xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision b0d29bc47dba79f6f38e67eabadfb4b32ffd9390)
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 VIRF_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 VIRF_DOOMED
155 		 * flag isn't set now.
156 		 */
157 		VI_LOCK(nvp);
158 		if (VN_IS_DOOMED(nvp)) {
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 	vp->v_vflag |= VV_VMSIZEVNLOCK;
251 
252 	np->n_fhp = nfhp;
253 	/*
254 	 * For NFSv4, we have to attach the directory file handle and
255 	 * file name, so that Open Ops can be done later.
256 	 */
257 	if (nmp->nm_flag & NFSMNT_NFSV4) {
258 		np->n_v4 = malloc(sizeof (struct nfsv4node)
259 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
260 		    M_WAITOK);
261 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
262 		np->n_v4->n4_namelen = cnp->cn_namelen;
263 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
264 		    dnp->n_fhp->nfh_len);
265 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
266 		    cnp->cn_namelen);
267 	} else {
268 		np->n_v4 = NULL;
269 	}
270 
271 	/*
272 	 * NFS supports recursive and shared locking.
273 	 */
274 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
275 	VN_LOCK_AREC(vp);
276 	VN_LOCK_ASHARE(vp);
277 	error = insmntque(vp, mntp);
278 	if (error != 0) {
279 		*npp = NULL;
280 		mtx_destroy(&np->n_mtx);
281 		lockdestroy(&np->n_excl);
282 		free(nfhp, M_NFSFH);
283 		if (np->n_v4 != NULL)
284 			free(np->n_v4, M_NFSV4NODE);
285 		uma_zfree(newnfsnode_zone, np);
286 		return (error);
287 	}
288 	error = vfs_hash_insert(vp, hash, lkflags,
289 	    td, &nvp, newnfs_vncmpf, nfhp);
290 	if (error)
291 		return (error);
292 	if (nvp != NULL) {
293 		*npp = VTONFS(nvp);
294 		/* vfs_hash_insert() vput()'s the losing vnode */
295 		return (0);
296 	}
297 	*npp = np;
298 
299 	return (0);
300 }
301 
302 /*
303  * Another variant of nfs_nget(). This one is only used by reopen. It
304  * takes almost the same args as nfs_nget(), but only succeeds if an entry
305  * exists in the cache. (Since files should already be "open" with a
306  * vnode ref cnt on the node when reopen calls this, it should always
307  * succeed.)
308  * Also, don't get a vnode lock, since it may already be locked by some
309  * other process that is handling it. This is ok, since all other threads
310  * on the client are blocked by the nfsc_lock being exclusively held by the
311  * caller of this function.
312  */
313 int
314 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
315     struct thread *td, struct nfsnode **npp)
316 {
317 	struct vnode *nvp;
318 	u_int hash;
319 	struct nfsfh *nfhp;
320 	int error;
321 
322 	*npp = NULL;
323 	/* For forced dismounts, just return error. */
324 	if (NFSCL_FORCEDISM(mntp))
325 		return (EINTR);
326 	nfhp = malloc(sizeof (struct nfsfh) + fhsize,
327 	    M_NFSFH, M_WAITOK);
328 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
329 	nfhp->nfh_len = fhsize;
330 
331 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
332 
333 	/*
334 	 * First, try to get the vnode locked, but don't block for the lock.
335 	 */
336 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
337 	    newnfs_vncmpf, nfhp);
338 	if (error == 0 && nvp != NULL) {
339 		NFSVOPUNLOCK(nvp);
340 	} else if (error == EBUSY) {
341 		/*
342 		 * It is safe so long as a vflush() with
343 		 * FORCECLOSE has not been done. Since the Renew thread is
344 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
345 		 * a vflush() with FORCECLOSE, we should be ok here.
346 		 */
347 		if (NFSCL_FORCEDISM(mntp))
348 			error = EINTR;
349 		else {
350 			vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
351 			if (nvp == NULL) {
352 				error = ENOENT;
353 			} else if (VN_IS_DOOMED(nvp)) {
354 				error = ENOENT;
355 				vrele(nvp);
356 			} else {
357 				error = 0;
358 			}
359 		}
360 	}
361 	free(nfhp, M_NFSFH);
362 	if (error)
363 		return (error);
364 	if (nvp != NULL) {
365 		*npp = VTONFS(nvp);
366 		return (0);
367 	}
368 	return (EINVAL);
369 }
370 
371 static void
372 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
373     struct nfsvattr *newnap)
374 {
375 	int off;
376 
377 	if (ncl_fileid_maxwarnings >= 0 &&
378 	    ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
379 		return;
380 	off = 0;
381 	if (ncl_fileid_maxwarnings >= 0) {
382 		if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
383 			off = 1;
384 	}
385 
386 	printf("newnfs: server '%s' error: fileid changed. "
387 	    "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
388 	    "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
389 	    nmp->nm_com.nmcom_hostname,
390 	    (uintmax_t)nmp->nm_fsid[0],
391 	    (uintmax_t)nmp->nm_fsid[1],
392 	    (uintmax_t)oldnap->na_fileid,
393 	    (uintmax_t)newnap->na_fileid);
394 
395 	if (off)
396 		printf("newnfs: Logged %d times about fileid corruption; "
397 		    "going quiet to avoid spamming logs excessively. (Limit "
398 		    "is: %d).\n", ncl_fileid_nwarnings,
399 		    ncl_fileid_maxwarnings);
400 }
401 
402 /*
403  * Load the attribute cache (that lives in the nfsnode entry) with
404  * the attributes of the second argument and
405  * Iff vaper not NULL
406  *    copy the attributes to *vaper
407  * Similar to nfs_loadattrcache(), except the attributes are passed in
408  * instead of being parsed out of the mbuf list.
