xref: /freebsd/sys/fs/nfsclient/nfs_clport.c (revision d9a42747950146bf03cda7f6e25d219253f8a57a)
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 	/*
144 	 * If this is the mount point fh and NFSMNTP_FAKEROOT is set, replace
145 	 * it with the fake fh.
146 	 */
147 	if ((nmp->nm_privflag & NFSMNTP_FAKEROOTFH) != 0 &&
148 	    nmp->nm_fhsize > 0 && nmp->nm_fhsize == nfhp->nfh_len &&
149 	    !NFSBCMP(nmp->nm_fh, nfhp->nfh_fh, nmp->nm_fhsize)) {
150 		free(nfhp, M_NFSFH);
151 		nfhp = malloc(sizeof(struct nfsfh) + NFSX_FHMAX + 1,
152 		    M_NFSFH, M_WAITOK | M_ZERO);
153 		nfhp->nfh_len = NFSX_FHMAX + 1;
154 	}
155 
156 	hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
157 
158 	error = vfs_hash_get(mntp, hash, lkflags,
159 	    td, &nvp, newnfs_vncmpf, nfhp);
160 	if (error == 0 && nvp != NULL) {
161 		/*
162 		 * I believe there is a slight chance that vgonel() could
163 		 * get called on this vnode between when NFSVOPLOCK() drops
164 		 * the VI_LOCK() and vget() acquires it again, so that it
165 		 * hasn't yet had v_usecount incremented. If this were to
166 		 * happen, the VIRF_DOOMED flag would be set, so check for
167 		 * that here. Since we now have the v_usecount incremented,
168 		 * we should be ok until we vrele() it, if the VIRF_DOOMED
169 		 * flag isn't set now.
170 		 */
171 		VI_LOCK(nvp);
172 		if (VN_IS_DOOMED(nvp)) {
173 			VI_UNLOCK(nvp);
174 			vrele(nvp);
175 			error = ENOENT;
176 		} else {
177 			VI_UNLOCK(nvp);
178 		}
179 	}
180 	if (error) {
181 		free(nfhp, M_NFSFH);
182 		return (error);
183 	}
184 	if (nvp != NULL) {
185 		np = VTONFS(nvp);
186 		/*
187 		 * For NFSv4, check to see if it is the same name and
188 		 * replace the name, if it is different.
189 		 */
190 		oldd = newd = NULL;
191 		if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
192 		    nvp->v_type == VREG &&
193 		    (np->n_v4->n4_namelen != cnp->cn_namelen ||
194 		     NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
195 		     cnp->cn_namelen) ||
196 		     dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
197 		     NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
198 		     dnp->n_fhp->nfh_len))) {
199 		    newd = malloc(
200 			sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
201 			+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
202 		    NFSLOCKNODE(np);
203 		    if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
204 			&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
205 			 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
206 			 cnp->cn_namelen) ||
207 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
208 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
209 			 dnp->n_fhp->nfh_len))) {
210 			oldd = np->n_v4;
211 			np->n_v4 = newd;
212 			newd = NULL;
213 			np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
214 			np->n_v4->n4_namelen = cnp->cn_namelen;
215 			NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
216 			    dnp->n_fhp->nfh_len);
217 			NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
218 			    cnp->cn_namelen);
219 		    }
220 		    NFSUNLOCKNODE(np);
221 		}
222 		if (newd != NULL)
223 			free(newd, M_NFSV4NODE);
224 		if (oldd != NULL)
225 			free(oldd, M_NFSV4NODE);
226 		*npp = np;
227 		free(nfhp, M_NFSFH);
228 		return (0);
229 	}
230 	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
231 
232 	error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
233 	if (error) {
234 		uma_zfree(newnfsnode_zone, np);
235 		free(nfhp, M_NFSFH);
236 		return (error);
237 	}
238 	vp = nvp;
239 	KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
240 	vp->v_bufobj.bo_ops = &buf_ops_newnfs;
241 	vp->v_data = np;
242 	np->n_vnode = vp;
243 	/*
244 	 * Initialize the mutex even if the vnode is going to be a loser.
245 	 * This simplifies the logic in reclaim, which can then unconditionally
246 	 * destroy the mutex (in the case of the loser, or if hash_insert
247 	 * happened to return an error no special casing is needed).
248 	 */
249 	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
250 	lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE |
251 	    LK_CANRECURSE);
252 
253 	/*
254 	 * Are we getting the root? If so, make sure the vnode flags
255 	 * are correct
256 	 */
257 	if (nfhp->nfh_len == NFSX_FHMAX + 1 ||
258 	    (nfhp->nfh_len == nmp->nm_fhsize &&
259 	     !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len))) {
260 		if (vp->v_type == VNON)
261 			vp->v_type = VDIR;
262 		vp->v_vflag |= VV_ROOT;
263 	}
264 
265 	vp->v_vflag |= VV_VMSIZEVNLOCK;
266 
267 	np->n_fhp = nfhp;
268 	/*
269 	 * For NFSv4, we have to attach the directory file handle and
270 	 * file name, so that Open Ops can be done later.
271 	 */
272 	if (nmp->nm_flag & NFSMNT_NFSV4) {
273 		np->n_v4 = malloc(sizeof (struct nfsv4node)
274 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
275 		    M_WAITOK);
276 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
277 		np->n_v4->n4_namelen = cnp->cn_namelen;
278 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
279 		    dnp->n_fhp->nfh_len);
280 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
281 		    cnp->cn_namelen);
282 	} else {
283 		np->n_v4 = NULL;
284 	}
285 
286 	/*
287 	 * NFS supports recursive and shared locking.
