xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision c8e7f78a3d28ff6e6223ed136ada8e1e2f34965e)
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 #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 #include <net/route/nhop.h>
55 
56 #include <fs/nfsclient/nfs_kdtrace.h>
57 
58 #ifdef KDTRACE_HOOKS
59 dtrace_nfsclient_attrcache_flush_probe_func_t
60 		dtrace_nfscl_attrcache_flush_done_probe;
61 uint32_t	nfscl_attrcache_flush_done_id;
62 
63 dtrace_nfsclient_attrcache_get_hit_probe_func_t
64 		dtrace_nfscl_attrcache_get_hit_probe;
65 uint32_t	nfscl_attrcache_get_hit_id;
66 
67 dtrace_nfsclient_attrcache_get_miss_probe_func_t
68 		dtrace_nfscl_attrcache_get_miss_probe;
69 uint32_t	nfscl_attrcache_get_miss_id;
70 
71 dtrace_nfsclient_attrcache_load_probe_func_t
72 		dtrace_nfscl_attrcache_load_done_probe;
73 uint32_t	nfscl_attrcache_load_done_id;
74 #endif /* !KDTRACE_HOOKS */
75 
76 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
77 extern struct vop_vector newnfs_vnodeops;
78 extern struct vop_vector newnfs_fifoops;
79 extern uma_zone_t newnfsnode_zone;
80 extern uma_zone_t ncl_pbuf_zone;
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 extern struct mtx nfsrv_dslock_mtx;
88 
89 extern void (*ncl_call_invalcaches)(struct vnode *);
90 
91 SYSCTL_DECL(_vfs_nfs);
92 static int ncl_fileid_maxwarnings = 10;
93 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
94     &ncl_fileid_maxwarnings, 0,
95     "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
96 static volatile int ncl_fileid_nwarnings;
97 
98 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
99     struct nfsvattr *);
100 
101 /*
102  * Comparison function for vfs_hash functions.
103  */
104 int
105 newnfs_vncmpf(struct vnode *vp, void *arg)
106 {
107 	struct nfsfh *nfhp = (struct nfsfh *)arg;
108 	struct nfsnode *np = VTONFS(vp);
109 
110 	if (np->n_fhp->nfh_len != nfhp->nfh_len ||
111 	    NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
112 		return (1);
113 	return (0);
114 }
115 
116 /*
117  * Look up a vnode/nfsnode by file handle.
118  * Callers must check for mount points!!
119  * In all cases, a pointer to a
120  * nfsnode structure is returned.
121  * This variant takes a "struct nfsfh *" as second argument and uses
122  * that structure up, either by hanging off the nfsnode or FREEing it.
123  */
124 int
125 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
126     struct componentname *cnp, struct thread *td, struct nfsnode **npp,
127     int lkflags)
128 {
129 	struct nfsnode *np, *dnp;
130 	struct vnode *vp, *nvp;
131 	struct nfsv4node *newd, *oldd;
132 	int error;
133 	u_int hash;
134 	struct nfsmount *nmp;
135 
136 	nmp = VFSTONFS(mntp);
137 	dnp = VTONFS(dvp);
138 	*npp = NULL;
139 
140 	/*
141 	 * If this is the mount point fh and NFSMNTP_FAKEROOT is set, replace
142 	 * it with the fake fh.
143 	 */
144 	if ((nmp->nm_privflag & NFSMNTP_FAKEROOTFH) != 0 &&
145 	    nmp->nm_fhsize > 0 && nmp->nm_fhsize == nfhp->nfh_len &&
146 	    !NFSBCMP(nmp->nm_fh, nfhp->nfh_fh, nmp->nm_fhsize)) {
147 		free(nfhp, M_NFSFH);
148 		nfhp = malloc(sizeof(struct nfsfh) + NFSX_FHMAX + 1,
149 		    M_NFSFH, M_WAITOK | M_ZERO);
150 		nfhp->nfh_len = NFSX_FHMAX + 1;
151 	}
152 
153 	hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
154 
155 	error = vfs_hash_get(mntp, hash, lkflags,
156 	    td, &nvp, newnfs_vncmpf, nfhp);
157 	if (error == 0 && nvp != NULL) {
158 		/*
159 		 * I believe there is a slight chance that vgonel() could
160 		 * get called on this vnode between when NFSVOPLOCK() drops
161 		 * the VI_LOCK() and vget() acquires it again, so that it
162 		 * hasn't yet had v_usecount incremented. If this were to
163 		 * happen, the VIRF_DOOMED flag would be set, so check for
164 		 * that here. Since we now have the v_usecount incremented,
165 		 * we should be ok until we vrele() it, if the VIRF_DOOMED
166 		 * flag isn't set now.
167 		 */
168 		VI_LOCK(nvp);
169 		if (VN_IS_DOOMED(nvp)) {
170 			VI_UNLOCK(nvp);
171 			vrele(nvp);
172 			error = ENOENT;
173 		} else {
174 			VI_UNLOCK(nvp);
175 		}
176 	}
177 	if (error) {
178 		free(nfhp, M_NFSFH);
179 		return (error);
180 	}
181 	if (nvp != NULL) {
182 		np = VTONFS(nvp);
183 		/*
184 		 * For NFSv4, check to see if it is the same name and
185 		 * replace the name, if it is different.
186 		 */
187 		oldd = newd = NULL;
188 		if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
189 		    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 		    newd = malloc(
197 			sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
198 			+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
199 		    NFSLOCKNODE(np);
200 		    if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
201 			&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
202 			 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
203 			 cnp->cn_namelen) ||
204 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
205 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
206 			 dnp->n_fhp->nfh_len))) {
207 			oldd = np->n_v4;
208 			np->n_v4 = newd;
209 			newd = NULL;
210 			np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
211 			np->n_v4->n4_namelen = cnp->cn_namelen;
212 			NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
213 			    dnp->n_fhp->nfh_len);
214 			NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
215 			    cnp->cn_namelen);
216 		    }
217 		    NFSUNLOCKNODE(np);
218 		}
219 		if (newd != NULL)
220 			free(newd, M_NFSV4NODE);
221 		if (oldd != NULL)
222 			free(oldd, M_NFSV4NODE);
223 		*npp = np;
224 		free(nfhp, M_NFSFH);
225 		return (0);
226 	}
227 	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
228 
229 	error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
230 	if (error) {
231 		uma_zfree(newnfsnode_zone, np);
232 		free(nfhp, M_NFSFH);
233 		return (error);
234 	}
235 	vp = nvp;
236 	KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
237 	vp->v_data = np;
238 	np->n_vnode = vp;
239 	/*
240 	 * Initialize the mutex even if the vnode is going to be a loser.
