NFS - Implement statvfs to take advantage of 64-bit fields
[dragonfly.git] / sys / vfs / nfs / nfs_vfsops.c
CommitLineData
984263bc
MD
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 $
df0a038e 38 * $DragonFly: src/sys/vfs/nfs/nfs_vfsops.c,v 1.54 2008/07/31 20:23:40 swildner Exp $
984263bc
MD
39 */
40
41#include "opt_bootp.h"
0e8ff41a 42#include "opt_nfsroot.h"
984263bc
MD
43
44#include <sys/param.h>
45#include <sys/sockio.h>
46#include <sys/proc.h>
47#include <sys/vnode.h>
9ba76b73 48#include <sys/fcntl.h>
984263bc
MD
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
165dba55 66#include <sys/thread2.h>
8684e6f9 67#include <sys/mutex2.h>
165dba55 68
1f2de5d4
MD
69#include "rpcv2.h"
70#include "nfsproto.h"
71#include "nfs.h"
1f2de5d4 72#include "nfsmount.h"
c1cf1e59 73#include "nfsnode.h"
1f2de5d4
MD
74#include "xdr_subs.h"
75#include "nfsm_subs.h"
76#include "nfsdiskless.h"
0e8ff41a 77#include "nfsmountrpc.h"
984263bc 78
c8a30e1e
JS
79extern int nfs_mountroot(struct mount *mp);
80extern void bootpc_init(void);
984263bc 81
66a1ddf5
MD
82extern struct vop_ops nfsv2_vnode_vops;
83extern struct vop_ops nfsv2_fifo_vops;
84extern struct vop_ops nfsv2_spec_vops;
984263bc
MD
85
86MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
87MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
88MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
89MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
90MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
91MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
984263bc
MD
92MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
93
94vm_zone_t nfsmount_zone;
95
96struct nfsstats nfsstats;
97SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
98SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
99 &nfsstats, nfsstats, "");
100static int nfs_ip_paranoia = 1;
101SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
102 &nfs_ip_paranoia, 0, "");
103#ifdef NFS_DEBUG
104int nfs_debug;
105SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
106#endif
107
eefb74c1
MD
108/*
109 * Tunable to determine the Read/Write unit size. Maximum value
110 * is NFS_MAXDATA. We also default to NFS_MAXDATA.
111 */
112static int nfs_io_size = NFS_MAXDATA;
113SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
114 &nfs_io_size, 0, "NFS optimal I/O unit size");
115
a6ee311a
RG
116static void nfs_decode_args (struct nfsmount *nmp,
117 struct nfs_args *argp);
118static int mountnfs (struct nfs_args *,struct mount *,
119 struct sockaddr *,char *,char *,struct vnode **);
120static int nfs_mount ( struct mount *mp, char *path, caddr_t data,
acde96db
MD
121 struct ucred *cred);
122static int nfs_unmount ( struct mount *mp, int mntflags);
a6ee311a
RG
123static int nfs_root ( struct mount *mp, struct vnode **vpp);
124static int nfs_statfs ( struct mount *mp, struct statfs *sbp,
acde96db 125 struct ucred *cred);
2ef35c61
AH
126static int nfs_statvfs(struct mount *mp, struct statvfs *sbp,
127 struct ucred *cred);
87de5057 128static int nfs_sync ( struct mount *mp, int waitfor);
984263bc
MD
129
130/*
131 * nfs vfs operations.
132 */
133static struct vfsops nfs_vfsops = {
43c45e8f
HP
134 .vfs_mount = nfs_mount,
135 .vfs_unmount = nfs_unmount,
136 .vfs_root = nfs_root,
137 .vfs_statfs = nfs_statfs,
2ef35c61 138 .vfs_statvfs = nfs_statvfs,
43c45e8f
HP
139 .vfs_sync = nfs_sync,
140 .vfs_init = nfs_init,
141 .vfs_uninit = nfs_uninit
984263bc
MD
142};
143VFS_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 */
150struct nfs_diskless nfs_diskless = { { { 0 } } };
151struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
152int nfs_diskless_valid = 0;
153
2ef35c61 154SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
984263bc
MD
155 &nfs_diskless_valid, 0, "");
156
157SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
158 nfsv3_diskless.root_hostnam, 0, "");
159
160SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
161 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
162 "%Ssockaddr_in", "");
163
164SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
165 nfsv3_diskless.swap_hostnam, 0, "");
166
167SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
2ef35c61 168 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
984263bc
MD
169 "%Ssockaddr_in","");
170
171
a6ee311a
RG
172void nfsargs_ntoh (struct nfs_args *);
173static int nfs_mountdiskless (char *, char *, int,
984263bc 174 struct sockaddr_in *, struct nfs_args *,
dadab5e9 175 struct thread *, struct vnode **,
a6ee311a
RG
176 struct mount **);
177static void nfs_convert_diskless (void);
178static void nfs_convert_oargs (struct nfs_args *args,
179 struct onfs_args *oargs);
984263bc 180
eefb74c1
MD
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 */
3fcb1ab8 189static int
eefb74c1 190nfs_iosize(int v3, int sotype)
984263bc
MD
191{
192 int iosize;
eefb74c1 193 int iomax;
984263bc 194
eefb74c1
MD
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;
8d429613
MD
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;
984263bc
MD
221 return iosize;
222}
223
224static void
3fcb1ab8 225nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
984263bc
MD
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;
984263bc
MD
242 args->deadthresh = oargs->deadthresh;