409  */
410 int
411 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
412     void *stuff, int writeattr, int dontshrink)
413 {
414 	struct vnode *vp = *vpp;
415 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
416 	struct nfsnode *np;
417 	struct nfsmount *nmp;
418 	struct timespec mtime_save;
419 	int error, force_fid_err;
420 
421 	error = 0;
422 
423 	/*
424 	 * If v_type == VNON it is a new node, so fill in the v_type,
425 	 * n_mtime fields. Check to see if it represents a special
426 	 * device, and if so, check for a possible alias. Once the
427 	 * correct vnode has been obtained, fill in the rest of the
428 	 * information.
429 	 */
430 	np = VTONFS(vp);
431 	NFSLOCKNODE(np);
432 	if (vp->v_type != nvap->va_type) {
433 		vp->v_type = nvap->va_type;
434 		if (vp->v_type == VFIFO)
435 			vp->v_op = &newnfs_fifoops;
436 		np->n_mtime = nvap->va_mtime;
437 	}
438 	nmp = VFSTONFS(vp->v_mount);
439 	vap = &np->n_vattr.na_vattr;
440 	mtime_save = vap->va_mtime;
441 	if (writeattr) {
442 		np->n_vattr.na_filerev = nap->na_filerev;
443 		np->n_vattr.na_size = nap->na_size;
444 		np->n_vattr.na_mtime = nap->na_mtime;
445 		np->n_vattr.na_ctime = nap->na_ctime;
446 		np->n_vattr.na_fsid = nap->na_fsid;
447 		np->n_vattr.na_mode = nap->na_mode;
448 	} else {
449 		force_fid_err = 0;
450 		KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
451 		    force_fid_err);
452 		/*
453 		 * BROKEN NFS SERVER OR MIDDLEWARE
454 		 *
455 		 * Certain NFS servers (certain old proprietary filers ca.
456 		 * 2006) or broken middleboxes (e.g. WAN accelerator products)
457 		 * will respond to GETATTR requests with results for a
458 		 * different fileid.
459 		 *
460 		 * The WAN accelerator we've observed not only serves stale
461 		 * cache results for a given file, it also occasionally serves
462 		 * results for wholly different files.  This causes surprising
463 		 * problems; for example the cached size attribute of a file
464 		 * may truncate down and then back up, resulting in zero
465 		 * regions in file contents read by applications.  We observed
466 		 * this reliably with Clang and .c files during parallel build.
467 		 * A pcap revealed packet fragmentation and GETATTR RPC
468 		 * responses with wholly wrong fileids.
469 		 */
470 		if ((np->n_vattr.na_fileid != 0 &&
471 		     np->n_vattr.na_fileid != nap->na_fileid) ||
472 		    force_fid_err) {
473 			nfscl_warn_fileid(nmp, &np->n_vattr, nap);
474 			error = EIDRM;
475 			goto out;
476 		}
477 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
478 		    sizeof (struct nfsvattr));
479 	}
480 
481 	/*
482 	 * For NFSv4, if the node's fsid is not equal to the mount point's
483 	 * fsid, return the low order 32bits of the node's fsid. This
484 	 * allows getcwd(3) to work. There is a chance that the fsid might
485 	 * be the same as a local fs, but since this is in an NFS mount
486 	 * point, I don't think that will cause any problems?
487 	 */
488 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
489 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
490 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
491 		/*
492 		 * va_fsid needs to be set to some value derived from
493 		 * np->n_vattr.na_filesid that is not equal
494 		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
495 		 * from the value used for the top level server volume
496 		 * in the mounted subtree.
497 		 */
498 		vn_fsid(vp, vap);
499 		if ((uint32_t)vap->va_fsid == np->n_vattr.na_filesid[0])
500 			vap->va_fsid = hash32_buf(
501 			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
502 	} else
503 		vn_fsid(vp, vap);
504 	np->n_attrstamp = time_second;
505 	if (vap->va_size != np->n_size) {
506 		if (vap->va_type == VREG) {
507 			if (dontshrink && vap->va_size < np->n_size) {
508 				/*
509 				 * We've been told not to shrink the file;
510 				 * zero np->n_attrstamp to indicate that
511 				 * the attributes are stale.
