NFS - Remove old nfsiod, start adding new kernel thread infrastructure
[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
MD
81
82extern int nfs_ticks;
66a1ddf5
MD
83extern struct vop_ops nfsv2_vnode_vops;
84extern struct vop_ops nfsv2_fifo_vops;
85extern struct vop_ops nfsv2_spec_vops;
984263bc
MD
86
87MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
88MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
89MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
90MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
91MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
92MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
984263bc
MD
93MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
94
95vm_zone_t nfsmount_zone;
96
97struct nfsstats nfsstats;
98SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
99SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
100 &nfsstats, nfsstats, "");
101static int nfs_ip_paranoia = 1;
102SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
103 &nfs_ip_paranoia, 0, "");
104#ifdef NFS_DEBUG
105int nfs_debug;
106SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
107#endif
108
eefb74c1
MD
109/*
110 * Tunable to determine the Read/Write unit size. Maximum value
111 * is NFS_MAXDATA. We also default to NFS_MAXDATA.
112 */
113static int nfs_io_size = NFS_MAXDATA;
114SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
115 &nfs_io_size, 0, "NFS optimal I/O unit size");
116
a6ee311a
RG
117static void nfs_decode_args (struct nfsmount *nmp,
118 struct nfs_args *argp);
119static int mountnfs (struct nfs_args *,struct mount *,
120 struct sockaddr *,char *,char *,struct vnode **);
121static int nfs_mount ( struct mount *mp, char *path, caddr_t data,
acde96db
MD
122 struct ucred *cred);
123static int nfs_unmount ( struct mount *mp, int mntflags);
a6ee311a
RG
124static int nfs_root ( struct mount *mp, struct vnode **vpp);
125static int nfs_statfs ( struct mount *mp, struct statfs *sbp,
acde96db 126 struct ucred *cred);
87de5057 127static int nfs_sync ( struct mount *mp, int waitfor);
984263bc
MD
128
129/*
130 * nfs vfs operations.
131 */
132static struct vfsops nfs_vfsops = {
43c45e8f
HP
133 .vfs_mount = nfs_mount,
134 .vfs_unmount = nfs_unmount,
135 .vfs_root = nfs_root,
136 .vfs_statfs = nfs_statfs,
137 .vfs_sync = nfs_sync,
138 .vfs_init = nfs_init,
139 .vfs_uninit = nfs_uninit
984263bc
MD
140};
141VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
142
143/*
144 * This structure must be filled in by a primary bootstrap or bootstrap
145 * server for a diskless/dataless machine. It is initialized below just
146 * to ensure that it is allocated to initialized data (.data not .bss).
147 */
148struct nfs_diskless nfs_diskless = { { { 0 } } };
149struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
150int nfs_diskless_valid = 0;
151
152SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
153 &nfs_diskless_valid, 0, "");
154
155SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
156 nfsv3_diskless.root_hostnam, 0, "");
157
158SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
159 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
160 "%Ssockaddr_in", "");
161
162SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
163 nfsv3_diskless.swap_hostnam, 0, "");
164
165SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
166 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
167 "%Ssockaddr_in","");
168
169
a6ee311a
RG
170void nfsargs_ntoh (struct nfs_args *);
171static int nfs_mountdiskless (char *, char *, int,
984263bc 172 struct sockaddr_in *, struct nfs_args *,
dadab5e9 173 struct thread *, struct vnode **,
a6ee311a
RG
174 struct mount **);
175static void nfs_convert_diskless (void);
176static void nfs_convert_oargs (struct nfs_args *args,
177 struct onfs_args *oargs);
984263bc 178
eefb74c1
MD
179/*
180 * Calculate the buffer I/O block size to use. The maximum V2 block size
181 * is typically 8K, the maximum datagram size is typically 16K, and the
182 * maximum V3 block size is typically 32K. The buffer cache tends to work
183 * best with 16K blocks but we allow 32K for TCP connections.
184 *
185 * We force the block size to be at least a page for buffer cache efficiency.
186 */
3fcb1ab8 187static int
eefb74c1 188nfs_iosize(int v3, int sotype)
984263bc
MD
189{
190 int iosize;
eefb74c1 191 int iomax;
984263bc 192
eefb74c1
MD
193 if (v3) {
194 if (sotype == SOCK_STREAM)
195 iomax = NFS_MAXDATA;
196 else
197 iomax = NFS_MAXDGRAMDATA;
198 } else {
199 iomax = NFS_V2MAXDATA;
200 }
201 if ((iosize = nfs_io_size) > iomax)
202 iosize = iomax;
203 if (iosize < PAGE_SIZE)
204 iosize = PAGE_SIZE;
8d429613
MD
205
206 /*
207 * This is an aweful hack but until the buffer cache is rewritten
208 * we need it. The problem is that when you combine write() with
209 * mmap() the vm_page->valid bits can become weird looking
210 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers
211 * at the file EOF. To solve the problem the BIO system needs to
212 * be guarenteed that the NFS iosize for regular files will be a
213 * multiple of PAGE_SIZE so it can invalidate the whole page
214 * rather then just the piece of it owned by the buffer when
215 * NFS does vinvalbuf() calls.
