xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision af23369a6deaaeb612ab266eb88b8bb8d560c322)
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 #include <net/route/nhop.h>
57 
58 #include <fs/nfsclient/nfs_kdtrace.h>
59 
60 #ifdef KDTRACE_HOOKS
61 dtrace_nfsclient_attrcache_flush_probe_func_t
62 		dtrace_nfscl_attrcache_flush_done_probe;
63 uint32_t	nfscl_attrcache_flush_done_id;
64 
65 dtrace_nfsclient_attrcache_get_hit_probe_func_t
66 		dtrace_nfscl_attrcache_get_hit_probe;
67 uint32_t	nfscl_attrcache_get_hit_id;
68 
69 dtrace_nfsclient_attrcache_get_miss_probe_func_t
70 		dtrace_nfscl_attrcache_get_miss_probe;
71 uint32_t	nfscl_attrcache_get_miss_id;
72 
73 dtrace_nfsclient_attrcache_load_probe_func_t
74 		dtrace_nfscl_attrcache_load_done_probe;
75 uint32_t	nfscl_attrcache_load_done_id;
76 #endif /* !KDTRACE_HOOKS */
77 
78 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
79 extern struct vop_vector newnfs_vnodeops;
80 extern struct vop_vector newnfs_fifoops;
81 extern uma_zone_t newnfsnode_zone;
82 extern struct buf_ops buf_ops_newnfs;
83 extern uma_zone_t ncl_pbuf_zone;
84 extern short nfsv4_cbport;
85 extern int nfscl_enablecallb;
86 extern int nfs_numnfscbd;
87 extern int nfscl_inited;
88 struct mtx ncl_iod_mutex;
89 NFSDLOCKMUTEX;
90 extern struct mtx nfsrv_dslock_mtx;
91 
92 extern void (*ncl_call_invalcaches)(struct vnode *);
93 
94 SYSCTL_DECL(_vfs_nfs);
95 static int ncl_fileid_maxwarnings = 10;
96 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
97     &ncl_fileid_maxwarnings, 0,
98     "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
99 static volatile int ncl_fileid_nwarnings;
100 
101 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
102     struct nfsvattr *);
103 
104 /*
105  * Comparison function for vfs_hash functions.
106  */
107 int
108 newnfs_vncmpf(struct vnode *vp, void *arg)
109 {
110 	struct nfsfh *nfhp = (struct nfsfh *)arg;
111 	struct nfsnode *np = VTONFS(vp);
112 
113 	if (np->n_fhp->nfh_len != nfhp->nfh_len ||
114 	    NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
115 		return (1);
116 	return (0);
117 }
118 
119 /*
120  * Look up a vnode/nfsnode by file handle.
121  * Callers must check for mount points!!
122  * In all cases, a pointer to a
123  * nfsnode structure is returned.
124  * This variant takes a "struct nfsfh *" as second argument and uses
125  * that structure up, either by hanging off the nfsnode or FREEing it.
126  */
127 int
128 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
129     struct componentname *cnp, struct thread *td, struct nfsnode **npp,
130     int lkflags)
131 {
132 	struct nfsnode *np, *dnp;
133 	struct vnode *vp, *nvp;
134 	struct nfsv4node *newd, *oldd;
135 	int error;
136 	u_int hash;
137 	struct nfsmount *nmp;
138 
139 	nmp = VFSTONFS(mntp);
140 	dnp = VTONFS(dvp);
141 	*npp = NULL;
142 
143 	hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
144 
145 	error = vfs_hash_get(mntp, hash, lkflags,
146 	    td, &nvp, newnfs_vncmpf, nfhp);
147 	if (error == 0 && nvp != NULL) {
148 		/*
149 		 * I believe there is a slight chance that vgonel() could
150 		 * get called on this vnode between when NFSVOPLOCK() drops
151 		 * the VI_LOCK() and vget() acquires it again, so that it
152 		 * hasn't yet had v_usecount incremented. If this were to
153 		 * happen, the VIRF_DOOMED flag would be set, so check for
154 		 * that here. Since we now have the v_usecount incremented,
155 		 * we should be ok until we vrele() it, if the VIRF_DOOMED
156 		 * flag isn't set now.
157 		 */
158 		VI_LOCK(nvp);
159 		if (VN_IS_DOOMED(nvp)) {
160 			VI_UNLOCK(nvp);
161 			vrele(nvp);
162 			error = ENOENT;
163 		} else {
164 			VI_UNLOCK(nvp);
165 		}
166 	}
167 	if (error) {
168 		free(nfhp, M_NFSFH);
169 		return (error);
170 	}
171 	if (nvp != NULL) {
172 		np = VTONFS(nvp);
173 		/*
174 		 * For NFSv4, check to see if it is the same name and
175 		 * replace the name, if it is different.
176 		 */
177 		oldd = newd = NULL;
178 		if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
179 		    nvp->v_type == VREG &&
180 		    (np->n_v4->n4_namelen != cnp->cn_namelen ||
181 		     NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
182 		     cnp->cn_namelen) ||
183 		     dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
184 		     NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
185 		     dnp->n_fhp->nfh_len))) {
186 		    newd = malloc(
187 			sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
188 			+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
189 		    NFSLOCKNODE(np);
190 		    if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
191 			&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
192 			 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
193 			 cnp->cn_namelen) ||
194 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
195 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
196 			 dnp->n_fhp->nfh_len))) {
197 			oldd = np->n_v4;
198 			np->n_v4 = newd;
199 			newd = NULL;
200 			np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
201 			np->n_v4->n4_namelen = cnp->cn_namelen;
202 			NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
203 			    dnp->n_fhp->nfh_len);
204 			NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
205 			    cnp->cn_namelen);
206 		    }
207 		    NFSUNLOCKNODE(np);
208 		}
209 		if (newd != NULL)
210 			free(newd, M_NFSV4NODE);
211 		if (oldd != NULL)
212 			free(oldd, M_NFSV4NODE);
213 		*npp = np;
214 		free(nfhp, M_NFSFH);
215 		return (0);
216 	}
217 	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
218 
219 	error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
220 	if (error) {
221 		uma_zfree(newnfsnode_zone, np);
222 		free(nfhp, M_NFSFH);
223 		return (error);
224 	}
225 	vp = nvp;
226 	KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
227 	vp->v_bufobj.bo_ops = &buf_ops_newnfs;
228 	vp->v_data = np;
229 	np->n_vnode = vp;
230 	/*
231 	 * Initialize the mutex even if the vnode is going to be a loser.
