fid));
2598 error = VFS_VPTOFH(vp, &fhp->fh_fid);
2600 error = VOP_GETATTR(vp, vap);
2604 diraft_ret = VOP_GETATTR(dirp, &diraft);
2605 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2608 nfsm_srvpostop_fh(fhp);
2609 nfsm_srvpostop_attr(0, vap);
2611 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2613 nfsm_srvfhtom(fhp, v3);
2614 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2615 nfsm_srvfillattr(vap, fp);
2639 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2640 struct thread *td, struct mbuf **mrq)
2642 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2643 struct sockaddr *nam = nfsd->nd_nam;
2644 caddr_t dpos = nfsd->nd_dpos;
2645 struct ucred *cred = &nfsd->nd_cr;
2649 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2650 int v3 = (nfsd->nd_flag & ND_NFSV3);
2652 struct mbuf *mb, *mreq;
2656 struct vattr dirfor, diraft;
2659 struct nlookupdata nd;
2661 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2667 fhp = &nfh.fh_generic;
2669 nfsm_srvnamesiz(len);
2671 error = nfs_namei(&nd, cred, NLC_DELETE, &dvp, &vp,
2672 fhp, len, slp, nam, &md, &dpos, &dirp,
2673 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2676 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2679 nfsm_reply(NFSX_WCCDATA(v3));
2680 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2684 if (vp->v_type != VDIR) {
2690 * The root of a mounted filesystem cannot be deleted.
2692 if (vp->v_flag & VROOT)
2696 * Issue or abort op. Since SAVESTART is not set, path name
2697 * component is freed by the VOP after either.
2704 error = VOP_NRMDIR(&nd.nl_nch, dvp, nd.nl_cred);
2711 diraft_ret = VOP_GETATTR(dirp, &diraft);
2712 nfsm_reply(NFSX_WCCDATA(v3));
2714 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2735 * nfs readdir service
2736 * - mallocs what it thinks is enough to read
2737 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2738 * - calls VOP_READDIR()
2739 * - loops around building the reply
2740 * if the output generated exceeds count break out of loop
2741 * The nfsm_clget macro is used here so that the reply will be packed
2742 * tightly in mbuf clusters.
2743 * - it only knows that it has encountered eof when the VOP_READDIR()
2745 * - as such one readdir rpc will return eof false although you are there
2746 * and then the next will return eof
2747 * - it trims out records with d_fileno == 0
2748 * this doesn't matter for Unix clients, but they might confuse clients
2750 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2751 * than requested, but this may not apply to all filesystems. For
2752 * example, client NFS does not { although it is never remote mounted
2754 * The alternate call nfsrv_readdirplus() does lookups as well.
2755 * PS: The NFS protocol spec. does not clarify what the "count" byte
2756 * argument is a count of.. just name strings and file id's or the
2757 * entire reply rpc or ...
2758 * I tried just file name and id sizes and it confused the Sun client,
2759 * so I am using the full rpc size now. The "paranoia.." comment refers
2760 * to including the status longwords that are not a part of the dir.
2761 * "entry" structures, but are in the rpc.
2765 u_int32_t fl_postopok;
2766 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2768 u_int32_t fl_fhsize;
2769 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2773 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2774 struct thread *td, struct mbuf **mrq)
2776 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2777 struct sockaddr *nam = nfsd->nd_nam;
2778 caddr_t dpos = nfsd->nd_dpos;
2779 struct ucred *cred = &nfsd->nd_cr;
2786 struct mbuf *mb, *mb2, *mreq, *mp1, *mp2;
2787 char *cpos, *cend, *cp2, *rbuf;
2788 struct vnode *vp = NULL;
2789 struct mount *mp = NULL;
2795 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2796 int siz, cnt, fullsiz, eofflag, rdonly, ncookies;
2797 int v3 = (nfsd->nd_flag & ND_NFSV3);
2798 u_quad_t off, toff, verf;
2799 off_t *cookies = NULL, *cookiep;
2801 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2802 fhp = &nfh.fh_generic;
2805 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2806 toff = fxdr_hyper(tl);
2808 verf = fxdr_hyper(tl);
2811 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2812 toff = fxdr_unsigned(u_quad_t, *tl++);
2813 verf = 0; /* shut up gcc */
2816 cnt = fxdr_unsigned(int, *tl);
2817 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2818 xfer = NFS_SRVMAXDATA(nfsd);
2824 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
2825 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
2826 if (!error && vp->v_type != VDIR) {
2832 nfsm_reply(NFSX_UNSIGNED);
2833 nfsm_srvpostop_attr(getret, &at);
2839 * Obtain lock on vnode for this section of the code
2843 error = getret = VOP_GETATTR(vp, &at);
2846 * XXX This check may be too strict for Solaris 2.5 clients.
