xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision 2a3e3873a1e4cd958f2b0f85d3b10cfa40575d30)
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 	u_quad_t nsize;
365 	int setnsize;
366 
367 	/*
368 	 * If v_type == VNON it is a new node, so fill in the v_type,
369 	 * n_mtime fields. Check to see if it represents a special
370 	 * device, and if so, check for a possible alias. Once the
371 	 * correct vnode has been obtained, fill in the rest of the
372 	 * information.
373 	 */
374 	np = VTONFS(vp);
375 	NFSLOCKNODE(np);
376 	if (vp->v_type != nvap->va_type) {
377 		vp->v_type = nvap->va_type;
378 		if (vp->v_type == VFIFO)
379 			vp->v_op = &newnfs_fifoops;
380 		np->n_mtime = nvap->va_mtime;
381 	}
382 	nmp = VFSTONFS(vp->v_mount);
383 	vap = &np->n_vattr.na_vattr;
384 	mtime_save = vap->va_mtime;
385 	if (writeattr) {
386 		np->n_vattr.na_filerev = nap->na_filerev;
387 		np->n_vattr.na_size = nap->na_size;
388 		np->n_vattr.na_mtime = nap->na_mtime;
389 		np->n_vattr.na_ctime = nap->na_ctime;
390 		np->n_vattr.na_fsid = nap->na_fsid;
391 		np->n_vattr.na_mode = nap->na_mode;
392 	} else {
393 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
394 		    sizeof (struct nfsvattr));
395 	}
396 
397 	/*
398 	 * For NFSv4, if the node's fsid is not equal to the mount point's
399 	 * fsid, return the low order 32bits of the node's fsid. This
400 	 * allows getcwd(3) to work. There is a chance that the fsid might
401 	 * be the same as a local fs, but since this is in an NFS mount
402 	 * point, I don't think that will cause any problems?
403 	 */
404 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
405 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
406 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
407 		/*
408 		 * va_fsid needs to be set to some value derived from
409 		 * np->n_vattr.na_filesid that is not equal
410 		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
411 		 * from the value used for the top level server volume
412 		 * in the mounted subtree.
413 		 */
414 		if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
415 		    (uint32_t)np->n_vattr.na_filesid[0])
416 			vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
417 		else
418 			vap->va_fsid = (uint32_t)hash32_buf(
419 			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
420 	} else
421 		vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
422 	np->n_attrstamp = time_second;
423 	setnsize = 0;
424 	nsize = 0;
425 	if (vap->va_size != np->n_size) {
426 		if (vap->va_type == VREG) {
427 			if (dontshrink && vap->va_size < np->n_size) {
428 				/*
429 				 * We've been told not to shrink the file;
430 				 * zero np->n_attrstamp to indicate that
431 				 * the attributes are stale.
432 				 */
433 				vap->va_size = np->n_size;
434 				np->n_attrstamp = 0;
435 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
436 			} else if (np->n_flag & NMODIFIED) {
437 				/*
438 				 * We've modified the file: Use the larger
439 				 * of our size, and the server's size.
440 				 */
441 				if (vap->va_size < np->n_size) {
442 					vap->va_size = np->n_size;
443 				} else {
444 					np->n_size = vap->va_size;
445 					np->n_flag |= NSIZECHANGED;
446 				}
447 			} else {
448 				np->n_size = vap->va_size;
449 				np->n_flag |= NSIZECHANGED;
450 			}
451 			setnsize = 1;
452 			nsize = vap->va_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 	if (setnsize)
489 		vnode_pager_setsize(vp, nsize);
490 	return (0);
491 }
492 
493 /*
494  * Fill in the client id name. For these bytes:
495  * 1 - they must be unique
496  * 2 - they should be persistent across client reboots
497  * 1 is more critical than 2
498  * Use the mount point's unique id plus either the uuid or, if that
499  * isn't set, random junk.
500  */
501 void
502 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
503 {
504 	int uuidlen;
505 
506 	/*
507 	 * First, put in the 64bit mount point identifier.
