xref: /netbsd-src/sys/nfs/nfs_serv.c (revision e6c7e151de239c49d2e38720a061ed9d1fa99309)
1 /*	$NetBSD: nfs_serv.c,v 1.179 2020/01/17 20:08:09 ad Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_serv.c	8.8 (Berkeley) 7/31/95
35  */
36 
37 /*
38  * nfs version 2 and 3 server calls to vnode ops
39  * - these routines generally have 3 phases
40  *   1 - break down and validate rpc request in mbuf list
41  *   2 - do the vnode ops for the request
42  *       (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
43  *   3 - build the rpc reply in an mbuf list
44  *   nb:
45  *	- do not mix the phases, since the nfsm_?? macros can return failures
46  *	  on a bad rpc or similar and do not do any vrele() or vput()'s
47  *
48  *      - the nfsm_reply() macro generates an nfs rpc reply with the nfs
49  *	error number iff error != 0 whereas
50  *	returning an error from the server function implies a fatal error
51  *	such as a badly constructed rpc request that should be dropped without
52  *	a reply.
53  *	For Version 3, nfsm_reply() does not return for the error case, since
54  *	most version 3 rpcs return more than the status for error cases.
55  */
56 
57 #include <sys/cdefs.h>
58 __KERNEL_RCSID(0, "$NetBSD: nfs_serv.c,v 1.179 2020/01/17 20:08:09 ad Exp $");
59 
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/proc.h>
63 #include <sys/file.h>
64 #include <sys/namei.h>
65 #include <sys/vnode.h>
66 #include <sys/mount.h>
67 #include <sys/fstrans.h>
68 #include <sys/socket.h>
69 #include <sys/socketvar.h>
70 #include <sys/mbuf.h>
71 #include <sys/dirent.h>
72 #include <sys/stat.h>
73 #include <sys/kernel.h>
74 #include <sys/hash.h>
75 #include <sys/kauth.h>
76 #include <sys/module.h>
77 #include <sys/syscall.h>
78 #include <sys/syscallargs.h>
79 #include <sys/syscallvar.h>
80 
81 #include <uvm/uvm.h>
82 
83 #include <nfs/nfsproto.h>
84 #include <nfs/rpcv2.h>
85 #include <nfs/nfs.h>
86 #include <nfs/xdr_subs.h>
87 #include <nfs/nfsm_subs.h>
88 #include <nfs/nfs_var.h>
89 
90 MODULE(MODULE_CLASS_MISC, nfsserver, "nfs");
91 
92 /* Global vars */
93 extern u_int32_t nfs_xdrneg1;
94 extern u_int32_t nfs_false, nfs_true;
95 extern const enum vtype nv3tov_type[8];
96 extern struct nfsstats nfsstats;
97 extern const nfstype nfsv2_type[9];
98 extern const nfstype nfsv3_type[9];
99 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
100 bool nfsd_use_loan = true;	/* use page-loan for READ OP */
101 
102 #define	nqsrv_getl(vp, rw)	/* nothing */
103 
104 static const struct syscall_package nfsserver_syscalls[] = {
105 	{ SYS_nfssvc, 0, (sy_call_t *)sys_nfssvc },
106 	{ 0, 0, NULL },
107 };
108 
109 static int
110 nfsserver_modcmd(modcmd_t cmd, void *arg)
111 {
112 	extern struct vfs_hooks nfs_export_hooks;	/* XXX */
113 	int error;
114 
115 	switch (cmd) {
116 	case MODULE_CMD_INIT:
117 		error = syscall_establish(NULL, nfsserver_syscalls);
118 		if (error != 0) {
119 			return error;
120 		}
121 		nfs_init();	/* XXX for monolithic kernel */
122 		netexport_init();
123 		nfsrv_initcache();	/* Init the server request cache */
124 		nfsrv_init(0);		/* Init server data structures */
125 		vfs_hooks_attach(&nfs_export_hooks);
126 		nfs_timer_srvinit(nfsrv_timer);
127 		return 0;
128 	case MODULE_CMD_FINI:
129 		error = syscall_disestablish(NULL, nfsserver_syscalls);
130 		if (error != 0) {
131 			return error;
132 		}
133 		/*
134 		 * Kill export list before detaching VFS hooks, so we
135 		 * we don't leak state due to a concurrent umount().
136 		 */
137 		netexport_fini();
138 		vfs_hooks_detach(&nfs_export_hooks);
139 
140 		/* Kill timer before server goes away. */
141 		nfs_timer_srvfini();
142 		nfsrv_fini();
143 
144 		/* Server uses server cache, so kill cache last. */
145 		nfsrv_finicache();
146 		nfs_fini();
147 		return 0;
148 	case MODULE_CMD_AUTOUNLOAD:
149 		if (netexport_hasexports())
150 			return EBUSY;
151 		/*FALLTHROUGH*/
152 	default:
153 		return ENOTTY;
154 	}
155 }
156 
157 /*
158  * nfs v3 access service
159  */
160 int
161 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
162 {
163 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
164 	struct mbuf *nam = nfsd->nd_nam;
165 	char *dpos = nfsd->nd_dpos;
166 	kauth_cred_t cred = nfsd->nd_cr;
167 	struct vnode *vp;
168 	nfsrvfh_t nsfh;
169 	u_int32_t *tl;
170 	int32_t t1;
171 	char *bpos;
172 	int error = 0, rdonly, cache = 0, getret;
173 	char *cp2;
174 	struct mbuf *mb, *mreq __unused;
175 	struct vattr va;
176 	u_long inmode, testmode, outmode;
177 	u_quad_t frev;
178 
179 	nfsm_srvmtofh(&nsfh);
180 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
181 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly,
182 	    (nfsd->nd_flag & ND_KERBAUTH), false);
183 	if (error) {
184 		nfsm_reply(NFSX_UNSIGNED);
185 		nfsm_srvpostop_attr(1, (struct vattr *)0);
186 		return (0);
187 	}
188 	inmode = fxdr_unsigned(u_int32_t, *tl);
189 	outmode = 0;
190 	if ((inmode & NFSV3ACCESS_READ) &&
191 	    nfsrv_access(vp, VREAD, cred, rdonly, lwp, 0) == 0)
192 		outmode |= NFSV3ACCESS_READ;
193 	if (vp->v_type != VDIR) {
194 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
195 		if (testmode &&
196 		    nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0)
197 			outmode |= testmode;
198 		if ((inmode & NFSV3ACCESS_EXECUTE) &&
199 		    nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0)
200 			outmode |= NFSV3ACCESS_EXECUTE;
201 	} else {
202 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
203 		    NFSV3ACCESS_DELETE);
204 		if (testmode &&
205 		    nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0)
206 			outmode |= testmode;
207 		if ((inmode & NFSV3ACCESS_LOOKUP) &&
208 		    nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0)
209 			outmode |= NFSV3ACCESS_LOOKUP;
210 	}
211 	getret = VOP_GETATTR(vp, &va, cred);
212 	vput(vp);
213 	nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
214 	nfsm_srvpostop_attr(getret, &va);
215 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
216 	*tl = txdr_unsigned(outmode);
217 	nfsm_srvdone;
218 }
219 
220 /*
221  * nfs getattr service
222  */
223 int
224 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
225 {
226 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
227 	struct mbuf *nam = nfsd->nd_nam;
228 	char *dpos = nfsd->nd_dpos;
229 	kauth_cred_t cred = nfsd->nd_cr;
230 	struct nfs_fattr *fp;
231 	struct vattr va;
232 	struct vnode *vp;
233 	nfsrvfh_t nsfh;
234 	u_int32_t *tl;
235 	int32_t t1;
236 	char *bpos;
237 	int error = 0, rdonly, cache = 0;
238 	char *cp2;
239 	struct mbuf *mb, *mreq __unused;
240 	u_quad_t frev;
241 
242 	nfsm_srvmtofh(&nsfh);
243 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly,
244 	    (nfsd->nd_flag & ND_KERBAUTH), false);
245 	if (error) {
246 		nfsm_reply(0);
247 		return (0);
248 	}
249 	nqsrv_getl(vp, ND_READ);
250 	error = VOP_GETATTR(vp, &va, cred);
251 	vput(vp);
252 	nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
253 	if (error)
254 		return (0);
255 	nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
256 	nfsm_srvfillattr(&va, fp);
257 	nfsm_srvdone;
258 }
259 
260 /*
261  * nfs setattr service
262  */
263 int
264 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
265 {
266 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
267 	struct mbuf *nam = nfsd->nd_nam;
268 	char *dpos = nfsd->nd_dpos;
269 	kauth_cred_t cred = nfsd->nd_cr;
270 	struct vattr va, preat;
271 	struct nfsv2_sattr *sp;
272 	struct nfs_fattr *fp;
273 	struct vnode *vp;
274 	nfsrvfh_t nsfh;
275 	u_int32_t *tl;
276 	int32_t t1;
277 	char *bpos;
278 	int error = 0, rdonly, cache = 0, preat_ret = 1, postat_ret = 1;
279 	int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
280 	char *cp2;
281 	struct mbuf *mb, *mreq __unused;
282 	u_quad_t frev;
283 	struct timespec guard;
284 
285 	memset(&guard, 0, sizeof guard);	/* XXX gcc */
286 
287 	nfsm_srvmtofh(&nsfh);
288 	vattr_null(&va);
289 	if (v3) {
290 		nfsm_srvsattr(&va);
291 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
292 		gcheck = fxdr_unsigned(int, *tl);
293 		if (gcheck) {
294 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
295 			fxdr_nfsv3time(tl, &guard);
296 		}
297 	} else {
298 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
299 		/*
300 		 * Nah nah nah nah na nah
301 		 * There is a bug in the Sun client that puts 0xffff in the mode
302 		 * field of sattr when it should put in 0xffffffff. The u_short
303 		 * doesn't sign extend.
304 		 * --> check the low order 2 bytes for 0xffff
305 		 */
306 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
307 			va.va_mode = nfstov_mode(sp->sa_mode);
308 		if (sp->sa_uid != nfs_xdrneg1)
309 			va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
310 		if (sp->sa_gid != nfs_xdrneg1)
311 			va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
312 		if (sp->sa_size != nfs_xdrneg1)
313 			va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
314 		if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
315 #ifdef notyet
316 			fxdr_nfsv2time(&sp->sa_atime, &va.va_atime);
317 #else
318 			va.va_atime.tv_sec =
319 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
320 			va.va_atime.tv_nsec = 0;
321 #endif
322 		}
323 		if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
324 			fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime);
325 
326 	}
327 
328 	/*
329 	 * Now that we have all the fields, lets do it.
330 	 */
331 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly,
332 	    (nfsd->nd_flag & ND_KERBAUTH), false);
333 	if (error) {
334 		nfsm_reply(2 * NFSX_UNSIGNED);
335 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
336 		return (0);
337 	}
338 	nqsrv_getl(vp, ND_WRITE);
339 	if (v3) {
340 		error = preat_ret = VOP_GETATTR(vp, &preat, cred);
341 		if (!error && gcheck &&
342 			(preat.va_ctime.tv_sec != guard.tv_sec ||
343 			 preat.va_ctime.tv_nsec != guard.tv_nsec))
344 			error = NFSERR_NOT_SYNC;
345 		if (error) {
346 			vput(vp);
347 			nfsm_reply(NFSX_WCCDATA(v3));
348 			nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
349 			return (0);
350 		}
351 	}
352 
353 	/*
354 	 * If the size is being changed write acces is required, otherwise
355 	 * just check for a read only file system.
356 	 */
357 	if (va.va_size == ((u_quad_t)((quad_t) -1))) {
358 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
359 			error = EROFS;
360 			goto out;
361 		}
362 	} else {
363 		if (vp->v_type == VDIR) {
364 			error = EISDIR;
365 			goto out;
366 		} else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
367 			lwp, 0)) != 0)
368 			goto out;
369 	}
370 	error = VOP_SETATTR(vp, &va, cred);
371 	postat_ret = VOP_GETATTR(vp, &va, cred);
372 	if (!error)
373 		error = postat_ret;
374 out:
375 	vput(vp);
376 	nfsm_reply(NFSX_WCCORFATTR(v3));
377 	if (v3) {
378 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
379 		return (0);
380 	} else {
381 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
382 		nfsm_srvfillattr(&va, fp);
383 	}
384 	nfsm_srvdone;
385 }
386 
387 /*
388  * nfs lookup rpc
389  */
390 int
391 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
392 {
393 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
394 	struct mbuf *nam = nfsd->nd_nam;
395 	char *dpos = nfsd->nd_dpos;
396 	kauth_cred_t cred = nfsd->nd_cr;
397 	struct nfs_fattr *fp;
398 	struct nameidata nd, ind, *ndp = &nd;
399 	struct pathbuf *ipb = NULL;
400 	struct vnode *vp, *dirp;
401 	nfsrvfh_t nsfh;
402 	char *cp;
403 	u_int32_t *tl;
404 	int32_t t1;
405 	char *bpos;
406 	int error = 0, cache = 0, dirattr_ret = 1;
407 	uint32_t len;
408 	int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
409 	char *cp2;
410 	struct mbuf *mb, *mreq __unused;
411 	struct vattr va, dirattr;
412 	u_quad_t frev;
413 
414 	nfsm_srvmtofh(&nsfh);
415 	nfsm_srvnamesiz(len);
416 
417 	pubflag = nfs_ispublicfh(&nsfh);
418 
419 	nd.ni_cnd.cn_cred = cred;
420 	nd.ni_cnd.cn_nameiop = LOOKUP;
421 	nd.ni_cnd.cn_flags = LOCKLEAF;
422 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
423 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
424 
425 	if (!error && pubflag) {
426 		if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL &&
427 		    (ipb = pathbuf_create(nfs_pub.np_index)) != NULL) {
428 			/*
429 			 * Setup call to lookup() to see if we can find
430 			 * the index file. Arguably, this doesn't belong
431 			 * in a kernel.. Ugh.
432 			 */
433 			ind = nd;
434 			VOP_UNLOCK(nd.ni_vp);
435 			ind.ni_pathbuf = ipb;
436 
437 			error = lookup_for_nfsd_index(&ind, nd.ni_vp);
438 			if (!error) {
439 				/*
440 				 * Found an index file. Get rid of
441 				 * the old references.
442 				 */
443 				if (dirp)
444 					vrele(dirp);
445 				dirp = nd.ni_vp;
446 				ndp = &ind;
447 			} else
448 				error = 0;
449 		}
450 		/*
451 		 * If the public filehandle was used, check that this lookup
452 		 * didn't result in a filehandle outside the publicly exported
453 		 * filesystem.
