xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision d34048812292b714a0bf99967270d18fe3097c62)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include "opt_inet.h"
40 #include "opt_inet6.h"
41 
42 #include <sys/capsicum.h>
43 
44 /*
45  * generally, I don't like #includes inside .h files, but it seems to
46  * be the easiest way to handle the port.
47  */
48 #include <sys/fail.h>
49 #include <sys/hash.h>
50 #include <sys/sysctl.h>
51 #include <fs/nfs/nfsport.h>
52 #include <netinet/in_fib.h>
53 #include <netinet/if_ether.h>
54 #include <netinet6/ip6_var.h>
55 #include <net/if_types.h>
56 
57 #include <fs/nfsclient/nfs_kdtrace.h>
58 
59 #ifdef KDTRACE_HOOKS
60 dtrace_nfsclient_attrcache_flush_probe_func_t
61 		dtrace_nfscl_attrcache_flush_done_probe;
62 uint32_t	nfscl_attrcache_flush_done_id;
63 
64 dtrace_nfsclient_attrcache_get_hit_probe_func_t
65 		dtrace_nfscl_attrcache_get_hit_probe;
66 uint32_t	nfscl_attrcache_get_hit_id;
67 
68 dtrace_nfsclient_attrcache_get_miss_probe_func_t
69 		dtrace_nfscl_attrcache_get_miss_probe;
70 uint32_t	nfscl_attrcache_get_miss_id;
71 
72 dtrace_nfsclient_attrcache_load_probe_func_t
73 		dtrace_nfscl_attrcache_load_done_probe;
74 uint32_t	nfscl_attrcache_load_done_id;
75 #endif /* !KDTRACE_HOOKS */
76 
77 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
78 extern struct vop_vector newnfs_vnodeops;
79 extern struct vop_vector newnfs_fifoops;
80 extern uma_zone_t newnfsnode_zone;
81 extern struct buf_ops buf_ops_newnfs;
82 extern int ncl_pbuf_freecnt;
83 extern short nfsv4_cbport;
84 extern int nfscl_enablecallb;
85 extern int nfs_numnfscbd;
86 extern int nfscl_inited;
87 struct mtx ncl_iod_mutex;
88 NFSDLOCKMUTEX;
89 extern struct mtx nfsrv_dslock_mtx;
90 
91 extern void (*ncl_call_invalcaches)(struct vnode *);
92 
93 SYSCTL_DECL(_vfs_nfs);
94 static int ncl_fileid_maxwarnings = 10;
95 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
96     &ncl_fileid_maxwarnings, 0,
97     "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
98 static volatile int ncl_fileid_nwarnings;
99 
100 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
101     struct nfsvattr *);
102 
103 /*
104  * Comparison function for vfs_hash functions.
105  */
106 int
107 newnfs_vncmpf(struct vnode *vp, void *arg)
108 {
109 	struct nfsfh *nfhp = (struct nfsfh *)arg;
110 	struct nfsnode *np = VTONFS(vp);
111 
112 	if (np->n_fhp->nfh_len != nfhp->nfh_len ||
113 	    NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
114 		return (1);
115 	return (0);
116 }
117 
118 /*
119  * Look up a vnode/nfsnode by file handle.
120  * Callers must check for mount points!!
121  * In all cases, a pointer to a
122  * nfsnode structure is returned.
123  * This variant takes a "struct nfsfh *" as second argument and uses
124  * that structure up, either by hanging off the nfsnode or FREEing it.
125  */
126 int
127 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
128     struct componentname *cnp, struct thread *td, struct nfsnode **npp,
129     void *stuff, int lkflags)
130 {
131 	struct nfsnode *np, *dnp;
132 	struct vnode *vp, *nvp;
133 	struct nfsv4node *newd, *oldd;
134 	int error;
135 	u_int hash;
136 	struct nfsmount *nmp;
137 
138 	nmp = VFSTONFS(mntp);
139 	dnp = VTONFS(dvp);
140 	*npp = NULL;
141 
142 	hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
143 
144 	error = vfs_hash_get(mntp, hash, lkflags,
145 	    td, &nvp, newnfs_vncmpf, nfhp);
146 	if (error == 0 && nvp != NULL) {
147 		/*
148 		 * I believe there is a slight chance that vgonel() could
149 		 * get called on this vnode between when NFSVOPLOCK() drops
150 		 * the VI_LOCK() and vget() acquires it again, so that it
151 		 * hasn't yet had v_usecount incremented. If this were to
152 		 * happen, the VI_DOOMED flag would be set, so check for
153 		 * that here. Since we now have the v_usecount incremented,
154 		 * we should be ok until we vrele() it, if the VI_DOOMED
155 		 * flag isn't set now.
156 		 */
157 		VI_LOCK(nvp);
158 		if ((nvp->v_iflag & VI_DOOMED)) {
159 			VI_UNLOCK(nvp);
160 			vrele(nvp);
161 			error = ENOENT;
162 		} else {
163 			VI_UNLOCK(nvp);
164 		}
165 	}
166 	if (error) {
167 		free(nfhp, M_NFSFH);
168 		return (error);
169 	}
170 	if (nvp != NULL) {
171 		np = VTONFS(nvp);
172 		/*
173 		 * For NFSv4, check to see if it is the same name and
174 		 * replace the name, if it is different.
