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
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
43 * nfs version 2 and 3 server calls to vnode ops
44 * - these routines generally have 3 phases
45 * 1 - break down and validate rpc request in mbuf list
46 * 2 - do the vnode ops for the request
47 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
48 * 3 - build the rpc reply in an mbuf list
50 * - do not mix the phases, since the nfsm_?? macros can return failures
51 * on a bad rpc or similar and do not do any vrele() or vput()'s
53 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs
54 * error number iff error != 0 whereas
55 * returning an error from the server function implies a fatal error
56 * such as a badly constructed rpc request that should be dropped without
58 * For nfsm_reply(), the case where error == EBADRPC is treated
59 * specially; after constructing a reply, it does an immediate
60 * `goto nfsmout' to avoid getting any V3 post-op status appended.
63 * Warning: always pay careful attention to resource cleanup on return
64 * and note that nfsm_*() macros can terminate a procedure on certain
67 * lookup() and namei()
68 * may return garbage in various structural fields/return elements
69 * if an error is returned, and may garbage up nd.ni_dvp even if no
70 * error is returned and you did not request LOCKPARENT or WANTPARENT.
72 * We use the ni_cnd.cn_flags 'HASBUF' flag to track whether the name
73 * buffer has been freed or not.
76 #include <sys/param.h>
77 #include <sys/systm.h>
79 #include <sys/namei.h>
80 #include <sys/unistd.h>
81 #include <sys/vnode.h>
82 #include <sys/mount.h>
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/malloc.h>
87 #include <sys/dirent.h>
89 #include <sys/kernel.h>
90 #include <sys/sysctl.h>
95 #include <vm/vm_extern.h>
96 #include <vm/vm_object.h>
98 #include <nfs/nfsproto.h>
99 #include <nfs/rpcv2.h>
100 #include <nfsserver/nfs.h>
101 #include <nfs/xdr_subs.h>
102 #include <nfsserver/nfsm_subs.h>
105 #define nfsdbprintf(info) printf info
107 #define nfsdbprintf(info)
110 #define MAX_COMMIT_COUNT (1024 * 1024)
112 #define NUM_HEURISTIC 64
113 #define NHUSE_INIT 64
115 #define NHUSE_MAX 2048
117 static struct nfsheur {
118 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
119 off_t nh_nextr; /* next offset for sequential detection */
120 int nh_use; /* use count for selection */
121 int nh_seqcount; /* heuristic */
122 } nfsheur[NUM_HEURISTIC];
126 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
127 int nfsrvw_procrastinate_v3 = 0;
129 static struct timeval nfsver = { 0 };
131 SYSCTL_NODE(_vfs, OID_AUTO, nfsrv, CTLFLAG_RW, 0, "NFS server");
133 static int nfs_async;
134 static int nfs_commit_blks;
135 static int nfs_commit_miss;
136 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0, "");
137 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, "");
138 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, "");
140 struct nfsrvstats nfsrvstats;
141 SYSCTL_STRUCT(_vfs_nfsrv, NFS_NFSRVSTATS, nfsrvstats, CTLFLAG_RD,
142 &nfsrvstats, nfsrvstats, "S,nfsrvstats");
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 * Clear nameidata fields that are tested in nsfmout cleanup code prior
151 * to using first nfsm macro (that might jump to the cleanup code).
155 ndclear(struct nameidata *nd)
158 nd->ni_cnd.cn_flags = 0;
161 nd->ni_startdir = NULL;
165 * nfs v3 access service
168 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
169 struct thread *td, struct mbuf **mrq)
171 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
172 struct sockaddr *nam = nfsd->nd_nam;
173 caddr_t dpos = nfsd->nd_dpos;
174 struct ucred *cred = &nfsd->nd_cr;
175 struct vnode *vp = NULL;
180 int error = 0, rdonly, getret;
181 struct mbuf *mb, *mreq;
182 struct vattr vattr, *vap = &vattr;
183 u_long testmode, nfsmode;
184 int v3 = (nfsd->nd_flag & ND_NFSV3);
186 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
188 panic("nfsrv3_access: v3 proc called on a v2 connection");
189 fhp = &nfh.fh_generic;
191 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
192 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
194 nfsm_reply(NFSX_UNSIGNED);
195 nfsm_srvpostop_attr(1, (struct vattr *)0);
199 nfsmode = fxdr_unsigned(u_int32_t, *tl);
200 if ((nfsmode & NFSV3ACCESS_READ) &&
201 nfsrv_access(vp, VREAD, cred, rdonly, td, 0))
202 nfsmode &= ~NFSV3ACCESS_READ;
203 if (vp->v_type == VDIR)
204 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
207 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
208 if ((nfsmode & testmode) &&
209 nfsrv_access(vp, VWRITE, cred, rdonly, td, 0))
210 nfsmode &= ~testmode;
211 if (vp->v_type == VDIR)
212 testmode = NFSV3ACCESS_LOOKUP;
214 testmode = NFSV3ACCESS_EXECUTE;
215 if ((nfsmode & testmode) &&
216 nfsrv_access(vp, VEXEC, cred, rdonly, td, 0))
217 nfsmode &= ~testmode;
218 getret = VOP_GETATTR(vp, vap, cred, td);
221 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
222 nfsm_srvpostop_attr(getret, vap);
223 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
224 *tl = txdr_unsigned(nfsmode);
232 * nfs getattr service
235 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
236 struct thread *td, struct mbuf **mrq)
238 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
239 struct sockaddr *nam = nfsd->nd_nam;
240 caddr_t dpos = nfsd->nd_dpos;
241 struct ucred *cred = &nfsd->nd_cr;
242 struct nfs_fattr *fp;
244 struct vattr *vap = &va;
245 struct vnode *vp = NULL;
249 int error = 0, rdonly;
250 struct mbuf *mb, *mreq;
252 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
253 fhp = &nfh.fh_generic;
255 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
261 error = VOP_GETATTR(vp, vap, cred, td);
264 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
269 fp = nfsm_build(struct nfs_fattr *,
270 NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
271 nfsm_srvfillattr(vap, fp);
281 * nfs setattr service
284 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
285 struct thread *td, struct mbuf **mrq)
287 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
288 struct sockaddr *nam = nfsd->nd_nam;
289 caddr_t dpos = nfsd->nd_dpos;
290 struct ucred *cred = &nfsd->nd_cr;
291 struct vattr va, preat;
292 struct vattr *vap = &va;
293 struct nfsv2_sattr *sp;
294 struct nfs_fattr *fp;
295 struct vnode *vp = NULL;
300 int error = 0, rdonly, preat_ret = 1, postat_ret = 1;
301 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
302 struct mbuf *mb, *mreq;
303 struct timespec guard;
304 struct mount *mp = NULL;
306 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
307 fhp = &nfh.fh_generic;
309 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
313 if ((error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp)) != 0) {
317 (void) vn_start_write(vp, &mp, V_WAIT);
323 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
324 gcheck = fxdr_unsigned(int, *tl);
326 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
327 fxdr_nfsv3time(tl, &guard);
330 sp = nfsm_dissect(struct nfsv2_sattr *, NFSX_V2SATTR);
332 * Nah nah nah nah na nah
333 * There is a bug in the Sun client that puts 0xffff in the mode
334 * field of sattr when it should put in 0xffffffff. The u_short
335 * doesn't sign extend.
336 * --> check the low order 2 bytes for 0xffff
338 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
339 vap->va_mode = nfstov_mode(sp->sa_mode);
340 if (sp->sa_uid != nfsrv_nfs_xdrneg1)
341 vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
342 if (sp->sa_gid != nfsrv_nfs_xdrneg1)
343 vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
344 if (sp->sa_size != nfsrv_nfs_xdrneg1)
345 vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
346 if (sp->sa_atime.nfsv2_sec != nfsrv_nfs_xdrneg1) {
348 fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
350 vap->va_atime.tv_sec =
351 fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
352 vap->va_atime.tv_nsec = 0;
355 if (sp->sa_mtime.nfsv2_sec != nfsrv_nfs_xdrneg1)
356 fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
361 * Now that we have all the fields, lets do it.
363 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
365 nfsm_reply(2 * NFSX_UNSIGNED);
367 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
373 * vp now an active resource, pay careful attention to cleanup
376 error = preat_ret = VOP_GETATTR(vp, &preat, cred, td);
377 if (!error && gcheck &&
378 (preat.va_ctime.tv_sec != guard.tv_sec ||
379 preat.va_ctime.tv_nsec != guard.tv_nsec))
380 error = NFSERR_NOT_SYNC;
384 nfsm_reply(NFSX_WCCDATA(v3));
386 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
393 * If the size is being changed write acces is required, otherwise
394 * just check for a read only file system.
396 if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
397 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
402 if (vp->v_type == VDIR) {
405 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
409 error = VOP_SETATTR(vp, vap, cred, td);
410 postat_ret = VOP_GETATTR(vp, vap, cred, td);
417 nfsm_reply(NFSX_WCCORFATTR(v3));
419 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
421 /* v2 non-error case. */
422 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
423 nfsm_srvfillattr(vap, fp);
431 vn_finished_write(mp);
439 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
440 struct thread *td, struct mbuf **mrq)
442 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
443 struct sockaddr *nam = nfsd->nd_nam;
444 caddr_t dpos = nfsd->nd_dpos;
445 struct ucred *cred = &nfsd->nd_cr;
446 struct nfs_fattr *fp;
447 struct nameidata nd, ind, *ndp = &nd;
448 struct vnode *vp, *dirp = NULL;
452 int error = 0, len, dirattr_ret = 1;
453 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
454 struct mbuf *mb, *mreq;
455 struct vattr va, dirattr, *vap = &va;
457 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
460 fhp = &nfh.fh_generic;
462 nfsm_srvnamesiz(len);
464 pubflag = nfs_ispublicfh(fhp);
466 nd.ni_cnd.cn_cred = cred;
467 nd.ni_cnd.cn_nameiop = LOOKUP;
468 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART;
469 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
473 * namei failure, only dirp to cleanup. Clear out garbarge from
474 * structure in case macros jump to nfsmout.