288 	 */
289 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
290 	VN_LOCK_AREC(vp);
291 	VN_LOCK_ASHARE(vp);
292 	error = insmntque(vp, mntp);
293 	if (error != 0) {
294 		*npp = NULL;
295 		mtx_destroy(&np->n_mtx);
296 		lockdestroy(&np->n_excl);
297 		free(nfhp, M_NFSFH);
298 		if (np->n_v4 != NULL)
299 			free(np->n_v4, M_NFSV4NODE);
300 		uma_zfree(newnfsnode_zone, np);
301 		return (error);
302 	}
303 	vn_set_state(vp, VSTATE_CONSTRUCTED);
304 	error = vfs_hash_insert(vp, hash, lkflags,
305 	    td, &nvp, newnfs_vncmpf, nfhp);
306 	if (error)
307 		return (error);
308 	if (nvp != NULL) {
309 		*npp = VTONFS(nvp);
310 		/* vfs_hash_insert() vput()'s the losing vnode */
311 		return (0);
312 	}
313 	*npp = np;
314 
315 	return (0);
316 }
317 
318 /*
319  * Another variant of nfs_nget(). This one is only used by reopen. It
320  * takes almost the same args as nfs_nget(), but only succeeds if an entry
321  * exists in the cache. (Since files should already be "open" with a
322  * vnode ref cnt on the node when reopen calls this, it should always
323  * succeed.)
324  * Also, don't get a vnode lock, since it may already be locked by some
325  * other process that is handling it. This is ok, since all other threads
326  * on the client are blocked by the nfsc_lock being exclusively held by the
327  * caller of this function.
328  */
329 int
330 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
331     struct thread *td, struct nfsnode **npp)
332 {
333 	struct vnode *nvp;
334 	u_int hash;
335 	struct nfsfh *nfhp;
336 	int error;
337 
338 	*npp = NULL;
339 	/* For forced dismounts, just return error. */
340 	if (NFSCL_FORCEDISM(mntp))
341 		return (EINTR);
342 	nfhp = malloc(sizeof (struct nfsfh) + fhsize,
343 	    M_NFSFH, M_WAITOK);
344 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
345 	nfhp->nfh_len = fhsize;
346 
347 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
348 
349 	/*
350 	 * First, try to get the vnode locked, but don't block for the lock.
351 	 */
352 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
353 	    newnfs_vncmpf, nfhp);
354 	if (error == 0 && nvp != NULL) {
355 		NFSVOPUNLOCK(nvp);
356 	} else if (error == EBUSY) {
357 		/*
358 		 * It is safe so long as a vflush() with
359 		 * FORCECLOSE has not been done. Since the Renew thread is
360 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
361 		 * a vflush() with FORCECLOSE, we should be ok here.
362 		 */
363 		if (NFSCL_FORCEDISM(mntp))
364 			error = EINTR;
365 		else {
366 			vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
367 			if (nvp == NULL) {
368 				error = ENOENT;
369 			} else if (VN_IS_DOOMED(nvp)) {
370 				error = ENOENT;
371 				vrele(nvp);
372 			} else {
373 				error = 0;
374 			}
375 		}
376 	}
377 	free(nfhp, M_NFSFH);
378 	if (error)
379 		return (error);
380 	if (nvp != NULL) {
381 		*npp = VTONFS(nvp);
382 		return (0);
383 	}
384 	return (EINVAL);
385 }
386 
387 static void
388 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
389     struct nfsvattr *newnap)
390 {
391 	int off;
392 
393 	if (ncl_fileid_maxwarnings >= 0 &&
394 	    ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
395 		return;
396 	off = 0;
397 	if (ncl_fileid_maxwarnings >= 0) {
398 		if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
399 			off = 1;
400 	}
401 
402 	printf("newnfs: server '%s' error: fileid changed. "
403 	    "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
404 	    "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
405 	    nmp->nm_com.nmcom_hostname,
406 	    (uintmax_t)nmp->nm_fsid[0],
407 	    (uintmax_t)nmp->nm_fsid[1],
408 	    (uintmax_t)oldnap->na_fileid,
409 	    (uintmax_t)newnap->na_fileid);
410 
411 	if (off)
412 		printf("newnfs: Logged %d times about fileid corruption; "
413 		    "going quiet to avoid spamming logs excessively. (Limit "
414 		    "is: %d).\n", ncl_fileid_nwarnings,
415 		    ncl_fileid_maxwarnings);
416 }
417 
418 void
419 ncl_copy_vattr(struct vattr *dst, struct vattr *src)
420 {
421 	dst->va_type = src->va_type;
422 	dst->va_mode = src->va_mode;
423 	dst->va_nlink = src->va_nlink;
424 	dst->va_uid = src->va_uid;
425 	dst->va_gid = src->va_gid;
426 	dst->va_fsid = src->va_fsid;
427 	dst->va_fileid = src->va_fileid;
428 	dst->va_size = src->va_size;
429 	dst->va_blocksize = src->va_blocksize;
430 	dst->va_atime = src->va_atime;
431 	dst->va_mtime = src->va_mtime;
432 	dst->va_ctime = src->va_ctime;
433 	dst->va_birthtime = src->va_birthtime;
434 	dst->va_gen = src->va_gen;
435 	dst->va_flags = src->va_flags;
436 	dst->va_rdev = src->va_rdev;
437 	dst->va_bytes = src->va_bytes;
438 	dst->va_filerev = src->va_filerev;
439 }
440 
441 /*
442  * Load the attribute cache (that lives in the nfsnode entry) with
443  * the attributes of the second argument and
444  * Iff vaper not NULL
445  *    copy the attributes to *vaper
446  * Similar to nfs_loadattrcache(), except the attributes are passed in
447  * instead of being parsed out of the mbuf list.
448  */
449 int
450 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
451     int writeattr, int dontshrink)
452 {
453 	struct vnode *vp = *vpp;
454 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
455 	struct nfsnode *np;
456 	struct nfsmount *nmp;
457 	struct timespec mtime_save;
458 	int error, force_fid_err;
459 	dev_t topfsid;
460 
461 	error = 0;
462 
463 	/*
464 	 * If v_type == VNON it is a new node, so fill in the v_type,
465 	 * n_mtime fields. Check to see if it represents a special
466 	 * device, and if so, check for a possible alias. Once the
467 	 * correct vnode has been obtained, fill in the rest of the
468 	 * information.