175 		 */
176 		oldd = newd = NULL;
177 		if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
178 		    nvp->v_type == VREG &&
179 		    (np->n_v4->n4_namelen != cnp->cn_namelen ||
180 		     NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
181 		     cnp->cn_namelen) ||
182 		     dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
183 		     NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
184 		     dnp->n_fhp->nfh_len))) {
185 		    newd = malloc(
186 			sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
187 			+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
188 		    NFSLOCKNODE(np);
189 		    if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
190 			&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
191 			 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
192 			 cnp->cn_namelen) ||
193 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
194 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
195 			 dnp->n_fhp->nfh_len))) {
196 			oldd = np->n_v4;
197 			np->n_v4 = newd;
198 			newd = NULL;
199 			np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
200 			np->n_v4->n4_namelen = cnp->cn_namelen;
201 			NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
202 			    dnp->n_fhp->nfh_len);
203 			NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
204 			    cnp->cn_namelen);
205 		    }
206 		    NFSUNLOCKNODE(np);
207 		}
208 		if (newd != NULL)
209 			free(newd, M_NFSV4NODE);
210 		if (oldd != NULL)
211 			free(oldd, M_NFSV4NODE);
212 		*npp = np;
213 		free(nfhp, M_NFSFH);
214 		return (0);
215 	}
216 	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
217 
218 	error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
219 	if (error) {
220 		uma_zfree(newnfsnode_zone, np);
221 		free(nfhp, M_NFSFH);
222 		return (error);
223 	}
224 	vp = nvp;
225 	KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
226 	vp->v_bufobj.bo_ops = &buf_ops_newnfs;
227 	vp->v_data = np;
228 	np->n_vnode = vp;
229 	/*
230 	 * Initialize the mutex even if the vnode is going to be a loser.
231 	 * This simplifies the logic in reclaim, which can then unconditionally
232 	 * destroy the mutex (in the case of the loser, or if hash_insert
233 	 * happened to return an error no special casing is needed).
234 	 */
235 	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
236 	lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE |
237 	    LK_CANRECURSE);
238 
239 	/*
240 	 * Are we getting the root? If so, make sure the vnode flags
241 	 * are correct
242 	 */
243 	if ((nfhp->nfh_len == nmp->nm_fhsize) &&
244 	    !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
245 		if (vp->v_type == VNON)
246 			vp->v_type = VDIR;
247 		vp->v_vflag |= VV_ROOT;
248 	}
249 
250 	np->n_fhp = nfhp;
251 	/*
252 	 * For NFSv4, we have to attach the directory file handle and
253 	 * file name, so that Open Ops can be done later.
254 	 */
255 	if (nmp->nm_flag & NFSMNT_NFSV4) {
256 		np->n_v4 = malloc(sizeof (struct nfsv4node)
257 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
258 		    M_WAITOK);
259 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
260 		np->n_v4->n4_namelen = cnp->cn_namelen;
261 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
262 		    dnp->n_fhp->nfh_len);
263 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
264 		    cnp->cn_namelen);
265 	} else {
266 		np->n_v4 = NULL;
267 	}
268 
269 	/*
270 	 * NFS supports recursive and shared locking.
271 	 */
272 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
273 	VN_LOCK_AREC(vp);
274 	VN_LOCK_ASHARE(vp);
275 	error = insmntque(vp, mntp);
276 	if (error != 0) {
277 		*npp = NULL;
278 		mtx_destroy(&np->n_mtx);
279 		lockdestroy(&np->n_excl);
280 		free(nfhp, M_NFSFH);
281 		if (np->n_v4 != NULL)
282 			free(np->n_v4, M_NFSV4NODE);
283 		uma_zfree(newnfsnode_zone, np);
284 		return (error);
285 	}
286 	error = vfs_hash_insert(vp, hash, lkflags,
287 	    td, &nvp, newnfs_vncmpf, nfhp);
288 	if (error)
289 		return (error);
290 	if (nvp != NULL) {
291 		*npp = VTONFS(nvp);
292 		/* vfs_hash_insert() vput()'s the losing vnode */
293 		return (0);
294 	}
295 	*npp = np;
296 
297 	return (0);
298 }
299 
300 /*
301  * Another variant of nfs_nget(). This one is only used by reopen. It
302  * takes almost the same args as nfs_nget(), but only succeeds if an entry
303  * exists in the cache. (Since files should already be "open" with a
304  * vnode ref cnt on the node when reopen calls this, it should always
305  * succeed.)
306  * Also, don't get a vnode lock, since it may already be locked by some
307  * other process that is handling it. This is ok, since all other threads
308  * on the client are blocked by the nfsc_lock being exclusively held by the
309  * caller of this function.
310  */
311 int
312 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
313     struct thread *td, struct nfsnode **npp)
314 {
315 	struct vnode *nvp;
316 	u_int hash;
317 	struct nfsfh *nfhp;
318 	int error;
319 
320 	*npp = NULL;
321 	/* For forced dismounts, just return error. */
322 	if (NFSCL_FORCEDISM(mntp))
323 		return (EINTR);
324 	nfhp = malloc(sizeof (struct nfsfh) + fhsize,
325 	    M_NFSFH, M_WAITOK);
326 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
327 	nfhp->nfh_len = fhsize;
328 
329 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
330 
331 	/*
332 	 * First, try to get the vnode locked, but don't block for the lock.
333 	 */
334 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
335 	    newnfs_vncmpf, nfhp);
336 	if (error == 0 && nvp != NULL) {
337 		NFSVOPUNLOCK(nvp, 0);
338 	} else if (error == EBUSY) {
339 		/*
340 		 * It is safe so long as a vflush() with
341 		 * FORCECLOSE has not been done. Since the Renew thread is
342 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
343 		 * a vflush() with FORCECLOSE, we should be ok here.