216 */
217 if (iosize & PAGE_MASK)
218 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
984263bc
MD
219 return iosize;
220}
221
222static void
3fcb1ab8 223nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
984263bc
MD
224{
225 args->version = NFS_ARGSVERSION;
226 args->addr = oargs->addr;
227 args->addrlen = oargs->addrlen;
228 args->sotype = oargs->sotype;
229 args->proto = oargs->proto;
230 args->fh = oargs->fh;
231 args->fhsize = oargs->fhsize;
232 args->flags = oargs->flags;
233 args->wsize = oargs->wsize;
234 args->rsize = oargs->rsize;
235 args->readdirsize = oargs->readdirsize;
236 args->timeo = oargs->timeo;
237 args->retrans = oargs->retrans;
238 args->maxgrouplist = oargs->maxgrouplist;
239 args->readahead = oargs->readahead;
984263bc
MD
240 args->deadthresh = oargs->deadthresh;
241 args->hostname = oargs->hostname;
242}
243
244static void
3fcb1ab8 245nfs_convert_diskless(void)
984263bc 246{
0e8ff41a
MD
247 int i;
248
984263bc
MD
249 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
250 sizeof(struct ifaliasreq));
251 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
252 sizeof(struct sockaddr_in));
253 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
0e8ff41a 254
984263bc 255 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
0e8ff41a
MD
256 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
257 for (i = NFSX_V2FH - 1; i >= 0; --i) {
258 if (nfs_diskless.swap_fh[i])
259 break;
260 }
261 if (i < 0)
262 nfsv3_diskless.swap_fhsize = 0;
263
984263bc
MD
264 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
265 sizeof(struct sockaddr_in));
266 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
267 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
268 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
269 sizeof(struct ucred));
270 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
0e8ff41a 271
984263bc 272 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
0e8ff41a
MD
273 nfsv3_diskless.root_fhsize = NFSX_V2FH;
274 for (i = NFSX_V2FH - 1; i >= 0; --i) {
275 if (nfs_diskless.root_fh[i])
276 break;
277 }
278 if (i < 0)
279 nfsv3_diskless.root_fhsize = 0;
280
984263bc
MD
281 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
282 sizeof(struct sockaddr_in));
283 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
284 nfsv3_diskless.root_time = nfs_diskless.root_time;
285 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
286 MAXHOSTNAMELEN);
287 nfs_diskless_valid = 3;
288}
289
290/*
291 * nfs statfs call
292 */
293int
acde96db 294nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
984263bc 295{
40393ded
RG
296 struct vnode *vp;
297 struct nfs_statfs *sfp;
298 caddr_t cp;
299 u_int32_t *tl;
300 int32_t t1, t2;
984263bc
MD
301 caddr_t bpos, dpos, cp2;
302 struct nfsmount *nmp = VFSTONFS(mp);
acde96db 303 thread_t td = curthread;
984263bc
MD
304 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
305 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
984263bc
MD
306 struct nfsnode *np;
307 u_quad_t tquad;
308
309#ifndef nolint
60233e58 310 sfp = NULL;
984263bc
MD
311#endif
312 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
313 if (error)
314 return (error);
315 vp = NFSTOV(np);
acde96db 316 /* ignore the passed cred */
984263bc
MD
317 cred = crget();
318 cred->cr_ngroups = 1;
319 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
3b568787 320 (void)nfs_fsinfo(nmp, vp, td);
984263bc
MD
321 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
322 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
323 nfsm_fhtom(vp, v3);
dadab5e9 324 nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
984263bc 325 if (v3)
5a9187cb 326 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
984263bc
MD
327 if (error) {
328 if (mrep != NULL)
329 m_freem(mrep);
330 goto nfsmout;
331 }
332 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
333 sbp->f_flags = nmp->nm_flag;
eefb74c1
MD
334 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
335
984263bc
MD
336 if (v3) {
337 sbp->f_bsize = NFS_FABLKSIZE;
338 tquad = fxdr_hyper(&sfp->sf_tbytes);
339 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
340 tquad = fxdr_hyper(&sfp->sf_fbytes);
341 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
342 tquad = fxdr_hyper(&sfp->sf_abytes);
343 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
344 sbp->f_files = (fxdr_unsigned(int32_t,
345 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
346 sbp->f_ffree = (fxdr_unsigned(int32_t,
347 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
348 } else {
349 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
350 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
351 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
352 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
353 sbp->f_files = 0;
354 sbp->f_ffree = 0;
355 }
356 if (sbp != &mp->mnt_stat) {
357 sbp->f_type = mp->mnt_vfc->vfc_typenum;
984263bc
MD
358 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
359 }
6b08710e
MD
360 m_freem(mrep);
361nfsmout:
984263bc
MD
362 vput(vp);
