Rework NFS Duplicate Request Cache cleanup logic.

 - Introduce additional hash to group requests by hash of sockref.  This
allows to process TCP acknowledgements without looping though all the cache,
and as result allows to do it every time.
 - Indroduce additional callbacks to notify application layer about sockets
disconnection.  Without this last few requests processed just before socket
disconnection never processed their ACKs and stuck in cache for many hours.
 - Implement transport-specific method for tracking reply acknowledgements.
New implementation does not cross multiple stack layers to get the data and
does not have race conditions that previously made some requests stuck
in cache.  This could be done more efficiently at sockbuf layer, but that
would broke some KBIs, while I don't know other consumers for it aside NFS.
 - Instead of traversing all DRC twice per request, run cleaning only once
per request, and except in some conditions traverse only single hash slot
at a time.

Together this limits NFS DRC growth only to situations of real connectivity
problems.  If network is working well, and so all replies are acknowledged,
cache remains almost empty even after hours of heavy load.  Without this
change on the same test cache was growing to many thousand requests even
with perfectly working local network.

As another result this reduces CPU time spent on the DRC handling during
SPEC NFS benchmark from about 10% to 0.5%.

Sponsored by:	iXsystems, Inc.

1 files changed, 48 insertions, 3 deletions

diff --git a/sys/rpc/svc.h b/sys/rpc/svc.h
index 7d9428ef498f..bcd2d79a0b10 100644
--- a/sys/rpc/svc.h
+++ b/sys/rpc/svc.h
@@ -103,9 +103,11 @@ struct xp_ops {
 	    struct sockaddr **, struct mbuf **);
 	/* get transport status */
 	enum xprt_stat (*xp_stat)(struct __rpc_svcxprt *);
+	/* get transport acknowledge sequence */
+	bool_t (*xp_ack)(struct __rpc_svcxprt *, uint32_t *);
 	/* send reply */
 	bool_t	(*xp_reply)(struct __rpc_svcxprt *, struct rpc_msg *,
-	    struct sockaddr *, struct mbuf *);
+	    struct sockaddr *, struct mbuf *, uint32_t *);
 	/* destroy this struct */
 	void	(*xp_destroy)(struct __rpc_svcxprt *);
 	/* catch-all function */
@@ -166,6 +168,8 @@ typedef struct __rpc_svcxprt {
 	time_t		xp_lastactive;	/* time of last RPC */
 	u_int64_t	xp_sockref;	/* set by nfsv4 to identify socket */
 	int		xp_upcallset;	/* socket upcall is set up */
+	uint32_t	xp_snd_cnt;	/* # of bytes sent to socket */
+	struct sx	xp_snd_lock;	/* protects xp_snd_cnt & sb_cc */
 #else
 	int		xp_fd;
 	u_short		xp_port;	 /* associated port number */
@@ -230,6 +234,17 @@ struct svc_callout {
 };
 TAILQ_HEAD(svc_callout_list, svc_callout);
 
+/*
+ * The services connection loss list
+ * The dispatch routine takes request structs and runs the
+ * apropriate procedure.
+ */
+struct svc_loss_callout {
+	TAILQ_ENTRY(svc_loss_callout) slc_link;
+	void		    (*slc_dispatch)(SVCXPRT *);
+};
+TAILQ_HEAD(svc_loss_callout_list, svc_loss_callout);
+
 struct __rpc_svcthread;
 
 /*
@@ -253,6 +268,7 @@ struct svc_req {
 	void		*rq_p1;		/* application workspace */
 	int		rq_p2;		/* application workspace */
 	uint64_t	rq_p3;		/* application workspace */
+	uint32_t	rq_reply_seq;	/* reply socket sequence # */
 	char		rq_credarea[3*MAX_AUTH_BYTES];
 };
 STAILQ_HEAD(svc_reqlist, svc_req);
@@ -318,6 +334,7 @@ typedef struct __rpc_svcpool {
 	struct svcxprt_list sp_xlist;	/* all transports in the pool */
 	struct svcxprt_list sp_active;	/* transports needing service */
 	struct svc_callout_list sp_callouts; /* (prog,vers)->dispatch list */
+	struct svc_loss_callout_list sp_lcallouts; /* loss->dispatch list */
 	struct svcthread_list sp_threads; /* service threads */
 	struct svcthread_list sp_idlethreads; /* idle service threads */
 	int		sp_minthreads;	/* minimum service thread count */
@@ -393,8 +410,12 @@ struct svc_req {
 #define SVC_STAT(xprt)					\
 	(*(xprt)->xp_ops->xp_stat)(xprt)
 
-#define SVC_REPLY(xprt, msg, addr, m)			\
-	(*(xprt)->xp_ops->xp_reply) ((xprt), (msg), (addr), (m))
+#define SVC_ACK(xprt, ack)				\
+	((xprt)->xp_ops->xp_stat == NULL ? FALSE :	\
+	    ((ack) == NULL ? TRUE : (*(xprt)->xp_ops->xp_ack)((xprt), (ack))))
+
+#define SVC_REPLY(xprt, msg, addr, m, seq)			\
+	(*(xprt)->xp_ops->xp_reply) ((xprt), (msg), (addr), (m), (seq))
 
 #define SVC_DESTROY(xprt)				\
 	(*(xprt)->xp_ops->xp_destroy)(xprt)
@@ -496,6 +517,30 @@ extern void	svc_unreg(const rpcprog_t, const rpcvers_t);
 __END_DECLS
 
 /*
+ * Service connection loss registration
+ *
+ * svc_loss_reg(xprt, dispatch)
+ *	const SVCXPRT *xprt;
+ *	const void (*dispatch)();
+ */
+
+__BEGIN_DECLS
+extern bool_t	svc_loss_reg(SVCXPRT *, void (*)(SVCXPRT *));
+__END_DECLS
+
+/*
+ * Service connection loss un-registration
+ *
+ * svc_loss_unreg(xprt, dispatch)
+ *	const SVCXPRT *xprt;
+ *	const void (*dispatch)();
+ */
+
+__BEGIN_DECLS
+extern void	svc_loss_unreg(SVCPOOL *, void (*)(SVCXPRT *));
+__END_DECLS
+
+/*
  * Transport registration.
  *
  * xprt_register(xprt)