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>
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);
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 mount *, struct vnode *, int,
145 struct ucred *, int, 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;
161 struct mount *mp = NULL;
167 int error = 0, rdonly, getret;
169 struct mbuf *mb, *mreq, *mb2;
170 struct vattr vattr, *vap = &vattr;
171 u_long testmode, nfsmode;
173 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
174 fhp = &nfh.fh_generic;
176 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
177 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam, &rdonly,
178 (nfsd->nd_flag & ND_KERBAUTH), TRUE);
180 nfsm_reply(NFSX_UNSIGNED);
181 nfsm_srvpostop_attr(1, (struct vattr *)0);
185 nfsmode = fxdr_unsigned(u_int32_t, *tl);
186 if ((nfsmode & NFSV3ACCESS_READ) &&
187 nfsrv_access(mp, vp, VREAD, cred, rdonly, td, 0))
188 nfsmode &= ~NFSV3ACCESS_READ;
189 if (vp->v_type == VDIR)
190 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
193 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
194 if ((nfsmode & testmode) &&
195 nfsrv_access(mp, vp, VWRITE, cred, rdonly, td, 0))
196 nfsmode &= ~testmode;
197 if (vp->v_type == VDIR)
198 testmode = NFSV3ACCESS_LOOKUP;
200 testmode = NFSV3ACCESS_EXECUTE;
201 if ((nfsmode & testmode) &&
202 nfsrv_access(mp, vp, VEXEC, cred, rdonly, td, 0))
203 nfsmode &= ~testmode;
204 getret = VOP_GETATTR(vp, vap);
207 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
208 nfsm_srvpostop_attr(getret, vap);
209 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
210 *tl = txdr_unsigned(nfsmode);
218 * nfs getattr service
221 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
222 struct thread *td, struct mbuf **mrq)
224 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
225 struct sockaddr *nam = nfsd->nd_nam;
226 caddr_t dpos = nfsd->nd_dpos;
227 struct ucred *cred = &nfsd->nd_cr;
228 struct nfs_fattr *fp;
230 struct vattr *vap = &va;
231 struct vnode *vp = NULL;
232 struct mount *mp = NULL;
238 int error = 0, rdonly;
240 struct mbuf *mb, *mb2, *mreq;
242 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
243 fhp = &nfh.fh_generic;
245 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
246 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
252 error = VOP_GETATTR(vp, vap);
255 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
260 nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
261 nfsm_srvfillattr(vap, fp);
271 * nfs setattr service
274 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
275 struct thread *td, struct mbuf **mrq)
277 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
278 struct sockaddr *nam = nfsd->nd_nam;
279 caddr_t dpos = nfsd->nd_dpos;
280 struct ucred *cred = &nfsd->nd_cr;
281 struct vattr va, preat;
282 struct vattr *vap = &va;
283 struct nfsv2_sattr *sp;
284 struct nfs_fattr *fp;
285 struct vnode *vp = NULL;
286 struct mount *mp = NULL;
292 int error = 0, rdonly, preat_ret = 1, postat_ret = 1;
293 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
295 struct mbuf *mb, *mb2, *mreq;
296 struct timespec guard;
298 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
299 fhp = &nfh.fh_generic;
304 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
305 gcheck = fxdr_unsigned(int, *tl);
307 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
308 fxdr_nfsv3time(tl, &guard);
311 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
313 * Nah nah nah nah na nah
314 * There is a bug in the Sun client that puts 0xffff in the mode
315 * field of sattr when it should put in 0xffffffff. The u_short
316 * doesn't sign extend.
317 * --> check the low order 2 bytes for 0xffff
319 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
320 vap->va_mode = nfstov_mode(sp->sa_mode);
321 if (sp->sa_uid != nfs_xdrneg1)
322 vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
323 if (sp->sa_gid != nfs_xdrneg1)
324 vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
325 if (sp->sa_size != nfs_xdrneg1)
326 vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
327 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
329 fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
331 vap->va_atime.tv_sec =
332 fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
333 vap->va_atime.tv_nsec = 0;
336 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
337 fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
342 * Now that we have all the fields, lets do it.
344 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam, &rdonly,
345 (nfsd->nd_flag & ND_KERBAUTH), TRUE);
347 nfsm_reply(2 * NFSX_UNSIGNED);
348 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
354 * vp now an active resource, pay careful attention to cleanup
358 error = preat_ret = VOP_GETATTR(vp, &preat);
359 if (!error && gcheck &&
360 (preat.va_ctime.tv_sec != guard.tv_sec ||
361 preat.va_ctime.tv_nsec != guard.tv_nsec))
362 error = NFSERR_NOT_SYNC;
366 nfsm_reply(NFSX_WCCDATA(v3));
367 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
374 * If the size is being changed write acces is required, otherwise
375 * just check for a read only file system.
377 if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
378 if (rdonly || (mp->mnt_flag & MNT_RDONLY)) {
383 if (vp->v_type == VDIR) {
386 } else if ((error = nfsrv_access(mp, vp, VWRITE, cred, rdonly,
391 error = VOP_SETATTR(vp, vap, cred);
392 postat_ret = VOP_GETATTR(vp, vap);
398 nfsm_reply(NFSX_WCCORFATTR(v3));
400 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
404 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
405 nfsm_srvfillattr(vap, fp);
419 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
420 struct thread *td, struct mbuf **mrq)
422 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
423 struct sockaddr *nam = nfsd->nd_nam;
424 caddr_t dpos = nfsd->nd_dpos;
425 struct ucred *cred = &nfsd->nd_cr;
426 struct nfs_fattr *fp;
427 struct nlookupdata nd;
437 int error = 0, len, dirattr_ret = 1;
438 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
440 struct mbuf *mb, *mb2, *mreq;
441 struct vattr va, dirattr, *vap = &va;
443 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
448 fhp = &nfh.fh_generic;
450 nfsm_srvnamesiz(len);
452 pubflag = nfs_ispublicfh(fhp);
454 error = nfs_namei(&nd, cred, NAMEI_LOOKUP, NULL, &vp,
455 fhp, len, slp, nam, &md, &dpos,
456 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
459 * namei failure, only dirp to cleanup. Clear out garbarge from
460 * structure in case macros jump to nfsmout.
466 dirattr_ret = VOP_GETATTR(dirp, &dirattr);
470 nfsm_reply(NFSX_POSTOPATTR(v3));
471 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
477 * Locate index file for public filehandle
479 * error is 0 on entry and 0 on exit from this block.
483 if (vp->v_type == VDIR && nfs_pub.np_index != NULL) {
485 * Setup call to lookup() to see if we can find
486 * the index file. Arguably, this doesn't belong
487 * in a kernel.. Ugh. If an error occurs, do not
488 * try to install an index file and then clear the
491 * When we replace nd with ind and redirect ndp,
492 * maintenance of ni_startdir and ni_vp shift to
493 * ind and we have to clean them up in the old nd.
494 * However, the cnd resource continues to be maintained
495 * via the original nd. Confused? You aren't alone!
498 cache_copy(&nd.nl_nch, &nch);
500 error = nlookup_init_raw(&nd, nfs_pub.np_index,
501 UIO_SYSSPACE, 0, cred, &nch);
504 error = nlookup(&nd);
508 * Found an index file. Get rid of
509 * the old references. transfer vp and
510 * load up the new vp. Fortunately we do
511 * not have to deal with dvp, that would be
518 error = cache_vget(&nd.nl_nch, nd.nl_cred,
520 KKASSERT(error == 0);
525 * If the public filehandle was used, check that this lookup
526 * didn't result in a filehandle outside the publicly exported
527 * filesystem. We clear the poor vp here to avoid lockups due
531 if (vp->v_mount != nfs_pub.np_mount) {
540 dirattr_ret = VOP_GETATTR(dirp, &dirattr);
546 * Resources at this point:
547 * ndp->ni_vp may not be NULL
552 nfsm_reply(NFSX_POSTOPATTR(v3));
553 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
559 * Clear out some resources prior to potentially blocking. This
560 * is not as critical as ni_dvp resources in other routines, but
566 * Get underlying attribute, then release remaining resources ( for
567 * the same potential blocking reason ) and reply.
