xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision b5677b36047b601b9addaaa494a58ceae82c2a6c)
1 /*	$NetBSD: nfs_vfsops.c,v 1.210 2009/03/15 17:20:10 cegger 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 1.210 2009/03/15 17:20:10 cegger Exp $");
39 
40 #if defined(_KERNEL_OPT)
41 #include "opt_nfs.h"
42 #endif
43 
44 #include <sys/param.h>
45 #include <sys/ioctl.h>
46 #include <sys/signal.h>
47 #include <sys/proc.h>
48 #include <sys/namei.h>
49 #include <sys/device.h>
50 #include <sys/vnode.h>
51 #include <sys/kernel.h>
52 #include <sys/mount.h>
53 #include <sys/buf.h>
54 #include <sys/mbuf.h>
55 #include <sys/dirent.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/sysctl.h>
59 #include <sys/systm.h>
60 #include <sys/timetc.h>
61 #include <sys/kauth.h>
62 #include <sys/module.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 MODULE(MODULE_CLASS_VFS, nfs, NULL);
79 
80 extern struct nfsstats nfsstats;
81 extern int nfs_ticks;
82 
83 /*
84  * keep a count of the nfs mounts to generate ficticious drive names
85  * for the per drive stats.
86  */
87 unsigned int nfs_mount_count = 0;
88 
89 /*
90  * nfs vfs operations.
91  */
92 
93 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
94 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
95 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
96 
97 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = {
98 	&nfsv2_vnodeop_opv_desc,
99 	&spec_nfsv2nodeop_opv_desc,
100 	&fifo_nfsv2nodeop_opv_desc,
101 	NULL,
102 };
103 
104 struct vfsops nfs_vfsops = {
105 	MOUNT_NFS,
106 	sizeof (struct nfs_args),
107 	nfs_mount,
108 	nfs_start,
109 	nfs_unmount,
110 	nfs_root,
111 	(void *)eopnotsupp,	/* vfs_quotactl */
112 	nfs_statvfs,
113 	nfs_sync,
114 	nfs_vget,
115 	nfs_fhtovp,
116 	nfs_vptofh,
117 	nfs_vfs_init,
118 	NULL,
119 	nfs_vfs_done,
120 	nfs_mountroot,
121 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
122 	vfs_stdextattrctl,
123 	(void *)eopnotsupp,	/* vfs_suspendctl */
124 	genfs_renamelock_enter,
125 	genfs_renamelock_exit,
126 	(void *)eopnotsupp,
127 	nfs_vnodeopv_descs,
128 	0,
129 	{ NULL, NULL },
130 };
131 
132 extern u_int32_t nfs_procids[NFS_NPROCS];
133 extern u_int32_t nfs_prog, nfs_vers;
134 static struct sysctllog *nfs_clog;
135 
136 static int nfs_mount_diskless(struct nfs_dlmount *, const char *,
137     struct mount **, struct vnode **, struct lwp *);
138 static void nfs_sysctl_init(void);
139 static void nfs_sysctl_fini(void);
140 
141 static int
142 nfs_modcmd(modcmd_t cmd, void *arg)
143 {
144 	int error;
145 
146 	switch (cmd) {
147 	case MODULE_CMD_INIT:
148 		error = vfs_attach(&nfs_vfsops);
149 		if (error == 0) {
150 			nfs_sysctl_init();
151 		}
152 		return error;
153 	case MODULE_CMD_FINI:
154 		error = vfs_detach(&nfs_vfsops);
155 		if (error == 0) {
156 			nfs_sysctl_fini();
157 		}
158 		return error;
159 	default:
160 		return ENOTTY;
161 	}
162 }
163 
164 /*
165  * nfs statvfs call
166  */
167 int
168 nfs_statvfs(struct mount *mp, struct statvfs *sbp)
169 {
170 	struct lwp *l = curlwp;
171 	struct vnode *vp;
172 	struct nfs_statfs *sfp;
173 	char *cp;
174 	u_int32_t *tl;
175 	int32_t t1, t2;
176 	char *bpos, *dpos, *cp2;
177 	struct nfsmount *nmp = VFSTONFS(mp);
178 	int error = 0, retattr;
179 #ifdef NFS_V2_ONLY
180 	const int v3 = 0;
181 #else
182 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
183 #endif
184 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
185 	kauth_cred_t cred;
186 	u_quad_t tquad;
187 	struct nfsnode *np;
188 
189 #ifndef nolint
190 	sfp = (struct nfs_statfs *)0;
191 #endif
192 	vp = nmp->nm_vnode;
193 	np = VTONFS(vp);
194 	cred = kauth_cred_alloc();
195 #ifndef NFS_V2_ONLY
196 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
197 		(void)nfs_fsinfo(nmp, vp, cred, l);
198 #endif
199 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
200 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
201 	nfsm_fhtom(np, v3);
202 	nfsm_request(np, NFSPROC_FSSTAT, l, cred);
203 	if (v3)
204 		nfsm_postop_attr(vp, retattr, 0);
205 	if (error) {
206 		if (mrep != NULL) {
207 			if (mrep->m_next != NULL)
208 				printf("nfs_vfsops: nfs_statvfs would lose buffers\n");
209 			m_freem(mrep);
210 		}
211 		goto nfsmout;
212 	}
213 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
