No caller or macro uses the return value so make all
the functions return void.
Signed-off-by: NJoe Perches <joe@perches.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The hci_recv_fragment function is no longer used by any driver and thus
do not export it. In fact it is not even needed by the core and it can
be removed altogether.
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>

The Bluetooth core already does processing of the HCI command header
and puts it together before sending it to the driver. It is not really
efficient for the driver to look at the HCI command header again in
case it has to make certain decisions about certain commands. To make
this easier, just provide the opcode as part of the SKB control buffer
information. The extra information about the opcode is optional and
only provided for HCI commands.
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>

Whether through HCI with BR/EDR or SMP with LE when authentication fails
we should also notify any pending Pair Device mgmt command. This patch
updates the mgmt_auth_failed function to take the actual hci_conn object
and makes sure that any pending pairing command is notified and cleaned
up appropriately.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Before the move the l2cap_chan the SMP context (smp_chan) didn't have
any kind of proper locking. The best there existed was the
HCI_CONN_LE_SMP_PEND flag which was used to enable mutual exclusion for
potential multiple creators of the SMP context.

Now that SMP has been converted to use the l2cap_chan infrastructure and
since the SMP context is directly mapped to a corresponding l2cap_chan
we get the SMP context locking essentially for free through the
l2cap_chan lock. For all callbacks that l2cap_core.c makes for each
channel implementation (smp.c in the case of SMP) the l2cap_chan lock is
held through l2cap_chan_lock(chan).

Since the calls from l2cap_core.c to smp.c are covered the only missing
piece to have the locking implemented properly is to ensure that the
lock is held for any other call path that may access the SMP context.
This means user responses through mgmt.c, requests to elevate the
security of a connection through hci_conn.c, as well as any deferred
work through workqueues.

This patch adds the necessary locking to all these other code paths that
try to access the SMP context. Since mutual exclusion for the l2cap_chan
access is now covered from all directions the patch also removes
unnecessary HCI_CONN_LE_SMP_PEND flag (once we've acquired the chan lock
we can simply check whether chan->smp is set to know if there's an SMP
context).
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The identity address update of all channels for an l2cap_conn needs to
take the lock for each channel, i.e. it's safest to do this by a
separate workqueue callback.

Previously this was partially solved by moving the entire SMP key
distribution behind a workqueue. However, if we want SMP context locking
to be correct and safe we should always use the l2cap_chan lock when
accessing it, meaning even smp_distribute_keys needs to take that lock
which would once again create a dead lock when updating the identity
address.

The simplest way to solve this is to have l2cap_conn manage the deferred
work which is what this patch does. A subsequent patch will remove the
now unnecessary SMP key distribution work struct.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

We'll soon use hci_disconnect() from places that are interested to know
whether the hci_send_cmd() really succeeded or not. This patch updates
hci_disconnect() to pass on any error returned from hci_send_cmd().
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Now that there are no more users of the l2cap_conn_shutdown API (since
smp.c switched to using hci_disconnect) we can simply remove it along
with all of it's l2cap_conn variables.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

When hci_chan_del is called the disconnection routines get scheduled
through a workqueue. If there's any incoming ACL data before the
routines get executed there's a chance that a new hci_chan is created
and the disconnection never happens. This patch adds a new hci_conn flag
to indicate that we're in the process of driving the connection down. We
set the flag in hci_chan_del and check for it in hci_chan_create so that
no new channels are created for the same connection.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

There's no point in passing a "small" timeout to queue_delayed_work() to
try to get the callback faster scheduled. Passing 0 is perfectly valid
and will cause a shortcut to a direct queue_work().
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

It's natural to have *_get() functions that increment the reference
count of an object to return the object type itself. This way it's
simple to make a copy of the object pointer and increase the reference
count in a single step. This patch updates two such get() functions,
namely hci_conn_get() and l2cap_conn_get(), and updates the users to
take advantage of the new API.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Now that SMP has been converted to use fixed channels we've got a bit of
a problem with the hci_conn reference counting. So far the L2CAP code
has kept a reference for each L2CAP channel that was notified of the
connection. With SMP however this would mean that the connection is
never dropped even though there are no other users of it. Furthermore,
SMP already does its own hci_conn reference counting internally,
starting from a security or pairing request and ending with the key
distribution.