241 	 * This simplifies the logic in reclaim, which can then unconditionally
242 	 * destroy the mutex (in the case of the loser, or if hash_insert
243 	 * happened to return an error no special casing is needed).
244 	 */
245 	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
246 	lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE |
247 	    LK_CANRECURSE);
248 
249 	/*
250 	 * Are we getting the root? If so, make sure the vnode flags
251 	 * are correct
252 	 */
253 	if (nfhp->nfh_len == NFSX_FHMAX + 1 ||
254 	    (nfhp->nfh_len == nmp->nm_fhsize &&
255 	     !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len))) {
256 		if (vp->v_type == VNON)
257 			vp->v_type = VDIR;
258 		vp->v_vflag |= VV_ROOT;
259 	}
260 
261 	vp->v_vflag |= VV_VMSIZEVNLOCK;
262 
263 	np->n_fhp = nfhp;
264 	/*
265 	 * For NFSv4.0, we have to attach the directory file handle and
266 	 * file name, so that Open Ops can be done later.
267 	 */
268 	if (NFSHASNFSV4(nmp) && !NFSHASNFSV4N(nmp)) {
269 		np->n_v4 = malloc(sizeof (struct nfsv4node)
270 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
271 		    M_WAITOK);
272 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
273 		np->n_v4->n4_namelen = cnp->cn_namelen;
274 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
275 		    dnp->n_fhp->nfh_len);
276 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
277 		    cnp->cn_namelen);
278 	} else {
279 		np->n_v4 = NULL;
280 	}
281 
282 	/*
283 	 * NFS supports recursive and shared locking.
284 	 */
285 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
286 	VN_LOCK_AREC(vp);
287 	VN_LOCK_ASHARE(vp);
288 	error = insmntque(vp, mntp);
289 	if (error != 0) {
290 		*npp = NULL;
291 		mtx_destroy(&np->n_mtx);
292 		lockdestroy(&np->n_excl);
293 		free(nfhp, M_NFSFH);
294 		if (np->n_v4 != NULL)
295 			free(np->n_v4, M_NFSV4NODE);
296 		uma_zfree(newnfsnode_zone, np);
297 		return (error);
298 	}
299 	vn_set_state(vp, VSTATE_CONSTRUCTED);
300 	error = vfs_hash_insert(vp, hash, lkflags,
301 	    td, &nvp, newnfs_vncmpf, nfhp);
302 	if (error)
303 		return (error);
304 	if (nvp != NULL) {
305 		*npp = VTONFS(nvp);
306 		/* vfs_hash_insert() vput()'s the losing vnode */
307 		return (0);
308 	}
309 	*npp = np;
310 
311 	return (0);
312 }
313 
314 /*
315  * Another variant of nfs_nget(). This one is only used by reopen. It
316  * takes almost the same args as nfs_nget(), but only succeeds if an entry
317  * exists in the cache. (Since files should already be "open" with a
318  * vnode ref cnt on the node when reopen calls this, it should always
319  * succeed.)
320  * Also, don't get a vnode lock, since it may already be locked by some
321  * other process that is handling it. This is ok, since all other threads
322  * on the client are blocked by the nfsc_lock being exclusively held by the
323  * caller of this function.
324  */
325 int
326 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
327     struct thread *td, struct nfsnode **npp)
328 {
329 	struct vnode *nvp;
330 	u_int hash;
331 	struct nfsfh *nfhp;
332 	int error;
333 
334 	*npp = NULL;
335 	/* For forced dismounts, just return error. */
336 	if (NFSCL_FORCEDISM(mntp))
337 		return (EINTR);
338 	nfhp = malloc(sizeof (struct nfsfh) + fhsize,
339 	    M_NFSFH, M_WAITOK);
340 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
341 	nfhp->nfh_len = fhsize;
342 
343 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
344 
345 	/*
346 	 * First, try to get the vnode locked, but don't block for the lock.
347 	 */
348 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
349 	    newnfs_vncmpf, nfhp);
350 	if (error == 0 && nvp != NULL) {
351 		NFSVOPUNLOCK(nvp);
352 	} else if (error == EBUSY) {
353 		/*
354 		 * It is safe so long as a vflush() with
355 		 * FORCECLOSE has not been done. Since the Renew thread is
356 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
357 		 * a vflush() with FORCECLOSE, we should be ok here.
358 		 */
359 		if (NFSCL_FORCEDISM(mntp))
360 			error = EINTR;
361 		else {
362 			vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
363 			if (nvp == NULL) {
364 				error = ENOENT;
365 			} else if (VN_IS_DOOMED(nvp)) {
366 				error = ENOENT;
367 				vrele(nvp);
368 			} else {
369 				error = 0;
370 			}
371 		}
372 	}
373 	free(nfhp, M_NFSFH);
374 	if (error)
375 		return (error);
376 	if (nvp != NULL) {
377 		*npp = VTONFS(nvp);
378 		return (0);
379 	}
380 	return (EINVAL);
381 }
382 
383 static void
384 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
385     struct nfsvattr *newnap)
386 {
387 	int off;
388 
389 	if (ncl_fileid_maxwarnings >= 0 &&
390 	    ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
391 		return;
392 	off = 0;
393 	if (ncl_fileid_maxwarnings >= 0) {
394 		if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
395 			off = 1;
396 	}
397 
398 	printf("newnfs: server '%s' error: fileid changed. "
399 	    "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
400 	    "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
401 	    nmp->nm_com.nmcom_hostname,
402 	    (uintmax_t)nmp->nm_fsid[0],
403 	    (uintmax_t)nmp->nm_fsid[1],
404 	    (uintmax_t)oldnap->na_fileid,
405 	    (uintmax_t)newnap->na_fileid);
406 
407 	if (off)
408 		printf("newnfs: Logged %d times about fileid corruption; "
409 		    "going quiet to avoid spamming logs excessively. (Limit "
410 		    "is: %d).\n", ncl_fileid_nwarnings,
411 		    ncl_fileid_maxwarnings);
412 }
413 
414 void
415 ncl_copy_vattr(struct vattr *dst, struct vattr *src)
416 {
417 	dst->va_type = src->va_type;
418 	dst->va_mode = src->va_mode;
419 	dst->va_nlink = src->va_nlink;
420 	dst->va_uid = src->va_uid;
421 	dst->va_gid = src->va_gid;
422 	dst->va_fsid = src->va_fsid;
423 	dst->va_fileid = src->va_fileid;
424 	dst->va_size = src->va_size;
425 	dst->va_blocksize = src->va_blocksize;
426 	dst->va_atime = src->va_atime;
427 	dst->va_mtime = src->va_mtime;
428 	dst->va_ctime = src->va_ctime;
429 	dst->va_birthtime = src->va_birthtime;
430 	dst->va_gen = src->va_gen;
431 	dst->va_flags = src->va_flags;
432 	dst->va_rdev = src->va_rdev;
433 	dst->va_bytes = src->va_bytes;
434 	dst->va_filerev = src->va_filerev;
435 }
436 
437 /*
438  * Load the attribute cache (that lives in the nfsnode entry) with
439  * the attributes of the second argument and
440  * Iff vaper not NULL
441  *    copy the attributes to *vaper
442  * Similar to nfs_loadattrcache(), except the attributes are passed in
443  * instead of being parsed out of the mbuf list.