214 	sbp->f_flag = nmp->nm_flag;
215 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
216 	if (v3) {
217 		sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE;
218 		tquad = fxdr_hyper(&sfp->sf_tbytes);
219 		sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
220 		tquad = fxdr_hyper(&sfp->sf_fbytes);
221 		sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
222 		tquad = fxdr_hyper(&sfp->sf_abytes);
223 		tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
224 		sbp->f_bresvd = sbp->f_bfree - tquad;
225 		sbp->f_bavail = tquad;
226 		/* Handle older NFS servers returning negative values */
227 		if ((quad_t)sbp->f_bavail < 0)
228 			sbp->f_bavail = 0;
229 		tquad = fxdr_hyper(&sfp->sf_tfiles);
230 		sbp->f_files = tquad;
231 		tquad = fxdr_hyper(&sfp->sf_ffiles);
232 		sbp->f_ffree = tquad;
233 		sbp->f_favail = tquad;
234 		sbp->f_fresvd = 0;
235 		sbp->f_namemax = MAXNAMLEN;
236 	} else {
237 		sbp->f_bsize = NFS_FABLKSIZE;
238 		sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
239 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
240 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
241 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
242 		sbp->f_fresvd = 0;
243 		sbp->f_files = 0;
244 		sbp->f_ffree = 0;
245 		sbp->f_favail = 0;
246 		sbp->f_fresvd = 0;
247 		sbp->f_namemax = MAXNAMLEN;
248 	}
249 	copy_statvfs_info(sbp, mp);
250 	nfsm_reqdone;
251 	kauth_cred_free(cred);
252 	return (error);
253 }
254 
255 #ifndef NFS_V2_ONLY
256 /*
257  * nfs version 3 fsinfo rpc call
258  */
259 int
260 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, kauth_cred_t cred, struct lwp *l)
261 {
262 	struct nfsv3_fsinfo *fsp;
263 	char *cp;
264 	int32_t t1, t2;
265 	u_int32_t *tl, pref, xmax;
266 	char *bpos, *dpos, *cp2;
267 	int error = 0, retattr;
268 	struct mbuf *mreq, *mrep, *md, *mb;
269 	u_int64_t maxfsize;
270 	struct nfsnode *np = VTONFS(vp);
271 
272 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
273 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
274 	nfsm_fhtom(np, 1);
275 	nfsm_request(np, NFSPROC_FSINFO, l, cred);
276 	nfsm_postop_attr(vp, retattr, 0);
277 	if (!error) {
278 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
279 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
280 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
281 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
282 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
283 				~(NFS_FABLKSIZE - 1);
284 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
285 		if (xmax < nmp->nm_wsize && xmax > 0) {
286 			nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1);
287 			if (nmp->nm_wsize == 0)
288 				nmp->nm_wsize = xmax;
289 		}
290 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
291 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
292 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
293 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
294 				~(NFS_FABLKSIZE - 1);
295 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
296 		if (xmax < nmp->nm_rsize && xmax > 0) {
297 			nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1);
298 			if (nmp->nm_rsize == 0)
299 				nmp->nm_rsize = xmax;
300 		}
301 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
302 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
303 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
304 				~(NFS_DIRFRAGSIZ - 1);
305 		if (xmax < nmp->nm_readdirsize && xmax > 0) {
306 			nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1);
307 			if (nmp->nm_readdirsize == 0)
308 				nmp->nm_readdirsize = xmax;
309 		}
310 		/* XXX */
311 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
312 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
313 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
314 			nmp->nm_maxfilesize = maxfsize;
315 		nmp->nm_mountp->mnt_fs_bshift =
316 		    ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
317 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
318 	}
319 	nfsm_reqdone;
320 	return (error);
321 }
322 #endif
323 
324 /*
325  * Mount a remote root fs via. NFS.  It goes like this:
326  * - Call nfs_boot_init() to fill in the nfs_diskless struct
327  * - build the rootfs mount point and call mountnfs() to do the rest.
