xref: /netbsd-src/sys/nfs/nfs_serv.c (revision e39ef1d61eee3ccba837ee281f1e098c864487aa)
1 /*	$NetBSD: nfs_serv.c,v 1.162 2011/11/21 09:07:59 hannken 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 1.162 2011/11/21 09:07:59 hannken 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 pathbuf *ipb = NULL;
394 	struct vnode *vp, *dirp;
395 	nfsrvfh_t nsfh;
396 	char *cp;
397 	u_int32_t *tl;
398 	int32_t t1;
399 	char *bpos;
400 	int error = 0, cache = 0, dirattr_ret = 1;
401 	uint32_t len;
402 	int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
403 	char *cp2;
404 	struct mbuf *mb, *mreq;
405 	struct vattr va, dirattr;
406 	u_quad_t frev;
407 
408 	nfsm_srvmtofh(&nsfh);
409 	nfsm_srvnamesiz(len);
410 
411 	pubflag = nfs_ispublicfh(&nsfh);
412 
413 	nd.ni_cnd.cn_cred = cred;
414 	nd.ni_cnd.cn_nameiop = LOOKUP;
415 	nd.ni_cnd.cn_flags = LOCKLEAF;
416 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
417 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
418 
419 	if (!error && pubflag) {
420 		if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL &&
421 		    (ipb = pathbuf_create(nfs_pub.np_index)) != NULL) {
422 			/*
423 			 * Setup call to lookup() to see if we can find
424 			 * the index file. Arguably, this doesn't belong
425 			 * in a kernel.. Ugh.
426 			 */
427 			ind = nd;
428 			VOP_UNLOCK(nd.ni_vp);
429 			ind.ni_pathbuf = ipb;
430 
431 			error = lookup_for_nfsd_index(&ind, nd.ni_vp);
432 			if (!error) {
433 				/*
434 				 * Found an index file. Get rid of
435 				 * the old references.
436 				 */
437 				if (dirp)
438 					vrele(dirp);
439 				dirp = nd.ni_vp;
440 				ndp = &ind;
441 			} else
442 				error = 0;
443 		}
444 		/*
445 		 * If the public filehandle was used, check that this lookup
446 		 * didn't result in a filehandle outside the publicly exported
447 		 * filesystem.
448 		 */
449 
450 		if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) {
451 			vput(nd.ni_vp);
452 			error = EPERM;
453 		}
454 	}
455 
456 	if (error) {
457 		if (nd.ni_pathbuf != NULL) {
458 			pathbuf_destroy(nd.ni_pathbuf);
459 		}
460 		if (ipb != NULL) {
461 			pathbuf_destroy(ipb);
462 		}
463 		if (dirp) {
464 			if (v3) {
465 				vn_lock(dirp, LK_SHARED | LK_RETRY);
466 				dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred);
467 				vput(dirp);
468 			} else
469 				vrele(dirp);
470 		}
471 		nfsm_reply(NFSX_POSTOPATTR(v3));
472 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
473 		return (0);
474 	}
475 
476 	nqsrv_getl(ndp->ni_startdir, ND_READ);
477 	pathbuf_destroy(nd.ni_pathbuf);
478 	if (ipb != NULL) {
479 		pathbuf_destroy(ipb);
480 	}
481 	vp = ndp->ni_vp;
482 	error = nfsrv_composefh(vp, &nsfh, v3);
483 	if (!error)
484 		error = VOP_GETATTR(vp, &va, cred);
485 	vput(vp);
486 	if (dirp) {
487 		if (v3) {
488 			vn_lock(dirp, LK_SHARED | LK_RETRY);
489 			dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred);
490 			vput(dirp);
491 		} else
492 			vrele(dirp);
493 	}
494 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPORFATTR(v3) +
495 	    NFSX_POSTOPATTR(v3));
496 	if (error) {
497 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
498 		return (0);
499 	}
500 	nfsm_srvfhtom(&nsfh, v3);
501 	if (v3) {
502 		nfsm_srvpostop_attr(0, &va);
503 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
504 	} else {
505 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
506 		nfsm_srvfillattr(&va, fp);
507 	}
508 	nfsm_srvdone;
509 }
510 
511 /*
512  * nfs readlink service
513  */
514 int
515 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
516 {
517 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
518 	struct mbuf *nam = nfsd->nd_nam;
519 	char *dpos = nfsd->nd_dpos;
520 	kauth_cred_t cred = nfsd->nd_cr;
521 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
522 	struct iovec *ivp = iv;
523 	struct mbuf *mp;
524 	u_int32_t *tl;
525 	int32_t t1;
526 	char *bpos;
527 	int error = 0, rdonly, cache = 0, i, padlen, getret;
528 	uint32_t len;
529 	int v3 = (nfsd->nd_flag & ND_NFSV3);
530 	char *cp2;
531 	struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq;
532 	struct vnode *vp;
533 	struct vattr attr;
534 	nfsrvfh_t nsfh;
535 	struct uio io, *uiop = &io;
536 	u_quad_t frev;
537 
538 	nfsm_srvmtofh(&nsfh);
539 	len = 0;
540 	i = 0;
541 	while (len < NFS_MAXPATHLEN) {
542 		mp = m_get(M_WAIT, MT_DATA);
543 		MCLAIM(mp, &nfs_mowner);
544 		m_clget(mp, M_WAIT);
545 		mp->m_len = NFSMSIZ(mp);
546 		if (len == 0)
547 			mp3 = mp2 = mp;
548 		else {
549 			mp2->m_next = mp;
550 			mp2 = mp;
551 		}
552 		if ((len+mp->m_len) > NFS_MAXPATHLEN) {
553 			mp->m_len = NFS_MAXPATHLEN-len;
554 			len = NFS_MAXPATHLEN;
555 		} else
556 			len += mp->m_len;
557 		ivp->iov_base = mtod(mp, void *);
558 		ivp->iov_len = mp->m_len;
559 		i++;
560 		ivp++;
561 	}
562 	uiop->uio_iov = iv;
563 	uiop->uio_iovcnt = i;
564 	uiop->uio_offset = 0;
565 	uiop->uio_resid = len;
566 	uiop->uio_rw = UIO_READ;
567 	UIO_SETUP_SYSSPACE(uiop);
568 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
569 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
570 	if (error) {
571 		m_freem(mp3);
572 		nfsm_reply(2 * NFSX_UNSIGNED);
573 		nfsm_srvpostop_attr(1, (struct vattr *)0);
574 		return (0);
575 	}
576 	if (vp->v_type != VLNK) {
577 		if (v3)
578 			error = EINVAL;
579 		else
580 			error = ENXIO;
581 		goto out;
582 	}
583 	nqsrv_getl(vp, ND_READ);
584 	error = VOP_READLINK(vp, uiop, cred);
585 out:
586 	getret = VOP_GETATTR(vp, &attr, cred);
587 	vput(vp);
588 	if (error)
589 		m_freem(mp3);
590 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
591 	if (v3) {
592 		nfsm_srvpostop_attr(getret, &attr);
593 		if (error)
594 			return (0);
595 	}
596 	len -= uiop->uio_resid;
597 	padlen = nfsm_padlen(len);
598 	if (uiop->uio_resid || padlen)
599 		nfs_zeropad(mp3, uiop->uio_resid, padlen);
600 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
601 	*tl = txdr_unsigned(len);
602 	mb->m_next = mp3;
603 	nfsm_srvdone;
604 }
605 
606 /*
607  * nfs read service
608  */
609 int
610 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
611 {
612 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
613 	struct mbuf *nam = nfsd->nd_nam;
614 	char *dpos = nfsd->nd_dpos;
615 	kauth_cred_t cred = nfsd->nd_cr;
616 	struct mbuf *m;
617 	struct nfs_fattr *fp;
618 	u_int32_t *tl;
619 	int32_t t1;
620 	int i;
621 	char *bpos;
622 	int error = 0, rdonly, cache = 0, getret;
623 	int v3 = (nfsd->nd_flag & ND_NFSV3);
624 	uint32_t reqlen, len, cnt, left;
625 	int padlen;
626 	char *cp2;
627 	struct mbuf *mb, *mreq;
628 	struct vnode *vp;
629 	nfsrvfh_t nsfh;
630 	struct uio io, *uiop = &io;
631 	struct vattr va;
632 	off_t off;
633 	u_quad_t frev;
634 
635 	nfsm_srvmtofh(&nsfh);
636 	if (v3) {
637 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
638 		off = fxdr_hyper(tl);
639 	} else {
640 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
641 		off = (off_t)fxdr_unsigned(u_int32_t, *tl);
642 	}
643 	nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
644 	reqlen = fxdr_unsigned(uint32_t, *tl);
645 	reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd));
646 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
647 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
648 	if (error) {
649 		nfsm_reply(2 * NFSX_UNSIGNED);
650 		nfsm_srvpostop_attr(1, (struct vattr *)0);
651 		return (0);
652 	}
653 	if (vp->v_type != VREG) {
654 		if (v3)
655 			error = EINVAL;
656 		else
657 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
658 	}
659 	if (!error) {
660 	    nqsrv_getl(vp, ND_READ);
661 	    if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0)
662 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1);
663 	}
664 	getret = VOP_GETATTR(vp, &va, cred);
665 	if (!error)
666 		error = getret;
667 	if (error) {
668 		vput(vp);
669 		nfsm_reply(NFSX_POSTOPATTR(v3));
670 		nfsm_srvpostop_attr(getret, &va);
671 		return (0);
672 	}
673 	if (off >= va.va_size)
674 		cnt = 0;
675 	else if ((off + reqlen) > va.va_size)
676 		cnt = va.va_size - off;
677 	else
678 		cnt = reqlen;
679 	nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
680 	if (v3) {
681 		nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
682 		*tl++ = nfs_true;
683 		fp = (struct nfs_fattr *)tl;
684 		tl += (NFSX_V3FATTR / sizeof (u_int32_t));
685 	} else {
686 		nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
687 		fp = (struct nfs_fattr *)tl;
688 		tl += (NFSX_V2FATTR / sizeof (u_int32_t));
689 	}
690 	len = left = cnt;
691 	if (cnt > 0) {
692 		if (nfsd_use_loan) {
693 			struct vm_page **pgpp;
694 			voff_t pgoff = trunc_page(off);
695 			int npages;
696 			vaddr_t lva;
697 
698 			npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT;
699 			KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */
700 
701 			/* allocate kva for mbuf data */
702 			lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so);
703 			if (lva == 0) {
704 				/* fall back to VOP_READ */
705 				goto loan_fail;
706 			}
707 
708 			/* allocate mbuf */
709 			m = m_get(M_WAIT, MT_DATA);
710 			MCLAIM(m, &nfs_mowner);
711 			pgpp = m->m_ext.ext_pgs;
712 
713 			/* loan pages */
714 			error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages,
715 			    pgpp);
716 			if (error) {
717 				sokvafree(lva, npages << PAGE_SHIFT);
718 				m_free(m);
719 				if (error == EBUSY)
720 					goto loan_fail;
721 				goto read_error;
722 			}
723 
724 			/* associate kva to mbuf */
725 			MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)),
726 			    cnt, M_MBUF, soloanfree, slp->ns_so);
727 			m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
728 			m->m_len = cnt;
729 
730 			/* map pages */
731 			for (i = 0; i < npages; i++) {
732 				pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]),
733 				    VM_PROT_READ, 0);
734 				lva += PAGE_SIZE;
735 			}
736 
737 			pmap_update(pmap_kernel());
738 
739 			mb->m_next = m;
740 			mb = m;
741 			error = 0;
742 			uiop->uio_resid = 0;
743 		} else {
744 			struct iovec *iv;
745 			struct iovec *iv2;
746 			struct mbuf *m2;
747 			int siz;
748 loan_fail:
749 			/*
750 			 * Generate the mbuf list with the uio_iov ref. to it.
