xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: nfs_vfsops.c,v 1.196 2008/02/13 09:51:37 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993, 1995
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  *	@(#)nfs_vfsops.c	8.12 (Berkeley) 5/20/95
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: nfs_vfsops.c,v 1.196 2008/02/13 09:51:37 yamt Exp $");
39 
40 #if defined(_KERNEL_OPT)
41 #include "opt_compat_netbsd.h"
42 #include "opt_nfs.h"
43 #endif
44 
45 #include <sys/param.h>
46 #include <sys/ioctl.h>
47 #include <sys/signal.h>
48 #include <sys/proc.h>
49 #include <sys/namei.h>
50 #include <sys/device.h>
51 #include <sys/vnode.h>
52 #include <sys/kernel.h>
53 #include <sys/mount.h>
54 #include <sys/buf.h>
55 #include <sys/mbuf.h>
56 #include <sys/dirent.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/sysctl.h>
60 #include <sys/systm.h>
61 #include <sys/timetc.h>
62 #include <sys/kauth.h>
63 
64 #include <net/if.h>
65 #include <net/route.h>
66 #include <netinet/in.h>
67 
68 #include <nfs/rpcv2.h>
69 #include <nfs/nfsproto.h>
70 #include <nfs/nfsnode.h>
71 #include <nfs/nfs.h>
72 #include <nfs/nfsmount.h>
73 #include <nfs/xdr_subs.h>
74 #include <nfs/nfsm_subs.h>
75 #include <nfs/nfsdiskless.h>
76 #include <nfs/nfs_var.h>
77 
78 extern struct nfsstats nfsstats;
79 extern int nfs_ticks;
80 
81 /*
82  * keep a count of the nfs mounts to generate ficticious drive names
83  * for the per drive stats.
84  */
85 unsigned int nfs_mount_count = 0;
86 
87 /*
88  * nfs vfs operations.
89  */
90 
91 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
92 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
93 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
94 
95 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = {
96 	&nfsv2_vnodeop_opv_desc,
97 	&spec_nfsv2nodeop_opv_desc,
98 	&fifo_nfsv2nodeop_opv_desc,
99 	NULL,
100 };
101 
102 struct vfsops nfs_vfsops = {
103 	MOUNT_NFS,
104 	sizeof (struct nfs_args),
105 	nfs_mount,
106 	nfs_start,
107 	nfs_unmount,
108 	nfs_root,
109 	(void *)eopnotsupp,	/* vfs_quotactl */
110 	nfs_statvfs,
111 	nfs_sync,
112 	nfs_vget,
113 	nfs_fhtovp,
114 	nfs_vptofh,
115 	nfs_vfs_init,
116 	nfs_vfs_reinit,
117 	nfs_vfs_done,
118 	nfs_mountroot,
119 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
120 	vfs_stdextattrctl,
121 	(void *)eopnotsupp,	/* vfs_suspendctl */
122 	genfs_renamelock_enter,
123 	genfs_renamelock_exit,
124 	nfs_vnodeopv_descs,
125 	0,
126 	{ NULL, NULL },
127 };
128 VFS_ATTACH(nfs_vfsops);
129 
130 extern u_int32_t nfs_procids[NFS_NPROCS];
131 extern u_int32_t nfs_prog, nfs_vers;
132 
133 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
134     struct mount **, struct vnode **, struct lwp *));
135 
136 /*
137  * nfs statvfs call
138  */
139 int
140 nfs_statvfs(mp, sbp)
141 	struct mount *mp;
142 	struct statvfs *sbp;
143 {
144 	struct lwp *l = curlwp;
145 	struct vnode *vp;
146 	struct nfs_statfs *sfp;
147 	char *cp;
148 	u_int32_t *tl;
149 	int32_t t1, t2;
150 	char *bpos, *dpos, *cp2;
151 	struct nfsmount *nmp = VFSTONFS(mp);
152 	int error = 0, retattr;
153 #ifdef NFS_V2_ONLY
154 	const int v3 = 0;
155 #else
156 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
157 #endif
158 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
159 	kauth_cred_t cred;
160 	u_quad_t tquad;
161 	struct nfsnode *np;
162 
163 #ifndef nolint
164 	sfp = (struct nfs_statfs *)0;
165 #endif
166 	vp = nmp->nm_vnode;
167 	np = VTONFS(vp);
168 	cred = kauth_cred_alloc();
169 #ifndef NFS_V2_ONLY
170 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
171 		(void)nfs_fsinfo(nmp, vp, cred, l);
172 #endif
173 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
174 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
175 	nfsm_fhtom(np, v3);
176 	nfsm_request(np, NFSPROC_FSSTAT, l, cred);
177 	if (v3)
178 		nfsm_postop_attr(vp, retattr, 0);
179 	if (error) {
180 		if (mrep != NULL) {
181 			if (mrep->m_next != NULL)
182 				printf("nfs_vfsops: nfs_statvfs would lose buffers\n");
183 			m_freem(mrep);
184 		}
185 		goto nfsmout;
186 	}
187 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
188 	sbp->f_flag = nmp->nm_flag;
189 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
190 	if (v3) {
