freebsd-nq/sys/net/if_ethersubr.c

1207 lines
30 KiB
C
Raw Normal View History

/*-
1994-05-24 10:09:53 +00:00
* Copyright (c) 1982, 1989, 1993
* The Regents of the University of California. All rights reserved.
*
* 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.
* 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.
*
* @(#)if_ethersubr.c 8.1 (Berkeley) 6/10/93
1999-08-28 01:08:13 +00:00
* $FreeBSD$
1994-05-24 10:09:53 +00:00
*/
#include "opt_inet.h"
#include "opt_inet6.h"
#include "opt_netgraph.h"
#include "opt_mbuf_profiling.h"
Several years after initial development, merge prototype support for linking NIC Receive Side Scaling (RSS) to the network stack's connection-group implementation. This prototype (and derived patches) are in use at Juniper and several other FreeBSD-using companies, so despite some reservations about its maturity, merge the patch to the base tree so that it can be iteratively refined in collaboration rather than maintained as a set of gradually diverging patch sets. (1) Merge a software implementation of the Toeplitz hash specified in RSS implemented by David Malone. This is used to allow suitable pcbgroup placement of connections before the first packet is received from the NIC. Software hashing is generally avoided, however, due to high cost of the hash on general-purpose CPUs. (2) In in_rss.c, maintain authoritative versions of RSS state intended to be pushed to each NIC, including keying material, hash algorithm/ configuration, and buckets. Provide software-facing interfaces to hash 2- and 4-tuples for IPv4 and IPv6 using both the RSS standardised Toeplitz and a 'naive' variation with a hash efficient in software but with poor distribution properties. Implement rss_m2cpuid()to be used by netisr and other load balancing code to look up the CPU on which an mbuf should be processed. (3) In the Ethernet link layer, allow netisr distribution using RSS as a source of policy as an alternative to source ordering; continue to default to direct dispatch (i.e., don't try and requeue packets for processing on the 'right' CPU if they arrive in a directly dispatchable context). (4) Allow RSS to control tuning of connection groups in order to align groups with RSS buckets. If a packet arrives on a protocol using connection groups, and contains a suitable hardware-generated hash, use that hash value to select the connection group for pcb lookup for both IPv4 and IPv6. If no hardware-generated Toeplitz hash is available, we fall back on regular PCB lookup risking contention rather than pay the cost of Toeplitz in software -- this is a less scalable but, at my last measurement, faster approach. As core counts go up, we may want to revise this strategy despite CPU overhead. Where device drivers suitably configure NICs, and connection groups / RSS are enabled, this should avoid both lock and line contention during connection lookup for TCP. This commit does not modify any device drivers to tune device RSS configuration to the global RSS configuration; patches are in circulation to do this for at least Chelsio T3 and Intel 1G/10G drivers. Currently, the KPI for device drivers is not particularly robust, nor aware of more advanced features such as runtime reconfiguration/rebalancing. This will hopefully prove a useful starting point for refinement. No MFC is scheduled as we will first want to nail down a more mature and maintainable KPI/KBI for device drivers. Sponsored by: Juniper Networks (original work) Sponsored by: EMC/Isilon (patch update and merge)
2014-03-15 00:57:50 +00:00
#include "opt_rss.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
Conditionally compile out V_ globals while instantiating the appropriate container structures, depending on VIMAGE_GLOBALS compile time option. Make VIMAGE_GLOBALS a new compile-time option, which by default will not be defined, resulting in instatiations of global variables selected for V_irtualization (enclosed in #ifdef VIMAGE_GLOBALS blocks) to be effectively compiled out. Instantiate new global container structures to hold V_irtualized variables: vnet_net_0, vnet_inet_0, vnet_inet6_0, vnet_ipsec_0, vnet_netgraph_0, and vnet_gif_0. Update the VSYM() macro so that depending on VIMAGE_GLOBALS the V_ macros resolve either to the original globals, or to fields inside container structures, i.e. effectively #ifdef VIMAGE_GLOBALS #define V_rt_tables rt_tables #else #define V_rt_tables vnet_net_0._rt_tables #endif Update SYSCTL_V_*() macros to operate either on globals or on fields inside container structs. Extend the internal kldsym() lookups with the ability to resolve selected fields inside the virtualization container structs. This applies only to the fields which are explicitly registered for kldsym() visibility via VNET_MOD_DECLARE() and vnet_mod_register(), currently this is done only in sys/net/if.c. Fix a few broken instances of MODULE_GLOBAL() macro use in SCTP code, and modify the MODULE_GLOBAL() macro to resolve to V_ macros, which in turn result in proper code being generated depending on VIMAGE_GLOBALS. De-virtualize local static variables in sys/contrib/pf/net/pf_subr.c which were prematurely V_irtualized by automated V_ prepending scripts during earlier merging steps. PF virtualization will be done separately, most probably after next PF import. Convert a few variable initializations at instantiation to initialization in init functions, most notably in ipfw. Also convert TUNABLE_INT() initializers for V_ variables to TUNABLE_FETCH_INT() in initializer functions. Discussed at: devsummit Strassburg Reviewed by: bz, julian Approved by: julian (mentor) Obtained from: //depot/projects/vimage-commit2/... X-MFC after: never Sponsored by: NLnet Foundation, The FreeBSD Foundation
2008-12-10 23:12:39 +00:00
#include <sys/lock.h>
1994-05-24 10:09:53 +00:00
#include <sys/malloc.h>
#include <sys/module.h>
1994-05-24 10:09:53 +00:00
#include <sys/mbuf.h>
#include <sys/random.h>
1994-05-24 10:09:53 +00:00
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <sys/uuid.h>
1994-05-24 10:09:53 +00:00
#include <net/if.h>
#include <net/if_var.h>
#include <net/if_arp.h>
1994-05-24 10:09:53 +00:00
#include <net/netisr.h>
#include <net/route.h>
#include <net/if_llc.h>
#include <net/if_dl.h>
#include <net/if_types.h>
#include <net/bpf.h>
#include <net/ethernet.h>
#include <net/if_bridgevar.h>
#include <net/if_vlan_var.h>
#include <net/if_llatbl.h>
#include <net/pfil.h>
#include <net/vnet.h>
1994-05-24 10:09:53 +00:00
#include <netpfil/pf/pf_mtag.h>
#if defined(INET) || defined(INET6)
1994-05-24 10:09:53 +00:00
#include <netinet/in.h>
#include <netinet/in_var.h>
#include <netinet/if_ether.h>
Several years after initial development, merge prototype support for linking NIC Receive Side Scaling (RSS) to the network stack's connection-group implementation. This prototype (and derived patches) are in use at Juniper and several other FreeBSD-using companies, so despite some reservations about its maturity, merge the patch to the base tree so that it can be iteratively refined in collaboration rather than maintained as a set of gradually diverging patch sets. (1) Merge a software implementation of the Toeplitz hash specified in RSS implemented by David Malone. This is used to allow suitable pcbgroup placement of connections before the first packet is received from the NIC. Software hashing is generally avoided, however, due to high cost of the hash on general-purpose CPUs. (2) In in_rss.c, maintain authoritative versions of RSS state intended to be pushed to each NIC, including keying material, hash algorithm/ configuration, and buckets. Provide software-facing interfaces to hash 2- and 4-tuples for IPv4 and IPv6 using both the RSS standardised Toeplitz and a 'naive' variation with a hash efficient in software but with poor distribution properties. Implement rss_m2cpuid()to be used by netisr and other load balancing code to look up the CPU on which an mbuf should be processed. (3) In the Ethernet link layer, allow netisr distribution using RSS as a source of policy as an alternative to source ordering; continue to default to direct dispatch (i.e., don't try and requeue packets for processing on the 'right' CPU if they arrive in a directly dispatchable context). (4) Allow RSS to control tuning of connection groups in order to align groups with RSS buckets. If a packet arrives on a protocol using connection groups, and contains a suitable hardware-generated hash, use that hash value to select the connection group for pcb lookup for both IPv4 and IPv6. If no hardware-generated Toeplitz hash is available, we fall back on regular PCB lookup risking contention rather than pay the cost of Toeplitz in software -- this is a less scalable but, at my last measurement, faster approach. As core counts go up, we may want to revise this strategy despite CPU overhead. Where device drivers suitably configure NICs, and connection groups / RSS are enabled, this should avoid both lock and line contention during connection lookup for TCP. This commit does not modify any device drivers to tune device RSS configuration to the global RSS configuration; patches are in circulation to do this for at least Chelsio T3 and Intel 1G/10G drivers. Currently, the KPI for device drivers is not particularly robust, nor aware of more advanced features such as runtime reconfiguration/rebalancing. This will hopefully prove a useful starting point for refinement. No MFC is scheduled as we will first want to nail down a more mature and maintainable KPI/KBI for device drivers. Sponsored by: Juniper Networks (original work) Sponsored by: EMC/Isilon (patch update and merge)