480 dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred,
485 nfsm_reply(NFSX_POSTOPATTR(v3));
487 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
493 * Locate index file for public filehandle
495 * error is 0 on entry and 0 on exit from this block.
499 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
501 * Setup call to lookup() to see if we can find
502 * the index file. Arguably, this doesn't belong
503 * in a kernel.. Ugh. If an error occurs, do not
504 * try to install an index file and then clear the
507 * When we replace nd with ind and redirect ndp,
508 * maintenance of ni_startdir and ni_vp shift to
509 * ind and we have to clean them up in the old nd.
510 * However, the cnd resource continues to be maintained
511 * via the original nd. Confused? You aren't alone!
514 VOP_UNLOCK(nd.ni_vp, 0, td);
515 ind.ni_pathlen = strlen(nfs_pub.np_index);
516 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
518 ind.ni_startdir = nd.ni_vp;
519 VREF(ind.ni_startdir);
521 error = lookup(&ind);
526 * Found an index file. Get rid of
527 * the old references. transfer nd.ni_vp'
533 vrele(nd.ni_startdir);
534 nd.ni_startdir = NULL;
540 * If the public filehandle was used, check that this lookup
541 * didn't result in a filehandle outside the publicly exported
542 * filesystem. We clear the poor vp here to avoid lockups due
546 if (ndp->ni_vp->v_mount != nfs_pub.np_mount) {
555 dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred,
562 * Resources at this point:
563 * ndp->ni_vp may not be NULL
568 nfsm_reply(NFSX_POSTOPATTR(v3));
570 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
576 * Clear out some resources prior to potentially blocking. This
577 * is not as critical as ni_dvp resources in other routines, but
580 vrele(ndp->ni_startdir);
581 ndp->ni_startdir = NULL;
582 NDFREE(&nd, NDF_ONLY_PNBUF);
585 * Get underlying attribute, then release remaining resources ( for
586 * the same potential blocking reason ) and reply.
589 bzero((caddr_t)fhp, sizeof(nfh));
590 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
591 error = VFS_VPTOFH(vp, &fhp->fh_fid);
593 error = VOP_GETATTR(vp, vap, cred, td);
597 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
600 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
604 nfsm_srvfhtom(fhp, v3);
606 nfsm_srvpostop_attr(0, vap);
607 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
609 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
610 nfsm_srvfillattr(vap, fp);
616 NDFREE(&nd, NDF_ONLY_PNBUF);
617 if (ndp->ni_startdir)
618 vrele(ndp->ni_startdir);
625 * nfs readlink service
628 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
629 struct thread *td, struct mbuf **mrq)
631 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
632 struct sockaddr *nam = nfsd->nd_nam;
633 caddr_t dpos = nfsd->nd_dpos;
634 struct ucred *cred = &nfsd->nd_cr;
635 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
636 struct iovec *ivp = iv;
640 int error = 0, rdonly, i, tlen, len, getret;
641 int v3 = (nfsd->nd_flag & ND_NFSV3);
642 struct mbuf *mb, *mp3, *nmp, *mreq;
643 struct vnode *vp = NULL;
647 struct uio io, *uiop = &io;
649 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
651 mp = (struct mbuf *)0;
654 fhp = &nfh.fh_generic;
658 while (len < NFS_MAXPATHLEN) {
659 MGET(nmp, M_TRYWAIT, MT_DATA);
660 MCLGET(nmp, M_TRYWAIT);
661 nmp->m_len = NFSMSIZ(nmp);
668 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
669 mp->m_len = NFS_MAXPATHLEN - len;
670 len = NFS_MAXPATHLEN;
673 ivp->iov_base = mtod(mp, caddr_t);
674 ivp->iov_len = mp->m_len;
679 uiop->uio_iovcnt = i;
680 uiop->uio_offset = 0;
681 uiop->uio_resid = len;
682 uiop->uio_rw = UIO_READ;
683 uiop->uio_segflg = UIO_SYSSPACE;
684 uiop->uio_td = (struct thread *)0;
685 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
687 nfsm_reply(2 * NFSX_UNSIGNED);
689 nfsm_srvpostop_attr(1, (struct vattr *)0);
693 if (vp->v_type != VLNK) {
700 error = VOP_READLINK(vp, uiop, cred);
702 getret = VOP_GETATTR(vp, &attr, cred, td);
705 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
707 nfsm_srvpostop_attr(getret, &attr);
712 if (uiop->uio_resid > 0) {
713 len -= uiop->uio_resid;
714 tlen = nfsm_rndup(len);
715 nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
717 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
718 *tl = txdr_unsigned(len);
733 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
734 struct thread *td, struct mbuf **mrq)
736 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
737 struct sockaddr *nam = nfsd->nd_nam;
738 caddr_t dpos = nfsd->nd_dpos;
739 struct ucred *cred = &nfsd->nd_cr;
743 struct nfs_fattr *fp;
747 int error = 0, rdonly, cnt, len, left, siz, tlen, getret;
748 int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
749 struct mbuf *mb, *mreq;
751 struct vnode *vp = NULL;
754 struct uio io, *uiop = &io;
755 struct vattr va, *vap = &va;
760 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
761 fhp = &nfh.fh_generic;
764 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
765 off = fxdr_hyper(tl);
767 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
768 off = (off_t)fxdr_unsigned(u_int32_t, *tl);
770 nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
773 * Reference vp. If an error occurs, vp will be invalid, but we
774 * have to NULL it just in case. The macros might goto nfsmout
778 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
781 nfsm_reply(2 * NFSX_UNSIGNED);
783 nfsm_srvpostop_attr(1, (struct vattr *)0);
788 if (vp->v_type != VREG) {
792 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
795 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, td, 1)) != 0)
796 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 1);
798 getret = VOP_GETATTR(vp, vap, cred, td);
804 nfsm_reply(NFSX_POSTOPATTR(v3));
806 nfsm_srvpostop_attr(getret, vap);
812 * Calculate byte count to read
815 if (off >= vap->va_size)
817 else if ((off + reqlen) > vap->va_size)
818 cnt = vap->va_size - off;
823 * Calculate seqcount for heuristic
831 * Locate best candidate
834 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) & (NUM_HEURISTIC - 1);
838 if (nfsheur[hi].nh_vp == vp) {
842 if (nfsheur[hi].nh_use > 0)
843 --nfsheur[hi].nh_use;
844 hi = (hi + 1) & (NUM_HEURISTIC - 1);
845 if (nfsheur[hi].nh_use < nh->nh_use)
849 if (nh->nh_vp != vp) {
852 nh->nh_use = NHUSE_INIT;
860 * Calculate heuristic
863 if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
864 if (++nh->nh_seqcount > 127)
865 nh->nh_seqcount = 127;
866 } else if (nh->nh_seqcount > 1) {
871 nh->nh_use += NHUSE_INC;
872 if (nh->nh_use > NHUSE_MAX)
873 nh->nh_use = NHUSE_MAX;
874 ioflag |= nh->nh_seqcount << 16;
877 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
879 tl = nfsm_build(u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
880 *tl++ = nfsrv_nfs_true;
881 fp = (struct nfs_fattr *)tl;
882 tl += (NFSX_V3FATTR / sizeof (u_int32_t));
884 tl = nfsm_build(u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
885 fp = (struct nfs_fattr *)tl;
886 tl += (NFSX_V2FATTR / sizeof (u_int32_t));
888 len = left = nfsm_rndup(cnt);
891 * Generate the mbuf list with the uio_iov ref. to it.
896 siz = min(M_TRAILINGSPACE(m), left);
902 MGET(m, M_TRYWAIT, MT_DATA);
903 MCLGET(m, M_TRYWAIT);
909 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
911 uiop->uio_iov = iv2 = iv;
917 panic("nfsrv_read iov");
918 siz = min(M_TRAILINGSPACE(m), left);
920 iv->iov_base = mtod(m, caddr_t) + m->m_len;
929 uiop->uio_iovcnt = i;
930 uiop->uio_offset = off;
931 uiop->uio_resid = len;
932 uiop->uio_rw = UIO_READ;
933 uiop->uio_segflg = UIO_SYSSPACE;
934 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
935 off = uiop->uio_offset;
937 FREE((caddr_t)iv2, M_TEMP);
938 if (error || (getret = VOP_GETATTR(vp, vap, cred, td))) {
944 nfsm_reply(NFSX_POSTOPATTR(v3));
946 nfsm_srvpostop_attr(getret, vap);
955 nfsm_srvfillattr(vap, fp);
956 tlen = len - uiop->uio_resid;
957 cnt = cnt < tlen ? cnt : tlen;
958 tlen = nfsm_rndup(cnt);
959 if (len != tlen || tlen != cnt)
960 nfsm_adj(mb, len - tlen, tlen - cnt);
962 *tl++ = txdr_unsigned(cnt);
964 *tl++ = nfsrv_nfs_true;
966 *tl++ = nfsrv_nfs_false;
968 *tl = txdr_unsigned(cnt);
979 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
980 struct thread *td, struct mbuf **mrq)
982 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
983 struct sockaddr *nam = nfsd->nd_nam;
984 caddr_t dpos = nfsd->nd_dpos;
985 struct ucred *cred = &nfsd->nd_cr;
989 struct nfs_fattr *fp;
991 struct vattr va, forat;
992 struct vattr *vap = &va;
995 int error = 0, rdonly, len, forat_ret = 1;
996 int ioflags, aftat_ret = 1, retlen = 0, zeroing, adjust;
997 int stable = NFSV3WRITE_FILESYNC;
998 int v3 = (nfsd->nd_flag & ND_NFSV3);
999 struct mbuf *mb, *mreq;
1000 struct vnode *vp = NULL;
1003 struct uio io, *uiop = &io;
1005 struct mount *mntp = NULL;
1007 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1013 fhp = &nfh.fh_generic;
1015 if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1019 if ((error = VFS_FHTOVP(mntp, &fhp->fh_fid, &vp)) != 0) {
1023 (void) vn_start_write(vp, &mntp, V_WAIT);
1027 tl = nfsm_dissect(u_int32_t *, 5 * NFSX_UNSIGNED);
1028 off = fxdr_hyper(tl);
1030 stable = fxdr_unsigned(int, *tl++);
1032 tl = nfsm_dissect(u_int32_t *, 4 * NFSX_UNSIGNED);
1033 off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1036 stable = NFSV3WRITE_UNSTABLE;
1038 retlen = len = fxdr_unsigned(int32_t, *tl);
1042 * For NFS Version 2, it is not obvious what a write of zero length
1043 * should do, but I might as well be consistent with Version 3,
1044 * which is to return ok so long as there are no permission problems.
