xref: /dragonfly/sys/vfs/nfs/nfs_vfsops.c (revision 2702099d)
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  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)nfs_vfsops.c	8.12 (Berkeley) 5/20/95
33  * $FreeBSD: src/sys/nfs/nfs_vfsops.c,v 1.91.2.7 2003/01/27 20:04:08 dillon Exp $
34  */
35 
36 #include "opt_bootp.h"
37 #include "opt_nfsroot.h"
38 
39 #include <sys/param.h>
40 #include <sys/sockio.h>
41 #include <sys/proc.h>
42 #include <sys/vnode.h>
43 #include <sys/fcntl.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/systm.h>
52 #include <sys/objcache.h>
53 
54 #include <vm/vm.h>
55 #include <vm/vm_extern.h>
56 
57 #include <net/if.h>
58 #include <net/route.h>
59 #include <netinet/in.h>
60 
61 #include <sys/thread2.h>
62 #include <sys/mutex2.h>
63 
64 #include "rpcv2.h"
65 #include "nfsproto.h"
66 #include "nfs.h"
67 #include "nfsmount.h"
68 #include "nfsnode.h"
69 #include "xdr_subs.h"
70 #include "nfsm_subs.h"
71 #include "nfsdiskless.h"
72 #include "nfsmountrpc.h"
73 
74 extern int	nfs_mountroot(struct mount *mp);
75 extern void	bootpc_init(void);
76 
77 extern struct vop_ops nfsv2_vnode_vops;
78 extern struct vop_ops nfsv2_fifo_vops;
79 extern struct vop_ops nfsv2_spec_vops;
80 
81 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
82 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
83 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
84 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
85 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
86 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
87 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
88 
89 struct objcache *nfsmount_objcache;
90 
91 struct nfsstats	nfsstats;
92 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
93 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD, &nfsstats, nfsstats,
94     "Nfs stats structure");
95 static int nfs_ip_paranoia = 1;
96 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW, &nfs_ip_paranoia, 0,
97     "Enable no-connection mode for protocols that support no-connection mode");
98 #ifdef NFS_DEBUG
99 int nfs_debug;
100 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
101 #endif
102 
103 /*
104  * Tunable to determine the Read/Write unit size.  Maximum value
105  * is NFS_MAXDATA.  We also default to NFS_MAXDATA.
106  */
107 static int nfs_io_size = NFS_MAXDATA;
108 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
109 	&nfs_io_size, 0, "NFS optimal I/O unit size");
110 
111 static void	nfs_decode_args (struct nfsmount *nmp,
112 			struct nfs_args *argp);
113 static int	mountnfs (struct nfs_args *,struct mount *,
114 			struct sockaddr *,char *,char *,struct vnode **);
115 static int	nfs_mount ( struct mount *mp, char *path, caddr_t data,
116 			struct ucred *cred);
117 static int	nfs_unmount ( struct mount *mp, int mntflags);
118 static int	nfs_root ( struct mount *mp, struct vnode **vpp);
119 static int	nfs_statfs ( struct mount *mp, struct statfs *sbp,
120 			struct ucred *cred);
121 static int	nfs_statvfs(struct mount *mp, struct statvfs *sbp,
122 				struct ucred *cred);
123 static int	nfs_sync ( struct mount *mp, int waitfor);
124 
125 /*
126  * nfs vfs operations.
127  */
128 static struct vfsops nfs_vfsops = {
129 	.vfs_mount =    	nfs_mount,
130 	.vfs_unmount =  	nfs_unmount,
131 	.vfs_root =     	nfs_root,
132 	.vfs_statfs =    	nfs_statfs,
133 	.vfs_statvfs =   	nfs_statvfs,
134 	.vfs_sync =     	nfs_sync,
135 	.vfs_init =     	nfs_init,
136 	.vfs_uninit =    	nfs_uninit
137 };
138 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
139 MODULE_VERSION(nfs, 1);
140 
141 /*
142  * This structure must be filled in by a primary bootstrap or bootstrap
143  * server for a diskless/dataless machine. It is initialized below just
144  * to ensure that it is allocated to initialized data (.data not .bss).
145  */
146 struct nfs_diskless nfs_diskless = { { { 0 } } };
147 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
148 int nfs_diskless_valid = 0;
149 
150 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
151 	&nfs_diskless_valid, 0,
152 	"NFS diskless params were obtained");
153 
154 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
155 	nfsv3_diskless.root_hostnam, 0,
156 	"Host name for mount point");
157 
158 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
159 	&nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
160 	"%Ssockaddr_in", "Address of root server");
161 
162 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
163 	nfsv3_diskless.swap_hostnam, 0,
164 	"Host name for mount ppoint");
165 
166 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
167 	&nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
168 	"%Ssockaddr_in", "Address of swap server");
169 
170 
171 void nfsargs_ntoh (struct nfs_args *);
172 static int nfs_mountdiskless (char *, char *, int,
173 				  struct sockaddr_in *, struct nfs_args *,
174 				  struct thread *, struct vnode **,
175 				  struct mount **);
176 static void nfs_convert_diskless (void);
177 static void nfs_convert_oargs (struct nfs_args *args,
178 				   struct onfs_args *oargs);
179 
180 /*
181  * Calculate the buffer I/O block size to use.  The maximum V2 block size
182  * is typically 8K, the maximum datagram size is typically 16K, and the
183  * maximum V3 block size is typically 32K.  The buffer cache tends to work
184  * best with 16K blocks but we allow 32K for TCP connections.
185  *
186  * We force the block size to be at least a page for buffer cache efficiency.