444  */
445 int
446 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
447     int writeattr, int dontshrink)
448 {
449 	struct vnode *vp = *vpp;
450 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
451 	struct nfsnode *np;
452 	struct nfsmount *nmp;
453 	struct timespec mtime_save;
454 	int error, force_fid_err;
455 	dev_t topfsid;
456 
457 	error = 0;
458 
459 	/*
460 	 * If v_type == VNON it is a new node, so fill in the v_type,
461 	 * n_mtime fields. Check to see if it represents a special
462 	 * device, and if so, check for a possible alias. Once the
463 	 * correct vnode has been obtained, fill in the rest of the
464 	 * information.
465 	 */
466 	np = VTONFS(vp);
467 	NFSLOCKNODE(np);
468 	if (vp->v_type != nvap->va_type) {
469 		vp->v_type = nvap->va_type;
470 		if (vp->v_type == VFIFO)
471 			vp->v_op = &newnfs_fifoops;
472 		np->n_mtime = nvap->va_mtime;
473 	}
474 	nmp = VFSTONFS(vp->v_mount);
475 	vap = &np->n_vattr.na_vattr;
476 	mtime_save = vap->va_mtime;
477 	if (writeattr) {
478 		np->n_vattr.na_filerev = nap->na_filerev;
479 		np->n_vattr.na_size = nap->na_size;
480 		np->n_vattr.na_mtime = nap->na_mtime;
481 		np->n_vattr.na_ctime = nap->na_ctime;
482 		np->n_vattr.na_btime = nap->na_btime;
483 		np->n_vattr.na_fsid = nap->na_fsid;
484 		np->n_vattr.na_mode = nap->na_mode;
485 	} else {
486 		force_fid_err = 0;
487 		KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
488 		    force_fid_err);
489 		/*
490 		 * BROKEN NFS SERVER OR MIDDLEWARE
491 		 *
492 		 * Certain NFS servers (certain old proprietary filers ca.
493 		 * 2006) or broken middleboxes (e.g. WAN accelerator products)
494 		 * will respond to GETATTR requests with results for a
495 		 * different fileid.
496 		 *
497 		 * The WAN accelerator we've observed not only serves stale
498 		 * cache results for a given file, it also occasionally serves
499 		 * results for wholly different files.  This causes surprising
500 		 * problems; for example the cached size attribute of a file
501 		 * may truncate down and then back up, resulting in zero
502 		 * regions in file contents read by applications.  We observed
503 		 * this reliably with Clang and .c files during parallel build.
504 		 * A pcap revealed packet fragmentation and GETATTR RPC
505 		 * responses with wholly wrong fileids.
506 		 * For the case where the file handle is a fake one
507 		 * generated via the "syskrb5" mount option and
508 		 * the old fileid is 2, ignore the test, since this might
509 		 * be replacing the fake attributes with correct ones.
510 		 */
511 		if ((np->n_vattr.na_fileid != 0 &&
512 		     np->n_vattr.na_fileid != nap->na_fileid &&
513 		     (np->n_vattr.na_fileid != 2 || !NFSHASSYSKRB5(nmp) ||
514 		      np->n_fhp->nfh_len != NFSX_FHMAX + 1)) ||
515 		    force_fid_err) {
516 			nfscl_warn_fileid(nmp, &np->n_vattr, nap);
517 			error = EIDRM;
518 			goto out;
519 		}
520 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
521 		    sizeof (struct nfsvattr));
522 	}
523 
524 	/*
525 	 * For NFSv4, the server's export may be a tree of file systems
526 	 * where a fileno is a unique value within each file system.
527 	 * na_filesid[0,1] uniquely identify the server file system
528 	 * and nm_fsid[0,1] is the value for the root file system mounted.
529 	 * As such, the value of va_fsid generated by vn_fsid() represents
530 	 * the root file system on the server and a different value for
531 	 * va_fsid is needed for the other server file systems.  This
532 	 * va_fsid is ideally unique for all of the server file systems,
533 	 * so a 64bit hash on na_filesid[0,1] is calculated.
534 	 * Although highly unlikely that the fnv_64_hash() will be
535 	 * the same as the root, test for this case and recalculate the hash.
536 	 */
537 	vn_fsid(vp, vap);
538 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
539 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
540 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
541 		topfsid = vap->va_fsid;
542 		vap->va_fsid = FNV1_64_INIT;
543 		do {
544 			vap->va_fsid = fnv_64_buf(np->n_vattr.na_filesid,
545 			    sizeof(np->n_vattr.na_filesid), vap->va_fsid);
546 		} while (vap->va_fsid == topfsid);
547 	}
548 
549 	np->n_attrstamp = time_second;
550 	if (vap->va_size != np->n_size) {
551 		if (vap->va_type == VREG) {
552 			if (dontshrink && vap->va_size < np->n_size) {
553 				/*
554 				 * We've been told not to shrink the file;
555 				 * zero np->n_attrstamp to indicate that
556 				 * the attributes are stale.
