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 1.19 2004/06/06 19:16:11 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)
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()
253 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
254 sizeof(struct ifaliasreq));
255 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
256 sizeof(struct sockaddr_in));
257 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
258 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
259 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
260 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
261 sizeof(struct sockaddr_in));
262 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
263 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
264 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
265 sizeof(struct ucred));
266 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
267 nfsv3_diskless.root_fhsize = NFSX_V2FH;
268 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
269 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
270 sizeof(struct sockaddr_in));
271 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
272 nfsv3_diskless.root_time = nfs_diskless.root_time;
273 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
275 nfs_diskless_valid = 3;
282 nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
285 struct nfs_statfs *sfp;
289 caddr_t bpos, dpos, cp2;
290 struct nfsmount *nmp = VFSTONFS(mp);
291 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
292 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
298 sfp = (struct nfs_statfs *)0;
300 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
305 cred->cr_ngroups = 1;
306 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
307 (void)nfs_fsinfo(nmp, vp, td);
308 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
309 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
311 nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
313 nfsm_postop_attr(vp, retattr);
319 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
320 sbp->f_flags = nmp->nm_flag;
321 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
324 sbp->f_bsize = NFS_FABLKSIZE;
325 tquad = fxdr_hyper(&sfp->sf_tbytes);
326 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
327 tquad = fxdr_hyper(&sfp->sf_fbytes);
328 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
329 tquad = fxdr_hyper(&sfp->sf_abytes);
330 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
331 sbp->f_files = (fxdr_unsigned(int32_t,
332 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
333 sbp->f_ffree = (fxdr_unsigned(int32_t,
334 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
336 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
337 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
338 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
339 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
343 if (sbp != &mp->mnt_stat) {
344 sbp->f_type = mp->mnt_vfc->vfc_typenum;
345 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
346 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
356 * nfs version 3 fsinfo rpc call
359 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
361 struct nfsv3_fsinfo *fsp;
364 u_int32_t *tl, pref, max;
365 caddr_t bpos, dpos, cp2;
366 int error = 0, retattr;
367 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
370 nfsstats.rpccnt[NFSPROC_FSINFO]++;
371 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
373 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
374 nfsm_postop_attr(vp, retattr);
376 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
377 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
378 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
379 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
380 ~(NFS_FABLKSIZE - 1);
381 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
382 if (max < nmp->nm_wsize && max > 0) {
383 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
384 if (nmp->nm_wsize == 0)
387 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
388 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
389 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
390 ~(NFS_FABLKSIZE - 1);
391 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
392 if (max < nmp->nm_rsize && max > 0) {
393 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
394 if (nmp->nm_rsize == 0)
397 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
398 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
399 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
400 ~(NFS_DIRBLKSIZ - 1);
401 if (max < nmp->nm_readdirsize && max > 0) {
402 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
403 if (nmp->nm_readdirsize == 0)
404 nmp->nm_readdirsize = max;
406 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
407 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
408 nmp->nm_maxfilesize = maxfsize;
409 nmp->nm_state |= NFSSTA_GOTFSINFO;
417 * Mount a remote root fs via. nfs. This depends on the info in the
418 * nfs_diskless structure that has been filled in properly by some primary
420 * It goes something like this:
421 * - do enough of "ifconfig" by calling ifioctl() so that the system
422 * can talk to the server
423 * - If nfs_diskless.mygateway is filled in, use that address as
425 * - build the rootfs mount point and call mountnfs() to do the rest.
431 struct mount *swap_mp;
432 struct nfsv3_diskless *nd = &nfsv3_diskless;
435 struct thread *td = curthread; /* XXX */
440 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
441 bootpc_init(); /* use bootp to get nfs_diskless filled in */
445 * XXX time must be non-zero when we init the interface or else
446 * the arp code will wedge...
448 while (mycpu->gd_time_seconds == 0)
449 tsleep(mycpu, 0, "arpkludge", 10);
451 if (nfs_diskless_valid==1)
452 nfs_convert_diskless();
455 * XXX splnet, so networks will receive...
460 /* Set up swap credentials. */
461 proc0.p_ucred->cr_uid = ntohl(nd->swap_ucred.cr_uid);
462 proc0.p_ucred->cr_gid = ntohl(nd->swap_ucred.cr_gid);
463 if ((proc0.p_ucred->cr_ngroups = ntohs(nd->swap_ucred.cr_ngroups)) >
465 proc0.p_ucred->cr_ngroups = NGROUPS;
466 for (i = 0; i < proc0.p_ucred->cr_ngroups; i++)
467 proc0.p_ucred->cr_groups[i] = ntohl(nd->swap_ucred.cr_groups[i]);
471 * Do enough of ifconfig(8) so that the critical net interface can
472 * talk to the server.
