xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision 9a14aa017b21c292740c00ee098195cd46642730)
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 
201 	/*
202 	 * Allocate before getnewvnode since doing so afterward
203 	 * might cause a bogus v_data pointer to get dereferenced
204 	 * elsewhere if zalloc should block.
205 	 */
206 	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
207 
208 	error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
209 	if (error) {
210 		uma_zfree(newnfsnode_zone, np);
211 		FREE((caddr_t)nfhp, M_NFSFH);
212 		return (error);
213 	}
214 	vp = nvp;
215 	vp->v_bufobj.bo_ops = &buf_ops_newnfs;
216 	vp->v_data = np;
217 	np->n_vnode = vp;
218 	/*
219 	 * Initialize the mutex even if the vnode is going to be a loser.
220 	 * This simplifies the logic in reclaim, which can then unconditionally
221 	 * destroy the mutex (in the case of the loser, or if hash_insert
222 	 * happened to return an error no special casing is needed).
223 	 */
224 	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
225 
226 	/*
227 	 * Are we getting the root? If so, make sure the vnode flags
228 	 * are correct
229 	 */
230 	if ((nfhp->nfh_len == nmp->nm_fhsize) &&
231 	    !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
232 		if (vp->v_type == VNON)
233 			vp->v_type = VDIR;
234 		vp->v_vflag |= VV_ROOT;
235 	}
236 
237 	np->n_fhp = nfhp;
238 	/*
239 	 * For NFSv4, we have to attach the directory file handle and
240 	 * file name, so that Open Ops can be done later.
241 	 */
242 	if (nmp->nm_flag & NFSMNT_NFSV4) {
243 		MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
244 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
245 		    M_WAITOK);
246 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
247 		np->n_v4->n4_namelen = cnp->cn_namelen;
248 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
249 		    dnp->n_fhp->nfh_len);
250 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
251 		    cnp->cn_namelen);
252 	} else {
253 		np->n_v4 = NULL;
254 	}
255 
256 	/*
257 	 * NFS supports recursive and shared locking.
258 	 */
259 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
260 	VN_LOCK_AREC(vp);
261 	VN_LOCK_ASHARE(vp);
262 	error = insmntque(vp, mntp);
263 	if (error != 0) {
264 		*npp = NULL;
265 		mtx_destroy(&np->n_mtx);
266 		FREE((caddr_t)nfhp, M_NFSFH);
267 		if (np->n_v4 != NULL)
268 			FREE((caddr_t)np->n_v4, M_NFSV4NODE);
269 		uma_zfree(newnfsnode_zone, np);
270 		return (error);
271 	}
272 	error = vfs_hash_insert(vp, hash, lkflags,
273 	    td, &nvp, newnfs_vncmpf, nfhp);
274 	if (error)
275 		return (error);
276 	if (nvp != NULL) {
277 		*npp = VTONFS(nvp);
278 		/* vfs_hash_insert() vput()'s the losing vnode */
279 		return (0);
280 	}
281 	*npp = np;
282 
283 	return (0);
284 }
285 
286 /*
287  * Anothe variant of nfs_nget(). This one is only used by reopen. It
288  * takes almost the same args as nfs_nget(), but only succeeds if an entry
289  * exists in the cache. (Since files should already be "open" with a
290  * vnode ref cnt on the node when reopen calls this, it should always
291  * succeed.)
292  * Also, don't get a vnode lock, since it may already be locked by some
293  * other process that is handling it. This is ok, since all other threads
294  * on the client are blocked by the nfsc_lock being exclusively held by the
295  * caller of this function.
296  */
297 int
298 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
299     struct thread *td, struct nfsnode **npp)
300 {
301 	struct vnode *nvp;
302 	u_int hash;
303 	struct nfsfh *nfhp;
304 	int error;
305 
306 	*npp = NULL;
307 	/* For forced dismounts, just return error. */
308 	if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
309 		return (EINTR);
310 	MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
311 	    M_NFSFH, M_WAITOK);
312 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
313 	nfhp->nfh_len = fhsize;
314 
315 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
316 
317 	/*
318 	 * First, try to get the vnode locked, but don't block for the lock.
319 	 */
320 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
321 	    newnfs_vncmpf, nfhp);
322 	if (error == 0 && nvp != NULL) {
323 		NFSVOPUNLOCK(nvp, 0);
324 	} else if (error == EBUSY) {
325 		/*
326 		 * The LK_EXCLOTHER lock type tells nfs_lock1() to not try
327 		 * and lock the vnode, but just get a v_usecount on it.
