openwrt/staging/blogic.git
6 years agonet/mlx4_en: optimizes get_fixed_ipv6_csum()
Eric Dumazet [Thu, 19 Apr 2018 15:49:29 +0000 (08:49 -0700)]
net/mlx4_en: optimizes get_fixed_ipv6_csum()

While trying to support CHECKSUM_COMPLETE for IPV6 fragments,
I had to experiments various hacks in get_fixed_ipv6_csum().
I must admit I could not find how to implement this :/

However, get_fixed_ipv6_csum() does a lot of redundant operations,
calling csum_partial() twice.

First csum_partial() computes the checksum of saddr and daddr,
put in @csum_pseudo_hdr. Undone later in the second csum_partial()
computed on whole ipv6 header.

Then nexthdr is added once, added a second time, then substracted.

payload_len is added once, then substracted.

Really all this can be reduced to two add_csum(), to add back 6 bytes
that were removed by mlx4 when providing hw_checksum in RX descriptor.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Saeed Mahameed <saeedm@mellanox.com>
Cc: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'smc-splice-implementation'
David S. Miller [Fri, 4 May 2018 15:45:12 +0000 (11:45 -0400)]
Merge branch 'smc-splice-implementation'

Ursula Braun says:

====================
net/smc: splice implementation

Stefan comes up with an smc implementation for splice(). The first
three patches are preparational patches, the 4th patch implements
splice().
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosmc: add support for splice()
Stefan Raspl [Thu, 3 May 2018 16:12:39 +0000 (18:12 +0200)]
smc: add support for splice()

Provide an implementation for splice() when we are using SMC. See
smc_splice_read() for further details.

Signed-off-by: Stefan Raspl <raspl@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com><
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosmc: allocate RMBs as compound pages
Stefan Raspl [Thu, 3 May 2018 16:12:38 +0000 (18:12 +0200)]
smc: allocate RMBs as compound pages

Preparatory work for splice() support.

Signed-off-by: Stefan Raspl <raspl@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com><
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosmc: make smc_rx_wait_data() generic
Stefan Raspl [Thu, 3 May 2018 16:12:37 +0000 (18:12 +0200)]
smc: make smc_rx_wait_data() generic

Turn smc_rx_wait_data into a generic function that can be used at various
instances to wait on traffic to complete with varying criteria.

Signed-off-by: Stefan Raspl <raspl@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com><
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosmc: simplify abort logic
Stefan Raspl [Thu, 3 May 2018 16:12:36 +0000 (18:12 +0200)]
smc: simplify abort logic

Some of the conditions to exit recv() are common in two pathes - cleaning up
code by moving the check up so we have it only once.

Signed-off-by: Stefan Raspl <raspl@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.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
David S. Miller [Fri, 4 May 2018 13:58:56 +0000 (09:58 -0400)]
Merge git://git./linux/kernel/git/davem/net

Overlapping changes in selftests Makefile.

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'sh_eth-complain-on-access-to-unimplemented-TSU-registers'
David S. Miller [Fri, 4 May 2018 13:11:50 +0000 (09:11 -0400)]
Merge branch 'sh_eth-complain-on-access-to-unimplemented-TSU-registers'

Sergei Shtylyov says:

====================
sh_eth: complain on access to unimplemented TSU registers

Here's a set of 2 patches against DaveM's 'net-next.git' repo. The 1st patch
routes TSU_POST<n> register accesses thru sh_eth_tsu_{read|write}() and the 2nd
added WARN_ON() unimplemented register to those functions. I'm going to deal with
TSU_ADR{H|L}<n> registers in a later series...
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosh_eth: WARN_ON() access to unimplemented TSU register
Sergei Shtylyov [Wed, 2 May 2018 19:55:52 +0000 (22:55 +0300)]
sh_eth: WARN_ON() access to unimplemented TSU register

Commit 3365711df024 ("sh_eth: WARN on access to a register not implemented
in a particular chip") added  WARN_ON() to sh_eth_{read|write}() but not
to sh_eth_tsu_{read|write}(). Now that we've routed almost all TSU register
accesses  (except TSU_ADR{H|L}<n> -- which are special) thru the latter
pair of accessors, it makes sense to check for the unimplemented TSU
registers as well...

Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosh_eth: use TSU register accessors for TSU_POST<n>
Sergei Shtylyov [Wed, 2 May 2018 19:54:48 +0000 (22:54 +0300)]
sh_eth: use TSU register accessors for TSU_POST<n>

There's no particularly good reason TSU_POST<n> registers get accessed
circumventing sh_eth_tsu_{read|write}() -- start using those, removing
(badly named) sh_eth_tsu_get_post_reg_offset(),  while at it...

Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'linux-kselftest-4.17-rc4' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 4 May 2018 05:26:51 +0000 (19:26 -1000)]
Merge tag 'linux-kselftest-4.17-rc4' of git://git./linux/kernel/git/shuah/linux-kselftest

Pull kselftest fixes from Shuah Khan:
 "This Kselftest update for 4.17-rc4 consists of a fix for a syntax
  error in the script that runs selftests. Mathieu Desnoyers found this
  bug in the script on systems running GNU Make 3.8 or older"

* tag 'linux-kselftest-4.17-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest:
  selftests: Fix lib.mk run_tests target shell script

6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Fri, 4 May 2018 04:57:03 +0000 (18:57 -1000)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Various sockmap fixes from John Fastabend (pinned map handling,
    blocking in recvmsg, double page put, error handling during redirect
    failures, etc.)

 2) Fix dead code handling in x86-64 JIT, from Gianluca Borello.

 3) Missing device put in RDS IB code, from Dag Moxnes.

 4) Don't process fast open during repair mode in TCP< from Yuchung
    Cheng.

 5) Move address/port comparison fixes in SCTP, from Xin Long.

 6) Handle add a bond slave's master into a bridge properly, from
    Hangbin Liu.

 7) IPv6 multipath code can operate on unitialized memory due to an
    assumption that the icmp header is in the linear SKB area. Fix from
    Eric Dumazet.

 8) Don't invoke do_tcp_sendpages() recursively via TLS, from Dave
    Watson.

9) Fix memory leaks in x86-64 JIT, from Daniel Borkmann.

10) RDS leaks kernel memory to userspace, from Eric Dumazet.

