]> git.proxmox.com Git - mirror_ubuntu-kernels.git/log
mirror_ubuntu-kernels.git
6 years agobpf: allow xadd only on aligned memory
Daniel Borkmann [Fri, 23 Feb 2018 21:29:05 +0000 (22:29 +0100)]
bpf: allow xadd only on aligned memory

The requirements around atomic_add() / atomic64_add() resp. their
JIT implementations differ across architectures. E.g. while x86_64
seems just fine with BPF's xadd on unaligned memory, on arm64 it
triggers via interpreter but also JIT the following crash:

  [  830.864985] Unable to handle kernel paging request at virtual address ffff8097d7ed6703
  [...]
  [  830.916161] Internal error: Oops: 96000021 [#1] SMP
  [  830.984755] CPU: 37 PID: 2788 Comm: test_verifier Not tainted 4.16.0-rc2+ #8
  [  830.991790] Hardware name: Huawei TaiShan 2280 /BC11SPCD, BIOS 1.29 07/17/2017
  [  830.998998] pstate: 80400005 (Nzcv daif +PAN -UAO)
  [  831.003793] pc : __ll_sc_atomic_add+0x4/0x18
  [  831.008055] lr : ___bpf_prog_run+0x1198/0x1588
  [  831.012485] sp : ffff00001ccabc20
  [  831.015786] x29: ffff00001ccabc20 x28: ffff8017d56a0f00
  [  831.021087] x27: 0000000000000001 x26: 0000000000000000
  [  831.026387] x25: 000000c168d9db98 x24: 0000000000000000
  [  831.031686] x23: ffff000008203878 x22: ffff000009488000
  [  831.036986] x21: ffff000008b14e28 x20: ffff00001ccabcb0
  [  831.042286] x19: ffff0000097b5080 x18: 0000000000000a03
  [  831.047585] x17: 0000000000000000 x16: 0000000000000000
  [  831.052885] x15: 0000ffffaeca8000 x14: 0000000000000000
  [  831.058184] x13: 0000000000000000 x12: 0000000000000000
  [  831.063484] x11: 0000000000000001 x10: 0000000000000000
  [  831.068783] x9 : 0000000000000000 x8 : 0000000000000000
  [  831.074083] x7 : 0000000000000000 x6 : 000580d428000000
  [  831.079383] x5 : 0000000000000018 x4 : 0000000000000000
  [  831.084682] x3 : ffff00001ccabcb0 x2 : 0000000000000001
  [  831.089982] x1 : ffff8097d7ed6703 x0 : 0000000000000001
  [  831.095282] Process test_verifier (pid: 2788, stack limit = 0x0000000018370044)
  [  831.102577] Call trace:
  [  831.105012]  __ll_sc_atomic_add+0x4/0x18
  [  831.108923]  __bpf_prog_run32+0x4c/0x70
  [  831.112748]  bpf_test_run+0x78/0xf8
  [  831.116224]  bpf_prog_test_run_xdp+0xb4/0x120
  [  831.120567]  SyS_bpf+0x77c/0x1110
  [  831.123873]  el0_svc_naked+0x30/0x34
  [  831.127437] Code: 97fffe97 17ffffec 00000000 f9800031 (885f7c31)

Reason for this is because memory is required to be aligned. In
case of BPF, we always enforce alignment in terms of stack access,
but not when accessing map values or packet data when the underlying
arch (e.g. arm64) has CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS set.

xadd on packet data that is local to us anyway is just wrong, so
forbid this case entirely. The only place where xadd makes sense in
fact are map values; xadd on stack is wrong as well, but it's been
around for much longer. Specifically enforce strict alignment in case
of xadd, so that we handle this case generically and avoid such crashes
in the first place.

Fixes: 17a5267067f3 ("bpf: verifier (add verifier core)")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agonet_sched: gen_estimator: fix broken estimators based on percpu stats
Eric Dumazet [Fri, 23 Feb 2018 03:45:27 +0000 (19:45 -0800)]
net_sched: gen_estimator: fix broken estimators based on percpu stats

pfifo_fast got percpu stats lately, uncovering a bug I introduced last
year in linux-4.10.

I missed the fact that we have to clear our temporary storage
before calling __gnet_stats_copy_basic() in the case of percpu stats.

Without this fix, rate estimators (tc qd replace dev xxx root est 1sec
4sec pfifo_fast) are utterly broken.

Fixes: 1c0d32fde5bd ("net_sched: gen_estimator: complete rewrite of rate estimators")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Fri, 23 Feb 2018 17:34:18 +0000 (12:34 -0500)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf

Alexei Starovoitov says:

====================
pull-request: bpf 2018-02-22

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

The main changes are:

1) two urgent fixes for bpf_tail_call logic for x64 and arm64 JITs, from Daniel.

2) cond_resched points in percpu array alloc/free paths, from Eric.

3) lockdep and other minor fixes, from Yonghong, Arnd, Anders, Li.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agogianfar: simplify FCS handling and fix memory leak
Andy Spencer [Thu, 22 Feb 2018 19:05:33 +0000 (11:05 -0800)]
gianfar: simplify FCS handling and fix memory leak

Previously, buffer descriptors containing only the frame check sequence
(FCS) were skipped and not added to the skb. However, the page reference
count was still incremented, leading to a memory leak.

Fixing this inside gfar_add_rx_frag() is difficult due to reserved
memory handling and page reuse. Instead, move the FCS handling to
gfar_process_frame() and trim off the FCS before passing the skb up the
networking stack.

Signed-off-by: Andy Spencer <aspencer@spacex.com>
Signed-off-by: Jim Gruen <jgruen@spacex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6 sit: work around bogus gcc-8 -Wrestrict warning
Arnd Bergmann [Thu, 22 Feb 2018 15:55:34 +0000 (16:55 +0100)]
ipv6 sit: work around bogus gcc-8 -Wrestrict warning

gcc-8 has a new warning that detects overlapping input and output arguments
in memcpy(). It triggers for sit_init_net() calling ipip6_tunnel_clone_6rd(),
which is actually correct:

net/ipv6/sit.c: In function 'sit_init_net':
net/ipv6/sit.c:192:3: error: 'memcpy' source argument is the same as destination [-Werror=restrict]

The problem here is that the logic detecting the memcpy() arguments finds them
to be the same, but the conditional that tests for the input and output of
ipip6_tunnel_clone_6rd() to be identical is not a compile-time constant.

We know that netdev_priv(t->dev) is the same as t for a tunnel device,
and comparing "dev" directly here lets the compiler figure out as well
that 'dev == sitn->fb_tunnel_dev' when called from sit_init_net(), so
it no longer warns.

This code is old, so Cc stable to make sure that we don't get the warning
for older kernels built with new gcc.

Cc: Martin Sebor <msebor@gmail.com>
Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83456
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomacvlan: fix use-after-free in macvlan_common_newlink()
Alexey Kodanev [Thu, 22 Feb 2018 15:20:30 +0000 (18:20 +0300)]
macvlan: fix use-after-free in macvlan_common_newlink()

The following use-after-free was reported by KASan when running
LTP macvtap01 test on 4.16-rc2:

[10642.528443] BUG: KASAN: use-after-free in
               macvlan_common_newlink+0x12ef/0x14a0 [macvlan]
[10642.626607] Read of size 8 at addr ffff880ba49f2100 by task ip/18450
...
[10642.963873] Call Trace:
[10642.994352]  dump_stack+0x5c/0x7c
[10643.035325]  print_address_description+0x75/0x290
[10643.092938]  kasan_report+0x28d/0x390
[10643.137971]  ? macvlan_common_newlink+0x12ef/0x14a0 [macvlan]
[10643.207963]  macvlan_common_newlink+0x12ef/0x14a0 [macvlan]
[10643.275978]  macvtap_newlink+0x171/0x260 [macvtap]
[10643.334532]  rtnl_newlink+0xd4f/0x1300
...
[10646.256176] Allocated by task 18450:
[10646.299964]  kasan_kmalloc+0xa6/0xd0
[10646.343746]  kmem_cache_alloc_trace+0xf1/0x210
[10646.397826]  macvlan_common_newlink+0x6de/0x14a0 [macvlan]
[10646.464386]  macvtap_newlink+0x171/0x260 [macvtap]
[10646.522728]  rtnl_newlink+0xd4f/0x1300
...
[10647.022028] Freed by task 18450:
[10647.061549]  __kasan_slab_free+0x138/0x180
[10647.111468]  kfree+0x9e/0x1c0
[10647.147869]  macvlan_port_destroy+0x3db/0x650 [macvlan]
[10647.211411]  rollback_registered_many+0x5b9/0xb10
[10647.268715]  rollback_registered+0xd9/0x190
[10647.319675]  register_netdevice+0x8eb/0xc70
[10647.370635]  macvlan_common_newlink+0xe58/0x14a0 [macvlan]
[10647.437195]  macvtap_newlink+0x171/0x260 [macvtap]

Commit d02fd6e7d293 ("macvlan: Fix one possible double free") handles
the case when register_netdevice() invokes ndo_uninit() on error and
as a result free the port. But 'macvlan_port_get_rtnl(dev))' check
(returns dev->rx_handler_data), which was added by this commit in order
to prevent double free, is not quite correct:

* for macvlan it always returns NULL because 'lowerdev' is the one that
  was used to register rx handler (port) in macvlan_port_create() as
  well as to unregister it in macvlan_port_destroy().
* for macvtap it always returns a valid pointer because macvtap registers
  its own rx handler before macvlan_common_newlink().

Fixes: d02fd6e7d293 ("macvlan: Fix one possible double free")
Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpf, arm64: fix out of bounds access in tail call
Daniel Borkmann [Fri, 23 Feb 2018 00:03:43 +0000 (01:03 +0100)]
bpf, arm64: fix out of bounds access in tail call

I recently noticed a crash on arm64 when feeding a bogus index
into BPF tail call helper. The crash would not occur when the
interpreter is used, but only in case of JIT. Output looks as
follows:

  [  347.007486] Unable to handle kernel paging request at virtual address fffb850e96492510
  [...]
  [  347.043065] [fffb850e96492510] address between user and kernel address ranges
  [  347.050205] Internal error: Oops: 96000004 [#1] SMP
  [...]
  [  347.190829] x13: 0000000000000000 x12: 0000000000000000
  [  347.196128] x11: fffc047ebe782800 x10: ffff808fd7d0fd10
  [  347.201427] x9 : 0000000000000000 x8 : 0000000000000000
  [  347.206726] x7 : 0000000000000000 x6 : 001c991738000000
  [  347.212025] x5 : 0000000000000018 x4 : 000000000000ba5a
  [  347.217325] x3 : 00000000000329c4 x2 : ffff808fd7cf0500
  [  347.222625] x1 : ffff808fd7d0fc00 x0 : ffff808fd7cf0500
  [  347.227926] Process test_verifier (pid: 4548, stack limit = 0x000000007467fa61)
  [  347.235221] Call trace:
  [  347.237656]  0xffff000002f3a4fc
  [  347.240784]  bpf_test_run+0x78/0xf8
  [  347.244260]  bpf_prog_test_run_skb+0x148/0x230
  [  347.248694]  SyS_bpf+0x77c/0x1110
  [  347.251999]  el0_svc_naked+0x30/0x34
  [  347.255564] Code: 9100075a d280220a 8b0a002a d37df04b (f86b694b)
  [...]

In this case the index used in BPF r3 is the same as in r1
at the time of the call, meaning we fed a pointer as index;
here, it had the value 0xffff808fd7cf0500 which sits in x2.

While I found tail calls to be working in general (also for
hitting the error cases), I noticed the following in the code
emission:

  # bpftool p d j i 988
  [...]
  38:   ldr     w10, [x1,x10]
  3c:   cmp     w2, w10
  40:   b.ge    0x000000000000007c              <-- signed cmp
  44:   mov     x10, #0x20                      // #32
  48:   cmp     x26, x10
  4c:   b.gt    0x000000000000007c
  50:   add     x26, x26, #0x1
  54:   mov     x10, #0x110                     // #272
  58:   add     x10, x1, x10
  5c:   lsl     x11, x2, #3
  60:   ldr     x11, [x10,x11]                  <-- faulting insn (f86b694b)
  64:   cbz     x11, 0x000000000000007c
  [...]

Meaning, the tests passed because commit ddb55992b04d ("arm64:
bpf: implement bpf_tail_call() helper") was using signed compares
instead of unsigned which as a result had the test wrongly passing.

Change this but also the tail call count test both into unsigned
and cap the index as u32. Latter we did as well in 90caccdd8cc0
("bpf: fix bpf_tail_call() x64 JIT") and is needed in addition here,
too. Tested on HiSilicon Hi1616.

Result after patch:

  # bpftool p d j i 268
  [...]
  38: ldr w10, [x1,x10]
  3c: add w2, w2, #0x0
  40: cmp w2, w10
  44: b.cs 0x0000000000000080
  48: mov x10, #0x20                   // #32
  4c: cmp x26, x10
  50: b.hi 0x0000000000000080
  54: add x26, x26, #0x1
  58: mov x10, #0x110                  // #272
  5c: add x10, x1, x10
  60: lsl x11, x2, #3
  64: ldr x11, [x10,x11]
  68: cbz x11, 0x0000000000000080
  [...]

