xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision ba65fde2d7fefa7d39838fa5fa855e62bd606b5e)
1 /*	$NetBSD: nfs_vfsops.c,v 1.221 2013/01/22 09:39:14 dholland 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 1.221 2013/01/22 09:39:14 dholland 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 int nfs_commitsize;
90 
91 /*
92  * nfs vfs operations.
93  */
94 
95 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
96 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
97 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
98 
99 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = {
100 	&nfsv2_vnodeop_opv_desc,
101 	&spec_nfsv2nodeop_opv_desc,
102 	&fifo_nfsv2nodeop_opv_desc,
103 	NULL,
104 };
105 
106 struct vfsops nfs_vfsops = {
107 	MOUNT_NFS,
108 	sizeof (struct nfs_args),
109 	nfs_mount,
110 	nfs_start,
111 	nfs_unmount,
112 	nfs_root,
113 	(void *)eopnotsupp,	/* vfs_quotactl */
114 	nfs_statvfs,
115 	nfs_sync,
116 	nfs_vget,
117 	nfs_fhtovp,
118 	nfs_vptofh,
119 	nfs_vfs_init,
120 	NULL,
121 	nfs_vfs_done,
122 	nfs_mountroot,
123 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
124 	vfs_stdextattrctl,
125 	(void *)eopnotsupp,	/* vfs_suspendctl */
126 	genfs_renamelock_enter,
127 	genfs_renamelock_exit,
128 	(void *)eopnotsupp,
129 	nfs_vnodeopv_descs,
130 	0,
131 	{ NULL, NULL },
132 };
133 
134 extern u_int32_t nfs_procids[NFS_NPROCS];
135 extern u_int32_t nfs_prog, nfs_vers;
136 static struct sysctllog *nfs_clog;
137 
138 static int nfs_mount_diskless(struct nfs_dlmount *, const char *,
139     struct mount **, struct vnode **, struct lwp *);
140 static void nfs_sysctl_init(void);
141 static void nfs_sysctl_fini(void);
142 
143 static int
144 nfs_modcmd(modcmd_t cmd, void *arg)
145 {
146 	int error;
147 
148 	switch (cmd) {
149 	case MODULE_CMD_INIT:
150 		error = vfs_attach(&nfs_vfsops);
151 		if (error == 0) {
152 			nfs_sysctl_init();
153 		}
154 		return error;
155 	case MODULE_CMD_FINI:
156 		error = vfs_detach(&nfs_vfsops);
157 		if (error == 0) {
158 			nfs_sysctl_fini();
159 		}
160 		return error;
161 	default:
162 		return ENOTTY;
163 	}
164 }
165 
166 /*
167  * nfs statvfs call
168  */
169 int
170 nfs_statvfs(struct mount *mp, struct statvfs *sbp)
171 {
172 	struct lwp *l = curlwp;
173 	struct vnode *vp;
174 	struct nfs_statfs *sfp;
175 	char *cp;
176 	u_int32_t *tl;
177 	int32_t t1, t2;
178 	char *bpos, *dpos, *cp2;
179 	struct nfsmount *nmp = VFSTONFS(mp);
180 	int error = 0, retattr;
181 #ifdef NFS_V2_ONLY
182 	const int v3 = 0;
183 #else
184 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
185 #endif
186 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
187 	kauth_cred_t cred;
188 	u_quad_t tquad;
189 	struct nfsnode *np;
190 
191 #ifndef nolint
192 	sfp = (struct nfs_statfs *)0;
193 #endif
194 	vp = nmp->nm_vnode;
195 	np = VTONFS(vp);
196 	cred = kauth_cred_alloc();
197 #ifndef NFS_V2_ONLY
198 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
199 		(void)nfs_fsinfo(nmp, vp, cred, l);
200 #endif
201 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
202 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
203 	nfsm_fhtom(np, v3);
204 	nfsm_request(np, NFSPROC_FSSTAT, l, cred);
205 	if (v3)
206 		nfsm_postop_attr(vp, retattr, 0);
207 	if (error) {
208 		if (mrep != NULL) {
209 			if (mrep->m_next != NULL)
210 				printf("nfs_vfsops: nfs_statvfs would lose buffers\n");
211 			m_freem(mrep);
212 		}
213 		goto nfsmout;
214 	}
215 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
216 	sbp->f_flag = nmp->nm_flag;
217 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
218 	if (v3) {
219 		sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE;
220 		tquad = fxdr_hyper(&sfp->sf_tbytes);
221 		sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
222 		tquad = fxdr_hyper(&sfp->sf_fbytes);
223 		sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
224 		tquad = fxdr_hyper(&sfp->sf_abytes);
225 		tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
226 		sbp->f_bresvd = sbp->f_bfree - tquad;
227 		sbp->f_bavail = tquad;
228 		/* Handle older NFS servers returning negative values */
229 		if ((quad_t)sbp->f_bavail < 0)
230 			sbp->f_bavail = 0;
231 		tquad = fxdr_hyper(&sfp->sf_tfiles);
232 		sbp->f_files = tquad;
233 		tquad = fxdr_hyper(&sfp->sf_ffiles);
234 		sbp->f_ffree = tquad;