512 				 */
513 				vap->va_size = np->n_size;
514 				np->n_attrstamp = 0;
515 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
516 			} else if (np->n_flag & NMODIFIED) {
517 				/*
518 				 * We've modified the file: Use the larger
519 				 * of our size, and the server's size.
520 				 */
521 				if (vap->va_size < np->n_size) {
522 					vap->va_size = np->n_size;
523 				} else {
524 					np->n_size = vap->va_size;
525 					np->n_flag |= NSIZECHANGED;
526 				}
527 			} else {
528 				np->n_size = vap->va_size;
529 				np->n_flag |= NSIZECHANGED;
530 			}
531 		} else {
532 			np->n_size = vap->va_size;
533 		}
534 	}
535 	/*
536 	 * The following checks are added to prevent a race between (say)
537 	 * a READDIR+ and a WRITE.
538 	 * READDIR+, WRITE requests sent out.
539 	 * READDIR+ resp, WRITE resp received on client.
540 	 * However, the WRITE resp was handled before the READDIR+ resp
541 	 * causing the post op attrs from the write to be loaded first
542 	 * and the attrs from the READDIR+ to be loaded later. If this
543 	 * happens, we have stale attrs loaded into the attrcache.
544 	 * We detect this by for the mtime moving back. We invalidate the
545 	 * attrcache when this happens.
546 	 */
547 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
548 		/* Size changed or mtime went backwards */
549 		np->n_attrstamp = 0;
550 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
551 	}
552 	if (vaper != NULL) {
553 		NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
554 		if (np->n_flag & NCHG) {
555 			if (np->n_flag & NACC)
556 				vaper->va_atime = np->n_atim;
557 			if (np->n_flag & NUPD)
558 				vaper->va_mtime = np->n_mtim;
559 		}
560 	}
561 
562 out:
563 #ifdef KDTRACE_HOOKS
564 	if (np->n_attrstamp != 0)
565 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
566 #endif
567 	(void)ncl_pager_setsize(vp, NULL);
568 	return (error);
569 }
570 
571 /*
572  * Call vnode_pager_setsize() if the size of the node changed, as
573  * recorded in nfsnode vs. v_object, or delay the call if notifying
574  * the pager is not possible at the moment.
575  *
576  * If nsizep is non-NULL, the call is delayed and the new node size is
577  * provided.  Caller should itself call vnode_pager_setsize() if
578  * function returned true.  If nsizep is NULL, function tries to call
579  * vnode_pager_setsize() itself if needed and possible, and the nfs
580  * node is unlocked unconditionally, the return value is not useful.
581  */
582 bool
583 ncl_pager_setsize(struct vnode *vp, u_quad_t *nsizep)
584 {
585 	struct nfsnode *np;
586 	vm_object_t object;
587 	struct vattr *vap;
588 	u_quad_t nsize;
589 	bool setnsize;
590 
591 	np = VTONFS(vp);
592 	NFSASSERTNODE(np);
593 
594 	vap = &np->n_vattr.na_vattr;
595 	nsize = vap->va_size;
596 	object = vp->v_object;
597 	setnsize = false;
598 
599 	if (object != NULL && nsize != object->un_pager.vnp.vnp_size) {
600 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
601 		    (curthread->td_pflags2 & TDP2_SBPAGES) == 0)
602 			setnsize = true;
603 		else
604 			np->n_flag |= NVNSETSZSKIP;
605 	}
606 	if (nsizep == NULL) {
607 		NFSUNLOCKNODE(np);
608 		if (setnsize)
609 			vnode_pager_setsize(vp, nsize);
610 		setnsize = false;
611 	} else {
612 		*nsizep = nsize;
613 	}
614 	return (setnsize);
615 }
616 
617 /*
618  * Fill in the client id name. For these bytes:
619  * 1 - they must be unique
620  * 2 - they should be persistent across client reboots
621  * 1 is more critical than 2
622  * Use the mount point's unique id plus either the uuid or, if that
623  * isn't set, random junk.
624  */
625 void
626 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
627 {
628 	int uuidlen;
629 
630 	/*
631 	 * First, put in the 64bit mount point identifier.
632 	 */
633 	if (idlen >= sizeof (u_int64_t)) {
634 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
635 		cp += sizeof (u_int64_t);
636 		idlen -= sizeof (u_int64_t);
637 	}
638 
639 	/*
640 	 * If uuid is non-zero length, use it.
641 	 */
642 	uuidlen = strlen(uuid);
643 	if (uuidlen > 0 && idlen >= uuidlen) {
644 		NFSBCOPY(uuid, cp, uuidlen);
645 		cp += uuidlen;
646 		idlen -= uuidlen;
647 	}
648 
649 	/*
650 	 * This only normally happens if the uuid isn't set.
651 	 */
652 	while (idlen > 0) {
653 		*cp++ = (u_int8_t)(arc4random() % 256);
654 		idlen--;
655 	}
656 }
657 
658 /*
659  * Fill in a lock owner name. For now, pid + the process's creation time.
