I always had thought that collapsing up to two at a time was
intentional decision to avoid excessive processing if 1 byte
sized skbs are to be combined for a full mtu, and consecutive
retransmissions would make the size of the retransmittee
double each round anyway, but some recent discussion made me
to understand that was not the case. Thus make collapse work
more and wait less.
It would be possible to take advantage of the shifting
machinery (added in the later patch) in the case of paged
data but that can be implemented on top of this change.
tcp_skb_is_last check is now provided by the loop.
I tested a bit (ss-after-idle-off, fill 4096x4096B xfer,
10s sleep + 4096 x 1byte writes while dropping them for
some a while with netem):
. 16774097:16775545(1448) ack 1 win 46
. 16775545:16776993(1448) ack 1 win 46
. ack 16759617 win 2399
P 16776993:16777217(224) ack 1 win 46
. ack 16762513 win 2399
. ack 16765409 win 2399
. ack 16768305 win 2399
. ack 16771201 win 2399
. ack 16774097 win 2399
. ack 16776993 win 2399
. ack 16777217 win 2399
P 16777217:16777257(40) ack 1 win 46
. ack 16777257 win 2399
P 16777257:16778705(1448) ack 1 win 46
P 16778705:16780153(1448) ack 1 win 46
FP 16780153:16781313(1160) ack 1 win 46
. ack 16778705 win 2399
. ack 16780153 win 2399
F 1:1(0) ack 16781314 win 2399
While without drop-all period I get this:
. 16773585:16775033(1448) ack 1 win 46
. ack 16764897 win 9367
. ack 16767793 win 9367
. ack 16770689 win 9367
. ack 16773585 win 9367
. 16775033:16776481(1448) ack 1 win 46
P 16776481:16777217(736) ack 1 win 46
. ack 16776481 win 9367
. ack 16777217 win 9367
P 16777217:16777218(1) ack 1 win 46
P 16777218:16777219(1) ack 1 win 46
P 16777219:16777220(1) ack 1 win 46
...
P 16777247:16777248(1) ack 1 win 46
. ack 16777218 win 9367
. ack 16777219 win 9367
...
. ack 16777233 win 9367
. ack 16777248 win 9367
P 16777248:16778696(1448) ack 1 win 46
P 16778696:16780144(1448) ack 1 win 46
FP 16780144:16781313(1169) ack 1 win 46
. ack 16780144 win 9367
F 1:1(0) ack 16781314 win 9367
The window seems to be 30-40 segments, which were successfully
combined into: P 16777217:16777257(40) ack 1 win 46
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes http://bugzilla.kernel.org/show_bug.cgi?id=12014
Since most (if not all) implementations of TSO and even the in-kernel
software GSO do not update the urgent pointer when splitting a large
segment, it is necessary to turn off TSO/GSO for all outgoing traffic
with the URG pointer set.
Looking at tcp_current_mss (and the preceding comment) I even think
this was the original intention. However, this approach is insufficient,
because TSO/GSO is turned off only for newly created frames, not for
frames which were already pending at the arrival of a message with
MSG_OOB set. These frames were created when TSO/GSO was enabled,
so they may be large, and they will have the urgent pointer set
in tcp_transmit_skb().
With this patch, such large packets will be fragmented again before
going to the transmit routine.
As a side note, at least the following NICs are known to screw up
the urgent pointer in the TCP header when doing TSO:
Intel 82566MM (PCI ID 8086:1049)
Intel 82566DC (PCI ID 8086:104b)
Intel 82541GI (PCI ID 8086:1076)
Broadcom NetXtreme II BCM5708 (PCI ID 14e4:164c)
Signed-off-by: Petr Tesarik <ptesarik@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
David Miller noticed that commit
33ad798c92 '(tcp: options clean up')
did not move the req->cookie_ts check.
This essentially disabled commit 4dfc281702
'[Syncookies]: Add support for TCP options via timestamps.'.
