xref: /dflybsd-src/sys/vfs/nfs/nfs_vfsops.c (revision 90ea502b8c5d21f908cedff6680ee2bc9e74ce74)
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  * $DragonFly: src/sys/vfs/nfs/nfs_vfsops.c,v 1.54 2008/07/31 20:23:40 swildner Exp $
39  */
40 
41 #include "opt_bootp.h"
42 #include "opt_nfsroot.h"
43 
44 #include <sys/param.h>
45 #include <sys/sockio.h>
46 #include <sys/proc.h>
47 #include <sys/vnode.h>
48 #include <sys/fcntl.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/mbuf.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/systm.h>
57 
58 #include <vm/vm.h>
59 #include <vm/vm_extern.h>
60 #include <vm/vm_zone.h>
61 
62 #include <net/if.h>
63 #include <net/route.h>
64 #include <netinet/in.h>
65 
66 #include <sys/thread2.h>
67 #include <sys/mutex2.h>
68 
69 #include "rpcv2.h"
70 #include "nfsproto.h"
71 #include "nfs.h"
72 #include "nfsmount.h"
73 #include "nfsnode.h"
74 #include "xdr_subs.h"
75 #include "nfsm_subs.h"
76 #include "nfsdiskless.h"
77 #include "nfsmountrpc.h"
78 
79 extern int	nfs_mountroot(struct mount *mp);
80 extern void	bootpc_init(void);
81 
82 extern struct vop_ops nfsv2_vnode_vops;
83 extern struct vop_ops nfsv2_fifo_vops;
84 extern struct vop_ops nfsv2_spec_vops;
85 
86 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
87 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
88 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
89 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
90 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
91 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
92 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
93 
94 vm_zone_t nfsmount_zone;
95 
96 struct nfsstats	nfsstats;
97 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
98 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
99 	&nfsstats, nfsstats, "");
100 static int nfs_ip_paranoia = 1;
101 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
102 	&nfs_ip_paranoia, 0, "");
103 #ifdef NFS_DEBUG
104 int nfs_debug;
105 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
106 #endif
107 
108 /*
109  * Tunable to determine the Read/Write unit size.  Maximum value
110  * is NFS_MAXDATA.  We also default to NFS_MAXDATA.
111  */
112 static int nfs_io_size = NFS_MAXDATA;
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
114 	&nfs_io_size, 0, "NFS optimal I/O unit size");
115 
116 static void	nfs_decode_args (struct nfsmount *nmp,
117 			struct nfs_args *argp);
118 static int	mountnfs (struct nfs_args *,struct mount *,
119 			struct sockaddr *,char *,char *,struct vnode **);
120 static int	nfs_mount ( struct mount *mp, char *path, caddr_t data,
121 			struct ucred *cred);
122 static int	nfs_unmount ( struct mount *mp, int mntflags);
123 static int	nfs_root ( struct mount *mp, struct vnode **vpp);
124 static int	nfs_statfs ( struct mount *mp, struct statfs *sbp,
125 			struct ucred *cred);
126 static int	nfs_statvfs(struct mount *mp, struct statvfs *sbp,
127 				struct ucred *cred);
128 static int	nfs_sync ( struct mount *mp, int waitfor);
129 
130 /*
131  * nfs vfs operations.
132  */
133 static struct vfsops nfs_vfsops = {
134 	.vfs_mount =    	nfs_mount,
135 	.vfs_unmount =  	nfs_unmount,
136 	.vfs_root =     	nfs_root,
137 	.vfs_statfs =    	nfs_statfs,
138 	.vfs_statvfs =   	nfs_statvfs,
139 	.vfs_sync =     	nfs_sync,
140 	.vfs_init =     	nfs_init,
141 	.vfs_uninit =    	nfs_uninit
142 };
143 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
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 
157 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
158 	nfsv3_diskless.root_hostnam, 0, "");
159 
160 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
161 	&nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
162 	"%Ssockaddr_in", "");
163 
164 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
165 	nfsv3_diskless.swap_hostnam, 0, "");
166 
167 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
168 	&nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
169 	"%Ssockaddr_in","");
170 
171 
172 void nfsargs_ntoh (struct nfs_args *);
173 static int nfs_mountdiskless (char *, char *, int,
174 				  struct sockaddr_in *, struct nfs_args *,
175 				  struct thread *, struct vnode **,
176 				  struct mount **);
177 static void nfs_convert_diskless (void);
178 static void nfs_convert_oargs (struct nfs_args *args,
179 				   struct onfs_args *oargs);
180 
181 /*
182  * Calculate the buffer I/O block size to use.  The maximum V2 block size
183  * is typically 8K, the maximum datagram size is typically 16K, and the
184  * maximum V3 block size is typically 32K.  The buffer cache tends to work
185  * best with 16K blocks but we allow 32K for TCP connections.
186  *
187  * We force the block size to be at least a page for buffer cache efficiency.
188  */
189 static int
190 nfs_iosize(int v3, int sotype)
191 {
192 	int iosize;
193 	int iomax;
194 
195 	if (v3) {
196 		if (sotype == SOCK_STREAM)
197 			iomax = NFS_MAXDATA;
198 		else
199 			iomax = NFS_MAXDGRAMDATA;
200 	} else {
201 		iomax = NFS_V2MAXDATA;
202 	}
203 	if ((iosize = nfs_io_size) > iomax)
204 		iosize = iomax;
205 	if (iosize < PAGE_SIZE)
206 		iosize = PAGE_SIZE;
207 
208 	/*
209 	 * This is an aweful hack but until the buffer cache is rewritten
210 	 * we need it.  The problem is that when you combine write() with
211 	 * mmap() the vm_page->valid bits can become weird looking
212 	 * (e.g. 0xfc).  This occurs because NFS uses piecemeal buffers
213 	 * at the file EOF.  To solve the problem the BIO system needs to
214 	 * be guarenteed that the NFS iosize for regular files will be a
215 	 * multiple of PAGE_SIZE so it can invalidate the whole page
216 	 * rather then just the piece of it owned by the buffer when
217 	 * NFS does vinvalbuf() calls.
