Commit Graph

378 Commits

Author SHA1 Message Date
Andrew Gallatin
50575ce11c Track TCP connection's NUMA domain in the inpcb
Drivers can now pass up numa domain information via the
mbuf numa domain field.  This information is then used
by TCP syncache_socket() to associate that information
with the inpcb. The domain information is then fed back
into transmitted mbufs in ip{6}_output(). This mechanism
is nearly identical to what is done to track RSS hash values
in the inp_flowid.

Follow on changes will use this information for lacp egress
port selection, binding TCP pacers to the appropriate NUMA
domain, etc.

Reviewed by:	markj, kib, slavash, bz, scottl, jtl, tuexen
Sponsored by:	Netflix
Differential Revision:	https://reviews.freebsd.org/D20028
2019-04-25 15:37:28 +00:00
John Baldwin
43b65e3c98 Don't check the inp socket pointer in in_pcboutput_eagain.
Reviewed by:	hps (by saying it was ok to be removed)
MFC after:	1 month
Sponsored by:	Netflix
2019-03-29 19:47:42 +00:00
Andrey V. Elsukov
c7ee62fcd5 In r335015 PCB destroing was made deferred using epoch_call().
But ipsec_delete_pcbpolicy() uses some VNET-virtualized variables,
and thus it needs VNET context, that is missing during gtaskqueue
executing. Use inp_vnet context to set curvnet in in_pcbfree_deferred().

PR:		235684
MFC after:	1 week
2019-02-13 15:46:05 +00:00
Gleb Smirnoff
a68cc38879 Mechanical cleanup of epoch(9) usage in network stack.
- Remove macros that covertly create epoch_tracker on thread stack. Such
  macros a quite unsafe, e.g. will produce a buggy code if same macro is
  used in embedded scopes. Explicitly declare epoch_tracker always.

- Unmask interface list IFNET_RLOCK_NOSLEEP(), interface address list
  IF_ADDR_RLOCK() and interface AF specific data IF_AFDATA_RLOCK() read
  locking macros to what they actually are - the net_epoch.
  Keeping them as is is very misleading. They all are named FOO_RLOCK(),
  while they no longer have lock semantics. Now they allow recursion and
  what's more important they now no longer guarantee protection against
  their companion WLOCK macros.
  Note: INP_HASH_RLOCK() has same problems, but not touched by this commit.

This is non functional mechanical change. The only functionally changed
functions are ni6_addrs() and ni6_store_addrs(), where we no longer enter
epoch recursively.

Discussed with:	jtl, gallatin
2019-01-09 01:11:19 +00:00
Mateusz Guzik
cc426dd319 Remove unused argument to priv_check_cred.
Patch mostly generated with cocinnelle:

@@
expression E1,E2;
@@

- priv_check_cred(E1,E2,0)
+ priv_check_cred(E1,E2)

Sponsored by:	The FreeBSD Foundation
2018-12-11 19:32:16 +00:00
Mark Johnston
9d2877fc3d Clamp the INPCB port hash tables to IPPORT_MAX + 1 chains.
Memory beyond that limit was previously unused, wasting roughly 1MB per
8GB of RAM.  Also retire INP_PCBLBGROUP_PORTHASH, which was identical to
INP_PCBPORTHASH.

Reviewed by:	glebius
MFC after:	2 weeks
Differential Revision:	https://reviews.freebsd.org/D17803
2018-12-05 17:06:00 +00:00
Mark Johnston
79db6fe7aa Plug some networking sysctl leaks.
Various network protocol sysctl handlers were not zero-filling their
output buffers and thus would export uninitialized stack memory to
userland.  Fix a number of such handlers.

Reported by:	Thomas Barabosch, Fraunhofer FKIE
Reviewed by:	tuexen
MFC after:	3 days
Security:	kernel memory disclosure
Sponsored by:	The FreeBSD Foundation
Differential Revision:	https://reviews.freebsd.org/D18301
2018-11-22 20:49:41 +00:00
Mark Johnston
d9ff5789be Remove redundant checks for a NULL lbgroup table.
No functional change intended.

MFC after:	2 weeks
Sponsored by:	The FreeBSD Foundation
Differential Revision:	https://reviews.freebsd.org/D17108
2018-11-01 15:52:49 +00:00
Mark Johnston
79ee680b65 Improve style in in_pcbinslbgrouphash() and related subroutines.
No functional change intended.

