xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision f18976136625a7d016e97bfd9eabddf640b3e06d)
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 uma_zone_t ncl_pbuf_zone;
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 	vm_object_t object;
418 	u_quad_t nsize;
419 	int error, force_fid_err;
420 	bool setnsize;
421 
422 	error = 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 			} else if (np->n_flag & NMODIFIED) {
518 				/*
519 				 * We've modified the file: Use the larger
520 				 * of our size, and the server's size.
521 				 */
522 				if (vap->va_size < np->n_size) {
523 					vap->va_size = np->n_size;
524 				} else {
525 					np->n_size = vap->va_size;
526 					np->n_flag |= NSIZECHANGED;
527 				}
528 			} else {
529 				np->n_size = vap->va_size;
530 				np->n_flag |= NSIZECHANGED;
531 			}
532 		} else {
533 			np->n_size = vap->va_size;
534 		}
535 	}
536 	/*
537 	 * The following checks are added to prevent a race between (say)
538 	 * a READDIR+ and a WRITE.
539 	 * READDIR+, WRITE requests sent out.
540 	 * READDIR+ resp, WRITE resp received on client.
541 	 * However, the WRITE resp was handled before the READDIR+ resp
542 	 * causing the post op attrs from the write to be loaded first
543 	 * and the attrs from the READDIR+ to be loaded later. If this
544 	 * happens, we have stale attrs loaded into the attrcache.
545 	 * We detect this by for the mtime moving back. We invalidate the
546 	 * attrcache when this happens.
547 	 */
548 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
549 		/* Size changed or mtime went backwards */
550 		np->n_attrstamp = 0;
551 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
552 	}
553 	if (vaper != NULL) {
554 		NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
555 		if (np->n_flag & NCHG) {
556 			if (np->n_flag & NACC)
557 				vaper->va_atime = np->n_atim;
558 			if (np->n_flag & NUPD)
559 				vaper->va_mtime = np->n_mtim;
560 		}
561 	}
562 
563 out:
564 #ifdef KDTRACE_HOOKS
565 	if (np->n_attrstamp != 0)
566 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
567 #endif
568 	nsize = vap->va_size;
569 	object = vp->v_object;
570 	setnsize = false;
571 	if (object != NULL) {
572 		if (OFF_TO_IDX(nsize + PAGE_MASK) < object->size) {
573 			/*
574 			 * When shrinking the size, the call to
575 			 * vnode_pager_setsize() cannot be done with
576 			 * the mutex held, because we might need to
577 			 * wait for a busy page.  Delay it until after
578 			 * the node is unlocked.
579 			 */
580 			setnsize = true;
581 		} else {
582 			vnode_pager_setsize(vp, nsize);
583 		}
584 	}
585 	NFSUNLOCKNODE(np);
586 	if (setnsize)
587 		vnode_pager_setsize(vp, nsize);
588 	return (error);
589 }
590 
591 /*
592  * Fill in the client id name. For these bytes:
593  * 1 - they must be unique
594  * 2 - they should be persistent across client reboots
595  * 1 is more critical than 2
596  * Use the mount point's unique id plus either the uuid or, if that
597  * isn't set, random junk.
598  */
599 void
600 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
601 {
602 	int uuidlen;
603 
604 	/*
605 	 * First, put in the 64bit mount point identifier.
606 	 */
607 	if (idlen >= sizeof (u_int64_t)) {
608 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
609 		cp += sizeof (u_int64_t);
610 		idlen -= sizeof (u_int64_t);
611 	}
612 
613 	/*
614 	 * If uuid is non-zero length, use it.
615 	 */
616 	uuidlen = strlen(uuid);
617 	if (uuidlen > 0 && idlen >= uuidlen) {
618 		NFSBCOPY(uuid, cp, uuidlen);
619 		cp += uuidlen;
620 		idlen -= uuidlen;
621 	}
622 
623 	/*
624 	 * This only normally happens if the uuid isn't set.
625 	 */
626 	while (idlen > 0) {
627 		*cp++ = (u_int8_t)(arc4random() % 256);
628 		idlen--;
629 	}
630 }
631 
632 /*
633  * Fill in a lock owner name. For now, pid + the process's creation time.
