xref: /netbsd-src/sys/nfs/nfs_serv.c (revision b757af438b42b93f8c6571f026d8b8ef3eaf5fc9)
1 /*	$NetBSD: nfs_serv.c,v 1.163 2012/02/01 02:27:24 matt 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 1.163 2012/02/01 02:27:24 matt 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(pgoff, npages << PAGE_SHIFT,
703 			    slp->ns_so);
704 			if (lva == 0) {
705 				/* fall back to VOP_READ */
706 				goto loan_fail;
707 			}
708 
709 			/* allocate mbuf */
710 			m = m_get(M_WAIT, MT_DATA);
711 			MCLAIM(m, &nfs_mowner);
712 			pgpp = m->m_ext.ext_pgs;
713 
714 			/* loan pages */
715 			error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages,
716 			    pgpp);
717 			if (error) {
718 				sokvafree(lva, npages << PAGE_SHIFT);
719 				m_free(m);
720 				if (error == EBUSY)
721 					goto loan_fail;
722 				goto read_error;
723 			}
724 
725 			/* associate kva to mbuf */
726 			MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)),
727 			    cnt, M_MBUF, soloanfree, slp->ns_so);
728 			m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
729 			m->m_len = cnt;
730 
731 			/* map pages */
732 			for (i = 0; i < npages; i++) {
733 				pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]),
734 				    VM_PROT_READ, 0);
735 				lva += PAGE_SIZE;
736 			}
737 
738 			pmap_update(pmap_kernel());
739 
740 			mb->m_next = m;
741 			mb = m;
742 			error = 0;
743 			uiop->uio_resid = 0;
744 		} else {
745 			struct iovec *iv;
746 			struct iovec *iv2;
747 			struct mbuf *m2;
748 			int siz;
749 loan_fail:
750 			/*
751 			 * Generate the mbuf list with the uio_iov ref. to it.
752 			 */
753 			i = 0;
754 			m = m2 = mb;
755 			while (left > 0) {
756 				siz = min(M_TRAILINGSPACE(m), left);
757 				if (siz > 0) {
758 					left -= siz;
759 					i++;
760 				}
761 				if (left > 0) {
762 					m = m_get(M_WAIT, MT_DATA);
763 					MCLAIM(m, &nfs_mowner);
764 					m_clget(m, M_WAIT);
765 					m->m_len = 0;
766 					m2->m_next = m;
767 					m2 = m;
768 				}
769 			}
770 			iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK);
771 			uiop->uio_iov = iv2 = iv;
772 			m = mb;
773 			left = cnt;
774 			i = 0;
775 			while (left > 0) {
776 				if (m == NULL)
777 					panic("nfsrv_read iov");
778 				siz = min(M_TRAILINGSPACE(m), left);
779 				if (siz > 0) {
780 					iv->iov_base = mtod(m, char *) +
781 					    m->m_len;
782 					iv->iov_len = siz;
783 					m->m_len += siz;
784 					left -= siz;
785 					iv++;
786 					i++;
787 				}
788 				m = m->m_next;
789 			}
790 			uiop->uio_iovcnt = i;
791 			uiop->uio_offset = off;
792 			uiop->uio_resid = cnt;
793 			uiop->uio_rw = UIO_READ;
794 			UIO_SETUP_SYSSPACE(uiop);
795 			error = VOP_READ(vp, uiop, IO_NODELOCKED, cred);
796 			free((void *)iv2, M_TEMP);
797 		}
798 read_error:
799 		if (error || (getret = VOP_GETATTR(vp, &va, cred)) != 0){
800 			if (!error)
801 				error = getret;
802 			m_freem(mreq);
803 			vput(vp);
804 			nfsm_reply(NFSX_POSTOPATTR(v3));
805 			nfsm_srvpostop_attr(getret, &va);
806 			return (0);
807 		}
808 	} else {
809 		uiop->uio_resid = 0;
810 	}
811 	vput(vp);
812 	nfsm_srvfillattr(&va, fp);
813 	len -= uiop->uio_resid;
814 	padlen = nfsm_padlen(len);
815 	if (uiop->uio_resid || padlen)
816 		nfs_zeropad(mb, uiop->uio_resid, padlen);
817 	if (v3) {
818 		/* count */
819 		*tl++ = txdr_unsigned(len);
820 		/* eof */
821 		if (off + len >= va.va_size)
822 			*tl++ = nfs_true;
823 		else
824 			*tl++ = nfs_false;
825 	}
826 	*tl = txdr_unsigned(len);
827 	nfsm_srvdone;
828 }
829 
830 /*
831  * nfs write service
832  */
833 int
834 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
835 {
836 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
837 	struct mbuf *nam = nfsd->nd_nam;
838 	char *dpos = nfsd->nd_dpos;
839 	kauth_cred_t cred = nfsd->nd_cr;
840 	struct iovec *ivp;
841 	int i, cnt;
842 	struct mbuf *mp;
843 	struct nfs_fattr *fp;
844 	struct iovec *iv;
845 	struct vattr va, forat;
846 	u_int32_t *tl;
847 	int32_t t1;
848 	char *bpos;
849 	int error = 0, rdonly, cache = 0, len, forat_ret = 1;
850 	int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
851 	int stable = NFSV3WRITE_FILESYNC;
852 	int v3 = (nfsd->nd_flag & ND_NFSV3);
853 	char *cp2;
854 	struct mbuf *mb, *mreq;
855 	struct vnode *vp;
856 	nfsrvfh_t nsfh;
857 	struct uio io, *uiop = &io;
858 	off_t off;
859 	u_quad_t frev;
860 
861 	if (mrep == NULL) {
862 		*mrq = NULL;
863 		return (0);
864 	}
865 	nfsm_srvmtofh(&nsfh);
866 	if (v3) {
867 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
868 		off = fxdr_hyper(tl);
869 		tl += 3;
870 		stable = fxdr_unsigned(int, *tl++);
871 	} else {
872 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
873 		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
874 		tl += 2;
875 	}
876 	retlen = len = fxdr_unsigned(int32_t, *tl);
877 	cnt = i = 0;
878 
879 	/*
880 	 * For NFS Version 2, it is not obvious what a write of zero length
881 	 * should do, but I might as well be consistent with Version 3,
882 	 * which is to return ok so long as there are no permission problems.
883 	 */
884 	if (len > 0) {
885 		zeroing = 1;
886 		mp = mrep;
887 		while (mp) {
888 			if (mp == md) {
889 				zeroing = 0;
890 				adjust = dpos - mtod(mp, char *);
891 				mp->m_len -= adjust;
892 				if (mp->m_len > 0 && adjust > 0)
893 					NFSMADV(mp, adjust);
894 			}
895 			if (zeroing)
896 				mp->m_len = 0;
897 			else if (mp->m_len > 0) {
898 				i += mp->m_len;
899 				if (i > len) {
900 					mp->m_len -= (i - len);
901 					zeroing	= 1;
902 				}
903 				if (mp->m_len > 0)
904 					cnt++;
905 			}
906 			mp = mp->m_next;
907 		}
908 	}
909 	if (len > NFS_MAXDATA || len < 0 || i < len) {
910 		error = EIO;
911 		nfsm_reply(2 * NFSX_UNSIGNED);
912 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
913 		return (0);
914 	}
915 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
916 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
917 	if (error) {
918 		nfsm_reply(2 * NFSX_UNSIGNED);
919 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
920 		return (0);
921 	}
922 	if (v3)
923 		forat_ret = VOP_GETATTR(vp, &forat, cred);
924 	if (vp->v_type != VREG) {
925 		if (v3)
926 			error = EINVAL;
927 		else
928 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
929 	}
930 	if (!error) {
931 		nqsrv_getl(vp, ND_WRITE);
932 		error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1);
933 	}
934 	if (error) {
935 		vput(vp);
936 		nfsm_reply(NFSX_WCCDATA(v3));
937 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
938 		return (0);
939 	}
940 
941 	if (len > 0) {
942 		ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK);
943 		uiop->uio_iov = iv = ivp;
944 		uiop->uio_iovcnt = cnt;
945 		mp = mrep;
946 		while (mp) {
947 			if (mp->m_len > 0) {
948 				ivp->iov_base = mtod(mp, void *);
949 				ivp->iov_len = mp->m_len;
950 				ivp++;
951 			}
952 			mp = mp->m_next;
953 		}
954 
955 		/*
956 		 * XXX
957 		 * The IO_METASYNC flag indicates that all metadata (and not
958 		 * just enough to ensure data integrity) must be written to
959 		 * stable storage synchronously.
960 		 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
961 		 */
962 		if (stable == NFSV3WRITE_UNSTABLE)
963 			ioflags = IO_NODELOCKED;
964 		else if (stable == NFSV3WRITE_DATASYNC)
965 			ioflags = (IO_SYNC | IO_NODELOCKED);
966 		else
967 			ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
968 		uiop->uio_resid = len;
969 		uiop->uio_rw = UIO_WRITE;
970 		uiop->uio_offset = off;
971 		UIO_SETUP_SYSSPACE(uiop);
972 		error = VOP_WRITE(vp, uiop, ioflags, cred);
973 		nfsstats.srvvop_writes++;
974 		free(iv, M_TEMP);
975 	}
976 	aftat_ret = VOP_GETATTR(vp, &va, cred);
977 	vput(vp);
978 	if (!error)
979 		error = aftat_ret;
980 	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
981 		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
982 	if (v3) {
983 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
984 		if (error)
985 			return (0);
986 		nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
987 		*tl++ = txdr_unsigned(retlen);
988 		if (stable == NFSV3WRITE_UNSTABLE)
989 			*tl++ = txdr_unsigned(stable);
990 		else
991 			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
992 		/*
993 		 * Actually, there is no need to txdr these fields,
994 		 * but it may make the values more human readable,
995 		 * for debugging purposes.
996 		 */
997 		*tl++ = txdr_unsigned(boottime.tv_sec);
998 		*tl = txdr_unsigned(boottime.tv_nsec / 1000);
999 	} else {
1000 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1001 		nfsm_srvfillattr(&va, fp);
1002 	}
1003 	nfsm_srvdone;
1004 }
1005 
1006 /*
1007  * XXX elad: the original NFSW_SAMECRED() macro also made sure the
1008  *	     two nd_flag fields of the descriptors contained
1009  *	     ND_KERBAUTH.
1010  */
1011 static int
1012 nfsrv_samecred(kauth_cred_t cred1, kauth_cred_t cred2)
1013 {
1014 	int i, do_ngroups;
1015 
1016 	if (kauth_cred_geteuid(cred1) != kauth_cred_geteuid(cred2))
1017 		return (0);
1018 	if (kauth_cred_ngroups(cred1) != kauth_cred_ngroups(cred2))
1019 		return (0);
1020 	do_ngroups = kauth_cred_ngroups(cred1);
1021 	for (i = 0; i < do_ngroups; i++)
1022 		if (kauth_cred_group(cred1, i) !=
1023 		    kauth_cred_group(cred2, i))
1024 			return (0);
1025 
1026 	return (1);
1027 }
1028 
1029 static struct nfsrvw_delayhash *
1030 nfsrv_nwdelayhash(struct nfssvc_sock *slp, const nfsrvfh_t *nsfh)
1031 {
1032 	uint32_t hash;
1033 
1034 	hash = hash32_buf(NFSRVFH_DATA(nsfh), NFSRVFH_SIZE(nsfh),
1035 	    HASH32_BUF_INIT);
1036 	return &slp->ns_wdelayhashtbl[hash % NFS_WDELAYHASHSIZ];
1037 }
1038 
1039 /*
1040  * NFS write service with write gathering support. Called when
1041  * nfsrvw_procrastinate > 0.
1042  * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1043  * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1044  * Jan. 1994.
1045  */
1046 int
1047 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1048 {
1049 	struct timeval now;
1050 	struct iovec *ivp;
1051 	struct mbuf *mp;
1052 	struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1053 	struct nfs_fattr *fp;
1054 	int i = 0;
1055 	struct iovec *iov;
1056 	struct nfsrvw_delayhash *wpp;
1057 	kauth_cred_t cred;
1058 	struct vattr va, forat;
1059 	u_int32_t *tl;
1060 	int32_t t1;
1061 	char *bpos, *dpos;
1062 	int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1;
1063 	int ioflags, aftat_ret = 1, adjust, v3, zeroing;
1064 	char *cp2;
1065 	struct mbuf *mb, *mreq, *mrep, *md;
1066 	struct vnode *vp;
1067 	struct uio io, *uiop = &io;
1068 	u_quad_t frev, cur_usec;
1069 
1070 	*mrq = NULL;
1071 	if (*ndp) {
1072 	    nfsd = *ndp;
1073 	    *ndp = NULL;
1074 	    mrep = nfsd->nd_mrep;
1075 	    md = nfsd->nd_md;
1076 	    dpos = nfsd->nd_dpos;
1077 	    cred = nfsd->nd_cr;
1078 	    v3 = (nfsd->nd_flag & ND_NFSV3);
1079 	    LIST_INIT(&nfsd->nd_coalesce);
1080 	    nfsd->nd_mreq = NULL;
1081 	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1082 	    getmicrotime(&now);
1083 	    cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec;
1084 	    nfsd->nd_time = cur_usec + nfsrvw_procrastinate;
1085 
1086 	    /*
1087 	     * Now, get the write header..
