xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision 1f4bcc459a76b7aa664f3fd557684cd0ba6da352)
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_inet.h"
38 #include "opt_inet6.h"
39 
40 #include <sys/capsicum.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/fail.h>
47 #include <sys/hash.h>
48 #include <sys/sysctl.h>
49 #include <fs/nfs/nfsport.h>
50 #include <netinet/in_fib.h>
51 #include <netinet/if_ether.h>
52 #include <netinet6/ip6_var.h>
53 #include <net/if_types.h>
54 
55 #include <fs/nfsclient/nfs_kdtrace.h>
56 
57 #ifdef KDTRACE_HOOKS
58 dtrace_nfsclient_attrcache_flush_probe_func_t
59 		dtrace_nfscl_attrcache_flush_done_probe;
60 uint32_t	nfscl_attrcache_flush_done_id;
61 
62 dtrace_nfsclient_attrcache_get_hit_probe_func_t
63 		dtrace_nfscl_attrcache_get_hit_probe;
64 uint32_t	nfscl_attrcache_get_hit_id;
65 
66 dtrace_nfsclient_attrcache_get_miss_probe_func_t
67 		dtrace_nfscl_attrcache_get_miss_probe;
68 uint32_t	nfscl_attrcache_get_miss_id;
69 
70 dtrace_nfsclient_attrcache_load_probe_func_t
71 		dtrace_nfscl_attrcache_load_done_probe;
72 uint32_t	nfscl_attrcache_load_done_id;
73 #endif /* !KDTRACE_HOOKS */
74 
75 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
76 extern struct vop_vector newnfs_vnodeops;
77 extern struct vop_vector newnfs_fifoops;
78 extern uma_zone_t newnfsnode_zone;
79 extern struct buf_ops buf_ops_newnfs;
80 extern int ncl_pbuf_freecnt;
81 extern short nfsv4_cbport;
82 extern int nfscl_enablecallb;
83 extern int nfs_numnfscbd;
84 extern int nfscl_inited;
85 struct mtx nfs_clstate_mutex;
86 struct mtx ncl_iod_mutex;
87 NFSDLOCKMUTEX;
88 
89 extern void (*ncl_call_invalcaches)(struct vnode *);
90 
91 SYSCTL_DECL(_vfs_nfs);
92 static int ncl_fileid_maxwarnings = 10;
93 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
94     &ncl_fileid_maxwarnings, 0,
95     "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
96 static volatile int ncl_fileid_nwarnings;
97 
98 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
99     struct nfsvattr *);
100 
101 /*
102  * Comparison function for vfs_hash functions.
103  */
104 int
105 newnfs_vncmpf(struct vnode *vp, void *arg)
106 {
107 	struct nfsfh *nfhp = (struct nfsfh *)arg;
108 	struct nfsnode *np = VTONFS(vp);
109 
110 	if (np->n_fhp->nfh_len != nfhp->nfh_len ||
111 	    NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
112 		return (1);
113 	return (0);
114 }
115 
116 /*
117  * Look up a vnode/nfsnode by file handle.
118  * Callers must check for mount points!!
119  * In all cases, a pointer to a
120  * nfsnode structure is returned.
121  * This variant takes a "struct nfsfh *" as second argument and uses
122  * that structure up, either by hanging off the nfsnode or FREEing it.
123  */
124 int
125 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
126     struct componentname *cnp, struct thread *td, struct nfsnode **npp,
127     void *stuff, int lkflags)
128 {
129 	struct nfsnode *np, *dnp;
130 	struct vnode *vp, *nvp;
131 	struct nfsv4node *newd, *oldd;
132 	int error;
133 	u_int hash;
134 	struct nfsmount *nmp;
135 
136 	nmp = VFSTONFS(mntp);
137 	dnp = VTONFS(dvp);
138 	*npp = NULL;
139 
140 	hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
141 
142 	error = vfs_hash_get(mntp, hash, lkflags,
143 	    td, &nvp, newnfs_vncmpf, nfhp);
144 	if (error == 0 && nvp != NULL) {
145 		/*
146 		 * I believe there is a slight chance that vgonel() could
147 		 * get called on this vnode between when NFSVOPLOCK() drops
148 		 * the VI_LOCK() and vget() acquires it again, so that it
149 		 * hasn't yet had v_usecount incremented. If this were to
150 		 * happen, the VI_DOOMED flag would be set, so check for
151 		 * that here. Since we now have the v_usecount incremented,
152 		 * we should be ok until we vrele() it, if the VI_DOOMED
153 		 * flag isn't set now.
154 		 */
155 		VI_LOCK(nvp);
156 		if ((nvp->v_iflag & VI_DOOMED)) {
157 			VI_UNLOCK(nvp);
158 			vrele(nvp);
159 			error = ENOENT;
160 		} else {
161 			VI_UNLOCK(nvp);
162 		}
163 	}
164 	if (error) {
165 		FREE((caddr_t)nfhp, M_NFSFH);
166 		return (error);
167 	}
168 	if (nvp != NULL) {
169 		np = VTONFS(nvp);
170 		/*
171 		 * For NFSv4, check to see if it is the same name and
172 		 * replace the name, if it is different.