243 args->hostname = oargs->hostname;
244}
245
246static void
3fcb1ab8 247nfs_convert_diskless(void)
984263bc 248{
0e8ff41a
MD
249 int i;
250
984263bc
MD
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);
0e8ff41a 256
984263bc 257 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
0e8ff41a
MD
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
984263bc
MD
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);
0e8ff41a 273
984263bc 274 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
0e8ff41a
MD
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
984263bc
MD
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 */
295int
acde96db 296nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
984263bc 297{
40393ded
RG
298 struct vnode *vp;
299 struct nfs_statfs *sfp;
984263bc 300 struct nfsmount *nmp = VFSTONFS(mp);
acde96db 301 thread_t td = curthread;
42edf14f 302 int error = 0, retattr;
984263bc
MD
303 struct nfsnode *np;
304 u_quad_t tquad;
42edf14f
MD
305 struct nfsm_info info;
306
307 info.mrep = NULL;
308 info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
984263bc
MD
309
310#ifndef nolint
60233e58 311 sfp = NULL;
984263bc
MD
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);
acde96db 317 /* ignore the passed cred */
984263bc
MD
318 cred = crget();
319 cred->cr_ngroups = 1;
42edf14f 320 if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
3b568787 321 (void)nfs_fsinfo(nmp, vp, td);
984263bc 322 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
42edf14f
MD
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 }
984263bc 330 if (error) {
42edf14f
MD
331 if (info.mrep != NULL)
332 m_freem(info.mrep);
984263bc
MD
333 goto nfsmout;
334 }
42edf14f 335 NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
984263bc 336 sbp->f_flags = nmp->nm_flag;
42edf14f 337 sbp->f_iosize = nfs_iosize(info.v3, nmp->nm_sotype);
eefb74c1 338
42edf14f 339 if (info.v3) {
984263bc
MD
340 sbp->f_bsize = NFS_FABLKSIZE;
341 tquad = fxdr_hyper(&sfp->sf_tbytes);
342 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
343 tquad = fxdr_hyper(&sfp->sf_fbytes);
344 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
345 tquad = fxdr_hyper(&sfp->sf_abytes);
346 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
347 sbp->f_files = (fxdr_unsigned(int32_t,
348 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
349 sbp->f_ffree = (fxdr_unsigned(int32_t,
350 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
351 } else {
352 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
353 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
354 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
355 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
356 sbp->f_files = 0;
357 sbp->f_ffree = 0;
358 }
359 if (sbp != &mp->mnt_stat) {
360 sbp->f_type = mp->mnt_vfc->vfc_typenum;
984263bc
MD
361 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
362 }
42edf14f
MD
363 m_freem(info.mrep);
364 info.mrep = NULL;
6b08710e 365nfsmout:
984263bc
MD
366 vput(vp);
367 crfree(cred);
368 return (error);
369}
370
2ef35c61
AH
371static int
372nfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
373{
374 struct vnode *vp;
375 struct nfs_statfs *sfp;
376 struct nfsmount *nmp = VFSTONFS(mp);
377 thread_t td = curthread;
378 int error = 0, retattr;
379 struct nfsnode *np;
380 struct nfsm_info info;
381
382 info.mrep = NULL;
383 info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
384
385#ifndef nolint
386 sfp = NULL;
387#endif
388 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
389 if (error)
390 return (error);
391 vp = NFSTOV(np);
392 /* ignore the passed cred */
393 cred = crget();
394 cred->cr_ngroups = 1;
395 if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
396 (void)nfs_fsinfo(nmp, vp, td);
397 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
398 nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
399 ERROROUT(nfsm_fhtom(&info, vp));
400 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
401 if (info.v3) {
402 ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
403 NFS_LATTR_NOSHRINK));
404 }
405 if (error) {
406 if (info.mrep != NULL)
407 m_freem(info.mrep);
408 goto nfsmout;
409 }
410 NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
411 sbp->f_flag = nmp->nm_flag;
412 sbp->f_owner = nmp->nm_cred->cr_ruid;
413
414 if (info.v3) {
415 sbp->f_bsize = NFS_FABLKSIZE;
416 sbp->f_frsize = NFS_FABLKSIZE;
417 sbp->f_blocks = (fxdr_hyper(&sfp->sf_tbytes) /
418 ((u_quad_t)NFS_FABLKSIZE));
419 sbp->f_bfree = (fxdr_hyper(&sfp->sf_fbytes) /
420 ((u_quad_t)NFS_FABLKSIZE));
421 sbp->f_bavail = (fxdr_hyper(&sfp->sf_abytes) /
422 ((u_quad_t)NFS_FABLKSIZE));
423 sbp->f_files = fxdr_hyper(&sfp->sf_tfiles);
424 sbp->f_ffree = fxdr_hyper(&sfp->sf_ffiles);
425 sbp->f_favail = fxdr_hyper(&sfp->sf_afiles);
426 } else {
427 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
428 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
429 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
430 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
431 sbp->f_files = 0;
432 sbp->f_ffree = 0;
433 sbp->f_favail = 0;
434 }
435 sbp->f_syncreads = 0;
436 sbp->f_syncwrites = 0;
437 sbp->f_asyncreads = 0;
438 sbp->f_asyncwrites = 0;
439 sbp->f_type = mp->mnt_vfc->vfc_typenum;
