kernel - Fix sack NULL pointer dereference

* sack_block_lookup() can get confused when the passed-in sequence
number appears to be less than sblk_start and greater than sblk_end.
This situation

kernel - Fix sack NULL pointer dereference

* sack_block_lookup() can get confused when the passed-in sequence
number appears to be less than sblk_start and greater than sblk_end.
This situation can occur when the signed integer delta test has an
overflow due to (sblk_end - seq) overflowing the sign bit verses
(sblk_start - seq).

The result is that sack_block_lookup() can crash on a NULL pointer
indirection.

* Check for the case, complain, and try to allow it. Though I suspect
if the case occurs at all SACK will wind up with a broken list anyway.

* I don't think this case can occur under normal conditions since TCP
buffers do not grow to 2GB+ in size, so the crash we got was triggered
by either an accidently malformed packet or an intentional one.

show more ...

merge

tcp: Improve sender-sender and sender-receiver fairness on the same netisr

Yield to other senders or receivers on the same netisr if the current TCP
stream has sent certain amount of segments (curre

tcp: Improve sender-sender and sender-receiver fairness on the same netisr

Yield to other senders or receivers on the same netisr if the current TCP
stream has sent certain amount of segments (currently 4) and is going to
burst more segments. sysctl net.inet.tcp.fairsend could be used to tune
how many segements are allowed to burst. For TSO capable devices, their
TSO aggregate size limit could also affect the number of segments allowed
to burst. Set net.inet.tcp.fairsend to 0 will allow single TCP stream to
burst as much as it wants (the old TCP sender's behaviour).

"Fairsend" is performed at the places that do not affect segment sending
during congestion control:
- User requested output path
- ACK input path

Measured improvement in the following setup:

+---+ +---+
| |<-----------| B |
| | +---+
| A |
| | +---+
| |----------->| C |
+---+ +---+

A (i7-2600, w/ HT enabled), 82571EB
B (e3-1230, w/ HT enabled), 82574L
C (e3-1230, w/ HT enabled), 82574L
The performance stats are gathered from 'systat -if 1'

When A runs 8 TCP senders to C and 8 TCP receivers from B, sending
performance are same ~975Mbps, however, the receiving performance before
this commit stumbles between 670Mbps and 850Mbps; w/ "fairsend" receiving
performance stays at 981Mbps.

When A runs 16 TCP senders to C and 16 TCP receivers from B, sending
performance are same ~975Mbps, however, the receiving performance before
this commit goes from 960Mbps to 980Mbps; w/ "fairsend" receiving
performance stays at 981Mbps stably.

When there are more senders and receivers running on A, there is no
noticable performance difference on either sending or receiving between
non-"fairsend" and "fairsend", because senders are no longer being able
to do continuous large burst.

"Fairsend" also improves Jain's fairness index between various amount of
senders (8 ~ 128) a little bit (sending only tests).

show more ...

Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2

tcp: RFC3517bis is now officially RFC6675

Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2

tcp/sack: Take bwnd into consideration when calculate length of new segment

tcp/sack: Discard HighRxt, RescueRxt and LostSeq along with SACK scoreboard

Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2

tcp/sack: If other side reneged, discard the current SACK scoreboard

Other side reneging is detected using the first SACK record:
If its left edge is less than or equal to the cumulative ACK of the

tcp/sack: If other side reneged, discard the current SACK scoreboard

Other side reneging is detected using the first SACK record:
If its left edge is less than or equal to the cumulative ACK of the
incoming segment, other side probably reneged.

