xref: /csrg-svn/sys/nfs/nfs_srvcache.c (revision 68653)

/*
 * Copyright (c) 1989, 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.
 *
 * %sccs.include.redist.c%
 *
 *	@(#)nfs_srvcache.c	8.3 (Berkeley) 03/30/95
 */

/*
 * Reference: Chet Juszczak, "Improving the Performance and Correctness
 *		of an NFS Server", in Proc. Winter 1989 USENIX Conference,
 *		pages 53-63. San Diego, February 1989.
 */
#include <sys/param.h>
#include <sys/vnode.h>
#include <sys/mount.h>
#include <sys/kernel.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/mbuf.h>
#include <sys/malloc.h>
#include <sys/socket.h>
#include <sys/socketvar.h>

#include <netinet/in.h>
#ifdef ISO
#include <netiso/iso.h>
#endif
#include <nfs/nfsm_subs.h>
#include <nfs/rpcv2.h>
#include <nfs/nfsproto.h>
#include <nfs/nfs.h>
#include <nfs/nfsrvcache.h>
#include <nfs/nqnfs.h>

extern struct nfsstats nfsstats;
extern int nfsv2_procid[NFS_NPROCS];
long numnfsrvcache, desirednfsrvcache = NFSRVCACHESIZ;

#define	NFSRCHASH(xid) \
	(&nfsrvhashtbl[((xid) + ((xid) >> 24)) & nfsrvhash])
LIST_HEAD(nfsrvhash, nfsrvcache) *nfsrvhashtbl;
TAILQ_HEAD(nfsrvlru, nfsrvcache) nfsrvlruhead;
u_long nfsrvhash;

#define TRUE	1
#define	FALSE	0

#define	NETFAMILY(rp) \
		(((rp)->rc_flag & RC_INETADDR) ? AF_INET : AF_ISO)

/*
 * Static array that defines which nfs rpc's are nonidempotent
 */
int nonidempotent[NFS_NPROCS] = {
	FALSE,
	FALSE,
	TRUE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	TRUE,
	TRUE,
	TRUE,
	TRUE,
	TRUE,
	TRUE,
	TRUE,
	TRUE,
	TRUE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
};

/* True iff the rpc reply is an nfs status ONLY! */
static int nfsv2_repstat[NFS_NPROCS] = {
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	FALSE,
	TRUE,
	TRUE,
	TRUE,
	TRUE,
	FALSE,
	TRUE,
	FALSE,
	FALSE,
};

/*
 * Initialize the server request cache list
 */
void
nfsrv_initcache()
{

	nfsrvhashtbl = hashinit(desirednfsrvcache, M_NFSD, &nfsrvhash);
	TAILQ_INIT(&nfsrvlruhead);
}

/*
 * Look for the request in the cache
 * If found then
 *    return action and optionally reply
 * else
 *    insert it in the cache
 *
 * The rules are as follows:
 * - if in progress, return DROP request
 * - if completed within DELAY of the current time, return DROP it
 * - if completed a longer time ago return REPLY if the reply was cached or
 *   return DOIT
 * Update/add new request at end of lru list
 */
int
nfsrv_getcache(nd, slp, repp)
	register struct nfsrv_descript *nd;
	struct nfssvc_sock *slp;
	struct mbuf **repp;
{
	register struct nfsrvcache *rp;
	struct mbuf *mb;
	struct sockaddr_in *saddr;
	caddr_t bpos;
	int ret;