557 				 */
558 				vap->va_size = np->n_size;
559 				np->n_attrstamp = 0;
560 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
561 			} else if (np->n_flag & NMODIFIED) {
562 				/*
563 				 * We've modified the file: Use the larger
564 				 * of our size, and the server's size.
565 				 */
566 				if (vap->va_size < np->n_size) {
567 					vap->va_size = np->n_size;
568 				} else {
569 					np->n_size = vap->va_size;
570 					np->n_flag |= NSIZECHANGED;
571 				}
572 			} else {
573 				np->n_size = vap->va_size;
574 				np->n_flag |= NSIZECHANGED;
575 			}
576 		} else {
577 			np->n_size = vap->va_size;
578 		}
579 	}
580 	/*
581 	 * The following checks are added to prevent a race between (say)
582 	 * a READDIR+ and a WRITE.
583 	 * READDIR+, WRITE requests sent out.
584 	 * READDIR+ resp, WRITE resp received on client.
585 	 * However, the WRITE resp was handled before the READDIR+ resp
586 	 * causing the post op attrs from the write to be loaded first
587 	 * and the attrs from the READDIR+ to be loaded later. If this
588 	 * happens, we have stale attrs loaded into the attrcache.
589 	 * We detect this by for the mtime moving back. We invalidate the
590 	 * attrcache when this happens.
591 	 */
592 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
593 		/* Size changed or mtime went backwards */
594 		np->n_attrstamp = 0;
595 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
596 	}
597 	if (vaper != NULL) {
598 		ncl_copy_vattr(vaper, vap);
599 		if (np->n_flag & NCHG) {
600 			if (np->n_flag & NACC)
601 				vaper->va_atime = np->n_atim;
602 			if (np->n_flag & NUPD)
603 				vaper->va_mtime = np->n_mtim;
604 		}
605 	}
606 
607 out:
608 #ifdef KDTRACE_HOOKS
609 	if (np->n_attrstamp != 0)
610 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
611 #endif
612 	(void)ncl_pager_setsize(vp, NULL);
613 	return (error);
614 }
615 
616 /*
617  * Call vnode_pager_setsize() if the size of the node changed, as
618  * recorded in nfsnode vs. v_object, or delay the call if notifying
619  * the pager is not possible at the moment.
620  *
621  * If nsizep is non-NULL, the call is delayed and the new node size is
622  * provided.  Caller should itself call vnode_pager_setsize() if
623  * function returned true.  If nsizep is NULL, function tries to call
624  * vnode_pager_setsize() itself if needed and possible, and the nfs
625  * node is unlocked unconditionally, the return value is not useful.
626  */
627 bool
628 ncl_pager_setsize(struct vnode *vp, u_quad_t *nsizep)
629 {
630 	struct nfsnode *np;
631 	vm_object_t object;
632 	struct vattr *vap;
633 	u_quad_t nsize;
634 	bool setnsize;
635 
636 	np = VTONFS(vp);
637 	NFSASSERTNODE(np);
638 
639 	vap = &np->n_vattr.na_vattr;
640 	nsize = vap->va_size;
641 	object = vp->v_object;
642 	setnsize = false;
643 
644 	if (object != NULL && nsize != object->un_pager.vnp.vnp_size) {
645 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
646 		    (curthread->td_pflags2 & TDP2_SBPAGES) == 0)
647 			setnsize = true;
648 		else
649 			np->n_flag |= NVNSETSZSKIP;
650 	}
651 	if (nsizep == NULL) {
652 		NFSUNLOCKNODE(np);
653 		if (setnsize)
654 			vnode_pager_setsize(vp, nsize);
655 		setnsize = false;
656 	} else {
657 		*nsizep = nsize;
658 	}
659 	return (setnsize);
660 }
661 
662 /*
663  * Fill in the client id name. For these bytes:
664  * 1 - they must be unique
665  * 2 - they should be persistent across client reboots
666  * 1 is more critical than 2
667  * Use the mount point's unique id plus either the uuid or, if that
668  * isn't set, random junk.
669  */
670 void
671 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
672 {
673 	int uuidlen;
674 
675 	/*
676 	 * First, put in the 64bit mount point identifier.
677 	 */
678 	if (idlen >= sizeof (u_int64_t)) {
679 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
680 		cp += sizeof (u_int64_t);
681 		idlen -= sizeof (u_int64_t);
682 	}
683 
684 	/*
685 	 * If uuid is non-zero length, use it.
686 	 */
687 	uuidlen = strlen(uuid);
688 	if (uuidlen > 0 && idlen >= uuidlen) {
689 		NFSBCOPY(uuid, cp, uuidlen);
690 		cp += uuidlen;
691 		idlen -= uuidlen;
692 	}
693 
694 	/*
695 	 * This only normally happens if the uuid isn't set.
696 	 */
697 	while (idlen > 0) {
698 		*cp++ = (u_int8_t)(arc4random() % 256);
699 		idlen--;
700 	}
701 }
702 
703 /*
704  * Fill in a lock owner name. For now, pid + the process's creation time.
705  */
706 void
707 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
708 {
709 	union {
710 		u_int32_t	lval;
711 		u_int8_t	cval[4];
712 	} tl;
713 	struct proc *p;
714 
715 	if (id == NULL) {
716 		/* Return the single open_owner of all 0 bytes. */
717 		bzero(cp, NFSV4CL_LOCKNAMELEN);
718 		return;
719 	}
720 	if ((flags & F_POSIX) != 0) {
721 		p = (struct proc *)id;
722 		tl.lval = p->p_pid;
723 		*cp++ = tl.cval[0];
724 		*cp++ = tl.cval[1];
725 		*cp++ = tl.cval[2];
726 		*cp++ = tl.cval[3];
727 		tl.lval = p->p_stats->p_start.tv_sec;
728 		*cp++ = tl.cval[0];
729 		*cp++ = tl.cval[1];
730 		*cp++ = tl.cval[2];
731 		*cp++ = tl.cval[3];
732 		tl.lval = p->p_stats->p_start.tv_usec;
733 		*cp++ = tl.cval[0];
734 		*cp++ = tl.cval[1];
735 		*cp++ = tl.cval[2];
736 		*cp = tl.cval[3];
737 	} else if ((flags & F_FLOCK) != 0) {
738 		bcopy(&id, cp, sizeof(id));
739 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
740 	} else {
741 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
742 		bzero(cp, NFSV4CL_LOCKNAMELEN);
743 	}
744 }
745 
746 /*
747  * Find the parent process for the thread passed in as an argument.