This restores the original logic.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is not our bug! Sadly some devices cannot cope with the change
of TCP option ordering which was a result of the recent rewrite of
the option code (not that there was some particular reason steming
from the rewrite for the reordering) though any ordering of TCP
options is perfectly legal. Thus we restore the original ordering
to allow interoperability with/through such broken devices and add
some warning about this trap. Since the reordering just happened
without any particular reason, this change shouldn't cost us
anything.
There are already couple of known failure reports (within close
proximity of the last release), so the problem might be more
wide-spread than a single device. And other reports which may
be due to the same problem though the symptoms were less obvious.
Analysis of one of the case revealed (with very high probability)
that sack capability cannot be negotiated as the first option
(SYN never got a response).
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Reported-by: Aldo Maggi <sentiniate@tiscali.it>
Tested-by: Aldo Maggi <sentiniate@tiscali.it>
Signed-off-by: David S. Miller <davem@davemloft.net>
While looking for the recent "sack issue" I also read all eff_sacks
usage that was played around by some relevant commit. I found
out that there's another thing that is asking for a fix (unrelated
to the "sack issue" though).
This feature has probably very little significance in practice.
Opposite direction timeout with bidirectional tcp comes to me as
the most likely scenario though there might be other cases as
well related to non-data segments we send (e.g., response to the
opposite direction segment). Also some ACK losses or option space
wasted for other purposes is necessary to prevent the earlier
SACK feedback getting to the sender.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
It all started from me noticing that this urgent check in
tcp_clean_rtx_queue is unnecessarily inside the loop. Then
I took a longer look to it and found out that the users of
urg_mode can trivially do without, well almost, there was
one gotcha.
Bonus: those funny people who use urg with >= 2^31 write_seq -
snd_una could now rejoice too (that's the only purpose for the
between being there, otherwise a simple compare would have done
the thing). Not that I assume that the rest of the tcp code
happily lives with such mind-boggling numbers :-). Alas, it
turned out to be impossible to set wmem to such numbers anyway,
yes I really tried a big sendfile after setting some wmem but
nothing happened :-). ...Tcp_wmem is int and so is sk_sndbuf...
So I hacked a bit variable to long and found out that it seems
to work... :-)
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
Current TCP code relies on the local port of the listening socket
being the same as the destination address of the incoming
connection. Port redirection used by many transparent proxying
techniques obviously breaks this, so we have to store the original
destination port address.
This patch extends struct inet_request_sock and stores the incoming
destination port value there. It also modifies the handshake code to
use that value as the source port when sending reply packets.
Signed-off-by: KOVACS Krisztian <hidden@sch.bme.hu>
Signed-off-by: David S. Miller <davem@davemloft.net>
I'm trying to use the TCP_MAXSEG option to setsockopt() to set the MSS
for both sides of a bidirectional connection.
man tcp says: "If this option is set before connection establishment, it
also changes the MSS value announced to the other end in the initial
packet."
However, the kernel only uses the MTU/route cache to set the advertised
MSS. That means if I set the MSS to, say, 500 before calling connect(),
I will send at most 500-byte packets, but I will still receive 1500-byte
packets in reply.
This is a bug, either in the kernel or the documentation.
This patch (applies to latest net-2.6) reduces the advertised value to
that requested by the user as long as setsockopt() is called before
connect() or accept(). This seems like the behavior that one would
expect as well as that which is documented.
I've tried to make sure that things that depend on the advertised MSS
are set correctly.
Signed-off-by: Tom Quetchenbach <virtualphtn@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If lost skb is sacked, we might have nothing to retransmit
as high as the retransmit_high is pointing to, so place
it lower to avoid unnecessary walking.
This is mainly for the case where high L'ed skbs gets sacked.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
Most importantly avoid doing it with cumulative ACK. Not clearing
means that we no longer need n^2 processing in resolution of each
fast recovery.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
This doesn't much sense here afaict, probably never has. Since
fragmenting and collapsing deal the hints by themselves, there
should be very little reason for the rexmit loop to do that.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
Both loops are quite similar, so they can be combined
with little effort. As a result, forward_skb_hint becomes
obsolete as well.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
The validity of the retransmit_high must then be ensured
if no L'ed skb exits!