344 		 */
345 		if (NFSCL_FORCEDISM(mntp))
346 			error = EINTR;
347 		else {
348 			vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
349 			if (nvp == NULL) {
350 				error = ENOENT;
351 			} else if ((nvp->v_iflag & VI_DOOMED) != 0) {
352 				error = ENOENT;
353 				vrele(nvp);
354 			} else {
355 				error = 0;
356 			}
357 		}
358 	}
359 	free(nfhp, M_NFSFH);
360 	if (error)
361 		return (error);
362 	if (nvp != NULL) {
363 		*npp = VTONFS(nvp);
364 		return (0);
365 	}
366 	return (EINVAL);
367 }
368 
369 static void
370 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
371     struct nfsvattr *newnap)
372 {
373 	int off;
374 
375 	if (ncl_fileid_maxwarnings >= 0 &&
376 	    ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
377 		return;
378 	off = 0;
379 	if (ncl_fileid_maxwarnings >= 0) {
380 		if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
381 			off = 1;
382 	}
383 
384 	printf("newnfs: server '%s' error: fileid changed. "
385 	    "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
386 	    "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
387 	    nmp->nm_com.nmcom_hostname,
388 	    (uintmax_t)nmp->nm_fsid[0],
389 	    (uintmax_t)nmp->nm_fsid[1],
390 	    (uintmax_t)oldnap->na_fileid,
391 	    (uintmax_t)newnap->na_fileid);
392 
393 	if (off)
394 		printf("newnfs: Logged %d times about fileid corruption; "
395 		    "going quiet to avoid spamming logs excessively. (Limit "
396 		    "is: %d).\n", ncl_fileid_nwarnings,
397 		    ncl_fileid_maxwarnings);
398 }
399 
400 /*
401  * Load the attribute cache (that lives in the nfsnode entry) with
402  * the attributes of the second argument and
403  * Iff vaper not NULL
404  *    copy the attributes to *vaper
405  * Similar to nfs_loadattrcache(), except the attributes are passed in
406  * instead of being parsed out of the mbuf list.
407  */
408 int
409 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
410     void *stuff, int writeattr, int dontshrink)
411 {
412 	struct vnode *vp = *vpp;
413 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
414 	struct nfsnode *np;
415 	struct nfsmount *nmp;
416 	struct timespec mtime_save;
417 	u_quad_t nsize;
418 	int setnsize, error, force_fid_err;
419 
420 	error = 0;
421 	setnsize = 0;
422 	nsize = 0;
423 
424 	/*
425 	 * If v_type == VNON it is a new node, so fill in the v_type,
426 	 * n_mtime fields. Check to see if it represents a special
427 	 * device, and if so, check for a possible alias. Once the
428 	 * correct vnode has been obtained, fill in the rest of the
429 	 * information.
430 	 */
431 	np = VTONFS(vp);
432 	NFSLOCKNODE(np);
433 	if (vp->v_type != nvap->va_type) {
434 		vp->v_type = nvap->va_type;
435 		if (vp->v_type == VFIFO)
436 			vp->v_op = &newnfs_fifoops;
437 		np->n_mtime = nvap->va_mtime;
438 	}
439 	nmp = VFSTONFS(vp->v_mount);
440 	vap = &np->n_vattr.na_vattr;
441 	mtime_save = vap->va_mtime;
442 	if (writeattr) {
443 		np->n_vattr.na_filerev = nap->na_filerev;
444 		np->n_vattr.na_size = nap->na_size;
445 		np->n_vattr.na_mtime = nap->na_mtime;
446 		np->n_vattr.na_ctime = nap->na_ctime;
447 		np->n_vattr.na_fsid = nap->na_fsid;
448 		np->n_vattr.na_mode = nap->na_mode;
449 	} else {
450 		force_fid_err = 0;
451 		KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
452 		    force_fid_err);
453 		/*
454 		 * BROKEN NFS SERVER OR MIDDLEWARE
455 		 *
456 		 * Certain NFS servers (certain old proprietary filers ca.
457 		 * 2006) or broken middleboxes (e.g. WAN accelerator products)
458 		 * will respond to GETATTR requests with results for a
459 		 * different fileid.
460 		 *
461 		 * The WAN accelerator we've observed not only serves stale
462 		 * cache results for a given file, it also occasionally serves
463 		 * results for wholly different files.  This causes surprising
464 		 * problems; for example the cached size attribute of a file
465 		 * may truncate down and then back up, resulting in zero
466 		 * regions in file contents read by applications.  We observed
467 		 * this reliably with Clang and .c files during parallel build.
468 		 * A pcap revealed packet fragmentation and GETATTR RPC
469 		 * responses with wholly wrong fileids.
470 		 */
471 		if ((np->n_vattr.na_fileid != 0 &&
472 		     np->n_vattr.na_fileid != nap->na_fileid) ||
473 		    force_fid_err) {
474 			nfscl_warn_fileid(nmp, &np->n_vattr, nap);
475 			error = EIDRM;
476 			goto out;
477 		}
478 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
479 		    sizeof (struct nfsvattr));
480 	}
481 
482 	/*
483 	 * For NFSv4, if the node's fsid is not equal to the mount point's
484 	 * fsid, return the low order 32bits of the node's fsid. This
485 	 * allows getcwd(3) to work. There is a chance that the fsid might
486 	 * be the same as a local fs, but since this is in an NFS mount
487 	 * point, I don't think that will cause any problems?
488 	 */
489 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
490 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
491 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
492 		/*
493 		 * va_fsid needs to be set to some value derived from
494 		 * np->n_vattr.na_filesid that is not equal
495 		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
496 		 * from the value used for the top level server volume
497 		 * in the mounted subtree.
498 		 */
499 		vn_fsid(vp, vap);
500 		if ((uint32_t)vap->va_fsid == np->n_vattr.na_filesid[0])
501 			vap->va_fsid = hash32_buf(
502 			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
503 	} else
504 		vn_fsid(vp, vap);
505 	np->n_attrstamp = time_second;
506 	if (vap->va_size != np->n_size) {
507 		if (vap->va_type == VREG) {
508 			if (dontshrink && vap->va_size < np->n_size) {
509 				/*
510 				 * We've been told not to shrink the file;
511 				 * zero np->n_attrstamp to indicate that
512 				 * the attributes are stale.