748  * If none exists, return NULL, otherwise return a thread for the parent.
749  * (Can be any of the threads, since it is only used for td->td_proc.)
750  */
751 NFSPROC_T *
752 nfscl_getparent(struct thread *td)
753 {
754 	struct proc *p;
755 	struct thread *ptd;
756 
757 	if (td == NULL)
758 		return (NULL);
759 	p = td->td_proc;
760 	if (p->p_pid == 0)
761 		return (NULL);
762 	p = p->p_pptr;
763 	if (p == NULL)
764 		return (NULL);
765 	ptd = TAILQ_FIRST(&p->p_threads);
766 	return (ptd);
767 }
768 
769 /*
770  * Start up the renew kernel thread.
771  */
772 static void
773 start_nfscl(void *arg)
774 {
775 	struct nfsclclient *clp;
776 	struct thread *td;
777 
778 	clp = (struct nfsclclient *)arg;
779 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
780 	nfscl_renewthread(clp, td);
781 	kproc_exit(0);
782 }
783 
784 void
785 nfscl_start_renewthread(struct nfsclclient *clp)
786 {
787 
788 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
789 	    "nfscl");
790 }
791 
792 /*
793  * Handle wcc_data.
794  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
795  * as the first Op after PutFH.
796  * (For NFSv4, the postop attributes are after the Op, so they can't be
797  *  parsed here. A separate call to nfscl_postop_attr() is required.)
798  */
799 int
800 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
801     struct nfsvattr *nap, int *flagp, int *wccflagp, uint64_t *repsizep)
802 {
803 	u_int32_t *tl;
804 	struct nfsnode *np = VTONFS(vp);
805 	struct nfsvattr nfsva;
806 	int error = 0;
807 
808 	if (wccflagp != NULL)
809 		*wccflagp = 0;
810 	if (nd->nd_flag & ND_NFSV3) {
811 		*flagp = 0;
812 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
813 		if (*tl == newnfs_true) {
814 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
815 			if (wccflagp != NULL) {
816 				NFSLOCKNODE(np);
817 				*wccflagp = (np->n_mtime.tv_sec ==
818 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
819 				    np->n_mtime.tv_nsec ==
820 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
821 				NFSUNLOCKNODE(np);
822 			}
823 		}
824 		error = nfscl_postop_attr(nd, nap, flagp);
825 		if (wccflagp != NULL && *flagp == 0)
826 			*wccflagp = 0;
827 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
828 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
829 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
830 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
831 		    NULL, NULL, NULL, NULL, NULL);
832 		if (error)
833 			return (error);
834 		/*
835 		 * Get rid of Op# and status for next op.
836 		 */
837 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
838 		if (*++tl)
839 			nd->nd_flag |= ND_NOMOREDATA;
840 		if (repsizep != NULL)
841 			*repsizep = nfsva.na_size;
842 		if (wccflagp != NULL &&
843 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
844 			NFSLOCKNODE(np);
845 			*wccflagp = (np->n_mtime.tv_sec ==
846 			    nfsva.na_vattr.va_mtime.tv_sec &&
847 			    np->n_mtime.tv_nsec ==
848 			    nfsva.na_vattr.va_mtime.tv_sec);
849 			NFSUNLOCKNODE(np);
850 		}
851 	}
852 nfsmout:
853 	return (error);
854 }
855 
856 /*
857  * Get postop attributes.
858  */
859 int
860 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp)
861 {
862 	u_int32_t *tl;
863 	int error = 0;
864 
865 	*retp = 0;
866 	if (nd->nd_flag & ND_NOMOREDATA)
867 		return (error);
868 	if (nd->nd_flag & ND_NFSV3) {
869 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
870 		*retp = fxdr_unsigned(int, *tl);
871 	} else if (nd->nd_flag & ND_NFSV4) {
872 		/*
873 		 * For NFSv4, the postop attr are at the end, so no point
874 		 * in looking if nd_repstat != 0.
875 		 */
876 		if (!nd->nd_repstat) {
877 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
878 			if (*(tl + 1))
879 				/* should never happen since nd_repstat != 0 */
880 				nd->nd_flag |= ND_NOMOREDATA;
881 			else
882 				*retp = 1;
883 		}
884 	} else if (!nd->nd_repstat) {
885 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
886 		*retp = 1;
887 	}
888 	if (*retp) {
889 		error = nfsm_loadattr(nd, nap);
890 		if (error)
891 			*retp = 0;
892 	}
893 nfsmout:
894 	return (error);
895 }
896 
897 /*
898  * nfscl_request() - mostly a wrapper for newnfs_request().
899  */
900 int
901 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
902     struct ucred *cred)
903 {
904 	int ret, vers;
905 	struct nfsmount *nmp;
906 
907 	nmp = VFSTONFS(vp->v_mount);
908 	if (nd->nd_flag & ND_NFSV4)
909 		vers = NFS_VER4;
910 	else if (nd->nd_flag & ND_NFSV3)
911 		vers = NFS_VER3;
912 	else
913 		vers = NFS_VER2;
914 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
915 		NFS_PROG, vers, NULL, 1, NULL, NULL);
916 	return (ret);
917 }
918 
919 /*
920  * fill in this bsden's variant of statfs using nfsstatfs.
921  */
922 void
923 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
924 {
925 	struct statfs *sbp = (struct statfs *)statfs;
926 
927 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
928 		sbp->f_bsize = NFS_FABLKSIZE;
929 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
930 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
931 		/*
932 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
933 		 * the value after dividing by NFS_FABLKSIZE is small
934 		 * enough that it will fit in 63bits, so it is ok to
935 		 * assign it to f_bavail without fear that it will become
936 		 * negative.
937 		 */
938 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
939 		sbp->f_files = sfp->sf_tfiles;
940 		/* Since f_ffree is int64_t, clip it to 63bits. */
941 		if (sfp->sf_ffiles > INT64_MAX)
942 			sbp->f_ffree = INT64_MAX;
943 		else
944 			sbp->f_ffree = sfp->sf_ffiles;
945 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
946 		/*
947 		 * The type casts to (int32_t) ensure that this code is
948 		 * compatible with the old NFS client, in that it will
949 		 * propagate bit31 to the high order bits. This may or may
950 		 * not be correct for NFSv2, but since it is a legacy
951 		 * environment, I'd rather retain backwards compatibility.
