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 1.27 2005/06/06 15:09:38 drhodus 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 = {
141 vfs_stdfhtovp, /* shouldn't happen */
143 vfs_stdvptofh, /* shouldn't happen */
148 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
151 * This structure must be filled in by a primary bootstrap or bootstrap
152 * server for a diskless/dataless machine. It is initialized below just
153 * to ensure that it is allocated to initialized data (.data not .bss).
155 struct nfs_diskless nfs_diskless = { { { 0 } } };
156 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
157 int nfs_diskless_valid = 0;
159 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
160 &nfs_diskless_valid, 0, "");
162 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
163 nfsv3_diskless.root_hostnam, 0, "");
165 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
166 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
167 "%Ssockaddr_in", "");
169 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
170 nfsv3_diskless.swap_hostnam, 0, "");
172 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
173 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
177 void nfsargs_ntoh (struct nfs_args *);
178 static int nfs_mountdiskless (char *, char *, int,
179 struct sockaddr_in *, struct nfs_args *,
180 struct thread *, struct vnode **,
182 static void nfs_convert_diskless (void);
183 static void nfs_convert_oargs (struct nfs_args *args,
184 struct onfs_args *oargs);
187 * Calculate the buffer I/O block size to use. The maximum V2 block size
188 * is typically 8K, the maximum datagram size is typically 16K, and the
189 * maximum V3 block size is typically 32K. The buffer cache tends to work
190 * best with 16K blocks but we allow 32K for TCP connections.
192 * We force the block size to be at least a page for buffer cache efficiency.
196 nfs_iosize(int v3, int sotype)
202 if (sotype == SOCK_STREAM)
205 iomax = NFS_MAXDGRAMDATA;
207 iomax = NFS_V2MAXDATA;
209 if ((iosize = nfs_io_size) > iomax)
211 if (iosize < PAGE_SIZE)
215 * This is an aweful hack but until the buffer cache is rewritten
216 * we need it. The problem is that when you combine write() with
217 * mmap() the vm_page->valid bits can become weird looking
218 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers
219 * at the file EOF. To solve the problem the BIO system needs to
220 * be guarenteed that the NFS iosize for regular files will be a
221 * multiple of PAGE_SIZE so it can invalidate the whole page
222 * rather then just the piece of it owned by the buffer when
223 * NFS does vinvalbuf() calls.
225 if (iosize & PAGE_MASK)
226 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
231 nfs_convert_oargs(args, oargs)
232 struct nfs_args *args;
233 struct onfs_args *oargs;
235 args->version = NFS_ARGSVERSION;
236 args->addr = oargs->addr;
237 args->addrlen = oargs->addrlen;
238 args->sotype = oargs->sotype;
239 args->proto = oargs->proto;
240 args->fh = oargs->fh;
241 args->fhsize = oargs->fhsize;
242 args->flags = oargs->flags;
243 args->wsize = oargs->wsize;
244 args->rsize = oargs->rsize;
245 args->readdirsize = oargs->readdirsize;
246 args->timeo = oargs->timeo;
247 args->retrans = oargs->retrans;
248 args->maxgrouplist = oargs->maxgrouplist;
249 args->readahead = oargs->readahead;
250 args->leaseterm = oargs->leaseterm;
251 args->deadthresh = oargs->deadthresh;
252 args->hostname = oargs->hostname;
256 nfs_convert_diskless()
258 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
259 sizeof(struct ifaliasreq));
260 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
261 sizeof(struct sockaddr_in));
262 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
263 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
264 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
265 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
266 sizeof(struct sockaddr_in));
267 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
268 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
269 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
270 sizeof(struct ucred));
271 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
272 nfsv3_diskless.root_fhsize = NFSX_V2FH;
273 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
274 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
275 sizeof(struct sockaddr_in));
276 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
277 nfsv3_diskless.root_time = nfs_diskless.root_time;
278 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
280 nfs_diskless_valid = 3;
287 nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
290 struct nfs_statfs *sfp;
294 caddr_t bpos, dpos, cp2;
295 struct nfsmount *nmp = VFSTONFS(mp);
296 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
297 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
303 sfp = (struct nfs_statfs *)0;
305 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
310 cred->cr_ngroups = 1;
311 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
312 (void)nfs_fsinfo(nmp, vp, td);
313 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
314 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
