nfs_socket.c

早期freebsd实现

/*
 * Copyright (c) 1989, 1991, 1993
 *	The Regents of the University of California.  All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * Rick Macklem at The University of Guelph.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *	This product includes software developed by the University of
 *	California, Berkeley and its contributors.
 * 4. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *	@(#)nfs_socket.c	8.3 (Berkeley) 1/12/94
 */

/*
 * Socket operations for use by nfs
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/mount.h>
#include <sys/kernel.h>
#include <sys/mbuf.h>
#include <sys/vnode.h>
#include <sys/domain.h>
#include <sys/protosw.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/syslog.h>
#include <sys/tprintf.h>

#include <netinet/in.h>
#include <netinet/tcp.h>
#include <nfs/rpcv2.h>
#include <nfs/nfsv2.h>
#include <nfs/nfs.h>
#include <nfs/xdr_subs.h>
#include <nfs/nfsm_subs.h>
#include <nfs/nfsmount.h>
#include <nfs/nfsnode.h>
#include <nfs/nfsrtt.h>
#include <nfs/nqnfs.h>

#define	TRUE	1
#define	FALSE	0

/*
 * Estimate rto for an nfs rpc sent via. an unreliable datagram.
 * Use the mean and mean deviation of rtt for the appropriate type of rpc
 * for the frequent rpcs and a default for the others.
 * The justification for doing "other" this way is that these rpcs
 * happen so infrequently that timer est. would probably be stale.
 * Also, since many of these rpcs are
 * non-idempotent, a conservative timeout is desired.
 * getattr, lookup - A+2D
 * read, write     - A+4D
 * other           - nm_timeo
 */
#define	NFS_RTO(n, t) \
	((t) == 0 ? (n)->nm_timeo : \
	 ((t) < 3 ? \
	  (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \
	  ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1)))
#define	NFS_SRTT(r)	(r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1]
#define	NFS_SDRTT(r)	(r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1]
/*
 * External data, mostly RPC constants in XDR form
 */
extern u_long rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix,
	rpc_msgaccepted, rpc_call, rpc_autherr, rpc_rejectedcred,
	rpc_auth_kerb;
extern u_long nfs_prog, nfs_vers, nqnfs_prog, nqnfs_vers;
extern time_t nqnfsstarttime;
extern int nonidempotent[NFS_NPROCS];

/*
 * Maps errno values to nfs error numbers.
 * Use NFSERR_IO as the catch all for ones not specifically defined in
 * RFC 1094.
 */
static int nfsrv_errmap[ELAST] = {
  NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
  NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
  NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
  NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
  NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
  NFSERR_IO,
};

/*
 * Defines which timer to use for the procnum.
 * 0 - default
 * 1 - getattr
 * 2 - lookup
 * 3 - read
 * 4 - write
 */
static int proct[NFS_NPROCS] = {
	0, 1, 0, 0, 2, 3, 3, 0, 4, 0, 0, 0, 0, 0, 0, 0, 3, 0, 3, 0, 0, 0, 0,
};

/*
 * There is a congestion window for outstanding rpcs maintained per mount
 * point. The cwnd size is adjusted in roughly the way that:
 * Van Jacobson, Congestion avoidance and Control, In "Proceedings of
 * SIGCOMM '88". ACM, August 1988.
 * describes for TCP. The cwnd size is chopped in half on a retransmit timeout
 * and incremented by 1/cwnd when each rpc reply is received and a full cwnd
 * of rpcs is in progress.
 * (The sent count and cwnd are scaled for integer arith.)
 * Variants of "slow start" were tried and were found to be too much of a
 * performance hit (ave. rtt 3 times larger),
 * I suspect due to the large rtt that nfs rpcs have.
 */
#define	NFS_CWNDSCALE	256
#define	NFS_MAXCWND	(NFS_CWNDSCALE * 32)
static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, };
int	nfs_sbwait();
void	nfs_disconnect(), nfs_realign(), nfsrv_wakenfsd(), nfs_sndunlock();
void	nfs_rcvunlock(), nqnfs_serverd(), nqnfs_clientlease();
struct mbuf *nfsm_rpchead();
int nfsrtton = 0;
struct nfsrtt nfsrtt;
struct nfsd nfsd_head;

int	nfsrv_null(),
	nfsrv_getattr(),
	nfsrv_setattr(),
	nfsrv_lookup(),
	nfsrv_readlink(),
	nfsrv_read(),
	nfsrv_write(),
	nfsrv_create(),
	nfsrv_remove(),
	nfsrv_rename(),
	nfsrv_link(),
	nfsrv_symlink(),
	nfsrv_mkdir(),
	nfsrv_rmdir(),
	nfsrv_readdir(),
	nfsrv_statfs(),
	nfsrv_noop(),
	nqnfsrv_readdirlook(),
	nqnfsrv_getlease(),
	nqnfsrv_vacated(),
	nqnfsrv_access();

