Implement a CPU-affine TCP and UDP connection lookup data structure,

struct inpcbgroup.  pcbgroups, or "connection groups", supplement the
existing inpcbinfo connection hash table, which when pcbgroups are
enabled, might now be thought of more usefully as a per-protocol
4-tuple reservation table.

Connections are assigned to connection groups base on a hash of their
4-tuple; wildcard sockets require special handling, and are members
of all connection groups.  During a connection lookup, a
per-connection group lock is employed rather than the global pcbinfo
lock.  By aligning connection groups with input path processing,
connection groups take on an effective CPU affinity, especially when
aligned with RSS work placement (see a forthcoming commit for
details).  This eliminates cache line migration associated with
global, protocol-layer data structures in steady state TCP and UDP
processing (with the exception of protocol-layer statistics; further
commit to follow).

Elements of this approach were inspired by Willman, Rixner, and Cox's
2006 USENIX paper, "An Evaluation of Network Stack Parallelization
Strategies in Modern Operating Systems".  However, there are also
significant differences: we maintain the inpcb lock, rather than using
the connection group lock for per-connection state.

Likewise, the focus of this implementation is alignment with NIC
packet distribution strategies such as RSS, rather than pure software
strategies.  Despite that focus, software distribution is supported
through the parallel netisr implementation, and works well in
configurations where the number of hardware threads is greater than
the number of NIC input queues, such as in the RMI XLR threaded MIPS
architecture.

Another important difference is the continued maintenance of existing
hash tables as "reservation tables" -- these are useful both to
distinguish the resource allocation aspect of protocol name management
and the more common-case lookup aspect.  In configurations where
connection tables are aligned with hardware hashes, it is desirable to
use the traditional lookup tables for loopback or encapsulated traffic
rather than take the expense of hardware hashes that are hard to
implement efficiently in software (such as RSS Toeplitz).

Connection group support is enabled by compiling "options PCBGROUP"
into your kernel configuration; for the time being, this is an
experimental feature, and hence is not enabled by default.

Subject to the limited MFCability of change dependencies in inpcb,
and its change to the inpcbinfo init function signature, this change
in principle could be merged to FreeBSD 8.x.

Reviewed by:    bz
Sponsored by:   Juniper Networks, Inc.

1 files changed, 167 insertions, 0 deletions

diff --git a/sys/netinet6/in6_pcb.c b/sys/netinet6/in6_pcb.c
index da73f21..d15c605 100644
--- a/sys/netinet6/in6_pcb.c
+++ b/sys/netinet6/in6_pcb.c
@@ -70,6 +70,7 @@ __FBSDID("$FreeBSD$");
 #include "opt_inet.h"
 #include "opt_inet6.h"
 #include "opt_ipsec.h"
+#include "opt_pcbgroup.h"
 
 #include <sys/param.h>
 #include <sys/systm.h>
@@ -827,6 +828,141 @@ in6_rtchange(struct inpcb *inp, int errno)
 	return inp;
 }
 