328  */
329 int
330 nfs_mountroot(void)
331 {
332 	struct timespec ts;
333 	struct nfs_diskless *nd;
334 	struct vattr attr;
335 	struct mount *mp;
336 	struct vnode *vp;
337 	struct lwp *l;
338 	long n;
339 	int error;
340 
341 	l = curlwp; /* XXX */
342 
343 	if (device_class(root_device) != DV_IFNET)
344 		return (ENODEV);
345 
346 	/*
347 	 * XXX time must be non-zero when we init the interface or else
348 	 * the arp code will wedge.  [Fixed now in if_ether.c]
349 	 * However, the NFS attribute cache gives false "hits" when the
350 	 * current time < nfs_attrtimeo(nmp, np) so keep this in for now.
351 	 */
352 	if (time_second < NFS_MAXATTRTIMO) {
353 		ts.tv_sec = NFS_MAXATTRTIMO;
354 		ts.tv_nsec = 0;
355 		tc_setclock(&ts);
356 	}
357 
358 	/*
359 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
360 	 * Side effect:  Finds and configures a network interface.
361 	 */
362 	nd = kmem_zalloc(sizeof(*nd), KM_SLEEP);
363 	error = nfs_boot_init(nd, l);
364 	if (error) {
365 		kmem_free(nd, sizeof(*nd));
366 		return (error);
367 	}
368 
369 	/*
370 	 * Create the root mount point.
371 	 */
372 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l);
373 	if (error)
374 		goto out;
375 	printf("root on %s\n", nd->nd_root.ndm_host);
376 
377 	/*
378 	 * Link it into the mount list.
379 	 */
380 	mutex_enter(&mountlist_lock);
381 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
382 	mutex_exit(&mountlist_lock);
383 	rootvp = vp;
384 	mp->mnt_vnodecovered = NULLVP;
385 	vfs_unbusy(mp, false, NULL);
386 
387 	/* Get root attributes (for the time). */
388 	error = VOP_GETATTR(vp, &attr, l->l_cred);
389 	if (error)
390 		panic("nfs_mountroot: getattr for root");
391 	n = attr.va_atime.tv_sec;
392 #ifdef	DEBUG
393 	printf("root time: 0x%lx\n", n);
394 #endif
395 	setrootfstime(n);
396 
397 out:
398 	if (error)
399 		nfs_boot_cleanup(nd, l);
400 	kmem_free(nd, sizeof(*nd));
401 	return (error);
402 }
403 
404 /*
405  * Internal version of mount system call for diskless setup.
406  * Separate function because we used to call it twice.
407  * (once for root and once for swap)
408  */
409 static int
410 nfs_mount_diskless(struct nfs_dlmount *ndmntp, const char *mntname, struct mount **mpp, struct vnode **vpp, struct lwp *l)
411 	/* mntname:	 mount point name */
412 {
413 	struct mount *mp;
414 	struct mbuf *m;
415 	int error;
416 
417 	vfs_rootmountalloc(MOUNT_NFS, mntname, &mp);
418 
419 	mp->mnt_op = &nfs_vfsops;
420 
421 	/*
422 	 * Historical practice expects NFS root file systems to
423 	 * be initially mounted r/w.
424 	 */
425 	mp->mnt_flag &= ~MNT_RDONLY;
426 
427 	/* Get mbuf for server sockaddr. */
428 	m = m_get(M_WAIT, MT_SONAME);
429 	if (m == NULL)
430 		panic("nfs_mountroot: mget soname for %s", mntname);
431 	MCLAIM(m, &nfs_mowner);
432 	memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr,
433 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
434 
435 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
436 			 ndmntp->ndm_args.hostname, vpp, l);
437 	if (error) {
438 		vfs_unbusy(mp, false, NULL);
439 		vfs_destroy(mp);
440 		printf("nfs_mountroot: mount %s failed: %d\n",
441 		       mntname, error);
442 	} else
443 		*mpp = mp;
444 
445 	return (error);
446 }
447 
448 void
449 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp, struct lwp *l)
450 {
451 	int s;
452 	int adjsock;
453 	int maxio;
454 
455 	s = splsoftnet();
456 
457 	/*
458 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
459 	 * no sense in that context.
460 	 */
461 	if (argp->sotype == SOCK_STREAM)
462 		argp->flags &= ~NFSMNT_NOCONN;
463 
464 	/*
465 	 * Cookie translation is not needed for v2, silently ignore it.