173 		 */
174 		oldd = newd = NULL;
175 		if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
176 		    nvp->v_type == VREG &&
177 		    (np->n_v4->n4_namelen != cnp->cn_namelen ||
178 		     NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
179 		     cnp->cn_namelen) ||
180 		     dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
181 		     NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
182 		     dnp->n_fhp->nfh_len))) {
183 		    MALLOC(newd, struct nfsv4node *,
184 			sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
185 			+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
186 		    NFSLOCKNODE(np);
187 		    if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
188 			&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
189 			 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
190 			 cnp->cn_namelen) ||
191 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
192 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
193 			 dnp->n_fhp->nfh_len))) {
194 			oldd = np->n_v4;
195 			np->n_v4 = newd;
196 			newd = NULL;
197 			np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
198 			np->n_v4->n4_namelen = cnp->cn_namelen;
199 			NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
200 			    dnp->n_fhp->nfh_len);
201 			NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
202 			    cnp->cn_namelen);
203 		    }
204 		    NFSUNLOCKNODE(np);
205 		}
206 		if (newd != NULL)
207 			FREE((caddr_t)newd, M_NFSV4NODE);
208 		if (oldd != NULL)
209 			FREE((caddr_t)oldd, M_NFSV4NODE);
210 		*npp = np;
211 		FREE((caddr_t)nfhp, M_NFSFH);
212 		return (0);
213 	}
214 	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
215 
216 	error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
217 	if (error) {
218 		uma_zfree(newnfsnode_zone, np);
219 		FREE((caddr_t)nfhp, M_NFSFH);
220 		return (error);
221 	}
222 	vp = nvp;
223 	KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
224 	vp->v_bufobj.bo_ops = &buf_ops_newnfs;
225 	vp->v_data = np;
226 	np->n_vnode = vp;
227 	/*
228 	 * Initialize the mutex even if the vnode is going to be a loser.
229 	 * This simplifies the logic in reclaim, which can then unconditionally
230 	 * destroy the mutex (in the case of the loser, or if hash_insert
231 	 * happened to return an error no special casing is needed).
232 	 */
233 	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
234 
235 	/*
236 	 * Are we getting the root? If so, make sure the vnode flags
237 	 * are correct
238 	 */
239 	if ((nfhp->nfh_len == nmp->nm_fhsize) &&
240 	    !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
241 		if (vp->v_type == VNON)
242 			vp->v_type = VDIR;
243 		vp->v_vflag |= VV_ROOT;
244 	}
245 
246 	np->n_fhp = nfhp;
247 	/*
248 	 * For NFSv4, we have to attach the directory file handle and
249 	 * file name, so that Open Ops can be done later.
250 	 */
251 	if (nmp->nm_flag & NFSMNT_NFSV4) {
252 		MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
253 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
254 		    M_WAITOK);
255 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
256 		np->n_v4->n4_namelen = cnp->cn_namelen;
257 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
258 		    dnp->n_fhp->nfh_len);
259 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
260 		    cnp->cn_namelen);
261 	} else {
262 		np->n_v4 = NULL;
263 	}
264 
265 	/*
266 	 * NFS supports recursive and shared locking.
267 	 */
268 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
269 	VN_LOCK_AREC(vp);
270 	VN_LOCK_ASHARE(vp);
271 	error = insmntque(vp, mntp);
272 	if (error != 0) {
273 		*npp = NULL;
274 		mtx_destroy(&np->n_mtx);
275 		FREE((caddr_t)nfhp, M_NFSFH);
276 		if (np->n_v4 != NULL)
277 			FREE((caddr_t)np->n_v4, M_NFSV4NODE);
278 		uma_zfree(newnfsnode_zone, np);
279 		return (error);
280 	}
281 	error = vfs_hash_insert(vp, hash, lkflags,
282 	    td, &nvp, newnfs_vncmpf, nfhp);
283 	if (error)
284 		return (error);
285 	if (nvp != NULL) {
286 		*npp = VTONFS(nvp);
287 		/* vfs_hash_insert() vput()'s the losing vnode */
288 		return (0);
289 	}
290 	*npp = np;
291 
292 	return (0);
293 }
294 
295 /*
296  * Anothe variant of nfs_nget(). This one is only used by reopen. It
297  * takes almost the same args as nfs_nget(), but only succeeds if an entry
298  * exists in the cache. (Since files should already be "open" with a
299  * vnode ref cnt on the node when reopen calls this, it should always
300  * succeed.)
301  * Also, don't get a vnode lock, since it may already be locked by some
302  * other process that is handling it. This is ok, since all other threads
303  * on the client are blocked by the nfsc_lock being exclusively held by the
304  * caller of this function.
305  */
306 int
307 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
308     struct thread *td, struct nfsnode **npp)
309 {
310 	struct vnode *nvp;
311 	u_int hash;
312 	struct nfsfh *nfhp;
313 	int error;
314 
315 	*npp = NULL;
316 	/* For forced dismounts, just return error. */
317 	if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
318 		return (EINTR);
319 	MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
320 	    M_NFSFH, M_WAITOK);
321 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
322 	nfhp->nfh_len = fhsize;
323 
324 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
325 
326 	/*
327 	 * First, try to get the vnode locked, but don't block for the lock.
328 	 */
329 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
330 	    newnfs_vncmpf, nfhp);
331 	if (error == 0 && nvp != NULL) {
332 		NFSVOPUNLOCK(nvp, 0);
333 	} else if (error == EBUSY) {
334 		/*
335 		 * The LK_EXCLOTHER lock type tells nfs_lock1() to not try
336 		 * and lock the vnode, but just get a v_usecount on it.
337 		 * LK_NOWAIT is set so that when vget() returns ENOENT,
338 		 * vfs_hash_get() fails instead of looping.
339 		 * If this succeeds, it is safe so long as a vflush() with
340 		 * FORCECLOSE has not been done. Since the Renew thread is
341 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
342 		 * a vflush() with FORCECLOSE, we should be ok here.