+#ifdef PCBGROUP
+/*
+ * Lookup PCB in hash list, using pcbgroup tables.
+ */
+static struct inpcb *
+in6_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
+    struct in6_addr *faddr, u_int fport_arg, struct in6_addr *laddr,
+    u_int lport_arg, int lookupflags, struct ifnet *ifp)
+{
+	struct inpcbhead *head;
+	struct inpcb *inp, *tmpinp;
+	u_short fport = fport_arg, lport = lport_arg;
+	int faith;
+
+	if (faithprefix_p != NULL)
+		faith = (*faithprefix_p)(laddr);
+	else
+		faith = 0;
+
+	/*
+	 * First look for an exact match.
+	 */
+	tmpinp = NULL;
+	INP_GROUP_LOCK(pcbgroup);
+	head = &pcbgroup->ipg_hashbase[
+	    INP_PCBHASH(faddr->s6_addr32[3] /* XXX */, lport, fport,
+	    pcbgroup->ipg_hashmask)];
+	LIST_FOREACH(inp, head, inp_pcbgrouphash) {
+		/* XXX inp locking */
+		if ((inp->inp_vflag & INP_IPV6) == 0)
+			continue;
+		if (IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr, faddr) &&
+		    IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr) &&
+		    inp->inp_fport == fport &&
+		    inp->inp_lport == lport) {
+			/*
+			 * XXX We should be able to directly return
+			 * the inp here, without any checks.
+			 * Well unless both bound with SO_REUSEPORT?
+			 */
+			if (prison_flag(inp->inp_cred, PR_IP6))
+				goto found;
+			if (tmpinp == NULL)
+				tmpinp = inp;
+		}
+	}
+	if (tmpinp != NULL) {
+		inp = tmpinp;
+		goto found;
+	}
+
+	/*
+	 * Then look for a wildcard match, if requested.
+	 */
+	if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
+		struct inpcb *local_wild = NULL, *local_exact = NULL;
+		struct inpcb *jail_wild = NULL;
+		int injail;
+
+		/*
+		 * Order of socket selection - we always prefer jails.
+		 *      1. jailed, non-wild.
+		 *      2. jailed, wild.
+		 *      3. non-jailed, non-wild.
+		 *      4. non-jailed, wild.
+		 */
+		head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
+		    0, pcbinfo->ipi_wildmask)];
+		LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
+			/* XXX inp locking */
+			if ((inp->inp_vflag & INP_IPV6) == 0)
+				continue;
+
+			if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) ||
+			    inp->inp_lport != lport) {
+				continue;
+			}
+
+			/* XXX inp locking */
+			if (faith && (inp->inp_flags & INP_FAITH) == 0)
+				continue;
+
+			injail = prison_flag(inp->inp_cred, PR_IP6);
+			if (injail) {
+				if (prison_check_ip6(inp->inp_cred,
+				    laddr) != 0)
+					continue;
+			} else {
+				if (local_exact != NULL)
+					continue;
+			}
+
+			if (IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr)) {
+				if (injail)
+					goto found;
+				else
+					local_exact = inp;
+			} else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) {
+				if (injail)
+					jail_wild = inp;
+				else
+					local_wild = inp;
+			}
+		} /* LIST_FOREACH */
+
+		inp = jail_wild;
+		if (inp == NULL)
+			inp = jail_wild;
+		if (inp == NULL)
+			inp = local_exact;
+		if (inp == NULL)
+			inp = local_wild;
+		if (inp != NULL)
+			goto found;
+	} /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
+	INP_GROUP_UNLOCK(pcbgroup);
+	return (NULL);
+
+found:
+	in_pcbref(inp);
+	INP_GROUP_UNLOCK(pcbgroup);
+	if (lookupflags & INPLOOKUP_WLOCKPCB) {
+		INP_WLOCK(inp);
+		if (in_pcbrele_wlocked(inp))
+			return (NULL);
+	} else if (lookupflags & INPLOOKUP_RLOCKPCB) {
+		INP_RLOCK(inp);
+		if (in_pcbrele_rlocked(inp))
+			return (NULL);
+	} else
+		panic("%s: locking buf", __func__);
+	return (inp);
+}
+#endif /* PCBGROUP */
+
 /*
  * Lookup PCB in hash list.
  */
@@ -983,16 +1119,30 @@ in6_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in6_addr *faddr,
 /*
  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
  * from which a pre-calculated hash value may be extracted.
+ *
+ * Possibly more of this logic should be in in6_pcbgroup.c.
  */
 struct inpcb *
 in6_pcblookup(struct inpcbinfo *pcbinfo, struct in6_addr *faddr, u_int fport,
     struct in6_addr *laddr, u_int lport, int lookupflags, struct ifnet *ifp)
 {
+#if defined(PCBGROUP)
+	struct inpcbgroup *pcbgroup;
+#endif
+
 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
 	    ("%s: LOCKPCB not set", __func__));
 
+#if defined(PCBGROUP)
+	if (in_pcbgroup_enabled(pcbinfo)) {
+		pcbgroup = in6_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
+		    fport);
+		return (in6_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
+		    laddr, lport, lookupflags, ifp));
+	}
+#endif
 	return (in6_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
 	    lookupflags, ifp));
 }
@@ -1002,11 +1152,28 @@ in6_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in6_addr *faddr,
     u_int fport, struct in6_addr *laddr, u_int lport, int lookupflags,
     struct ifnet *ifp, struct mbuf *m)
 {
+#ifdef PCBGROUP
+	struct inpcbgroup *pcbgroup;
+#endif
+
 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
 	    ("%s: LOCKPCB not set", __func__));
 
+#ifdef PCBGROUP
+	if (in_pcbgroup_enabled(pcbinfo)) {
+		pcbgroup = in6_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
+		    m->m_pkthdr.flowid);
+		if (pcbgroup != NULL)
+			return (in6_pcblookup_group(pcbinfo, pcbgroup, faddr,
+			    fport, laddr, lport, lookupflags, ifp));
+		pcbgroup = in6_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
+		    fport);
+		return (in6_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
+		    laddr, lport, lookupflags, ifp));
+	}
+#endif
 	return (in6_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
 	    lookupflags, ifp));
 }