xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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 /*
38  * generally, I don't like #includes inside .h files, but it seems to
39  * be the easiest way to handle the port.
40  */
41 #include <sys/hash.h>
42 #include <fs/nfs/nfsport.h>
43 #include <netinet/if_ether.h>
44 #include <net/if_types.h>
45 
46 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
47 extern struct vop_vector newnfs_vnodeops;
48 extern struct vop_vector newnfs_fifoops;
49 extern uma_zone_t newnfsnode_zone;
50 extern struct buf_ops buf_ops_newnfs;
51 extern int ncl_pbuf_freecnt;
52 extern short nfsv4_cbport;
53 extern int nfscl_enablecallb;
54 extern int nfs_numnfscbd;
55 extern int nfscl_inited;
56 struct mtx nfs_clstate_mutex;
57 struct mtx ncl_iod_mutex;
58 NFSDLOCKMUTEX;
59 
60 extern void (*ncl_call_invalcaches)(struct vnode *);
61 
62 /*
63  * Comparison function for vfs_hash functions.
64  */
65 int
66 newnfs_vncmpf(struct vnode *vp, void *arg)
67 {
68 	struct nfsfh *nfhp = (struct nfsfh *)arg;
69 	struct nfsnode *np = VTONFS(vp);
70 
71 	if (np->n_fhp->nfh_len != nfhp->nfh_len ||
72 	    NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
73 		return (1);
74 	return (0);
75 }
76 
77 /*
78  * Look up a vnode/nfsnode by file handle.
79  * Callers must check for mount points!!
80  * In all cases, a pointer to a
81  * nfsnode structure is returned.
82  * This variant takes a "struct nfsfh *" as second argument and uses
83  * that structure up, either by hanging off the nfsnode or FREEing it.
84  */
85 int
86 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
87     struct componentname *cnp, struct thread *td, struct nfsnode **npp,
88     void *stuff, int lkflags)
89 {
90 	struct nfsnode *np, *dnp;
91 	struct vnode *vp, *nvp;
92 	struct nfsv4node *newd, *oldd;
93 	int error;
94 	u_int hash;
95 	struct nfsmount *nmp;
96 
97 	nmp = VFSTONFS(mntp);
98 	dnp = VTONFS(dvp);
99 	*npp = NULL;
100 
101 	hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
102 
103 	error = vfs_hash_get(mntp, hash, lkflags,
104 	    td, &nvp, newnfs_vncmpf, nfhp);
105 	if (error == 0 && nvp != NULL) {
106 		/*
107 		 * I believe there is a slight chance that vgonel() could
108 		 * get called on this vnode between when vn_lock() drops
109 		 * the VI_LOCK() and vget() acquires it again, so that it
110 		 * hasn't yet had v_usecount incremented. If this were to
111 		 * happen, the VI_DOOMED flag would be set, so check for
112 		 * that here. Since we now have the v_usecount incremented,
113 		 * we should be ok until we vrele() it, if the VI_DOOMED
114 		 * flag isn't set now.
115 		 */
116 		VI_LOCK(nvp);
117 		if ((nvp->v_iflag & VI_DOOMED)) {
118 			VI_UNLOCK(nvp);
119 			vrele(nvp);
120 			error = ENOENT;
121 		} else {
122 			VI_UNLOCK(nvp);
123 		}
124 	}
125 	if (error) {
126 		FREE((caddr_t)nfhp, M_NFSFH);
127 		return (error);
128 	}
129 	if (nvp != NULL) {
130 		np = VTONFS(nvp);
131 		/*
132 		 * For NFSv4, check to see if it is the same name and
133 		 * replace the name, if it is different.
134 		 */
135 		oldd = newd = NULL;
136 		if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
137 		    nvp->v_type == VREG &&
138 		    (np->n_v4->n4_namelen != cnp->cn_namelen ||
139 		     NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
140 		     cnp->cn_namelen) ||
141 		     dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
142 		     NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
143 		     dnp->n_fhp->nfh_len))) {
144 		    MALLOC(newd, struct nfsv4node *,
145 			sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
146 			+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
147 		    NFSLOCKNODE(np);
148 		    if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
149 			&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
150 			 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
151 			 cnp->cn_namelen) ||
152 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
153 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
154 			 dnp->n_fhp->nfh_len))) {
155 			oldd = np->n_v4;
156 			np->n_v4 = newd;
157 			newd = NULL;
158 			np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
159 			np->n_v4->n4_namelen = cnp->cn_namelen;
160 			NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
161 			    dnp->n_fhp->nfh_len);
162 			NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
163 			    cnp->cn_namelen);
164 		    }
165 		    NFSUNLOCKNODE(np);
166 		}
167 		if (newd != NULL)
168 			FREE((caddr_t)newd, M_NFSV4NODE);
169 		if (oldd != NULL)
170 			FREE((caddr_t)oldd, M_NFSV4NODE);
171 		*npp = np;
172 		FREE((caddr_t)nfhp, M_NFSFH);
173 		return (0);
174 	}
175 
176 	/*
177 	 * Allocate before getnewvnode since doing so afterward
178 	 * might cause a bogus v_data pointer to get dereferenced
179 	 * elsewhere if zalloc should block.
