[MAC80211]: remove unused ioctls (3)

The ioctls
 * PRISM2_PARAM_RADAR_DETECT
 * PRISM2_PARAM_SPECTRUM_MGMT

are not used by hostapd or wpa_supplicant,

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[MAC80211]: remove unused ioctls (2)

The ioctls

 * PRISM2_PARAM_STA_ANTENNA_SEL
 * PRISM2_PARAM_TX_POWER_REDUCTION
 * PRISM2_PARAM_DEFAULT_WEP_ONLY

are not used by hostapd or wpa_supplicant.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[MAC80211]: remove unused ioctls (1)

The ioctls

 * PRISM2_PARAM_ANTENNA_MODE
 * PRISM2_PARAM_STAT_TIME

are not used by hostapd or wpa_supplicant.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[MAC80211]: fix software decryption

When doing key selection for software decryption, mac80211 gets
a few things wrong: it always uses pairwise keys if configured,
even if the frame is addressed to a multicast address. Also, it
doesn't allow using a key index of zero if a pairwise key has
also been found.

This patch changes the key selection code to be (more) in line
with the 802.11 specification. I have confirmed that with this,
multicast frames are correctly decrypted and I've tested with
WEP as well.

While at it, I've cleaned up the semantics of the hardware flags
IEEE80211_HW_WEP_INCLUDE_IV and IEEE80211_HW_DEVICE_HIDES_WEP
and clarified them in the mac80211.h header; it is also now
allowed to set the IEEE80211_HW_DEVICE_HIDES_WEP option even if
it only applies to frames that have been decrypted by the hw,
unencrypted frames must be dropped but encrypted frames that
the hardware couldn't handle can be passed up unmodified.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

[MAC80211]: remove radar stuff

Unused in drivers, userspace and mac80211.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[MAC80211]: remove ieee80211_msg_wep_frame_unknown_key

Neither hostapd nor wpa_supplicant really use it.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

[MAC80211]: ratelimit some RX messages

Many if not all of these messages can be triggered by sending
a few rogue frames which is trivially done and then we overflow
our logs.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[MAC80211]: remove IEEE80211_HW_DATA_NULLFUNC_ACK

Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[MAC80211]: remove PRISM2_PARAM_RADIO_ENABLED

This now is unused in hostapd/wpa_supplicant.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[MAC80211]: remove IEEE80211_HW_HOST_GEN_BEACON flag

The flag is never checked because drivers can simply call
ieee80211_beacon_get() regardless of setting this flag.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[MAC80211]: remove reset callback

The callback isn't used so remove it.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[MAC80211]: fix key debugfs

This fixes two issues with the key debugfs:
 1) key index obviously isn't unique
 2) various missing break statements led to bogus output

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[MAC80211]: avoid copying packets to interfaces that are down

David Woodhouse noticed that under some circumstances the number of slab
allocations kept growing. After looking a bit, this seemed to happen
when you had a management mode interface that was *down*.

The reason for this is that when the device is down, all management
frames get queued to the in-kernel MLME (via ieee80211_sta_rx_mgmt) but
then the sta work is invoked but doesn't run when the netif is down.
When you then bring the interface up, all such frames are freed, but if
you change the mode all of them are lost because the skb queue is
reinitialised as soon as you go back to managed mode. The skb queue is
correctly cleared when the interface is brought down, but the code
doesn't account for the fact that it may be filled while it is not up.

This patch should fix the issue by simply ignoring all interfaces that
are down when going through the RX handlers.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Michael Wu <flamingice@sourmilk.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IrDA]: Kingsun KS-959 IrDA USB driver

This dongle does not follow the usb-irda specification, so it needs its own
special driver. First, it uses control URBs for data transfer, instead of
bulk or interrupt transfers; the only interrupt endpoint exposed seems to
be a dummy to prevent the interface from being rejected. Second, it uses
obfuscation and padding at the USB traffic level, for no apparent reason
other than to make reverse engineering harder (full details on obfuscation
in comments at beginning of source). Although it is advertised as a "4 Mbps
FIR dongle", it apparently loses packets at speeds greater than 57600 bps.

On plugin, this dongle reports vendor and device IDs: 0x07d0:0x4959 .