454 		 */
455 
456 		if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) {
457 			vput(nd.ni_vp);
458 			error = EPERM;
459 		}
460 	}
461 
462 	if (error) {
463 		if (nd.ni_pathbuf != NULL) {
464 			pathbuf_destroy(nd.ni_pathbuf);
465 		}
466 		if (ipb != NULL) {
467 			pathbuf_destroy(ipb);
468 		}
469 		if (dirp) {
470 			if (v3) {
471 				vn_lock(dirp, LK_SHARED | LK_RETRY);
472 				dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred);
473 				vput(dirp);
474 			} else
475 				vrele(dirp);
476 		}
477 		nfsm_reply(NFSX_POSTOPATTR(v3));
478 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
479 		return (0);
480 	}
481 
482 	nqsrv_getl(ndp->ni_startdir, ND_READ);
483 	pathbuf_destroy(nd.ni_pathbuf);
484 	if (ipb != NULL) {
485 		pathbuf_destroy(ipb);
486 	}
487 	vp = ndp->ni_vp;
488 	error = nfsrv_composefh(vp, &nsfh, v3);
489 	if (!error)
490 		error = VOP_GETATTR(vp, &va, cred);
491 	vput(vp);
492 	if (dirp) {
493 		if (v3) {
494 			vn_lock(dirp, LK_SHARED | LK_RETRY);
495 			dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred);
496 			vput(dirp);
497 		} else
498 			vrele(dirp);
499 	}
500 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPORFATTR(v3) +
501 	    NFSX_POSTOPATTR(v3));
502 	if (error) {
503 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
504 		return (0);
505 	}
506 	nfsm_srvfhtom(&nsfh, v3);
507 	if (v3) {
508 		nfsm_srvpostop_attr(0, &va);
509 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
510 	} else {
511 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
512 		nfsm_srvfillattr(&va, fp);
513 	}
514 	nfsm_srvdone;
515 }
516 
517 /*
518  * nfs readlink service
519  */
520 int
521 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
522 {
523 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
524 	struct mbuf *nam = nfsd->nd_nam;
525 	char *dpos = nfsd->nd_dpos;
526 	kauth_cred_t cred = nfsd->nd_cr;
527 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
528 	struct iovec *ivp = iv;
529 	struct mbuf *mp;
530 	u_int32_t *tl;
531 	int32_t t1;
532 	char *bpos;
533 	int error = 0, rdonly, cache = 0, i, padlen, getret;
534 	uint32_t len;
535 	int v3 = (nfsd->nd_flag & ND_NFSV3);
536 	char *cp2;
537 	struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq __unused;
538 	struct vnode *vp;
539 	struct vattr attr;
540 	nfsrvfh_t nsfh;
541 	struct uio io, *uiop = &io;
542 	u_quad_t frev;
543 
544 	nfsm_srvmtofh(&nsfh);
545 	len = 0;
546 	i = 0;
547 	while (len < NFS_MAXPATHLEN) {
548 		mp = m_get(M_WAIT, MT_DATA);
549 		MCLAIM(mp, &nfs_mowner);
550 		m_clget(mp, M_WAIT);
551 		mp->m_len = NFSMSIZ(mp);
552 		if (len == 0)
553 			mp3 = mp2 = mp;
554 		else {
555 			mp2->m_next = mp;
556 			mp2 = mp;
557 		}
558 		if ((len+mp->m_len) > NFS_MAXPATHLEN) {
559 			mp->m_len = NFS_MAXPATHLEN-len;
560 			len = NFS_MAXPATHLEN;
561 		} else
562 			len += mp->m_len;
563 		ivp->iov_base = mtod(mp, void *);
564 		ivp->iov_len = mp->m_len;
565 		i++;
566 		ivp++;
567 	}
568 	uiop->uio_iov = iv;
569 	uiop->uio_iovcnt = i;
570 	uiop->uio_offset = 0;
571 	uiop->uio_resid = len;
572 	uiop->uio_rw = UIO_READ;
573 	UIO_SETUP_SYSSPACE(uiop);
574 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
575 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
576 	if (error) {
577 		m_freem(mp3);
578 		nfsm_reply(2 * NFSX_UNSIGNED);
579 		nfsm_srvpostop_attr(1, (struct vattr *)0);
580 		return (0);
581 	}
582 	if (vp->v_type != VLNK) {
583 		if (v3)
584 			error = EINVAL;
585 		else
586 			error = ENXIO;
587 		goto out;
588 	}
589 	nqsrv_getl(vp, ND_READ);
590 	error = VOP_READLINK(vp, uiop, cred);
591 out:
592 	getret = VOP_GETATTR(vp, &attr, cred);
593 	vput(vp);
594 	if (error)
595 		m_freem(mp3);
596 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
597 	if (v3) {
598 		nfsm_srvpostop_attr(getret, &attr);
599 		if (error)
600 			return (0);
601 	}
602 	len -= uiop->uio_resid;
603 	padlen = nfsm_padlen(len);
604 	if (len == 0) {
605 		m_freem(mp3);
606 		mp3 = NULL;
607 	} else if (uiop->uio_resid || padlen)
608 		nfs_zeropad(mp3, uiop->uio_resid, padlen);
609 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
610 	*tl = txdr_unsigned(len);
611 	mb->m_next = mp3;
612 	nfsm_srvdone;
613 }
614 
615 /*
616  * nfs read service
617  */
618 int
619 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
620 {
621 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
622 	struct mbuf *nam = nfsd->nd_nam;
623 	char *dpos = nfsd->nd_dpos;
624 	kauth_cred_t cred = nfsd->nd_cr;
625 	struct mbuf *m;
626 	struct nfs_fattr *fp;
627 	u_int32_t *tl;
628 	int32_t t1;
629 	int i;
630 	char *bpos;
631 	int error = 0, rdonly, cache = 0, getret;
632 	int v3 = (nfsd->nd_flag & ND_NFSV3);
633 	uint32_t reqlen, len, cnt, left;
634 	int padlen;
635 	char *cp2;
636 	struct mbuf *mb, *mreq;
637 	struct vnode *vp;
638 	nfsrvfh_t nsfh;
639 	struct uio io, *uiop = &io;
640 	struct vattr va;
641 	off_t off;
642 	u_quad_t frev;
643 
644 	nfsm_srvmtofh(&nsfh);
645 	if (v3) {
646 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
647 		off = fxdr_hyper(tl);
648 	} else {
649 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
650 		off = (off_t)fxdr_unsigned(u_int32_t, *tl);
651 	}
652 	nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
653 	reqlen = fxdr_unsigned(uint32_t, *tl);
654 	reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd));
655 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
656 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
657 	if (error) {
658 		nfsm_reply(2 * NFSX_UNSIGNED);
659 		nfsm_srvpostop_attr(1, (struct vattr *)0);
660 		return (0);
661 	}
662 	if (vp->v_type != VREG) {
663 		if (v3)
664 			error = EINVAL;
665 		else
666 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
667 	}
668 	if (!error) {
669 	    nqsrv_getl(vp, ND_READ);
670 	    if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0)
671 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1);
672 	}
673 	getret = VOP_GETATTR(vp, &va, cred);
674 	if (!error)
675 		error = getret;
676 	if (error) {
677 		vput(vp);
678 		nfsm_reply(NFSX_POSTOPATTR(v3));
679 		nfsm_srvpostop_attr(getret, &va);
680 		return (0);
681 	}
682 	if (off >= va.va_size)
683 		cnt = 0;
684 	else if ((off + reqlen) > va.va_size)
685 		cnt = va.va_size - off;
686 	else
687 		cnt = reqlen;
688 	nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
689 	if (v3) {
690 		nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
691 		*tl++ = nfs_true;
692 		fp = (struct nfs_fattr *)tl;
693 		tl += (NFSX_V3FATTR / sizeof (u_int32_t));
694 	} else {
695 		nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
696 		fp = (struct nfs_fattr *)tl;
697 		tl += (NFSX_V2FATTR / sizeof (u_int32_t));
698 	}
699 	len = left = cnt;
700 	if (cnt > 0) {
701 		if (nfsd_use_loan) {
702 			struct vm_page **pgpp;
703 			voff_t pgoff = trunc_page(off);
704 			int npages;
705 			vaddr_t lva;
706 
707 			npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT;
708 			KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */
709 
710 			/* allocate kva for mbuf data */
711 			lva = sokvaalloc(pgoff, npages << PAGE_SHIFT,
712 			    slp->ns_so);
713 			if (lva == 0) {
714 				/* fall back to VOP_READ */
715 				goto loan_fail;
716 			}
717 
718 			/* allocate mbuf */
719 			m = m_get(M_WAIT, MT_DATA);
720 			MCLAIM(m, &nfs_mowner);
721 			pgpp = m->m_ext.ext_pgs;
722 
723 			/* loan pages */
724 			error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages,
725 			    pgpp);
726 			if (error) {
727 				sokvafree(lva, npages << PAGE_SHIFT);
728 				m_free(m);
729 				if (error == EBUSY)
730 					goto loan_fail;
731 				goto read_error;
732 			}
733 
734 			/* associate kva to mbuf */
735 			MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)),
736 			    cnt, M_MBUF, soloanfree, slp->ns_so);
737 			m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
738 			m->m_len = cnt;
739 
740 			/* map pages */
741 			for (i = 0; i < npages; i++) {
742 				pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]),
743 				    VM_PROT_READ, 0);
744 				lva += PAGE_SIZE;
745 			}
746 
747 			pmap_update(pmap_kernel());
748 
749 			mb->m_next = m;
750 			mb = m;
751 			error = 0;
752 			uiop->uio_resid = 0;
753 		} else {
754 			struct iovec *iv;
755 			struct iovec *iv2;
756 			struct mbuf *m2;
757 			int siz;
758 loan_fail:
759 			/*
760 			 * Generate the mbuf list with the uio_iov ref. to it.
761 			 */
762 			i = 0;
763 			m = m2 = mb;
764 			while (left > 0) {
765 				siz = uimin(M_TRAILINGSPACE(m), left);
766 				if (siz > 0) {
767 					left -= siz;
768 					i++;
769 				}
770 				if (left > 0) {
771 					m = m_get(M_WAIT, MT_DATA);
772 					MCLAIM(m, &nfs_mowner);
773 					m_clget(m, M_WAIT);
774 					m->m_len = 0;
775 					m2->m_next = m;
776 					m2 = m;
777 				}
778 			}
779 			iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK);
780 			uiop->uio_iov = iv2 = iv;
781 			m = mb;
782 			left = cnt;
783 			i = 0;
784 			while (left > 0) {
785 				if (m == NULL)
786 					panic("nfsrv_read iov");
787 				siz = uimin(M_TRAILINGSPACE(m), left);
788 				if (siz > 0) {
789 					iv->iov_base = mtod(m, char *) +
790 					    m->m_len;
791 					iv->iov_len = siz;
792 					m->m_len += siz;
793 					left -= siz;
794 					iv++;
795 					i++;
796 				}
797 				m = m->m_next;
798 			}
799 			uiop->uio_iovcnt = i;
800 			uiop->uio_offset = off;
801 			uiop->uio_resid = cnt;
802 			uiop->uio_rw = UIO_READ;
803 			UIO_SETUP_SYSSPACE(uiop);
804 			error = VOP_READ(vp, uiop, IO_NODELOCKED, cred);
805 			free((void *)iv2, M_TEMP);
806 		}
807 read_error:
808 		if (error || (getret = VOP_GETATTR(vp, &va, cred)) != 0){
809 			if (!error)
810 				error = getret;
811 			m_freem(mreq);
812 			vput(vp);
813 			nfsm_reply(NFSX_POSTOPATTR(v3));
814 			nfsm_srvpostop_attr(getret, &va);
815 			return (0);
816 		}
817 	} else {
818 		uiop->uio_resid = 0;
819 	}
820 	vput(vp);
821 	nfsm_srvfillattr(&va, fp);
822 	len -= uiop->uio_resid;
823 	padlen = nfsm_padlen(len);
824 	if (uiop->uio_resid || padlen)
825 		nfs_zeropad(mb, uiop->uio_resid, padlen);
826 	if (v3) {
827 		/* count */
828 		*tl++ = txdr_unsigned(len);
829 		/* eof */
830 		if (off + len >= va.va_size)
831 			*tl++ = nfs_true;
832 		else
833 			*tl++ = nfs_false;
834 	}
835 	*tl = txdr_unsigned(len);
836 	nfsm_srvdone;
837 }
838 
839 /*
840  * nfs write service
841  */
842 int
843 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
844 {
845 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
846 	struct mbuf *nam = nfsd->nd_nam;
847 	char *dpos = nfsd->nd_dpos;
848 	kauth_cred_t cred = nfsd->nd_cr;
849 	struct iovec *ivp;
850 	int i, cnt;
851 	struct mbuf *mp;
852 	struct nfs_fattr *fp;
853 	struct iovec *iv;
854 	struct vattr va, forat;
855 	u_int32_t *tl;
856 	int32_t t1;
857 	char *bpos;
858 	int error = 0, rdonly, cache = 0, len, forat_ret = 1;
859 	int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
860 	int stable = NFSV3WRITE_FILESYNC;
861 	int v3 = (nfsd->nd_flag & ND_NFSV3);
862 	char *cp2;
863 	struct mbuf *mb, *mreq __unused;
864 	struct vnode *vp;
865 	nfsrvfh_t nsfh;
866 	struct uio io, *uiop = &io;
867 	off_t off;
868 	u_quad_t frev;
869 
870 	if (mrep == NULL) {
871 		*mrq = NULL;
872 		return (0);
873 	}
874 	nfsm_srvmtofh(&nsfh);
875 	if (v3) {
876 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
877 		off = fxdr_hyper(tl);
878 		tl += 3;
879 		stable = fxdr_unsigned(int, *tl++);
880 	} else {
881 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
882 		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
883 		tl += 2;
884 	}
885 	retlen = len = fxdr_unsigned(int32_t, *tl);
886 	cnt = i = 0;
887 
888 	/*
889 	 * For NFS Version 2, it is not obvious what a write of zero length
890 	 * should do, but I might as well be consistent with Version 3,
891 	 * which is to return ok so long as there are no permission problems.
892 	 */
893 	if (len > 0) {
894 		zeroing = 1;
895 		mp = mrep;
896 		while (mp) {
897 			if (mp == md) {
898 				zeroing = 0;
899 				adjust = dpos - mtod(mp, char *);
900 				mp->m_len -= adjust;
901 				if (mp->m_len > 0 && adjust > 0)
902 					NFSMADV(mp, adjust);
903 			}
904 			if (zeroing)
905 				mp->m_len = 0;
906 			else if (mp->m_len > 0) {
907 				i += mp->m_len;
908 				if (i > len) {
909 					mp->m_len -= (i - len);
910 					zeroing	= 1;
911 				}
912 				if (mp->m_len > 0)
913 					cnt++;
914 			}
915 			mp = mp->m_next;
916 		}
917 	}
918 	if (len > NFS_MAXDATA || len < 0 || i < len) {
919 		error = EIO;
920 		nfsm_reply(2 * NFSX_UNSIGNED);
921 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
922 		return (0);
923 	}
924 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
925 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
926 	if (error) {
927 		nfsm_reply(2 * NFSX_UNSIGNED);
928 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
929 		return (0);
930 	}
931 	if (v3)
932 		forat_ret = VOP_GETATTR(vp, &forat, cred);
933 	if (vp->v_type != VREG) {
934 		if (v3)
935 			error = EINVAL;
936 		else
937 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
938 	}
939 	if (!error) {
940 		nqsrv_getl(vp, ND_WRITE);
941 		error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1);
942 	}
943 	if (error) {
944 		vput(vp);
945 		nfsm_reply(NFSX_WCCDATA(v3));
946 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
947 		return (0);
948 	}
949 
950 	if (len > 0) {
951 		ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK);
952 		uiop->uio_iov = iv = ivp;
953 		uiop->uio_iovcnt = cnt;
954 		mp = mrep;
955 		while (mp) {
956 			if (mp->m_len > 0) {
957 				ivp->iov_base = mtod(mp, void *);
958 				ivp->iov_len = mp->m_len;
959 				ivp++;
960 			}
961 			mp = mp->m_next;
962 		}
963 
964 		/*
965 		 * XXX
966 		 * The IO_METASYNC flag indicates that all metadata (and not
967 		 * just enough to ensure data integrity) must be written to
968 		 * stable storage synchronously.
