suggested by RFC 6675.
Currently differnt places in the stack tries to guess this in suboptimal ways.
The main problem is that current calculations don't take sacked bytes into
account. Sacked bytes are the bytes receiver acked via SACK option. This is
suboptimal because it assumes that network has more outstanding (unacked) bytes
than the actual value and thus sends less data by setting congestion window
lower than what's possible which in turn may cause slower recovery from losses.
As an example, one of the current calculations looks something like this:
snd_nxt - snd_fack + sackhint.sack_bytes_rexmit
New proposal from RFC 6675 is:
snd_max - snd_una - sackhint.sacked_bytes + sackhint.sack_bytes_rexmit
which takes sacked bytes into account which is a new addition to the sackhint
struct. Only thing we are missing from RFC 6675 is isLost() i.e. segment being
considered lost and thus adjusting pipe based on that which makes this
calculation a bit on conservative side.
The approach is very simple. We already process each ack with sack info in
tcp_sack_doack() and extract sack blocks/holes out of it. We'd now also track
this new variable sacked_bytes which keeps track of total sacked bytes reported.
One downside to this approach is that we may get incorrect count of sacked_bytes
if the other end decides to drop sack info in the ack because of memory pressure
or some other reasons. But in this (not very likely) case also the pipe
calculation would be conservative which is okay as opposed to being aggressive
in sending packets into the network.
Next step is to use this more accurate pipe estimation to drive congestion
window adjustments.
In collaboration with: rrs
Reviewed by: jason_eggnet dot com, rrs
MFC after: 2 weeks
Sponsored by: Limelight Networks
Differential Revision: https://reviews.freebsd.org/D3971
window in number of segments on fly. It is set to 10 segments by default.
Remove net.inet.tcp.experimental.initcwnd10 which is now redundant. Also remove
the parent node net.inet.tcp.experimental as it's not needed anymore and also
because it was not well thought out.
Differential Revision: https://reviews.freebsd.org/D3858
In collaboration with: lstewart
Reviewed by: gnn (prev version), rwatson, allanjude, wblock (man page)
MFC after: 2 weeks
Relnotes: yes
Sponsored by: Limelight Networks
degradation (7%) for host host TCP connections over 10Gbps links,
even when there were no secuirty policies in place. There is no
change in performance on 1Gbps network links. Testing GENERIC vs.
GENERIC-NOIPSEC vs. GENERIC with this change shows that the new
code removes any overhead introduced by having IPSEC always in the
kernel.
Differential Revision: D3993
MFC after: 1 month
Sponsored by: Rubicon Communications (Netgate)
down state.
Regression appeared in r287789, where the "prefix has no corresponding
installed route" case was forgotten. Additionally, lltable_delete_addr()
was called with incorrect byte order (default is network for lltable code).
While here, improve comments on given cases and byte order.
PR: 203573
Submitted by: phk
retransmission timeout (rto) when blackhole detection is enabled. Make
sure it only happens when the second attempt to send the same segment also fails
with rto.
Also make sure that each mtu probing stage (usually 1448 -> 1188 -> 524) follows
the same pattern and gets 2 chances (rto) before further clamping down.
Note: RFC4821 doesn't specify implementation details on how this situation
should be handled.
Differential Revision: https://reviews.freebsd.org/D3434
Reviewed by: sbruno, gnn (previous version)
MFC after: 2 weeks
Sponsored by: Limelight Networks
packets and/or state transitions from each TCP socket. That would help with
narrowing down certain problems we see in the field that are hard to reproduce
without understanding the history of how we got into a certain state. This
change provides just that.
It saves copies of the last N packets in a list in the tcpcb. When the tcpcb is
destroyed, the list is freed. I thought this was likely to be more
performance-friendly than saving copies of the tcpcb. Plus, with the packets,
you should be able to reverse-engineer what happened to the tcpcb.
To enable the feature, you will need to compile a kernel with the TCPPCAP
option. Even then, the feature defaults to being deactivated. You can activate
it by setting a positive value for the number of captured packets. You can do
that on either a global basis or on a per-socket basis (via a setsockopt call).
There is no way to get the packets out of the kernel other than using kmem or
getting a coredump. I thought that would help some of the legal/privacy concerns
regarding such a feature. However, it should be possible to add a future effort
to export them in PCAP format.
I tested this at low scale, and found that there were no mbuf leaks and the peak
mbuf usage appeared to be unchanged with and without the feature.
The main performance concern I can envision is the number of mbufs that would be
used on systems with a large number of sockets. If you save five packets per
direction per socket and have 3,000 sockets, that will consume at least 30,000
mbufs just to keep these packets. I tried to reduce the concerns associated with
this by limiting the number of clusters (not mbufs) that could be used for this
feature. Again, in my testing, that appears to work correctly.
Differential Revision: D3100
Submitted by: Jonathan Looney <jlooney at juniper dot net>
Reviewed by: gnn, hiren
unconditionally, we end up with an mbuf chain of two mbufs, which
later in in_arpreply() is rewritten from ARP request to ARP reply
and is sent out. Looks like igb(4) (at least mine, and at least
at my network) fails on such mbuf chain, so ARP reply doesn't go
out wire. Thus, make the m_pullup() call conditional, as it is
everywhere. Of course, the bug in igb(?) should be investigated,
but better first fix the head. And unconditional m_pullup() was
suboptimal, anyway.
Currently we perform crypto requests for IPSEC synchronous for most of
crypto providers (software, aesni) and only VIA padlock calls crypto
callback asynchronous. In synchronous mode it is possible, that security
policy will be removed during the processing crypto request. And crypto
callback will release the last reference to SP. Then upon return into
ipsec[46]_process_packet() IPSECREQUEST_UNLOCK() will be called to already
freed request. To prevent this we will take extra reference to SP.
