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 1.23 2004/12/17 00:18:28 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);
78 extern struct vnodeopv_entry_desc nfsv2_vnodeop_entries[];
79 extern struct vnodeopv_entry_desc nfsv2_fifoop_entries[];
80 extern struct vnodeopv_entry_desc nfsv2_specop_entries[];
82 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
83 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
84 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
85 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
86 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
87 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
88 MALLOC_DEFINE(M_NQLEASE, "NQNFS Lease", "Nqnfs lease");
89 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
91 vm_zone_t nfsmount_zone;
93 struct nfsstats nfsstats;
94 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
95 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
96 &nfsstats, nfsstats, "");
97 static int nfs_ip_paranoia = 1;
98 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
99 &nfs_ip_paranoia, 0, "");
102 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
106 * Tunable to determine the Read/Write unit size. Maximum value
107 * is NFS_MAXDATA. We also default to NFS_MAXDATA.
109 static int nfs_io_size = NFS_MAXDATA;
110 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
111 &nfs_io_size, 0, "NFS optimal I/O unit size");
113 static void nfs_decode_args (struct nfsmount *nmp,
114 struct nfs_args *argp);
115 static int mountnfs (struct nfs_args *,struct mount *,
116 struct sockaddr *,char *,char *,struct vnode **);
117 static int nfs_mount ( struct mount *mp, char *path, caddr_t data,
119 static int nfs_unmount ( struct mount *mp, int mntflags,
121 static int nfs_root ( struct mount *mp, struct vnode **vpp);
122 static int nfs_statfs ( struct mount *mp, struct statfs *sbp,
124 static int nfs_sync ( struct mount *mp, int waitfor,
128 * nfs vfs operations.
130 static struct vfsops nfs_vfsops = {
139 vfs_stdfhtovp, /* shouldn't happen */
141 vfs_stdvptofh, /* shouldn't happen */
146 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
149 * This structure must be filled in by a primary bootstrap or bootstrap
150 * server for a diskless/dataless machine. It is initialized below just
151 * to ensure that it is allocated to initialized data (.data not .bss).
153 struct nfs_diskless nfs_diskless = { { { 0 } } };
154 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
155 int nfs_diskless_valid = 0;
157 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
158 &nfs_diskless_valid, 0, "");
160 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
161 nfsv3_diskless.root_hostnam, 0, "");
163 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
164 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
165 "%Ssockaddr_in", "");
167 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
168 nfsv3_diskless.swap_hostnam, 0, "");
170 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
171 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
175 void nfsargs_ntoh (struct nfs_args *);
176 static int nfs_mountdiskless (char *, char *, int,
177 struct sockaddr_in *, struct nfs_args *,
178 struct thread *, struct vnode **,
180 static void nfs_convert_diskless (void);
181 static void nfs_convert_oargs (struct nfs_args *args,
182 struct onfs_args *oargs);
185 * Calculate the buffer I/O block size to use. The maximum V2 block size
186 * is typically 8K, the maximum datagram size is typically 16K, and the
187 * maximum V3 block size is typically 32K. The buffer cache tends to work
188 * best with 16K blocks but we allow 32K for TCP connections.
190 * We force the block size to be at least a page for buffer cache efficiency.
194 nfs_iosize(int v3, int sotype)
200 if (sotype == SOCK_STREAM)
203 iomax = NFS_MAXDGRAMDATA;
205 iomax = NFS_V2MAXDATA;
207 if ((iosize = nfs_io_size) > iomax)
209 if (iosize < PAGE_SIZE)
213 * This is an aweful hack but until the buffer cache is rewritten
214 * we need it. The problem is that when you combine write() with
215 * mmap() the vm_page->valid bits can become weird looking
216 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers
217 * at the file EOF. To solve the problem the BIO system needs to
218 * be guarenteed that the NFS iosize for regular files will be a
219 * multiple of PAGE_SIZE so it can invalidate the whole page
220 * rather then just the piece of it owned by the buffer when
221 * NFS does vinvalbuf() calls.