343 		 */
344 		if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
345 			error = EINTR;
346 		else
347 			error = vfs_hash_get(mntp, hash,
348 			    (LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
349 			    newnfs_vncmpf, nfhp);
350 	}
351 	FREE(nfhp, M_NFSFH);
352 	if (error)
353 		return (error);
354 	if (nvp != NULL) {
355 		*npp = VTONFS(nvp);
356 		return (0);
357 	}
358 	return (EINVAL);
359 }
360 
361 static void
362 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
363     struct nfsvattr *newnap)
364 {
365 	int off;
366 
367 	if (ncl_fileid_maxwarnings >= 0 &&
368 	    ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
369 		return;
370 	off = 0;
371 	if (ncl_fileid_maxwarnings >= 0) {
372 		if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
373 			off = 1;
374 	}
375 
376 	printf("newnfs: server '%s' error: fileid changed. "
377 	    "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
378 	    "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
379 	    nmp->nm_com.nmcom_hostname,
380 	    (uintmax_t)nmp->nm_fsid[0],
381 	    (uintmax_t)nmp->nm_fsid[1],
382 	    (uintmax_t)oldnap->na_fileid,
383 	    (uintmax_t)newnap->na_fileid);
384 
385 	if (off)
386 		printf("newnfs: Logged %d times about fileid corruption; "
387 		    "going quiet to avoid spamming logs excessively. (Limit "
388 		    "is: %d).\n", ncl_fileid_nwarnings,
389 		    ncl_fileid_maxwarnings);
390 }
391 
392 /*
393  * Load the attribute cache (that lives in the nfsnode entry) with
394  * the attributes of the second argument and
395  * Iff vaper not NULL
396  *    copy the attributes to *vaper
397  * Similar to nfs_loadattrcache(), except the attributes are passed in
398  * instead of being parsed out of the mbuf list.
399  */
400 int
401 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
402     void *stuff, int writeattr, int dontshrink)
403 {
404 	struct vnode *vp = *vpp;
405 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
406 	struct nfsnode *np;
407 	struct nfsmount *nmp;
408 	struct timespec mtime_save;
409 	u_quad_t nsize;
410 	int setnsize, error, force_fid_err;
411 
412 	error = 0;
413 	setnsize = 0;
414 	nsize = 0;
415 
416 	/*
417 	 * If v_type == VNON it is a new node, so fill in the v_type,
418 	 * n_mtime fields. Check to see if it represents a special
419 	 * device, and if so, check for a possible alias. Once the
420 	 * correct vnode has been obtained, fill in the rest of the
421 	 * information.
422 	 */
423 	np = VTONFS(vp);
424 	NFSLOCKNODE(np);
425 	if (vp->v_type != nvap->va_type) {
426 		vp->v_type = nvap->va_type;
427 		if (vp->v_type == VFIFO)
428 			vp->v_op = &newnfs_fifoops;
429 		np->n_mtime = nvap->va_mtime;
430 	}
431 	nmp = VFSTONFS(vp->v_mount);
432 	vap = &np->n_vattr.na_vattr;
433 	mtime_save = vap->va_mtime;
434 	if (writeattr) {
435 		np->n_vattr.na_filerev = nap->na_filerev;
436 		np->n_vattr.na_size = nap->na_size;
437 		np->n_vattr.na_mtime = nap->na_mtime;
438 		np->n_vattr.na_ctime = nap->na_ctime;
439 		np->n_vattr.na_fsid = nap->na_fsid;
440 		np->n_vattr.na_mode = nap->na_mode;
441 	} else {
442 		force_fid_err = 0;
443 		KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
444 		    force_fid_err);
445 		/*
446 		 * BROKEN NFS SERVER OR MIDDLEWARE
447 		 *
448 		 * Certain NFS servers (certain old proprietary filers ca.
449 		 * 2006) or broken middleboxes (e.g. WAN accelerator products)
450 		 * will respond to GETATTR requests with results for a
451 		 * different fileid.
452 		 *
453 		 * The WAN accelerator we've observed not only serves stale
454 		 * cache results for a given file, it also occasionally serves
455 		 * results for wholly different files.  This causes surprising
456 		 * problems; for example the cached size attribute of a file
457 		 * may truncate down and then back up, resulting in zero
458 		 * regions in file contents read by applications.  We observed
459 		 * this reliably with Clang and .c files during parallel build.
460 		 * A pcap revealed packet fragmentation and GETATTR RPC
461 		 * responses with wholly wrong fileids.
462 		 */
463 		if ((np->n_vattr.na_fileid != 0 &&
464 		     np->n_vattr.na_fileid != nap->na_fileid) ||
465 		    force_fid_err) {
466 			nfscl_warn_fileid(nmp, &np->n_vattr, nap);
467 			error = EIDRM;
468 			goto out;
469 		}
470 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
471 		    sizeof (struct nfsvattr));
472 	}
473 
474 	/*
475 	 * For NFSv4, if the node's fsid is not equal to the mount point's
476 	 * fsid, return the low order 32bits of the node's fsid. This
477 	 * allows getcwd(3) to work. There is a chance that the fsid might
478 	 * be the same as a local fs, but since this is in an NFS mount
479 	 * point, I don't think that will cause any problems?
480 	 */
481 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
482 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
483 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
484 		/*
485 		 * va_fsid needs to be set to some value derived from
486 		 * np->n_vattr.na_filesid that is not equal
487 		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
488 		 * from the value used for the top level server volume
489 		 * in the mounted subtree.
490 		 */
491 		if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
492 		    (uint32_t)np->n_vattr.na_filesid[0])
493 			vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
494 		else
495 			vap->va_fsid = (uint32_t)hash32_buf(
496 			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
497 	} else
498 		vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
499 	np->n_attrstamp = time_second;
500 	if (vap->va_size != np->n_size) {
501 		if (vap->va_type == VREG) {
502 			if (dontshrink && vap->va_size < np->n_size) {
503 				/*
504 				 * We've been told not to shrink the file;
505 				 * zero np->n_attrstamp to indicate that
506 				 * the attributes are stale.
507 				 */
508 				vap->va_size = np->n_size;
509 				np->n_attrstamp = 0;
510 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
511 				vnode_pager_setsize(vp, np->n_size);
512 			} else if (np->n_flag & NMODIFIED) {
513 				/*
514 				 * We've modified the file: Use the larger
515 				 * of our size, and the server's size.
