xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision 53d1339bf7f9c7367b35a9e1ebe693f9b047a47b)
1 /*	$NetBSD: nfs_vfsops.c,v 1.242 2021/04/02 03:07:54 christos 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 1.242 2021/04/02 03:07:54 christos 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 	.vfs_name = MOUNT_NFS,
108 	.vfs_min_mount_data = sizeof (struct nfs_args),
109 	.vfs_mount = nfs_mount,
110 	.vfs_start = nfs_start,
111 	.vfs_unmount = nfs_unmount,
112 	.vfs_root = nfs_root,
113 	.vfs_quotactl = (void *)eopnotsupp,
114 	.vfs_statvfs = nfs_statvfs,
115 	.vfs_sync = nfs_sync,
116 	.vfs_loadvnode = nfs_loadvnode,
117 	.vfs_vget = nfs_vget,
118 	.vfs_fhtovp = nfs_fhtovp,
119 	.vfs_vptofh = nfs_vptofh,
120 	.vfs_init = nfs_vfs_init,
121 	.vfs_done = nfs_vfs_done,
122 	.vfs_mountroot = nfs_mountroot,
123 	.vfs_snapshot = (void *)eopnotsupp,
124 	.vfs_extattrctl = vfs_stdextattrctl,
125 	.vfs_suspendctl = genfs_suspendctl,
126 	.vfs_renamelock_enter = genfs_renamelock_enter,
127 	.vfs_renamelock_exit = genfs_renamelock_exit,
128 	.vfs_fsync = (void *)eopnotsupp,
129 	.vfs_opv_descs = nfs_vnodeopv_descs
130 };
131 
132 extern u_int32_t nfs_procids[NFS_NPROCS];
133 extern u_int32_t nfs_prog, nfs_vers;
134 
135 static int nfs_mount_diskless(struct nfs_dlmount *, const char *,
136     struct mount **, struct vnode **, struct lwp *);
137 
138 static int
139 nfs_modcmd(modcmd_t cmd, void *arg)
140 {
141 	int error;
142 
143 	switch (cmd) {
144 	case MODULE_CMD_INIT:
145 		error = vfs_attach(&nfs_vfsops);
146 		return error;
147 	case MODULE_CMD_FINI:
148 		error = vfs_detach(&nfs_vfsops);
149 		return error;
150 	default:
151 		return ENOTTY;
152 	}
153 }
154 
155 /*
156  * nfs statvfs call
157  */
158 int
159 nfs_statvfs(struct mount *mp, struct statvfs *sbp)
160 {
161 	struct lwp *l = curlwp;
162 	struct vnode *vp;
163 	struct nfs_statfs *sfp;
164 	char *cp;
165 	u_int32_t *tl;
166 	int32_t t1, t2;
167 	char *bpos, *dpos, *cp2;
168 	struct nfsmount *nmp = VFSTONFS(mp);
169 	int error = 0, retattr;
170 #ifdef NFS_V2_ONLY
171 	const int v3 = 0;
172 #else
173 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
174 #endif
175 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
176 	kauth_cred_t cred;
177 	u_quad_t tquad;
178 	struct nfsnode *np;
179 
180 #ifndef nolint
181 	sfp = (struct nfs_statfs *)0;
182 #endif
183 	vp = nmp->nm_vnode;
184 	np = VTONFS(vp);
185 	cred = kauth_cred_alloc();
186 #ifndef NFS_V2_ONLY
187 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
188 		(void)nfs_fsinfo(nmp, vp, cred, l);
189 #endif
190 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
191 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
192 	nfsm_fhtom(np, v3);
193 	nfsm_request(np, NFSPROC_FSSTAT, l, cred);
194 	if (v3)
195 		nfsm_postop_attr(vp, retattr, 0);
196 	if (error) {
197 		if (mrep != NULL) {
198 			if (mrep->m_next != NULL)
199 				printf("nfs_vfsops: nfs_statvfs would lose buffers\n");
200 			m_freem(mrep);
201 		}
202 		goto nfsmout;
203 	}
204 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
205 	sbp->f_flag = nmp->nm_flag;
206 	sbp->f_iosize = uimin(nmp->nm_rsize, nmp->nm_wsize);
207 	if (v3) {
208 		sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE;
209 		tquad = fxdr_hyper(&sfp->sf_tbytes);
210 		sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
211 		tquad = fxdr_hyper(&sfp->sf_fbytes);
212 		sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
213 		tquad = fxdr_hyper(&sfp->sf_abytes);
214 		tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
215 		sbp->f_bresvd = sbp->f_bfree - tquad;
216 		sbp->f_bavail = tquad;
217 		/* Handle older NFS servers returning negative values */
218 		if ((quad_t)sbp->f_bavail < 0)
219 			sbp->f_bavail = 0;
220 		tquad = fxdr_hyper(&sfp->sf_tfiles);
221 		sbp->f_files = tquad;
222 		tquad = fxdr_hyper(&sfp->sf_ffiles);
223 		sbp->f_ffree = tquad;
224 		sbp->f_favail = tquad;
225 		sbp->f_fresvd = 0;
226 	} else {
227 		sbp->f_bsize = NFS_FABLKSIZE;
228 		sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
229 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
230 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
