platform/kernel/linux-rpi.git
7 years agonetlink: Allow ext_ack to carry non-error messages
David Ahern [Sat, 28 Oct 2017 00:37:12 +0000 (17:37 -0700)]
netlink: Allow ext_ack to carry non-error messages

The NLMSGERR API already carries data (eg, a cookie) on the success path.
Allow a message string to be returned as well.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'dsa-port-parsing'
David S. Miller [Wed, 1 Nov 2017 02:47:45 +0000 (11:47 +0900)]
Merge branch 'dsa-port-parsing'

Vivien Didelot says:

====================
net: dsa: add port parsing functions

This patchset adds port parsing functions called early in the new
bindings parsing stage, which regroup all the fetching of static data
available at the port level, including the port's type, name and CPU
master interface.

This simplifies the rest of the code which does not need to dig into
device tree or platform data again in order to check a port's type or
name.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: remove name arg from slave create
Vivien Didelot [Fri, 27 Oct 2017 19:55:19 +0000 (15:55 -0400)]
net: dsa: remove name arg from slave create

Now that slave dsa_port always have their name set, there is no need to
pass it to dsa_slave_create() anymore. Remove this argument.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: get port name at parse time
Vivien Didelot [Fri, 27 Oct 2017 19:55:18 +0000 (15:55 -0400)]
net: dsa: get port name at parse time

Get the optional "label" property and assign a default one directly at
parse time instead of doing it when creating the slave.

For legacy, simply assign the port name stored in cd->port_names.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: get master device at port parsing time
Vivien Didelot [Fri, 27 Oct 2017 19:55:17 +0000 (15:55 -0400)]
net: dsa: get master device at port parsing time

Fetching the master device can be done directly when a port is parsed
from device tree or pdata, instead of waiting until dsa_dst_parse.

Now that -EPROBE_DEFER is returned before we add the switch to the tree,
there is no need to check for this error after dsa_dst_parse.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: get port type at parse time
Vivien Didelot [Fri, 27 Oct 2017 19:55:15 +0000 (15:55 -0400)]
net: dsa: get port type at parse time

Assign a port's type at parsed time instead of waiting for the tree to
be completed.

Because this is now done earlier, we can use the port's type in
dsa_port_is_* helpers instead of digging again in topology description.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: add port parse functions
Vivien Didelot [Fri, 27 Oct 2017 19:55:14 +0000 (15:55 -0400)]
net: dsa: add port parse functions

Add symmetrical DSA port parsing functions for pdata and device tree,
used to parse and validate a given port node or platform data.

They don't do much for the moment but will be extended later on to
assign a port type and get device references.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: get ports within parsing code
Vivien Didelot [Fri, 27 Oct 2017 19:55:13 +0000 (15:55 -0400)]
net: dsa: get ports within parsing code

There is no point into hiding the -EINVAL error code in ERR_PTR from a
dsa_get_ports function, simply get the "ports" node directly from within
the dsa_parse_ports_dn function.

This also has the effect to make the pdata and device tree handling code
symmetrical inside _dsa_register_switch.

At the same time, rename dsa_parse_ports_dn to dsa_parse_ports_of
because _of is a more common suffix for device tree parsing functions.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: reduce verifier memory consumption
Alexei Starovoitov [Wed, 1 Nov 2017 01:16:05 +0000 (18:16 -0700)]
bpf: reduce verifier memory consumption

the verifier got progressively smarter over time and size of its internal
state grew as well. Time to reduce the memory consumption.

Before:
sizeof(struct bpf_verifier_state) = 6520
After:
sizeof(struct bpf_verifier_state) = 896

It's done by observing that majority of BPF programs use little to
no stack whereas verifier kept all of 512 stack slots ready always.
Instead dynamically reallocate struct verifier state when stack
access is detected.
Runtime difference before vs after is within a noise.
The number of processed instructions stays the same.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'liquidio-switchdev-support'
David S. Miller [Wed, 1 Nov 2017 02:39:52 +0000 (11:39 +0900)]
Merge branch 'liquidio-switchdev-support'

Vijaya Mohan Guvva says:

====================
liquidio: switchdev support for LiquidIO NIC

patch1 of this patch set adds switchdev support for SRIOV capable
LiquidIO NIC, so that for every SRIOV VF on LiquidIO, a representor
netdev is created on hypervisor. It also has changes to send representor
interface configurations like admin state and MTU to LiquidIO firmware and
to retrieve HW counted VF stats for VF representor.

patch2 adds support for switchdev enable/disable from devlink

Patchset Change Log:
  V2 -> V3:
    * Use mac address as the physical switchID.
    * Check for eswitch_mode before returning switchID

  V1 -> V2:
    * Name the representors "pfXvfY".
    * Drop patch3 (ethtool support for switchdev ports) that was in V1
      because it's not necessary.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio: Configure switchdev with devlink
Vijaya Mohan Guvva [Tue, 31 Oct 2017 23:04:57 +0000 (16:04 -0700)]
liquidio: Configure switchdev with devlink

Enable and disable switchdev on SRIOV capable LiquidIO NIC with devlink.
Create representor netdev for each SRIOV VF function on SRIOV enable and
and do the cleanup on SRIOV disable.

Signed-off-by: Vijaya Mohan Guvva <vijaya.guvva@cavium.com>
Signed-off-by: Satanand Burla <satananda.burla@cavium.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio: switchdev support for LiquidIO NIC
Vijaya Mohan Guvva [Tue, 31 Oct 2017 23:04:53 +0000 (16:04 -0700)]
liquidio: switchdev support for LiquidIO NIC

Enable switchdev for SRIOV capable LiquidIO NIC. It registers
a representor netdev (with switchdev_ops) for each SRIOV VF created.
It also has changes to send representor interface configurations like
admin state and MTU to LiquidIO firmware and to retrieve HW counted
VF stats for VF representor.

Signed-off-by: Vijaya Mohan Guvva <vijaya.guvva@cavium.com>
Signed-off-by: Satanand Burla <satananda.burla@cavium.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'mlx5-updates-2017-10-31' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Wed, 1 Nov 2017 02:37:17 +0000 (11:37 +0900)]
Merge tag 'mlx5-updates-2017-10-31' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2017-10-31 mlx5e stats groups

This series from Kamal introduces an important refactoring for mlx5e stats
handling, which groups the stats into generic groups structure which allows
to control the behavior and stats reporting per group in a modular way.

In the first patch Kamal introduces a new data type "mlx5e_stats_grp" This change
defines a new API to create a group of stats and simplifies the way of handling them.
This struct will define the following behavior per group:
    - get_num_stats() - return the number of counters in the group.
    - fill_strings() - fill counters strings within the group.
    - fill_stats() - fill counters values within the group.

All other patches will be straight forward refactoring per stats group,
where Kamal will move each mlx5e stats group to use the new API.

The idea is to have better flexibility and modularity to add new counters,
all ethtool logic was rendered generic and loops through the generic stats groups and
calls the groups callbacks to figure out how and what to report back to user space.

Introducing new file en_stats.c to hold all the new stat groups logic and implementation.
Static structures (counters descriptors) moved from en_stats.h to en_stats.c which reduces
the mlx5_core binary footprint, originally reported and addressed by Stephen Hemminger:
("mlx5: fix space waste from ethtool descriptions") which was waived due to this re-design.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Wed, 1 Nov 2017 02:08:18 +0000 (11:08 +0900)]
Merge branch '40GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2017-10-31

This series contains updates to i40e, i40evf and net/sched.

Arnd Bergmann cleans up the power management code to resolve a build
warning.

Shannon Nelson fixes i40e to only redistribute our vectors when we did
not get the full count that we requested.

Alex reverts a previous commit because it potentially causes a memory leak
when combined with the current page recycling scheme.

Amritha enables configuring cloud filters in i40e using the tc-flower
classifier.  The classification function of the filter is to match a
packet to a traffic class.  cls_flower is extended to offload classid to
hardware.  Hardware traffic classes are identified using classid values
reserved in the range :ffe0 - :ffef.
The cloud filters are added for a VSI and are cleaned up when the VSI is
deleted. The filters that match on L4 ports needs enhanced admin queue
functions with big buffer support for extended fields in cloud filter
commands.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotc-testing: better test case file error reporting
Brenda J. Butler [Tue, 31 Oct 2017 18:29:03 +0000 (14:29 -0400)]
tc-testing: better test case file error reporting

tdc.py reads a bunch of test cases in json files.  When a json file
cannot be parsed, tdc just exits and does not run any tests.

This patch will cause tdc to print a message with the file name and
line number, then that file will be ignored and the rest of the tests
will be processed.