1052 adjust = dpos - mtod(mp, caddr_t);
1053 mp->m_len -= adjust;
1054 if (mp->m_len > 0 && adjust > 0)
1055 mp->m_data += adjust;
1059 else if (mp->m_len > 0) {
1062 mp->m_len -= (i - len);
1071 if (len > NFS_MAXDATA || len < 0 || i < len) {
1073 nfsm_reply(2 * NFSX_UNSIGNED);
1075 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1079 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
1082 nfsm_reply(2 * NFSX_UNSIGNED);
1084 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1089 forat_ret = VOP_GETATTR(vp, &forat, cred, td);
1090 if (vp->v_type != VREG) {
1094 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1097 error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
1101 nfsm_reply(NFSX_WCCDATA(v3));
1103 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1109 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
1111 uiop->uio_iov = iv = ivp;
1112 uiop->uio_iovcnt = cnt;
1115 if (mp->m_len > 0) {
1116 ivp->iov_base = mtod(mp, caddr_t);
1117 ivp->iov_len = mp->m_len;
1125 * The IO_METASYNC flag indicates that all metadata (and not just
1126 * enough to ensure data integrity) mus be written to stable storage
1128 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1130 if (stable == NFSV3WRITE_UNSTABLE)
1131 ioflags = IO_NODELOCKED;
1132 else if (stable == NFSV3WRITE_DATASYNC)
1133 ioflags = (IO_SYNC | IO_NODELOCKED);
1135 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1136 uiop->uio_resid = len;
1137 uiop->uio_rw = UIO_WRITE;
1138 uiop->uio_segflg = UIO_SYSSPACE;
1139 uiop->uio_td = (struct thread *)0;
1140 uiop->uio_offset = off;
1141 error = VOP_WRITE(vp, uiop, ioflags, cred);
1142 nfsrvstats.srvvop_writes++;
1143 FREE((caddr_t)iv, M_TEMP);
1145 aftat_ret = VOP_GETATTR(vp, vap, cred, td);
1151 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1152 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1154 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1159 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1160 *tl++ = txdr_unsigned(retlen);
1162 * If nfs_async is set, then pretend the write was FILESYNC.
1164 if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1165 *tl++ = txdr_unsigned(stable);
1167 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1169 * Actually, there is no need to txdr these fields,
1170 * but it may make the values more human readable,
1171 * for debugging purposes.
1173 if (nfsver.tv_sec == 0)
1175 *tl++ = txdr_unsigned(nfsver.tv_sec);
1176 *tl = txdr_unsigned(nfsver.tv_usec);
1177 } else if (!error) {
1178 /* v2 non-error case. */
1179 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1180 nfsm_srvfillattr(vap, fp);
1186 vn_finished_write(mntp);
1191 * NFS write service with write gathering support. Called when
1192 * nfsrvw_procrastinate > 0.
1193 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1194 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1198 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp,
1199 struct thread *td, struct mbuf **mrq)
1203 struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1204 struct nfs_fattr *fp;
1207 struct nfsrvw_delayhash *wpp;
1209 struct vattr va, forat;
1212 int error = 0, rdonly, len, forat_ret = 1;
1213 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
1214 struct mbuf *mb, *mreq, *mrep, *md;
1215 struct vnode *vp = NULL;
1216 struct uio io, *uiop = &io;
1218 struct mount *mntp = NULL;
1220 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1229 mrep = nfsd->nd_mrep;
1231 dpos = nfsd->nd_dpos;
1232 cred = &nfsd->nd_cr;
1233 v3 = (nfsd->nd_flag & ND_NFSV3);
1234 LIST_INIT(&nfsd->nd_coalesce);
1235 nfsd->nd_mreq = NULL;
1236 nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1237 cur_usec = nfs_curusec();
1238 nfsd->nd_time = cur_usec +
1239 (v3 ? nfsrvw_procrastinate_v3 : nfsrvw_procrastinate);
1242 * Now, get the write header..
1244 nfsm_srvmtofh(&nfsd->nd_fh);
1246 tl = nfsm_dissect(u_int32_t *, 5 * NFSX_UNSIGNED);
1247 nfsd->nd_off = fxdr_hyper(tl);
1249 nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1251 tl = nfsm_dissect(u_int32_t *, 4 * NFSX_UNSIGNED);
1252 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1255 nfsd->nd_stable = NFSV3WRITE_UNSTABLE;
1257 len = fxdr_unsigned(int32_t, *tl);
1259 nfsd->nd_eoff = nfsd->nd_off + len;
1262 * Trim the header out of the mbuf list and trim off any trailing
1263 * junk so that the mbuf list has only the write data.
1271 adjust = dpos - mtod(mp, caddr_t);
1272 mp->m_len -= adjust;
1273 if (mp->m_len > 0 && adjust > 0)
1274 mp->m_data += adjust;
1281 mp->m_len -= (i - len);
1287 if (len > NFS_MAXDATA || len < 0 || i < len) {
1291 nfsm_writereply(2 * NFSX_UNSIGNED);
1293 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1294 nfsd->nd_mreq = mreq;
1295 nfsd->nd_mrep = NULL;
1300 * Add this entry to the hash and time queues.
1304 wp = LIST_FIRST(&slp->ns_tq);
1305 while (wp && wp->nd_time < nfsd->nd_time) {
1307 wp = LIST_NEXT(wp, nd_tq);
1309 NFS_DPF(WG, ("Q%03x", nfsd->nd_retxid & 0xfff));
1311 LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1313 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1315 if (nfsd->nd_mrep) {
1316 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1318 wp = LIST_FIRST(wpp);
1320 bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh, NFSX_V3FH)){
1322 wp = LIST_NEXT(wp, nd_hash);
1324 while (wp && wp->nd_off < nfsd->nd_off &&
1325 !bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1327 wp = LIST_NEXT(wp, nd_hash);
1330 LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1333 * Search the hash list for overlapping entries and
1336 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1337 wp = LIST_NEXT(nfsd, nd_hash);
1338 if (NFSW_SAMECRED(owp, nfsd))
1339 nfsrvw_coalesce(owp, nfsd);
1342 LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1349 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1350 * and generate the associated reply mbuf list(s).
1353 cur_usec = nfs_curusec();
1355 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1356 owp = LIST_NEXT(nfsd, nd_tq);
1357 if (nfsd->nd_time > cur_usec)
1361 NFS_DPF(WG, ("P%03x", nfsd->nd_retxid & 0xfff));
1362 LIST_REMOVE(nfsd, nd_tq);
1363 LIST_REMOVE(nfsd, nd_hash);
1365 mrep = nfsd->nd_mrep;
1366 nfsd->nd_mrep = NULL;
1367 cred = &nfsd->nd_cr;
1368 v3 = (nfsd->nd_flag & ND_NFSV3);
1369 forat_ret = aftat_ret = 1;
1370 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1371 nfsd->nd_nam, &rdonly, TRUE);
1374 forat_ret = VOP_GETATTR(vp, &forat, cred, td);
1375 if (vp->v_type != VREG) {
1379 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1385 error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
1386 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1387 ioflags = IO_NODELOCKED;
1388 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1389 ioflags = (IO_SYNC | IO_NODELOCKED);
1391 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1392 uiop->uio_rw = UIO_WRITE;
1393 uiop->uio_segflg = UIO_SYSSPACE;
1394 uiop->uio_td = (struct thread *)0;
1395 uiop->uio_offset = nfsd->nd_off;
1396 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1397 if (uiop->uio_resid > 0) {
1405 uiop->uio_iovcnt = i;
1406 MALLOC(iov, struct iovec *, i * sizeof (struct iovec),
1408 uiop->uio_iov = ivp = iov;
1411 if (mp->m_len > 0) {
1412 ivp->iov_base = mtod(mp, caddr_t);
1413 ivp->iov_len = mp->m_len;
1419 if (vn_start_write(vp, &mntp, V_NOWAIT) != 0) {
1420 VOP_UNLOCK(vp, 0, td);
1421 error = vn_start_write(NULL, &mntp, V_WAIT);
1422 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1426 error = VOP_WRITE(vp, uiop, ioflags, cred);
1427 nfsrvstats.srvvop_writes++;
1428 vn_finished_write(mntp);
1430 FREE((caddr_t)iov, M_TEMP);
1434 aftat_ret = VOP_GETATTR(vp, &va, cred, td);
1440 * Loop around generating replies for all write rpcs that have
1441 * now been completed.
1445 NFS_DPF(WG, ("R%03x", nfsd->nd_retxid & 0xfff));
1447 nfsm_writereply(NFSX_WCCDATA(v3));
1449 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1452 nfsm_writereply(NFSX_PREOPATTR(v3) +
1453 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1454 NFSX_WRITEVERF(v3));
1456 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1457 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1458 *tl++ = txdr_unsigned(nfsd->nd_len);
1459 *tl++ = txdr_unsigned(swp->nd_stable);
1461 * Actually, there is no need to txdr these fields,
1462 * but it may make the values more human readable,
1463 * for debugging purposes.
1465 if (nfsver.tv_sec == 0)
1467 *tl++ = txdr_unsigned(nfsver.tv_sec);
1468 *tl = txdr_unsigned(nfsver.tv_usec);
1470 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1471 nfsm_srvfillattr(&va, fp);
1474 nfsd->nd_mreq = mreq;
1476 panic("nfsrv_write: nd_mrep not free");
1479 * Done. Put it at the head of the timer queue so that
1480 * the final phase can return the reply.
1485 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1487 nfsd = LIST_FIRST(&swp->nd_coalesce);
1489 LIST_REMOVE(nfsd, nd_tq);
1495 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1502 * Search for a reply to return.
