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 1.14 2004/03/10 02:07:52 dillon Exp $
41 #include "opt_bootp.h"
43 #include <sys/param.h>
44 #include <sys/sockio.h>
46 #include <sys/vnode.h>
47 #include <sys/kernel.h>
48 #include <sys/sysctl.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/systm.h>
57 #include <vm/vm_extern.h>
58 #include <vm/vm_zone.h>
61 #include <net/route.h>
62 #include <netinet/in.h>
70 #include "nfsm_subs.h"
71 #include "nfsdiskless.h"
74 extern int nfs_mountroot(struct mount *mp);
75 extern void bootpc_init(void);
79 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
80 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
81 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
82 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
83 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
84 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
85 MALLOC_DEFINE(M_NQLEASE, "NQNFS Lease", "Nqnfs lease");
86 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
88 vm_zone_t nfsmount_zone;
90 struct nfsstats nfsstats;
91 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
92 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
93 &nfsstats, nfsstats, "");
94 static int nfs_ip_paranoia = 1;
95 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
96 &nfs_ip_paranoia, 0, "");
99 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
103 * Tunable to determine the Read/Write unit size. Maximum value
104 * is NFS_MAXDATA. We also default to NFS_MAXDATA.
106 static int nfs_io_size = NFS_MAXDATA;
107 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
108 &nfs_io_size, 0, "NFS optimal I/O unit size");
110 static void nfs_decode_args (struct nfsmount *nmp,
111 struct nfs_args *argp);
112 static int mountnfs (struct nfs_args *,struct mount *,
113 struct sockaddr *,char *,char *,struct vnode **);
114 static int nfs_mount ( struct mount *mp, char *path, caddr_t data,
115 struct nameidata *ndp, struct thread *td);
116 static int nfs_unmount ( struct mount *mp, int mntflags,
118 static int nfs_root ( struct mount *mp, struct vnode **vpp);
119 static int nfs_statfs ( struct mount *mp, struct statfs *sbp,
121 static int nfs_sync ( struct mount *mp, int waitfor,
125 * nfs vfs operations.
127 static struct vfsops nfs_vfsops = {
136 vfs_stdfhtovp, /* shouldn't happen */
138 vfs_stdvptofh, /* shouldn't happen */
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)
212 nfs_convert_oargs(args, oargs)
213 struct nfs_args *args;
214 struct onfs_args *oargs;
216 args->version = NFS_ARGSVERSION;
217 args->addr = oargs->addr;
218 args->addrlen = oargs->addrlen;
219 args->sotype = oargs->sotype;
220 args->proto = oargs->proto;
221 args->fh = oargs->fh;
222 args->fhsize = oargs->fhsize;
223 args->flags = oargs->flags;
224 args->wsize = oargs->wsize;
225 args->rsize = oargs->rsize;
226 args->readdirsize = oargs->readdirsize;
227 args->timeo = oargs->timeo;
228 args->retrans = oargs->retrans;
229 args->maxgrouplist = oargs->maxgrouplist;
230 args->readahead = oargs->readahead;
231 args->leaseterm = oargs->leaseterm;
232 args->deadthresh = oargs->deadthresh;
233 args->hostname = oargs->hostname;
237 nfs_convert_diskless()
239 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
240 sizeof(struct ifaliasreq));
241 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
242 sizeof(struct sockaddr_in));
243 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
244 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
245 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
246 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
247 sizeof(struct sockaddr_in));
248 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
249 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
250 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
251 sizeof(struct ucred));
252 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
253 nfsv3_diskless.root_fhsize = NFSX_V2FH;
254 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
255 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
256 sizeof(struct sockaddr_in));
257 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
258 nfsv3_diskless.root_time = nfs_diskless.root_time;
259 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
261 nfs_diskless_valid = 3;
268 nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
271 struct nfs_statfs *sfp;
275 caddr_t bpos, dpos, cp2;
276 struct nfsmount *nmp = VFSTONFS(mp);
277 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
278 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
284 sfp = (struct nfs_statfs *)0;
286 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
291 cred->cr_ngroups = 1;
292 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
293 (void)nfs_fsinfo(nmp, vp, td);
294 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