218 	 */
219 	if (iosize & PAGE_MASK)
220 		iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
221 	return iosize;
222 }
223 
224 static void
225 nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
226 {
227 	args->version = NFS_ARGSVERSION;
228 	args->addr = oargs->addr;
229 	args->addrlen = oargs->addrlen;
230 	args->sotype = oargs->sotype;
231 	args->proto = oargs->proto;
232 	args->fh = oargs->fh;
233 	args->fhsize = oargs->fhsize;
234 	args->flags = oargs->flags;
235 	args->wsize = oargs->wsize;
236 	args->rsize = oargs->rsize;
237 	args->readdirsize = oargs->readdirsize;
238 	args->timeo = oargs->timeo;
239 	args->retrans = oargs->retrans;
240 	args->maxgrouplist = oargs->maxgrouplist;
241 	args->readahead = oargs->readahead;
242 	args->deadthresh = oargs->deadthresh;
243 	args->hostname = oargs->hostname;
244 }
245 
246 static void
247 nfs_convert_diskless(void)
248 {
249 	int i;
250 
251 	bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
252 		sizeof(struct ifaliasreq));
253 	bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
254 		sizeof(struct sockaddr_in));
255 	nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
256 
257 	bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
258 	nfsv3_diskless.swap_fhsize = NFSX_V2FH;
259 	for (i = NFSX_V2FH - 1; i >= 0; --i) {
260 		if (nfs_diskless.swap_fh[i])
261 			break;
262 	}
263 	if (i < 0)
264 		nfsv3_diskless.swap_fhsize = 0;
265 
266 	bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
267 		sizeof(struct sockaddr_in));
268 	bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
269 	nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
270 	bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
271 		sizeof(struct ucred));
272 	nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
273 
274 	bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
275 	nfsv3_diskless.root_fhsize = NFSX_V2FH;
276 	for (i = NFSX_V2FH - 1; i >= 0; --i) {
277 		if (nfs_diskless.root_fh[i])
278 			break;
279 	}
280 	if (i < 0)
281 		nfsv3_diskless.root_fhsize = 0;
282 
283 	bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
284 		sizeof(struct sockaddr_in));
285 	bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
286 	nfsv3_diskless.root_time = nfs_diskless.root_time;
287 	bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
288 		MAXHOSTNAMELEN);
289 	nfs_diskless_valid = 3;
290 }
291 
292 /*
293  * nfs statfs call
294  */
295 int
296 nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
297 {
298 	struct vnode *vp;
299 	struct nfs_statfs *sfp;
300 	struct nfsmount *nmp = VFSTONFS(mp);
301 	thread_t td = curthread;
302 	int error = 0, retattr;
303 	struct nfsnode *np;
304 	u_quad_t tquad;
305 	struct nfsm_info info;
306 
307 	info.mrep = NULL;
308 	info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
309 
310 #ifndef nolint
311 	sfp = NULL;
312 #endif
313 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
314 	if (error)
315 		return (error);
316 	vp = NFSTOV(np);
317 	/* ignore the passed cred */
318 	cred = crget();
319 	cred->cr_ngroups = 1;
320 	if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
321 		(void)nfs_fsinfo(nmp, vp, td);
322 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
323 	nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
324 	ERROROUT(nfsm_fhtom(&info, vp));
325 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
326 	if (info.v3) {
327 		ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
328 					 NFS_LATTR_NOSHRINK));
329 	}
330 	if (error) {
331 		if (info.mrep != NULL)
332 			m_freem(info.mrep);
333 		goto nfsmout;
334 	}
335 	NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
336 	sbp->f_flags = nmp->nm_flag;
337 
338 	if (info.v3) {
339 		sbp->f_bsize = NFS_FABLKSIZE;
340 		tquad = fxdr_hyper(&sfp->sf_tbytes);
341 		sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
342 		tquad = fxdr_hyper(&sfp->sf_fbytes);
343 		sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
344 		tquad = fxdr_hyper(&sfp->sf_abytes);
345 		sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
346 		sbp->f_files = (fxdr_unsigned(int32_t,
347 		    sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
348 		sbp->f_ffree = (fxdr_unsigned(int32_t,
349 		    sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
350 	} else {
351 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
352 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
353 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
354 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
355 		sbp->f_files = 0;
356 		sbp->f_ffree = 0;
357 	}
358 
359 	/*
360 	 * Some values are pre-set in mnt_stat.  Note in particular f_iosize
361 	 * cannot be changed once the filesystem is mounted as it is used
362 	 * as the basis for BIOs.