1505 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq)
1506 if (nfsd->nd_mreq) {
1507 NFS_DPF(WG, ("X%03x", nfsd->nd_retxid & 0xfff));
1508 LIST_REMOVE(nfsd, nd_tq);
1509 *mrq = nfsd->nd_mreq;
1518 * Coalesce the write request nfsd into owp. To do this we must:
1519 * - remove nfsd from the queues
1520 * - merge nfsd->nd_mrep into owp->nd_mrep
1521 * - update the nd_eoff and nd_stable for owp
1522 * - put nfsd on owp's nd_coalesce list
1523 * NB: Must be called at splsoftclock().
1526 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1530 struct nfsrv_descript *p;
1532 NFS_DPF(WG, ("C%03x-%03x",
1533 nfsd->nd_retxid & 0xfff, owp->nd_retxid & 0xfff));
1534 LIST_REMOVE(nfsd, nd_hash);
1535 LIST_REMOVE(nfsd, nd_tq);
1536 if (owp->nd_eoff < nfsd->nd_eoff) {
1537 overlap = owp->nd_eoff - nfsd->nd_off;
1539 panic("nfsrv_coalesce: bad off");
1541 m_adj(nfsd->nd_mrep, overlap);
1545 mp->m_next = nfsd->nd_mrep;
1546 owp->nd_eoff = nfsd->nd_eoff;
1548 m_freem(nfsd->nd_mrep);
1549 nfsd->nd_mrep = NULL;
1550 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1551 owp->nd_stable = NFSV3WRITE_FILESYNC;
1552 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1553 owp->nd_stable == NFSV3WRITE_UNSTABLE)
1554 owp->nd_stable = NFSV3WRITE_DATASYNC;
1555 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1558 * If nfsd had anything else coalesced into it, transfer them
1559 * to owp, otherwise their replies will never get sent.
1561 for (p = LIST_FIRST(&nfsd->nd_coalesce); p;
1562 p = LIST_FIRST(&nfsd->nd_coalesce)) {
1563 LIST_REMOVE(p, nd_tq);
1564 LIST_INSERT_HEAD(&owp->nd_coalesce, p, nd_tq);
1569 * nfs create service
1570 * now does a truncate to 0 length via. setattr if it already exists
1573 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1574 struct thread *td, struct mbuf **mrq)
1576 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1577 struct sockaddr *nam = nfsd->nd_nam;
1578 caddr_t dpos = nfsd->nd_dpos;
1579 struct ucred *cred = &nfsd->nd_cr;
1580 struct nfs_fattr *fp;
1581 struct vattr va, dirfor, diraft;
1582 struct vattr *vap = &va;
1583 struct nfsv2_sattr *sp;
1585 struct nameidata nd;
1587 int error = 0, rdev, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1588 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1590 struct mbuf *mb, *mreq;
1591 struct vnode *dirp = (struct vnode *)0;
1595 u_char cverf[NFSX_V3CREATEVERF];
1596 struct mount *mp = NULL;
1599 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1605 fhp = &nfh.fh_generic;
1607 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1611 if ((error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp)) != 0) {
1615 (void) vn_start_write(vp, &mp, V_WAIT);
1617 nfsm_srvnamesiz(len);
1619 nd.ni_cnd.cn_cred = cred;
1620 nd.ni_cnd.cn_nameiop = CREATE;
1621 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART;
1624 * Call namei and do initial cleanup to get a few things
1625 * out of the way. If we get an initial error we cleanup
1626 * and return here to avoid special-casing the invalid nd
1627 * structure through the rest of the case. dirp may be
1628 * set even if an error occurs, but the nd structure will not
1629 * be valid at all if an error occurs so we have to invalidate it
1630 * prior to calling nfsm_reply ( which might goto nfsmout ).
1632 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1636 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1644 nfsm_reply(NFSX_WCCDATA(v3));
1646 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1652 * No error. Continue. State:
1654 * startdir is valid ( we release this immediately )
1656 * nd.ni_vp may be valid
1657 * nd.ni_dvp is valid
1659 * The error state is set through the code and we may also do some
1660 * opportunistic releasing of vnodes to avoid holding locks through
1661 * NFS I/O. The cleanup at the end is a catch-all
1666 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1667 how = fxdr_unsigned(int, *tl);
1669 case NFSV3CREATE_GUARDED:
1675 case NFSV3CREATE_UNCHECKED:
1678 case NFSV3CREATE_EXCLUSIVE:
1679 cp = nfsm_dissect(caddr_t, NFSX_V3CREATEVERF);
1680 bcopy(cp, cverf, NFSX_V3CREATEVERF);
1684 vap->va_type = VREG;
1686 sp = nfsm_dissect(struct nfsv2_sattr *, NFSX_V2SATTR);
1687 vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1688 if (vap->va_type == VNON)
1689 vap->va_type = VREG;
1690 vap->va_mode = nfstov_mode(sp->sa_mode);
1691 switch (vap->va_type) {
1693 tsize = fxdr_unsigned(int32_t, sp->sa_size);
1695 vap->va_size = (u_quad_t)tsize;
1700 rdev = fxdr_unsigned(long, sp->sa_size);
1708 * Iff doesn't exist, create it
1709 * otherwise just truncate to 0 length
1710 * should I set the mode too ?
1712 * The only possible error we can have at this point is EEXIST.
1713 * nd.ni_vp will also be non-NULL in that case.
1715 if (nd.ni_vp == NULL) {
1716 if (vap->va_mode == (mode_t)VNOVAL)
1718 if (vap->va_type == VREG || vap->va_type == VSOCK) {
1719 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1721 NDFREE(&nd, NDF_ONLY_PNBUF);
1723 nfsrv_object_create(nd.ni_vp);
1724 if (exclusive_flag) {
1727 bcopy(cverf, (caddr_t)&vap->va_atime,
1729 error = VOP_SETATTR(nd.ni_vp, vap, cred,
1733 } else if (vap->va_type == VCHR || vap->va_type == VBLK ||
1734 vap->va_type == VFIFO) {
1736 * NFSv2-specific code for creating device nodes
1739 * Handle SysV FIFO node special cases. All other
1740 * devices require super user to access.
1742 if (vap->va_type == VCHR && rdev == 0xffffffff)
1743 vap->va_type = VFIFO;
1744 if (vap->va_type != VFIFO &&
1745 (error = suser_xxx(cred, 0, 0))) {
1748 vap->va_rdev = rdev;
1749 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1751 NDFREE(&nd, NDF_ONLY_PNBUF);
1758 * release dvp prior to lookup
1766 * Even though LOCKPARENT was cleared, ni_dvp may
1769 nd.ni_cnd.cn_nameiop = LOOKUP;
1770 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
1771 nd.ni_cnd.cn_thread = td;
1772 nd.ni_cnd.cn_cred = cred;
1774 error = lookup(&nd);
1779 nfsrv_object_create(nd.ni_vp);
1780 if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1788 if (vap->va_size != -1) {
1789 error = nfsrv_access(nd.ni_vp, VWRITE, cred,
1790 (nd.ni_cnd.cn_flags & RDONLY), td, 0);
1792 tempsize = vap->va_size;
1794 vap->va_size = tempsize;
1795 error = VOP_SETATTR(nd.ni_vp, vap, cred,
1802 bzero((caddr_t)fhp, sizeof(nfh));
1803 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
1804 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
1806 error = VOP_GETATTR(nd.ni_vp, vap, cred, td);
1809 if (exclusive_flag && !error &&
1810 bcmp(cverf, (caddr_t)&vap->va_atime, NFSX_V3CREATEVERF))
1812 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
1817 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1820 nfsm_srvpostop_fh(fhp);
1821 nfsm_srvpostop_attr(0, vap);
1823 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1824 } else if (!error) {
1825 /* v2 non-error case. */
1826 nfsm_srvfhtom(fhp, v3);
1827 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1828 nfsm_srvfillattr(vap, fp);
1833 if (nd.ni_startdir) {
1834 vrele(nd.ni_startdir);
1835 nd.ni_startdir = NULL;
1839 NDFREE(&nd, NDF_ONLY_PNBUF);
1841 if (nd.ni_dvp == nd.ni_vp)
1848 vn_finished_write(mp);
1853 * nfs v3 mknod service
1856 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1857 struct thread *td, struct mbuf **mrq)
1859 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1860 struct sockaddr *nam = nfsd->nd_nam;
1861 caddr_t dpos = nfsd->nd_dpos;
1862 struct ucred *cred = &nfsd->nd_cr;
1863 struct vattr va, dirfor, diraft;
1864 struct vattr *vap = &va;
1866 struct nameidata nd;
1868 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1869 u_int32_t major, minor;
1871 struct mbuf *mb, *mreq;
1872 struct vnode *vp, *dirp = (struct vnode *)0;
1875 struct mount *mp = NULL;
1876 int v3 = (nfsd->nd_flag & ND_NFSV3);
1878 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1880 panic("nfsrv_mknod: v3 proc called on a v2 connection");
1883 fhp = &nfh.fh_generic;
1885 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1889 if ((error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp)) != 0) {
1893 (void) vn_start_write(vp, &mp, V_WAIT);
1896 nfsm_srvnamesiz(len);
1898 nd.ni_cnd.cn_cred = cred;
1899 nd.ni_cnd.cn_nameiop = CREATE;
1900 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART;
1903 * Handle nfs_namei() call. If an error occurs, the nd structure
1904 * is not valid. However, nfsm_*() routines may still jump to
1908 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1911 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, td);
1913 nfsm_reply(NFSX_WCCDATA(1));
1914 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1918 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1919 vtyp = nfsv3tov_type(*tl);
1920 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1921 error = NFSERR_BADTYPE;
1926 if (vtyp == VCHR || vtyp == VBLK) {
1927 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1928 major = fxdr_unsigned(u_int32_t, *tl++);
1929 minor = fxdr_unsigned(u_int32_t, *tl);
1930 vap->va_rdev = makeudev(major, minor);
1934 * Iff doesn't exist, create it.