MFC after:	2 weeks
Sponsored by:	The FreeBSD Foundation
Differential Revision:	https://reviews.freebsd.org/D17107
2018-11-01 15:51:49 +00:00
Mark Johnston
54af3d0dac Fix synchronization of LB group access.
Lookups are protected by an epoch section, so the LB group linkage must
be a CK_LIST rather than a plain LIST.  Furthermore, we were not
deferring LB group frees, so in_pcbremlbgrouphash() could race with
readers and cause a use-after-free.

Reviewed by:	sbruno, Johannes Lundberg <johalun0@gmail.com>
Tested by:	gallatin
Approved by:	re (gjb)
Sponsored by:	The FreeBSD Foundation
Differential Revision:	https://reviews.freebsd.org/D17031
2018-09-10 19:00:29 +00:00
Mark Johnston
a7026c7fd9 Use ratecheck(9) in in_pcbinslbgrouphash().
Reviewed by:	bz, Johannes Lundberg <johalun0@gmail.com>
Approved by:	re (kib)
Sponsored by:	The FreeBSD Foundation
Differential revision:	https://reviews.freebsd.org/D17065
2018-09-07 21:11:41 +00:00
Mark Johnston
8be02ee4da Fix style bugs in in_pcblookup_lbgroup().
No functional change intended.

Reviewed by:	bz, Johannes Lundberg <johalun0@gmail.com>
Approved by:	re (rgrimes)
Sponsored by:	The FreeBSD Foundation
Differential Revision:	https://reviews.freebsd.org/D17030
2018-09-05 15:04:11 +00:00
Mark Johnston
73ad0b6abf Use the correct malloc type in in_pcblbgroup_free().
Approved by:	re (kib)
Sponsored by:	The FreeBSD Foundation
2018-09-03 17:39:09 +00:00
Matt Macy
f9be038601 Fix in6_multi double free
This is actually several different bugs:
- The code is not designed to handle inpcb deletion after interface deletion
  - add reference for inpcb membership
- The multicast address has to be removed from interface lists when the refcount
  goes to zero OR when the interface goes away
  - decouple list disconnect from refcount (v6 only for now)
- ifmultiaddr can exist past being on interface lists
  - add flag for tracking whether or not it's enqueued
- deferring freeing moptions makes the incpb cleanup code simpler but opens the
  door wider still to races
  - call inp_gcmoptions synchronously after dropping the the inpcb lock

Fundamentally multicast needs a rewrite - but keep applying band-aids for now.

Tested by: kp
Reported by: novel, kp, lwhsu
2018-08-15 20:23:08 +00:00
Andrew Turner
5f901c92a8 Use the new VNET_DEFINE_STATIC macro when we are defining static VNET
variables.

Reviewed by:	bz
Sponsored by:	DARPA, AFRL
Differential Revision:	https://reviews.freebsd.org/D16147
2018-07-24 16:35:52 +00:00
Brooks Davis
3a20f06a1c Use uintptr_t alone when assigning to kvaddr_t variables.
Suggested by:	jhb
2018-07-10 13:03:06 +00:00
Brooks Davis
7524b4c14b Correct breakage on 32-bit platforms from r335979. 2018-07-06 10:03:33 +00:00
Brooks Davis
f38b68ae8a Make struct xinpcb and friends word-size independent.
Replace size_t members with ksize_t (uint64_t) and pointer members
(never used as pointers in userspace, but instead as unique
idenitifiers) with kvaddr_t (uint64_t). This makes the structs
identical between 32-bit and 64-bit ABIs.

On 64-bit bit systems, the ABI is maintained. On 32-bit systems,
this is an ABI breaking change. The ABI of most of these structs
was previously broken in r315662.  This also imposes a small API
change on userspace consumers who must handle kernel pointers
becoming virtual addresses.

PR:		228301 (exp-run by antoine)
Reviewed by:	jtl, kib, rwatson (various versions)
Sponsored by:	DARPA, AFRL
Differential Revision:	https://reviews.freebsd.org/D15386
2018-07-05 13:13:48 +00:00
Matt Macy
6573d7580b epoch(9): allow preemptible epochs to compose
- Add tracker argument to preemptible epochs
- Inline epoch read path in kernel and tied modules
- Change in_epoch to take an epoch as argument
- Simplify tfb_tcp_do_segment to not take a ti_locked argument,
  there's no longer any benefit to dropping the pcbinfo lock
  and trying to do so just adds an error prone branchfest to
  these functions
- Remove cases of same function recursion on the epoch as
  recursing is no longer free.
- Remove the the TAILQ_ENTRY and epoch_section from struct
  thread as the tracker field is now stack or heap allocated
  as appropriate.