231 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
232 		sbp->f_fresvd = 0;
233 		sbp->f_files = 0;
234 		sbp->f_ffree = 0;
235 		sbp->f_favail = 0;
236 		sbp->f_fresvd = 0;
237 	}
238 	copy_statvfs_info(sbp, mp);
239 	nfsm_reqdone;
240 	kauth_cred_free(cred);
241 	return (error);
242 }
243 
244 #ifndef NFS_V2_ONLY
245 /*
246  * nfs version 3 fsinfo rpc call
247  */
248 int
249 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, kauth_cred_t cred, struct lwp *l)
250 {
251 	struct nfsv3_fsinfo *fsp;
252 	char *cp;
253 	int32_t t1, t2;
254 	u_int32_t *tl, pref, xmax;
255 	char *bpos, *dpos, *cp2;
256 	int error = 0, retattr;
257 	struct mbuf *mreq, *mrep, *md, *mb;
258 	u_int64_t maxfsize;
259 	struct nfsnode *np = VTONFS(vp);
260 
261 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
262 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
263 	nfsm_fhtom(np, 1);
264 	nfsm_request(np, NFSPROC_FSINFO, l, cred);
265 	nfsm_postop_attr(vp, retattr, 0);
266 	if (!error) {
267 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
268 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
269 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
270 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
271 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
272 				~(NFS_FABLKSIZE - 1);
273 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
274 		if (xmax < nmp->nm_wsize && xmax > 0) {
275 			nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1);
276 			if (nmp->nm_wsize == 0)
277 				nmp->nm_wsize = xmax;
278 		}
279 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
280 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
281 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
282 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
283 				~(NFS_FABLKSIZE - 1);
284 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
285 		if (xmax < nmp->nm_rsize && xmax > 0) {
286 			nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1);
287 			if (nmp->nm_rsize == 0)
288 				nmp->nm_rsize = xmax;
289 		}
290 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
291 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
292 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
293 				~(NFS_DIRFRAGSIZ - 1);
294 		if (xmax < nmp->nm_readdirsize && xmax > 0) {
295 			nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1);
296 			if (nmp->nm_readdirsize == 0)
297 				nmp->nm_readdirsize = xmax;
298 		}
299 		/* XXX */
300 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
301 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
302 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
303 			nmp->nm_maxfilesize = maxfsize;
304 		nmp->nm_mountp->mnt_fs_bshift =
305 		    ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
306 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
307 	}
308 	nfsm_reqdone;
309 	return (error);
310 }
311 #endif
312 
313 /*
314  * Mount a remote root fs via. NFS.  It goes like this:
315  * - Call nfs_boot_init() to fill in the nfs_diskless struct
316  * - build the rootfs mount point and call mountnfs() to do the rest.
317  */
318 int
319 nfs_mountroot(void)
320 {
321 	struct timespec ts;
322 	struct nfs_diskless *nd;
323 	struct vattr attr;
324 	struct mount *mp;
325 	struct vnode *vp;
326 	struct lwp *l;
327 	long n;
328 	int error;
329 
330 	l = curlwp; /* XXX */
331 
332 	if (device_class(root_device) != DV_IFNET)
333 		return (ENODEV);
334 
335 	/*
336 	 * XXX time must be non-zero when we init the interface or else
337 	 * the arp code will wedge.  [Fixed now in if_ether.c]
338 	 * However, the NFS attribute cache gives false "hits" when the
339 	 * current time < nfs_attrtimeo(nmp, np) so keep this in for now.
340 	 */
341 	if (time_second < NFS_MAXATTRTIMO) {
342 		ts.tv_sec = NFS_MAXATTRTIMO;
343 		ts.tv_nsec = 0;
344 		tc_setclock(&ts);
345 	}
346 
347 	/*
348 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
349 	 * Side effect:  Finds and configures a network interface.
350 	 */
351 	nd = kmem_zalloc(sizeof(*nd), KM_SLEEP);
352 	error = nfs_boot_init(nd, l);
353 	if (error) {
354 		kmem_free(nd, sizeof(*nd));
355 		return (error);
356 	}
357 
358 	/*
359 	 * Create the root mount point.
