xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision 243e928310d073338c5ec089f0dce238a80b9866)
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  * Another 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 		 * It is safe so long as a vflush() with
336 		 * FORCECLOSE has not been done. Since the Renew thread is
337 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
338 		 * a vflush() with FORCECLOSE, we should be ok here.
339 		 */
340 		if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
341 			error = EINTR;
342 		else {
343 			vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
344 			if (nvp == NULL) {
345 				error = ENOENT;
346 			} else if ((nvp->v_iflag & VI_DOOMED) != 0) {
347 				error = ENOENT;
348 				vrele(nvp);
349 			} else {
350 				error = 0;
351 			}
352 		}
353 	}
354 	FREE(nfhp, M_NFSFH);
355 	if (error)
356 		return (error);
357 	if (nvp != NULL) {
358 		*npp = VTONFS(nvp);
359 		return (0);
360 	}
361 	return (EINVAL);
362 }
363 
364 static void
365 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
366     struct nfsvattr *newnap)
367 {
368 	int off;
369 
370 	if (ncl_fileid_maxwarnings >= 0 &&
371 	    ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
372 		return;
373 	off = 0;
374 	if (ncl_fileid_maxwarnings >= 0) {
375 		if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
376 			off = 1;
377 	}
378 
379 	printf("newnfs: server '%s' error: fileid changed. "
380 	    "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
381 	    "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
382 	    nmp->nm_com.nmcom_hostname,
383 	    (uintmax_t)nmp->nm_fsid[0],
384 	    (uintmax_t)nmp->nm_fsid[1],
385 	    (uintmax_t)oldnap->na_fileid,
386 	    (uintmax_t)newnap->na_fileid);
387 
388 	if (off)
389 		printf("newnfs: Logged %d times about fileid corruption; "
390 		    "going quiet to avoid spamming logs excessively. (Limit "
391 		    "is: %d).\n", ncl_fileid_nwarnings,
392 		    ncl_fileid_maxwarnings);
393 }
394 
395 /*
396  * Load the attribute cache (that lives in the nfsnode entry) with
397  * the attributes of the second argument and
398  * Iff vaper not NULL
399  *    copy the attributes to *vaper
400  * Similar to nfs_loadattrcache(), except the attributes are passed in
401  * instead of being parsed out of the mbuf list.
402  */
403 int
404 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
405     void *stuff, int writeattr, int dontshrink)
406 {
407 	struct vnode *vp = *vpp;
408 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
409 	struct nfsnode *np;
410 	struct nfsmount *nmp;
411 	struct timespec mtime_save;
412 	u_quad_t nsize;
413 	int setnsize, error, force_fid_err;
414 
415 	error = 0;
416 	setnsize = 0;
417 	nsize = 0;
418 
419 	/*
420 	 * If v_type == VNON it is a new node, so fill in the v_type,
421 	 * n_mtime fields. Check to see if it represents a special
422 	 * device, and if so, check for a possible alias. Once the
423 	 * correct vnode has been obtained, fill in the rest of the
424 	 * information.
425 	 */
426 	np = VTONFS(vp);
427 	NFSLOCKNODE(np);
428 	if (vp->v_type != nvap->va_type) {
429 		vp->v_type = nvap->va_type;
430 		if (vp->v_type == VFIFO)
431 			vp->v_op = &newnfs_fifoops;
432 		np->n_mtime = nvap->va_mtime;
433 	}
434 	nmp = VFSTONFS(vp->v_mount);
435 	vap = &np->n_vattr.na_vattr;
436 	mtime_save = vap->va_mtime;
437 	if (writeattr) {
438 		np->n_vattr.na_filerev = nap->na_filerev;
439 		np->n_vattr.na_size = nap->na_size;
440 		np->n_vattr.na_mtime = nap->na_mtime;
441 		np->n_vattr.na_ctime = nap->na_ctime;
442 		np->n_vattr.na_fsid = nap->na_fsid;
443 		np->n_vattr.na_mode = nap->na_mode;
444 	} else {
445 		force_fid_err = 0;
446 		KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
447 		    force_fid_err);
448 		/*
449 		 * BROKEN NFS SERVER OR MIDDLEWARE
450 		 *
451 		 * Certain NFS servers (certain old proprietary filers ca.
452 		 * 2006) or broken middleboxes (e.g. WAN accelerator products)
453 		 * will respond to GETATTR requests with results for a
454 		 * different fileid.
