d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
c5390e5ab4
freebsd-nq/sys/net/flowtable.c

1185 lines
28 KiB
C
Raw Normal View History

Bring copyright notice to standard style.
2014-02-17 12:01:50 +00:00
/*-
Add my copyright to flowtable.
2014-02-17 12:07:17 +00:00
* Copyright (c) 2014 Gleb Smirnoff <glebius@FreeBSD.org>
Bring copyright notice to standard style.
2014-02-17 12:01:50 +00:00
* Copyright (c) 2008-2010, BitGravity Inc.
* 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. Neither the name of the BitGravity Corporation 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 COPYRIGHT HOLDERS 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 COPYRIGHT OWNER 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.
*/
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include "opt_route.h"
#include "opt_mpath.h"
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
#include "opt_ddb.h"
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
#include "opt_inet.h"
#include "opt_inet6.h"
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Spacing.
2014-02-07 10:05:12 +00:00
#include <sys/param.h>
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include <sys/types.h>
#include <sys/bitstring.h>
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
#include <sys/condvar.h>
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include <sys/callout.h>
- Move jenkins.h to jenkins_hash.c - Provide missing function that can do hashing of arbitrary sized buffer. - Refetch lookup3.c and do only minimal edits to it, so that diff between our jenkins_hash.c and lookup3.c is minimal. - Add declarations for jenkins_hash(), jenkins_hash32() to sys/hash.h. - Document these functions in hash(9) Obtained from: http://burtleburtle.net/bob/c/lookup3.c
2012-09-04 12:07:33 +00:00
#include <sys/hash.h>
Spacing.
2014-02-07 10:05:12 +00:00
#include <sys/kernel.h>
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include <sys/kthread.h>
#include <sys/limits.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
#include <sys/pcpu.h>
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include <sys/proc.h>
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
#include <sys/queue.h>
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
#include <sys/sbuf.h>
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include <sys/sched.h>
#include <sys/smp.h>
#include <sys/socket.h>
#include <sys/syslog.h>
#include <sys/sysctl.h>
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
#include <vm/uma.h>
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include <net/if.h>
#include <net/if_llatbl.h>
#include <net/if_var.h>
Spacing.
2014-02-07 10:05:12 +00:00
#include <net/route.h>
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include <net/flowtable.h>
Introduce and use a sysinit-based initialization scheme for virtual network stacks, VNET_SYSINIT: - Add VNET_SYSINIT and VNET_SYSUNINIT macros to declare events that will occur each time a network stack is instantiated and destroyed. In the !VIMAGE case, these are simply mapped into regular SYSINIT/SYSUNINIT. For the VIMAGE case, we instead use SYSINIT's to track their order and properties on registration, using them for each vnet when created/ destroyed, or immediately on module load for already-started vnets. - Remove vnet_modinfo mechanism that existed to serve this purpose previously, as well as its dependency scheme: we now just use the SYSINIT ordering scheme. - Implement VNET_DOMAIN_SET() to allow protocol domains to declare that they want init functions to be called for each virtual network stack rather than just once at boot, compiling down to DOMAIN_SET() in the non-VIMAGE case. - Walk all virtualized kernel subsystems and make use of these instead of modinfo or DOMAIN_SET() for init/uninit events. In some cases, convert modular components from using modevent to using sysinit (where appropriate). In some cases, do minor rejuggling of SYSINIT ordering to make room for or better manage events. Portions submitted by: jhb (VNET_SYSINIT), bz (cleanup) Discussed with: jhb, bz, julian, zec Reviewed by: bz Approved by: re (VIMAGE blanket)
2009-07-23 20:46:49 +00:00
#include <net/vnet.h>
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/in_var.h>
#include <netinet/if_ether.h>
#include <netinet/ip.h>
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
#ifdef INET6
#include <netinet/ip6.h>
#endif
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef FLOWTABLE_HASH_ALL
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
#include <netinet/tcp.h>
#include <netinet/udp.h>
#include <netinet/sctp.h>
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#endif
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
#include <ddb/ddb.h>
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef FLOWTABLE_HASH_ALL
#define KEY_PORTS (sizeof(uint16_t) * 2)
#define KEY_ADDRS 2
#else
#define KEY_PORTS 0
#define KEY_ADDRS 1
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
#endif
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef INET6
#define KEY_ADDR_LEN sizeof(struct in6_addr)
#else
#define KEY_ADDR_LEN sizeof(struct in_addr)
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
#endif
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#define KEYLEN ((KEY_ADDR_LEN * KEY_ADDRS + KEY_PORTS) / sizeof(uint32_t))
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
struct flentry {
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
uint32_t f_hash; /* hash flowing forward */
uint32_t f_key[KEYLEN]; /* address(es and ports) */
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
uint32_t f_uptime; /* uptime at last access */
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
uint16_t f_fibnum; /* fib index */
#ifdef FLOWTABLE_HASH_ALL
uint8_t f_proto; /* protocol */
uint8_t f_flags; /* stale? */
#define FL_STALE 1
#endif
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
SLIST_ENTRY(flentry) f_next; /* pointer to collision entry */
struct rtentry *f_rt; /* rtentry for flow */
struct llentry *f_lle; /* llentry for flow */
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
};
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#undef KEYLEN
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
SLIST_HEAD(flist, flentry);
/* Make sure we can use pcpu_zone_ptr for struct flist. */
CTASSERT(sizeof(struct flist) == sizeof(void *));
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
struct flowtable {
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
counter_u64_t *ft_stat;
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
int ft_size;
- boot-time size the ipv4 flowtable and the maximum number of flows - increase flow cleaning frequency and decrease flow caching time when near the flow limit - stop allocating new flows when within 3% of maxflows don't start allocating again until below 12.5% MFC after: 7 days
2010-03-22 23:04:12 +00:00
/*
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
* ft_table is a malloc(9)ed array of pointers. Pointers point to
* memory from UMA_ZONE_PCPU zone.
* ft_masks is per-cpu pointer itself. Each instance points
* to a malloc(9)ed bitset, that is private to corresponding CPU.
Spacing.
2014-02-07 10:05:12 +00:00
*/
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
struct flist **ft_table;
bitstr_t **ft_masks;
simplify code by removing bit_fns and replacing with the use of a temporary mask
2009-04-20 16:16:43 +00:00
bitstr_t *ft_tmpmask;
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
};
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
#define FLOWSTAT_ADD(ft, name, v) \
counter_u64_add((ft)->ft_stat[offsetof(struct flowtable_stat, name) / sizeof(uint64_t)], (v))
#define FLOWSTAT_INC(ft, name) FLOWSTAT_ADD(ft, name, 1)
V_irtualize flowtable state. This change should make options VIMAGE kernel builds usable again, to some extent at least. Note that the size of struct vnet_inet has changed, though in accordance with one-bump-per-day policy we didn't update the __FreeBSD_version number, given that it has already been touched by r194640 a few hours ago. Reviewed by: bz Approved by: julian (mentor)
2009-06-22 21:19:24 +00:00
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
static struct proc *flowcleanerproc;
static uint32_t flow_hashjitter;
- boot-time size the ipv4 flowtable and the maximum number of flows - increase flow cleaning frequency and decrease flow caching time when near the flow limit - stop allocating new flows when within 3% of maxflows don't start allocating again until below 12.5% MFC after: 7 days
2010-03-22 23:04:12 +00:00
Move the increment operation under the lock and split the condition variable into two so that we can see on which one we are waiting. This might also more properly propagate the update of the flowclean_cycles flag and avoid "hangs" people were seeing. Suggested by: rwatson [1] Sponsored by: ISPsystem [1] Reviewed by: julian [1] Updated by: Mikolaj Golub (to.my.trociny gmail.com) Tested by: Mikolaj Golub (to.my.trociny gmail.com) MFC After: 1 week [1] Early 2010, initial version.
2010-12-31 21:06:52 +00:00
static struct cv flowclean_f_cv;
static struct cv flowclean_c_cv;
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
static struct mtx flowclean_lock;
static uint32_t flowclean_cycles;
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
/*
* TODO:
Spacing.