1940 vap->va_type = vtyp;
1941 if (vap->va_mode == (mode_t)VNOVAL)
1943 if (vtyp == VSOCK) {
1944 vrele(nd.ni_startdir);
1945 nd.ni_startdir = NULL;
1946 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1948 NDFREE(&nd, NDF_ONLY_PNBUF);
1950 if (vtyp != VFIFO && (error = suser_xxx(cred, 0, 0)))
1952 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1954 NDFREE(&nd, NDF_ONLY_PNBUF);
1961 * Release dvp prior to lookup
1966 nd.ni_cnd.cn_nameiop = LOOKUP;
1967 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
1968 nd.ni_cnd.cn_thread = td;
1969 nd.ni_cnd.cn_cred = td->td_proc->p_ucred;
1971 error = lookup(&nd);
1976 if (nd.ni_cnd.cn_flags & ISSYMLINK)
1981 * send response, cleanup, return.
1984 if (nd.ni_startdir) {
1985 vrele(nd.ni_startdir);
1986 nd.ni_startdir = NULL;
1988 NDFREE(&nd, NDF_ONLY_PNBUF);
1990 if (nd.ni_dvp == nd.ni_vp)
1998 bzero((caddr_t)fhp, sizeof(nfh));
1999 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2000 error = VFS_VPTOFH(vp, &fhp->fh_fid);
2002 error = VOP_GETATTR(vp, vap, cred, td);
2007 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2013 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
2016 nfsm_srvpostop_fh(fhp);
2017 nfsm_srvpostop_attr(0, vap);
2019 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2021 vn_finished_write(mp);
2027 vrele(nd.ni_startdir);
2028 NDFREE(&nd, NDF_ONLY_PNBUF);
2030 if (nd.ni_dvp == nd.ni_vp)
2037 vn_finished_write(mp);
2042 * nfs remove service
2045 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2046 struct thread *td, struct mbuf **mrq)
2048 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2049 struct sockaddr *nam = nfsd->nd_nam;
2050 caddr_t dpos = nfsd->nd_dpos;
2051 struct ucred *cred = &nfsd->nd_cr;
2052 struct nameidata nd;
2054 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2055 int v3 = (nfsd->nd_flag & ND_NFSV3);
2056 struct mbuf *mb, *mreq;
2058 struct vattr dirfor, diraft;
2061 struct mount *mp = NULL;
2064 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2067 fhp = &nfh.fh_generic;
2069 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2073 if ((error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp)) != 0) {
2077 (void) vn_start_write(vp, &mp, V_WAIT);
2080 nfsm_srvnamesiz(len);
2082 nd.ni_cnd.cn_cred = cred;
2083 nd.ni_cnd.cn_nameiop = DELETE;
2084 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2085 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2089 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2097 if (nd.ni_vp->v_type == VDIR) {
2098 error = EPERM; /* POSIX */
2102 * The root of a mounted filesystem cannot be deleted.
2104 if (nd.ni_vp->v_flag & VROOT) {
2110 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2111 NDFREE(&nd, NDF_ONLY_PNBUF);
2115 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2120 nfsm_reply(NFSX_WCCDATA(v3));
2122 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2126 NDFREE(&nd, NDF_ONLY_PNBUF);
2128 if (nd.ni_dvp == nd.ni_vp)
2135 vn_finished_write(mp);
2140 * nfs rename service
2143 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2144 struct thread *td, struct mbuf **mrq)
2146 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2147 struct sockaddr *nam = nfsd->nd_nam;
2148 caddr_t dpos = nfsd->nd_dpos;
2149 struct ucred *cred = &nfsd->nd_cr;
2151 int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
2152 int tdirfor_ret = 1, tdiraft_ret = 1;
2153 int v3 = (nfsd->nd_flag & ND_NFSV3);
2154 struct mbuf *mb, *mreq;
2155 struct nameidata fromnd, tond;
2156 struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0;
2157 struct vnode *tdirp = (struct vnode *)0;
2158 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
2160 fhandle_t *ffhp, *tfhp;
2162 struct mount *mp = NULL;
2165 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2167 fvp = (struct vnode *)0;
2169 ffhp = &fnfh.fh_generic;
2170 tfhp = &tnfh.fh_generic;
2173 * Clear fields incase goto nfsmout occurs from macro.
2179 nfsm_srvmtofh(ffhp);
2180 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) {
2184 if ((error = VFS_FHTOVP(mp, &ffhp->fh_fid, &vp)) != 0) {
2188 (void) vn_start_write(vp, &mp, V_WAIT);
2191 nfsm_srvnamesiz(len);
2193 * Remember our original uid so that we can reset cr_uid before
2194 * the second nfs_namei() call, in case it is remapped.
2196 saved_uid = cred->cr_uid;
2197 fromnd.ni_cnd.cn_cred = cred;
2198 fromnd.ni_cnd.cn_nameiop = DELETE;
2199 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART;
2200 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
2201 &dpos, &fdirp, td, FALSE);
2204 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred,
2212 nfsm_reply(2 * NFSX_WCCDATA(v3));
2214 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2215 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2221 nfsm_srvmtofh(tfhp);
2222 nfsm_strsiz(len2, NFS_MAXNAMLEN);
2223 cred->cr_uid = saved_uid;
2224 tond.ni_cnd.cn_cred = cred;
2225 tond.ni_cnd.cn_nameiop = RENAME;
2226 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART;
2227 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
2228 &dpos, &tdirp, td, FALSE);
2231 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred,
2244 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2250 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2257 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
2265 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
2272 if (fvp->v_mount != tdvp->v_mount) {
2286 * If source is the same as the destination (that is the
2287 * same vnode with the same name in the same directory),
2288 * then there is nothing to do.
2290 if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2291 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2292 !bcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2293 fromnd.ni_cnd.cn_namelen))
2298 * The VOP_RENAME function releases all vnode references &
2299 * locks prior to returning so we need to clear the pointers
2300 * to bypass cleanup code later on.
2302 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2303 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2304 fromnd.ni_dvp = NULL;
2305 fromnd.ni_vp = NULL;
2309 fromnd.ni_cnd.cn_flags &= ~HASBUF;
2310 tond.ni_cnd.cn_flags &= ~HASBUF;
2320 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, td);
2322 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, td);
2323 nfsm_reply(2 * NFSX_WCCDATA(v3));
2325 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2326 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2333 * Clear out tond related fields
2337 if (tond.ni_startdir)
2338 vrele(tond.ni_startdir);
2339 NDFREE(&tond, NDF_ONLY_PNBUF);
2341 if (tond.ni_dvp == tond.ni_vp)
2350 * Clear out fromnd related fields
2354 if (fromnd.ni_startdir)
2355 vrele(fromnd.ni_startdir);
2356 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2358 vrele(fromnd.ni_dvp);
2360 vrele(fromnd.ni_vp);
2362 vn_finished_write(mp);
2370 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2371 struct thread *td, struct mbuf **mrq)
2373 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2374 struct sockaddr *nam = nfsd->nd_nam;
2375 caddr_t dpos = nfsd->nd_dpos;
2376 struct ucred *cred = &nfsd->nd_cr;
2377 struct nameidata nd;
2379 int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1;
2380 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2381 struct mbuf *mb, *mreq;
2382 struct vnode *vp = NULL, *xp, *dirp = (struct vnode *)0;
2383 struct vattr dirfor, diraft, at;
2385 fhandle_t *fhp, *dfhp;
2386 struct mount *mp = NULL;
2388 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2391 fhp = &nfh.fh_generic;
2392 dfhp = &dnfh.fh_generic;
2394 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2398 if ((error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp)) != 0) {
2402 (void) vn_start_write(vp, &mp, V_WAIT);
2405 nfsm_srvmtofh(dfhp);
2406 nfsm_srvnamesiz(len);
2408 error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam, &rdonly, TRUE);
2410 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2412 nfsm_srvpostop_attr(getret, &at);
2413 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2419 if (vp->v_type == VDIR) {
2420 error = EPERM; /* POSIX */
2423 nd.ni_cnd.cn_cred = cred;
2424 nd.ni_cnd.cn_nameiop = CREATE;
2425 nd.ni_cnd.cn_flags = LOCKPARENT;
2426 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
2430 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2446 if (vp->v_mount != xp->v_mount)
2450 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2451 NDFREE(&nd, NDF_ONLY_PNBUF);
2457 getret = VOP_GETATTR(vp, &at, cred, td);
2459 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2461 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2463 nfsm_srvpostop_attr(getret, &at);
2464 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2470 NDFREE(&nd, NDF_ONLY_PNBUF);
2476 if (nd.ni_dvp == nd.ni_vp)
2483 vn_finished_write(mp);
2488 * nfs symbolic link service
2491 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2492 struct thread *td, struct mbuf **mrq)
2494 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2495 struct sockaddr *nam = nfsd->nd_nam;
2496 caddr_t dpos = nfsd->nd_dpos;
2497 struct ucred *cred = &nfsd->nd_cr;
2498 struct vattr va, dirfor, diraft;
2499 struct nameidata nd;
2500 struct vattr *vap = &va;
2501 struct nfsv2_sattr *sp;
2502 char *bpos, *pathcp = (char *)0;
2505 int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2506 int v3 = (nfsd->nd_flag & ND_NFSV3);
2507 struct mbuf *mb, *mreq;
2508 struct vnode *dirp = (struct vnode *)0;
2511 struct mount *mp = NULL;
2514 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2517 fhp = &nfh.fh_generic;
2519 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2523 if ((error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp)) != 0) {
2527 (void) vn_start_write(vp, &mp, V_WAIT);
2530 nfsm_srvnamesiz(len);
2531 nd.ni_cnd.cn_cred = cred;
2532 nd.ni_cnd.cn_nameiop = CREATE;
2533 nd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART;
2534 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2538 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2551 nfsm_strsiz(len2, NFS_MAXPATHLEN);
2552 MALLOC(pathcp, caddr_t, len2 + 1, M_TEMP, M_WAITOK);
2553 iv.iov_base = pathcp;
2555 io.uio_resid = len2;
2559 io.uio_segflg = UIO_SYSSPACE;
2560 io.uio_rw = UIO_READ;
2561 io.uio_td = (struct thread *)0;
2562 nfsm_mtouio(&io, len2);
2564 sp = nfsm_dissect(struct nfsv2_sattr *, NFSX_V2SATTR);
2565 vap->va_mode = nfstov_mode(sp->sa_mode);
2567 *(pathcp + len2) = '\0';
2574 * issue symlink op. SAVESTART is set so the underlying path component
2575 * is only freed by the VOP if an error occurs.
