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 1.28 2005/07/26 15:43:35 hmp 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>
64 #include <sys/thread2.h>
72 #include "nfsm_subs.h"
73 #include "nfsdiskless.h"
76 extern int nfs_mountroot(struct mount *mp);
77 extern void bootpc_init(void);
80 extern struct vnodeopv_entry_desc nfsv2_vnodeop_entries[];
81 extern struct vnodeopv_entry_desc nfsv2_fifoop_entries[];
82 extern struct vnodeopv_entry_desc nfsv2_specop_entries[];
84 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
85 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
86 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
87 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
88 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
89 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
90 MALLOC_DEFINE(M_NQLEASE, "NQNFS Lease", "Nqnfs lease");
91 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
93 vm_zone_t nfsmount_zone;
95 struct nfsstats nfsstats;
96 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
97 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
98 &nfsstats, nfsstats, "");
99 static int nfs_ip_paranoia = 1;
100 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
101 &nfs_ip_paranoia, 0, "");
104 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
108 * Tunable to determine the Read/Write unit size. Maximum value
109 * is NFS_MAXDATA. We also default to NFS_MAXDATA.
111 static int nfs_io_size = NFS_MAXDATA;
112 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
113 &nfs_io_size, 0, "NFS optimal I/O unit size");
115 static void nfs_decode_args (struct nfsmount *nmp,
116 struct nfs_args *argp);
117 static int mountnfs (struct nfs_args *,struct mount *,
118 struct sockaddr *,char *,char *,struct vnode **);
119 static int nfs_mount ( struct mount *mp, char *path, caddr_t data,
121 static int nfs_unmount ( struct mount *mp, int mntflags,
123 static int nfs_root ( struct mount *mp, struct vnode **vpp);
124 static int nfs_statfs ( struct mount *mp, struct statfs *sbp,
126 static int nfs_sync ( struct mount *mp, int waitfor,
130 * nfs vfs operations.
132 static struct vfsops nfs_vfsops = {
133 .vfs_mount = nfs_mount,
134 .vfs_unmount = nfs_unmount,
135 .vfs_root = nfs_root,
136 .vfs_statfs = nfs_statfs,
137 .vfs_sync = nfs_sync,
138 .vfs_init = nfs_init,
139 .vfs_uninit = nfs_uninit
141 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
144 * This structure must be filled in by a primary bootstrap or bootstrap
145 * server for a diskless/dataless machine. It is initialized below just
146 * to ensure that it is allocated to initialized data (.data not .bss).
148 struct nfs_diskless nfs_diskless = { { { 0 } } };
149 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
150 int nfs_diskless_valid = 0;
152 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
153 &nfs_diskless_valid, 0, "");
155 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
156 nfsv3_diskless.root_hostnam, 0, "");
158 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
159 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
160 "%Ssockaddr_in", "");
162 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
163 nfsv3_diskless.swap_hostnam, 0, "");
165 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
166 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
170 void nfsargs_ntoh (struct nfs_args *);
171 static int nfs_mountdiskless (char *, char *, int,
172 struct sockaddr_in *, struct nfs_args *,
173 struct thread *, struct vnode **,
175 static void nfs_convert_diskless (void);
176 static void nfs_convert_oargs (struct nfs_args *args,
177 struct onfs_args *oargs);
180 * Calculate the buffer I/O block size to use. The maximum V2 block size
181 * is typically 8K, the maximum datagram size is typically 16K, and the
182 * maximum V3 block size is typically 32K. The buffer cache tends to work
183 * best with 16K blocks but we allow 32K for TCP connections.
185 * We force the block size to be at least a page for buffer cache efficiency.