360 	 */
361 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l);
362 	if (error)
363 		goto out;
364 	printf("root on %s\n", nd->nd_root.ndm_host);
365 
366 	/*
367 	 * Link it into the mount list.
368 	 */
369 	mountlist_append(mp);
370 	rootvp = vp;
371 	mp->mnt_vnodecovered = NULLVP;
372 	vfs_unbusy(mp);
373 
374 	/* Get root attributes (for the time). */
375 	vn_lock(vp, LK_SHARED | LK_RETRY);
376 	error = VOP_GETATTR(vp, &attr, l->l_cred);
377 	VOP_UNLOCK(vp);
378 	if (error)
379 		panic("nfs_mountroot: getattr for root");
380 	n = attr.va_atime.tv_sec;
381 #ifdef	DEBUG
382 	printf("root time: 0x%lx\n", n);
383 #endif
384 	setrootfstime(n);
385 
386 out:
387 	if (error)
388 		nfs_boot_cleanup(nd, l);
389 	kmem_free(nd, sizeof(*nd));
390 	return (error);
391 }
392 
393 /*
394  * Internal version of mount system call for diskless setup.
395  * Separate function because we used to call it twice.
396  * (once for root and once for swap)
397  */
398 static int
399 nfs_mount_diskless(struct nfs_dlmount *ndmntp, const char *mntname, struct mount **mpp, struct vnode **vpp, struct lwp *l)
400 	/* mntname:	 mount point name */
401 {
402 	struct mount *mp;
403 	struct mbuf *m;
404 	int error;
405 
406 	vfs_rootmountalloc(MOUNT_NFS, mntname, &mp);
407 
408 	mp->mnt_op = &nfs_vfsops;
409 
410 	/*
411 	 * Historical practice expects NFS root file systems to
412 	 * be initially mounted r/w.
413 	 */
414 	mp->mnt_flag &= ~MNT_RDONLY;
415 
416 	/* Get mbuf for server sockaddr. */
417 	m = m_get(M_WAIT, MT_SONAME);
418 	if (m == NULL)
419 		panic("nfs_mountroot: mget soname for %s", mntname);
420 	MCLAIM(m, &nfs_mowner);
421 	memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr,
422 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
423 
424 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
425 			 ndmntp->ndm_args.hostname, vpp, l);
426 	if (error) {
427 		vfs_unbusy(mp);
428 		vfs_rele(mp);
429 		printf("nfs_mountroot: mount %s failed: %d\n",
430 		       mntname, error);
431 	} else
432 		*mpp = mp;
433 
434 	return (error);
435 }
436 
437 void
438 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp, struct lwp *l)
439 {
440 	int s;
441 	int adjsock;
442 	int maxio;
443 
444 	s = splsoftnet();
445 
446 	/*
447 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
448 	 * no sense in that context.
449 	 */
450 	if (argp->sotype == SOCK_STREAM)
451 		argp->flags &= ~NFSMNT_NOCONN;
452 
453 	/*
454 	 * Cookie translation is not needed for v2, silently ignore it.
455 	 */
456 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
457 	    NFSMNT_XLATECOOKIE)
458 		argp->flags &= ~NFSMNT_XLATECOOKIE;
459 
460 	/* Re-bind if rsrvd port requested and wasn't on one */
461 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
462 		  && (argp->flags & NFSMNT_RESVPORT);
463 	/* Also re-bind if we're switching to/from a connected UDP socket */
464 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
465 		    (argp->flags & NFSMNT_NOCONN));
466 
467 	/* Update flags. */
468 	nmp->nm_flag = argp->flags;
469 	splx(s);
470 
471 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
472 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
473 		if (nmp->nm_timeo < NFS_MINTIMEO)
474 			nmp->nm_timeo = NFS_MINTIMEO;
475 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
476 			nmp->nm_timeo = NFS_MAXTIMEO;
477 	}
478 
479 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
480 		nmp->nm_retry = argp->retrans;
481 		if (nmp->nm_retry > NFS_MAXREXMIT)
482 			nmp->nm_retry = NFS_MAXREXMIT;
483 	}
484 
485 #ifndef NFS_V2_ONLY
486 	if (argp->flags & NFSMNT_NFSV3) {
487 		if (argp->sotype == SOCK_DGRAM)
488 			maxio = NFS_MAXDGRAMDATA;
489 		else
490 			maxio = NFS_MAXDATA;
491 	} else
492 #endif
493 		maxio = NFS_V2MAXDATA;
494 
