1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD: stable/9/sys/fs/nfsclient/nfs_clport.c 317477 2017-04-26 23:32:57Z rmacklem $");
36 
37 #include "opt_inet6.h"
38 #include "opt_kdtrace.h"
39 
40 #include <sys/capability.h>
41 
42 /*
43  * generally, I don't like #includes inside .h files, but it seems to
44  * be the easiest way to handle the port.
45  */
46 #include <sys/hash.h>
47 #include <fs/nfs/nfsport.h>
48 #include <netinet/if_ether.h>
49 #include <net/if_types.h>
50 
51 #include <fs/nfsclient/nfs_kdtrace.h>
52 
53 #ifdef KDTRACE_HOOKS
54 dtrace_nfsclient_attrcache_flush_probe_func_t
55 		dtrace_nfscl_attrcache_flush_done_probe;
56 uint32_t	nfscl_attrcache_flush_done_id;
57 
58 dtrace_nfsclient_attrcache_get_hit_probe_func_t
59 		dtrace_nfscl_attrcache_get_hit_probe;
60 uint32_t	nfscl_attrcache_get_hit_id;
61 
62 dtrace_nfsclient_attrcache_get_miss_probe_func_t
63 		dtrace_nfscl_attrcache_get_miss_probe;
64 uint32_t	nfscl_attrcache_get_miss_id;
65 
66 dtrace_nfsclient_attrcache_load_probe_func_t
67 		dtrace_nfscl_attrcache_load_done_probe;
68 uint32_t	nfscl_attrcache_load_done_id;
69 #endif /* !KDTRACE_HOOKS */
70 
71 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
72 extern struct vop_vector newnfs_vnodeops;
73 extern struct vop_vector newnfs_fifoops;
74 extern uma_zone_t newnfsnode_zone;
75 extern struct buf_ops buf_ops_newnfs;
76 extern int ncl_pbuf_freecnt;
77 extern short nfsv4_cbport;
78 extern int nfscl_enablecallb;
79 extern int nfs_numnfscbd;
80 extern int nfscl_inited;
81 struct mtx nfs_clstate_mutex;
82 struct mtx ncl_iod_mutex;
83 NFSDLOCKMUTEX;
84 
85 extern void (*ncl_call_invalcaches)(struct vnode *);
86 
87 /*
88  * Comparison function for vfs_hash functions.
89  */
90 int
newnfs_vncmpf(struct vnode * vp,void * arg)91 newnfs_vncmpf(struct vnode *vp, void *arg)
92 {
93 	struct nfsfh *nfhp = (struct nfsfh *)arg;
94 	struct nfsnode *np = VTONFS(vp);
95 
96 	if (np->n_fhp->nfh_len != nfhp->nfh_len ||
97 	    NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
98 		return (1);
99 	return (0);
100 }
101 
102 /*
103  * Look up a vnode/nfsnode by file handle.
104  * Callers must check for mount points!!
105  * In all cases, a pointer to a
106  * nfsnode structure is returned.
107  * This variant takes a "struct nfsfh *" as second argument and uses
108  * that structure up, either by hanging off the nfsnode or FREEing it.
109  */
110 int
nfscl_nget(struct mount * mntp,struct vnode * dvp,struct nfsfh * nfhp,struct componentname * cnp,struct thread * td,struct nfsnode ** npp,void * stuff,int lkflags)111 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
112     struct componentname *cnp, struct thread *td, struct nfsnode **npp,
113     void *stuff, int lkflags)
114 {
115 	struct nfsnode *np, *dnp;
116 	struct vnode *vp, *nvp;
117 	struct nfsv4node *newd, *oldd;
118 	int error;
119 	u_int hash;
120 	struct nfsmount *nmp;
121 
122 	nmp = VFSTONFS(mntp);
123 	dnp = VTONFS(dvp);
124 	*npp = NULL;
125 
126 	hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
127 
128 	error = vfs_hash_get(mntp, hash, lkflags,
129 	    td, &nvp, newnfs_vncmpf, nfhp);
130 	if (error == 0 && nvp != NULL) {
131 		/*
132 		 * I believe there is a slight chance that vgonel() could
133 		 * get called on this vnode between when NFSVOPLOCK() drops
134 		 * the VI_LOCK() and vget() acquires it again, so that it
135 		 * hasn't yet had v_usecount incremented. If this were to
136 		 * happen, the VI_DOOMED flag would be set, so check for
137 		 * that here. Since we now have the v_usecount incremented,
138 		 * we should be ok until we vrele() it, if the VI_DOOMED
139 		 * flag isn't set now.
140 		 */
141 		VI_LOCK(nvp);
142 		if ((nvp->v_iflag & VI_DOOMED)) {
143 			VI_UNLOCK(nvp);
144 			vrele(nvp);
145 			error = ENOENT;
146 		} else {
147 			VI_UNLOCK(nvp);
148 		}
149 	}
150 	if (error) {
151 		FREE((caddr_t)nfhp, M_NFSFH);
152 		return (error);
153 	}
154 	if (nvp != NULL) {
155 		np = VTONFS(nvp);
156 		/*
157 		 * For NFSv4, check to see if it is the same name and
158 		 * replace the name, if it is different.
159 		 */
160 		oldd = newd = NULL;
161 		if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
162 		    nvp->v_type == VREG &&
163 		    (np->n_v4->n4_namelen != cnp->cn_namelen ||
164 		     NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
165 		     cnp->cn_namelen) ||
166 		     dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
167 		     NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
168 		     dnp->n_fhp->nfh_len))) {
169 		    MALLOC(newd, struct nfsv4node *,
170 			sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
171 			+ cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
172 		    NFSLOCKNODE(np);
173 		    if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
174 			&& (np->n_v4->n4_namelen != cnp->cn_namelen ||
175 			 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
176 			 cnp->cn_namelen) ||
177 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
178 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
179 			 dnp->n_fhp->nfh_len))) {
180 			oldd = np->n_v4;
181 			np->n_v4 = newd;
182 			newd = NULL;
183 			np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
184 			np->n_v4->n4_namelen = cnp->cn_namelen;
185 			NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
186 			    dnp->n_fhp->nfh_len);
187 			NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
188 			    cnp->cn_namelen);
189 		    }
190 		    NFSUNLOCKNODE(np);
191 		}
192 		if (newd != NULL)
193 			FREE((caddr_t)newd, M_NFSV4NODE);
194 		if (oldd != NULL)
195 			FREE((caddr_t)oldd, M_NFSV4NODE);
196 		*npp = np;
197 		FREE((caddr_t)nfhp, M_NFSFH);
198 		return (0);
199 	}
200 
201 	/*
202 	 * Allocate before getnewvnode since doing so afterward
203 	 * might cause a bogus v_data pointer to get dereferenced
204 	 * elsewhere if zalloc should block.