223 if (iosize & PAGE_MASK)
224 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
229 nfs_convert_oargs(args, oargs)
230 struct nfs_args *args;
231 struct onfs_args *oargs;
233 args->version = NFS_ARGSVERSION;
234 args->addr = oargs->addr;
235 args->addrlen = oargs->addrlen;
236 args->sotype = oargs->sotype;
237 args->proto = oargs->proto;
238 args->fh = oargs->fh;
239 args->fhsize = oargs->fhsize;
240 args->flags = oargs->flags;
241 args->wsize = oargs->wsize;
242 args->rsize = oargs->rsize;
243 args->readdirsize = oargs->readdirsize;
244 args->timeo = oargs->timeo;
245 args->retrans = oargs->retrans;
246 args->maxgrouplist = oargs->maxgrouplist;
247 args->readahead = oargs->readahead;
248 args->leaseterm = oargs->leaseterm;
249 args->deadthresh = oargs->deadthresh;
250 args->hostname = oargs->hostname;
254 nfs_convert_diskless()
256 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
257 sizeof(struct ifaliasreq));
258 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
259 sizeof(struct sockaddr_in));
260 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
261 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
262 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
263 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
264 sizeof(struct sockaddr_in));
265 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
266 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
267 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
268 sizeof(struct ucred));
269 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
270 nfsv3_diskless.root_fhsize = NFSX_V2FH;
271 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
272 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
273 sizeof(struct sockaddr_in));
274 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
275 nfsv3_diskless.root_time = nfs_diskless.root_time;
276 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
278 nfs_diskless_valid = 3;
285 nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
288 struct nfs_statfs *sfp;
292 caddr_t bpos, dpos, cp2;
293 struct nfsmount *nmp = VFSTONFS(mp);
294 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
295 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
301 sfp = (struct nfs_statfs *)0;
303 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
308 cred->cr_ngroups = 1;
309 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
310 (void)nfs_fsinfo(nmp, vp, td);
311 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
312 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
314 nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
316 nfsm_postop_attr(vp, retattr);
322 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
323 sbp->f_flags = nmp->nm_flag;
324 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
327 sbp->f_bsize = NFS_FABLKSIZE;
328 tquad = fxdr_hyper(&sfp->sf_tbytes);
329 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
330 tquad = fxdr_hyper(&sfp->sf_fbytes);
331 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
332 tquad = fxdr_hyper(&sfp->sf_abytes);
333 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
334 sbp->f_files = (fxdr_unsigned(int32_t,
335 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
336 sbp->f_ffree = (fxdr_unsigned(int32_t,
337 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
339 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
340 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
341 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
342 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
346 if (sbp != &mp->mnt_stat) {
347 sbp->f_type = mp->mnt_vfc->vfc_typenum;
348 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
349 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
359 * nfs version 3 fsinfo rpc call
362 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
364 struct nfsv3_fsinfo *fsp;
367 u_int32_t *tl, pref, max;
368 caddr_t bpos, dpos, cp2;
369 int error = 0, retattr;
370 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
373 nfsstats.rpccnt[NFSPROC_FSINFO]++;
374 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
376 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
377 nfsm_postop_attr(vp, retattr);
379 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
380 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
381 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
382 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
383 ~(NFS_FABLKSIZE - 1);
384 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
385 if (max < nmp->nm_wsize && max > 0) {
386 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
387 if (nmp->nm_wsize == 0)
390 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
391 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
392 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
393 ~(NFS_FABLKSIZE - 1);
394 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
395 if (max < nmp->nm_rsize && max > 0) {
396 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
397 if (nmp->nm_rsize == 0)
400 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
401 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
402 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
403 ~(NFS_DIRBLKSIZ - 1);
404 if (max < nmp->nm_readdirsize && max > 0) {
405 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
406 if (nmp->nm_readdirsize == 0)
407 nmp->nm_readdirsize = max;
409 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
410 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
411 nmp->nm_maxfilesize = maxfsize;
412 nmp->nm_state |= NFSSTA_GOTFSINFO;
420 * Mount a remote root fs via. nfs. This depends on the info in the
421 * nfs_diskless structure that has been filled in properly by some primary
423 * It goes something like this:
424 * - do enough of "ifconfig" by calling ifioctl() so that the system
425 * can talk to the server
426 * - If nfs_diskless.mygateway is filled in, use that address as
428 * - build the rootfs mount point and call mountnfs() to do the rest.