180 	 */
181 	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
182 
183 	error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
184 	if (error) {
185 		uma_zfree(newnfsnode_zone, np);
186 		FREE((caddr_t)nfhp, M_NFSFH);
187 		return (error);
188 	}
189 	vp = nvp;
190 	vp->v_bufobj.bo_ops = &buf_ops_newnfs;
191 	vp->v_data = np;
192 	np->n_vnode = vp;
193 	/*
194 	 * Initialize the mutex even if the vnode is going to be a loser.
195 	 * This simplifies the logic in reclaim, which can then unconditionally
196 	 * destroy the mutex (in the case of the loser, or if hash_insert
197 	 * happened to return an error no special casing is needed).
198 	 */
199 	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
200 
201 	/*
202 	 * Are we getting the root? If so, make sure the vnode flags
203 	 * are correct
204 	 */
205 	if ((nfhp->nfh_len == nmp->nm_fhsize) &&
206 	    !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
207 		if (vp->v_type == VNON)
208 			vp->v_type = VDIR;
209 		vp->v_vflag |= VV_ROOT;
210 	}
211 
212 	np->n_fhp = nfhp;
213 	/*
214 	 * For NFSv4, we have to attach the directory file handle and
215 	 * file name, so that Open Ops can be done later.
216 	 */
217 	if (nmp->nm_flag & NFSMNT_NFSV4) {
218 		MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
219 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
220 		    M_WAITOK);
221 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
222 		np->n_v4->n4_namelen = cnp->cn_namelen;
223 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
224 		    dnp->n_fhp->nfh_len);
225 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
226 		    cnp->cn_namelen);
227 	} else {
228 		np->n_v4 = NULL;
229 	}
230 
231 	/*
232 	 * NFS supports recursive and shared locking.
233 	 */
234 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
235 	VN_LOCK_AREC(vp);
236 	VN_LOCK_ASHARE(vp);
237 	error = insmntque(vp, mntp);
238 	if (error != 0) {
239 		*npp = NULL;
240 		mtx_destroy(&np->n_mtx);
241 		FREE((caddr_t)nfhp, M_NFSFH);
242 		if (np->n_v4 != NULL)
243 			FREE((caddr_t)np->n_v4, M_NFSV4NODE);
244 		uma_zfree(newnfsnode_zone, np);
245 		return (error);
246 	}
247 	error = vfs_hash_insert(vp, hash, lkflags,
248 	    td, &nvp, newnfs_vncmpf, nfhp);
249 	if (error)
250 		return (error);
251 	if (nvp != NULL) {
252 		*npp = VTONFS(nvp);
253 		/* vfs_hash_insert() vput()'s the losing vnode */
254 		return (0);
255 	}
256 	*npp = np;
257 
258 	return (0);
259 }
260 
261 /*
262  * Anothe variant of nfs_nget(). This one is only used by reopen. It
263  * takes almost the same args as nfs_nget(), but only succeeds if an entry
264  * exists in the cache. (Since files should already be "open" with a
265  * vnode ref cnt on the node when reopen calls this, it should always
266  * succeed.)
267  * Also, don't get a vnode lock, since it may already be locked by some
268  * other process that is handling it. This is ok, since all other threads
269  * on the client are blocked by the nfsc_lock being exclusively held by the
270  * caller of this function.
271  */
272 int
273 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
274     struct thread *td, struct nfsnode **npp)
275 {
276 	struct vnode *nvp;
277 	u_int hash;
278 	struct nfsfh *nfhp;
279 	int error;
280 
281 	*npp = NULL;
282 	/* For forced dismounts, just return error. */
283 	if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
284 		return (EINTR);
285 	MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
286 	    M_NFSFH, M_WAITOK);
287 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
288 	nfhp->nfh_len = fhsize;
289 
290 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
291 
292 	/*
293 	 * First, try to get the vnode locked, but don't block for the lock.
294 	 */
295 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
296 	    newnfs_vncmpf, nfhp);
297 	if (error == 0 && nvp != NULL) {
298 		VOP_UNLOCK(nvp, 0);
299 	} else if (error == EBUSY) {
300 		/*
301 		 * The LK_EXCLOTHER lock type tells nfs_lock1() to not try
302 		 * and lock the vnode, but just get a v_usecount on it.
303 		 * LK_NOWAIT is set so that when vget() returns ENOENT,
304 		 * vfs_hash_get() fails instead of looping.
305 		 * If this succeeds, it is safe so long as a vflush() with
306 		 * FORCECLOSE has not been done. Since the Renew thread is
307 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
308 		 * a vflush() with FORCECLOSE, we should be ok here.
309 		 */
310 		if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
311 			error = EINTR;
312 		else
313 			error = vfs_hash_get(mntp, hash,
314 			    (LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
315 			    newnfs_vncmpf, nfhp);
316 	}
317 	FREE(nfhp, M_NFSFH);
318 	if (error)
319 		return (error);
320 	if (nvp != NULL) {
321 		*npp = VTONFS(nvp);
322 		return (0);
323 	}
324 	return (EINVAL);
325 }
326 
327 /*
328  * Load the attribute cache (that lives in the nfsnode entry) with
329  * the attributes of the second argument and
330  * Iff vaper not NULL
331  *    copy the attributes to *vaper
332  * Similar to nfs_loadattrcache(), except the attributes are passed in
333  * instead of being parsed out of the mbuf list.