205 	 */
206 	np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
207 
208 	error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
209 	if (error) {
210 		uma_zfree(newnfsnode_zone, np);
211 		FREE((caddr_t)nfhp, M_NFSFH);
212 		return (error);
213 	}
214 	vp = nvp;
215 	KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
216 	vp->v_bufobj.bo_ops = &buf_ops_newnfs;
217 	vp->v_data = np;
218 	np->n_vnode = vp;
219 	/*
220 	 * Initialize the mutex even if the vnode is going to be a loser.
221 	 * This simplifies the logic in reclaim, which can then unconditionally
222 	 * destroy the mutex (in the case of the loser, or if hash_insert
223 	 * happened to return an error no special casing is needed).
224 	 */
225 	mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
226 
227 	/*
228 	 * Are we getting the root? If so, make sure the vnode flags
229 	 * are correct
230 	 */
231 	if ((nfhp->nfh_len == nmp->nm_fhsize) &&
232 	    !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
233 		if (vp->v_type == VNON)
234 			vp->v_type = VDIR;
235 		vp->v_vflag |= VV_ROOT;
236 	}
237 
238 	np->n_fhp = nfhp;
239 	/*
240 	 * For NFSv4, we have to attach the directory file handle and
241 	 * file name, so that Open Ops can be done later.
242 	 */
243 	if (nmp->nm_flag & NFSMNT_NFSV4) {
244 		MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
245 		    + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
246 		    M_WAITOK);
247 		np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
248 		np->n_v4->n4_namelen = cnp->cn_namelen;
249 		NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
250 		    dnp->n_fhp->nfh_len);
251 		NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
252 		    cnp->cn_namelen);
253 	} else {
254 		np->n_v4 = NULL;
255 	}
256 
257 	/*
258 	 * NFS supports recursive and shared locking.
259 	 */
260 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
261 	VN_LOCK_AREC(vp);
262 	VN_LOCK_ASHARE(vp);
263 	error = insmntque(vp, mntp);
264 	if (error != 0) {
265 		*npp = NULL;
266 		mtx_destroy(&np->n_mtx);
267 		FREE((caddr_t)nfhp, M_NFSFH);
268 		if (np->n_v4 != NULL)
269 			FREE((caddr_t)np->n_v4, M_NFSV4NODE);
270 		uma_zfree(newnfsnode_zone, np);
271 		return (error);
272 	}
273 	error = vfs_hash_insert(vp, hash, lkflags,
274 	    td, &nvp, newnfs_vncmpf, nfhp);
275 	if (error)
276 		return (error);
277 	if (nvp != NULL) {
278 		*npp = VTONFS(nvp);
279 		/* vfs_hash_insert() vput()'s the losing vnode */
280 		return (0);
281 	}
282 	*npp = np;
283 
284 	return (0);
285 }
286 
287 /*
288  * Anothe variant of nfs_nget(). This one is only used by reopen. It
289  * takes almost the same args as nfs_nget(), but only succeeds if an entry
290  * exists in the cache. (Since files should already be "open" with a
291  * vnode ref cnt on the node when reopen calls this, it should always
292  * succeed.)
293  * Also, don't get a vnode lock, since it may already be locked by some
294  * other process that is handling it. This is ok, since all other threads
295  * on the client are blocked by the nfsc_lock being exclusively held by the
296  * caller of this function.
297  */
298 int
nfscl_ngetreopen(struct mount * mntp,u_int8_t * fhp,int fhsize,struct thread * td,struct nfsnode ** npp)299 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
300     struct thread *td, struct nfsnode **npp)
301 {
302 	struct vnode *nvp;
303 	u_int hash;
304 	struct nfsfh *nfhp;
305 	int error;
306 
307 	*npp = NULL;
308 	/* For forced dismounts, just return error. */
309 	if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
310 		return (EINTR);
311 	MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
312 	    M_NFSFH, M_WAITOK);
313 	bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
314 	nfhp->nfh_len = fhsize;
315 
316 	hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
317 
318 	/*
319 	 * First, try to get the vnode locked, but don't block for the lock.
320 	 */
321 	error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
322 	    newnfs_vncmpf, nfhp);
323 	if (error == 0 && nvp != NULL) {
324 		NFSVOPUNLOCK(nvp, 0);
325 	} else if (error == EBUSY) {
326 		/*
327 		 * The LK_EXCLOTHER lock type tells nfs_lock1() to not try
328 		 * and lock the vnode, but just get a v_usecount on it.
329 		 * LK_NOWAIT is set so that when vget() returns ENOENT,
330 		 * vfs_hash_get() fails instead of looping.
331 		 * If this succeeds, it is safe so long as a vflush() with
332 		 * FORCECLOSE has not been done. Since the Renew thread is
333 		 * stopped and the MNTK_UNMOUNTF flag is set before doing
334 		 * a vflush() with FORCECLOSE, we should be ok here.
335 		 */
336 		if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
337 			error = EINTR;
338 		else
339 			error = vfs_hash_get(mntp, hash,
340 			    (LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
341 			    newnfs_vncmpf, nfhp);
342 	}
343 	FREE(nfhp, M_NFSFH);
344 	if (error)
345 		return (error);
346 	if (nvp != NULL) {
347 		*npp = VTONFS(nvp);
348 		return (0);
349 	}
350 	return (EINVAL);
351 }
352 
353 /*
354  * Load the attribute cache (that lives in the nfsnode entry) with
355  * the attributes of the second argument and
356  * Iff vaper not NULL
357  *    copy the attributes to *vaper
358  * Similar to nfs_loadattrcache(), except the attributes are passed in
359  * instead of being parsed out of the mbuf list.