This makes a minor change to behavior, we now have to
iterate the head to find out that the loop terminates.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
Main benefit in this is that we can then freely point
the retransmit_skb_hint to anywhere we want to because
there's no longer need to know what would be the count
changes involve, and since this is really used only as a
terminator, unnecessary work is one time walk at most,
and if some retransmissions are necessary after that
point later on, the walk is not full waste of time
anyway.
Since retransmit_high must be kept valid, all lost
markers must ensure that.
Now I also have learned how those "holes" in the
rexmittable skbs can appear, mtu probe does them. So
I removed the misleading comment as well.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ie., the difference between partial and all clearing doesn't
exists anymore since the SACK optimizations got dropped by
an sacktag rewrite.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
The size of the TCP header is miscalculated when the window scale ends
up being 0. Additionally, this can be induced by sending a SYN to a
passive open port with a window scale option with value 0.
Signed-off-by: Philip Love <love_phil@emc.com>
Signed-off-by: Adam Langley <agl@imperialviolet.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
As reported by Alexey Dobriyan:
CHECK net/ipv4/tcp_output.c
net/ipv4/tcp_output.c:475:7: warning: dubious: !x & y
And sparse is damn right!
if (unlikely(!OPTION_TS & opts->options))
^^^
size += TCPOLEN_SACKPERM_ALIGNED;
OPTION_TS is (1 << 1), so condition will never trigger.
Signed-off-by: David S. Miller <davem@davemloft.net>
This should fix the following bugs:
* Connections with MD5 signatures produce invalid packets whenever SACK
options are included
* MD5 signatures are counted twice in the MSS calculations
Behaviour changes:
* A SYN with MD5 + SACK + TS elicits a SYNACK with MD5 + SACK
This is because we can't fit any SACK blocks in a packet with MD5 + TS
options. There was discussion about disabling SACK rather than TS in
order to fit in better with old, buggy kernels, but that was deemed to
be unnecessary.
* SYNs with MD5 don't include a TS option
See above.
Additionally, it removes a bunch of duplicated logic for calculating options,
which should help avoid these sort of issues in the future.
Signed-off-by: Adam Langley <agl@imperialviolet.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, the MD5 code assumes that the SKBs are linear and, in the case
that they aren't, happily goes off and hashes off the end of the SKB and
into random memory.
Reported by Stephen Hemminger in [1]. Advice thanks to Stephen and Evgeniy
Polyakov. Also includes a couple of missed route_caps from Stephen's patch
in [2].
[1] http://marc.info/?l=linux-netdev&m=121445989106145&w=2
[2] http://marc.info/?l=linux-netdev&m=121459157816964&w=2
Signed-off-by: Adam Langley <agl@imperialviolet.org>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fortunately (almost) all the TCP code has a sock to get the net from :)
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are some places in TCP that select one MIB index to
bump snmp statistics like this:
if (<something>)
NET_INC_STATS_BH(<some_id>);
else if (<something_else>)
NET_INC_STATS_BH(<some_other_id>);
...
else
NET_INC_STATS_BH(<default_id>);
or in a more tricky but still similar way.
On the other hand, this NET_INC_STATS_BH is a camouflaged
increment of percpu variable, which is not that small.
Factoring those cases out de-bloats 235 bytes on non-preemptible
i386 config and drives parts of the code into 80 columns.
add/remove: 0/0 grow/shrink: 0/7 up/down: 0/-235 (-235)
function old new delta
tcp_fastretrans_alert 1437 1424 -13
tcp_dsack_set 137 124 -13
tcp_xmit_retransmit_queue 690 676 -14
tcp_try_undo_recovery 283 265 -18
tcp_sacktag_write_queue 1550 1515 -35
tcp_update_reordering 162 106 -56
tcp_retransmit_timer 990 904 -86
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes CVS keywords that weren't updated for a long time
from comments.