Fixes: ddb55992b04d ("arm64: bpf: implement bpf_tail_call() helper")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agobpf, x64: implement retpoline for tail call
Daniel Borkmann [Thu, 22 Feb 2018 14:12:53 +0000 (15:12 +0100)]
bpf, x64: implement retpoline for tail call

Implement a retpoline [0] for the BPF tail call JIT'ing that converts
the indirect jump via jmp %rax that is used to make the long jump into
another JITed BPF image. Since this is subject to speculative execution,
we need to control the transient instruction sequence here as well
when CONFIG_RETPOLINE is set, and direct it into a pause + lfence loop.
The latter aligns also with what gcc / clang emits (e.g. [1]).

JIT dump after patch:

  # bpftool p d x i 1
   0: (18) r2 = map[id:1]
   2: (b7) r3 = 0
   3: (85) call bpf_tail_call#12
   4: (b7) r0 = 2
   5: (95) exit

With CONFIG_RETPOLINE:

  # bpftool p d j i 1
  [...]
  33: cmp    %edx,0x24(%rsi)
  36: jbe    0x0000000000000072  |*
  38: mov    0x24(%rbp),%eax
  3e: cmp    $0x20,%eax
  41: ja     0x0000000000000072  |
  43: add    $0x1,%eax
  46: mov    %eax,0x24(%rbp)
  4c: mov    0x90(%rsi,%rdx,8),%rax
  54: test   %rax,%rax
  57: je     0x0000000000000072  |
  59: mov    0x28(%rax),%rax
  5d: add    $0x25,%rax
  61: callq  0x000000000000006d  |+
  66: pause                      |
  68: lfence                     |
  6b: jmp    0x0000000000000066  |
  6d: mov    %rax,(%rsp)         |
  71: retq                       |
  72: mov    $0x2,%eax
  [...]

  * relative fall-through jumps in error case
  + retpoline for indirect jump

Without CONFIG_RETPOLINE:

  # bpftool p d j i 1
  [...]
  33: cmp    %edx,0x24(%rsi)
  36: jbe    0x0000000000000063  |*
  38: mov    0x24(%rbp),%eax
  3e: cmp    $0x20,%eax
  41: ja     0x0000000000000063  |
  43: add    $0x1,%eax
  46: mov    %eax,0x24(%rbp)
  4c: mov    0x90(%rsi,%rdx,8),%rax
  54: test   %rax,%rax
  57: je     0x0000000000000063  |
  59: mov    0x28(%rax),%rax
  5d: add    $0x25,%rax
  61: jmpq   *%rax               |-
  63: mov    $0x2,%eax
  [...]

  * relative fall-through jumps in error case
  - plain indirect jump as before

  [0] https://support.google.com/faqs/answer/7625886
  [1] https://github.com/gcc-mirror/gcc/commit/a31e654fa107be968b802786d747e962c2fcdb2b

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agorxrpc: Fix send in rxrpc_send_data_packet()
David Howells [Thu, 22 Feb 2018 14:38:14 +0000 (14:38 +0000)]
rxrpc: Fix send in rxrpc_send_data_packet()

All the kernel_sendmsg() calls in rxrpc_send_data_packet() need to send
both parts of the iov[] buffer, but one of them does not.  Fix it so that
it does.

Without this, short IPv6 rxrpc DATA packets may be seen that have the rxrpc
header included, but no payload.

Fixes: 5a924b8951f8 ("rxrpc: Don't store the rxrpc header in the Tx queue sk_buffs")
Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: aquantia: Fix error handling in aq_pci_probe()
Dan Carpenter [Thu, 22 Feb 2018 09:11:55 +0000 (12:11 +0300)]
net: aquantia: Fix error handling in aq_pci_probe()

We should check "self->aq_hw" for allocation failure, and also we should
free it on the error paths.

Fixes: 23ee07ad3c2f ("net: aquantia: Cleanup pci functions module")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpf: fix rcu lockdep warning for lpm_trie map_free callback
Yonghong Song [Thu, 22 Feb 2018 18:10:35 +0000 (10:10 -0800)]
bpf: fix rcu lockdep warning for lpm_trie map_free callback

Commit 9a3efb6b661f ("bpf: fix memory leak in lpm_trie map_free callback function")
fixed a memory leak and removed unnecessary locks in map_free callback function.
Unfortrunately, it introduced a lockdep warning. When lockdep checking is turned on,
running tools/testing/selftests/bpf/test_lpm_map will have:

  [   98.294321] =============================
  [   98.294807] WARNING: suspicious RCU usage
  [   98.295359] 4.16.0-rc2+ #193 Not tainted
  [   98.295907] -----------------------------
  [   98.296486] /home/yhs/work/bpf/kernel/bpf/lpm_trie.c:572 suspicious rcu_dereference_check() usage!
  [   98.297657]
  [   98.297657] other info that might help us debug this:
  [   98.297657]
  [   98.298663]
  [   98.298663] rcu_scheduler_active = 2, debug_locks = 1
  [   98.299536] 2 locks held by kworker/2:1/54:
  [   98.300152]  #0:  ((wq_completion)"events"){+.+.}, at: [<00000000196bc1f0>] process_one_work+0x157/0x5c0
  [   98.301381]  #1:  ((work_completion)(&map->work)){+.+.}, at: [<00000000196bc1f0>] process_one_work+0x157/0x5c0

Since actual trie tree removal happens only after no other
accesses to the tree are possible, replacing
  rcu_dereference_protected(*slot, lockdep_is_held(&trie->lock))
with
  rcu_dereference_protected(*slot, 1)
fixed the issue.

Fixes: 9a3efb6b661f ("bpf: fix memory leak in lpm_trie map_free callback function")
Reported-by: Eric Dumazet <edumazet@google.com>
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agobpf: add schedule points in percpu arrays management
Eric Dumazet [Thu, 22 Feb 2018 16:33:24 +0000 (08:33 -0800)]
bpf: add schedule points in percpu arrays management

syszbot managed to trigger RCU detected stalls in
bpf_array_free_percpu()

It takes time to allocate a huge percpu map, but even more time to free
it.

Since we run in process context, use cond_resched() to yield cpu if
needed.

Fixes: a10423b87a7e ("bpf: introduce BPF_MAP_TYPE_PERCPU_ARRAY map")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agoMerge tag 'mac80211-for-davem-2018-02-22' of git://git.kernel.org/pub/scm/linux/kerne...
David S. Miller [Thu, 22 Feb 2018 20:17:01 +0000 (15:17 -0500)]
Merge tag 'mac80211-for-davem-2018-02-22' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211

Johannes Berg says:

====================
Various fixes across the tree, the shortlog basically says it all:

  cfg80211: fix cfg80211_beacon_dup
  -> old bug in this code

  cfg80211: clear wep keys after disconnection
  -> certain ways of disconnecting left the keys

  mac80211: round IEEE80211_TX_STATUS_HEADROOM up to multiple of 4
  -> alignment issues with using 14 bytes

  mac80211: Do not disconnect on invalid operating class
  -> if the AP has a bogus operating class, let it be

  mac80211: Fix sending ADDBA response for an ongoing session
  -> don't send the same frame twice

  cfg80211: use only 1Mbps for basic rates in mesh
  -> interop issue with old versions of our code

  mac80211_hwsim: don't use WQ_MEM_RECLAIM
  -> it causes splats because it flushes work on a non-reclaim WQ

  regulatory: add NUL to request alpha2
  -> nla_put_string() issue from Kees

  mac80211: mesh: fix wrong mesh TTL offset calculation
  -> protocol issue

  mac80211: fix a possible leak of station stats
  -> error path might leak memory

  mac80211: fix calling sleeping function in atomic context
  -> percpu allocations need to be made with gfp flags
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoregulatory: add NUL to request alpha2
Johannes Berg [Thu, 22 Feb 2018 19:55:28 +0000 (20:55 +0100)]
regulatory: add NUL to request alpha2

Similar to the ancient commit a5fe8e7695dc ("regulatory: add NUL
to alpha2"), add another byte to alpha2 in the request struct so
that when we use nla_put_string(), we don't overrun anything.

Fixes: 73d54c9e74c4 ("cfg80211: add regulatory netlink multicast group")
Reported-by: Kees Cook <keescook@google.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
6 years agoibmvnic: Fix early release of login buffer
Thomas Falcon [Thu, 22 Feb 2018 00:18:30 +0000 (18:18 -0600)]
ibmvnic: Fix early release of login buffer

The login buffer is released before the driver can perform
sanity checks between resources the driver requested and what
firmware will provide. Don't release the login buffer until
the sanity check is performed.

Fixes: 34f0f4e3f488 ("ibmvnic: Fix login buffer memory leaks")
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/smc9194: Remove bogus CONFIG_MAC reference
Finn Thain [Wed, 21 Feb 2018 22:24:59 +0000 (09:24 +1100)]
net/smc9194: Remove bogus CONFIG_MAC reference

AFAIK the only version of smc9194.c with Mac support is the one in the
linux-mac68k CVS repo, which never made it to the mainline.

Despite that, from v2.3.45, arch/m68k/config.in listed CONFIG_SMC9194
under CONFIG_MAC. This mistake got carried over into Kconfig in v2.5.55.
(See pre-git era "[PATCH] add m68k dependencies to net driver config".)

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ipv4: Set addr_type in hash_keys for forwarded case
David Ahern [Wed, 21 Feb 2018 19:00:54 +0000 (11:00 -0800)]
net: ipv4: Set addr_type in hash_keys for forwarded case

The result of the skb flow dissect is copied from keys to hash_keys to
ensure only the intended data is hashed. The original L4 hash patch
overlooked setting the addr_type for this case; add it.

Fixes: bf4e0a3db97eb ("net: ipv4: add support for ECMP hash policy choice")
Reported-by: Ido Schimmel <idosch@idosch.org>
Signed-off-by: David Ahern <dsahern@gmail.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp_bbr: better deal with suboptimal GSO
Eric Dumazet [Wed, 21 Feb 2018 14:43:03 +0000 (06:43 -0800)]
tcp_bbr: better deal with suboptimal GSO

BBR uses tcp_tso_autosize() in an attempt to probe what would be the
burst sizes and to adjust cwnd in bbr_target_cwnd() with following
gold formula :

/* Allow enough full-sized skbs in flight to utilize end systems. */
cwnd += 3 * bbr->tso_segs_goal;

But GSO can be lacking or be constrained to very small
units (ip link set dev ... gso_max_segs 2)

What we really want is to have enough packets in flight so that both
GSO and GRO are efficient.

So in the case GSO is off or downgraded, we still want to have the same
number of packets in flight as if GSO/TSO was fully operational, so
that GRO can hopefully be working efficiently.

To fix this issue, we make tcp_tso_autosize() unaware of
sk->sk_gso_max_segs

Only tcp_tso_segs() has to enforce the gso_max_segs limit.

Tested:

ethtool -K eth0 tso off gso off
tc qd replace dev eth0 root pfifo_fast

Before patch:
for f in {1..5}; do ./super_netperf 1 -H lpaa24 -- -K bbr; done
    691  (ss -temoi shows cwnd is stuck around 6 )
    667
    651
    631
    517

After patch :
# for f in {1..5}; do ./super_netperf 1 -H lpaa24 -- -K bbr; done
   1733 (ss -temoi shows cwnd is around 386 )
   1778
   1746
   1781
   1718

Fixes: 0f8782ea1497 ("tcp_bbr: add BBR congestion control")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosmsc75xx: fix smsc75xx_set_features()
Eric Dumazet [Wed, 21 Feb 2018 05:42:26 +0000 (21:42 -0800)]
smsc75xx: fix smsc75xx_set_features()

If an attempt is made to disable RX checksums, USB adapter is changed
but netdev->features is not, because smsc75xx_set_features() returns a
non zero value.

This throws errors from netdev_rx_csum_fault() :
<devname>: hw csum failure

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Steve Glendinning <steve.glendinning@shawell.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonetlink: put module reference if dump start fails
Jason A. Donenfeld [Wed, 21 Feb 2018 03:41:59 +0000 (04:41 +0100)]
netlink: put module reference if dump start fails

Before, if cb->start() failed, the module reference would never be put,
because cb->cb_running is intentionally false at this point. Users are
generally annoyed by this because they can no longer unload modules that
leak references. Also, it may be possible to tediously wrap a reference
counter back to zero, especially since module.c still uses atomic_inc
instead of refcount_inc.

This patch expands the error path to simply call module_put if
cb->start() fails.

Fixes: 41c87425a1ac ("netlink: do not set cb_running if dump's start() errs")
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests/bpf/test_maps: exit child process without error in ENOMEM case
Li Zhijian [Thu, 22 Feb 2018 02:34:02 +0000 (10:34 +0800)]
selftests/bpf/test_maps: exit child process without error in ENOMEM case

test_maps contains a series of stress tests, and previously it will break the
rest tests when it failed to alloc memory.
-----------------------
Failed to create hashmap key=8 value=262144 'Cannot allocate memory'
Failed to create hashmap key=16 value=262144 'Cannot allocate memory'
Failed to create hashmap key=8 value=262144 'Cannot allocate memory'
Failed to create hashmap key=8 value=262144 'Cannot allocate memory'
test_maps: test_maps.c:955: run_parallel: Assertion `status == 0' failed.
Aborted
not ok 1..3 selftests:  test_maps [FAIL]
-----------------------
after this patch, the rest tests will be continue when it occurs an ENOMEM failure

CC: Alexei Starovoitov <alexei.starovoitov@gmail.com>
CC: Philip Li <philip.li@intel.com>
Suggested-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Li Zhijian <zhijianx.li@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agoselftests/bpf: update gitignore with test_libbpf_open
Anders Roxell [Wed, 21 Feb 2018 21:30:01 +0000 (22:30 +0100)]
selftests/bpf: update gitignore with test_libbpf_open

bpf builds a test program for loading BPF ELF files. Add the executable
to the .gitignore list.

Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
Tested-by: Daniel Díaz <daniel.diaz@linaro.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Shuah Khan <shuahkh@osg.samsung.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agoselftests/bpf: tcpbpf_kern: use in6_* macros from glibc
Anders Roxell [Wed, 21 Feb 2018 16:51:16 +0000 (17:51 +0100)]
selftests/bpf: tcpbpf_kern: use in6_* macros from glibc

Both glibc and the kernel have in6_* macros definitions. Build fails
because it picks up wrong in6_* macro from the kernel header and not the
header from glibc.

Fixes build error below:
clang -I. -I./include/uapi -I../../../include/uapi
     -Wno-compare-distinct-pointer-types \
         -O2 -target bpf -emit-llvm -c test_tcpbpf_kern.c -o - |      \
llc -march=bpf -mcpu=generic -filetype=obj
     -o .../tools/testing/selftests/bpf/test_tcpbpf_kern.o
In file included from test_tcpbpf_kern.c:12:
.../netinet/in.h:101:5: error: expected identifier
    IPPROTO_HOPOPTS = 0,   /* IPv6 Hop-by-Hop options.  */
    ^
.../linux/in6.h:131:26: note: expanded from macro 'IPPROTO_HOPOPTS'
                                ^
In file included from test_tcpbpf_kern.c:12:
/usr/include/netinet/in.h:103:5: error: expected identifier
    IPPROTO_ROUTING = 43,  /* IPv6 routing header.  */
    ^
.../linux/in6.h:132:26: note: expanded from macro 'IPPROTO_ROUTING'
                                ^
In file included from test_tcpbpf_kern.c:12:
.../netinet/in.h:105:5: error: expected identifier
    IPPROTO_FRAGMENT = 44, /* IPv6 fragmentation header.  */
    ^

Since both glibc and the kernel have in6_* macros definitions, use the
one from glibc.  Kernel headers will check for previous libc definitions
by including include/linux/libc-compat.h.

Reported-by: Daniel Díaz <daniel.diaz@linaro.org>
Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
Tested-by: Daniel Díaz <daniel.diaz@linaro.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agobpf: clean up unused-variable warning
Arnd Bergmann [Tue, 20 Feb 2018 22:07:33 +0000 (23:07 +0100)]
bpf: clean up unused-variable warning

The only user of this variable is inside of an #ifdef, causing
a warning without CONFIG_INET:

net/core/filter.c: In function '____bpf_sock_ops_cb_flags_set':
net/core/filter.c:3382:6: error: unused variable 'val' [-Werror=unused-variable]
  int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;

This replaces the #ifdef with a nicer IS_ENABLED() check that
makes the code more readable and avoids the warning.

Fixes: b13d88072172 ("bpf: Adds field bpf_sock_ops_cb_flags to tcp_sock")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agoamd-xgbe: Restore PCI interrupt enablement setting on resume
Tom Lendacky [Tue, 20 Feb 2018 21:22:05 +0000 (15:22 -0600)]
amd-xgbe: Restore PCI interrupt enablement setting on resume

After resuming from suspend, the PCI device support must re-enable the
interrupt setting so that interrupts are actually delivered.

Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Wed, 21 Feb 2018 20:37:37 +0000 (15:37 -0500)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2018-02-20

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

The main changes are:

1) Fix a memory leak in LPM trie's map_free() callback function, where
   the trie structure itself was not freed since initial implementation.
   Also a synchronize_rcu() was needed in order to wait for outstanding
   programs accessing the trie to complete, from Yonghong.

2) Fix sock_map_alloc()'s error path in order to correctly propagate
   the -EINVAL error in case of too large allocation requests. This
   was just recently introduced when fixing close hooks via ULP layer,
   fix from Eric.

3) Do not use GFP_ATOMIC in __cpu_map_entry_alloc(). Reason is that this
   will not work with the recent __ptr_ring_init_queue_alloc() conversion
   to kvmalloc_array(), where in case of fallback to vmalloc() that GFP
   flag is invalid, from Jason.

4) Fix two recent syzkaller warnings: i) fix bpf_prog_array_copy_to_user()
   when a prog query with a big number of ids was performed where we'd
   otherwise trigger a warning from allocator side, ii) fix a missing
   mlock precharge on arraymaps, from Daniel.

5) Two fixes for bpftool in order to avoid breaking JSON output when used
   in batch mode, from Quentin.

6) Move a pr_debug() in libbpf in order to avoid having an otherwise
   uninitialized variable in bpf_program__reloc_text(), from Jeremy.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'virtio_net-XDP-fixes'
David S. Miller [Wed, 21 Feb 2018 20:09:30 +0000 (15:09 -0500)]
Merge branch 'virtio_net-XDP-fixes'

Jesper Dangaard Brouer says:

====================
virtio_net: several bugs in XDP code for driver virtio_net

The virtio_net driver actually violates the original memory model of
XDP causing hard to debug crashes.  Per request of John Fastabend,
instead of removing the XDP feature I'm fixing as much as possible.
While testing virtio_net with XDP_REDIRECT I found 4 different bugs.

Patch-1: not enough tail-room for build_skb in receive_mergeable()
 only option is to disable XDP_REDIRECT in receive_mergeable()

Patch-2: XDP in receive_small() basically never worked (check wrong flag)

Patch-3: fix memory leak for XDP_REDIRECT in error cases

Patch-4: avoid crash when ndo_xdp_xmit is called on dev not ready for XDP

In the longer run, we should consider introducing a separate receive
function when attaching an XDP program, and also change the memory
model to be compatible with XDP when attaching an XDP prog.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovirtio_net: fix ndo_xdp_xmit crash towards dev not ready for XDP
Jesper Dangaard Brouer [Tue, 20 Feb 2018 13:32:20 +0000 (14:32 +0100)]
virtio_net: fix ndo_xdp_xmit crash towards dev not ready for XDP

When a driver implements the ndo_xdp_xmit() function, there is
(currently) no generic way to determine whether it is safe to call.

It is e.g. unsafe to call the drivers ndo_xdp_xmit, if it have not
allocated the needed XDP TX queues yet.  This is the case for
virtio_net, which first allocates the XDP TX queues once an XDP/bpf
prog is attached (in virtnet_xdp_set()).

Thus, a crash will occur for virtio_net when redirecting to another
virtio_net device's ndo_xdp_xmit, which have not attached a XDP prog.
The sample xdp_redirect_map tries to attach a dummy XDP prog to take
this into account, but it can also easily fail if the virtio_net (or
actually underlying vhost driver) have not allocated enough extra
queues for the device.

Allocating more queue this is currently a manual config.
Hint for libvirt XML add:

  <driver name='vhost' queues='16'>
    <host mrg_rxbuf='off'/>
    <guest tso4='off' tso6='off' ecn='off' ufo='off'/>
  </driver>

The solution in this patch is to check that the device have loaded an
XDP/bpf prog before proceeding.  This is similar to the check
performed in driver ixgbe.

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovirtio_net: fix memory leak in XDP_REDIRECT
Jesper Dangaard Brouer [Tue, 20 Feb 2018 13:32:15 +0000 (14:32 +0100)]
virtio_net: fix memory leak in XDP_REDIRECT

XDP_REDIRECT calling xdp_do_redirect() can fail for multiple reasons
(which can be inspected by tracepoints). The current semantics is that
on failure the driver calling xdp_do_redirect() must handle freeing or
recycling the page associated with this frame.  This can be seen as an
optimization, as drivers usually have an optimized XDP_DROP code path
for frame recycling in place already.

The virtio_net driver didn't handle when xdp_do_redirect() failed.
This caused a memory leak as the page refcnt wasn't decremented on
failures.

The function __virtnet_xdp_xmit() did handle one type of failure,
when the xmit queue virtqueue_add_outbuf() is full, which "hides"
releasing a refcnt on the page.  Instead the function __virtnet_xdp_xmit()
must follow API of xdp_do_redirect(), which on errors leave it up to
the caller to free the page, of the failed send operation.

Fixes: 186b3c998c50 ("virtio-net: support XDP_REDIRECT")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovirtio_net: fix XDP code path in receive_small()
Jesper Dangaard Brouer [Tue, 20 Feb 2018 13:32:10 +0000 (14:32 +0100)]
virtio_net: fix XDP code path in receive_small()

When configuring virtio_net to use the code path 'receive_small()',
in-order to get correct XDP_REDIRECT support, I discovered TCP packets
would get silently dropped when loading an XDP program action XDP_PASS.

The bug seems to be that receive_small() when XDP is loaded check that
hdr->hdr.flags is zero, which seems wrong as hdr.flags contains the
flags VIRTIO_NET_HDR_F_* :
 #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /* Use csum_start, csum_offset */
 #define VIRTIO_NET_HDR_F_DATA_VALID 2 /* Csum is valid */

TCP got dropped as it had the VIRTIO_NET_HDR_F_DATA_VALID flag set.

The flags that are relevant here are the VIRTIO_NET_HDR_GSO_* flags
stored in hdr->hdr.gso_type. Thus, the fix is just check that none of
the gso_type flags have been set.

Fixes: bb91accf2733 ("virtio-net: XDP support for small buffers")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agovirtio_net: disable XDP_REDIRECT in receive_mergeable() case
Jesper Dangaard Brouer [Tue, 20 Feb 2018 13:32:04 +0000 (14:32 +0100)]
virtio_net: disable XDP_REDIRECT in receive_mergeable() case

The virtio_net code have three different RX code-paths in receive_buf().
Two of these code paths can handle XDP, but one of them is broken for
at least XDP_REDIRECT.

Function(1): receive_big() does not support XDP.
Function(2): receive_small() support XDP fully and uses build_skb().
Function(3): receive_mergeable() broken XDP_REDIRECT uses napi_alloc_skb().

The simple explanation is that receive_mergeable() is broken because
it uses napi_alloc_skb(), which violates XDP given XDP assumes packet
header+data in single page and enough tail room for skb_shared_info.

The longer explaination is that receive_mergeable() tries to
work-around and satisfy these XDP requiresments e.g. by having a
function xdp_linearize_page() that allocates and memcpy RX buffers
around (in case packet is scattered across multiple rx buffers).  This
does currently satisfy XDP_PASS, XDP_DROP and XDP_TX (but only because
we have not implemented bpf_xdp_adjust_tail yet).

The XDP_REDIRECT action combined with cpumap is broken, and cause hard
to debug crashes.  The main issue is that the RX packet does not have
the needed tail-room (SKB_DATA_ALIGN(skb_shared_info)), causing
skb_shared_info to overlap the next packets head-room (in which cpumap
stores info).

Reproducing depend on the packet payload length and if RX-buffer size
happened to have tail-room for skb_shared_info or not.  But to make
this even harder to troubleshoot, the RX-buffer size is runtime
dynamically change based on an Exponentially Weighted Moving Average
(EWMA) over the packet length, when refilling RX rings.

This patch only disable XDP_REDIRECT support in receive_mergeable()
case, because it can cause a real crash.

IMHO we should consider NOT supporting XDP in receive_mergeable() at
all, because the principles behind XDP are to gain speed by (1) code
simplicity, (2) sacrificing memory and (3) where possible moving
runtime checks to setup time.  These principles are clearly being
violated in receive_mergeable(), that e.g. runtime track average
buffer size to save memory consumption.

In the longer run, we should consider introducing a separate receive
function when attaching an XDP program, and also change the memory
model to be compatible with XDP when attaching an XDP prog.

Fixes: 186b3c998c50 ("virtio-net: support XDP_REDIRECT")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'mlx5-fixes-2018-02-20' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Wed, 21 Feb 2018 19:57:35 +0000 (14:57 -0500)]
Merge tag 'mlx5-fixes-2018-02-20' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
Mellanox, mlx5 fixes 2018-02-20

The following pull request includes some fixes for the mlx5 core and
netdevice driver.

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

-stable 4.10.y:
('net/mlx5e: Fix loopback self test when GRO is off')

-stable 4.12.y:
('net/mlx5e: Specify numa node when allocating drop rq')

-stable 4.13.y:
('net/mlx5e: Verify inline header size do not exceed SKB linear size')

-stable 4.15.y:
('net/mlx5e: Fix TCP checksum in LRO buffers')
('net/mlx5: Fix error handling when adding flow rules')
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf
David S. Miller [Wed, 21 Feb 2018 19:49:55 +0000 (14:49 -0500)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf

Pablo Neira Ayuso says:

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

The following patchset contains large batch with Netfilter fixes for
your net tree, mostly due to syzbot report fixups and pr_err()
ratelimiting, more specifically, they are:

1) Get rid of superfluous unnecessary check in x_tables before vmalloc(),
   we don't hit BUG there anymore, patch from Michal Hock, suggested by
   Andrew Morton.