328 		 * LK_NOWAIT is set so that when vget() returns ENOENT,
329 		 * vfs_hash_get() fails instead of looping.
330 		 * If this succeeds, it is safe so long as a vflush() with
331 		 * FORCECLOSE has not been done. Since the Renew thread is
332 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
333 		 * a vflush() with FORCECLOSE, we should be ok here.
334 		 */
335 		if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
336 			error = EINTR;
337 		else
338 			error = vfs_hash_get(mntp, hash,
339 			    (LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
340 			    newnfs_vncmpf, nfhp);
341 	}
342 	FREE(nfhp, M_NFSFH);
343 	if (error)
344 		return (error);
345 	if (nvp != NULL) {
346 		*npp = VTONFS(nvp);
347 		return (0);
348 	}
349 	return (EINVAL);
350 }
351 
352 /*
353  * Load the attribute cache (that lives in the nfsnode entry) with
354  * the attributes of the second argument and
355  * Iff vaper not NULL
356  *    copy the attributes to *vaper
357  * Similar to nfs_loadattrcache(), except the attributes are passed in
358  * instead of being parsed out of the mbuf list.
359  */
360 int
361 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
362     void *stuff, int writeattr, int dontshrink)
363 {
364 	struct vnode *vp = *vpp;
365 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
366 	struct nfsnode *np;
367 	struct nfsmount *nmp;
368 	struct timespec mtime_save;
369 
370 	/*
371 	 * If v_type == VNON it is a new node, so fill in the v_type,
372 	 * n_mtime fields. Check to see if it represents a special
373 	 * device, and if so, check for a possible alias. Once the
374 	 * correct vnode has been obtained, fill in the rest of the
375 	 * information.
376 	 */
377 	np = VTONFS(vp);
378 	NFSLOCKNODE(np);
379 	if (vp->v_type != nvap->va_type) {
380 		vp->v_type = nvap->va_type;
381 		if (vp->v_type == VFIFO)
382 			vp->v_op = &newnfs_fifoops;
383 		np->n_mtime = nvap->va_mtime;
384 	}
385 	nmp = VFSTONFS(vp->v_mount);
386 	vap = &np->n_vattr.na_vattr;
387 	mtime_save = vap->va_mtime;
388 	if (writeattr) {
389 		np->n_vattr.na_filerev = nap->na_filerev;
390 		np->n_vattr.na_size = nap->na_size;
391 		np->n_vattr.na_mtime = nap->na_mtime;
392 		np->n_vattr.na_ctime = nap->na_ctime;
393 		np->n_vattr.na_fsid = nap->na_fsid;
394 		np->n_vattr.na_mode = nap->na_mode;
395 	} else {
396 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
397 		    sizeof (struct nfsvattr));
398 	}
399 
400 	/*
401 	 * For NFSv4, if the node's fsid is not equal to the mount point's
402 	 * fsid, return the low order 32bits of the node's fsid. This
403 	 * allows getcwd(3) to work. There is a chance that the fsid might
404 	 * be the same as a local fs, but since this is in an NFS mount
405 	 * point, I don't think that will cause any problems?
406 	 */
407 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
408 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
409 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
410 		/*
411 		 * va_fsid needs to be set to some value derived from
412 		 * np->n_vattr.na_filesid that is not equal
413 		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
414 		 * from the value used for the top level server volume
415 		 * in the mounted subtree.
416 		 */
417 		if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
418 		    (uint32_t)np->n_vattr.na_filesid[0])
419 			vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
420 		else
421 			vap->va_fsid = (uint32_t)hash32_buf(
422 			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
423 	} else
424 		vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
425 	np->n_attrstamp = time_second;
426 	if (vap->va_size != np->n_size) {
427 		if (vap->va_type == VREG) {
428 			if (dontshrink && vap->va_size < np->n_size) {
429 				/*
430 				 * We've been told not to shrink the file;
431 				 * zero np->n_attrstamp to indicate that
432 				 * the attributes are stale.
433 				 */
434 				vap->va_size = np->n_size;
435 				np->n_attrstamp = 0;
436 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
437 			} else if (np->n_flag & NMODIFIED) {
438 				/*
439 				 * We've modified the file: Use the larger
440 				 * of our size, and the server's size.