952 		 */
953 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
954 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
955 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
956 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
957 		sbp->f_files = 0;
958 		sbp->f_ffree = 0;
959 	}
960 }
961 
962 /*
963  * Use the fsinfo stuff to update the mount point.
964  */
965 void
966 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
967 {
968 
969 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
970 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
971 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
972 		    ~(NFS_FABLKSIZE - 1);
973 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
974 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
975 		if (nmp->nm_wsize == 0)
976 			nmp->nm_wsize = fsp->fs_wtmax;
977 	}
978 	if (nmp->nm_wsize < NFS_FABLKSIZE)
979 		nmp->nm_wsize = NFS_FABLKSIZE;
980 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
981 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
982 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
983 		    ~(NFS_FABLKSIZE - 1);
984 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
985 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
986 		if (nmp->nm_rsize == 0)
987 			nmp->nm_rsize = fsp->fs_rtmax;
988 	}
989 	if (nmp->nm_rsize < NFS_FABLKSIZE)
990 		nmp->nm_rsize = NFS_FABLKSIZE;
991 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
992 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
993 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
994 		    ~(NFS_DIRBLKSIZ - 1);
995 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
996 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
997 		if (nmp->nm_readdirsize == 0)
998 			nmp->nm_readdirsize = fsp->fs_rtmax;
999 	}
1000 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
1001 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
1002 	if (fsp->fs_maxfilesize > 0 &&
1003 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
1004 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
1005 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
1006 	nmp->nm_state |= NFSSTA_GOTFSINFO;
1007 }
1008 
1009 /*
1010  * Lookups source address which should be used to communicate with
1011  * @nmp and stores it inside @pdst.
1012  *
1013  * Returns 0 on success.
1014  */
1015 u_int8_t *
1016 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
1017 {
1018 #if defined(INET6) || defined(INET)
1019 	int fibnum;
1020 
1021 	fibnum = curthread->td_proc->p_fibnum;
1022 #endif
1023 #ifdef INET
1024 	if (nmp->nm_nam->sa_family == AF_INET) {
1025 		struct epoch_tracker et;
1026 		struct nhop_object *nh;
1027 		struct sockaddr_in *sin;
1028 		struct in_addr addr = {};
1029 
1030 		sin = (struct sockaddr_in *)nmp->nm_nam;
1031 		NET_EPOCH_ENTER(et);
1032 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1033 		nh = fib4_lookup(fibnum, sin->sin_addr, 0, NHR_NONE, 0);
1034 		if (nh != NULL) {
1035 			addr = IA_SIN(ifatoia(nh->nh_ifa))->sin_addr;
1036 			if (IN_LOOPBACK(ntohl(addr.s_addr))) {
1037 				/* Ignore loopback addresses */
1038 				nh = NULL;
1039 			}
1040 		}
1041 		CURVNET_RESTORE();
1042 		NET_EPOCH_EXIT(et);
1043 
1044 		if (nh == NULL)
1045 			return (NULL);
1046 		*isinet6p = 0;
1047 		*((struct in_addr *)paddr) = addr;
1048 
1049 		return (u_int8_t *)paddr;
1050 	}
1051 #endif
1052 #ifdef INET6
1053 	if (nmp->nm_nam->sa_family == AF_INET6) {
1054 		struct epoch_tracker et;
1055 		struct sockaddr_in6 *sin6;
1056 		int error;
1057 
1058 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1059 
1060 		NET_EPOCH_ENTER(et);
1061 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1062 		error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1063 		    sin6->sin6_scope_id, NULL, paddr, NULL);
1064 		CURVNET_RESTORE();
1065 		NET_EPOCH_EXIT(et);
1066 		if (error != 0)
1067 			return (NULL);
1068 
1069 		if (IN6_IS_ADDR_LOOPBACK(paddr))
1070 			return (NULL);
1071 
1072 		/* Scope is embedded in */
1073 		*isinet6p = 1;
1074 
1075 		return (u_int8_t *)paddr;
1076 	}
1077 #endif
1078 	return (NULL);
1079 }
1080 
1081 /*
1082  * Copy NFS uid, gids from the cred structure.
1083  */
1084 void
1085 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1086 {
1087 	int i;
1088 
1089 	KASSERT(cr->cr_ngroups >= 0,
1090 	    ("newnfs_copyincred: negative cr_ngroups"));
1091 	nfscr->nfsc_uid = cr->cr_uid;
1092 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1093 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1094 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1095 }
1096 
1097 /*
1098  * Do any client specific initialization.
1099  */
1100 void
1101 nfscl_init(void)
1102 {
1103 	static int inited = 0;
1104 
1105 	if (inited)
1106 		return;
1107 	inited = 1;
1108 	nfscl_inited = 1;
1109 	ncl_pbuf_zone = pbuf_zsecond_create("nfspbuf", nswbuf / 2);
1110 }
1111 
1112 /*
1113  * Check each of the attributes to be set, to ensure they aren't already
1114  * the correct value. Disable setting ones already correct.
1115  */
1116 int
1117 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1118 {
1119 
1120 	if (vap->va_mode != (mode_t)VNOVAL) {
1121 		if (vap->va_mode == nvap->na_mode)
1122 			vap->va_mode = (mode_t)VNOVAL;
1123 	}
1124 	if (vap->va_uid != (uid_t)VNOVAL) {
1125 		if (vap->va_uid == nvap->na_uid)
1126 			vap->va_uid = (uid_t)VNOVAL;
1127 	}
1128 	if (vap->va_gid != (gid_t)VNOVAL) {
1129 		if (vap->va_gid == nvap->na_gid)
1130 			vap->va_gid = (gid_t)VNOVAL;
1131 	}
1132 	if (vap->va_size != VNOVAL) {
1133 		if (vap->va_size == nvap->na_size)
1134 			vap->va_size = VNOVAL;
1135 	}
1136 
1137 	/*
1138 	 * We are normally called with only a partially initialized
1139 	 * VAP.  Since the NFSv3 spec says that server may use the
1140 	 * file attributes to store the verifier, the spec requires
1141 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1142 	 * in atime, but we can't really assume that all servers will
1143 	 * so we ensure that our SETATTR sets both atime and mtime.