235 		sbp->f_favail = tquad;
236 		sbp->f_fresvd = 0;
237 		sbp->f_namemax = NFS_MAXNAMLEN;
238 	} else {
239 		sbp->f_bsize = NFS_FABLKSIZE;
240 		sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
241 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
242 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
243 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
244 		sbp->f_fresvd = 0;
245 		sbp->f_files = 0;
246 		sbp->f_ffree = 0;
247 		sbp->f_favail = 0;
248 		sbp->f_fresvd = 0;
249 		sbp->f_namemax = NFS_MAXNAMLEN;
250 	}
251 	copy_statvfs_info(sbp, mp);
252 	nfsm_reqdone;
253 	kauth_cred_free(cred);
254 	return (error);
255 }
256 
257 #ifndef NFS_V2_ONLY
258 /*
259  * nfs version 3 fsinfo rpc call
260  */
261 int
262 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, kauth_cred_t cred, struct lwp *l)
263 {
264 	struct nfsv3_fsinfo *fsp;
265 	char *cp;
266 	int32_t t1, t2;
267 	u_int32_t *tl, pref, xmax;
268 	char *bpos, *dpos, *cp2;
269 	int error = 0, retattr;
270 	struct mbuf *mreq, *mrep, *md, *mb;
271 	u_int64_t maxfsize;
272 	struct nfsnode *np = VTONFS(vp);
273 
274 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
275 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
276 	nfsm_fhtom(np, 1);
277 	nfsm_request(np, NFSPROC_FSINFO, l, cred);
278 	nfsm_postop_attr(vp, retattr, 0);
279 	if (!error) {
280 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
281 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
282 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
283 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
284 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
285 				~(NFS_FABLKSIZE - 1);
286 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
287 		if (xmax < nmp->nm_wsize && xmax > 0) {
288 			nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1);
289 			if (nmp->nm_wsize == 0)
290 				nmp->nm_wsize = xmax;
291 		}
292 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
293 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
294 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
295 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
296 				~(NFS_FABLKSIZE - 1);
297 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
298 		if (xmax < nmp->nm_rsize && xmax > 0) {
299 			nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1);
300 			if (nmp->nm_rsize == 0)
301 				nmp->nm_rsize = xmax;
302 		}
303 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
304 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
305 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
306 				~(NFS_DIRFRAGSIZ - 1);
307 		if (xmax < nmp->nm_readdirsize && xmax > 0) {
308 			nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1);
309 			if (nmp->nm_readdirsize == 0)
310 				nmp->nm_readdirsize = xmax;
311 		}
312 		/* XXX */
313 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
314 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
315 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
316 			nmp->nm_maxfilesize = maxfsize;
317 		nmp->nm_mountp->mnt_fs_bshift =
318 		    ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
319 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
320 	}
321 	nfsm_reqdone;
322 	return (error);
323 }
324 #endif
325 
326 /*
327  * Mount a remote root fs via. NFS.  It goes like this:
328  * - Call nfs_boot_init() to fill in the nfs_diskless struct
329  * - build the rootfs mount point and call mountnfs() to do the rest.
330  */
331 int
332 nfs_mountroot(void)
333 {
334 	struct timespec ts;
335 	struct nfs_diskless *nd;
336 	struct vattr attr;
337 	struct mount *mp;
338 	struct vnode *vp;
339 	struct lwp *l;
340 	long n;
341 	int error;
342 
343 	l = curlwp; /* XXX */
344 
345 	if (device_class(root_device) != DV_IFNET)
346 		return (ENODEV);
347 
348 	/*
349 	 * XXX time must be non-zero when we init the interface or else
350 	 * the arp code will wedge.  [Fixed now in if_ether.c]
351 	 * However, the NFS attribute cache gives false "hits" when the
352 	 * current time < nfs_attrtimeo(nmp, np) so keep this in for now.
353 	 */
354 	if (time_second < NFS_MAXATTRTIMO) {
355 		ts.tv_sec = NFS_MAXATTRTIMO;
356 		ts.tv_nsec = 0;
357 		tc_setclock(&ts);
358 	}
359 
360 	/*
361 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
362 	 * Side effect:  Finds and configures a network interface.
