OSDN Git Service

uclinux-h8/linux.git
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Thu, 3 May 2018 00:42:44 +0000 (20:42 -0400)]
Merge git://git./pub/scm/linux/kernel/git/bpf/bpf

Daniel Borkmann says:

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

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

The main changes are:

1) Several BPF sockmap fixes mostly related to bugs in error path
   handling, that is, a bug in updating the scatterlist length /
   offset accounting, a missing sk_mem_uncharge() in redirect
   error handling, and a bug where the outstanding bytes counter
   sg_size was not zeroed, from John.

2) Fix two memory leaks in the x86-64 BPF JIT, one in an error
   path where we still don't converge after image was allocated
   and another one where BPF calls are used and JIT passes don't
   converge, from Daniel.

3) Minor fix in BPF selftests where in test_stacktrace_build_id()
   we drop useless args in urandom_read and we need to add a missing
   newline in a CHECK() error message, from Song.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'bpf-sockmap-fixes'
Alexei Starovoitov [Wed, 2 May 2018 22:30:45 +0000 (15:30 -0700)]
Merge branch 'bpf-sockmap-fixes'

John Fastabend says:

====================
When I added the test_sockmap to selftests I mistakenly changed the
test logic a bit. The result of this was on redirect cases we ended up
choosing the wrong sock from the BPF program and ended up sending to a
socket that had no receive handler. The result was the actual receive
handler, running on a different socket, is timing out and closing the
socket. This results in errors (-EPIPE to be specific) on the sending
side. Typically happening if the sender does not complete the send
before the receive side times out. So depending on timing and the size
of the send we may get errors. This exposed some bugs in the sockmap
error path handling.

This series fixes the errors. The primary issue is we did not do proper
memory accounting in these cases which resulted in missing a
sk_mem_uncharge(). This happened in the redirect path and in one case
on the normal send path. See the three patches for the details.

The other take-away from this is we need to fix the test_sockmap and
also add more negative test cases. That will happen in bpf-next.

Finally, I tested this using the existing test_sockmap program, the
older sockmap sample test script, and a few real use cases with
Cilium. All of these seem to be in working correctly.

v2: fix compiler warning, drop iterator variable 'i' that is no longer
    used in patch 3.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agobpf: sockmap, fix error handling in redirect failures
John Fastabend [Wed, 2 May 2018 20:50:29 +0000 (13:50 -0700)]
bpf: sockmap, fix error handling in redirect failures

When a redirect failure happens we release the buffers in-flight
without calling a sk_mem_uncharge(), the uncharge is called before
dropping the sock lock for the redirecte, however we missed updating
the ring start index. When no apply actions are in progress this
is OK because we uncharge the entire buffer before the redirect.
But, when we have apply logic running its possible that only a
portion of the buffer is being redirected. In this case we only
do memory accounting for the buffer slice being redirected and
expect to be able to loop over the BPF program again and/or if
a sock is closed uncharge the memory at sock destruct time.

With an invalid start index however the program logic looks at
the start pointer index, checks the length, and when seeing the
length is zero (from the initial release and failure to update
the pointer) aborts without uncharging/releasing the remaining
memory.

The fix for this is simply to update the start index. To avoid
fixing this error in two locations we do a small refactor and
remove one case where it is open-coded. Then fix it in the
single function.

Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agobpf: sockmap, zero sg_size on error when buffer is released
John Fastabend [Wed, 2 May 2018 20:50:24 +0000 (13:50 -0700)]
bpf: sockmap, zero sg_size on error when buffer is released

When an error occurs during a redirect we have two cases that need
to be handled (i) we have a cork'ed buffer (ii) we have a normal
sendmsg buffer.

In the cork'ed buffer case we don't currently support recovering from
errors in a redirect action. So the buffer is released and the error
should _not_ be pushed back to the caller of sendmsg/sendpage. The
rationale here is the user will get an error that relates to old
data that may have been sent by some arbitrary thread on that sock.
Instead we simple consume the data and tell the user that the data
has been consumed. We may add proper error recovery in the future.
However, this patch fixes a bug where the bytes outstanding counter
sg_size was not zeroed. This could result in a case where if the user
has both a cork'ed action and apply action in progress we may
incorrectly call into the BPF program when the user expected an
old verdict to be applied via the apply action. I don't have a use
case where using apply and cork at the same time is valid but we
never explicitly reject it because it should work fine. This patch
ensures the sg_size is zeroed so we don't have this case.

In the normal sendmsg buffer case (no cork data) we also do not
zero sg_size. Again this can confuse the apply logic when the logic
calls into the BPF program when the BPF programmer expected the old
verdict to remain. So ensure we set sg_size to zero here as well. And
additionally to keep the psock state in-sync with the sk_msg_buff
release all the memory as well. Previously we did this before
returning to the user but this left a gap where psock and sk_msg_buff
states were out of sync which seems fragile. No additional overhead
is taken here except for a call to check the length and realize its
already been freed. This is in the error path as well so in my
opinion lets have robust code over optimized error paths.

Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agobpf: sockmap, fix scatterlist update on error path in send with apply
John Fastabend [Wed, 2 May 2018 20:50:19 +0000 (13:50 -0700)]
bpf: sockmap, fix scatterlist update on error path in send with apply

When the call to do_tcp_sendpage() fails to send the complete block
requested we either retry if only a partial send was completed or
abort if we receive a error less than or equal to zero. Before
returning though we must update the scatterlist length/offset to
account for any partial send completed.

Before this patch we did this at the end of the retry loop, but
this was buggy when used while applying a verdict to fewer bytes
than in the scatterlist. When the scatterlist length was being set
we forgot to account for the apply logic reducing the size variable.
So the result was we chopped off some bytes in the scatterlist without
doing proper cleanup on them. This results in a WARNING when the
sock is tore down because the bytes have previously been charged to
the socket but are never uncharged.

The simple fix is to simply do the accounting inside the retry loop
subtracting from the absolute scatterlist values rather than trying
to accumulate the totals and subtract at the end.

Reported-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agonet_sched: fq: take care of throttled flows before reuse
Eric Dumazet [Wed, 2 May 2018 17:03:30 +0000 (10:03 -0700)]
net_sched: fq: take care of throttled flows before reuse

Normally, a socket can not be freed/reused unless all its TX packets
left qdisc and were TX-completed. However connect(AF_UNSPEC) allows
this to happen.

With commit fc59d5bdf1e3 ("pkt_sched: fq: clear time_next_packet for
reused flows") we cleared f->time_next_packet but took no special
action if the flow was still in the throttled rb-tree.

Since f->time_next_packet is the key used in the rb-tree searches,
blindly clearing it might break rb-tree integrity. We need to make
sure the flow is no longer in the rb-tree to avoid this problem.

Fixes: fc59d5bdf1e3 ("pkt_sched: fq: clear time_next_packet for reused flows")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: Revert "ipv6: Allow non-gateway ECMP for IPv6"
Ido Schimmel [Wed, 2 May 2018 19:41:56 +0000 (22:41 +0300)]
ipv6: Revert "ipv6: Allow non-gateway ECMP for IPv6"

This reverts commit edd7ceb78296 ("ipv6: Allow non-gateway ECMP for
IPv6").

Eric reported a division by zero in rt6_multipath_rebalance() which is
caused by above commit that considers identical local routes to be
siblings. The division by zero happens because a nexthop weight is not
set for local routes.

Revert the commit as it does not fix a bug and has side effects.

To reproduce:

# ip -6 address add 2001:db8::1/64 dev dummy0
# ip -6 address add 2001:db8::1/64 dev dummy1

Fixes: edd7ceb78296 ("ipv6: Allow non-gateway ECMP for IPv6")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Tested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'x86-bpf-jit-fixes'
Alexei Starovoitov [Wed, 2 May 2018 19:35:47 +0000 (12:35 -0700)]
Merge branch 'x86-bpf-jit-fixes'

Daniel Borkmann says:

====================
Fix two memory leaks in x86 JIT. For details, please see
individual patches in this series. Thanks!
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agobpf, x64: fix memleak when not converging on calls
Daniel Borkmann [Wed, 2 May 2018 18:12:23 +0000 (20:12 +0200)]
bpf, x64: fix memleak when not converging on calls

The JIT logic in jit_subprogs() is as follows: for all subprogs we
allocate a bpf_prog_alloc(), populate it (prog->is_func = 1 here),
and pass it to bpf_int_jit_compile(). If a failure occurred during
JIT and prog->jited is not set, then we bail out from attempting to
JIT the whole program, and punt to the interpreter instead. In case
JITing went successful, we fixup BPF call offsets and do another
pass to bpf_int_jit_compile() (extra_pass is true at that point) to
complete JITing calls. Given that requires to pass JIT context around
addrs and jit_data from x86 JIT are freed in the extra_pass in
bpf_int_jit_compile() when calls are involved (if not, they can
be freed immediately). However, if in the original pass, the JIT
image didn't converge then we leak addrs and jit_data since image
itself is NULL, the prog->is_func is set and extra_pass is false
in that case, meaning both will become unreachable and are never
cleaned up, therefore we need to free as well on !image. Only x64
JIT is affected.

Fixes: 1c2a088a6626 ("bpf: x64: add JIT support for multi-function programs")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agobpf, x64: fix memleak when not converging after image
Daniel Borkmann [Wed, 2 May 2018 18:12:22 +0000 (20:12 +0200)]
bpf, x64: fix memleak when not converging after image

While reviewing x64 JIT code, I noticed that we leak the prior allocated
JIT image in the case where proglen != oldproglen during the JIT passes.
Prior to the commit e0ee9c12157d ("x86: bpf_jit: fix two bugs in eBPF JIT
compiler") we would just break out of the loop, and using the image as the
JITed prog since it could only shrink in size anyway. After e0ee9c12157d,
we would bail out to out_addrs label where we free addrs and jit_data but
not the image coming from bpf_jit_binary_alloc().

Fixes: e0ee9c12157d ("x86: bpf_jit: fix two bugs in eBPF JIT compiler")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agonet/smc: restrict non-blocking connect finish
Ursula Braun [Wed, 2 May 2018 14:53:56 +0000 (16:53 +0200)]
net/smc: restrict non-blocking connect finish

The smc_poll code tries to finish connect() if the socket is in
state SMC_INIT and polling of the internal CLC-socket returns with
EPOLLOUT. This makes sense for a select/poll call following a connect
call, but not without preceding connect().
With this patch smc_poll starts connect logic only, if the CLC-socket
is no longer in its initial state TCP_CLOSE.

In addition, a poll error on the internal CLC-socket is always
propagated to the SMC socket.

With this patch the code path mentioned by syzbot
https://syzkaller.appspot.com/bug?extid=03faa2dc16b8b64be396
is no longer possible.

Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Reported-by: syzbot+03faa2dc16b8b64be396@syzkaller.appspotmail.com
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years ago8139too: Use disable_irq_nosync() in rtl8139_poll_controller()
Ingo Molnar [Wed, 2 May 2018 11:30:57 +0000 (13:30 +0200)]
8139too: Use disable_irq_nosync() in rtl8139_poll_controller()

Use disable_irq_nosync() instead of disable_irq() as this might be
called in atomic context with netpoll.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: fix the issue that the cookie-ack with auth can't get processed
Xin Long [Wed, 2 May 2018 05:45:12 +0000 (13:45 +0800)]
sctp: fix the issue that the cookie-ack with auth can't get processed

When auth is enabled for cookie-ack chunk, in sctp_inq_pop, sctp
processes auth chunk first, then continues to the next chunk in
this packet if chunk_end + chunk_hdr size < skb_tail_pointer().
Otherwise, it will go to the next packet or discard this chunk.

However, it missed the fact that cookie-ack chunk's size is equal
to chunk_hdr size, which couldn't match that check, and thus this
chunk would not get processed.

This patch fixes it by changing the check to chunk_end + chunk_hdr
size <= skb_tail_pointer().

Fixes: 26b87c788100 ("net: sctp: fix remote memory pressure from excessive queueing")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: use the old asoc when making the cookie-ack chunk in dupcook_d
Xin Long [Wed, 2 May 2018 05:39:46 +0000 (13:39 +0800)]
sctp: use the old asoc when making the cookie-ack chunk in dupcook_d

When processing a duplicate cookie-echo chunk, for case 'D', sctp will
not process the param from this chunk. It means old asoc has nothing
to be updated, and the new temp asoc doesn't have the complete info.

So there's no reason to use the new asoc when creating the cookie-ack
chunk. Otherwise, like when auth is enabled for cookie-ack, the chunk
can not be set with auth, and it will definitely be dropped by peer.

This issue is there since very beginning, and we fix it by using the
old asoc instead.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: init active key for the new asoc in dupcook_a and dupcook_b
Xin Long [Wed, 2 May 2018 05:37:44 +0000 (13:37 +0800)]
sctp: init active key for the new asoc in dupcook_a and dupcook_b

When processing a duplicate cookie-echo chunk, for case 'A' and 'B',
after sctp_process_init for the new asoc, if auth is enabled for the
cookie-ack chunk, the active key should also be initialized.

Otherwise, the cookie-ack chunk made later can not be set with auth
shkey properly, and a crash can even be caused by this, as after
Commit 1b1e0bc99474 ("sctp: add refcnt support for sh_key"), sctp
needs to hold the shkey when making control chunks.

Fixes: 1b1e0bc99474 ("sctp: add refcnt support for sh_key")
Reported-by: Jianwen Ji <jiji@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp_bbr: fix to zero idle_restart only upon S/ACKed data
Neal Cardwell [Wed, 2 May 2018 01:45:41 +0000 (21:45 -0400)]
tcp_bbr: fix to zero idle_restart only upon S/ACKed data

Previously the bbr->idle_restart tracking was zeroing out the
bbr->idle_restart bit upon ACKs that did not SACK or ACK anything,
e.g. receiving incoming data or receiver window updates. In such
situations BBR would forget that this was a restart-from-idle
situation, and if the min_rtt had expired it would unnecessarily enter
PROBE_RTT (even though we were actually restarting from idle but had
merely forgotten that fact).

The fix is simple: we need to remember we are restarting from idle
until we receive a S/ACK for some data (a S/ACK for the first flight
of data we send as we are restarting).

This commit is a stable candidate for kernels back as far as 4.9.

Fixes: 0f8782ea1497 ("tcp_bbr: add BBR congestion control")
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Yousuk Seung <ysseung@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethernet: ti: cpsw: fix packet leaking in dual_mac mode
Grygorii Strashko [Tue, 1 May 2018 17:41:22 +0000 (12:41 -0500)]
net: ethernet: ti: cpsw: fix packet leaking in dual_mac mode

In dual_mac mode packets arrived on one port should not be forwarded by
switch hw to another port. Only Linux Host can forward packets between
ports. The below test case (reported in [1]) shows that packet arrived on
one port can be leaked to anoter (reproducible with dual port evms):
 - connect port 1 (eth0) to linux Host 0 and run tcpdump or Wireshark
 - connect port 2 (eth1) to linux Host 1 with vlan 1 configured
 - ping <IPx> from Host 1 through vlan 1 interface.
ARP packets will be seen on Host 0.

Issue happens because dual_mac mode is implemnted using two vlans: 1 (Port
1+Port 0) and 2 (Port 2+Port 0), so there are vlan records created for for
each vlan. By default, the ALE will find valid vlan record in its table
when vlan 1 tagged packet arrived on Port 2 and so forwards packet to all
ports which are vlan 1 members (like Port.

To avoid such behaviorr the ALE VLAN ID Ingress Check need to be enabled
for each external CPSW port (ALE_PORTCTLn.VID_INGRESS_CHECK) so ALE will
drop ingress packets if Rx port is not VLAN member.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoRevert "vhost: make msg padding explicit"
Michael S. Tsirkin [Wed, 2 May 2018 14:19:05 +0000 (17:19 +0300)]
Revert "vhost: make msg padding explicit"

This reverts commit 93c0d549c4c5a7382ad70de6b86610b7aae57406.

Unfortunately the padding will break 32 bit userspace.
Ouch. Need to add some compat code, revert for now.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: flower: set tunnel ttl value to net default
John Hurley [Tue, 1 May 2018 22:49:49 +0000 (15:49 -0700)]
nfp: flower: set tunnel ttl value to net default

Firmware requires that the ttl value for an encapsulating ipv4 tunnel
header be included as an action field. Prior to the support of Geneve
tunnel encap (when ttl set was removed completely), ttl value was
extracted from the tunnel key. However, tests have shown that this can
still produce a ttl of 0.

Fix the issue by setting the namespace default value for each new tunnel.
Follow up patch for net-next will do a full route lookup.

Fixes: 3ca3059dc3a9 ("nfp: flower: compile Geneve encap actions")
Fixes: b27d6a95a70d ("nfp: compile flower vxlan tunnel set actions")
Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/tls: Don't recursively call push_record during tls_write_space callbacks
Dave Watson [Tue, 1 May 2018 20:05:39 +0000 (13:05 -0700)]
net/tls: Don't recursively call push_record during tls_write_space callbacks

It is reported that in some cases, write_space may be called in
do_tcp_sendpages, such that we recursively invoke do_tcp_sendpages again:

[  660.468802]  ? do_tcp_sendpages+0x8d/0x580
[  660.468826]  ? tls_push_sg+0x74/0x130 [tls]
[  660.468852]  ? tls_push_record+0x24a/0x390 [tls]
[  660.468880]  ? tls_write_space+0x6a/0x80 [tls]
...

tls_push_sg already does a loop over all sending sg's, so ignore
any tls_write_space notifications until we are done sending.
We then have to call the previous write_space to wake up
poll() waiters after we are done with the send loop.

Reported-by: Andre Tomt <andre@tomt.net>
Signed-off-by: Dave Watson <davejwatson@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpf: minor fix to selftest test_stacktrace_build_id()
Song Liu [Tue, 1 May 2018 17:20:24 +0000 (10:20 -0700)]
bpf: minor fix to selftest test_stacktrace_build_id()

1. remove useless parameter list to ./urandom_read
2. add missing "\n" to the end of an error message

Fixes: 81f77fd0deeb ("bpf: add selftest for stackmap with BPF_F_STACK_BUILD_ID")
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agoipv6: Allow non-gateway ECMP for IPv6
Thomas Winter [Mon, 30 Apr 2018 21:15:29 +0000 (09:15 +1200)]
ipv6: Allow non-gateway ECMP for IPv6

It is valid to have static routes where the nexthop
is an interface not an address such as tunnels.
For IPv4 it was possible to use ECMP on these routes
but not for IPv6.

Signed-off-by: Thomas Winter <Thomas.Winter@alliedtelesis.co.nz>
Cc: David Ahern <dsahern@gmail.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoethtool: fix a potential missing-check bug
Wenwen Wang [Mon, 30 Apr 2018 17:31:13 +0000 (12:31 -0500)]
ethtool: fix a potential missing-check bug

In ethtool_get_rxnfc(), the object "info" is firstly copied from
user-space. If the FLOW_RSS flag is set in the member field flow_type of
"info" (and cmd is ETHTOOL_GRXFH), info needs to be copied again from
user-space because FLOW_RSS is newer and has new definition, as mentioned
in the comment. However, given that the user data resides in user-space, a
malicious user can race to change the data after the first copy. By doing
so, the user can inject inconsistent data. For example, in the second
copy, the FLOW_RSS flag could be cleared in the field flow_type of "info".
In the following execution, "info" will be used in the function
ops->get_rxnfc(). Such inconsistent data can potentially lead to unexpected
information leakage since ops->get_rxnfc() will prepare various types of
data according to flow_type, and the prepared data will be eventually
copied to user-space. This inconsistent data may also cause undefined
behaviors based on how ops->get_rxnfc() is implemented.

This patch simply re-verifies the flow_type field of "info" after the
second copy. If the value is not as expected, an error code will be
returned.

Signed-off-by: Wenwen Wang <wang6495@umn.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/mlx4: fix spelling mistake: "failedi" -> "failed"
Colin Ian King [Mon, 30 Apr 2018 16:29:45 +0000 (17:29 +0100)]
net/mlx4: fix spelling mistake: "failedi" -> "failed"

trivial fix to spelling mistake in mlx4_warn message.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovhost: make msg padding explicit
Michael S. Tsirkin [Fri, 27 Apr 2018 16:02:05 +0000 (19:02 +0300)]
vhost: make msg padding explicit

There's a 32 bit hole just after type. It's best to
give it a name, this way compiler is forced to initialize
it with rest of the structure.

Reported-by: Kevin Easton <kevin@guarana.org>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: fix TCP_REPAIR_QUEUE bound checking
Eric Dumazet [Mon, 30 Apr 2018 01:55:20 +0000 (18:55 -0700)]
tcp: fix TCP_REPAIR_QUEUE bound checking

syzbot is able to produce a nasty WARN_ON() in tcp_verify_left_out()
with following C-repro :

socket(PF_INET, SOCK_STREAM, IPPROTO_IP) = 3
setsockopt(3, SOL_TCP, TCP_REPAIR, [1], 4) = 0
setsockopt(3, SOL_TCP, TCP_REPAIR_QUEUE, [-1], 4) = 0
bind(3, {sa_family=AF_INET, sin_port=htons(20002), sin_addr=inet_addr("0.0.0.0")}, 16) = 0
sendto(3, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"...,
1242, MSG_FASTOPEN, {sa_family=AF_INET, sin_port=htons(20002), sin_addr=inet_addr("127.0.0.1")}, 16) = 1242
setsockopt(3, SOL_TCP, TCP_REPAIR_WINDOW, "\4\0\0@+\205\0\0\377\377\0\0\377\377\377\177\0\0\0\0", 20) = 0
writev(3, [{"\270", 1}], 1)             = 1
setsockopt(3, SOL_TCP, TCP_REPAIR_OPTIONS, "\10\0\0\0\0\0\0\0\0\0\0\0|\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 386) = 0
writev(3, [{"\210v\r[\226\320t\231qwQ\204\264l\254\t\1\20\245\214p\350H\223\254;\\\37\345\307p$"..., 3144}], 1) = 3144

The 3rd system call looks odd :
setsockopt(3, SOL_TCP, TCP_REPAIR_QUEUE, [-1], 4) = 0

This patch makes sure bound checking is using an unsigned compare.