189 nfs_iosize(int v3, int sotype)
195 if (sotype == SOCK_STREAM)
198 iomax = NFS_MAXDGRAMDATA;
200 iomax = NFS_V2MAXDATA;
202 if ((iosize = nfs_io_size) > iomax)
204 if (iosize < PAGE_SIZE)
208 * This is an aweful hack but until the buffer cache is rewritten
209 * we need it. The problem is that when you combine write() with
210 * mmap() the vm_page->valid bits can become weird looking
211 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers
212 * at the file EOF. To solve the problem the BIO system needs to
213 * be guarenteed that the NFS iosize for regular files will be a
214 * multiple of PAGE_SIZE so it can invalidate the whole page
215 * rather then just the piece of it owned by the buffer when
216 * NFS does vinvalbuf() calls.
218 if (iosize & PAGE_MASK)
219 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
224 nfs_convert_oargs(args, oargs)
225 struct nfs_args *args;
226 struct onfs_args *oargs;
228 args->version = NFS_ARGSVERSION;
229 args->addr = oargs->addr;
230 args->addrlen = oargs->addrlen;
231 args->sotype = oargs->sotype;
232 args->proto = oargs->proto;
233 args->fh = oargs->fh;
234 args->fhsize = oargs->fhsize;
235 args->flags = oargs->flags;
236 args->wsize = oargs->wsize;
237 args->rsize = oargs->rsize;
238 args->readdirsize = oargs->readdirsize;
239 args->timeo = oargs->timeo;
240 args->retrans = oargs->retrans;
241 args->maxgrouplist = oargs->maxgrouplist;
242 args->readahead = oargs->readahead;
243 args->leaseterm = oargs->leaseterm;
244 args->deadthresh = oargs->deadthresh;
245 args->hostname = oargs->hostname;
249 nfs_convert_diskless()
251 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
252 sizeof(struct ifaliasreq));
253 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
254 sizeof(struct sockaddr_in));
255 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
256 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
257 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
258 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
259 sizeof(struct sockaddr_in));
260 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
261 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
262 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
263 sizeof(struct ucred));
264 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
265 nfsv3_diskless.root_fhsize = NFSX_V2FH;
266 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
267 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
268 sizeof(struct sockaddr_in));
269 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
270 nfsv3_diskless.root_time = nfs_diskless.root_time;
271 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
273 nfs_diskless_valid = 3;
280 nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
283 struct nfs_statfs *sfp;
287 caddr_t bpos, dpos, cp2;
288 struct nfsmount *nmp = VFSTONFS(mp);
289 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
290 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
296 sfp = (struct nfs_statfs *)0;
298 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
303 cred->cr_ngroups = 1;
304 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
305 (void)nfs_fsinfo(nmp, vp, td);
306 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
307 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
309 nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
311 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
317 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
318 sbp->f_flags = nmp->nm_flag;
319 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
322 sbp->f_bsize = NFS_FABLKSIZE;
323 tquad = fxdr_hyper(&sfp->sf_tbytes);
324 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
325 tquad = fxdr_hyper(&sfp->sf_fbytes);
326 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
327 tquad = fxdr_hyper(&sfp->sf_abytes);
328 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
329 sbp->f_files = (fxdr_unsigned(int32_t,
330 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
331 sbp->f_ffree = (fxdr_unsigned(int32_t,
332 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
334 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
335 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
336 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
337 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
341 if (sbp != &mp->mnt_stat) {
342 sbp->f_type = mp->mnt_vfc->vfc_typenum;
343 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
353 * nfs version 3 fsinfo rpc call
356 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
358 struct nfsv3_fsinfo *fsp;
361 u_int32_t *tl, pref, max;
362 caddr_t bpos, dpos, cp2;
363 int error = 0, retattr;
364 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
367 nfsstats.rpccnt[NFSPROC_FSINFO]++;
368 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
370 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
371 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
373 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
374 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
375 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
376 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
377 ~(NFS_FABLKSIZE - 1);
378 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
379 if (max < nmp->nm_wsize && max > 0) {
380 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
381 if (nmp->nm_wsize == 0)
384 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
385 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
386 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
387 ~(NFS_FABLKSIZE - 1);
388 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
389 if (max < nmp->nm_rsize && max > 0) {
390 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
391 if (nmp->nm_rsize == 0)
394 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
395 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
396 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
397 ~(NFS_DIRBLKSIZ - 1);
398 if (max < nmp->nm_readdirsize && max > 0) {
399 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
400 if (nmp->nm_readdirsize == 0)
401 nmp->nm_readdirsize = max;
403 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
404 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
405 nmp->nm_maxfilesize = maxfsize;
406 nmp->nm_state |= NFSSTA_GOTFSINFO;
414 * Mount a remote root fs via. nfs. This depends on the info in the
415 * nfs_diskless structure that has been filled in properly by some primary
417 * It goes something like this:
418 * - do enough of "ifconfig" by calling ifioctl() so that the system
419 * can talk to the server
420 * - If nfs_diskless.mygateway is filled in, use that address as
422 * - build the rootfs mount point and call mountnfs() to do the rest.