466 	 */
467 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
468 	    NFSMNT_XLATECOOKIE)
469 		argp->flags &= ~NFSMNT_XLATECOOKIE;
470 
471 	/* Re-bind if rsrvd port requested and wasn't on one */
472 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
473 		  && (argp->flags & NFSMNT_RESVPORT);
474 	/* Also re-bind if we're switching to/from a connected UDP socket */
475 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
476 		    (argp->flags & NFSMNT_NOCONN));
477 
478 	/* Update flags. */
479 	nmp->nm_flag = argp->flags;
480 	splx(s);
481 
482 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
483 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
484 		if (nmp->nm_timeo < NFS_MINTIMEO)
485 			nmp->nm_timeo = NFS_MINTIMEO;
486 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
487 			nmp->nm_timeo = NFS_MAXTIMEO;
488 	}
489 
490 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
491 		nmp->nm_retry = argp->retrans;
492 		if (nmp->nm_retry > NFS_MAXREXMIT)
493 			nmp->nm_retry = NFS_MAXREXMIT;
494 	}
495 
496 #ifndef NFS_V2_ONLY
497 	if (argp->flags & NFSMNT_NFSV3) {
498 		if (argp->sotype == SOCK_DGRAM)
499 			maxio = NFS_MAXDGRAMDATA;
500 		else
501 			maxio = NFS_MAXDATA;
502 	} else
503 #endif
504 		maxio = NFS_V2MAXDATA;
505 
506 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
507 		int osize = nmp->nm_wsize;
508 		nmp->nm_wsize = argp->wsize;
509 		/* Round down to multiple of blocksize */
510 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
511 		if (nmp->nm_wsize <= 0)
512 			nmp->nm_wsize = NFS_FABLKSIZE;
513 		adjsock |= (nmp->nm_wsize != osize);
514 	}
515 	if (nmp->nm_wsize > maxio)
516 		nmp->nm_wsize = maxio;
517 	if (nmp->nm_wsize > MAXBSIZE)
518 		nmp->nm_wsize = MAXBSIZE;
519 
520 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
521 		int osize = nmp->nm_rsize;
522 		nmp->nm_rsize = argp->rsize;
523 		/* Round down to multiple of blocksize */
524 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
525 		if (nmp->nm_rsize <= 0)
526 			nmp->nm_rsize = NFS_FABLKSIZE;
527 		adjsock |= (nmp->nm_rsize != osize);
528 	}
529 	if (nmp->nm_rsize > maxio)
530 		nmp->nm_rsize = maxio;
531 	if (nmp->nm_rsize > MAXBSIZE)
532 		nmp->nm_rsize = MAXBSIZE;
533 
534 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
535 		nmp->nm_readdirsize = argp->readdirsize;
536 		/* Round down to multiple of minimum blocksize */
537 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
538 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
539 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
540 		/* Bigger than buffer size makes no sense */
541 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
542 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
543 	} else if (argp->flags & NFSMNT_RSIZE)
544 		nmp->nm_readdirsize = nmp->nm_rsize;
545 
546 	if (nmp->nm_readdirsize > maxio)
547 		nmp->nm_readdirsize = maxio;
548 
549 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
550 		argp->maxgrouplist <= NFS_MAXGRPS)
551 		nmp->nm_numgrps = argp->maxgrouplist;
552 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
553 		argp->readahead <= NFS_MAXRAHEAD)
554 		nmp->nm_readahead = argp->readahead;
555 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
556 		argp->deadthresh <= NFS_NEVERDEAD)
557 		nmp->nm_deadthresh = argp->deadthresh;
558 
559 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
560 		    (nmp->nm_soproto != argp->proto));
561 	nmp->nm_sotype = argp->sotype;
562 	nmp->nm_soproto = argp->proto;
563 
564 	if (nmp->nm_so && adjsock) {
565 		nfs_safedisconnect(nmp);
566 		if (nmp->nm_sotype == SOCK_DGRAM)
567 			while (nfs_connect(nmp, (struct nfsreq *)0, l)) {
568 				printf("nfs_args: retrying connect\n");
569 				kpause("nfscn3", false, hz, NULL);
570 			}
571 	}
572 }
573 
574 /*
575  * VFS Operations.
576  *
577  * mount system call
578  * It seems a bit dumb to copyinstr() the host and path here and then
579  * memcpy() them in mountnfs(), but I wanted to detect errors before
580  * doing the sockargs() call because sockargs() allocates an mbuf and
581  * an error after that means that I have to release the mbuf.
