xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision 8657387683946d0c03e09fe77029edfe309eeb20)
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  * 3. 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 ncl_iod_mutex;
86 NFSDLOCKMUTEX;
87 
88 extern void (*ncl_call_invalcaches)(struct vnode *);
89 
90 SYSCTL_DECL(_vfs_nfs);
91 static int ncl_fileid_maxwarnings = 10;
92 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
93     &ncl_fileid_maxwarnings, 0,
94     "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
95 static volatile int ncl_fileid_nwarnings;
96 
97 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
98     struct nfsvattr *);
99 
100 /*
101  * Comparison function for vfs_hash functions.
102  */
103 int
104 newnfs_vncmpf(struct vnode *vp, void *arg)
105 {
106 	struct nfsfh *nfhp = (struct nfsfh *)arg;
107 	struct nfsnode *np = VTONFS(vp);
108 
109 	if (np->n_fhp->nfh_len != nfhp->nfh_len ||
110 	    NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
111 		return (1);
112 	return (0);
113 }
114 
115 /*
116  * Look up a vnode/nfsnode by file handle.
117  * Callers must check for mount points!!
118  * In all cases, a pointer to a
119  * nfsnode structure is returned.
120  * This variant takes a "struct nfsfh *" as second argument and uses
121  * that structure up, either by hanging off the nfsnode or FREEing it.
122  */
123 int
124 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
125     struct componentname *cnp, struct thread *td, struct nfsnode **npp,
126     void *stuff, int lkflags)
127 {
128 	struct nfsnode *np, *dnp;
129 	struct vnode *vp, *nvp;
130 	struct nfsv4node *newd, *oldd;
131 	int error;
132 	u_int hash;
133 	struct nfsmount *nmp;
134 
135 	nmp = VFSTONFS(mntp);
136 	dnp = VTONFS(dvp);
137 	*npp = NULL;
138 
139 	hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
140 
141 	error = vfs_hash_get(mntp, hash, lkflags,
142 	    td, &nvp, newnfs_vncmpf, nfhp);
143 	if (error == 0 && nvp != NULL) {
144 		/*
145 		 * I believe there is a slight chance that vgonel() could
146 		 * get called on this vnode between when NFSVOPLOCK() drops
147 		 * the VI_LOCK() and vget() acquires it again, so that it
148 		 * hasn't yet had v_usecount incremented. If this were to
149 		 * happen, the VI_DOOMED flag would be set, so check for
150 		 * that here. Since we now have the v_usecount incremented,
151 		 * we should be ok until we vrele() it, if the VI_DOOMED
152 		 * flag isn't set now.
153 		 */
154 		VI_LOCK(nvp);
155 		if ((nvp->v_iflag & VI_DOOMED)) {
156 			VI_UNLOCK(nvp);
157 			vrele(nvp);
158 			error = ENOENT;
159 		} else {
160 			VI_UNLOCK(nvp);
161 		}
162 	}
163 	if (error) {
164 		FREE((caddr_t)nfhp, M_NFSFH);
165 		return (error);
166 	}
167 	if (nvp != NULL) {
168 		np = VTONFS(nvp);
169 		/*
170 		 * For NFSv4, check to see if it is the same name and
171 		 * replace the name, if it is different.
172 		 */
173 		oldd = newd = NULL;
174 		if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
175 		    nvp->v_type == VREG &&
176 		    (np->n_v4->n4_namelen != cnp->cn_namelen ||
177 		     NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
178 		     cnp->cn_namelen) ||
179 		     dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
180 		     NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
181 		     dnp->n_fhp->nfh_len))) {
182 		    MALLOC(newd, struct nfsv4node *,
183 			sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
184 			+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
185 		    NFSLOCKNODE(np);
186 		    if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
187 			&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
188 			 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
189 			 cnp->cn_namelen) ||
190 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
191 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
192 			 dnp->n_fhp->nfh_len))) {
193 			oldd = np->n_v4;
194 			np->n_v4 = newd;
195 			newd = NULL;
196 			np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
197 			np->n_v4->n4_namelen = cnp->cn_namelen;
198 			NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
199 			    dnp->n_fhp->nfh_len);
200 			NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
201 			    cnp->cn_namelen);
202 		    }
203 		    NFSUNLOCKNODE(np);
204 		}
205 		if (newd != NULL)
206 			FREE((caddr_t)newd, M_NFSV4NODE);
207 		if (oldd != NULL)
208 			FREE((caddr_t)oldd, M_NFSV4NODE);
209 		*npp = np;
210 		FREE((caddr_t)nfhp, M_NFSFH);
211 		return (0);
212 	}
213 	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
214 
215 	error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
216 	if (error) {
217 		uma_zfree(newnfsnode_zone, np);
218 		FREE((caddr_t)nfhp, M_NFSFH);
219 		return (error);
220 	}
221 	vp = nvp;
222 	KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
223 	vp->v_bufobj.bo_ops = &buf_ops_newnfs;
224 	vp->v_data = np;
225 	np->n_vnode = vp;
226 	/*
227 	 * Initialize the mutex even if the vnode is going to be a loser.
