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 1.46 2007/11/20 21:36:08 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>
69 #include <sys/nlookup.h>
70 #include <sys/namei.h>
71 #include <sys/unistd.h>
72 #include <sys/vnode.h>
73 #include <sys/mount.h>
74 #include <sys/socket.h>
75 #include <sys/socketvar.h>
76 #include <sys/malloc.h>
78 #include <sys/dirent.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_zone.h>
87 #include <vm/vm_object.h>
91 #include <sys/thread2.h>
97 #include "nfsm_subs.h"
100 #define nfsdbprintf(info) kprintf info
102 #define nfsdbprintf(info)
105 #define MAX_COMMIT_COUNT (1024 * 1024)
107 #define NUM_HEURISTIC 1017
108 #define NHUSE_INIT 64
110 #define NHUSE_MAX 2048
112 static struct nfsheur {
113 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
114 off_t nh_nextr; /* next offset for sequential detection */
115 int nh_use; /* use count for selection */
116 int nh_seqcount; /* heuristic */
117 } nfsheur[NUM_HEURISTIC];
119 nfstype nfsv3_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK,
122 nfstype nfsv2_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON,
125 extern u_int32_t nfs_xdrneg1;
126 extern u_int32_t nfs_false, nfs_true;
127 extern enum vtype nv3tov_type[8];
128 extern struct nfsstats nfsstats;
130 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
131 int nfsrvw_procrastinate_v3 = 0;
133 static struct timespec nfsver;
135 SYSCTL_DECL(_vfs_nfs);
137 static int nfs_async;
138 SYSCTL_INT(_vfs_nfs, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0, "");
139 static int nfs_commit_blks;
140 static int nfs_commit_miss;
141 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, "");
142 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, "");
144 static int nfsrv_access (struct vnode *,int,struct ucred *,int,
145 struct thread *, int);
146 static void nfsrvw_coalesce (struct nfsrv_descript *,
147 struct nfsrv_descript *);
150 * nfs v3 access service
153 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
154 struct thread *td, struct mbuf **mrq)
156 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
157 struct sockaddr *nam = nfsd->nd_nam;
158 caddr_t dpos = nfsd->nd_dpos;
159 struct ucred *cred = &nfsd->nd_cr;
160 struct vnode *vp = NULL;
166 int error = 0, rdonly, getret;
168 struct mbuf *mb, *mreq, *mb2;
169 struct vattr vattr, *vap = &vattr;
170 u_long testmode, nfsmode;
172 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
173 fhp = &nfh.fh_generic;
175 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
176 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
177 (nfsd->nd_flag & ND_KERBAUTH), TRUE);
179 nfsm_reply(NFSX_UNSIGNED);
180 nfsm_srvpostop_attr(1, (struct vattr *)0);
184 nfsmode = fxdr_unsigned(u_int32_t, *tl);
185 if ((nfsmode & NFSV3ACCESS_READ) &&
186 nfsrv_access(vp, VREAD, cred, rdonly, td, 0))
187 nfsmode &= ~NFSV3ACCESS_READ;
188 if (vp->v_type == VDIR)
189 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
192 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
193 if ((nfsmode & testmode) &&
194 nfsrv_access(vp, VWRITE, cred, rdonly, td, 0))
195 nfsmode &= ~testmode;
196 if (vp->v_type == VDIR)
197 testmode = NFSV3ACCESS_LOOKUP;
199 testmode = NFSV3ACCESS_EXECUTE;
200 if ((nfsmode & testmode) &&
201 nfsrv_access(vp, VEXEC, cred, rdonly, td, 0))
202 nfsmode &= ~testmode;
203 getret = VOP_GETATTR(vp, vap);
206 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
207 nfsm_srvpostop_attr(getret, vap);
208 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
209 *tl = txdr_unsigned(nfsmode);
217 * nfs getattr service
220 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
221 struct thread *td, struct mbuf **mrq)
223 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
224 struct sockaddr *nam = nfsd->nd_nam;
225 caddr_t dpos = nfsd->nd_dpos;
226 struct ucred *cred = &nfsd->nd_cr;
227 struct nfs_fattr *fp;
229 struct vattr *vap = &va;
230 struct vnode *vp = NULL;
236 int error = 0, rdonly;
238 struct mbuf *mb, *mb2, *mreq;
240 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
241 fhp = &nfh.fh_generic;
243 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
244 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
250 error = VOP_GETATTR(vp, vap);
253 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
258 nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
259 nfsm_srvfillattr(vap, fp);
269 * nfs setattr service
272 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
273 struct thread *td, struct mbuf **mrq)
275 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
276 struct sockaddr *nam = nfsd->nd_nam;
277 caddr_t dpos = nfsd->nd_dpos;
278 struct ucred *cred = &nfsd->nd_cr;
279 struct vattr va, preat;
280 struct vattr *vap = &va;
281 struct nfsv2_sattr *sp;
282 struct nfs_fattr *fp;
283 struct vnode *vp = NULL;
289 int error = 0, rdonly, preat_ret = 1, postat_ret = 1;
290 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
292 struct mbuf *mb, *mb2, *mreq;
293 struct timespec guard;
295 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
296 fhp = &nfh.fh_generic;
301 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
302 gcheck = fxdr_unsigned(int, *tl);
304 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
305 fxdr_nfsv3time(tl, &guard);
308 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
310 * Nah nah nah nah na nah
311 * There is a bug in the Sun client that puts 0xffff in the mode
312 * field of sattr when it should put in 0xffffffff. The u_short
313 * doesn't sign extend.
314 * --> check the low order 2 bytes for 0xffff
316 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
317 vap->va_mode = nfstov_mode(sp->sa_mode);
318 if (sp->sa_uid != nfs_xdrneg1)
319 vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
320 if (sp->sa_gid != nfs_xdrneg1)
321 vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
322 if (sp->sa_size != nfs_xdrneg1)
323 vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
324 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
326 fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
328 vap->va_atime.tv_sec =
329 fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
330 vap->va_atime.tv_nsec = 0;
333 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
334 fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
339 * Now that we have all the fields, lets do it.
341 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
342 (nfsd->nd_flag & ND_KERBAUTH), TRUE);
344 nfsm_reply(2 * NFSX_UNSIGNED);
345 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
351 * vp now an active resource, pay careful attention to cleanup
355 error = preat_ret = VOP_GETATTR(vp, &preat);
356 if (!error && gcheck &&
357 (preat.va_ctime.tv_sec != guard.tv_sec ||
358 preat.va_ctime.tv_nsec != guard.tv_nsec))
359 error = NFSERR_NOT_SYNC;
363 nfsm_reply(NFSX_WCCDATA(v3));
364 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
371 * If the size is being changed write acces is required, otherwise
372 * just check for a read only file system.
374 if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
375 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
380 if (vp->v_type == VDIR) {
383 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
388 error = VOP_SETATTR(vp, vap, cred);
389 postat_ret = VOP_GETATTR(vp, vap);
395 nfsm_reply(NFSX_WCCORFATTR(v3));
397 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
401 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
402 nfsm_srvfillattr(vap, fp);
416 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
417 struct thread *td, struct mbuf **mrq)
419 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
420 struct sockaddr *nam = nfsd->nd_nam;
421 caddr_t dpos = nfsd->nd_dpos;
422 struct ucred *cred = &nfsd->nd_cr;
423 struct nfs_fattr *fp;
424 struct nlookupdata nd;
434 int error = 0, len, dirattr_ret = 1;
435 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
437 struct mbuf *mb, *mb2, *mreq;
438 struct vattr va, dirattr, *vap = &va;
440 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
445 fhp = &nfh.fh_generic;
447 nfsm_srvnamesiz(len);
449 pubflag = nfs_ispublicfh(fhp);
451 error = nfs_namei(&nd, cred, NAMEI_LOOKUP, NULL, &vp,
452 fhp, len, slp, nam, &md, &dpos,
453 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
456 * namei failure, only dirp to cleanup. Clear out garbarge from
457 * structure in case macros jump to nfsmout.
463 dirattr_ret = VOP_GETATTR(dirp, &dirattr);
467 nfsm_reply(NFSX_POSTOPATTR(v3));
468 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
474 * Locate index file for public filehandle
476 * error is 0 on entry and 0 on exit from this block.
480 if (vp->v_type == VDIR && nfs_pub.np_index != NULL) {
482 * Setup call to lookup() to see if we can find
483 * the index file. Arguably, this doesn't belong
484 * in a kernel.. Ugh. If an error occurs, do not
485 * try to install an index file and then clear the
488 * When we replace nd with ind and redirect ndp,
489 * maintenance of ni_startdir and ni_vp shift to
490 * ind and we have to clean them up in the old nd.
491 * However, the cnd resource continues to be maintained
492 * via the original nd. Confused? You aren't alone!
