xref: /csrg-svn/sys/nfs/nfs_vfsops.c (revision 40553)
138414Smckusick /*
238414Smckusick  * Copyright (c) 1989 The Regents of the University of California.
338414Smckusick  * All rights reserved.
438414Smckusick  *
538414Smckusick  * This code is derived from software contributed to Berkeley by
638414Smckusick  * Rick Macklem at The University of Guelph.
738414Smckusick  *
838414Smckusick  * Redistribution and use in source and binary forms are permitted
938414Smckusick  * provided that the above copyright notice and this paragraph are
1038414Smckusick  * duplicated in all such forms and that any documentation,
1138414Smckusick  * advertising materials, and other materials related to such
1238414Smckusick  * distribution and use acknowledge that the software was developed
1338414Smckusick  * by the University of California, Berkeley.  The name of the
1438414Smckusick  * University may not be used to endorse or promote products derived
1538414Smckusick  * from this software without specific prior written permission.
1638414Smckusick  * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
1738414Smckusick  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
1838414Smckusick  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
1938414Smckusick  *
20*40553Smckusick  *	@(#)nfs_vfsops.c	7.18 (Berkeley) 03/20/90
2138414Smckusick  */
2238414Smckusick 
2338414Smckusick #include "param.h"
2438414Smckusick #include "signal.h"
2538414Smckusick #include "user.h"
2638414Smckusick #include "proc.h"
2738414Smckusick #include "vnode.h"
2838414Smckusick #include "mount.h"
2938414Smckusick #include "errno.h"
3040120Smckusick #include "buf.h"
3138414Smckusick #include "mbuf.h"
3238414Smckusick #undef	m_data
3338414Smckusick #include "socket.h"
3440120Smckusick #include "systm.h"
3538414Smckusick #include "nfsv2.h"
3638414Smckusick #include "nfsnode.h"
3738414Smckusick #include "nfsmount.h"
3838414Smckusick #include "nfs.h"
3938414Smckusick 
4038414Smckusick /*
4138414Smckusick  * nfs vfs operations.
4238414Smckusick  */
4338414Smckusick int nfs_mount();
4438874Smckusick int nfs_start();
4538414Smckusick int nfs_unmount();
4638414Smckusick int nfs_root();
4739443Smckusick int nfs_statfs();
4838414Smckusick int nfs_sync();
4938414Smckusick int nfs_fhtovp();
5038414Smckusick int nfs_vptofh();
5139443Smckusick int nfs_init();
5238414Smckusick 
5338414Smckusick struct vfsops nfs_vfsops = {
5438414Smckusick 	nfs_mount,
5538874Smckusick 	nfs_start,
5638414Smckusick 	nfs_unmount,
5738414Smckusick 	nfs_root,
5838414Smckusick 	nfs_statfs,
5938414Smckusick 	nfs_sync,
6038414Smckusick 	nfs_fhtovp,
6138414Smckusick 	nfs_vptofh,
6239443Smckusick 	nfs_init,
6338414Smckusick };
6438414Smckusick 
6539757Smckusick static u_char nfs_mntid;
6638414Smckusick 
6738414Smckusick /*
6838414Smckusick  * Called by vfs_mountroot when nfs is going to be mounted as root
6938414Smckusick  * Not Yet (By a LONG shot)
7038414Smckusick  */
7138414Smckusick nfs_mountroot()
7238414Smckusick {
7338414Smckusick 	return (ENODEV);
7438414Smckusick }
7538414Smckusick 
7638414Smckusick /*
7738414Smckusick  * VFS Operations.
7838414Smckusick  *
7938414Smckusick  * mount system call
8038414Smckusick  * It seems a bit dumb to copyinstr() the host and path here and then
8138414Smckusick  * bcopy() them in mountnfs(), but I wanted to detect errors before
8238414Smckusick  * doing the sockargs() call because sockargs() allocates an mbuf and
8338414Smckusick  * an error after that means that I have to release the mbuf.