187  */
188 static int
189 nfs_iosize(int v3, int sotype)
190 {
191 	int iosize;
192 	int iomax;
193 
194 	if (v3) {
195 		if (sotype == SOCK_STREAM)
196 			iomax = NFS_MAXDATA;
197 		else
198 			iomax = NFS_MAXDGRAMDATA;
199 	} else {
200 		iomax = NFS_V2MAXDATA;
201 	}
202 	if ((iosize = nfs_io_size) > iomax)
203 		iosize = iomax;
204 	if (iosize < PAGE_SIZE)
205 		iosize = PAGE_SIZE;
206 
207 	/*
208 	 * This is an aweful hack but until the buffer cache is rewritten
209 	 * we need it.  The problem is that when you combine write() with
210 	 * mmap() the vm_page->valid bits can become weird looking
211 	 * (e.g. 0xfc).  This occurs because NFS uses piecemeal buffers
212 	 * at the file EOF.  To solve the problem the BIO system needs to
213 	 * be guarenteed that the NFS iosize for regular files will be a
214 	 * multiple of PAGE_SIZE so it can invalidate the whole page
215 	 * rather then just the piece of it owned by the buffer when
216 	 * NFS does vinvalbuf() calls.
217 	 */
218 	if (iosize & PAGE_MASK)
219 		iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
220 	return iosize;
221 }
222 
223 static void
224 nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
225 {
226 	args->version = NFS_ARGSVERSION;
227 	args->addr = oargs->addr;
228 	args->addrlen = oargs->addrlen;
229 	args->sotype = oargs->sotype;
230 	args->proto = oargs->proto;
231 	args->fh = oargs->fh;
232 	args->fhsize = oargs->fhsize;
233 	args->flags = oargs->flags;
234 	args->wsize = oargs->wsize;
235 	args->rsize = oargs->rsize;
236 	args->readdirsize = oargs->readdirsize;
237 	args->timeo = oargs->timeo;
238 	args->retrans = oargs->retrans;
239 	args->maxgrouplist = oargs->maxgrouplist;
240 	args->readahead = oargs->readahead;
241 	args->deadthresh = oargs->deadthresh;
242 	args->hostname = oargs->hostname;
243 }
244 
245 static void
246 nfs_convert_diskless(void)
247 {
248 	int i;
249 
250 	bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
251 		sizeof(struct ifaliasreq));
252 	bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
253 		sizeof(struct sockaddr_in));
254 	nfs_convert_oargs(&nfsv3_diskless.swap_args, &nfs_diskless.swap_args);
255 
256 	/*
257 	 * Copy the NFS handle passed from the diskless code.
258 	 *
259 	 * XXX CURRENTLY DISABLED - bootp passes us a NFSv2 handle which
260 	 * will fail utterly with HAMMER due to limitations with NFSv2
261 	 * directory cookies.
262 	 */
263 	bcopy(nfs_diskless.swap_fh, nfsv3_diskless.swap_fh, NFSX_V2FH);
264 	nfsv3_diskless.swap_fhsize = NFSX_V2FH;
265 	for (i = NFSX_V2FH - 1; i >= 0; --i) {
266 		if (nfs_diskless.swap_fh[i])
267 			break;
268 	}
269 	if (i < 0)
270 		nfsv3_diskless.swap_fhsize = 0;
271 	nfsv3_diskless.swap_fhsize = 0;		/* FORCE DISABLE */
272 
273 	bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
274 		sizeof(struct sockaddr_in));
275 	bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
276 	nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
277 	bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
278 		sizeof(struct ucred));
279 	nfs_convert_oargs(&nfsv3_diskless.root_args, &nfs_diskless.root_args);
280 
281 	/*
282 	 * Copy the NFS handle passed from the diskless code.
283 	 *
284 	 * XXX CURRENTLY DISABLED - bootp passes us a NFSv2 handle which
285 	 * will fail utterly with HAMMER due to limitations with NFSv2
286 	 * directory cookies.
287 	 */
288 	bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V2FH);
289 	nfsv3_diskless.root_fhsize = NFSX_V2FH;
290 	for (i = NFSX_V2FH - 1; i >= 0; --i) {
291 		if (nfs_diskless.root_fh[i])
292 			break;
293 	}
294 	if (i < 0)
295 		nfsv3_diskless.root_fhsize = 0;
296 	nfsv3_diskless.root_fhsize = 0;		/* FORCE DISABLE */
297 
298 	bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
299 		sizeof(struct sockaddr_in));
300 	bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
301 	nfsv3_diskless.root_time = nfs_diskless.root_time;
302 	bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
303 		MAXHOSTNAMELEN);
304 	nfs_diskless_valid = 3;
305 }
306 
307 /*
308  * nfs statfs call
309  */
310 int
311 nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
312 {
313 	struct vnode *vp;
314 	struct nfs_statfs *sfp;
315 	struct nfsmount *nmp = VFSTONFS(mp);
316 	thread_t td = curthread;
317 	int error = 0, retattr;
318 	struct nfsnode *np;
319 	u_quad_t tquad;
320 	struct nfsm_info info;
321 
322 	info.mrep = NULL;
323 	info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
324 
325 	lwkt_gettoken(&nmp->nm_token);
326 
327 #ifndef nolint
328 	sfp = NULL;
329 #endif
330 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
331 	if (error) {
332 		lwkt_reltoken(&nmp->nm_token);
333 		return (error);
334 	}
335 	vp = NFSTOV(np);
336 	/* ignore the passed cred */
337 	cred = crget();
338 	cred->cr_ngroups = 1;
339 	if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
340 		(void)nfs_fsinfo(nmp, vp, td);
341 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
342 	nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
343 	ERROROUT(nfsm_fhtom(&info, vp));
344 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
345 	if (info.v3) {
346 		ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
347 					 NFS_LATTR_NOSHRINK));
348 	}
349 	if (error) {
350 		if (info.mrep != NULL)
351 			m_freem(info.mrep);
352 		goto nfsmout;
353 	}
354 	NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
355 	sbp->f_flags = nmp->nm_flag;
356 
357 	if (info.v3) {
358 		sbp->f_bsize = NFS_FABLKSIZE;
359 		tquad = fxdr_hyper(&sfp->sf_tbytes);
360 		sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
361 		tquad = fxdr_hyper(&sfp->sf_fbytes);
362 		sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
363 		tquad = fxdr_hyper(&sfp->sf_abytes);
364 		sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
365 		sbp->f_files = (fxdr_unsigned(int32_t,
366 		    sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
367 		sbp->f_ffree = (fxdr_unsigned(int32_t,
368 		    sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
369 	} else {
370 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
371 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
372 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
373 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
374 		sbp->f_files = 0;
375 		sbp->f_ffree = 0;
376 	}
377 
378 	/*
379 	 * Some values are pre-set in mnt_stat.  Note in particular f_iosize
380 	 * cannot be changed once the filesystem is mounted as it is used
381 	 * as the basis for BIOs.