334  */
335 int
336 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
337     void *stuff, int writeattr, int dontshrink)
338 {
339 	struct vnode *vp = *vpp;
340 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
341 	struct nfsnode *np;
342 	struct nfsmount *nmp;
343 	struct timespec mtime_save;
344 
345 	/*
346 	 * If v_type == VNON it is a new node, so fill in the v_type,
347 	 * n_mtime fields. Check to see if it represents a special
348 	 * device, and if so, check for a possible alias. Once the
349 	 * correct vnode has been obtained, fill in the rest of the
350 	 * information.
351 	 */
352 	np = VTONFS(vp);
353 	NFSLOCKNODE(np);
354 	if (vp->v_type != nvap->va_type) {
355 		vp->v_type = nvap->va_type;
356 		if (vp->v_type == VFIFO)
357 			vp->v_op = &newnfs_fifoops;
358 		np->n_mtime = nvap->va_mtime;
359 	}
360 	nmp = VFSTONFS(vp->v_mount);
361 	vap = &np->n_vattr.na_vattr;
362 	mtime_save = vap->va_mtime;
363 	if (writeattr) {
364 		np->n_vattr.na_filerev = nap->na_filerev;
365 		np->n_vattr.na_size = nap->na_size;
366 		np->n_vattr.na_mtime = nap->na_mtime;
367 		np->n_vattr.na_ctime = nap->na_ctime;
368 		np->n_vattr.na_fsid = nap->na_fsid;
369 	} else {
370 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
371 		    sizeof (struct nfsvattr));
372 	}
373 
374 	/*
375 	 * For NFSv4, if the node's fsid is not equal to the mount point's
376 	 * fsid, return the low order 32bits of the node's fsid. This
377 	 * allows getcwd(3) to work. There is a chance that the fsid might
378 	 * be the same as a local fs, but since this is in an NFS mount
379 	 * point, I don't think that will cause any problems?
380 	 */
381 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
382 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
383 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
384 		/*
385 		 * va_fsid needs to be set to some value derived from
386 		 * np->n_vattr.na_filesid that is not equal
387 		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
388 		 * from the value used for the top level server volume
389 		 * in the mounted subtree.
390 		 */
391 		if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
392 		    (uint32_t)np->n_vattr.na_filesid[0])
393 			vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
394 		else
395 			vap->va_fsid = (uint32_t)hash32_buf(
396 			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
397 	} else
398 		vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
399 	np->n_attrstamp = time_second;
400 	if (vap->va_size != np->n_size) {
401 		if (vap->va_type == VREG) {
402 			if (dontshrink && vap->va_size < np->n_size) {
403 				/*
404 				 * We've been told not to shrink the file;
405 				 * zero np->n_attrstamp to indicate that
406 				 * the attributes are stale.
407 				 */
408 				vap->va_size = np->n_size;
409 				np->n_attrstamp = 0;
410 			} else if (np->n_flag & NMODIFIED) {
411 				/*
412 				 * We've modified the file: Use the larger
413 				 * of our size, and the server's size.
414 				 */
415 				if (vap->va_size < np->n_size) {
416 					vap->va_size = np->n_size;
417 				} else {
418 					np->n_size = vap->va_size;
419 					np->n_flag |= NSIZECHANGED;
420 				}
421 			} else {
422 				np->n_size = vap->va_size;
423 				np->n_flag |= NSIZECHANGED;
424 			}
425 			vnode_pager_setsize(vp, np->n_size);
426 		} else {
427 			np->n_size = vap->va_size;
428 		}
429 	}
430 	/*
431 	 * The following checks are added to prevent a race between (say)
432 	 * a READDIR+ and a WRITE.
433 	 * READDIR+, WRITE requests sent out.
434 	 * READDIR+ resp, WRITE resp received on client.
435 	 * However, the WRITE resp was handled before the READDIR+ resp
436 	 * causing the post op attrs from the write to be loaded first
437 	 * and the attrs from the READDIR+ to be loaded later. If this
438 	 * happens, we have stale attrs loaded into the attrcache.
439 	 * We detect this by for the mtime moving back. We invalidate the
440 	 * attrcache when this happens.
441 	 */
442 	if (timespeccmp(&mtime_save, &vap->va_mtime, >))
443 		/* Size changed or mtime went backwards */
444 		np->n_attrstamp = 0;
445 	if (vaper != NULL) {
446 		NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
447 		if (np->n_flag & NCHG) {
448 			if (np->n_flag & NACC)
449 				vaper->va_atime = np->n_atim;
450 			if (np->n_flag & NUPD)
451 				vaper->va_mtime = np->n_mtim;
452 		}
453 	}
454 	NFSUNLOCKNODE(np);
455 	return (0);
456 }
457 
458 /*
459  * Fill in the client id name. For these bytes:
460  * 1 - they must be unique
461  * 2 - they should be persistent across client reboots
462  * 1 is more critical than 2
463  * Use the mount point's unique id plus either the uuid or, if that
464  * isn't set, random junk.
465  */
466 void
467 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
468 {
469 	int uuidlen;
470 
471 	/*
472 	 * First, put in the 64bit mount point identifier.