2) Race condition in proc file creation in ipt_CLUSTERIP, from Cong Wang.

3) Drop socket lock that results in circular locking dependency, patch
   from Paolo Abeni.

4) Drop packet if case of malformed blob that makes backpointer jump
   in x_tables, from Florian Westphal.

5) Fix refcount leak due to race in ipt_CLUSTERIP in
   clusterip_config_find_get(), from Cong Wang.

6) Several patches to ratelimit pr_err() for x_tables since this can be
   a problem where CAP_NET_ADMIN semantics can protect us in untrusted
   namespace, from Florian Westphal.

7) Missing .gitignore update for new autogenerated asn1 state machine
   for the SNMP NAT helper, from Zhu Lingshan.

8) Missing timer initialization in xt_LED, from Paolo Abeni.

9) Do not allow negative port range in NAT, also from Paolo.

10) Lock imbalance in the xt_hashlimit rate match mode, patch from
    Eric Dumazet.

11) Initialize workqueue before timer in the idletimer match,
    from Eric Dumazet.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: report if filter is too large to dump
Roman Kapl [Mon, 19 Feb 2018 20:32:51 +0000 (21:32 +0100)]
net: sched: report if filter is too large to dump

So far, if the filter was too large to fit in the allocated skb, the
kernel did not return any error and stopped dumping. Modify the dumper
so that it returns -EMSGSIZE when a filter fails to dump and it is the
first filter in the skb. If we are not first, we will get a next chance
with more room.

I understand this is pretty near to being an API change, but the
original design (silent truncation) can be considered a bug.

Note: The error case can happen pretty easily if you create a filter
with 32 actions and have 4kb pages. Also recent versions of iproute try
to be clever with their buffer allocation size, which in turn leads to

Signed-off-by: Roman Kapl <code@rkapl.cz>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/mlx5: Fix error handling when adding flow rules
Vlad Buslov [Tue, 6 Feb 2018 08:52:19 +0000 (10:52 +0200)]
net/mlx5: Fix error handling when adding flow rules

If building match list or adding existing fg fails when
node is locked, function returned without unlocking it.
This happened if node version changed or adding existing fg
returned with EAGAIN after jumping to search_again_locked label.

Fixes: bd71b08ec2ee ("net/mlx5: Support multiple updates of steering rules in parallel")
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: E-Switch, Fix drop counters use before creation
Eugenia Emantayev [Wed, 31 Jan 2018 07:36:29 +0000 (09:36 +0200)]
net/mlx5: E-Switch, Fix drop counters use before creation

First use of drop counters happens in esw_apply_vport_conf function,
while they are allocated later in the flow. Fix that by moving
esw_vport_create_drop_counters function to be called before the first use.

Fixes: b8a0dbe3a90b ("net/mlx5e: E-switch, Add steering drop counters")
Signed-off-by: Eugenia Emantayev <eugenia@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: Add header re-write to the checks for conflicting actions
Or Gerlitz [Sun, 11 Feb 2018 11:26:06 +0000 (13:26 +0200)]
net/mlx5: Add header re-write to the checks for conflicting actions

We can't allow only some of the rules sharing an FTE to ask for
header re-write, add it to the conflicting action checks.

Fixes: 0d235c3fabb7 ('net/mlx5: Add hash table to search FTEs in a flow-group')
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: Use 128B cacheline size for 128B or larger cachelines
Daniel Jurgens [Fri, 2 Feb 2018 15:32:53 +0000 (09:32 -0600)]
net/mlx5: Use 128B cacheline size for 128B or larger cachelines

The adapter uses the cache_line_128byte setting to set the bounds for
end padding. On systems where the cacheline size is greater than 128B
use 128B instead of the default of 64B. This results in fewer partial
cacheline writes. There's a 50% chance it will pad to the end of a 256B
cache line vs only 25% when using 64B.

Fixes: f32f5bd2eb7e ("net/mlx5: Configure cache line size for start and end padding")
Signed-off-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5e: Specify numa node when allocating drop rq
Gal Pressman [Thu, 25 Jan 2018 16:00:41 +0000 (18:00 +0200)]
net/mlx5e: Specify numa node when allocating drop rq

When allocating a drop rq, no numa node is explicitly set which means
allocations are done on node zero. This is not necessarily the nearest
numa node to the HCA, and even worse, might even be a memoryless numa
node.

Choose the numa_node given to us by the pci device in order to properly
allocate the coherent dma memory instead of assuming zero is valid.

Fixes: 556dd1b9c313 ("net/mlx5e: Set drop RQ's necessary parameters only")
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5e: Return error if prio is specified when offloading eswitch vlan push
Or Gerlitz [Tue, 30 Jan 2018 11:16:58 +0000 (13:16 +0200)]
net/mlx5e: Return error if prio is specified when offloading eswitch vlan push

This isn't supported when we emulate eswitch vlan push action which
is the current state of things.

Fixes: 8b32580df1cb ('net/mlx5e: Add TC vlan action for SRIOV offloads')
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: Address static checker warnings on non-constant initializers
Or Gerlitz [Mon, 1 Jan 2018 13:29:53 +0000 (13:29 +0000)]
net/mlx5: Address static checker warnings on non-constant initializers

Address these sparse warnings on drivers/net/ethernet/mellanox/mlx5

[..]/core/diag/fs_tracepoint.c:99:53: warning: non-constant initializer for static object
[..]/core/diag/fs_tracepoint.c:102:53: warning: non-constant initializer for static object

etc

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5e: Eliminate build warnings on no previous prototype
Or Gerlitz [Mon, 1 Jan 2018 13:19:51 +0000 (13:19 +0000)]
net/mlx5e: Eliminate build warnings on no previous prototype

Fix these gcc warnings on drivers/net/ethernet/mellanox/mlx5:

[..]/core/lib/clock.c:454:6: warning: no previous prototype for 'mlx5_init_clock' [-Wmissing-prototypes]
[..]/core/lib/clock.c:510:6: warning: no previous prototype for 'mlx5_cleanup_clock' [-Wmissing-prototypes]
[..]/core/en_main.c:3141:5: warning: no previous prototype for 'mlx5e_setup_tc' [-Wmissing-prototypes]

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5e: Verify inline header size do not exceed SKB linear size
Eran Ben Elisha [Thu, 25 Jan 2018 09:18:09 +0000 (11:18 +0200)]
net/mlx5e: Verify inline header size do not exceed SKB linear size

Driver tries to copy at least MLX5E_MIN_INLINE bytes into the control
segment of the WQE. It assumes that the linear part contains at least
MLX5E_MIN_INLINE bytes, which can be wrong.

Cited commit verified that driver will not copy more bytes into the
inline header part that the actual size of the packet. Re-factor this
check to make sure we do not exceed the linear part as well.

This fix is aligned with the current driver's assumption that the entire
L2 will be present in the linear part of the SKB.

Fixes: 6aace17e64f4 ("net/mlx5e: Fix inline header size for small packets")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5e: Fix loopback self test when GRO is off
Inbar Karmy [Thu, 7 Dec 2017 15:26:33 +0000 (17:26 +0200)]
net/mlx5e: Fix loopback self test when GRO is off

When GRO is off, the transport header pointer in sk_buff is
initialized to network's header.

To find the udp header, instead of using udp_hdr() which assumes
skb_network_header was set, manually calculate the udp header offset.

Fixes: 0952da791c97 ("net/mlx5e: Add support for loopback selftest")
Signed-off-by: Inbar Karmy <inbark@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5e: Fix TCP checksum in LRO buffers
Gal Pressman [Wed, 20 Dec 2017 06:48:24 +0000 (08:48 +0200)]
net/mlx5e: Fix TCP checksum in LRO buffers

When receiving an LRO packet, the checksum field is set by the hardware
to the checksum of the first coalesced packet. Obviously, this checksum
is not valid for the merged LRO packet and should be fixed.  We can use
the CQE checksum which covers the checksum of the entire merged packet
TCP payload to help us calculate the checksum incrementally.

Tested by sending IPv4/6 traffic with LRO enabled, RX checksum disabled
and watching nstat checksum error counters (in addition to the obvious
bandwidth drop caused by checksum errors).

This bug is usually "hidden" since LRO packets would go through the
CHECKSUM_UNNECESSARY flow which does not validate the packet checksum.

It's important to note that previous to this patch, LRO packets provided
with CHECKSUM_UNNECESSARY are indeed packets with a correct validated
checksum (even though the checksum inside the TCP header is incorrect),
since the hardware LRO aggregation is terminated upon receiving a packet
with bad checksum.

Fixes: e586b3b0baee ("net/mlx5: Ethernet Datapath files")
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agotools/libbpf: Avoid possibly using uninitialized variable
Jeremy Cline [Tue, 20 Feb 2018 01:00:07 +0000 (01:00 +0000)]
tools/libbpf: Avoid possibly using uninitialized variable

Fixes a GCC maybe-uninitialized warning introduced by 48cca7e44f9f.
"text" is only initialized inside the if statement so only print debug
info there.

Fixes: 48cca7e44f9f ("libbpf: add support for bpf_call")
Signed-off-by: Jeremy Cline <jeremy@jcline.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
6 years agoibmvnic: Check for NULL skb's in NAPI poll routine
Thomas Falcon [Tue, 20 Feb 2018 02:12:57 +0000 (20:12 -0600)]
ibmvnic: Check for NULL skb's in NAPI poll routine

After introduction of commit d0869c0071e4, there were some instances of
RX queue entries from a previous session (before the device was closed
and reopened) returned to the NAPI polling routine. Since the corresponding
socket buffers were freed, this resulted in a panic on reopen. Include
a check for a NULL skb here to avoid this.

Fixes: d0869c0071e4 ("ibmvnic: Clean RX pool buffers during device close")
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Mon, 19 Feb 2018 19:58:19 +0000 (11:58 -0800)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Prevent index integer overflow in ptr_ring, from Jason Wang.

 2) Program mvpp2 multicast filter properly, from Mikulas Patocka.

 3) The bridge brport attribute file is write only and doesn't have a
    ->show() method, don't blindly invoke it. From Xin Long.

 4) Inverted mask used in genphy_setup_forced(), from Ingo van Lil.

 5) Fix multiple definition issue with if_ether.h UAPI header, from
    Hauke Mehrtens.

 6) Fix GFP_KERNEL usage in atomic in RDS protocol code, from Sowmini
    Varadhan.

 7) Revert XDP redirect support from thunderx driver, it is not
    implemented properly. From Jesper Dangaard Brouer.

 8) Fix missing RTNL protection across some tipc operations, from Ying
    Xue.

 9) Return the correct IV bytes in the TLS getsockopt code, from Boris
    Pismenny.

10) Take tclassid into consideration properly when doing FIB rule
    matching. From Stefano Brivio.

11) cxgb4 device needs more PCI VPD quirks, from Casey Leedom.

12) TUN driver doesn't align frags properly, and we can end up doing
    unaligned atomics on misaligned metadata. From Eric Dumazet.

13) Fix various crashes found using DEBUG_PREEMPT in rmnet driver, from
    Subash Abhinov Kasiviswanathan.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (56 commits)
  tg3: APE heartbeat changes
  mlxsw: spectrum_router: Do not unconditionally clear route offload indication
  net: qualcomm: rmnet: Fix possible null dereference in command processing
  net: qualcomm: rmnet: Fix warning seen with 64 bit stats
  net: qualcomm: rmnet: Fix crash on real dev unregistration
  sctp: remove the left unnecessary check for chunk in sctp_renege_events
  rxrpc: Work around usercopy check
  tun: fix tun_napi_alloc_frags() frag allocator
  udplite: fix partial checksum initialization
  skbuff: Fix comment mis-spelling.
  dn_getsockoptdecnet: move nf_{get/set}sockopt outside sock lock
  PCI/cxgb4: Extend T3 PCI quirk to T4+ devices
  cxgb4: fix trailing zero in CIM LA dump
  cxgb4: free up resources of pf 0-3
  fib_semantics: Don't match route with mismatching tclassid
  NFC: llcp: Limit size of SDP URI
  tls: getsockopt return record sequence number
  tls: reset the crypto info if copy_from_user fails
  tls: retrun the correct IV in getsockopt
  docs: segmentation-offloads.txt: add SCTP info
  ...

6 years agotg3: APE heartbeat changes
Prashant Sreedharan [Mon, 19 Feb 2018 06:57:04 +0000 (12:27 +0530)]
tg3: APE heartbeat changes

In ungraceful host shutdown or driver crash case BMC connectivity is
lost. APE firmware is missing the driver state in this
case to keep the BMC connectivity alive.
This patch has below change to address this issue.

Heartbeat mechanism with APE firmware. This heartbeat mechanism
is needed to notify the APE firmware about driver state.

This patch also has the change in wait time for APE event from
1ms to 20ms as there can be some delay in getting response.

v2: Drop inline keyword as per David suggestion.