363 	 */
364 	nd = kmem_zalloc(sizeof(*nd), KM_SLEEP);
365 	error = nfs_boot_init(nd, l);
366 	if (error) {
367 		kmem_free(nd, sizeof(*nd));
368 		return (error);
369 	}
370 
371 	/*
372 	 * Create the root mount point.
373 	 */
374 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l);
375 	if (error)
376 		goto out;
377 	printf("root on %s\n", nd->nd_root.ndm_host);
378 
379 	/*
380 	 * Link it into the mount list.
381 	 */
382 	mutex_enter(&mountlist_lock);
383 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
384 	mutex_exit(&mountlist_lock);
385 	rootvp = vp;
386 	mp->mnt_vnodecovered = NULLVP;
387 	vfs_unbusy(mp, false, NULL);
388 
389 	/* Get root attributes (for the time). */
390 	vn_lock(vp, LK_SHARED | LK_RETRY);
391 	error = VOP_GETATTR(vp, &attr, l->l_cred);
392 	VOP_UNLOCK(vp);
393 	if (error)
394 		panic("nfs_mountroot: getattr for root");
395 	n = attr.va_atime.tv_sec;
396 #ifdef	DEBUG
397 	printf("root time: 0x%lx\n", n);
398 #endif
399 	setrootfstime(n);
400 
401 out:
402 	if (error)
403 		nfs_boot_cleanup(nd, l);
404 	kmem_free(nd, sizeof(*nd));
405 	return (error);
406 }
407 
408 /*
409  * Internal version of mount system call for diskless setup.
410  * Separate function because we used to call it twice.
411  * (once for root and once for swap)
412  */
413 static int
414 nfs_mount_diskless(struct nfs_dlmount *ndmntp, const char *mntname, struct mount **mpp, struct vnode **vpp, struct lwp *l)
415 	/* mntname:	 mount point name */
416 {
417 	struct mount *mp;
418 	struct mbuf *m;
419 	int error;
420 
421 	vfs_rootmountalloc(MOUNT_NFS, mntname, &mp);
422 
423 	mp->mnt_op = &nfs_vfsops;
424 
425 	/*
426 	 * Historical practice expects NFS root file systems to
427 	 * be initially mounted r/w.
428 	 */
429 	mp->mnt_flag &= ~MNT_RDONLY;
430 
431 	/* Get mbuf for server sockaddr. */
432 	m = m_get(M_WAIT, MT_SONAME);
433 	if (m == NULL)
434 		panic("nfs_mountroot: mget soname for %s", mntname);
435 	MCLAIM(m, &nfs_mowner);
436 	memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr,
437 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
438 
439 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
440 			 ndmntp->ndm_args.hostname, vpp, l);
441 	if (error) {
442 		vfs_unbusy(mp, false, NULL);
443 		vfs_destroy(mp);
444 		printf("nfs_mountroot: mount %s failed: %d\n",
445 		       mntname, error);
446 	} else
447 		*mpp = mp;
448 
449 	return (error);
450 }
451 
452 void
453 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp, struct lwp *l)
454 {
455 	int s;
456 	int adjsock;
457 	int maxio;
458 
459 	s = splsoftnet();
460 
461 	/*
462 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
463 	 * no sense in that context.
464 	 */
465 	if (argp->sotype == SOCK_STREAM)
466 		argp->flags &= ~NFSMNT_NOCONN;
467 
468 	/*
469 	 * Cookie translation is not needed for v2, silently ignore it.