1088 	     */
1089 	    nfsm_srvmtofh(&nfsd->nd_fh);
1090 	    if (v3) {
1091 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1092 		nfsd->nd_off = fxdr_hyper(tl);
1093 		tl += 3;
1094 		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1095 	    } else {
1096 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1097 		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1098 		tl += 2;
1099 	    }
1100 	    len = fxdr_unsigned(int32_t, *tl);
1101 	    nfsd->nd_len = len;
1102 	    nfsd->nd_eoff = nfsd->nd_off + len;
1103 
1104 	    /*
1105 	     * Trim the header out of the mbuf list and trim off any trailing
1106 	     * junk so that the mbuf list has only the write data.
1107 	     */
1108 	    zeroing = 1;
1109 	    i = 0;
1110 	    mp = mrep;
1111 	    while (mp) {
1112 		if (mp == md) {
1113 		    zeroing = 0;
1114 		    adjust = dpos - mtod(mp, char *);
1115 		    mp->m_len -= adjust;
1116 		    if (mp->m_len > 0 && adjust > 0)
1117 			NFSMADV(mp, adjust);
1118 		}
1119 		if (zeroing)
1120 		    mp->m_len = 0;
1121 		else {
1122 		    i += mp->m_len;
1123 		    if (i > len) {
1124 			mp->m_len -= (i - len);
1125 			zeroing = 1;
1126 		    }
1127 		}
1128 		mp = mp->m_next;
1129 	    }
1130 	    if (len > NFS_MAXDATA || len < 0  || i < len) {
1131 nfsmout:
1132 		m_freem(mrep);
1133 		error = EIO;
1134 		nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1135 		if (v3)
1136 		    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1137 		nfsd->nd_mreq = mreq;
1138 		nfsd->nd_mrep = NULL;
1139 		nfsd->nd_time = 0;
1140 	    }
1141 
1142 	    /*
1143 	     * Add this entry to the hash and time queues.
1144 	     */
1145 	    owp = NULL;
1146 	    mutex_enter(&nfsd_lock);
1147 	    wp = LIST_FIRST(&slp->ns_tq);
1148 	    while (wp && wp->nd_time < nfsd->nd_time) {
1149 		owp = wp;
1150 		wp = LIST_NEXT(wp, nd_tq);
1151 	    }
1152 	    if (owp) {
1153 		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1154 	    } else {
1155 		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1156 	    }
1157 	    if (nfsd->nd_mrep) {
1158 		wpp = nfsrv_nwdelayhash(slp, &nfsd->nd_fh);
1159 		owp = NULL;
1160 		wp = LIST_FIRST(wpp);
1161 		while (wp && nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) {
1162 		    owp = wp;
1163 		    wp = LIST_NEXT(wp, nd_hash);
1164 		}
1165 		while (wp && wp->nd_off < nfsd->nd_off &&
1166 		    !nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) {
1167 		    owp = wp;
1168 		    wp = LIST_NEXT(wp, nd_hash);
1169 		}
1170 		if (owp) {
1171 		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1172 
1173 		    /*
1174 		     * Search the hash list for overlapping entries and
1175 		     * coalesce.
1176 		     */
1177 		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1178 			wp = LIST_NEXT(nfsd, nd_hash);
1179 			if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr))
1180 			    nfsrvw_coalesce(owp, nfsd);
1181 		    }
1182 		} else {
1183 		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1184 		}
1185 	    }
1186 	    mutex_exit(&nfsd_lock);
1187 	}
1188 
1189 	/*
1190 	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1191 	 * and generate the associated reply mbuf list(s).
1192 	 */
1193 loop1:
1194 	getmicrotime(&now);
1195 	cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec;
1196 	mutex_enter(&nfsd_lock);
1197 	for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1198 		owp = LIST_NEXT(nfsd, nd_tq);
1199 		if (nfsd->nd_time > cur_usec)
1200 		    break;
1201 		if (nfsd->nd_mreq)
1202 		    continue;
1203 		LIST_REMOVE(nfsd, nd_tq);
1204 		LIST_REMOVE(nfsd, nd_hash);
1205 		mutex_exit(&nfsd_lock);
1206 
1207 		mrep = nfsd->nd_mrep;
1208 		nfsd->nd_mrep = NULL;
1209 		cred = nfsd->nd_cr;
1210 		v3 = (nfsd->nd_flag & ND_NFSV3);
1211 		forat_ret = aftat_ret = 1;
1212 		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1213 		    nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH),
1214 		    false);
1215 		if (!error) {
1216 		    if (v3)
1217 			forat_ret = VOP_GETATTR(vp, &forat, cred);
1218 		    if (vp->v_type != VREG) {
1219 			if (v3)
1220 			    error = EINVAL;
1221 			else
1222 			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1223 		    }
1224 		} else
1225 		    vp = NULL;
1226 		if (!error) {
1227 		    nqsrv_getl(vp, ND_WRITE);
1228 		    error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1);
1229 		}
1230 
1231 		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1232 		    ioflags = IO_NODELOCKED;
1233 		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1234 		    ioflags = (IO_SYNC | IO_NODELOCKED);
1235 		else
1236 		    ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1237 		uiop->uio_rw = UIO_WRITE;
1238 		uiop->uio_offset = nfsd->nd_off;
1239 		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1240 		UIO_SETUP_SYSSPACE(uiop);
1241 		if (uiop->uio_resid > 0) {
1242 		    mp = mrep;
1243 		    i = 0;
1244 		    while (mp) {
1245 			if (mp->m_len > 0)
1246 			    i++;
1247 			mp = mp->m_next;
1248 		    }
1249 		    uiop->uio_iovcnt = i;
1250 		    iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1251 		    uiop->uio_iov = ivp = iov;
1252 		    mp = mrep;
1253 		    while (mp) {
1254 			if (mp->m_len > 0) {
1255 			    ivp->iov_base = mtod(mp, void *);
1256 			    ivp->iov_len = mp->m_len;
1257 			    ivp++;
1258 			}
1259 			mp = mp->m_next;
1260 		    }
1261 		    if (!error) {
1262 			error = VOP_WRITE(vp, uiop, ioflags, cred);
1263 			nfsstats.srvvop_writes++;
1264 		    }
1265 		    free((void *)iov, M_TEMP);
1266 		}
1267 		m_freem(mrep);
1268 		if (vp) {
1269 		    aftat_ret = VOP_GETATTR(vp, &va, cred);
1270 		    vput(vp);
1271 		}
1272 
1273 		/*
1274 		 * Loop around generating replies for all write rpcs that have
1275 		 * now been completed.
1276 		 */
1277 		swp = nfsd;
1278 		do {
1279 		    if (error) {
1280 			nfsm_writereply(NFSX_WCCDATA(v3), v3);
1281 			if (v3) {
1282 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1283 			}
1284 		    } else {
1285 			nfsm_writereply(NFSX_PREOPATTR(v3) +
1286 			    NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1287 			    NFSX_WRITEVERF(v3), v3);
1288 			if (v3) {
1289 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1290 			    nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1291 			    *tl++ = txdr_unsigned(nfsd->nd_len);
1292 			    *tl++ = txdr_unsigned(swp->nd_stable);
1293 			    /*
1294 			     * Actually, there is no need to txdr these fields,
1295 			     * but it may make the values more human readable,
1296 			     * for debugging purposes.
1297 			     */
1298 			    *tl++ = txdr_unsigned(boottime.tv_sec);
1299 			    *tl = txdr_unsigned(boottime.tv_nsec / 1000);
1300 			} else {
1301 			    nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1302 			    nfsm_srvfillattr(&va, fp);
1303 			}
1304 		    }
1305 		    nfsd->nd_mreq = mreq;
1306 		    if (nfsd->nd_mrep)
1307 			panic("nfsrv_write: nd_mrep not free");
1308 
1309 		    /*
1310 		     * Done. Put it at the head of the timer queue so that
1311 		     * the final phase can return the reply.
1312 		     */
1313 		    mutex_enter(&nfsd_lock);
1314 		    if (nfsd != swp) {
1315 			nfsd->nd_time = 0;
1316 			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1317 		    }
1318 		    nfsd = LIST_FIRST(&swp->nd_coalesce);
1319 		    if (nfsd) {
1320 			LIST_REMOVE(nfsd, nd_tq);
1321 		    }
1322 		    mutex_exit(&nfsd_lock);
1323 		} while (nfsd);
1324 		swp->nd_time = 0;
1325 
1326 		mutex_enter(&nfsd_lock);
1327 		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1328 		mutex_exit(&nfsd_lock);
1329 		goto loop1;
1330 	}
1331 	mutex_exit(&nfsd_lock);
1332 	nfs_timer_start();
1333 
1334 	/*
1335 	 * Search for a reply to return.
1336 	 */
1337 	mutex_enter(&nfsd_lock);
1338 	LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) {
1339 		if (nfsd->nd_mreq) {
1340 		    LIST_REMOVE(nfsd, nd_tq);
1341 		    *mrq = nfsd->nd_mreq;
1342 		    *ndp = nfsd;
1343 		    break;
1344 		}
1345 	}
1346 	mutex_exit(&nfsd_lock);
1347 	return (0);
1348 }
1349 
1350 /*
1351  * Coalesce the write request nfsd into owp. To do this we must:
1352  * - remove nfsd from the queues
1353  * - merge nfsd->nd_mrep into owp->nd_mrep
1354  * - update the nd_eoff and nd_stable for owp
1355  * - put nfsd on owp's nd_coalesce list
1356  * NB: Must be called at splsoftclock().
1357  */
1358 void
1359 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1360 {
1361         int overlap;
1362         struct mbuf *mp;
1363 	struct nfsrv_descript *m;
1364 
1365 	KASSERT(mutex_owned(&nfsd_lock));
1366 
1367         LIST_REMOVE(nfsd, nd_hash);
1368         LIST_REMOVE(nfsd, nd_tq);
1369         if (owp->nd_eoff < nfsd->nd_eoff) {
1370             overlap = owp->nd_eoff - nfsd->nd_off;
1371             if (overlap < 0)
1372                 panic("nfsrv_coalesce: bad off");
1373             if (overlap > 0)
1374                 m_adj(nfsd->nd_mrep, overlap);
1375             mp = owp->nd_mrep;
1376             while (mp->m_next)
1377                 mp = mp->m_next;
1378             mp->m_next = nfsd->nd_mrep;
1379             owp->nd_eoff = nfsd->nd_eoff;
1380         } else
1381             m_freem(nfsd->nd_mrep);
1382         nfsd->nd_mrep = NULL;
1383         if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1384             owp->nd_stable = NFSV3WRITE_FILESYNC;
1385         else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1386             owp->nd_stable == NFSV3WRITE_UNSTABLE)
1387             owp->nd_stable = NFSV3WRITE_DATASYNC;
1388         LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1389  	/*
1390  	 * nfsd might hold coalesce elements! Move them to owp.
1391  	 * Otherwise, requests may be lost and clients will be stuck.
