xref: /dragonfly/sys/vfs/nfs/nfs_vfsops.c (revision 25a2db75)
1 /*
2  * Copyright (c) 1989, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)nfs_vfsops.c	8.12 (Berkeley) 5/20/95
37  * $FreeBSD: src/sys/nfs/nfs_vfsops.c,v 1.91.2.7 2003/01/27 20:04:08 dillon Exp $
38  */
39 
40 #include "opt_bootp.h"
41 #include "opt_nfsroot.h"
42 
43 #include <sys/param.h>
44 #include <sys/sockio.h>
45 #include <sys/proc.h>
46 #include <sys/vnode.h>
47 #include <sys/fcntl.h>
48 #include <sys/kernel.h>
49 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/mbuf.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/systm.h>
56 #include <sys/objcache.h>
57 
58 #include <vm/vm.h>
59 #include <vm/vm_extern.h>
60 
61 #include <net/if.h>
62 #include <net/route.h>
63 #include <netinet/in.h>
64 
65 #include <sys/thread2.h>
66 #include <sys/mutex2.h>
67 
68 #include "rpcv2.h"
69 #include "nfsproto.h"
70 #include "nfs.h"
71 #include "nfsmount.h"
72 #include "nfsnode.h"
73 #include "xdr_subs.h"
74 #include "nfsm_subs.h"
75 #include "nfsdiskless.h"
76 #include "nfsmountrpc.h"
77 
78 extern int	nfs_mountroot(struct mount *mp);
79 extern void	bootpc_init(void);
80 
81 extern struct vop_ops nfsv2_vnode_vops;
82 extern struct vop_ops nfsv2_fifo_vops;
83 extern struct vop_ops nfsv2_spec_vops;
84 
85 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
86 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
87 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
88 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
89 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
90 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
91 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
92 
93 struct objcache *nfsmount_objcache;
94 
95 struct nfsstats	nfsstats;
96 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
97 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD, &nfsstats, nfsstats,
98     "Nfs stats structure");
99 static int nfs_ip_paranoia = 1;
100 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW, &nfs_ip_paranoia, 0,
101     "Enable no-connection mode for protocols that support no-connection mode");
102 #ifdef NFS_DEBUG
103 int nfs_debug;
104 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
105 #endif
106 
107 /*
108  * Tunable to determine the Read/Write unit size.  Maximum value
109  * is NFS_MAXDATA.  We also default to NFS_MAXDATA.
110  */
111 static int nfs_io_size = NFS_MAXDATA;
112 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
113 	&nfs_io_size, 0, "NFS optimal I/O unit size");
114 
115 static void	nfs_decode_args (struct nfsmount *nmp,
116 			struct nfs_args *argp);
117 static int	mountnfs (struct nfs_args *,struct mount *,
118 			struct sockaddr *,char *,char *,struct vnode **);
119 static int	nfs_mount ( struct mount *mp, char *path, caddr_t data,
120 			struct ucred *cred);
121 static int	nfs_unmount ( struct mount *mp, int mntflags);
122 static int	nfs_root ( struct mount *mp, struct vnode **vpp);
123 static int	nfs_statfs ( struct mount *mp, struct statfs *sbp,
124 			struct ucred *cred);
125 static int	nfs_statvfs(struct mount *mp, struct statvfs *sbp,
126 				struct ucred *cred);
127 static int	nfs_sync ( struct mount *mp, int waitfor);
128 
129 /*
130  * nfs vfs operations.
131  */
132 static struct vfsops nfs_vfsops = {
133 	.vfs_mount =    	nfs_mount,
134 	.vfs_unmount =  	nfs_unmount,
135 	.vfs_root =     	nfs_root,
136 	.vfs_statfs =    	nfs_statfs,
137 	.vfs_statvfs =   	nfs_statvfs,
138 	.vfs_sync =     	nfs_sync,
139 	.vfs_init =     	nfs_init,
140 	.vfs_uninit =    	nfs_uninit
141 };
142 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
143 MODULE_VERSION(nfs, 1);
144 
145 /*
146  * This structure must be filled in by a primary bootstrap or bootstrap
147  * server for a diskless/dataless machine. It is initialized below just
148  * to ensure that it is allocated to initialized data (.data not .bss).
149  */
150 struct nfs_diskless nfs_diskless = { { { 0 } } };
151 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
152 int nfs_diskless_valid = 0;
153 
154 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
155 	&nfs_diskless_valid, 0,
156 	"NFS diskless params were obtained");
157 
158 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
159 	nfsv3_diskless.root_hostnam, 0,
160 	"Host name for mount point");
161 
162 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
163 	&nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
164 	"%Ssockaddr_in", "Address of root server");
165 
166 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
167 	nfsv3_diskless.swap_hostnam, 0,
168 	"Host name for mount ppoint");
169 
170 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
171 	&nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
172 	"%Ssockaddr_in", "Address of swap server");
173 
174 
175 void nfsargs_ntoh (struct nfs_args *);
176 static int nfs_mountdiskless (char *, char *, int,
177 				  struct sockaddr_in *, struct nfs_args *,
178 				  struct thread *, struct vnode **,
179 				  struct mount **);
180 static void nfs_convert_diskless (void);
181 static void nfs_convert_oargs (struct nfs_args *args,
182 				   struct onfs_args *oargs);
183 
184 /*
185  * Calculate the buffer I/O block size to use.  The maximum V2 block size
186  * is typically 8K, the maximum datagram size is typically 16K, and the
187  * maximum V3 block size is typically 32K.  The buffer cache tends to work
188  * best with 16K blocks but we allow 32K for TCP connections.
189  *
190  * We force the block size to be at least a page for buffer cache efficiency.
