Eric Dumazet [Tue, 2 Oct 2018 22:47:35 +0000 (15:47 -0700)]
rtnl: limit IFLA_NUM_TX_QUEUES and IFLA_NUM_RX_QUEUES to 4096
We have an impressive number of syzkaller bugs that are linked
to the fact that syzbot was able to create a networking device
with millions of TX (or RX) queues.
Let's limit the number of RX/TX queues to 4096, this really should
cover all known cases.
A separate patch will add various cond_resched() in the loops
handling sysfs entries at device creation and dismantle.
Tested:
lpaa6:~# ip link add gre-4097 numtxqueues 4097 numrxqueues 4097 type ip6gretap
RTNETLINK answers: Invalid argument
lpaa6:~# time ip link add gre-4096 numtxqueues 4096 numrxqueues 4096 type ip6gretap
real 0m0.180s
user 0m0.000s
sys 0m0.107s
Fixes: 76ff5cc91935 ("rtnl: allow to specify number of rx and tx queues on device creation") Signed-off-by: Eric Dumazet <[email protected]> Reported-by: syzbot <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Mahesh Bandewar [Tue, 2 Oct 2018 19:14:34 +0000 (12:14 -0700)]
bonding: fix warning message
RX queue config for bonding master could be different from its slave
device(s). With the commit 6a9e461f6fe4 ("bonding: pass link-local
packets to bonding master also."), the packet is reinjected into stack
with skb->dev as bonding master. This potentially triggers the
message:
"bondX received packet on queue Y, but number of RX queues is Z"
whenever the queue that packet is received on is higher than the
numrxqueues on bonding master (Y > Z).
Fixes: 6a9e461f6fe4 ("bonding: pass link-local packets to bonding master also.") Reported-by: John Sperbeck <[email protected]> Signed-off-by: Eric Dumazet <[email protected]> Signed-off-by: Mahesh Bandewar <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Eric Dumazet [Tue, 2 Oct 2018 19:35:05 +0000 (12:35 -0700)]
inet: make sure to grab rcu_read_lock before using ireq->ireq_opt
Timer handlers do not imply rcu_read_lock(), so my recent fix
triggered a LOCKDEP warning when SYNACK is retransmit.
Lets add rcu_read_lock()/rcu_read_unlock() pairs around ireq->ireq_opt
usages instead of guessing what is done by callers, since it is
not worth the pain.
Get rid of ireq_opt_deref() helper since it hides the logic
without real benefit, since it is now a standard rcu_dereference().
Fixes: 1ad98e9d1bdf ("tcp/dccp: fix lockdep issue when SYN is backlogged") Signed-off-by: Eric Dumazet <[email protected]> Reported-by: Willem de Bruijn <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Jakub Kicinski [Tue, 2 Oct 2018 17:10:14 +0000 (10:10 -0700)]
nfp: avoid soft lockups under control message storm
When FW floods the driver with control messages try to exit the cmsg
processing loop every now and then to avoid soft lockups. Cmsg
processing is generally very lightweight so 512 seems like a reasonable
budget, which should not be exceeded under normal conditions.
declance: Fix continuation with the adapter identification message
Fix a commit 4bcc595ccd80 ("printk: reinstate KERN_CONT for printing
continuation lines") regression with the `declance' driver, which caused
the adapter identification message to be split between two lines, e.g.:
declance.c: v0.011 by Linux MIPS DECstation task force
tc6: PMAD-AA
, addr = 08:00:2b:1b:2a:6a, irq = 14
tc6: registered as eth0.
Address that properly, by printing identification with a single call,
making the messages now look like:
declance.c: v0.011 by Linux MIPS DECstation task force
tc6: PMAD-AA, addr = 08:00:2b:1b:2a:6a, irq = 14
tc6: registered as eth0.
Signed-off-by: Maciej W. Rozycki <[email protected]> Fixes: 4bcc595ccd80 ("printk: reinstate KERN_CONT for printing continuation lines") Signed-off-by: David S. Miller <[email protected]>
During certain heavy network loads TX could time out
with TX ring dump.
TX is sometimes never restarted after reaching
"tx_stop_threshold" because function "fec_enet_tx_queue"
only tests the first queue.
In addition the TX timeout callback function failed to
recover because it also operated only on the first queue.
r8169: fix network stalls due to missing bit TXCFG_AUTO_FIFO
Some of the chip-specific hw_start functions set bit TXCFG_AUTO_FIFO
in register TxConfig. The original patch changed the order of some
calls resulting in these changes being overwritten by
rtl_set_tx_config_registers() in rtl_hw_start(). This eventually
resulted in network stalls especially under high load.
Analyzing the chip-specific hw_start functions all chip version from
34, with the exception of version 39, need this bit set.
This patch moves setting this bit to rtl_set_tx_config_registers().
David S. Miller [Tue, 2 Oct 2018 06:27:28 +0000 (23:27 -0700)]
Merge branch 'tun-races'
Eric Dumazet says:
====================
tun: address two syzbot reports
Small changes addressing races discovered by syzbot.
First patch is a cleanup.
Second patch moves a mutex init sooner.
Third patch makes sure each tfile gets its own napi enable flags.
====================
Fixes: 90e33d459407 ("tun: enable napi_gro_frags() for TUN/TAP driver") Signed-off-by: Eric Dumazet <[email protected]> Reported-by: syzbot <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Dave Jones [Fri, 28 Sep 2018 20:26:08 +0000 (16:26 -0400)]
bond: take rcu lock in netpoll_send_skb_on_dev
The bonding driver lacks the rcu lock when it calls down into
netdev_lower_get_next_private_rcu from bond_poll_controller, which
results in a trace like:
David Ahern [Fri, 28 Sep 2018 19:28:41 +0000 (12:28 -0700)]
rtnetlink: Fail dump if target netnsid is invalid
Link dumps can return results from a target namespace. If the namespace id
is invalid, then the dump request should fail if get_target_net fails
rather than continuing with a dump of the current namespace.