2014-02-07 10:05:12 +00:00
* - add sysctls to resize && flush flow tables
update TODO list
2009-04-19 04:39:42 +00:00
* - Add per flowtable sysctls for statistics and configuring timeouts
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
* - add saturation counter to rtentry to support per-packet load-balancing
* add flag to indicate round-robin flow, add list lookup from head
for flows
* - add sysctl / device node / syscall to support exporting and importing
* of flows with flag to indicate that a flow was imported so should
* not be considered for auto-cleaning
* - support explicit connection state (currently only ad-hoc for DSR)
V_irtualize flowtable state. This change should make options VIMAGE kernel builds usable again, to some extent at least. Note that the size of struct vnet_inet has changed, though in accordance with one-bump-per-day policy we didn't update the __FreeBSD_version number, given that it has already been touched by r194640 a few hours ago. Reviewed by: bz Approved by: julian (mentor)
2009-06-22 21:19:24 +00:00
* - idetach() cleanup for options VIMAGE builds.
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
*/
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
#ifdef INET
static VNET_DEFINE(struct flowtable, ip4_ft);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
#define V_ip4_ft VNET(ip4_ft)
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
#endif
#ifdef INET6
static VNET_DEFINE(struct flowtable, ip6_ft);
#define V_ip6_ft VNET(ip6_ft)
#endif
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
static uma_zone_t flow_zone;
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
static VNET_DEFINE(int, flowtable_enable) = 1;
Remove unused VNET_SET() and related macros; only VNET_GET() is ever actually used. Rename VNET_GET() to VNET() to shorten variable references. Discussed with: bz, julian Reviewed by: bz Approved by: re (kensmith, kib)
2009-07-16 21:13:04 +00:00
#define V_flowtable_enable VNET(flowtable_enable)
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
static SYSCTL_NODE(_net, OID_AUTO, flowtable, CTLFLAG_RD, NULL,
Mark all SYSCTL_NODEs static that have no corresponding SYSCTL_DECLs. The SYSCTL_NODE macro defines a list that stores all child-elements of that node. If there's no SYSCTL_DECL macro anywhere else, there's no reason why it shouldn't be static.
2011-11-07 15:43:11 +00:00
"flowtable");
Remove SYSCTL_VNET_* macros, and simply put CTLFLAG_VNET where needed. Sponsored by: Nginx, Inc.
2014-11-07 09:39:05 +00:00
SYSCTL_INT(_net_flowtable, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
&VNET_NAME(flowtable_enable), 0, "enable flowtable caching.");
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
SYSCTL_UMA_MAX(_net_flowtable, OID_AUTO, maxflows, CTLFLAG_RW,
&flow_zone, "Maximum number of flows allowed");
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
static MALLOC_DEFINE(M_FTABLE, "flowtable", "flowtable hashes and bitstrings");
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
static struct flentry *
flowtable_lookup_common(struct flowtable *, uint32_t *, int, uint32_t);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
#ifdef INET
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
static struct flentry *
flowtable_lookup_ipv4(struct mbuf *m, struct route *ro)
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
{
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
struct flentry *fle;
struct sockaddr_in *sin;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
struct ip *ip;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
uint32_t fibnum;
#ifdef FLOWTABLE_HASH_ALL
uint32_t key[3];
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
int iphlen;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
uint16_t sport, dport;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
uint8_t proto;
#endif
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
ip = mtod(m, struct ip *);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (ip->ip_src.s_addr == ip->ip_dst.s_addr ||
(ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
(ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
return (NULL);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
fibnum = M_GETFIB(m);
#ifdef FLOWTABLE_HASH_ALL
iphlen = ip->ip_hl << 2;
proto = ip->ip_p;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
switch (proto) {
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
case IPPROTO_TCP: {
struct tcphdr *th;
th = (struct tcphdr *)((char *)ip + iphlen);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
sport = th->th_sport;
dport = th->th_dport;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (th->th_flags & (TH_RST|TH_FIN))
fibnum |= (FL_STALE << 24);
Spacing.
2014-02-07 10:05:12 +00:00
break;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
}
case IPPROTO_UDP: {
struct udphdr *uh;
uh = (struct udphdr *)((char *)ip + iphlen);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
sport = uh->uh_sport;
dport = uh->uh_dport;
Spacing.
2014-02-07 10:05:12 +00:00
break;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
}
case IPPROTO_SCTP: {
struct sctphdr *sh;
sh = (struct sctphdr *)((char *)ip + iphlen);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
sport = sh->src_port;
dport = sh->dest_port;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
/* XXXGL: handle stale? */
Spacing.
2014-02-07 10:05:12 +00:00
break;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
}
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
default:
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
sport = dport = 0;
Remove extraneous semicolons, no functional changes. Submitted by: Marc Balmer <marc@msys.ch> MFC after: 1 week
2010-01-07 21:01:37 +00:00
break;
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
key[0] = ip->ip_dst.s_addr;
key[1] = ip->ip_src.s_addr;
key[2] = (dport << 16) | sport;
fibnum |= proto << 16;
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
fle = flowtable_lookup_common(&V_ip4_ft, key, 3 * sizeof(uint32_t),
fibnum);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#else /* !FLOWTABLE_HASH_ALL */
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
fle = flowtable_lookup_common(&V_ip4_ft, (uint32_t *)&ip->ip_dst,
sizeof(struct in_addr), fibnum);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#endif /* FLOWTABLE_HASH_ALL */
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (fle == NULL)
return (NULL);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
sin = (struct sockaddr_in *)&ro->ro_dst;
sin->sin_family = AF_INET;
sin->sin_len = sizeof(*sin);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
sin->sin_addr = ip->ip_dst;
return (fle);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
}
#endif /* INET */
#ifdef INET6
/*
* PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
* then it sets p to point at the offset "len" in the mbuf. WARNING: the
* pointer might become stale after other pullups (but we never use it
* this way).
*/
#define PULLUP_TO(_len, p, T) \
do { \
int x = (_len) + sizeof(T); \
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if ((m)->m_len < x) \
return (NULL); \
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
p = (mtod(m, char *) + (_len)); \
} while (0)
#define TCP(p) ((struct tcphdr *)(p))
#define SCTP(p) ((struct sctphdr *)(p))
#define UDP(p) ((struct udphdr *)(p))
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
static struct flentry *
flowtable_lookup_ipv6(struct mbuf *m, struct route *ro)
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
{
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
struct flentry *fle;
struct sockaddr_in6 *sin6;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
struct ip6_hdr *ip6;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
uint32_t fibnum;
#ifdef FLOWTABLE_HASH_ALL
uint32_t key[9];
void *ulp;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
int hlen;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
uint16_t sport, dport;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
u_short offset;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
uint8_t proto;
#else
uint32_t key[4];
#endif
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
ip6 = mtod(m, struct ip6_hdr *);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (in6_localaddr(&ip6->ip6_dst))
return (NULL);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
fibnum = M_GETFIB(m);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef FLOWTABLE_HASH_ALL
hlen = sizeof(struct ip6_hdr);
proto = ip6->ip6_nxt;
offset = sport = dport = 0;
ulp = NULL;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
while (ulp == NULL) {
switch (proto) {
case IPPROTO_ICMPV6:
case IPPROTO_OSPFIGP:
case IPPROTO_PIM:
case IPPROTO_CARP:
case IPPROTO_ESP:
case IPPROTO_NONE:
ulp = ip6;
break;
case IPPROTO_TCP:
PULLUP_TO(hlen, ulp, struct tcphdr);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
dport = TCP(ulp)->th_dport;
sport = TCP(ulp)->th_sport;
if (TCP(ulp)->th_flags & (TH_RST|TH_FIN))
fibnum |= (FL_STALE << 24);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
break;
case IPPROTO_SCTP:
PULLUP_TO(hlen, ulp, struct sctphdr);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
dport = SCTP(ulp)->src_port;
sport = SCTP(ulp)->dest_port;
/* XXXGL: handle stale? */
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
break;
case IPPROTO_UDP:
PULLUP_TO(hlen, ulp, struct udphdr);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
dport = UDP(ulp)->uh_dport;
sport = UDP(ulp)->uh_sport;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
break;
case IPPROTO_HOPOPTS: /* RFC 2460 */
PULLUP_TO(hlen, ulp, struct ip6_hbh);
hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
ulp = NULL;
break;
case IPPROTO_ROUTING: /* RFC 2460 */
Spacing.