474 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
476 panic("nfs_mountroot: socreate(%04x): %d",
477 nd->myif.ifra_addr.sa_family, error);
479 #if 0 /* XXX Bad idea */
481 * We might not have been told the right interface, so we pass
482 * over the first ten interfaces of the same kind, until we get
483 * one of them configured.
486 for (i = strlen(nd->myif.ifra_name) - 1;
487 nd->myif.ifra_name[i] >= '0' &&
488 nd->myif.ifra_name[i] <= '9';
489 nd->myif.ifra_name[i] ++) {
490 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
495 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
497 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
501 * If the gateway field is filled in, set it as the default route.
503 if (nd->mygateway.sin_len != 0) {
504 struct sockaddr_in mask, sin;
506 bzero((caddr_t)&mask, sizeof(mask));
508 sin.sin_family = AF_INET;
509 sin.sin_len = sizeof(sin);
510 error = rtrequest(RTM_ADD, (struct sockaddr *)&sin,
511 (struct sockaddr *)&nd->mygateway,
512 (struct sockaddr *)&mask,
513 RTF_UP | RTF_GATEWAY, (struct rtentry **)0);
515 panic("nfs_mountroot: RTM_ADD: %d", error);
519 * Create the rootfs mount point.
521 nd->root_args.fh = nd->root_fh;
522 nd->root_args.fhsize = nd->root_fhsize;
523 l = ntohl(nd->root_saddr.sin_addr.s_addr);
524 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
525 (l >> 24) & 0xff, (l >> 16) & 0xff,
526 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
527 printf("NFS ROOT: %s\n",buf);
528 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
529 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
531 mp->mnt_vfc->vfc_refcount--;
532 free(swap_mp, M_MOUNT);
538 if (nd->swap_nblks) {
540 /* Convert to DEV_BSIZE instead of Kilobyte */
544 * Create a fake mount point just for the swap vnode so that the
545 * swap file can be on a different server from the rootfs.
547 nd->swap_args.fh = nd->swap_fh;
548 nd->swap_args.fhsize = nd->swap_fhsize;
549 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
550 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
551 (l >> 24) & 0xff, (l >> 16) & 0xff,
552 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
553 printf("NFS SWAP: %s\n",buf);
554 if ((error = nfs_mountdiskless(buf, "/swap", 0,
555 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0)
557 vfs_unbusy(swap_mp, td);
559 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
560 nd->swap_nblks * DEV_BSIZE ;
563 * Since the swap file is not the root dir of a file system,
564 * hack it to a regular file.
569 swaponvp(td, vp, nd->swap_nblks);
572 mp->mnt_flag |= MNT_ROOTFS;
573 mp->mnt_vnodecovered = NULLVP;
578 * This is not really an nfs issue, but it is much easier to
579 * set hostname here and then let the "/etc/rc.xxx" files
580 * mount the right /var based upon its preset value.
582 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
583 hostname[MAXHOSTNAMELEN - 1] = '\0';
584 for (i = 0; i < MAXHOSTNAMELEN; i++)
585 if (hostname[i] == '\0')
587 inittodr(ntohl(nd->root_time));
592 * Internal version of mount system call for diskless setup.
595 nfs_mountdiskless(char *path, char *which, int mountflag,
596 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
597 struct vnode **vpp, struct mount **mpp)
600 struct sockaddr *nam;
607 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
608 printf("nfs_mountroot: NFS not configured");
614 mp->mnt_kern_flag = 0;
615 mp->mnt_flag = mountflag;
616 nam = dup_sockaddr((struct sockaddr *)sin);
617 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
618 printf("nfs_mountroot: mount %s on %s: %d", path, which, error);
619 mp->mnt_vfc->vfc_refcount--;
626 (void) copystr(which, mp->mnt_stat.f_mntonname, MNAMELEN - 1, 0);
632 nfs_decode_args(nmp, argp)
633 struct nfsmount *nmp;
634 struct nfs_args *argp;
642 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
643 * no sense in that context.