470 	 */
471 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
472 	    NFSMNT_XLATECOOKIE)
473 		argp->flags &= ~NFSMNT_XLATECOOKIE;
474 
475 	/* Re-bind if rsrvd port requested and wasn't on one */
476 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
477 		  && (argp->flags & NFSMNT_RESVPORT);
478 	/* Also re-bind if we're switching to/from a connected UDP socket */
479 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
480 		    (argp->flags & NFSMNT_NOCONN));
481 
482 	/* Update flags. */
483 	nmp->nm_flag = argp->flags;
484 	splx(s);
485 
486 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
487 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
488 		if (nmp->nm_timeo < NFS_MINTIMEO)
489 			nmp->nm_timeo = NFS_MINTIMEO;
490 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
491 			nmp->nm_timeo = NFS_MAXTIMEO;
492 	}
493 
494 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
495 		nmp->nm_retry = argp->retrans;
496 		if (nmp->nm_retry > NFS_MAXREXMIT)
497 			nmp->nm_retry = NFS_MAXREXMIT;
498 	}
499 
500 #ifndef NFS_V2_ONLY
501 	if (argp->flags & NFSMNT_NFSV3) {
502 		if (argp->sotype == SOCK_DGRAM)
503 			maxio = NFS_MAXDGRAMDATA;
504 		else
505 			maxio = NFS_MAXDATA;
506 	} else
507 #endif
508 		maxio = NFS_V2MAXDATA;
509 
510 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
511 		int osize = nmp->nm_wsize;
512 		nmp->nm_wsize = argp->wsize;
513 		/* Round down to multiple of blocksize */
514 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
515 		if (nmp->nm_wsize <= 0)
516 			nmp->nm_wsize = NFS_FABLKSIZE;
517 		adjsock |= (nmp->nm_wsize != osize);
518 	}
519 	if (nmp->nm_wsize > maxio)
520 		nmp->nm_wsize = maxio;
521 	if (nmp->nm_wsize > MAXBSIZE)
522 		nmp->nm_wsize = MAXBSIZE;
523 
524 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
525 		int osize = nmp->nm_rsize;
526 		nmp->nm_rsize = argp->rsize;
527 		/* Round down to multiple of blocksize */
528 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
529 		if (nmp->nm_rsize <= 0)
530 			nmp->nm_rsize = NFS_FABLKSIZE;
531 		adjsock |= (nmp->nm_rsize != osize);
532 	}
533 	if (nmp->nm_rsize > maxio)
534 		nmp->nm_rsize = maxio;
535 	if (nmp->nm_rsize > MAXBSIZE)
536 		nmp->nm_rsize = MAXBSIZE;
537 
538 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
539 		nmp->nm_readdirsize = argp->readdirsize;
540 		/* Round down to multiple of minimum blocksize */
541 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
542 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
543 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
544 		/* Bigger than buffer size makes no sense */
545 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
546 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
547 	} else if (argp->flags & NFSMNT_RSIZE)
548 		nmp->nm_readdirsize = nmp->nm_rsize;
549 
550 	if (nmp->nm_readdirsize > maxio)
551 		nmp->nm_readdirsize = maxio;
552 
553 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
554 		argp->maxgrouplist <= NFS_MAXGRPS)
555 		nmp->nm_numgrps = argp->maxgrouplist;
556 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
557 		argp->readahead <= NFS_MAXRAHEAD)
558 		nmp->nm_readahead = argp->readahead;
559 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
560 		argp->deadthresh <= NFS_NEVERDEAD)
561 		nmp->nm_deadthresh = argp->deadthresh;
562 
563 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
564 		    (nmp->nm_soproto != argp->proto));
565 	nmp->nm_sotype = argp->sotype;
566 	nmp->nm_soproto = argp->proto;
567 
568 	if (nmp->nm_so && adjsock) {
569 		nfs_safedisconnect(nmp);
570 		if (nmp->nm_sotype == SOCK_DGRAM)
571 			while (nfs_connect(nmp, (struct nfsreq *)0, l)) {
572 				printf("nfs_args: retrying connect\n");
573 				kpause("nfscn3", false, hz, NULL);
574 			}
575 	}
576 }
577 
578 /*
579  * VFS Operations.
580  *
581  * mount system call
582  * It seems a bit dumb to copyinstr() the host and path here and then
583  * memcpy() them in mountnfs(), but I wanted to detect errors before
584  * doing the sockargs() call because sockargs() allocates an mbuf and
585  * an error after that means that I have to release the mbuf.
586  */
587 /* ARGSUSED */
588 int
589 nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
590 {
591 	struct lwp *l = curlwp;
592 	int error;
593 	struct nfs_args *args = data;
594 	struct mbuf *nam;
595 	struct nfsmount *nmp = VFSTONFS(mp);
596 	struct sockaddr *sa;
597 	struct vnode *vp;
598 	char *pth, *hst;