11) DCCP can invoke a tasklet on a freed socket, take a refcount. Also
    from Eric Dumazet.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (78 commits)
  dccp: fix tasklet usage
  smc: fix sendpage() call
  net/smc: handle unregistered buffers
  net/smc: call consolidation
  qed: fix spelling mistake: "offloded" -> "offloaded"
  net/mlx5e: fix spelling mistake: "loobpack" -> "loopback"
  tcp: restore autocorking
  rds: do not leak kernel memory to user land
  qmi_wwan: do not steal interfaces from class drivers
  ipv4: fix fnhe usage by non-cached routes
  bpf: sockmap, fix error handling in redirect failures
  bpf: sockmap, zero sg_size on error when buffer is released
  bpf: sockmap, fix scatterlist update on error path in send with apply
  net_sched: fq: take care of throttled flows before reuse
  ipv6: Revert "ipv6: Allow non-gateway ECMP for IPv6"
  bpf, x64: fix memleak when not converging on calls
  bpf, x64: fix memleak when not converging after image
  net/smc: restrict non-blocking connect finish
  8139too: Use disable_irq_nosync() in rtl8139_poll_controller()
  sctp: fix the issue that the cookie-ack with auth can't get processed
  ...

6 years agoMerge branch 'parisc-4.17-4' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Fri, 4 May 2018 04:31:19 +0000 (18:31 -1000)]
Merge branch 'parisc-4.17-4' of git://git./linux/kernel/git/deller/parisc-linux

Pull parisc fixes from Helge Deller:
 "Fix two section mismatches, convert to read_persistent_clock64(), add
  further documentation regarding the HPMC crash handler and make
  bzImage the default build target"

* 'parisc-4.17-4' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Fix section mismatches
  parisc: drivers.c: Fix section mismatches
  parisc: time: Convert read_persistent_clock() to read_persistent_clock64()
  parisc: Document rules regarding checksum of HPMC handler
  parisc: Make bzImage default build target

6 years agodccp: fix tasklet usage
Eric Dumazet [Thu, 3 May 2018 16:39:20 +0000 (09:39 -0700)]
dccp: fix tasklet usage

syzbot reported a crash in tasklet_action_common() caused by dccp.

dccp needs to make sure socket wont disappear before tasklet handler
has completed.

This patch takes a reference on the socket when arming the tasklet,
and moves the sock_put() from dccp_write_xmit_timer() to dccp_write_xmitlet()

kernel BUG at kernel/softirq.c:514!
invalid opcode: 0000 [#1] SMP KASAN
Dumping ftrace buffer:
   (ftrace buffer empty)
Modules linked in:
CPU: 1 PID: 17 Comm: ksoftirqd/1 Not tainted 4.17.0-rc3+ #30
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
RIP: 0010:tasklet_action_common.isra.19+0x6db/0x700 kernel/softirq.c:515
RSP: 0018:ffff8801d9b3faf8 EFLAGS: 00010246
dccp_close: ABORT with 65423 bytes unread
RAX: 1ffff1003b367f6b RBX: ffff8801daf1f3f0 RCX: 0000000000000000
RDX: ffff8801cf895498 RSI: 0000000000000004 RDI: 0000000000000000
RBP: ffff8801d9b3fc40 R08: ffffed0039f12a95 R09: ffffed0039f12a94
dccp_close: ABORT with 65423 bytes unread
R10: ffffed0039f12a94 R11: ffff8801cf8954a3 R12: 0000000000000000
R13: ffff8801d9b3fc18 R14: dffffc0000000000 R15: ffff8801cf895490
FS:  0000000000000000(0000) GS:ffff8801daf00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000001b2bc28000 CR3: 00000001a08a9000 CR4: 00000000001406e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 tasklet_action+0x1d/0x20 kernel/softirq.c:533
 __do_softirq+0x2e0/0xaf5 kernel/softirq.c:285
dccp_close: ABORT with 65423 bytes unread
 run_ksoftirqd+0x86/0x100 kernel/softirq.c:646
 smpboot_thread_fn+0x417/0x870 kernel/smpboot.c:164
 kthread+0x345/0x410 kernel/kthread.c:238
 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:412
Code: 48 8b 85 e8 fe ff ff 48 8b 95 f0 fe ff ff e9 94 fb ff ff 48 89 95 f0 fe ff ff e8 81 53 6e 00 48 8b 95 f0 fe ff ff e9 62 fb ff ff <0f> 0b 48 89 cf 48 89 8d e8 fe ff ff e8 64 53 6e 00 48 8b 8d e8
RIP: tasklet_action_common.isra.19+0x6db/0x700 kernel/softirq.c:515 RSP: ffff8801d9b3faf8

Fixes: dc841e30eaea ("dccp: Extend CCID packet dequeueing interface")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Cc: dccp@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'smc-fixes'
David S. Miller [Thu, 3 May 2018 18:47:32 +0000 (14:47 -0400)]
Merge branch 'smc-fixes'

Ursula Braun says:

====================
net/smc: fixes 2018/05/03

here are smc fixes for 2 problems:
 * receive buffers in SMC must be registered. If registration fails
   these buffers must not be kept within the link group for reuse.
   Patch 1 is a preparational patch; patch 2 contains the fix.
 * sendpage: do not hold the sock lock when calling kernel_sendpage()
             or sock_no_sendpage()
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosmc: fix sendpage() call
Stefan Raspl [Thu, 3 May 2018 15:57:39 +0000 (17:57 +0200)]
smc: fix sendpage() call

The sendpage() call grabs the sock lock before calling the default
implementation - which tries to grab it once again.

Signed-off-by: Stefan Raspl <raspl@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com><
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/smc: handle unregistered buffers
Karsten Graul [Thu, 3 May 2018 15:57:38 +0000 (17:57 +0200)]
net/smc: handle unregistered buffers

When smc_wr_reg_send() fails then tag (regerr) the affected buffer and
free it in smc_buf_unuse().

Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/smc: call consolidation
Karsten Graul [Thu, 3 May 2018 15:57:37 +0000 (17:57 +0200)]
net/smc: call consolidation

Consolidate the call to smc_wr_reg_send() in a new function.
No functional changes.

Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoqed: fix spelling mistake: "offloded" -> "offloaded"
Colin Ian King [Thu, 3 May 2018 15:19:32 +0000 (16:19 +0100)]
qed: fix spelling mistake: "offloded" -> "offloaded"

Trivial fix to spelling mistake in DP_NOTICE message

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'bridge-FDB-Notify-about-removal-of-non-user-added-entries'
David S. Miller [Thu, 3 May 2018 17:46:48 +0000 (13:46 -0400)]
Merge branch 'bridge-FDB-Notify-about-removal-of-non-user-added-entries'

Petr Machata says:

====================
bridge: FDB: Notify about removal of non-user-added entries

Device drivers may generally need to keep in sync with bridge's FDB. In
particular, for its offload of tc mirror action where the mirrored-to
device is a gretap device, mlxsw needs to listen to a number of events,
FDB events among the others. SWITCHDEV_FDB_{ADD,DEL}_TO_DEVICE would be
a natural notification in that case.

However, for removal of FDB entries added due to device activity (as
opposed to explicit addition through "bridge fdb add" or similar), there
are no notifications.

Thus in patch #1, add the "added_by_user" field to switchdev
notifications sent for FDB activity. Adapt drivers to ignore activity on
non-user-added entries, to maintain the current behavior. Specifically
in case of mlxsw, allow mlxsw_sp_span_respin() call for any and all FDB
updates.

In patch #2, change the bridge driver to actually emit notifications for
these FDB entries. Take care not to send notification for bridge
updates that itself originate in SWITCHDEV_FDB_*_TO_BRIDGE events.

Changes from v1 to v2:
- Instead of introducing a new variant of fdb_delete(), add a new
  parameter to the existing function.
- Name the parameter swdev_notify, not notify.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: bridge: Notify about !added_by_user FDB entries
Petr Machata [Thu, 3 May 2018 12:43:53 +0000 (14:43 +0200)]
net: bridge: Notify about !added_by_user FDB entries

Do not automatically bail out on sending notifications about activity on
non-user-added FDB entries. Instead, notify about this activity except
for cases where the activity itself originates in a notification, to
avoid sending duplicate notifications.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Acked-by: Ivan Vecera <ivecera@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoswitchdev: Add fdb.added_by_user to switchdev notifications
Petr Machata [Thu, 3 May 2018 12:43:46 +0000 (14:43 +0200)]
switchdev: Add fdb.added_by_user to switchdev notifications

The following patch enables sending notifications also for events on FDB
entries that weren't added by the user. Give the drivers the information
necessary to distinguish between the two origins of FDB entries.

To maintain the current behavior, have switchdev-implementing drivers
bail out on notifications about non-user-added FDB entries. In case of
mlxsw driver, allow a call to mlxsw_sp_span_respin() so that SPAN over
bridge catches up with the changed FDB.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Acked-by: Ivan Vecera <ivecera@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'mlxsw-Introduce-support-for-CQEv1-2'
David S. Miller [Thu, 3 May 2018 17:44:43 +0000 (13:44 -0400)]
Merge branch 'mlxsw-Introduce-support-for-CQEv1-2'

Ido Schimmel says:

====================
mlxsw: Introduce support for CQEv1/2

Jiri says:

Current SwitchX2 and Spectrum FWs support CQEv0 and that is what we
implement in mlxsw. Spectrum FW also supports CQE v1 and v2.
However, Spectrum-2 won't support CQEv0. Prepare for it and setup the
CQE versions to use according to what is queried from FW.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: pci: Check number of CQEs for CQE version 2
Jiri Pirko [Thu, 3 May 2018 11:59:42 +0000 (14:59 +0300)]
mlxsw: pci: Check number of CQEs for CQE version 2

Check number of CQEs for CQE version 2 reported by QUERY_AQ_CAP command.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: pci: Allow to use CQEs of version 1 and version 2
Jiri Pirko [Thu, 3 May 2018 11:59:41 +0000 (14:59 +0300)]
mlxsw: pci: Allow to use CQEs of version 1 and version 2

Use previously added resources to query FW support for multiple versions
of CQEs. Use the biggest version supported. For SDQs, it has no sense to
use version 2 as it does not introduce any new features, but it is
twice the size of CQE version 1.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: pci: Introduce helpers to work with multiple CQE versions
Jiri Pirko [Thu, 3 May 2018 11:59:40 +0000 (14:59 +0300)]
mlxsw: pci: Introduce helpers to work with multiple CQE versions

Introduce definitions of fields in CQE version 1 and 2. Also, introduce
common helpers that would call appropriate version-specific helpers
according to the version enum passed.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: resources: Add CQE versions resources
Jiri Pirko [Thu, 3 May 2018 11:59:39 +0000 (14:59 +0300)]
mlxsw: resources: Add CQE versions resources

Add resources that FW uses to report supported CQE versions.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: bridge: avoid duplicate notification on up/down/change netdev events
Nikolay Aleksandrov [Thu, 3 May 2018 10:47:24 +0000 (13:47 +0300)]
net: bridge: avoid duplicate notification on up/down/change netdev events

While handling netdevice events, br_device_event() sometimes uses
br_stp_(disable|enable)_port which unconditionally send a notification,
but then a second notification for the same event is sent at the end of
the br_device_event() function. To avoid sending duplicate notifications
in such cases, check if one has already been sent (i.e.
br_stp_enable/disable_port have been called).
The patch is based on a change by Satish Ashok.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'selftests-forwarding-sysctl'
David S. Miller [Thu, 3 May 2018 17:37:03 +0000 (13:37 -0400)]
Merge branch 'selftests-forwarding-sysctl'

Petr Machata says:

====================
selftests: forwarding: Updates to sysctl handling

Some selftests need to adjust sysctl settings. In order to be neutral to
the system that the test is run on, it is a good practice to change back
to the original setting after the test ends. That involves some
boilerplate that can be abstracted away.

In patch #1, introduce two functions, sysctl_set() and sysctl_restore().
The former stores the current value of a given setting, and sets a new
value. The latter restores the setting to the previously-stored value.

In patch #2, use these wrappers in a number of tests.