751 			 */
752 			i = 0;
753 			m = m2 = mb;
754 			while (left > 0) {
755 				siz = min(M_TRAILINGSPACE(m), left);
756 				if (siz > 0) {
757 					left -= siz;
758 					i++;
759 				}
760 				if (left > 0) {
761 					m = m_get(M_WAIT, MT_DATA);
762 					MCLAIM(m, &nfs_mowner);
763 					m_clget(m, M_WAIT);
764 					m->m_len = 0;
765 					m2->m_next = m;
766 					m2 = m;
767 				}
768 			}
769 			iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK);
770 			uiop->uio_iov = iv2 = iv;
771 			m = mb;
772 			left = cnt;
773 			i = 0;
774 			while (left > 0) {
775 				if (m == NULL)
776 					panic("nfsrv_read iov");
777 				siz = min(M_TRAILINGSPACE(m), left);
778 				if (siz > 0) {
779 					iv->iov_base = mtod(m, char *) +
780 					    m->m_len;
781 					iv->iov_len = siz;
782 					m->m_len += siz;
783 					left -= siz;
784 					iv++;
785 					i++;
786 				}
787 				m = m->m_next;
788 			}
789 			uiop->uio_iovcnt = i;
790 			uiop->uio_offset = off;
791 			uiop->uio_resid = cnt;
792 			uiop->uio_rw = UIO_READ;
793 			UIO_SETUP_SYSSPACE(uiop);
794 			error = VOP_READ(vp, uiop, IO_NODELOCKED, cred);
795 			free((void *)iv2, M_TEMP);
796 		}
797 read_error:
798 		if (error || (getret = VOP_GETATTR(vp, &va, cred)) != 0){
799 			if (!error)
800 				error = getret;
801 			m_freem(mreq);
802 			vput(vp);
803 			nfsm_reply(NFSX_POSTOPATTR(v3));
804 			nfsm_srvpostop_attr(getret, &va);
805 			return (0);
806 		}
807 	} else {
808 		uiop->uio_resid = 0;
809 	}
810 	vput(vp);
811 	nfsm_srvfillattr(&va, fp);
812 	len -= uiop->uio_resid;
813 	padlen = nfsm_padlen(len);
814 	if (uiop->uio_resid || padlen)
815 		nfs_zeropad(mb, uiop->uio_resid, padlen);
816 	if (v3) {
817 		/* count */
818 		*tl++ = txdr_unsigned(len);
819 		/* eof */
820 		if (off + len >= va.va_size)
821 			*tl++ = nfs_true;
822 		else
823 			*tl++ = nfs_false;
824 	}
825 	*tl = txdr_unsigned(len);
826 	nfsm_srvdone;
827 }
828 
829 /*
830  * nfs write service
831  */
832 int
833 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
834 {
835 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
836 	struct mbuf *nam = nfsd->nd_nam;
837 	char *dpos = nfsd->nd_dpos;
838 	kauth_cred_t cred = nfsd->nd_cr;
839 	struct iovec *ivp;
840 	int i, cnt;
841 	struct mbuf *mp;
842 	struct nfs_fattr *fp;
843 	struct iovec *iv;
844 	struct vattr va, forat;
845 	u_int32_t *tl;
846 	int32_t t1;
847 	char *bpos;
848 	int error = 0, rdonly, cache = 0, len, forat_ret = 1;
849 	int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
850 	int stable = NFSV3WRITE_FILESYNC;
851 	int v3 = (nfsd->nd_flag & ND_NFSV3);
852 	char *cp2;
853 	struct mbuf *mb, *mreq;
854 	struct vnode *vp;
855 	nfsrvfh_t nsfh;
856 	struct uio io, *uiop = &io;
857 	off_t off;
858 	u_quad_t frev;
859 
860 	if (mrep == NULL) {
861 		*mrq = NULL;
862 		return (0);
863 	}
864 	nfsm_srvmtofh(&nsfh);
865 	if (v3) {
866 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
867 		off = fxdr_hyper(tl);
868 		tl += 3;
869 		stable = fxdr_unsigned(int, *tl++);
870 	} else {
871 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
872 		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
873 		tl += 2;
874 	}
875 	retlen = len = fxdr_unsigned(int32_t, *tl);
876 	cnt = i = 0;
877 
878 	/*
879 	 * For NFS Version 2, it is not obvious what a write of zero length
880 	 * should do, but I might as well be consistent with Version 3,
881 	 * which is to return ok so long as there are no permission problems.
882 	 */
883 	if (len > 0) {
884 		zeroing = 1;
885 		mp = mrep;
886 		while (mp) {
887 			if (mp == md) {
888 				zeroing = 0;
889 				adjust = dpos - mtod(mp, char *);
890 				mp->m_len -= adjust;
891 				if (mp->m_len > 0 && adjust > 0)
892 					NFSMADV(mp, adjust);
893 			}
894 			if (zeroing)
895 				mp->m_len = 0;
896 			else if (mp->m_len > 0) {
897 				i += mp->m_len;
898 				if (i > len) {
899 					mp->m_len -= (i - len);
900 					zeroing	= 1;
901 				}
902 				if (mp->m_len > 0)
903 					cnt++;
904 			}
905 			mp = mp->m_next;
906 		}
907 	}
908 	if (len > NFS_MAXDATA || len < 0 || i < len) {
909 		error = EIO;
910 		nfsm_reply(2 * NFSX_UNSIGNED);
911 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
912 		return (0);
913 	}
914 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
915 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
916 	if (error) {
917 		nfsm_reply(2 * NFSX_UNSIGNED);
918 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
919 		return (0);
920 	}
921 	if (v3)
922 		forat_ret = VOP_GETATTR(vp, &forat, cred);
923 	if (vp->v_type != VREG) {
924 		if (v3)
925 			error = EINVAL;
926 		else
927 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
928 	}
929 	if (!error) {
930 		nqsrv_getl(vp, ND_WRITE);
931 		error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1);
932 	}
933 	if (error) {
934 		vput(vp);
935 		nfsm_reply(NFSX_WCCDATA(v3));
936 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
937 		return (0);
938 	}
939 
940 	if (len > 0) {
941 		ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK);
942 		uiop->uio_iov = iv = ivp;
943 		uiop->uio_iovcnt = cnt;
944 		mp = mrep;
945 		while (mp) {
946 			if (mp->m_len > 0) {
947 				ivp->iov_base = mtod(mp, void *);
948 				ivp->iov_len = mp->m_len;
949 				ivp++;
950 			}
951 			mp = mp->m_next;
952 		}
953 
954 		/*
955 		 * XXX
956 		 * The IO_METASYNC flag indicates that all metadata (and not
957 		 * just enough to ensure data integrity) must be written to
958 		 * stable storage synchronously.
959 		 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
960 		 */
961 		if (stable == NFSV3WRITE_UNSTABLE)
962 			ioflags = IO_NODELOCKED;
963 		else if (stable == NFSV3WRITE_DATASYNC)
964 			ioflags = (IO_SYNC | IO_NODELOCKED);
965 		else
966 			ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
967 		uiop->uio_resid = len;
968 		uiop->uio_rw = UIO_WRITE;
969 		uiop->uio_offset = off;
970 		UIO_SETUP_SYSSPACE(uiop);
971 		error = VOP_WRITE(vp, uiop, ioflags, cred);
972 		nfsstats.srvvop_writes++;
973 		free(iv, M_TEMP);
974 	}
975 	aftat_ret = VOP_GETATTR(vp, &va, cred);
976 	vput(vp);
977 	if (!error)
978 		error = aftat_ret;
979 	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
980 		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
981 	if (v3) {
982 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
983 		if (error)
984 			return (0);
985 		nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
986 		*tl++ = txdr_unsigned(retlen);
987 		if (stable == NFSV3WRITE_UNSTABLE)
988 			*tl++ = txdr_unsigned(stable);
989 		else
990 			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
991 		/*
992 		 * Actually, there is no need to txdr these fields,
993 		 * but it may make the values more human readable,
994 		 * for debugging purposes.
995 		 */
996 		*tl++ = txdr_unsigned(boottime.tv_sec);
997 		*tl = txdr_unsigned(boottime.tv_nsec / 1000);
998 	} else {
999 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1000 		nfsm_srvfillattr(&va, fp);
1001 	}
1002 	nfsm_srvdone;
1003 }
1004 
1005 /*
1006  * XXX elad: the original NFSW_SAMECRED() macro also made sure the
1007  *	     two nd_flag fields of the descriptors contained
1008  *	     ND_KERBAUTH.
1009  */
1010 static int
1011 nfsrv_samecred(kauth_cred_t cred1, kauth_cred_t cred2)
1012 {
1013 	int i, do_ngroups;
1014 
1015 	if (kauth_cred_geteuid(cred1) != kauth_cred_geteuid(cred2))
1016 		return (0);
1017 	if (kauth_cred_ngroups(cred1) != kauth_cred_ngroups(cred2))
1018 		return (0);
1019 	do_ngroups = kauth_cred_ngroups(cred1);
1020 	for (i = 0; i < do_ngroups; i++)
1021 		if (kauth_cred_group(cred1, i) !=
1022 		    kauth_cred_group(cred2, i))
1023 			return (0);
1024 
1025 	return (1);
1026 }
1027 
1028 static struct nfsrvw_delayhash *
1029 nfsrv_nwdelayhash(struct nfssvc_sock *slp, const nfsrvfh_t *nsfh)
1030 {
1031 	uint32_t hash;
1032 
1033 	hash = hash32_buf(NFSRVFH_DATA(nsfh), NFSRVFH_SIZE(nsfh),
1034 	    HASH32_BUF_INIT);
1035 	return &slp->ns_wdelayhashtbl[hash % NFS_WDELAYHASHSIZ];
1036 }
1037 
1038 /*
1039  * NFS write service with write gathering support. Called when
1040  * nfsrvw_procrastinate > 0.
1041  * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1042  * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1043  * Jan. 1994.
1044  */
1045 int
1046 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1047 {
1048 	struct timeval now;
1049 	struct iovec *ivp;
1050 	struct mbuf *mp;
1051 	struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1052 	struct nfs_fattr *fp;
1053 	int i = 0;
1054 	struct iovec *iov;
1055 	struct nfsrvw_delayhash *wpp;
1056 	kauth_cred_t cred;
1057 	struct vattr va, forat;
1058 	u_int32_t *tl;
1059 	int32_t t1;
1060 	char *bpos, *dpos;
1061 	int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1;
1062 	int ioflags, aftat_ret = 1, adjust, v3, zeroing;
1063 	char *cp2;
1064 	struct mbuf *mb, *mreq, *mrep, *md;
1065 	struct vnode *vp;
1066 	struct uio io, *uiop = &io;
1067 	u_quad_t frev, cur_usec;
1068 
1069 	*mrq = NULL;
1070 	if (*ndp) {
1071 	    nfsd = *ndp;
1072 	    *ndp = NULL;
1073 	    mrep = nfsd->nd_mrep;
1074 	    md = nfsd->nd_md;
1075 	    dpos = nfsd->nd_dpos;
1076 	    cred = nfsd->nd_cr;
1077 	    v3 = (nfsd->nd_flag & ND_NFSV3);
1078 	    LIST_INIT(&nfsd->nd_coalesce);
1079 	    nfsd->nd_mreq = NULL;
1080 	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1081 	    getmicrotime(&now);
1082 	    cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec;
1083 	    nfsd->nd_time = cur_usec + nfsrvw_procrastinate;
1084 
1085 	    /*
1086 	     * Now, get the write header..
1087 	     */
1088 	    nfsm_srvmtofh(&nfsd->nd_fh);
1089 	    if (v3) {
1090 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1091 		nfsd->nd_off = fxdr_hyper(tl);
1092 		tl += 3;
1093 		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1094 	    } else {
1095 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1096 		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1097 		tl += 2;
1098 	    }
1099 	    len = fxdr_unsigned(int32_t, *tl);
1100 	    nfsd->nd_len = len;
1101 	    nfsd->nd_eoff = nfsd->nd_off + len;
1102 
1103 	    /*
1104 	     * Trim the header out of the mbuf list and trim off any trailing
1105 	     * junk so that the mbuf list has only the write data.
1106 	     */
1107 	    zeroing = 1;
1108 	    i = 0;
1109 	    mp = mrep;
1110 	    while (mp) {
1111 		if (mp == md) {
1112 		    zeroing = 0;
1113 		    adjust = dpos - mtod(mp, char *);
1114 		    mp->m_len -= adjust;
1115 		    if (mp->m_len > 0 && adjust > 0)
1116 			NFSMADV(mp, adjust);
1117 		}
1118 		if (zeroing)
1119 		    mp->m_len = 0;
1120 		else {
1121 		    i += mp->m_len;
1122 		    if (i > len) {
1123 			mp->m_len -= (i - len);
1124 			zeroing = 1;
1125 		    }
1126 		}
1127 		mp = mp->m_next;
1128 	    }
1129 	    if (len > NFS_MAXDATA || len < 0  || i < len) {
1130 nfsmout:
1131 		m_freem(mrep);
1132 		error = EIO;
1133 		nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1134 		if (v3)
1135 		    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1136 		nfsd->nd_mreq = mreq;
1137 		nfsd->nd_mrep = NULL;
1138 		nfsd->nd_time = 0;
1139 	    }
1140 
1141 	    /*
1142 	     * Add this entry to the hash and time queues.
1143 	     */
1144 	    owp = NULL;
1145 	    mutex_enter(&nfsd_lock);
1146 	    wp = LIST_FIRST(&slp->ns_tq);
1147 	    while (wp && wp->nd_time < nfsd->nd_time) {
1148 		owp = wp;
1149 		wp = LIST_NEXT(wp, nd_tq);
1150 	    }
1151 	    if (owp) {
1152 		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1153 	    } else {
1154 		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1155 	    }
1156 	    if (nfsd->nd_mrep) {
1157 		wpp = nfsrv_nwdelayhash(slp, &nfsd->nd_fh);
1158 		owp = NULL;
1159 		wp = LIST_FIRST(wpp);
1160 		while (wp && nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) {
1161 		    owp = wp;
1162 		    wp = LIST_NEXT(wp, nd_hash);
1163 		}
1164 		while (wp && wp->nd_off < nfsd->nd_off &&
1165 		    !nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) {
1166 		    owp = wp;
1167 		    wp = LIST_NEXT(wp, nd_hash);
1168 		}
1169 		if (owp) {
1170 		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1171 
1172 		    /*
1173 		     * Search the hash list for overlapping entries and
1174 		     * coalesce.
1175 		     */
1176 		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1177 			wp = LIST_NEXT(nfsd, nd_hash);
1178 			if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr))
1179 			    nfsrvw_coalesce(owp, nfsd);
1180 		    }
1181 		} else {
1182 		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1183 		}
1184 	    }
1185 	    mutex_exit(&nfsd_lock);
1186 	}
1187 
1188 	/*
1189 	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1190 	 * and generate the associated reply mbuf list(s).
