xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision d11b170b9000ada93db553723522a63d5deac310)
1 /*	$NetBSD: nfs_vfsops.c,v 1.228 2014/05/24 16:34:04 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 1.228 2014/05/24 16:34:04 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_vget = nfs_vget,
117 	.vfs_fhtovp = nfs_fhtovp,
118 	.vfs_vptofh = nfs_vptofh,
119 	.vfs_init = nfs_vfs_init,
120 	.vfs_done = nfs_vfs_done,
121 	.vfs_mountroot = nfs_mountroot,
122 	.vfs_snapshot = (void *)eopnotsupp,
123 	.vfs_extattrctl = vfs_stdextattrctl,
124 	.vfs_suspendctl = (void *)eopnotsupp,
125 	.vfs_renamelock_enter = genfs_renamelock_enter,
126 	.vfs_renamelock_exit = genfs_renamelock_exit,
127 	.vfs_fsync = (void *)eopnotsupp,
128 	.vfs_opv_descs = nfs_vnodeopv_descs
129 };
130 
131 extern u_int32_t nfs_procids[NFS_NPROCS];
132 extern u_int32_t nfs_prog, nfs_vers;
133 static struct sysctllog *nfs_clog;
134 
135 static int nfs_mount_diskless(struct nfs_dlmount *, const char *,
136     struct mount **, struct vnode **, struct lwp *);
137 static void nfs_sysctl_init(void);
138 static void nfs_sysctl_fini(void);
139 
140 static int
141 nfs_modcmd(modcmd_t cmd, void *arg)
142 {
143 	int error;
144 
145 	switch (cmd) {
146 	case MODULE_CMD_INIT:
147 		error = vfs_attach(&nfs_vfsops);
148 		if (error == 0) {
149 			nfs_sysctl_init();
150 		}
151 		return error;
152 	case MODULE_CMD_FINI:
153 		error = vfs_detach(&nfs_vfsops);
154 		if (error == 0) {
155 			nfs_sysctl_fini();
156 		}
157 		return error;
158 	default:
159 		return ENOTTY;
160 	}
161 }
162 
163 /*
164  * nfs statvfs call
165  */
166 int
167 nfs_statvfs(struct mount *mp, struct statvfs *sbp)
168 {
169 	struct lwp *l = curlwp;
170 	struct vnode *vp;
171 	struct nfs_statfs *sfp;
172 	char *cp;
173 	u_int32_t *tl;
174 	int32_t t1, t2;
175 	char *bpos, *dpos, *cp2;
176 	struct nfsmount *nmp = VFSTONFS(mp);
177 	int error = 0, retattr;
178 #ifdef NFS_V2_ONLY
179 	const int v3 = 0;
180 #else
181 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
182 #endif
183 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
184 	kauth_cred_t cred;
185 	u_quad_t tquad;
186 	struct nfsnode *np;
187 
188 #ifndef nolint
189 	sfp = (struct nfs_statfs *)0;
190 #endif
191 	vp = nmp->nm_vnode;
192 	np = VTONFS(vp);
193 	cred = kauth_cred_alloc();
194 #ifndef NFS_V2_ONLY
195 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
196 		(void)nfs_fsinfo(nmp, vp, cred, l);
197 #endif
198 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
199 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
200 	nfsm_fhtom(np, v3);
201 	nfsm_request(np, NFSPROC_FSSTAT, l, cred);
202 	if (v3)
203 		nfsm_postop_attr(vp, retattr, 0);
204 	if (error) {
205 		if (mrep != NULL) {
206 			if (mrep->m_next != NULL)
207 				printf("nfs_vfsops: nfs_statvfs would lose buffers\n");
208 			m_freem(mrep);
209 		}
210 		goto nfsmout;
211 	}
212 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
213 	sbp->f_flag = nmp->nm_flag;
214 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
215 	if (v3) {
216 		sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE;
217 		tquad = fxdr_hyper(&sfp->sf_tbytes);
218 		sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
219 		tquad = fxdr_hyper(&sfp->sf_fbytes);
220 		sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
221 		tquad = fxdr_hyper(&sfp->sf_abytes);
222 		tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
223 		sbp->f_bresvd = sbp->f_bfree - tquad;
224 		sbp->f_bavail = tquad;
225 		/* Handle older NFS servers returning negative values */
226 		if ((quad_t)sbp->f_bavail < 0)
227 			sbp->f_bavail = 0;
228 		tquad = fxdr_hyper(&sfp->sf_tfiles);
229 		sbp->f_files = tquad;
230 		tquad = fxdr_hyper(&sfp->sf_ffiles);
231 		sbp->f_ffree = tquad;
232 		sbp->f_favail = tquad;
233 		sbp->f_fresvd = 0;
234 		sbp->f_namemax = NFS_MAXNAMLEN;
235 	} else {
236 		sbp->f_bsize = NFS_FABLKSIZE;
237 		sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
238 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
239 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
240 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
241 		sbp->f_fresvd = 0;
