xref: /netbsd-src/sys/nfs/nfs_serv.c (revision da9817918ec7e88db2912a2882967c7570a83f47)
1 /*	$NetBSD: nfs_serv.c,v 1.146 2009/05/23 15:31:21 ad Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_serv.c	8.8 (Berkeley) 7/31/95
35  */
36 
37 /*
38  * nfs version 2 and 3 server calls to vnode ops
39  * - these routines generally have 3 phases
40  *   1 - break down and validate rpc request in mbuf list
41  *   2 - do the vnode ops for the request
42  *       (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
43  *   3 - build the rpc reply in an mbuf list
44  *   nb:
45  *	- do not mix the phases, since the nfsm_?? macros can return failures
46  *	  on a bad rpc or similar and do not do any vrele() or vput()'s
47  *
48  *      - the nfsm_reply() macro generates an nfs rpc reply with the nfs
49  *	error number iff error != 0 whereas
50  *	returning an error from the server function implies a fatal error
51  *	such as a badly constructed rpc request that should be dropped without
52  *	a reply.
53  *	For Version 3, nfsm_reply() does not return for the error case, since
54  *	most version 3 rpcs return more than the status for error cases.
55  */
56 
57 #include <sys/cdefs.h>
58 __KERNEL_RCSID(0, "$NetBSD: nfs_serv.c,v 1.146 2009/05/23 15:31:21 ad Exp $");
59 
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/proc.h>
63 #include <sys/file.h>
64 #include <sys/namei.h>
65 #include <sys/vnode.h>
66 #include <sys/mount.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/mbuf.h>
70 #include <sys/dirent.h>
71 #include <sys/stat.h>
72 #include <sys/kernel.h>
73 #include <sys/hash.h>
74 #include <sys/kauth.h>
75 #include <sys/module.h>
76 #include <sys/syscall.h>
77 #include <sys/syscallargs.h>
78 #include <sys/syscallvar.h>
79 
80 #include <uvm/uvm.h>
81 
82 #include <nfs/nfsproto.h>
83 #include <nfs/rpcv2.h>
84 #include <nfs/nfs.h>
85 #include <nfs/xdr_subs.h>
86 #include <nfs/nfsm_subs.h>
87 #include <nfs/nfs_var.h>
88 
89 MODULE(MODULE_CLASS_MISC, nfsserver, "nfs");
90 
91 /* Global vars */
92 extern u_int32_t nfs_xdrneg1;
93 extern u_int32_t nfs_false, nfs_true;
94 extern const enum vtype nv3tov_type[8];
95 extern struct nfsstats nfsstats;
96 extern const nfstype nfsv2_type[9];
97 extern const nfstype nfsv3_type[9];
98 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
99 bool nfsd_use_loan = true;	/* use page-loan for READ OP */
100 
101 #define	nqsrv_getl(vp, rw)	/* nothing */
102 
103 static const struct syscall_package nfsserver_syscalls[] = {
104 	{ SYS_nfssvc, 0, (sy_call_t *)sys_nfssvc },
105 	{ 0, 0, NULL },
106 };
107 
108 static int
109 nfsserver_modcmd(modcmd_t cmd, void *arg)
110 {
111 	extern struct vfs_hooks nfs_export_hooks;	/* XXX */
112 	int error;
113 
114 	switch (cmd) {
115 	case MODULE_CMD_INIT:
116 		error = syscall_establish(NULL, nfsserver_syscalls);
117 		if (error != 0) {
118 			return error;
119 		}
120 		nfs_init();	/* XXX for monolithic kernel */
121 		netexport_init();
122 		nfsrv_initcache();	/* Init the server request cache */
123 		nfsrv_init(0);		/* Init server data structures */
124 		vfs_hooks_attach(&nfs_export_hooks);
125 		nfs_timer_srvinit(nfsrv_timer);
126 		return 0;
127 	case MODULE_CMD_FINI:
128 		error = syscall_disestablish(NULL, nfsserver_syscalls);
129 		if (error != 0) {
130 			return error;
131 		}
132 		/*
133 		 * Kill export list before detaching VFS hooks, so we
134 		 * we don't leak state due to a concurrent umount().
135 		 */
136 		netexport_fini();
137 		vfs_hooks_detach(&nfs_export_hooks);
138 
139 		/* Kill timer before server goes away. */
140 		nfs_timer_srvfini();
141 		nfsrv_fini();
142 
143 		/* Server uses server cache, so kill cache last. */
144 		nfsrv_finicache();
145 		return 0;
146 	default:
147 		return ENOTTY;
148 	}
149 }
150 
151 /*
152  * nfs v3 access service
153  */
154 int
155 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
156 {
157 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
158 	struct mbuf *nam = nfsd->nd_nam;
159 	char *dpos = nfsd->nd_dpos;
160 	kauth_cred_t cred = nfsd->nd_cr;
161 	struct vnode *vp;
162 	nfsrvfh_t nsfh;
163 	u_int32_t *tl;
164 	int32_t t1;
165 	char *bpos;
166 	int error = 0, rdonly, cache = 0, getret;
167 	char *cp2;
168 	struct mbuf *mb, *mreq;
169 	struct vattr va;
170 	u_long inmode, testmode, outmode;
171 	u_quad_t frev;
172 
173 	nfsm_srvmtofh(&nsfh);
174 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
175 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly,
176 	    (nfsd->nd_flag & ND_KERBAUTH), false);
177 	if (error) {
178 		nfsm_reply(NFSX_UNSIGNED);
179 		nfsm_srvpostop_attr(1, (struct vattr *)0);
180 		return (0);
181 	}
182 	inmode = fxdr_unsigned(u_int32_t, *tl);
183 	outmode = 0;
184 	if ((inmode & NFSV3ACCESS_READ) &&
185 	    nfsrv_access(vp, VREAD, cred, rdonly, lwp, 0) == 0)
186 		outmode |= NFSV3ACCESS_READ;
187 	if (vp->v_type != VDIR) {
188 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
189 		if (testmode &&
190 		    nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0)
191 			outmode |= testmode;
192 		if ((inmode & NFSV3ACCESS_EXECUTE) &&
193 		    nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0)
194 			outmode |= NFSV3ACCESS_EXECUTE;
195 	} else {
196 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
197 		    NFSV3ACCESS_DELETE);
198 		if (testmode &&
199 		    nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0)
200 			outmode |= testmode;
201 		if ((inmode & NFSV3ACCESS_LOOKUP) &&
202 		    nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0)
203 			outmode |= NFSV3ACCESS_LOOKUP;
204 	}
205 	getret = VOP_GETATTR(vp, &va, cred);
206 	vput(vp);
207 	nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
208 	nfsm_srvpostop_attr(getret, &va);
209 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
210 	*tl = txdr_unsigned(outmode);
211 	nfsm_srvdone;
212 }
213 
214 /*
215  * nfs getattr service
216  */
217 int
218 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
219 {
220 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
221 	struct mbuf *nam = nfsd->nd_nam;
222 	char *dpos = nfsd->nd_dpos;
223 	kauth_cred_t cred = nfsd->nd_cr;
224 	struct nfs_fattr *fp;
225 	struct vattr va;
226 	struct vnode *vp;
227 	nfsrvfh_t nsfh;
228 	u_int32_t *tl;
229 	int32_t t1;
230 	char *bpos;
231 	int error = 0, rdonly, cache = 0;
232 	char *cp2;
233 	struct mbuf *mb, *mreq;
234 	u_quad_t frev;
235 
236 	nfsm_srvmtofh(&nsfh);
237 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly,
238 	    (nfsd->nd_flag & ND_KERBAUTH), false);
239 	if (error) {
240 		nfsm_reply(0);
241 		return (0);
242 	}
243 	nqsrv_getl(vp, ND_READ);
244 	error = VOP_GETATTR(vp, &va, cred);
245 	vput(vp);
246 	nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
247 	if (error)
248 		return (0);
249 	nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
250 	nfsm_srvfillattr(&va, fp);
251 	nfsm_srvdone;
252 }
253 
254 /*
255  * nfs setattr service
256  */
257 int
258 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
259 {
260 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
261 	struct mbuf *nam = nfsd->nd_nam;
262 	char *dpos = nfsd->nd_dpos;
263 	kauth_cred_t cred = nfsd->nd_cr;
264 	struct vattr va, preat;
265 	struct nfsv2_sattr *sp;
266 	struct nfs_fattr *fp;
267 	struct vnode *vp;
268 	nfsrvfh_t nsfh;
269 	u_int32_t *tl;
270 	int32_t t1;
271 	char *bpos;
272 	int error = 0, rdonly, cache = 0, preat_ret = 1, postat_ret = 1;
273 	int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
274 	char *cp2;
275 	struct mbuf *mb, *mreq;
276 	u_quad_t frev;
277 	struct timespec guard;
278 
279 	memset(&guard, 0, sizeof guard);	/* XXX gcc */
280 
281 	nfsm_srvmtofh(&nsfh);
282 	VATTR_NULL(&va);
283 	if (v3) {
284 		nfsm_srvsattr(&va);
285 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
286 		gcheck = fxdr_unsigned(int, *tl);
287 		if (gcheck) {
288 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
289 			fxdr_nfsv3time(tl, &guard);
290 		}
291 	} else {
292 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
293 		/*
294 		 * Nah nah nah nah na nah
295 		 * There is a bug in the Sun client that puts 0xffff in the mode
296 		 * field of sattr when it should put in 0xffffffff. The u_short
297 		 * doesn't sign extend.
298 		 * --> check the low order 2 bytes for 0xffff
299 		 */
300 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
301 			va.va_mode = nfstov_mode(sp->sa_mode);
302 		if (sp->sa_uid != nfs_xdrneg1)
303 			va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
304 		if (sp->sa_gid != nfs_xdrneg1)
305 			va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
306 		if (sp->sa_size != nfs_xdrneg1)
307 			va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
308 		if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
309 #ifdef notyet
310 			fxdr_nfsv2time(&sp->sa_atime, &va.va_atime);
311 #else
312 			va.va_atime.tv_sec =
313 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
314 			va.va_atime.tv_nsec = 0;
315 #endif
316 		}
317 		if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
318 			fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime);
319 
320 	}
321 
322 	/*
323 	 * Now that we have all the fields, lets do it.
324 	 */
325 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly,
326 	    (nfsd->nd_flag & ND_KERBAUTH), false);
327 	if (error) {
328 		nfsm_reply(2 * NFSX_UNSIGNED);
329 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
330 		return (0);
331 	}
332 	nqsrv_getl(vp, ND_WRITE);
333 	if (v3) {
334 		error = preat_ret = VOP_GETATTR(vp, &preat, cred);
335 		if (!error && gcheck &&
336 			(preat.va_ctime.tv_sec != guard.tv_sec ||
337 			 preat.va_ctime.tv_nsec != guard.tv_nsec))
338 			error = NFSERR_NOT_SYNC;
339 		if (error) {
340 			vput(vp);
341 			nfsm_reply(NFSX_WCCDATA(v3));
342 			nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
343 			return (0);
344 		}
345 	}
346 
347 	/*
348 	 * If the size is being changed write acces is required, otherwise
349 	 * just check for a read only file system.
350 	 */
351 	if (va.va_size == ((u_quad_t)((quad_t) -1))) {
352 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
353 			error = EROFS;
354 			goto out;
355 		}
356 	} else {
357 		if (vp->v_type == VDIR) {
358 			error = EISDIR;
359 			goto out;
360 		} else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
361 			lwp, 0)) != 0)
362 			goto out;
363 	}
364 	error = VOP_SETATTR(vp, &va, cred);
365 	postat_ret = VOP_GETATTR(vp, &va, cred);
366 	if (!error)
367 		error = postat_ret;
368 out:
369 	vput(vp);
370 	nfsm_reply(NFSX_WCCORFATTR(v3));
371 	if (v3) {
372 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
373 		return (0);
374 	} else {
375 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
376 		nfsm_srvfillattr(&va, fp);
377 	}
378 	nfsm_srvdone;
379 }
380 
381 /*
382  * nfs lookup rpc
383  */
384 int
385 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
386 {
387 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
388 	struct mbuf *nam = nfsd->nd_nam;
389 	char *dpos = nfsd->nd_dpos;
390 	kauth_cred_t cred = nfsd->nd_cr;
391 	struct nfs_fattr *fp;
392 	struct nameidata nd, ind, *ndp = &nd;
393 	struct vnode *vp, *dirp;
394 	nfsrvfh_t nsfh;
395 	char *cp;
396 	u_int32_t *tl;
397 	int32_t t1;
398 	char *bpos;
399 	int error = 0, cache = 0, dirattr_ret = 1;
400 	uint32_t len;
401 	int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
402 	char *cp2;
403 	struct mbuf *mb, *mreq;
404 	struct vattr va, dirattr;
405 	u_quad_t frev;
406 
407 	nfsm_srvmtofh(&nsfh);
408 	nfsm_srvnamesiz(len);
409 
410 	pubflag = nfs_ispublicfh(&nsfh);
411 
412 	nd.ni_cnd.cn_cred = cred;
413 	nd.ni_cnd.cn_nameiop = LOOKUP;
414 	nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART;
415 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
416 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
417 
418 	if (!error && pubflag) {
419 		if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
420 			/*
421 			 * Setup call to lookup() to see if we can find
422 			 * the index file. Arguably, this doesn't belong
423 			 * in a kernel.. Ugh.
424 			 */
425 			ind = nd;
426 			VOP_UNLOCK(nd.ni_vp, 0);
427 			ind.ni_pathlen = strlen(nfs_pub.np_index);
428 			ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
429 			    nfs_pub.np_index;
430 			ind.ni_startdir = nd.ni_vp;
431 			VREF(ind.ni_startdir);
432 			error = lookup(&ind);
433 			if (!error) {
434 				/*
435 				 * Found an index file. Get rid of
436 				 * the old references.
437 				 */
438 				if (dirp)
439 					vrele(dirp);
440 				dirp = nd.ni_vp;
441 				vrele(nd.ni_startdir);
442 				ndp = &ind;
443 			} else
444 				error = 0;
445 		}
446 		/*
447 		 * If the public filehandle was used, check that this lookup
448 		 * didn't result in a filehandle outside the publicly exported
449 		 * filesystem.