441 				 */
442 				if (vap->va_size < np->n_size) {
443 					vap->va_size = np->n_size;
444 				} else {
445 					np->n_size = vap->va_size;
446 					np->n_flag |= NSIZECHANGED;
447 				}
448 			} else {
449 				np->n_size = vap->va_size;
450 				np->n_flag |= NSIZECHANGED;
451 			}
452 			vnode_pager_setsize(vp, np->n_size);
453 		} else {
454 			np->n_size = vap->va_size;
455 		}
456 	}
457 	/*
458 	 * The following checks are added to prevent a race between (say)
459 	 * a READDIR+ and a WRITE.
460 	 * READDIR+, WRITE requests sent out.
461 	 * READDIR+ resp, WRITE resp received on client.
462 	 * However, the WRITE resp was handled before the READDIR+ resp
463 	 * causing the post op attrs from the write to be loaded first
464 	 * and the attrs from the READDIR+ to be loaded later. If this
465 	 * happens, we have stale attrs loaded into the attrcache.
466 	 * We detect this by for the mtime moving back. We invalidate the
467 	 * attrcache when this happens.
468 	 */
469 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
470 		/* Size changed or mtime went backwards */
471 		np->n_attrstamp = 0;
472 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
473 	}
474 	if (vaper != NULL) {
475 		NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
476 		if (np->n_flag & NCHG) {
477 			if (np->n_flag & NACC)
478 				vaper->va_atime = np->n_atim;
479 			if (np->n_flag & NUPD)
480 				vaper->va_mtime = np->n_mtim;
481 		}
482 	}
483 #ifdef KDTRACE_HOOKS
484 	if (np->n_attrstamp != 0)
485 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, 0);
486 #endif
487 	NFSUNLOCKNODE(np);
488 	return (0);
489 }
490 
491 /*
492  * Fill in the client id name. For these bytes:
493  * 1 - they must be unique
494  * 2 - they should be persistent across client reboots
495  * 1 is more critical than 2
496  * Use the mount point's unique id plus either the uuid or, if that
497  * isn't set, random junk.
498  */
499 void
500 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
501 {
502 	int uuidlen;
503 
504 	/*
505 	 * First, put in the 64bit mount point identifier.
506 	 */
507 	if (idlen >= sizeof (u_int64_t)) {
508 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
509 		cp += sizeof (u_int64_t);
510 		idlen -= sizeof (u_int64_t);
511 	}
512 
513 	/*
514 	 * If uuid is non-zero length, use it.
515 	 */
516 	uuidlen = strlen(uuid);
517 	if (uuidlen > 0 && idlen >= uuidlen) {
518 		NFSBCOPY(uuid, cp, uuidlen);
519 		cp += uuidlen;
520 		idlen -= uuidlen;
521 	}
522 
523 	/*
524 	 * This only normally happens if the uuid isn't set.
525 	 */
526 	while (idlen > 0) {
527 		*cp++ = (u_int8_t)(arc4random() % 256);
528 		idlen--;
529 	}
530 }
531 
532 /*
533  * Fill in a lock owner name. For now, pid + the process's creation time.
534  */
535 void
536 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
537 {
538 	union {
539 		u_int32_t	lval;
540 		u_int8_t	cval[4];
541 	} tl;
542 	struct proc *p;
543 
544 	if (id == NULL) {
545 		printf("NULL id\n");
546 		bzero(cp, NFSV4CL_LOCKNAMELEN);
547 		return;
548 	}
549 	if ((flags & F_POSIX) != 0) {
550 		p = (struct proc *)id;
551 		tl.lval = p->p_pid;
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_sec;
557 		*cp++ = tl.cval[0];
558 		*cp++ = tl.cval[1];
559 		*cp++ = tl.cval[2];
560 		*cp++ = tl.cval[3];
561 		tl.lval = p->p_stats->p_start.tv_usec;
562 		*cp++ = tl.cval[0];
563 		*cp++ = tl.cval[1];
564 		*cp++ = tl.cval[2];
565 		*cp = tl.cval[3];
566 	} else if ((flags & F_FLOCK) != 0) {
567 		bcopy(&id, cp, sizeof(id));
568 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
569 	} else {
570 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
571 		bzero(cp, NFSV4CL_LOCKNAMELEN);
572 	}
573 }
574 
575 /*
576  * Find the parent process for the thread passed in as an argument.
577  * If none exists, return NULL, otherwise return a thread for the parent.
