xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision 298cf604ccf133b101c6fad42d1a078a1fac58ca)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet6.h"
38 #include "opt_kdtrace.h"
39 
40 #include <sys/capability.h>
41 
42 /*
43  * generally, I don't like #includes inside .h files, but it seems to
44  * be the easiest way to handle the port.
45  */
46 #include <sys/hash.h>
47 #include <fs/nfs/nfsport.h>
48 #include <netinet/if_ether.h>
49 #include <net/if_types.h>
50 
51 #include <fs/nfsclient/nfs_kdtrace.h>
52 
53 #ifdef KDTRACE_HOOKS
54 dtrace_nfsclient_attrcache_flush_probe_func_t
55 		dtrace_nfscl_attrcache_flush_done_probe;
56 uint32_t	nfscl_attrcache_flush_done_id;
57 
58 dtrace_nfsclient_attrcache_get_hit_probe_func_t
59 		dtrace_nfscl_attrcache_get_hit_probe;
60 uint32_t	nfscl_attrcache_get_hit_id;
61 
62 dtrace_nfsclient_attrcache_get_miss_probe_func_t
63 		dtrace_nfscl_attrcache_get_miss_probe;
64 uint32_t	nfscl_attrcache_get_miss_id;
65 
66 dtrace_nfsclient_attrcache_load_probe_func_t
67 		dtrace_nfscl_attrcache_load_done_probe;
68 uint32_t	nfscl_attrcache_load_done_id;
69 #endif /* !KDTRACE_HOOKS */
70 
71 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
72 extern struct vop_vector newnfs_vnodeops;
73 extern struct vop_vector newnfs_fifoops;
74 extern uma_zone_t newnfsnode_zone;
75 extern struct buf_ops buf_ops_newnfs;
76 extern int ncl_pbuf_freecnt;
77 extern short nfsv4_cbport;
78 extern int nfscl_enablecallb;
79 extern int nfs_numnfscbd;
80 extern int nfscl_inited;
81 struct mtx nfs_clstate_mutex;
82 struct mtx ncl_iod_mutex;
83 NFSDLOCKMUTEX;
84 
85 extern void (*ncl_call_invalcaches)(struct vnode *);
86 
87 /*
88  * Comparison function for vfs_hash functions.
89  */
90 int
91 newnfs_vncmpf(struct vnode *vp, void *arg)
92 {
93 	struct nfsfh *nfhp = (struct nfsfh *)arg;
94 	struct nfsnode *np = VTONFS(vp);
95 
96 	if (np->n_fhp->nfh_len != nfhp->nfh_len ||
97 	    NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
98 		return (1);
99 	return (0);
100 }
101 
102 /*
103  * Look up a vnode/nfsnode by file handle.
104  * Callers must check for mount points!!
105  * In all cases, a pointer to a
106  * nfsnode structure is returned.
107  * This variant takes a "struct nfsfh *" as second argument and uses
108  * that structure up, either by hanging off the nfsnode or FREEing it.
109  */
110 int
111 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
112     struct componentname *cnp, struct thread *td, struct nfsnode **npp,
113     void *stuff, int lkflags)
114 {
115 	struct nfsnode *np, *dnp;
116 	struct vnode *vp, *nvp;
117 	struct nfsv4node *newd, *oldd;
118 	int error;
119 	u_int hash;
120 	struct nfsmount *nmp;
121 
122 	nmp = VFSTONFS(mntp);
123 	dnp = VTONFS(dvp);
124 	*npp = NULL;
125 
126 	hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
127 
128 	error = vfs_hash_get(mntp, hash, lkflags,
129 	    td, &nvp, newnfs_vncmpf, nfhp);
130 	if (error == 0 && nvp != NULL) {
131 		/*
132 		 * I believe there is a slight chance that vgonel() could
133 		 * get called on this vnode between when NFSVOPLOCK() drops
134 		 * the VI_LOCK() and vget() acquires it again, so that it
135 		 * hasn't yet had v_usecount incremented. If this were to
136 		 * happen, the VI_DOOMED flag would be set, so check for
137 		 * that here. Since we now have the v_usecount incremented,
138 		 * we should be ok until we vrele() it, if the VI_DOOMED
139 		 * flag isn't set now.
140 		 */
141 		VI_LOCK(nvp);
142 		if ((nvp->v_iflag & VI_DOOMED)) {
143 			VI_UNLOCK(nvp);
144 			vrele(nvp);
145 			error = ENOENT;
146 		} else {
147 			VI_UNLOCK(nvp);
148 		}
149 	}
150 	if (error) {
151 		FREE((caddr_t)nfhp, M_NFSFH);
152 		return (error);
153 	}
154 	if (nvp != NULL) {
155 		np = VTONFS(nvp);
156 		/*
157 		 * For NFSv4, check to see if it is the same name and
158 		 * replace the name, if it is different.