Fixes: 79e1ad148c844 ("rtnetlink: use netnsid to query interface") Signed-off-by: David Ahern <[email protected]> Signed-off-by: David S. Miller <[email protected]>
When the commit was merged, the code used nf_ct_put() to free
the entry, but later on commit 76644232e612 ("openvswitch: Free
tmpl with tmpl_free.") replaced that with nf_ct_tmpl_free which
is a more appropriate. Now the original problem is removed.
Then 44d6e2f27328 ("net: Replace NF_CT_ASSERT() with WARN_ON().")
replaced a debug assert with a WARN_ON() which is trigged now.
The initial session number when a link is created is based on a random
value, taken from struct tipc_net->random. It is then incremented for
each link reset to avoid mixing protocol messages from different link
sessions.
However, when a bearer is reset all its links are deleted, and will
later be re-created using the same random value as the first time.
This means that if the link never went down between creation and
deletion we will still sometimes have two subsequent sessions with
the same session number. In virtual environments with potentially
long transmission times this has turned out to be a real problem.
We now fix this by randomizing the session number each time a link
is created.
With a session number size of 16 bits this gives a risk of session
collision of 1/64k. To reduce this further, we also introduce a sanity
check on the very first STATE message arriving at a link. If this has
an acknowledge value differing from 0, which is logically impossible,
we ignore the message. The final risk for session collision is hence
reduced to 1/4G, which should be sufficient.
Dan Carpenter [Sat, 22 Sep 2018 13:46:48 +0000 (16:46 +0300)]
net: sched: act_ipt: check for underflow in __tcf_ipt_init()
If "td->u.target_size" is larger than sizeof(struct xt_entry_target) we
return -EINVAL. But we don't check whether it's smaller than
sizeof(struct xt_entry_target) and that could lead to an out of bounds
read.
Fixes: 7ba699c604ab ("[NET_SCHED]: Convert actions from rtnetlink to new netlink API") Signed-off-by: Dan Carpenter <[email protected]> Signed-off-by: David S. Miller <[email protected]>
1) Validate address prefix lengths in the xfrm selector,
otherwise we may hit undefined behaviour in the
address matching functions if the prefix is too
big for the given address family.
2) Fix skb leak on local message size errors.
From Thadeu Lima de Souza Cascardo.
3) We currently reset the transport header back to the network
header after a transport mode transformation is applied. This
leads to an incorrect transport header when multiple transport
mode transformations are applied. Reset the transport header
only after all transformations are already applied to fix this.
From Sowmini Varadhan.
4) We only support one offloaded xfrm, so reset crypto_done after
the first transformation in xfrm_input(). Otherwise we may call
the wrong input method for subsequent transformations.
From Sowmini Varadhan.
5) Fix NULL pointer dereference when skb_dst_force clears the dst_entry.
skb_dst_force does not really force a dst refcount anymore, it might
clear it instead. xfrm code did not expect this, add a check to not
dereference skb_dst() if it was cleared by skb_dst_force.
6) Validate xfrm template mode, otherwise we can get a stack-out-of-bounds
read in xfrm_state_find. From Sean Tranchetti.
Please pull or let me know if there are problems.
====================
Eric Dumazet [Mon, 1 Oct 2018 22:02:26 +0000 (15:02 -0700)]
tcp/dccp: fix lockdep issue when SYN is backlogged
In normal SYN processing, packets are handled without listener
lock and in RCU protected ingress path.
But syzkaller is known to be able to trick us and SYN
packets might be processed in process context, after being
queued into socket backlog.
In commit 06f877d613be ("tcp/dccp: fix other lockdep splats
accessing ireq_opt") I made a very stupid fix, that happened
to work mostly because of the regular path being RCU protected.
Really the thing protecting ireq->ireq_opt is RCU read lock,
and the pseudo request refcnt is not relevant.
This patch extends what I did in commit 449809a66c1d ("tcp/dccp:
block BH for SYN processing") by adding an extra rcu_read_{lock|unlock}
pair in the paths that might be taken when processing SYN from
socket backlog (thus possibly in process context)
Fixes: 06f877d613be ("tcp/dccp: fix other lockdep splats accessing ireq_opt") Signed-off-by: Eric Dumazet <[email protected]> Reported-by: syzbot <[email protected]> Signed-off-by: David S. Miller <[email protected]>
qlcnic: fix Tx descriptor corruption on 82xx devices
In regular NIC transmission flow, driver always configures MAC using
Tx queue zero descriptor as a part of MAC learning flow.
But with multi Tx queue supported NIC, regular transmission can occur on
any non-zero Tx queue and from that context it uses
Tx queue zero descriptor to configure MAC, at the same time TX queue
zero could be used by another CPU for regular transmission
which could lead to Tx queue zero descriptor corruption and cause FW
abort.
This patch fixes this in such a way that driver always configures
learned MAC address from the same Tx queue which is used for
regular transmission.
Fixes: 7e2cf4feba05 ("qlcnic: change driver hardware interface mechanism") Signed-off-by: Shahed Shaikh <[email protected]> Signed-off-by: David S. Miller <[email protected]>
We see the following scenario:
1) Link endpoint B on node 1 discovers that its peer endpoint is gone.
Since there is a second working link, failover procedure is started.
2) Link endpoint A on node 1 sends a FAILOVER message to peer endpoint
A on node 2. The node item 1->2 goes to state FAILINGOVER.
3) Linke endpoint A/2 receives the failover, and is supposed to take
down its parallell link endpoint B/2, while producing a FAILOVER
message to send back to A/1.
4) However, B/2 has already been deleted, so no FAILOVER message can
created.
5) Node 1->2 remains in state FAILINGOVER forever, refusing to receive
any messages that can bring B/1 up again. We are left with a non-
redundant link between node 1 and 2.
We fix this with letting endpoint A/2 build a dummy FAILOVER message
to send to back to A/1, so that the situation can be resolved.