191 		sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE;
192 		tquad = fxdr_hyper(&sfp->sf_tbytes);
193 		sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
194 		tquad = fxdr_hyper(&sfp->sf_fbytes);
195 		sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
196 		tquad = fxdr_hyper(&sfp->sf_abytes);
197 		tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
198 		sbp->f_bresvd = sbp->f_bfree - tquad;
199 		sbp->f_bavail = tquad;
200 #ifdef COMPAT_20
201 		/* Handle older NFS servers returning negative values */
202 		if ((quad_t)sbp->f_bavail < 0)
203 			sbp->f_bavail = 0;
204 #endif
205 		tquad = fxdr_hyper(&sfp->sf_tfiles);
206 		sbp->f_files = tquad;
207 		tquad = fxdr_hyper(&sfp->sf_ffiles);
208 		sbp->f_ffree = tquad;
209 		sbp->f_favail = tquad;
210 		sbp->f_fresvd = 0;
211 		sbp->f_namemax = MAXNAMLEN;
212 	} else {
213 		sbp->f_bsize = NFS_FABLKSIZE;
214 		sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
215 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
216 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
217 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
218 		sbp->f_fresvd = 0;
219 		sbp->f_files = 0;
220 		sbp->f_ffree = 0;
221 		sbp->f_favail = 0;
222 		sbp->f_fresvd = 0;
223 		sbp->f_namemax = MAXNAMLEN;
224 	}
225 	copy_statvfs_info(sbp, mp);
226 	nfsm_reqdone;
227 	kauth_cred_free(cred);
228 	return (error);
229 }
230 
231 #ifndef NFS_V2_ONLY
232 /*
233  * nfs version 3 fsinfo rpc call
234  */
235 int
236 nfs_fsinfo(nmp, vp, cred, l)
237 	struct nfsmount *nmp;
238 	struct vnode *vp;
239 	kauth_cred_t cred;
240 	struct lwp *l;
241 {
242 	struct nfsv3_fsinfo *fsp;
243 	char *cp;
244 	int32_t t1, t2;
245 	u_int32_t *tl, pref, xmax;
246 	char *bpos, *dpos, *cp2;
247 	int error = 0, retattr;
248 	struct mbuf *mreq, *mrep, *md, *mb;
249 	u_int64_t maxfsize;
250 	struct nfsnode *np = VTONFS(vp);
251 
252 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
253 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
254 	nfsm_fhtom(np, 1);
255 	nfsm_request(np, NFSPROC_FSINFO, l, cred);
256 	nfsm_postop_attr(vp, retattr, 0);
257 	if (!error) {
258 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
259 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
260 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
261 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
262 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
263 				~(NFS_FABLKSIZE - 1);
264 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
265 		if (xmax < nmp->nm_wsize && xmax > 0) {
266 			nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1);
267 			if (nmp->nm_wsize == 0)
268 				nmp->nm_wsize = xmax;
269 		}
270 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
271 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
272 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
273 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
274 				~(NFS_FABLKSIZE - 1);
275 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
276 		if (xmax < nmp->nm_rsize && xmax > 0) {
277 			nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1);
278 			if (nmp->nm_rsize == 0)
279 				nmp->nm_rsize = xmax;
280 		}
281 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
282 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
283 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
284 				~(NFS_DIRFRAGSIZ - 1);
285 		if (xmax < nmp->nm_readdirsize && xmax > 0) {
286 			nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1);
287 			if (nmp->nm_readdirsize == 0)
288 				nmp->nm_readdirsize = xmax;
289 		}
290 		/* XXX */
291 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
292 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
293 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
294 			nmp->nm_maxfilesize = maxfsize;
295 		nmp->nm_mountp->mnt_fs_bshift =
296 		    ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
297 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
298 	}
299 	nfsm_reqdone;
300 	return (error);
301 }
302 #endif
303 
304 /*
305  * Mount a remote root fs via. NFS.  It goes like this:
306  * - Call nfs_boot_init() to fill in the nfs_diskless struct
307  * - build the rootfs mount point and call mountnfs() to do the rest.