469 	 */
470 	np = VTONFS(vp);
471 	NFSLOCKNODE(np);
472 	if (vp->v_type != nvap->va_type) {
473 		vp->v_type = nvap->va_type;
474 		if (vp->v_type == VFIFO)
475 			vp->v_op = &newnfs_fifoops;
476 		np->n_mtime = nvap->va_mtime;
477 	}
478 	nmp = VFSTONFS(vp->v_mount);
479 	vap = &np->n_vattr.na_vattr;
480 	mtime_save = vap->va_mtime;
481 	if (writeattr) {
482 		np->n_vattr.na_filerev = nap->na_filerev;
483 		np->n_vattr.na_size = nap->na_size;
484 		np->n_vattr.na_mtime = nap->na_mtime;
485 		np->n_vattr.na_ctime = nap->na_ctime;
486 		np->n_vattr.na_btime = nap->na_btime;
487 		np->n_vattr.na_fsid = nap->na_fsid;
488 		np->n_vattr.na_mode = nap->na_mode;
489 	} else {
490 		force_fid_err = 0;
491 		KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
492 		    force_fid_err);
493 		/*
494 		 * BROKEN NFS SERVER OR MIDDLEWARE
495 		 *
496 		 * Certain NFS servers (certain old proprietary filers ca.
497 		 * 2006) or broken middleboxes (e.g. WAN accelerator products)
498 		 * will respond to GETATTR requests with results for a
499 		 * different fileid.
500 		 *
501 		 * The WAN accelerator we've observed not only serves stale
502 		 * cache results for a given file, it also occasionally serves
503 		 * results for wholly different files.  This causes surprising
504 		 * problems; for example the cached size attribute of a file
505 		 * may truncate down and then back up, resulting in zero
506 		 * regions in file contents read by applications.  We observed
507 		 * this reliably with Clang and .c files during parallel build.
508 		 * A pcap revealed packet fragmentation and GETATTR RPC
509 		 * responses with wholly wrong fileids.
510 		 * For the case where the file handle is a fake one
511 		 * generated via the "syskrb5" mount option and
512 		 * the old fileid is 2, ignore the test, since this might
513 		 * be replacing the fake attributes with correct ones.
514 		 */
515 		if ((np->n_vattr.na_fileid != 0 &&
516 		     np->n_vattr.na_fileid != nap->na_fileid &&
517 		     (np->n_vattr.na_fileid != 2 || !NFSHASSYSKRB5(nmp) ||
518 		      np->n_fhp->nfh_len != NFSX_FHMAX + 1)) ||
519 		    force_fid_err) {
520 			nfscl_warn_fileid(nmp, &np->n_vattr, nap);
521 			error = EIDRM;
522 			goto out;
523 		}
524 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
525 		    sizeof (struct nfsvattr));
526 	}
527 
528 	/*
529 	 * For NFSv4, the server's export may be a tree of file systems
530 	 * where a fileno is a unique value within each file system.
531 	 * na_filesid[0,1] uniquely identify the server file system
532 	 * and nm_fsid[0,1] is the value for the root file system mounted.
533 	 * As such, the value of va_fsid generated by vn_fsid() represents
534 	 * the root file system on the server and a different value for
535 	 * va_fsid is needed for the other server file systems.  This
536 	 * va_fsid is ideally unique for all of the server file systems,
537 	 * so a 64bit hash on na_filesid[0,1] is calculated.
538 	 * Although highly unlikely that the fnv_64_hash() will be
539 	 * the same as the root, test for this case and recalculate the hash.
540 	 */
541 	vn_fsid(vp, vap);
542 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
543 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
544 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
545 		topfsid = vap->va_fsid;
546 		vap->va_fsid = FNV1_64_INIT;
547 		do {
548 			vap->va_fsid = fnv_64_buf(np->n_vattr.na_filesid,
549 			    sizeof(np->n_vattr.na_filesid), vap->va_fsid);
550 		} while (vap->va_fsid == topfsid);
551 	}
552 
553 	np->n_attrstamp = time_second;
554 	if (vap->va_size != np->n_size) {
555 		if (vap->va_type == VREG) {
556 			if (dontshrink && vap->va_size < np->n_size) {
557 				/*
558 				 * We've been told not to shrink the file;
559 				 * zero np->n_attrstamp to indicate that
560 				 * the attributes are stale.
561 				 */
562 				vap->va_size = np->n_size;
563 				np->n_attrstamp = 0;
564 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
565 			} else if (np->n_flag & NMODIFIED) {
566 				/*
567 				 * We've modified the file: Use the larger
568 				 * of our size, and the server's size.
569 				 */
570 				if (vap->va_size < np->n_size) {
571 					vap->va_size = np->n_size;
572 				} else {
573 					np->n_size = vap->va_size;
574 					np->n_flag |= NSIZECHANGED;
575 				}
576 			} else {
577 				np->n_size = vap->va_size;
578 				np->n_flag |= NSIZECHANGED;
579 			}
580 		} else {
581 			np->n_size = vap->va_size;
582 		}
583 	}
584 	/*
585 	 * The following checks are added to prevent a race between (say)
586 	 * a READDIR+ and a WRITE.
587 	 * READDIR+, WRITE requests sent out.
588 	 * READDIR+ resp, WRITE resp received on client.
589 	 * However, the WRITE resp was handled before the READDIR+ resp
590 	 * causing the post op attrs from the write to be loaded first
591 	 * and the attrs from the READDIR+ to be loaded later. If this
592 	 * happens, we have stale attrs loaded into the attrcache.
593 	 * We detect this by for the mtime moving back. We invalidate the
594 	 * attrcache when this happens.