382 	 */
383 	if (sbp != &mp->mnt_stat) {
384 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
385 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
386 		sbp->f_iosize = mp->mnt_stat.f_iosize;
387 	}
388 	m_freem(info.mrep);
389 	info.mrep = NULL;
390 nfsmout:
391 	vput(vp);
392 	crfree(cred);
393 	lwkt_reltoken(&nmp->nm_token);
394 	return (error);
395 }
396 
397 static int
398 nfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
399 {
400 	struct vnode *vp;
401 	struct nfs_statfs *sfp;
402 	struct nfsmount *nmp = VFSTONFS(mp);
403 	thread_t td = curthread;
404 	int error = 0, retattr;
405 	struct nfsnode *np;
406 	struct nfsm_info info;
407 
408 	info.mrep = NULL;
409 	info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
410 	lwkt_gettoken(&nmp->nm_token);
411 
412 #ifndef nolint
413 	sfp = NULL;
414 #endif
415 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
416 	if (error) {
417 		lwkt_reltoken(&nmp->nm_token);
418 		return (error);
419 	}
420 	vp = NFSTOV(np);
421 	/* ignore the passed cred */
422 	cred = crget();
423 	cred->cr_ngroups = 1;
424 	if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
425 		(void)nfs_fsinfo(nmp, vp, td);
426 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
427 	nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
428 	ERROROUT(nfsm_fhtom(&info, vp));
429 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
430 	if (info.v3) {
431 		ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
432 					 NFS_LATTR_NOSHRINK));
433 	}
434 	if (error) {
435 		if (info.mrep != NULL)
436 			m_freem(info.mrep);
437 		goto nfsmout;
438 	}
439 	NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
440 	sbp->f_flag = nmp->nm_flag;
441 	sbp->f_owner = nmp->nm_cred->cr_ruid;
442 
443 	if (info.v3) {
444 		sbp->f_bsize = NFS_FABLKSIZE;
445 		sbp->f_frsize = NFS_FABLKSIZE;
446 		sbp->f_blocks = (fxdr_hyper(&sfp->sf_tbytes) /
447 				((u_quad_t)NFS_FABLKSIZE));
448 		sbp->f_bfree = (fxdr_hyper(&sfp->sf_fbytes) /
449 				((u_quad_t)NFS_FABLKSIZE));
450 		sbp->f_bavail = (fxdr_hyper(&sfp->sf_abytes) /
451 				((u_quad_t)NFS_FABLKSIZE));
452 		sbp->f_files = fxdr_hyper(&sfp->sf_tfiles);
453 		sbp->f_ffree = fxdr_hyper(&sfp->sf_ffiles);
454 		sbp->f_favail = fxdr_hyper(&sfp->sf_afiles);
455 	} else {
456 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
457 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
458 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
459 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
460 		sbp->f_files = 0;
461 		sbp->f_ffree = 0;
462 		sbp->f_favail = 0;
463 	}
464 	sbp->f_syncreads = 0;
465 	sbp->f_syncwrites = 0;
466 	sbp->f_asyncreads = 0;
467 	sbp->f_asyncwrites = 0;
468 	sbp->f_type = mp->mnt_vfc->vfc_typenum;
469 
470 	m_freem(info.mrep);
471 	info.mrep = NULL;
472 nfsmout:
473 	vput(vp);
474 	crfree(cred);
475 	lwkt_reltoken(&nmp->nm_token);
476 	return (error);
477 }
478 
479 /*
480  * nfs version 3 fsinfo rpc call
481  */
482 int
483 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
484 {
485 	struct nfsv3_fsinfo *fsp;
486 	u_int32_t pref, max;
487 	int error = 0, retattr;
488 	u_int64_t maxfsize;
489 	struct nfsm_info info;
490 
491 	info.v3 = 1;
492 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
493 	nfsm_reqhead(&info, vp, NFSPROC_FSINFO, NFSX_FH(1));
494 	ERROROUT(nfsm_fhtom(&info, vp));
495 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSINFO, td,
496 				nfs_vpcred(vp, ND_READ), &error));
497 	ERROROUT(nfsm_postop_attr(&info, vp, &retattr, NFS_LATTR_NOSHRINK));
498 	if (error == 0) {
499 		NULLOUT(fsp = nfsm_dissect(&info, NFSX_V3FSINFO));
500 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
501 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
502 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
503 				~(NFS_FABLKSIZE - 1);
504 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
505 		if (max < nmp->nm_wsize && max > 0) {
506 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
507 			if (nmp->nm_wsize == 0)
508 				nmp->nm_wsize = max;
509 		}
510 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
511 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
512 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
513 				~(NFS_FABLKSIZE - 1);
514 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
515 		if (max < nmp->nm_rsize && max > 0) {
516 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
517 			if (nmp->nm_rsize == 0)
518 				nmp->nm_rsize = max;
519 		}
520 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
521 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
522 			nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
523 				~(NFS_DIRBLKSIZ - 1);
524 		if (max < nmp->nm_readdirsize && max > 0) {
525 			nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
526 			if (nmp->nm_readdirsize == 0)
527 				nmp->nm_readdirsize = max;
528 		}
529 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
530 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
531 			nmp->nm_maxfilesize = maxfsize;
532 		nmp->nm_state |= NFSSTA_GOTFSINFO;
533 
534 		/*
535 		 * Use the smaller of rsize/wsize for the biosize.
536 		 */
537 		if (nmp->nm_rsize < nmp->nm_wsize)
538 			nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_rsize;
539 		else
540 			nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_wsize;
541 	}
542 	m_freem(info.mrep);
543 	info.mrep = NULL;
544 nfsmout:
545 	return (error);
546 }
547 
548 /*
549  * Mount a remote root fs via. nfs. This depends on the info in the
550  * nfs_diskless structure that has been filled in properly by some primary
551  * bootstrap.
