xref: /dragonfly/sys/vfs/nfs/nfs_vfsops.c (revision 8bf5b238)
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. 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 | VFCF_MPSAFE);
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 = roundup2(pref, NFS_FABLKSIZE);
503 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
504 		if (max < nmp->nm_wsize && max > 0) {
505 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
506 			if (nmp->nm_wsize == 0)
507 				nmp->nm_wsize = max;
508 		}
509 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
510 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
511 			nmp->nm_rsize = roundup2(pref, NFS_FABLKSIZE);
512 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
513 		if (max < nmp->nm_rsize && max > 0) {
514 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
515 			if (nmp->nm_rsize == 0)
516 				nmp->nm_rsize = max;
517 		}
518 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
519 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
520 			nmp->nm_readdirsize = roundup2(pref, NFS_DIRBLKSIZ);
521 		if (max < nmp->nm_readdirsize && max > 0) {
522 			nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
523 			if (nmp->nm_readdirsize == 0)
524 				nmp->nm_readdirsize = max;
525 		}
526 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
527 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
528 			nmp->nm_maxfilesize = maxfsize;
529 		nmp->nm_state |= NFSSTA_GOTFSINFO;
530 
531 		/*
532 		 * Use the smaller of rsize/wsize for the biosize.
533 		 */
534 		if (nmp->nm_rsize < nmp->nm_wsize)
535 			nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_rsize;
536 		else
537 			nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_wsize;
538 	}
539 	m_freem(info.mrep);
540 	info.mrep = NULL;
541 nfsmout:
542 	return (error);
543 }
544 
545 /*
546  * Mount a remote root fs via. nfs. This depends on the info in the
547  * nfs_diskless structure that has been filled in properly by some primary
548  * bootstrap.
549  * It goes something like this:
550  * - do enough of "ifconfig" by calling ifioctl() so that the system
551  *   can talk to the server
552  * - If nfs_diskless.mygateway is filled in, use that address as
553  *   a default gateway.
554  * - build the rootfs mount point and call mountnfs() to do the rest.
555  */
556 int
557 nfs_mountroot(struct mount *mp)
558 {
559 	struct mount  *swap_mp;
560 	struct nfsv3_diskless *nd = &nfsv3_diskless;
561 	struct socket *so;
562 	struct vnode *vp;
563 	struct thread *td = curthread;		/* XXX */
564 	int error, i;
565 	u_long l;
566 	char buf[128], addr[INET_ADDRSTRLEN];
567 
568 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
569 	bootpc_init();		/* use bootp to get nfs_diskless filled in */
570 #endif
571 
572 	/*
573 	 * XXX time must be non-zero when we init the interface or else
574 	 * the arp code will wedge...
575 	 */
576 	while (mycpu->gd_time_seconds == 0)
577 		tsleep(mycpu, 0, "arpkludge", 10);
578 
579 	/*
580 	 * The boot code may have passed us a diskless structure.
581 	 */
582 	kprintf("DISKLESS %d\n", nfs_diskless_valid);
583 	if (nfs_diskless_valid == 1)
584 		nfs_convert_diskless();
585 
586 	/*
587 	 * NFSv3 is required.
588 	 */
589 	nd->root_args.flags |= NFSMNT_NFSV3 | NFSMNT_RDIRPLUS;
590 	nd->swap_args.flags |= NFSMNT_NFSV3;
591 
592 #define SINP(sockaddr)	((struct sockaddr_in *)(sockaddr))
593 	kprintf("nfs_mountroot: interface %s ip %s",
594 		nd->myif.ifra_name,
595 		kinet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr, addr));
596 	kprintf(" bcast %s",
597 		kinet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr, addr));
598 	kprintf(" mask %s\n",
599 		kinet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr, addr));
600 #undef SINP
601 
602 	/*
603 	 * XXX splnet, so networks will receive...
604 	 */
605 	crit_enter();
606 
607 	/*
608 	 * BOOTP does not necessarily have to be compiled into the kernel
609 	 * for an NFS root to work.  If we inherited the network
610 	 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
611 	 * out our interface for us and all we need to do is ifconfig the
612 	 * interface.  We only do this if the interface has not already been
613 	 * ifconfig'd by e.g. BOOTP.