495 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
496 		int osize = nmp->nm_wsize;
497 		nmp->nm_wsize = argp->wsize;
498 		/* Round down to multiple of blocksize */
499 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
500 		if (nmp->nm_wsize <= 0)
501 			nmp->nm_wsize = NFS_FABLKSIZE;
502 		adjsock |= (nmp->nm_wsize != osize);
503 	}
504 	if (nmp->nm_wsize > maxio)
505 		nmp->nm_wsize = maxio;
506 	if (nmp->nm_wsize > MAXBSIZE)
507 		nmp->nm_wsize = MAXBSIZE;
508 
509 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
510 		int osize = nmp->nm_rsize;
511 		nmp->nm_rsize = argp->rsize;
512 		/* Round down to multiple of blocksize */
513 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
514 		if (nmp->nm_rsize <= 0)
515 			nmp->nm_rsize = NFS_FABLKSIZE;
516 		adjsock |= (nmp->nm_rsize != osize);
517 	}
518 	if (nmp->nm_rsize > maxio)
519 		nmp->nm_rsize = maxio;
520 	if (nmp->nm_rsize > MAXBSIZE)
521 		nmp->nm_rsize = MAXBSIZE;
522 
523 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
524 		nmp->nm_readdirsize = argp->readdirsize;
525 		/* Round down to multiple of minimum blocksize */
526 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
527 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
528 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
529 		/* Bigger than buffer size makes no sense */
530 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
531 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
532 	} else if (argp->flags & NFSMNT_RSIZE)
533 		nmp->nm_readdirsize = nmp->nm_rsize;
534 
535 	if (nmp->nm_readdirsize > maxio)
536 		nmp->nm_readdirsize = maxio;
537 
538 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
539 		argp->maxgrouplist <= NFS_MAXGRPS)
540 		nmp->nm_numgrps = argp->maxgrouplist;
541 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
542 		argp->readahead <= NFS_MAXRAHEAD)
543 		nmp->nm_readahead = argp->readahead;
544 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
545 		argp->deadthresh <= NFS_NEVERDEAD)
546 		nmp->nm_deadthresh = argp->deadthresh;
547 
548 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
549 		    (nmp->nm_soproto != argp->proto));
550 	nmp->nm_sotype = argp->sotype;
551 	nmp->nm_soproto = argp->proto;
552 
553 	if (nmp->nm_so && adjsock) {
554 		nfs_safedisconnect(nmp);
555 		if (nmp->nm_sotype == SOCK_DGRAM)
556 			while (nfs_connect(nmp, (struct nfsreq *)0, l)) {
557 				printf("nfs_args: retrying connect\n");
558 				kpause("nfscn3", false, hz, NULL);
559 			}
560 	}
561 }
562 
563 /*
564  * VFS Operations.
565  *
566  * mount system call
567  * It seems a bit dumb to copyinstr() the host and path here and then
568  * memcpy() them in mountnfs(), but I wanted to detect errors before
569  * doing the sockargs() call because sockargs() allocates an mbuf and
570  * an error after that means that I have to release the mbuf.
571  */
572 /* ARGSUSED */
573 int
574 nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
575 {
576 	struct lwp *l = curlwp;
577 	int error;
578 	struct nfs_args *args = data;
579 	struct mbuf *nam;
580 	struct nfsmount *nmp = VFSTONFS(mp);
581 	struct sockaddr *sa;
582 	struct vnode *vp;
583 	char *pth, *hst;
584 	size_t len;
585 	u_char *nfh;
586 
587 	if (args == NULL)
588 		return EINVAL;
589 	if (*data_len < sizeof *args)
590 		return EINVAL;
591 
592 	if (mp->mnt_flag & MNT_GETARGS) {
593 
594 		if (nmp == NULL)
595 			return (EIO);
596 		if (args->addr != NULL) {
597 			sa = mtod(nmp->nm_nam, struct sockaddr *);
598 			error = copyout(sa, args->addr, sa->sa_len);
599 			if (error)
600 				return (error);
601 			args->addrlen = sa->sa_len;
602 		} else
603 			args->addrlen = 0;
604 
605 		args->version = NFS_ARGSVERSION;
606 		args->sotype = nmp->nm_sotype;
607 		args->proto = nmp->nm_soproto;
608 		args->fh = NULL;
609 		args->fhsize = 0;