363 	 */
364 	if (sbp != &mp->mnt_stat) {
365 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
366 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
367 		sbp->f_iosize = mp->mnt_stat.f_iosize;
368 	}
369 	m_freem(info.mrep);
370 	info.mrep = NULL;
371 nfsmout:
372 	vput(vp);
373 	crfree(cred);
374 	return (error);
375 }
376 
377 static int
378 nfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
379 {
380 	struct vnode *vp;
381 	struct nfs_statfs *sfp;
382 	struct nfsmount *nmp = VFSTONFS(mp);
383 	thread_t td = curthread;
384 	int error = 0, retattr;
385 	struct nfsnode *np;
386 	struct nfsm_info info;
387 
388 	info.mrep = NULL;
389 	info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
390 
391 #ifndef nolint
392 	sfp = NULL;
393 #endif
394 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
395 	if (error)
396 		return (error);
397 	vp = NFSTOV(np);
398 	/* ignore the passed cred */
399 	cred = crget();
400 	cred->cr_ngroups = 1;
401 	if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
402 		(void)nfs_fsinfo(nmp, vp, td);
403 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
404 	nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
405 	ERROROUT(nfsm_fhtom(&info, vp));
406 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
407 	if (info.v3) {
408 		ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
409 					 NFS_LATTR_NOSHRINK));
410 	}
411 	if (error) {
412 		if (info.mrep != NULL)
413 			m_freem(info.mrep);
414 		goto nfsmout;
415 	}
416 	NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
417 	sbp->f_flag = nmp->nm_flag;
418 	sbp->f_owner = nmp->nm_cred->cr_ruid;
419 
420 	if (info.v3) {
421 		sbp->f_bsize = NFS_FABLKSIZE;
422 		sbp->f_frsize = NFS_FABLKSIZE;
423 		sbp->f_blocks = (fxdr_hyper(&sfp->sf_tbytes) /
424 				((u_quad_t)NFS_FABLKSIZE));
425 		sbp->f_bfree = (fxdr_hyper(&sfp->sf_fbytes) /
426 				((u_quad_t)NFS_FABLKSIZE));
427 		sbp->f_bavail = (fxdr_hyper(&sfp->sf_abytes) /
428 				((u_quad_t)NFS_FABLKSIZE));
429 		sbp->f_files = fxdr_hyper(&sfp->sf_tfiles);
430 		sbp->f_ffree = fxdr_hyper(&sfp->sf_ffiles);
431 		sbp->f_favail = fxdr_hyper(&sfp->sf_afiles);
432 	} else {
433 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
434 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
435 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
436 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
437 		sbp->f_files = 0;
438 		sbp->f_ffree = 0;
439 		sbp->f_favail = 0;
440 	}
441 	sbp->f_syncreads = 0;
442 	sbp->f_syncwrites = 0;
443 	sbp->f_asyncreads = 0;
444 	sbp->f_asyncwrites = 0;
445 	sbp->f_type = mp->mnt_vfc->vfc_typenum;
446 
447 	m_freem(info.mrep);
448 	info.mrep = NULL;
449 nfsmout:
450 	vput(vp);
451 	crfree(cred);
452 	return (error);
453 }
454 
455 /*
456  * nfs version 3 fsinfo rpc call
457  */
458 int
459 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
460 {
461 	struct nfsv3_fsinfo *fsp;
462 	u_int32_t pref, max;
463 	int error = 0, retattr;
464 	u_int64_t maxfsize;
465 	struct nfsm_info info;
466 
467 	info.v3 = 1;
468 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
469 	nfsm_reqhead(&info, vp, NFSPROC_FSINFO, NFSX_FH(1));
470 	ERROROUT(nfsm_fhtom(&info, vp));
471 	NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSINFO, td,
472 				nfs_vpcred(vp, ND_READ), &error));
473 	ERROROUT(nfsm_postop_attr(&info, vp, &retattr, NFS_LATTR_NOSHRINK));
474 	if (error == 0) {
475 		NULLOUT(fsp = nfsm_dissect(&info, NFSX_V3FSINFO));
476 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
477 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
478 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
479 				~(NFS_FABLKSIZE - 1);
480 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
481 		if (max < nmp->nm_wsize && max > 0) {
482 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
483 			if (nmp->nm_wsize == 0)
484 				nmp->nm_wsize = max;
485 		}
486 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
487 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
488 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
489 				~(NFS_FABLKSIZE - 1);
490 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
491 		if (max < nmp->nm_rsize && max > 0) {
492 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
493 			if (nmp->nm_rsize == 0)
494 				nmp->nm_rsize = max;
495 		}
496 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
497 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
498 			nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
499 				~(NFS_DIRBLKSIZ - 1);
500 		if (max < nmp->nm_readdirsize && max > 0) {
501 			nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
502 			if (nmp->nm_readdirsize == 0)
503 				nmp->nm_readdirsize = max;
504 		}
505 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
506 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
507 			nmp->nm_maxfilesize = maxfsize;
508 		nmp->nm_state |= NFSSTA_GOTFSINFO;
509 
510 		/*
511 		 * Use the smaller of rsize/wsize for the biosize.
512 		 */
513 		if (nmp->nm_rsize < nmp->nm_wsize)
514 			nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_rsize;
515 		else
516 			nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_wsize;
517 	}
518 	m_freem(info.mrep);
519 	info.mrep = NULL;
520 nfsmout:
521 	return (error);
522 }
523 
524 /*
525  * Mount a remote root fs via. nfs. This depends on the info in the
526  * nfs_diskless structure that has been filled in properly by some primary
527  * bootstrap.
528  * It goes something like this:
529  * - do enough of "ifconfig" by calling ifioctl() so that the system
530  *   can talk to the server
531  * - If nfs_diskless.mygateway is filled in, use that address as
532  *   a default gateway.
533  * - build the rootfs mount point and call mountnfs() to do the rest.