634  */
635 void
636 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
637 {
638 	union {
639 		u_int32_t	lval;
640 		u_int8_t	cval[4];
641 	} tl;
642 	struct proc *p;
643 
644 	if (id == NULL) {
645 		/* Return the single open_owner of all 0 bytes. */
646 		bzero(cp, NFSV4CL_LOCKNAMELEN);
647 		return;
648 	}
649 	if ((flags & F_POSIX) != 0) {
650 		p = (struct proc *)id;
651 		tl.lval = p->p_pid;
652 		*cp++ = tl.cval[0];
653 		*cp++ = tl.cval[1];
654 		*cp++ = tl.cval[2];
655 		*cp++ = tl.cval[3];
656 		tl.lval = p->p_stats->p_start.tv_sec;
657 		*cp++ = tl.cval[0];
658 		*cp++ = tl.cval[1];
659 		*cp++ = tl.cval[2];
660 		*cp++ = tl.cval[3];
661 		tl.lval = p->p_stats->p_start.tv_usec;
662 		*cp++ = tl.cval[0];
663 		*cp++ = tl.cval[1];
664 		*cp++ = tl.cval[2];
665 		*cp = tl.cval[3];
666 	} else if ((flags & F_FLOCK) != 0) {
667 		bcopy(&id, cp, sizeof(id));
668 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
669 	} else {
670 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
671 		bzero(cp, NFSV4CL_LOCKNAMELEN);
672 	}
673 }
674 
675 /*
676  * Find the parent process for the thread passed in as an argument.
677  * If none exists, return NULL, otherwise return a thread for the parent.
678  * (Can be any of the threads, since it is only used for td->td_proc.)
679  */
680 NFSPROC_T *
681 nfscl_getparent(struct thread *td)
682 {
683 	struct proc *p;
684 	struct thread *ptd;
685 
686 	if (td == NULL)
687 		return (NULL);
688 	p = td->td_proc;
689 	if (p->p_pid == 0)
690 		return (NULL);
691 	p = p->p_pptr;
692 	if (p == NULL)
693 		return (NULL);
694 	ptd = TAILQ_FIRST(&p->p_threads);
695 	return (ptd);
696 }
697 
698 /*
699  * Start up the renew kernel thread.
700  */
701 static void
702 start_nfscl(void *arg)
703 {
704 	struct nfsclclient *clp;
705 	struct thread *td;
706 
707 	clp = (struct nfsclclient *)arg;
708 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
709 	nfscl_renewthread(clp, td);
710 	kproc_exit(0);
711 }
712 
713 void
714 nfscl_start_renewthread(struct nfsclclient *clp)
715 {
716 
717 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
718 	    "nfscl");
719 }
720 
721 /*
722  * Handle wcc_data.
723  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
724  * as the first Op after PutFH.
725  * (For NFSv4, the postop attributes are after the Op, so they can't be
726  *  parsed here. A separate call to nfscl_postop_attr() is required.)
727  */
728 int
729 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
730     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
731 {
732 	u_int32_t *tl;
733 	struct nfsnode *np = VTONFS(vp);
734 	struct nfsvattr nfsva;
735 	int error = 0;
736 
737 	if (wccflagp != NULL)
738 		*wccflagp = 0;
739 	if (nd->nd_flag & ND_NFSV3) {
740 		*flagp = 0;
741 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
742 		if (*tl == newnfs_true) {
743 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
744 			if (wccflagp != NULL) {
745 				NFSLOCKNODE(np);
746 				*wccflagp = (np->n_mtime.tv_sec ==
747 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
748 				    np->n_mtime.tv_nsec ==
749 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
750 				NFSUNLOCKNODE(np);
751 			}
752 		}
753 		error = nfscl_postop_attr(nd, nap, flagp, stuff);
754 		if (wccflagp != NULL && *flagp == 0)
755 			*wccflagp = 0;
756 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
757 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
758 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
759 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
760 		    NULL, NULL, NULL, NULL, NULL);
761 		if (error)
762 			return (error);
763 		/*
764 		 * Get rid of Op# and status for next op.