Tested by: pho and Limelight Networks
Reviewed by: kbowling at llnw dot com
Sponsored by: Limelight Networks
Differential Revision: https://reviews.freebsd.org/D16066
2018-07-04 02:47:16 +00:00
Sean Bruno
45fc0718d8 Reap unused variable and assignment that had no effect. Noted by cross
compiling with gcc on mips.

Reviewed by:	mmacy
2018-06-24 21:36:37 +00:00
Matt Macy
3d348772e7 in_pcblookup_hash: validate inp before return
Post r335356 it is possible to have an inpcb on the hash lists that is
partially torn down. Validate before using. Also as a side effect of this
change the lock ordering issue between hash lock and inpcb no longer exists
allowing some simplification.

Reported by:	pho@
2018-06-21 18:40:15 +00:00
Matt Macy
feeef8509b Fix PCBGROUPS build post CK conversion of pcbinfo 2018-06-13 23:19:54 +00:00
Matt Macy
483305b99c Handle INP_FREED when looking up an inpcb
When hash table lookups are not serialized with in_pcbfree it will be
possible for callers to find an inpcb that has been marked free. We
need to check for this and return NULL.
2018-06-13 04:23:49 +00:00
Matt Macy
700e893c34 Defer inpcbport free in in_pcbremlists as well 2018-06-12 23:26:25 +00:00
Matt Macy
f09ee4fc01 Defer inpcbport free until after a grace period has elapsed
This is a dependency for inpcbinfo rlock conversion to epoch
2018-06-12 22:18:27 +00:00
Matt Macy
b872626dbe mechanical CK macro conversion of inpcbinfo lists
This is a dependency for converting the inpcbinfo hash and info rlocks
to epoch.
2018-06-12 22:18:20 +00:00
Matt Macy
addf2b2009 Defer inpcb deletion until after a grace period has elapsed
Deferring the actual free of the inpcb until after a grace
period has elapsed will allow us to convert the inpcbinfo
info and hash read locks to epoch.

Reviewed by: gallatin, jtl
Sponsored by: Limelight Networks
Differential Revision: https://reviews.freebsd.org/D15510
2018-06-12 22:18:15 +00:00
Sean Bruno
1a43cff92a Load balance sockets with new SO_REUSEPORT_LB option.
This patch adds a new socket option, SO_REUSEPORT_LB, which allow multiple
programs or threads to bind to the same port and incoming connections will be
load balanced using a hash function.

Most of the code was copied from a similar patch for DragonflyBSD.

However, in DragonflyBSD, load balancing is a global on/off setting and can not
be set per socket. This patch allows for simultaneous use of both the current
SO_REUSEPORT and the new SO_REUSEPORT_LB options on the same system.

Required changes to structures:
Globally change so_options from 16 to 32 bit value to allow for more options.
Add hashtable in pcbinfo to hold all SO_REUSEPORT_LB sockets.

Limitations:
As DragonflyBSD, a load balance group is limited to 256 pcbs (256 programs or
threads sharing the same socket).

This is a substantially different contribution as compared to its original
incarnation at svn r332894 and reverted at svn r332967.  Thanks to rwatson@
for the substantive feedback that is included in this commit.

Submitted by:	Johannes Lundberg <johalun0@gmail.com>
Obtained from:	DragonflyBSD
Relnotes:	Yes
Sponsored by:	Limelight Networks
Differential Revision:	https://reviews.freebsd.org/D11003
2018-06-06 15:45:57 +00:00
Matt Macy
62d733a11d in_pcbladdr: remove debug code that snuck in with ifa epoch conversion r334118 2018-05-27 06:47:09 +00:00
Matt Macy
4f6c66cc9c UDP: further performance improvements on tx
Cumulative throughput while running 64
  netperf -H $DUT -t UDP_STREAM -- -m 1
on a 2x8x2 SKL went from 1.1Mpps to 2.5Mpps

Single stream throughput increases from 910kpps to 1.18Mpps

Baseline:
https://people.freebsd.org/~mmacy/2018.05.11/udpsender2.svg

- Protect read access to global ifnet list with epoch
https://people.freebsd.org/~mmacy/2018.05.11/udpsender3.svg

- Protect short lived ifaddr references with epoch
https://people.freebsd.org/~mmacy/2018.05.11/udpsender4.svg

- Convert if_afdata read lock path to epoch
https://people.freebsd.org/~mmacy/2018.05.11/udpsender5.svg

A fix for the inpcbhash contention is pending sufficient time
on a canary at LLNW.

