RACK skips an ACK unless it advances the most recently delivered
TX timestamp (rack.mstamp). Since RACK also uses the most recent
RTT to decide if a packet is lost, RACK should still run the
loss detection whenever the most recent RTT changes. For example,
an ACK that does not advance the timestamp but triggers the cwnd
undo due to reordering, would then use the most recent (higher)
RTT measurement to detect further losses.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Priyaranjan Jha <priyarjha@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When sender detects spurious retransmission, all packets
marked lost are remarked to be in-flight. However some may
be considered lost based on its timestamps in RACK. This patch
forces RACK to re-evaluate, which may be skipped previously if
the ACK does not advance RACK timestamp.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Priyaranjan Jha <priyarjha@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RACK does not test the loss recovery state correctly to compute
the reordering window. It assumes if lost_out is zero then TCP is
not in loss recovery. But it can be zero during recovery before
calling tcp_rack_detect_loss(): when an ACK acknowledges all
packets marked lost before receiving this ACK, but has not yet
to discover new ones by tcp_rack_detect_loss(). The fix is to
simply test the congestion state directly.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Priyaranjan Jha <priyarjha@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After applying 2270bc5da3 ("bnxt_en: Fix netpoll handling") and
903649e718 ("bnxt_en: Improve -ENOMEM logic in NAPI poll loop."),
we still see the following WARN fire:
------------[ cut here ]------------
WARNING: CPU: 0 PID: 1875170 at net/core/netpoll.c:165 netpoll_poll_dev+0x15a/0x160
bnxt_poll+0x0/0xd0 exceeded budget in poll
<snip>
Call Trace:
[<ffffffff814be5cd>] dump_stack+0x4d/0x70
[<ffffffff8107e013>] __warn+0xd3/0xf0
[<ffffffff8107e07f>] warn_slowpath_fmt+0x4f/0x60
[<ffffffff8179519a>] netpoll_poll_dev+0x15a/0x160
[<ffffffff81795f38>] netpoll_send_skb_on_dev+0x168/0x250
[<ffffffff817962fc>] netpoll_send_udp+0x2dc/0x440
[<ffffffff815fa9be>] write_ext_msg+0x20e/0x250
[<ffffffff810c8125>] call_console_drivers.constprop.23+0xa5/0x110
[<ffffffff810c9549>] console_unlock+0x339/0x5b0
[<ffffffff810c9a88>] vprintk_emit+0x2c8/0x450
[<ffffffff810c9d5f>] vprintk_default+0x1f/0x30
[<ffffffff81173df5>] printk+0x48/0x50
[<ffffffffa0197713>] edac_raw_mc_handle_error+0x563/0x5c0 [edac_core]
[<ffffffffa0197b9b>] edac_mc_handle_error+0x42b/0x6e0 [edac_core]
[<ffffffffa01c3a60>] sbridge_mce_output_error+0x410/0x10d0 [sb_edac]
[<ffffffffa01c47cc>] sbridge_check_error+0xac/0x130 [sb_edac]
[<ffffffffa0197f3c>] edac_mc_workq_function+0x3c/0x90 [edac_core]
[<ffffffff81095f8b>] process_one_work+0x19b/0x480
[<ffffffff810967ca>] worker_thread+0x6a/0x520
[<ffffffff8109c7c4>] kthread+0xe4/0x100
[<ffffffff81884c52>] ret_from_fork+0x22/0x40
This happens because we increment rx_pkts on -ENOMEM and -EIO, resulting
in rx_pkts > 0. Fix this by only bumping rx_pkts if we were actually
given a non-zero budget.
Signed-off-by: Calvin Owens <calvinowens@fb.com>
Acked-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Neal Cardwell says:
====================
TCP BBR sampling fixes for loss recovery undo
This patch series has a few minor bug fixes for cases where spurious
loss recoveries can trick BBR estimators into estimating that the
available bandwidth is much lower than the true available bandwidth.
In both cases the fix here is to just reset the estimator upon loss
recovery undo.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix BBR so that upon notification of a loss recovery undo BBR resets
long-term bandwidth sampling.
Under high reordering, reordering events can be interpreted as loss.