1144 	 * Set the VA_UTIMES_NULL flag for this case, so that
1145 	 * the server's time will be used.  This is needed to
1146 	 * work around a bug in some Solaris servers, where
1147 	 * setting the time TOCLIENT causes the Setattr RPC
1148 	 * to return NFS_OK, but not set va_mode.
1149 	 */
1150 	if (vap->va_mtime.tv_sec == VNOVAL) {
1151 		vfs_timestamp(&vap->va_mtime);
1152 		vap->va_vaflags |= VA_UTIMES_NULL;
1153 	}
1154 	if (vap->va_atime.tv_sec == VNOVAL)
1155 		vap->va_atime = vap->va_mtime;
1156 	return (1);
1157 }
1158 
1159 /*
1160  * Map nfsv4 errors to errno.h errors.
1161  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1162  * error should only be returned for the Open, Create and Setattr Ops.
1163  * As such, most calls can just pass in 0 for those arguments.
1164  */
1165 int
1166 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1167 {
1168 	struct proc *p;
1169 
1170 	if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1171 		return (error);
1172 	if (td != NULL)
1173 		p = td->td_proc;
1174 	else
1175 		p = NULL;
1176 	switch (error) {
1177 	case NFSERR_BADOWNER:
1178 		tprintf(p, LOG_INFO,
1179 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1180 		    uid, gid);
1181 		return (EPERM);
1182 	case NFSERR_BADNAME:
1183 	case NFSERR_BADCHAR:
1184 		printf("nfsv4 char/name not handled by server\n");
1185 		return (ENOENT);
1186 	case NFSERR_STALECLIENTID:
1187 	case NFSERR_STALESTATEID:
1188 	case NFSERR_EXPIRED:
1189 	case NFSERR_BADSTATEID:
1190 	case NFSERR_BADSESSION:
1191 		printf("nfsv4 recover err returned %d\n", error);
1192 		return (EIO);
1193 	case NFSERR_BADHANDLE:
1194 	case NFSERR_SERVERFAULT:
1195 	case NFSERR_BADTYPE:
1196 	case NFSERR_FHEXPIRED:
1197 	case NFSERR_RESOURCE:
1198 	case NFSERR_MOVED:
1199 	case NFSERR_MINORVERMISMATCH:
1200 	case NFSERR_OLDSTATEID:
1201 	case NFSERR_BADSEQID:
1202 	case NFSERR_LEASEMOVED:
1203 	case NFSERR_RECLAIMBAD:
1204 	case NFSERR_BADXDR:
1205 	case NFSERR_OPILLEGAL:
1206 		printf("nfsv4 client/server protocol prob err=%d\n",
1207 		    error);
1208 		return (EIO);
1209 	case NFSERR_NOFILEHANDLE:
1210 		printf("nfsv4 no file handle: usually means the file "
1211 		    "system is not exported on the NFSv4 server\n");
1212 		return (EIO);
1213 	case NFSERR_WRONGSEC:
1214 		tprintf(p, LOG_INFO, "NFSv4 error WrongSec: You probably need a"
1215 		    " Kerberos TGT\n");
1216 		return (EIO);
1217 	default:
1218 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1219 		return (EIO);
1220 	};
1221 }
1222 
1223 /*
1224  * Check to see if the process for this owner exists. Return 1 if it doesn't
1225  * and 0 otherwise.
1226  */
1227 int
1228 nfscl_procdoesntexist(u_int8_t *own)
1229 {
1230 	union {
1231 		u_int32_t	lval;
1232 		u_int8_t	cval[4];
1233 	} tl;
1234 	struct proc *p;
1235 	pid_t pid;
1236 	int i, ret = 0;
1237 
1238 	/* For the single open_owner of all 0 bytes, just return 0. */
1239 	for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1240 		if (own[i] != 0)
1241 			break;
1242 	if (i == NFSV4CL_LOCKNAMELEN)
1243 		return (0);
1244 
1245 	tl.cval[0] = *own++;
1246 	tl.cval[1] = *own++;
1247 	tl.cval[2] = *own++;
1248 	tl.cval[3] = *own++;
1249 	pid = tl.lval;
1250 	p = pfind_any_locked(pid);
1251 	if (p == NULL)
1252 		return (1);
1253 	if (p->p_stats == NULL) {
1254 		PROC_UNLOCK(p);
1255 		return (0);
1256 	}
1257 	tl.cval[0] = *own++;
1258 	tl.cval[1] = *own++;
1259 	tl.cval[2] = *own++;
1260 	tl.cval[3] = *own++;
1261 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1262 		ret = 1;
1263 	} else {
1264 		tl.cval[0] = *own++;
1265 		tl.cval[1] = *own++;
1266 		tl.cval[2] = *own++;
1267 		tl.cval[3] = *own;
1268 		if (tl.lval != p->p_stats->p_start.tv_usec)
1269 			ret = 1;
1270 	}
1271 	PROC_UNLOCK(p);
1272 	return (ret);
1273 }
1274 
1275 /*
1276  * - nfs pseudo system call for the client
1277  */
1278 /*
1279  * MPSAFE
1280  */
1281 static int
1282 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1283 {
1284 	struct file *fp;
1285 	struct nfscbd_args nfscbdarg;
1286 	struct nfsd_nfscbd_args nfscbdarg2;
1287 	struct nameidata nd;
1288 	struct nfscl_dumpmntopts dumpmntopts;
1289 	cap_rights_t rights;
1290 	char *buf;
1291 	int error;
1292 	struct mount *mp;
1293 	struct nfsmount *nmp;
1294 
1295 	NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td));
1296 	if (uap->flag & NFSSVC_CBADDSOCK) {
1297 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1298 		if (error)
1299 			goto out;
1300 		/*
1301 		 * Since we don't know what rights might be required,
1302 		 * pretend that we need them all. It is better to be too
1303 		 * careful than too reckless.