Reviewed by:	gallatin
Sponsored by:	Limelight Networks
Differential Revision:	https://reviews.freebsd.org/D15409
2018-05-23 21:02:14 +00:00
Matt Macy
f42a83f2a6 inpcb: revert deferred inpcb free pending further review 2018-05-21 16:13:43 +00:00
Matt Macy
1a3d880c26 in(s)_moptions: free before tearing down inpcb 2018-05-20 20:08:21 +00:00
Matt Macy
47d2a58560 inpcb: defer destruction of inpcb until after a grace period has elapsed
in_pcbfree will remove the incpb from the list and release the rtentry
while the vnet is set, but the actual destruction will be deferred
until any threads in a (not yet used) epoch section, no longer potentially
have references.
2018-05-20 04:38:04 +00:00
Matt Macy
ddece765f3 in_pcb: add helper for deferring inpcb rele calls from list functions 2018-05-20 02:17:30 +00:00
Matt Macy
cb6bb2303e ip(6)_freemoptions: defer imo destruction to epoch callback task
Avoid the ugly unlock / lock of the inpcbinfo where we need to
figure out what kind of lock we hold by simply deferring the
operation to another context. (Also a small dependency for
converting the pcbinfo read lock to epoch)
2018-05-20 00:22:28 +00:00
Matt Macy
d7c5a620e2 ifnet: Replace if_addr_lock rwlock with epoch + mutex
Run on LLNW canaries and tested by pho@

gallatin:
Using a 14-core, 28-HTT single socket E5-2697 v3 with a 40GbE MLX5
based ConnectX 4-LX NIC, I see an almost 12% improvement in received
packet rate, and a larger improvement in bytes delivered all the way
to userspace.

When the host receiving 64 streams of netperf -H $DUT -t UDP_STREAM -- -m 1,
I see, using nstat -I mce0 1 before the patch:

InMpps OMpps  InGbs  OGbs err TCP Est %CPU syscalls csw     irq GBfree
4.98   0.00   4.42   0.00 4235592     33   83.80 4720653 2149771   1235 247.32
4.73   0.00   4.20   0.00 4025260     33   82.99 4724900 2139833   1204 247.32
4.72   0.00   4.20   0.00 4035252     33   82.14 4719162 2132023   1264 247.32
4.71   0.00   4.21   0.00 4073206     33   83.68 4744973 2123317   1347 247.32
4.72   0.00   4.21   0.00 4061118     33   80.82 4713615 2188091   1490 247.32
4.72   0.00   4.21   0.00 4051675     33   85.29 4727399 2109011   1205 247.32
4.73   0.00   4.21   0.00 4039056     33   84.65 4724735 2102603   1053 247.32

After the patch

InMpps OMpps  InGbs  OGbs err TCP Est %CPU syscalls csw     irq GBfree
5.43   0.00   4.20   0.00 3313143     33   84.96 5434214 1900162   2656 245.51
5.43   0.00   4.20   0.00 3308527     33   85.24 5439695 1809382   2521 245.51
5.42   0.00   4.19   0.00 3316778     33   87.54 5416028 1805835   2256 245.51
5.42   0.00   4.19   0.00 3317673     33   90.44 5426044 1763056   2332 245.51
5.42   0.00   4.19   0.00 3314839     33   88.11 5435732 1792218   2499 245.52
5.44   0.00   4.19   0.00 3293228     33   91.84 5426301 1668597   2121 245.52

Similarly, netperf reports 230Mb/s before the patch, and 270Mb/s after the patch

Reviewed by:	gallatin
Sponsored by:	Limelight Networks
Differential Revision:	https://reviews.freebsd.org/D15366
2018-05-18 20:13:34 +00:00
Matt Macy
28c001002a Currently in_pcbfree will unconditionally wunlock the pcbinfo lock
to avoid a LOR on the multicast list lock in the freemoptions routines.
As it turns out, tcp_usr_detach can acquire the tcbinfo lock readonly.
Trying to wunlock the pcbinfo lock in that context has caused a number
of reported crashes.