610 		args->flags = nmp->nm_flag;
611 		args->wsize = nmp->nm_wsize;
612 		args->rsize = nmp->nm_rsize;
613 		args->readdirsize = nmp->nm_readdirsize;
614 		args->timeo = nmp->nm_timeo;
615 		args->retrans = nmp->nm_retry;
616 		args->maxgrouplist = nmp->nm_numgrps;
617 		args->readahead = nmp->nm_readahead;
618 		args->leaseterm = 0; /* dummy */
619 		args->deadthresh = nmp->nm_deadthresh;
620 		args->hostname = NULL;
621 		*data_len = sizeof *args;
622 		return 0;
623 	}
624 
625 	if (args->version != NFS_ARGSVERSION)
626 		return (EPROGMISMATCH);
627 	if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB))
628 		return (EPROGUNAVAIL);
629 #ifdef NFS_V2_ONLY
630 	if (args->flags & NFSMNT_NFSV3)
631 		return (EPROGMISMATCH);
632 #endif
633 	if (mp->mnt_flag & MNT_UPDATE) {
634 		if (nmp == NULL)
635 			return (EIO);
636 		/*
637 		 * When doing an update, we can't change from or to
638 		 * v3, or change cookie translation
639 		 */
640 		args->flags = (args->flags & ~(NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)) |
641 		    (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_XLATECOOKIE));
642 		nfs_decode_args(nmp, args, l);
643 		return (0);
644 	}
645 	if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX)
646 		return (EINVAL);
647 	nfh = malloc(NFSX_V3FHMAX, M_TEMP, M_WAITOK);
648 	error = copyin(args->fh, nfh, args->fhsize);
649 	if (error)
650 		goto free_nfh;
651 	pth = malloc(MNAMELEN, M_TEMP, M_WAITOK);
652 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
653 	if (error)
654 		goto free_pth;
655 	memset(&pth[len], 0, MNAMELEN - len);
656 	hst = malloc(MNAMELEN, M_TEMP, M_WAITOK);
657 	error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len);
658 	if (error)
659 		goto free_hst;
660 	memset(&hst[len], 0, MNAMELEN - len);
661 	/* sockargs() call must be after above copyin() calls */
662 	error = sockargs(&nam, args->addr, args->addrlen, UIO_USERSPACE,
663 	    MT_SONAME);
664 	if (error)
665 		goto free_hst;
666 	MCLAIM(nam, &nfs_mowner);
667 	args->fh = nfh;
668 	error = mountnfs(args, mp, nam, pth, hst, &vp, l);
669 
670 free_hst:
671 	free(hst, M_TEMP);
672 free_pth:
673 	free(pth, M_TEMP);
674 free_nfh:
675 	free(nfh, M_TEMP);
676 
677 	return (error);
678 }
679 
680 /*
681  * Common code for mount and mountroot
682  */
683 int
684 mountnfs(struct nfs_args *argp, struct mount *mp, struct mbuf *nam, const char *pth, const char *hst, struct vnode **vpp, struct lwp *l)
685 {
686 	struct nfsmount *nmp;
687 	struct nfsnode *np;
688 	struct vnode *vp;
689 	int error;
690 	struct vattr *attrs;
691 	kauth_cred_t cr;
692 	char iosname[IOSTATNAMELEN];
693 
694 	/*
695 	 * If the number of nfs iothreads to use has never
696 	 * been set, create a reasonable number of them.
697 	 */
698 
699 	if (nfs_niothreads < 0) {
700 		nfs_set_niothreads(NFS_DEFAULT_NIOTHREADS);
701 	}
702 
703 	if (mp->mnt_flag & MNT_UPDATE) {
704 		nmp = VFSTONFS(mp);
705 		/* update paths, file handles, etc, here	XXX */
706 		m_freem(nam);
707 		return 0;
708 	}
709 	nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP);
710 	TAILQ_INIT(&nmp->nm_uidlruhead);
711 	TAILQ_INIT(&nmp->nm_bufq);
712 	rw_init(&nmp->nm_writeverflock);
713 	mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE);
714 	cv_init(&nmp->nm_rcvcv, "nfsrcv");
715 	cv_init(&nmp->nm_sndcv, "nfssnd");
716 	cv_init(&nmp->nm_aiocv, "nfsaio");
717 	cv_init(&nmp->nm_disconcv, "nfsdis");
718 
719 	mp->mnt_data = nmp;
720 	mp->mnt_stat.f_namemax = NFS_MAXNAMLEN;
721 	vfs_getnewfsid(mp);
722 	nmp->nm_mountp = mp;
723 
724 #ifndef NFS_V2_ONLY
725 	if ((argp->flags & NFSMNT_NFSV3) == 0)
726 #endif
727 	{
728 		if (argp->fhsize != NFSX_V2FH) {
729 			return EINVAL;
730 		}
731 	}
732 
733 	/*
734 	 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
735 	 * will overwrite this.