====================
net: usb: Check for Wake-on-LAN modes
Most of our USB Ethernet drivers don't seem to be checking properly
whether the user is supplying a correct Wake-on-LAN mode to enter, so
the experience as an user could be confusing, since it would generally
lead to either no wake-up, or the device not being marked for wake-up.
The driver does not check for Wake-on-LAN modes specified by an user,
but will conditionally set the device as wake-up enabled or not based on
that, which could be a very confusing user experience.
Fixes: e0e474a83c18 ("smsc95xx: add wol magic packet support") Signed-off-by: Florian Fainelli <[email protected]> Signed-off-by: David S. Miller <[email protected]>
The driver does not check for Wake-on-LAN modes specified by an user,
but will conditionally set the device as wake-up enabled or not based on
that, which could be a very confusing user experience.
Fixes: 6c636503260d ("smsc75xx: add wol magic packet support") Signed-off-by: Florian Fainelli <[email protected]> Signed-off-by: David S. Miller <[email protected]>
The driver does not check for Wake-on-LAN modes specified by an user,
but will conditionally set the device as wake-up enabled or not based on
that, which could be a very confusing user experience.
Fixes: 21ff2e8976b1 ("r8152: support WOL") Signed-off-by: Florian Fainelli <[email protected]> Signed-off-by: David S. Miller <[email protected]>
The driver currently silently accepts unsupported Wake-on-LAN modes
(other than WAKE_PHY or WAKE_MAGIC) without reporting that to the user,
which is confusing.
Fixes: 19a38d8e0aa3 ("USB2NET : SR9800 : One chip USB2.0 USB2NET SR9800 Device Driver Support") Signed-off-by: Florian Fainelli <[email protected]> Signed-off-by: David S. Miller <[email protected]>
The driver supports a fair amount of Wake-on-LAN modes, but is not
checking that the user specified one that is supported.
Fixes: 55d7de9de6c3 ("Microchip's LAN7800 family USB 2/3 to 10/100/1000 Ethernet device driver") Signed-off-by: Florian Fainelli <[email protected]> Reviewed-by: Woojung Huh <[email protected]> Signed-off-by: David S. Miller <[email protected]>
ax88179_178a: Check for supported Wake-on-LAN modes
The driver currently silently accepts unsupported Wake-on-LAN modes
(other than WAKE_PHY or WAKE_MAGIC) without reporting that to the user,
which is confusing.
Fixes: e2ca90c276e1 ("ax88179_178a: ASIX AX88179_178A USB 3.0/2.0 to gigabit ethernet adapter driver") Signed-off-by: Florian Fainelli <[email protected]> Signed-off-by: David S. Miller <[email protected]>
The driver currently silently accepts unsupported Wake-on-LAN modes
(other than WAKE_PHY or WAKE_MAGIC) without reporting that to the user,
which is confusing.
Fixes: 2e55cc7210fe ("[PATCH] USB: usbnet (3/9) module for ASIX Ethernet adapters") Signed-off-by: Florian Fainelli <[email protected]> Signed-off-by: David S. Miller <[email protected]>
David S. Miller [Sat, 29 Sep 2018 18:28:49 +0000 (11:28 -0700)]
Merge branch 'ieee802154-for-davem-2018-09-28' of git://git.kernel.org/pub/scm/linux/kernel/git/sschmidt/wpan
Stefan Schmidt says:
====================
pull-request: ieee802154 for net 2018-09-28
An update from ieee802154 for your *net* tree.
Some cleanup patches throughout the drivers from the Huawei tag team
Yue Haibing and Zhong Jiang.
Xue is replacing some magic numbers with defines in his mcr20a driver.
====================
David S. Miller [Sat, 29 Sep 2018 18:28:17 +0000 (11:28 -0700)]
Merge tag 'rxrpc-fixes-20180928' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs
David Howells says:
====================
rxrpc: Fixes
Here are some miscellaneous fixes for AF_RXRPC:
(1) Remove a duplicate variable initialisation.
(2) Fix one of the checks made when we decide to set up a new incoming
service call in which a flag is being checked in the wrong field of
the packet header. This check is abstracted out into helper
functions.
(3) Fix RTT gathering. The code has been trying to make use of socket
timestamps, but wasn't actually enabling them. The code has also been
recording a transmit time for the outgoing packet for which we're
going to measure the RTT after sending the message - but we can get
the incoming packet before we get to that and record a negative RTT.
(4) Fix the emission of BUSY packets (we are emitting ABORTs instead).
(5) Improve error checking on incoming packets.
(6) Try to fix a bug in new service call handling whereby a BUG we should
never be able to reach somehow got triggered. Do this by moving much
of the checking as early as possible and not repeating it later
(depends on (5) above).
(7) Fix the sockopts set on a UDP6 socket to include the ones set on a
UDP4 socket so that we receive UDP4 errors and packet-too-large
notifications too.
(8) Fix the distribution of errors so that we do it at the point of
receiving an error in the UDP callback rather than deferring it
thereby cutting short any transmissions that would otherwise occur in
the window.
====================
David S. Miller [Fri, 28 Sep 2018 18:12:29 +0000 (11:12 -0700)]
Merge branch 'netpoll-second-round-of-fixes'
Eric Dumazet says:
====================
netpoll: second round of fixes.
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC).
This capture, showing one ksoftirqd eating all cycles
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
It seems that all networking drivers that do use NAPI
for their TX completions, should not provide a ndo_poll_controller() :
Most NAPI drivers have netpoll support already handled
in core networking stack, since netpoll_poll_dev()
uses poll_napi(dev) to iterate through registered
NAPI contexts for a device.
First patch is a fix in poll_one_napi().
Then following patches take care of ten drivers.