969 		 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
970 		 */
971 		if (stable == NFSV3WRITE_UNSTABLE)
972 			ioflags = IO_NODELOCKED;
973 		else if (stable == NFSV3WRITE_DATASYNC)
974 			ioflags = (IO_SYNC | IO_NODELOCKED);
975 		else
976 			ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
977 		uiop->uio_resid = len;
978 		uiop->uio_rw = UIO_WRITE;
979 		uiop->uio_offset = off;
980 		UIO_SETUP_SYSSPACE(uiop);
981 		error = VOP_WRITE(vp, uiop, ioflags, cred);
982 		nfsstats.srvvop_writes++;
983 		free(iv, M_TEMP);
984 	}
985 	aftat_ret = VOP_GETATTR(vp, &va, cred);
986 	vput(vp);
987 	if (!error)
988 		error = aftat_ret;
989 	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
990 		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
991 	if (v3) {
992 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
993 		if (error)
994 			return (0);
995 		nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
996 		*tl++ = txdr_unsigned(retlen);
997 		if (stable == NFSV3WRITE_UNSTABLE)
998 			*tl++ = txdr_unsigned(stable);
999 		else
1000 			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1001 		/*
1002 		 * Actually, there is no need to txdr these fields,
1003 		 * but it may make the values more human readable,
1004 		 * for debugging purposes.
1005 		 */
1006 		struct timeval btv;
1007 		getmicroboottime(&btv);
1008 		*tl++ = txdr_unsigned(btv.tv_sec);
1009 		*tl = txdr_unsigned(btv.tv_usec);
1010 	} else {
1011 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1012 		nfsm_srvfillattr(&va, fp);
1013 	}
1014 	nfsm_srvdone;
1015 }
1016 
1017 /*
1018  * XXX elad: the original NFSW_SAMECRED() macro also made sure the
1019  *	     two nd_flag fields of the descriptors contained
1020  *	     ND_KERBAUTH.
1021  */
1022 static int
1023 nfsrv_samecred(kauth_cred_t cred1, kauth_cred_t cred2)
1024 {
1025 	int i, do_ngroups;
1026 
1027 	if (kauth_cred_geteuid(cred1) != kauth_cred_geteuid(cred2))
1028 		return (0);
1029 	if (kauth_cred_ngroups(cred1) != kauth_cred_ngroups(cred2))
1030 		return (0);
1031 	do_ngroups = kauth_cred_ngroups(cred1);
1032 	for (i = 0; i < do_ngroups; i++)
1033 		if (kauth_cred_group(cred1, i) !=
1034 		    kauth_cred_group(cred2, i))
1035 			return (0);
1036 
1037 	return (1);
1038 }
1039 
1040 static struct nfsrvw_delayhash *
1041 nfsrv_nwdelayhash(struct nfssvc_sock *slp, const nfsrvfh_t *nsfh)
1042 {
1043 	uint32_t hash;
1044 
1045 	hash = hash32_buf(NFSRVFH_DATA(nsfh), NFSRVFH_SIZE(nsfh),
1046 	    HASH32_BUF_INIT);
1047 	return &slp->ns_wdelayhashtbl[hash % NFS_WDELAYHASHSIZ];
1048 }
1049 
1050 /*
1051  * NFS write service with write gathering support. Called when
1052  * nfsrvw_procrastinate > 0.
1053  * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1054  * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1055  * Jan. 1994.
1056  */
1057 int
1058 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1059 {
1060 	struct timeval now;
1061 	struct iovec *ivp;
1062 	struct mbuf *mp;
1063 	struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1064 	struct nfs_fattr *fp;
1065 	int i = 0;
1066 	struct iovec *iov;
1067 	struct nfsrvw_delayhash *wpp;
1068 	kauth_cred_t cred;
1069 	struct vattr va, forat;
1070 	u_int32_t *tl;
1071 	int32_t t1;
1072 	char *bpos, *dpos;
1073 	int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1;
1074 	int ioflags, aftat_ret = 1, adjust, v3, zeroing;
1075 	char *cp2;
1076 	struct mbuf *mb, *mreq, *mrep, *md;
1077 	struct vnode *vp;
1078 	struct uio io, *uiop = &io;
1079 	u_quad_t frev, cur_usec;
1080 
1081 	*mrq = NULL;
1082 	if (*ndp) {
1083 	    nfsd = *ndp;
1084 	    *ndp = NULL;
1085 	    mrep = nfsd->nd_mrep;
1086 	    md = nfsd->nd_md;
1087 	    dpos = nfsd->nd_dpos;
1088 	    cred = nfsd->nd_cr;
1089 	    v3 = (nfsd->nd_flag & ND_NFSV3);
1090 	    LIST_INIT(&nfsd->nd_coalesce);
1091 	    nfsd->nd_mreq = NULL;
1092 	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1093 	    getmicrotime(&now);
1094 	    cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec;
1095 	    nfsd->nd_time = cur_usec + nfsrvw_procrastinate;
1096 
1097 	    /*
1098 	     * Now, get the write header..
1099 	     */
1100 	    nfsm_srvmtofh(&nfsd->nd_fh);
1101 	    if (v3) {
1102 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1103 		nfsd->nd_off = fxdr_hyper(tl);
1104 		tl += 3;
1105 		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1106 	    } else {
1107 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1108 		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1109 		tl += 2;
1110 	    }
1111 	    len = fxdr_unsigned(int32_t, *tl);
1112 	    nfsd->nd_len = len;
1113 	    nfsd->nd_eoff = nfsd->nd_off + len;
1114 
1115 	    /*
1116 	     * Trim the header out of the mbuf list and trim off any trailing
1117 	     * junk so that the mbuf list has only the write data.
1118 	     */
1119 	    zeroing = 1;
1120 	    i = 0;
1121 	    mp = mrep;
1122 	    while (mp) {
1123 		if (mp == md) {
1124 		    zeroing = 0;
1125 		    adjust = dpos - mtod(mp, char *);
1126 		    mp->m_len -= adjust;
1127 		    if (mp->m_len > 0 && adjust > 0)
1128 			NFSMADV(mp, adjust);
1129 		}
1130 		if (zeroing)
1131 		    mp->m_len = 0;
1132 		else {
1133 		    i += mp->m_len;
1134 		    if (i > len) {
1135 			mp->m_len -= (i - len);
1136 			zeroing = 1;
1137 		    }
1138 		}
1139 		mp = mp->m_next;
1140 	    }
1141 	    if (len > NFS_MAXDATA || len < 0  || i < len) {
1142 nfsmout:
1143 		m_freem(mrep);
1144 		error = EIO;
1145 		nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1146 		if (v3)
1147 		    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1148 		nfsd->nd_mreq = mreq;
1149 		nfsd->nd_mrep = NULL;
1150 		nfsd->nd_time = 0;
1151 	    }
1152 
1153 	    /*
1154 	     * Add this entry to the hash and time queues.
1155 	     */
1156 	    owp = NULL;
1157 	    mutex_enter(&nfsd_lock);
1158 	    wp = LIST_FIRST(&slp->ns_tq);
1159 	    while (wp && wp->nd_time < nfsd->nd_time) {
1160 		owp = wp;
1161 		wp = LIST_NEXT(wp, nd_tq);
1162 	    }
1163 	    if (owp) {
1164 		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1165 	    } else {
1166 		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1167 	    }
1168 	    if (nfsd->nd_mrep) {
1169 		wpp = nfsrv_nwdelayhash(slp, &nfsd->nd_fh);
1170 		owp = NULL;
1171 		wp = LIST_FIRST(wpp);
1172 		while (wp && nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) {
1173 		    owp = wp;
1174 		    wp = LIST_NEXT(wp, nd_hash);
1175 		}
1176 		while (wp && wp->nd_off < nfsd->nd_off &&
1177 		    !nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) {
1178 		    owp = wp;
1179 		    wp = LIST_NEXT(wp, nd_hash);
1180 		}
1181 		if (owp) {
1182 		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1183 
1184 		    /*
1185 		     * Search the hash list for overlapping entries and
1186 		     * coalesce.
1187 		     */
1188 		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1189 			wp = LIST_NEXT(nfsd, nd_hash);
1190 			if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr))
1191 			    nfsrvw_coalesce(owp, nfsd);
1192 		    }
1193 		} else {
1194 		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1195 		}
1196 	    }
1197 	    mutex_exit(&nfsd_lock);
1198 	}
1199 
1200 	/*
1201 	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1202 	 * and generate the associated reply mbuf list(s).
1203 	 */
1204 loop1:
1205 	getmicrotime(&now);
1206 	cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec;
1207 	mutex_enter(&nfsd_lock);
1208 	for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1209 		owp = LIST_NEXT(nfsd, nd_tq);
1210 		if (nfsd->nd_time > cur_usec)
1211 		    break;
1212 		if (nfsd->nd_mreq)
1213 		    continue;
1214 		LIST_REMOVE(nfsd, nd_tq);
1215 		LIST_REMOVE(nfsd, nd_hash);
1216 		mutex_exit(&nfsd_lock);
1217 
1218 		mrep = nfsd->nd_mrep;
1219 		nfsd->nd_mrep = NULL;
1220 		cred = nfsd->nd_cr;
1221 		v3 = (nfsd->nd_flag & ND_NFSV3);
1222 		forat_ret = aftat_ret = 1;
1223 		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1224 		    nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH),
1225 		    false);
1226 		if (!error) {
1227 		    if (v3)
1228 			forat_ret = VOP_GETATTR(vp, &forat, cred);
1229 		    if (vp->v_type != VREG) {
1230 			if (v3)
1231 			    error = EINVAL;
1232 			else
1233 			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1234 		    }
1235 		} else
1236 		    vp = NULL;
1237 		if (!error) {
1238 		    nqsrv_getl(vp, ND_WRITE);
1239 		    error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1);
1240 		}
1241 
1242 		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1243 		    ioflags = IO_NODELOCKED;
1244 		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1245 		    ioflags = (IO_SYNC | IO_NODELOCKED);
1246 		else
1247 		    ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1248 		uiop->uio_rw = UIO_WRITE;
1249 		uiop->uio_offset = nfsd->nd_off;
1250 		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1251 		UIO_SETUP_SYSSPACE(uiop);
1252 		if (uiop->uio_resid > 0) {
1253 		    mp = mrep;
1254 		    i = 0;
1255 		    while (mp) {
1256 			if (mp->m_len > 0)
1257 			    i++;
1258 			mp = mp->m_next;
1259 		    }
1260 		    uiop->uio_iovcnt = i;
1261 		    iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1262 		    uiop->uio_iov = ivp = iov;
1263 		    mp = mrep;
1264 		    while (mp) {
1265 			if (mp->m_len > 0) {
1266 			    ivp->iov_base = mtod(mp, void *);
1267 			    ivp->iov_len = mp->m_len;
1268 			    ivp++;
1269 			}
1270 			mp = mp->m_next;
1271 		    }
1272 		    if (!error) {
1273 			error = VOP_WRITE(vp, uiop, ioflags, cred);
1274 			nfsstats.srvvop_writes++;
1275 		    }
1276 		    free((void *)iov, M_TEMP);
1277 		}
1278 		m_freem(mrep);
1279 		if (vp) {
1280 		    aftat_ret = VOP_GETATTR(vp, &va, cred);
1281 		    vput(vp);
1282 		}
1283 
1284 		/*
1285 		 * Loop around generating replies for all write rpcs that have
1286 		 * now been completed.
1287 		 */
1288 		swp = nfsd;
1289 		do {
1290 		    if (error) {
1291 			nfsm_writereply(NFSX_WCCDATA(v3), v3);
1292 			if (v3) {
1293 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1294 			}
1295 		    } else {
1296 			nfsm_writereply(NFSX_PREOPATTR(v3) +
1297 			    NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1298 			    NFSX_WRITEVERF(v3), v3);
1299 			if (v3) {
1300 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1301 			    nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1302 			    *tl++ = txdr_unsigned(nfsd->nd_len);
1303 			    *tl++ = txdr_unsigned(swp->nd_stable);
1304 			    /*
1305 			     * Actually, there is no need to txdr these fields,
1306 			     * but it may make the values more human readable,
1307 			     * for debugging purposes.
1308 			     */
1309 			    struct timeval btv;
1310 			    getmicroboottime(&btv);
1311 			    *tl++ = txdr_unsigned(btv.tv_sec);
1312 			    *tl = txdr_unsigned(btv.tv_usec);
1313 			} else {
1314 			    nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1315 			    nfsm_srvfillattr(&va, fp);
1316 			}
1317 		    }
1318 		    nfsd->nd_mreq = mreq;
1319 		    if (nfsd->nd_mrep)
1320 			panic("nfsrv_write: nd_mrep not free");
1321 
1322 		    /*
1323 		     * Done. Put it at the head of the timer queue so that
1324 		     * the final phase can return the reply.
1325 		     */
1326 		    mutex_enter(&nfsd_lock);
1327 		    if (nfsd != swp) {
1328 			nfsd->nd_time = 0;
1329 			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1330 		    }
1331 		    nfsd = LIST_FIRST(&swp->nd_coalesce);
1332 		    if (nfsd) {
1333 			LIST_REMOVE(nfsd, nd_tq);
1334 		    }
1335 		    mutex_exit(&nfsd_lock);
1336 		} while (nfsd);
1337 		swp->nd_time = 0;
1338 
1339 		mutex_enter(&nfsd_lock);
1340 		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1341 		mutex_exit(&nfsd_lock);
1342 		goto loop1;
1343 	}
1344 	mutex_exit(&nfsd_lock);
1345 	nfs_timer_start();
1346 
1347 	/*
1348 	 * Search for a reply to return.
1349 	 */
1350 	mutex_enter(&nfsd_lock);
1351 	LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) {
1352 		if (nfsd->nd_mreq) {
1353 		    LIST_REMOVE(nfsd, nd_tq);
1354 		    *mrq = nfsd->nd_mreq;
1355 		    *ndp = nfsd;
1356 		    break;
1357 		}
1358 	}
1359 	mutex_exit(&nfsd_lock);
1360 	return (0);
1361 }
1362 
1363 /*
1364  * Coalesce the write request nfsd into owp. To do this we must:
1365  * - remove nfsd from the queues
1366  * - merge nfsd->nd_mrep into owp->nd_mrep
1367  * - update the nd_eoff and nd_stable for owp
1368  * - put nfsd on owp's nd_coalesce list
1369  * NB: Must be called at splsoftclock().
1370  */
1371 void
1372 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1373 {
1374         int overlap;
1375         struct mbuf *mp;
1376 	struct nfsrv_descript *m;
1377 
1378 	KASSERT(mutex_owned(&nfsd_lock));
1379 
1380         LIST_REMOVE(nfsd, nd_hash);
1381         LIST_REMOVE(nfsd, nd_tq);
1382         if (owp->nd_eoff < nfsd->nd_eoff) {
1383             overlap = owp->nd_eoff - nfsd->nd_off;
1384             if (overlap < 0)
1385                 panic("nfsrv_coalesce: bad off");
1386             if (overlap > 0)
1387                 m_adj(nfsd->nd_mrep, overlap);
1388             mp = owp->nd_mrep;
1389             while (mp->m_next)
1390                 mp = mp->m_next;
1391             mp->m_next = nfsd->nd_mrep;
1392             owp->nd_eoff = nfsd->nd_eoff;
1393         } else
1394             m_freem(nfsd->nd_mrep);
1395         nfsd->nd_mrep = NULL;
1396         if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1397             owp->nd_stable = NFSV3WRITE_FILESYNC;
1398         else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1399             owp->nd_stable == NFSV3WRITE_UNSTABLE)
1400             owp->nd_stable = NFSV3WRITE_DATASYNC;
1401         LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1402  	/*
1403  	 * nfsd might hold coalesce elements! Move them to owp.