242 		sbp->f_files = 0;
243 		sbp->f_ffree = 0;
244 		sbp->f_favail = 0;
245 		sbp->f_fresvd = 0;
246 		sbp->f_namemax = NFS_MAXNAMLEN;
247 	}
248 	copy_statvfs_info(sbp, mp);
249 	nfsm_reqdone;
250 	kauth_cred_free(cred);
251 	return (error);
252 }
253 
254 #ifndef NFS_V2_ONLY
255 /*
256  * nfs version 3 fsinfo rpc call
257  */
258 int
259 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, kauth_cred_t cred, struct lwp *l)
260 {
261 	struct nfsv3_fsinfo *fsp;
262 	char *cp;
263 	int32_t t1, t2;
264 	u_int32_t *tl, pref, xmax;
265 	char *bpos, *dpos, *cp2;
266 	int error = 0, retattr;
267 	struct mbuf *mreq, *mrep, *md, *mb;
268 	u_int64_t maxfsize;
269 	struct nfsnode *np = VTONFS(vp);
270 
271 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
272 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
273 	nfsm_fhtom(np, 1);
274 	nfsm_request(np, NFSPROC_FSINFO, l, cred);
275 	nfsm_postop_attr(vp, retattr, 0);
276 	if (!error) {
277 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
278 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
279 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
280 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
281 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
282 				~(NFS_FABLKSIZE - 1);
283 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
284 		if (xmax < nmp->nm_wsize && xmax > 0) {
285 			nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1);
286 			if (nmp->nm_wsize == 0)
287 				nmp->nm_wsize = xmax;
288 		}
289 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
290 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
291 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
292 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
293 				~(NFS_FABLKSIZE - 1);
294 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
295 		if (xmax < nmp->nm_rsize && xmax > 0) {
296 			nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1);
297 			if (nmp->nm_rsize == 0)
298 				nmp->nm_rsize = xmax;
299 		}
300 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
301 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
302 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
303 				~(NFS_DIRFRAGSIZ - 1);
304 		if (xmax < nmp->nm_readdirsize && xmax > 0) {
305 			nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1);
306 			if (nmp->nm_readdirsize == 0)
307 				nmp->nm_readdirsize = xmax;
308 		}
309 		/* XXX */
310 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
311 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
312 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
313 			nmp->nm_maxfilesize = maxfsize;
314 		nmp->nm_mountp->mnt_fs_bshift =
315 		    ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
316 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
317 	}
318 	nfsm_reqdone;
319 	return (error);
320 }
321 #endif
322 
323 /*
324  * Mount a remote root fs via. NFS.  It goes like this:
325  * - Call nfs_boot_init() to fill in the nfs_diskless struct
326  * - build the rootfs mount point and call mountnfs() to do the rest.
327  */
328 int
329 nfs_mountroot(void)
330 {
331 	struct timespec ts;
332 	struct nfs_diskless *nd;
333 	struct vattr attr;
334 	struct mount *mp;
335 	struct vnode *vp;
336 	struct lwp *l;
337 	long n;
338 	int error;
339 
340 	l = curlwp; /* XXX */
341 
342 	if (device_class(root_device) != DV_IFNET)
343 		return (ENODEV);
344 
345 	/*
346 	 * XXX time must be non-zero when we init the interface or else
347 	 * the arp code will wedge.  [Fixed now in if_ether.c]
348 	 * However, the NFS attribute cache gives false "hits" when the
349 	 * current time < nfs_attrtimeo(nmp, np) so keep this in for now.
350 	 */
351 	if (time_second < NFS_MAXATTRTIMO) {
352 		ts.tv_sec = NFS_MAXATTRTIMO;
353 		ts.tv_nsec = 0;
354 		tc_setclock(&ts);
355 	}
356 
357 	/*
358 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
359 	 * Side effect:  Finds and configures a network interface.
360 	 */
361 	nd = kmem_zalloc(sizeof(*nd), KM_SLEEP);
362 	error = nfs_boot_init(nd, l);
363 	if (error) {
364 		kmem_free(nd, sizeof(*nd));
365 		return (error);
366 	}
367 
368 	/*
369 	 * Create the root mount point.