1191 	 */
1192 loop1:
1193 	getmicrotime(&now);
1194 	cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec;
1195 	mutex_enter(&nfsd_lock);
1196 	for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1197 		owp = LIST_NEXT(nfsd, nd_tq);
1198 		if (nfsd->nd_time > cur_usec)
1199 		    break;
1200 		if (nfsd->nd_mreq)
1201 		    continue;
1202 		LIST_REMOVE(nfsd, nd_tq);
1203 		LIST_REMOVE(nfsd, nd_hash);
1204 		mutex_exit(&nfsd_lock);
1205 
1206 		mrep = nfsd->nd_mrep;
1207 		nfsd->nd_mrep = NULL;
1208 		cred = nfsd->nd_cr;
1209 		v3 = (nfsd->nd_flag & ND_NFSV3);
1210 		forat_ret = aftat_ret = 1;
1211 		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1212 		    nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH),
1213 		    false);
1214 		if (!error) {
1215 		    if (v3)
1216 			forat_ret = VOP_GETATTR(vp, &forat, cred);
1217 		    if (vp->v_type != VREG) {
1218 			if (v3)
1219 			    error = EINVAL;
1220 			else
1221 			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1222 		    }
1223 		} else
1224 		    vp = NULL;
1225 		if (!error) {
1226 		    nqsrv_getl(vp, ND_WRITE);
1227 		    error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1);
1228 		}
1229 
1230 		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1231 		    ioflags = IO_NODELOCKED;
1232 		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1233 		    ioflags = (IO_SYNC | IO_NODELOCKED);
1234 		else
1235 		    ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1236 		uiop->uio_rw = UIO_WRITE;
1237 		uiop->uio_offset = nfsd->nd_off;
1238 		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1239 		UIO_SETUP_SYSSPACE(uiop);
1240 		if (uiop->uio_resid > 0) {
1241 		    mp = mrep;
1242 		    i = 0;
1243 		    while (mp) {
1244 			if (mp->m_len > 0)
1245 			    i++;
1246 			mp = mp->m_next;
1247 		    }
1248 		    uiop->uio_iovcnt = i;
1249 		    iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1250 		    uiop->uio_iov = ivp = iov;
1251 		    mp = mrep;
1252 		    while (mp) {
1253 			if (mp->m_len > 0) {
1254 			    ivp->iov_base = mtod(mp, void *);
1255 			    ivp->iov_len = mp->m_len;
1256 			    ivp++;
1257 			}
1258 			mp = mp->m_next;
1259 		    }
1260 		    if (!error) {
1261 			error = VOP_WRITE(vp, uiop, ioflags, cred);
1262 			nfsstats.srvvop_writes++;
1263 		    }
1264 		    free((void *)iov, M_TEMP);
1265 		}
1266 		m_freem(mrep);
1267 		if (vp) {
1268 		    aftat_ret = VOP_GETATTR(vp, &va, cred);
1269 		    vput(vp);
1270 		}
1271 
1272 		/*
1273 		 * Loop around generating replies for all write rpcs that have
1274 		 * now been completed.
1275 		 */
1276 		swp = nfsd;
1277 		do {
1278 		    if (error) {
1279 			nfsm_writereply(NFSX_WCCDATA(v3), v3);
1280 			if (v3) {
1281 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1282 			}
1283 		    } else {
1284 			nfsm_writereply(NFSX_PREOPATTR(v3) +
1285 			    NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1286 			    NFSX_WRITEVERF(v3), v3);
1287 			if (v3) {
1288 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1289 			    nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1290 			    *tl++ = txdr_unsigned(nfsd->nd_len);
1291 			    *tl++ = txdr_unsigned(swp->nd_stable);
1292 			    /*
1293 			     * Actually, there is no need to txdr these fields,
1294 			     * but it may make the values more human readable,
1295 			     * for debugging purposes.
1296 			     */
1297 			    *tl++ = txdr_unsigned(boottime.tv_sec);
1298 			    *tl = txdr_unsigned(boottime.tv_nsec / 1000);
1299 			} else {
1300 			    nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1301 			    nfsm_srvfillattr(&va, fp);
1302 			}
1303 		    }
1304 		    nfsd->nd_mreq = mreq;
1305 		    if (nfsd->nd_mrep)
1306 			panic("nfsrv_write: nd_mrep not free");
1307 
1308 		    /*
1309 		     * Done. Put it at the head of the timer queue so that
1310 		     * the final phase can return the reply.
1311 		     */
1312 		    mutex_enter(&nfsd_lock);
1313 		    if (nfsd != swp) {
1314 			nfsd->nd_time = 0;
1315 			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1316 		    }
1317 		    nfsd = LIST_FIRST(&swp->nd_coalesce);
1318 		    if (nfsd) {
1319 			LIST_REMOVE(nfsd, nd_tq);
1320 		    }
1321 		    mutex_exit(&nfsd_lock);
1322 		} while (nfsd);
1323 		swp->nd_time = 0;
1324 
1325 		mutex_enter(&nfsd_lock);
1326 		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1327 		mutex_exit(&nfsd_lock);
1328 		goto loop1;
1329 	}
1330 	mutex_exit(&nfsd_lock);
1331 	nfs_timer_start();
1332 
1333 	/*
1334 	 * Search for a reply to return.
1335 	 */
1336 	mutex_enter(&nfsd_lock);
1337 	LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) {
1338 		if (nfsd->nd_mreq) {
1339 		    LIST_REMOVE(nfsd, nd_tq);
1340 		    *mrq = nfsd->nd_mreq;
1341 		    *ndp = nfsd;
1342 		    break;
1343 		}
1344 	}
1345 	mutex_exit(&nfsd_lock);
1346 	return (0);
1347 }
1348 
1349 /*
1350  * Coalesce the write request nfsd into owp. To do this we must:
1351  * - remove nfsd from the queues
1352  * - merge nfsd->nd_mrep into owp->nd_mrep
1353  * - update the nd_eoff and nd_stable for owp
1354  * - put nfsd on owp's nd_coalesce list
1355  * NB: Must be called at splsoftclock().
1356  */
1357 void
1358 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1359 {
1360         int overlap;
1361         struct mbuf *mp;
1362 	struct nfsrv_descript *m;
1363 
1364 	KASSERT(mutex_owned(&nfsd_lock));
1365 
1366         LIST_REMOVE(nfsd, nd_hash);
1367         LIST_REMOVE(nfsd, nd_tq);
1368         if (owp->nd_eoff < nfsd->nd_eoff) {
1369             overlap = owp->nd_eoff - nfsd->nd_off;
1370             if (overlap < 0)
1371                 panic("nfsrv_coalesce: bad off");
1372             if (overlap > 0)
1373                 m_adj(nfsd->nd_mrep, overlap);
1374             mp = owp->nd_mrep;
1375             while (mp->m_next)
1376                 mp = mp->m_next;
1377             mp->m_next = nfsd->nd_mrep;
1378             owp->nd_eoff = nfsd->nd_eoff;
1379         } else
1380             m_freem(nfsd->nd_mrep);
1381         nfsd->nd_mrep = NULL;
1382         if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1383             owp->nd_stable = NFSV3WRITE_FILESYNC;
1384         else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1385             owp->nd_stable == NFSV3WRITE_UNSTABLE)
1386             owp->nd_stable = NFSV3WRITE_DATASYNC;
1387         LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1388  	/*
1389  	 * nfsd might hold coalesce elements! Move them to owp.
1390  	 * Otherwise, requests may be lost and clients will be stuck.
1391  	 */
1392 	while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) {
1393 		LIST_REMOVE(m, nd_tq);
1394 		LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
1395 	}
1396 }
1397 
1398 /*
1399  * nfs create service
1400  * now does a truncate to 0 length via. setattr if it already exists
1401  */
1402 int
1403 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1404 {
1405 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1406 	struct mbuf *nam = nfsd->nd_nam;
1407 	char *dpos = nfsd->nd_dpos;
1408 	kauth_cred_t cred = nfsd->nd_cr;
1409 	struct nfs_fattr *fp;
1410 	struct vattr va, dirfor, diraft;
1411 	struct nfsv2_sattr *sp;
1412 	u_int32_t *tl;
1413 	struct nameidata nd;
1414 	char *cp;
1415 	int32_t t1;
1416 	char *bpos;
1417 	int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1418 	int rdev = 0, abort = 0;
1419 	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1420 	char *cp2;
1421 	struct mbuf *mb, *mreq;
1422 	struct vnode *vp = NULL, *dirp = NULL;
1423 	nfsrvfh_t nsfh;
1424 	u_quad_t frev, tempsize;
1425 	u_char cverf[NFSX_V3CREATEVERF];
1426 
1427 	nd.ni_cnd.cn_nameiop = 0;
1428 	nfsm_srvmtofh(&nsfh);
1429 	nfsm_srvnamesiz(len);
1430 	nd.ni_cnd.cn_cred = cred;
1431 	nd.ni_cnd.cn_nameiop = CREATE;
1432 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1433 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1434 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1435 	if (dirp && v3) {
1436 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1437 	}
1438 	if (error) {
1439 		nfsm_reply(NFSX_WCCDATA(v3));
1440 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1441 		if (dirp)
1442 			vrele(dirp);
1443 		if (nd.ni_pathbuf != NULL) {
1444 			pathbuf_destroy(nd.ni_pathbuf);
1445 			nd.ni_pathbuf = NULL;
1446 		}
1447 		return (0);
1448 	}
1449 	abort = 1;
1450 	vattr_null(&va);
1451 	if (v3) {
1452 		va.va_mode = 0;
1453 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1454 		how = fxdr_unsigned(int, *tl);
1455 		switch (how) {
1456 		case NFSV3CREATE_GUARDED:
1457 			if (nd.ni_vp) {
1458 				error = EEXIST;
1459 				break;
1460 			}
1461 		case NFSV3CREATE_UNCHECKED:
1462 			nfsm_srvsattr(&va);
1463 			break;
1464 		case NFSV3CREATE_EXCLUSIVE:
1465 			nfsm_dissect(cp, void *, NFSX_V3CREATEVERF);
1466 			memcpy(cverf, cp, NFSX_V3CREATEVERF);
1467 			exclusive_flag = 1;
1468 			break;
1469 		};
1470 		va.va_type = VREG;
1471 	} else {
1472 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1473 		va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1474 		if (va.va_type == VNON)
1475 			va.va_type = VREG;
1476 		va.va_mode = nfstov_mode(sp->sa_mode);
1477 		switch (va.va_type) {
1478 		case VREG:
1479 			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1480 			if (tsize != -1)
1481 				va.va_size = (u_quad_t)tsize;
1482 			break;
1483 		case VCHR:
1484 		case VBLK:
1485 		case VFIFO:
1486 			rdev = fxdr_unsigned(int32_t, sp->sa_size);
1487 			break;
1488 		default:
1489 			break;
1490 		};
1491 	}
1492 
1493 	/*
1494 	 * Iff doesn't exist, create it
1495 	 * otherwise just truncate to 0 length
1496 	 *   should I set the mode too ??