569 bzero(&fhp->fh_fid, sizeof(fhp->fh_fid));
570 error = VFS_VPTOFH(vp, &fhp->fh_fid);
572 error = VOP_GETATTR(vp, vap);
576 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
578 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
582 nfsm_srvfhtom(fhp, v3);
584 nfsm_srvpostop_attr(0, vap);
585 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
587 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
588 nfsm_srvfillattr(vap, fp);
594 nlookup_done(&nd); /* may be called twice */
601 * nfs readlink service
604 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
605 struct thread *td, struct mbuf **mrq)
607 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
608 struct sockaddr *nam = nfsd->nd_nam;
609 caddr_t dpos = nfsd->nd_dpos;
610 struct ucred *cred = &nfsd->nd_cr;
611 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
612 struct iovec *ivp = iv;
616 int error = 0, rdonly, i, tlen, len, getret;
617 int v3 = (nfsd->nd_flag & ND_NFSV3);
619 struct mbuf *mb, *mb2, *mp1, *mp2, *mp3, *mreq;
620 struct vnode *vp = NULL;
621 struct mount *mp = NULL;
625 struct uio io, *uiop = &io;
627 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
629 mp2 = (struct mbuf *)0;
632 fhp = &nfh.fh_generic;
636 while (len < NFS_MAXPATHLEN) {
637 mp1 = m_getcl(MB_WAIT, MT_DATA, 0);
638 mp1->m_len = MCLBYTES;
645 if ((len + mp1->m_len) > NFS_MAXPATHLEN) {
646 mp1->m_len = NFS_MAXPATHLEN-len;
647 len = NFS_MAXPATHLEN;
650 ivp->iov_base = mtod(mp1, caddr_t);
651 ivp->iov_len = mp1->m_len;
656 uiop->uio_iovcnt = i;
657 uiop->uio_offset = 0;
658 uiop->uio_resid = len;
659 uiop->uio_rw = UIO_READ;
660 uiop->uio_segflg = UIO_SYSSPACE;
662 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
663 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
665 nfsm_reply(2 * NFSX_UNSIGNED);
666 nfsm_srvpostop_attr(1, (struct vattr *)0);
670 if (vp->v_type != VLNK) {
677 error = VOP_READLINK(vp, uiop, cred);
679 getret = VOP_GETATTR(vp, &attr);
682 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
684 nfsm_srvpostop_attr(getret, &attr);
690 if (uiop->uio_resid > 0) {
691 len -= uiop->uio_resid;
692 tlen = nfsm_rndup(len);
693 nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
695 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
696 *tl = txdr_unsigned(len);
711 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
712 struct thread *td, struct mbuf **mrq)
714 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
715 struct sockaddr *nam = nfsd->nd_nam;
716 caddr_t dpos = nfsd->nd_dpos;
717 struct ucred *cred = &nfsd->nd_cr;
721 struct nfs_fattr *fp;
726 int error = 0, rdonly, cnt, len, left, siz, tlen, getret;
727 int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
729 struct mbuf *mb, *mb2, *mreq;
731 struct vnode *vp = NULL;
732 struct mount *mp = NULL;
735 struct uio io, *uiop = &io;
736 struct vattr va, *vap = &va;
741 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
742 fhp = &nfh.fh_generic;
745 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
746 off = fxdr_hyper(tl);
748 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
749 off = (off_t)fxdr_unsigned(u_int32_t, *tl);
751 nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
754 * Reference vp. If an error occurs, vp will be invalid, but we
755 * have to NULL it just in case. The macros might goto nfsmout
759 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
760 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
763 nfsm_reply(2 * NFSX_UNSIGNED);
764 nfsm_srvpostop_attr(1, (struct vattr *)0);
769 if (vp->v_type != VREG) {
773 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
776 if ((error = nfsrv_access(mp, vp, VREAD, cred, rdonly, td, 1)) != 0)
777 error = nfsrv_access(mp, vp, VEXEC, cred, rdonly, td, 1);
779 getret = VOP_GETATTR(vp, vap);
785 nfsm_reply(NFSX_POSTOPATTR(v3));
786 nfsm_srvpostop_attr(getret, vap);
792 * Calculate byte count to read
795 if (off >= vap->va_size)
797 else if ((off + reqlen) > vap->va_size)
798 cnt = vap->va_size - off;
803 * Calculate seqcount for heuristic
811 * Locate best candidate
814 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
818 if (nfsheur[hi].nh_vp == vp) {
822 if (nfsheur[hi].nh_use > 0)
823 --nfsheur[hi].nh_use;
824 hi = (hi + 1) % NUM_HEURISTIC;
825 if (nfsheur[hi].nh_use < nh->nh_use)
829 if (nh->nh_vp != vp) {
832 nh->nh_use = NHUSE_INIT;
840 * Calculate heuristic
843 if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
844 if (++nh->nh_seqcount > IO_SEQMAX)
845 nh->nh_seqcount = IO_SEQMAX;
846 } else if (nh->nh_seqcount > 1) {
851 nh->nh_use += NHUSE_INC;
852 if (nh->nh_use > NHUSE_MAX)
853 nh->nh_use = NHUSE_MAX;
854 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
857 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
859 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
861 fp = (struct nfs_fattr *)tl;
862 tl += (NFSX_V3FATTR / sizeof (u_int32_t));
864 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
865 fp = (struct nfs_fattr *)tl;
866 tl += (NFSX_V2FATTR / sizeof (u_int32_t));
868 len = left = nfsm_rndup(cnt);
871 * Generate the mbuf list with the uio_iov ref. to it.
876 siz = min(M_TRAILINGSPACE(m), left);
882 m = m_getcl(MB_WAIT, MT_DATA, 0);
888 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
890 uiop->uio_iov = iv2 = iv;
896 panic("nfsrv_read iov");
897 siz = min(M_TRAILINGSPACE(m), left);
899 iv->iov_base = mtod(m, caddr_t) + m->m_len;
908 uiop->uio_iovcnt = i;
909 uiop->uio_offset = off;
910 uiop->uio_resid = len;
911 uiop->uio_rw = UIO_READ;
912 uiop->uio_segflg = UIO_SYSSPACE;
913 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
914 off = uiop->uio_offset;
916 FREE((caddr_t)iv2, M_TEMP);
917 if (error || (getret = VOP_GETATTR(vp, vap))) {
923 nfsm_reply(NFSX_POSTOPATTR(v3));
924 nfsm_srvpostop_attr(getret, vap);
933 nfsm_srvfillattr(vap, fp);
934 tlen = len - uiop->uio_resid;
935 cnt = cnt < tlen ? cnt : tlen;
936 tlen = nfsm_rndup(cnt);
937 if (len != tlen || tlen != cnt)
938 nfsm_adj(mb, len - tlen, tlen - cnt);
940 *tl++ = txdr_unsigned(cnt);
946 *tl = txdr_unsigned(cnt);
957 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
958 struct thread *td, struct mbuf **mrq)
960 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
961 struct sockaddr *nam = nfsd->nd_nam;
962 caddr_t dpos = nfsd->nd_dpos;
963 struct ucred *cred = &nfsd->nd_cr;
967 struct nfs_fattr *fp;
969 struct vattr va, forat;
970 struct vattr *vap = &va;
974 int error = 0, rdonly, len, forat_ret = 1;
975 int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
976 int stable = NFSV3WRITE_FILESYNC;
977 int v3 = (nfsd->nd_flag & ND_NFSV3);
979 struct mbuf *mb, *mb2, *mreq;
980 struct vnode *vp = NULL;
981 struct mount *mp = NULL;
984 struct uio io, *uiop = &io;
987 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
993 fhp = &nfh.fh_generic;
996 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
997 off = fxdr_hyper(tl);
999 stable = fxdr_unsigned(int, *tl++);
1001 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1002 off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1005 stable = NFSV3WRITE_UNSTABLE;
1007 retlen = len = fxdr_unsigned(int32_t, *tl);
1011 * For NFS Version 2, it is not obvious what a write of zero length
1012 * should do, but I might as well be consistent with Version 3,
1013 * which is to return ok so long as there are no permission problems.
1021 adjust = dpos - mtod(mp1, caddr_t);
1022 mp1->m_len -= adjust;
1023 if (mp1->m_len > 0 && adjust > 0)
1024 NFSMADV(mp1, adjust);
1028 else if (mp1->m_len > 0) {
1031 mp1->m_len -= (i - len);
1040 if (len > NFS_MAXDATA || len < 0 || i < len) {
1042 nfsm_reply(2 * NFSX_UNSIGNED);
1043 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1047 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
1048 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
1051 nfsm_reply(2 * NFSX_UNSIGNED);
1052 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1057 forat_ret = VOP_GETATTR(vp, &forat);
1058 if (vp->v_type != VREG) {
1062 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1065 error = nfsrv_access(mp, vp, VWRITE, cred, rdonly, td, 1);
1070 nfsm_reply(NFSX_WCCDATA(v3));
1071 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1077 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
1079 uiop->uio_iov = iv = ivp;
1080 uiop->uio_iovcnt = cnt;
1083 if (mp1->m_len > 0) {
1084 ivp->iov_base = mtod(mp1, caddr_t);
1085 ivp->iov_len = mp1->m_len;
1093 * The IO_METASYNC flag indicates that all metadata (and not just
1094 * enough to ensure data integrity) mus be written to stable storage
1096 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1098 if (stable == NFSV3WRITE_UNSTABLE)
1099 ioflags = IO_NODELOCKED;
1100 else if (stable == NFSV3WRITE_DATASYNC)
1101 ioflags = (IO_SYNC | IO_NODELOCKED);
1103 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1104 uiop->uio_resid = len;
1105 uiop->uio_rw = UIO_WRITE;
1106 uiop->uio_segflg = UIO_SYSSPACE;
1107 uiop->uio_td = NULL;
1108 uiop->uio_offset = off;
1109 error = VOP_WRITE(vp, uiop, ioflags, cred);
1110 nfsstats.srvvop_writes++;
1111 FREE((caddr_t)iv, M_TEMP);
1113 aftat_ret = VOP_GETATTR(vp, vap);
1118 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1119 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1121 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1126 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1127 *tl++ = txdr_unsigned(retlen);
1129 * If nfs_async is set, then pretend the write was FILESYNC.