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP "resets sent" counter is not incremented when a TCP Reset is
sent via tcp_send_active_reset().
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We are seeing an issue with TCP in handling an ICMP frag needed
message that is received after net.ipv4.tcp_retries1 retransmits.
The default value of retries1 is 3. So if the path mtu changes
and ICMP frag needed is lost for the first 3 retransmits or if
it gets delayed until 3 retransmits are done, TCP doesn't update
MSS correctly and continues to retransmit the orginal message
until it timesout after tcp_retries2 retransmits.
I am seeing this issue even with the latest 2.6.25.4 kernel.
In tcp_retransmit_timer(), when retransmits counter exceeds
tcp_retries1 value, the dst cache entry of the socket is reset.
At this time, if we receive an ICMP frag needed message, the
dst entry gets updated with the new MTU, but the TCP sockets
dst_cache entry remains NULL.
So the next time when we try to retransmit after the ICMP frag
needed is received, tcp_retransmit_skb() gets called. Here the
cur_mss value is calculated at the start of the routine with
a NULL sk_dst_cache. Instead we should call tcp_current_mss after
the rebuild_header that caches the dst entry with the updated mtu.
Also the rebuild_header should be called before tcp_fragment
so that skb is fragmented if the mss goes down.
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Needed can only be more strict than what was checked by the
earlier common case check for non-tail skbs, thus
cwnd_len <= needed will never match in that case anyway.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow the use of SACK and window scaling when syncookies are used
and the client supports tcp timestamps. Options are encoded into
the timestamp sent in the syn-ack and restored from the timestamp
echo when the ack is received.
Based on earlier work by Glenn Griffin.
This patch avoids increasing the size of structs by encoding TCP
options into the least significant bits of the timestamp and
by not using any 'timestamp offset'.
The downside is that the timestamp sent in the packet after the synack
will increase by several seconds.
changes since v1:
don't duplicate timestamp echo decoding function, put it into ipv4/syncookie.c
and have ipv6/syncookies.c use it.
Feedback from Glenn Griffin: fix line indented with spaces, kill redundant if ()
Reviewed-by: Hagen Paul Pfeifer <hagen@jauu.net>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
This fixes Bugzilla #10384
tcp_simple_retransmit does L increment without any checking
whatsoever for overflowing S+L when Reno is in use.
The simplest scenario I can currently think of is rather
complex in practice (there might be some more straightforward
cases though). Ie., if mss is reduced during mtu probing, it
may end up marking everything lost and if some duplicate ACKs
arrived prior to that sacked_out will be non-zero as well,
leading to S+L > packets_out, tcp_clean_rtx_queue on the next
cumulative ACK or tcp_fastretrans_alert on the next duplicate
ACK will fix the S counter.
More straightforward (but questionable) solution would be to
just call tcp_reset_reno_sack() in tcp_simple_retransmit but
it would negatively impact the probe's retransmission, ie.,
the retransmissions would not occur if some duplicate ACKs
had arrived.
So I had to add reno sacked_out reseting to CA_Loss state
when the first cumulative ACK arrives (this stale sacked_out
might actually be the explanation for the reports of left_out
overflows in kernel prior to 2.6.23 and S+L overflow reports
of 2.6.24). However, this alone won't be enough to fix kernel
before 2.6.24 because it is building on top of the commit
1b6d427bb7 ([TCP]: Reduce sacked_out with reno when purging
write_queue) to keep the sacked_out from overflowing.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Reported-by: Alessandro Suardi <alessandro.suardi@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Update: My mailer ate one of Jarek's feedback mails... Fixed the
parameter in netif_set_gso_max_size() to be u32, not u16. Fixed the
whitespace issue due to a patch import botch. Changed the types from
u32 to unsigned int to be more consistent with other variables in the
area. Also brought the patch up to the latest net-2.6.26 tree.
Update: Made gso_max_size container 32 bits, not 16. Moved the
location of gso_max_size within netdev to be less hotpath. Made more
consistent names between the sock and netdev layers, and added a
define for the max GSO size.