====================
Eric Dumazet [Thu, 27 Sep 2018 16:32:01 +0000 (09:32 -0700)]
ibmvnic: remove ndo_poll_controller
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC). This capture
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
ibmvnic uses NAPI for TX completions, so we better let core
networking stack call the napi->poll() to avoid the capture.
ibmvnic_netpoll_controller() was completely wrong anyway,
as it was scheduling NAPI to service RX queues (instead of TX),
so I doubt netpoll ever worked on this driver.
Eric Dumazet [Thu, 27 Sep 2018 16:32:00 +0000 (09:32 -0700)]
sfc-falcon: remove ndo_poll_controller
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC). This capture
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
sfc-falcon uses NAPI for TX completions, so we better let core
networking stack call the napi->poll() to avoid the capture.
Eric Dumazet [Thu, 27 Sep 2018 16:31:59 +0000 (09:31 -0700)]
sfc: remove ndo_poll_controller
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC). This capture
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
sfc uses NAPI for TX completions, so we better let core
networking stack call the napi->poll() to avoid the capture.
Eric Dumazet [Thu, 27 Sep 2018 16:31:58 +0000 (09:31 -0700)]
net: ena: remove ndo_poll_controller
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC). This capture
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
ena uses NAPI for TX completions, so we better let core
networking stack call the napi->poll() to avoid the capture.
Eric Dumazet [Thu, 27 Sep 2018 16:31:57 +0000 (09:31 -0700)]
qlogic: netxen: remove ndo_poll_controller
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC). This capture
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
netxen uses NAPI for TX completions, so we better let core
networking stack call the napi->poll() to avoid the capture.
Eric Dumazet [Thu, 27 Sep 2018 16:31:56 +0000 (09:31 -0700)]
qlcnic: remove ndo_poll_controller
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC). This capture
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
qlcnic uses NAPI for TX completions, so we better let core
networking stack call the napi->poll() to avoid the capture.
Eric Dumazet [Thu, 27 Sep 2018 16:31:55 +0000 (09:31 -0700)]
virtio_net: remove ndo_poll_controller
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC). This capture
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
virto_net uses NAPI for TX completions, so we better let core
networking stack call the napi->poll() to avoid the capture.
Eric Dumazet [Thu, 27 Sep 2018 16:31:54 +0000 (09:31 -0700)]
net: hns: remove ndo_poll_controller
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC). This capture
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
hns uses NAPI for TX completions, so we better let core
networking stack call the napi->poll() to avoid the capture.
Eric Dumazet [Thu, 27 Sep 2018 16:31:53 +0000 (09:31 -0700)]
ehea: remove ndo_poll_controller
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC). This capture
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
ehea uses NAPI for TX completions, so we better let core
networking stack call the napi->poll() to avoid the capture.
Eric Dumazet [Thu, 27 Sep 2018 16:31:52 +0000 (09:31 -0700)]
hinic: remove ndo_poll_controller
As diagnosed by Song Liu, ndo_poll_controller() can
be very dangerous on loaded hosts, since the cpu
calling ndo_poll_controller() might steal all NAPI
contexts (for all RX/TX queues of the NIC). This capture
can last for unlimited amount of time, since one
cpu is generally not able to drain all the queues under load.
hinic uses NAPI for TX completions, so we better let core
networking stack call the napi->poll() to avoid the capture.
Note that hinic_netpoll() was incorrectly scheduling NAPI
on both RX and TX queues.
Eric Dumazet [Thu, 27 Sep 2018 16:31:51 +0000 (09:31 -0700)]
netpoll: do not test NAPI_STATE_SCHED in poll_one_napi()
Since we do no longer require NAPI drivers to provide
an ndo_poll_controller(), napi_schedule() has not been done
before poll_one_napi() invocation.
So testing NAPI_STATE_SCHED is likely to cause early returns.
While we are at it, remove outdated comment.
Note to future bisections : This change might surface prior
bugs in drivers. See commit 73f21c653f93 ("bnxt_en: Fix TX
timeout during netpoll.") for one occurrence.
David S. Miller [Fri, 28 Sep 2018 17:41:59 +0000 (10:41 -0700)]
Merge tag 'mac80211-for-davem-2018-09-27' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211
Johannes Berg says:
====================
More patches than I'd like perhaps, but each seems reasonable:
* two new spectre-v1 mitigations in nl80211
* TX status fix in general, and mesh in particular
* powersave vs. offchannel fix
* regulatory initialization fix
* fix for a queue hang due to a bad return value
* allocate TXQs for active monitor interfaces, fixing my
earlier patch to avoid unnecessary allocations where I
missed this case needed them
* fix TDLS data frames priority assignment
* fix scan results processing to take into account duplicate
channel numbers (over different operating classes, but we
don't necessarily know the operating class)
* various hwsim fixes for radio destruction and new radio
announcement messages
* remove an extraneous kernel-doc line
====================
qed: Fix shmem structure inconsistency between driver and the mfw.
The structure shared between driver and the management FW (mfw) differ in
sizes. This would lead to issues when driver try to access the structure
members which are not-aligned with the mfw copy e.g., data_ptr usage in the
case of mfw_tlv request.
Align the driver structure with mfw copy, add reserved field(s) to driver
structure for the members not used by the driver.
I haven't been doing reviews only but not active development on bridge
code for several years. Roopa and Nikolay have been doing most of
the new features and have agreed to take over as new co-maintainers.
David S. Miller [Fri, 28 Sep 2018 17:27:33 +0000 (10:27 -0700)]
Merge branch 's390-qeth-fixes'
Julian Wiedmann says:
====================
s390/qeth: fixes 2019-09-26
please apply two qeth patches for -net. The first is a trivial cleanup
required for patch #2 by Jean, which fixes a potential endless loop.
====================
Jean Delvare [Wed, 26 Sep 2018 16:07:10 +0000 (18:07 +0200)]
s390: qeth: Fix potential array overrun in cmd/rc lookup
Functions qeth_get_ipa_msg and qeth_get_ipa_cmd_name are modifying
the last member of global arrays without any locking that I can see.