513 				 */
514 				vap->va_size = np->n_size;
515 				np->n_attrstamp = 0;
516 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
517 				vnode_pager_setsize(vp, np->n_size);
518 			} else if (np->n_flag & NMODIFIED) {
519 				/*
520 				 * We've modified the file: Use the larger
521 				 * of our size, and the server's size.
522 				 */
523 				if (vap->va_size < np->n_size) {
524 					vap->va_size = np->n_size;
525 				} else {
526 					np->n_size = vap->va_size;
527 					np->n_flag |= NSIZECHANGED;
528 				}
529 				vnode_pager_setsize(vp, np->n_size);
530 			} else if (vap->va_size < np->n_size) {
531 				/*
532 				 * When shrinking the size, the call to
533 				 * vnode_pager_setsize() cannot be done
534 				 * with the mutex held, so delay it until
535 				 * after the mtx_unlock call.
536 				 */
537 				nsize = np->n_size = vap->va_size;
538 				np->n_flag |= NSIZECHANGED;
539 				setnsize = 1;
540 			} else {
541 				np->n_size = vap->va_size;
542 				np->n_flag |= NSIZECHANGED;
543 				vnode_pager_setsize(vp, np->n_size);
544 			}
545 		} else {
546 			np->n_size = vap->va_size;
547 		}
548 	}
549 	/*
550 	 * The following checks are added to prevent a race between (say)
551 	 * a READDIR+ and a WRITE.
552 	 * READDIR+, WRITE requests sent out.
553 	 * READDIR+ resp, WRITE resp received on client.
554 	 * However, the WRITE resp was handled before the READDIR+ resp
555 	 * causing the post op attrs from the write to be loaded first
556 	 * and the attrs from the READDIR+ to be loaded later. If this
557 	 * happens, we have stale attrs loaded into the attrcache.
558 	 * We detect this by for the mtime moving back. We invalidate the
559 	 * attrcache when this happens.
560 	 */
561 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
562 		/* Size changed or mtime went backwards */
563 		np->n_attrstamp = 0;
564 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
565 	}
566 	if (vaper != NULL) {
567 		NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
568 		if (np->n_flag & NCHG) {
569 			if (np->n_flag & NACC)
570 				vaper->va_atime = np->n_atim;
571 			if (np->n_flag & NUPD)
572 				vaper->va_mtime = np->n_mtim;
573 		}
574 	}
575 
576 out:
577 #ifdef KDTRACE_HOOKS
578 	if (np->n_attrstamp != 0)
579 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
580 #endif
581 	NFSUNLOCKNODE(np);
582 	if (setnsize)
583 		vnode_pager_setsize(vp, nsize);
584 	return (error);
585 }
586 
587 /*
588  * Fill in the client id name. For these bytes:
589  * 1 - they must be unique
590  * 2 - they should be persistent across client reboots
591  * 1 is more critical than 2
592  * Use the mount point's unique id plus either the uuid or, if that
593  * isn't set, random junk.
594  */
595 void
596 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
597 {
598 	int uuidlen;
599 
600 	/*
601 	 * First, put in the 64bit mount point identifier.
602 	 */
603 	if (idlen >= sizeof (u_int64_t)) {
604 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
605 		cp += sizeof (u_int64_t);
606 		idlen -= sizeof (u_int64_t);
607 	}
608 
609 	/*
610 	 * If uuid is non-zero length, use it.
611 	 */
612 	uuidlen = strlen(uuid);
613 	if (uuidlen > 0 && idlen >= uuidlen) {
614 		NFSBCOPY(uuid, cp, uuidlen);
615 		cp += uuidlen;
616 		idlen -= uuidlen;
617 	}
618 
619 	/*
620 	 * This only normally happens if the uuid isn't set.
621 	 */
622 	while (idlen > 0) {
623 		*cp++ = (u_int8_t)(arc4random() % 256);
624 		idlen--;
625 	}
626 }
627 
628 /*
629  * Fill in a lock owner name. For now, pid + the process's creation time.
630  */
631 void
632 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
633 {
634 	union {
635 		u_int32_t	lval;
636 		u_int8_t	cval[4];
637 	} tl;
638 	struct proc *p;
639 
640 	if (id == NULL) {
641 		/* Return the single open_owner of all 0 bytes. */
642 		bzero(cp, NFSV4CL_LOCKNAMELEN);
643 		return;
644 	}
645 	if ((flags & F_POSIX) != 0) {
646 		p = (struct proc *)id;
647 		tl.lval = p->p_pid;
648 		*cp++ = tl.cval[0];
649 		*cp++ = tl.cval[1];
650 		*cp++ = tl.cval[2];
651 		*cp++ = tl.cval[3];
652 		tl.lval = p->p_stats->p_start.tv_sec;
653 		*cp++ = tl.cval[0];
654 		*cp++ = tl.cval[1];
655 		*cp++ = tl.cval[2];
656 		*cp++ = tl.cval[3];
657 		tl.lval = p->p_stats->p_start.tv_usec;
658 		*cp++ = tl.cval[0];
659 		*cp++ = tl.cval[1];
660 		*cp++ = tl.cval[2];
661 		*cp = tl.cval[3];
662 	} else if ((flags & F_FLOCK) != 0) {
663 		bcopy(&id, cp, sizeof(id));
664 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
665 	} else {
666 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
667 		bzero(cp, NFSV4CL_LOCKNAMELEN);
668 	}
669 }
670 
671 /*
672  * Find the parent process for the thread passed in as an argument.
673  * If none exists, return NULL, otherwise return a thread for the parent.
674  * (Can be any of the threads, since it is only used for td->td_proc.)