159 		 */
160 		oldd = newd = NULL;
161 		if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
162 		    nvp->v_type == VREG &&
163 		    (np->n_v4->n4_namelen != cnp->cn_namelen ||
164 		     NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
165 		     cnp->cn_namelen) ||
166 		     dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
167 		     NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
168 		     dnp->n_fhp->nfh_len))) {
169 		    MALLOC(newd, struct nfsv4node *,
170 			sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
171 			+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
172 		    NFSLOCKNODE(np);
173 		    if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
174 			&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
175 			 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
176 			 cnp->cn_namelen) ||
177 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
178 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
179 			 dnp->n_fhp->nfh_len))) {
180 			oldd = np->n_v4;
181 			np->n_v4 = newd;
182 			newd = NULL;
183 			np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
184 			np->n_v4->n4_namelen = cnp->cn_namelen;
185 			NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
186 			    dnp->n_fhp->nfh_len);
187 			NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
188 			    cnp->cn_namelen);
189 		    }
190 		    NFSUNLOCKNODE(np);
191 		}
192 		if (newd != NULL)
193 			FREE((caddr_t)newd, M_NFSV4NODE);
194 		if (oldd != NULL)
195 			FREE((caddr_t)oldd, M_NFSV4NODE);
196 		*npp = np;
197 		FREE((caddr_t)nfhp, M_NFSFH);
198 		return (0);
199 	}
200 	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
201 
202 	error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
203 	if (error) {
204 		uma_zfree(newnfsnode_zone, np);
205 		FREE((caddr_t)nfhp, M_NFSFH);
206 		return (error);
207 	}
208 	vp = nvp;
209 	KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
210 	vp->v_bufobj.bo_ops = &buf_ops_newnfs;
211 	vp->v_data = np;
212 	np->n_vnode = vp;
213 	/*
214 	 * Initialize the mutex even if the vnode is going to be a loser.
215 	 * This simplifies the logic in reclaim, which can then unconditionally
216 	 * destroy the mutex (in the case of the loser, or if hash_insert
217 	 * happened to return an error no special casing is needed).
218 	 */
219 	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
220 
221 	/*
222 	 * Are we getting the root? If so, make sure the vnode flags
223 	 * are correct
224 	 */
225 	if ((nfhp->nfh_len == nmp->nm_fhsize) &&
226 	    !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
227 		if (vp->v_type == VNON)
228 			vp->v_type = VDIR;
229 		vp->v_vflag |= VV_ROOT;
230 	}
231 
232 	np->n_fhp = nfhp;
233 	/*
234 	 * For NFSv4, we have to attach the directory file handle and
235 	 * file name, so that Open Ops can be done later.
236 	 */
237 	if (nmp->nm_flag & NFSMNT_NFSV4) {
238 		MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
239 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
240 		    M_WAITOK);
241 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
242 		np->n_v4->n4_namelen = cnp->cn_namelen;
243 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
244 		    dnp->n_fhp->nfh_len);
245 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
246 		    cnp->cn_namelen);
247 	} else {
248 		np->n_v4 = NULL;
249 	}
250 
251 	/*
252 	 * NFS supports recursive and shared locking.
253 	 */
254 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
255 	VN_LOCK_AREC(vp);
256 	VN_LOCK_ASHARE(vp);
257 	error = insmntque(vp, mntp);
258 	if (error != 0) {
259 		*npp = NULL;
260 		mtx_destroy(&np->n_mtx);
261 		FREE((caddr_t)nfhp, M_NFSFH);
262 		if (np->n_v4 != NULL)
263 			FREE((caddr_t)np->n_v4, M_NFSV4NODE);
264 		uma_zfree(newnfsnode_zone, np);
265 		return (error);
266 	}
267 	error = vfs_hash_insert(vp, hash, lkflags,
268 	    td, &nvp, newnfs_vncmpf, nfhp);
269 	if (error)
270 		return (error);
271 	if (nvp != NULL) {
272 		*npp = VTONFS(nvp);
273 		/* vfs_hash_insert() vput()'s the losing vnode */
274 		return (0);
275 	}
276 	*npp = np;
277 
278 	return (0);
279 }
280 
281 /*
282  * Anothe variant of nfs_nget(). This one is only used by reopen. It
283  * takes almost the same args as nfs_nget(), but only succeeds if an entry
284  * exists in the cache. (Since files should already be "open" with a
285  * vnode ref cnt on the node when reopen calls this, it should always
286  * succeed.)
287  * Also, don't get a vnode lock, since it may already be locked by some
288  * other process that is handling it. This is ok, since all other threads
289  * on the client are blocked by the nfsc_lock being exclusively held by the
290  * caller of this function.
291  */
292 int
293 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
294     struct thread *td, struct nfsnode **npp)
295 {
296 	struct vnode *nvp;
297 	u_int hash;
298 	struct nfsfh *nfhp;
299 	int error;
300 
301 	*npp = NULL;
302 	/* For forced dismounts, just return error. */
303 	if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
304 		return (EINTR);
305 	MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
306 	    M_NFSFH, M_WAITOK);
307 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
308 	nfhp->nfh_len = fhsize;
309 
310 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
311 
312 	/*
313 	 * First, try to get the vnode locked, but don't block for the lock.
314 	 */
315 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
316 	    newnfs_vncmpf, nfhp);
317 	if (error == 0 && nvp != NULL) {
318 		NFSVOPUNLOCK(nvp, 0);
319 	} else if (error == EBUSY) {
320 		/*
321 		 * The LK_EXCLOTHER lock type tells nfs_lock1() to not try
322 		 * and lock the vnode, but just get a v_usecount on it.
323 		 * LK_NOWAIT is set so that when vget() returns ENOENT,
324 		 * vfs_hash_get() fails instead of looping.
325 		 * If this succeeds, it is safe so long as a vflush() with
326 		 * FORCECLOSE has not been done. Since the Renew thread is
327 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
328 		 * a vflush() with FORCECLOSE, we should be ok here.