440
441 m_freem(info.mrep);
442 info.mrep = NULL;
443nfsmout:
444 vput(vp);
445 crfree(cred);
446 return (error);
447}
448
984263bc
MD
449/*
450 * nfs version 3 fsinfo rpc call
451 */
452int
3b568787 453nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
984263bc 454{
40393ded 455 struct nfsv3_fsinfo *fsp;
42edf14f 456 u_int32_t pref, max;
984263bc 457 int error = 0, retattr;
984263bc 458 u_int64_t maxfsize;
42edf14f 459 struct nfsm_info info;
984263bc 460
42edf14f 461 info.v3 = 1;
984263bc 462 nfsstats.rpccnt[NFSPROC_FSINFO]++;
42edf14f
MD
463 nfsm_reqhead(&info, vp, NFSPROC_FSINFO, NFSX_FH(1));
464 ERROROUT(nfsm_fhtom(&info, vp));
465 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSINFO, td,
466 nfs_vpcred(vp, ND_READ), &error));
467 ERROROUT(nfsm_postop_attr(&info, vp, &retattr, NFS_LATTR_NOSHRINK));
468 if (error == 0) {
469 NULLOUT(fsp = nfsm_dissect(&info, NFSX_V3FSINFO));
984263bc
MD
470 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
471 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
472 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
473 ~(NFS_FABLKSIZE - 1);
474 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
475 if (max < nmp->nm_wsize && max > 0) {
476 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
477 if (nmp->nm_wsize == 0)
478 nmp->nm_wsize = max;
479 }
480 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
481 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
482 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
483 ~(NFS_FABLKSIZE - 1);
484 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
485 if (max < nmp->nm_rsize && max > 0) {
486 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
487 if (nmp->nm_rsize == 0)
488 nmp->nm_rsize = max;
489 }
490 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
491 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
492 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
493 ~(NFS_DIRBLKSIZ - 1);
494 if (max < nmp->nm_readdirsize && max > 0) {
495 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
496 if (nmp->nm_readdirsize == 0)
497 nmp->nm_readdirsize = max;
498 }
499 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
500 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
501 nmp->nm_maxfilesize = maxfsize;
502 nmp->nm_state |= NFSSTA_GOTFSINFO;
503 }
42edf14f
MD
504 m_freem(info.mrep);
505 info.mrep = NULL;
6b08710e 506nfsmout:
984263bc
MD
507 return (error);
508}
509
510/*
511 * Mount a remote root fs via. nfs. This depends on the info in the
512 * nfs_diskless structure that has been filled in properly by some primary
513 * bootstrap.
514 * It goes something like this:
515 * - do enough of "ifconfig" by calling ifioctl() so that the system
516 * can talk to the server
517 * - If nfs_diskless.mygateway is filled in, use that address as
518 * a default gateway.
519 * - build the rootfs mount point and call mountnfs() to do the rest.
520 */
521int
acde96db 522nfs_mountroot(struct mount *mp)
984263bc
MD
523{
524 struct mount *swap_mp;
525 struct nfsv3_diskless *nd = &nfsv3_diskless;
526 struct socket *so;
527 struct vnode *vp;
dadab5e9 528 struct thread *td = curthread; /* XXX */
984263bc
MD
529 int error, i;
530 u_long l;
531 char buf[128];
532
533#if defined(BOOTP_NFSROOT) && defined(BOOTP)
534 bootpc_init(); /* use bootp to get nfs_diskless filled in */
535#endif
536
537 /*
538 * XXX time must be non-zero when we init the interface or else
539 * the arp code will wedge...
540 */
3a6f9faf
MD
541 while (mycpu->gd_time_seconds == 0)
542 tsleep(mycpu, 0, "arpkludge", 10);
984263bc 543
a2103e90
MD
544 /*
545 * The boot code may have passed us a diskless structure.
546 */
2ef35c61 547 if (nfs_diskless_valid == 1)
0b9984b8
MD
548 nfs_convert_diskless();
549
a2103e90 550#define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr))
086c1d7e 551 kprintf("nfs_mountroot: interface %s ip %s",
2ef35c61 552 nd->myif.ifra_name,
a2103e90 553 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
2ef35c61 554 kprintf(" bcast %s",
a2103e90 555 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
2ef35c61 556 kprintf(" mask %s\n",
a2103e90
MD
557 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
558#undef SINP
984263bc
MD
559
560 /*
561 * XXX splnet, so networks will receive...
562 */
165dba55 563 crit_enter();
984263bc 564
984263bc 565 /*
a2103e90 566 * BOOTP does not necessarily have to be compiled into the kernel
2ef35c61 567 * for an NFS root to work. If we inherited the network
a2103e90
MD
568 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
569 * out our interface for us and all we need to do is ifconfig the
570 * interface. We only do this if the interface has not already been
571 * ifconfig'd by e.g. BOOTP.
984263bc 572 */
dadab5e9 573 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
a2103e90 574 if (error) {
984263bc
MD
575 panic("nfs_mountroot: socreate(%04x): %d",
576 nd->myif.ifra_addr.sa_family, error);
984263bc 577 }
a2103e90 578
87de5057 579 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
984263bc
MD
580 if (error)
581 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
a2103e90 582
9ba76b73 583 soclose(so, FNONBLOCK);
984263bc
MD
584
585 /*
586 * If the gateway field is filled in, set it as the default route.