8438414Smckusick  */
8539494Smckusick /* ARGSUSED */
8638414Smckusick nfs_mount(mp, path, data, ndp)
8738414Smckusick 	struct mount *mp;
8838414Smckusick 	char *path;
8938414Smckusick 	caddr_t data;
9038414Smckusick 	struct nameidata *ndp;
9138414Smckusick {
9238414Smckusick 	int error;
9338414Smckusick 	struct nfs_args args;
9438414Smckusick 	struct mbuf *saddr;
9538414Smckusick 	char pth[MNAMELEN], hst[MNAMELEN];
9638414Smckusick 	int len;
9738414Smckusick 	nfsv2fh_t nfh;
9838414Smckusick 
9939460Smckusick 	if (mp->m_flag & M_UPDATE)
10039460Smckusick 		return (0);
10138414Smckusick 	if (error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)))
10238414Smckusick 		return (error);
10338414Smckusick 	if (error=copyin((caddr_t)args.fh, (caddr_t)&nfh, sizeof (nfsv2fh_t)))
10438414Smckusick 		return (error);
10538414Smckusick 	if (error = copyinstr(path, pth, MNAMELEN-1, &len))
10638414Smckusick 		return (error);
10738414Smckusick 	bzero(&pth[len], MNAMELEN-len);
10838414Smckusick 	if (error = copyinstr(args.hostname, hst, MNAMELEN-1, &len))
10938414Smckusick 		return (error);
11038414Smckusick 	bzero(&hst[len], MNAMELEN-len);
11138414Smckusick 	/* sockargs() call must be after above copyin() calls */
11238414Smckusick 	if (error = sockargs(&saddr, (caddr_t)args.addr,
11340120Smckusick 		sizeof (struct sockaddr), MT_SONAME))
11438414Smckusick 		return (error);
11538414Smckusick 	args.fh = &nfh;
11638414Smckusick 	error = mountnfs(&args, mp, saddr, pth, hst);
11738414Smckusick 	return (error);
11838414Smckusick }
11938414Smckusick 
12038414Smckusick /*
12138414Smckusick  * Common code for mount and mountroot
12238414Smckusick  */
12338414Smckusick mountnfs(argp, mp, saddr, pth, hst)
12438414Smckusick 	register struct nfs_args *argp;
12538414Smckusick 	register struct mount *mp;
12638414Smckusick 	register struct mbuf *saddr;
12738414Smckusick 	char *pth, *hst;
12838414Smckusick {
12938414Smckusick 	register struct nfsmount *nmp;
13040010Smckusick 	struct nfsnode *np;
13140120Smckusick 	int error;
13239757Smckusick 	fsid_t tfsid;
13338414Smckusick 
13440120Smckusick 	MALLOC(nmp, struct nfsmount *, sizeof *nmp, M_NFSMNT, M_WAITOK);
13540120Smckusick 	bzero((caddr_t)nmp, sizeof *nmp);
13638414Smckusick 	mp->m_data = (qaddr_t)nmp;
13739757Smckusick 	/*
13839757Smckusick 	 * Generate a unique nfs mount id. The problem is that a dev number
13939757Smckusick 	 * is not unique across multiple systems. The techique is as follows:
14039757Smckusick 	 * 1) Set to nblkdev,0 which will never be used otherwise
14139757Smckusick 	 * 2) Generate a first guess as nblkdev,nfs_mntid where nfs_mntid is
14239757Smckusick 	 *	NOT 0
14339757Smckusick 	 * 3) Loop searching the mount list for another one with same id
14439757Smckusick 	 *	If a match, increment val[0] and try again
14539757Smckusick 	 * NB: I increment val[0] { a long } instead of nfs_mntid { a u_char }
14639757Smckusick 	 *	so that nfs is not limited to 255 mount points
14739757Smckusick 	 *     Incrementing the high order bits does no real harm, since it
14839757Smckusick 	 *     simply makes the major dev number tick up. The upper bound is
14939757Smckusick 	 *     set to major dev 127 to avoid any sign extention problems
15039757Smckusick 	 */
15140353Smckusick 	mp->m_stat.f_fsid.val[0] = makedev(nblkdev, 0);
15240353Smckusick 	mp->m_stat.f_fsid.val[1] = MOUNT_NFS;
15339757Smckusick 	if (++nfs_mntid == 0)
15439757Smckusick 		++nfs_mntid;
15539757Smckusick 	tfsid.val[0] = makedev(nblkdev, nfs_mntid);