450 		 */
451 
452 		if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) {
453 			vput(nd.ni_vp);
454 			error = EPERM;
455 		}
456 	}
457 
458 	if (dirp) {
459 		if (v3)
460 			dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred);
461 		vrele(dirp);
462 	}
463 
464 	if (error) {
465 		nfsm_reply(NFSX_POSTOPATTR(v3));
466 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
467 		return (0);
468 	}
469 
470 	nqsrv_getl(ndp->ni_startdir, ND_READ);
471 	PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
472 	vp = ndp->ni_vp;
473 	error = nfsrv_composefh(vp, &nsfh, v3);
474 	if (!error)
475 		error = VOP_GETATTR(vp, &va, cred);
476 	vput(vp);
477 	vrele(ndp->ni_startdir);
478 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPORFATTR(v3) +
479 	    NFSX_POSTOPATTR(v3));
480 	if (error) {
481 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
482 		return (0);
483 	}
484 	nfsm_srvfhtom(&nsfh, v3);
485 	if (v3) {
486 		nfsm_srvpostop_attr(0, &va);
487 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
488 	} else {
489 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
490 		nfsm_srvfillattr(&va, fp);
491 	}
492 	nfsm_srvdone;
493 }
494 
495 /*
496  * nfs readlink service
497  */
498 int
499 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
500 {
501 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
502 	struct mbuf *nam = nfsd->nd_nam;
503 	char *dpos = nfsd->nd_dpos;
504 	kauth_cred_t cred = nfsd->nd_cr;
505 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
506 	struct iovec *ivp = iv;
507 	struct mbuf *mp;
508 	u_int32_t *tl;
509 	int32_t t1;
510 	char *bpos;
511 	int error = 0, rdonly, cache = 0, i, padlen, getret;
512 	uint32_t len;
513 	int v3 = (nfsd->nd_flag & ND_NFSV3);
514 	char *cp2;
515 	struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq;
516 	struct vnode *vp;
517 	struct vattr attr;
518 	nfsrvfh_t nsfh;
519 	struct uio io, *uiop = &io;
520 	u_quad_t frev;
521 
522 	nfsm_srvmtofh(&nsfh);
523 	len = 0;
524 	i = 0;
525 	while (len < NFS_MAXPATHLEN) {
526 		mp = m_get(M_WAIT, MT_DATA);
527 		MCLAIM(mp, &nfs_mowner);
528 		m_clget(mp, M_WAIT);
529 		mp->m_len = NFSMSIZ(mp);
530 		if (len == 0)
531 			mp3 = mp2 = mp;
532 		else {
533 			mp2->m_next = mp;
534 			mp2 = mp;
535 		}
536 		if ((len+mp->m_len) > NFS_MAXPATHLEN) {
537 			mp->m_len = NFS_MAXPATHLEN-len;
538 			len = NFS_MAXPATHLEN;
539 		} else
540 			len += mp->m_len;
541 		ivp->iov_base = mtod(mp, void *);
542 		ivp->iov_len = mp->m_len;
543 		i++;
544 		ivp++;
545 	}
546 	uiop->uio_iov = iv;
547 	uiop->uio_iovcnt = i;
548 	uiop->uio_offset = 0;
549 	uiop->uio_resid = len;
550 	uiop->uio_rw = UIO_READ;
551 	UIO_SETUP_SYSSPACE(uiop);
552 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
553 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
554 	if (error) {
555 		m_freem(mp3);
556 		nfsm_reply(2 * NFSX_UNSIGNED);
557 		nfsm_srvpostop_attr(1, (struct vattr *)0);
558 		return (0);
559 	}
560 	if (vp->v_type != VLNK) {
561 		if (v3)
562 			error = EINVAL;
563 		else
564 			error = ENXIO;
565 		goto out;
566 	}
567 	nqsrv_getl(vp, ND_READ);
568 	error = VOP_READLINK(vp, uiop, cred);
569 out:
570 	getret = VOP_GETATTR(vp, &attr, cred);
571 	vput(vp);
572 	if (error)
573 		m_freem(mp3);
574 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
575 	if (v3) {
576 		nfsm_srvpostop_attr(getret, &attr);
577 		if (error)
578 			return (0);
579 	}
580 	len -= uiop->uio_resid;
581 	padlen = nfsm_padlen(len);
582 	if (uiop->uio_resid || padlen)
583 		nfs_zeropad(mp3, uiop->uio_resid, padlen);
584 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
585 	*tl = txdr_unsigned(len);
586 	mb->m_next = mp3;
587 	nfsm_srvdone;
588 }
589 
590 /*
591  * nfs read service
592  */
593 int
594 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
595 {
596 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
597 	struct mbuf *nam = nfsd->nd_nam;
598 	char *dpos = nfsd->nd_dpos;
599 	kauth_cred_t cred = nfsd->nd_cr;
600 	struct mbuf *m;
601 	struct nfs_fattr *fp;
602 	u_int32_t *tl;
603 	int32_t t1;
604 	int i;
605 	char *bpos;
606 	int error = 0, rdonly, cache = 0, getret;
607 	int v3 = (nfsd->nd_flag & ND_NFSV3);
608 	uint32_t reqlen, len, cnt, left;
609 	int padlen;
610 	char *cp2;
611 	struct mbuf *mb, *mreq;
612 	struct vnode *vp;
613 	nfsrvfh_t nsfh;
614 	struct uio io, *uiop = &io;
615 	struct vattr va;
616 	off_t off;
617 	u_quad_t frev;
618 
619 	nfsm_srvmtofh(&nsfh);
620 	if (v3) {
621 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
622 		off = fxdr_hyper(tl);
623 	} else {
624 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
625 		off = (off_t)fxdr_unsigned(u_int32_t, *tl);
626 	}
627 	nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
628 	reqlen = fxdr_unsigned(uint32_t, *tl);
629 	reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd));
630 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
631 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
632 	if (error) {
633 		nfsm_reply(2 * NFSX_UNSIGNED);
634 		nfsm_srvpostop_attr(1, (struct vattr *)0);
635 		return (0);
636 	}
637 	if (vp->v_type != VREG) {
638 		if (v3)
639 			error = EINVAL;
640 		else
641 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
642 	}
643 	if (!error) {
644 	    nqsrv_getl(vp, ND_READ);
645 	    if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0)
646 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1);
647 	}
648 	getret = VOP_GETATTR(vp, &va, cred);
649 	if (!error)
650 		error = getret;
651 	if (error) {
652 		vput(vp);
653 		nfsm_reply(NFSX_POSTOPATTR(v3));
654 		nfsm_srvpostop_attr(getret, &va);
655 		return (0);
656 	}
657 	if (off >= va.va_size)
658 		cnt = 0;
659 	else if ((off + reqlen) > va.va_size)
660 		cnt = va.va_size - off;
661 	else
662 		cnt = reqlen;
663 	nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
664 	if (v3) {
665 		nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
666 		*tl++ = nfs_true;
667 		fp = (struct nfs_fattr *)tl;
668 		tl += (NFSX_V3FATTR / sizeof (u_int32_t));
669 	} else {
670 		nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
671 		fp = (struct nfs_fattr *)tl;
672 		tl += (NFSX_V2FATTR / sizeof (u_int32_t));
673 	}
674 	len = left = cnt;
675 	if (cnt > 0) {
676 		if (nfsd_use_loan) {
677 			struct vm_page **pgpp;
678 			voff_t pgoff = trunc_page(off);
679 			int npages;
680 			vaddr_t lva;
681 
682 			npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT;
683 			KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */
684 
685 			/* allocate kva for mbuf data */
686 			lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so);
687 			if (lva == 0) {
688 				/* fall back to VOP_READ */
689 				goto loan_fail;
690 			}
691 
692 			/* allocate mbuf */
693 			m = m_get(M_WAIT, MT_DATA);
694 			MCLAIM(m, &nfs_mowner);
695 			pgpp = m->m_ext.ext_pgs;
696 
697 			/* loan pages */
698 			error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages,
699 			    pgpp);
700 			if (error) {
701 				sokvafree(lva, npages << PAGE_SHIFT);
702 				m_free(m);
703 				if (error == EBUSY)
704 					goto loan_fail;
705 				goto read_error;
706 			}
707 
708 			/* associate kva to mbuf */
709 			MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)),
710 			    cnt, M_MBUF, soloanfree, slp->ns_so);
711 			m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
712 			m->m_len = cnt;
713 
714 			/* map pages */
715 			for (i = 0; i < npages; i++) {
716 				pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]),
717 				    VM_PROT_READ);
718 				lva += PAGE_SIZE;
719 			}
720 
721 			pmap_update(pmap_kernel());
722 
723 			mb->m_next = m;
724 			mb = m;
725 			error = 0;
726 			uiop->uio_resid = 0;
727 		} else {
728 			struct iovec *iv;
729 			struct iovec *iv2;
730 			struct mbuf *m2;
731 			int siz;
732 loan_fail:
733 			/*
734 			 * Generate the mbuf list with the uio_iov ref. to it.
735 			 */
736 			i = 0;
737 			m = m2 = mb;
738 			while (left > 0) {
739 				siz = min(M_TRAILINGSPACE(m), left);
740 				if (siz > 0) {
741 					left -= siz;
742 					i++;
743 				}
744 				if (left > 0) {
745 					m = m_get(M_WAIT, MT_DATA);
746 					MCLAIM(m, &nfs_mowner);
747 					m_clget(m, M_WAIT);
748 					m->m_len = 0;
749 					m2->m_next = m;
750 					m2 = m;
751 				}
752 			}
753 			iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK);
754 			uiop->uio_iov = iv2 = iv;
755 			m = mb;
756 			left = cnt;
757 			i = 0;
758 			while (left > 0) {
759 				if (m == NULL)
760 					panic("nfsrv_read iov");
761 				siz = min(M_TRAILINGSPACE(m), left);
762 				if (siz > 0) {
763 					iv->iov_base = mtod(m, char *) +
764 					    m->m_len;
765 					iv->iov_len = siz;
766 					m->m_len += siz;
767 					left -= siz;
768 					iv++;
769 					i++;
770 				}
771 				m = m->m_next;
772 			}
773 			uiop->uio_iovcnt = i;
774 			uiop->uio_offset = off;
775 			uiop->uio_resid = cnt;
776 			uiop->uio_rw = UIO_READ;
777 			UIO_SETUP_SYSSPACE(uiop);
778 			error = VOP_READ(vp, uiop, IO_NODELOCKED, cred);
779 			free((void *)iv2, M_TEMP);
780 		}
781 read_error:
782 		if (error || (getret = VOP_GETATTR(vp, &va, cred)) != 0){
783 			if (!error)
784 				error = getret;
785 			m_freem(mreq);
786 			vput(vp);
787 			nfsm_reply(NFSX_POSTOPATTR(v3));
788 			nfsm_srvpostop_attr(getret, &va);
789 			return (0);
790 		}
791 	} else {
792 		uiop->uio_resid = 0;
793 	}
794 	vput(vp);
795 	nfsm_srvfillattr(&va, fp);
796 	len -= uiop->uio_resid;
797 	padlen = nfsm_padlen(len);
798 	if (uiop->uio_resid || padlen)
799 		nfs_zeropad(mb, uiop->uio_resid, padlen);
800 	if (v3) {
801 		/* count */
802 		*tl++ = txdr_unsigned(len);
803 		/* eof */
804 		if (off + len >= va.va_size)
805 			*tl++ = nfs_true;
806 		else
807 			*tl++ = nfs_false;
808 	}
809 	*tl = txdr_unsigned(len);
810 	nfsm_srvdone;
811 }
812 
813 /*
814  * nfs write service
815  */
816 int
817 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
818 {
819 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
820 	struct mbuf *nam = nfsd->nd_nam;
821 	char *dpos = nfsd->nd_dpos;
822 	kauth_cred_t cred = nfsd->nd_cr;
823 	struct iovec *ivp;
824 	int i, cnt;
825 	struct mbuf *mp;
826 	struct nfs_fattr *fp;
827 	struct iovec *iv;
828 	struct vattr va, forat;
829 	u_int32_t *tl;
830 	int32_t t1;
831 	char *bpos;
832 	int error = 0, rdonly, cache = 0, len, forat_ret = 1;
833 	int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
834 	int stable = NFSV3WRITE_FILESYNC;
835 	int v3 = (nfsd->nd_flag & ND_NFSV3);
836 	char *cp2;
837 	struct mbuf *mb, *mreq;
838 	struct vnode *vp;
839 	nfsrvfh_t nsfh;
840 	struct uio io, *uiop = &io;
841 	off_t off;
842 	u_quad_t frev;
843 
844 	if (mrep == NULL) {
845 		*mrq = NULL;
846 		return (0);
847 	}
848 	nfsm_srvmtofh(&nsfh);
849 	if (v3) {
850 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
851 		off = fxdr_hyper(tl);
852 		tl += 3;
853 		stable = fxdr_unsigned(int, *tl++);
854 	} else {
855 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
856 		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
857 		tl += 2;
858 	}
859 	retlen = len = fxdr_unsigned(int32_t, *tl);
860 	cnt = i = 0;
861 
862 	/*
863 	 * For NFS Version 2, it is not obvious what a write of zero length
864 	 * should do, but I might as well be consistent with Version 3,
865 	 * which is to return ok so long as there are no permission problems.
866 	 */
867 	if (len > 0) {
868 		zeroing = 1;
869 		mp = mrep;
870 		while (mp) {
871 			if (mp == md) {
872 				zeroing = 0;
873 				adjust = dpos - mtod(mp, char *);
874 				mp->m_len -= adjust;
875 				if (mp->m_len > 0 && adjust > 0)
876 					NFSMADV(mp, adjust);
877 			}
878 			if (zeroing)
879 				mp->m_len = 0;
880 			else if (mp->m_len > 0) {
881 				i += mp->m_len;
882 				if (i > len) {
883 					mp->m_len -= (i - len);
884 					zeroing	= 1;
885 				}
886 				if (mp->m_len > 0)
887 					cnt++;
888 			}
889 			mp = mp->m_next;
890 		}
891 	}
892 	if (len > NFS_MAXDATA || len < 0 || i < len) {
893 		error = EIO;
894 		nfsm_reply(2 * NFSX_UNSIGNED);
895 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
896 		return (0);
897 	}
898 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
899 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
900 	if (error) {
901 		nfsm_reply(2 * NFSX_UNSIGNED);
902 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
903 		return (0);
904 	}
905 	if (v3)
906 		forat_ret = VOP_GETATTR(vp, &forat, cred);
907 	if (vp->v_type != VREG) {
908 		if (v3)
909 			error = EINVAL;
910 		else
911 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
912 	}
913 	if (!error) {
914 		nqsrv_getl(vp, ND_WRITE);
915 		error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1);
916 	}
917 	if (error) {
918 		vput(vp);
919 		nfsm_reply(NFSX_WCCDATA(v3));
920 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
921 		return (0);
922 	}
923 
924 	if (len > 0) {
925 		ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK);
926 		uiop->uio_iov = iv = ivp;
927 		uiop->uio_iovcnt = cnt;
928 		mp = mrep;
929 		while (mp) {
930 			if (mp->m_len > 0) {
931 				ivp->iov_base = mtod(mp, void *);
932 				ivp->iov_len = mp->m_len;
933 				ivp++;
934 			}
935 			mp = mp->m_next;
936 		}
937 
938 		/*
939 		 * XXX
940 		 * The IO_METASYNC flag indicates that all metadata (and not
941 		 * just enough to ensure data integrity) must be written to
942 		 * stable storage synchronously.