Update: Respun for net-2.6.26 tree.
Update: changed max_gso_frame_size and sk_gso_max_size from signed to
unsigned - thanks Stephen!
This patch adds the ability for device drivers to control the size of
the TSO frames being sent to them, per TCP connection. By setting the
netdevice's gso_max_size value, the socket layer will set the GSO
frame size based on that value. This will propogate into the TCP
layer, and send TSO's of that size to the hardware.
This can be desirable to help tune the bursty nature of TSO on a
per-adapter basis, where one may have 1 GbE and 10 GbE devices
coexisting in a system, one running multiqueue and the other not, etc.
This can also be desirable for devices that cannot support full 64 KB
TSO's, but still want to benefit from some level of segmentation
offloading.
Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When selecting a new window, tcp_select_window() tries not to shrink
the offered window by using the maximum of the remaining offered window
size and the newly calculated window size. The newly calculated window
size is always a multiple of the window scaling factor, the remaining
window size however might not be since it depends on rcv_wup/rcv_nxt.
This means we're effectively shrinking the window when scaling it down.
The dump below shows the problem (scaling factor 2^7):
- Window size of 557 (71296) is advertised, up to 3111907257:
IP 172.2.2.3.33000 > 172.2.2.2.33000: . ack 3111835961 win 557 <...>
- New window size of 514 (65792) is advertised, up to 3111907217, 40 bytes
below the last end:
IP 172.2.2.3.33000 > 172.2.2.2.33000: . 3113575668:3113577116(1448) ack 3111841425 win 514 <...>
The number 40 results from downscaling the remaining window:
3111907257 - 3111841425 = 65832
65832 / 2^7 = 514
65832 % 2^7 = 40
If the sender uses up the entire window before it is shrunk, this can have
chaotic effects on the connection. When sending ACKs, tcp_acceptable_seq()
will notice that the window has been shrunk since tcp_wnd_end() is before
tp->snd_nxt, which makes it choose tcp_wnd_end() as sequence number.
This will fail the receivers checks in tcp_sequence() however since it
is before it's tp->rcv_wup, making it respond with a dupack.
If both sides are in this condition, this leads to a constant flood of
ACKs until the connection times out.
Make sure the window is never shrunk by aligning the remaining window to
the window scaling factor.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
With TSO it was possible to send past the receiver window when the skb
to be sent was the last in the write queue while the receiver window
is the limiting factor. One can notice that there's a loophole in the
tcp_mss_split_point that lacked a receiver window check for the
tcp_write_queue_tail() if also cwnd was smaller than the full skb.
Noticed by Thomas Gleixner <tglx@linutronix.de> in form of "Treason
uncloaked! Peer ... shrinks window .... Repaired." messages (the peer
didn't actually shrink its window as the message suggests, we had just
sent something past it without a permission to do so).
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Tested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Updated to incorporate Eric's suggestion of using a per cpu buffer
rather than allocating on the stack. Just a two line change, but will
resend in it's entirety.
Signed-off-by: Glenn Griffin <ggriffin.kernel@gmail.com>
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
This patch removes the no longer used
EXPORT_SYMBOL(sysctl_tcp_tso_win_divisor).
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case of segments which are purely for control without any
data (SYN/ACK/FIN/RST), many fields are set to common values
in multiple places.
i386 results:
$ gcc --version
gcc (GCC) 4.1.2 20070626 (Red Hat 4.1.2-13)
$ codiff tcp_output.o.old tcp_output.o.new
net/ipv4/tcp_output.c:
tcp_xmit_probe_skb | -48
tcp_send_ack | -56
tcp_retransmit_skb | -79
tcp_connect | -43
tcp_send_active_reset | -35
tcp_make_synack | -42
tcp_send_fin | -48
7 functions changed, 351 bytes removed
net/ipv4/tcp_output.c:
tcp_init_nondata_skb | +90
1 function changed, 90 bytes added
tcp_output.o.mid:
8 functions changed, 90 bytes added, 351 bytes removed, diff: -261
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