765 		 */
766 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
767 		if (*++tl)
768 			nd->nd_flag |= ND_NOMOREDATA;
769 		if (wccflagp != NULL &&
770 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
771 			NFSLOCKNODE(np);
772 			*wccflagp = (np->n_mtime.tv_sec ==
773 			    nfsva.na_vattr.va_mtime.tv_sec &&
774 			    np->n_mtime.tv_nsec ==
775 			    nfsva.na_vattr.va_mtime.tv_sec);
776 			NFSUNLOCKNODE(np);
777 		}
778 	}
779 nfsmout:
780 	return (error);
781 }
782 
783 /*
784  * Get postop attributes.
785  */
786 int
787 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
788     void *stuff)
789 {
790 	u_int32_t *tl;
791 	int error = 0;
792 
793 	*retp = 0;
794 	if (nd->nd_flag & ND_NOMOREDATA)
795 		return (error);
796 	if (nd->nd_flag & ND_NFSV3) {
797 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
798 		*retp = fxdr_unsigned(int, *tl);
799 	} else if (nd->nd_flag & ND_NFSV4) {
800 		/*
801 		 * For NFSv4, the postop attr are at the end, so no point
802 		 * in looking if nd_repstat != 0.
803 		 */
804 		if (!nd->nd_repstat) {
805 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
806 			if (*(tl + 1))
807 				/* should never happen since nd_repstat != 0 */
808 				nd->nd_flag |= ND_NOMOREDATA;
809 			else
810 				*retp = 1;
811 		}
812 	} else if (!nd->nd_repstat) {
813 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
814 		*retp = 1;
815 	}
816 	if (*retp) {
817 		error = nfsm_loadattr(nd, nap);
818 		if (error)
819 			*retp = 0;
820 	}
821 nfsmout:
822 	return (error);
823 }
824 
825 /*
826  * nfscl_request() - mostly a wrapper for newnfs_request().
827  */
828 int
829 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
830     struct ucred *cred, void *stuff)
831 {
832 	int ret, vers;
833 	struct nfsmount *nmp;
834 
835 	nmp = VFSTONFS(vp->v_mount);
836 	if (nd->nd_flag & ND_NFSV4)
837 		vers = NFS_VER4;
838 	else if (nd->nd_flag & ND_NFSV3)
839 		vers = NFS_VER3;
840 	else
841 		vers = NFS_VER2;
842 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
843 		NFS_PROG, vers, NULL, 1, NULL, NULL);
844 	return (ret);
845 }
846 
847 /*
848  * fill in this bsden's variant of statfs using nfsstatfs.
849  */
850 void
851 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
852 {
853 	struct statfs *sbp = (struct statfs *)statfs;
854 
855 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
856 		sbp->f_bsize = NFS_FABLKSIZE;
857 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
858 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
859 		/*
860 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
861 		 * the value after dividing by NFS_FABLKSIZE is small
862 		 * enough that it will fit in 63bits, so it is ok to
863 		 * assign it to f_bavail without fear that it will become
864 		 * negative.
865 		 */
866 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
867 		sbp->f_files = sfp->sf_tfiles;
868 		/* Since f_ffree is int64_t, clip it to 63bits. */
869 		if (sfp->sf_ffiles > INT64_MAX)
870 			sbp->f_ffree = INT64_MAX;
871 		else
872 			sbp->f_ffree = sfp->sf_ffiles;
873 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
874 		/*
875 		 * The type casts to (int32_t) ensure that this code is
876 		 * compatible with the old NFS client, in that it will
877 		 * propagate bit31 to the high order bits. This may or may
878 		 * not be correct for NFSv2, but since it is a legacy
879 		 * environment, I'd rather retain backwards compatibility.
880 		 */
881 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
882 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
883 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
884 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
885 		sbp->f_files = 0;
886 		sbp->f_ffree = 0;
887 	}
888 }
889 
890 /*
891  * Use the fsinfo stuff to update the mount point.
