xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision 8a166cafe0965f6bd72cd3d2f5372704f05cb5e8)
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 
730 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
731 	case ND_NFSV2:
732 		NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
733 		if (vap->va_mode == (mode_t)VNOVAL)
734 			sp->sa_mode = newnfs_xdrneg1;
735 		else
736 			sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
737 		if (vap->va_uid == (uid_t)VNOVAL)
738 			sp->sa_uid = newnfs_xdrneg1;
739 		else
740 			sp->sa_uid = txdr_unsigned(vap->va_uid);
741 		if (vap->va_gid == (gid_t)VNOVAL)
742 			sp->sa_gid = newnfs_xdrneg1;
743 		else
744 			sp->sa_gid = txdr_unsigned(vap->va_gid);
745 		if (flags & NFSSATTR_SIZE0)
746 			sp->sa_size = 0;
747 		else if (flags & NFSSATTR_SIZENEG1)
748 			sp->sa_size = newnfs_xdrneg1;
749 		else if (flags & NFSSATTR_SIZERDEV)
750 			sp->sa_size = txdr_unsigned(rdev);
751 		else
752 			sp->sa_size = txdr_unsigned(vap->va_size);
753 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
754 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
755 		break;
756 	case ND_NFSV3:
757 		if (vap->va_mode != (mode_t)VNOVAL) {
758 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
759 			*tl++ = newnfs_true;
760 			*tl = txdr_unsigned(vap->va_mode);
761 		} else {
762 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
763 			*tl = newnfs_false;
764 		}
765 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
766 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
767 			*tl++ = newnfs_true;
768 			*tl = txdr_unsigned(vap->va_uid);
769 		} else {
770 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
771 			*tl = newnfs_false;
772 		}
773 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
774 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
775 			*tl++ = newnfs_true;
776 			*tl = txdr_unsigned(vap->va_gid);
777 		} else {
778 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
779 			*tl = newnfs_false;
780 		}
781 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
782 			NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
783 			*tl++ = newnfs_true;
784 			txdr_hyper(vap->va_size, tl);
785 		} else {
786 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
787 			*tl = newnfs_false;
788 		}
789 		if (vap->va_atime.tv_sec != VNOVAL) {
790 			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
791 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
792 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
793 				txdr_nfsv3time(&vap->va_atime, tl);
794 			} else {
795 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
796 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
797 			}
798 		} else {
799 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
800 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
801 		}
802 		if (vap->va_mtime.tv_sec != VNOVAL) {
803 			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
804 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
805 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
806 				txdr_nfsv3time(&vap->va_mtime, tl);
807 			} else {
808 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
809 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
810 			}
811 		} else {
812 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
813 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
814 		}
815 		break;
816 	case ND_NFSV4:
817 		NFSZERO_ATTRBIT(&attrbits);
818 		if (vap->va_mode != (mode_t)VNOVAL)
819 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
820 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
821 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
822 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
823 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
824 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
825 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
826 		if (vap->va_atime.tv_sec != VNOVAL)
827 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
828 		if (vap->va_mtime.tv_sec != VNOVAL)
829 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
830 		(void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
831 		    &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
832 		break;
833 	};
834 }
835 
836 /*
837  * nfscl_request() - mostly a wrapper for newnfs_request().
838  */
839 int
840 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
841     struct ucred *cred, void *stuff)
842 {
843 	int ret, vers;
844 	struct nfsmount *nmp;
845 
846 	nmp = VFSTONFS(vp->v_mount);
847 	if (nd->nd_flag & ND_NFSV4)
848 		vers = NFS_VER4;
849 	else if (nd->nd_flag & ND_NFSV3)
850 		vers = NFS_VER3;
851 	else
852 		vers = NFS_VER2;
853 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
854 		NFS_PROG, vers, NULL, 1, NULL, NULL);
855 	return (ret);
856 }
857 
858 /*
859  * fill in this bsden's variant of statfs using nfsstatfs.
860  */
861 void
862 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
863 {
864 	struct statfs *sbp = (struct statfs *)statfs;
865 
866 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
867 		sbp->f_bsize = NFS_FABLKSIZE;
868 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
869 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
870 		/*
871 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
872 		 * the value after dividing by NFS_FABLKSIZE is small
873 		 * enough that it will fit in 63bits, so it is ok to
874 		 * assign it to f_bavail without fear that it will become
875 		 * negative.
876 		 */
877 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
878 		sbp->f_files = sfp->sf_tfiles;
879 		/* Since f_ffree is int64_t, clip it to 63bits. */
880 		if (sfp->sf_ffiles > INT64_MAX)
881 			sbp->f_ffree = INT64_MAX;
882 		else
883 			sbp->f_ffree = sfp->sf_ffiles;
884 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
885 		/*
886 		 * The type casts to (int32_t) ensure that this code is
887 		 * compatible with the old NFS client, in that it will
888 		 * propagate bit31 to the high order bits. This may or may
889 		 * not be correct for NFSv2, but since it is a legacy
890 		 * environment, I'd rather retain backwards compatibility.