360  */
361 int
nfscl_loadattrcache(struct vnode ** vpp,struct nfsvattr * nap,void * nvaper,void * stuff,int writeattr,int dontshrink)362 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
363     void *stuff, int writeattr, int dontshrink)
364 {
365 	struct vnode *vp = *vpp;
366 	struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
367 	struct nfsnode *np;
368 	struct nfsmount *nmp;
369 	struct timespec mtime_save;
370 	u_quad_t nsize;
371 	int setnsize;
372 
373 	/*
374 	 * If v_type == VNON it is a new node, so fill in the v_type,
375 	 * n_mtime fields. Check to see if it represents a special
376 	 * device, and if so, check for a possible alias. Once the
377 	 * correct vnode has been obtained, fill in the rest of the
378 	 * information.
379 	 */
380 	np = VTONFS(vp);
381 	NFSLOCKNODE(np);
382 	if (vp->v_type != nvap->va_type) {
383 		vp->v_type = nvap->va_type;
384 		if (vp->v_type == VFIFO)
385 			vp->v_op = &newnfs_fifoops;
386 		np->n_mtime = nvap->va_mtime;
387 	}
388 	nmp = VFSTONFS(vp->v_mount);
389 	vap = &np->n_vattr.na_vattr;
390 	mtime_save = vap->va_mtime;
391 	if (writeattr) {
392 		np->n_vattr.na_filerev = nap->na_filerev;
393 		np->n_vattr.na_size = nap->na_size;
394 		np->n_vattr.na_mtime = nap->na_mtime;
395 		np->n_vattr.na_ctime = nap->na_ctime;
396 		np->n_vattr.na_fsid = nap->na_fsid;
397 		np->n_vattr.na_mode = nap->na_mode;
398 	} else {
399 		NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
400 		    sizeof (struct nfsvattr));
401 	}
402 
403 	/*
404 	 * For NFSv4, if the node's fsid is not equal to the mount point's
405 	 * fsid, return the low order 32bits of the node's fsid. This
406 	 * allows getcwd(3) to work. There is a chance that the fsid might
407 	 * be the same as a local fs, but since this is in an NFS mount
408 	 * point, I don't think that will cause any problems?
409 	 */
410 	if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
411 	    (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
412 	     nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
413 		/*
414 		 * va_fsid needs to be set to some value derived from
415 		 * np->n_vattr.na_filesid that is not equal
416 		 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
417 		 * from the value used for the top level server volume
418 		 * in the mounted subtree.
419 		 */
420 		if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
421 		    (uint32_t)np->n_vattr.na_filesid[0])
422 			vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
423 		else
424 			vap->va_fsid = (uint32_t)hash32_buf(
425 			    np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
426 	} else
427 		vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
428 	np->n_attrstamp = time_second;
429 	setnsize = 0;
430 	nsize = 0;
431 	if (vap->va_size != np->n_size) {
432 		if (vap->va_type == VREG) {
433 			if (dontshrink && vap->va_size < np->n_size) {
434 				/*
435 				 * We've been told not to shrink the file;
436 				 * zero np->n_attrstamp to indicate that
437 				 * the attributes are stale.
438 				 */
439 				vap->va_size = np->n_size;
440 				np->n_attrstamp = 0;
441 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
442 				vnode_pager_setsize(vp, np->n_size);
443 			} else if (np->n_flag & NMODIFIED) {
444 				/*
445 				 * We've modified the file: Use the larger
446 				 * of our size, and the server's size.
447 				 */
448 				if (vap->va_size < np->n_size) {
449 					vap->va_size = np->n_size;
450 				} else {
451 					np->n_size = vap->va_size;
452 					np->n_flag |= NSIZECHANGED;
453 				}
454 				vnode_pager_setsize(vp, np->n_size);
455 			} else if (vap->va_size < np->n_size) {
456 				/*
457 				 * When shrinking the size, the call to
458 				 * vnode_pager_setsize() cannot be done
459 				 * with the mutex held, so delay it until
460 				 * after the mtx_unlock call.
461 				 */
462 				nsize = np->n_size = vap->va_size;
463 				np->n_flag |= NSIZECHANGED;
464 				setnsize = 1;
465 			} else {
466 				np->n_size = vap->va_size;
467 				np->n_flag |= NSIZECHANGED;
468 				vnode_pager_setsize(vp, np->n_size);
469 			}
470 		} else {
471 			np->n_size = vap->va_size;
472 		}
473 	}
474 	/*
475 	 * The following checks are added to prevent a race between (say)
476 	 * a READDIR+ and a WRITE.
477 	 * READDIR+, WRITE requests sent out.
478 	 * READDIR+ resp, WRITE resp received on client.
479 	 * However, the WRITE resp was handled before the READDIR+ resp
480 	 * causing the post op attrs from the write to be loaded first
481 	 * and the attrs from the READDIR+ to be loaded later. If this
482 	 * happens, we have stale attrs loaded into the attrcache.
483 	 * We detect this by for the mtime moving back. We invalidate the
484 	 * attrcache when this happens.
485 	 */
486 	if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
487 		/* Size changed or mtime went backwards */
488 		np->n_attrstamp = 0;
489 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
490 	}
491 	if (vaper != NULL) {
492 		NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
493 		if (np->n_flag & NCHG) {
494 			if (np->n_flag & NACC)
495 				vaper->va_atime = np->n_atim;
496 			if (np->n_flag & NUPD)
497 				vaper->va_mtime = np->n_mtim;
498 		}
499 	}
500 #ifdef KDTRACE_HOOKS
501 	if (np->n_attrstamp != 0)
502 		KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, 0);
503 #endif
504 	NFSUNLOCKNODE(np);
505 	if (setnsize)
506 		vnode_pager_setsize(vp, nsize);
507 	return (0);
508 }
509 
510 /*
511  * Fill in the client id name. For these bytes:
512  * 1 - they must be unique
513  * 2 - they should be persistent across client reboots
514  * 1 is more critical than 2
515  * Use the mount point's unique id plus either the uuid or, if that
516  * isn't set, random junk.