2848 if (!error && toff && verf && verf != at.va_filerev)
2849 error = NFSERR_BAD_COOKIE;
2853 error = nfsrv_access(mp, vp, VEXEC, cred, rdonly, td, 0);
2857 nfsm_reply(NFSX_POSTOPATTR(v3));
2858 nfsm_srvpostop_attr(getret, &at);
2865 * end section. Allocate rbuf and continue
2867 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
2870 iv.iov_len = fullsiz;
2873 io.uio_offset = (off_t)off;
2874 io.uio_resid = fullsiz;
2875 io.uio_segflg = UIO_SYSSPACE;
2876 io.uio_rw = UIO_READ;
2880 kfree((caddr_t)cookies, M_TEMP);
2883 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
2884 off = (off_t)io.uio_offset;
2885 if (!cookies && !error)
2886 error = NFSERR_PERM;
2888 getret = VOP_GETATTR(vp, &at);
2895 kfree((caddr_t)rbuf, M_TEMP);
2897 kfree((caddr_t)cookies, M_TEMP);
2898 nfsm_reply(NFSX_POSTOPATTR(v3));
2899 nfsm_srvpostop_attr(getret, &at);
2904 siz -= io.uio_resid;
2907 * If nothing read, return eof
2913 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2916 nfsm_srvpostop_attr(getret, &at);
2917 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2918 txdr_hyper(at.va_filerev, tl);
2921 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2924 FREE((caddr_t)rbuf, M_TEMP);
2925 FREE((caddr_t)cookies, M_TEMP);
2932 * Check for degenerate cases of nothing useful read.
2933 * If so go try again
2937 dp = (struct dirent *)cpos;
2940 * For some reason FreeBSD's ufs_readdir() chooses to back the
2941 * directory offset up to a block boundary, so it is necessary to
2942 * skip over the records that preceed the requested offset. This
2943 * requires the assumption that file offset cookies monotonically
2946 while (cpos < cend && ncookies > 0 &&
2947 (dp->d_ino == 0 || dp->d_type == DT_WHT ||
2948 ((u_quad_t)(*cookiep)) <= toff)) {
2949 dp = _DIRENT_NEXT(dp);
2954 if (cpos >= cend || ncookies == 0) {
2960 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
2961 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2963 nfsm_srvpostop_attr(getret, &at);
2964 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2965 txdr_hyper(at.va_filerev, tl);
2969 be = bp + M_TRAILINGSPACE(mp1);
2971 /* Loop through the records and build reply */
2972 while (cpos < cend && ncookies > 0) {
2973 if (dp->d_ino != 0 && dp->d_type != DT_WHT) {
2974 nlen = dp->d_namlen;
2975 rem = nfsm_rndup(nlen) - nlen;
2976 len += (4 * NFSX_UNSIGNED + nlen + rem);
2978 len += 2 * NFSX_UNSIGNED;
2984 * Build the directory record xdr from
2987 nfsm_clget(mp1, mp2, mb, bp, be, tl);
2989 bp += NFSX_UNSIGNED;
2991 nfsm_clget(mp1, mp2, mb, bp, be, tl);
2993 bp += NFSX_UNSIGNED;
2995 nfsm_clget(mp1, mp2, mb, bp, be, tl);
2996 *tl = txdr_unsigned(dp->d_ino);
2997 bp += NFSX_UNSIGNED;
2998 nfsm_clget(mp1, mp2, mb, bp, be, tl);
2999 *tl = txdr_unsigned(nlen);
3000 bp += NFSX_UNSIGNED;
3002 /* And loop around copying the name */
3006 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3011 bcopy(cp, bp, tsiz);
3017 /* And null pad to a int32_t boundary */
3018 for (i = 0; i < rem; i++)
3020 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3022 /* Finish off the record */
3024 *tl = txdr_unsigned(*cookiep >> 32);