Signed-off-by: Brenda J. Butler <bjb@mojatatu.com>
Acked-by: Lucas Bates <lucasb@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotc-testing: better check if thing is list
Brenda J. Butler [Tue, 31 Oct 2017 18:28:35 +0000 (14:28 -0400)]
tc-testing: better check if thing is list

Check if tcase[k] is an instance of a list (is or is derived from list)
instead of checking if it is a list.

This will be useful if the data structures change to be something
that implements list, instead of being an actual list.  In that
case, this code will not have to change.

Signed-off-by: Brenda J. Butler <bjb@mojatatu.com>
Acked-by: Lucas Bates <lucasb@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotc-testing: correction to docstring in get_unique_item
Brenda J. Butler [Tue, 31 Oct 2017 18:28:02 +0000 (14:28 -0400)]
tc-testing: correction to docstring in get_unique_item

Signed-off-by: Brenda J. Butler <bjb@mojatatu.com>
Acked-by: Lucas Bates <lucasb@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotc-testing: split config file
Brenda J. Butler [Tue, 31 Oct 2017 18:27:28 +0000 (14:27 -0400)]
tc-testing: split config file

Move the config customization into a site-local file
tdc_config_local.py, so that updates of the tdc test
software does not require hand-editing of the config.
This patch includes a template for the site-local
customization file.

In addition, this makes it easy to revert to a stock
tdc environment for testing the test framework and/or
the core tests.

Also it makes it harder for any custom config to be
submitted back to the kernel tdc.

Signed-off-by: Brenda J. Butler <bjb@mojatatu.com>
Acked-by: Lucas Bates <lucasb@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotc-testing: gitignore, ignore standard python artifacts
Brenda J. Butler [Tue, 31 Oct 2017 18:26:54 +0000 (14:26 -0400)]
tc-testing: gitignore, ignore standard python artifacts

Ignore .pyc files, "python compiled" files, that get created
when a python script is run.  They should never be committed.

Signed-off-by: Brenda J. Butler <bjb@mojatatu.com>
Acked-by: Lucas Bates <lucasb@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotc-testing: very simple example test cases
Brenda J. Butler [Tue, 31 Oct 2017 18:25:46 +0000 (14:25 -0400)]
tc-testing: very simple example test cases

As part of documentation, supply some very simple test cases
to illustrate how test cases work.  One test case shows
commands in the setup, command, verify and teardown stages.
Other test cases show how to have a working test case that
does not have commands in the setup, verify and/or teardown
stages.

Specifically, the command lists for setup and teardown can
be empty.  And the verify command must have a command, but
it can be /bin/true.  The regex must have a string, we
recommend a single space, and the count of matches must be
zero if you do not want to use the match feature of verify.
Verify will always look for a return code of success (0)
so we give /bin/true when we do not want to make a check
there.

Also, update the documentation for testcases to be more
specific in the cases of:

    - accepting non-success return codes in setup and
      teardown stages
    - how to write the test when no setup, teardown
      and/or verify are desired.

To run the example test cases:

    $ sudo -E ./tdc.py -f creating-testcases/example.json -l
    1f: (example) simple test to test framework
    2f: (example) simple test, no need for verify
    3f: (example) simple test, no need for setup or teardown (or verify)
    $ sudo -E ./tdc.py -f creating-testcases/example.json
    Test 1f: simple test to test framework
    Test 2f: simple test, no need for verify
    Test 3f: simple test, no need for setup or teardown (or verify)
    All test results:

    1..3
    ok 1 1f simple test to test framework
    ok 2 2f simple test, no need for verify
    ok 3 3f simple test, no need for setup or teardown (or verify)

    $

Signed-off-by: Brenda J. Butler <bjb@mojatatu.com>
Acked-by: Lucas Bates <lucasb@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'l2tp-remove-unused-code'
David S. Miller [Wed, 1 Nov 2017 01:57:24 +0000 (10:57 +0900)]
Merge branch 'l2tp-remove-unused-code'

Guillaume Nault says:

====================
l2tp: remove unused code

Patch #1 removes the ref/deref mechanism that was originally used to
prevent ppp pseudowires from dropping their sockets. This mechanism
was error prone and isn't used anymore.

Patch #2 removes some module specific refcount debugging.
Patches #3 and #4 take care of some dead code.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agol2tp: remove field 'dev' from struct l2tp_eth
Guillaume Nault [Tue, 31 Oct 2017 16:36:46 +0000 (17:36 +0100)]
l2tp: remove field 'dev' from struct l2tp_eth

This field has never been used.

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agol2tp: remove l2tp_tunnel_count and l2tp_session_count
Guillaume Nault [Tue, 31 Oct 2017 16:36:45 +0000 (17:36 +0100)]
l2tp: remove l2tp_tunnel_count and l2tp_session_count

These variables have never been used.

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agol2tp: remove l2tp specific refcount debugging
Guillaume Nault [Tue, 31 Oct 2017 16:36:44 +0000 (17:36 +0100)]
l2tp: remove l2tp specific refcount debugging

With conversion to refcount_t, such manual debugging code doesn't make
sense anymore.
The tunnel part was already dropped by
54652eb12c1b ("l2tp: hold tunnel while looking up sessions in l2tp_netlink").

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agol2tp: remove ->ref() and ->deref()
Guillaume Nault [Tue, 31 Oct 2017 16:36:42 +0000 (17:36 +0100)]
l2tp: remove ->ref() and ->deref()

The ->ref() and ->deref() callbacks are unused since PPP stopped using
them in ee40fb2e1eb5 ("l2tp: protect sock pointer of struct pppol2tp_session with RCU").

We can thus remove them from struct l2tp_session and drop the do_ref
parameter of l2tp_session_get*().

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/mlx5e: Switch channels counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 12:08:19 +0000 (15:08 +0300)]
net/mlx5e: Switch channels counters to use stats group API

Switch the channels counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch ipsec counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 12:07:20 +0000 (15:07 +0300)]
net/mlx5e: Switch ipsec counters to use stats group API

Switch the ipsec counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch pme counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 12:05:58 +0000 (15:05 +0300)]
net/mlx5e: Switch pme counters to use stats group API

Switch the pme counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch per prio pfc counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 12:04:47 +0000 (15:04 +0300)]
net/mlx5e: Switch per prio pfc counters to use stats group API

Switch the per prio pfc counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch per prio traffic counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 12:03:03 +0000 (15:03 +0300)]
net/mlx5e: Switch per prio traffic counters to use stats group API

Switch the per prio traffic counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch pcie counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 12:01:58 +0000 (15:01 +0300)]
net/mlx5e: Switch pcie counters to use stats group API

Switch the pcie counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch ethernet extended counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 12:00:40 +0000 (15:00 +0300)]
net/mlx5e: Switch ethernet extended counters to use stats group API

Switch the ethernet extended counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch physical statistical counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 11:58:47 +0000 (14:58 +0300)]
net/mlx5e: Switch physical statistical counters to use stats group API

Switch the physical statistical counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch RFC 2819 counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 11:56:48 +0000 (14:56 +0300)]
net/mlx5e: Switch RFC 2819 counters to use stats group API

Switch the RFC 2819 counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch RFC 2863 counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 11:51:27 +0000 (14:51 +0300)]
net/mlx5e: Switch RFC 2863 counters to use stats group API

Switch the RFC 2863 counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch IEEE 802.3 counters to use stats group API
Kamal Heib [Wed, 23 Aug 2017 11:49:57 +0000 (14:49 +0300)]
net/mlx5e: Switch IEEE 802.3 counters to use stats group API

Switch the IEEE 802.3 counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch vport counters to use the stats group API
Kamal Heib [Wed, 23 Aug 2017 11:47:58 +0000 (14:47 +0300)]
net/mlx5e: Switch vport counters to use the stats group API

Switch the vport counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Switch Q counters to use the stats group API
Kamal Heib [Wed, 23 Aug 2017 11:46:33 +0000 (14:46 +0300)]
net/mlx5e: Switch Q counters to use the stats group API

Switch the Q counters to use the new stats group API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agonet/mlx5e: Introduce stats group API
Kamal Heib [Wed, 23 Aug 2017 11:42:03 +0000 (14:42 +0300)]
net/mlx5e: Introduce stats group API

Currently the mlx5e driver has multiple groups of stats, each group is
used for different purposes and it may depend on hardware capabilities
or not. The problem with the current implementation is that there is no
clear API to create a new group of stats.

This change define a new API to create a group of stats and simplifies
the way of handling them by defining a new struct "mlx5e_stats_grp" which
have the following three function pointers:
- get_num_stats() - return the number of counters in the group.
- fill_strings() - fill counters strings within the group.
- fill_stats() - fill counters values within the group.