329 		 */
330 		if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
331 			error = EINTR;
332 		else
333 			error = vfs_hash_get(mntp, hash,
334 			    (LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
335 			    newnfs_vncmpf, nfhp);
336 	}
337 	FREE(nfhp, M_NFSFH);
338 	if (error)
339 		return (error);
340 	if (nvp != NULL) {
341 		*npp = VTONFS(nvp);
342 		return (0);
343 	}
344 	return (EINVAL);
345 }
346 
347 /*
348  * Load the attribute cache (that lives in the nfsnode entry) with
349  * the attributes of the second argument and
350  * Iff vaper not NULL
351  *    copy the attributes to *vaper
352  * Similar to nfs_loadattrcache(), except the attributes are passed in
353  * instead of being parsed out of the mbuf list.
354  */
355 int
356 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
357     void *stuff, int writeattr, int dontshrink)
358 {
359 	struct vnode *vp = *vpp;
360 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
361 	struct nfsnode *np;
362 	struct nfsmount *nmp;
363 	struct timespec mtime_save;
364 
365 	/*
366 	 * If v_type == VNON it is a new node, so fill in the v_type,
367 	 * n_mtime fields. Check to see if it represents a special
368 	 * device, and if so, check for a possible alias. Once the
369 	 * correct vnode has been obtained, fill in the rest of the
370 	 * information.
371 	 */
372 	np = VTONFS(vp);
373 	NFSLOCKNODE(np);
374 	if (vp->v_type != nvap->va_type) {
375 		vp->v_type = nvap->va_type;
376 		if (vp->v_type == VFIFO)
377 			vp->v_op = &newnfs_fifoops;
378 		np->n_mtime = nvap->va_mtime;
379 	}
380 	nmp = VFSTONFS(vp->v_mount);
381 	vap = &np->n_vattr.na_vattr;
382 	mtime_save = vap->va_mtime;
383 	if (writeattr) {
384 		np->n_vattr.na_filerev = nap->na_filerev;
385 		np->n_vattr.na_size = nap->na_size;
386 		np->n_vattr.na_mtime = nap->na_mtime;
387 		np->n_vattr.na_ctime = nap->na_ctime;
388 		np->n_vattr.na_fsid = nap->na_fsid;
389 		np->n_vattr.na_mode = nap->na_mode;
390 	} else {
391 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
392 		    sizeof (struct nfsvattr));
393 	}
394 
395 	/*
396 	 * For NFSv4, if the node's fsid is not equal to the mount point's
397 	 * fsid, return the low order 32bits of the node's fsid. This
398 	 * allows getcwd(3) to work. There is a chance that the fsid might
399 	 * be the same as a local fs, but since this is in an NFS mount
400 	 * point, I don't think that will cause any problems?
401 	 */
402 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
403 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
404 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
405 		/*
406 		 * va_fsid needs to be set to some value derived from
407 		 * np->n_vattr.na_filesid that is not equal
408 		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
409 		 * from the value used for the top level server volume
410 		 * in the mounted subtree.
411 		 */
412 		if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
413 		    (uint32_t)np->n_vattr.na_filesid[0])
414 			vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
415 		else
416 			vap->va_fsid = (uint32_t)hash32_buf(
417 			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
418 	} else
419 		vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
420 	np->n_attrstamp = time_second;
421 	if (vap->va_size != np->n_size) {
422 		if (vap->va_type == VREG) {
423 			if (dontshrink && vap->va_size < np->n_size) {
424 				/*
425 				 * We've been told not to shrink the file;
426 				 * zero np->n_attrstamp to indicate that
427 				 * the attributes are stale.
428 				 */
429 				vap->va_size = np->n_size;
430 				np->n_attrstamp = 0;
431 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
432 			} else if (np->n_flag & NMODIFIED) {
433 				/*
434 				 * We've modified the file: Use the larger
435 				 * of our size, and the server's size.
436 				 */
437 				if (vap->va_size < np->n_size) {
438 					vap->va_size = np->n_size;
439 				} else {
440 					np->n_size = vap->va_size;
441 					np->n_flag |= NSIZECHANGED;
442 				}
443 			} else {
444 				np->n_size = vap->va_size;
445 				np->n_flag |= NSIZECHANGED;
446 			}
447 			vnode_pager_setsize(vp, np->n_size);
448 		} else {
449 			np->n_size = vap->va_size;
450 		}
451 	}
452 	/*
453 	 * The following checks are added to prevent a race between (say)
454 	 * a READDIR+ and a WRITE.
455 	 * READDIR+, WRITE requests sent out.
456 	 * READDIR+ resp, WRITE resp received on client.
457 	 * However, the WRITE resp was handled before the READDIR+ resp
458 	 * causing the post op attrs from the write to be loaded first
459 	 * and the attrs from the READDIR+ to be loaded later. If this
460 	 * happens, we have stale attrs loaded into the attrcache.
461 	 * We detect this by for the mtime moving back. We invalidate the
462 	 * attrcache when this happens.
463 	 */
464 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
465 		/* Size changed or mtime went backwards */
466 		np->n_attrstamp = 0;
467 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
468 	}
469 	if (vaper != NULL) {
470 		NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
471 		if (np->n_flag & NCHG) {
472 			if (np->n_flag & NACC)
473 				vaper->va_atime = np->n_atim;
474 			if (np->n_flag & NUPD)
475 				vaper->va_mtime = np->n_mtim;
476 		}
477 	}
478 #ifdef KDTRACE_HOOKS
479 	if (np->n_attrstamp != 0)
480 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, 0);
481 #endif
482 	NFSUNLOCKNODE(np);
483 	return (0);
484 }
485 
486 /*
487  * Fill in the client id name. For these bytes:
488  * 1 - they must be unique
489  * 2 - they should be persistent across client reboots
490  * 1 is more critical than 2
491  * Use the mount point's unique id plus either the uuid or, if that
492  * isn't set, random junk.