1404  	 * Otherwise, requests may be lost and clients will be stuck.
1405  	 */
1406 	while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) {
1407 		LIST_REMOVE(m, nd_tq);
1408 		LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
1409 	}
1410 }
1411 
1412 /*
1413  * nfs create service
1414  * now does a truncate to 0 length via. setattr if it already exists
1415  */
1416 int
1417 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1418 {
1419 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1420 	struct mbuf *nam = nfsd->nd_nam;
1421 	char *dpos = nfsd->nd_dpos;
1422 	kauth_cred_t cred = nfsd->nd_cr;
1423 	struct nfs_fattr *fp;
1424 	struct vattr va, dirfor, diraft;
1425 	struct nfsv2_sattr *sp;
1426 	u_int32_t *tl;
1427 	struct nameidata nd;
1428 	char *cp;
1429 	int32_t t1;
1430 	char *bpos;
1431 	int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1432 	int rdev = 0, abort = 0;
1433 	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1434 	char *cp2;
1435 	struct mbuf *mb, *mreq __unused;
1436 	struct vnode *vp = NULL, *dirp = NULL;
1437 	nfsrvfh_t nsfh;
1438 	u_quad_t frev, tempsize;
1439 	u_char cverf[NFSX_V3CREATEVERF];
1440 
1441 	nd.ni_cnd.cn_nameiop = 0;
1442 	nfsm_srvmtofh(&nsfh);
1443 	nfsm_srvnamesiz(len);
1444 	nd.ni_cnd.cn_cred = cred;
1445 	nd.ni_cnd.cn_nameiop = CREATE;
1446 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1447 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1448 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1449 	if (dirp && v3) {
1450 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1451 	}
1452 	if (error) {
1453 		nfsm_reply(NFSX_WCCDATA(v3));
1454 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1455 		if (dirp)
1456 			vrele(dirp);
1457 		if (nd.ni_pathbuf != NULL) {
1458 			pathbuf_destroy(nd.ni_pathbuf);
1459 			nd.ni_pathbuf = NULL;
1460 		}
1461 		return (0);
1462 	}
1463 	abort = 1;
1464 	vattr_null(&va);
1465 	if (v3) {
1466 		va.va_mode = 0;
1467 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1468 		how = fxdr_unsigned(int, *tl);
1469 		switch (how) {
1470 		case NFSV3CREATE_GUARDED:
1471 			if (nd.ni_vp) {
1472 				error = EEXIST;
1473 				break;
1474 			}
1475 			/* FALLTHROUGH */
1476 		case NFSV3CREATE_UNCHECKED:
1477 			nfsm_srvsattr(&va);
1478 			break;
1479 		case NFSV3CREATE_EXCLUSIVE:
1480 			nfsm_dissect(cp, void *, NFSX_V3CREATEVERF);
1481 			memcpy(cverf, cp, NFSX_V3CREATEVERF);
1482 			exclusive_flag = 1;
1483 			break;
1484 		};
1485 		va.va_type = VREG;
1486 	} else {
1487 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1488 		va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1489 		if (va.va_type == VNON)
1490 			va.va_type = VREG;
1491 		va.va_mode = nfstov_mode(sp->sa_mode);
1492 		switch (va.va_type) {
1493 		case VREG:
1494 			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1495 			if (tsize != -1)
1496 				va.va_size = (u_quad_t)tsize;
1497 			break;
1498 		case VCHR:
1499 		case VBLK:
1500 		case VFIFO:
1501 			rdev = fxdr_unsigned(int32_t, sp->sa_size);
1502 			break;
1503 		default:
1504 			break;
1505 		};
1506 	}
1507 
1508 	/*
1509 	 * Iff doesn't exist, create it
1510 	 * otherwise just truncate to 0 length
1511 	 *   should I set the mode too ??
1512 	 */
1513 	if (nd.ni_vp == NULL) {
1514 		if (va.va_type == VREG || va.va_type == VSOCK) {
1515 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1516 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1517 			if (!error) {
1518 				vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1519 				if (exclusive_flag) {
1520 					exclusive_flag = 0;
1521 					vattr_null(&va);
1522 					/*
1523 					 * XXX
1524 					 * assuming NFSX_V3CREATEVERF
1525 					 * == sizeof(nfstime3)
1526 					 */
1527 					fxdr_nfsv3time(cverf, &va.va_atime);
1528 					error = VOP_SETATTR(nd.ni_vp, &va,
1529 						cred);
1530 				}
1531 			}
1532 		} else if (va.va_type == VCHR || va.va_type == VBLK ||
1533 			va.va_type == VFIFO) {
1534 			if (va.va_type == VCHR && rdev == 0xffffffff)
1535 				va.va_type = VFIFO;
1536 			if (va.va_type != VFIFO &&
1537 			    (error = kauth_authorize_system(cred,
1538 			    KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) {
1539 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1540 				vput(nd.ni_dvp);
1541 				abort = 0;
1542 				nfsm_reply(0);
1543 				if (nd.ni_pathbuf != NULL) {
1544 					pathbuf_destroy(nd.ni_pathbuf);
1545 					nd.ni_pathbuf = NULL;
1546 				}
1547 				return (error);
1548 			} else
1549 				va.va_rdev = (dev_t)rdev;
1550 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1551 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd,
1552 			    &va);
1553 			if (!error) {
1554 				vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1555 			} else {
1556 				nfsm_reply(0);
1557 			}
1558 		} else {
1559 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1560 			if (nd.ni_pathbuf != NULL) {
1561 				pathbuf_destroy(nd.ni_pathbuf);
1562 				nd.ni_pathbuf = NULL;
1563 			}
1564 			error = ENXIO;
1565 			abort = 0;
1566 		}
1567 		vp = nd.ni_vp;
1568 	} else {
1569 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1570 		if (nd.ni_pathbuf != NULL) {
1571 			pathbuf_destroy(nd.ni_pathbuf);
1572 			nd.ni_pathbuf = NULL;
1573 		}
1574 		vp = nd.ni_vp;
1575 		abort = 0;
1576 		if (!error && va.va_size != -1) {
1577 			error = nfsrv_access(vp, VWRITE, cred,
1578 			    (nd.ni_cnd.cn_flags & RDONLY), lwp, 0);
1579 			if (!error) {
1580 				nqsrv_getl(vp, ND_WRITE);
1581 				tempsize = va.va_size;
1582 				vattr_null(&va);
1583 				va.va_size = tempsize;
1584 				error = VOP_SETATTR(vp, &va, cred);
1585 			}
1586 		}
1587 		if (error)
1588 			vput(vp);
1589 	}
1590 	if (!error) {
1591 		error = nfsrv_composefh(vp, &nsfh, v3);
1592 		if (!error)
1593 			error = VOP_GETATTR(vp, &va, cred);
1594 		vput(vp);
1595 	}
1596 	if (nd.ni_dvp == vp)
1597 		vrele(nd.ni_dvp);
1598 	else
1599 		vput(nd.ni_dvp);
1600 	if (v3) {
1601 		if (exclusive_flag && !error) {
1602 			/*
1603 			 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3)
1604 			 */
1605 			char oldverf[NFSX_V3CREATEVERF];
1606 
1607 			txdr_nfsv3time(&va.va_atime, oldverf);
1608 			if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF))
1609 				error = EEXIST;
1610 		}
1611 		if (dirp) {
1612 			vn_lock(dirp, LK_SHARED | LK_RETRY);
1613 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1614 			VOP_UNLOCK(dirp);
1615 		}
1616 	}
1617 	if (dirp) {
1618 		vrele(dirp);
1619 		dirp = NULL;
1620 	}
1621 	if (nd.ni_pathbuf != NULL) {
1622 		pathbuf_destroy(nd.ni_pathbuf);
1623 		nd.ni_pathbuf = NULL;
1624 	}
1625 	abort = 0;
1626 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1627 	if (v3) {
1628 		if (!error) {
1629 			nfsm_srvpostop_fh(&nsfh);
1630 			nfsm_srvpostop_attr(0, &va);
1631 		}
1632 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1633 	} else {
1634 		nfsm_srvfhtom(&nsfh, v3);
1635 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1636 		nfsm_srvfillattr(&va, fp);
1637 	}
1638 	return (0);
1639 nfsmout:
1640 	if (dirp)
1641 		vrele(dirp);
1642 	if (abort) {
1643 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1644 		if (nd.ni_dvp == nd.ni_vp)
1645 			vrele(nd.ni_dvp);
1646 		else
1647 			vput(nd.ni_dvp);
1648 		if (nd.ni_vp)
1649 			vput(nd.ni_vp);
1650 	}
1651 	if (nd.ni_pathbuf != NULL) {
1652 		pathbuf_destroy(nd.ni_pathbuf);
1653 		nd.ni_pathbuf = NULL;
1654 	}
1655 	return (error);
1656 }
1657 
1658 /*
1659  * nfs v3 mknod service
1660  */
1661 int
1662 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1663 {
1664 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1665 	struct mbuf *nam = nfsd->nd_nam;
1666 	char *dpos = nfsd->nd_dpos;
1667 	kauth_cred_t cred = nfsd->nd_cr;
1668 	struct vattr va, dirfor, diraft;
1669 	u_int32_t *tl;
1670 	struct nameidata nd;
1671 	int32_t t1;
1672 	char *bpos;
1673 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
1674 	int abort = 0;
1675 	u_int32_t major, minor;
1676 	enum vtype vtyp;
1677 	char *cp2;
1678 	struct mbuf *mb, *mreq __unused;
1679 	struct vnode *vp, *dirp = (struct vnode *)0;
1680 	nfsrvfh_t nsfh;
1681 	u_quad_t frev;
1682 
1683 	nd.ni_cnd.cn_nameiop = 0;
1684 	nfsm_srvmtofh(&nsfh);
1685 	nfsm_srvnamesiz(len);
1686 	nd.ni_cnd.cn_cred = cred;
1687 	nd.ni_cnd.cn_nameiop = CREATE;
1688 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1689 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1690 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1691 	if (dirp)
1692 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1693 	if (error) {
1694 		nfsm_reply(NFSX_WCCDATA(1));
1695 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1696 		if (dirp)
1697 			vrele(dirp);
1698 		if (nd.ni_pathbuf != NULL) {
1699 			pathbuf_destroy(nd.ni_pathbuf);
1700 			nd.ni_pathbuf = NULL;
1701 		}
1702 		return (0);
1703 	}
1704 	abort = 1;
1705 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1706 	vtyp = nfsv3tov_type(*tl);
1707 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1708 		error = NFSERR_BADTYPE;
1709 		goto abort;
1710 	}
1711 	vattr_null(&va);
1712 	va.va_mode = 0;
1713 	nfsm_srvsattr(&va);
1714 	if (vtyp == VCHR || vtyp == VBLK) {
1715 		dev_t rdev;
1716 
1717 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1718 		major = fxdr_unsigned(u_int32_t, *tl++);
1719 		minor = fxdr_unsigned(u_int32_t, *tl);
1720 		rdev = makedev(major, minor);
1721 		if (major(rdev) != major || minor(rdev) != minor) {
1722 			error = EINVAL;
1723 			goto abort;
1724 		}
1725 		va.va_rdev = rdev;
1726 	}
1727 
1728 	/*
1729 	 * Iff doesn't exist, create it.
1730 	 */
1731 	if (nd.ni_vp) {
1732 		error = EEXIST;
1733 abort:
1734 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1735 		if (nd.ni_vp)
1736 			vput(nd.ni_vp);
1737 		if (nd.ni_pathbuf != NULL) {
1738 			pathbuf_destroy(nd.ni_pathbuf);
1739 			nd.ni_pathbuf = NULL;
1740 		}
1741 		goto out;
1742 	}
1743 	va.va_type = vtyp;
1744 	if (vtyp == VSOCK) {
1745 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1746 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1747 	} else {
1748 		if (va.va_type != VFIFO &&
1749 		    (error = kauth_authorize_system(cred,
1750 		    KAUTH_SYSTEM_MKNOD, 0, NULL, NULL, NULL))) {
1751 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1752 			vput(nd.ni_dvp);
1753 			goto out;
1754 		}
1755 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1756 		error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1757 		if (error)
1758 			goto out;
1759 	}
1760 	if (!error) {
1761 		vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1762 	}
1763 out:
1764 	vp = nd.ni_vp;
1765 	if (!error) {
1766 		error = nfsrv_composefh(vp, &nsfh, true);
1767 		if (!error)
1768 			error = VOP_GETATTR(vp, &va, cred);
1769 		vput(vp);
1770 	}
1771 	if (nd.ni_dvp == nd.ni_vp)
1772 		vrele(nd.ni_dvp);
1773 	else
1774 		vput(nd.ni_dvp);
1775 	if (dirp) {
1776 		vn_lock(dirp, LK_SHARED | LK_RETRY);
1777 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1778 		VOP_UNLOCK(dirp);
1779 		vrele(dirp);
1780 		dirp = NULL;
1781 	}
1782 	if (nd.ni_pathbuf != NULL) {
1783 		pathbuf_destroy(nd.ni_pathbuf);
1784 		nd.ni_pathbuf = NULL;
1785 	}
1786 	abort = 0;
1787 	nfsm_reply(NFSX_SRVFH(&nsfh, true) + NFSX_POSTOPATTR(1) +
1788 	    NFSX_WCCDATA(1));
1789 	if (!error) {
1790 		nfsm_srvpostop_fh(&nsfh);
1791 		nfsm_srvpostop_attr(0, &va);
1792 	}
1793 	nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1794 	return (0);
1795 nfsmout:
1796 	if (abort) {
1797 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1798 		if (nd.ni_dvp == nd.ni_vp)
1799 			vrele(nd.ni_dvp);
1800 		else
1801 			vput(nd.ni_dvp);
1802 		if (nd.ni_vp)
1803 			vput(nd.ni_vp);
1804 	}
1805 	if (nd.ni_pathbuf != NULL) {
1806 		pathbuf_destroy(nd.ni_pathbuf);
1807 		nd.ni_pathbuf = NULL;
1808 	}
1809 	if (dirp)
1810 		vrele(dirp);
1811 	return (error);
1812 }
1813 
1814 /*
1815  * nfs remove service
1816  */
1817 int
1818 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1819 {
1820 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1821 	struct mbuf *nam = nfsd->nd_nam;
1822 	char *dpos = nfsd->nd_dpos;
1823 	kauth_cred_t cred = nfsd->nd_cr;
1824 	struct nameidata nd;
1825 	u_int32_t *tl;
1826 	int32_t t1;
1827 	char *bpos;
1828 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
1829 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1830 	char *cp2;
1831 	struct mbuf *mb, *mreq __unused;
1832 	struct vnode *vp, *dirp;
1833 	struct vattr dirfor, diraft;
1834 	nfsrvfh_t nsfh;
1835 	u_quad_t frev;
1836 
1837 #ifndef nolint
1838 	vp = (struct vnode *)0;
1839 #endif
1840 	nfsm_srvmtofh(&nsfh);
1841 	nfsm_srvnamesiz(len);
1842 	nd.ni_cnd.cn_cred = cred;
1843 	nd.ni_cnd.cn_nameiop = DELETE;
1844 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1845 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1846 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1847 	if (error == 0 && dirp && v3) {
1848 		if (nd.ni_dvp == nd.ni_vp)
1849 			vn_lock(dirp, LK_SHARED | LK_RETRY);
1850 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
1851 		if (nd.ni_dvp == nd.ni_vp)
1852 			VOP_UNLOCK(dirp);
1853 	}
1854 	if (!error) {
1855 		vp = nd.ni_vp;
1856 		if (vp->v_type == VDIR) {
1857 			error = EPERM;
1858 			goto out;
1859 		}
1860 		/*
1861 		 * The root of a mounted filesystem cannot be deleted.