308  */
309 int
310 nfs_mountroot()
311 {
312 	struct timespec ts;
313 	struct nfs_diskless *nd;
314 	struct vattr attr;
315 	struct mount *mp;
316 	struct vnode *vp;
317 	struct lwp *l;
318 	long n;
319 	int error;
320 
321 	l = curlwp; /* XXX */
322 
323 	if (device_class(root_device) != DV_IFNET)
324 		return (ENODEV);
325 
326 	/*
327 	 * XXX time must be non-zero when we init the interface or else
328 	 * the arp code will wedge.  [Fixed now in if_ether.c]
329 	 * However, the NFS attribute cache gives false "hits" when the
330 	 * current time < nfs_attrtimeo(nmp, np) so keep this in for now.
331 	 */
332 	if (time_second < NFS_MAXATTRTIMO) {
333 		ts.tv_sec = NFS_MAXATTRTIMO;
334 		ts.tv_nsec = 0;
335 		tc_setclock(&ts);
336 	}
337 
338 	/*
339 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
340 	 * Side effect:  Finds and configures a network interface.
341 	 */
342 	nd = kmem_zalloc(sizeof(*nd), KM_SLEEP);
343 	error = nfs_boot_init(nd, l);
344 	if (error) {
345 		kmem_free(nd, sizeof(*nd));
346 		return (error);
347 	}
348 
349 	/*
350 	 * Create the root mount point.
351 	 */
352 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l);
353 	if (error)
354 		goto out;
355 	printf("root on %s\n", nd->nd_root.ndm_host);
356 
357 	/*
358 	 * Link it into the mount list.
359 	 */
360 	mutex_enter(&mountlist_lock);
361 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
362 	mutex_exit(&mountlist_lock);
363 	rootvp = vp;
364 	mp->mnt_vnodecovered = NULLVP;
365 	vfs_unbusy(mp, false);
366 
367 	/* Get root attributes (for the time). */
368 	error = VOP_GETATTR(vp, &attr, l->l_cred);
369 	if (error)
370 		panic("nfs_mountroot: getattr for root");
371 	n = attr.va_atime.tv_sec;
372 #ifdef	DEBUG
373 	printf("root time: 0x%lx\n", n);
374 #endif
375 	setrootfstime(n);
376 
377 out:
378 	if (error)
379 		nfs_boot_cleanup(nd, l);
380 	kmem_free(nd, sizeof(*nd));
381 	return (error);
382 }
383 
384 /*
385  * Internal version of mount system call for diskless setup.
386  * Separate function because we used to call it twice.
387  * (once for root and once for swap)
388  */
389 static int
390 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, l)
391 	struct nfs_dlmount *ndmntp;
392 	const char *mntname;	/* mount point name */
393 	struct mount **mpp;
394 	struct vnode **vpp;
395 	struct lwp *l;
396 {
397 	struct mount *mp;
398 	struct mbuf *m;
399 	int error;
400 
401 	vfs_rootmountalloc(MOUNT_NFS, mntname, &mp);
402 
403 	mp->mnt_op = &nfs_vfsops;
404 
405 	/*
406 	 * Historical practice expects NFS root file systems to
407 	 * be initially mounted r/w.
408 	 */
409 	mp->mnt_flag &= ~MNT_RDONLY;
410 
411 	/* Get mbuf for server sockaddr. */
412 	m = m_get(M_WAIT, MT_SONAME);
413 	if (m == NULL)
414 		panic("nfs_mountroot: mget soname for %s", mntname);
415 	MCLAIM(m, &nfs_mowner);
416 	memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr,
417 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
418 
419 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
420 			 ndmntp->ndm_args.hostname, vpp, l);
421 	if (error) {
422 		vfs_unbusy(mp, false);
423 		vfs_destroy(mp);
424 		printf("nfs_mountroot: mount %s failed: %d\n",
425 		       mntname, error);
426 	} else
427 		*mpp = mp;
428 
429 	return (error);
430 }
431 
432 void
433 nfs_decode_args(nmp, argp, l)
434 	struct nfsmount *nmp;
435 	struct nfs_args *argp;
436 	struct lwp *l;
437 {
438 	int s;
439 	int adjsock;
440 	int maxio;
441 
442 	s = splsoftnet();
443 
444 	/*
445 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
446 	 * no sense in that context.
447 	 */
448 	if (argp->sotype == SOCK_STREAM)
449 		argp->flags &= ~NFSMNT_NOCONN;
450 
451 	/*
452 	 * Cookie translation is not needed for v2, silently ignore it.