578  * (Can be any of the threads, since it is only used for td->td_proc.)
579  */
580 NFSPROC_T *
581 nfscl_getparent(struct thread *td)
582 {
583 	struct proc *p;
584 	struct thread *ptd;
585 
586 	if (td == NULL)
587 		return (NULL);
588 	p = td->td_proc;
589 	if (p->p_pid == 0)
590 		return (NULL);
591 	p = p->p_pptr;
592 	if (p == NULL)
593 		return (NULL);
594 	ptd = TAILQ_FIRST(&p->p_threads);
595 	return (ptd);
596 }
597 
598 /*
599  * Start up the renew kernel thread.
600  */
601 static void
602 start_nfscl(void *arg)
603 {
604 	struct nfsclclient *clp;
605 	struct thread *td;
606 
607 	clp = (struct nfsclclient *)arg;
608 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
609 	nfscl_renewthread(clp, td);
610 	kproc_exit(0);
611 }
612 
613 void
614 nfscl_start_renewthread(struct nfsclclient *clp)
615 {
616 
617 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
618 	    "nfscl");
619 }
620 
621 /*
622  * Handle wcc_data.
623  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
624  * as the first Op after PutFH.
625  * (For NFSv4, the postop attributes are after the Op, so they can't be
626  *  parsed here. A separate call to nfscl_postop_attr() is required.)
627  */
628 int
629 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
630     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
631 {
632 	u_int32_t *tl;
633 	struct nfsnode *np = VTONFS(vp);
634 	struct nfsvattr nfsva;
635 	int error = 0;
636 
637 	if (wccflagp != NULL)
638 		*wccflagp = 0;
639 	if (nd->nd_flag & ND_NFSV3) {
640 		*flagp = 0;
641 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
642 		if (*tl == newnfs_true) {
643 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
644 			if (wccflagp != NULL) {
645 				mtx_lock(&np->n_mtx);
646 				*wccflagp = (np->n_mtime.tv_sec ==
647 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
648 				    np->n_mtime.tv_nsec ==
649 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
650 				mtx_unlock(&np->n_mtx);
651 			}
652 		}
653 		error = nfscl_postop_attr(nd, nap, flagp, stuff);
654 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
655 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
656 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
657 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
658 		    NULL, NULL, NULL, NULL, NULL);
659 		if (error)
660 			return (error);
661 		/*
662 		 * Get rid of Op# and status for next op.
663 		 */
664 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
665 		if (*++tl)
666 			nd->nd_flag |= ND_NOMOREDATA;
667 		if (wccflagp != NULL &&
668 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
669 			mtx_lock(&np->n_mtx);
670 			*wccflagp = (np->n_mtime.tv_sec ==
671 			    nfsva.na_vattr.va_mtime.tv_sec &&
672 			    np->n_mtime.tv_nsec ==
673 			    nfsva.na_vattr.va_mtime.tv_sec);
674 			mtx_unlock(&np->n_mtx);
675 		}
676 	}
677 nfsmout:
678 	return (error);
679 }
680 
681 /*
682  * Get postop attributes.
683  */
684 int
685 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
686     void *stuff)
687 {
688 	u_int32_t *tl;
689 	int error = 0;
690 
691 	*retp = 0;
692 	if (nd->nd_flag & ND_NOMOREDATA)
693 		return (error);
694 	if (nd->nd_flag & ND_NFSV3) {
695 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
696 		*retp = fxdr_unsigned(int, *tl);
697 	} else if (nd->nd_flag & ND_NFSV4) {
698 		/*
699 		 * For NFSv4, the postop attr are at the end, so no point
700 		 * in looking if nd_repstat != 0.
701 		 */
702 		if (!nd->nd_repstat) {
703 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
704 			if (*(tl + 1))
705 				/* should never happen since nd_repstat != 0 */
706 				nd->nd_flag |= ND_NOMOREDATA;
707 			else
708 				*retp = 1;
709 		}
710 	} else if (!nd->nd_repstat) {
711 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
712 		*retp = 1;
713 	}
714 	if (*retp) {
715 		error = nfsm_loadattr(nd, nap);
716 		if (error)
717 			*retp = 0;
718 	}
719 nfsmout:
720 	return (error);
721 }
722 
723 /*
724  * Fill in the setable attributes. The full argument indicates whether
725  * to fill in them all or just mode and time.