891 		 */
892 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
893 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
894 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
895 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
896 		sbp->f_files = 0;
897 		sbp->f_ffree = 0;
898 	}
899 }
900 
901 /*
902  * Use the fsinfo stuff to update the mount point.
903  */
904 void
905 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
906 {
907 
908 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
909 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
910 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
911 		    ~(NFS_FABLKSIZE - 1);
912 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
913 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
914 		if (nmp->nm_wsize == 0)
915 			nmp->nm_wsize = fsp->fs_wtmax;
916 	}
917 	if (nmp->nm_wsize < NFS_FABLKSIZE)
918 		nmp->nm_wsize = NFS_FABLKSIZE;
919 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
920 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
921 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
922 		    ~(NFS_FABLKSIZE - 1);
923 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
924 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
925 		if (nmp->nm_rsize == 0)
926 			nmp->nm_rsize = fsp->fs_rtmax;
927 	}
928 	if (nmp->nm_rsize < NFS_FABLKSIZE)
929 		nmp->nm_rsize = NFS_FABLKSIZE;
930 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
931 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
932 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
933 		    ~(NFS_DIRBLKSIZ - 1);
934 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
935 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
936 		if (nmp->nm_readdirsize == 0)
937 			nmp->nm_readdirsize = fsp->fs_rtmax;
938 	}
939 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
940 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
941 	if (fsp->fs_maxfilesize > 0 &&
942 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
943 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
944 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
945 	nmp->nm_state |= NFSSTA_GOTFSINFO;
946 }
947 
948 /*
949  * Get a pointer to my IP addrress and return it.
950  * Return NULL if you can't find one.
951  */
952 u_int8_t *
953 nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
954 {
955 	struct sockaddr_in sad, *sin;
956 	struct rtentry *rt;
957 	u_int8_t *retp = NULL;
958 	static struct in_addr laddr;
959 
960 	*isinet6p = 0;
961 	/*
962 	 * Loop up a route for the destination address.
963 	 */
964 	if (nmp->nm_nam->sa_family == AF_INET) {
965 		bzero(&sad, sizeof (sad));
966 		sin = (struct sockaddr_in *)nmp->nm_nam;
967 		sad.sin_family = AF_INET;
968 		sad.sin_len = sizeof (struct sockaddr_in);
969 		sad.sin_addr.s_addr = sin->sin_addr.s_addr;
970 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
971 		rt = rtalloc1_fib((struct sockaddr *)&sad, 0, 0UL,
972 		     curthread->td_proc->p_fibnum);
973 		if (rt != NULL) {
974 			if (rt->rt_ifp != NULL &&
975 			    rt->rt_ifa != NULL &&
976 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
977 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
978 				sin = (struct sockaddr_in *)
979 				    rt->rt_ifa->ifa_addr;
980 				laddr.s_addr = sin->sin_addr.s_addr;
981 				retp = (u_int8_t *)&laddr;
982 			}
983 			RTFREE_LOCKED(rt);
984 		}
985 		CURVNET_RESTORE();
986 #ifdef INET6
987 	} else if (nmp->nm_nam->sa_family == AF_INET6) {
988 		struct sockaddr_in6 sad6, *sin6;
989 		static struct in6_addr laddr6;
990 
991 		bzero(&sad6, sizeof (sad6));
992 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
993 		sad6.sin6_family = AF_INET6;
994 		sad6.sin6_len = sizeof (struct sockaddr_in6);
995 		sad6.sin6_addr = sin6->sin6_addr;
996 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
997 		rt = rtalloc1_fib((struct sockaddr *)&sad6, 0, 0UL,
998 		     curthread->td_proc->p_fibnum);
999 		if (rt != NULL) {
1000 			if (rt->rt_ifp != NULL &&
1001 			    rt->rt_ifa != NULL &&
1002 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1003 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
1004 				sin6 = (struct sockaddr_in6 *)
1005 				    rt->rt_ifa->ifa_addr;
1006 				laddr6 = sin6->sin6_addr;
1007 				retp = (u_int8_t *)&laddr6;
1008 				*isinet6p = 1;
1009 			}
1010 			RTFREE_LOCKED(rt);
1011 		}
1012 		CURVNET_RESTORE();
1013 #endif
1014 	}
1015 	return (retp);
1016 }
1017 
1018 /*
1019  * Copy NFS uid, gids from the cred structure.