If two instances of either function are running at the same time,
it could cause a race ultimately leading to an array overrun (the
contents of the last entry of the array is the only guarantee that
the loop will ever stop).
Performing the lookups without modifying the arrays is admittedly
slower (two comparisons per iteration instead of one) but these
are operations which are rare (should only be needed in error
cases or when debugging, not during successful operation) and it
seems still less costly than introducing a mutex to protect the
arrays in question.
As a side bonus, it allows us to declare both arrays as const data.
Unfortunately some versions of gcc emit following warning:
$ make net/xfrm/xfrm_output.o
linux/compiler.h:252:20: warning: array subscript is above array bounds [-Warray-bounds]
hook_head = rcu_dereference(net->nf.hooks_arp[hook]);
^~~~~~~~~~~~~~~~~~~~~
xfrm_output_resume passes skb_dst(skb)->ops->family as its 'pf' arg so compiler
can't know that we'll never access hooks_arp[].
(NFPROTO_IPV4 or NFPROTO_IPV6 are only possible cases).
Avoid this by adding an explicit WARN_ON_ONCE() check.
This patch has no effect if the family is a compile-time constant as gcc
will remove the switch() construct entirely.
Taehee Yoo [Thu, 30 Aug 2018 08:56:52 +0000 (17:56 +0900)]
netfilter: nft_set_rbtree: add missing rb_erase() in GC routine
The nft_set_gc_batch_check() checks whether gc buffer is full.
If gc buffer is full, gc buffer is released by
the nft_set_gc_batch_complete() internally.
In case of rbtree, the rb_erase() should be called before calling the
nft_set_gc_batch_complete(). therefore the rb_erase() should
be called before calling the nft_set_gc_batch_check() too.
test commands:
table ip filter {
set set1 {
type ipv4_addr; flags interval, timeout;
gc-interval 10s;
timeout 1s;
elements = {
1-2,
3-4,
5-6,
...
10000-10001,
}
}
}
%nft -f test.nft
David Howells [Thu, 27 Sep 2018 14:13:09 +0000 (15:13 +0100)]
rxrpc: Fix error distribution
Fix error distribution by immediately delivering the errors to all the
affected calls rather than deferring them to a worker thread. The problem
with the latter is that retries and things can happen in the meantime when we
want to stop that sooner.
To this end:
(1) Stop the error distributor from removing calls from the error_targets
list so that peer->lock isn't needed to synchronise against other adds
and removals.
(2) Require the peer's error_targets list to be accessed with RCU, thereby
avoiding the need to take peer->lock over distribution.
(3) Don't attempt to affect a call's state if it is already marked complete.
David Howells [Thu, 27 Sep 2018 14:13:09 +0000 (15:13 +0100)]
rxrpc: Fix transport sockopts to get IPv4 errors on an IPv6 socket
It seems that enabling IPV6_RECVERR on an IPv6 socket doesn't also turn on
IP_RECVERR, so neither local errors nor ICMP-transported remote errors from
IPv4 peer addresses are returned to the AF_RXRPC protocol.
Make the sockopt setting code in rxrpc_open_socket() fall through from the
AF_INET6 case to the AF_INET case to turn on all the AF_INET options too in
the AF_INET6 case.
Fixes: f2aeed3a591f ("rxrpc: Fix error reception on AF_INET6 sockets") Signed-off-by: David Howells <[email protected]>
David Howells [Thu, 27 Sep 2018 14:13:09 +0000 (15:13 +0100)]
rxrpc: Make service call handling more robust
Make the following changes to improve the robustness of the code that sets
up a new service call:
(1) Cache the rxrpc_sock struct obtained in rxrpc_data_ready() to do a
service ID check and pass that along to rxrpc_new_incoming_call().
This means that I can remove the check from rxrpc_new_incoming_call()
without the need to worry about the socket attached to the local
endpoint getting replaced - which would invalidate the check.
(2) Cache the rxrpc_peer struct, thereby allowing the peer search to be
done once. The peer is passed to rxrpc_new_incoming_call(), thereby
saving the need to repeat the search.
This also reduces the possibility of rxrpc_publish_service_conn()
BUG()'ing due to the detection of a duplicate connection, despite the
initial search done by rxrpc_find_connection_rcu() having turned up
nothing.
This BUG() shouldn't ever get hit since rxrpc_data_ready() *should* be
non-reentrant and the result of the initial search should still hold
true, but it has proven possible to hit.
I *think* this may be due to __rxrpc_lookup_peer_rcu() cutting short
the iteration over the hash table if it finds a matching peer with a
zero usage count, but I don't know for sure since it's only ever been
hit once that I know of.
Another possibility is that a bug in rxrpc_data_ready() that checked
the wrong byte in the header for the RXRPC_CLIENT_INITIATED flag
might've let through a packet that caused a spurious and invalid call
to be set up. That is addressed in another patch.
(3) Fix __rxrpc_lookup_peer_rcu() to skip peer records that have a zero
usage count rather than stopping and returning not found, just in case
there's another peer record behind it in the bucket.
(4) Don't search the peer records in rxrpc_alloc_incoming_call(), but
rather either use the peer cached in (2) or, if one wasn't found,
preemptively install a new one.
Fixes: 8496af50eb38 ("rxrpc: Use RCU to access a peer's service connection tree") Signed-off-by: David Howells <[email protected]>
David Howells [Thu, 27 Sep 2018 14:13:08 +0000 (15:13 +0100)]
rxrpc: Improve up-front incoming packet checking
Do more up-front checking on incoming packets to weed out invalid ones and
also ones aimed at services that we don't support.
Whilst we're at it, replace the clearing of call and skew if we don't find
a connection with just initialising the variables to zero at the top of the
function.