Additionally in patch #3, fix a problem in mirror_gre_nh.sh, which
neglected to set a sysctl that's crucial for the test to work.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: forwarding: mirror_gre_nh: Unset RP filter
Petr Machata [Thu, 3 May 2018 10:37:21 +0000 (12:37 +0200)]
selftests: forwarding: mirror_gre_nh: Unset RP filter

The test fails to work if reverse-path filtering is in effect on the
mirrored-to host interface, or for all interfaces.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: forwarding: Use sysctl_set(), sysctl_restore()
Petr Machata [Thu, 3 May 2018 10:37:13 +0000 (12:37 +0200)]
selftests: forwarding: Use sysctl_set(), sysctl_restore()

Instead of hand-managing the sysctl set and restore, use the wrappers
sysctl_set() and sysctl_restore() to do the bookkeeping automatically.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: forwarding: lib: Add sysctl_set(), sysctl_restore()
Petr Machata [Thu, 3 May 2018 10:36:59 +0000 (12:36 +0200)]
selftests: forwarding: lib: Add sysctl_set(), sysctl_restore()

Add two helper functions: sysctl_set() to change the value of a given
sysctl setting, and sysctl_restore() to change it back to what it was.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/mlx5e: fix spelling mistake: "loobpack" -> "loopback"
Colin Ian King [Thu, 3 May 2018 09:12:53 +0000 (10:12 +0100)]
net/mlx5e: fix spelling mistake: "loobpack" -> "loopback"

Trivial fix to spelling mistake in netdev_err error message

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'selftests-forwarding-Two-enhancements'
David S. Miller [Thu, 3 May 2018 16:54:32 +0000 (12:54 -0400)]
Merge branch 'selftests-forwarding-Two-enhancements'

Ido Schimmel says:

====================
selftests: forwarding: Two enhancements

First patch increases the maximum deviation in the multipath tests which
proved to be too low in some cases.

Second patch allows user to run only specific tests from each file using
the TESTS environment variable. This granularity is needed in setups
where not all the tests can pass.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: forwarding: Allow running specific tests
Ido Schimmel [Thu, 3 May 2018 07:51:33 +0000 (10:51 +0300)]
selftests: forwarding: Allow running specific tests

Similar to commit a511858c7536 ("selftests: fib_tests: Allow user to run
a specific test"), allow user to run only a subset of the tests using
the TESTS environment variable.

This is useful when not all the tests can pass on a given system.

Example:
# export TESTS="ping_ipv4 ping_ipv6"
# ./bridge_vlan_aware.sh
TEST: ping [PASS]
TEST: ping6 [PASS]

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftests: forwarding: Increase maximum deviation in multipath test
Ido Schimmel [Thu, 3 May 2018 07:51:32 +0000 (10:51 +0300)]
selftests: forwarding: Increase maximum deviation in multipath test

We sometimes observe failures in the test due to too large discrepancy
between the measured and expected ratios. For example:

TEST: ECMP                                                          [FAIL]
        Too large discrepancy between expected and measured ratios
        INFO: Expected ratio 1.00 Measured ratio 1.11

Fix this by allowing an up to 15% deviation between both ratios.

Another possibility is to increase the number of generated flows, but
this will prolong the execution time of the test, which is already quite
high.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4: update latest firmware version supported
Ganesh Goudar [Thu, 3 May 2018 06:24:23 +0000 (11:54 +0530)]
cxgb4: update latest firmware version supported

Change t4fw_version.h to update latest firmware version
number to 1.19.1.0.

Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'dma-mapping-4.17-4' of git://git.infradead.org/users/hch/dma-mapping
Linus Torvalds [Thu, 3 May 2018 16:27:39 +0000 (06:27 -1000)]
Merge tag 'dma-mapping-4.17-4' of git://git.infradead.org/users/hch/dma-mapping

Pull dma-mapping fix from Christoph Hellwig:
 "Fix an incorrect warning selection introduced in the last merge
  window"

* tag 'dma-mapping-4.17-4' of git://git.infradead.org/users/hch/dma-mapping:
  swiotlb: fix inversed DMA_ATTR_NO_WARN test

6 years agotcp: restore autocorking
Eric Dumazet [Thu, 3 May 2018 03:25:13 +0000 (20:25 -0700)]
tcp: restore autocorking

When adding rb-tree for TCP retransmit queue, we inadvertently broke
TCP autocorking.

tcp_should_autocork() should really check if the rtx queue is not empty.

Tested:

Before the fix :
$ nstat -n;./netperf -H 10.246.7.152 -Cc -- -m 500;nstat | grep AutoCork
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.246.7.152 () port 0 AF_INET
Recv   Send    Send                          Utilization       Service Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local   remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % S      us/KB   us/KB

540000 262144    500    10.00      2682.85   2.47     1.59     3.618   2.329
TcpExtTCPAutoCorking            33                 0.0

// Same test, but forcing TCP_NODELAY
$ nstat -n;./netperf -H 10.246.7.152 -Cc -- -D -m 500;nstat | grep AutoCork
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.246.7.152 () port 0 AF_INET : nodelay
Recv   Send    Send                          Utilization       Service Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local   remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % S      us/KB   us/KB

540000 262144    500    10.00      1408.75   2.44     2.96     6.802   8.259
TcpExtTCPAutoCorking            1                  0.0

After the fix :
$ nstat -n;./netperf -H 10.246.7.152 -Cc -- -m 500;nstat | grep AutoCork
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.246.7.152 () port 0 AF_INET
Recv   Send    Send                          Utilization       Service Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local   remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % S      us/KB   us/KB

540000 262144    500    10.00      5472.46   2.45     1.43     1.761   1.027
TcpExtTCPAutoCorking            361293             0.0

// With TCP_NODELAY option
$ nstat -n;./netperf -H 10.246.7.152 -Cc -- -D -m 500;nstat | grep AutoCork
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.246.7.152 () port 0 AF_INET : nodelay
Recv   Send    Send                          Utilization       Service Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local   remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % S      us/KB   us/KB

540000 262144    500    10.00      5454.96   2.46     1.63     1.775   1.174
TcpExtTCPAutoCorking            315448             0.0

Fixes: 75c119afe14f ("tcp: implement rb-tree based retransmit queue")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Michael Wenig <mwenig@vmware.com>
Tested-by: Michael Wenig <mwenig@vmware.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Michael Wenig <mwenig@vmware.com>
Tested-by: Michael Wenig <mwenig@vmware.com>
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 agoip6_gre: correct the function name in ip6gre_tnl_addr_conflict() comment
Sun Lianwen [Thu, 3 May 2018 01:34:29 +0000 (09:34 +0800)]
ip6_gre: correct the function name in ip6gre_tnl_addr_conflict() comment

The function name is wrong in ip6gre_tnl_addr_conflict() comment, which
use ip6_tnl_addr_conflict instead of ip6gre_tnl_addr_conflict.