495 cache_copy(&nd.nl_nch, &nch);
497 error = nlookup_init_raw(&nd, nfs_pub.np_index,
498 UIO_SYSSPACE, 0, cred, &nch);
501 error = nlookup(&nd);
505 * Found an index file. Get rid of
506 * the old references. transfer vp and
507 * load up the new vp. Fortunately we do
508 * not have to deal with dvp, that would be
515 error = cache_vget(&nd.nl_nch, nd.nl_cred,
517 KKASSERT(error == 0);
522 * If the public filehandle was used, check that this lookup
523 * didn't result in a filehandle outside the publicly exported
524 * filesystem. We clear the poor vp here to avoid lockups due
528 if (vp->v_mount != nfs_pub.np_mount) {
537 dirattr_ret = VOP_GETATTR(dirp, &dirattr);
543 * Resources at this point:
544 * ndp->ni_vp may not be NULL
549 nfsm_reply(NFSX_POSTOPATTR(v3));
550 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
556 * Clear out some resources prior to potentially blocking. This
557 * is not as critical as ni_dvp resources in other routines, but
563 * Get underlying attribute, then release remaining resources ( for
564 * the same potential blocking reason ) and reply.
566 bzero((caddr_t)fhp, sizeof(nfh));
567 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
568 error = VFS_VPTOFH(vp, &fhp->fh_fid);
570 error = VOP_GETATTR(vp, vap);
574 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
576 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
580 nfsm_srvfhtom(fhp, v3);
582 nfsm_srvpostop_attr(0, vap);
583 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
585 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
586 nfsm_srvfillattr(vap, fp);
592 nlookup_done(&nd); /* may be called twice */
599 * nfs readlink service
602 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
603 struct thread *td, struct mbuf **mrq)
605 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
606 struct sockaddr *nam = nfsd->nd_nam;
607 caddr_t dpos = nfsd->nd_dpos;
608 struct ucred *cred = &nfsd->nd_cr;
609 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
610 struct iovec *ivp = iv;
615 int error = 0, rdonly, i, tlen, len, getret;
616 int v3 = (nfsd->nd_flag & ND_NFSV3);
618 struct mbuf *mb, *mb2, *mp2, *mp3, *mreq;
619 struct vnode *vp = NULL;
623 struct uio io, *uiop = &io;
625 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
627 mp2 = (struct mbuf *)0;
630 fhp = &nfh.fh_generic;
634 while (len < NFS_MAXPATHLEN) {
635 mp = m_getcl(MB_WAIT, MT_DATA, 0);
636 mp->m_len = MCLBYTES;
643 if ((len+mp->m_len) > NFS_MAXPATHLEN) {
644 mp->m_len = NFS_MAXPATHLEN-len;
645 len = NFS_MAXPATHLEN;
648 ivp->iov_base = mtod(mp, caddr_t);
649 ivp->iov_len = mp->m_len;
654 uiop->uio_iovcnt = i;
655 uiop->uio_offset = 0;
656 uiop->uio_resid = len;
657 uiop->uio_rw = UIO_READ;
658 uiop->uio_segflg = UIO_SYSSPACE;
660 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
661 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
663 nfsm_reply(2 * NFSX_UNSIGNED);
664 nfsm_srvpostop_attr(1, (struct vattr *)0);
668 if (vp->v_type != VLNK) {
675 error = VOP_READLINK(vp, uiop, cred);
677 getret = VOP_GETATTR(vp, &attr);
680 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
682 nfsm_srvpostop_attr(getret, &attr);
688 if (uiop->uio_resid > 0) {
689 len -= uiop->uio_resid;
690 tlen = nfsm_rndup(len);
691 nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
693 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
694 *tl = txdr_unsigned(len);
709 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
710 struct thread *td, struct mbuf **mrq)
712 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
713 struct sockaddr *nam = nfsd->nd_nam;
714 caddr_t dpos = nfsd->nd_dpos;
715 struct ucred *cred = &nfsd->nd_cr;
719 struct nfs_fattr *fp;
724 int error = 0, rdonly, cnt, len, left, siz, tlen, getret;
725 int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
727 struct mbuf *mb, *mb2, *mreq;
729 struct vnode *vp = NULL;
732 struct uio io, *uiop = &io;
733 struct vattr va, *vap = &va;
738 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
739 fhp = &nfh.fh_generic;
742 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
743 off = fxdr_hyper(tl);
745 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
746 off = (off_t)fxdr_unsigned(u_int32_t, *tl);
748 nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
751 * Reference vp. If an error occurs, vp will be invalid, but we
752 * have to NULL it just in case. The macros might goto nfsmout
756 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
757 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
760 nfsm_reply(2 * NFSX_UNSIGNED);
761 nfsm_srvpostop_attr(1, (struct vattr *)0);
766 if (vp->v_type != VREG) {
770 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
773 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, td, 1)) != 0)
774 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 1);
776 getret = VOP_GETATTR(vp, vap);
782 nfsm_reply(NFSX_POSTOPATTR(v3));
783 nfsm_srvpostop_attr(getret, vap);
789 * Calculate byte count to read
792 if (off >= vap->va_size)
794 else if ((off + reqlen) > vap->va_size)
795 cnt = vap->va_size - off;
800 * Calculate seqcount for heuristic
808 * Locate best candidate
811 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
815 if (nfsheur[hi].nh_vp == vp) {
819 if (nfsheur[hi].nh_use > 0)
820 --nfsheur[hi].nh_use;
821 hi = (hi + 1) % NUM_HEURISTIC;
822 if (nfsheur[hi].nh_use < nh->nh_use)
826 if (nh->nh_vp != vp) {
829 nh->nh_use = NHUSE_INIT;
837 * Calculate heuristic
840 if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
841 if (++nh->nh_seqcount > IO_SEQMAX)
842 nh->nh_seqcount = IO_SEQMAX;
843 } else if (nh->nh_seqcount > 1) {
848 nh->nh_use += NHUSE_INC;
849 if (nh->nh_use > NHUSE_MAX)
850 nh->nh_use = NHUSE_MAX;
851 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
854 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
856 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
858 fp = (struct nfs_fattr *)tl;
859 tl += (NFSX_V3FATTR / sizeof (u_int32_t));
861 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
862 fp = (struct nfs_fattr *)tl;
863 tl += (NFSX_V2FATTR / sizeof (u_int32_t));
865 len = left = nfsm_rndup(cnt);
868 * Generate the mbuf list with the uio_iov ref. to it.
873 siz = min(M_TRAILINGSPACE(m), left);
879 m = m_getcl(MB_WAIT, MT_DATA, 0);
885 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
887 uiop->uio_iov = iv2 = iv;
893 panic("nfsrv_read iov");
894 siz = min(M_TRAILINGSPACE(m), left);
896 iv->iov_base = mtod(m, caddr_t) + m->m_len;
905 uiop->uio_iovcnt = i;
906 uiop->uio_offset = off;
907 uiop->uio_resid = len;
908 uiop->uio_rw = UIO_READ;
909 uiop->uio_segflg = UIO_SYSSPACE;
910 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
911 off = uiop->uio_offset;
913 FREE((caddr_t)iv2, M_TEMP);
914 if (error || (getret = VOP_GETATTR(vp, vap))) {
920 nfsm_reply(NFSX_POSTOPATTR(v3));
921 nfsm_srvpostop_attr(getret, vap);
930 nfsm_srvfillattr(vap, fp);
931 tlen = len - uiop->uio_resid;
932 cnt = cnt < tlen ? cnt : tlen;
933 tlen = nfsm_rndup(cnt);
934 if (len != tlen || tlen != cnt)
935 nfsm_adj(mb, len - tlen, tlen - cnt);
937 *tl++ = txdr_unsigned(cnt);
943 *tl = txdr_unsigned(cnt);
954 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
955 struct thread *td, struct mbuf **mrq)
957 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
958 struct sockaddr *nam = nfsd->nd_nam;
959 caddr_t dpos = nfsd->nd_dpos;
960 struct ucred *cred = &nfsd->nd_cr;
964 struct nfs_fattr *fp;
966 struct vattr va, forat;
967 struct vattr *vap = &va;
971 int error = 0, rdonly, len, forat_ret = 1;
972 int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
973 int stable = NFSV3WRITE_FILESYNC;
974 int v3 = (nfsd->nd_flag & ND_NFSV3);
976 struct mbuf *mb, *mb2, *mreq;
977 struct vnode *vp = NULL;
980 struct uio io, *uiop = &io;
983 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
989 fhp = &nfh.fh_generic;
992 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
993 off = fxdr_hyper(tl);
995 stable = fxdr_unsigned(int, *tl++);
997 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
998 off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1001 stable = NFSV3WRITE_UNSTABLE;
1003 retlen = len = fxdr_unsigned(int32_t, *tl);
1007 * For NFS Version 2, it is not obvious what a write of zero length
1008 * should do, but I might as well be consistent with Version 3,
1009 * which is to return ok so long as there are no permission problems.
