OSDN Git Service

tomoyo/tomoyo-test1.git
6 years agovti6: Fix dev->max_mtu setting
Stefano Brivio [Thu, 15 Mar 2018 16:17:13 +0000 (17:17 +0100)]
vti6: Fix dev->max_mtu setting

We shouldn't allow a tunnel to have IP_MAX_MTU as MTU, because
another IPv6 header is going on top of our packets. Without this
patch, we might end up building packets bigger than IP_MAX_MTU.

Fixes: b96f9afee4eb ("ipv4/6: use core net MTU range checking")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Acked-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
6 years agovti6: Keep set MTU on link creation or change, validate it
Stefano Brivio [Thu, 15 Mar 2018 16:17:12 +0000 (17:17 +0100)]
vti6: Keep set MTU on link creation or change, validate it

In vti6_link_config(), if MTU is already given on link creation
or change, validate and use it instead of recomputing it. To do
that, we need to propagate the knowledge that MTU was set by
userspace all the way down to vti6_link_config().

To keep this simple, vti6_dev_init() sets the new 'keep_mtu'
argument of vti6_link_config() to true: on initialization, we
don't have convenient access to netlink attributes there, but we
will anyway check whether dev->mtu is set in vti6_link_config().
If it's non-zero, it was set to the value of the IFLA_MTU
attribute during creation. Otherwise, determine a reasonable
value.

Fixes: ed1efb2aefbb ("ipv6: Add support for IPsec virtual tunnel interfaces")
Fixes: 53c81e95df17 ("ip6_vti: adjust vti mtu according to mtu of lower device")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Acked-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
6 years agovti6: Properly adjust vti6 MTU from MTU of lower device
Stefano Brivio [Thu, 15 Mar 2018 16:17:11 +0000 (17:17 +0100)]
vti6: Properly adjust vti6 MTU from MTU of lower device

If a lower device is found, we don't need to subtract
LL_MAX_HEADER to calculate our MTU: just use its MTU, the link
layer headers are already taken into account by it.

If the lower device is not found, start from ETH_DATA_LEN
instead, and only in this case subtract a worst-case
LL_MAX_HEADER.

We then need to subtract our additional IPv6 header from the
calculation.

While at it, note that vti6 doesn't have a hardware header, so
it doesn't need to set dev->hard_header_len. And as
vti6_link_config() now always sets the MTU, there's no need to
set a default value in vti6_dev_setup().

This makes the behaviour consistent with IPv4 vti, after
commit a32452366b72 ("vti4: Don't count header length twice."),
which was accidentally reverted by merge commit f895f0cfbb77
("Merge branch 'master' of
git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec").

While commit 53c81e95df17 ("ip6_vti: adjust vti mtu according to
mtu of lower device") improved on the original situation, this
was still not ideal. As reported in that commit message itself,
if we start from an underlying veth MTU of 9000, we end up with
an MTU of 8832, that is, 9000 - LL_MAX_HEADER - sizeof(ipv6hdr).
This should simply be 8880, or 9000 - sizeof(ipv6hdr) instead:
we found the lower device (veth) and we know we don't have any
additional link layer header, so there's no need to subtract an
hypothetical worst-case number.

Fixes: 53c81e95df17 ("ip6_vti: adjust vti mtu according to mtu of lower device")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Acked-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
6 years agovti4: Don't override MTU passed on link creation via IFLA_MTU
Stefano Brivio [Thu, 15 Mar 2018 16:16:29 +0000 (17:16 +0100)]
vti4: Don't override MTU passed on link creation via IFLA_MTU

Don't hardcode a MTU value on vti tunnel initialization,
ip_tunnel_newlink() is able to deal with this already. See also
commit ffc2b6ee4174 ("ip_gre: fix IFLA_MTU ignored on NEWLINK").

Fixes: 1181412c1a67 ("net/ipv4: VTI support new module for ip_vti.")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Acked-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
6 years agoip_tunnel: Clamp MTU to bounds on new link
Stefano Brivio [Thu, 15 Mar 2018 16:16:28 +0000 (17:16 +0100)]
ip_tunnel: Clamp MTU to bounds on new link

Otherwise, it's possible to specify invalid MTU values directly
on creation of a link (via 'ip link add'). This is already
prevented on subsequent MTU changes by commit b96f9afee4eb
("ipv4/6: use core net MTU range checking").

Fixes: c54419321455 ("GRE: Refactor GRE tunneling code.")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Acked-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
6 years agovti4: Don't count header length twice on tunnel setup
Stefano Brivio [Thu, 15 Mar 2018 16:16:27 +0000 (17:16 +0100)]
vti4: Don't count header length twice on tunnel setup

This re-introduces the effect of commit a32452366b72 ("vti4:
Don't count header length twice.") which was accidentally
reverted by merge commit f895f0cfbb77 ("Merge branch 'master' of
git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec").

The commit message from Steffen Klassert said:

    We currently count the size of LL_MAX_HEADER and struct iphdr
    twice for vti4 devices, this leads to a wrong device mtu.
    The size of LL_MAX_HEADER and struct iphdr is already counted in
    ip_tunnel_bind_dev(), so don't do it again in vti_tunnel_init().

And this is still the case now: ip_tunnel_bind_dev() already
accounts for the header length of the link layer (not
necessarily LL_MAX_HEADER, if the output device is found), plus
one IP header.

For example, with a vti device on top of veth, with MTU of 1500,
the existing implementation would set the initial vti MTU to
1332, accounting once for LL_MAX_HEADER (128, included in
hard_header_len by vti) and twice for the same IP header (once
from hard_header_len, once from ip_tunnel_bind_dev()).

It should instead be 1480, because ip_tunnel_bind_dev() is able
to figure out that the output device is veth, so no additional
link layer header is attached, and will properly count one
single IP header.

The existing issue had the side effect of avoiding PMTUD for
most xfrm policies, by arbitrarily lowering the initial MTU.
However, the only way to get a consistent PMTU value is to let
the xfrm PMTU discovery do its course, and commit d6af1a31cc72
("vti: Add pmtu handling to vti_xmit.") now takes care of local
delivery cases where the application ignores local socket
notifications.

Fixes: b9959fd3b0fa ("vti: switch to new ip tunnel code")
Fixes: f895f0cfbb77 ("Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Acked-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
6 years agoxfrm: fix rcu_read_unlock usage in xfrm_local_error
Taehee Yoo [Fri, 16 Mar 2018 02:35:51 +0000 (11:35 +0900)]
xfrm: fix rcu_read_unlock usage in xfrm_local_error

In the xfrm_local_error, rcu_read_unlock should be called when afinfo
is not NULL. because xfrm_state_get_afinfo calls rcu_read_unlock
if afinfo is NULL.

Fixes: af5d27c4e12b ("xfrm: remove xfrm_state_put_afinfo")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
6 years agonet sched actions: return explicit error when tunnel_key mode is not specified
Roman Mashak [Mon, 12 Mar 2018 20:20:58 +0000 (16:20 -0400)]
net sched actions: return explicit error when tunnel_key mode is not specified

If set/unset mode of the tunnel_key action is not provided, ->init() still
returns 0, and the caller proceeds with bogus 'struct tc_action *' object,
this results in crash:

% tc actions add action tunnel_key src_ip 1.1.1.1 dst_ip 2.2.2.1 id 7 index 1

[   35.805515] general protection fault: 0000 [#1] SMP PTI
[   35.806161] Modules linked in: act_tunnel_key kvm_intel kvm irqbypass
crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcbc aesni_intel aes_x86_64
crypto_simd glue_helper cryptd serio_raw
[   35.808233] CPU: 1 PID: 428 Comm: tc Not tainted 4.16.0-rc4+ #286
[   35.808929] RIP: 0010:tcf_action_init+0x90/0x190
[   35.809457] RSP: 0018:ffffb8edc068b9a0 EFLAGS: 00010206
[   35.810053] RAX: 1320c000000a0003 RBX: 0000000000000001 RCX: 0000000000000000
[   35.810866] RDX: 0000000000000070 RSI: 0000000000007965 RDI: ffffb8edc068b910
[   35.811660] RBP: ffffb8edc068b9d0 R08: 0000000000000000 R09: ffffb8edc068b808
[   35.812463] R10: ffffffffc02bf040 R11: 0000000000000040 R12: ffffb8edc068bb38
[   35.813235] R13: 0000000000000000 R14: 0000000000000000 R15: ffffb8edc068b910
[   35.814006] FS:  00007f3d0d8556c0(0000) GS:ffff91d1dbc40000(0000)
knlGS:0000000000000000
[   35.814881] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   35.815540] CR2: 000000000043f720 CR3: 0000000019248001 CR4: 00000000001606a0
[   35.816457] Call Trace:
[   35.817158]  tc_ctl_action+0x11a/0x220
[   35.817795]  rtnetlink_rcv_msg+0x23d/0x2e0
[   35.818457]  ? __slab_alloc+0x1c/0x30
[   35.819079]  ? __kmalloc_node_track_caller+0xb1/0x2b0
[   35.819544]  ? rtnl_calcit.isra.30+0xe0/0xe0
[   35.820231]  netlink_rcv_skb+0xce/0x100
[   35.820744]  netlink_unicast+0x164/0x220
[   35.821500]  netlink_sendmsg+0x293/0x370
[   35.822040]  sock_sendmsg+0x30/0x40
[   35.822508]  ___sys_sendmsg+0x2c5/0x2e0
[   35.823149]  ? pagecache_get_page+0x27/0x220
[   35.823714]  ? filemap_fault+0xa2/0x640
[   35.824423]  ? page_add_file_rmap+0x108/0x200
[   35.825065]  ? alloc_set_pte+0x2aa/0x530
[   35.825585]  ? finish_fault+0x4e/0x70
[   35.826140]  ? __handle_mm_fault+0xbc1/0x10d0
[   35.826723]  ? __sys_sendmsg+0x41/0x70
[   35.827230]  __sys_sendmsg+0x41/0x70
[   35.827710]  do_syscall_64+0x68/0x120
[   35.828195]  entry_SYSCALL_64_after_hwframe+0x3d/0xa2
[   35.828859] RIP: 0033:0x7f3d0ca4da67
[   35.829331] RSP: 002b:00007ffc9f284338 EFLAGS: 00000246 ORIG_RAX:
000000000000002e
[   35.830304] RAX: ffffffffffffffda RBX: 00007ffc9f284460 RCX: 00007f3d0ca4da67
[   35.831247] RDX: 0000000000000000 RSI: 00007ffc9f2843b0 RDI: 0000000000000003
[   35.832167] RBP: 000000005aa6a7a9 R08: 0000000000000001 R09: 0000000000000000
[   35.833075] R10: 00000000000005f1 R11: 0000000000000246 R12: 0000000000000000
[   35.833997] R13: 00007ffc9f2884c0 R14: 0000000000000001 R15: 0000000000674640
[   35.834923] Code: 24 30 bb 01 00 00 00 45 31 f6 eb 5e 8b 50 08 83 c2 07 83 e2
fc 83 c2 70 49 8b 07 48 8b 40 70 48 85 c0 74 10 48 89 14 24 4c 89 ff <ff> d0 48
8b 14 24 48 01 c2 49 01 d6 45 85 ed 74 05 41 83 47 2c
[   35.837442] RIP: tcf_action_init+0x90/0x190 RSP: ffffb8edc068b9a0
[   35.838291] ---[ end trace a095c06ee4b97a26 ]---

Fixes: d0f6dd8a914f ("net/sched: Introduce act_tunnel_key")
Signed-off-by: Roman Mashak <mrv@mojatatu.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/smc: simplify wait when closing listen socket
Ursula Braun [Tue, 13 Mar 2018 09:41:54 +0000 (10:41 +0100)]
net/smc: simplify wait when closing listen socket

Closing of a listen socket wakes up kernel_accept() of
smc_tcp_listen_worker(), and then has to wait till smc_tcp_listen_worker()
gives up the internal clcsock. The wait logic introduced with
commit 127f49705823 ("net/smc: release clcsock from tcp_listen_worker")
might wait longer than necessary. This patch implements the idea to
implement the wait just with flush_work(), and gets rid of the extra
smc_close_wait_listen_clcsock() function.