Signed-off-by: Sun Lianwen <sunlw.fnst@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agords: do not leak kernel memory to user land
Eric Dumazet [Wed, 2 May 2018 21:53:39 +0000 (14:53 -0700)]
rds: do not leak kernel memory to user land

syzbot/KMSAN reported an uninit-value in put_cmsg(), originating
from rds_cmsg_recv().

Simply clear the structure, since we have holes there, or since
rx_traces might be smaller than RDS_MSG_RX_DGRAM_TRACE_MAX.

BUG: KMSAN: uninit-value in copy_to_user include/linux/uaccess.h:184 [inline]
BUG: KMSAN: uninit-value in put_cmsg+0x600/0x870 net/core/scm.c:242
CPU: 0 PID: 4459 Comm: syz-executor582 Not tainted 4.16.0+ #87
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:17 [inline]
 dump_stack+0x185/0x1d0 lib/dump_stack.c:53
 kmsan_report+0x142/0x240 mm/kmsan/kmsan.c:1067
 kmsan_internal_check_memory+0x135/0x1e0 mm/kmsan/kmsan.c:1157
 kmsan_copy_to_user+0x69/0x160 mm/kmsan/kmsan.c:1199
 copy_to_user include/linux/uaccess.h:184 [inline]
 put_cmsg+0x600/0x870 net/core/scm.c:242
 rds_cmsg_recv net/rds/recv.c:570 [inline]
 rds_recvmsg+0x2db5/0x3170 net/rds/recv.c:657
 sock_recvmsg_nosec net/socket.c:803 [inline]
 sock_recvmsg+0x1d0/0x230 net/socket.c:810
 ___sys_recvmsg+0x3fb/0x810 net/socket.c:2205
 __sys_recvmsg net/socket.c:2250 [inline]
 SYSC_recvmsg+0x298/0x3c0 net/socket.c:2262
 SyS_recvmsg+0x54/0x80 net/socket.c:2257
 do_syscall_64+0x309/0x430 arch/x86/entry/common.c:287
 entry_SYSCALL_64_after_hwframe+0x3d/0xa2

Fixes: 3289025aedc0 ("RDS: add receive message trace used by application")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Cc: linux-rdma <linux-rdma@vger.kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoqmi_wwan: do not steal interfaces from class drivers
Bjørn Mork [Wed, 2 May 2018 20:22:54 +0000 (22:22 +0200)]
qmi_wwan: do not steal interfaces from class drivers

The USB_DEVICE_INTERFACE_NUMBER matching macro assumes that
the { vendorid, productid, interfacenumber } set uniquely
identifies one specific function.  This has proven to fail
for some configurable devices. One example is the Quectel
EM06/EP06 where the same interface number can be either
QMI or MBIM, without the device ID changing either.

Fix by requiring the vendor-specific class for interface number
based matching.  Functions of other classes can and should use
class based matching instead.

Fixes: 03304bcb5ec4 ("net: qmi_wwan: use fixed interface number matching")
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'act_csum-get_fill_size'
David S. Miller [Thu, 3 May 2018 15:15:59 +0000 (11:15 -0400)]
Merge branch 'act_csum-get_fill_size'

Craig Dillabaugh says:

====================
Update csum tc action for batch operation.

This patchset includes two patches the first updating act_csum.c
to include the get_fill_size routine required for batch operation, and
the second including updated TDC tests for the feature.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotc-testing: Updated csum action tests batch create w/wo cookies.
Craig Dillabaugh [Tue, 1 May 2018 14:17:44 +0000 (10:17 -0400)]
tc-testing: Updated csum action tests batch create w/wo cookies.

Signed-off-by: Craig Dillabaugh <cdillaba@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet sched: Implemented get_fill_size routine for act_csum.
Craig Dillabaugh [Tue, 1 May 2018 14:17:43 +0000 (10:17 -0400)]
net sched: Implemented get_fill_size routine for act_csum.

Signed-off-by: Craig Dillabaugh <cdillaba@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'trace-v4.17-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rosted...
Linus Torvalds [Thu, 3 May 2018 03:38:37 +0000 (17:38 -1000)]
Merge tag 'trace-v4.17-rc1-2' of git://git./linux/kernel/git/rostedt/linux-trace

Pull tracing fixes from Steven Rostedt:
 "Various fixes in tracing:

   - Tracepoints should not give warning on OOM failures

   - Use special field for function pointer in trace event

   - Fix igrab issues in uprobes

   - Fixes to the new histogram triggers"

* tag 'trace-v4.17-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
  tracepoint: Do not warn on ENOMEM
  tracing: Add field modifier parsing hist error for hist triggers
  tracing: Add field parsing hist error for hist triggers
  tracing: Restore proper field flag printing when displaying triggers
  tracing: initcall: Ordered comparison of function pointers
  tracing: Remove igrab() iput() call from uprobes.c
  tracing: Fix bad use of igrab in trace_uprobe.c

6 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Thu, 3 May 2018 03:34:42 +0000 (17:34 -1000)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Pull input updates from Dmitry Torokhov:
 "Just a few driver fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: atmel_mxt_ts - add missing compatible strings to OF device table
  Input: atmel_mxt_ts - fix the firmware update
  Input: atmel_mxt_ts - add touchpad button mapping for Samsung Chromebook Pro
  MAINTAINERS: Rakesh Iyer can't be reached anymore
  Input: hideep_ts - fix a typo in Kconfig
  Input: alps - fix reporting pressure of v3 trackstick
  Input: leds - fix out of bound access
  Input: synaptics-rmi4 - fix an unchecked out of memory error path

6 years agoipv4: fix fnhe usage by non-cached routes
Julian Anastasov [Wed, 2 May 2018 06:41:19 +0000 (09:41 +0300)]
ipv4: fix fnhe usage by non-cached routes

Allow some non-cached routes to use non-expired fnhe:

1. ip_del_fnhe: moved above and now called by find_exception.
The 4.5+ commit deed49df7390 expires fnhe only when caching
routes. Change that to:

1.1. use fnhe for non-cached local output routes, with the help
from (2)

1.2. allow __mkroute_input to detect expired fnhe (outdated
fnhe_gw, for example) when do_cache is false, eg. when itag!=0
for unicast destinations.