726  */
727 void
728 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
729     struct vnode *vp, int flags, u_int32_t rdev)
730 {
731 	u_int32_t *tl;
732 	struct nfsv2_sattr *sp;
733 	nfsattrbit_t attrbits;
734 	struct timeval curtime;
735 
736 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
737 	case ND_NFSV2:
738 		NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
739 		if (vap->va_mode == (mode_t)VNOVAL)
740 			sp->sa_mode = newnfs_xdrneg1;
741 		else
742 			sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
743 		if (vap->va_uid == (uid_t)VNOVAL)
744 			sp->sa_uid = newnfs_xdrneg1;
745 		else
746 			sp->sa_uid = txdr_unsigned(vap->va_uid);
747 		if (vap->va_gid == (gid_t)VNOVAL)
748 			sp->sa_gid = newnfs_xdrneg1;
749 		else
750 			sp->sa_gid = txdr_unsigned(vap->va_gid);
751 		if (flags & NFSSATTR_SIZE0)
752 			sp->sa_size = 0;
753 		else if (flags & NFSSATTR_SIZENEG1)
754 			sp->sa_size = newnfs_xdrneg1;
755 		else if (flags & NFSSATTR_SIZERDEV)
756 			sp->sa_size = txdr_unsigned(rdev);
757 		else
758 			sp->sa_size = txdr_unsigned(vap->va_size);
759 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
760 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
761 		break;
762 	case ND_NFSV3:
763 		getmicrotime(&curtime);
764 		if (vap->va_mode != (mode_t)VNOVAL) {
765 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
766 			*tl++ = newnfs_true;
767 			*tl = txdr_unsigned(vap->va_mode);
768 		} else {
769 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
770 			*tl = newnfs_false;
771 		}
772 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
773 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
774 			*tl++ = newnfs_true;
775 			*tl = txdr_unsigned(vap->va_uid);
776 		} else {
777 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
778 			*tl = newnfs_false;
779 		}
780 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
781 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
782 			*tl++ = newnfs_true;
783 			*tl = txdr_unsigned(vap->va_gid);
784 		} else {
785 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
786 			*tl = newnfs_false;
787 		}
788 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
789 			NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
790 			*tl++ = newnfs_true;
791 			txdr_hyper(vap->va_size, tl);
792 		} else {
793 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
794 			*tl = newnfs_false;
795 		}
796 		if (vap->va_atime.tv_sec != VNOVAL) {
797 			if (vap->va_atime.tv_sec != curtime.tv_sec) {
798 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
799 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
800 				txdr_nfsv3time(&vap->va_atime, tl);
801 			} else {
802 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
803 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
804 			}
805 		} else {
806 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
807 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
808 		}
809 		if (vap->va_mtime.tv_sec != VNOVAL) {
810 			if (vap->va_mtime.tv_sec != curtime.tv_sec) {
811 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
812 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
813 				txdr_nfsv3time(&vap->va_mtime, tl);
814 			} else {
815 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
816 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
817 			}
818 		} else {
819 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
820 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
821 		}
822 		break;
823 	case ND_NFSV4:
824 		NFSZERO_ATTRBIT(&attrbits);
825 		if (vap->va_mode != (mode_t)VNOVAL)
826 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
827 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
828 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
829 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
830 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
831 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
832 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
833 		if (vap->va_atime.tv_sec != VNOVAL)
834 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
835 		if (vap->va_mtime.tv_sec != VNOVAL)
836 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
837 		(void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
838 		    &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
839 		break;
840 	};
841 }
842 
843 /*
844  * nfscl_request() - mostly a wrapper for newnfs_request().
845  */
846 int
847 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
848     struct ucred *cred, void *stuff)
849 {
850 	int ret, vers;
851 	struct nfsmount *nmp;
852 
853 	nmp = VFSTONFS(vp->v_mount);
854 	if (nd->nd_flag & ND_NFSV4)
855 		vers = NFS_VER4;
856 	else if (nd->nd_flag & ND_NFSV3)
857 		vers = NFS_VER3;
858 	else
859 		vers = NFS_VER2;
860 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
861 		NFS_PROG, vers, NULL, 1, NULL);
862 	return (ret);
863 }
864 
865 /*
866  * fill in this bsden's variant of statfs using nfsstatfs.
867  */
868 void
869 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
870 {
871 	struct statfs *sbp = (struct statfs *)statfs;
872 
873 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
874 		sbp->f_bsize = NFS_FABLKSIZE;
875 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
876 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
877 		/*
878 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
879 		 * the value after dividing by NFS_FABLKSIZE is small
880 		 * enough that it will fit in 63bits, so it is ok to
881 		 * assign it to f_bavail without fear that it will become
882 		 * negative.