455 		 *
456 		 * The WAN accelerator we've observed not only serves stale
457 		 * cache results for a given file, it also occasionally serves
458 		 * results for wholly different files.  This causes surprising
459 		 * problems; for example the cached size attribute of a file
460 		 * may truncate down and then back up, resulting in zero
461 		 * regions in file contents read by applications.  We observed
462 		 * this reliably with Clang and .c files during parallel build.
463 		 * A pcap revealed packet fragmentation and GETATTR RPC
464 		 * responses with wholly wrong fileids.
465 		 */
466 		if ((np->n_vattr.na_fileid != 0 &&
467 		     np->n_vattr.na_fileid != nap->na_fileid) ||
468 		    force_fid_err) {
469 			nfscl_warn_fileid(nmp, &np->n_vattr, nap);
470 			error = EIDRM;
471 			goto out;
472 		}
473 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
474 		    sizeof (struct nfsvattr));
475 	}
476 
477 	/*
478 	 * For NFSv4, if the node's fsid is not equal to the mount point's
479 	 * fsid, return the low order 32bits of the node's fsid. This
480 	 * allows getcwd(3) to work. There is a chance that the fsid might
481 	 * be the same as a local fs, but since this is in an NFS mount
482 	 * point, I don't think that will cause any problems?
483 	 */
484 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
485 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
486 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
487 		/*
488 		 * va_fsid needs to be set to some value derived from
489 		 * np->n_vattr.na_filesid that is not equal
490 		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
491 		 * from the value used for the top level server volume
492 		 * in the mounted subtree.
493 		 */
494 		if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
495 		    (uint32_t)np->n_vattr.na_filesid[0])
496 			vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
497 		else
498 			vap->va_fsid = (uint32_t)hash32_buf(
499 			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
500 	} else
501 		vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
502 	np->n_attrstamp = time_second;
503 	if (vap->va_size != np->n_size) {
504 		if (vap->va_type == VREG) {
505 			if (dontshrink && vap->va_size < np->n_size) {
506 				/*
507 				 * We've been told not to shrink the file;
508 				 * zero np->n_attrstamp to indicate that
509 				 * the attributes are stale.
510 				 */
511 				vap->va_size = np->n_size;
512 				np->n_attrstamp = 0;
513 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
514 				vnode_pager_setsize(vp, np->n_size);
515 			} else if (np->n_flag & NMODIFIED) {
516 				/*
517 				 * We've modified the file: Use the larger
518 				 * of our size, and the server's size.
519 				 */
520 				if (vap->va_size < np->n_size) {
521 					vap->va_size = np->n_size;
522 				} else {
523 					np->n_size = vap->va_size;
524 					np->n_flag |= NSIZECHANGED;
525 				}
526 				vnode_pager_setsize(vp, np->n_size);
527 			} else if (vap->va_size < np->n_size) {
528 				/*
529 				 * When shrinking the size, the call to
530 				 * vnode_pager_setsize() cannot be done
531 				 * with the mutex held, so delay it until
532 				 * after the mtx_unlock call.
533 				 */
534 				nsize = np->n_size = vap->va_size;
535 				np->n_flag |= NSIZECHANGED;
536 				setnsize = 1;
537 			} else {
538 				np->n_size = vap->va_size;
539 				np->n_flag |= NSIZECHANGED;
540 				vnode_pager_setsize(vp, np->n_size);
541 			}
542 		} else {
543 			np->n_size = vap->va_size;
544 		}
545 	}
546 	/*
547 	 * The following checks are added to prevent a race between (say)
548 	 * a READDIR+ and a WRITE.
549 	 * READDIR+, WRITE requests sent out.
550 	 * READDIR+ resp, WRITE resp received on client.
551 	 * However, the WRITE resp was handled before the READDIR+ resp
552 	 * causing the post op attrs from the write to be loaded first
553 	 * and the attrs from the READDIR+ to be loaded later. If this
554 	 * happens, we have stale attrs loaded into the attrcache.
555 	 * We detect this by for the mtime moving back. We invalidate the
556 	 * attrcache when this happens.
557 	 */
558 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
559 		/* Size changed or mtime went backwards */
560 		np->n_attrstamp = 0;
561 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
562 	}
563 	if (vaper != NULL) {
564 		NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
565 		if (np->n_flag & NCHG) {
566 			if (np->n_flag & NACC)
567 				vaper->va_atime = np->n_atim;
568 			if (np->n_flag & NUPD)
569 				vaper->va_mtime = np->n_mtim;
570 		}
571 	}
572 
573 out:
574 #ifdef KDTRACE_HOOKS
575 	if (np->n_attrstamp != 0)
576 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
577 #endif
578 	NFSUNLOCKNODE(np);
579 	if (setnsize)
580 		vnode_pager_setsize(vp, nsize);
581 	return (error);
582 }
583 
584 /*
585  * Fill in the client id name. For these bytes:
586  * 1 - they must be unique
587  * 2 - they should be persistent across client reboots
588  * 1 is more critical than 2
589  * Use the mount point's unique id plus either the uuid or, if that
590  * isn't set, random junk.