1497 	 */
1498 	if (nd.ni_vp == NULL) {
1499 		if (va.va_type == VREG || va.va_type == VSOCK) {
1500 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1501 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1502 			if (!error) {
1503 				if (exclusive_flag) {
1504 					exclusive_flag = 0;
1505 					vattr_null(&va);
1506 					/*
1507 					 * XXX
1508 					 * assuming NFSX_V3CREATEVERF
1509 					 * == sizeof(nfstime3)
1510 					 */
1511 					fxdr_nfsv3time(cverf, &va.va_atime);
1512 					error = VOP_SETATTR(nd.ni_vp, &va,
1513 						cred);
1514 				}
1515 			}
1516 		} else if (va.va_type == VCHR || va.va_type == VBLK ||
1517 			va.va_type == VFIFO) {
1518 			if (va.va_type == VCHR && rdev == 0xffffffff)
1519 				va.va_type = VFIFO;
1520 			if (va.va_type != VFIFO &&
1521 			    (error = kauth_authorize_system(cred,
1522 			    KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) {
1523 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1524 				vput(nd.ni_dvp);
1525 				abort = 0;
1526 				nfsm_reply(0);
1527 				if (nd.ni_pathbuf != NULL) {
1528 					pathbuf_destroy(nd.ni_pathbuf);
1529 					nd.ni_pathbuf = NULL;
1530 				}
1531 				return (error);
1532 			} else
1533 				va.va_rdev = (dev_t)rdev;
1534 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1535 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd,
1536 			    &va);
1537 			if (error) {
1538 				nfsm_reply(0);
1539 			}
1540 		} else {
1541 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1542 			if (nd.ni_pathbuf != NULL) {
1543 				pathbuf_destroy(nd.ni_pathbuf);
1544 				nd.ni_pathbuf = NULL;
1545 			}
1546 			vput(nd.ni_dvp);
1547 			error = ENXIO;
1548 			abort = 0;
1549 		}
1550 		vp = nd.ni_vp;
1551 	} else {
1552 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1553 		if (nd.ni_pathbuf != NULL) {
1554 			pathbuf_destroy(nd.ni_pathbuf);
1555 			nd.ni_pathbuf = NULL;
1556 		}
1557 		vp = nd.ni_vp;
1558 		if (nd.ni_dvp == vp)
1559 			vrele(nd.ni_dvp);
1560 		else
1561 			vput(nd.ni_dvp);
1562 		abort = 0;
1563 		if (!error && va.va_size != -1) {
1564 			error = nfsrv_access(vp, VWRITE, cred,
1565 			    (nd.ni_cnd.cn_flags & RDONLY), lwp, 0);
1566 			if (!error) {
1567 				nqsrv_getl(vp, ND_WRITE);
1568 				tempsize = va.va_size;
1569 				vattr_null(&va);
1570 				va.va_size = tempsize;
1571 				error = VOP_SETATTR(vp, &va, cred);
1572 			}
1573 		}
1574 		if (error)
1575 			vput(vp);
1576 	}
1577 	if (!error) {
1578 		error = nfsrv_composefh(vp, &nsfh, v3);
1579 		if (!error)
1580 			error = VOP_GETATTR(vp, &va, cred);
1581 		vput(vp);
1582 	}
1583 	if (v3) {
1584 		if (exclusive_flag && !error) {
1585 			/*
1586 			 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3)
1587 			 */
1588 			char oldverf[NFSX_V3CREATEVERF];
1589 
1590 			txdr_nfsv3time(&va.va_atime, oldverf);
1591 			if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF))
1592 				error = EEXIST;
1593 		}
1594 		if (dirp) {
1595 			vn_lock(dirp, LK_SHARED | LK_RETRY);
1596 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1597 			VOP_UNLOCK(dirp);
1598 		}
1599 	}
1600 	if (dirp) {
1601 		vrele(dirp);
1602 		dirp = NULL;
1603 	}
1604 	if (nd.ni_pathbuf != NULL) {
1605 		pathbuf_destroy(nd.ni_pathbuf);
1606 		nd.ni_pathbuf = NULL;
1607 	}
1608 	abort = 0;
1609 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1610 	if (v3) {
1611 		if (!error) {
1612 			nfsm_srvpostop_fh(&nsfh);
1613 			nfsm_srvpostop_attr(0, &va);
1614 		}
1615 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1616 	} else {
1617 		nfsm_srvfhtom(&nsfh, v3);
1618 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1619 		nfsm_srvfillattr(&va, fp);
1620 	}
1621 	return (0);
1622 nfsmout:
1623 	if (dirp)
1624 		vrele(dirp);
1625 	if (abort) {
1626 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1627 		if (nd.ni_dvp == nd.ni_vp)
1628 			vrele(nd.ni_dvp);
1629 		else
1630 			vput(nd.ni_dvp);
1631 		if (nd.ni_vp)
1632 			vput(nd.ni_vp);
1633 	}
1634 	if (nd.ni_pathbuf != NULL) {
1635 		pathbuf_destroy(nd.ni_pathbuf);
1636 		nd.ni_pathbuf = NULL;
1637 	}
1638 	return (error);
1639 }
1640 
1641 /*
1642  * nfs v3 mknod service
1643  */
1644 int
1645 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1646 {
1647 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1648 	struct mbuf *nam = nfsd->nd_nam;
1649 	char *dpos = nfsd->nd_dpos;
1650 	kauth_cred_t cred = nfsd->nd_cr;
1651 	struct vattr va, dirfor, diraft;
1652 	u_int32_t *tl;
1653 	struct nameidata nd;
1654 	int32_t t1;
1655 	char *bpos;
1656 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
1657 	int abort = 0;
1658 	u_int32_t major, minor;
1659 	enum vtype vtyp;
1660 	char *cp2;
1661 	struct mbuf *mb, *mreq;
1662 	struct vnode *vp, *dirp = (struct vnode *)0;
1663 	nfsrvfh_t nsfh;
1664 	u_quad_t frev;
1665 
1666 	nd.ni_cnd.cn_nameiop = 0;
1667 	nfsm_srvmtofh(&nsfh);
1668 	nfsm_srvnamesiz(len);
1669 	nd.ni_cnd.cn_cred = cred;
1670 	nd.ni_cnd.cn_nameiop = CREATE;
1671 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1672 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1673 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1674 	if (dirp)
1675 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1676 	if (error) {
1677 		nfsm_reply(NFSX_WCCDATA(1));
1678 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1679 		if (dirp)
1680 			vrele(dirp);
1681 		if (nd.ni_pathbuf != NULL) {
1682 			pathbuf_destroy(nd.ni_pathbuf);
1683 			nd.ni_pathbuf = NULL;
1684 		}
1685 		return (0);
1686 	}
1687 	abort = 1;
1688 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1689 	vtyp = nfsv3tov_type(*tl);
1690 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1691 		error = NFSERR_BADTYPE;
1692 		goto abort;
1693 	}
1694 	vattr_null(&va);
1695 	va.va_mode = 0;
1696 	nfsm_srvsattr(&va);
1697 	if (vtyp == VCHR || vtyp == VBLK) {
1698 		dev_t rdev;
1699 
1700 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1701 		major = fxdr_unsigned(u_int32_t, *tl++);
1702 		minor = fxdr_unsigned(u_int32_t, *tl);
1703 		rdev = makedev(major, minor);
1704 		if (major(rdev) != major || minor(rdev) != minor) {
1705 			error = EINVAL;
1706 			goto abort;
1707 		}
1708 		va.va_rdev = rdev;
1709 	}
1710 
1711 	/*
1712 	 * Iff doesn't exist, create it.
1713 	 */
1714 	if (nd.ni_vp) {
1715 		error = EEXIST;
1716 abort:
1717 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1718 		if (nd.ni_dvp == nd.ni_vp)
1719 			vrele(nd.ni_dvp);
1720 		else
1721 			vput(nd.ni_dvp);
1722 		if (nd.ni_vp)
1723 			vput(nd.ni_vp);
1724 		if (nd.ni_pathbuf != NULL) {
1725 			pathbuf_destroy(nd.ni_pathbuf);
1726 			nd.ni_pathbuf = NULL;
1727 		}
1728 		goto out;
1729 	}
1730 	va.va_type = vtyp;
1731 	if (vtyp == VSOCK) {
1732 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1733 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1734 	} else {
1735 		if (va.va_type != VFIFO &&
1736 		    (error = kauth_authorize_system(cred,
1737 		    KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) {
1738 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1739 			vput(nd.ni_dvp);
1740 			goto out;
1741 		}
1742 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1743 		error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1744 		if (error)
1745 			goto out;
1746 	}
1747 out:
1748 	vp = nd.ni_vp;
1749 	if (!error) {
1750 		error = nfsrv_composefh(vp, &nsfh, true);
1751 		if (!error)
1752 			error = VOP_GETATTR(vp, &va, cred);
1753 		vput(vp);
1754 	}
1755 	if (dirp) {
1756 		vn_lock(dirp, LK_SHARED | LK_RETRY);
1757 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1758 		VOP_UNLOCK(dirp);
1759 		vrele(dirp);
1760 		dirp = NULL;
1761 	}
1762 	if (nd.ni_pathbuf != NULL) {
1763 		pathbuf_destroy(nd.ni_pathbuf);
1764 		nd.ni_pathbuf = NULL;
1765 	}
1766 	abort = 0;
1767 	nfsm_reply(NFSX_SRVFH(&nsfh, true) + NFSX_POSTOPATTR(1) +
1768 	    NFSX_WCCDATA(1));
1769 	if (!error) {
1770 		nfsm_srvpostop_fh(&nsfh);
1771 		nfsm_srvpostop_attr(0, &va);
1772 	}
1773 	nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1774 	return (0);
1775 nfsmout:
1776 	if (abort) {
1777 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1778 		if (nd.ni_dvp == nd.ni_vp)
1779 			vrele(nd.ni_dvp);
1780 		else
1781 			vput(nd.ni_dvp);
1782 		if (nd.ni_vp)
1783 			vput(nd.ni_vp);
1784 	}
1785 	if (nd.ni_pathbuf != NULL) {
1786 		pathbuf_destroy(nd.ni_pathbuf);
1787 		nd.ni_pathbuf = NULL;
1788 	}
1789 	if (dirp)
1790 		vrele(dirp);
1791 	return (error);
1792 }
1793 
1794 /*
1795  * nfs remove service
1796  */
1797 int
1798 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1799 {
1800 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1801 	struct mbuf *nam = nfsd->nd_nam;
1802 	char *dpos = nfsd->nd_dpos;
1803 	kauth_cred_t cred = nfsd->nd_cr;
1804 	struct nameidata nd;
1805 	u_int32_t *tl;
1806 	int32_t t1;
1807 	char *bpos;
1808 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
1809 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1810 	char *cp2;
1811 	struct mbuf *mb, *mreq;
1812 	struct vnode *vp, *dirp;
1813 	struct vattr dirfor, diraft;
1814 	nfsrvfh_t nsfh;
1815 	u_quad_t frev;
1816 
1817 #ifndef nolint
1818 	vp = (struct vnode *)0;
1819 #endif
1820 	nfsm_srvmtofh(&nsfh);
1821 	nfsm_srvnamesiz(len);
1822 	nd.ni_cnd.cn_cred = cred;
1823 	nd.ni_cnd.cn_nameiop = DELETE;
1824 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1825 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1826 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1827 	if (error == 0 && dirp && v3) {
1828 		if (nd.ni_dvp == nd.ni_vp)
1829 			vn_lock(dirp, LK_SHARED | LK_RETRY);
1830 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1831 		if (nd.ni_dvp == nd.ni_vp)
1832 			VOP_UNLOCK(dirp);
1833 	}
1834 	if (!error) {
1835 		vp = nd.ni_vp;
1836 		if (vp->v_type == VDIR) {
1837 			error = EPERM;
1838 			goto out;
1839 		}
1840 		/*
1841 		 * The root of a mounted filesystem cannot be deleted.
1842 		 */
1843 		if (vp->v_vflag & VV_ROOT) {
1844 			error = EBUSY;
1845 		}
1846 out:
1847 		if (!error) {
1848 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1849 			nqsrv_getl(vp, ND_WRITE);
1850 			error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1851 		} else {
1852 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1853 			if (nd.ni_dvp == vp)
1854 				vrele(nd.ni_dvp);
1855 			else
1856 				vput(nd.ni_dvp);
1857 			vput(vp);
1858 		}
1859 	}
1860 	if (nd.ni_pathbuf != NULL) {
1861 		pathbuf_destroy(nd.ni_pathbuf);
1862 		nd.ni_pathbuf = NULL;
1863 	}
1864 	if (dirp) {
1865 		if (v3) {
1866 			vn_lock(dirp, LK_SHARED | LK_RETRY);
1867 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1868 			VOP_UNLOCK(dirp);
1869 		}
1870 		vrele(dirp);
1871 	}
1872 	nfsm_reply(NFSX_WCCDATA(v3));
1873 	if (v3) {
1874 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1875 		return (0);
1876 	}
1877 	nfsm_srvdone;
1878 }
1879 
1880 /*
1881  * nfs rename service
1882  */
1883 int
1884 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1885 {
1886 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1887 	struct mbuf *nam = nfsd->nd_nam;
1888 	char *dpos = nfsd->nd_dpos;
1889 	kauth_cred_t cred = nfsd->nd_cr;
1890 	u_int32_t *tl;
1891 	int32_t t1;
1892 	char *bpos;
1893 	int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1;
1894 	uint32_t len, len2;
1895 	int tdirfor_ret = 1, tdiraft_ret = 1;
1896 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1897 	char *cp2;
1898 	struct mbuf *mb, *mreq;
1899 	struct nameidata fromnd, tond;
1900 	struct vnode *fvp, *tvp, *tdvp;
1901 	struct vnode *fdirp = NULL, *tdirp = NULL;
1902 	struct mount *localfs = NULL;
1903 	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1904 	nfsrvfh_t fnsfh, tnsfh;
1905 	u_quad_t frev;
1906 	uid_t saved_uid;
1907 
1908 #ifndef nolint
1909 	fvp = (struct vnode *)0;
1910 #endif
1911 	fromnd.ni_cnd.cn_nameiop = 0;
1912 	tond.ni_cnd.cn_nameiop = 0;
1913 	nfsm_srvmtofh(&fnsfh);
1914 	nfsm_srvnamesiz(len);
1915 	/*
1916 	 * Remember our original uid so that we can reset cr_uid before
1917 	 * the second nfs_namei() call, in case it is remapped.
1918 	 */
1919 	saved_uid = kauth_cred_geteuid(cred);
1920 	fromnd.ni_cnd.cn_cred = cred;
1921 	fromnd.ni_cnd.cn_nameiop = DELETE;
1922 	fromnd.ni_cnd.cn_flags = LOCKPARENT | INRENAME;
1923 	error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md,
1924 		&dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1925 	if (error == 0 && fdirp && v3) {
1926 		if (fromnd.ni_dvp == fromnd.ni_vp)
1927 			vn_lock(fdirp, LK_SHARED | LK_RETRY);
1928 		fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred);
1929 		if (fromnd.ni_dvp == fromnd.ni_vp)
1930 			VOP_UNLOCK(fdirp);
1931 	}
1932 	if (error) {
1933 		nfsm_reply(2 * NFSX_WCCDATA(v3));
1934 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1935 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1936 		if (fdirp)
1937 			vrele(fdirp);
1938 		if (fromnd.ni_pathbuf != NULL) {
1939 			pathbuf_destroy(fromnd.ni_pathbuf);
1940 		}
1941 		return (0);
1942 	}
1943 	if (fromnd.ni_dvp != fromnd.ni_vp) {
1944 		VOP_UNLOCK(fromnd.ni_dvp);
1945 	}
1946 	fvp = fromnd.ni_vp;
1947 
1948 	localfs = fvp->v_mount;
1949 	error = VFS_RENAMELOCK_ENTER(localfs);
1950 	if (error) {
1951 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1952 		vrele(fromnd.ni_dvp);
1953 		vrele(fvp);
1954 		goto out1;
1955 	}
1956 
1957 	/* Copied, regrettably, from vfs_syscalls.c (q.v.) */
1958 	vrele(fvp);
1959 	if ((fromnd.ni_cnd.cn_namelen == 1 &&
1960 	     fromnd.ni_cnd.cn_nameptr[0] == '.') ||
1961 	    (fromnd.ni_cnd.cn_namelen == 2 &&
1962 	     fromnd.ni_cnd.cn_nameptr[0] == '.' &&
1963 	     fromnd.ni_cnd.cn_nameptr[1] == '.')) {
1964 		error = EINVAL;
1965 		VFS_RENAMELOCK_EXIT(localfs);
1966 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1967 		vrele(fromnd.ni_dvp);
1968 		goto out1;
1969 	}
1970 	vn_lock(fromnd.ni_dvp, LK_EXCLUSIVE | LK_RETRY);
1971 	error = relookup(fromnd.ni_dvp, &fromnd.ni_vp, &fromnd.ni_cnd, 0);
1972 	if (error) {
1973 		VOP_UNLOCK(fromnd.ni_dvp);
1974 		VFS_RENAMELOCK_EXIT(localfs);
1975 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1976 		vrele(fromnd.ni_dvp);
1977 		goto out1;
1978 	}
1979 	VOP_UNLOCK(fromnd.ni_vp);
1980 	if (fromnd.ni_dvp != fromnd.ni_vp)
1981 		VOP_UNLOCK(fromnd.ni_dvp);
1982 	fvp = fromnd.ni_vp;
1983 
1984 	nfsm_srvmtofh(&tnsfh);
1985 	if (v3) {
1986 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1987 		len2 = fxdr_unsigned(uint32_t, *tl);
1988 		/* len2 will be checked by nfs_namei */
1989 	}
1990 	else {
1991 		/* NFSv2 */
1992 		nfsm_strsiz(len2, NFS_MAXNAMLEN);
1993 	}
1994 	kauth_cred_seteuid(cred, saved_uid);
1995 	tond.ni_cnd.cn_cred = cred;
1996 	tond.ni_cnd.cn_nameiop = RENAME;
1997 	tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | INRENAME;
1998 	error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md,
1999 		&dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2000 	if (tdirp && v3) {
2001 		tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred);
2002 	}
2003 	if (error) {
2004 		VFS_RENAMELOCK_EXIT(localfs);
2005 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2006 		vrele(fromnd.ni_dvp);
2007 		vrele(fvp);
2008 		goto out1;
2009 	}
2010 	tdvp = tond.ni_dvp;
2011 	tvp = tond.ni_vp;
2012 	if (tvp != NULL) {
2013 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2014 			if (v3)
2015 				error = EEXIST;
2016 			else
2017 				error = EISDIR;
2018 			goto out;
2019 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2020 			if (v3)
2021 				error = EEXIST;
2022 			else
2023 				error = ENOTDIR;
2024 			goto out;
2025 		}
2026 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
2027 			if (v3)
2028 				error = EXDEV;
2029 			else
2030 				error = ENOTEMPTY;
2031 			goto out;
2032 		}
2033 	}
2034 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
2035 		if (v3)
2036 			error = EXDEV;
2037 		else
2038 			error = ENOTEMPTY;
2039 		goto out;
2040 	}
2041 	if (fvp->v_mount != tdvp->v_mount) {
2042 		if (v3)
2043 			error = EXDEV;
2044 		else
2045 			error = ENOTEMPTY;
2046 		goto out;
2047 	}
2048 	if (fvp == tdvp) {
2049 		if (v3)
2050 			error = EINVAL;
2051 		else
2052 			error = ENOTEMPTY;
2053 	}
2054 	/*
2055 	 * If source is the same as the destination (that is the
2056 	 * same vnode with the same name in the same directory),
2057 	 * then there is nothing to do.