552  * It goes something like this:
553  * - do enough of "ifconfig" by calling ifioctl() so that the system
554  *   can talk to the server
555  * - If nfs_diskless.mygateway is filled in, use that address as
556  *   a default gateway.
557  * - build the rootfs mount point and call mountnfs() to do the rest.
558  */
559 int
560 nfs_mountroot(struct mount *mp)
561 {
562 	struct mount  *swap_mp;
563 	struct nfsv3_diskless *nd = &nfsv3_diskless;
564 	struct socket *so;
565 	struct vnode *vp;
566 	struct thread *td = curthread;		/* XXX */
567 	int error, i;
568 	u_long l;
569 	char buf[128];
570 
571 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
572 	bootpc_init();		/* use bootp to get nfs_diskless filled in */
573 #endif
574 
575 	/*
576 	 * XXX time must be non-zero when we init the interface or else
577 	 * the arp code will wedge...
578 	 */
579 	while (mycpu->gd_time_seconds == 0)
580 		tsleep(mycpu, 0, "arpkludge", 10);
581 
582 	/*
583 	 * The boot code may have passed us a diskless structure.
584 	 */
585 	kprintf("DISKLESS %d\n", nfs_diskless_valid);
586 	if (nfs_diskless_valid == 1)
587 		nfs_convert_diskless();
588 
589 	/*
590 	 * NFSv3 is required.
591 	 */
592 	nd->root_args.flags |= NFSMNT_NFSV3 | NFSMNT_RDIRPLUS;
593 	nd->swap_args.flags |= NFSMNT_NFSV3;
594 
595 #define SINP(sockaddr)	((struct sockaddr_in *)(sockaddr))
596 	kprintf("nfs_mountroot: interface %s ip %s",
597 		nd->myif.ifra_name,
598 		inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
599 	kprintf(" bcast %s",
600 		inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
601 	kprintf(" mask %s\n",
602 		inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
603 #undef SINP
604 
605 	/*
606 	 * XXX splnet, so networks will receive...
607 	 */
608 	crit_enter();
609 
610 	/*
611 	 * BOOTP does not necessarily have to be compiled into the kernel
612 	 * for an NFS root to work.  If we inherited the network
613 	 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
614 	 * out our interface for us and all we need to do is ifconfig the
615 	 * interface.  We only do this if the interface has not already been
616 	 * ifconfig'd by e.g. BOOTP.
617 	 */
618 	error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
619 	if (error) {
620 		panic("nfs_mountroot: socreate(%04x): %d",
621 			nd->myif.ifra_addr.sa_family, error);
622 	}
623 
624 	error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
625 	if (error)
626 		panic("nfs_mountroot: SIOCAIFADDR: %d", error);
627 
628 	soclose(so, FNONBLOCK);
629 
630 	/*
631 	 * If the gateway field is filled in, set it as the default route.
632 	 */
633 	if (nd->mygateway.sin_len != 0) {
634 		struct sockaddr_in mask, sin;
635 
636 		bzero((caddr_t)&mask, sizeof(mask));
637 		sin = mask;
638 		sin.sin_family = AF_INET;
639 		sin.sin_len = sizeof(sin);
640 		kprintf("nfs_mountroot: gateway %s\n",
641 			inet_ntoa(nd->mygateway.sin_addr));
642 		error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
643 					(struct sockaddr *)&nd->mygateway,
644 					(struct sockaddr *)&mask,
645 					RTF_UP | RTF_GATEWAY);
646 		if (error)
647 			kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
648 	}
649 
650 	/*
651 	 * Create the rootfs mount point.
652 	 */
653 	nd->root_args.fh = nd->root_fh;
654 	nd->root_args.fhsize = nd->root_fhsize;
655 	l = ntohl(nd->root_saddr.sin_addr.s_addr);
656 	ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
657 		(l >> 24) & 0xff, (l >> 16) & 0xff,
658 		(l >>  8) & 0xff, (l >>  0) & 0xff,nd->root_hostnam);
659 	kprintf("NFS_ROOT: %s\n",buf);
660 	error = nfs_mountdiskless(buf, "/", MNT_RDONLY, &nd->root_saddr,
661 				  &nd->root_args, td, &vp, &mp);
662 	if (error) {
663 		mp->mnt_vfc->vfc_refcount--;
664 		crit_exit();
665 		return (error);
666 	}
667 
668 	swap_mp = NULL;
669 	if (nd->swap_nblks) {
670 
671 		/* Convert to DEV_BSIZE instead of Kilobyte */
672 		nd->swap_nblks *= 2;
673 
674 		/*
675 		 * Create a fake mount point just for the swap vnode so that the
676 		 * swap file can be on a different server from the rootfs.
677 		 */
678 		nd->swap_args.fh = nd->swap_fh;
679 		nd->swap_args.fhsize = nd->swap_fhsize;
680 		l = ntohl(nd->swap_saddr.sin_addr.s_addr);
681 		ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
682 			(l >> 24) & 0xff, (l >> 16) & 0xff,
683 			(l >>  8) & 0xff, (l >>  0) & 0xff,nd->swap_hostnam);
684 		kprintf("NFS SWAP: %s\n",buf);
685 		error = nfs_mountdiskless(buf, "/swap", 0, &nd->swap_saddr,
686 					  &nd->swap_args, td, &vp, &swap_mp);
687 		if (error) {
688 			crit_exit();
689 			return (error);
690 		}
691 		vfs_unbusy(swap_mp);
692 
693 		VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
694 				nd->swap_nblks * DEV_BSIZE ;
695 
696 		/*
697 		 * Since the swap file is not the root dir of a file system,
698 		 * hack it to a regular file.
699 		 */
700 		vclrflags(vp, VROOT);
701 		vref(vp);
702 		nfs_setvtype(vp, VREG);
703 		swaponvp(td, vp, nd->swap_nblks);
704 	}
705 
706 	mp->mnt_flag |= MNT_ROOTFS;
707 
708 	/*
709 	 * This is not really an nfs issue, but it is much easier to
710 	 * set hostname here and then let the "/etc/rc.xxx" files
711 	 * mount the right /var based upon its preset value.