15639757Smckusick 	tfsid.val[1] = MOUNT_NFS;
15739757Smckusick 	while (getvfs(&tfsid)) {
15839757Smckusick 		tfsid.val[0]++;
15939757Smckusick 		nfs_mntid++;
16039757Smckusick 	}
16139757Smckusick 	if (major(tfsid.val[0]) > 127) {
16239757Smckusick 		error = ENOENT;
163*40553Smckusick 		m_freem(saddr);
16439757Smckusick 		goto bad;
16539757Smckusick 	}
16640353Smckusick 	mp->m_stat.f_fsid.val[0] = tfsid.val[0];
16738414Smckusick 	nmp->nm_mountp = mp;
16838414Smckusick 	nmp->nm_flag = argp->flags;
16940120Smckusick 	nmp->nm_rto = NFS_TIMEO;
17040120Smckusick 	nmp->nm_rtt = -1;
17140120Smckusick 	nmp->nm_rttvar = nmp->nm_rto << 1;
17240120Smckusick 	nmp->nm_retry = NFS_RETRANS;
17340120Smckusick 	nmp->nm_wsize = NFS_WSIZE;
17440120Smckusick 	nmp->nm_rsize = NFS_RSIZE;
17540120Smckusick 	bcopy((caddr_t)argp->fh, (caddr_t)&nmp->nm_fh, sizeof(nfsv2fh_t));
17640353Smckusick 	bcopy(hst, mp->m_stat.f_mntfromname, MNAMELEN);
17740353Smckusick 	bcopy(pth, mp->m_stat.f_mntonname, MNAMELEN);
17840120Smckusick 
17940120Smckusick 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
18040120Smckusick 		nmp->nm_rto = argp->timeo;
18140120Smckusick 		/* NFS timeouts are specified in 1/10 sec. */
18240120Smckusick 		nmp->nm_rto = (nmp->nm_rto * 10) / NFS_HZ;
18340120Smckusick 		if (nmp->nm_rto < NFS_MINTIMEO)
18440120Smckusick 			nmp->nm_rto = NFS_MINTIMEO;
18540120Smckusick 		else if (nmp->nm_rto > NFS_MAXTIMEO)
18640120Smckusick 			nmp->nm_rto = NFS_MAXTIMEO;
18740120Smckusick 		nmp->nm_rttvar = nmp->nm_rto << 1;
18840120Smckusick 	}
18940120Smckusick 
19040120Smckusick 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans >= 0) {
19140120Smckusick 		nmp->nm_retry = argp->retrans;
19240120Smckusick 		if (nmp->nm_retry > NFS_MAXREXMIT)
19340120Smckusick 			nmp->nm_retry = NFS_MAXREXMIT;
19440120Smckusick 	}
19540120Smckusick 
19640120Smckusick 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
19738414Smckusick 		nmp->nm_wsize = argp->wsize;
19840120Smckusick 		/* Round down to multiple of blocksize */
19940120Smckusick 		nmp->nm_wsize &= ~0x1ff;
20040120Smckusick 		if (nmp->nm_wsize <= 0)
20140120Smckusick 			nmp->nm_wsize = 512;
20240120Smckusick 		else if (nmp->nm_wsize > NFS_MAXDATA)
20340120Smckusick 			nmp->nm_wsize = NFS_MAXDATA;
20440120Smckusick 	}
20540120Smckusick 
20640120Smckusick 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
20738414Smckusick 		nmp->nm_rsize = argp->rsize;
20840120Smckusick 		/* Round down to multiple of blocksize */
20940120Smckusick 		nmp->nm_rsize &= ~0x1ff;
21040120Smckusick 		if (nmp->nm_rsize <= 0)
21140120Smckusick 			nmp->nm_rsize = 512;
21240120Smckusick 		else if (nmp->nm_rsize > NFS_MAXDATA)
21340120Smckusick 			nmp->nm_rsize = NFS_MAXDATA;
21440120Smckusick 	}
21540120Smckusick 	/* Set up the sockets and per-host congestion */
216*40553Smckusick 	if (error = nfs_connect(nmp, saddr)) {
217*40553Smckusick 		m_freem(saddr);
21840120Smckusick 		goto bad;
219*40553Smckusick 	}
22040120Smckusick 
22140353Smckusick 	if (error = nfs_statfs(mp, &mp->m_stat))
22240353Smckusick 		goto bad;
22338414Smckusick 	/*
22440010Smckusick 	 * A reference count is needed on the nfsnode representing the
22540010Smckusick 	 * remote root.  If this object is not persistent, then backward
22640010Smckusick 	 * traversals of the mount point (i.e. "..") will not work if
22740010Smckusick 	 * the nfsnode gets flushed out of the cache. Ufs does not have
22840010Smckusick 	 * this problem, because one can identify root inodes by their
22940010Smckusick 	 * number == ROOTINO (2).