This change unclutters in_pcbfree and moves the handling of wunlock vs
runlock of pcbinfo to the freemoptions routine.

Reported by:	mjg@, bde@, o.hartmann at walstatt.org
Approved by:	sbruno
2018-05-05 22:40:40 +00:00
Stephen Hurd
f3e1324b41 Separate list manipulation locking from state change in multicast
Multicast incorrectly calls in to drivers with a mutex held causing drivers
to have to go through all manner of contortions to use a non sleepable lock.
Serialize multicast updates instead.

Submitted by:	mmacy <mmacy@mattmacy.io>
Reviewed by:	shurd, sbruno
Sponsored by:	Limelight Networks
Differential Revision:	https://reviews.freebsd.org/D14969
2018-05-02 19:36:29 +00:00
Sean Bruno
7875017ca9 Revert r332894 at the request of the submitter.
Submitted by:	Johannes Lundberg <johalun0_gmail.com>
Sponsored by:	Limelight Networks
2018-04-24 19:55:12 +00:00
Sean Bruno
7b7796eea5 Load balance sockets with new SO_REUSEPORT_LB option
This patch adds a new socket option, SO_REUSEPORT_LB, which allow multiple
programs or threads to bind to the same port and incoming connections will be
load balanced using a hash function.

Most of the code was copied from a similar patch for DragonflyBSD.

However, in DragonflyBSD, load balancing is a global on/off setting and can not
be set per socket. This patch allows for simultaneous use of both the current
SO_REUSEPORT and the new SO_REUSEPORT_LB options on the same system.

Required changes to structures
Globally change so_options from 16 to 32 bit value to allow for more options.
Add hashtable in pcbinfo to hold all SO_REUSEPORT_LB sockets.

Limitations
As DragonflyBSD, a load balance group is limited to 256 pcbs
(256 programs or threads sharing the same socket).

Submitted by:	Johannes Lundberg <johanlun0@gmail.com>
Sponsored by:	Limelight Networks
Differential Revision:	https://reviews.freebsd.org/D11003
2018-04-23 19:51:00 +00:00
Randall Stewart
3ee9c3c4eb This commit brings in the TCP high precision timer system (tcp_hpts).
It is the forerunner/foundational work of bringing in both Rack and BBR
which use hpts for pacing out packets. The feature is optional and requires
the TCPHPTS option to be enabled before the feature will be active. TCP
modules that use it must assure that the base component is compile in
the kernel in which they are loaded.

MFC after:	Never
Sponsored by:	Netflix Inc.
Differential Revision:	https://reviews.freebsd.org/D15020
2018-04-19 13:37:59 +00:00
Jonathan T. Looney
45a48d07c3 Check that in_pcbfree() is only called once for each PCB. If that
assumption is violated, "bad things" could follow.

I believe such an assert would have detected some of the problems jch@
was chasing in PR 203175 (see r307551).  We also use it in our internal
TCP development efforts.  And, in case a bug does slip through to
released code, this change silently ignores subsequent calls to
in_pcbfree().

Reviewed by:	rrs
Sponsored by:	Netflix, Inc.
Differential Revision:	https://reviews.freebsd.org/D14990
2018-04-06 16:48:11 +00:00
Navdeep Parhar
b9a9e8e9bd Fix RSS build (broken in r331309).
Sponsored by:	Chelsio Communications
2018-03-29 19:48:17 +00:00
Jonathan T. Looney
7fb2986ff6 If the INP lock is uncontested, avoid taking a reference and jumping
through the lock-switching hoops.

A few of the INP lookup operations that lock INPs after the lookup do
so using this mechanism (to maintain lock ordering):

1. Lock lookup structure.
2. Find INP.
3. Acquire reference on INP.
4. Drop lock on lookup structure.
5. Acquire INP lock.
6. Drop reference on INP.

This change provides a slightly shorter path for cases where the INP
lock is uncontested:

1. Lock lookup structure.
2. Find INP.
3. Try to acquire the INP lock.
4. If successful, drop lock on lookup structure.

Of course, if the INP lock is contested, the functions will need to
revert to the previous way of switching locks safely.