2. __mkroute_output: keep fi to allow local routes with orig_oif != 0
to use fnhe info even when the new route will not be cached into fnhe.
After commit 839da4d98960 ("net: ipv4: set orig_oif based on fib
result for local traffic") it means all local routes will be affected
because they are not cached. This change is used to solve a PMTU
problem with IPVS (and probably Netfilter DNAT) setups that redirect
local clients from target local IP (local route to Virtual IP)
to new remote IP target, eg. IPVS TUN real server. Loopback has
64K MTU and we need to create fnhe on the local route that will
keep the reduced PMTU for the Virtual IP. Without this change
fnhe_pmtu is updated from ICMP but never exposed to non-cached
local routes. This includes routes with flowi4_oif!=0 for 4.6+ and
with flowi4_oif=any for 4.14+).

3. update_or_create_fnhe: make sure fnhe_expires is not 0 for
new entries

Fixes: 839da4d98960 ("net: ipv4: set orig_oif based on fib result for local traffic")
Fixes: d6d5e999e5df ("route: do not cache fib route info on local routes with oif")
Fixes: deed49df7390 ("route: check and remove route cache when we get route")
Cc: David Ahern <dsahern@gmail.com>
Cc: Xin Long <lucien.xin@gmail.com>
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Thu, 3 May 2018 02:38:17 +0000 (16:38 -1000)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Three small bug fixes: an illegally overlapping memcmp in target code,
  a potential infinite loop in isci under certain rare phy conditions
  and an ATA queue depth (performance) correction for storvsc"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: target: Fix fortify_panic kernel exception
  scsi: isci: Fix infinite loop in while loop
  scsi: storvsc: Set up correct queue depth values for IDE devices

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

Daniel Borkmann says:

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

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

The main changes are:

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

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

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

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

John Fastabend says:

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

To reproduce:

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

Fixes: edd7ceb78296 ("ipv6: Allow non-gateway ECMP for IPv6")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Tested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'r8169-series-with-further-improvements'
David S. Miller [Wed, 2 May 2018 20:23:50 +0000 (16:23 -0400)]
Merge branch 'r8169-series-with-further-improvements'

Heiner Kallweit says:

====================
r8169: series with further improvements

I thought I'm more or less done with the basic refactoring. But again
I stumbled across things that can be improved / simplified.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: replace get_protocol with vlan_get_protocol
Heiner Kallweit [Wed, 2 May 2018 19:40:02 +0000 (21:40 +0200)]
r8169: replace get_protocol with vlan_get_protocol

This patch is basically the same as 6e74d1749a33 ("r8152: replace
get_protocol with vlan_get_protocol"). Use vlan_get_protocol
instead of duplicating the functionality.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: avoid potentially misaligned access when getting mac address
Heiner Kallweit [Wed, 2 May 2018 19:39:59 +0000 (21:39 +0200)]
r8169: avoid potentially misaligned access when getting mac address

Interpreting a member of an u16 array as u32 may result in a misaligned
access. Also it's not really intuitive to define a mac address variable
as array of three u16 words. Therefore use an array of six bytes that
is properly aligned for 32 bit access.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: improve PCI config space access
Heiner Kallweit [Wed, 2 May 2018 19:39:56 +0000 (21:39 +0200)]
r8169: improve PCI config space access

Some chips have a non-zero function id, however instead of hardcoding
the id's (CSIAR_FUNC_NIC and CSIAR_FUNC_NIC2) we can get them
dynamically via PCI_FUNC(pci_dev->devfn). This way we can get rid
of the csi_ops.

In general csi is just a fallback mechanism for PCI config space
access in case no native access is supported. Therefore let's
try native access first.

I checked with Realtek regarding the functionality of config space
byte 0x070f and according to them it controls the L0s/L1
entrance latency.
Currently ASPM is disabled in general and therefore this value
isn't used. However we may introduce a whitelist for chips
where ASPM is known to work, therefore let's keep this code.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: drop rtl_generic_op
Heiner Kallweit [Wed, 2 May 2018 19:39:54 +0000 (21:39 +0200)]
r8169: drop rtl_generic_op

Only two places are left where rtl_generic_op() is used, so we can
inline it and simplify the code a little.
This change also avoids the overhead of unlocking/locking in case
the respective operation isn't set.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: replace longer if statements with switch statements
Heiner Kallweit [Wed, 2 May 2018 19:39:52 +0000 (21:39 +0200)]
r8169: replace longer if statements with switch statements

Some longer if statements can be simplified by using switch
statements instead.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: simplify code by using ranges in switch clauses
Heiner Kallweit [Wed, 2 May 2018 19:39:49 +0000 (21:39 +0200)]
r8169: simplify code by using ranges in switch clauses

Several switch statements can be significantly simplified by using
case ranges.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: drop member pll_power_ops from struct rtl8169_private
Heiner Kallweit [Wed, 2 May 2018 19:39:47 +0000 (21:39 +0200)]
r8169: drop member pll_power_ops from struct rtl8169_private

After merging r810x_pll_power_down/up and r8168_pll_power_down/up we
don't need member pll_power_ops any longer and can drop it, thus
simplifying the code.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: merge r810x_pll_power_down/up into r8168_pll_power_down/up
Heiner Kallweit [Wed, 2 May 2018 19:39:45 +0000 (21:39 +0200)]
r8169: merge r810x_pll_power_down/up into r8168_pll_power_down/up

r810x_pll_power_down/up and r8168_pll_power_down/up have a lot in common,
so we can simplify the code by merging the former into the latter.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: remove 810x_phy_power_up/down
Heiner Kallweit [Wed, 2 May 2018 19:39:40 +0000 (21:39 +0200)]
r8169: remove 810x_phy_power_up/down

The functionality of 810x_phy_power_up/down is covered by the default
clause in 8168_phy_power_up/down. Therefore we don't need these
functions.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: remove unneeded check in r8168_pll_power_down
Heiner Kallweit [Wed, 2 May 2018 19:39:35 +0000 (21:39 +0200)]
r8169: remove unneeded check in r8168_pll_power_down

RTL_GIGA_MAC_VER_23/24 are configured by rtl_hw_start_8168cp_2()
and rtl_hw_start_8168cp_3() respectively which both apply
CPCMD_QUIRK_MASK, thus clearing bit ASF.