370 	 */
371 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l);
372 	if (error)
373 		goto out;
374 	printf("root on %s\n", nd->nd_root.ndm_host);
375 
376 	/*
377 	 * Link it into the mount list.
378 	 */
379 	mountlist_append(mp);
380 	rootvp = vp;
381 	mp->mnt_vnodecovered = NULLVP;
382 	vfs_unbusy(mp, false, NULL);
383 
384 	/* Get root attributes (for the time). */
385 	vn_lock(vp, LK_SHARED | LK_RETRY);
386 	error = VOP_GETATTR(vp, &attr, l->l_cred);
387 	VOP_UNLOCK(vp);
388 	if (error)
389 		panic("nfs_mountroot: getattr for root");
390 	n = attr.va_atime.tv_sec;
391 #ifdef	DEBUG
392 	printf("root time: 0x%lx\n", n);
393 #endif
394 	setrootfstime(n);
395 
396 out:
397 	if (error)
398 		nfs_boot_cleanup(nd, l);
399 	kmem_free(nd, sizeof(*nd));
400 	return (error);
401 }
402 
403 /*
404  * Internal version of mount system call for diskless setup.
405  * Separate function because we used to call it twice.
406  * (once for root and once for swap)
407  */
408 static int
409 nfs_mount_diskless(struct nfs_dlmount *ndmntp, const char *mntname, struct mount **mpp, struct vnode **vpp, struct lwp *l)
410 	/* mntname:	 mount point name */
411 {
412 	struct mount *mp;
413 	struct mbuf *m;
414 	int error;
415 
416 	vfs_rootmountalloc(MOUNT_NFS, mntname, &mp);
417 
418 	mp->mnt_op = &nfs_vfsops;
419 
420 	/*
421 	 * Historical practice expects NFS root file systems to
422 	 * be initially mounted r/w.
423 	 */
424 	mp->mnt_flag &= ~MNT_RDONLY;
425 
426 	/* Get mbuf for server sockaddr. */
427 	m = m_get(M_WAIT, MT_SONAME);
428 	if (m == NULL)
429 		panic("nfs_mountroot: mget soname for %s", mntname);
430 	MCLAIM(m, &nfs_mowner);
431 	memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr,
432 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
433 
434 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
435 			 ndmntp->ndm_args.hostname, vpp, l);
436 	if (error) {
437 		vfs_unbusy(mp, false, NULL);
438 		vfs_destroy(mp);
439 		printf("nfs_mountroot: mount %s failed: %d\n",
440 		       mntname, error);
441 	} else
442 		*mpp = mp;
443 
444 	return (error);
445 }
446 
447 void
448 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp, struct lwp *l)
449 {
450 	int s;
451 	int adjsock;
452 	int maxio;
453 
454 	s = splsoftnet();
455 
456 	/*
457 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
458 	 * no sense in that context.
459 	 */
460 	if (argp->sotype == SOCK_STREAM)
461 		argp->flags &= ~NFSMNT_NOCONN;
462 
463 	/*
464 	 * Cookie translation is not needed for v2, silently ignore it.
465 	 */
466 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
467 	    NFSMNT_XLATECOOKIE)
468 		argp->flags &= ~NFSMNT_XLATECOOKIE;
469 
470 	/* Re-bind if rsrvd port requested and wasn't on one */
471 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
472 		  && (argp->flags & NFSMNT_RESVPORT);
473 	/* Also re-bind if we're switching to/from a connected UDP socket */
474 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
475 		    (argp->flags & NFSMNT_NOCONN));
476 
477 	/* Update flags. */
478 	nmp->nm_flag = argp->flags;
479 	splx(s);
480 
481 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
482 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
483 		if (nmp->nm_timeo < NFS_MINTIMEO)
484 			nmp->nm_timeo = NFS_MINTIMEO;
485 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
486 			nmp->nm_timeo = NFS_MAXTIMEO;
487 	}
488 
489 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
490 		nmp->nm_retry = argp->retrans;
491 		if (nmp->nm_retry > NFS_MAXREXMIT)
492 			nmp->nm_retry = NFS_MAXREXMIT;
493 	}
494 
495 #ifndef NFS_V2_ONLY
496 	if (argp->flags & NFSMNT_NFSV3) {
497 		if (argp->sotype == SOCK_DGRAM)
498 			maxio = NFS_MAXDGRAMDATA;
499 		else
500 			maxio = NFS_MAXDATA;
501 	} else
502 #endif
503 		maxio = NFS_V2MAXDATA;
504 