316 nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
318 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
324 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
325 sbp->f_flags = nmp->nm_flag;
326 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
329 sbp->f_bsize = NFS_FABLKSIZE;
330 tquad = fxdr_hyper(&sfp->sf_tbytes);
331 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
332 tquad = fxdr_hyper(&sfp->sf_fbytes);
333 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
334 tquad = fxdr_hyper(&sfp->sf_abytes);
335 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
336 sbp->f_files = (fxdr_unsigned(int32_t,
337 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
338 sbp->f_ffree = (fxdr_unsigned(int32_t,
339 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
341 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
342 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
343 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
344 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
348 if (sbp != &mp->mnt_stat) {
349 sbp->f_type = mp->mnt_vfc->vfc_typenum;
350 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
360 * nfs version 3 fsinfo rpc call
363 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
365 struct nfsv3_fsinfo *fsp;
368 u_int32_t *tl, pref, max;
369 caddr_t bpos, dpos, cp2;
370 int error = 0, retattr;
371 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
374 nfsstats.rpccnt[NFSPROC_FSINFO]++;
375 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
377 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
378 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
380 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
381 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
382 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
383 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
384 ~(NFS_FABLKSIZE - 1);
385 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
386 if (max < nmp->nm_wsize && max > 0) {
387 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
388 if (nmp->nm_wsize == 0)
391 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
392 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
393 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
394 ~(NFS_FABLKSIZE - 1);
395 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
396 if (max < nmp->nm_rsize && max > 0) {
397 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
398 if (nmp->nm_rsize == 0)
401 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
402 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
403 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
404 ~(NFS_DIRBLKSIZ - 1);
405 if (max < nmp->nm_readdirsize && max > 0) {
406 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
407 if (nmp->nm_readdirsize == 0)
408 nmp->nm_readdirsize = max;
410 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
411 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
412 nmp->nm_maxfilesize = maxfsize;
413 nmp->nm_state |= NFSSTA_GOTFSINFO;
421 * Mount a remote root fs via. nfs. This depends on the info in the
422 * nfs_diskless structure that has been filled in properly by some primary
424 * It goes something like this:
425 * - do enough of "ifconfig" by calling ifioctl() so that the system
426 * can talk to the server
427 * - If nfs_diskless.mygateway is filled in, use that address as
429 * - build the rootfs mount point and call mountnfs() to do the rest.
435 struct mount *swap_mp;
436 struct nfsv3_diskless *nd = &nfsv3_diskless;
439 struct thread *td = curthread; /* XXX */
444 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
445 bootpc_init(); /* use bootp to get nfs_diskless filled in */
449 * XXX time must be non-zero when we init the interface or else
450 * the arp code will wedge...
452 while (mycpu->gd_time_seconds == 0)
453 tsleep(mycpu, 0, "arpkludge", 10);
455 if (nfs_diskless_valid==1)
456 nfs_convert_diskless();
459 * XXX splnet, so networks will receive...
464 /* Set up swap credentials. */
465 proc0.p_ucred->cr_uid = ntohl(nd->swap_ucred.cr_uid);
466 proc0.p_ucred->cr_gid = ntohl(nd->swap_ucred.cr_gid);
467 if ((proc0.p_ucred->cr_ngroups = ntohs(nd->swap_ucred.cr_ngroups)) >
469 proc0.p_ucred->cr_ngroups = NGROUPS;
470 for (i = 0; i < proc0.p_ucred->cr_ngroups; i++)
471 proc0.p_ucred->cr_groups[i] = ntohl(nd->swap_ucred.cr_groups[i]);
475 * Do enough of ifconfig(8) so that the critical net interface can
476 * talk to the server.
478 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
480 panic("nfs_mountroot: socreate(%04x): %d",
481 nd->myif.ifra_addr.sa_family, error);
483 #if 0 /* XXX Bad idea */
485 * We might not have been told the right interface, so we pass
486 * over the first ten interfaces of the same kind, until we get
487 * one of them configured.
490 for (i = strlen(nd->myif.ifra_name) - 1;
491 nd->myif.ifra_name[i] >= '0' &&
492 nd->myif.ifra_name[i] <= '9';
493 nd->myif.ifra_name[i] ++) {
494 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
499 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
501 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
505 * If the gateway field is filled in, set it as the default route.