587 */
588 if (nd->mygateway.sin_len != 0) {
589 struct sockaddr_in mask, sin;
590
591 bzero((caddr_t)&mask, sizeof(mask));
592 sin = mask;
593 sin.sin_family = AF_INET;
594 sin.sin_len = sizeof(sin);
086c1d7e 595 kprintf("nfs_mountroot: gateway %s\n",
0b9984b8 596 inet_ntoa(nd->mygateway.sin_addr));
ecdefdda
MD
597 error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
598 (struct sockaddr *)&nd->mygateway,
599 (struct sockaddr *)&mask,
600 RTF_UP | RTF_GATEWAY);
984263bc 601 if (error)
086c1d7e 602 kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
984263bc
MD
603 }
604
605 /*
606 * Create the rootfs mount point.
607 */
608 nd->root_args.fh = nd->root_fh;
609 nd->root_args.fhsize = nd->root_fhsize;
610 l = ntohl(nd->root_saddr.sin_addr.s_addr);
f8c7a42d 611 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
984263bc
MD
612 (l >> 24) & 0xff, (l >> 16) & 0xff,
613 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
5dacbb2e 614 kprintf("NFS_ROOT: %s\n",buf);
984263bc 615 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
dadab5e9 616 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
5dacbb2e 617 mp->mnt_vfc->vfc_refcount--;
0b9984b8 618 crit_exit();
984263bc
MD
619 return (error);
620 }
621
622 swap_mp = NULL;
623 if (nd->swap_nblks) {
624
625 /* Convert to DEV_BSIZE instead of Kilobyte */
626 nd->swap_nblks *= 2;
627
628 /*
629 * Create a fake mount point just for the swap vnode so that the
630 * swap file can be on a different server from the rootfs.
631 */
632 nd->swap_args.fh = nd->swap_fh;
633 nd->swap_args.fhsize = nd->swap_fhsize;
634 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
f8c7a42d 635 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
984263bc
MD
636 (l >> 24) & 0xff, (l >> 16) & 0xff,
637 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
086c1d7e 638 kprintf("NFS SWAP: %s\n",buf);
984263bc 639 if ((error = nfs_mountdiskless(buf, "/swap", 0,
0b9984b8
MD
640 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) {
641 crit_exit();
984263bc 642 return (error);
0b9984b8 643 }
f9642f56 644 vfs_unbusy(swap_mp);
984263bc 645
2ef35c61 646 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
984263bc 647 nd->swap_nblks * DEV_BSIZE ;
2ef35c61 648
984263bc
MD
649 /*
650 * Since the swap file is not the root dir of a file system,
651 * hack it to a regular file.
652 */
984263bc 653 vp->v_flag = 0;
597aea93 654 vref(vp);
1c843a13 655 nfs_setvtype(vp, VREG);
e4c9c0c8 656 swaponvp(td, vp, nd->swap_nblks);
984263bc
MD
657 }
658
659 mp->mnt_flag |= MNT_ROOTFS;
f9642f56 660 vfs_unbusy(mp);
984263bc
MD
661
662 /*
663 * This is not really an nfs issue, but it is much easier to
664 * set hostname here and then let the "/etc/rc.xxx" files
665 * mount the right /var based upon its preset value.
666 */
667 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
668 hostname[MAXHOSTNAMELEN - 1] = '\0';
669 for (i = 0; i < MAXHOSTNAMELEN; i++)
670 if (hostname[i] == '\0')
671 break;
672 inittodr(ntohl(nd->root_time));
0b9984b8 673 crit_exit();
984263bc
MD
674 return (0);
675}
676
677/*
678 * Internal version of mount system call for diskless setup.
679 */
680static int
dadab5e9
MD
681nfs_mountdiskless(char *path, char *which, int mountflag,
682 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
683 struct vnode **vpp, struct mount **mpp)
984263bc
MD
684{
685 struct mount *mp;
686 struct sockaddr *nam;
984263bc 687 int didalloc = 0;
0e8ff41a 688 int error;
984263bc
MD
689
690 mp = *mpp;
691
692 if (mp == NULL) {
693 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
086c1d7e 694 kprintf("nfs_mountroot: NFS not configured");
984263bc
MD
695 return (error);
696 }
697 didalloc = 1;
698 }
984263bc
MD
699 mp->mnt_kern_flag = 0;
700 mp->mnt_flag = mountflag;
cfa2ba21 701 nam = dup_sockaddr((struct sockaddr *)sin);
0e8ff41a
MD
702
703#if defined(BOOTP) || defined(NFS_ROOT)
704 if (args->fhsize == 0) {
df0a038e
SW
705 char *xpath = path;
706
086c1d7e 707 kprintf("NFS_ROOT: No FH passed from loader, attempting mount rpc...");
5dacbb2e
MD
708 while (*xpath && *xpath != ':')
709 ++xpath;
710 if (*xpath)
711 ++xpath;
0e8ff41a 712 args->fhsize = 0;
5dacbb2e 713 error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td);
0e8ff41a 714 if (error) {
5dacbb2e 715 kprintf("failed error %d.\n", error);
0e8ff41a
MD
716 goto haderror;
717 }
086c1d7e 718 kprintf("success!\n");
0e8ff41a
MD
719 }
720#endif
721
984263bc 722 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
04ef15fb 723#if defined(BOOTP) || defined(NFS_ROOT)
0e8ff41a 724haderror:
04ef15fb 725#endif
086c1d7e 726 kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error);
984263bc 727 mp->mnt_vfc->vfc_refcount--;
f9642f56 728 vfs_unbusy(mp);
984263bc 729 if (didalloc)
efda3bd0 730 kfree(mp, M_MOUNT);
984263bc
MD
731 FREE(nam, M_SONAME);
732 return (error);
733 }
984263bc
MD
734 *mpp = mp;
735 return (0);
736}
737
738static void
3fcb1ab8 739nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp)
984263bc 740{
984263bc
MD
741 int adjsock;
742 int maxio;
743
165dba55 744 crit_enter();
984263bc
MD
745 /*
746 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
747 * no sense in that context.