493  */
494 void
495 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
496 {
497 	int uuidlen;
498 
499 	/*
500 	 * First, put in the 64bit mount point identifier.
501 	 */
502 	if (idlen >= sizeof (u_int64_t)) {
503 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
504 		cp += sizeof (u_int64_t);
505 		idlen -= sizeof (u_int64_t);
506 	}
507 
508 	/*
509 	 * If uuid is non-zero length, use it.
510 	 */
511 	uuidlen = strlen(uuid);
512 	if (uuidlen > 0 && idlen >= uuidlen) {
513 		NFSBCOPY(uuid, cp, uuidlen);
514 		cp += uuidlen;
515 		idlen -= uuidlen;
516 	}
517 
518 	/*
519 	 * This only normally happens if the uuid isn't set.
520 	 */
521 	while (idlen > 0) {
522 		*cp++ = (u_int8_t)(arc4random() % 256);
523 		idlen--;
524 	}
525 }
526 
527 /*
528  * Fill in a lock owner name. For now, pid + the process's creation time.
529  */
530 void
531 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
532 {
533 	union {
534 		u_int32_t	lval;
535 		u_int8_t	cval[4];
536 	} tl;
537 	struct proc *p;
538 
539 	if (id == NULL) {
540 		printf("NULL id\n");
541 		bzero(cp, NFSV4CL_LOCKNAMELEN);
542 		return;
543 	}
544 	if ((flags & F_POSIX) != 0) {
545 		p = (struct proc *)id;
546 		tl.lval = p->p_pid;
547 		*cp++ = tl.cval[0];
548 		*cp++ = tl.cval[1];
549 		*cp++ = tl.cval[2];
550 		*cp++ = tl.cval[3];
551 		tl.lval = p->p_stats->p_start.tv_sec;
552 		*cp++ = tl.cval[0];
553 		*cp++ = tl.cval[1];
554 		*cp++ = tl.cval[2];
555 		*cp++ = tl.cval[3];
556 		tl.lval = p->p_stats->p_start.tv_usec;
557 		*cp++ = tl.cval[0];
558 		*cp++ = tl.cval[1];
559 		*cp++ = tl.cval[2];
560 		*cp = tl.cval[3];
561 	} else if ((flags & F_FLOCK) != 0) {
562 		bcopy(&id, cp, sizeof(id));
563 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
564 	} else {
565 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
566 		bzero(cp, NFSV4CL_LOCKNAMELEN);
567 	}
568 }
569 
570 /*
571  * Find the parent process for the thread passed in as an argument.
572  * If none exists, return NULL, otherwise return a thread for the parent.
573  * (Can be any of the threads, since it is only used for td->td_proc.)
574  */
575 NFSPROC_T *
576 nfscl_getparent(struct thread *td)
577 {
578 	struct proc *p;
579 	struct thread *ptd;
580 
581 	if (td == NULL)
582 		return (NULL);
583 	p = td->td_proc;
584 	if (p->p_pid == 0)
585 		return (NULL);
586 	p = p->p_pptr;
587 	if (p == NULL)
588 		return (NULL);
589 	ptd = TAILQ_FIRST(&p->p_threads);
590 	return (ptd);
591 }
592 
593 /*
594  * Start up the renew kernel thread.
595  */
596 static void
597 start_nfscl(void *arg)
598 {
599 	struct nfsclclient *clp;
600 	struct thread *td;
601 
602 	clp = (struct nfsclclient *)arg;
603 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
604 	nfscl_renewthread(clp, td);
605 	kproc_exit(0);
606 }
607 
608 void
609 nfscl_start_renewthread(struct nfsclclient *clp)
610 {
611 
612 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
613 	    "nfscl");
614 }
615 
616 /*
617  * Handle wcc_data.
618  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
619  * as the first Op after PutFH.
620  * (For NFSv4, the postop attributes are after the Op, so they can't be
621  *  parsed here. A separate call to nfscl_postop_attr() is required.)
622  */
623 int
624 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
625     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
626 {
627 	u_int32_t *tl;
628 	struct nfsnode *np = VTONFS(vp);
629 	struct nfsvattr nfsva;
630 	int error = 0;
631 
632 	if (wccflagp != NULL)
633 		*wccflagp = 0;
634 	if (nd->nd_flag & ND_NFSV3) {
635 		*flagp = 0;
636 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
637 		if (*tl == newnfs_true) {
638 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
639 			if (wccflagp != NULL) {
640 				mtx_lock(&np->n_mtx);
641 				*wccflagp = (np->n_mtime.tv_sec ==
642 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
643 				    np->n_mtime.tv_nsec ==
644 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
645 				mtx_unlock(&np->n_mtx);
646 			}
647 		}
648 		error = nfscl_postop_attr(nd, nap, flagp, stuff);
649 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
650 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
651 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
652 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
653 		    NULL, NULL, NULL, NULL, NULL);
654 		if (error)
655 			return (error);
656 		/*
657 		 * Get rid of Op# and status for next op.