507 if (nd->mygateway.sin_len != 0) {
508 struct sockaddr_in mask, sin;
510 bzero((caddr_t)&mask, sizeof(mask));
512 sin.sin_family = AF_INET;
513 sin.sin_len = sizeof(sin);
514 error = rtrequest(RTM_ADD, (struct sockaddr *)&sin,
515 (struct sockaddr *)&nd->mygateway,
516 (struct sockaddr *)&mask,
517 RTF_UP | RTF_GATEWAY, (struct rtentry **)0);
519 panic("nfs_mountroot: RTM_ADD: %d", error);
523 * Create the rootfs mount point.
525 nd->root_args.fh = nd->root_fh;
526 nd->root_args.fhsize = nd->root_fhsize;
527 l = ntohl(nd->root_saddr.sin_addr.s_addr);
528 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
529 (l >> 24) & 0xff, (l >> 16) & 0xff,
530 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
531 printf("NFS ROOT: %s\n",buf);
532 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
533 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
535 mp->mnt_vfc->vfc_refcount--;
536 free(swap_mp, M_MOUNT);
542 if (nd->swap_nblks) {
544 /* Convert to DEV_BSIZE instead of Kilobyte */
548 * Create a fake mount point just for the swap vnode so that the
549 * swap file can be on a different server from the rootfs.
551 nd->swap_args.fh = nd->swap_fh;
552 nd->swap_args.fhsize = nd->swap_fhsize;
553 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
554 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
555 (l >> 24) & 0xff, (l >> 16) & 0xff,
556 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
557 printf("NFS SWAP: %s\n",buf);
558 if ((error = nfs_mountdiskless(buf, "/swap", 0,
559 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0)
561 vfs_unbusy(swap_mp, td);
563 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
564 nd->swap_nblks * DEV_BSIZE ;
567 * Since the swap file is not the root dir of a file system,
568 * hack it to a regular file.
573 swaponvp(td, vp, nd->swap_nblks);
576 mp->mnt_flag |= MNT_ROOTFS;
577 mp->mnt_vnodecovered = NULLVP;
581 * This is not really an nfs issue, but it is much easier to
582 * set hostname here and then let the "/etc/rc.xxx" files
583 * mount the right /var based upon its preset value.
585 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
586 hostname[MAXHOSTNAMELEN - 1] = '\0';
587 for (i = 0; i < MAXHOSTNAMELEN; i++)
588 if (hostname[i] == '\0')
590 inittodr(ntohl(nd->root_time));
595 * Internal version of mount system call for diskless setup.
598 nfs_mountdiskless(char *path, char *which, int mountflag,
599 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
600 struct vnode **vpp, struct mount **mpp)
603 struct sockaddr *nam;
610 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
611 printf("nfs_mountroot: NFS not configured");
617 mp->mnt_kern_flag = 0;
618 mp->mnt_flag = mountflag;
619 nam = dup_sockaddr((struct sockaddr *)sin);
620 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
621 printf("nfs_mountroot: mount %s on %s: %d", path, which, error);
622 mp->mnt_vfc->vfc_refcount--;
634 nfs_decode_args(nmp, argp)
635 struct nfsmount *nmp;
636 struct nfs_args *argp;
643 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
644 * no sense in that context.
