2 * Copyright (c) 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
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 1.33 2005/09/17 07:43:12 dillon Exp $
41 #include "opt_bootp.h"
42 #include "opt_nfsroot.h"
44 #include <sys/param.h>
45 #include <sys/sockio.h>
47 #include <sys/vnode.h>
48 #include <sys/kernel.h>
49 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/systm.h>
58 #include <vm/vm_extern.h>
59 #include <vm/vm_zone.h>
62 #include <net/route.h>
63 #include <netinet/in.h>
65 #include <sys/thread2.h>
73 #include "nfsm_subs.h"
74 #include "nfsdiskless.h"
76 #include "nfsmountrpc.h"
78 extern int nfs_mountroot(struct mount *mp);
79 extern void bootpc_init(void);
82 extern struct vnodeopv_entry_desc nfsv2_vnodeop_entries[];
83 extern struct vnodeopv_entry_desc nfsv2_fifoop_entries[];
84 extern struct vnodeopv_entry_desc nfsv2_specop_entries[];
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_NQLEASE, "NQNFS Lease", "Nqnfs lease");
93 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
95 vm_zone_t nfsmount_zone;
97 struct nfsstats nfsstats;
98 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
99 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
100 &nfsstats, nfsstats, "");
101 static int nfs_ip_paranoia = 1;
102 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
103 &nfs_ip_paranoia, 0, "");
106 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
110 * Tunable to determine the Read/Write unit size. Maximum value
111 * is NFS_MAXDATA. We also default to NFS_MAXDATA.
113 static int nfs_io_size = NFS_MAXDATA;
114 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
115 &nfs_io_size, 0, "NFS optimal I/O unit size");
117 static void nfs_decode_args (struct nfsmount *nmp,
118 struct nfs_args *argp);
119 static int mountnfs (struct nfs_args *,struct mount *,
120 struct sockaddr *,char *,char *,struct vnode **);
121 static int nfs_mount ( struct mount *mp, char *path, caddr_t data,
123 static int nfs_unmount ( struct mount *mp, int mntflags,
125 static int nfs_root ( struct mount *mp, struct vnode **vpp);
126 static int nfs_statfs ( struct mount *mp, struct statfs *sbp,
128 static int nfs_sync ( struct mount *mp, int waitfor,
132 * nfs vfs operations.
134 static struct vfsops nfs_vfsops = {
135 .vfs_mount = nfs_mount,
136 .vfs_unmount = nfs_unmount,
137 .vfs_root = nfs_root,
138 .vfs_statfs = nfs_statfs,
139 .vfs_sync = nfs_sync,
140 .vfs_init = nfs_init,
141 .vfs_uninit = nfs_uninit
143 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
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).
150 struct nfs_diskless nfs_diskless = { { { 0 } } };
151 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
152 int nfs_diskless_valid = 0;
154 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
155 &nfs_diskless_valid, 0, "");
157 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
158 nfsv3_diskless.root_hostnam, 0, "");
160 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
161 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
162 "%Ssockaddr_in", "");
164 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
165 nfsv3_diskless.swap_hostnam, 0, "");
167 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
168 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
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 **,
177 static void nfs_convert_diskless (void);
178 static void nfs_convert_oargs (struct nfs_args *args,
179 struct onfs_args *oargs);
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.
187 * We force the block size to be at least a page for buffer cache efficiency.
191 nfs_iosize(int v3, int sotype)
197 if (sotype == SOCK_STREAM)
200 iomax = NFS_MAXDGRAMDATA;
202 iomax = NFS_V2MAXDATA;
204 if ((iosize = nfs_io_size) > iomax)
206 if (iosize < PAGE_SIZE)
210 * This is an aweful hack but until the buffer cache is rewritten
211 * we need it. The problem is that when you combine write() with
212 * mmap() the vm_page->valid bits can become weird looking
213 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers
214 * at the file EOF. To solve the problem the BIO system needs to
215 * be guarenteed that the NFS iosize for regular files will be a
216 * multiple of PAGE_SIZE so it can invalidate the whole page
217 * rather then just the piece of it owned by the buffer when
218 * NFS does vinvalbuf() calls.