1020  */
1021 void
1022 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1023 {
1024 	int i;
1025 
1026 	KASSERT(cr->cr_ngroups >= 0,
1027 	    ("newnfs_copyincred: negative cr_ngroups"));
1028 	nfscr->nfsc_uid = cr->cr_uid;
1029 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1030 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1031 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1032 }
1033 
1034 
1035 /*
1036  * Do any client specific initialization.
1037  */
1038 void
1039 nfscl_init(void)
1040 {
1041 	static int inited = 0;
1042 
1043 	if (inited)
1044 		return;
1045 	inited = 1;
1046 	nfscl_inited = 1;
1047 	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1048 }
1049 
1050 /*
1051  * Check each of the attributes to be set, to ensure they aren't already
1052  * the correct value. Disable setting ones already correct.
1053  */
1054 int
1055 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1056 {
1057 
1058 	if (vap->va_mode != (mode_t)VNOVAL) {
1059 		if (vap->va_mode == nvap->na_mode)
1060 			vap->va_mode = (mode_t)VNOVAL;
1061 	}
1062 	if (vap->va_uid != (uid_t)VNOVAL) {
1063 		if (vap->va_uid == nvap->na_uid)
1064 			vap->va_uid = (uid_t)VNOVAL;
1065 	}
1066 	if (vap->va_gid != (gid_t)VNOVAL) {
1067 		if (vap->va_gid == nvap->na_gid)
1068 			vap->va_gid = (gid_t)VNOVAL;
1069 	}
1070 	if (vap->va_size != VNOVAL) {
1071 		if (vap->va_size == nvap->na_size)
1072 			vap->va_size = VNOVAL;
1073 	}
1074 
1075 	/*
1076 	 * We are normally called with only a partially initialized
1077 	 * VAP.  Since the NFSv3 spec says that server may use the
1078 	 * file attributes to store the verifier, the spec requires
1079 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1080 	 * in atime, but we can't really assume that all servers will
1081 	 * so we ensure that our SETATTR sets both atime and mtime.
1082 	 */
1083 	if (vap->va_mtime.tv_sec == VNOVAL)
1084 		vfs_timestamp(&vap->va_mtime);
1085 	if (vap->va_atime.tv_sec == VNOVAL)
1086 		vap->va_atime = vap->va_mtime;
1087 	return (1);
1088 }
1089 
1090 /*
1091  * Map nfsv4 errors to errno.h errors.
1092  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1093  * error should only be returned for the Open, Create and Setattr Ops.
1094  * As such, most calls can just pass in 0 for those arguments.
1095  */
1096 APPLESTATIC int
1097 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1098 {
1099 	struct proc *p;
1100 
1101 	if (error < 10000)
1102 		return (error);
1103 	if (td != NULL)
1104 		p = td->td_proc;
1105 	else
1106 		p = NULL;
1107 	switch (error) {
1108 	case NFSERR_BADOWNER:
1109 		tprintf(p, LOG_INFO,
1110 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1111 		    uid, gid);
1112 		return (EPERM);
1113 	case NFSERR_BADNAME:
1114 	case NFSERR_BADCHAR:
1115 		printf("nfsv4 char/name not handled by server\n");
1116 		return (ENOENT);
1117 	case NFSERR_STALECLIENTID:
1118 	case NFSERR_STALESTATEID:
1119 	case NFSERR_EXPIRED:
1120 	case NFSERR_BADSTATEID:
1121 	case NFSERR_BADSESSION:
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_OPILLEGAL:
1138 		printf("nfsv4 client/server protocol prob err=%d\n",
1139 		    error);
1140 		return (EIO);
1141 	default:
1142 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1143 		return (EIO);
1144 	};
1145 }
1146 
1147 /*
1148  * Check to see if the process for this owner exists. Return 1 if it doesn't
1149  * and 0 otherwise.
1150  */
1151 int
1152 nfscl_procdoesntexist(u_int8_t *own)
1153 {
1154 	union {
1155 		u_int32_t	lval;
1156 		u_int8_t	cval[4];
1157 	} tl;
1158 	struct proc *p;
1159 	pid_t pid;
1160 	int ret = 0;
1161 
1162 	tl.cval[0] = *own++;
1163 	tl.cval[1] = *own++;
1164 	tl.cval[2] = *own++;
1165 	tl.cval[3] = *own++;
1166 	pid = tl.lval;
1167 	p = pfind_locked(pid);
1168 	if (p == NULL)
1169 		return (1);
1170 	if (p->p_stats == NULL) {
1171 		PROC_UNLOCK(p);
1172 		return (0);
1173 	}
1174 	tl.cval[0] = *own++;
1175 	tl.cval[1] = *own++;
1176 	tl.cval[2] = *own++;
1177 	tl.cval[3] = *own++;
1178 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1179 		ret = 1;
1180 	} else {
1181 		tl.cval[0] = *own++;
1182 		tl.cval[1] = *own++;
1183 		tl.cval[2] = *own++;
1184 		tl.cval[3] = *own;
1185 		if (tl.lval != p->p_stats->p_start.tv_usec)
1186 			ret = 1;
1187 	}
1188 	PROC_UNLOCK(p);
1189 	return (ret);
1190 }
1191 
1192 /*
1193  * - nfs pseudo system call for the client
1194  */
1195 /*
1196  * MPSAFE
1197  */
1198 static int
1199 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1200 {
1201 	struct file *fp;
1202 	struct nfscbd_args nfscbdarg;
1203 	struct nfsd_nfscbd_args nfscbdarg2;
1204 	int error;
1205 	struct nameidata nd;
1206 	struct nfscl_dumpmntopts dumpmntopts;
1207 	char *buf;
1208 
1209 	if (uap->flag & NFSSVC_CBADDSOCK) {
1210 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1211 		if (error)
1212 			return (error);
1213 		/*
1214 		 * Since we don't know what rights might be required,
1215 		 * pretend that we need them all. It is better to be too
1216 		 * careful than too reckless.