2014-02-08 22:10:53 +00:00
PULLUP_TO(hlen, ulp, struct ip6_rthdr);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
ulp = NULL;
break;
case IPPROTO_FRAGMENT: /* RFC 2460 */
PULLUP_TO(hlen, ulp, struct ip6_frag);
hlen += sizeof (struct ip6_frag);
proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
IP6F_OFF_MASK;
ulp = NULL;
break;
case IPPROTO_DSTOPTS: /* RFC 2460 */
PULLUP_TO(hlen, ulp, struct ip6_hbh);
hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
ulp = NULL;
break;
case IPPROTO_AH: /* RFC 2402 */
PULLUP_TO(hlen, ulp, struct ip6_ext);
hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
ulp = NULL;
break;
default:
PULLUP_TO(hlen, ulp, struct ip6_ext);
break;
}
}
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
bcopy(&ip6->ip6_dst, &key[0], sizeof(struct in6_addr));
bcopy(&ip6->ip6_src, &key[4], sizeof(struct in6_addr));
key[8] = (dport << 16) | sport;
fibnum |= proto << 16;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
fle = flowtable_lookup_common(&V_ip6_ft, key, 9 * sizeof(uint32_t),
fibnum);
#else /* !FLOWTABLE_HASH_ALL */
bcopy(&ip6->ip6_dst, &key[0], sizeof(struct in6_addr));
fle = flowtable_lookup_common(&V_ip6_ft, key, sizeof(struct in6_addr),
fibnum);
#endif /* FLOWTABLE_HASH_ALL */
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (fle == NULL)
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
return (NULL);
sin6 = (struct sockaddr_in6 *)&ro->ro_dst;
sin6->sin6_family = AF_INET6;
sin6->sin6_len = sizeof(*sin6);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
bcopy(&ip6->ip6_dst, &sin6->sin6_addr, sizeof(struct in6_addr));
return (fle);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
}
#endif /* INET6 */
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
static bitstr_t *
flowtable_mask(struct flowtable *ft)
{
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
Whitespace.
2014-02-17 12:02:44 +00:00
/*
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
* flowtable_free_stale() calls w/o critical section, but
* with sched_bind(). Since pointer is stable throughout
* ft lifetime, it is safe, otherwise...
*
* CRITICAL_ASSERT(curthread);
*/
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
return (*(bitstr_t **)zpcpu_get(ft->ft_masks));
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
static struct flist *
flowtable_list(struct flowtable *ft, uint32_t hash)
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
{
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
CRITICAL_ASSERT(curthread);
return (zpcpu_get(ft->ft_table[hash % ft->ft_size]));
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
static int
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
flow_stale(struct flowtable *ft, struct flentry *fle, int maxidle)
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
{
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
Fix three flowtable bugs, a) one lookup issue, b) a two cleaner issue. MFC after: 3 days Sponsored by: Netflix Inc. Differential Revision: https://reviews.freebsd.org/D4014
2015-11-02 21:21:00 +00:00
if (((fle->f_rt->rt_flags & RTF_UP) == 0) ||
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
(fle->f_rt->rt_ifp == NULL) ||
!RT_LINK_IS_UP(fle->f_rt->rt_ifp) ||
(fle->f_lle->la_flags & LLE_VALID) == 0)
return (1);
if (time_uptime - fle->f_uptime > maxidle)
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
return (1);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef FLOWTABLE_HASH_ALL
if (fle->f_flags & FL_STALE)
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
return (1);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#endif
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
return (0);
}
- boot-time size the ipv4 flowtable and the maximum number of flows - increase flow cleaning frequency and decrease flow caching time when near the flow limit - stop allocating new flows when within 3% of maxflows don't start allocating again until below 12.5% MFC after: 7 days
2010-03-22 23:04:12 +00:00
static int
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
flow_full(void)
- boot-time size the ipv4 flowtable and the maximum number of flows - increase flow cleaning frequency and decrease flow caching time when near the flow limit - stop allocating new flows when within 3% of maxflows don't start allocating again until below 12.5% MFC after: 7 days
2010-03-22 23:04:12 +00:00
{
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
int count, max;
Spacing.
2014-02-08 22:10:53 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
count = uma_zone_get_cur(flow_zone);
max = uma_zone_get_max(flow_zone);
- boot-time size the ipv4 flowtable and the maximum number of flows - increase flow cleaning frequency and decrease flow caching time when near the flow limit - stop allocating new flows when within 3% of maxflows don't start allocating again until below 12.5% MFC after: 7 days
2010-03-22 23:04:12 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
return (count > (max - (max >> 3)));
- boot-time size the ipv4 flowtable and the maximum number of flows - increase flow cleaning frequency and decrease flow caching time when near the flow limit - stop allocating new flows when within 3% of maxflows don't start allocating again until below 12.5% MFC after: 7 days
2010-03-22 23:04:12 +00:00
}
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
static int
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
flow_matches(struct flentry *fle, uint32_t *key, int keylen, uint32_t fibnum)
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
{
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef FLOWTABLE_HASH_ALL
uint8_t proto;
proto = (fibnum >> 16) & 0xff;
fibnum &= 0xffff;
#endif
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
CRITICAL_ASSERT(curthread);
/* Microoptimization for IPv4: don't use bcmp(). */
Fix three flowtable bugs, a) one lookup issue, b) a two cleaner issue. MFC after: 3 days Sponsored by: Netflix Inc. Differential Revision: https://reviews.freebsd.org/D4014
2015-11-02 21:21:00 +00:00
if (((keylen == sizeof(uint32_t) && (fle->f_key[0] == key[0])) ||
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
(bcmp(fle->f_key, key, keylen) == 0)) &&
fibnum == fle->f_fibnum &&
#ifdef FLOWTABLE_HASH_ALL
proto == fle->f_proto &&
#endif
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
(fle->f_rt->rt_flags & RTF_UP) &&
fle->f_rt->rt_ifp != NULL &&
(fle->f_lle->la_flags & LLE_VALID))
return (1);
return (0);
}
static struct flentry *
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
flowtable_insert(struct flowtable *ft, uint32_t hash, uint32_t *key,
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
int keylen, uint32_t fibnum0)
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
{
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef INET6
Whitespace.
2014-02-17 12:02:44 +00:00
struct route_in6 sro6;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#endif
#ifdef INET
Whitespace.
2014-02-17 12:02:44 +00:00
struct route sro;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#endif
struct route *ro = NULL;
struct rtentry *rt;
struct lltable *lt = NULL;
struct llentry *lle;
struct sockaddr_storage *l3addr;
struct ifnet *ifp;
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
struct flist *flist;
struct flentry *fle, *iter;
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
bitstr_t *mask;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
uint16_t fibnum = fibnum0;
#ifdef FLOWTABLE_HASH_ALL
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
uint8_t proto;
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
proto = (fibnum0 >> 16) & 0xff;
fibnum = fibnum0 & 0xffff;
#endif
/*
* This bit of code ends up locking the
* same route 3 times (just like ip_output + ether_output)
* - at lookup
* - in rt_check when called by arpresolve
* - dropping the refcount for the rtentry
*
* This could be consolidated to one if we wrote a variant
* of arpresolve with an rt_check variant that expected to
* receive the route locked
*/
#ifdef INET
if (ft == &V_ip4_ft) {
struct sockaddr_in *sin;
ro = &sro;
bzero(&sro.ro_dst, sizeof(sro.ro_dst));
sin = (struct sockaddr_in *)&sro.ro_dst;
sin->sin_family = AF_INET;
sin->sin_len = sizeof(*sin);
sin->sin_addr.s_addr = key[0];
}
#endif
#ifdef INET6
if (ft == &V_ip6_ft) {
struct sockaddr_in6 *sin6;
ro = (struct route *)&sro6;
sin6 = &sro6.ro_dst;
bzero(sin6, sizeof(*sin6));
sin6->sin6_family = AF_INET6;
sin6->sin6_len = sizeof(*sin6);
bcopy(key, &sin6->sin6_addr, sizeof(struct in6_addr));
}
#endif
ro->ro_rt = NULL;
#ifdef RADIX_MPATH
rtalloc_mpath_fib(ro, hash, fibnum);
#else
rtalloc_ign_fib(ro, 0, fibnum);
#endif
if (ro->ro_rt == NULL)
return (NULL);
rt = ro->ro_rt;
ifp = rt->rt_ifp;
if (ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) {
RTFREE(rt);
return (NULL);
}
#ifdef INET
if (ft == &V_ip4_ft)
lt = LLTABLE(ifp);
#endif
#ifdef INET6
if (ft == &V_ip6_ft)
lt = LLTABLE6(ifp);
#endif
if (rt->rt_flags & RTF_GATEWAY)
l3addr = (struct sockaddr_storage *)rt->rt_gateway;
else
l3addr = (struct sockaddr_storage *)&ro->ro_dst;
lle = llentry_alloc(ifp, lt, l3addr);
if (lle == NULL) {
RTFREE(rt);
return (NULL);
}
/* Don't insert the entry if the ARP hasn't yet finished resolving. */
if ((lle->la_flags & LLE_VALID) == 0) {
RTFREE(rt);
LLE_FREE(lle);
FLOWSTAT_INC(ft, ft_fail_lle_invalid);
return (NULL);
}
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
fle = uma_zalloc(flow_zone, M_NOWAIT | M_ZERO);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (fle == NULL) {
RTFREE(rt);
LLE_FREE(lle);
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
return (NULL);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
}
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
fle->f_hash = hash;
bcopy(key, &fle->f_key, keylen);
fle->f_rt = rt;
fle->f_lle = lle;
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
fle->f_fibnum = fibnum;
fle->f_uptime = time_uptime;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef FLOWTABLE_HASH_ALL
fle->f_proto = proto;
fle->f_flags = fibnum0 >> 24;
#endif
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
critical_enter();
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
mask = flowtable_mask(ft);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
flist = flowtable_list(ft, hash);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
if (SLIST_EMPTY(flist)) {
bit_set(mask, (hash % ft->ft_size));
SLIST_INSERT_HEAD(flist, fle, f_next);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
goto skip;
Spacing.