220 if (iosize & PAGE_MASK)
221 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
226 nfs_convert_oargs(args, oargs)
227 struct nfs_args *args;
228 struct onfs_args *oargs;
230 args->version = NFS_ARGSVERSION;
231 args->addr = oargs->addr;
232 args->addrlen = oargs->addrlen;
233 args->sotype = oargs->sotype;
234 args->proto = oargs->proto;
235 args->fh = oargs->fh;
236 args->fhsize = oargs->fhsize;
237 args->flags = oargs->flags;
238 args->wsize = oargs->wsize;
239 args->rsize = oargs->rsize;
240 args->readdirsize = oargs->readdirsize;
241 args->timeo = oargs->timeo;
242 args->retrans = oargs->retrans;
243 args->maxgrouplist = oargs->maxgrouplist;
244 args->readahead = oargs->readahead;
245 args->leaseterm = oargs->leaseterm;
246 args->deadthresh = oargs->deadthresh;
247 args->hostname = oargs->hostname;
251 nfs_convert_diskless()
255 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
256 sizeof(struct ifaliasreq));
257 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
258 sizeof(struct sockaddr_in));
259 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
261 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
262 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
263 for (i = NFSX_V2FH - 1; i >= 0; --i) {
264 if (nfs_diskless.swap_fh[i])
268 nfsv3_diskless.swap_fhsize = 0;
270 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
271 sizeof(struct sockaddr_in));
272 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
273 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
274 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
275 sizeof(struct ucred));
276 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
278 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
279 nfsv3_diskless.root_fhsize = NFSX_V2FH;
280 for (i = NFSX_V2FH - 1; i >= 0; --i) {
281 if (nfs_diskless.root_fh[i])
285 nfsv3_diskless.root_fhsize = 0;
287 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
288 sizeof(struct sockaddr_in));
289 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
290 nfsv3_diskless.root_time = nfs_diskless.root_time;
291 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
293 nfs_diskless_valid = 3;
300 nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
303 struct nfs_statfs *sfp;
307 caddr_t bpos, dpos, cp2;
308 struct nfsmount *nmp = VFSTONFS(mp);
309 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
310 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
316 sfp = (struct nfs_statfs *)0;
318 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
323 cred->cr_ngroups = 1;
324 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
325 (void)nfs_fsinfo(nmp, vp, td);
326 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
327 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
329 nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
331 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
337 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
338 sbp->f_flags = nmp->nm_flag;
339 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
342 sbp->f_bsize = NFS_FABLKSIZE;
343 tquad = fxdr_hyper(&sfp->sf_tbytes);
344 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
345 tquad = fxdr_hyper(&sfp->sf_fbytes);
346 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
347 tquad = fxdr_hyper(&sfp->sf_abytes);
348 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
349 sbp->f_files = (fxdr_unsigned(int32_t,
350 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
351 sbp->f_ffree = (fxdr_unsigned(int32_t,
352 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
354 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
355 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
356 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
357 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
361 if (sbp != &mp->mnt_stat) {
362 sbp->f_type = mp->mnt_vfc->vfc_typenum;
363 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
373 * nfs version 3 fsinfo rpc call
376 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
378 struct nfsv3_fsinfo *fsp;
381 u_int32_t *tl, pref, max;
382 caddr_t bpos, dpos, cp2;
383 int error = 0, retattr;
384 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
387 nfsstats.rpccnt[NFSPROC_FSINFO]++;
388 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
390 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
391 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
393 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
394 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
395 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
396 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
397 ~(NFS_FABLKSIZE - 1);
398 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
399 if (max < nmp->nm_wsize && max > 0) {
400 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
401 if (nmp->nm_wsize == 0)
404 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
405 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
406 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
407 ~(NFS_FABLKSIZE - 1);
408 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
409 if (max < nmp->nm_rsize && max > 0) {
410 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
411 if (nmp->nm_rsize == 0)
414 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
415 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
416 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
417 ~(NFS_DIRBLKSIZ - 1);
418 if (max < nmp->nm_readdirsize && max > 0) {
419 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
420 if (nmp->nm_readdirsize == 0)
421 nmp->nm_readdirsize = max;
423 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
424 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
425 nmp->nm_maxfilesize = maxfsize;
426 nmp->nm_state |= NFSSTA_GOTFSINFO;
434 * Mount a remote root fs via. nfs. This depends on the info in the
435 * nfs_diskless structure that has been filled in properly by some primary
437 * It goes something like this:
438 * - do enough of "ifconfig" by calling ifioctl() so that the system
439 * can talk to the server
440 * - If nfs_diskless.mygateway is filled in, use that address as
442 * - build the rootfs mount point and call mountnfs() to do the rest.