675  */
676 NFSPROC_T *
677 nfscl_getparent(struct thread *td)
678 {
679 	struct proc *p;
680 	struct thread *ptd;
681 
682 	if (td == NULL)
683 		return (NULL);
684 	p = td->td_proc;
685 	if (p->p_pid == 0)
686 		return (NULL);
687 	p = p->p_pptr;
688 	if (p == NULL)
689 		return (NULL);
690 	ptd = TAILQ_FIRST(&p->p_threads);
691 	return (ptd);
692 }
693 
694 /*
695  * Start up the renew kernel thread.
696  */
697 static void
698 start_nfscl(void *arg)
699 {
700 	struct nfsclclient *clp;
701 	struct thread *td;
702 
703 	clp = (struct nfsclclient *)arg;
704 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
705 	nfscl_renewthread(clp, td);
706 	kproc_exit(0);
707 }
708 
709 void
710 nfscl_start_renewthread(struct nfsclclient *clp)
711 {
712 
713 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
714 	    "nfscl");
715 }
716 
717 /*
718  * Handle wcc_data.
719  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
720  * as the first Op after PutFH.
721  * (For NFSv4, the postop attributes are after the Op, so they can't be
722  *  parsed here. A separate call to nfscl_postop_attr() is required.)
723  */
724 int
725 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
726     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
727 {
728 	u_int32_t *tl;
729 	struct nfsnode *np = VTONFS(vp);
730 	struct nfsvattr nfsva;
731 	int error = 0;
732 
733 	if (wccflagp != NULL)
734 		*wccflagp = 0;
735 	if (nd->nd_flag & ND_NFSV3) {
736 		*flagp = 0;
737 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
738 		if (*tl == newnfs_true) {
739 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
740 			if (wccflagp != NULL) {
741 				mtx_lock(&np->n_mtx);
742 				*wccflagp = (np->n_mtime.tv_sec ==
743 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
744 				    np->n_mtime.tv_nsec ==
745 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
746 				mtx_unlock(&np->n_mtx);
747 			}
748 		}
749 		error = nfscl_postop_attr(nd, nap, flagp, stuff);
750 		if (wccflagp != NULL && *flagp == 0)
751 			*wccflagp = 0;
752 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
753 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
754 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
755 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
756 		    NULL, NULL, NULL, NULL, NULL);
757 		if (error)
758 			return (error);
759 		/*
760 		 * Get rid of Op# and status for next op.
761 		 */
762 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
763 		if (*++tl)
764 			nd->nd_flag |= ND_NOMOREDATA;
765 		if (wccflagp != NULL &&
766 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
767 			mtx_lock(&np->n_mtx);
768 			*wccflagp = (np->n_mtime.tv_sec ==
769 			    nfsva.na_vattr.va_mtime.tv_sec &&
770 			    np->n_mtime.tv_nsec ==
771 			    nfsva.na_vattr.va_mtime.tv_sec);
772 			mtx_unlock(&np->n_mtx);
773 		}
774 	}
775 nfsmout:
776 	return (error);
777 }
778 
779 /*
780  * Get postop attributes.
781  */
782 int
783 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
784     void *stuff)
785 {
786 	u_int32_t *tl;
787 	int error = 0;
788 
789 	*retp = 0;
790 	if (nd->nd_flag & ND_NOMOREDATA)
791 		return (error);
792 	if (nd->nd_flag & ND_NFSV3) {
793 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
794 		*retp = fxdr_unsigned(int, *tl);
795 	} else if (nd->nd_flag & ND_NFSV4) {
796 		/*
797 		 * For NFSv4, the postop attr are at the end, so no point
798 		 * in looking if nd_repstat != 0.
799 		 */
800 		if (!nd->nd_repstat) {
801 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
802 			if (*(tl + 1))
803 				/* should never happen since nd_repstat != 0 */
804 				nd->nd_flag |= ND_NOMOREDATA;
805 			else
806 				*retp = 1;
807 		}
808 	} else if (!nd->nd_repstat) {
809 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
810 		*retp = 1;
811 	}
812 	if (*retp) {
813 		error = nfsm_loadattr(nd, nap);
814 		if (error)
815 			*retp = 0;
816 	}
817 nfsmout:
818 	return (error);
819 }
820 
821 /*
822  * nfscl_request() - mostly a wrapper for newnfs_request().
823  */
824 int
825 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
826     struct ucred *cred, void *stuff)
827 {
828 	int ret, vers;
829 	struct nfsmount *nmp;
830 
831 	nmp = VFSTONFS(vp->v_mount);
832 	if (nd->nd_flag & ND_NFSV4)
833 		vers = NFS_VER4;
834 	else if (nd->nd_flag & ND_NFSV3)
835 		vers = NFS_VER3;
836 	else
837 		vers = NFS_VER2;
838 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
839 		NFS_PROG, vers, NULL, 1, NULL, NULL);
840 	return (ret);
841 }
842 
843 /*
844  * fill in this bsden's variant of statfs using nfsstatfs.
845  */
846 void
847 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
848 {
849 	struct statfs *sbp = (struct statfs *)statfs;
850 
851 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
852 		sbp->f_bsize = NFS_FABLKSIZE;
853 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
854 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
855 		/*
856 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
857 		 * the value after dividing by NFS_FABLKSIZE is small
858 		 * enough that it will fit in 63bits, so it is ok to
859 		 * assign it to f_bavail without fear that it will become
860 		 * negative.
861 		 */
862 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
863 		sbp->f_files = sfp->sf_tfiles;
864 		/* Since f_ffree is int64_t, clip it to 63bits. */
865 		if (sfp->sf_ffiles > INT64_MAX)
866 			sbp->f_ffree = INT64_MAX;
867 		else
868 			sbp->f_ffree = sfp->sf_ffiles;
869 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
870 		/*
871 		 * The type casts to (int32_t) ensure that this code is
872 		 * compatible with the old NFS client, in that it will
873 		 * propagate bit31 to the high order bits. This may or may
874 		 * not be correct for NFSv2, but since it is a legacy
875 		 * environment, I'd rather retain backwards compatibility.