232 	 * This simplifies the logic in reclaim, which can then unconditionally
233 	 * destroy the mutex (in the case of the loser, or if hash_insert
234 	 * happened to return an error no special casing is needed).
235 	 */
236 	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
237 	lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE |
238 	    LK_CANRECURSE);
239 
240 	/*
241 	 * Are we getting the root? If so, make sure the vnode flags
242 	 * are correct
243 	 */
244 	if ((nfhp->nfh_len == nmp->nm_fhsize) &&
245 	    !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
246 		if (vp->v_type == VNON)
247 			vp->v_type = VDIR;
248 		vp->v_vflag |= VV_ROOT;
249 	}
250 
251 	vp->v_vflag |= VV_VMSIZEVNLOCK;
252 
253 	np->n_fhp = nfhp;
254 	/*
255 	 * For NFSv4, we have to attach the directory file handle and
256 	 * file name, so that Open Ops can be done later.
257 	 */
258 	if (nmp->nm_flag & NFSMNT_NFSV4) {
259 		np->n_v4 = malloc(sizeof (struct nfsv4node)
260 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
261 		    M_WAITOK);
262 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
263 		np->n_v4->n4_namelen = cnp->cn_namelen;
264 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
265 		    dnp->n_fhp->nfh_len);
266 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
267 		    cnp->cn_namelen);
268 	} else {
269 		np->n_v4 = NULL;
270 	}
271 
272 	/*
273 	 * NFS supports recursive and shared locking.
274 	 */
275 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
276 	VN_LOCK_AREC(vp);
277 	VN_LOCK_ASHARE(vp);
278 	error = insmntque(vp, mntp);
279 	if (error != 0) {
280 		*npp = NULL;
281 		mtx_destroy(&np->n_mtx);
282 		lockdestroy(&np->n_excl);
283 		free(nfhp, M_NFSFH);
284 		if (np->n_v4 != NULL)
285 			free(np->n_v4, M_NFSV4NODE);
286 		uma_zfree(newnfsnode_zone, np);
287 		return (error);
288 	}
289 	vn_set_state(vp, VSTATE_CONSTRUCTED);
290 	error = vfs_hash_insert(vp, hash, lkflags,
291 	    td, &nvp, newnfs_vncmpf, nfhp);
292 	if (error)
293 		return (error);
294 	if (nvp != NULL) {
295 		*npp = VTONFS(nvp);
296 		/* vfs_hash_insert() vput()'s the losing vnode */
297 		return (0);
298 	}
299 	*npp = np;
300 
301 	return (0);
302 }
303 
304 /*
305  * Another variant of nfs_nget(). This one is only used by reopen. It
306  * takes almost the same args as nfs_nget(), but only succeeds if an entry
307  * exists in the cache. (Since files should already be "open" with a
308  * vnode ref cnt on the node when reopen calls this, it should always
309  * succeed.)
310  * Also, don't get a vnode lock, since it may already be locked by some
311  * other process that is handling it. This is ok, since all other threads
312  * on the client are blocked by the nfsc_lock being exclusively held by the
313  * caller of this function.
314  */
315 int
316 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
317     struct thread *td, struct nfsnode **npp)
318 {
319 	struct vnode *nvp;
320 	u_int hash;
321 	struct nfsfh *nfhp;
322 	int error;
323 
324 	*npp = NULL;
325 	/* For forced dismounts, just return error. */
326 	if (NFSCL_FORCEDISM(mntp))
327 		return (EINTR);
328 	nfhp = malloc(sizeof (struct nfsfh) + fhsize,
329 	    M_NFSFH, M_WAITOK);
330 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
331 	nfhp->nfh_len = fhsize;
332 
333 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
334 
335 	/*
336 	 * First, try to get the vnode locked, but don't block for the lock.
337 	 */
338 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
339 	    newnfs_vncmpf, nfhp);
340 	if (error == 0 && nvp != NULL) {
341 		NFSVOPUNLOCK(nvp);
342 	} else if (error == EBUSY) {
343 		/*
344 		 * It is safe so long as a vflush() with
345 		 * FORCECLOSE has not been done. Since the Renew thread is
346 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
347 		 * a vflush() with FORCECLOSE, we should be ok here.
348 		 */
349 		if (NFSCL_FORCEDISM(mntp))
350 			error = EINTR;
351 		else {
352 			vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
353 			if (nvp == NULL) {
354 				error = ENOENT;
355 			} else if (VN_IS_DOOMED(nvp)) {
356 				error = ENOENT;
357 				vrele(nvp);
358 			} else {
359 				error = 0;
360 			}
361 		}
362 	}
363 	free(nfhp, M_NFSFH);
364 	if (error)
365 		return (error);
366 	if (nvp != NULL) {
367 		*npp = VTONFS(nvp);
368 		return (0);
369 	}
370 	return (EINVAL);
371 }
372 
373 static void
374 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
375     struct nfsvattr *newnap)
376 {
377 	int off;
378 
379 	if (ncl_fileid_maxwarnings >= 0 &&
380 	    ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
381 		return;
382 	off = 0;
383 	if (ncl_fileid_maxwarnings >= 0) {
384 		if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
385 			off = 1;
386 	}
387 
388 	printf("newnfs: server '%s' error: fileid changed. "
389 	    "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
390 	    "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
391 	    nmp->nm_com.nmcom_hostname,
392 	    (uintmax_t)nmp->nm_fsid[0],
393 	    (uintmax_t)nmp->nm_fsid[1],
394 	    (uintmax_t)oldnap->na_fileid,
395 	    (uintmax_t)newnap->na_fileid);
396 
397 	if (off)
398 		printf("newnfs: Logged %d times about fileid corruption; "
399 		    "going quiet to avoid spamming logs excessively. (Limit "
400 		    "is: %d).\n", ncl_fileid_nwarnings,
401 		    ncl_fileid_maxwarnings);
402 }
403 
404 void
405 ncl_copy_vattr(struct vattr *dst, struct vattr *src)
406 {
407 	dst->va_type = src->va_type;
408 	dst->va_mode = src->va_mode;
409 	dst->va_nlink = src->va_nlink;
410 	dst->va_uid = src->va_uid;
411 	dst->va_gid = src->va_gid;
412 	dst->va_fsid = src->va_fsid;
413 	dst->va_fileid = src->va_fileid;
414 	dst->va_size = src->va_size;
415 	dst->va_blocksize = src->va_blocksize;
416 	dst->va_atime = src->va_atime;
417 	dst->va_mtime = src->va_mtime;
418 	dst->va_ctime = src->va_ctime;
419 	dst->va_birthtime = src->va_birthtime;
420 	dst->va_gen = src->va_gen;
421 	dst->va_flags = src->va_flags;
422 	dst->va_rdev = src->va_rdev;
423 	dst->va_bytes = src->va_bytes;
424 	dst->va_filerev = src->va_filerev;
425 }
426 
427 /*
428  * Load the attribute cache (that lives in the nfsnode entry) with
429  * the attributes of the second argument and
430  * Iff vaper not NULL
431  *    copy the attributes to *vaper
432  * Similar to nfs_loadattrcache(), except the attributes are passed in
433  * instead of being parsed out of the mbuf list.