363 crfree(cred);
364 return (error);
365}
366
367/*
368 * nfs version 3 fsinfo rpc call
369 */
370int
3b568787 371nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
984263bc 372{
40393ded
RG
373 struct nfsv3_fsinfo *fsp;
374 caddr_t cp;
375 int32_t t1, t2;
376 u_int32_t *tl, pref, max;
984263bc
MD
377 caddr_t bpos, dpos, cp2;
378 int error = 0, retattr;
379 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
380 u_int64_t maxfsize;
381
382 nfsstats.rpccnt[NFSPROC_FSINFO]++;
383 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
384 nfsm_fhtom(vp, 1);
c1cf1e59 385 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
5a9187cb 386 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
984263bc
MD
387 if (!error) {
388 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
389 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
390 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
391 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
392 ~(NFS_FABLKSIZE - 1);
393 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
394 if (max < nmp->nm_wsize && max > 0) {
395 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
396 if (nmp->nm_wsize == 0)
397 nmp->nm_wsize = max;
398 }
399 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
400 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
401 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
402 ~(NFS_FABLKSIZE - 1);
403 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
404 if (max < nmp->nm_rsize && max > 0) {
405 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
406 if (nmp->nm_rsize == 0)
407 nmp->nm_rsize = max;
408 }
409 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
410 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
411 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
412 ~(NFS_DIRBLKSIZ - 1);
413 if (max < nmp->nm_readdirsize && max > 0) {
414 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
415 if (nmp->nm_readdirsize == 0)
416 nmp->nm_readdirsize = max;
417 }
418 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
419 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
420 nmp->nm_maxfilesize = maxfsize;
421 nmp->nm_state |= NFSSTA_GOTFSINFO;
422 }
6b08710e
MD
423 m_freem(mrep);
424nfsmout:
984263bc
MD
425 return (error);
426}
427
428/*
429 * Mount a remote root fs via. nfs. This depends on the info in the
430 * nfs_diskless structure that has been filled in properly by some primary
431 * bootstrap.
432 * It goes something like this:
433 * - do enough of "ifconfig" by calling ifioctl() so that the system
434 * can talk to the server
435 * - If nfs_diskless.mygateway is filled in, use that address as
436 * a default gateway.
437 * - build the rootfs mount point and call mountnfs() to do the rest.
438 */
439int
acde96db 440nfs_mountroot(struct mount *mp)
984263bc
MD
441{
442 struct mount *swap_mp;
443 struct nfsv3_diskless *nd = &nfsv3_diskless;
444 struct socket *so;
445 struct vnode *vp;
dadab5e9 446 struct thread *td = curthread; /* XXX */
984263bc
MD
447 int error, i;
448 u_long l;
449 char buf[128];
450
451#if defined(BOOTP_NFSROOT) && defined(BOOTP)
452 bootpc_init(); /* use bootp to get nfs_diskless filled in */
453#endif
454
455 /*
456 * XXX time must be non-zero when we init the interface or else
457 * the arp code will wedge...
458 */
3a6f9faf
MD
459 while (mycpu->gd_time_seconds == 0)
460 tsleep(mycpu, 0, "arpkludge", 10);
984263bc 461
a2103e90
MD
462 /*
463 * The boot code may have passed us a diskless structure.
464 */
465 if (nfs_diskless_valid == 1)
0b9984b8
MD
466 nfs_convert_diskless();
467
a2103e90 468#define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr))
086c1d7e 469 kprintf("nfs_mountroot: interface %s ip %s",
a2103e90
MD
470 nd->myif.ifra_name,
471 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
086c1d7e 472 kprintf(" bcast %s",
a2103e90 473 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
086c1d7e 474 kprintf(" mask %s\n",
a2103e90
MD
475 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
476#undef SINP
984263bc
MD
477
478 /*
479 * XXX splnet, so networks will receive...
480 */
165dba55 481 crit_enter();
984263bc 482
984263bc 483 /*
a2103e90
MD
484 * BOOTP does not necessarily have to be compiled into the kernel
485 * for an NFS root to work. If we inherited the network
486 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
487 * out our interface for us and all we need to do is ifconfig the
488 * interface. We only do this if the interface has not already been
489 * ifconfig'd by e.g. BOOTP.
984263bc 490 */
dadab5e9 491 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
a2103e90 492 if (error) {
984263bc
MD
493 panic("nfs_mountroot: socreate(%04x): %d",
494 nd->myif.ifra_addr.sa_family, error);
984263bc 495 }
a2103e90 496
87de5057 497 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
984263bc
MD
498 if (error)
499 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
a2103e90 500
9ba76b73 501 soclose(so, FNONBLOCK);
984263bc
MD
502
503 /*
504 * If the gateway field is filled in, set it as the default route.