228 	 * This simplifies the logic in reclaim, which can then unconditionally
229 	 * destroy the mutex (in the case of the loser, or if hash_insert
230 	 * happened to return an error no special casing is needed).
231 	 */
232 	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
233 	lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE |
234 	    LK_CANRECURSE);
235 
236 	/*
237 	 * Are we getting the root? If so, make sure the vnode flags
238 	 * are correct
239 	 */
240 	if ((nfhp->nfh_len == nmp->nm_fhsize) &&
241 	    !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
242 		if (vp->v_type == VNON)
243 			vp->v_type = VDIR;
244 		vp->v_vflag |= VV_ROOT;
245 	}
246 
247 	np->n_fhp = nfhp;
248 	/*
249 	 * For NFSv4, we have to attach the directory file handle and
250 	 * file name, so that Open Ops can be done later.
251 	 */
252 	if (nmp->nm_flag & NFSMNT_NFSV4) {
253 		MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
254 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
255 		    M_WAITOK);
256 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
257 		np->n_v4->n4_namelen = cnp->cn_namelen;
258 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
259 		    dnp->n_fhp->nfh_len);
260 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
261 		    cnp->cn_namelen);
262 	} else {
263 		np->n_v4 = NULL;
264 	}
265 
266 	/*
267 	 * NFS supports recursive and shared locking.
268 	 */
269 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
270 	VN_LOCK_AREC(vp);
271 	VN_LOCK_ASHARE(vp);
272 	error = insmntque(vp, mntp);
273 	if (error != 0) {
274 		*npp = NULL;
275 		mtx_destroy(&np->n_mtx);
276 		lockdestroy(&np->n_excl);
277 		FREE((caddr_t)nfhp, M_NFSFH);
278 		if (np->n_v4 != NULL)
279 			FREE((caddr_t)np->n_v4, M_NFSV4NODE);
280 		uma_zfree(newnfsnode_zone, np);
281 		return (error);
282 	}
283 	error = vfs_hash_insert(vp, hash, lkflags,
284 	    td, &nvp, newnfs_vncmpf, nfhp);
285 	if (error)
286 		return (error);
287 	if (nvp != NULL) {
288 		*npp = VTONFS(nvp);
289 		/* vfs_hash_insert() vput()'s the losing vnode */
290 		return (0);
291 	}
292 	*npp = np;
293 
294 	return (0);
295 }
296 
297 /*
298  * Another variant of nfs_nget(). This one is only used by reopen. It
299  * takes almost the same args as nfs_nget(), but only succeeds if an entry
300  * exists in the cache. (Since files should already be "open" with a
301  * vnode ref cnt on the node when reopen calls this, it should always
302  * succeed.)
303  * Also, don't get a vnode lock, since it may already be locked by some
304  * other process that is handling it. This is ok, since all other threads
305  * on the client are blocked by the nfsc_lock being exclusively held by the
306  * caller of this function.
307  */
308 int
309 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
310     struct thread *td, struct nfsnode **npp)
311 {
312 	struct vnode *nvp;
313 	u_int hash;
314 	struct nfsfh *nfhp;
315 	int error;
316 
317 	*npp = NULL;
318 	/* For forced dismounts, just return error. */
319 	if (NFSCL_FORCEDISM(mntp))
320 		return (EINTR);
321 	MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
322 	    M_NFSFH, M_WAITOK);
323 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
324 	nfhp->nfh_len = fhsize;
325 
326 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
327 
328 	/*
329 	 * First, try to get the vnode locked, but don't block for the lock.
330 	 */
331 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
332 	    newnfs_vncmpf, nfhp);
333 	if (error == 0 && nvp != NULL) {
334 		NFSVOPUNLOCK(nvp, 0);
335 	} else if (error == EBUSY) {
336 		/*
337 		 * It is safe so long as a vflush() with
338 		 * FORCECLOSE has not been done. Since the Renew thread is
339 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
340 		 * a vflush() with FORCECLOSE, we should be ok here.
