other end till it reaches predetermined threshold which is 3 for us right now.
Once that happens, we trigger fast-retransmit to do loss recovery.
Main problem with the current implementation is that we don't honor SACK
information well to detect whether an incoming ack is a dupack or not. RFC6675
has latest recommendations for that. According to it, dupack is a segment that
arrives carrying a SACK block that identifies previously unknown information
between snd_una and snd_max even if it carries new data, changes the advertised
window, or moves the cumulative acknowledgment point.
With the prevalence of Selective ACK (SACK) these days, improper handling can
lead to delayed loss recovery.
With the fix, new behavior looks like following:
0) th_ack < snd_una --> ignore
Old acks are ignored.
1) th_ack == snd_una, !sack_changed --> ignore
Acks with SACK enabled but without any new SACK info in them are ignored.
2) th_ack == snd_una, window == old_window --> increment
Increment on a good dupack.
3) th_ack == snd_una, window != old_window, sack_changed --> increment
When SACK enabled, it's okay to have advertized window changed if the ack has
new SACK info.
4) th_ack > snd_una --> reset to 0
Reset to 0 when left edge moves.
5) th_ack > snd_una, sack_changed --> increment
Increment if left edge moves but there is new SACK info.
Here, sack_changed is the indicator that incoming ack has previously unknown
SACK info in it.
Note: This fix is not fully compliant to RFC6675. That may require a few
changes to current implementation in order to keep per-sackhole dupack counter
and change to the way we mark/handle sack holes.
PR: 203663
Reviewed by: jtl
MFC after: 3 weeks
Sponsored by: Limelight Networks
Differential Revision: https://reviews.freebsd.org/D4225
Vast majority of rtalloc(9) users require only basic info from
route table (e.g. "does the rtentry interface match with the interface
I have?". "what is the MTU?", "Give me the IPv4 source address to use",
etc..).
Instead of hand-rolling lookups, checking if rtentry is up, valid,
dealing with IPv6 mtu, finding "address" ifp (almost never done right),
provide easy-to-use API hiding all the complexity and returning the
needed info into small on-stack structure.
This change also helps hiding route subsystem internals (locking, direct
rtentry accesses).
Additionaly, using this API improves lookup performance since rtentry is not
locked.
(This is safe, since all the rtentry changes happens under both radix WLOCK
and rtentry WLOCK).
Sponsored by: Yandex LLC
* When processing a cookie, use the number of
streams announced in the INIT-ACK.
* When sending an INIT-ACK for an existing
association, use the value from the association,
not from the end-point.
MFC after: 1 week
LLE structure is mostly unchanged during its lifecycle.
To be more specific, there are 2 things relevant for fast path
lookup code:
1) link-level address change. Since r286722, these updates are performed
under AFDATA WLOCK.
2) Some sort of feedback indicating that this particular entry is used so
we re-send arp request to perform reachability verification instead of
expiring entry. The only signal that is needed from fast path is something
like binary yes/no.
The latter is solved by the following changes:
1) introduce special r_skip_req field which is read lockless by fast path,
but updated under (new) req_mutex mutex. If this field is non-zero, then
fast path will acquire lock and set it back to 0.
2) introduce simple state machine: incomplete->reachable<->verify->deleted.
Before that we implicitely had incomplete->reachable->deleted state machine,
with V_arpt_keep between "reachable" and "deleted". Verification was performed
in runtime 5 seconds before V_arpt_keep expire.
This is changed to "change state to verify 5 seconds before V_arpt_keep,
set r_skip_req to non-zero value and check it every second". If the value
is zero - then send arp verification probe.
These changes do not introduce any signifficant control plane overhead:
typically lle callout timer would fire 1 time more each V_arpt_keep (1200s)
for used lles and up to arp_maxtries (5) for dead lles.
As a result, all packets towards "reachable" lle are handled by fast path without
acquiring lle read lock.
Additional "req_mutex" is needed because callout / arpresolve_slow() or eventhandler
might keep LLE lock for signifficant amount of time, which might not be feasible
for fast path locking (e.g. having rmlock as ether AFDATA or lltable own lock).
Differential Revision: https://reviews.freebsd.org/D3688
In case when router has a lot of multicast groups, the reply can take
several packets due to MTU limitation.
