Florian Lohoff noticed a bug in mac80211: when bringing the
master interface down while other virtual interfaces are up
we call dev_close() under a spinlock which is not allowed.
This patch removes the sub_if_lock used by mac80211 in favour
of using an RCU list. All list manipulations are already done
under rtnl so are well protected against each other, and the
read-side locks we took in the RX and TX code are already in
RCU read-side critical sections.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: Florian Lohoff <flo@rfc822.org>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Michal Piotrowski <michal.k.k.piotrowski@gmail.com>
Cc: Satyam Sharma <satyam@infradead.org>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The typedef is not required, we can just use "enum ieee80211_key_alg"
instead of "ieee80211_key_alg"
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch changes mac80211 to verify that VLAN interfaces
are valid and not bother drivers about them any more.
VLAN interfaces are now only valid when an AP interface
is up with the same MAC address, and are automatically
turned off when the AP interface is set down.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: Jouni Malinen <j@w1.fi>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Drivers are currently supposed to keep track of monitor
interfaces if they allow so-called "hard" monitor, and
they are also supposed to keep track of multicast etc.
This patch changes that, replaces the set_multicast_list()
callback with a new configure_filter() callback that takes
filter flags (FIF_*) instead of interface flags (IFF_*).
For a driver, this means it should open the filter as much
as necessary to get all frames requested by the filter flags.
Accordingly, the filter flags are named "positively", e.g.
FIF_ALLMULTI.
Multicast filtering is a bit special in that drivers that
have no multicast address filters need to allow multicast
frames through when either the FIF_ALLMULTI flag is set or
when the mc_count value is positive.
At the same time, drivers are no longer notified about
monitor interfaces at all, this means they now need to
implement the start() and stop() callbacks and the new
change_filter_flags() callback. Also, the start()/stop()
ordering changed, start() is now called *before* any
add_interface() as it really should be, and stop() after
any remove_interface().
The patch also changes the behaviour of setting the bssid
to multicast for scanning when IEEE80211_HW_NO_PROBE_FILTERING
is set; the IEEE80211_HW_NO_PROBE_FILTERING flag is removed
and the filter flag FIF_BCN_PRBRESP_PROMISC introduced.
This is a lot more efficient for hardware like b43 that
supports it and other hardware can still set the BSSID
to all-ones.
Driver modifications by Johannes Berg (b43 & iwlwifi), Michael Wu
(rtl8187, adm8211, and p54), Larry Finger (b43legacy), and
Ivo van Doorn (rt2x00).
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Denis V. Lunev <den@sw.ru> noticed that the locking rules
for the network namespace list are over complicated and broken.
In particular the current register_netdev_notifier currently
does not take any lock making the for_each_net iteration racy
with network namespace creation and destruction. Oops.
The fact that we need to use for_each_net in rtnl_unlock() when
the rtnetlink support becomes per network namespace makes designing
the proper locking tricky. In addition we need to be able to call
rtnl_lock() and rtnl_unlock() when we have the net_mutex held.
After thinking about it and looking at the alternatives carefully
it looks like the simplest and most maintainable solution is
to remove net_list_mutex altogether, and to use the rtnl_mutex instead.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/atm/lec.c: In function 'lec_start_xmit':
net/atm/lec.c:371: warning: format '%x' expects type 'unsigned int', but argument 4 has type 'long unsigned int'
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Seems that a bare "depends" is no longer allowed in Sam's kbuild tree.
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since hardware header operations are part of the protocol class
not the device instance, make them into a separate object and
save memory.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Wrap the hard_header_parse function to simplify next step of
header_ops conversion.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add inline for common usage of hardware header creation, and
fix bug in IPV6 mcast where the assumption about negative return is
an errno. Negative return from hard_header means not enough space
was available,(ie -N bytes).
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch makes loopback_dev per network namespace. Adding
code to create a different loopback device for each network
namespace and adding the code to free a loopback device
when a network namespace exits.
This patch modifies all users the loopback_dev so they
access it as init_net.loopback_dev, keeping all of the
code compiling and working. A later pass will be needed to
update the users to use something other than the initial network
namespace.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that multiple loopback devices are becoming possible it makes
the code a little cleaner and more maintainable to test if a deivice
is th a loopback device by testing dev->flags & IFF_LOOPBACK instead
of dev == loopback_dev.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently we never call unregister_netdev for the loopback device so
it is impossible for us to reach inetdev_destroy with the loopback
device. So the test in inetdev_destroy is unnecessary.
Further when testing with my network namespace patches removing
unregistering the loopback device and calling inetdev_destroy works
fine so there appears to be no reason for avoiding unregistering the
loopback device.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch allows you to create a new network namespace
using sys_clone, or sys_unshare.