892  */
893 void
894 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
895 {
896 
897 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
898 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
899 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
900 		    ~(NFS_FABLKSIZE - 1);
901 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
902 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
903 		if (nmp->nm_wsize == 0)
904 			nmp->nm_wsize = fsp->fs_wtmax;
905 	}
906 	if (nmp->nm_wsize < NFS_FABLKSIZE)
907 		nmp->nm_wsize = NFS_FABLKSIZE;
908 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
909 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
910 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
911 		    ~(NFS_FABLKSIZE - 1);
912 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
913 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
914 		if (nmp->nm_rsize == 0)
915 			nmp->nm_rsize = fsp->fs_rtmax;
916 	}
917 	if (nmp->nm_rsize < NFS_FABLKSIZE)
918 		nmp->nm_rsize = NFS_FABLKSIZE;
919 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
920 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
921 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
922 		    ~(NFS_DIRBLKSIZ - 1);
923 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
924 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
925 		if (nmp->nm_readdirsize == 0)
926 			nmp->nm_readdirsize = fsp->fs_rtmax;
927 	}
928 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
929 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
930 	if (fsp->fs_maxfilesize > 0 &&
931 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
932 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
933 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
934 	nmp->nm_state |= NFSSTA_GOTFSINFO;
935 }
936 
937 /*
938  * Lookups source address which should be used to communicate with
939  * @nmp and stores it inside @pdst.
940  *
941  * Returns 0 on success.
942  */
943 u_int8_t *
944 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
945 {
946 #if defined(INET6) || defined(INET)
947 	int error, fibnum;
948 
949 	fibnum = curthread->td_proc->p_fibnum;
950 #endif
951 #ifdef INET
952 	if (nmp->nm_nam->sa_family == AF_INET) {
953 		struct sockaddr_in *sin;
954 		struct nhop4_extended nh_ext;
955 
956 		sin = (struct sockaddr_in *)nmp->nm_nam;
957 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
958 		error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
959 		    &nh_ext);
960 		CURVNET_RESTORE();
961 		if (error != 0)
962 			return (NULL);
963 
964 		if (IN_LOOPBACK(ntohl(nh_ext.nh_src.s_addr))) {
965 			/* Ignore loopback addresses */
966 			return (NULL);
967 		}
968 
969 		*isinet6p = 0;
970 		*((struct in_addr *)paddr) = nh_ext.nh_src;
971 
972 		return (u_int8_t *)paddr;
973 	}
974 #endif
975 #ifdef INET6
976 	if (nmp->nm_nam->sa_family == AF_INET6) {
977 		struct sockaddr_in6 *sin6;
978 
979 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
980 
981 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
982 		error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
983 		    sin6->sin6_scope_id, NULL, paddr, NULL);
984 		CURVNET_RESTORE();
985 		if (error != 0)
986 			return (NULL);
987 
988 		if (IN6_IS_ADDR_LOOPBACK(paddr))
989 			return (NULL);
990 
991 		/* Scope is embedded in */
992 		*isinet6p = 1;
993 
994 		return (u_int8_t *)paddr;
995 	}
996 #endif
997 	return (NULL);
998 }
999 
1000 /*
1001  * Copy NFS uid, gids from the cred structure.
1002  */
1003 void
1004 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1005 {
1006 	int i;
1007 
1008 	KASSERT(cr->cr_ngroups >= 0,
1009 	    ("newnfs_copyincred: negative cr_ngroups"));
1010 	nfscr->nfsc_uid = cr->cr_uid;
1011 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1012 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1013 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1014 }
1015 
1016 
1017 /*
1018  * Do any client specific initialization.
1019  */
1020 void
1021 nfscl_init(void)
1022 {
1023 	static int inited = 0;
1024 
1025 	if (inited)
1026 		return;
1027 	inited = 1;
1028 	nfscl_inited = 1;
1029 	ncl_pbuf_zone = pbuf_zsecond_create("nfspbuf", nswbuf / 2);
1030 }
1031 
1032 /*
1033  * Check each of the attributes to be set, to ensure they aren't already
1034  * the correct value. Disable setting ones already correct.