473 	 */
474 	if (idlen >= sizeof (u_int64_t)) {
475 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
476 		cp += sizeof (u_int64_t);
477 		idlen -= sizeof (u_int64_t);
478 	}
479 
480 	/*
481 	 * If uuid is non-zero length, use it.
482 	 */
483 	uuidlen = strlen(uuid);
484 	if (uuidlen > 0 && idlen >= uuidlen) {
485 		NFSBCOPY(uuid, cp, uuidlen);
486 		cp += uuidlen;
487 		idlen -= uuidlen;
488 	}
489 
490 	/*
491 	 * This only normally happens if the uuid isn't set.
492 	 */
493 	while (idlen > 0) {
494 		*cp++ = (u_int8_t)(arc4random() % 256);
495 		idlen--;
496 	}
497 }
498 
499 /*
500  * Fill in a lock owner name. For now, pid + the process's creation time.
501  */
502 void
503 nfscl_filllockowner(struct thread *td, u_int8_t *cp)
504 {
505 	union {
506 		u_int32_t	lval;
507 		u_int8_t	cval[4];
508 	} tl;
509 	struct proc *p;
510 
511 if (td == NULL) {
512 	printf("NULL td\n");
513 	bzero(cp, 12);
514 	return;
515 }
516 	p = td->td_proc;
517 if (p == NULL) {
518 	printf("NULL pid\n");
519 	bzero(cp, 12);
520 	return;
521 }
522 	tl.lval = p->p_pid;
523 	*cp++ = tl.cval[0];
524 	*cp++ = tl.cval[1];
525 	*cp++ = tl.cval[2];
526 	*cp++ = tl.cval[3];
527 if (p->p_stats == NULL) {
528 	printf("pstats null\n");
529 	bzero(cp, 8);
530 	return;
531 }
532 	tl.lval = p->p_stats->p_start.tv_sec;
533 	*cp++ = tl.cval[0];
534 	*cp++ = tl.cval[1];
535 	*cp++ = tl.cval[2];
536 	*cp++ = tl.cval[3];
537 	tl.lval = p->p_stats->p_start.tv_usec;
538 	*cp++ = tl.cval[0];
539 	*cp++ = tl.cval[1];
540 	*cp++ = tl.cval[2];
541 	*cp = tl.cval[3];
542 }
543 
544 /*
545  * Find the parent process for the thread passed in as an argument.
546  * If none exists, return NULL, otherwise return a thread for the parent.
547  * (Can be any of the threads, since it is only used for td->td_proc.)
548  */
549 NFSPROC_T *
550 nfscl_getparent(struct thread *td)
551 {
552 	struct proc *p;
553 	struct thread *ptd;
554 
555 	if (td == NULL)
556 		return (NULL);
557 	p = td->td_proc;
558 	if (p->p_pid == 0)
559 		return (NULL);
560 	p = p->p_pptr;
561 	if (p == NULL)
562 		return (NULL);
563 	ptd = TAILQ_FIRST(&p->p_threads);
564 	return (ptd);
565 }
566 
567 /*
568  * Start up the renew kernel thread.
569  */
570 static void
571 start_nfscl(void *arg)
572 {
573 	struct nfsclclient *clp;
574 	struct thread *td;
575 
576 	clp = (struct nfsclclient *)arg;
577 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
578 	nfscl_renewthread(clp, td);
579 	kproc_exit(0);
580 }
581 
582 void
583 nfscl_start_renewthread(struct nfsclclient *clp)
584 {
585 
586 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
587 	    "nfscl");
588 }
589 
590 /*
591  * Handle wcc_data.
592  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
593  * as the first Op after PutFH.
594  * (For NFSv4, the postop attributes are after the Op, so they can't be
595  *  parsed here. A separate call to nfscl_postop_attr() is required.)
596  */
597 int
598 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
599     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
600 {
601 	u_int32_t *tl;
602 	struct nfsnode *np = VTONFS(vp);
603 	struct nfsvattr nfsva;
604 	int error = 0;
605 
606 	if (wccflagp != NULL)
607 		*wccflagp = 0;
608 	if (nd->nd_flag & ND_NFSV3) {
609 		*flagp = 0;
610 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
611 		if (*tl == newnfs_true) {
612 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
613 			if (wccflagp != NULL) {
614 				mtx_lock(&np->n_mtx);
615 				*wccflagp = (np->n_mtime.tv_sec ==
616 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
617 				    np->n_mtime.tv_nsec ==
618 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
619 				mtx_unlock(&np->n_mtx);
620 			}
621 		}
622 		error = nfscl_postop_attr(nd, nap, flagp, stuff);
623 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
624 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
625 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
626 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
627 		    NULL, NULL, NULL, NULL, NULL);
628 		if (error)
629 			return (error);
630 		/*
631 		 * Get rid of Op# and status for next op.
632 		 */
633 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
634 		if (*++tl)
635 			nd->nd_flag |= ND_NOMOREDATA;
636 		if (wccflagp != NULL &&
637 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
638 			mtx_lock(&np->n_mtx);
639 			*wccflagp = (np->n_mtime.tv_sec ==
640 			    nfsva.na_vattr.va_mtime.tv_sec &&
641 			    np->n_mtime.tv_nsec ==
642 			    nfsva.na_vattr.va_mtime.tv_sec);
643 			mtx_unlock(&np->n_mtx);
644 		}
645 	}
646 nfsmout:
647 	return (error);
648 }
649 
650 /*
651  * Get postop attributes.