453 	 */
454 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
455 	    NFSMNT_XLATECOOKIE)
456 		argp->flags &= ~NFSMNT_XLATECOOKIE;
457 
458 	/* Re-bind if rsrvd port requested and wasn't on one */
459 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
460 		  && (argp->flags & NFSMNT_RESVPORT);
461 	/* Also re-bind if we're switching to/from a connected UDP socket */
462 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
463 		    (argp->flags & NFSMNT_NOCONN));
464 
465 	/* Update flags. */
466 	nmp->nm_flag = argp->flags;
467 	splx(s);
468 
469 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
470 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
471 		if (nmp->nm_timeo < NFS_MINTIMEO)
472 			nmp->nm_timeo = NFS_MINTIMEO;
473 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
474 			nmp->nm_timeo = NFS_MAXTIMEO;
475 	}
476 
477 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
478 		nmp->nm_retry = argp->retrans;
479 		if (nmp->nm_retry > NFS_MAXREXMIT)
480 			nmp->nm_retry = NFS_MAXREXMIT;
481 	}
482 
483 #ifndef NFS_V2_ONLY
484 	if (argp->flags & NFSMNT_NFSV3) {
485 		if (argp->sotype == SOCK_DGRAM)
486 			maxio = NFS_MAXDGRAMDATA;
487 		else
488 			maxio = NFS_MAXDATA;
489 	} else
490 #endif
491 		maxio = NFS_V2MAXDATA;
492 
493 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
494 		int osize = nmp->nm_wsize;
495 		nmp->nm_wsize = argp->wsize;
496 		/* Round down to multiple of blocksize */
497 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
498 		if (nmp->nm_wsize <= 0)
499 			nmp->nm_wsize = NFS_FABLKSIZE;
500 		adjsock |= (nmp->nm_wsize != osize);
501 	}
502 	if (nmp->nm_wsize > maxio)
503 		nmp->nm_wsize = maxio;
504 	if (nmp->nm_wsize > MAXBSIZE)
505 		nmp->nm_wsize = MAXBSIZE;
506 
507 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
508 		int osize = nmp->nm_rsize;
509 		nmp->nm_rsize = argp->rsize;
510 		/* Round down to multiple of blocksize */
511 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
512 		if (nmp->nm_rsize <= 0)
513 			nmp->nm_rsize = NFS_FABLKSIZE;
514 		adjsock |= (nmp->nm_rsize != osize);
515 	}
516 	if (nmp->nm_rsize > maxio)
517 		nmp->nm_rsize = maxio;
518 	if (nmp->nm_rsize > MAXBSIZE)
519 		nmp->nm_rsize = MAXBSIZE;
520 
521 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
522 		nmp->nm_readdirsize = argp->readdirsize;
523 		/* Round down to multiple of minimum blocksize */
524 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
525 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
526 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
527 		/* Bigger than buffer size makes no sense */
528 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
529 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
530 	} else if (argp->flags & NFSMNT_RSIZE)
531 		nmp->nm_readdirsize = nmp->nm_rsize;
532 
533 	if (nmp->nm_readdirsize > maxio)
534 		nmp->nm_readdirsize = maxio;
535 
536 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
537 		argp->maxgrouplist <= NFS_MAXGRPS)
538 		nmp->nm_numgrps = argp->maxgrouplist;
539 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
540 		argp->readahead <= NFS_MAXRAHEAD)
541 		nmp->nm_readahead = argp->readahead;
542 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
543 		argp->deadthresh <= NFS_NEVERDEAD)
544 		nmp->nm_deadthresh = argp->deadthresh;
545 
546 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
547 		    (nmp->nm_soproto != argp->proto));
548 	nmp->nm_sotype = argp->sotype;
549 	nmp->nm_soproto = argp->proto;
550 
551 	if (nmp->nm_so && adjsock) {
552 		nfs_safedisconnect(nmp);
553 		if (nmp->nm_sotype == SOCK_DGRAM)
554 			while (nfs_connect(nmp, (struct nfsreq *)0, l)) {
555 				printf("nfs_args: retrying connect\n");
556 				kpause("nfscn3", false, hz, NULL);
557 			}
558 	}
559 }
560 
561 /*
562  * VFS Operations.
563  *
564  * mount system call
565  * It seems a bit dumb to copyinstr() the host and path here and then
566  * memcpy() them in mountnfs(), but I wanted to detect errors before
567  * doing the sockargs() call because sockargs() allocates an mbuf and
568  * an error after that means that I have to release the mbuf.