Signed-off-by: Prashant Sreedharan <prashant.sreedharan@broadcom.com>
Signed-off-by: Satish Baddipadige <satish.baddipadige@broadcom.com>
Signed-off-by: Siva Reddy Kallam <siva.kallam@broadcom.com>
Acked-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonetfilter: IDLETIMER: be syzkaller friendly
Eric Dumazet [Sat, 17 Feb 2018 03:36:28 +0000 (19:36 -0800)]
netfilter: IDLETIMER: be syzkaller friendly

We had one report from syzkaller [1]

First issue is that INIT_WORK() should be done before mod_timer()
or we risk timer being fired too soon, even with a 1 second timer.

Second issue is that we need to reject too big info->timeout
to avoid overflows in msecs_to_jiffies(info->timeout * 1000), or
risk looping, if result after overflow is 0.

[1]
WARNING: CPU: 1 PID: 5129 at kernel/workqueue.c:1444 __queue_work+0xdf4/0x1230 kernel/workqueue.c:1444
Kernel panic - not syncing: panic_on_warn set ...

CPU: 1 PID: 5129 Comm: syzkaller159866 Not tainted 4.16.0-rc1+ #230
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 <IRQ>
 __dump_stack lib/dump_stack.c:17 [inline]
 dump_stack+0x194/0x257 lib/dump_stack.c:53
 panic+0x1e4/0x41c kernel/panic.c:183
 __warn+0x1dc/0x200 kernel/panic.c:547
 report_bug+0x211/0x2d0 lib/bug.c:184
 fixup_bug.part.11+0x37/0x80 arch/x86/kernel/traps.c:178
 fixup_bug arch/x86/kernel/traps.c:247 [inline]
 do_error_trap+0x2d7/0x3e0 arch/x86/kernel/traps.c:296
 do_invalid_op+0x1b/0x20 arch/x86/kernel/traps.c:315
 invalid_op+0x22/0x40 arch/x86/entry/entry_64.S:988
RIP: 0010:__queue_work+0xdf4/0x1230 kernel/workqueue.c:1444
RSP: 0018:ffff8801db507538 EFLAGS: 00010006
RAX: ffff8801aeb46080 RBX: ffff8801db530200 RCX: ffffffff81481404
RDX: 0000000000000100 RSI: ffffffff86b42640 RDI: 0000000000000082
RBP: ffff8801db507758 R08: 1ffff1003b6a0de5 R09: 000000000000000c
R10: ffff8801db5073f0 R11: 0000000000000020 R12: 1ffff1003b6a0eb6
R13: ffff8801b1067ae0 R14: 00000000000001f8 R15: dffffc0000000000
 queue_work_on+0x16a/0x1c0 kernel/workqueue.c:1488
 queue_work include/linux/workqueue.h:488 [inline]
 schedule_work include/linux/workqueue.h:546 [inline]
 idletimer_tg_expired+0x44/0x60 net/netfilter/xt_IDLETIMER.c:116
 call_timer_fn+0x228/0x820 kernel/time/timer.c:1326
 expire_timers kernel/time/timer.c:1363 [inline]
 __run_timers+0x7ee/0xb70 kernel/time/timer.c:1666
 run_timer_softirq+0x4c/0x70 kernel/time/timer.c:1692
 __do_softirq+0x2d7/0xb85 kernel/softirq.c:285
 invoke_softirq kernel/softirq.c:365 [inline]
 irq_exit+0x1cc/0x200 kernel/softirq.c:405
 exiting_irq arch/x86/include/asm/apic.h:541 [inline]
 smp_apic_timer_interrupt+0x16b/0x700 arch/x86/kernel/apic/apic.c:1052
 apic_timer_interrupt+0xa9/0xb0 arch/x86/entry/entry_64.S:829
 </IRQ>
RIP: 0010:arch_local_irq_restore arch/x86/include/asm/paravirt.h:777 [inline]
RIP: 0010:__raw_spin_unlock_irqrestore include/linux/spinlock_api_smp.h:160 [inline]
RIP: 0010:_raw_spin_unlock_irqrestore+0x5e/0xba kernel/locking/spinlock.c:184
RSP: 0018:ffff8801c20173c8 EFLAGS: 00000282 ORIG_RAX: ffffffffffffff12
RAX: dffffc0000000000 RBX: 0000000000000282 RCX: 0000000000000006
RDX: 1ffffffff0d592cd RSI: 1ffff10035d68d23 RDI: 0000000000000282
RBP: ffff8801c20173d8 R08: 1ffff10038402e47 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000000 R12: ffffffff8820e5c8
R13: ffff8801b1067ad8 R14: ffff8801aea7c268 R15: ffff8801aea7c278
 __debug_object_init+0x235/0x1040 lib/debugobjects.c:378
 debug_object_init+0x17/0x20 lib/debugobjects.c:391
 __init_work+0x2b/0x60 kernel/workqueue.c:506
 idletimer_tg_create net/netfilter/xt_IDLETIMER.c:152 [inline]
 idletimer_tg_checkentry+0x691/0xb00 net/netfilter/xt_IDLETIMER.c:213
 xt_check_target+0x22c/0x7d0 net/netfilter/x_tables.c:850
 check_target net/ipv6/netfilter/ip6_tables.c:533 [inline]
 find_check_entry.isra.7+0x935/0xcf0 net/ipv6/netfilter/ip6_tables.c:575
 translate_table+0xf52/0x1690 net/ipv6/netfilter/ip6_tables.c:744
 do_replace net/ipv6/netfilter/ip6_tables.c:1160 [inline]
 do_ip6t_set_ctl+0x370/0x5f0 net/ipv6/netfilter/ip6_tables.c:1686
 nf_sockopt net/netfilter/nf_sockopt.c:106 [inline]
 nf_setsockopt+0x67/0xc0 net/netfilter/nf_sockopt.c:115
 ipv6_setsockopt+0x10b/0x130 net/ipv6/ipv6_sockglue.c:927
 udpv6_setsockopt+0x45/0x80 net/ipv6/udp.c:1422
 sock_common_setsockopt+0x95/0xd0 net/core/sock.c:2976
 SYSC_setsockopt net/socket.c:1850 [inline]
 SyS_setsockopt+0x189/0x360 net/socket.c:1829
 do_syscall_64+0x282/0x940 arch/x86/entry/common.c:287

Fixes: 0902b469bd25 ("netfilter: xtables: idletimer target implementation")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzkaller <syzkaller@googlegroups.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
6 years agomlxsw: spectrum_router: Do not unconditionally clear route offload indication
Ido Schimmel [Fri, 16 Feb 2018 23:30:44 +0000 (00:30 +0100)]
mlxsw: spectrum_router: Do not unconditionally clear route offload indication

When mlxsw replaces (or deletes) a route it removes the offload
indication from the replaced route. This is problematic for IPv4 routes,
as the offload indication is stored in the fib_info which is usually
shared between multiple routes.

Instead of unconditionally clearing the offload indication, only clear
it if no other route is using the fib_info.

Fixes: 3984d1a89fe7 ("mlxsw: spectrum_router: Provide offload indication using nexthop flags")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reported-by: Alexander Petrovskiy <alexpe@mellanox.com>
Tested-by: Alexander Petrovskiy <alexpe@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'qualcomm-rmnet-Fix-issues-with-CONFIG_DEBUG_PREEMPT-enabled'
David S. Miller [Mon, 19 Feb 2018 16:17:34 +0000 (11:17 -0500)]
Merge branch 'qualcomm-rmnet-Fix-issues-with-CONFIG_DEBUG_PREEMPT-enabled'

Subash Abhinov Kasiviswanathan says:

====================
net: qualcomm: rmnet: Fix issues with CONFIG_DEBUG_PREEMPT enabled

Patch 1 and 2 fixes issues identified when CONFIG_DEBUG_PREEMPT was
enabled. These involve APIs which were called in invalid contexts.

Patch 3 is a null derefence fix identified by code inspection.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: qualcomm: rmnet: Fix possible null dereference in command processing
Subash Abhinov Kasiviswanathan [Fri, 16 Feb 2018 22:56:39 +0000 (15:56 -0700)]
net: qualcomm: rmnet: Fix possible null dereference in command processing

If a command packet with invalid mux id is received, the packet would
not have a valid endpoint. This invalid endpoint maybe dereferenced
leading to a crash. Identified by manual code inspection.

Fixes: 3352e6c45760 ("net: qualcomm: rmnet: Convert the muxed endpoint to hlist")
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: qualcomm: rmnet: Fix warning seen with 64 bit stats
Subash Abhinov Kasiviswanathan [Fri, 16 Feb 2018 22:56:38 +0000 (15:56 -0700)]
net: qualcomm: rmnet: Fix warning seen with 64 bit stats

With CONFIG_DEBUG_PREEMPT enabled, a warning was seen on device
creation. This occurs due to the incorrect cpu API usage in
ndo_get_stats64 handler.

BUG: using smp_processor_id() in preemptible [00000000] code: rmnetcli/5743
caller is debug_smp_processor_id+0x1c/0x24
Call trace:
[<ffffff9d48c8967c>] dump_backtrace+0x0/0x2a8
[<ffffff9d48c89bbc>] show_stack+0x20/0x28
[<ffffff9d4901fff8>] dump_stack+0xa8/0xe0
[<ffffff9d490421e0>] check_preemption_disabled+0x104/0x108
[<ffffff9d49042200>] debug_smp_processor_id+0x1c/0x24
[<ffffff9d494a36b0>] rmnet_get_stats64+0x64/0x13c
[<ffffff9d49b014e0>] dev_get_stats+0x68/0xd8
[<ffffff9d49d58df8>] rtnl_fill_stats+0x54/0x140
[<ffffff9d49b1f0b8>] rtnl_fill_ifinfo+0x428/0x9cc
[<ffffff9d49b23834>] rtmsg_ifinfo_build_skb+0x80/0xf4
[<ffffff9d49b23930>] rtnetlink_event+0x88/0xb4
[<ffffff9d48cd21b4>] raw_notifier_call_chain+0x58/0x78
[<ffffff9d49b028a4>] call_netdevice_notifiers_info+0x48/0x78
[<ffffff9d49b08bf8>] __netdev_upper_dev_link+0x290/0x5e8
[<ffffff9d49b08fcc>] netdev_master_upper_dev_link+0x3c/0x48
[<ffffff9d494a2e74>] rmnet_newlink+0xf0/0x1c8
[<ffffff9d49b23360>] rtnl_newlink+0x57c/0x6c8
[<ffffff9d49b2355c>] rtnetlink_rcv_msg+0xb0/0x244
[<ffffff9d49b5230c>] netlink_rcv_skb+0xb4/0xdc
[<ffffff9d49b204f4>] rtnetlink_rcv+0x34/0x44
[<ffffff9d49b51af0>] netlink_unicast+0x1ec/0x294
[<ffffff9d49b51fdc>] netlink_sendmsg+0x320/0x390
[<ffffff9d49ae6858>] sock_sendmsg+0x54/0x60
[<ffffff9d49ae91bc>] SyS_sendto+0x1a0/0x1e4
[<ffffff9d48c83770>] el0_svc_naked+0x24/0x28

Fixes: 192c4b5d48f2 ("net: qualcomm: rmnet: Add support for 64 bit stats")
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: qualcomm: rmnet: Fix crash on real dev unregistration
Subash Abhinov Kasiviswanathan [Fri, 16 Feb 2018 22:56:37 +0000 (15:56 -0700)]
net: qualcomm: rmnet: Fix crash on real dev unregistration

With CONFIG_DEBUG_PREEMPT enabled, a crash with the following call
stack was observed when removing a real dev which had rmnet devices
attached to it.
To fix this, remove the netdev_upper link APIs and instead use the
existing information in rmnet_port and rmnet_priv to get the
association between real and rmnet devs.

BUG: sleeping function called from invalid context
in_atomic(): 0, irqs_disabled(): 0, pid: 5762, name: ip
Preemption disabled at:
[<ffffff9d49043564>] debug_object_active_state+0xa4/0x16c
Internal error: Oops - BUG: 0 [#1] PREEMPT SMP
Modules linked in:
PC is at ___might_sleep+0x13c/0x180
LR is at ___might_sleep+0x17c/0x180
[<ffffff9d48ce0924>] ___might_sleep+0x13c/0x180
[<ffffff9d48ce09c0>] __might_sleep+0x58/0x8c
[<ffffff9d49d6253c>] mutex_lock+0x2c/0x48
[<ffffff9d48ed4840>] kernfs_remove_by_name_ns+0x48/0xa8
[<ffffff9d48ed6ec8>] sysfs_remove_link+0x30/0x58
[<ffffff9d49b05840>] __netdev_adjacent_dev_remove+0x14c/0x1e0
[<ffffff9d49b05914>] __netdev_adjacent_dev_unlink_lists+0x40/0x68
[<ffffff9d49b08820>] netdev_upper_dev_unlink+0xb4/0x1fc
[<ffffff9d494a29f0>] rmnet_dev_walk_unreg+0x6c/0xc8
[<ffffff9d49b00b40>] netdev_walk_all_lower_dev_rcu+0x58/0xb4
[<ffffff9d494a30fc>] rmnet_config_notify_cb+0xf4/0x134
[<ffffff9d48cd21b4>] raw_notifier_call_chain+0x58/0x78
[<ffffff9d49b028a4>] call_netdevice_notifiers_info+0x48/0x78
[<ffffff9d49b0b568>] rollback_registered_many+0x230/0x3c8
[<ffffff9d49b0b738>] unregister_netdevice_many+0x38/0x94
[<ffffff9d49b1e110>] rtnl_delete_link+0x58/0x88
[<ffffff9d49b201dc>] rtnl_dellink+0xbc/0x1cc
[<ffffff9d49b2355c>] rtnetlink_rcv_msg+0xb0/0x244
[<ffffff9d49b5230c>] netlink_rcv_skb+0xb4/0xdc
[<ffffff9d49b204f4>] rtnetlink_rcv+0x34/0x44
[<ffffff9d49b51af0>] netlink_unicast+0x1ec/0x294
[<ffffff9d49b51fdc>] netlink_sendmsg+0x320/0x390
[<ffffff9d49ae6858>] sock_sendmsg+0x54/0x60
[<ffffff9d49ae6f94>] ___sys_sendmsg+0x298/0x2b0
[<ffffff9d49ae98f8>] SyS_sendmsg+0xb4/0xf0
[<ffffff9d48c83770>] el0_svc_naked+0x24/0x28

Fixes: ceed73a2cf4a ("drivers: net: ethernet: qualcomm: rmnet: Initial implementation")
Fixes: 60d58f971c10 ("net: qualcomm: rmnet: Implement bridge mode")
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomac80211: Fix sending ADDBA response for an ongoing session
Ilan Peer [Mon, 19 Feb 2018 12:48:43 +0000 (14:48 +0200)]
mac80211: Fix sending ADDBA response for an ongoing session

In case an ADDBA request is received while there is already
an ongoing BA sessions with the same parameters, i.e., update
flow, an ADBBA response with decline status was sent twice. Fix it.

Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
6 years agomac80211: Do not disconnect on invalid operating class
Ilan Peer [Mon, 19 Feb 2018 12:48:42 +0000 (14:48 +0200)]
mac80211: Do not disconnect on invalid operating class

Some APs include a non global operating class in their extended channel
switch information element. In such a case, as the operating class is not
known, mac80211 would decide to disconnect.

However the specification states that the operating class needs to be
taken from Annex E, but it does not specify from which table it should be
taken, so it is valid for an AP to use a non global operating class.

To avoid possibly unneeded disconnection, in such a case ignore the
operating class and assume that the current band is used, and if the
resulting channel and band configuration is invalid disconnect.

Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
6 years agocfg80211: clear wep keys after disconnection
Avraham Stern [Mon, 19 Feb 2018 12:48:38 +0000 (14:48 +0200)]
cfg80211: clear wep keys after disconnection

When a low level driver calls cfg80211_disconnected(), wep keys are
not cleared. As a result, following connection requests will fail
since cfg80211 internal state shows a connection is still in progress.

Fix this by clearing the wep keys when disconnecting.

Signed-off-by: Avraham Stern <avraham.stern@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
6 years agomac80211: fix calling sleeping function in atomic context
Sara Sharon [Mon, 19 Feb 2018 12:48:37 +0000 (14:48 +0200)]
mac80211: fix calling sleeping function in atomic context

sta_info_alloc can be called from atomic paths (such as RX path)
so we need to call pcpu_alloc with the correct gfp.

Fixes: c9c5962b56c1 ("mac80211: enable collecting station statistics per-CPU")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
6 years agomac80211: fix a possible leak of station stats
Sara Sharon [Mon, 19 Feb 2018 12:48:35 +0000 (14:48 +0200)]
mac80211: fix a possible leak of station stats

If sta_info_alloc fails after allocating the per CPU statistics,
they are not properly freed.

Fixes: c9c5962b56c1 ("mac80211: enable collecting station statistics per-CPU")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
6 years agomac80211: round IEEE80211_TX_STATUS_HEADROOM up to multiple of 4
Felix Fietkau [Sat, 10 Feb 2018 12:20:34 +0000 (13:20 +0100)]
mac80211: round IEEE80211_TX_STATUS_HEADROOM up to multiple of 4

This ensures that mac80211 allocated management frames are properly
aligned, which makes copying them more efficient.
For instance, mt76 uses iowrite32_copy to copy beacon frames to beacon
template memory on the chip.
Misaligned 32-bit accesses cause CPU exceptions on MIPS and should be
avoided.

Signed-off-by: Felix Fietkau <nbd@nbd.name>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
6 years agomac80211_hwsim: don't use WQ_MEM_RECLAIM
Johannes Berg [Wed, 24 Jan 2018 07:40:51 +0000 (08:40 +0100)]
mac80211_hwsim: don't use WQ_MEM_RECLAIM

We're obviously not part of a memory reclaim path, so don't set the flag.

This also causes a warning in check_flush_dependency() since we end up
in a code path that flushes a non-reclaim workqueue, and we shouldn't do
that if we were really part of reclaim.

Reported-by: syzbot+41cdaf4232c50e658934@syzkaller.appspotmail.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
6 years agocfg80211: fix cfg80211_beacon_dup
Arnd Bergmann [Fri, 2 Feb 2018 15:31:23 +0000 (16:31 +0100)]
cfg80211: fix cfg80211_beacon_dup

gcc-8 warns about some obviously incorrect code:

net/mac80211/cfg.c: In function 'cfg80211_beacon_dup':
net/mac80211/cfg.c:2896:3: error: 'memcpy' source argument is the same as destination [-Werror=restrict]

From the context, I conclude that we want to copy from beacon into
new_beacon, as we do in the rest of the function.

Cc: stable@vger.kernel.org
Fixes: 73da7d5bab79 ("mac80211: add channel switch command and beacon callbacks")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
6 years agoLinux 4.16-rc2
Linus Torvalds [Mon, 19 Feb 2018 01:29:42 +0000 (17:29 -0800)]
Linux 4.16-rc2

6 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 18 Feb 2018 20:56:41 +0000 (12:56 -0800)]
Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull x86 Kconfig fixes from Thomas Gleixner:
 "Three patchlets to correct HIGHMEM64G and CMPXCHG64 dependencies in
  Kconfig when CPU selections are explicitely set to M586 or M686"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/Kconfig: Explicitly enumerate i686-class CPUs in Kconfig
  x86/Kconfig: Exclude i586-class CPUs lacking PAE support from the HIGHMEM64G Kconfig group
  x86/Kconfig: Add missing i586-class CPUs to the X86_CMPXCHG64 Kconfig group

6 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 18 Feb 2018 20:38:40 +0000 (12:38 -0800)]
Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull perf updates from Thomas Gleixner:
 "Perf tool updates and kprobe fixes:

   - perf_mmap overwrite mode fixes/overhaul, prep work to get 'perf
     top' using it, making it bearable to use it in large core count
     systems such as Knights Landing/Mill Intel systems (Kan Liang)

   - s/390 now uses syscall.tbl, just like x86-64 to generate the
     syscall table id -> string tables used by 'perf trace' (Hendrik
     Brueckner)

   - Use strtoull() instead of home grown function (Andy Shevchenko)

   - Synchronize kernel ABI headers, v4.16-rc1 (Ingo Molnar)

   - Document missing 'perf data --force' option (Sangwon Hong)

   - Add perf vendor JSON metrics for ARM Cortex-A53 Processor (William
     Cohen)

   - Improve error handling and error propagation of ftrace based
     kprobes so failures when installing kprobes are not silently
     ignored and create disfunctional tracepoints"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (27 commits)
  kprobes: Propagate error from disarm_kprobe_ftrace()
  kprobes: Propagate error from arm_kprobe_ftrace()
  Revert "tools include s390: Grab a copy of arch/s390/include/uapi/asm/unistd.h"
  perf s390: Rework system call table creation by using syscall.tbl
  perf s390: Grab a copy of arch/s390/kernel/syscall/syscall.tbl
  tools/headers: Synchronize kernel ABI headers, v4.16-rc1
  perf test: Fix test trace+probe_libc_inet_pton.sh for s390x
  perf data: Document missing --force option
  perf tools: Substitute yet another strtoull()
  perf top: Check the latency of perf_top__mmap_read()
  perf top: Switch default mode to overwrite mode
  perf top: Remove lost events checking
  perf hists browser: Add parameter to disable lost event warning
  perf top: Add overwrite fall back
  perf evsel: Expose the perf_missing_features struct
  perf top: Check per-event overwrite term
  perf mmap: Discard legacy interface for mmap read
  perf test: Update mmap read functions for backward-ring-buffer test
  perf mmap: Introduce perf_mmap__read_event()
  perf mmap: Introduce perf_mmap__read_done()
  ...

6 years agoMerge branch 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 18 Feb 2018 20:22:04 +0000 (12:22 -0800)]
Merge branch 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull irq updates from Thomas Gleixner:
 "A small set of updates mostly for irq chip drivers:

   - MIPS GIC fix for spurious, masked interrupts

   - fix for a subtle IPI bug in GICv3

   - do not probe GICv3 ITSs that are marked as disabled

   - multi-MSI support for GICv2m

   - various small cleanups"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqdomain: Re-use DEFINE_SHOW_ATTRIBUTE() macro
  irqchip/bcm: Remove hashed address printing
  irqchip/gic-v2m: Add PCI Multi-MSI support
  irqchip/gic-v3: Ignore disabled ITS nodes
  irqchip/gic-v3: Use wmb() instead of smb_wmb() in gic_raise_softirq()
  irqchip/gic-v3: Change pr_debug message to pr_devel
  irqchip/mips-gic: Avoid spuriously handling masked interrupts

6 years agoMerge branch 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 18 Feb 2018 19:54:22 +0000 (11:54 -0800)]
Merge branch 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull core fix from Thomas Gleixner:
 "A small fix which adds the missing for_each_cpu_wrap() stub for the UP
  case to avoid build failures"

* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  cpumask: Make for_each_cpu_wrap() available on UP as well

6 years agoMerge tag 'for-linus-20180217' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 17 Feb 2018 18:20:47 +0000 (10:20 -0800)]
Merge tag 'for-linus-20180217' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:

 - NVMe pull request from Keith, with fixes all over the map for nvme.
   From various folks.

 - Classic polling fix, that avoids a latency issue where we still end
   up waiting for an interrupt in some cases. From Nitesh Shetty.

 - Comment typo fix from Minwoo Im.

* tag 'for-linus-20180217' of git://git.kernel.dk/linux-block:
  block: fix a typo in comment of BLK_MQ_POLL_STATS_BKTS
  nvme-rdma: fix sysfs invoked reset_ctrl error flow
  nvmet: Change return code of discard command if not supported
  nvme-pci: Fix timeouts in connecting state
  nvme-pci: Remap CMB SQ entries on every controller reset
  nvme: fix the deadlock in nvme_update_formats
  blk: optimization for classic polling
  nvme: Don't use a stack buffer for keep-alive command
  nvme_fc: cleanup io completion
  nvme_fc: correct abort race condition on resets
  nvme: Fix discard buffer overrun
  nvme: delete NVME_CTRL_LIVE --> NVME_CTRL_CONNECTING transition
  nvme-rdma: use NVME_CTRL_CONNECTING state to mark init process
  nvme: rename NVME_CTRL_RECONNECTING state to NVME_CTRL_CONNECTING

6 years agoMerge tag 'mmc-v4.16-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc
Linus Torvalds [Sat, 17 Feb 2018 18:08:28 +0000 (10:08 -0800)]
Merge tag 'mmc-v4.16-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc

Pull MMC fixes from Ulf Hansson:

 - meson-gx: Revert to earlier tuning process

 - bcm2835: Don't overwrite max frequency unconditionally

* tag 'mmc-v4.16-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
  mmc: bcm2835: Don't overwrite max frequency unconditionally
  Revert "mmc: meson-gx: include tx phase in the tuning process"

6 years agoMerge tag 'mtd/fixes-for-4.16-rc2' of git://git.infradead.org/linux-mtd
Linus Torvalds [Sat, 17 Feb 2018 18:06:13 +0000 (10:06 -0800)]
Merge tag 'mtd/fixes-for-4.16-rc2' of git://git.infradead.org/linux-mtd

Pull mtd fixes from Boris Brezillon:

 - add missing dependency to NAND_MARVELL Kconfig entry

 - use the appropriate OOB layout in the VF610 driver

* tag 'mtd/fixes-for-4.16-rc2' of git://git.infradead.org/linux-mtd:
  mtd: nand: MTD_NAND_MARVELL should depend on HAS_DMA
  mtd: nand: vf610: set correct ooblayout

6 years agoMerge tag 'powerpc-4.16-3' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Sat, 17 Feb 2018 17:48:26 +0000 (09:48 -0800)]
Merge tag 'powerpc-4.16-3' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:
 "The main attraction is a fix for a bug in the new drmem code, which
  was causing an oops on boot on some versions of Qemu.

  There's also a fix for XIVE (Power9 interrupt controller) on KVM, as
  well as a few other minor fixes.