Bit ASF isn't set at any other place in the driver, therefore this
check can be removed.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoparisc: Fix section mismatches
Helge Deller [Fri, 20 Apr 2018 21:23:37 +0000 (23:23 +0200)]
parisc: Fix section mismatches

Fix three section mismatches:
1) Section mismatch in reference from the function ioread8() to the
   function .init.text:pcibios_init_bridge()
2) Section mismatch in reference from the function free_initmem() to the
   function .init.text:map_pages()
3) Section mismatch in reference from the function ccio_ioc_init() to
   the function .init.text:count_parisc_driver()

Signed-off-by: Helge Deller <deller@gmx.de>
6 years agoparisc: drivers.c: Fix section mismatches
Helge Deller [Fri, 20 Apr 2018 21:19:17 +0000 (23:19 +0200)]
parisc: drivers.c: Fix section mismatches

Fix two section mismatches in drivers.c:
1) Section mismatch in reference from the function alloc_tree_node() to
   the function .init.text:create_tree_node().
2) Section mismatch in reference from the function walk_native_bus() to
   the function .init.text:alloc_pa_dev().

Signed-off-by: Helge Deller <deller@gmx.de>
6 years agoMerge branch 'x86-bpf-jit-fixes'
Alexei Starovoitov [Wed, 2 May 2018 19:35:47 +0000 (12:35 -0700)]
Merge branch 'x86-bpf-jit-fixes'

Daniel Borkmann says:

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

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

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

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

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

Fixes: e0ee9c12157d ("x86: bpf_jit: fix two bugs in eBPF JIT compiler")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
6 years agoMerge branch 'net-smc-small-features'
David S. Miller [Wed, 2 May 2018 17:29:13 +0000 (13:29 -0400)]
Merge branch 'net-smc-small-features'

Ursula Braun says:

====================
net/smc: small features 2018/04/30

here are 4 smc patches for net-next covering small new features
in different areas:
   * link health check
   * diagnostics for IPv6 smc sockets
   * ioctl
   * improvement for vlan determination

v2 changes:
   * better title
   * patch 2 - remove compile problem for disabled CONFIG_IPV6
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/smc: determine vlan_id of stacked net_device
Ursula Braun [Wed, 2 May 2018 14:56:47 +0000 (16:56 +0200)]
net/smc: determine vlan_id of stacked net_device

An SMC link group is bound to a specific vlan_id. Its link uses
the RoCE-GIDs established for the specific vlan_id. This patch makes
sure the appropriate vlan_id is determined for stacked scenarios like
for instance a master bonding device with vlan devices enslaved.

Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/smc: handle ioctls SIOCINQ, SIOCOUTQ, and SIOCOUTQNSD
Ursula Braun [Wed, 2 May 2018 14:56:46 +0000 (16:56 +0200)]
net/smc: handle ioctls SIOCINQ, SIOCOUTQ, and SIOCOUTQNSD

SIOCINQ returns the amount of unread data in the RMB.
SIOCOUTQ returns the amount of unsent or unacked sent data in the send
buffer.
SIOCOUTQNSD returns the amount of data prepared for sending, but
not yet sent.

Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/smc: ipv6 support for smc_diag.c
Karsten Graul [Wed, 2 May 2018 14:56:45 +0000 (16:56 +0200)]
net/smc: ipv6 support for smc_diag.c

Update smc_diag.c to support ipv6 addresses on the diagnosis interface.

Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/smc: periodic testlink support
Karsten Graul [Wed, 2 May 2018 14:56:44 +0000 (16:56 +0200)]
net/smc: periodic testlink support

Add periodic LLC testlink support to ensure the link is still active.
The interval time is initialized using the value of
sysctl_tcp_keepalive_time.

Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/smc: restrict non-blocking connect finish
Ursula Braun [Wed, 2 May 2018 14:53:56 +0000 (16:53 +0200)]
net/smc: restrict non-blocking connect finish

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

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

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

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

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

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'mlxsw-Reject-unsupported-FIB-configurations'
David S. Miller [Wed, 2 May 2018 17:15:18 +0000 (13:15 -0400)]
Merge branch 'mlxsw-Reject-unsupported-FIB-configurations'

Ido Schimmel says:

====================
mlxsw: Reject unsupported FIB configurations

Recently it became possible for listeners of the FIB notification chain
to veto operations such as addition of routes and rules.

Adjust the mlxsw driver to take advantage of it and return an error for
unsupported FIB rules and for routes configured after the abort
mechanism was triggered (due to exceeded resources for example).

v2:
* Change error code in first patch to -EOPNOTSUPP (David Ahern).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum_router: Return an error for routes added after abort
Ido Schimmel [Wed, 2 May 2018 07:17:35 +0000 (10:17 +0300)]
mlxsw: spectrum_router: Return an error for routes added after abort

We currently do not perform accounting in the driver and thus can't
reject routes before resources are exceeded.

However, in order to make users aware of the fact that routes are no
longer offloaded we can return an error for routes configured after the
abort mechanism was triggered.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum_router: Return an error for non-default FIB rules
Ido Schimmel [Wed, 2 May 2018 07:17:34 +0000 (10:17 +0300)]
mlxsw: spectrum_router: Return an error for non-default FIB rules

Since commit 9776d32537d2 ("net: Move call_fib_rule_notifiers up in
fib_nl_newrule") it is possible to forbid the installation of
unsupported FIB rules.

Have mlxsw return an error for non-default FIB rules in addition to the
existing extack message.

Example:
# ip rule add from 198.51.100.1 table 10
Error: mlxsw_spectrum: FIB rules not supported.

Note that offload is only aborted when non-default FIB rules are already
installed and merely replayed during module initialization.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4: add new T5 device id's
Ganesh Goudar [Wed, 2 May 2018 06:17:15 +0000 (11:47 +0530)]
cxgb4: add new T5 device id's

Add device id's 0x5019, 0x501a and 0x501b for T5
cards.

Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: fix the issue that the cookie-ack with auth can't get processed
Xin Long [Wed, 2 May 2018 05:45:12 +0000 (13:45 +0800)]
sctp: fix the issue that the cookie-ack with auth can't get processed

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

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

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

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

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

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

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

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

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

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

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

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

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

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

Fixes: 0f8782ea1497 ("tcp_bbr: add BBR congestion control")
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Yousuk Seung <ysseung@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: stmmac: Avoid VLA usage
Kees Cook [Tue, 1 May 2018 21:01:30 +0000 (14:01 -0700)]
net: stmmac: Avoid VLA usage

In the quest to remove all stack VLAs from the kernel[1], this switches
the "status" stack buffer to use the existing small (8) upper bound on
how many queues can be checked for DMA, and adds a sanity-check just to
make sure it doesn't operate under pathological conditions.