1017 adjust = dpos - mtod(mp, caddr_t);
1018 mp->m_len -= adjust;
1019 if (mp->m_len > 0 && adjust > 0)
1020 NFSMADV(mp, adjust);
1024 else if (mp->m_len > 0) {
1027 mp->m_len -= (i - len);
1036 if (len > NFS_MAXDATA || len < 0 || i < len) {
1038 nfsm_reply(2 * NFSX_UNSIGNED);
1039 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1043 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
1044 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
1047 nfsm_reply(2 * NFSX_UNSIGNED);
1048 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1053 forat_ret = VOP_GETATTR(vp, &forat);
1054 if (vp->v_type != VREG) {
1058 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1061 error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
1066 nfsm_reply(NFSX_WCCDATA(v3));
1067 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1073 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
1075 uiop->uio_iov = iv = ivp;
1076 uiop->uio_iovcnt = cnt;
1079 if (mp->m_len > 0) {
1080 ivp->iov_base = mtod(mp, caddr_t);
1081 ivp->iov_len = mp->m_len;
1089 * The IO_METASYNC flag indicates that all metadata (and not just
1090 * enough to ensure data integrity) mus be written to stable storage
1092 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1094 if (stable == NFSV3WRITE_UNSTABLE)
1095 ioflags = IO_NODELOCKED;
1096 else if (stable == NFSV3WRITE_DATASYNC)
1097 ioflags = (IO_SYNC | IO_NODELOCKED);
1099 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1100 uiop->uio_resid = len;
1101 uiop->uio_rw = UIO_WRITE;
1102 uiop->uio_segflg = UIO_SYSSPACE;
1103 uiop->uio_td = NULL;
1104 uiop->uio_offset = off;
1105 error = VOP_WRITE(vp, uiop, ioflags, cred);
1106 nfsstats.srvvop_writes++;
1107 FREE((caddr_t)iv, M_TEMP);
1109 aftat_ret = VOP_GETATTR(vp, vap);
1114 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1115 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1117 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1122 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1123 *tl++ = txdr_unsigned(retlen);
1125 * If nfs_async is set, then pretend the write was FILESYNC.
1127 if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1128 *tl++ = txdr_unsigned(stable);
1130 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1132 * Actually, there is no need to txdr these fields,
1133 * but it may make the values more human readable,
1134 * for debugging purposes.
1136 if (nfsver.tv_sec == 0)
1138 *tl++ = txdr_unsigned(nfsver.tv_sec);
1139 *tl = txdr_unsigned(nfsver.tv_nsec / 1000);
1141 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1142 nfsm_srvfillattr(vap, fp);
1151 * NFS write service with write gathering support. Called when
1152 * nfsrvw_procrastinate > 0.
1153 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1154 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1158 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp,
1159 struct thread *td, struct mbuf **mrq)
1163 struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1164 struct nfs_fattr *fp;
1167 struct nfsrvw_delayhash *wpp;
1169 struct vattr va, forat;
1173 int error = 0, rdonly, len, forat_ret = 1;
1174 int ioflags, aftat_ret = 1, adjust, v3, zeroing;
1176 struct mbuf *mb, *mb2, *mreq, *mrep, *md;
1177 struct vnode *vp = NULL;
1178 struct uio io, *uiop = &io;
1181 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1190 mrep = nfsd->nd_mrep;
1192 dpos = nfsd->nd_dpos;
1193 cred = &nfsd->nd_cr;
1194 v3 = (nfsd->nd_flag & ND_NFSV3);
1195 LIST_INIT(&nfsd->nd_coalesce);
1196 nfsd->nd_mreq = NULL;
1197 nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1198 cur_usec = nfs_curusec();
1199 nfsd->nd_time = cur_usec +
1200 (v3 ? nfsrvw_procrastinate_v3 : nfsrvw_procrastinate);
1203 * Now, get the write header..
1205 nfsm_srvmtofh(&nfsd->nd_fh);
1207 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1208 nfsd->nd_off = fxdr_hyper(tl);
1210 nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1212 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1213 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1216 nfsd->nd_stable = NFSV3WRITE_UNSTABLE;
1218 len = fxdr_unsigned(int32_t, *tl);
1220 nfsd->nd_eoff = nfsd->nd_off + len;
1223 * Trim the header out of the mbuf list and trim off any trailing
1224 * junk so that the mbuf list has only the write data.
1232 adjust = dpos - mtod(mp, caddr_t);
1233 mp->m_len -= adjust;
1234 if (mp->m_len > 0 && adjust > 0)
1235 NFSMADV(mp, adjust);
1242 mp->m_len -= (i - len);
1248 if (len > NFS_MAXDATA || len < 0 || i < len) {
1252 nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1254 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1255 nfsd->nd_mreq = mreq;
1256 nfsd->nd_mrep = NULL;
1261 * Add this entry to the hash and time queues.
1265 wp = slp->ns_tq.lh_first;
1266 while (wp && wp->nd_time < nfsd->nd_time) {
1268 wp = wp->nd_tq.le_next;
1270 NFS_DPF(WG, ("Q%03x", nfsd->nd_retxid & 0xfff));
1272 LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1274 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1276 if (nfsd->nd_mrep) {
1277 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1281 bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
1283 wp = wp->nd_hash.le_next;
1285 while (wp && wp->nd_off < nfsd->nd_off &&
1286 !bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
1288 wp = wp->nd_hash.le_next;
1291 LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1294 * Search the hash list for overlapping entries and
1297 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1298 wp = nfsd->nd_hash.le_next;
1299 if (NFSW_SAMECRED(owp, nfsd))
1300 nfsrvw_coalesce(owp, nfsd);
1303 LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1310 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1311 * and generate the associated reply mbuf list(s).
1314 cur_usec = nfs_curusec();
1316 for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = owp) {
1317 owp = nfsd->nd_tq.le_next;
1318 if (nfsd->nd_time > cur_usec)
1322 NFS_DPF(WG, ("P%03x", nfsd->nd_retxid & 0xfff));
1323 LIST_REMOVE(nfsd, nd_tq);
1324 LIST_REMOVE(nfsd, nd_hash);
1326 mrep = nfsd->nd_mrep;
1327 nfsd->nd_mrep = NULL;
1328 cred = &nfsd->nd_cr;
1329 v3 = (nfsd->nd_flag & ND_NFSV3);
1330 forat_ret = aftat_ret = 1;
1331 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1332 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
1335 forat_ret = VOP_GETATTR(vp, &forat);
1336 if (vp->v_type != VREG) {
1340 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1346 error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
1349 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1350 ioflags = IO_NODELOCKED;
1351 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1352 ioflags = (IO_SYNC | IO_NODELOCKED);
1354 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1355 uiop->uio_rw = UIO_WRITE;
1356 uiop->uio_segflg = UIO_SYSSPACE;
1357 uiop->uio_td = NULL;
1358 uiop->uio_offset = nfsd->nd_off;
1359 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1360 if (uiop->uio_resid > 0) {
1368 uiop->uio_iovcnt = i;
1369 MALLOC(iov, struct iovec *, i * sizeof (struct iovec),
1371 uiop->uio_iov = ivp = iov;
1374 if (mp->m_len > 0) {
1375 ivp->iov_base = mtod(mp, caddr_t);
1376 ivp->iov_len = mp->m_len;
1382 error = VOP_WRITE(vp, uiop, ioflags, cred);
1383 nfsstats.srvvop_writes++;
1385 FREE((caddr_t)iov, M_TEMP);
1389 aftat_ret = VOP_GETATTR(vp, &va);
1395 * Loop around generating replies for all write rpcs that have
1396 * now been completed.
1400 NFS_DPF(WG, ("R%03x", nfsd->nd_retxid & 0xfff));
1402 nfsm_writereply(NFSX_WCCDATA(v3), v3);
1404 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1407 nfsm_writereply(NFSX_PREOPATTR(v3) +
1408 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1409 NFSX_WRITEVERF(v3), v3);
1411 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1412 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1413 *tl++ = txdr_unsigned(nfsd->nd_len);
1414 *tl++ = txdr_unsigned(swp->nd_stable);
1416 * Actually, there is no need to txdr these fields,
1417 * but it may make the values more human readable,
1418 * for debugging purposes.
1420 if (nfsver.tv_sec == 0)
1422 *tl++ = txdr_unsigned(nfsver.tv_sec);
1423 *tl = txdr_unsigned(nfsver.tv_nsec / 1000);
1425 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1426 nfsm_srvfillattr(&va, fp);
1429 nfsd->nd_mreq = mreq;
1431 panic("nfsrv_write: nd_mrep not free");
1434 * Done. Put it at the head of the timer queue so that
1435 * the final phase can return the reply.
1440 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1442 nfsd = swp->nd_coalesce.lh_first;
1444 LIST_REMOVE(nfsd, nd_tq);
1450 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1457 * Search for a reply to return.
1460 for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = nfsd->nd_tq.le_next)
1461 if (nfsd->nd_mreq) {
1462 NFS_DPF(WG, ("X%03x", nfsd->nd_retxid & 0xfff));
1463 LIST_REMOVE(nfsd, nd_tq);
1464 *mrq = nfsd->nd_mreq;
1473 * Coalesce the write request nfsd into owp. To do this we must:
1474 * - remove nfsd from the queues
1475 * - merge nfsd->nd_mrep into owp->nd_mrep
1476 * - update the nd_eoff and nd_stable for owp
1477 * - put nfsd on owp's nd_coalesce list
1478 * NB: Must be called at splsoftclock().