3025 bp += NFSX_UNSIGNED;
3026 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3028 *tl = txdr_unsigned(*cookiep);
3029 bp += NFSX_UNSIGNED;
3031 dp = _DIRENT_NEXT(dp);
3038 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3040 bp += NFSX_UNSIGNED;
3041 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3046 bp += NFSX_UNSIGNED;
3049 mp1->m_len = bp - mtod(mp1, caddr_t);
3051 mp1->m_len += bp - bpos;
3052 FREE((caddr_t)rbuf, M_TEMP);
3053 FREE((caddr_t)cookies, M_TEMP);
3062 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3063 struct thread *td, struct mbuf **mrq)
3065 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3066 struct sockaddr *nam = nfsd->nd_nam;
3067 caddr_t dpos = nfsd->nd_dpos;
3068 struct ucred *cred = &nfsd->nd_cr;
3075 struct mbuf *mb, *mb2, *mreq, *mp1, *mp2;
3076 char *cpos, *cend, *cp2, *rbuf;
3077 struct vnode *vp = NULL, *nvp;
3078 struct mount *mp = NULL;
3081 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3084 struct vattr va, at, *vap = &va;
3085 struct nfs_fattr *fp;
3086 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3087 int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies;
3088 u_quad_t off, toff, verf;
3089 off_t *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3091 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3092 fhp = &nfh.fh_generic;
3094 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
3095 toff = fxdr_hyper(tl);
3097 verf = fxdr_hyper(tl);
3099 siz = fxdr_unsigned(int, *tl++);
3100 cnt = fxdr_unsigned(int, *tl);
3102 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3103 xfer = NFS_SRVMAXDATA(nfsd);
3109 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
3110 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3111 if (!error && vp->v_type != VDIR) {
3117 nfsm_reply(NFSX_UNSIGNED);
3118 nfsm_srvpostop_attr(getret, &at);
3122 error = getret = VOP_GETATTR(vp, &at);
3125 * XXX This check may be too strict for Solaris 2.5 clients.
3127 if (!error && toff && verf && verf != at.va_filerev)
3128 error = NFSERR_BAD_COOKIE;
3131 error = nfsrv_access(mp, vp, VEXEC, cred, rdonly, td, 0);
3136 nfsm_reply(NFSX_V3POSTOPATTR);
3137 nfsm_srvpostop_attr(getret, &at);
3142 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3145 iv.iov_len = fullsiz;
3148 io.uio_offset = (off_t)off;
3149 io.uio_resid = fullsiz;
3150 io.uio_segflg = UIO_SYSSPACE;
3151 io.uio_rw = UIO_READ;
3155 kfree((caddr_t)cookies, M_TEMP);
3158 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3159 off = (u_quad_t)io.uio_offset;
3160 getret = VOP_GETATTR(vp, &at);
3161 if (!cookies && !error)
3162 error = NFSERR_PERM;
3169 kfree((caddr_t)cookies, M_TEMP);
3170 kfree((caddr_t)rbuf, M_TEMP);
3171 nfsm_reply(NFSX_V3POSTOPATTR);
3172 nfsm_srvpostop_attr(getret, &at);
3177 siz -= io.uio_resid;
3180 * If nothing read, return eof
3186 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3188 nfsm_srvpostop_attr(getret, &at);
3189 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3190 txdr_hyper(at.va_filerev, tl);
3194 FREE((caddr_t)cookies, M_TEMP);
3195 FREE((caddr_t)rbuf, M_TEMP);
3202 * Check for degenerate cases of nothing useful read.