The above function pointers are used within the ethtool callbaks while
calling "ethtool -S" from userspace. This change also switch the SW
group to use the new API.

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
7 years agoi40e: Enable cloud filters via tc-flower
Amritha Nambiar [Fri, 27 Oct 2017 09:36:01 +0000 (02:36 -0700)]
i40e: Enable cloud filters via tc-flower

This patch enables tc-flower based hardware offloads. tc flower
filter provided by the kernel is configured as driver specific
cloud filter. The patch implements functions and admin queue
commands needed to support cloud filters in the driver and
adds cloud filters to configure these tc-flower filters.

The classification function of the filter is to direct matched
packets to a traffic class. The hardware traffic class is set
based on the the classid reserved in the range :ffe0 - :ffef.

Match Dst MAC and route to TC0:
  prio 1 flower dst_mac 3c:fd:fe:a0:d6:70 skip_sw\
  hw_tc 1

Match Dst IPv4,Dst Port and route to TC1:
  prio 2 flower dst_ip 192.168.3.5/32\
  ip_proto udp dst_port 25 skip_sw\
  hw_tc 2

Match Dst IPv6,Dst Port and route to TC1:
  prio 3 flower dst_ip fe8::200:1\
  ip_proto udp dst_port 66 skip_sw\
  hw_tc 2

Delete tc flower filter:
Example:

Flow Director Sideband is disabled while configuring cloud filters
via tc-flower and until any cloud filter exists.

Unsupported matches when cloud filters are added using enhanced
big buffer cloud filter mode of underlying switch include:
1. source port and source IP
2. Combined MAC address and IP fields.
3. Not specifying L4 port

These filter matches can however be used to redirect traffic to
the main VSI (tc 0) which does not require the enhanced big buffer
cloud filter support.

Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Signed-off-by: Kiran Patil <kiran.patil@intel.com>
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Shannon Nelson <shannon.nelson@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Clean up of cloud filters
Amritha Nambiar [Fri, 27 Oct 2017 09:35:56 +0000 (02:35 -0700)]
i40e: Clean up of cloud filters

Introduce the cloud filter data structure and cleanup of cloud
filters associated with the device.

Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Acked-by: Shannon Nelson <shannon.nelson@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Admin queue definitions for cloud filters
Amritha Nambiar [Fri, 27 Oct 2017 09:35:51 +0000 (02:35 -0700)]
i40e: Admin queue definitions for cloud filters

Add new admin queue definitions and extended fields for cloud
filter support. Define big buffer for extended general fields
in Add/Remove Cloud filters command.

Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Signed-off-by: Kiran Patil <kiran.patil@intel.com>
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Shannon Nelson <shannon.nelson@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Cloud filter mode for set_switch_config command
Amritha Nambiar [Fri, 27 Oct 2017 09:35:45 +0000 (02:35 -0700)]
i40e: Cloud filter mode for set_switch_config command

Add definitions for L4 filters and switch modes based on cloud filters
modes and extend the set switch config command to include the
additional cloud filter mode.

Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Signed-off-by: Kiran Patil <kiran.patil@intel.com>
Acked-by: Shannon Nelson <shannon.nelson@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Map TCs with the VSI seids
Amritha Nambiar [Fri, 27 Oct 2017 09:35:40 +0000 (02:35 -0700)]
i40e: Map TCs with the VSI seids

Add mapping of TCs with the seids of the channel VSIs. TC0
will be mapped to the main VSI seid and all other TCs are
mapped to the seid of the corresponding channel VSI.

Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Acked-by: Shannon Nelson <shannon.nelson@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agonet: sched: Identify hardware traffic classes using classid
Amritha Nambiar [Fri, 27 Oct 2017 09:35:34 +0000 (02:35 -0700)]
net: sched: Identify hardware traffic classes using classid

This patch offloads the classid to hardware and uses the classid
reserved in the range :ffe0 - :ffef to identify hardware traffic
classes reported via dev->num_tc.

tcf_result structure contains the class ID of the class to which
the packet belongs and is offloaded to hardware via flower filter.
A new helper function is introduced to represent HW traffic
classes 0 through 15 using the reserved classid values :ffe0 - :ffef.

Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Acked-by: Shannon Nelson <shannon.nelson@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/i40evf: Revert "i40e/i40evf: bump tail only in multiples of 8"
Alexander Duyck [Sun, 22 Oct 2017 00:51:50 +0000 (17:51 -0700)]
i40e/i40evf: Revert "i40e/i40evf: bump tail only in multiples of 8"

This reverts commit 11f29003d6376fb123b7c3779dba49bb56fb0815.

I am reverting this as I am fairly certain this can result in a memory leak
when combined with the current page recycling scheme. Specifically we end
up attempting to allocate fewer buffers than we recycled and this results
in us rewinding the next to alloc pointer which leads to leaks when we
overwrite the rx_buffer_info when processing the next frame.

Fixes: 11f29003d637 ("i40e/i40evf: bump tail only in multiples of 8")
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: only redistribute MSI-X vectors when needed
Shannon Nelson [Tue, 10 Oct 2017 21:56:58 +0000 (14:56 -0700)]
i40e: only redistribute MSI-X vectors when needed

Whether or not there are vectors_left, we only need to redistribute
our vectors if we didn't get as many as we requested.  With the current
check, the code will try to redistribute even if we did in fact get all
the vectors we requested - this can happen when we have more CPUs than
we do vectors.  This restores an earlier check to be sure we only
redistribute if we didn't get the full count we requested.

Fixes: 4ce20abc645f (i40e: fix MSI-X vector redistribution if hw limit is reached)
Signed-off-by: Shannon Nelson <shannon.nelson@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: mark PM functions as __maybe_unused
Arnd Bergmann [Tue, 10 Oct 2017 08:17:38 +0000 (10:17 +0200)]
i40e: mark PM functions as __maybe_unused

A cleanup of the PM code left an incorrect #ifdef in place, leading
to a harmless build warning:

drivers/net/ethernet/intel/i40e/i40e_main.c:12223:12: error: 'i40e_resume' defined but not used [-Werror=unused-function]
drivers/net/ethernet/intel/i40e/i40e_main.c:12185:12: error: 'i40e_suspend' defined but not used [-Werror=unused-function]

It's easier to use __maybe_unused attributes here, since you
can't pick the wrong one.

Fixes: 0e5d3da40055 ("i40e: use newer generic PM support instead of legacy PM callbacks")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Acked-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agonet: filter: remove unused variable and fix warning
Jakub Kicinski [Mon, 30 Oct 2017 20:46:47 +0000 (13:46 -0700)]
net: filter: remove unused variable and fix warning

bpf_getsockopt bpf call sets the ret variable to zero and
never changes it.  What's worse in case CONFIG_INET is
not selected the variable is completely unused generating
a warning.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Lawrence Brakmo <brakmo@fb.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Mon, 30 Oct 2017 05:10:01 +0000 (14:10 +0900)]
Merge git://git./linux/kernel/git/davem/net

Several conflicts here.

NFP driver bug fix adding nfp_netdev_is_nfp_repr() check to
nfp_fl_output() needed some adjustments because the code block is in
an else block now.

Parallel additions to net/pkt_cls.h and net/sch_generic.h

A bug fix in __tcp_retransmit_skb() conflicted with some of
the rbtree changes in net-next.

The tc action RCU callback fixes in 'net' had some overlap with some
of the recent tcf_block reworking.

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoLinux 4.14-rc7 v4.14-rc7
Linus Torvalds [Sun, 29 Oct 2017 20:58:38 +0000 (13:58 -0700)]
Linux 4.14-rc7

7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Sun, 29 Oct 2017 15:11:49 +0000 (08:11 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Fix route leak in xfrm_bundle_create().

 2) In mac80211, validate user rate mask before configuring it. From
    Johannes Berg.

 3) Properly enforce memory limits in fair queueing code, from Toke
    Hoiland-Jorgensen.

 4) Fix lockdep splat in inet_csk_route_req(), from Eric Dumazet.

 5) Fix TSO header allocation and management in mvpp2 driver, from Yan
    Markman.

 6) Don't take socket lock in BH handler in strparser code, from Tom
    Herbert.

 7) Don't show sockets from other namespaces in AF_UNIX code, from
    Andrei Vagin.

 8) Fix double free in error path of tap_open(), from Girish Moodalbail.

 9) Fix TX map failure path in igb and ixgbe, from Jean-Philippe Brucker
    and Alexander Duyck.

10) Fix DCB mode programming in stmmac driver, from Jose Abreu.

11) Fix err_count handling in various tunnels (ipip, ip6_gre). From Xin
    Long.

12) Properly align SKB head before building SKB in tuntap, from Jason
    Wang.