1392  	 */
1393 	while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) {
1394 		LIST_REMOVE(m, nd_tq);
1395 		LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
1396 	}
1397 }
1398 
1399 /*
1400  * nfs create service
1401  * now does a truncate to 0 length via. setattr if it already exists
1402  */
1403 int
1404 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1405 {
1406 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1407 	struct mbuf *nam = nfsd->nd_nam;
1408 	char *dpos = nfsd->nd_dpos;
1409 	kauth_cred_t cred = nfsd->nd_cr;
1410 	struct nfs_fattr *fp;
1411 	struct vattr va, dirfor, diraft;
1412 	struct nfsv2_sattr *sp;
1413 	u_int32_t *tl;
1414 	struct nameidata nd;
1415 	char *cp;
1416 	int32_t t1;
1417 	char *bpos;
1418 	int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1419 	int rdev = 0, abort = 0;
1420 	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1421 	char *cp2;
1422 	struct mbuf *mb, *mreq;
1423 	struct vnode *vp = NULL, *dirp = NULL;
1424 	nfsrvfh_t nsfh;
1425 	u_quad_t frev, tempsize;
1426 	u_char cverf[NFSX_V3CREATEVERF];
1427 
1428 	nd.ni_cnd.cn_nameiop = 0;
1429 	nfsm_srvmtofh(&nsfh);
1430 	nfsm_srvnamesiz(len);
1431 	nd.ni_cnd.cn_cred = cred;
1432 	nd.ni_cnd.cn_nameiop = CREATE;
1433 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1434 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1435 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1436 	if (dirp && v3) {
1437 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1438 	}
1439 	if (error) {
1440 		nfsm_reply(NFSX_WCCDATA(v3));
1441 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1442 		if (dirp)
1443 			vrele(dirp);
1444 		if (nd.ni_pathbuf != NULL) {
1445 			pathbuf_destroy(nd.ni_pathbuf);
1446 			nd.ni_pathbuf = NULL;
1447 		}
1448 		return (0);
1449 	}
1450 	abort = 1;
1451 	vattr_null(&va);
1452 	if (v3) {
1453 		va.va_mode = 0;
1454 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1455 		how = fxdr_unsigned(int, *tl);
1456 		switch (how) {
1457 		case NFSV3CREATE_GUARDED:
1458 			if (nd.ni_vp) {
1459 				error = EEXIST;
1460 				break;
1461 			}
1462 		case NFSV3CREATE_UNCHECKED:
1463 			nfsm_srvsattr(&va);
1464 			break;
1465 		case NFSV3CREATE_EXCLUSIVE:
1466 			nfsm_dissect(cp, void *, NFSX_V3CREATEVERF);
1467 			memcpy(cverf, cp, NFSX_V3CREATEVERF);
1468 			exclusive_flag = 1;
1469 			break;
1470 		};
1471 		va.va_type = VREG;
1472 	} else {
1473 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1474 		va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1475 		if (va.va_type == VNON)
1476 			va.va_type = VREG;
1477 		va.va_mode = nfstov_mode(sp->sa_mode);
1478 		switch (va.va_type) {
1479 		case VREG:
1480 			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1481 			if (tsize != -1)
1482 				va.va_size = (u_quad_t)tsize;
1483 			break;
1484 		case VCHR:
1485 		case VBLK:
1486 		case VFIFO:
1487 			rdev = fxdr_unsigned(int32_t, sp->sa_size);
1488 			break;
1489 		default:
1490 			break;
1491 		};
1492 	}
1493 
1494 	/*
1495 	 * Iff doesn't exist, create it
1496 	 * otherwise just truncate to 0 length
1497 	 *   should I set the mode too ??
1498 	 */
1499 	if (nd.ni_vp == NULL) {
1500 		if (va.va_type == VREG || va.va_type == VSOCK) {
1501 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1502 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1503 			if (!error) {
1504 				if (exclusive_flag) {
1505 					exclusive_flag = 0;
1506 					vattr_null(&va);
1507 					/*
1508 					 * XXX
1509 					 * assuming NFSX_V3CREATEVERF
1510 					 * == sizeof(nfstime3)
1511 					 */
1512 					fxdr_nfsv3time(cverf, &va.va_atime);
1513 					error = VOP_SETATTR(nd.ni_vp, &va,
1514 						cred);
1515 				}
1516 			}
1517 		} else if (va.va_type == VCHR || va.va_type == VBLK ||
1518 			va.va_type == VFIFO) {
1519 			if (va.va_type == VCHR && rdev == 0xffffffff)
1520 				va.va_type = VFIFO;
1521 			if (va.va_type != VFIFO &&
1522 			    (error = kauth_authorize_system(cred,
1523 			    KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) {
1524 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1525 				vput(nd.ni_dvp);
1526 				abort = 0;
1527 				nfsm_reply(0);
1528 				if (nd.ni_pathbuf != NULL) {
1529 					pathbuf_destroy(nd.ni_pathbuf);
1530 					nd.ni_pathbuf = NULL;
1531 				}
1532 				return (error);
1533 			} else
1534 				va.va_rdev = (dev_t)rdev;
1535 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1536 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd,
1537 			    &va);
1538 			if (error) {
1539 				nfsm_reply(0);
1540 			}
1541 		} else {
1542 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1543 			if (nd.ni_pathbuf != NULL) {
1544 				pathbuf_destroy(nd.ni_pathbuf);
1545 				nd.ni_pathbuf = NULL;
1546 			}
1547 			vput(nd.ni_dvp);
1548 			error = ENXIO;
1549 			abort = 0;
1550 		}
1551 		vp = nd.ni_vp;
1552 	} else {
1553 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1554 		if (nd.ni_pathbuf != NULL) {
1555 			pathbuf_destroy(nd.ni_pathbuf);
1556 			nd.ni_pathbuf = NULL;
1557 		}
1558 		vp = nd.ni_vp;
1559 		if (nd.ni_dvp == vp)
1560 			vrele(nd.ni_dvp);
1561 		else
1562 			vput(nd.ni_dvp);
1563 		abort = 0;
1564 		if (!error && va.va_size != -1) {
1565 			error = nfsrv_access(vp, VWRITE, cred,
1566 			    (nd.ni_cnd.cn_flags & RDONLY), lwp, 0);
1567 			if (!error) {
1568 				nqsrv_getl(vp, ND_WRITE);
1569 				tempsize = va.va_size;
1570 				vattr_null(&va);
1571 				va.va_size = tempsize;
1572 				error = VOP_SETATTR(vp, &va, cred);
1573 			}
1574 		}
1575 		if (error)
1576 			vput(vp);
1577 	}
1578 	if (!error) {
1579 		error = nfsrv_composefh(vp, &nsfh, v3);
1580 		if (!error)
1581 			error = VOP_GETATTR(vp, &va, cred);
1582 		vput(vp);
1583 	}
1584 	if (v3) {
1585 		if (exclusive_flag && !error) {
1586 			/*
1587 			 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3)
1588 			 */
1589 			char oldverf[NFSX_V3CREATEVERF];
1590 
1591 			txdr_nfsv3time(&va.va_atime, oldverf);
1592 			if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF))
1593 				error = EEXIST;
1594 		}
1595 		if (dirp) {
1596 			vn_lock(dirp, LK_SHARED | LK_RETRY);
1597 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1598 			VOP_UNLOCK(dirp);
1599 		}
1600 	}
1601 	if (dirp) {
1602 		vrele(dirp);
1603 		dirp = NULL;
1604 	}
1605 	if (nd.ni_pathbuf != NULL) {
1606 		pathbuf_destroy(nd.ni_pathbuf);
1607 		nd.ni_pathbuf = NULL;
1608 	}
1609 	abort = 0;
1610 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1611 	if (v3) {
1612 		if (!error) {
1613 			nfsm_srvpostop_fh(&nsfh);
1614 			nfsm_srvpostop_attr(0, &va);
1615 		}
1616 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1617 	} else {
1618 		nfsm_srvfhtom(&nsfh, v3);
1619 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1620 		nfsm_srvfillattr(&va, fp);
1621 	}
1622 	return (0);
1623 nfsmout:
1624 	if (dirp)
1625 		vrele(dirp);
1626 	if (abort) {
1627 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1628 		if (nd.ni_dvp == nd.ni_vp)
1629 			vrele(nd.ni_dvp);
1630 		else
1631 			vput(nd.ni_dvp);
1632 		if (nd.ni_vp)
1633 			vput(nd.ni_vp);
1634 	}
1635 	if (nd.ni_pathbuf != NULL) {
1636 		pathbuf_destroy(nd.ni_pathbuf);
1637 		nd.ni_pathbuf = NULL;
1638 	}
1639 	return (error);
1640 }
1641 
1642 /*
1643  * nfs v3 mknod service
1644  */
1645 int
1646 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1647 {
1648 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1649 	struct mbuf *nam = nfsd->nd_nam;
1650 	char *dpos = nfsd->nd_dpos;
1651 	kauth_cred_t cred = nfsd->nd_cr;
1652 	struct vattr va, dirfor, diraft;
1653 	u_int32_t *tl;
1654 	struct nameidata nd;
1655 	int32_t t1;
1656 	char *bpos;
1657 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
1658 	int abort = 0;
1659 	u_int32_t major, minor;
1660 	enum vtype vtyp;
1661 	char *cp2;
1662 	struct mbuf *mb, *mreq;
1663 	struct vnode *vp, *dirp = (struct vnode *)0;
1664 	nfsrvfh_t nsfh;
1665 	u_quad_t frev;
1666 
1667 	nd.ni_cnd.cn_nameiop = 0;
1668 	nfsm_srvmtofh(&nsfh);
1669 	nfsm_srvnamesiz(len);
1670 	nd.ni_cnd.cn_cred = cred;
1671 	nd.ni_cnd.cn_nameiop = CREATE;
1672 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1673 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1674 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1675 	if (dirp)
1676 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1677 	if (error) {
1678 		nfsm_reply(NFSX_WCCDATA(1));
1679 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1680 		if (dirp)
1681 			vrele(dirp);
1682 		if (nd.ni_pathbuf != NULL) {
1683 			pathbuf_destroy(nd.ni_pathbuf);
1684 			nd.ni_pathbuf = NULL;
1685 		}
1686 		return (0);
1687 	}
1688 	abort = 1;
1689 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1690 	vtyp = nfsv3tov_type(*tl);
1691 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1692 		error = NFSERR_BADTYPE;
1693 		goto abort;
1694 	}
1695 	vattr_null(&va);
1696 	va.va_mode = 0;
1697 	nfsm_srvsattr(&va);
1698 	if (vtyp == VCHR || vtyp == VBLK) {
1699 		dev_t rdev;
1700 
1701 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1702 		major = fxdr_unsigned(u_int32_t, *tl++);
1703 		minor = fxdr_unsigned(u_int32_t, *tl);
1704 		rdev = makedev(major, minor);
1705 		if (major(rdev) != major || minor(rdev) != minor) {
1706 			error = EINVAL;
1707 			goto abort;
1708 		}
1709 		va.va_rdev = rdev;
1710 	}
1711 
1712 	/*
1713 	 * Iff doesn't exist, create it.
1714 	 */
1715 	if (nd.ni_vp) {
1716 		error = EEXIST;
1717 abort:
1718 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1719 		if (nd.ni_dvp == nd.ni_vp)
1720 			vrele(nd.ni_dvp);
1721 		else
1722 			vput(nd.ni_dvp);
1723 		if (nd.ni_vp)
1724 			vput(nd.ni_vp);
1725 		if (nd.ni_pathbuf != NULL) {
1726 			pathbuf_destroy(nd.ni_pathbuf);
1727 			nd.ni_pathbuf = NULL;
1728 		}
1729 		goto out;
1730 	}
1731 	va.va_type = vtyp;
1732 	if (vtyp == VSOCK) {
1733 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1734 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1735 	} else {
1736 		if (va.va_type != VFIFO &&
1737 		    (error = kauth_authorize_system(cred,
1738 		    KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) {
1739 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1740 			vput(nd.ni_dvp);
1741 			goto out;
1742 		}
1743 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1744 		error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1745 		if (error)
1746 			goto out;
1747 	}
1748 out:
1749 	vp = nd.ni_vp;
1750 	if (!error) {
1751 		error = nfsrv_composefh(vp, &nsfh, true);
1752 		if (!error)
1753 			error = VOP_GETATTR(vp, &va, cred);
1754 		vput(vp);
1755 	}
1756 	if (dirp) {
1757 		vn_lock(dirp, LK_SHARED | LK_RETRY);
1758 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1759 		VOP_UNLOCK(dirp);
1760 		vrele(dirp);
1761 		dirp = NULL;
1762 	}
1763 	if (nd.ni_pathbuf != NULL) {
1764 		pathbuf_destroy(nd.ni_pathbuf);
1765 		nd.ni_pathbuf = NULL;
1766 	}
1767 	abort = 0;
1768 	nfsm_reply(NFSX_SRVFH(&nsfh, true) + NFSX_POSTOPATTR(1) +
1769 	    NFSX_WCCDATA(1));
1770 	if (!error) {
1771 		nfsm_srvpostop_fh(&nsfh);
1772 		nfsm_srvpostop_attr(0, &va);
1773 	}
1774 	nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1775 	return (0);
1776 nfsmout:
1777 	if (abort) {
1778 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1779 		if (nd.ni_dvp == nd.ni_vp)
1780 			vrele(nd.ni_dvp);
1781 		else
1782 			vput(nd.ni_dvp);
1783 		if (nd.ni_vp)
1784 			vput(nd.ni_vp);
1785 	}
1786 	if (nd.ni_pathbuf != NULL) {
1787 		pathbuf_destroy(nd.ni_pathbuf);
1788 		nd.ni_pathbuf = NULL;
1789 	}
1790 	if (dirp)
1791 		vrele(dirp);
1792 	return (error);
1793 }
1794 
1795 /*
1796  * nfs remove service
1797  */
1798 int
1799 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1800 {
1801 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1802 	struct mbuf *nam = nfsd->nd_nam;
1803 	char *dpos = nfsd->nd_dpos;
1804 	kauth_cred_t cred = nfsd->nd_cr;
1805 	struct nameidata nd;
1806 	u_int32_t *tl;
1807 	int32_t t1;
1808 	char *bpos;
1809 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
1810 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1811 	char *cp2;
1812 	struct mbuf *mb, *mreq;
1813 	struct vnode *vp, *dirp;
1814 	struct vattr dirfor, diraft;
1815 	nfsrvfh_t nsfh;
1816 	u_quad_t frev;
1817 
1818 #ifndef nolint
1819 	vp = (struct vnode *)0;
1820 #endif
1821 	nfsm_srvmtofh(&nsfh);
1822 	nfsm_srvnamesiz(len);
1823 	nd.ni_cnd.cn_cred = cred;
1824 	nd.ni_cnd.cn_nameiop = DELETE;
1825 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1826 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1827 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1828 	if (error == 0 && dirp && v3) {
1829 		if (nd.ni_dvp == nd.ni_vp)
1830 			vn_lock(dirp, LK_SHARED | LK_RETRY);
1831 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1832 		if (nd.ni_dvp == nd.ni_vp)
1833 			VOP_UNLOCK(dirp);
1834 	}
1835 	if (!error) {
1836 		vp = nd.ni_vp;
1837 		if (vp->v_type == VDIR) {
1838 			error = EPERM;
1839 			goto out;
1840 		}
1841 		/*
1842 		 * The root of a mounted filesystem cannot be deleted.