516 				 */
517 				if (vap->va_size < np->n_size) {
518 					vap->va_size = np->n_size;
519 				} else {
520 					np->n_size = vap->va_size;
521 					np->n_flag |= NSIZECHANGED;
522 				}
523 				vnode_pager_setsize(vp, np->n_size);
524 			} else if (vap->va_size < np->n_size) {
525 				/*
526 				 * When shrinking the size, the call to
527 				 * vnode_pager_setsize() cannot be done
528 				 * with the mutex held, so delay it until
529 				 * after the mtx_unlock call.
530 				 */
531 				nsize = np->n_size = vap->va_size;
532 				np->n_flag |= NSIZECHANGED;
533 				setnsize = 1;
534 			} else {
535 				np->n_size = vap->va_size;
536 				np->n_flag |= NSIZECHANGED;
537 				vnode_pager_setsize(vp, np->n_size);
538 			}
539 		} else {
540 			np->n_size = vap->va_size;
541 		}
542 	}
543 	/*
544 	 * The following checks are added to prevent a race between (say)
545 	 * a READDIR+ and a WRITE.
546 	 * READDIR+, WRITE requests sent out.
547 	 * READDIR+ resp, WRITE resp received on client.
548 	 * However, the WRITE resp was handled before the READDIR+ resp
549 	 * causing the post op attrs from the write to be loaded first
550 	 * and the attrs from the READDIR+ to be loaded later. If this
551 	 * happens, we have stale attrs loaded into the attrcache.
552 	 * We detect this by for the mtime moving back. We invalidate the
553 	 * attrcache when this happens.
554 	 */
555 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
556 		/* Size changed or mtime went backwards */
557 		np->n_attrstamp = 0;
558 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
559 	}
560 	if (vaper != NULL) {
561 		NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
562 		if (np->n_flag & NCHG) {
563 			if (np->n_flag & NACC)
564 				vaper->va_atime = np->n_atim;
565 			if (np->n_flag & NUPD)
566 				vaper->va_mtime = np->n_mtim;
567 		}
568 	}
569 
570 out:
571 #ifdef KDTRACE_HOOKS
572 	if (np->n_attrstamp != 0)
573 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
574 #endif
575 	NFSUNLOCKNODE(np);
576 	if (setnsize)
577 		vnode_pager_setsize(vp, nsize);
578 	return (error);
579 }
580 
581 /*
582  * Fill in the client id name. For these bytes:
583  * 1 - they must be unique
584  * 2 - they should be persistent across client reboots
585  * 1 is more critical than 2
586  * Use the mount point's unique id plus either the uuid or, if that
587  * isn't set, random junk.
588  */
589 void
590 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
591 {
592 	int uuidlen;
593 
594 	/*
595 	 * First, put in the 64bit mount point identifier.
596 	 */
597 	if (idlen >= sizeof (u_int64_t)) {
598 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
599 		cp += sizeof (u_int64_t);
600 		idlen -= sizeof (u_int64_t);
601 	}
602 
603 	/*
604 	 * If uuid is non-zero length, use it.
605 	 */
606 	uuidlen = strlen(uuid);
607 	if (uuidlen > 0 && idlen >= uuidlen) {
608 		NFSBCOPY(uuid, cp, uuidlen);
609 		cp += uuidlen;
610 		idlen -= uuidlen;
611 	}
612 
613 	/*
614 	 * This only normally happens if the uuid isn't set.
615 	 */
616 	while (idlen > 0) {
617 		*cp++ = (u_int8_t)(arc4random() % 256);
618 		idlen--;
619 	}
620 }
621 
622 /*
623  * Fill in a lock owner name. For now, pid + the process's creation time.
624  */
625 void
626 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
627 {
628 	union {
629 		u_int32_t	lval;
630 		u_int8_t	cval[4];
631 	} tl;
632 	struct proc *p;
633 
634 	if (id == NULL) {
635 		printf("NULL id\n");
636 		bzero(cp, NFSV4CL_LOCKNAMELEN);
637 		return;
638 	}
639 	if ((flags & F_POSIX) != 0) {
640 		p = (struct proc *)id;
641 		tl.lval = p->p_pid;
642 		*cp++ = tl.cval[0];
643 		*cp++ = tl.cval[1];
644 		*cp++ = tl.cval[2];
645 		*cp++ = tl.cval[3];
646 		tl.lval = p->p_stats->p_start.tv_sec;
647 		*cp++ = tl.cval[0];
648 		*cp++ = tl.cval[1];
649 		*cp++ = tl.cval[2];
650 		*cp++ = tl.cval[3];
651 		tl.lval = p->p_stats->p_start.tv_usec;
652 		*cp++ = tl.cval[0];
653 		*cp++ = tl.cval[1];
654 		*cp++ = tl.cval[2];
655 		*cp = tl.cval[3];
656 	} else if ((flags & F_FLOCK) != 0) {
657 		bcopy(&id, cp, sizeof(id));
658 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
659 	} else {
660 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
661 		bzero(cp, NFSV4CL_LOCKNAMELEN);
662 	}
663 }
664 
665 /*
666  * Find the parent process for the thread passed in as an argument.
667  * If none exists, return NULL, otherwise return a thread for the parent.
668  * (Can be any of the threads, since it is only used for td->td_proc.)
669  */
670 NFSPROC_T *
671 nfscl_getparent(struct thread *td)
672 {
673 	struct proc *p;
674 	struct thread *ptd;
675 
676 	if (td == NULL)
677 		return (NULL);
678 	p = td->td_proc;
679 	if (p->p_pid == 0)
680 		return (NULL);
681 	p = p->p_pptr;
682 	if (p == NULL)
683 		return (NULL);
684 	ptd = TAILQ_FIRST(&p->p_threads);
685 	return (ptd);
686 }
687 
688 /*
689  * Start up the renew kernel thread.