int (*nfsrv_procs[NFS_NPROCS])() = {
	nfsrv_null,
	nfsrv_getattr,
	nfsrv_setattr,
	nfsrv_noop,
	nfsrv_lookup,
	nfsrv_readlink,
	nfsrv_read,
	nfsrv_noop,
	nfsrv_write,
	nfsrv_create,
	nfsrv_remove,
	nfsrv_rename,
	nfsrv_link,
	nfsrv_symlink,
	nfsrv_mkdir,
	nfsrv_rmdir,
	nfsrv_readdir,
	nfsrv_statfs,
	nqnfsrv_readdirlook,
	nqnfsrv_getlease,
	nqnfsrv_vacated,
	nfsrv_noop,
	nqnfsrv_access,
};

struct nfsreq nfsreqh;

/*
 * Initialize sockets and congestion for a new NFS connection.
 * We do not free the sockaddr if error.
 */
nfs_connect(nmp, rep)
	register struct nfsmount *nmp;
	struct nfsreq *rep;
{
	register struct socket *so;
	int s, error, rcvreserve, sndreserve;
	struct sockaddr *saddr;
	struct sockaddr_in *sin;
	struct mbuf *m;
	u_short tport;

	nmp->nm_so = (struct socket *)0;
	saddr = mtod(nmp->nm_nam, struct sockaddr *);
	if (error = socreate(saddr->sa_family,
		&nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto))
		goto bad;
	so = nmp->nm_so;
	nmp->nm_soflags = so->so_proto->pr_flags;

	/*
	 * Some servers require that the client port be a reserved port number.
	 */
	if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) {
		MGET(m, M_WAIT, MT_SONAME);
		sin = mtod(m, struct sockaddr_in *);
		sin->sin_len = m->m_len = sizeof (struct sockaddr_in);
		sin->sin_family = AF_INET;
		sin->sin_addr.s_addr = INADDR_ANY;
		tport = IPPORT_RESERVED - 1;
		sin->sin_port = htons(tport);
		while ((error = sobind(so, m)) == EADDRINUSE &&
		       --tport > IPPORT_RESERVED / 2)
			sin->sin_port = htons(tport);
		m_freem(m);
		if (error)
			goto bad;
	}

	/*
	 * Protocols that do not require connections may be optionally left
	 * unconnected for servers that reply from a port other than NFS_PORT.
	 */
	if (nmp->nm_flag & NFSMNT_NOCONN) {
		if (nmp->nm_soflags & PR_CONNREQUIRED) {
			error = ENOTCONN;
			goto bad;
		}
	} else {
		if (error = soconnect(so, nmp->nm_nam))
			goto bad;

		/*
		 * Wait for the connection to complete. Cribbed from the
		 * connect system call but with the wait timing out so
		 * that interruptible mounts don't hang here for a long time.
		 */
		s = splnet();
		while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
			(void) tsleep((caddr_t)&so->so_timeo, PSOCK,
				"nfscon", 2 * hz);
			if ((so->so_state & SS_ISCONNECTING) &&
			    so->so_error == 0 && rep &&
			    (error = nfs_sigintr(nmp, rep, rep->r_procp))) {
				so->so_state &= ~SS_ISCONNECTING;
				splx(s);
				goto bad;
			}
		}
		if (so->so_error) {
			error = so->so_error;
			so->so_error = 0;
			splx(s);
			goto bad;
		}
		splx(s);
	}
	if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
		so->so_rcv.sb_timeo = (5 * hz);
		so->so_snd.sb_timeo = (5 * hz);
	} else {
		so->so_rcv.sb_timeo = 0;
		so->so_snd.sb_timeo = 0;
	}
	if (nmp->nm_sotype == SOCK_DGRAM) {
		sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR;
		rcvreserve = nmp->nm_rsize + NFS_MAXPKTHDR;
	} else if (nmp->nm_sotype == SOCK_SEQPACKET) {
		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2;
		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * 2;
	} else {
		if (nmp->nm_sotype != SOCK_STREAM)
			panic("nfscon sotype");
		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
			MGET(m, M_WAIT, MT_SOOPTS);
			*mtod(m, int *) = 1;
			m->m_len = sizeof(int);
			sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
		}
		if (so->so_proto->pr_protocol == IPPROTO_TCP) {
			MGET(m, M_WAIT, MT_SOOPTS);
			*mtod(m, int *) = 1;
			m->m_len = sizeof(int);
			sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
		}
		sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof (u_long))
				* 2;
		rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + sizeof (u_long))
				* 2;
	}
	if (error = soreserve(so, sndreserve, rcvreserve))
		goto bad;
	so->so_rcv.sb_flags |= SB_NOINTR;
	so->so_snd.sb_flags |= SB_NOINTR;