658 		 */
659 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
660 		if (*++tl)
661 			nd->nd_flag |= ND_NOMOREDATA;
662 		if (wccflagp != NULL &&
663 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
664 			mtx_lock(&np->n_mtx);
665 			*wccflagp = (np->n_mtime.tv_sec ==
666 			    nfsva.na_vattr.va_mtime.tv_sec &&
667 			    np->n_mtime.tv_nsec ==
668 			    nfsva.na_vattr.va_mtime.tv_sec);
669 			mtx_unlock(&np->n_mtx);
670 		}
671 	}
672 nfsmout:
673 	return (error);
674 }
675 
676 /*
677  * Get postop attributes.
678  */
679 int
680 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
681     void *stuff)
682 {
683 	u_int32_t *tl;
684 	int error = 0;
685 
686 	*retp = 0;
687 	if (nd->nd_flag & ND_NOMOREDATA)
688 		return (error);
689 	if (nd->nd_flag & ND_NFSV3) {
690 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
691 		*retp = fxdr_unsigned(int, *tl);
692 	} else if (nd->nd_flag & ND_NFSV4) {
693 		/*
694 		 * For NFSv4, the postop attr are at the end, so no point
695 		 * in looking if nd_repstat != 0.
696 		 */
697 		if (!nd->nd_repstat) {
698 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
699 			if (*(tl + 1))
700 				/* should never happen since nd_repstat != 0 */
701 				nd->nd_flag |= ND_NOMOREDATA;
702 			else
703 				*retp = 1;
704 		}
705 	} else if (!nd->nd_repstat) {
706 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
707 		*retp = 1;
708 	}
709 	if (*retp) {
710 		error = nfsm_loadattr(nd, nap);
711 		if (error)
712 			*retp = 0;
713 	}
714 nfsmout:
715 	return (error);
716 }
717 
718 /*
719  * Fill in the setable attributes. The full argument indicates whether
720  * to fill in them all or just mode and time.
721  */
722 void
723 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
724     struct vnode *vp, int flags, u_int32_t rdev)
725 {
726 	u_int32_t *tl;
727 	struct nfsv2_sattr *sp;
728 	nfsattrbit_t attrbits;
729 	struct timeval curtime;
730 
731 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
732 	case ND_NFSV2:
733 		NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
734 		if (vap->va_mode == (mode_t)VNOVAL)
735 			sp->sa_mode = newnfs_xdrneg1;
736 		else
737 			sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
738 		if (vap->va_uid == (uid_t)VNOVAL)
739 			sp->sa_uid = newnfs_xdrneg1;
740 		else
741 			sp->sa_uid = txdr_unsigned(vap->va_uid);
742 		if (vap->va_gid == (gid_t)VNOVAL)
743 			sp->sa_gid = newnfs_xdrneg1;
744 		else
745 			sp->sa_gid = txdr_unsigned(vap->va_gid);
746 		if (flags & NFSSATTR_SIZE0)
747 			sp->sa_size = 0;
748 		else if (flags & NFSSATTR_SIZENEG1)
749 			sp->sa_size = newnfs_xdrneg1;
750 		else if (flags & NFSSATTR_SIZERDEV)
751 			sp->sa_size = txdr_unsigned(rdev);
752 		else
753 			sp->sa_size = txdr_unsigned(vap->va_size);
754 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
755 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
756 		break;
757 	case ND_NFSV3:
758 		getmicrotime(&curtime);
759 		if (vap->va_mode != (mode_t)VNOVAL) {
760 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
761 			*tl++ = newnfs_true;
762 			*tl = txdr_unsigned(vap->va_mode);
763 		} else {
764 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
765 			*tl = newnfs_false;
766 		}
767 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
768 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
769 			*tl++ = newnfs_true;
770 			*tl = txdr_unsigned(vap->va_uid);
771 		} else {
772 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
773 			*tl = newnfs_false;
774 		}
775 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
776 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
777 			*tl++ = newnfs_true;
778 			*tl = txdr_unsigned(vap->va_gid);
779 		} else {
780 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
781 			*tl = newnfs_false;
782 		}
783 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
784 			NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
785 			*tl++ = newnfs_true;
786 			txdr_hyper(vap->va_size, tl);
787 		} else {
788 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
789 			*tl = newnfs_false;
790 		}
791 		if (vap->va_atime.tv_sec != VNOVAL) {
792 			if (vap->va_atime.tv_sec != curtime.tv_sec) {
793 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
794 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
795 				txdr_nfsv3time(&vap->va_atime, tl);
796 			} else {
797 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
798 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
799 			}
800 		} else {
801 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
802 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
803 		}
804 		if (vap->va_mtime.tv_sec != VNOVAL) {
805 			if (vap->va_mtime.tv_sec != curtime.tv_sec) {
806 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
807 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
808 				txdr_nfsv3time(&vap->va_mtime, tl);
809 			} else {
810 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
811 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
812 			}
813 		} else {
814 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
815 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
816 		}
817 		break;
818 	case ND_NFSV4:
819 		NFSZERO_ATTRBIT(&attrbits);
820 		if (vap->va_mode != (mode_t)VNOVAL)
821 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
822 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
823 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
824 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
825 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
826 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
827 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
828 		if (vap->va_atime.tv_sec != VNOVAL)
829 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
830 		if (vap->va_mtime.tv_sec != VNOVAL)
831 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
832 		(void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
833 		    &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
834 		break;
835 	};
836 }
837 
838 /*
839  * nfscl_request() - mostly a wrapper for newnfs_request().