434 struct mount *swap_mp;
435 struct nfsv3_diskless *nd = &nfsv3_diskless;
438 struct thread *td = curthread; /* XXX */
443 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
444 bootpc_init(); /* use bootp to get nfs_diskless filled in */
448 * XXX time must be non-zero when we init the interface or else
449 * the arp code will wedge...
451 while (mycpu->gd_time_seconds == 0)
452 tsleep(mycpu, 0, "arpkludge", 10);
454 if (nfs_diskless_valid==1)
455 nfs_convert_diskless();
458 * XXX splnet, so networks will receive...
463 /* Set up swap credentials. */
464 proc0.p_ucred->cr_uid = ntohl(nd->swap_ucred.cr_uid);
465 proc0.p_ucred->cr_gid = ntohl(nd->swap_ucred.cr_gid);
466 if ((proc0.p_ucred->cr_ngroups = ntohs(nd->swap_ucred.cr_ngroups)) >
468 proc0.p_ucred->cr_ngroups = NGROUPS;
469 for (i = 0; i < proc0.p_ucred->cr_ngroups; i++)
470 proc0.p_ucred->cr_groups[i] = ntohl(nd->swap_ucred.cr_groups[i]);
474 * Do enough of ifconfig(8) so that the critical net interface can
475 * talk to the server.
477 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
479 panic("nfs_mountroot: socreate(%04x): %d",
480 nd->myif.ifra_addr.sa_family, error);
482 #if 0 /* XXX Bad idea */
484 * We might not have been told the right interface, so we pass
485 * over the first ten interfaces of the same kind, until we get
486 * one of them configured.
489 for (i = strlen(nd->myif.ifra_name) - 1;
490 nd->myif.ifra_name[i] >= '0' &&
491 nd->myif.ifra_name[i] <= '9';
492 nd->myif.ifra_name[i] ++) {
493 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
498 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
500 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
504 * If the gateway field is filled in, set it as the default route.
506 if (nd->mygateway.sin_len != 0) {
507 struct sockaddr_in mask, sin;
509 bzero((caddr_t)&mask, sizeof(mask));
511 sin.sin_family = AF_INET;
512 sin.sin_len = sizeof(sin);
513 error = rtrequest(RTM_ADD, (struct sockaddr *)&sin,
514 (struct sockaddr *)&nd->mygateway,
515 (struct sockaddr *)&mask,
516 RTF_UP | RTF_GATEWAY, (struct rtentry **)0);
518 panic("nfs_mountroot: RTM_ADD: %d", error);
522 * Create the rootfs mount point.
524 nd->root_args.fh = nd->root_fh;
525 nd->root_args.fhsize = nd->root_fhsize;
526 l = ntohl(nd->root_saddr.sin_addr.s_addr);
527 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
528 (l >> 24) & 0xff, (l >> 16) & 0xff,
529 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
530 printf("NFS ROOT: %s\n",buf);
531 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
532 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
534 mp->mnt_vfc->vfc_refcount--;
535 free(swap_mp, M_MOUNT);
541 if (nd->swap_nblks) {
543 /* Convert to DEV_BSIZE instead of Kilobyte */
547 * Create a fake mount point just for the swap vnode so that the
548 * swap file can be on a different server from the rootfs.