428 struct mount *swap_mp;
429 struct nfsv3_diskless *nd = &nfsv3_diskless;
432 struct thread *td = curthread; /* XXX */
437 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
438 bootpc_init(); /* use bootp to get nfs_diskless filled in */
442 * XXX time must be non-zero when we init the interface or else
443 * the arp code will wedge...
445 while (mycpu->gd_time_seconds == 0)
446 tsleep(mycpu, 0, "arpkludge", 10);
448 if (nfs_diskless_valid==1)
449 nfs_convert_diskless();
452 * XXX splnet, so networks will receive...
457 /* Set up swap credentials. */
458 proc0.p_ucred->cr_uid = ntohl(nd->swap_ucred.cr_uid);
459 proc0.p_ucred->cr_gid = ntohl(nd->swap_ucred.cr_gid);
460 if ((proc0.p_ucred->cr_ngroups = ntohs(nd->swap_ucred.cr_ngroups)) >
462 proc0.p_ucred->cr_ngroups = NGROUPS;
463 for (i = 0; i < proc0.p_ucred->cr_ngroups; i++)
464 proc0.p_ucred->cr_groups[i] = ntohl(nd->swap_ucred.cr_groups[i]);
468 * Do enough of ifconfig(8) so that the critical net interface can
469 * talk to the server.
471 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
473 panic("nfs_mountroot: socreate(%04x): %d",
474 nd->myif.ifra_addr.sa_family, error);
476 #if 0 /* XXX Bad idea */
478 * We might not have been told the right interface, so we pass
479 * over the first ten interfaces of the same kind, until we get
480 * one of them configured.
483 for (i = strlen(nd->myif.ifra_name) - 1;
484 nd->myif.ifra_name[i] >= '0' &&
485 nd->myif.ifra_name[i] <= '9';
486 nd->myif.ifra_name[i] ++) {
487 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
492 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
494 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
498 * If the gateway field is filled in, set it as the default route.
500 if (nd->mygateway.sin_len != 0) {
501 struct sockaddr_in mask, sin;
503 bzero((caddr_t)&mask, sizeof(mask));
505 sin.sin_family = AF_INET;
506 sin.sin_len = sizeof(sin);
507 error = rtrequest(RTM_ADD, (struct sockaddr *)&sin,
508 (struct sockaddr *)&nd->mygateway,
509 (struct sockaddr *)&mask,
510 RTF_UP | RTF_GATEWAY, (struct rtentry **)0);
512 panic("nfs_mountroot: RTM_ADD: %d", error);
516 * Create the rootfs mount point.
518 nd->root_args.fh = nd->root_fh;
519 nd->root_args.fhsize = nd->root_fhsize;
520 l = ntohl(nd->root_saddr.sin_addr.s_addr);
521 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
522 (l >> 24) & 0xff, (l >> 16) & 0xff,
523 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
524 printf("NFS ROOT: %s\n",buf);
525 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
526 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
528 mp->mnt_vfc->vfc_refcount--;
529 free(swap_mp, M_MOUNT);
535 if (nd->swap_nblks) {
537 /* Convert to DEV_BSIZE instead of Kilobyte */
541 * Create a fake mount point just for the swap vnode so that the
542 * swap file can be on a different server from the rootfs.