1843 		 */
1844 		if (vp->v_vflag & VV_ROOT) {
1845 			error = EBUSY;
1846 		}
1847 out:
1848 		if (!error) {
1849 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1850 			nqsrv_getl(vp, ND_WRITE);
1851 			error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1852 		} else {
1853 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1854 			if (nd.ni_dvp == vp)
1855 				vrele(nd.ni_dvp);
1856 			else
1857 				vput(nd.ni_dvp);
1858 			vput(vp);
1859 		}
1860 	}
1861 	if (nd.ni_pathbuf != NULL) {
1862 		pathbuf_destroy(nd.ni_pathbuf);
1863 		nd.ni_pathbuf = NULL;
1864 	}
1865 	if (dirp) {
1866 		if (v3) {
1867 			vn_lock(dirp, LK_SHARED | LK_RETRY);
1868 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1869 			VOP_UNLOCK(dirp);
1870 		}
1871 		vrele(dirp);
1872 	}
1873 	nfsm_reply(NFSX_WCCDATA(v3));
1874 	if (v3) {
1875 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1876 		return (0);
1877 	}
1878 	nfsm_srvdone;
1879 }
1880 
1881 /*
1882  * nfs rename service
1883  */
1884 int
1885 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1886 {
1887 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1888 	struct mbuf *nam = nfsd->nd_nam;
1889 	char *dpos = nfsd->nd_dpos;
1890 	kauth_cred_t cred = nfsd->nd_cr;
1891 	u_int32_t *tl;
1892 	int32_t t1;
1893 	char *bpos;
1894 	int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1;
1895 	uint32_t len, len2;
1896 	int tdirfor_ret = 1, tdiraft_ret = 1;
1897 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1898 	char *cp2;
1899 	struct mbuf *mb, *mreq;
1900 	struct nameidata fromnd, tond;
1901 	struct vnode *fvp, *tvp, *tdvp;
1902 	struct vnode *fdirp = NULL, *tdirp = NULL;
1903 	struct mount *localfs = NULL;
1904 	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1905 	nfsrvfh_t fnsfh, tnsfh;
1906 	u_quad_t frev;
1907 	uid_t saved_uid;
1908 
1909 #ifndef nolint
1910 	fvp = (struct vnode *)0;
1911 #endif
1912 	fromnd.ni_cnd.cn_nameiop = 0;
1913 	tond.ni_cnd.cn_nameiop = 0;
1914 	nfsm_srvmtofh(&fnsfh);
1915 	nfsm_srvnamesiz(len);
1916 	/*
1917 	 * Remember our original uid so that we can reset cr_uid before
1918 	 * the second nfs_namei() call, in case it is remapped.
1919 	 */
1920 	saved_uid = kauth_cred_geteuid(cred);
1921 	fromnd.ni_cnd.cn_cred = cred;
1922 	fromnd.ni_cnd.cn_nameiop = DELETE;
1923 	fromnd.ni_cnd.cn_flags = LOCKPARENT | INRENAME;
1924 	error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md,
1925 		&dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1926 	if (error == 0 && fdirp && v3) {
1927 		if (fromnd.ni_dvp == fromnd.ni_vp)
1928 			vn_lock(fdirp, LK_SHARED | LK_RETRY);
1929 		fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred);
1930 		if (fromnd.ni_dvp == fromnd.ni_vp)
1931 			VOP_UNLOCK(fdirp);
1932 	}
1933 	if (error) {
1934 		nfsm_reply(2 * NFSX_WCCDATA(v3));
1935 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1936 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1937 		if (fdirp)
1938 			vrele(fdirp);
1939 		if (fromnd.ni_pathbuf != NULL) {
1940 			pathbuf_destroy(fromnd.ni_pathbuf);
1941 		}
1942 		return (0);
1943 	}
1944 	if (fromnd.ni_dvp != fromnd.ni_vp) {
1945 		VOP_UNLOCK(fromnd.ni_dvp);
1946 	}
1947 	fvp = fromnd.ni_vp;
1948 
1949 	localfs = fvp->v_mount;
1950 	error = VFS_RENAMELOCK_ENTER(localfs);
1951 	if (error) {
1952 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1953 		vrele(fromnd.ni_dvp);
1954 		vrele(fvp);
1955 		goto out1;
1956 	}
1957 
1958 	/* Copied, regrettably, from vfs_syscalls.c (q.v.) */
1959 	vrele(fvp);
1960 	if ((fromnd.ni_cnd.cn_namelen == 1 &&
1961 	     fromnd.ni_cnd.cn_nameptr[0] == '.') ||
1962 	    (fromnd.ni_cnd.cn_namelen == 2 &&
1963 	     fromnd.ni_cnd.cn_nameptr[0] == '.' &&
1964 	     fromnd.ni_cnd.cn_nameptr[1] == '.')) {
1965 		error = EINVAL;
1966 		VFS_RENAMELOCK_EXIT(localfs);
1967 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1968 		vrele(fromnd.ni_dvp);
1969 		goto out1;
1970 	}
1971 	vn_lock(fromnd.ni_dvp, LK_EXCLUSIVE | LK_RETRY);
1972 	error = relookup(fromnd.ni_dvp, &fromnd.ni_vp, &fromnd.ni_cnd, 0);
1973 	if (error) {
1974 		VOP_UNLOCK(fromnd.ni_dvp);
1975 		VFS_RENAMELOCK_EXIT(localfs);
1976 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1977 		vrele(fromnd.ni_dvp);
1978 		goto out1;
1979 	}
1980 	VOP_UNLOCK(fromnd.ni_vp);
1981 	if (fromnd.ni_dvp != fromnd.ni_vp)
1982 		VOP_UNLOCK(fromnd.ni_dvp);
1983 	fvp = fromnd.ni_vp;
1984 
1985 	nfsm_srvmtofh(&tnsfh);
1986 	if (v3) {
1987 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1988 		len2 = fxdr_unsigned(uint32_t, *tl);
1989 		/* len2 will be checked by nfs_namei */
1990 	}
1991 	else {
1992 		/* NFSv2 */
1993 		nfsm_strsiz(len2, NFS_MAXNAMLEN);
1994 	}
1995 	kauth_cred_seteuid(cred, saved_uid);
1996 	tond.ni_cnd.cn_cred = cred;
1997 	tond.ni_cnd.cn_nameiop = RENAME;
1998 	tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | INRENAME;
1999 	error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md,
2000 		&dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2001 	if (tdirp && v3) {
2002 		tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred);
2003 	}
2004 	if (error) {
2005 		VFS_RENAMELOCK_EXIT(localfs);
2006 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2007 		vrele(fromnd.ni_dvp);
2008 		vrele(fvp);
2009 		goto out1;
2010 	}
2011 	tdvp = tond.ni_dvp;
2012 	tvp = tond.ni_vp;
2013 	if (tvp != NULL) {
2014 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2015 			if (v3)
2016 				error = EEXIST;
2017 			else
2018 				error = EISDIR;
2019 			goto out;
2020 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2021 			if (v3)
2022 				error = EEXIST;
2023 			else
2024 				error = ENOTDIR;
2025 			goto out;
2026 		}
2027 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
2028 			if (v3)
2029 				error = EXDEV;
2030 			else
2031 				error = ENOTEMPTY;
2032 			goto out;
2033 		}
2034 	}
2035 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
2036 		if (v3)
2037 			error = EXDEV;
2038 		else
2039 			error = ENOTEMPTY;
2040 		goto out;
2041 	}
2042 	if (fvp->v_mount != tdvp->v_mount) {
2043 		if (v3)
2044 			error = EXDEV;
2045 		else
2046 			error = ENOTEMPTY;
2047 		goto out;
2048 	}
2049 	if (fvp == tdvp) {
2050 		if (v3)
2051 			error = EINVAL;
2052 		else
2053 			error = ENOTEMPTY;
2054 	}
2055 	/*
2056 	 * If source is the same as the destination (that is the
2057 	 * same vnode with the same name in the same directory),
2058 	 * then there is nothing to do.