591  */
592 void
593 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
594 {
595 	int uuidlen;
596 
597 	/*
598 	 * First, put in the 64bit mount point identifier.
599 	 */
600 	if (idlen >= sizeof (u_int64_t)) {
601 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
602 		cp += sizeof (u_int64_t);
603 		idlen -= sizeof (u_int64_t);
604 	}
605 
606 	/*
607 	 * If uuid is non-zero length, use it.
608 	 */
609 	uuidlen = strlen(uuid);
610 	if (uuidlen > 0 && idlen >= uuidlen) {
611 		NFSBCOPY(uuid, cp, uuidlen);
612 		cp += uuidlen;
613 		idlen -= uuidlen;
614 	}
615 
616 	/*
617 	 * This only normally happens if the uuid isn't set.
618 	 */
619 	while (idlen > 0) {
620 		*cp++ = (u_int8_t)(arc4random() % 256);
621 		idlen--;
622 	}
623 }
624 
625 /*
626  * Fill in a lock owner name. For now, pid + the process's creation time.
627  */
628 void
629 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
630 {
631 	union {
632 		u_int32_t	lval;
633 		u_int8_t	cval[4];
634 	} tl;
635 	struct proc *p;
636 
637 	if (id == NULL) {
638 		printf("NULL id\n");
639 		bzero(cp, NFSV4CL_LOCKNAMELEN);
640 		return;
641 	}
642 	if ((flags & F_POSIX) != 0) {
643 		p = (struct proc *)id;
644 		tl.lval = p->p_pid;
645 		*cp++ = tl.cval[0];
646 		*cp++ = tl.cval[1];
647 		*cp++ = tl.cval[2];
648 		*cp++ = tl.cval[3];
649 		tl.lval = p->p_stats->p_start.tv_sec;
650 		*cp++ = tl.cval[0];
651 		*cp++ = tl.cval[1];
652 		*cp++ = tl.cval[2];
653 		*cp++ = tl.cval[3];
654 		tl.lval = p->p_stats->p_start.tv_usec;
655 		*cp++ = tl.cval[0];
656 		*cp++ = tl.cval[1];
657 		*cp++ = tl.cval[2];
658 		*cp = tl.cval[3];
659 	} else if ((flags & F_FLOCK) != 0) {
660 		bcopy(&id, cp, sizeof(id));
661 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
662 	} else {
663 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
664 		bzero(cp, NFSV4CL_LOCKNAMELEN);
665 	}
666 }
667 
668 /*
669  * Find the parent process for the thread passed in as an argument.
670  * If none exists, return NULL, otherwise return a thread for the parent.
671  * (Can be any of the threads, since it is only used for td->td_proc.)
672  */
673 NFSPROC_T *
674 nfscl_getparent(struct thread *td)
675 {
676 	struct proc *p;
677 	struct thread *ptd;
678 
679 	if (td == NULL)
680 		return (NULL);
681 	p = td->td_proc;
682 	if (p->p_pid == 0)
683 		return (NULL);
684 	p = p->p_pptr;
685 	if (p == NULL)
686 		return (NULL);
687 	ptd = TAILQ_FIRST(&p->p_threads);
688 	return (ptd);
689 }
690 
691 /*
692  * Start up the renew kernel thread.
693  */
694 static void
695 start_nfscl(void *arg)
696 {
697 	struct nfsclclient *clp;
698 	struct thread *td;
699 
700 	clp = (struct nfsclclient *)arg;
701 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
702 	nfscl_renewthread(clp, td);
703 	kproc_exit(0);
704 }
705 
706 void
707 nfscl_start_renewthread(struct nfsclclient *clp)
708 {
709 
710 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
711 	    "nfscl");
712 }
713 
714 /*
715  * Handle wcc_data.
716  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
717  * as the first Op after PutFH.
718  * (For NFSv4, the postop attributes are after the Op, so they can't be
719  *  parsed here. A separate call to nfscl_postop_attr() is required.)