The Windows driver that is used normally to control this dongle has a
filename of KS-959.SYS .

Signed-off-by: Alex Villacís Lasso <a_villacis@palosanto.com>
Signed-off-by: Samuel Ortiz <samuel@sortiz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IrDA]: Kingsun Dazzle IrDA USB driver

This dongle does not follow the usb-irda specification, so it needs its own
special driver. Just like the Kingsun/Donshine dongle, it exposes two
interrupt endpoints. Reception is performed through direct reads from the
input endpoint. Transmission requires splitting the IrDA frames into 8-byte
segments, in which the first byte encodes how many of the remaining 7 bytes
are used as data. Speed change is made with a control URB just like the one
in cypress_m8, and it seems to support up to 115200 bps.

On plugin, this dongle reports vendor and device IDs: 0x07d0:0x4100

Signed-off-by: Alex Villacís Lasso <a_villacis@palosanto.com>
Signed-off-by: Samuel Ortiz <samuel@sortiz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IrDA]: MSG_NOSIGNAL support for IrDA sockets

Signed-off-by: Samuel Ortiz <samuel@sortiz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: DIV_ROUND_UP cleanup (part two)

Hopefully captured all single statement cases under net/. I'm
not too sure if there is some policy about #includes that are
"guaranteed" (ie., in the current tree) to be available through
some other #included header, so I just added linux/kernel.h to
each changed file that didn't #include it previously.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SOFTIRQ]: Remove do_softirq() symbol export.

As noted by Christoph Hellwig, pktgen was the only user so
it can now be removed.

[ Add missing cases caught by Adrian Bunk. -DaveM ]

Signed-off-by: Robert Olsson <robert.olsson@its.uu.se>
Signed-off-by: David S. Miller <davem@davemloft.net>

[PKTGEN]: Remove softirq scheduling.

It's not a job for pktgen.

Signed-off-by: Robert Olsson <robert.olsson@its.uu.se>
Signed-off-by: David S. Miller <davem@davemloft.net>

[PKTGEN]: Multiqueue support.

Below some pktgen support to send into different TX queues.
This can of course be feed into input queues on other machines

Signed-off-by: Robert Olsson <robert.olsson@its.uu.se>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETLINK]: use container_of instead

This could make future redesign of struct netlink_sock easier.

Signed-off-by: Denis Cheng <crquan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] policy: Replace magic number with XFRM_POLICY_OUT

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] IPSEC: Omit redirect for tunnelled packet.

IPv4 IPsec tunnel gateway incorrectly sends redirect to
sender if it is onlink host when network device the IPsec tunnelled
packet is arrived is the same as the one the decapsulated packet
is sent.

With this patch, it omits to send the redirect when the forwarding
skbuff carries secpath, since such skbuff should be assumed as
a decapsulated packet from IPsec tunnel by own.

Request for comments:
Alternatively we'd have another way to change net/ipv4/route.c
(__mkroute_input) to use RTCF_DOREDIRECT flag unless skbuff
has no secpath. It is better than this patch at performance
point of view because IPv4 redirect judgement is done at
routing slow-path. However, it should be taken care of resource
changes between SAD(XFRM states) and routing table. In other words,
When IPv4 SAD is changed does the related routing entry go to its
slow-path? If not, it is reasonable to apply this patch.

Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6] IPSEC: Omit redirect for tunnelled packet.

IPv6 IPsec tunnel gateway incorrectly sends redirect to
router or sender when network device the IPsec tunnelled packet
is arrived is the same as the one the decapsulated packet
is sent.

With this patch, it omits to send the redirect when the forwarding
skbuff carries secpath, since such skbuff should be assumed as
a decapsulated packet from IPsec tunnel by own.

It may be a rare case for an IPsec security gateway, however
it is not rare when the gateway is MIPv6 Home Agent since
the another tunnel end-point is Mobile Node and it changes
the attached network.

Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6] XFRM: Fix connected socket to use transformation.

When XFRM policy and state are ready after TCP connection is started,
the traffic should be transformed immediately, however it does not
on IPv6 TCP.

It depends on a dst cache replacement policy with connected socket.
It seems that the replacement is always done for IPv4, however, on
IPv6 case it is done only when routing cookie is changed.