582  */
583 /* ARGSUSED */
584 int
585 nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
586 {
587 	struct lwp *l = curlwp;
588 	int error;
589 	struct nfs_args *args = data;
590 	struct mbuf *nam;
591 	struct nfsmount *nmp = VFSTONFS(mp);
592 	struct sockaddr *sa;
593 	struct vnode *vp;
594 	char *pth, *hst;
595 	struct proc *p;
596 	size_t len;
597 	u_char *nfh;
598 
599 	if (*data_len < sizeof *args)
600 		return EINVAL;
601 
602 	p = l->l_proc;
603 	if (mp->mnt_flag & MNT_GETARGS) {
604 
605 		if (nmp == NULL)
606 			return (EIO);
607 		if (args->addr != NULL) {
608 			sa = mtod(nmp->nm_nam, struct sockaddr *);
609 			error = copyout(sa, args->addr, sa->sa_len);
610 			if (error)
611 				return (error);
612 			args->addrlen = sa->sa_len;
613 		} else
614 			args->addrlen = 0;
615 
616 		args->version = NFS_ARGSVERSION;
617 		args->sotype = nmp->nm_sotype;
618 		args->proto = nmp->nm_soproto;
619 		args->fh = NULL;
620 		args->fhsize = 0;
621 		args->flags = nmp->nm_flag;
622 		args->wsize = nmp->nm_wsize;
623 		args->rsize = nmp->nm_rsize;
624 		args->readdirsize = nmp->nm_readdirsize;
625 		args->timeo = nmp->nm_timeo;
626 		args->retrans = nmp->nm_retry;
627 		args->maxgrouplist = nmp->nm_numgrps;
628 		args->readahead = nmp->nm_readahead;
629 		args->leaseterm = 0; /* dummy */
630 		args->deadthresh = nmp->nm_deadthresh;
631 		args->hostname = NULL;
632 		*data_len = sizeof *args;
633 		return 0;
634 	}
635 
636 	if (args->version != NFS_ARGSVERSION)
637 		return (EPROGMISMATCH);
638 	if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB))
639 		return (EPROGUNAVAIL);
640 #ifdef NFS_V2_ONLY
641 	if (args->flags & NFSMNT_NFSV3)
642 		return (EPROGMISMATCH);
643 #endif
644 	if (mp->mnt_flag & MNT_UPDATE) {
645 		if (nmp == NULL)
646 			return (EIO);
647 		/*
648 		 * When doing an update, we can't change from or to
649 		 * v3, or change cookie translation
650 		 */
651 		args->flags = (args->flags & ~(NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)) |
652 		    (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_XLATECOOKIE));
653 		nfs_decode_args(nmp, args, l);
654 		return (0);
655 	}
656 	if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX)
657 		return (EINVAL);
658 	nfh = malloc(NFSX_V3FHMAX, M_TEMP, M_WAITOK);
659 	error = copyin(args->fh, nfh, args->fhsize);
660 	if (error)
661 		goto free_nfh;
662 	pth = malloc(MNAMELEN, M_TEMP, M_WAITOK);
663 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
664 	if (error)
665 		goto free_pth;
666 	memset(&pth[len], 0, MNAMELEN - len);
667 	hst = malloc(MNAMELEN, M_TEMP, M_WAITOK);
668 	error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len);
669 	if (error)
670 		goto free_hst;
671 	memset(&hst[len], 0, MNAMELEN - len);
672 	/* sockargs() call must be after above copyin() calls */
673 	error = sockargs(&nam, args->addr, args->addrlen, MT_SONAME);
674 	if (error)
675 		goto free_hst;
676 	MCLAIM(nam, &nfs_mowner);
677 	args->fh = nfh;
678 	error = mountnfs(args, mp, nam, pth, hst, &vp, l);
679 
680 free_hst:
681 	free(hst, M_TEMP);
682 free_pth:
683 	free(pth, M_TEMP);
684 free_nfh:
685 	free(nfh, M_TEMP);
686 
687 	return (error);
688 }
689 
690 /*
691  * Common code for mount and mountroot
692  */
693 int
694 mountnfs(struct nfs_args *argp, struct mount *mp, struct mbuf *nam, const char *pth, const char *hst, struct vnode **vpp, struct lwp *l)
695 {
696 	struct nfsmount *nmp;
697 	struct nfsnode *np;
698 	struct vnode *vp;
699 	int error;
700 	struct vattr *attrs;
701 	kauth_cred_t cr;
702 	char iosname[IOSTATNAMELEN];
703 
704 	/*
705 	 * If the number of nfs iothreads to use has never
706 	 * been set, create a reasonable number of them.
707 	 */
708 
709 	if (nfs_niothreads < 0) {
710 		nfs_set_niothreads(NFS_DEFAULT_NIOTHREADS);
711 	}
712 
713 	if (mp->mnt_flag & MNT_UPDATE) {
714 		nmp = VFSTONFS(mp);
715 		/* update paths, file handles, etc, here	XXX */
716 		m_freem(nam);
717 		return (0);
718 	} else {
719 		nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP);
720 		mp->mnt_data = nmp;
721 		TAILQ_INIT(&nmp->nm_uidlruhead);
722 		TAILQ_INIT(&nmp->nm_bufq);
723 		rw_init(&nmp->nm_writeverflock);
724 		mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE);
725 		rw_init(&nmp->nm_rbtlock);
726 		cv_init(&nmp->nm_rcvcv, "nfsrcv");
727 		cv_init(&nmp->nm_sndcv, "nfssnd");
728 		cv_init(&nmp->nm_aiocv, "nfsaio");
729 		cv_init(&nmp->nm_disconcv, "nfsdis");
730 		nfs_rbtinit(nmp);
731 	}
732 	vfs_getnewfsid(mp);
733 	nmp->nm_mountp = mp;
734 
735 #ifndef NFS_V2_ONLY
736 	if ((argp->flags & NFSMNT_NFSV3) == 0)
737 #endif
738 	{
739 		if (argp->fhsize != NFSX_V2FH) {
740 			return EINVAL;
741 		}
742 	}
743 
744 	/*
745 	 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
746 	 * will overwrite this.