1862 		 */
1863 		if (vp->v_vflag & VV_ROOT) {
1864 			error = EBUSY;
1865 		}
1866 out:
1867 		if (!error) {
1868 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1869 			nqsrv_getl(vp, ND_WRITE);
1870 			error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1871 			vput(nd.ni_dvp);
1872 		} else {
1873 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1874 			if (nd.ni_dvp == vp)
1875 				vrele(nd.ni_dvp);
1876 			else
1877 				vput(nd.ni_dvp);
1878 			vput(vp);
1879 		}
1880 	}
1881 	if (nd.ni_pathbuf != NULL) {
1882 		pathbuf_destroy(nd.ni_pathbuf);
1883 		nd.ni_pathbuf = NULL;
1884 	}
1885 	if (dirp) {
1886 		if (v3) {
1887 			vn_lock(dirp, LK_SHARED | LK_RETRY);
1888 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1889 			VOP_UNLOCK(dirp);
1890 		}
1891 		vrele(dirp);
1892 	}
1893 	nfsm_reply(NFSX_WCCDATA(v3));
1894 	if (v3) {
1895 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1896 		return (0);
1897 	}
1898 	nfsm_srvdone;
1899 }
1900 
1901 /*
1902  * nfs rename service
1903  */
1904 int
1905 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
1906 {
1907 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1908 	struct mbuf *nam = nfsd->nd_nam;
1909 	char *dpos = nfsd->nd_dpos;
1910 	kauth_cred_t cred = nfsd->nd_cr;
1911 	u_int32_t *tl;
1912 	int32_t t1;
1913 	char *bpos;
1914 	int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1;
1915 	uint32_t len, len2;
1916 	int tdirfor_ret = 1, tdiraft_ret = 1;
1917 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1918 	char *cp2;
1919 	struct mbuf *mb, *mreq __unused;
1920 	struct nameidata fromnd, tond;
1921 	struct vnode *fvp, *tvp, *tdvp;
1922 	struct vnode *fdirp = NULL, *tdirp = NULL;
1923 	struct mount *localfs = NULL;
1924 	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1925 	nfsrvfh_t fnsfh, tnsfh;
1926 	u_quad_t frev;
1927 	uid_t saved_uid;
1928 
1929 #ifndef nolint
1930 	fvp = (struct vnode *)0;
1931 #endif
1932 	fromnd.ni_cnd.cn_nameiop = 0;
1933 	tond.ni_cnd.cn_nameiop = 0;
1934 	nfsm_srvmtofh(&fnsfh);
1935 	nfsm_srvnamesiz(len);
1936 	/*
1937 	 * Remember our original uid so that we can reset cr_uid before
1938 	 * the second nfs_namei() call, in case it is remapped.
1939 	 */
1940 	saved_uid = kauth_cred_geteuid(cred);
1941 	fromnd.ni_cnd.cn_cred = cred;
1942 	fromnd.ni_cnd.cn_nameiop = DELETE;
1943 	fromnd.ni_cnd.cn_flags = LOCKPARENT;
1944 	error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md,
1945 		&dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
1946 	if (error == 0 && fdirp && v3) {
1947 		if (fromnd.ni_dvp == fromnd.ni_vp)
1948 			vn_lock(fdirp, LK_SHARED | LK_RETRY);
1949 		fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred);
1950 		if (fromnd.ni_dvp == fromnd.ni_vp)
1951 			VOP_UNLOCK(fdirp);
1952 	}
1953 	if (error) {
1954 		fromnd.ni_cnd.cn_nameiop = 0;
1955 		nfsm_reply(2 * NFSX_WCCDATA(v3));
1956 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1957 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1958 		if (fdirp)
1959 			vrele(fdirp);
1960 		if (fromnd.ni_pathbuf != NULL) {
1961 			pathbuf_destroy(fromnd.ni_pathbuf);
1962 		}
1963 		return (0);
1964 	}
1965 	localfs = fromnd.ni_dvp->v_mount;
1966 	fstrans_start(localfs);
1967 	if (fromnd.ni_dvp != fromnd.ni_vp) {
1968 		VOP_UNLOCK(fromnd.ni_dvp);
1969 	}
1970 	fvp = fromnd.ni_vp;
1971 
1972 	error = VFS_RENAMELOCK_ENTER(localfs);
1973 	if (error) {
1974 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1975 		vrele(fromnd.ni_dvp);
1976 		vrele(fvp);
1977 		goto out1;
1978 	}
1979 
1980 	/* Copied, regrettably, from vfs_syscalls.c (q.v.) */
1981 	vrele(fvp);
1982 	if ((fromnd.ni_cnd.cn_namelen == 1 &&
1983 	     fromnd.ni_cnd.cn_nameptr[0] == '.') ||
1984 	    (fromnd.ni_cnd.cn_namelen == 2 &&
1985 	     fromnd.ni_cnd.cn_nameptr[0] == '.' &&
1986 	     fromnd.ni_cnd.cn_nameptr[1] == '.')) {
1987 		error = EINVAL;
1988 		VFS_RENAMELOCK_EXIT(localfs);
1989 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1990 		vrele(fromnd.ni_dvp);
1991 		goto out1;
1992 	}
1993 	vn_lock(fromnd.ni_dvp, LK_EXCLUSIVE | LK_RETRY);
1994 	error = relookup(fromnd.ni_dvp, &fromnd.ni_vp, &fromnd.ni_cnd, 0);
1995 	if (error) {
1996 		VOP_UNLOCK(fromnd.ni_dvp);
1997 		VFS_RENAMELOCK_EXIT(localfs);
1998 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1999 		vrele(fromnd.ni_dvp);
2000 		goto out1;
2001 	}
2002 	VOP_UNLOCK(fromnd.ni_vp);
2003 	if (fromnd.ni_dvp != fromnd.ni_vp)
2004 		VOP_UNLOCK(fromnd.ni_dvp);
2005 	fvp = fromnd.ni_vp;
2006 
2007 	nfsm_srvmtofh(&tnsfh);
2008 	if (v3) {
2009 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
2010 		len2 = fxdr_unsigned(uint32_t, *tl);
2011 		/* len2 will be checked by nfs_namei */
2012 	}
2013 	else {
2014 		/* NFSv2 */
2015 		nfsm_strsiz(len2, NFS_MAXNAMLEN);
2016 	}
2017 	kauth_cred_seteuid(cred, saved_uid);
2018 	tond.ni_cnd.cn_cred = cred;
2019 	tond.ni_cnd.cn_nameiop = RENAME;
2020 	tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE;
2021 	error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md,
2022 		&dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2023 	if (tdirp && v3) {
2024 		tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred);
2025 	}
2026 	if (error) {
2027 		VFS_RENAMELOCK_EXIT(localfs);
2028 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2029 		vrele(fromnd.ni_dvp);
2030 		vrele(fvp);
2031 		goto out1;
2032 	}
2033 	tdvp = tond.ni_dvp;
2034 	tvp = tond.ni_vp;
2035 	if (tvp != NULL) {
2036 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2037 			if (v3)
2038 				error = EEXIST;
2039 			else
2040 				error = EISDIR;
2041 			goto out;
2042 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2043 			if (v3)
2044 				error = EEXIST;
2045 			else
2046 				error = ENOTDIR;
2047 			goto out;
2048 		}
2049 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
2050 			if (v3)
2051 				error = EXDEV;
2052 			else
2053 				error = ENOTEMPTY;
2054 			goto out;
2055 		}
2056 	}
2057 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
2058 		if (v3)
2059 			error = EXDEV;
2060 		else
2061 			error = ENOTEMPTY;
2062 		goto out;
2063 	}
2064 	if (fvp->v_mount != tdvp->v_mount) {
2065 		if (v3)
2066 			error = EXDEV;
2067 		else
2068 			error = ENOTEMPTY;
2069 		goto out;
2070 	}
2071 	if (fvp == tdvp) {
2072 		if (v3)
2073 			error = EINVAL;
2074 		else
2075 			error = ENOTEMPTY;
2076 	}
2077 	/*
2078 	 * If source is the same as the destination (that is the
2079 	 * same vnode with the same name in the same directory),
2080 	 * then there is nothing to do.
2081 	 */
2082 	if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2083 	    fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2084 	    !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2085 	      fromnd.ni_cnd.cn_namelen))
2086 		error = -1;
2087 out:
2088 	if (!error) {
2089 		nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
2090 		nqsrv_getl(tdvp, ND_WRITE);
2091 		if (tvp) {
2092 			nqsrv_getl(tvp, ND_WRITE);
2093 		}
2094 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2095 				   tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2096 		VFS_RENAMELOCK_EXIT(localfs);
2097 	} else {
2098 		VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd);
2099 		if (tdvp == tvp)
2100 			vrele(tdvp);
2101 		else
2102 			vput(tdvp);
2103 		if (tvp)
2104 			vput(tvp);
2105 		VFS_RENAMELOCK_EXIT(localfs);
2106 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2107 		vrele(fromnd.ni_dvp);
2108 		vrele(fvp);
2109 		if (error == -1)
2110 			error = 0;
2111 	}
2112 	if (tond.ni_pathbuf != NULL) {
2113 		pathbuf_destroy(tond.ni_pathbuf);
2114 		tond.ni_pathbuf = NULL;
2115 	}
2116 	tond.ni_cnd.cn_nameiop = 0;
2117 out1:
2118 	if (fdirp) {
2119 		if (v3) {
2120 			vn_lock(fdirp, LK_SHARED | LK_RETRY);
2121 			fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred);
2122 			VOP_UNLOCK(fdirp);
2123 		}
2124 		vrele(fdirp);
2125 		fdirp = NULL;
2126 	}
2127 	if (tdirp) {
2128 		if (v3) {
2129 			vn_lock(tdirp, LK_SHARED | LK_RETRY);
2130 			tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred);
2131 			VOP_UNLOCK(tdirp);
2132 		}
2133 		vrele(tdirp);
2134 		tdirp = NULL;
2135 	}
2136 	pathbuf_destroy(fromnd.ni_pathbuf);
2137 	fromnd.ni_pathbuf = NULL;
2138 	fromnd.ni_cnd.cn_nameiop = 0;
2139 	fstrans_done(localfs);
2140 	localfs = NULL;
2141 	nfsm_reply(2 * NFSX_WCCDATA(v3));
2142 	if (v3) {
2143 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2144 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2145 	}
2146 	return (0);
2147 
2148 nfsmout:
2149 	if (fdirp)
2150 		vrele(fdirp);
2151 #ifdef notdef
2152 	if (tdirp)
2153 		vrele(tdirp);
2154 #endif
2155 	if (tond.ni_cnd.cn_nameiop) {
2156 		if (tond.ni_pathbuf != NULL) {
2157 			pathbuf_destroy(tond.ni_pathbuf);
2158 			tond.ni_pathbuf = NULL;
2159 		}
2160 	}
2161 	if (localfs) {
2162 		VFS_RENAMELOCK_EXIT(localfs);
2163 		fstrans_done(localfs);
2164 	}
2165 	if (fromnd.ni_cnd.cn_nameiop) {
2166 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2167 		if (fromnd.ni_pathbuf != NULL) {
2168 			pathbuf_destroy(fromnd.ni_pathbuf);
2169 			fromnd.ni_pathbuf = NULL;
2170 		}
2171 		vrele(fromnd.ni_dvp);
2172 		vrele(fvp);
2173 	}
2174 	return (error);
2175 }
2176 
2177 /*
2178  * nfs link service
2179  */
2180 int
2181 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2182 {
2183 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2184 	struct mbuf *nam = nfsd->nd_nam;
2185 	char *dpos = nfsd->nd_dpos;
2186 	kauth_cred_t cred = nfsd->nd_cr;
2187 	struct nameidata nd;
2188 	u_int32_t *tl;
2189 	int32_t t1;
2190 	char *bpos;
2191 	int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2192 	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2193 	char *cp2;
2194 	struct mbuf *mb, *mreq __unused;
2195 	struct vnode *vp, *xp, *dirp = (struct vnode *)0;
2196 	struct vattr dirfor, diraft, at;
2197 	nfsrvfh_t nsfh, dnsfh;
2198 	u_quad_t frev;
2199 
2200 	nfsm_srvmtofh(&nsfh);
2201 	nfsm_srvmtofh(&dnsfh);
2202 	nfsm_srvnamesiz(len);
2203 	error = nfsrv_fhtovp(&nsfh, false, &vp, cred, slp, nam,
2204 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2205 	if (error) {
2206 		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2207 		nfsm_srvpostop_attr(getret, &at);
2208 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2209 		return (0);
2210 	}
2211 	if (vp->v_type == VDIR) {
2212 		error = EPERM;
2213 		goto out1;
2214 	}
2215 	nd.ni_cnd.cn_cred = cred;
2216 	nd.ni_cnd.cn_nameiop = CREATE;
2217 	nd.ni_cnd.cn_flags = LOCKPARENT;
2218 	error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos,
2219 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2220 	if (dirp && v3) {
2221 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2222 	}
2223 	if (error)
2224 		goto out1;
2225 	xp = nd.ni_vp;
2226 	if (xp != NULL) {
2227 		error = EEXIST;
2228 		goto out;
2229 	}
2230 	xp = nd.ni_dvp;
2231 	if (vp->v_mount != xp->v_mount)
2232 		error = EXDEV;
2233 out:
2234 	if (!error) {
2235 		nqsrv_getl(vp, ND_WRITE);
2236 		nqsrv_getl(xp, ND_WRITE);
2237 		error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2238 		if (nd.ni_dvp != nd.ni_vp)
2239 			VOP_UNLOCK(nd.ni_dvp);
2240 		vrele(nd.ni_dvp);
2241 	} else {
2242 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2243 		if (nd.ni_dvp == nd.ni_vp)
2244 			vrele(nd.ni_dvp);
2245 		else
2246 			vput(nd.ni_dvp);
2247 		if (nd.ni_vp)
2248 			vrele(nd.ni_vp);
2249 	}
2250 out1:
2251 	if (v3) {
2252 		vn_lock(vp, LK_SHARED | LK_RETRY);
2253 		getret = VOP_GETATTR(vp, &at, cred);
2254 		VOP_UNLOCK(vp);
2255 	}
2256 	if (dirp) {
2257 		if (v3) {
2258 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2259 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2260 			VOP_UNLOCK(dirp);
2261 		}
2262 		vrele(dirp);
2263 	}
2264 	vrele(vp);
2265 	if (nd.ni_pathbuf != NULL) {
2266 		pathbuf_destroy(nd.ni_pathbuf);
2267 		nd.ni_pathbuf = NULL;
2268 	}
2269 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2270 	if (v3) {
2271 		nfsm_srvpostop_attr(getret, &at);
2272 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2273 		return (0);
2274 	}
2275 	nfsm_srvdone;
2276 }
2277 
2278 /*
2279  * nfs symbolic link service
2280  */
2281 int
2282 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2283 {
2284 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2285 	struct mbuf *nam = nfsd->nd_nam;
2286 	char *dpos = nfsd->nd_dpos;
2287 	kauth_cred_t cred = nfsd->nd_cr;
2288 	struct vattr va, dirfor, diraft;
2289 	struct nameidata nd;
2290 	u_int32_t *tl;
2291 	int32_t t1;
2292 	struct nfsv2_sattr *sp;
2293 	char *bpos, *pathcp = NULL, *cp2;
2294 	struct uio io;
2295 	struct iovec iv;
2296 	int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1, abort = 0;
2297 	uint32_t len, len2;
2298 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2299 	struct mbuf *mb, *mreq __unused;
2300 	struct vnode *dirp = (struct vnode *)0;
2301 	nfsrvfh_t nsfh;
2302 	u_quad_t frev;
2303 
2304 	nd.ni_cnd.cn_nameiop = 0;
2305 	nfsm_srvmtofh(&nsfh);
2306 	nfsm_srvnamesiz(len);
2307 	nd.ni_cnd.cn_cred = cred;
2308 	nd.ni_cnd.cn_nameiop = CREATE;
2309 	nd.ni_cnd.cn_flags = LOCKPARENT;