508 	 */
509 	if (idlen >= sizeof (u_int64_t)) {
510 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
511 		cp += sizeof (u_int64_t);
512 		idlen -= sizeof (u_int64_t);
513 	}
514 
515 	/*
516 	 * If uuid is non-zero length, use it.
517 	 */
518 	uuidlen = strlen(uuid);
519 	if (uuidlen > 0 && idlen >= uuidlen) {
520 		NFSBCOPY(uuid, cp, uuidlen);
521 		cp += uuidlen;
522 		idlen -= uuidlen;
523 	}
524 
525 	/*
526 	 * This only normally happens if the uuid isn't set.
527 	 */
528 	while (idlen > 0) {
529 		*cp++ = (u_int8_t)(arc4random() % 256);
530 		idlen--;
531 	}
532 }
533 
534 /*
535  * Fill in a lock owner name. For now, pid + the process's creation time.
536  */
537 void
538 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
539 {
540 	union {
541 		u_int32_t	lval;
542 		u_int8_t	cval[4];
543 	} tl;
544 	struct proc *p;
545 
546 	if (id == NULL) {
547 		printf("NULL id\n");
548 		bzero(cp, NFSV4CL_LOCKNAMELEN);
549 		return;
550 	}
551 	if ((flags & F_POSIX) != 0) {
552 		p = (struct proc *)id;
553 		tl.lval = p->p_pid;
554 		*cp++ = tl.cval[0];
555 		*cp++ = tl.cval[1];
556 		*cp++ = tl.cval[2];
557 		*cp++ = tl.cval[3];
558 		tl.lval = p->p_stats->p_start.tv_sec;
559 		*cp++ = tl.cval[0];
560 		*cp++ = tl.cval[1];
561 		*cp++ = tl.cval[2];
562 		*cp++ = tl.cval[3];
563 		tl.lval = p->p_stats->p_start.tv_usec;
564 		*cp++ = tl.cval[0];
565 		*cp++ = tl.cval[1];
566 		*cp++ = tl.cval[2];
567 		*cp = tl.cval[3];
568 	} else if ((flags & F_FLOCK) != 0) {
569 		bcopy(&id, cp, sizeof(id));
570 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
571 	} else {
572 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
573 		bzero(cp, NFSV4CL_LOCKNAMELEN);
574 	}
575 }
576 
577 /*
578  * Find the parent process for the thread passed in as an argument.
579  * If none exists, return NULL, otherwise return a thread for the parent.
580  * (Can be any of the threads, since it is only used for td->td_proc.)
581  */
582 NFSPROC_T *
583 nfscl_getparent(struct thread *td)
584 {
585 	struct proc *p;
586 	struct thread *ptd;
587 
588 	if (td == NULL)
589 		return (NULL);
590 	p = td->td_proc;
591 	if (p->p_pid == 0)
592 		return (NULL);
593 	p = p->p_pptr;
594 	if (p == NULL)
595 		return (NULL);
596 	ptd = TAILQ_FIRST(&p->p_threads);
597 	return (ptd);
598 }
599 
600 /*
601  * Start up the renew kernel thread.
602  */
603 static void
604 start_nfscl(void *arg)
605 {
606 	struct nfsclclient *clp;
607 	struct thread *td;
608 
609 	clp = (struct nfsclclient *)arg;
610 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
611 	nfscl_renewthread(clp, td);
612 	kproc_exit(0);
613 }
614 
615 void
616 nfscl_start_renewthread(struct nfsclclient *clp)
617 {
618 
619 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
620 	    "nfscl");
621 }
622 
623 /*
624  * Handle wcc_data.
625  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
626  * as the first Op after PutFH.
627  * (For NFSv4, the postop attributes are after the Op, so they can't be
628  *  parsed here. A separate call to nfscl_postop_attr() is required.)