534  */
535 int
536 nfs_mountroot(struct mount *mp)
537 {
538 	struct mount  *swap_mp;
539 	struct nfsv3_diskless *nd = &nfsv3_diskless;
540 	struct socket *so;
541 	struct vnode *vp;
542 	struct thread *td = curthread;		/* XXX */
543 	int error, i;
544 	u_long l;
545 	char buf[128];
546 
547 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
548 	bootpc_init();		/* use bootp to get nfs_diskless filled in */
549 #endif
550 
551 	/*
552 	 * XXX time must be non-zero when we init the interface or else
553 	 * the arp code will wedge...
554 	 */
555 	while (mycpu->gd_time_seconds == 0)
556 		tsleep(mycpu, 0, "arpkludge", 10);
557 
558 	/*
559 	 * The boot code may have passed us a diskless structure.
560 	 */
561 	if (nfs_diskless_valid == 1)
562 		nfs_convert_diskless();
563 
564 #define SINP(sockaddr)	((struct sockaddr_in *)(sockaddr))
565 	kprintf("nfs_mountroot: interface %s ip %s",
566 		nd->myif.ifra_name,
567 		inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
568 	kprintf(" bcast %s",
569 		inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
570 	kprintf(" mask %s\n",
571 		inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
572 #undef SINP
573 
574 	/*
575 	 * XXX splnet, so networks will receive...
576 	 */
577 	crit_enter();
578 
579 	/*
580 	 * BOOTP does not necessarily have to be compiled into the kernel
581 	 * for an NFS root to work.  If we inherited the network
582 	 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
583 	 * out our interface for us and all we need to do is ifconfig the
584 	 * interface.  We only do this if the interface has not already been
585 	 * ifconfig'd by e.g. BOOTP.
586 	 */
587 	error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
588 	if (error) {
589 		panic("nfs_mountroot: socreate(%04x): %d",
590 			nd->myif.ifra_addr.sa_family, error);
591 	}
592 
593 	error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
594 	if (error)
595 		panic("nfs_mountroot: SIOCAIFADDR: %d", error);
596 
597 	soclose(so, FNONBLOCK);
598 
599 	/*
600 	 * If the gateway field is filled in, set it as the default route.
601 	 */
602 	if (nd->mygateway.sin_len != 0) {
603 		struct sockaddr_in mask, sin;
604 
605 		bzero((caddr_t)&mask, sizeof(mask));
606 		sin = mask;
607 		sin.sin_family = AF_INET;
608 		sin.sin_len = sizeof(sin);
609 		kprintf("nfs_mountroot: gateway %s\n",
610 			inet_ntoa(nd->mygateway.sin_addr));
611 		error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
612 					(struct sockaddr *)&nd->mygateway,
613 					(struct sockaddr *)&mask,
614 					RTF_UP | RTF_GATEWAY);
615 		if (error)
616 			kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
617 	}
618 
619 	/*
620 	 * Create the rootfs mount point.
621 	 */
622 	nd->root_args.fh = nd->root_fh;
623 	nd->root_args.fhsize = nd->root_fhsize;
624 	l = ntohl(nd->root_saddr.sin_addr.s_addr);
625 	ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
626 		(l >> 24) & 0xff, (l >> 16) & 0xff,
627 		(l >>  8) & 0xff, (l >>  0) & 0xff,nd->root_hostnam);
628 	kprintf("NFS_ROOT: %s\n",buf);
629 	if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
630 	    &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
631 		mp->mnt_vfc->vfc_refcount--;
632 		crit_exit();
633 		return (error);
634 	}
635 
636 	swap_mp = NULL;
637 	if (nd->swap_nblks) {
638 
639 		/* Convert to DEV_BSIZE instead of Kilobyte */
640 		nd->swap_nblks *= 2;
641 
642 		/*
643 		 * Create a fake mount point just for the swap vnode so that the
644 		 * swap file can be on a different server from the rootfs.
645 		 */
646 		nd->swap_args.fh = nd->swap_fh;
647 		nd->swap_args.fhsize = nd->swap_fhsize;
648 		l = ntohl(nd->swap_saddr.sin_addr.s_addr);
649 		ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
650 			(l >> 24) & 0xff, (l >> 16) & 0xff,
651 			(l >>  8) & 0xff, (l >>  0) & 0xff,nd->swap_hostnam);
652 		kprintf("NFS SWAP: %s\n",buf);
653 		if ((error = nfs_mountdiskless(buf, "/swap", 0,
654 		    &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) {
655 			crit_exit();
656 			return (error);
657 		}
658 		vfs_unbusy(swap_mp);
659 
660 		VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
661 				nd->swap_nblks * DEV_BSIZE ;
662 
663 		/*
664 		 * Since the swap file is not the root dir of a file system,
665 		 * hack it to a regular file.
666 		 */
667 		vp->v_flag &= ~VROOT;
668 		vref(vp);
669 		nfs_setvtype(vp, VREG);
670 		swaponvp(td, vp, nd->swap_nblks);
671 	}
672 
673 	mp->mnt_flag |= MNT_ROOTFS;
674 	vfs_unbusy(mp);
675 
676 	/*
677 	 * This is not really an nfs issue, but it is much easier to
678 	 * set hostname here and then let the "/etc/rc.xxx" files
679 	 * mount the right /var based upon its preset value.
680 	 */
681 	bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
682 	hostname[MAXHOSTNAMELEN - 1] = '\0';
683 	for (i = 0; i < MAXHOSTNAMELEN; i++)
684 		if (hostname[i] == '\0')
685 			break;
686 	inittodr(ntohl(nd->root_time));
687 	crit_exit();
688 	return (0);
689 }
690 
691 /*
692  * Internal version of mount system call for diskless setup.