448 struct mount *swap_mp;
449 struct nfsv3_diskless *nd = &nfsv3_diskless;
452 struct thread *td = curthread; /* XXX */
457 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
458 bootpc_init(); /* use bootp to get nfs_diskless filled in */
462 * XXX time must be non-zero when we init the interface or else
463 * the arp code will wedge...
465 while (mycpu->gd_time_seconds == 0)
466 tsleep(mycpu, 0, "arpkludge", 10);
469 * The boot code may have passed us a diskless structure.
471 if (nfs_diskless_valid == 1)
472 nfs_convert_diskless();
474 #define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr))
475 printf("nfs_mountroot: interface %s ip %s",
477 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
479 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
481 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
485 * XXX splnet, so networks will receive...
490 * BOOTP does not necessarily have to be compiled into the kernel
491 * for an NFS root to work. If we inherited the network
492 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
493 * out our interface for us and all we need to do is ifconfig the
494 * interface. We only do this if the interface has not already been
495 * ifconfig'd by e.g. BOOTP.
497 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
499 panic("nfs_mountroot: socreate(%04x): %d",
500 nd->myif.ifra_addr.sa_family, error);
503 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
505 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
510 * If the gateway field is filled in, set it as the default route.
512 if (nd->mygateway.sin_len != 0) {
513 struct sockaddr_in mask, sin;
515 bzero((caddr_t)&mask, sizeof(mask));
517 sin.sin_family = AF_INET;
518 sin.sin_len = sizeof(sin);
519 printf("nfs_mountroot: gateway %s\n",
520 inet_ntoa(nd->mygateway.sin_addr));
521 error = rtrequest(RTM_ADD, (struct sockaddr *)&sin,
522 (struct sockaddr *)&nd->mygateway,
523 (struct sockaddr *)&mask,
524 RTF_UP | RTF_GATEWAY, (struct rtentry **)0);
526 printf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
530 * Create the rootfs mount point.
532 nd->root_args.fh = nd->root_fh;
533 nd->root_args.fhsize = nd->root_fhsize;
534 l = ntohl(nd->root_saddr.sin_addr.s_addr);
535 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
536 (l >> 24) & 0xff, (l >> 16) & 0xff,
537 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
538 printf("NFS ROOT: %s\n",buf);
539 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
540 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
542 mp->mnt_vfc->vfc_refcount--;
543 free(swap_mp, M_MOUNT);
550 if (nd->swap_nblks) {
552 /* Convert to DEV_BSIZE instead of Kilobyte */
556 * Create a fake mount point just for the swap vnode so that the
557 * swap file can be on a different server from the rootfs.
559 nd->swap_args.fh = nd->swap_fh;
560 nd->swap_args.fhsize = nd->swap_fhsize;
561 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
562 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
563 (l >> 24) & 0xff, (l >> 16) & 0xff,
564 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
565 printf("NFS SWAP: %s\n",buf);
566 if ((error = nfs_mountdiskless(buf, "/swap", 0,
567 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) {
571 vfs_unbusy(swap_mp, td);
573 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
574 nd->swap_nblks * DEV_BSIZE ;
577 * Since the swap file is not the root dir of a file system,
578 * hack it to a regular file.