This patch makes L2CAP fixed channels default to the L2CAP core not
keeping a hci_conn reference for them. A new FLAG_HOLD_HCI_CONN flag is
added so that L2CAP users can declare an exception to this rule and hold
a reference even for their fixed channels. One such exception is the
L2CAP socket layer which does want a reference for each socket (e.g. an
ATT socket which uses a fixed channel).
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

When using the cfg80211_inform_bss[_width]() functions drivers
cannot currently indicate whether the data was received in a
beacon or probe response. Fix that by passing a new enum that
indicates such (or unknown).

For good measure, use it in ath6kl.

Acked-by: Kalle Valo <kvalo@qca.qualcomm.com> [ath6kl]
Acked-by: Arend van Spriel <arend@broadcom.com> [brcmfmac]
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

There are a few possible cases of where BSS data came from:
 1) only a beacon has been received
 2) only a probe response has been received
 3) the driver didn't report what it received (this happens when
    using cfg80211_inform_bss[_width]())
 4) both probe response and beacon data has been received

Unfortunately, in the userspace API, a few things weren't there:
 a) there was no way to differentiate cases 1) and 4) above
    without comparing the data of the IEs
 b) the TSF was always from the last frame, instead of being
    exposed for beacon/probe response separately like IEs

Fix this by
   i) exporting a new flag attribute that indicates whether or
      not probe response data has been received - this addresses (a)
  ii) exporting a BEACON_TSF attribute that holds the beacon's TSF
      if a beacon has been received
 iii) not exporting the beacon attributes in case (3) above as that
      would just lead userspace into thinking the data actually came
      from a beacon when that isn't clear

To implement this, track inside the IEs struct whether or not it
(definitely) came from a beacon.

Reported-by: William Seto
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Header-less cloned skbs with sufficient headroom need not be cloned
unless the tailroom is going to be modified.

Fix ieee80211_skb_resize so it would only resize cloned skbs if either
the header isn't released or the tailroom is going to be modified.

Some drivers might have assumed that skbs are never cloned, so add a HW
flag that explicitly permits cloned TX skbs. Drivers which do not modify
TX skbs should set this flag to avoid copying skbs.
Signed-off-by: NIdo Yariv <idox.yariv@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

When hw acceleration is enabled, the GENERATE_IV or PUT_IV_SPACE flags
will only require headroom space. Consequently, the tailroom-needed
counter can safely be decremented.
Signed-off-by: NIdo Yariv <idox.yariv@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

In the cfg80211_rx_mgmt(), parameter @gfp was used for the memory allocation.
But, memory get allocated under spin_lock_bh(), this implies atomic context.
So, one can't use GFP_KERNEL, only variants with no __GFP_WAIT. Actually, in all
occurrences GFP_ATOMIC is used (wil6210 use GFP_KERNEL by mistake),
and it should be this way or warning triggered in the memory allocation code.

Remove @gfp parameter as no actual choice exist, and use hard coded
GFP_ATOMIC for memory allocation.
Signed-off-by: NVladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Recently the LE passive scanning and auto-connections feature was
introduced. It uses the hci_connect_le() API which returns a hci_conn
along with a reference count to that object. All previous users would
tie this returned reference to some existing object, such as an L2CAP
channel, and there'd be no leaked references this way. For
auto-connections however the reference was returned but not stored
anywhere, leaving established connections with one higher reference
count than they should have.

Instead of playing special tricks with hci_conn_hold/drop this patch
associates the returned reference from hci_connect_le() with the object
that in practice does own this reference, i.e. the hci_conn_params
struct that caused us to initiate a connection in the first place. Once
the connection is established or fails to establish this reference is
removed appropriately.

One extra thing needed is to call hci_pend_le_actions_clear() before
calling hci_conn_hash_flush() so that the reference is cleared before
the hci_conn objects are fully removed.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

This patch drops the userspace accessable sysfs entry for the maximum
datagram size of a 6LoWPAN fragment packet.

A fragment should not have a datagram size value greater than 1280 byte.
Instead of make this value configurable, we accept 1280 datagram size
fragment packets only.
Signed-off-by: NMartin Townsend <martin.townsend@xsilon.com>
Signed-off-by: NAlexander Aring <alex.aring@gmail.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

We currently track the QoS capability twice: for all peer stations
in the WLAN_STA_WME flag, and for any clients associated to an AP
interface separately for drivers in the sta->sta.wme field.

Remove the WLAN_STA_WME flag and track the capability only in the
driver-visible field, getting rid of the limitation that the field
is only valid in AP mode.
Reviewed-by: NArik Nemtsov <arik@wizery.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

alpha2 is defined as 2-chars array, but is used in multiple
places as string (e.g. with nla_put_string calls), which
might leak kernel data.