876 		 */
877 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
878 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
879 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
880 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
881 		sbp->f_files = 0;
882 		sbp->f_ffree = 0;
883 	}
884 }
885 
886 /*
887  * Use the fsinfo stuff to update the mount point.
888  */
889 void
890 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
891 {
892 
893 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
894 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
895 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
896 		    ~(NFS_FABLKSIZE - 1);
897 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
898 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
899 		if (nmp->nm_wsize == 0)
900 			nmp->nm_wsize = fsp->fs_wtmax;
901 	}
902 	if (nmp->nm_wsize < NFS_FABLKSIZE)
903 		nmp->nm_wsize = NFS_FABLKSIZE;
904 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
905 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
906 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
907 		    ~(NFS_FABLKSIZE - 1);
908 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
909 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
910 		if (nmp->nm_rsize == 0)
911 			nmp->nm_rsize = fsp->fs_rtmax;
912 	}
913 	if (nmp->nm_rsize < NFS_FABLKSIZE)
914 		nmp->nm_rsize = NFS_FABLKSIZE;
915 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
916 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
917 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
918 		    ~(NFS_DIRBLKSIZ - 1);
919 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
920 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
921 		if (nmp->nm_readdirsize == 0)
922 			nmp->nm_readdirsize = fsp->fs_rtmax;
923 	}
924 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
925 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
926 	if (fsp->fs_maxfilesize > 0 &&
927 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
928 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
929 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
930 	nmp->nm_state |= NFSSTA_GOTFSINFO;
931 }
932 
933 /*
934  * Lookups source address which should be used to communicate with
935  * @nmp and stores it inside @pdst.
936  *
937  * Returns 0 on success.
938  */
939 u_int8_t *
940 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
941 {
942 #if defined(INET6) || defined(INET)
943 	int error, fibnum;
944 
945 	fibnum = curthread->td_proc->p_fibnum;
946 #endif
947 #ifdef INET
948 	if (nmp->nm_nam->sa_family == AF_INET) {
949 		struct sockaddr_in *sin;
950 		struct nhop4_extended nh_ext;
951 
952 		sin = (struct sockaddr_in *)nmp->nm_nam;
953 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
954 		error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
955 		    &nh_ext);
956 		CURVNET_RESTORE();
957 		if (error != 0)
958 			return (NULL);
959 
960 		if ((ntohl(nh_ext.nh_src.s_addr) >> IN_CLASSA_NSHIFT) ==
961 		    IN_LOOPBACKNET) {
962 			/* Ignore loopback addresses */
963 			return (NULL);
964 		}
965 
966 		*isinet6p = 0;
967 		*((struct in_addr *)paddr) = nh_ext.nh_src;
968 
969 		return (u_int8_t *)paddr;
970 	}
971 #endif
972 #ifdef INET6
973 	if (nmp->nm_nam->sa_family == AF_INET6) {
974 		struct sockaddr_in6 *sin6;
975 
976 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
977 
978 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
979 		error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
980 		    sin6->sin6_scope_id, NULL, paddr, NULL);
981 		CURVNET_RESTORE();
982 		if (error != 0)
983 			return (NULL);
984 
985 		if (IN6_IS_ADDR_LOOPBACK(paddr))
986 			return (NULL);
987 
988 		/* Scope is embedded in */
989 		*isinet6p = 1;
990 
991 		return (u_int8_t *)paddr;
992 	}
993 #endif
994 	return (NULL);
995 }
996 
997 /*
998  * Copy NFS uid, gids from the cred structure.
999  */
1000 void
1001 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1002 {
1003 	int i;
1004 
1005 	KASSERT(cr->cr_ngroups >= 0,
1006 	    ("newnfs_copyincred: negative cr_ngroups"));
1007 	nfscr->nfsc_uid = cr->cr_uid;
1008 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1009 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1010 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1011 }
1012 
1013 
1014 /*
1015  * Do any client specific initialization.
1016  */
1017 void
1018 nfscl_init(void)
1019 {
1020 	static int inited = 0;
1021 
1022 	if (inited)
1023 		return;
1024 	inited = 1;
1025 	nfscl_inited = 1;
1026 	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1027 }
1028 
1029 /*
1030  * Check each of the attributes to be set, to ensure they aren't already
1031  * the correct value. Disable setting ones already correct.
1032  */
1033 int
1034 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1035 {
1036 
1037 	if (vap->va_mode != (mode_t)VNOVAL) {
1038 		if (vap->va_mode == nvap->na_mode)
1039 			vap->va_mode = (mode_t)VNOVAL;
1040 	}
1041 	if (vap->va_uid != (uid_t)VNOVAL) {
1042 		if (vap->va_uid == nvap->na_uid)
1043 			vap->va_uid = (uid_t)VNOVAL;
1044 	}
1045 	if (vap->va_gid != (gid_t)VNOVAL) {
1046 		if (vap->va_gid == nvap->na_gid)
1047 			vap->va_gid = (gid_t)VNOVAL;
1048 	}
1049 	if (vap->va_size != VNOVAL) {
1050 		if (vap->va_size == nvap->na_size)
1051 			vap->va_size = VNOVAL;
1052 	}
1053 
1054 	/*
1055 	 * We are normally called with only a partially initialized
1056 	 * VAP.  Since the NFSv3 spec says that server may use the
1057 	 * file attributes to store the verifier, the spec requires
1058 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1059 	 * in atime, but we can't really assume that all servers will
1060 	 * so we ensure that our SETATTR sets both atime and mtime.