544 nd->swap_args.fh = nd->swap_fh;
545 nd->swap_args.fhsize = nd->swap_fhsize;
546 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
547 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
548 (l >> 24) & 0xff, (l >> 16) & 0xff,
549 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
550 printf("NFS SWAP: %s\n",buf);
551 if ((error = nfs_mountdiskless(buf, "/swap", 0,
552 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0)
554 vfs_unbusy(swap_mp, td);
556 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
557 nd->swap_nblks * DEV_BSIZE ;
560 * Since the swap file is not the root dir of a file system,
561 * hack it to a regular file.
566 swaponvp(td, vp, nd->swap_nblks);
569 mp->mnt_flag |= MNT_ROOTFS;
570 mp->mnt_vnodecovered = NULLVP;
574 * This is not really an nfs issue, but it is much easier to
575 * set hostname here and then let the "/etc/rc.xxx" files
576 * mount the right /var based upon its preset value.
578 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
579 hostname[MAXHOSTNAMELEN - 1] = '\0';
580 for (i = 0; i < MAXHOSTNAMELEN; i++)
581 if (hostname[i] == '\0')
583 inittodr(ntohl(nd->root_time));
588 * Internal version of mount system call for diskless setup.
591 nfs_mountdiskless(char *path, char *which, int mountflag,
592 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
593 struct vnode **vpp, struct mount **mpp)
596 struct sockaddr *nam;
603 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
604 printf("nfs_mountroot: NFS not configured");
610 mp->mnt_kern_flag = 0;
611 mp->mnt_flag = mountflag;
612 nam = dup_sockaddr((struct sockaddr *)sin);
613 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
614 printf("nfs_mountroot: mount %s on %s: %d", path, which, error);
615 mp->mnt_vfc->vfc_refcount--;
627 nfs_decode_args(nmp, argp)
628 struct nfsmount *nmp;
629 struct nfs_args *argp;
636 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
637 * no sense in that context.
639 if (argp->sotype == SOCK_STREAM)
640 nmp->nm_flag &= ~NFSMNT_NOCONN;
642 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
643 if ((argp->flags & NFSMNT_NFSV3) == 0)
644 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
646 /* Re-bind if rsrvd port requested and wasn't on one */
647 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
648 && (argp->flags & NFSMNT_RESVPORT);
649 /* Also re-bind if we're switching to/from a connected UDP socket */
650 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
651 (argp->flags & NFSMNT_NOCONN));
653 /* Update flags atomically. Don't change the lock bits. */
654 nmp->nm_flag = argp->flags | nmp->nm_flag;
657 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
658 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
659 if (nmp->nm_timeo < NFS_MINTIMEO)
660 nmp->nm_timeo = NFS_MINTIMEO;
661 else if (nmp->nm_timeo > NFS_MAXTIMEO)
662 nmp->nm_timeo = NFS_MAXTIMEO;
665 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
666 nmp->nm_retry = argp->retrans;
667 if (nmp->nm_retry > NFS_MAXREXMIT)
668 nmp->nm_retry = NFS_MAXREXMIT;
671 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
673 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