569  */
570 /* ARGSUSED */
571 int
572 nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
573 {
574 	struct lwp *l = curlwp;
575 	int error;
576 	struct nfs_args *args = data;
577 	struct mbuf *nam;
578 	struct nfsmount *nmp = VFSTONFS(mp);
579 	struct sockaddr *sa;
580 	struct vnode *vp;
581 	char *pth, *hst;
582 	struct proc *p;
583 	size_t len;
584 	u_char *nfh;
585 
586 	if (*data_len < sizeof *args)
587 		return EINVAL;
588 
589 	p = l->l_proc;
590 	if (mp->mnt_flag & MNT_GETARGS) {
591 
592 		if (nmp == NULL)
593 			return (EIO);
594 		if (args->addr != NULL) {
595 			sa = mtod(nmp->nm_nam, struct sockaddr *);
596 			error = copyout(sa, args->addr, sa->sa_len);
597 			if (error)
598 				return (error);
599 			args->addrlen = sa->sa_len;
600 		} else
601 			args->addrlen = 0;
602 
603 		args->version = NFS_ARGSVERSION;
604 		args->sotype = nmp->nm_sotype;
605 		args->proto = nmp->nm_soproto;
606 		args->fh = NULL;
607 		args->fhsize = 0;
608 		args->flags = nmp->nm_flag;
609 		args->wsize = nmp->nm_wsize;
610 		args->rsize = nmp->nm_rsize;
611 		args->readdirsize = nmp->nm_readdirsize;
612 		args->timeo = nmp->nm_timeo;
613 		args->retrans = nmp->nm_retry;
614 		args->maxgrouplist = nmp->nm_numgrps;
615 		args->readahead = nmp->nm_readahead;
616 		args->leaseterm = 0; /* dummy */
617 		args->deadthresh = nmp->nm_deadthresh;
618 		args->hostname = NULL;
619 		*data_len = sizeof *args;
620 		return 0;
621 	}
622 
623 	if (args->version != NFS_ARGSVERSION)
624 		return (EPROGMISMATCH);
625 	if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB))
626 		return (EPROGUNAVAIL);
627 #ifdef NFS_V2_ONLY
628 	if (args->flags & NFSMNT_NFSV3)
629 		return (EPROGMISMATCH);
630 #endif
631 	if (mp->mnt_flag & MNT_UPDATE) {
632 		if (nmp == NULL)
633 			return (EIO);
634 		/*
635 		 * When doing an update, we can't change from or to
636 		 * v3, or change cookie translation
637 		 */
638 		args->flags = (args->flags & ~(NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)) |
639 		    (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_XLATECOOKIE));
640 		nfs_decode_args(nmp, args, l);
641 		return (0);
642 	}
643 	if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX)
644 		return (EINVAL);
645 	MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK);
646 	error = copyin(args->fh, nfh, args->fhsize);
647 	if (error)
648 		goto free_nfh;
649 	MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK);
650 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
651 	if (error)
652 		goto free_pth;
653 	memset(&pth[len], 0, MNAMELEN - len);
654 	MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK);
655 	error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len);
656 	if (error)
657 		goto free_hst;
658 	memset(&hst[len], 0, MNAMELEN - len);
659 	/* sockargs() call must be after above copyin() calls */
660 	error = sockargs(&nam, args->addr, args->addrlen, MT_SONAME);
661 	if (error)
662 		goto free_hst;
663 	MCLAIM(nam, &nfs_mowner);
664 	args->fh = nfh;
665 	error = mountnfs(args, mp, nam, pth, hst, &vp, l);
666 
667 free_hst:
668 	FREE(hst, M_TEMP);
669 free_pth:
670 	FREE(pth, M_TEMP);
671 free_nfh:
672 	FREE(nfh, M_TEMP);
673 
674 	return (error);
675 }
676 
677 /*
678  * Common code for mount and mountroot
679  */
680 int
681 mountnfs(argp, mp, nam, pth, hst, vpp, l)
682 	struct nfs_args *argp;
683 	struct mount *mp;
684 	struct mbuf *nam;
685 	const char *pth, *hst;
686 	struct vnode **vpp;
687 	struct lwp *l;
688 {
689 	struct nfsmount *nmp;
690 	struct nfsnode *np;
691 	int error;
692 	struct vattr *attrs;
693 	kauth_cred_t cr;
694 	char iosname[IOSTATNAMELEN];
695 
696 	/*
697 	 * If the number of nfs iothreads to use has never
698 	 * been set, create a reasonable number of them.