If the reordering and spurious loss estimates are high enough, this
can cause BBR to spuriously estimate that we are seeing loss rates
high enough to trigger long-term bandwidth estimation. To avoid that
problem, this commit resets long-term bandwidth sampling on loss
recovery undo events.
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix BBR so that upon notification of a loss recovery undo BBR resets
the full pipe detection (STARTUP exit) state machine.
Under high reordering, reordering events can be interpreted as loss.
If the reordering and spurious loss estimates are high enough, this
could previously cause BBR to spuriously estimate that the pipe is
full.
Since spurious loss recovery means that our overall sending will have
slowed down spuriously, this commit gives a flow more time to probe
robustly for bandwidth and decide the pipe is really full.
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit records the "full bw reached" decision in a new
full_bw_reached bit. This is a pure refactor that does not change the
current behavior, but enables subsequent fixes and improvements.
In particular, this enables simple and clean fixes because the full_bw
and full_bw_cnt can be unconditionally zeroed without worrying about
forgetting that we estimated we filled the pipe in Startup. And it
enables future improvements because multiple code paths can be used
for estimating that we filled the pipe in Startup; any new code paths
only need to set this bit when they think the pipe is full.
Note that this fix intentionally reduces the width of the full_bw_cnt
counter, since we have never used the most significant bit.
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The bytes_compl and pkts_compl pointers passed to efx_dequeue_buffers
cannot be NULL. Add a paranoid warning to check this condition and fix
the one case where they were NULL.
efx_enqueue_unwind() is called very rarely, during error handling.
Without this fix it would fail with a NULL pointer dereference in
efx_dequeue_buffer, with efx_enqueue_skb in the call stack.
Fixes: e9117e5099 ("sfc: Firmware-Assisted TSO version 2")
Reported-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Tested-by: Jarod Wilson <jarod@redhat.com>
Acked-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This controller does not support EEE, but it may connect to a PHY
which supports EEE and advertises EEE by default, while its link
partner also advertises EEE. If this happens, the PHY enters low
power mode when the traffic rate is low and causes packet loss.
This patch disables EEE advertisement by default for any PHY that
gianfar connects to, to prevent the above unwanted outcome.
Signed-off-by: Shaohui Xie <Shaohui.Xie@nxp.com>
Tested-by: Yangbo Lu <Yangbo.lu@nxp.com>
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Mark tcp_sock during a SACK reneging event and invalidate rate samples
while marked. Such rate samples may overestimate bw by including packets
that were SACKed before reneging.
< ack 6001 win 10000 sack 7001:38001
< ack 7001 win 0 sack 8001:38001 // Reneg detected
> seq 7001:8001 // RTO, SACK cleared.
< ack 38001 win 10000
In above example the rate sample taken after the last ack will count
7001-38001 as delivered while the actual delivery rate likely could
be much lower i.e. 7001-8001.
This patch adds a new field tcp_sock.sack_reneg and marks it when we
declare SACK reneging and entering TCP_CA_Loss, and unmarks it after
the last rate sample was taken before moving back to TCP_CA_Open. This
patch also invalidates rate samples taken while tcp_sock.is_sack_reneg
is set.
Fixes: b9f64820fb ("tcp: track data delivery rate for a TCP connection")
Signed-off-by: Yousuk Seung <ysseung@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Daniel Borkmann says:
====================
pull-request: bpf 2017-12-06
The following pull-request contains BPF updates for your *net* tree.
The main changes are:
1) Fixing broken uapi for BPF tracing programs for s390 and arm64
architectures due to pt_regs being in-kernel only, and not part
of uapi right now. A wrapper is added that exports pt_regs in
an asm-generic way. For arm64 this maps to existing user_pt_regs
structure and for s390 a user_pt_regs structure exporting the
beginning of pt_regs is added and uapi-exported, thus fixing the
BPF issues seen in perf (and BPF selftests), all from Hendrik.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The qmi_wwan minidriver support a 'raw-ip' mode where frames are
received without any ethernet header. This causes alignment issues
because the skbs allocated by usbnet are "IP aligned".
Fix by allowing minidrivers to disable the additional alignment
offset. This is implemented using a per-device flag, since the same
minidriver also supports 'ethernet' mode.