943 		 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
944 		 */
945 		if (stable == NFSV3WRITE_UNSTABLE)
946 			ioflags = IO_NODELOCKED;
947 		else if (stable == NFSV3WRITE_DATASYNC)
948 			ioflags = (IO_SYNC | IO_NODELOCKED);
949 		else
950 			ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
951 		uiop->uio_resid = len;
952 		uiop->uio_rw = UIO_WRITE;
953 		uiop->uio_offset = off;
954 		UIO_SETUP_SYSSPACE(uiop);
955 		error = VOP_WRITE(vp, uiop, ioflags, cred);
956 		nfsstats.srvvop_writes++;
957 		free(iv, M_TEMP);
958 	}
959 	aftat_ret = VOP_GETATTR(vp, &va, cred);
960 	vput(vp);
961 	if (!error)
962 		error = aftat_ret;
963 	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
964 		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
965 	if (v3) {
966 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
967 		if (error)
968 			return (0);
969 		nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
970 		*tl++ = txdr_unsigned(retlen);
971 		if (stable == NFSV3WRITE_UNSTABLE)
972 			*tl++ = txdr_unsigned(stable);
973 		else
974 			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
975 		/*
976 		 * Actually, there is no need to txdr these fields,
977 		 * but it may make the values more human readable,
978 		 * for debugging purposes.
979 		 */
980 		*tl++ = txdr_unsigned(boottime.tv_sec);
981 		*tl = txdr_unsigned(boottime.tv_nsec / 1000);
982 	} else {
983 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
984 		nfsm_srvfillattr(&va, fp);
985 	}
986 	nfsm_srvdone;
987 }
988 
989 /*
990  * XXX elad: the original NFSW_SAMECRED() macro also made sure the
991  *	     two nd_flag fields of the descriptors contained
992  *	     ND_KERBAUTH.
993  */
994 static int
995 nfsrv_samecred(kauth_cred_t cred1, kauth_cred_t cred2)
996 {
997 	int i, do_ngroups;
998 
999 	if (kauth_cred_geteuid(cred1) != kauth_cred_geteuid(cred2))
1000 		return (0);
1001 	if (kauth_cred_ngroups(cred1) != kauth_cred_ngroups(cred2))
1002 		return (0);
1003 	do_ngroups = kauth_cred_ngroups(cred1);
1004 	for (i = 0; i < do_ngroups; i++)
1005 		if (kauth_cred_group(cred1, i) !=
1006 		    kauth_cred_group(cred2, i))
1007 			return (0);
1008 
1009 	return (1);
1010 }
1011 
1012 static struct nfsrvw_delayhash *
1013 nfsrv_nwdelayhash(struct nfssvc_sock *slp, const nfsrvfh_t *nsfh)
1014 {
1015 	uint32_t hash;
1016 
1017 	hash = hash32_buf(NFSRVFH_DATA(nsfh), NFSRVFH_SIZE(nsfh),
1018 	    HASH32_BUF_INIT);
1019 	return &slp->ns_wdelayhashtbl[hash % NFS_WDELAYHASHSIZ];
1020 }
1021 
1022 /*
1023  * NFS write service with write gathering support. Called when
1024  * nfsrvw_procrastinate > 0.
1025  * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1026  * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1027  * Jan. 1994.
1028  */
1029 int
1030 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1031 {
1032 	struct timeval now;
1033 	struct iovec *ivp;
1034 	struct mbuf *mp;
1035 	struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1036 	struct nfs_fattr *fp;
1037 	int i = 0;
1038 	struct iovec *iov;
1039 	struct nfsrvw_delayhash *wpp;
1040 	kauth_cred_t cred;
1041 	struct vattr va, forat;
1042 	u_int32_t *tl;
1043 	int32_t t1;
1044 	char *bpos, *dpos;
1045 	int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1;
1046 	int ioflags, aftat_ret = 1, adjust, v3, zeroing;
1047 	char *cp2;
1048 	struct mbuf *mb, *mreq, *mrep, *md;
1049 	struct vnode *vp;
1050 	struct uio io, *uiop = &io;
1051 	u_quad_t frev, cur_usec;
1052 
1053 	*mrq = NULL;
1054 	if (*ndp) {
1055 	    nfsd = *ndp;
1056 	    *ndp = NULL;
1057 	    mrep = nfsd->nd_mrep;
1058 	    md = nfsd->nd_md;
1059 	    dpos = nfsd->nd_dpos;
1060 	    cred = nfsd->nd_cr;
1061 	    v3 = (nfsd->nd_flag & ND_NFSV3);
1062 	    LIST_INIT(&nfsd->nd_coalesce);
1063 	    nfsd->nd_mreq = NULL;
1064 	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1065 	    getmicrotime(&now);
1066 	    cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec;
1067 	    nfsd->nd_time = cur_usec + nfsrvw_procrastinate;
1068 
1069 	    /*
1070 	     * Now, get the write header..
1071 	     */
1072 	    nfsm_srvmtofh(&nfsd->nd_fh);
1073 	    if (v3) {
1074 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1075 		nfsd->nd_off = fxdr_hyper(tl);
1076 		tl += 3;
1077 		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1078 	    } else {
1079 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1080 		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1081 		tl += 2;
1082 	    }
1083 	    len = fxdr_unsigned(int32_t, *tl);
1084 	    nfsd->nd_len = len;
1085 	    nfsd->nd_eoff = nfsd->nd_off + len;
1086 
1087 	    /*
1088 	     * Trim the header out of the mbuf list and trim off any trailing
1089 	     * junk so that the mbuf list has only the write data.
1090 	     */
1091 	    zeroing = 1;
1092 	    i = 0;
1093 	    mp = mrep;
1094 	    while (mp) {
1095 		if (mp == md) {
1096 		    zeroing = 0;
1097 		    adjust = dpos - mtod(mp, char *);
1098 		    mp->m_len -= adjust;
1099 		    if (mp->m_len > 0 && adjust > 0)
1100 			NFSMADV(mp, adjust);
1101 		}
1102 		if (zeroing)
1103 		    mp->m_len = 0;
1104 		else {
1105 		    i += mp->m_len;
1106 		    if (i > len) {
1107 			mp->m_len -= (i - len);
1108 			zeroing = 1;
1109 		    }
1110 		}
1111 		mp = mp->m_next;
1112 	    }
1113 	    if (len > NFS_MAXDATA || len < 0  || i < len) {
1114 nfsmout:
1115 		m_freem(mrep);
1116 		error = EIO;
1117 		nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1118 		if (v3)
1119 		    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1120 		nfsd->nd_mreq = mreq;
1121 		nfsd->nd_mrep = NULL;
1122 		nfsd->nd_time = 0;
1123 	    }
1124 
1125 	    /*
1126 	     * Add this entry to the hash and time queues.
1127 	     */
1128 	    owp = NULL;
1129 	    mutex_enter(&nfsd_lock);
1130 	    wp = LIST_FIRST(&slp->ns_tq);
1131 	    while (wp && wp->nd_time < nfsd->nd_time) {
1132 		owp = wp;
1133 		wp = LIST_NEXT(wp, nd_tq);
1134 	    }
1135 	    if (owp) {
1136 		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1137 	    } else {
1138 		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1139 	    }
1140 	    if (nfsd->nd_mrep) {
1141 		wpp = nfsrv_nwdelayhash(slp, &nfsd->nd_fh);
1142 		owp = NULL;
1143 		wp = LIST_FIRST(wpp);
1144 		while (wp && nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) {
1145 		    owp = wp;
1146 		    wp = LIST_NEXT(wp, nd_hash);
1147 		}
1148 		while (wp && wp->nd_off < nfsd->nd_off &&
1149 		    !nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) {
1150 		    owp = wp;
1151 		    wp = LIST_NEXT(wp, nd_hash);
1152 		}
1153 		if (owp) {
1154 		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1155 
1156 		    /*
1157 		     * Search the hash list for overlapping entries and
1158 		     * coalesce.
1159 		     */
1160 		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1161 			wp = LIST_NEXT(nfsd, nd_hash);
1162 			if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr))
1163 			    nfsrvw_coalesce(owp, nfsd);
1164 		    }
1165 		} else {
1166 		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1167 		}
1168 	    }
1169 	    mutex_exit(&nfsd_lock);
1170 	}
1171 
1172 	/*
1173 	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1174 	 * and generate the associated reply mbuf list(s).
1175 	 */
1176 loop1:
1177 	getmicrotime(&now);
1178 	cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec;
1179 	mutex_enter(&nfsd_lock);
1180 	for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1181 		owp = LIST_NEXT(nfsd, nd_tq);
1182 		if (nfsd->nd_time > cur_usec)
1183 		    break;
1184 		if (nfsd->nd_mreq)
1185 		    continue;
1186 		LIST_REMOVE(nfsd, nd_tq);
1187 		LIST_REMOVE(nfsd, nd_hash);
1188 		mutex_exit(&nfsd_lock);
1189 
1190 		mrep = nfsd->nd_mrep;
1191 		nfsd->nd_mrep = NULL;
1192 		cred = nfsd->nd_cr;
1193 		v3 = (nfsd->nd_flag & ND_NFSV3);
1194 		forat_ret = aftat_ret = 1;
1195 		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1196 		    nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH),
1197 		    false);
1198 		if (!error) {
1199 		    if (v3)
1200 			forat_ret = VOP_GETATTR(vp, &forat, cred);
1201 		    if (vp->v_type != VREG) {
1202 			if (v3)
1203 			    error = EINVAL;
1204 			else
1205 			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1206 		    }
1207 		} else
1208 		    vp = NULL;
1209 		if (!error) {
1210 		    nqsrv_getl(vp, ND_WRITE);
1211 		    error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1);
1212 		}
1213 
1214 		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1215 		    ioflags = IO_NODELOCKED;
1216 		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1217 		    ioflags = (IO_SYNC | IO_NODELOCKED);
1218 		else
1219 		    ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1220 		uiop->uio_rw = UIO_WRITE;
1221 		uiop->uio_offset = nfsd->nd_off;
1222 		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1223 		UIO_SETUP_SYSSPACE(uiop);
1224 		if (uiop->uio_resid > 0) {
1225 		    mp = mrep;
1226 		    i = 0;
1227 		    while (mp) {
1228 			if (mp->m_len > 0)
1229 			    i++;
1230 			mp = mp->m_next;
1231 		    }
1232 		    uiop->uio_iovcnt = i;
1233 		    iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1234 		    uiop->uio_iov = ivp = iov;
1235 		    mp = mrep;
1236 		    while (mp) {
1237 			if (mp->m_len > 0) {
1238 			    ivp->iov_base = mtod(mp, void *);
1239 			    ivp->iov_len = mp->m_len;
1240 			    ivp++;
1241 			}
1242 			mp = mp->m_next;
1243 		    }
1244 		    if (!error) {
1245 			error = VOP_WRITE(vp, uiop, ioflags, cred);
1246 			nfsstats.srvvop_writes++;
1247 		    }
1248 		    free((void *)iov, M_TEMP);
1249 		}
1250 		m_freem(mrep);
1251 		if (vp) {
1252 		    aftat_ret = VOP_GETATTR(vp, &va, cred);
1253 		    vput(vp);
1254 		}
1255 
1256 		/*
1257 		 * Loop around generating replies for all write rpcs that have
1258 		 * now been completed.
1259 		 */
1260 		swp = nfsd;
1261 		do {
1262 		    if (error) {
1263 			nfsm_writereply(NFSX_WCCDATA(v3), v3);
1264 			if (v3) {
1265 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1266 			}
1267 		    } else {
1268 			nfsm_writereply(NFSX_PREOPATTR(v3) +
1269 			    NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1270 			    NFSX_WRITEVERF(v3), v3);
1271 			if (v3) {
1272 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1273 			    nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1274 			    *tl++ = txdr_unsigned(nfsd->nd_len);
1275 			    *tl++ = txdr_unsigned(swp->nd_stable);
1276 			    /*
1277 			     * Actually, there is no need to txdr these fields,
1278 			     * but it may make the values more human readable,
1279 			     * for debugging purposes.
1280 			     */
1281 			    *tl++ = txdr_unsigned(boottime.tv_sec);
1282 			    *tl = txdr_unsigned(boottime.tv_nsec / 1000);
1283 			} else {
1284 			    nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1285 			    nfsm_srvfillattr(&va, fp);
1286 			}
1287 		    }
1288 		    nfsd->nd_mreq = mreq;
1289 		    if (nfsd->nd_mrep)
1290 			panic("nfsrv_write: nd_mrep not free");
1291 
1292 		    /*
1293 		     * Done. Put it at the head of the timer queue so that
1294 		     * the final phase can return the reply.
1295 		     */
1296 		    mutex_enter(&nfsd_lock);
1297 		    if (nfsd != swp) {
1298 			nfsd->nd_time = 0;
1299 			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1300 		    }
1301 		    nfsd = LIST_FIRST(&swp->nd_coalesce);
1302 		    if (nfsd) {
1303 			LIST_REMOVE(nfsd, nd_tq);
1304 		    }
1305 		    mutex_exit(&nfsd_lock);
1306 		} while (nfsd);
1307 		swp->nd_time = 0;
1308 
1309 		mutex_enter(&nfsd_lock);
1310 		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1311 		mutex_exit(&nfsd_lock);
1312 		goto loop1;
1313 	}
1314 	mutex_exit(&nfsd_lock);
1315 	nfs_timer_start();
1316 
1317 	/*
1318 	 * Search for a reply to return.
