Rusty Russell [Wed, 28 Mar 2007 21:29:08 +0000 (14:29 -0700)]
[NET]: Inline net_device_stats
Network drivers which keep stats allocate their own stats structure
then write a get_stats() function to return them. It would be nice if
this were done by default.
1) Add a new "stats" field to "struct net_device".
2) Add a new feature field to say "this driver uses the internal one"
3) Have a default "get_stats" which returns NULL if that feature not set.
4) Change callers to check result of get_stats call for NULL, not if
->get_stats is set.
This should not break backwards compatibility with older drivers, yet
allow modern drivers to shed some boilerplate code.
Lightly tested: works for a modified lguest network driver.
Eric Dumazet [Wed, 28 Mar 2007 21:22:33 +0000 (14:22 -0700)]
[NET]: random functions can use nsec resolution instead of usec
In order to get more randomness for secure_tcpv6_sequence_number(),
secure_tcp_sequence_number(), secure_dccp_sequence_number() functions,
we can use the high resolution time services, providing nanosec
resolution.
I've also done two kmalloc()/kzalloc() conversions.
Eric Dumazet [Tue, 27 Mar 2007 20:58:31 +0000 (13:58 -0700)]
[TCP]: tcp_memory_pressure and tcp_socket are__read_mostly candidates
tcp_memory_pressure and tcp_socket currently share a cache line with tcp_memory_allocated, tcp_sockets_allocated.
(Very hot cache line)
It makes sense to declare these variables as __read_mostly, to avoid false sharing on SMP.
Thomas Graf [Tue, 27 Mar 2007 20:56:52 +0000 (13:56 -0700)]
[NET] fib_rules: Flush route cache after rule modifications
The results of FIB rules lookups are cached in the routing cache
except for IPv6 as no such cache exists. So far, it was the
responsibility of the user to flush the cache after modifying any
rules. This lead to many false bug reports due to misunderstanding
of this concept.
This patch automatically flushes the route cache after inserting
or deleting a rule.
Thanks to Muli Ben-Yehuda <[email protected]> for catching a bug
in the previous patch.
Herbert Xu [Tue, 27 Mar 2007 06:22:20 +0000 (23:22 -0700)]
[NET]: Allow forwarding of ip_summed except CHECKSUM_COMPLETE
Right now Xen has a horrible hack that lets it forward packets with
partial checksums. One of the reasons that CHECKSUM_PARTIAL and
CHECKSUM_COMPLETE were added is so that we can get rid of this hack
(where it creates two extra bits in the skbuff to essentially mirror
ip_summed without being destroyed by the forwarding code).
I had forgotten that I've already gone through all the deivce drivers
last time around to make sure that they're looking at ip_summed ==
CHECKSUM_PARTIAL rather than ip_summed != 0 on transmit. In any case,
I've now done that again so it should definitely be safe.
Unfortunately nobody has yet added any code to update CHECKSUM_COMPLETE
values on forward so we I'm setting that to CHECKSUM_NONE. This should
be safe to remove for bridging but I'd like to check that code path
first.
So here is the patch that lets us get rid of the hack by preserving
ip_summed (mostly) on forwarded packets.
[IPV4] LVS: Allow to send ICMP unreachable responses when real-servers are removed
this is a small patch by Janusz Krzysztofik to ip_route_output_slow()
that allows VIP-less LVS linux director to generate packets
originating >From VIP if sysctl_ip_nonlocal_bind is set.
In a nutshell, the intention is for an LVS linux director to be able
to send ICMP unreachable responses to end-users when real-servers are
removed.
Thomas Graf [Tue, 27 Mar 2007 00:38:53 +0000 (17:38 -0700)]
[NET] fib_rules: Add no-operation action
The use of nop rules simplifies the usage of goto rules
and adds more flexibility as they allow targets to remain
while the actual content of the branches can change easly.
Thomas Graf [Tue, 27 Mar 2007 00:37:59 +0000 (17:37 -0700)]
[NET] fib_rules: Mark rules detached from the device
Rules which match against device names in their selector can
remain while the device itself disappears, in fact the device
doesn't have to present when the rule is added in the first
place. The device name is resolved by trying when the rule is
added and later by listening to NETDEV_REGISTER/UNREGISTER
notifications.
This patch adds the flag FIB_RULE_DEV_DETACHED which is set
towards userspace when a rule contains a device match which
is unresolved at the moment. This eases spotting the reason
why certain rules seem not to function properly.
Thomas Graf [Tue, 27 Mar 2007 00:14:15 +0000 (17:14 -0700)]
[NET] fib_rules: goto rule action
This patch adds a new rule action FR_ACT_GOTO which allows
to skip a set of rules by jumping to another rule. The rule
to jump to is specified via the FRA_GOTO attribute which
carries a rule preference.
Referring to a rule which doesn't exists is explicitely allowed.
Such goto rules are marked with the flag FIB_RULE_UNRESOLVED
and will act like a rule with a non-matching selector. The rule
will become functional as soon as its target is present.
The goto action enables performance optimizations by reducing
the average number of rules that have to be passed per lookup.