Fixes: 32f7adf633 ("net: qmi_wwan: support "raw IP" mode")
Reported-and-tested-by: Jay Foster <jay@systech.com>
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
When I switched rcv_rtt_est to high resolution timestamps, I forgot
that tp->tcp_mstamp needed to be refreshed in tcp_rcv_space_adjust()
Using an old timestamp leads to autotuning lags.
Fixes: 645f4c6f2e ("tcp: switch rcv_rtt_est and rcvq_space to high resolution timestamps")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Wei Wang <weiwan@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 28033ae4e0 ("net: netlink: Update attr validation to require
exact length for some types") requires attributes using types NLA_U* and
NLA_S* to have an exact length. This change is exposing bugs in various
userspace commands that are sending attributes with an invalid length
(e.g., attribute has type NLA_U8 and userspace sends NLA_U32). While
the commands are clearly broken and need to be fixed, users are arguing
that the sudden change in enforcement is breaking older commands on
newer kernels for use cases that otherwise "worked".
Relax the validation to print a warning mesage similar to what is done
for messages containing extra bytes after parsing.
Fixes: 28033ae4e0 ("net: netlink: Update attr validation to require exact length for some types")
Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
commit 8d79266bc4 ("ip6_tunnel: add collect_md mode to IPv6 tunnels")
introduced new exit point in ipxip6_rcv. however rcu_read_unlock is
missing there. this diff is fixing this
v1->v2:
instead of doing rcu_read_unlock in place, we are going to "drop"
section (to prevent skb leakage)
Fixes: 8d79266bc4 ("ip6_tunnel: add collect_md mode to IPv6 tunnels")
Signed-off-by: Nikita V. Shirokov <tehnerd@fb.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Andrew Lunn says:
====================
mv88e6xxx error patch fixes
While trying to bring up a new PHY on a board, i exercised the error
paths a bit, and discovered some bugs. The unwind for interrupt
handling deadlocks, and the MDIO code hits a BUG() when a registered
MDIO device is freed without first being unregistered.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The MDIO busses need to be unregistered before they are freed,
otherwise BUG() is called. Add a call to the unregister code if the
registration fails, since we can have multiple busses, of which some
may correctly register before one fails. This requires moving the code
around a little.
Fixes: a3c53be55c ("net: dsa: mv88e6xxx: Support multiple MDIO busses")
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When removing the interrupt handling code, we should mask the
generation of interrupts. The code however unmasked all
interrupts. This can then cause a new interrupt. We then get into a
deadlock where the interrupt thread is waiting to run, and the code
continues, trying to remove the interrupt handler, which means waiting
for the thread to complete. On a UP machine this deadlocks.
Fix so we really mask interrupts in the hardware. The same error is
made in the error path when install the interrupt handling code.
Fixes: 3460a5770c ("net: dsa: mv88e6xxx: Mask g1 interrupts and free interrupt")
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If clk_set_rate() fails, we should disable clk before return.
Found by Linux Driver Verification project (linuxtesting.org).
Signed-off-by: Branislav Radocaj <branislav@radocaj.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
add appropriate calls to clk_disable_unprepare() by jumping to out_mdio
in case orion_mdio_probe() returns -EPROBE_DEFER.
Found by Linux Driver Verification project (linuxtesting.org).
Fixes: 3d604da1e9 ("net: mvmdio: get and enable optional clock")
Signed-off-by: Tobias Jordan <Tobias.Jordan@elektrobit.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
These duplicate includes have been found with scripts/checkincludes.pl but
they have been removed manually to avoid removing false positives.
Signed-off-by: Pravin Shedge <pravin.shedge4linux@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Macvlan devices are similar to vlans and do not update their
own trans_start. In order for arp monitoring to work for a bond device
when the slaves are macvlans, obtain its real device.
Signed-off-by: Chris Dion <christopher.dion@dell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Yuchung Cheng
Calvin Owens
David S. Miller
Neal Cardwell
Bert Kenward
Claudiu Manoil
Yousuk Seung
Bjørn Mork
Eric Dumazet
David Ahern
Nikita V. Shirokov
Andrew Lunn
Branislav Radocaj
Tobias Jordan
Pravin Shedge
Håkon Bugge
Chris Dion