505 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
506 		int osize = nmp->nm_wsize;
507 		nmp->nm_wsize = argp->wsize;
508 		/* Round down to multiple of blocksize */
509 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
510 		if (nmp->nm_wsize <= 0)
511 			nmp->nm_wsize = NFS_FABLKSIZE;
512 		adjsock |= (nmp->nm_wsize != osize);
513 	}
514 	if (nmp->nm_wsize > maxio)
515 		nmp->nm_wsize = maxio;
516 	if (nmp->nm_wsize > MAXBSIZE)
517 		nmp->nm_wsize = MAXBSIZE;
518 
519 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
520 		int osize = nmp->nm_rsize;
521 		nmp->nm_rsize = argp->rsize;
522 		/* Round down to multiple of blocksize */
523 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
524 		if (nmp->nm_rsize <= 0)
525 			nmp->nm_rsize = NFS_FABLKSIZE;
526 		adjsock |= (nmp->nm_rsize != osize);
527 	}
528 	if (nmp->nm_rsize > maxio)
529 		nmp->nm_rsize = maxio;
530 	if (nmp->nm_rsize > MAXBSIZE)
531 		nmp->nm_rsize = MAXBSIZE;
532 
533 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
534 		nmp->nm_readdirsize = argp->readdirsize;
535 		/* Round down to multiple of minimum blocksize */
536 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
537 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
538 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
539 		/* Bigger than buffer size makes no sense */
540 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
541 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
542 	} else if (argp->flags & NFSMNT_RSIZE)
543 		nmp->nm_readdirsize = nmp->nm_rsize;
544 
545 	if (nmp->nm_readdirsize > maxio)
546 		nmp->nm_readdirsize = maxio;
547 
548 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
549 		argp->maxgrouplist <= NFS_MAXGRPS)
550 		nmp->nm_numgrps = argp->maxgrouplist;
551 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
552 		argp->readahead <= NFS_MAXRAHEAD)
553 		nmp->nm_readahead = argp->readahead;
554 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
555 		argp->deadthresh <= NFS_NEVERDEAD)
556 		nmp->nm_deadthresh = argp->deadthresh;
557 
558 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
559 		    (nmp->nm_soproto != argp->proto));
560 	nmp->nm_sotype = argp->sotype;
561 	nmp->nm_soproto = argp->proto;
562 
563 	if (nmp->nm_so && adjsock) {
564 		nfs_safedisconnect(nmp);
565 		if (nmp->nm_sotype == SOCK_DGRAM)
566 			while (nfs_connect(nmp, (struct nfsreq *)0, l)) {
567 				printf("nfs_args: retrying connect\n");
568 				kpause("nfscn3", false, hz, NULL);
569 			}
570 	}
571 }
572 
573 /*
574  * VFS Operations.
575  *
576  * mount system call
577  * It seems a bit dumb to copyinstr() the host and path here and then
578  * memcpy() them in mountnfs(), but I wanted to detect errors before
579  * doing the sockargs() call because sockargs() allocates an mbuf and
580  * an error after that means that I have to release the mbuf.
581  */
582 /* ARGSUSED */
583 int
584 nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
585 {
586 	struct lwp *l = curlwp;
587 	int error;
588 	struct nfs_args *args = data;
589 	struct mbuf *nam;
590 	struct nfsmount *nmp = VFSTONFS(mp);
591 	struct sockaddr *sa;
592 	struct vnode *vp;
593 	char *pth, *hst;
594 	size_t len;
595 	u_char *nfh;
596 
597 	if (args == NULL)
598 		return EINVAL;
599 	if (*data_len < sizeof *args)
600 		return EINVAL;
601 
602 	if (mp->mnt_flag & MNT_GETARGS) {
603 
604 		if (nmp == NULL)
605 			return (EIO);
606 		if (args->addr != NULL) {
607 			sa = mtod(nmp->nm_nam, struct sockaddr *);
608 			error = copyout(sa, args->addr, sa->sa_len);
609 			if (error)
610 				return (error);
611 			args->addrlen = sa->sa_len;
612 		} else
613 			args->addrlen = 0;
614 
615 		args->version = NFS_ARGSVERSION;
616 		args->sotype = nmp->nm_sotype;
617 		args->proto = nmp->nm_soproto;
618 		args->fh = NULL;
619 		args->fhsize = 0;
