or to compute the total retransmitted bytes in this sack recovery
episode, the scoreboard is traversed. While in sack recovery, this
traversal occurs on every call to tcp_output(), every dupack and
every partial ack. The scoreboard could potentially get quite large,
making this traversal expensive.
This change optimizes this by storing hints (for the next hole to
retransmit and the total retransmitted bytes in this sack recovery
episode) reducing the complexity to find these values from O(n) to
constant time.
The debug code that sanity checks the hints against the computed
value will be removed eventually.
Submitted by: Mohan Srinivasan, Noritoshi Demizu, Raja Mukerji.
in flight in SACK recovery.
Found by: Noritoshi Demizu
Submitted by: Mohan Srinivasan <mohans at yahoo-inc dot com>
Noritoshi Demizu <demizu at dd dot ij4u dot or dot jp>
Raja Mukerji <raja at moselle dot com>
setting ts_recent to an arbitrary value, stopping further
communication between the two hosts.
- If the Echoed Timestamp is greater than the current time,
fall back to the non RFC 1323 RTT calculation.
Submitted by: Raja Mukerji (raja at moselle dot com)
Reviewed by: Noritoshi Demizu, Mohan Srinivasan
a reassembly queue state structure, don't update (receiver) sack
report.
- Similarly, if tcp_drain() is called, freeing up all items on the
reassembly queue, clean the sack report.
Found, Submitted by: Noritoshi Demizu <demizu at dd dot iij4u dot or dot jp>
Reviewed by: Mohan Srinivasan (mohans at yahoo-inc dot com),
Raja Mukerji (raja at moselle dot com).
reported to the sender - in the case where the sender sends data
outside the window (as WinXP does :().
Reported by: Sam Jensen <sam at wand dot net dot nz>
Submitted by: Mohan Srinivasan
Remove the SACK "initburst" sysctl.
- Fix bugs in SACK dupack and partialack handling that can cause
large bursts while in SACK recovery.
Submitted by: Mohan Srinivasan
that the RFC 793 specification for accepting RST packets should be
following. When followed, this makes one vulnerable to the attacks
described in "slipping in the window", but it may be necessary in
some odd circumstances.
cases for tcp_input():
While it is true that the pcbinfo lock provides a pseudo-reference to
inpcbs, both the inpcb and pcbinfo locks are required to free an
un-referenced inpcb. As such, we can release the pcbinfo lock as
long as the inpcb remains locked with the confidence that it will not
be garbage-collected. This leads to a less conservative locking
strategy that should reduce contention on the TCP pcbinfo lock.
Discussed with: sam
pointer updates: test available space while holding the socket buffer
mutex, and continue to hold until until the pointer update has been
performed.
MFC after: 2 weeks
window was 0 bytes in size. This may have been the cause of unsolved
"connection not closing" reports over the years.
Thanks to Michiel Boland for providing the fix and providing a concise
test program for the problem.
Submitted by: Michiel Boland
MFC after: 2 weeks
contents of the tcpcb are read and modified in volume.
In tcp_input(), replace th comparison with 0 with a comparison with
NULL.
At the 'findpcb', 'dropafterack', and 'dropwithreset' labels in
tcp_input(), assert 'headlocked'. Try to improve consistency between
various assertions regarding headlocked to be more informative.
MFC after: 2 weeks
structure, so assert the inpcb lock associated with the tcptw.
Also assert the tcbinfo lock, as tcp_timewait() may call
tcp_twclose() or tcp_2msl_rest(), which require it. Since
tcp_timewait() is already called with that lock from tcp_input(),
this doesn't change current locking, merely documents reasons for
it.
In tcp_twstart(), assert the tcbinfo lock, as tcp_timer_2msl_rest()
is called, which requires that lock.
In tcp_twclose(), assert the tcbinfo lock, as tcp_timer_2msl_stop()
is called, which requires that lock.