David Howells [Thu, 27 Sep 2018 14:13:08 +0000 (15:13 +0100)]
rxrpc: Emit BUSY packets when supposed to rather than ABORTs
In the input path, a received sk_buff can be marked for rejection by
setting RXRPC_SKB_MARK_* in skb->mark and, if needed, some auxiliary data
(such as an abort code) in skb->priority. The rejection is handled by
queueing the sk_buff up for dealing with in process context. The output
code reads the mark and priority and, theoretically, generates an
appropriate response packet.
However, if RXRPC_SKB_MARK_BUSY is set, this isn't noticed and an ABORT
message with a random abort code is generated (since skb->priority wasn't
set to anything).
Fix this by outputting the appropriate sort of packet.
Also, whilst we're at it, most of the marks are no longer used, so remove
them and rename the remaining two to something more obvious.
Fixes: 248f219cb8bc ("rxrpc: Rewrite the data and ack handling code") Signed-off-by: David Howells <[email protected]>
but that only really applies to service calls that we're handling,
since the client side gets to send the final ACK (or not).
(4) When about to transmit an ACK or DATA packet, record the Tx timestamp
before only; don't update the timestamp afterwards.
(5) Switch the ordering between recording the serial and recording the
timestamp to always set the serial number first. The serial number
shouldn't be seen referenced by an ACK packet until we've transmitted
the packet bearing it - so in the Rx path, we don't need the timestamp
until we've checked the serial number.
nl80211: Fix possible Spectre-v1 for CQM RSSI thresholds
Use array_index_nospec() to sanitize i with respect to speculation.
Note that the user doesn't control i directly, but can make it out
of bounds by not finding a threshold in the array.
Signed-off-by: Masashi Honma <[email protected]>
[add note about user control, as explained by Masashi] Signed-off-by: Johannes Berg <[email protected]>
Michael Chan [Wed, 26 Sep 2018 04:41:04 +0000 (00:41 -0400)]
bnxt_en: Fix TX timeout during netpoll.
The current netpoll implementation in the bnxt_en driver has problems
that may miss TX completion events. bnxt_poll_work() in effect is
only handling at most 1 TX packet before exiting. In addition,
there may be in flight TX completions that ->poll() may miss even
after we fix bnxt_poll_work() to handle all visible TX completions.
netpoll may not call ->poll() again and HW may not generate IRQ
because the driver does not ARM the IRQ when the budget (0 for netpoll)
is reached.
We fix it by handling all TX completions and to always ARM the IRQ
when we exit ->poll() with 0 budget.
Also, the logic to ACK the completion ring in case it is almost filled
with TX completions need to be adjusted to take care of the 0 budget
case, as discussed with Eric Dumazet <[email protected]>
Hangbin Liu [Wed, 26 Sep 2018 02:35:42 +0000 (10:35 +0800)]
vxlan: fill ttl inherit info
When add vxlan ttl inherit support, I forgot to fill it when dump
vlxan info. Fix it now.
Fixes: 72f6d71e491e6 ("vxlan: add ttl inherit support") Signed-off-by: Hangbin Liu <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Andrew Lunn [Mon, 24 Sep 2018 23:50:00 +0000 (01:50 +0200)]
net: phy: sfp: Fix unregistering of HWMON SFP device
A HWMON device is only registered is the SFP module supports the
diagnostic page and is complient to SFF8472. Don't unconditionally
unregister the hwmon device when the SFP module is remove, otherwise
we access data structures which don't exist.
qed: Avoid implicit enum conversion in qed_iwarp_parse_rx_pkt
Clang warns when one enumerated type is implicitly converted to another.
drivers/net/ethernet/qlogic/qed/qed_iwarp.c:1713:25: warning: implicit
conversion from enumeration type 'enum tcp_ip_version' to different
enumeration type 'enum qed_tcp_ip_version' [-Wenum-conversion]
cm_info->ip_version = TCP_IPV4;
~ ^~~~~~~~
drivers/net/ethernet/qlogic/qed/qed_iwarp.c:1733:25: warning: implicit
conversion from enumeration type 'enum tcp_ip_version' to different
enumeration type 'enum qed_tcp_ip_version' [-Wenum-conversion]
cm_info->ip_version = TCP_IPV6;
~ ^~~~~~~~
2 warnings generated.
Use the appropriate values from the expected type, qed_tcp_ip_version:
qed: Avoid constant logical operation warning in qed_vf_pf_acquire
Clang warns when a constant is used in a boolean context as it thinks a
bitwise operation may have been intended.
drivers/net/ethernet/qlogic/qed/qed_vf.c:415:27: warning: use of logical
'&&' with constant operand [-Wconstant-logical-operand]
if (!p_iov->b_pre_fp_hsi &&
^
drivers/net/ethernet/qlogic/qed/qed_vf.c:415:27: note: use '&' for a
bitwise operation
if (!p_iov->b_pre_fp_hsi &&
^~
&
drivers/net/ethernet/qlogic/qed/qed_vf.c:415:27: note: remove constant
to silence this warning
if (!p_iov->b_pre_fp_hsi &&
~^~
1 warning generated.
This has been here since commit 1fe614d10f45 ("qed: Relax VF firmware
requirements") and I am not entirely sure why since 0 isn't a special
case. Just remove the statement causing Clang to warn since it isn't
required.
bonding: pass link-local packets to bonding master also.
Commit b89f04c61efe ("bonding: deliver link-local packets with
skb->dev set to link that packets arrived on") changed the behavior
of how link-local-multicast packets are processed. The change in
the behavior broke some legacy use cases where these packets are
expected to arrive on bonding master device also.
This patch passes the packet to the stack with the link it arrived
on as well as passes to the bonding-master device to preserve the
legacy use case.