191  */
192 static int
193 nfs_iosize(int v3, int sotype)
194 {
195 	int iosize;
196 	int iomax;
197 
198 	if (v3) {
199 		if (sotype == SOCK_STREAM)
200 			iomax = NFS_MAXDATA;
201 		else
202 			iomax = NFS_MAXDGRAMDATA;
203 	} else {
204 		iomax = NFS_V2MAXDATA;
205 	}
206 	if ((iosize = nfs_io_size) > iomax)
207 		iosize = iomax;
208 	if (iosize < PAGE_SIZE)
209 		iosize = PAGE_SIZE;
210 
211 	/*
212 	 * This is an aweful hack but until the buffer cache is rewritten
213 	 * we need it.  The problem is that when you combine write() with
214 	 * mmap() the vm_page->valid bits can become weird looking
215 	 * (e.g. 0xfc).  This occurs because NFS uses piecemeal buffers
216 	 * at the file EOF.  To solve the problem the BIO system needs to
217 	 * be guarenteed that the NFS iosize for regular files will be a
218 	 * multiple of PAGE_SIZE so it can invalidate the whole page
219 	 * rather then just the piece of it owned by the buffer when
220 	 * NFS does vinvalbuf() calls.
221 	 */
222 	if (iosize & PAGE_MASK)
223 		iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
224 	return iosize;
225 }
226 
227 static void
228 nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
229 {
230 	args->version = NFS_ARGSVERSION;
231 	args->addr = oargs->addr;
232 	args->addrlen = oargs->addrlen;
233 	args->sotype = oargs->sotype;
234 	args->proto = oargs->proto;
235 	args->fh = oargs->fh;
236 	args->fhsize = oargs->fhsize;
237 	args->flags = oargs->flags;
238 	args->wsize = oargs->wsize;
239 	args->rsize = oargs->rsize;
240 	args->readdirsize = oargs->readdirsize;
241 	args->timeo = oargs->timeo;
242 	args->retrans = oargs->retrans;
243 	args->maxgrouplist = oargs->maxgrouplist;
244 	args->readahead = oargs->readahead;
245 	args->deadthresh = oargs->deadthresh;
246 	args->hostname = oargs->hostname;
247 }
248 
249 static void
250 nfs_convert_diskless(void)
251 {
252 	int i;
253 
254 	bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
255 		sizeof(struct ifaliasreq));
256 	bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
257 		sizeof(struct sockaddr_in));
258 	nfs_convert_oargs(&nfsv3_diskless.swap_args, &nfs_diskless.swap_args);
259 
260 	/*
261 	 * Copy the NFS handle passed from the diskless code.
262 	 *
263 	 * XXX CURRENTLY DISABLED - bootp passes us a NFSv2 handle which
264 	 * will fail utterly with HAMMER due to limitations with NFSv2
265 	 * directory cookies.
266 	 */
267 	bcopy(nfs_diskless.swap_fh, nfsv3_diskless.swap_fh, NFSX_V2FH);
268 	nfsv3_diskless.swap_fhsize = NFSX_V2FH;
269 	for (i = NFSX_V2FH - 1; i >= 0; --i) {
270 		if (nfs_diskless.swap_fh[i])
271 			break;
272 	}
273 	if (i < 0)
274 		nfsv3_diskless.swap_fhsize = 0;
275 	nfsv3_diskless.swap_fhsize = 0;		/* FORCE DISABLE */
276 
277 	bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
278 		sizeof(struct sockaddr_in));
279 	bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
280 	nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
281 	bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
282 		sizeof(struct ucred));
283 	nfs_convert_oargs(&nfsv3_diskless.root_args, &nfs_diskless.root_args);
284 
285 	/*
286 	 * Copy the NFS handle passed from the diskless code.
287 	 *
288 	 * XXX CURRENTLY DISABLED - bootp passes us a NFSv2 handle which
289 	 * will fail utterly with HAMMER due to limitations with NFSv2
290 	 * directory cookies.
291 	 */
292 	bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V2FH);
293 	nfsv3_diskless.root_fhsize = NFSX_V2FH;
294 	for (i = NFSX_V2FH - 1; i >= 0; --i) {
295 		if (nfs_diskless.root_fh[i])
296 			break;
297 	}
298 	if (i < 0)
299 		nfsv3_diskless.root_fhsize = 0;
300 	nfsv3_diskless.root_fhsize = 0;		/* FORCE DISABLE */
301 
302 	bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
303 		sizeof(struct sockaddr_in));
304 	bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
305 	nfsv3_diskless.root_time = nfs_diskless.root_time;
306 	bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
307 		MAXHOSTNAMELEN);
308 	nfs_diskless_valid = 3;
309 }
310 
311 /*
312  * nfs statfs call
313  */
314 int
315 nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
316 {
317 	struct vnode *vp;
318 	struct nfs_statfs *sfp;
319 	struct nfsmount *nmp = VFSTONFS(mp);
320 	thread_t td = curthread;
321 	int error = 0, retattr;
322 	struct nfsnode *np;
323 	u_quad_t tquad;
324 	struct nfsm_info info;
325 
326 	info.mrep = NULL;
327 	info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
328 
329 	lwkt_gettoken(&nmp->nm_token);
330 
331 #ifndef nolint
332 	sfp = NULL;
333 #endif
334 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
335 	if (error) {
336 		lwkt_reltoken(&nmp->nm_token);
337 		return (error);
338 	}
339 	vp = NFSTOV(np);
340 	/* ignore the passed cred */
341 	cred = crget();
342 	cred->cr_ngroups = 1;
343 	if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
344 		(void)nfs_fsinfo(nmp, vp, td);
345 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
346 	nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
347 	ERROROUT(nfsm_fhtom(&info, vp));
348 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
349 	if (info.v3) {
350 		ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
351 					 NFS_LATTR_NOSHRINK));
352 	}
353 	if (error) {
354 		if (info.mrep != NULL)
355 			m_freem(info.mrep);
356 		goto nfsmout;
357 	}
358 	NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
359 	sbp->f_flags = nmp->nm_flag;
360 
361 	if (info.v3) {
362 		sbp->f_bsize = NFS_FABLKSIZE;
363 		tquad = fxdr_hyper(&sfp->sf_tbytes);
364 		sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
365 		tquad = fxdr_hyper(&sfp->sf_fbytes);
366 		sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
367 		tquad = fxdr_hyper(&sfp->sf_abytes);
368 		sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
369 		sbp->f_files = (fxdr_unsigned(int32_t,
370 		    sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
371 		sbp->f_ffree = (fxdr_unsigned(int32_t,
372 		    sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
373 	} else {
374 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
375 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
376 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
377 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
378 		sbp->f_files = 0;
379 		sbp->f_ffree = 0;
380 	}
381 
382 	/*
383 	 * Some values are pre-set in mnt_stat.  Note in particular f_iosize
384 	 * cannot be changed once the filesystem is mounted as it is used
385 	 * as the basis for BIOs.