This patch fix that non-transformation dst can be changed to
transformation one.
This behavior is required by MIPv6 and improves IPv6 IPsec.

Fixes by Masahide NAKAMURA.

Signed-off-by: Noriaki TAKAMIYA <takamiya@po.ntts.co.jp>
Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6]: Add v4mapped address inline

Add v4mapped address inline to avoid calls to ipv6_addr_type().

Signed-off-by: Brian Haley <brian.haley@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SHAPER]: Mark for removal.

This driver has been marked obsolete for a long time and
is superseded by traffic schedulers.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[UDP]: Randomize port selection.

This patch causes UDP port allocation to be randomized like TCP.
The earlier code would always choose same port (ie first empty list).

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] Cleanup: DIV_ROUND_UP

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP] MIB: Add counters for discarded SACK blocks

In DSACK case, some events are not extraordinary, such as packet
duplication generated DSACK. They can arrive easily below
snd_una when undo_marker is not set (TCP being in CA_Open),
counting such DSACKs amoung SACK discards will likely just
mislead if they occur in some scenario when there are other
problems as well. Similarly, excessively delayed packets could
cause "normal" DSACKs. Therefore, separate counters are
allocated for DSACK events.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Discard fuzzy SACK blocks

SACK processing code has been a sort of russian roulette as no
validation of SACK blocks is previously attempted. Besides, it
is not very clear what all kinds of broken SACK blocks really
mean (e.g., one that has start and end sequence numbers
reversed). So now close the roulette once and for all.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Rename tcp_ack_packets_out -> tcp_rearm_rto

Only thing that tiny function does is rearming the RTO (if
necessary), name it accordingly.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: tcp_packets_out_inc to tcp_output.c (no callers elsewhere)

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Remove unnecessary wrapper tcp_packets_out_dec

Makes caller side more obvious, there's no need to have
a wrapper for this oneliner!

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM]: Replace DPRINTK() with pr_debug().

Get rid of using DPRINTK macro in ATM and use pr_debug (in kernel.h).
Using the standard macro is cleaner and forces code to check for bad arguments
and formatting.

Fixes from Thomas Graf.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] ethernet: optimize memcpy and memset

The ethernet header management only needs to handle a fixed
size address (6 bytes). If the memcpy/memset are changed to
be passed a constant length, then compiler can optimize for
this case (and if it is smart eliminate string instructions).

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Inline attach_encap_tmpl(), attach_sec_ctx(), and attach_one_addr()

These functions are only used once and are a lot easier to understand if
inlined directly into the function.

Fixes by Masahide NAKAMURA.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Remove dependency on rtnetlink

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Use nlattr instead of rtattr

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Rename attribute array from xfrma[] to attrs[]

Increases readability a lot.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Enhance indexing of the attribute array

nlmsg_parse() puts attributes at array[type] so the indexing
method can be simpilfied by removing the obscuring "- 1".

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Establish an attribute policy

Adds a policy defining the minimal payload lengths for all the attributes
allowing for most attribute validation checks to be removed from in
the middle of the code path. Makes updates more consistent as many format
errors are recognised earlier, before any changes have been attempted.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Use nlmsg_parse() to parse attributes

Uses nlmsg_parse() to parse the attributes. This actually changes
behaviour as unknown attributes (type > MAXTYPE) no longer cause
an error. Instead unknown attributes will be ignored henceforth
to keep older kernels compatible with more recent userspace tools.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Use nlmsg_new() and type-safe size calculation helpers

Moves all complex message size calculation into own inlined helper
functions and makes use of the type-safe netlink interface.

Using nlmsg_new() simplifies the calculation itself as it takes care
of the netlink header length by itself.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Clear up some of the CONFIG_XFRM_SUB_POLICY ifdef mess

Moves all of the SUB_POLICY ifdefs related to the attribute size
calculation into a function.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Move algorithm length calculation to its own function

Adds alg_len() to calculate the properly padded length of an
algorithm attribute to simplify the code.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Use nla_put()/NLA_PUT() variantes

Also makes use of copy_sec_ctx() in another place and removes
duplicated code.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Use nlmsg_broadcast() and nlmsg_unicast()

This simplifies successful return codes from >0 to 0.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Use nlmsg_data() instead of NLMSG_DATA()