Document the locking strategy for the time wait queues in tcp_timer.c,
which consists of protecting the time wait queues in the same manner
as the tcbinfo structure (using the tcbinfo lock).
In tcp_timer_2msl_reset(), assert the tcbinfo lock, as the time wait
queues are modified.
In tcp_timer_2msl_stop(), assert the tcbinfo lock, as the time wait
queues may be modified.
In tcp_timer_2msl_tw(), assert the tcbinfo lock, as the time wait
queues may be modified.
MFC after: 2 weeks
retain the pcbinfo lock until we're done using a pcb in the in-bound
path, as the pcbinfo lock acts as a pseuo-reference to prevent the pcb
from potentially being recycled. Clean up assertions and make sure to
assert that the pcbinfo is locked at the head of code subsections where
it is needed. Free the mbuf at the end of tcp_input after releasing
any held locks to reduce the time the locks are held.
MFC after: 3 weeks
A complete rationale and discussion is given in this message
and the resulting discussion:
http://docs.freebsd.org/cgi/mid.cgi?4177C8AD.6060706
Note that this commit removes only the functional part of T/TCP
from the tcp_* related functions in the kernel. Other features
introduced with RFC1644 are left intact (socket layer changes,
sendmsg(2) on connection oriented protocols) and are meant to
be reused by a simpler and less intrusive reimplemention of the
previous T/TCP functionality.
Discussed on: -arch
to control the packets injected while in sack recovery (for both
retransmissions and new data).
- Cleanups to the sack codepaths in tcp_output.c and tcp_sack.c.
- Add a new sysctl (net.inet.tcp.sack.initburst) that controls the
number of sack retransmissions done upon initiation of sack recovery.
Submitted by: Mohan Srinivasan <mohans@yahoo-inc.com>
and preserves the ipfw ABI. The ipfw core packet inspection and filtering
functions have not been changed, only how ipfw is invoked is different.
However there are many changes how ipfw is and its add-on's are handled:
In general ipfw is now called through the PFIL_HOOKS and most associated
magic, that was in ip_input() or ip_output() previously, is now done in
ipfw_check_[in|out]() in the ipfw PFIL handler.
IPDIVERT is entirely handled within the ipfw PFIL handlers. A packet to
be diverted is checked if it is fragmented, if yes, ip_reass() gets in for
reassembly. If not, or all fragments arrived and the packet is complete,
divert_packet is called directly. For 'tee' no reassembly attempt is made
and a copy of the packet is sent to the divert socket unmodified. The
original packet continues its way through ip_input/output().
ipfw 'forward' is done via m_tag's. The ipfw PFIL handlers tag the packet
with the new destination sockaddr_in. A check if the new destination is a
local IP address is made and the m_flags are set appropriately. ip_input()
and ip_output() have some more work to do here. For ip_input() the m_flags
are checked and a packet for us is directly sent to the 'ours' section for
further processing. Destination changes on the input path are only tagged
and the 'srcrt' flag to ip_forward() is set to disable destination checks
and ICMP replies at this stage. The tag is going to be handled on output.
ip_output() again checks for m_flags and the 'ours' tag. If found, the
packet will be dropped back to the IP netisr where it is going to be picked
up by ip_input() again and the directly sent to the 'ours' section. When
only the destination changes, the route's 'dst' is overwritten with the
new destination from the forward m_tag. Then it jumps back at the route
lookup again and skips the firewall check because it has been marked with
M_SKIP_FIREWALL. ipfw 'forward' has to be compiled into the kernel with
'option IPFIREWALL_FORWARD' to enable it.
DUMMYNET is entirely handled within the ipfw PFIL handlers. A packet for
a dummynet pipe or queue is directly sent to dummynet_io(). Dummynet will
then inject it back into ip_input/ip_output() after it has served its time.
Dummynet packets are tagged and will continue from the next rule when they
hit the ipfw PFIL handlers again after re-injection.