  Thanks to: Corentin Labbe, Cyril Bur, Cédric Le Goater, Daniel Black,
  Nathan Fontenot, Nicholas Piggin"

* tag 'powerpc-4.16-3' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/pseries: Check for zero filled ibm,dynamic-memory property
  powerpc/pseries: Add empty update_numa_cpu_lookup_table() for NUMA=n
  powerpc/powernv: IMC fix out of bounds memory access at shutdown
  powerpc/xive: Use hw CPU ids when configuring the CPU queues
  powerpc: Expose TSCR via sysfs only on powernv

6 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Sat, 17 Feb 2018 17:46:18 +0000 (09:46 -0800)]
Merge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux

Pull arm64 fixes from Catalin Marinas:
 "The bulk of this is the pte accessors annotation to READ/WRITE_ONCE
  (we tried to avoid pushing this during the merge window to avoid
  conflicts)

   - Updated the page table accessors to use READ/WRITE_ONCE and prevent
     compiler transformation that could lead to an apparent loss of
     coherency

   - Enabled branch predictor hardening for the Falkor CPU

   - Fix interaction between kpti enabling and KASan causing the
     recursive page table walking to take a significant time

   - Fix some sparse warnings"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: cputype: Silence Sparse warnings
  arm64: mm: Use READ_ONCE/WRITE_ONCE when accessing page tables
  arm64: proc: Set PTE_NG for table entries to avoid traversing them twice
  arm64: Add missing Falkor part number for branch predictor hardening

6 years agoMerge tag 'for-linus-4.16a-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 17 Feb 2018 17:16:09 +0000 (09:16 -0800)]
Merge tag 'for-linus-4.16a-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip

Pull xen fixes from Juergen Gross:

 - fixes for the Xen pvcalls frontend driver

 - fix for booting Xen pv domains

 - fix for the xenbus driver user interface

* tag 'for-linus-4.16a-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  pvcalls-front: wait for other operations to return when release passive sockets
  pvcalls-front: introduce a per sock_mapping refcount
  x86/xen: Calculate __max_logical_packages on PV domains
  xenbus: track caller request id

6 years agopvcalls-front: wait for other operations to return when release passive sockets
Stefano Stabellini [Wed, 14 Feb 2018 18:28:24 +0000 (10:28 -0800)]
pvcalls-front: wait for other operations to return when release passive sockets

Passive sockets can have ongoing operations on them, specifically, we
have two wait_event_interruptable calls in pvcalls_front_accept.

Add two wake_up calls in pvcalls_front_release, then wait for the
potential waiters to return and release the sock_mapping refcount.

Signed-off-by: Stefano Stabellini <stefano@aporeto.com>
Acked-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
6 years agopvcalls-front: introduce a per sock_mapping refcount
Stefano Stabellini [Wed, 14 Feb 2018 18:28:23 +0000 (10:28 -0800)]
pvcalls-front: introduce a per sock_mapping refcount

Introduce a per sock_mapping refcount, in addition to the existing
global refcount. Thanks to the sock_mapping refcount, we can safely wait
for it to be 1 in pvcalls_front_release before freeing an active socket,
instead of waiting for the global refcount to be 1.

Signed-off-by: Stefano Stabellini <stefano@aporeto.com>
Acked-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
6 years agox86/xen: Calculate __max_logical_packages on PV domains
Prarit Bhargava [Wed, 7 Feb 2018 23:49:23 +0000 (18:49 -0500)]
x86/xen: Calculate __max_logical_packages on PV domains

The kernel panics on PV domains because native_smp_cpus_done() is
only called for HVM domains.

Calculate __max_logical_packages for PV domains.

Fixes: b4c0a7326f5d ("x86/smpboot: Fix __max_logical_packages estimate")
Signed-off-by: Prarit Bhargava <prarit@redhat.com>
Tested-and-reported-by: Simon Gaiser <simon@invisiblethingslab.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: x86@kernel.org
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Dou Liyang <douly.fnst@cn.fujitsu.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Kate Stewart <kstewart@linuxfoundation.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: xen-devel@lists.xenproject.org
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
6 years agoxenbus: track caller request id
Joao Martins [Fri, 2 Feb 2018 17:42:33 +0000 (17:42 +0000)]
xenbus: track caller request id

Commit fd8aa9095a95 ("xen: optimize xenbus driver for multiple concurrent
xenstore accesses") optimized xenbus concurrent accesses but in doing so
broke UABI of /dev/xen/xenbus. Through /dev/xen/xenbus applications are in
charge of xenbus message exchange with the correct header and body. Now,
after the mentioned commit the replies received by application will no
longer have the header req_id echoed back as it was on request (see
specification below for reference), because that particular field is being
overwritten by kernel.

struct xsd_sockmsg
{
  uint32_t type;  /* XS_??? */
  uint32_t req_id;/* Request identifier, echoed in daemon's response.  */
  uint32_t tx_id; /* Transaction id (0 if not related to a transaction). */
  uint32_t len;   /* Length of data following this. */

  /* Generally followed by nul-terminated string(s). */
};

Before there was only one request at a time so req_id could simply be
forwarded back and forth. To allow simultaneous requests we need a
different req_id for each message thus kernel keeps a monotonic increasing
counter for this field and is written on every request irrespective of
userspace value.

Forwarding again the req_id on userspace requests is not a solution because
we would open the possibility of userspace-generated req_id colliding with
kernel ones. So this patch instead takes another route which is to
artificially keep user req_id while keeping the xenbus logic as is. We do
that by saving the original req_id before xs_send(), use the private kernel
counter as req_id and then once reply comes and was validated, we restore
back the original req_id.

Cc: <stable@vger.kernel.org> # 4.11
Fixes: fd8aa9095a ("xen: optimize xenbus driver for multiple concurrent xenstore accesses")
Reported-by: Bhavesh Davda <bhavesh.davda@oracle.com>
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
6 years agoarm64: cputype: Silence Sparse warnings
Robin Murphy [Fri, 16 Feb 2018 17:04:23 +0000 (17:04 +0000)]
arm64: cputype: Silence Sparse warnings

Sparse makes a fair bit of noise about our MPIDR mask being implicitly
long - let's explicitly describe it as such rather than just relying on
the value forcing automatic promotion.

Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
6 years agosctp: remove the left unnecessary check for chunk in sctp_renege_events
Xin Long [Fri, 16 Feb 2018 09:18:33 +0000 (17:18 +0800)]
sctp: remove the left unnecessary check for chunk in sctp_renege_events

Commit fb23403536ea ("sctp: remove the useless check in
sctp_renege_events") forgot to remove another check for
chunk in sctp_renege_events.

Dan found this when doing a static check.

This patch is to remove that check, and also to merge
two checks into one 'if statement'.

Fixes: fb23403536ea ("sctp: remove the useless check in sctp_renege_events")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-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 agorxrpc: Work around usercopy check
David Howells [Thu, 15 Feb 2018 22:59:00 +0000 (22:59 +0000)]
rxrpc: Work around usercopy check

Due to a check recently added to copy_to_user(), it's now not permitted to
copy from slab-held data to userspace unless the slab is whitelisted.  This
affects rxrpc_recvmsg() when it attempts to place an RXRPC_USER_CALL_ID
control message in the userspace control message buffer.  A warning is
generated by usercopy_warn() because the source is the copy of the
user_call_ID retained in the rxrpc_call struct.

Work around the issue by copying the user_call_ID to a variable on the
stack and passing that to put_cmsg().

The warning generated looks like:

Bad or missing usercopy whitelist? Kernel memory exposure attempt detected from SLUB object 'dmaengine-unmap-128' (offset 680, size 8)!
WARNING: CPU: 0 PID: 1401 at mm/usercopy.c:81 usercopy_warn+0x7e/0xa0
...
RIP: 0010:usercopy_warn+0x7e/0xa0
...
Call Trace:
 __check_object_size+0x9c/0x1a0
 put_cmsg+0x98/0x120
 rxrpc_recvmsg+0x6fc/0x1010 [rxrpc]
 ? finish_wait+0x80/0x80
 ___sys_recvmsg+0xf8/0x240
 ? __clear_rsb+0x25/0x3d
 ? __clear_rsb+0x15/0x3d
 ? __clear_rsb+0x25/0x3d
 ? __clear_rsb+0x15/0x3d
 ? __clear_rsb+0x25/0x3d
 ? __clear_rsb+0x15/0x3d
 ? __clear_rsb+0x25/0x3d
 ? __clear_rsb+0x15/0x3d
 ? finish_task_switch+0xa6/0x2b0
 ? trace_hardirqs_on_caller+0xed/0x180
 ? _raw_spin_unlock_irq+0x29/0x40
 ? __sys_recvmsg+0x4e/0x90
 __sys_recvmsg+0x4e/0x90
 do_syscall_64+0x7a/0x220
 entry_SYSCALL_64_after_hwframe+0x26/0x9b

Reported-by: Jonathan Billings <jsbillings@jsbillings.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Tested-by: Jonathan Billings <jsbillings@jsbillings.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotun: fix tun_napi_alloc_frags() frag allocator
Eric Dumazet [Thu, 15 Feb 2018 22:47:15 +0000 (14:47 -0800)]
tun: fix tun_napi_alloc_frags() frag allocator

<Mark Rutland reported>
    While fuzzing arm64 v4.16-rc1 with Syzkaller, I've been hitting a
    misaligned atomic in __skb_clone:

        atomic_inc(&(skb_shinfo(skb)->dataref));

   where dataref doesn't have the required natural alignment, and the
   atomic operation faults. e.g. i often see it aligned to a single
   byte boundary rather than a four byte boundary.

   AFAICT, the skb_shared_info is misaligned at the instant it's
   allocated in __napi_alloc_skb()  __napi_alloc_skb()
</end of report>

Problem is caused by tun_napi_alloc_frags() using
napi_alloc_frag() with user provided seg sizes,
leading to other users of this API getting unaligned
page fragments.

Since we would like to not necessarily add paddings or alignments to
the frags that tun_napi_alloc_frags() attaches to the skb, switch to
another page frag allocator.

As a bonus skb_page_frag_refill() can use GFP_KERNEL allocations,
meaning that we can not deplete memory reserves as easily.

Fixes: 90e33d459407 ("tun: enable napi_gro_frags() for TUN/TAP driver")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoudplite: fix partial checksum initialization
Alexey Kodanev [Thu, 15 Feb 2018 17:18:43 +0000 (20:18 +0300)]
udplite: fix partial checksum initialization

Since UDP-Lite is always using checksum, the following path is
triggered when calculating pseudo header for it:

  udp4_csum_init() or udp6_csum_init()
    skb_checksum_init_zero_check()
      __skb_checksum_validate_complete()

The problem can appear if skb->len is less than CHECKSUM_BREAK. In
this particular case __skb_checksum_validate_complete() also invokes
__skb_checksum_complete(skb). If UDP-Lite is using partial checksum
that covers only part of a packet, the function will return bad
checksum and the packet will be dropped.

It can be fixed if we skip skb_checksum_init_zero_check() and only
set the required pseudo header checksum for UDP-Lite with partial
checksum before udp4_csum_init()/udp6_csum_init() functions return.

Fixes: ed70fcfcee95 ("net: Call skb_checksum_init in IPv4")
Fixes: e4f45b7f40bd ("net: Call skb_checksum_init in IPv6")
Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoskbuff: Fix comment mis-spelling.
David S. Miller [Fri, 16 Feb 2018 20:52:42 +0000 (15:52 -0500)]
skbuff: Fix comment mis-spelling.

'peform' --> 'perform'

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodn_getsockoptdecnet: move nf_{get/set}sockopt outside sock lock
Paolo Abeni [Thu, 15 Feb 2018 15:59:49 +0000 (16:59 +0100)]
dn_getsockoptdecnet: move nf_{get/set}sockopt outside sock lock

After commit 3f34cfae1238 ("netfilter: on sockopt() acquire sock lock
only in the required scope"), the caller of nf_{get/set}sockopt() must
not hold any lock, but, in such changeset, I forgot to cope with DECnet.

This commit addresses the issue moving the nf call outside the lock,
in the dn_{get,set}sockopt() with the same schema currently used by
ipv4 and ipv6. Also moves the unhandled sockopts of the end of the main
switch statements, to improve code readability.

Reported-by: Petr Vandrovec <petr@vandrovec.name>
BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=198791#c2
Fixes: 3f34cfae1238 ("netfilter: on sockopt() acquire sock lock only in the required scope")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoPCI/cxgb4: Extend T3 PCI quirk to T4+ devices
Casey Leedom [Thu, 15 Feb 2018 14:33:18 +0000 (20:03 +0530)]
PCI/cxgb4: Extend T3 PCI quirk to T4+ devices

We've run into a problem where our device is attached
to a Virtual Machine and the use of the new pci_set_vpd_size()
API doesn't help.  The VM kernel has been informed that
the accesses are okay, but all of the actual VPD Capability
Accesses are trapped down into the KVM Hypervisor where it
goes ahead and imposes the silent denials.

The right idea is to follow the kernel.org
commit 1c7de2b4ff88 ("PCI: Enable access to non-standard VPD for
Chelsio devices (cxgb3)") which Alexey Kardashevskiy authored
to establish a PCI Quirk for our T3-based adapters. This commit
extends that PCI Quirk to cover Chelsio T4 devices and later.

The advantage of this approach is that the VPD Size gets set early
in the Base OS/Hypervisor Boot and doesn't require that the cxgb4
driver even be available in the Base OS/Hypervisor.  Thus PF4 can
be exported to a Virtual Machine and everything should work.

Fixes: 67e658794ca1 ("cxgb4: Set VPD size so we can read both VPD structures")
Cc: <stable@vger.kernel.org> # v4.9+
Signed-off-by: Casey Leedom <leedom@chelsio.com>
Signed-off-by: Arjun Vynipadath <arjun@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4: fix trailing zero in CIM LA dump
Rahul Lakkireddy [Thu, 15 Feb 2018 12:50:01 +0000 (18:20 +0530)]
cxgb4: fix trailing zero in CIM LA dump

Set correct size of the CIM LA dump for T6.