614 	 */
615 	error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
616 	if (error) {
617 		panic("nfs_mountroot: socreate(%04x): %d",
618 			nd->myif.ifra_addr.sa_family, error);
619 	}
620 
621 	error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
622 	if (error)
623 		panic("nfs_mountroot: SIOCAIFADDR: %d", error);
624 
625 	soclose(so, FNONBLOCK);
626 
627 	/*
628 	 * If the gateway field is filled in, set it as the default route.
629 	 */
630 	if (nd->mygateway.sin_len != 0) {
631 		struct sockaddr_in mask, sin;
632 
633 		bzero((caddr_t)&mask, sizeof(mask));
634 		sin = mask;
635 		sin.sin_family = AF_INET;
636 		sin.sin_len = sizeof(sin);
637 		kprintf("nfs_mountroot: gateway %s\n",
638 			kinet_ntoa(nd->mygateway.sin_addr, addr));
639 		error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
640 					(struct sockaddr *)&nd->mygateway,
641 					(struct sockaddr *)&mask,
642 					RTF_UP | RTF_GATEWAY);
643 		if (error)
644 			kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
645 	}
646 
647 	/*
648 	 * Create the rootfs mount point.
649 	 */
650 	nd->root_args.fh = nd->root_fh;
651 	nd->root_args.fhsize = nd->root_fhsize;
652 	l = ntohl(nd->root_saddr.sin_addr.s_addr);
653 	ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
654 		(l >> 24) & 0xff, (l >> 16) & 0xff,
655 		(l >>  8) & 0xff, (l >>  0) & 0xff,nd->root_hostnam);
656 	kprintf("NFS_ROOT: %s\n",buf);
657 	error = nfs_mountdiskless(buf, "/", MNT_RDONLY, &nd->root_saddr,
658 				  &nd->root_args, td, &vp, &mp);
659 	if (error) {
660 		mp->mnt_vfc->vfc_refcount--;
661 		crit_exit();
662 		return (error);
663 	}
664 
665 	swap_mp = NULL;
666 	if (nd->swap_nblks) {
667 
668 		/* Convert to DEV_BSIZE instead of Kilobyte */
669 		nd->swap_nblks *= 2;
670 
671 		/*
672 		 * Create a fake mount point just for the swap vnode so that the
673 		 * swap file can be on a different server from the rootfs.
674 		 */
675 		nd->swap_args.fh = nd->swap_fh;
676 		nd->swap_args.fhsize = nd->swap_fhsize;
677 		l = ntohl(nd->swap_saddr.sin_addr.s_addr);
678 		ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
679 			(l >> 24) & 0xff, (l >> 16) & 0xff,
680 			(l >>  8) & 0xff, (l >>  0) & 0xff,nd->swap_hostnam);
681 		kprintf("NFS SWAP: %s\n",buf);
682 		vp = NULL;	/* avoid gcc warnings */
683 		error = nfs_mountdiskless(buf, "/swap", 0, &nd->swap_saddr,
684 					  &nd->swap_args, td, &vp, &swap_mp);
685 		if (error) {
686 			crit_exit();
687 			return (error);
688 		}
689 		vfs_unbusy(swap_mp);
690 
691 		VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
692 				nd->swap_nblks * DEV_BSIZE ;
693 
694 		/*
695 		 * Since the swap file is not the root dir of a file system,
696 		 * hack it to a regular file.
697 		 */
698 		vclrflags(vp, VROOT);
699 		vref(vp);
700 		nfs_setvtype(vp, VREG);
701 		swaponvp(td, vp, nd->swap_nblks);
702 	}
703 
704 	mp->mnt_flag |= MNT_ROOTFS;
705 
706 	/*
707 	 * This is not really an nfs issue, but it is much easier to
708 	 * set hostname here and then let the "/etc/rc.xxx" files
709 	 * mount the right /var based upon its preset value.