1481 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1485 struct nfsrv_descript *p;
1487 NFS_DPF(WG, ("C%03x-%03x",
1488 nfsd->nd_retxid & 0xfff, owp->nd_retxid & 0xfff));
1489 LIST_REMOVE(nfsd, nd_hash);
1490 LIST_REMOVE(nfsd, nd_tq);
1491 if (owp->nd_eoff < nfsd->nd_eoff) {
1492 overlap = owp->nd_eoff - nfsd->nd_off;
1494 panic("nfsrv_coalesce: bad off");
1496 m_adj(nfsd->nd_mrep, overlap);
1500 mp->m_next = nfsd->nd_mrep;
1501 owp->nd_eoff = nfsd->nd_eoff;
1503 m_freem(nfsd->nd_mrep);
1504 nfsd->nd_mrep = NULL;
1505 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1506 owp->nd_stable = NFSV3WRITE_FILESYNC;
1507 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1508 owp->nd_stable == NFSV3WRITE_UNSTABLE)
1509 owp->nd_stable = NFSV3WRITE_DATASYNC;
1510 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1513 * If nfsd had anything else coalesced into it, transfer them
1514 * to owp, otherwise their replies will never get sent.
1516 for (p = nfsd->nd_coalesce.lh_first; p;
1517 p = nfsd->nd_coalesce.lh_first) {
1518 LIST_REMOVE(p, nd_tq);
1519 LIST_INSERT_HEAD(&owp->nd_coalesce, p, nd_tq);
1524 * nfs create service
1525 * now does a truncate to 0 length via. setattr if it already exists
1528 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1529 struct thread *td, struct mbuf **mrq)
1531 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1532 struct sockaddr *nam = nfsd->nd_nam;
1533 caddr_t dpos = nfsd->nd_dpos;
1534 struct ucred *cred = &nfsd->nd_cr;
1535 struct nfs_fattr *fp;
1536 struct vattr va, dirfor, diraft;
1537 struct vattr *vap = &va;
1538 struct nfsv2_sattr *sp;
1540 struct nlookupdata nd;
1543 int error = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1544 udev_t rdev = NOUDEV;
1545 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1548 struct mbuf *mb, *mb2, *mreq;
1555 u_char cverf[NFSX_V3CREATEVERF];
1557 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1563 fhp = &nfh.fh_generic;
1565 nfsm_srvnamesiz(len);
1568 * Call namei and do initial cleanup to get a few things
1569 * out of the way. If we get an initial error we cleanup
1570 * and return here to avoid special-casing the invalid nd
1571 * structure through the rest of the case. dirp may be
1572 * set even if an error occurs, but the nd structure will not
1573 * be valid at all if an error occurs so we have to invalidate it
1574 * prior to calling nfsm_reply ( which might goto nfsmout ).
1576 error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
1577 fhp, len, slp, nam, &md, &dpos, &dirp,
1578 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1581 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
1588 nfsm_reply(NFSX_WCCDATA(v3));
1589 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1595 * No error. Continue. State:
1598 * vp may be valid or NULL if the target does not
1602 * The error state is set through the code and we may also do some
1603 * opportunistic releasing of vnodes to avoid holding locks through
1604 * NFS I/O. The cleanup at the end is a catch-all
1609 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1610 how = fxdr_unsigned(int, *tl);
1612 case NFSV3CREATE_GUARDED:
1618 case NFSV3CREATE_UNCHECKED:
1621 case NFSV3CREATE_EXCLUSIVE:
1622 nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1623 bcopy(cp, cverf, NFSX_V3CREATEVERF);
1627 vap->va_type = VREG;
1629 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1630 vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1631 if (vap->va_type == VNON)
1632 vap->va_type = VREG;
1633 vap->va_mode = nfstov_mode(sp->sa_mode);
1634 switch (vap->va_type) {
1636 tsize = fxdr_unsigned(int32_t, sp->sa_size);
1638 vap->va_size = (u_quad_t)tsize;
1643 rdev = fxdr_unsigned(long, sp->sa_size);
1651 * Iff doesn't exist, create it
1652 * otherwise just truncate to 0 length
1653 * should I set the mode too ?
1655 * The only possible error we can have at this point is EEXIST.
1656 * nd.ni_vp will also be non-NULL in that case.
1659 if (vap->va_mode == (mode_t)VNOVAL)
1661 if (vap->va_type == VREG || vap->va_type == VSOCK) {
1663 error = VOP_NCREATE(&nd.nl_nch, dvp, &vp,
1668 if (exclusive_flag) {
1671 bcopy(cverf, (caddr_t)&vap->va_atime,
1673 error = VOP_SETATTR(vp, vap, cred);
1677 vap->va_type == VCHR ||
1678 vap->va_type == VBLK ||
1679 vap->va_type == VFIFO
1682 * Handle SysV FIFO node special cases. All other
1683 * devices require super user to access.
1685 if (vap->va_type == VCHR && rdev == 0xffffffff)
1686 vap->va_type = VFIFO;
1687 if (vap->va_type != VFIFO &&
1688 (error = suser_cred(cred, 0))) {
1691 vap->va_rmajor = umajor(rdev);
1692 vap->va_rminor = uminor(rdev);
1695 error = VOP_NMKNOD(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
1702 * XXX what is this junk supposed to do ?
1709 * release dvp prior to lookup
1717 * Even though LOCKPARENT was cleared, ni_dvp may
1720 nd.ni_cnd.cn_nameiop = NAMEI_LOOKUP;
1721 nd.ni_cnd.cn_flags &= ~(CNP_LOCKPARENT);
1722 nd.ni_cnd.cn_td = td;
1723 nd.ni_cnd.cn_cred = cred;
1725 error = lookup(&nd);
1730 /* fall through on certain errors */
1732 nfsrv_object_create(nd.ni_vp);
1733 if (nd.ni_cnd.cn_flags & CNP_ISSYMLINK) {
1742 if (vap->va_size != -1) {
1743 error = nfsrv_access(vp, VWRITE, cred,
1744 (nd.nl_flags & NLC_NFS_RDONLY), td, 0);
1746 tempsize = vap->va_size;
1748 vap->va_size = tempsize;
1749 error = VOP_SETATTR(vp, vap, cred);
1755 bzero((caddr_t)fhp, sizeof(nfh));
1756 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1757 error = VFS_VPTOFH(vp, &fhp->fh_fid);
1759 error = VOP_GETATTR(vp, vap);
1762 if (exclusive_flag && !error &&
1763 bcmp(cverf, (caddr_t)&vap->va_atime, NFSX_V3CREATEVERF))
1765 diraft_ret = VOP_GETATTR(dirp, &diraft);
1769 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1772 nfsm_srvpostop_fh(fhp);
1773 nfsm_srvpostop_attr(0, vap);
1775 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1778 nfsm_srvfhtom(fhp, v3);
1779 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1780 nfsm_srvfillattr(vap, fp);
1805 * nfs v3 mknod service
1808 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1809 struct thread *td, struct mbuf **mrq)
1811 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1812 struct sockaddr *nam = nfsd->nd_nam;
1813 caddr_t dpos = nfsd->nd_dpos;
1814 struct ucred *cred = &nfsd->nd_cr;
1815 struct vattr va, dirfor, diraft;
1816 struct vattr *vap = &va;
1818 struct nlookupdata nd;
1821 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1824 struct mbuf *mb, *mb2, *mreq;
1831 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1837 fhp = &nfh.fh_generic;
1839 nfsm_srvnamesiz(len);
1842 * Handle nfs_namei() call. If an error occurs, the nd structure
1843 * is not valid. However, nfsm_*() routines may still jump to
1847 error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
1848 fhp, len, slp, nam, &md, &dpos, &dirp,
1849 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1851 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
1853 nfsm_reply(NFSX_WCCDATA(1));
1854 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1858 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1859 vtyp = nfsv3tov_type(*tl);
1860 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1861 error = NFSERR_BADTYPE;
1866 if (vtyp == VCHR || vtyp == VBLK) {
1867 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1868 vap->va_rmajor = fxdr_unsigned(u_int32_t, *tl++);
1869 vap->va_rminor = fxdr_unsigned(u_int32_t, *tl);
1873 * Iff doesn't exist, create it.
1879 vap->va_type = vtyp;
1880 if (vap->va_mode == (mode_t)VNOVAL)
1882 if (vtyp == VSOCK) {
1884 error = VOP_NCREATE(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
1888 if (vtyp != VFIFO && (error = suser_cred(cred, 0)))
1892 error = VOP_NMKNOD(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
1900 * send response, cleanup, return.