712 	 */
713 	bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
714 	hostname[MAXHOSTNAMELEN - 1] = '\0';
715 	for (i = 0; i < MAXHOSTNAMELEN; i++)
716 		if (hostname[i] == '\0')
717 			break;
718 	inittodr(ntohl(nd->root_time));
719 	crit_exit();
720 	return (0);
721 }
722 
723 /*
724  * Internal version of mount system call for diskless setup.
725  */
726 static int
727 nfs_mountdiskless(char *path, char *which, int mountflag,
728 	struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
729 	struct vnode **vpp, struct mount **mpp)
730 {
731 	struct mount *mp;
732 	struct sockaddr *nam;
733 	int didalloc = 0;
734 	int error;
735 
736 	mp = *mpp;
737 
738 	if (mp == NULL) {
739 		if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
740 			kprintf("nfs_mountroot: NFS not configured");
741 			return (error);
742 		}
743 		didalloc = 1;
744 	}
745 	mp->mnt_kern_flag = 0;
746 	mp->mnt_flag = mountflag;
747 	nam = dup_sockaddr((struct sockaddr *)sin);
748 
749 #if defined(BOOTP) || defined(NFS_ROOT)
750 	if (args->fhsize == 0) {
751 		char *xpath = path;
752 
753 		kprintf("NFS_ROOT: No FH passed from loader, attempting "
754 			"mount rpc...");
755 		while (*xpath && *xpath != ':')
756 			++xpath;
757 		if (*xpath)
758 			++xpath;
759 		args->fhsize = 0;
760 		error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td);
761 		if (error) {
762 			kprintf("failed error %d.\n", error);
763 			goto haderror;
764 		}
765 		kprintf("success!\n");
766 	}
767 #endif
768 
769 	if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
770 #if defined(BOOTP) || defined(NFS_ROOT)
771 haderror:
772 #endif
773 		kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error);
774 		mp->mnt_vfc->vfc_refcount--;
775 		if (didalloc)
776 			kfree(mp, M_MOUNT);
777 		kfree(nam, M_SONAME);
778 		return (error);
779 	}
780 	*mpp = mp;
781 	return (0);
782 }
783 
784 static void
785 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp)
786 {
787 	int adjsock;
788 	int maxio;
789 
790 	crit_enter();
791 	/*
792 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
793 	 * no sense in that context.
794 	 */
795 	if (nmp->nm_sotype == SOCK_STREAM) {
796 		nmp->nm_flag &= ~NFSMNT_NOCONN;
797 		argp->flags &= ~NFSMNT_NOCONN;
798 	}
799 
800 	/*
801 	 * readdirplus is NFSv3 only.
802 	 */
803 	if ((argp->flags & NFSMNT_NFSV3) == 0) {
804 		nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
805 		argp->flags &= ~NFSMNT_RDIRPLUS;
806 	}
807 
808 	/*
809 	 * Re-bind if rsrvd port flag has changed
810 	 */
811 	adjsock = (nmp->nm_flag & NFSMNT_RESVPORT) !=
812 		  (argp->flags & NFSMNT_RESVPORT);
813 
814 	/* Update flags atomically.  Don't change the lock bits. */
815 	nmp->nm_flag = argp->flags | nmp->nm_flag;
816 	crit_exit();
817 
818 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
819 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
820 		if (nmp->nm_timeo < NFS_MINTIMEO)
821 			nmp->nm_timeo = NFS_MINTIMEO;
822 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
823 			nmp->nm_timeo = NFS_MAXTIMEO;
824 	}
825 
826 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
827 		nmp->nm_retry = argp->retrans;
828 		if (nmp->nm_retry > NFS_MAXREXMIT)
829 			nmp->nm_retry = NFS_MAXREXMIT;
830 	}
831 
832 	/*
833 	 * These parameters effect the buffer cache and cannot be changed
834 	 * once we've successfully mounted.
835 	 */
836 	if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
837 		maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, nmp->nm_sotype);
838 
839 		if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
840 			nmp->nm_wsize = argp->wsize;
841 			/* Round down to multiple of blocksize */
842 			nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
843 			if (nmp->nm_wsize <= 0)
844 				nmp->nm_wsize = NFS_FABLKSIZE;
845 		}
846 		if (nmp->nm_wsize > maxio)
847 			nmp->nm_wsize = maxio;
848 		if (nmp->nm_wsize > MAXBSIZE)
849 			nmp->nm_wsize = MAXBSIZE;
850 
851 		if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
852 			nmp->nm_rsize = argp->rsize;
853 			/* Round down to multiple of blocksize */
854 			nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
855 			if (nmp->nm_rsize <= 0)
856 				nmp->nm_rsize = NFS_FABLKSIZE;
857 		}
858 		if (nmp->nm_rsize > maxio)
859 			nmp->nm_rsize = maxio;
860 		if (nmp->nm_rsize > MAXBSIZE)
861 			nmp->nm_rsize = MAXBSIZE;
862 
863 		if ((argp->flags & NFSMNT_READDIRSIZE) &&
864 		    argp->readdirsize > 0) {
865 			nmp->nm_readdirsize = argp->readdirsize;
866 		}
867 		if (nmp->nm_readdirsize > maxio)
868 			nmp->nm_readdirsize = maxio;
869 		if (nmp->nm_readdirsize > nmp->nm_rsize)
870 			nmp->nm_readdirsize = nmp->nm_rsize;
871 	}
872 
873 	if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
874 		nmp->nm_acregmin = argp->acregmin;
875 	else
876 		nmp->nm_acregmin = NFS_MINATTRTIMO;
877 	if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
878 		nmp->nm_acregmax = argp->acregmax;
879 	else
880 		nmp->nm_acregmax = NFS_MAXATTRTIMO;
881 	if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
882 		nmp->nm_acdirmin = argp->acdirmin;
883 	else
884 		nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
885 	if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
886 		nmp->nm_acdirmax = argp->acdirmax;
887 	else
888 		nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
889 	if (nmp->nm_acdirmin > nmp->nm_acdirmax)
890 		nmp->nm_acdirmin = nmp->nm_acdirmax;
891 	if (nmp->nm_acregmin > nmp->nm_acregmax)
892 		nmp->nm_acregmin = nmp->nm_acregmax;
893 
894 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
895 		if (argp->maxgrouplist <= NFS_MAXGRPS)
896 			nmp->nm_numgrps = argp->maxgrouplist;
897 		else
898 			nmp->nm_numgrps = NFS_MAXGRPS;
899 	}
900 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
901 		if (argp->readahead <= NFS_MAXRAHEAD)
902 			nmp->nm_readahead = argp->readahead;
903 		else
904 			nmp->nm_readahead = NFS_MAXRAHEAD;
905 	}
906 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
907 		if (argp->deadthresh <= NFS_NEVERDEAD)
908 			nmp->nm_deadthresh = argp->deadthresh;
909 		else
910 			nmp->nm_deadthresh = NFS_NEVERDEAD;
911 	}
912 
913 	if (nmp->nm_so && adjsock) {
914 		nfs_safedisconnect(nmp);
915 		if (nmp->nm_sotype == SOCK_DGRAM)
916 			while (nfs_connect(nmp, NULL)) {
917 				kprintf("nfs_args: retrying connect\n");
918 				(void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
919 			}
920 	}
921 }
922 
923 /*
924  * VFS Operations.