595 	 */
596 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
597 		/* Size changed or mtime went backwards */
598 		np->n_attrstamp = 0;
599 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
600 	}
601 	if (vaper != NULL) {
602 		ncl_copy_vattr(vaper, vap);
603 		if (np->n_flag & NCHG) {
604 			if (np->n_flag & NACC)
605 				vaper->va_atime = np->n_atim;
606 			if (np->n_flag & NUPD)
607 				vaper->va_mtime = np->n_mtim;
608 		}
609 	}
610 
611 out:
612 #ifdef KDTRACE_HOOKS
613 	if (np->n_attrstamp != 0)
614 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
615 #endif
616 	(void)ncl_pager_setsize(vp, NULL);
617 	return (error);
618 }
619 
620 /*
621  * Call vnode_pager_setsize() if the size of the node changed, as
622  * recorded in nfsnode vs. v_object, or delay the call if notifying
623  * the pager is not possible at the moment.
624  *
625  * If nsizep is non-NULL, the call is delayed and the new node size is
626  * provided.  Caller should itself call vnode_pager_setsize() if
627  * function returned true.  If nsizep is NULL, function tries to call
628  * vnode_pager_setsize() itself if needed and possible, and the nfs
629  * node is unlocked unconditionally, the return value is not useful.
630  */
631 bool
632 ncl_pager_setsize(struct vnode *vp, u_quad_t *nsizep)
633 {
634 	struct nfsnode *np;
635 	vm_object_t object;
636 	struct vattr *vap;
637 	u_quad_t nsize;
638 	bool setnsize;
639 
640 	np = VTONFS(vp);
641 	NFSASSERTNODE(np);
642 
643 	vap = &np->n_vattr.na_vattr;
644 	nsize = vap->va_size;
645 	object = vp->v_object;
646 	setnsize = false;
647 
648 	if (object != NULL && nsize != object->un_pager.vnp.vnp_size) {
649 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
650 		    (curthread->td_pflags2 & TDP2_SBPAGES) == 0)
651 			setnsize = true;
652 		else
653 			np->n_flag |= NVNSETSZSKIP;
654 	}
655 	if (nsizep == NULL) {
656 		NFSUNLOCKNODE(np);
657 		if (setnsize)
658 			vnode_pager_setsize(vp, nsize);
659 		setnsize = false;
660 	} else {
661 		*nsizep = nsize;
662 	}
663 	return (setnsize);
664 }
665 
666 /*
667  * Fill in the client id name. For these bytes:
668  * 1 - they must be unique
669  * 2 - they should be persistent across client reboots
670  * 1 is more critical than 2
671  * Use the mount point's unique id plus either the uuid or, if that
672  * isn't set, random junk.
673  */
674 void
675 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
676 {
677 	int uuidlen;
678 
679 	/*
680 	 * First, put in the 64bit mount point identifier.
681 	 */
682 	if (idlen >= sizeof (u_int64_t)) {
683 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
684 		cp += sizeof (u_int64_t);
685 		idlen -= sizeof (u_int64_t);
686 	}
687 
688 	/*
689 	 * If uuid is non-zero length, use it.
690 	 */
691 	uuidlen = strlen(uuid);
692 	if (uuidlen > 0 && idlen >= uuidlen) {
693 		NFSBCOPY(uuid, cp, uuidlen);
694 		cp += uuidlen;
695 		idlen -= uuidlen;
696 	}
697 
698 	/*
699 	 * This only normally happens if the uuid isn't set.
700 	 */
701 	while (idlen > 0) {
702 		*cp++ = (u_int8_t)(arc4random() % 256);
703 		idlen--;
704 	}
705 }
706 
707 /*
708  * Fill in a lock owner name. For now, pid + the process's creation time.
709  */
710 void
711 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
712 {
713 	union {
714 		u_int32_t	lval;
715 		u_int8_t	cval[4];
716 	} tl;
717 	struct proc *p;
718 
719 	if (id == NULL) {
720 		/* Return the single open_owner of all 0 bytes. */
721 		bzero(cp, NFSV4CL_LOCKNAMELEN);
722 		return;
723 	}
724 	if ((flags & F_POSIX) != 0) {
725 		p = (struct proc *)id;
726 		tl.lval = p->p_pid;
727 		*cp++ = tl.cval[0];
728 		*cp++ = tl.cval[1];
729 		*cp++ = tl.cval[2];
730 		*cp++ = tl.cval[3];
731 		tl.lval = p->p_stats->p_start.tv_sec;
732 		*cp++ = tl.cval[0];
733 		*cp++ = tl.cval[1];
734 		*cp++ = tl.cval[2];
735 		*cp++ = tl.cval[3];
736 		tl.lval = p->p_stats->p_start.tv_usec;
737 		*cp++ = tl.cval[0];
738 		*cp++ = tl.cval[1];
739 		*cp++ = tl.cval[2];
740 		*cp = tl.cval[3];
741 	} else if ((flags & F_FLOCK) != 0) {
742 		bcopy(&id, cp, sizeof(id));
743 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
744 	} else {
745 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
746 		bzero(cp, NFSV4CL_LOCKNAMELEN);
747 	}
748 }
749 
750 /*
751  * Find the parent process for the thread passed in as an argument.
752  * If none exists, return NULL, otherwise return a thread for the parent.
753  * (Can be any of the threads, since it is only used for td->td_proc.)
754  */
755 NFSPROC_T *
756 nfscl_getparent(struct thread *td)
757 {
758 	struct proc *p;
759 	struct thread *ptd;
760 
761 	if (td == NULL)
762 		return (NULL);
763 	p = td->td_proc;
764 	if (p->p_pid == 0)
765 		return (NULL);
766 	p = p->p_pptr;
767 	if (p == NULL)
768 		return (NULL);
769 	ptd = TAILQ_FIRST(&p->p_threads);
770 	return (ptd);
771 }
772 
773 /*
774  * Start up the renew kernel thread.
775  */
776 static void
777 start_nfscl(void *arg)
778 {
779 	struct nfsclclient *clp;
780 	struct thread *td;
781 
782 	clp = (struct nfsclclient *)arg;
783 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
784 	nfscl_renewthread(clp, td);
785 	kproc_exit(0);
786 }
787 
788 void
789 nfscl_start_renewthread(struct nfsclclient *clp)
790 {
791 
792 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
793 	    "nfscl");
794 }
795 
796 /*
797  * Handle wcc_data.