Fixes: ee9952831cfd ("tcp: Initial repair mode")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: fix uninit-value in ip6_multipath_l3_keys()
Eric Dumazet [Sun, 29 Apr 2018 16:54:59 +0000 (09:54 -0700)]
ipv6: fix uninit-value in ip6_multipath_l3_keys()

syzbot/KMSAN reported an uninit-value in ip6_multipath_l3_keys(),
root caused to a bad assumption of ICMP header being already
pulled in skb->head

ip_multipath_l3_keys() does the correct thing, so it is an IPv6 only bug.

BUG: KMSAN: uninit-value in ip6_multipath_l3_keys net/ipv6/route.c:1830 [inline]
BUG: KMSAN: uninit-value in rt6_multipath_hash+0x5c4/0x640 net/ipv6/route.c:1858
CPU: 0 PID: 4507 Comm: syz-executor661 Not tainted 4.16.0+ #87
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+0x185/0x1d0 lib/dump_stack.c:53
 kmsan_report+0x142/0x240 mm/kmsan/kmsan.c:1067
 __msan_warning_32+0x6c/0xb0 mm/kmsan/kmsan_instr.c:683
 ip6_multipath_l3_keys net/ipv6/route.c:1830 [inline]
 rt6_multipath_hash+0x5c4/0x640 net/ipv6/route.c:1858
 ip6_route_input+0x65a/0x920 net/ipv6/route.c:1884
 ip6_rcv_finish+0x413/0x6e0 net/ipv6/ip6_input.c:69
 NF_HOOK include/linux/netfilter.h:288 [inline]
 ipv6_rcv+0x1e16/0x2340 net/ipv6/ip6_input.c:208
 __netif_receive_skb_core+0x47df/0x4a90 net/core/dev.c:4562
 __netif_receive_skb net/core/dev.c:4627 [inline]
 netif_receive_skb_internal+0x49d/0x630 net/core/dev.c:4701
 netif_receive_skb+0x230/0x240 net/core/dev.c:4725
 tun_rx_batched drivers/net/tun.c:1555 [inline]
 tun_get_user+0x740f/0x7c60 drivers/net/tun.c:1962
 tun_chr_write_iter+0x1d4/0x330 drivers/net/tun.c:1990
 call_write_iter include/linux/fs.h:1782 [inline]
 new_sync_write fs/read_write.c:469 [inline]
 __vfs_write+0x7fb/0x9f0 fs/read_write.c:482
 vfs_write+0x463/0x8d0 fs/read_write.c:544
 SYSC_write+0x172/0x360 fs/read_write.c:589
 SyS_write+0x55/0x80 fs/read_write.c:581
 do_syscall_64+0x309/0x430 arch/x86/entry/common.c:287
 entry_SYSCALL_64_after_hwframe+0x3d/0xa2

Fixes: 23aebdacb05d ("ipv6: Compute multipath hash for ICMP errors from offending packet")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Jakub Sitnicki <jkbs@redhat.com>
Acked-by: Jakub Sitnicki <jkbs@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethernet: ucc: fix spelling mistake: "tx-late-collsion" -> "tx-late-collision"
Colin Ian King [Sat, 28 Apr 2018 09:57:07 +0000 (10:57 +0100)]
net: ethernet: ucc: fix spelling mistake: "tx-late-collsion" -> "tx-late-collision"

Trivial fix to spelling mistake in tx_fw_stat_gstrings text

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoqed: fix spelling mistake: "checksumed" -> "checksummed"
Colin Ian King [Sat, 28 Apr 2018 09:43:20 +0000 (10:43 +0100)]
qed: fix spelling mistake: "checksumed" -> "checksummed"

Trivial fix to spelling mistake in DP_INFO message text

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMAINTAINERS: add myself as SCTP co-maintainer
Marcelo Ricardo Leitner [Fri, 27 Apr 2018 19:46:11 +0000 (16:46 -0300)]
MAINTAINERS: add myself as SCTP co-maintainer

Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: systemport: fix spelling mistake: "asymetric" -> "asymmetric"
Colin Ian King [Fri, 27 Apr 2018 19:09:25 +0000 (20:09 +0100)]
net: systemport: fix spelling mistake: "asymetric" -> "asymmetric"

Trivial fix to spelling mistake in netdev_warn warning message

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobridge: check iface upper dev when setting master via ioctl
Hangbin Liu [Fri, 27 Apr 2018 12:59:24 +0000 (20:59 +0800)]
bridge: check iface upper dev when setting master via ioctl

When we set a bond slave's master to bridge via ioctl, we only check
the IFF_BRIDGE_PORT flag. Although we will find the slave's real master
at netdev_master_upper_dev_link() later, it already does some settings
and allocates some resources. It would be better to return as early
as possible.

v1 -> v2:
use netdev_master_upper_dev_get() instead of netdev_has_any_upper_dev()
to check if we have a master, because not all upper devs are masters,
e.g. vlan device.

Reported-by: syzbot+de73361ee4971b6e6f75@syzkaller.appspotmail.com
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'sfc-more-ARFS-fixes'
David S. Miller [Sat, 28 Apr 2018 00:21:08 +0000 (20:21 -0400)]
Merge branch 'sfc-more-ARFS-fixes'

Edward Cree says:

====================
sfc: more ARFS fixes

A couple more bits of breakage in my recent ARFS and async filters work.
Patch #1 in particular fixes a bug that leads to memory trampling and
 consequent crashes.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosfc: fix ARFS expiry check on EF10
Edward Cree [Fri, 27 Apr 2018 14:08:57 +0000 (15:08 +0100)]
sfc: fix ARFS expiry check on EF10

Owing to a missing conditional, the result of rps_may_expire_flow() was
 being ignored and filters were being removed even if we'd decided not to
 expire them.

Fixes: f8d6203780b7 ("sfc: ARFS filter IDs")
Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosfc: Use filter index rather than ID for rps_flow_id table
Edward Cree [Fri, 27 Apr 2018 14:08:41 +0000 (15:08 +0100)]
sfc: Use filter index rather than ID for rps_flow_id table

efx->type->filter_insert() returns an ID rather than the index that
 efx->type->filter_async_insert() used to, which causes it to exceed
 efx->type->max_rx_ip_filters on some EF10 configurations, leading to out-
 of-bounds array writes.
So, in efx_filter_rfs_work(), convert this back into an index (which is
 what the remove call in the expiry path expects, anyway).

Fixes: 3af0f34290f6 ("sfc: replace asynchronous filter operations")
Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: support compat 64-bit time in {s,g}etsockopt
Lance Richardson [Wed, 25 Apr 2018 14:21:54 +0000 (10:21 -0400)]
net: support compat 64-bit time in {s,g}etsockopt

For the x32 ABI, struct timeval has two 64-bit fields. However
the kernel currently interprets the user-space values used for
the SO_RCVTIMEO and SO_SNDTIMEO socket options as having a pair
of 32-bit fields.

When the seconds portion of the requested timeout is less than 2**32,
the seconds portion of the effective timeout is correct but the
microseconds portion is zero.  When the seconds portion of the
requested timeout is zero and the microseconds portion is non-zero,
the kernel interprets the timeout as zero (never timeout).

Fix by using 64-bit time for SO_RCVTIMEO/SO_SNDTIMEO as required
for the ABI.

The code included below demonstrates the problem.

Results before patch:
    $ gcc -m64 -Wall -O2 -o socktmo socktmo.c && ./socktmo
    recv time: 2.008181 seconds
    send time: 2.015985 seconds

    $ gcc -m32 -Wall -O2 -o socktmo socktmo.c && ./socktmo
    recv time: 2.016763 seconds
    send time: 2.016062 seconds

    $ gcc -mx32 -Wall -O2 -o socktmo socktmo.c && ./socktmo
    recv time: 1.007239 seconds
    send time: 1.023890 seconds

Results after patch:
    $ gcc -m64 -O2 -Wall -o socktmo socktmo.c && ./socktmo
    recv time: 2.010062 seconds
    send time: 2.015836 seconds

    $ gcc -m32 -O2 -Wall -o socktmo socktmo.c && ./socktmo
    recv time: 2.013974 seconds
    send time: 2.015981 seconds

    $ gcc -mx32 -O2 -Wall -o socktmo socktmo.c && ./socktmo
    recv time: 2.030257 seconds
    send time: 2.013383 seconds

 #include <stdio.h>
 #include <stdlib.h>
 #include <sys/socket.h>
 #include <sys/types.h>
 #include <sys/time.h>

 void checkrc(char *str, int rc)
 {
         if (rc >= 0)
                 return;

         perror(str);
         exit(1);
 }

 static char buf[1024];
 int main(int argc, char **argv)
 {
         int rc;
         int socks[2];
         struct timeval tv;
         struct timeval start, end, delta;

         rc = socketpair(AF_UNIX, SOCK_STREAM, 0, socks);
         checkrc("socketpair", rc);

         /* set timeout to 1.999999 seconds */
         tv.tv_sec = 1;
         tv.tv_usec = 999999;
         rc = setsockopt(socks[0], SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof tv);
         rc = setsockopt(socks[0], SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof tv);
         checkrc("setsockopt", rc);

         /* measure actual receive timeout */
         gettimeofday(&start, NULL);
         rc = recv(socks[0], buf, sizeof buf, 0);
         gettimeofday(&end, NULL);
         timersub(&end, &start, &delta);

         printf("recv time: %ld.%06ld seconds\n",
                (long)delta.tv_sec, (long)delta.tv_usec);

         /* fill send buffer */
         do {
                 rc = send(socks[0], buf, sizeof buf, 0);
         } while (rc > 0);

         /* measure actual send timeout */
         gettimeofday(&start, NULL);
         rc = send(socks[0], buf, sizeof buf, 0);
         gettimeofday(&end, NULL);
         timersub(&end, &start, &delta);

         printf("send time: %ld.%06ld seconds\n",
                (long)delta.tv_sec, (long)delta.tv_usec);
         exit(0);
 }

Fixes: 515c7af85ed9 ("x32: Use compat shims for {g,s}etsockopt")
Reported-by: Gopal RajagopalSai <gopalsr83@gmail.com>
Signed-off-by: Lance Richardson <lance.richardson.net@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMAINTAINERS: add davem in NETWORKING DRIVERS
Vivien Didelot [Thu, 26 Apr 2018 23:47:35 +0000 (19:47 -0400)]
MAINTAINERS: add davem in NETWORKING DRIVERS

"./scripts/get_maintainer.pl -f" does not actually show us David as the
maintainer of drivers/net directories such as team, bonding, phy or dsa.
Adding him in an M: entry of NETWORKING DRIVERS fixes this.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'mlx5-fixes-2018-04-25' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Fri, 27 Apr 2018 18:30:38 +0000 (14:30 -0400)]
Merge tag 'mlx5-fixes-2018-04-25' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
Mellanox, mlx5 fixes 2018-04-26

This pull request includes fixes for mlx5 core and netdev driver.