674 nmp->nm_wsize = argp->wsize;
675 /* Round down to multiple of blocksize */
676 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
677 if (nmp->nm_wsize <= 0)
678 nmp->nm_wsize = NFS_FABLKSIZE;
680 if (nmp->nm_wsize > maxio)
681 nmp->nm_wsize = maxio;
682 if (nmp->nm_wsize > MAXBSIZE)
683 nmp->nm_wsize = MAXBSIZE;
685 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
686 nmp->nm_rsize = argp->rsize;
687 /* Round down to multiple of blocksize */
688 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
689 if (nmp->nm_rsize <= 0)
690 nmp->nm_rsize = NFS_FABLKSIZE;
692 if (nmp->nm_rsize > maxio)
693 nmp->nm_rsize = maxio;
694 if (nmp->nm_rsize > MAXBSIZE)
695 nmp->nm_rsize = MAXBSIZE;
697 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
698 nmp->nm_readdirsize = argp->readdirsize;
700 if (nmp->nm_readdirsize > maxio)
701 nmp->nm_readdirsize = maxio;
702 if (nmp->nm_readdirsize > nmp->nm_rsize)
703 nmp->nm_readdirsize = nmp->nm_rsize;
705 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
706 nmp->nm_acregmin = argp->acregmin;
708 nmp->nm_acregmin = NFS_MINATTRTIMO;
709 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
710 nmp->nm_acregmax = argp->acregmax;
712 nmp->nm_acregmax = NFS_MAXATTRTIMO;
713 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
714 nmp->nm_acdirmin = argp->acdirmin;
716 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
717 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
718 nmp->nm_acdirmax = argp->acdirmax;
720 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
721 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
722 nmp->nm_acdirmin = nmp->nm_acdirmax;
723 if (nmp->nm_acregmin > nmp->nm_acregmax)
724 nmp->nm_acregmin = nmp->nm_acregmax;
726 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
727 if (argp->maxgrouplist <= NFS_MAXGRPS)
728 nmp->nm_numgrps = argp->maxgrouplist;
730 nmp->nm_numgrps = NFS_MAXGRPS;
732 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
733 if (argp->readahead <= NFS_MAXRAHEAD)
734 nmp->nm_readahead = argp->readahead;
736 nmp->nm_readahead = NFS_MAXRAHEAD;
738 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2) {
739 if (argp->leaseterm <= NQ_MAXLEASE)
740 nmp->nm_leaseterm = argp->leaseterm;
742 nmp->nm_leaseterm = NQ_MAXLEASE;
744 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
745 if (argp->deadthresh <= NQ_NEVERDEAD)
746 nmp->nm_deadthresh = argp->deadthresh;
748 nmp->nm_deadthresh = NQ_NEVERDEAD;
751 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
752 (nmp->nm_soproto != argp->proto));
753 nmp->nm_sotype = argp->sotype;
754 nmp->nm_soproto = argp->proto;
756 if (nmp->nm_so && adjsock) {
757 nfs_safedisconnect(nmp);
758 if (nmp->nm_sotype == SOCK_DGRAM)
759 while (nfs_connect(nmp, (struct nfsreq *)0)) {
760 printf("nfs_args: retrying connect\n");
761 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
770 * It seems a bit dumb to copyinstr() the host and path here and then
771 * bcopy() them in mountnfs(), but I wanted to detect errors before
772 * doing the sockargs() call because sockargs() allocates an mbuf and
773 * an error after that means that I have to release the mbuf.