693  */
694 static int
695 nfs_mountdiskless(char *path, char *which, int mountflag,
696 	struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
697 	struct vnode **vpp, struct mount **mpp)
698 {
699 	struct mount *mp;
700 	struct sockaddr *nam;
701 	int didalloc = 0;
702 	int error;
703 
704 	mp = *mpp;
705 
706 	if (mp == NULL) {
707 		if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
708 			kprintf("nfs_mountroot: NFS not configured");
709 			return (error);
710 		}
711 		didalloc = 1;
712 	}
713 	mp->mnt_kern_flag = 0;
714 	mp->mnt_flag = mountflag;
715 	nam = dup_sockaddr((struct sockaddr *)sin);
716 
717 #if defined(BOOTP) || defined(NFS_ROOT)
718 	if (args->fhsize == 0) {
719 		char *xpath = path;
720 
721 		kprintf("NFS_ROOT: No FH passed from loader, attempting mount rpc...");
722 		while (*xpath && *xpath != ':')
723 			++xpath;
724 		if (*xpath)
725 			++xpath;
726 		args->fhsize = 0;
727 		error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td);
728 		if (error) {
729 			kprintf("failed error %d.\n", error);
730 			goto haderror;
731 		}
732 		kprintf("success!\n");
733 	}
734 #endif
735 
736 	if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
737 #if defined(BOOTP) || defined(NFS_ROOT)
738 haderror:
739 #endif
740 		kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error);
741 		mp->mnt_vfc->vfc_refcount--;
742 		vfs_unbusy(mp);
743 		if (didalloc)
744 			kfree(mp, M_MOUNT);
745 		FREE(nam, M_SONAME);
746 		return (error);
747 	}
748 	*mpp = mp;
749 	return (0);
750 }
751 
752 static void
753 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp)
754 {
755 	int adjsock;
756 	int maxio;
757 
758 	crit_enter();
759 	/*
760 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
761 	 * no sense in that context.
762 	 */
763 	if (nmp->nm_sotype == SOCK_STREAM)
764 		nmp->nm_flag &= ~NFSMNT_NOCONN;
765 
766 	/* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
767 	if ((argp->flags & NFSMNT_NFSV3) == 0)
768 		nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
769 
770 	/*
771 	 * Re-bind if rsrvd port flag has changed
772 	 */
773 	adjsock = (nmp->nm_flag & NFSMNT_RESVPORT) !=
774 		  (argp->flags & NFSMNT_RESVPORT);
775 
776 	/* Update flags atomically.  Don't change the lock bits. */
777 	nmp->nm_flag = argp->flags | nmp->nm_flag;
778 	crit_exit();
779 
780 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
781 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
782 		if (nmp->nm_timeo < NFS_MINTIMEO)
783 			nmp->nm_timeo = NFS_MINTIMEO;
784 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
785 			nmp->nm_timeo = NFS_MAXTIMEO;
786 	}
787 
788 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
789 		nmp->nm_retry = argp->retrans;
790 		if (nmp->nm_retry > NFS_MAXREXMIT)
791 			nmp->nm_retry = NFS_MAXREXMIT;
792 	}
793 
794 	/*
795 	 * These parameters effect the buffer cache and cannot be changed
796 	 * once we've successfully mounted.
797 	 */
798 	if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
799 		maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, nmp->nm_sotype);
800 
801 		if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
802 			nmp->nm_wsize = argp->wsize;
803 			/* Round down to multiple of blocksize */
804 			nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
805 			if (nmp->nm_wsize <= 0)
806 				nmp->nm_wsize = NFS_FABLKSIZE;
807 		}
808 		if (nmp->nm_wsize > maxio)
809 			nmp->nm_wsize = maxio;
810 		if (nmp->nm_wsize > MAXBSIZE)
811 			nmp->nm_wsize = MAXBSIZE;
812 
813 		if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
814 			nmp->nm_rsize = argp->rsize;
815 			/* Round down to multiple of blocksize */
816 			nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
817 			if (nmp->nm_rsize <= 0)
818 				nmp->nm_rsize = NFS_FABLKSIZE;
819 		}
820 		if (nmp->nm_rsize > maxio)
821 			nmp->nm_rsize = maxio;
822 		if (nmp->nm_rsize > MAXBSIZE)
823 			nmp->nm_rsize = MAXBSIZE;
824 
825 		if ((argp->flags & NFSMNT_READDIRSIZE) &&
826 		    argp->readdirsize > 0) {
827 			nmp->nm_readdirsize = argp->readdirsize;
828 		}
829 		if (nmp->nm_readdirsize > maxio)
830 			nmp->nm_readdirsize = maxio;
831 		if (nmp->nm_readdirsize > nmp->nm_rsize)
832 			nmp->nm_readdirsize = nmp->nm_rsize;
833 	}
834 
835 	if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
836 		nmp->nm_acregmin = argp->acregmin;
837 	else
838 		nmp->nm_acregmin = NFS_MINATTRTIMO;
839 	if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
840 		nmp->nm_acregmax = argp->acregmax;
841 	else
842 		nmp->nm_acregmax = NFS_MAXATTRTIMO;
843 	if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
844 		nmp->nm_acdirmin = argp->acdirmin;
845 	else
846 		nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
847 	if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
848 		nmp->nm_acdirmax = argp->acdirmax;
849 	else
850 		nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
851 	if (nmp->nm_acdirmin > nmp->nm_acdirmax)
852 		nmp->nm_acdirmin = nmp->nm_acdirmax;
853 	if (nmp->nm_acregmin > nmp->nm_acregmax)
854 		nmp->nm_acregmin = nmp->nm_acregmax;
855 
856 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
857 		if (argp->maxgrouplist <= NFS_MAXGRPS)
858 			nmp->nm_numgrps = argp->maxgrouplist;
859 		else
860 			nmp->nm_numgrps = NFS_MAXGRPS;
861 	}
862 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
863 		if (argp->readahead <= NFS_MAXRAHEAD)
864 			nmp->nm_readahead = argp->readahead;
865 		else
866 			nmp->nm_readahead = NFS_MAXRAHEAD;
867 	}
868 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
869 		if (argp->deadthresh <= NFS_NEVERDEAD)
870 			nmp->nm_deadthresh = argp->deadthresh;
871 		else
872 			nmp->nm_deadthresh = NFS_NEVERDEAD;
873 	}
874 
875 	if (nmp->nm_so && adjsock) {
876 		nfs_safedisconnect(nmp);
877 		if (nmp->nm_sotype == SOCK_DGRAM)
878 			while (nfs_connect(nmp, NULL)) {
879 				kprintf("nfs_args: retrying connect\n");
880 				(void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
881 			}
882 	}
883 }
884 
885 /*
886  * VFS Operations.