Fixes: 127f49705823 ("net/smc: release clcsock from tcp_listen_worker")
Reported-by: Hans Wippel <hwippel@linux.vnet.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosunvnet: does not support GSO for sctp
Cathy Zhou [Wed, 14 Mar 2018 17:56:07 +0000 (10:56 -0700)]
sunvnet: does not support GSO for sctp

The NETIF_F_GSO_SOFTWARE implies support for GSO on SCTP, but the
sunvnet driver does not support GSO for sctp.  Here we remove the
NETIF_F_GSO_SOFTWARE feature flag and only report NETIF_F_ALL_TSO
instead.

Signed-off-by: Cathy Zhou <Cathy.Zhou@Oracle.COM>
Signed-off-by: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'linux-can-fixes-for-4.16-20180314' of ssh://gitolite.kernel.org/pub/scm...
David S. Miller [Wed, 14 Mar 2018 17:50:47 +0000 (13:50 -0400)]
Merge tag 'linux-can-fixes-for-4.16-20180314' of ssh://gitolite./linux/kernel/git/mkl/linux-can

Marc Kleine-Budde says:

====================
pull-request: can 2018-03-14

this is a pull request of two patches for net/master.

Both patches are by Andri Yngvason and fix problems in the cc770 driver,
that show up quite fast on RT systems, but also on non RT setups.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotg3: prevent scheduling while atomic splat
Jonathan Toppins [Wed, 14 Mar 2018 16:36:25 +0000 (12:36 -0400)]
tg3: prevent scheduling while atomic splat

The problem was introduced in commit
506b0a395f26 ("[netdrv] tg3: APE heartbeat changes"). The bug occurs
because tp->lock spinlock is held which is obtained in tg3_start
by way of tg3_full_lock(), line 11571. The documentation for usleep_range()
specifically states it cannot be used inside a spinlock.

Fixes: 506b0a395f26 ("[netdrv] tg3: APE heartbeat changes")
Signed-off-by: Jonathan Toppins <jtoppins@redhat.com>
Acked-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv4: lock mtu in fnhe when received PMTU < net.ipv4.route.min_pmtu
Sabrina Dubroca [Wed, 14 Mar 2018 09:21:14 +0000 (10:21 +0100)]
ipv4: lock mtu in fnhe when received PMTU < net.ipv4.route.min_pmtu

Prior to the rework of PMTU information storage in commit
2c8cec5c10bc ("ipv4: Cache learned PMTU information in inetpeer."),
when a PMTU event advertising a PMTU smaller than
net.ipv4.route.min_pmtu was received, we would disable setting the DF
flag on packets by locking the MTU metric, and set the PMTU to
net.ipv4.route.min_pmtu.

Since then, we don't disable DF, and set PMTU to
net.ipv4.route.min_pmtu, so the intermediate router that has this link
with a small MTU will have to drop the packets.

This patch reestablishes pre-2.6.39 behavior by splitting
rtable->rt_pmtu into a bitfield with rt_mtu_locked and rt_pmtu.
rt_mtu_locked indicates that we shouldn't set the DF bit on that path,
and is checked in ip_dont_fragment().

One possible workaround is to set net.ipv4.route.min_pmtu to a value low
enough to accommodate the lowest MTU encountered.

Fixes: 2c8cec5c10bc ("ipv4: Cache learned PMTU information in inetpeer.")
Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'DPAA-Ethernet-fixes'
David S. Miller [Wed, 14 Mar 2018 17:16:23 +0000 (13:16 -0400)]
Merge branch 'DPAA-Ethernet-fixes'

Madalin Bucur says:

====================
DPAA Ethernet fixes

This patch set is addressing several issues in the DPAA Ethernet
driver suite:

 - module unload crash caused by wrong reference to device being left
   in the cleanup code after the DSA related changes
 - scheduling wile atomic bug in QMan code revealed during dpaa_eth
   module unload
 - a couple of error counter fixes, a duplicated init in dpaa_eth.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodpaa_eth: remove duplicate increment of the tx_errors counter
Camelia Groza [Wed, 14 Mar 2018 13:37:32 +0000 (08:37 -0500)]
dpaa_eth: remove duplicate increment of the tx_errors counter

The tx_errors counter is incremented by the dpaa_xmit caller.

Signed-off-by: Camelia Groza <camelia.groza@nxp.com>
Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodpaa_eth: increment the RX dropped counter when needed
Camelia Groza [Wed, 14 Mar 2018 13:37:31 +0000 (08:37 -0500)]
dpaa_eth: increment the RX dropped counter when needed

Signed-off-by: Camelia Groza <camelia.groza@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodpaa_eth: remove duplicate initialization
Camelia Groza [Wed, 14 Mar 2018 13:37:30 +0000 (08:37 -0500)]
dpaa_eth: remove duplicate initialization

The fd_format has already been initialized at this point.

Signed-off-by: Camelia Groza <camelia.groza@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodpaa_eth: fix error in dpaa_remove()
Madalin Bucur [Wed, 14 Mar 2018 13:37:29 +0000 (08:37 -0500)]
dpaa_eth: fix error in dpaa_remove()

The recent changes that make the driver probing compatible with DSA
were not propagated in the dpa_remove() function, breaking the
module unload function. Using the proper device to address the issue.

Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosoc/fsl/qbman: fix issue in qman_delete_cgr_safe()
Madalin Bucur [Wed, 14 Mar 2018 13:37:28 +0000 (08:37 -0500)]
soc/fsl/qbman: fix issue in qman_delete_cgr_safe()

The wait_for_completion() call in qman_delete_cgr_safe()
was triggering a scheduling while atomic bug, replacing the
kthread with a smp_call_function_single() call to fix it.

Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Signed-off-by: Roy Pledge <roy.pledge@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: use skb_to_full_sk() in skb_update_prio()
Eric Dumazet [Wed, 14 Mar 2018 16:04:16 +0000 (09:04 -0700)]
net: use skb_to_full_sk() in skb_update_prio()

Andrei Vagin reported a KASAN: slab-out-of-bounds error in
skb_update_prio()

Since SYNACK might be attached to a request socket, we need to
get back to the listener socket.
Since this listener is manipulated without locks, add const
qualifiers to sock_cgroup_prioidx() so that the const can also
be used in skb_update_prio()

Also add the const qualifier to sock_cgroup_classid() for consistency.

Fixes: ca6fb0651883 ("tcp: attach SYNACK messages to request sockets instead of listener")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocan: cc770: Fix queue stall & dropped RTR reply
Andri Yngvason [Wed, 14 Mar 2018 11:52:57 +0000 (11:52 +0000)]
can: cc770: Fix queue stall & dropped RTR reply

While waiting for the TX object to send an RTR, an external message with a
matching id can overwrite the TX data. In this case we must call the rx
routine and then try transmitting the message that was overwritten again.

The queue was being stalled because the RX event did not generate an
interrupt to wake up the queue again and the TX event did not happen
because the TXRQST flag is reset by the chip when new data is received.

According to the CC770 datasheet the id of a message object should not be
changed while the MSGVAL bit is set. This has been fixed by resetting the
MSGVAL bit before modifying the object in the transmit function and setting
it after. It is not enough to set & reset CPUUPD.

It is important to keep the MSGVAL bit reset while the message object is
being modified. Otherwise, during RTR transmission, a frame with matching
id could trigger an rx-interrupt, which would cause a race condition
between the interrupt routine and the transmit function.

Signed-off-by: Andri Yngvason <andri.yngvason@marel.com>
Tested-by: Richard Weinberger <richard@nod.at>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
6 years agocan: cc770: Fix stalls on rt-linux, remove redundant IRQ ack
Andri Yngvason [Wed, 14 Mar 2018 11:52:56 +0000 (11:52 +0000)]
can: cc770: Fix stalls on rt-linux, remove redundant IRQ ack

This has been reported to cause stalls on rt-linux.

Suggested-by: Richard Weinberger <richard@nod.at>
Tested-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Andri Yngvason <andri.yngvason@marel.com>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
6 years agoqed: Use after free in qed_rdma_free()
Dan Carpenter [Tue, 13 Mar 2018 09:09:38 +0000 (12:09 +0300)]
qed: Use after free in qed_rdma_free()

We're dereferencing "p_hwfn->p_rdma_info" but that is freed on the line
before in qed_rdma_resc_free(p_hwfn).

Fixes: 9de506a547c0 ("qed: Free RoCE ILT Memory on rmmod qedr")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec
David S. Miller [Tue, 13 Mar 2018 14:38:07 +0000 (10:38 -0400)]
Merge branch 'master' of git://git./linux/kernel/git/klassert/ipsec

Steffen Klassert says:

====================
pull request (net): ipsec 2018-03-13

1) Refuse to insert 32 bit userspace socket policies on 64
   bit systems like we do it for standard policies. We don't
   have a compat layer, so inserting socket policies from
   32 bit userspace will lead to a broken configuration.

2) Make the policy hold queue work without the flowcache.
   Dummy bundles are not chached anymore, so we need to
   generate a new one on each lookup as long as the SAs
   are not yet in place.

3) Fix the validation of the esn replay attribute. The
   The sanity check in verify_replay() is bypassed if
   the XFRM_STATE_ESN flag is not set. Fix this by doing
   the sanity check uncoditionally.
   From Florian Westphal.

4) After most of the dst_entry garbage collection code
   is removed, we may leak xfrm_dst entries as they are
   neither cached nor tracked somewhere. Fix this by
   reusing the 'uncached_list' to track xfrm_dst entries
   too. From Xin Long.

5) Fix a rcu_read_lock/rcu_read_unlock imbalance in
   xfrm_get_tos() From Xin Long.

6) Fix an infinite loop in xfrm_get_dst_nexthop. On
   transport mode we fetch the child dst_entry after
   we continue, so this pointer is never updated.
   Fix this by fetching it before we continue.

7) Fix ESN sequence number gap after IPsec GSO packets.
    We accidentally increment the sequence number counter
    on the xfrm_state by one packet too much in the ESN
    case. Fix this by setting the sequence number to the
    correct value.

8) Reset the ethernet protocol after decapsulation only if a
   mac header was set. Otherwise it breaks configurations
   with TUN devices. From Yossi Kuperman.

9) Fix __this_cpu_read() usage in preemptible code. Use
   this_cpu_read() instead in ipcomp_alloc_tfms().
   From Greg Hackmann.

Please pull or let me know if there are problems.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: xfrm: use preempt-safe this_cpu_read() in ipcomp_alloc_tfms()
Greg Hackmann [Wed, 7 Mar 2018 22:42:53 +0000 (14:42 -0800)]
net: xfrm: use preempt-safe this_cpu_read() in ipcomp_alloc_tfms()

f7c83bcbfaf5 ("net: xfrm: use __this_cpu_read per-cpu helper") added a
__this_cpu_read() call inside ipcomp_alloc_tfms().