Please pull and let me know if there's any problems.

For -stable v4.12
    net/mlx5e: TX, Use correct counter in dma_map error flow
For -stable v4.13
    net/mlx5: Avoid cleaning flow steering table twice during error flow
For -stable v4.14
    net/mlx5e: Allow offloading ipv4 header re-write for icmp
For -stable v4.15
    net/mlx5e: DCBNL fix min inline header size for dscp
For -stable v4.16
    net/mlx5: Fix mlx5_get_vector_affinity function
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'wireless-drivers-for-davem-2018-04-26' of git://git.kernel.org/pub/scm...
David S. Miller [Fri, 27 Apr 2018 17:48:37 +0000 (13:48 -0400)]
Merge tag 'wireless-drivers-for-davem-2018-04-26' of git://git./linux/kernel/git/kvalo/wireless-drivers

Kalle Valo says:

====================
wireless-drivers fixes for 4.17

A few fixes for 4.17 but nothing really special. The new ETSI WMM
parameter support for iwlwifi is not technically a bugfix but
important for regulatory compliance.

iwlwifi

* use new ETSI WMM parameters from regulatory database

* fix a regression with the older firmware API 31 (eg. 31.560484.0)

brcmfmac

* fix a double free in nvmam loading fails

rtlwifi

* yet another fix for ant_sel module parameter
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum_switchdev: Do not remove mrouter port from MDB's ports list
Ido Schimmel [Thu, 26 Apr 2018 08:46:29 +0000 (11:46 +0300)]
mlxsw: spectrum_switchdev: Do not remove mrouter port from MDB's ports list

When IGMP snooping is enabled on a bridge, traffic forwarded by an MDB
entry should be sent to both ports member in the MDB's ports list and
mrouter ports.

In case a port needs to be removed from an MDB's ports list, but this
port is also configured as an mrouter port, then do not update the
device so that it will continue to forward traffic through that port.

Fix a copy-paste error that checked that IGMP snooping is enabled twice
instead of checking the port's mrouter state.

Fixes: ded711c87a04 ("mlxsw: spectrum_switchdev: Consider mrouter status for mdb changes")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reported-by: Colin King <colin.king@canonical.com>
Reviewed-by: Nogah Frankel <nogahf@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: clear the new asoc's stream outcnt in sctp_stream_update
Xin Long [Thu, 26 Apr 2018 07:21:44 +0000 (15:21 +0800)]
sctp: clear the new asoc's stream outcnt in sctp_stream_update

When processing a duplicate cookie-echo chunk, sctp moves the new
temp asoc's stream out/in into the old asoc, and later frees this
new temp asoc.

But now after this move, the new temp asoc's stream->outcnt is not
cleared while stream->out is set to NULL, which would cause a same
crash as the one fixed in Commit 79d0895140e9 ("sctp: fix error
path in sctp_stream_init") when freeing this asoc later.

This fix is to clear this outcnt in sctp_stream_update.

Fixes: f952be79cebd ("sctp: introduce struct sctp_stream_out_ext")
Reported-by: Jianwen Ji <jiji@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoNET: usb: qmi_wwan: add support for ublox R410M PID 0x90b2
SZ Lin (林上智) [Thu, 26 Apr 2018 06:30:13 +0000 (14:30 +0800)]
NET: usb: qmi_wwan: add support for ublox R410M PID 0x90b2

This patch adds support for PID 0x90b2 of ublox R410M.

qmicli -d /dev/cdc-wdm0 --dms-get-manufacturer
[/dev/cdc-wdm0] Device manufacturer retrieved:
        Manufacturer: 'u-blox'

qmicli -d /dev/cdc-wdm0 --dms-get-model
[/dev/cdc-wdm0] Device model retrieved:
        Model: 'SARA-R410M-02B'

Signed-off-by: SZ Lin (林上智) <sz.lin@moxa.com>
Cc: stable <stable@vger.kernel.org>
Acked-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: handle two v4 addrs comparison in sctp_inet6_cmp_addr
Xin Long [Thu, 26 Apr 2018 06:13:57 +0000 (14:13 +0800)]
sctp: handle two v4 addrs comparison in sctp_inet6_cmp_addr

Since sctp ipv6 socket also supports v4 addrs, it's possible to
compare two v4 addrs in pf v6 .cmp_addr, sctp_inet6_cmp_addr.

However after Commit 1071ec9d453a ("sctp: do not check port in
sctp_inet6_cmp_addr"), it no longer calls af1->cmp_addr, which
in this case is sctp_v4_cmp_addr, but calls __sctp_v6_cmp_addr
where it handles them as two v6 addrs. It would cause a out of
bounds crash.

syzbot found this crash when trying to bind two v4 addrs to a
v6 socket.

This patch fixes it by adding the process for two v4 addrs in
sctp_inet6_cmp_addr.

Fixes: 1071ec9d453a ("sctp: do not check port in sctp_inet6_cmp_addr")
Reported-by: syzbot+cd494c1dd681d4d93ebb@syzkaller.appspotmail.com
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: systemport: Correclty disambiguate driver instances
Florian Fainelli [Wed, 25 Apr 2018 23:21:51 +0000 (16:21 -0700)]
net: systemport: Correclty disambiguate driver instances

While adding the DSA notifier, we will be sending DSA notifications with
info->master that is going to point to a particular net_device instance.

Our logic in bcm_sysport_map_queues() correctly disambiguates net_device
instances that are not covered by our own driver, but it will not make
sure that info->master points to a particular driver instance that we
are interested in. In a system where e.g: two or more SYSTEMPORT
instances are registered, this would lead in programming two or more
times the queue mapping, completely messing with the logic which does
the queue/port allocation and tracking.

Fix this by looking at the notifier_block pointer which is unique per
instance and allows us to go back to our driver private structure, and
in turn to the backing net_device instance.

Fixes: d156576362c0 ("net: systemport: Establish lower/upper queue mapping")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: net: add in_netns.sh TEST_GEN_PROGS_EXTENDED
Anders Roxell [Wed, 25 Apr 2018 21:32:06 +0000 (23:32 +0200)]
selftests: net: add in_netns.sh TEST_GEN_PROGS_EXTENDED

Script in_netns.sh is a utility function and not its own test so it
shouldn't be part of the TEST_PROGS. The in_netns.sh get used by
run_afpackettests.
To install in_netns.sh without being added to the main run_kselftest.sh
script use the TEST_GEN_PROGS_EXTENDED variable.

Fixes: 5ff9c1a3dd92 ("selftests: net: add in_netns.sh to TEST_PROGS")
Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: ignore Fast Open on repair mode
Yuchung Cheng [Wed, 25 Apr 2018 18:33:08 +0000 (11:33 -0700)]
tcp: ignore Fast Open on repair mode

The TCP repair sequence of operation is to first set the socket in
repair mode, then inject the TCP stats into the socket with repair
socket options, then call connect() to re-activate the socket. The
connect syscall simply returns and set state to ESTABLISHED
mode. As a result Fast Open is meaningless for TCP repair.

However allowing sendto() system call with MSG_FASTOPEN flag half-way
during the repair operation could unexpectedly cause data to be
sent, before the operation finishes changing the internal TCP stats
(e.g. MSS).  This in turn triggers TCP warnings on inconsistent
packet accounting.

The fix is to simply disallow Fast Open operation once the socket
is in the repair mode.

Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'mvpp2-fixes'
David S. Miller [Fri, 27 Apr 2018 15:22:56 +0000 (11:22 -0400)]
Merge branch 'mvpp2-fixes'

Maxime Chevallier says:

====================
net: mvpp2: Fix hangs when starting some interfaces on 7k/8k

Armada 7K / 8K clock management has recently been reworked, see :

commit c7e92def1ef4 ("clk: mvebu: cp110: Fix clock tree representation")

I have been experiencing overall system hangs on MacchiatoBin when starting
the eth1 interface since then. It turns out some clocks dependencies were
missing in the PPv2 and xmdio driver, the clock rework made this visible.

This is the V2 series, that adds support for the missing 'MG Core clock' in
mvpp2, and fixes an issue with the error path for the axi_clk.

Thanks to Gregory Clement for finding the root cause of this bug.

V2 : Remove all DT patches from this series, they will be merged through
     the mvebu tree.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: mvpp2: Fix clock resource by adding missing mg_core_clk
Maxime Chevallier [Wed, 25 Apr 2018 18:21:17 +0000 (20:21 +0200)]
net: mvpp2: Fix clock resource by adding missing mg_core_clk

Marvell's PPv2.2 IP needs an additional clock named "MG Core clock".
This is required on Armada 7K and 8K.

This commit adds the required clock in mvpp2, making sure it's only
used on PPv2.2.

Fixes: c7e92def1ef4 ("clk: mvebu: cp110: Fix clock tree representation")
Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: mvpp2: Fix clk error path in mvpp2_probe
Maxime Chevallier [Wed, 25 Apr 2018 18:21:16 +0000 (20:21 +0200)]
net: mvpp2: Fix clk error path in mvpp2_probe

When clk_prepare_enable fails for the axi_clk, the mg_clk isn't properly
cleaned up. Add another jump label to handle that case, and make sure we
jump to it in the later error cases.

Fixes: 4792ea04bcd0 ("net: mvpp2: Fix clock resource by adding an optional bus clock")
Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Acked-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: don't depend on eth_tbl being available
Jakub Kicinski [Wed, 25 Apr 2018 18:21:08 +0000 (11:21 -0700)]
nfp: don't depend on eth_tbl being available

For very very old generation of the management FW Ethernet port
information table may theoretically not be available.  This in
turn will cause the nfp_port structures to not be allocated.

Make sure we don't crash the kernel when there is no eth_tbl:

RIP: 0010:nfp_net_pci_probe+0xf2/0xb40 [nfp]
...
Call Trace:
  nfp_pci_probe+0x6de/0xab0 [nfp]
  local_pci_probe+0x47/0xa0
  work_for_cpu_fn+0x1a/0x30
  process_one_work+0x1de/0x3e0

Found while working with broken/development version of management FW.

Fixes: a5950182c00e ("nfp: map mac_stats and vf_cfg BARs")
Fixes: 93da7d9660ee ("nfp: provide nfp_port to of nfp_net_get_mac_addr()")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotipc: fix bug in function tipc_nl_node_dump_monitor
Jon Maloy [Wed, 25 Apr 2018 16:29:25 +0000 (18:29 +0200)]
tipc: fix bug in function tipc_nl_node_dump_monitor

Commit 36a50a989ee8 ("tipc: fix infinite loop when dumping link monitor
summary") intended to fix a problem with user tool looping when max
number of bearers are enabled.

Unfortunately, the wrong version of the commit was posted, so the
problem was not solved at all.

This commit adds the missing part.