2058 	 */
2059 	if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2060 	    fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2061 	    !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2062 	      fromnd.ni_cnd.cn_namelen))
2063 		error = -1;
2064 out:
2065 	if (!error) {
2066 		nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
2067 		nqsrv_getl(tdvp, ND_WRITE);
2068 		if (tvp) {
2069 			nqsrv_getl(tvp, ND_WRITE);
2070 		}
2071 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2072 				   tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2073 		VFS_RENAMELOCK_EXIT(localfs);
2074 	} else {
2075 		VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd);
2076 		if (tdvp == tvp)
2077 			vrele(tdvp);
2078 		else
2079 			vput(tdvp);
2080 		if (tvp)
2081 			vput(tvp);
2082 		VFS_RENAMELOCK_EXIT(localfs);
2083 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2084 		vrele(fromnd.ni_dvp);
2085 		vrele(fvp);
2086 		if (error == -1)
2087 			error = 0;
2088 	}
2089 	if (tond.ni_pathbuf != NULL) {
2090 		pathbuf_destroy(tond.ni_pathbuf);
2091 		tond.ni_pathbuf = NULL;
2092 	}
2093 	tond.ni_cnd.cn_nameiop = 0;
2094 out1:
2095 	if (fdirp) {
2096 		if (v3) {
2097 			vn_lock(fdirp, LK_SHARED | LK_RETRY);
2098 			fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred);
2099 			VOP_UNLOCK(fdirp);
2100 		}
2101 		vrele(fdirp);
2102 		fdirp = NULL;
2103 	}
2104 	if (tdirp) {
2105 		if (v3) {
2106 			vn_lock(tdirp, LK_SHARED | LK_RETRY);
2107 			tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred);
2108 			VOP_UNLOCK(tdirp);
2109 		}
2110 		vrele(tdirp);
2111 		tdirp = NULL;
2112 	}
2113 	pathbuf_destroy(fromnd.ni_pathbuf);
2114 	fromnd.ni_pathbuf = NULL;
2115 	fromnd.ni_cnd.cn_nameiop = 0;
2116 	localfs = NULL;
2117 	nfsm_reply(2 * NFSX_WCCDATA(v3));
2118 	if (v3) {
2119 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2120 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2121 	}
2122 	return (0);
2123 
2124 nfsmout:
2125 	if (fdirp)
2126 		vrele(fdirp);
2127 #ifdef notdef
2128 	if (tdirp)
2129 		vrele(tdirp);
2130 #endif
2131 	if (tond.ni_cnd.cn_nameiop) {
2132 		if (tond.ni_pathbuf != NULL) {
2133 			pathbuf_destroy(tond.ni_pathbuf);
2134 			tond.ni_pathbuf = NULL;
2135 		}
2136 	}
2137 	if (localfs) {
2138 		VFS_RENAMELOCK_EXIT(localfs);
2139 	}
2140 	if (fromnd.ni_cnd.cn_nameiop) {
2141 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2142 		if (fromnd.ni_pathbuf != NULL) {
2143 			pathbuf_destroy(fromnd.ni_pathbuf);
2144 			fromnd.ni_pathbuf = NULL;
2145 		}
2146 		vrele(fromnd.ni_dvp);
2147 		vrele(fvp);
2148 	}
2149 	return (error);
2150 }
2151 
2152 /*
2153  * nfs link service
2154  */
2155 int
2156 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2157 {
2158 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2159 	struct mbuf *nam = nfsd->nd_nam;
2160 	char *dpos = nfsd->nd_dpos;
2161 	kauth_cred_t cred = nfsd->nd_cr;
2162 	struct nameidata nd;
2163 	u_int32_t *tl;
2164 	int32_t t1;
2165 	char *bpos;
2166 	int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2167 	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2168 	char *cp2;
2169 	struct mbuf *mb, *mreq;
2170 	struct vnode *vp, *xp, *dirp = (struct vnode *)0;
2171 	struct vattr dirfor, diraft, at;
2172 	nfsrvfh_t nsfh, dnsfh;
2173 	u_quad_t frev;
2174 
2175 	nfsm_srvmtofh(&nsfh);
2176 	nfsm_srvmtofh(&dnsfh);
2177 	nfsm_srvnamesiz(len);
2178 	error = nfsrv_fhtovp(&nsfh, false, &vp, cred, slp, nam,
2179 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2180 	if (error) {
2181 		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2182 		nfsm_srvpostop_attr(getret, &at);
2183 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2184 		return (0);
2185 	}
2186 	if (vp->v_type == VDIR) {
2187 		error = EPERM;
2188 		goto out1;
2189 	}
2190 	nd.ni_cnd.cn_cred = cred;
2191 	nd.ni_cnd.cn_nameiop = CREATE;
2192 	nd.ni_cnd.cn_flags = LOCKPARENT;
2193 	error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos,
2194 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2195 	if (dirp && v3) {
2196 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2197 	}
2198 	if (error)
2199 		goto out1;
2200 	xp = nd.ni_vp;
2201 	if (xp != NULL) {
2202 		error = EEXIST;
2203 		goto out;
2204 	}
2205 	xp = nd.ni_dvp;
2206 	if (vp->v_mount != xp->v_mount)
2207 		error = EXDEV;
2208 out:
2209 	if (!error) {
2210 		nqsrv_getl(vp, ND_WRITE);
2211 		nqsrv_getl(xp, ND_WRITE);
2212 		error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2213 	} else {
2214 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2215 		if (nd.ni_dvp == nd.ni_vp)
2216 			vrele(nd.ni_dvp);
2217 		else
2218 			vput(nd.ni_dvp);
2219 		if (nd.ni_vp)
2220 			vrele(nd.ni_vp);
2221 	}
2222 out1:
2223 	if (v3) {
2224 		vn_lock(vp, LK_SHARED | LK_RETRY);
2225 		getret = VOP_GETATTR(vp, &at, cred);
2226 		VOP_UNLOCK(vp);
2227 	}
2228 	if (dirp) {
2229 		if (v3) {
2230 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2231 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2232 			VOP_UNLOCK(dirp);
2233 		}
2234 		vrele(dirp);
2235 	}
2236 	vrele(vp);
2237 	if (nd.ni_pathbuf != NULL) {
2238 		pathbuf_destroy(nd.ni_pathbuf);
2239 		nd.ni_pathbuf = NULL;
2240 	}
2241 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2242 	if (v3) {
2243 		nfsm_srvpostop_attr(getret, &at);
2244 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2245 		return (0);
2246 	}
2247 	nfsm_srvdone;
2248 }
2249 
2250 /*
2251  * nfs symbolic link service
2252  */
2253 int
2254 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2255 {
2256 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2257 	struct mbuf *nam = nfsd->nd_nam;
2258 	char *dpos = nfsd->nd_dpos;
2259 	kauth_cred_t cred = nfsd->nd_cr;
2260 	struct vattr va, dirfor, diraft;
2261 	struct nameidata nd;
2262 	u_int32_t *tl;
2263 	int32_t t1;
2264 	struct nfsv2_sattr *sp;
2265 	char *bpos, *pathcp = NULL, *cp2;
2266 	struct uio io;
2267 	struct iovec iv;
2268 	int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1, abort = 0;
2269 	uint32_t len, len2;
2270 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2271 	struct mbuf *mb, *mreq;
2272 	struct vnode *dirp = (struct vnode *)0;
2273 	nfsrvfh_t nsfh;
2274 	u_quad_t frev;
2275 
2276 	nd.ni_cnd.cn_nameiop = 0;
2277 	nfsm_srvmtofh(&nsfh);
2278 	nfsm_srvnamesiz(len);
2279 	nd.ni_cnd.cn_cred = cred;
2280 	nd.ni_cnd.cn_nameiop = CREATE;
2281 	nd.ni_cnd.cn_flags = LOCKPARENT;
2282 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2283 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2284 	if (dirp && v3) {
2285 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2286 	}
2287 	if (error)
2288 		goto out;
2289 	abort = 1;
2290 	vattr_null(&va);
2291 	va.va_type = VLNK;
2292 	if (v3) {
2293 		va.va_mode = 0;
2294 		nfsm_srvsattr(&va);
2295 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
2296 		len2 = fxdr_unsigned(uint32_t, *tl);
2297 		if (len2 > PATH_MAX) {
2298 			/* XXX should check _PC_NO_TRUNC */
2299 			error = ENAMETOOLONG;
2300 			goto abortop;
2301 		}
2302 	}
2303 	else {
2304 		/* NFSv2 */
2305 		nfsm_strsiz(len2, NFS_MAXPATHLEN);
2306 	}
2307 	pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2308 	iv.iov_base = pathcp;
2309 	iv.iov_len = len2;
2310 	io.uio_resid = len2;
2311 	io.uio_offset = 0;
2312 	io.uio_iov = &iv;
2313 	io.uio_iovcnt = 1;
2314 	io.uio_rw = UIO_READ;
2315 	UIO_SETUP_SYSSPACE(&io);
2316 	nfsm_mtouio(&io, len2);
2317 	if (!v3) {
2318 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2319 		va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
2320 	}
2321 	*(pathcp + len2) = '\0';
2322 	if (nd.ni_vp) {
2323 		error = EEXIST;
2324 abortop:
2325 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2326 		if (nd.ni_dvp == nd.ni_vp)
2327 			vrele(nd.ni_dvp);
2328 		else
2329 			vput(nd.ni_dvp);
2330 		if (nd.ni_vp)
2331 			vrele(nd.ni_vp);
2332 		goto out;
2333 	}
2334 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2335 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp);
2336 	if (!error) {
2337 	    if (v3) {
2338 		error = nfsrv_composefh(nd.ni_vp, &nsfh, v3);
2339 		if (!error)
2340 		    error = VOP_GETATTR(nd.ni_vp, &va, cred);
2341 		vput(nd.ni_vp);
2342 	    } else {
2343 		vput(nd.ni_vp);
2344 	    }
2345 	}
2346 out:
2347 	if (pathcp)
2348 		free(pathcp, M_TEMP);
2349 	if (dirp) {
2350 		if (v3) {
2351 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2352 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2353 			VOP_UNLOCK(dirp);
2354 		}
2355 		vrele(dirp);
2356 		dirp = NULL;
2357 	}
2358 	if (nd.ni_pathbuf != NULL) {
2359 		pathbuf_destroy(nd.ni_pathbuf);
2360 		nd.ni_pathbuf = NULL;
2361 	}
2362 	abort = 0;
2363 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) +
2364 	    NFSX_WCCDATA(v3));
2365 	if (v3) {
2366 		if (!error) {
2367 			nfsm_srvpostop_fh(&nsfh);
2368 			nfsm_srvpostop_attr(0, &va);
2369 		}
2370 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2371 	}
2372 	return (0);
2373 nfsmout:
2374 	if (abort) {
2375 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2376 		if (nd.ni_dvp == nd.ni_vp)
2377 			vrele(nd.ni_dvp);
2378 		else
2379 			vput(nd.ni_dvp);
2380 		if (nd.ni_vp)
2381 			vrele(nd.ni_vp);
2382 		if (nd.ni_pathbuf != NULL) {
2383 			pathbuf_destroy(nd.ni_pathbuf);
2384 			nd.ni_pathbuf = NULL;
2385 		}
2386 	}
2387 	if (dirp)
2388 		vrele(dirp);
2389 	if (pathcp)
2390 		free(pathcp, M_TEMP);
2391 	return (error);
2392 }
2393 
2394 /*
2395  * nfs mkdir service
2396  */
2397 int
2398 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2399 {
2400 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2401 	struct mbuf *nam = nfsd->nd_nam;
2402 	char *dpos = nfsd->nd_dpos;
2403 	kauth_cred_t cred = nfsd->nd_cr;
2404 	struct vattr va, dirfor, diraft;
2405 	struct nfs_fattr *fp;
2406 	struct nameidata nd;
2407 	char *cp;
2408 	u_int32_t *tl;
2409 	int32_t t1;
2410 	char *bpos;
2411 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2412 	int abort = 0;
2413 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2414 	char *cp2;
2415 	struct mbuf *mb, *mreq;
2416 	struct vnode *vp, *dirp = (struct vnode *)0;
2417 	nfsrvfh_t nsfh;
2418 	u_quad_t frev;
2419 
2420 	nfsm_srvmtofh(&nsfh);
2421 	nfsm_srvnamesiz(len);
2422 	nd.ni_cnd.cn_cred = cred;
2423 	nd.ni_cnd.cn_nameiop = CREATE;
2424 	nd.ni_cnd.cn_flags = LOCKPARENT;
2425 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2426 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2427 	if (dirp && v3) {
2428 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2429 	}
2430 	if (error) {
2431 		if (nd.ni_pathbuf != NULL) {
2432 			pathbuf_destroy(nd.ni_pathbuf);
2433 			nd.ni_pathbuf = NULL;
2434 		}