Also we have a limit IGMP_MAX_RESPONSE_BURST == 4, that limits the number
of packets we send in one shot. Then we recalculate the timer value and
schedule the remaining packets for sending.
The problem is that when we call igmp_v3_dispatch_general_query() to send
remaining packets, we queue new reply in the same mbuf queue. And when
number of packets is bigger than IGMP_MAX_RESPONSE_BURST, we get endless
reply of IGMPv3 reports.
To fix this, add the check for remaining packets in the queue.
MFC after: 1 week
Sponsored by: Yandex LLC
by filter function instead of picking into routing table details in
each consumer.
Remove now-unused rt_expunge() (eliminating last external RTF_RNH_LOCKED
user).
This simplifies future nexthops/mulitipath changes and rtrequest1_fib()
locking refactoring.
Actual changes:
Add "rt_chain" field to permit rte grouping while doing batched delete
from routing table (thus growing rte 200->208 on amd64).
Add "rti_filter" / "rti_filterdata" / "rti_spare" fields to rt_addrinfo
to pass filter function to various routing subsystems in standard way.
Convert all rt_expunge() customers to new rt_addinfo-based api and eliminate
rt_expunge().
the UDP checksum computation and signals that it was OK,
clear this bit when passing the packet to SCTP. Since the
bits indicating a valid UDP checksum and a valid SCTP
checksum are the same, the SCTP stack would assume
that also an SCTP checksum check has been performed.
MFC after: 1 week
only for read locks on pcbs. The same race can happen with write
lock semantics as well.
The race scenario:
- Two threads (1 and 2) locate pcb with writer semantics (INPLOOKUP_WLOCKPCB)
and do in_pcbref() on it.
- 1 and 2 both drop the inp hash lock.
- Another thread (3) grabs the inp hash lock. Then it runs in_pcbfree(),
which wlocks the pcb. They must happen faster than 1 or 2 come INP_WLOCK()!
- 1 and 2 congest in INP_WLOCK().
- 3 does in_pcbremlists(), drops hash lock, and runs in_pcbrele_wlocked(),
which doesn't free the pcb due to two references on it.
Then it unlocks the pcb.
- 1 (or 2) gets wlock on the pcb, runs in_pcbrele_wlocked(), which doesn't
report inp as freed, due to 2 (or 1) still helding extra reference on it.
The thread tries to do smth with a disconnected pcb and crashes.
Submitted by: emeric.poupon@stormshield.eu
Reviewed by: gleb@
MFC after: 1 week
Sponsored by: Stormshield
Tested by: Cassiano Peixoto, Stormshield
Use hhook(9) framework to achieve ability of loading and unloading
if_enc(4) kernel module. INET and INET6 code on initialization registers
two helper hooks points in the kernel. if_enc(4) module uses these helper
hook points and registers its hooks. IPSEC code uses these hhook points
to call helper hooks implemented in if_enc(4).
When the M_NOWAIT allocation fails, we recurse the if_addr_lock trying
to clean up. Reorder the cleanup after dropping the if_addr_lock. The
obvious race is already possible between if_addmulti and IF_ADDR_WLOCK
above, so it must be ok.
Submitted by: Ryan Libby <rlibby@gmail.com>
Reviewed by: jhb
Found with: M_NOWAIT failure injection testing
Sponsored by: EMC / Isilon Storage Division
Differential Revision: https://reviews.freebsd.org/D4138
new return codes of -1 were mistakenly being considered "true". Callout_stop
now returns -1 to indicate the callout had either already completed or
was not running and 0 to indicate it could not be stopped. Also update
the manual page to make it more consistent no non-zero in the callout_stop
or callout_reset descriptions.
MFC after: 1 Month with associated callout change.
Rename arp_ifinit2() into arp_announce_ifaddr().
Eliminate zeroing ifa_rtrequest: it was used for calling arp_rtrequest()
which was responsible for handling route cloning requests. It became
obsolete since r186119 (L2/L3 split).
sysctl and will always be on. The former split between default and
fast forwarding is removed by this commit while preserving the ability
to use all network stack features.
Differential Revision: https://reviews.freebsd.org/D4042
Reviewed by: ae, melifaro, olivier, rwatson
MFC after: 1 month
Sponsored by: Rubicon Communications (Netgate)
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.