1061 	 * Set the VA_UTIMES_NULL flag for this case, so that
1062 	 * the server's time will be used.  This is needed to
1063 	 * work around a bug in some Solaris servers, where
1064 	 * setting the time TOCLIENT causes the Setattr RPC
1065 	 * to return NFS_OK, but not set va_mode.
1066 	 */
1067 	if (vap->va_mtime.tv_sec == VNOVAL) {
1068 		vfs_timestamp(&vap->va_mtime);
1069 		vap->va_vaflags |= VA_UTIMES_NULL;
1070 	}
1071 	if (vap->va_atime.tv_sec == VNOVAL)
1072 		vap->va_atime = vap->va_mtime;
1073 	return (1);
1074 }
1075 
1076 /*
1077  * Map nfsv4 errors to errno.h errors.
1078  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1079  * error should only be returned for the Open, Create and Setattr Ops.
1080  * As such, most calls can just pass in 0 for those arguments.
1081  */
1082 APPLESTATIC int
1083 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1084 {
1085 	struct proc *p;
1086 
1087 	if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1088 		return (error);
1089 	if (td != NULL)
1090 		p = td->td_proc;
1091 	else
1092 		p = NULL;
1093 	switch (error) {
1094 	case NFSERR_BADOWNER:
1095 		tprintf(p, LOG_INFO,
1096 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1097 		    uid, gid);
1098 		return (EPERM);
1099 	case NFSERR_BADNAME:
1100 	case NFSERR_BADCHAR:
1101 		printf("nfsv4 char/name not handled by server\n");
1102 		return (ENOENT);
1103 	case NFSERR_STALECLIENTID:
1104 	case NFSERR_STALESTATEID:
1105 	case NFSERR_EXPIRED:
1106 	case NFSERR_BADSTATEID:
1107 	case NFSERR_BADSESSION:
1108 		printf("nfsv4 recover err returned %d\n", error);
1109 		return (EIO);
1110 	case NFSERR_BADHANDLE:
1111 	case NFSERR_SERVERFAULT:
1112 	case NFSERR_BADTYPE:
1113 	case NFSERR_FHEXPIRED:
1114 	case NFSERR_RESOURCE:
1115 	case NFSERR_MOVED:
1116 	case NFSERR_NOFILEHANDLE:
1117 	case NFSERR_MINORVERMISMATCH:
1118 	case NFSERR_OLDSTATEID:
1119 	case NFSERR_BADSEQID:
1120 	case NFSERR_LEASEMOVED:
1121 	case NFSERR_RECLAIMBAD:
1122 	case NFSERR_BADXDR:
1123 	case NFSERR_OPILLEGAL:
1124 		printf("nfsv4 client/server protocol prob err=%d\n",
1125 		    error);
1126 		return (EIO);
1127 	default:
1128 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1129 		return (EIO);
1130 	};
1131 }
1132 
1133 /*
1134  * Check to see if the process for this owner exists. Return 1 if it doesn't
1135  * and 0 otherwise.
1136  */
1137 int
1138 nfscl_procdoesntexist(u_int8_t *own)
1139 {
1140 	union {
1141 		u_int32_t	lval;
1142 		u_int8_t	cval[4];
1143 	} tl;
1144 	struct proc *p;
1145 	pid_t pid;
1146 	int i, ret = 0;
1147 
1148 	/* For the single open_owner of all 0 bytes, just return 0. */
1149 	for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1150 		if (own[i] != 0)
1151 			break;
1152 	if (i == NFSV4CL_LOCKNAMELEN)
1153 		return (0);
1154 
1155 	tl.cval[0] = *own++;
1156 	tl.cval[1] = *own++;
1157 	tl.cval[2] = *own++;
1158 	tl.cval[3] = *own++;
1159 	pid = tl.lval;
1160 	p = pfind(pid);
1161 	if (p == NULL)
1162 		return (1);
1163 	if (p->p_stats == NULL) {
1164 		PROC_UNLOCK(p);
1165 		return (0);
1166 	}
1167 	tl.cval[0] = *own++;
1168 	tl.cval[1] = *own++;
1169 	tl.cval[2] = *own++;
1170 	tl.cval[3] = *own++;
1171 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1172 		ret = 1;
1173 	} else {
1174 		tl.cval[0] = *own++;
1175 		tl.cval[1] = *own++;
1176 		tl.cval[2] = *own++;
1177 		tl.cval[3] = *own;
1178 		if (tl.lval != p->p_stats->p_start.tv_usec)
1179 			ret = 1;
1180 	}
1181 	PROC_UNLOCK(p);
1182 	return (ret);
1183 }
1184 
1185 /*
1186  * - nfs pseudo system call for the client
1187  */
1188 /*
1189  * MPSAFE
1190  */
1191 static int
1192 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1193 {
1194 	struct file *fp;
1195 	struct nfscbd_args nfscbdarg;
1196 	struct nfsd_nfscbd_args nfscbdarg2;
1197 	struct nameidata nd;
1198 	struct nfscl_dumpmntopts dumpmntopts;
1199 	cap_rights_t rights;
1200 	char *buf;
1201 	int error;
1202 	struct mount *mp;
1203 	struct nfsmount *nmp;
1204 
1205 	if (uap->flag & NFSSVC_CBADDSOCK) {
1206 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1207 		if (error)
1208 			return (error);
1209 		/*
1210 		 * Since we don't know what rights might be required,
1211 		 * pretend that we need them all. It is better to be too
1212 		 * careful than too reckless.