517  */
518 void
nfscl_fillclid(u_int64_t clval,char * uuid,u_int8_t * cp,u_int16_t idlen)519 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
520 {
521 	int uuidlen;
522 
523 	/*
524 	 * First, put in the 64bit mount point identifier.
525 	 */
526 	if (idlen >= sizeof (u_int64_t)) {
527 		NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
528 		cp += sizeof (u_int64_t);
529 		idlen -= sizeof (u_int64_t);
530 	}
531 
532 	/*
533 	 * If uuid is non-zero length, use it.
534 	 */
535 	uuidlen = strlen(uuid);
536 	if (uuidlen > 0 && idlen >= uuidlen) {
537 		NFSBCOPY(uuid, cp, uuidlen);
538 		cp += uuidlen;
539 		idlen -= uuidlen;
540 	}
541 
542 	/*
543 	 * This only normally happens if the uuid isn't set.
544 	 */
545 	while (idlen > 0) {
546 		*cp++ = (u_int8_t)(arc4random() % 256);
547 		idlen--;
548 	}
549 }
550 
551 /*
552  * Fill in a lock owner name. For now, pid + the process's creation time.
553  */
554 void
nfscl_filllockowner(void * id,u_int8_t * cp,int flags)555 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
556 {
557 	union {
558 		u_int32_t	lval;
559 		u_int8_t	cval[4];
560 	} tl;
561 	struct proc *p;
562 
563 	if (id == NULL) {
564 		printf("NULL id\n");
565 		bzero(cp, NFSV4CL_LOCKNAMELEN);
566 		return;
567 	}
568 	if ((flags & F_POSIX) != 0) {
569 		p = (struct proc *)id;
570 		tl.lval = p->p_pid;
571 		*cp++ = tl.cval[0];
572 		*cp++ = tl.cval[1];
573 		*cp++ = tl.cval[2];
574 		*cp++ = tl.cval[3];
575 		tl.lval = p->p_stats->p_start.tv_sec;
576 		*cp++ = tl.cval[0];
577 		*cp++ = tl.cval[1];
578 		*cp++ = tl.cval[2];
579 		*cp++ = tl.cval[3];
580 		tl.lval = p->p_stats->p_start.tv_usec;
581 		*cp++ = tl.cval[0];
582 		*cp++ = tl.cval[1];
583 		*cp++ = tl.cval[2];
584 		*cp = tl.cval[3];
585 	} else if ((flags & F_FLOCK) != 0) {
586 		bcopy(&id, cp, sizeof(id));
587 		bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
588 	} else {
589 		printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
590 		bzero(cp, NFSV4CL_LOCKNAMELEN);
591 	}
592 }
593 
594 /*
595  * Find the parent process for the thread passed in as an argument.
596  * If none exists, return NULL, otherwise return a thread for the parent.
597  * (Can be any of the threads, since it is only used for td->td_proc.)
598  */
599 NFSPROC_T *
nfscl_getparent(struct thread * td)600 nfscl_getparent(struct thread *td)
601 {
602 	struct proc *p;
603 	struct thread *ptd;
604 
605 	if (td == NULL)
606 		return (NULL);
607 	p = td->td_proc;
608 	if (p->p_pid == 0)
609 		return (NULL);
610 	p = p->p_pptr;
611 	if (p == NULL)
612 		return (NULL);
613 	ptd = TAILQ_FIRST(&p->p_threads);
614 	return (ptd);
615 }
616 
617 /*
618  * Start up the renew kernel thread.
619  */
620 static void
start_nfscl(void * arg)621 start_nfscl(void *arg)
622 {
623 	struct nfsclclient *clp;
624 	struct thread *td;
625 
626 	clp = (struct nfsclclient *)arg;
627 	td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
628 	nfscl_renewthread(clp, td);
629 	kproc_exit(0);
630 }
631 
632 void
nfscl_start_renewthread(struct nfsclclient * clp)633 nfscl_start_renewthread(struct nfsclclient *clp)
634 {
635 
636 	kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
637 	    "nfscl");
638 }
639 
640 /*
641  * Handle wcc_data.
642  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
643  * as the first Op after PutFH.
644  * (For NFSv4, the postop attributes are after the Op, so they can't be
645  *  parsed here. A separate call to nfscl_postop_attr() is required.)
646  */
647 int
nfscl_wcc_data(struct nfsrv_descript * nd,struct vnode * vp,struct nfsvattr * nap,int * flagp,int * wccflagp,void * stuff)648 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
649     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
650 {
651 	u_int32_t *tl;
652 	struct nfsnode *np = VTONFS(vp);
653 	struct nfsvattr nfsva;
654 	int error = 0;
655 
656 	if (wccflagp != NULL)
657 		*wccflagp = 0;
658 	if (nd->nd_flag & ND_NFSV3) {
659 		*flagp = 0;
660 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
661 		if (*tl == newnfs_true) {
662 			NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
663 			if (wccflagp != NULL) {
664 				mtx_lock(&np->n_mtx);
665 				*wccflagp = (np->n_mtime.tv_sec ==
666 				    fxdr_unsigned(u_int32_t, *(tl + 2)) &&
667 				    np->n_mtime.tv_nsec ==
668 				    fxdr_unsigned(u_int32_t, *(tl + 3)));
669 				mtx_unlock(&np->n_mtx);
670 			}
671 		}
672 		error = nfscl_postop_attr(nd, nap, flagp, stuff);
673 	} else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
674 	    == (ND_NFSV4 | ND_V4WCCATTR)) {
675 		error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
676 		    NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
677 		    NULL, NULL, NULL, NULL, NULL);
678 		if (error)
679 			return (error);
680 		/*
681 		 * Get rid of Op# and status for next op.