295 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
297 nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
299 nfsm_postop_attr(vp, retattr);
305 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
306 sbp->f_flags = nmp->nm_flag;
307 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
310 sbp->f_bsize = NFS_FABLKSIZE;
311 tquad = fxdr_hyper(&sfp->sf_tbytes);
312 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
313 tquad = fxdr_hyper(&sfp->sf_fbytes);
314 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
315 tquad = fxdr_hyper(&sfp->sf_abytes);
316 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
317 sbp->f_files = (fxdr_unsigned(int32_t,
318 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
319 sbp->f_ffree = (fxdr_unsigned(int32_t,
320 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
322 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
323 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
324 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
325 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
329 if (sbp != &mp->mnt_stat) {
330 sbp->f_type = mp->mnt_vfc->vfc_typenum;
331 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
332 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
342 * nfs version 3 fsinfo rpc call
345 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
347 struct nfsv3_fsinfo *fsp;
350 u_int32_t *tl, pref, max;
351 caddr_t bpos, dpos, cp2;
352 int error = 0, retattr;
353 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
356 nfsstats.rpccnt[NFSPROC_FSINFO]++;
357 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
359 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
360 nfsm_postop_attr(vp, retattr);
362 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
363 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
364 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
365 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
366 ~(NFS_FABLKSIZE - 1);
367 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
368 if (max < nmp->nm_wsize && max > 0) {
369 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
370 if (nmp->nm_wsize == 0)
373 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
374 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
375 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
376 ~(NFS_FABLKSIZE - 1);
377 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
378 if (max < nmp->nm_rsize && max > 0) {
379 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
380 if (nmp->nm_rsize == 0)
383 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
384 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
385 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
386 ~(NFS_DIRBLKSIZ - 1);
387 if (max < nmp->nm_readdirsize && max > 0) {
388 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
389 if (nmp->nm_readdirsize == 0)
390 nmp->nm_readdirsize = max;
392 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
393 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
394 nmp->nm_maxfilesize = maxfsize;
395 nmp->nm_state |= NFSSTA_GOTFSINFO;
403 * Mount a remote root fs via. nfs. This depends on the info in the
404 * nfs_diskless structure that has been filled in properly by some primary
406 * It goes something like this:
407 * - do enough of "ifconfig" by calling ifioctl() so that the system
408 * can talk to the server
409 * - If nfs_diskless.mygateway is filled in, use that address as
411 * - build the rootfs mount point and call mountnfs() to do the rest.
417 struct mount *swap_mp;
418 struct nfsv3_diskless *nd = &nfsv3_diskless;
421 struct thread *td = curthread; /* XXX */
426 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
427 bootpc_init(); /* use bootp to get nfs_diskless filled in */
431 * XXX time must be non-zero when we init the interface or else
432 * the arp code will wedge...
434 while (mycpu->gd_time_seconds == 0)
435 tsleep(mycpu, 0, "arpkludge", 10);
437 if (nfs_diskless_valid==1)
438 nfs_convert_diskless();
441 * XXX splnet, so networks will receive...
446 /* Set up swap credentials. */
447 proc0.p_ucred->cr_uid = ntohl(nd->swap_ucred.cr_uid);
448 proc0.p_ucred->cr_gid = ntohl(nd->swap_ucred.cr_gid);
449 if ((proc0.p_ucred->cr_ngroups = ntohs(nd->swap_ucred.cr_ngroups)) >
451 proc0.p_ucred->cr_ngroups = NGROUPS;
452 for (i = 0; i < proc0.p_ucred->cr_ngroups; i++)
453 proc0.p_ucred->cr_groups[i] = ntohl(nd->swap_ucred.cr_groups[i]);
457 * Do enough of ifconfig(8) so that the critical net interface can
458 * talk to the server.
460 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
462 panic("nfs_mountroot: socreate(%04x): %d",
463 nd->myif.ifra_addr.sa_family, error);
465 #if 0 /* XXX Bad idea */
467 * We might not have been told the right interface, so we pass
468 * over the first ten interfaces of the same kind, until we get
469 * one of them configured.