341 		 */
342 		if (NFSCL_FORCEDISM(mntp))
343 			error = EINTR;
344 		else {
345 			vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
346 			if (nvp == NULL) {
347 				error = ENOENT;
348 			} else if ((nvp->v_iflag & VI_DOOMED) != 0) {
349 				error = ENOENT;
350 				vrele(nvp);
351 			} else {
352 				error = 0;
353 			}
354 		}
355 	}
356 	FREE(nfhp, M_NFSFH);
357 	if (error)
358 		return (error);
359 	if (nvp != NULL) {
360 		*npp = VTONFS(nvp);
361 		return (0);
362 	}
363 	return (EINVAL);
364 }
365 
366 static void
367 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
368     struct nfsvattr *newnap)
369 {
370 	int off;
371 
372 	if (ncl_fileid_maxwarnings >= 0 &&
373 	    ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
374 		return;
375 	off = 0;
376 	if (ncl_fileid_maxwarnings >= 0) {
377 		if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
378 			off = 1;
379 	}
380 
381 	printf("newnfs: server '%s' error: fileid changed. "
382 	    "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
383 	    "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
384 	    nmp->nm_com.nmcom_hostname,
385 	    (uintmax_t)nmp->nm_fsid[0],
386 	    (uintmax_t)nmp->nm_fsid[1],
387 	    (uintmax_t)oldnap->na_fileid,
388 	    (uintmax_t)newnap->na_fileid);
389 
390 	if (off)
391 		printf("newnfs: Logged %d times about fileid corruption; "
392 		    "going quiet to avoid spamming logs excessively. (Limit "
393 		    "is: %d).\n", ncl_fileid_nwarnings,
394 		    ncl_fileid_maxwarnings);
395 }
396 
397 /*
398  * Load the attribute cache (that lives in the nfsnode entry) with
399  * the attributes of the second argument and
400  * Iff vaper not NULL
401  *    copy the attributes to *vaper
402  * Similar to nfs_loadattrcache(), except the attributes are passed in
403  * instead of being parsed out of the mbuf list.
404  */
405 int
406 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
407     void *stuff, int writeattr, int dontshrink)
408 {
409 	struct vnode *vp = *vpp;
410 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
411 	struct nfsnode *np;
412 	struct nfsmount *nmp;
413 	struct timespec mtime_save;
414 	u_quad_t nsize;
415 	int setnsize, error, force_fid_err;
416 
417 	error = 0;
418 	setnsize = 0;
419 	nsize = 0;
420 
421 	/*
422 	 * If v_type == VNON it is a new node, so fill in the v_type,
423 	 * n_mtime fields. Check to see if it represents a special
424 	 * device, and if so, check for a possible alias. Once the
425 	 * correct vnode has been obtained, fill in the rest of the
426 	 * information.
427 	 */
428 	np = VTONFS(vp);
429 	NFSLOCKNODE(np);
430 	if (vp->v_type != nvap->va_type) {
431 		vp->v_type = nvap->va_type;
432 		if (vp->v_type == VFIFO)
433 			vp->v_op = &newnfs_fifoops;
434 		np->n_mtime = nvap->va_mtime;
435 	}
436 	nmp = VFSTONFS(vp->v_mount);
437 	vap = &np->n_vattr.na_vattr;
438 	mtime_save = vap->va_mtime;
439 	if (writeattr) {
440 		np->n_vattr.na_filerev = nap->na_filerev;
441 		np->n_vattr.na_size = nap->na_size;
442 		np->n_vattr.na_mtime = nap->na_mtime;
443 		np->n_vattr.na_ctime = nap->na_ctime;
444 		np->n_vattr.na_fsid = nap->na_fsid;
445 		np->n_vattr.na_mode = nap->na_mode;
446 	} else {
447 		force_fid_err = 0;
448 		KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
449 		    force_fid_err);
450 		/*
451 		 * BROKEN NFS SERVER OR MIDDLEWARE
452 		 *
453 		 * Certain NFS servers (certain old proprietary filers ca.
454 		 * 2006) or broken middleboxes (e.g. WAN accelerator products)
455 		 * will respond to GETATTR requests with results for a
456 		 * different fileid.
457 		 *
458 		 * The WAN accelerator we've observed not only serves stale
459 		 * cache results for a given file, it also occasionally serves
460 		 * results for wholly different files.  This causes surprising
461 		 * problems; for example the cached size attribute of a file
462 		 * may truncate down and then back up, resulting in zero
463 		 * regions in file contents read by applications.  We observed
464 		 * this reliably with Clang and .c files during parallel build.
465 		 * A pcap revealed packet fragmentation and GETATTR RPC
466 		 * responses with wholly wrong fileids.
467 		 */
468 		if ((np->n_vattr.na_fileid != 0 &&
469 		     np->n_vattr.na_fileid != nap->na_fileid) ||
470 		    force_fid_err) {
471 			nfscl_warn_fileid(nmp, &np->n_vattr, nap);
472 			error = EIDRM;
473 			goto out;
474 		}
475 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
476 		    sizeof (struct nfsvattr));
477 	}
478 
479 	/*
480 	 * For NFSv4, if the node's fsid is not equal to the mount point's
481 	 * fsid, return the low order 32bits of the node's fsid. This
482 	 * allows getcwd(3) to work. There is a chance that the fsid might
483 	 * be the same as a local fs, but since this is in an NFS mount
484 	 * point, I don't think that will cause any problems?
485 	 */
486 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
487 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
488 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
489 		/*
490 		 * va_fsid needs to be set to some value derived from
491 		 * np->n_vattr.na_filesid that is not equal
492 		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
493 		 * from the value used for the top level server volume
494 		 * in the mounted subtree.