583 swaponvp(td, vp, nd->swap_nblks);
586 mp->mnt_flag |= MNT_ROOTFS;
587 mp->mnt_vnodecovered = NULLVP;
591 * This is not really an nfs issue, but it is much easier to
592 * set hostname here and then let the "/etc/rc.xxx" files
593 * mount the right /var based upon its preset value.
595 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
596 hostname[MAXHOSTNAMELEN - 1] = '\0';
597 for (i = 0; i < MAXHOSTNAMELEN; i++)
598 if (hostname[i] == '\0')
600 inittodr(ntohl(nd->root_time));
606 * Internal version of mount system call for diskless setup.
609 nfs_mountdiskless(char *path, char *which, int mountflag,
610 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
611 struct vnode **vpp, struct mount **mpp)
614 struct sockaddr *nam;
621 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
622 printf("nfs_mountroot: NFS not configured");
627 mp->mnt_kern_flag = 0;
628 mp->mnt_flag = mountflag;
629 nam = dup_sockaddr((struct sockaddr *)sin);
631 #if defined(BOOTP) || defined(NFS_ROOT)
632 if (args->fhsize == 0) {
633 printf("NFS_ROOT: No FH passed from loader, attempting mount rpc...");
635 error = md_mount(sin, which, args->fh, &args->fhsize, args, td);
640 printf("success!\n");
644 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
645 #if defined(BOOTP) || defined(NFS_ROOT)
648 printf("nfs_mountroot: mount %s on %s: %d", path, which, error);
649 mp->mnt_vfc->vfc_refcount--;
661 nfs_decode_args(nmp, argp)
662 struct nfsmount *nmp;
663 struct nfs_args *argp;
670 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
671 * no sense in that context.
673 if (argp->sotype == SOCK_STREAM)
674 nmp->nm_flag &= ~NFSMNT_NOCONN;
676 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
677 if ((argp->flags & NFSMNT_NFSV3) == 0)
678 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
680 /* Re-bind if rsrvd port requested and wasn't on one */
681 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
682 && (argp->flags & NFSMNT_RESVPORT);
683 /* Also re-bind if we're switching to/from a connected UDP socket */
684 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
685 (argp->flags & NFSMNT_NOCONN));
687 /* Update flags atomically. Don't change the lock bits. */
688 nmp->nm_flag = argp->flags | nmp->nm_flag;
691 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
692 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
693 if (nmp->nm_timeo < NFS_MINTIMEO)
694 nmp->nm_timeo = NFS_MINTIMEO;
695 else if (nmp->nm_timeo > NFS_MAXTIMEO)
696 nmp->nm_timeo = NFS_MAXTIMEO;
699 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
700 nmp->nm_retry = argp->retrans;
701 if (nmp->nm_retry > NFS_MAXREXMIT)
702 nmp->nm_retry = NFS_MAXREXMIT;
705 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
707 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
708 nmp->nm_wsize = argp->wsize;
709 /* Round down to multiple of blocksize */
710 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
711 if (nmp->nm_wsize <= 0)
712 nmp->nm_wsize = NFS_FABLKSIZE;
714 if (nmp->nm_wsize > maxio)
715 nmp->nm_wsize = maxio;
716 if (nmp->nm_wsize > MAXBSIZE)
717 nmp->nm_wsize = MAXBSIZE;
719 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
720 nmp->nm_rsize = argp->rsize;
721 /* Round down to multiple of blocksize */
722 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
723 if (nmp->nm_rsize <= 0)
724 nmp->nm_rsize = NFS_FABLKSIZE;
726 if (nmp->nm_rsize > maxio)
727 nmp->nm_rsize = maxio;
728 if (nmp->nm_rsize > MAXBSIZE)
729 nmp->nm_rsize = MAXBSIZE;
731 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
732 nmp->nm_readdirsize = argp->readdirsize;
734 if (nmp->nm_readdirsize > maxio)
735 nmp->nm_readdirsize = maxio;
736 if (nmp->nm_readdirsize > nmp->nm_rsize)
