This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/intel/e1000e/netdev.c
Minor conflict in e1000e, a line that got fixed in 'net'
has been removed in 'net-next'.
Signed-off-by: David S. Miller <davem@davemloft.net>
The bridge multicast fast leave feature was added sufficient space
was not reserved in the netlink message. This means the flag may be
lost in netlink events and results of queries.
Found by observation while looking up some netlink stuff for discussion with Vlad.
Problem introduced by commit c2d3babfaf
Author: David S. Miller <davem@davemloft.net>
Date: Wed Dec 5 16:24:45 2012 -0500
bridge: implement multicast fast leave
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is needed in order to detect if the timestamp option appears
more than once in a packet, to remove the option if the packet is
fragmented, etc. My previous change neglected to store the option
location when the router addresses were prespecified and Pointer >
Length. But now the option location is also stored when Flag is an
unrecognized value, to ensure these option handling behaviors are
still performed.
Signed-off-by: David Ward <david.ward@ll.mit.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>
batadv_iv_ogm_process() accesses the packet using the tt_num_changes
attribute regardless of the real packet len (assuming the length check
was done before). Therefore a length check is needed to avoid reading
random memory.
Signed-off-by: Marek Lindner <lindner_marek@yahoo.de>
Signed-off-by: Antonio Quartulli <ordex@autistici.org>
In 4f6a7e5ee1 we effectively dropped support
for the legacy encoding for the OSDMap and incremental. However, we didn't
fix the decoding for the pgid.
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
Pull networking fixes from David Miller:
1) Missing cancel of work items in mac80211 MLME, from Ben Greear.
2) Fix DMA mapping handling in iwlwifi by using coherent DMA for
command headers, from Johannes Berg.
3) Decrease the amount of pressure on the page allocator by using order
1 pages less in iwlwifi, from Emmanuel Grumbach.
4) Fix mesh PS broadcast OOPS in mac80211, from Marco Porsch.
5) Don't forget to recalculate idle state in mac80211 monitor
interface, from Felix Fietkau.
6) Fix varargs in netfilter conntrack handler, from Joe Perches.
7) Need to reset entire chip when command queue fills up in iwlwifi,
from Emmanuel Grumbach.
8) The TX antenna value must be valid when calibrations are performed
in iwlwifi, fix from Dor Shaish.
9) Don't generate netfilter audit log entries when audit is disabled,
from Gao Feng.
10) Deal with DMA unit hang on e1000e during power state transitions,
from Bruce Allan.
11) Remove BUILD_BUG_ON check from igb driver, from Alexander Duyck.
12) Fix lockdep warning on i2c handling of igb driver, from Carolyn
Wyborny.
13) Fix several TTY handling issues in IRDA ircomm tty driver, from
Peter Hurley.
14) Several QFQ packet scheduler fixes from Paolo Valente.
15) When VXLAN encapsulates on transmit, we have to reset the netfilter
state. From Zang MingJie.
16) Fix jiffie check in net_rx_action() so that we really cap the
processing at 2HZ. From Eric Dumazet.
17) Fix erroneous trigger of IP option space exhaustion, when routers
are pre-specified and we are looking to see if we can insert a
timestamp, we will have the space. From David Ward.
18) Fix various issues in benet driver wrt waiting for firmware to
finish POST after resets or errors. From Gavin Shan and Sathya
Perla.
19) Fix TX locking in SFC driver, from Ben Hutchings.
20) Like the VXLAN fix above, when we encap in a TUN device we have to
reset the netfilter state. This should fix several strange crashes
reported by Dave Jones and others. From Eric Dumazet.
21) Don't forget to clean up MAC address resources when shutting down a
port in mlx4 driver, from Yan Burman.
22) Fix divide by zero in vmxnet3 driver, from Bhavesh Davda.
23) Fix device statistic regression in tg3 when the driver is using
phylib, from Nithin Sujir.