645 if (argp->sotype == SOCK_STREAM)
646 nmp->nm_flag &= ~NFSMNT_NOCONN;
648 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
649 if ((argp->flags & NFSMNT_NFSV3) == 0)
650 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
652 /* Re-bind if rsrvd port requested and wasn't on one */
653 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
654 && (argp->flags & NFSMNT_RESVPORT);
655 /* Also re-bind if we're switching to/from a connected UDP socket */
656 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
657 (argp->flags & NFSMNT_NOCONN));
659 /* Update flags atomically. Don't change the lock bits. */
660 nmp->nm_flag = argp->flags | nmp->nm_flag;
663 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
664 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
665 if (nmp->nm_timeo < NFS_MINTIMEO)
666 nmp->nm_timeo = NFS_MINTIMEO;
667 else if (nmp->nm_timeo > NFS_MAXTIMEO)
668 nmp->nm_timeo = NFS_MAXTIMEO;
671 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
672 nmp->nm_retry = argp->retrans;
673 if (nmp->nm_retry > NFS_MAXREXMIT)
674 nmp->nm_retry = NFS_MAXREXMIT;
677 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
679 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
680 nmp->nm_wsize = argp->wsize;
681 /* Round down to multiple of blocksize */
682 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
683 if (nmp->nm_wsize <= 0)
684 nmp->nm_wsize = NFS_FABLKSIZE;
686 if (nmp->nm_wsize > maxio)
687 nmp->nm_wsize = maxio;
688 if (nmp->nm_wsize > MAXBSIZE)
689 nmp->nm_wsize = MAXBSIZE;
691 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
692 nmp->nm_rsize = argp->rsize;
693 /* Round down to multiple of blocksize */
694 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
695 if (nmp->nm_rsize <= 0)
696 nmp->nm_rsize = NFS_FABLKSIZE;
698 if (nmp->nm_rsize > maxio)
699 nmp->nm_rsize = maxio;
700 if (nmp->nm_rsize > MAXBSIZE)
701 nmp->nm_rsize = MAXBSIZE;
703 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
704 nmp->nm_readdirsize = argp->readdirsize;
706 if (nmp->nm_readdirsize > maxio)
707 nmp->nm_readdirsize = maxio;
708 if (nmp->nm_readdirsize > nmp->nm_rsize)
709 nmp->nm_readdirsize = nmp->nm_rsize;
711 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
712 nmp->nm_acregmin = argp->acregmin;
714 nmp->nm_acregmin = NFS_MINATTRTIMO;
715 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
716 nmp->nm_acregmax = argp->acregmax;
718 nmp->nm_acregmax = NFS_MAXATTRTIMO;
719 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
720 nmp->nm_acdirmin = argp->acdirmin;
722 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
723 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
724 nmp->nm_acdirmax = argp->acdirmax;
726 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
727 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
728 nmp->nm_acdirmin = nmp->nm_acdirmax;
729 if (nmp->nm_acregmin > nmp->nm_acregmax)
730 nmp->nm_acregmin = nmp->nm_acregmax;
732 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
733 if (argp->maxgrouplist <= NFS_MAXGRPS)
734 nmp->nm_numgrps = argp->maxgrouplist;
736 nmp->nm_numgrps = NFS_MAXGRPS;
738 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
739 if (argp->readahead <= NFS_MAXRAHEAD)
740 nmp->nm_readahead = argp->readahead;
742 nmp->nm_readahead = NFS_MAXRAHEAD;
744 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2) {
745 if (argp->leaseterm <= NQ_MAXLEASE)
746 nmp->nm_leaseterm = argp->leaseterm;
748 nmp->nm_leaseterm = NQ_MAXLEASE;
750 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
751 if (argp->deadthresh <= NQ_NEVERDEAD)
752 nmp->nm_deadthresh = argp->deadthresh;
754 nmp->nm_deadthresh = NQ_NEVERDEAD;
757 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
758 (nmp->nm_soproto != argp->proto));
759 nmp->nm_sotype = argp->sotype;
760 nmp->nm_soproto = argp->proto;
762 if (nmp->nm_so && adjsock) {
763 nfs_safedisconnect(nmp);
764 if (nmp->nm_sotype == SOCK_DGRAM)
765 while (nfs_connect(nmp, (struct nfsreq *)0)) {
766 printf("nfs_args: retrying connect\n");
767 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
776 * It seems a bit dumb to copyinstr() the host and path here and then
777 * bcopy() them in mountnfs(), but I wanted to detect errors before
778 * doing the sockargs() call because sockargs() allocates an mbuf and
779 * an error after that means that I have to release the mbuf.