925  *
926  * mount system call
927  * It seems a bit dumb to copyinstr() the host and path here and then
928  * bcopy() them in mountnfs(), but I wanted to detect errors before
929  * doing the sockargs() call because sockargs() allocates an mbuf and
930  * an error after that means that I have to release the mbuf.
931  */
932 /* ARGSUSED */
933 static int
934 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
935 {
936 	int error;
937 	struct nfs_args args;
938 	struct sockaddr *nam;
939 	struct vnode *vp;
940 	char pth[MNAMELEN], hst[MNAMELEN];
941 	size_t len;
942 	u_char nfh[NFSX_V3FHMAX];
943 
944 	if (path == NULL) {
945 		nfs_mountroot(mp);
946 		return (0);
947 	}
948 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
949 	if (error)
950 		return (error);
951 	if (args.version != NFS_ARGSVERSION) {
952 #ifdef COMPAT_PRELITE2
953 		/*
954 		 * If the argument version is unknown, then assume the
955 		 * caller is a pre-lite2 4.4BSD client and convert its
956 		 * arguments.
957 		 */
958 		struct onfs_args oargs;
959 		error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
960 		if (error)
961 			return (error);
962 		nfs_convert_oargs(&args,&oargs);
963 #else /* !COMPAT_PRELITE2 */
964 		return (EPROGMISMATCH);
965 #endif /* COMPAT_PRELITE2 */
966 	}
967 	if (mp->mnt_flag & MNT_UPDATE) {
968 		struct nfsmount *nmp = VFSTONFS(mp);
969 
970 		if (nmp == NULL)
971 			return (EIO);
972 		/*
973 		 * When doing an update, we can't change from or to
974 		 * v3, or change cookie translation, or rsize or wsize.
975 		 */
976 		args.flags &= ~(NFSMNT_NFSV3 | NFSMNT_RSIZE | NFSMNT_WSIZE);
977 		args.flags |= nmp->nm_flag & (NFSMNT_NFSV3);
978 		nfs_decode_args(nmp, &args);
979 		return (0);
980 	}
981 
982 	/*
983 	 * Make the nfs_ip_paranoia sysctl serve as the default connection
984 	 * or no-connection mode for those protocols that support
985 	 * no-connection mode (the flag will be cleared later for protocols
986 	 * that do not support no-connection mode).  This will allow a client
987 	 * to receive replies from a different IP then the request was
988 	 * sent to.  Note: default value for nfs_ip_paranoia is 1 (paranoid),
989 	 * not 0.
990 	 */
991 	if (nfs_ip_paranoia == 0)
992 		args.flags |= NFSMNT_NOCONN;
993 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
994 		return (EINVAL);
995 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
996 	if (error)
997 		return (error);
998 	error = copyinstr(path, pth, MNAMELEN-1, &len);
999 	if (error)
1000 		return (error);
1001 	bzero(&pth[len], MNAMELEN - len);
1002 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
1003 	if (error)
1004 		return (error);
1005 	bzero(&hst[len], MNAMELEN - len);
1006 	/* sockargs() call must be after above copyin() calls */
1007 	error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
1008 	if (error)
1009 		return (error);
1010 	args.fh = nfh;
1011 	error = mountnfs(&args, mp, nam, pth, hst, &vp);
1012 	return (error);
1013 }
1014 
1015 /*
1016  * Common code for mount and mountroot
1017  */
1018 static int
1019 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
1020 	char *pth, char *hst, struct vnode **vpp)
1021 {
1022 	struct nfsmount *nmp;
1023 	struct nfsnode *np;
1024 	int error;
1025 	int rxcpu;
1026 	int txcpu;
1027 
1028 	if (mp->mnt_flag & MNT_UPDATE) {
1029 		nmp = VFSTONFS(mp);
1030 		/* update paths, file handles, etc, here	XXX */
1031 		kfree(nam, M_SONAME);
1032 		return (0);
1033 	} else {
1034 		nmp = objcache_get(nfsmount_objcache, M_WAITOK);
1035 		bzero((caddr_t)nmp, sizeof (struct nfsmount));
1036 		mtx_init(&nmp->nm_rxlock);
1037 		mtx_init(&nmp->nm_txlock);
1038 		TAILQ_INIT(&nmp->nm_uidlruhead);
1039 		TAILQ_INIT(&nmp->nm_bioq);
1040 		TAILQ_INIT(&nmp->nm_reqq);
1041 		TAILQ_INIT(&nmp->nm_reqtxq);
1042 		TAILQ_INIT(&nmp->nm_reqrxq);
1043 		mp->mnt_data = (qaddr_t)nmp;
1044 		lwkt_token_init(&nmp->nm_token, "nfs_token");
1045 	}
1046 	vfs_getnewfsid(mp);
1047 	nmp->nm_mountp = mp;
1048 	mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
1049 	mp->mnt_kern_flag |= MNTK_THR_SYNC;
1050 
1051 	lwkt_gettoken(&nmp->nm_token);
1052 
1053 	/*
1054 	 * V2 can only handle 32 bit filesizes.  A 4GB-1 limit may be too
1055 	 * high, depending on whether we end up with negative offsets in
1056 	 * the client or server somewhere.  2GB-1 may be safer.