748 */
749 if (argp->sotype == SOCK_STREAM)
750 nmp->nm_flag &= ~NFSMNT_NOCONN;
751
752 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
753 if ((argp->flags & NFSMNT_NFSV3) == 0)
754 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
755
756 /* Re-bind if rsrvd port requested and wasn't on one */
757 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
758 && (argp->flags & NFSMNT_RESVPORT);
759 /* Also re-bind if we're switching to/from a connected UDP socket */
760 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
761 (argp->flags & NFSMNT_NOCONN));
762
763 /* Update flags atomically. Don't change the lock bits. */
764 nmp->nm_flag = argp->flags | nmp->nm_flag;
165dba55 765 crit_exit();
984263bc
MD
766
767 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
768 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
769 if (nmp->nm_timeo < NFS_MINTIMEO)
770 nmp->nm_timeo = NFS_MINTIMEO;
771 else if (nmp->nm_timeo > NFS_MAXTIMEO)
772 nmp->nm_timeo = NFS_MAXTIMEO;
773 }
774
775 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
776 nmp->nm_retry = argp->retrans;
777 if (nmp->nm_retry > NFS_MAXREXMIT)
778 nmp->nm_retry = NFS_MAXREXMIT;
779 }
780
eefb74c1 781 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
984263bc
MD
782
783 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
784 nmp->nm_wsize = argp->wsize;
785 /* Round down to multiple of blocksize */
786 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
787 if (nmp->nm_wsize <= 0)
788 nmp->nm_wsize = NFS_FABLKSIZE;
789 }
790 if (nmp->nm_wsize > maxio)
791 nmp->nm_wsize = maxio;
792 if (nmp->nm_wsize > MAXBSIZE)
793 nmp->nm_wsize = MAXBSIZE;
794
795 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
796 nmp->nm_rsize = argp->rsize;
797 /* Round down to multiple of blocksize */
798 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
799 if (nmp->nm_rsize <= 0)
800 nmp->nm_rsize = NFS_FABLKSIZE;
801 }
802 if (nmp->nm_rsize > maxio)
803 nmp->nm_rsize = maxio;
804 if (nmp->nm_rsize > MAXBSIZE)
805 nmp->nm_rsize = MAXBSIZE;
806
807 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
808 nmp->nm_readdirsize = argp->readdirsize;
809 }
810 if (nmp->nm_readdirsize > maxio)
811 nmp->nm_readdirsize = maxio;
812 if (nmp->nm_readdirsize > nmp->nm_rsize)
813 nmp->nm_readdirsize = nmp->nm_rsize;
814
815 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
816 nmp->nm_acregmin = argp->acregmin;
817 else
818 nmp->nm_acregmin = NFS_MINATTRTIMO;
819 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
820 nmp->nm_acregmax = argp->acregmax;
821 else
822 nmp->nm_acregmax = NFS_MAXATTRTIMO;
823 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
824 nmp->nm_acdirmin = argp->acdirmin;
825 else
826 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
827 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
828 nmp->nm_acdirmax = argp->acdirmax;
829 else
830 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
831 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
832 nmp->nm_acdirmin = nmp->nm_acdirmax;
833 if (nmp->nm_acregmin > nmp->nm_acregmax)
834 nmp->nm_acregmin = nmp->nm_acregmax;
835
836 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
837 if (argp->maxgrouplist <= NFS_MAXGRPS)
838 nmp->nm_numgrps = argp->maxgrouplist;
839 else
840 nmp->nm_numgrps = NFS_MAXGRPS;
841 }
842 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
843 if (argp->readahead <= NFS_MAXRAHEAD)
844 nmp->nm_readahead = argp->readahead;
845 else
846 nmp->nm_readahead = NFS_MAXRAHEAD;
847 }
984263bc 848 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
e07fef60 849 if (argp->deadthresh <= NFS_NEVERDEAD)
984263bc
MD
850 nmp->nm_deadthresh = argp->deadthresh;
851 else
e07fef60 852 nmp->nm_deadthresh = NFS_NEVERDEAD;
984263bc
MD
853 }
854
855 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
856 (nmp->nm_soproto != argp->proto));
857 nmp->nm_sotype = argp->sotype;
858 nmp->nm_soproto = argp->proto;
859
860 if (nmp->nm_so && adjsock) {
861 nfs_safedisconnect(nmp);
862 if (nmp->nm_sotype == SOCK_DGRAM)
60233e58 863 while (nfs_connect(nmp, NULL)) {
086c1d7e 864 kprintf("nfs_args: retrying connect\n");
377d4740 865 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
984263bc
MD
866 }
867 }
868}
869
870/*
871 * VFS Operations.
872 *
873 * mount system call
874 * It seems a bit dumb to copyinstr() the host and path here and then
875 * bcopy() them in mountnfs(), but I wanted to detect errors before
876 * doing the sockargs() call because sockargs() allocates an mbuf and
877 * an error after that means that I have to release the mbuf.