783 nfs_mount(struct mount *mp, char *path, caddr_t data,
784 struct nameidata *ndp, struct thread *td)
787 struct nfs_args args;
788 struct sockaddr *nam;
790 char pth[MNAMELEN], hst[MNAMELEN];
792 u_char nfh[NFSX_V3FHMAX];
798 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
801 if (args.version != NFS_ARGSVERSION) {
802 #ifdef COMPAT_PRELITE2
804 * If the argument version is unknown, then assume the
805 * caller is a pre-lite2 4.4BSD client and convert its
808 struct onfs_args oargs;
809 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
812 nfs_convert_oargs(&args,&oargs);
813 #else /* !COMPAT_PRELITE2 */
814 return (EPROGMISMATCH);
815 #endif /* COMPAT_PRELITE2 */
817 if (mp->mnt_flag & MNT_UPDATE) {
818 struct nfsmount *nmp = VFSTONFS(mp);
823 * When doing an update, we can't change from or to
824 * v3 and/or nqnfs, or change cookie translation
826 args.flags = (args.flags &
827 ~(NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)) |
829 (NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/));
830 nfs_decode_args(nmp, &args);
835 * Make the nfs_ip_paranoia sysctl serve as the default connection
836 * or no-connection mode for those protocols that support
837 * no-connection mode (the flag will be cleared later for protocols
838 * that do not support no-connection mode). This will allow a client
839 * to receive replies from a different IP then the request was
840 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
843 if (nfs_ip_paranoia == 0)
844 args.flags |= NFSMNT_NOCONN;
845 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
847 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
850 error = copyinstr(path, pth, MNAMELEN-1, &len);
853 bzero(&pth[len], MNAMELEN - len);
854 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
857 bzero(&hst[len], MNAMELEN - len);
858 /* sockargs() call must be after above copyin() calls */
859 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
863 error = mountnfs(&args, mp, nam, pth, hst, &vp);
868 * Common code for mount and mountroot
871 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
872 char *pth, char *hst, struct vnode **vpp)
874 struct nfsmount *nmp;
878 if (mp->mnt_flag & MNT_UPDATE) {
880 /* update paths, file handles, etc, here XXX */
884 nmp = zalloc(nfsmount_zone);
885 bzero((caddr_t)nmp, sizeof (struct nfsmount));
886 TAILQ_INIT(&nmp->nm_uidlruhead);
887 TAILQ_INIT(&nmp->nm_bufq);
888 mp->mnt_data = (qaddr_t)nmp;
892 if (argp->flags & NFSMNT_NQNFS)
894 * We have to set mnt_maxsymlink to a non-zero value so
895 * that COMPAT_43 routines will know that we are setting
896 * the d_type field in directories (and can zero it for
897 * unsuspecting binaries).
899 mp->mnt_maxsymlinklen = 1;
902 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
903 * high, depending on whether we end up with negative offsets in
904 * the client or server somewhere. 2GB-1 may be safer.
906 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
907 * that we can handle until we find out otherwise.
908 * XXX Our "safe" limit on the client is what we can store in our
909 * buffer cache using signed(!) block numbers.
911 if ((argp->flags & NFSMNT_NFSV3) == 0)
912 nmp->nm_maxfilesize = 0xffffffffLL;
914 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
916 nmp->nm_timeo = NFS_TIMEO;
917 nmp->nm_retry = NFS_RETRANS;
918 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
919 nmp->nm_rsize = nmp->nm_wsize;
920 nmp->nm_readdirsize = NFS_READDIRSIZE;
921 nmp->nm_numgrps = NFS_MAXGRPS;
922 nmp->nm_readahead = NFS_DEFRAHEAD;
923 nmp->nm_leaseterm = NQ_DEFLEASE;
924 nmp->nm_deadthresh = NQ_DEADTHRESH;
925 CIRCLEQ_INIT(&nmp->nm_timerhead);
926 nmp->nm_inprog = NULLVP;
927 nmp->nm_fhsize = argp->fhsize;
928 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
929 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
930 bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN);
932 /* Set up the sockets and per-host congestion */
933 nmp->nm_sotype = argp->sotype;
934 nmp->nm_soproto = argp->proto;
935 nmp->nm_cred = crhold(proc0.p_ucred);
937 nfs_decode_args(nmp, argp);
940 * For Connection based sockets (TCP,...) defer the connect until
941 * the first request, in case the server is not responding.