As the network namespace is still experimental and under development
clone and unshare support is only made available when CONFIG_NET_NS is
selected at compile time.
As this patch introduces network namespace support into code paths
that exist when the CONFIG_NET is not selected there are a few
additions made to net_namespace.h to allow a few more functions
to be used when the networking stack is not compiled in.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When sysfs support is compiled out the kernel still keeps and maintains
the kobject tree. So it is not safe to skip our kobject reference counting or
to avoid becoming members of the kobject tree. It is safe to not add
the networking specific sysfs attributes.
This patch removes the sysfs special cases from net/core/dev.c
renames functions from netdev_sysfs_xxxx to netdev_kobject_xxxx
and always compiles in net-sysfs.c
net-sysfs.c is modified with a CONFIG_SYSFS guard around the parts
that are actually sysfs specific.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since DCCP requires to close both ends of a connection simultaneously,
permission to write in state DCCP_CLOSING is removed in dccp_sendmsg():
* if the sending end closed, it would encounter a write error anyhow;
* if the other end has closed the connection, it accepts no more data.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
This factors code common to dccp_v{4,6}_ctl_send_reset into a separate function,
and adds support for filling in the Data 1 ... Data 3 fields from RFC 4340, 5.6.
It is useful to have this separate, since the following Reset codes will always
be generated from the control socket rather than via dccp_send_reset:
* Code 3, "No Connection", cf. 8.3.1;
* Code 4, "Packet Error" (identification for Data 1 added);
* Code 5, "Option Error" (identification for Data 1..3 added, will be used later);
* Code 6, "Mandatory Error" (same as Option Error);
* Code 7, "Connection Refused" (what on Earth is the difference to "No Connection"?);
* Code 8, "Bad Service Code";
* Code 9, "Too Busy";
* Code 10, "Bad Init Cookie" (not used).
Code 0 is not recommended by the RFC, the following codes would be used in
dccp_send_reset() instead, since they all relate to an established DCCP connection:
* Code 1, "Closed";
* Code 2, "Aborted";
* Code 11, "Aggression Penalty" (12.3).
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
This replaces normal addition with mod-48 addition so that sequence number
wraparound is respected.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
This implements a SHOULD from RFC 4340, 7.5.4:
"To protect against denial-of-service attacks, DCCP implementations SHOULD
impose a rate limit on DCCP-Syncs sent in response to sequence-invalid packets,
such as not more than eight DCCP-Syncs per second."
The rate-limit is maintained on a per-socket basis. This is a more stringent
policy than enforcing the rate-limit on a per-source-address basis and
protects against attacks with forged source addresses.
Moreover, the mechanism is deliberately kept simple. In contrast to
xrlim_allow(), bursts of Sync packets in reply to sequence-invalid packets
are not supported. This foils such attacks where the receipt of a Sync
triggers further sequence-invalid packets. (I have tested this mechanism against
xrlim_allow algorithm for Syncs, permitting bursts just increases the problems.)
In order to keep flexibility, the timeout parameter can be set via sysctl; and
the whole mechanism can even be disabled (which is however not recommended).
The algorithm in this patch has been improved with regard to wrapping issues
thanks to a suggestion by Arnaldo.
Commiter note: Rate limited the step 6 DCCP_WARN too, as it says we're
sending a sync.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
In this patch, duplicated code is removed for the case when a Reset packet is
sent from a connected socket. This code duplication is between dccp_make_reset
and dccp_transmit_skb, which already contained an (up to now entirely unused)
switch statement to fill in the reset code from the DCCP_SKB_CB.
The only thing that has been removed is the call to dst_clone(dst), since
the queue_xmit functions use sk_dst_cache anyway.
I wasn't sure which purpose inet_sk_rebuild_header served, so I left it in.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
This adds fields to support the informational Data 1..3 fields of the
DCCP-Reset packets (RFC 4340, 5.6), and makes minor cosmetic changes
to documentation.
Code which fills in these fields follows in subsequent patches, it is
primarily used for reporting option-processing and feature-negotiation
errors.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
This adds a FIXME to signal that the function dccp_send_delayed_ack is nowhere
used in the entire DCCP/CCID code.
Using a delayed Ack timer is suggested in 11.3 of RFC 4340, but it has also
rather subtle implications for the Ack-Ratio-accounting.
CCID2 does not use this (maybe it should).
I think leaving the function in is good, in case someone wants to implement
this.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
This moves several instances of testing against NULL into the function which is
used to de-reference the CCID-private data.