2577 if (vap->va_mode == (mode_t)VNOVAL)
2579 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap, pathcp);
2581 NDFREE(&nd, NDF_ONLY_PNBUF);
2586 * releases directory prior to potential lookup op.
2594 * Issue lookup. Leave SAVESTART set so we can easily free
2595 * the name buffer later on.
2597 * since LOCKPARENT is not set, ni_dvp will be garbage on
2598 * return whether an error occurs or not.
2600 nd.ni_cnd.cn_nameiop = LOOKUP;
2601 nd.ni_cnd.cn_flags &= ~(LOCKPARENT | FOLLOW);
2602 nd.ni_cnd.cn_flags |= (NOFOLLOW | LOCKLEAF);
2603 nd.ni_cnd.cn_thread = td;
2604 nd.ni_cnd.cn_cred = cred;
2606 error = lookup(&nd);
2610 bzero((caddr_t)fhp, sizeof(nfh));
2611 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2612 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2614 error = VOP_GETATTR(nd.ni_vp, vap, cred,
2623 * These releases aren't strictly required, does even doing them
2624 * make any sense? XXX can nfsm_reply() block?
2627 FREE(pathcp, M_TEMP);
2631 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2635 if (nd.ni_startdir) {
2636 vrele(nd.ni_startdir);
2637 nd.ni_startdir = NULL;
2639 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2642 nfsm_srvpostop_fh(fhp);
2643 nfsm_srvpostop_attr(0, vap);
2645 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2651 NDFREE(&nd, NDF_ONLY_PNBUF);
2653 if (nd.ni_dvp == nd.ni_vp)
2661 vrele(nd.ni_startdir);
2665 FREE(pathcp, M_TEMP);
2667 vn_finished_write(mp);
2675 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2676 struct thread *td, struct mbuf **mrq)
2678 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2679 struct sockaddr *nam = nfsd->nd_nam;
2680 caddr_t dpos = nfsd->nd_dpos;
2681 struct ucred *cred = &nfsd->nd_cr;
2682 struct vattr va, dirfor, diraft;
2683 struct vattr *vap = &va;
2684 struct nfs_fattr *fp;
2685 struct nameidata nd;
2688 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2689 int v3 = (nfsd->nd_flag & ND_NFSV3);
2690 struct mbuf *mb, *mreq;
2691 struct vnode *dirp = NULL;
2695 struct mount *mp = NULL;
2698 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2701 fhp = &nfh.fh_generic;
2703 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2707 if ((error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp)) != 0) {
2711 (void) vn_start_write(vp, &mp, V_WAIT);
2714 nfsm_srvnamesiz(len);
2715 nd.ni_cnd.cn_cred = cred;
2716 nd.ni_cnd.cn_nameiop = CREATE;
2717 nd.ni_cnd.cn_flags = LOCKPARENT;
2719 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2723 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2731 nfsm_reply(NFSX_WCCDATA(v3));
2733 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2741 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2742 vap->va_mode = nfstov_mode(*tl++);
2746 * At this point nd.ni_dvp is referenced and exclusively locked and
2747 * nd.ni_vp, if it exists, is referenced but not locked.
2750 vap->va_type = VDIR;
2751 if (nd.ni_vp != NULL) {
2752 NDFREE(&nd, NDF_ONLY_PNBUF);
2758 * Issue mkdir op. Since SAVESTART is not set, the pathname
2759 * component is freed by the VOP call. This will fill-in
2760 * nd.ni_vp, reference, and exclusively lock it.
2762 if (vap->va_mode == (mode_t)VNOVAL)
2764 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2765 NDFREE(&nd, NDF_ONLY_PNBUF);
2772 bzero((caddr_t)fhp, sizeof(nfh));
2773 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2774 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2776 error = VOP_GETATTR(nd.ni_vp, vap, cred, td);
2780 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2781 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2784 nfsm_srvpostop_fh(fhp);
2785 nfsm_srvpostop_attr(0, vap);
2787 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2788 } else if (!error) {
2789 /* v2 non-error case. */
2790 nfsm_srvfhtom(fhp, v3);
2791 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
2792 nfsm_srvfillattr(vap, fp);
2801 NDFREE(&nd, NDF_ONLY_PNBUF);
2802 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2813 vn_finished_write(mp);
2821 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2822 struct thread *td, struct mbuf **mrq)
2824 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2825 struct sockaddr *nam = nfsd->nd_nam;
2826 caddr_t dpos = nfsd->nd_dpos;
2827 struct ucred *cred = &nfsd->nd_cr;
2829 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2830 int v3 = (nfsd->nd_flag & ND_NFSV3);
2831 struct mbuf *mb, *mreq;
2832 struct vnode *vp, *dirp = (struct vnode *)0;
2833 struct vattr dirfor, diraft;
2836 struct nameidata nd;
2837 struct mount *mp = NULL;
2839 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2842 fhp = &nfh.fh_generic;
2844 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2848 if ((error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp)) != 0) {
2852 (void) vn_start_write(vp, &mp, V_WAIT);
2855 nfsm_srvnamesiz(len);
2856 nd.ni_cnd.cn_cred = cred;
2857 nd.ni_cnd.cn_nameiop = DELETE;
2858 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2859 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2863 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2871 nfsm_reply(NFSX_WCCDATA(v3));
2873 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2878 if (vp->v_type != VDIR) {
2883 * No rmdir "." please.
2885 if (nd.ni_dvp == vp) {
2890 * The root of a mounted filesystem cannot be deleted.
2892 if (vp->v_flag & VROOT)
2896 * Issue or abort op. Since SAVESTART is not set, path name
2897 * component is freed by the VOP after either.
2900 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2901 NDFREE(&nd, NDF_ONLY_PNBUF);
2904 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2905 nfsm_reply(NFSX_WCCDATA(v3));
2907 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2913 NDFREE(&nd, NDF_ONLY_PNBUF);
2917 if (nd.ni_dvp == nd.ni_vp)
2925 vn_finished_write(mp);
2930 * nfs readdir service
2931 * - mallocs what it thinks is enough to read
2932 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2933 * - calls VOP_READDIR()
2934 * - loops around building the reply
2935 * if the output generated exceeds count break out of loop
2936 * The nfsm_clget macro is used here so that the reply will be packed
2937 * tightly in mbuf clusters.
2938 * - it only knows that it has encountered eof when the VOP_READDIR()
2940 * - as such one readdir rpc will return eof false although you are there
2941 * and then the next will return eof
2942 * - it trims out records with d_fileno == 0
2943 * this doesn't matter for Unix clients, but they might confuse clients
2945 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2946 * than requested, but this may not apply to all filesystems. For
2947 * example, client NFS does not { although it is never remote mounted
2949 * The alternate call nfsrv_readdirplus() does lookups as well.
2950 * PS: The NFS protocol spec. does not clarify what the "count" byte
2951 * argument is a count of.. just name strings and file id's or the
2952 * entire reply rpc or ...
2953 * I tried just file name and id sizes and it confused the Sun client,
2954 * so I am using the full rpc size now. The "paranoia.." comment refers
2955 * to including the status longwords that are not a part of the dir.
2956 * "entry" structures, but are in the rpc.
2960 u_int32_t fl_postopok;
2961 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2963 u_int32_t fl_fhsize;
2964 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2968 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2969 struct thread *td, struct mbuf **mrq)
2971 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2972 struct sockaddr *nam = nfsd->nd_nam;
2973 caddr_t dpos = nfsd->nd_dpos;
2974 struct ucred *cred = &nfsd->nd_cr;
2981 struct mbuf *mb, *mreq;
2982 char *cpos, *cend, *rbuf;
2983 struct vnode *vp = NULL;
2989 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2990 int siz, cnt, fullsiz, eofflag, rdonly, ncookies;
2991 int v3 = (nfsd->nd_flag & ND_NFSV3);
2992 u_quad_t off, toff, verf;
2993 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
2995 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2996 fhp = &nfh.fh_generic;
2999 tl = nfsm_dissect(u_int32_t *, 5 * NFSX_UNSIGNED);
3000 toff = fxdr_hyper(tl);
3002 verf = fxdr_hyper(tl);
3005 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
3006 toff = fxdr_unsigned(u_quad_t, *tl++);
3007 verf = 0; /* shut up gcc */
3010 cnt = fxdr_unsigned(int, *tl);
3011 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3012 xfer = NFS_SRVMAXDATA(nfsd);
3016 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
3017 if (!error && vp->v_type != VDIR) {
3023 nfsm_reply(NFSX_UNSIGNED);
3025 nfsm_srvpostop_attr(getret, &at);
3031 * Obtain lock on vnode for this section of the code
3034 error = getret = VOP_GETATTR(vp, &at, cred, td);
3037 * XXX This check may be too strict for Solaris 2.5 clients.
3039 if (!error && toff && verf && verf != at.va_filerev)
3040 error = NFSERR_BAD_COOKIE;
3044 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3048 nfsm_reply(NFSX_POSTOPATTR(v3));
3050 nfsm_srvpostop_attr(getret, &at);
3054 VOP_UNLOCK(vp, 0, td);
3057 * end section. Allocate rbuf and continue
3059 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3062 iv.iov_len = fullsiz;
3065 io.uio_offset = (off_t)off;
3066 io.uio_resid = fullsiz;
3067 io.uio_segflg = UIO_SYSSPACE;
3068 io.uio_rw = UIO_READ;
3069 io.uio_td = (struct thread *)0;
3071 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
3073 free((caddr_t)cookies, M_TEMP);
3076 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3077 off = (off_t)io.uio_offset;
3078 if (!cookies && !error)
3079 error = NFSERR_PERM;
3081 getret = VOP_GETATTR(vp, &at, cred, td);
3085 VOP_UNLOCK(vp, 0, td);
3089 free((caddr_t)rbuf, M_TEMP);
3091 free((caddr_t)cookies, M_TEMP);
3092 nfsm_reply(NFSX_POSTOPATTR(v3));
3094 nfsm_srvpostop_attr(getret, &at);
3099 siz -= io.uio_resid;
3102 * If nothing read, return eof
3108 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
3111 nfsm_srvpostop_attr(getret, &at);
3112 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3113 txdr_hyper(at.va_filerev, tl);
3116 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3117 *tl++ = nfsrv_nfs_false;
3118 *tl = nfsrv_nfs_true;
3119 FREE((caddr_t)rbuf, M_TEMP);
3120 FREE((caddr_t)cookies, M_TEMP);
3127 * Check for degenerate cases of nothing useful read.