495 		 */
496 		vn_fsid(vp, vap);
497 		if ((uint32_t)vap->va_fsid == np->n_vattr.na_filesid[0])
498 			vap->va_fsid = hash32_buf(
499 			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
500 	} else
501 		vn_fsid(vp, vap);
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 		/* Return the single open_owner of all 0 bytes. */
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 		if (wccflagp != NULL && *flagp == 0)
748 			*wccflagp = 0;
749 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
750 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
751 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
752 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
753 		    NULL, NULL, NULL, NULL, NULL);
754 		if (error)
755 			return (error);
756 		/*
757 		 * Get rid of Op# and status for next op.
758 		 */
759 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
760 		if (*++tl)
761 			nd->nd_flag |= ND_NOMOREDATA;
762 		if (wccflagp != NULL &&
763 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
764 			mtx_lock(&np->n_mtx);
765 			*wccflagp = (np->n_mtime.tv_sec ==
766 			    nfsva.na_vattr.va_mtime.tv_sec &&
767 			    np->n_mtime.tv_nsec ==
768 			    nfsva.na_vattr.va_mtime.tv_sec);
769 			mtx_unlock(&np->n_mtx);
770 		}
771 	}
772 nfsmout:
773 	return (error);
774 }
775 
776 /*
777  * Get postop attributes.
778  */
779 int
780 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
781     void *stuff)
782 {
783 	u_int32_t *tl;
784 	int error = 0;
785 
786 	*retp = 0;
787 	if (nd->nd_flag & ND_NOMOREDATA)
788 		return (error);
789 	if (nd->nd_flag & ND_NFSV3) {
790 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
791 		*retp = fxdr_unsigned(int, *tl);
792 	} else if (nd->nd_flag & ND_NFSV4) {
793 		/*
794 		 * For NFSv4, the postop attr are at the end, so no point
795 		 * in looking if nd_repstat != 0.
796 		 */
797 		if (!nd->nd_repstat) {
798 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
799 			if (*(tl + 1))
800 				/* should never happen since nd_repstat != 0 */
801 				nd->nd_flag |= ND_NOMOREDATA;
802 			else
803 				*retp = 1;
804 		}
805 	} else if (!nd->nd_repstat) {
806 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
807 		*retp = 1;
808 	}
809 	if (*retp) {
810 		error = nfsm_loadattr(nd, nap);
811 		if (error)
812 			*retp = 0;
813 	}
814 nfsmout:
815 	return (error);
816 }
817 
818 /*
819  * Fill in the setable attributes. The full argument indicates whether
820  * to fill in them all or just mode and time.
821  */
822 void
823 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
824     struct vnode *vp, int flags, u_int32_t rdev)
825 {
826 	u_int32_t *tl;
827 	struct nfsv2_sattr *sp;
828 	nfsattrbit_t attrbits;
829 
830 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
831 	case ND_NFSV2:
832 		NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
833 		if (vap->va_mode == (mode_t)VNOVAL)
834 			sp->sa_mode = newnfs_xdrneg1;
835 		else
836 			sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
837 		if (vap->va_uid == (uid_t)VNOVAL)
838 			sp->sa_uid = newnfs_xdrneg1;
839 		else
840 			sp->sa_uid = txdr_unsigned(vap->va_uid);
841 		if (vap->va_gid == (gid_t)VNOVAL)
842 			sp->sa_gid = newnfs_xdrneg1;
843 		else
844 			sp->sa_gid = txdr_unsigned(vap->va_gid);
845 		if (flags & NFSSATTR_SIZE0)
846 			sp->sa_size = 0;
847 		else if (flags & NFSSATTR_SIZENEG1)
848 			sp->sa_size = newnfs_xdrneg1;
849 		else if (flags & NFSSATTR_SIZERDEV)
850 			sp->sa_size = txdr_unsigned(rdev);
851 		else
852 			sp->sa_size = txdr_unsigned(vap->va_size);
853 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
854 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
855 		break;
856 	case ND_NFSV3:
857 		if (vap->va_mode != (mode_t)VNOVAL) {
858 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
859 			*tl++ = newnfs_true;
860 			*tl = txdr_unsigned(vap->va_mode);
861 		} else {
862 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
863 			*tl = newnfs_false;
864 		}
865 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
866 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
867 			*tl++ = newnfs_true;
868 			*tl = txdr_unsigned(vap->va_uid);
869 		} else {
870 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
871 			*tl = newnfs_false;
872 		}
873 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
874 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
875 			*tl++ = newnfs_true;
876 			*tl = txdr_unsigned(vap->va_gid);
877 		} else {
878 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
879 			*tl = newnfs_false;
880 		}
881 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
882 			NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
883 			*tl++ = newnfs_true;
884 			txdr_hyper(vap->va_size, tl);
885 		} else {
886 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
887 			*tl = newnfs_false;
888 		}
889 		if (vap->va_atime.tv_sec != VNOVAL) {
890 			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
891 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
892 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
893 				txdr_nfsv3time(&vap->va_atime, tl);
894 			} else {
895 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
896 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
897 			}
898 		} else {
899 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
900 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
901 		}
902 		if (vap->va_mtime.tv_sec != VNOVAL) {
903 			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
904 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
905 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
906 				txdr_nfsv3time(&vap->va_mtime, tl);
907 			} else {
908 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
909 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
910 			}
911 		} else {
912 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
913 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
914 		}
915 		break;
916 	case ND_NFSV4:
917 		NFSZERO_ATTRBIT(&attrbits);
918 		if (vap->va_mode != (mode_t)VNOVAL)
919 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
920 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
921 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
922 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
923 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
924 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
925 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
926 		if (vap->va_atime.tv_sec != VNOVAL)
927 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
928 		if (vap->va_mtime.tv_sec != VNOVAL)
929 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
930 		(void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
931 		    &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
932 		break;
933 	}
934 }
935 
936 /*
937  * nfscl_request() - mostly a wrapper for newnfs_request().