599 	struct proc *p;
600 	size_t len;
601 	u_char *nfh;
602 
603 	if (*data_len < sizeof *args)
604 		return EINVAL;
605 
606 	p = l->l_proc;
607 	if (mp->mnt_flag & MNT_GETARGS) {
608 
609 		if (nmp == NULL)
610 			return (EIO);
611 		if (args->addr != NULL) {
612 			sa = mtod(nmp->nm_nam, struct sockaddr *);
613 			error = copyout(sa, args->addr, sa->sa_len);
614 			if (error)
615 				return (error);
616 			args->addrlen = sa->sa_len;
617 		} else
618 			args->addrlen = 0;
619 
620 		args->version = NFS_ARGSVERSION;
621 		args->sotype = nmp->nm_sotype;
622 		args->proto = nmp->nm_soproto;
623 		args->fh = NULL;
624 		args->fhsize = 0;
625 		args->flags = nmp->nm_flag;
626 		args->wsize = nmp->nm_wsize;
627 		args->rsize = nmp->nm_rsize;
628 		args->readdirsize = nmp->nm_readdirsize;
629 		args->timeo = nmp->nm_timeo;
630 		args->retrans = nmp->nm_retry;
631 		args->maxgrouplist = nmp->nm_numgrps;
632 		args->readahead = nmp->nm_readahead;
633 		args->leaseterm = 0; /* dummy */
634 		args->deadthresh = nmp->nm_deadthresh;
635 		args->hostname = NULL;
636 		*data_len = sizeof *args;
637 		return 0;
638 	}
639 
640 	if (args->version != NFS_ARGSVERSION)
641 		return (EPROGMISMATCH);
642 	if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB))
643 		return (EPROGUNAVAIL);
644 #ifdef NFS_V2_ONLY
645 	if (args->flags & NFSMNT_NFSV3)
646 		return (EPROGMISMATCH);
647 #endif
648 	if (mp->mnt_flag & MNT_UPDATE) {
649 		if (nmp == NULL)
650 			return (EIO);
651 		/*
652 		 * When doing an update, we can't change from or to
653 		 * v3, or change cookie translation
654 		 */
655 		args->flags = (args->flags & ~(NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)) |
656 		    (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_XLATECOOKIE));
657 		nfs_decode_args(nmp, args, l);
658 		return (0);
659 	}
660 	if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX)
661 		return (EINVAL);
662 	nfh = malloc(NFSX_V3FHMAX, M_TEMP, M_WAITOK);
663 	error = copyin(args->fh, nfh, args->fhsize);
664 	if (error)
665 		goto free_nfh;
666 	pth = malloc(MNAMELEN, M_TEMP, M_WAITOK);
667 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
668 	if (error)
669 		goto free_pth;
670 	memset(&pth[len], 0, MNAMELEN - len);
671 	hst = malloc(MNAMELEN, M_TEMP, M_WAITOK);
672 	error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len);
673 	if (error)
674 		goto free_hst;
675 	memset(&hst[len], 0, MNAMELEN - len);
676 	/* sockargs() call must be after above copyin() calls */
677 	error = sockargs(&nam, args->addr, args->addrlen, MT_SONAME);
678 	if (error)
679 		goto free_hst;
680 	MCLAIM(nam, &nfs_mowner);
681 	args->fh = nfh;
682 	error = mountnfs(args, mp, nam, pth, hst, &vp, l);
683 
684 free_hst:
685 	free(hst, M_TEMP);
686 free_pth:
687 	free(pth, M_TEMP);
688 free_nfh:
689 	free(nfh, M_TEMP);
690 
691 	return (error);
692 }
693 
694 /*
695  * Common code for mount and mountroot
696  */
697 int
698 mountnfs(struct nfs_args *argp, struct mount *mp, struct mbuf *nam, const char *pth, const char *hst, struct vnode **vpp, struct lwp *l)
699 {
700 	struct nfsmount *nmp;
701 	struct nfsnode *np;
702 	struct vnode *vp;
703 	int error;
704 	struct vattr *attrs;
705 	kauth_cred_t cr;
706 	char iosname[IOSTATNAMELEN];
707 
708 	/*
709 	 * If the number of nfs iothreads to use has never
710 	 * been set, create a reasonable number of them.
711 	 */
712 
713 	if (nfs_niothreads < 0) {
714 		nfs_set_niothreads(NFS_DEFAULT_NIOTHREADS);
715 	}
716 
717 	if (mp->mnt_flag & MNT_UPDATE) {
718 		nmp = VFSTONFS(mp);
719 		/* update paths, file handles, etc, here	XXX */
720 		m_freem(nam);
721 		return (0);
722 	} else {
723 		nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP);
724 		mp->mnt_data = nmp;
725 		TAILQ_INIT(&nmp->nm_uidlruhead);
726 		TAILQ_INIT(&nmp->nm_bufq);
727 		rw_init(&nmp->nm_writeverflock);
728 		mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE);
729 		rw_init(&nmp->nm_rbtlock);
730 		cv_init(&nmp->nm_rcvcv, "nfsrcv");
731 		cv_init(&nmp->nm_sndcv, "nfssnd");
732 		cv_init(&nmp->nm_aiocv, "nfsaio");
733 		cv_init(&nmp->nm_disconcv, "nfsdis");
734 		nfs_rbtinit(nmp);
735 	}
736 	vfs_getnewfsid(mp);
737 	nmp->nm_mountp = mp;
738 
739 #ifndef NFS_V2_ONLY
740 	if ((argp->flags & NFSMNT_NFSV3) == 0)
741 #endif
742 	{
743 		if (argp->fhsize != NFSX_V2FH) {
744 			return EINVAL;
745 		}
746 	}
747 
748 	/*
749 	 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
750 	 * will overwrite this.