1131 if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1132 *tl++ = txdr_unsigned(stable);
1134 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1136 * Actually, there is no need to txdr these fields,
1137 * but it may make the values more human readable,
1138 * for debugging purposes.
1140 if (nfsver.tv_sec == 0)
1142 *tl++ = txdr_unsigned(nfsver.tv_sec);
1143 *tl = txdr_unsigned(nfsver.tv_nsec / 1000);
1145 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1146 nfsm_srvfillattr(vap, fp);
1155 * NFS write service with write gathering support. Called when
1156 * nfsrvw_procrastinate > 0.
1157 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1158 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1162 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp,
1163 struct thread *td, struct mbuf **mrq)
1166 struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1167 struct nfs_fattr *fp;
1170 struct nfsrvw_delayhash *wpp;
1172 struct vattr va, forat;
1176 int error = 0, rdonly, len, forat_ret = 1;
1177 int ioflags, aftat_ret = 1, adjust, v3, zeroing;
1179 struct mbuf *mb, *mb2, *mreq, *mrep, *md, *mp1;
1180 struct vnode *vp = NULL;
1181 struct mount *mp = NULL;
1182 struct uio io, *uiop = &io;
1185 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1194 mrep = nfsd->nd_mrep;
1196 dpos = nfsd->nd_dpos;
1197 cred = &nfsd->nd_cr;
1198 v3 = (nfsd->nd_flag & ND_NFSV3);
1199 LIST_INIT(&nfsd->nd_coalesce);
1200 nfsd->nd_mreq = NULL;
1201 nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1202 cur_usec = nfs_curusec();
1203 nfsd->nd_time = cur_usec +
1204 (v3 ? nfsrvw_procrastinate_v3 : nfsrvw_procrastinate);
1207 * Now, get the write header..
1209 nfsm_srvmtofh(&nfsd->nd_fh);
1211 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1212 nfsd->nd_off = fxdr_hyper(tl);
1214 nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1216 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1217 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1220 nfsd->nd_stable = NFSV3WRITE_UNSTABLE;
1222 len = fxdr_unsigned(int32_t, *tl);
1224 nfsd->nd_eoff = nfsd->nd_off + len;
1227 * Trim the header out of the mbuf list and trim off any trailing
1228 * junk so that the mbuf list has only the write data.
1236 adjust = dpos - mtod(mp1, caddr_t);
1237 mp1->m_len -= adjust;
1238 if (mp1->m_len > 0 && adjust > 0)
1239 NFSMADV(mp1, adjust);
1246 mp1->m_len -= (i - len);
1252 if (len > NFS_MAXDATA || len < 0 || i < len) {
1256 nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1258 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1259 nfsd->nd_mreq = mreq;
1260 nfsd->nd_mrep = NULL;
1265 * Add this entry to the hash and time queues.
1269 wp = slp->ns_tq.lh_first;
1270 while (wp && wp->nd_time < nfsd->nd_time) {
1272 wp = wp->nd_tq.le_next;
1274 NFS_DPF(WG, ("Q%03x", nfsd->nd_retxid & 0xfff));
1276 LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1278 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1280 if (nfsd->nd_mrep) {
1281 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1285 bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
1287 wp = wp->nd_hash.le_next;
1289 while (wp && wp->nd_off < nfsd->nd_off &&
1290 !bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
1292 wp = wp->nd_hash.le_next;
1295 LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1298 * Search the hash list for overlapping entries and
1301 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1302 wp = nfsd->nd_hash.le_next;
1303 if (NFSW_SAMECRED(owp, nfsd))
1304 nfsrvw_coalesce(owp, nfsd);
1307 LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1314 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1315 * and generate the associated reply mbuf list(s).
1318 cur_usec = nfs_curusec();
1320 for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = owp) {
1321 owp = nfsd->nd_tq.le_next;
1322 if (nfsd->nd_time > cur_usec)
1326 NFS_DPF(WG, ("P%03x", nfsd->nd_retxid & 0xfff));
1327 LIST_REMOVE(nfsd, nd_tq);
1328 LIST_REMOVE(nfsd, nd_hash);
1330 mrep = nfsd->nd_mrep;
1331 nfsd->nd_mrep = NULL;
1332 cred = &nfsd->nd_cr;
1333 v3 = (nfsd->nd_flag & ND_NFSV3);
1334 forat_ret = aftat_ret = 1;
1335 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &mp, &vp, cred, slp,
1336 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
1339 forat_ret = VOP_GETATTR(vp, &forat);
1340 if (vp->v_type != VREG) {
1344 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1350 error = nfsrv_access(mp, vp, VWRITE, cred, rdonly, td, 1);
1353 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1354 ioflags = IO_NODELOCKED;
1355 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1356 ioflags = (IO_SYNC | IO_NODELOCKED);
1358 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1359 uiop->uio_rw = UIO_WRITE;
1360 uiop->uio_segflg = UIO_SYSSPACE;
1361 uiop->uio_td = NULL;
1362 uiop->uio_offset = nfsd->nd_off;
1363 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1364 if (uiop->uio_resid > 0) {
1372 uiop->uio_iovcnt = i;
1373 MALLOC(iov, struct iovec *, i * sizeof (struct iovec),
1375 uiop->uio_iov = ivp = iov;
1378 if (mp1->m_len > 0) {
1379 ivp->iov_base = mtod(mp1, caddr_t);
1380 ivp->iov_len = mp1->m_len;
1386 error = VOP_WRITE(vp, uiop, ioflags, cred);
1387 nfsstats.srvvop_writes++;
1389 FREE((caddr_t)iov, M_TEMP);
1393 aftat_ret = VOP_GETATTR(vp, &va);
1399 * Loop around generating replies for all write rpcs that have
1400 * now been completed.
1404 NFS_DPF(WG, ("R%03x", nfsd->nd_retxid & 0xfff));
1406 nfsm_writereply(NFSX_WCCDATA(v3), v3);
1408 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1411 nfsm_writereply(NFSX_PREOPATTR(v3) +
1412 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1413 NFSX_WRITEVERF(v3), v3);
1415 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1416 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1417 *tl++ = txdr_unsigned(nfsd->nd_len);
1418 *tl++ = txdr_unsigned(swp->nd_stable);
1420 * Actually, there is no need to txdr these fields,
1421 * but it may make the values more human readable,
1422 * for debugging purposes.
1424 if (nfsver.tv_sec == 0)
1426 *tl++ = txdr_unsigned(nfsver.tv_sec);
1427 *tl = txdr_unsigned(nfsver.tv_nsec / 1000);
1429 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1430 nfsm_srvfillattr(&va, fp);
1433 nfsd->nd_mreq = mreq;
1435 panic("nfsrv_write: nd_mrep not free");
1438 * Done. Put it at the head of the timer queue so that
1439 * the final phase can return the reply.
1444 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1446 nfsd = swp->nd_coalesce.lh_first;
1448 LIST_REMOVE(nfsd, nd_tq);
1454 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1461 * Search for a reply to return.
1464 for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = nfsd->nd_tq.le_next)
1465 if (nfsd->nd_mreq) {
1466 NFS_DPF(WG, ("X%03x", nfsd->nd_retxid & 0xfff));
1467 LIST_REMOVE(nfsd, nd_tq);
1468 *mrq = nfsd->nd_mreq;
1477 * Coalesce the write request nfsd into owp. To do this we must:
1478 * - remove nfsd from the queues
1479 * - merge nfsd->nd_mrep into owp->nd_mrep
1480 * - update the nd_eoff and nd_stable for owp
1481 * - put nfsd on owp's nd_coalesce list
1482 * NB: Must be called at splsoftclock().
1485 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1489 struct nfsrv_descript *p;
1491 NFS_DPF(WG, ("C%03x-%03x",
1492 nfsd->nd_retxid & 0xfff, owp->nd_retxid & 0xfff));
1493 LIST_REMOVE(nfsd, nd_hash);
1494 LIST_REMOVE(nfsd, nd_tq);
1495 if (owp->nd_eoff < nfsd->nd_eoff) {
1496 overlap = owp->nd_eoff - nfsd->nd_off;
1498 panic("nfsrv_coalesce: bad off");
1500 m_adj(nfsd->nd_mrep, overlap);
1504 mp1->m_next = nfsd->nd_mrep;
1505 owp->nd_eoff = nfsd->nd_eoff;
1507 m_freem(nfsd->nd_mrep);
1508 nfsd->nd_mrep = NULL;
1509 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1510 owp->nd_stable = NFSV3WRITE_FILESYNC;
1511 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1512 owp->nd_stable == NFSV3WRITE_UNSTABLE)
1513 owp->nd_stable = NFSV3WRITE_DATASYNC;
1514 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1517 * If nfsd had anything else coalesced into it, transfer them
1518 * to owp, otherwise their replies will never get sent.