883 		 */
884 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
885 		sbp->f_files = sfp->sf_tfiles;
886 		/* Since f_ffree is int64_t, clip it to 63bits. */
887 		if (sfp->sf_ffiles > INT64_MAX)
888 			sbp->f_ffree = INT64_MAX;
889 		else
890 			sbp->f_ffree = sfp->sf_ffiles;
891 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
892 		/*
893 		 * The type casts to (int32_t) ensure that this code is
894 		 * compatible with the old NFS client, in that it will
895 		 * propagate bit31 to the high order bits. This may or may
896 		 * not be correct for NFSv2, but since it is a legacy
897 		 * environment, I'd rather retain backwards compatibility.
898 		 */
899 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
900 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
901 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
902 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
903 		sbp->f_files = 0;
904 		sbp->f_ffree = 0;
905 	}
906 }
907 
908 /*
909  * Use the fsinfo stuff to update the mount point.
910  */
911 void
912 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
913 {
914 
915 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
916 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
917 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
918 		    ~(NFS_FABLKSIZE - 1);
919 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
920 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
921 		if (nmp->nm_wsize == 0)
922 			nmp->nm_wsize = fsp->fs_wtmax;
923 	}
924 	if (nmp->nm_wsize < NFS_FABLKSIZE)
925 		nmp->nm_wsize = NFS_FABLKSIZE;
926 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
927 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
928 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
929 		    ~(NFS_FABLKSIZE - 1);
930 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
931 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
932 		if (nmp->nm_rsize == 0)
933 			nmp->nm_rsize = fsp->fs_rtmax;
934 	}
935 	if (nmp->nm_rsize < NFS_FABLKSIZE)
936 		nmp->nm_rsize = NFS_FABLKSIZE;
937 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
938 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
939 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
940 		    ~(NFS_DIRBLKSIZ - 1);
941 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
942 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
943 		if (nmp->nm_readdirsize == 0)
944 			nmp->nm_readdirsize = fsp->fs_rtmax;
945 	}
946 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
947 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
948 	if (fsp->fs_maxfilesize > 0 &&
949 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
950 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
951 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
952 	nmp->nm_state |= NFSSTA_GOTFSINFO;
953 }
954 
955 /*
956  * Get a pointer to my IP addrress and return it.
957  * Return NULL if you can't find one.
958  */
959 u_int8_t *
960 nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
961 {
962 	struct sockaddr_in sad, *sin;
963 	struct rtentry *rt;
964 	u_int8_t *retp = NULL;
965 	static struct in_addr laddr;
966 
967 	*isinet6p = 0;
968 	/*
969 	 * Loop up a route for the destination address.
970 	 */
971 	if (nmp->nm_nam->sa_family == AF_INET) {
972 		bzero(&sad, sizeof (sad));
973 		sin = (struct sockaddr_in *)nmp->nm_nam;
974 		sad.sin_family = AF_INET;
975 		sad.sin_len = sizeof (struct sockaddr_in);
976 		sad.sin_addr.s_addr = sin->sin_addr.s_addr;
977 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
978 		rt = rtalloc1((struct sockaddr *)&sad, 0, 0UL);
979 		if (rt != NULL) {
980 			if (rt->rt_ifp != NULL &&
981 			    rt->rt_ifa != NULL &&
982 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
983 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
984 				sin = (struct sockaddr_in *)
985 				    rt->rt_ifa->ifa_addr;
986 				laddr.s_addr = sin->sin_addr.s_addr;
987 				retp = (u_int8_t *)&laddr;
988 			}
989 			RTFREE_LOCKED(rt);
990 		}
991 		CURVNET_RESTORE();
992 #ifdef INET6
993 	} else if (nmp->nm_nam->sa_family == AF_INET6) {
994 		struct sockaddr_in6 sad6, *sin6;
995 		static struct in6_addr laddr6;
996 
997 		bzero(&sad6, sizeof (sad6));
998 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
999 		sad6.sin6_family = AF_INET6;
1000 		sad6.sin6_len = sizeof (struct sockaddr_in6);
1001 		sad6.sin6_addr = sin6->sin6_addr;
1002 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1003 		rt = rtalloc1((struct sockaddr *)&sad6, 0, 0UL);
1004 		if (rt != NULL) {
1005 			if (rt->rt_ifp != NULL &&
1006 			    rt->rt_ifa != NULL &&
1007 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1008 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
1009 				sin6 = (struct sockaddr_in6 *)
1010 				    rt->rt_ifa->ifa_addr;
1011 				laddr6 = sin6->sin6_addr;
1012 				retp = (u_int8_t *)&laddr6;
1013 				*isinet6p = 1;
1014 			}
1015 			RTFREE_LOCKED(rt);
1016 		}
1017 		CURVNET_RESTORE();
1018 #endif
1019 	}
1020 	return (retp);
1021 }
1022 
1023 /*
1024  * Copy NFS uid, gids from the cred structure.