24) Fix info leak in several netlink handlers, from Mathias Krause.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (79 commits)
6lowpan: Fix endianness issue in is_addr_link_local().
rrunner.c: fix possible memory leak in rr_init_one()
dcbnl: fix various netlink info leaks
rtnl: fix info leak on RTM_GETLINK request for VF devices
bridge: fix mdb info leaks
tg3: Update link_up flag for phylib devices
ipv6: stop multicast forwarding to process interface scoped addresses
bridging: fix rx_handlers return code
netlabel: fix build problems when CONFIG_IPV6=n
drivers/isdn: checkng length to be sure not memory overflow
net/rds: zero last byte for strncpy
bnx2x: Fix SFP+ misconfiguration in iSCSI boot scenario
bnx2x: Fix intermittent long KR2 link up time
macvlan: Set IFF_UNICAST_FLT flag to prevent unnecessary promisc mode.
team: unsyc the devices addresses when port is removed
bridge: add missing vid to br_mdb_get()
Fix: sparse warning in inet_csk_prepare_forced_close
afkey: fix a typo
MAINTAINERS: Update qlcnic maintainers list
netlabel: correctly list all the static label mappings
...
Some devices have multicast filter capability for each individual
virtual interface rather than just a global one. Add an interface
specific driver callback allowing such drivers to configure this.
Signed-off-by: Alexander Bondar <alexander.bondar@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Instead of calling synchronize_net() for every key
on an interface or when a station is removed, do it
only once for all keys in both of these cases.
As a side-effect, removing station keys now always
calls synchronize_net() even if there are no keys,
which fixes an issue with station removal happening
in the driver while the station could still be used
for TX.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
When a key is allocated but not really added, there's no
need to go through the entire teardown process. Also, if
adding a key fails, ieee80211_key_link() can take care of
freeing it instead of the (only) caller.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
If a P2P Device interface receives an unhandled action
frame, we attempt to return it. This crashes because it
doesn't have a channel context. Fix the crash by using
status->band and properly mark the return frame as an
off-channel frame.
Reported-by: Ilan Peer <ilan.peer@intel.com>
Reviewed-by: Ilan Peer <ilan.peer@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
If the remote LLC doesn't reply in time to our SNL requests we remove
them from the list of pending requests. The timeout is fixed to an
arbitrary value of 3 times remote_lto.
When not replied, the local LLC broadcasts NFC_EVENT_LLC_SDRES nl events for
the concerned uris with sap values set to LLCP_SDP_UNBOUND (which is 65).
Signed-off-by: Thierry Escande <thierry.escande@linux.intel.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
This adds a netlink interface for service name lookup support.
Multiple URIs can be passed nested into the NFC_ATTR_LLC_SDP attribute
using the NFC_CMD_LLC_SDREQ netlink command.
When the SNL reply is received, a NFC_EVENT_LLC_SDRES event is sent to
the user space. URI and SAP tuples are passed back, nested into
NFC_ATTR_LLC_SDP attribute.
Signed-off-by: Thierry Escande <thierry.escande@linux.intel.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
This modifies the way SDRES PDUs are sent back. If multiple SDREQs are
received within a single SNL PDU, all SDRES replies are sent packed in
one SNL PDU too.
Signed-off-by: Thierry Escande <thierry.escande@linux.intel.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
Some LLCP services (e.g. the validation ones) require some control over
the LLCP link parameters like the receive window (RW) or the MIU extension
(MIUX). This can only be done through socket options.
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
They really are remote peer parameters, and we need to distinguish them
from the local ones as we'll modify the latter with socket options.
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
Similar to iptunnel_xmit(), group these operations into a
helper function.
This by the way fixes the missing u64_stats_update_begin()
and u64_stats_update_end() for 32 bit arch.
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Pravin B Shelar <pshelar@nicira.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
nla_data() cannot return NULL, so these NULL pointer checks are
superfluous.
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The dcb netlink interface leaks stack memory in various places:
* perm_addr[] buffer is only filled at max with 12 of the 32 bytes but
copied completely,
* no in-kernel driver fills all fields of an IEEE 802.1Qaz subcommand,
so we're leaking up to 58 bytes for ieee_ets structs, up to 136 bytes
for ieee_pfc structs, etc.,
* the same is true for CEE -- no in-kernel driver fills the whole
struct,
Prevent all of the above stack info leaks by properly initializing the
buffers/structures involved.
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Initialize the mac address buffer with 0 as the driver specific function
will probably not fill the whole buffer. In fact, all in-kernel drivers
fill only ETH_ALEN of the MAX_ADDR_LEN bytes, i.e. 6 of the 32 possible
bytes. Therefore we currently leak 26 bytes of stack memory to userland
via the netlink interface.