938  */
939 int
940 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
941     struct ucred *cred, void *stuff)
942 {
943 	int ret, vers;
944 	struct nfsmount *nmp;
945 
946 	nmp = VFSTONFS(vp->v_mount);
947 	if (nd->nd_flag & ND_NFSV4)
948 		vers = NFS_VER4;
949 	else if (nd->nd_flag & ND_NFSV3)
950 		vers = NFS_VER3;
951 	else
952 		vers = NFS_VER2;
953 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
954 		NFS_PROG, vers, NULL, 1, NULL, NULL);
955 	return (ret);
956 }
957 
958 /*
959  * fill in this bsden's variant of statfs using nfsstatfs.
960  */
961 void
962 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
963 {
964 	struct statfs *sbp = (struct statfs *)statfs;
965 
966 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
967 		sbp->f_bsize = NFS_FABLKSIZE;
968 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
969 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
970 		/*
971 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
972 		 * the value after dividing by NFS_FABLKSIZE is small
973 		 * enough that it will fit in 63bits, so it is ok to
974 		 * assign it to f_bavail without fear that it will become
975 		 * negative.
976 		 */
977 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
978 		sbp->f_files = sfp->sf_tfiles;
979 		/* Since f_ffree is int64_t, clip it to 63bits. */
980 		if (sfp->sf_ffiles > INT64_MAX)
981 			sbp->f_ffree = INT64_MAX;
982 		else
983 			sbp->f_ffree = sfp->sf_ffiles;
984 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
985 		/*
986 		 * The type casts to (int32_t) ensure that this code is
987 		 * compatible with the old NFS client, in that it will
988 		 * propagate bit31 to the high order bits. This may or may
989 		 * not be correct for NFSv2, but since it is a legacy
990 		 * environment, I'd rather retain backwards compatibility.
991 		 */
992 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
993 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
994 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
995 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
996 		sbp->f_files = 0;
997 		sbp->f_ffree = 0;
998 	}
999 }
1000 
1001 /*
1002  * Use the fsinfo stuff to update the mount point.
1003  */
1004 void
1005 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
1006 {
1007 
1008 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
1009 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
1010 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
1011 		    ~(NFS_FABLKSIZE - 1);
1012 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
1013 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
1014 		if (nmp->nm_wsize == 0)
1015 			nmp->nm_wsize = fsp->fs_wtmax;
1016 	}
1017 	if (nmp->nm_wsize < NFS_FABLKSIZE)
1018 		nmp->nm_wsize = NFS_FABLKSIZE;
1019 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
1020 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
1021 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
1022 		    ~(NFS_FABLKSIZE - 1);
1023 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
1024 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
1025 		if (nmp->nm_rsize == 0)
1026 			nmp->nm_rsize = fsp->fs_rtmax;
1027 	}
1028 	if (nmp->nm_rsize < NFS_FABLKSIZE)
1029 		nmp->nm_rsize = NFS_FABLKSIZE;
1030 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
1031 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
1032 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
1033 		    ~(NFS_DIRBLKSIZ - 1);
1034 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
1035 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
1036 		if (nmp->nm_readdirsize == 0)
1037 			nmp->nm_readdirsize = fsp->fs_rtmax;
1038 	}
1039 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
1040 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
1041 	if (fsp->fs_maxfilesize > 0 &&
1042 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
1043 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
1044 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
1045 	nmp->nm_state |= NFSSTA_GOTFSINFO;
1046 }
1047 
1048 /*
1049  * Lookups source address which should be used to communicate with
1050  * @nmp and stores it inside @pdst.
1051  *
1052  * Returns 0 on success.