751 	 */
752 	nmp->nm_maxfilesize = 0xffffffffLL;
753 
754 	nmp->nm_timeo = NFS_TIMEO;
755 	nmp->nm_retry = NFS_RETRANS;
756 	nmp->nm_wsize = NFS_WSIZE;
757 	nmp->nm_rsize = NFS_RSIZE;
758 	nmp->nm_readdirsize = NFS_READDIRSIZE;
759 	nmp->nm_numgrps = NFS_MAXGRPS;
760 	nmp->nm_readahead = NFS_DEFRAHEAD;
761 	nmp->nm_deadthresh = NFS_DEFDEADTHRESH;
762 	error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE,
763 	    mp->mnt_op->vfs_name, mp, l);
764 	if (error)
765 		goto bad;
766 	nmp->nm_nam = nam;
767 
768 	/* Set up the sockets and per-host congestion */
769 	nmp->nm_sotype = argp->sotype;
770 	nmp->nm_soproto = argp->proto;
771 
772 	nfs_decode_args(nmp, argp, l);
773 
774 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
775 	mp->mnt_dev_bshift = DEV_BSHIFT;
776 
777 	/*
778 	 * For Connection based sockets (TCP,...) defer the connect until
779 	 * the first request, in case the server is not responding.
780 	 */
781 	if (nmp->nm_sotype == SOCK_DGRAM &&
782 		(error = nfs_connect(nmp, (struct nfsreq *)0, l)))
783 		goto bad;
784 
785 	/*
786 	 * This is silly, but it has to be set so that vinifod() works.
787 	 * We do not want to do an nfs_statvfs() here since we can get
788 	 * stuck on a dead server and we are holding a lock on the mount
789 	 * point.
790 	 */
791 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
792 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
793 	if (error)
794 		goto bad;
795 	vp = NFSTOV(np);
796 	attrs = malloc(sizeof(struct vattr), M_TEMP, M_WAITOK);
797 	VOP_GETATTR(vp, attrs, l->l_cred);
798 	if ((nmp->nm_flag & NFSMNT_NFSV3) && (vp->v_type == VDIR)) {
799 		cr = kauth_cred_alloc();
800 		kauth_cred_setuid(cr, attrs->va_uid);
801 		kauth_cred_seteuid(cr, attrs->va_uid);
802 		kauth_cred_setsvuid(cr, attrs->va_uid);
803 		kauth_cred_setgid(cr, attrs->va_gid);
804 		kauth_cred_setegid(cr, attrs->va_gid);
805 		kauth_cred_setsvgid(cr, attrs->va_gid);
806 		nfs_cookieheuristic(vp, &nmp->nm_iflag, l, cr);
807 		kauth_cred_free(cr);
808 	}
809 	free(attrs, M_TEMP);
810 
811 	/*
812 	 * A reference count is needed on the nfsnode representing the
813 	 * remote root.  If this object is not persistent, then backward
814 	 * traversals of the mount point (i.e. "..") will not work if
815 	 * the nfsnode gets flushed out of the cache. Ufs does not have
816 	 * this problem, because one can identify root inodes by their
817 	 * number == UFS_ROOTINO (2). So, just unlock, but no rele.
818 	 */
819 
820 	nmp->nm_vnode = vp;
821 	if (vp->v_type == VNON)
822 		vp->v_type = VDIR;
823 	vp->v_vflag |= VV_ROOT;
824 	VOP_UNLOCK(vp);
825 	*vpp = vp;
826 
827 	snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++);
828 	nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname);
829 
830 	return (0);
831 bad:
832 	nfs_disconnect(nmp);
833 	rw_destroy(&nmp->nm_writeverflock);
834 	rw_destroy(&nmp->nm_rbtlock);
835 	mutex_destroy(&nmp->nm_lock);
836 	cv_destroy(&nmp->nm_rcvcv);
837 	cv_destroy(&nmp->nm_sndcv);
838 	cv_destroy(&nmp->nm_aiocv);
839 	cv_destroy(&nmp->nm_disconcv);
840 	kmem_free(nmp, sizeof(*nmp));
841 	m_freem(nam);
842 	return (error);
843 }
844 
845 /*
846  * unmount system call
847  */
848 int
849 nfs_unmount(struct mount *mp, int mntflags)
850 {
851 	struct nfsmount *nmp;
852 	struct vnode *vp;
853 	int error, flags = 0;
854 
855 	if (mntflags & MNT_FORCE)
856 		flags |= FORCECLOSE;
857 	nmp = VFSTONFS(mp);
858 	/*
859 	 * Goes something like this..
860 	 * - Check for activity on the root vnode (other than ourselves).
861 	 * - Call vflush() to clear out vnodes for this file system,
862 	 *   except for the root vnode.
863 	 * - Decrement reference on the vnode representing remote root.
864 	 * - Close the socket
865 	 * - Free up the data structures
866 	 */
867 	/*
868 	 * We need to decrement the ref. count on the nfsnode representing
869 	 * the remote root.  See comment in mountnfs().