710 	 */
711 	bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
712 	hostname[MAXHOSTNAMELEN - 1] = '\0';
713 	for (i = 0; i < MAXHOSTNAMELEN; i++)
714 		if (hostname[i] == '\0')
715 			break;
716 	inittodr(ntohl(nd->root_time));
717 	crit_exit();
718 	return (0);
719 }
720 
721 /*
722  * Internal version of mount system call for diskless setup.
723  */
724 static int
725 nfs_mountdiskless(char *path, char *which, int mountflag,
726 	struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
727 	struct vnode **vpp, struct mount **mpp)
728 {
729 	struct mount *mp;
730 	struct sockaddr *nam;
731 	int didalloc = 0;
732 	int error;
733 
734 	mp = *mpp;
735 
736 	if (mp == NULL) {
737 		if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
738 			kprintf("nfs_mountroot: NFS not configured");
739 			return (error);
740 		}
741 		didalloc = 1;
742 	}
743 	mp->mnt_kern_flag = 0;
744 	mp->mnt_flag = mountflag;
745 	nam = dup_sockaddr((struct sockaddr *)sin);
746 
747 #if defined(BOOTP) || defined(NFS_ROOT)
748 	if (args->fhsize == 0) {
749 		char *xpath = path;
750 
751 		kprintf("NFS_ROOT: No FH passed from loader, attempting "
752 			"mount rpc...");
753 		while (*xpath && *xpath != ':')
754 			++xpath;
755 		if (*xpath)
756 			++xpath;
757 		args->fhsize = 0;
758 		error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td);
759 		if (error) {
760 			kprintf("failed error %d.\n", error);
761 			goto haderror;
762 		}
763 		kprintf("success!\n");
764 	}
765 #endif
766 
767 	if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
768 #if defined(BOOTP) || defined(NFS_ROOT)
769 haderror:
770 #endif
771 		kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error);
772 		mp->mnt_vfc->vfc_refcount--;
773 		if (didalloc)
774 			kfree(mp, M_MOUNT);
775 		kfree(nam, M_SONAME);
776 		return (error);
777 	}
778 	*mpp = mp;
779 	return (0);
780 }
781 
782 static void
783 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp)
784 {
785 	int adjsock;
786 	int maxio;
787 
788 	crit_enter();
789 	/*
790 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
791 	 * no sense in that context.
792 	 */
793 	if (nmp->nm_sotype == SOCK_STREAM) {
794 		nmp->nm_flag &= ~NFSMNT_NOCONN;
795 		argp->flags &= ~NFSMNT_NOCONN;
796 	}
797 
798 	/*
799 	 * readdirplus is NFSv3 only.
800 	 */
801 	if ((argp->flags & NFSMNT_NFSV3) == 0) {
802 		nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
803 		argp->flags &= ~NFSMNT_RDIRPLUS;
804 	}
805 
806 	/*
807 	 * Re-bind if rsrvd port flag has changed
808 	 */
809 	adjsock = (nmp->nm_flag & NFSMNT_RESVPORT) !=
810 		  (argp->flags & NFSMNT_RESVPORT);
811 
812 	/* Update flags atomically.  Don't change the lock bits. */
813 	nmp->nm_flag = argp->flags | nmp->nm_flag;
814 	crit_exit();
815 
816 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
817 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
818 		if (nmp->nm_timeo < NFS_MINTIMEO)
819 			nmp->nm_timeo = NFS_MINTIMEO;
820 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
821 			nmp->nm_timeo = NFS_MAXTIMEO;
822 	}
823 
824 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
825 		nmp->nm_retry = argp->retrans;
826 		if (nmp->nm_retry > NFS_MAXREXMIT)
827 			nmp->nm_retry = NFS_MAXREXMIT;
828 	}
829 
830 	/*
831 	 * These parameters effect the buffer cache and cannot be changed
832 	 * once we've successfully mounted.