3203 * If so go try again
3207 dp = (struct dirent *)cpos;
3210 * For some reason FreeBSD's ufs_readdir() chooses to back the
3211 * directory offset up to a block boundary, so it is necessary to
3212 * skip over the records that preceed the requested offset. This
3213 * requires the assumption that file offset cookies monotonically
3216 while (cpos < cend && ncookies > 0 &&
3217 (dp->d_ino == 0 || dp->d_type == DT_WHT ||
3218 ((u_quad_t)(*cookiep)) <= toff)) {
3219 dp = _DIRENT_NEXT(dp);
3224 if (cpos >= cend || ncookies == 0) {
3231 * Probe one of the directory entries to see if the filesystem
3234 if (VFS_VGET(vp->v_mount, dp->d_ino, &nvp) == EOPNOTSUPP) {
3235 error = NFSERR_NOTSUPP;
3238 kfree((caddr_t)cookies, M_TEMP);
3239 kfree((caddr_t)rbuf, M_TEMP);
3240 nfsm_reply(NFSX_V3POSTOPATTR);
3241 nfsm_srvpostop_attr(getret, &at);
3250 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
3252 nfsm_srvpostop_attr(getret, &at);
3253 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3254 txdr_hyper(at.va_filerev, tl);
3257 be = bp + M_TRAILINGSPACE(mp1);
3259 /* Loop through the records and build reply */
3260 while (cpos < cend && ncookies > 0) {
3261 if (dp->d_ino != 0 && dp->d_type != DT_WHT) {
3262 nlen = dp->d_namlen;
3263 rem = nfsm_rndup(nlen)-nlen;
3266 * For readdir_and_lookup get the vnode using
3269 if (VFS_VGET(vp->v_mount, dp->d_ino, &nvp))
3271 bzero((caddr_t)nfhp, NFSX_V3FH);
3272 nfhp->fh_fsid = fhp->fh_fsid;
3273 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
3278 if (VOP_GETATTR(nvp, vap)) {
3287 * If either the dircount or maxcount will be
3288 * exceeded, get out now. Both of these lengths
3289 * are calculated conservatively, including all
3292 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3294 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3295 if (len > cnt || dirlen > fullsiz) {
3301 * Build the directory record xdr from
3304 fp = (struct nfs_fattr *)&fl.fl_fattr;
3305 nfsm_srvfillattr(vap, fp);
3306 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3307 fl.fl_fhok = nfs_true;
3308 fl.fl_postopok = nfs_true;
3309 fl.fl_off.nfsuquad[0] = txdr_unsigned(*cookiep >> 32);
3310 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3312 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3314 bp += NFSX_UNSIGNED;
3315 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3317 bp += NFSX_UNSIGNED;
3318 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3319 *tl = txdr_unsigned(dp->d_ino);
3320 bp += NFSX_UNSIGNED;
3321 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3322 *tl = txdr_unsigned(nlen);
3323 bp += NFSX_UNSIGNED;
3325 /* And loop around copying the name */
3329 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3330 if ((bp + xfer) > be)
3334 bcopy(cp, bp, tsiz);
3340 /* And null pad to a int32_t boundary */
3341 for (i = 0; i < rem; i++)
3345 * Now copy the flrep structure out.
3347 xfer = sizeof (struct flrep);
3350 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3351 if ((bp + xfer) > be)
3355 bcopy(cp, bp, tsiz);
3363 dp = _DIRENT_NEXT(dp);
3370 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3372 bp += NFSX_UNSIGNED;
3373 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3378 bp += NFSX_UNSIGNED;
3381 mp1->m_len = bp - mtod(mp1, caddr_t);
3383 mp1->m_len += bp - bpos;
3384 FREE((caddr_t)cookies, M_TEMP);
3385 FREE((caddr_t)rbuf, M_TEMP);
3393 * nfs commit service
3396 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3397 struct thread *td, struct mbuf **mrq)
3399 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3400 struct sockaddr *nam = nfsd->nd_nam;
3401 caddr_t dpos = nfsd->nd_dpos;
3402 struct ucred *cred = &nfsd->nd_cr;
3403 struct vattr bfor, aft;
3404 struct vnode *vp = NULL;
3405 struct mount *mp = NULL;
3411 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt;
3413 struct mbuf *mb, *mb2, *mreq;
3416 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3417 fhp = &nfh.fh_generic;
3419 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3422 * XXX At this time VOP_FSYNC() does not accept offset and byte
3423 * count parameters, so these arguments are useless (someday maybe).
3425 off = fxdr_hyper(tl);
3427 cnt = fxdr_unsigned(int, *tl);
3428 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
3429 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3431 nfsm_reply(2 * NFSX_UNSIGNED);
3432 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3436 for_ret = VOP_GETATTR(vp, &bfor);
3438 if (cnt > MAX_COMMIT_COUNT) {
3440 * Give up and do the whole thing
3443 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3444 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3446 error = VOP_FSYNC(vp, MNT_WAIT);
3449 * Locate and synchronously write any buffers that fall
3450 * into the requested range. Note: we are assuming that
3451 * f_iosize is a power of 2.