652  */
653 int
654 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
655     void *stuff)
656 {
657 	u_int32_t *tl;
658 	int error = 0;
659 
660 	*retp = 0;
661 	if (nd->nd_flag & ND_NOMOREDATA)
662 		return (error);
663 	if (nd->nd_flag & ND_NFSV3) {
664 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
665 		*retp = fxdr_unsigned(int, *tl);
666 	} else if (nd->nd_flag & ND_NFSV4) {
667 		/*
668 		 * For NFSv4, the postop attr are at the end, so no point
669 		 * in looking if nd_repstat != 0.
670 		 */
671 		if (!nd->nd_repstat) {
672 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
673 			if (*(tl + 1))
674 				/* should never happen since nd_repstat != 0 */
675 				nd->nd_flag |= ND_NOMOREDATA;
676 			else
677 				*retp = 1;
678 		}
679 	} else if (!nd->nd_repstat) {
680 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
681 		*retp = 1;
682 	}
683 	if (*retp) {
684 		error = nfsm_loadattr(nd, nap);
685 		if (error)
686 			*retp = 0;
687 	}
688 nfsmout:
689 	return (error);
690 }
691 
692 /*
693  * Fill in the setable attributes. The full argument indicates whether
694  * to fill in them all or just mode and time.
695  */
696 void
697 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
698     struct vnode *vp, int flags, u_int32_t rdev)
699 {
700 	u_int32_t *tl;
701 	struct nfsv2_sattr *sp;
702 	nfsattrbit_t attrbits;
703 	struct timeval curtime;
704 
705 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
706 	case ND_NFSV2:
707 		NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
708 		if (vap->va_mode == (mode_t)VNOVAL)
709 			sp->sa_mode = newnfs_xdrneg1;
710 		else
711 			sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
712 		if (vap->va_uid == (uid_t)VNOVAL)
713 			sp->sa_uid = newnfs_xdrneg1;
714 		else
715 			sp->sa_uid = txdr_unsigned(vap->va_uid);
716 		if (vap->va_gid == (gid_t)VNOVAL)
717 			sp->sa_gid = newnfs_xdrneg1;
718 		else
719 			sp->sa_gid = txdr_unsigned(vap->va_gid);
720 		if (flags & NFSSATTR_SIZE0)
721 			sp->sa_size = 0;
722 		else if (flags & NFSSATTR_SIZENEG1)
723 			sp->sa_size = newnfs_xdrneg1;
724 		else if (flags & NFSSATTR_SIZERDEV)
725 			sp->sa_size = txdr_unsigned(rdev);
726 		else
727 			sp->sa_size = txdr_unsigned(vap->va_size);
728 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
729 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
730 		break;
731 	case ND_NFSV3:
732 		getmicrotime(&curtime);
733 		if (vap->va_mode != (mode_t)VNOVAL) {
734 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
735 			*tl++ = newnfs_true;
736 			*tl = txdr_unsigned(vap->va_mode);
737 		} else {
738 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
739 			*tl = newnfs_false;
740 		}
741 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
742 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
743 			*tl++ = newnfs_true;
744 			*tl = txdr_unsigned(vap->va_uid);
745 		} else {
746 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
747 			*tl = newnfs_false;
748 		}
749 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
750 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
751 			*tl++ = newnfs_true;
752 			*tl = txdr_unsigned(vap->va_gid);
753 		} else {
754 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
755 			*tl = newnfs_false;
756 		}
757 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
758 			NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
759 			*tl++ = newnfs_true;
760 			txdr_hyper(vap->va_size, tl);
761 		} else {
762 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
763 			*tl = newnfs_false;
764 		}
765 		if (vap->va_atime.tv_sec != VNOVAL) {
766 			if (vap->va_atime.tv_sec != curtime.tv_sec) {
767 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
768 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
769 				txdr_nfsv3time(&vap->va_atime, tl);
770 			} else {
771 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
772 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
773 			}
774 		} else {
775 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
776 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
777 		}
778 		if (vap->va_mtime.tv_sec != VNOVAL) {
779 			if (vap->va_mtime.tv_sec != curtime.tv_sec) {
780 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
781 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
782 				txdr_nfsv3time(&vap->va_mtime, tl);
783 			} else {
784 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
785 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
786 			}
787 		} else {
788 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
789 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
790 		}
791 		break;
792 	case ND_NFSV4:
793 		NFSZERO_ATTRBIT(&attrbits);
794 		if (vap->va_mode != (mode_t)VNOVAL)
795 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
796 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
797 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
798 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
799 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
800 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
801 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
802 		if (vap->va_atime.tv_sec != VNOVAL)
803 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
804 		if (vap->va_mtime.tv_sec != VNOVAL)
805 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
806 		(void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
807 		    &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
808 		break;
809 	};
810 }
811 
812 /*
813  * nfscl_request() - mostly a wrapper for newnfs_request().