Example:
0: from all lookup local
40: not from all to 192.168.23.128 goto 32766
41: from all fwmark 0xa blackhole
42: from all fwmark 0xff blackhole
32766: from all lookup main
The following update received from Injong updates TCP cubic to the latest
version. I am running more complete tests and will have results after 4/1.
According to Injong: the new version improves on its scalability,
fairness and stability. So in all properties, we confirmed it shows better
performance.
NCSU results (for 2.6.18 and 2.6.20) available:
http://netsrv.csc.ncsu.edu/wiki/index.php/TCP_Testing
This version is described in a new Internet draft for CUBIC.
http://www.ietf.org/internet-drafts/draft-rhee-tcp-cubic-00.txt
David S. Miller [Fri, 23 Mar 2007 18:40:27 +0000 (11:40 -0700)]
[INET]: Use jhash + random secret for ehash.
The days are gone when this was not an issue, there are folks out
there with huge bot networks that can be used to attack the
established hash tables on remote systems.
So just like the routing cache and connection tracking
hash, use Jenkins hash with random secret input.
Vlad Yasevich [Fri, 23 Mar 2007 18:34:08 +0000 (11:34 -0700)]
[SCTP]: Implement sac_info field in SCTP_ASSOC_CHANGE notification.
As stated in the sctp socket api draft:
sac_info: variable
If the sac_state is SCTP_COMM_LOST and an ABORT chunk was received
for this association, sac_info[] contains the complete ABORT chunk as
defined in the SCTP specification RFC2960 [RFC2960] section 3.3.7.
We now save received ABORT chunks into the sac_info field and pass that
to the user.
Vlad Yasevich [Fri, 23 Mar 2007 18:33:12 +0000 (11:33 -0700)]
[SCTP]: Honor flags when setting peer address parameters
Parameters only take effect when a corresponding flag bit is set
and a value is specified. This means we need to check the flags
in addition to checking for non-zero value.
This option induces partial delivery to run as soon
as the specified amount of data has been accumulated on
the association. However, we give preference to fully
reassembled messages over PD messages. In any case,
window and buffer is freed up.
This option was introduced in draft-ietf-tsvwg-sctpsocket-13. It
prevents head-of-line blocking in the case of one-to-many endpoint.
Applications enabling this option really must enable SCTP_SNDRCV event
so that they would know where the data belongs. Based on an
earlier patch by Ivan Skytte Jørgensen.
Additionally, this functionality now permits multiple associations
on the same endpoint to enter Partial Delivery. Applications should
be extra careful, when using this functionality, to track EOR indicators.
Patrick McHardy [Fri, 23 Mar 2007 18:27:04 +0000 (11:27 -0700)]
[NET_SCHED]: sch_netem: fix off-by-one in send time comparison
netem checks PSCHED_TLESS(cb->time_to_send, now) to find out whether it is
allowed to send a packet, which is equivalent to cb->time_to_send < now.
Use !PSCHED_TLESS(now, cb->time_to_send) instead to properly handle
cb->time_to_send == now.
Thomas Graf [Fri, 23 Mar 2007 18:17:57 +0000 (11:17 -0700)]
[NETFILTER] nfnetlink: netlink_run_queue() already checks for NLM_F_REQUEST
Patrick has made use of netlink_run_queue() in nfnetlink while my patches
have been waiting for net-2.6.22 to open. So this check for NLM_F_REQUEST
can go as well.
Thomas Graf [Fri, 23 Mar 2007 06:30:55 +0000 (23:30 -0700)]
[NETLINK]: Directly return -EINTR from netlink_dump_start()
Now that all users of netlink_dump_start() use netlink_run_queue()
to process the receive queue, it is possible to return -EINTR from
netlink_dump_start() directly, therefore simplying the callers.
Thomas Graf [Fri, 23 Mar 2007 06:30:12 +0000 (23:30 -0700)]
[NETLINK]: Remove error pointer from netlink message handler
The error pointer argument in netlink message handlers is used
to signal the special case where processing has to be interrupted
because a dump was started but no error happened. Instead it is
simpler and more clear to return -EINTR and have netlink_run_queue()
deal with getting the queue right.
nfnetlink passed on this error pointer to its subsystem handlers
but only uses it to signal the start of a netlink dump. Therefore
it can be removed there as well.
This patch also cleans up the error handling in the affected
message handlers to be consistent since it had to be touched anyway.
Thomas Graf [Fri, 23 Mar 2007 06:28:46 +0000 (23:28 -0700)]
[NETLINK]: Ignore !NLM_F_REQUEST messages directly in netlink_run_queue()
netlink_rcv_skb() is changed to skip messages which don't have the
NLM_F_REQUEST bit to avoid every netlink family having to perform this
check on their own.
[NET_SCHED] qdisc: avoid transmit softirq on watchdog wakeup
If possible, avoid having to do a transmit softirq when a qdisc
watchdog decides to re-enable. The watchdog routine runs off
a timer, so it is already in the same effective context as
the softirq.
The netem code would call getnstimeofday() and dequeue/requeue after
every packet, even if it was waiting. Avoid this overhead by using
the throttled flag.