13) Avoid matching qdiscs with a zero handle during lookups, from Cong
    Wang.

14) Fix various endianness bugs in sctp, from Xin Long.

15) Fix tc filter callback races and add selftests which trigger the
    problem, from Cong Wang.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (73 commits)
  selftests: Introduce a new test case to tc testsuite
  selftests: Introduce a new script to generate tc batch file
  net_sched: fix call_rcu() race on act_sample module removal
  net_sched: add rtnl assertion to tcf_exts_destroy()
  net_sched: use tcf_queue_work() in tcindex filter
  net_sched: use tcf_queue_work() in rsvp filter
  net_sched: use tcf_queue_work() in route filter
  net_sched: use tcf_queue_work() in u32 filter
  net_sched: use tcf_queue_work() in matchall filter
  net_sched: use tcf_queue_work() in fw filter
  net_sched: use tcf_queue_work() in flower filter
  net_sched: use tcf_queue_work() in flow filter
  net_sched: use tcf_queue_work() in cgroup filter
  net_sched: use tcf_queue_work() in bpf filter
  net_sched: use tcf_queue_work() in basic filter
  net_sched: introduce a workqueue for RCU callbacks of tc filter
  sctp: fix some type cast warnings introduced since very beginning
  sctp: fix a type cast warnings that causes a_rwnd gets the wrong value
  sctp: fix some type cast warnings introduced by transport rhashtable
  sctp: fix some type cast warnings introduced by stream reconf
  ...

7 years agoMerge branch 'net_sched-fix-races-with-RCU-callbacks'
David S. Miller [Sun, 29 Oct 2017 13:49:32 +0000 (22:49 +0900)]
Merge branch 'net_sched-fix-races-with-RCU-callbacks'

Cong Wang says:

====================
net_sched: fix races with RCU callbacks

Recently, the RCU callbacks used in TC filters and TC actions keep
drawing my attention, they introduce at least 4 race condition bugs:

1. A simple one fixed by Daniel:

commit c78e1746d3ad7d548bdf3fe491898cc453911a49
Author: Daniel Borkmann <daniel@iogearbox.net>
Date:   Wed May 20 17:13:33 2015 +0200

    net: sched: fix call_rcu() race on classifier module unloads

2. A very nasty one fixed by me:

commit 1697c4bb5245649a23f06a144cc38c06715e1b65
Author: Cong Wang <xiyou.wangcong@gmail.com>
Date:   Mon Sep 11 16:33:32 2017 -0700

    net_sched: carefully handle tcf_block_put()

3. Two more bugs found by Chris:
https://patchwork.ozlabs.org/patch/826696/
https://patchwork.ozlabs.org/patch/826695/

Usually RCU callbacks are simple, however for TC filters and actions,
they are complex because at least TC actions could be destroyed
together with the TC filter in one callback. And RCU callbacks are
invoked in BH context, without locking they are parallel too. All of
these contribute to the cause of these nasty bugs.

Alternatively, we could also:

a) Introduce a spinlock to serialize these RCU callbacks. But as I
said in commit 1697c4bb5245 ("net_sched: carefully handle
tcf_block_put()"), it is very hard to do because of tcf_chain_dump().
Potentially we need to do a lot of work to make it possible (if not
impossible).

b) Just get rid of these RCU callbacks, because they are not
necessary at all, callers of these call_rcu() are all on slow paths
and holding RTNL lock, so blocking is allowed in their contexts.
However, David and Eric dislike adding synchronize_rcu() here.

As suggested by Paul, we could defer the work to a workqueue and
gain the permission of holding RTNL again without any performance
impact, however, in tcf_block_put() we could have a deadlock when
flushing workqueue while hodling RTNL lock, the trick here is to
defer the work itself in workqueue and make it queued after all
other works so that we keep the same ordering to avoid any
use-after-free. Please see the first patch for details.

Patch 1 introduces the infrastructure, patch 2~12 move each
tc filter to the new tc filter workqueue, patch 13 adds
an assertion to catch potential bugs like this, patch 14
closes another rcu callback race, patch 15 and patch 16 add
new test cases.
====================

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoselftests: Introduce a new test case to tc testsuite
Chris Mi [Fri, 27 Oct 2017 01:24:43 +0000 (18:24 -0700)]
selftests: Introduce a new test case to tc testsuite

In this patchset, we fixed a tc bug. This patch adds the test case
that reproduces the bug. To run this test case, user should specify
an existing NIC device:
  # sudo ./tdc.py -d enp4s0f0

This test case belongs to category "flower". If user doesn't specify
a NIC device, the test cases belong to "flower" will not be run.

In this test case, we create 1M filters and all filters share the same
action. When destroying all filters, kernel should not panic. It takes
about 18s to run it.

Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Lucas Bates <lucasb@mojatatu.com>
Signed-off-by: Chris Mi <chrism@mellanox.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoselftests: Introduce a new script to generate tc batch file
Chris Mi [Fri, 27 Oct 2017 01:24:42 +0000 (18:24 -0700)]
selftests: Introduce a new script to generate tc batch file

  # ./tdc_batch.py -h
  usage: tdc_batch.py [-h] [-n NUMBER] [-o] [-s] [-p] device file

  TC batch file generator

  positional arguments:
    device                device name
    file                  batch file name

  optional arguments:
    -h, --help            show this help message and exit
    -n NUMBER, --number NUMBER
                          how many lines in batch file
    -o, --skip_sw         skip_sw (offload), by default skip_hw
    -s, --share_action    all filters share the same action
    -p, --prio            all filters have different prio

Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Lucas Bates <lucasb@mojatatu.com>
Signed-off-by: Chris Mi <chrism@mellanox.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: fix call_rcu() race on act_sample module removal
Cong Wang [Fri, 27 Oct 2017 01:24:41 +0000 (18:24 -0700)]
net_sched: fix call_rcu() race on act_sample module removal

Similar to commit c78e1746d3ad
("net: sched: fix call_rcu() race on classifier module unloads"),
we need to wait for flying RCU callback tcf_sample_cleanup_rcu().

Cc: Yotam Gigi <yotamg@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: add rtnl assertion to tcf_exts_destroy()
Cong Wang [Fri, 27 Oct 2017 01:24:40 +0000 (18:24 -0700)]
net_sched: add rtnl assertion to tcf_exts_destroy()

After previous patches, it is now safe to claim that
tcf_exts_destroy() is always called with RTNL lock.

Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in tcindex filter
Cong Wang [Fri, 27 Oct 2017 01:24:39 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in tcindex filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in rsvp filter
Cong Wang [Fri, 27 Oct 2017 01:24:38 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in rsvp filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in route filter
Cong Wang [Fri, 27 Oct 2017 01:24:37 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in route filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in u32 filter
Cong Wang [Fri, 27 Oct 2017 01:24:36 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in u32 filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in matchall filter
Cong Wang [Fri, 27 Oct 2017 01:24:35 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in matchall filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in fw filter
Cong Wang [Fri, 27 Oct 2017 01:24:34 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in fw filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in flower filter
Cong Wang [Fri, 27 Oct 2017 01:24:33 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in flower filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in flow filter
Cong Wang [Fri, 27 Oct 2017 01:24:32 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in flow filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in cgroup filter
Cong Wang [Fri, 27 Oct 2017 01:24:31 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in cgroup filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in bpf filter
Cong Wang [Fri, 27 Oct 2017 01:24:30 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in bpf filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: use tcf_queue_work() in basic filter
Cong Wang [Fri, 27 Oct 2017 01:24:29 +0000 (18:24 -0700)]
net_sched: use tcf_queue_work() in basic filter

Defer the tcf_exts_destroy() in RCU callback to
tc filter workqueue and get RTNL lock.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: introduce a workqueue for RCU callbacks of tc filter
Cong Wang [Fri, 27 Oct 2017 01:24:28 +0000 (18:24 -0700)]
net_sched: introduce a workqueue for RCU callbacks of tc filter

This patch introduces a dedicated workqueue for tc filters
so that each tc filter's RCU callback could defer their
action destroy work to this workqueue. The helper
tcf_queue_work() is introduced for them to use.

Because we hold RTNL lock when calling tcf_block_put(), we
can not simply flush works inside it, therefore we have to
defer it again to this workqueue and make sure all flying RCU
callbacks have already queued their work before this one, in
other words, to ensure this is the last one to execute to
prevent any use-after-free.

On the other hand, this makes tcf_block_put() ugly and
harder to understand. Since David and Eric strongly dislike
adding synchronize_rcu(), this is probably the only
solution that could make everyone happy.

Please also see the code comments below.