Committer note: Made the BUG_ON depend on having CONFIG_IP_DCCP_CCID3_DEBUG, as it
is too much to have this on production code. Also made sure that
the macro is used only after checking if sk_state is not LISTEN,
to make it equivalent to what we had before.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
This fixes the code to correspond to RFC 4340, 7.5.4, which states the
exception that a Sync received in state REQUEST generates a Reset (not
a SyncAck).
To achieve this, only a small change is required. Since
dccp_rcv_request_sent_state_process() already uses the correct Reset Code
number 4 ("Packet Error"), we only need to shift the if-statement a few
lines further down.
(To test this case: replace DCCP_PKT_RESPONSE with DCCP_PKT_SYNC
in dccp_make_response.)
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
This patch:
- makes hidp_setup_input() return int to indicate errors;
- checks its return value to handle errors.
And this time it is against -rc7-mm1 tree.
Thanks to roel and Marcel Holtmann for comments.
Signed-off-by: WANG Cong <xiyou.wangcong@gmail.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case of ACK reordering, the SACK block might be valid in it's
time but is already obsoleted since we've received another kind
of confirmation about arrival of the segments through snd_una
advancement of an earlier packet.
I didn't bother to build distinguishing of valid and invalid
SACK blocks but simply made reordered SACK blocks that are too
old always not counted regardless of their "real" validity which
could be determined by using the ack field of the reordered
packet (won't be significant IMHO).
DSACKs can very well be considered useful even in this situation,
so won't do any of this for them.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
I previously added checking to position that is rather poor as
state has already been adjusted quite a bit. Re-placing it above
all state changes should be more robust though the return should
never ever get executed regardless of its place :-).
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
The parameter `seq' of dccp_send_sync() is in fact an acknowledgement number
and not a sequence number - thus renamed by this patch into `ackno'.
Secondly, a `critical' warning is added when a Sync/SyncAck could not be sent.
Sanity: I have checked all other functions that are called in dccp_transmit_skb,
there are no clashes with the use of dccpd_ack_seq; no other function is
using this slot at the same time.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This updates sequence number checking with regard to RFC 4340, 7.5.4.
Missing in the code was an exception for sequence-invalid Reset packets,
which get a Sync acknowledging GSR, instead of (as usual) P.seqno.
This can lead to an oscillating ping-pong flood of Reset packets.
In fact, it has been observed on the wire as follows:
1. client establishes connection to server;
2. before server can write to client, client crashes without notifying
the server (NB: now no longer possible due to ABORT function);
3. server sends DCCP-Data packet (has no ackno);
4. client generates Reset "No Connection", seqno=0, increments seqno;
5. server replies with Sync, using ackno = P.seqno;
6. client generates Reset "No Connection" with seqno = ackno + 1;
7. goto (5).
The difference is that now in (5) the server uses GSR. This causes the
Reset sent by the client in (6) to become sequence-valid, so that in (7)
the vicious circle is broken; the Reset is then enqueued and causes the
socket to enter TIMEWAIT state.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch is in part required by the next patch; it
* replaces 6 instances of `DCCP_SKB_CB(skb)->dccpd_seq' with `seqno';
* replaces 7 instances of `DCCP_SKB_CB(skb)->dccpd_ack_seq' with `ackno';
* replaces 1 use of dccp_inc_seqno() by unfolding `ADD48' macro in place.
No changes in algorithm, all changes are text replacement/substitution.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
The third parameter of dccp_sample_rtt now becomes useless and is removed.
Also combined the subtraction of the timestamp echo and the elapsed time.
This is safe, since (a) presence of timestamp echo is tested first and (b)
elapsed time is either present and non-zero or it is not set and equals 0
due to the memset in dccp_parse_options.
To avoid measuring option-processing time, the timestamp for measuring the
initial Request/Response RTT sample is taken directly when the function is
called (the Linux implementation always adds a timestamp on the Request,
so there is no loss in doing this).
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This provides a timesource, conveniently used for DCCP timestamps, which
returns the elapsed time in 10s of microseconds since initialisation.
This makes for a wrap-around time of about 11.9 hours, which should be
sufficient for most applications.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch reduces the number of timestamps taken in the receive path
for each packet.
The ccid3_hc_tx_update_x() routine is called in
* the receive path for each CCID3-controlled packet
* for the nofeedback timer (if no feedback arrives during 4 RTT)
Currently, when there is no loss, each packet gets timestamped twice.
The patch resolves this by recycling the first timestamp taken on packet
reception for RTT sampling.
When the no_feedback_timer() is called, then the timestamp argument is
simply set to NULL - so that ccid3_hc_tx_update_x() takes care of the logic.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Might as well rename ieee80211_cfg.h to cfg.h to keep things consistent.