386 	 */
387 	if (sbp != &mp->mnt_stat) {
388 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
389 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
390 		sbp->f_iosize = mp->mnt_stat.f_iosize;
391 	}
392 	m_freem(info.mrep);
393 	info.mrep = NULL;
394 nfsmout:
395 	vput(vp);
396 	crfree(cred);
397 	lwkt_reltoken(&nmp->nm_token);
398 	return (error);
399 }
400 
401 static int
402 nfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
403 {
404 	struct vnode *vp;
405 	struct nfs_statfs *sfp;
406 	struct nfsmount *nmp = VFSTONFS(mp);
407 	thread_t td = curthread;
408 	int error = 0, retattr;
409 	struct nfsnode *np;
410 	struct nfsm_info info;
411 
412 	info.mrep = NULL;
413 	info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
414 	lwkt_gettoken(&nmp->nm_token);
415 
416 #ifndef nolint
417 	sfp = NULL;
418 #endif
419 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
420 	if (error) {
421 		lwkt_reltoken(&nmp->nm_token);
422 		return (error);
423 	}
424 	vp = NFSTOV(np);
425 	/* ignore the passed cred */
426 	cred = crget();
427 	cred->cr_ngroups = 1;
428 	if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
429 		(void)nfs_fsinfo(nmp, vp, td);
430 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
431 	nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
432 	ERROROUT(nfsm_fhtom(&info, vp));
433 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
434 	if (info.v3) {
435 		ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
436 					 NFS_LATTR_NOSHRINK));
437 	}
438 	if (error) {
439 		if (info.mrep != NULL)
440 			m_freem(info.mrep);
441 		goto nfsmout;
442 	}
443 	NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
444 	sbp->f_flag = nmp->nm_flag;
445 	sbp->f_owner = nmp->nm_cred->cr_ruid;
446 
447 	if (info.v3) {
448 		sbp->f_bsize = NFS_FABLKSIZE;
449 		sbp->f_frsize = NFS_FABLKSIZE;
450 		sbp->f_blocks = (fxdr_hyper(&sfp->sf_tbytes) /
451 				((u_quad_t)NFS_FABLKSIZE));
452 		sbp->f_bfree = (fxdr_hyper(&sfp->sf_fbytes) /
453 				((u_quad_t)NFS_FABLKSIZE));
454 		sbp->f_bavail = (fxdr_hyper(&sfp->sf_abytes) /
455 				((u_quad_t)NFS_FABLKSIZE));
456 		sbp->f_files = fxdr_hyper(&sfp->sf_tfiles);
457 		sbp->f_ffree = fxdr_hyper(&sfp->sf_ffiles);
458 		sbp->f_favail = fxdr_hyper(&sfp->sf_afiles);
459 	} else {
460 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
461 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
462 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
463 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
464 		sbp->f_files = 0;
465 		sbp->f_ffree = 0;
466 		sbp->f_favail = 0;
467 	}
468 	sbp->f_syncreads = 0;
469 	sbp->f_syncwrites = 0;
470 	sbp->f_asyncreads = 0;
471 	sbp->f_asyncwrites = 0;
472 	sbp->f_type = mp->mnt_vfc->vfc_typenum;
473 
474 	m_freem(info.mrep);
475 	info.mrep = NULL;
476 nfsmout:
477 	vput(vp);
478 	crfree(cred);
479 	lwkt_reltoken(&nmp->nm_token);
480 	return (error);
481 }
482 
483 /*
484  * nfs version 3 fsinfo rpc call
485  */
486 int
487 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
488 {
489 	struct nfsv3_fsinfo *fsp;
490 	u_int32_t pref, max;
491 	int error = 0, retattr;
492 	u_int64_t maxfsize;
493 	struct nfsm_info info;
494 
495 	info.v3 = 1;
496 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
497 	nfsm_reqhead(&info, vp, NFSPROC_FSINFO, NFSX_FH(1));
498 	ERROROUT(nfsm_fhtom(&info, vp));
499 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSINFO, td,
500 				nfs_vpcred(vp, ND_READ), &error));
501 	ERROROUT(nfsm_postop_attr(&info, vp, &retattr, NFS_LATTR_NOSHRINK));
502 	if (error == 0) {
503 		NULLOUT(fsp = nfsm_dissect(&info, NFSX_V3FSINFO));
504 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
505 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
506 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
507 				~(NFS_FABLKSIZE - 1);
508 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
509 		if (max < nmp->nm_wsize && max > 0) {
510 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
511 			if (nmp->nm_wsize == 0)
512 				nmp->nm_wsize = max;
513 		}
514 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
515 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
516 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
517 				~(NFS_FABLKSIZE - 1);
518 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
519 		if (max < nmp->nm_rsize && max > 0) {
520 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
521 			if (nmp->nm_rsize == 0)
522 				nmp->nm_rsize = max;
523 		}
524 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
525 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
526 			nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
527 				~(NFS_DIRBLKSIZ - 1);
528 		if (max < nmp->nm_readdirsize && max > 0) {
529 			nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
530 			if (nmp->nm_readdirsize == 0)
531 				nmp->nm_readdirsize = max;
532 		}
533 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
534 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
535 			nmp->nm_maxfilesize = maxfsize;
536 		nmp->nm_state |= NFSSTA_GOTFSINFO;
537 
538 		/*
539 		 * Use the smaller of rsize/wsize for the biosize.
540 		 */
541 		if (nmp->nm_rsize < nmp->nm_wsize)
542 			nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_rsize;
543 		else
544 			nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_wsize;
545 	}
546 	m_freem(info.mrep);
547 	info.mrep = NULL;
548 nfsmout:
549 	return (error);
550 }
551 
552 /*
553  * Mount a remote root fs via. nfs. This depends on the info in the
554  * nfs_diskless structure that has been filled in properly by some primary
555  * bootstrap.