737 nmp->nm_readdirsize = nmp->nm_rsize;
739 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
740 nmp->nm_acregmin = argp->acregmin;
742 nmp->nm_acregmin = NFS_MINATTRTIMO;
743 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
744 nmp->nm_acregmax = argp->acregmax;
746 nmp->nm_acregmax = NFS_MAXATTRTIMO;
747 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
748 nmp->nm_acdirmin = argp->acdirmin;
750 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
751 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
752 nmp->nm_acdirmax = argp->acdirmax;
754 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
755 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
756 nmp->nm_acdirmin = nmp->nm_acdirmax;
757 if (nmp->nm_acregmin > nmp->nm_acregmax)
758 nmp->nm_acregmin = nmp->nm_acregmax;
760 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
761 if (argp->maxgrouplist <= NFS_MAXGRPS)
762 nmp->nm_numgrps = argp->maxgrouplist;
764 nmp->nm_numgrps = NFS_MAXGRPS;
766 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
767 if (argp->readahead <= NFS_MAXRAHEAD)
768 nmp->nm_readahead = argp->readahead;
770 nmp->nm_readahead = NFS_MAXRAHEAD;
772 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2) {
773 if (argp->leaseterm <= NQ_MAXLEASE)
774 nmp->nm_leaseterm = argp->leaseterm;
776 nmp->nm_leaseterm = NQ_MAXLEASE;
778 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
779 if (argp->deadthresh <= NQ_NEVERDEAD)
780 nmp->nm_deadthresh = argp->deadthresh;
782 nmp->nm_deadthresh = NQ_NEVERDEAD;
785 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
786 (nmp->nm_soproto != argp->proto));
787 nmp->nm_sotype = argp->sotype;
788 nmp->nm_soproto = argp->proto;
790 if (nmp->nm_so && adjsock) {
791 nfs_safedisconnect(nmp);
792 if (nmp->nm_sotype == SOCK_DGRAM)
793 while (nfs_connect(nmp, (struct nfsreq *)0)) {
794 printf("nfs_args: retrying connect\n");
795 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
804 * It seems a bit dumb to copyinstr() the host and path here and then
805 * bcopy() them in mountnfs(), but I wanted to detect errors before
806 * doing the sockargs() call because sockargs() allocates an mbuf and
807 * an error after that means that I have to release the mbuf.
811 nfs_mount(struct mount *mp, char *path, caddr_t data, struct thread *td)
814 struct nfs_args args;
815 struct sockaddr *nam;
817 char pth[MNAMELEN], hst[MNAMELEN];
819 u_char nfh[NFSX_V3FHMAX];
825 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
828 if (args.version != NFS_ARGSVERSION) {
829 #ifdef COMPAT_PRELITE2
831 * If the argument version is unknown, then assume the
832 * caller is a pre-lite2 4.4BSD client and convert its
835 struct onfs_args oargs;
836 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
839 nfs_convert_oargs(&args,&oargs);
840 #else /* !COMPAT_PRELITE2 */
841 return (EPROGMISMATCH);
842 #endif /* COMPAT_PRELITE2 */
844 if (mp->mnt_flag & MNT_UPDATE) {
845 struct nfsmount *nmp = VFSTONFS(mp);
850 * When doing an update, we can't change from or to
851 * v3 and/or nqnfs, or change cookie translation
853 args.flags = (args.flags &
854 ~(NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)) |
856 (NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/));
857 nfs_decode_args(nmp, &args);
862 * Make the nfs_ip_paranoia sysctl serve as the default connection
863 * or no-connection mode for those protocols that support
864 * no-connection mode (the flag will be cleared later for protocols
865 * that do not support no-connection mode). This will allow a client
866 * to receive replies from a different IP then the request was
867 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
870 if (nfs_ip_paranoia == 0)