720  */
721 int
722 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
723     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
724 {
725 	u_int32_t *tl;
726 	struct nfsnode *np = VTONFS(vp);
727 	struct nfsvattr nfsva;
728 	int error = 0;
729 
730 	if (wccflagp != NULL)
731 		*wccflagp = 0;
732 	if (nd->nd_flag & ND_NFSV3) {
733 		*flagp = 0;
734 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
735 		if (*tl == newnfs_true) {
736 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
737 			if (wccflagp != NULL) {
738 				mtx_lock(&np->n_mtx);
739 				*wccflagp = (np->n_mtime.tv_sec ==
740 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
741 				    np->n_mtime.tv_nsec ==
742 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
743 				mtx_unlock(&np->n_mtx);
744 			}
745 		}
746 		error = nfscl_postop_attr(nd, nap, flagp, stuff);
747 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
748 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
749 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
750 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
751 		    NULL, NULL, NULL, NULL, NULL);
752 		if (error)
753 			return (error);
754 		/*
755 		 * Get rid of Op# and status for next op.
756 		 */
757 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
758 		if (*++tl)
759 			nd->nd_flag |= ND_NOMOREDATA;
760 		if (wccflagp != NULL &&
761 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
762 			mtx_lock(&np->n_mtx);
763 			*wccflagp = (np->n_mtime.tv_sec ==
764 			    nfsva.na_vattr.va_mtime.tv_sec &&
765 			    np->n_mtime.tv_nsec ==
766 			    nfsva.na_vattr.va_mtime.tv_sec);
767 			mtx_unlock(&np->n_mtx);
768 		}
769 	}
770 nfsmout:
771 	return (error);
772 }
773 
774 /*
775  * Get postop attributes.
776  */
777 int
778 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
779     void *stuff)
780 {
781 	u_int32_t *tl;
782 	int error = 0;
783 
784 	*retp = 0;
785 	if (nd->nd_flag & ND_NOMOREDATA)
786 		return (error);
787 	if (nd->nd_flag & ND_NFSV3) {
788 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
789 		*retp = fxdr_unsigned(int, *tl);
790 	} else if (nd->nd_flag & ND_NFSV4) {
791 		/*
792 		 * For NFSv4, the postop attr are at the end, so no point
793 		 * in looking if nd_repstat != 0.
794 		 */
795 		if (!nd->nd_repstat) {
796 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
797 			if (*(tl + 1))
798 				/* should never happen since nd_repstat != 0 */
799 				nd->nd_flag |= ND_NOMOREDATA;
800 			else
801 				*retp = 1;
802 		}
803 	} else if (!nd->nd_repstat) {
804 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
805 		*retp = 1;
806 	}
807 	if (*retp) {
808 		error = nfsm_loadattr(nd, nap);
809 		if (error)
810 			*retp = 0;
811 	}
812 nfsmout:
813 	return (error);
814 }
815 
816 /*
817  * Fill in the setable attributes. The full argument indicates whether
818  * to fill in them all or just mode and time.
819  */
820 void
821 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
822     struct vnode *vp, int flags, u_int32_t rdev)
823 {
824 	u_int32_t *tl;
825 	struct nfsv2_sattr *sp;
826 	nfsattrbit_t attrbits;
827 
828 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
829 	case ND_NFSV2:
830 		NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
831 		if (vap->va_mode == (mode_t)VNOVAL)
832 			sp->sa_mode = newnfs_xdrneg1;
833 		else
834 			sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
835 		if (vap->va_uid == (uid_t)VNOVAL)
836 			sp->sa_uid = newnfs_xdrneg1;
837 		else
838 			sp->sa_uid = txdr_unsigned(vap->va_uid);
839 		if (vap->va_gid == (gid_t)VNOVAL)
840 			sp->sa_gid = newnfs_xdrneg1;
841 		else
842 			sp->sa_gid = txdr_unsigned(vap->va_gid);
843 		if (flags & NFSSATTR_SIZE0)
844 			sp->sa_size = 0;
845 		else if (flags & NFSSATTR_SIZENEG1)
846 			sp->sa_size = newnfs_xdrneg1;
847 		else if (flags & NFSSATTR_SIZERDEV)
848 			sp->sa_size = txdr_unsigned(rdev);
849 		else
850 			sp->sa_size = txdr_unsigned(vap->va_size);
851 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
852 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
853 		break;
854 	case ND_NFSV3:
855 		if (vap->va_mode != (mode_t)VNOVAL) {
856 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
857 			*tl++ = newnfs_true;
858 			*tl = txdr_unsigned(vap->va_mode);
859 		} else {
860 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
861 			*tl = newnfs_false;
862 		}
863 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
864 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
865 			*tl++ = newnfs_true;
866 			*tl = txdr_unsigned(vap->va_uid);
867 		} else {
868 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
869 			*tl = newnfs_false;
870 		}
871 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
872 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
873 			*tl++ = newnfs_true;
874 			*tl = txdr_unsigned(vap->va_gid);
875 		} else {
876 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
877 			*tl = newnfs_false;
878 		}
879 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
880 			NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
881 			*tl++ = newnfs_true;
882 			txdr_hyper(vap->va_size, tl);
883 		} else {
884 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
885 			*tl = newnfs_false;
886 		}
887 		if (vap->va_atime.tv_sec != VNOVAL) {
888 			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
889 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
890 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
891 				txdr_nfsv3time(&vap->va_atime, tl);
892 			} else {
893 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
894 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
895 			}
896 		} else {
897 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