556  * It goes something like this:
557  * - do enough of "ifconfig" by calling ifioctl() so that the system
558  *   can talk to the server
559  * - If nfs_diskless.mygateway is filled in, use that address as
560  *   a default gateway.
561  * - build the rootfs mount point and call mountnfs() to do the rest.
562  */
563 int
564 nfs_mountroot(struct mount *mp)
565 {
566 	struct mount  *swap_mp;
567 	struct nfsv3_diskless *nd = &nfsv3_diskless;
568 	struct socket *so;
569 	struct vnode *vp;
570 	struct thread *td = curthread;		/* XXX */
571 	int error, i;
572 	u_long l;
573 	char buf[128];
574 
575 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
576 	bootpc_init();		/* use bootp to get nfs_diskless filled in */
577 #endif
578 
579 	/*
580 	 * XXX time must be non-zero when we init the interface or else
581 	 * the arp code will wedge...
582 	 */
583 	while (mycpu->gd_time_seconds == 0)
584 		tsleep(mycpu, 0, "arpkludge", 10);
585 
586 	/*
587 	 * The boot code may have passed us a diskless structure.
588 	 */
589 	kprintf("DISKLESS %d\n", nfs_diskless_valid);
590 	if (nfs_diskless_valid == 1)
591 		nfs_convert_diskless();
592 
593 	/*
594 	 * NFSv3 is required.
595 	 */
596 	nd->root_args.flags |= NFSMNT_NFSV3 | NFSMNT_RDIRPLUS;
597 	nd->swap_args.flags |= NFSMNT_NFSV3;
598 
599 #define SINP(sockaddr)	((struct sockaddr_in *)(sockaddr))
600 	kprintf("nfs_mountroot: interface %s ip %s",
601 		nd->myif.ifra_name,
602 		inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
603 	kprintf(" bcast %s",
604 		inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
605 	kprintf(" mask %s\n",
606 		inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
607 #undef SINP
608 
609 	/*
610 	 * XXX splnet, so networks will receive...
611 	 */
612 	crit_enter();
613 
614 	/*
615 	 * BOOTP does not necessarily have to be compiled into the kernel
616 	 * for an NFS root to work.  If we inherited the network
617 	 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
618 	 * out our interface for us and all we need to do is ifconfig the
619 	 * interface.  We only do this if the interface has not already been
620 	 * ifconfig'd by e.g. BOOTP.
621 	 */
622 	error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
623 	if (error) {
624 		panic("nfs_mountroot: socreate(%04x): %d",
625 			nd->myif.ifra_addr.sa_family, error);
626 	}
627 
628 	error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
629 	if (error)
630 		panic("nfs_mountroot: SIOCAIFADDR: %d", error);
631 
632 	soclose(so, FNONBLOCK);
633 
634 	/*
635 	 * If the gateway field is filled in, set it as the default route.
636 	 */
637 	if (nd->mygateway.sin_len != 0) {
638 		struct sockaddr_in mask, sin;
639 
640 		bzero((caddr_t)&mask, sizeof(mask));
641 		sin = mask;
642 		sin.sin_family = AF_INET;
643 		sin.sin_len = sizeof(sin);
644 		kprintf("nfs_mountroot: gateway %s\n",
645 			inet_ntoa(nd->mygateway.sin_addr));
646 		error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
647 					(struct sockaddr *)&nd->mygateway,
648 					(struct sockaddr *)&mask,
649 					RTF_UP | RTF_GATEWAY);
650 		if (error)
651 			kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
652 	}
653 
654 	/*
655 	 * Create the rootfs mount point.
656 	 */
657 	nd->root_args.fh = nd->root_fh;
658 	nd->root_args.fhsize = nd->root_fhsize;
659 	l = ntohl(nd->root_saddr.sin_addr.s_addr);
660 	ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
661 		(l >> 24) & 0xff, (l >> 16) & 0xff,
662 		(l >>  8) & 0xff, (l >>  0) & 0xff,nd->root_hostnam);
663 	kprintf("NFS_ROOT: %s\n",buf);
664 	error = nfs_mountdiskless(buf, "/", MNT_RDONLY, &nd->root_saddr,
665 				  &nd->root_args, td, &vp, &mp);
666 	if (error) {
667 		mp->mnt_vfc->vfc_refcount--;
668 		crit_exit();
669 		return (error);
670 	}
671 
672 	swap_mp = NULL;
673 	if (nd->swap_nblks) {
674 
675 		/* Convert to DEV_BSIZE instead of Kilobyte */
676 		nd->swap_nblks *= 2;
677 
678 		/*
679 		 * Create a fake mount point just for the swap vnode so that the
680 		 * swap file can be on a different server from the rootfs.
681 		 */
682 		nd->swap_args.fh = nd->swap_fh;
683 		nd->swap_args.fhsize = nd->swap_fhsize;
684 		l = ntohl(nd->swap_saddr.sin_addr.s_addr);
685 		ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
686 			(l >> 24) & 0xff, (l >> 16) & 0xff,
687 			(l >>  8) & 0xff, (l >>  0) & 0xff,nd->swap_hostnam);
688 		kprintf("NFS SWAP: %s\n",buf);
689 		error = nfs_mountdiskless(buf, "/swap", 0, &nd->swap_saddr,
690 					  &nd->swap_args, td, &vp, &swap_mp);
691 		if (error) {
692 			crit_exit();
693 			return (error);
694 		}
695 		vfs_unbusy(swap_mp);
696 
697 		VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
698 				nd->swap_nblks * DEV_BSIZE ;
699 
700 		/*
701 		 * Since the swap file is not the root dir of a file system,
702 		 * hack it to a regular file.
703 		 */
704 		vclrflags(vp, VROOT);
705 		vref(vp);
706 		nfs_setvtype(vp, VREG);
707 		swaponvp(td, vp, nd->swap_nblks);
708 	}
709 
710 	mp->mnt_flag |= MNT_ROOTFS;
711 
712 	/*
713 	 * This is not really an nfs issue, but it is much easier to
714 	 * set hostname here and then let the "/etc/rc.xxx" files
715 	 * mount the right /var based upon its preset value.