682 		 */
683 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
684 		if (*++tl)
685 			nd->nd_flag |= ND_NOMOREDATA;
686 		if (wccflagp != NULL &&
687 		    nfsva.na_vattr.va_mtime.tv_sec != 0) {
688 			mtx_lock(&np->n_mtx);
689 			*wccflagp = (np->n_mtime.tv_sec ==
690 			    nfsva.na_vattr.va_mtime.tv_sec &&
691 			    np->n_mtime.tv_nsec ==
692 			    nfsva.na_vattr.va_mtime.tv_sec);
693 			mtx_unlock(&np->n_mtx);
694 		}
695 	}
696 nfsmout:
697 	return (error);
698 }
699 
700 /*
701  * Get postop attributes.
702  */
703 int
nfscl_postop_attr(struct nfsrv_descript * nd,struct nfsvattr * nap,int * retp,void * stuff)704 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
705     void *stuff)
706 {
707 	u_int32_t *tl;
708 	int error = 0;
709 
710 	*retp = 0;
711 	if (nd->nd_flag & ND_NOMOREDATA)
712 		return (error);
713 	if (nd->nd_flag & ND_NFSV3) {
714 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
715 		*retp = fxdr_unsigned(int, *tl);
716 	} else if (nd->nd_flag & ND_NFSV4) {
717 		/*
718 		 * For NFSv4, the postop attr are at the end, so no point
719 		 * in looking if nd_repstat != 0.
720 		 */
721 		if (!nd->nd_repstat) {
722 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
723 			if (*(tl + 1))
724 				/* should never happen since nd_repstat != 0 */
725 				nd->nd_flag |= ND_NOMOREDATA;
726 			else
727 				*retp = 1;
728 		}
729 	} else if (!nd->nd_repstat) {
730 		/* For NFSv2, the attributes are here iff nd_repstat == 0 */
731 		*retp = 1;
732 	}
733 	if (*retp) {
734 		error = nfsm_loadattr(nd, nap);
735 		if (error)
736 			*retp = 0;
737 	}
738 nfsmout:
739 	return (error);
740 }
741 
742 /*
743  * Fill in the setable attributes. The full argument indicates whether
744  * to fill in them all or just mode and time.
745  */
746 void
nfscl_fillsattr(struct nfsrv_descript * nd,struct vattr * vap,struct vnode * vp,int flags,u_int32_t rdev)747 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
748     struct vnode *vp, int flags, u_int32_t rdev)
749 {
750 	u_int32_t *tl;
751 	struct nfsv2_sattr *sp;
752 	nfsattrbit_t attrbits;
753 
754 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
755 	case ND_NFSV2:
756 		NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
757 		if (vap->va_mode == (mode_t)VNOVAL)
758 			sp->sa_mode = newnfs_xdrneg1;
759 		else
760 			sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
761 		if (vap->va_uid == (uid_t)VNOVAL)
762 			sp->sa_uid = newnfs_xdrneg1;
763 		else
764 			sp->sa_uid = txdr_unsigned(vap->va_uid);
765 		if (vap->va_gid == (gid_t)VNOVAL)
766 			sp->sa_gid = newnfs_xdrneg1;
767 		else
768 			sp->sa_gid = txdr_unsigned(vap->va_gid);
769 		if (flags & NFSSATTR_SIZE0)
770 			sp->sa_size = 0;
771 		else if (flags & NFSSATTR_SIZENEG1)
772 			sp->sa_size = newnfs_xdrneg1;
773 		else if (flags & NFSSATTR_SIZERDEV)
774 			sp->sa_size = txdr_unsigned(rdev);
775 		else
776 			sp->sa_size = txdr_unsigned(vap->va_size);
777 		txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
778 		txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
779 		break;
780 	case ND_NFSV3:
781 		if (vap->va_mode != (mode_t)VNOVAL) {
782 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
783 			*tl++ = newnfs_true;
784 			*tl = txdr_unsigned(vap->va_mode);
785 		} else {
786 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
787 			*tl = newnfs_false;
788 		}
789 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
790 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
791 			*tl++ = newnfs_true;
792 			*tl = txdr_unsigned(vap->va_uid);
793 		} else {
794 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
795 			*tl = newnfs_false;
796 		}
797 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
798 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
799 			*tl++ = newnfs_true;
800 			*tl = txdr_unsigned(vap->va_gid);
801 		} else {
802 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
803 			*tl = newnfs_false;
804 		}
805 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
806 			NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
807 			*tl++ = newnfs_true;
808 			txdr_hyper(vap->va_size, tl);
809 		} else {
810 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
811 			*tl = newnfs_false;
812 		}
813 		if (vap->va_atime.tv_sec != VNOVAL) {
814 			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
815 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
816 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
817 				txdr_nfsv3time(&vap->va_atime, tl);
818 			} else {
819 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
820 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
821 			}
822 		} else {
823 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
824 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
825 		}
826 		if (vap->va_mtime.tv_sec != VNOVAL) {
827 			if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
828 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
829 				*tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
830 				txdr_nfsv3time(&vap->va_mtime, tl);
831 			} else {
832 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
833 				*tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
834 			}
835 		} else {
836 			NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
837 			*tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
838 		}
839 		break;
840 	case ND_NFSV4:
841 		NFSZERO_ATTRBIT(&attrbits);
842 		if (vap->va_mode != (mode_t)VNOVAL)
843 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
844 		if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
845 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
846 		if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
847 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
848 		if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
849 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
850 		if (vap->va_atime.tv_sec != VNOVAL)
851 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
852 		if (vap->va_mtime.tv_sec != VNOVAL)
853 			NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
854 		(void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
855 		    &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
856 		break;
857 	};
858 }
859 
860 /*
861  * nfscl_request() - mostly a wrapper for newnfs_request().
862  */
863 int
nfscl_request(struct nfsrv_descript * nd,struct vnode * vp,NFSPROC_T * p,struct ucred * cred,void * stuff)864 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
865     struct ucred *cred, void *stuff)
866 {
867 	int ret, vers;
868 	struct nfsmount *nmp;
869 
870 	nmp = VFSTONFS(vp->v_mount);
871 	if (nd->nd_flag & ND_NFSV4)
872 		vers = NFS_VER4;
873 	else if (nd->nd_flag & ND_NFSV3)
874 		vers = NFS_VER3;
875 	else
876 		vers = NFS_VER2;
877 	ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
878 		NFS_PROG, vers, NULL, 1, NULL);
879 	return (ret);
880 }
881 
882 /*
883  * fill in this bsden's variant of statfs using nfsstatfs.