PR: 201876
Sponsored by: Yandex LLC
is smaller than the current one for this connection. This is behavior
specified by RFC 1191, and this is how original BSD stack behaved, but this
was unintentionally regressed in r182851.
Reported & tested by: Richard Russo <russor whatsapp.com>
Differential Revision: D3567
Sponsored by: Nginx, Inc.
Initially function was introduced in r53541 (KAME initial commit) to
"provide hints from upper layer protocols that indicate a connection
is making "forward progress"" (quote from RFC 2461 7.3.1 Reachability
Confirmation).
However, it was converted to do nothing (e.g. just return) in r122922
(tcp_hostcache implementation) back in 2003. Some defines were moved
to tcp_var.h in r169541. Then, it was broken (for non-corner cases)
by r186119 (L2<>L3 split) in 2008 (NULL ifp in nd6_lookup). So,
right now this code is broken and has no "real" base users.
Differential Revision: https://reviews.freebsd.org/D3699
instead of old "ignore-and-return 0" in r287789. This broke arp -da /
ndp -cn behavior (they exit on rtsock command failure). Fix this by
translating LLE_IFADDR to RTM_PINNED flag, passing it to userland and
making arp/ndp ignore these entries in batched delete.
MFC after: 2 weeks
function. The change is mostly mechanical with the following exception:
Last piece of nd6_resolve_slow() was refactored: ND6_LLINFO_PERMANENT
condition was removed as always-true, explicit ND6_LLINFO_NOSTATE ->
ND6_LLINFO_INCOMPLETE state transition was removed as duplicate.
Reviewed by: ae
Sponsored by: Yandex LLC
* prepare gateway before insertion
* use RTM_CHANGE instead of explicit find/change route
* Remove fib argument from ifa_switch_loopback_route added in r264887:
if old ifp fib differes from new one, that the caller
is doing something wrong
* Make ifa_*_loopback_route call single ifa_maintain_loopback_route().
without holding afdata wlock
* convert per-af delete_address callback to global lltable_delete_entry() and
more low-level "delete this lle" per-af callback
* fix some bugs/inconsistencies in IPv4/IPv6 ifscrub procedures
Sponsored by: Yandex LLC
Differential Revision: https://reviews.freebsd.org/D3573
To make driver programming easier the TSO limits are changed to
reflect the values used in the BUSDMA tag a network adapter driver is
using. The TCP/IP network stack will subtract space for all linklevel
and protocol level headers and ensure that the full mbuf chain passed
to the network adapter fits within the given limits.
Implementation notes:
If a network adapter driver needs to fixup the first mbuf in order to
support VLAN tag insertion, the size of the VLAN tag should be
subtracted from the TSO limit. Else not.
Network adapters which typically inline the complete header mbuf could
technically transmit one more segment. This patch does not implement a
mechanism to recover the last segment for data transmission. It is
believed when sufficiently large mbuf clusters are used, the segment
limit will not be reached and recovering the last segment will not
have any effect.
The current TSO algorithm tries to send MTU-sized packets, where the
MTU typically is 1500 bytes, which gives 1448 bytes of TCP data
payload per packet for IPv4. That means if the TSO length limitiation
is set to 65536 bytes, there will be a data payload remainder of
(65536 - 1500) mod 1448 bytes which is equal to 324 bytes. Trying to
recover total TSO length due to inlining mbuf header data will not
have any effect, because adding or removing the ETH/IP/TCP headers
to or from 324 bytes will not cause more or less TCP payload to be
TSO'ed.
Existing network adapter limits will be updated separately.
Differential Revision: https://reviews.freebsd.org/D3458
Reviewed by: rmacklem
MFC after: 2 weeks
to provide the TCPDEBUG functionality with pure DTrace.
Reviewed by: rwatson
MFC after: 2 weeks
Sponsored by: Limelight Networks
Differential Revision: D3530
in the routine, which queues an ERROR chunk, instead on relyinh
on the callers to do so. Since one caller missed this, this actially
fixes a bug.
MFC after: 1 week
because the RSS hash may need to be recalculated.
Submitted by: Tiwei Bie <btw@mail.ustc.edu.cn>
Differential Revision: https://reviews.freebsd.org/D3564
o Unlike xor, in Jenkins hash every bit of input affects virtually
every bit of output, thus salting the hash actually works. With
xor salting only provides a false sense of security, since if
hash(x) collides with hash(y), then of course, hash(x) ^ salt
would also collide with hash(y) ^ salt. [1]
o Jenkins provides much better distribution than xor, very close to
ideal.
TCP connection setup/teardown benchmark has shown a 10% increase
with default hash size, and with bigger hashes that still provide
possibility for collisions. With enormous hash size, when dataset is
by an order of magnitude smaller than hash size, the benchmark has
shown 4% decrease in performance decrease, which is expected and
acceptable.
Noticed by: Jeffrey Knockel <jeffk cs.unm.edu> [1]
Benchmarks by: jch
Reviewed by: jch, pkelsey, delphij
Security: strengthens protection against hash collision DoS
Sponsored by: Nginx, Inc.
workaround for a callout(9) issue, it turns out it is instead the right
way to use callout in mpsafe mode without using callout_drain().
r284245 commit message:
Fix a callout race condition introduced in TCP timers callouts with r281599.
In TCP timer context, it is not enough to check callout_stop() return value
to decide if a callout is still running or not, previous callout_reset()
return values have also to be checked.
Differential Revision: https://reviews.freebsd.org/D2763