870 	 */
871 	vp = nmp->nm_vnode;
872 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
873 	if (error != 0)
874 		return error;
875 
876 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 1) {
877 		VOP_UNLOCK(vp);
878 		return (EBUSY);
879 	}
880 
881 	error = vflush(mp, vp, flags);
882 	if (error) {
883 		VOP_UNLOCK(vp);
884 		return (error);
885 	}
886 
887 	/*
888 	 * We are now committed to the unmount; mark the mount structure
889 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
890 	 * will go away cleanly.
891 	 */
892 	nmp->nm_iflag |= NFSMNT_DISMNT;
893 
894 	/*
895 	 * Clean up the stats... note that we carefully avoid decrementing
896 	 * nfs_mount_count here for good reason - we may not be unmounting
897 	 * the last thing mounted.
898 	 */
899 	iostat_free(nmp->nm_stats);
900 
901 	/*
902 	 * There is one reference count to get rid of here
903 	 * (see comment in mountnfs()).
904 	 */
905 	VOP_UNLOCK(vp);
906 	vgone(vp);
907 	nfs_disconnect(nmp);
908 	m_freem(nmp->nm_nam);
909 
910 	rw_destroy(&nmp->nm_writeverflock);
911 	rw_destroy(&nmp->nm_rbtlock);
912 	mutex_destroy(&nmp->nm_lock);
913 	cv_destroy(&nmp->nm_rcvcv);
914 	cv_destroy(&nmp->nm_sndcv);
915 	cv_destroy(&nmp->nm_aiocv);
916 	cv_destroy(&nmp->nm_disconcv);
917 	kmem_free(nmp, sizeof(*nmp));
918 	return (0);
919 }
920 
921 /*
922  * Return root of a filesystem
923  */
924 int
925 nfs_root(struct mount *mp, struct vnode **vpp)
926 {
927 	struct vnode *vp;
928 	struct nfsmount *nmp;
929 	int error;
930 
931 	nmp = VFSTONFS(mp);
932 	vp = nmp->nm_vnode;
933 	vref(vp);
934 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
935 	if (error != 0) {
936 		vrele(vp);
937 		return error;
938 	}
939 	*vpp = vp;
940 	return (0);
941 }
942 
943 extern int syncprt;
944 
945 /*
946  * Flush out the buffer cache
947  */
948 /* ARGSUSED */
949 int
950 nfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
951 {
952 	struct vnode *vp, *mvp;
953 	int error, allerror = 0;
954 
955 	/*
956 	 * Force stale buffer cache information to be flushed.
957 	 */
958 	mvp = vnalloc(mp);
959 loop:
960 	/*
961 	 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
962 	 * and vclean() can be called indirectly
963 	 */
964 	mutex_enter(&mntvnode_lock);
965 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) {
966 		vmark(mvp, vp);
967 		if (vp->v_mount != mp || vismarker(vp))
968 			continue;
969 		mutex_enter(vp->v_interlock);
970 		/* XXX MNT_LAZY cannot be right? */
971 		if (waitfor == MNT_LAZY ||
972 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
973 		     UVM_OBJ_IS_CLEAN(&vp->v_uobj))) {
974 			mutex_exit(vp->v_interlock);
975 			continue;
976 		}
977 		mutex_exit(&mntvnode_lock);
978 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT);
979 		if (error != 0) {
980 			if (error != ENOENT) {
981 				mutex_enter(&mntvnode_lock);
982 				continue;
983 			}
984 			(void)vunmark(mvp);
985 			goto loop;
986 		}
987 		error = VOP_FSYNC(vp, cred,
988 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0);
989 		if (error)
990 			allerror = error;
991 		vput(vp);
992 		mutex_enter(&mntvnode_lock);
993 	}
994 	mutex_exit(&mntvnode_lock);
995 	vnfree(mvp);
996 	return (allerror);
997 }
998 
999 /*
1000  * NFS flat namespace lookup.
1001  * Currently unsupported.
1002  */
1003 /* ARGSUSED */
1004 int
1005 nfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1006 {
1007 
1008 	return (EOPNOTSUPP);
1009 }
1010 
1011 /*
1012  * Do that sysctl thang...