660  */
661 void
662 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
663 {
664 	union {
665 		u_int32_t	lval;
666 		u_int8_t	cval[4];
667 	} tl;
668 	struct proc *p;
669 
670 	if (id == NULL) {
671 		/* Return the single open_owner of all 0 bytes. */
672 		bzero(cp, NFSV4CL_LOCKNAMELEN);
673 		return;
674 	}
675 	if ((flags & F_POSIX) != 0) {
676 		p = (struct proc *)id;
677 		tl.lval = p->p_pid;
678 		*cp++ = tl.cval[0];
679 		*cp++ = tl.cval[1];
680 		*cp++ = tl.cval[2];
681 		*cp++ = tl.cval[3];
682 		tl.lval = p->p_stats->p_start.tv_sec;
683 		*cp++ = tl.cval[0];
684 		*cp++ = tl.cval[1];
685 		*cp++ = tl.cval[2];
686 		*cp++ = tl.cval[3];
687 		tl.lval = p->p_stats->p_start.tv_usec;
688 		*cp++ = tl.cval[0];
689 		*cp++ = tl.cval[1];
690 		*cp++ = tl.cval[2];
691 		*cp = tl.cval[3];
692 	} else if ((flags & F_FLOCK) != 0) {
693 		bcopy(&id, cp, sizeof(id));
694 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
695 	} else {
696 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
697 		bzero(cp, NFSV4CL_LOCKNAMELEN);
698 	}
699 }
700 
701 /*
702  * Find the parent process for the thread passed in as an argument.
703  * If none exists, return NULL, otherwise return a thread for the parent.
704  * (Can be any of the threads, since it is only used for td->td_proc.)
705  */
706 NFSPROC_T *
707 nfscl_getparent(struct thread *td)
708 {
709 	struct proc *p;
710 	struct thread *ptd;
711 
712 	if (td == NULL)
713 		return (NULL);
714 	p = td->td_proc;
715 	if (p->p_pid == 0)
716 		return (NULL);
717 	p = p->p_pptr;
718 	if (p == NULL)
719 		return (NULL);
720 	ptd = TAILQ_FIRST(&p->p_threads);
721 	return (ptd);
722 }
723 
724 /*
725  * Start up the renew kernel thread.
726  */
727 static void
728 start_nfscl(void *arg)
729 {
730 	struct nfsclclient *clp;
731 	struct thread *td;
732 
733 	clp = (struct nfsclclient *)arg;
734 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
735 	nfscl_renewthread(clp, td);
736 	kproc_exit(0);
737 }
738 
739 void
740 nfscl_start_renewthread(struct nfsclclient *clp)
741 {
742 
743 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
744 	    "nfscl");
745 }
746 
747 /*
748  * Handle wcc_data.
749  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
750  * as the first Op after PutFH.
751  * (For NFSv4, the postop attributes are after the Op, so they can't be
752  *  parsed here. A separate call to nfscl_postop_attr() is required.)
753  */
754 int
755 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
756     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
757 {
758 	u_int32_t *tl;
759 	struct nfsnode *np = VTONFS(vp);
760 	struct nfsvattr nfsva;
761 	int error = 0;
762 
763 	if (wccflagp != NULL)
764 		*wccflagp = 0;
765 	if (nd->nd_flag & ND_NFSV3) {
766 		*flagp = 0;
767 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
768 		if (*tl == newnfs_true) {
769 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
770 			if (wccflagp != NULL) {
771 				NFSLOCKNODE(np);
772 				*wccflagp = (np->n_mtime.tv_sec ==
773 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
774 				    np->n_mtime.tv_nsec ==
775 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
776 				NFSUNLOCKNODE(np);
777 			}
778 		}
779 		error = nfscl_postop_attr(nd, nap, flagp, stuff);
780 		if (wccflagp != NULL && *flagp == 0)
781 			*wccflagp = 0;
782 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
783 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
784 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
785 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
786 		    NULL, NULL, NULL, NULL, NULL);
787 		if (error)
788 			return (error);
789 		/*
790 		 * Get rid of Op# and status for next op.
791 		 */
792 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
793 		if (*++tl)
794 			nd->nd_flag |= ND_NOMOREDATA;
795 		if (wccflagp != NULL &&
796 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
797 			NFSLOCKNODE(np);
798 			*wccflagp = (np->n_mtime.tv_sec ==
799 			    nfsva.na_vattr.va_mtime.tv_sec &&
800 			    np->n_mtime.tv_nsec ==
801 			    nfsva.na_vattr.va_mtime.tv_sec);
802 			NFSUNLOCKNODE(np);
803 		}
804 	}
805 nfsmout:
806 	return (error);
807 }
808 
809 /*
810  * Get postop attributes.
811  */
812 int
813 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
814     void *stuff)
815 {
816 	u_int32_t *tl;
817 	int error = 0;
818 
819 	*retp = 0;
820 	if (nd->nd_flag & ND_NOMOREDATA)
821 		return (error);
822 	if (nd->nd_flag & ND_NFSV3) {
823 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
824 		*retp = fxdr_unsigned(int, *tl);
825 	} else if (nd->nd_flag & ND_NFSV4) {
826 		/*
827 		 * For NFSv4, the postop attr are at the end, so no point
828 		 * in looking if nd_repstat != 0.