2059 	 */
2060 	if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2061 	    fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2062 	    !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2063 	      fromnd.ni_cnd.cn_namelen))
2064 		error = -1;
2065 out:
2066 	if (!error) {
2067 		nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
2068 		nqsrv_getl(tdvp, ND_WRITE);
2069 		if (tvp) {
2070 			nqsrv_getl(tvp, ND_WRITE);
2071 		}
2072 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2073 				   tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2074 		VFS_RENAMELOCK_EXIT(localfs);
2075 	} else {
2076 		VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd);
2077 		if (tdvp == tvp)
2078 			vrele(tdvp);
2079 		else
2080 			vput(tdvp);
2081 		if (tvp)
2082 			vput(tvp);
2083 		VFS_RENAMELOCK_EXIT(localfs);
2084 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2085 		vrele(fromnd.ni_dvp);
2086 		vrele(fvp);
2087 		if (error == -1)
2088 			error = 0;
2089 	}
2090 	if (tond.ni_pathbuf != NULL) {
2091 		pathbuf_destroy(tond.ni_pathbuf);
2092 		tond.ni_pathbuf = NULL;
2093 	}
2094 	tond.ni_cnd.cn_nameiop = 0;
2095 out1:
2096 	if (fdirp) {
2097 		if (v3) {
2098 			vn_lock(fdirp, LK_SHARED | LK_RETRY);
2099 			fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred);
2100 			VOP_UNLOCK(fdirp);
2101 		}
2102 		vrele(fdirp);
2103 		fdirp = NULL;
2104 	}
2105 	if (tdirp) {
2106 		if (v3) {
2107 			vn_lock(tdirp, LK_SHARED | LK_RETRY);
2108 			tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred);
2109 			VOP_UNLOCK(tdirp);
2110 		}
2111 		vrele(tdirp);
2112 		tdirp = NULL;
2113 	}
2114 	pathbuf_destroy(fromnd.ni_pathbuf);
2115 	fromnd.ni_pathbuf = NULL;
2116 	fromnd.ni_cnd.cn_nameiop = 0;
2117 	localfs = NULL;
2118 	nfsm_reply(2 * NFSX_WCCDATA(v3));
2119 	if (v3) {
2120 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2121 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2122 	}
2123 	return (0);
2124 
2125 nfsmout:
2126 	if (fdirp)
2127 		vrele(fdirp);
2128 #ifdef notdef
2129 	if (tdirp)
2130 		vrele(tdirp);
2131 #endif
2132 	if (tond.ni_cnd.cn_nameiop) {
2133 		if (tond.ni_pathbuf != NULL) {
2134 			pathbuf_destroy(tond.ni_pathbuf);
2135 			tond.ni_pathbuf = NULL;
2136 		}
2137 	}
2138 	if (localfs) {
2139 		VFS_RENAMELOCK_EXIT(localfs);
2140 	}
2141 	if (fromnd.ni_cnd.cn_nameiop) {
2142 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2143 		if (fromnd.ni_pathbuf != NULL) {
2144 			pathbuf_destroy(fromnd.ni_pathbuf);
2145 			fromnd.ni_pathbuf = NULL;
2146 		}
2147 		vrele(fromnd.ni_dvp);
2148 		vrele(fvp);
2149 	}
2150 	return (error);
2151 }
2152 
2153 /*
2154  * nfs link service
2155  */
2156 int
2157 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2158 {
2159 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2160 	struct mbuf *nam = nfsd->nd_nam;
2161 	char *dpos = nfsd->nd_dpos;
2162 	kauth_cred_t cred = nfsd->nd_cr;
2163 	struct nameidata nd;
2164 	u_int32_t *tl;
2165 	int32_t t1;
2166 	char *bpos;
2167 	int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2168 	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2169 	char *cp2;
2170 	struct mbuf *mb, *mreq;
2171 	struct vnode *vp, *xp, *dirp = (struct vnode *)0;
2172 	struct vattr dirfor, diraft, at;
2173 	nfsrvfh_t nsfh, dnsfh;
2174 	u_quad_t frev;
2175 
2176 	nfsm_srvmtofh(&nsfh);
2177 	nfsm_srvmtofh(&dnsfh);
2178 	nfsm_srvnamesiz(len);
2179 	error = nfsrv_fhtovp(&nsfh, false, &vp, cred, slp, nam,
2180 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2181 	if (error) {
2182 		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2183 		nfsm_srvpostop_attr(getret, &at);
2184 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2185 		return (0);
2186 	}
2187 	if (vp->v_type == VDIR) {
2188 		error = EPERM;
2189 		goto out1;
2190 	}
2191 	nd.ni_cnd.cn_cred = cred;
2192 	nd.ni_cnd.cn_nameiop = CREATE;
2193 	nd.ni_cnd.cn_flags = LOCKPARENT;
2194 	error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos,
2195 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2196 	if (dirp && v3) {
2197 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2198 	}
2199 	if (error)
2200 		goto out1;
2201 	xp = nd.ni_vp;
2202 	if (xp != NULL) {
2203 		error = EEXIST;
2204 		goto out;
2205 	}
2206 	xp = nd.ni_dvp;
2207 	if (vp->v_mount != xp->v_mount)
2208 		error = EXDEV;
2209 out:
2210 	if (!error) {
2211 		nqsrv_getl(vp, ND_WRITE);
2212 		nqsrv_getl(xp, ND_WRITE);
2213 		error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2214 	} else {
2215 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2216 		if (nd.ni_dvp == nd.ni_vp)
2217 			vrele(nd.ni_dvp);
2218 		else
2219 			vput(nd.ni_dvp);
2220 		if (nd.ni_vp)
2221 			vrele(nd.ni_vp);
2222 	}
2223 out1:
2224 	if (v3) {
2225 		vn_lock(vp, LK_SHARED | LK_RETRY);
2226 		getret = VOP_GETATTR(vp, &at, cred);
2227 		VOP_UNLOCK(vp);
2228 	}
2229 	if (dirp) {
2230 		if (v3) {
2231 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2232 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2233 			VOP_UNLOCK(dirp);
2234 		}
2235 		vrele(dirp);
2236 	}
2237 	vrele(vp);
2238 	if (nd.ni_pathbuf != NULL) {
2239 		pathbuf_destroy(nd.ni_pathbuf);
2240 		nd.ni_pathbuf = NULL;
2241 	}
2242 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2243 	if (v3) {
2244 		nfsm_srvpostop_attr(getret, &at);
2245 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2246 		return (0);
2247 	}
2248 	nfsm_srvdone;
2249 }
2250 
2251 /*
2252  * nfs symbolic link service
2253  */
2254 int
2255 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2256 {
2257 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2258 	struct mbuf *nam = nfsd->nd_nam;
2259 	char *dpos = nfsd->nd_dpos;
2260 	kauth_cred_t cred = nfsd->nd_cr;
2261 	struct vattr va, dirfor, diraft;
2262 	struct nameidata nd;
2263 	u_int32_t *tl;
2264 	int32_t t1;
2265 	struct nfsv2_sattr *sp;
2266 	char *bpos, *pathcp = NULL, *cp2;
2267 	struct uio io;
2268 	struct iovec iv;
2269 	int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1, abort = 0;
2270 	uint32_t len, len2;
2271 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2272 	struct mbuf *mb, *mreq;
2273 	struct vnode *dirp = (struct vnode *)0;
2274 	nfsrvfh_t nsfh;
2275 	u_quad_t frev;
2276 
2277 	nd.ni_cnd.cn_nameiop = 0;
2278 	nfsm_srvmtofh(&nsfh);
2279 	nfsm_srvnamesiz(len);
2280 	nd.ni_cnd.cn_cred = cred;
2281 	nd.ni_cnd.cn_nameiop = CREATE;
2282 	nd.ni_cnd.cn_flags = LOCKPARENT;
2283 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2284 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2285 	if (dirp && v3) {
2286 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2287 	}
2288 	if (error)
2289 		goto out;
2290 	abort = 1;
2291 	vattr_null(&va);
2292 	va.va_type = VLNK;
2293 	if (v3) {
2294 		va.va_mode = 0;
2295 		nfsm_srvsattr(&va);
2296 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
2297 		len2 = fxdr_unsigned(uint32_t, *tl);
2298 		if (len2 > PATH_MAX) {
2299 			/* XXX should check _PC_NO_TRUNC */
2300 			error = ENAMETOOLONG;
2301 			goto abortop;
2302 		}
2303 	}
2304 	else {
2305 		/* NFSv2 */
2306 		nfsm_strsiz(len2, NFS_MAXPATHLEN);
2307 	}
2308 	pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2309 	iv.iov_base = pathcp;
2310 	iv.iov_len = len2;
2311 	io.uio_resid = len2;
2312 	io.uio_offset = 0;
2313 	io.uio_iov = &iv;
2314 	io.uio_iovcnt = 1;
2315 	io.uio_rw = UIO_READ;
2316 	UIO_SETUP_SYSSPACE(&io);
2317 	nfsm_mtouio(&io, len2);
2318 	if (!v3) {
2319 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2320 		va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
2321 	}
2322 	*(pathcp + len2) = '\0';
2323 	if (nd.ni_vp) {
2324 		error = EEXIST;
2325 abortop:
2326 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2327 		if (nd.ni_dvp == nd.ni_vp)
2328 			vrele(nd.ni_dvp);
2329 		else
2330 			vput(nd.ni_dvp);
2331 		if (nd.ni_vp)
2332 			vrele(nd.ni_vp);
2333 		goto out;
2334 	}
2335 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2336 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp);
2337 	if (!error) {
2338 	    if (v3) {
2339 		error = nfsrv_composefh(nd.ni_vp, &nsfh, v3);
2340 		if (!error)
2341 		    error = VOP_GETATTR(nd.ni_vp, &va, cred);
2342 		vput(nd.ni_vp);
2343 	    } else {
2344 		vput(nd.ni_vp);
2345 	    }
2346 	}
2347 out:
2348 	if (pathcp)
2349 		free(pathcp, M_TEMP);
2350 	if (dirp) {
2351 		if (v3) {
2352 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2353 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2354 			VOP_UNLOCK(dirp);
2355 		}
2356 		vrele(dirp);
2357 		dirp = NULL;
2358 	}
2359 	if (nd.ni_pathbuf != NULL) {
2360 		pathbuf_destroy(nd.ni_pathbuf);
2361 		nd.ni_pathbuf = NULL;
2362 	}
2363 	abort = 0;
2364 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) +
2365 	    NFSX_WCCDATA(v3));
2366 	if (v3) {
2367 		if (!error) {
2368 			nfsm_srvpostop_fh(&nsfh);
2369 			nfsm_srvpostop_attr(0, &va);
2370 		}
2371 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2372 	}
2373 	return (0);
2374 nfsmout:
2375 	if (abort) {
2376 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2377 		if (nd.ni_dvp == nd.ni_vp)
2378 			vrele(nd.ni_dvp);
2379 		else
2380 			vput(nd.ni_dvp);
2381 		if (nd.ni_vp)
2382 			vrele(nd.ni_vp);
2383 		if (nd.ni_pathbuf != NULL) {
2384 			pathbuf_destroy(nd.ni_pathbuf);
2385 			nd.ni_pathbuf = NULL;
2386 		}
2387 	}
2388 	if (dirp)
2389 		vrele(dirp);
2390 	if (pathcp)
2391 		free(pathcp, M_TEMP);
2392 	return (error);
2393 }
2394 
2395 /*
2396  * nfs mkdir service
2397  */
2398 int
2399 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2400 {
2401 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2402 	struct mbuf *nam = nfsd->nd_nam;
2403 	char *dpos = nfsd->nd_dpos;
2404 	kauth_cred_t cred = nfsd->nd_cr;
2405 	struct vattr va, dirfor, diraft;
2406 	struct nfs_fattr *fp;
2407 	struct nameidata nd;
2408 	char *cp;
2409 	u_int32_t *tl;
2410 	int32_t t1;
2411 	char *bpos;
2412 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2413 	int abort = 0;
2414 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2415 	char *cp2;
2416 	struct mbuf *mb, *mreq;
2417 	struct vnode *vp, *dirp = (struct vnode *)0;
2418 	nfsrvfh_t nsfh;
2419 	u_quad_t frev;
2420 
2421 	nfsm_srvmtofh(&nsfh);
2422 	nfsm_srvnamesiz(len);
2423 	nd.ni_cnd.cn_cred = cred;
2424 	nd.ni_cnd.cn_nameiop = CREATE;
2425 	nd.ni_cnd.cn_flags = LOCKPARENT;
2426 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2427 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2428 	if (dirp && v3) {
2429 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2430 	}
2431 	if (error) {
2432 		if (nd.ni_pathbuf != NULL) {
2433 			pathbuf_destroy(nd.ni_pathbuf);
2434 			nd.ni_pathbuf = NULL;
2435 		}
2436 		nfsm_reply(NFSX_WCCDATA(v3));
2437 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2438 		if (dirp)
2439 			vrele(dirp);
2440 		return (0);
2441 	}
2442 	abort = 1;
2443 	vattr_null(&va);
2444 	if (v3) {
2445 		va.va_mode = 0;
2446 		nfsm_srvsattr(&va);
2447 	} else {
2448 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2449 		va.va_mode = nfstov_mode(*tl++);
2450 	}
2451 	va.va_type = VDIR;
2452 	vp = nd.ni_vp;
2453 	if (vp != NULL) {
2454 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2455 		if (nd.ni_dvp == vp)
2456 			vrele(nd.ni_dvp);
2457 		else
2458 			vput(nd.ni_dvp);
2459 		vrele(vp);
2460 		error = EEXIST;
2461 		goto out;
2462 	}
2463 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2464 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
2465 	if (!error) {
2466 		vp = nd.ni_vp;
2467 		error = nfsrv_composefh(vp, &nsfh, v3);
2468 		if (!error)
2469 			error = VOP_GETATTR(vp, &va, cred);
2470 		vput(vp);
2471 	}
2472 out:
2473 	if (dirp) {
2474 		if (v3) {
2475 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2476 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2477 			VOP_UNLOCK(dirp);
2478 		}
2479 		vrele(dirp);
2480 		dirp = NULL;
2481 	}
2482 	if (nd.ni_pathbuf != NULL) {
2483 		pathbuf_destroy(nd.ni_pathbuf);
2484 		nd.ni_pathbuf = NULL;
2485 	}
2486 	abort = 0;
2487 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) +
2488 	    NFSX_WCCDATA(v3));
2489 	if (v3) {
2490 		if (!error) {
2491 			nfsm_srvpostop_fh(&nsfh);
2492 			nfsm_srvpostop_attr(0, &va);
2493 		}
2494 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2495 	} else {
2496 		nfsm_srvfhtom(&nsfh, v3);
2497 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2498 		nfsm_srvfillattr(&va, fp);
2499 	}
2500 	return (0);
2501 nfsmout:
2502 	if (abort) {
2503 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2504 		if (nd.ni_dvp == nd.ni_vp)
2505 			vrele(nd.ni_dvp);
2506 		else
2507 			vput(nd.ni_dvp);
2508 		if (nd.ni_vp)
2509 			vrele(nd.ni_vp);
2510 		if (nd.ni_pathbuf != NULL) {
2511 			pathbuf_destroy(nd.ni_pathbuf);
2512 			nd.ni_pathbuf = NULL;
2513 		}
2514 	}
2515 	if (dirp)
2516 		vrele(dirp);
2517 	return (error);
2518 }
2519 
2520 /*
2521  * nfs rmdir service
2522  */
2523 int
2524 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2525 {
2526 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2527 	struct mbuf *nam = nfsd->nd_nam;
2528 	char *dpos = nfsd->nd_dpos;
2529 	kauth_cred_t cred = nfsd->nd_cr;
2530 	u_int32_t *tl;
2531 	int32_t t1;
2532 	char *bpos;
2533 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2534 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2535 	char *cp2;
2536 	struct mbuf *mb, *mreq;
2537 	struct vnode *vp, *dirp = (struct vnode *)0;
2538 	struct vattr dirfor, diraft;
2539 	nfsrvfh_t nsfh;
2540 	struct nameidata nd;
2541 	u_quad_t frev;
2542 
2543 	nfsm_srvmtofh(&nsfh);
2544 	nfsm_srvnamesiz(len);
2545 	nd.ni_cnd.cn_cred = cred;
2546 	nd.ni_cnd.cn_nameiop = DELETE;
2547 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2548 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2549 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2550 	if (dirp && v3) {
2551 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2552 	}
2553 	if (error) {
2554 		if (nd.ni_pathbuf != NULL) {
2555 			pathbuf_destroy(nd.ni_pathbuf);
2556 			nd.ni_pathbuf = NULL;
2557 		}
2558 		nfsm_reply(NFSX_WCCDATA(v3));
2559 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2560 		if (dirp)
2561 			vrele(dirp);
2562 		return (0);
2563 	}
2564 	vp = nd.ni_vp;
2565 	if (vp->v_type != VDIR) {
2566 		error = ENOTDIR;
2567 		goto out;
2568 	}
2569 	/*
2570 	 * No rmdir "." please.