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The bridging code discloses heap and stack bytes via the RTM_GETMDB
netlink interface and via the notify messages send to group RTNLGRP_MDB
afer a successful add/del.
Fix both cases by initializing all unset members/padding bytes with
memset(0).
Cc: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With recent patches from Pravin, most tunnels can't use iptunnel_xmit()
any more, due to ip_select_ident() and skb->ip_summed. But we can just
move these operations out of iptunnel_xmit(), so that tunnels can
use it again.
This by the way fixes a bug in vxlan (missing nf_reset()) for net-next.
Cc: Pravin B Shelar <pshelar@nicira.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull namespace bugfixes from Eric Biederman:
"This is three simple fixes against 3.9-rc1. I have tested each of
these fixes and verified they work correctly.
The userns oops in key_change_session_keyring and the BUG_ON triggered
by proc_ns_follow_link were found by Dave Jones.
I am including the enhancement for mount to only trigger requests of
filesystem modules here instead of delaying this for the 3.10 merge
window because it is both trivial and the kind of change that tends to
bit-rot if left untouched for two months."
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
proc: Use nd_jump_link in proc_ns_follow_link
fs: Limit sys_mount to only request filesystem modules (Part 2).
fs: Limit sys_mount to only request filesystem modules.
userns: Stop oopsing in key_change_session_keyring
Allow IPIP to make use of tx-checksum offloading.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
tunnel_ip_select_ident() is more efficient when generating ip-header
id given inner packet is of ipv4 type.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add header with function definitions to quiet warnings and avoid future errors.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adds generic tunneling offloading support for IPv4-UDP based
tunnels.
GSO type is added to request this offload for a skb.
netdev feature NETIF_F_UDP_TUNNEL is added for hardware offloaded
udp-tunnel support. Currently no device supports this feature,
software offload is used.
This can be used by tunneling protocols like VXLAN.
CC: Jesse Gross <jesse@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds inner mac header. This will be used in next patch
to find tunner header length. Header len is required to copy tunnel
header to each gso segment.
This patch does not change any functionality.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This function will be used in next VXLAN_GSO patch. This patch does
not change any functionality.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Inherit scatergather feature for tunnel devices to avoid
copy for TSO packets of tunneling device like GRE.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Earlier SG was unset if CSUM was not available for given device to
force skb copy to avoid sending inconsistent csum.
Commit c9af6db4c1 (net: Fix possible wrong checksum generation)
added explicit flag to force copy to fix this issue. Therefore
there is no need to link SG and CSUM, following patch kills this
link between there two features.
This patch is also required following patch in series.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The opcode in cmd_complete and cmd_status events is 16 bits, so we
should only be comparing it after having converted it to the host
endianness. There's already an opcode variable in both functions which
is in host endiannes so the right fix is to just start using it instead
of ev->opcode.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Since no one checks the returning value of hci_req_add and HCI
request errors are now handled in hci_req_run, we can make hci_
req_add returning void.
Signed-off-by: Andre Guedes <andre.guedes@openbossa.org>
Acked-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
When we are building a HCI request with more than one HCI command
and one of the hci_req_add calls fail, we should have some cleanup
routine so the HCI commands already queued on HCI request can be
deleted. Otherwise, we will face some memory leaks issues.
This patch implements the HCI request error handling which is the
following: If a hci_req_add fails, we save the error code in hci_
request. Once hci_req_run is called, we verify the error field. If
it is different from zero, we delete all HCI commands already queued
and return the error code.
Signed-off-by: Andre Guedes <andre.guedes@openbossa.org>
Acked-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
In case the HCI request queue is empty, hci_req_run should return
ENODATA instead of EINVAL. This way, hci_req_run returns a more
meaningful error value.
Signed-off-by: Andre Guedes <andre.guedes@openbossa.org>
Acked-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
If hci_req_run returns error, we erroneously leave the current
process in TASK_INTERRUPTABLE state. If we leave the process in
TASK_INTERRUPTABLE and it is preempted, this process will never
be scheduled again.
This patch fixes this issue by moving the preparation for scheduling
(add to waitqueue and set process state) to just after the hci_req_run
call.
Signed-off-by: Andre Guedes <andre.guedes@openbossa.org>
Acked-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