829 		 */
830 		if (!nd->nd_repstat) {
831 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
832 			if (*(tl + 1))
833 				/* should never happen since nd_repstat != 0 */
834 				nd->nd_flag |= ND_NOMOREDATA;
835 			else
836 				*retp = 1;
837 		}
838 	} else if (!nd->nd_repstat) {
839 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
840 		*retp = 1;
841 	}
842 	if (*retp) {
843 		error = nfsm_loadattr(nd, nap);
844 		if (error)
845 			*retp = 0;
846 	}
847 nfsmout:
848 	return (error);
849 }
850 
851 /*
852  * nfscl_request() - mostly a wrapper for newnfs_request().
853  */
854 int
855 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
856     struct ucred *cred, void *stuff)
857 {
858 	int ret, vers;
859 	struct nfsmount *nmp;
860 
861 	nmp = VFSTONFS(vp->v_mount);
862 	if (nd->nd_flag & ND_NFSV4)
863 		vers = NFS_VER4;
864 	else if (nd->nd_flag & ND_NFSV3)
865 		vers = NFS_VER3;
866 	else
867 		vers = NFS_VER2;
868 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
869 		NFS_PROG, vers, NULL, 1, NULL, NULL);
870 	return (ret);
871 }
872 
873 /*
874  * fill in this bsden's variant of statfs using nfsstatfs.
875  */
876 void
877 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
878 {
879 	struct statfs *sbp = (struct statfs *)statfs;
880 
881 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
882 		sbp->f_bsize = NFS_FABLKSIZE;
883 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
884 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
885 		/*
886 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
887 		 * the value after dividing by NFS_FABLKSIZE is small
888 		 * enough that it will fit in 63bits, so it is ok to
889 		 * assign it to f_bavail without fear that it will become
890 		 * negative.
891 		 */
892 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
893 		sbp->f_files = sfp->sf_tfiles;
894 		/* Since f_ffree is int64_t, clip it to 63bits. */
895 		if (sfp->sf_ffiles > INT64_MAX)
896 			sbp->f_ffree = INT64_MAX;
897 		else
898 			sbp->f_ffree = sfp->sf_ffiles;
899 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
900 		/*
901 		 * The type casts to (int32_t) ensure that this code is
902 		 * compatible with the old NFS client, in that it will
903 		 * propagate bit31 to the high order bits. This may or may
904 		 * not be correct for NFSv2, but since it is a legacy
905 		 * environment, I'd rather retain backwards compatibility.
906 		 */
907 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
908 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
909 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
910 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
911 		sbp->f_files = 0;
912 		sbp->f_ffree = 0;
913 	}
914 }
915 
916 /*
917  * Use the fsinfo stuff to update the mount point.
918  */
919 void
920 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
921 {
922 
923 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
924 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
925 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
926 		    ~(NFS_FABLKSIZE - 1);
927 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
928 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
929 		if (nmp->nm_wsize == 0)
930 			nmp->nm_wsize = fsp->fs_wtmax;
931 	}
932 	if (nmp->nm_wsize < NFS_FABLKSIZE)
933 		nmp->nm_wsize = NFS_FABLKSIZE;
934 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
935 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
936 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
937 		    ~(NFS_FABLKSIZE - 1);
938 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
939 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
940 		if (nmp->nm_rsize == 0)
941 			nmp->nm_rsize = fsp->fs_rtmax;
942 	}
943 	if (nmp->nm_rsize < NFS_FABLKSIZE)
944 		nmp->nm_rsize = NFS_FABLKSIZE;
945 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
946 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
947 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
948 		    ~(NFS_DIRBLKSIZ - 1);
949 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
950 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
951 		if (nmp->nm_readdirsize == 0)
952 			nmp->nm_readdirsize = fsp->fs_rtmax;
953 	}
954 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
955 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
956 	if (fsp->fs_maxfilesize > 0 &&
957 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
958 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
959 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
960 	nmp->nm_state |= NFSSTA_GOTFSINFO;
961 }
962 
963 /*
964  * Lookups source address which should be used to communicate with
965  * @nmp and stores it inside @pdst.
966  *
967  * Returns 0 on success.