2571 	 */
2572 	if (nd.ni_dvp == vp) {
2573 		error = EINVAL;
2574 		goto out;
2575 	}
2576 	/*
2577 	 * The root of a mounted filesystem cannot be deleted.
2578 	 */
2579 	if (vp->v_vflag & VV_ROOT)
2580 		error = EBUSY;
2581 out:
2582 	if (!error) {
2583 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
2584 		nqsrv_getl(vp, ND_WRITE);
2585 		error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2586 	} else {
2587 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2588 		if (nd.ni_dvp == nd.ni_vp)
2589 			vrele(nd.ni_dvp);
2590 		else
2591 			vput(nd.ni_dvp);
2592 		vput(vp);
2593 	}
2594 	if (nd.ni_pathbuf != NULL) {
2595 		pathbuf_destroy(nd.ni_pathbuf);
2596 		nd.ni_pathbuf = NULL;
2597 	}
2598 	if (dirp) {
2599 		if (v3) {
2600 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2601 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2602 			VOP_UNLOCK(dirp);
2603 		}
2604 		vrele(dirp);
2605 	}
2606 	nfsm_reply(NFSX_WCCDATA(v3));
2607 	if (v3) {
2608 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2609 		return (0);
2610 	}
2611 	nfsm_srvdone;
2612 }
2613 
2614 /*
2615  * nfs readdir service
2616  * - mallocs what it thinks is enough to read
2617  *	count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR
2618  * - calls VOP_READDIR()
2619  * - loops around building the reply
2620  *	if the output generated exceeds count break out of loop
2621  *	The nfsm_clget macro is used here so that the reply will be packed
2622  *	tightly in mbuf clusters.
2623  * - it only knows that it has encountered eof when the VOP_READDIR()
2624  *	reads nothing
2625  * - as such one readdir rpc will return eof false although you are there
2626  *	and then the next will return eof
2627  * - it trims out records with d_fileno == 0
2628  *	this doesn't matter for Unix clients, but they might confuse clients
2629  *	for other os'.
2630  * - it trims out records with d_type == DT_WHT
2631  *	these cannot be seen through NFS (unless we extend the protocol)
2632  * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2633  *	than requested, but this may not apply to all filesystems. For
2634  *	example, client NFS does not { although it is never remote mounted
2635  *	anyhow }
2636  *     The alternate call nfsrv_readdirplus() does lookups as well.
2637  * PS: The NFS protocol spec. does not clarify what the "count" byte
2638  *	argument is a count of.. just name strings and file id's or the
2639  *	entire reply rpc or ...
2640  *	I tried just file name and id sizes and it confused the Sun client,
2641  *	so I am using the full rpc size now. The "paranoia.." comment refers
2642  *	to including the status longwords that are not a part of the dir.
2643  *	"entry" structures, but are in the rpc.
2644  */
2645 
2646 #define	NFS_SRVDIRBLKSIZ	1024
2647 
2648 struct flrep {
2649 	nfsuint64 fl_off;
2650 	u_int32_t fl_postopok;
2651 	struct nfs_fattr fl_fattr; /* XXX: must be of fattr3 size */
2652 	u_int32_t fl_fhok;
2653 	u_int32_t fl_fhsize;
2654 	/* handle comes here, filled in dynamically */
2655 };
2656 
2657 int
2658 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2659 {
2660 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2661 	struct mbuf *nam = nfsd->nd_nam;
2662 	char *dpos = nfsd->nd_dpos;
2663 	kauth_cred_t cred = nfsd->nd_cr;
2664 	char *bp, *be;
2665 	struct mbuf *mp;
2666 	struct dirent *dp;
2667 	char *cp;
2668 	u_int32_t *tl;
2669 	int32_t t1;
2670 	char *bpos;
2671 	struct mbuf *mb, *mreq, *mp2;
2672 	char *cpos, *cend, *cp2, *rbuf;
2673 	struct vnode *vp;
2674 	struct vattr at;
2675 	nfsrvfh_t nsfh;
2676 	struct uio io;
2677 	struct iovec iv;
2678 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2679 	int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies;
2680 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2681 	u_quad_t frev, off, toff, verf;
2682 	off_t *cookies = NULL, *cookiep;
2683 	nfsuint64 jar;
2684 
2685 	nfsm_srvmtofh(&nsfh);
2686 	if (v3) {
2687 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2688 		toff = fxdr_hyper(tl);
2689 		tl += 2;
2690 		verf = fxdr_hyper(tl);
2691 		tl += 2;
2692 	} else {
2693 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2694 		toff = fxdr_unsigned(u_quad_t, *tl++);
2695 	}
2696 	off = toff;
2697 	cnt = fxdr_unsigned(int, *tl);
2698 	siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1));
2699 	xfer = NFS_SRVMAXDATA(nfsd);
2700 	if (siz > xfer)
2701 		siz = xfer;
2702 	fullsiz = siz;
2703 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
2704 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2705 	if (!error && vp->v_type != VDIR) {
2706 		error = ENOTDIR;
2707 		vput(vp);
2708 	}
2709 	if (error) {
2710 		nfsm_reply(NFSX_UNSIGNED);
2711 		nfsm_srvpostop_attr(getret, &at);
2712 		return (0);
2713 	}
2714 	nqsrv_getl(vp, ND_READ);
2715 	if (v3) {
2716 		error = getret = VOP_GETATTR(vp, &at, cred);
2717 #ifdef NFS3_STRICTVERF
2718 		/*
2719 		 * XXX This check is too strict for Solaris 2.5 clients.
2720 		 */
2721 		if (!error && toff && verf != at.va_filerev)
2722 			error = NFSERR_BAD_COOKIE;
2723 #endif
2724 	}
2725 	if (!error)
2726 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0);
2727 	if (error) {
2728 		vput(vp);
2729 		nfsm_reply(NFSX_POSTOPATTR(v3));
2730 		nfsm_srvpostop_attr(getret, &at);
2731 		return (0);
2732 	}
2733 	VOP_UNLOCK(vp);
2734 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2735 again:
2736 	iv.iov_base = rbuf;
2737 	iv.iov_len = fullsiz;
2738 	io.uio_iov = &iv;
2739 	io.uio_iovcnt = 1;
2740 	io.uio_offset = (off_t)off;
2741 	io.uio_resid = fullsiz;
2742 	io.uio_rw = UIO_READ;
2743 	UIO_SETUP_SYSSPACE(&io);
2744 	eofflag = 0;
2745 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2746 
2747 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2748 
2749 	off = (off_t)io.uio_offset;
2750 	if (!cookies && !error)
2751 		error = NFSERR_PERM;
2752 	if (v3) {
2753 		getret = VOP_GETATTR(vp, &at, cred);
2754 		if (!error)
2755 			error = getret;
2756 	}
2757 
2758 	VOP_UNLOCK(vp);
2759 	if (error) {
2760 		vrele(vp);
2761 		free((void *)rbuf, M_TEMP);
2762 		if (cookies)
2763 			free((void *)cookies, M_TEMP);
2764 		nfsm_reply(NFSX_POSTOPATTR(v3));
2765 		nfsm_srvpostop_attr(getret, &at);
2766 		return (0);
2767 	}
2768 	if (io.uio_resid) {
2769 		siz -= io.uio_resid;
2770 
2771 		/*
2772 		 * If nothing read, return eof
2773 		 * rpc reply
2774 		 */
2775 		if (siz == 0) {
2776 			vrele(vp);
2777 			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2778 				2 * NFSX_UNSIGNED);
2779 			if (v3) {
2780 				nfsm_srvpostop_attr(getret, &at);
2781 				nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2782 				txdr_hyper(at.va_filerev, tl);
2783 				tl += 2;
2784 			} else
2785 				nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2786 			*tl++ = nfs_false;
2787 			*tl = nfs_true;
2788 			free((void *)rbuf, M_TEMP);
2789 			free((void *)cookies, M_TEMP);
2790 			return (0);
2791 		}
2792 	}
2793 
2794 	/*
2795 	 * Check for degenerate cases of nothing useful read.
2796 	 * If so go try again
2797 	 */
2798 	cpos = rbuf;
2799 	cend = rbuf + siz;
2800 	dp = (struct dirent *)cpos;
2801 	cookiep = cookies;
2802 
2803 	while (cpos < cend && ncookies > 0 &&
2804 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2805 		cpos += dp->d_reclen;
2806 		dp = (struct dirent *)cpos;
2807 		cookiep++;
2808 		ncookies--;
2809 	}
2810 	if (cpos >= cend || ncookies == 0) {
2811 		toff = off;
2812 		siz = fullsiz;
2813 		free(cookies, M_TEMP);
2814 		cookies = NULL;
2815 		goto again;
2816 	}
2817 
2818 	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2819 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2820 	if (v3) {
2821 		nfsm_srvpostop_attr(getret, &at);
2822 		nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2823 		txdr_hyper(at.va_filerev, tl);
2824 	}
2825 	mp = mp2 = mb;
2826 	bp = bpos;
2827 	be = bp + M_TRAILINGSPACE(mp);
2828 
2829 	/* Loop through the records and build reply */
2830 	while (cpos < cend && ncookies > 0) {
2831 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2832 			nlen = dp->d_namlen;
2833 			rem = nfsm_rndup(nlen)-nlen;
2834 			len += (4 * NFSX_UNSIGNED + nlen + rem);
2835 			if (v3)
2836 				len += 2 * NFSX_UNSIGNED;
2837 			if (len > cnt) {
2838 				eofflag = 0;
2839 				break;
2840 			}
2841 			/*
2842 			 * Build the directory record xdr from
2843 			 * the dirent entry.