This saves a few atomic operations when the INP lock is uncontested.

Discussed with:	gallatin, rrs, rwatson
MFC after:	2 weeks
Sponsored by:	Netflix, Inc.
Differential Revision:	https://reviews.freebsd.org/D12911
2018-03-21 15:54:46 +00:00
Ryan Stone
fc21c53f63 Reduce code duplication for inpcb route caching
Add a new macro to clear both the L3 and L2 route caches, to
hopefully prevent future instances where only the L3 cache was
cleared when both should have been.

MFC after:	1 week
Sponsored by:	Dell EMC Isilon
Differential Revision:	https://reviews.freebsd.org/D13989
Reviewed by:	karels
2018-01-23 03:15:39 +00:00
Pedro F. Giffuni
51369649b0 sys: further adoption of SPDX licensing ID tags.
Mainly focus on files that use BSD 3-Clause license.

The Software Package Data Exchange (SPDX) group provides a specification
to make it easier for automated tools to detect and summarize well known
opensource licenses. We are gradually adopting the specification, noting
that the tags are considered only advisory and do not, in any way,
superceed or replace the license texts.

Special thanks to Wind River for providing access to "The Duke of
Highlander" tool: an older (2014) run over FreeBSD tree was useful as a
starting point.
2017-11-20 19:43:44 +00:00
Hans Petter Selasky
95ed5015ec Add support for generic backpressure indicator for ratelimited
transmit queues aswell as non-ratelimited ones.

Add the required structure bits in order to support a backpressure
indication with ratelimited connections aswell as non-ratelimited
ones. The backpressure indicator is a value between zero and 65535
inclusivly, indicating if the destination transmit queue is empty or
full respectivly. Applications can use this value as a decision point
for when to stop transmitting data to avoid endless ENOBUFS error
codes upon transmitting an mbuf. This indicator is also useful to
reduce the latency for ratelimited queues.

Reviewed by:		gallatin, kib, gnn
Differential Revision:	https://reviews.freebsd.org/D11518
Sponsored by:		Mellanox Technologies
2017-09-06 13:56:18 +00:00
Gleb Smirnoff
6a6cefac3d o Rearrange struct inpcb fields to optimize the TCP output code path
considering cache line hits and misses.  Put the lock and hash list
  glue into the first cache line, put inp_refcount inp_flags inp_socket
  into the second cache line.
o On allocation zero out entire structure except the lock and list entries,
  including inp_route inp_lle inp_gencnt.  When inp_route and inp_lle were
  introduced, they were added below inp_zero_size, resulting on not being
  cleared after free/alloc.  This definitely was a source of bugs with route
  caching.  Could be that r315956 has just fixed one of them.
  The inp_gencnt is reinitialized on every alloc, so it is safe to clear it.

This has been proved to improve TCP performance at Netflix.

Obtained from:		rrs
Differential Revision:	D10686
2017-05-24 17:47:16 +00:00
Gleb Smirnoff
cc487c1697 Reduce in_pcbinfo_init() by two params. No users supply any flags to this
function (they used to say UMA_ZONE_NOFREE), so flag parameter goes away.
The zone_fini parameter also goes away.  Previously no protocols (except
divert) supplied zone_fini function, so inpcb locks were leaked with slabs.
This was okay while zones were allocated with UMA_ZONE_NOFREE flag, but now
this is a leak.  Fix that by suppling inpcb_fini() function as fini method
for all inpcb zones.
2017-05-15 21:58:36 +00:00
Mike Karels
8c1960d506 Fix reference count leak with L2 caching.
ip_forward, TCP/IPv6, and probably SCTP leaked references to L2 cache
entry because they used their own routes on the stack, not in_pcb routes.
The original model for route caching was callers that provided a route
structure to ip{,6}input() would keep the route, and this model was used
for L2 caching as well. Instead, change L2 caching to be done by default
only when using a route structure in the in_pcb; the pcb deallocation
code frees L2 as well as L3 cacches. A separate change will add route
caching to TCP/IPv6.

Another suggestion was to have the transport protocols indicate willingness
to use L2 caching, but this approach keeps the changes in the network
level

Reviewed by:    ae gnn
MFC after:      2 weeks
Differential Revision:  https://reviews.freebsd.org/D10059
2017-03-25 15:06:28 +00:00