798  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
799  * as the first Op after PutFH.
800  * (For NFSv4, the postop attributes are after the Op, so they can't be
801  *  parsed here. A separate call to nfscl_postop_attr() is required.)
802  */
803 int
804 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
805     struct nfsvattr *nap, int *flagp, int *wccflagp, uint64_t *repsizep)
806 {
807 	u_int32_t *tl;
808 	struct nfsnode *np = VTONFS(vp);
809 	struct nfsvattr nfsva;
810 	int error = 0;
811 
812 	if (wccflagp != NULL)
813 		*wccflagp = 0;
814 	if (nd->nd_flag & ND_NFSV3) {
815 		*flagp = 0;
816 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
817 		if (*tl == newnfs_true) {
818 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
819 			if (wccflagp != NULL) {
820 				NFSLOCKNODE(np);
821 				*wccflagp = (np->n_mtime.tv_sec ==
822 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
823 				    np->n_mtime.tv_nsec ==
824 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
825 				NFSUNLOCKNODE(np);
826 			}
827 		}
828 		error = nfscl_postop_attr(nd, nap, flagp);
829 		if (wccflagp != NULL && *flagp == 0)
830 			*wccflagp = 0;
831 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
832 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
833 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
834 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
835 		    NULL, NULL, NULL, NULL, NULL);
836 		if (error)
837 			return (error);
838 		/*
839 		 * Get rid of Op# and status for next op.
840 		 */
841 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
842 		if (*++tl)
843 			nd->nd_flag |= ND_NOMOREDATA;
844 		if (repsizep != NULL)
845 			*repsizep = nfsva.na_size;
846 		if (wccflagp != NULL &&
847 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
848 			NFSLOCKNODE(np);
849 			*wccflagp = (np->n_mtime.tv_sec ==
850 			    nfsva.na_vattr.va_mtime.tv_sec &&
851 			    np->n_mtime.tv_nsec ==
852 			    nfsva.na_vattr.va_mtime.tv_sec);
853 			NFSUNLOCKNODE(np);
854 		}
855 	}
856 nfsmout:
857 	return (error);
858 }
859 
860 /*
861  * Get postop attributes.
862  */
863 int
864 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp)
865 {
866 	u_int32_t *tl;
867 	int error = 0;
868 
869 	*retp = 0;
870 	if (nd->nd_flag & ND_NOMOREDATA)
871 		return (error);
872 	if (nd->nd_flag & ND_NFSV3) {
873 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
874 		*retp = fxdr_unsigned(int, *tl);
875 	} else if (nd->nd_flag & ND_NFSV4) {
876 		/*
877 		 * For NFSv4, the postop attr are at the end, so no point
878 		 * in looking if nd_repstat != 0.
879 		 */
880 		if (!nd->nd_repstat) {
881 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
882 			if (*(tl + 1))
883 				/* should never happen since nd_repstat != 0 */
884 				nd->nd_flag |= ND_NOMOREDATA;
885 			else
886 				*retp = 1;
887 		}
888 	} else if (!nd->nd_repstat) {
889 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
890 		*retp = 1;
891 	}
892 	if (*retp) {
893 		error = nfsm_loadattr(nd, nap);
894 		if (error)
895 			*retp = 0;
896 	}
897 nfsmout:
898 	return (error);
899 }
900 
901 /*
902  * nfscl_request() - mostly a wrapper for newnfs_request().
903  */
904 int
905 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
906     struct ucred *cred)
907 {
908 	int ret, vers;
909 	struct nfsmount *nmp;
910 
911 	nmp = VFSTONFS(vp->v_mount);
912 	if (nd->nd_flag & ND_NFSV4)
913 		vers = NFS_VER4;
914 	else if (nd->nd_flag & ND_NFSV3)
915 		vers = NFS_VER3;
916 	else
917 		vers = NFS_VER2;
918 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
919 		NFS_PROG, vers, NULL, 1, NULL, NULL);
920 	return (ret);
921 }
922 
923 /*
924  * fill in this bsden's variant of statfs using nfsstatfs.
925  */
926 void
927 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
928 {
929 	struct statfs *sbp = (struct statfs *)statfs;
930 
931 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
932 		sbp->f_bsize = NFS_FABLKSIZE;
933 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
934 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
935 		/*
936 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
937 		 * the value after dividing by NFS_FABLKSIZE is small
938 		 * enough that it will fit in 63bits, so it is ok to
939 		 * assign it to f_bavail without fear that it will become
940 		 * negative.
941 		 */
942 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
943 		sbp->f_files = sfp->sf_tfiles;
944 		/* Since f_ffree is int64_t, clip it to 63bits. */
945 		if (sfp->sf_ffiles > INT64_MAX)
946 			sbp->f_ffree = INT64_MAX;
947 		else
948 			sbp->f_ffree = sfp->sf_ffiles;
949 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
950 		/*
951 		 * The type casts to (int32_t) ensure that this code is
952 		 * compatible with the old NFS client, in that it will
953 		 * propagate bit31 to the high order bits. This may or may
954 		 * not be correct for NFSv2, but since it is a legacy
955 		 * environment, I'd rather retain backwards compatibility.
956 		 */
957 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
958 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
959 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
960 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
961 		sbp->f_files = 0;
962 		sbp->f_ffree = 0;
963 	}
964 }
965 
966 /*
967  * Use the fsinfo stuff to update the mount point.