550 nd->swap_args.fh = nd->swap_fh;
551 nd->swap_args.fhsize = nd->swap_fhsize;
552 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
553 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
554 (l >> 24) & 0xff, (l >> 16) & 0xff,
555 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
556 printf("NFS SWAP: %s\n",buf);
557 if ((error = nfs_mountdiskless(buf, "/swap", 0,
558 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0)
560 vfs_unbusy(swap_mp, td);
562 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
563 nd->swap_nblks * DEV_BSIZE ;
566 * Since the swap file is not the root dir of a file system,
567 * hack it to a regular file.
572 swaponvp(td, vp, nd->swap_nblks);
575 mp->mnt_flag |= MNT_ROOTFS;
576 mp->mnt_vnodecovered = NULLVP;
580 * This is not really an nfs issue, but it is much easier to
581 * set hostname here and then let the "/etc/rc.xxx" files
582 * mount the right /var based upon its preset value.
584 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
585 hostname[MAXHOSTNAMELEN - 1] = '\0';
586 for (i = 0; i < MAXHOSTNAMELEN; i++)
587 if (hostname[i] == '\0')
589 inittodr(ntohl(nd->root_time));
594 * Internal version of mount system call for diskless setup.
597 nfs_mountdiskless(char *path, char *which, int mountflag,
598 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
599 struct vnode **vpp, struct mount **mpp)
602 struct sockaddr *nam;
609 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
610 printf("nfs_mountroot: NFS not configured");
616 mp->mnt_kern_flag = 0;
617 mp->mnt_flag = mountflag;
618 nam = dup_sockaddr((struct sockaddr *)sin);
619 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
620 printf("nfs_mountroot: mount %s on %s: %d", path, which, error);
621 mp->mnt_vfc->vfc_refcount--;
628 (void) copystr(which, mp->mnt_stat.f_mntonname, MNAMELEN - 1, 0);
634 nfs_decode_args(nmp, argp)
635 struct nfsmount *nmp;
636 struct nfs_args *argp;
644 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
645 * no sense in that context.
647 if (argp->sotype == SOCK_STREAM)
648 nmp->nm_flag &= ~NFSMNT_NOCONN;
650 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
651 if ((argp->flags & NFSMNT_NFSV3) == 0)
652 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
654 /* Re-bind if rsrvd port requested and wasn't on one */
655 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
656 && (argp->flags & NFSMNT_RESVPORT);
657 /* Also re-bind if we're switching to/from a connected UDP socket */
658 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
659 (argp->flags & NFSMNT_NOCONN));
661 /* Update flags atomically. Don't change the lock bits. */
662 nmp->nm_flag = argp->flags | nmp->nm_flag;
665 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
666 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
667 if (nmp->nm_timeo < NFS_MINTIMEO)
668 nmp->nm_timeo = NFS_MINTIMEO;
669 else if (nmp->nm_timeo > NFS_MAXTIMEO)
670 nmp->nm_timeo = NFS_MAXTIMEO;
673 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
674 nmp->nm_retry = argp->retrans;
675 if (nmp->nm_retry > NFS_MAXREXMIT)
676 nmp->nm_retry = NFS_MAXREXMIT;
679 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