2014-03-15 00:57:50 +00:00
#include <netinet/in_rss.h>
#include <netinet/ip_carp.h>
#include <netinet/ip_var.h>
#endif
#ifdef INET6
#include <netinet6/nd6.h>
#endif
1994-05-24 10:09:53 +00:00
int (*ef_inputp)(struct ifnet*, struct ether_header *eh, struct mbuf *m);
int (*ef_outputp)(struct ifnet *ifp, struct mbuf **mp,
const struct sockaddr *dst, short *tp, int *hlen);
#include <security/mac/mac_framework.h>
#ifdef CTASSERT
CTASSERT(sizeof (struct ether_header) == ETHER_ADDR_LEN * 2 + 2);
CTASSERT(sizeof (struct ether_addr) == ETHER_ADDR_LEN);
#endif
VNET_DEFINE(struct pfil_head, link_pfil_hook); /* Packet filter hooks */
/* netgraph node hooks for ng_ether(4) */
void (*ng_ether_input_p)(struct ifnet *ifp, struct mbuf **mp);
void (*ng_ether_input_orphan_p)(struct ifnet *ifp, struct mbuf *m);
int (*ng_ether_output_p)(struct ifnet *ifp, struct mbuf **mp);
void (*ng_ether_attach_p)(struct ifnet *ifp);
void (*ng_ether_detach_p)(struct ifnet *ifp);
void (*vlan_input_p)(struct ifnet *, struct mbuf *);
/* if_bridge(4) support */
struct mbuf *(*bridge_input_p)(struct ifnet *, struct mbuf *);
int (*bridge_output_p)(struct ifnet *, struct mbuf *,
struct sockaddr *, struct rtentry *);
void (*bridge_dn_p)(struct mbuf *, struct ifnet *);
/* if_lagg(4) support */
struct mbuf *(*lagg_input_p)(struct ifnet *, struct mbuf *);
static const u_char etherbroadcastaddr[ETHER_ADDR_LEN] =
{ 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
2002-03-19 21:54:18 +00:00
static int ether_resolvemulti(struct ifnet *, struct sockaddr **,
struct sockaddr *);
#ifdef VIMAGE
static void ether_reassign(struct ifnet *, struct vnet *, char *);
#endif
/* XXX: should be in an arp support file, not here */
static MALLOC_DEFINE(M_ARPCOM, "arpcom", "802.* interface internals");
#define ETHER_IS_BROADCAST(addr) \
(bcmp(etherbroadcastaddr, (addr), ETHER_ADDR_LEN) == 0)
#define senderr(e) do { error = (e); goto bad;} while (0)
1994-05-24 10:09:53 +00:00
static void
update_mbuf_csumflags(struct mbuf *src, struct mbuf *dst)
{
int csum_flags = 0;
if (src->m_pkthdr.csum_flags & CSUM_IP)
csum_flags |= (CSUM_IP_CHECKED|CSUM_IP_VALID);
if (src->m_pkthdr.csum_flags & CSUM_DELAY_DATA)
csum_flags |= (CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
if (src->m_pkthdr.csum_flags & CSUM_SCTP)
csum_flags |= CSUM_SCTP_VALID;
dst->m_pkthdr.csum_flags |= csum_flags;
if (csum_flags & CSUM_DATA_VALID)
dst->m_pkthdr.csum_data = 0xffff;
}
1994-05-24 10:09:53 +00:00
/*
* Ethernet output routine.
* Encapsulate a packet of type family for the local net.
* Use trailer local net encapsulation if enough data in first
* packet leaves a multiple of 512 bytes of data in remainder.
*/
int
ether_output(struct ifnet *ifp, struct mbuf *m,
const struct sockaddr *dst, struct route *ro)
1994-05-24 10:09:53 +00:00
{
short type;
int error = 0, hdrcmplt = 0;
u_char esrc[ETHER_ADDR_LEN], edst[ETHER_ADDR_LEN];
struct llentry *lle = NULL;
struct rtentry *rt0 = NULL;
2003-03-03 00:21:52 +00:00
struct ether_header *eh;
struct pf_mtag *t;
int loop_copy = 1;
2003-10-12 20:51:26 +00:00
int hlen; /* link layer header length */
1994-05-24 10:09:53 +00:00
if (ro != NULL) {
if (!(m->m_flags & (M_BCAST | M_MCAST)))
lle = ro->ro_lle;
rt0 = ro->ro_rt;
}
#ifdef MAC
error = mac_ifnet_check_transmit(ifp, m);
if (error)
senderr(error);
#endif
M_PROFILE(m);
if (ifp->if_flags & IFF_MONITOR)
senderr(ENETDOWN);
if (!((ifp->if_flags & IFF_UP) &&
(ifp->if_drv_flags & IFF_DRV_RUNNING)))
1994-05-24 10:09:53 +00:00
senderr(ENETDOWN);
hlen = ETHER_HDR_LEN;
1994-05-24 10:09:53 +00:00
switch (dst->sa_family) {
#ifdef INET
case AF_INET:
if (lle != NULL && (lle->la_flags & LLE_VALID))
memcpy(edst, &lle->ll_addr.mac16, sizeof(edst));
else
error = arpresolve(ifp, rt0, m, dst, edst, &lle);
This commit does two things: 1. rt_check() cleanup: rt_check() is only necessary for some address families to gain access to the corresponding arp entry, so call it only in/near the *resolve() routines where it is actually used -- at the moment this is arpresolve(), nd6_storelladdr() (the call is embedded here), and atmresolve() (the call is just before atmresolve to reduce the number of changes). This change will make it a lot easier to decouple the arp table from the routing table. There is an extra call to rt_check() in if_iso88025subr.c to determine the routing info length. I have left it alone for the time being. The interface of arpresolve() and nd6_storelladdr() now changes slightly: + the 'rtentry' parameter (really a hint from the upper level layer) is now passed unchanged from *_output(), so it becomes the route to the final destination and not to the gateway. + the routines will return 0 if resolution is possible, non-zero otherwise. + arpresolve() returns EWOULDBLOCK in case the mbuf is being held waiting for an arp reply -- in this case the error code is masked in the caller so the upper layer protocol will not see a failure. 2. arpcom untangling Where possible, use 'struct ifnet' instead of 'struct arpcom' variables, and use the IFP2AC macro to access arpcom fields. This mostly affects the netatalk code. === Detailed changes: === net/if_arcsubr.c rt_check() cleanup, remove a useless variable net/if_atmsubr.c rt_check() cleanup net/if_ethersubr.c rt_check() cleanup, arpcom untangling net/if_fddisubr.c rt_check() cleanup, arpcom untangling net/if_iso88025subr.c rt_check() cleanup netatalk/aarp.c arpcom untangling, remove a block of duplicated code netatalk/at_extern.h arpcom untangling netinet/if_ether.c rt_check() cleanup (change arpresolve) netinet6/nd6.c rt_check() cleanup (change nd6_storelladdr)
2004-04-25 09:24:52 +00:00
if (error)
return (error == EWOULDBLOCK ? 0 : error);
type = htons(ETHERTYPE_IP);
1994-05-24 10:09:53 +00:00
break;
case AF_ARP:
{
struct arphdr *ah;
ah = mtod(m, struct arphdr *);
ah->ar_hrd = htons(ARPHRD_ETHER);
loop_copy = 0; /* if this is for us, don't do it */
switch(ntohs(ah->ar_op)) {
case ARPOP_REVREQUEST:
case ARPOP_REVREPLY:
type = htons(ETHERTYPE_REVARP);
break;
case ARPOP_REQUEST:
case ARPOP_REPLY:
default:
type = htons(ETHERTYPE_ARP);
break;
}
if (m->m_flags & M_BCAST)
bcopy(ifp->if_broadcastaddr, edst, ETHER_ADDR_LEN);
else
bcopy(ar_tha(ah), edst, ETHER_ADDR_LEN);
}
break;
1994-05-24 10:09:53 +00:00
#endif
#ifdef INET6
case AF_INET6:
if (lle != NULL && (lle->la_flags & LLE_VALID))
memcpy(edst, &lle->ll_addr.mac16, sizeof(edst));
else
error = nd6_storelladdr(ifp, m, dst, (u_char *)edst, &lle);
This commit does two things: 1. rt_check() cleanup: rt_check() is only necessary for some address families to gain access to the corresponding arp entry, so call it only in/near the *resolve() routines where it is actually used -- at the moment this is arpresolve(), nd6_storelladdr() (the call is embedded here), and atmresolve() (the call is just before atmresolve to reduce the number of changes). This change will make it a lot easier to decouple the arp table from the routing table. There is an extra call to rt_check() in if_iso88025subr.c to determine the routing info length. I have left it alone for the time being. The interface of arpresolve() and nd6_storelladdr() now changes slightly: + the 'rtentry' parameter (really a hint from the upper level layer) is now passed unchanged from *_output(), so it becomes the route to the final destination and not to the gateway. + the routines will return 0 if resolution is possible, non-zero otherwise. + arpresolve() returns EWOULDBLOCK in case the mbuf is being held waiting for an arp reply -- in this case the error code is masked in the caller so the upper layer protocol will not see a failure. 2. arpcom untangling Where possible, use 'struct ifnet' instead of 'struct arpcom' variables, and use the IFP2AC macro to access arpcom fields. This mostly affects the netatalk code. === Detailed changes: === net/if_arcsubr.c rt_check() cleanup, remove a useless variable net/if_atmsubr.c rt_check() cleanup net/if_ethersubr.c rt_check() cleanup, arpcom untangling net/if_fddisubr.c rt_check() cleanup, arpcom untangling net/if_iso88025subr.c rt_check() cleanup netatalk/aarp.c arpcom untangling, remove a block of duplicated code netatalk/at_extern.h arpcom untangling netinet/if_ether.c rt_check() cleanup (change arpresolve) netinet6/nd6.c rt_check() cleanup (change nd6_storelladdr)
2004-04-25 09:24:52 +00:00
if (error)
return error;
type = htons(ETHERTYPE_IPV6);
break;
#endif
case pseudo_AF_HDRCMPLT:
{
const struct ether_header *eh;
hdrcmplt = 1;
eh = (const struct ether_header *)dst->sa_data;
(void)memcpy(esrc, eh->ether_shost, sizeof (esrc));
/* FALLTHROUGH */
1994-05-24 10:09:53 +00:00
case AF_UNSPEC:
loop_copy = 0; /* if this is for us, don't do it */
eh = (const struct ether_header *)dst->sa_data;
(void)memcpy(edst, eh->ether_dhost, sizeof (edst));
1994-05-24 10:09:53 +00:00
type = eh->ether_type;
break;
}
1994-05-24 10:09:53 +00:00
default:
if_printf(ifp, "can't handle af%d\n", dst->sa_family);
1994-05-24 10:09:53 +00:00
senderr(EAFNOSUPPORT);
}
if (lle != NULL && (lle->la_flags & LLE_IFADDR)) {
update_mbuf_csumflags(m, m);
return (if_simloop(ifp, m, dst->sa_family, 0));
}
1994-05-24 10:09:53 +00:00
/*
* Add local net header. If no space in first mbuf,
* allocate another.