690  */
691 static void
692 start_nfscl(void *arg)
693 {
694 	struct nfsclclient *clp;
695 	struct thread *td;
696 
697 	clp = (struct nfsclclient *)arg;
698 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
699 	nfscl_renewthread(clp, td);
700 	kproc_exit(0);
701 }
702 
703 void
704 nfscl_start_renewthread(struct nfsclclient *clp)
705 {
706 
707 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
708 	    "nfscl");
709 }
710 
711 /*
712  * Handle wcc_data.
713  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
714  * as the first Op after PutFH.
715  * (For NFSv4, the postop attributes are after the Op, so they can't be
716  *  parsed here. A separate call to nfscl_postop_attr() is required.)
717  */
718 int
719 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
720     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
721 {
722 	u_int32_t *tl;
723 	struct nfsnode *np = VTONFS(vp);
724 	struct nfsvattr nfsva;
725 	int error = 0;
726 
727 	if (wccflagp != NULL)
728 		*wccflagp = 0;
729 	if (nd->nd_flag & ND_NFSV3) {
730 		*flagp = 0;
731 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
732 		if (*tl == newnfs_true) {
733 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
734 			if (wccflagp != NULL) {
735 				mtx_lock(&np->n_mtx);
736 				*wccflagp = (np->n_mtime.tv_sec ==
737 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
738 				    np->n_mtime.tv_nsec ==
739 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
740 				mtx_unlock(&np->n_mtx);
741 			}
742 		}
743 		error = nfscl_postop_attr(nd, nap, flagp, stuff);
744 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
745 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
746 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
747 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
748 		    NULL, NULL, NULL, NULL, NULL);
749 		if (error)
750 			return (error);
751 		/*
752 		 * Get rid of Op# and status for next op.
753 		 */
754 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
755 		if (*++tl)
756 			nd->nd_flag |= ND_NOMOREDATA;
757 		if (wccflagp != NULL &&
758 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
759 			mtx_lock(&np->n_mtx);
760 			*wccflagp = (np->n_mtime.tv_sec ==
761 			    nfsva.na_vattr.va_mtime.tv_sec &&
762 			    np->n_mtime.tv_nsec ==
763 			    nfsva.na_vattr.va_mtime.tv_sec);
764 			mtx_unlock(&np->n_mtx);
765 		}
766 	}
767 nfsmout:
768 	return (error);
769 }
770 
771 /*
772  * Get postop attributes.
773  */
774 int
775 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
776     void *stuff)
777 {
778 	u_int32_t *tl;
779 	int error = 0;
780 
781 	*retp = 0;
782 	if (nd->nd_flag & ND_NOMOREDATA)
783 		return (error);
784 	if (nd->nd_flag & ND_NFSV3) {
785 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
786 		*retp = fxdr_unsigned(int, *tl);
787 	} else if (nd->nd_flag & ND_NFSV4) {
788 		/*
789 		 * For NFSv4, the postop attr are at the end, so no point
790 		 * in looking if nd_repstat != 0.
791 		 */
792 		if (!nd->nd_repstat) {
793 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
794 			if (*(tl + 1))
795 				/* should never happen since nd_repstat != 0 */
796 				nd->nd_flag |= ND_NOMOREDATA;
797 			else
798 				*retp = 1;
799 		}
800 	} else if (!nd->nd_repstat) {
801 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
802 		*retp = 1;
803 	}
804 	if (*retp) {
805 		error = nfsm_loadattr(nd, nap);
806 		if (error)
807 			*retp = 0;
808 	}
809 nfsmout:
810 	return (error);
811 }
812 
813 /*
814  * Fill in the setable attributes. The full argument indicates whether
815  * to fill in them all or just mode and time.
816  */
817 void
818 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
819     struct vnode *vp, int flags, u_int32_t rdev)
820 {
821 	u_int32_t *tl;
822 	struct nfsv2_sattr *sp;
823 	nfsattrbit_t attrbits;
824 
825 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
826 	case ND_NFSV2:
827 		NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
828 		if (vap->va_mode == (mode_t)VNOVAL)
829 			sp->sa_mode = newnfs_xdrneg1;
830 		else
831 			sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
832 		if (vap->va_uid == (uid_t)VNOVAL)
833 			sp->sa_uid = newnfs_xdrneg1;
834 		else
835 			sp->sa_uid = txdr_unsigned(vap->va_uid);
836 		if (vap->va_gid == (gid_t)VNOVAL)
837 			sp->sa_gid = newnfs_xdrneg1;
838 		else
839 			sp->sa_gid = txdr_unsigned(vap->va_gid);
840 		if (flags & NFSSATTR_SIZE0)
841 			sp->sa_size = 0;
842 		else if (flags & NFSSATTR_SIZENEG1)
843 			sp->sa_size = newnfs_xdrneg1;
844 		else if (flags & NFSSATTR_SIZERDEV)
845 			sp->sa_size = txdr_unsigned(rdev);
846 		else
847 			sp->sa_size = txdr_unsigned(vap->va_size);
848 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
849 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
850 		break;
851 	case ND_NFSV3:
852 		if (vap->va_mode != (mode_t)VNOVAL) {
853 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
854 			*tl++ = newnfs_true;
855 			*tl = txdr_unsigned(vap->va_mode);
856 		} else {
857 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
858 			*tl = newnfs_false;
859 		}
860 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
861 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
862 			*tl++ = newnfs_true;
863 			*tl = txdr_unsigned(vap->va_uid);
864 		} else {
865 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
866 			*tl = newnfs_false;
867 		}
868 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
869 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
870 			*tl++ = newnfs_true;
871 			*tl = txdr_unsigned(vap->va_gid);
872 		} else {
873 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
874 			*tl = newnfs_false;
875 		}
876 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
877 			NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
878 			*tl++ = newnfs_true;
879 			txdr_hyper(vap->va_size, tl);
880 		} else {
881 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
882 			*tl = newnfs_false;
883 		}
884 		if (vap->va_atime.tv_sec != VNOVAL) {
885 			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
886 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
887 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
888 				txdr_nfsv3time(&vap->va_atime, tl);
889 			} else {
890 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
891 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
892 			}
893 		} else {
894 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
895 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
896 		}
897 		if (vap->va_mtime.tv_sec != VNOVAL) {
898 			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
899 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
900 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
901 				txdr_nfsv3time(&vap->va_mtime, tl);
902 			} else {
903 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
904 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
905 			}
906 		} else {
907 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
908 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
909 		}
910 		break;
911 	case ND_NFSV4:
912 		NFSZERO_ATTRBIT(&attrbits);
913 		if (vap->va_mode != (mode_t)VNOVAL)
914 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
915 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
916 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
917 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
918 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
919 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
920 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
921 		if (vap->va_atime.tv_sec != VNOVAL)
922 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
923 		if (vap->va_mtime.tv_sec != VNOVAL)
924 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
925 		(void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
926 		    &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
927 		break;
928 	};
929 }
930 
931 /*
932  * nfscl_request() - mostly a wrapper for newnfs_request().