Solve it by simply adding an extra char for the NULL
terminator, making such operations safe.

Cc: stable@vger.kernel.org
Signed-off-by: NEliad Peller <eliadx.peller@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

tcp_tw_recycle heavily relies on tcp timestamps to build a per-host
ordering of incoming connections and teardowns without the need to
hold state on a specific quadruple for TCP_TIMEWAIT_LEN, but only for
the last measured RTO. To do so, we keep the last seen timestamp in a
per-host indexed data structure and verify if the incoming timestamp
in a connection request is strictly greater than the saved one during
last connection teardown. Thus we can verify later on that no old data
packets will be accepted by the new connection.

During moving a socket to time-wait state we already verify if timestamps
where seen on a connection. Only if that was the case we let the
time-wait socket expire after the RTO, otherwise normal TCP_TIMEWAIT_LEN
will be used. But we don't verify this on incoming SYN packets. If a
connection teardown was less than TCP_PAWS_MSL seconds in the past we
cannot guarantee to not accept data packets from an old connection if
no timestamps are present. We should drop this SYN packet. This patch
closes this loophole.

Please note, this patch does not make tcp_tw_recycle in any way more
usable but only adds another safety check:
Sporadic drops of SYN packets because of reordering in the network or
in the socket backlog queues can happen. Users behing NAT trying to
connect to a tcp_tw_recycle enabled server can get caught in blackholes
and their connection requests may regullary get dropped because hosts
behind an address translator don't have synchronized tcp timestamp clocks.
tcp_tw_recycle cannot work if peers don't have tcp timestamps enabled.

In general, use of tcp_tw_recycle is disadvised.

Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Florian Westphal <fw@strlen.de>
Signed-off-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Make sure we use the correct address-family-specific function for
handling MTU reductions from within tcp_release_cb().

Previously AF_INET6 sockets were incorrectly always using the IPv6
code path when sometimes they were handling IPv4 traffic and thus had
an IPv4 dst.
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Diagnosed-by: NWillem de Bruijn <willemb@google.com>
Fixes: 563d34d0 ("tcp: dont drop MTU reduction indications")
Reviewed-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We don't know right timestamp for repaired skb-s. Wrong RTT estimations
isn't good, because some congestion modules heavily depends on it.

This patch adds the TCPCB_REPAIRED flag, which is included in
TCPCB_RETRANS.

Thanks to Eric for the advice how to fix this issue.

This patch fixes the warning:
[  879.562947] WARNING: CPU: 0 PID: 2825 at net/ipv4/tcp_input.c:3078 tcp_ack+0x11f5/0x1380()
[  879.567253] CPU: 0 PID: 2825 Comm: socket-tcpbuf-l Not tainted 3.16.0-next-20140811 #1
[  879.567829] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[  879.568177]  0000000000000000 00000000c532680c ffff880039643d00 ffffffff817aa2d2
[  879.568776]  0000000000000000 ffff880039643d38 ffffffff8109afbd ffff880039d6ba80
[  879.569386]  ffff88003a449800 000000002983d6bd 0000000000000000 000000002983d6bc
[  879.569982] Call Trace:
[  879.570264]  [<ffffffff817aa2d2>] dump_stack+0x4d/0x66
[  879.570599]  [<ffffffff8109afbd>] warn_slowpath_common+0x7d/0xa0
[  879.570935]  [<ffffffff8109b0ea>] warn_slowpath_null+0x1a/0x20
[  879.571292]  [<ffffffff816d0a05>] tcp_ack+0x11f5/0x1380
[  879.571614]  [<ffffffff816d10bd>] tcp_rcv_established+0x1ed/0x710
[  879.571958]  [<ffffffff816dc9da>] tcp_v4_do_rcv+0x10a/0x370
[  879.572315]  [<ffffffff81657459>] release_sock+0x89/0x1d0
[  879.572642]  [<ffffffff816c81a0>] do_tcp_setsockopt.isra.36+0x120/0x860
[  879.573000]  [<ffffffff8110a52e>] ? rcu_read_lock_held+0x6e/0x80
[  879.573352]  [<ffffffff816c8912>] tcp_setsockopt+0x32/0x40
[  879.573678]  [<ffffffff81654ac4>] sock_common_setsockopt+0x14/0x20
[  879.574031]  [<ffffffff816537b0>] SyS_setsockopt+0x80/0xf0
[  879.574393]  [<ffffffff817b40a9>] system_call_fastpath+0x16/0x1b
[  879.574730] ---[ end trace a17cbc38eb8c5c00 ]---

v2: moving setting of skb->when for repaired skb-s in tcp_write_xmit,
    where it's set for other skb-s.