23040010Smckusick 	 */
23140010Smckusick 	if (error = nfs_nget(mp, &nmp->nm_fh, &np))
23240010Smckusick 		goto bad;
23340010Smckusick 	/*
23440010Smckusick 	 * Unlock it, but keep the reference count.
23540010Smckusick 	 */
23640010Smckusick 	nfs_unlock(NFSTOV(np));
23740353Smckusick 	return (0);
23840120Smckusick 
23938414Smckusick bad:
24040120Smckusick 	nfs_disconnect(nmp);
24140120Smckusick 	FREE(nmp, M_NFSMNT);
24238414Smckusick 	return (error);
24338414Smckusick }
24438414Smckusick 
24538414Smckusick /*
24638414Smckusick  * unmount system call
24738414Smckusick  */
24838414Smckusick nfs_unmount(mp, flags)
24938414Smckusick 	struct mount *mp;
25038414Smckusick 	int flags;
25138414Smckusick {
25238414Smckusick 	register struct nfsmount *nmp;
25338414Smckusick 	register struct nfsreq *rep;
25438414Smckusick 	struct nfsreq *rep2;
25540010Smckusick 	struct nfsnode *np;
25640120Smckusick 	struct vnode *vp;
25738414Smckusick 	int error;
25838414Smckusick 	int s;
25938414Smckusick 
26038414Smckusick 	if (flags & MNT_FORCE)
26138414Smckusick 		return (EINVAL);
26238414Smckusick 	nmp = vfs_to_nfs(mp);
26338414Smckusick 	/*
26438884Smacklem 	 * Clear out the buffer cache
26538884Smacklem 	 */
26639669Smckusick 	mntflushbuf(mp, 0);
26739669Smckusick 	if (mntinvalbuf(mp))
26838884Smacklem 		return (EBUSY);
26938884Smacklem 	/*
27038414Smckusick 	 * Goes something like this..
27140120Smckusick 	 * - Check for activity on the root vnode (other than ourselves).
27240120Smckusick 	 * - Call vflush() to clear out vnodes for this file system,
27340120Smckusick 	 *   except for the root vnode.
27440120Smckusick 	 * - Decrement reference on the vnode representing remote root.
27538414Smckusick 	 * - Close the socket
27638414Smckusick 	 * - Free up the data structures
27738414Smckusick 	 */
27840010Smckusick 	/*
27940010Smckusick 	 * We need to decrement the ref. count on the nfsnode representing
28040010Smckusick 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
28140010Smckusick 	 * has done vput on this vnode, otherwise we would get deadlock!
28240010Smckusick 	 */
28340010Smckusick 	if (error = nfs_nget(mp, &nmp->nm_fh, &np))
28440010Smckusick 		return(error);
28540120Smckusick 	vp = NFSTOV(np);
28640120Smckusick 	if (vp->v_usecount > 2) {
28740120Smckusick 		vput(vp);
28840120Smckusick 		return (EBUSY);
28940120Smckusick 	}
29040120Smckusick 	if (error = vflush(mp, vp, flags)) {
29140120Smckusick 		vput(vp);
29240120Smckusick 		return (error);
29340120Smckusick 	}
29440010Smckusick 	/*
29540010Smckusick 	 * Get rid of two reference counts, and unlock it on the second.