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Each station has a vlan_id that is useless. Remove it.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The parse result typedef isn't needed.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It's just painful to have the extra ieee80211_ prefix.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This makes mac80211 print out the wiphy name instead of the
master device name where appropriate.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This fixes a warning about NUM_IEEE80211_MODES missing
in a switch statement. Intentionally do not add a default
case so we get warnings at these places if we need to add
new modes.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes the key threshold stuff from mac80211.
I have patches for later that add it as a per-key setting
to nl/cfg80211.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch allows drivers to indicate bad FCS/PLCP CRC to the stack and
have the stack drop packets like that except for monitor interfaces.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds support for disabling the radio and setting the TXpower
through wext.
This also fixes the prism TXpower ioctl (It always overwrote the TXpower
value in ieee80211_hw_config())
Signed-off-by: Michael Buesch <mb@bu3sch.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It seems I was actually able to hit this deadlock, on my quad G5 softmac
locks up more often than not. This fixes it by using an own workqueue
that can safely be flushed under RTNL.
Not sure if the patch is correct with the workqueue naming. And don't
think with the patch it doesn't continually lock up. It still does, just
doesn't invoke lockdep warnings all the time.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For N cpus, with full throttle traffic on all N CPUs, funneling traffic
to the same ethernet device, the devices queue lock is contended by all
N CPUs constantly. The TX lock is only contended by a max of 2 CPUS.
In the current mode of operation, after all the work of entering the
dequeue region, we may endup aborting the path if we are unable to get
the tx lock and go back to contend for the queue lock. As N goes up,
this gets worse.
The changes in this patch result in a small increase in performance
with a 4CPU (2xdual-core) with no irq binding. Both e1000 and tg3
showed similar behavior;
Signed-off-by: Jamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch replaces all occurences to the static variable
loopback_dev to a pointer loopback_dev. That provides the
mindless, trivial, uninteressting change part for the dynamic
allocation for the loopback.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Acked-By: Kirill Korotaev <dev@sw.ru>
Acked-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There's no reason to clear the sacktag skb hint when small part
of the rexmit queue changes. Account changes (if any) instead when
fragmenting/collapsing. RTO/FRTO do not touch SACKED_ACKED bits so
no need to discard SACK tag hint at all.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
Most of the description that follows comes from my mail to
netdev (some editing done):
Main obstacle to FRTO use is its deployment as it has to be on
the sender side where as wireless link is often the receiver's
access link. Take initiative on behalf of unlucky receivers and
enable it by default in future Linux TCP senders. Also IETF
seems to interested in advancing FRTO from experimental [1].
How does FRTO help?
===================
FRTO detects spurious RTOs and avoids a number of unnecessary
retransmissions and a couple of other problems that can arise
due to incorrect guess made at RTO (i.e., that segments were
lost when they actually got delayed which is likely to occur
e.g. in wireless environments with link-layer retransmission).
Though FRTO cannot prevent the first (potentially unnecessary)
retransmission at RTO, I suspect that it won't cost that much
even if you have to pay for each bit (won't be that high
percentage out of all packets after all :-)). However, usually
when you have a spurious RTO, not only the first segment
unnecessarily retransmitted but the *whole window*. It goes like
this: all cumulative ACKs got delayed due to in-order delivery,
then TCP will actually send 1.5*original cwnd worth of data in
the RTO's slow-start when the delayed ACKs arrive (basically the
original cwnd worth of it unnecessarily). In case one is
interested in minimizing unnecessary retransmissions e.g. due to
cost, those rexmissions must never see daylight. Besides, in the
worst case the generated burst overloads the bottleneck buffers
which is likely to significantly delay the further progress of
the flow. In case of ll rexmissions, ACK compression often
occurs at the same time making the burst very "sharp edged" (in
that case TCP often loses most of the segments above high_seq
=> very bad performance too). When FRTO is enabled, those
unnecessary retransmissions are fully avoided except for the
first segment and the cwnd behavior after detected spurious RTO
is determined by the response (one can tune that by sysctl).
Basic version (non-SACK enhanced one), FRTO can fail to detect
spurious RTO as spurious and falls back to conservative
behavior. ACK lossage is much less significant than reordering,
usually the FRTO can detect spurious RTO if at least 2
cumulative ACKs from original window are preserved (excluding
the ACK that advances to high_seq). With SACK-enhanced version,
the detection is quite robust.
FRTO should remove the need to set a high lower bound for the
RTO estimator due to delay spikes that occur relatively common
in some environments (esp. in wireless/cellular ones).
[1] http://www1.ietf.org/mail-archive/web/tcpm/current/msg02862.html
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