884  */
885 void
nfscl_loadsbinfo(struct nfsmount * nmp,struct nfsstatfs * sfp,void * statfs)886 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
887 {
888 	struct statfs *sbp = (struct statfs *)statfs;
889 
890 	if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
891 		sbp->f_bsize = NFS_FABLKSIZE;
892 		sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
893 		sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
894 		/*
895 		 * Although sf_abytes is uint64_t and f_bavail is int64_t,
896 		 * the value after dividing by NFS_FABLKSIZE is small
897 		 * enough that it will fit in 63bits, so it is ok to
898 		 * assign it to f_bavail without fear that it will become
899 		 * negative.
900 		 */
901 		sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
902 		sbp->f_files = sfp->sf_tfiles;
903 		/* Since f_ffree is int64_t, clip it to 63bits. */
904 		if (sfp->sf_ffiles > INT64_MAX)
905 			sbp->f_ffree = INT64_MAX;
906 		else
907 			sbp->f_ffree = sfp->sf_ffiles;
908 	} else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
909 		/*
910 		 * The type casts to (int32_t) ensure that this code is
911 		 * compatible with the old NFS client, in that it will
912 		 * propagate bit31 to the high order bits. This may or may
913 		 * not be correct for NFSv2, but since it is a legacy
914 		 * environment, I'd rather retain backwards compatibility.
915 		 */
916 		sbp->f_bsize = (int32_t)sfp->sf_bsize;
917 		sbp->f_blocks = (int32_t)sfp->sf_blocks;
918 		sbp->f_bfree = (int32_t)sfp->sf_bfree;
919 		sbp->f_bavail = (int32_t)sfp->sf_bavail;
920 		sbp->f_files = 0;
921 		sbp->f_ffree = 0;
922 	}
923 }
924 
925 /*
926  * Use the fsinfo stuff to update the mount point.
927  */
928 void
nfscl_loadfsinfo(struct nfsmount * nmp,struct nfsfsinfo * fsp)929 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
930 {
931 
932 	if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
933 	    fsp->fs_wtpref >= NFS_FABLKSIZE)
934 		nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
935 		    ~(NFS_FABLKSIZE - 1);
936 	if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
937 		nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
938 		if (nmp->nm_wsize == 0)
939 			nmp->nm_wsize = fsp->fs_wtmax;
940 	}
941 	if (nmp->nm_wsize < NFS_FABLKSIZE)
942 		nmp->nm_wsize = NFS_FABLKSIZE;
943 	if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
944 	    fsp->fs_rtpref >= NFS_FABLKSIZE)
945 		nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
946 		    ~(NFS_FABLKSIZE - 1);
947 	if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
948 		nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
949 		if (nmp->nm_rsize == 0)
950 			nmp->nm_rsize = fsp->fs_rtmax;
951 	}
952 	if (nmp->nm_rsize < NFS_FABLKSIZE)
953 		nmp->nm_rsize = NFS_FABLKSIZE;
954 	if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
955 	    && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
956 		nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
957 		    ~(NFS_DIRBLKSIZ - 1);
958 	if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
959 		nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
960 		if (nmp->nm_readdirsize == 0)
961 			nmp->nm_readdirsize = fsp->fs_rtmax;
962 	}
963 	if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
964 		nmp->nm_readdirsize = NFS_DIRBLKSIZ;
965 	if (fsp->fs_maxfilesize > 0 &&
966 	    fsp->fs_maxfilesize < nmp->nm_maxfilesize)
967 		nmp->nm_maxfilesize = fsp->fs_maxfilesize;
968 	nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
969 	nmp->nm_state |= NFSSTA_GOTFSINFO;
970 }
971 
972 /*
973  * Get a pointer to my IP addrress and return it.
974  * Return NULL if you can't find one.
975  */
976 u_int8_t *
nfscl_getmyip(struct nfsmount * nmp,int * isinet6p)977 nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
978 {
979 	struct sockaddr_in sad, *sin;
980 	struct rtentry *rt;
981 	u_int8_t *retp = NULL;
982 	static struct in_addr laddr;
983 
984 	*isinet6p = 0;
985 	/*
986 	 * Loop up a route for the destination address.
987 	 */
988 	if (nmp->nm_nam->sa_family == AF_INET) {
989 		bzero(&sad, sizeof (sad));
990 		sin = (struct sockaddr_in *)nmp->nm_nam;
991 		sad.sin_family = AF_INET;
992 		sad.sin_len = sizeof (struct sockaddr_in);
993 		sad.sin_addr.s_addr = sin->sin_addr.s_addr;
994 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
995 		rt = rtalloc1_fib((struct sockaddr *)&sad, 0, 0UL,
996 		     curthread->td_proc->p_fibnum);
997 		if (rt != NULL) {
998 			if (rt->rt_ifp != NULL &&
999 			    rt->rt_ifa != NULL &&
1000 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1001 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
1002 				sin = (struct sockaddr_in *)
1003 				    rt->rt_ifa->ifa_addr;
1004 				laddr.s_addr = sin->sin_addr.s_addr;
1005 				retp = (u_int8_t *)&laddr;
1006 			}
1007 			RTFREE_LOCKED(rt);
1008 		}
1009 		CURVNET_RESTORE();
1010 #ifdef INET6
1011 	} else if (nmp->nm_nam->sa_family == AF_INET6) {
1012 		struct sockaddr_in6 sad6, *sin6;
1013 		static struct in6_addr laddr6;
1014 
1015 		bzero(&sad6, sizeof (sad6));
1016 		sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1017 		sad6.sin6_family = AF_INET6;
1018 		sad6.sin6_len = sizeof (struct sockaddr_in6);
1019 		sad6.sin6_addr = sin6->sin6_addr;
1020 		CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1021 		rt = rtalloc1_fib((struct sockaddr *)&sad6, 0, 0UL,
1022 		     curthread->td_proc->p_fibnum);
1023 		if (rt != NULL) {
1024 			if (rt->rt_ifp != NULL &&
1025 			    rt->rt_ifa != NULL &&
1026 			    ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1027 			    rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
1028 				sin6 = (struct sockaddr_in6 *)
1029 				    rt->rt_ifa->ifa_addr;
1030 				laddr6 = sin6->sin6_addr;
1031 				retp = (u_int8_t *)&laddr6;
1032 				*isinet6p = 1;
1033 			}
1034 			RTFREE_LOCKED(rt);
1035 		}
1036 		CURVNET_RESTORE();
1037 #endif
1038 	}
1039 	return (retp);
1040 }
1041 
1042 /*
1043  * Copy NFS uid, gids from the cred structure.