2014-02-07 10:05:12 +00:00
}
Spacing.
2014-02-08 22:10:53 +00:00
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
/*
* find end of list and make sure that we were not
* preempted by another thread handling this flow
*/
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
SLIST_FOREACH(iter, flist, f_next) {
Fix incorrect assertions.
2014-02-18 14:21:26 +00:00
KASSERT(iter->f_hash % ft->ft_size == hash % ft->ft_size,
("%s: wrong hash", __func__));
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (flow_matches(iter, key, keylen, fibnum)) {
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
/*
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
* We probably migrated to an other CPU after
* lookup in flowtable_lookup_common() failed.
* It appeared that this CPU already has flow
* entry.
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
*/
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
iter->f_uptime = time_uptime;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef FLOWTABLE_HASH_ALL
iter->f_flags |= fibnum >> 24;
#endif
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
critical_exit();
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
FLOWSTAT_INC(ft, ft_collisions);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
uma_zfree(flow_zone, fle);
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
return (iter);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
Spacing.
2014-02-07 10:05:12 +00:00
}
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
SLIST_INSERT_HEAD(flist, fle, f_next);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
skip:
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
critical_exit();
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
FLOWSTAT_INC(ft, ft_inserts);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
return (fle);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
int
flowtable_lookup(sa_family_t sa, struct mbuf *m, struct route *ro)
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
{
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
struct flentry *fle;
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
struct llentry *lle;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (V_flowtable_enable == 0)
return (ENXIO);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
switch (sa) {
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
#ifdef INET
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
case AF_INET:
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
fle = flowtable_lookup_ipv4(m, ro);
break;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
#endif
#ifdef INET6
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
case AF_INET6:
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
fle = flowtable_lookup_ipv6(m, ro);
break;
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
#endif
default:
panic("%s: sa %d", __func__, sa);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
}
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (fle == NULL)
return (EHOSTUNREACH);
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
Start process of removing the use of the deprecated "M_FLOWID" flag from the FreeBSD network code. The flag is still kept around in the "sys/mbuf.h" header file, but does no longer have any users. Instead the "m_pkthdr.rsstype" field in the mbuf structure is now used to decide the meaning of the "m_pkthdr.flowid" field. To modify the "m_pkthdr.rsstype" field please use the existing "M_HASHTYPE_XXX" macros as defined in the "sys/mbuf.h" header file. This patch introduces new behaviour in the transmit direction. Previously network drivers checked if "M_FLOWID" was set in "m_flags" before using the "m_pkthdr.flowid" field. This check has now now been replaced by checking if "M_HASHTYPE_GET(m)" is different from "M_HASHTYPE_NONE". In the future more hashtypes will be added, for example hashtypes for hardware dedicated flows. "M_HASHTYPE_OPAQUE" indicates that the "m_pkthdr.flowid" value is valid and has no particular type. This change removes the need for an "if" statement in TCP transmit code checking for the presence of a valid flowid value. The "if" statement mentioned above is now a direct variable assignment which is then later checked by the respective network drivers like before. Additional notes: - The SCTP code changes will be committed as a separate patch. - Removal of the "M_FLOWID" flag will also be done separately. - The FreeBSD version has been bumped. MFC after: 1 month Sponsored by: Mellanox Technologies
2014-12-01 11:45:24 +00:00
if (M_HASHTYPE_GET(m) == M_HASHTYPE_NONE) {
net: Use M_HASHTYPE_OPAQUE_HASH if the mbuf flowid has hash properties Reviewed by: hps, erj, tuexen Sponsored by: Microsoft OSTC Differential Revision: https://reviews.freebsd.org/D6688
2016-06-07 04:51:50 +00:00
M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE_HASH);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
m->m_pkthdr.flowid = fle->f_hash;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
}
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
ro->ro_rt = fle->f_rt;
ro->ro_flags |= RT_NORTREF;
Implement interface link header precomputation API. Add if_requestencap() interface method which is capable of calculating various link headers for given interface. Right now there is support for INET/INET6/ARP llheader calculation (IFENCAP_LL type request). Other types are planned to support more complex calculation (L2 multipath lagg nexthops, tunnel encap nexthops, etc..). Reshape 'struct route' to be able to pass additional data (with is length) to prepend to mbuf. These two changes permits routing code to pass pre-calculated nexthop data (like L2 header for route w/gateway) down to the stack eliminating the need for other lookups. It also brings us closer to more complex scenarios like transparently handling MPLS nexthops and tunnel interfaces. Last, but not least, it removes layering violation introduced by flowtable code (ro_lle) and simplifies handling of existing if_output consumers. ARP/ND changes: Make arp/ndp stack pre-calculate link header upon installing/updating lle record. Interface link address change are handled by re-calculating headers for all lles based on if_lladdr event. After these changes, arpresolve()/nd6_resolve() returns full pre-calculated header for supported interfaces thus simplifying if_output(). Move these lookups to separate ether_resolve_addr() function which ether returs error or fully-prepared link header. Add <arp|nd6_>resolve_addr() compat versions to return link addresses instead of pre-calculated data. BPF changes: Raw bpf writes occupied _two_ cases: AF_UNSPEC and pseudo_AF_HDRCMPLT. Despite the naming, both of there have ther header "complete". The only difference is that interface source mac has to be filled by OS for AF_UNSPEC (controlled via BIOCGHDRCMPLT). This logic has to stay inside BPF and not pollute if_output() routines. Convert BPF to pass prepend data via new 'struct route' mechanism. Note that it does not change non-optimized if_output(): ro_prepend handling is purely optional. Side note: hackish pseudo_AF_HDRCMPLT is supported for ethernet and FDDI. It is not needed for ethernet anymore. The only remaining FDDI user is dev/pdq mostly untouched since 2007. FDDI support was eliminated from OpenBSD in 2013 (sys/net/if_fddisubr.c rev 1.65). Flowtable changes: Flowtable violates layering by saving (and not correctly managing) rtes/lles. Instead of passing lle pointer, pass pointer to pre-calculated header data from that lle. Differential Revision: https://reviews.freebsd.org/D4102
2015-12-31 05:03:27 +00:00
lle = fle->f_lle;
This change re-adds L2 caching for TCP and UDP, as originally added in D4306 but removed due to other changes in the system. Restore the llentry pointer to the "struct route", and use it to cache the L2 lookup (ARP or ND6) as appropriate. Submitted by: Mike Karels Differential Revision: https://reviews.freebsd.org/D6262
2016-06-02 17:51:29 +00:00
if (lle != NULL && (lle->la_flags & LLE_VALID))
ro->ro_lle = lle; /* share ref with fle->f_lle */
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
return (0);
}
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
static struct flentry *
flowtable_lookup_common(struct flowtable *ft, uint32_t *key, int keylen,
uint32_t fibnum)
{
struct flist *flist;
struct flentry *fle;
uint32_t hash;
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
FLOWSTAT_INC(ft, ft_lookups);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
hash = jenkins_hash32(key, keylen / sizeof(uint32_t), flow_hashjitter);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
critical_enter();
flist = flowtable_list(ft, hash);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
SLIST_FOREACH(fle, flist, f_next) {
Fix incorrect assertions.