2435 		nfsm_reply(NFSX_WCCDATA(v3));
2436 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2437 		if (dirp)
2438 			vrele(dirp);
2439 		return (0);
2440 	}
2441 	abort = 1;
2442 	vattr_null(&va);
2443 	if (v3) {
2444 		va.va_mode = 0;
2445 		nfsm_srvsattr(&va);
2446 	} else {
2447 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2448 		va.va_mode = nfstov_mode(*tl++);
2449 	}
2450 	va.va_type = VDIR;
2451 	vp = nd.ni_vp;
2452 	if (vp != NULL) {
2453 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2454 		if (nd.ni_dvp == vp)
2455 			vrele(nd.ni_dvp);
2456 		else
2457 			vput(nd.ni_dvp);
2458 		vrele(vp);
2459 		error = EEXIST;
2460 		goto out;
2461 	}
2462 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2463 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
2464 	if (!error) {
2465 		vp = nd.ni_vp;
2466 		error = nfsrv_composefh(vp, &nsfh, v3);
2467 		if (!error)
2468 			error = VOP_GETATTR(vp, &va, cred);
2469 		vput(vp);
2470 	}
2471 out:
2472 	if (dirp) {
2473 		if (v3) {
2474 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2475 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2476 			VOP_UNLOCK(dirp);
2477 		}
2478 		vrele(dirp);
2479 		dirp = NULL;
2480 	}
2481 	if (nd.ni_pathbuf != NULL) {
2482 		pathbuf_destroy(nd.ni_pathbuf);
2483 		nd.ni_pathbuf = NULL;
2484 	}
2485 	abort = 0;
2486 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) +
2487 	    NFSX_WCCDATA(v3));
2488 	if (v3) {
2489 		if (!error) {
2490 			nfsm_srvpostop_fh(&nsfh);
2491 			nfsm_srvpostop_attr(0, &va);
2492 		}
2493 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2494 	} else {
2495 		nfsm_srvfhtom(&nsfh, v3);
2496 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2497 		nfsm_srvfillattr(&va, fp);
2498 	}
2499 	return (0);
2500 nfsmout:
2501 	if (abort) {
2502 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2503 		if (nd.ni_dvp == nd.ni_vp)
2504 			vrele(nd.ni_dvp);
2505 		else
2506 			vput(nd.ni_dvp);
2507 		if (nd.ni_vp)
2508 			vrele(nd.ni_vp);
2509 		if (nd.ni_pathbuf != NULL) {
2510 			pathbuf_destroy(nd.ni_pathbuf);
2511 			nd.ni_pathbuf = NULL;
2512 		}
2513 	}
2514 	if (dirp)
2515 		vrele(dirp);
2516 	return (error);
2517 }
2518 
2519 /*
2520  * nfs rmdir service
2521  */
2522 int
2523 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2524 {
2525 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2526 	struct mbuf *nam = nfsd->nd_nam;
2527 	char *dpos = nfsd->nd_dpos;
2528 	kauth_cred_t cred = nfsd->nd_cr;
2529 	u_int32_t *tl;
2530 	int32_t t1;
2531 	char *bpos;
2532 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2533 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2534 	char *cp2;
2535 	struct mbuf *mb, *mreq;
2536 	struct vnode *vp, *dirp = (struct vnode *)0;
2537 	struct vattr dirfor, diraft;
2538 	nfsrvfh_t nsfh;
2539 	struct nameidata nd;
2540 	u_quad_t frev;
2541 
2542 	nfsm_srvmtofh(&nsfh);
2543 	nfsm_srvnamesiz(len);
2544 	nd.ni_cnd.cn_cred = cred;
2545 	nd.ni_cnd.cn_nameiop = DELETE;
2546 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2547 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2548 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2549 	if (dirp && v3) {
2550 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2551 	}
2552 	if (error) {
2553 		if (nd.ni_pathbuf != NULL) {
2554 			pathbuf_destroy(nd.ni_pathbuf);
2555 			nd.ni_pathbuf = NULL;
2556 		}
2557 		nfsm_reply(NFSX_WCCDATA(v3));
2558 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2559 		if (dirp)
2560 			vrele(dirp);
2561 		return (0);
2562 	}
2563 	vp = nd.ni_vp;
2564 	if (vp->v_type != VDIR) {
2565 		error = ENOTDIR;
2566 		goto out;
2567 	}
2568 	/*
2569 	 * No rmdir "." please.
2570 	 */
2571 	if (nd.ni_dvp == vp) {
2572 		error = EINVAL;
2573 		goto out;
2574 	}
2575 	/*
2576 	 * The root of a mounted filesystem cannot be deleted.
2577 	 */
2578 	if (vp->v_vflag & VV_ROOT)
2579 		error = EBUSY;
2580 out:
2581 	if (!error) {
2582 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
2583 		nqsrv_getl(vp, ND_WRITE);
2584 		error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2585 	} else {
2586 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2587 		if (nd.ni_dvp == nd.ni_vp)
2588 			vrele(nd.ni_dvp);
2589 		else
2590 			vput(nd.ni_dvp);
2591 		vput(vp);
2592 	}
2593 	if (nd.ni_pathbuf != NULL) {
2594 		pathbuf_destroy(nd.ni_pathbuf);
2595 		nd.ni_pathbuf = NULL;
2596 	}
2597 	if (dirp) {
2598 		if (v3) {
2599 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2600 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2601 			VOP_UNLOCK(dirp);
2602 		}
2603 		vrele(dirp);
2604 	}
2605 	nfsm_reply(NFSX_WCCDATA(v3));
2606 	if (v3) {
2607 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2608 		return (0);
2609 	}
2610 	nfsm_srvdone;
2611 }
2612 
2613 /*
2614  * nfs readdir service
2615  * - mallocs what it thinks is enough to read
2616  *	count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR
2617  * - calls VOP_READDIR()
2618  * - loops around building the reply
2619  *	if the output generated exceeds count break out of loop
2620  *	The nfsm_clget macro is used here so that the reply will be packed
2621  *	tightly in mbuf clusters.
2622  * - it only knows that it has encountered eof when the VOP_READDIR()
2623  *	reads nothing
2624  * - as such one readdir rpc will return eof false although you are there
2625  *	and then the next will return eof
2626  * - it trims out records with d_fileno == 0
2627  *	this doesn't matter for Unix clients, but they might confuse clients
2628  *	for other os'.
2629  * - it trims out records with d_type == DT_WHT
2630  *	these cannot be seen through NFS (unless we extend the protocol)
2631  * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2632  *	than requested, but this may not apply to all filesystems. For
2633  *	example, client NFS does not { although it is never remote mounted
2634  *	anyhow }
2635  *     The alternate call nfsrv_readdirplus() does lookups as well.
2636  * PS: The NFS protocol spec. does not clarify what the "count" byte
2637  *	argument is a count of.. just name strings and file id's or the
2638  *	entire reply rpc or ...
2639  *	I tried just file name and id sizes and it confused the Sun client,
2640  *	so I am using the full rpc size now. The "paranoia.." comment refers
2641  *	to including the status longwords that are not a part of the dir.
2642  *	"entry" structures, but are in the rpc.
2643  */
2644 
2645 #define	NFS_SRVDIRBLKSIZ	1024
2646 
2647 struct flrep {
2648 	nfsuint64 fl_off;
2649 	u_int32_t fl_postopok;
2650 	struct nfs_fattr fl_fattr; /* XXX: must be of fattr3 size */
2651 	u_int32_t fl_fhok;
2652 	u_int32_t fl_fhsize;
2653 	/* handle comes here, filled in dynamically */
2654 };
2655 
2656 int
2657 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2658 {
2659 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2660 	struct mbuf *nam = nfsd->nd_nam;
2661 	char *dpos = nfsd->nd_dpos;
2662 	kauth_cred_t cred = nfsd->nd_cr;
2663 	char *bp, *be;
2664 	struct mbuf *mp;
2665 	struct dirent *dp;
2666 	char *cp;
2667 	u_int32_t *tl;
2668 	int32_t t1;
2669 	char *bpos;
2670 	struct mbuf *mb, *mreq, *mp2;
2671 	char *cpos, *cend, *cp2, *rbuf;
2672 	struct vnode *vp;
2673 	struct vattr at;
2674 	nfsrvfh_t nsfh;
2675 	struct uio io;
2676 	struct iovec iv;
2677 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2678 	int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies;
2679 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2680 	u_quad_t frev, off, toff, verf;
2681 	off_t *cookies = NULL, *cookiep;
2682 	nfsuint64 jar;
2683 
2684 	nfsm_srvmtofh(&nsfh);
2685 	if (v3) {
2686 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2687 		toff = fxdr_hyper(tl);
2688 		tl += 2;
2689 		verf = fxdr_hyper(tl);
2690 		tl += 2;
2691 	} else {
2692 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2693 		toff = fxdr_unsigned(u_quad_t, *tl++);
2694 	}
2695 	off = toff;
2696 	cnt = fxdr_unsigned(int, *tl);
2697 	siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1));
2698 	xfer = NFS_SRVMAXDATA(nfsd);
2699 	if (siz > xfer)
2700 		siz = xfer;
2701 	fullsiz = siz;
2702 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
2703 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2704 	if (!error && vp->v_type != VDIR) {
2705 		error = ENOTDIR;
2706 		vput(vp);
2707 	}
2708 	if (error) {
2709 		nfsm_reply(NFSX_UNSIGNED);
2710 		nfsm_srvpostop_attr(getret, &at);
2711 		return (0);
2712 	}
2713 	nqsrv_getl(vp, ND_READ);
2714 	if (v3) {
2715 		error = getret = VOP_GETATTR(vp, &at, cred);
2716 #ifdef NFS3_STRICTVERF
2717 		/*
2718 		 * XXX This check is too strict for Solaris 2.5 clients.
2719 		 */
2720 		if (!error && toff && verf != at.va_filerev)
2721 			error = NFSERR_BAD_COOKIE;
2722 #endif
2723 	}
2724 	if (!error)
2725 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0);
2726 	if (error) {
2727 		vput(vp);
2728 		nfsm_reply(NFSX_POSTOPATTR(v3));
2729 		nfsm_srvpostop_attr(getret, &at);
2730 		return (0);
2731 	}
2732 	VOP_UNLOCK(vp);
2733 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2734 again:
2735 	iv.iov_base = rbuf;
2736 	iv.iov_len = fullsiz;
2737 	io.uio_iov = &iv;
2738 	io.uio_iovcnt = 1;
2739 	io.uio_offset = (off_t)off;
2740 	io.uio_resid = fullsiz;
2741 	io.uio_rw = UIO_READ;
2742 	UIO_SETUP_SYSSPACE(&io);
2743 	eofflag = 0;
2744 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2745 
2746 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2747 
2748 	off = (off_t)io.uio_offset;
2749 	if (!cookies && !error)
2750 		error = NFSERR_PERM;
2751 	if (v3) {
2752 		getret = VOP_GETATTR(vp, &at, cred);
2753 		if (!error)
2754 			error = getret;
2755 	}
2756 
2757 	VOP_UNLOCK(vp);
2758 	if (error) {
2759 		vrele(vp);
2760 		free((void *)rbuf, M_TEMP);
2761 		if (cookies)
2762 			free((void *)cookies, M_TEMP);
2763 		nfsm_reply(NFSX_POSTOPATTR(v3));
2764 		nfsm_srvpostop_attr(getret, &at);
2765 		return (0);
2766 	}
2767 	if (io.uio_resid) {
2768 		siz -= io.uio_resid;
2769 
2770 		/*
2771 		 * If nothing read, return eof
2772 		 * rpc reply
2773 		 */
2774 		if (siz == 0) {
2775 			vrele(vp);
2776 			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2777 				2 * NFSX_UNSIGNED);
2778 			if (v3) {
2779 				nfsm_srvpostop_attr(getret, &at);
2780 				nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2781 				txdr_hyper(at.va_filerev, tl);
2782 				tl += 2;
2783 			} else
2784 				nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2785 			*tl++ = nfs_false;
2786 			*tl = nfs_true;
2787 			free((void *)rbuf, M_TEMP);
2788 			free((void *)cookies, M_TEMP);
2789 			return (0);
2790 		}
2791 	}
2792 
2793 	/*
2794 	 * Check for degenerate cases of nothing useful read.