833 	 */
834 	if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
835 		maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, nmp->nm_sotype);
836 
837 		if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
838 			nmp->nm_wsize = argp->wsize;
839 			/* Round down to multiple of blocksize */
840 			nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
841 			if (nmp->nm_wsize <= 0)
842 				nmp->nm_wsize = NFS_FABLKSIZE;
843 		}
844 		if (nmp->nm_wsize > maxio)
845 			nmp->nm_wsize = maxio;
846 		if (nmp->nm_wsize > MAXBSIZE)
847 			nmp->nm_wsize = MAXBSIZE;
848 
849 		if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
850 			nmp->nm_rsize = argp->rsize;
851 			/* Round down to multiple of blocksize */
852 			nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
853 			if (nmp->nm_rsize <= 0)
854 				nmp->nm_rsize = NFS_FABLKSIZE;
855 		}
856 		if (nmp->nm_rsize > maxio)
857 			nmp->nm_rsize = maxio;
858 		if (nmp->nm_rsize > MAXBSIZE)
859 			nmp->nm_rsize = MAXBSIZE;
860 
861 		if ((argp->flags & NFSMNT_READDIRSIZE) &&
862 		    argp->readdirsize > 0) {
863 			nmp->nm_readdirsize = argp->readdirsize;
864 		}
865 		if (nmp->nm_readdirsize > maxio)
866 			nmp->nm_readdirsize = maxio;
867 		if (nmp->nm_readdirsize > nmp->nm_rsize)
868 			nmp->nm_readdirsize = nmp->nm_rsize;
869 	}
870 
871 	if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
872 		nmp->nm_acregmin = argp->acregmin;
873 	else
874 		nmp->nm_acregmin = NFS_MINATTRTIMO;
875 	if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
876 		nmp->nm_acregmax = argp->acregmax;
877 	else
878 		nmp->nm_acregmax = NFS_MAXATTRTIMO;
879 	if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
880 		nmp->nm_acdirmin = argp->acdirmin;
881 	else
882 		nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
883 	if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
884 		nmp->nm_acdirmax = argp->acdirmax;
885 	else
886 		nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
887 	if (nmp->nm_acdirmin > nmp->nm_acdirmax)
888 		nmp->nm_acdirmin = nmp->nm_acdirmax;
889 	if (nmp->nm_acregmin > nmp->nm_acregmax)
890 		nmp->nm_acregmin = nmp->nm_acregmax;
891 
892 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
893 		if (argp->maxgrouplist <= NFS_MAXGRPS)
894 			nmp->nm_numgrps = argp->maxgrouplist;
895 		else
896 			nmp->nm_numgrps = NFS_MAXGRPS;
897 	}
898 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
899 		if (argp->readahead <= NFS_MAXRAHEAD)
900 			nmp->nm_readahead = argp->readahead;
901 		else
902 			nmp->nm_readahead = NFS_MAXRAHEAD;
903 	}
904 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
905 		if (argp->deadthresh <= NFS_NEVERDEAD)
906 			nmp->nm_deadthresh = argp->deadthresh;
907 		else
908 			nmp->nm_deadthresh = NFS_NEVERDEAD;
909 	}
910 
911 	if (nmp->nm_so && adjsock) {
912 		nfs_safedisconnect(nmp);
913 		if (nmp->nm_sotype == SOCK_DGRAM)
914 			while (nfs_connect(nmp, NULL)) {
915 				kprintf("nfs_args: retrying connect\n");
916 				(void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
917 			}
918 	}
919 }
920 
921 /*
922  * VFS Operations.
923  *
924  * mount system call
925  * It seems a bit dumb to copyinstr() the host and path here and then
926  * bcopy() them in mountnfs(), but I wanted to detect errors before
927  * doing the sockargs() call because sockargs() allocates an mbuf and
928  * an error after that means that I have to release the mbuf.
929  */
930 /* ARGSUSED */
931 static int
932 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
933 {
934 	int error;
935 	struct nfs_args args;
936 	struct sockaddr *nam;
937 	struct vnode *vp;
938 	char pth[MNAMELEN], hst[MNAMELEN];
939 	size_t len;
940 	u_char nfh[NFSX_V3FHMAX];
941 
942 	if (path == NULL) {
943 		nfs_mountroot(mp);
944 		return (0);
945 	}
946 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
947 	if (error)
948 		return (error);
949 	if (args.version != NFS_ARGSVERSION) {
950 #ifdef COMPAT_PRELITE2
951 		/*
952 		 * If the argument version is unknown, then assume the
953 		 * caller is a pre-lite2 4.4BSD client and convert its
954 		 * arguments.