2014-02-18 14:21:26 +00:00
KASSERT(fle->f_hash % ft->ft_size == hash % ft->ft_size,
("%s: wrong hash", __func__));
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (flow_matches(fle, key, keylen, fibnum)) {
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
fle->f_uptime = time_uptime;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef FLOWTABLE_HASH_ALL
fle->f_flags |= fibnum >> 24;
#endif
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
critical_exit();
FLOWSTAT_INC(ft, ft_hits);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
return (fle);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
}
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
}
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
critical_exit();
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
FLOWSTAT_INC(ft, ft_misses);
Whenever flowtable lookup fails, we do route lookup and then try to insert flow entry. During the route lookup the critical section is exited. It may happen, that after route lookup we will be executed on an other CPU that already has such flowentry. Before this change we simply freed the flowentry and returned to ip_output() with failure. Actually there is nothing wrong with using previously allocated flow entry, updating it properly. Thus, make flowentry_insert() return the new either old fle, and make use of it. Count reuses as "collisions" and real inserts as "inserts". Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-14 10:56:26 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
return (flowtable_insert(ft, hash, key, keylen, fibnum));
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
static void
flowtable_alloc(struct flowtable *ft)
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
{
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
ft->ft_table = malloc(ft->ft_size * sizeof(struct flist),
M_FTABLE, M_WAITOK);
for (int i = 0; i < ft->ft_size; i++)
ft->ft_table[i] = uma_zalloc(pcpu_zone_ptr, M_WAITOK | M_ZERO);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
ft->ft_masks = uma_zalloc(pcpu_zone_ptr, M_WAITOK);
for (int i = 0; i < mp_ncpus; i++) {
bitstr_t **b;
b = zpcpu_get_cpu(ft->ft_masks, i);
Improve performance and functionality of the bitstring(3) api Two new functions are provided, bit_ffs_at() and bit_ffc_at(), which allow for efficient searching of set or cleared bits starting from any bit offset within the bit string. Performance is improved by operating on longs instead of bytes and using ffsl() for searches within a long. ffsl() is a compiler builtin in both clang and gcc for most architectures, converting what was a brute force while loop search into a couple of instructions. All of the bitstring(3) API continues to be contained in the header file. Some of the functions are large enough that perhaps they should be uninlined and moved to a library, but that is beyond the scope of this commit. sys/sys/bitstring.h: Convert the majority of the existing bit string implementation from macros to inline functions. Properly protect the implementation from inadvertant macro expansion when included in a user's program by prefixing all private macros/functions and local variables with '_'. Add bit_ffs_at() and bit_ffc_at(). Implement bit_ffs() and bit_ffc() in terms of their "at" counterparts. Provide a kernel implementation of bit_alloc(), making the full API usable in the kernel. Improve code documenation. share/man/man3/bitstring.3: Add pre-exisiting API bit_ffc() to the synopsis. Document new APIs. Document the initialization state of the bit strings allocated/declared by bit_alloc() and bit_decl(). Correct documentation for bitstr_size(). The original code comments indicate the size is in bytes, not "elements of bitstr_t". The new implementation follows this lead. Only hastd assumed "elements" rather than bytes and it has been corrected. etc/mtree/BSD.tests.dist: tests/sys/Makefile: tests/sys/sys/Makefile: tests/sys/sys/bitstring.c: Add tests for all existing and new functionality. include/bitstring.h Include all headers needed by sys/bitstring.h lib/libbluetooth/bluetooth.h: usr.sbin/bluetooth/hccontrol/le.c: Include bitstring.h instead of sys/bitstring.h. sbin/hastd/activemap.c: Correct usage of bitstr_size(). sys/dev/xen/blkback/blkback.c Use new bit_alloc. sys/kern/subr_unit.c: Remove hard-coded assumption that sizeof(bitstr_t) is 1. Get rid of unrb.busy, which caches the number of bits set in unrb.map. When INVARIANTS are disabled, nothing needs to know that information. callapse_unr can be adapted to use bit_ffs and bit_ffc instead. Eliminating unrb.busy saves memory, simplifies the code, and provides a slight speedup when INVARIANTS are disabled. sys/net/flowtable.c: Use the new kernel implementation of bit-alloc, instead of hacking the old libc-dependent macro. sys/sys/param.h Update __FreeBSD_version to indicate availability of new API Submitted by: gibbs, asomers Reviewed by: gibbs, ngie MFC after: 4 weeks Sponsored by: Spectra Logic Corp Differential Revision: https://reviews.freebsd.org/D6004
2016-05-04 22:34:11 +00:00
*b = bit_alloc(ft->ft_size, M_FTABLE, M_WAITOK);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
Improve performance and functionality of the bitstring(3) api Two new functions are provided, bit_ffs_at() and bit_ffc_at(), which allow for efficient searching of set or cleared bits starting from any bit offset within the bit string. Performance is improved by operating on longs instead of bytes and using ffsl() for searches within a long. ffsl() is a compiler builtin in both clang and gcc for most architectures, converting what was a brute force while loop search into a couple of instructions. All of the bitstring(3) API continues to be contained in the header file. Some of the functions are large enough that perhaps they should be uninlined and moved to a library, but that is beyond the scope of this commit. sys/sys/bitstring.h: Convert the majority of the existing bit string implementation from macros to inline functions. Properly protect the implementation from inadvertant macro expansion when included in a user's program by prefixing all private macros/functions and local variables with '_'. Add bit_ffs_at() and bit_ffc_at(). Implement bit_ffs() and bit_ffc() in terms of their "at" counterparts. Provide a kernel implementation of bit_alloc(), making the full API usable in the kernel. Improve code documenation. share/man/man3/bitstring.3: Add pre-exisiting API bit_ffc() to the synopsis. Document new APIs. Document the initialization state of the bit strings allocated/declared by bit_alloc() and bit_decl(). Correct documentation for bitstr_size(). The original code comments indicate the size is in bytes, not "elements of bitstr_t". The new implementation follows this lead. Only hastd assumed "elements" rather than bytes and it has been corrected. etc/mtree/BSD.tests.dist: tests/sys/Makefile: tests/sys/sys/Makefile: tests/sys/sys/bitstring.c: Add tests for all existing and new functionality. include/bitstring.h Include all headers needed by sys/bitstring.h lib/libbluetooth/bluetooth.h: usr.sbin/bluetooth/hccontrol/le.c: Include bitstring.h instead of sys/bitstring.h. sbin/hastd/activemap.c: Correct usage of bitstr_size(). sys/dev/xen/blkback/blkback.c Use new bit_alloc. sys/kern/subr_unit.c: Remove hard-coded assumption that sizeof(bitstr_t) is 1. Get rid of unrb.busy, which caches the number of bits set in unrb.map. When INVARIANTS are disabled, nothing needs to know that information. callapse_unr can be adapted to use bit_ffs and bit_ffc instead. Eliminating unrb.busy saves memory, simplifies the code, and provides a slight speedup when INVARIANTS are disabled. sys/net/flowtable.c: Use the new kernel implementation of bit-alloc, instead of hacking the old libc-dependent macro. sys/sys/param.h Update __FreeBSD_version to indicate availability of new API Submitted by: gibbs, asomers Reviewed by: gibbs, ngie MFC after: 4 weeks Sponsored by: Spectra Logic Corp Differential Revision: https://reviews.freebsd.org/D6004
2016-05-04 22:34:11 +00:00
ft->ft_tmpmask = bit_alloc(ft->ft_size, M_FTABLE, M_WAITOK);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
static void
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
flowtable_free_stale(struct flowtable *ft, struct rtentry *rt, int maxidle)
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
{
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
struct flist *flist, freelist;
struct flentry *fle, *fle1, *fleprev;
simplify code by removing bit_fns and replacing with the use of a temporary mask
2009-04-20 16:16:43 +00:00
bitstr_t *mask, *tmpmask;
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
int curbit, tmpsize;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
SLIST_INIT(&freelist);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
mask = flowtable_mask(ft);
simplify code by removing bit_fns and replacing with the use of a temporary mask
2009-04-20 16:16:43 +00:00
tmpmask = ft->ft_tmpmask;
In the flowtable scanner, restart the scan at the last found position, not at position 0. Changes the scanner from O(N^2) to O(N). Submitted by: scottl Obtained from: Netflix, Inc MFC after: 3 weeks
2013-10-15 21:28:51 +00:00
tmpsize = ft->ft_size;
simplify code by removing bit_fns and replacing with the use of a temporary mask
2009-04-20 16:16:43 +00:00
memcpy(tmpmask, mask, ft->ft_size/8);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
curbit = 0;
Pacify gcc.