	/*
	 * Don't cache recent requests for reliable transport protocols.
	 * (Maybe we should for the case of a reconnect, but..)
	 */
	if (!nd->nd_nam2)
		return (RC_DOIT);
loop:
	for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != 0;
	    rp = rp->rc_hash.le_next) {
	    if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
		netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
			if ((rp->rc_flag & RC_LOCKED) != 0) {
				rp->rc_flag |= RC_WANTED;
				(void) tsleep((caddr_t)rp, PZERO-1, "nfsrc", 0);
				goto loop;
			}
			rp->rc_flag |= RC_LOCKED;
			/* If not at end of LRU chain, move it there */
			if (rp->rc_lru.tqe_next) {
				TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
				TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
			}
			if (rp->rc_state == RC_UNUSED)
				panic("nfsrv cache");
			if (rp->rc_state == RC_INPROG) {
				nfsstats.srvcache_inproghits++;
				ret = RC_DROPIT;
			} else if (rp->rc_flag & RC_REPSTATUS) {
				nfsstats.srvcache_nonidemdonehits++;
				nfs_rephead(0, nd, slp, rp->rc_status,
				   0, (u_quad_t *)0, repp, &mb, &bpos);
				ret = RC_REPLY;
			} else if (rp->rc_flag & RC_REPMBUF) {
				nfsstats.srvcache_nonidemdonehits++;
				*repp = m_copym(rp->rc_reply, 0, M_COPYALL,
						M_WAIT);
				ret = RC_REPLY;
			} else {
				nfsstats.srvcache_idemdonehits++;
				rp->rc_state = RC_INPROG;
				ret = RC_DOIT;
			}
			rp->rc_flag &= ~RC_LOCKED;
			if (rp->rc_flag & RC_WANTED) {
				rp->rc_flag &= ~RC_WANTED;
				wakeup((caddr_t)rp);
			}
			return (ret);
		}
	}
	nfsstats.srvcache_misses++;
	if (numnfsrvcache < desirednfsrvcache) {
		rp = (struct nfsrvcache *)malloc((u_long)sizeof *rp,
		    M_NFSD, M_WAITOK);
		bzero((char *)rp, sizeof *rp);
		numnfsrvcache++;
		rp->rc_flag = RC_LOCKED;
	} else {
		rp = nfsrvlruhead.tqh_first;
		while ((rp->rc_flag & RC_LOCKED) != 0) {
			rp->rc_flag |= RC_WANTED;
			(void) tsleep((caddr_t)rp, PZERO-1, "nfsrc", 0);
			rp = nfsrvlruhead.tqh_first;
		}
		rp->rc_flag |= RC_LOCKED;
		LIST_REMOVE(rp, rc_hash);
		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
		if (rp->rc_flag & RC_REPMBUF)
			m_freem(rp->rc_reply);
		if (rp->rc_flag & RC_NAM)
			MFREE(rp->rc_nam, mb);
		rp->rc_flag &= (RC_LOCKED | RC_WANTED);
	}
	TAILQ_INSERT_TAIL(&nfsrvlruhead, rp, rc_lru);
	rp->rc_state = RC_INPROG;
	rp->rc_xid = nd->nd_retxid;
	saddr = mtod(nd->nd_nam, struct sockaddr_in *);
	switch (saddr->sin_family) {
	case AF_INET:
		rp->rc_flag |= RC_INETADDR;
		rp->rc_inetaddr = saddr->sin_addr.s_addr;
		break;
	case AF_ISO:
	default:
		rp->rc_flag |= RC_NAM;
		rp->rc_nam = m_copym(nd->nd_nam, 0, M_COPYALL, M_WAIT);
		break;
	};
	rp->rc_proc = nd->nd_procnum;
	LIST_INSERT_HEAD(NFSRCHASH(nd->nd_retxid), rp, rc_hash);
	rp->rc_flag &= ~RC_LOCKED;
	if (rp->rc_flag & RC_WANTED) {
		rp->rc_flag &= ~RC_WANTED;
		wakeup((caddr_t)rp);
	}
	return (RC_DOIT);
}

/*
 * Update a request cache entry after the rpc has been done
 */
void
nfsrv_updatecache(nd, repvalid, repmbuf)
	register struct nfsrv_descript *nd;
	int repvalid;
	struct mbuf *repmbuf;
{
	register struct nfsrvcache *rp;

	if (!nd->nd_nam2)
		return;
loop:
	for (rp = NFSRCHASH(nd->nd_retxid)->lh_first; rp != 0;
	    rp = rp->rc_hash.le_next) {
	    if (nd->nd_retxid == rp->rc_xid && nd->nd_procnum == rp->rc_proc &&
		netaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) {
			if ((rp->rc_flag & RC_LOCKED) != 0) {
				rp->rc_flag |= RC_WANTED;
				(void) tsleep((caddr_t)rp, PZERO-1, "nfsrc", 0);
				goto loop;
			}
			rp->rc_flag |= RC_LOCKED;
			rp->rc_state = RC_DONE;
			/*
			 * If we have a valid reply update status and save
			 * the reply for non-idempotent rpc's.
			 */
			if (repvalid && nonidempotent[nd->nd_procnum]) {
				if ((nd->nd_flag & ND_NFSV3) == 0 &&
				  nfsv2_repstat[nfsv2_procid[nd->nd_procnum]]) {
					rp->rc_status = nd->nd_repstat;
					rp->rc_flag |= RC_REPSTATUS;
				} else {
					rp->rc_reply = m_copym(repmbuf,
						0, M_COPYALL, M_WAIT);
					rp->rc_flag |= RC_REPMBUF;
				}
			}
			rp->rc_flag &= ~RC_LOCKED;
			if (rp->rc_flag & RC_WANTED) {
				rp->rc_flag &= ~RC_WANTED;
				wakeup((caddr_t)rp);
			}
			return;
		}
	}
}

/*
 * Clean out the cache. Called when the last nfsd terminates.
 */
void
nfsrv_cleancache()
{
	register struct nfsrvcache *rp, *nextrp;

	for (rp = nfsrvlruhead.tqh_first; rp != 0; rp = nextrp) {
		nextrp = rp->rc_lru.tqe_next;
		LIST_REMOVE(rp, rc_hash);
		TAILQ_REMOVE(&nfsrvlruhead, rp, rc_lru);
		free(rp, M_NFSD);
	}
	numnfsrvcache = 0;
}