1025  */
1026 void
1027 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1028 {
1029 	int i;
1030 
1031 	KASSERT(cr->cr_ngroups >= 0,
1032 	    ("newnfs_copyincred: negative cr_ngroups"));
1033 	nfscr->nfsc_uid = cr->cr_uid;
1034 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1035 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1036 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1037 }
1038 
1039 
1040 /*
1041  * Do any client specific initialization.
1042  */
1043 void
1044 nfscl_init(void)
1045 {
1046 	static int inited = 0;
1047 
1048 	if (inited)
1049 		return;
1050 	inited = 1;
1051 	nfscl_inited = 1;
1052 	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1053 }
1054 
1055 /*
1056  * Check each of the attributes to be set, to ensure they aren't already
1057  * the correct value. Disable setting ones already correct.
1058  */
1059 int
1060 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1061 {
1062 
1063 	if (vap->va_mode != (mode_t)VNOVAL) {
1064 		if (vap->va_mode == nvap->na_mode)
1065 			vap->va_mode = (mode_t)VNOVAL;
1066 	}
1067 	if (vap->va_uid != (uid_t)VNOVAL) {
1068 		if (vap->va_uid == nvap->na_uid)
1069 			vap->va_uid = (uid_t)VNOVAL;
1070 	}
1071 	if (vap->va_gid != (gid_t)VNOVAL) {
1072 		if (vap->va_gid == nvap->na_gid)
1073 			vap->va_gid = (gid_t)VNOVAL;
1074 	}
1075 	if (vap->va_size != VNOVAL) {
1076 		if (vap->va_size == nvap->na_size)
1077 			vap->va_size = VNOVAL;
1078 	}
1079 
1080 	/*
1081 	 * We are normally called with only a partially initialized
1082 	 * VAP.  Since the NFSv3 spec says that server may use the
1083 	 * file attributes to store the verifier, the spec requires
1084 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1085 	 * in atime, but we can't really assume that all servers will
1086 	 * so we ensure that our SETATTR sets both atime and mtime.
1087 	 */
1088 	if (vap->va_mtime.tv_sec == VNOVAL)
1089 		vfs_timestamp(&vap->va_mtime);
1090 	if (vap->va_atime.tv_sec == VNOVAL)
1091 		vap->va_atime = vap->va_mtime;
1092 	return (1);
1093 }
1094 
1095 /*
1096  * Map nfsv4 errors to errno.h errors.
1097  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1098  * error should only be returned for the Open, Create and Setattr Ops.
1099  * As such, most calls can just pass in 0 for those arguments.
1100  */
1101 APPLESTATIC int
1102 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1103 {
1104 	struct proc *p;
1105 
1106 	if (error < 10000)
1107 		return (error);
1108 	if (td != NULL)
1109 		p = td->td_proc;
1110 	else
1111 		p = NULL;
1112 	switch (error) {
1113 	case NFSERR_BADOWNER:
1114 		tprintf(p, LOG_INFO,
1115 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1116 		    uid, gid);
1117 		return (EPERM);
1118 	case NFSERR_STALECLIENTID:
1119 	case NFSERR_STALESTATEID:
1120 	case NFSERR_EXPIRED:
1121 	case NFSERR_BADSTATEID:
1122 		printf("nfsv4 recover err returned %d\n", error);
1123 		return (EIO);
1124 	case NFSERR_BADHANDLE:
1125 	case NFSERR_SERVERFAULT:
1126 	case NFSERR_BADTYPE:
1127 	case NFSERR_FHEXPIRED:
1128 	case NFSERR_RESOURCE:
1129 	case NFSERR_MOVED:
1130 	case NFSERR_NOFILEHANDLE:
1131 	case NFSERR_MINORVERMISMATCH:
1132 	case NFSERR_OLDSTATEID:
1133 	case NFSERR_BADSEQID:
1134 	case NFSERR_LEASEMOVED:
1135 	case NFSERR_RECLAIMBAD:
1136 	case NFSERR_BADXDR:
1137 	case NFSERR_BADCHAR:
1138 	case NFSERR_BADNAME:
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  * Locate a process by number; return only "live" processes -- i.e., neither
1151  * zombies nor newly born but incompletely initialized processes.  By not
1152  * returning processes in the PRS_NEW state, we allow callers to avoid
1153  * testing for that condition to avoid dereferencing p_ucred, et al.