472 for (i = strlen(nd->myif.ifra_name) - 1;
473 nd->myif.ifra_name[i] >= '0' &&
474 nd->myif.ifra_name[i] <= '9';
475 nd->myif.ifra_name[i] ++) {
476 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
481 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
483 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
487 * If the gateway field is filled in, set it as the default route.
489 if (nd->mygateway.sin_len != 0) {
490 struct sockaddr_in mask, sin;
492 bzero((caddr_t)&mask, sizeof(mask));
494 sin.sin_family = AF_INET;
495 sin.sin_len = sizeof(sin);
496 error = rtrequest(RTM_ADD, (struct sockaddr *)&sin,
497 (struct sockaddr *)&nd->mygateway,
498 (struct sockaddr *)&mask,
499 RTF_UP | RTF_GATEWAY, (struct rtentry **)0);
501 panic("nfs_mountroot: RTM_ADD: %d", error);
505 * Create the rootfs mount point.
507 nd->root_args.fh = nd->root_fh;
508 nd->root_args.fhsize = nd->root_fhsize;
509 l = ntohl(nd->root_saddr.sin_addr.s_addr);
510 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
511 (l >> 24) & 0xff, (l >> 16) & 0xff,
512 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
513 printf("NFS ROOT: %s\n",buf);
514 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
515 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
517 mp->mnt_vfc->vfc_refcount--;
518 free(swap_mp, M_MOUNT);
524 if (nd->swap_nblks) {
526 /* Convert to DEV_BSIZE instead of Kilobyte */
530 * Create a fake mount point just for the swap vnode so that the
531 * swap file can be on a different server from the rootfs.
533 nd->swap_args.fh = nd->swap_fh;
534 nd->swap_args.fhsize = nd->swap_fhsize;
535 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
536 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
537 (l >> 24) & 0xff, (l >> 16) & 0xff,
538 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
539 printf("NFS SWAP: %s\n",buf);
540 if ((error = nfs_mountdiskless(buf, "/swap", 0,
541 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0)
543 vfs_unbusy(swap_mp, td);
545 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
546 nd->swap_nblks * DEV_BSIZE ;
549 * Since the swap file is not the root dir of a file system,
550 * hack it to a regular file.
555 swaponvp(td, vp, NODEV, nd->swap_nblks);
558 mp->mnt_flag |= MNT_ROOTFS;
559 mp->mnt_vnodecovered = NULLVP;
564 * This is not really an nfs issue, but it is much easier to
565 * set hostname here and then let the "/etc/rc.xxx" files
566 * mount the right /var based upon its preset value.
568 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
569 hostname[MAXHOSTNAMELEN - 1] = '\0';
570 for (i = 0; i < MAXHOSTNAMELEN; i++)
571 if (hostname[i] == '\0')
573 inittodr(ntohl(nd->root_time));
578 * Internal version of mount system call for diskless setup.
581 nfs_mountdiskless(char *path, char *which, int mountflag,
582 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
583 struct vnode **vpp, struct mount **mpp)
586 struct sockaddr *nam;
593 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
594 printf("nfs_mountroot: NFS not configured");
600 mp->mnt_kern_flag = 0;
601 mp->mnt_flag = mountflag;
602 nam = dup_sockaddr((struct sockaddr *)sin, 1);
603 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
604 printf("nfs_mountroot: mount %s on %s: %d", path, which, error);
605 mp->mnt_vfc->vfc_refcount--;
612 (void) copystr(which, mp->mnt_stat.f_mntonname, MNAMELEN - 1, 0);
618 nfs_decode_args(nmp, argp)
619 struct nfsmount *nmp;
620 struct nfs_args *argp;
628 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
629 * no sense in that context.