747 	 */
748 	nmp->nm_maxfilesize = 0xffffffffLL;
749 
750 	nmp->nm_timeo = NFS_TIMEO;
751 	nmp->nm_retry = NFS_RETRANS;
752 	nmp->nm_wsize = NFS_WSIZE;
753 	nmp->nm_rsize = NFS_RSIZE;
754 	nmp->nm_readdirsize = NFS_READDIRSIZE;
755 	nmp->nm_numgrps = NFS_MAXGRPS;
756 	nmp->nm_readahead = NFS_DEFRAHEAD;
757 	nmp->nm_deadthresh = NFS_DEFDEADTHRESH;
758 	error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE,
759 	    mp->mnt_op->vfs_name, mp, l);
760 	if (error)
761 		goto bad;
762 	nmp->nm_nam = nam;
763 
764 	/* Set up the sockets and per-host congestion */
765 	nmp->nm_sotype = argp->sotype;
766 	nmp->nm_soproto = argp->proto;
767 
768 	nfs_decode_args(nmp, argp, l);
769 
770 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
771 	mp->mnt_dev_bshift = DEV_BSHIFT;
772 
773 	/*
774 	 * For Connection based sockets (TCP,...) defer the connect until
775 	 * the first request, in case the server is not responding.
776 	 */
777 	if (nmp->nm_sotype == SOCK_DGRAM &&
778 		(error = nfs_connect(nmp, (struct nfsreq *)0, l)))
779 		goto bad;
780 
781 	/*
782 	 * This is silly, but it has to be set so that vinifod() works.
783 	 * We do not want to do an nfs_statvfs() here since we can get
784 	 * stuck on a dead server and we are holding a lock on the mount
785 	 * point.
786 	 */
787 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
788 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
789 	if (error)
790 		goto bad;
791 	vp = NFSTOV(np);
792 	attrs = malloc(sizeof(struct vattr), M_TEMP, M_WAITOK);
793 	VOP_GETATTR(vp, attrs, l->l_cred);
794 	if ((nmp->nm_flag & NFSMNT_NFSV3) && (vp->v_type == VDIR)) {
795 		cr = kauth_cred_alloc();
796 		kauth_cred_setuid(cr, attrs->va_uid);
797 		kauth_cred_seteuid(cr, attrs->va_uid);
798 		kauth_cred_setsvuid(cr, attrs->va_uid);
799 		kauth_cred_setgid(cr, attrs->va_gid);
800 		kauth_cred_setegid(cr, attrs->va_gid);
801 		kauth_cred_setsvgid(cr, attrs->va_gid);
802 		nfs_cookieheuristic(vp, &nmp->nm_iflag, l, cr);
803 		kauth_cred_free(cr);
804 	}
805 	free(attrs, M_TEMP);
806 
807 	/*
808 	 * A reference count is needed on the nfsnode representing the
809 	 * remote root.  If this object is not persistent, then backward
810 	 * traversals of the mount point (i.e. "..") will not work if
811 	 * the nfsnode gets flushed out of the cache. Ufs does not have
812 	 * this problem, because one can identify root inodes by their
813 	 * number == ROOTINO (2). So, just unlock, but no rele.
814 	 */
815 
816 	nmp->nm_vnode = vp;
817 	if (vp->v_type == VNON)
818 		vp->v_type = VDIR;
819 	vp->v_vflag |= VV_ROOT;
820 	VOP_UNLOCK(vp, 0);
821 	*vpp = vp;
822 
823 	snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++);
824 	nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname);
825 
826 	return (0);
827 bad:
828 	nfs_disconnect(nmp);
829 	rw_destroy(&nmp->nm_writeverflock);
830 	rw_destroy(&nmp->nm_rbtlock);
831 	mutex_destroy(&nmp->nm_lock);
832 	cv_destroy(&nmp->nm_rcvcv);
833 	cv_destroy(&nmp->nm_sndcv);
834 	cv_destroy(&nmp->nm_aiocv);
835 	cv_destroy(&nmp->nm_disconcv);
836 	kmem_free(nmp, sizeof(*nmp));
837 	m_freem(nam);
838 	return (error);
839 }
840 
841 /*
842  * unmount system call
843  */
844 int
845 nfs_unmount(struct mount *mp, int mntflags)
846 {
847 	struct nfsmount *nmp;
848 	struct vnode *vp;
849 	int error, flags = 0;
850 
851 	if (mntflags & MNT_FORCE)
852 		flags |= FORCECLOSE;
853 	nmp = VFSTONFS(mp);
854 	/*
855 	 * Goes something like this..