933  */
934 int
935 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
936     struct ucred *cred, void *stuff)
937 {
938 	int ret, vers;
939 	struct nfsmount *nmp;
940 
941 	nmp = VFSTONFS(vp->v_mount);
942 	if (nd->nd_flag & ND_NFSV4)
943 		vers = NFS_VER4;
944 	else if (nd->nd_flag & ND_NFSV3)
945 		vers = NFS_VER3;
946 	else
947 		vers = NFS_VER2;
948 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
949 		NFS_PROG, vers, NULL, 1, NULL, NULL);
950 	return (ret);
951 }
952 
953 /*
954  * fill in this bsden's variant of statfs using nfsstatfs.
955  */
956 void
957 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
958 {
959 	struct statfs *sbp = (struct statfs *)statfs;
960 
961 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
962 		sbp->f_bsize = NFS_FABLKSIZE;
963 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
964 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
965 		/*
966 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
967 		 * the value after dividing by NFS_FABLKSIZE is small
968 		 * enough that it will fit in 63bits, so it is ok to
969 		 * assign it to f_bavail without fear that it will become
970 		 * negative.
971 		 */
972 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
973 		sbp->f_files = sfp->sf_tfiles;
974 		/* Since f_ffree is int64_t, clip it to 63bits. */
975 		if (sfp->sf_ffiles > INT64_MAX)
976 			sbp->f_ffree = INT64_MAX;
977 		else
978 			sbp->f_ffree = sfp->sf_ffiles;
979 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
980 		/*
981 		 * The type casts to (int32_t) ensure that this code is
982 		 * compatible with the old NFS client, in that it will
983 		 * propagate bit31 to the high order bits. This may or may
984 		 * not be correct for NFSv2, but since it is a legacy
985 		 * environment, I'd rather retain backwards compatibility.
986 		 */
987 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
988 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
989 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
990 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
991 		sbp->f_files = 0;
992 		sbp->f_ffree = 0;
993 	}
994 }
995 
996 /*
997  * Use the fsinfo stuff to update the mount point.
998  */
999 void
1000 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
1001 {
1002 
1003 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
1004 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
1005 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
1006 		    ~(NFS_FABLKSIZE - 1);
1007 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
1008 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
1009 		if (nmp->nm_wsize == 0)
1010 			nmp->nm_wsize = fsp->fs_wtmax;
1011 	}
1012 	if (nmp->nm_wsize < NFS_FABLKSIZE)
1013 		nmp->nm_wsize = NFS_FABLKSIZE;
1014 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
1015 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
1016 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
1017 		    ~(NFS_FABLKSIZE - 1);
1018 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
1019 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
1020 		if (nmp->nm_rsize == 0)
1021 			nmp->nm_rsize = fsp->fs_rtmax;
1022 	}
1023 	if (nmp->nm_rsize < NFS_FABLKSIZE)
1024 		nmp->nm_rsize = NFS_FABLKSIZE;
1025 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
1026 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
1027 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
1028 		    ~(NFS_DIRBLKSIZ - 1);
1029 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
1030 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
1031 		if (nmp->nm_readdirsize == 0)
1032 			nmp->nm_readdirsize = fsp->fs_rtmax;
1033 	}
1034 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
1035 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
1036 	if (fsp->fs_maxfilesize > 0 &&
1037 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
1038 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
1039 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
1040 	nmp->nm_state |= NFSSTA_GOTFSINFO;
1041 }
1042 
1043 /*
1044  * Lookups source address which should be used to communicate with
1045  * @nmp and stores it inside @pdst.
1046  *
1047  * Returns 0 on success.
1048  */
1049 u_int8_t *
1050 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
1051 {
1052 #if defined(INET6) || defined(INET)
1053 	int error, fibnum;
1054 
1055 	fibnum = curthread->td_proc->p_fibnum;
1056 #endif
1057 #ifdef INET
1058 	if (nmp->nm_nam->sa_family == AF_INET) {
1059 		struct sockaddr_in *sin;
1060 		struct nhop4_extended nh_ext;
1061 
1062 		sin = (struct sockaddr_in *)nmp->nm_nam;
1063 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1064 		error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
1065 		    &nh_ext);
1066 		CURVNET_RESTORE();
1067 		if (error != 0)
1068 			return (NULL);
1069 
1070 		if ((ntohl(nh_ext.nh_src.s_addr) >> IN_CLASSA_NSHIFT) ==
1071 		    IN_LOOPBACKNET) {
1072 			/* Ignore loopback addresses */
1073 			return (NULL);
1074 		}
1075 
1076 		*isinet6p = 0;
1077 		*((struct in_addr *)paddr) = nh_ext.nh_src;
1078 
1079 		return (u_int8_t *)paddr;
1080 	}
1081 #endif
1082 #ifdef INET6
1083 	if (nmp->nm_nam->sa_family == AF_INET6) {
1084 		struct sockaddr_in6 *sin6;
1085 
1086 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1087 
1088 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1089 		error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1090 		    sin6->sin6_scope_id, NULL, paddr, NULL);
1091 		CURVNET_RESTORE();
1092 		if (error != 0)
1093 			return (NULL);
1094 
1095 		if (IN6_IS_ADDR_LOOPBACK(paddr))
1096 			return (NULL);
1097 
1098 		/* Scope is embedded in */
1099 		*isinet6p = 1;
1100 
1101 		return (u_int8_t *)paddr;
1102 	}
1103 #endif
1104 	return (NULL);
1105 }
1106 
1107 /*
1108  * Copy NFS uid, gids from the cred structure.