Fixes: 431a9124 ("tcp: timestamp SYN+DATA messages")
Fixes: 740b0f18 ("tcp: switch rtt estimations to usec resolution")
Cc: Eric Dumazet <edumazet@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: NAndrey Vagin <avagin@openvz.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now that there are no-longer any users for l2cap_conn->security_timer we
can go ahead and simply remove it. The patch makes initialization of the
conn->info_timer unconditional since it's better not to leave any
l2cap_conn data structures uninitialized no matter what the underlying
transport.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Since we no-longer do special handling of SMP within l2cap_core.c we
don't have any code for calling l2cap_conn_del() when smp.c doesn't like
the data it gets. At the same time we cannot simply export
l2cap_conn_del() since it will try to lock the channels it calls into
whereas we already hold the lock in the smp.c l2cap_chan callbacks (i.e.
it'd lead to a deadlock).

This patch adds a new l2cap_conn_shutdown() API which is very similar to
l2cap_conn_del() except that it defers the call to l2cap_conn_del()
through a workqueue, thereby making it safe to use it from an L2CAP
channel callback.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Now that we have all the necessary pieces in place we can fully convert
SMP to use the L2CAP channel infrastructure. This patch adds the
necessary callbacks and removes the now unneeded conn->smp_chan pointer.

One notable behavioral change in this patch comes from the following
code snippet:

-       case L2CAP_CID_SMP:
-               if (smp_sig_channel(conn, skb))
-                       l2cap_conn_del(conn->hcon, EACCES);

This piece of code was essentially forcing a disconnection if garbage
SMP data was received. The l2cap_conn_del() function is private to
l2cap_conn.c so we don't have access to it anymore when using the L2CAP
channel callbacks. Therefore, the behavior of the new code is simply to
return errors in the recv() callback (which is simply the old
smp_sig_channel()), but no disconnection will occur.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Now that we have per-adapter SMP data thanks to the root SMP L2CAP
channel we can take advantage of it and attach the AES crypto context
(only used for SMP) to it. This means that the smp_irk_matches() and
smp_generate_rpa() function can be converted to internally handle the
AES context.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

This patch creates the initial SMP L2CAP channels and a skeleton for
their callbacks. There is one per-adapter channel created upon adapter
registration, and then one channel per-connection created through the
new_connection callback. The channels are registered with the reserved
CID 0x1f for now in order to not conflict with existing SMP
functionality. Once everything is in place the value can be changed to
what it should be, i.e. L2CAP_CID_SMP.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

In preparation for converting SMP to use l2cap_chan it's useful to add a
few more callback helpers so that smp.c won't need to define all of its
own.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

The LE ATT socket uses a special trick where it temporarily sets
BT_CONFIG state for the duration of a security level elevation. In order
to not require special hacks for going back to BT_CONNECTED state in the
l2cap_core.c code the most reasonable place to resume the state is the
resume callback. This patch adds a new flag to track the pending
security level change and ensures that the state is set back to
BT_CONNECTED in the resume callback in case the flag is set.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

Similar to our hci_update_background_scan() function we can simplify a
lot of code by creating a unified helper function for doing page scan
updates. This patch adds such a function to hci_core.c and updates all
the relevant places to use it.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

There are several situations where we're interested in knowing whether
we're currently in the process of powering off an adapter. This patch
adds a convenience function for the purpose and makes it public since
we'll soon need to access it from hci_event.c as well.
Signed-off-by: NJohan Hedberg <johan.hedberg@intel.com>
Signed-off-by: NMarcel Holtmann <marcel@holtmann.org>

When I added the netlbl_catmap_setlong() function I mistakenly forgot
to mark the associated dummy function as an inline.
Reported-by: NStephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: NPaul Moore <pmoore@redhat.com>

sock_tx_timestamp() should not ignore initial *tx_flags value, as TCP
stack can store SKBTX_SHARED_FRAG in it.

Also first argument (struct sock *) can be const.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Fixes: 4ed2d765 ("net-timestamp: TCP timestamping")
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Datagrams timestamped on transmission can coexist in the kernel stack
and be reordered in packet scheduling. When reading looped datagrams
from the socket error queue it is not always possible to unique
correlate looped data with original send() call (for application
level retransmits). Even if possible, it may be expensive and complex,
requiring packet inspection.