943 if (nmp->nm_sotype == SOCK_DGRAM &&
944 (error = nfs_connect(nmp, (struct nfsreq *)0)))
948 * This is silly, but it has to be set so that vinifod() works.
949 * We do not want to do an nfs_statfs() here since we can get
950 * stuck on a dead server and we are holding a lock on the mount
953 mp->mnt_stat.f_iosize =
954 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
957 * A reference count is needed on the nfsnode representing the
958 * remote root. If this object is not persistent, then backward
959 * traversals of the mount point (i.e. "..") will not work if
960 * the nfsnode gets flushed out of the cache. Ufs does not have
961 * this problem, because one can identify root inodes by their
962 * number == ROOTINO (2).
964 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
970 * Retrieval of mountpoint attributes is delayed until nfs_rot
971 * or nfs_statfs are first called. This will happen either when
972 * we first traverse the mount point or if somebody does a df(1).
974 * NFSSTA_GOTFSINFO is used to flag if we have successfully
975 * retrieved mountpoint attributes. In the case of NFSv3 we
976 * also flag static fsinfo.
979 (*vpp)->v_type = VNON;
982 * Lose the lock but keep the ref.
984 VOP_UNLOCK(*vpp, NULL, 0, curthread);
995 * unmount system call
998 nfs_unmount(struct mount *mp, int mntflags, struct thread *td)
1000 struct nfsmount *nmp;
1001 int error, flags = 0;
1003 if (mntflags & MNT_FORCE)
1004 flags |= FORCECLOSE;
1007 * Goes something like this..
1008 * - Call vflush() to clear out vnodes for this file system
1009 * - Close the socket
1010 * - Free up the data structures
1012 /* In the forced case, cancel any outstanding requests. */
1013 if (flags & FORCECLOSE) {
1014 error = nfs_nmcancelreqs(nmp);
1019 * Must handshake with nqnfs_clientd() if it is active.
1021 nmp->nm_state |= NFSSTA_DISMINPROG;
1022 while (nmp->nm_inprog != NULLVP)
1023 (void) tsleep((caddr_t)&lbolt, 0, "nfsdism", 0);
1025 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1026 error = vflush(mp, 1, flags);
1028 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1033 * We are now committed to the unmount.
1034 * For NQNFS, let the server daemon free the nfsmount structure.
1036 if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB))
1037 nmp->nm_state |= NFSSTA_DISMNT;
1039 nfs_disconnect(nmp);
1040 FREE(nmp->nm_nam, M_SONAME);
1042 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
1043 nfs_free_mount(nmp);
1048 nfs_free_mount(struct nfsmount *nmp)
1051 crfree(nmp->nm_cred);
1052 nmp->nm_cred = NULL;
1054 zfree(nfsmount_zone, nmp);
1058 * Return root of a filesystem
1066 struct nfsmount *nmp;
1072 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1078 * Get transfer parameters and root vnode attributes
1080 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1081 if (nmp->nm_flag & NFSMNT_NFSV3) {
1082 nfs_fsinfo(nmp, vp, curthread);
1083 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1085 if ((error = VOP_GETATTR(vp, &attrs, curthread)) == 0)
1086 nmp->nm_state |= NFSSTA_GOTFSINFO;
1090 if (vp->v_type == VNON)
1106 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1107 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data);
1110 * Flush out the buffer cache
1114 nfs_sync(struct mount *mp, int waitfor, struct thread *td)
1116 struct scaninfo scaninfo;
1119 scaninfo.rescan = 0;
1121 scaninfo.waitfor = waitfor;
1122 scaninfo.allerror = 0;
1125 * Force stale buffer cache information to be flushed.
1128 while (error == 0 && scaninfo.rescan) {
1129 scaninfo.rescan = 0;
1130 error = vmntvnodescan(mp, nfs_sync_scan1,
1131 nfs_sync_scan2, &scaninfo);
1138 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1140 struct scaninfo *info = data;
1142 if (VOP_ISLOCKED(vp, NULL) || TAILQ_EMPTY(&vp->v_dirtyblkhd))
1144 if (info->waitfor == MNT_LAZY)
1151 nfs_sync_scan2(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
1153 struct scaninfo *info = data;
1156 if (vget(vp, vlock, LK_EXCLUSIVE | LK_INTERLOCK, info->td)) {
1160 error = VOP_FSYNC(vp, info->waitfor, info->td);
1162 info->allerror = error;