968  */
969 u_int8_t *
970 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
971 {
972 #if defined(INET6) || defined(INET)
973 	int error, fibnum;
974 
975 	fibnum = curthread->td_proc->p_fibnum;
976 #endif
977 #ifdef INET
978 	if (nmp->nm_nam->sa_family == AF_INET) {
979 		struct sockaddr_in *sin;
980 		struct nhop4_extended nh_ext;
981 
982 		sin = (struct sockaddr_in *)nmp->nm_nam;
983 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
984 		error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
985 		    &nh_ext);
986 		CURVNET_RESTORE();
987 		if (error != 0)
988 			return (NULL);
989 
990 		if (IN_LOOPBACK(ntohl(nh_ext.nh_src.s_addr))) {
991 			/* Ignore loopback addresses */
992 			return (NULL);
993 		}
994 
995 		*isinet6p = 0;
996 		*((struct in_addr *)paddr) = nh_ext.nh_src;
997 
998 		return (u_int8_t *)paddr;
999 	}
1000 #endif
1001 #ifdef INET6
1002 	if (nmp->nm_nam->sa_family == AF_INET6) {
1003 		struct sockaddr_in6 *sin6;
1004 
1005 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1006 
1007 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1008 		error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1009 		    sin6->sin6_scope_id, NULL, paddr, NULL);
1010 		CURVNET_RESTORE();
1011 		if (error != 0)
1012 			return (NULL);
1013 
1014 		if (IN6_IS_ADDR_LOOPBACK(paddr))
1015 			return (NULL);
1016 
1017 		/* Scope is embedded in */
1018 		*isinet6p = 1;
1019 
1020 		return (u_int8_t *)paddr;
1021 	}
1022 #endif
1023 	return (NULL);
1024 }
1025 
1026 /*
1027  * Copy NFS uid, gids from the cred structure.
1028  */
1029 void
1030 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1031 {
1032 	int i;
1033 
1034 	KASSERT(cr->cr_ngroups >= 0,
1035 	    ("newnfs_copyincred: negative cr_ngroups"));
1036 	nfscr->nfsc_uid = cr->cr_uid;
1037 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1038 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1039 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1040 }
1041 
1042 
1043 /*
1044  * Do any client specific initialization.
1045  */
1046 void
1047 nfscl_init(void)
1048 {
1049 	static int inited = 0;
1050 
1051 	if (inited)
1052 		return;
1053 	inited = 1;
1054 	nfscl_inited = 1;
1055 	ncl_pbuf_zone = pbuf_zsecond_create("nfspbuf", nswbuf / 2);
1056 }
1057 
1058 /*
1059  * Check each of the attributes to be set, to ensure they aren't already
1060  * the correct value. Disable setting ones already correct.
1061  */
1062 int
1063 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1064 {
1065 
1066 	if (vap->va_mode != (mode_t)VNOVAL) {
1067 		if (vap->va_mode == nvap->na_mode)
1068 			vap->va_mode = (mode_t)VNOVAL;
1069 	}
1070 	if (vap->va_uid != (uid_t)VNOVAL) {
1071 		if (vap->va_uid == nvap->na_uid)
1072 			vap->va_uid = (uid_t)VNOVAL;
1073 	}
1074 	if (vap->va_gid != (gid_t)VNOVAL) {
1075 		if (vap->va_gid == nvap->na_gid)
1076 			vap->va_gid = (gid_t)VNOVAL;
1077 	}
1078 	if (vap->va_size != VNOVAL) {
1079 		if (vap->va_size == nvap->na_size)
1080 			vap->va_size = VNOVAL;
1081 	}
1082 
1083 	/*
1084 	 * We are normally called with only a partially initialized
1085 	 * VAP.  Since the NFSv3 spec says that server may use the
1086 	 * file attributes to store the verifier, the spec requires
1087 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1088 	 * in atime, but we can't really assume that all servers will
1089 	 * so we ensure that our SETATTR sets both atime and mtime.
1090 	 * Set the VA_UTIMES_NULL flag for this case, so that
1091 	 * the server's time will be used.  This is needed to
1092 	 * work around a bug in some Solaris servers, where
1093 	 * setting the time TOCLIENT causes the Setattr RPC
1094 	 * to return NFS_OK, but not set va_mode.
1095 	 */
1096 	if (vap->va_mtime.tv_sec == VNOVAL) {
1097 		vfs_timestamp(&vap->va_mtime);
1098 		vap->va_vaflags |= VA_UTIMES_NULL;
1099 	}
1100 	if (vap->va_atime.tv_sec == VNOVAL)
1101 		vap->va_atime = vap->va_mtime;
1102 	return (1);
1103 }
1104 
1105 /*
1106  * Map nfsv4 errors to errno.h errors.
1107  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1108  * error should only be returned for the Open, Create and Setattr Ops.
1109  * As such, most calls can just pass in 0 for those arguments.