*/
M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT);
if (m == NULL)
1994-05-24 10:09:53 +00:00
senderr(ENOBUFS);
eh = mtod(m, struct ether_header *);
(void)memcpy(&eh->ether_type, &type,
1994-05-24 10:09:53 +00:00
sizeof(eh->ether_type));
(void)memcpy(eh->ether_dhost, edst, sizeof (edst));
if (hdrcmplt)
(void)memcpy(eh->ether_shost, esrc,
sizeof(eh->ether_shost));
else
(void)memcpy(eh->ether_shost, IF_LLADDR(ifp),
sizeof(eh->ether_shost));
/*
* If a simplex interface, and the packet is being sent to our
* Ethernet address or a broadcast address, loopback a copy.
* XXX To make a simplex device behave exactly like a duplex
* device, we should copy in the case of sending to our own
* ethernet address (thus letting the original actually appear
* on the wire). However, we don't do that here for security
* reasons and compatibility with the original behavior.
*/
if ((ifp->if_flags & IFF_SIMPLEX) && loop_copy &&
((t = pf_find_mtag(m)) == NULL || !t->routed)) {
if (m->m_flags & M_BCAST) {
2003-03-15 19:37:44 +00:00
struct mbuf *n;
/*
* Because if_simloop() modifies the packet, we need a
* writable copy through m_dup() instead of a readonly
* one as m_copy[m] would give us. The alternative would
* be to modify if_simloop() to handle the readonly mbuf,
* but performancewise it is mostly equivalent (trading
* extra data copying vs. extra locking).
*
* XXX This is a local workaround. A number of less
* often used kernel parts suffer from the same bug.
* See PR kern/105943 for a proposed general solution.
*/
if ((n = m_dup(m, M_NOWAIT)) != NULL) {
update_mbuf_csumflags(m, n);
(void)if_simloop(ifp, n, dst->sa_family, hlen);
} else
if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
2003-03-15 19:37:44 +00:00
} else if (bcmp(eh->ether_dhost, eh->ether_shost,
ETHER_ADDR_LEN) == 0) {
update_mbuf_csumflags(m, m);
(void) if_simloop(ifp, m, dst->sa_family, hlen);
return (0); /* XXX */
}
}
/*
* Bridges require special output handling.
*/
if (ifp->if_bridge) {
BRIDGE_OUTPUT(ifp, m, error);
return (error);
}
#if defined(INET) || defined(INET6)
if (ifp->if_carp &&
(error = (*carp_output_p)(ifp, m, dst)))
goto bad;
#endif
/* Handle ng_ether(4) processing, if any */
if (IFP2AC(ifp)->ac_netgraph != NULL) {
KASSERT(ng_ether_output_p != NULL,
("ng_ether_output_p is NULL"));
if ((error = (*ng_ether_output_p)(ifp, &m)) != 0) {
bad: if (m != NULL)
m_freem(m);
return (error);
}
if (m == NULL)
return (0);
}
/* Continue with link-layer output */
return ether_output_frame(ifp, m);
}
/*
* Ethernet link layer output routine to send a raw frame to the device.
*
* This assumes that the 14 byte Ethernet header is present and contiguous
* in the first mbuf (if BRIDGE'ing).
*/
int
ether_output_frame(struct ifnet *ifp, struct mbuf *m)
{
int i;
Remove (almost all) global variables that were used to hold packet forwarding state ("annotations") during ip processing. The code is considerably cleaner now. The variables removed by this change are: ip_divert_cookie used by divert sockets ip_fw_fwd_addr used for transparent ip redirection last_pkt used by dynamic pipes in dummynet Removal of the first two has been done by carrying the annotations into volatile structs prepended to the mbuf chains, and adding appropriate code to add/remove annotations in the routines which make use of them, i.e. ip_input(), ip_output(), tcp_input(), bdg_forward(), ether_demux(), ether_output_frame(), div_output(). On passing, remove a bug in divert handling of fragmented packet. Now it is the fragment at offset 0 which sets the divert status of the whole packet, whereas formerly it was the last incoming fragment to decide. Removal of last_pkt required a change in the interface of ip_fw_chk() and dummynet_io(). On passing, use the same mechanism for dummynet annotations and for divert/forward annotations. option IPFIREWALL_FORWARD is effectively useless, the code to implement it is very small and is now in by default to avoid the obfuscation of conditionally compiled code. NOTES: * there is at least one global variable left, sro_fwd, in ip_output(). I am not sure if/how this can be removed. * I have deliberately avoided gratuitous style changes in this commit to avoid cluttering the diffs. Minor stule cleanup will likely be necessary * this commit only focused on the IP layer. I am sure there is a number of global variables used in the TCP and maybe UDP stack. * despite the number of files touched, there are absolutely no API's or data structures changed by this commit (except the interfaces of ip_fw_chk() and dummynet_io(), which are internal anyways), so an MFC is quite safe and unintrusive (and desirable, given the improved readability of the code). MFC after: 10 days
2002-06-22 11:51:02 +00:00
if (PFIL_HOOKED(&V_link_pfil_hook)) {
i = pfil_run_hooks(&V_link_pfil_hook, &m, ifp, PFIL_OUT, NULL);
if (i != 0)
return (EACCES);
if (m == NULL)
return (0);
}
Remove (almost all) global variables that were used to hold packet forwarding state ("annotations") during ip processing. The code is considerably cleaner now. The variables removed by this change are: ip_divert_cookie used by divert sockets ip_fw_fwd_addr used for transparent ip redirection last_pkt used by dynamic pipes in dummynet Removal of the first two has been done by carrying the annotations into volatile structs prepended to the mbuf chains, and adding appropriate code to add/remove annotations in the routines which make use of them, i.e. ip_input(), ip_output(), tcp_input(), bdg_forward(), ether_demux(), ether_output_frame(), div_output(). On passing, remove a bug in divert handling of fragmented packet. Now it is the fragment at offset 0 which sets the divert status of the whole packet, whereas formerly it was the last incoming fragment to decide. Removal of last_pkt required a change in the interface of ip_fw_chk() and dummynet_io(). On passing, use the same mechanism for dummynet annotations and for divert/forward annotations. option IPFIREWALL_FORWARD is effectively useless, the code to implement it is very small and is now in by default to avoid the obfuscation of conditionally compiled code. NOTES: * there is at least one global variable left, sro_fwd, in ip_output(). I am not sure if/how this can be removed. * I have deliberately avoided gratuitous style changes in this commit to avoid cluttering the diffs. Minor stule cleanup will likely be necessary * this commit only focused on the IP layer. I am sure there is a number of global variables used in the TCP and maybe UDP stack. * despite the number of files touched, there are absolutely no API's or data structures changed by this commit (except the interfaces of ip_fw_chk() and dummynet_io(), which are internal anyways), so an MFC is quite safe and unintrusive (and desirable, given the improved readability of the code). MFC after: 10 days
2002-06-22 11:51:02 +00:00
1994-05-24 10:09:53 +00:00
/*
* Queue message on interface, update output statistics if
* successful, and start output if interface not yet active.