716 	 */
717 	bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
718 	hostname[MAXHOSTNAMELEN - 1] = '\0';
719 	for (i = 0; i < MAXHOSTNAMELEN; i++)
720 		if (hostname[i] == '\0')
721 			break;
722 	inittodr(ntohl(nd->root_time));
723 	crit_exit();
724 	return (0);
725 }
726 
727 /*
728  * Internal version of mount system call for diskless setup.
729  */
730 static int
731 nfs_mountdiskless(char *path, char *which, int mountflag,
732 	struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
733 	struct vnode **vpp, struct mount **mpp)
734 {
735 	struct mount *mp;
736 	struct sockaddr *nam;
737 	int didalloc = 0;
738 	int error;
739 
740 	mp = *mpp;
741 
742 	if (mp == NULL) {
743 		if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
744 			kprintf("nfs_mountroot: NFS not configured");
745 			return (error);
746 		}
747 		didalloc = 1;
748 	}
749 	mp->mnt_kern_flag = 0;
750 	mp->mnt_flag = mountflag;
751 	nam = dup_sockaddr((struct sockaddr *)sin);
752 
753 #if defined(BOOTP) || defined(NFS_ROOT)
754 	if (args->fhsize == 0) {
755 		char *xpath = path;
756 
757 		kprintf("NFS_ROOT: No FH passed from loader, attempting "
758 			"mount rpc...");
759 		while (*xpath && *xpath != ':')
760 			++xpath;
761 		if (*xpath)
762 			++xpath;
763 		args->fhsize = 0;
764 		error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td);
765 		if (error) {
766 			kprintf("failed error %d.\n", error);
767 			goto haderror;
768 		}
769 		kprintf("success!\n");
770 	}
771 #endif
772 
773 	if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
774 #if defined(BOOTP) || defined(NFS_ROOT)
775 haderror:
776 #endif
777 		kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error);
778 		mp->mnt_vfc->vfc_refcount--;
779 		if (didalloc)
780 			kfree(mp, M_MOUNT);
781 		kfree(nam, M_SONAME);
782 		return (error);
783 	}
784 	*mpp = mp;
785 	return (0);
786 }
787 
788 static void
789 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp)
790 {
791 	int adjsock;
792 	int maxio;
793 
794 	crit_enter();
795 	/*
796 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
797 	 * no sense in that context.
798 	 */
799 	if (nmp->nm_sotype == SOCK_STREAM) {
800 		nmp->nm_flag &= ~NFSMNT_NOCONN;
801 		argp->flags &= ~NFSMNT_NOCONN;
802 	}
803 
804 	/*
805 	 * readdirplus is NFSv3 only.
806 	 */
807 	if ((argp->flags & NFSMNT_NFSV3) == 0) {
808 		nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
809 		argp->flags &= ~NFSMNT_RDIRPLUS;
810 	}
811 
812 	/*
813 	 * Re-bind if rsrvd port flag has changed
814 	 */
815 	adjsock = (nmp->nm_flag & NFSMNT_RESVPORT) !=
816 		  (argp->flags & NFSMNT_RESVPORT);
817 
818 	/* Update flags atomically.  Don't change the lock bits. */
819 	nmp->nm_flag = argp->flags | nmp->nm_flag;
820 	crit_exit();
821 
822 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
823 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
824 		if (nmp->nm_timeo < NFS_MINTIMEO)
825 			nmp->nm_timeo = NFS_MINTIMEO;
826 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
827 			nmp->nm_timeo = NFS_MAXTIMEO;
828 	}
829 
830 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
831 		nmp->nm_retry = argp->retrans;
832 		if (nmp->nm_retry > NFS_MAXREXMIT)
833 			nmp->nm_retry = NFS_MAXREXMIT;
834 	}
835 
836 	/*
837 	 * These parameters effect the buffer cache and cannot be changed
838 	 * once we've successfully mounted.
839 	 */
840 	if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
841 		maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, nmp->nm_sotype);
842 
843 		if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
844 			nmp->nm_wsize = argp->wsize;
845 			/* Round down to multiple of blocksize */
846 			nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
847 			if (nmp->nm_wsize <= 0)
848 				nmp->nm_wsize = NFS_FABLKSIZE;
849 		}
850 		if (nmp->nm_wsize > maxio)
851 			nmp->nm_wsize = maxio;
852 		if (nmp->nm_wsize > MAXBSIZE)
853 			nmp->nm_wsize = MAXBSIZE;
854 
855 		if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
856 			nmp->nm_rsize = argp->rsize;
857 			/* Round down to multiple of blocksize */
858 			nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
859 			if (nmp->nm_rsize <= 0)
860 				nmp->nm_rsize = NFS_FABLKSIZE;
861 		}
862 		if (nmp->nm_rsize > maxio)
863 			nmp->nm_rsize = maxio;
864 		if (nmp->nm_rsize > MAXBSIZE)
865 			nmp->nm_rsize = MAXBSIZE;
866 
867 		if ((argp->flags & NFSMNT_READDIRSIZE) &&
868 		    argp->readdirsize > 0) {
869 			nmp->nm_readdirsize = argp->readdirsize;
870 		}
871 		if (nmp->nm_readdirsize > maxio)
872 			nmp->nm_readdirsize = maxio;
873 		if (nmp->nm_readdirsize > nmp->nm_rsize)
874 			nmp->nm_readdirsize = nmp->nm_rsize;
875 	}
876 
877 	if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
878 		nmp->nm_acregmin = argp->acregmin;
879 	else
880 		nmp->nm_acregmin = NFS_MINATTRTIMO;
881 	if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
882 		nmp->nm_acregmax = argp->acregmax;
883 	else
884 		nmp->nm_acregmax = NFS_MAXATTRTIMO;
885 	if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
886 		nmp->nm_acdirmin = argp->acdirmin;
887 	else
888 		nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
889 	if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
890 		nmp->nm_acdirmax = argp->acdirmax;
891 	else
892 		nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
893 	if (nmp->nm_acdirmin > nmp->nm_acdirmax)
894 		nmp->nm_acdirmin = nmp->nm_acdirmax;
895 	if (nmp->nm_acregmin > nmp->nm_acregmax)
896 		nmp->nm_acregmin = nmp->nm_acregmax;
897 
898 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
899 		if (argp->maxgrouplist <= NFS_MAXGRPS)
900 			nmp->nm_numgrps = argp->maxgrouplist;
901 		else
902 			nmp->nm_numgrps = NFS_MAXGRPS;
903 	}
904 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
905 		if (argp->readahead <= NFS_MAXRAHEAD)
906 			nmp->nm_readahead = argp->readahead;
907 		else
908 			nmp->nm_readahead = NFS_MAXRAHEAD;
909 	}
910 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
911 		if (argp->deadthresh <= NFS_NEVERDEAD)
912 			nmp->nm_deadthresh = argp->deadthresh;
913 		else
914 			nmp->nm_deadthresh = NFS_NEVERDEAD;
915 	}
916 
917 	if (nmp->nm_so && adjsock) {
918 		nfs_safedisconnect(nmp);
919 		if (nmp->nm_sotype == SOCK_DGRAM)
920 			while (nfs_connect(nmp, NULL)) {
921 				kprintf("nfs_args: retrying connect\n");
922 				(void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
923 			}
924 	}
925 }
926 
927 /*
928  * VFS Operations.