1053  */
1054 u_int8_t *
1055 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
1056 {
1057 #if defined(INET6) || defined(INET)
1058 	int error, fibnum;
1059 
1060 	fibnum = curthread->td_proc->p_fibnum;
1061 #endif
1062 #ifdef INET
1063 	if (nmp->nm_nam->sa_family == AF_INET) {
1064 		struct sockaddr_in *sin;
1065 		struct nhop4_extended nh_ext;
1066 
1067 		sin = (struct sockaddr_in *)nmp->nm_nam;
1068 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1069 		error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
1070 		    &nh_ext);
1071 		CURVNET_RESTORE();
1072 		if (error != 0)
1073 			return (NULL);
1074 
1075 		if ((ntohl(nh_ext.nh_src.s_addr) >> IN_CLASSA_NSHIFT) ==
1076 		    IN_LOOPBACKNET) {
1077 			/* Ignore loopback addresses */
1078 			return (NULL);
1079 		}
1080 
1081 		*isinet6p = 0;
1082 		*((struct in_addr *)paddr) = nh_ext.nh_src;
1083 
1084 		return (u_int8_t *)paddr;
1085 	}
1086 #endif
1087 #ifdef INET6
1088 	if (nmp->nm_nam->sa_family == AF_INET6) {
1089 		struct sockaddr_in6 *sin6;
1090 
1091 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1092 
1093 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1094 		error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1095 		    sin6->sin6_scope_id, NULL, paddr, NULL);
1096 		CURVNET_RESTORE();
1097 		if (error != 0)
1098 			return (NULL);
1099 
1100 		if (IN6_IS_ADDR_LOOPBACK(paddr))
1101 			return (NULL);
1102 
1103 		/* Scope is embedded in */
1104 		*isinet6p = 1;
1105 
1106 		return (u_int8_t *)paddr;
1107 	}
1108 #endif
1109 	return (NULL);
1110 }
1111 
1112 /*
1113  * Copy NFS uid, gids from the cred structure.
1114  */
1115 void
1116 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1117 {
1118 	int i;
1119 
1120 	KASSERT(cr->cr_ngroups >= 0,
1121 	    ("newnfs_copyincred: negative cr_ngroups"));
1122 	nfscr->nfsc_uid = cr->cr_uid;
1123 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1124 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1125 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1126 }
1127 
1128 
1129 /*
1130  * Do any client specific initialization.
1131  */
1132 void
1133 nfscl_init(void)
1134 {
1135 	static int inited = 0;
1136 
1137 	if (inited)
1138 		return;
1139 	inited = 1;
1140 	nfscl_inited = 1;
1141 	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1142 }
1143 
1144 /*
1145  * Check each of the attributes to be set, to ensure they aren't already
1146  * the correct value. Disable setting ones already correct.
1147  */
1148 int
1149 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1150 {
1151 
1152 	if (vap->va_mode != (mode_t)VNOVAL) {
1153 		if (vap->va_mode == nvap->na_mode)
1154 			vap->va_mode = (mode_t)VNOVAL;
1155 	}
1156 	if (vap->va_uid != (uid_t)VNOVAL) {
1157 		if (vap->va_uid == nvap->na_uid)
1158 			vap->va_uid = (uid_t)VNOVAL;
1159 	}
1160 	if (vap->va_gid != (gid_t)VNOVAL) {
1161 		if (vap->va_gid == nvap->na_gid)
1162 			vap->va_gid = (gid_t)VNOVAL;
1163 	}
1164 	if (vap->va_size != VNOVAL) {
1165 		if (vap->va_size == nvap->na_size)
1166 			vap->va_size = VNOVAL;
1167 	}
1168 
1169 	/*
1170 	 * We are normally called with only a partially initialized
1171 	 * VAP.  Since the NFSv3 spec says that server may use the
1172 	 * file attributes to store the verifier, the spec requires
1173 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1174 	 * in atime, but we can't really assume that all servers will
1175 	 * so we ensure that our SETATTR sets both atime and mtime.
1176 	 * Set the VA_UTIMES_NULL flag for this case, so that
1177 	 * the server's time will be used.  This is needed to
1178 	 * work around a bug in some Solaris servers, where
1179 	 * setting the time TOCLIENT causes the Setattr RPC
1180 	 * to return NFS_OK, but not set va_mode.
1181 	 */
1182 	if (vap->va_mtime.tv_sec == VNOVAL) {
1183 		vfs_timestamp(&vap->va_mtime);
1184 		vap->va_vaflags |= VA_UTIMES_NULL;
1185 	}
1186 	if (vap->va_atime.tv_sec == VNOVAL)
1187 		vap->va_atime = vap->va_mtime;
1188 	return (1);
1189 }
1190 
1191 /*
1192  * Map nfsv4 errors to errno.h errors.
1193  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1194  * error should only be returned for the Open, Create and Setattr Ops.
1195  * As such, most calls can just pass in 0 for those arguments.