620 		args->flags = nmp->nm_flag;
621 		args->wsize = nmp->nm_wsize;
622 		args->rsize = nmp->nm_rsize;
623 		args->readdirsize = nmp->nm_readdirsize;
624 		args->timeo = nmp->nm_timeo;
625 		args->retrans = nmp->nm_retry;
626 		args->maxgrouplist = nmp->nm_numgrps;
627 		args->readahead = nmp->nm_readahead;
628 		args->leaseterm = 0; /* dummy */
629 		args->deadthresh = nmp->nm_deadthresh;
630 		args->hostname = NULL;
631 		*data_len = sizeof *args;
632 		return 0;
633 	}
634 
635 	if (args->version != NFS_ARGSVERSION)
636 		return (EPROGMISMATCH);
637 	if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB))
638 		return (EPROGUNAVAIL);
639 #ifdef NFS_V2_ONLY
640 	if (args->flags & NFSMNT_NFSV3)
641 		return (EPROGMISMATCH);
642 #endif
643 	if (mp->mnt_flag & MNT_UPDATE) {
644 		if (nmp == NULL)
645 			return (EIO);
646 		/*
647 		 * When doing an update, we can't change from or to
648 		 * v3, or change cookie translation
649 		 */
650 		args->flags = (args->flags & ~(NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)) |
651 		    (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_XLATECOOKIE));
652 		nfs_decode_args(nmp, args, l);
653 		return (0);
654 	}
655 	if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX)
656 		return (EINVAL);
657 	nfh = malloc(NFSX_V3FHMAX, M_TEMP, M_WAITOK);
658 	error = copyin(args->fh, nfh, args->fhsize);
659 	if (error)
660 		goto free_nfh;
661 	pth = malloc(MNAMELEN, M_TEMP, M_WAITOK);
662 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
663 	if (error)
664 		goto free_pth;
665 	memset(&pth[len], 0, MNAMELEN - len);
666 	hst = malloc(MNAMELEN, M_TEMP, M_WAITOK);
667 	error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len);
668 	if (error)
669 		goto free_hst;
670 	memset(&hst[len], 0, MNAMELEN - len);
671 	/* sockargs() call must be after above copyin() calls */
672 	error = sockargs(&nam, args->addr, args->addrlen, MT_SONAME);
673 	if (error)
674 		goto free_hst;
675 	MCLAIM(nam, &nfs_mowner);
676 	args->fh = nfh;
677 	error = mountnfs(args, mp, nam, pth, hst, &vp, l);
678 
679 free_hst:
680 	free(hst, M_TEMP);
681 free_pth:
682 	free(pth, M_TEMP);
683 free_nfh:
684 	free(nfh, M_TEMP);
685 
686 	return (error);
687 }
688 
689 /*
690  * Common code for mount and mountroot
691  */
692 int
693 mountnfs(struct nfs_args *argp, struct mount *mp, struct mbuf *nam, const char *pth, const char *hst, struct vnode **vpp, struct lwp *l)
694 {
695 	struct nfsmount *nmp;
696 	struct nfsnode *np;
697 	struct vnode *vp;
698 	int error;
699 	struct vattr *attrs;
700 	kauth_cred_t cr;
701 	char iosname[IOSTATNAMELEN];
702 
703 	/*
704 	 * If the number of nfs iothreads to use has never
705 	 * been set, create a reasonable number of them.
706 	 */
707 
708 	if (nfs_niothreads < 0) {
709 		nfs_set_niothreads(NFS_DEFAULT_NIOTHREADS);
710 	}
711 
712 	if (mp->mnt_flag & MNT_UPDATE) {
713 		nmp = VFSTONFS(mp);
714 		/* update paths, file handles, etc, here	XXX */
715 		m_freem(nam);
716 		return (0);
717 	} else {
718 		nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP);
719 		mp->mnt_data = nmp;
720 		TAILQ_INIT(&nmp->nm_uidlruhead);
721 		TAILQ_INIT(&nmp->nm_bufq);
722 		rw_init(&nmp->nm_writeverflock);
723 		mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE);
724 		rw_init(&nmp->nm_rbtlock);
725 		cv_init(&nmp->nm_rcvcv, "nfsrcv");
726 		cv_init(&nmp->nm_sndcv, "nfssnd");
727 		cv_init(&nmp->nm_aiocv, "nfsaio");
728 		cv_init(&nmp->nm_disconcv, "nfsdis");
729 		nfs_rbtinit(nmp);
730 	}
731 	vfs_getnewfsid(mp);
732 	nmp->nm_mountp = mp;
733 
734 #ifndef NFS_V2_ONLY
735 	if ((argp->flags & NFSMNT_NFSV3) == 0)
736 #endif
737 	{
738 		if (argp->fhsize != NFSX_V2FH) {
739 			return EINVAL;
740 		}
741 	}
742 
743 	/*
744 	 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
745 	 * will overwrite this.