434  */
435 int
436 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
437     int writeattr, int dontshrink)
438 {
439 	struct vnode *vp = *vpp;
440 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
441 	struct nfsnode *np;
442 	struct nfsmount *nmp;
443 	struct timespec mtime_save;
444 	int error, force_fid_err;
445 	dev_t topfsid;
446 
447 	error = 0;
448 
449 	/*
450 	 * If v_type == VNON it is a new node, so fill in the v_type,
451 	 * n_mtime fields. Check to see if it represents a special
452 	 * device, and if so, check for a possible alias. Once the
453 	 * correct vnode has been obtained, fill in the rest of the
454 	 * information.
455 	 */
456 	np = VTONFS(vp);
457 	NFSLOCKNODE(np);
458 	if (vp->v_type != nvap->va_type) {
459 		vp->v_type = nvap->va_type;
460 		if (vp->v_type == VFIFO)
461 			vp->v_op = &newnfs_fifoops;
462 		np->n_mtime = nvap->va_mtime;
463 	}
464 	nmp = VFSTONFS(vp->v_mount);
465 	vap = &np->n_vattr.na_vattr;
466 	mtime_save = vap->va_mtime;
467 	if (writeattr) {
468 		np->n_vattr.na_filerev = nap->na_filerev;
469 		np->n_vattr.na_size = nap->na_size;
470 		np->n_vattr.na_mtime = nap->na_mtime;
471 		np->n_vattr.na_ctime = nap->na_ctime;
472 		np->n_vattr.na_btime = nap->na_btime;
473 		np->n_vattr.na_fsid = nap->na_fsid;
474 		np->n_vattr.na_mode = nap->na_mode;
475 	} else {
476 		force_fid_err = 0;
477 		KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
478 		    force_fid_err);
479 		/*
480 		 * BROKEN NFS SERVER OR MIDDLEWARE
481 		 *
482 		 * Certain NFS servers (certain old proprietary filers ca.
483 		 * 2006) or broken middleboxes (e.g. WAN accelerator products)
484 		 * will respond to GETATTR requests with results for a
485 		 * different fileid.
486 		 *
487 		 * The WAN accelerator we've observed not only serves stale
488 		 * cache results for a given file, it also occasionally serves
489 		 * results for wholly different files.  This causes surprising
490 		 * problems; for example the cached size attribute of a file
491 		 * may truncate down and then back up, resulting in zero
492 		 * regions in file contents read by applications.  We observed
493 		 * this reliably with Clang and .c files during parallel build.
494 		 * A pcap revealed packet fragmentation and GETATTR RPC
495 		 * responses with wholly wrong fileids.
496 		 */
497 		if ((np->n_vattr.na_fileid != 0 &&
498 		     np->n_vattr.na_fileid != nap->na_fileid) ||
499 		    force_fid_err) {
500 			nfscl_warn_fileid(nmp, &np->n_vattr, nap);
501 			error = EIDRM;
502 			goto out;
503 		}
504 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
505 		    sizeof (struct nfsvattr));
506 	}
507 
508 	/*
509 	 * For NFSv4, the server's export may be a tree of file systems
510 	 * where a fileno is a unique value within each file system.
511 	 * na_filesid[0,1] uniquely identify the server file system
512 	 * and nm_fsid[0,1] is the value for the root file system mounted.
513 	 * As such, the value of va_fsid generated by vn_fsid() represents
514 	 * the root file system on the server and a different value for
515 	 * va_fsid is needed for the other server file systems.  This
516 	 * va_fsid is ideally unique for all of the server file systems,
517 	 * so a 64bit hash on na_filesid[0,1] is calculated.
518 	 * Although highly unlikely that the fnv_64_hash() will be
519 	 * the same as the root, test for this case and recalculate the hash.
520 	 */
521 	vn_fsid(vp, vap);
522 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
523 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
524 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
525 		topfsid = vap->va_fsid;
526 		vap->va_fsid = FNV1_64_INIT;
527 		do {
528 			vap->va_fsid = fnv_64_buf(np->n_vattr.na_filesid,
529 			    sizeof(np->n_vattr.na_filesid), vap->va_fsid);
530 		} while (vap->va_fsid == topfsid);
531 	}
532 
533 	np->n_attrstamp = time_second;
534 	if (vap->va_size != np->n_size) {
535 		if (vap->va_type == VREG) {
536 			if (dontshrink && vap->va_size < np->n_size) {
537 				/*
538 				 * We've been told not to shrink the file;
539 				 * zero np->n_attrstamp to indicate that
540 				 * the attributes are stale.
541 				 */
542 				vap->va_size = np->n_size;
543 				np->n_attrstamp = 0;
544 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
545 			} else if (np->n_flag & NMODIFIED) {
546 				/*
547 				 * We've modified the file: Use the larger
548 				 * of our size, and the server's size.