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Use nlmsg_end() and nlmsg_cancel()

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] netlink: Use nlmsg_put() instead of NLMSG_PUT()

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP]: Nuke the timeval helpers now that we fully converted to ktime_t

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP]: Nuke dccp_timestamp and dccps_epoch, not used anymore

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP] options: convert dccp_insert_option_timestamp to ktime_t

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP]: Convert dccps_timestamp_time to ktime_t

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP] packet_history: Convert dccphtx_tstamp to ktime_t

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP] packet_history: convert dccphrx_tstamp to ktime_t

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP] ackvec: Convert to ktime_t

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP] CCID3: Stop using dccp_timestamp

Now to convert the ackvec code to ktime_t so that we can get rid of
dccp_timestamp and the epoch thing in dccp_sock.

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP]: Convert dccp_sample_rtt to ktime_t

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[KTIME]: Introduce ktime_sub_ns and ktime_sub_us

First user will be the DCCP transport networking protocol.

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP]: Convert ccid3hcrx_tstamp_last_feedback to ktime_t

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP]: Convert ccid3hcrx_tstamp_last_ack to ktime_t

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP]: Convert ccid3hctx_t_ld to ktime_t

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP]: Make ccid3_hc_tx_update_x get a timestamp if needed

The code was too complicated, if p > 0 in ccid3_hc_tx_no_feedback_timer the
timestamp was being obtained to be passed to ccid3_hc_tx_update_x, where only
if p > 0 the timestamp was needed, so just leave it to ccid3_hc_tx_update_x to
obtain the timestamp if needed.

This will help in the upcoming changesets where we'll convert t_ld to ktime_t.
We'll eventually try to reuse ktime_get_real() calls again.

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Rewrite of sctp buffer management code

This patch introduces autotuning to the sctp buffer management code
similar to the TCP.  The buffer space can be grown if the advertised
receive window still has room.  This might happen if small message
sizes are used, which is common in telecom environmens.
New tunables are introduced that provide limits to buffer growth
and memory pressure is entered if to much buffer spaces is used.

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ETHTOOL]: Internal cleanup of ethtool_value-related handlers

Several get/set functions can be handled by a passing the ethtool_op
function pointer directly to a generic function.  This permits deletion
of a fair bit of redundant code.

Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ETHTOOL]: Introduce ->{get,set}_priv_flags, ETHTOOL_[GS]PFLAGS

Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ETHTOOL]: Introduce get_sset_count. Obsolete get_stats_count, self_test_count

Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ETHTOOL]: Add ETHTOOL_[GS]FLAGS sub-ioctls

Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] netconsole: Support dynamic reconfiguration using configfs

Based upon initial work by Keiichi Kii <k-keiichi@bx.jp.nec.com>.

This patch introduces support for dynamic reconfiguration (adding, removing
and/or modifying parameters of netconsole targets at runtime) using a
userspace interface exported via configfs.  Documentation is also updated
accordingly.

Issues and brief design overview:

(1) Kernel-initiated creation / destruction of kernel objects is not
    possible with configfs -- the lifetimes of the "config items" is managed
    exclusively from userspace.  But netconsole must support boot/module
    params too, and these are parsed in kernel and hence netpolls must be
    setup from the kernel.  Joel Becker suggested to separately manage the
    lifetimes of the two kinds of netconsole_target objects -- those created
    via configfs mkdir(2) from userspace and those specified from the
    boot/module option string.  This adds complexity and some redundancy here
    and also means that boot/module param-created targets are not exposed
    through the configfs namespace (and hence cannot be updated / destroyed
    dynamically).  However, this saves us from locking / refcounting
    complexities that would need to be introduced in configfs to support
    kernel-initiated item creation / destroy there.

(2) In configfs, item creation takes place in the call chain of the
    mkdir(2) syscall in the driver subsystem.  If we used an ioctl(2) to
    create / destroy objects from userspace, the special userspace program is
    able to fill out the structure to be passed into the ioctl and hence
    specify attributes such as local interface that are required at the time
    we set up the netpoll.  For configfs, this information is not available at
    the time of mkdir(2).  So, we keep all newly-created targets (via
    configfs) disabled by default.  The user is expected to set various
    attributes appropriately (including the local network interface if
    required) and then write(2) "1" to the "enabled" attribute.  Thus,
    netpoll_setup() is then called on the set parameters in the context of
    _this_ write(2) on the "enabled" attribute itself.  This design enables
    the user to reconfigure existing netconsole targets at runtime to be
    attached to newly-come-up interfaces that may not have existed when
    netconsole was loaded or when the targets were actually created.  All this
    effectively enables us to get rid of custom ioctls.