629  */
630 int
631 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
632     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
633 {
634 	u_int32_t *tl;
635 	struct nfsnode *np = VTONFS(vp);
636 	struct nfsvattr nfsva;
637 	int error = 0;
638 
639 	if (wccflagp != NULL)
640 		*wccflagp = 0;
641 	if (nd->nd_flag & ND_NFSV3) {
642 		*flagp = 0;
643 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
644 		if (*tl == newnfs_true) {
645 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
646 			if (wccflagp != NULL) {
647 				mtx_lock(&np->n_mtx);
648 				*wccflagp = (np->n_mtime.tv_sec ==
649 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
650 				    np->n_mtime.tv_nsec ==
651 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
652 				mtx_unlock(&np->n_mtx);
653 			}
654 		}
655 		error = nfscl_postop_attr(nd, nap, flagp, stuff);
656 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
657 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
658 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
659 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
660 		    NULL, NULL, NULL, NULL, NULL);
661 		if (error)
662 			return (error);
663 		/*
664 		 * Get rid of Op# and status for next op.
665 		 */
666 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
667 		if (*++tl)
668 			nd->nd_flag |= ND_NOMOREDATA;
669 		if (wccflagp != NULL &&
670 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
671 			mtx_lock(&np->n_mtx);
672 			*wccflagp = (np->n_mtime.tv_sec ==
673 			    nfsva.na_vattr.va_mtime.tv_sec &&
674 			    np->n_mtime.tv_nsec ==
675 			    nfsva.na_vattr.va_mtime.tv_sec);
676 			mtx_unlock(&np->n_mtx);
677 		}
678 	}
679 nfsmout:
680 	return (error);
681 }
682 
683 /*
684  * Get postop attributes.
685  */
686 int
687 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
688     void *stuff)
689 {
690 	u_int32_t *tl;
691 	int error = 0;
692 
693 	*retp = 0;
694 	if (nd->nd_flag & ND_NOMOREDATA)
695 		return (error);
696 	if (nd->nd_flag & ND_NFSV3) {
697 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
698 		*retp = fxdr_unsigned(int, *tl);
699 	} else if (nd->nd_flag & ND_NFSV4) {
700 		/*
701 		 * For NFSv4, the postop attr are at the end, so no point
702 		 * in looking if nd_repstat != 0.
703 		 */
704 		if (!nd->nd_repstat) {
705 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
706 			if (*(tl + 1))
707 				/* should never happen since nd_repstat != 0 */
708 				nd->nd_flag |= ND_NOMOREDATA;
709 			else
710 				*retp = 1;
711 		}
712 	} else if (!nd->nd_repstat) {
713 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
714 		*retp = 1;
715 	}
716 	if (*retp) {
717 		error = nfsm_loadattr(nd, nap);
718 		if (error)
719 			*retp = 0;
720 	}
721 nfsmout:
722 	return (error);
723 }
724 
725 /*
726  * Fill in the setable attributes. The full argument indicates whether
727  * to fill in them all or just mode and time.
728  */
729 void
730 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
731     struct vnode *vp, int flags, u_int32_t rdev)
732 {
733 	u_int32_t *tl;
734 	struct nfsv2_sattr *sp;
735 	nfsattrbit_t attrbits;
736 
737 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
738 	case ND_NFSV2:
739 		NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
740 		if (vap->va_mode == (mode_t)VNOVAL)
741 			sp->sa_mode = newnfs_xdrneg1;
742 		else
743 			sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
744 		if (vap->va_uid == (uid_t)VNOVAL)
745 			sp->sa_uid = newnfs_xdrneg1;
746 		else
747 			sp->sa_uid = txdr_unsigned(vap->va_uid);
748 		if (vap->va_gid == (gid_t)VNOVAL)
749 			sp->sa_gid = newnfs_xdrneg1;
750 		else
751 			sp->sa_gid = txdr_unsigned(vap->va_gid);
752 		if (flags & NFSSATTR_SIZE0)
753 			sp->sa_size = 0;
754 		else if (flags & NFSSATTR_SIZENEG1)
755 			sp->sa_size = newnfs_xdrneg1;
756 		else if (flags & NFSSATTR_SIZERDEV)
757 			sp->sa_size = txdr_unsigned(rdev);
758 		else