At the time, __this_cpu_read() required the caller to either not care
about races or to handle preemption/interrupt issues.  3.15 tightened
the rules around some per-cpu operations, and now __this_cpu_read()
should never be used in a preemptible context.  On 3.15 and later, we
need to use this_cpu_read() instead.

syzkaller reported this leading to the following kernel BUG while
fuzzing sendmsg:

BUG: using __this_cpu_read() in preemptible [00000000] code: repro/3101
caller is ipcomp_init_state+0x185/0x990
CPU: 3 PID: 3101 Comm: repro Not tainted 4.16.0-rc4-00123-g86f84779d8e9 #154
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1 04/01/2014
Call Trace:
 dump_stack+0xb9/0x115
 check_preemption_disabled+0x1cb/0x1f0
 ipcomp_init_state+0x185/0x990
 ? __xfrm_init_state+0x876/0xc20
 ? lock_downgrade+0x5e0/0x5e0
 ipcomp4_init_state+0xaa/0x7c0
 __xfrm_init_state+0x3eb/0xc20
 xfrm_init_state+0x19/0x60
 pfkey_add+0x20df/0x36f0
 ? pfkey_broadcast+0x3dd/0x600
 ? pfkey_sock_destruct+0x340/0x340
 ? pfkey_seq_stop+0x80/0x80
 ? __skb_clone+0x236/0x750
 ? kmem_cache_alloc+0x1f6/0x260
 ? pfkey_sock_destruct+0x340/0x340
 ? pfkey_process+0x62a/0x6f0
 pfkey_process+0x62a/0x6f0
 ? pfkey_send_new_mapping+0x11c0/0x11c0
 ? mutex_lock_io_nested+0x1390/0x1390
 pfkey_sendmsg+0x383/0x750
 ? dump_sp+0x430/0x430
 sock_sendmsg+0xc0/0x100
 ___sys_sendmsg+0x6c8/0x8b0
 ? copy_msghdr_from_user+0x3b0/0x3b0
 ? pagevec_lru_move_fn+0x144/0x1f0
 ? find_held_lock+0x32/0x1c0
 ? do_huge_pmd_anonymous_page+0xc43/0x11e0
 ? lock_downgrade+0x5e0/0x5e0
 ? get_kernel_page+0xb0/0xb0
 ? _raw_spin_unlock+0x29/0x40
 ? do_huge_pmd_anonymous_page+0x400/0x11e0
 ? __handle_mm_fault+0x553/0x2460
 ? __fget_light+0x163/0x1f0
 ? __sys_sendmsg+0xc7/0x170
 __sys_sendmsg+0xc7/0x170
 ? SyS_shutdown+0x1a0/0x1a0
 ? __do_page_fault+0x5a0/0xca0
 ? lock_downgrade+0x5e0/0x5e0
 SyS_sendmsg+0x27/0x40
 ? __sys_sendmsg+0x170/0x170
 do_syscall_64+0x19f/0x640
 entry_SYSCALL_64_after_hwframe+0x42/0xb7
RIP: 0033:0x7f0ee73dfb79
RSP: 002b:00007ffe14fc15a8 EFLAGS: 00000207 ORIG_RAX: 000000000000002e
RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0ee73dfb79
RDX: 0000000000000000 RSI: 00000000208befc8 RDI: 0000000000000004
RBP: 00007ffe14fc15b0 R08: 00007ffe14fc15c0 R09: 00007ffe14fc15c0
R10: 0000000000000000 R11: 0000000000000207 R12: 0000000000400440
R13: 00007ffe14fc16b0 R14: 0000000000000000 R15: 0000000000000000

Signed-off-by: Greg Hackmann <ghackmann@google.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
6 years agonet: dsa: Fix dsa_is_user_port() test inversion
Florian Fainelli [Mon, 12 Mar 2018 23:00:40 +0000 (16:00 -0700)]
net: dsa: Fix dsa_is_user_port() test inversion

During the conversion to dsa_is_user_port(), a condition ended up being
reversed, which would prevent the creation of any user port when using
the legacy binding and/or platform data, fix that.

Fixes: 4a5b85ffe2a0 ("net: dsa: use dsa_is_user_port everywhere")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net...
David S. Miller [Mon, 12 Mar 2018 19:21:08 +0000 (15:21 -0400)]
Merge branch '1GbE' of git://git./linux/kernel/git/jkirsher/net-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2018-03-12

This series contains fixes to e1000e only.

Benjamin Poirier provides two fixes, first reverts commits that changed
what happens to the link status when there is an error.  These commits
were to resolve a race condition, but in the process of fixing the race
condition, they changed the behavior when an error occurred.  Second fix
resolves a race condition by not setting "get_link_status" to false
after checking the link.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'l2tp-fix-races-with-ipv4-mapped-ipv6-addresses'
David S. Miller [Mon, 12 Mar 2018 19:11:20 +0000 (15:11 -0400)]
Merge branch 'l2tp-fix-races-with-ipv4-mapped-ipv6-addresses'

Paolo Abeni says:

====================
l2tp: fix races with ipv4-mapped ipv6 addresses

The syzbot reported an l2tp oops that uncovered some races in the l2tp xmit
path and a partially related issue in the generic ipv6 code.

We need to address them separately.

v1 -> v2:
 - add missing fixes tag in patch 1
 - fix several issues in patch 2

v2 -> v3:
 - dropped some unneeded chunks in patch 2
====================

Reviewed-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agol2tp: fix races with ipv4-mapped ipv6 addresses
Paolo Abeni [Mon, 12 Mar 2018 13:54:24 +0000 (14:54 +0100)]
l2tp: fix races with ipv4-mapped ipv6 addresses

The l2tp_tunnel_create() function checks for v4mapped ipv6
sockets and cache that flag, so that l2tp core code can
reusing it at xmit time.

If the socket is provided by the userspace, the connection
status of the tunnel sockets can change between the tunnel
creation and the xmit call, so that syzbot is able to
trigger the following splat:

BUG: KASAN: use-after-free in ip6_dst_idev include/net/ip6_fib.h:192
[inline]
BUG: KASAN: use-after-free in ip6_xmit+0x1f76/0x2260
net/ipv6/ip6_output.c:264
Read of size 8 at addr ffff8801bd949318 by task syz-executor4/23448

CPU: 0 PID: 23448 Comm: syz-executor4 Not tainted 4.16.0-rc4+ #65
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS
Google 01/01/2011
Call Trace:
  __dump_stack lib/dump_stack.c:17 [inline]
  dump_stack+0x194/0x24d lib/dump_stack.c:53
  print_address_description+0x73/0x250 mm/kasan/report.c:256
  kasan_report_error mm/kasan/report.c:354 [inline]
  kasan_report+0x23c/0x360 mm/kasan/report.c:412
  __asan_report_load8_noabort+0x14/0x20 mm/kasan/report.c:433
  ip6_dst_idev include/net/ip6_fib.h:192 [inline]
  ip6_xmit+0x1f76/0x2260 net/ipv6/ip6_output.c:264
  inet6_csk_xmit+0x2fc/0x580 net/ipv6/inet6_connection_sock.c:139
  l2tp_xmit_core net/l2tp/l2tp_core.c:1053 [inline]
  l2tp_xmit_skb+0x105f/0x1410 net/l2tp/l2tp_core.c:1148
  pppol2tp_sendmsg+0x470/0x670 net/l2tp/l2tp_ppp.c:341
  sock_sendmsg_nosec net/socket.c:630 [inline]
  sock_sendmsg+0xca/0x110 net/socket.c:640
  ___sys_sendmsg+0x767/0x8b0 net/socket.c:2046
  __sys_sendmsg+0xe5/0x210 net/socket.c:2080
  SYSC_sendmsg net/socket.c:2091 [inline]
  SyS_sendmsg+0x2d/0x50 net/socket.c:2087
  do_syscall_64+0x281/0x940 arch/x86/entry/common.c:287
  entry_SYSCALL_64_after_hwframe+0x42/0xb7
RIP: 0033:0x453e69
RSP: 002b:00007f819593cc68 EFLAGS: 00000246 ORIG_RAX: 000000000000002e
RAX: ffffffffffffffda RBX: 00007f819593d6d4 RCX: 0000000000453e69
RDX: 0000000000000081 RSI: 000000002037ffc8 RDI: 0000000000000004
RBP: 000000000072bea0 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff
R13: 00000000000004c3 R14: 00000000006f72e8 R15: 0000000000000000

This change addresses the issues:
* explicitly checking for TCP_ESTABLISHED for user space provided sockets
* dropping the v4mapped flag usage - it can become outdated - and
  explicitly invoking ipv6_addr_v4mapped() instead

The issue is apparently there since ancient times.

v1 -> v2: (many thanks to Guillaume)
 - with csum issue introduced in v1
 - replace pr_err with pr_debug
 - fix build issue with IPV6 disabled
 - move l2tp_sk_is_v4mapped in l2tp_core.c

v2 -> v3:
 - don't update inet_daddr for v4mapped address, unneeded
 - drop rendundant check at creation time

Reported-and-tested-by: syzbot+92fa328176eb07e4ac1a@syzkaller.appspotmail.com
Fixes: 3557baabf280 ("[L2TP]: PPP over L2TP driver core")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ipv6: keep sk status consistent after datagram connect failure
Paolo Abeni [Mon, 12 Mar 2018 13:54:23 +0000 (14:54 +0100)]
net: ipv6: keep sk status consistent after datagram connect failure

On unsuccesful ip6_datagram_connect(), if the failure is caused by
ip6_datagram_dst_update(), the sk peer information are cleared, but
the sk->sk_state is preserved.

If the socket was already in an established status, the overall sk
status is inconsistent and fouls later checks in datagram code.

Fix this saving the old peer information and restoring them in
case of failure. This also aligns ipv6 datagram connect() behavior
with ipv4.

v1 -> v2:
 - added missing Fixes tag

Fixes: 85cb73ff9b74 ("net: ipv6: reset daddr and dport in sk if connect() fails")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoe1000e: Fix link check race condition
Benjamin Poirier [Tue, 6 Mar 2018 01:55:53 +0000 (10:55 +0900)]
e1000e: Fix link check race condition

Alex reported the following race condition:

/* link goes up... interrupt... schedule watchdog */
\ e1000_watchdog_task
\ e1000e_has_link
\ hw->mac.ops.check_for_link() === e1000e_check_for_copper_link
\ e1000e_phy_has_link_generic(..., &link)
link = true

 /* link goes down... interrupt */
 \ e1000_msix_other
 hw->mac.get_link_status = true

/* link is up */
mac->get_link_status = false

link_active = true
/* link_active is true, wrongly, and stays so because
 * get_link_status is false */

Avoid this problem by making sure that we don't set get_link_status = false
after having checked the link.

It seems this problem has been present since the introduction of e1000e.

Link: https://lkml.org/lkml/2018/1/29/338
Reported-by: Alexander Duyck <alexander.duyck@gmail.com>
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Acked-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoRevert "e1000e: Separate signaling for link check/link up"
Benjamin Poirier [Tue, 6 Mar 2018 01:55:52 +0000 (10:55 +0900)]
Revert "e1000e: Separate signaling for link check/link up"

This reverts commit 19110cfbb34d4af0cdfe14cd243f3b09dc95b013.
This reverts commit 4110e02eb45ea447ec6f5459c9934de0a273fb91.
This reverts commit d3604515c9eda464a92e8e67aae82dfe07fe3c98.

Commit 19110cfbb34d ("e1000e: Separate signaling for link check/link up")
changed what happens to the link status when there is an error which
happens after "get_link_status = false" in the copper check_for_link
callbacks. Previously, such an error would be ignored and the link
considered up. After that commit, any error implies that the link is down.

Revert commit 19110cfbb34d ("e1000e: Separate signaling for link check/link
up") and its followups. After reverting, the race condition described in
the log of commit 19110cfbb34d is reintroduced. It may still be triggered
by LSC events but this should keep the link down in case the link is
electrically unstable, as discussed. The race may no longer be
triggered by RXO events because commit 4aea7a5c5e94 ("e1000e: Avoid
receiver overrun interrupt bursts") restored reading icr in the Other
handler.

Link: https://lkml.org/lkml/2018/3/1/789
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Acked-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf
David S. Miller [Mon, 12 Mar 2018 16:49:30 +0000 (12:49 -0400)]
Merge git://git./pub/scm/linux/kernel/git/pablo/nf

Pablo Neira Ayuso says:

====================
Netfilter fixes for net

The following patchset contains Netfilter fixes for your net tree, they are:

1) Fixed hashtable representation doesn't support timeout flag, skip it
   otherwise rules to add elements from the packet fail bogusly fail with
   EOPNOTSUPP.

2) Fix bogus error with 32-bits ebtables userspace and 64-bits kernel,
   patch from Florian Westphal.

3) Sanitize proc names in several x_tables extensions, also from Florian.

4) Add sanitization to ebt_among wormhash logic, from Florian.

5) Missing release of hook array in flowtable.
====================

6 years agoMerge tag 'linux-can-fixes-for-4.16-20180312' of ssh://gitolite.kernel.org/pub/scm...
David S. Miller [Mon, 12 Mar 2018 15:16:40 +0000 (11:16 -0400)]
Merge tag 'linux-can-fixes-for-4.16-20180312' of ssh://gitolite./linux/kernel/git/mkl/linux-can

Marc Kleine-Budde says:

====================
pull-request: can 2018-03-12

this is a pull reqeust of 6 patches for net/master.

The first patch is by Wolfram Sang and fixes a bitshift vs. comparison mistake
in the m_can driver. Two patches of Marek Vasut repair the error handling in
the ifi driver. The two patches by Stephane Grosjean fix a "echo_skb is
occupied!" bug in the peak/pcie_fd driver. Bich HEMON's patch adds pinctrl
select state calls to the m_can's driver to further improve power saving during
suspend.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosock_diag: request _diag module only when the family or proto has been registered
Xin Long [Sat, 10 Mar 2018 10:57:50 +0000 (18:57 +0800)]
sock_diag: request _diag module only when the family or proto has been registered

Now when using 'ss' in iproute, kernel would try to load all _diag
modules, which also causes corresponding family and proto modules
to be loaded as well due to module dependencies.