1057 	 *
1058 	 * For V3, nfs_fsinfo will adjust this as necessary.  Assume maximum
1059 	 * that we can handle until we find out otherwise.  Note that seek
1060 	 * offsets are signed.
1061 	 */
1062 	if ((argp->flags & NFSMNT_NFSV3) == 0)
1063 		nmp->nm_maxfilesize = 0xffffffffLL;
1064 	else
1065 		nmp->nm_maxfilesize = 0x7fffffffffffffffLL;
1066 
1067 	nmp->nm_timeo = NFS_TIMEO;
1068 	nmp->nm_retry = NFS_RETRANS;
1069 	nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
1070 	nmp->nm_rsize = nmp->nm_wsize;
1071 	nmp->nm_readdirsize = NFS_READDIRSIZE;
1072 	nmp->nm_numgrps = NFS_MAXGRPS;
1073 	nmp->nm_readahead = NFS_DEFRAHEAD;
1074 	nmp->nm_deadthresh = NFS_DEADTHRESH;
1075 	nmp->nm_fhsize = argp->fhsize;
1076 	bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
1077 	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
1078 	nmp->nm_nam = nam;
1079 	/* Set up the sockets and per-host congestion */
1080 	nmp->nm_sotype = argp->sotype;
1081 	nmp->nm_soproto = argp->proto;
1082 	nmp->nm_cred = crhold(proc0.p_ucred);
1083 
1084 	nfs_decode_args(nmp, argp);
1085 
1086 	/*
1087 	 * For Connection based sockets (TCP,...) defer the connect until
1088 	 * the first request, in case the server is not responding.
1089 	 */
1090 	if (nmp->nm_sotype == SOCK_DGRAM &&
1091 		(error = nfs_connect(nmp, NULL)))
1092 		goto bad;
1093 
1094 	/*
1095 	 * This is silly, but it has to be set so that vinifod() works.
1096 	 * We do not want to do an nfs_statfs() here since we can get
1097 	 * stuck on a dead server and we are holding a lock on the mount
1098 	 * point.
1099 	 */
1100 	mp->mnt_stat.f_iosize =
1101 		nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
1102 
1103 	/*
1104 	 * Install vop_ops for our vnops
1105 	 */
1106 	vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
1107 	vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
1108 	vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
1109 
1110 	/*
1111 	 * A reference count is needed on the nfsnode representing the
1112 	 * remote root.  If this object is not persistent, then backward
1113 	 * traversals of the mount point (i.e. "..") will not work if
1114 	 * the nfsnode gets flushed out of the cache. Ufs does not have
1115 	 * this problem, because one can identify root inodes by their
1116 	 * number == ROOTINO (2).
1117 	 */
1118 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
1119 	if (error)
1120 		goto bad;
1121 	*vpp = NFSTOV(np);
1122 
1123 	/*
1124 	 * Retrieval of mountpoint attributes is delayed until nfs_rot
1125 	 * or nfs_statfs are first called.  This will happen either when
1126 	 * we first traverse the mount point or if somebody does a df(1).
1127 	 *
1128 	 * NFSSTA_GOTFSINFO is used to flag if we have successfully
1129 	 * retrieved mountpoint attributes.  In the case of NFSv3 we
1130 	 * also flag static fsinfo.
1131 	 */
1132 	if (*vpp != NULL)
1133 		(*vpp)->v_type = VNON;
1134 
1135 	/*
1136 	 * Lose the lock but keep the ref.
1137 	 */
1138 	vn_unlock(*vpp);
1139 	lwkt_gettoken(&nfs_token);
1140 	TAILQ_INSERT_TAIL(&nfs_mountq, nmp, nm_entry);
1141 	lwkt_reltoken(&nfs_token);
1142 
1143 	switch(ncpus) {
1144 	case 0:
1145 	case 1:
1146 		rxcpu = 0;
1147 		txcpu = 0;
1148 		break;
1149 	case 2:
1150 		rxcpu = 0;
1151 		txcpu = 1;
1152 		break;
1153 	default:
1154 		rxcpu = -1;
1155 		txcpu = -1;
1156 		break;
1157 	}
1158 
1159 	/*
1160 	 * Start the reader and writer threads.
1161 	 */
1162 	lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread,
1163 		    NULL, 0, rxcpu, "nfsiod_rx");
1164 	lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread,
1165 		    NULL, 0, txcpu, "nfsiod_tx");
1166 	lwkt_reltoken(&nmp->nm_token);
1167 	return (0);
1168 bad:
1169 	nfs_disconnect(nmp);
1170 	lwkt_reltoken(&nmp->nm_token);
1171 	nfs_free_mount(nmp);
1172 	return (error);
1173 }
1174 
1175 /*
1176  * unmount system call
1177  */
1178 static int
1179 nfs_unmount(struct mount *mp, int mntflags)
1180 {
1181 	struct nfsmount *nmp;
1182 	int error, flags = 0;
1183 
1184 	nmp = VFSTONFS(mp);
1185 	lwkt_gettoken(&nmp->nm_token);
1186 	if (mntflags & MNT_FORCE) {
1187 		flags |= FORCECLOSE;
1188 		nmp->nm_flag |= NFSMNT_FORCE;
1189 	}
1190 
1191 	/*
1192 	 * Goes something like this..