Introduce a data-independent ID mechanism to associate timestamps with
send calls. Pass an ID alongside the timestamp in field ee_data of
sock_extended_err.

The ID is a simple 32 bit unsigned int that is associated with the
socket and incremented on each send() call for which software tx
timestamp generation is enabled.

The feature is enabled only if SOF_TIMESTAMPING_OPT_ID is set, to
avoid changing ee_data for existing applications that expect it 0.
The counter is reset each time the flag is reenabled. Reenabling
does not change the ID of already submitted data. It is possible
to receive out of order IDs if the timestamp stream is not quiesced
first.
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

sk_flags is reaching its limit. New timestamping options will not fit.
Move all of them into a new field sk->sk_tsflags.

Added benefit is that this removes boilerplate code to convert between
SOF_TIMESTAMPING_.. and SOCK_TIMESTAMPING_.. in getsockopt/setsockopt.

SOCK_TIMESTAMPING_RX_SOFTWARE is also used to toggle the receive
timestamp logic (netstamp_needed). That can be simplified and this
last key removed, but will leave that for a separate patch.
Signed-off-by: NWillem de Bruijn <willemb@google.com>

----

The u16 in sock can be moved into a 16-bit hole below sk_gso_max_segs,
though that scatters tstamp fields throughout the struct.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Applications that request kernel tx timestamps with SO_TIMESTAMPING
read timestamps as recvmsg() ancillary data. The response is defined
implicitly as timespec[3].

1) define struct scm_timestamping explicitly and

2) add support for new tstamp types. On tx, scm_timestamping always
   accompanies a sock_extended_err. Define previously unused field
   ee_info to signal the type of ts[0]. Introduce SCM_TSTAMP_SND to
   define the existing behavior.

The reception path is not modified. On rx, no struct similar to
sock_extended_err is passed along with SCM_TIMESTAMPING.
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This commit reduces spurious retransmits due to apparent SACK reneging
by only reacting to SACK reneging that persists for a short delay.

When a sequence space hole at snd_una is filled, some TCP receivers
send a series of ACKs as they apparently scan their out-of-order queue
and cumulatively ACK all the packets that have now been consecutiveyly
received. This is essentially misbehavior B in "Misbehaviors in TCP
SACK generation" ACM SIGCOMM Computer Communication Review, April
2011, so we suspect that this is from several common OSes (Windows
2000, Windows Server 2003, Windows XP). However, this issue has also
been seen in other cases, e.g. the netdev thread "TCP being hoodwinked
into spurious retransmissions by lack of timestamps?" from March 2014,
where the receiver was thought to be a BSD box.

Since snd_una would temporarily be adjacent to a previously SACKed
range in these scenarios, this receiver behavior triggered the Linux
SACK reneging code path in the sender. This led the sender to clear
the SACK scoreboard, enter CA_Loss, and spuriously retransmit
(potentially) every packet from the entire write queue at line rate
just a few milliseconds before the ACK for each packet arrives at the
sender.

To avoid such situations, now when a sender sees apparent reneging it
does not yet retransmit, but rather adjusts the RTO timer to give the
receiver a little time (max(RTT/2, 10ms)) to send us some more ACKs
that will restore sanity to the SACK scoreboard. If the reneging
persists until this RTO then, as before, we clear the SACK scoreboard
and enter CA_Loss.

A 10ms delay tolerates a receiver sending such a stream of ACKs at
56Kbit/sec. And to allow for receivers with slower or more congested
paths, we wait for at least RTT/2.

We validated the resulting max(RTT/2, 10ms) delay formula with a mix
of North American and South American Google web server traffic, and
found that for ACKs displaying transient reneging:

 (1) 90% of inter-ACK delays were less than 10ms
 (2) 99% of inter-ACK delays were less than RTT/2

In tests on Google web servers this commit reduced reneging events by
75%-90% (as measured by the TcpExtTCPSACKReneging counter), without
any measurable impact on latency for user HTTP and SPDY requests.
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Use kmem_cache to allocate/free inet_frag_queue objects since they're
all the same size per inet_frags user and are alloced/freed in high volumes
thus making it a perfect case for kmem_cache.
Signed-off-by: NNikolay Aleksandrov <nikolay@redhat.com>
Acked-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>