878 */
879/* ARGSUSED */
880static int
acde96db 881nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
984263bc
MD
882{
883 int error;
884 struct nfs_args args;
885 struct sockaddr *nam;
886 struct vnode *vp;
887 char pth[MNAMELEN], hst[MNAMELEN];
888 size_t len;
889 u_char nfh[NFSX_V3FHMAX];
890
891 if (path == NULL) {
892 nfs_mountroot(mp);
893 return (0);
894 }
895 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
896 if (error)
897 return (error);
898 if (args.version != NFS_ARGSVERSION) {
899#ifdef COMPAT_PRELITE2
900 /*
901 * If the argument version is unknown, then assume the
902 * caller is a pre-lite2 4.4BSD client and convert its
903 * arguments.
904 */
905 struct onfs_args oargs;
906 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
907 if (error)
908 return (error);
909 nfs_convert_oargs(&args,&oargs);
910#else /* !COMPAT_PRELITE2 */
911 return (EPROGMISMATCH);
912#endif /* COMPAT_PRELITE2 */
913 }
914 if (mp->mnt_flag & MNT_UPDATE) {
40393ded 915 struct nfsmount *nmp = VFSTONFS(mp);
984263bc
MD
916
917 if (nmp == NULL)
918 return (EIO);
919 /*
920 * When doing an update, we can't change from or to
e07fef60 921 * v3, or change cookie translation
984263bc
MD
922 */
923 args.flags = (args.flags &
e07fef60 924 ~(NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/)) |
984263bc 925 (nmp->nm_flag &
e07fef60 926 (NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/));
984263bc
MD
927 nfs_decode_args(nmp, &args);
928 return (0);
929 }
930
931 /*
932 * Make the nfs_ip_paranoia sysctl serve as the default connection
2ef35c61 933 * or no-connection mode for those protocols that support
984263bc
MD
934 * no-connection mode (the flag will be cleared later for protocols
935 * that do not support no-connection mode). This will allow a client
936 * to receive replies from a different IP then the request was
937 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
938 * not 0.
939 */
940 if (nfs_ip_paranoia == 0)
941 args.flags |= NFSMNT_NOCONN;
942 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
943 return (EINVAL);
944 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
945 if (error)
946 return (error);
947 error = copyinstr(path, pth, MNAMELEN-1, &len);
948 if (error)
949 return (error);
950 bzero(&pth[len], MNAMELEN - len);
951 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
952 if (error)
953 return (error);
954 bzero(&hst[len], MNAMELEN - len);
955 /* sockargs() call must be after above copyin() calls */
956 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
957 if (error)
958 return (error);
959 args.fh = nfh;
960 error = mountnfs(&args, mp, nam, pth, hst, &vp);
961 return (error);
962}
963
964/*
965 * Common code for mount and mountroot
966 */
967static int
dadab5e9
MD
968mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
969 char *pth, char *hst, struct vnode **vpp)
984263bc 970{
40393ded 971 struct nfsmount *nmp;
984263bc
MD
972 struct nfsnode *np;
973 int error;
13ddc895
MD
974 int rxcpu;
975 int txcpu;
984263bc
MD
976
977 if (mp->mnt_flag & MNT_UPDATE) {
978 nmp = VFSTONFS(mp);
979 /* update paths, file handles, etc, here XXX */
980 FREE(nam, M_SONAME);
981 return (0);
982 } else {
983 nmp = zalloc(nfsmount_zone);
984 bzero((caddr_t)nmp, sizeof (struct nfsmount));
8684e6f9
MD
985 mtx_init(&nmp->nm_rxlock);
986 mtx_init(&nmp->nm_txlock);
984263bc 987 TAILQ_INIT(&nmp->nm_uidlruhead);
81b5c339 988 TAILQ_INIT(&nmp->nm_bioq);
e21aec5b 989 TAILQ_INIT(&nmp->nm_reqq);
edb90c22
MD
990 TAILQ_INIT(&nmp->nm_reqtxq);
991 TAILQ_INIT(&nmp->nm_reqrxq);
984263bc
MD
992 mp->mnt_data = (qaddr_t)nmp;
993 }
994 vfs_getnewfsid(mp);
995 nmp->nm_mountp = mp;
984263bc
MD
996
997 /*
998 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
999 * high, depending on whether we end up with negative offsets in
1000 * the client or server somewhere. 2GB-1 may be safer.
1001 *
1002 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
1003 * that we can handle until we find out otherwise.
1004 * XXX Our "safe" limit on the client is what we can store in our
1005 * buffer cache using signed(!) block numbers.
1006 */
1007 if ((argp->flags & NFSMNT_NFSV3) == 0)
1008 nmp->nm_maxfilesize = 0xffffffffLL;
1009 else
1010 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
1011
1012 nmp->nm_timeo = NFS_TIMEO;
1013 nmp->nm_retry = NFS_RETRANS;
eefb74c1
MD
1014 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
1015 nmp->nm_rsize = nmp->nm_wsize;
984263bc
MD
1016 nmp->nm_readdirsize = NFS_READDIRSIZE;
1017 nmp->nm_numgrps = NFS_MAXGRPS;
1018 nmp->nm_readahead = NFS_DEFRAHEAD;
0639df29 1019 nmp->nm_deadthresh = NFS_DEADTHRESH;
984263bc
MD
1020 nmp->nm_fhsize = argp->fhsize;
1021 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
1022 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
984263bc
MD
1023 nmp->nm_nam = nam;
1024 /* Set up the sockets and per-host congestion */
1025 nmp->nm_sotype = argp->sotype;
1026 nmp->nm_soproto = argp->proto;
3b568787 1027 nmp->nm_cred = crhold(proc0.p_ucred);
984263bc
MD
1028
1029 nfs_decode_args(nmp, argp);
1030
1031 /*
1032 * For Connection based sockets (TCP,...) defer the connect until
1033 * the first request, in case the server is not responding.