2014-03-05 02:35:15 +00:00
fleprev = NULL; /* pacify gcc */
simplify code by removing bit_fns and replacing with the use of a temporary mask
2009-04-20 16:16:43 +00:00
/*
* XXX Note to self, bit_ffs operates at the byte level
* and thus adds gratuitous overhead
*/
bit_ffs(tmpmask, ft->ft_size, &curbit);
while (curbit != -1) {
- put larger flowtable members at the end - fix bug where tail pointer of the free list would not get advanced - clear entry's next pointer when it is added to the freelist to avoid freeing an entry that it still points to
2009-04-19 04:24:56 +00:00
if (curbit >= ft->ft_size || curbit < -1) {
log(LOG_ALERT,
"warning: bad curbit value %d \n",
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
curbit);
- put larger flowtable members at the end - fix bug where tail pointer of the free list would not get advanced - clear entry's next pointer when it is added to the freelist to avoid freeing an entry that it still points to
2009-04-19 04:24:56 +00:00
break;
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
FLOWSTAT_INC(ft, ft_free_checks);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
critical_enter();
flist = flowtable_list(ft, curbit);
- put larger flowtable members at the end - fix bug where tail pointer of the free list would not get advanced - clear entry's next pointer when it is added to the freelist to avoid freeing an entry that it still points to
2009-04-19 04:24:56 +00:00
#ifdef DIAGNOSTIC
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
if (SLIST_EMPTY(flist) && curbit > 0) {
- put larger flowtable members at the end - fix bug where tail pointer of the free list would not get advanced - clear entry's next pointer when it is added to the freelist to avoid freeing an entry that it still points to
2009-04-19 04:24:56 +00:00
log(LOG_ALERT,
"warning bit=%d set, but no fle found\n",
curbit);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
Spacing.
2014-02-08 22:10:53 +00:00
#endif
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
SLIST_FOREACH_SAFE(fle, flist, f_next, fle1) {
if (rt != NULL && fle->f_rt != rt) {
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
fleprev = fle;
continue;
}
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (!flow_stale(ft, fle, maxidle)) {
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
fleprev = fle;
continue;
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
if (fle == SLIST_FIRST(flist))
SLIST_REMOVE_HEAD(flist, f_next);
else
SLIST_REMOVE_AFTER(fleprev, f_next);
SLIST_INSERT_HEAD(&freelist, fle, f_next);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
if (SLIST_EMPTY(flist))
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
bit_clear(mask, curbit);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
critical_exit();
simplify code by removing bit_fns and replacing with the use of a temporary mask
2009-04-20 16:16:43 +00:00
bit_clear(tmpmask, curbit);
In the flowtable scanner, restart the scan at the last found position, not at position 0. Changes the scanner from O(N^2) to O(N). Submitted by: scottl Obtained from: Netflix, Inc MFC after: 3 weeks
2013-10-15 21:28:51 +00:00
bit_ffs(tmpmask, tmpsize, &curbit);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
SLIST_FOREACH_SAFE(fle, &freelist, f_next, fle1) {
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
FLOWSTAT_INC(ft, ft_frees);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
if (fle->f_rt != NULL)
RTFREE(fle->f_rt);
if (fle->f_lle != NULL)
LLE_FREE(fle->f_lle);
uma_zfree(flow_zone, fle);
}
}
static void
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
flowtable_clean_vnet(struct flowtable *ft, struct rtentry *rt, int maxidle)
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
{
int i;
CPU_FOREACH(i) {
if (smp_started == 1) {
thread_lock(curthread);
sched_bind(curthread, i);
thread_unlock(curthread);
}
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
flowtable_free_stale(ft, rt, maxidle);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
if (smp_started == 1) {
thread_lock(curthread);
sched_unbind(curthread);
thread_unlock(curthread);
}
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
}
The flow-table associates TCP/UDP flows and IP destinations with specific routes. When the routing table changes, for example, when a new route with a more specific prefix is inserted into the routing table, the flow-table is not updated to reflect that change. As such existing connections cannot take advantage of the new path. In some cases the path is broken. This patch will update the affected flow-table entries when a more specific route is added. The route entry is properly marked when a route is deleted from the table. In this case, when the flow-table performs a search, the stale entry is updated automatically. Therefore this patch is not necessary for route deletion. Submitted by: simon, phk Reviewed by: bz, kmacy MFC after: 3 days
2009-10-01 20:32:29 +00:00
void
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
flowtable_route_flush(sa_family_t sa, struct rtentry *rt)
The flow-table associates TCP/UDP flows and IP destinations with specific routes. When the routing table changes, for example, when a new route with a more specific prefix is inserted into the routing table, the flow-table is not updated to reflect that change. As such existing connections cannot take advantage of the new path. In some cases the path is broken. This patch will update the affected flow-table entries when a more specific route is added. The route entry is properly marked when a route is deleted from the table. In this case, when the flow-table performs a search, the stale entry is updated automatically. Therefore this patch is not necessary for route deletion. Submitted by: simon, phk Reviewed by: bz, kmacy MFC after: 3 days
2009-10-01 20:32:29 +00:00
{
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
struct flowtable *ft;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
switch (sa) {
#ifdef INET
case AF_INET:
ft = &V_ip4_ft;
break;
#endif
#ifdef INET6
case AF_INET6:
ft = &V_ip6_ft;
break;
#endif
default:
panic("%s: sa %d", __func__, sa);
}
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
flowtable_clean_vnet(ft, rt, 0);
V_irtualize flowtable state. This change should make options VIMAGE kernel builds usable again, to some extent at least. Note that the size of struct vnet_inet has changed, though in accordance with one-bump-per-day policy we didn't update the __FreeBSD_version number, given that it has already been touched by r194640 a few hours ago. Reviewed by: bz Approved by: julian (mentor)
2009-06-22 21:19:24 +00:00
}
static void
flowtable_cleaner(void)
{
VNET_ITERATOR_DECL(vnet_iter);
- Restore dropping the priority of syncer down to PPAUSE when it is idle. This was lost when it was converted to using a condition variable instead of lbolt. - Drop the priority of flowtable down to PPAUSE when it is idle as well since it is a similar background task. MFC after: 2 weeks
2011-01-06 22:17:07 +00:00
struct thread *td;
V_irtualize flowtable state. This change should make options VIMAGE kernel builds usable again, to some extent at least. Note that the size of struct vnet_inet has changed, though in accordance with one-bump-per-day policy we didn't update the __FreeBSD_version number, given that it has already been touched by r194640 a few hours ago. Reviewed by: bz Approved by: julian (mentor)
2009-06-22 21:19:24 +00:00
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
if (bootverbose)
log(LOG_INFO, "flowtable cleaner started\n");
- Restore dropping the priority of syncer down to PPAUSE when it is idle. This was lost when it was converted to using a condition variable instead of lbolt. - Drop the priority of flowtable down to PPAUSE when it is idle as well since it is a similar background task. MFC after: 2 weeks
2011-01-06 22:17:07 +00:00
td = curthread;
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
while (1) {
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
uint32_t flowclean_freq, maxidle;
/*
* The maximum idle time, as well as frequency are arbitrary.