Like after running 'ss', sctp, dccp, af_packet (if it works as a module)
would be loaded.

For example:

  $ lsmod|grep sctp
  $ ss
  $ lsmod|grep sctp
  sctp_diag              16384  0
  sctp                  323584  5 sctp_diag
  inet_diag              24576  4 raw_diag,tcp_diag,sctp_diag,udp_diag
  libcrc32c              16384  3 nf_conntrack,nf_nat,sctp

As these family and proto modules are loaded unintentionally, it
could cause some problems, like:

- Some debug tools use 'ss' to collect the socket info, which loads all
  those diag and family and protocol modules. It's noisy for identifying
  issues.

- Users usually expect to drop sctp init packet silently when they
  have no sense of sctp protocol instead of sending abort back.

- It wastes resources (especially with multiple netns), and SCTP module
  can't be unloaded once it's loaded.

...

In short, it's really inappropriate to have these family and proto
modules loaded unexpectedly when just doing debugging with inet_diag.

This patch is to introduce sock_load_diag_module() where it loads
the _diag module only when it's corresponding family or proto has
been already registered.

Note that we can't just load _diag module without the family or
proto loaded, as some symbols used in _diag module are from the
family or proto module.

v1->v2:
  - move inet proto check to inet_diag to avoid a compiling err.
v2->v3:
  - define sock_load_diag_module in sock.c and export one symbol
    only.
  - improve the changelog.

Reported-by: Sabrina Dubroca <sd@queasysnail.net>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Acked-by: Phil Sutter <phil@nwl.cc>
Acked-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'bnxt_en-Bug-fixes'
David S. Miller [Mon, 12 Mar 2018 14:58:28 +0000 (10:58 -0400)]
Merge branch 'bnxt_en-Bug-fixes'

Michael Chan says:

====================
bnxt_en: Bug fixes.

There are 3 bug fixes in this series to fix regressions recently
introduced when adding the new ring reservations scheme.  2 minor
fixes in the TC Flower code to return standard errno values and
to elide some unnecessary warning dmesg.  One Fixes the VLAN TCI
value passed to the stack by including the entire 16-bit VLAN TCI,
and the last fix is to check for valid VNIC ID before setting up or
shutting down LRO/GRO.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobnxt_en: Check valid VNIC ID in bnxt_hwrm_vnic_set_tpa().
Michael Chan [Sat, 10 Mar 2018 04:46:10 +0000 (23:46 -0500)]
bnxt_en: Check valid VNIC ID in bnxt_hwrm_vnic_set_tpa().

During initialization, if we encounter errors, there is a code path that
calls bnxt_hwrm_vnic_set_tpa() with invalid VNIC ID.  This may cause a
warning in firmware logs.

Fixes: c0c050c58d84 ("bnxt_en: New Broadcom ethernet driver.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobnxt_en: close & open NIC, only when the interface is in running state.
Venkat Duvvuru [Sat, 10 Mar 2018 04:46:09 +0000 (23:46 -0500)]
bnxt_en: close & open NIC, only when the interface is in running state.

bnxt_restore_pf_fw_resources routine frees PF resources by calling
close_nic and allocates the resources back, by doing open_nic. However,
this is not needed, if the PF is already in closed state.

This bug causes the driver to call open the device and call request_irq()
when it is not needed.  Ultimately, pci_disable_msix() will crash
when bnxt_en is unloaded.

This patch fixes the problem by skipping __bnxt_close_nic and
__bnxt_open_nic inside bnxt_restore_pf_fw_resources routine, if the
interface is not running.

Fixes: 80fcaf46c092 ("bnxt_en: Restore MSIX after disabling SRIOV.")
Signed-off-by: Venkat Duvvuru <venkatkumar.duvvuru@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobnxt_en: Return standard Linux error codes for hwrm flow cmds.
Venkat Duvvuru [Sat, 10 Mar 2018 04:46:08 +0000 (23:46 -0500)]
bnxt_en: Return standard Linux error codes for hwrm flow cmds.

Currently, internal error value is returned by the driver, when
hwrm_cfa_flow_alloc() fails due lack of resources.  We should be returning
Linux errno value -ENOSPC instead.

This patch also converts other similar command errors to standard Linux errno
code (-EIO) in bnxt_tc.c

Fixes: db1d36a27324 ("bnxt_en: add TC flower offload flow_alloc/free FW cmds")
Signed-off-by: Venkat Duvvuru <venkatkumar.duvvuru@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobnxt_en: Fix regressions when setting up MQPRIO TX rings.
Michael Chan [Sat, 10 Mar 2018 04:46:07 +0000 (23:46 -0500)]
bnxt_en: Fix regressions when setting up MQPRIO TX rings.

Recent changes added the bnxt_init_int_mode() call in the driver's open
path whenever ring reservations are changed.  This call was previously
only called in the probe path.  In the open path, if MQPRIO TC has been
setup, the bnxt_init_int_mode() call would reset and mess up the MQPRIO
per TC rings.

Fix it by not re-initilizing bp->tx_nr_rings_per_tc in
bnxt_init_int_mode().  Instead, initialize it in the probe path only
after the bnxt_init_int_mode() call.

Fixes: 674f50a5b026 ("bnxt_en: Implement new method to reserve rings.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobnxt_en: Pass complete VLAN TCI to the stack.
Michael Chan [Sat, 10 Mar 2018 04:46:06 +0000 (23:46 -0500)]
bnxt_en: Pass complete VLAN TCI to the stack.

When receiving a packet with VLAN tag, pass the entire 16-bit TCI to the
stack when calling __vlan_hwaccel_put_tag().  The current code is only
passing the 12-bit tag and it is missing the priority bits.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobnxt_en: Remove unwanted ovs-offload messages in some conditions
Sriharsha Basavapatna [Sat, 10 Mar 2018 04:46:05 +0000 (23:46 -0500)]
bnxt_en: Remove unwanted ovs-offload messages in some conditions

In some conditions when the driver fails to add a flow in HW and returns
an error back to the stack, the stack continues to invoke get_flow_stats()
and/or del_flow() on it. The driver fails these APIs with an error message
"no flow_node for cookie". The message gets logged repeatedly as long as
the stack keeps invoking these functions.

Fix this by removing the corresponding netdev_info() calls from these
functions.

Fixes: d7bc73053024 ("bnxt_en: add code to query TC flower offload stats")
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobnxt_en: Fix vnic accounting in the bnxt_check_rings() path.
Eddie Wai [Sat, 10 Mar 2018 04:46:04 +0000 (23:46 -0500)]
bnxt_en: Fix vnic accounting in the bnxt_check_rings() path.

The number of vnics to check must be determined ahead of time because
only standard RX rings require vnics to support RFS.  The logic is
similar to the ring reservation logic and we can now use the
refactored common functions to do most of the work in setting up
the firmware message.

Fixes: 8f23d638b36b ("bnxt_en: Expand bnxt_check_rings() to check all resources.")
Signed-off-by: Eddie Wai <eddie.wai@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobnxt_en: Refactor the functions to reserve hardware rings.
Michael Chan [Sat, 10 Mar 2018 04:46:03 +0000 (23:46 -0500)]
bnxt_en: Refactor the functions to reserve hardware rings.

The bnxt_hwrm_reserve_{pf|vf}_rings() functions are very similar to
the bnxt_hwrm_check_{pf|vf}_rings() functions.  Refactor the former
so that the latter can make use of common code in the next patch.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: Tell caller result of phy_change()
Brad Mouring [Thu, 8 Mar 2018 22:23:03 +0000 (16:23 -0600)]
net: phy: Tell caller result of phy_change()

In 664fcf123a30e (net: phy: Threaded interrupts allow some simplification)
the phy_interrupt system was changed to use a traditional threaded
interrupt scheme instead of a workqueue approach.

With this change, the phy status check moved into phy_change, which
did not report back to the caller whether or not the interrupt was
handled. This means that, in the case of a shared phy interrupt,
only the first phydev's interrupt registers are checked (since
phy_interrupt() would always return IRQ_HANDLED). This leads to
interrupt storms when it is a secondary device that's actually the
interrupt source.

Signed-off-by: Brad Mouring <brad.mouring@ni.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocan: m_can: select pinctrl state in each suspend/resume function
Bich HEMON [Mon, 12 Mar 2018 08:52:37 +0000 (08:52 +0000)]
can: m_can: select pinctrl state in each suspend/resume function

Make sure to apply the correct pin state in suspend/resume callbacks.
Putting pins in sleep state saves power.

Signed-off-by: Bich Hemon <bich.hemon@st.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
6 years agocan: peak/pcie_fd: remove useless code when interface starts
Stephane Grosjean [Thu, 8 Mar 2018 08:30:29 +0000 (09:30 +0100)]
can: peak/pcie_fd: remove useless code when interface starts

When an interface starts, the echo_skb array is empty and the network
queue should be started only. This patch replaces useless code and locks
when the internal RX_BARRIER message is received from the IP core, telling
the driver that tx may start.

Signed-off-by: Stephane Grosjean <s.grosjean@peak-system.com>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
6 years agocan: peak/pcie_fd: fix echo_skb is occupied! bug
Stephane Grosjean [Thu, 8 Mar 2018 08:30:28 +0000 (09:30 +0100)]
can: peak/pcie_fd: fix echo_skb is occupied! bug

This patch makes atomic the handling of the linux-can echo_skb array and
the network tx queue. This prevents from the "BUG! echo_skb is occupied!"
message to be printed by the linux-can core, in SMP environments.

Reported-by: Diana Burgess <diana@peloton-tech.com>
Signed-off-by: Stephane Grosjean <s.grosjean@peak-system.com>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
6 years agocan: ifi: Repair the error handling
Marek Vasut [Thu, 1 Mar 2018 18:34:00 +0000 (19:34 +0100)]
can: ifi: Repair the error handling

The new version of the IFI CANFD core has significantly less complex
error state indication logic. In particular, the warning/error state
bits are no longer all over the place, but are all present in the
STATUS register. Moreover, there is a new IRQ register bit indicating
transition between error states (active/warning/passive/busoff).

This patch makes use of this bit to weed out the obscure selective
INTERRUPT register clearing, which was used to carry over the error
state indication into the poll function. While at it, this patch
fixes the handling of the ACTIVE state, since the hardware provides
indication of the core being in ACTIVE state and that in turn fixes
the state transition indication toward userspace. Finally, register
reads in the poll function are moved to the matching subfunctions
since those are also no longer needed in the poll function.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Heiko Schocher <hs@denx.de>
Cc: Markus Marb <markus@marb.org>
Cc: Marc Kleine-Budde <mkl@pengutronix.de>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
6 years agocan: ifi: Check core revision upon probe
Marek Vasut [Mon, 5 Mar 2018 20:29:52 +0000 (21:29 +0100)]
can: ifi: Check core revision upon probe

Older versions of the core are not compatible with the driver due
to various intrusive fixes of the core. Read out the VER register,
check the core revision bitfield and verify if the core in use is
new enough (rev 2.1 or newer) to work correctly with this driver.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Heiko Schocher <hs@denx.de>
Cc: Markus Marb <markus@marb.org>
Cc: Marc Kleine-Budde <mkl@pengutronix.de>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
6 years agocan: m_can: change comparison to bitshift when dealing with a mask
Wolfram Sang [Tue, 6 Feb 2018 08:52:07 +0000 (09:52 +0100)]
can: m_can: change comparison to bitshift when dealing with a mask

Due to a typo, the mask was destroyed by a comparison instead of a bit
shift.

Reported-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
6 years agoopenvswitch: meter: fix the incorrect calculation of max delta_t
zhangliping [Fri, 9 Mar 2018 02:08:50 +0000 (10:08 +0800)]
openvswitch: meter: fix the incorrect calculation of max delta_t

Max delat_t should be the full_bucket/rate instead of the full_bucket.
Also report EINVAL if the rate is zero.

Fixes: 96fbc13d7e77 ("openvswitch: Add meter infrastructure")
Cc: Andy Zhou <azhou@ovn.org>
Signed-off-by: zhangliping <zhangliping02@baidu.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomacvlan: filter out unsupported feature flags
Shannon Nelson [Fri, 9 Mar 2018 00:17:23 +0000 (16:17 -0800)]
macvlan: filter out unsupported feature flags

Adding a macvlan device on top of a lowerdev that supports
the xfrm offloads fails with a new regression:
  # ip link add link ens1f0 mv0 type macvlan
  RTNETLINK answers: Operation not permitted

Tracing down the failure shows that the macvlan device inherits
the NETIF_F_HW_ESP and NETIF_F_HW_ESP_TX_CSUM feature flags
from the lowerdev, but with no dev->xfrmdev_ops API filled
in, it doesn't actually support xfrm.  When the request is
made to add the new macvlan device, the XFRM listener for
NETDEV_REGISTER calls xfrm_api_check() which fails the new
registration because dev->xfrmdev_ops is NULL.