631 if (argp->sotype == SOCK_STREAM)
632 nmp->nm_flag &= ~NFSMNT_NOCONN;
634 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
635 if ((argp->flags & NFSMNT_NFSV3) == 0)
636 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
638 /* Re-bind if rsrvd port requested and wasn't on one */
639 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
640 && (argp->flags & NFSMNT_RESVPORT);
641 /* Also re-bind if we're switching to/from a connected UDP socket */
642 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
643 (argp->flags & NFSMNT_NOCONN));
645 /* Update flags atomically. Don't change the lock bits. */
646 nmp->nm_flag = argp->flags | nmp->nm_flag;
649 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
650 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
651 if (nmp->nm_timeo < NFS_MINTIMEO)
652 nmp->nm_timeo = NFS_MINTIMEO;
653 else if (nmp->nm_timeo > NFS_MAXTIMEO)
654 nmp->nm_timeo = NFS_MAXTIMEO;
657 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
658 nmp->nm_retry = argp->retrans;
659 if (nmp->nm_retry > NFS_MAXREXMIT)
660 nmp->nm_retry = NFS_MAXREXMIT;
663 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
665 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
666 nmp->nm_wsize = argp->wsize;
667 /* Round down to multiple of blocksize */
668 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
669 if (nmp->nm_wsize <= 0)
670 nmp->nm_wsize = NFS_FABLKSIZE;
672 if (nmp->nm_wsize > maxio)
673 nmp->nm_wsize = maxio;
674 if (nmp->nm_wsize > MAXBSIZE)
675 nmp->nm_wsize = MAXBSIZE;
677 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
678 nmp->nm_rsize = argp->rsize;
679 /* Round down to multiple of blocksize */
680 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
681 if (nmp->nm_rsize <= 0)
682 nmp->nm_rsize = NFS_FABLKSIZE;
684 if (nmp->nm_rsize > maxio)
685 nmp->nm_rsize = maxio;
686 if (nmp->nm_rsize > MAXBSIZE)
687 nmp->nm_rsize = MAXBSIZE;
689 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
690 nmp->nm_readdirsize = argp->readdirsize;
692 if (nmp->nm_readdirsize > maxio)
693 nmp->nm_readdirsize = maxio;
694 if (nmp->nm_readdirsize > nmp->nm_rsize)
695 nmp->nm_readdirsize = nmp->nm_rsize;
697 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
698 nmp->nm_acregmin = argp->acregmin;
700 nmp->nm_acregmin = NFS_MINATTRTIMO;
701 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
702 nmp->nm_acregmax = argp->acregmax;
704 nmp->nm_acregmax = NFS_MAXATTRTIMO;
705 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
706 nmp->nm_acdirmin = argp->acdirmin;
708 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
709 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
710 nmp->nm_acdirmax = argp->acdirmax;
712 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
713 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
714 nmp->nm_acdirmin = nmp->nm_acdirmax;
715 if (nmp->nm_acregmin > nmp->nm_acregmax)
716 nmp->nm_acregmin = nmp->nm_acregmax;
718 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
719 if (argp->maxgrouplist <= NFS_MAXGRPS)
720 nmp->nm_numgrps = argp->maxgrouplist;
722 nmp->nm_numgrps = NFS_MAXGRPS;
724 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
725 if (argp->readahead <= NFS_MAXRAHEAD)
726 nmp->nm_readahead = argp->readahead;
728 nmp->nm_readahead = NFS_MAXRAHEAD;
730 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2) {
731 if (argp->leaseterm <= NQ_MAXLEASE)
732 nmp->nm_leaseterm = argp->leaseterm;
734 nmp->nm_leaseterm = NQ_MAXLEASE;
736 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
737 if (argp->deadthresh <= NQ_NEVERDEAD)
738 nmp->nm_deadthresh = argp->deadthresh;
740 nmp->nm_deadthresh = NQ_NEVERDEAD;
743 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
744 (nmp->nm_soproto != argp->proto));
745 nmp->nm_sotype = argp->sotype;
746 nmp->nm_soproto = argp->proto;
748 if (nmp->nm_so && adjsock) {
749 nfs_safedisconnect(nmp);
750 if (nmp->nm_sotype == SOCK_DGRAM)
751 while (nfs_connect(nmp, (struct nfsreq *)0)) {
752 printf("nfs_args: retrying connect\n");
753 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
762 * It seems a bit dumb to copyinstr() the host and path here and then
763 * bcopy() them in mountnfs(), but I wanted to detect errors before
764 * doing the sockargs() call because sockargs() allocates an mbuf and
765 * an error after that means that I have to release the mbuf.