3453 int iosize = vp->v_mount->mnt_stat.f_iosize;
3454 int iomask = iosize - 1;
3458 * Align to iosize boundry, super-align to page boundry.
3461 cnt += off & iomask;
3462 off &= ~(u_quad_t)iomask;
3464 if (off & PAGE_MASK) {
3465 cnt += off & PAGE_MASK;
3466 off &= ~(u_quad_t)PAGE_MASK;
3471 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3472 vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
3480 * If we have a buffer and it is marked B_DELWRI we
3481 * have to lock and write it. Otherwise the prior
3482 * write is assumed to have already been committed.
3484 if ((bp = findblk(vp, loffset)) != NULL && (bp->b_flags & B_DELWRI)) {
3485 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
3486 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL) == 0)
3488 continue; /* retry */
3491 bp->b_flags &= ~B_ASYNC;
3504 aft_ret = VOP_GETATTR(vp, &aft);
3507 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3508 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3510 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3511 if (nfsver.tv_sec == 0)
3513 *tl++ = txdr_unsigned(nfsver.tv_sec);
3514 *tl = txdr_unsigned(nfsver.tv_nsec / 1000);
3525 * nfs statfs service
3528 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3529 struct thread *td, struct mbuf **mrq)
3531 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3532 struct sockaddr *nam = nfsd->nd_nam;
3533 caddr_t dpos = nfsd->nd_dpos;
3534 struct ucred *cred = &nfsd->nd_cr;
3536 struct nfs_statfs *sfp;
3540 int error = 0, rdonly, getret = 1;
3541 int v3 = (nfsd->nd_flag & ND_NFSV3);
3543 struct mbuf *mb, *mb2, *mreq;
3544 struct vnode *vp = NULL;
3545 struct mount *mp = NULL;
3549 struct statfs statfs;
3552 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3553 fhp = &nfh.fh_generic;
3555 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
3556 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3558 nfsm_reply(NFSX_UNSIGNED);
3559 nfsm_srvpostop_attr(getret, &at);
3564 error = VFS_STATFS(vp->v_mount, sf, proc0.p_ucred);
3565 getret = VOP_GETATTR(vp, &at);
3568 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3570 nfsm_srvpostop_attr(getret, &at);
3575 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3577 tval = (u_quad_t)sf->f_blocks;
3578 tval *= (u_quad_t)sf->f_bsize;
3579 txdr_hyper(tval, &sfp->sf_tbytes);
3580 tval = (u_quad_t)sf->f_bfree;
3581 tval *= (u_quad_t)sf->f_bsize;
3582 txdr_hyper(tval, &sfp->sf_fbytes);
3583 tval = (u_quad_t)sf->f_bavail;
3584 tval *= (u_quad_t)sf->f_bsize;
3585 txdr_hyper(tval, &sfp->sf_abytes);
3586 sfp->sf_tfiles.nfsuquad[0] = 0;
3587 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
3588 sfp->sf_ffiles.nfsuquad[0] = 0;
3589 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3590 sfp->sf_afiles.nfsuquad[0] = 0;
3591 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3592 sfp->sf_invarsec = 0;
3594 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3595 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3596 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3597 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3598 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3607 * nfs fsinfo service
3610 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3611 struct thread *td, struct mbuf **mrq)
3613 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3614 struct sockaddr *nam = nfsd->nd_nam;
3615 caddr_t dpos = nfsd->nd_dpos;
3616 struct ucred *cred = &nfsd->nd_cr;
3618 struct nfsv3_fsinfo *sip;
3621 int error = 0, rdonly, getret = 1, pref;
3623 struct mbuf *mb, *mb2, *mreq;
3624 struct vnode *vp = NULL;
3625 struct mount *mp = NULL;
3632 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3633 fhp = &nfh.fh_generic;
3635 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
3636 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3638 nfsm_reply(NFSX_UNSIGNED);
3639 nfsm_srvpostop_attr(getret, &at);
3644 /* XXX Try to make a guess on the max file size. */
3645 VFS_STATFS(vp->v_mount, &sb, proc0.p_ucred);
3646 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3648 getret = VOP_GETATTR(vp, &at);
3651 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3652 nfsm_srvpostop_attr(getret, &at);
3653 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3657 * There should be file system VFS OP(s) to get this information.
3658 * For now, assume ufs.