1109  */
1110 void
1111 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1112 {
1113 	int i;
1114 
1115 	KASSERT(cr->cr_ngroups >= 0,
1116 	    ("newnfs_copyincred: negative cr_ngroups"));
1117 	nfscr->nfsc_uid = cr->cr_uid;
1118 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1119 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1120 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1121 }
1122 
1123 
1124 /*
1125  * Do any client specific initialization.
1126  */
1127 void
1128 nfscl_init(void)
1129 {
1130 	static int inited = 0;
1131 
1132 	if (inited)
1133 		return;
1134 	inited = 1;
1135 	nfscl_inited = 1;
1136 	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1137 }
1138 
1139 /*
1140  * Check each of the attributes to be set, to ensure they aren't already
1141  * the correct value. Disable setting ones already correct.
1142  */
1143 int
1144 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1145 {
1146 
1147 	if (vap->va_mode != (mode_t)VNOVAL) {
1148 		if (vap->va_mode == nvap->na_mode)
1149 			vap->va_mode = (mode_t)VNOVAL;
1150 	}
1151 	if (vap->va_uid != (uid_t)VNOVAL) {
1152 		if (vap->va_uid == nvap->na_uid)
1153 			vap->va_uid = (uid_t)VNOVAL;
1154 	}
1155 	if (vap->va_gid != (gid_t)VNOVAL) {
1156 		if (vap->va_gid == nvap->na_gid)
1157 			vap->va_gid = (gid_t)VNOVAL;
1158 	}
1159 	if (vap->va_size != VNOVAL) {
1160 		if (vap->va_size == nvap->na_size)
1161 			vap->va_size = VNOVAL;
1162 	}
1163 
1164 	/*
1165 	 * We are normally called with only a partially initialized
1166 	 * VAP.  Since the NFSv3 spec says that server may use the
1167 	 * file attributes to store the verifier, the spec requires
1168 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1169 	 * in atime, but we can't really assume that all servers will
1170 	 * so we ensure that our SETATTR sets both atime and mtime.
1171 	 * Set the VA_UTIMES_NULL flag for this case, so that
1172 	 * the server's time will be used.  This is needed to
1173 	 * work around a bug in some Solaris servers, where
1174 	 * setting the time TOCLIENT causes the Setattr RPC
1175 	 * to return NFS_OK, but not set va_mode.
1176 	 */
1177 	if (vap->va_mtime.tv_sec == VNOVAL) {
1178 		vfs_timestamp(&vap->va_mtime);
1179 		vap->va_vaflags |= VA_UTIMES_NULL;
1180 	}
1181 	if (vap->va_atime.tv_sec == VNOVAL)
1182 		vap->va_atime = vap->va_mtime;
1183 	return (1);
1184 }
1185 
1186 /*
1187  * Map nfsv4 errors to errno.h errors.
1188  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1189  * error should only be returned for the Open, Create and Setattr Ops.
1190  * As such, most calls can just pass in 0 for those arguments.
1191  */
1192 APPLESTATIC int
1193 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1194 {
1195 	struct proc *p;
1196 
1197 	if (error < 10000)
1198 		return (error);
1199 	if (td != NULL)
1200 		p = td->td_proc;
1201 	else
1202 		p = NULL;
1203 	switch (error) {
1204 	case NFSERR_BADOWNER:
1205 		tprintf(p, LOG_INFO,
1206 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1207 		    uid, gid);
1208 		return (EPERM);
1209 	case NFSERR_BADNAME:
1210 	case NFSERR_BADCHAR:
1211 		printf("nfsv4 char/name not handled by server\n");
1212 		return (ENOENT);
1213 	case NFSERR_STALECLIENTID:
1214 	case NFSERR_STALESTATEID:
1215 	case NFSERR_EXPIRED:
1216 	case NFSERR_BADSTATEID:
1217 	case NFSERR_BADSESSION:
1218 		printf("nfsv4 recover err returned %d\n", error);
1219 		return (EIO);
1220 	case NFSERR_BADHANDLE:
1221 	case NFSERR_SERVERFAULT:
1222 	case NFSERR_BADTYPE:
1223 	case NFSERR_FHEXPIRED:
1224 	case NFSERR_RESOURCE:
1225 	case NFSERR_MOVED:
1226 	case NFSERR_NOFILEHANDLE:
1227 	case NFSERR_MINORVERMISMATCH:
1228 	case NFSERR_OLDSTATEID:
1229 	case NFSERR_BADSEQID:
1230 	case NFSERR_LEASEMOVED:
1231 	case NFSERR_RECLAIMBAD:
1232 	case NFSERR_BADXDR:
1233 	case NFSERR_OPILLEGAL:
1234 		printf("nfsv4 client/server protocol prob err=%d\n",
1235 		    error);
1236 		return (EIO);
1237 	default:
1238 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1239 		return (EIO);
1240 	};
1241 }
1242 
1243 /*
1244  * Check to see if the process for this owner exists. Return 1 if it doesn't
1245  * and 0 otherwise.