769 nfs_mount(struct mount *mp, char *path, caddr_t data,
770 struct nameidata *ndp, struct thread *td)
773 struct nfs_args args;
774 struct sockaddr *nam;
776 char pth[MNAMELEN], hst[MNAMELEN];
778 u_char nfh[NFSX_V3FHMAX];
784 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
787 if (args.version != NFS_ARGSVERSION) {
788 #ifdef COMPAT_PRELITE2
790 * If the argument version is unknown, then assume the
791 * caller is a pre-lite2 4.4BSD client and convert its
794 struct onfs_args oargs;
795 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
798 nfs_convert_oargs(&args,&oargs);
799 #else /* !COMPAT_PRELITE2 */
800 return (EPROGMISMATCH);
801 #endif /* COMPAT_PRELITE2 */
803 if (mp->mnt_flag & MNT_UPDATE) {
804 struct nfsmount *nmp = VFSTONFS(mp);
809 * When doing an update, we can't change from or to
810 * v3 and/or nqnfs, or change cookie translation
812 args.flags = (args.flags &
813 ~(NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)) |
815 (NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/));
816 nfs_decode_args(nmp, &args);
821 * Make the nfs_ip_paranoia sysctl serve as the default connection
822 * or no-connection mode for those protocols that support
823 * no-connection mode (the flag will be cleared later for protocols
824 * that do not support no-connection mode). This will allow a client
825 * to receive replies from a different IP then the request was
826 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
829 if (nfs_ip_paranoia == 0)
830 args.flags |= NFSMNT_NOCONN;
831 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
833 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
836 error = copyinstr(path, pth, MNAMELEN-1, &len);
839 bzero(&pth[len], MNAMELEN - len);
840 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
843 bzero(&hst[len], MNAMELEN - len);
844 /* sockargs() call must be after above copyin() calls */
845 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
849 error = mountnfs(&args, mp, nam, pth, hst, &vp);
854 * Common code for mount and mountroot
857 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
858 char *pth, char *hst, struct vnode **vpp)
860 struct nfsmount *nmp;
864 if (mp->mnt_flag & MNT_UPDATE) {
866 /* update paths, file handles, etc, here XXX */
870 nmp = zalloc(nfsmount_zone);
871 bzero((caddr_t)nmp, sizeof (struct nfsmount));
872 TAILQ_INIT(&nmp->nm_uidlruhead);
873 TAILQ_INIT(&nmp->nm_bufq);
874 mp->mnt_data = (qaddr_t)nmp;
878 if (argp->flags & NFSMNT_NQNFS)
880 * We have to set mnt_maxsymlink to a non-zero value so
881 * that COMPAT_43 routines will know that we are setting
882 * the d_type field in directories (and can zero it for
883 * unsuspecting binaries).
885 mp->mnt_maxsymlinklen = 1;
888 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
889 * high, depending on whether we end up with negative offsets in
890 * the client or server somewhere. 2GB-1 may be safer.
892 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
893 * that we can handle until we find out otherwise.
894 * XXX Our "safe" limit on the client is what we can store in our
895 * buffer cache using signed(!) block numbers.
897 if ((argp->flags & NFSMNT_NFSV3) == 0)
898 nmp->nm_maxfilesize = 0xffffffffLL;
900 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
902 nmp->nm_timeo = NFS_TIMEO;
903 nmp->nm_retry = NFS_RETRANS;
904 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
905 nmp->nm_rsize = nmp->nm_wsize;
906 nmp->nm_readdirsize = NFS_READDIRSIZE;
907 nmp->nm_numgrps = NFS_MAXGRPS;
908 nmp->nm_readahead = NFS_DEFRAHEAD;
909 nmp->nm_leaseterm = NQ_DEFLEASE;
910 nmp->nm_deadthresh = NQ_DEADTHRESH;
911 CIRCLEQ_INIT(&nmp->nm_timerhead);
912 nmp->nm_inprog = NULLVP;
913 nmp->nm_fhsize = argp->fhsize;
914 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
915 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
916 bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN);
918 /* Set up the sockets and per-host congestion */
919 nmp->nm_sotype = argp->sotype;
920 nmp->nm_soproto = argp->proto;
921 nmp->nm_cred = crhold(proc0.p_ucred);
923 nfs_decode_args(nmp, argp);
926 * For Connection based sockets (TCP,...) defer the connect until
927 * the first request, in case the server is not responding.