1520 for (p = nfsd->nd_coalesce.lh_first; p;
1521 p = nfsd->nd_coalesce.lh_first) {
1522 LIST_REMOVE(p, nd_tq);
1523 LIST_INSERT_HEAD(&owp->nd_coalesce, p, nd_tq);
1528 * nfs create service
1529 * now does a truncate to 0 length via. setattr if it already exists
1532 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1533 struct thread *td, struct mbuf **mrq)
1535 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1536 struct sockaddr *nam = nfsd->nd_nam;
1537 caddr_t dpos = nfsd->nd_dpos;
1538 struct ucred *cred = &nfsd->nd_cr;
1539 struct nfs_fattr *fp;
1540 struct vattr va, dirfor, diraft;
1541 struct vattr *vap = &va;
1542 struct nfsv2_sattr *sp;
1544 struct nlookupdata nd;
1547 int error = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1548 udev_t rdev = NOUDEV;
1549 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1552 struct mbuf *mb, *mb2, *mreq;
1560 u_char cverf[NFSX_V3CREATEVERF];
1562 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1568 fhp = &nfh.fh_generic;
1570 nfsm_srvnamesiz(len);
1573 * Call namei and do initial cleanup to get a few things
1574 * out of the way. If we get an initial error we cleanup
1575 * and return here to avoid special-casing the invalid nd
1576 * structure through the rest of the case. dirp may be
1577 * set even if an error occurs, but the nd structure will not
1578 * be valid at all if an error occurs so we have to invalidate it
1579 * prior to calling nfsm_reply ( which might goto nfsmout ).
1581 error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
1582 fhp, len, slp, nam, &md, &dpos, &dirp,
1583 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1584 mp = vfs_getvfs(&fhp->fh_fsid);
1588 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
1595 nfsm_reply(NFSX_WCCDATA(v3));
1596 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1602 * No error. Continue. State:
1605 * vp may be valid or NULL if the target does not
1609 * The error state is set through the code and we may also do some
1610 * opportunistic releasing of vnodes to avoid holding locks through
1611 * NFS I/O. The cleanup at the end is a catch-all
1616 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1617 how = fxdr_unsigned(int, *tl);
1619 case NFSV3CREATE_GUARDED:
1625 case NFSV3CREATE_UNCHECKED:
1628 case NFSV3CREATE_EXCLUSIVE:
1629 nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1630 bcopy(cp, cverf, NFSX_V3CREATEVERF);
1634 vap->va_type = VREG;
1636 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1637 vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1638 if (vap->va_type == VNON)
1639 vap->va_type = VREG;
1640 vap->va_mode = nfstov_mode(sp->sa_mode);
1641 switch (vap->va_type) {
1643 tsize = fxdr_unsigned(int32_t, sp->sa_size);
1645 vap->va_size = (u_quad_t)tsize;
1650 rdev = fxdr_unsigned(long, sp->sa_size);
1658 * Iff doesn't exist, create it
1659 * otherwise just truncate to 0 length
1660 * should I set the mode too ?
1662 * The only possible error we can have at this point is EEXIST.
1663 * nd.ni_vp will also be non-NULL in that case.
1666 if (vap->va_mode == (mode_t)VNOVAL)
1668 if (vap->va_type == VREG || vap->va_type == VSOCK) {
1670 error = VOP_NCREATE(&nd.nl_nch, dvp, &vp,
1675 if (exclusive_flag) {
1678 bcopy(cverf, (caddr_t)&vap->va_atime,
1680 error = VOP_SETATTR(vp, vap, cred);
1684 vap->va_type == VCHR ||
1685 vap->va_type == VBLK ||
1686 vap->va_type == VFIFO
1689 * Handle SysV FIFO node special cases. All other
1690 * devices require super user to access.
1692 if (vap->va_type == VCHR && rdev == 0xffffffff)
1693 vap->va_type = VFIFO;
1694 if (vap->va_type != VFIFO &&
1695 (error = suser_cred(cred, 0))) {
1698 vap->va_rmajor = umajor(rdev);
1699 vap->va_rminor = uminor(rdev);
1702 error = VOP_NMKNOD(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
1709 * XXX what is this junk supposed to do ?
1716 * release dvp prior to lookup
1724 * Even though LOCKPARENT was cleared, ni_dvp may
1727 nd.ni_cnd.cn_nameiop = NAMEI_LOOKUP;
1728 nd.ni_cnd.cn_flags &= ~(CNP_LOCKPARENT);
1729 nd.ni_cnd.cn_td = td;
1730 nd.ni_cnd.cn_cred = cred;
1732 error = lookup(&nd);
1737 /* fall through on certain errors */
1739 nfsrv_object_create(nd.ni_vp);
1740 if (nd.ni_cnd.cn_flags & CNP_ISSYMLINK) {
1749 if (vap->va_size != -1) {
1750 error = nfsrv_access(mp, vp, VWRITE, cred,
1751 (nd.nl_flags & NLC_NFS_RDONLY), td, 0);
1753 tempsize = vap->va_size;
1755 vap->va_size = tempsize;
1756 error = VOP_SETATTR(vp, vap, cred);
1762 bzero(&fhp->fh_fid, sizeof(fhp->fh_fid));
1763 error = VFS_VPTOFH(vp, &fhp->fh_fid);
1765 error = VOP_GETATTR(vp, vap);
1768 if (exclusive_flag && !error &&
1769 bcmp(cverf, (caddr_t)&vap->va_atime, NFSX_V3CREATEVERF))
1771 diraft_ret = VOP_GETATTR(dirp, &diraft);
1775 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1778 nfsm_srvpostop_fh(fhp);
1779 nfsm_srvpostop_attr(0, vap);
1781 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1784 nfsm_srvfhtom(fhp, v3);
1785 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1786 nfsm_srvfillattr(vap, fp);
1811 * nfs v3 mknod service
1814 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1815 struct thread *td, struct mbuf **mrq)
1817 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1818 struct sockaddr *nam = nfsd->nd_nam;
1819 caddr_t dpos = nfsd->nd_dpos;
1820 struct ucred *cred = &nfsd->nd_cr;
1821 struct vattr va, dirfor, diraft;
1822 struct vattr *vap = &va;
1824 struct nlookupdata nd;
1827 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1830 struct mbuf *mb, *mb2, *mreq;
1837 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1843 fhp = &nfh.fh_generic;
1845 nfsm_srvnamesiz(len);
1848 * Handle nfs_namei() call. If an error occurs, the nd structure
1849 * is not valid. However, nfsm_*() routines may still jump to
1853 error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
1854 fhp, len, slp, nam, &md, &dpos, &dirp,
1855 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1857 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
1859 nfsm_reply(NFSX_WCCDATA(1));
1860 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1864 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1865 vtyp = nfsv3tov_type(*tl);
1866 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1867 error = NFSERR_BADTYPE;
1872 if (vtyp == VCHR || vtyp == VBLK) {
1873 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1874 vap->va_rmajor = fxdr_unsigned(u_int32_t, *tl++);
1875 vap->va_rminor = fxdr_unsigned(u_int32_t, *tl);
1879 * Iff doesn't exist, create it.
1885 vap->va_type = vtyp;
1886 if (vap->va_mode == (mode_t)VNOVAL)
1888 if (vtyp == VSOCK) {
1890 error = VOP_NCREATE(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
1894 if (vtyp != VFIFO && (error = suser_cred(cred, 0)))
1898 error = VOP_NMKNOD(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
1906 * send response, cleanup, return.
1918 bzero(&fhp->fh_fid, sizeof(fhp->fh_fid));
1919 error = VFS_VPTOFH(vp, &fhp->fh_fid);
1921 error = VOP_GETATTR(vp, vap);
1927 diraft_ret = VOP_GETATTR(dirp, &diraft);
1932 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1934 nfsm_srvpostop_fh(fhp);
1935 nfsm_srvpostop_attr(0, vap);
1937 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1955 * nfs remove service
1958 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1959 struct thread *td, struct mbuf **mrq)
1961 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1962 struct sockaddr *nam = nfsd->nd_nam;
1963 caddr_t dpos = nfsd->nd_dpos;
1964 struct ucred *cred = &nfsd->nd_cr;
1965 struct nlookupdata nd;
1969 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1970 int v3 = (nfsd->nd_flag & ND_NFSV3);
1972 struct mbuf *mb, *mreq;
1976 struct vattr dirfor, diraft;
1980 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1986 fhp = &nfh.fh_generic;
1988 nfsm_srvnamesiz(len);
1990 error = nfs_namei(&nd, cred, NAMEI_DELETE, &dvp, &vp,
1991 fhp, len, slp, nam, &md, &dpos, &dirp,
1992 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1995 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
1998 if (vp->v_type == VDIR) {
1999 error = EPERM; /* POSIX */
2003 * The root of a mounted filesystem cannot be deleted.