Fixes: b89f04c61efe ("bonding: deliver link-local packets with skb->dev set to link that packets arrived on") Reported-by: Michal Soltys <[email protected]> Signed-off-by: Mahesh Bandewar <[email protected]> Signed-off-by: David S. Miller <[email protected]>
qed: Avoid implicit enum conversion in qed_roce_mode_to_flavor
Clang warns when one enumerated type is implicitly converted to another.
drivers/net/ethernet/qlogic/qed/qed_roce.c:153:12: warning: implicit
conversion from enumeration type 'enum roce_mode' to different
enumeration type 'enum roce_flavor' [-Wenum-conversion]
flavor = ROCE_V2_IPV6;
~ ^~~~~~~~~~~~
drivers/net/ethernet/qlogic/qed/qed_roce.c:156:12: warning: implicit
conversion from enumeration type 'enum roce_mode' to different
enumeration type 'enum roce_flavor' [-Wenum-conversion]
flavor = MAX_ROCE_MODE;
~ ^~~~~~~~~~~~~
2 warnings generated.
Use the appropriate values from the expected type, roce_flavor:
qed: Fix mask parameter in qed_vf_prep_tunn_req_tlv
Clang complains when one enumerated type is implicitly converted to
another.
drivers/net/ethernet/qlogic/qed/qed_vf.c:686:6: warning: implicit
conversion from enumeration type 'enum qed_tunn_mode' to different
enumeration type 'enum qed_tunn_clss' [-Wenum-conversion]
QED_MODE_L2GENEVE_TUNN,
^~~~~~~~~~~~~~~~~~~~~~
Update mask's parameter to expect qed_tunn_mode, which is what was
intended.
qed: Avoid implicit enum conversion in qed_set_tunn_cls_info
Clang warns when one enumerated type is implicitly converted to another.
drivers/net/ethernet/qlogic/qed/qed_sp_commands.c:163:25: warning:
implicit conversion from enumeration type 'enum tunnel_clss' to
different enumeration type 'enum qed_tunn_clss' [-Wenum-conversion]
p_tun->vxlan.tun_cls = type;
~ ^~~~
drivers/net/ethernet/qlogic/qed/qed_sp_commands.c:165:26: warning:
implicit conversion from enumeration type 'enum tunnel_clss' to
different enumeration type 'enum qed_tunn_clss' [-Wenum-conversion]
p_tun->l2_gre.tun_cls = type;
~ ^~~~
drivers/net/ethernet/qlogic/qed/qed_sp_commands.c:167:26: warning:
implicit conversion from enumeration type 'enum tunnel_clss' to
different enumeration type 'enum qed_tunn_clss' [-Wenum-conversion]
p_tun->ip_gre.tun_cls = type;
~ ^~~~
drivers/net/ethernet/qlogic/qed/qed_sp_commands.c:169:29: warning:
implicit conversion from enumeration type 'enum tunnel_clss' to
different enumeration type 'enum qed_tunn_clss' [-Wenum-conversion]
p_tun->l2_geneve.tun_cls = type;
~ ^~~~
drivers/net/ethernet/qlogic/qed/qed_sp_commands.c:171:29: warning:
implicit conversion from enumeration type 'enum tunnel_clss' to
different enumeration type 'enum qed_tunn_clss' [-Wenum-conversion]
p_tun->ip_geneve.tun_cls = type;
~ ^~~~
5 warnings generated.
Trivial fix to spelling mistake in ms_to_errno array of error messages
and remove confusing "not" from the error text since the error code
refers to an uninitialized error code.
====================
net: phy: fix WoL handling when suspending the PHY
phy_suspend doesn't always recognize that WoL is enabled and therefore
suspends the PHY when it should not. First idea to address the issue
was to reuse checks used in mdio_bus_phy_may_suspend which check
whether relevant devices are wakeup-enabled.
Florian raised some concerns because drivers may enable wakeup even if
WoL isn't enabled (e.g. certain USB network drivers).
The new approach focuses on reducing the risk to break existing stuff.
We add a flag wol_enabled to struct net_device which is set in
ethtool_set_wol(). Then this flag is checked in phy_suspend().
This doesn't cover 100% of the cases yet (e.g. if WoL is enabled w/o
explicit configuration), but it covers the most relevant cases with
very little risk of regressions.
net: phy: fix WoL handling when suspending the PHY
Core of the problem is that phy_suspend() suspends the PHY when it
should not because of WoL. phy_suspend() checks for WoL already, but
this works only if the PHY driver handles WoL (what is rarely the case).
Typically WoL is handled by the MAC driver.
This patch uses new member wol_enabled of struct net_device as
additional criteria in the check when not to suspend the PHY because
of WoL.
Last but not least change phy_detach() to call phy_suspend() before
attached_dev is set to NULL. phy_suspend() accesses attached_dev
when checking whether the MAC driver activated WoL.
Fixes: f1e911d5d0df ("r8169: add basic phylib support") Fixes: e8cfd9d6c772 ("net: phy: call state machine synchronously in phy_stop") Signed-off-by: Heiner Kallweit <[email protected]> Reviewed-by: Florian Fainelli <[email protected]> Signed-off-by: David S. Miller <[email protected]>
net: core: add member wol_enabled to struct net_device
Add flag wol_enabled to struct net_device indicating whether
Wake-on-LAN is enabled. As first user phy_suspend() will use it to
decide whether PHY can be suspended or not.
Fixes: f1e911d5d0df ("r8169: add basic phylib support") Fixes: e8cfd9d6c772 ("net: phy: call state machine synchronously in phy_stop") Signed-off-by: Heiner Kallweit <[email protected]> Reviewed-by: Florian Fainelli <[email protected]> Signed-off-by: David S. Miller <[email protected]>
net: phy: fix WoL handling when suspending the PHY
Actually there's nothing wrong with the two changes marked as "Fixes",
they just revealed a problem which has been existing before.
After having switched r8169 to phylib it was reported that WoL from
shutdown doesn't work any longer (WoL from suspend isn't affected).
Reason is that during shutdown phy_disconnect()->phy_detach()->
phy_suspend() is called.
A similar issue occurs when the phylib state machine calls
phy_suspend() when handling state PHY_HALTED.
Core of the problem is that phy_suspend() suspends the PHY when it
should not due to WoL. phy_suspend() checks for WoL already, but this
works only if the PHY driver handles WoL (what is rarely the case).