Fixes: 36a50a989ee8 ("tipc: fix infinite loop when dumping link monitor summary")
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpf: fix uninitialized variable in bpf tools
John Fastabend [Wed, 25 Apr 2018 22:08:53 +0000 (15:08 -0700)]
bpf: fix uninitialized variable in bpf tools

Here the variable cont is used as the saved_pointer for a call to
strtok_r(). It is safe to use the value uninitialized in this
context however and the later reference is only ever used if
the strtok_r is successful. But, 'gcc-5' at least doesn't have all
this knowledge so initialize cont to NULL. Additionally, do the
natural NULL check before accessing just for completness.

The warning is the following:

./bpf/tools/bpf/bpf_dbg.c: In function ‘cmd_load’:
./bpf/tools/bpf/bpf_dbg.c:1077:13: warning: ‘cont’ may be used uninitialized in this function [-Wmaybe-uninitialized]
  } else if (matches(subcmd, "pcap") == 0) {

Fixes: fd981e3c321a "filter: bpf_dbg: add minimal bpf debugger"
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agonet/mlx5: Properly deal with flow counters when deleting rules
Chris Mi [Mon, 16 Apr 2018 01:36:51 +0000 (10:36 +0900)]
net/mlx5: Properly deal with flow counters when deleting rules

When deleting a flow counter, the modify mask should be the action and
the flow counter. Otherwise the flow counter is not deleted and we'll
get a firmware warning when deleting the remaining destinations on the
same FTE.

It only happens in the presence of flow counter and multiple vport
destinations. If there is only one vport destination, there is no
need to update the FTE when deleting the only vport destination,
we just delete the FTE.

Fixes: ae05831424ed ("net/mlx5: Add option to add fwd rule with counter")
Signed-off-by: Chris Mi <chrism@mellanox.com>
Signed-off-by: Jianbo Liu <jianbol@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5e: Fix traffic between VF and representor
Shahar Klein [Sun, 8 Apr 2018 06:50:53 +0000 (09:50 +0300)]
net/mlx5e: Fix traffic between VF and representor

After the cited commit, WQE RQ size is calculated based on sw_mtu but it
was not set for representors. This commit fixes that.

Fixes: 472a1e44b349 ("net/mlx5e: Save MTU in channels params")
Signed-off-by: Shahar Klein <shahark@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: Avoid cleaning flow steering table twice during error flow
Talat Batheesh [Sun, 15 Apr 2018 08:26:19 +0000 (11:26 +0300)]
net/mlx5: Avoid cleaning flow steering table twice during error flow

When we fail to initialize the RX root namespace, we need
to clean only that and not the entire flow steering.

Currently the code may try to clean the flow steering twice
on error witch leads to null pointer deference.
Make sure we clean correctly.

Fixes: fba53f7b5719 ("net/mlx5: Introduce mlx5_flow_steering structure")
Signed-off-by: Talat Batheesh <talatb@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5e: TX, Use correct counter in dma_map error flow
Tariq Toukan [Tue, 20 Mar 2018 16:17:25 +0000 (18:17 +0200)]
net/mlx5e: TX, Use correct counter in dma_map error flow

In case of a dma_mapping_error, do not use wi->num_dma
as a parameter for dma unmap function because it's yet
to be set, and holds an out-of-date value.
Use actual value (local variable num_dma) instead.

Fixes: 34802a42b352 ("net/mlx5e: Do not modify the TX SKB")
Fixes: e586b3b0baee ("net/mlx5: Ethernet Datapath files")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: Fix mlx5_get_vector_affinity function
Israel Rukshin [Thu, 12 Apr 2018 09:49:11 +0000 (09:49 +0000)]
net/mlx5: Fix mlx5_get_vector_affinity function

Adding the vector offset when calling to mlx5_vector2eqn() is wrong.
This is because mlx5_vector2eqn() checks if EQ index is equal to vector number
and the fact that the internal completion vectors that mlx5 allocates
don't get an EQ index.

The second problem here is that using effective_affinity_mask gives the same
CPU for different vectors.
This leads to unmapped queues when calling it from blk_mq_rdma_map_queues().
This doesn't happen when using affinity_hint mask.

Fixes: 2572cf57d75a ("mlx5: fix mlx5_get_vector_affinity to start from completion vector 0")
Fixes: 05e0cc84e00c ("net/mlx5: Fix get vector affinity helper function")
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
6 years agonet/mlx5e: DCBNL fix min inline header size for dscp
Huy Nguyen [Thu, 21 Dec 2017 21:15:24 +0000 (15:15 -0600)]
net/mlx5e: DCBNL fix min inline header size for dscp

When the trust state is set to dscp and the netdev is down, the inline
header size is not updated. When netdev is up, the inline header size
stays at L2 instead of IP.

Fix this issue by updating the private parameter when the netdev is in
down so that when netdev is up, it picks up the right header size.

Fixes: fbcb127e89ba ("net/mlx5e: Support DSCP trust state ...")
Signed-off-by: Huy Nguyen <huyn@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5e: Allow offloading ipv4 header re-write for icmp
Jianbo Liu [Tue, 27 Mar 2018 09:22:16 +0000 (09:22 +0000)]
net/mlx5e: Allow offloading ipv4 header re-write for icmp

For ICMPv4, the checksum is calculated from the ICMP headers and data.
Since the ICMPv4 checksum doesn't cover the IP header, we can allow to
do L3 header re-write for this protocol.

Fixes: bdd66ac0aeed ('net/mlx5e: Disallow TC offloading of unsupported match/action combinations')
Signed-off-by: Jianbo Liu <jianbol@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Thu, 26 Apr 2018 02:55:33 +0000 (22:55 -0400)]
Merge git://git./pub/scm/linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2018-04-25

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

The main changes are:

1) Fix to clear the percpu metadata_dst that could otherwise carry
   stale ip_tunnel_info, from William.

2) Fix that reduces the number of passes in x64 JIT with regards to
   dead code sanitation to avoid risk of prog rejection, from Gianluca.

3) Several fixes of sockmap programs, besides others, fixing a double
   page_put() in error path, missing refcount hold for pinned sockmap,
   adding required -target bpf for clang in sample Makefile, from John.

4) Fix to disable preemption in __BPF_PROG_RUN_ARRAY() paths, from Roman.

5) Fix tools/bpf/ Makefile with regards to a lex/yacc build error
   seen on older gcc-5, from John.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpf: fix for lex/yacc build error with gcc-5
John Fastabend [Wed, 25 Apr 2018 21:22:45 +0000 (14:22 -0700)]
bpf: fix for lex/yacc build error with gcc-5

Fix build error found with Ubuntu shipped gcc-5

~/git/bpf/tools/bpf$ make all

Auto-detecting system features:
...                        libbfd: [ OFF ]
...        disassembler-four-args: [ OFF ]

  CC       bpf_jit_disasm.o
  LINK     bpf_jit_disasm
  CC       bpf_dbg.o
/home/john/git/bpf/tools/bpf/bpf_dbg.c: In function ‘cmd_load’:
/home/john/git/bpf/tools/bpf/bpf_dbg.c:1077:13: warning: ‘cont’ may be used uninitialized in this function [-Wmaybe-uninitialized]
  } else if (matches(subcmd, "pcap") == 0) {
             ^
  LINK     bpf_dbg
  CC       bpf_asm.o
make: *** No rule to make target `bpf_exp.yacc.o', needed by `bpf_asm'.  Stop.

Fixes: 5a8997f20715 ("tools: bpf: respect output directory during build")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agords: ib: Fix missing call to rds_ib_dev_put in rds_ib_setup_qp
Dag Moxnes [Wed, 25 Apr 2018 11:22:01 +0000 (13:22 +0200)]
rds: ib: Fix missing call to rds_ib_dev_put in rds_ib_setup_qp

The function rds_ib_setup_qp is calling rds_ib_get_client_data and
should correspondingly call rds_ib_dev_put. This call was lost in
the non-error path with the introduction of error handling done in
commit 3b12f73a5c29 ("rds: ib: add error handle")

Signed-off-by: Dag Moxnes <dag.moxnes@oracle.com>
Reviewed-by: Håkon Bugge <haakon.bugge@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/smc: keep clcsock reference in smc_tcp_listen_work()
Ursula Braun [Wed, 25 Apr 2018 10:48:58 +0000 (12:48 +0200)]
net/smc: keep clcsock reference in smc_tcp_listen_work()

The internal CLC socket should exist till the SMC-socket is released.
Function tcp_listen_worker() releases the internal CLC socket of a
listen socket, if an smc_close_active() is called. This function
is called for the final release(), but it is called for shutdown
SHUT_RDWR as well. This opens a door for protection faults, if
socket calls using the internal CLC socket are called for a
shutdown listen socket.

With the changes of
commit 3d502067599f ("net/smc: simplify wait when closing listen socket")
there is no need anymore to release the internal CLC socket in
function tcp_listen_worker((). It is sufficient to release it in
smc_release().

Fixes: 127f49705823 ("net/smc: release clcsock from tcp_listen_worker")
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Reported-by: syzbot+9045fc589fcd196ef522@syzkaller.appspotmail.com
Reported-by: syzbot+28a2c86cf19c81d871fa@syzkaller.appspotmail.com
Reported-by: syzbot+9605e6cace1b5efd4a0a@syzkaller.appspotmail.com
Reported-by: syzbot+cf9012c597c8379d535c@syzkaller.appspotmail.com
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: allow scanning busses with missing phys
Alexandre Belloni [Tue, 24 Apr 2018 16:09:04 +0000 (18:09 +0200)]
net: phy: allow scanning busses with missing phys

Some MDIO busses will error out when trying to read a phy address with no
phy present at that address. In that case, probing the bus will fail
because __mdiobus_register() is scanning the bus for all possible phys
addresses.

In case MII_PHYSID1 returns -EIO or -ENODEV, consider there is no phy at
this address and set the phy ID to 0xffffffff which is then properly
handled in get_phy_device().

Suggested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpf, x64: fix JIT emission for dead code
Gianluca Borello [Wed, 25 Apr 2018 05:42:16 +0000 (05:42 +0000)]
bpf, x64: fix JIT emission for dead code

Commit 2a5418a13fcf ("bpf: improve dead code sanitizing") replaced dead
code with a series of ja-1 instructions, for safety. That made JIT
compilation much more complex for some BPF programs. One instance of such
programs is, for example:

bool flag = false
...
/* A bunch of other code */
...
if (flag)
        do_something()

In some cases llvm is not able to remove at compile time the code for
do_something(), so the generated BPF program ends up with a large amount
of dead instructions. In one specific real life example, there are two
series of ~500 and ~1000 dead instructions in the program. When the
verifier replaces them with a series of ja-1 instructions, it causes an
interesting behavior at JIT time.

During the first pass, since all the instructions are estimated at 64
bytes, the ja-1 instructions end up being translated as 5 bytes JMP
instructions (0xE9), since the jump offsets become increasingly large (>
127) as each instruction gets discovered to be 5 bytes instead of the
estimated 64.