2005 if (vp->v_flag & VROOT) {
2017 error = VOP_NREMOVE(&nd.nl_nch, dvp, nd.nl_cred);
2023 diraft_ret = VOP_GETATTR(dirp, &diraft);
2024 nfsm_reply(NFSX_WCCDATA(v3));
2026 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2045 * nfs rename service
2048 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2049 struct thread *td, struct mbuf **mrq)
2051 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2052 struct sockaddr *nam = nfsd->nd_nam;
2053 caddr_t dpos = nfsd->nd_dpos;
2054 struct ucred *cred = &nfsd->nd_cr;
2058 int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
2059 int tdirfor_ret = 1, tdiraft_ret = 1;
2060 int v3 = (nfsd->nd_flag & ND_NFSV3);
2062 struct mbuf *mb, *mreq;
2063 struct nlookupdata fromnd, tond;
2064 struct vnode *fvp, *fdirp, *fdvp;
2065 struct vnode *tvp, *tdirp, *tdvp;
2066 struct namecache *ncp;
2067 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
2069 fhandle_t *ffhp, *tfhp;
2072 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2074 fvp = (struct vnode *)0;
2076 ffhp = &fnfh.fh_generic;
2077 tfhp = &tnfh.fh_generic;
2080 * Clear fields incase goto nfsmout occurs from macro.
2083 nlookup_zero(&fromnd);
2084 nlookup_zero(&tond);
2088 nfsm_srvmtofh(ffhp);
2089 nfsm_srvnamesiz(len);
2091 * Remember our original uid so that we can reset cr_uid before
2092 * the second nfs_namei() call, in case it is remapped.
2094 saved_uid = cred->cr_uid;
2095 error = nfs_namei(&fromnd, cred, NAMEI_DELETE, NULL, NULL,
2096 ffhp, len, slp, nam, &md, &dpos, &fdirp,
2097 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2100 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor);
2103 nfsm_reply(2 * NFSX_WCCDATA(v3));
2104 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2105 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2111 * We have to unlock the from ncp before we can safely lookup
2114 KKASSERT(fromnd.nl_flags & NLC_NCPISLOCKED);
2115 cache_unlock(&fromnd.nl_nch);
2116 fromnd.nl_flags &= ~NLC_NCPISLOCKED;
2117 nfsm_srvmtofh(tfhp);
2118 nfsm_strsiz(len2, NFS_MAXNAMLEN);
2119 cred->cr_uid = saved_uid;
2121 error = nfs_namei(&tond, cred, NAMEI_RENAME, NULL, NULL,
2122 tfhp, len2, slp, nam, &md, &dpos, &tdirp,
2123 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2126 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor);
2134 if (cache_lock_nonblock(&fromnd.nl_nch) == 0) {
2135 cache_resolve(&fromnd.nl_nch, fromnd.nl_cred);
2136 } else if (fromnd.nl_nch.ncp > tond.nl_nch.ncp) {
2137 cache_lock(&fromnd.nl_nch);
2138 cache_resolve(&fromnd.nl_nch, fromnd.nl_cred);
2140 cache_unlock(&tond.nl_nch);
2141 cache_lock(&fromnd.nl_nch);
2142 cache_resolve(&fromnd.nl_nch, fromnd.nl_cred);
2143 cache_lock(&tond.nl_nch);
2144 cache_resolve(&tond.nl_nch, tond.nl_cred);
2146 fromnd.nl_flags |= NLC_NCPISLOCKED;
2148 fvp = fromnd.nl_nch.ncp->nc_vp;
2149 tvp = tond.nl_nch.ncp->nc_vp;
2152 * Set fdvp and tdvp. We haven't done all the topology checks
2153 * so these can wind up NULL (e.g. if either fvp or tvp is a mount
2154 * point). If we get through the checks these will be guarenteed
2157 * Holding the children ncp's should be sufficient to prevent
2158 * fdvp and tdvp ripouts.
2160 if (fromnd.nl_nch.ncp->nc_parent)
2161 fdvp = fromnd.nl_nch.ncp->nc_parent->nc_vp;
2164 if (tond.nl_nch.ncp->nc_parent)
2165 tdvp = tond.nl_nch.ncp->nc_parent->nc_vp;
2170 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2176 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2183 if (tvp->v_type == VDIR && (tond.nl_nch.ncp->nc_flag & NCF_ISMOUNTPT)) {
2191 if (fvp->v_type == VDIR && (fromnd.nl_nch.ncp->nc_flag & NCF_ISMOUNTPT)) {
2198 if (fromnd.nl_nch.mount != tond.nl_nch.mount) {
2205 if (fromnd.nl_nch.ncp == tond.nl_nch.ncp->nc_parent) {
2213 * You cannot rename a source into itself or a subdirectory of itself.
2214 * We check this by travsering the target directory upwards looking
2215 * for a match against the source.
2218 for (ncp = tond.nl_nch.ncp; ncp; ncp = ncp->nc_parent) {
2219 if (fromnd.nl_nch.ncp == ncp) {
2227 * If source is the same as the destination (that is the
2228 * same vnode with the same name in the same directory),
2229 * then there is nothing to do.
2231 if (fromnd.nl_nch.ncp == tond.nl_nch.ncp)
2236 * The VOP_NRENAME function releases all vnode references &
2237 * locks prior to returning so we need to clear the pointers
2238 * to bypass cleanup code later on.
2240 error = VOP_NRENAME(&fromnd.nl_nch, &tond.nl_nch,
2241 fdvp, tdvp, tond.nl_cred);
2250 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft);
2252 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft);
2253 nfsm_reply(2 * NFSX_WCCDATA(v3));
2255 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2256 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2264 nlookup_done(&tond);
2267 nlookup_done(&fromnd);
2275 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2276 struct thread *td, struct mbuf **mrq)
2278 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2279 struct sockaddr *nam = nfsd->nd_nam;
2280 caddr_t dpos = nfsd->nd_dpos;
2281 struct ucred *cred = &nfsd->nd_cr;
2282 struct nlookupdata nd;
2286 int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1;
2287 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2289 struct mbuf *mb, *mreq;
2296 struct vattr dirfor, diraft, at;
2298 fhandle_t *fhp, *dfhp;
2300 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2302 dirp = dvp = vp = xp = NULL;
2305 fhp = &nfh.fh_generic;
2306 dfhp = &dnfh.fh_generic;
2308 nfsm_srvmtofh(dfhp);
2309 nfsm_srvnamesiz(len);
2311 error = nfsrv_fhtovp(fhp, FALSE, &xmp, &xp, cred, slp, nam,
2312 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
2314 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2315 nfsm_srvpostop_attr(getret, &at);
2316 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2321 if (xp->v_type == VDIR) {
2322 error = EPERM; /* POSIX */
2326 error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
2327 dfhp, len, slp, nam, &md, &dpos, &dirp,
2328 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2331 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2340 if (xp->v_mount != dvp->v_mount)
2345 error = VOP_NLINK(&nd.nl_nch, dvp, xp, nd.nl_cred);
2353 getret = VOP_GETATTR(xp, &at);
2355 diraft_ret = VOP_GETATTR(dirp, &diraft);
2356 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2358 nfsm_srvpostop_attr(getret, &at);
2359 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2382 * nfs symbolic link service
2385 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2386 struct thread *td, struct mbuf **mrq)
2388 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2389 struct sockaddr *nam = nfsd->nd_nam;
2390 caddr_t dpos = nfsd->nd_dpos;
2391 struct ucred *cred = &nfsd->nd_cr;
2392 struct vattr va, dirfor, diraft;
2393 struct nlookupdata nd;
2394 struct vattr *vap = &va;
2397 struct nfsv2_sattr *sp;
2398 char *bpos, *pathcp = (char *)0, *cp2;
2401 int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2402 int v3 = (nfsd->nd_flag & ND_NFSV3);
2403 struct mbuf *mb, *mreq, *mb2;
2410 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2416 fhp = &nfh.fh_generic;
2418 nfsm_srvnamesiz(len);
2420 error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
2421 fhp, len, slp, nam, &md, &dpos, &dirp,
2422 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2425 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2433 nfsm_strsiz(len2, NFS_MAXPATHLEN);
2434 MALLOC(pathcp, caddr_t, len2 + 1, M_TEMP, M_WAITOK);
2435 iv.iov_base = pathcp;
2437 io.uio_resid = len2;
2441 io.uio_segflg = UIO_SYSSPACE;
2442 io.uio_rw = UIO_READ;
2444 nfsm_mtouio(&io, len2);
2446 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2447 vap->va_mode = nfstov_mode(sp->sa_mode);
2449 *(pathcp + len2) = '\0';
2455 if (vap->va_mode == (mode_t)VNOVAL)
2459 error = VOP_NSYMLINK(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap, pathcp);
2463 bzero(&fhp->fh_fid, sizeof(fhp->fh_fid));
2464 error = VFS_VPTOFH(vp, &fhp->fh_fid);
2466 error = VOP_GETATTR(vp, vap);
2481 FREE(pathcp, M_TEMP);
2485 diraft_ret = VOP_GETATTR(dirp, &diraft);
2489 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2492 nfsm_srvpostop_fh(fhp);
2493 nfsm_srvpostop_attr(0, vap);
2495 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2507 FREE(pathcp, M_TEMP);
2515 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2516 struct thread *td, struct mbuf **mrq)
2518 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2519 struct sockaddr *nam = nfsd->nd_nam;
2520 caddr_t dpos = nfsd->nd_dpos;
2521 struct ucred *cred = &nfsd->nd_cr;
2522 struct vattr va, dirfor, diraft;
2523 struct vattr *vap = &va;
2524 struct nfs_fattr *fp;
2525 struct nlookupdata nd;
2530 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2531 int v3 = (nfsd->nd_flag & ND_NFSV3);
2533 struct mbuf *mb, *mb2, *mreq;
2540 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2546 fhp = &nfh.fh_generic;
2548 nfsm_srvnamesiz(len);
2550 error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
2551 fhp, len, slp, nam, &md, &dpos, &dirp,
2552 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2555 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2558 nfsm_reply(NFSX_WCCDATA(v3));
2559 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2567 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2568 vap->va_mode = nfstov_mode(*tl++);
2572 * At this point nd.ni_dvp is referenced and exclusively locked and
2573 * nd.ni_vp, if it exists, is referenced but not locked.