1110  */
1111 APPLESTATIC int
1112 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1113 {
1114 	struct proc *p;
1115 
1116 	if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1117 		return (error);
1118 	if (td != NULL)
1119 		p = td->td_proc;
1120 	else
1121 		p = NULL;
1122 	switch (error) {
1123 	case NFSERR_BADOWNER:
1124 		tprintf(p, LOG_INFO,
1125 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1126 		    uid, gid);
1127 		return (EPERM);
1128 	case NFSERR_BADNAME:
1129 	case NFSERR_BADCHAR:
1130 		printf("nfsv4 char/name not handled by server\n");
1131 		return (ENOENT);
1132 	case NFSERR_STALECLIENTID:
1133 	case NFSERR_STALESTATEID:
1134 	case NFSERR_EXPIRED:
1135 	case NFSERR_BADSTATEID:
1136 	case NFSERR_BADSESSION:
1137 		printf("nfsv4 recover err returned %d\n", error);
1138 		return (EIO);
1139 	case NFSERR_BADHANDLE:
1140 	case NFSERR_SERVERFAULT:
1141 	case NFSERR_BADTYPE:
1142 	case NFSERR_FHEXPIRED:
1143 	case NFSERR_RESOURCE:
1144 	case NFSERR_MOVED:
1145 	case NFSERR_NOFILEHANDLE:
1146 	case NFSERR_MINORVERMISMATCH:
1147 	case NFSERR_OLDSTATEID:
1148 	case NFSERR_BADSEQID:
1149 	case NFSERR_LEASEMOVED:
1150 	case NFSERR_RECLAIMBAD:
1151 	case NFSERR_BADXDR:
1152 	case NFSERR_OPILLEGAL:
1153 		printf("nfsv4 client/server protocol prob err=%d\n",
1154 		    error);
1155 		return (EIO);
1156 	default:
1157 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1158 		return (EIO);
1159 	};
1160 }
1161 
1162 /*
1163  * Check to see if the process for this owner exists. Return 1 if it doesn't
1164  * and 0 otherwise.
1165  */
1166 int
1167 nfscl_procdoesntexist(u_int8_t *own)
1168 {
1169 	union {
1170 		u_int32_t	lval;
1171 		u_int8_t	cval[4];
1172 	} tl;
1173 	struct proc *p;
1174 	pid_t pid;
1175 	int i, ret = 0;
1176 
1177 	/* For the single open_owner of all 0 bytes, just return 0. */
1178 	for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1179 		if (own[i] != 0)
1180 			break;
1181 	if (i == NFSV4CL_LOCKNAMELEN)
1182 		return (0);
1183 
1184 	tl.cval[0] = *own++;
1185 	tl.cval[1] = *own++;
1186 	tl.cval[2] = *own++;
1187 	tl.cval[3] = *own++;
1188 	pid = tl.lval;
1189 	p = pfind_any_locked(pid);
1190 	if (p == NULL)
1191 		return (1);
1192 	if (p->p_stats == NULL) {
1193 		PROC_UNLOCK(p);
1194 		return (0);
1195 	}
1196 	tl.cval[0] = *own++;
1197 	tl.cval[1] = *own++;
1198 	tl.cval[2] = *own++;
1199 	tl.cval[3] = *own++;
1200 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1201 		ret = 1;
1202 	} else {
1203 		tl.cval[0] = *own++;
1204 		tl.cval[1] = *own++;
1205 		tl.cval[2] = *own++;
1206 		tl.cval[3] = *own;
1207 		if (tl.lval != p->p_stats->p_start.tv_usec)
1208 			ret = 1;
1209 	}
1210 	PROC_UNLOCK(p);
1211 	return (ret);
1212 }
1213 
1214 /*
1215  * - nfs pseudo system call for the client
1216  */
1217 /*
1218  * MPSAFE
1219  */
1220 static int
1221 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1222 {
1223 	struct file *fp;
1224 	struct nfscbd_args nfscbdarg;
1225 	struct nfsd_nfscbd_args nfscbdarg2;
1226 	struct nameidata nd;
1227 	struct nfscl_dumpmntopts dumpmntopts;
1228 	cap_rights_t rights;
1229 	char *buf;
1230 	int error;
1231 	struct mount *mp;
1232 	struct nfsmount *nmp;
1233 
1234 	if (uap->flag & NFSSVC_CBADDSOCK) {
1235 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1236 		if (error)
1237 			return (error);
1238 		/*
1239 		 * Since we don't know what rights might be required,
1240 		 * pretend that we need them all. It is better to be too
1241 		 * careful than too reckless.