746 	 */
747 	nmp->nm_maxfilesize = 0xffffffffLL;
748 
749 	nmp->nm_timeo = NFS_TIMEO;
750 	nmp->nm_retry = NFS_RETRANS;
751 	nmp->nm_wsize = NFS_WSIZE;
752 	nmp->nm_rsize = NFS_RSIZE;
753 	nmp->nm_readdirsize = NFS_READDIRSIZE;
754 	nmp->nm_numgrps = NFS_MAXGRPS;
755 	nmp->nm_readahead = NFS_DEFRAHEAD;
756 	nmp->nm_deadthresh = NFS_DEFDEADTHRESH;
757 	error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE,
758 	    mp->mnt_op->vfs_name, mp, l);
759 	if (error)
760 		goto bad;
761 	nmp->nm_nam = nam;
762 
763 	/* Set up the sockets and per-host congestion */
764 	nmp->nm_sotype = argp->sotype;
765 	nmp->nm_soproto = argp->proto;
766 
767 	nfs_decode_args(nmp, argp, l);
768 
769 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
770 	mp->mnt_dev_bshift = DEV_BSHIFT;
771 
772 	/*
773 	 * For Connection based sockets (TCP,...) defer the connect until
774 	 * the first request, in case the server is not responding.
775 	 */
776 	if (nmp->nm_sotype == SOCK_DGRAM &&
777 		(error = nfs_connect(nmp, (struct nfsreq *)0, l)))
778 		goto bad;
779 
780 	/*
781 	 * This is silly, but it has to be set so that vinifod() works.
782 	 * We do not want to do an nfs_statvfs() here since we can get
783 	 * stuck on a dead server and we are holding a lock on the mount
784 	 * point.
785 	 */
786 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
787 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
788 	if (error)
789 		goto bad;
790 	vp = NFSTOV(np);
791 	attrs = malloc(sizeof(struct vattr), M_TEMP, M_WAITOK);
792 	VOP_GETATTR(vp, attrs, l->l_cred);
793 	if ((nmp->nm_flag & NFSMNT_NFSV3) && (vp->v_type == VDIR)) {
794 		cr = kauth_cred_alloc();
795 		kauth_cred_setuid(cr, attrs->va_uid);
796 		kauth_cred_seteuid(cr, attrs->va_uid);
797 		kauth_cred_setsvuid(cr, attrs->va_uid);
798 		kauth_cred_setgid(cr, attrs->va_gid);
799 		kauth_cred_setegid(cr, attrs->va_gid);
800 		kauth_cred_setsvgid(cr, attrs->va_gid);
801 		nfs_cookieheuristic(vp, &nmp->nm_iflag, l, cr);
802 		kauth_cred_free(cr);
803 	}
804 	free(attrs, M_TEMP);
805 
806 	/*
807 	 * A reference count is needed on the nfsnode representing the
808 	 * remote root.  If this object is not persistent, then backward
809 	 * traversals of the mount point (i.e. "..") will not work if
810 	 * the nfsnode gets flushed out of the cache. Ufs does not have
811 	 * this problem, because one can identify root inodes by their
812 	 * number == UFS_ROOTINO (2). So, just unlock, but no rele.
813 	 */
814 
815 	nmp->nm_vnode = vp;
816 	if (vp->v_type == VNON)
817 		vp->v_type = VDIR;
818 	vp->v_vflag |= VV_ROOT;
819 	VOP_UNLOCK(vp);
820 	*vpp = vp;
821 
822 	snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++);
823 	nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname);
824 
825 	return (0);
826 bad:
827 	nfs_disconnect(nmp);
828 	rw_destroy(&nmp->nm_writeverflock);
829 	rw_destroy(&nmp->nm_rbtlock);
830 	mutex_destroy(&nmp->nm_lock);
831 	cv_destroy(&nmp->nm_rcvcv);
832 	cv_destroy(&nmp->nm_sndcv);
833 	cv_destroy(&nmp->nm_aiocv);
834 	cv_destroy(&nmp->nm_disconcv);
835 	kmem_free(nmp, sizeof(*nmp));
836 	m_freem(nam);
837 	return (error);
838 }
839 
840 /*
841  * unmount system call
842  */
843 int
844 nfs_unmount(struct mount *mp, int mntflags)
845 {
846 	struct nfsmount *nmp;
847 	struct vnode *vp;
848 	int error, flags = 0;
849 
850 	if (mntflags & MNT_FORCE)
851 		flags |= FORCECLOSE;
852 	nmp = VFSTONFS(mp);
853 	/*
854 	 * Goes something like this..
855 	 * - Check for activity on the root vnode (other than ourselves).