856 	 * - Check for activity on the root vnode (other than ourselves).
857 	 * - Call vflush() to clear out vnodes for this file system,
858 	 *   except for the root vnode.
859 	 * - Decrement reference on the vnode representing remote root.
860 	 * - Close the socket
861 	 * - Free up the data structures
862 	 */
863 	/*
864 	 * We need to decrement the ref. count on the nfsnode representing
865 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
866 	 * has done vput on this vnode, otherwise we would get deadlock!
867 	 */
868 	vp = nmp->nm_vnode;
869 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
870 	if (error != 0)
871 		return error;
872 
873 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
874 		vput(vp);
875 		return (EBUSY);
876 	}
877 
878 	error = vflush(mp, vp, flags);
879 	if (error) {
880 		vput(vp);
881 		return (error);
882 	}
883 
884 	/*
885 	 * We are now committed to the unmount; mark the mount structure
886 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
887 	 * will go away cleanly.
888 	 */
889 	nmp->nm_iflag |= NFSMNT_DISMNT;
890 
891 	/*
892 	 * Clean up the stats... note that we carefully avoid decrementing
893 	 * nfs_mount_count here for good reason - we may not be unmounting
894 	 * the last thing mounted.
895 	 */
896 	iostat_free(nmp->nm_stats);
897 
898 	/*
899 	 * There are two reference counts to get rid of here
900 	 * (see comment in mountnfs()).
901 	 */
902 	vput(vp);
903 	vgone(vp);
904 	nfs_disconnect(nmp);
905 	m_freem(nmp->nm_nam);
906 
907 	rw_destroy(&nmp->nm_writeverflock);
908 	rw_destroy(&nmp->nm_rbtlock);
909 	mutex_destroy(&nmp->nm_lock);
910 	cv_destroy(&nmp->nm_rcvcv);
911 	cv_destroy(&nmp->nm_sndcv);
912 	cv_destroy(&nmp->nm_aiocv);
913 	cv_destroy(&nmp->nm_disconcv);
914 	kmem_free(nmp, sizeof(*nmp));
915 	return (0);
916 }
917 
918 /*
919  * Return root of a filesystem
920  */
921 int
922 nfs_root(struct mount *mp, struct vnode **vpp)
923 {
924 	struct vnode *vp;
925 	struct nfsmount *nmp;
926 	int error;
927 
928 	nmp = VFSTONFS(mp);
929 	vp = nmp->nm_vnode;
930 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
931 	if (error != 0)
932 		return error;
933 	*vpp = vp;
934 	return (0);
935 }
936 
937 extern int syncprt;
938 
939 /*
940  * Flush out the buffer cache
941  */
942 /* ARGSUSED */
943 int
944 nfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
945 {
946 	struct vnode *vp, *mvp;
947 	int error, allerror = 0;
948 
949 	/*
950 	 * Force stale buffer cache information to be flushed.
951 	 */
952 	if ((mvp = vnalloc(mp)) == NULL)
953 		return (ENOMEM);
954 loop:
955 	/*
956 	 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
957 	 * and vclean() can be called indirectly
958 	 */
959 	mutex_enter(&mntvnode_lock);
960 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) {
961 		vmark(mvp, vp);
962 		if (vp->v_mount != mp || vismarker(vp))
963 			continue;
964 		mutex_enter(&vp->v_interlock);
965 		/* XXX MNT_LAZY cannot be right? */
966 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
967 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
968 		     UVM_OBJ_IS_CLEAN(&vp->v_uobj))) {
969 			mutex_exit(&vp->v_interlock);
970 			continue;
971 		}
972 		mutex_exit(&mntvnode_lock);
973 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
974 			(void)vunmark(mvp);
975 			goto loop;
976 		}
977 		error = VOP_FSYNC(vp, cred,
978 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0);
979 		if (error)
980 			allerror = error;
981 		vput(vp);
982 		mutex_enter(&mntvnode_lock);
983 	}
984 	mutex_exit(&mntvnode_lock);
985 	vnfree(mvp);
986 	return (allerror);
987 }
988 
989 /*
990  * NFS flat namespace lookup.