1912 bzero((caddr_t)fhp, sizeof(nfh));
1913 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1914 error = VFS_VPTOFH(vp, &fhp->fh_fid);
1916 error = VOP_GETATTR(vp, vap);
1922 diraft_ret = VOP_GETATTR(dirp, &diraft);
1927 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1929 nfsm_srvpostop_fh(fhp);
1930 nfsm_srvpostop_attr(0, vap);
1932 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1950 * nfs remove service
1953 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1954 struct thread *td, struct mbuf **mrq)
1956 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1957 struct sockaddr *nam = nfsd->nd_nam;
1958 caddr_t dpos = nfsd->nd_dpos;
1959 struct ucred *cred = &nfsd->nd_cr;
1960 struct nlookupdata nd;
1964 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1965 int v3 = (nfsd->nd_flag & ND_NFSV3);
1967 struct mbuf *mb, *mreq;
1971 struct vattr dirfor, diraft;
1975 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1981 fhp = &nfh.fh_generic;
1983 nfsm_srvnamesiz(len);
1985 error = nfs_namei(&nd, cred, NAMEI_DELETE, &dvp, &vp,
1986 fhp, len, slp, nam, &md, &dpos, &dirp,
1987 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1990 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
1993 if (vp->v_type == VDIR) {
1994 error = EPERM; /* POSIX */
1998 * The root of a mounted filesystem cannot be deleted.
2000 if (vp->v_flag & VROOT) {
2012 error = VOP_NREMOVE(&nd.nl_nch, dvp, nd.nl_cred);
2018 diraft_ret = VOP_GETATTR(dirp, &diraft);
2019 nfsm_reply(NFSX_WCCDATA(v3));
2021 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2040 * nfs rename service
2043 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2044 struct thread *td, struct mbuf **mrq)
2046 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2047 struct sockaddr *nam = nfsd->nd_nam;
2048 caddr_t dpos = nfsd->nd_dpos;
2049 struct ucred *cred = &nfsd->nd_cr;
2053 int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
2054 int tdirfor_ret = 1, tdiraft_ret = 1;
2055 int v3 = (nfsd->nd_flag & ND_NFSV3);
2057 struct mbuf *mb, *mreq;
2058 struct nlookupdata fromnd, tond;
2059 struct vnode *fvp, *fdirp, *fdvp;
2060 struct vnode *tvp, *tdirp, *tdvp;
2061 struct namecache *ncp;
2062 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
2064 fhandle_t *ffhp, *tfhp;
2067 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2069 fvp = (struct vnode *)0;
2071 ffhp = &fnfh.fh_generic;
2072 tfhp = &tnfh.fh_generic;
2075 * Clear fields incase goto nfsmout occurs from macro.
2078 nlookup_zero(&fromnd);
2079 nlookup_zero(&tond);
2083 nfsm_srvmtofh(ffhp);
2084 nfsm_srvnamesiz(len);
2086 * Remember our original uid so that we can reset cr_uid before
2087 * the second nfs_namei() call, in case it is remapped.
2089 saved_uid = cred->cr_uid;
2090 error = nfs_namei(&fromnd, cred, NAMEI_DELETE, NULL, NULL,
2091 ffhp, len, slp, nam, &md, &dpos, &fdirp,
2092 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2095 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor);
2098 nfsm_reply(2 * NFSX_WCCDATA(v3));
2099 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2100 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2106 * We have to unlock the from ncp before we can safely lookup
2109 KKASSERT(fromnd.nl_flags & NLC_NCPISLOCKED);
2110 cache_unlock(&fromnd.nl_nch);
2111 fromnd.nl_flags &= ~NLC_NCPISLOCKED;
2112 nfsm_srvmtofh(tfhp);
2113 nfsm_strsiz(len2, NFS_MAXNAMLEN);
2114 cred->cr_uid = saved_uid;
2116 error = nfs_namei(&tond, cred, NAMEI_RENAME, NULL, NULL,
2117 tfhp, len2, slp, nam, &md, &dpos, &tdirp,
2118 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2121 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor);
2129 if (cache_lock_nonblock(&fromnd.nl_nch) == 0) {
2130 cache_resolve(&fromnd.nl_nch, fromnd.nl_cred);
2131 } else if (fromnd.nl_nch.ncp > tond.nl_nch.ncp) {
2132 cache_lock(&fromnd.nl_nch);
2133 cache_resolve(&fromnd.nl_nch, fromnd.nl_cred);
2135 cache_unlock(&tond.nl_nch);
2136 cache_lock(&fromnd.nl_nch);
2137 cache_resolve(&fromnd.nl_nch, fromnd.nl_cred);
2138 cache_lock(&tond.nl_nch);
2139 cache_resolve(&tond.nl_nch, tond.nl_cred);
2141 fromnd.nl_flags |= NLC_NCPISLOCKED;
2143 fvp = fromnd.nl_nch.ncp->nc_vp;
2144 tvp = tond.nl_nch.ncp->nc_vp;
2147 * Set fdvp and tdvp. We haven't done all the topology checks
2148 * so these can wind up NULL (e.g. if either fvp or tvp is a mount
2149 * point). If we get through the checks these will be guarenteed
2152 * Holding the children ncp's should be sufficient to prevent
2153 * fdvp and tdvp ripouts.
2155 if (fromnd.nl_nch.ncp->nc_parent)
2156 fdvp = fromnd.nl_nch.ncp->nc_parent->nc_vp;
2159 if (tond.nl_nch.ncp->nc_parent)
2160 tdvp = tond.nl_nch.ncp->nc_parent->nc_vp;
2165 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2171 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2178 if (tvp->v_type == VDIR && (tond.nl_nch.ncp->nc_flag & NCF_ISMOUNTPT)) {
2186 if (fvp->v_type == VDIR && (fromnd.nl_nch.ncp->nc_flag & NCF_ISMOUNTPT)) {
2193 if (fromnd.nl_nch.mount != tond.nl_nch.mount) {
2200 if (fromnd.nl_nch.ncp == tond.nl_nch.ncp->nc_parent) {
2208 * You cannot rename a source into itself or a subdirectory of itself.
2209 * We check this by travsering the target directory upwards looking
2210 * for a match against the source.
2213 for (ncp = tond.nl_nch.ncp; ncp; ncp = ncp->nc_parent) {
2214 if (fromnd.nl_nch.ncp == ncp) {
2222 * If source is the same as the destination (that is the
2223 * same vnode with the same name in the same directory),
2224 * then there is nothing to do.
2226 if (fromnd.nl_nch.ncp == tond.nl_nch.ncp)
2231 * The VOP_NRENAME function releases all vnode references &
2232 * locks prior to returning so we need to clear the pointers
2233 * to bypass cleanup code later on.
2235 error = VOP_NRENAME(&fromnd.nl_nch, &tond.nl_nch,
2236 fdvp, tdvp, tond.nl_cred);
2245 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft);
2247 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft);
2248 nfsm_reply(2 * NFSX_WCCDATA(v3));
2250 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2251 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2259 nlookup_done(&tond);
2262 nlookup_done(&fromnd);
2270 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2271 struct thread *td, struct mbuf **mrq)
2273 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2274 struct sockaddr *nam = nfsd->nd_nam;
2275 caddr_t dpos = nfsd->nd_dpos;
2276 struct ucred *cred = &nfsd->nd_cr;
2277 struct nlookupdata nd;
2281 int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1;
2282 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2284 struct mbuf *mb, *mreq;
2289 struct vattr dirfor, diraft, at;
2291 fhandle_t *fhp, *dfhp;
2293 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2295 dirp = dvp = vp = xp = NULL;
2297 fhp = &nfh.fh_generic;
2298 dfhp = &dnfh.fh_generic;
2300 nfsm_srvmtofh(dfhp);
2301 nfsm_srvnamesiz(len);
2303 error = nfsrv_fhtovp(fhp, FALSE, &xp, cred, slp, nam,
2304 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
2306 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2307 nfsm_srvpostop_attr(getret, &at);
2308 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2313 if (xp->v_type == VDIR) {
2314 error = EPERM; /* POSIX */
2318 error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
2319 dfhp, len, slp, nam, &md, &dpos, &dirp,
2320 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2323 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2332 if (xp->v_mount != dvp->v_mount)
2337 error = VOP_NLINK(&nd.nl_nch, dvp, xp, nd.nl_cred);
2345 getret = VOP_GETATTR(xp, &at);
2347 diraft_ret = VOP_GETATTR(dirp, &diraft);
2348 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2350 nfsm_srvpostop_attr(getret, &at);
2351 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2374 * nfs symbolic link service
2377 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2378 struct thread *td, struct mbuf **mrq)
2380 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2381 struct sockaddr *nam = nfsd->nd_nam;
2382 caddr_t dpos = nfsd->nd_dpos;
2383 struct ucred *cred = &nfsd->nd_cr;
2384 struct vattr va, dirfor, diraft;
2385 struct nlookupdata nd;
2386 struct vattr *vap = &va;
2389 struct nfsv2_sattr *sp;
2390 char *bpos, *pathcp = (char *)0, *cp2;
2393 int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2394 int v3 = (nfsd->nd_flag & ND_NFSV3);
2395 struct mbuf *mb, *mreq, *mb2;
2402 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2408 fhp = &nfh.fh_generic;
2410 nfsm_srvnamesiz(len);
2412 error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
2413 fhp, len, slp, nam, &md, &dpos, &dirp,
2414 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2417 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2425 nfsm_strsiz(len2, NFS_MAXPATHLEN);
2426 MALLOC(pathcp, caddr_t, len2 + 1, M_TEMP, M_WAITOK);
2427 iv.iov_base = pathcp;
2429 io.uio_resid = len2;
2433 io.uio_segflg = UIO_SYSSPACE;
2434 io.uio_rw = UIO_READ;
2436 nfsm_mtouio(&io, len2);
2438 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2439 vap->va_mode = nfstov_mode(sp->sa_mode);
2441 *(pathcp + len2) = '\0';
2447 if (vap->va_mode == (mode_t)VNOVAL)
2451 error = VOP_NSYMLINK(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap, pathcp);
2455 bzero((caddr_t)fhp, sizeof(nfh));
2456 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2457 error = VFS_VPTOFH(vp, &fhp->fh_fid);
2459 error = VOP_GETATTR(vp, vap);
2474 FREE(pathcp, M_TEMP);
2478 diraft_ret = VOP_GETATTR(dirp, &diraft);
2482 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2485 nfsm_srvpostop_fh(fhp);
2486 nfsm_srvpostop_attr(0, vap);
2488 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2500 FREE(pathcp, M_TEMP);
2508 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2509 struct thread *td, struct mbuf **mrq)
2511 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2512 struct sockaddr *nam = nfsd->nd_nam;
2513 caddr_t dpos = nfsd->nd_dpos;
2514 struct ucred *cred = &nfsd->nd_cr;
2515 struct vattr va, dirfor, diraft;
2516 struct vattr *vap = &va;
2517 struct nfs_fattr *fp;
2518 struct nlookupdata nd;
2523 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2524 int v3 = (nfsd->nd_flag & ND_NFSV3);
2526 struct mbuf *mb, *mb2, *mreq;
2533 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2539 fhp = &nfh.fh_generic;
2541 nfsm_srvnamesiz(len);
2543 error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
2544 fhp, len, slp, nam, &md, &dpos, &dirp,
2545 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2548 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2551 nfsm_reply(NFSX_WCCDATA(v3));
2552 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2560 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2561 vap->va_mode = nfstov_mode(*tl++);
2565 * At this point nd.ni_dvp is referenced and exclusively locked and
2566 * nd.ni_vp, if it exists, is referenced but not locked.