898 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
899 		}
900 		if (vap->va_mtime.tv_sec != VNOVAL) {
901 			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
902 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
903 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
904 				txdr_nfsv3time(&vap->va_mtime, tl);
905 			} else {
906 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
907 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
908 			}
909 		} else {
910 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
911 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
912 		}
913 		break;
914 	case ND_NFSV4:
915 		NFSZERO_ATTRBIT(&attrbits);
916 		if (vap->va_mode != (mode_t)VNOVAL)
917 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
918 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
919 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
920 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
921 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
922 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
923 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
924 		if (vap->va_atime.tv_sec != VNOVAL)
925 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
926 		if (vap->va_mtime.tv_sec != VNOVAL)
927 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
928 		(void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
929 		    &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
930 		break;
931 	}
932 }
933 
934 /*
935  * nfscl_request() - mostly a wrapper for newnfs_request().
936  */
937 int
938 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
939     struct ucred *cred, void *stuff)
940 {
941 	int ret, vers;
942 	struct nfsmount *nmp;
943 
944 	nmp = VFSTONFS(vp->v_mount);
945 	if (nd->nd_flag & ND_NFSV4)
946 		vers = NFS_VER4;
947 	else if (nd->nd_flag & ND_NFSV3)
948 		vers = NFS_VER3;
949 	else
950 		vers = NFS_VER2;
951 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
952 		NFS_PROG, vers, NULL, 1, NULL, NULL);
953 	return (ret);
954 }
955 
956 /*
957  * fill in this bsden's variant of statfs using nfsstatfs.
958  */
959 void
960 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
961 {
962 	struct statfs *sbp = (struct statfs *)statfs;
963 
964 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
965 		sbp->f_bsize = NFS_FABLKSIZE;
966 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
967 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
968 		/*
969 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
970 		 * the value after dividing by NFS_FABLKSIZE is small
971 		 * enough that it will fit in 63bits, so it is ok to
972 		 * assign it to f_bavail without fear that it will become
973 		 * negative.
974 		 */
975 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
976 		sbp->f_files = sfp->sf_tfiles;
977 		/* Since f_ffree is int64_t, clip it to 63bits. */
978 		if (sfp->sf_ffiles > INT64_MAX)
979 			sbp->f_ffree = INT64_MAX;
980 		else
981 			sbp->f_ffree = sfp->sf_ffiles;
982 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
983 		/*
984 		 * The type casts to (int32_t) ensure that this code is
985 		 * compatible with the old NFS client, in that it will
986 		 * propagate bit31 to the high order bits. This may or may
987 		 * not be correct for NFSv2, but since it is a legacy
988 		 * environment, I'd rather retain backwards compatibility.
989 		 */
990 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
991 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
992 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
993 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
994 		sbp->f_files = 0;
995 		sbp->f_ffree = 0;
996 	}
997 }
998 
999 /*
1000  * Use the fsinfo stuff to update the mount point.
1001  */
1002 void
1003 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
1004 {
1005 
1006 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
1007 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
1008 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
1009 		    ~(NFS_FABLKSIZE - 1);
1010 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
1011 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
1012 		if (nmp->nm_wsize == 0)
1013 			nmp->nm_wsize = fsp->fs_wtmax;
1014 	}
1015 	if (nmp->nm_wsize < NFS_FABLKSIZE)
1016 		nmp->nm_wsize = NFS_FABLKSIZE;
1017 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
1018 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
1019 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
1020 		    ~(NFS_FABLKSIZE - 1);
1021 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
1022 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
1023 		if (nmp->nm_rsize == 0)
1024 			nmp->nm_rsize = fsp->fs_rtmax;
1025 	}
1026 	if (nmp->nm_rsize < NFS_FABLKSIZE)
1027 		nmp->nm_rsize = NFS_FABLKSIZE;
1028 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
1029 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
1030 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
1031 		    ~(NFS_DIRBLKSIZ - 1);
1032 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
1033 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
1034 		if (nmp->nm_readdirsize == 0)
1035 			nmp->nm_readdirsize = fsp->fs_rtmax;
1036 	}
1037 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
1038 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
1039 	if (fsp->fs_maxfilesize > 0 &&
1040 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
1041 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
1042 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
1043 	nmp->nm_state |= NFSSTA_GOTFSINFO;
1044 }
1045 
1046 /*
1047  * Lookups source address which should be used to communicate with
1048  * @nmp and stores it inside @pdst.