736 	 */
737 	nmp->nm_maxfilesize = 0xffffffffLL;
738 
739 	nmp->nm_timeo = NFS_TIMEO;
740 	nmp->nm_retry = NFS_RETRANS;
741 	nmp->nm_wsize = NFS_WSIZE;
742 	nmp->nm_rsize = NFS_RSIZE;
743 	nmp->nm_readdirsize = NFS_READDIRSIZE;
744 	nmp->nm_numgrps = NFS_MAXGRPS;
745 	nmp->nm_readahead = NFS_DEFRAHEAD;
746 	nmp->nm_deadthresh = NFS_DEFDEADTHRESH;
747 	error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE,
748 	    mp->mnt_op->vfs_name, mp, l);
749 	if (error)
750 		goto bad;
751 	nmp->nm_nam = nam;
752 
753 	/* Set up the sockets and per-host congestion */
754 	nmp->nm_sotype = argp->sotype;
755 	nmp->nm_soproto = argp->proto;
756 
757 	nfs_decode_args(nmp, argp, l);
758 
759 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
760 	mp->mnt_dev_bshift = DEV_BSHIFT;
761 
762 	/*
763 	 * For Connection based sockets (TCP,...) defer the connect until
764 	 * the first request, in case the server is not responding.
765 	 */
766 	if (nmp->nm_sotype == SOCK_DGRAM &&
767 		(error = nfs_connect(nmp, (struct nfsreq *)0, l)))
768 		goto bad;
769 
770 	/*
771 	 * This is silly, but it has to be set so that vinifod() works.
772 	 * We do not want to do an nfs_statvfs() here since we can get
773 	 * stuck on a dead server and we are holding a lock on the mount
774 	 * point.
775 	 */
776 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
777 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
778 	if (error)
779 		goto bad;
780 	vp = NFSTOV(np);
781 	attrs = malloc(sizeof(struct vattr), M_TEMP, M_WAITOK);
782 	VOP_GETATTR(vp, attrs, l->l_cred);
783 	if ((nmp->nm_flag & NFSMNT_NFSV3) && (vp->v_type == VDIR)) {
784 		cr = kauth_cred_alloc();
785 		kauth_cred_setuid(cr, attrs->va_uid);
786 		kauth_cred_seteuid(cr, attrs->va_uid);
787 		kauth_cred_setsvuid(cr, attrs->va_uid);
788 		kauth_cred_setgid(cr, attrs->va_gid);
789 		kauth_cred_setegid(cr, attrs->va_gid);
790 		kauth_cred_setsvgid(cr, attrs->va_gid);
791 		nfs_cookieheuristic(vp, &nmp->nm_iflag, l, cr);
792 		kauth_cred_free(cr);
793 	}
794 	free(attrs, M_TEMP);
795 
796 	/*
797 	 * A reference count is needed on the nfsnode representing the
798 	 * remote root.  If this object is not persistent, then backward
799 	 * traversals of the mount point (i.e. "..") will not work if
800 	 * the nfsnode gets flushed out of the cache. Ufs does not have
801 	 * this problem, because one can identify root inodes by their
802 	 * number == UFS_ROOTINO (2). So, just unlock, but no rele.
803 	 */
804 
805 	nmp->nm_vnode = vp;
806 	if (vp->v_type == VNON)
807 		vp->v_type = VDIR;
808 	vp->v_vflag |= VV_ROOT;
809 	VOP_UNLOCK(vp);
810 	*vpp = vp;
811 
812 	snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++);
813 	nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname);
814 
815 	return (0);
816 bad:
817 	nfs_disconnect(nmp);
818 	rw_destroy(&nmp->nm_writeverflock);
819 	mutex_destroy(&nmp->nm_lock);
820 	cv_destroy(&nmp->nm_rcvcv);
821 	cv_destroy(&nmp->nm_sndcv);
822 	cv_destroy(&nmp->nm_aiocv);
823 	cv_destroy(&nmp->nm_disconcv);
824 	kmem_free(nmp, sizeof(*nmp));
825 	m_freem(nam);
826 	return (error);
827 }
828 
829 /*
830  * unmount system call
831  */
832 int
833 nfs_unmount(struct mount *mp, int mntflags)
834 {
835 	struct nfsmount *nmp = VFSTONFS(mp);
836 	struct vnode *vp;
837 	int error, flags = 0;
838 
839 	if (mntflags & MNT_FORCE) {
840 		mutex_enter(&nmp->nm_lock);
841 		flags |= FORCECLOSE;
842 		nmp->nm_iflag |= NFSMNT_DISMNTFORCE;
843 		mutex_exit(&nmp->nm_lock);
844 
845 	}
846 
847 	/*
848 	 * Goes something like this..
849 	 * - Check for activity on the root vnode (other than ourselves).
850 	 * - Call vflush() to clear out vnodes for this file system,
851 	 *   except for the root vnode.
852 	 * - Decrement reference on the vnode representing remote root.
853 	 * - Close the socket
854 	 * - Free up the data structures
855 	 */
856 	/*
857 	 * We need to decrement the ref. count on the nfsnode representing
858 	 * the remote root.  See comment in mountnfs().