1242 		 */
1243 		error = fget(td, nfscbdarg.sock,
1244 		    cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1245 		if (error)
1246 			return (error);
1247 		if (fp->f_type != DTYPE_SOCKET) {
1248 			fdrop(fp, td);
1249 			return (EPERM);
1250 		}
1251 		error = nfscbd_addsock(fp);
1252 		fdrop(fp, td);
1253 		if (!error && nfscl_enablecallb == 0) {
1254 			nfsv4_cbport = nfscbdarg.port;
1255 			nfscl_enablecallb = 1;
1256 		}
1257 	} else if (uap->flag & NFSSVC_NFSCBD) {
1258 		if (uap->argp == NULL)
1259 			return (EINVAL);
1260 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1261 		    sizeof(nfscbdarg2));
1262 		if (error)
1263 			return (error);
1264 		error = nfscbd_nfsd(td, &nfscbdarg2);
1265 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1266 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1267 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1268 		    dumpmntopts.ndmnt_blen > 1024))
1269 			error = EINVAL;
1270 		if (error == 0)
1271 			error = nfsrv_lookupfilename(&nd,
1272 			    dumpmntopts.ndmnt_fname, td);
1273 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1274 		    "nfs") != 0) {
1275 			vput(nd.ni_vp);
1276 			error = EINVAL;
1277 		}
1278 		if (error == 0) {
1279 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1280 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1281 			    dumpmntopts.ndmnt_blen);
1282 			vput(nd.ni_vp);
1283 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1284 			    dumpmntopts.ndmnt_blen);
1285 			free(buf, M_TEMP);
1286 		}
1287 	} else if (uap->flag & NFSSVC_FORCEDISM) {
1288 		buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
1289 		error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
1290 		if (error == 0) {
1291 			nmp = NULL;
1292 			mtx_lock(&mountlist_mtx);
1293 			TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1294 				if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
1295 				    0 && strcmp(mp->mnt_stat.f_fstypename,
1296 				    "nfs") == 0 && mp->mnt_data != NULL) {
1297 					nmp = VFSTONFS(mp);
1298 					NFSDDSLOCK();
1299 					if (nfsv4_findmirror(nmp) != NULL) {
1300 						NFSDDSUNLOCK();
1301 						error = ENXIO;
1302 						nmp = NULL;
1303 						break;
1304 					}
1305 					mtx_lock(&nmp->nm_mtx);
1306 					if ((nmp->nm_privflag &
1307 					    NFSMNTP_FORCEDISM) == 0) {
1308 						nmp->nm_privflag |=
1309 						   (NFSMNTP_FORCEDISM |
1310 						    NFSMNTP_CANCELRPCS);
1311 						mtx_unlock(&nmp->nm_mtx);
1312 					} else {
1313 						mtx_unlock(&nmp->nm_mtx);
1314 						nmp = NULL;
1315 					}
1316 					NFSDDSUNLOCK();
1317 					break;
1318 				}
1319 			}
1320 			mtx_unlock(&mountlist_mtx);
1321 
1322 			if (nmp != NULL) {
1323 				/*
1324 				 * Call newnfs_nmcancelreqs() to cause
1325 				 * any RPCs in progress on the mount point to
1326 				 * fail.
1327 				 * This will cause any process waiting for an
1328 				 * RPC to complete while holding a vnode lock
1329 				 * on the mounted-on vnode (such as "df" or
1330 				 * a non-forced "umount") to fail.
1331 				 * This will unlock the mounted-on vnode so
1332 				 * a forced dismount can succeed.
1333 				 * Then clear NFSMNTP_CANCELRPCS and wakeup(),
1334 				 * so that nfs_unmount() can complete.
1335 				 */
1336 				newnfs_nmcancelreqs(nmp);
1337 				mtx_lock(&nmp->nm_mtx);
1338 				nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
1339 				wakeup(nmp);
1340 				mtx_unlock(&nmp->nm_mtx);
1341 			} else if (error == 0)
1342 				error = EINVAL;
1343 		}
1344 		free(buf, M_TEMP);
1345 	} else {
1346 		error = EINVAL;
1347 	}
1348 	return (error);
1349 }
1350 
1351 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1352 
1353 /*
1354  * Called once to initialize data structures...
1355  */
1356 static int
1357 nfscl_modevent(module_t mod, int type, void *data)
1358 {
1359 	int error = 0;
1360 	static int loaded = 0;
1361 
1362 	switch (type) {
1363 	case MOD_LOAD:
1364 		if (loaded)
1365 			return (0);
1366 		newnfs_portinit();
1367 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1368 		nfscl_init();
1369 		NFSD_LOCK();
1370 		nfsrvd_cbinit(0);
1371 		NFSD_UNLOCK();
1372 		ncl_call_invalcaches = ncl_invalcaches;
1373 		nfsd_call_nfscl = nfssvc_nfscl;
1374 		loaded = 1;
1375 		break;
1376 
1377 	case MOD_UNLOAD:
1378 		if (nfs_numnfscbd != 0) {
1379 			error = EBUSY;
1380 			break;
1381 		}
1382 
1383 		/*
1384 		 * XXX: Unloading of nfscl module is unsupported.
1385 		 */
1386 #if 0
1387 		ncl_call_invalcaches = NULL;
1388 		nfsd_call_nfscl = NULL;
1389 		uma_zdestroy(ncl_pbuf_zone);
1390 		/* and get rid of the mutexes */
1391 		mtx_destroy(&ncl_iod_mutex);
1392 		loaded = 0;
1393 		break;
1394 #else
1395 		/* FALLTHROUGH */
1396 #endif
1397 	default:
1398 		error = EOPNOTSUPP;
1399 		break;
1400 	}
1401 	return error;
1402 }
1403 static moduledata_t nfscl_mod = {
1404 	"nfscl",
1405 	nfscl_modevent,
1406 	NULL,
1407 };
1408 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1409 
1410 /* So that loader and kldload(2) can find us, wherever we are.. */
1411 MODULE_VERSION(nfscl, 1);
1412 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1413 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1414 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1415 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1416 
1417