3660 if (slp->ns_so->so_type == SOCK_DGRAM)
3661 pref = NFS_MAXDGRAMDATA;
3664 sip->fs_rtmax = txdr_unsigned(NFS_MAXDATA);
3665 sip->fs_rtpref = txdr_unsigned(pref);
3666 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3667 sip->fs_wtmax = txdr_unsigned(NFS_MAXDATA);
3668 sip->fs_wtpref = txdr_unsigned(pref);
3669 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3670 sip->fs_dtpref = txdr_unsigned(pref);
3671 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3672 sip->fs_timedelta.nfsv3_sec = 0;
3673 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3674 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3675 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3676 NFSV3FSINFO_CANSETTIME);
3684 * nfs pathconf service
3687 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3688 struct thread *td, struct mbuf **mrq)
3690 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3691 struct sockaddr *nam = nfsd->nd_nam;
3692 caddr_t dpos = nfsd->nd_dpos;
3693 struct ucred *cred = &nfsd->nd_cr;
3695 struct nfsv3_pathconf *pc;
3698 int error = 0, rdonly, getret = 1;
3699 register_t linkmax, namemax, chownres, notrunc;
3701 struct mbuf *mb, *mb2, *mreq;
3702 struct vnode *vp = NULL;
3703 struct mount *mp = NULL;
3708 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3709 fhp = &nfh.fh_generic;
3711 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
3712 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3714 nfsm_reply(NFSX_UNSIGNED);
3715 nfsm_srvpostop_attr(getret, &at);
3719 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3721 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3723 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3725 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc);
3726 getret = VOP_GETATTR(vp, &at);
3729 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3730 nfsm_srvpostop_attr(getret, &at);
3735 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3737 pc->pc_linkmax = txdr_unsigned(linkmax);
3738 pc->pc_namemax = txdr_unsigned(namemax);
3739 pc->pc_notrunc = txdr_unsigned(notrunc);
3740 pc->pc_chownrestricted = txdr_unsigned(chownres);
3743 * These should probably be supported by VOP_PATHCONF(), but
3744 * until msdosfs is exportable (why would you want to?), the
3745 * Unix defaults should be ok.
3747 pc->pc_caseinsensitive = nfs_false;
3748 pc->pc_casepreserving = nfs_true;
3756 * Null operation, used by clients to ping server
3760 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3761 struct thread *td, struct mbuf **mrq)
3763 struct mbuf *mrep = nfsd->nd_mrep;
3765 int error = NFSERR_RETVOID;
3766 struct mbuf *mb, *mreq;
3768 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3774 * No operation, used for obsolete procedures
3778 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3779 struct thread *td, struct mbuf **mrq)
3781 struct mbuf *mrep = nfsd->nd_mrep;
3784 struct mbuf *mb, *mreq;
3786 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3787 if (nfsd->nd_repstat)
3788 error = nfsd->nd_repstat;
3790 error = EPROCUNAVAIL;
3797 * Perform access checking for vnodes obtained from file handles that would
3798 * refer to files already opened by a Unix client. You cannot just use
3799 * vn_writechk() and VOP_ACCESS() for two reasons.
3800 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3801 * 2 - The owner is to be given access irrespective of mode bits for some
3802 * operations, so that processes that chmod after opening a file don't
3803 * break. I don't like this because it opens a security hole, but since
3804 * the nfs server opens a security hole the size of a barn door anyhow,
3807 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3808 * will return EPERM instead of EACCESS. EPERM is always an error.
3811 nfsrv_access(struct mount *mp, struct vnode *vp, int flags, struct ucred *cred,
3812 int rdonly, struct thread *td, int override)
3817 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3818 if (flags & VWRITE) {
3819 /* Just vn_writechk() changed to check rdonly */
3821 * Disallow write attempts on read-only file systems;
3822 * unless the file is a socket or a block or character
3823 * device resident on the file system.
3826 ((mp->mnt_flag | vp->v_mount->mnt_flag) & MNT_RDONLY)) {
3827 switch (vp->v_type) {
3837 * If there's shared text associated with
3838 * the inode, we can't allow writing.
3840 if (vp->v_flag & VTEXT)
3843 error = VOP_GETATTR(vp, &vattr);
3846 error = VOP_ACCESS(vp, flags, cred);
3848 * Allow certain operations for the owner (reads and writes
3849 * on files that are already open).
3851 if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3855 #endif /* NFS_NOSERVER */