1319 	 */
1320 	mutex_enter(&nfsd_lock);
1321 	LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) {
1322 		if (nfsd->nd_mreq) {
1323 		    LIST_REMOVE(nfsd, nd_tq);
1324 		    *mrq = nfsd->nd_mreq;
1325 		    *ndp = nfsd;
1326 		    break;
1327 		}
1328 	}
1329 	mutex_exit(&nfsd_lock);
1330 	return (0);
1331 }
1332 
1333 /*
1334  * Coalesce the write request nfsd into owp. To do this we must:
1335  * - remove nfsd from the queues
1336  * - merge nfsd->nd_mrep into owp->nd_mrep
1337  * - update the nd_eoff and nd_stable for owp
1338  * - put nfsd on owp's nd_coalesce list
1339  * NB: Must be called at splsoftclock().
1340  */
1341 void
1342 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1343 {
1344         int overlap;
1345         struct mbuf *mp;
1346 	struct nfsrv_descript *m;
1347 
1348 	KASSERT(mutex_owned(&nfsd_lock));
1349 
1350         LIST_REMOVE(nfsd, nd_hash);
1351         LIST_REMOVE(nfsd, nd_tq);
1352         if (owp->nd_eoff < nfsd->nd_eoff) {
1353             overlap = owp->nd_eoff - nfsd->nd_off;
1354             if (overlap < 0)
1355                 panic("nfsrv_coalesce: bad off");
1356             if (overlap > 0)
1357                 m_adj(nfsd->nd_mrep, overlap);
1358             mp = owp->nd_mrep;
1359             while (mp->m_next)
1360                 mp = mp->m_next;
1361             mp->m_next = nfsd->nd_mrep;
1362             owp->nd_eoff = nfsd->nd_eoff;
1363         } else
1364             m_freem(nfsd->nd_mrep);
1365         nfsd->nd_mrep = NULL;
1366         if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1367             owp->nd_stable = NFSV3WRITE_FILESYNC;
1368         else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1369             owp->nd_stable == NFSV3WRITE_UNSTABLE)
1370             owp->nd_stable = NFSV3WRITE_DATASYNC;
1371         LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1372  	/*
1373  	 * nfsd might hold coalesce elements! Move them to owp.
1374  	 * Otherwise, requests may be lost and clients will be stuck.
1375  	 */
1376 	while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) {
1377 		LIST_REMOVE(m, nd_tq);
1378 		LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
1379 	}
1380 }
1381 
1382 /*
1383  * nfs create service
1384  * now does a truncate to 0 length via. setattr if it already exists
1385  */
1386 int
1387 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1388 {
1389 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1390 	struct mbuf *nam = nfsd->nd_nam;
1391 	char *dpos = nfsd->nd_dpos;
1392 	kauth_cred_t cred = nfsd->nd_cr;
1393 	struct nfs_fattr *fp;
1394 	struct vattr va, dirfor, diraft;
1395 	struct nfsv2_sattr *sp;
1396 	u_int32_t *tl;
1397 	struct nameidata nd;
1398 	char *cp;
1399 	int32_t t1;
1400 	char *bpos;
1401 	int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1402 	int rdev = 0, abort = 0;
1403 	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1404 	char *cp2;
1405 	struct mbuf *mb, *mreq;
1406 	struct vnode *vp = NULL, *dirp = NULL;
1407 	nfsrvfh_t nsfh;
1408 	u_quad_t frev, tempsize;
1409 	u_char cverf[NFSX_V3CREATEVERF];
1410 
1411 	nd.ni_cnd.cn_nameiop = 0;
1412 	nfsm_srvmtofh(&nsfh);
1413 	nfsm_srvnamesiz(len);
1414 	nd.ni_cnd.cn_cred = cred;
1415 	nd.ni_cnd.cn_nameiop = CREATE;
1416 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1417 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1418 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1419 	if (dirp && v3) {
1420 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1421 	}
1422 	if (error) {
1423 		nfsm_reply(NFSX_WCCDATA(v3));
1424 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1425 		if (dirp)
1426 			vrele(dirp);
1427 		return (0);
1428 	}
1429 	abort = 1;
1430 	VATTR_NULL(&va);
1431 	if (v3) {
1432 		va.va_mode = 0;
1433 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1434 		how = fxdr_unsigned(int, *tl);
1435 		switch (how) {
1436 		case NFSV3CREATE_GUARDED:
1437 			if (nd.ni_vp) {
1438 				error = EEXIST;
1439 				break;
1440 			}
1441 		case NFSV3CREATE_UNCHECKED:
1442 			nfsm_srvsattr(&va);
1443 			break;
1444 		case NFSV3CREATE_EXCLUSIVE:
1445 			nfsm_dissect(cp, void *, NFSX_V3CREATEVERF);
1446 			memcpy(cverf, cp, NFSX_V3CREATEVERF);
1447 			exclusive_flag = 1;
1448 			break;
1449 		};
1450 		va.va_type = VREG;
1451 	} else {
1452 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1453 		va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1454 		if (va.va_type == VNON)
1455 			va.va_type = VREG;
1456 		va.va_mode = nfstov_mode(sp->sa_mode);
1457 		switch (va.va_type) {
1458 		case VREG:
1459 			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1460 			if (tsize != -1)
1461 				va.va_size = (u_quad_t)tsize;
1462 			break;
1463 		case VCHR:
1464 		case VBLK:
1465 		case VFIFO:
1466 			rdev = fxdr_unsigned(int32_t, sp->sa_size);
1467 			break;
1468 		default:
1469 			break;
1470 		};
1471 	}
1472 
1473 	/*
1474 	 * Iff doesn't exist, create it
1475 	 * otherwise just truncate to 0 length
1476 	 *   should I set the mode too ??
1477 	 */
1478 	if (nd.ni_vp == NULL) {
1479 		if (va.va_type == VREG || va.va_type == VSOCK) {
1480 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1481 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1482 			if (!error) {
1483 				if (exclusive_flag) {
1484 					exclusive_flag = 0;
1485 					VATTR_NULL(&va);
1486 					/*
1487 					 * XXX
1488 					 * assuming NFSX_V3CREATEVERF
1489 					 * == sizeof(nfstime3)
1490 					 */
1491 					fxdr_nfsv3time(cverf, &va.va_atime);
1492 					error = VOP_SETATTR(nd.ni_vp, &va,
1493 						cred);
1494 				}
1495 			}
1496 		} else if (va.va_type == VCHR || va.va_type == VBLK ||
1497 			va.va_type == VFIFO) {
1498 			if (va.va_type == VCHR && rdev == 0xffffffff)
1499 				va.va_type = VFIFO;
1500 			if (va.va_type != VFIFO &&
1501 			    (error = kauth_authorize_system(cred,
1502 			    KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) {
1503 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1504 				vput(nd.ni_dvp);
1505 				abort = 0;
1506 				nfsm_reply(0);
1507 				return (error);
1508 			} else
1509 				va.va_rdev = (dev_t)rdev;
1510 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1511 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd,
1512 			    &va);
1513 			if (error) {
1514 				nfsm_reply(0);
1515 			}
1516 			if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1517 				vput(nd.ni_vp);
1518 				vrele(nd.ni_dvp);
1519 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1520 				error = EINVAL;
1521 				abort = 0;
1522 				nfsm_reply(0);
1523 			}
1524 		} else {
1525 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1526 			vput(nd.ni_dvp);
1527 			error = ENXIO;
1528 			abort = 0;
1529 		}
1530 		vp = nd.ni_vp;
1531 	} else {
1532 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1533 		vp = nd.ni_vp;
1534 		if (nd.ni_dvp == vp)
1535 			vrele(nd.ni_dvp);
1536 		else
1537 			vput(nd.ni_dvp);
1538 		abort = 0;
1539 		if (!error && va.va_size != -1) {
1540 			error = nfsrv_access(vp, VWRITE, cred,
1541 			    (nd.ni_cnd.cn_flags & RDONLY), lwp, 0);
1542 			if (!error) {
1543 				nqsrv_getl(vp, ND_WRITE);
1544 				tempsize = va.va_size;
1545 				VATTR_NULL(&va);
1546 				va.va_size = tempsize;
1547 				error = VOP_SETATTR(vp, &va, cred);
1548 			}
1549 		}
1550 		if (error)
1551 			vput(vp);
1552 	}
1553 	if (!error) {
1554 		error = nfsrv_composefh(vp, &nsfh, v3);
1555 		if (!error)
1556 			error = VOP_GETATTR(vp, &va, cred);
1557 		vput(vp);
1558 	}
1559 	if (v3) {
1560 		if (exclusive_flag && !error) {
1561 			/*
1562 			 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3)
1563 			 */
1564 			char oldverf[NFSX_V3CREATEVERF];
1565 
1566 			txdr_nfsv3time(&va.va_atime, oldverf);
1567 			if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF))
1568 				error = EEXIST;
1569 		}
1570 		if (dirp) {
1571 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1572 		}
1573 	}
1574 	if (dirp) {
1575 		vrele(dirp);
1576 		dirp = NULL;
1577 	}
1578 	abort = 0;
1579 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1580 	if (v3) {
1581 		if (!error) {
1582 			nfsm_srvpostop_fh(&nsfh);
1583 			nfsm_srvpostop_attr(0, &va);
1584 		}
1585 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1586 	} else {
1587 		nfsm_srvfhtom(&nsfh, v3);
1588 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1589 		nfsm_srvfillattr(&va, fp);
1590 	}
1591 	return (0);
1592 nfsmout:
1593 	if (dirp)
1594 		vrele(dirp);
1595 	if (abort) {
1596 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1597 		if (nd.ni_dvp == nd.ni_vp)
1598 			vrele(nd.ni_dvp);
1599 		else
1600 			vput(nd.ni_dvp);
1601 		if (nd.ni_vp)
1602 			vput(nd.ni_vp);
1603 	}
1604 	return (error);
1605 }
1606 
1607 /*
1608  * nfs v3 mknod service
1609  */
1610 int
1611 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1612 {
1613 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1614 	struct mbuf *nam = nfsd->nd_nam;
1615 	char *dpos = nfsd->nd_dpos;
1616 	kauth_cred_t cred = nfsd->nd_cr;
1617 	struct vattr va, dirfor, diraft;
1618 	u_int32_t *tl;
1619 	struct nameidata nd;
1620 	int32_t t1;
1621 	char *bpos;
1622 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
1623 	int abort = 0;
1624 	u_int32_t major, minor;
1625 	enum vtype vtyp;
1626 	char *cp2;
1627 	struct mbuf *mb, *mreq;
1628 	struct vnode *vp, *dirp = (struct vnode *)0;
1629 	nfsrvfh_t nsfh;
1630 	u_quad_t frev;
1631 
1632 	nd.ni_cnd.cn_nameiop = 0;
1633 	nfsm_srvmtofh(&nsfh);
1634 	nfsm_srvnamesiz(len);
1635 	nd.ni_cnd.cn_cred = cred;
1636 	nd.ni_cnd.cn_nameiop = CREATE;
1637 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1638 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1639 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1640 	if (dirp)
1641 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1642 	if (error) {
1643 		nfsm_reply(NFSX_WCCDATA(1));
1644 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1645 		if (dirp)
1646 			vrele(dirp);
1647 		return (0);
1648 	}
1649 	abort = 1;
1650 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1651 	vtyp = nfsv3tov_type(*tl);
1652 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1653 		error = NFSERR_BADTYPE;
1654 		goto abort;
1655 	}
1656 	VATTR_NULL(&va);
1657 	va.va_mode = 0;
1658 	nfsm_srvsattr(&va);
1659 	if (vtyp == VCHR || vtyp == VBLK) {
1660 		dev_t rdev;
1661 
1662 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1663 		major = fxdr_unsigned(u_int32_t, *tl++);
1664 		minor = fxdr_unsigned(u_int32_t, *tl);
1665 		rdev = makedev(major, minor);
1666 		if (major(rdev) != major || minor(rdev) != minor) {
1667 			error = EINVAL;
1668 			goto abort;
1669 		}
1670 		va.va_rdev = rdev;
1671 	}
1672 
1673 	/*
1674 	 * Iff doesn't exist, create it.
1675 	 */
1676 	if (nd.ni_vp) {
1677 		error = EEXIST;
1678 abort:
1679 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1680 		if (nd.ni_dvp == nd.ni_vp)
1681 			vrele(nd.ni_dvp);
1682 		else
1683 			vput(nd.ni_dvp);
1684 		if (nd.ni_vp)
1685 			vput(nd.ni_vp);
1686 		goto out;
1687 	}
1688 	va.va_type = vtyp;
1689 	if (vtyp == VSOCK) {
1690 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1691 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1692 	} else {
1693 		if (va.va_type != VFIFO &&
1694 		    (error = kauth_authorize_system(cred,
1695 		    KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) {
1696 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1697 			vput(nd.ni_dvp);
1698 			goto out;
1699 		}
1700 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1701 		error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1702 		if (error)
1703 			goto out;
1704 		if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1705 			vput(nd.ni_vp);
1706 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1707 			error = EINVAL;
1708 		}
1709 	}
1710 out:
1711 	vp = nd.ni_vp;
1712 	if (!error) {
1713 		error = nfsrv_composefh(vp, &nsfh, true);
1714 		if (!error)
1715 			error = VOP_GETATTR(vp, &va, cred);
1716 		vput(vp);
1717 	}
1718 	if (dirp) {
1719 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1720 		vrele(dirp);
1721 		dirp = NULL;
1722 	}
1723 	abort = 0;
1724 	nfsm_reply(NFSX_SRVFH(&nsfh, true) + NFSX_POSTOPATTR(1) +
1725 	    NFSX_WCCDATA(1));
1726 	if (!error) {
1727 		nfsm_srvpostop_fh(&nsfh);
1728 		nfsm_srvpostop_attr(0, &va);
1729 	}
1730 	nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1731 	return (0);
1732 nfsmout:
1733 	if (abort) {
1734 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1735 		if (nd.ni_dvp == nd.ni_vp)
1736 			vrele(nd.ni_dvp);
1737 		else
1738 			vput(nd.ni_dvp);
1739 		if (nd.ni_vp)
1740 			vput(nd.ni_vp);
1741 	}
1742 	if (dirp)
1743 		vrele(dirp);
1744 	return (error);
1745 }
1746 
1747 /*
1748  * nfs remove service
1749  */
1750 int
1751 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1752 {
1753 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1754 	struct mbuf *nam = nfsd->nd_nam;
1755 	char *dpos = nfsd->nd_dpos;
1756 	kauth_cred_t cred = nfsd->nd_cr;
1757 	struct nameidata nd;
1758 	u_int32_t *tl;
1759 	int32_t t1;
1760 	char *bpos;
1761 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
1762 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1763 	char *cp2;
1764 	struct mbuf *mb, *mreq;
1765 	struct vnode *vp, *dirp;
1766 	struct vattr dirfor, diraft;
1767 	nfsrvfh_t nsfh;
1768 	u_quad_t frev;
1769 
1770 #ifndef nolint
1771 	vp = (struct vnode *)0;
1772 #endif
1773 	nfsm_srvmtofh(&nsfh);
1774 	nfsm_srvnamesiz(len);
1775 	nd.ni_cnd.cn_cred = cred;
1776 	nd.ni_cnd.cn_nameiop = DELETE;
1777 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1778 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1779 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1780 	if (dirp && v3) {
1781 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1782 	}
1783 	if (!error) {
1784 		vp = nd.ni_vp;
1785 		if (vp->v_type == VDIR) {
1786 			error = EPERM;
1787 			goto out;
1788 		}
1789 		/*
1790 		 * The root of a mounted filesystem cannot be deleted.