	/* Initialize other non-zero congestion variables */
	nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =
		nmp->nm_srtt[4] = (NFS_TIMEO << 3);
	nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] =
		nmp->nm_sdrtt[3] = nmp->nm_sdrtt[4] = 0;
	nmp->nm_cwnd = NFS_MAXCWND / 2;	    /* Initial send window */
	nmp->nm_sent = 0;
	nmp->nm_timeouts = 0;
	return (0);

bad:
	nfs_disconnect(nmp);
	return (error);
}

/*
 * Reconnect routine:
 * Called when a connection is broken on a reliable protocol.
 * - clean up the old socket
 * - nfs_connect() again
 * - set R_MUSTRESEND for all outstanding requests on mount point
 * If this fails the mount point is DEAD!
 * nb: Must be called with the nfs_sndlock() set on the mount point.
 */
nfs_reconnect(rep)
	register struct nfsreq *rep;
{
	register struct nfsreq *rp;
	register struct nfsmount *nmp = rep->r_nmp;
	int error;

	nfs_disconnect(nmp);
	while (error = nfs_connect(nmp, rep)) {
		if (error == EINTR || error == ERESTART)
			return (EINTR);
		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0);
	}

	/*
	 * Loop through outstanding request list and fix up all requests
	 * on old socket.
	 */
	rp = nfsreqh.r_next;
	while (rp != &nfsreqh) {
		if (rp->r_nmp == nmp)
			rp->r_flags |= R_MUSTRESEND;
		rp = rp->r_next;
	}
	return (0);
}

/*
 * NFS disconnect. Clean up and unlink.
 */
void
nfs_disconnect(nmp)
	register struct nfsmount *nmp;
{
	register struct socket *so;

	if (nmp->nm_so) {
		so = nmp->nm_so;
		nmp->nm_so = (struct socket *)0;
		soshutdown(so, 2);
		soclose(so);
	}
}

/*
 * This is the nfs send routine. For connection based socket types, it
 * must be called with an nfs_sndlock() on the socket.
 * "rep == NULL" indicates that it has been called from a server.
 * For the client side:
 * - return EINTR if the RPC is terminated, 0 otherwise
 * - set R_MUSTRESEND if the send fails for any reason
 * - do any cleanup required by recoverable socket errors (???)
 * For the server side:
 * - return EINTR or ERESTART if interrupted by a signal
 * - return EPIPE if a connection is lost for connection based sockets (TCP...)
 * - do any cleanup required by recoverable socket errors (???)
 */
nfs_send(so, nam, top, rep)
	register struct socket *so;
	struct mbuf *nam;
	register struct mbuf *top;
	struct nfsreq *rep;
{
	struct mbuf *sendnam;
	int error, soflags, flags;

	if (rep) {
		if (rep->r_flags & R_SOFTTERM) {
			m_freem(top);
			return (EINTR);
		}
		if ((so = rep->r_nmp->nm_so) == NULL) {
			rep->r_flags |= R_MUSTRESEND;
			m_freem(top);
			return (0);
		}
		rep->r_flags &= ~R_MUSTRESEND;
		soflags = rep->r_nmp->nm_soflags;
	} else
		soflags = so->so_proto->pr_flags;
	if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
		sendnam = (struct mbuf *)0;
	else
		sendnam = nam;
	if (so->so_type == SOCK_SEQPACKET)
		flags = MSG_EOR;
	else
		flags = 0;

	error = sosend(so, sendnam, (struct uio *)0, top,
		(struct mbuf *)0, flags);
	if (error) {
		if (rep) {
			log(LOG_INFO, "nfs send error %d for server %s\n",error,
			    rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname);
			/*
			 * Deal with errors for the client side.
			 */
			if (rep->r_flags & R_SOFTTERM)
				error = EINTR;
			else
				rep->r_flags |= R_MUSTRESEND;
		} else
			log(LOG_INFO, "nfsd send error %d\n", error);

		/*
		 * Handle any recoverable (soft) socket errors here. (???)
		 */
		if (error != EINTR && error != ERESTART &&
			error != EWOULDBLOCK && error != EPIPE)
			error = 0;
	}
	return (error);
}

/*
 * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all
 * done by soreceive(), but for SOCK_STREAM we must deal with the Record
 * Mark and consolidate the data into a new mbuf list.
 * nb: Sometimes TCP passes the data up to soreceive() in long lists of
 *     small mbufs.
 * For SOCK_STREAM we must be very careful to read an entire record once
 * we have read any of it, even if the system call has been interrupted.
 */
nfs_receive(rep, aname, mp)
	register struct nfsreq *rep;
	struct mbuf **aname;
	struct mbuf **mp;
{
	register struct socket *so;
	struct uio auio;
	struct iovec aio;
	register struct mbuf *m;
	struct mbuf *control;
	u_long len;
	struct mbuf **getnam;
	int error, sotype, rcvflg;
	struct proc *p = curproc;	/* XXX */

	/*
	 * Set up arguments for soreceive()
	 */
	*mp = (struct mbuf *)0;
	*aname = (struct mbuf *)0;
	sotype = rep->r_nmp->nm_sotype;