1035  */
1036 int
1037 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1038 {
1039 
1040 	if (vap->va_mode != (mode_t)VNOVAL) {
1041 		if (vap->va_mode == nvap->na_mode)
1042 			vap->va_mode = (mode_t)VNOVAL;
1043 	}
1044 	if (vap->va_uid != (uid_t)VNOVAL) {
1045 		if (vap->va_uid == nvap->na_uid)
1046 			vap->va_uid = (uid_t)VNOVAL;
1047 	}
1048 	if (vap->va_gid != (gid_t)VNOVAL) {
1049 		if (vap->va_gid == nvap->na_gid)
1050 			vap->va_gid = (gid_t)VNOVAL;
1051 	}
1052 	if (vap->va_size != VNOVAL) {
1053 		if (vap->va_size == nvap->na_size)
1054 			vap->va_size = VNOVAL;
1055 	}
1056 
1057 	/*
1058 	 * We are normally called with only a partially initialized
1059 	 * VAP.  Since the NFSv3 spec says that server may use the
1060 	 * file attributes to store the verifier, the spec requires
1061 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1062 	 * in atime, but we can't really assume that all servers will
1063 	 * so we ensure that our SETATTR sets both atime and mtime.
1064 	 * Set the VA_UTIMES_NULL flag for this case, so that
1065 	 * the server's time will be used.  This is needed to
1066 	 * work around a bug in some Solaris servers, where
1067 	 * setting the time TOCLIENT causes the Setattr RPC
1068 	 * to return NFS_OK, but not set va_mode.
1069 	 */
1070 	if (vap->va_mtime.tv_sec == VNOVAL) {
1071 		vfs_timestamp(&vap->va_mtime);
1072 		vap->va_vaflags |= VA_UTIMES_NULL;
1073 	}
1074 	if (vap->va_atime.tv_sec == VNOVAL)
1075 		vap->va_atime = vap->va_mtime;
1076 	return (1);
1077 }
1078 
1079 /*
1080  * Map nfsv4 errors to errno.h errors.
1081  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1082  * error should only be returned for the Open, Create and Setattr Ops.
1083  * As such, most calls can just pass in 0 for those arguments.
1084  */
1085 APPLESTATIC int
1086 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1087 {
1088 	struct proc *p;
1089 
1090 	if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1091 		return (error);
1092 	if (td != NULL)
1093 		p = td->td_proc;
1094 	else
1095 		p = NULL;
1096 	switch (error) {
1097 	case NFSERR_BADOWNER:
1098 		tprintf(p, LOG_INFO,
1099 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1100 		    uid, gid);
1101 		return (EPERM);
1102 	case NFSERR_BADNAME:
1103 	case NFSERR_BADCHAR:
1104 		printf("nfsv4 char/name not handled by server\n");
1105 		return (ENOENT);
1106 	case NFSERR_STALECLIENTID:
1107 	case NFSERR_STALESTATEID:
1108 	case NFSERR_EXPIRED:
1109 	case NFSERR_BADSTATEID:
1110 	case NFSERR_BADSESSION:
1111 		printf("nfsv4 recover err returned %d\n", error);
1112 		return (EIO);
1113 	case NFSERR_BADHANDLE:
1114 	case NFSERR_SERVERFAULT:
1115 	case NFSERR_BADTYPE:
1116 	case NFSERR_FHEXPIRED:
1117 	case NFSERR_RESOURCE:
1118 	case NFSERR_MOVED:
1119 	case NFSERR_NOFILEHANDLE:
1120 	case NFSERR_MINORVERMISMATCH:
1121 	case NFSERR_OLDSTATEID:
1122 	case NFSERR_BADSEQID:
1123 	case NFSERR_LEASEMOVED:
1124 	case NFSERR_RECLAIMBAD:
1125 	case NFSERR_BADXDR:
1126 	case NFSERR_OPILLEGAL:
1127 		printf("nfsv4 client/server protocol prob err=%d\n",
1128 		    error);
1129 		return (EIO);
1130 	default:
1131 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1132 		return (EIO);
1133 	};
1134 }
1135 
1136 /*
1137  * Check to see if the process for this owner exists. Return 1 if it doesn't
1138  * and 0 otherwise.