929  *
930  * mount system call
931  * It seems a bit dumb to copyinstr() the host and path here and then
932  * bcopy() them in mountnfs(), but I wanted to detect errors before
933  * doing the sockargs() call because sockargs() allocates an mbuf and
934  * an error after that means that I have to release the mbuf.
935  */
936 /* ARGSUSED */
937 static int
938 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
939 {
940 	int error;
941 	struct nfs_args args;
942 	struct sockaddr *nam;
943 	struct vnode *vp;
944 	char pth[MNAMELEN], hst[MNAMELEN];
945 	size_t len;
946 	u_char nfh[NFSX_V3FHMAX];
947 
948 	if (path == NULL) {
949 		nfs_mountroot(mp);
950 		return (0);
951 	}
952 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
953 	if (error)
954 		return (error);
955 	if (args.version != NFS_ARGSVERSION) {
956 #ifdef COMPAT_PRELITE2
957 		/*
958 		 * If the argument version is unknown, then assume the
959 		 * caller is a pre-lite2 4.4BSD client and convert its
960 		 * arguments.
961 		 */
962 		struct onfs_args oargs;
963 		error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
964 		if (error)
965 			return (error);
966 		nfs_convert_oargs(&args,&oargs);
967 #else /* !COMPAT_PRELITE2 */
968 		return (EPROGMISMATCH);
969 #endif /* COMPAT_PRELITE2 */
970 	}
971 	if (mp->mnt_flag & MNT_UPDATE) {
972 		struct nfsmount *nmp = VFSTONFS(mp);
973 
974 		if (nmp == NULL)
975 			return (EIO);
976 		/*
977 		 * When doing an update, we can't change from or to
978 		 * v3, or change cookie translation, or rsize or wsize.
979 		 */
980 		args.flags &= ~(NFSMNT_NFSV3 | NFSMNT_RSIZE | NFSMNT_WSIZE);
981 		args.flags |= nmp->nm_flag & (NFSMNT_NFSV3);
982 		nfs_decode_args(nmp, &args);
983 		return (0);
984 	}
985 
986 	/*
987 	 * Make the nfs_ip_paranoia sysctl serve as the default connection
988 	 * or no-connection mode for those protocols that support
989 	 * no-connection mode (the flag will be cleared later for protocols
990 	 * that do not support no-connection mode).  This will allow a client
991 	 * to receive replies from a different IP then the request was
992 	 * sent to.  Note: default value for nfs_ip_paranoia is 1 (paranoid),
993 	 * not 0.
994 	 */
995 	if (nfs_ip_paranoia == 0)
996 		args.flags |= NFSMNT_NOCONN;
997 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
998 		return (EINVAL);
999 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
1000 	if (error)
1001 		return (error);
1002 	error = copyinstr(path, pth, MNAMELEN-1, &len);
1003 	if (error)
1004 		return (error);
1005 	bzero(&pth[len], MNAMELEN - len);
1006 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
1007 	if (error)
1008 		return (error);
1009 	bzero(&hst[len], MNAMELEN - len);
1010 	/* sockargs() call must be after above copyin() calls */
1011 	error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
1012 	if (error)
1013 		return (error);
1014 	args.fh = nfh;
1015 	error = mountnfs(&args, mp, nam, pth, hst, &vp);
1016 	return (error);
1017 }
1018 
1019 /*
1020  * Common code for mount and mountroot
1021  */
1022 static int
1023 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
1024 	char *pth, char *hst, struct vnode **vpp)
1025 {
1026 	struct nfsmount *nmp;
1027 	struct nfsnode *np;
1028 	int error;
1029 	int rxcpu;
1030 	int txcpu;
1031 
1032 	if (mp->mnt_flag & MNT_UPDATE) {
1033 		nmp = VFSTONFS(mp);
1034 		/* update paths, file handles, etc, here	XXX */
1035 		kfree(nam, M_SONAME);
1036 		return (0);
1037 	} else {
1038 		nmp = objcache_get(nfsmount_objcache, M_WAITOK);
1039 		bzero((caddr_t)nmp, sizeof (struct nfsmount));
1040 		mtx_init(&nmp->nm_rxlock);
1041 		mtx_init(&nmp->nm_txlock);
1042 		TAILQ_INIT(&nmp->nm_uidlruhead);
1043 		TAILQ_INIT(&nmp->nm_bioq);
1044 		TAILQ_INIT(&nmp->nm_reqq);
1045 		TAILQ_INIT(&nmp->nm_reqtxq);
1046 		TAILQ_INIT(&nmp->nm_reqrxq);
1047 		mp->mnt_data = (qaddr_t)nmp;
1048 		lwkt_token_init(&nmp->nm_token, "nfs_token");
1049 	}
1050 	vfs_getnewfsid(mp);
1051 	nmp->nm_mountp = mp;
1052 	mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
1053 	mp->mnt_kern_flag |= MNTK_THR_SYNC;
1054 
1055 	lwkt_gettoken(&nmp->nm_token);
1056 
1057 	/*
1058 	 * V2 can only handle 32 bit filesizes.  A 4GB-1 limit may be too
1059 	 * high, depending on whether we end up with negative offsets in
1060 	 * the client or server somewhere.  2GB-1 may be safer.