1994-05-24 10:09:53 +00:00
*/
return ((ifp->if_transmit)(ifp, m));
1994-05-24 10:09:53 +00:00
}
#if defined(INET) || defined(INET6)
#endif
1994-05-24 10:09:53 +00:00
/*
* Process a received Ethernet packet; the packet is in the
* mbuf chain m with the ethernet header at the front.
1994-05-24 10:09:53 +00:00
*/
static void
ether_input_internal(struct ifnet *ifp, struct mbuf *m)
1994-05-24 10:09:53 +00:00
{
struct ether_header *eh;
u_short etype;
1994-05-24 10:09:53 +00:00
if ((ifp->if_flags & IFF_UP) == 0) {
m_freem(m);
return;
}
#ifdef DIAGNOSTIC
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
if_printf(ifp, "discard frame at !IFF_DRV_RUNNING\n");
m_freem(m);
return;
}
#endif
/*
* Do consistency checks to verify assumptions
* made by code past this point.
*/
if ((m->m_flags & M_PKTHDR) == 0) {
if_printf(ifp, "discard frame w/o packet header\n");
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
m_freem(m);
return;
}
if (m->m_len < ETHER_HDR_LEN) {
/* XXX maybe should pullup? */
if_printf(ifp, "discard frame w/o leading ethernet "
"header (len %u pkt len %u)\n",
m->m_len, m->m_pkthdr.len);
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
m_freem(m);
return;
}
eh = mtod(m, struct ether_header *);
etype = ntohs(eh->ether_type);
if (m->m_pkthdr.rcvif == NULL) {
if_printf(ifp, "discard frame w/o interface pointer\n");
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
m_freem(m);
return;
}
#ifdef DIAGNOSTIC
if (m->m_pkthdr.rcvif != ifp) {
if_printf(ifp, "Warning, frame marked as received on %s\n",
m->m_pkthdr.rcvif->if_xname);
}
#endif
Change the curvnet variable from a global const struct vnet *, previously always pointing to the default vnet context, to a dynamically changing thread-local one. The currvnet context should be set on entry to networking code via CURVNET_SET() macros, and reverted to previous state via CURVNET_RESTORE(). Recursions on curvnet are permitted, though strongly discuouraged. This change should have no functional impact on nooptions VIMAGE kernel builds, where CURVNET_* macros expand to whitespace. The curthread->td_vnet (aka curvnet) variable's purpose is to be an indicator of the vnet context in which the current network-related operation takes place, in case we cannot deduce the current vnet context from any other source, such as by looking at mbuf's m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so far curvnet has turned out to be an invaluable consistency checking aid: it helps to catch cases when sockets, ifnets or any other vnet-aware structures may have leaked from one vnet to another. The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros was a result of an empirical iterative process, whith an aim to reduce recursions on CURVNET_SET() to a minimum, while still reducing the scope of CURVNET_SET() to networking only operations - the alternative would be calling CURVNET_SET() on each system call entry. In general, curvnet has to be set in three typicall cases: when processing socket-related requests from userspace or from within the kernel; when processing inbound traffic flowing from device drivers to upper layers of the networking stack, and when executing timer-driven networking functions. This change also introduces a DDB subcommand to show the list of all vnet instances. Approved by: julian (mentor)
2009-05-05 10:56:12 +00:00
CURVNET_SET_QUIET(ifp->if_vnet);
if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
if (ETHER_IS_BROADCAST(eh->ether_dhost))
m->m_flags |= M_BCAST;
else
m->m_flags |= M_MCAST;
if_inc_counter(ifp, IFCOUNTER_IMCASTS, 1);
}
#ifdef MAC
/*
* Tag the mbuf with an appropriate MAC label before any other
* consumers can get to it.
*/
mac_ifnet_create_mbuf(ifp, m);
#endif
/*
* Give bpf a chance at the packet.
*/
ETHER_BPF_MTAP(ifp, m);
/*
* If the CRC is still on the packet, trim it off. We do this once
* and once only in case we are re-entered. Nothing else on the
* Ethernet receive path expects to see the FCS.
*/
if (m->m_flags & M_HASFCS) {
m_adj(m, -ETHER_CRC_LEN);
m->m_flags &= ~M_HASFCS;
}
if (!(ifp->if_capenable & IFCAP_HWSTATS))
if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
/* Allow monitor mode to claim this frame, after stats are updated. */
if (ifp->if_flags & IFF_MONITOR) {
m_freem(m);
Change the curvnet variable from a global const struct vnet *, previously always pointing to the default vnet context, to a dynamically changing thread-local one. The currvnet context should be set on entry to networking code via CURVNET_SET() macros, and reverted to previous state via CURVNET_RESTORE(). Recursions on curvnet are permitted, though strongly discuouraged. This change should have no functional impact on nooptions VIMAGE kernel builds, where CURVNET_* macros expand to whitespace. The curthread->td_vnet (aka curvnet) variable's purpose is to be an indicator of the vnet context in which the current network-related operation takes place, in case we cannot deduce the current vnet context from any other source, such as by looking at mbuf's m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so far curvnet has turned out to be an invaluable consistency checking aid: it helps to catch cases when sockets, ifnets or any other vnet-aware structures may have leaked from one vnet to another. The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros was a result of an empirical iterative process, whith an aim to reduce recursions on CURVNET_SET() to a minimum, while still reducing the scope of CURVNET_SET() to networking only operations - the alternative would be calling CURVNET_SET() on each system call entry. In general, curvnet has to be set in three typicall cases: when processing socket-related requests from userspace or from within the kernel; when processing inbound traffic flowing from device drivers to upper layers of the networking stack, and when executing timer-driven networking functions. This change also introduces a DDB subcommand to show the list of all vnet instances. Approved by: julian (mentor)
2009-05-05 10:56:12 +00:00
CURVNET_RESTORE();
return;
}
/* Handle input from a lagg(4) port */
if (ifp->if_type == IFT_IEEE8023ADLAG) {
KASSERT(lagg_input_p != NULL,
("%s: if_lagg not loaded!", __func__));
m = (*lagg_input_p)(ifp, m);
if (m != NULL)
ifp = m->m_pkthdr.rcvif;
else {
CURVNET_RESTORE();
return;
}
}
/*
* If the hardware did not process an 802.1Q tag, do this now,
* to allow 802.1P priority frames to be passed to the main input
* path correctly.
* TODO: Deal with Q-in-Q frames, but not arbitrary nesting levels.
*/
if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_VLAN) {
struct ether_vlan_header *evl;
if (m->m_len < sizeof(*evl) &&
(m = m_pullup(m, sizeof(*evl))) == NULL) {
#ifdef DIAGNOSTIC
if_printf(ifp, "cannot pullup VLAN header\n");
#endif
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
m_freem(m);
CURVNET_RESTORE();
return;
}
evl = mtod(m, struct ether_vlan_header *);
m->m_pkthdr.ether_vtag = ntohs(evl->evl_tag);
m->m_flags |= M_VLANTAG;
bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN,
ETHER_HDR_LEN - ETHER_TYPE_LEN);
m_adj(m, ETHER_VLAN_ENCAP_LEN);
eh = mtod(m, struct ether_header *);
}
M_SETFIB(m, ifp->if_fib);
/* Allow ng_ether(4) to claim this frame. */
if (IFP2AC(ifp)->ac_netgraph != NULL) {
KASSERT(ng_ether_input_p != NULL,
("%s: ng_ether_input_p is NULL", __func__));
m->m_flags &= ~M_PROMISC;
(*ng_ether_input_p)(ifp, &m);
Change the curvnet variable from a global const struct vnet *, previously always pointing to the default vnet context, to a dynamically changing thread-local one. The currvnet context should be set on entry to networking code via CURVNET_SET() macros, and reverted to previous state via CURVNET_RESTORE(). Recursions on curvnet are permitted, though strongly discuouraged. This change should have no functional impact on nooptions VIMAGE kernel builds, where CURVNET_* macros expand to whitespace. The curthread->td_vnet (aka curvnet) variable's purpose is to be an indicator of the vnet context in which the current network-related operation takes place, in case we cannot deduce the current vnet context from any other source, such as by looking at mbuf's m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so far curvnet has turned out to be an invaluable consistency checking aid: it helps to catch cases when sockets, ifnets or any other vnet-aware structures may have leaked from one vnet to another. The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros was a result of an empirical iterative process, whith an aim to reduce recursions on CURVNET_SET() to a minimum, while still reducing the scope of CURVNET_SET() to networking only operations - the alternative would be calling CURVNET_SET() on each system call entry. In general, curvnet has to be set in three typicall cases: when processing socket-related requests from userspace or from within the kernel; when processing inbound traffic flowing from device drivers to upper layers of the networking stack, and when executing timer-driven networking functions. This change also introduces a DDB subcommand to show the list of all vnet instances. Approved by: julian (mentor)
2009-05-05 10:56:12 +00:00
if (m == NULL) {
CURVNET_RESTORE();
return;
Change the curvnet variable from a global const struct vnet *, previously always pointing to the default vnet context, to a dynamically changing thread-local one. The currvnet context should be set on entry to networking code via CURVNET_SET() macros, and reverted to previous state via CURVNET_RESTORE(). Recursions on curvnet are permitted, though strongly discuouraged. This change should have no functional impact on nooptions VIMAGE kernel builds, where CURVNET_* macros expand to whitespace. The curthread->td_vnet (aka curvnet) variable's purpose is to be an indicator of the vnet context in which the current network-related operation takes place, in case we cannot deduce the current vnet context from any other source, such as by looking at mbuf's m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so far curvnet has turned out to be an invaluable consistency checking aid: it helps to catch cases when sockets, ifnets or any other vnet-aware structures may have leaked from one vnet to another. The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros was a result of an empirical iterative process, whith an aim to reduce recursions on CURVNET_SET() to a minimum, while still reducing the scope of CURVNET_SET() to networking only operations - the alternative would be calling CURVNET_SET() on each system call entry. In general, curvnet has to be set in three typicall cases: when processing socket-related requests from userspace or from within the kernel; when processing inbound traffic flowing from device drivers to upper layers of the networking stack, and when executing timer-driven networking functions. This change also introduces a DDB subcommand to show the list of all vnet instances. Approved by: julian (mentor)
2009-05-05 10:56:12 +00:00
}
eh = mtod(m, struct ether_header *);
}
/*
* Allow if_bridge(4) to claim this frame.