929 if (nmp->nm_sotype == SOCK_DGRAM &&
930 (error = nfs_connect(nmp, (struct nfsreq *)0)))
934 * This is silly, but it has to be set so that vinifod() works.
935 * We do not want to do an nfs_statfs() here since we can get
936 * stuck on a dead server and we are holding a lock on the mount
939 mp->mnt_stat.f_iosize =
940 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
943 * A reference count is needed on the nfsnode representing the
944 * remote root. If this object is not persistent, then backward
945 * traversals of the mount point (i.e. "..") will not work if
946 * the nfsnode gets flushed out of the cache. Ufs does not have
947 * this problem, because one can identify root inodes by their
948 * number == ROOTINO (2).
950 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
956 * Retrieval of mountpoint attributes is delayed until nfs_rot
957 * or nfs_statfs are first called. This will happen either when
958 * we first traverse the mount point or if somebody does a df(1).
960 * NFSSTA_GOTFSINFO is used to flag if we have successfully
961 * retrieved mountpoint attributes. In the case of NFSv3 we
962 * also flag static fsinfo.
965 (*vpp)->v_type = VNON;
968 * Lose the lock but keep the ref.
970 VOP_UNLOCK(*vpp, NULL, 0, curthread);
981 * unmount system call
984 nfs_unmount(struct mount *mp, int mntflags, struct thread *td)
986 struct nfsmount *nmp;
987 int error, flags = 0;
989 if (mntflags & MNT_FORCE)
993 * Goes something like this..
994 * - Call vflush() to clear out vnodes for this file system
996 * - Free up the data structures
998 /* In the forced case, cancel any outstanding requests. */
999 if (flags & FORCECLOSE) {
1000 error = nfs_nmcancelreqs(nmp);
1005 * Must handshake with nqnfs_clientd() if it is active.
1007 nmp->nm_state |= NFSSTA_DISMINPROG;
1008 while (nmp->nm_inprog != NULLVP)
1009 (void) tsleep((caddr_t)&lbolt, 0, "nfsdism", 0);
1011 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1012 error = vflush(mp, 1, flags);
1014 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1019 * We are now committed to the unmount.
1020 * For NQNFS, let the server daemon free the nfsmount structure.
1022 if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB))
1023 nmp->nm_state |= NFSSTA_DISMNT;
1025 nfs_disconnect(nmp);
1026 FREE(nmp->nm_nam, M_SONAME);
1028 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
1029 nfs_free_mount(nmp);
1034 nfs_free_mount(struct nfsmount *nmp)
1037 crfree(nmp->nm_cred);
1038 nmp->nm_cred = NULL;
1040 zfree(nfsmount_zone, nmp);
1044 * Return root of a filesystem
1052 struct nfsmount *nmp;
1058 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1064 * Get transfer parameters and root vnode attributes
1066 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1067 if (nmp->nm_flag & NFSMNT_NFSV3) {
1068 nfs_fsinfo(nmp, vp, curthread);
1069 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1071 if ((error = VOP_GETATTR(vp, &attrs, curthread)) == 0)
1072 nmp->nm_state |= NFSSTA_GOTFSINFO;
1076 if (vp->v_type == VNON)
1086 * Flush out the buffer cache
1090 nfs_sync(struct mount *mp, int waitfor, struct thread *td)
1093 int error, allerror = 0;
1096 * Force stale buffer cache information to be flushed.
1099 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1101 vp = TAILQ_NEXT(vp, v_nmntvnodes)) {
1102 if (vp->v_flag & VPLACEMARKER) /* ZZZ */
1105 * If the vnode that we are about to sync is no longer
1106 * associated with this mount point, start over.
1108 if (vp->v_mount != mp)
1110 if (VOP_ISLOCKED(vp, NULL) || TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
1111 waitfor == MNT_LAZY)
1113 if (vget(vp, NULL, LK_EXCLUSIVE, td))
1115 error = VOP_FSYNC(vp, waitfor, td);