The macvlan creation succeeds when we filter out the ESP
feature flags in macvlan_fix_features(), so let's filter them
out like we're already filtering out ~NETIF_F_NETNS_LOCAL.
When XFRM support is added in the future, we can add the flags
into MACVLAN_FEATURES.

This same problem could crop up in the future with any other
new feature flags, so let's filter out any flags that aren't
defined as supported in macvlan.

Fixes: d77e38e612a0 ("xfrm: Add an IPsec hardware offloading API")
Reported-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Signed-off-by: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonetfilter: nf_tables: release flowtable hooks
Pablo Neira Ayuso [Fri, 9 Mar 2018 17:56:26 +0000 (18:56 +0100)]
netfilter: nf_tables: release flowtable hooks

Otherwise we leak this array.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: bridge: ebt_among: add more missing match size checks
Florian Westphal [Fri, 9 Mar 2018 13:27:31 +0000 (14:27 +0100)]
netfilter: bridge: ebt_among: add more missing match size checks

ebt_among is special, it has a dynamic match size and is exempt
from the central size checks.

commit c4585a2823edf ("bridge: ebt_among: add missing match size checks")
added validation for pool size, but missed fact that the macros
ebt_among_wh_src/dst can already return out-of-bound result because
they do not check value of wh_src/dst_ofs (an offset) vs. the size
of the match that userspace gave to us.

v2:
check that offset has correct alignment.
Paolo Abeni points out that we should also check that src/dst
wormhash arrays do not overlap, and src + length lines up with
start of dst (or vice versa).
v3: compact wormhash_sizes_valid() part

NB: Fixes tag is intentionally wrong, this bug exists from day
one when match was added for 2.6 kernel. Tag is there so stable
maintainers will notice this one too.

Tested with same rules from the earlier patch.

Fixes: c4585a2823edf ("bridge: ebt_among: add missing match size checks")
Reported-by: <syzbot+bdabab6f1983a03fc009@syzkaller.appspotmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: x_tables: add and use xt_check_proc_name
Florian Westphal [Sat, 10 Mar 2018 00:15:45 +0000 (01:15 +0100)]
netfilter: x_tables: add and use xt_check_proc_name

recent and hashlimit both create /proc files, but only check that
name is 0 terminated.

This can trigger WARN() from procfs when name is "" or "/".
Add helper for this and then use it for both.

Cc: Eric Dumazet <eric.dumazet@gmail.com>
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Reported-by: <syzbot+0502b00edac2a0680b61@syzkaller.appspotmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agonetfilter: ebtables: fix erroneous reject of last rule
Florian Westphal [Thu, 8 Mar 2018 11:54:19 +0000 (12:54 +0100)]
netfilter: ebtables: fix erroneous reject of last rule

The last rule in the blob has next_entry offset that is same as total size.
This made "ebtables32 -A OUTPUT -d de:ad:be:ef:01:02" fail on 64 bit kernel.

Fixes: b71812168571fa ("netfilter: ebtables: CONFIG_COMPAT: don't trust userland offsets")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agoMerge branch 'erspan-fixes'
David S. Miller [Fri, 9 Mar 2018 18:03:57 +0000 (13:03 -0500)]
Merge branch 'erspan-fixes'

William Tu says:

====================
a couple of erspan fixes

The series fixes a couple of erspan issues.
The first patch adds the erspan v2 proto type to the ip6 tunnel lookup.
The second patch improves the error handling when users screws the
version number in metadata.  The final patch makes sure the skb has
enough headroom for pushing erspan header when xmit.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoip6erspan: make sure enough headroom at xmit.
William Tu [Fri, 9 Mar 2018 15:34:42 +0000 (07:34 -0800)]
ip6erspan: make sure enough headroom at xmit.

The patch adds skb_cow_header() to ensure enough headroom
at ip6erspan_tunnel_xmit before pushing the erspan header
to the skb.

Signed-off-by: William Tu <u9012063@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoip6erspan: improve error handling for erspan version number.
William Tu [Fri, 9 Mar 2018 15:34:41 +0000 (07:34 -0800)]
ip6erspan: improve error handling for erspan version number.

When users fill in incorrect erspan version number through
the struct erspan_metadata uapi, current code skips pushing
the erspan header but continue pushing the gre header, which
is incorrect.  The patch fixes it by returning error.

Signed-off-by: William Tu <u9012063@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoip6gre: add erspan v2 to tunnel lookup
William Tu [Fri, 9 Mar 2018 15:34:40 +0000 (07:34 -0800)]
ip6gre: add erspan v2 to tunnel lookup

The patch adds the erspan v2 proto in ip6gre_tunnel_lookup
so the erspan v2 tunnel can be found correctly.

Signed-off-by: William Tu <u9012063@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'mlxsw-ACL-and-mirroring-fixes'
David S. Miller [Fri, 9 Mar 2018 18:02:14 +0000 (13:02 -0500)]
Merge branch 'mlxsw-ACL-and-mirroring-fixes'

Ido Schimmel says:

====================
mlxsw: ACL and mirroring fixes

The first patch fixes offload of rules using the 'pass' action. Instead
of continuing to evaluate lower priority rules, the binding is
terminated and the packet proceeds to the bridge and router blocks on
ingress, or goes out of the port on egress.

Second patch prevents the user from mirroring more than once from a
given {Port, Direction} as this is not supported by the device.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum: Prevent duplicate mirrors
Petr Machata [Fri, 9 Mar 2018 13:33:53 +0000 (15:33 +0200)]
mlxsw: spectrum: Prevent duplicate mirrors

The Spectrum ASIC doesn't support mirroring more than once from a single
binding point (which is a port-direction pair). Therefore detect that a
second binding of a given binding point is attempted.

To that end, extend struct mlxsw_sp_span_inspected_port to track whether
a given binding point is bound or not. Extend
mlxsw_sp_span_entry_port_find() to look for ports based on the full
unique key: port number, direction, and boundness.

Besides fixing the overt bug where configured mirrors are not offloaded,
this also fixes a more subtle bug: mlxsw_sp_span_inspected_port_del()
just defers to mlxsw_sp_span_entry_bound_port_find(), and that used to
find the first port with the right number (disregarding the type). Thus
by adding and removing egress and ingress mirrors in the right order,
one could trick the system into believing it has no egress mirrors when
in fact it did have some. That then caused that
mlxsw_sp_span_port_mtu_update() didn't update mirroring buffer when MTU
was changed.

Fixes: 763b4b70afcd ("mlxsw: spectrum: Add support in matchall mirror TC offloading")
Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum: Fix gact_ok offloading
Jiri Pirko [Fri, 9 Mar 2018 13:33:52 +0000 (15:33 +0200)]
mlxsw: spectrum: Fix gact_ok offloading

For ok GACT action, TERMINATE binding_cmd should be used in action set
passed down to HW.

Fixes: b2925957ec1a9 ("mlxsw: spectrum_flower: Offload "ok" termination action")
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reported-by: Alexander Petrovskiy <alexpe@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'vhost_net-ptr_ring-fixes'
David S. Miller [Fri, 9 Mar 2018 17:03:00 +0000 (12:03 -0500)]
Merge branch 'vhost_net-ptr_ring-fixes'

Jason Wang says:

====================
Several fixes for vhost_net ptr_ring usage

This small series try to fix several bugs of ptr_ring usage in
vhost_net. Please review.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovhost_net: examine pointer types during un-producing
Jason Wang [Fri, 9 Mar 2018 06:50:34 +0000 (14:50 +0800)]
vhost_net: examine pointer types during un-producing

After commit fc72d1d54dd9 ("tuntap: XDP transmission"), we can
actually queueing XDP pointers in the pointer ring, so we should
examine the pointer type before freeing the pointer.

Fixes: fc72d1d54dd9 ("tuntap: XDP transmission")
Reported-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovhost_net: keep private_data and rx_ring synced
Jason Wang [Fri, 9 Mar 2018 06:50:33 +0000 (14:50 +0800)]
vhost_net: keep private_data and rx_ring synced

We get pointer ring from the exported sock, this means we should keep
rx_ring and vq->private synced during both vq stop and backend set,
otherwise we may see stale rx_ring.

Fixes: c67df11f6e480 ("vhost_net: try batch dequing from skb array")
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovhost_net: initialize rx_ring in vhost_net_open()
Alexander Potapenko [Fri, 9 Mar 2018 06:50:32 +0000 (14:50 +0800)]
vhost_net: initialize rx_ring in vhost_net_open()

KMSAN reported a use of uninit memory in vhost_net_buf_unproduce()
while trying to access n->vqs[VHOST_NET_VQ_TX].rx_ring:

==================================================================
BUG: KMSAN: use of uninitialized memory in vhost_net_buf_unproduce+0x7bb/0x9a0 drivers/vho
et.c:170
CPU: 0 PID: 3021 Comm: syz-fuzzer Not tainted 4.16.0-rc4+ #3853
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1 04/01/2014
Call Trace:
 __dump_stack lib/dump_stack.c:17 [inline]
 dump_stack+0x185/0x1d0 lib/dump_stack.c:53
 kmsan_report+0x142/0x1f0 mm/kmsan/kmsan.c:1093
 __msan_warning_32+0x6c/0xb0 mm/kmsan/kmsan_instr.c:676
 vhost_net_buf_unproduce+0x7bb/0x9a0 drivers/vhost/net.c:170
 vhost_net_stop_vq drivers/vhost/net.c:974 [inline]
 vhost_net_stop+0x146/0x380 drivers/vhost/net.c:982
 vhost_net_release+0xb1/0x4f0 drivers/vhost/net.c:1015
 __fput+0x49f/0xa00 fs/file_table.c:209
 ____fput+0x37/0x40 fs/file_table.c:243
 task_work_run+0x243/0x2c0 kernel/task_work.c:113
 tracehook_notify_resume include/linux/tracehook.h:191 [inline]
 exit_to_usermode_loop arch/x86/entry/common.c:166 [inline]
 prepare_exit_to_usermode+0x349/0x3b0 arch/x86/entry/common.c:196
 syscall_return_slowpath+0xf3/0x6d0 arch/x86/entry/common.c:265
 do_syscall_64+0x34d/0x450 arch/x86/entry/common.c:292
...
origin:
 kmsan_save_stack_with_flags mm/kmsan/kmsan.c:303 [inline]
 kmsan_internal_poison_shadow+0xb8/0x1b0 mm/kmsan/kmsan.c:213
 kmsan_kmalloc_large+0x6f/0xd0 mm/kmsan/kmsan.c:392
 kmalloc_large_node_hook mm/slub.c:1366 [inline]
 kmalloc_large_node mm/slub.c:3808 [inline]
 __kmalloc_node+0x100e/0x1290 mm/slub.c:3818
 kmalloc_node include/linux/slab.h:554 [inline]
 kvmalloc_node+0x1a5/0x2e0 mm/util.c:419
 kvmalloc include/linux/mm.h:541 [inline]
 vhost_net_open+0x64/0x5f0 drivers/vhost/net.c:921
 misc_open+0x7b5/0x8b0 drivers/char/misc.c:154
 chrdev_open+0xc28/0xd90 fs/char_dev.c:417
 do_dentry_open+0xccb/0x1430 fs/open.c:752
 vfs_open+0x272/0x2e0 fs/open.c:866
 do_last fs/namei.c:3378 [inline]
 path_openat+0x49ad/0x6580 fs/namei.c:3519
 do_filp_open+0x267/0x640 fs/namei.c:3553
 do_sys_open+0x6ad/0x9c0 fs/open.c:1059
 SYSC_openat+0xc7/0xe0 fs/open.c:1086
 SyS_openat+0x63/0x90 fs/open.c:1080
 do_syscall_64+0x2f1/0x450 arch/x86/entry/common.c:287
==================================================================

Fixes: c67df11f6e480 ("vhost_net: try batch dequing from skb array")
Signed-off-by: Alexander Potapenko <glider@google.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethernet: ave: enable Rx drop interrupt
Kunihiko Hayashi [Fri, 9 Mar 2018 06:27:20 +0000 (15:27 +0900)]
net: ethernet: ave: enable Rx drop interrupt

This enables AVE_GI_RXDROP interrupt factor. This factor indicates
depletion of Rx descriptors and the handler counts the number
of dropped packets.

Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: use skb_is_gso_sctp() instead of open-coding
Daniel Axtens [Fri, 9 Mar 2018 03:06:09 +0000 (14:06 +1100)]
net: use skb_is_gso_sctp() instead of open-coding

As well as the basic conversion, I noticed that a lot of the
SCTP code checks gso_type without first checking skb_is_gso()
so I have added that where appropriate.

Also, document the helper.

Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoieee802154: 6lowpan: fix possible NULL deref in lowpan_device_event()
Eric Dumazet [Mon, 5 Mar 2018 16:51:03 +0000 (08:51 -0800)]
ieee802154: 6lowpan: fix possible NULL deref in lowpan_device_event()

A tun device type can trivially be set to arbitrary value using
TUNSETLINK ioctl().

Therefore, lowpan_device_event() must really check that ieee802154_ptr
is not NULL.

Fixes: 2c88b5283f60d ("ieee802154: 6lowpan: remove check on null")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Alexander Aring <alex.aring@gmail.com>
Cc: Stefan Schmidt <stefan@osg.samsung.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Acked-by: Stefan Schmidt <stefan@osg.samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: fix access to non-linear packet in ndisc_fill_redirect_hdr_option()
Lorenzo Bianconi [Thu, 8 Mar 2018 16:00:02 +0000 (17:00 +0100)]
ipv6: fix access to non-linear packet in ndisc_fill_redirect_hdr_option()

Fix the following slab-out-of-bounds kasan report in
ndisc_fill_redirect_hdr_option when the incoming ipv6 packet is not
linear and the accessed data are not in the linear data region of orig_skb.

[ 1503.122508] ==================================================================
[ 1503.122832] BUG: KASAN: slab-out-of-bounds in ndisc_send_redirect+0x94e/0x990
[ 1503.123036] Read of size 1184 at addr ffff8800298ab6b0 by task netperf/1932

[ 1503.123220] CPU: 0 PID: 1932 Comm: netperf Not tainted 4.16.0-rc2+ #124
[ 1503.123347] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.10.2-2.fc27 04/01/2014
[ 1503.123527] Call Trace:
[ 1503.123579]  <IRQ>
[ 1503.123638]  print_address_description+0x6e/0x280
[ 1503.123849]  kasan_report+0x233/0x350
[ 1503.123946]  memcpy+0x1f/0x50
[ 1503.124037]  ndisc_send_redirect+0x94e/0x990
[ 1503.125150]  ip6_forward+0x1242/0x13b0
[...]
[ 1503.153890] Allocated by task 1932:
[ 1503.153982]  kasan_kmalloc+0x9f/0xd0
[ 1503.154074]  __kmalloc_track_caller+0xb5/0x160
[ 1503.154198]  __kmalloc_reserve.isra.41+0x24/0x70
[ 1503.154324]  __alloc_skb+0x130/0x3e0
[ 1503.154415]  sctp_packet_transmit+0x21a/0x1810
[ 1503.154533]  sctp_outq_flush+0xc14/0x1db0
[ 1503.154624]  sctp_do_sm+0x34e/0x2740
[ 1503.154715]  sctp_primitive_SEND+0x57/0x70
[ 1503.154807]  sctp_sendmsg+0xaa6/0x1b10
[ 1503.154897]  sock_sendmsg+0x68/0x80
[ 1503.154987]  ___sys_sendmsg+0x431/0x4b0
[ 1503.155078]  __sys_sendmsg+0xa4/0x130
[ 1503.155168]  do_syscall_64+0x171/0x3f0
[ 1503.155259]  entry_SYSCALL_64_after_hwframe+0x42/0xb7

[ 1503.155436] Freed by task 1932:
[ 1503.155527]  __kasan_slab_free+0x134/0x180
[ 1503.155618]  kfree+0xbc/0x180
[ 1503.155709]  skb_release_data+0x27f/0x2c0
[ 1503.155800]  consume_skb+0x94/0xe0
[ 1503.155889]  sctp_chunk_put+0x1aa/0x1f0
[ 1503.155979]  sctp_inq_pop+0x2f8/0x6e0
[ 1503.156070]  sctp_assoc_bh_rcv+0x6a/0x230
[ 1503.156164]  sctp_inq_push+0x117/0x150
[ 1503.156255]  sctp_backlog_rcv+0xdf/0x4a0
[ 1503.156346]  __release_sock+0x142/0x250
[ 1503.156436]  release_sock+0x80/0x180
[ 1503.156526]  sctp_sendmsg+0xbb0/0x1b10
[ 1503.156617]  sock_sendmsg+0x68/0x80
[ 1503.156708]  ___sys_sendmsg+0x431/0x4b0
[ 1503.156799]  __sys_sendmsg+0xa4/0x130
[ 1503.156889]  do_syscall_64+0x171/0x3f0
[ 1503.156980]  entry_SYSCALL_64_after_hwframe+0x42/0xb7

[ 1503.157158] The buggy address belongs to the object at ffff8800298ab600
                which belongs to the cache kmalloc-1024 of size 1024
[ 1503.157444] The buggy address is located 176 bytes inside of
                1024-byte region [ffff8800298ab600ffff8800298aba00)
[ 1503.157702] The buggy address belongs to the page:
[ 1503.157820] page:ffffea0000a62a00 count:1 mapcount:0 mapping:0000000000000000 index:0x0 compound_mapcount: 0
[ 1503.158053] flags: 0x4000000000008100(slab|head)
[ 1503.158171] raw: 4000000000008100 0000000000000000 0000000000000000 00000001800e000e
[ 1503.158350] raw: dead000000000100 dead000000000200 ffff880036002600 0000000000000000
[ 1503.158523] page dumped because: kasan: bad access detected

[ 1503.158698] Memory state around the buggy address:
[ 1503.158816]  ffff8800298ab900: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[ 1503.158988]  ffff8800298ab980: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[ 1503.159165] >ffff8800298aba00: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[ 1503.159338]                    ^
[ 1503.159436]  ffff8800298aba80: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[ 1503.159610]  ffff8800298abb00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[ 1503.159785] ==================================================================
[ 1503.159964] Disabling lock debugging due to kernel taint

The test scenario to trigger the issue consists of 4 devices:
- H0: data sender, connected to LAN0
- H1: data receiver, connected to LAN1
- GW0 and GW1: routers between LAN0 and LAN1. Both of them have an
  ethernet connection on LAN0 and LAN1
On H{0,1} set GW0 as default gateway while on GW0 set GW1 as next hop for
data from LAN0 to LAN1.
Moreover create an ip6ip6 tunnel between H0 and H1 and send 3 concurrent
data streams (TCP/UDP/SCTP) from H0 to H1 through ip6ip6 tunnel (send
buffer size is set to 16K). While data streams are active flush the route
cache on HA multiple times.
I have not been able to identify a given commit that introduced the issue
since, using the reproducer described above, the kasan report has been
triggered from 4.14 and I have not gone back further.

Reported-by: Jianlin Shi <jishi@redhat.com>
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'hv_netvsc-fix-multicast-flags-and-sync'
David S. Miller [Thu, 8 Mar 2018 17:48:57 +0000 (12:48 -0500)]
Merge branch 'hv_netvsc-fix-multicast-flags-and-sync'

Stephen Hemminger says:

====================
hv_netvsc: fix multicast flags and sync

This set of patches deals with the handling of multicast flags
and addresses in transparent VF mode. The recent set of patches
(in linux-net) had a couple of bugs.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agohv_netvsc: fix locking during VF setup
Stephen Hemminger [Wed, 7 Mar 2018 21:49:12 +0000 (13:49 -0800)]
hv_netvsc: fix locking during VF setup

The dev_uc/mc_sync calls need to have the device address list
locked. This was spotted by running with lockdep enabled.

Fixes: bee9d41b37ea ("hv_netvsc: propagate rx filters to VF")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agohv_netvsc: fix locking for rx_mode
Stephen Hemminger [Wed, 7 Mar 2018 21:49:11 +0000 (13:49 -0800)]
hv_netvsc: fix locking for rx_mode

The rx_mode operation handler is different than other callbacks
in that is not always called with rtnl held. Therefore use
RCU to ensure that references are valid.

Fixes: bee9d41b37ea ("hv_netvsc: propagate rx filters to VF")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agohv_netvsc: avoid repeated updates of packet filter
Stephen Hemminger [Wed, 7 Mar 2018 21:49:10 +0000 (13:49 -0800)]
hv_netvsc: avoid repeated updates of packet filter

The netvsc driver can get repeated calls to netvsc_rx_mode during
network setup; each of these calls ends up scheduling the lower
layers to update tha packet filter. This update requires an
request/response to the host. So avoid doing this if we already
know that the correct packet filter value is set.

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agohv_netvsc: fix filter flags
Stephen Hemminger [Wed, 7 Mar 2018 21:49:09 +0000 (13:49 -0800)]
hv_netvsc: fix filter flags

The recent change to not always enable all multicast and broadcast
was broken; meant to set filter, not change flags.

Fixes: 009f766ca238 ("hv_netvsc: filter multicast/broadcast")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'wireless-drivers-for-davem-2018-03-08' of git://git.kernel.org/pub/scm...
David S. Miller [Thu, 8 Mar 2018 16:44:04 +0000 (11:44 -0500)]
Merge tag 'wireless-drivers-for-davem-2018-03-08' of git://git./linux/kernel/git/kvalo/wireless-drivers

Kalle Valo:

====================
wireless-drivers fixes for 4.16

Quote a few fixes as I have not been able to send a pull request
earlier. Most of the fixes for iwlwifi but also few others, nothing
really standing out though.

iwlwifi

* fix a bogus warning when freeing a TFD

* fix severe throughput problem with 9000 series

* fix for a bug that caused queue hangs in certain situations

* fix for an issue with IBSS

* fix an issue with rate-scaling in AP-mode

* fix Channel Switch Announcement (CSA) issues with count 0 and 1

* some firmware debugging fixes

* remov a wrong error message when removing keys

* fix a firmware sysassert most usually triggered in IBSS

* a couple of fixes on multicast queues

* a fix with CCMP 256

rtlwifi

* fix loss of signal for rtl8723be

brcmfmac

* add possibility to obtain firmware error

* fix P2P_DEVICE ethernet address generation
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodocs: segmentation-offloads.txt: Correct TCP gso_types
Daniel Axtens [Thu, 8 Mar 2018 12:34:35 +0000 (23:34 +1100)]
docs: segmentation-offloads.txt: Correct TCP gso_types

Pretty minor: just SKB_GSO_TCP  -> SKB_GSO_TCPV4 and
                   SKB_GSO_TCP6 -> SKB_GSO_TCPV6.

Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoteam: Fix double free in error path
Arkadi Sharshevsky [Thu, 8 Mar 2018 10:42:10 +0000 (12:42 +0200)]
team: Fix double free in error path

The __send_and_alloc_skb() receives a skb ptr as a parameter but in
case it fails the skb is not valid:
- Send failed and released the skb internally.
- Allocation failed.

The current code tries to release the skb in case of failure which
causes redundant freeing.

Fixes: 9b00cf2d1024 ("team: implement multipart netlink messages for options transfers")
Signed-off-by: Arkadi Sharshevsky <arkadis@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Thu, 8 Mar 2018 01:27:51 +0000 (20:27 -0500)]
Merge git://git./pub/scm/linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2018-03-08

The following pull-request contains BPF updates for your *net* tree.

The main changes are:

1) Fix various BPF helpers which adjust the skb and its GSO information
   with regards to SCTP GSO. The latter is a special case where gso_size
   is of value GSO_BY_FRAGS, so mangling that will end up corrupting
   the skb, thus bail out when seeing SCTP GSO packets, from Daniel(s).

2) Fix a compilation error in bpftool where BPF_FS_MAGIC is not defined
   due to too old kernel headers in the system, from Jiri.