1034 */
1035 if (nmp->nm_sotype == SOCK_DGRAM &&
60233e58 1036 (error = nfs_connect(nmp, NULL)))
984263bc
MD
1037 goto bad;
1038
1039 /*
1040 * This is silly, but it has to be set so that vinifod() works.
1041 * We do not want to do an nfs_statfs() here since we can get
1042 * stuck on a dead server and we are holding a lock on the mount
1043 * point.
1044 */
2ef35c61 1045 mp->mnt_stat.f_iosize =
eefb74c1
MD
1046 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
1047
984263bc 1048 /*
0961aa92
MD
1049 * Install vop_ops for our vnops
1050 */
66a1ddf5
MD
1051 vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
1052 vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
1053 vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
0961aa92
MD
1054
1055 /*
984263bc
MD
1056 * A reference count is needed on the nfsnode representing the
1057 * remote root. If this object is not persistent, then backward
1058 * traversals of the mount point (i.e. "..") will not work if
1059 * the nfsnode gets flushed out of the cache. Ufs does not have
1060 * this problem, because one can identify root inodes by their
1061 * number == ROOTINO (2).
1062 */
1063 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1064 if (error)
1065 goto bad;
1066 *vpp = NFSTOV(np);
1067
1068 /*
eefb74c1
MD
1069 * Retrieval of mountpoint attributes is delayed until nfs_rot
1070 * or nfs_statfs are first called. This will happen either when
1071 * we first traverse the mount point or if somebody does a df(1).
1072 *
1073 * NFSSTA_GOTFSINFO is used to flag if we have successfully
1074 * retrieved mountpoint attributes. In the case of NFSv3 we
1075 * also flag static fsinfo.
984263bc 1076 */
eefb74c1
MD
1077 if (*vpp != NULL)
1078 (*vpp)->v_type = VNON;
984263bc
MD
1079
1080 /*
1081 * Lose the lock but keep the ref.
1082 */
a11aaa81 1083 vn_unlock(*vpp);
9a702b27 1084 TAILQ_INSERT_TAIL(&nfs_mountq, nmp, nm_entry);
984263bc 1085
13ddc895
MD
1086#ifdef SMP
1087 switch(ncpus) {
1088 case 0:
f8565b0f 1089 case 1:
13ddc895
MD
1090 rxcpu = 0;
1091 txcpu = 0;
1092 break;
f8565b0f 1093 case 2:
13ddc895
MD
1094 rxcpu = 0;
1095 txcpu = 1;
1096 break;
1097 default:
1098 rxcpu = 1;
1099 txcpu = 2;
1100 break;
1101 }
1102#else
1103 rxcpu = 0;
1104 txcpu = 0;
1105#endif
1106
52e1cf57
MD
1107 /*
1108 * Start the reader and writer threads.
1109 */
1110 lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread,
13ddc895 1111 NULL, 0, rxcpu, "nfsiod_rx");
52e1cf57 1112 lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread,
13ddc895 1113 NULL, 0, txcpu, "nfsiod_tx");
52e1cf57 1114
984263bc
MD
1115 return (0);
1116bad:
1117 nfs_disconnect(nmp);
3b568787 1118 nfs_free_mount(nmp);
984263bc
MD
1119 return (error);
1120}
1121
1122/*
1123 * unmount system call
1124 */
1125static int
acde96db 1126nfs_unmount(struct mount *mp, int mntflags)
984263bc 1127{
40393ded 1128 struct nfsmount *nmp;
984263bc
MD
1129 int error, flags = 0;
1130
984263bc 1131 nmp = VFSTONFS(mp);
d9adbeaf
MD
1132 if (mntflags & MNT_FORCE) {
1133 flags |= FORCECLOSE;
1134 nmp->nm_flag |= NFSMNT_FORCE;
1135 }
1136
984263bc
MD
1137 /*
1138 * Goes something like this..
1139 * - Call vflush() to clear out vnodes for this file system
1140 * - Close the socket
1141 * - Free up the data structures
1142 */
1143 /* In the forced case, cancel any outstanding requests. */
1144 if (flags & FORCECLOSE) {
1145 error = nfs_nmcancelreqs(nmp);
d9adbeaf
MD
1146 if (error) {
1147 kprintf("NFS: %s: Unable to cancel all requests\n",
1148 mp->mnt_stat.f_mntfromname);
1149 /* continue anyway */
1150 }
984263bc 1151 }
d9adbeaf 1152
984263bc 1153 /*
e07fef60 1154 * Must handshake with nfs_clientd() if it is active. XXX
984263bc
MD
1155 */
1156 nmp->nm_state |= NFSSTA_DISMINPROG;
984263bc 1157
d9adbeaf
MD
1158 /*
1159 * We hold 1 extra ref on the root vnode; see comment in mountnfs().
1160 *
1161 * If this doesn't work and we are doing a forced unmount we continue
1162 * anyway.