549 				 */
550 				if (vap->va_size < np->n_size) {
551 					vap->va_size = np->n_size;
552 				} else {
553 					np->n_size = vap->va_size;
554 					np->n_flag |= NSIZECHANGED;
555 				}
556 			} else {
557 				np->n_size = vap->va_size;
558 				np->n_flag |= NSIZECHANGED;
559 			}
560 		} else {
561 			np->n_size = vap->va_size;
562 		}
563 	}
564 	/*
565 	 * The following checks are added to prevent a race between (say)
566 	 * a READDIR+ and a WRITE.
567 	 * READDIR+, WRITE requests sent out.
568 	 * READDIR+ resp, WRITE resp received on client.
569 	 * However, the WRITE resp was handled before the READDIR+ resp
570 	 * causing the post op attrs from the write to be loaded first
571 	 * and the attrs from the READDIR+ to be loaded later. If this
572 	 * happens, we have stale attrs loaded into the attrcache.
573 	 * We detect this by for the mtime moving back. We invalidate the
574 	 * attrcache when this happens.
575 	 */
576 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
577 		/* Size changed or mtime went backwards */
578 		np->n_attrstamp = 0;
579 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
580 	}
581 	if (vaper != NULL) {
582 		ncl_copy_vattr(vaper, vap);
583 		if (np->n_flag & NCHG) {
584 			if (np->n_flag & NACC)
585 				vaper->va_atime = np->n_atim;
586 			if (np->n_flag & NUPD)
587 				vaper->va_mtime = np->n_mtim;
588 		}
589 	}
590 
591 out:
592 #ifdef KDTRACE_HOOKS
593 	if (np->n_attrstamp != 0)
594 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
595 #endif
596 	(void)ncl_pager_setsize(vp, NULL);
597 	return (error);
598 }
599 
600 /*
601  * Call vnode_pager_setsize() if the size of the node changed, as
602  * recorded in nfsnode vs. v_object, or delay the call if notifying
603  * the pager is not possible at the moment.
604  *
605  * If nsizep is non-NULL, the call is delayed and the new node size is
606  * provided.  Caller should itself call vnode_pager_setsize() if
607  * function returned true.  If nsizep is NULL, function tries to call
608  * vnode_pager_setsize() itself if needed and possible, and the nfs
609  * node is unlocked unconditionally, the return value is not useful.
610  */
611 bool
612 ncl_pager_setsize(struct vnode *vp, u_quad_t *nsizep)
613 {
614 	struct nfsnode *np;
615 	vm_object_t object;
616 	struct vattr *vap;
617 	u_quad_t nsize;
618 	bool setnsize;
619 
620 	np = VTONFS(vp);
621 	NFSASSERTNODE(np);
622 
623 	vap = &np->n_vattr.na_vattr;
624 	nsize = vap->va_size;
625 	object = vp->v_object;
626 	setnsize = false;
627 
628 	if (object != NULL && nsize != object->un_pager.vnp.vnp_size) {
629 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
630 		    (curthread->td_pflags2 & TDP2_SBPAGES) == 0)
631 			setnsize = true;
632 		else
633 			np->n_flag |= NVNSETSZSKIP;
634 	}
635 	if (nsizep == NULL) {
636 		NFSUNLOCKNODE(np);
637 		if (setnsize)
638 			vnode_pager_setsize(vp, nsize);
639 		setnsize = false;
640 	} else {
641 		*nsizep = nsize;
642 	}
643 	return (setnsize);
644 }
645 
646 /*
647  * Fill in the client id name. For these bytes:
648  * 1 - they must be unique
649  * 2 - they should be persistent across client reboots
650  * 1 is more critical than 2
651  * Use the mount point's unique id plus either the uuid or, if that
652  * isn't set, random junk.
653  */
654 void
655 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
656 {
657 	int uuidlen;
658 
659 	/*
660 	 * First, put in the 64bit mount point identifier.
661 	 */
662 	if (idlen >= sizeof (u_int64_t)) {
663 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
664 		cp += sizeof (u_int64_t);
665 		idlen -= sizeof (u_int64_t);
666 	}
667 
668 	/*
669 	 * If uuid is non-zero length, use it.
670 	 */
671 	uuidlen = strlen(uuid);
672 	if (uuidlen > 0 && idlen >= uuidlen) {
673 		NFSBCOPY(uuid, cp, uuidlen);
674 		cp += uuidlen;
675 		idlen -= uuidlen;
676 	}
677 
678 	/*
679 	 * This only normally happens if the uuid isn't set.
680 	 */
681 	while (idlen > 0) {
682 		*cp++ = (u_int8_t)(arc4random() % 256);
683 		idlen--;
684 	}
685 }
686 
687 /*
688  * Fill in a lock owner name. For now, pid + the process's creation time.
689  */
690 void
691 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
692 {
693 	union {
694 		u_int32_t	lval;
695 		u_int8_t	cval[4];
696 	} tl;
697 	struct proc *p;
698 
699 	if (id == NULL) {
700 		/* Return the single open_owner of all 0 bytes. */
701 		bzero(cp, NFSV4CL_LOCKNAMELEN);
702 		return;
703 	}
704 	if ((flags & F_POSIX) != 0) {
705 		p = (struct proc *)id;
706 		tl.lval = p->p_pid;
707 		*cp++ = tl.cval[0];
708 		*cp++ = tl.cval[1];
709 		*cp++ = tl.cval[2];
710 		*cp++ = tl.cval[3];
711 		tl.lval = p->p_stats->p_start.tv_sec;
712 		*cp++ = tl.cval[0];
713 		*cp++ = tl.cval[1];
714 		*cp++ = tl.cval[2];
715 		*cp++ = tl.cval[3];
716 		tl.lval = p->p_stats->p_start.tv_usec;
717 		*cp++ = tl.cval[0];
718 		*cp++ = tl.cval[1];
719 		*cp++ = tl.cval[2];
720 		*cp = tl.cval[3];
721 	} else if ((flags & F_FLOCK) != 0) {
722 		bcopy(&id, cp, sizeof(id));
723 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
724 	} else {
725 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
726 		bzero(cp, NFSV4CL_LOCKNAMELEN);
727 	}
728 }
729 
730 /*
731  * Find the parent process for the thread passed in as an argument.
732  * If none exists, return NULL, otherwise return a thread for the parent.
733  * (Can be any of the threads, since it is only used for td->td_proc.)