814  */
815 int
816 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
817     struct ucred *cred, void *stuff)
818 {
819 	int ret, vers;
820 	struct nfsmount *nmp;
821 
822 	nmp = VFSTONFS(vp->v_mount);
823 	if (nd->nd_flag & ND_NFSV4)
824 		vers = NFS_VER4;
825 	else if (nd->nd_flag & ND_NFSV3)
826 		vers = NFS_VER3;
827 	else
828 		vers = NFS_VER2;
829 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
830 		NFS_PROG, vers, NULL, 1, NULL);
831 	return (ret);
832 }
833 
834 /*
835  * fill in this bsden's variant of statfs using nfsstatfs.
836  */
837 void
838 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
839 {
840 	struct statfs *sbp = (struct statfs *)statfs;
841 
842 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
843 		sbp->f_bsize = NFS_FABLKSIZE;
844 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
845 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
846 		/*
847 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
848 		 * the value after dividing by NFS_FABLKSIZE is small
849 		 * enough that it will fit in 63bits, so it is ok to
850 		 * assign it to f_bavail without fear that it will become
851 		 * negative.
852 		 */
853 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
854 		sbp->f_files = sfp->sf_tfiles;
855 		/* Since f_ffree is int64_t, clip it to 63bits. */
856 		if (sfp->sf_ffiles > INT64_MAX)
857 			sbp->f_ffree = INT64_MAX;
858 		else
859 			sbp->f_ffree = sfp->sf_ffiles;
860 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
861 		/*
862 		 * The type casts to (int32_t) ensure that this code is
863 		 * compatible with the old NFS client, in that it will
864 		 * propagate bit31 to the high order bits. This may or may
865 		 * not be correct for NFSv2, but since it is a legacy
866 		 * environment, I'd rather retain backwards compatibility.
867 		 */
868 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
869 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
870 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
871 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
872 		sbp->f_files = 0;
873 		sbp->f_ffree = 0;
874 	}
875 }
876 
877 /*
878  * Use the fsinfo stuff to update the mount point.
879  */
880 void
881 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
882 {
883 
884 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
885 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
886 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
887 		    ~(NFS_FABLKSIZE - 1);
888 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
889 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
890 		if (nmp->nm_wsize == 0)
891 			nmp->nm_wsize = fsp->fs_wtmax;
892 	}
893 	if (nmp->nm_wsize < NFS_FABLKSIZE)
894 		nmp->nm_wsize = NFS_FABLKSIZE;
895 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
896 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
897 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
898 		    ~(NFS_FABLKSIZE - 1);
899 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
900 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
901 		if (nmp->nm_rsize == 0)
902 			nmp->nm_rsize = fsp->fs_rtmax;
903 	}
904 	if (nmp->nm_rsize < NFS_FABLKSIZE)
905 		nmp->nm_rsize = NFS_FABLKSIZE;
906 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
907 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
908 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
909 		    ~(NFS_DIRBLKSIZ - 1);
910 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
911 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
912 		if (nmp->nm_readdirsize == 0)
913 			nmp->nm_readdirsize = fsp->fs_rtmax;
914 	}
915 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
916 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
917 	if (fsp->fs_maxfilesize > 0 &&
918 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
919 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
920 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
921 	nmp->nm_state |= NFSSTA_GOTFSINFO;
922 }
923 
924 /*
925  * Get a pointer to my IP addrress and return it.
926  * Return NULL if you can't find one.
927  */
928 u_int8_t *
929 nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
930 {
931 	struct sockaddr_in sad, *sin;
932 	struct rtentry *rt;
933 	u_int8_t *retp = NULL;
934 	static struct in_addr laddr;
935 
936 	*isinet6p = 0;
937 	/*
938 	 * Loop up a route for the destination address.
939 	 */
940 	if (nmp->nm_nam->sa_family == AF_INET) {
941 		bzero(&sad, sizeof (sad));
942 		sin = (struct sockaddr_in *)nmp->nm_nam;
943 		sad.sin_family = AF_INET;
944 		sad.sin_len = sizeof (struct sockaddr_in);
945 		sad.sin_addr.s_addr = sin->sin_addr.s_addr;
946 		rt = rtalloc1((struct sockaddr *)&sad, 0, 0UL);
947 		if (rt != NULL) {
948 			if (rt->rt_ifp != NULL &&
949 			    rt->rt_ifa != NULL &&
950 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
951 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
952 				sin = (struct sockaddr_in *)
953 				    rt->rt_ifa->ifa_addr;
954 				laddr.s_addr = sin->sin_addr.s_addr;
955 				retp = (u_int8_t *)&laddr;
956 			}
957 			RTFREE_LOCKED(rt);
958 		}
959 #ifdef INET6
960 	} else if (nmp->nm_nam->sa_family == AF_INET6) {
961 		struct sockaddr_in6 sad6, *sin6;
962 		static struct in6_addr laddr6;
963 
964 		bzero(&sad6, sizeof (sad6));
965 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
966 		sad6.sin6_family = AF_INET6;
967 		sad6.sin6_len = sizeof (struct sockaddr_in6);
968 		sad6.sin6_addr = sin6->sin6_addr;
969 		rt = rtalloc1((struct sockaddr *)&sad6, 0, 0UL);
970 		if (rt != NULL) {
971 			if (rt->rt_ifp != NULL &&
972 			    rt->rt_ifa != NULL &&
973 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
974 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
975 				sin6 = (struct sockaddr_in6 *)
976 				    rt->rt_ifa->ifa_addr;
977 				laddr6 = sin6->sin6_addr;
978 				retp = (u_int8_t *)&laddr6;
979 				*isinet6p = 1;
980 			}
981 			RTFREE_LOCKED(rt);
982 		}
983 #endif
984 	}
985 	return (retp);
986 }
987 
988 /*
989  * Copy NFS uid, gids from the cred structure.