840  */
841 int
842 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
843     struct ucred *cred, void *stuff)
844 {
845 	int ret, vers;
846 	struct nfsmount *nmp;
847 
848 	nmp = VFSTONFS(vp->v_mount);
849 	if (nd->nd_flag & ND_NFSV4)
850 		vers = NFS_VER4;
851 	else if (nd->nd_flag & ND_NFSV3)
852 		vers = NFS_VER3;
853 	else
854 		vers = NFS_VER2;
855 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
856 		NFS_PROG, vers, NULL, 1, NULL, NULL);
857 	return (ret);
858 }
859 
860 /*
861  * fill in this bsden's variant of statfs using nfsstatfs.
862  */
863 void
864 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
865 {
866 	struct statfs *sbp = (struct statfs *)statfs;
867 
868 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
869 		sbp->f_bsize = NFS_FABLKSIZE;
870 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
871 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
872 		/*
873 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
874 		 * the value after dividing by NFS_FABLKSIZE is small
875 		 * enough that it will fit in 63bits, so it is ok to
876 		 * assign it to f_bavail without fear that it will become
877 		 * negative.
878 		 */
879 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
880 		sbp->f_files = sfp->sf_tfiles;
881 		/* Since f_ffree is int64_t, clip it to 63bits. */
882 		if (sfp->sf_ffiles > INT64_MAX)
883 			sbp->f_ffree = INT64_MAX;
884 		else
885 			sbp->f_ffree = sfp->sf_ffiles;
886 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
887 		/*
888 		 * The type casts to (int32_t) ensure that this code is
889 		 * compatible with the old NFS client, in that it will
890 		 * propagate bit31 to the high order bits. This may or may
891 		 * not be correct for NFSv2, but since it is a legacy
892 		 * environment, I'd rather retain backwards compatibility.
893 		 */
894 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
895 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
896 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
897 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
898 		sbp->f_files = 0;
899 		sbp->f_ffree = 0;
900 	}
901 }
902 
903 /*
904  * Use the fsinfo stuff to update the mount point.
905  */
906 void
907 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
908 {
909 
910 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
911 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
912 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
913 		    ~(NFS_FABLKSIZE - 1);
914 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
915 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
916 		if (nmp->nm_wsize == 0)
917 			nmp->nm_wsize = fsp->fs_wtmax;
918 	}
919 	if (nmp->nm_wsize < NFS_FABLKSIZE)
920 		nmp->nm_wsize = NFS_FABLKSIZE;
921 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
922 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
923 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
924 		    ~(NFS_FABLKSIZE - 1);
925 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
926 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
927 		if (nmp->nm_rsize == 0)
928 			nmp->nm_rsize = fsp->fs_rtmax;
929 	}
930 	if (nmp->nm_rsize < NFS_FABLKSIZE)
931 		nmp->nm_rsize = NFS_FABLKSIZE;
932 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
933 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
934 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
935 		    ~(NFS_DIRBLKSIZ - 1);
936 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
937 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
938 		if (nmp->nm_readdirsize == 0)
939 			nmp->nm_readdirsize = fsp->fs_rtmax;
940 	}
941 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
942 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
943 	if (fsp->fs_maxfilesize > 0 &&
944 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
945 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
946 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
947 	nmp->nm_state |= NFSSTA_GOTFSINFO;
948 }
949 
950 /*
951  * Get a pointer to my IP addrress and return it.
952  * Return NULL if you can't find one.
953  */
954 u_int8_t *
955 nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
956 {
957 	struct sockaddr_in sad, *sin;
958 	struct rtentry *rt;
959 	u_int8_t *retp = NULL;
960 	static struct in_addr laddr;
961 
962 	*isinet6p = 0;
963 	/*
964 	 * Loop up a route for the destination address.
965 	 */
966 	if (nmp->nm_nam->sa_family == AF_INET) {
967 		bzero(&sad, sizeof (sad));
968 		sin = (struct sockaddr_in *)nmp->nm_nam;
969 		sad.sin_family = AF_INET;
970 		sad.sin_len = sizeof (struct sockaddr_in);
971 		sad.sin_addr.s_addr = sin->sin_addr.s_addr;
972 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
973 		rt = rtalloc1_fib((struct sockaddr *)&sad, 0, 0UL,
974 		     curthread->td_proc->p_fibnum);
975 		if (rt != NULL) {
976 			if (rt->rt_ifp != NULL &&
977 			    rt->rt_ifa != NULL &&
978 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
979 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
980 				sin = (struct sockaddr_in *)
981 				    rt->rt_ifa->ifa_addr;
982 				laddr.s_addr = sin->sin_addr.s_addr;
983 				retp = (u_int8_t *)&laddr;
984 			}
985 			RTFREE_LOCKED(rt);
986 		}
987 		CURVNET_RESTORE();
988 #ifdef INET6
989 	} else if (nmp->nm_nam->sa_family == AF_INET6) {
990 		struct sockaddr_in6 sad6, *sin6;
991 		static struct in6_addr laddr6;
992 
993 		bzero(&sad6, sizeof (sad6));
994 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
995 		sad6.sin6_family = AF_INET6;
996 		sad6.sin6_len = sizeof (struct sockaddr_in6);
997 		sad6.sin6_addr = sin6->sin6_addr;
998 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
999 		rt = rtalloc1_fib((struct sockaddr *)&sad6, 0, 0UL,
1000 		     curthread->td_proc->p_fibnum);
1001 		if (rt != NULL) {
1002 			if (rt->rt_ifp != NULL &&
1003 			    rt->rt_ifa != NULL &&
1004 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1005 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
1006 				sin6 = (struct sockaddr_in6 *)
1007 				    rt->rt_ifa->ifa_addr;
1008 				laddr6 = sin6->sin6_addr;
1009 				retp = (u_int8_t *)&laddr6;
1010 				*isinet6p = 1;
1011 			}
1012 			RTFREE_LOCKED(rt);
1013 		}
1014 		CURVNET_RESTORE();
1015 #endif
1016 	}
1017 	return (retp);
1018 }
1019 
1020 /*
1021  * Copy NFS uid, gids from the cred structure.