734  */
735 NFSPROC_T *
736 nfscl_getparent(struct thread *td)
737 {
738 	struct proc *p;
739 	struct thread *ptd;
740 
741 	if (td == NULL)
742 		return (NULL);
743 	p = td->td_proc;
744 	if (p->p_pid == 0)
745 		return (NULL);
746 	p = p->p_pptr;
747 	if (p == NULL)
748 		return (NULL);
749 	ptd = TAILQ_FIRST(&p->p_threads);
750 	return (ptd);
751 }
752 
753 /*
754  * Start up the renew kernel thread.
755  */
756 static void
757 start_nfscl(void *arg)
758 {
759 	struct nfsclclient *clp;
760 	struct thread *td;
761 
762 	clp = (struct nfsclclient *)arg;
763 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
764 	nfscl_renewthread(clp, td);
765 	kproc_exit(0);
766 }
767 
768 void
769 nfscl_start_renewthread(struct nfsclclient *clp)
770 {
771 
772 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
773 	    "nfscl");
774 }
775 
776 /*
777  * Handle wcc_data.
778  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
779  * as the first Op after PutFH.
780  * (For NFSv4, the postop attributes are after the Op, so they can't be
781  *  parsed here. A separate call to nfscl_postop_attr() is required.)
782  */
783 int
784 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
785     struct nfsvattr *nap, int *flagp, int *wccflagp, uint64_t *repsizep)
786 {
787 	u_int32_t *tl;
788 	struct nfsnode *np = VTONFS(vp);
789 	struct nfsvattr nfsva;
790 	int error = 0;
791 
792 	if (wccflagp != NULL)
793 		*wccflagp = 0;
794 	if (nd->nd_flag & ND_NFSV3) {
795 		*flagp = 0;
796 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
797 		if (*tl == newnfs_true) {
798 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
799 			if (wccflagp != NULL) {
800 				NFSLOCKNODE(np);
801 				*wccflagp = (np->n_mtime.tv_sec ==
802 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
803 				    np->n_mtime.tv_nsec ==
804 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
805 				NFSUNLOCKNODE(np);
806 			}
807 		}
808 		error = nfscl_postop_attr(nd, nap, flagp);
809 		if (wccflagp != NULL && *flagp == 0)
810 			*wccflagp = 0;
811 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
812 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
813 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
814 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
815 		    NULL, NULL, NULL, NULL, NULL);
816 		if (error)
817 			return (error);
818 		/*
819 		 * Get rid of Op# and status for next op.
820 		 */
821 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
822 		if (*++tl)
823 			nd->nd_flag |= ND_NOMOREDATA;
824 		if (repsizep != NULL)
825 			*repsizep = nfsva.na_size;
826 		if (wccflagp != NULL &&
827 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
828 			NFSLOCKNODE(np);
829 			*wccflagp = (np->n_mtime.tv_sec ==
830 			    nfsva.na_vattr.va_mtime.tv_sec &&
831 			    np->n_mtime.tv_nsec ==
832 			    nfsva.na_vattr.va_mtime.tv_sec);
833 			NFSUNLOCKNODE(np);
834 		}
835 	}
836 nfsmout:
837 	return (error);
838 }
839 
840 /*
841  * Get postop attributes.
842  */
843 int
844 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp)
845 {
846 	u_int32_t *tl;
847 	int error = 0;
848 
849 	*retp = 0;
850 	if (nd->nd_flag & ND_NOMOREDATA)
851 		return (error);
852 	if (nd->nd_flag & ND_NFSV3) {
853 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
854 		*retp = fxdr_unsigned(int, *tl);
855 	} else if (nd->nd_flag & ND_NFSV4) {
856 		/*
857 		 * For NFSv4, the postop attr are at the end, so no point
858 		 * in looking if nd_repstat != 0.
859 		 */
860 		if (!nd->nd_repstat) {
861 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
862 			if (*(tl + 1))
863 				/* should never happen since nd_repstat != 0 */
864 				nd->nd_flag |= ND_NOMOREDATA;
865 			else
866 				*retp = 1;
867 		}
868 	} else if (!nd->nd_repstat) {
869 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
870 		*retp = 1;
871 	}
872 	if (*retp) {
873 		error = nfsm_loadattr(nd, nap);
874 		if (error)
875 			*retp = 0;
876 	}
877 nfsmout:
878 	return (error);
879 }
880 
881 /*
882  * nfscl_request() - mostly a wrapper for newnfs_request().
883  */
884 int
885 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
886     struct ucred *cred)
887 {
888 	int ret, vers;
889 	struct nfsmount *nmp;
890 
891 	nmp = VFSTONFS(vp->v_mount);
892 	if (nd->nd_flag & ND_NFSV4)
893 		vers = NFS_VER4;
894 	else if (nd->nd_flag & ND_NFSV3)
895 		vers = NFS_VER3;
896 	else
897 		vers = NFS_VER2;
898 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
899 		NFS_PROG, vers, NULL, 1, NULL, NULL);
900 	return (ret);
901 }
902 
903 /*
904  * fill in this bsden's variant of statfs using nfsstatfs.
905  */
906 void
907 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
908 {
909 	struct statfs *sbp = (struct statfs *)statfs;
910 
911 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
912 		sbp->f_bsize = NFS_FABLKSIZE;
913 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
914 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
915 		/*
916 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
917 		 * the value after dividing by NFS_FABLKSIZE is small
918 		 * enough that it will fit in 63bits, so it is ok to
919 		 * assign it to f_bavail without fear that it will become
920 		 * negative.
921 		 */
922 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
923 		sbp->f_files = sfp->sf_tfiles;
924 		/* Since f_ffree is int64_t, clip it to 63bits. */
925 		if (sfp->sf_ffiles > INT64_MAX)
926 			sbp->f_ffree = INT64_MAX;
927 		else
928 			sbp->f_ffree = sfp->sf_ffiles;
929 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
930 		/*
931 		 * The type casts to (int32_t) ensure that this code is
932 		 * compatible with the old NFS client, in that it will
933 		 * propagate bit31 to the high order bits. This may or may
934 		 * not be correct for NFSv2, but since it is a legacy
935 		 * environment, I'd rather retain backwards compatibility.