2795 	 * If so go try again
2796 	 */
2797 	cpos = rbuf;
2798 	cend = rbuf + siz;
2799 	dp = (struct dirent *)cpos;
2800 	cookiep = cookies;
2801 
2802 	while (cpos < cend && ncookies > 0 &&
2803 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2804 		cpos += dp->d_reclen;
2805 		dp = (struct dirent *)cpos;
2806 		cookiep++;
2807 		ncookies--;
2808 	}
2809 	if (cpos >= cend || ncookies == 0) {
2810 		toff = off;
2811 		siz = fullsiz;
2812 		free(cookies, M_TEMP);
2813 		cookies = NULL;
2814 		goto again;
2815 	}
2816 
2817 	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2818 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2819 	if (v3) {
2820 		nfsm_srvpostop_attr(getret, &at);
2821 		nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2822 		txdr_hyper(at.va_filerev, tl);
2823 	}
2824 	mp = mp2 = mb;
2825 	bp = bpos;
2826 	be = bp + M_TRAILINGSPACE(mp);
2827 
2828 	/* Loop through the records and build reply */
2829 	while (cpos < cend && ncookies > 0) {
2830 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2831 			nlen = dp->d_namlen;
2832 			rem = nfsm_rndup(nlen)-nlen;
2833 			len += (4 * NFSX_UNSIGNED + nlen + rem);
2834 			if (v3)
2835 				len += 2 * NFSX_UNSIGNED;
2836 			if (len > cnt) {
2837 				eofflag = 0;
2838 				break;
2839 			}
2840 			/*
2841 			 * Build the directory record xdr from
2842 			 * the dirent entry.
2843 			 */
2844 			nfsm_clget;
2845 			*tl = nfs_true;
2846 			bp += NFSX_UNSIGNED;
2847 			if (v3) {
2848 				nfsm_clget;
2849 				*tl = txdr_unsigned(dp->d_fileno >> 32);
2850 				bp += NFSX_UNSIGNED;
2851 			}
2852 			nfsm_clget;
2853 			*tl = txdr_unsigned(dp->d_fileno);
2854 			bp += NFSX_UNSIGNED;
2855 			nfsm_clget;
2856 			*tl = txdr_unsigned(nlen);
2857 			bp += NFSX_UNSIGNED;
2858 
2859 			/* And loop around copying the name */
2860 			xfer = nlen;
2861 			cp = dp->d_name;
2862 			while (xfer > 0) {
2863 				nfsm_clget;
2864 				if ((bp+xfer) > be)
2865 					tsiz = be-bp;
2866 				else
2867 					tsiz = xfer;
2868 				memcpy(bp, cp, tsiz);
2869 				bp += tsiz;
2870 				xfer -= tsiz;
2871 				if (xfer > 0)
2872 					cp += tsiz;
2873 			}
2874 			/* And null pad to an int32_t boundary */
2875 			for (i = 0; i < rem; i++)
2876 				*bp++ = '\0';
2877 			nfsm_clget;
2878 
2879 			/* Finish off the record */
2880 			txdr_hyper(*cookiep, &jar);
2881 			if (v3) {
2882 				*tl = jar.nfsuquad[0];
2883 				bp += NFSX_UNSIGNED;
2884 				nfsm_clget;
2885 			}
2886 			*tl = jar.nfsuquad[1];
2887 			bp += NFSX_UNSIGNED;
2888 		}
2889 		cpos += dp->d_reclen;
2890 		dp = (struct dirent *)cpos;
2891 		cookiep++;
2892 		ncookies--;
2893 	}
2894 	vrele(vp);
2895 	nfsm_clget;
2896 	*tl = nfs_false;
2897 	bp += NFSX_UNSIGNED;
2898 	nfsm_clget;
2899 	if (eofflag)
2900 		*tl = nfs_true;
2901 	else
2902 		*tl = nfs_false;
2903 	bp += NFSX_UNSIGNED;
2904 	if (mp != mb) {
2905 		if (bp < be)
2906 			mp->m_len = bp - mtod(mp, char *);
2907 	} else
2908 		mp->m_len += bp - bpos;
2909 	free((void *)rbuf, M_TEMP);
2910 	free((void *)cookies, M_TEMP);
2911 	nfsm_srvdone;
2912 }
2913 
2914 int
2915 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2916 {
2917 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2918 	struct mbuf *nam = nfsd->nd_nam;
2919 	char *dpos = nfsd->nd_dpos;
2920 	kauth_cred_t cred = nfsd->nd_cr;
2921 	char *bp, *be;
2922 	struct mbuf *mp;
2923 	struct dirent *dp;
2924 	char *cp;
2925 	u_int32_t *tl;
2926 	int32_t t1;
2927 	char *bpos;
2928 	struct mbuf *mb, *mreq, *mp2;
2929 	char *cpos, *cend, *cp2, *rbuf;
2930 	struct vnode *vp, *nvp;
2931 	struct flrep fl;
2932 	nfsrvfh_t nsfh;
2933 	struct uio io;
2934 	struct iovec iv;
2935 	struct vattr va, at, *vap = &va;
2936 	struct nfs_fattr *fp;
2937 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2938 	int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies;
2939 	u_quad_t frev, off, toff, verf;
2940 	off_t *cookies = NULL, *cookiep;
2941 
2942 	nfsm_srvmtofh(&nsfh);
2943 	nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2944 	toff = fxdr_hyper(tl);
2945 	tl += 2;
2946 	verf = fxdr_hyper(tl);
2947 	tl += 2;
2948 	siz = fxdr_unsigned(int, *tl++);
2949 	cnt = fxdr_unsigned(int, *tl);
2950 	off = toff;
2951 	siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1));
2952 	xfer = NFS_SRVMAXDATA(nfsd);
2953 	if (siz > xfer)
2954 		siz = xfer;
2955 	fullsiz = siz;
2956 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
2957 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2958 	if (!error && vp->v_type != VDIR) {
2959 		error = ENOTDIR;
2960 		vput(vp);
2961 	}
2962 	if (error) {
2963 		nfsm_reply(NFSX_UNSIGNED);
2964 		nfsm_srvpostop_attr(getret, &at);
2965 		return (0);
2966 	}
2967 	error = getret = VOP_GETATTR(vp, &at, cred);
2968 #ifdef NFS3_STRICTVERF
2969 	/*
2970 	 * XXX This check is too strict for Solaris 2.5 clients.
2971 	 */
2972 	if (!error && toff && verf != at.va_filerev)
2973 		error = NFSERR_BAD_COOKIE;
2974 #endif
2975 	if (!error) {
2976 		nqsrv_getl(vp, ND_READ);
2977 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0);
2978 	}
2979 	if (error) {
2980 		vput(vp);
2981 		nfsm_reply(NFSX_V3POSTOPATTR);
2982 		nfsm_srvpostop_attr(getret, &at);
2983 		return (0);
2984 	}
2985 	VOP_UNLOCK(vp);
2986 
2987 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2988 again:
2989 	iv.iov_base = rbuf;
2990 	iv.iov_len = fullsiz;
2991 	io.uio_iov = &iv;
2992 	io.uio_iovcnt = 1;
2993 	io.uio_offset = (off_t)off;
2994 	io.uio_resid = fullsiz;
2995 	io.uio_rw = UIO_READ;
2996 	UIO_SETUP_SYSSPACE(&io);
2997 	eofflag = 0;
2998 
2999 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3000 
3001 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
3002 
3003 	off = (u_quad_t)io.uio_offset;
3004 	getret = VOP_GETATTR(vp, &at, cred);
3005 
3006 	VOP_UNLOCK(vp);
3007 
3008 	/*
3009 	 * If the VGET operation doesn't work for this filesystem,
3010 	 * we can't support readdirplus. Returning NOTSUPP should
3011 	 * make clients fall back to plain readdir.
3012 	 * There's no need to check for VPTOFH as well, we wouldn't
3013 	 * even be here otherwise.
3014 	 */
3015 	if (!getret) {
3016 		if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp)))
3017 			getret = (getret == EOPNOTSUPP) ?
3018 				NFSERR_NOTSUPP : NFSERR_IO;
3019 		else
3020 			vput(nvp);
3021 	}
3022 
3023 	if (!cookies && !error)
3024 		error = NFSERR_PERM;
3025 	if (!error)
3026 		error = getret;
3027 	if (error) {
3028 		vrele(vp);
3029 		if (cookies)
3030 			free((void *)cookies, M_TEMP);
3031 		free((void *)rbuf, M_TEMP);
3032 		nfsm_reply(NFSX_V3POSTOPATTR);
3033 		nfsm_srvpostop_attr(getret, &at);
3034 		return (0);
3035 	}
3036 	if (io.uio_resid) {
3037 		siz -= io.uio_resid;
3038 
3039 		/*
3040 		 * If nothing read, return eof
3041 		 * rpc reply
3042 		 */
3043 		if (siz == 0) {
3044 			vrele(vp);
3045 			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3046 				2 * NFSX_UNSIGNED);
3047 			nfsm_srvpostop_attr(getret, &at);
3048 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3049 			txdr_hyper(at.va_filerev, tl);
3050 			tl += 2;
3051 			*tl++ = nfs_false;
3052 			*tl = nfs_true;
3053 			free((void *)cookies, M_TEMP);
3054 			free((void *)rbuf, M_TEMP);
3055 			return (0);
3056 		}
3057 	}
3058 
3059 	/*
3060 	 * Check for degenerate cases of nothing useful read.
3061 	 * If so go try again
3062 	 */
3063 	cpos = rbuf;
3064 	cend = rbuf + siz;
3065 	dp = (struct dirent *)cpos;
3066 	cookiep = cookies;
3067 
3068 	while (cpos < cend && ncookies > 0 &&
3069 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
3070 		cpos += dp->d_reclen;
3071 		dp = (struct dirent *)cpos;
3072 		cookiep++;
3073 		ncookies--;
3074 	}
3075 	if (cpos >= cend || ncookies == 0) {
3076 		toff = off;
3077 		siz = fullsiz;
3078 		free(cookies, M_TEMP);
3079 		cookies = NULL;
3080 		goto again;
3081 	}
3082 
3083 	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
3084 	nfsm_reply(cnt);
3085 	nfsm_srvpostop_attr(getret, &at);
3086 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3087 	txdr_hyper(at.va_filerev, tl);
3088 	mp = mp2 = mb;
3089 	bp = bpos;
3090 	be = bp + M_TRAILINGSPACE(mp);
3091 
3092 	/* Loop through the records and build reply */
3093 	while (cpos < cend && ncookies > 0) {
3094 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3095 			nfsrvfh_t nnsfh;
3096 
3097 			nlen = dp->d_namlen;
3098 			rem = nfsm_rndup(nlen)-nlen;
3099 
3100 			/*
3101 			 * For readdir_and_lookup get the vnode using
3102 			 * the file number.
3103 			 */
3104 			if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
3105 				goto invalid;
3106 			if (nfsrv_composefh(nvp, &nnsfh, true)) {
3107 				vput(nvp);
3108 				goto invalid;
3109 			}
3110 			if (VOP_GETATTR(nvp, vap, cred)) {
3111 				vput(nvp);
3112 				goto invalid;
3113 			}
3114 			vput(nvp);
3115 
3116 			/*
3117 			 * If either the dircount or maxcount will be
3118 			 * exceeded, get out now. Both of these lengths
3119 			 * are calculated conservatively, including all
3120 			 * XDR overheads.
3121 			 */
3122 			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3123 				NFSX_V3POSTOPATTR);
3124 			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3125 			if (len > cnt || dirlen > fullsiz) {
3126 				eofflag = 0;
3127 				break;
3128 			}
3129 
3130 			/*
3131 			 * Build the directory record xdr from
3132 			 * the dirent entry.
3133 			 */
3134 			fp = (struct nfs_fattr *)&fl.fl_fattr;
3135 			nfsm_srvfillattr(vap, fp);
3136 			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3137 			fl.fl_fhok = nfs_true;
3138 			fl.fl_postopok = nfs_true;
3139 			txdr_hyper(*cookiep, fl.fl_off.nfsuquad);
3140 
3141 			nfsm_clget;
3142 			*tl = nfs_true;
3143 			bp += NFSX_UNSIGNED;
3144 			nfsm_clget;
3145 			*tl = txdr_unsigned(dp->d_fileno >> 32);
3146 			bp += NFSX_UNSIGNED;
3147 			nfsm_clget;
3148 			*tl = txdr_unsigned(dp->d_fileno);
3149 			bp += NFSX_UNSIGNED;
3150 			nfsm_clget;
3151 			*tl = txdr_unsigned(nlen);
3152 			bp += NFSX_UNSIGNED;
3153 
3154 			/* And loop around copying the name */
3155 			xfer = nlen;
3156 			cp = dp->d_name;
3157 			while (xfer > 0) {
3158 				nfsm_clget;
3159 				if ((bp + xfer) > be)
3160 					tsiz = be - bp;
3161 				else
3162 					tsiz = xfer;
3163 				memcpy(bp, cp, tsiz);
3164 				bp += tsiz;
3165 				xfer -= tsiz;
3166 				if (xfer > 0)
3167 					cp += tsiz;
3168 			}
3169 			/* And null pad to an int32_t boundary */
3170 			for (i = 0; i < rem; i++)
3171 				*bp++ = '\0';
3172 
3173 			/*
3174 			 * Now copy the flrep structure out.
3175 			 */
3176 			xfer = sizeof(struct flrep);
3177 			cp = (void *)&fl;
3178 			while (xfer > 0) {
3179 				nfsm_clget;
3180 				if ((bp + xfer) > be)
3181 					tsiz = be - bp;
3182 				else
3183 					tsiz = xfer;
3184 				memcpy(bp, cp, tsiz);
3185 				bp += tsiz;
3186 				xfer -= tsiz;
3187 				if (xfer > 0)
3188 					cp += tsiz;
3189 			}
3190 
3191 			/*
3192 			 * ... and filehandle.