2576 vap->va_type = VDIR;
2583 * Issue mkdir op. Since SAVESTART is not set, the pathname
2584 * component is freed by the VOP call. This will fill-in
2585 * nd.ni_vp, reference, and exclusively lock it.
2587 if (vap->va_mode == (mode_t)VNOVAL)
2590 error = VOP_NMKDIR(&nd.nl_nch, dvp, &vp, nd.nl_cred, vap);
2595 bzero(&fhp->fh_fid, sizeof(fhp->fh_fid));
2596 error = VFS_VPTOFH(vp, &fhp->fh_fid);
2598 error = VOP_GETATTR(vp, vap);
2602 diraft_ret = VOP_GETATTR(dirp, &diraft);
2603 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2606 nfsm_srvpostop_fh(fhp);
2607 nfsm_srvpostop_attr(0, vap);
2609 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2611 nfsm_srvfhtom(fhp, v3);
2612 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2613 nfsm_srvfillattr(vap, fp);
2637 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2638 struct thread *td, struct mbuf **mrq)
2640 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2641 struct sockaddr *nam = nfsd->nd_nam;
2642 caddr_t dpos = nfsd->nd_dpos;
2643 struct ucred *cred = &nfsd->nd_cr;
2647 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2648 int v3 = (nfsd->nd_flag & ND_NFSV3);
2650 struct mbuf *mb, *mreq;
2654 struct vattr dirfor, diraft;
2657 struct nlookupdata nd;
2659 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2665 fhp = &nfh.fh_generic;
2667 nfsm_srvnamesiz(len);
2669 error = nfs_namei(&nd, cred, NAMEI_DELETE, &dvp, &vp,
2670 fhp, len, slp, nam, &md, &dpos, &dirp,
2671 td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2674 dirfor_ret = VOP_GETATTR(dirp, &dirfor);
2677 nfsm_reply(NFSX_WCCDATA(v3));
2678 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2682 if (vp->v_type != VDIR) {
2688 * The root of a mounted filesystem cannot be deleted.
2690 if (vp->v_flag & VROOT)
2694 * Issue or abort op. Since SAVESTART is not set, path name
2695 * component is freed by the VOP after either.
2702 error = VOP_NRMDIR(&nd.nl_nch, dvp, nd.nl_cred);
2709 diraft_ret = VOP_GETATTR(dirp, &diraft);
2710 nfsm_reply(NFSX_WCCDATA(v3));
2712 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2733 * nfs readdir service
2734 * - mallocs what it thinks is enough to read
2735 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2736 * - calls VOP_READDIR()
2737 * - loops around building the reply
2738 * if the output generated exceeds count break out of loop
2739 * The nfsm_clget macro is used here so that the reply will be packed
2740 * tightly in mbuf clusters.
2741 * - it only knows that it has encountered eof when the VOP_READDIR()
2743 * - as such one readdir rpc will return eof false although you are there
2744 * and then the next will return eof
2745 * - it trims out records with d_fileno == 0
2746 * this doesn't matter for Unix clients, but they might confuse clients
2748 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2749 * than requested, but this may not apply to all filesystems. For
2750 * example, client NFS does not { although it is never remote mounted
2752 * The alternate call nfsrv_readdirplus() does lookups as well.
2753 * PS: The NFS protocol spec. does not clarify what the "count" byte
2754 * argument is a count of.. just name strings and file id's or the
2755 * entire reply rpc or ...
2756 * I tried just file name and id sizes and it confused the Sun client,
2757 * so I am using the full rpc size now. The "paranoia.." comment refers
2758 * to including the status longwords that are not a part of the dir.
2759 * "entry" structures, but are in the rpc.
2763 u_int32_t fl_postopok;
2764 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2766 u_int32_t fl_fhsize;
2767 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2771 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2772 struct thread *td, struct mbuf **mrq)
2774 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2775 struct sockaddr *nam = nfsd->nd_nam;
2776 caddr_t dpos = nfsd->nd_dpos;
2777 struct ucred *cred = &nfsd->nd_cr;
2784 struct mbuf *mb, *mb2, *mreq, *mp1, *mp2;
2785 char *cpos, *cend, *cp2, *rbuf;
2786 struct vnode *vp = NULL;
2787 struct mount *mp = NULL;
2793 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2794 int siz, cnt, fullsiz, eofflag, rdonly, ncookies;
2795 int v3 = (nfsd->nd_flag & ND_NFSV3);
2796 u_quad_t off, toff, verf;
2797 off_t *cookies = NULL, *cookiep;
2799 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2800 fhp = &nfh.fh_generic;
2803 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2804 toff = fxdr_hyper(tl);
2806 verf = fxdr_hyper(tl);
2809 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2810 toff = fxdr_unsigned(u_quad_t, *tl++);
2811 verf = 0; /* shut up gcc */
2814 cnt = fxdr_unsigned(int, *tl);
2815 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2816 xfer = NFS_SRVMAXDATA(nfsd);
2822 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
2823 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
2824 if (!error && vp->v_type != VDIR) {
2830 nfsm_reply(NFSX_UNSIGNED);
2831 nfsm_srvpostop_attr(getret, &at);
2837 * Obtain lock on vnode for this section of the code
2841 error = getret = VOP_GETATTR(vp, &at);
2844 * XXX This check may be too strict for Solaris 2.5 clients.
2846 if (!error && toff && verf && verf != at.va_filerev)
2847 error = NFSERR_BAD_COOKIE;
2851 error = nfsrv_access(mp, vp, VEXEC, cred, rdonly, td, 0);
2855 nfsm_reply(NFSX_POSTOPATTR(v3));
2856 nfsm_srvpostop_attr(getret, &at);
2863 * end section. Allocate rbuf and continue
2865 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
2868 iv.iov_len = fullsiz;
2871 io.uio_offset = (off_t)off;
2872 io.uio_resid = fullsiz;
2873 io.uio_segflg = UIO_SYSSPACE;
2874 io.uio_rw = UIO_READ;
2878 kfree((caddr_t)cookies, M_TEMP);
2881 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
2882 off = (off_t)io.uio_offset;
2883 if (!cookies && !error)
2884 error = NFSERR_PERM;
2886 getret = VOP_GETATTR(vp, &at);
2893 kfree((caddr_t)rbuf, M_TEMP);
2895 kfree((caddr_t)cookies, M_TEMP);
2896 nfsm_reply(NFSX_POSTOPATTR(v3));
2897 nfsm_srvpostop_attr(getret, &at);
2902 siz -= io.uio_resid;
2905 * If nothing read, return eof
2911 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2914 nfsm_srvpostop_attr(getret, &at);
2915 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2916 txdr_hyper(at.va_filerev, tl);
2919 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2922 FREE((caddr_t)rbuf, M_TEMP);
2923 FREE((caddr_t)cookies, M_TEMP);
2930 * Check for degenerate cases of nothing useful read.