Fixes: 27887bc7cb7f ("cxgb4: collect hardware LA dumps")
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4: free up resources of pf 0-3
Ganesh Goudar [Thu, 15 Feb 2018 12:46:57 +0000 (18:16 +0530)]
cxgb4: free up resources of pf 0-3

free pf 0-3 resources, commit baf5086840ab ("cxgb4:
restructure VF mgmt code") erroneously removed the
code which frees the pf 0-3 resources, causing the
probe of pf 0-3 to fail in case of driver reload.

Fixes: baf5086840ab ("cxgb4: restructure VF mgmt code")
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'dma-mapping-4.16-2' of git://git.infradead.org/users/hch/dma-mapping
Linus Torvalds [Fri, 16 Feb 2018 20:22:33 +0000 (12:22 -0800)]
Merge tag 'dma-mapping-4.16-2' of git://git.infradead.org/users/hch/dma-mapping

Pull dma-mapping fixes from Christoph Hellwig:
 "A few dma-mapping fixes for the fallout from the changes in rc1"

* tag 'dma-mapping-4.16-2' of git://git.infradead.org/users/hch/dma-mapping:
  powerpc/macio: set a proper dma_coherent_mask
  dma-mapping: fix a comment typo
  dma-direct: comment the dma_direct_free calling convention
  dma-direct: mark as is_phys
  ia64: fix build failure with CONFIG_SWIOTLB

6 years agofib_semantics: Don't match route with mismatching tclassid
Stefano Brivio [Thu, 15 Feb 2018 08:46:03 +0000 (09:46 +0100)]
fib_semantics: Don't match route with mismatching tclassid

In fib_nh_match(), if output interface or gateway are passed in
the FIB configuration, we don't have to check next hops of
multipath routes to conclude whether we have a match or not.

However, we might still have routes with different realms
matching the same output interface and gateway configuration,
and this needs to cause the match to fail. Otherwise the first
route inserted in the FIB will match, regardless of the realms:

 # ip route add 1.1.1.1 dev eth0 table 1234 realms 1/2
 # ip route append 1.1.1.1 dev eth0 table 1234 realms 3/4
 # ip route list table 1234
 1.1.1.1 dev eth0 scope link realms 1/2
 1.1.1.1 dev eth0 scope link realms 3/4
 # ip route del 1.1.1.1 dev ens3 table 1234 realms 3/4
 # ip route list table 1234
 1.1.1.1 dev ens3 scope link realms 3/4

whereas route with realms 3/4 should have been deleted instead.

Explicitly check for fc_flow passed in the FIB configuration
(this comes from RTA_FLOW extracted by rtm_to_fib_config()) and
fail matching if it differs from nh_tclassid.

The handling of RTA_FLOW for multipath routes later in
fib_nh_match() is still needed, as we can have multiple RTA_FLOW
attributes that need to be matched against the tclassid of each
next hop.

v2: Check that fc_flow is set before discarding the match, so
    that the user can still select the first matching rule by
    not specifying any realm, as suggested by David Ahern.

Reported-by: Jianlin Shi <jishi@redhat.com>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoNFC: llcp: Limit size of SDP URI
Kees Cook [Wed, 14 Feb 2018 23:45:07 +0000 (15:45 -0800)]
NFC: llcp: Limit size of SDP URI

The tlv_len is u8, so we need to limit the size of the SDP URI. Enforce
this both in the NLA policy and in the code that performs the allocation
and copy, to avoid writing past the end of the allocated buffer.

Fixes: d9b8d8e19b073 ("NFC: llcp: Service Name Lookup netlink interface")
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoarm64: mm: Use READ_ONCE/WRITE_ONCE when accessing page tables
Will Deacon [Thu, 15 Feb 2018 11:14:56 +0000 (11:14 +0000)]
arm64: mm: Use READ_ONCE/WRITE_ONCE when accessing page tables

In many cases, page tables can be accessed concurrently by either another
CPU (due to things like fast gup) or by the hardware page table walker
itself, which may set access/dirty bits. In such cases, it is important
to use READ_ONCE/WRITE_ONCE when accessing page table entries so that
entries cannot be torn, merged or subject to apparent loss of coherence
due to compiler transformations.

Whilst there are some scenarios where this cannot happen (e.g. pinned
kernel mappings for the linear region), the overhead of using READ_ONCE
/WRITE_ONCE everywhere is minimal and makes the code an awful lot easier
to reason about. This patch consistently uses these macros in the arch
code, as well as explicitly namespacing pointers to page table entries
from the entries themselves by using adopting a 'p' suffix for the former
(as is sometimes used elsewhere in the kernel source).

Tested-by: Yury Norov <ynorov@caviumnetworks.com>
Tested-by: Richard Ruigrok <rruigrok@codeaurora.org>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
6 years agomm: hide a #warning for COMPILE_TEST
Arnd Bergmann [Fri, 16 Feb 2018 15:25:53 +0000 (16:25 +0100)]
mm: hide a #warning for COMPILE_TEST

We get a warning about some slow configurations in randconfig kernels:

  mm/memory.c:83:2: error: #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. [-Werror=cpp]

The warning is reasonable by itself, but gets in the way of randconfig
build testing, so I'm hiding it whenever CONFIG_COMPILE_TEST is set.

The warning was added in 2013 in commit 75980e97dacc ("mm: fold
page->_last_nid into page->flags where possible").

Cc: stable@vger.kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoMerge tag 'mips_fixes_4.16_2' of git://git.kernel.org/pub/scm/linux/kernel/git/jhogan...
Linus Torvalds [Fri, 16 Feb 2018 17:31:37 +0000 (09:31 -0800)]
Merge tag 'mips_fixes_4.16_2' of git://git.kernel.org/pub/scm/linux/kernel/git/jhogan/mips

Pull MIPS fixes from James Hogan:
 "A few fixes for outstanding MIPS issues:

   - an __init section mismatch warning when brcmstb_pm is enabled

   - a regression handling multiple mem=X@Y arguments (4.11)

   - a USB Kconfig select warning, and related sparc cleanup (4.16)"

* tag 'mips_fixes_4.16_2' of git://git.kernel.org/pub/scm/linux/kernel/git/jhogan/mips:
  sparc,leon: Select USB_UHCI_BIG_ENDIAN_{MMIO,DESC}
  usb: Move USB_UHCI_BIG_ENDIAN_* out of USB_SUPPORT
  MIPS: Fix incorrect mem=X@Y handling
  MIPS: BMIPS: Fix section mismatch warning

6 years agoMerge tag 'for-4.16-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
Linus Torvalds [Fri, 16 Feb 2018 17:26:18 +0000 (09:26 -0800)]
Merge tag 'for-4.16-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux

Pull btrfs fixes from David Sterba:
 "We have a few assorted fixes, some of them show up during fstests so I
  gave them more testing"

* tag 'for-4.16-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: Fix use-after-free when cleaning up fs_devs with a single stale device
  Btrfs: fix null pointer dereference when replacing missing device
  btrfs: remove spurious WARN_ON(ref->count < 0) in find_parent_nodes
  btrfs: Ignore errors from btrfs_qgroup_trace_extent_post
  Btrfs: fix unexpected -EEXIST when creating new inode
  Btrfs: fix use-after-free on root->orphan_block_rsv
  Btrfs: fix btrfs_evict_inode to handle abnormal inodes correctly
  Btrfs: fix extent state leak from tree log
  Btrfs: fix crash due to not cleaning up tree log block's dirty bits
  Btrfs: fix deadlock in run_delalloc_nocow

6 years agoMerge tag 'for-4.16/dm-chained-bios-fix' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 16 Feb 2018 17:23:36 +0000 (09:23 -0800)]
Merge tag 'for-4.16/dm-chained-bios-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

Pull device mapper fix from Mike Snitzer:
 "Fix for DM core to properly propagate errors (avoids overriding
  non-zero error with 0). This is particularly important given DM core's
  increased use of chained bios"

* tag 'for-4.16/dm-chained-bios-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm: correctly handle chained bios in dec_pending()

6 years agoMerge tag 'platform-drivers-x86-v4.16-4' of git://git.infradead.org/linux-platform...
Linus Torvalds [Fri, 16 Feb 2018 17:20:00 +0000 (09:20 -0800)]
Merge tag 'platform-drivers-x86-v4.16-4' of git://git.infradead.org/linux-platform-drivers-x86

Pull x86 platform driver fixes from Andy Shevchenko:

 - regression fix in keyboard support for Dell laptops

 - prevent out-of-boundary write in WMI bus driver

 - increase timeout to read functional key status on Lenovo laptops

* tag 'platform-drivers-x86-v4.16-4' of git://git.infradead.org/linux-platform-drivers-x86:
  platform/x86: dell-laptop: Removed duplicates in DMI whitelist
  platform/x86: dell-laptop: fix kbd_get_state's request value
  platform/x86: ideapad-laptop: Increase timeout to wait for EC answer
  platform/x86: wmi: fix off-by-one write in wmi_dev_probe()

6 years agoMerge tag 'sound-4.16-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Fri, 16 Feb 2018 17:11:30 +0000 (09:11 -0800)]
Merge tag 'sound-4.16-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "A collection of usual suspects:

   - a handful USB-audio and HD-audio device-specific quirks

   - some trivial fixes for the new AC97 bus stuff

   - another race fix in ALSA sequencer core"

* tag 'sound-4.16-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: hda/realtek: PCI quirk for Fujitsu U7x7
  ALSA: seq: Fix racy pool initializations
  ALSA: usb: add more device quirks for USB DSD devices
  ALSA: usb-audio: Fix UAC2 get_ctl request with a RANGE attribute
  ALSA: ac97: Fix copy and paste typo in documentation
  ALSA: usb-audio: add implicit fb quirk for Behringer UFX1204
  ALSA: ac97: kconfig: Remove select of undefined symbol AC97
  ALSA: hda/realtek - Enable Thinkpad Dock device for ALC298 platform
  ALSA: hda/realtek - Add headset mode support for Dell laptop
  ALSA: hda - Fix headset mic detection problem for two Dell machines

6 years agoMerge tag 'drm-fixes-for-v4.16-rc2' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Fri, 16 Feb 2018 17:08:59 +0000 (09:08 -0800)]
Merge tag 'drm-fixes-for-v4.16-rc2' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "One nouveau regression fix, one AMD quirk and a full set of i915
  fixes.

  The i915 fixes are mostly for things caught by their CI system, main
  ones being DSI panel fixes and GEM fixes"

* tag 'drm-fixes-for-v4.16-rc2' of git://people.freedesktop.org/~airlied/linux:
  drm/nouveau: Make clock gate support conditional
  drm/i915: Fix DSI panels with v1 MIPI sequences without a DEASSERT sequence v3
  drm/i915: Free memdup-ed DSI VBT data structures on driver_unload
  drm/i915: Add intel_bios_cleanup() function
  drm/i915/vlv: Add cdclk workaround for DSI
  drm/i915/gvt: fix one typo of render_mmio trace
  drm/i915/gvt: Support BAR0 8-byte reads/writes
  drm/i915/gvt: add 0xe4f0 into gen9 render list
  drm/i915/pmu: Fix building without CONFIG_PM
  drm/i915/pmu: Fix sleep under atomic in RC6 readout
  drm/i915/pmu: Fix PMU enable vs execlists tasklet race
  drm/i915: Lock out execlist tasklet while peeking inside for busy-stats
  drm/i915/breadcrumbs: Ignore unsubmitted signalers
  drm/i915: Don't wake the device up to check if the engine is asleep
  drm/i915: Avoid truncation before clamping userspace's priority value
  drm/i915/perf: Fix compiler warning for string truncation
  drm/i915/perf: Fix compiler warning for string truncation
  drm/amdgpu: add new device to use atpx quirk

6 years agodm: correctly handle chained bios in dec_pending()
NeilBrown [Thu, 15 Feb 2018 09:00:15 +0000 (20:00 +1100)]
dm: correctly handle chained bios in dec_pending()

dec_pending() is given an error status (possibly 0) to be recorded
against a bio.  It can be called several times on the one 'struct
dm_io', and it is careful to only assign a non-zero error to
io->status.  However when it then assigned io->status to bio->bi_status,
it is not careful and could overwrite a genuine error status with 0.

This can happen when chained bios are in use.  If a bio is chained
beneath the bio that this dm_io is handling, the child bio might
complete and set bio->bi_status before the dm_io completes.

This has been possible since chained bios were introduced in 3.14, and
has become a lot easier to trigger with commit 18a25da84354 ("dm: ensure
bio submission follows a depth-first tree walk") as that commit caused
dm to start using chained bios itself.

A particular failure mode is that if a bio spans an 'error' target and a
working target, the 'error' fragment will complete instantly and set the
->bi_status, and the other fragment will normally complete a little
later, and will clear ->bi_status.

The fix is simply to only assign io_error to bio->bi_status when
io_error is not zero.

Reported-and-tested-by: Milan Broz <gmazyland@gmail.com>
Cc: stable@vger.kernel.org (v3.14+)
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>