1791 		 */
1792 		if (vp->v_vflag & VV_ROOT) {
1793 			error = EBUSY;
1794 		}
1795 out:
1796 		if (!error) {
1797 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1798 			nqsrv_getl(vp, ND_WRITE);
1799 			error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1800 		} else {
1801 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1802 			if (nd.ni_dvp == vp)
1803 				vrele(nd.ni_dvp);
1804 			else
1805 				vput(nd.ni_dvp);
1806 			vput(vp);
1807 		}
1808 	}
1809 	if (dirp) {
1810 		if (v3) {
1811 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1812 		}
1813 		vrele(dirp);
1814 	}
1815 	nfsm_reply(NFSX_WCCDATA(v3));
1816 	if (v3) {
1817 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1818 		return (0);
1819 	}
1820 	nfsm_srvdone;
1821 }
1822 
1823 /*
1824  * nfs rename service
1825  */
1826 int
1827 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1828 {
1829 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1830 	struct mbuf *nam = nfsd->nd_nam;
1831 	char *dpos = nfsd->nd_dpos;
1832 	kauth_cred_t cred = nfsd->nd_cr;
1833 	u_int32_t *tl;
1834 	int32_t t1;
1835 	char *bpos;
1836 	int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1;
1837 	uint32_t len, len2;
1838 	int tdirfor_ret = 1, tdiraft_ret = 1;
1839 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1840 	char *cp2;
1841 	struct mbuf *mb, *mreq;
1842 	struct nameidata fromnd, tond;
1843 	struct vnode *fvp, *tvp, *tdvp;
1844 	struct vnode *fdirp = NULL, *tdirp = NULL;
1845 	struct mount *localfs = NULL;
1846 	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1847 	nfsrvfh_t fnsfh, tnsfh;
1848 	u_quad_t frev;
1849 	uid_t saved_uid;
1850 	uint32_t saveflag;
1851 
1852 #ifndef nolint
1853 	fvp = (struct vnode *)0;
1854 #endif
1855 	fromnd.ni_cnd.cn_nameiop = 0;
1856 	tond.ni_cnd.cn_nameiop = 0;
1857 	nfsm_srvmtofh(&fnsfh);
1858 	nfsm_srvnamesiz(len);
1859 	/*
1860 	 * Remember our original uid so that we can reset cr_uid before
1861 	 * the second nfs_namei() call, in case it is remapped.
1862 	 */
1863 	saved_uid = kauth_cred_geteuid(cred);
1864 	fromnd.ni_cnd.cn_cred = cred;
1865 	fromnd.ni_cnd.cn_nameiop = DELETE;
1866 	fromnd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART;
1867 	error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md,
1868 		&dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1869 	if (fdirp && v3) {
1870 		fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred);
1871 	}
1872 	if (error) {
1873 		nfsm_reply(2 * NFSX_WCCDATA(v3));
1874 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1875 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1876 		if (fdirp)
1877 			vrele(fdirp);
1878 		return (0);
1879 	}
1880 	if (fromnd.ni_dvp != fromnd.ni_vp) {
1881 		VOP_UNLOCK(fromnd.ni_dvp, 0);
1882 	}
1883 	fvp = fromnd.ni_vp;
1884 
1885 	localfs = fvp->v_mount;
1886 	error = VFS_RENAMELOCK_ENTER(localfs);
1887 	if (error) {
1888 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1889 		vrele(fromnd.ni_dvp);
1890 		vrele(fvp);
1891 		goto out1;
1892 	}
1893 
1894 	/* Copied, regrettably, from vfs_syscalls.c (q.v.) */
1895 	vrele(fvp);
1896 	if ((fromnd.ni_cnd.cn_namelen == 1 &&
1897 	     fromnd.ni_cnd.cn_nameptr[0] == '.') ||
1898 	    (fromnd.ni_cnd.cn_namelen == 2 &&
1899 	     fromnd.ni_cnd.cn_nameptr[0] == '.' &&
1900 	     fromnd.ni_cnd.cn_nameptr[1] == '.')) {
1901 		error = EINVAL;
1902 		VFS_RENAMELOCK_EXIT(localfs);
1903 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1904 		vrele(fromnd.ni_dvp);
1905 		goto out1;
1906 	}
1907 	saveflag = fromnd.ni_cnd.cn_flags & SAVESTART;
1908 	fromnd.ni_cnd.cn_flags &= ~SAVESTART;
1909 	vn_lock(fromnd.ni_dvp, LK_EXCLUSIVE | LK_RETRY);
1910 	error = relookup(fromnd.ni_dvp, &fromnd.ni_vp, &fromnd.ni_cnd);
1911 	fromnd.ni_cnd.cn_flags |= saveflag;
1912 	if (error) {
1913 		VOP_UNLOCK(fromnd.ni_dvp, 0);
1914 		VFS_RENAMELOCK_EXIT(localfs);
1915 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1916 		vrele(fromnd.ni_dvp);
1917 		goto out1;
1918 	}
1919 	VOP_UNLOCK(fromnd.ni_vp, 0);
1920 	if (fromnd.ni_dvp != fromnd.ni_vp)
1921 		VOP_UNLOCK(fromnd.ni_dvp, 0);
1922 	fvp = fromnd.ni_vp;
1923 
1924 	nfsm_srvmtofh(&tnsfh);
1925 	if (v3) {
1926 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1927 		len2 = fxdr_unsigned(uint32_t, *tl);
1928 		/* len2 will be checked by nfs_namei */
1929 	}
1930 	else {
1931 		/* NFSv2 */
1932 		nfsm_strsiz(len2, NFS_MAXNAMLEN);
1933 	}
1934 	kauth_cred_seteuid(cred, saved_uid);
1935 	tond.ni_cnd.cn_cred = cred;
1936 	tond.ni_cnd.cn_nameiop = RENAME;
1937 	tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART;
1938 	error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md,
1939 		&dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1940 	if (tdirp && v3) {
1941 		tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred);
1942 	}
1943 	if (error) {
1944 		VFS_RENAMELOCK_EXIT(localfs);
1945 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1946 		vrele(fromnd.ni_dvp);
1947 		vrele(fvp);
1948 		goto out1;
1949 	}
1950 	tdvp = tond.ni_dvp;
1951 	tvp = tond.ni_vp;
1952 	if (tvp != NULL) {
1953 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1954 			if (v3)
1955 				error = EEXIST;
1956 			else
1957 				error = EISDIR;
1958 			goto out;
1959 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1960 			if (v3)
1961 				error = EEXIST;
1962 			else
1963 				error = ENOTDIR;
1964 			goto out;
1965 		}
1966 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1967 			if (v3)
1968 				error = EXDEV;
1969 			else
1970 				error = ENOTEMPTY;
1971 			goto out;
1972 		}
1973 	}
1974 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1975 		if (v3)
1976 			error = EXDEV;
1977 		else
1978 			error = ENOTEMPTY;
1979 		goto out;
1980 	}
1981 	if (fvp->v_mount != tdvp->v_mount) {
1982 		if (v3)
1983 			error = EXDEV;
1984 		else
1985 			error = ENOTEMPTY;
1986 		goto out;
1987 	}
1988 	if (fvp == tdvp) {
1989 		if (v3)
1990 			error = EINVAL;
1991 		else
1992 			error = ENOTEMPTY;
1993 	}
1994 	/*
1995 	 * If source is the same as the destination (that is the
1996 	 * same vnode with the same name in the same directory),
1997 	 * then there is nothing to do.
1998 	 */
1999 	if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2000 	    fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2001 	    !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2002 	      fromnd.ni_cnd.cn_namelen))
2003 		error = -1;
2004 out:
2005 	if (!error) {
2006 		nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
2007 		nqsrv_getl(tdvp, ND_WRITE);
2008 		if (tvp) {
2009 			nqsrv_getl(tvp, ND_WRITE);
2010 		}
2011 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2012 				   tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2013 		VFS_RENAMELOCK_EXIT(localfs);
2014 	} else {
2015 		VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd);
2016 		if (tdvp == tvp)
2017 			vrele(tdvp);
2018 		else
2019 			vput(tdvp);
2020 		if (tvp)
2021 			vput(tvp);
2022 		VFS_RENAMELOCK_EXIT(localfs);
2023 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2024 		vrele(fromnd.ni_dvp);
2025 		vrele(fvp);
2026 		if (error == -1)
2027 			error = 0;
2028 	}
2029 	vrele(tond.ni_startdir);
2030 	PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
2031 	tond.ni_cnd.cn_nameiop = 0;
2032 out1:
2033 	if (fdirp) {
2034 		if (v3) {
2035 			fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred);
2036 		}
2037 		vrele(fdirp);
2038 		fdirp = NULL;
2039 	}
2040 	if (tdirp) {
2041 		if (v3) {
2042 			tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred);
2043 		}
2044 		vrele(tdirp);
2045 		tdirp = NULL;
2046 	}
2047 	vrele(fromnd.ni_startdir);
2048 	PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
2049 	fromnd.ni_cnd.cn_nameiop = 0;
2050 	localfs = NULL;
2051 	nfsm_reply(2 * NFSX_WCCDATA(v3));
2052 	if (v3) {
2053 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2054 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2055 	}
2056 	return (0);
2057 
2058 nfsmout:
2059 	if (fdirp)
2060 		vrele(fdirp);
2061 #ifdef notdef
2062 	if (tdirp)
2063 		vrele(tdirp);
2064 #endif
2065 	if (tond.ni_cnd.cn_nameiop) {
2066 		vrele(tond.ni_startdir);
2067 		PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
2068 	}
2069 	if (localfs) {
2070 		VFS_RENAMELOCK_EXIT(localfs);
2071 	}
2072 	if (fromnd.ni_cnd.cn_nameiop) {
2073 		vrele(fromnd.ni_startdir);
2074 		PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
2075 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2076 		vrele(fromnd.ni_dvp);
2077 		vrele(fvp);
2078 	}
2079 	return (error);
2080 }
2081 
2082 /*
2083  * nfs link service
2084  */
2085 int
2086 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2087 {
2088 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2089 	struct mbuf *nam = nfsd->nd_nam;
2090 	char *dpos = nfsd->nd_dpos;
2091 	kauth_cred_t cred = nfsd->nd_cr;
2092 	struct nameidata nd;
2093 	u_int32_t *tl;
2094 	int32_t t1;
2095 	char *bpos;
2096 	int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2097 	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2098 	char *cp2;
2099 	struct mbuf *mb, *mreq;
2100 	struct vnode *vp, *xp, *dirp = (struct vnode *)0;
2101 	struct vattr dirfor, diraft, at;
2102 	nfsrvfh_t nsfh, dnsfh;
2103 	u_quad_t frev;
2104 
2105 	nfsm_srvmtofh(&nsfh);
2106 	nfsm_srvmtofh(&dnsfh);
2107 	nfsm_srvnamesiz(len);
2108 	error = nfsrv_fhtovp(&nsfh, false, &vp, cred, slp, nam,
2109 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2110 	if (error) {
2111 		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2112 		nfsm_srvpostop_attr(getret, &at);
2113 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2114 		return (0);
2115 	}
2116 	if (vp->v_type == VDIR) {
2117 		error = EPERM;
2118 		goto out1;
2119 	}
2120 	nd.ni_cnd.cn_cred = cred;
2121 	nd.ni_cnd.cn_nameiop = CREATE;
2122 	nd.ni_cnd.cn_flags = LOCKPARENT;
2123 	error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos,
2124 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2125 	if (dirp && v3) {
2126 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2127 	}
2128 	if (error)
2129 		goto out1;
2130 	xp = nd.ni_vp;
2131 	if (xp != NULL) {
2132 		error = EEXIST;
2133 		goto out;
2134 	}
2135 	xp = nd.ni_dvp;
2136 	if (vp->v_mount != xp->v_mount)
2137 		error = EXDEV;
2138 out:
2139 	if (!error) {
2140 		nqsrv_getl(vp, ND_WRITE);
2141 		nqsrv_getl(xp, ND_WRITE);
2142 		error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2143 	} else {
2144 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2145 		if (nd.ni_dvp == nd.ni_vp)
2146 			vrele(nd.ni_dvp);
2147 		else
2148 			vput(nd.ni_dvp);
2149 		if (nd.ni_vp)
2150 			vrele(nd.ni_vp);
2151 	}
2152 out1:
2153 	if (v3)
2154 		getret = VOP_GETATTR(vp, &at, cred);
2155 	if (dirp) {
2156 		if (v3) {
2157 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2158 		}
2159 		vrele(dirp);
2160 	}
2161 	vrele(vp);
2162 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2163 	if (v3) {
2164 		nfsm_srvpostop_attr(getret, &at);
2165 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2166 		return (0);
2167 	}
2168 	nfsm_srvdone;
2169 }
2170 
2171 /*
2172  * nfs symbolic link service
2173  */
2174 int
2175 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2176 {
2177 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2178 	struct mbuf *nam = nfsd->nd_nam;
2179 	char *dpos = nfsd->nd_dpos;
2180 	kauth_cred_t cred = nfsd->nd_cr;
2181 	struct vattr va, dirfor, diraft;
2182 	struct nameidata nd;
2183 	u_int32_t *tl;
2184 	int32_t t1;
2185 	struct nfsv2_sattr *sp;
2186 	char *bpos, *pathcp = NULL, *cp2;
2187 	struct uio io;
2188 	struct iovec iv;
2189 	int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1, abort = 0;
2190 	uint32_t len, len2;
2191 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2192 	struct mbuf *mb, *mreq;
2193 	struct vnode *dirp = (struct vnode *)0;
2194 	nfsrvfh_t nsfh;
2195 	u_quad_t frev;
2196 
2197 	nd.ni_cnd.cn_nameiop = 0;
2198 	nfsm_srvmtofh(&nsfh);