2931 * If so go try again
2935 dp = (struct dirent *)cpos;
2938 * For some reason FreeBSD's ufs_readdir() chooses to back the
2939 * directory offset up to a block boundary, so it is necessary to
2940 * skip over the records that preceed the requested offset. This
2941 * requires the assumption that file offset cookies monotonically
2944 while (cpos < cend && ncookies > 0 &&
2945 (dp->d_ino == 0 || dp->d_type == DT_WHT ||
2946 ((u_quad_t)(*cookiep)) <= toff)) {
2947 dp = _DIRENT_NEXT(dp);
2952 if (cpos >= cend || ncookies == 0) {
2958 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
2959 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2961 nfsm_srvpostop_attr(getret, &at);
2962 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2963 txdr_hyper(at.va_filerev, tl);
2967 be = bp + M_TRAILINGSPACE(mp1);
2969 /* Loop through the records and build reply */
2970 while (cpos < cend && ncookies > 0) {
2971 if (dp->d_ino != 0 && dp->d_type != DT_WHT) {
2972 nlen = dp->d_namlen;
2973 rem = nfsm_rndup(nlen) - nlen;
2974 len += (4 * NFSX_UNSIGNED + nlen + rem);
2976 len += 2 * NFSX_UNSIGNED;
2982 * Build the directory record xdr from
2985 nfsm_clget(mp1, mp2, mb, bp, be, tl);
2987 bp += NFSX_UNSIGNED;
2989 nfsm_clget(mp1, mp2, mb, bp, be, tl);
2991 bp += NFSX_UNSIGNED;
2993 nfsm_clget(mp1, mp2, mb, bp, be, tl);
2994 *tl = txdr_unsigned(dp->d_ino);
2995 bp += NFSX_UNSIGNED;
2996 nfsm_clget(mp1, mp2, mb, bp, be, tl);
2997 *tl = txdr_unsigned(nlen);
2998 bp += NFSX_UNSIGNED;
3000 /* And loop around copying the name */
3004 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3009 bcopy(cp, bp, tsiz);
3015 /* And null pad to a int32_t boundary */
3016 for (i = 0; i < rem; i++)
3018 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3020 /* Finish off the record */
3022 *tl = txdr_unsigned(*cookiep >> 32);
3023 bp += NFSX_UNSIGNED;
3024 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3026 *tl = txdr_unsigned(*cookiep);
3027 bp += NFSX_UNSIGNED;
3029 dp = _DIRENT_NEXT(dp);
3036 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3038 bp += NFSX_UNSIGNED;
3039 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3044 bp += NFSX_UNSIGNED;
3047 mp1->m_len = bp - mtod(mp1, caddr_t);
3049 mp1->m_len += bp - bpos;
3050 FREE((caddr_t)rbuf, M_TEMP);
3051 FREE((caddr_t)cookies, M_TEMP);
3060 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3061 struct thread *td, struct mbuf **mrq)
3063 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3064 struct sockaddr *nam = nfsd->nd_nam;
3065 caddr_t dpos = nfsd->nd_dpos;
3066 struct ucred *cred = &nfsd->nd_cr;
3073 struct mbuf *mb, *mb2, *mreq, *mp1, *mp2;
3074 char *cpos, *cend, *cp2, *rbuf;
3075 struct vnode *vp = NULL, *nvp;
3076 struct mount *mp = NULL;
3079 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3082 struct vattr va, at, *vap = &va;
3083 struct nfs_fattr *fp;
3084 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3085 int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies;
3086 u_quad_t off, toff, verf;
3087 off_t *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3089 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3090 fhp = &nfh.fh_generic;
3092 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
3093 toff = fxdr_hyper(tl);
3095 verf = fxdr_hyper(tl);
3097 siz = fxdr_unsigned(int, *tl++);
3098 cnt = fxdr_unsigned(int, *tl);
3100 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3101 xfer = NFS_SRVMAXDATA(nfsd);
3107 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
3108 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3109 if (!error && vp->v_type != VDIR) {
3115 nfsm_reply(NFSX_UNSIGNED);
3116 nfsm_srvpostop_attr(getret, &at);
3120 error = getret = VOP_GETATTR(vp, &at);
3123 * XXX This check may be too strict for Solaris 2.5 clients.
3125 if (!error && toff && verf && verf != at.va_filerev)
3126 error = NFSERR_BAD_COOKIE;
3129 error = nfsrv_access(mp, vp, VEXEC, cred, rdonly, td, 0);
3134 nfsm_reply(NFSX_V3POSTOPATTR);
3135 nfsm_srvpostop_attr(getret, &at);
3140 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3143 iv.iov_len = fullsiz;
3146 io.uio_offset = (off_t)off;
3147 io.uio_resid = fullsiz;
3148 io.uio_segflg = UIO_SYSSPACE;
3149 io.uio_rw = UIO_READ;
3153 kfree((caddr_t)cookies, M_TEMP);
3156 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3157 off = (u_quad_t)io.uio_offset;
3158 getret = VOP_GETATTR(vp, &at);
3159 if (!cookies && !error)
3160 error = NFSERR_PERM;
3167 kfree((caddr_t)cookies, M_TEMP);
3168 kfree((caddr_t)rbuf, M_TEMP);
3169 nfsm_reply(NFSX_V3POSTOPATTR);
3170 nfsm_srvpostop_attr(getret, &at);
3175 siz -= io.uio_resid;
3178 * If nothing read, return eof
3184 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3186 nfsm_srvpostop_attr(getret, &at);
3187 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3188 txdr_hyper(at.va_filerev, tl);
3192 FREE((caddr_t)cookies, M_TEMP);
3193 FREE((caddr_t)rbuf, M_TEMP);
3200 * Check for degenerate cases of nothing useful read.
3201 * If so go try again
3205 dp = (struct dirent *)cpos;
3208 * For some reason FreeBSD's ufs_readdir() chooses to back the
3209 * directory offset up to a block boundary, so it is necessary to
3210 * skip over the records that preceed the requested offset. This
3211 * requires the assumption that file offset cookies monotonically
3214 while (cpos < cend && ncookies > 0 &&
3215 (dp->d_ino == 0 || dp->d_type == DT_WHT ||
3216 ((u_quad_t)(*cookiep)) <= toff)) {
3217 dp = _DIRENT_NEXT(dp);
3222 if (cpos >= cend || ncookies == 0) {
3229 * Probe one of the directory entries to see if the filesystem
3232 if (VFS_VGET(vp->v_mount, dp->d_ino, &nvp) == EOPNOTSUPP) {
3233 error = NFSERR_NOTSUPP;
3236 kfree((caddr_t)cookies, M_TEMP);
3237 kfree((caddr_t)rbuf, M_TEMP);
3238 nfsm_reply(NFSX_V3POSTOPATTR);
3239 nfsm_srvpostop_attr(getret, &at);
3248 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
3250 nfsm_srvpostop_attr(getret, &at);
3251 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3252 txdr_hyper(at.va_filerev, tl);
3255 be = bp + M_TRAILINGSPACE(mp1);
3257 /* Loop through the records and build reply */
3258 while (cpos < cend && ncookies > 0) {
3259 if (dp->d_ino != 0 && dp->d_type != DT_WHT) {
3260 nlen = dp->d_namlen;
3261 rem = nfsm_rndup(nlen)-nlen;
3264 * For readdir_and_lookup get the vnode using
3267 if (VFS_VGET(vp->v_mount, dp->d_ino, &nvp))
3269 bzero((caddr_t)nfhp, NFSX_V3FH);
3270 nfhp->fh_fsid = fhp->fh_fsid;
3271 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
3276 if (VOP_GETATTR(nvp, vap)) {
3285 * If either the dircount or maxcount will be
3286 * exceeded, get out now. Both of these lengths
3287 * are calculated conservatively, including all
3290 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3292 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3293 if (len > cnt || dirlen > fullsiz) {
3299 * Build the directory record xdr from
3302 fp = (struct nfs_fattr *)&fl.fl_fattr;
3303 nfsm_srvfillattr(vap, fp);
3304 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3305 fl.fl_fhok = nfs_true;
3306 fl.fl_postopok = nfs_true;
3307 fl.fl_off.nfsuquad[0] = txdr_unsigned(*cookiep >> 32);
3308 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3310 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3312 bp += NFSX_UNSIGNED;
3313 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3315 bp += NFSX_UNSIGNED;
3316 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3317 *tl = txdr_unsigned(dp->d_ino);
3318 bp += NFSX_UNSIGNED;
3319 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3320 *tl = txdr_unsigned(nlen);
3321 bp += NFSX_UNSIGNED;
3323 /* And loop around copying the name */
3327 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3328 if ((bp + xfer) > be)
3332 bcopy(cp, bp, tsiz);
3338 /* And null pad to a int32_t boundary */
3339 for (i = 0; i < rem; i++)
3343 * Now copy the flrep structure out.
3345 xfer = sizeof (struct flrep);
3348 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3349 if ((bp + xfer) > be)
3353 bcopy(cp, bp, tsiz);
3361 dp = _DIRENT_NEXT(dp);
3368 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3370 bp += NFSX_UNSIGNED;
3371 nfsm_clget(mp1, mp2, mb, bp, be, tl);
3376 bp += NFSX_UNSIGNED;
3379 mp1->m_len = bp - mtod(mp1, caddr_t);
3381 mp1->m_len += bp - bpos;
3382 FREE((caddr_t)cookies, M_TEMP);
3383 FREE((caddr_t)rbuf, M_TEMP);
3391 * nfs commit service
3394 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3395 struct thread *td, struct mbuf **mrq)
3397 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3398 struct sockaddr *nam = nfsd->nd_nam;
3399 caddr_t dpos = nfsd->nd_dpos;
3400 struct ucred *cred = &nfsd->nd_cr;
3401 struct vattr bfor, aft;
3402 struct vnode *vp = NULL;
3403 struct mount *mp = NULL;
3409 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt;
3411 struct mbuf *mb, *mb2, *mreq;
3414 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3415 fhp = &nfh.fh_generic;
3417 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3420 * XXX At this time VOP_FSYNC() does not accept offset and byte
3421 * count parameters, so these arguments are useless (someday maybe).