936 		 */
937 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
938 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
939 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
940 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
941 		sbp->f_files = 0;
942 		sbp->f_ffree = 0;
943 	}
944 }
945 
946 /*
947  * Use the fsinfo stuff to update the mount point.
948  */
949 void
950 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
951 {
952 
953 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
954 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
955 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
956 		    ~(NFS_FABLKSIZE - 1);
957 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
958 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
959 		if (nmp->nm_wsize == 0)
960 			nmp->nm_wsize = fsp->fs_wtmax;
961 	}
962 	if (nmp->nm_wsize < NFS_FABLKSIZE)
963 		nmp->nm_wsize = NFS_FABLKSIZE;
964 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
965 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
966 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
967 		    ~(NFS_FABLKSIZE - 1);
968 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
969 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
970 		if (nmp->nm_rsize == 0)
971 			nmp->nm_rsize = fsp->fs_rtmax;
972 	}
973 	if (nmp->nm_rsize < NFS_FABLKSIZE)
974 		nmp->nm_rsize = NFS_FABLKSIZE;
975 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
976 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
977 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
978 		    ~(NFS_DIRBLKSIZ - 1);
979 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
980 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
981 		if (nmp->nm_readdirsize == 0)
982 			nmp->nm_readdirsize = fsp->fs_rtmax;
983 	}
984 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
985 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
986 	if (fsp->fs_maxfilesize > 0 &&
987 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
988 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
989 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
990 	nmp->nm_state |= NFSSTA_GOTFSINFO;
991 }
992 
993 /*
994  * Lookups source address which should be used to communicate with
995  * @nmp and stores it inside @pdst.
996  *
997  * Returns 0 on success.
998  */
999 u_int8_t *
1000 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
1001 {
1002 #if defined(INET6) || defined(INET)
1003 	int fibnum;
1004 
1005 	fibnum = curthread->td_proc->p_fibnum;
1006 #endif
1007 #ifdef INET
1008 	if (nmp->nm_nam->sa_family == AF_INET) {
1009 		struct epoch_tracker et;
1010 		struct nhop_object *nh;
1011 		struct sockaddr_in *sin;
1012 		struct in_addr addr = {};
1013 
1014 		sin = (struct sockaddr_in *)nmp->nm_nam;
1015 		NET_EPOCH_ENTER(et);
1016 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1017 		nh = fib4_lookup(fibnum, sin->sin_addr, 0, NHR_NONE, 0);
1018 		if (nh != NULL) {
1019 			addr = IA_SIN(ifatoia(nh->nh_ifa))->sin_addr;
1020 			if (IN_LOOPBACK(ntohl(addr.s_addr))) {
1021 				/* Ignore loopback addresses */
1022 				nh = NULL;
1023 			}
1024 		}
1025 		CURVNET_RESTORE();
1026 		NET_EPOCH_EXIT(et);
1027 
1028 		if (nh == NULL)
1029 			return (NULL);
1030 		*isinet6p = 0;
1031 		*((struct in_addr *)paddr) = addr;
1032 
1033 		return (u_int8_t *)paddr;
1034 	}
1035 #endif
1036 #ifdef INET6
1037 	if (nmp->nm_nam->sa_family == AF_INET6) {
1038 		struct epoch_tracker et;
1039 		struct sockaddr_in6 *sin6;
1040 		int error;
1041 
1042 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1043 
1044 		NET_EPOCH_ENTER(et);
1045 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1046 		error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1047 		    sin6->sin6_scope_id, NULL, paddr, NULL);
1048 		CURVNET_RESTORE();
1049 		NET_EPOCH_EXIT(et);
1050 		if (error != 0)
1051 			return (NULL);
1052 
1053 		if (IN6_IS_ADDR_LOOPBACK(paddr))
1054 			return (NULL);
1055 
1056 		/* Scope is embedded in */
1057 		*isinet6p = 1;
1058 
1059 		return (u_int8_t *)paddr;
1060 	}
1061 #endif
1062 	return (NULL);
1063 }
1064 
1065 /*
1066  * Copy NFS uid, gids from the cred structure.
1067  */
1068 void
1069 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1070 {
1071 	int i;
1072 
1073 	KASSERT(cr->cr_ngroups >= 0,
1074 	    ("newnfs_copyincred: negative cr_ngroups"));
1075 	nfscr->nfsc_uid = cr->cr_uid;
1076 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1077 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1078 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1079 }
1080 
1081 /*
1082  * Do any client specific initialization.
1083  */
1084 void
1085 nfscl_init(void)
1086 {
1087 	static int inited = 0;
1088 
1089 	if (inited)
1090 		return;
1091 	inited = 1;
1092 	nfscl_inited = 1;
1093 	ncl_pbuf_zone = pbuf_zsecond_create("nfspbuf", nswbuf / 2);
1094 }
1095 
1096 /*
1097  * Check each of the attributes to be set, to ensure they aren't already
1098  * the correct value. Disable setting ones already correct.
1099  */
1100 int
1101 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1102 {
1103 
1104 	if (vap->va_mode != (mode_t)VNOVAL) {
1105 		if (vap->va_mode == nvap->na_mode)
1106 			vap->va_mode = (mode_t)VNOVAL;
1107 	}
1108 	if (vap->va_uid != (uid_t)VNOVAL) {
1109 		if (vap->va_uid == nvap->na_uid)
1110 			vap->va_uid = (uid_t)VNOVAL;
1111 	}
1112 	if (vap->va_gid != (gid_t)VNOVAL) {
1113 		if (vap->va_gid == nvap->na_gid)
1114 			vap->va_gid = (gid_t)VNOVAL;
1115 	}
1116 	if (vap->va_size != VNOVAL) {
1117 		if (vap->va_size == nvap->na_size)
1118 			vap->va_size = VNOVAL;
1119 	}
1120 
1121 	/*
1122 	 * We are normally called with only a partially initialized
1123 	 * VAP.  Since the NFSv3 spec says that server may use the
1124 	 * file attributes to store the verifier, the spec requires
1125 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1126 	 * in atime, but we can't really assume that all servers will
1127 	 * so we ensure that our SETATTR sets both atime and mtime.