955 		 */
956 		struct onfs_args oargs;
957 		error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
958 		if (error)
959 			return (error);
960 		nfs_convert_oargs(&args,&oargs);
961 #else /* !COMPAT_PRELITE2 */
962 		return (EPROGMISMATCH);
963 #endif /* COMPAT_PRELITE2 */
964 	}
965 	if (mp->mnt_flag & MNT_UPDATE) {
966 		struct nfsmount *nmp = VFSTONFS(mp);
967 
968 		if (nmp == NULL)
969 			return (EIO);
970 		/*
971 		 * When doing an update, we can't change from or to
972 		 * v3, or change cookie translation, or rsize or wsize.
973 		 */
974 		args.flags &= ~(NFSMNT_NFSV3 | NFSMNT_RSIZE | NFSMNT_WSIZE);
975 		args.flags |= nmp->nm_flag & (NFSMNT_NFSV3);
976 		nfs_decode_args(nmp, &args);
977 		return (0);
978 	}
979 
980 	/*
981 	 * Make the nfs_ip_paranoia sysctl serve as the default connection
982 	 * or no-connection mode for those protocols that support
983 	 * no-connection mode (the flag will be cleared later for protocols
984 	 * that do not support no-connection mode).  This will allow a client
985 	 * to receive replies from a different IP then the request was
986 	 * sent to.  Note: default value for nfs_ip_paranoia is 1 (paranoid),
987 	 * not 0.
988 	 */
989 	if (nfs_ip_paranoia == 0)
990 		args.flags |= NFSMNT_NOCONN;
991 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
992 		return (EINVAL);
993 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
994 	if (error)
995 		return (error);
996 	error = copyinstr(path, pth, MNAMELEN-1, &len);
997 	if (error)
998 		return (error);
999 	bzero(&pth[len], MNAMELEN - len);
1000 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
1001 	if (error)
1002 		return (error);
1003 	bzero(&hst[len], MNAMELEN - len);
1004 	/* sockargs() call must be after above copyin() calls */
1005 	error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
1006 	if (error)
1007 		return (error);
1008 	args.fh = nfh;
1009 	error = mountnfs(&args, mp, nam, pth, hst, &vp);
1010 	return (error);
1011 }
1012 
1013 /*
1014  * Common code for mount and mountroot
1015  */
1016 static int
1017 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
1018 	char *pth, char *hst, struct vnode **vpp)
1019 {
1020 	struct nfsmount *nmp;
1021 	struct nfsnode *np;
1022 	int error;
1023 	int rxcpu;
1024 	int txcpu;
1025 
1026 	if (mp->mnt_flag & MNT_UPDATE) {
1027 		nmp = VFSTONFS(mp);
1028 		/* update paths, file handles, etc, here	XXX */
1029 		kfree(nam, M_SONAME);
1030 		return (0);
1031 	} else {
1032 		nmp = objcache_get(nfsmount_objcache, M_WAITOK);
1033 		bzero((caddr_t)nmp, sizeof (struct nfsmount));
1034 		mtx_init_flags(&nmp->nm_rxlock, "nfsrx", MTXF_NOCOLLSTATS);
1035 		mtx_init_flags(&nmp->nm_txlock, "nfstx", MTXF_NOCOLLSTATS);
1036 		TAILQ_INIT(&nmp->nm_uidlruhead);
1037 		TAILQ_INIT(&nmp->nm_bioq);
1038 		TAILQ_INIT(&nmp->nm_reqq);
1039 		TAILQ_INIT(&nmp->nm_reqtxq);
1040 		TAILQ_INIT(&nmp->nm_reqrxq);
1041 		mp->mnt_data = (qaddr_t)nmp;
1042 		lwkt_token_init(&nmp->nm_token, "nfs_token");
1043 	}
1044 	vfs_getnewfsid(mp);
1045 	nmp->nm_mountp = mp;
1046 	mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
1047 	mp->mnt_kern_flag |= MNTK_THR_SYNC;	/* new vsyncscan semantics */
1048 
1049 	lwkt_gettoken(&nmp->nm_token);
1050 
1051 	/*
1052 	 * V2 can only handle 32 bit filesizes.  A 4GB-1 limit may be too
1053 	 * high, depending on whether we end up with negative offsets in
1054 	 * the client or server somewhere.  2GB-1 may be safer.