Typically WoL is handled by the MAC driver.
phylib knows about this and handles it in mdio_bus_phy_may_suspend(),
but that's used only when suspending the system, not in other cases
like shutdown.
Therefore factor out the relevant check from
mdio_bus_phy_may_suspend() to a new function phy_may_suspend() and
use it in phy_suspend().
Last but not least change phy_detach() to call phy_suspend() before
attached_dev is set to NULL. phy_suspend() accesses attached_dev
when checking whether the MAC driver activated WoL.
Fixes: f1e911d5d0df ("r8169: add basic phylib support") Fixes: e8cfd9d6c772 ("net: phy: call state machine synchronously in phy_stop") Signed-off-by: Heiner Kallweit <[email protected]> Signed-off-by: David S. Miller <[email protected]>
David Ahern [Thu, 27 Sep 2018 00:35:14 +0000 (17:35 -0700)]
net/ipv6: Remove extra call to ip6_convert_metrics for multipath case
The change to move metrics from the dst to rt6_info moved the call
to ip6_convert_metrics from ip6_route_add to ip6_route_info_create. In
doing so it makes the call in ip6_route_info_append redundant and
actually leaks the metrics installed as part of the ip6_route_info_create.
Remove the now unnecessary call.
Fixes: d4ead6b34b67f ("net/ipv6: move metrics from dst to rt6_info") Signed-off-by: David Ahern <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Crash happened because list_del_rcu() was called twice for smp->ltk. This
was possible if unpair_device was called right after ltk was generated
but before keys were distributed.
In this commit smp_cancel_pairing was refactored to cancel pairing if it
is in progress and otherwise just removes keys. Once keys are removed from
rcu list, pointers to smp context's keys are set to NULL to make sure
removed list items are not accessed later.
This commit also adjusts the functionality of mgmt unpair_device() little
bit. Previously pairing was canceled only if pairing was in state that
keys were already generated. With this commit unpair_device() cancels
pairing already in earlier states.
Bug was found by fuzzing kernel SMP implementation using Synopsys
Defensics.
Martin Willi [Tue, 25 Sep 2018 07:51:02 +0000 (09:51 +0200)]
mac80211_hwsim: do not omit multicast announce of first added radio
The allocation of hwsim radio identifiers uses a post-increment from 0,
so the first radio has idx 0. This idx is explicitly excluded from
multicast announcements ever since, but it is unclear why.
Drop that idx check and announce the first radio as well. This makes
userspace happy if it relies on these events.
Martin Willi [Tue, 25 Sep 2018 07:41:14 +0000 (09:41 +0200)]
mac80211_hwsim: fix race in radio destruction from netlink notifier
The asynchronous destruction from a work-queue of radios tagged with
destroy-on-close may race with the owning namespace about to exit,
resulting in potential use-after-free of that namespace.
Instead of using a work-queue, move radios about to destroy to a
temporary list, which can be worked on synchronously after releasing
the lock. This should be safe to do from the netlink socket notifier,
as the namespace is guaranteed to not get released.
Martin Willi [Tue, 25 Sep 2018 07:41:13 +0000 (09:41 +0200)]
mac80211_hwsim: fix locking when iterating radios during ns exit
The cleanup of radios during namespace exit has recently been reworked
to directly delete a radio while temporarily releasing the spinlock,
fixing a race condition between the work-queue execution and namespace
exits. However, the temporary unlock allows unsafe modifications on the
iterated list, resulting in a potential crash when continuing the
iteration of additional radios.
Move radios about to destroy to a temporary list, and clean that up
after releasing the spinlock once iteration is complete.
Fixes: 8cfd36a0b53a ("mac80211_hwsim: fix use-after-free bug in hwsim_exit_net") Signed-off-by: Martin Willi <[email protected]> Signed-off-by: Johannes Berg <[email protected]>
Randy Dunlap [Sun, 23 Sep 2018 18:59:13 +0000 (11:59 -0700)]
cfg80211: fix reg_query_regdb_wmm kernel-doc
Drop @ptr from kernel-doc for function reg_query_regdb_wmm().
This function parameter was recently removed so update the
kernel-doc to match that and remove the kernel-doc warnings.
Removes 109 occurrences of this warning message:
../include/net/cfg80211.h:4869: warning: Excess function parameter 'ptr' description in 'reg_query_regdb_wmm'
Felix Fietkau [Sat, 22 Sep 2018 16:35:31 +0000 (18:35 +0200)]
mac80211: allocate TXQs for active monitor interfaces
Monitor mode interfaces with the active flag are passed down to the driver.
Drivers using TXQ expect that all interfaces have allocated TXQs before
they get added.
In tipc_link_reset() we copy the wakeup queue to input queue using
skb_queue_splice_init(link->wakeupq, link->inputq).
This is performed without holding any locks. The lists might be
simultaneously be accessed by other cpu threads in tipc_sk_rcv(),
something leading to to random missing packets.
tipc: fix flow control accounting for implicit connect
In the case of implicit connect message with data > 1K, the flow
control accounting is incorrect. At this state, the socket does not
know the peer nodes capability and falls back to legacy flow control
by return 1, however the receiver of this message will perform the
new block accounting. This leads to a slack and eventually traffic
disturbance.
In this commit, we perform tipc_node_get_capabilities() at implicit
connect and perform accounting based on the peer's capability.
Yunsheng Lin [Tue, 25 Sep 2018 09:21:55 +0000 (10:21 +0100)]
net: hns: fix for unmapping problem when SMMU is on
If SMMU is on, there is more likely that skb_shinfo(skb)->frags[i]
can not send by a single BD. when this happen, the
hns_nic_net_xmit_hw function map the whole data in a frags using
skb_frag_dma_map, but unmap each BD' data individually when tx is
done, which causes problem when SMMU is on.
This patch fixes this problem by ummapping the whole data in a
frags when tx is done.
projects / linux.git / log