1044  */
1045 void
newnfs_copyincred(struct ucred * cr,struct nfscred * nfscr)1046 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1047 {
1048 	int i;
1049 
1050 	KASSERT(cr->cr_ngroups >= 0,
1051 	    ("newnfs_copyincred: negative cr_ngroups"));
1052 	nfscr->nfsc_uid = cr->cr_uid;
1053 	nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1054 	for (i = 0; i < nfscr->nfsc_ngroups; i++)
1055 		nfscr->nfsc_groups[i] = cr->cr_groups[i];
1056 }
1057 
1058 
1059 /*
1060  * Do any client specific initialization.
1061  */
1062 void
nfscl_init(void)1063 nfscl_init(void)
1064 {
1065 	static int inited = 0;
1066 
1067 	if (inited)
1068 		return;
1069 	inited = 1;
1070 	nfscl_inited = 1;
1071 	ncl_pbuf_freecnt = nswbuf / 2 + 1;
1072 }
1073 
1074 /*
1075  * Check each of the attributes to be set, to ensure they aren't already
1076  * the correct value. Disable setting ones already correct.
1077  */
1078 int
nfscl_checksattr(struct vattr * vap,struct nfsvattr * nvap)1079 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1080 {
1081 
1082 	if (vap->va_mode != (mode_t)VNOVAL) {
1083 		if (vap->va_mode == nvap->na_mode)
1084 			vap->va_mode = (mode_t)VNOVAL;
1085 	}
1086 	if (vap->va_uid != (uid_t)VNOVAL) {
1087 		if (vap->va_uid == nvap->na_uid)
1088 			vap->va_uid = (uid_t)VNOVAL;
1089 	}
1090 	if (vap->va_gid != (gid_t)VNOVAL) {
1091 		if (vap->va_gid == nvap->na_gid)
1092 			vap->va_gid = (gid_t)VNOVAL;
1093 	}
1094 	if (vap->va_size != VNOVAL) {
1095 		if (vap->va_size == nvap->na_size)
1096 			vap->va_size = VNOVAL;
1097 	}
1098 
1099 	/*
1100 	 * We are normally called with only a partially initialized
1101 	 * VAP.  Since the NFSv3 spec says that server may use the
1102 	 * file attributes to store the verifier, the spec requires
1103 	 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1104 	 * in atime, but we can't really assume that all servers will
1105 	 * so we ensure that our SETATTR sets both atime and mtime.
1106 	 * Set the VA_UTIMES_NULL flag for this case, so that
1107 	 * the server's time will be used.  This is needed to
1108 	 * work around a bug in some Solaris servers, where
1109 	 * setting the time TOCLIENT causes the Setattr RPC
1110 	 * to return NFS_OK, but not set va_mode.
1111 	 */
1112 	if (vap->va_mtime.tv_sec == VNOVAL) {
1113 		vfs_timestamp(&vap->va_mtime);
1114 		vap->va_vaflags |= VA_UTIMES_NULL;
1115 	}
1116 	if (vap->va_atime.tv_sec == VNOVAL)
1117 		vap->va_atime = vap->va_mtime;
1118 	return (1);
1119 }
1120 
1121 /*
1122  * Map nfsv4 errors to errno.h errors.
1123  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1124  * error should only be returned for the Open, Create and Setattr Ops.
1125  * As such, most calls can just pass in 0 for those arguments.
1126  */
1127 APPLESTATIC int
nfscl_maperr(struct thread * td,int error,uid_t uid,gid_t gid)1128 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1129 {
1130 	struct proc *p;
1131 
1132 	if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1133 		return (error);
1134 	if (td != NULL)
1135 		p = td->td_proc;
1136 	else
1137 		p = NULL;
1138 	switch (error) {
1139 	case NFSERR_BADOWNER:
1140 		tprintf(p, LOG_INFO,
1141 		    "No name and/or group mapping for uid,gid:(%d,%d)\n",
1142 		    uid, gid);
1143 		return (EPERM);
1144 	case NFSERR_STALECLIENTID:
1145 	case NFSERR_STALESTATEID:
1146 	case NFSERR_EXPIRED:
1147 	case NFSERR_BADSTATEID:
1148 		printf("nfsv4 recover err returned %d\n", error);
1149 		return (EIO);
1150 	case NFSERR_BADHANDLE:
1151 	case NFSERR_SERVERFAULT:
1152 	case NFSERR_BADTYPE:
1153 	case NFSERR_FHEXPIRED:
1154 	case NFSERR_RESOURCE:
1155 	case NFSERR_MOVED:
1156 	case NFSERR_NOFILEHANDLE:
1157 	case NFSERR_MINORVERMISMATCH:
1158 	case NFSERR_OLDSTATEID:
1159 	case NFSERR_BADSEQID:
1160 	case NFSERR_LEASEMOVED:
1161 	case NFSERR_RECLAIMBAD:
1162 	case NFSERR_BADXDR:
1163 	case NFSERR_BADCHAR:
1164 	case NFSERR_BADNAME:
1165 	case NFSERR_OPILLEGAL:
1166 		printf("nfsv4 client/server protocol prob err=%d\n",
1167 		    error);
1168 		return (EIO);
1169 	default:
1170 		tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1171 		return (EIO);
1172 	};
1173 }
1174 
1175 /*
1176  * Check to see if the process for this owner exists. Return 1 if it doesn't
1177  * and 0 otherwise.