3128 * If so go try again
3132 dp = (struct dirent *)cpos;
3135 * For some reason FreeBSD's ufs_readdir() chooses to back the
3136 * directory offset up to a block boundary, so it is necessary to
3137 * skip over the records that precede the requested offset. This
3138 * requires the assumption that file offset cookies monotonically
3141 while (cpos < cend && ncookies > 0 &&
3142 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3143 ((u_quad_t)(*cookiep)) <= toff)) {
3144 cpos += dp->d_reclen;
3145 dp = (struct dirent *)cpos;
3149 if (cpos >= cend || ncookies == 0) {
3155 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
3156 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
3158 nfsm_srvpostop_attr(getret, &at);
3159 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3160 txdr_hyper(at.va_filerev, tl);
3164 be = bp + M_TRAILINGSPACE(mp);
3166 /* Loop through the records and build reply */
3167 while (cpos < cend && ncookies > 0) {
3168 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3169 nlen = dp->d_namlen;
3170 rem = nfsm_rndup(nlen) - nlen;
3171 len += (4 * NFSX_UNSIGNED + nlen + rem);
3173 len += 2 * NFSX_UNSIGNED;
3179 * Build the directory record xdr from
3183 *tl = nfsrv_nfs_true;
3184 bp += NFSX_UNSIGNED;
3188 bp += NFSX_UNSIGNED;
3191 *tl = txdr_unsigned(dp->d_fileno);
3192 bp += NFSX_UNSIGNED;
3194 *tl = txdr_unsigned(nlen);
3195 bp += NFSX_UNSIGNED;
3197 /* And loop around copying the name */
3206 bcopy(cp, bp, tsiz);
3212 /* And null pad to a int32_t boundary */
3213 for (i = 0; i < rem; i++)
3217 /* Finish off the record */
3220 bp += NFSX_UNSIGNED;
3223 *tl = txdr_unsigned(*cookiep);
3224 bp += NFSX_UNSIGNED;
3226 cpos += dp->d_reclen;
3227 dp = (struct dirent *)cpos;
3234 *tl = nfsrv_nfs_false;
3235 bp += NFSX_UNSIGNED;
3238 *tl = nfsrv_nfs_true;
3240 *tl = nfsrv_nfs_false;
3241 bp += NFSX_UNSIGNED;
3244 mp->m_len = bp - mtod(mp, caddr_t);
3246 mp->m_len += bp - bpos;
3247 FREE((caddr_t)rbuf, M_TEMP);
3248 FREE((caddr_t)cookies, M_TEMP);
3257 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3258 struct thread *td, struct mbuf **mrq)
3260 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3261 struct sockaddr *nam = nfsd->nd_nam;
3262 caddr_t dpos = nfsd->nd_dpos;
3263 struct ucred *cred = &nfsd->nd_cr;
3270 struct mbuf *mb, *mreq;
3271 char *cpos, *cend, *rbuf;
3272 struct vnode *vp = NULL, *nvp;
3275 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3278 struct vattr va, at, *vap = &va;
3279 struct nfs_fattr *fp;
3280 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3281 int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies;
3282 u_quad_t off, toff, verf;
3283 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3284 int v3 = (nfsd->nd_flag & ND_NFSV3);
3286 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3288 panic("nfsrv_readdirplus: v3 proc called on a v2 connection");
3289 fhp = &nfh.fh_generic;
3291 tl = nfsm_dissect(u_int32_t *, 6 * NFSX_UNSIGNED);
3292 toff = fxdr_hyper(tl);
3294 verf = fxdr_hyper(tl);
3296 siz = fxdr_unsigned(int, *tl++);
3297 cnt = fxdr_unsigned(int, *tl);
3299 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3300 xfer = NFS_SRVMAXDATA(nfsd);
3304 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
3305 if (!error && vp->v_type != VDIR) {
3311 nfsm_reply(NFSX_UNSIGNED);
3312 nfsm_srvpostop_attr(getret, &at);
3316 error = getret = VOP_GETATTR(vp, &at, cred, td);
3319 * XXX This check may be too strict for Solaris 2.5 clients.
3321 if (!error && toff && verf && verf != at.va_filerev)
3322 error = NFSERR_BAD_COOKIE;
3325 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3329 nfsm_reply(NFSX_V3POSTOPATTR);
3330 nfsm_srvpostop_attr(getret, &at);
3334 VOP_UNLOCK(vp, 0, td);
3335 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3338 iv.iov_len = fullsiz;
3341 io.uio_offset = (off_t)off;
3342 io.uio_resid = fullsiz;
3343 io.uio_segflg = UIO_SYSSPACE;
3344 io.uio_rw = UIO_READ;
3345 io.uio_td = (struct thread *)0;
3347 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
3349 free((caddr_t)cookies, M_TEMP);
3352 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3353 off = (u_quad_t)io.uio_offset;
3354 getret = VOP_GETATTR(vp, &at, cred, td);
3355 VOP_UNLOCK(vp, 0, td);
3356 if (!cookies && !error)
3357 error = NFSERR_PERM;
3364 free((caddr_t)cookies, M_TEMP);
3365 free((caddr_t)rbuf, M_TEMP);
3366 nfsm_reply(NFSX_V3POSTOPATTR);
3367 nfsm_srvpostop_attr(getret, &at);
3372 siz -= io.uio_resid;
3375 * If nothing read, return eof
3381 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3383 nfsm_srvpostop_attr(getret, &at);
3384 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3385 txdr_hyper(at.va_filerev, tl);
3387 *tl++ = nfsrv_nfs_false;
3388 *tl = nfsrv_nfs_true;
3389 FREE((caddr_t)cookies, M_TEMP);
3390 FREE((caddr_t)rbuf, M_TEMP);
3397 * Check for degenerate cases of nothing useful read.
3398 * If so go try again
3402 dp = (struct dirent *)cpos;
3405 * For some reason FreeBSD's ufs_readdir() chooses to back the
3406 * directory offset up to a block boundary, so it is necessary to
3407 * skip over the records that precede the requested offset. This
3408 * requires the assumption that file offset cookies monotonically
3411 while (cpos < cend && ncookies > 0 &&
3412 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3413 ((u_quad_t)(*cookiep)) <= toff)) {
3414 cpos += dp->d_reclen;
3415 dp = (struct dirent *)cpos;
3419 if (cpos >= cend || ncookies == 0) {
3426 * Probe one of the directory entries to see if the filesystem
3429 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp) == EOPNOTSUPP) {
3430 error = NFSERR_NOTSUPP;
3433 free((caddr_t)cookies, M_TEMP);
3434 free((caddr_t)rbuf, M_TEMP);
3435 nfsm_reply(NFSX_V3POSTOPATTR);
3436 nfsm_srvpostop_attr(getret, &at);
3443 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3446 nfsm_srvpostop_attr(getret, &at);
3447 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3448 txdr_hyper(at.va_filerev, tl);
3451 be = bp + M_TRAILINGSPACE(mp);
3453 /* Loop through the records and build reply */
3454 while (cpos < cend && ncookies > 0) {
3455 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3456 nlen = dp->d_namlen;
3457 rem = nfsm_rndup(nlen)-nlen;
3460 * For readdir_and_lookup get the vnode using
3463 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
3465 bzero((caddr_t)nfhp, NFSX_V3FH);
3467 nvp->v_mount->mnt_stat.f_fsid;
3468 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
3473 if (VOP_GETATTR(nvp, vap, cred, td)) {
3482 * If either the dircount or maxcount will be
3483 * exceeded, get out now. Both of these lengths
3484 * are calculated conservatively, including all
3487 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3489 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3490 if (len > cnt || dirlen > fullsiz) {
3496 * Build the directory record xdr from
3499 fp = (struct nfs_fattr *)&fl.fl_fattr;
3500 nfsm_srvfillattr(vap, fp);
3501 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3502 fl.fl_fhok = nfsrv_nfs_true;
3503 fl.fl_postopok = nfsrv_nfs_true;
3504 fl.fl_off.nfsuquad[0] = 0;
3505 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3508 *tl = nfsrv_nfs_true;
3509 bp += NFSX_UNSIGNED;
3512 bp += NFSX_UNSIGNED;
3514 *tl = txdr_unsigned(dp->d_fileno);
3515 bp += NFSX_UNSIGNED;
3517 *tl = txdr_unsigned(nlen);
3518 bp += NFSX_UNSIGNED;
3520 /* And loop around copying the name */
3525 if ((bp + xfer) > be)
3529 bcopy(cp, bp, tsiz);
3535 /* And null pad to a int32_t boundary */
3536 for (i = 0; i < rem; i++)
3540 * Now copy the flrep structure out.
3542 xfer = sizeof (struct flrep);
3546 if ((bp + xfer) > be)
3550 bcopy(cp, bp, tsiz);
3558 cpos += dp->d_reclen;
3559 dp = (struct dirent *)cpos;
3566 *tl = nfsrv_nfs_false;
3567 bp += NFSX_UNSIGNED;
3570 *tl = nfsrv_nfs_true;
3572 *tl = nfsrv_nfs_false;
3573 bp += NFSX_UNSIGNED;
3576 mp->m_len = bp - mtod(mp, caddr_t);
3578 mp->m_len += bp - bpos;
3579 FREE((caddr_t)cookies, M_TEMP);
3580 FREE((caddr_t)rbuf, M_TEMP);
3588 * nfs commit service
3591 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3592 struct thread *td, struct mbuf **mrq)
3594 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3595 struct sockaddr *nam = nfsd->nd_nam;
3596 caddr_t dpos = nfsd->nd_dpos;
3597 struct ucred *cred = &nfsd->nd_cr;
3598 struct vattr bfor, aft;
3599 struct vnode *vp = NULL;
3604 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt;
3605 struct mbuf *mb, *mreq;
3607 struct mount *mp = NULL;
3608 int v3 = (nfsd->nd_flag & ND_NFSV3);
3610 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3612 panic("nfsrv_commit: v3 proc called on a v2 connection");
3613 fhp = &nfh.fh_generic;
3615 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
3619 if ((error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp)) != 0) {
3623 (void) vn_start_write(vp, &mp, V_WAIT);
3626 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
3629 * XXX At this time VOP_FSYNC() does not accept offset and byte
3630 * count parameters, so these arguments are useless (someday maybe).