Starting from the second pass, the first N instructions of the ja-1
sequence get translated into 2 bytes JMPs (0xEB) because the jump offsets
become <= 127 this time. In particular, N is defined as roughly 127 / (5
- 2) ~= 42. So, each further pass will make the subsequent N JMP
instructions shrink from 5 to 2 bytes, making the image shrink every time.
This means that in order to have the entire program converge, there need
to be, in the real example above, at least ~1000 / 42 ~= 24 passes just
for translating the dead code. If we add this number to the passes needed
to translate the other non dead code, it brings such program to 40+
passes, and JIT doesn't complete. Ultimately the userspace loader fails
because such BPF program was supposed to be part of a prog array owner
being JITed.

While it is certainly possible to try to refactor such programs to help
the compiler remove dead code, the behavior is not really intuitive and it
puts further burden on the BPF developer who is not expecting such
behavior. To make things worse, such programs are working just fine in all
the kernel releases prior to the ja-1 fix.

A possible approach to mitigate this behavior consists into noticing that
for ja-1 instructions we don't really need to rely on the estimated size
of the previous and current instructions, we know that a -1 BPF jump
offset can be safely translated into a 0xEB instruction with a jump offset
of -2.

Such fix brings the BPF program in the previous example to complete again
in ~9 passes.

Fixes: 2a5418a13fcf ("bpf: improve dead code sanitizing")
Signed-off-by: Gianluca Borello <g.borello@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agobpf: clear the ip_tunnel_info.
William Tu [Wed, 25 Apr 2018 06:46:59 +0000 (23:46 -0700)]
bpf: clear the ip_tunnel_info.

The percpu metadata_dst might carry the stale ip_tunnel_info
and cause incorrect behavior.  When mixing tests using ipv4/ipv6
bpf vxlan and geneve tunnel, the ipv6 tunnel info incorrectly uses
ipv4's src ip addr as its ipv6 src address, because the previous
tunnel info does not clean up.  The patch zeros the fields in
ip_tunnel_info.

Signed-off-by: William Tu <u9012063@gmail.com>
Reported-by: Yifeng Sun <pkusunyifeng@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agoMerge branch 'userns-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
Linus Torvalds [Wed, 25 Apr 2018 00:58:51 +0000 (17:58 -0700)]
Merge branch 'userns-linus' of git://git./linux/kernel/git/ebiederm/user-namespace

Pull userns bug fix from Eric Biederman:
 "Just a small fix to properly set the return code on error"

* 'userns-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
  commoncap: Handle memory allocation failure.

6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Tue, 24 Apr 2018 21:16:40 +0000 (14:16 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Fix rtnl deadlock in ipvs, from Julian Anastasov.

 2) s390 qeth fixes from Julian Wiedmann (control IO completion stalls,
    bad MAC address update sequence, request side races on command IO
    timeouts).

 3) Handle seq_file overflow properly in l2tp, from Guillaume Nault.

 4) Fix VLAN priority mappings in cpsw driver, from Ivan Khoronzhuk.

 5) Packet scheduler ife action fixes (malformed TLV lengths, etc.) from
    Alexander Aring.

 6) Fix out of bounds access in tcp md5 option parser, from Jann Horn.

 7) Missing netlink attribute policies in rtm_ipv6_policy table, from
    Eric Dumazet.

 8) Missing socket address length checks in l2tp and pppoe connect, from
    Guillaume Nault.

 9) Fix netconsole over team and bonding, from Xin Long.

10) Fix race with AF_PACKET socket state bitfields, from Willem de
    Bruijn.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (51 commits)
  ice: Fix insufficient memory issue in ice_aq_manage_mac_read
  sfc: ARFS filter IDs
  net: ethtool: Add missing kernel doc for FEC parameters
  packet: fix bitfield update race
  ice: Do not check INTEVENT bit for OICR interrupts
  ice: Fix incorrect comment for action type
  ice: Fix initialization for num_nodes_added
  igb: Fix the transmission mode of queue 0 for Qav mode
  ixgbevf: ensure xdp_ring resources are free'd on error exit
  team: fix netconsole setup over team
  amd-xgbe: Only use the SFP supported transceiver signals
  amd-xgbe: Improve KR auto-negotiation and training
  amd-xgbe: Add pre/post auto-negotiation phy hooks
  pppoe: check sockaddr length in pppoe_connect()
  l2tp: check sockaddr length in pppol2tp_connect()
  net: phy: marvell: clear wol event before setting it
  ipv6: add RTA_TABLE and RTA_PREFSRC to rtm_ipv6_policy
  bonding: do not set slave_dev npinfo before slave_enable_netpoll in bond_enslave
  tcp: don't read out-of-bounds opsize
  ibmvnic: Clean actual number of RX or TX pools
  ...

6 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net...
David S. Miller [Tue, 24 Apr 2018 20:17:59 +0000 (16:17 -0400)]
Merge branch '1GbE' of git://git./linux/kernel/git/jkirsher/net-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2018-04-24

This series contains fixes to ixgbevf, igb and ice drivers.

Colin Ian King fixes the return value on error for the new XDP support
that went into ixgbevf for 4.17.

Vinicius provides a fix for queue 0 for igb, which was not receiving all
the credits it needed when QAV mode was enabled.

Anirudh provides several fixes for the new ice driver, starting with
properly initializing num_nodes_added to zero.  Fixed up a code comment
to better reflect what is really going on in the code.  Fixed how to
detect if an OICR interrupt has occurred to a more reliable method.

Md Fahad fixes the ice driver to allocate the right amount of memory
when reading and storing the devices MAC addresses.  The device can have
up to 2 MAC addresses (LAN and WoL), while WoL is currently not
supported, we need to ensure it can be properly handled when support is
added.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoice: Fix insufficient memory issue in ice_aq_manage_mac_read
Md Fahad Iqbal Polash [Mon, 16 Apr 2018 17:07:03 +0000 (10:07 -0700)]
ice: Fix insufficient memory issue in ice_aq_manage_mac_read

For the MAC read operation, the device can return up to two (LAN and WoL)
MAC addresses. Without access to adequate memory, the device will return
an error. Fixed this by allocating the right amount of memory. Also, logic
to detect and copy the LAN MAC address into the port_info structure has
been added. Note that the WoL MAC address is ignored currently as the WoL
feature isn't supported yet.

Fixes: dc49c7723676 ("ice: Get MAC/PHY/link info and scheduler topology")
Signed-off-by: Md Fahad Iqbal Polash <md.fahad.iqbal.polash@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agosfc: ARFS filter IDs
Edward Cree [Tue, 24 Apr 2018 16:09:30 +0000 (17:09 +0100)]
sfc: ARFS filter IDs

Associate an arbitrary ID with each ARFS filter, allowing to properly query
 for expiry.  The association is maintained in a hash table, which is
 protected by a spinlock.

v3: fix build warnings when CONFIG_RFS_ACCEL is disabled (thanks lkp-robot).
v2: fixed uninitialised variable (thanks davem and lkp-robot).

Fixes: 3af0f34290f6 ("sfc: replace asynchronous filter operations")
Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethtool: Add missing kernel doc for FEC parameters
Florian Fainelli [Mon, 23 Apr 2018 22:51:38 +0000 (15:51 -0700)]
net: ethtool: Add missing kernel doc for FEC parameters

While adding support for ethtool::get_fecparam and set_fecparam, kernel
doc for these functions was missed, add those.

Fixes: 1a5f3da20bd9 ("net: ethtool: add support for forward error correction modes")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agopacket: fix bitfield update race
Willem de Bruijn [Mon, 23 Apr 2018 21:37:03 +0000 (17:37 -0400)]
packet: fix bitfield update race

Updates to the bitfields in struct packet_sock are not atomic.
Serialize these read-modify-write cycles.

Move po->running into a separate variable. Its writes are protected by
po->bind_lock (except for one startup case at packet_create). Also
replace a textual precondition warning with lockdep annotation.

All others are set only in packet_setsockopt. Serialize these
updates by holding the socket lock. Analogous to other field updates,
also hold the lock when testing whether a ring is active (pg_vec).

Fixes: 8dc419447415 ("[PACKET]: Add optional checksum computation for recvmsg")
Reported-by: DaeRyong Jeong <threeearcat@gmail.com>
Reported-by: Byoungyoung Lee <byoungyoung@purdue.edu>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoice: Do not check INTEVENT bit for OICR interrupts
Ben Shelton [Wed, 11 Apr 2018 19:21:33 +0000 (12:21 -0700)]
ice: Do not check INTEVENT bit for OICR interrupts

According to the hardware spec, checking the INTEVENT bit isn't a
reliable way to detect if an OICR interrupt has occurred. This is
because this bit can be cleared by the hardware/firmware before the
interrupt service routine has run. So instead, just check for OICR
events every time.

Fixes: 940b61af02f4 ("ice: Initialize PF and setup miscellaneous interrupt")
Signed-off-by: Ben Shelton <benjamin.h.shelton@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoice: Fix incorrect comment for action type
Anirudh Venkataramanan [Wed, 11 Apr 2018 17:41:47 +0000 (10:41 -0700)]
ice: Fix incorrect comment for action type

Action type 5 defines large action generic values. Fix comment to
reflect that better.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoice: Fix initialization for num_nodes_added
Anirudh Venkataramanan [Tue, 10 Apr 2018 17:49:49 +0000 (10:49 -0700)]
ice: Fix initialization for num_nodes_added

ice_sched_add_nodes_to_layer is used recursively, and so we start
with num_nodes_added being 0. This way, in case of an error or if
num_nodes is NULL, the function just returns 0 to indicate that no
nodes were added.

Fixes: 5513b920a4f7 ("ice: Update Tx scheduler tree for VSI multi-Tx queue support")
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoigb: Fix the transmission mode of queue 0 for Qav mode
Vinicius Costa Gomes [Sat, 31 Mar 2018 00:06:52 +0000 (17:06 -0700)]
igb: Fix the transmission mode of queue 0 for Qav mode

When Qav mode is enabled, queue 0 should be kept on Stream Reservation
mode. From the i210 datasheet, section 8.12.19:

"Note: Queue0 QueueMode must be set to 1b when TransmitMode is set to
Qav." ("QueueMode 1b" represents the Stream Reservation mode)

The solution is to give queue 0 the all the credits it might need, so
it has priority over queue 1.

A situation where this can happen is when cbs is "installed" only on
queue 1, leaving queue 0 alone. For example:

$ tc qdisc replace dev enp2s0 handle 100: parent root mqprio num_tc 3 \
         map 2 2 1 0 2 2 2 2 2 2 2 2 2 2 2 2 queues 1@0 1@1 2@2 hw 0

$ tc qdisc replace dev enp2s0 parent 100:2 cbs locredit -1470 \
         hicredit 30 sendslope -980000 idleslope 20000 offload 1

Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoixgbevf: ensure xdp_ring resources are free'd on error exit
Colin Ian King [Tue, 27 Mar 2018 14:21:48 +0000 (15:21 +0100)]
ixgbevf: ensure xdp_ring resources are free'd on error exit

The current error handling for failed resource setup for xdp_ring
data is a break out of the loop and returning 0 indicated everything
was OK, when in fact it is not.  Fix this by exiting via the
error exit label err_setup_tx that will clean up the resources
correctly and return and error status.