759 			sp->sa_size = txdr_unsigned(vap->va_size);
760 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
761 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
762 		break;
763 	case ND_NFSV3:
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_vaflags & VA_UTIMES_NULL) == 0) {
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_vaflags & VA_UTIMES_NULL) == 0) {
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, 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_fib((struct sockaddr *)&sad, 0, 0UL,
979 		     curthread->td_proc->p_fibnum);
980 		if (rt != NULL) {
981 			if (rt->rt_ifp != NULL &&
982 			    rt->rt_ifa != NULL &&
983 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
984 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
985 				sin = (struct sockaddr_in *)
986 				    rt->rt_ifa->ifa_addr;
987 				laddr.s_addr = sin->sin_addr.s_addr;
988 				retp = (u_int8_t *)&laddr;
989 			}
990 			RTFREE_LOCKED(rt);
991 		}
992 		CURVNET_RESTORE();
993 #ifdef INET6
994 	} else if (nmp->nm_nam->sa_family == AF_INET6) {
995 		struct sockaddr_in6 sad6, *sin6;
996 		static struct in6_addr laddr6;
997 
998 		bzero(&sad6, sizeof (sad6));
999 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1000 		sad6.sin6_family = AF_INET6;
1001 		sad6.sin6_len = sizeof (struct sockaddr_in6);
1002 		sad6.sin6_addr = sin6->sin6_addr;
1003 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1004 		rt = rtalloc1_fib((struct sockaddr *)&sad6, 0, 0UL,
1005 		     curthread->td_proc->p_fibnum);
1006 		if (rt != NULL) {
1007 			if (rt->rt_ifp != NULL &&
1008 			    rt->rt_ifa != NULL &&
1009 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1010 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
1011 				sin6 = (struct sockaddr_in6 *)
1012 				    rt->rt_ifa->ifa_addr;
1013 				laddr6 = sin6->sin6_addr;
1014 				retp = (u_int8_t *)&laddr6;
1015 				*isinet6p = 1;
1016 			}
1017 			RTFREE_LOCKED(rt);
1018 		}
1019 		CURVNET_RESTORE();
1020 #endif
1021 	}
1022 	return (retp);
1023 }
1024 
1025 /*
1026  * Copy NFS uid, gids from the cred structure.
1027  */
1028 void
1029 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1030 {
1031 	int i;
1032 
1033 	KASSERT(cr->cr_ngroups >= 0,
1034 	    ("newnfs_copyincred: negative cr_ngroups"));
1035 	nfscr->nfsc_uid = cr->cr_uid;
1036 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1037 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1038 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1039 }
1040 
1041 
1042 /*
1043  * Do any client specific initialization.
1044  */
1045 void
1046 nfscl_init(void)
1047 {
1048 	static int inited = 0;
1049 
1050 	if (inited)
1051 		return;
1052 	inited = 1;
1053 	nfscl_inited = 1;
1054 	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1055 }
1056 
1057 /*
1058  * Check each of the attributes to be set, to ensure they aren't already
1059  * the correct value. Disable setting ones already correct.
1060  */
1061 int
1062 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1063 {
1064 
1065 	if (vap->va_mode != (mode_t)VNOVAL) {
1066 		if (vap->va_mode == nvap->na_mode)
1067 			vap->va_mode = (mode_t)VNOVAL;
1068 	}
1069 	if (vap->va_uid != (uid_t)VNOVAL) {
1070 		if (vap->va_uid == nvap->na_uid)
1071 			vap->va_uid = (uid_t)VNOVAL;
1072 	}
1073 	if (vap->va_gid != (gid_t)VNOVAL) {
1074 		if (vap->va_gid == nvap->na_gid)
1075 			vap->va_gid = (gid_t)VNOVAL;
1076 	}
1077 	if (vap->va_size != VNOVAL) {
1078 		if (vap->va_size == nvap->na_size)
1079 			vap->va_size = VNOVAL;
1080 	}
1081 
1082 	/*
1083 	 * We are normally called with only a partially initialized
1084 	 * VAP.  Since the NFSv3 spec says that server may use the
1085 	 * file attributes to store the verifier, the spec requires
1086 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1087 	 * in atime, but we can't really assume that all servers will
1088 	 * so we ensure that our SETATTR sets both atime and mtime.