1178  */
1179 int
nfscl_procdoesntexist(u_int8_t * own)1180 nfscl_procdoesntexist(u_int8_t *own)
1181 {
1182 	union {
1183 		u_int32_t	lval;
1184 		u_int8_t	cval[4];
1185 	} tl;
1186 	struct proc *p;
1187 	pid_t pid;
1188 	int ret = 0;
1189 
1190 	tl.cval[0] = *own++;
1191 	tl.cval[1] = *own++;
1192 	tl.cval[2] = *own++;
1193 	tl.cval[3] = *own++;
1194 	pid = tl.lval;
1195 	p = pfind_locked(pid);
1196 	if (p == NULL)
1197 		return (1);
1198 	if (p->p_stats == NULL) {
1199 		PROC_UNLOCK(p);
1200 		return (0);
1201 	}
1202 	tl.cval[0] = *own++;
1203 	tl.cval[1] = *own++;
1204 	tl.cval[2] = *own++;
1205 	tl.cval[3] = *own++;
1206 	if (tl.lval != p->p_stats->p_start.tv_sec) {
1207 		ret = 1;
1208 	} else {
1209 		tl.cval[0] = *own++;
1210 		tl.cval[1] = *own++;
1211 		tl.cval[2] = *own++;
1212 		tl.cval[3] = *own;
1213 		if (tl.lval != p->p_stats->p_start.tv_usec)
1214 			ret = 1;
1215 	}
1216 	PROC_UNLOCK(p);
1217 	return (ret);
1218 }
1219 
1220 /*
1221  * - nfs pseudo system call for the client
1222  */
1223 /*
1224  * MPSAFE
1225  */
1226 static int
nfssvc_nfscl(struct thread * td,struct nfssvc_args * uap)1227 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1228 {
1229 	struct file *fp;
1230 	struct nfscbd_args nfscbdarg;
1231 	struct nfsd_nfscbd_args nfscbdarg2;
1232 	int error;
1233 	struct nameidata nd;
1234 	struct nfscl_dumpmntopts dumpmntopts;
1235 	char *buf;
1236 
1237 	if (uap->flag & NFSSVC_CBADDSOCK) {
1238 		error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1239 		if (error)
1240 			return (error);
1241 		/*
1242 		 * Since we don't know what rights might be required,
1243 		 * pretend that we need them all. It is better to be too
1244 		 * careful than too reckless.
1245 		 */
1246 		if ((error = fget(td, nfscbdarg.sock, CAP_SOCK_ALL, &fp))
1247 		    != 0) {
1248 			return (error);
1249 		}
1250 		if (fp->f_type != DTYPE_SOCKET) {
1251 			fdrop(fp, td);
1252 			return (EPERM);
1253 		}
1254 		error = nfscbd_addsock(fp);
1255 		fdrop(fp, td);
1256 		if (!error && nfscl_enablecallb == 0) {
1257 			nfsv4_cbport = nfscbdarg.port;
1258 			nfscl_enablecallb = 1;
1259 		}
1260 	} else if (uap->flag & NFSSVC_NFSCBD) {
1261 		if (uap->argp == NULL)
1262 			return (EINVAL);
1263 		error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1264 		    sizeof(nfscbdarg2));
1265 		if (error)
1266 			return (error);
1267 		error = nfscbd_nfsd(td, &nfscbdarg2);
1268 	} else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1269 		error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1270 		if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1271 		    dumpmntopts.ndmnt_blen > 1024))
1272 			error = EINVAL;
1273 		if (error == 0)
1274 			error = nfsrv_lookupfilename(&nd,
1275 			    dumpmntopts.ndmnt_fname, td);
1276 		if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1277 		    "nfs") != 0) {
1278 			vput(nd.ni_vp);
1279 			error = EINVAL;
1280 		}
1281 		if (error == 0) {
1282 			buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1283 			nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1284 			    dumpmntopts.ndmnt_blen);
1285 			vput(nd.ni_vp);
1286 			error = copyout(buf, dumpmntopts.ndmnt_buf,
1287 			    dumpmntopts.ndmnt_blen);
1288 			free(buf, M_TEMP);
1289 		}
1290 	} else {
1291 		error = EINVAL;
1292 	}
1293 	return (error);
1294 }
1295 
1296 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1297 
1298 /*
1299  * Called once to initialize data structures...
1300  */
1301 static int
nfscl_modevent(module_t mod,int type,void * data)1302 nfscl_modevent(module_t mod, int type, void *data)
1303 {
1304 	int error = 0;
1305 	static int loaded = 0;
1306 
1307 	switch (type) {
1308 	case MOD_LOAD:
1309 		if (loaded)
1310 			return (0);
1311 		newnfs_portinit();
1312 		mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1313 		    MTX_DEF);
1314 		mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1315 		nfscl_init();
1316 		NFSD_LOCK();
1317 		nfsrvd_cbinit(0);
1318 		NFSD_UNLOCK();
1319 		ncl_call_invalcaches = ncl_invalcaches;
1320 		nfsd_call_nfscl = nfssvc_nfscl;
1321 		loaded = 1;
1322 		break;
1323 
1324 	case MOD_UNLOAD:
1325 		if (nfs_numnfscbd != 0) {
1326 			error = EBUSY;
1327 			break;
1328 		}
1329 
1330 		/*
1331 		 * XXX: Unloading of nfscl module is unsupported.
1332 		 */
1333 #if 0
1334 		ncl_call_invalcaches = NULL;
1335 		nfsd_call_nfscl = NULL;
1336 		/* and get rid of the mutexes */
1337 		mtx_destroy(&nfs_clstate_mutex);
1338 		mtx_destroy(&ncl_iod_mutex);
1339 		loaded = 0;
1340 		break;
1341 #else
1342 		/* FALLTHROUGH */
1343 #endif
1344 	default:
1345 		error = EOPNOTSUPP;
1346 		break;
1347 	}
1348 	return error;
1349 }
1350 static moduledata_t nfscl_mod = {
1351 	"nfscl",
1352 	nfscl_modevent,
1353 	NULL,
1354 };
1355 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1356 
1357 /* So that loader and kldload(2) can find us, wherever we are.. */
1358 MODULE_VERSION(nfscl, 1);
1359 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1360 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1361 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1362 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1363 
1364