Detected by CoverityScan, CID#1466879 ("Logically dead code")

Fixes: 21092e9ce8b1 ("ixgbevf: Add support for XDP_TX action")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
6 years agoteam: fix netconsole setup over team
Xin Long [Tue, 24 Apr 2018 06:33:37 +0000 (14:33 +0800)]
team: fix netconsole setup over team

The same fix in Commit dbe173079ab5 ("bridge: fix netconsole
setup over bridge") is also needed for team driver.

While at it, remove the unnecessary parameter *team from
team_port_enable_netpoll().

v1->v2:
  - fix it in a better way, as does bridge.

Fixes: 0fb52a27a04a ("team: cleanup netpoll clode")
Reported-by: João Avelino Bellomo Filho <jbellomo@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agortlwifi: cleanup 8723be ant_sel definition
Ping-Ke Shih [Fri, 20 Apr 2018 02:30:09 +0000 (10:30 +0800)]
rtlwifi: cleanup 8723be ant_sel definition

Some HP laptops have only a single wifi antenna. This would not be a
problem except that they were shipped with an incorrectly encoded
EFUSE. It should have been possible to open the computer and transfer
the antenna connection to the other terminal except that such action
might void the warranty, and moving the antenna broke the Windows
driver. The fix was to add a module option that would override the
EFUSE encoding. That was done with commit c18d8f509571 ("rtlwifi:
rtl8723be: Add antenna select module parameter"). There was still a
problem with Bluetooth coexistence, which was addressed with commit
baa170229095 ("rtlwifi: btcoexist: Implement antenna selection").
There were still problems, thus there were commit 0ff78adeef11
("rtlwifi: rtl8723be: fix ant_sel code") and commit 6d6226928369
("rtlwifi: btcoexist: Fix antenna selection code"). Despite all these
attempts at fixing the problem, the code is not yet right. A proper
fix is important as there are now instances of laptops having
RTL8723DE chips with the same problem.

The module parameter ant_sel is used to control antenna number and path.
At present enum ANT_{X2,X1} is used to define the antenna number, but
this choice is not intuitive, thus change to a new enum ANT_{MAIN,AUX}
to make it more readable. This change showed examples where incorrect
values were used. It was also possible to remove a workaround in
halbtcoutsrc.c.

The experimental results with single antenna connected to specific path
are now as follows:
  ant_sel  ANT_MAIN(#1)  ANT_AUX(#2)
     0        -8            -62
     1        -62           -10
     2        -6            -60

Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Fixes: c18d8f509571 ("rtlwifi: rtl8723be: Add antenna select module parameter")
Fixes: baa170229095 ("rtlwifi: btcoexist: Implement antenna selection")
Fixes: 0ff78adeef11 ("rtlwifi: rtl8723be: fix ant_sel code")
Fixes: 6d6226928369 ("rtlwifi: btcoexist: Fix antenna selection code")
Cc: Stable <stable@vger.kernel.org> # 4.7+
Reviewed-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
6 years agoiwlwifi: mvm: fix old scan version sizes
Luca Coelho [Mon, 23 Apr 2018 13:01:31 +0000 (16:01 +0300)]
iwlwifi: mvm: fix old scan version sizes

When version 8 of the scan command API was introduced, only the size
of version 7 was updated, causing older versions of the firmware to
throw BAD_COMMAND errors.

Calculating the old version based on the size of the latest version
got too complicated and the size of the older versions will never
change anyway, so it's better to just hardcoded the sizes.

Fixes: 66fa2424df16 ("iwlwifi: fw api: support the new scan request FW API version")
Reported-by: Scott Register <sreg@sreg.io>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
6 years agoMerge branch 'amd-xgbe-fixes'
David S. Miller [Tue, 24 Apr 2018 01:24:23 +0000 (21:24 -0400)]
Merge branch 'amd-xgbe-fixes'

aTom Lendacky says:

====================
amd-xgbe: AMD XGBE driver fixes 2018-04-23

This patch series addresses some issues in the AMD XGBE driver.

The following fixes are included in this driver update series:

- Improve KR auto-negotiation and training (2 patches)
  - Add pre and post auto-negotiation hooks
  - Use the pre and post auto-negotiation hooks to disable CDR tracking
    during auto-negotiation page exchange in KR mode
- Check for SFP tranceiver signal support and only use the signal if the
  SFP indicates that it is supported

This patch series is based on net.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoamd-xgbe: Only use the SFP supported transceiver signals
Tom Lendacky [Mon, 23 Apr 2018 16:43:34 +0000 (11:43 -0500)]
amd-xgbe: Only use the SFP supported transceiver signals

The SFP eeprom indicates the transceiver signals (Rx LOS, Tx Fault, etc.)
that it supports.  Update the driver to include checking the eeprom data
when deciding whether to use a transceiver signal.

Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoamd-xgbe: Improve KR auto-negotiation and training
Tom Lendacky [Mon, 23 Apr 2018 16:43:17 +0000 (11:43 -0500)]
amd-xgbe: Improve KR auto-negotiation and training

Update xgbe-phy-v2.c to make use of the auto-negotiation (AN) phy hooks
to improve the ability to successfully complete Clause 73 AN when running
at 10gbps.  Hardware can sometimes have issues with CDR lock when the
AN DME page exchange is being performed.

The AN and KR training hooks are used as follows:
- The pre AN hook is used to disable CDR tracking in the PHY so that the
  DME page exchange can be successfully and consistently completed.
- The post KR training hook is used to re-enable the CDR tracking so that
  KR training can successfully complete.
- The post AN hook is used to check for an unsuccessful AN which will
  increase a CDR tracking enablement delay (up to a maximum value).

Add two debugfs entries to allow control over use of the CDR tracking
workaround.  The debugfs entries allow the CDR tracking workaround to
be disabled and determine whether to re-enable CDR tracking before or
after link training has been initiated.

Also, with these changes the receiver reset cycle that is performed during
the link status check can be performed less often.

Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoamd-xgbe: Add pre/post auto-negotiation phy hooks
Tom Lendacky [Mon, 23 Apr 2018 16:43:08 +0000 (11:43 -0500)]
amd-xgbe: Add pre/post auto-negotiation phy hooks

Add hooks to the driver auto-negotiation (AN) flow to allow the different
phy implementations to perform any steps necessary to improve AN.

Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agopppoe: check sockaddr length in pppoe_connect()
Guillaume Nault [Mon, 23 Apr 2018 14:38:27 +0000 (16:38 +0200)]
pppoe: check sockaddr length in pppoe_connect()

We must validate sockaddr_len, otherwise userspace can pass fewer data
than we expect and we end up accessing invalid data.

Fixes: 224cf5ad14c0 ("ppp: Move the PPP drivers")
Reported-by: syzbot+4f03bdf92fdf9ef5ddab@syzkaller.appspotmail.com
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agol2tp: check sockaddr length in pppol2tp_connect()
Guillaume Nault [Mon, 23 Apr 2018 14:15:14 +0000 (16:15 +0200)]
l2tp: check sockaddr length in pppol2tp_connect()

Check sockaddr_len before dereferencing sp->sa_protocol, to ensure that
it actually points to valid data.

Fixes: fd558d186df2 ("l2tp: Split pppol2tp patch into separate l2tp and ppp parts")
Reported-by: syzbot+a70ac890b23b1bf29f5c@syzkaller.appspotmail.com
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: marvell: clear wol event before setting it
Jingju Hou [Mon, 23 Apr 2018 07:22:49 +0000 (15:22 +0800)]
net: phy: marvell: clear wol event before setting it

If WOL event happened once, the LED[2] interrupt pin will not be
cleared unless we read the CSISR register. If interrupts are in use,
the normal interrupt handling will clear the WOL event. Let's clear the
WOL event before enabling it if !phy_interrupt_is_valid().

Signed-off-by: Jingju Hou <Jingju.Hou@synaptics.com>
Signed-off-by: Jisheng Zhang <Jisheng.Zhang@synaptics.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'bpf-sockmap-fixes'
Daniel Borkmann [Mon, 23 Apr 2018 22:49:46 +0000 (00:49 +0200)]
Merge branch 'bpf-sockmap-fixes'

John Fastabend says:

====================
While testing sockmap with more programs (besides our test programs)
I found a couple issues.

The attached series fixes an issue where pinned maps were not
working correctly, blocking sockets returned zero, and an error
path that when the sock hit an out of memory case resulted in a
double page_put() while doing ingress redirects.

See individual patches for more details.

v2: Incorporated Daniel's feedback to use map ops for uref put op
    which also fixed the build error discovered in v1.
v3: rename map_put_uref to map_release_uref
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agobpf: sockmap, fix double page_put on ENOMEM error in redirect path
John Fastabend [Mon, 23 Apr 2018 22:39:33 +0000 (15:39 -0700)]
bpf: sockmap, fix double page_put on ENOMEM error in redirect path

In the case where the socket memory boundary is hit the redirect
path returns an ENOMEM error. However, before checking for this
condition the redirect scatterlist buffer is setup with a valid
page and length. This is never unwound so when the buffers are
released latter in the error path we do a put_page() and clear
the scatterlist fields. But, because the initial error happens
before completing the scatterlist buffer we end up with both the
original buffer and the redirect buffer pointing to the same page
resulting in duplicate put_page() calls.

To fix this simply move the initial configuration of the redirect
scatterlist buffer below the sock memory check.

Found this while running TCP_STREAM test with netperf using Cilium.

Fixes: fa246693a111 ("bpf: sockmap, BPF_F_INGRESS flag for BPF_SK_SKB_STREAM_VERDICT")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agobpf: sockmap, sk_wait_event needed to handle blocking cases
John Fastabend [Mon, 23 Apr 2018 22:39:28 +0000 (15:39 -0700)]
bpf: sockmap, sk_wait_event needed to handle blocking cases

In the recvmsg handler we need to add a wait event to support the
blocking use cases. Without this we return zero and may confuse
user applications. In the wait event any data received on the
sk either via sk_receive_queue or the psock ingress list will
wake up the sock.

Fixes: fa246693a111 ("bpf: sockmap, BPF_F_INGRESS flag for BPF_SK_SKB_STREAM_VERDICT")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agobpf: sockmap, map_release does not hold refcnt for pinned maps
John Fastabend [Mon, 23 Apr 2018 22:39:23 +0000 (15:39 -0700)]
bpf: sockmap, map_release does not hold refcnt for pinned maps

Relying on map_release hook to decrement the reference counts when a
map is removed only works if the map is not being pinned. In the
pinned case the ref is decremented immediately and the BPF programs
released. After this BPF programs may not be in-use which is not
what the user would expect.

This patch moves the release logic into bpf_map_put_uref() and brings
sockmap in-line with how a similar case is handled in prog array maps.