887  *
888  * mount system call
889  * It seems a bit dumb to copyinstr() the host and path here and then
890  * bcopy() them in mountnfs(), but I wanted to detect errors before
891  * doing the sockargs() call because sockargs() allocates an mbuf and
892  * an error after that means that I have to release the mbuf.
893  */
894 /* ARGSUSED */
895 static int
896 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
897 {
898 	int error;
899 	struct nfs_args args;
900 	struct sockaddr *nam;
901 	struct vnode *vp;
902 	char pth[MNAMELEN], hst[MNAMELEN];
903 	size_t len;
904 	u_char nfh[NFSX_V3FHMAX];
905 
906 	if (path == NULL) {
907 		nfs_mountroot(mp);
908 		return (0);
909 	}
910 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
911 	if (error)
912 		return (error);
913 	if (args.version != NFS_ARGSVERSION) {
914 #ifdef COMPAT_PRELITE2
915 		/*
916 		 * If the argument version is unknown, then assume the
917 		 * caller is a pre-lite2 4.4BSD client and convert its
918 		 * arguments.
919 		 */
920 		struct onfs_args oargs;
921 		error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
922 		if (error)
923 			return (error);
924 		nfs_convert_oargs(&args,&oargs);
925 #else /* !COMPAT_PRELITE2 */
926 		return (EPROGMISMATCH);
927 #endif /* COMPAT_PRELITE2 */
928 	}
929 	if (mp->mnt_flag & MNT_UPDATE) {
930 		struct nfsmount *nmp = VFSTONFS(mp);
931 
932 		if (nmp == NULL)
933 			return (EIO);
934 		/*
935 		 * When doing an update, we can't change from or to
936 		 * v3, or change cookie translation, or rsize or wsize.
937 		 */
938 		args.flags &= ~(NFSMNT_NFSV3 | NFSMNT_RSIZE | NFSMNT_WSIZE);
939 		args.flags |= nmp->nm_flag & (NFSMNT_NFSV3);
940 		nfs_decode_args(nmp, &args);
941 		return (0);
942 	}
943 
944 	/*
945 	 * Make the nfs_ip_paranoia sysctl serve as the default connection
946 	 * or no-connection mode for those protocols that support
947 	 * no-connection mode (the flag will be cleared later for protocols
948 	 * that do not support no-connection mode).  This will allow a client
949 	 * to receive replies from a different IP then the request was
950 	 * sent to.  Note: default value for nfs_ip_paranoia is 1 (paranoid),
951 	 * not 0.
952 	 */
953 	if (nfs_ip_paranoia == 0)
954 		args.flags |= NFSMNT_NOCONN;
955 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
956 		return (EINVAL);
957 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
958 	if (error)
959 		return (error);
960 	error = copyinstr(path, pth, MNAMELEN-1, &len);
961 	if (error)
962 		return (error);
963 	bzero(&pth[len], MNAMELEN - len);
964 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
965 	if (error)
966 		return (error);
967 	bzero(&hst[len], MNAMELEN - len);
968 	/* sockargs() call must be after above copyin() calls */
969 	error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
970 	if (error)
971 		return (error);
972 	args.fh = nfh;
973 	error = mountnfs(&args, mp, nam, pth, hst, &vp);
974 	return (error);
975 }
976 
977 /*
978  * Common code for mount and mountroot
979  */
980 static int
981 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
982 	char *pth, char *hst, struct vnode **vpp)
983 {
984 	struct nfsmount *nmp;
985 	struct nfsnode *np;
986 	int error;
987 	int rxcpu;
988 	int txcpu;
989 
990 	if (mp->mnt_flag & MNT_UPDATE) {
991 		nmp = VFSTONFS(mp);
992 		/* update paths, file handles, etc, here	XXX */
993 		FREE(nam, M_SONAME);
994 		return (0);
995 	} else {
996 		nmp = zalloc(nfsmount_zone);
997 		bzero((caddr_t)nmp, sizeof (struct nfsmount));
998 		mtx_init(&nmp->nm_rxlock);
999 		mtx_init(&nmp->nm_txlock);
1000 		TAILQ_INIT(&nmp->nm_uidlruhead);
1001 		TAILQ_INIT(&nmp->nm_bioq);
1002 		TAILQ_INIT(&nmp->nm_reqq);
1003 		TAILQ_INIT(&nmp->nm_reqtxq);
1004 		TAILQ_INIT(&nmp->nm_reqrxq);
1005 		mp->mnt_data = (qaddr_t)nmp;
1006 	}
1007 	vfs_getnewfsid(mp);
1008 	nmp->nm_mountp = mp;
1009 
1010 	/*
1011 	 * V2 can only handle 32 bit filesizes.  A 4GB-1 limit may be too
1012 	 * high, depending on whether we end up with negative offsets in
1013 	 * the client or server somewhere.  2GB-1 may be safer.