646 if (argp->sotype == SOCK_STREAM)
647 nmp->nm_flag &= ~NFSMNT_NOCONN;
649 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
650 if ((argp->flags & NFSMNT_NFSV3) == 0)
651 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
653 /* Re-bind if rsrvd port requested and wasn't on one */
654 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
655 && (argp->flags & NFSMNT_RESVPORT);
656 /* Also re-bind if we're switching to/from a connected UDP socket */
657 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
658 (argp->flags & NFSMNT_NOCONN));
660 /* Update flags atomically. Don't change the lock bits. */
661 nmp->nm_flag = argp->flags | nmp->nm_flag;
664 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
665 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
666 if (nmp->nm_timeo < NFS_MINTIMEO)
667 nmp->nm_timeo = NFS_MINTIMEO;
668 else if (nmp->nm_timeo > NFS_MAXTIMEO)
669 nmp->nm_timeo = NFS_MAXTIMEO;
672 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
673 nmp->nm_retry = argp->retrans;
674 if (nmp->nm_retry > NFS_MAXREXMIT)
675 nmp->nm_retry = NFS_MAXREXMIT;
678 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
680 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
681 nmp->nm_wsize = argp->wsize;
682 /* Round down to multiple of blocksize */
683 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
684 if (nmp->nm_wsize <= 0)
685 nmp->nm_wsize = NFS_FABLKSIZE;
687 if (nmp->nm_wsize > maxio)
688 nmp->nm_wsize = maxio;
689 if (nmp->nm_wsize > MAXBSIZE)
690 nmp->nm_wsize = MAXBSIZE;
692 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
693 nmp->nm_rsize = argp->rsize;
694 /* Round down to multiple of blocksize */
695 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
696 if (nmp->nm_rsize <= 0)
697 nmp->nm_rsize = NFS_FABLKSIZE;
699 if (nmp->nm_rsize > maxio)
700 nmp->nm_rsize = maxio;
701 if (nmp->nm_rsize > MAXBSIZE)
702 nmp->nm_rsize = MAXBSIZE;
704 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
705 nmp->nm_readdirsize = argp->readdirsize;
707 if (nmp->nm_readdirsize > maxio)
708 nmp->nm_readdirsize = maxio;
709 if (nmp->nm_readdirsize > nmp->nm_rsize)
710 nmp->nm_readdirsize = nmp->nm_rsize;
712 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
713 nmp->nm_acregmin = argp->acregmin;
715 nmp->nm_acregmin = NFS_MINATTRTIMO;
716 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
717 nmp->nm_acregmax = argp->acregmax;
719 nmp->nm_acregmax = NFS_MAXATTRTIMO;
720 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
721 nmp->nm_acdirmin = argp->acdirmin;
723 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
724 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
725 nmp->nm_acdirmax = argp->acdirmax;
727 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
728 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
729 nmp->nm_acdirmin = nmp->nm_acdirmax;
730 if (nmp->nm_acregmin > nmp->nm_acregmax)
731 nmp->nm_acregmin = nmp->nm_acregmax;
733 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
734 if (argp->maxgrouplist <= NFS_MAXGRPS)
735 nmp->nm_numgrps = argp->maxgrouplist;
737 nmp->nm_numgrps = NFS_MAXGRPS;
739 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
740 if (argp->readahead <= NFS_MAXRAHEAD)
741 nmp->nm_readahead = argp->readahead;
743 nmp->nm_readahead = NFS_MAXRAHEAD;
745 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2) {
746 if (argp->leaseterm <= NQ_MAXLEASE)
747 nmp->nm_leaseterm = argp->leaseterm;
749 nmp->nm_leaseterm = NQ_MAXLEASE;
751 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
752 if (argp->deadthresh <= NQ_NEVERDEAD)
753 nmp->nm_deadthresh = argp->deadthresh;
755 nmp->nm_deadthresh = NQ_NEVERDEAD;
758 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
759 (nmp->nm_soproto != argp->proto));
760 nmp->nm_sotype = argp->sotype;
761 nmp->nm_soproto = argp->proto;
763 if (nmp->nm_so && adjsock) {
764 nfs_safedisconnect(nmp);
765 if (nmp->nm_sotype == SOCK_DGRAM)
766 while (nfs_connect(nmp, (struct nfsreq *)0)) {
767 printf("nfs_args: retrying connect\n");
768 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
777 * It seems a bit dumb to copyinstr() the host and path here and then
778 * bcopy() them in mountnfs(), but I wanted to detect errors before
779 * doing the sockargs() call because sockargs() allocates an mbuf and
780 * an error after that means that I have to release the mbuf.
784 nfs_mount(struct mount *mp, char *path, caddr_t data, 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);
931 /* Set up the sockets and per-host congestion */
932 nmp->nm_sotype = argp->sotype;
933 nmp->nm_soproto = argp->proto;
934 nmp->nm_cred = crhold(proc0.p_ucred);
936 nfs_decode_args(nmp, argp);
939 * For Connection based sockets (TCP,...) defer the connect until
940 * the first request, in case the server is not responding.
942 if (nmp->nm_sotype == SOCK_DGRAM &&
943 (error = nfs_connect(nmp, (struct nfsreq *)0)))
947 * This is silly, but it has to be set so that vinifod() works.