Reported-by: Chris Mi <chrism@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'ipvlan-private-vepa'
David S. Miller [Sun, 29 Oct 2017 09:39:58 +0000 (18:39 +0900)]
Merge branch 'ipvlan-private-vepa'

Mahesh Bandewar says:

====================
add 'private' and 'vepa' attributes to ipvlan modes

IPvlan has always been operating in bridge-mode for its supported modes i.e.
if the packets are destined to the adjacent neighbor dev, then IPvlan driver
will switch the packet internally without needing the packets to hit the
wire or get routed. However, there are situations where this bridge-mode is
not needed. e.g. two private processes running inside two namespaces which
are having one IPvlan slave each for its namespace but sharing the master. These
processes should reach the outside world through the master device but at
the same time the bridge function should not work. Currently that's not
possible hence the private attribute for the selected mode comes in play.

VEPA or 802.1Qbg on the other hand has limited appeal with IPvlan since IPvlan
uses the mac-address of the lower device. So packets that are destined to
the adjacent neighbor slave-dev will have same src and dest mac. When these
packets reach the external switch/router, they will send you the redirect
message which the host will have to deal with. Having said that this attribute
will have appeal in debugging as IPvlan will not switch / short-circuit
packets internally. e.g. using VEPA mode with lower-device in loopback mode
will avoid some complicated set-ups that use non-local-bind with some route
jugglery.

This patch-set implements these attributes for the existing modes that
IPvlan has. Please see individual patches for their detailed implementation.
A subsequent ip-utils patch is needed and will be sent soon.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipvlan: implement VEPA mode
Mahesh Bandewar [Thu, 26 Oct 2017 22:09:25 +0000 (15:09 -0700)]
ipvlan: implement VEPA mode

This is very similar to the Macvlan VEPA mode, however, there is some
difference. IPvlan uses the mac-address of the lower device, so the VEPA
mode has implications of ICMP-redirects for packets destined for its
immediate neighbors sharing same master since the packets will have same
source and dest mac. The external switch/router will send redirect msg.

Having said that, this will be useful tool in terms of debugging
since IPvlan will not switch packets within its slaves and rely completely
on the external entity as intended in 802.1Qbg.

Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipvlan: introduce 'private' attribute for all existing modes.
Mahesh Bandewar [Thu, 26 Oct 2017 22:09:21 +0000 (15:09 -0700)]
ipvlan: introduce 'private' attribute for all existing modes.

IPvlan has always operated in bridge mode. However there are scenarios
where each slave should be able to talk through the master device but
not necessarily across each other. Think of an environment where each
of a namespace is a private and independant customer. In this scenario
the machine which is hosting these namespaces neither want to tell who
their neighbor is nor the individual namespaces care to talk to neighbor
on short-circuited network path.

This patch implements the mode that is very similar to the 'private' mode
in macvlan where individual slaves can send and receive traffic through
the master device, just that they can not talk among slave devices.

Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotools: bpftool: add bash completion for bpftool
Quentin Monnet [Thu, 26 Oct 2017 21:16:05 +0000 (14:16 -0700)]
tools: bpftool: add bash completion for bpftool

Add a completion file for bash. The completion function runs bpftool
when needed, making it smart enough to help users complete ids or tags
for eBPF programs and maps currently on the system.

Update Makefile to install completion file to
/usr/share/bash-completion/completions when running `make install`.

Emacs file mode and (at the end) Vim modeline have been added, to keep
the style in use for most existing bash completion files. In this, it
differs from tools/perf/perf-completion.sh, which seems to be the only
other completion file among the kernel sources repository. This is also
valid for indent style: 4-space indents, as in other completion files.

Signed-off-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'sctp-endianness-fixes'
David S. Miller [Sun, 29 Oct 2017 09:03:25 +0000 (18:03 +0900)]
Merge branch 'sctp-endianness-fixes'

Xin Long says:

====================
sctp: a bunch of fixes for some sparse warnings

As Eric noticed, when running 'make C=2 M=net/sctp/', a plenty of
warnings or errors checked by sparse appear. They are all problems
about Endian and type cast.

Most of them are just warnings by which no issues could be caused
while some might be bugs.

This patchset fixes them with four patches basically according to
how they are introduced.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosctp: fix some type cast warnings introduced since very beginning
Xin Long [Sat, 28 Oct 2017 11:43:57 +0000 (19:43 +0800)]
sctp: fix some type cast warnings introduced since very beginning

These warnings were found by running 'make C=2 M=net/sctp/'.
They are there since very beginning.

Note after this patch, there still one warning left in
sctp_outq_flush():
  sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM)

Since it has been moved to sctp_stream_outq_migrate on net-next,
to avoid the extra job when merging net-next to net, I will post
the fix for it after the merging is done.

Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosctp: fix a type cast warnings that causes a_rwnd gets the wrong value
Xin Long [Sat, 28 Oct 2017 11:43:56 +0000 (19:43 +0800)]
sctp: fix a type cast warnings that causes a_rwnd gets the wrong value

These warnings were found by running 'make C=2 M=net/sctp/'.

Commit d4d6fb5787a6 ("sctp: Try not to change a_rwnd when faking a
SACK from SHUTDOWN.") expected to use the peers old rwnd and add
our flight size to the a_rwnd. But with the wrong Endian, it may
not work as well as expected.

So fix it by converting to the right value.

Fixes: d4d6fb5787a6 ("sctp: Try not to change a_rwnd when faking a SACK from SHUTDOWN.")
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosctp: fix some type cast warnings introduced by transport rhashtable
Xin Long [Sat, 28 Oct 2017 11:43:55 +0000 (19:43 +0800)]
sctp: fix some type cast warnings introduced by transport rhashtable

These warnings were found by running 'make C=2 M=net/sctp/'.

They are introduced by not aware of Endian for the port when
coding transport rhashtable patches.

Fixes: 7fda702f9315 ("sctp: use new rhlist interface on sctp transport rhashtable")
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosctp: fix some type cast warnings introduced by stream reconf
Xin Long [Sat, 28 Oct 2017 11:43:54 +0000 (19:43 +0800)]
sctp: fix some type cast warnings introduced by stream reconf

These warnings were found by running 'make C=2 M=net/sctp/'.

They are introduced by not aware of Endian when coding stream
reconf patches.

Since commit c0d8bab6ae51 ("sctp: add get and set sockopt for
reconf_enable") enabled stream reconf feature for users, the
Fixes tag below would use it.

Fixes: c0d8bab6ae51 ("sctp: add get and set sockopt for reconf_enable")
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: avoid matching qdisc with zero handle
Cong Wang [Sat, 28 Oct 2017 05:08:56 +0000 (22:08 -0700)]
net_sched: avoid matching qdisc with zero handle

Davide found the following script triggers a NULL pointer
dereference:

ip l a name eth0 type dummy
tc q a dev eth0 parent :1 handle 1: htb

This is because for a freshly created netdevice noop_qdisc
is attached and when passing 'parent :1', kernel actually
tries to match the major handle which is 0 and noop_qdisc
has handle 0 so is matched by mistake. Commit 69012ae425d7
tries to fix a similar bug but still misses this case.

Handle 0 is not a valid one, should be just skipped. In
fact, kernel uses it as TC_H_UNSPEC.

Fixes: 69012ae425d7 ("net: sched: fix handling of singleton qdiscs with qdisc_hash")
Fixes: 59cc1f61f09c ("net: sched:convert qdisc linked list to hashtable")
Reported-by: Davide Caratti <dcaratti@redhat.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: aquantia: Make local functions static
Wei Yongjun [Sat, 28 Oct 2017 05:03:38 +0000 (05:03 +0000)]
net: aquantia: Make local functions static

Fixes the following sparse warnings:

drivers/net/ethernet/aquantia/atlantic/aq_ethtool.c:224:5: warning:
 symbol 'aq_ethtool_get_coalesce' was not declared. Should it be static?
drivers/net/ethernet/aquantia/atlantic/aq_ethtool.c:245:5: warning:
 symbol 'aq_ethtool_set_coalesce' was not declared. Should it be static?

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipv6: prevent user from adding cached routes
Wei Wang [Sat, 28 Oct 2017 00:30:12 +0000 (17:30 -0700)]
ipv6: prevent user from adding cached routes

Cached routes should only be created by the system when receiving pmtu
discovery or ip redirect msg. Users should not be allowed to create
cached routes.

Furthermore, after the patch series to move cached routes into exception
table, user added cached routes will trigger the following warning in
fib6_add():

WARNING: CPU: 0 PID: 2985 at net/ipv6/ip6_fib.c:1137
fib6_add+0x20d9/0x2c10 net/ipv6/ip6_fib.c:1137
Kernel panic - not syncing: panic_on_warn set ...