1139  */
1140 int
1141 nfscl_procdoesntexist(u_int8_t *own)
1142 {
1143 	union {
1144 		u_int32_t	lval;
1145 		u_int8_t	cval[4];
1146 	} tl;
1147 	struct proc *p;
1148 	pid_t pid;
1149 	int i, ret = 0;
1150 
1151 	/* For the single open_owner of all 0 bytes, just return 0. */
1152 	for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1153 		if (own[i] != 0)
1154 			break;
1155 	if (i == NFSV4CL_LOCKNAMELEN)
1156 		return (0);
1157 
1158 	tl.cval[0] = *own++;
1159 	tl.cval[1] = *own++;
1160 	tl.cval[2] = *own++;
1161 	tl.cval[3] = *own++;
1162 	pid = tl.lval;
1163 	p = pfind_any_locked(pid);
1164 	if (p == NULL)
1165 		return (1);
1166 	if (p->p_stats == NULL) {
1167 		PROC_UNLOCK(p);
1168 		return (0);
1169 	}
1170 	tl.cval[0] = *own++;
1171 	tl.cval[1] = *own++;
1172 	tl.cval[2] = *own++;
1173 	tl.cval[3] = *own++;
1174 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1175 		ret = 1;
1176 	} else {
1177 		tl.cval[0] = *own++;
1178 		tl.cval[1] = *own++;
1179 		tl.cval[2] = *own++;
1180 		tl.cval[3] = *own;
1181 		if (tl.lval != p->p_stats->p_start.tv_usec)
1182 			ret = 1;
1183 	}
1184 	PROC_UNLOCK(p);
1185 	return (ret);
1186 }
1187 
1188 /*
1189  * - nfs pseudo system call for the client
1190  */
1191 /*
1192  * MPSAFE
1193  */
1194 static int
1195 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1196 {
1197 	struct file *fp;
1198 	struct nfscbd_args nfscbdarg;
1199 	struct nfsd_nfscbd_args nfscbdarg2;
1200 	struct nameidata nd;
1201 	struct nfscl_dumpmntopts dumpmntopts;
1202 	cap_rights_t rights;
1203 	char *buf;
1204 	int error;
1205 	struct mount *mp;
1206 	struct nfsmount *nmp;
1207 
1208 	if (uap->flag & NFSSVC_CBADDSOCK) {
1209 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1210 		if (error)
1211 			return (error);
1212 		/*
1213 		 * Since we don't know what rights might be required,
1214 		 * pretend that we need them all. It is better to be too
1215 		 * careful than too reckless.
1216 		 */
1217 		error = fget(td, nfscbdarg.sock,
1218 		    cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1219 		if (error)
1220 			return (error);
1221 		if (fp->f_type != DTYPE_SOCKET) {
1222 			fdrop(fp, td);
1223 			return (EPERM);
1224 		}
1225 		error = nfscbd_addsock(fp);
1226 		fdrop(fp, td);
1227 		if (!error && nfscl_enablecallb == 0) {
1228 			nfsv4_cbport = nfscbdarg.port;
1229 			nfscl_enablecallb = 1;
1230 		}
1231 	} else if (uap->flag & NFSSVC_NFSCBD) {
1232 		if (uap->argp == NULL)
1233 			return (EINVAL);
1234 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1235 		    sizeof(nfscbdarg2));
1236 		if (error)
1237 			return (error);
1238 		error = nfscbd_nfsd(td, &nfscbdarg2);
1239 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1240 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1241 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1242 		    dumpmntopts.ndmnt_blen > 1024))
1243 			error = EINVAL;
1244 		if (error == 0)
1245 			error = nfsrv_lookupfilename(&nd,
1246 			    dumpmntopts.ndmnt_fname, td);
1247 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1248 		    "nfs") != 0) {
1249 			vput(nd.ni_vp);
1250 			error = EINVAL;
1251 		}
1252 		if (error == 0) {
1253 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1254 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1255 			    dumpmntopts.ndmnt_blen);
1256 			vput(nd.ni_vp);
1257 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1258 			    dumpmntopts.ndmnt_blen);
1259 			free(buf, M_TEMP);
1260 		}
1261 	} else if (uap->flag & NFSSVC_FORCEDISM) {
1262 		buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
1263 		error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