948 * We do not want to do an nfs_statfs() here since we can get
949 * stuck on a dead server and we are holding a lock on the mount
952 mp->mnt_stat.f_iosize =
953 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
956 * Install vop_ops for our vnops
958 vfs_add_vnodeops(mp, &mp->mnt_vn_norm_ops, nfsv2_vnodeop_entries);
959 vfs_add_vnodeops(mp, &mp->mnt_vn_spec_ops, nfsv2_specop_entries);
960 vfs_add_vnodeops(mp, &mp->mnt_vn_fifo_ops, nfsv2_fifoop_entries);
963 * A reference count is needed on the nfsnode representing the
964 * remote root. If this object is not persistent, then backward
965 * traversals of the mount point (i.e. "..") will not work if
966 * the nfsnode gets flushed out of the cache. Ufs does not have
967 * this problem, because one can identify root inodes by their
968 * number == ROOTINO (2).
970 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
976 * Retrieval of mountpoint attributes is delayed until nfs_rot
977 * or nfs_statfs are first called. This will happen either when
978 * we first traverse the mount point or if somebody does a df(1).
980 * NFSSTA_GOTFSINFO is used to flag if we have successfully
981 * retrieved mountpoint attributes. In the case of NFSv3 we
982 * also flag static fsinfo.
985 (*vpp)->v_type = VNON;
988 * Lose the lock but keep the ref.
990 VOP_UNLOCK(*vpp, 0, curthread);
1001 * unmount system call
1004 nfs_unmount(struct mount *mp, int mntflags, struct thread *td)
1006 struct nfsmount *nmp;
1007 int error, flags = 0;
1009 if (mntflags & MNT_FORCE)
1010 flags |= FORCECLOSE;
1013 * Goes something like this..
1014 * - Call vflush() to clear out vnodes for this file system
1015 * - Close the socket
1016 * - Free up the data structures
1018 /* In the forced case, cancel any outstanding requests. */
1019 if (flags & FORCECLOSE) {
1020 error = nfs_nmcancelreqs(nmp);
1025 * Must handshake with nqnfs_clientd() if it is active.
1027 nmp->nm_state |= NFSSTA_DISMINPROG;
1028 while (nmp->nm_inprog != NULLVP)
1029 (void) tsleep((caddr_t)&lbolt, 0, "nfsdism", 0);
1031 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1032 error = vflush(mp, 1, flags);
1034 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1039 * We are now committed to the unmount.
1040 * For NQNFS, let the server daemon free the nfsmount structure.
1042 if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB))
1043 nmp->nm_state |= NFSSTA_DISMNT;
1045 nfs_disconnect(nmp);
1046 FREE(nmp->nm_nam, M_SONAME);
1048 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
1049 nfs_free_mount(nmp);
1054 nfs_free_mount(struct nfsmount *nmp)
1057 crfree(nmp->nm_cred);
1058 nmp->nm_cred = NULL;
1060 zfree(nfsmount_zone, nmp);
1064 * Return root of a filesystem
1072 struct nfsmount *nmp;
1078 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1084 * Get transfer parameters and root vnode attributes
1086 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1087 if (nmp->nm_flag & NFSMNT_NFSV3) {
1088 nfs_fsinfo(nmp, vp, curthread);
1089 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1091 if ((error = VOP_GETATTR(vp, &attrs, curthread)) == 0)
1092 nmp->nm_state |= NFSSTA_GOTFSINFO;
1096 if (vp->v_type == VNON)
1112 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1113 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1116 * Flush out the buffer cache
1120 nfs_sync(struct mount *mp, int waitfor, struct thread *td)
1122 struct scaninfo scaninfo;
1125 scaninfo.rescan = 0;
1127 scaninfo.waitfor = waitfor;
1128 scaninfo.allerror = 0;
1131 * Force stale buffer cache information to be flushed.
1134 while (error == 0 && scaninfo.rescan) {
1135 scaninfo.rescan = 0;
1136 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1137 nfs_sync_scan2, &scaninfo);
1144 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1146 struct scaninfo *info = data;
1148 if (VOP_ISLOCKED(vp, NULL) || RB_EMPTY(&vp->v_rbdirty_tree))
1150 if (info->waitfor == MNT_LAZY)
1157 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1159 struct scaninfo *info = data;
1162 error = VOP_FSYNC(vp, info->waitfor, info->td);
1164 info->allerror = error;