CPU: 0 PID: 2985 Comm: syzkaller320388 Not tainted 4.14.0-rc3+ #74
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:16 [inline]
 dump_stack+0x194/0x257 lib/dump_stack.c:52
 panic+0x1e4/0x417 kernel/panic.c:181
 __warn+0x1c4/0x1d9 kernel/panic.c:542
 report_bug+0x211/0x2d0 lib/bug.c:183
 fixup_bug+0x40/0x90 arch/x86/kernel/traps.c:178
 do_trap_no_signal arch/x86/kernel/traps.c:212 [inline]
 do_trap+0x260/0x390 arch/x86/kernel/traps.c:261
 do_error_trap+0x120/0x390 arch/x86/kernel/traps.c:298
 do_invalid_op+0x1b/0x20 arch/x86/kernel/traps.c:311
 invalid_op+0x18/0x20 arch/x86/entry/entry_64.S:905
RIP: 0010:fib6_add+0x20d9/0x2c10 net/ipv6/ip6_fib.c:1137
RSP: 0018:ffff8801cf09f6a0 EFLAGS: 00010297
RAX: ffff8801ce45e340 RBX: 1ffff10039e13eec RCX: ffff8801d749c814
RDX: 0000000000000000 RSI: ffff8801d749c700 RDI: ffff8801d749c780
RBP: ffff8801cf09fa08 R08: 0000000000000000 R09: ffff8801cf09f360
R10: ffff8801cf09f2d8 R11: 1ffff10039c8befb R12: 0000000000000001
R13: dffffc0000000000 R14: ffff8801d749c700 R15: ffffffff860655c0
 __ip6_ins_rt+0x6c/0x90 net/ipv6/route.c:1011
 ip6_route_add+0x148/0x1a0 net/ipv6/route.c:2782
 ipv6_route_ioctl+0x4d5/0x690 net/ipv6/route.c:3291
 inet6_ioctl+0xef/0x1e0 net/ipv6/af_inet6.c:521
 sock_do_ioctl+0x65/0xb0 net/socket.c:961
 sock_ioctl+0x2c2/0x440 net/socket.c:1058
 vfs_ioctl fs/ioctl.c:45 [inline]
 do_vfs_ioctl+0x1b1/0x1530 fs/ioctl.c:685
 SYSC_ioctl fs/ioctl.c:700 [inline]
 SyS_ioctl+0x8f/0xc0 fs/ioctl.c:691
 entry_SYSCALL_64_fastpath+0x1f/0xbe

So we fix this by failing the attemp to add cached routes from userspace
with returning EINVAL error.

Fixes: 2b760fcf5cfb ("ipv6: hook up exception table to store dst cache")
Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosamples/bpf: adjust rlimit RLIMIT_MEMLOCK for xdp_redirect_map
Tushar Dave [Sat, 28 Oct 2017 00:28:22 +0000 (17:28 -0700)]
samples/bpf: adjust rlimit RLIMIT_MEMLOCK for xdp_redirect_map

Default rlimit RLIMIT_MEMLOCK is 64KB, causes bpf map failure.
e.g.
[root@labbpf]# ./xdp_redirect_map $(</sys/class/net/eth2/ifindex) \
> $(</sys/class/net/eth3/ifindex)
failed to create a map: 1 Operation not permitted

The failure is 100% when multiple xdp programs are running. Fix it.

Signed-off-by: Tushar Dave <tushar.n.dave@oracle.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosamples/bpf: adjust rlimit RLIMIT_MEMLOCK for xdp1
Tushar Dave [Fri, 27 Oct 2017 23:12:30 +0000 (16:12 -0700)]
samples/bpf: adjust rlimit RLIMIT_MEMLOCK for xdp1

Default rlimit RLIMIT_MEMLOCK is 64KB, causes bpf map failure.
e.g.
[root@lab bpf]#./xdp1 -N $(</sys/class/net/eth2/ifindex)
failed to create a map: 1 Operation not permitted

Fix it.

Signed-off-by: Tushar Dave <tushar.n.dave@oracle.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: b53: Export b53_configure_vlan()
Florian Fainelli [Fri, 27 Oct 2017 22:56:01 +0000 (15:56 -0700)]
net: dsa: b53: Export b53_configure_vlan()

bcm_sf2 and b53 replicate the same operations: clear all VLANs and set
their ports to the default VLAN tag (1 for these devices) so export the
b53 function doing just that.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio: get rid of false alarm "Unknown cmd 27" in dmesg
Felix Manlunas [Fri, 27 Oct 2017 21:37:03 +0000 (14:37 -0700)]
liquidio: get rid of false alarm "Unknown cmd 27" in dmesg

Creating a macvtap interface with the liquidio VF driver as lower device
causes this alarming message to show up in dmesg:

    liquidio_link_ctrl_cmd_completion Unknown cmd 27

That's actually a false alarm because cmd 27 is the value of the macro
OCTNET_CMD_SET_UC_LIST which is known.  It's a control command sent from
host to NIC firmware to set the unicast MAC address list of the macvtap
lower device.

Make the false alarm go away by adding a case for OCTNET_CMD_SET_UC_LIST
in liquidio_link_ctrl_cmd_completion().

Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agohv_netvsc: Set tx_table to equal weight after subchannels open
Haiyang Zhang [Fri, 27 Oct 2017 19:36:38 +0000 (12:36 -0700)]
hv_netvsc: Set tx_table to equal weight after subchannels open

In some cases, like internal vSwitch, the host doesn't provide
send indirection table updates. This patch sets the table to be
equal weight after subchannels are all open. Otherwise, all workload
will be on one TX channel.

As tested, this patch has largely increased the throughput over
internal vSwitch.

Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosctp: reset owner sk for data chunks on out queues when migrating a sock
Xin Long [Fri, 27 Oct 2017 18:13:29 +0000 (02:13 +0800)]
sctp: reset owner sk for data chunks on out queues when migrating a sock

Now when migrating sock to another one in sctp_sock_migrate(), it only
resets owner sk for the data in receive queues, not the chunks on out
queues.

It would cause that data chunks length on the sock is not consistent
with sk sk_wmem_alloc. When closing the sock or freeing these chunks,
the old sk would never be freed, and the new sock may crash due to
the overflow sk_wmem_alloc.

syzbot found this issue with this series:

  r0 = socket$inet_sctp()
  sendto$inet(r0)
  listen(r0)
  accept4(r0)
  close(r0)

Although listen() should have returned error when one TCP-style socket
is in connecting (I may fix this one in another patch), it could also
be reproduced by peeling off an assoc.

This issue is there since very beginning.

This patch is to reset owner sk for the chunks on out queues so that
sk sk_wmem_alloc has correct value after accept one sock or peeloff
an assoc to one sock.

Note that when resetting owner sk for chunks on outqueue, it has to
sctp_clear_owner_w/skb_orphan chunks before changing assoc->base.sk
first and then sctp_set_owner_w them after changing assoc->base.sk,
due to that sctp_wfree and it's callees are using assoc->base.sk.

Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoppp: allow usage in namespaces
Matteo Croce [Fri, 27 Oct 2017 18:08:23 +0000 (20:08 +0200)]
ppp: allow usage in namespaces

Check for CAP_NET_ADMIN with ns_capable() instead of capable()
to allow usage of ppp in user namespace other than the init one.

Signed-off-by: Matteo Croce <mcroce@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Sun, 29 Oct 2017 02:20:28 +0000 (11:20 +0900)]
Merge branch '1GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
1GbE Intel Wired LAN Driver Updates 2017-10-27

This patchset is a proposal of how the Traffic Control subsystem can
be used to offload the configuration of the Credit Based Shaper
(defined in the IEEE 802.1Q-2014 Section 8.6.8.2) into supported
network devices.

As part of this work, we've assessed previous public discussions
related to TSN enabling: patches from Henrik Austad (Cisco), the
presentation from Eric Mann at Linux Plumbers 2012, patches from
Gangfeng Huang (National Instruments) and the current state of the
OpenAVNU project (https://github.com/AVnu/OpenAvnu/).

Overview
========

Time-sensitive Networking (TSN) is a set of standards that aim to
address resources availability for providing bandwidth reservation and
bounded latency on Ethernet based LANs. The proposal described here
aims to cover mainly what is needed to enable the following standards:
802.1Qat and 802.1Qav.

The initial target of this work is the Intel i210 NIC, but other
controllers' datasheet were also taken into account, like the Renesas
RZ/A1H RZ/A1M group and the Synopsis DesignWare Ethernet QoS
controller.

Proposal
========

Feature-wise, what is covered here is the configuration interfaces for
HW implementations of the Credit-Based shaper (CBS, 802.1Qav). CBS is
a per-queue shaper. Given that this feature is related to traffic
shaping, and that the traffic control subsystem already provides a
queueing discipline that offloads config into the device driver (i.e.
mqprio), designing a new qdisc for the specific purpose of offloading
the config for the CBS shaper seemed like a good fit.