*/
if (flow_full())
maxidle = 5;
else
maxidle = 30;
V_irtualize flowtable state. This change should make options VIMAGE kernel builds usable again, to some extent at least. Note that the size of struct vnet_inet has changed, though in accordance with one-bump-per-day policy we didn't update the __FreeBSD_version number, given that it has already been touched by r194640 a few hours ago. Reviewed by: bz Approved by: julian (mentor)
2009-06-22 21:19:24 +00:00
VNET_LIST_RLOCK();
VNET_FOREACH(vnet_iter) {
CURVNET_SET(vnet_iter);
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
#ifdef INET
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
flowtable_clean_vnet(&V_ip4_ft, NULL, maxidle);
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
#endif
#ifdef INET6
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
flowtable_clean_vnet(&V_ip6_ft, NULL, maxidle);
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
#endif
V_irtualize flowtable state. This change should make options VIMAGE kernel builds usable again, to some extent at least. Note that the size of struct vnet_inet has changed, though in accordance with one-bump-per-day policy we didn't update the __FreeBSD_version number, given that it has already been touched by r194640 a few hours ago. Reviewed by: bz Approved by: julian (mentor)
2009-06-22 21:19:24 +00:00
CURVNET_RESTORE();
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
V_irtualize flowtable state. This change should make options VIMAGE kernel builds usable again, to some extent at least. Note that the size of struct vnet_inet has changed, though in accordance with one-bump-per-day policy we didn't update the __FreeBSD_version number, given that it has already been touched by r194640 a few hours ago. Reviewed by: bz Approved by: julian (mentor)
2009-06-22 21:19:24 +00:00
VNET_LIST_RUNLOCK();
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
if (flow_full())
flowclean_freq = 4*hz;
else
flowclean_freq = 20*hz;
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
mtx_lock(&flowclean_lock);
- Restore dropping the priority of syncer down to PPAUSE when it is idle. This was lost when it was converted to using a condition variable instead of lbolt. - Drop the priority of flowtable down to PPAUSE when it is idle as well since it is a similar background task. MFC after: 2 weeks
2011-01-06 22:17:07 +00:00
thread_lock(td);
sched_prio(td, PPAUSE);
thread_unlock(td);
Move the increment operation under the lock and split the condition variable into two so that we can see on which one we are waiting. This might also more properly propagate the update of the flowclean_cycles flag and avoid "hangs" people were seeing. Suggested by: rwatson [1] Sponsored by: ISPsystem [1] Reviewed by: julian [1] Updated by: Mikolaj Golub (to.my.trociny gmail.com) Tested by: Mikolaj Golub (to.my.trociny gmail.com) MFC After: 1 week [1] Early 2010, initial version.
2010-12-31 21:06:52 +00:00
flowclean_cycles++;
cv_broadcast(&flowclean_f_cv);
cv_timedwait(&flowclean_c_cv, &flowclean_lock, flowclean_freq);
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
mtx_unlock(&flowclean_lock);
}
}
static void
flowtable_flush(void *unused __unused)
{
uint64_t start;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
mtx_lock(&flowclean_lock);
start = flowclean_cycles;
while (start == flowclean_cycles) {
Move the increment operation under the lock and split the condition variable into two so that we can see on which one we are waiting. This might also more properly propagate the update of the flowclean_cycles flag and avoid "hangs" people were seeing. Suggested by: rwatson [1] Sponsored by: ISPsystem [1] Reviewed by: julian [1] Updated by: Mikolaj Golub (to.my.trociny gmail.com) Tested by: Mikolaj Golub (to.my.trociny gmail.com) MFC After: 1 week [1] Early 2010, initial version.
2010-12-31 21:06:52 +00:00
cv_broadcast(&flowclean_c_cv);
cv_wait(&flowclean_f_cv, &flowclean_lock);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
mtx_unlock(&flowclean_lock);
- Import infrastructure for caching flows as a means of accelerating L3 and L2 lookups as well as providing stateful load balancing when used with RADIX_MPATH. - Currently compiled in to i386 and amd64 but disabled by default, it can be enabled at runtime with 'sysctl net.inet.flowtable.enable=1'. - Embedded users can remove it entirely from the kernel by adding 'nooption FLOWTABLE' to their kernel config files. - A minimal hookup will be added to ip_output in a subsequent commit. I would like to see more review before bringing in changes that require more churn. Supported by: Bitgravity Inc.
2009-04-19 00:16:04 +00:00
}
static struct kproc_desc flow_kp = {
"flowcleaner",
flowtable_cleaner,
&flowcleanerproc
};
SYSINIT(flowcleaner, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, kproc_start, &flow_kp);
revert to opt-in flowtable
2009-06-09 21:55:28 +00:00
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
static int
flowtable_get_size(char *name)
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
{
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
int size;
if (TUNABLE_INT_FETCH(name, &size)) {
if (size < 256)
size = 256;
if (!powerof2(size)) {
printf("%s must be power of 2\n", name);
size = 2048;
}
} else {
/*
* round up to the next power of 2
*/
size = 1 << fls((1024 + maxusers * 64) - 1);
}
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
return (size);
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
}
static void
flowtable_init(const void *unused __unused)
{
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
flow_hashjitter = arc4random();
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
flow_zone = uma_zcreate("flows", sizeof(struct flentry),
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
NULL, NULL, NULL, NULL, (64-1), UMA_ZONE_MAXBUCKET);
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
uma_zone_set_max(flow_zone, 1024 + maxusers * 64 * mp_ncpus);
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
Move the increment operation under the lock and split the condition variable into two so that we can see on which one we are waiting. This might also more properly propagate the update of the flowclean_cycles flag and avoid "hangs" people were seeing. Suggested by: rwatson [1] Sponsored by: ISPsystem [1] Reviewed by: julian [1] Updated by: Mikolaj Golub (to.my.trociny gmail.com) Tested by: Mikolaj Golub (to.my.trociny gmail.com) MFC After: 1 week [1] Early 2010, initial version.
2010-12-31 21:06:52 +00:00
cv_init(&flowclean_c_cv, "c_flowcleanwait");
cv_init(&flowclean_f_cv, "f_flowcleanwait");
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
mtx_init(&flowclean_lock, "flowclean lock", NULL, MTX_DEF);
EVENTHANDLER_REGISTER(ifnet_departure_event, flowtable_flush, NULL,
EVENTHANDLER_PRI_ANY);
}
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
SYSINIT(flowtable_init, SI_SUB_PROTO_BEGIN, SI_ORDER_FIRST,
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
flowtable_init, NULL);
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
#ifdef INET
static SYSCTL_NODE(_net_flowtable, OID_AUTO, ip4, CTLFLAG_RD, NULL,
"Flowtable for IPv4");
static VNET_PCPUSTAT_DEFINE(struct flowtable_stat, ip4_ftstat);
VNET_PCPUSTAT_SYSINIT(ip4_ftstat);
VNET_PCPUSTAT_SYSUNINIT(ip4_ftstat);
SYSCTL_VNET_PCPUSTAT(_net_flowtable_ip4, OID_AUTO, stat, struct flowtable_stat,
ip4_ftstat, "Flowtable statistics for IPv4 "
"(struct flowtable_stat, net/flowtable.h)");
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
static void
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
flowtable_init_vnet_v4(const void *unused __unused)
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
{
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
V_ip4_ft.ft_size = flowtable_get_size("net.flowtable.ip4.size");
V_ip4_ft.ft_stat = VNET(ip4_ftstat);
flowtable_alloc(&V_ip4_ft);
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
}
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
VNET_SYSINIT(ft_vnet_v4, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
flowtable_init_vnet_v4, NULL);
#endif /* INET */
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
#ifdef INET6
static SYSCTL_NODE(_net_flowtable, OID_AUTO, ip6, CTLFLAG_RD, NULL,
"Flowtable for IPv6");
static VNET_PCPUSTAT_DEFINE(struct flowtable_stat, ip6_ftstat);
VNET_PCPUSTAT_SYSINIT(ip6_ftstat);
VNET_PCPUSTAT_SYSUNINIT(ip6_ftstat);
SYSCTL_VNET_PCPUSTAT(_net_flowtable_ip6, OID_AUTO, stat, struct flowtable_stat,
ip6_ftstat, "Flowtable statistics for IPv6 "
"(struct flowtable_stat, net/flowtable.h)");
static void
flowtable_init_vnet_v6(const void *unused __unused)
{
V_ip6_ft.ft_size = flowtable_get_size("net.flowtable.ip6.size");
V_ip6_ft.ft_stat = VNET(ip6_ftstat);
flowtable_alloc(&V_ip6_ft);
}
VNET_SYSINIT(flowtable_init_vnet_v6, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
flowtable_init_vnet_v6, NULL);
#endif /* INET6 */
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
#ifdef DDB
static bitstr_t *
flowtable_mask_pcpu(struct flowtable *ft, int cpuid)
{
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
return (zpcpu_get_cpu(*ft->ft_masks, cpuid));
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
}
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
static struct flist *
flowtable_list_pcpu(struct flowtable *ft, uint32_t hash, int cpuid)
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
{
Spacing.
2014-02-08 22:10:53 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
return (zpcpu_get_cpu(&ft->ft_table[hash % ft->ft_size], cpuid));
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
}
static void
flow_show(struct flowtable *ft, struct flentry *fle)
{
int idle_time;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
int rt_valid, ifp_valid;
volatile struct rtentry *rt;
struct ifnet *ifp = NULL;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
uint32_t *hashkey = fle->f_key;
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
idle_time = (int)(time_uptime - fle->f_uptime);
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
rt = fle->f_rt;
rt_valid = rt != NULL;
Spacing.