(3) Ultra-paranoid configfs attribute show() and store() operations, with
    sanity and input range checking, using only safe string primitives, and
    compliant with the recommendations in Documentation/filesystems/sysfs.txt.

(4) A new function netpoll_print_options() is created in the netpoll API,
    that just prints out the configured parameters for a netpoll structure.
    netpoll_parse_options() is modified to use that and it is also exported to
    be used from netconsole.

Signed-off-by: Satyam Sharma <satyam@infradead.org>
Acked-by: Keiichi Kii <k-keiichi@bx.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] netconsole: Support multiple logging targets

Based upon initial work by Keiichi Kii <k-keiichi@bx.jp.nec.com>.

This patch introduces support for multiple targets, independent of
CONFIG_NETCONSOLE_DYNAMIC -- this is useful even in the default case and
(including the infrastructure introduced in previous patches) doesn't really
add too many bytes to module text.  All the complexity (and size) comes with
the dynamic reconfigurability / userspace interface patch, and so it's
plausible users may want to keep this enabled but that disabled (say to avoid
a dependency on CONFIG_CONFIGFS_FS too).

Also update documentation to mention the use of ";" separator to specify
multiple logging targets in the boot/module option string.

Brief overview:

We maintain a target_list (and corresponding lock).  Get rid of the static
"default_target" and introduce allocation and release functions for our
netconsole_target objects (but keeping sure to preserve previous behaviour
such as default values).  During init_netconsole(), ";" is used as the
separator to identify multiple target specifications in the boot/module option
string.  The target specifications are parsed and netpolls setup.  During
exit, the target_list is torn down and all items released.

Signed-off-by: Satyam Sharma <satyam@infradead.org>
Signed-off-by: Keiichi Kii <k-keiichi@bx.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] netconsole: Introduce netconsole_netdev_notifier

Based upon initial work by Keiichi Kii <k-keiichi@bx.jp.nec.com>.

To update fields of underlying netpoll structure at runtime on corresponding
NETDEV_CHANGEADDR or NETDEV_CHANGENAME notifications.

ioctl(SIOCSIFHWADDR or SIOCSIFNAME) could be used to change the hardware/MAC
address or name of the local interface that our netpoll is attached to.
Whenever this happens, netdev notifier chain is called out with the
NETDEV_CHANGEADDR or NETDEV_CHANGENAME event message.  We respond to that and
update the local_mac or dev_name field of the struct netpoll.  This makes
sense anyway, but is especially required for dynamic netconsole because the
netpoll structure's internal members become user visible files when either
sysfs or configfs are used.  So this helps us to keep up with the MAC
address/name changes and keep values in struct netpoll uptodate.

[ Note that ioctl(SIOCSIFADDR) to change IP address of interface at
  runtime is not handled (to update local_ip of netpoll) on purpose --
  some setups may set the local_ip to a private address, not necessary
  the actual IP address of the sender host, as presently allowed. ]

Signed-off-by: Satyam Sharma <satyam@infradead.org>
Signed-off-by: Keiichi Kii <k-keiichi@bx.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] netconsole: Introduce netconsole_target

Based upon initial work by Keiichi Kii <k-keiichi@bx.jp.nec.com>.

Introduce a wrapper structure over netpoll to represent logging targets
configured in netconsole.  This will get extended with other members in
further patches.

This is done independent of the (to-be-introduced) NETCONSOLE_DYNAMIC config
option so that we're able to drastically cut down on the #ifdef complexity of
final netconsole.c.  Also, struct netconsole_target would be required for
multiple targets support also, and not just dynamic reconfigurability.

Signed-off-by: Satyam Sharma <satyam@infradead.org>
Signed-off-by: Keiichi Kii <k-keiichi@bx.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] netconsole: Add some useful tips to documentation

Based upon initial work by Keiichi Kii <k-keiichi@bx.jp.nec.com>.