505 */
506 if (nd->mygateway.sin_len != 0) {
507 struct sockaddr_in mask, sin;
508
509 bzero((caddr_t)&mask, sizeof(mask));
510 sin = mask;
511 sin.sin_family = AF_INET;
512 sin.sin_len = sizeof(sin);
086c1d7e 513 kprintf("nfs_mountroot: gateway %s\n",
0b9984b8 514 inet_ntoa(nd->mygateway.sin_addr));
ecdefdda
MD
515 error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
516 (struct sockaddr *)&nd->mygateway,
517 (struct sockaddr *)&mask,
518 RTF_UP | RTF_GATEWAY);
984263bc 519 if (error)
086c1d7e 520 kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
984263bc
MD
521 }
522
523 /*
524 * Create the rootfs mount point.
525 */
526 nd->root_args.fh = nd->root_fh;
527 nd->root_args.fhsize = nd->root_fhsize;
528 l = ntohl(nd->root_saddr.sin_addr.s_addr);
f8c7a42d 529 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
984263bc
MD
530 (l >> 24) & 0xff, (l >> 16) & 0xff,
531 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
5dacbb2e 532 kprintf("NFS_ROOT: %s\n",buf);
984263bc 533 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
dadab5e9 534 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
5dacbb2e 535 mp->mnt_vfc->vfc_refcount--;
0b9984b8 536 crit_exit();
984263bc
MD
537 return (error);
538 }
539
540 swap_mp = NULL;
541 if (nd->swap_nblks) {
542
543 /* Convert to DEV_BSIZE instead of Kilobyte */
544 nd->swap_nblks *= 2;
545
546 /*
547 * Create a fake mount point just for the swap vnode so that the
548 * swap file can be on a different server from the rootfs.
549 */
550 nd->swap_args.fh = nd->swap_fh;
551 nd->swap_args.fhsize = nd->swap_fhsize;
552 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
f8c7a42d 553 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
984263bc
MD
554 (l >> 24) & 0xff, (l >> 16) & 0xff,
555 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
086c1d7e 556 kprintf("NFS SWAP: %s\n",buf);
984263bc 557 if ((error = nfs_mountdiskless(buf, "/swap", 0,
0b9984b8
MD
558 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) {
559 crit_exit();
984263bc 560 return (error);
0b9984b8 561 }
f9642f56 562 vfs_unbusy(swap_mp);
984263bc
MD
563
564 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
565 nd->swap_nblks * DEV_BSIZE ;
566
567 /*
568 * Since the swap file is not the root dir of a file system,
569 * hack it to a regular file.
570 */
984263bc 571 vp->v_flag = 0;
597aea93 572 vref(vp);
1c843a13 573 nfs_setvtype(vp, VREG);
e4c9c0c8 574 swaponvp(td, vp, nd->swap_nblks);
984263bc
MD
575 }
576
577 mp->mnt_flag |= MNT_ROOTFS;
f9642f56 578 vfs_unbusy(mp);
984263bc
MD
579
580 /*
581 * This is not really an nfs issue, but it is much easier to
582 * set hostname here and then let the "/etc/rc.xxx" files
583 * mount the right /var based upon its preset value.
584 */
585 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
586 hostname[MAXHOSTNAMELEN - 1] = '\0';
587 for (i = 0; i < MAXHOSTNAMELEN; i++)
588 if (hostname[i] == '\0')
589 break;
590 inittodr(ntohl(nd->root_time));
0b9984b8 591 crit_exit();
984263bc
MD
592 return (0);
593}
594
595/*
596 * Internal version of mount system call for diskless setup.
597 */
598static int
dadab5e9
MD
599nfs_mountdiskless(char *path, char *which, int mountflag,
600 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
601 struct vnode **vpp, struct mount **mpp)
984263bc
MD
602{
603 struct mount *mp;
604 struct sockaddr *nam;
984263bc 605 int didalloc = 0;
0e8ff41a 606 int error;
984263bc
MD
607
608 mp = *mpp;
609
610 if (mp == NULL) {
611 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
086c1d7e 612 kprintf("nfs_mountroot: NFS not configured");
984263bc
MD
613 return (error);
614 }
615 didalloc = 1;
616 }
984263bc
MD
617 mp->mnt_kern_flag = 0;
618 mp->mnt_flag = mountflag;
cfa2ba21 619 nam = dup_sockaddr((struct sockaddr *)sin);
0e8ff41a
MD
620
621#if defined(BOOTP) || defined(NFS_ROOT)
622 if (args->fhsize == 0) {
df0a038e
SW
623 char *xpath = path;
624
086c1d7e 625 kprintf("NFS_ROOT: No FH passed from loader, attempting mount rpc...");
5dacbb2e
MD
626 while (*xpath && *xpath != ':')
627 ++xpath;
628 if (*xpath)
629 ++xpath;
0e8ff41a 630 args->fhsize = 0;
5dacbb2e 631 error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td);
0e8ff41a 632 if (error) {
5dacbb2e 633 kprintf("failed error %d.\n", error);
0e8ff41a
MD
634 goto haderror;
635 }
086c1d7e 636 kprintf("success!\n");
0e8ff41a
MD
637 }
638#endif
639
984263bc 640 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
04ef15fb 641#if defined(BOOTP) || defined(NFS_ROOT)
0e8ff41a 642haderror:
04ef15fb 643#endif
086c1d7e 644 kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error);
984263bc 645 mp->mnt_vfc->vfc_refcount--;
f9642f56 646 vfs_unbusy(mp);
984263bc 647 if (didalloc)
efda3bd0 648 kfree(mp, M_MOUNT);
984263bc
MD
649 FREE(nam, M_SONAME);
650 return (error);
651 }
984263bc
MD
652 *mpp = mp;
653 return (0);
654}
655
656static void
3fcb1ab8 657nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp)
984263bc 658{
984263bc
MD
659 int adjsock;
660 int maxio;
661
165dba55 662 crit_enter();
984263bc
MD
663 /*
664 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
665 * no sense in that context.