1061 	 *
1062 	 * For V3, nfs_fsinfo will adjust this as necessary.  Assume maximum
1063 	 * that we can handle until we find out otherwise.  Note that seek
1064 	 * offsets are signed.
1065 	 */
1066 	if ((argp->flags & NFSMNT_NFSV3) == 0)
1067 		nmp->nm_maxfilesize = 0xffffffffLL;
1068 	else
1069 		nmp->nm_maxfilesize = 0x7fffffffffffffffLL;
1070 
1071 	nmp->nm_timeo = NFS_TIMEO;
1072 	nmp->nm_retry = NFS_RETRANS;
1073 	nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
1074 	nmp->nm_rsize = nmp->nm_wsize;
1075 	nmp->nm_readdirsize = NFS_READDIRSIZE;
1076 	nmp->nm_numgrps = NFS_MAXGRPS;
1077 	nmp->nm_readahead = NFS_DEFRAHEAD;
1078 	nmp->nm_deadthresh = NFS_DEADTHRESH;
1079 	nmp->nm_fhsize = argp->fhsize;
1080 	bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
1081 	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
1082 	nmp->nm_nam = nam;
1083 	/* Set up the sockets and per-host congestion */
1084 	nmp->nm_sotype = argp->sotype;
1085 	nmp->nm_soproto = argp->proto;
1086 	nmp->nm_cred = crhold(proc0.p_ucred);
1087 
1088 	nfs_decode_args(nmp, argp);
1089 
1090 	/*
1091 	 * For Connection based sockets (TCP,...) defer the connect until
1092 	 * the first request, in case the server is not responding.
1093 	 */
1094 	if (nmp->nm_sotype == SOCK_DGRAM &&
1095 		(error = nfs_connect(nmp, NULL)))
1096 		goto bad;
1097 
1098 	/*
1099 	 * This is silly, but it has to be set so that vinifod() works.
1100 	 * We do not want to do an nfs_statfs() here since we can get
1101 	 * stuck on a dead server and we are holding a lock on the mount
1102 	 * point.
1103 	 */
1104 	mp->mnt_stat.f_iosize =
1105 		nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
1106 
1107 	/*
1108 	 * Install vop_ops for our vnops
1109 	 */
1110 	vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
1111 	vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
1112 	vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
1113 
1114 	/*
1115 	 * A reference count is needed on the nfsnode representing the
1116 	 * remote root.  If this object is not persistent, then backward
1117 	 * traversals of the mount point (i.e. "..") will not work if
1118 	 * the nfsnode gets flushed out of the cache. Ufs does not have
1119 	 * this problem, because one can identify root inodes by their
1120 	 * number == ROOTINO (2).
1121 	 */
1122 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
1123 	if (error)
1124 		goto bad;
1125 	*vpp = NFSTOV(np);
1126 
1127 	/*
1128 	 * Retrieval of mountpoint attributes is delayed until nfs_rot
1129 	 * or nfs_statfs are first called.  This will happen either when
1130 	 * we first traverse the mount point or if somebody does a df(1).
1131 	 *
1132 	 * NFSSTA_GOTFSINFO is used to flag if we have successfully
1133 	 * retrieved mountpoint attributes.  In the case of NFSv3 we
1134 	 * also flag static fsinfo.
1135 	 */
1136 	if (*vpp != NULL)
1137 		(*vpp)->v_type = VNON;
1138 
1139 	/*
1140 	 * Lose the lock but keep the ref.
1141 	 */
1142 	vn_unlock(*vpp);
1143 	lwkt_gettoken(&nfs_token);
1144 	TAILQ_INSERT_TAIL(&nfs_mountq, nmp, nm_entry);
1145 	lwkt_reltoken(&nfs_token);
1146 
1147 	switch(ncpus) {
1148 	case 0:
1149 	case 1:
1150 		rxcpu = 0;
1151 		txcpu = 0;
1152 		break;
1153 	case 2:
1154 		rxcpu = 0;
1155 		txcpu = 1;
1156 		break;
1157 	default:
1158 		rxcpu = -1;
1159 		txcpu = -1;
1160 		break;
1161 	}
1162 
1163 	/*
1164 	 * Start the reader and writer threads.
1165 	 */
1166 	lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread,
1167 		    NULL, 0, rxcpu, "nfsiod_rx");
1168 	lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread,
1169 		    NULL, 0, txcpu, "nfsiod_tx");
1170 	lwkt_reltoken(&nmp->nm_token);
1171 	return (0);
1172 bad:
1173 	nfs_disconnect(nmp);
1174 	lwkt_reltoken(&nmp->nm_token);
1175 	nfs_free_mount(nmp);
1176 	return (error);
1177 }
1178 
1179 /*
1180  * unmount system call
1181  */
1182 static int
1183 nfs_unmount(struct mount *mp, int mntflags)
1184 {
1185 	struct nfsmount *nmp;
1186 	int error, flags = 0;
1187 
1188 	nmp = VFSTONFS(mp);
1189 	lwkt_gettoken(&nmp->nm_token);
1190 	if (mntflags & MNT_FORCE) {
1191 		flags |= FORCECLOSE;
1192 		nmp->nm_flag |= NFSMNT_FORCE;
1193 	}
1194 
1195 	/*
1196 	 * Goes something like this..