2014-02-07 10:05:12 +00:00
if (rt_valid)
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
ifp = rt->rt_ifp;
ifp_valid = ifp != NULL;
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#ifdef INET
if (ft == &V_ip4_ft) {
char daddr[4*sizeof "123"];
#ifdef FLOWTABLE_HASH_ALL
char saddr[4*sizeof "123"];
uint16_t sport, dport;
#endif
inet_ntoa_r(*(struct in_addr *) &hashkey[0], daddr);
#ifdef FLOWTABLE_HASH_ALL
Spacing.
2014-02-08 22:10:53 +00:00
inet_ntoa_r(*(struct in_addr *) &hashkey[1], saddr);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
dport = ntohs((uint16_t)(hashkey[2] >> 16));
sport = ntohs((uint16_t)(hashkey[2] & 0xffff));
db_printf("%s:%d->%s:%d", saddr, sport, daddr, dport);
#else
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
db_printf("%s ", daddr);
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#endif
}
#endif /* INET */
#ifdef INET6
if (ft == &V_ip6_ft) {
#ifdef FLOWTABLE_HASH_ALL
db_printf("\n\tkey=%08x:%08x:%08x%08x:%08x:%08x%08x:%08x:%08x",
hashkey[0], hashkey[1], hashkey[2],
hashkey[3], hashkey[4], hashkey[5],
hashkey[6], hashkey[7], hashkey[8]);
#else
db_printf("\n\tkey=%08x:%08x:%08x ",
hashkey[0], hashkey[1], hashkey[2]);
#endif
}
#endif /* INET6 */
Spacing.
2014-02-07 10:05:12 +00:00
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
db_printf("hash=%08x idle_time=%03d"
"\n\tfibnum=%02d rt=%p",
fle->f_hash, idle_time, fle->f_fibnum, fle->f_rt);
#ifdef FLOWTABLE_HASH_ALL
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
if (fle->f_flags & FL_STALE)
db_printf(" FL_STALE ");
o Remove at compile time the HASH_ALL code, that was never tested and is unfinished. However, I've tested my version, it works okay. As before it is unfinished: timeout aren't driven by TCP session state. To enable the HASH_ALL mode, one needs in kernel config: options FLOWTABLE_HASH_ALL o Reduce the alignment on flentry to 64 bytes. Without the FLOWTABLE_HASH_ALL option, twice less memory would be consumed by flows. o API to ip_output()/ip6_output() got even more thin: 1 liner. o Remove unused unions. Simply use fle->f_key[]. o Merge all IPv4 code into flowtable_lookup_ipv4(), and do same flowtable_lookup_ipv6(). Stop copying data to on stack sockaddr structures, simply use key[] on stack. o Move code from flowtable_lookup_common() that actually works on insertion into flowtable_insert(). Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-17 11:50:56 +00:00
#endif
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
if (rt_valid) {
if (rt->rt_flags & RTF_UP)
db_printf(" RTF_UP ");
}
if (ifp_valid) {
if (ifp->if_flags & IFF_LOOPBACK)
db_printf(" IFF_LOOPBACK ");
if (ifp->if_flags & IFF_UP)
Spacing.
2014-02-08 22:10:53 +00:00
db_printf(" IFF_UP ");
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
if (ifp->if_flags & IFF_POINTOPOINT)
Spacing.
2014-02-08 22:10:53 +00:00
db_printf(" IFF_POINTOPOINT ");
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
}
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
db_printf("\n");
}
static void
flowtable_show(struct flowtable *ft, int cpuid)
{
int curbit = 0;
bitstr_t *mask, *tmpmask;
- restructure flowtable to support ipv6 - add a name argument to flowtable_alloc for printing with ddb commands - extend ddb commands to print destination address or 4-tuples - don't parse ports in ulp header if FL_HASH_ALL is not passed - add kern_flowtable_insert to enable more generic use of flowtable (e.g. system calls for adding entries) - don't hash loopback addresses - cleanup whitespace - keep statistics per-cpu for per-cpu flowtables to avoid cache line contention - add sysctls to accumulate stats and report aggregate MFC after: 7 days
2010-03-12 05:03:26 +00:00
if (cpuid != -1)
db_printf("cpu: %d\n", cpuid);
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
mask = flowtable_mask_pcpu(ft, cpuid);
tmpmask = ft->ft_tmpmask;
memcpy(tmpmask, mask, ft->ft_size/8);
/*
* XXX Note to self, bit_ffs operates at the byte level
* and thus adds gratuitous overhead
*/
bit_ffs(tmpmask, ft->ft_size, &curbit);
while (curbit != -1) {
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
struct flist *flist;
struct flentry *fle;
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
if (curbit >= ft->ft_size || curbit < -1) {
db_printf("warning: bad curbit value %d \n",
curbit);
break;
}
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
flist = flowtable_list_pcpu(ft, curbit, cpuid);
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
SLIST_FOREACH(fle, flist, f_next)
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
flow_show(ft, fle);
bit_clear(tmpmask, curbit);
bit_ffs(tmpmask, ft->ft_size, &curbit);
}
}
static void
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
flowtable_show_vnet(struct flowtable *ft)
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
{
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
int i;
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
o Axe non-pcpu flowtable implementation. It wasn't enabled or used, and probably is a leftover from first prototyping by Kip. The non-pcpu implementation used mutexes, so it doubtfully worked better than simple routing lookup. o Use UMA_ZONE_PCPU zone for pointers instead of [MAXCPU] arrays, use zpcpu_get() to access data in there. o Substitute own single list implementation with SLIST(). This has two functional side effects: - new flows go into head of a list, before they went to tail. - a bug when incorrect flow was deleted in flow cleaner is fixed. o Due to cache line alignment, there is no reason to keep different zones for IPv4 and IPv6 flows. Both consume one cache line, real size of allocation is equal. o Rely on that f_hash, f_rt, f_lle are stable during fle lifetime, remove useless volatile quilifiers. o More INET/INET6 splitting. Reviewed by: adrian Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-13 04:59:18 +00:00
CPU_FOREACH(i)
flowtable_show(ft, i);
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
}
DB_SHOW_COMMAND(flowtables, db_show_flowtables)
{
VNET_ITERATOR_DECL(vnet_iter);
VNET_FOREACH(vnet_iter) {
CURVNET_SET(vnet_iter);
Print the vnet pointer under DDB when iterating over flowtables of each virtual network stack instance. Sponsored by: ISPsystem [1] Reviewed by: julian [1] MFC after: 1 week [1] Early 2010.
2010-12-31 21:20:32 +00:00
#ifdef VIMAGE
db_printf("vnet %p\n", vnet_iter);
#endif
o Revamp API between flowtable and netinet, netinet6. - ip_output() and ip_output6() simply call flowtable_lookup(), passing mbuf and address family. That's the only code under #ifdef FLOWTABLE in the protocols code now. o Revamp statistics gathering and export. - Remove hand made pcpu stats, and utilize counter(9). - Snapshot of statistics is available via 'netstat -rs'. - All sysctls are moved into net.flowtable namespace, since spreading them over net.inet isn't correct. o Properly separate at compile time INET and INET6 parts. o General cleanup. - Remove chain of multiple flowtables. We simply have one for IPv4 and one for IPv6. - Flowtables are allocated in flowtable.c, symbols are static. - With proper argument to SYSINIT() we no longer need flowtable_ready. - Hash salt doesn't need to be per-VNET. - Removed rudimentary debugging, which use quite useless in dtrace era. The runtime behavior of flowtable shouldn't be changed by this commit. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-02-07 15:18:23 +00:00
#ifdef INET
printf("IPv4:\n");
flowtable_show_vnet(&V_ip4_ft);
#endif
#ifdef INET6
printf("IPv6:\n");
flowtable_show_vnet(&V_ip6_ft);
#endif
- change the interface to flowtable_lookup so that we don't rely on the mbuf for obtaining the fib index - check that a cached flow corresponds to the same fib index as the packet for which we are doing the lookup - at interface detach time flush any flows referencing stale rtentrys associated with the interface that is going away (fixes reported panics) - reduce the time between cleans in case the cleaner is running at the time the eventhandler is called and the wakeup is missed less time will elapse before the eventhandler returns - separate per-vnet initialization from global initialization (pointed out by jeli@) Reviewed by: sam@ Approved by: re@
2009-08-18 20:28:58 +00:00
CURVNET_RESTORE();
}
}
#endif
Reference in New Issue Copy Permalink