3193 			 */
3194 			xfer = NFSRVFH_SIZE(&nnsfh);
3195 			cp = NFSRVFH_DATA(&nnsfh);
3196 			while (xfer > 0) {
3197 				nfsm_clget;
3198 				if ((bp + xfer) > be)
3199 					tsiz = be - bp;
3200 				else
3201 					tsiz = xfer;
3202 				memcpy(bp, cp, tsiz);
3203 				bp += tsiz;
3204 				xfer -= tsiz;
3205 				if (xfer > 0)
3206 					cp += tsiz;
3207 			}
3208 		}
3209 invalid:
3210 		cpos += dp->d_reclen;
3211 		dp = (struct dirent *)cpos;
3212 		cookiep++;
3213 		ncookies--;
3214 	}
3215 	vrele(vp);
3216 	nfsm_clget;
3217 	*tl = nfs_false;
3218 	bp += NFSX_UNSIGNED;
3219 	nfsm_clget;
3220 	if (eofflag)
3221 		*tl = nfs_true;
3222 	else
3223 		*tl = nfs_false;
3224 	bp += NFSX_UNSIGNED;
3225 	if (mp != mb) {
3226 		if (bp < be)
3227 			mp->m_len = bp - mtod(mp, char *);
3228 	} else
3229 		mp->m_len += bp - bpos;
3230 	free((void *)cookies, M_TEMP);
3231 	free((void *)rbuf, M_TEMP);
3232 	nfsm_srvdone;
3233 }
3234 
3235 /*
3236  * nfs commit service
3237  */
3238 int
3239 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3240 {
3241 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3242 	struct mbuf *nam = nfsd->nd_nam;
3243 	char *dpos = nfsd->nd_dpos;
3244 	kauth_cred_t cred = nfsd->nd_cr;
3245 	struct vattr bfor, aft;
3246 	struct vnode *vp;
3247 	nfsrvfh_t nsfh;
3248 	u_int32_t *tl;
3249 	int32_t t1;
3250 	char *bpos;
3251 	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0;
3252 	uint32_t cnt;
3253 	char *cp2;
3254 	struct mbuf *mb, *mreq;
3255 	u_quad_t frev, off, end;
3256 
3257 	nfsm_srvmtofh(&nsfh);
3258 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3259 
3260 	off = fxdr_hyper(tl);
3261 	tl += 2;
3262 	cnt = fxdr_unsigned(uint32_t, *tl);
3263 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3264 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3265 	if (error) {
3266 		nfsm_reply(2 * NFSX_UNSIGNED);
3267 		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3268 		return (0);
3269 	}
3270 	for_ret = VOP_GETATTR(vp, &bfor, cred);
3271 	end = (cnt > 0) ? off + cnt : vp->v_size;
3272 	if (end < off || end > vp->v_size)
3273 		end = vp->v_size;
3274 	if (off < vp->v_size)
3275 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end);
3276 	/* else error == 0, from nfsrv_fhtovp() */
3277 	aft_ret = VOP_GETATTR(vp, &aft, cred);
3278 	vput(vp);
3279 	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3280 	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3281 	if (!error) {
3282 		nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3283 		*tl++ = txdr_unsigned(boottime.tv_sec);
3284 		*tl = txdr_unsigned(boottime.tv_nsec / 1000);
3285 	} else {
3286 		return (0);
3287 	}
3288 	nfsm_srvdone;
3289 }
3290 
3291 /*
3292  * nfs statfs service
3293  */
3294 int
3295 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3296 {
3297 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3298 	struct mbuf *nam = nfsd->nd_nam;
3299 	char *dpos = nfsd->nd_dpos;
3300 	kauth_cred_t cred = nfsd->nd_cr;
3301 	struct statvfs *sf = NULL;
3302 	struct nfs_statfs *sfp;
3303 	u_int32_t *tl;
3304 	int32_t t1;
3305 	char *bpos;
3306 	int error = 0, rdonly, cache = 0, getret = 1;
3307 	int v3 = (nfsd->nd_flag & ND_NFSV3);
3308 	char *cp2;
3309 	struct mbuf *mb, *mreq;
3310 	struct vnode *vp;
3311 	struct vattr at;
3312 	nfsrvfh_t nsfh;
3313 	u_quad_t frev, tval;
3314 
3315 	nfsm_srvmtofh(&nsfh);
3316 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3317 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3318 	if (error) {
3319 		nfsm_reply(NFSX_UNSIGNED);
3320 		nfsm_srvpostop_attr(getret, &at);
3321 		return (0);
3322 	}
3323 	sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK);
3324 	error = VFS_STATVFS(vp->v_mount, sf);
3325 	getret = VOP_GETATTR(vp, &at, cred);
3326 	vput(vp);
3327 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3328 	if (v3)
3329 		nfsm_srvpostop_attr(getret, &at);
3330 	if (error) {
3331 		free(sf, M_TEMP);
3332 		return (0);
3333 	}
3334 	nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3335 	if (v3) {
3336 		tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize);
3337 		txdr_hyper(tval, &sfp->sf_tbytes);
3338 		tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize);
3339 		txdr_hyper(tval, &sfp->sf_fbytes);
3340 		tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize);
3341 		txdr_hyper(tval, &sfp->sf_abytes);
3342 		tval = (u_quad_t)sf->f_files;
3343 		txdr_hyper(tval, &sfp->sf_tfiles);
3344 		tval = (u_quad_t)sf->f_ffree;
3345 		txdr_hyper(tval, &sfp->sf_ffiles);
3346 		txdr_hyper(tval, &sfp->sf_afiles);
3347 		sfp->sf_invarsec = 0;
3348 	} else {
3349 		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3350 		sfp->sf_bsize = txdr_unsigned(sf->f_frsize);
3351 		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3352 		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3353 		sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3354 	}
3355 nfsmout:
3356 	if (sf)
3357 	    free(sf, M_TEMP);
3358 	return error;
3359 }
3360 
3361 /*
3362  * nfs fsinfo service
3363  */
3364 int
3365 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3366 {
3367 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3368 	struct mbuf *nam = nfsd->nd_nam;
3369 	char *dpos = nfsd->nd_dpos;
3370 	kauth_cred_t cred = nfsd->nd_cr;
3371 	u_int32_t *tl;
3372 	struct nfsv3_fsinfo *sip;
3373 	int32_t t1;
3374 	char *bpos;
3375 	int error = 0, rdonly, cache = 0, getret = 1;
3376 	uint32_t maxdata;
3377 	char *cp2;
3378 	struct mbuf *mb, *mreq;
3379 	struct vnode *vp;
3380 	struct vattr at;
3381 	nfsrvfh_t nsfh;
3382 	u_quad_t frev, maxfsize;
3383 	struct statvfs *sb;
3384 
3385 	nfsm_srvmtofh(&nsfh);
3386 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3387 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3388 	if (error) {
3389 		nfsm_reply(NFSX_UNSIGNED);
3390 		nfsm_srvpostop_attr(getret, &at);
3391 		return (0);
3392 	}
3393 
3394 	/* XXX Try to make a guess on the max file size. */
3395 	sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK);
3396 	VFS_STATVFS(vp->v_mount, sb);
3397 	maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1;
3398 	free(sb, M_TEMP);
3399 
3400 	getret = VOP_GETATTR(vp, &at, cred);
3401 	vput(vp);
3402 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3403 	nfsm_srvpostop_attr(getret, &at);
3404 	nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3405 
3406 	/*
3407 	 * XXX
3408 	 * There should be file system VFS OP(s) to get this information.
3409 	 * For now, assume ufs.
3410 	 */
3411 	if (slp->ns_so->so_type == SOCK_DGRAM)
3412 		maxdata = NFS_MAXDGRAMDATA;
3413 	else
3414 		maxdata = NFS_MAXDATA;
3415 	sip->fs_rtmax = txdr_unsigned(maxdata);
3416 	sip->fs_rtpref = txdr_unsigned(maxdata);
3417 	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3418 	sip->fs_wtmax = txdr_unsigned(maxdata);
3419 	sip->fs_wtpref = txdr_unsigned(maxdata);
3420 	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3421 	sip->fs_dtpref = txdr_unsigned(maxdata);
3422 	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3423 	sip->fs_timedelta.nfsv3_sec = 0;
3424 	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3425 	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3426 		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3427 		NFSV3FSINFO_CANSETTIME);
3428 	nfsm_srvdone;
3429 }
3430 
3431 /*
3432  * nfs pathconf service
3433  */
3434 int
3435 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3436 {
3437 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3438 	struct mbuf *nam = nfsd->nd_nam;
3439 	char *dpos = nfsd->nd_dpos;
3440 	kauth_cred_t cred = nfsd->nd_cr;
3441 	u_int32_t *tl;
3442 	struct nfsv3_pathconf *pc;
3443 	int32_t t1;
3444 	char *bpos;
3445 	int error = 0, rdonly, cache = 0, getret = 1;
3446 	register_t linkmax, namemax, chownres, notrunc;
3447 	char *cp2;
3448 	struct mbuf *mb, *mreq;
3449 	struct vnode *vp;
3450 	struct vattr at;
3451 	nfsrvfh_t nsfh;
3452 	u_quad_t frev;
3453 
3454 	nfsm_srvmtofh(&nsfh);
3455 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3456 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3457 	if (error) {
3458 		nfsm_reply(NFSX_UNSIGNED);
3459 		nfsm_srvpostop_attr(getret, &at);
3460 		return (0);
3461 	}
3462 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3463 	if (!error)
3464 		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3465 	if (!error)
3466 		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3467 	if (!error)
3468 		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3469 	getret = VOP_GETATTR(vp, &at, cred);
3470 	vput(vp);
3471 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3472 	nfsm_srvpostop_attr(getret, &at);
3473 	if (error)
3474 		return (0);
3475 	nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3476 
3477 	pc->pc_linkmax = txdr_unsigned(linkmax);
3478 	pc->pc_namemax = txdr_unsigned(namemax);
3479 	pc->pc_notrunc = txdr_unsigned(notrunc);
3480 	pc->pc_chownrestricted = txdr_unsigned(chownres);
3481 
3482 	/*
3483 	 * These should probably be supported by VOP_PATHCONF(), but
3484 	 * until msdosfs is exportable (why would you want to?), the
3485 	 * Unix defaults should be ok.
3486 	 */
3487 	pc->pc_caseinsensitive = nfs_false;
3488 	pc->pc_casepreserving = nfs_true;
3489 	nfsm_srvdone;
3490 }
3491 
3492 /*
3493  * Null operation, used by clients to ping server
3494  */
3495 /* ARGSUSED */
3496 int
3497 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3498     struct lwp *lwp, struct mbuf **mrq)
3499 {
3500 	struct mbuf *mrep = nfsd->nd_mrep;
3501 	char *bpos;
3502 	int error = NFSERR_RETVOID, cache = 0;
3503 	struct mbuf *mb, *mreq;
3504 	u_quad_t frev;
3505 
3506 	nfsm_reply(0);
3507 nfsmout:
3508 	return (0);
3509 }
3510 
3511 /*
3512  * No operation, used for obsolete procedures
3513  */
3514 /* ARGSUSED */
3515 int
3516 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3517     struct lwp *lwp, struct mbuf **mrq)
3518 {
3519 	struct mbuf *mrep = nfsd->nd_mrep;
3520 	char *bpos;
3521 	int error, cache = 0;
3522 	struct mbuf *mb, *mreq;
3523 	u_quad_t frev;
3524 
3525 	if (nfsd->nd_repstat)
3526 		error = nfsd->nd_repstat;
3527 	else
3528 		error = EPROCUNAVAIL;
3529 	nfsm_reply(0);
3530 nfsmout:
3531 	return (0);
3532 }
3533 
3534 /*
3535  * Perform access checking for vnodes obtained from file handles that would
3536  * refer to files already opened by a Unix client. You cannot just use
3537  * vn_writechk() and VOP_ACCESS() for two reasons.
3538  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3539  * 2 - The owner is to be given access irrespective of mode bits for some
3540  *     operations, so that processes that chmod after opening a file don't
3541  *     break. I don't like this because it opens a security hole, but since
3542  *     the nfs server opens a security hole the size of a barn door anyhow,
3543  *     what the heck.
3544  *
3545  * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3546  * will return EPERM instead of EACCES. EPERM is always an error.
3547  */
3548 int
3549 nfsrv_access(struct vnode *vp, int flags, kauth_cred_t cred, int rdonly, struct lwp *lwp, int override)
3550 {
3551 	struct vattr vattr;
3552 	int error;
3553 	if (flags & VWRITE) {
3554 		/* Just vn_writechk() changed to check rdonly */
3555 		/*
3556 		 * Disallow write attempts on read-only file systems;
3557 		 * unless the file is a socket or a block or character
3558 		 * device resident on the file system.
3559 		 */
3560 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3561 			switch (vp->v_type) {
3562 			case VREG:
3563 			case VDIR:
3564 			case VLNK:
3565 				return (EROFS);
3566 			default:
3567 				break;
3568 			}
3569 		}
3570 
3571 		/*
3572 		 * If the vnode is in use as a process's text,
3573 		 * we can't allow writing.
3574 		 */
3575 		if (vp->v_iflag & VI_TEXT)
3576 			return (ETXTBSY);
3577 	}
3578 	error = VOP_GETATTR(vp, &vattr, cred);
3579 	if (error)
3580 		return (error);
3581 	error = VOP_ACCESS(vp, flags, cred);
3582 	/*
3583 	 * Allow certain operations for the owner (reads and writes
3584 	 * on files that are already open).
3585 	 */
3586 	if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid)
3587 		error = 0;
3588 	return error;
3589 }
3590