For steering traffic into the correct queues, we use the socket option
SO_PRIORITY and then a mechanism to map priority to traffic classes /
Tx queues. The qdisc mqprio is currently used in our tests.

As for the CBS config interface, this patchset is proposing a new
qdisc called 'cbs'. Its 'tc' cmd line is:

$ tc qdisc add dev IFACE parent ID cbs locredit N hicredit M sendslope S \
     idleslope I

   Note that the parameters for this qdisc are the ones defined by the
   802.1Q-2014 spec, so no hardware specific functionality is exposed here.

Per-stream shaping, as defined by IEEE 802.1Q-2014 Section 34.6.1, is
not yet covered by this proposal.

v2: Merged patch 6 of the original series into patch 4 based on feedback
    from David Miller.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'sockmap-fixes'
David S. Miller [Sun, 29 Oct 2017 02:18:49 +0000 (11:18 +0900)]
Merge branch 'sockmap-fixes'

John Fastabend says:

====================
net: sockmap fixes

Last two fixes (as far as I know) for sockmap code this round.

First, we are using the qdisc cb structure when making the data end
calculation. This is really just wrong so, store it with the other
metadata in the correct tcp_skb_cb sturct to avoid breaking things.

Next, with recent work to attach multiple programs to a cgroup a
specific enumeration of return codes was agreed upon. However,
I wrote the sk_skb program types before seeing this work and used
a different convention. Patch 2 in the series aligns the return
codes to avoid breaking with this infrastructure and also aligns
with other programming conventions to avoid being the odd duck out
forcing programs to remember SK_SKB programs are different. Pusing
to net because its a user visible change. With this SK_SKB program
return codes are the same as other cgroup program types.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: rename sk_actions to align with bpf infrastructure
John Fastabend [Fri, 27 Oct 2017 16:45:53 +0000 (09:45 -0700)]
bpf: rename sk_actions to align with bpf infrastructure

Recent additions to support multiple programs in cgroups impose
a strict requirement, "all yes is yes, any no is no". To enforce
this the infrastructure requires the 'no' return code, SK_DROP in
this case, to be 0.

To apply these rules to SK_SKB program types the sk_actions return
codes need to be adjusted.

This fix adds SK_PASS and makes 'SK_DROP = 0'. Finally, remove
SK_ABORTED to remove any chance that the API may allow aborted
program flows to be passed up the stack. This would be incorrect
behavior and allow programs to break existing policies.

Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: bpf_compute_data uses incorrect cb structure
John Fastabend [Fri, 27 Oct 2017 16:45:34 +0000 (09:45 -0700)]
bpf: bpf_compute_data uses incorrect cb structure

SK_SKB program types use bpf_compute_data to store the end of the
packet data. However, bpf_compute_data assumes the cb is stored in the
qdisc layer format. But, for SK_SKB this is the wrong layer of the
stack for this type.

It happens to work (sort of!) because in most cases nothing happens
to be overwritten today. This is very fragile and error prone.
Fortunately, we have another hole in tcp_skb_cb we can use so lets
put the data_end value there.

Note, SK_SKB program types do not use data_meta, they are failed by
sk_skb_is_valid_access().

Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'l2tp-register-sessions-atomically'
David S. Miller [Sun, 29 Oct 2017 02:16:22 +0000 (11:16 +0900)]
Merge branch 'l2tp-register-sessions-atomically'

Guillaume Nault says:

====================
l2tp: register sessions atomically

Currently l2tp_session_create() allocates a session, partially
initialises it and finally registers it. It therefore exposes sessions
that aren't fully initialised to the rest of the system, because
pseudo-wire specific initialisation can only happen after
l2tp_session_create() returns.
This leads to several crashes when these sessions are used or deleted.

This series starts by splitting session registration out of
l2tp_session_create() (patch #1). Thus allowing pseudo-wires code to
terminate the initialisation phase before registration.

Then patch #2 fixes the eth pseudo-wire code. This requires protecting
the session's netdevice pointer with RCU, because it still needs to be
updated concurrently after the session got registered.

Remaining patches take care of ppp pseudo-wires. RCU protection is
needed there too, for the same reasons. This time it's the pppol2tp
socket pointer that gets protected. For clarity, and since the
conversion requires more modifications, introducing RCU is done in
its own patch (#3). Then patch #4 only has to take care of fixing
sessions initialisation and registration (and adapting part of the
deletion process).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agol2tp: initialise PPP sessions before registering them
Guillaume Nault [Fri, 27 Oct 2017 14:51:52 +0000 (16:51 +0200)]
l2tp: initialise PPP sessions before registering them

pppol2tp_connect() initialises L2TP sessions after they've been exposed
to the rest of the system by l2tp_session_register(). This puts
sessions into transient states that are the source of several races, in
particular with session's deletion path.

This patch centralises the initialisation code into
pppol2tp_session_init(), which is called before the registration phase.
The only field that can't be set before session registration is the
pppol2tp socket pointer, which has already been converted to RCU. So
pppol2tp_connect() should now be race-free.

The session's .session_close() callback is now set before registration.
Therefore, it's always called when l2tp_core deletes the session, even
if it was created by pppol2tp_session_create() and hasn't been plugged
to a pppol2tp socket yet. That'd prevent session free because the extra
reference taken by pppol2tp_session_close() wouldn't be dropped by the
socket's ->sk_destruct() callback (pppol2tp_session_destruct()).
We could set .session_close() only while connecting a session to its
pppol2tp socket, or teach pppol2tp_session_close() to avoid grabbing a
reference when the session isn't connected, but that'd require adding
some form of synchronisation to be race free.

Instead of that, we can just let the pppol2tp socket hold a reference
on the session as soon as it starts depending on it (that is, in
pppol2tp_connect()). Then we don't need to utilise
pppol2tp_session_close() to hold a reference at the last moment to
prevent l2tp_core from dropping it.

When releasing the socket, pppol2tp_release() now deletes the session
using the standard l2tp_session_delete() function, instead of merely
removing it from hash tables. l2tp_session_delete() drops the reference
the sessions holds on itself, but also makes sure it doesn't remove a
session twice. So it can safely be called, even if l2tp_core already
tried, or is concurrently trying, to remove the session.
Finally, pppol2tp_session_destruct() drops the reference held by the
socket.

Fixes: fd558d186df2 ("l2tp: Split pppol2tp patch into separate l2tp and ppp parts")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agol2tp: protect sock pointer of struct pppol2tp_session with RCU
Guillaume Nault [Fri, 27 Oct 2017 14:51:52 +0000 (16:51 +0200)]
l2tp: protect sock pointer of struct pppol2tp_session with RCU

pppol2tp_session_create() registers sessions that can't have their
corresponding socket initialised. This socket has to be created by
userspace, then connected to the session by pppol2tp_connect().
Therefore, we need to protect the pppol2tp socket pointer of L2TP
sessions, so that it can safely be updated when userspace is connecting
or closing the socket. This will eventually allow pppol2tp_connect()
to avoid generating transient states while initialising its parts of the
session.

To this end, this patch protects the pppol2tp socket pointer using RCU.

The pppol2tp socket pointer is still set in pppol2tp_connect(), but
only once we know the function isn't going to fail. It's eventually
reset by pppol2tp_release(), which now has to wait for a grace period
to elapse before it can drop the last reference on the socket. This
ensures that pppol2tp_session_get_sock() can safely grab a reference
on the socket, even after ps->sk is reset to NULL but before this
operation actually gets visible from pppol2tp_session_get_sock().

The rest is standard RCU conversion: pppol2tp_recv(), which already
runs in atomic context, is simply enclosed by rcu_read_lock() and
rcu_read_unlock(), while other functions are converted to use
pppol2tp_session_get_sock() followed by sock_put().
pppol2tp_session_setsockopt() is a special case. It used to retrieve
the pppol2tp socket from the L2TP session, which itself was retrieved
from the pppol2tp socket. Therefore we can just avoid dereferencing
ps->sk and directly use the original socket pointer instead.

With all users of ps->sk now handling NULL and concurrent updates, the
L2TP ->ref() and ->deref() callbacks aren't needed anymore. Therefore,
rather than converting pppol2tp_session_sock_hold() and
pppol2tp_session_sock_put(), we can just drop them.

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agol2tp: initialise l2tp_eth sessions before registering them
Guillaume Nault [Fri, 27 Oct 2017 14:51:51 +0000 (16:51 +0200)]
l2tp: initialise l2tp_eth sessions before registering them

Sessions must be initialised before being made externally visible by
l2tp_session_register(). Otherwise the session may be concurrently
deleted before being initialised, which can confuse the deletion path
and eventually lead to kernel oops.