BRIDGING and IPFW_ETHER are not changed yet and use ipfw_chk() directly as
they did before. Later this will be changed to dedicated ETHER PFIL_HOOKS.
More detailed changes to the code:
conf/files
Add netinet/ip_fw_pfil.c.
conf/options
Add IPFIREWALL_FORWARD option.
modules/ipfw/Makefile
Add ip_fw_pfil.c.
net/bridge.c
Disable PFIL_HOOKS if ipfw for bridging is active. Bridging ipfw
is still directly invoked to handle layer2 headers and packets would
get a double ipfw when run through PFIL_HOOKS as well.
netinet/ip_divert.c
Removed divert_clone() function. It is no longer used.
netinet/ip_dummynet.[ch]
Neither the route 'ro' nor the destination 'dst' need to be stored
while in dummynet transit. Structure members and associated macros
are removed.
netinet/ip_fastfwd.c
Removed all direct ipfw handling code and replace it with the new
'ipfw forward' handling code.
netinet/ip_fw.h
Removed 'ro' and 'dst' from struct ip_fw_args.
netinet/ip_fw2.c
(Re)moved some global variables and the module handling.
netinet/ip_fw_pfil.c
New file containing the ipfw PFIL handlers and module initialization.
netinet/ip_input.c
Removed all direct ipfw handling code and replace it with the new
'ipfw forward' handling code. ip_forward() does not longer require
the 'next_hop' struct sockaddr_in argument. Disable early checks
if 'srcrt' is set.
netinet/ip_output.c
Removed all direct ipfw handling code and replace it with the new
'ipfw forward' handling code.
netinet/ip_var.h
Add ip_reass() as general function. (Used from ipfw PFIL handlers
for IPDIVERT.)
netinet/raw_ip.c
Directly check if ipfw and dummynet control pointers are active.
netinet/tcp_input.c
Rework the 'ipfw forward' to local code to work with the new way of
forward tags.
netinet/tcp_sack.c
Remove include 'opt_ipfw.h' which is not needed here.
sys/mbuf.h
Remove m_claim_next() macro which was exclusively for ipfw 'forward'
and is no longer needed.
Approved by: re (scottl)
- Trailing tab/space cleanup
- Remove spurious spaces between or before tabs
This change avoids touching files that Andre likely has in his working
set for PFIL hooks changes for IPFW/DUMMYNET.
Approved by: re (scottl)
Submitted by: Xin LI <delphij@frontfree.net>
Fix this problem by separating out the SACK and the newreno cases. Also, check
if we are in FASTRECOVERY for the sack case and if so, turn off dupacks.
Fix an issue where the congestion window was not being incremented by ssthresh.
Thanks to Mohan Srinivasan for finding this problem.
associated with performing a wakeup on the socket buffer:
- When performing an sbappend*() followed by a so[rw]wakeup(), explicitly
acquire the socket buffer lock and use the _locked() variants of both
calls. Note that the _locked() sowakeup() versions unlock the mutex on
return. This is done in uipc_send(), divert_packet(), mroute
socket_send(), raw_append(), tcp_reass(), tcp_input(), and udp_append().
- When the socket buffer lock is dropped before a sowakeup(), remove the
explicit unlock and use the _locked() sowakeup() variant. This is done
in soisdisconnecting(), soisdisconnected() when setting the can't send/
receive flags and dropping data, and in uipc_rcvd() which adjusting
back-pressure on the sockets.
For UNIX domain sockets running mpsafe with a contention-intensive SMP
mysql benchmark, this results in a 1.6% query rate improvement due to
reduce mutex costs.
locking in tcp_input() for TCP packets with urgent data pointers to
hold the socket buffer lock across testing and updating oobmark
from just protecting sb_state.