1014 	 *
1015 	 * For V3, nfs_fsinfo will adjust this as necessary.  Assume maximum
1016 	 * that we can handle until we find out otherwise.
1017 	 * XXX Our "safe" limit on the client is what we can store in our
1018 	 * buffer cache using signed(!) block numbers.
1019 	 */
1020 	if ((argp->flags & NFSMNT_NFSV3) == 0)
1021 		nmp->nm_maxfilesize = 0xffffffffLL;
1022 	else
1023 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
1024 
1025 	nmp->nm_timeo = NFS_TIMEO;
1026 	nmp->nm_retry = NFS_RETRANS;
1027 	nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
1028 	nmp->nm_rsize = nmp->nm_wsize;
1029 	nmp->nm_readdirsize = NFS_READDIRSIZE;
1030 	nmp->nm_numgrps = NFS_MAXGRPS;
1031 	nmp->nm_readahead = NFS_DEFRAHEAD;
1032 	nmp->nm_deadthresh = NFS_DEADTHRESH;
1033 	nmp->nm_fhsize = argp->fhsize;
1034 	bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
1035 	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
1036 	nmp->nm_nam = nam;
1037 	/* Set up the sockets and per-host congestion */
1038 	nmp->nm_sotype = argp->sotype;
1039 	nmp->nm_soproto = argp->proto;
1040 	nmp->nm_cred = crhold(proc0.p_ucred);
1041 
1042 	nfs_decode_args(nmp, argp);
1043 
1044 	/*
1045 	 * For Connection based sockets (TCP,...) defer the connect until
1046 	 * the first request, in case the server is not responding.
1047 	 */
1048 	if (nmp->nm_sotype == SOCK_DGRAM &&
1049 		(error = nfs_connect(nmp, NULL)))
1050 		goto bad;
1051 
1052 	/*
1053 	 * This is silly, but it has to be set so that vinifod() works.
1054 	 * We do not want to do an nfs_statfs() here since we can get
1055 	 * stuck on a dead server and we are holding a lock on the mount
1056 	 * point.
1057 	 */
1058 	mp->mnt_stat.f_iosize =
1059 		nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
1060 
1061 	/*
1062 	 * Install vop_ops for our vnops
1063 	 */
1064 	vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
1065 	vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
1066 	vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
1067 
1068 	/*
1069 	 * A reference count is needed on the nfsnode representing the
1070 	 * remote root.  If this object is not persistent, then backward
1071 	 * traversals of the mount point (i.e. "..") will not work if
1072 	 * the nfsnode gets flushed out of the cache. Ufs does not have
1073 	 * this problem, because one can identify root inodes by their
1074 	 * number == ROOTINO (2).
1075 	 */
1076 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1077 	if (error)
1078 		goto bad;
1079 	*vpp = NFSTOV(np);
1080 
1081 	/*
1082 	 * Retrieval of mountpoint attributes is delayed until nfs_rot
1083 	 * or nfs_statfs are first called.  This will happen either when
1084 	 * we first traverse the mount point or if somebody does a df(1).
1085 	 *
1086 	 * NFSSTA_GOTFSINFO is used to flag if we have successfully
1087 	 * retrieved mountpoint attributes.  In the case of NFSv3 we
1088 	 * also flag static fsinfo.
1089 	 */
1090 	if (*vpp != NULL)
1091 		(*vpp)->v_type = VNON;
1092 
1093 	/*
1094 	 * Lose the lock but keep the ref.
1095 	 */
1096 	vn_unlock(*vpp);
1097 	TAILQ_INSERT_TAIL(&nfs_mountq, nmp, nm_entry);
1098 
1099 #ifdef SMP
1100 	switch(ncpus) {
1101 	case 0:
1102 	case 1:
1103 		rxcpu = 0;
1104 		txcpu = 0;
1105 		break;
1106 	case 2:
1107 		rxcpu = 0;
1108 		txcpu = 1;
1109 		break;
1110 	default:
1111 		rxcpu = 1;
1112 		txcpu = 2;
1113 		break;
1114 	}
1115 #else
1116 	rxcpu = 0;
1117 	txcpu = 0;
1118 #endif
1119 
1120 	/*
1121 	 * Start the reader and writer threads.
1122 	 */
1123 	lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread,
1124 		    NULL, 0, rxcpu, "nfsiod_rx");
1125 	lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread,
1126 		    NULL, 0, txcpu, "nfsiod_tx");
1127 
1128 	return (0);
1129 bad:
1130 	nfs_disconnect(nmp);
1131 	nfs_free_mount(nmp);
1132 	return (error);
1133 }
1134 
1135 /*
1136  * unmount system call
1137  */
1138 static int
1139 nfs_unmount(struct mount *mp, int mntflags)
1140 {
1141 	struct nfsmount *nmp;
1142 	int error, flags = 0;
1143 
1144 	nmp = VFSTONFS(mp);
1145 	if (mntflags & MNT_FORCE) {
1146 		flags |= FORCECLOSE;
1147 		nmp->nm_flag |= NFSMNT_FORCE;
1148 	}
1149 
1150 	/*
1151 	 * Goes something like this..