968  */
969 void
970 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
971 {
972 
973 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
974 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
975 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
976 		    ~(NFS_FABLKSIZE - 1);
977 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
978 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
979 		if (nmp->nm_wsize == 0)
980 			nmp->nm_wsize = fsp->fs_wtmax;
981 	}
982 	if (nmp->nm_wsize < NFS_FABLKSIZE)
983 		nmp->nm_wsize = NFS_FABLKSIZE;
984 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
985 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
986 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
987 		    ~(NFS_FABLKSIZE - 1);
988 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
989 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
990 		if (nmp->nm_rsize == 0)
991 			nmp->nm_rsize = fsp->fs_rtmax;
992 	}
993 	if (nmp->nm_rsize < NFS_FABLKSIZE)
994 		nmp->nm_rsize = NFS_FABLKSIZE;
995 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
996 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
997 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
998 		    ~(NFS_DIRBLKSIZ - 1);
999 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
1000 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
1001 		if (nmp->nm_readdirsize == 0)
1002 			nmp->nm_readdirsize = fsp->fs_rtmax;
1003 	}
1004 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
1005 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
1006 	if (fsp->fs_maxfilesize > 0 &&
1007 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
1008 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
1009 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
1010 	nmp->nm_state |= NFSSTA_GOTFSINFO;
1011 }
1012 
1013 /*
1014  * Lookups source address which should be used to communicate with
1015  * @nmp and stores it inside @pdst.
1016  *
1017  * Returns 0 on success.
1018  */
1019 u_int8_t *
1020 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
1021 {
1022 #if defined(INET6) || defined(INET)
1023 	int fibnum;
1024 
1025 	fibnum = curthread->td_proc->p_fibnum;
1026 #endif
1027 #ifdef INET
1028 	if (nmp->nm_nam->sa_family == AF_INET) {
1029 		struct epoch_tracker et;
1030 		struct nhop_object *nh;
1031 		struct sockaddr_in *sin;
1032 		struct in_addr addr = {};
1033 
1034 		sin = (struct sockaddr_in *)nmp->nm_nam;
1035 		NET_EPOCH_ENTER(et);
1036 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1037 		nh = fib4_lookup(fibnum, sin->sin_addr, 0, NHR_NONE, 0);
1038 		if (nh != NULL) {
1039 			addr = IA_SIN(ifatoia(nh->nh_ifa))->sin_addr;
1040 			if (IN_LOOPBACK(ntohl(addr.s_addr))) {
1041 				/* Ignore loopback addresses */
1042 				nh = NULL;
1043 			}
1044 		}
1045 		CURVNET_RESTORE();
1046 		NET_EPOCH_EXIT(et);
1047 
1048 		if (nh == NULL)
1049 			return (NULL);
1050 		*isinet6p = 0;
1051 		*((struct in_addr *)paddr) = addr;
1052 
1053 		return (u_int8_t *)paddr;
1054 	}
1055 #endif
1056 #ifdef INET6
1057 	if (nmp->nm_nam->sa_family == AF_INET6) {
1058 		struct epoch_tracker et;
1059 		struct sockaddr_in6 *sin6;
1060 		int error;
1061 
1062 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1063 
1064 		NET_EPOCH_ENTER(et);
1065 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1066 		error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1067 		    sin6->sin6_scope_id, NULL, paddr, NULL);
1068 		CURVNET_RESTORE();
1069 		NET_EPOCH_EXIT(et);
1070 		if (error != 0)
1071 			return (NULL);
1072 
1073 		if (IN6_IS_ADDR_LOOPBACK(paddr))
1074 			return (NULL);
1075 
1076 		/* Scope is embedded in */
1077 		*isinet6p = 1;
1078 
1079 		return (u_int8_t *)paddr;
1080 	}
1081 #endif
1082 	return (NULL);
1083 }
1084 
1085 /*
1086  * Copy NFS uid, gids from the cred structure.
1087  */
1088 void
1089 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1090 {
1091 	int i;
1092 
1093 	KASSERT(cr->cr_ngroups >= 0,
1094 	    ("newnfs_copyincred: negative cr_ngroups"));
1095 	nfscr->nfsc_uid = cr->cr_uid;
1096 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1097 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1098 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1099 }
1100 
1101 /*
1102  * Do any client specific initialization.
1103  */
1104 void
1105 nfscl_init(void)
1106 {
1107 	static int inited = 0;
1108 
1109 	if (inited)
1110 		return;
1111 	inited = 1;
1112 	nfscl_inited = 1;
1113 	ncl_pbuf_zone = pbuf_zsecond_create("nfspbuf", nswbuf / 2);
1114 }
1115 
1116 /*
1117  * Check each of the attributes to be set, to ensure they aren't already
1118  * the correct value. Disable setting ones already correct.
1119  */
1120 int
1121 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1122 {
1123 
1124 	if (vap->va_mode != (mode_t)VNOVAL) {
1125 		if (vap->va_mode == nvap->na_mode)
1126 			vap->va_mode = (mode_t)VNOVAL;
1127 	}
1128 	if (vap->va_uid != (uid_t)VNOVAL) {
1129 		if (vap->va_uid == nvap->na_uid)
1130 			vap->va_uid = (uid_t)VNOVAL;
1131 	}
1132 	if (vap->va_gid != (gid_t)VNOVAL) {
1133 		if (vap->va_gid == nvap->na_gid)
1134 			vap->va_gid = (gid_t)VNOVAL;
1135 	}
1136 	if (vap->va_size != VNOVAL) {
1137 		if (vap->va_size == nvap->na_size)
1138 			vap->va_size = VNOVAL;
1139 	}
1140 
1141 	/*
1142 	 * We are normally called with only a partially initialized
1143 	 * VAP.  Since the NFSv3 spec says that server may use the
1144 	 * file attributes to store the verifier, the spec requires
1145 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1146 	 * in atime, but we can't really assume that all servers will
1147 	 * so we ensure that our SETATTR sets both atime and mtime.
1148 	 * Set the VA_UTIMES_NULL flag for this case, so that
1149 	 * the server's time will be used.  This is needed to
1150 	 * work around a bug in some Solaris servers, where
1151 	 * setting the time TOCLIENT causes the Setattr RPC
1152 	 * to return NFS_OK, but not set va_mode.
1153 	 */
1154 	if (vap->va_mtime.tv_sec == VNOVAL) {
1155 		vfs_timestamp(&vap->va_mtime);
1156 		vap->va_vaflags |= VA_UTIMES_NULL;
1157 	}
1158 	if (vap->va_atime.tv_sec == VNOVAL)
1159 		vap->va_atime = vap->va_mtime;
1160 	return (1);
1161 }
1162 
1163 /*
1164  * Map nfsv4 errors to errno.h errors.