1246  */
1247 int
1248 nfscl_procdoesntexist(u_int8_t *own)
1249 {
1250 	union {
1251 		u_int32_t	lval;
1252 		u_int8_t	cval[4];
1253 	} tl;
1254 	struct proc *p;
1255 	pid_t pid;
1256 	int ret = 0;
1257 
1258 	tl.cval[0] = *own++;
1259 	tl.cval[1] = *own++;
1260 	tl.cval[2] = *own++;
1261 	tl.cval[3] = *own++;
1262 	pid = tl.lval;
1263 	p = pfind_locked(pid);
1264 	if (p == NULL)
1265 		return (1);
1266 	if (p->p_stats == NULL) {
1267 		PROC_UNLOCK(p);
1268 		return (0);
1269 	}
1270 	tl.cval[0] = *own++;
1271 	tl.cval[1] = *own++;
1272 	tl.cval[2] = *own++;
1273 	tl.cval[3] = *own++;
1274 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1275 		ret = 1;
1276 	} else {
1277 		tl.cval[0] = *own++;
1278 		tl.cval[1] = *own++;
1279 		tl.cval[2] = *own++;
1280 		tl.cval[3] = *own;
1281 		if (tl.lval != p->p_stats->p_start.tv_usec)
1282 			ret = 1;
1283 	}
1284 	PROC_UNLOCK(p);
1285 	return (ret);
1286 }
1287 
1288 /*
1289  * - nfs pseudo system call for the client
1290  */
1291 /*
1292  * MPSAFE
1293  */
1294 static int
1295 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1296 {
1297 	struct file *fp;
1298 	struct nfscbd_args nfscbdarg;
1299 	struct nfsd_nfscbd_args nfscbdarg2;
1300 	struct nameidata nd;
1301 	struct nfscl_dumpmntopts dumpmntopts;
1302 	cap_rights_t rights;
1303 	char *buf;
1304 	int error;
1305 
1306 	if (uap->flag & NFSSVC_CBADDSOCK) {
1307 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1308 		if (error)
1309 			return (error);
1310 		/*
1311 		 * Since we don't know what rights might be required,
1312 		 * pretend that we need them all. It is better to be too
1313 		 * careful than too reckless.
1314 		 */
1315 		error = fget(td, nfscbdarg.sock,
1316 		    cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1317 		if (error)
1318 			return (error);
1319 		if (fp->f_type != DTYPE_SOCKET) {
1320 			fdrop(fp, td);
1321 			return (EPERM);
1322 		}
1323 		error = nfscbd_addsock(fp);
1324 		fdrop(fp, td);
1325 		if (!error && nfscl_enablecallb == 0) {
1326 			nfsv4_cbport = nfscbdarg.port;
1327 			nfscl_enablecallb = 1;
1328 		}
1329 	} else if (uap->flag & NFSSVC_NFSCBD) {
1330 		if (uap->argp == NULL)
1331 			return (EINVAL);
1332 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1333 		    sizeof(nfscbdarg2));
1334 		if (error)
1335 			return (error);
1336 		error = nfscbd_nfsd(td, &nfscbdarg2);
1337 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1338 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1339 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1340 		    dumpmntopts.ndmnt_blen > 1024))
1341 			error = EINVAL;
1342 		if (error == 0)
1343 			error = nfsrv_lookupfilename(&nd,
1344 			    dumpmntopts.ndmnt_fname, td);
1345 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1346 		    "nfs") != 0) {
1347 			vput(nd.ni_vp);
1348 			error = EINVAL;
1349 		}
1350 		if (error == 0) {
1351 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1352 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1353 			    dumpmntopts.ndmnt_blen);
1354 			vput(nd.ni_vp);
1355 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1356 			    dumpmntopts.ndmnt_blen);
1357 			free(buf, M_TEMP);
1358 		}
1359 	} else {
1360 		error = EINVAL;
1361 	}
1362 	return (error);
1363 }
1364 
1365 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1366 
1367 /*
1368  * Called once to initialize data structures...
1369  */
1370 static int
1371 nfscl_modevent(module_t mod, int type, void *data)
1372 {
1373 	int error = 0;
1374 	static int loaded = 0;
1375 
1376 	switch (type) {
1377 	case MOD_LOAD:
1378 		if (loaded)
1379 			return (0);
1380 		newnfs_portinit();
1381 		mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1382 		    MTX_DEF);
1383 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1384 		nfscl_init();
1385 		NFSD_LOCK();
1386 		nfsrvd_cbinit(0);
1387 		NFSD_UNLOCK();
1388 		ncl_call_invalcaches = ncl_invalcaches;
1389 		nfsd_call_nfscl = nfssvc_nfscl;
1390 		loaded = 1;
1391 		break;
1392 
1393 	case MOD_UNLOAD:
1394 		if (nfs_numnfscbd != 0) {
1395 			error = EBUSY;
1396 			break;
1397 		}
1398 
1399 		/*
1400 		 * XXX: Unloading of nfscl module is unsupported.
1401 		 */
1402 #if 0
1403 		ncl_call_invalcaches = NULL;
1404 		nfsd_call_nfscl = NULL;
1405 		/* and get rid of the mutexes */
1406 		mtx_destroy(&nfs_clstate_mutex);
1407 		mtx_destroy(&ncl_iod_mutex);
1408 		loaded = 0;
1409 		break;
1410 #else
1411 		/* FALLTHROUGH */
1412 #endif
1413 	default:
1414 		error = EOPNOTSUPP;
1415 		break;
1416 	}
1417 	return error;
1418 }
1419 static moduledata_t nfscl_mod = {
1420 	"nfscl",
1421 	nfscl_modevent,
1422 	NULL,
1423 };
1424 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1425 
1426 /* So that loader and kldload(2) can find us, wherever we are.. */
1427 MODULE_VERSION(nfscl, 1);
1428 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1429 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1430 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1431 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1432 
1433