* The BRIDGE_INPUT() macro will update ifp if the bridge changed it
* and the frame should be delivered locally.
*/
if (ifp->if_bridge != NULL) {
m->m_flags &= ~M_PROMISC;
BRIDGE_INPUT(ifp, m);
Change the curvnet variable from a global const struct vnet *, previously always pointing to the default vnet context, to a dynamically changing thread-local one. The currvnet context should be set on entry to networking code via CURVNET_SET() macros, and reverted to previous state via CURVNET_RESTORE(). Recursions on curvnet are permitted, though strongly discuouraged. This change should have no functional impact on nooptions VIMAGE kernel builds, where CURVNET_* macros expand to whitespace. The curthread->td_vnet (aka curvnet) variable's purpose is to be an indicator of the vnet context in which the current network-related operation takes place, in case we cannot deduce the current vnet context from any other source, such as by looking at mbuf's m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so far curvnet has turned out to be an invaluable consistency checking aid: it helps to catch cases when sockets, ifnets or any other vnet-aware structures may have leaked from one vnet to another. The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros was a result of an empirical iterative process, whith an aim to reduce recursions on CURVNET_SET() to a minimum, while still reducing the scope of CURVNET_SET() to networking only operations - the alternative would be calling CURVNET_SET() on each system call entry. In general, curvnet has to be set in three typicall cases: when processing socket-related requests from userspace or from within the kernel; when processing inbound traffic flowing from device drivers to upper layers of the networking stack, and when executing timer-driven networking functions. This change also introduces a DDB subcommand to show the list of all vnet instances. Approved by: julian (mentor)
2009-05-05 10:56:12 +00:00
if (m == NULL) {
CURVNET_RESTORE();
return;
Change the curvnet variable from a global const struct vnet *, previously always pointing to the default vnet context, to a dynamically changing thread-local one. The currvnet context should be set on entry to networking code via CURVNET_SET() macros, and reverted to previous state via CURVNET_RESTORE(). Recursions on curvnet are permitted, though strongly discuouraged. This change should have no functional impact on nooptions VIMAGE kernel builds, where CURVNET_* macros expand to whitespace. The curthread->td_vnet (aka curvnet) variable's purpose is to be an indicator of the vnet context in which the current network-related operation takes place, in case we cannot deduce the current vnet context from any other source, such as by looking at mbuf's m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so far curvnet has turned out to be an invaluable consistency checking aid: it helps to catch cases when sockets, ifnets or any other vnet-aware structures may have leaked from one vnet to another. The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros was a result of an empirical iterative process, whith an aim to reduce recursions on CURVNET_SET() to a minimum, while still reducing the scope of CURVNET_SET() to networking only operations - the alternative would be calling CURVNET_SET() on each system call entry. In general, curvnet has to be set in three typicall cases: when processing socket-related requests from userspace or from within the kernel; when processing inbound traffic flowing from device drivers to upper layers of the networking stack, and when executing timer-driven networking functions. This change also introduces a DDB subcommand to show the list of all vnet instances. Approved by: julian (mentor)
2009-05-05 10:56:12 +00:00
}
eh = mtod(m, struct ether_header *);
}
#if defined(INET) || defined(INET6)
/*
* Clear M_PROMISC on frame so that carp(4) will see it when the
* mbuf flows up to Layer 3.
* FreeBSD's implementation of carp(4) uses the inprotosw
* to dispatch IPPROTO_CARP. carp(4) also allocates its own
* Ethernet addresses of the form 00:00:5e:00:01:xx, which
* is outside the scope of the M_PROMISC test below.
* TODO: Maintain a hash table of ethernet addresses other than
* ether_dhost which may be active on this ifp.
*/
if (ifp->if_carp && (*carp_forus_p)(ifp, eh->ether_dhost)) {
m->m_flags &= ~M_PROMISC;
} else
#endif
{
/*
* If the frame received was not for our MAC address, set the
* M_PROMISC flag on the mbuf chain. The frame may need to
* be seen by the rest of the Ethernet input path in case of
* re-entry (e.g. bridge, vlan, netgraph) but should not be
* seen by upper protocol layers.
*/
if (!ETHER_IS_MULTICAST(eh->ether_dhost) &&
bcmp(IF_LLADDR(ifp), eh->ether_dhost, ETHER_ADDR_LEN) != 0)
m->m_flags |= M_PROMISC;
}
This is the much-discussed major upgrade to the random(4) device, known to you all as /dev/random. This code has had an extensive rewrite and a good series of reviews, both by the author and other parties. This means a lot of code has been simplified. Pluggable structures for high-rate entropy generators are available, and it is most definitely not the case that /dev/random can be driven by only a hardware souce any more. This has been designed out of the device. Hardware sources are stirred into the CSPRNG (Yarrow, Fortuna) like any other entropy source. Pluggable modules may be written by third parties for additional sources. The harvesting structures and consequently the locking have been simplified. Entropy harvesting is done in a more general way (the documentation for this will follow). There is some GREAT entropy to be had in the UMA allocator, but it is disabled for now as messing with that is likely to annoy many people. The venerable (but effective) Yarrow algorithm, which is no longer supported by its authors now has an alternative, Fortuna. For now, Yarrow is retained as the default algorithm, but this may be changed using a kernel option. It is intended to make Fortuna the default algorithm for 11.0. Interested parties are encouraged to read ISBN 978-0-470-47424-2 "Cryptography Engineering" By Ferguson, Schneier and Kohno for Fortuna's gory details. Heck, read it anyway. Many thanks to Arthur Mesh who did early grunt work, and who got caught in the crossfire rather more than he deserved to. My thanks also to folks who helped me thresh this out on whiteboards and in the odd "Hallway track", or otherwise. My Nomex pants are on. Let the feedback commence! Reviewed by: trasz,des(partial),imp(partial?),rwatson(partial?) Approved by: so(des)
2014-10-30 21:21:53 +00:00
random_harvest(&(m->m_data), 12, 2, RANDOM_NET_ETHER);
ether_demux(ifp, m);
Change the curvnet variable from a global const struct vnet *, previously always pointing to the default vnet context, to a dynamically changing thread-local one. The currvnet context should be set on entry to networking code via CURVNET_SET() macros, and reverted to previous state via CURVNET_RESTORE(). Recursions on curvnet are permitted, though strongly discuouraged. This change should have no functional impact on nooptions VIMAGE kernel builds, where CURVNET_* macros expand to whitespace. The curthread->td_vnet (aka curvnet) variable's purpose is to be an indicator of the vnet context in which the current network-related operation takes place, in case we cannot deduce the current vnet context from any other source, such as by looking at mbuf's m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so far curvnet has turned out to be an invaluable consistency checking aid: it helps to catch cases when sockets, ifnets or any other vnet-aware structures may have leaked from one vnet to another. The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros was a result of an empirical iterative process, whith an aim to reduce recursions on CURVNET_SET() to a minimum, while still reducing the scope of CURVNET_SET() to networking only operations - the alternative would be calling CURVNET_SET() on each system call entry. In general, curvnet has to be set in three typicall cases: when processing socket-related requests from userspace or from within the kernel; when processing inbound traffic flowing from device drivers to upper layers of the networking stack, and when executing timer-driven networking functions. This change also introduces a DDB subcommand to show the list of all vnet instances. Approved by: julian (mentor)
2009-05-05 10:56:12 +00:00
CURVNET_RESTORE();
}
/*
* Ethernet input dispatch; by default, direct dispatch here regardless of