1128 	 * Set the VA_UTIMES_NULL flag for this case, so that
1129 	 * the server's time will be used.  This is needed to
1130 	 * work around a bug in some Solaris servers, where
1131 	 * setting the time TOCLIENT causes the Setattr RPC
1132 	 * to return NFS_OK, but not set va_mode.
1133 	 */
1134 	if (vap->va_mtime.tv_sec == VNOVAL) {
1135 		vfs_timestamp(&vap->va_mtime);
1136 		vap->va_vaflags |= VA_UTIMES_NULL;
1137 	}
1138 	if (vap->va_atime.tv_sec == VNOVAL)
1139 		vap->va_atime = vap->va_mtime;
1140 	return (1);
1141 }
1142 
1143 /*
1144  * Map nfsv4 errors to errno.h errors.
1145  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1146  * error should only be returned for the Open, Create and Setattr Ops.
1147  * As such, most calls can just pass in 0 for those arguments.
1148  */
1149 int
1150 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1151 {
1152 	struct proc *p;
1153 
1154 	if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1155 		return (error);
1156 	if (td != NULL)
1157 		p = td->td_proc;
1158 	else
1159 		p = NULL;
1160 	switch (error) {
1161 	case NFSERR_BADOWNER:
1162 		tprintf(p, LOG_INFO,
1163 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1164 		    uid, gid);
1165 		return (EPERM);
1166 	case NFSERR_BADNAME:
1167 	case NFSERR_BADCHAR:
1168 		printf("nfsv4 char/name not handled by server\n");
1169 		return (ENOENT);
1170 	case NFSERR_STALECLIENTID:
1171 	case NFSERR_STALESTATEID:
1172 	case NFSERR_EXPIRED:
1173 	case NFSERR_BADSTATEID:
1174 	case NFSERR_BADSESSION:
1175 		printf("nfsv4 recover err returned %d\n", error);
1176 		return (EIO);
1177 	case NFSERR_BADHANDLE:
1178 	case NFSERR_SERVERFAULT:
1179 	case NFSERR_BADTYPE:
1180 	case NFSERR_FHEXPIRED:
1181 	case NFSERR_RESOURCE:
1182 	case NFSERR_MOVED:
1183 	case NFSERR_MINORVERMISMATCH:
1184 	case NFSERR_OLDSTATEID:
1185 	case NFSERR_BADSEQID:
1186 	case NFSERR_LEASEMOVED:
1187 	case NFSERR_RECLAIMBAD:
1188 	case NFSERR_BADXDR:
1189 	case NFSERR_OPILLEGAL:
1190 		printf("nfsv4 client/server protocol prob err=%d\n",
1191 		    error);
1192 		return (EIO);
1193 	case NFSERR_NOFILEHANDLE:
1194 		printf("nfsv4 no file handle: usually means the file "
1195 		    "system is not exported on the NFSv4 server\n");
1196 		return (EIO);
1197 	case NFSERR_WRONGSEC:
1198 		tprintf(p, LOG_INFO, "NFSv4 error WrongSec: You probably need a"
1199 		    " Kerberos TGT\n");
1200 		return (EIO);
1201 	default:
1202 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1203 		return (EIO);
1204 	};
1205 }
1206 
1207 /*
1208  * Check to see if the process for this owner exists. Return 1 if it doesn't
1209  * and 0 otherwise.
1210  */
1211 int
1212 nfscl_procdoesntexist(u_int8_t *own)
1213 {
1214 	union {
1215 		u_int32_t	lval;
1216 		u_int8_t	cval[4];
1217 	} tl;
1218 	struct proc *p;
1219 	pid_t pid;
1220 	int i, ret = 0;
1221 
1222 	/* For the single open_owner of all 0 bytes, just return 0. */
1223 	for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1224 		if (own[i] != 0)
1225 			break;
1226 	if (i == NFSV4CL_LOCKNAMELEN)
1227 		return (0);
1228 
1229 	tl.cval[0] = *own++;
1230 	tl.cval[1] = *own++;
1231 	tl.cval[2] = *own++;
1232 	tl.cval[3] = *own++;
1233 	pid = tl.lval;
1234 	p = pfind_any_locked(pid);
1235 	if (p == NULL)
1236 		return (1);
1237 	if (p->p_stats == NULL) {
1238 		PROC_UNLOCK(p);
1239 		return (0);
1240 	}
1241 	tl.cval[0] = *own++;
1242 	tl.cval[1] = *own++;
1243 	tl.cval[2] = *own++;
1244 	tl.cval[3] = *own++;
1245 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1246 		ret = 1;
1247 	} else {
1248 		tl.cval[0] = *own++;
1249 		tl.cval[1] = *own++;
1250 		tl.cval[2] = *own++;
1251 		tl.cval[3] = *own;
1252 		if (tl.lval != p->p_stats->p_start.tv_usec)
1253 			ret = 1;
1254 	}
1255 	PROC_UNLOCK(p);
1256 	return (ret);
1257 }
1258 
1259 /*
1260  * - nfs pseudo system call for the client
1261  */
1262 /*
1263  * MPSAFE
1264  */
1265 static int
1266 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1267 {
1268 	struct file *fp;
1269 	struct nfscbd_args nfscbdarg;
1270 	struct nfsd_nfscbd_args nfscbdarg2;
1271 	struct nameidata nd;
1272 	struct nfscl_dumpmntopts dumpmntopts;
1273 	cap_rights_t rights;
1274 	char *buf;
1275 	int error;
1276 	struct mount *mp;
1277 	struct nfsmount *nmp;
1278 
1279 	NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td));
1280 	if (uap->flag & NFSSVC_CBADDSOCK) {
1281 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1282 		if (error)
1283 			goto out;
1284 		/*
1285 		 * Since we don't know what rights might be required,
1286 		 * pretend that we need them all. It is better to be too
1287 		 * careful than too reckless.