2569 vap->va_type = VDIR;
2576 * Issue mkdir op. Since SAVESTART is not set, the pathname
2577 * component is freed by the VOP call. This will fill-in
2578 * nd.ni_vp, reference, and exclusively lock it.
2580 if (vap->va_mode == (mode_t)VNOVAL)
2583 error = VOP_NMKDIR(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
2588 bzero((caddr_t)fhp, sizeof(nfh));
2589 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2590 error = VFS_VPTOFH(vp, &fhp->fh_fid);
2592 error = VOP_GETATTR(vp, vap);
2596 diraft_ret = VOP_GETATTR(dirp, &diraft);
2597 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2600 nfsm_srvpostop_fh(fhp);
2601 nfsm_srvpostop_attr(0, vap);
2603 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2605 nfsm_srvfhtom(fhp, v3);
2606 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2607 nfsm_srvfillattr(vap, fp);
2631 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2632 struct thread *td, struct mbuf **mrq)
2634 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2635 struct sockaddr *nam = nfsd->nd_nam;
2636 caddr_t dpos = nfsd->nd_dpos;
2637 struct ucred *cred = &nfsd->nd_cr;
2641 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2642 int v3 = (nfsd->nd_flag & ND_NFSV3);
2644 struct mbuf *mb, *mreq;
2648 struct vattr dirfor, diraft;
2651 struct nlookupdata nd;
2653 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2659 fhp = &nfh.fh_generic;
2661 nfsm_srvnamesiz(len);
2663 error = nfs_namei(&nd, cred, NAMEI_DELETE, &dvp, &vp,
2664 fhp, len, slp, nam, &md, &dpos, &dirp,
2665 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2668 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2671 nfsm_reply(NFSX_WCCDATA(v3));
2672 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2676 if (vp->v_type != VDIR) {
2682 * The root of a mounted filesystem cannot be deleted.
2684 if (vp->v_flag & VROOT)
2688 * Issue or abort op. Since SAVESTART is not set, path name
2689 * component is freed by the VOP after either.
2696 error = VOP_NRMDIR(&nd.nl_nch, dvp, nd.nl_cred);
2703 diraft_ret = VOP_GETATTR(dirp, &diraft);
2704 nfsm_reply(NFSX_WCCDATA(v3));
2706 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2727 * nfs readdir service
2728 * - mallocs what it thinks is enough to read
2729 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2730 * - calls VOP_READDIR()
2731 * - loops around building the reply
2732 * if the output generated exceeds count break out of loop
2733 * The nfsm_clget macro is used here so that the reply will be packed
2734 * tightly in mbuf clusters.
2735 * - it only knows that it has encountered eof when the VOP_READDIR()
2737 * - as such one readdir rpc will return eof false although you are there
2738 * and then the next will return eof
2739 * - it trims out records with d_fileno == 0
2740 * this doesn't matter for Unix clients, but they might confuse clients
2742 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2743 * than requested, but this may not apply to all filesystems. For
2744 * example, client NFS does not { although it is never remote mounted
2746 * The alternate call nfsrv_readdirplus() does lookups as well.
2747 * PS: The NFS protocol spec. does not clarify what the "count" byte
2748 * argument is a count of.. just name strings and file id's or the
2749 * entire reply rpc or ...
2750 * I tried just file name and id sizes and it confused the Sun client,
2751 * so I am using the full rpc size now. The "paranoia.." comment refers
2752 * to including the status longwords that are not a part of the dir.
2753 * "entry" structures, but are in the rpc.
2757 u_int32_t fl_postopok;
2758 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2760 u_int32_t fl_fhsize;
2761 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2765 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2766 struct thread *td, struct mbuf **mrq)
2768 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2769 struct sockaddr *nam = nfsd->nd_nam;
2770 caddr_t dpos = nfsd->nd_dpos;
2771 struct ucred *cred = &nfsd->nd_cr;
2779 struct mbuf *mb, *mb2, *mreq, *mp2;
2780 char *cpos, *cend, *cp2, *rbuf;
2781 struct vnode *vp = NULL;
2787 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2788 int siz, cnt, fullsiz, eofflag, rdonly, ncookies;
2789 int v3 = (nfsd->nd_flag & ND_NFSV3);
2790 u_quad_t off, toff, verf;
2791 off_t *cookies = NULL, *cookiep;
2793 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2794 fhp = &nfh.fh_generic;
2797 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2798 toff = fxdr_hyper(tl);
2800 verf = fxdr_hyper(tl);
2803 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2804 toff = fxdr_unsigned(u_quad_t, *tl++);
2805 verf = 0; /* shut up gcc */
2808 cnt = fxdr_unsigned(int, *tl);
2809 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2810 xfer = NFS_SRVMAXDATA(nfsd);
2816 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2817 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
2818 if (!error && vp->v_type != VDIR) {
2824 nfsm_reply(NFSX_UNSIGNED);
2825 nfsm_srvpostop_attr(getret, &at);
2831 * Obtain lock on vnode for this section of the code
2835 error = getret = VOP_GETATTR(vp, &at);
2838 * XXX This check may be too strict for Solaris 2.5 clients.
2840 if (!error && toff && verf && verf != at.va_filerev)
2841 error = NFSERR_BAD_COOKIE;
2845 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
2849 nfsm_reply(NFSX_POSTOPATTR(v3));
2850 nfsm_srvpostop_attr(getret, &at);
2857 * end section. Allocate rbuf and continue
2859 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
2862 iv.iov_len = fullsiz;
2865 io.uio_offset = (off_t)off;
2866 io.uio_resid = fullsiz;
2867 io.uio_segflg = UIO_SYSSPACE;
2868 io.uio_rw = UIO_READ;
2872 kfree((caddr_t)cookies, M_TEMP);
2875 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
2876 off = (off_t)io.uio_offset;
2877 if (!cookies && !error)
2878 error = NFSERR_PERM;
2880 getret = VOP_GETATTR(vp, &at);
2887 kfree((caddr_t)rbuf, M_TEMP);
2889 kfree((caddr_t)cookies, M_TEMP);
2890 nfsm_reply(NFSX_POSTOPATTR(v3));
2891 nfsm_srvpostop_attr(getret, &at);
2896 siz -= io.uio_resid;
2899 * If nothing read, return eof
2905 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2908 nfsm_srvpostop_attr(getret, &at);
2909 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2910 txdr_hyper(at.va_filerev, tl);
2913 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2916 FREE((caddr_t)rbuf, M_TEMP);
2917 FREE((caddr_t)cookies, M_TEMP);
2924 * Check for degenerate cases of nothing useful read.