1055 	 *
1056 	 * For V3, nfs_fsinfo will adjust this as necessary.  Assume maximum
1057 	 * that we can handle until we find out otherwise.  Note that seek
1058 	 * offsets are signed.
1059 	 */
1060 	if ((argp->flags & NFSMNT_NFSV3) == 0)
1061 		nmp->nm_maxfilesize = 0xffffffffLL;
1062 	else
1063 		nmp->nm_maxfilesize = 0x7fffffffffffffffLL;
1064 
1065 	nmp->nm_timeo = NFS_TIMEO;
1066 	nmp->nm_retry = NFS_RETRANS;
1067 	nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
1068 	nmp->nm_rsize = nmp->nm_wsize;
1069 	nmp->nm_readdirsize = NFS_READDIRSIZE;
1070 	nmp->nm_numgrps = NFS_MAXGRPS;
1071 	nmp->nm_readahead = NFS_DEFRAHEAD;
1072 	nmp->nm_deadthresh = NFS_DEADTHRESH;
1073 	nmp->nm_fhsize = argp->fhsize;
1074 	bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
1075 	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
1076 	nmp->nm_nam = nam;
1077 	/* Set up the sockets and per-host congestion */
1078 	nmp->nm_sotype = argp->sotype;
1079 	nmp->nm_soproto = argp->proto;
1080 	nmp->nm_cred = crhold(proc0.p_ucred);
1081 
1082 	nfs_decode_args(nmp, argp);
1083 
1084 	/*
1085 	 * For Connection based sockets (TCP,...) defer the connect until
1086 	 * the first request, in case the server is not responding.
1087 	 */
1088 	if (nmp->nm_sotype == SOCK_DGRAM &&
1089 		(error = nfs_connect(nmp, NULL)))
1090 		goto bad;
1091 
1092 	/*
1093 	 * This is silly, but it has to be set so that vinifod() works.
1094 	 * We do not want to do an nfs_statfs() here since we can get
1095 	 * stuck on a dead server and we are holding a lock on the mount
1096 	 * point.
1097 	 */
1098 	mp->mnt_stat.f_iosize =
1099 		nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
1100 
1101 	/*
1102 	 * Install vop_ops for our vnops
1103 	 */
1104 	vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
1105 	vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
1106 	vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
1107 
1108 	/*
1109 	 * A reference count is needed on the nfsnode representing the
1110 	 * remote root.  If this object is not persistent, then backward
1111 	 * traversals of the mount point (i.e. "..") will not work if
1112 	 * the nfsnode gets flushed out of the cache. Ufs does not have
1113 	 * this problem, because one can identify root inodes by their
1114 	 * number == UFS_ROOTINO (2).
1115 	 */
1116 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
1117 	if (error)
1118 		goto bad;
1119 	*vpp = NFSTOV(np);
1120 
1121 	/*
1122 	 * Retrieval of mountpoint attributes is delayed until nfs_rot
1123 	 * or nfs_statfs are first called.  This will happen either when
1124 	 * we first traverse the mount point or if somebody does a df(1).
1125 	 *
1126 	 * NFSSTA_GOTFSINFO is used to flag if we have successfully
1127 	 * retrieved mountpoint attributes.  In the case of NFSv3 we
1128 	 * also flag static fsinfo.
1129 	 */
1130 	if (*vpp != NULL)
1131 		(*vpp)->v_type = VNON;
1132 
1133 	/*
1134 	 * Lose the lock but keep the ref.