3) Increase the number of x64 JIT passes in order to allow larger images
   to converge instead of punting them to interpreter or having them
   rejected when the interpreter is not built into the kernel, from Daniel.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpf, x64: increase number of passes
Daniel Borkmann [Wed, 7 Mar 2018 21:10:01 +0000 (22:10 +0100)]
bpf, x64: increase number of passes

In Cilium some of the main programs we run today are hitting 9 passes
on x64's JIT compiler, and we've had cases already where we surpassed
the limit where the JIT then punts the program to the interpreter
instead, leading to insertion failures due to CONFIG_BPF_JIT_ALWAYS_ON
or insertion failures due to the prog array owner being JITed but the
program to insert not (both must have the same JITed/non-JITed property).

One concrete case the program image shrunk from 12,767 bytes down to
10,288 bytes where the image converged after 16 steps. I've measured
that this took 340us in the JIT until it converges on my i7-6600U. Thus,
increase the original limit we had from day one where the JIT covered
cBPF only back then before we run into the case (as similar with the
complexity limit) where we trip over this and hit program rejections.
Also add a cond_resched() into the compilation loop, the JIT process
runs without any locks and may sleep anyway.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agocxgb4: do not set needs_free_netdev for mgmt dev's
Ganesh Goudar [Wed, 7 Mar 2018 07:40:24 +0000 (13:10 +0530)]
cxgb4: do not set needs_free_netdev for mgmt dev's

Do not set 'needs_free_netdev' as we do call free_netdev
for mgmt net devices, doing both hits BUG_ON.

Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4: copy adap index to PF0-3 adapter instances
Ganesh Goudar [Wed, 7 Mar 2018 07:38:45 +0000 (13:08 +0530)]
cxgb4: copy adap index to PF0-3 adapter instances

instantiation of VF's on different adapters fails, copy
adapter index and chip type to PF0-3 adapter instances
to fix the issue.

Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: don't unnecessarily load kernel modules in dev_ioctl()
Paul Moore [Tue, 6 Mar 2018 22:27:44 +0000 (17:27 -0500)]
net: don't unnecessarily load kernel modules in dev_ioctl()

Starting with v4.16-rc1 we've been seeing a higher than usual number
of requests for the kernel to load networking modules, even on events
which shouldn't trigger a module load (e.g. ioctl(TCGETS)).  Stephen
Smalley suggested the problem may lie in commit 44c02a2c3dc5
("dev_ioctl(): move copyin/copyout to callers") which moves changes
the network dev_ioctl() function to always call dev_load(),
regardless of the requested ioctl.

This patch moves the dev_load() calls back into the individual ioctls
while preserving the rest of the original patch.

Reported-by: Dominick Grift <dac.override@gmail.com>
Suggested-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: Paul Moore <paul@paul-moore.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: purge write queue upon aborting the connection
Soheil Hassas Yeganeh [Tue, 6 Mar 2018 22:15:12 +0000 (17:15 -0500)]
tcp: purge write queue upon aborting the connection

When the connection is aborted, there is no point in
keeping the packets on the write queue until the connection
is closed.

Similar to a27fd7a8ed38 ('tcp: purge write queue upon RST'),
this is essential for a correct MSG_ZEROCOPY implementation,
because userspace cannot call close(fd) before receiving
zerocopy signals even when the connection is aborted.

Fixes: f214f915e7db ("tcp: enable MSG_ZEROCOPY")
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodccp: check sk for closed state in dccp_sendmsg()
Alexey Kodanev [Tue, 6 Mar 2018 19:57:01 +0000 (22:57 +0300)]
dccp: check sk for closed state in dccp_sendmsg()

dccp_disconnect() sets 'dp->dccps_hc_tx_ccid' tx handler to NULL,
therefore if DCCP socket is disconnected and dccp_sendmsg() is
called after it, it will cause a NULL pointer dereference in
dccp_write_xmit().

This crash and the reproducer was reported by syzbot. Looks like
it is reproduced if commit 69c64866ce07 ("dccp: CVE-2017-8824:
use-after-free in DCCP code") is applied.

Reported-by: syzbot+f99ab3887ab65d70f816@syzkaller.appspotmail.com
Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agol2tp: do not accept arbitrary sockets
Eric Dumazet [Tue, 6 Mar 2018 15:54:53 +0000 (07:54 -0800)]
l2tp: do not accept arbitrary sockets

syzkaller found an issue caused by lack of sufficient checks
in l2tp_tunnel_create()

RAW sockets can not be considered as UDP ones for instance.

In another patch, we shall replace all pr_err() by less intrusive
pr_debug() so that syzkaller can find other bugs faster.
Acked-by: Guillaume Nault <g.nault@alphalink.fr>
Acked-by: James Chapman <jchapman@katalix.com>
==================================================================
BUG: KASAN: slab-out-of-bounds in setup_udp_tunnel_sock+0x3ee/0x5f0 net/ipv4/udp_tunnel.c:69
dst_release: dst:00000000d53d0d0f refcnt:-1
Write of size 1 at addr ffff8801d013b798 by task syz-executor3/6242

CPU: 1 PID: 6242 Comm: syz-executor3 Not tainted 4.16.0-rc2+ #253
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:17 [inline]
 dump_stack+0x194/0x24d lib/dump_stack.c:53
 print_address_description+0x73/0x250 mm/kasan/report.c:256
 kasan_report_error mm/kasan/report.c:354 [inline]
 kasan_report+0x23b/0x360 mm/kasan/report.c:412
 __asan_report_store1_noabort+0x17/0x20 mm/kasan/report.c:435
 setup_udp_tunnel_sock+0x3ee/0x5f0 net/ipv4/udp_tunnel.c:69
 l2tp_tunnel_create+0x1354/0x17f0 net/l2tp/l2tp_core.c:1596
 pppol2tp_connect+0x14b1/0x1dd0 net/l2tp/l2tp_ppp.c:707
 SYSC_connect+0x213/0x4a0 net/socket.c:1640
 SyS_connect+0x24/0x30 net/socket.c:1621
 do_syscall_64+0x280/0x940 arch/x86/entry/common.c:287
 entry_SYSCALL_64_after_hwframe+0x42/0xb7

Fixes: fd558d186df2 ("l2tp: Split pppol2tp patch into separate l2tp and ppp parts")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: Fix hlist corruptions in inet_evict_bucket()
Kirill Tkhai [Tue, 6 Mar 2018 15:46:39 +0000 (18:46 +0300)]
net: Fix hlist corruptions in inet_evict_bucket()

inet_evict_bucket() iterates global list, and
several tasks may call it in parallel. All of
them hash the same fq->list_evictor to different
lists, which leads to list corruption.

This patch makes fq be hashed to expired list
only if this has not been made yet by another
task. Since inet_frag_alloc() allocates fq
using kmem_cache_zalloc(), we may rely on
list_evictor is initially unhashed.

The problem seems to exist before async
pernet_operations, as there was possible to have
exit method to be executed in parallel with
inet_frags::frags_work, so I add two Fixes tags.
This also may go to stable.

Fixes: d1fe19444d82 "inet: frag: don't re-use chainlist for evictor"
Fixes: f84c6821aa54 "net: Convert pernet_subsys, registered from inet_init()"
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: smsc911x: Fix unload crash when link is up
Jeremy Linton [Tue, 6 Mar 2018 15:00:06 +0000 (09:00 -0600)]
net: smsc911x: Fix unload crash when link is up

The smsc911x driver will crash if it is rmmod'ed while the netdev
is up like:

Call trace:
  phy_detach+0x94/0x150
  phy_disconnect+0x40/0x50
  smsc911x_stop+0x104/0x128 [smsc911x]
  __dev_close_many+0xb4/0x138
  dev_close_many+0xbc/0x190
  rollback_registered_many+0x140/0x460
  rollback_registered+0x68/0xb0
  unregister_netdevice_queue+0x100/0x118
  unregister_netdev+0x28/0x38
  smsc911x_drv_remove+0x58/0x130 [smsc911x]
  platform_drv_remove+0x30/0x50
  device_release_driver_internal+0x15c/0x1f8
  driver_detach+0x54/0x98
  bus_remove_driver+0x64/0xe8
  driver_unregister+0x34/0x60
  platform_driver_unregister+0x20/0x30
  smsc911x_cleanup_module+0x14/0xbca8 [smsc911x]
  SyS_delete_module+0x1e8/0x238
  __sys_trace_return+0x0/0x4

This is caused by the mdiobus being unregistered/free'd
and the code in phy_detach() attempting to manipulate mdio
related structures from unregister_netdev() calling close()

To fix this, we delay the mdiobus teardown until after
the netdev is deregistered.

Reported-by: Matt Sealey <matt.sealey@arm.com>
Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: Reflect MTU changes on PMTU of exceptions for MTU-less routes
Stefano Brivio [Tue, 6 Mar 2018 10:10:19 +0000 (11:10 +0100)]
ipv6: Reflect MTU changes on PMTU of exceptions for MTU-less routes

Currently, administrative MTU changes on a given netdevice are
not reflected on route exceptions for MTU-less routes, with a
set PMTU value, for that device:

 # ip -6 route get 2001:db8::b
 2001:db8::b from :: dev vti_a proto kernel src 2001:db8::a metric 256 pref medium
 # ping6 -c 1 -q -s10000 2001:db8::b > /dev/null
 # ip netns exec a ip -6 route get 2001:db8::b
 2001:db8::b from :: dev vti_a src 2001:db8::a metric 0
     cache expires 571sec mtu 4926 pref medium
 # ip link set dev vti_a mtu 3000
 # ip -6 route get 2001:db8::b
 2001:db8::b from :: dev vti_a src 2001:db8::a metric 0
     cache expires 571sec mtu 4926 pref medium
 # ip link set dev vti_a mtu 9000
 # ip -6 route get 2001:db8::b
 2001:db8::b from :: dev vti_a src 2001:db8::a metric 0
     cache expires 571sec mtu 4926 pref medium

The first issue is that since commit fb56be83e43d ("net-ipv6: on
device mtu change do not add mtu to mtu-less routes") we don't
call rt6_exceptions_update_pmtu() from rt6_mtu_change_route(),
which handles administrative MTU changes, if the regular route
is MTU-less.

However, PMTU exceptions should be always updated, as long as
RTAX_MTU is not locked. Keep the check for MTU-less main route,
as introduced by that commit, but, for exceptions,
call rt6_exceptions_update_pmtu() regardless of that check.

Once that is fixed, one problem remains: MTU changes are not
reflected if the new MTU is higher than the previous one,
because rt6_exceptions_update_pmtu() doesn't allow that. We
should instead allow PMTU increase if the old PMTU matches the
local MTU, as that implies that the old MTU was the lowest in the
path, and PMTU discovery might lead to different results.

The existing check in rt6_mtu_change_route() correctly took that
case into account (for regular routes only), so factor it out
and re-use it also in rt6_exceptions_update_pmtu().

While at it, fix comments style and grammar, and try to be a bit
more descriptive.

Reported-by: Xiumei Mu <xmu@redhat.com>
Fixes: fb56be83e43d ("net-ipv6: on device mtu change do not add mtu to mtu-less routes")
Fixes: f5bbe7ee79c2 ("ipv6: prepare rt6_mtu_change() for exception table")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: qcom/emac: Use proper free methods during TX
Hemanth Puranik [Tue, 6 Mar 2018 02:48:06 +0000 (08:18 +0530)]
net: qcom/emac: Use proper free methods during TX

This patch fixes the warning messages/call traces seen if DMA debug is
enabled, In case of fragmented skb's memory was allocated using
dma_map_page but freed using dma_unmap_single. This patch modifies buffer
allocations in TX path to use dma_map_page in all the places and
dma_unmap_page while freeing the buffers.

Signed-off-by: Hemanth Puranik <hpuranik@codeaurora.org>
Acked-by: Timur Tabi <timur@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoqed: Free RoCE ILT Memory on rmmod qedr
Michal Kalderon [Mon, 5 Mar 2018 21:50:46 +0000 (23:50 +0200)]
qed: Free RoCE ILT Memory on rmmod qedr

Rdma requires ILT Memory to be allocated for it's QPs.
Each ILT entry points to a page used by several Rdma QPs.
To avoid allocating all the memory in advance, the rdma
implementation dynamically allocates memory as more QPs are
added, however it does not dynamically free the memory.
The memory should have been freed on rmmod qedr, but isn't.
This patch adds the memory freeing on rmmod qedr (currently
it will be freed with qed is removed).

An outcome of this bug, is that if qedr is unloaded and loaded
without unloaded qed, there will be no more RoCE traffic.

The reason these are related, is that the logic of detecting the
first QP ever opened is by asking whether ILT memory for RoCE has
been allocated.