1049  *
1050  * Returns 0 on success.
1051  */
1052 u_int8_t *
1053 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
1054 {
1055 #if defined(INET6) || defined(INET)
1056 	int error, fibnum;
1057 
1058 	fibnum = curthread->td_proc->p_fibnum;
1059 #endif
1060 #ifdef INET
1061 	if (nmp->nm_nam->sa_family == AF_INET) {
1062 		struct sockaddr_in *sin;
1063 		struct nhop4_extended nh_ext;
1064 
1065 		sin = (struct sockaddr_in *)nmp->nm_nam;
1066 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1067 		error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
1068 		    &nh_ext);
1069 		CURVNET_RESTORE();
1070 		if (error != 0)
1071 			return (NULL);
1072 
1073 		if ((ntohl(nh_ext.nh_src.s_addr) >> IN_CLASSA_NSHIFT) ==
1074 		    IN_LOOPBACKNET) {
1075 			/* Ignore loopback addresses */
1076 			return (NULL);
1077 		}
1078 
1079 		*isinet6p = 0;
1080 		*((struct in_addr *)paddr) = nh_ext.nh_src;
1081 
1082 		return (u_int8_t *)paddr;
1083 	}
1084 #endif
1085 #ifdef INET6
1086 	if (nmp->nm_nam->sa_family == AF_INET6) {
1087 		struct sockaddr_in6 *sin6;
1088 
1089 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1090 
1091 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1092 		error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1093 		    sin6->sin6_scope_id, NULL, paddr, NULL);
1094 		CURVNET_RESTORE();
1095 		if (error != 0)
1096 			return (NULL);
1097 
1098 		if (IN6_IS_ADDR_LOOPBACK(paddr))
1099 			return (NULL);
1100 
1101 		/* Scope is embedded in */
1102 		*isinet6p = 1;
1103 
1104 		return (u_int8_t *)paddr;
1105 	}
1106 #endif
1107 	return (NULL);
1108 }
1109 
1110 /*
1111  * Copy NFS uid, gids from the cred structure.
1112  */
1113 void
1114 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1115 {
1116 	int i;
1117 
1118 	KASSERT(cr->cr_ngroups >= 0,
1119 	    ("newnfs_copyincred: negative cr_ngroups"));
1120 	nfscr->nfsc_uid = cr->cr_uid;
1121 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1122 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1123 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1124 }
1125 
1126 
1127 /*
1128  * Do any client specific initialization.
1129  */
1130 void
1131 nfscl_init(void)
1132 {
1133 	static int inited = 0;
1134 
1135 	if (inited)
1136 		return;
1137 	inited = 1;
1138 	nfscl_inited = 1;
1139 	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1140 }
1141 
1142 /*
1143  * Check each of the attributes to be set, to ensure they aren't already
1144  * the correct value. Disable setting ones already correct.
1145  */
1146 int
1147 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1148 {
1149 
1150 	if (vap->va_mode != (mode_t)VNOVAL) {
1151 		if (vap->va_mode == nvap->na_mode)
1152 			vap->va_mode = (mode_t)VNOVAL;
1153 	}
1154 	if (vap->va_uid != (uid_t)VNOVAL) {
1155 		if (vap->va_uid == nvap->na_uid)
1156 			vap->va_uid = (uid_t)VNOVAL;
1157 	}
1158 	if (vap->va_gid != (gid_t)VNOVAL) {
1159 		if (vap->va_gid == nvap->na_gid)
1160 			vap->va_gid = (gid_t)VNOVAL;
1161 	}
1162 	if (vap->va_size != VNOVAL) {
1163 		if (vap->va_size == nvap->na_size)
1164 			vap->va_size = VNOVAL;
1165 	}
1166 
1167 	/*
1168 	 * We are normally called with only a partially initialized
1169 	 * VAP.  Since the NFSv3 spec says that server may use the
1170 	 * file attributes to store the verifier, the spec requires
1171 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1172 	 * in atime, but we can't really assume that all servers will
1173 	 * so we ensure that our SETATTR sets both atime and mtime.