1152 	 * - Call vflush() to clear out vnodes for this file system
1153 	 * - Close the socket
1154 	 * - Free up the data structures
1155 	 */
1156 	/* In the forced case, cancel any outstanding requests. */
1157 	if (flags & FORCECLOSE) {
1158 		error = nfs_nmcancelreqs(nmp);
1159 		if (error) {
1160 			kprintf("NFS: %s: Unable to cancel all requests\n",
1161 				mp->mnt_stat.f_mntfromname);
1162 			/* continue anyway */
1163 		}
1164 	}
1165 
1166 	/*
1167 	 * Must handshake with nfs_clientd() if it is active. XXX
1168 	 */
1169 	nmp->nm_state |= NFSSTA_DISMINPROG;
1170 
1171 	/*
1172 	 * We hold 1 extra ref on the root vnode; see comment in mountnfs().
1173 	 *
1174 	 * If this doesn't work and we are doing a forced unmount we continue
1175 	 * anyway.
1176 	 */
1177 	error = vflush(mp, 1, flags);
1178 	if (error) {
1179 		nmp->nm_state &= ~NFSSTA_DISMINPROG;
1180 		if ((flags & FORCECLOSE) == 0)
1181 			return (error);
1182 	}
1183 
1184 	/*
1185 	 * We are now committed to the unmount.
1186 	 * For NQNFS, let the server daemon free the nfsmount structure.
1187 	 */
1188 	if (nmp->nm_flag & NFSMNT_KERB)
1189 		nmp->nm_state |= NFSSTA_DISMNT;
1190 	nfssvc_iod_stop1(nmp);
1191 	nfs_disconnect(nmp);
1192 	nfssvc_iod_stop2(nmp);
1193 	TAILQ_REMOVE(&nfs_mountq, nmp, nm_entry);
1194 
1195 	if ((nmp->nm_flag & NFSMNT_KERB) == 0) {
1196 		nfs_free_mount(nmp);
1197 	}
1198 	return (0);
1199 }
1200 
1201 void
1202 nfs_free_mount(struct nfsmount *nmp)
1203 {
1204 	if (nmp->nm_cred)  {
1205 		crfree(nmp->nm_cred);
1206 		nmp->nm_cred = NULL;
1207 	}
1208 	if (nmp->nm_nam) {
1209 		FREE(nmp->nm_nam, M_SONAME);
1210 		nmp->nm_nam = NULL;
1211 	}
1212 	zfree(nfsmount_zone, nmp);
1213 }
1214 
1215 /*
1216  * Return root of a filesystem
1217  */
1218 static int
1219 nfs_root(struct mount *mp, struct vnode **vpp)
1220 {
1221 	struct vnode *vp;
1222 	struct nfsmount *nmp;
1223 	struct vattr attrs;
1224 	struct nfsnode *np;
1225 	int error;
1226 
1227 	nmp = VFSTONFS(mp);
1228 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1229 	if (error)
1230 		return (error);
1231 	vp = NFSTOV(np);
1232 
1233 	/*
1234 	 * Get transfer parameters and root vnode attributes
1235 	 *
1236 	 * NOTE: nfs_fsinfo() is expected to override the default
1237 	 *	 f_iosize we set.
1238 	 */
1239 	if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1240 	    if (nmp->nm_flag & NFSMNT_NFSV3) {
1241 		mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1242 		error = nfs_fsinfo(nmp, vp, curthread);
1243 	    } else {
1244 		if ((error = VOP_GETATTR(vp, &attrs)) == 0)
1245 			nmp->nm_state |= NFSSTA_GOTFSINFO;
1246 
1247 	    }
1248 	} else {
1249 	    /*
1250 	     * The root vnode is usually cached by the namecache so do not
1251 	     * try to avoid going over the wire even if we have previous
1252 	     * information cached.  A stale NFS mount can loop
1253 	     * forever resolving the root vnode if we return no-error when
1254 	     * there is in fact an error.
1255 	     */
1256 	    np->n_attrstamp = 0;
1257 	    error = VOP_GETATTR(vp, &attrs);
1258 	}
1259 	if (vp->v_type == VNON)
1260 	    nfs_setvtype(vp, VDIR);
1261 	vp->v_flag |= VROOT;
1262 	if (error)
1263 		vput(vp);
1264 	else
1265 		*vpp = vp;
1266 	return (error);
1267 }
1268 
1269 struct scaninfo {
1270 	int rescan;
1271 	int waitfor;
1272 	int allerror;
1273 };
1274 
1275 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1276 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1277 
1278 /*
1279  * Flush out the buffer cache
1280  */
1281 /* ARGSUSED */
1282 static int
1283 nfs_sync(struct mount *mp, int waitfor)
1284 {
1285 	struct scaninfo scaninfo;
1286 	int error;
1287 
1288 	scaninfo.rescan = 1;
1289 	scaninfo.waitfor = waitfor;
1290 	scaninfo.allerror = 0;
1291 
1292 	/*
1293 	 * Force stale buffer cache information to be flushed.
1294 	 */
1295 	error = 0;
1296 	while (error == 0 && scaninfo.rescan) {
1297 		scaninfo.rescan = 0;
1298 		error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1299 					nfs_sync_scan2, &scaninfo);
1300 	}
1301 	return(error);
1302 }
1303 
1304 static int
1305 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1306 {
1307     struct scaninfo *info = data;
1308 
1309     if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree))
1310 	return(-1);
1311     if (info->waitfor == MNT_LAZY)
1312 	return(-1);
1313     return(0);
1314 }
1315 
1316 static int
1317 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1318 {
1319     struct scaninfo *info = data;
1320     int error;
1321 
1322     error = VOP_FSYNC(vp, info->waitfor, 0);
1323     if (error)
1324 	info->allerror = error;
1325     return(0);
1326 }
1327 
1328