859 	 */
860 	vp = nmp->nm_vnode;
861 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
862 	if (error != 0)
863 		goto err;
864 
865 	if ((mntflags & MNT_FORCE) == 0 && vrefcnt(vp) > 1) {
866 		VOP_UNLOCK(vp);
867 		error = EBUSY;
868 		goto err;
869 	}
870 
871 	error = vflush(mp, vp, flags);
872 	if (error) {
873 		VOP_UNLOCK(vp);
874 		goto err;
875 	}
876 
877 	/*
878 	 * We are now committed to the unmount; mark the mount structure
879 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
880 	 * will go away cleanly.
881 	 */
882 	nmp->nm_iflag |= NFSMNT_DISMNT;
883 
884 	/*
885 	 * No new async I/O will be added, but await for pending
886 	 * ones to drain.
887 	 */
888 	while (nfs_iodbusy(nmp))
889 		kpause("nfsumnt", false, hz, NULL);
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 is one reference count to get rid of here
900 	 * (see comment in mountnfs()).
901 	 */
902 	VOP_UNLOCK(vp);
903 	vgone(vp);
904 	nfs_disconnect(nmp);
905 	m_freem(nmp->nm_nam);
906 
907 	rw_destroy(&nmp->nm_writeverflock);
908 	mutex_destroy(&nmp->nm_lock);
909 	cv_destroy(&nmp->nm_rcvcv);
910 	cv_destroy(&nmp->nm_sndcv);
911 	cv_destroy(&nmp->nm_aiocv);
912 	cv_destroy(&nmp->nm_disconcv);
913 	kmem_free(nmp, sizeof(*nmp));
914 	return (0);
915 
916 err:
917 	if (mntflags & MNT_FORCE) {
918 		mutex_enter(&nmp->nm_lock);
919 		nmp->nm_iflag &= ~NFSMNT_DISMNTFORCE;
920 		mutex_exit(&nmp->nm_lock);
921 	}
922 
923 	return error;
924 }
925 
926 /*
927  * Return root of a filesystem
928  */
929 int
930 nfs_root(struct mount *mp, int lktype, struct vnode **vpp)
931 {
932 	struct vnode *vp;
933 	struct nfsmount *nmp;
934 	int error;
935 
936 	nmp = VFSTONFS(mp);
937 	vp = nmp->nm_vnode;
938 	vref(vp);
939 	error = vn_lock(vp, lktype | LK_RETRY);
940 	if (error != 0) {
941 		vrele(vp);
942 		return error;
943 	}
944 	*vpp = vp;
945 	return (0);
946 }
947 
948 extern int syncprt;
949 
950 static bool
951 nfs_sync_selector(void *cl, struct vnode *vp)
952 {
953 
954 	KASSERT(mutex_owned(vp->v_interlock));
955 
956 	return !LIST_EMPTY(&vp->v_dirtyblkhd) ||
957 	    (vp->v_iflag & VI_ONWORKLST) != 0;
958 }
959 
960 /*
961  * Flush out the buffer cache
962  */
963 /* ARGSUSED */
964 int
965 nfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
966 {
967 	struct vnode *vp;
968 	struct vnode_iterator *marker;
969 	int error, allerror = 0;
970 
971 	/*
972 	 * Force stale buffer cache information to be flushed.
973 	 */
974 	vfs_vnode_iterator_init(mp, &marker);
975 	while ((vp = vfs_vnode_iterator_next(marker, nfs_sync_selector,
976 	    NULL)))
977 	{
978 		error = vn_lock(vp, LK_EXCLUSIVE);
979 		if (error) {
980 			vrele(vp);
981 			continue;
982 		}
983 		error = VOP_FSYNC(vp, cred,
984 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0);
985 		if (error)
986 			allerror = error;
987 		vput(vp);
988 	}
989 	vfs_vnode_iterator_destroy(marker);
990 	return allerror;
991 }
992 
993 /*
994  * NFS flat namespace lookup.
995  * Currently unsupported.
996  */
997 /* ARGSUSED */
998 int
999 nfs_vget(struct mount *mp, ino_t ino, int lktype, struct vnode **vpp)
1000 {
1001 
1002 	return (EOPNOTSUPP);
1003 }
1004 
1005 /*
1006  * Do that sysctl thang...