666 */
667 if (argp->sotype == SOCK_STREAM)
668 nmp->nm_flag &= ~NFSMNT_NOCONN;
669
670 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
671 if ((argp->flags & NFSMNT_NFSV3) == 0)
672 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
673
674 /* Re-bind if rsrvd port requested and wasn't on one */
675 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
676 && (argp->flags & NFSMNT_RESVPORT);
677 /* Also re-bind if we're switching to/from a connected UDP socket */
678 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
679 (argp->flags & NFSMNT_NOCONN));
680
681 /* Update flags atomically. Don't change the lock bits. */
682 nmp->nm_flag = argp->flags | nmp->nm_flag;
165dba55 683 crit_exit();
984263bc
MD
684
685 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
686 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
687 if (nmp->nm_timeo < NFS_MINTIMEO)
688 nmp->nm_timeo = NFS_MINTIMEO;
689 else if (nmp->nm_timeo > NFS_MAXTIMEO)
690 nmp->nm_timeo = NFS_MAXTIMEO;
691 }
692
693 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
694 nmp->nm_retry = argp->retrans;
695 if (nmp->nm_retry > NFS_MAXREXMIT)
696 nmp->nm_retry = NFS_MAXREXMIT;
697 }
698
eefb74c1 699 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
984263bc
MD
700
701 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
702 nmp->nm_wsize = argp->wsize;
703 /* Round down to multiple of blocksize */
704 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
705 if (nmp->nm_wsize <= 0)
706 nmp->nm_wsize = NFS_FABLKSIZE;
707 }
708 if (nmp->nm_wsize > maxio)
709 nmp->nm_wsize = maxio;
710 if (nmp->nm_wsize > MAXBSIZE)
711 nmp->nm_wsize = MAXBSIZE;
712
713 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
714 nmp->nm_rsize = argp->rsize;
715 /* Round down to multiple of blocksize */
716 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
717 if (nmp->nm_rsize <= 0)
718 nmp->nm_rsize = NFS_FABLKSIZE;
719 }
720 if (nmp->nm_rsize > maxio)
721 nmp->nm_rsize = maxio;
722 if (nmp->nm_rsize > MAXBSIZE)
723 nmp->nm_rsize = MAXBSIZE;
724
725 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
726 nmp->nm_readdirsize = argp->readdirsize;
727 }
728 if (nmp->nm_readdirsize > maxio)
729 nmp->nm_readdirsize = maxio;
730 if (nmp->nm_readdirsize > nmp->nm_rsize)
731 nmp->nm_readdirsize = nmp->nm_rsize;
732
733 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
734 nmp->nm_acregmin = argp->acregmin;
735 else
736 nmp->nm_acregmin = NFS_MINATTRTIMO;
737 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
738 nmp->nm_acregmax = argp->acregmax;
739 else
740 nmp->nm_acregmax = NFS_MAXATTRTIMO;
741 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
742 nmp->nm_acdirmin = argp->acdirmin;
743 else
744 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
745 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
746 nmp->nm_acdirmax = argp->acdirmax;
747 else
748 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
749 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
750 nmp->nm_acdirmin = nmp->nm_acdirmax;
751 if (nmp->nm_acregmin > nmp->nm_acregmax)
752 nmp->nm_acregmin = nmp->nm_acregmax;
753
754 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
755 if (argp->maxgrouplist <= NFS_MAXGRPS)
756 nmp->nm_numgrps = argp->maxgrouplist;
757 else
758 nmp->nm_numgrps = NFS_MAXGRPS;
759 }
760 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
761 if (argp->readahead <= NFS_MAXRAHEAD)
762 nmp->nm_readahead = argp->readahead;
763 else
764 nmp->nm_readahead = NFS_MAXRAHEAD;
765 }
984263bc 766 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
e07fef60 767 if (argp->deadthresh <= NFS_NEVERDEAD)
984263bc
MD
768 nmp->nm_deadthresh = argp->deadthresh;
769 else
e07fef60 770 nmp->nm_deadthresh = NFS_NEVERDEAD;
984263bc
MD
771 }
772
773 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
774 (nmp->nm_soproto != argp->proto));
775 nmp->nm_sotype = argp->sotype;
776 nmp->nm_soproto = argp->proto;
777
778 if (nmp->nm_so && adjsock) {
779 nfs_safedisconnect(nmp);
780 if (nmp->nm_sotype == SOCK_DGRAM)
60233e58 781 while (nfs_connect(nmp, NULL)) {
086c1d7e 782 kprintf("nfs_args: retrying connect\n");
377d4740 783 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
984263bc
MD
784 }
785 }
786}
787
788/*
789 * VFS Operations.