[1] http://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com

Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethernet: ti: cpsw: fix packet leaking in dual_mac mode
Grygorii Strashko [Tue, 1 May 2018 17:41:22 +0000 (12:41 -0500)]
net: ethernet: ti: cpsw: fix packet leaking in dual_mac mode

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

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

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

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoliquidio VF: indicate that disabling rx vlan offload is not allowed
Raghu Vatsavayi [Tue, 1 May 2018 17:32:10 +0000 (10:32 -0700)]
liquidio VF: indicate that disabling rx vlan offload is not allowed

NIC firmware does not support disabling rx vlan offload, but the VF driver
incorrectly indicates that it is supported.  The PF driver already does the
correct indication by clearing the NETIF_F_HW_VLAN_CTAG_RX bit in its
netdev->hw_features.  So just do the same thing in the VF.

Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@cavium.com>
Acked-by: Prasad Kanneganti <prasad.kanneganti@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoudp: Complement partial checksum for GSO packet
Sean Tranchetti [Tue, 1 May 2018 00:01:02 +0000 (18:01 -0600)]
udp: Complement partial checksum for GSO packet

Using the udp_v4_check() function to calculate the pseudo header
for the newly segmented UDP packets results in assigning the complement
of the value to the UDP header checksum field.

Always undo the complement the partial checksum value in order to
match the case where GSO is not used on the UDP transmit path.

Fixes: ee80d1ebe5ba ("udp: add udp gso")
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoRevert "vhost: make msg padding explicit"
Michael S. Tsirkin [Wed, 2 May 2018 14:19:05 +0000 (17:19 +0300)]
Revert "vhost: make msg padding explicit"

This reverts commit 93c0d549c4c5a7382ad70de6b86610b7aae57406.

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

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoswiotlb: fix inversed DMA_ATTR_NO_WARN test
Michel Dänzer [Tue, 1 May 2018 13:24:11 +0000 (15:24 +0200)]
swiotlb: fix inversed DMA_ATTR_NO_WARN test

The result was printing the warning only when we were explicitly asked
not to.

Cc: stable@vger.kernel.org
Fixes: 0176adb004065d6815a8e67946752df4cd947c5b "swiotlb: refactor
 coherent buffer allocation"
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>.
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rkuo/linux...
Linus Torvalds [Wed, 2 May 2018 02:54:22 +0000 (19:54 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/rkuo/linux-hexagon-kernel

Pull hexagon fixes from Richard Kuo:
 "Some small fixes for module compilation"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rkuo/linux-hexagon-kernel:
  hexagon: export csum_partial_copy_nocheck
  hexagon: add memset_io() helper

6 years agonfp: flower: set tunnel ttl value to net default
John Hurley [Tue, 1 May 2018 22:49:49 +0000 (15:49 -0700)]
nfp: flower: set tunnel ttl value to net default

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

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

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

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

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

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

Reported-by: Andre Tomt <andre@tomt.net>
Signed-off-by: Dave Watson <davejwatson@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoselftest: add test for TCP_INQ
Soheil Hassas Yeganeh [Tue, 1 May 2018 19:39:16 +0000 (15:39 -0400)]
selftest: add test for TCP_INQ

Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: send in-queue bytes in cmsg upon read
Soheil Hassas Yeganeh [Tue, 1 May 2018 19:39:15 +0000 (15:39 -0400)]
tcp: send in-queue bytes in cmsg upon read

Applications with many concurrent connections, high variance
in receive queue length and tight memory bounds cannot
allocate worst-case buffer size to drain sockets. Knowing
the size of receive queue length, applications can optimize
how they allocate buffers to read from the socket.

The number of bytes pending on the socket is directly
available through ioctl(FIONREAD/SIOCINQ) and can be
approximated using getsockopt(MEMINFO) (rmem_alloc includes
skb overheads in addition to application data). But, both of
these options add an extra syscall per recvmsg. Moreover,
ioctl(FIONREAD/SIOCINQ) takes the socket lock.

Add the TCP_INQ socket option to TCP. When this socket
option is set, recvmsg() relays the number of bytes available
on the socket for reading to the application via the
TCP_CM_INQ control message.

Calculate the number of bytes after releasing the socket lock
to include the processed backlog, if any. To avoid an extra
branch in the hot path of recvmsg() for this new control
message, move all cmsg processing inside an existing branch for
processing receive timestamps. Since the socket lock is not held
when calculating the size of receive queue, TCP_INQ is a hint.
For example, it can overestimate the queue size by one byte,
if FIN is received.

With this method, applications can start reading from the socket
using a small buffer, and then use larger buffers based on the
remaining data when needed.

V3 change-log:
As suggested by David Miller, added loads with barrier
to check whether we have multiple threads calling recvmsg
in parallel. When that happens we lock the socket to
calculate inq.
V4 change-log:
Removed inline from a static function.

Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Suggested-by: David Miller <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoInput: atmel_mxt_ts - add missing compatible strings to OF device table
Javier Martinez Canillas [Tue, 1 May 2018 18:44:53 +0000 (11:44 -0700)]
Input: atmel_mxt_ts - add missing compatible strings to OF device table

Commit af503716ac14 ("i2c: core: report OF style module alias for devices
registered via OF") fixed how the I2C core reports the module alias when
devices are registered via OF.

But the atmel_mxt_ts driver only has an "atmel,maxtouch" compatible in its
OF device ID table, so if a Device Tree is using a different one, autoload
won't be working for the module (the matching works because the I2C device
ID table is used as a fallback).

So add compatible strings for each of the entries in the I2C device table.

Fixes: af503716ac14 ("i2c: core: report OF style module alias for devices registered via OF")
Reported-by: Enric Balletbo i Serra <enric.balletbo@collabora.com>
Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Tested-by: Enric Balletbo i Serra <enric.balletbo@collabora.com>
Reviewed-by: Rob Herring <robh@kernel.org>
[dtor: document which compatibles are deprecated and should not be used]
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
6 years agobpf: minor fix to selftest test_stacktrace_build_id()
Song Liu [Tue, 1 May 2018 17:20:24 +0000 (10:20 -0700)]
bpf: minor fix to selftest test_stacktrace_build_id()

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

Fixes: 81f77fd0deeb ("bpf: add selftest for stackmap with BPF_F_STACK_BUILD_ID")
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>