1154  * Identical to pfind() in kern_proc.c, except it assume the list is
1155  * already locked.
1156  */
1157 static struct proc *
1158 pfind_locked(pid_t pid)
1159 {
1160 	struct proc *p;
1161 
1162 	LIST_FOREACH(p, PIDHASH(pid), p_hash)
1163 		if (p->p_pid == pid) {
1164 			PROC_LOCK(p);
1165 			if (p->p_state == PRS_NEW) {
1166 				PROC_UNLOCK(p);
1167 				p = NULL;
1168 			}
1169 			break;
1170 		}
1171 	return (p);
1172 }
1173 
1174 /*
1175  * Check to see if the process for this owner exists. Return 1 if it doesn't
1176  * and 0 otherwise.
1177  */
1178 int
1179 nfscl_procdoesntexist(u_int8_t *own)
1180 {
1181 	union {
1182 		u_int32_t	lval;
1183 		u_int8_t	cval[4];
1184 	} tl;
1185 	struct proc *p;
1186 	pid_t pid;
1187 	int ret = 0;
1188 
1189 	tl.cval[0] = *own++;
1190 	tl.cval[1] = *own++;
1191 	tl.cval[2] = *own++;
1192 	tl.cval[3] = *own++;
1193 	pid = tl.lval;
1194 	p = pfind_locked(pid);
1195 	if (p == NULL)
1196 		return (1);
1197 	if (p->p_stats == NULL) {
1198 		PROC_UNLOCK(p);
1199 		return (0);
1200 	}
1201 	tl.cval[0] = *own++;
1202 	tl.cval[1] = *own++;
1203 	tl.cval[2] = *own++;
1204 	tl.cval[3] = *own++;
1205 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1206 		ret = 1;
1207 	} else {
1208 		tl.cval[0] = *own++;
1209 		tl.cval[1] = *own++;
1210 		tl.cval[2] = *own++;
1211 		tl.cval[3] = *own;
1212 		if (tl.lval != p->p_stats->p_start.tv_usec)
1213 			ret = 1;
1214 	}
1215 	PROC_UNLOCK(p);
1216 	return (ret);
1217 }
1218 
1219 /*
1220  * - nfs pseudo system call for the client
1221  */
1222 /*
1223  * MPSAFE
1224  */
1225 static int
1226 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1227 {
1228 	struct file *fp;
1229 	struct nfscbd_args nfscbdarg;
1230 	struct nfsd_nfscbd_args nfscbdarg2;
1231 	int error;
1232 
1233 	if (uap->flag & NFSSVC_CBADDSOCK) {
1234 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1235 		if (error)
1236 			return (error);
1237 		/*
1238 		 * Since we don't know what rights might be required,
1239 		 * pretend that we need them all. It is better to be too
1240 		 * careful than too reckless.
1241 		 */
1242 		if ((error = fget(td, nfscbdarg.sock, CAP_SOCK_ALL, &fp))
1243 		    != 0) {
1244 			return (error);
1245 		}
1246 		if (fp->f_type != DTYPE_SOCKET) {
1247 			fdrop(fp, td);
1248 			return (EPERM);
1249 		}
1250 		error = nfscbd_addsock(fp);
1251 		fdrop(fp, td);
1252 		if (!error && nfscl_enablecallb == 0) {
1253 			nfsv4_cbport = nfscbdarg.port;
1254 			nfscl_enablecallb = 1;
1255 		}
1256 	} else if (uap->flag & NFSSVC_NFSCBD) {
1257 		if (uap->argp == NULL)
1258 			return (EINVAL);
1259 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1260 		    sizeof(nfscbdarg2));
1261 		if (error)
1262 			return (error);
1263 		error = nfscbd_nfsd(td, &nfscbdarg2);
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