681 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
682 nmp->nm_wsize = argp->wsize;
683 /* Round down to multiple of blocksize */
684 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
685 if (nmp->nm_wsize <= 0)
686 nmp->nm_wsize = NFS_FABLKSIZE;
688 if (nmp->nm_wsize > maxio)
689 nmp->nm_wsize = maxio;
690 if (nmp->nm_wsize > MAXBSIZE)
691 nmp->nm_wsize = MAXBSIZE;
693 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
694 nmp->nm_rsize = argp->rsize;
695 /* Round down to multiple of blocksize */
696 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
697 if (nmp->nm_rsize <= 0)
698 nmp->nm_rsize = NFS_FABLKSIZE;
700 if (nmp->nm_rsize > maxio)
701 nmp->nm_rsize = maxio;
702 if (nmp->nm_rsize > MAXBSIZE)
703 nmp->nm_rsize = MAXBSIZE;
705 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
706 nmp->nm_readdirsize = argp->readdirsize;
708 if (nmp->nm_readdirsize > maxio)
709 nmp->nm_readdirsize = maxio;
710 if (nmp->nm_readdirsize > nmp->nm_rsize)
711 nmp->nm_readdirsize = nmp->nm_rsize;
713 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
714 nmp->nm_acregmin = argp->acregmin;
716 nmp->nm_acregmin = NFS_MINATTRTIMO;
717 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
718 nmp->nm_acregmax = argp->acregmax;
720 nmp->nm_acregmax = NFS_MAXATTRTIMO;
721 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
722 nmp->nm_acdirmin = argp->acdirmin;
724 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
725 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
726 nmp->nm_acdirmax = argp->acdirmax;
728 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
729 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
730 nmp->nm_acdirmin = nmp->nm_acdirmax;
731 if (nmp->nm_acregmin > nmp->nm_acregmax)
732 nmp->nm_acregmin = nmp->nm_acregmax;
734 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
735 if (argp->maxgrouplist <= NFS_MAXGRPS)
736 nmp->nm_numgrps = argp->maxgrouplist;
738 nmp->nm_numgrps = NFS_MAXGRPS;
740 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
741 if (argp->readahead <= NFS_MAXRAHEAD)
742 nmp->nm_readahead = argp->readahead;
744 nmp->nm_readahead = NFS_MAXRAHEAD;
746 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2) {
747 if (argp->leaseterm <= NQ_MAXLEASE)
748 nmp->nm_leaseterm = argp->leaseterm;
750 nmp->nm_leaseterm = NQ_MAXLEASE;
752 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
753 if (argp->deadthresh <= NQ_NEVERDEAD)
754 nmp->nm_deadthresh = argp->deadthresh;
756 nmp->nm_deadthresh = NQ_NEVERDEAD;
759 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
760 (nmp->nm_soproto != argp->proto));
761 nmp->nm_sotype = argp->sotype;
762 nmp->nm_soproto = argp->proto;
764 if (nmp->nm_so && adjsock) {
765 nfs_safedisconnect(nmp);
766 if (nmp->nm_sotype == SOCK_DGRAM)
767 while (nfs_connect(nmp, (struct nfsreq *)0)) {
768 printf("nfs_args: retrying connect\n");
769 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
778 * It seems a bit dumb to copyinstr() the host and path here and then
779 * bcopy() them in mountnfs(), but I wanted to detect errors before
780 * doing the sockargs() call because sockargs() allocates an mbuf and
781 * an error after that means that I have to release the mbuf.