2844 			 */
2845 			nfsm_clget;
2846 			*tl = nfs_true;
2847 			bp += NFSX_UNSIGNED;
2848 			if (v3) {
2849 				nfsm_clget;
2850 				*tl = txdr_unsigned(dp->d_fileno >> 32);
2851 				bp += NFSX_UNSIGNED;
2852 			}
2853 			nfsm_clget;
2854 			*tl = txdr_unsigned(dp->d_fileno);
2855 			bp += NFSX_UNSIGNED;
2856 			nfsm_clget;
2857 			*tl = txdr_unsigned(nlen);
2858 			bp += NFSX_UNSIGNED;
2859 
2860 			/* And loop around copying the name */
2861 			xfer = nlen;
2862 			cp = dp->d_name;
2863 			while (xfer > 0) {
2864 				nfsm_clget;
2865 				if ((bp+xfer) > be)
2866 					tsiz = be-bp;
2867 				else
2868 					tsiz = xfer;
2869 				memcpy(bp, cp, tsiz);
2870 				bp += tsiz;
2871 				xfer -= tsiz;
2872 				if (xfer > 0)
2873 					cp += tsiz;
2874 			}
2875 			/* And null pad to an int32_t boundary */
2876 			for (i = 0; i < rem; i++)
2877 				*bp++ = '\0';
2878 			nfsm_clget;
2879 
2880 			/* Finish off the record */
2881 			txdr_hyper(*cookiep, &jar);
2882 			if (v3) {
2883 				*tl = jar.nfsuquad[0];
2884 				bp += NFSX_UNSIGNED;
2885 				nfsm_clget;
2886 			}
2887 			*tl = jar.nfsuquad[1];
2888 			bp += NFSX_UNSIGNED;
2889 		}
2890 		cpos += dp->d_reclen;
2891 		dp = (struct dirent *)cpos;
2892 		cookiep++;
2893 		ncookies--;
2894 	}
2895 	vrele(vp);
2896 	nfsm_clget;
2897 	*tl = nfs_false;
2898 	bp += NFSX_UNSIGNED;
2899 	nfsm_clget;
2900 	if (eofflag)
2901 		*tl = nfs_true;
2902 	else
2903 		*tl = nfs_false;
2904 	bp += NFSX_UNSIGNED;
2905 	if (mp != mb) {
2906 		if (bp < be)
2907 			mp->m_len = bp - mtod(mp, char *);
2908 	} else
2909 		mp->m_len += bp - bpos;
2910 	free((void *)rbuf, M_TEMP);
2911 	free((void *)cookies, M_TEMP);
2912 	nfsm_srvdone;
2913 }
2914 
2915 int
2916 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2917 {
2918 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2919 	struct mbuf *nam = nfsd->nd_nam;
2920 	char *dpos = nfsd->nd_dpos;
2921 	kauth_cred_t cred = nfsd->nd_cr;
2922 	char *bp, *be;
2923 	struct mbuf *mp;
2924 	struct dirent *dp;
2925 	char *cp;
2926 	u_int32_t *tl;
2927 	int32_t t1;
2928 	char *bpos;
2929 	struct mbuf *mb, *mreq, *mp2;
2930 	char *cpos, *cend, *cp2, *rbuf;
2931 	struct vnode *vp, *nvp;
2932 	struct flrep fl;
2933 	nfsrvfh_t nsfh;
2934 	struct uio io;
2935 	struct iovec iv;
2936 	struct vattr va, at, *vap = &va;
2937 	struct nfs_fattr *fp;
2938 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2939 	int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies;
2940 	u_quad_t frev, off, toff, verf;
2941 	off_t *cookies = NULL, *cookiep;
2942 
2943 	nfsm_srvmtofh(&nsfh);
2944 	nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2945 	toff = fxdr_hyper(tl);
2946 	tl += 2;
2947 	verf = fxdr_hyper(tl);
2948 	tl += 2;
2949 	siz = fxdr_unsigned(int, *tl++);
2950 	cnt = fxdr_unsigned(int, *tl);
2951 	off = toff;
2952 	siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1));
2953 	xfer = NFS_SRVMAXDATA(nfsd);
2954 	if (siz > xfer)
2955 		siz = xfer;
2956 	fullsiz = siz;
2957 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
2958 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2959 	if (!error && vp->v_type != VDIR) {
2960 		error = ENOTDIR;
2961 		vput(vp);
2962 	}
2963 	if (error) {
2964 		nfsm_reply(NFSX_UNSIGNED);
2965 		nfsm_srvpostop_attr(getret, &at);
2966 		return (0);
2967 	}
2968 	error = getret = VOP_GETATTR(vp, &at, cred);
2969 #ifdef NFS3_STRICTVERF
2970 	/*
2971 	 * XXX This check is too strict for Solaris 2.5 clients.
2972 	 */
2973 	if (!error && toff && verf != at.va_filerev)
2974 		error = NFSERR_BAD_COOKIE;
2975 #endif
2976 	if (!error) {
2977 		nqsrv_getl(vp, ND_READ);
2978 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0);
2979 	}
2980 	if (error) {
2981 		vput(vp);
2982 		nfsm_reply(NFSX_V3POSTOPATTR);
2983 		nfsm_srvpostop_attr(getret, &at);
2984 		return (0);
2985 	}
2986 	VOP_UNLOCK(vp);
2987 
2988 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2989 again:
2990 	iv.iov_base = rbuf;
2991 	iv.iov_len = fullsiz;
2992 	io.uio_iov = &iv;
2993 	io.uio_iovcnt = 1;
2994 	io.uio_offset = (off_t)off;
2995 	io.uio_resid = fullsiz;
2996 	io.uio_rw = UIO_READ;
2997 	UIO_SETUP_SYSSPACE(&io);
2998 	eofflag = 0;
2999 
3000 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3001 
3002 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
3003 
3004 	off = (u_quad_t)io.uio_offset;
3005 	getret = VOP_GETATTR(vp, &at, cred);
3006 
3007 	VOP_UNLOCK(vp);
3008 
3009 	/*
3010 	 * If the VGET operation doesn't work for this filesystem,
3011 	 * we can't support readdirplus. Returning NOTSUPP should
3012 	 * make clients fall back to plain readdir.
3013 	 * There's no need to check for VPTOFH as well, we wouldn't
3014 	 * even be here otherwise.
3015 	 */
3016 	if (!getret) {
3017 		if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp)))
3018 			getret = (getret == EOPNOTSUPP) ?
3019 				NFSERR_NOTSUPP : NFSERR_IO;
3020 		else
3021 			vput(nvp);
3022 	}
3023 
3024 	if (!cookies && !error)
3025 		error = NFSERR_PERM;
3026 	if (!error)
3027 		error = getret;
3028 	if (error) {
3029 		vrele(vp);
3030 		if (cookies)
3031 			free((void *)cookies, M_TEMP);
3032 		free((void *)rbuf, M_TEMP);
3033 		nfsm_reply(NFSX_V3POSTOPATTR);
3034 		nfsm_srvpostop_attr(getret, &at);
3035 		return (0);
3036 	}
3037 	if (io.uio_resid) {
3038 		siz -= io.uio_resid;
3039 
3040 		/*
3041 		 * If nothing read, return eof
3042 		 * rpc reply
3043 		 */
3044 		if (siz == 0) {
3045 			vrele(vp);
3046 			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3047 				2 * NFSX_UNSIGNED);
3048 			nfsm_srvpostop_attr(getret, &at);
3049 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3050 			txdr_hyper(at.va_filerev, tl);
3051 			tl += 2;
3052 			*tl++ = nfs_false;
3053 			*tl = nfs_true;
3054 			free((void *)cookies, M_TEMP);
3055 			free((void *)rbuf, M_TEMP);
3056 			return (0);
3057 		}
3058 	}
3059 
3060 	/*
3061 	 * Check for degenerate cases of nothing useful read.
3062 	 * If so go try again
3063 	 */
3064 	cpos = rbuf;
3065 	cend = rbuf + siz;
3066 	dp = (struct dirent *)cpos;
3067 	cookiep = cookies;
3068 
3069 	while (cpos < cend && ncookies > 0 &&
3070 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
3071 		cpos += dp->d_reclen;
3072 		dp = (struct dirent *)cpos;
3073 		cookiep++;
3074 		ncookies--;
3075 	}
3076 	if (cpos >= cend || ncookies == 0) {
3077 		toff = off;
3078 		siz = fullsiz;
3079 		free(cookies, M_TEMP);
3080 		cookies = NULL;
3081 		goto again;
3082 	}
3083 
3084 	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
3085 	nfsm_reply(cnt);
3086 	nfsm_srvpostop_attr(getret, &at);
3087 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3088 	txdr_hyper(at.va_filerev, tl);
3089 	mp = mp2 = mb;
3090 	bp = bpos;
3091 	be = bp + M_TRAILINGSPACE(mp);
3092 
3093 	/* Loop through the records and build reply */
3094 	while (cpos < cend && ncookies > 0) {
3095 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3096 			nfsrvfh_t nnsfh;
3097 
3098 			nlen = dp->d_namlen;
3099 			rem = nfsm_rndup(nlen)-nlen;
3100 
3101 			/*
3102 			 * For readdir_and_lookup get the vnode using
3103 			 * the file number.
3104 			 */
3105 			if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
3106 				goto invalid;
3107 			if (nfsrv_composefh(nvp, &nnsfh, true)) {
3108 				vput(nvp);
3109 				goto invalid;
3110 			}
3111 			if (VOP_GETATTR(nvp, vap, cred)) {
3112 				vput(nvp);
3113 				goto invalid;
3114 			}
3115 			vput(nvp);
3116 
3117 			/*
3118 			 * If either the dircount or maxcount will be
3119 			 * exceeded, get out now. Both of these lengths
3120 			 * are calculated conservatively, including all
3121 			 * XDR overheads.
3122 			 */
3123 			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3124 				NFSX_V3POSTOPATTR);
3125 			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3126 			if (len > cnt || dirlen > fullsiz) {
3127 				eofflag = 0;
3128 				break;
3129 			}
3130 
3131 			/*
3132 			 * Build the directory record xdr from
3133 			 * the dirent entry.
3134 			 */
3135 			fp = (struct nfs_fattr *)&fl.fl_fattr;
3136 			nfsm_srvfillattr(vap, fp);
3137 			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3138 			fl.fl_fhok = nfs_true;
3139 			fl.fl_postopok = nfs_true;
3140 			txdr_hyper(*cookiep, fl.fl_off.nfsuquad);
3141 
3142 			nfsm_clget;
3143 			*tl = nfs_true;
3144 			bp += NFSX_UNSIGNED;
3145 			nfsm_clget;
3146 			*tl = txdr_unsigned(dp->d_fileno >> 32);
3147 			bp += NFSX_UNSIGNED;
3148 			nfsm_clget;
3149 			*tl = txdr_unsigned(dp->d_fileno);
3150 			bp += NFSX_UNSIGNED;
3151 			nfsm_clget;
3152 			*tl = txdr_unsigned(nlen);
3153 			bp += NFSX_UNSIGNED;
3154 
3155 			/* And loop around copying the name */
3156 			xfer = nlen;
3157 			cp = dp->d_name;
3158 			while (xfer > 0) {
3159 				nfsm_clget;
3160 				if ((bp + xfer) > be)
3161 					tsiz = be - bp;
3162 				else
3163 					tsiz = xfer;
3164 				memcpy(bp, cp, tsiz);
3165 				bp += tsiz;
3166 				xfer -= tsiz;
3167 				if (xfer > 0)
3168 					cp += tsiz;
3169 			}
3170 			/* And null pad to an int32_t boundary */
3171 			for (i = 0; i < rem; i++)
3172 				*bp++ = '\0';
3173 
3174 			/*
3175 			 * Now copy the flrep structure out.
3176 			 */
3177 			xfer = sizeof(struct flrep);
3178 			cp = (void *)&fl;
3179 			while (xfer > 0) {
3180 				nfsm_clget;
3181 				if ((bp + xfer) > be)
3182 					tsiz = be - bp;
3183 				else
3184 					tsiz = xfer;
3185 				memcpy(bp, cp, tsiz);
3186 				bp += tsiz;
3187 				xfer -= tsiz;
3188 				if (xfer > 0)
3189 					cp += tsiz;
3190 			}
3191 
3192 			/*
3193 			 * ... and filehandle.