990  */
991 void
992 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
993 {
994 	int i;
995 
996 	KASSERT(cr->cr_ngroups >= 0,
997 	    ("newnfs_copyincred: negative cr_ngroups"));
998 	nfscr->nfsc_uid = cr->cr_uid;
999 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1000 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1001 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1002 }
1003 
1004 
1005 /*
1006  * Do any client specific initialization.
1007  */
1008 void
1009 nfscl_init(void)
1010 {
1011 	static int inited = 0;
1012 
1013 	if (inited)
1014 		return;
1015 	inited = 1;
1016 	nfscl_inited = 1;
1017 	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1018 }
1019 
1020 /*
1021  * Check each of the attributes to be set, to ensure they aren't already
1022  * the correct value. Disable setting ones already correct.
1023  */
1024 int
1025 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1026 {
1027 
1028 	if (vap->va_mode != (mode_t)VNOVAL) {
1029 		if (vap->va_mode == nvap->na_mode)
1030 			vap->va_mode = (mode_t)VNOVAL;
1031 	}
1032 	if (vap->va_uid != (uid_t)VNOVAL) {
1033 		if (vap->va_uid == nvap->na_uid)
1034 			vap->va_uid = (uid_t)VNOVAL;
1035 	}
1036 	if (vap->va_gid != (gid_t)VNOVAL) {
1037 		if (vap->va_gid == nvap->na_gid)
1038 			vap->va_gid = (gid_t)VNOVAL;
1039 	}
1040 	if (vap->va_size != VNOVAL) {
1041 		if (vap->va_size == nvap->na_size)
1042 			vap->va_size = VNOVAL;
1043 	}
1044 
1045 	/*
1046 	 * We are normally called with only a partially initialized
1047 	 * VAP.  Since the NFSv3 spec says that server may use the
1048 	 * file attributes to store the verifier, the spec requires
1049 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1050 	 * in atime, but we can't really assume that all servers will
1051 	 * so we ensure that our SETATTR sets both atime and mtime.
1052 	 */
1053 	if (vap->va_mtime.tv_sec == VNOVAL)
1054 		vfs_timestamp(&vap->va_mtime);
1055 	if (vap->va_atime.tv_sec == VNOVAL)
1056 		vap->va_atime = vap->va_mtime;
1057 	return (1);
1058 }
1059 
1060 /*
1061  * Map nfsv4 errors to errno.h errors.
1062  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1063  * error should only be returned for the Open, Create and Setattr Ops.
1064  * As such, most calls can just pass in 0 for those arguments.
1065  */
1066 APPLESTATIC int
1067 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1068 {
1069 	struct proc *p;
1070 
1071 	if (error < 10000)
1072 		return (error);
1073 	if (td != NULL)
1074 		p = td->td_proc;
1075 	else
1076 		p = NULL;
1077 	switch (error) {
1078 	case NFSERR_BADOWNER:
1079 		tprintf(p, LOG_INFO,
1080 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1081 		    uid, gid);
1082 		return (EPERM);
1083 	case NFSERR_STALECLIENTID:
1084 	case NFSERR_STALESTATEID:
1085 	case NFSERR_EXPIRED:
1086 	case NFSERR_BADSTATEID:
1087 		printf("nfsv4 recover err returned %d\n", error);
1088 		return (EIO);
1089 	case NFSERR_BADHANDLE:
1090 	case NFSERR_SERVERFAULT:
1091 	case NFSERR_BADTYPE:
1092 	case NFSERR_FHEXPIRED:
1093 	case NFSERR_RESOURCE:
1094 	case NFSERR_MOVED:
1095 	case NFSERR_NOFILEHANDLE:
1096 	case NFSERR_MINORVERMISMATCH:
1097 	case NFSERR_OLDSTATEID:
1098 	case NFSERR_BADSEQID:
1099 	case NFSERR_LEASEMOVED:
1100 	case NFSERR_RECLAIMBAD:
1101 	case NFSERR_BADXDR:
1102 	case NFSERR_BADCHAR:
1103 	case NFSERR_BADNAME:
1104 	case NFSERR_OPILLEGAL:
1105 		printf("nfsv4 client/server protocol prob err=%d\n",
1106 		    error);
1107 		return (EIO);
1108 	default:
1109 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1110 		return (EIO);
1111 	};
1112 }
1113 
1114 /*
1115  * Locate a process by number; return only "live" processes -- i.e., neither
1116  * zombies nor newly born but incompletely initialized processes.  By not
1117  * returning processes in the PRS_NEW state, we allow callers to avoid
1118  * testing for that condition to avoid dereferencing p_ucred, et al.
1119  * Identical to pfind() in kern_proc.c, except it assume the list is
1120  * already locked.