1197 	 * - Call vflush() to clear out vnodes for this file system
1198 	 * - Close the socket
1199 	 * - Free up the data structures
1200 	 */
1201 	/* In the forced case, cancel any outstanding requests. */
1202 	if (flags & FORCECLOSE) {
1203 		error = nfs_nmcancelreqs(nmp);
1204 		if (error) {
1205 			kprintf("NFS: %s: Unable to cancel all requests\n",
1206 				mp->mnt_stat.f_mntfromname);
1207 			/* continue anyway */
1208 		}
1209 	}
1210 
1211 	/*
1212 	 * Must handshake with nfs_clientd() if it is active. XXX
1213 	 */
1214 	nmp->nm_state |= NFSSTA_DISMINPROG;
1215 
1216 	/*
1217 	 * We hold 1 extra ref on the root vnode; see comment in mountnfs().
1218 	 *
1219 	 * If this doesn't work and we are doing a forced unmount we continue
1220 	 * anyway.
1221 	 */
1222 	error = vflush(mp, 1, flags);
1223 	if (error) {
1224 		nmp->nm_state &= ~NFSSTA_DISMINPROG;
1225 		if ((flags & FORCECLOSE) == 0) {
1226 			lwkt_reltoken(&nmp->nm_token);
1227 			return (error);
1228 		}
1229 	}
1230 
1231 	/*
1232 	 * We are now committed to the unmount.
1233 	 * For NQNFS, let the server daemon free the nfsmount structure.
1234 	 */
1235 	if (nmp->nm_flag & NFSMNT_KERB)
1236 		nmp->nm_state |= NFSSTA_DISMNT;
1237 	nfssvc_iod_stop1(nmp);
1238 	nfs_disconnect(nmp);
1239 	nfssvc_iod_stop2(nmp);
1240 
1241 	lwkt_gettoken(&nfs_token);
1242 	TAILQ_REMOVE(&nfs_mountq, nmp, nm_entry);
1243 	lwkt_reltoken(&nfs_token);
1244 
1245 	lwkt_reltoken(&nmp->nm_token);
1246 
1247 	if ((nmp->nm_flag & NFSMNT_KERB) == 0) {
1248 		nfs_free_mount(nmp);
1249 	}
1250 	return (0);
1251 }
1252 
1253 void
1254 nfs_free_mount(struct nfsmount *nmp)
1255 {
1256 	if (nmp->nm_cred)  {
1257 		crfree(nmp->nm_cred);
1258 		nmp->nm_cred = NULL;
1259 	}
1260 	if (nmp->nm_nam) {
1261 		kfree(nmp->nm_nam, M_SONAME);
1262 		nmp->nm_nam = NULL;
1263 	}
1264 	objcache_put(nfsmount_objcache, nmp);
1265 }
1266 
1267 /*
1268  * Return root of a filesystem
1269  */
1270 static int
1271 nfs_root(struct mount *mp, struct vnode **vpp)
1272 {
1273 	struct vnode *vp;
1274 	struct nfsmount *nmp;
1275 	struct vattr attrs;
1276 	struct nfsnode *np;
1277 	int error;
1278 
1279 	nmp = VFSTONFS(mp);
1280 	lwkt_gettoken(&nmp->nm_token);
1281 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
1282 	if (error) {
1283 		lwkt_reltoken(&nmp->nm_token);
1284 		return (error);
1285 	}
1286 	vp = NFSTOV(np);
1287 
1288 	/*
1289 	 * Get transfer parameters and root vnode attributes
1290 	 *
1291 	 * NOTE: nfs_fsinfo() is expected to override the default
1292 	 *	 f_iosize we set.
1293 	 */
1294 	if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1295 	    if (nmp->nm_flag & NFSMNT_NFSV3) {
1296 		mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1297 		error = nfs_fsinfo(nmp, vp, curthread);
1298 	    } else {
1299 		if ((error = VOP_GETATTR(vp, &attrs)) == 0)
1300 			nmp->nm_state |= NFSSTA_GOTFSINFO;
1301 
1302 	    }
1303 	} else {
1304 	    /*
1305 	     * The root vnode is usually cached by the namecache so do not
1306 	     * try to avoid going over the wire even if we have previous
1307 	     * information cached.  A stale NFS mount can loop
1308 	     * forever resolving the root vnode if we return no-error when
1309 	     * there is in fact an error.
1310 	     */
1311 	    np->n_attrstamp = 0;
1312 	    error = VOP_GETATTR(vp, &attrs);
1313 	}
1314 	if (vp->v_type == VNON)
1315 	    nfs_setvtype(vp, VDIR);
1316 	vsetflags(vp, VROOT);
1317 	if (error)
1318 		vput(vp);
1319 	else
1320 		*vpp = vp;
1321 	lwkt_reltoken(&nmp->nm_token);
1322 	return (error);
1323 }
1324 
1325 struct scaninfo {
1326 	int rescan;
1327 	int waitfor;
1328 	int allerror;
1329 };
1330 
1331 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1332 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1333 
1334 /*
1335  * Flush out the buffer cache
1336  */
1337 /* ARGSUSED */
1338 static int
1339 nfs_sync(struct mount *mp, int waitfor)
1340 {
1341 	struct nfsmount *nmp = VFSTONFS(mp);
1342 	struct scaninfo scaninfo;
1343 	int error;
1344 
1345 	scaninfo.rescan = 1;
1346 	scaninfo.waitfor = waitfor;
1347 	scaninfo.allerror = 0;
1348 
1349 	/*
1350 	 * Force stale buffer cache information to be flushed.
1351 	 */
1352 	lwkt_gettoken(&nmp->nm_token);
1353 	error = 0;
1354 	while (error == 0 && scaninfo.rescan) {
1355 		scaninfo.rescan = 0;
1356 		error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1357 					nfs_sync_scan2, &scaninfo);
1358 	}
1359 	lwkt_reltoken(&nmp->nm_token);
1360 	return(error);
1361 }
1362 
1363 static int
1364 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1365 {
1366     struct scaninfo *info = data;
1367 
1368     if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree))
1369 	return(-1);
1370     if (info->waitfor & MNT_LAZY)
1371 	return(-1);
1372     return(0);
1373 }
1374 
1375 static int
1376 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1377 {
1378     struct scaninfo *info = data;
1379     int error;
1380 
1381     error = VOP_FSYNC(vp, info->waitfor, 0);
1382     if (error)
1383 	info->allerror = error;
1384     return(0);
1385 }
1386 
1387