2310 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2311 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2312 	if (dirp && v3) {
2313 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2314 	}
2315 	if (error)
2316 		goto out;
2317 	abort = 1;
2318 	vattr_null(&va);
2319 	va.va_type = VLNK;
2320 	if (v3) {
2321 		va.va_mode = 0;
2322 		nfsm_srvsattr(&va);
2323 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
2324 		len2 = fxdr_unsigned(uint32_t, *tl);
2325 		if (len2 > PATH_MAX) {
2326 			/* XXX should check _PC_NO_TRUNC */
2327 			error = ENAMETOOLONG;
2328 			goto abortop;
2329 		}
2330 	}
2331 	else {
2332 		/* NFSv2 */
2333 		nfsm_strsiz(len2, NFS_MAXPATHLEN);
2334 	}
2335 	pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2336 	iv.iov_base = pathcp;
2337 	iv.iov_len = len2;
2338 	io.uio_resid = len2;
2339 	io.uio_offset = 0;
2340 	io.uio_iov = &iv;
2341 	io.uio_iovcnt = 1;
2342 	io.uio_rw = UIO_READ;
2343 	UIO_SETUP_SYSSPACE(&io);
2344 	nfsm_mtouio(&io, len2);
2345 	if (!v3) {
2346 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2347 		va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
2348 	}
2349 	*(pathcp + len2) = '\0';
2350 	if (nd.ni_vp) {
2351 		error = EEXIST;
2352 abortop:
2353 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2354 		if (nd.ni_dvp == nd.ni_vp)
2355 			vrele(nd.ni_dvp);
2356 		else
2357 			vput(nd.ni_dvp);
2358 		if (nd.ni_vp)
2359 			vrele(nd.ni_vp);
2360 		goto out;
2361 	}
2362 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2363 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp);
2364 	if (!error) {
2365 	    if (v3) {
2366 		vn_lock(nd.ni_vp, LK_SHARED | LK_RETRY);
2367 		error = nfsrv_composefh(nd.ni_vp, &nsfh, v3);
2368 		if (!error)
2369 		    error = VOP_GETATTR(nd.ni_vp, &va, cred);
2370 		vput(nd.ni_vp);
2371 	    } else {
2372 		vrele(nd.ni_vp);
2373 	    }
2374 	}
2375 	vput(nd.ni_dvp);
2376 out:
2377 	if (pathcp)
2378 		free(pathcp, M_TEMP);
2379 	if (dirp) {
2380 		if (v3) {
2381 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2382 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2383 			VOP_UNLOCK(dirp);
2384 		}
2385 		vrele(dirp);
2386 		dirp = NULL;
2387 	}
2388 	if (nd.ni_pathbuf != NULL) {
2389 		pathbuf_destroy(nd.ni_pathbuf);
2390 		nd.ni_pathbuf = NULL;
2391 	}
2392 	abort = 0;
2393 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) +
2394 	    NFSX_WCCDATA(v3));
2395 	if (v3) {
2396 		if (!error) {
2397 			nfsm_srvpostop_fh(&nsfh);
2398 			nfsm_srvpostop_attr(0, &va);
2399 		}
2400 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2401 	}
2402 	return (0);
2403 nfsmout:
2404 	if (abort) {
2405 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2406 		if (nd.ni_dvp == nd.ni_vp)
2407 			vrele(nd.ni_dvp);
2408 		else
2409 			vput(nd.ni_dvp);
2410 		if (nd.ni_vp)
2411 			vrele(nd.ni_vp);
2412 		if (nd.ni_pathbuf != NULL) {
2413 			pathbuf_destroy(nd.ni_pathbuf);
2414 			nd.ni_pathbuf = NULL;
2415 		}
2416 	}
2417 	if (dirp)
2418 		vrele(dirp);
2419 	if (pathcp)
2420 		free(pathcp, M_TEMP);
2421 	return (error);
2422 }
2423 
2424 /*
2425  * nfs mkdir service
2426  */
2427 int
2428 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2429 {
2430 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2431 	struct mbuf *nam = nfsd->nd_nam;
2432 	char *dpos = nfsd->nd_dpos;
2433 	kauth_cred_t cred = nfsd->nd_cr;
2434 	struct vattr va, dirfor, diraft;
2435 	struct nfs_fattr *fp;
2436 	struct nameidata nd;
2437 	char *cp;
2438 	u_int32_t *tl;
2439 	int32_t t1;
2440 	char *bpos;
2441 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2442 	int abort = 0;
2443 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2444 	char *cp2;
2445 	struct mbuf *mb, *mreq __unused;
2446 	struct vnode *vp, *dirp = (struct vnode *)0;
2447 	nfsrvfh_t nsfh;
2448 	u_quad_t frev;
2449 
2450 	nfsm_srvmtofh(&nsfh);
2451 	nfsm_srvnamesiz(len);
2452 	nd.ni_cnd.cn_cred = cred;
2453 	nd.ni_cnd.cn_nameiop = CREATE;
2454 	nd.ni_cnd.cn_flags = LOCKPARENT;
2455 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2456 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2457 	if (dirp && v3) {
2458 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2459 	}
2460 	if (error) {
2461 		if (nd.ni_pathbuf != NULL) {
2462 			pathbuf_destroy(nd.ni_pathbuf);
2463 			nd.ni_pathbuf = NULL;
2464 		}
2465 		nfsm_reply(NFSX_WCCDATA(v3));
2466 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2467 		if (dirp)
2468 			vrele(dirp);
2469 		return (0);
2470 	}
2471 	abort = 1;
2472 	vattr_null(&va);
2473 	if (v3) {
2474 		va.va_mode = 0;
2475 		nfsm_srvsattr(&va);
2476 	} else {
2477 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2478 		va.va_mode = nfstov_mode(*tl++);
2479 	}
2480 	va.va_type = VDIR;
2481 	vp = nd.ni_vp;
2482 	if (vp != NULL) {
2483 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2484 		if (nd.ni_dvp == vp)
2485 			vrele(nd.ni_dvp);
2486 		else
2487 			vput(nd.ni_dvp);
2488 		vrele(vp);
2489 		error = EEXIST;
2490 		goto out;
2491 	}
2492 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2493 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
2494 	if (!error) {
2495 		vp = nd.ni_vp;
2496 		vn_lock(vp, LK_SHARED | LK_RETRY);
2497 		error = nfsrv_composefh(vp, &nsfh, v3);
2498 		if (!error)
2499 			error = VOP_GETATTR(vp, &va, cred);
2500 		vput(vp);
2501 	}
2502 	vput(nd.ni_dvp);
2503 out:
2504 	if (dirp) {
2505 		if (v3) {
2506 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2507 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2508 			VOP_UNLOCK(dirp);
2509 		}
2510 		vrele(dirp);
2511 		dirp = NULL;
2512 	}
2513 	if (nd.ni_pathbuf != NULL) {
2514 		pathbuf_destroy(nd.ni_pathbuf);
2515 		nd.ni_pathbuf = NULL;
2516 	}
2517 	abort = 0;
2518 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) +
2519 	    NFSX_WCCDATA(v3));
2520 	if (v3) {
2521 		if (!error) {
2522 			nfsm_srvpostop_fh(&nsfh);
2523 			nfsm_srvpostop_attr(0, &va);
2524 		}
2525 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2526 	} else {
2527 		nfsm_srvfhtom(&nsfh, v3);
2528 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2529 		nfsm_srvfillattr(&va, fp);
2530 	}
2531 	return (0);
2532 nfsmout:
2533 	if (abort) {
2534 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2535 		if (nd.ni_dvp == nd.ni_vp)
2536 			vrele(nd.ni_dvp);
2537 		else
2538 			vput(nd.ni_dvp);
2539 		if (nd.ni_vp)
2540 			vrele(nd.ni_vp);
2541 		if (nd.ni_pathbuf != NULL) {
2542 			pathbuf_destroy(nd.ni_pathbuf);
2543 			nd.ni_pathbuf = NULL;
2544 		}
2545 	}
2546 	if (dirp)
2547 		vrele(dirp);
2548 	return (error);
2549 }
2550 
2551 /*
2552  * nfs rmdir service
2553  */
2554 int
2555 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2556 {
2557 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2558 	struct mbuf *nam = nfsd->nd_nam;
2559 	char *dpos = nfsd->nd_dpos;
2560 	kauth_cred_t cred = nfsd->nd_cr;
2561 	u_int32_t *tl;
2562 	int32_t t1;
2563 	char *bpos;
2564 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2565 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2566 	char *cp2;
2567 	struct mbuf *mb, *mreq __unused;
2568 	struct vnode *vp, *dirp = (struct vnode *)0;
2569 	struct vattr dirfor, diraft;
2570 	nfsrvfh_t nsfh;
2571 	struct nameidata nd;
2572 	u_quad_t frev;
2573 
2574 	nfsm_srvmtofh(&nsfh);
2575 	nfsm_srvnamesiz(len);
2576 	nd.ni_cnd.cn_cred = cred;
2577 	nd.ni_cnd.cn_nameiop = DELETE;
2578 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2579 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2580 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), false);
2581 	if (dirp && v3) {
2582 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred);
2583 	}
2584 	if (error) {
2585 		if (nd.ni_pathbuf != NULL) {
2586 			pathbuf_destroy(nd.ni_pathbuf);
2587 			nd.ni_pathbuf = NULL;
2588 		}
2589 		nfsm_reply(NFSX_WCCDATA(v3));
2590 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2591 		if (dirp)
2592 			vrele(dirp);
2593 		return (0);
2594 	}
2595 	vp = nd.ni_vp;
2596 	if (vp->v_type != VDIR) {
2597 		error = ENOTDIR;
2598 		goto out;
2599 	}
2600 	/*
2601 	 * No rmdir "." please.
2602 	 */
2603 	if (nd.ni_dvp == vp) {
2604 		error = EINVAL;
2605 		goto out;
2606 	}
2607 	/*
2608 	 * The root of a mounted filesystem cannot be deleted.
2609 	 */
2610 	if (vp->v_vflag & VV_ROOT)
2611 		error = EBUSY;
2612 out:
2613 	if (!error) {
2614 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
2615 		nqsrv_getl(vp, ND_WRITE);
2616 		error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2617 		vput(nd.ni_dvp);
2618 	} else {
2619 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2620 		if (nd.ni_dvp == nd.ni_vp)
2621 			vrele(nd.ni_dvp);
2622 		else
2623 			vput(nd.ni_dvp);
2624 		vput(vp);
2625 	}
2626 	if (nd.ni_pathbuf != NULL) {
2627 		pathbuf_destroy(nd.ni_pathbuf);
2628 		nd.ni_pathbuf = NULL;
2629 	}
2630 	if (dirp) {
2631 		if (v3) {
2632 			vn_lock(dirp, LK_SHARED | LK_RETRY);
2633 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2634 			VOP_UNLOCK(dirp);
2635 		}
2636 		vrele(dirp);
2637 	}
2638 	nfsm_reply(NFSX_WCCDATA(v3));
2639 	if (v3) {
2640 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2641 		return (0);
2642 	}
2643 	nfsm_srvdone;
2644 }
2645 
2646 /*
2647  * nfs readdir service
2648  * - mallocs what it thinks is enough to read
2649  *	count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR
2650  * - calls VOP_READDIR()
2651  * - loops around building the reply
2652  *	if the output generated exceeds count break out of loop
2653  *	The nfsm_clget macro is used here so that the reply will be packed
2654  *	tightly in mbuf clusters.
2655  * - it only knows that it has encountered eof when the VOP_READDIR()
2656  *	reads nothing
2657  * - as such one readdir rpc will return eof false although you are there
2658  *	and then the next will return eof
2659  * - it trims out records with d_fileno == 0
2660  *	this doesn't matter for Unix clients, but they might confuse clients
2661  *	for other os'.
2662  * - it trims out records with d_type == DT_WHT
2663  *	these cannot be seen through NFS (unless we extend the protocol)
2664  * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2665  *	than requested, but this may not apply to all filesystems. For
2666  *	example, client NFS does not { although it is never remote mounted
2667  *	anyhow }
2668  *     The alternate call nfsrv_readdirplus() does lookups as well.
2669  * PS: The NFS protocol spec. does not clarify what the "count" byte
2670  *	argument is a count of.. just name strings and file id's or the
2671  *	entire reply rpc or ...
2672  *	I tried just file name and id sizes and it confused the Sun client,
2673  *	so I am using the full rpc size now. The "paranoia.." comment refers
2674  *	to including the status longwords that are not a part of the dir.
2675  *	"entry" structures, but are in the rpc.
2676  */
2677 
2678 #define	NFS_SRVDIRBLKSIZ	1024
2679 
2680 struct flrep {
2681 	nfsuint64 fl_off;
2682 	u_int32_t fl_postopok;
2683 	struct nfs_fattr fl_fattr; /* XXX: must be of fattr3 size */
2684 	u_int32_t fl_fhok;
2685 	u_int32_t fl_fhsize;
2686 	/* handle comes here, filled in dynamically */
2687 };
2688 
2689 int
2690 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2691 {
2692 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2693 	struct mbuf *nam = nfsd->nd_nam;
2694 	char *dpos = nfsd->nd_dpos;
2695 	kauth_cred_t cred = nfsd->nd_cr;
2696 	char *bp, *be;
2697 	struct mbuf *mp;
2698 	struct dirent *dp;
2699 	char *cp;
2700 	u_int32_t *tl;
2701 	int32_t t1;
2702 	char *bpos;
2703 	struct mbuf *mb, *mreq __unused, *mp2;
2704 	char *cpos, *cend, *cp2, *rbuf;
2705 	struct vnode *vp;
2706 	struct vattr at;
2707 	nfsrvfh_t nsfh;
2708 	struct uio io;
2709 	struct iovec iv;
2710 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2711 	int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies;
2712 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2713 	u_quad_t frev, off, toff;
2714 #ifdef NFS3_STRICTVERF
2715 	u_quad_t verf;
2716 #endif
2717 	off_t *cookies = NULL, *cookiep;
2718 	nfsuint64 jar;
2719 
2720 	nfsm_srvmtofh(&nsfh);
2721 	if (v3) {
2722 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2723 		toff = fxdr_hyper(tl);
2724 		tl += 2;
2725 #ifdef NFS3_STRICTVERF
2726 		verf = fxdr_hyper(tl);
2727 #endif
2728 		tl += 2;
2729 	} else {
2730 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2731 		toff = fxdr_unsigned(u_quad_t, *tl++);
2732 	}
2733 	off = toff;
2734 	cnt = fxdr_unsigned(int, *tl);
2735 	siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1));
2736 	xfer = NFS_SRVMAXDATA(nfsd);
2737 	if (siz > xfer)
2738 		siz = xfer;
2739 	fullsiz = siz;
2740 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
2741 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
2742 	if (!error && vp->v_type != VDIR) {
2743 		error = ENOTDIR;
2744 		vput(vp);
2745 	}
2746 	if (error) {
2747 		nfsm_reply(NFSX_UNSIGNED);
2748 		nfsm_srvpostop_attr(getret, &at);
2749 		return (0);
2750 	}
2751 	nqsrv_getl(vp, ND_READ);
2752 	if (v3) {
2753 		error = getret = VOP_GETATTR(vp, &at, cred);
2754 #ifdef NFS3_STRICTVERF
2755 		/*
2756 		 * XXX This check is too strict for Solaris 2.5 clients.