29640010Smckusick 	 */
29740120Smckusick 	vrele(vp);
29840120Smckusick 	vput(vp);
29940120Smckusick 	nfs_disconnect(nmp);
30038414Smckusick 	free((caddr_t)nmp, M_NFSMNT);
30138414Smckusick 	return (0);
30238414Smckusick }
30338414Smckusick 
30438414Smckusick /*
30538414Smckusick  * Return root of a filesystem
30638414Smckusick  */
30738414Smckusick nfs_root(mp, vpp)
30838414Smckusick 	struct mount *mp;
30938414Smckusick 	struct vnode **vpp;
31038414Smckusick {
31138414Smckusick 	register struct vnode *vp;
31238414Smckusick 	struct nfsmount *nmp;
31338414Smckusick 	struct nfsnode *np;
31438414Smckusick 	int error;
31538414Smckusick 
31638414Smckusick 	nmp = vfs_to_nfs(mp);
31738414Smckusick 	if (error = nfs_nget(mp, &nmp->nm_fh, &np))
31838414Smckusick 		return (error);
31938414Smckusick 	vp = NFSTOV(np);
32038414Smckusick 	vp->v_type = VDIR;
32138414Smckusick 	vp->v_flag = VROOT;
32238414Smckusick 	*vpp = vp;
32338414Smckusick 	return (0);
32438414Smckusick }
32538414Smckusick 
32638884Smacklem extern int syncprt;
32738884Smacklem 
32838414Smckusick /*
32938884Smacklem  * Flush out the buffer cache
33038414Smckusick  */
33139494Smckusick /* ARGSUSED */
33238414Smckusick nfs_sync(mp, waitfor)
33338414Smckusick 	struct mount *mp;
33438414Smckusick 	int waitfor;
33538414Smckusick {
33638884Smacklem 	if (syncprt)
33738884Smacklem 		bufstats();
33838884Smacklem 	/*
33938884Smacklem 	 * Force stale buffer cache information to be flushed.
34038884Smacklem 	 */
34140035Smckusick 	mntflushbuf(mp, waitfor == MNT_WAIT ? B_SYNC : 0);
34238414Smckusick 	return (0);
34338414Smckusick }
34438414Smckusick 
34538414Smckusick /*
34638414Smckusick  * At this point, this should never happen
34738414Smckusick  */
34839494Smckusick /* ARGSUSED */
34938414Smckusick nfs_fhtovp(mp, fhp, vpp)
35038414Smckusick 	struct mount *mp;
35138414Smckusick 	struct fid *fhp;
35238414Smckusick 	struct vnode **vpp;
35338414Smckusick {
35439494Smckusick 
35538414Smckusick 	return (EINVAL);
35638414Smckusick }
35738414Smckusick 
35838414Smckusick /*
35938414Smckusick  * Vnode pointer to File handle, should never happen either
36038414Smckusick  */
36139494Smckusick /* ARGSUSED */
36238414Smckusick nfs_vptofh(mp, fhp, vpp)
36338414Smckusick 	struct mount *mp;
36438414Smckusick 	struct fid *fhp;
36538414Smckusick 	struct vnode **vpp;
36638414Smckusick {
36739494Smckusick 
36838414Smckusick 	return (EINVAL);
36938414Smckusick }
37038884Smacklem 
37138884Smacklem /*
37238884Smacklem  * Vfs start routine, a no-op.
37338884Smacklem  */
37439494Smckusick /* ARGSUSED */
37538884Smacklem nfs_start(mp, flags)
37638884Smacklem 	struct mount *mp;
37738884Smacklem 	int flags;
37838884Smacklem {
37939494Smckusick 
38038884Smacklem 	return (0);
38138884Smacklem }
382