1196  */
1197 APPLESTATIC int
1198 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1199 {
1200 	struct proc *p;
1201 
1202 	if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1203 		return (error);
1204 	if (td != NULL)
1205 		p = td->td_proc;
1206 	else
1207 		p = NULL;
1208 	switch (error) {
1209 	case NFSERR_BADOWNER:
1210 		tprintf(p, LOG_INFO,
1211 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1212 		    uid, gid);
1213 		return (EPERM);
1214 	case NFSERR_BADNAME:
1215 	case NFSERR_BADCHAR:
1216 		printf("nfsv4 char/name not handled by server\n");
1217 		return (ENOENT);
1218 	case NFSERR_STALECLIENTID:
1219 	case NFSERR_STALESTATEID:
1220 	case NFSERR_EXPIRED:
1221 	case NFSERR_BADSTATEID:
1222 	case NFSERR_BADSESSION:
1223 		printf("nfsv4 recover err returned %d\n", error);
1224 		return (EIO);
1225 	case NFSERR_BADHANDLE:
1226 	case NFSERR_SERVERFAULT:
1227 	case NFSERR_BADTYPE:
1228 	case NFSERR_FHEXPIRED:
1229 	case NFSERR_RESOURCE:
1230 	case NFSERR_MOVED:
1231 	case NFSERR_NOFILEHANDLE:
1232 	case NFSERR_MINORVERMISMATCH:
1233 	case NFSERR_OLDSTATEID:
1234 	case NFSERR_BADSEQID:
1235 	case NFSERR_LEASEMOVED:
1236 	case NFSERR_RECLAIMBAD:
1237 	case NFSERR_BADXDR:
1238 	case NFSERR_OPILLEGAL:
1239 		printf("nfsv4 client/server protocol prob err=%d\n",
1240 		    error);
1241 		return (EIO);
1242 	default:
1243 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1244 		return (EIO);
1245 	};
1246 }
1247 
1248 /*
1249  * Check to see if the process for this owner exists. Return 1 if it doesn't
1250  * and 0 otherwise.
1251  */
1252 int
1253 nfscl_procdoesntexist(u_int8_t *own)
1254 {
1255 	union {
1256 		u_int32_t	lval;
1257 		u_int8_t	cval[4];
1258 	} tl;
1259 	struct proc *p;
1260 	pid_t pid;
1261 	int i, ret = 0;
1262 
1263 	/* For the single open_owner of all 0 bytes, just return 0. */
1264 	for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1265 		if (own[i] != 0)
1266 			break;
1267 	if (i == NFSV4CL_LOCKNAMELEN)
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 	pid = tl.lval;
1275 	p = pfind_locked(pid);
1276 	if (p == NULL)
1277 		return (1);
1278 	if (p->p_stats == NULL) {
1279 		PROC_UNLOCK(p);
1280 		return (0);
1281 	}
1282 	tl.cval[0] = *own++;
1283 	tl.cval[1] = *own++;
1284 	tl.cval[2] = *own++;
1285 	tl.cval[3] = *own++;
1286 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1287 		ret = 1;
1288 	} else {
1289 		tl.cval[0] = *own++;
1290 		tl.cval[1] = *own++;
1291 		tl.cval[2] = *own++;
1292 		tl.cval[3] = *own;
1293 		if (tl.lval != p->p_stats->p_start.tv_usec)
1294 			ret = 1;
1295 	}
1296 	PROC_UNLOCK(p);
1297 	return (ret);
1298 }
1299 
1300 /*
1301  * - nfs pseudo system call for the client
1302  */
1303 /*
1304  * MPSAFE
1305  */
1306 static int
1307 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1308 {
1309 	struct file *fp;
1310 	struct nfscbd_args nfscbdarg;
1311 	struct nfsd_nfscbd_args nfscbdarg2;
1312 	struct nameidata nd;
1313 	struct nfscl_dumpmntopts dumpmntopts;
1314 	cap_rights_t rights;
1315 	char *buf;
1316 	int error;
1317 	struct mount *mp;
1318 	struct nfsmount *nmp;
1319 
1320 	if (uap->flag & NFSSVC_CBADDSOCK) {
1321 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1322 		if (error)
1323 			return (error);
1324 		/*
1325 		 * Since we don't know what rights might be required,
1326 		 * pretend that we need them all. It is better to be too
1327 		 * careful than too reckless.