If you setup netem to just delay packets; "tc qdisc ls" will report
the reordering as 100%. Well it's a lie, reorder isn't used unless
gap is set, so just set value to 0 so the output of utility
is correct.
Thomas Graf [Mon, 26 Mar 2007 06:24:24 +0000 (23:24 -0700)]
[NET] rules: Unified rules dumping
Implements a unified, protocol independant rules dumping function
which is capable of both, dumping a specific protocol family or
all of them. This speeds up dumping as less lookups are required.
Thomas Graf [Thu, 22 Mar 2007 18:48:11 +0000 (11:48 -0700)]
[RTNL]: Message handler registration interface
This patch adds a new interface to register rtnetlink message
handlers replacing the exported rtnl_links[] array which
required many message handlers to be exported unnecessarly.
Gerrit Renker [Tue, 20 Mar 2007 18:31:56 +0000 (15:31 -0300)]
[CCID3]: Use initial RTT sample from SYN exchange
The patch follows the following recommendation made in an erratum to RFC 4342:
"Senders MAY additionally make use of other available RTT measurements,
including those from the initial Request-Response packet exchange."
It implements larger initial windows with regard to this inital RTT measurement,
using the mechanism suggested in draft-ietf-dccp-rfc3448bis, section 4.2.
Gerrit Renker [Tue, 20 Mar 2007 18:23:18 +0000 (15:23 -0300)]
[DCCP]: Provide function for RTT sampling
A recurring problem, in particular in the CCID code, is that RTT samples
from packets with timestamp echo and elapsed time options need to be taken.
This service is provided via a new function dccp_sample_rtt in this patch.
Furthermore, to protect against `insane' RTT samples, the sampled value
is bounded between 100 microseconds and 4 seconds - for which u32 is sufficient.
Gerrit Renker [Tue, 20 Mar 2007 18:04:30 +0000 (15:04 -0300)]
[CCID3]: Remove build warnings for 64bit
This clears the following sparc64 build warnings:
1) warning: format "%ld" expects type "long int", but argument 3 has type "suseconds_t"
2) warning: format "%llu" expects type "long long unsigned int", but argument 3 has type "__u64"
Fixed by using typecast to unsigned. This is argued to be safe, since the quantities, after
de-scaling (factor 2^6) fit all in u32.
Gerrit Renker [Tue, 20 Mar 2007 18:00:28 +0000 (15:00 -0300)]
[DCCP]: More debug information for dccp_wait_for_ccid
This adds more detail in the wait_for_ccid packet scheduling loop.
In particular, it informs about (i) when delay is used and (ii) why
a packet is discarded.
Gerrit Renker [Tue, 20 Mar 2007 17:59:23 +0000 (14:59 -0300)]
[DCCP]: Always use debug-toggle parameters
Currently debugging output (when configured) is automatically enabled when
DCCP modules are compiled into the kernel rather than built as loadable modules.
This is not necessary, since the module parameters in this case become kernel
commandline parameters, e.g. DCCP or CCID3 debug output can be enabled for a
static build by appending the following at the boot prompt:
dccp.dccp_debug=1 dccp_ccid3.ccid3_debug=1
This patch therefore does away with the more complicated way of always enabling
debug output for static builds
Gerrit Renker [Tue, 20 Mar 2007 17:28:44 +0000 (14:28 -0300)]
[CCID3]: Use MSS for larger initial windows
This improves the slow-start phase by using the MSS
(as suggested in RFC 4342, sec. 5) instead of the packet size s.
Also figured out that __u32 is ample resource enough.
Gerrit Renker [Tue, 20 Mar 2007 16:11:24 +0000 (13:11 -0300)]
[CCID3]: Re-order CCID 3 source file
No code change at all.
This splits ccid3.c into a RX and a TX section, so that the file has an
organisation similar to the other ones (e.g. packet_history.{h,c}).
Gerrit Renker [Tue, 20 Mar 2007 16:10:15 +0000 (13:10 -0300)]
[CCID3]: Remove redundant `len' test
Since CCID3 avoids sending 0-byte data packets (cf. ccid3_hc_tx_send_packet),
testing for zero-payload length, as performed by ccid3_hc_tx_update_s, is
redundant - hence removed by this patch.
Gerrit Renker [Tue, 20 Mar 2007 16:08:19 +0000 (13:08 -0300)]
[DCCP]: Remove ambiguity in the way before48 is used
This removes two ambiguities in employing the new definition of before48,
following the analysis on http://www.mail-archive.com/[email protected]/msg01295.html
(1) Updating GSR when P.seqno >= S.SWL
With the old definition we did not update when P.seqno and S.SWL are 2^47 apart. To
ensure the same behaviour as with the old definition, this is replaced with the
equivalent condition dccp_delta_seqno(S.SWL, P.seqno) >= 0
(2) Sending SYNC when P.seqno >= S.OSR
Here it is debatable whether the new definition causes an ambiguity: the case is
similar to (1); and to have consistency with the case (1), we use the equivalent
condition dccp_delta_seqno(S.OSR, P.seqno) >= 0
projects / linux.git / log