1165  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1166  * error should only be returned for the Open, Create and Setattr Ops.
1167  * As such, most calls can just pass in 0 for those arguments.
1168  */
1169 int
1170 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1171 {
1172 	struct proc *p;
1173 
1174 	if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1175 		return (error);
1176 	if (td != NULL)
1177 		p = td->td_proc;
1178 	else
1179 		p = NULL;
1180 	switch (error) {
1181 	case NFSERR_BADOWNER:
1182 		tprintf(p, LOG_INFO,
1183 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1184 		    uid, gid);
1185 		return (EPERM);
1186 	case NFSERR_BADNAME:
1187 	case NFSERR_BADCHAR:
1188 		printf("nfsv4 char/name not handled by server\n");
1189 		return (ENOENT);
1190 	case NFSERR_STALECLIENTID:
1191 	case NFSERR_STALESTATEID:
1192 	case NFSERR_EXPIRED:
1193 	case NFSERR_BADSTATEID:
1194 	case NFSERR_BADSESSION:
1195 		printf("nfsv4 recover err returned %d\n", error);
1196 		return (EIO);
1197 	case NFSERR_BADHANDLE:
1198 	case NFSERR_SERVERFAULT:
1199 	case NFSERR_BADTYPE:
1200 	case NFSERR_FHEXPIRED:
1201 	case NFSERR_RESOURCE:
1202 	case NFSERR_MOVED:
1203 	case NFSERR_MINORVERMISMATCH:
1204 	case NFSERR_OLDSTATEID:
1205 	case NFSERR_BADSEQID:
1206 	case NFSERR_LEASEMOVED:
1207 	case NFSERR_RECLAIMBAD:
1208 	case NFSERR_BADXDR:
1209 	case NFSERR_OPILLEGAL:
1210 		printf("nfsv4 client/server protocol prob err=%d\n",
1211 		    error);
1212 		return (EIO);
1213 	case NFSERR_NOFILEHANDLE:
1214 		printf("nfsv4 no file handle: usually means the file "
1215 		    "system is not exported on the NFSv4 server\n");
1216 		return (EIO);
1217 	case NFSERR_WRONGSEC:
1218 		tprintf(p, LOG_INFO, "NFSv4 error WrongSec: You probably need a"
1219 		    " Kerberos TGT\n");
1220 		return (EIO);
1221 	default:
1222 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1223 		return (EIO);
1224 	};
1225 }
1226 
1227 /*
1228  * Check to see if the process for this owner exists. Return 1 if it doesn't
1229  * and 0 otherwise.
1230  */
1231 int
1232 nfscl_procdoesntexist(u_int8_t *own)
1233 {
1234 	union {
1235 		u_int32_t	lval;
1236 		u_int8_t	cval[4];
1237 	} tl;
1238 	struct proc *p;
1239 	pid_t pid;
1240 	int i, ret = 0;
1241 
1242 	/* For the single open_owner of all 0 bytes, just return 0. */
1243 	for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1244 		if (own[i] != 0)
1245 			break;
1246 	if (i == NFSV4CL_LOCKNAMELEN)
1247 		return (0);
1248 
1249 	tl.cval[0] = *own++;
1250 	tl.cval[1] = *own++;
1251 	tl.cval[2] = *own++;
1252 	tl.cval[3] = *own++;
1253 	pid = tl.lval;
1254 	p = pfind_any_locked(pid);
1255 	if (p == NULL)
1256 		return (1);
1257 	if (p->p_stats == NULL) {
1258 		PROC_UNLOCK(p);
1259 		return (0);
1260 	}
1261 	tl.cval[0] = *own++;
1262 	tl.cval[1] = *own++;
1263 	tl.cval[2] = *own++;
1264 	tl.cval[3] = *own++;
1265 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1266 		ret = 1;
1267 	} else {
1268 		tl.cval[0] = *own++;
1269 		tl.cval[1] = *own++;
1270 		tl.cval[2] = *own++;
1271 		tl.cval[3] = *own;
1272 		if (tl.lval != p->p_stats->p_start.tv_usec)
1273 			ret = 1;
1274 	}
1275 	PROC_UNLOCK(p);
1276 	return (ret);
1277 }
1278 
1279 /*
1280  * - nfs pseudo system call for the client
1281  */
1282 /*
1283  * MPSAFE
1284  */
1285 static int
1286 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1287 {
1288 	struct file *fp;
1289 	struct nfscbd_args nfscbdarg;
1290 	struct nfsd_nfscbd_args nfscbdarg2;
1291 	struct nameidata nd;
1292 	struct nfscl_dumpmntopts dumpmntopts;
1293 	cap_rights_t rights;
1294 	char *buf;
1295 	int error;
1296 	struct mount *mp;
1297 	struct nfsmount *nmp;
1298 
1299 	NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td));
1300 	if (uap->flag & NFSSVC_CBADDSOCK) {
1301 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1302 		if (error)
1303 			goto out;
1304 		/*
1305 		 * Since we don't know what rights might be required,
1306 		 * pretend that we need them all. It is better to be too
1307 		 * careful than too reckless.