785 nfs_mount(struct mount *mp, char *path, caddr_t data, struct thread *td)
788 struct nfs_args args;
789 struct sockaddr *nam;
791 char pth[MNAMELEN], hst[MNAMELEN];
793 u_char nfh[NFSX_V3FHMAX];
799 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
802 if (args.version != NFS_ARGSVERSION) {
803 #ifdef COMPAT_PRELITE2
805 * If the argument version is unknown, then assume the
806 * caller is a pre-lite2 4.4BSD client and convert its
809 struct onfs_args oargs;
810 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
813 nfs_convert_oargs(&args,&oargs);
814 #else /* !COMPAT_PRELITE2 */
815 return (EPROGMISMATCH);
816 #endif /* COMPAT_PRELITE2 */
818 if (mp->mnt_flag & MNT_UPDATE) {
819 struct nfsmount *nmp = VFSTONFS(mp);
824 * When doing an update, we can't change from or to
825 * v3 and/or nqnfs, or change cookie translation
827 args.flags = (args.flags &
828 ~(NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)) |
830 (NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/));
831 nfs_decode_args(nmp, &args);
836 * Make the nfs_ip_paranoia sysctl serve as the default connection
837 * or no-connection mode for those protocols that support
838 * no-connection mode (the flag will be cleared later for protocols
839 * that do not support no-connection mode). This will allow a client
840 * to receive replies from a different IP then the request was
841 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
844 if (nfs_ip_paranoia == 0)
845 args.flags |= NFSMNT_NOCONN;
846 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
848 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
851 error = copyinstr(path, pth, MNAMELEN-1, &len);
854 bzero(&pth[len], MNAMELEN - len);
855 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
858 bzero(&hst[len], MNAMELEN - len);
859 /* sockargs() call must be after above copyin() calls */
860 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
864 error = mountnfs(&args, mp, nam, pth, hst, &vp);
869 * Common code for mount and mountroot
872 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
873 char *pth, char *hst, struct vnode **vpp)
875 struct nfsmount *nmp;
879 if (mp->mnt_flag & MNT_UPDATE) {
881 /* update paths, file handles, etc, here XXX */
885 nmp = zalloc(nfsmount_zone);
886 bzero((caddr_t)nmp, sizeof (struct nfsmount));
887 TAILQ_INIT(&nmp->nm_uidlruhead);
888 TAILQ_INIT(&nmp->nm_bufq);
889 mp->mnt_data = (qaddr_t)nmp;
893 if (argp->flags & NFSMNT_NQNFS)
895 * We have to set mnt_maxsymlink to a non-zero value so
896 * that COMPAT_43 routines will know that we are setting
897 * the d_type field in directories (and can zero it for
898 * unsuspecting binaries).
900 mp->mnt_maxsymlinklen = 1;
903 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
904 * high, depending on whether we end up with negative offsets in
905 * the client or server somewhere. 2GB-1 may be safer.
907 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
908 * that we can handle until we find out otherwise.
909 * XXX Our "safe" limit on the client is what we can store in our
910 * buffer cache using signed(!) block numbers.
912 if ((argp->flags & NFSMNT_NFSV3) == 0)
913 nmp->nm_maxfilesize = 0xffffffffLL;
915 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
917 nmp->nm_timeo = NFS_TIMEO;
918 nmp->nm_retry = NFS_RETRANS;
919 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
920 nmp->nm_rsize = nmp->nm_wsize;
921 nmp->nm_readdirsize = NFS_READDIRSIZE;
922 nmp->nm_numgrps = NFS_MAXGRPS;
923 nmp->nm_readahead = NFS_DEFRAHEAD;
924 nmp->nm_leaseterm = NQ_DEFLEASE;
925 nmp->nm_deadthresh = NQ_DEADTHRESH;
926 CIRCLEQ_INIT(&nmp->nm_timerhead);
927 nmp->nm_inprog = NULLVP;
928 nmp->nm_fhsize = argp->fhsize;
929 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
930 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
931 bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN);
933 /* Set up the sockets and per-host congestion */
934 nmp->nm_sotype = argp->sotype;
935 nmp->nm_soproto = argp->proto;
936 nmp->nm_cred = crhold(proc0.p_ucred);
938 nfs_decode_args(nmp, argp);
941 * For Connection based sockets (TCP,...) defer the connect until
942 * the first request, in case the server is not responding.
944 if (nmp->nm_sotype == SOCK_DGRAM &&
945 (error = nfs_connect(nmp, (struct nfsreq *)0)))
949 * This is silly, but it has to be set so that vinifod() works.
950 * We do not want to do an nfs_statfs() here since we can get
951 * stuck on a dead server and we are holding a lock on the mount
954 mp->mnt_stat.f_iosize =
955 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
958 * Install vop_ops for our vnops
960 vfs_add_vnodeops(mp, &mp->mnt_vn_norm_ops, nfsv2_vnodeop_entries);
961 vfs_add_vnodeops(mp, &mp->mnt_vn_spec_ops, nfsv2_specop_entries);
962 vfs_add_vnodeops(mp, &mp->mnt_vn_fifo_ops, nfsv2_fifoop_entries);
965 * A reference count is needed on the nfsnode representing the
966 * remote root. If this object is not persistent, then backward
967 * traversals of the mount point (i.e. "..") will not work if
968 * the nfsnode gets flushed out of the cache. Ufs does not have
969 * this problem, because one can identify root inodes by their
970 * number == ROOTINO (2).