3194 			 */
3195 			xfer = NFSRVFH_SIZE(&nnsfh);
3196 			cp = NFSRVFH_DATA(&nnsfh);
3197 			while (xfer > 0) {
3198 				nfsm_clget;
3199 				if ((bp + xfer) > be)
3200 					tsiz = be - bp;
3201 				else
3202 					tsiz = xfer;
3203 				memcpy(bp, cp, tsiz);
3204 				bp += tsiz;
3205 				xfer -= tsiz;
3206 				if (xfer > 0)
3207 					cp += tsiz;
3208 			}
3209 		}
3210 invalid:
3211 		cpos += dp->d_reclen;
3212 		dp = (struct dirent *)cpos;
3213 		cookiep++;
3214 		ncookies--;
3215 	}
3216 	vrele(vp);
3217 	nfsm_clget;
3218 	*tl = nfs_false;
3219 	bp += NFSX_UNSIGNED;
3220 	nfsm_clget;
3221 	if (eofflag)
3222 		*tl = nfs_true;
3223 	else
3224 		*tl = nfs_false;
3225 	bp += NFSX_UNSIGNED;
3226 	if (mp != mb) {
3227 		if (bp < be)
3228 			mp->m_len = bp - mtod(mp, char *);
3229 	} else
3230 		mp->m_len += bp - bpos;
3231 	free((void *)cookies, M_TEMP);
3232 	free((void *)rbuf, M_TEMP);
3233 	nfsm_srvdone;
3234 }
3235 
3236 /*
3237  * nfs commit service
3238  */
3239 int
3240 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3241 {
3242 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3243 	struct mbuf *nam = nfsd->nd_nam;
3244 	char *dpos = nfsd->nd_dpos;
3245 	kauth_cred_t cred = nfsd->nd_cr;
3246 	struct vattr bfor, aft;
3247 	struct vnode *vp;
3248 	nfsrvfh_t nsfh;
3249 	u_int32_t *tl;
3250 	int32_t t1;
3251 	char *bpos;
3252 	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0;
3253 	uint32_t cnt;
3254 	char *cp2;
3255 	struct mbuf *mb, *mreq;
3256 	u_quad_t frev, off, end;
3257 
3258 	nfsm_srvmtofh(&nsfh);
3259 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3260 
3261 	off = fxdr_hyper(tl);
3262 	tl += 2;
3263 	cnt = fxdr_unsigned(uint32_t, *tl);
3264 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3265 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3266 	if (error) {
3267 		nfsm_reply(2 * NFSX_UNSIGNED);
3268 		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3269 		return (0);
3270 	}
3271 	for_ret = VOP_GETATTR(vp, &bfor, cred);
3272 	end = (cnt > 0) ? off + cnt : vp->v_size;
3273 	if (end < off || end > vp->v_size)
3274 		end = vp->v_size;
3275 	if (off < vp->v_size)
3276 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end);
3277 	/* else error == 0, from nfsrv_fhtovp() */
3278 	aft_ret = VOP_GETATTR(vp, &aft, cred);
3279 	vput(vp);
3280 	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3281 	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3282 	if (!error) {
3283 		nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3284 		*tl++ = txdr_unsigned(boottime.tv_sec);
3285 		*tl = txdr_unsigned(boottime.tv_nsec / 1000);
3286 	} else {
3287 		return (0);
3288 	}
3289 	nfsm_srvdone;
3290 }
3291 
3292 /*
3293  * nfs statfs service
3294  */
3295 int
3296 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3297 {
3298 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3299 	struct mbuf *nam = nfsd->nd_nam;
3300 	char *dpos = nfsd->nd_dpos;
3301 	kauth_cred_t cred = nfsd->nd_cr;
3302 	struct statvfs *sf = NULL;
3303 	struct nfs_statfs *sfp;
3304 	u_int32_t *tl;
3305 	int32_t t1;
3306 	char *bpos;
3307 	int error = 0, rdonly, cache = 0, getret = 1;
3308 	int v3 = (nfsd->nd_flag & ND_NFSV3);
3309 	char *cp2;
3310 	struct mbuf *mb, *mreq;
3311 	struct vnode *vp;
3312 	struct vattr at;
3313 	nfsrvfh_t nsfh;
3314 	u_quad_t frev, tval;
3315 
3316 	nfsm_srvmtofh(&nsfh);
3317 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3318 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3319 	if (error) {
3320 		nfsm_reply(NFSX_UNSIGNED);
3321 		nfsm_srvpostop_attr(getret, &at);
3322 		return (0);
3323 	}
3324 	sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK);
3325 	error = VFS_STATVFS(vp->v_mount, sf);
3326 	getret = VOP_GETATTR(vp, &at, cred);
3327 	vput(vp);
3328 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3329 	if (v3)
3330 		nfsm_srvpostop_attr(getret, &at);
3331 	if (error) {
3332 		free(sf, M_TEMP);
3333 		return (0);
3334 	}
3335 	nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3336 	if (v3) {
3337 		tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize);
3338 		txdr_hyper(tval, &sfp->sf_tbytes);
3339 		tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize);
3340 		txdr_hyper(tval, &sfp->sf_fbytes);
3341 		tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize);
3342 		txdr_hyper(tval, &sfp->sf_abytes);
3343 		tval = (u_quad_t)sf->f_files;
3344 		txdr_hyper(tval, &sfp->sf_tfiles);
3345 		tval = (u_quad_t)sf->f_ffree;
3346 		txdr_hyper(tval, &sfp->sf_ffiles);
3347 		txdr_hyper(tval, &sfp->sf_afiles);
3348 		sfp->sf_invarsec = 0;
3349 	} else {
3350 		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3351 		sfp->sf_bsize = txdr_unsigned(sf->f_frsize);
3352 		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3353 		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3354 		sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3355 	}
3356 nfsmout:
3357 	if (sf)
3358 	    free(sf, M_TEMP);
3359 	return error;
3360 }
3361 
3362 /*
3363  * nfs fsinfo service
3364  */
3365 int
3366 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3367 {
3368 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3369 	struct mbuf *nam = nfsd->nd_nam;
3370 	char *dpos = nfsd->nd_dpos;
3371 	kauth_cred_t cred = nfsd->nd_cr;
3372 	u_int32_t *tl;
3373 	struct nfsv3_fsinfo *sip;
3374 	int32_t t1;
3375 	char *bpos;
3376 	int error = 0, rdonly, cache = 0, getret = 1;
3377 	uint32_t maxdata;
3378 	char *cp2;
3379 	struct mbuf *mb, *mreq;
3380 	struct vnode *vp;
3381 	struct vattr at;
3382 	nfsrvfh_t nsfh;
3383 	u_quad_t frev, maxfsize;
3384 	struct statvfs *sb;
3385 
3386 	nfsm_srvmtofh(&nsfh);
3387 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3388 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3389 	if (error) {
3390 		nfsm_reply(NFSX_UNSIGNED);
3391 		nfsm_srvpostop_attr(getret, &at);
3392 		return (0);
3393 	}
3394 
3395 	/* XXX Try to make a guess on the max file size. */
3396 	sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK);
3397 	VFS_STATVFS(vp->v_mount, sb);
3398 	maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1;
3399 	free(sb, M_TEMP);
3400 
3401 	getret = VOP_GETATTR(vp, &at, cred);
3402 	vput(vp);
3403 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3404 	nfsm_srvpostop_attr(getret, &at);
3405 	nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3406 
3407 	/*
3408 	 * XXX
3409 	 * There should be file system VFS OP(s) to get this information.
3410 	 * For now, assume ufs.
3411 	 */
3412 	if (slp->ns_so->so_type == SOCK_DGRAM)
3413 		maxdata = NFS_MAXDGRAMDATA;
3414 	else
3415 		maxdata = NFS_MAXDATA;
3416 	sip->fs_rtmax = txdr_unsigned(maxdata);
3417 	sip->fs_rtpref = txdr_unsigned(maxdata);
3418 	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3419 	sip->fs_wtmax = txdr_unsigned(maxdata);
3420 	sip->fs_wtpref = txdr_unsigned(maxdata);
3421 	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3422 	sip->fs_dtpref = txdr_unsigned(maxdata);
3423 	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3424 	sip->fs_timedelta.nfsv3_sec = 0;
3425 	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3426 	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3427 		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3428 		NFSV3FSINFO_CANSETTIME);
3429 	nfsm_srvdone;
3430 }
3431 
3432 /*
3433  * nfs pathconf service
3434  */
3435 int
3436 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3437 {
3438 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3439 	struct mbuf *nam = nfsd->nd_nam;
3440 	char *dpos = nfsd->nd_dpos;
3441 	kauth_cred_t cred = nfsd->nd_cr;
3442 	u_int32_t *tl;
3443 	struct nfsv3_pathconf *pc;
3444 	int32_t t1;
3445 	char *bpos;
3446 	int error = 0, rdonly, cache = 0, getret = 1;
3447 	register_t linkmax, namemax, chownres, notrunc;
3448 	char *cp2;
3449 	struct mbuf *mb, *mreq;
3450 	struct vnode *vp;
3451 	struct vattr at;
3452 	nfsrvfh_t nsfh;
3453 	u_quad_t frev;
3454 
3455 	nfsm_srvmtofh(&nsfh);
3456 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3457 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3458 	if (error) {
3459 		nfsm_reply(NFSX_UNSIGNED);
3460 		nfsm_srvpostop_attr(getret, &at);
3461 		return (0);
3462 	}
3463 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3464 	if (!error)
3465 		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3466 	if (!error)
3467 		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3468 	if (!error)
3469 		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3470 	getret = VOP_GETATTR(vp, &at, cred);
3471 	vput(vp);
3472 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3473 	nfsm_srvpostop_attr(getret, &at);
3474 	if (error)
3475 		return (0);
3476 	nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3477 
3478 	pc->pc_linkmax = txdr_unsigned(linkmax);
3479 	pc->pc_namemax = txdr_unsigned(namemax);
3480 	pc->pc_notrunc = txdr_unsigned(notrunc);
3481 	pc->pc_chownrestricted = txdr_unsigned(chownres);
3482 
3483 	/*
3484 	 * These should probably be supported by VOP_PATHCONF(), but
3485 	 * until msdosfs is exportable (why would you want to?), the
3486 	 * Unix defaults should be ok.
3487 	 */
3488 	pc->pc_caseinsensitive = nfs_false;
3489 	pc->pc_casepreserving = nfs_true;
3490 	nfsm_srvdone;
3491 }
3492 
3493 /*
3494  * Null operation, used by clients to ping server
3495  */
3496 /* ARGSUSED */
3497 int
3498 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3499     struct lwp *lwp, struct mbuf **mrq)
3500 {
3501 	struct mbuf *mrep = nfsd->nd_mrep;
3502 	char *bpos;
3503 	int error = NFSERR_RETVOID, cache = 0;
3504 	struct mbuf *mb, *mreq;
3505 	u_quad_t frev;
3506 
3507 	nfsm_reply(0);
3508 nfsmout:
3509 	return (0);
3510 }
3511 
3512 /*
3513  * No operation, used for obsolete procedures
3514  */
3515 /* ARGSUSED */
3516 int
3517 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3518     struct lwp *lwp, struct mbuf **mrq)
3519 {
3520 	struct mbuf *mrep = nfsd->nd_mrep;
3521 	char *bpos;
3522 	int error, cache = 0;
3523 	struct mbuf *mb, *mreq;
3524 	u_quad_t frev;
3525 
3526 	if (nfsd->nd_repstat)
3527 		error = nfsd->nd_repstat;
3528 	else
3529 		error = EPROCUNAVAIL;
3530 	nfsm_reply(0);
3531 nfsmout:
3532 	return (0);
3533 }
3534 
3535 /*
3536  * Perform access checking for vnodes obtained from file handles that would
3537  * refer to files already opened by a Unix client. You cannot just use
3538  * vn_writechk() and VOP_ACCESS() for two reasons.
3539  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3540  * 2 - The owner is to be given access irrespective of mode bits for some
3541  *     operations, so that processes that chmod after opening a file don't
3542  *     break. I don't like this because it opens a security hole, but since
3543  *     the nfs server opens a security hole the size of a barn door anyhow,
3544  *     what the heck.
3545  *
3546  * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3547  * will return EPERM instead of EACCES. EPERM is always an error.
3548  */
3549 int
3550 nfsrv_access(struct vnode *vp, int flags, kauth_cred_t cred, int rdonly, struct lwp *lwp, int override)
3551 {
3552 	struct vattr vattr;
3553 	int error;
3554 	if (flags & VWRITE) {
3555 		/* Just vn_writechk() changed to check rdonly */
3556 		/*
3557 		 * Disallow write attempts on read-only file systems;
3558 		 * unless the file is a socket or a block or character
3559 		 * device resident on the file system.
3560 		 */
3561 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3562 			switch (vp->v_type) {
3563 			case VREG:
3564 			case VDIR:
3565 			case VLNK:
3566 				return (EROFS);
3567 			default:
3568 				break;
3569 			}
3570 		}
3571 
3572 		/*
3573 		 * If the vnode is in use as a process's text,
3574 		 * we can't allow writing.
3575 		 */
3576 		if (vp->v_iflag & VI_TEXT)
3577 			return (ETXTBSY);
3578 	}
3579 	error = VOP_GETATTR(vp, &vattr, cred);
3580 	if (error)
3581 		return (error);
3582 	error = VOP_ACCESS(vp, flags, cred);
3583 	/*
3584 	 * Allow certain operations for the owner (reads and writes
3585 	 * on files that are already open).
3586 	 */
3587 	if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid)
3588 		error = 0;
3589 	return error;
3590 }
3591