3632 off = fxdr_hyper(tl);
3634 cnt = fxdr_unsigned(int, *tl);
3635 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
3637 nfsm_reply(2 * NFSX_UNSIGNED);
3638 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3642 for_ret = VOP_GETATTR(vp, &bfor, cred, td);
3644 if (cnt > MAX_COMMIT_COUNT) {
3646 * Give up and do the whole thing
3649 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3650 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3652 error = VOP_FSYNC(vp, cred, MNT_WAIT, td);
3655 * Locate and synchronously write any buffers that fall
3656 * into the requested range. Note: we are assuming that
3657 * f_iosize is a power of 2.
3659 int iosize = vp->v_mount->mnt_stat.f_iosize;
3660 int iomask = iosize - 1;
3665 * Align to iosize boundry, super-align to page boundry.
3668 cnt += off & iomask;
3669 off &= ~(u_quad_t)iomask;
3671 if (off & PAGE_MASK) {
3672 cnt += off & PAGE_MASK;
3673 off &= ~(u_quad_t)PAGE_MASK;
3675 lblkno = off / iosize;
3678 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3679 vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
3687 * If we have a buffer and it is marked B_DELWRI we
3688 * have to lock and write it. Otherwise the prior
3689 * write is assumed to have already been committed.
3691 if ((bp = gbincore(vp, lblkno)) != NULL && (bp->b_flags & B_DELWRI)) {
3692 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
3693 BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL);
3694 continue; /* retry */
3697 bp->b_flags &= ~B_ASYNC;
3710 aft_ret = VOP_GETATTR(vp, &aft, cred, td);
3714 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3715 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3717 tl = nfsm_build(u_int32_t *, NFSX_V3WRITEVERF);
3718 if (nfsver.tv_sec == 0)
3720 *tl++ = txdr_unsigned(nfsver.tv_sec);
3721 *tl = txdr_unsigned(nfsver.tv_usec);
3728 vn_finished_write(mp);
3733 * nfs statfs service
3736 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3737 struct thread *td, struct mbuf **mrq)
3739 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3740 struct sockaddr *nam = nfsd->nd_nam;
3741 caddr_t dpos = nfsd->nd_dpos;
3742 struct ucred *cred = &nfsd->nd_cr;
3744 struct nfs_statfs *sfp;
3746 int error = 0, rdonly, getret = 1;
3747 int v3 = (nfsd->nd_flag & ND_NFSV3);
3748 struct mbuf *mb, *mreq;
3749 struct vnode *vp = NULL;
3753 struct statfs statfs;
3756 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3757 fhp = &nfh.fh_generic;
3759 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
3761 nfsm_reply(NFSX_UNSIGNED);
3763 nfsm_srvpostop_attr(getret, &at);
3768 error = VFS_STATFS(vp->v_mount, sf, td);
3769 getret = VOP_GETATTR(vp, &at, cred, td);
3772 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3774 nfsm_srvpostop_attr(getret, &at);
3779 sfp = nfsm_build(struct nfs_statfs *, NFSX_STATFS(v3));
3781 tval = (u_quad_t)sf->f_blocks;
3782 tval *= (u_quad_t)sf->f_bsize;
3783 txdr_hyper(tval, &sfp->sf_tbytes);
3784 tval = (u_quad_t)sf->f_bfree;
3785 tval *= (u_quad_t)sf->f_bsize;
3786 txdr_hyper(tval, &sfp->sf_fbytes);
3787 tval = (u_quad_t)sf->f_bavail;
3788 tval *= (u_quad_t)sf->f_bsize;
3789 txdr_hyper(tval, &sfp->sf_abytes);
3790 sfp->sf_tfiles.nfsuquad[0] = 0;
3791 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
3792 sfp->sf_ffiles.nfsuquad[0] = 0;
3793 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3794 sfp->sf_afiles.nfsuquad[0] = 0;
3795 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3796 sfp->sf_invarsec = 0;
3798 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3799 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3800 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3801 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3802 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3811 * nfs fsinfo service
3814 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3815 struct thread *td, struct mbuf **mrq)
3817 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3818 struct sockaddr *nam = nfsd->nd_nam;
3819 caddr_t dpos = nfsd->nd_dpos;
3820 struct ucred *cred = &nfsd->nd_cr;
3821 struct nfsv3_fsinfo *sip;
3823 int error = 0, rdonly, getret = 1, pref;
3824 struct mbuf *mb, *mreq;
3825 struct vnode *vp = NULL;
3831 int v3 = (nfsd->nd_flag & ND_NFSV3);
3833 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3835 panic("nfsrv_fsinfo: v3 proc called on a v2 connection");
3836 fhp = &nfh.fh_generic;
3838 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
3840 nfsm_reply(NFSX_UNSIGNED);
3841 nfsm_srvpostop_attr(getret, &at);
3846 /* XXX Try to make a guess on the max file size. */
3847 VFS_STATFS(vp->v_mount, &sb, td);
3848 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3850 getret = VOP_GETATTR(vp, &at, cred, td);
3853 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3854 nfsm_srvpostop_attr(getret, &at);
3855 sip = nfsm_build(struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3859 * There should be file system VFS OP(s) to get this information.
3860 * For now, assume ufs.
3862 if (slp->ns_so->so_type == SOCK_DGRAM)
3863 pref = NFS_MAXDGRAMDATA;
3866 sip->fs_rtmax = txdr_unsigned(NFS_MAXDATA);
3867 sip->fs_rtpref = txdr_unsigned(pref);
3868 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3869 sip->fs_wtmax = txdr_unsigned(NFS_MAXDATA);
3870 sip->fs_wtpref = txdr_unsigned(pref);
3871 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3872 sip->fs_dtpref = txdr_unsigned(pref);
3873 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3874 sip->fs_timedelta.nfsv3_sec = 0;
3875 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3876 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3877 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3878 NFSV3FSINFO_CANSETTIME);
3886 * nfs pathconf service
3889 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3890 struct thread *td, struct mbuf **mrq)
3892 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3893 struct sockaddr *nam = nfsd->nd_nam;
3894 caddr_t dpos = nfsd->nd_dpos;
3895 struct ucred *cred = &nfsd->nd_cr;
3896 struct nfsv3_pathconf *pc;
3898 int error = 0, rdonly, getret = 1;
3899 register_t linkmax, namemax, chownres, notrunc;
3900 struct mbuf *mb, *mreq;
3901 struct vnode *vp = NULL;
3905 int v3 = (nfsd->nd_flag & ND_NFSV3);
3907 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3909 panic("nfsrv_pathconf: v3 proc called on a v2 connection");
3910 fhp = &nfh.fh_generic;
3912 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly, TRUE);
3914 nfsm_reply(NFSX_UNSIGNED);
3915 nfsm_srvpostop_attr(getret, &at);
3919 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3921 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3923 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3925 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc);
3926 getret = VOP_GETATTR(vp, &at, cred, td);
3929 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3930 nfsm_srvpostop_attr(getret, &at);
3935 pc = nfsm_build(struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3937 pc->pc_linkmax = txdr_unsigned(linkmax);
3938 pc->pc_namemax = txdr_unsigned(namemax);
3939 pc->pc_notrunc = txdr_unsigned(notrunc);
3940 pc->pc_chownrestricted = txdr_unsigned(chownres);
3943 * These should probably be supported by VOP_PATHCONF(), but
3944 * until msdosfs is exportable (why would you want to?), the
3945 * Unix defaults should be ok.
3947 pc->pc_caseinsensitive = nfsrv_nfs_false;
3948 pc->pc_casepreserving = nfsrv_nfs_true;
3956 * Null operation, used by clients to ping server
3960 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3961 struct thread *td, struct mbuf **mrq)
3963 struct mbuf *mrep = nfsd->nd_mrep;
3965 int error = NFSERR_RETVOID;
3966 struct mbuf *mb, *mreq;
3968 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3975 * No operation, used for obsolete procedures
3979 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3980 struct thread *td, struct mbuf **mrq)
3982 struct mbuf *mrep = nfsd->nd_mrep;
3985 struct mbuf *mb, *mreq;
3987 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3988 if (nfsd->nd_repstat)
3989 error = nfsd->nd_repstat;
3991 error = EPROCUNAVAIL;
3999 * Perform access checking for vnodes obtained from file handles that would
4000 * refer to files already opened by a Unix client. You cannot just use
4001 * vn_writechk() and VOP_ACCESS() for two reasons.
4002 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
4003 * 2 - The owner is to be given access irrespective of mode bits for some
4004 * operations, so that processes that chmod after opening a file don't
4005 * break. I don't like this because it opens a security hole, but since
4006 * the nfs server opens a security hole the size of a barn door anyhow,
4009 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
4010 * will return EPERM instead of EACCESS. EPERM is always an error.
4013 nfsrv_access(struct vnode *vp, int flags, struct ucred *cred, int rdonly,
4014 struct thread *td, int override)
4019 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4020 if (flags & VWRITE) {
4021 /* Just vn_writechk() changed to check rdonly */
4023 * Disallow write attempts on read-only file systems;
4024 * unless the file is a socket or a block or character
4025 * device resident on the file system.
4027 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
4028 switch (vp->v_type) {
4038 * If there's shared text associated with
4039 * the inode, we can't allow writing.
4041 if (vp->v_flag & VTEXT)
4044 error = VOP_GETATTR(vp, &vattr, cred, td);
4047 error = VOP_ACCESS(vp, flags, cred, td);
4049 * Allow certain operations for the owner (reads and writes
4050 * on files that are already open).
4052 if (override && error == EACCES && cred->cr_uid == vattr.va_uid)