This fixes the later assertion that the first SACK record's left edge
must be above snd_una in tcp_sack_first_unsacked_len()

Add statistics about other side reneging

show more ...

tcp/sack: Constify function arguments if possible

tcp/sack: Fix off-by-one bug when updating rescue SACK information

tcp/sack: Force out more segments allowed by "pipe" during fast recovery

If some segments are cumulatively acked or SACKed, and HighRxt equals
snd_una, one segment (new or retransmit) will be forced

tcp/sack: Force out more segments allowed by "pipe" during fast recovery

If some segments are cumulatively acked or SACKed, and HighRxt equals
snd_una, one segment (new or retransmit) will be forced out even if cwnd
and pipe don't allow it. When large amount of segments are lost, i.e.
computed pipe could be large, this avoids unnecessary retransmit timeout
and could perform as good as NewReno.

Sysctl node net.inet.tcp.force_sackrxt could be tuned to burst out several
retransmits, default is 1 (should be good enough). Set this sysctl to 0,
SACK based fast recovery will obey the computed pipe.

Several unnecessary retransmit timeout graph as described above:
http://leaf.dragonflybsd.org/~sephe/no_force_sack_rexmt2_15.xpl (starts @15s)
http://leaf.dragonflybsd.org/~sephe/no_force_sack_rexmt_54.xpl (starts @54s)

show more ...

tcp/sack: Use RFC3517bis IsLost(snd_una) as fallback of early retransmit

Since we are less certain about whether is segment is lost or not when
using IsLost(snd_una), we do not send out other unSACK

tcp/sack: Use RFC3517bis IsLost(snd_una) as fallback of early retransmit

Since we are less certain about whether is segment is lost or not when
using IsLost(snd_una), we do not send out other unSACKed segments except
the first unSACKed segment under this condition. Sending out other
unSACKed segments could be too aggressive here; just wait for another
ACK to tick out more unSACKed segments.

show more ...

tcp/sack: Fix the condition that SACK rescue retransmit can't be done

If we have nothing left above the HighRxt, the first unSACKed segment
will be used as the SACK rescue retransmit.

tcp: Optimize SACK scoreboard records consolidation a little bit

If the SACK block and SACK scoreboard record are matched exactly,
SACK scoreboard records consolidation is not needed at all.

Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2

tcp: Implement RFC4653 Non-Congestion Robustness (NCR)

It is enabled by default and can be disabled using sysctl node:
net.inet.tcp.ncr

As far as I have tested on heavily reordered network path, th

tcp: Implement RFC4653 Non-Congestion Robustness (NCR)

It is enabled by default and can be disabled using sysctl node:
net.inet.tcp.ncr

As far as I have tested on heavily reordered network path, this
algorithm does avoid most of the spurious fast retransmits. While
on the normal network path, the fast retransmits stil could be
triggered properly.

show more ...

Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2

tcp: Use TAILQ for segments reassemble queue

So the last segment of the reassemble queue could be peeked w/ minimal cost

Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2

tcp: Add sack_flags for SACK related operations

This saves us 4 bits in the crowded t_flags

Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly into hammer2

tcp/sack: Don't reduce retransmit threshold as recommended by RFC3517bis

- Don't reduce byte threshold in IsLost()
- Don't retransmit if IsLost(tcpcb.snd_una) is true

They cause spurious retransmit

tcp/sack: Don't reduce retransmit threshold as recommended by RFC3517bis

- Don't reduce byte threshold in IsLost()
- Don't retransmit if IsLost(tcpcb.snd_una) is true

They cause spurious retransmits.

Add sysctl node net.inet.tcp.rfc3517bis_rxt to enable the RFC3517bis
recommended retransmit threshold reduction. It is disabled by default.

show more ...

tcp/sack: Implement RFC3517bis

http://tools.ietf.org/html/draft-ietf-tcpm-3517bis-02, which will
be become "Standards Track" soon.

net.inet.tcp.rfc3517bis sysctl node is added to enable this update

tcp/sack: Implement RFC3517bis

http://tools.ietf.org/html/draft-ietf-tcpm-3517bis-02, which will
be become "Standards Track" soon.

net.inet.tcp.rfc3517bis sysctl node is added to enable this update.
It is off by default.

show more ...