790 *
791 * mount system call
792 * It seems a bit dumb to copyinstr() the host and path here and then
793 * bcopy() them in mountnfs(), but I wanted to detect errors before
794 * doing the sockargs() call because sockargs() allocates an mbuf and
795 * an error after that means that I have to release the mbuf.
796 */
797/* ARGSUSED */
798static int
acde96db 799nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
984263bc
MD
800{
801 int error;
802 struct nfs_args args;
803 struct sockaddr *nam;
804 struct vnode *vp;
805 char pth[MNAMELEN], hst[MNAMELEN];
806 size_t len;
807 u_char nfh[NFSX_V3FHMAX];
808
809 if (path == NULL) {
810 nfs_mountroot(mp);
811 return (0);
812 }
813 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
814 if (error)
815 return (error);
816 if (args.version != NFS_ARGSVERSION) {
817#ifdef COMPAT_PRELITE2
818 /*
819 * If the argument version is unknown, then assume the
820 * caller is a pre-lite2 4.4BSD client and convert its
821 * arguments.
822 */
823 struct onfs_args oargs;
824 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
825 if (error)
826 return (error);
827 nfs_convert_oargs(&args,&oargs);
828#else /* !COMPAT_PRELITE2 */
829 return (EPROGMISMATCH);
830#endif /* COMPAT_PRELITE2 */
831 }
832 if (mp->mnt_flag & MNT_UPDATE) {
40393ded 833 struct nfsmount *nmp = VFSTONFS(mp);
984263bc
MD
834
835 if (nmp == NULL)
836 return (EIO);
837 /*
838 * When doing an update, we can't change from or to
e07fef60 839 * v3, or change cookie translation
984263bc
MD
840 */
841 args.flags = (args.flags &
e07fef60 842 ~(NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/)) |
984263bc 843 (nmp->nm_flag &
e07fef60 844 (NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/));
984263bc
MD
845 nfs_decode_args(nmp, &args);
846 return (0);
847 }
848
849 /*
850 * Make the nfs_ip_paranoia sysctl serve as the default connection
851 * or no-connection mode for those protocols that support
852 * no-connection mode (the flag will be cleared later for protocols
853 * that do not support no-connection mode). This will allow a client
854 * to receive replies from a different IP then the request was
855 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
856 * not 0.
857 */
858 if (nfs_ip_paranoia == 0)
859 args.flags |= NFSMNT_NOCONN;
860 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
861 return (EINVAL);
862 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
863 if (error)
864 return (error);
865 error = copyinstr(path, pth, MNAMELEN-1, &len);
866 if (error)
867 return (error);
868 bzero(&pth[len], MNAMELEN - len);
869 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
870 if (error)
871 return (error);
872 bzero(&hst[len], MNAMELEN - len);
873 /* sockargs() call must be after above copyin() calls */
874 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
875 if (error)
876 return (error);
877 args.fh = nfh;
878 error = mountnfs(&args, mp, nam, pth, hst, &vp);
879 return (error);
880}
881
882/*
883 * Common code for mount and mountroot
884 */
885static int
dadab5e9
MD
886mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
887 char *pth, char *hst, struct vnode **vpp)
984263bc 888{
40393ded 889 struct nfsmount *nmp;
984263bc
MD
890 struct nfsnode *np;
891 int error;
984263bc
MD
892
893 if (mp->mnt_flag & MNT_UPDATE) {
894 nmp = VFSTONFS(mp);
895 /* update paths, file handles, etc, here XXX */
896 FREE(nam, M_SONAME);
897 return (0);
898 } else {
899 nmp = zalloc(nfsmount_zone);
900 bzero((caddr_t)nmp, sizeof (struct nfsmount));
8684e6f9
MD
901 mtx_init(&nmp->nm_rxlock);
902 mtx_init(&nmp->nm_txlock);
984263bc 903 TAILQ_INIT(&nmp->nm_uidlruhead);
81b5c339 904 TAILQ_INIT(&nmp->nm_bioq);
e21aec5b 905 TAILQ_INIT(&nmp->nm_reqq);
984263bc
MD
906 mp->mnt_data = (qaddr_t)nmp;
907 }
908 vfs_getnewfsid(mp);
909 nmp->nm_mountp = mp;
984263bc
MD
910
911 /*
912 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
913 * high, depending on whether we end up with negative offsets in
914 * the client or server somewhere. 2GB-1 may be safer.