1022  */
1023 void
1024 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1025 {
1026 	int i;
1027 
1028 	KASSERT(cr->cr_ngroups >= 0,
1029 	    ("newnfs_copyincred: negative cr_ngroups"));
1030 	nfscr->nfsc_uid = cr->cr_uid;
1031 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1032 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1033 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1034 }
1035 
1036 
1037 /*
1038  * Do any client specific initialization.
1039  */
1040 void
1041 nfscl_init(void)
1042 {
1043 	static int inited = 0;
1044 
1045 	if (inited)
1046 		return;
1047 	inited = 1;
1048 	nfscl_inited = 1;
1049 	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1050 }
1051 
1052 /*
1053  * Check each of the attributes to be set, to ensure they aren't already
1054  * the correct value. Disable setting ones already correct.
1055  */
1056 int
1057 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1058 {
1059 
1060 	if (vap->va_mode != (mode_t)VNOVAL) {
1061 		if (vap->va_mode == nvap->na_mode)
1062 			vap->va_mode = (mode_t)VNOVAL;
1063 	}
1064 	if (vap->va_uid != (uid_t)VNOVAL) {
1065 		if (vap->va_uid == nvap->na_uid)
1066 			vap->va_uid = (uid_t)VNOVAL;
1067 	}
1068 	if (vap->va_gid != (gid_t)VNOVAL) {
1069 		if (vap->va_gid == nvap->na_gid)
1070 			vap->va_gid = (gid_t)VNOVAL;
1071 	}
1072 	if (vap->va_size != VNOVAL) {
1073 		if (vap->va_size == nvap->na_size)
1074 			vap->va_size = VNOVAL;
1075 	}
1076 
1077 	/*
1078 	 * We are normally called with only a partially initialized
1079 	 * VAP.  Since the NFSv3 spec says that server may use the
1080 	 * file attributes to store the verifier, the spec requires
1081 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1082 	 * in atime, but we can't really assume that all servers will
1083 	 * so we ensure that our SETATTR sets both atime and mtime.
1084 	 */
1085 	if (vap->va_mtime.tv_sec == VNOVAL)
1086 		vfs_timestamp(&vap->va_mtime);
1087 	if (vap->va_atime.tv_sec == VNOVAL)
1088 		vap->va_atime = vap->va_mtime;
1089 	return (1);
1090 }
1091 
1092 /*
1093  * Map nfsv4 errors to errno.h errors.
1094  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1095  * error should only be returned for the Open, Create and Setattr Ops.
1096  * As such, most calls can just pass in 0 for those arguments.
1097  */
1098 APPLESTATIC int
1099 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1100 {
1101 	struct proc *p;
1102 
1103 	if (error < 10000)
1104 		return (error);
1105 	if (td != NULL)
1106 		p = td->td_proc;
1107 	else
1108 		p = NULL;
1109 	switch (error) {
1110 	case NFSERR_BADOWNER:
1111 		tprintf(p, LOG_INFO,
1112 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1113 		    uid, gid);
1114 		return (EPERM);
1115 	case NFSERR_BADNAME:
1116 	case NFSERR_BADCHAR:
1117 		printf("nfsv4 char/name not handled by server\n");
1118 		return (ENOENT);
1119 	case NFSERR_STALECLIENTID:
1120 	case NFSERR_STALESTATEID:
1121 	case NFSERR_EXPIRED:
1122 	case NFSERR_BADSTATEID:
1123 	case NFSERR_BADSESSION:
1124 		printf("nfsv4 recover err returned %d\n", error);
1125 		return (EIO);
1126 	case NFSERR_BADHANDLE:
1127 	case NFSERR_SERVERFAULT:
1128 	case NFSERR_BADTYPE:
1129 	case NFSERR_FHEXPIRED:
1130 	case NFSERR_RESOURCE:
1131 	case NFSERR_MOVED:
1132 	case NFSERR_NOFILEHANDLE:
1133 	case NFSERR_MINORVERMISMATCH:
1134 	case NFSERR_OLDSTATEID:
1135 	case NFSERR_BADSEQID:
1136 	case NFSERR_LEASEMOVED:
1137 	case NFSERR_RECLAIMBAD:
1138 	case NFSERR_BADXDR:
1139 	case NFSERR_OPILLEGAL:
1140 		printf("nfsv4 client/server protocol prob err=%d\n",
1141 		    error);
1142 		return (EIO);
1143 	default:
1144 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1145 		return (EIO);
1146 	};
1147 }
1148 
1149 /*
1150  * Check to see if the process for this owner exists. Return 1 if it doesn't
1151  * and 0 otherwise.