1135 	 */
1136 	vn_unlock(*vpp);
1137 	lwkt_gettoken(&nfs_token);
1138 	TAILQ_INSERT_TAIL(&nfs_mountq, nmp, nm_entry);
1139 	lwkt_reltoken(&nfs_token);
1140 
1141 	switch(ncpus) {
1142 	case 0:
1143 	case 1:
1144 		rxcpu = 0;
1145 		txcpu = 0;
1146 		break;
1147 	case 2:
1148 		rxcpu = 0;
1149 		txcpu = 1;
1150 		break;
1151 	default:
1152 		rxcpu = -1;
1153 		txcpu = -1;
1154 		break;
1155 	}
1156 
1157 	/*
1158 	 * Start the reader and writer threads.
1159 	 */
1160 	lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread,
1161 		    NULL, 0, rxcpu, "nfsiod_rx");
1162 	lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread,
1163 		    NULL, 0, txcpu, "nfsiod_tx");
1164 	lwkt_reltoken(&nmp->nm_token);
1165 	return (0);
1166 bad:
1167 	nfs_disconnect(nmp);
1168 	lwkt_reltoken(&nmp->nm_token);
1169 	nfs_free_mount(nmp);
1170 	return (error);
1171 }
1172 
1173 /*
1174  * unmount system call
1175  */
1176 static int
1177 nfs_unmount(struct mount *mp, int mntflags)
1178 {
1179 	struct nfsmount *nmp;
1180 	int error, flags = 0;
1181 
1182 	nmp = VFSTONFS(mp);
1183 	lwkt_gettoken(&nmp->nm_token);
1184 	if (mntflags & MNT_FORCE) {
1185 		flags |= FORCECLOSE;
1186 		nmp->nm_flag |= NFSMNT_FORCE;
1187 	}
1188 
1189 	/*
1190 	 * Goes something like this..
1191 	 * - Call vflush() to clear out vnodes for this file system
1192 	 * - Close the socket
1193 	 * - Free up the data structures
1194 	 */
1195 	/* In the forced case, cancel any outstanding requests. */
1196 	if (flags & FORCECLOSE) {
1197 		error = nfs_nmcancelreqs(nmp);
1198 		if (error) {
1199 			kprintf("NFS: %s: Unable to cancel all requests\n",
1200 				mp->mnt_stat.f_mntfromname);
1201 			/* continue anyway */
1202 		}
1203 	}
1204 
1205 	/*
1206 	 * Must handshake with nfs_clientd() if it is active. XXX
1207 	 */
1208 	nmp->nm_state |= NFSSTA_DISMINPROG;
1209 
1210 	/*
1211 	 * We hold 1 extra ref on the root vnode; see comment in mountnfs().
1212 	 *
1213 	 * If this doesn't work and we are doing a forced unmount we continue
1214 	 * anyway.
1215 	 */
1216 	error = vflush(mp, 1, flags);
1217 	if (error) {
1218 		nmp->nm_state &= ~NFSSTA_DISMINPROG;
1219 		if ((flags & FORCECLOSE) == 0) {
1220 			lwkt_reltoken(&nmp->nm_token);
1221 			return (error);
1222 		}
1223 	}
1224 
1225 	/*
1226 	 * We are now committed to the unmount.
1227 	 * For NQNFS, let the server daemon free the nfsmount structure.