1013  */
1014 static int
1015 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS)
1016 {
1017 	struct sysctlnode node;
1018 	int val;
1019 	int error;
1020 
1021 	val = nfs_niothreads;
1022 	node = *rnode;
1023 	node.sysctl_data = &val;
1024         error = sysctl_lookup(SYSCTLFN_CALL(&node));
1025 	if (error || newp == NULL)
1026 		return error;
1027 
1028 	return nfs_set_niothreads(val);
1029 }
1030 
1031 static void
1032 nfs_sysctl_init(void)
1033 {
1034 
1035 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1036 		       CTLFLAG_PERMANENT,
1037 		       CTLTYPE_NODE, "vfs", NULL,
1038 		       NULL, 0, NULL, 0,
1039 		       CTL_VFS, CTL_EOL);
1040 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1041 		       CTLFLAG_PERMANENT,
1042 		       CTLTYPE_NODE, "nfs",
1043 		       SYSCTL_DESCR("NFS vfs options"),
1044 		       NULL, 0, NULL, 0,
1045 		       CTL_VFS, 2, CTL_EOL);
1046 	/*
1047 	 * XXX the "2" above could be dynamic, thereby eliminating one
1048 	 * more instance of the "number to vfs" mapping problem, but
1049 	 * "2" is the order as taken from sys/mount.h
1050 	 */
1051 
1052 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1053 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1054 		       CTLTYPE_STRUCT, "nfsstats",
1055 		       SYSCTL_DESCR("NFS operation statistics"),
1056 		       NULL, 0, &nfsstats, sizeof(nfsstats),
1057 		       CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL);
1058 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1059 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1060 		       CTLTYPE_INT, "iothreads",
1061 		       SYSCTL_DESCR("Number of NFS client processes desired"),
1062 		       sysctl_vfs_nfs_iothreads, 0, NULL, 0,
1063 		       CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL);
1064 }
1065 
1066 static void
1067 nfs_sysctl_fini(void)
1068 {
1069 
1070 	sysctl_teardown(&nfs_clog);
1071 }
1072 
1073 /* ARGSUSED */
1074 int
1075 nfs_fhtovp(struct mount *mp, struct fid *fid, struct vnode **vpp)
1076 {
1077 	size_t fidsize;
1078 	size_t fhsize;
1079 	struct nfsnode *np;
1080 	int error;
1081 	struct vattr va;
1082 
1083 	fidsize = fid->fid_len;
1084 	if (fidsize < sizeof(*fid)) {
1085 		return EINVAL;
1086 	}
1087 	fhsize = fidsize - sizeof(*fid);
1088 	if ((fhsize % NFSX_UNSIGNED) != 0) {
1089 		return EINVAL;
1090 	}
1091 	if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) {
1092 		if (fhsize > NFSX_V3FHMAX || fhsize == 0) {
1093 			return EINVAL;
1094 		}
1095 	} else {
1096 		if (fhsize != NFSX_V2FH) {
1097 			return EINVAL;
1098 		}
1099 	}
1100 	error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np);
1101 	if (error) {
1102 		return error;
1103 	}
1104 	*vpp = NFSTOV(np);
1105 	error = VOP_GETATTR(*vpp, &va, kauth_cred_get());
1106 	if (error != 0) {
1107 		vput(*vpp);
1108 		*vpp = NULLVP;
1109 	}
1110 	return error;
1111 }
1112 
1113 /* ARGSUSED */
1114 int
1115 nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize)
1116 {
1117 	struct nfsnode *np;
1118 	struct fid *fid;
1119 	size_t fidsize;
1120 	int error = 0;
1121 
1122 	np = VTONFS(vp);
1123 	fidsize = sizeof(*fid) + np->n_fhsize;
1124 	if (*bufsize < fidsize) {
1125 		error = E2BIG;
1126 	}
1127 	*bufsize = fidsize;
1128 	if (error == 0) {
1129 		struct fid fid_store;
1130 
1131 		fid = &fid_store;
1132 		memset(fid, 0, sizeof(*fid));
1133 		fid->fid_len = fidsize;
1134 		memcpy(buf, fid, sizeof(*fid));
1135 		memcpy(buf->fid_data, np->n_fhp, np->n_fhsize);
1136 	}
1137 	return error;
1138 }
1139 
1140 /*
1141  * Vfs start routine, a no-op.
1142  */
1143 /* ARGSUSED */
1144 int
1145 nfs_start(struct mount *mp, int flags)
1146 {
1147 
1148 	return (0);
1149 }
1150 
1151 /*
1152  * Called once at VFS init to initialize client-specific data structures.
1153  */
1154 void
1155 nfs_vfs_init(void)
1156 {
1157 
1158 	/* Initialize NFS server / client shared data. */
1159 	nfs_init();
1160 	nfs_node_init();
1161 
1162 	/* Initialize the kqueue structures */
1163 	nfs_kqinit();
1164 	/* Initialize the iod structures */
1165 	nfs_iodinit();
1166 
1167 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1168 }
1169 
1170 void
1171 nfs_vfs_done(void)
1172 {
1173 
1174 	nfs_node_done();
1175 	nfs_kqfini();
1176 	nfs_iodfini();
1177 }
1178