1089 	 */
1090 	if (vap->va_mtime.tv_sec == VNOVAL)
1091 		vfs_timestamp(&vap->va_mtime);
1092 	if (vap->va_atime.tv_sec == VNOVAL)
1093 		vap->va_atime = vap->va_mtime;
1094 	return (1);
1095 }
1096 
1097 /*
1098  * Map nfsv4 errors to errno.h errors.
1099  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1100  * error should only be returned for the Open, Create and Setattr Ops.
1101  * As such, most calls can just pass in 0 for those arguments.
1102  */
1103 APPLESTATIC int
1104 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1105 {
1106 	struct proc *p;
1107 
1108 	if (error < 10000)
1109 		return (error);
1110 	if (td != NULL)
1111 		p = td->td_proc;
1112 	else
1113 		p = NULL;
1114 	switch (error) {
1115 	case NFSERR_BADOWNER:
1116 		tprintf(p, LOG_INFO,
1117 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1118 		    uid, gid);
1119 		return (EPERM);
1120 	case NFSERR_BADNAME:
1121 	case NFSERR_BADCHAR:
1122 		printf("nfsv4 char/name not handled by server\n");
1123 		return (ENOENT);
1124 	case NFSERR_STALECLIENTID:
1125 	case NFSERR_STALESTATEID:
1126 	case NFSERR_EXPIRED:
1127 	case NFSERR_BADSTATEID:
1128 	case NFSERR_BADSESSION:
1129 		printf("nfsv4 recover err returned %d\n", error);
1130 		return (EIO);
1131 	case NFSERR_BADHANDLE:
1132 	case NFSERR_SERVERFAULT:
1133 	case NFSERR_BADTYPE:
1134 	case NFSERR_FHEXPIRED:
1135 	case NFSERR_RESOURCE:
1136 	case NFSERR_MOVED:
1137 	case NFSERR_NOFILEHANDLE:
1138 	case NFSERR_MINORVERMISMATCH:
1139 	case NFSERR_OLDSTATEID:
1140 	case NFSERR_BADSEQID:
1141 	case NFSERR_LEASEMOVED:
1142 	case NFSERR_RECLAIMBAD:
1143 	case NFSERR_BADXDR:
1144 	case NFSERR_OPILLEGAL:
1145 		printf("nfsv4 client/server protocol prob err=%d\n",
1146 		    error);
1147 		return (EIO);
1148 	default:
1149 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1150 		return (EIO);
1151 	};
1152 }
1153 
1154 /*
1155  * Check to see if the process for this owner exists. Return 1 if it doesn't
1156  * and 0 otherwise.
1157  */
1158 int
1159 nfscl_procdoesntexist(u_int8_t *own)
1160 {
1161 	union {
1162 		u_int32_t	lval;
1163 		u_int8_t	cval[4];
1164 	} tl;
1165 	struct proc *p;
1166 	pid_t pid;
1167 	int ret = 0;
1168 
1169 	tl.cval[0] = *own++;
1170 	tl.cval[1] = *own++;
1171 	tl.cval[2] = *own++;
1172 	tl.cval[3] = *own++;
1173 	pid = tl.lval;
1174 	p = pfind_locked(pid);
1175 	if (p == NULL)
1176 		return (1);
1177 	if (p->p_stats == NULL) {
1178 		PROC_UNLOCK(p);
1179 		return (0);
1180 	}
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_sec) {
1186 		ret = 1;
1187 	} else {
1188 		tl.cval[0] = *own++;
1189 		tl.cval[1] = *own++;
1190 		tl.cval[2] = *own++;
1191 		tl.cval[3] = *own;
1192 		if (tl.lval != p->p_stats->p_start.tv_usec)
1193 			ret = 1;
1194 	}
1195 	PROC_UNLOCK(p);
1196 	return (ret);
1197 }
1198 
1199 /*
1200  * - nfs pseudo system call for the client
1201  */
1202 /*
1203  * MPSAFE
1204  */
1205 static int
1206 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1207 {
1208 	struct file *fp;
1209 	struct nfscbd_args nfscbdarg;
1210 	struct nfsd_nfscbd_args nfscbdarg2;
1211 	int error;
1212 	struct nameidata nd;
1213 	struct nfscl_dumpmntopts dumpmntopts;
1214 	char *buf;
1215 
1216 	if (uap->flag & NFSSVC_CBADDSOCK) {
1217 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1218 		if (error)
1219 			return (error);
1220 		/*
1221 		 * Since we don't know what rights might be required,
1222 		 * pretend that we need them all. It is better to be too
1223 		 * careful than too reckless.