1217 		 */
1218 		if ((error = fget(td, nfscbdarg.sock, CAP_SOCK_ALL, &fp))
1219 		    != 0) {
1220 			return (error);
1221 		}
1222 		if (fp->f_type != DTYPE_SOCKET) {
1223 			fdrop(fp, td);
1224 			return (EPERM);
1225 		}
1226 		error = nfscbd_addsock(fp);
1227 		fdrop(fp, td);
1228 		if (!error && nfscl_enablecallb == 0) {
1229 			nfsv4_cbport = nfscbdarg.port;
1230 			nfscl_enablecallb = 1;
1231 		}
1232 	} else if (uap->flag & NFSSVC_NFSCBD) {
1233 		if (uap->argp == NULL)
1234 			return (EINVAL);
1235 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1236 		    sizeof(nfscbdarg2));
1237 		if (error)
1238 			return (error);
1239 		error = nfscbd_nfsd(td, &nfscbdarg2);
1240 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1241 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1242 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1243 		    dumpmntopts.ndmnt_blen > 1024))
1244 			error = EINVAL;
1245 		if (error == 0)
1246 			error = nfsrv_lookupfilename(&nd,
1247 			    dumpmntopts.ndmnt_fname, td);
1248 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1249 		    "nfs") != 0) {
1250 			vput(nd.ni_vp);
1251 			error = EINVAL;
1252 		}
1253 		if (error == 0) {
1254 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1255 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1256 			    dumpmntopts.ndmnt_blen);
1257 			vput(nd.ni_vp);
1258 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1259 			    dumpmntopts.ndmnt_blen);
1260 			free(buf, M_TEMP);
1261 		}
1262 	} else {
1263 		error = EINVAL;
1264 	}
1265 	return (error);
1266 }
1267 
1268 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1269 
1270 /*
1271  * Called once to initialize data structures...
1272  */
1273 static int
1274 nfscl_modevent(module_t mod, int type, void *data)
1275 {
1276 	int error = 0;
1277 	static int loaded = 0;
1278 
1279 	switch (type) {
1280 	case MOD_LOAD:
1281 		if (loaded)
1282 			return (0);
1283 		newnfs_portinit();
1284 		mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1285 		    MTX_DEF);
1286 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1287 		nfscl_init();
1288 		NFSD_LOCK();
1289 		nfsrvd_cbinit(0);
1290 		NFSD_UNLOCK();
1291 		ncl_call_invalcaches = ncl_invalcaches;
1292 		nfsd_call_nfscl = nfssvc_nfscl;
1293 		loaded = 1;
1294 		break;
1295 
1296 	case MOD_UNLOAD:
1297 		if (nfs_numnfscbd != 0) {
1298 			error = EBUSY;
1299 			break;
1300 		}
1301 
1302 		/*
1303 		 * XXX: Unloading of nfscl module is unsupported.
1304 		 */
1305 #if 0
1306 		ncl_call_invalcaches = NULL;
1307 		nfsd_call_nfscl = NULL;
1308 		/* and get rid of the mutexes */
1309 		mtx_destroy(&nfs_clstate_mutex);
1310 		mtx_destroy(&ncl_iod_mutex);
1311 		loaded = 0;
1312 		break;
1313 #else
1314 		/* FALLTHROUGH */
1315 #endif
1316 	default:
1317 		error = EOPNOTSUPP;
1318 		break;
1319 	}
1320 	return error;
1321 }
1322 static moduledata_t nfscl_mod = {
1323 	"nfscl",
1324 	nfscl_modevent,
1325 	NULL,
1326 };
1327 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1328 
1329 /* So that loader and kldload(2) can find us, wherever we are.. */
1330 MODULE_VERSION(nfscl, 1);
1331 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1332 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1333 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1334 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1335 
1336