991  * Currently unsupported.
992  */
993 /* ARGSUSED */
994 int
995 nfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
996 {
997 
998 	return (EOPNOTSUPP);
999 }
1000 
1001 /*
1002  * Do that sysctl thang...
1003  */
1004 static int
1005 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS)
1006 {
1007 	struct sysctlnode node;
1008 	int val;
1009 	int error;
1010 
1011 	val = nfs_niothreads;
1012 	node = *rnode;
1013 	node.sysctl_data = &val;
1014         error = sysctl_lookup(SYSCTLFN_CALL(&node));
1015 	if (error || newp == NULL)
1016 		return error;
1017 
1018 	return nfs_set_niothreads(val);
1019 }
1020 
1021 static void
1022 nfs_sysctl_init(void)
1023 {
1024 
1025 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1026 		       CTLFLAG_PERMANENT,
1027 		       CTLTYPE_NODE, "vfs", NULL,
1028 		       NULL, 0, NULL, 0,
1029 		       CTL_VFS, CTL_EOL);
1030 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1031 		       CTLFLAG_PERMANENT,
1032 		       CTLTYPE_NODE, "nfs",
1033 		       SYSCTL_DESCR("NFS vfs options"),
1034 		       NULL, 0, NULL, 0,
1035 		       CTL_VFS, 2, CTL_EOL);
1036 	/*
1037 	 * XXX the "2" above could be dynamic, thereby eliminating one
1038 	 * more instance of the "number to vfs" mapping problem, but
1039 	 * "2" is the order as taken from sys/mount.h
1040 	 */
1041 
1042 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1043 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1044 		       CTLTYPE_STRUCT, "nfsstats",
1045 		       SYSCTL_DESCR("NFS operation statistics"),
1046 		       NULL, 0, &nfsstats, sizeof(nfsstats),
1047 		       CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL);
1048 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1049 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1050 		       CTLTYPE_INT, "iothreads",
1051 		       SYSCTL_DESCR("Number of NFS client processes desired"),
1052 		       sysctl_vfs_nfs_iothreads, 0, NULL, 0,
1053 		       CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL);
1054 }
1055 
1056 static void
1057 nfs_sysctl_fini(void)
1058 {
1059 
1060 	sysctl_teardown(&nfs_clog);
1061 }
1062 
1063 /* ARGSUSED */
1064 int
1065 nfs_fhtovp(struct mount *mp, struct fid *fid, struct vnode **vpp)
1066 {
1067 	size_t fidsize;
1068 	size_t fhsize;
1069 	struct nfsnode *np;
1070 	int error;
1071 	struct vattr va;
1072 
1073 	fidsize = fid->fid_len;
1074 	if (fidsize < sizeof(*fid)) {
1075 		return EINVAL;
1076 	}
1077 	fhsize = fidsize - sizeof(*fid);
1078 	if ((fhsize % NFSX_UNSIGNED) != 0) {
1079 		return EINVAL;
1080 	}
1081 	if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) {
1082 		if (fhsize > NFSX_V3FHMAX || fhsize == 0) {
1083 			return EINVAL;
1084 		}
1085 	} else {
1086 		if (fhsize != NFSX_V2FH) {
1087 			return EINVAL;
1088 		}
1089 	}
1090 	error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np);
1091 	if (error) {
1092 		return error;
1093 	}
1094 	*vpp = NFSTOV(np);
1095 	error = VOP_GETATTR(*vpp, &va, kauth_cred_get());
1096 	if (error != 0) {
1097 		vput(*vpp);
1098 	}
1099 	return error;
1100 }
1101 
1102 /* ARGSUSED */
1103 int
1104 nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize)
1105 {
1106 	struct nfsnode *np;
1107 	struct fid *fid;
1108 	size_t fidsize;
1109 	int error = 0;
1110 
1111 	np = VTONFS(vp);
1112 	fidsize = sizeof(*fid) + np->n_fhsize;
1113 	if (*bufsize < fidsize) {
1114 		error = E2BIG;
1115 	}
1116 	*bufsize = fidsize;
1117 	if (error == 0) {
1118 		struct fid fid_store;
1119 
1120 		fid = &fid_store;
1121 		memset(fid, 0, sizeof(*fid));
1122 		fid->fid_len = fidsize;
1123 		memcpy(buf, fid, sizeof(*fid));
1124 		memcpy(buf->fid_data, np->n_fhp, np->n_fhsize);
1125 	}
1126 	return error;
1127 }
1128 
1129 /*
1130  * Vfs start routine, a no-op.
1131  */
1132 /* ARGSUSED */
1133 int
1134 nfs_start(struct mount *mp, int flags)
1135 {
1136 
1137 	return (0);
1138 }
1139