1152  */
1153 int
1154 nfscl_procdoesntexist(u_int8_t *own)
1155 {
1156 	union {
1157 		u_int32_t	lval;
1158 		u_int8_t	cval[4];
1159 	} tl;
1160 	struct proc *p;
1161 	pid_t pid;
1162 	int ret = 0;
1163 
1164 	tl.cval[0] = *own++;
1165 	tl.cval[1] = *own++;
1166 	tl.cval[2] = *own++;
1167 	tl.cval[3] = *own++;
1168 	pid = tl.lval;
1169 	p = pfind_locked(pid);
1170 	if (p == NULL)
1171 		return (1);
1172 	if (p->p_stats == NULL) {
1173 		PROC_UNLOCK(p);
1174 		return (0);
1175 	}
1176 	tl.cval[0] = *own++;
1177 	tl.cval[1] = *own++;
1178 	tl.cval[2] = *own++;
1179 	tl.cval[3] = *own++;
1180 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1181 		ret = 1;
1182 	} else {
1183 		tl.cval[0] = *own++;
1184 		tl.cval[1] = *own++;
1185 		tl.cval[2] = *own++;
1186 		tl.cval[3] = *own;
1187 		if (tl.lval != p->p_stats->p_start.tv_usec)
1188 			ret = 1;
1189 	}
1190 	PROC_UNLOCK(p);
1191 	return (ret);
1192 }
1193 
1194 /*
1195  * - nfs pseudo system call for the client
1196  */
1197 /*
1198  * MPSAFE
1199  */
1200 static int
1201 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1202 {
1203 	struct file *fp;
1204 	struct nfscbd_args nfscbdarg;
1205 	struct nfsd_nfscbd_args nfscbdarg2;
1206 	int error;
1207 	struct nameidata nd;
1208 	struct nfscl_dumpmntopts dumpmntopts;
1209 	char *buf;
1210 
1211 	if (uap->flag & NFSSVC_CBADDSOCK) {
1212 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1213 		if (error)
1214 			return (error);
1215 		/*
1216 		 * Since we don't know what rights might be required,
1217 		 * pretend that we need them all. It is better to be too
1218 		 * careful than too reckless.
1219 		 */
1220 		if ((error = fget(td, nfscbdarg.sock, CAP_SOCK_ALL, &fp))
1221 		    != 0) {
1222 			return (error);
1223 		}
1224 		if (fp->f_type != DTYPE_SOCKET) {
1225 			fdrop(fp, td);
1226 			return (EPERM);
1227 		}
1228 		error = nfscbd_addsock(fp);
1229 		fdrop(fp, td);
1230 		if (!error && nfscl_enablecallb == 0) {
1231 			nfsv4_cbport = nfscbdarg.port;
1232 			nfscl_enablecallb = 1;
1233 		}
1234 	} else if (uap->flag & NFSSVC_NFSCBD) {
1235 		if (uap->argp == NULL)
1236 			return (EINVAL);
1237 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1238 		    sizeof(nfscbdarg2));
1239 		if (error)
1240 			return (error);
1241 		error = nfscbd_nfsd(td, &nfscbdarg2);
1242 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1243 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1244 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1245 		    dumpmntopts.ndmnt_blen > 1024))
1246 			error = EINVAL;
1247 		if (error == 0)
1248 			error = nfsrv_lookupfilename(&nd,
1249 			    dumpmntopts.ndmnt_fname, td);
1250 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1251 		    "nfs") != 0) {
1252 			vput(nd.ni_vp);
1253 			error = EINVAL;
1254 		}
1255 		if (error == 0) {
1256 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1257 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1258 			    dumpmntopts.ndmnt_blen);
1259 			vput(nd.ni_vp);
1260 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1261 			    dumpmntopts.ndmnt_blen);
1262 			free(buf, M_TEMP);
1263 		}
1264 	} else {
1265 		error = EINVAL;
1266 	}
1267 	return (error);
1268 }
1269 
1270 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1271 
1272 /*
1273  * Called once to initialize data structures...
1274  */
1275 static int
1276 nfscl_modevent(module_t mod, int type, void *data)
1277 {
1278 	int error = 0;
1279 	static int loaded = 0;
1280 
1281 	switch (type) {
1282 	case MOD_LOAD:
1283 		if (loaded)
1284 			return (0);
1285 		newnfs_portinit();
1286 		mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1287 		    MTX_DEF);
1288 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1289 		nfscl_init();
1290 		NFSD_LOCK();
1291 		nfsrvd_cbinit(0);
1292 		NFSD_UNLOCK();
1293 		ncl_call_invalcaches = ncl_invalcaches;
1294 		nfsd_call_nfscl = nfssvc_nfscl;
1295 		loaded = 1;
1296 		break;
1297 
1298 	case MOD_UNLOAD:
1299 		if (nfs_numnfscbd != 0) {
1300 			error = EBUSY;
1301 			break;
1302 		}
1303 
1304 		/*
1305 		 * XXX: Unloading of nfscl module is unsupported.
1306 		 */
1307 #if 0
1308 		ncl_call_invalcaches = NULL;
1309 		nfsd_call_nfscl = NULL;
1310 		/* and get rid of the mutexes */
1311 		mtx_destroy(&nfs_clstate_mutex);
1312 		mtx_destroy(&ncl_iod_mutex);
1313 		loaded = 0;
1314 		break;
1315 #else
1316 		/* FALLTHROUGH */
1317 #endif
1318 	default:
1319 		error = EOPNOTSUPP;
1320 		break;
1321 	}
1322 	return error;
1323 }
1324 static moduledata_t nfscl_mod = {
1325 	"nfscl",
1326 	nfscl_modevent,
1327 	NULL,
1328 };
1329 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1330 
1331 /* So that loader and kldload(2) can find us, wherever we are.. */
1332 MODULE_VERSION(nfscl, 1);
1333 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1334 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1335 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1336 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1337 
1338