2757 		 */
2758 		if (!error && toff && verf != at.va_filerev)
2759 			error = NFSERR_BAD_COOKIE;
2760 #endif
2761 	}
2762 	if (!error)
2763 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0);
2764 	if (error) {
2765 		vput(vp);
2766 		nfsm_reply(NFSX_POSTOPATTR(v3));
2767 		nfsm_srvpostop_attr(getret, &at);
2768 		return (0);
2769 	}
2770 	VOP_UNLOCK(vp);
2771 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2772 again:
2773 	iv.iov_base = rbuf;
2774 	iv.iov_len = fullsiz;
2775 	io.uio_iov = &iv;
2776 	io.uio_iovcnt = 1;
2777 	io.uio_offset = (off_t)off;
2778 	io.uio_resid = fullsiz;
2779 	io.uio_rw = UIO_READ;
2780 	UIO_SETUP_SYSSPACE(&io);
2781 	eofflag = 0;
2782 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2783 
2784 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2785 
2786 	off = (off_t)io.uio_offset;
2787 	if (!cookies && !error)
2788 		error = NFSERR_PERM;
2789 	if (v3) {
2790 		getret = VOP_GETATTR(vp, &at, cred);
2791 		if (!error)
2792 			error = getret;
2793 	}
2794 
2795 	VOP_UNLOCK(vp);
2796 	if (error) {
2797 		vrele(vp);
2798 		free((void *)rbuf, M_TEMP);
2799 		if (cookies)
2800 			free((void *)cookies, M_TEMP);
2801 		nfsm_reply(NFSX_POSTOPATTR(v3));
2802 		nfsm_srvpostop_attr(getret, &at);
2803 		return (0);
2804 	}
2805 	if (io.uio_resid) {
2806 		siz -= io.uio_resid;
2807 
2808 		/*
2809 		 * If nothing read, return eof
2810 		 * rpc reply
2811 		 */
2812 		if (siz == 0) {
2813 			vrele(vp);
2814 			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2815 				2 * NFSX_UNSIGNED);
2816 			if (v3) {
2817 				nfsm_srvpostop_attr(getret, &at);
2818 				nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2819 				txdr_hyper(at.va_filerev, tl);
2820 				tl += 2;
2821 			} else
2822 				nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2823 			*tl++ = nfs_false;
2824 			*tl = nfs_true;
2825 			free((void *)rbuf, M_TEMP);
2826 			free((void *)cookies, M_TEMP);
2827 			return (0);
2828 		}
2829 	}
2830 
2831 	/*
2832 	 * Check for degenerate cases of nothing useful read.
2833 	 * If so go try again
2834 	 */
2835 	cpos = rbuf;
2836 	cend = rbuf + siz;
2837 	dp = (struct dirent *)cpos;
2838 	cookiep = cookies;
2839 
2840 	while (cpos < cend && ncookies > 0 &&
2841 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2842 		cpos += dp->d_reclen;
2843 		dp = (struct dirent *)cpos;
2844 		cookiep++;
2845 		ncookies--;
2846 	}
2847 	if (cpos >= cend || ncookies == 0) {
2848 		toff = off;
2849 		siz = fullsiz;
2850 		free(cookies, M_TEMP);
2851 		cookies = NULL;
2852 		goto again;
2853 	}
2854 
2855 	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2856 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2857 	if (v3) {
2858 		nfsm_srvpostop_attr(getret, &at);
2859 		nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2860 		txdr_hyper(at.va_filerev, tl);
2861 	}
2862 	mp = mp2 = mb;
2863 	bp = bpos;
2864 	be = bp + M_TRAILINGSPACE(mp);
2865 
2866 	/* Loop through the records and build reply */
2867 	while (cpos < cend && ncookies > 0) {
2868 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2869 			nlen = dp->d_namlen;
2870 			rem = nfsm_rndup(nlen)-nlen;
2871 			len += (4 * NFSX_UNSIGNED + nlen + rem);
2872 			if (v3)
2873 				len += 2 * NFSX_UNSIGNED;
2874 			if (len > cnt) {
2875 				eofflag = 0;
2876 				break;
2877 			}
2878 			/*
2879 			 * Build the directory record xdr from
2880 			 * the dirent entry.
2881 			 */
2882 			nfsm_clget;
2883 			*tl = nfs_true;
2884 			bp += NFSX_UNSIGNED;
2885 			if (v3) {
2886 				nfsm_clget;
2887 				*tl = txdr_unsigned(dp->d_fileno >> 32);
2888 				bp += NFSX_UNSIGNED;
2889 			}
2890 			nfsm_clget;
2891 			*tl = txdr_unsigned(dp->d_fileno);
2892 			bp += NFSX_UNSIGNED;
2893 			nfsm_clget;
2894 			*tl = txdr_unsigned(nlen);
2895 			bp += NFSX_UNSIGNED;
2896 
2897 			/* And loop around copying the name */
2898 			xfer = nlen;
2899 			cp = dp->d_name;
2900 			while (xfer > 0) {
2901 				nfsm_clget;
2902 				if ((bp+xfer) > be)
2903 					tsiz = be-bp;
2904 				else
2905 					tsiz = xfer;
2906 				memcpy(bp, cp, tsiz);
2907 				bp += tsiz;
2908 				xfer -= tsiz;
2909 				if (xfer > 0)
2910 					cp += tsiz;
2911 			}
2912 			/* And null pad to an int32_t boundary */
2913 			for (i = 0; i < rem; i++)
2914 				*bp++ = '\0';
2915 			nfsm_clget;
2916 
2917 			/* Finish off the record */
2918 			txdr_hyper(*cookiep, &jar);
2919 			if (v3) {
2920 				*tl = jar.nfsuquad[0];
2921 				bp += NFSX_UNSIGNED;
2922 				nfsm_clget;
2923 			}
2924 			*tl = jar.nfsuquad[1];
2925 			bp += NFSX_UNSIGNED;
2926 		}
2927 		cpos += dp->d_reclen;
2928 		dp = (struct dirent *)cpos;
2929 		cookiep++;
2930 		ncookies--;
2931 	}
2932 	vrele(vp);
2933 	nfsm_clget;
2934 	*tl = nfs_false;
2935 	bp += NFSX_UNSIGNED;
2936 	nfsm_clget;
2937 	if (eofflag)
2938 		*tl = nfs_true;
2939 	else
2940 		*tl = nfs_false;
2941 	bp += NFSX_UNSIGNED;
2942 	if (mp != mb) {
2943 		if (bp < be)
2944 			mp->m_len = bp - mtod(mp, char *);
2945 	} else
2946 		mp->m_len += bp - bpos;
2947 	free((void *)rbuf, M_TEMP);
2948 	free((void *)cookies, M_TEMP);
2949 	nfsm_srvdone;
2950 }
2951 
2952 int
2953 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
2954 {
2955 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2956 	struct mbuf *nam = nfsd->nd_nam;
2957 	char *dpos = nfsd->nd_dpos;
2958 	kauth_cred_t cred = nfsd->nd_cr;
2959 	char *bp, *be;
2960 	struct mbuf *mp;
2961 	struct dirent *dp;
2962 	char *cp;
2963 	u_int32_t *tl;
2964 	int32_t t1;
2965 	char *bpos;
2966 	struct mbuf *mb, *mreq __unused, *mp2;
2967 	char *cpos, *cend, *cp2, *rbuf;
2968 	struct vnode *vp, *nvp;
2969 	struct flrep fl;
2970 	nfsrvfh_t nsfh;
2971 	struct uio io;
2972 	struct iovec iv;
2973 	struct vattr va, at, *vap = &va;
2974 	struct nfs_fattr *fp;
2975 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2976 	int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies;
2977 	u_quad_t frev, off, toff;
2978 #ifdef NFS3_STRICTVERF
2979 	u_quad_t verf;
2980 #endif
2981 	off_t *cookies = NULL, *cookiep;
2982 
2983 	nfsm_srvmtofh(&nsfh);
2984 	nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2985 	toff = fxdr_hyper(tl);
2986 	tl += 2;
2987 #ifdef NFS3_STRICTVERF
2988 	verf = fxdr_hyper(tl);
2989 #endif
2990 	tl += 2;
2991 	siz = fxdr_unsigned(int, *tl++);
2992 	cnt = fxdr_unsigned(int, *tl);
2993 	off = toff;
2994 	siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1));
2995 	xfer = NFS_SRVMAXDATA(nfsd);
2996 	if (siz > xfer)
2997 		siz = xfer;
2998 	fullsiz = siz;
2999 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3000 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3001 	if (!error && vp->v_type != VDIR) {
3002 		error = ENOTDIR;
3003 		vput(vp);
3004 	}
3005 	if (error) {
3006 		nfsm_reply(NFSX_UNSIGNED);
3007 		nfsm_srvpostop_attr(getret, &at);
3008 		return (0);
3009 	}
3010 	error = getret = VOP_GETATTR(vp, &at, cred);
3011 #ifdef NFS3_STRICTVERF
3012 	/*
3013 	 * XXX This check is too strict for Solaris 2.5 clients.
3014 	 */
3015 	if (!error && toff && verf != at.va_filerev)
3016 		error = NFSERR_BAD_COOKIE;
3017 #endif
3018 	if (!error) {
3019 		nqsrv_getl(vp, ND_READ);
3020 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0);
3021 	}
3022 	if (error) {
3023 		vput(vp);
3024 		nfsm_reply(NFSX_V3POSTOPATTR);
3025 		nfsm_srvpostop_attr(getret, &at);
3026 		return (0);
3027 	}
3028 	VOP_UNLOCK(vp);
3029 
3030 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
3031 again:
3032 	iv.iov_base = rbuf;
3033 	iv.iov_len = fullsiz;
3034 	io.uio_iov = &iv;
3035 	io.uio_iovcnt = 1;
3036 	io.uio_offset = (off_t)off;
3037 	io.uio_resid = fullsiz;
3038 	io.uio_rw = UIO_READ;
3039 	UIO_SETUP_SYSSPACE(&io);
3040 	eofflag = 0;
3041 
3042 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3043 
3044 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
3045 
3046 	off = (u_quad_t)io.uio_offset;
3047 	getret = VOP_GETATTR(vp, &at, cred);
3048 
3049 	VOP_UNLOCK(vp);
3050 
3051 	/*
3052 	 * If the VGET operation doesn't work for this filesystem,
3053 	 * we can't support readdirplus. Returning NOTSUPP should
3054 	 * make clients fall back to plain readdir.
3055 	 * There's no need to check for VPTOFH as well, we wouldn't
3056 	 * even be here otherwise.
3057 	 */
3058 	if (!getret) {
3059 		if ((getret = VFS_VGET(vp->v_mount, at.va_fileid,
3060 		    LK_EXCLUSIVE, &nvp)))
3061 			getret = (getret == EOPNOTSUPP) ?
3062 				NFSERR_NOTSUPP : NFSERR_IO;
3063 		else
3064 			vput(nvp);
3065 	}
3066 
3067 	if (!cookies && !error)
3068 		error = NFSERR_PERM;
3069 	if (!error)
3070 		error = getret;
3071 	if (error) {
3072 		vrele(vp);
3073 		if (cookies)
3074 			free((void *)cookies, M_TEMP);
3075 		free((void *)rbuf, M_TEMP);
3076 		nfsm_reply(NFSX_V3POSTOPATTR);
3077 		nfsm_srvpostop_attr(getret, &at);
3078 		return (0);
3079 	}
3080 	if (io.uio_resid) {
3081 		siz -= io.uio_resid;
3082 
3083 		/*
3084 		 * If nothing read, return eof
3085 		 * rpc reply
3086 		 */
3087 		if (siz == 0) {
3088 			vrele(vp);
3089 			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3090 				2 * NFSX_UNSIGNED);
3091 			nfsm_srvpostop_attr(getret, &at);
3092 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3093 			txdr_hyper(at.va_filerev, tl);
3094 			tl += 2;
3095 			*tl++ = nfs_false;
3096 			*tl = nfs_true;
3097 			free((void *)cookies, M_TEMP);
3098 			free((void *)rbuf, M_TEMP);
3099 			return (0);
3100 		}
3101 	}
3102 
3103 	/*
3104 	 * Check for degenerate cases of nothing useful read.
3105 	 * If so go try again
3106 	 */
3107 	cpos = rbuf;
3108 	cend = rbuf + siz;
3109 	dp = (struct dirent *)cpos;
3110 	cookiep = cookies;
3111 
3112 	while (cpos < cend && ncookies > 0 &&
3113 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
3114 		cpos += dp->d_reclen;
3115 		dp = (struct dirent *)cpos;
3116 		cookiep++;
3117 		ncookies--;
3118 	}
3119 	if (cpos >= cend || ncookies == 0) {
3120 		toff = off;
3121 		siz = fullsiz;
3122 		free(cookies, M_TEMP);
3123 		cookies = NULL;
3124 		goto again;
3125 	}
3126 
3127 	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
3128 	nfsm_reply(cnt);
3129 	nfsm_srvpostop_attr(getret, &at);
3130 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3131 	txdr_hyper(at.va_filerev, tl);
3132 	mp = mp2 = mb;
3133 	bp = bpos;
3134 	be = bp + M_TRAILINGSPACE(mp);
3135 
3136 	/* Loop through the records and build reply */
3137 	while (cpos < cend && ncookies > 0) {
3138 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3139 			nfsrvfh_t nnsfh;
3140 
3141 			nlen = dp->d_namlen;
3142 			rem = nfsm_rndup(nlen)-nlen;
3143 
3144 			/*
3145 			 * For readdir_and_lookup get the vnode using
3146 			 * the file number.
3147 			 */
3148 			if (VFS_VGET(vp->v_mount, dp->d_fileno, LK_EXCLUSIVE,
3149 			    &nvp))
3150 				goto invalid;
3151 			if (nfsrv_composefh(nvp, &nnsfh, true)) {
3152 				vput(nvp);
3153 				goto invalid;
3154 			}
3155 			if (VOP_GETATTR(nvp, vap, cred)) {
3156 				vput(nvp);
3157 				goto invalid;
3158 			}
3159 			vput(nvp);
3160 
3161 			/*
3162 			 * If either the dircount or maxcount will be
3163 			 * exceeded, get out now. Both of these lengths
3164 			 * are calculated conservatively, including all
3165 			 * XDR overheads.
3166 			 */
3167 			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3168 				NFSX_V3POSTOPATTR);
3169 			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3170 			if (len > cnt || dirlen > fullsiz) {
3171 				eofflag = 0;
3172 				break;
3173 			}
3174 
3175 			/*
3176 			 * Build the directory record xdr from
3177 			 * the dirent entry.