Add some useful general-purpose tips.  Also suggest solution for the frequent
problem of console loglevel set too low numerically (i.e.  for high priority
messages only) on the sender.

Signed-off-by: Satyam Sharma <satyam@infradead.org>
Acked-by: Keiichi Kii <k-keiichi@bx.jp.nec.com>
Acked-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] netconsole: Use netif_running() in write_msg()

Based upon initial work by Keiichi Kii <k-keiichi@bx.jp.nec.com>.

Avoid unnecessarily disabling interrupts and calling netpoll_send_udp() if the
corresponding local interface is not up.

Signed-off-by: Satyam Sharma <satyam@infradead.org>
Acked-by: Keiichi Kii <k-keiichi@bx.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] netconsole: Simplify boot/module option setup logic

Based upon initial work by Keiichi Kii <k-keiichi@bx.jp.nec.com>.

Presently, boot/module parameters are set up quite differently for the case of
built-in netconsole (__setup() -> obsolete_checksetup() ->
netpoll_parse_options() -> strlen(config) == 0 in init_netconsole()) vs
modular netconsole (module_param_string() -> string copied to the config
variable -> strlen(config) != 0 init_netconsole() -> netpoll_parse_options()).

This patch makes both of them similar by doing exactly the equivalent of a
module_param_string() in option_setup() also -- just copying the param string
passed from the kernel command line into "config" variable.  So,
strlen(config) != 0 in both cases, and netpoll_parse_options() is always
called from init_netconsole(), thus making the setup logic for both cases
similar.

Now, option_setup() is only ever called / used for the built-in case, so we
put it inside a #ifndef MODULE, otherwise gcc will complain about
option_setup() being "defined but not used".  Also, the "configured" variable
is redundant with this patch and hence removed.

Signed-off-by: Satyam Sharma <satyam@infradead.org>
Signed-off-by: Keiichi Kii <k-keiichi@bx.jp.nec.com>
Acked-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] netconsole: Remove bogus check

Based upon initial work by Keiichi Kii <k-keiichi@bx.jp.nec.com>.

The (!np.dev) check in write_msg() is bogus (always false), because: np.dev is
set by netpoll_setup(), which is called by init_netconsole() before
register_console(), so write_msg() cannot be triggered unless netpoll_setup()
successfully set np.dev.  Also np.dev cannot go away from under us, because
netpoll_setup() grabs us reference on it.  So let's remove the bogus check.

Signed-off-by: Satyam Sharma <satyam@infradead.org>
Acked-by: Keiichi Kii <k-keiichi@bx.jp.nec.com>
Acked-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] netconsole: Cleanups, codingstyle, prettyfication

Based upon initial work by Keiichi Kii <k-keiichi@bx.jp.nec.com>.

(1) Remove unwanted headers.
(2) Mark __init and __exit as appropriate.
(3) Various trivial codingstyle and prettification stuff.

Signed-off-by: Satyam Sharma <satyam@infradead.org>
Signed-off-by: Keiichi Kii <k-keiichi@bx.jp.nec.com>
Acked-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] fib_trie: macro cleanup

This patch converts the messy macro for MASK_PFX to inline function
and expands TKEY_GET_MASK in the one place it is used.

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] fib_trie: cleanup

Try this out:
     * replace macro's with inlines
     * get rid of places doing multiple evaluations of NODE_PARENT

[akpm@linux-foundation.org: rcu_dereference wants an lval]

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Update comment about highest_sack validity

This stale info came from the original idea, which proved to be
unnecessarily complex, sacked_out > 0 is easy to do and that when
it's going to be needed anyway (it _can_ be valid also when
sacked_out == 0 but there's not going to be a guarantee about it
for now).

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Move sack_ok access to obviously named funcs & cleanup

Previously code had IsReno/IsFack defined as macros that were
local to tcp_input.c though sack_ok field has user elsewhere too
for the same purpose. This changes them to static inlines as
preferred according the current coding style and unifies the
access to sack_ok across multiple files. Magic bitops of sack_ok
for FACK and DSACK are also abstracted to functions with
appropriate names.