1304 		 */
1305 		error = fget(td, nfscbdarg.sock,
1306 		    cap_rights_init_one(&rights, CAP_SOCK_CLIENT), &fp);
1307 		if (error)
1308 			goto out;
1309 		if (fp->f_type != DTYPE_SOCKET) {
1310 			fdrop(fp, td);
1311 			error = EPERM;
1312 			goto out;
1313 		}
1314 		error = nfscbd_addsock(fp);
1315 		fdrop(fp, td);
1316 		if (!error && nfscl_enablecallb == 0) {
1317 			nfsv4_cbport = nfscbdarg.port;
1318 			nfscl_enablecallb = 1;
1319 		}
1320 	} else if (uap->flag & NFSSVC_NFSCBD) {
1321 		if (uap->argp == NULL) {
1322 			error = EINVAL;
1323 			goto out;
1324 		}
1325 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1326 		    sizeof(nfscbdarg2));
1327 		if (error)
1328 			goto out;
1329 		error = nfscbd_nfsd(td, &nfscbdarg2);
1330 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1331 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1332 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1333 		    dumpmntopts.ndmnt_blen > 1024))
1334 			error = EINVAL;
1335 		if (error == 0)
1336 			error = nfsrv_lookupfilename(&nd,
1337 			    dumpmntopts.ndmnt_fname, td);
1338 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1339 		    "nfs") != 0) {
1340 			vput(nd.ni_vp);
1341 			error = EINVAL;
1342 		}
1343 		if (error == 0) {
1344 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK |
1345 			    M_ZERO);
1346 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1347 			    dumpmntopts.ndmnt_blen);
1348 			vput(nd.ni_vp);
1349 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1350 			    dumpmntopts.ndmnt_blen);
1351 			free(buf, M_TEMP);
1352 		}
1353 	} else if (uap->flag & NFSSVC_FORCEDISM) {
1354 		buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
1355 		error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
1356 		if (error == 0) {
1357 			nmp = NULL;
1358 			mtx_lock(&mountlist_mtx);
1359 			TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1360 				if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
1361 				    0 && strcmp(mp->mnt_stat.f_fstypename,
1362 				    "nfs") == 0 && mp->mnt_data != NULL) {
1363 					nmp = VFSTONFS(mp);
1364 					NFSDDSLOCK();
1365 					if (nfsv4_findmirror(nmp) != NULL) {
1366 						NFSDDSUNLOCK();
1367 						error = ENXIO;
1368 						nmp = NULL;
1369 						break;
1370 					}
1371 					mtx_lock(&nmp->nm_mtx);
1372 					if ((nmp->nm_privflag &
1373 					    NFSMNTP_FORCEDISM) == 0) {
1374 						nmp->nm_privflag |=
1375 						   (NFSMNTP_FORCEDISM |
1376 						    NFSMNTP_CANCELRPCS);
1377 						mtx_unlock(&nmp->nm_mtx);
1378 					} else {
1379 						mtx_unlock(&nmp->nm_mtx);
1380 						nmp = NULL;
1381 					}
1382 					NFSDDSUNLOCK();
1383 					break;
1384 				}
1385 			}
1386 			mtx_unlock(&mountlist_mtx);
1387 
1388 			if (nmp != NULL) {
1389 				/*
1390 				 * Call newnfs_nmcancelreqs() to cause
1391 				 * any RPCs in progress on the mount point to
1392 				 * fail.
1393 				 * This will cause any process waiting for an
1394 				 * RPC to complete while holding a vnode lock
1395 				 * on the mounted-on vnode (such as "df" or
1396 				 * a non-forced "umount") to fail.
1397 				 * This will unlock the mounted-on vnode so
1398 				 * a forced dismount can succeed.
1399 				 * Then clear NFSMNTP_CANCELRPCS and wakeup(),
1400 				 * so that nfs_unmount() can complete.
1401 				 */
1402 				newnfs_nmcancelreqs(nmp);
1403 				mtx_lock(&nmp->nm_mtx);
1404 				nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
1405 				wakeup(nmp);
1406 				mtx_unlock(&nmp->nm_mtx);
1407 			} else if (error == 0)
1408 				error = EINVAL;
1409 		}
1410 		free(buf, M_TEMP);
1411 	} else {
1412 		error = EINVAL;
1413 	}
1414 out:
1415 	NFSD_CURVNET_RESTORE();
1416 	return (error);
1417 }
1418 
1419 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1420 
1421 /*
1422  * Called once to initialize data structures...
1423  */
1424 static int
1425 nfscl_modevent(module_t mod, int type, void *data)
1426 {
1427 	int error = 0;
1428 	static int loaded = 0;
1429 
1430 	switch (type) {
1431 	case MOD_LOAD:
1432 		if (loaded)
1433 			return (0);
1434 		newnfs_portinit();
1435 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1436 		nfscl_init();
1437 		NFSD_LOCK();
1438 		nfsrvd_cbinit(0);
1439 		NFSD_UNLOCK();
1440 		ncl_call_invalcaches = ncl_invalcaches;
1441 		nfsd_call_nfscl = nfssvc_nfscl;
1442 		loaded = 1;
1443 		break;
1444 
1445 	case MOD_UNLOAD:
1446 		if (nfs_numnfscbd != 0) {
1447 			error = EBUSY;
1448 			break;
1449 		}
1450 
1451 		/*
1452 		 * XXX: Unloading of nfscl module is unsupported.
1453 		 */
1454 #if 0
1455 		ncl_call_invalcaches = NULL;
1456 		nfsd_call_nfscl = NULL;
1457 		uma_zdestroy(ncl_pbuf_zone);
1458 		/* and get rid of the mutexes */
1459 		mtx_destroy(&ncl_iod_mutex);
1460 		loaded = 0;
1461 		break;
1462 #else
1463 		/* FALLTHROUGH */
1464 #endif
1465 	default:
1466 		error = EOPNOTSUPP;
1467 		break;
1468 	}
1469 	return error;
1470 }
1471 static moduledata_t nfscl_mod = {
1472 	"nfscl",
1473 	nfscl_modevent,
1474 	NULL,
1475 };
1476 /*
1477  * This is the main module declaration for the NFS client.  The
1478  * nfscl_modevent() function is needed to ensure that the module
1479  * cannot be unloaded, among other things.
1480  * There is also a module declaration in sys/fs/nfsclient/nfs_clvfsops.c
1481  * for the name "nfs" within the VFS_SET() macro that defines the "nfs"
1482  * file system type.
1483  */
1484 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1485 
1486 /* So that loader and kldload(2) can find us, wherever we are.. */
1487 MODULE_VERSION(nfscl, 1);
1488 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1489 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1490 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1491 MODULE_DEPEND(nfscl, xdr, 1, 1, 1);
1492