methods called Vestigial Time-Wait (VTW) and Maximum Segment Lifetime
Truncation (MSLT).
MSLT and VTW were contributed by Coyote Point Systems, Inc.
Even after a TCP session enters the TIME_WAIT state, its corresponding
socket and protocol control blocks (PCBs) stick around until the TCP
Maximum Segment Lifetime (MSL) expires. On a host whose workload
necessarily creates and closes down many TCP sockets, the sockets & PCBs
for TCP sessions in TIME_WAIT state amount to many megabytes of dead
weight in RAM.
Maximum Segment Lifetimes Truncation (MSLT) assigns each TCP session to
a class based on the nearness of the peer. Corresponding to each class
is an MSL, and a session uses the MSL of its class. The classes are
loopback (local host equals remote host), local (local host and remote
host are on the same link/subnet), and remote (local host and remote
host communicate via one or more gateways). Classes corresponding to
nearer peers have lower MSLs by default: 2 seconds for loopback, 10
seconds for local, 60 seconds for remote. Loopback and local sessions
expire more quickly when MSLT is used.
Vestigial Time-Wait (VTW) replaces a TIME_WAIT session's PCB/socket
dead weight with a compact representation of the session, called a
"vestigial PCB". VTW data structures are designed to be very fast and
memory-efficient: for fast insertion and lookup of vestigial PCBs,
the PCBs are stored in a hash table that is designed to minimize the
number of cacheline visits per lookup/insertion. The memory both
for vestigial PCBs and for elements of the PCB hashtable come from
fixed-size pools, and linked data structures exploit this to conserve
memory by representing references with a narrow index/offset from the
start of a pool instead of a pointer. When space for new vestigial PCBs
runs out, VTW makes room by discarding old vestigial PCBs, oldest first.
VTW cooperates with MSLT.
It may help to think of VTW as a "FIN cache" by analogy to the SYN
cache.
A 2.8-GHz Pentium 4 running a test workload that creates TIME_WAIT
sessions as fast as it can is approximately 17% idle when VTW is active
versus 0% idle when VTW is inactive. It has 103 megabytes more free RAM
when VTW is active (approximately 64k vestigial PCBs are created) than
when it is inactive.
instead of adding/subtracting our own IPv4 header.
There are many benefits: gre(4) needn't grok the outer encapsulation
header any longer, so this simplifies the gre(4) code. The IP
stack needn't grok GRE, so it is simplified, too. gre(4) will
benefit from optimizations in the socket code. Eventually, gre(4)
will gain an IPv6 encapsulation with very few new lines of code.
There is a small performance loss. A 133 MHz, 486-class AMD Elan
sinks/sources a TCP stream over GRE with about 93% the throughput
of the old code. TCP throughput on a 266 MHz, 586-class AMD Geode
is about 96% the throughput of the old code. A 175-MHz ADM5120
(MIPS) only sinks a TCP stream over GRE at about 90% of the old
code; I am still investigating that.
I produced stripped-down versions of sosend() and soreceive() for
gre(4) to use. They are guaranteed not to block, so they can be
called from a software interrupt and from a socket upcall,
respectively.
A kernel thread is no longer necessary for socket transmit/receive,
but I didn't get around to removing it, yet.
Thanks to Matt Thomas for suggesting the use of stripped-down socket
code and software interrupts, and to Andrew Doran for advice and
answers concerning software interrupts, threads, and performance.
routing caused by stale route caches (struct route). Route caches
are sprinkled throughout PCBs, the IP fast-forwarding table, and
IP tunnel interfaces (gre, gif, stf).
Stale IPv6 and ISO route caches will be treated by separate patches.
Thank you to Christoph Badura for suggesting the general approach
to invalidating route caches that I take here.
Here are the details:
Add hooks to struct domain for tracking and for invalidating each
domain's route caches: dom_rtcache, dom_rtflush, and dom_rtflushall.
Introduce helper subroutines, rtflush(ro) for invalidating a route
cache, rtflushall(family) for invalidating all route caches in a
routing domain, and rtcache(ro) for notifying the domain of a new
cached route.
Chain together all IPv4 route caches where ro_rt != NULL. Provide
in_rtcache() for adding a route to the chain. Provide in_rtflush()
and in_rtflushall() for invalidating IPv4 route caches. In
in_rtflush(), set ro_rt to NULL, and remove the route from the
chain. In in_rtflushall(), walk the chain and remove every route
cache.
In rtrequest1(), call rtflushall() to invalidate route caches when
a route is added.
In gif(4), discard the workaround for stale caches that involves
expiring them every so often.
Replace the pattern 'RTFREE(ro->ro_rt); ro->ro_rt = NULL;' with a
call to rtflush(ro).
Update ipflow_fastforward() and all other users of route caches so
that they expect a cached route, ro->ro_rt, to turn to NULL.
Take care when moving a 'struct route' to rtflush() the source and
to rtcache() the destination.
In domain initializers, use .dom_xxx tags.
KNF here and there.
Also, add ioctls SIOCGIFADDRPREF/SIOCSIFADDRPREF to get/set preference
numbers for addresses. Make ifconfig(8) set/display preference
numbers.
To activate source-address selection policies in your kernel, add
'options IPSELSRC' to your kernel configuration.
Miscellaneous changes in support of source-address selection:
1 Factor out some common code, producing rt_replace_ifa().
2 Abbreviate a for-loop with TAILQ_FOREACH().
3 Add the predicates on IPv4 addresses IN_LINKLOCAL() and
IN_PRIVATE(), that are true for link-local unicast
(169.254/16) and RFC1918 private addresses, respectively.
Add the predicate IN_ANY_LOCAL() that is true for link-local
unicast and multicast.
4 Add IPv4-specific interface attach/detach routines,
in_domifattach and in_domifdetach, which build #ifdef
IPSELSRC.
See in_getifa(9) for a more thorough description of source-address
selection policy.
headers and LKM.
Add MKPF; if set to no, don't build and install the pf(4) programs,
headers, LKM and spamd.
Both options default to yes, so nothing changed in the default build.
Reviewed by lukem.
(Sorry for a big commit, I can't separate this into several pieces...)
Pls check sys/netinet6/TODO and sys/netinet6/IMPLEMENTATION for details.
- sys/kern: do not assume single mbuf, accept chained mbuf on passing
data from userland to kernel (or other way round).
- "midway" ATM card: ATM PVC pseudo device support, like those done in ALTQ
package (ftp://ftp.csl.sony.co.jp/pub/kjc/).
- sys/netinet/tcp*: IPv4/v6 dual stack tcp support.
- sys/netinet/{ip6,icmp6}.h, sys/net/pfkeyv2.h: IETF document assumes those
file to be there so we patch it up.
- sys/netinet: IPsec additions are here and there.
- sys/netinet6/*: most of IPv6 code sits here.
- sys/netkey: IPsec key management code
- dev/pci/pcidevs: regen
In my understanding no code here is subject to export control so it
should be safe.
as with user-land programs, include files are installed by each directory
in the tree that has includes to install. (This allows more flexibility
as to what gets installed, makes 'partial installs' easier, and gives us
more options as to which machines' includes get installed at any given
time.) The old SYS_INCLUDES={symlinks,copies} behaviours are _both_
still supported, though at least one bug in the 'symlinks' case is
fixed by this change. Include files can't be build before installation,
so directories that have includes as targets (e.g. dev/pci) have to move
those targets into a different Makefile.