1264 		if (error == 0) {
1265 			nmp = NULL;
1266 			mtx_lock(&mountlist_mtx);
1267 			TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1268 				if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
1269 				    0 && strcmp(mp->mnt_stat.f_fstypename,
1270 				    "nfs") == 0 && mp->mnt_data != NULL) {
1271 					nmp = VFSTONFS(mp);
1272 					NFSDDSLOCK();
1273 					if (nfsv4_findmirror(nmp) != NULL) {
1274 						NFSDDSUNLOCK();
1275 						error = ENXIO;
1276 						nmp = NULL;
1277 						break;
1278 					}
1279 					mtx_lock(&nmp->nm_mtx);
1280 					if ((nmp->nm_privflag &
1281 					    NFSMNTP_FORCEDISM) == 0) {
1282 						nmp->nm_privflag |=
1283 						   (NFSMNTP_FORCEDISM |
1284 						    NFSMNTP_CANCELRPCS);
1285 						mtx_unlock(&nmp->nm_mtx);
1286 					} else {
1287 						mtx_unlock(&nmp->nm_mtx);
1288 						nmp = NULL;
1289 					}
1290 					NFSDDSUNLOCK();
1291 					break;
1292 				}
1293 			}
1294 			mtx_unlock(&mountlist_mtx);
1295 
1296 			if (nmp != NULL) {
1297 				/*
1298 				 * Call newnfs_nmcancelreqs() to cause
1299 				 * any RPCs in progress on the mount point to
1300 				 * fail.
1301 				 * This will cause any process waiting for an
1302 				 * RPC to complete while holding a vnode lock
1303 				 * on the mounted-on vnode (such as "df" or
1304 				 * a non-forced "umount") to fail.
1305 				 * This will unlock the mounted-on vnode so
1306 				 * a forced dismount can succeed.
1307 				 * Then clear NFSMNTP_CANCELRPCS and wakeup(),
1308 				 * so that nfs_unmount() can complete.
1309 				 */
1310 				newnfs_nmcancelreqs(nmp);
1311 				mtx_lock(&nmp->nm_mtx);
1312 				nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
1313 				wakeup(nmp);
1314 				mtx_unlock(&nmp->nm_mtx);
1315 			} else if (error == 0)
1316 				error = EINVAL;
1317 		}
1318 		free(buf, M_TEMP);
1319 	} else {
1320 		error = EINVAL;
1321 	}
1322 	return (error);
1323 }
1324 
1325 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1326 
1327 /*
1328  * Called once to initialize data structures...
1329  */
1330 static int
1331 nfscl_modevent(module_t mod, int type, void *data)
1332 {
1333 	int error = 0;
1334 	static int loaded = 0;
1335 
1336 	switch (type) {
1337 	case MOD_LOAD:
1338 		if (loaded)
1339 			return (0);
1340 		newnfs_portinit();
1341 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1342 		nfscl_init();
1343 		NFSD_LOCK();
1344 		nfsrvd_cbinit(0);
1345 		NFSD_UNLOCK();
1346 		ncl_call_invalcaches = ncl_invalcaches;
1347 		nfsd_call_nfscl = nfssvc_nfscl;
1348 		loaded = 1;
1349 		break;
1350 
1351 	case MOD_UNLOAD:
1352 		if (nfs_numnfscbd != 0) {
1353 			error = EBUSY;
1354 			break;
1355 		}
1356 
1357 		/*
1358 		 * XXX: Unloading of nfscl module is unsupported.
1359 		 */
1360 #if 0
1361 		ncl_call_invalcaches = NULL;
1362 		nfsd_call_nfscl = NULL;
1363 		uma_zdestroy(ncl_pbuf_zone);
1364 		/* and get rid of the mutexes */
1365 		mtx_destroy(&ncl_iod_mutex);
1366 		loaded = 0;
1367 		break;
1368 #else
1369 		/* FALLTHROUGH */
1370 #endif
1371 	default:
1372 		error = EOPNOTSUPP;
1373 		break;
1374 	}
1375 	return error;
1376 }
1377 static moduledata_t nfscl_mod = {
1378 	"nfscl",
1379 	nfscl_modevent,
1380 	NULL,
1381 };
1382 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1383 
1384 /* So that loader and kldload(2) can find us, wherever we are.. */
1385 MODULE_VERSION(nfscl, 1);
1386 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1387 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1388 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1389 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1390 
1391