856 	 * - Call vflush() to clear out vnodes for this file system,
857 	 *   except for the root vnode.
858 	 * - Decrement reference on the vnode representing remote root.
859 	 * - Close the socket
860 	 * - Free up the data structures
861 	 */
862 	/*
863 	 * We need to decrement the ref. count on the nfsnode representing
864 	 * the remote root.  See comment in mountnfs().
865 	 */
866 	vp = nmp->nm_vnode;
867 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
868 	if (error != 0)
869 		return error;
870 
871 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 1) {
872 		VOP_UNLOCK(vp);
873 		return (EBUSY);
874 	}
875 
876 	error = vflush(mp, vp, flags);
877 	if (error) {
878 		VOP_UNLOCK(vp);
879 		return (error);
880 	}
881 
882 	/*
883 	 * We are now committed to the unmount; mark the mount structure
884 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
885 	 * will go away cleanly.
886 	 */
887 	nmp->nm_iflag |= NFSMNT_DISMNT;
888 
889 	/*
890 	 * Clean up the stats... note that we carefully avoid decrementing
891 	 * nfs_mount_count here for good reason - we may not be unmounting
892 	 * the last thing mounted.
893 	 */
894 	iostat_free(nmp->nm_stats);
895 
896 	/*
897 	 * There is one reference count to get rid of here
898 	 * (see comment in mountnfs()).
899 	 */
900 	VOP_UNLOCK(vp);
901 	vgone(vp);
902 	nfs_disconnect(nmp);
903 	m_freem(nmp->nm_nam);
904 
905 	rw_destroy(&nmp->nm_writeverflock);
906 	rw_destroy(&nmp->nm_rbtlock);
907 	mutex_destroy(&nmp->nm_lock);
908 	cv_destroy(&nmp->nm_rcvcv);
909 	cv_destroy(&nmp->nm_sndcv);
910 	cv_destroy(&nmp->nm_aiocv);
911 	cv_destroy(&nmp->nm_disconcv);
912 	kmem_free(nmp, sizeof(*nmp));
913 	return (0);
914 }
915 
916 /*
917  * Return root of a filesystem
918  */
919 int
920 nfs_root(struct mount *mp, struct vnode **vpp)
921 {
922 	struct vnode *vp;
923 	struct nfsmount *nmp;
924 	int error;
925 
926 	nmp = VFSTONFS(mp);
927 	vp = nmp->nm_vnode;
928 	vref(vp);
929 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
930 	if (error != 0) {
931 		vrele(vp);
932 		return error;
933 	}
934 	*vpp = vp;
935 	return (0);
936 }
937 
938 extern int syncprt;
939 
940 static bool
941 nfs_sync_selector(void *cl, struct vnode *vp)
942 {
943 
944 	return !LIST_EMPTY(&vp->v_dirtyblkhd) || !UVM_OBJ_IS_CLEAN(&vp->v_uobj);
945 }
946 
947 /*
948  * Flush out the buffer cache
949  */
950 /* ARGSUSED */
951 int
952 nfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
953 {
954 	struct vnode *vp;
955 	struct vnode_iterator *marker;
956 	int error, allerror = 0;
957 
958 	/*
959 	 * Force stale buffer cache information to be flushed.
960 	 */
961 	vfs_vnode_iterator_init(mp, &marker);
962 	while ((vp = vfs_vnode_iterator_next(marker, nfs_sync_selector,
963 	    NULL)))
964 	{
965 		error = vn_lock(vp, LK_EXCLUSIVE);
966 		if (error) {
967 			vrele(vp);
968 			continue;
969 		}
970 		error = VOP_FSYNC(vp, cred,
971 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0);
972 		if (error)
973 			allerror = error;
974 		vput(vp);
975 	}
976 	vfs_vnode_iterator_destroy(marker);
977 	return allerror;
978 }
979 
980 /*
981  * NFS flat namespace lookup.
982  * Currently unsupported.
983  */
984 /* ARGSUSED */
985 int
986 nfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
987 {
988 
989 	return (EOPNOTSUPP);
990 }
991 
992 /*
993  * Do that sysctl thang...