1308 		 */
1309 		error = fget(td, nfscbdarg.sock,
1310 		    cap_rights_init_one(&rights, CAP_SOCK_CLIENT), &fp);
1311 		if (error)
1312 			goto out;
1313 		if (fp->f_type != DTYPE_SOCKET) {
1314 			fdrop(fp, td);
1315 			error = EPERM;
1316 			goto out;
1317 		}
1318 		error = nfscbd_addsock(fp);
1319 		fdrop(fp, td);
1320 		if (!error && nfscl_enablecallb == 0) {
1321 			nfsv4_cbport = nfscbdarg.port;
1322 			nfscl_enablecallb = 1;
1323 		}
1324 	} else if (uap->flag & NFSSVC_NFSCBD) {
1325 		if (uap->argp == NULL) {
1326 			error = EINVAL;
1327 			goto out;
1328 		}
1329 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1330 		    sizeof(nfscbdarg2));
1331 		if (error)
1332 			goto out;
1333 		error = nfscbd_nfsd(td, &nfscbdarg2);
1334 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1335 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1336 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1337 		    dumpmntopts.ndmnt_blen > 1024))
1338 			error = EINVAL;
1339 		if (error == 0)
1340 			error = nfsrv_lookupfilename(&nd,
1341 			    dumpmntopts.ndmnt_fname, td);
1342 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1343 		    "nfs") != 0) {
1344 			vput(nd.ni_vp);
1345 			error = EINVAL;
1346 		}
1347 		if (error == 0) {
1348 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK |
1349 			    M_ZERO);
1350 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1351 			    dumpmntopts.ndmnt_blen);
1352 			vput(nd.ni_vp);
1353 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1354 			    dumpmntopts.ndmnt_blen);
1355 			free(buf, M_TEMP);
1356 		}
1357 	} else if (uap->flag & NFSSVC_FORCEDISM) {
1358 		buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
1359 		error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
1360 		if (error == 0) {
1361 			nmp = NULL;
1362 			mtx_lock(&mountlist_mtx);
1363 			TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1364 				if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
1365 				    0 && strcmp(mp->mnt_stat.f_fstypename,
1366 				    "nfs") == 0 && mp->mnt_data != NULL) {
1367 					nmp = VFSTONFS(mp);
1368 					NFSDDSLOCK();
1369 					if (nfsv4_findmirror(nmp) != NULL) {
1370 						NFSDDSUNLOCK();
1371 						error = ENXIO;
1372 						nmp = NULL;
1373 						break;
1374 					}
1375 					mtx_lock(&nmp->nm_mtx);
1376 					if ((nmp->nm_privflag &
1377 					    NFSMNTP_FORCEDISM) == 0) {
1378 						nmp->nm_privflag |=
1379 						   (NFSMNTP_FORCEDISM |
1380 						    NFSMNTP_CANCELRPCS);
1381 						mtx_unlock(&nmp->nm_mtx);
1382 					} else {
1383 						mtx_unlock(&nmp->nm_mtx);
1384 						nmp = NULL;
1385 					}
1386 					NFSDDSUNLOCK();
1387 					break;
1388 				}
1389 			}
1390 			mtx_unlock(&mountlist_mtx);
1391 
1392 			if (nmp != NULL) {
1393 				/*
1394 				 * Call newnfs_nmcancelreqs() to cause
1395 				 * any RPCs in progress on the mount point to
1396 				 * fail.
1397 				 * This will cause any process waiting for an
1398 				 * RPC to complete while holding a vnode lock
1399 				 * on the mounted-on vnode (such as "df" or
1400 				 * a non-forced "umount") to fail.
1401 				 * This will unlock the mounted-on vnode so
1402 				 * a forced dismount can succeed.
1403 				 * Then clear NFSMNTP_CANCELRPCS and wakeup(),
1404 				 * so that nfs_unmount() can complete.
1405 				 */
1406 				newnfs_nmcancelreqs(nmp);
1407 				mtx_lock(&nmp->nm_mtx);
1408 				nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
1409 				wakeup(nmp);
1410 				mtx_unlock(&nmp->nm_mtx);
1411 			} else if (error == 0)
1412 				error = EINVAL;
1413 		}
1414 		free(buf, M_TEMP);
1415 	} else {
1416 		error = EINVAL;
1417 	}
1418 out:
1419 	NFSD_CURVNET_RESTORE();
1420 	return (error);
1421 }
1422 
1423 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1424 
1425 /*
1426  * Called once to initialize data structures...
1427  */
1428 static int
1429 nfscl_modevent(module_t mod, int type, void *data)
1430 {
1431 	int error = 0;
1432 	static int loaded = 0;
1433 
1434 	switch (type) {
1435 	case MOD_LOAD:
1436 		if (loaded)
1437 			return (0);
1438 		newnfs_portinit();
1439 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1440 		nfscl_init();
1441 		NFSD_LOCK();
1442 		nfsrvd_cbinit(0);
1443 		NFSD_UNLOCK();
1444 		ncl_call_invalcaches = ncl_invalcaches;
1445 		nfsd_call_nfscl = nfssvc_nfscl;
1446 		loaded = 1;
1447 		break;
1448 
1449 	case MOD_UNLOAD:
1450 		if (nfs_numnfscbd != 0) {
1451 			error = EBUSY;
1452 			break;
1453 		}
1454 
1455 		/*
1456 		 * XXX: Unloading of nfscl module is unsupported.
1457 		 */
1458 #if 0
1459 		ncl_call_invalcaches = NULL;
1460 		nfsd_call_nfscl = NULL;
1461 		uma_zdestroy(ncl_pbuf_zone);
1462 		/* and get rid of the mutexes */
1463 		mtx_destroy(&ncl_iod_mutex);
1464 		loaded = 0;
1465 		break;
1466 #else
1467 		/* FALLTHROUGH */
1468 #endif
1469 	default:
1470 		error = EOPNOTSUPP;
1471 		break;
1472 	}
1473 	return error;
1474 }
1475 static moduledata_t nfscl_mod = {
1476 	"nfscl",
1477 	nfscl_modevent,
1478 	NULL,
1479 };
1480 /*
1481  * This is the main module declaration for the NFS client.  The
1482  * nfscl_modevent() function is needed to ensure that the module
1483  * cannot be unloaded, among other things.
1484  * There is also a module declaration in sys/fs/nfsclient/nfs_clvfsops.c
1485  * for the name "nfs" within the VFS_SET() macro that defines the "nfs"
1486  * file system type.
1487  */
1488 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1489 
1490 /* So that loader and kldload(2) can find us, wherever we are.. */
1491 MODULE_VERSION(nfscl, 1);
1492 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1493 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1494 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1495 MODULE_DEPEND(nfscl, xdr, 1, 1, 1);
1496