699 	 */
700 
701 	if (nfs_niothreads < 0) {
702 		nfs_set_niothreads(NFS_DEFAULT_NIOTHREADS);
703 	}
704 
705 	if (mp->mnt_flag & MNT_UPDATE) {
706 		nmp = VFSTONFS(mp);
707 		/* update paths, file handles, etc, here	XXX */
708 		m_freem(nam);
709 		return (0);
710 	} else {
711 		nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP);
712 		mp->mnt_data = nmp;
713 		TAILQ_INIT(&nmp->nm_uidlruhead);
714 		TAILQ_INIT(&nmp->nm_bufq);
715 		rw_init(&nmp->nm_writeverflock);
716 		mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE);
717 		cv_init(&nmp->nm_rcvcv, "nfsrcv");
718 		cv_init(&nmp->nm_sndcv, "nfssnd");
719 		cv_init(&nmp->nm_aiocv, "nfsaio");
720 		cv_init(&nmp->nm_disconcv, "nfsdis");
721 	}
722 	vfs_getnewfsid(mp);
723 	nmp->nm_mountp = mp;
724 
725 #ifndef NFS_V2_ONLY
726 	if ((argp->flags & NFSMNT_NFSV3) == 0)
727 #endif
728 	{
729 		if (argp->fhsize != NFSX_V2FH) {
730 			return EINVAL;
731 		}
732 	}
733 
734 	/*
735 	 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
736 	 * will overwrite this.
737 	 */
738 	nmp->nm_maxfilesize = 0xffffffffLL;
739 
740 	nmp->nm_timeo = NFS_TIMEO;
741 	nmp->nm_retry = NFS_RETRANS;
742 	nmp->nm_wsize = NFS_WSIZE;
743 	nmp->nm_rsize = NFS_RSIZE;
744 	nmp->nm_readdirsize = NFS_READDIRSIZE;
745 	nmp->nm_numgrps = NFS_MAXGRPS;
746 	nmp->nm_readahead = NFS_DEFRAHEAD;
747 	nmp->nm_deadthresh = NFS_DEFDEADTHRESH;
748 	error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE,
749 	    mp->mnt_op->vfs_name, mp, l);
750 	if (error)
751 		goto bad;
752 	nmp->nm_nam = nam;
753 
754 	/* Set up the sockets and per-host congestion */
755 	nmp->nm_sotype = argp->sotype;
756 	nmp->nm_soproto = argp->proto;
757 
758 	nfs_decode_args(nmp, argp, l);
759 
760 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
761 	mp->mnt_dev_bshift = DEV_BSHIFT;
762 
763 	/*
764 	 * For Connection based sockets (TCP,...) defer the connect until
765 	 * the first request, in case the server is not responding.
766 	 */
767 	if (nmp->nm_sotype == SOCK_DGRAM &&
768 		(error = nfs_connect(nmp, (struct nfsreq *)0, l)))
769 		goto bad;
770 
771 	/*
772 	 * This is silly, but it has to be set so that vinifod() works.
773 	 * We do not want to do an nfs_statvfs() here since we can get
774 	 * stuck on a dead server and we are holding a lock on the mount
775 	 * point.
776 	 */
777 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
778 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
779 	if (error)
780 		goto bad;
781 	*vpp = NFSTOV(np);
782 	MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK);
783 	VOP_GETATTR(*vpp, attrs, l->l_cred);
784 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
785 		cr = kauth_cred_alloc();
786 		kauth_cred_setuid(cr, attrs->va_uid);
787 		kauth_cred_seteuid(cr, attrs->va_uid);
788 		kauth_cred_setsvuid(cr, attrs->va_uid);
789 		kauth_cred_setgid(cr, attrs->va_gid);
790 		kauth_cred_setegid(cr, attrs->va_gid);
791 		kauth_cred_setsvgid(cr, attrs->va_gid);
792 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, l, cr);
793 		kauth_cred_free(cr);
794 	}
795 	FREE(attrs, M_TEMP);
796 
797 	/*
798 	 * A reference count is needed on the nfsnode representing the
799 	 * remote root.  If this object is not persistent, then backward
800 	 * traversals of the mount point (i.e. "..") will not work if
801 	 * the nfsnode gets flushed out of the cache. Ufs does not have
802 	 * this problem, because one can identify root inodes by their
803 	 * number == ROOTINO (2). So, just unlock, but no rele.
804 	 */
805 
806 	nmp->nm_vnode = *vpp;
807 	VOP_UNLOCK(*vpp, 0);
808 
809 	snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++);
810 	nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname);
811 
812 	return (0);
813 bad:
814 	nfs_disconnect(nmp);
815 	rw_destroy(&nmp->nm_writeverflock);
816 	mutex_destroy(&nmp->nm_lock);
817 	cv_destroy(&nmp->nm_rcvcv);
818 	cv_destroy(&nmp->nm_sndcv);
819 	cv_destroy(&nmp->nm_aiocv);
820 	cv_destroy(&nmp->nm_disconcv);
821 	kmem_free(nmp, sizeof(*nmp));
822 	m_freem(nam);
823 	return (error);
824 }
825 
826 /*
827  * unmount system call
828  */
829 int
830 nfs_unmount(struct mount *mp, int mntflags)
831 {
832 	struct nfsmount *nmp;
833 	struct vnode *vp;
834 	int error, flags = 0;
835 
836 	if (mntflags & MNT_FORCE)
837 		flags |= FORCECLOSE;
838 	nmp = VFSTONFS(mp);
839 	/*
840 	 * Goes something like this..