1288 		 */
1289 		error = fget(td, nfscbdarg.sock,
1290 		    cap_rights_init_one(&rights, CAP_SOCK_CLIENT), &fp);
1291 		if (error)
1292 			goto out;
1293 		if (fp->f_type != DTYPE_SOCKET) {
1294 			fdrop(fp, td);
1295 			error = EPERM;
1296 			goto out;
1297 		}
1298 		error = nfscbd_addsock(fp);
1299 		fdrop(fp, td);
1300 		if (!error && nfscl_enablecallb == 0) {
1301 			nfsv4_cbport = nfscbdarg.port;
1302 			nfscl_enablecallb = 1;
1303 		}
1304 	} else if (uap->flag & NFSSVC_NFSCBD) {
1305 		if (uap->argp == NULL) {
1306 			error = EINVAL;
1307 			goto out;
1308 		}
1309 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1310 		    sizeof(nfscbdarg2));
1311 		if (error)
1312 			goto out;
1313 		error = nfscbd_nfsd(td, &nfscbdarg2);
1314 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1315 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1316 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1317 		    dumpmntopts.ndmnt_blen > 1024))
1318 			error = EINVAL;
1319 		if (error == 0)
1320 			error = nfsrv_lookupfilename(&nd,
1321 			    dumpmntopts.ndmnt_fname, td);
1322 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1323 		    "nfs") != 0) {
1324 			vput(nd.ni_vp);
1325 			error = EINVAL;
1326 		}
1327 		if (error == 0) {
1328 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK |
1329 			    M_ZERO);
1330 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1331 			    dumpmntopts.ndmnt_blen);
1332 			vput(nd.ni_vp);
1333 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1334 			    dumpmntopts.ndmnt_blen);
1335 			free(buf, M_TEMP);
1336 		}
1337 	} else if (uap->flag & NFSSVC_FORCEDISM) {
1338 		buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
1339 		error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
1340 		if (error == 0) {
1341 			nmp = NULL;
1342 			mtx_lock(&mountlist_mtx);
1343 			TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1344 				if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
1345 				    0 && strcmp(mp->mnt_stat.f_fstypename,
1346 				    "nfs") == 0 && mp->mnt_data != NULL) {
1347 					nmp = VFSTONFS(mp);
1348 					NFSDDSLOCK();
1349 					if (nfsv4_findmirror(nmp) != NULL) {
1350 						NFSDDSUNLOCK();
1351 						error = ENXIO;
1352 						nmp = NULL;
1353 						break;
1354 					}
1355 					mtx_lock(&nmp->nm_mtx);
1356 					if ((nmp->nm_privflag &
1357 					    NFSMNTP_FORCEDISM) == 0) {
1358 						nmp->nm_privflag |=
1359 						   (NFSMNTP_FORCEDISM |
1360 						    NFSMNTP_CANCELRPCS);
1361 						mtx_unlock(&nmp->nm_mtx);
1362 					} else {
1363 						mtx_unlock(&nmp->nm_mtx);
1364 						nmp = NULL;
1365 					}
1366 					NFSDDSUNLOCK();
1367 					break;
1368 				}
1369 			}
1370 			mtx_unlock(&mountlist_mtx);
1371 
1372 			if (nmp != NULL) {
1373 				/*
1374 				 * Call newnfs_nmcancelreqs() to cause
1375 				 * any RPCs in progress on the mount point to
1376 				 * fail.
1377 				 * This will cause any process waiting for an
1378 				 * RPC to complete while holding a vnode lock
1379 				 * on the mounted-on vnode (such as "df" or
1380 				 * a non-forced "umount") to fail.
1381 				 * This will unlock the mounted-on vnode so
1382 				 * a forced dismount can succeed.
1383 				 * Then clear NFSMNTP_CANCELRPCS and wakeup(),
1384 				 * so that nfs_unmount() can complete.
1385 				 */
1386 				newnfs_nmcancelreqs(nmp);
1387 				mtx_lock(&nmp->nm_mtx);
1388 				nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
1389 				wakeup(nmp);
1390 				mtx_unlock(&nmp->nm_mtx);
1391 			} else if (error == 0)
1392 				error = EINVAL;
1393 		}
1394 		free(buf, M_TEMP);
1395 	} else {
1396 		error = EINVAL;
1397 	}
1398 out:
1399 	NFSD_CURVNET_RESTORE();
1400 	return (error);
1401 }
1402 
1403 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1404 
1405 /*
1406  * Called once to initialize data structures...
1407  */
1408 static int
1409 nfscl_modevent(module_t mod, int type, void *data)
1410 {
1411 	int error = 0;
1412 	static int loaded = 0;
1413 
1414 	switch (type) {
1415 	case MOD_LOAD:
1416 		if (loaded)
1417 			return (0);
1418 		newnfs_portinit();
1419 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1420 		nfscl_init();
1421 		NFSD_LOCK();
1422 		nfsrvd_cbinit(0);
1423 		NFSD_UNLOCK();
1424 		ncl_call_invalcaches = ncl_invalcaches;
1425 		nfsd_call_nfscl = nfssvc_nfscl;
1426 		loaded = 1;
1427 		break;
1428 
1429 	case MOD_UNLOAD:
1430 		if (nfs_numnfscbd != 0) {
1431 			error = EBUSY;
1432 			break;
1433 		}
1434 
1435 		/*
1436 		 * XXX: Unloading of nfscl module is unsupported.
1437 		 */
1438 #if 0
1439 		ncl_call_invalcaches = NULL;
1440 		nfsd_call_nfscl = NULL;
1441 		uma_zdestroy(ncl_pbuf_zone);
1442 		/* and get rid of the mutexes */
1443 		mtx_destroy(&ncl_iod_mutex);
1444 		loaded = 0;
1445 		break;
1446 #else
1447 		/* FALLTHROUGH */
1448 #endif
1449 	default:
1450 		error = EOPNOTSUPP;
1451 		break;
1452 	}
1453 	return error;
1454 }
1455 static moduledata_t nfscl_mod = {
1456 	"nfscl",
1457 	nfscl_modevent,
1458 	NULL,
1459 };
1460 /*
1461  * This is the main module declaration for the NFS client.  The
1462  * nfscl_modevent() function is needed to ensure that the module
1463  * cannot be unloaded, among other things.
1464  * There is also a module declaration in sys/fs/nfsclient/nfs_clvfsops.c
1465  * for the name "nfs" within the VFS_SET() macro that defines the "nfs"
1466  * file system type.
1467  */
1468 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1469 
1470 /* So that loader and kldload(2) can find us, wherever we are.. */
1471 MODULE_VERSION(nfscl, 1);
1472 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1473 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1474 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1475 MODULE_DEPEND(nfscl, xdr, 1, 1, 1);
1476