1163 */
984263bc
MD
1164 error = vflush(mp, 1, flags);
1165 if (error) {
1166 nmp->nm_state &= ~NFSSTA_DISMINPROG;
d9adbeaf
MD
1167 if ((flags & FORCECLOSE) == 0)
1168 return (error);
984263bc
MD
1169 }
1170
1171 /*
1172 * We are now committed to the unmount.
1173 * For NQNFS, let the server daemon free the nfsmount structure.
1174 */
e07fef60 1175 if (nmp->nm_flag & NFSMNT_KERB)
984263bc 1176 nmp->nm_state |= NFSSTA_DISMNT;
13ddc895 1177 nfssvc_iod_stop1(nmp);
984263bc 1178 nfs_disconnect(nmp);
13ddc895 1179 nfssvc_iod_stop2(nmp);
9a702b27 1180 TAILQ_REMOVE(&nfs_mountq, nmp, nm_entry);
984263bc 1181
52e1cf57 1182 if ((nmp->nm_flag & NFSMNT_KERB) == 0) {
3b568787 1183 nfs_free_mount(nmp);
52e1cf57 1184 }
984263bc
MD
1185 return (0);
1186}
1187
3b568787
MD
1188void
1189nfs_free_mount(struct nfsmount *nmp)
1190{
1191 if (nmp->nm_cred) {
1192 crfree(nmp->nm_cred);
1193 nmp->nm_cred = NULL;
1194 }
13ddc895
MD
1195 if (nmp->nm_nam) {
1196 FREE(nmp->nm_nam, M_SONAME);
1197 nmp->nm_nam = NULL;
1198 }
3b568787
MD
1199 zfree(nfsmount_zone, nmp);
1200}
1201
984263bc
MD
1202/*
1203 * Return root of a filesystem
1204 */
1205static int
3fcb1ab8 1206nfs_root(struct mount *mp, struct vnode **vpp)
984263bc 1207{
40393ded 1208 struct vnode *vp;
984263bc 1209 struct nfsmount *nmp;
eefb74c1 1210 struct vattr attrs;
984263bc
MD
1211 struct nfsnode *np;
1212 int error;
1213
1214 nmp = VFSTONFS(mp);
1215 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1216 if (error)
1217 return (error);
1218 vp = NFSTOV(np);
eefb74c1
MD
1219
1220 /*
1221 * Get transfer parameters and root vnode attributes
1222 */
1223 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1224 if (nmp->nm_flag & NFSMNT_NFSV3) {
14aa2dd7 1225 error = nfs_fsinfo(nmp, vp, curthread);
eefb74c1
MD
1226 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1227 } else {
87de5057 1228 if ((error = VOP_GETATTR(vp, &attrs)) == 0)
eefb74c1 1229 nmp->nm_state |= NFSSTA_GOTFSINFO;
2ef35c61 1230
eefb74c1 1231 }
14aa2dd7
MD
1232 } else {
1233 /*
1234 * The root vnode is usually cached by the namecache so do not
1235 * try to avoid going over the wire even if we have previous
1236 * information cached. A stale NFS mount can loop
1237 * forever resolving the root vnode if we return no-error when
1238 * there is in fact an error.
1239 */
1240 np->n_attrstamp = 0;
1241 error = VOP_GETATTR(vp, &attrs);
eefb74c1 1242 }
984263bc 1243 if (vp->v_type == VNON)
1c843a13 1244 nfs_setvtype(vp, VDIR);
984263bc 1245 vp->v_flag = VROOT;
14aa2dd7
MD
1246 if (error)
1247 vput(vp);
1248 else
1249 *vpp = vp;
1250 return (error);
984263bc
MD
1251}
1252
b2eb81cd
MD
1253struct scaninfo {
1254 int rescan;
b2eb81cd
MD
1255 int waitfor;
1256 int allerror;
1257};
1258
1259static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
5fd012e0 1260static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
b2eb81cd 1261
984263bc
MD
1262/*
1263 * Flush out the buffer cache
1264 */
1265/* ARGSUSED */
1266static int
87de5057 1267nfs_sync(struct mount *mp, int waitfor)
984263bc 1268{
b2eb81cd
MD
1269 struct scaninfo scaninfo;
1270 int error;
1271
a482a28a 1272 scaninfo.rescan = 1;
b2eb81cd
MD
1273 scaninfo.waitfor = waitfor;
1274 scaninfo.allerror = 0;
984263bc
MD
1275
1276 /*
1277 * Force stale buffer cache information to be flushed.
1278 */
b2eb81cd
MD
1279 error = 0;
1280 while (error == 0 && scaninfo.rescan) {
1281 scaninfo.rescan = 0;
5fd012e0 1282 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
b2eb81cd 1283 nfs_sync_scan2, &scaninfo);
984263bc 1284 }
b2eb81cd
MD
1285 return(error);
1286}
1287
3fcb1ab8 1288static int
b2eb81cd
MD
1289nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1290{
1291 struct scaninfo *info = data;
1292
a11aaa81 1293 if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree))
b2eb81cd
MD
1294 return(-1);
1295 if (info->waitfor == MNT_LAZY)
1296 return(-1);
1297 return(0);
1298}
1299
3fcb1ab8 1300static int
5fd012e0 1301nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
b2eb81cd
MD
1302{
1303 struct scaninfo *info = data;
1304 int error;
1305
87de5057 1306 error = VOP_FSYNC(vp, info->waitfor);
b2eb81cd
MD
1307 if (error)
1308 info->allerror = error;
b2eb81cd 1309 return(0);
984263bc
MD
1310}
1311