841 	 * - Check for activity on the root vnode (other than ourselves).
842 	 * - Call vflush() to clear out vnodes for this file system,
843 	 *   except for the root vnode.
844 	 * - Decrement reference on the vnode representing remote root.
845 	 * - Close the socket
846 	 * - Free up the data structures
847 	 */
848 	/*
849 	 * We need to decrement the ref. count on the nfsnode representing
850 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
851 	 * has done vput on this vnode, otherwise we would get deadlock!
852 	 */
853 	vp = nmp->nm_vnode;
854 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
855 	if (error != 0)
856 		return error;
857 
858 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
859 		vput(vp);
860 		return (EBUSY);
861 	}
862 
863 	error = vflush(mp, vp, flags);
864 	if (error) {
865 		vput(vp);
866 		return (error);
867 	}
868 
869 	/*
870 	 * We are now committed to the unmount; mark the mount structure
871 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
872 	 * will go away cleanly.
873 	 */
874 	nmp->nm_iflag |= NFSMNT_DISMNT;
875 
876 	/*
877 	 * Clean up the stats... note that we carefully avoid decrementing
878 	 * nfs_mount_count here for good reason - we may not be unmounting
879 	 * the last thing mounted.
880 	 */
881 	iostat_free(nmp->nm_stats);
882 
883 	/*
884 	 * There are two reference counts to get rid of here
885 	 * (see comment in mountnfs()).
886 	 */
887 	vput(vp);
888 	vgone(vp);
889 	nfs_disconnect(nmp);
890 	m_freem(nmp->nm_nam);
891 
892 	rw_destroy(&nmp->nm_writeverflock);
893 	mutex_destroy(&nmp->nm_lock);
894 	cv_destroy(&nmp->nm_rcvcv);
895 	cv_destroy(&nmp->nm_sndcv);
896 	cv_destroy(&nmp->nm_aiocv);
897 	cv_destroy(&nmp->nm_disconcv);
898 	kmem_free(nmp, sizeof(*nmp));
899 	return (0);
900 }
901 
902 /*
903  * Return root of a filesystem
904  */
905 int
906 nfs_root(mp, vpp)
907 	struct mount *mp;
908 	struct vnode **vpp;
909 {
910 	struct vnode *vp;
911 	struct nfsmount *nmp;
912 	int error;
913 
914 	nmp = VFSTONFS(mp);
915 	vp = nmp->nm_vnode;
916 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
917 	if (error != 0)
918 		return error;
919 	if (vp->v_type == VNON)
920 		vp->v_type = VDIR;
921 	vp->v_vflag = VV_ROOT;
922 	*vpp = vp;
923 	return (0);
924 }
925 
926 extern int syncprt;
927 
928 /*
929  * Flush out the buffer cache
930  */
931 /* ARGSUSED */
932 int
933 nfs_sync(mp, waitfor, cred)
934 	struct mount *mp;
935 	int waitfor;
936 	kauth_cred_t cred;
937 {
938 	struct vnode *vp, *mvp;
939 	int error, allerror = 0;
940 
941 	/*
942 	 * Force stale buffer cache information to be flushed.
943 	 */
944 	if ((mvp = vnalloc(mp)) == NULL)
945 		return (ENOMEM);
946 loop:
947 	/*
948 	 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
949 	 * and vclean() can be called indirectly
950 	 */
951 	mutex_enter(&mntvnode_lock);
952 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) {
953 		vmark(mvp, vp);
954 		if (vp->v_mount != mp || vismarker(vp))
955 			continue;
956 		mutex_enter(&vp->v_interlock);
957 		/* XXX MNT_LAZY cannot be right? */
958 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
959 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
960 		     UVM_OBJ_IS_CLEAN(&vp->v_uobj))) {
961 			mutex_exit(&vp->v_interlock);
962 			continue;
963 		}
964 		mutex_exit(&mntvnode_lock);
965 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
966 			(void)vunmark(mvp);
967 			goto loop;
968 		}
969 		error = VOP_FSYNC(vp, cred,
970 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0);
971 		if (error)
972 			allerror = error;
973 		vput(vp);
974 		mutex_enter(&mntvnode_lock);
975 	}
976 	mutex_exit(&mntvnode_lock);
977 	vnfree(mvp);
978 	return (allerror);
979 }
980 
981 /*
982  * NFS flat namespace lookup.
983  * Currently unsupported.