1224 		 */
1225 		if ((error = fget(td, nfscbdarg.sock, CAP_SOCK_CLIENT, &fp))
1226 		    != 0) {
1227 			return (error);
1228 		}
1229 		if (fp->f_type != DTYPE_SOCKET) {
1230 			fdrop(fp, td);
1231 			return (EPERM);
1232 		}
1233 		error = nfscbd_addsock(fp);
1234 		fdrop(fp, td);
1235 		if (!error && nfscl_enablecallb == 0) {
1236 			nfsv4_cbport = nfscbdarg.port;
1237 			nfscl_enablecallb = 1;
1238 		}
1239 	} else if (uap->flag & NFSSVC_NFSCBD) {
1240 		if (uap->argp == NULL)
1241 			return (EINVAL);
1242 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1243 		    sizeof(nfscbdarg2));
1244 		if (error)
1245 			return (error);
1246 		error = nfscbd_nfsd(td, &nfscbdarg2);
1247 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1248 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1249 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1250 		    dumpmntopts.ndmnt_blen > 1024))
1251 			error = EINVAL;
1252 		if (error == 0)
1253 			error = nfsrv_lookupfilename(&nd,
1254 			    dumpmntopts.ndmnt_fname, td);
1255 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1256 		    "nfs") != 0) {
1257 			vput(nd.ni_vp);
1258 			error = EINVAL;
1259 		}
1260 		if (error == 0) {
1261 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1262 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1263 			    dumpmntopts.ndmnt_blen);
1264 			vput(nd.ni_vp);
1265 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1266 			    dumpmntopts.ndmnt_blen);
1267 			free(buf, M_TEMP);
1268 		}
1269 	} else {
1270 		error = EINVAL;
1271 	}
1272 	return (error);
1273 }
1274 
1275 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1276 
1277 /*
1278  * Called once to initialize data structures...
1279  */
1280 static int
1281 nfscl_modevent(module_t mod, int type, void *data)
1282 {
1283 	int error = 0;
1284 	static int loaded = 0;
1285 
1286 	switch (type) {
1287 	case MOD_LOAD:
1288 		if (loaded)
1289 			return (0);
1290 		newnfs_portinit();
1291 		mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1292 		    MTX_DEF);
1293 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1294 		nfscl_init();
1295 		NFSD_LOCK();
1296 		nfsrvd_cbinit(0);
1297 		NFSD_UNLOCK();
1298 		ncl_call_invalcaches = ncl_invalcaches;
1299 		nfsd_call_nfscl = nfssvc_nfscl;
1300 		loaded = 1;
1301 		break;
1302 
1303 	case MOD_UNLOAD:
1304 		if (nfs_numnfscbd != 0) {
1305 			error = EBUSY;
1306 			break;
1307 		}
1308 
1309 		/*
1310 		 * XXX: Unloading of nfscl module is unsupported.
1311 		 */
1312 #if 0
1313 		ncl_call_invalcaches = NULL;
1314 		nfsd_call_nfscl = NULL;
1315 		/* and get rid of the mutexes */
1316 		mtx_destroy(&nfs_clstate_mutex);
1317 		mtx_destroy(&ncl_iod_mutex);
1318 		loaded = 0;
1319 		break;
1320 #else
1321 		/* FALLTHROUGH */
1322 #endif
1323 	default:
1324 		error = EOPNOTSUPP;
1325 		break;
1326 	}
1327 	return error;
1328 }
1329 static moduledata_t nfscl_mod = {
1330 	"nfscl",
1331 	nfscl_modevent,
1332 	NULL,
1333 };
1334 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1335 
1336 /* So that loader and kldload(2) can find us, wherever we are.. */
1337 MODULE_VERSION(nfscl, 1);
1338 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1339 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1340 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1341 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1342 
1343