972 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
978 * Retrieval of mountpoint attributes is delayed until nfs_rot
979 * or nfs_statfs are first called. This will happen either when
980 * we first traverse the mount point or if somebody does a df(1).
982 * NFSSTA_GOTFSINFO is used to flag if we have successfully
983 * retrieved mountpoint attributes. In the case of NFSv3 we
984 * also flag static fsinfo.
987 (*vpp)->v_type = VNON;
990 * Lose the lock but keep the ref.
992 VOP_UNLOCK(*vpp, 0, curthread);
1003 * unmount system call
1006 nfs_unmount(struct mount *mp, int mntflags, struct thread *td)
1008 struct nfsmount *nmp;
1009 int error, flags = 0;
1011 if (mntflags & MNT_FORCE)
1012 flags |= FORCECLOSE;
1015 * Goes something like this..
1016 * - Call vflush() to clear out vnodes for this file system
1017 * - Close the socket
1018 * - Free up the data structures
1020 /* In the forced case, cancel any outstanding requests. */
1021 if (flags & FORCECLOSE) {
1022 error = nfs_nmcancelreqs(nmp);
1027 * Must handshake with nqnfs_clientd() if it is active.
1029 nmp->nm_state |= NFSSTA_DISMINPROG;
1030 while (nmp->nm_inprog != NULLVP)
1031 (void) tsleep((caddr_t)&lbolt, 0, "nfsdism", 0);
1033 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1034 error = vflush(mp, 1, flags);
1036 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1041 * We are now committed to the unmount.
1042 * For NQNFS, let the server daemon free the nfsmount structure.
1044 if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB))
1045 nmp->nm_state |= NFSSTA_DISMNT;
1047 nfs_disconnect(nmp);
1048 FREE(nmp->nm_nam, M_SONAME);
1050 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
1051 nfs_free_mount(nmp);
1056 nfs_free_mount(struct nfsmount *nmp)
1059 crfree(nmp->nm_cred);
1060 nmp->nm_cred = NULL;
1062 zfree(nfsmount_zone, nmp);
1066 * Return root of a filesystem
1074 struct nfsmount *nmp;
1080 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1086 * Get transfer parameters and root vnode attributes
1088 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1089 if (nmp->nm_flag & NFSMNT_NFSV3) {
1090 nfs_fsinfo(nmp, vp, curthread);
1091 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1093 if ((error = VOP_GETATTR(vp, &attrs, curthread)) == 0)
1094 nmp->nm_state |= NFSSTA_GOTFSINFO;
1098 if (vp->v_type == VNON)
1114 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1115 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1118 * Flush out the buffer cache
1122 nfs_sync(struct mount *mp, int waitfor, struct thread *td)
1124 struct scaninfo scaninfo;
1127 scaninfo.rescan = 0;
1129 scaninfo.waitfor = waitfor;
1130 scaninfo.allerror = 0;
1133 * Force stale buffer cache information to be flushed.
1136 while (error == 0 && scaninfo.rescan) {
1137 scaninfo.rescan = 0;
1138 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1139 nfs_sync_scan2, &scaninfo);
1146 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1148 struct scaninfo *info = data;
1150 if (VOP_ISLOCKED(vp, NULL) || TAILQ_EMPTY(&vp->v_dirtyblkhd))
1152 if (info->waitfor == MNT_LAZY)
1159 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1161 struct scaninfo *info = data;
1164 error = VOP_FSYNC(vp, info->waitfor, info->td);
1166 info->allerror = error;