Update socket locking annotations
the socket buffer having its limits adjusted. sbreserve() now acquires
the lock before calling sbreserve_locked(). In soreserve(), acquire
socket buffer locks across read-modify-writes of socket buffer fields,
and calls into sbreserve/sbrelease; make sure to acquire in keeping
with the socket buffer lock order. In tcp_mss(), acquire the socket
buffer lock in the calling context so that we have atomic read-modify
-write on buffer sizes.
originated on RELENG_4 and was ported to -CURRENT.
The scoreboarding code was obtained from OpenBSD, and many
of the remaining changes were inspired by OpenBSD, but not
taken directly from there.
You can enable/disable sack using net.inet.tcp.do_sack. You can
also limit the number of sack holes that all senders can have in
the scoreboard with net.inet.tcp.sackhole_limit.
Reviewed by: gnn
Obtained from: Yahoo! (Mohan Srinivasan, Jayanth Vijayaraghavan)
flags relating to several aspects of socket functionality. This change
breaks out several bits relating to send and receive operation into a
new per-socket buffer field, sb_state, in order to facilitate locking.
This is required because, in order to provide more granular locking of
sockets, different state fields have different locking properties. The
following fields are moved to sb_state:
SS_CANTRCVMORE (so_state)
SS_CANTSENDMORE (so_state)
SS_RCVATMARK (so_state)
Rename respectively to:
SBS_CANTRCVMORE (so_rcv.sb_state)
SBS_CANTSENDMORE (so_snd.sb_state)
SBS_RCVATMARK (so_rcv.sb_state)
This facilitates locking by isolating fields to be located with other
identically locked fields, and permits greater granularity in socket
locking by avoiding storing fields with different locking semantics in
the same short (avoiding locking conflicts). In the future, we may
wish to coallesce sb_state and sb_flags; for the time being I leave
them separate and there is no additional memory overhead due to the
packing/alignment of shorts in the socket buffer structure.
SOCK_LOCK(so):
- Hold socket lock over calls to MAC entry points reading or
manipulating socket labels.
- Assert socket lock in MAC entry point implementations.
- When externalizing the socket label, first make a thread-local
copy while holding the socket lock, then release the socket lock
to externalize to userspace.
uncommitted):
Rename ip_claim_next_hop() to m_claim_next_hop(), give it an extra arg
(the type of tag to claim) and push it out of ip_var.h into mbuf.h
alongside all of the other macros that work ok mbuf's and tag's.
possible while maintaining compatibility with the widest range of TCP stacks.
The algorithm is as follows:
---
For connections in the ESTABLISHED state, only resets with
sequence numbers exactly matching last_ack_sent will cause a reset,
all other segments will be silently dropped.
For connections in all other states, a reset anywhere in the window
will cause the connection to be reset. All other segments will be
silently dropped.
---
The necessity of accepting all in-window resets was discovered
by jayanth and jlemon, both of whom have seen TCP stacks that
will respond to FIN-ACK packets with resets not meeting the
strict last_ack_sent check.
Idea by: Darren Reed
Reviewed by: truckman, jlemon, others(?)
from tcp_hostcache would have overridden a (now) lower MTU of
an interface or route that changed since first PMTU discovery.
The bug would have caused TCP to redo the PMTU discovery when
not strictly necessary.
Make a comment about already pre-initialized default values
more clear.
Reviewed by: sam
amount of segments it will hold.
The following tuneables and sysctls control the behaviour of the tcp
segment reassembly queue:
net.inet.tcp.reass.maxsegments (loader tuneable)
specifies the maximum number of segments all tcp reassemly queues can
hold (defaults to 1/16 of nmbclusters).
net.inet.tcp.reass.maxqlen
specifies the maximum number of segments any individual tcp session queue
can hold (defaults to 48).
net.inet.tcp.reass.cursegments (readonly)
counts the number of segments currently in all reassembly queues.
net.inet.tcp.reass.overflows (readonly)
counts how often either the global or local queue limit has been reached.
Tested by: bms, silby
Reviewed by: bms, silby