2199 	nfsm_srvnamesiz(len);
2200 	nd.ni_cnd.cn_cred = cred;
2201 	nd.ni_cnd.cn_nameiop = CREATE;
2202 	nd.ni_cnd.cn_flags = LOCKPARENT;
2203 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2204 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2205 	if (dirp && v3) {
2206 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2207 	}
2208 	if (error)
2209 		goto out;
2210 	abort = 1;
2211 	VATTR_NULL(&va);
2212 	va.va_type = VLNK;
2213 	if (v3) {
2214 		va.va_mode = 0;
2215 		nfsm_srvsattr(&va);
2216 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
2217 		len2 = fxdr_unsigned(uint32_t, *tl);
2218 		if (len2 > PATH_MAX) {
2219 			/* XXX should check _PC_NO_TRUNC */
2220 			error = ENAMETOOLONG;
2221 			goto abortop;
2222 		}
2223 	}
2224 	else {
2225 		/* NFSv2 */
2226 		nfsm_strsiz(len2, NFS_MAXPATHLEN);
2227 	}
2228 	pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2229 	iv.iov_base = pathcp;
2230 	iv.iov_len = len2;
2231 	io.uio_resid = len2;
2232 	io.uio_offset = 0;
2233 	io.uio_iov = &iv;
2234 	io.uio_iovcnt = 1;
2235 	io.uio_rw = UIO_READ;
2236 	UIO_SETUP_SYSSPACE(&io);
2237 	nfsm_mtouio(&io, len2);
2238 	if (!v3) {
2239 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2240 		va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
2241 	}
2242 	*(pathcp + len2) = '\0';
2243 	if (nd.ni_vp) {
2244 		error = EEXIST;
2245 abortop:
2246 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2247 		if (nd.ni_dvp == nd.ni_vp)
2248 			vrele(nd.ni_dvp);
2249 		else
2250 			vput(nd.ni_dvp);
2251 		if (nd.ni_vp)
2252 			vrele(nd.ni_vp);
2253 		goto out;
2254 	}
2255 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2256 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp);
2257 	if (!error) {
2258 	    if (v3) {
2259 		error = nfsrv_composefh(nd.ni_vp, &nsfh, v3);
2260 		if (!error)
2261 		    error = VOP_GETATTR(nd.ni_vp, &va, cred);
2262 		vput(nd.ni_vp);
2263 	    } else {
2264 		vput(nd.ni_vp);
2265 	    }
2266 	}
2267 out:
2268 	if (pathcp)
2269 		free(pathcp, M_TEMP);
2270 	if (dirp) {
2271 		if (v3) {
2272 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2273 		}
2274 		vrele(dirp);
2275 		dirp = NULL;
2276 	}
2277 	abort = 0;
2278 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) +
2279 	    NFSX_WCCDATA(v3));
2280 	if (v3) {
2281 		if (!error) {
2282 			nfsm_srvpostop_fh(&nsfh);
2283 			nfsm_srvpostop_attr(0, &va);
2284 		}
2285 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2286 	}
2287 	return (0);
2288 nfsmout:
2289 	if (abort) {
2290 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2291 		if (nd.ni_dvp == nd.ni_vp)
2292 			vrele(nd.ni_dvp);
2293 		else
2294 			vput(nd.ni_dvp);
2295 		if (nd.ni_vp)
2296 			vrele(nd.ni_vp);
2297 	}
2298 	if (dirp)
2299 		vrele(dirp);
2300 	if (pathcp)
2301 		free(pathcp, M_TEMP);
2302 	return (error);
2303 }
2304 
2305 /*
2306  * nfs mkdir service
2307  */
2308 int
2309 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2310 {
2311 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2312 	struct mbuf *nam = nfsd->nd_nam;
2313 	char *dpos = nfsd->nd_dpos;
2314 	kauth_cred_t cred = nfsd->nd_cr;
2315 	struct vattr va, dirfor, diraft;
2316 	struct nfs_fattr *fp;
2317 	struct nameidata nd;
2318 	char *cp;
2319 	u_int32_t *tl;
2320 	int32_t t1;
2321 	char *bpos;
2322 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2323 	int abort = 0;
2324 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2325 	char *cp2;
2326 	struct mbuf *mb, *mreq;
2327 	struct vnode *vp, *dirp = (struct vnode *)0;
2328 	nfsrvfh_t nsfh;
2329 	u_quad_t frev;
2330 
2331 	nfsm_srvmtofh(&nsfh);
2332 	nfsm_srvnamesiz(len);
2333 	nd.ni_cnd.cn_cred = cred;
2334 	nd.ni_cnd.cn_nameiop = CREATE;
2335 	nd.ni_cnd.cn_flags = LOCKPARENT;
2336 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2337 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2338 	if (dirp && v3) {
2339 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2340 	}
2341 	if (error) {
2342 		nfsm_reply(NFSX_WCCDATA(v3));
2343 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2344 		if (dirp)
2345 			vrele(dirp);
2346 		return (0);
2347 	}
2348 	abort = 1;
2349 	VATTR_NULL(&va);
2350 	if (v3) {
2351 		va.va_mode = 0;
2352 		nfsm_srvsattr(&va);
2353 	} else {
2354 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2355 		va.va_mode = nfstov_mode(*tl++);
2356 	}
2357 	va.va_type = VDIR;
2358 	vp = nd.ni_vp;
2359 	if (vp != NULL) {
2360 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2361 		if (nd.ni_dvp == vp)
2362 			vrele(nd.ni_dvp);
2363 		else
2364 			vput(nd.ni_dvp);
2365 		vrele(vp);
2366 		error = EEXIST;
2367 		goto out;
2368 	}
2369 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2370 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
2371 	if (!error) {
2372 		vp = nd.ni_vp;
2373 		error = nfsrv_composefh(vp, &nsfh, v3);
2374 		if (!error)
2375 			error = VOP_GETATTR(vp, &va, cred);
2376 		vput(vp);
2377 	}
2378 out:
2379 	if (dirp) {
2380 		if (v3) {
2381 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2382 		}
2383 		vrele(dirp);
2384 		dirp = NULL;
2385 	}
2386 	abort = 0;
2387 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) +
2388 	    NFSX_WCCDATA(v3));
2389 	if (v3) {
2390 		if (!error) {
2391 			nfsm_srvpostop_fh(&nsfh);
2392 			nfsm_srvpostop_attr(0, &va);
2393 		}
2394 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2395 	} else {
2396 		nfsm_srvfhtom(&nsfh, v3);
2397 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2398 		nfsm_srvfillattr(&va, fp);
2399 	}
2400 	return (0);
2401 nfsmout:
2402 	if (abort) {
2403 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2404 		if (nd.ni_dvp == nd.ni_vp)
2405 			vrele(nd.ni_dvp);
2406 		else
2407 			vput(nd.ni_dvp);
2408 		if (nd.ni_vp)
2409 			vrele(nd.ni_vp);
2410 	}
2411 	if (dirp)
2412 		vrele(dirp);
2413 	return (error);
2414 }
2415 
2416 /*
2417  * nfs rmdir service
2418  */
2419 int
2420 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2421 {
2422 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2423 	struct mbuf *nam = nfsd->nd_nam;
2424 	char *dpos = nfsd->nd_dpos;
2425 	kauth_cred_t cred = nfsd->nd_cr;
2426 	u_int32_t *tl;
2427 	int32_t t1;
2428 	char *bpos;
2429 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2430 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2431 	char *cp2;
2432 	struct mbuf *mb, *mreq;
2433 	struct vnode *vp, *dirp = (struct vnode *)0;
2434 	struct vattr dirfor, diraft;
2435 	nfsrvfh_t nsfh;
2436 	struct nameidata nd;
2437 	u_quad_t frev;
2438 
2439 	nfsm_srvmtofh(&nsfh);
2440 	nfsm_srvnamesiz(len);
2441 	nd.ni_cnd.cn_cred = cred;
2442 	nd.ni_cnd.cn_nameiop = DELETE;
2443 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2444 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2445 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2446 	if (dirp && v3) {
2447 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2448 	}
2449 	if (error) {
2450 		nfsm_reply(NFSX_WCCDATA(v3));
2451 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2452 		if (dirp)
2453 			vrele(dirp);
2454 		return (0);
2455 	}
2456 	vp = nd.ni_vp;
2457 	if (vp->v_type != VDIR) {
2458 		error = ENOTDIR;
2459 		goto out;
2460 	}
2461 	/*
2462 	 * No rmdir "." please.
2463 	 */
2464 	if (nd.ni_dvp == vp) {
2465 		error = EINVAL;
2466 		goto out;
2467 	}
2468 	/*
2469 	 * The root of a mounted filesystem cannot be deleted.
2470 	 */
2471 	if (vp->v_vflag & VV_ROOT)
2472 		error = EBUSY;
2473 out:
2474 	if (!error) {
2475 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
2476 		nqsrv_getl(vp, ND_WRITE);
2477 		error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2478 	} else {
2479 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2480 		if (nd.ni_dvp == nd.ni_vp)
2481 			vrele(nd.ni_dvp);
2482 		else
2483 			vput(nd.ni_dvp);
2484 		vput(vp);
2485 	}
2486 	if (dirp) {
2487 		if (v3) {
2488 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2489 		}
2490 		vrele(dirp);
2491 	}
2492 	nfsm_reply(NFSX_WCCDATA(v3));
2493 	if (v3) {
2494 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2495 		return (0);
2496 	}
2497 	nfsm_srvdone;
2498 }
2499 
2500 /*
2501  * nfs readdir service
2502  * - mallocs what it thinks is enough to read
2503  *	count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR
2504  * - calls VOP_READDIR()
2505  * - loops around building the reply
2506  *	if the output generated exceeds count break out of loop
2507  *	The nfsm_clget macro is used here so that the reply will be packed
2508  *	tightly in mbuf clusters.
2509  * - it only knows that it has encountered eof when the VOP_READDIR()
2510  *	reads nothing
2511  * - as such one readdir rpc will return eof false although you are there
2512  *	and then the next will return eof
2513  * - it trims out records with d_fileno == 0
2514  *	this doesn't matter for Unix clients, but they might confuse clients
2515  *	for other os'.
2516  * - it trims out records with d_type == DT_WHT
2517  *	these cannot be seen through NFS (unless we extend the protocol)
2518  * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2519  *	than requested, but this may not apply to all filesystems. For
2520  *	example, client NFS does not { although it is never remote mounted
2521  *	anyhow }
2522  *     The alternate call nfsrv_readdirplus() does lookups as well.
2523  * PS: The NFS protocol spec. does not clarify what the "count" byte
2524  *	argument is a count of.. just name strings and file id's or the
2525  *	entire reply rpc or ...
2526  *	I tried just file name and id sizes and it confused the Sun client,
2527  *	so I am using the full rpc size now. The "paranoia.." comment refers
2528  *	to including the status longwords that are not a part of the dir.
2529  *	"entry" structures, but are in the rpc.
2530  */
2531 
2532 #define	NFS_SRVDIRBLKSIZ	1024
2533 
2534 struct flrep {
2535 	nfsuint64 fl_off;
2536 	u_int32_t fl_postopok;
2537 	struct nfs_fattr fl_fattr; /* XXX: must be of fattr3 size */
2538 	u_int32_t fl_fhok;
2539 	u_int32_t fl_fhsize;
2540 	/* handle comes here, filled in dynamically */
2541 };
2542 
2543 int
2544 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2545 {
2546 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2547 	struct mbuf *nam = nfsd->nd_nam;
2548 	char *dpos = nfsd->nd_dpos;
2549 	kauth_cred_t cred = nfsd->nd_cr;
2550 	char *bp, *be;
2551 	struct mbuf *mp;
2552 	struct dirent *dp;
2553 	char *cp;
2554 	u_int32_t *tl;
2555 	int32_t t1;
2556 	char *bpos;
2557 	struct mbuf *mb, *mreq, *mp2;
2558 	char *cpos, *cend, *cp2, *rbuf;
2559 	struct vnode *vp;
2560 	struct vattr at;
2561 	nfsrvfh_t nsfh;
2562 	struct uio io;
2563 	struct iovec iv;
2564 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2565 	int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies;
2566 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2567 	u_quad_t frev, off, toff, verf;
2568 	off_t *cookies = NULL, *cookiep;
2569 	nfsuint64 jar;
2570 
2571 	nfsm_srvmtofh(&nsfh);
2572 	if (v3) {
2573 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2574 		toff = fxdr_hyper(tl);
2575 		tl += 2;
2576 		verf = fxdr_hyper(tl);
2577 		tl += 2;
2578 	} else {
2579 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2580 		toff = fxdr_unsigned(u_quad_t, *tl++);
2581 	}
2582 	off = toff;
2583 	cnt = fxdr_unsigned(int, *tl);
2584 	siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1));
2585 	xfer = NFS_SRVMAXDATA(nfsd);
2586 	if (siz > xfer)
2587 		siz = xfer;
2588 	fullsiz = siz;
2589 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
2590 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2591 	if (!error && vp->v_type != VDIR) {
2592 		error = ENOTDIR;
2593 		vput(vp);
2594 	}
2595 	if (error) {
2596 		nfsm_reply(NFSX_UNSIGNED);
2597 		nfsm_srvpostop_attr(getret, &at);
2598 		return (0);
2599 	}
2600 	nqsrv_getl(vp, ND_READ);
2601 	if (v3) {
2602 		error = getret = VOP_GETATTR(vp, &at, cred);
2603 #ifdef NFS3_STRICTVERF
2604 		/*
2605 		 * XXX This check is too strict for Solaris 2.5 clients.