1121  */
1122 static struct proc *
1123 pfind_locked(pid_t pid)
1124 {
1125 	struct proc *p;
1126 
1127 	LIST_FOREACH(p, PIDHASH(pid), p_hash)
1128 		if (p->p_pid == pid) {
1129 			PROC_LOCK(p);
1130 			if (p->p_state == PRS_NEW) {
1131 				PROC_UNLOCK(p);
1132 				p = NULL;
1133 			}
1134 			break;
1135 		}
1136 	return (p);
1137 }
1138 
1139 /*
1140  * Check to see if the process for this owner exists. Return 1 if it doesn't
1141  * and 0 otherwise.
1142  */
1143 int
1144 nfscl_procdoesntexist(u_int8_t *own)
1145 {
1146 	union {
1147 		u_int32_t	lval;
1148 		u_int8_t	cval[4];
1149 	} tl;
1150 	struct proc *p;
1151 	pid_t pid;
1152 	int ret = 0;
1153 
1154 	tl.cval[0] = *own++;
1155 	tl.cval[1] = *own++;
1156 	tl.cval[2] = *own++;
1157 	tl.cval[3] = *own++;
1158 	pid = tl.lval;
1159 	p = pfind_locked(pid);
1160 	if (p == NULL)
1161 		return (1);
1162 	if (p->p_stats == NULL) {
1163 		PROC_UNLOCK(p);
1164 		return (0);
1165 	}
1166 	tl.cval[0] = *own++;
1167 	tl.cval[1] = *own++;
1168 	tl.cval[2] = *own++;
1169 	tl.cval[3] = *own++;
1170 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1171 		ret = 1;
1172 	} else {
1173 		tl.cval[0] = *own++;
1174 		tl.cval[1] = *own++;
1175 		tl.cval[2] = *own++;
1176 		tl.cval[3] = *own;
1177 		if (tl.lval != p->p_stats->p_start.tv_usec)
1178 			ret = 1;
1179 	}
1180 	PROC_UNLOCK(p);
1181 	return (ret);
1182 }
1183 
1184 /*
1185  * - nfs pseudo system call for the client
1186  */
1187 /*
1188  * MPSAFE
1189  */
1190 static int
1191 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1192 {
1193 	struct file *fp;
1194 	struct nfscbd_args nfscbdarg;
1195 	struct nfsd_nfscbd_args nfscbdarg2;
1196 	int error;
1197 
1198 	if (uap->flag & NFSSVC_CBADDSOCK) {
1199 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1200 		if (error)
1201 			return (error);
1202 		if ((error = fget(td, nfscbdarg.sock, &fp)) != 0) {
1203 			return (error);
1204 		}
1205 		if (fp->f_type != DTYPE_SOCKET) {
1206 			fdrop(fp, td);
1207 			return (EPERM);
1208 		}
1209 		error = nfscbd_addsock(fp);
1210 		fdrop(fp, td);
1211 		if (!error && nfscl_enablecallb == 0) {
1212 			nfsv4_cbport = nfscbdarg.port;
1213 			nfscl_enablecallb = 1;
1214 		}
1215 	} else if (uap->flag & NFSSVC_NFSCBD) {
1216 		if (uap->argp == NULL)
1217 			return (EINVAL);
1218 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1219 		    sizeof(nfscbdarg2));
1220 		if (error)
1221 			return (error);
1222 		error = nfscbd_nfsd(td, &nfscbdarg2);
1223 	} else {
1224 		error = EINVAL;
1225 	}
1226 	return (error);
1227 }
1228 
1229 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1230 
1231 /*
1232  * Called once to initialize data structures...
1233  */
1234 static int
1235 nfscl_modevent(module_t mod, int type, void *data)
1236 {
1237 	int error = 0;
1238 	static int loaded = 0;
1239 
1240 	switch (type) {
1241 	case MOD_LOAD:
1242 		if (loaded)
1243 			return (0);
1244 		newnfs_portinit();
1245 		mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1246 		    MTX_DEF);
1247 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1248 		nfscl_init();
1249 		NFSD_LOCK();
1250 		nfsrvd_cbinit(0);
1251 		NFSD_UNLOCK();
1252 		ncl_call_invalcaches = ncl_invalcaches;
1253 		nfsd_call_nfscl = nfssvc_nfscl;
1254 		loaded = 1;
1255 		break;
1256 
1257 	case MOD_UNLOAD:
1258 		if (nfs_numnfscbd != 0) {
1259 			error = EBUSY;
1260 			break;
1261 		}
1262 
1263 		/*
1264 		 * XXX: Unloading of nfscl module is unsupported.
1265 		 */
1266 #if 0
1267 		ncl_call_invalcaches = NULL;
1268 		nfsd_call_nfscl = NULL;
1269 		/* and get rid of the mutexes */
1270 		mtx_destroy(&nfs_clstate_mutex);
1271 		mtx_destroy(&ncl_iod_mutex);
1272 		loaded = 0;
1273 		break;
1274 #else
1275 		/* FALLTHROUGH */
1276 #endif
1277 	default:
1278 		error = EOPNOTSUPP;
1279 		break;
1280 	}
1281 	return error;
1282 }
1283 static moduledata_t nfscl_mod = {
1284 	"nfscl",
1285 	nfscl_modevent,
1286 	NULL,
1287 };
1288 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1289 
1290 /* So that loader and kldload(2) can find us, wherever we are.. */
1291 MODULE_VERSION(nfscl, 1);
1292 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1293 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1294 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1295 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1296 
1297