871 args.flags |= NFSMNT_NOCONN;
872 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
874 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
877 error = copyinstr(path, pth, MNAMELEN-1, &len);
880 bzero(&pth[len], MNAMELEN - len);
881 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
884 bzero(&hst[len], MNAMELEN - len);
885 /* sockargs() call must be after above copyin() calls */
886 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
890 error = mountnfs(&args, mp, nam, pth, hst, &vp);
895 * Common code for mount and mountroot
898 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
899 char *pth, char *hst, struct vnode **vpp)
901 struct nfsmount *nmp;
905 if (mp->mnt_flag & MNT_UPDATE) {
907 /* update paths, file handles, etc, here XXX */
911 nmp = zalloc(nfsmount_zone);
912 bzero((caddr_t)nmp, sizeof (struct nfsmount));
913 TAILQ_INIT(&nmp->nm_uidlruhead);
914 TAILQ_INIT(&nmp->nm_bufq);
915 mp->mnt_data = (qaddr_t)nmp;
919 if (argp->flags & NFSMNT_NQNFS) {
921 * We have to set mnt_maxsymlink to a non-zero value so
922 * that COMPAT_43 routines will know that we are setting
923 * the d_type field in directories (and can zero it for
924 * unsuspecting binaries).
926 mp->mnt_maxsymlinklen = 1;
930 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
931 * high, depending on whether we end up with negative offsets in
932 * the client or server somewhere. 2GB-1 may be safer.
934 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
935 * that we can handle until we find out otherwise.
936 * XXX Our "safe" limit on the client is what we can store in our
937 * buffer cache using signed(!) block numbers.
939 if ((argp->flags & NFSMNT_NFSV3) == 0)
940 nmp->nm_maxfilesize = 0xffffffffLL;
942 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
944 nmp->nm_timeo = NFS_TIMEO;
945 nmp->nm_retry = NFS_RETRANS;
946 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
947 nmp->nm_rsize = nmp->nm_wsize;
948 nmp->nm_readdirsize = NFS_READDIRSIZE;
949 nmp->nm_numgrps = NFS_MAXGRPS;
950 nmp->nm_readahead = NFS_DEFRAHEAD;
951 nmp->nm_leaseterm = NQ_DEFLEASE;
952 nmp->nm_deadthresh = NQ_DEADTHRESH;
953 CIRCLEQ_INIT(&nmp->nm_timerhead);
954 nmp->nm_inprog = NULLVP;
955 nmp->nm_fhsize = argp->fhsize;
956 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
957 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
959 /* Set up the sockets and per-host congestion */
960 nmp->nm_sotype = argp->sotype;
961 nmp->nm_soproto = argp->proto;
962 nmp->nm_cred = crhold(proc0.p_ucred);
964 nfs_decode_args(nmp, argp);
967 * For Connection based sockets (TCP,...) defer the connect until
968 * the first request, in case the server is not responding.
970 if (nmp->nm_sotype == SOCK_DGRAM &&
971 (error = nfs_connect(nmp, (struct nfsreq *)0)))
975 * This is silly, but it has to be set so that vinifod() works.
976 * We do not want to do an nfs_statfs() here since we can get
977 * stuck on a dead server and we are holding a lock on the mount
980 mp->mnt_stat.f_iosize =
981 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
984 * Install vop_ops for our vnops
986 vfs_add_vnodeops(mp, &mp->mnt_vn_norm_ops,
987 nfsv2_vnodeop_entries, 0);
988 vfs_add_vnodeops(mp, &mp->mnt_vn_spec_ops,
989 nfsv2_specop_entries, 0);
990 vfs_add_vnodeops(mp, &mp->mnt_vn_fifo_ops,
991 nfsv2_fifoop_entries, 0);
994 * A reference count is needed on the nfsnode representing the
995 * remote root. If this object is not persistent, then backward
996 * traversals of the mount point (i.e. "..") will not work if
997 * the nfsnode gets flushed out of the cache. Ufs does not have
998 * this problem, because one can identify root inodes by their
999 * number == ROOTINO (2).