2606 		 */
2607 		if (!error && toff && verf != at.va_filerev)
2608 			error = NFSERR_BAD_COOKIE;
2609 #endif
2610 	}
2611 	if (!error)
2612 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0);
2613 	if (error) {
2614 		vput(vp);
2615 		nfsm_reply(NFSX_POSTOPATTR(v3));
2616 		nfsm_srvpostop_attr(getret, &at);
2617 		return (0);
2618 	}
2619 	VOP_UNLOCK(vp, 0);
2620 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2621 again:
2622 	iv.iov_base = rbuf;
2623 	iv.iov_len = fullsiz;
2624 	io.uio_iov = &iv;
2625 	io.uio_iovcnt = 1;
2626 	io.uio_offset = (off_t)off;
2627 	io.uio_resid = fullsiz;
2628 	io.uio_rw = UIO_READ;
2629 	UIO_SETUP_SYSSPACE(&io);
2630 	eofflag = 0;
2631 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2632 
2633 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2634 
2635 	off = (off_t)io.uio_offset;
2636 	if (!cookies && !error)
2637 		error = NFSERR_PERM;
2638 	if (v3) {
2639 		getret = VOP_GETATTR(vp, &at, cred);
2640 		if (!error)
2641 			error = getret;
2642 	}
2643 
2644 	VOP_UNLOCK(vp, 0);
2645 	if (error) {
2646 		vrele(vp);
2647 		free((void *)rbuf, M_TEMP);
2648 		if (cookies)
2649 			free((void *)cookies, M_TEMP);
2650 		nfsm_reply(NFSX_POSTOPATTR(v3));
2651 		nfsm_srvpostop_attr(getret, &at);
2652 		return (0);
2653 	}
2654 	if (io.uio_resid) {
2655 		siz -= io.uio_resid;
2656 
2657 		/*
2658 		 * If nothing read, return eof
2659 		 * rpc reply
2660 		 */
2661 		if (siz == 0) {
2662 			vrele(vp);
2663 			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2664 				2 * NFSX_UNSIGNED);
2665 			if (v3) {
2666 				nfsm_srvpostop_attr(getret, &at);
2667 				nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2668 				txdr_hyper(at.va_filerev, tl);
2669 				tl += 2;
2670 			} else
2671 				nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2672 			*tl++ = nfs_false;
2673 			*tl = nfs_true;
2674 			free((void *)rbuf, M_TEMP);
2675 			free((void *)cookies, M_TEMP);
2676 			return (0);
2677 		}
2678 	}
2679 
2680 	/*
2681 	 * Check for degenerate cases of nothing useful read.
2682 	 * If so go try again
2683 	 */
2684 	cpos = rbuf;
2685 	cend = rbuf + siz;
2686 	dp = (struct dirent *)cpos;
2687 	cookiep = cookies;
2688 
2689 	while (cpos < cend && ncookies > 0 &&
2690 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2691 		cpos += dp->d_reclen;
2692 		dp = (struct dirent *)cpos;
2693 		cookiep++;
2694 		ncookies--;
2695 	}
2696 	if (cpos >= cend || ncookies == 0) {
2697 		toff = off;
2698 		siz = fullsiz;
2699 		free(cookies, M_TEMP);
2700 		cookies = NULL;
2701 		goto again;
2702 	}
2703 
2704 	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2705 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2706 	if (v3) {
2707 		nfsm_srvpostop_attr(getret, &at);
2708 		nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2709 		txdr_hyper(at.va_filerev, tl);
2710 	}
2711 	mp = mp2 = mb;
2712 	bp = bpos;
2713 	be = bp + M_TRAILINGSPACE(mp);
2714 
2715 	/* Loop through the records and build reply */
2716 	while (cpos < cend && ncookies > 0) {
2717 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2718 			nlen = dp->d_namlen;
2719 			rem = nfsm_rndup(nlen)-nlen;
2720 			len += (4 * NFSX_UNSIGNED + nlen + rem);
2721 			if (v3)
2722 				len += 2 * NFSX_UNSIGNED;
2723 			if (len > cnt) {
2724 				eofflag = 0;
2725 				break;
2726 			}
2727 			/*
2728 			 * Build the directory record xdr from
2729 			 * the dirent entry.
2730 			 */
2731 			nfsm_clget;
2732 			*tl = nfs_true;
2733 			bp += NFSX_UNSIGNED;
2734 			if (v3) {
2735 				nfsm_clget;
2736 				*tl = txdr_unsigned(dp->d_fileno >> 32);
2737 				bp += NFSX_UNSIGNED;
2738 			}
2739 			nfsm_clget;
2740 			*tl = txdr_unsigned(dp->d_fileno);
2741 			bp += NFSX_UNSIGNED;
2742 			nfsm_clget;
2743 			*tl = txdr_unsigned(nlen);
2744 			bp += NFSX_UNSIGNED;
2745 
2746 			/* And loop around copying the name */
2747 			xfer = nlen;
2748 			cp = dp->d_name;
2749 			while (xfer > 0) {
2750 				nfsm_clget;
2751 				if ((bp+xfer) > be)
2752 					tsiz = be-bp;
2753 				else
2754 					tsiz = xfer;
2755 				memcpy(bp, cp, tsiz);
2756 				bp += tsiz;
2757 				xfer -= tsiz;
2758 				if (xfer > 0)
2759 					cp += tsiz;
2760 			}
2761 			/* And null pad to an int32_t boundary */
2762 			for (i = 0; i < rem; i++)
2763 				*bp++ = '\0';
2764 			nfsm_clget;
2765 
2766 			/* Finish off the record */
2767 			txdr_hyper(*cookiep, &jar);
2768 			if (v3) {
2769 				*tl = jar.nfsuquad[0];
2770 				bp += NFSX_UNSIGNED;
2771 				nfsm_clget;
2772 			}
2773 			*tl = jar.nfsuquad[1];
2774 			bp += NFSX_UNSIGNED;
2775 		}
2776 		cpos += dp->d_reclen;
2777 		dp = (struct dirent *)cpos;
2778 		cookiep++;
2779 		ncookies--;
2780 	}
2781 	vrele(vp);
2782 	nfsm_clget;
2783 	*tl = nfs_false;
2784 	bp += NFSX_UNSIGNED;
2785 	nfsm_clget;
2786 	if (eofflag)
2787 		*tl = nfs_true;
2788 	else
2789 		*tl = nfs_false;
2790 	bp += NFSX_UNSIGNED;
2791 	if (mp != mb) {
2792 		if (bp < be)
2793 			mp->m_len = bp - mtod(mp, char *);
2794 	} else
2795 		mp->m_len += bp - bpos;
2796 	free((void *)rbuf, M_TEMP);
2797 	free((void *)cookies, M_TEMP);
2798 	nfsm_srvdone;
2799 }
2800 
2801 int
2802 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2803 {
2804 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2805 	struct mbuf *nam = nfsd->nd_nam;
2806 	char *dpos = nfsd->nd_dpos;
2807 	kauth_cred_t cred = nfsd->nd_cr;
2808 	char *bp, *be;
2809 	struct mbuf *mp;
2810 	struct dirent *dp;
2811 	char *cp;
2812 	u_int32_t *tl;
2813 	int32_t t1;
2814 	char *bpos;
2815 	struct mbuf *mb, *mreq, *mp2;
2816 	char *cpos, *cend, *cp2, *rbuf;
2817 	struct vnode *vp, *nvp;
2818 	struct flrep fl;
2819 	nfsrvfh_t nsfh;
2820 	struct uio io;
2821 	struct iovec iv;
2822 	struct vattr va, at, *vap = &va;
2823 	struct nfs_fattr *fp;
2824 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2825 	int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies;
2826 	u_quad_t frev, off, toff, verf;
2827 	off_t *cookies = NULL, *cookiep;
2828 
2829 	nfsm_srvmtofh(&nsfh);
2830 	nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2831 	toff = fxdr_hyper(tl);
2832 	tl += 2;
2833 	verf = fxdr_hyper(tl);
2834 	tl += 2;
2835 	siz = fxdr_unsigned(int, *tl++);
2836 	cnt = fxdr_unsigned(int, *tl);
2837 	off = toff;
2838 	siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1));
2839 	xfer = NFS_SRVMAXDATA(nfsd);
2840 	if (siz > xfer)
2841 		siz = xfer;
2842 	fullsiz = siz;
2843 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
2844 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2845 	if (!error && vp->v_type != VDIR) {
2846 		error = ENOTDIR;
2847 		vput(vp);
2848 	}
2849 	if (error) {
2850 		nfsm_reply(NFSX_UNSIGNED);
2851 		nfsm_srvpostop_attr(getret, &at);
2852 		return (0);
2853 	}
2854 	error = getret = VOP_GETATTR(vp, &at, cred);
2855 #ifdef NFS3_STRICTVERF
2856 	/*
2857 	 * XXX This check is too strict for Solaris 2.5 clients.
2858 	 */
2859 	if (!error && toff && verf != at.va_filerev)
2860 		error = NFSERR_BAD_COOKIE;
2861 #endif
2862 	if (!error) {
2863 		nqsrv_getl(vp, ND_READ);
2864 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0);
2865 	}
2866 	if (error) {
2867 		vput(vp);
2868 		nfsm_reply(NFSX_V3POSTOPATTR);
2869 		nfsm_srvpostop_attr(getret, &at);
2870 		return (0);
2871 	}
2872 	VOP_UNLOCK(vp, 0);
2873 
2874 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2875 again:
2876 	iv.iov_base = rbuf;
2877 	iv.iov_len = fullsiz;
2878 	io.uio_iov = &iv;
2879 	io.uio_iovcnt = 1;
2880 	io.uio_offset = (off_t)off;
2881 	io.uio_resid = fullsiz;
2882 	io.uio_rw = UIO_READ;
2883 	UIO_SETUP_SYSSPACE(&io);
2884 	eofflag = 0;
2885 
2886 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2887 
2888 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2889 
2890 	off = (u_quad_t)io.uio_offset;
2891 	getret = VOP_GETATTR(vp, &at, cred);
2892 
2893 	VOP_UNLOCK(vp, 0);
2894 
2895 	/*
2896 	 * If the VGET operation doesn't work for this filesystem,
2897 	 * we can't support readdirplus. Returning NOTSUPP should
2898 	 * make clients fall back to plain readdir.
2899 	 * There's no need to check for VPTOFH as well, we wouldn't
2900 	 * even be here otherwise.
2901 	 */
2902 	if (!getret) {
2903 		if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp)))
2904 			getret = (getret == EOPNOTSUPP) ?
2905 				NFSERR_NOTSUPP : NFSERR_IO;
2906 		else
2907 			vput(nvp);
2908 	}
2909 
2910 	if (!cookies && !error)
2911 		error = NFSERR_PERM;
2912 	if (!error)
2913 		error = getret;
2914 	if (error) {
2915 		vrele(vp);
2916 		if (cookies)
2917 			free((void *)cookies, M_TEMP);
2918 		free((void *)rbuf, M_TEMP);
2919 		nfsm_reply(NFSX_V3POSTOPATTR);
2920 		nfsm_srvpostop_attr(getret, &at);
2921 		return (0);
2922 	}
2923 	if (io.uio_resid) {
2924 		siz -= io.uio_resid;
2925 
2926 		/*
2927 		 * If nothing read, return eof
2928 		 * rpc reply
2929 		 */
2930 		if (siz == 0) {
2931 			vrele(vp);
2932 			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
2933 				2 * NFSX_UNSIGNED);
2934 			nfsm_srvpostop_attr(getret, &at);
2935 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2936 			txdr_hyper(at.va_filerev, tl);
2937 			tl += 2;
2938 			*tl++ = nfs_false;
2939 			*tl = nfs_true;
2940 			free((void *)cookies, M_TEMP);
2941 			free((void *)rbuf, M_TEMP);
2942 			return (0);
2943 		}
2944 	}
2945 
2946 	/*
2947 	 * Check for degenerate cases of nothing useful read.
2948 	 * If so go try again
2949 	 */
2950 	cpos = rbuf;
2951 	cend = rbuf + siz;
2952 	dp = (struct dirent *)cpos;
2953 	cookiep = cookies;
2954 
2955 	while (cpos < cend && ncookies > 0 &&
2956 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2957 		cpos += dp->d_reclen;
2958 		dp = (struct dirent *)cpos;
2959 		cookiep++;
2960 		ncookies--;
2961 	}
2962 	if (cpos >= cend || ncookies == 0) {
2963 		toff = off;
2964 		siz = fullsiz;
2965 		free(cookies, M_TEMP);
2966 		cookies = NULL;
2967 		goto again;
2968 	}
2969 
2970 	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
2971 	nfsm_reply(cnt);
2972 	nfsm_srvpostop_attr(getret, &at);
2973 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2974 	txdr_hyper(at.va_filerev, tl);
2975 	mp = mp2 = mb;
2976 	bp = bpos;
2977 	be = bp + M_TRAILINGSPACE(mp);
2978 
2979 	/* Loop through the records and build reply */
2980 	while (cpos < cend && ncookies > 0) {
2981 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2982 			nfsrvfh_t nnsfh;
2983 
2984 			nlen = dp->d_namlen;
2985 			rem = nfsm_rndup(nlen)-nlen;
2986 
2987 			/*
2988 			 * For readdir_and_lookup get the vnode using
2989 			 * the file number.
2990 			 */
2991 			if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
2992 				goto invalid;
2993 			if (nfsrv_composefh(nvp, &nnsfh, true)) {
2994 				vput(nvp);
2995 				goto invalid;
2996 			}
2997 			if (VOP_GETATTR(nvp, vap, cred)) {
2998 				vput(nvp);
2999 				goto invalid;
3000 			}
3001 			vput(nvp);
3002 
3003 			/*
3004 			 * If either the dircount or maxcount will be
3005 			 * exceeded, get out now. Both of these lengths
3006 			 * are calculated conservatively, including all
3007 			 * XDR overheads.
3008 			 */
3009 			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3010 				NFSX_V3POSTOPATTR);
3011 			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3012 			if (len > cnt || dirlen > fullsiz) {
3013 				eofflag = 0;
3014 				break;
3015 			}
3016 
3017 			/*
3018 			 * Build the directory record xdr from
3019 			 * the dirent entry.
3020 			 */
3021 			fp = (struct nfs_fattr *)&fl.fl_fattr;
3022 			nfsm_srvfillattr(vap, fp);
3023 			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3024 			fl.fl_fhok = nfs_true;
3025 			fl.fl_postopok = nfs_true;
3026 			txdr_hyper(*cookiep, fl.fl_off.nfsuquad);
3027 
3028 			nfsm_clget;
3029 			*tl = nfs_true;
3030 			bp += NFSX_UNSIGNED;
3031 			nfsm_clget;
3032 			*tl = txdr_unsigned(dp->d_fileno >> 32);
3033 			bp += NFSX_UNSIGNED;
3034 			nfsm_clget;
3035 			*tl = txdr_unsigned(dp->d_fileno);
3036 			bp += NFSX_UNSIGNED;
3037 			nfsm_clget;
3038 			*tl = txdr_unsigned(nlen);
3039 			bp += NFSX_UNSIGNED;
3040 
3041 			/* And loop around copying the name */
3042 			xfer = nlen;
3043 			cp = dp->d_name;
3044 			while (xfer > 0) {
3045 				nfsm_clget;
3046 				if ((bp + xfer) > be)
3047 					tsiz = be - bp;
3048 				else
3049 					tsiz = xfer;
3050 				memcpy(bp, cp, tsiz);
3051 				bp += tsiz;
3052 				xfer -= tsiz;
3053 				if (xfer > 0)
3054 					cp += tsiz;
3055 			}
3056 			/* And null pad to an int32_t boundary */
3057 			for (i = 0; i < rem; i++)
3058 				*bp++ = '\0';
3059 
3060 			/*
3061 			 * Now copy the flrep structure out.