In addition, this patch modifies freeing of the Task context to
always use the PROTOCOLID_ROCE and not the protocol passed,
this is because task context for iWARP and ROCE both use the
ROCE protocol id, as opposed to the connection context.

Fixes: dbb799c39717 ("qed: Initialize hardware for new protocols")
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net...
David S. Miller [Wed, 7 Mar 2018 16:58:05 +0000 (11:58 -0500)]
Merge branch '1GbE' of git://git./linux/kernel/git/jkirsher/net-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2018-03-05

This series contains fixes to e1000e only.

Benjamin Poirier provides all but one fix in this series, starting with
workaround for a VMWare e1000e emulation issue where ICR reads 0x0 on
the emulated device.  Partially reverted a previous commit dealing with
the "Other" interrupt throttling to avoid unforeseen fallout from these
changes that are not strictly necessary.  Restored the ICS write for
receive and transmit queue interrupts in the case that txq or rxq bits
were set in ICR and the Other interrupt handler read and cleared ICR
before the queue interrupt was raised.  Fixed an bug where interrupts
may be missed if ICR is read while INT_ASSERTED is not set, so avoid the
problem by setting all bits related to events that can trigger the Other
interrupt in IMS.  Fixed the return value for check_for_link() when
auto-negotiation is off.

Pierre-Yves Kerbrat fixes e1000e to use dma_zalloc_coherent() to make
sure the ring is memset to 0 to prevent the area from containing
garbage.

v2: added an additional e1000e fix to the series
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: usbnet: fix potential deadlock on 32bit hosts
Eric Dumazet [Mon, 5 Mar 2018 19:41:13 +0000 (11:41 -0800)]
net: usbnet: fix potential deadlock on 32bit hosts

Marek reported a LOCKDEP issue occurring on 32bit host,
that we tracked down to the fact that usbnet could either
run from soft or hard irqs.

This patch adds u64_stats_update_begin_irqsave() and
u64_stats_update_end_irqrestore() helpers to solve this case.

[   17.768040] ================================
[   17.772239] WARNING: inconsistent lock state
[   17.776511] 4.16.0-rc3-next-20180227-00007-g876c53a7493c #453 Not tainted
[   17.783329] --------------------------------
[   17.787580] inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage.
[   17.793607] swapper/0/0 [HC0[0]:SC1[1]:HE1:SE0] takes:
[   17.798751]  (&syncp->seq#5){?.-.}, at: [<9b22e5f0>]
asix_rx_fixup_internal+0x188/0x288
[   17.806790] {IN-HARDIRQ-W} state was registered at:
[   17.811677]   tx_complete+0x100/0x208
[   17.815319]   __usb_hcd_giveback_urb+0x60/0xf0
[   17.819770]   xhci_giveback_urb_in_irq+0xa8/0x240
[   17.824469]   xhci_td_cleanup+0xf4/0x16c
[   17.828367]   xhci_irq+0xe74/0x2240
[   17.831827]   usb_hcd_irq+0x24/0x38
[   17.835343]   __handle_irq_event_percpu+0x98/0x510
[   17.840111]   handle_irq_event_percpu+0x1c/0x58
[   17.844623]   handle_irq_event+0x38/0x5c
[   17.848519]   handle_fasteoi_irq+0xa4/0x138
[   17.852681]   generic_handle_irq+0x18/0x28
[   17.856760]   __handle_domain_irq+0x6c/0xe4
[   17.860941]   gic_handle_irq+0x54/0xa0
[   17.864666]   __irq_svc+0x70/0xb0
[   17.867964]   arch_cpu_idle+0x20/0x3c
[   17.871578]   arch_cpu_idle+0x20/0x3c
[   17.875190]   do_idle+0x144/0x218
[   17.878468]   cpu_startup_entry+0x18/0x1c
[   17.882454]   start_kernel+0x394/0x400
[   17.886177] irq event stamp: 161912
[   17.889616] hardirqs last  enabled at (161912): [<7bedfacf>]
__netdev_alloc_skb+0xcc/0x140
[   17.897893] hardirqs last disabled at (161911): [<d58261d0>]
__netdev_alloc_skb+0x94/0x140
[   17.904903] exynos5-hsi2c 12ca0000.i2c: tx timeout
[   17.906116] softirqs last  enabled at (161904): [<387102ff>]
irq_enter+0x78/0x80
[   17.906123] softirqs last disabled at (161905): [<cf4c628e>]
irq_exit+0x134/0x158
[   17.925722].
[   17.925722] other info that might help us debug this:
[   17.933435]  Possible unsafe locking scenario:
[   17.933435].
[   17.940331]        CPU0
[   17.942488]        ----
[   17.944894]   lock(&syncp->seq#5);
[   17.948274]   <Interrupt>
[   17.950847]     lock(&syncp->seq#5);
[   17.954386].
[   17.954386]  *** DEADLOCK ***
[   17.954386].
[   17.962422] no locks held by swapper/0/0.

Fixes: c8b5d129ee29 ("net: usbnet: support 64bit stats")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Greg Ungerer <gerg@linux-m68k.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosch_netem: fix skb leak in netem_enqueue()
Alexey Kodanev [Mon, 5 Mar 2018 17:52:54 +0000 (20:52 +0300)]
sch_netem: fix skb leak in netem_enqueue()

When we exceed current packets limit and we have more than one
segment in the list returned by skb_gso_segment(), netem drops
only the first one, skipping the rest, hence kmemleak reports:

unreferenced object 0xffff880b5d23b600 (size 1024):
  comm "softirq", pid 0, jiffies 4384527763 (age 2770.629s)
  hex dump (first 32 bytes):
    00 80 23 5d 0b 88 ff ff 00 00 00 00 00 00 00 00  ..#]............
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<00000000d8a19b9d>] __alloc_skb+0xc9/0x520
    [<000000001709b32f>] skb_segment+0x8c8/0x3710
    [<00000000c7b9bb88>] tcp_gso_segment+0x331/0x1830
    [<00000000c921cba1>] inet_gso_segment+0x476/0x1370
    [<000000008b762dd4>] skb_mac_gso_segment+0x1f9/0x510
    [<000000002182660a>] __skb_gso_segment+0x1dd/0x620
    [<00000000412651b9>] netem_enqueue+0x1536/0x2590 [sch_netem]
    [<0000000005d3b2a9>] __dev_queue_xmit+0x1167/0x2120
    [<00000000fc5f7327>] ip_finish_output2+0x998/0xf00
    [<00000000d309e9d3>] ip_output+0x1aa/0x2c0
    [<000000007ecbd3a4>] tcp_transmit_skb+0x18db/0x3670
    [<0000000042d2a45f>] tcp_write_xmit+0x4d4/0x58c0
    [<0000000056a44199>] tcp_tasklet_func+0x3d9/0x540
    [<0000000013d06d02>] tasklet_action+0x1ca/0x250
    [<00000000fcde0b8b>] __do_softirq+0x1b4/0x5a3
    [<00000000e7ed027c>] irq_exit+0x1e2/0x210

Fix it by adding the rest of the segments, if any, to skb 'to_free'
list. Add new __qdisc_drop_all() and qdisc_drop_all() functions
because they can be useful in the future if we need to drop segmented
GSO packets in other places.

Fixes: 6071bd1aa13e ("netem: Segment GSO packets on enqueue")
Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'for-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetoot...
David S. Miller [Wed, 7 Mar 2018 15:51:14 +0000 (10:51 -0500)]
Merge branch 'for-upstream' of git://git./linux/kernel/git/bluetooth/bluetooth

Johan Hedberg says:

====================
pull request: bluetooth 2018-03-05

Here are a few more Bluetooth fixes for the 4.16 kernel:

 - btusb: reset/resume fixes for Yoga 920 and Dell OptiPlex 3060
 - Fix for missing encryption refresh with the Security Manager protocol
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agofsl/fman: avoid sleeping in atomic context while adding an address
Denis Kirjanov [Sun, 4 Mar 2018 18:48:17 +0000 (21:48 +0300)]
fsl/fman: avoid sleeping in atomic context while adding an address

__dev_mc_add grabs an adress spinlock so use
atomic context in kmalloc.

/ # ifconfig eth0 inet 192.168.0.111
[   89.331622] BUG: sleeping function called from invalid context at mm/slab.h:420
[   89.339002] in_atomic(): 1, irqs_disabled(): 0, pid: 1035, name: ifconfig
[   89.345799] 2 locks held by ifconfig/1035:
[   89.349908]  #0:  (rtnl_mutex){+.+.}, at: [<(ptrval)>] devinet_ioctl+0xc0/0x8a0
[   89.357258]  #1:  (_xmit_ETHER){+...}, at: [<(ptrval)>] __dev_mc_add+0x28/0x80
[   89.364520] CPU: 1 PID: 1035 Comm: ifconfig Not tainted 4.16.0-rc3-dirty #8
[   89.371464] Call Trace:
[   89.373908] [e959db60] [c066f948] dump_stack+0xa4/0xfc (unreliable)
[   89.380177] [e959db80] [c00671d8] ___might_sleep+0x248/0x280
[   89.385833] [e959dba0] [c01aec34] kmem_cache_alloc_trace+0x174/0x320
[   89.392179] [e959dbd0] [c04ab920] dtsec_add_hash_mac_address+0x130/0x240
[   89.398874] [e959dc00] [c04a9d74] set_multi+0x174/0x1b0
[   89.404093] [e959dc30] [c04afb68] dpaa_set_rx_mode+0x68/0xe0
[   89.409745] [e959dc40] [c057baf8] __dev_mc_add+0x58/0x80
[   89.415052] [e959dc60] [c060fd64] igmp_group_added+0x164/0x190
[   89.420878] [e959dca0] [c060ffa8] ip_mc_inc_group+0x218/0x460
[   89.426617] [e959dce0] [c06120fc] ip_mc_up+0x3c/0x190
[   89.431662] [e959dd10] [c0607270] inetdev_event+0x250/0x620
[   89.437227] [e959dd50] [c005f190] notifier_call_chain+0x80/0xf0
[   89.443138] [e959dd80] [c0573a74] __dev_notify_flags+0x54/0xf0
[   89.448964] [e959dda0] [c05743f8] dev_change_flags+0x48/0x60
[   89.454615] [e959ddc0] [c0606744] devinet_ioctl+0x544/0x8a0
[   89.460180] [e959de10] [c060987c] inet_ioctl+0x9c/0x1f0
[   89.465400] [e959de80] [c05479a8] sock_ioctl+0x168/0x460
[   89.470708] [e959ded0] [c01cf3ec] do_vfs_ioctl+0xac/0x8c0
[   89.476099] [e959df20] [c01cfc40] SyS_ioctl+0x40/0xc0
[   89.481147] [e959df40] [c0011318] ret_from_syscall+0x0/0x3c
[   89.486715] --- interrupt: c01 at 0x1006943c
[   89.486715]     LR = 0x100c45ec

Signed-off-by: Denis Kirjanov <kda@linux-powerpc.org>
Acked-by: Madalin Bucur <madalin.bucur@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'rhltable-dups'
David S. Miller [Wed, 7 Mar 2018 15:44:13 +0000 (10:44 -0500)]
Merge branch 'rhltable-dups'

Paul Blakey says:

====================
rhashtable: Fix rhltable duplicates insertion

On our mlx5 driver fs_core.c, we use the rhltable interface to store
flow groups. We noticed that sometimes we get a warning that flow group isn't
found at removal. This rare case was caused when a specific scenario happened,
insertion of a flow group with a similar match criteria (a duplicate),
but only where the flow group rhash_head was second (or not first)
on the relevant rhashtable bucket list.

The first patch fixes it, and the second one adds a test that show
it is now working.

Paul.

v3 --> v2 changes:
    * added missing fix in rhashtable_lookup_one code path as well.

v1 --> v2 changes:
    * Changed commit messages to better reflect the change
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotest_rhashtable: add test case for rhltable with duplicate objects
Paul Blakey [Sun, 4 Mar 2018 15:29:49 +0000 (17:29 +0200)]
test_rhashtable: add test case for rhltable with duplicate objects

Tries to insert duplicates in the middle of bucket's chain:
bucket 1:  [[val 21 (tid=1)]] -> [[ val 1 (tid=2),  val 1 (tid=0) ]]

Reuses tid to distinguish the elements insertion order.

Signed-off-by: Paul Blakey <paulb@mellanox.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>