1174 	 * Set the VA_UTIMES_NULL flag for this case, so that
1175 	 * the server's time will be used.  This is needed to
1176 	 * work around a bug in some Solaris servers, where
1177 	 * setting the time TOCLIENT causes the Setattr RPC
1178 	 * to return NFS_OK, but not set va_mode.
1179 	 */
1180 	if (vap->va_mtime.tv_sec == VNOVAL) {
1181 		vfs_timestamp(&vap->va_mtime);
1182 		vap->va_vaflags |= VA_UTIMES_NULL;
1183 	}
1184 	if (vap->va_atime.tv_sec == VNOVAL)
1185 		vap->va_atime = vap->va_mtime;
1186 	return (1);
1187 }
1188 
1189 /*
1190  * Map nfsv4 errors to errno.h errors.
1191  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1192  * error should only be returned for the Open, Create and Setattr Ops.
1193  * As such, most calls can just pass in 0 for those arguments.
1194  */
1195 APPLESTATIC int
1196 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1197 {
1198 	struct proc *p;
1199 
1200 	if (error < 10000)
1201 		return (error);
1202 	if (td != NULL)
1203 		p = td->td_proc;
1204 	else
1205 		p = NULL;
1206 	switch (error) {
1207 	case NFSERR_BADOWNER:
1208 		tprintf(p, LOG_INFO,
1209 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1210 		    uid, gid);
1211 		return (EPERM);
1212 	case NFSERR_BADNAME:
1213 	case NFSERR_BADCHAR:
1214 		printf("nfsv4 char/name not handled by server\n");
1215 		return (ENOENT);
1216 	case NFSERR_STALECLIENTID:
1217 	case NFSERR_STALESTATEID:
1218 	case NFSERR_EXPIRED:
1219 	case NFSERR_BADSTATEID:
1220 	case NFSERR_BADSESSION:
1221 		printf("nfsv4 recover err returned %d\n", error);
1222 		return (EIO);
1223 	case NFSERR_BADHANDLE:
1224 	case NFSERR_SERVERFAULT:
1225 	case NFSERR_BADTYPE:
1226 	case NFSERR_FHEXPIRED:
1227 	case NFSERR_RESOURCE:
1228 	case NFSERR_MOVED:
1229 	case NFSERR_NOFILEHANDLE:
1230 	case NFSERR_MINORVERMISMATCH:
1231 	case NFSERR_OLDSTATEID:
1232 	case NFSERR_BADSEQID:
1233 	case NFSERR_LEASEMOVED:
1234 	case NFSERR_RECLAIMBAD:
1235 	case NFSERR_BADXDR:
1236 	case NFSERR_OPILLEGAL:
1237 		printf("nfsv4 client/server protocol prob err=%d\n",
1238 		    error);
1239 		return (EIO);
1240 	default:
1241 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1242 		return (EIO);
1243 	};
1244 }
1245 
1246 /*
1247  * Check to see if the process for this owner exists. Return 1 if it doesn't
1248  * and 0 otherwise.
1249  */
1250 int
1251 nfscl_procdoesntexist(u_int8_t *own)
1252 {
1253 	union {
1254 		u_int32_t	lval;
1255 		u_int8_t	cval[4];
1256 	} tl;
1257 	struct proc *p;
1258 	pid_t pid;
1259 	int ret = 0;
1260 
1261 	tl.cval[0] = *own++;
1262 	tl.cval[1] = *own++;
1263 	tl.cval[2] = *own++;
1264 	tl.cval[3] = *own++;
1265 	pid = tl.lval;
1266 	p = pfind_locked(pid);
1267 	if (p == NULL)
1268 		return (1);
1269 	if (p->p_stats == NULL) {
1270 		PROC_UNLOCK(p);
1271 		return (0);
1272 	}
1273 	tl.cval[0] = *own++;
1274 	tl.cval[1] = *own++;
1275 	tl.cval[2] = *own++;
1276 	tl.cval[3] = *own++;
1277 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1278 		ret = 1;
1279 	} else {
1280 		tl.cval[0] = *own++;
1281 		tl.cval[1] = *own++;
1282 		tl.cval[2] = *own++;
1283 		tl.cval[3] = *own;
1284 		if (tl.lval != p->p_stats->p_start.tv_usec)
1285 			ret = 1;
1286 	}
1287 	PROC_UNLOCK(p);
1288 	return (ret);
1289 }
1290 
1291 /*
1292  * - nfs pseudo system call for the client
1293  */
1294 /*
1295  * MPSAFE
1296  */
1297 static int
1298 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1299 {
1300 	struct file *fp;
1301 	struct nfscbd_args nfscbdarg;
1302 	struct nfsd_nfscbd_args nfscbdarg2;
1303 	struct nameidata nd;
1304 	struct nfscl_dumpmntopts dumpmntopts;
1305 	cap_rights_t rights;
1306 	char *buf;
1307 	int error;
1308 
1309 	if (uap->flag & NFSSVC_CBADDSOCK) {
1310 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1311 		if (error)
1312 			return (error);
1313 		/*
1314 		 * Since we don't know what rights might be required,
1315 		 * pretend that we need them all. It is better to be too
1316 		 * careful than too reckless.