Note:
- One sack_ok = 1 remains but that's self explanary, i.e., it
  enables sack
- Couple of !IsReno cases are changed to tcp_is_sack
- There were no users for IsDSack => I dropped it

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Don't panic if S+L skb is detected

BUG_ON is an overkill. In fact, I was mislead by BUG_TRAP
severity (equals to WARN_ON) which is much lower than BUG_ON's
(that panics).

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Reduce sacked_out with reno when purging write_queue

Previously TCP had a transitional state during which reno
counted segments that are already below the current window into
sacked_out, which is now prevented. In addition, re-try now
the unconditional S+L skb catching.

This approach conservatively calls just remove_sack and leaves
reset_sack() calls alone. The best solution to the whole problem
would be to first calculate the new sacked_out fully (this patch
does not move reno_sack_reset calls from original sites and thus
does not implement this). However, that would require very
invasive change to fastretrans_alert (perhaps even slicing it to
two halves). Alternatively, all callers of tcp_packets_in_flight
(i.e., users that depend on sacked_out) should be postponed
until the new sacked_out has been calculated but it isn't any
simpler alternative.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Keep state in Disorder also if only lost_out > 0

This happens rather infrequently and is only possible during
FRTO. We must not allow TCP to slip to Open state because
tcp_fastretrans_alert might then not be called on it's time
when FRTO has exited. This become a problem when left_out
got removed and was replaced by just sacked_out.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Restore over-zealous tcp_sync_left_out-like removals

tcp_verify_left_out is useful for verifying S+L condition, so
add it back to couple of places in where the code was not
calling to tcp_sync_left_out but used own ad-hoc solution
(before the tcp_sync_left_out got removed).

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Left out sync->verify (the new meaning of it) & definify

Left_out was dropped a while ago, thus leaving verifying
consistency of the "left out" as only task for the function in
question. Thus make it's name more appropriate.

In addition, it is intentionally converted to #define instead
of static inline because the location of the invariant failure
is the most important thing to have if this ever triggers. I
think it would have been helpful e.g. in this case where the
location of the failure point had to be based on some quesswork:
    http://lkml.org/lkml/2007/5/2/464
...Luckily the guesswork seems to have proved to be correct.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Add tcp_left_out(tp) "back" to get cleaner looking lines

tp->left_out got removed but nothing came to replace it back
then (users just did addition by themselves), so add function
for users now.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Tighten tcp_sock's belt, drop left_out

It is easily calculable when needed and user are not that many
after all.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Remove num_acked>0 checks from cong.ctrl mods pkts_acked

There is no need for such check in pkts_acked because the
callback is not invoked unless at least one segment got fully
ACKed (i.e., the snd_una moved past skb's end_seq) by the
cumulative ACK's snd_una advancement.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Add tcp_dec_pcount_approx int variant

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Move code from tcp_ecn.h to tcp*.c and tcp.h & remove it

No other users exist for tcp_ecn.h. Very few things remain in
tcp.h, for most TCP ECN functions callers reside within a
single .c file and can be placed there.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Access to highest_sack obsoletes forward_cnt_hint

In addition, added a reference about the purpose of the loop.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP] FRTO: remove unnecessary fackets/sacked_out recounting

F-RTO does not touch SACKED_ACKED bits at all, so there is no
need to recount them in tcp_enter_frto_loss. After removal of
the else branch, nested ifs can be combined.

This must also reset sacked_out when SACK is not in use as TCP
could have received some duplicate ACKs prior RTO. To achieve
that in a sane manner, tcp_reset_reno_sack was re-placed by the
previous patch.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Move Reno SACKed_out counter functions earlier

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Extract DSACK detection code from tcp_sacktag_write_queue().

Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Rexmit hint must be cleared instead of setting it

Stupid error from my side. Even though now that I noticed this,
I hoped it would have been an optimization but no, the counter
hint is then incorrect. Thus clearing is necessary for now (I
still suspect though that this path is never executed).

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Extracted rexmit hint clearing from the LOST marking code

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Add highest_sack seqno, points to globally highest SACK

It is guaranteed to be valid only when !tp->sacked_out. In most
cases this seqno is available in the last ACK but there is no
guarantee for that. The new fast recovery loss marking algorithm
needs this as entry point.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>