777 nfs_mount(struct mount *mp, char *path, caddr_t data, struct thread *td)
780 struct nfs_args args;
781 struct sockaddr *nam;
783 char pth[MNAMELEN], hst[MNAMELEN];
785 u_char nfh[NFSX_V3FHMAX];
791 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
794 if (args.version != NFS_ARGSVERSION) {
795 #ifdef COMPAT_PRELITE2
797 * If the argument version is unknown, then assume the
798 * caller is a pre-lite2 4.4BSD client and convert its
801 struct onfs_args oargs;
802 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
805 nfs_convert_oargs(&args,&oargs);
806 #else /* !COMPAT_PRELITE2 */
807 return (EPROGMISMATCH);
808 #endif /* COMPAT_PRELITE2 */
810 if (mp->mnt_flag & MNT_UPDATE) {
811 struct nfsmount *nmp = VFSTONFS(mp);
816 * When doing an update, we can't change from or to
817 * v3 and/or nqnfs, or change cookie translation
819 args.flags = (args.flags &
820 ~(NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)) |
822 (NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/));
823 nfs_decode_args(nmp, &args);
828 * Make the nfs_ip_paranoia sysctl serve as the default connection
829 * or no-connection mode for those protocols that support
830 * no-connection mode (the flag will be cleared later for protocols
831 * that do not support no-connection mode). This will allow a client
832 * to receive replies from a different IP then the request was
833 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
836 if (nfs_ip_paranoia == 0)
837 args.flags |= NFSMNT_NOCONN;
838 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
840 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
843 error = copyinstr(path, pth, MNAMELEN-1, &len);
846 bzero(&pth[len], MNAMELEN - len);
847 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
850 bzero(&hst[len], MNAMELEN - len);
851 /* sockargs() call must be after above copyin() calls */
852 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
856 error = mountnfs(&args, mp, nam, pth, hst, &vp);
861 * Common code for mount and mountroot
864 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
865 char *pth, char *hst, struct vnode **vpp)
867 struct nfsmount *nmp;
871 if (mp->mnt_flag & MNT_UPDATE) {
873 /* update paths, file handles, etc, here XXX */
877 nmp = zalloc(nfsmount_zone);
878 bzero((caddr_t)nmp, sizeof (struct nfsmount));
879 TAILQ_INIT(&nmp->nm_uidlruhead);
880 TAILQ_INIT(&nmp->nm_bufq);
881 mp->mnt_data = (qaddr_t)nmp;
885 if (argp->flags & NFSMNT_NQNFS)
887 * We have to set mnt_maxsymlink to a non-zero value so
888 * that COMPAT_43 routines will know that we are setting
889 * the d_type field in directories (and can zero it for
890 * unsuspecting binaries).
892 mp->mnt_maxsymlinklen = 1;
895 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
896 * high, depending on whether we end up with negative offsets in
897 * the client or server somewhere. 2GB-1 may be safer.
899 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
900 * that we can handle until we find out otherwise.
901 * XXX Our "safe" limit on the client is what we can store in our
902 * buffer cache using signed(!) block numbers.
904 if ((argp->flags & NFSMNT_NFSV3) == 0)
905 nmp->nm_maxfilesize = 0xffffffffLL;
907 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
909 nmp->nm_timeo = NFS_TIMEO;
910 nmp->nm_retry = NFS_RETRANS;
911 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
912 nmp->nm_rsize = nmp->nm_wsize;
913 nmp->nm_readdirsize = NFS_READDIRSIZE;
914 nmp->nm_numgrps = NFS_MAXGRPS;
915 nmp->nm_readahead = NFS_DEFRAHEAD;
916 nmp->nm_leaseterm = NQ_DEFLEASE;
917 nmp->nm_deadthresh = NQ_DEADTHRESH;
918 CIRCLEQ_INIT(&nmp->nm_timerhead);
919 nmp->nm_inprog = NULLVP;
920 nmp->nm_fhsize = argp->fhsize;
921 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
922 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
924 /* Set up the sockets and per-host congestion */
925 nmp->nm_sotype = argp->sotype;
926 nmp->nm_soproto = argp->proto;
927 nmp->nm_cred = crhold(proc0.p_ucred);
929 nfs_decode_args(nmp, argp);
932 * For Connection based sockets (TCP,...) defer the connect until
933 * the first request, in case the server is not responding.
935 if (nmp->nm_sotype == SOCK_DGRAM &&
936 (error = nfs_connect(nmp, (struct nfsreq *)0)))
940 * This is silly, but it has to be set so that vinifod() works.