1317 		 */
1318 		error = fget(td, nfscbdarg.sock,
1319 		    cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1320 		if (error)
1321 			return (error);
1322 		if (fp->f_type != DTYPE_SOCKET) {
1323 			fdrop(fp, td);
1324 			return (EPERM);
1325 		}
1326 		error = nfscbd_addsock(fp);
1327 		fdrop(fp, td);
1328 		if (!error && nfscl_enablecallb == 0) {
1329 			nfsv4_cbport = nfscbdarg.port;
1330 			nfscl_enablecallb = 1;
1331 		}
1332 	} else if (uap->flag & NFSSVC_NFSCBD) {
1333 		if (uap->argp == NULL)
1334 			return (EINVAL);
1335 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1336 		    sizeof(nfscbdarg2));
1337 		if (error)
1338 			return (error);
1339 		error = nfscbd_nfsd(td, &nfscbdarg2);
1340 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1341 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1342 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1343 		    dumpmntopts.ndmnt_blen > 1024))
1344 			error = EINVAL;
1345 		if (error == 0)
1346 			error = nfsrv_lookupfilename(&nd,
1347 			    dumpmntopts.ndmnt_fname, td);
1348 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1349 		    "nfs") != 0) {
1350 			vput(nd.ni_vp);
1351 			error = EINVAL;
1352 		}
1353 		if (error == 0) {
1354 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1355 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1356 			    dumpmntopts.ndmnt_blen);
1357 			vput(nd.ni_vp);
1358 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1359 			    dumpmntopts.ndmnt_blen);
1360 			free(buf, M_TEMP);
1361 		}
1362 	} else {
1363 		error = EINVAL;
1364 	}
1365 	return (error);
1366 }
1367 
1368 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1369 
1370 /*
1371  * Called once to initialize data structures...
1372  */
1373 static int
1374 nfscl_modevent(module_t mod, int type, void *data)
1375 {
1376 	int error = 0;
1377 	static int loaded = 0;
1378 
1379 	switch (type) {
1380 	case MOD_LOAD:
1381 		if (loaded)
1382 			return (0);
1383 		newnfs_portinit();
1384 		mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1385 		    MTX_DEF);
1386 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1387 		nfscl_init();
1388 		NFSD_LOCK();
1389 		nfsrvd_cbinit(0);
1390 		NFSD_UNLOCK();
1391 		ncl_call_invalcaches = ncl_invalcaches;
1392 		nfsd_call_nfscl = nfssvc_nfscl;
1393 		loaded = 1;
1394 		break;
1395 
1396 	case MOD_UNLOAD:
1397 		if (nfs_numnfscbd != 0) {
1398 			error = EBUSY;
1399 			break;
1400 		}
1401 
1402 		/*
1403 		 * XXX: Unloading of nfscl module is unsupported.
1404 		 */
1405 #if 0
1406 		ncl_call_invalcaches = NULL;
1407 		nfsd_call_nfscl = NULL;
1408 		/* and get rid of the mutexes */
1409 		mtx_destroy(&nfs_clstate_mutex);
1410 		mtx_destroy(&ncl_iod_mutex);
1411 		loaded = 0;
1412 		break;
1413 #else
1414 		/* FALLTHROUGH */
1415 #endif
1416 	default:
1417 		error = EOPNOTSUPP;
1418 		break;
1419 	}
1420 	return error;
1421 }
1422 static moduledata_t nfscl_mod = {
1423 	"nfscl",
1424 	nfscl_modevent,
1425 	NULL,
1426 };
1427 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1428 
1429 /* So that loader and kldload(2) can find us, wherever we are.. */
1430 MODULE_VERSION(nfscl, 1);
1431 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1432 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1433 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1434 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1435 
1436