1001 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1007 * Retrieval of mountpoint attributes is delayed until nfs_rot
1008 * or nfs_statfs are first called. This will happen either when
1009 * we first traverse the mount point or if somebody does a df(1).
1011 * NFSSTA_GOTFSINFO is used to flag if we have successfully
1012 * retrieved mountpoint attributes. In the case of NFSv3 we
1013 * also flag static fsinfo.
1016 (*vpp)->v_type = VNON;
1019 * Lose the lock but keep the ref.
1021 VOP_UNLOCK(*vpp, 0, curthread);
1025 nfs_disconnect(nmp);
1026 nfs_free_mount(nmp);
1027 FREE(nam, M_SONAME);
1032 * unmount system call
1035 nfs_unmount(struct mount *mp, int mntflags, struct thread *td)
1037 struct nfsmount *nmp;
1038 int error, flags = 0;
1040 if (mntflags & MNT_FORCE)
1041 flags |= FORCECLOSE;
1044 * Goes something like this..
1045 * - Call vflush() to clear out vnodes for this file system
1046 * - Close the socket
1047 * - Free up the data structures
1049 /* In the forced case, cancel any outstanding requests. */
1050 if (flags & FORCECLOSE) {
1051 error = nfs_nmcancelreqs(nmp);
1056 * Must handshake with nqnfs_clientd() if it is active.
1058 nmp->nm_state |= NFSSTA_DISMINPROG;
1059 while (nmp->nm_inprog != NULLVP)
1060 (void) tsleep((caddr_t)&lbolt, 0, "nfsdism", 0);
1062 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1063 error = vflush(mp, 1, flags);
1065 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1070 * We are now committed to the unmount.
1071 * For NQNFS, let the server daemon free the nfsmount structure.
1073 if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB))
1074 nmp->nm_state |= NFSSTA_DISMNT;
1076 nfs_disconnect(nmp);
1077 FREE(nmp->nm_nam, M_SONAME);
1079 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
1080 nfs_free_mount(nmp);
1085 nfs_free_mount(struct nfsmount *nmp)
1088 crfree(nmp->nm_cred);
1089 nmp->nm_cred = NULL;
1091 zfree(nfsmount_zone, nmp);
1095 * Return root of a filesystem
1103 struct nfsmount *nmp;
1109 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1115 * Get transfer parameters and root vnode attributes
1117 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1118 if (nmp->nm_flag & NFSMNT_NFSV3) {
1119 nfs_fsinfo(nmp, vp, curthread);
1120 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1122 if ((error = VOP_GETATTR(vp, &attrs, curthread)) == 0)
1123 nmp->nm_state |= NFSSTA_GOTFSINFO;
1127 if (vp->v_type == VNON)
1143 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1144 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1147 * Flush out the buffer cache
1151 nfs_sync(struct mount *mp, int waitfor, struct thread *td)
1153 struct scaninfo scaninfo;
1156 scaninfo.rescan = 0;
1158 scaninfo.waitfor = waitfor;
1159 scaninfo.allerror = 0;
1162 * Force stale buffer cache information to be flushed.
1165 while (error == 0 && scaninfo.rescan) {
1166 scaninfo.rescan = 0;
1167 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1168 nfs_sync_scan2, &scaninfo);
1175 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1177 struct scaninfo *info = data;
1179 if (VOP_ISLOCKED(vp, NULL) || RB_EMPTY(&vp->v_rbdirty_tree))
1181 if (info->waitfor == MNT_LAZY)
1188 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1190 struct scaninfo *info = data;
1193 error = VOP_FSYNC(vp, info->waitfor, info->td);
1195 info->allerror = error;