941 * We do not want to do an nfs_statfs() here since we can get
942 * stuck on a dead server and we are holding a lock on the mount
945 mp->mnt_stat.f_iosize =
946 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
949 * Install vop_ops for our vnops
951 vfs_add_vnodeops(mp, &mp->mnt_vn_norm_ops, nfsv2_vnodeop_entries);
952 vfs_add_vnodeops(mp, &mp->mnt_vn_spec_ops, nfsv2_specop_entries);
953 vfs_add_vnodeops(mp, &mp->mnt_vn_fifo_ops, nfsv2_fifoop_entries);
956 * A reference count is needed on the nfsnode representing the
957 * remote root. If this object is not persistent, then backward
958 * traversals of the mount point (i.e. "..") will not work if
959 * the nfsnode gets flushed out of the cache. Ufs does not have
960 * this problem, because one can identify root inodes by their
961 * number == ROOTINO (2).
963 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
969 * Retrieval of mountpoint attributes is delayed until nfs_rot
970 * or nfs_statfs are first called. This will happen either when
971 * we first traverse the mount point or if somebody does a df(1).
973 * NFSSTA_GOTFSINFO is used to flag if we have successfully
974 * retrieved mountpoint attributes. In the case of NFSv3 we
975 * also flag static fsinfo.
978 (*vpp)->v_type = VNON;
981 * Lose the lock but keep the ref.
983 VOP_UNLOCK(*vpp, 0, curthread);
994 * unmount system call
997 nfs_unmount(struct mount *mp, int mntflags, struct thread *td)
999 struct nfsmount *nmp;
1000 int error, flags = 0;
1002 if (mntflags & MNT_FORCE)
1003 flags |= FORCECLOSE;
1006 * Goes something like this..
1007 * - Call vflush() to clear out vnodes for this file system
1008 * - Close the socket
1009 * - Free up the data structures
1011 /* In the forced case, cancel any outstanding requests. */
1012 if (flags & FORCECLOSE) {
1013 error = nfs_nmcancelreqs(nmp);
1018 * Must handshake with nqnfs_clientd() if it is active.
1020 nmp->nm_state |= NFSSTA_DISMINPROG;
1021 while (nmp->nm_inprog != NULLVP)
1022 (void) tsleep((caddr_t)&lbolt, 0, "nfsdism", 0);
1024 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1025 error = vflush(mp, 1, flags);
1027 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1032 * We are now committed to the unmount.
1033 * For NQNFS, let the server daemon free the nfsmount structure.
1035 if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB))
1036 nmp->nm_state |= NFSSTA_DISMNT;
1038 nfs_disconnect(nmp);
1039 FREE(nmp->nm_nam, M_SONAME);
1041 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
1042 nfs_free_mount(nmp);
1047 nfs_free_mount(struct nfsmount *nmp)
1050 crfree(nmp->nm_cred);
1051 nmp->nm_cred = NULL;
1053 zfree(nfsmount_zone, nmp);
1057 * Return root of a filesystem
1065 struct nfsmount *nmp;
1071 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1077 * Get transfer parameters and root vnode attributes
1079 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1080 if (nmp->nm_flag & NFSMNT_NFSV3) {
1081 nfs_fsinfo(nmp, vp, curthread);
1082 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1084 if ((error = VOP_GETATTR(vp, &attrs, curthread)) == 0)
1085 nmp->nm_state |= NFSSTA_GOTFSINFO;
1089 if (vp->v_type == VNON)
1105 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1106 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1109 * Flush out the buffer cache
1113 nfs_sync(struct mount *mp, int waitfor, struct thread *td)
1115 struct scaninfo scaninfo;
1118 scaninfo.rescan = 0;
1120 scaninfo.waitfor = waitfor;
1121 scaninfo.allerror = 0;
1124 * Force stale buffer cache information to be flushed.
1127 while (error == 0 && scaninfo.rescan) {
1128 scaninfo.rescan = 0;
1129 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1130 nfs_sync_scan2, &scaninfo);
1137 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1139 struct scaninfo *info = data;
1141 if (VOP_ISLOCKED(vp, NULL) || RB_EMPTY(&vp->v_rbdirty_tree))
1143 if (info->waitfor == MNT_LAZY)
1150 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1152 struct scaninfo *info = data;
1155 error = VOP_FSYNC(vp, info->waitfor, info->td);
1157 info->allerror = error;