Fixes: 3d9e952697de ("bpf: sockmap, fix leaking maps with attached but not detached progs")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agobpf: sockmap sample use clang flag, -target bpf
John Fastabend [Mon, 23 Apr 2018 19:11:08 +0000 (12:11 -0700)]
bpf: sockmap sample use clang flag, -target bpf

Per Documentation/bpf/bpf_devel_QA.txt add the -target flag to the
sockmap Makefile. Relevant text quoted here,

   Otherwise, you can use bpf target. Additionally, you _must_ use
   bpf target when:

 - Your program uses data structures with pointer or long / unsigned
   long types that interface with BPF helpers or context data
   structures. Access into these structures is verified by the BPF
   verifier and may result in verification failures if the native
   architecture is not aligned with the BPF architecture, e.g. 64-bit.
   An example of this is BPF_PROG_TYPE_SK_MSG require '-target bpf'

Fixes: 69e8cc134bcb ("bpf: sockmap sample program")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agobpf: Document sockmap '-target bpf' requirement for PROG_TYPE_SK_MSG
John Fastabend [Mon, 23 Apr 2018 19:11:02 +0000 (12:11 -0700)]
bpf: Document sockmap '-target bpf' requirement for PROG_TYPE_SK_MSG

BPF_PROG_TYPE_SK_MSG programs use a 'void *' for both data and the
data_end pointers. Additionally, the verifier ensures that every
accesses into the values is a __u64 read. This correctly maps on
to the BPF 64-bit architecture.

However, to ensure that when building on 32bit architectures that
clang uses correct types the '-target bpf' option _must_ be
specified. To make this clear add a note to the Documentation.

Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agobpf: disable and restore preemption in __BPF_PROG_RUN_ARRAY
Roman Gushchin [Mon, 23 Apr 2018 17:09:21 +0000 (18:09 +0100)]
bpf: disable and restore preemption in __BPF_PROG_RUN_ARRAY

Running bpf programs requires disabled preemption,
however at least some* of the BPF_PROG_RUN_ARRAY users
do not follow this rule.

To fix this bug, and also to make it not happen in the future,
let's add explicit preemption disabling/re-enabling
to the __BPF_PROG_RUN_ARRAY code.

* for example:
 [   17.624472] RIP: 0010:__cgroup_bpf_run_filter_sk+0x1c4/0x1d0
 ...
 [   17.640890]  inet6_create+0x3eb/0x520
 [   17.641405]  __sock_create+0x242/0x340
 [   17.641939]  __sys_socket+0x57/0xe0
 [   17.642370]  ? trace_hardirqs_off_thunk+0x1a/0x1c
 [   17.642944]  SyS_socket+0xa/0x10
 [   17.643357]  do_syscall_64+0x79/0x220
 [   17.643879]  entry_SYSCALL_64_after_hwframe+0x42/0xb7

Signed-off-by: Roman Gushchin <guro@fb.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf
David S. Miller [Mon, 23 Apr 2018 20:22:24 +0000 (16:22 -0400)]
Merge git://git./pub/scm/linux/kernel/git/pablo/nf

Pablo Neira Ayuso says:

====================
Netfilter/IPVS fixes for net

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

1) Fix SIP conntrack with phones sending session descriptions for different
   media types but same port numbers, from Florian Westphal.

2) Fix incorrect rtnl_lock mutex logic from IPVS sync thread, from Julian
   Anastasov.

3) Skip compat array allocation in ebtables if there is no entries, also
   from Florian.

4) Do not lose left/right bits when shifting marks from xt_connmark, from
   Jack Ma.

5) Silence false positive memleak in conntrack extensions, from Cong Wang.

6) Fix CONFIG_NF_REJECT_IPV6=m link problems, from Arnd Bergmann.

7) Cannot kfree rule that is already in list in nf_tables, switch order
   so this error handling is not required, from Florian Westphal.

8) Release set name in error path, from Florian.

9) include kmemleak.h in nf_conntrack_extend.c, from Stepheh Rothwell.

10) NAT chain and extensions depend on NF_TABLES.

11) Out of bound access when renaming chains, from Taehee Yoo.

12) Incorrect casting in xt_connmark leads to wrong bitshifting.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: add RTA_TABLE and RTA_PREFSRC to rtm_ipv6_policy
Eric Dumazet [Mon, 23 Apr 2018 01:29:23 +0000 (18:29 -0700)]
ipv6: add RTA_TABLE and RTA_PREFSRC to rtm_ipv6_policy

KMSAN reported use of uninit-value that I tracked to lack
of proper size check on RTA_TABLE attribute.

I also believe RTA_PREFSRC lacks a similar check.

Fixes: 86872cb57925 ("[IPv6] route: FIB6 configuration using struct fib6_config")
Fixes: c3968a857a6b ("ipv6: RTA_PREFSRC support for ipv6 route source address selection")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobonding: do not set slave_dev npinfo before slave_enable_netpoll in bond_enslave
Xin Long [Sun, 22 Apr 2018 11:11:50 +0000 (19:11 +0800)]
bonding: do not set slave_dev npinfo before slave_enable_netpoll in bond_enslave

After Commit 8a8efa22f51b ("bonding: sync netpoll code with bridge"), it
would set slave_dev npinfo in slave_enable_netpoll when enslaving a dev
if bond->dev->npinfo was set.

However now slave_dev npinfo is set with bond->dev->npinfo before calling
slave_enable_netpoll. With slave_dev npinfo set, __netpoll_setup called
in slave_enable_netpoll will not call slave dev's .ndo_netpoll_setup().
It causes that the lower dev of this slave dev can't set its npinfo.

One way to reproduce it:

  # modprobe bonding
  # brctl addbr br0
  # brctl addif br0 eth1
  # ifconfig bond0 192.168.122.1/24 up
  # ifenslave bond0 eth2
  # systemctl restart netconsole
  # ifenslave bond0 br0
  # ifconfig eth2 down
  # systemctl restart netconsole

The netpoll won't really work.

This patch is to remove that slave_dev npinfo setting in bond_enslave().

Fixes: 8a8efa22f51b ("bonding: sync netpoll code with bridge")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: don't read out-of-bounds opsize
Jann Horn [Fri, 20 Apr 2018 13:57:30 +0000 (15:57 +0200)]
tcp: don't read out-of-bounds opsize

The old code reads the "opsize" variable from out-of-bounds memory (first
byte behind the segment) if a broken TCP segment ends directly after an
opcode that is neither EOL nor NOP.

The result of the read isn't used for anything, so the worst thing that
could theoretically happen is a pagefault; and since the physmap is usually
mostly contiguous, even that seems pretty unlikely.

The following C reproducer triggers the uninitialized read - however, you
can't actually see anything happen unless you put something like a
pr_warn() in tcp_parse_md5sig_option() to print the opsize.

====================================
#define _GNU_SOURCE
#include <arpa/inet.h>
#include <stdlib.h>
#include <errno.h>
#include <stdarg.h>
#include <net/if.h>
#include <linux/if.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/in.h>
#include <linux/if_tun.h>
#include <err.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <string.h>
#include <stdio.h>
#include <unistd.h>
#include <sys/ioctl.h>
#include <assert.h>

void systemf(const char *command, ...) {
  char *full_command;
  va_list ap;
  va_start(ap, command);
  if (vasprintf(&full_command, command, ap) == -1)
    err(1, "vasprintf");
  va_end(ap);
  printf("systemf: <<<%s>>>\n", full_command);
  system(full_command);
}

char *devname;

int tun_alloc(char *name) {
  int fd = open("/dev/net/tun", O_RDWR);
  if (fd == -1)
    err(1, "open tun dev");
  static struct ifreq req = { .ifr_flags = IFF_TUN|IFF_NO_PI };
  strcpy(req.ifr_name, name);
  if (ioctl(fd, TUNSETIFF, &req))
    err(1, "TUNSETIFF");
  devname = req.ifr_name;
  printf("device name: %s\n", devname);
  return fd;
}

#define IPADDR(a,b,c,d) (((a)<<0)+((b)<<8)+((c)<<16)+((d)<<24))

void sum_accumulate(unsigned int *sum, void *data, int len) {
  assert((len&2)==0);
  for (int i=0; i<len/2; i++) {
    *sum += ntohs(((unsigned short *)data)[i]);
  }
}

unsigned short sum_final(unsigned int sum) {
  sum = (sum >> 16) + (sum & 0xffff);
  sum = (sum >> 16) + (sum & 0xffff);
  return htons(~sum);
}

void fix_ip_sum(struct iphdr *ip) {
  unsigned int sum = 0;
  sum_accumulate(&sum, ip, sizeof(*ip));
  ip->check = sum_final(sum);
}

void fix_tcp_sum(struct iphdr *ip, struct tcphdr *tcp) {
  unsigned int sum = 0;
  struct {
    unsigned int saddr;
    unsigned int daddr;
    unsigned char pad;
    unsigned char proto_num;
    unsigned short tcp_len;
  } fakehdr = {
    .saddr = ip->saddr,
    .daddr = ip->daddr,
    .proto_num = ip->protocol,
    .tcp_len = htons(ntohs(ip->tot_len) - ip->ihl*4)
  };
  sum_accumulate(&sum, &fakehdr, sizeof(fakehdr));
  sum_accumulate(&sum, tcp, tcp->doff*4);
  tcp->check = sum_final(sum);
}

int main(void) {
  int tun_fd = tun_alloc("inject_dev%d");
  systemf("ip link set %s up", devname);
  systemf("ip addr add 192.168.42.1/24 dev %s", devname);

  struct {
    struct iphdr ip;
    struct tcphdr tcp;
    unsigned char tcp_opts[20];
  } __attribute__((packed)) syn_packet = {
    .ip = {
      .ihl = sizeof(struct iphdr)/4,
      .version = 4,
      .tot_len = htons(sizeof(syn_packet)),
      .ttl = 30,
      .protocol = IPPROTO_TCP,
      /* FIXUP check */
      .saddr = IPADDR(192,168,42,2),
      .daddr = IPADDR(192,168,42,1)
    },
    .tcp = {
      .source = htons(1),
      .dest = htons(1337),
      .seq = 0x12345678,
      .doff = (sizeof(syn_packet.tcp)+sizeof(syn_packet.tcp_opts))/4,
      .syn = 1,
      .window = htons(64),
      .check = 0 /*FIXUP*/
    },
    .tcp_opts = {
      /* INVALID: trailing MD5SIG opcode after NOPs */
      1, 1, 1, 1, 1,
      1, 1, 1, 1, 1,
      1, 1, 1, 1, 1,
      1, 1, 1, 1, 19
    }
  };
  fix_ip_sum(&syn_packet.ip);
  fix_tcp_sum(&syn_packet.ip, &syn_packet.tcp);
  while (1) {
    int write_res = write(tun_fd, &syn_packet, sizeof(syn_packet));
    if (write_res != sizeof(syn_packet))
      err(1, "packet write failed");
  }
}
====================================

Fixes: cfb6eeb4c860 ("[TCP]: MD5 Signature Option (RFC2385) support.")
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoLinux 4.17-rc2 v4.17-rc2
Linus Torvalds [Mon, 23 Apr 2018 02:20:09 +0000 (19:20 -0700)]
Linux 4.17-rc2