984  */
985 /* ARGSUSED */
986 int
987 nfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
988 {
989 
990 	return (EOPNOTSUPP);
991 }
992 
993 /*
994  * Do that sysctl thang...
995  */
996 static int
997 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS)
998 {
999 	struct sysctlnode node;
1000 	int val;
1001 	int error;
1002 
1003 	val = nfs_niothreads;
1004 	node = *rnode;
1005 	node.sysctl_data = &val;
1006         error = sysctl_lookup(SYSCTLFN_CALL(&node));
1007 	if (error || newp == NULL)
1008 		return error;
1009 
1010 	return nfs_set_niothreads(val);
1011 }
1012 
1013 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup")
1014 {
1015 
1016 	sysctl_createv(clog, 0, NULL, NULL,
1017 		       CTLFLAG_PERMANENT,
1018 		       CTLTYPE_NODE, "vfs", NULL,
1019 		       NULL, 0, NULL, 0,
1020 		       CTL_VFS, CTL_EOL);
1021 	sysctl_createv(clog, 0, NULL, NULL,
1022 		       CTLFLAG_PERMANENT,
1023 		       CTLTYPE_NODE, "nfs",
1024 		       SYSCTL_DESCR("NFS vfs options"),
1025 		       NULL, 0, NULL, 0,
1026 		       CTL_VFS, 2, CTL_EOL);
1027 	/*
1028 	 * XXX the "2" above could be dynamic, thereby eliminating one
1029 	 * more instance of the "number to vfs" mapping problem, but
1030 	 * "2" is the order as taken from sys/mount.h
1031 	 */
1032 
1033 	sysctl_createv(clog, 0, NULL, NULL,
1034 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1035 		       CTLTYPE_STRUCT, "nfsstats",
1036 		       SYSCTL_DESCR("NFS operation statistics"),
1037 		       NULL, 0, &nfsstats, sizeof(nfsstats),
1038 		       CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL);
1039 	sysctl_createv(clog, 0, NULL, NULL,
1040 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1041 		       CTLTYPE_INT, "iothreads",
1042 		       SYSCTL_DESCR("Number of NFS client processes desired"),
1043 		       sysctl_vfs_nfs_iothreads, 0, NULL, 0,
1044 		       CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL);
1045 }
1046 
1047 /* ARGSUSED */
1048 int
1049 nfs_fhtovp(struct mount *mp, struct fid *fid, struct vnode **vpp)
1050 {
1051 	size_t fidsize;
1052 	size_t fhsize;
1053 	struct nfsnode *np;
1054 	int error;
1055 	struct vattr va;
1056 
1057 	fidsize = fid->fid_len;
1058 	if (fidsize < sizeof(*fid)) {
1059 		return EINVAL;
1060 	}
1061 	fhsize = fidsize - sizeof(*fid);
1062 	if ((fhsize % NFSX_UNSIGNED) != 0) {
1063 		return EINVAL;
1064 	}
1065 	if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) {
1066 		if (fhsize > NFSX_V3FHMAX || fhsize == 0) {
1067 			return EINVAL;
1068 		}
1069 	} else {
1070 		if (fhsize != NFSX_V2FH) {
1071 			return EINVAL;
1072 		}
1073 	}
1074 	error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np);
1075 	if (error) {
1076 		return error;
1077 	}
1078 	*vpp = NFSTOV(np);
1079 	error = VOP_GETATTR(*vpp, &va, kauth_cred_get());
1080 	if (error != 0) {
1081 		vput(*vpp);
1082 	}
1083 	return error;
1084 }
1085 
1086 /* ARGSUSED */
1087 int
1088 nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize)
1089 {
1090 	struct nfsnode *np;
1091 	struct fid *fid;
1092 	size_t fidsize;
1093 	int error = 0;
1094 
1095 	np = VTONFS(vp);
1096 	fidsize = sizeof(*fid) + np->n_fhsize;
1097 	if (*bufsize < fidsize) {
1098 		error = E2BIG;
1099 	}
1100 	*bufsize = fidsize;
1101 	if (error == 0) {
1102 		struct fid fid_store;
1103 
1104 		fid = &fid_store;
1105 		memset(fid, 0, sizeof(*fid));
1106 		fid->fid_len = fidsize;
1107 		memcpy(buf, fid, sizeof(*fid));
1108 		memcpy(buf->fid_data, np->n_fhp, np->n_fhsize);
1109 	}
1110 	return error;
1111 }
1112 
1113 /*
1114  * Vfs start routine, a no-op.
1115  */
1116 /* ARGSUSED */
1117 int
1118 nfs_start(struct mount *mp, int flags)
1119 {
1120 
1121 	return (0);
1122 }
1123