1228 	 */
1229 	if (nmp->nm_flag & NFSMNT_KERB)
1230 		nmp->nm_state |= NFSSTA_DISMNT;
1231 	nfssvc_iod_stop1(nmp);
1232 	nfs_disconnect(nmp);
1233 	nfssvc_iod_stop2(nmp);
1234 
1235 	lwkt_gettoken(&nfs_token);
1236 	TAILQ_REMOVE(&nfs_mountq, nmp, nm_entry);
1237 	lwkt_reltoken(&nfs_token);
1238 
1239 	lwkt_reltoken(&nmp->nm_token);
1240 
1241 	if ((nmp->nm_flag & NFSMNT_KERB) == 0) {
1242 		nfs_free_mount(nmp);
1243 	}
1244 	return (0);
1245 }
1246 
1247 void
1248 nfs_free_mount(struct nfsmount *nmp)
1249 {
1250 	if (nmp->nm_cred)  {
1251 		crfree(nmp->nm_cred);
1252 		nmp->nm_cred = NULL;
1253 	}
1254 	if (nmp->nm_nam) {
1255 		kfree(nmp->nm_nam, M_SONAME);
1256 		nmp->nm_nam = NULL;
1257 	}
1258 	objcache_put(nfsmount_objcache, nmp);
1259 }
1260 
1261 /*
1262  * Return root of a filesystem
1263  */
1264 static int
1265 nfs_root(struct mount *mp, struct vnode **vpp)
1266 {
1267 	struct vnode *vp;
1268 	struct nfsmount *nmp;
1269 	struct vattr attrs;
1270 	struct nfsnode *np;
1271 	int error;
1272 
1273 	nmp = VFSTONFS(mp);
1274 	lwkt_gettoken(&nmp->nm_token);
1275 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
1276 	if (error) {
1277 		lwkt_reltoken(&nmp->nm_token);
1278 		return (error);
1279 	}
1280 	vp = NFSTOV(np);
1281 
1282 	/*
1283 	 * Get transfer parameters and root vnode attributes
1284 	 *
1285 	 * NOTE: nfs_fsinfo() is expected to override the default
1286 	 *	 f_iosize we set.
1287 	 */
1288 	if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1289 	    if (nmp->nm_flag & NFSMNT_NFSV3) {
1290 		mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1291 		error = nfs_fsinfo(nmp, vp, curthread);
1292 	    } else {
1293 		if ((error = VOP_GETATTR(vp, &attrs)) == 0)
1294 			nmp->nm_state |= NFSSTA_GOTFSINFO;
1295 
1296 	    }
1297 	} else {
1298 	    /*
1299 	     * The root vnode is usually cached by the namecache so do not
1300 	     * try to avoid going over the wire even if we have previous
1301 	     * information cached.  A stale NFS mount can loop
1302 	     * forever resolving the root vnode if we return no-error when
1303 	     * there is in fact an error.
1304 	     */
1305 	    np->n_attrstamp = 0;
1306 	    error = VOP_GETATTR(vp, &attrs);
1307 	}
1308 	if (vp->v_type == VNON)
1309 	    nfs_setvtype(vp, VDIR);
1310 	vsetflags(vp, VROOT);
1311 	if (error)
1312 		vput(vp);
1313 	else
1314 		*vpp = vp;
1315 	lwkt_reltoken(&nmp->nm_token);
1316 	return (error);
1317 }
1318 
1319 struct scaninfo {
1320 	int rescan;
1321 	int waitfor;
1322 	int allerror;
1323 };
1324 
1325 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1326 
1327 /*
1328  * Flush out the buffer cache
1329  */
1330 /* ARGSUSED */
1331 static int
1332 nfs_sync(struct mount *mp, int waitfor)
1333 {
1334 	struct nfsmount *nmp = VFSTONFS(mp);
1335 	struct scaninfo scaninfo;
1336 	int error;
1337 
1338 	scaninfo.rescan = 1;
1339 	scaninfo.waitfor = waitfor;
1340 	scaninfo.allerror = 0;
1341 
1342 	/*
1343 	 * Force stale buffer cache information to be flushed.
1344 	 */
1345 	lwkt_gettoken(&nmp->nm_token);
1346 	error = 0;
1347 	if ((waitfor & MNT_LAZY) == 0) {
1348 		while (error == 0 && scaninfo.rescan) {
1349 			scaninfo.rescan = 0;
1350 			error = vsyncscan(mp, VMSC_GETVP,
1351 					  nfs_sync_scan2, &scaninfo);
1352 		}
1353 	}
1354 	lwkt_reltoken(&nmp->nm_token);
1355 	return(error);
1356 }
1357 
1358 static int
1359 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1360 {
1361     struct scaninfo *info = data;
1362     int error;
1363 
1364     if (vp->v_type == VNON || vp->v_type == VBAD)
1365 	return(0);
1366     error = VOP_FSYNC(vp, info->waitfor, 0);
1367     if (error)
1368 	info->allerror = error;
1369     return(0);
1370 }
1371 
1372