3423 off = fxdr_hyper(tl);
3425 cnt = fxdr_unsigned(int, *tl);
3426 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
3427 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3429 nfsm_reply(2 * NFSX_UNSIGNED);
3430 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3434 for_ret = VOP_GETATTR(vp, &bfor);
3436 if (cnt > MAX_COMMIT_COUNT) {
3438 * Give up and do the whole thing
3441 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3442 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3444 error = VOP_FSYNC(vp, MNT_WAIT);
3447 * Locate and synchronously write any buffers that fall
3448 * into the requested range. Note: we are assuming that
3449 * f_iosize is a power of 2.
3451 int iosize = vp->v_mount->mnt_stat.f_iosize;
3452 int iomask = iosize - 1;
3456 * Align to iosize boundry, super-align to page boundry.
3459 cnt += off & iomask;
3460 off &= ~(u_quad_t)iomask;
3462 if (off & PAGE_MASK) {
3463 cnt += off & PAGE_MASK;
3464 off &= ~(u_quad_t)PAGE_MASK;
3469 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3470 vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
3478 * If we have a buffer and it is marked B_DELWRI we
3479 * have to lock and write it. Otherwise the prior
3480 * write is assumed to have already been committed.
3482 if ((bp = findblk(vp, loffset)) != NULL && (bp->b_flags & B_DELWRI)) {
3483 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
3484 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL) == 0)
3486 continue; /* retry */
3489 bp->b_flags &= ~B_ASYNC;
3502 aft_ret = VOP_GETATTR(vp, &aft);
3505 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3506 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3508 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3509 if (nfsver.tv_sec == 0)
3511 *tl++ = txdr_unsigned(nfsver.tv_sec);
3512 *tl = txdr_unsigned(nfsver.tv_nsec / 1000);
3523 * nfs statfs service
3526 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3527 struct thread *td, struct mbuf **mrq)
3529 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3530 struct sockaddr *nam = nfsd->nd_nam;
3531 caddr_t dpos = nfsd->nd_dpos;
3532 struct ucred *cred = &nfsd->nd_cr;
3534 struct nfs_statfs *sfp;
3538 int error = 0, rdonly, getret = 1;
3539 int v3 = (nfsd->nd_flag & ND_NFSV3);
3541 struct mbuf *mb, *mb2, *mreq;
3542 struct vnode *vp = NULL;
3543 struct mount *mp = NULL;
3547 struct statfs statfs;
3550 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3551 fhp = &nfh.fh_generic;
3553 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
3554 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3556 nfsm_reply(NFSX_UNSIGNED);
3557 nfsm_srvpostop_attr(getret, &at);
3562 error = VFS_STATFS(vp->v_mount, sf, proc0.p_ucred);
3563 getret = VOP_GETATTR(vp, &at);
3566 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3568 nfsm_srvpostop_attr(getret, &at);
3573 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3575 tval = (u_quad_t)sf->f_blocks;
3576 tval *= (u_quad_t)sf->f_bsize;
3577 txdr_hyper(tval, &sfp->sf_tbytes);
3578 tval = (u_quad_t)sf->f_bfree;
3579 tval *= (u_quad_t)sf->f_bsize;
3580 txdr_hyper(tval, &sfp->sf_fbytes);
3581 tval = (u_quad_t)sf->f_bavail;
3582 tval *= (u_quad_t)sf->f_bsize;
3583 txdr_hyper(tval, &sfp->sf_abytes);
3584 sfp->sf_tfiles.nfsuquad[0] = 0;
3585 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
3586 sfp->sf_ffiles.nfsuquad[0] = 0;
3587 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3588 sfp->sf_afiles.nfsuquad[0] = 0;
3589 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3590 sfp->sf_invarsec = 0;
3592 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3593 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3594 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3595 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3596 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3605 * nfs fsinfo service
3608 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3609 struct thread *td, struct mbuf **mrq)
3611 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3612 struct sockaddr *nam = nfsd->nd_nam;
3613 caddr_t dpos = nfsd->nd_dpos;
3614 struct ucred *cred = &nfsd->nd_cr;
3616 struct nfsv3_fsinfo *sip;
3619 int error = 0, rdonly, getret = 1, pref;
3621 struct mbuf *mb, *mb2, *mreq;
3622 struct vnode *vp = NULL;
3623 struct mount *mp = NULL;
3630 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3631 fhp = &nfh.fh_generic;
3633 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
3634 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3636 nfsm_reply(NFSX_UNSIGNED);
3637 nfsm_srvpostop_attr(getret, &at);
3642 /* XXX Try to make a guess on the max file size. */
3643 VFS_STATFS(vp->v_mount, &sb, proc0.p_ucred);
3644 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3646 getret = VOP_GETATTR(vp, &at);
3649 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3650 nfsm_srvpostop_attr(getret, &at);
3651 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3655 * There should be file system VFS OP(s) to get this information.
3656 * For now, assume ufs.
3658 if (slp->ns_so->so_type == SOCK_DGRAM)
3659 pref = NFS_MAXDGRAMDATA;
3662 sip->fs_rtmax = txdr_unsigned(NFS_MAXDATA);
3663 sip->fs_rtpref = txdr_unsigned(pref);
3664 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3665 sip->fs_wtmax = txdr_unsigned(NFS_MAXDATA);
3666 sip->fs_wtpref = txdr_unsigned(pref);
3667 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3668 sip->fs_dtpref = txdr_unsigned(pref);
3669 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3670 sip->fs_timedelta.nfsv3_sec = 0;
3671 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3672 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3673 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3674 NFSV3FSINFO_CANSETTIME);
3682 * nfs pathconf service
3685 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3686 struct thread *td, struct mbuf **mrq)
3688 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3689 struct sockaddr *nam = nfsd->nd_nam;
3690 caddr_t dpos = nfsd->nd_dpos;
3691 struct ucred *cred = &nfsd->nd_cr;
3693 struct nfsv3_pathconf *pc;
3696 int error = 0, rdonly, getret = 1;
3697 register_t linkmax, namemax, chownres, notrunc;
3699 struct mbuf *mb, *mb2, *mreq;
3700 struct vnode *vp = NULL;
3701 struct mount *mp = NULL;
3706 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3707 fhp = &nfh.fh_generic;
3709 error = nfsrv_fhtovp(fhp, 1, &mp, &vp, cred, slp, nam,
3710 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3712 nfsm_reply(NFSX_UNSIGNED);
3713 nfsm_srvpostop_attr(getret, &at);
3717 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3719 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3721 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3723 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc);
3724 getret = VOP_GETATTR(vp, &at);
3727 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3728 nfsm_srvpostop_attr(getret, &at);
3733 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3735 pc->pc_linkmax = txdr_unsigned(linkmax);
3736 pc->pc_namemax = txdr_unsigned(namemax);
3737 pc->pc_notrunc = txdr_unsigned(notrunc);
3738 pc->pc_chownrestricted = txdr_unsigned(chownres);
3741 * These should probably be supported by VOP_PATHCONF(), but
3742 * until msdosfs is exportable (why would you want to?), the
3743 * Unix defaults should be ok.
3745 pc->pc_caseinsensitive = nfs_false;
3746 pc->pc_casepreserving = nfs_true;
3754 * Null operation, used by clients to ping server
3758 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3759 struct thread *td, struct mbuf **mrq)
3761 struct mbuf *mrep = nfsd->nd_mrep;
3763 int error = NFSERR_RETVOID;
3764 struct mbuf *mb, *mreq;
3766 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3772 * No operation, used for obsolete procedures
3776 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3777 struct thread *td, struct mbuf **mrq)
3779 struct mbuf *mrep = nfsd->nd_mrep;
3782 struct mbuf *mb, *mreq;
3784 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3785 if (nfsd->nd_repstat)
3786 error = nfsd->nd_repstat;
3788 error = EPROCUNAVAIL;
3795 * Perform access checking for vnodes obtained from file handles that would
3796 * refer to files already opened by a Unix client. You cannot just use
3797 * vn_writechk() and VOP_ACCESS() for two reasons.
3798 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3799 * 2 - The owner is to be given access irrespective of mode bits for some
3800 * operations, so that processes that chmod after opening a file don't
3801 * break. I don't like this because it opens a security hole, but since
3802 * the nfs server opens a security hole the size of a barn door anyhow,
3805 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3806 * will return EPERM instead of EACCESS. EPERM is always an error.
3809 nfsrv_access(struct mount *mp, struct vnode *vp, int flags, struct ucred *cred,
3810 int rdonly, struct thread *td, int override)
3815 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3816 if (flags & VWRITE) {
3817 /* Just vn_writechk() changed to check rdonly */
3819 * Disallow write attempts on read-only file systems;
3820 * unless the file is a socket or a block or character
3821 * device resident on the file system.
3824 ((mp->mnt_flag | vp->v_mount->mnt_flag) & MNT_RDONLY)) {
3825 switch (vp->v_type) {
3835 * If there's shared text associated with
3836 * the inode, we can't allow writing.
3838 if (vp->v_flag & VTEXT)
3841 error = VOP_GETATTR(vp, &vattr);
3844 error = VOP_ACCESS(vp, flags, cred);
3846 * Allow certain operations for the owner (reads and writes
3847 * on files that are already open).
3849 if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3853 #endif /* NFS_NOSERVER */