Several years after initial development, merge prototype support for linking NIC Receive Side Scaling (RSS) to the network stack's connection-group implementation. This prototype (and derived patches) are in use at Juniper and several other FreeBSD-using companies, so despite some reservations about its maturity, merge the patch to the base tree so that it can be iteratively refined in collaboration rather than maintained as a set of gradually diverging patch sets. (1) Merge a software implementation of the Toeplitz hash specified in RSS implemented by David Malone. This is used to allow suitable pcbgroup placement of connections before the first packet is received from the NIC. Software hashing is generally avoided, however, due to high cost of the hash on general-purpose CPUs. (2) In in_rss.c, maintain authoritative versions of RSS state intended to be pushed to each NIC, including keying material, hash algorithm/ configuration, and buckets. Provide software-facing interfaces to hash 2- and 4-tuples for IPv4 and IPv6 using both the RSS standardised Toeplitz and a 'naive' variation with a hash efficient in software but with poor distribution properties. Implement rss_m2cpuid()to be used by netisr and other load balancing code to look up the CPU on which an mbuf should be processed. (3) In the Ethernet link layer, allow netisr distribution using RSS as a source of policy as an alternative to source ordering; continue to default to direct dispatch (i.e., don't try and requeue packets for processing on the 'right' CPU if they arrive in a directly dispatchable context). (4) Allow RSS to control tuning of connection groups in order to align groups with RSS buckets. If a packet arrives on a protocol using connection groups, and contains a suitable hardware-generated hash, use that hash value to select the connection group for pcb lookup for both IPv4 and IPv6. If no hardware-generated Toeplitz hash is available, we fall back on regular PCB lookup risking contention rather than pay the cost of Toeplitz in software -- this is a less scalable but, at my last measurement, faster approach. As core counts go up, we may want to revise this strategy despite CPU overhead. Where device drivers suitably configure NICs, and connection groups / RSS are enabled, this should avoid both lock and line contention during connection lookup for TCP. This commit does not modify any device drivers to tune device RSS configuration to the global RSS configuration; patches are in circulation to do this for at least Chelsio T3 and Intel 1G/10G drivers. Currently, the KPI for device drivers is not particularly robust, nor aware of more advanced features such as runtime reconfiguration/rebalancing. This will hopefully prove a useful starting point for refinement. No MFC is scheduled as we will first want to nail down a more mature and maintainable KPI/KBI for device drivers. Sponsored by: Juniper Networks (original work) Sponsored by: EMC/Isilon (patch update and merge)
2014-03-15 00:57:50 +00:00
* global configuration. However, if RSS is enabled, hook up RSS affinity
* so that when deferred or hybrid dispatch is enabled, we can redistribute
* load based on RSS.
*
* XXXRW: Would be nice if the ifnet passed up a flag indicating whether or
* not it had already done work distribution via multi-queue. Then we could
* direct dispatch in the event load balancing was already complete and
* handle the case of interfaces with different capabilities better.
*
* XXXRW: Sort of want an M_DISTRIBUTED flag to avoid multiple distributions
* at multiple layers?
*
* XXXRW: For now, enable all this only if RSS is compiled in, although it
* works fine without RSS. Need to characterise the performance overhead
* of the detour through the netisr code in the event the result is always
* direct dispatch.
*/
static void
ether_nh_input(struct mbuf *m)
{
ether_input_internal(m->m_pkthdr.rcvif, m);
}
static struct netisr_handler ether_nh = {
.nh_name = "ether",
.nh_handler = ether_nh_input,
.nh_proto = NETISR_ETHER,
Several years after initial development, merge prototype support for linking NIC Receive Side Scaling (RSS) to the network stack's connection-group implementation. This prototype (and derived patches) are in use at Juniper and several other FreeBSD-using companies, so despite some reservations about its maturity, merge the patch to the base tree so that it can be iteratively refined in collaboration rather than maintained as a set of gradually diverging patch sets. (1) Merge a software implementation of the Toeplitz hash specified in RSS implemented by David Malone. This is used to allow suitable pcbgroup placement of connections before the first packet is received from the NIC. Software hashing is generally avoided, however, due to high cost of the hash on general-purpose CPUs. (2) In in_rss.c, maintain authoritative versions of RSS state intended to be pushed to each NIC, including keying material, hash algorithm/ configuration, and buckets. Provide software-facing interfaces to hash 2- and 4-tuples for IPv4 and IPv6 using both the RSS standardised Toeplitz and a 'naive' variation with a hash efficient in software but with poor distribution properties. Implement rss_m2cpuid()to be used by netisr and other load balancing code to look up the CPU on which an mbuf should be processed. (3) In the Ethernet link layer, allow netisr distribution using RSS as a source of policy as an alternative to source ordering; continue to default to direct dispatch (i.e., don't try and requeue packets for processing on the 'right' CPU if they arrive in a directly dispatchable context). (4) Allow RSS to control tuning of connection groups in order to align groups with RSS buckets. If a packet arrives on a protocol using connection groups, and contains a suitable hardware-generated hash, use that hash value to select the connection group for pcb lookup for both IPv4 and IPv6. If no hardware-generated Toeplitz hash is available, we fall back on regular PCB lookup risking contention rather than pay the cost of Toeplitz in software -- this is a less scalable but, at my last measurement, faster approach. As core counts go up, we may want to revise this strategy despite CPU overhead. Where device drivers suitably configure NICs, and connection groups / RSS are enabled, this should avoid both lock and line contention during connection lookup for TCP. This commit does not modify any device drivers to tune device RSS configuration to the global RSS configuration; patches are in circulation to do this for at least Chelsio T3 and Intel 1G/10G drivers. Currently, the KPI for device drivers is not particularly robust, nor aware of more advanced features such as runtime reconfiguration/rebalancing. This will hopefully prove a useful starting point for refinement. No MFC is scheduled as we will first want to nail down a more mature and maintainable KPI/KBI for device drivers. Sponsored by: Juniper Networks (original work) Sponsored by: EMC/Isilon (patch update and merge)
2014-03-15 00:57:50 +00:00
#ifdef RSS
.nh_policy = NETISR_POLICY_CPU,
.nh_dispatch = NETISR_DISPATCH_DIRECT,
.nh_m2cpuid = rss_m2cpuid,
#else
.nh_policy = NETISR_POLICY_SOURCE,
.nh_dispatch = NETISR_DISPATCH_DIRECT,
Several years after initial development, merge prototype support for linking NIC Receive Side Scaling (RSS) to the network stack's connection-group implementation. This prototype (and derived patches) are in use at Juniper and several other FreeBSD-using companies, so despite some reservations about its maturity, merge the patch to the base tree so that it can be iteratively refined in collaboration rather than maintained as a set of gradually diverging patch sets. (1) Merge a software implementation of the Toeplitz hash specified in RSS implemented by David Malone. This is used to allow suitable pcbgroup placement of connections before the first packet is received from the NIC. Software hashing is generally avoided, however, due to high cost of the hash on general-purpose CPUs. (2) In in_rss.c, maintain authoritative versions of RSS state intended to be pushed to each NIC, including keying material, hash algorithm/ configuration, and buckets. Provide software-facing interfaces to hash 2- and 4-tuples for IPv4 and IPv6 using both the RSS standardised Toeplitz and a 'naive' variation with a hash efficient in software but with poor distribution properties. Implement rss_m2cpuid()to be used by netisr and other load balancing code to look up the CPU on which an mbuf should be processed. (3) In the Ethernet link layer, allow netisr distribution using RSS as a source of policy as an alternative to source ordering; continue to default to direct dispatch (i.e., don't try and requeue packets for processing on the 'right' CPU if they arrive in a directly dispatchable context). (4) Allow RSS to control tuning of connection groups in order to align groups with RSS buckets. If a packet arrives on a protocol using connection groups, and contains a suitable hardware-generated hash, use that hash value to select the connection group for pcb lookup for both IPv4 and IPv6. If no hardware-generated Toeplitz hash is available, we fall back on regular PCB lookup risking contention rather than pay the cost of Toeplitz in software -- this is a less scalable but, at my last measurement, faster approach. As core counts go up, we may want to revise this strategy despite CPU overhead. Where device drivers suitably configure NICs, and connection groups / RSS are enabled, this should avoid both lock and line contention during connection lookup for TCP. This commit does not modify any device drivers to tune device RSS configuration to the global RSS configuration; patches are in circulation to do this for at least Chelsio T3 and Intel 1G/10G drivers. Currently, the KPI for device drivers is not particularly robust, nor aware of more advanced features such as runtime reconfiguration/rebalancing. This will hopefully prove a useful starting point for refinement. No MFC is scheduled as we will first want to nail down a more mature and maintainable KPI/KBI for device drivers. Sponsored by: Juniper Networks (original work) Sponsored by: EMC/Isilon (patch update and merge)
2014-03-15 00:57:50 +00:00
#endif
};
static void
ether_init(__unused void *arg)
{
netisr_register(&ether_nh);
}
SYSINIT(ether, SI_SUB_INIT_IF, SI_ORDER_ANY, ether_init, NULL);
static void
vnet_ether_init(__unused void *arg)
{
int i;
/* Initialize packet filter hooks. */
V_link_pfil_hook.ph_type = PFIL_TYPE_AF;
V_link_pfil_hook.ph_af = AF_LINK;
if ((i = pfil_head_register(&V_link_pfil_hook)) != 0)
printf("%s: WARNING: unable to register pfil link hook, "
"error %d\n", __func__, i);
}
VNET_SYSINIT(vnet_ether_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
vnet_ether_init, NULL);
static void
vnet_ether_destroy(__unused void *arg)
{
int i;
if ((i = pfil_head_unregister(&V_link_pfil_hook)) != 0)
printf("%s: WARNING: unable to unregister pfil link hook, "
"error %d\n", __func__, i);
}
VNET_SYSUNINIT(vnet_ether_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
vnet_ether_destroy, NULL);
static void
ether_input(struct ifnet *ifp, struct mbuf *m)
{
struct mbuf *mn;
/*
* The drivers are allowed to pass in a chain of packets linked with
* m_nextpkt. We split them up into separate packets here and pass
* them up. This allows the drivers to amortize the receive lock.