1007  */
1008 static int
1009 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS)
1010 {
1011 	struct sysctlnode node;
1012 	int val;
1013 	int error;
1014 
1015 	val = nfs_niothreads;
1016 	node = *rnode;
1017 	node.sysctl_data = &val;
1018         error = sysctl_lookup(SYSCTLFN_CALL(&node));
1019 	if (error || newp == NULL)
1020 		return error;
1021 
1022 	return nfs_set_niothreads(val);
1023 }
1024 
1025 SYSCTL_SETUP(nfs_sysctl_init, "nfs sysctl")
1026 {
1027 
1028 	sysctl_createv(clog, 0, NULL, NULL,
1029 		       CTLFLAG_PERMANENT,
1030 		       CTLTYPE_NODE, "nfs",
1031 		       SYSCTL_DESCR("NFS vfs options"),
1032 		       NULL, 0, NULL, 0,
1033 		       CTL_VFS, 2, CTL_EOL);
1034 	/*
1035 	 * XXX the "2" above could be dynamic, thereby eliminating one
1036 	 * more instance of the "number to vfs" mapping problem, but
1037 	 * "2" is the order as taken from sys/mount.h
1038 	 */
1039 
1040 	sysctl_createv(clog, 0, NULL, NULL,
1041 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1042 		       CTLTYPE_STRUCT, "nfsstats",
1043 		       SYSCTL_DESCR("NFS operation statistics"),
1044 		       NULL, 0, &nfsstats, sizeof(nfsstats),
1045 		       CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL);
1046 	sysctl_createv(clog, 0, NULL, NULL,
1047 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1048 		       CTLTYPE_INT, "iothreads",
1049 		       SYSCTL_DESCR("Number of NFS client processes desired"),
1050 		       sysctl_vfs_nfs_iothreads, 0, NULL, 0,
1051 		       CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL);
1052 }
1053 
1054 /* ARGSUSED */
1055 int
1056 nfs_fhtovp(struct mount *mp, struct fid *fid, int lktype, struct vnode **vpp)
1057 {
1058 	size_t fidsize;
1059 	size_t fhsize;
1060 	struct nfsnode *np;
1061 	int error;
1062 	struct vattr va;
1063 
1064 	fidsize = fid->fid_len;
1065 	if (fidsize < sizeof(*fid)) {
1066 		return EINVAL;
1067 	}
1068 	fhsize = fidsize - sizeof(*fid);
1069 	if ((fhsize % NFSX_UNSIGNED) != 0) {
1070 		return EINVAL;
1071 	}
1072 	if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) {
1073 		if (fhsize > NFSX_V3FHMAX || fhsize == 0) {
1074 			return EINVAL;
1075 		}
1076 	} else {
1077 		if (fhsize != NFSX_V2FH) {
1078 			return EINVAL;
1079 		}
1080 	}
1081 	/* XXX lktype ignored */
1082 	error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np);
1083 	if (error) {
1084 		return error;
1085 	}
1086 	*vpp = NFSTOV(np);
1087 	error = VOP_GETATTR(*vpp, &va, kauth_cred_get());
1088 	if (error != 0) {
1089 		vput(*vpp);
1090 		*vpp = NULLVP;
1091 	}
1092 	return error;
1093 }
1094 
1095 /* ARGSUSED */
1096 int
1097 nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize)
1098 {
1099 	struct nfsnode *np;
1100 	struct fid *fid;
1101 	size_t fidsize;
1102 	int error = 0;
1103 
1104 	np = VTONFS(vp);
1105 	fidsize = sizeof(*fid) + np->n_fhsize;
1106 	if (*bufsize < fidsize) {
1107 		error = E2BIG;
1108 	}
1109 	*bufsize = fidsize;
1110 	if (error == 0) {
1111 		struct fid fid_store;
1112 
1113 		fid = &fid_store;
1114 		memset(fid, 0, sizeof(*fid));
1115 		fid->fid_len = fidsize;
1116 		memcpy(buf, fid, sizeof(*fid));
1117 		memcpy(buf->fid_data, np->n_fhp, np->n_fhsize);
1118 	}
1119 	return error;
1120 }
1121 
1122 /*
1123  * Vfs start routine, a no-op.
1124  */
1125 /* ARGSUSED */
1126 int
1127 nfs_start(struct mount *mp, int flags)
1128 {
1129 
1130 	return (0);
1131 }
1132 
1133 /*
1134  * Called once at VFS init to initialize client-specific data structures.
1135  */
1136 void
1137 nfs_vfs_init(void)
1138 {
1139 
1140 	/* Initialize NFS server / client shared data. */
1141 	nfs_init();
1142 	nfs_node_init();
1143 
1144 	/* Initialize the kqueue structures */
1145 	nfs_kqinit();
1146 	/* Initialize the iod structures */
1147 	nfs_iodinit();
1148 
1149 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1150 }
1151 
1152 void
1153 nfs_vfs_done(void)
1154 {
1155 
1156 	nfs_node_done();
1157 	nfs_kqfini();
1158 	nfs_iodfini();
1159 	nfs_fini();
1160 }
1161