915 *
916 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
917 * that we can handle until we find out otherwise.
918 * XXX Our "safe" limit on the client is what we can store in our
919 * buffer cache using signed(!) block numbers.
920 */
921 if ((argp->flags & NFSMNT_NFSV3) == 0)
922 nmp->nm_maxfilesize = 0xffffffffLL;
923 else
924 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
925
926 nmp->nm_timeo = NFS_TIMEO;
927 nmp->nm_retry = NFS_RETRANS;
eefb74c1
MD
928 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
929 nmp->nm_rsize = nmp->nm_wsize;
984263bc
MD
930 nmp->nm_readdirsize = NFS_READDIRSIZE;
931 nmp->nm_numgrps = NFS_MAXGRPS;
932 nmp->nm_readahead = NFS_DEFRAHEAD;
0639df29 933 nmp->nm_deadthresh = NFS_DEADTHRESH;
984263bc
MD
934 nmp->nm_fhsize = argp->fhsize;
935 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
936 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
984263bc
MD
937 nmp->nm_nam = nam;
938 /* Set up the sockets and per-host congestion */
939 nmp->nm_sotype = argp->sotype;
940 nmp->nm_soproto = argp->proto;
3b568787 941 nmp->nm_cred = crhold(proc0.p_ucred);
984263bc
MD
942
943 nfs_decode_args(nmp, argp);
944
945 /*
946 * For Connection based sockets (TCP,...) defer the connect until
947 * the first request, in case the server is not responding.
948 */
949 if (nmp->nm_sotype == SOCK_DGRAM &&
60233e58 950 (error = nfs_connect(nmp, NULL)))
984263bc
MD
951 goto bad;
952
953 /*
954 * This is silly, but it has to be set so that vinifod() works.
955 * We do not want to do an nfs_statfs() here since we can get
956 * stuck on a dead server and we are holding a lock on the mount
957 * point.
958 */
eefb74c1
MD
959 mp->mnt_stat.f_iosize =
960 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
961
0961aa92
MD
962 /*
963 * Install vop_ops for our vnops
964 */
66a1ddf5
MD
965 vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
966 vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
967 vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
0961aa92 968
984263bc
MD
969 /*
970 * A reference count is needed on the nfsnode representing the
971 * remote root. If this object is not persistent, then backward
972 * traversals of the mount point (i.e. "..") will not work if
973 * the nfsnode gets flushed out of the cache. Ufs does not have
974 * this problem, because one can identify root inodes by their
975 * number == ROOTINO (2).
976 */
977 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
978 if (error)
979 goto bad;
980 *vpp = NFSTOV(np);
981
982 /*
eefb74c1
MD
983 * Retrieval of mountpoint attributes is delayed until nfs_rot
984 * or nfs_statfs are first called. This will happen either when
985 * we first traverse the mount point or if somebody does a df(1).
986 *
987 * NFSSTA_GOTFSINFO is used to flag if we have successfully
988 * retrieved mountpoint attributes. In the case of NFSv3 we
989 * also flag static fsinfo.
984263bc 990 */
eefb74c1
MD
991 if (*vpp != NULL)
992 (*vpp)->v_type = VNON;
984263bc
MD
993
994 /*
995 * Lose the lock but keep the ref.
996 */
a11aaa81 997 vn_unlock(*vpp);
984263bc 998
52e1cf57
MD
999 /*
1000 * Start the reader and writer threads.
1001 */
1002 lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread,
1003 NULL, 0, -1, "nfsiod_rx");
1004 lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread,
1005 NULL, 0, -1, "nfsiod_tx");
1006
984263bc
MD
1007 return (0);
1008bad:
1009 nfs_disconnect(nmp);
3b568787 1010 nfs_free_mount(nmp);
984263bc
MD
1011 FREE(nam, M_SONAME);
1012 return (error);
1013}
1014
1015/*
1016 * unmount system call
1017 */
1018static int
acde96db 1019nfs_unmount(struct mount *mp, int mntflags)
984263bc 1020{
40393ded 1021 struct nfsmount *nmp;
984263bc
MD
1022 int error, flags = 0;
1023
1024 if (mntflags & MNT_FORCE)
1025 flags |= FORCECLOSE;
1026 nmp = VFSTONFS(mp);
1027 /*
1028 * Goes something like this..
1029 * - Call vflush() to clear out vnodes for this file system
1030 * - Close the socket
1031 * - Free up the data structures
1032 */
1033 /* In the forced case, cancel any outstanding requests. */
1034 if (flags & FORCECLOSE) {
1035 error = nfs_nmcancelreqs(nmp);
1036 if (error)
1037 return (error);
1038 }
1039 /*
e07fef60 1040 * Must handshake with nfs_clientd() if it is active. XXX
984263bc
MD
1041 */
1042 nmp->nm_state |= NFSSTA_DISMINPROG;
984263bc
MD
1043
1044 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1045 error = vflush(mp, 1, flags);
1046 if (error) {
1047 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1048 return (error);
1049 }
1050
1051 /*
1052 * We are now committed to the unmount.