*/
while (m) {
mn = m->m_nextpkt;
m->m_nextpkt = NULL;
/*
* We will rely on rcvif being set properly in the deferred context,
* so assert it is correct here.
*/
KASSERT(m->m_pkthdr.rcvif == ifp, ("%s: ifnet mismatch", __func__));
netisr_dispatch(NETISR_ETHER, m);
m = mn;
}
}
/*
* Upper layer processing for a received Ethernet packet.
*/
void
ether_demux(struct ifnet *ifp, struct mbuf *m)
{
struct ether_header *eh;
int i, isr;
u_short ether_type;
KASSERT(ifp != NULL, ("%s: NULL interface pointer", __func__));
/* Do not grab PROMISC frames in case we are re-entered. */
if (PFIL_HOOKED(&V_link_pfil_hook) && !(m->m_flags & M_PROMISC)) {
i = pfil_run_hooks(&V_link_pfil_hook, &m, ifp, PFIL_IN, NULL);
if (i != 0 || m == NULL)
return;
}
eh = mtod(m, struct ether_header *);
ether_type = ntohs(eh->ether_type);
/*
* If this frame has a VLAN tag other than 0, call vlan_input()
* if its module is loaded. Otherwise, drop.
*/
if ((m->m_flags & M_VLANTAG) &&
EVL_VLANOFTAG(m->m_pkthdr.ether_vtag) != 0) {
2006-01-30 13:45:15 +00:00
if (ifp->if_vlantrunk == NULL) {
if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1);
m_freem(m);
return;
}
KASSERT(vlan_input_p != NULL,("%s: VLAN not loaded!",
__func__));
/* Clear before possibly re-entering ether_input(). */
m->m_flags &= ~M_PROMISC;
(*vlan_input_p)(ifp, m);
return;
}
/*
* Pass promiscuously received frames to the upper layer if the user
* requested this by setting IFF_PPROMISC. Otherwise, drop them.
*/
if ((ifp->if_flags & IFF_PPROMISC) == 0 && (m->m_flags & M_PROMISC)) {
m_freem(m);
return;
}
/*
* Reset layer specific mbuf flags to avoid confusing upper layers.
* Strip off Ethernet header.
*/
m->m_flags &= ~M_VLANTAG;
m_clrprotoflags(m);
m_adj(m, ETHER_HDR_LEN);
/*
* Dispatch frame to upper layer.
*/
switch (ether_type) {
1994-05-24 10:09:53 +00:00
#ifdef INET
case ETHERTYPE_IP:
if ((m = ip_fastforward(m)) == NULL)
return;
isr = NETISR_IP;
1994-05-24 10:09:53 +00:00
break;
case ETHERTYPE_ARP:
if (ifp->if_flags & IFF_NOARP) {
/* Discard packet if ARP is disabled on interface */
m_freem(m);
return;
}
isr = NETISR_ARP;
1994-05-24 10:09:53 +00:00
break;
#endif
#ifdef INET6
case ETHERTYPE_IPV6:
isr = NETISR_IPV6;
break;
#endif
1994-05-24 10:09:53 +00:00
default:
goto discard;
1994-05-24 10:09:53 +00:00
}
netisr_dispatch(isr, m);
return;
discard:
/*
* Packet is to be discarded. If netgraph is present,
* hand the packet to it for last chance processing;
* otherwise dispose of it.
*/
if (IFP2AC(ifp)->ac_netgraph != NULL) {
KASSERT(ng_ether_input_orphan_p != NULL,
("ng_ether_input_orphan_p is NULL"));
/*
* Put back the ethernet header so netgraph has a
* consistent view of inbound packets.
*/
M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT);
(*ng_ether_input_orphan_p)(ifp, m);
return;
}
m_freem(m);
}
/*
* Convert Ethernet address to printable (loggable) representation.
* This routine is for compatibility; it's better to just use
*
* printf("%6D", <pointer to address>, ":");
*
* since there's no static buffer involved.
*/
char *
ether_sprintf(const u_char *ap)
{
static char etherbuf[18];
snprintf(etherbuf, sizeof (etherbuf), "%6D", ap, ":");
return (etherbuf);
1994-05-24 10:09:53 +00:00
}
/*
* Perform common duties while attaching to interface list
*/
void
ether_ifattach(struct ifnet *ifp, const u_int8_t *lla)
1994-05-24 10:09:53 +00:00
{
int i;
2003-03-03 00:21:52 +00:00
struct ifaddr *ifa;
struct sockaddr_dl *sdl;
1994-05-24 10:09:53 +00:00
ifp->if_addrlen = ETHER_ADDR_LEN;
ifp->if_hdrlen = ETHER_HDR_LEN;
if_attach(ifp);
1994-05-24 10:09:53 +00:00
ifp->if_mtu = ETHERMTU;
ifp->if_output = ether_output;
ifp->if_input = ether_input;
ifp->if_resolvemulti = ether_resolvemulti;
#ifdef VIMAGE
ifp->if_reassign = ether_reassign;
#endif
if (ifp->if_baudrate == 0)
ifp->if_baudrate = IF_Mbps(10); /* just a default */
ifp->if_broadcastaddr = etherbroadcastaddr;
ifa = ifp->if_addr;
KASSERT(ifa != NULL, ("%s: no lladdr!\n", __func__));
sdl = (struct sockaddr_dl *)ifa->ifa_addr;
sdl->sdl_type = IFT_ETHER;
sdl->sdl_alen = ifp->if_addrlen;
bcopy(lla, LLADDR(sdl), ifp->if_addrlen);
bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN);
if (ng_ether_attach_p != NULL)
(*ng_ether_attach_p)(ifp);
/* Announce Ethernet MAC address if non-zero. */
for (i = 0; i < ifp->if_addrlen; i++)
if (lla[i] != 0)
break;
if (i != ifp->if_addrlen)
if_printf(ifp, "Ethernet address: %6D\n", lla, ":");
uuid_ether_add(LLADDR(sdl));
}
/*
* Perform common duties while detaching an Ethernet interface
*/
void
ether_ifdetach(struct ifnet *ifp)
{
struct sockaddr_dl *sdl;
sdl = (struct sockaddr_dl *)(ifp->if_addr->ifa_addr);
uuid_ether_del(LLADDR(sdl));
if (IFP2AC(ifp)->ac_netgraph != NULL) {
KASSERT(ng_ether_detach_p != NULL,
("ng_ether_detach_p is NULL"));
(*ng_ether_detach_p)(ifp);
}
bpfdetach(ifp);
if_detach(ifp);
}
#ifdef VIMAGE
void
ether_reassign(struct ifnet *ifp, struct vnet *new_vnet, char *unused __unused)
{
if (IFP2AC(ifp)->ac_netgraph != NULL) {
KASSERT(ng_ether_detach_p != NULL,
("ng_ether_detach_p is NULL"));
(*ng_ether_detach_p)(ifp);
}
if (ng_ether_attach_p != NULL) {
CURVNET_SET_QUIET(new_vnet);
(*ng_ether_attach_p)(ifp);
CURVNET_RESTORE();
}
}
#endif
SYSCTL_DECL(_net_link);
SYSCTL_NODE(_net_link, IFT_ETHER, ether, CTLFLAG_RW, 0, "Ethernet");
#if 0
/*
* This is for reference. We have a table-driven version
* of the little-endian crc32 generator, which is faster
* than the double-loop.