2925 * If so go try again
2929 dp = (struct dirent *)cpos;
2932 * For some reason FreeBSD's ufs_readdir() chooses to back the
2933 * directory offset up to a block boundary, so it is necessary to
2934 * skip over the records that preceed the requested offset. This
2935 * requires the assumption that file offset cookies monotonically
2938 while (cpos < cend && ncookies > 0 &&
2939 (dp->d_ino == 0 || dp->d_type == DT_WHT ||
2940 ((u_quad_t)(*cookiep)) <= toff)) {
2941 dp = _DIRENT_NEXT(dp);
2946 if (cpos >= cend || ncookies == 0) {
2952 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
2953 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2955 nfsm_srvpostop_attr(getret, &at);
2956 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2957 txdr_hyper(at.va_filerev, tl);
2961 be = bp + M_TRAILINGSPACE(mp);
2963 /* Loop through the records and build reply */
2964 while (cpos < cend && ncookies > 0) {
2965 if (dp->d_ino != 0 && dp->d_type != DT_WHT) {
2966 nlen = dp->d_namlen;
2967 rem = nfsm_rndup(nlen) - nlen;
2968 len += (4 * NFSX_UNSIGNED + nlen + rem);
2970 len += 2 * NFSX_UNSIGNED;
2976 * Build the directory record xdr from
2981 bp += NFSX_UNSIGNED;
2985 bp += NFSX_UNSIGNED;
2988 *tl = txdr_unsigned(dp->d_ino);
2989 bp += NFSX_UNSIGNED;
2991 *tl = txdr_unsigned(nlen);
2992 bp += NFSX_UNSIGNED;
2994 /* And loop around copying the name */
3003 bcopy(cp, bp, tsiz);
3009 /* And null pad to a int32_t boundary */
3010 for (i = 0; i < rem; i++)
3014 /* Finish off the record */
3016 *tl = txdr_unsigned(*cookiep >> 32);
3017 bp += NFSX_UNSIGNED;
3020 *tl = txdr_unsigned(*cookiep);
3021 bp += NFSX_UNSIGNED;
3023 dp = _DIRENT_NEXT(dp);
3032 bp += NFSX_UNSIGNED;
3038 bp += NFSX_UNSIGNED;
3041 mp->m_len = bp - mtod(mp, caddr_t);
3043 mp->m_len += bp - bpos;
3044 FREE((caddr_t)rbuf, M_TEMP);
3045 FREE((caddr_t)cookies, M_TEMP);
3054 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3055 struct thread *td, struct mbuf **mrq)
3057 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3058 struct sockaddr *nam = nfsd->nd_nam;
3059 caddr_t dpos = nfsd->nd_dpos;
3060 struct ucred *cred = &nfsd->nd_cr;
3068 struct mbuf *mb, *mb2, *mreq, *mp2;
3069 char *cpos, *cend, *cp2, *rbuf;
3070 struct vnode *vp = NULL, *nvp;
3073 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3076 struct vattr va, at, *vap = &va;
3077 struct nfs_fattr *fp;
3078 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3079 int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies;
3080 u_quad_t off, toff, verf;
3081 off_t *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3083 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3084 fhp = &nfh.fh_generic;
3086 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
3087 toff = fxdr_hyper(tl);
3089 verf = fxdr_hyper(tl);
3091 siz = fxdr_unsigned(int, *tl++);
3092 cnt = fxdr_unsigned(int, *tl);
3094 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3095 xfer = NFS_SRVMAXDATA(nfsd);
3101 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3102 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3103 if (!error && vp->v_type != VDIR) {
3109 nfsm_reply(NFSX_UNSIGNED);
3110 nfsm_srvpostop_attr(getret, &at);
3114 error = getret = VOP_GETATTR(vp, &at);
3117 * XXX This check may be too strict for Solaris 2.5 clients.
3119 if (!error && toff && verf && verf != at.va_filerev)
3120 error = NFSERR_BAD_COOKIE;
3123 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3128 nfsm_reply(NFSX_V3POSTOPATTR);
3129 nfsm_srvpostop_attr(getret, &at);
3134 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3137 iv.iov_len = fullsiz;
3140 io.uio_offset = (off_t)off;
3141 io.uio_resid = fullsiz;
3142 io.uio_segflg = UIO_SYSSPACE;
3143 io.uio_rw = UIO_READ;
3147 kfree((caddr_t)cookies, M_TEMP);
3150 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3151 off = (u_quad_t)io.uio_offset;
3152 getret = VOP_GETATTR(vp, &at);
3153 if (!cookies && !error)
3154 error = NFSERR_PERM;
3161 kfree((caddr_t)cookies, M_TEMP);
3162 kfree((caddr_t)rbuf, M_TEMP);
3163 nfsm_reply(NFSX_V3POSTOPATTR);
3164 nfsm_srvpostop_attr(getret, &at);
3169 siz -= io.uio_resid;
3172 * If nothing read, return eof
3178 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3180 nfsm_srvpostop_attr(getret, &at);
3181 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3182 txdr_hyper(at.va_filerev, tl);
3186 FREE((caddr_t)cookies, M_TEMP);
3187 FREE((caddr_t)rbuf, M_TEMP);
3194 * Check for degenerate cases of nothing useful read.
3195 * If so go try again
3199 dp = (struct dirent *)cpos;
3202 * For some reason FreeBSD's ufs_readdir() chooses to back the
3203 * directory offset up to a block boundary, so it is necessary to
3204 * skip over the records that preceed the requested offset. This
3205 * requires the assumption that file offset cookies monotonically
3208 while (cpos < cend && ncookies > 0 &&
3209 (dp->d_ino == 0 || dp->d_type == DT_WHT ||
3210 ((u_quad_t)(*cookiep)) <= toff)) {
3211 dp = _DIRENT_NEXT(dp);
3216 if (cpos >= cend || ncookies == 0) {
3223 * Probe one of the directory entries to see if the filesystem
3226 if (VFS_VGET(vp->v_mount, dp->d_ino, &nvp) == EOPNOTSUPP) {
3227 error = NFSERR_NOTSUPP;
3230 kfree((caddr_t)cookies, M_TEMP);
3231 kfree((caddr_t)rbuf, M_TEMP);
3232 nfsm_reply(NFSX_V3POSTOPATTR);
3233 nfsm_srvpostop_attr(getret, &at);
3242 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
3244 nfsm_srvpostop_attr(getret, &at);
3245 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3246 txdr_hyper(at.va_filerev, tl);
3249 be = bp + M_TRAILINGSPACE(mp);
3251 /* Loop through the records and build reply */
3252 while (cpos < cend && ncookies > 0) {
3253 if (dp->d_ino != 0 && dp->d_type != DT_WHT) {
3254 nlen = dp->d_namlen;
3255 rem = nfsm_rndup(nlen)-nlen;
3258 * For readdir_and_lookup get the vnode using
3261 if (VFS_VGET(vp->v_mount, dp->d_ino, &nvp))
3263 bzero((caddr_t)nfhp, NFSX_V3FH);
3265 nvp->v_mount->mnt_stat.f_fsid;
3266 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
3271 if (VOP_GETATTR(nvp, vap)) {
3280 * If either the dircount or maxcount will be
3281 * exceeded, get out now. Both of these lengths
3282 * are calculated conservatively, including all
3285 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3287 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3288 if (len > cnt || dirlen > fullsiz) {
3294 * Build the directory record xdr from
3297 fp = (struct nfs_fattr *)&fl.fl_fattr;
3298 nfsm_srvfillattr(vap, fp);
3299 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3300 fl.fl_fhok = nfs_true;
3301 fl.fl_postopok = nfs_true;
3302 fl.fl_off.nfsuquad[0] = txdr_unsigned(*cookiep >> 32);
3303 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3307 bp += NFSX_UNSIGNED;
3310 bp += NFSX_UNSIGNED;
3312 *tl = txdr_unsigned(dp->d_ino);
3313 bp += NFSX_UNSIGNED;
3315 *tl = txdr_unsigned(nlen);
3316 bp += NFSX_UNSIGNED;
3318 /* And loop around copying the name */
3323 if ((bp + xfer) > be)
3327 bcopy(cp, bp, tsiz);
3333 /* And null pad to a int32_t boundary */
3334 for (i = 0; i < rem; i++)
3338 * Now copy the flrep structure out.
3340 xfer = sizeof (struct flrep);
3344 if ((bp + xfer) > be)
3348 bcopy(cp, bp, tsiz);
3356 dp = _DIRENT_NEXT(dp);
3365 bp += NFSX_UNSIGNED;
3371 bp += NFSX_UNSIGNED;
3374 mp->m_len = bp - mtod(mp, caddr_t);
3376 mp->m_len += bp - bpos;
3377 FREE((caddr_t)cookies, M_TEMP);
3378 FREE((caddr_t)rbuf, M_TEMP);
3386 * nfs commit service
3389 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3390 struct thread *td, struct mbuf **mrq)
3392 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3393 struct sockaddr *nam = nfsd->nd_nam;
3394 caddr_t dpos = nfsd->nd_dpos;
3395 struct ucred *cred = &nfsd->nd_cr;
3396 struct vattr bfor, aft;
3397 struct vnode *vp = NULL;
3403 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt;
3405 struct mbuf *mb, *mb2, *mreq;
3408 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3409 fhp = &nfh.fh_generic;
3411 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3414 * XXX At this time VOP_FSYNC() does not accept offset and byte
3415 * count parameters, so these arguments are useless (someday maybe).