3178 			 */
3179 			fp = (struct nfs_fattr *)&fl.fl_fattr;
3180 			nfsm_srvfillattr(vap, fp);
3181 			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3182 			fl.fl_fhok = nfs_true;
3183 			fl.fl_postopok = nfs_true;
3184 			txdr_hyper(*cookiep, fl.fl_off.nfsuquad);
3185 
3186 			nfsm_clget;
3187 			*tl = nfs_true;
3188 			bp += NFSX_UNSIGNED;
3189 			nfsm_clget;
3190 			*tl = txdr_unsigned(dp->d_fileno >> 32);
3191 			bp += NFSX_UNSIGNED;
3192 			nfsm_clget;
3193 			*tl = txdr_unsigned(dp->d_fileno);
3194 			bp += NFSX_UNSIGNED;
3195 			nfsm_clget;
3196 			*tl = txdr_unsigned(nlen);
3197 			bp += NFSX_UNSIGNED;
3198 
3199 			/* And loop around copying the name */
3200 			xfer = nlen;
3201 			cp = dp->d_name;
3202 			while (xfer > 0) {
3203 				nfsm_clget;
3204 				if ((bp + xfer) > be)
3205 					tsiz = be - bp;
3206 				else
3207 					tsiz = xfer;
3208 				memcpy(bp, cp, tsiz);
3209 				bp += tsiz;
3210 				xfer -= tsiz;
3211 				if (xfer > 0)
3212 					cp += tsiz;
3213 			}
3214 			/* And null pad to an int32_t boundary */
3215 			for (i = 0; i < rem; i++)
3216 				*bp++ = '\0';
3217 
3218 			/*
3219 			 * Now copy the flrep structure out.
3220 			 */
3221 			xfer = sizeof(struct flrep);
3222 			cp = (void *)&fl;
3223 			while (xfer > 0) {
3224 				nfsm_clget;
3225 				if ((bp + xfer) > be)
3226 					tsiz = be - bp;
3227 				else
3228 					tsiz = xfer;
3229 				memcpy(bp, cp, tsiz);
3230 				bp += tsiz;
3231 				xfer -= tsiz;
3232 				if (xfer > 0)
3233 					cp += tsiz;
3234 			}
3235 
3236 			/*
3237 			 * ... and filehandle.
3238 			 */
3239 			xfer = NFSRVFH_SIZE(&nnsfh);
3240 			cp = NFSRVFH_DATA(&nnsfh);
3241 			while (xfer > 0) {
3242 				nfsm_clget;
3243 				if ((bp + xfer) > be)
3244 					tsiz = be - bp;
3245 				else
3246 					tsiz = xfer;
3247 				memcpy(bp, cp, tsiz);
3248 				bp += tsiz;
3249 				xfer -= tsiz;
3250 				if (xfer > 0)
3251 					cp += tsiz;
3252 			}
3253 		}
3254 invalid:
3255 		cpos += dp->d_reclen;
3256 		dp = (struct dirent *)cpos;
3257 		cookiep++;
3258 		ncookies--;
3259 	}
3260 	vrele(vp);
3261 	nfsm_clget;
3262 	*tl = nfs_false;
3263 	bp += NFSX_UNSIGNED;
3264 	nfsm_clget;
3265 	if (eofflag)
3266 		*tl = nfs_true;
3267 	else
3268 		*tl = nfs_false;
3269 	bp += NFSX_UNSIGNED;
3270 	if (mp != mb) {
3271 		if (bp < be)
3272 			mp->m_len = bp - mtod(mp, char *);
3273 	} else
3274 		mp->m_len += bp - bpos;
3275 	free((void *)cookies, M_TEMP);
3276 	free((void *)rbuf, M_TEMP);
3277 	nfsm_srvdone;
3278 }
3279 
3280 /*
3281  * nfs commit service
3282  */
3283 int
3284 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3285 {
3286 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3287 	struct mbuf *nam = nfsd->nd_nam;
3288 	char *dpos = nfsd->nd_dpos;
3289 	kauth_cred_t cred = nfsd->nd_cr;
3290 	struct vattr bfor, aft;
3291 	struct vnode *vp;
3292 	nfsrvfh_t nsfh;
3293 	u_int32_t *tl;
3294 	int32_t t1;
3295 	char *bpos;
3296 	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0;
3297 	uint32_t cnt;
3298 	char *cp2;
3299 	struct mbuf *mb, *mreq __unused;
3300 	u_quad_t frev, off, end;
3301 
3302 	nfsm_srvmtofh(&nsfh);
3303 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3304 
3305 	off = fxdr_hyper(tl);
3306 	tl += 2;
3307 	cnt = fxdr_unsigned(uint32_t, *tl);
3308 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3309 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3310 	if (error) {
3311 		nfsm_reply(2 * NFSX_UNSIGNED);
3312 		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3313 		return (0);
3314 	}
3315 	for_ret = VOP_GETATTR(vp, &bfor, cred);
3316 	end = (cnt > 0) ? off + cnt : vp->v_size;
3317 	if (end < off || end > vp->v_size)
3318 		end = vp->v_size;
3319 	if (off < vp->v_size)
3320 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end);
3321 	/* else error == 0, from nfsrv_fhtovp() */
3322 	aft_ret = VOP_GETATTR(vp, &aft, cred);
3323 	vput(vp);
3324 	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3325 	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3326 	if (!error) {
3327 		nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3328 		struct timeval btv;
3329 		getmicroboottime(&btv);
3330 		*tl++ = txdr_unsigned(btv.tv_sec);
3331 		*tl = txdr_unsigned(btv.tv_usec);
3332 	} else {
3333 		return (0);
3334 	}
3335 	nfsm_srvdone;
3336 }
3337 
3338 /*
3339  * nfs statfs service
3340  */
3341 int
3342 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3343 {
3344 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3345 	struct mbuf *nam = nfsd->nd_nam;
3346 	char *dpos = nfsd->nd_dpos;
3347 	kauth_cred_t cred = nfsd->nd_cr;
3348 	struct statvfs *sf = NULL;
3349 	struct nfs_statfs *sfp;
3350 	u_int32_t *tl;
3351 	int32_t t1;
3352 	char *bpos;
3353 	int error = 0, rdonly, cache = 0, getret = 1;
3354 	int v3 = (nfsd->nd_flag & ND_NFSV3);
3355 	char *cp2;
3356 	struct mbuf *mb, *mreq __unused;
3357 	struct vnode *vp;
3358 	struct vattr at;
3359 	nfsrvfh_t nsfh;
3360 	u_quad_t frev, tval;
3361 
3362 	nfsm_srvmtofh(&nsfh);
3363 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3364 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3365 	if (error) {
3366 		nfsm_reply(NFSX_UNSIGNED);
3367 		nfsm_srvpostop_attr(getret, &at);
3368 		return (0);
3369 	}
3370 	sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK);
3371 	error = VFS_STATVFS(vp->v_mount, sf);
3372 	getret = VOP_GETATTR(vp, &at, cred);
3373 	vput(vp);
3374 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3375 	if (v3)
3376 		nfsm_srvpostop_attr(getret, &at);
3377 	if (error) {
3378 		free(sf, M_TEMP);
3379 		return (0);
3380 	}
3381 	nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3382 	if (v3) {
3383 		tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize);
3384 		txdr_hyper(tval, &sfp->sf_tbytes);
3385 		tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize);
3386 		txdr_hyper(tval, &sfp->sf_fbytes);
3387 		tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize);
3388 		txdr_hyper(tval, &sfp->sf_abytes);
3389 		tval = (u_quad_t)sf->f_files;
3390 		txdr_hyper(tval, &sfp->sf_tfiles);
3391 		tval = (u_quad_t)sf->f_ffree;
3392 		txdr_hyper(tval, &sfp->sf_ffiles);
3393 		txdr_hyper(tval, &sfp->sf_afiles);
3394 		sfp->sf_invarsec = 0;
3395 	} else {
3396 		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3397 		sfp->sf_bsize = txdr_unsigned(sf->f_frsize);
3398 		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3399 		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3400 		sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3401 	}
3402 nfsmout:
3403 	if (sf)
3404 	    free(sf, M_TEMP);
3405 	return error;
3406 }
3407 
3408 /*
3409  * nfs fsinfo service
3410  */
3411 int
3412 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3413 {
3414 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3415 	struct mbuf *nam = nfsd->nd_nam;
3416 	char *dpos = nfsd->nd_dpos;
3417 	kauth_cred_t cred = nfsd->nd_cr;
3418 	u_int32_t *tl;
3419 	struct nfsv3_fsinfo *sip;
3420 	int32_t t1;
3421 	char *bpos;
3422 	int error = 0, rdonly, cache = 0, getret = 1;
3423 	uint32_t maxdata;
3424 	char *cp2;
3425 	struct mbuf *mb, *mreq __unused;
3426 	struct vnode *vp;
3427 	struct vattr at;
3428 	nfsrvfh_t nsfh;
3429 	u_quad_t frev, maxfsize;
3430 	struct statvfs *sb;
3431 
3432 	nfsm_srvmtofh(&nsfh);
3433 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3434 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3435 	if (error) {
3436 		nfsm_reply(NFSX_UNSIGNED);
3437 		nfsm_srvpostop_attr(getret, &at);
3438 		return (0);
3439 	}
3440 
3441 	/* XXX Try to make a guess on the max file size. */
3442 	sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK);
3443 	VFS_STATVFS(vp->v_mount, sb);
3444 	maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1;
3445 	free(sb, M_TEMP);
3446 
3447 	getret = VOP_GETATTR(vp, &at, cred);
3448 	vput(vp);
3449 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3450 	nfsm_srvpostop_attr(getret, &at);
3451 	nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3452 
3453 	/*
3454 	 * XXX
3455 	 * There should be file system VFS OP(s) to get this information.
3456 	 * For now, assume ufs.
3457 	 */
3458 	if (slp->ns_so->so_type == SOCK_DGRAM)
3459 		maxdata = NFS_MAXDGRAMDATA;
3460 	else
3461 		maxdata = NFS_MAXDATA;
3462 	sip->fs_rtmax = txdr_unsigned(maxdata);
3463 	sip->fs_rtpref = txdr_unsigned(maxdata);
3464 	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3465 	sip->fs_wtmax = txdr_unsigned(maxdata);
3466 	sip->fs_wtpref = txdr_unsigned(maxdata);
3467 	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3468 	sip->fs_dtpref = txdr_unsigned(maxdata);
3469 	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3470 	sip->fs_timedelta.nfsv3_sec = 0;
3471 	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3472 	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3473 		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3474 		NFSV3FSINFO_CANSETTIME);
3475 	nfsm_srvdone;
3476 }
3477 
3478 /*
3479  * nfs pathconf service
3480  */
3481 int
3482 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp, struct lwp *lwp, struct mbuf **mrq)
3483 {
3484 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3485 	struct mbuf *nam = nfsd->nd_nam;
3486 	char *dpos = nfsd->nd_dpos;
3487 	kauth_cred_t cred = nfsd->nd_cr;
3488 	u_int32_t *tl;
3489 	struct nfsv3_pathconf *pc;
3490 	int32_t t1;
3491 	char *bpos;
3492 	int error = 0, rdonly, cache = 0, getret = 1;
3493 	register_t linkmax, namemax, chownres, notrunc;
3494 	char *cp2;
3495 	struct mbuf *mb, *mreq __unused;
3496 	struct vnode *vp;
3497 	struct vattr at;
3498 	nfsrvfh_t nsfh;
3499 	u_quad_t frev;
3500 
3501 	nfsm_srvmtofh(&nsfh);
3502 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3503 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), false);
3504 	if (error) {
3505 		nfsm_reply(NFSX_UNSIGNED);
3506 		nfsm_srvpostop_attr(getret, &at);
3507 		return (0);
3508 	}
3509 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3510 	if (!error)
3511 		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3512 	if (!error)
3513 		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3514 	if (!error)
3515 		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3516 	getret = VOP_GETATTR(vp, &at, cred);
3517 	vput(vp);
3518 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3519 	nfsm_srvpostop_attr(getret, &at);
3520 	if (error)
3521 		return (0);
3522 	nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3523 
3524 	pc->pc_linkmax = txdr_unsigned(linkmax);
3525 	pc->pc_namemax = txdr_unsigned(namemax);
3526 	pc->pc_notrunc = txdr_unsigned(notrunc);
3527 	pc->pc_chownrestricted = txdr_unsigned(chownres);
3528 
3529 	/*
3530 	 * These should probably be supported by VOP_PATHCONF(), but
3531 	 * until msdosfs is exportable (why would you want to?), the
3532 	 * Unix defaults should be ok.
3533 	 */
3534 	pc->pc_caseinsensitive = nfs_false;
3535 	pc->pc_casepreserving = nfs_true;
3536 	nfsm_srvdone;
3537 }
3538 
3539 /*
3540  * Null operation, used by clients to ping server
3541  */
3542 /* ARGSUSED */
3543 int
3544 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3545     struct lwp *lwp, struct mbuf **mrq)
3546 {
3547 	struct mbuf *mrep = nfsd->nd_mrep;
3548 	char *bpos;
3549 	int error = NFSERR_RETVOID, cache = 0;
3550 	struct mbuf *mb, *mreq __unused;
3551 	u_quad_t frev;
3552 
3553 	nfsm_reply(0);
3554 nfsmout:
3555 	return (0);
3556 }
3557 
3558 /*
3559  * No operation, used for obsolete procedures
3560  */
3561 /* ARGSUSED */
3562 int
3563 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3564     struct lwp *lwp, struct mbuf **mrq)
3565 {
3566 	struct mbuf *mrep = nfsd->nd_mrep;
3567 	char *bpos;
3568 	int error, cache = 0;
3569 	struct mbuf *mb, *mreq __unused;
3570 	u_quad_t frev;
3571 
3572 	if (nfsd->nd_repstat)
3573 		error = nfsd->nd_repstat;
3574 	else
3575 		error = EPROCUNAVAIL;
3576 	nfsm_reply(0);
3577 nfsmout:
3578 	return (0);
3579 }
3580 
3581 /*
3582  * Perform access checking for vnodes obtained from file handles that would
3583  * refer to files already opened by a Unix client. You cannot just use
3584  * vn_writechk() and VOP_ACCESS() for two reasons.
3585  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3586  * 2 - The owner is to be given access irrespective of mode bits for some
3587  *     operations, so that processes that chmod after opening a file don't
3588  *     break. I don't like this because it opens a security hole, but since
3589  *     the nfs server opens a security hole the size of a barn door anyhow,
3590  *     what the heck.
3591  *
3592  * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3593  * will return EPERM instead of EACCES. EPERM is always an error.
3594  */
3595 int
3596 nfsrv_access(struct vnode *vp, int flags, kauth_cred_t cred, int rdonly, struct lwp *lwp, int override)
3597 {
3598 	struct vattr vattr;
3599 	int error;
3600 	if (flags & VWRITE) {
3601 		/* Just vn_writechk() changed to check rdonly */
3602 		/*
3603 		 * Disallow write attempts on read-only file systems;
3604 		 * unless the file is a socket or a block or character
3605 		 * device resident on the file system.
3606 		 */
3607 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3608 			switch (vp->v_type) {
3609 			case VREG:
3610 			case VDIR:
3611 			case VLNK:
3612 				return (EROFS);
3613 			default:
3614 				break;
3615 			}
3616 		}
3617 
3618 		/*
3619 		 * If the vnode is in use as a process's text,
3620 		 * we can't allow writing.
3621 		 */
3622 		if (vp->v_iflag & VI_TEXT)
3623 			return (ETXTBSY);
3624 	}
3625 	error = VOP_GETATTR(vp, &vattr, cred);
3626 	if (error)
3627 		return (error);
3628 	error = VOP_ACCESS(vp, flags, cred);
3629 	/*
3630 	 * Allow certain operations for the owner (reads and writes
3631 	 * on files that are already open).
3632 	 */
3633 	if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid)
3634 		error = 0;
3635 	return error;
3636 }
3637