1053 * For NQNFS, let the server daemon free the nfsmount structure.
1054 */
e07fef60 1055 if (nmp->nm_flag & NFSMNT_KERB)
984263bc
MD
1056 nmp->nm_state |= NFSSTA_DISMNT;
1057
1058 nfs_disconnect(nmp);
1059 FREE(nmp->nm_nam, M_SONAME);
1060
52e1cf57
MD
1061 if ((nmp->nm_flag & NFSMNT_KERB) == 0) {
1062 nfssvc_iod_stop(nmp);
3b568787 1063 nfs_free_mount(nmp);
52e1cf57 1064 }
984263bc
MD
1065 return (0);
1066}
1067
3b568787
MD
1068void
1069nfs_free_mount(struct nfsmount *nmp)
1070{
1071 if (nmp->nm_cred) {
1072 crfree(nmp->nm_cred);
1073 nmp->nm_cred = NULL;
1074 }
1075 zfree(nfsmount_zone, nmp);
1076}
1077
984263bc
MD
1078/*
1079 * Return root of a filesystem
1080 */
1081static int
3fcb1ab8 1082nfs_root(struct mount *mp, struct vnode **vpp)
984263bc 1083{
40393ded 1084 struct vnode *vp;
984263bc 1085 struct nfsmount *nmp;
eefb74c1 1086 struct vattr attrs;
984263bc
MD
1087 struct nfsnode *np;
1088 int error;
1089
1090 nmp = VFSTONFS(mp);
1091 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1092 if (error)
1093 return (error);
1094 vp = NFSTOV(np);
eefb74c1
MD
1095
1096 /*
1097 * Get transfer parameters and root vnode attributes
1098 */
1099 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1100 if (nmp->nm_flag & NFSMNT_NFSV3) {
14aa2dd7 1101 error = nfs_fsinfo(nmp, vp, curthread);
eefb74c1
MD
1102 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1103 } else {
87de5057 1104 if ((error = VOP_GETATTR(vp, &attrs)) == 0)
eefb74c1
MD
1105 nmp->nm_state |= NFSSTA_GOTFSINFO;
1106
1107 }
14aa2dd7
MD
1108 } else {
1109 /*
1110 * The root vnode is usually cached by the namecache so do not
1111 * try to avoid going over the wire even if we have previous
1112 * information cached. A stale NFS mount can loop
1113 * forever resolving the root vnode if we return no-error when
1114 * there is in fact an error.
1115 */
1116 np->n_attrstamp = 0;
1117 error = VOP_GETATTR(vp, &attrs);
eefb74c1 1118 }
984263bc 1119 if (vp->v_type == VNON)
1c843a13 1120 nfs_setvtype(vp, VDIR);
984263bc 1121 vp->v_flag = VROOT;
14aa2dd7
MD
1122 if (error)
1123 vput(vp);
1124 else
1125 *vpp = vp;
1126 return (error);
984263bc
MD
1127}
1128
1129extern int syncprt;
1130
b2eb81cd
MD
1131struct scaninfo {
1132 int rescan;
b2eb81cd
MD
1133 int waitfor;
1134 int allerror;
1135};
1136
1137static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
5fd012e0 1138static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
b2eb81cd 1139
984263bc
MD
1140/*
1141 * Flush out the buffer cache
1142 */
1143/* ARGSUSED */
1144static int
87de5057 1145nfs_sync(struct mount *mp, int waitfor)
984263bc 1146{
b2eb81cd
MD
1147 struct scaninfo scaninfo;
1148 int error;
1149
a482a28a 1150 scaninfo.rescan = 1;
b2eb81cd
MD
1151 scaninfo.waitfor = waitfor;
1152 scaninfo.allerror = 0;
984263bc
MD
1153
1154 /*
1155 * Force stale buffer cache information to be flushed.
1156 */
b2eb81cd
MD
1157 error = 0;
1158 while (error == 0 && scaninfo.rescan) {
1159 scaninfo.rescan = 0;
5fd012e0 1160 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
b2eb81cd 1161 nfs_sync_scan2, &scaninfo);
984263bc 1162 }
b2eb81cd
MD
1163 return(error);
1164}
1165
3fcb1ab8 1166static int
b2eb81cd
MD
1167nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1168{
1169 struct scaninfo *info = data;
1170
a11aaa81 1171 if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree))
b2eb81cd
MD
1172 return(-1);
1173 if (info->waitfor == MNT_LAZY)
1174 return(-1);
1175 return(0);
1176}
1177
3fcb1ab8 1178static int
5fd012e0 1179nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
b2eb81cd
MD
1180{
1181 struct scaninfo *info = data;
1182 int error;
1183
87de5057 1184 error = VOP_FSYNC(vp, info->waitfor);
b2eb81cd
MD
1185 if (error)
1186 info->allerror = error;
b2eb81cd 1187 return(0);
984263bc
MD
1188}
1189