3417 off = fxdr_hyper(tl);
3419 cnt = fxdr_unsigned(int, *tl);
3420 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3421 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3423 nfsm_reply(2 * NFSX_UNSIGNED);
3424 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3428 for_ret = VOP_GETATTR(vp, &bfor);
3430 if (cnt > MAX_COMMIT_COUNT) {
3432 * Give up and do the whole thing
3435 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3436 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3438 error = VOP_FSYNC(vp, MNT_WAIT);
3441 * Locate and synchronously write any buffers that fall
3442 * into the requested range. Note: we are assuming that
3443 * f_iosize is a power of 2.
3445 int iosize = vp->v_mount->mnt_stat.f_iosize;
3446 int iomask = iosize - 1;
3450 * Align to iosize boundry, super-align to page boundry.
3453 cnt += off & iomask;
3454 off &= ~(u_quad_t)iomask;
3456 if (off & PAGE_MASK) {
3457 cnt += off & PAGE_MASK;
3458 off &= ~(u_quad_t)PAGE_MASK;
3463 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3464 vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
3472 * If we have a buffer and it is marked B_DELWRI we
3473 * have to lock and write it. Otherwise the prior
3474 * write is assumed to have already been committed.
3476 if ((bp = findblk(vp, loffset)) != NULL && (bp->b_flags & B_DELWRI)) {
3477 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
3478 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL) == 0)
3480 continue; /* retry */
3483 bp->b_flags &= ~B_ASYNC;
3496 aft_ret = VOP_GETATTR(vp, &aft);
3499 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3500 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3502 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3503 if (nfsver.tv_sec == 0)
3505 *tl++ = txdr_unsigned(nfsver.tv_sec);
3506 *tl = txdr_unsigned(nfsver.tv_nsec / 1000);
3517 * nfs statfs service
3520 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3521 struct thread *td, struct mbuf **mrq)
3523 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3524 struct sockaddr *nam = nfsd->nd_nam;
3525 caddr_t dpos = nfsd->nd_dpos;
3526 struct ucred *cred = &nfsd->nd_cr;
3528 struct nfs_statfs *sfp;
3532 int error = 0, rdonly, getret = 1;
3533 int v3 = (nfsd->nd_flag & ND_NFSV3);
3535 struct mbuf *mb, *mb2, *mreq;
3536 struct vnode *vp = NULL;
3540 struct statfs statfs;
3543 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3544 fhp = &nfh.fh_generic;
3546 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3547 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3549 nfsm_reply(NFSX_UNSIGNED);
3550 nfsm_srvpostop_attr(getret, &at);
3555 error = VFS_STATFS(vp->v_mount, sf, proc0.p_ucred);
3556 getret = VOP_GETATTR(vp, &at);
3559 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3561 nfsm_srvpostop_attr(getret, &at);
3566 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3568 tval = (u_quad_t)sf->f_blocks;
3569 tval *= (u_quad_t)sf->f_bsize;
3570 txdr_hyper(tval, &sfp->sf_tbytes);
3571 tval = (u_quad_t)sf->f_bfree;
3572 tval *= (u_quad_t)sf->f_bsize;
3573 txdr_hyper(tval, &sfp->sf_fbytes);
3574 tval = (u_quad_t)sf->f_bavail;
3575 tval *= (u_quad_t)sf->f_bsize;
3576 txdr_hyper(tval, &sfp->sf_abytes);
3577 sfp->sf_tfiles.nfsuquad[0] = 0;
3578 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
3579 sfp->sf_ffiles.nfsuquad[0] = 0;
3580 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3581 sfp->sf_afiles.nfsuquad[0] = 0;
3582 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3583 sfp->sf_invarsec = 0;
3585 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3586 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3587 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3588 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3589 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3598 * nfs fsinfo service
3601 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3602 struct thread *td, struct mbuf **mrq)
3604 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3605 struct sockaddr *nam = nfsd->nd_nam;
3606 caddr_t dpos = nfsd->nd_dpos;
3607 struct ucred *cred = &nfsd->nd_cr;
3609 struct nfsv3_fsinfo *sip;
3612 int error = 0, rdonly, getret = 1, pref;
3614 struct mbuf *mb, *mb2, *mreq;
3615 struct vnode *vp = NULL;
3622 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3623 fhp = &nfh.fh_generic;
3625 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3626 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3628 nfsm_reply(NFSX_UNSIGNED);
3629 nfsm_srvpostop_attr(getret, &at);
3634 /* XXX Try to make a guess on the max file size. */
3635 VFS_STATFS(vp->v_mount, &sb, proc0.p_ucred);
3636 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3638 getret = VOP_GETATTR(vp, &at);
3641 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3642 nfsm_srvpostop_attr(getret, &at);
3643 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3647 * There should be file system VFS OP(s) to get this information.
3648 * For now, assume ufs.
3650 if (slp->ns_so->so_type == SOCK_DGRAM)
3651 pref = NFS_MAXDGRAMDATA;
3654 sip->fs_rtmax = txdr_unsigned(NFS_MAXDATA);
3655 sip->fs_rtpref = txdr_unsigned(pref);
3656 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3657 sip->fs_wtmax = txdr_unsigned(NFS_MAXDATA);
3658 sip->fs_wtpref = txdr_unsigned(pref);
3659 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3660 sip->fs_dtpref = txdr_unsigned(pref);
3661 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3662 sip->fs_timedelta.nfsv3_sec = 0;
3663 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3664 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3665 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3666 NFSV3FSINFO_CANSETTIME);
3674 * nfs pathconf service
3677 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3678 struct thread *td, struct mbuf **mrq)
3680 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3681 struct sockaddr *nam = nfsd->nd_nam;
3682 caddr_t dpos = nfsd->nd_dpos;
3683 struct ucred *cred = &nfsd->nd_cr;
3685 struct nfsv3_pathconf *pc;
3688 int error = 0, rdonly, getret = 1;
3689 register_t linkmax, namemax, chownres, notrunc;
3691 struct mbuf *mb, *mb2, *mreq;
3692 struct vnode *vp = NULL;
3697 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3698 fhp = &nfh.fh_generic;
3700 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3701 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3703 nfsm_reply(NFSX_UNSIGNED);
3704 nfsm_srvpostop_attr(getret, &at);
3708 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3710 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3712 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3714 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc);
3715 getret = VOP_GETATTR(vp, &at);
3718 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3719 nfsm_srvpostop_attr(getret, &at);
3724 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3726 pc->pc_linkmax = txdr_unsigned(linkmax);
3727 pc->pc_namemax = txdr_unsigned(namemax);
3728 pc->pc_notrunc = txdr_unsigned(notrunc);
3729 pc->pc_chownrestricted = txdr_unsigned(chownres);
3732 * These should probably be supported by VOP_PATHCONF(), but
3733 * until msdosfs is exportable (why would you want to?), the
3734 * Unix defaults should be ok.
3736 pc->pc_caseinsensitive = nfs_false;
3737 pc->pc_casepreserving = nfs_true;
3745 * Null operation, used by clients to ping server
3749 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3750 struct thread *td, struct mbuf **mrq)
3752 struct mbuf *mrep = nfsd->nd_mrep;
3754 int error = NFSERR_RETVOID;
3755 struct mbuf *mb, *mreq;
3757 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3763 * No operation, used for obsolete procedures
3767 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3768 struct thread *td, struct mbuf **mrq)
3770 struct mbuf *mrep = nfsd->nd_mrep;
3773 struct mbuf *mb, *mreq;
3775 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3776 if (nfsd->nd_repstat)
3777 error = nfsd->nd_repstat;
3779 error = EPROCUNAVAIL;
3786 * Perform access checking for vnodes obtained from file handles that would
3787 * refer to files already opened by a Unix client. You cannot just use
3788 * vn_writechk() and VOP_ACCESS() for two reasons.
3789 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3790 * 2 - The owner is to be given access irrespective of mode bits for some
3791 * operations, so that processes that chmod after opening a file don't
3792 * break. I don't like this because it opens a security hole, but since
3793 * the nfs server opens a security hole the size of a barn door anyhow,
3796 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3797 * will return EPERM instead of EACCESS. EPERM is always an error.
3800 nfsrv_access(struct vnode *vp, int flags, struct ucred *cred,
3801 int rdonly, struct thread *td, int override)
3806 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3807 if (flags & VWRITE) {
3808 /* Just vn_writechk() changed to check rdonly */
3810 * Disallow write attempts on read-only file systems;
3811 * unless the file is a socket or a block or character
3812 * device resident on the file system.
3814 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3815 switch (vp->v_type) {
3825 * If there's shared text associated with
3826 * the inode, we can't allow writing.
3828 if (vp->v_flag & VTEXT)
3831 error = VOP_GETATTR(vp, &vattr);
3834 error = VOP_ACCESS(vp, flags, cred);
3836 * Allow certain operations for the owner (reads and writes
3837 * on files that are already open).
3839 if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3843 #endif /* NFS_NOSERVER */