1328 		 */
1329 		error = fget(td, nfscbdarg.sock,
1330 		    cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1331 		if (error)
1332 			return (error);
1333 		if (fp->f_type != DTYPE_SOCKET) {
1334 			fdrop(fp, td);
1335 			return (EPERM);
1336 		}
1337 		error = nfscbd_addsock(fp);
1338 		fdrop(fp, td);
1339 		if (!error && nfscl_enablecallb == 0) {
1340 			nfsv4_cbport = nfscbdarg.port;
1341 			nfscl_enablecallb = 1;
1342 		}
1343 	} else if (uap->flag & NFSSVC_NFSCBD) {
1344 		if (uap->argp == NULL)
1345 			return (EINVAL);
1346 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1347 		    sizeof(nfscbdarg2));
1348 		if (error)
1349 			return (error);
1350 		error = nfscbd_nfsd(td, &nfscbdarg2);
1351 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1352 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1353 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1354 		    dumpmntopts.ndmnt_blen > 1024))
1355 			error = EINVAL;
1356 		if (error == 0)
1357 			error = nfsrv_lookupfilename(&nd,
1358 			    dumpmntopts.ndmnt_fname, td);
1359 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1360 		    "nfs") != 0) {
1361 			vput(nd.ni_vp);
1362 			error = EINVAL;
1363 		}
1364 		if (error == 0) {
1365 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1366 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1367 			    dumpmntopts.ndmnt_blen);
1368 			vput(nd.ni_vp);
1369 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1370 			    dumpmntopts.ndmnt_blen);
1371 			free(buf, M_TEMP);
1372 		}
1373 	} else if (uap->flag & NFSSVC_FORCEDISM) {
1374 		buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
1375 		error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
1376 		if (error == 0) {
1377 			nmp = NULL;
1378 			mtx_lock(&mountlist_mtx);
1379 			TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1380 				if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
1381 				    0 && strcmp(mp->mnt_stat.f_fstypename,
1382 				    "nfs") == 0 && mp->mnt_data != NULL) {
1383 					nmp = VFSTONFS(mp);
1384 					mtx_lock(&nmp->nm_mtx);
1385 					if ((nmp->nm_privflag &
1386 					    NFSMNTP_FORCEDISM) == 0) {
1387 						nmp->nm_privflag |=
1388 						   (NFSMNTP_FORCEDISM |
1389 						    NFSMNTP_CANCELRPCS);
1390 						mtx_unlock(&nmp->nm_mtx);
1391 					} else {
1392 						nmp = NULL;
1393 						mtx_unlock(&nmp->nm_mtx);
1394 					}
1395 					break;
1396 				}
1397 			}
1398 			mtx_unlock(&mountlist_mtx);
1399 
1400 			if (nmp != NULL) {
1401 				/*
1402 				 * Call newnfs_nmcancelreqs() to cause
1403 				 * any RPCs in progress on the mount point to
1404 				 * fail.
1405 				 * This will cause any process waiting for an
1406 				 * RPC to complete while holding a vnode lock
1407 				 * on the mounted-on vnode (such as "df" or
1408 				 * a non-forced "umount") to fail.
1409 				 * This will unlock the mounted-on vnode so
1410 				 * a forced dismount can succeed.
1411 				 * Then clear NFSMNTP_CANCELRPCS and wakeup(),
1412 				 * so that nfs_unmount() can complete.
1413 				 */
1414 				newnfs_nmcancelreqs(nmp);
1415 				mtx_lock(&nmp->nm_mtx);
1416 				nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
1417 				wakeup(nmp);
1418 				mtx_unlock(&nmp->nm_mtx);
1419 			} else
1420 				error = EINVAL;
1421 		}
1422 		free(buf, M_TEMP);
1423 	} else {
1424 		error = EINVAL;
1425 	}
1426 	return (error);
1427 }
1428 
1429 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1430 
1431 /*
1432  * Called once to initialize data structures...
1433  */
1434 static int
1435 nfscl_modevent(module_t mod, int type, void *data)
1436 {
1437 	int error = 0;
1438 	static int loaded = 0;
1439 
1440 	switch (type) {
1441 	case MOD_LOAD:
1442 		if (loaded)
1443 			return (0);
1444 		newnfs_portinit();
1445 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1446 		nfscl_init();
1447 		NFSD_LOCK();
1448 		nfsrvd_cbinit(0);
1449 		NFSD_UNLOCK();
1450 		ncl_call_invalcaches = ncl_invalcaches;
1451 		nfsd_call_nfscl = nfssvc_nfscl;
1452 		loaded = 1;
1453 		break;
1454 
1455 	case MOD_UNLOAD:
1456 		if (nfs_numnfscbd != 0) {
1457 			error = EBUSY;
1458 			break;
1459 		}
1460 
1461 		/*
1462 		 * XXX: Unloading of nfscl module is unsupported.
1463 		 */
1464 #if 0
1465 		ncl_call_invalcaches = NULL;
1466 		nfsd_call_nfscl = NULL;
1467 		/* and get rid of the mutexes */
1468 		mtx_destroy(&ncl_iod_mutex);
1469 		loaded = 0;
1470 		break;
1471 #else
1472 		/* FALLTHROUGH */
1473 #endif
1474 	default:
1475 		error = EOPNOTSUPP;
1476 		break;
1477 	}
1478 	return error;
1479 }
1480 static moduledata_t nfscl_mod = {
1481 	"nfscl",
1482 	nfscl_modevent,
1483 	NULL,
1484 };
1485 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1486 
1487 /* So that loader and kldload(2) can find us, wherever we are.. */
1488 MODULE_VERSION(nfscl, 1);
1489 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1490 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1491 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1492 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1493 
1494