994  */
995 static int
996 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS)
997 {
998 	struct sysctlnode node;
999 	int val;
1000 	int error;
1001 
1002 	val = nfs_niothreads;
1003 	node = *rnode;
1004 	node.sysctl_data = &val;
1005         error = sysctl_lookup(SYSCTLFN_CALL(&node));
1006 	if (error || newp == NULL)
1007 		return error;
1008 
1009 	return nfs_set_niothreads(val);
1010 }
1011 
1012 static void
1013 nfs_sysctl_init(void)
1014 {
1015 
1016 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1017 		       CTLFLAG_PERMANENT,
1018 		       CTLTYPE_NODE, "nfs",
1019 		       SYSCTL_DESCR("NFS vfs options"),
1020 		       NULL, 0, NULL, 0,
1021 		       CTL_VFS, 2, CTL_EOL);
1022 	/*
1023 	 * XXX the "2" above could be dynamic, thereby eliminating one
1024 	 * more instance of the "number to vfs" mapping problem, but
1025 	 * "2" is the order as taken from sys/mount.h
1026 	 */
1027 
1028 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1029 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1030 		       CTLTYPE_STRUCT, "nfsstats",
1031 		       SYSCTL_DESCR("NFS operation statistics"),
1032 		       NULL, 0, &nfsstats, sizeof(nfsstats),
1033 		       CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL);
1034 	sysctl_createv(&nfs_clog, 0, NULL, NULL,
1035 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1036 		       CTLTYPE_INT, "iothreads",
1037 		       SYSCTL_DESCR("Number of NFS client processes desired"),
1038 		       sysctl_vfs_nfs_iothreads, 0, NULL, 0,
1039 		       CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL);
1040 }
1041 
1042 static void
1043 nfs_sysctl_fini(void)
1044 {
1045 
1046 	sysctl_teardown(&nfs_clog);
1047 }
1048 
1049 /* ARGSUSED */
1050 int
1051 nfs_fhtovp(struct mount *mp, struct fid *fid, struct vnode **vpp)
1052 {
1053 	size_t fidsize;
1054 	size_t fhsize;
1055 	struct nfsnode *np;
1056 	int error;
1057 	struct vattr va;
1058 
1059 	fidsize = fid->fid_len;
1060 	if (fidsize < sizeof(*fid)) {
1061 		return EINVAL;
1062 	}
1063 	fhsize = fidsize - sizeof(*fid);
1064 	if ((fhsize % NFSX_UNSIGNED) != 0) {
1065 		return EINVAL;
1066 	}
1067 	if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) {
1068 		if (fhsize > NFSX_V3FHMAX || fhsize == 0) {
1069 			return EINVAL;
1070 		}
1071 	} else {
1072 		if (fhsize != NFSX_V2FH) {
1073 			return EINVAL;
1074 		}
1075 	}
1076 	error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np);
1077 	if (error) {
1078 		return error;
1079 	}
1080 	*vpp = NFSTOV(np);
1081 	error = VOP_GETATTR(*vpp, &va, kauth_cred_get());
1082 	if (error != 0) {
1083 		vput(*vpp);
1084 		*vpp = NULLVP;
1085 	}
1086 	return error;
1087 }
1088 
1089 /* ARGSUSED */
1090 int
1091 nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize)
1092 {
1093 	struct nfsnode *np;
1094 	struct fid *fid;
1095 	size_t fidsize;
1096 	int error = 0;
1097 
1098 	np = VTONFS(vp);
1099 	fidsize = sizeof(*fid) + np->n_fhsize;
1100 	if (*bufsize < fidsize) {
1101 		error = E2BIG;
1102 	}
1103 	*bufsize = fidsize;
1104 	if (error == 0) {
1105 		struct fid fid_store;
1106 
1107 		fid = &fid_store;
1108 		memset(fid, 0, sizeof(*fid));
1109 		fid->fid_len = fidsize;
1110 		memcpy(buf, fid, sizeof(*fid));
1111 		memcpy(buf->fid_data, np->n_fhp, np->n_fhsize);
1112 	}
1113 	return error;
1114 }
1115 
1116 /*
1117  * Vfs start routine, a no-op.
1118  */
1119 /* ARGSUSED */
1120 int
1121 nfs_start(struct mount *mp, int flags)
1122 {
1123 
1124 	return (0);
1125 }
1126 
1127 /*
1128  * Called once at VFS init to initialize client-specific data structures.
1129  */
1130 void
1131 nfs_vfs_init(void)
1132 {
1133 
1134 	/* Initialize NFS server / client shared data. */
1135 	nfs_init();
1136 	nfs_node_init();
1137 
1138 	/* Initialize the kqueue structures */
1139 	nfs_kqinit();
1140 	/* Initialize the iod structures */
1141 	nfs_iodinit();
1142 
1143 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1144 }
1145 
1146 void
1147 nfs_vfs_done(void)
1148 {
1149 
1150 	nfs_node_done();
1151 	nfs_kqfini();
1152 	nfs_iodfini();
1153 }
1154