3062 			 */
3063 			xfer = sizeof(struct flrep);
3064 			cp = (void *)&fl;
3065 			while (xfer > 0) {
3066 				nfsm_clget;
3067 				if ((bp + xfer) > be)
3068 					tsiz = be - bp;
3069 				else
3070 					tsiz = xfer;
3071 				memcpy(bp, cp, tsiz);
3072 				bp += tsiz;
3073 				xfer -= tsiz;
3074 				if (xfer > 0)
3075 					cp += tsiz;
3076 			}
3077 
3078 			/*
3079 			 * ... and filehandle.
3080 			 */
3081 			xfer = NFSRVFH_SIZE(&nnsfh);
3082 			cp = NFSRVFH_DATA(&nnsfh);
3083 			while (xfer > 0) {
3084 				nfsm_clget;
3085 				if ((bp + xfer) > be)
3086 					tsiz = be - bp;
3087 				else
3088 					tsiz = xfer;
3089 				memcpy(bp, cp, tsiz);
3090 				bp += tsiz;
3091 				xfer -= tsiz;
3092 				if (xfer > 0)
3093 					cp += tsiz;
3094 			}
3095 		}
3096 invalid:
3097 		cpos += dp->d_reclen;
3098 		dp = (struct dirent *)cpos;
3099 		cookiep++;
3100 		ncookies--;
3101 	}
3102 	vrele(vp);
3103 	nfsm_clget;
3104 	*tl = nfs_false;
3105 	bp += NFSX_UNSIGNED;
3106 	nfsm_clget;
3107 	if (eofflag)
3108 		*tl = nfs_true;
3109 	else
3110 		*tl = nfs_false;
3111 	bp += NFSX_UNSIGNED;
3112 	if (mp != mb) {
3113 		if (bp < be)
3114 			mp->m_len = bp - mtod(mp, char *);
3115 	} else
3116 		mp->m_len += bp - bpos;
3117 	free((void *)cookies, M_TEMP);
3118 	free((void *)rbuf, M_TEMP);
3119 	nfsm_srvdone;
3120 }
3121 
3122 /*
3123  * nfs commit service
3124  */
3125 int
3126 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3127 {
3128 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3129 	struct mbuf *nam = nfsd->nd_nam;
3130 	char *dpos = nfsd->nd_dpos;
3131 	kauth_cred_t cred = nfsd->nd_cr;
3132 	struct vattr bfor, aft;
3133 	struct vnode *vp;
3134 	nfsrvfh_t nsfh;
3135 	u_int32_t *tl;
3136 	int32_t t1;
3137 	char *bpos;
3138 	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0;
3139 	uint32_t cnt;
3140 	char *cp2;
3141 	struct mbuf *mb, *mreq;
3142 	u_quad_t frev, off, end;
3143 
3144 	nfsm_srvmtofh(&nsfh);
3145 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3146 
3147 	off = fxdr_hyper(tl);
3148 	tl += 2;
3149 	cnt = fxdr_unsigned(uint32_t, *tl);
3150 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3151 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3152 	if (error) {
3153 		nfsm_reply(2 * NFSX_UNSIGNED);
3154 		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3155 		return (0);
3156 	}
3157 	for_ret = VOP_GETATTR(vp, &bfor, cred);
3158 	end = (cnt > 0) ? off + cnt : vp->v_size;
3159 	if (end < off || end > vp->v_size)
3160 		end = vp->v_size;
3161 	if (off < vp->v_size)
3162 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end);
3163 	/* else error == 0, from nfsrv_fhtovp() */
3164 	aft_ret = VOP_GETATTR(vp, &aft, cred);
3165 	vput(vp);
3166 	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3167 	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3168 	if (!error) {
3169 		nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3170 		*tl++ = txdr_unsigned(boottime.tv_sec);
3171 		*tl = txdr_unsigned(boottime.tv_nsec / 1000);
3172 	} else {
3173 		return (0);
3174 	}
3175 	nfsm_srvdone;
3176 }
3177 
3178 /*
3179  * nfs statfs service
3180  */
3181 int
3182 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3183 {
3184 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3185 	struct mbuf *nam = nfsd->nd_nam;
3186 	char *dpos = nfsd->nd_dpos;
3187 	kauth_cred_t cred = nfsd->nd_cr;
3188 	struct statvfs *sf = NULL;
3189 	struct nfs_statfs *sfp;
3190 	u_int32_t *tl;
3191 	int32_t t1;
3192 	char *bpos;
3193 	int error = 0, rdonly, cache = 0, getret = 1;
3194 	int v3 = (nfsd->nd_flag & ND_NFSV3);
3195 	char *cp2;
3196 	struct mbuf *mb, *mreq;
3197 	struct vnode *vp;
3198 	struct vattr at;
3199 	nfsrvfh_t nsfh;
3200 	u_quad_t frev, tval;
3201 
3202 	nfsm_srvmtofh(&nsfh);
3203 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3204 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3205 	if (error) {
3206 		nfsm_reply(NFSX_UNSIGNED);
3207 		nfsm_srvpostop_attr(getret, &at);
3208 		return (0);
3209 	}
3210 	sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK);
3211 	error = VFS_STATVFS(vp->v_mount, sf);
3212 	getret = VOP_GETATTR(vp, &at, cred);
3213 	vput(vp);
3214 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3215 	if (v3)
3216 		nfsm_srvpostop_attr(getret, &at);
3217 	if (error) {
3218 		free(sf, M_TEMP);
3219 		return (0);
3220 	}
3221 	nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3222 	if (v3) {
3223 		tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize);
3224 		txdr_hyper(tval, &sfp->sf_tbytes);
3225 		tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize);
3226 		txdr_hyper(tval, &sfp->sf_fbytes);
3227 		tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize);
3228 		txdr_hyper(tval, &sfp->sf_abytes);
3229 		tval = (u_quad_t)sf->f_files;
3230 		txdr_hyper(tval, &sfp->sf_tfiles);
3231 		tval = (u_quad_t)sf->f_ffree;
3232 		txdr_hyper(tval, &sfp->sf_ffiles);
3233 		txdr_hyper(tval, &sfp->sf_afiles);
3234 		sfp->sf_invarsec = 0;
3235 	} else {
3236 		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3237 		sfp->sf_bsize = txdr_unsigned(sf->f_frsize);
3238 		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3239 		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3240 		sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3241 	}
3242 nfsmout:
3243 	if (sf)
3244 	    free(sf, M_TEMP);
3245 	return error;
3246 }
3247 
3248 /*
3249  * nfs fsinfo service
3250  */
3251 int
3252 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3253 {
3254 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3255 	struct mbuf *nam = nfsd->nd_nam;
3256 	char *dpos = nfsd->nd_dpos;
3257 	kauth_cred_t cred = nfsd->nd_cr;
3258 	u_int32_t *tl;
3259 	struct nfsv3_fsinfo *sip;
3260 	int32_t t1;
3261 	char *bpos;
3262 	int error = 0, rdonly, cache = 0, getret = 1;
3263 	uint32_t maxdata;
3264 	char *cp2;
3265 	struct mbuf *mb, *mreq;
3266 	struct vnode *vp;
3267 	struct vattr at;
3268 	nfsrvfh_t nsfh;
3269 	u_quad_t frev, maxfsize;
3270 	struct statvfs *sb;
3271 
3272 	nfsm_srvmtofh(&nsfh);
3273 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3274 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3275 	if (error) {
3276 		nfsm_reply(NFSX_UNSIGNED);
3277 		nfsm_srvpostop_attr(getret, &at);
3278 		return (0);
3279 	}
3280 
3281 	/* XXX Try to make a guess on the max file size. */
3282 	sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK);
3283 	VFS_STATVFS(vp->v_mount, sb);
3284 	maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1;
3285 	free(sb, M_TEMP);
3286 
3287 	getret = VOP_GETATTR(vp, &at, cred);
3288 	vput(vp);
3289 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3290 	nfsm_srvpostop_attr(getret, &at);
3291 	nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3292 
3293 	/*
3294 	 * XXX
3295 	 * There should be file system VFS OP(s) to get this information.
3296 	 * For now, assume ufs.
3297 	 */
3298 	if (slp->ns_so->so_type == SOCK_DGRAM)
3299 		maxdata = NFS_MAXDGRAMDATA;
3300 	else
3301 		maxdata = NFS_MAXDATA;
3302 	sip->fs_rtmax = txdr_unsigned(maxdata);
3303 	sip->fs_rtpref = txdr_unsigned(maxdata);
3304 	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3305 	sip->fs_wtmax = txdr_unsigned(maxdata);
3306 	sip->fs_wtpref = txdr_unsigned(maxdata);
3307 	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3308 	sip->fs_dtpref = txdr_unsigned(maxdata);
3309 	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3310 	sip->fs_timedelta.nfsv3_sec = 0;
3311 	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3312 	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3313 		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3314 		NFSV3FSINFO_CANSETTIME);
3315 	nfsm_srvdone;
3316 }
3317 
3318 /*
3319  * nfs pathconf service
3320  */
3321 int
3322 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3323 {
3324 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3325 	struct mbuf *nam = nfsd->nd_nam;
3326 	char *dpos = nfsd->nd_dpos;
3327 	kauth_cred_t cred = nfsd->nd_cr;
3328 	u_int32_t *tl;
3329 	struct nfsv3_pathconf *pc;
3330 	int32_t t1;
3331 	char *bpos;
3332 	int error = 0, rdonly, cache = 0, getret = 1;
3333 	register_t linkmax, namemax, chownres, notrunc;
3334 	char *cp2;
3335 	struct mbuf *mb, *mreq;
3336 	struct vnode *vp;
3337 	struct vattr at;
3338 	nfsrvfh_t nsfh;
3339 	u_quad_t frev;
3340 
3341 	nfsm_srvmtofh(&nsfh);
3342 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3343 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3344 	if (error) {
3345 		nfsm_reply(NFSX_UNSIGNED);
3346 		nfsm_srvpostop_attr(getret, &at);
3347 		return (0);
3348 	}
3349 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3350 	if (!error)
3351 		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3352 	if (!error)
3353 		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3354 	if (!error)
3355 		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3356 	getret = VOP_GETATTR(vp, &at, cred);
3357 	vput(vp);
3358 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3359 	nfsm_srvpostop_attr(getret, &at);
3360 	if (error)
3361 		return (0);
3362 	nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3363 
3364 	pc->pc_linkmax = txdr_unsigned(linkmax);
3365 	pc->pc_namemax = txdr_unsigned(namemax);
3366 	pc->pc_notrunc = txdr_unsigned(notrunc);
3367 	pc->pc_chownrestricted = txdr_unsigned(chownres);
3368 
3369 	/*
3370 	 * These should probably be supported by VOP_PATHCONF(), but
3371 	 * until msdosfs is exportable (why would you want to?), the
3372 	 * Unix defaults should be ok.
3373 	 */
3374 	pc->pc_caseinsensitive = nfs_false;
3375 	pc->pc_casepreserving = nfs_true;
3376 	nfsm_srvdone;
3377 }
3378 
3379 /*
3380  * Null operation, used by clients to ping server
3381  */
3382 /* ARGSUSED */
3383 int
3384 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3385     struct lwp *lwp, struct mbuf **mrq)
3386 {
3387 	struct mbuf *mrep = nfsd->nd_mrep;
3388 	char *bpos;
3389 	int error = NFSERR_RETVOID, cache = 0;
3390 	struct mbuf *mb, *mreq;
3391 	u_quad_t frev;
3392 
3393 	nfsm_reply(0);
3394 nfsmout:
3395 	return (0);
3396 }
3397 
3398 /*
3399  * No operation, used for obsolete procedures
3400  */
3401 /* ARGSUSED */
3402 int
3403 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3404     struct lwp *lwp, struct mbuf **mrq)
3405 {
3406 	struct mbuf *mrep = nfsd->nd_mrep;
3407 	char *bpos;
3408 	int error, cache = 0;
3409 	struct mbuf *mb, *mreq;
3410 	u_quad_t frev;
3411 
3412 	if (nfsd->nd_repstat)
3413 		error = nfsd->nd_repstat;
3414 	else
3415 		error = EPROCUNAVAIL;
3416 	nfsm_reply(0);
3417 nfsmout:
3418 	return (0);
3419 }
3420 
3421 /*
3422  * Perform access checking for vnodes obtained from file handles that would
3423  * refer to files already opened by a Unix client. You cannot just use
3424  * vn_writechk() and VOP_ACCESS() for two reasons.
3425  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3426  * 2 - The owner is to be given access irrespective of mode bits for some
3427  *     operations, so that processes that chmod after opening a file don't
3428  *     break. I don't like this because it opens a security hole, but since
3429  *     the nfs server opens a security hole the size of a barn door anyhow,
3430  *     what the heck.
3431  *
3432  * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3433  * will return EPERM instead of EACCESS. EPERM is always an error.
3434  */
3435 int
3436 nfsrv_access(struct vnode *vp, int flags, kauth_cred_t cred, int rdonly, struct lwp *lwp, int override)
3437 {
3438 	struct vattr vattr;
3439 	int error;
3440 	if (flags & VWRITE) {
3441 		/* Just vn_writechk() changed to check rdonly */
3442 		/*
3443 		 * Disallow write attempts on read-only file systems;
3444 		 * unless the file is a socket or a block or character
3445 		 * device resident on the file system.
3446 		 */
3447 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3448 			switch (vp->v_type) {
3449 			case VREG:
3450 			case VDIR:
3451 			case VLNK:
3452 				return (EROFS);
3453 			default:
3454 				break;
3455 			}
3456 		}
3457 
3458 		/*
3459 		 * If the vnode is in use as a process's text,
3460 		 * we can't allow writing.
3461 		 */
3462 		if (vp->v_iflag & VI_TEXT)
3463 			return (ETXTBSY);
3464 	}
3465 	error = VOP_GETATTR(vp, &vattr, cred);
3466 	if (error)
3467 		return (error);
3468 	error = VOP_ACCESS(vp, flags, cred);
3469 	/*
3470 	 * Allow certain operations for the owner (reads and writes
3471 	 * on files that are already open).
3472 	 */
3473 	if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid)
3474 		error = 0;
3475 	return error;
3476 }
3477