*/
uint32_t
ether_crc32_le(const uint8_t *buf, size_t len)
{
size_t i;
uint32_t crc;
int bit;
uint8_t data;
crc = 0xffffffff; /* initial value */
for (i = 0; i < len; i++) {
for (data = *buf++, bit = 0; bit < 8; bit++, data >>= 1) {
carry = (crc ^ data) & 1;
crc >>= 1;
if (carry)
crc = (crc ^ ETHER_CRC_POLY_LE);
}
}
return (crc);
}
#else
uint32_t
ether_crc32_le(const uint8_t *buf, size_t len)
{
static const uint32_t crctab[] = {
0x00000000, 0x1db71064, 0x3b6e20c8, 0x26d930ac,
0x76dc4190, 0x6b6b51f4, 0x4db26158, 0x5005713c,
0xedb88320, 0xf00f9344, 0xd6d6a3e8, 0xcb61b38c,
0x9b64c2b0, 0x86d3d2d4, 0xa00ae278, 0xbdbdf21c
};
size_t i;
uint32_t crc;
crc = 0xffffffff; /* initial value */
for (i = 0; i < len; i++) {
crc ^= buf[i];
crc = (crc >> 4) ^ crctab[crc & 0xf];
crc = (crc >> 4) ^ crctab[crc & 0xf];
}
return (crc);
}
#endif
uint32_t
ether_crc32_be(const uint8_t *buf, size_t len)
{
size_t i;
uint32_t crc, carry;
int bit;
uint8_t data;
crc = 0xffffffff; /* initial value */
for (i = 0; i < len; i++) {
for (data = *buf++, bit = 0; bit < 8; bit++, data >>= 1) {
carry = ((crc & 0x80000000) ? 1 : 0) ^ (data & 0x01);
crc <<= 1;
if (carry)
crc = (crc ^ ETHER_CRC_POLY_BE) | carry;
}
}
return (crc);
}
int
ether_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
{
struct ifaddr *ifa = (struct ifaddr *) data;
struct ifreq *ifr = (struct ifreq *) data;
int error = 0;
switch (command) {
case SIOCSIFADDR:
ifp->if_flags |= IFF_UP;
switch (ifa->ifa_addr->sa_family) {
#ifdef INET
case AF_INET:
ifp->if_init(ifp->if_softc); /* before arpwhohas */
arp_ifinit(ifp, ifa);
break;
#endif
default:
ifp->if_init(ifp->if_softc);
break;
}
break;
case SIOCGIFADDR:
{
struct sockaddr *sa;
sa = (struct sockaddr *) & ifr->ifr_data;
bcopy(IF_LLADDR(ifp),
(caddr_t) sa->sa_data, ETHER_ADDR_LEN);
}
break;
case SIOCSIFMTU:
/*
* Set the interface MTU.
*/
if (ifr->ifr_mtu > ETHERMTU) {
error = EINVAL;
} else {
ifp->if_mtu = ifr->ifr_mtu;
}
break;
default:
error = EINVAL; /* XXX netbsd has ENOTTY??? */
break;
}
return (error);
}
static int
ether_resolvemulti(struct ifnet *ifp, struct sockaddr **llsa,
struct sockaddr *sa)
{
struct sockaddr_dl *sdl;
#ifdef INET
struct sockaddr_in *sin;
#endif
#ifdef INET6
struct sockaddr_in6 *sin6;
#endif
u_char *e_addr;
switch(sa->sa_family) {
case AF_LINK:
/*
* No mapping needed. Just check that it's a valid MC address.
*/
sdl = (struct sockaddr_dl *)sa;
e_addr = LLADDR(sdl);
if (!ETHER_IS_MULTICAST(e_addr))
return EADDRNOTAVAIL;
*llsa = 0;
return 0;
#ifdef INET
case AF_INET:
sin = (struct sockaddr_in *)sa;
if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
return EADDRNOTAVAIL;
sdl = link_init_sdl(ifp, *llsa, IFT_ETHER);
sdl->sdl_alen = ETHER_ADDR_LEN;
e_addr = LLADDR(sdl);
ETHER_MAP_IP_MULTICAST(&sin->sin_addr, e_addr);
*llsa = (struct sockaddr *)sdl;
return 0;
#endif
#ifdef INET6
case AF_INET6:
sin6 = (struct sockaddr_in6 *)sa;
if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
/*
* An IP6 address of 0 means listen to all
* of the Ethernet multicast address used for IP6.
* (This is used for multicast routers.)
*/
ifp->if_flags |= IFF_ALLMULTI;
*llsa = 0;
return 0;
}
if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
return EADDRNOTAVAIL;
sdl = link_init_sdl(ifp, *llsa, IFT_ETHER);
sdl->sdl_alen = ETHER_ADDR_LEN;
e_addr = LLADDR(sdl);
ETHER_MAP_IPV6_MULTICAST(&sin6->sin6_addr, e_addr);
*llsa = (struct sockaddr *)sdl;
return 0;
#endif
default:
/*
* Well, the text isn't quite right, but it's the name
* that counts...
*/
return EAFNOSUPPORT;
}
}
2003-03-15 15:38:02 +00:00
static void*
ether_alloc(u_char type, struct ifnet *ifp)
{
struct arpcom *ac;
ac = malloc(sizeof(struct arpcom), M_ARPCOM, M_WAITOK | M_ZERO);
ac->ac_ifp = ifp;
return (ac);
}
static void
ether_free(void *com, u_char type)
{
free(com, M_ARPCOM);
}
static int
ether_modevent(module_t mod, int type, void *data)
{
switch (type) {
case MOD_LOAD:
if_register_com_alloc(IFT_ETHER, ether_alloc, ether_free);
break;
case MOD_UNLOAD:
if_deregister_com_alloc(IFT_ETHER);
break;
default:
return EOPNOTSUPP;
}
return (0);
}
2003-03-15 15:38:02 +00:00
static moduledata_t ether_mod = {
"ether",
ether_modevent,
0
2003-03-15 15:38:02 +00:00
};
void
ether_vlan_mtap(struct bpf_if *bp, struct mbuf *m, void *data, u_int dlen)
{
struct ether_vlan_header vlan;
struct mbuf mv, mb;
KASSERT((m->m_flags & M_VLANTAG) != 0,
("%s: vlan information not present", __func__));
KASSERT(m->m_len >= sizeof(struct ether_header),
("%s: mbuf not large enough for header", __func__));
bcopy(mtod(m, char *), &vlan, sizeof(struct ether_header));
vlan.evl_proto = vlan.evl_encap_proto;
vlan.evl_encap_proto = htons(ETHERTYPE_VLAN);
vlan.evl_tag = htons(m->m_pkthdr.ether_vtag);
m->m_len -= sizeof(struct ether_header);
m->m_data += sizeof(struct ether_header);
/*
* If a data link has been supplied by the caller, then we will need to
* re-create a stack allocated mbuf chain with the following structure:
*
* (1) mbuf #1 will contain the supplied data link
* (2) mbuf #2 will contain the vlan header
* (3) mbuf #3 will contain the original mbuf's packet data
*
* Otherwise, submit the packet and vlan header via bpf_mtap2().
*/
if (data != NULL) {
mv.m_next = m;
mv.m_data = (caddr_t)&vlan;
mv.m_len = sizeof(vlan);
mb.m_next = &mv;
mb.m_data = data;
mb.m_len = dlen;
bpf_mtap(bp, &mb);
} else
bpf_mtap2(bp, &vlan, sizeof(vlan), m);
m->m_len += sizeof(struct ether_header);
m->m_data -= sizeof(struct ether_header);
}
struct mbuf *
ether_vlanencap(struct mbuf *m, uint16_t tag)
{
struct ether_vlan_header *evl;
M_PREPEND(m, ETHER_VLAN_ENCAP_LEN, M_NOWAIT);
if (m == NULL)
return (NULL);
/* M_PREPEND takes care of m_len, m_pkthdr.len for us */
if (m->m_len < sizeof(*evl)) {
m = m_pullup(m, sizeof(*evl));
if (m == NULL)
return (NULL);
}
/*
* Transform the Ethernet header into an Ethernet header
* with 802.1Q encapsulation.
*/
evl = mtod(m, struct ether_vlan_header *);
bcopy((char *)evl + ETHER_VLAN_ENCAP_LEN,
(char *)evl, ETHER_HDR_LEN - ETHER_TYPE_LEN);
evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
evl->evl_tag = htons(tag);
return (m);
}
DECLARE_MODULE(ether, ether_mod, SI_SUB_INIT_IF, SI_ORDER_ANY);
2003-03-15 15:38:02 +00:00
MODULE_VERSION(ether, 1);