1193 	 * - Call vflush() to clear out vnodes for this file system
1194 	 * - Close the socket
1195 	 * - Free up the data structures
1196 	 */
1197 	/* In the forced case, cancel any outstanding requests. */
1198 	if (flags & FORCECLOSE) {
1199 		error = nfs_nmcancelreqs(nmp);
1200 		if (error) {
1201 			kprintf("NFS: %s: Unable to cancel all requests\n",
1202 				mp->mnt_stat.f_mntfromname);
1203 			/* continue anyway */
1204 		}
1205 	}
1206 
1207 	/*
1208 	 * Must handshake with nfs_clientd() if it is active. XXX
1209 	 */
1210 	nmp->nm_state |= NFSSTA_DISMINPROG;
1211 
1212 	/*
1213 	 * We hold 1 extra ref on the root vnode; see comment in mountnfs().
1214 	 *
1215 	 * If this doesn't work and we are doing a forced unmount we continue
1216 	 * anyway.
1217 	 */
1218 	error = vflush(mp, 1, flags);
1219 	if (error) {
1220 		nmp->nm_state &= ~NFSSTA_DISMINPROG;
1221 		if ((flags & FORCECLOSE) == 0) {
1222 			lwkt_reltoken(&nmp->nm_token);
1223 			return (error);
1224 		}
1225 	}
1226 
1227 	/*
1228 	 * We are now committed to the unmount.
1229 	 * For NQNFS, let the server daemon free the nfsmount structure.
1230 	 */
1231 	if (nmp->nm_flag & NFSMNT_KERB)
1232 		nmp->nm_state |= NFSSTA_DISMNT;
1233 	nfssvc_iod_stop1(nmp);
1234 	nfs_disconnect(nmp);
1235 	nfssvc_iod_stop2(nmp);
1236 
1237 	lwkt_gettoken(&nfs_token);
1238 	TAILQ_REMOVE(&nfs_mountq, nmp, nm_entry);
1239 	lwkt_reltoken(&nfs_token);
1240 
1241 	lwkt_reltoken(&nmp->nm_token);
1242 
1243 	if ((nmp->nm_flag & NFSMNT_KERB) == 0) {
1244 		nfs_free_mount(nmp);
1245 	}
1246 	return (0);
1247 }
1248 
1249 void
1250 nfs_free_mount(struct nfsmount *nmp)
1251 {
1252 	if (nmp->nm_cred)  {
1253 		crfree(nmp->nm_cred);
1254 		nmp->nm_cred = NULL;
1255 	}
1256 	if (nmp->nm_nam) {
1257 		kfree(nmp->nm_nam, M_SONAME);
1258 		nmp->nm_nam = NULL;
1259 	}
1260 	objcache_put(nfsmount_objcache, nmp);
1261 }
1262 
1263 /*
1264  * Return root of a filesystem
1265  */
1266 static int
1267 nfs_root(struct mount *mp, struct vnode **vpp)
1268 {
1269 	struct vnode *vp;
1270 	struct nfsmount *nmp;
1271 	struct vattr attrs;
1272 	struct nfsnode *np;
1273 	int error;
1274 
1275 	nmp = VFSTONFS(mp);
1276 	lwkt_gettoken(&nmp->nm_token);
1277 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
1278 	if (error) {
1279 		lwkt_reltoken(&nmp->nm_token);
1280 		return (error);
1281 	}
1282 	vp = NFSTOV(np);
1283 
1284 	/*
1285 	 * Get transfer parameters and root vnode attributes
1286 	 *
1287 	 * NOTE: nfs_fsinfo() is expected to override the default
1288 	 *	 f_iosize we set.
1289 	 */
1290 	if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1291 	    if (nmp->nm_flag & NFSMNT_NFSV3) {
1292 		mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1293 		error = nfs_fsinfo(nmp, vp, curthread);
1294 	    } else {
1295 		if ((error = VOP_GETATTR(vp, &attrs)) == 0)
1296 			nmp->nm_state |= NFSSTA_GOTFSINFO;
1297 
1298 	    }
1299 	} else {
1300 	    /*
1301 	     * The root vnode is usually cached by the namecache so do not
1302 	     * try to avoid going over the wire even if we have previous
1303 	     * information cached.  A stale NFS mount can loop
1304 	     * forever resolving the root vnode if we return no-error when
1305 	     * there is in fact an error.
1306 	     */
1307 	    np->n_attrstamp = 0;
1308 	    error = VOP_GETATTR(vp, &attrs);
1309 	}
1310 	if (vp->v_type == VNON)
1311 	    nfs_setvtype(vp, VDIR);
1312 	vsetflags(vp, VROOT);
1313 	if (error)
1314 		vput(vp);
1315 	else
1316 		*vpp = vp;
1317 	lwkt_reltoken(&nmp->nm_token);
1318 	return (error);
1319 }
1320 
1321 struct scaninfo {
1322 	int rescan;
1323 	int waitfor;
1324 	int allerror;
1325 };
1326 
1327 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1328 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1329 
1330 /*
1331  * Flush out the buffer cache
1332  */
1333 /* ARGSUSED */
1334 static int
1335 nfs_sync(struct mount *mp, int waitfor)
1336 {
1337 	struct nfsmount *nmp = VFSTONFS(mp);
1338 	struct scaninfo scaninfo;
1339 	int error;
1340 
1341 	scaninfo.rescan = 1;
1342 	scaninfo.waitfor = waitfor;
1343 	scaninfo.allerror = 0;
1344 
1345 	/*
1346 	 * Force stale buffer cache information to be flushed.
1347 	 */
1348 	lwkt_gettoken(&nmp->nm_token);
1349 	error = 0;
1350 	while (error == 0 && scaninfo.rescan) {
1351 		scaninfo.rescan = 0;
1352 		error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1353 					nfs_sync_scan2, &scaninfo);
1354 	}
1355 	lwkt_reltoken(&nmp->nm_token);
1356 	return(error);
1357 }
1358 
1359 static int
1360 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1361 {
1362     struct scaninfo *info = data;
1363 
1364     if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree))
1365 	return(-1);
1366     if (info->waitfor & MNT_LAZY)
1367 	return(-1);
1368     return(0);
1369 }
1370 
1371 static int
1372 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1373 {
1374     struct scaninfo *info = data;
1375     int error;
1376 
1377     error = VOP_FSYNC(vp, info->waitfor, 0);
1378     if (error)
1379 	info->allerror = error;
1380     return(0);
1381 }
1382 
1383