1213 		 */
1214 		error = fget(td, nfscbdarg.sock,
1215 		    cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1216 		if (error)
1217 			return (error);
1218 		if (fp->f_type != DTYPE_SOCKET) {
1219 			fdrop(fp, td);
1220 			return (EPERM);
1221 		}
1222 		error = nfscbd_addsock(fp);
1223 		fdrop(fp, td);
1224 		if (!error && nfscl_enablecallb == 0) {
1225 			nfsv4_cbport = nfscbdarg.port;
1226 			nfscl_enablecallb = 1;
1227 		}
1228 	} else if (uap->flag & NFSSVC_NFSCBD) {
1229 		if (uap->argp == NULL)
1230 			return (EINVAL);
1231 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1232 		    sizeof(nfscbdarg2));
1233 		if (error)
1234 			return (error);
1235 		error = nfscbd_nfsd(td, &nfscbdarg2);
1236 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1237 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1238 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1239 		    dumpmntopts.ndmnt_blen > 1024))
1240 			error = EINVAL;
1241 		if (error == 0)
1242 			error = nfsrv_lookupfilename(&nd,
1243 			    dumpmntopts.ndmnt_fname, td);
1244 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1245 		    "nfs") != 0) {
1246 			vput(nd.ni_vp);
1247 			error = EINVAL;
1248 		}
1249 		if (error == 0) {
1250 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1251 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1252 			    dumpmntopts.ndmnt_blen);
1253 			vput(nd.ni_vp);
1254 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1255 			    dumpmntopts.ndmnt_blen);
1256 			free(buf, M_TEMP);
1257 		}
1258 	} else if (uap->flag & NFSSVC_FORCEDISM) {
1259 		buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
1260 		error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
1261 		if (error == 0) {
1262 			nmp = NULL;
1263 			mtx_lock(&mountlist_mtx);
1264 			TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1265 				if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
1266 				    0 && strcmp(mp->mnt_stat.f_fstypename,
1267 				    "nfs") == 0 && mp->mnt_data != NULL) {
1268 					nmp = VFSTONFS(mp);
1269 					NFSDDSLOCK();
1270 					if (nfsv4_findmirror(nmp) != NULL) {
1271 						NFSDDSUNLOCK();
1272 						error = ENXIO;
1273 						nmp = NULL;
1274 						break;
1275 					}
1276 					mtx_lock(&nmp->nm_mtx);
1277 					if ((nmp->nm_privflag &
1278 					    NFSMNTP_FORCEDISM) == 0) {
1279 						nmp->nm_privflag |=
1280 						   (NFSMNTP_FORCEDISM |
1281 						    NFSMNTP_CANCELRPCS);
1282 						mtx_unlock(&nmp->nm_mtx);
1283 					} else {
1284 						mtx_unlock(&nmp->nm_mtx);
1285 						nmp = NULL;
1286 					}
1287 					NFSDDSUNLOCK();
1288 					break;
1289 				}
1290 			}
1291 			mtx_unlock(&mountlist_mtx);
1292 
1293 			if (nmp != NULL) {
1294 				/*
1295 				 * Call newnfs_nmcancelreqs() to cause
1296 				 * any RPCs in progress on the mount point to
1297 				 * fail.
1298 				 * This will cause any process waiting for an
1299 				 * RPC to complete while holding a vnode lock
1300 				 * on the mounted-on vnode (such as "df" or
1301 				 * a non-forced "umount") to fail.
1302 				 * This will unlock the mounted-on vnode so
1303 				 * a forced dismount can succeed.
1304 				 * Then clear NFSMNTP_CANCELRPCS and wakeup(),
1305 				 * so that nfs_unmount() can complete.
1306 				 */
1307 				newnfs_nmcancelreqs(nmp);
1308 				mtx_lock(&nmp->nm_mtx);
1309 				nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
1310 				wakeup(nmp);
1311 				mtx_unlock(&nmp->nm_mtx);
1312 			} else if (error == 0)
1313 				error = EINVAL;
1314 		}
1315 		free(buf, M_TEMP);
1316 	} else {
1317 		error = EINVAL;
1318 	}
1319 	return (error);
1320 }
1321 
1322 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1323 
1324 /*
1325  * Called once to initialize data structures...
1326  */
1327 static int
1328 nfscl_modevent(module_t mod, int type, void *data)
1329 {
1330 	int error = 0;
1331 	static int loaded = 0;
1332 
1333 	switch (type) {
1334 	case MOD_LOAD:
1335 		if (loaded)
1336 			return (0);
1337 		newnfs_portinit();
1338 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1339 		nfscl_init();
1340 		NFSD_LOCK();
1341 		nfsrvd_cbinit(0);
1342 		NFSD_UNLOCK();
1343 		ncl_call_invalcaches = ncl_invalcaches;
1344 		nfsd_call_nfscl = nfssvc_nfscl;
1345 		loaded = 1;
1346 		break;
1347 
1348 	case MOD_UNLOAD:
1349 		if (nfs_numnfscbd != 0) {
1350 			error = EBUSY;
1351 			break;
1352 		}
1353 
1354 		/*
1355 		 * XXX: Unloading of nfscl module is unsupported.
1356 		 */
1357 #if 0
1358 		ncl_call_invalcaches = NULL;
1359 		nfsd_call_nfscl = NULL;
1360 		/* and get rid of the mutexes */
1361 		mtx_destroy(&ncl_iod_mutex);
1362 		loaded = 0;
1363 		break;
1364 #else
1365 		/* FALLTHROUGH */
1366 #endif
1367 	default:
1368 		error = EOPNOTSUPP;
1369 		break;
1370 	}
1371 	return error;
1372 }
1373 static moduledata_t nfscl_mod = {
1374 	"nfscl",
1375 	nfscl_modevent,
1376 	NULL,
1377 };
1378 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1379 
1380 /* So that loader and kldload(2) can find us, wherever we are.. */
1381 MODULE_VERSION(nfscl, 1);
1382 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1383 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1384 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1385 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1386 
1387