Therefore, we need to move l2tp_session_register() down in
l2tp_eth_create(), but also handle the intermediate step where only the
session or the netdevice has been registered.

We can't just call l2tp_session_register() in ->ndo_init() because
we'd have no way to properly undo this operation in ->ndo_uninit().
Instead, let's register the session and the netdevice in two different
steps and protect the session's device pointer with RCU.

And now that we allow the session's .dev field to be NULL, we don't
need to prevent the netdevice from being removed anymore. So we can
drop the dev_hold() and dev_put() calls in l2tp_eth_create() and
l2tp_eth_dev_uninit().

Fixes: d9e31d17ceba ("l2tp: Add L2TP ethernet pseudowire support")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agol2tp: don't register sessions in l2tp_session_create()
Guillaume Nault [Fri, 27 Oct 2017 14:51:50 +0000 (16:51 +0200)]
l2tp: don't register sessions in l2tp_session_create()

Sessions created by l2tp_session_create() aren't fully initialised:
some pseudo-wire specific operations need to be done before making the
session usable. Therefore the PPP and Ethernet pseudo-wires continue
working on the returned l2tp session while it's already been exposed to
the rest of the system.
This can lead to various issues. In particular, the session may enter
the deletion process before having been fully initialised, which will
confuse the session removal code.

This patch moves session registration out of l2tp_session_create(), so
that callers can control when the session is exposed to the rest of the
system. This is done by the new l2tp_session_register() function.

Only pppol2tp_session_create() can be easily converted to avoid
modifying its session after registration (the debug message is dropped
in order to avoid the need for holding a reference on the session).

For pppol2tp_connect() and l2tp_eth_create()), more work is needed.
That'll be done in followup patches. For now, let's just register the
session right after its creation, like it was done before. The only
difference is that we can easily take a reference on the session before
registering it, so, at least, we're sure it's not going to be freed
while we're working on it.

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: Remove "linux/unaligned/access_ok.h" include.
David S. Miller [Sun, 29 Oct 2017 02:14:08 +0000 (11:14 +0900)]
tcp: Remove "linux/unaligned/access_ok.h" include.

This causes build failures:

In file included from net/ipv4/tcp_input.c:79:0:
./include/linux/unaligned/access_ok.h:7:28: error: redefinition of
'get_unaligned_le16'
In file included from ./include/asm-generic/unaligned.h:17:0,
                 from ./arch/arm/include/generated/asm/unaligned.h:1,
                 from net/ipv4/tcp_input.c:76:
./include/linux/unaligned/le_struct.h:6:19: note: previous definition
of 'get_unaligned_le16' was here
In file included from net/ipv4/tcp_input.c:79:0:
./include/linux/unaligned/access_ok.h:12:28: error: redefinition of
'get_unaligned_le32'

Plain "asm/access_ok.h", which is already included, is
sufficient.

Fixes: 60e2a7780793 ("tcp: TCP experimental option for SMC")
Reported-by: Egil Hjelmeland <privat@egil-hjelmeland.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocxgb3: Check and handle the dma mapping errors
Arjun Vynipadath [Fri, 27 Oct 2017 12:38:21 +0000 (18:08 +0530)]
cxgb3: Check and handle the dma mapping errors

This patch adds checks at approprate places whether *dma_map*() call has
succeeded or not.

Original Work by: Santosh Rastapur <santosh@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>
7 years agor8169: Add support for interrupt coalesce tuning (ethtool -C)
Francois Romieu [Fri, 27 Oct 2017 10:24:49 +0000 (13:24 +0300)]
r8169: Add support for interrupt coalesce tuning (ethtool -C)

Kirr: In particular with

ethtool -C <ifname> rx-usecs 0 rx-frames 0

now it is possible to disable RX delays when NIC usage requires low-latency.

See this thread for context:

https://www.spinics.net/lists/netdev/msg217665.html

My specific case is that:

We have many computers with gigabit Realtek NICs. For 2 such computers
connected to a gigabit store-and-forward switch the minimum round-trip
time for small pings (`ping -i 0 -w 3 -s 56 -q peer`) is ~ 30μs.

However it turned out that when Ethernet frame length transitions 127 ->
128 bytes (`ping -i 0 -w 3 -s {81 -> 82} -q peer`) the lowest RTT
transitions step-wise to ~ 270μs.

As David Light said this is RX interrupt mitigation done by NIC which creates
the latency. For workloads when low-latency is required with e.g. Intel,
BCM etc NIC drivers one just uses `ethtool -C rx-usecs ...` to reduce
the time NIC delays before interrupting CPU, but it turned out
`ethtool -C` is not supported by r8169 driver.

Like Stéphane ANCELOT I've traced the problem down to IntrMitigate being
hardcoded to != 0 for our chips (we have 8168 based NICs):

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/ethernet/realtek/r8169.c#n5460
static void rtl_hw_start_8169(struct net_device *dev) {
        ...
        /*
         * Undocumented corner. Supposedly:
         * (TxTimer << 12) | (TxPackets << 8) | (RxTimer << 4) | RxPackets
         */
        RTL_W16(IntrMitigate, 0x0000);

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/ethernet/realtek/r8169.c#n6346
static void rtl_hw_start_8168(struct net_device *dev) {
        ...
        RTL_W16(IntrMitigate, 0x5151);

and then I've also found

https://www.spinics.net/lists/netdev/msg217665.html

and original Francois' patch:

https://www.spinics.net/lists/netdev/msg217984.html
https://www.spinics.net/lists/netdev/msg218207.html

So could we please finally get support for tuning r8169 interrupt
coalescing in tree? (so that next poor soul who hits the problem does
not need to go all the way to dig into driver sources and internet
wildly and finally patch locally

        -RTL_W16(IntrMitigate, 0x5151);
        +RTL_W16(IntrMitigate, 0x5100);

guessing whether it is right or not and also having to care to deploy
the patch everywhere it needs to be used, etc...).

To do so I've took original Francois's patch from 2012 and reworked it a bit:

- updated to latest net-next.git;
- adjusted scaling setup based on feedback from Hayes to pick up scaling
  vector depending not only on link speed but also on CPlusCmd[0:1] and to
  adjust CPlusCmd[0:1] correspondingly when setting timings;
- improved a bit (I think so) error handling.

I've tested the patch on "RTL8168d/8111d" (XID 083000c0) and with it and
`ethtool -C rx-usecs 0 rx-frames 0` on both ends it improves:

- minimum RTT latency:

        ~270μs ->  ~30μs (small packet),
        ~330μs -> ~110μs (full 1.5K ethernet frame)

- average RTT latency:

        ~480μs ->  ~50μs (small packet),
        ~560μs -> ~125μs (full 1.5K ethernet frame)

( before:

        root@neo1:# ping -i 0 -w 3 -s 82 -q neo2
        PING neo2.kirr.nexedi.com (192.168.102.21) 82(110) bytes of data.

        --- neo2.kirr.nexedi.com ping statistics ---
        5906 packets transmitted, 5905 received, 0% packet loss, time 2999ms
        rtt min/avg/max/mdev = 0.274/0.485/0.607/0.026 ms, ipg/ewma 0.508/0.489 ms

        root@neo1:# ping -i 0 -w 3 -s 1472 -q neo2
        PING neo2.kirr.nexedi.com (192.168.102.21) 1472(1500) bytes of data.

        --- neo2.kirr.nexedi.com ping statistics ---
        5073 packets transmitted, 5073 received, 0% packet loss, time 2999ms
        rtt min/avg/max/mdev = 0.330/0.566/0.710/0.028 ms, ipg/ewma 0.591/0.544 ms

  after:

        root@neo1# ping -i 0 -w 3 -s 82 -q neo2
        PING neo2.kirr.nexedi.com (192.168.102.21) 82(110) bytes of data.

        --- neo2.kirr.nexedi.com ping statistics ---
        45815 packets transmitted, 45815 received, 0% packet loss, time 3000ms
        rtt min/avg/max/mdev = 0.036/0.051/0.368/0.010 ms, ipg/ewma 0.065/0.053 ms

        root@neo1:# ping -i 0 -w 3 -s 1472 -q neo2
        PING neo2.kirr.nexedi.com (192.168.102.21) 1472(1500) bytes of data.

        --- neo2.kirr.nexedi.com ping statistics ---
        21250 packets transmitted, 21250 received, 0% packet loss, time 3000ms
        rtt min/avg/max/mdev = 0.112/0.125/0.390/0.007 ms, ipg/ewma 0.141/0.125 ms

  the small -> 1.5K latency growth is understandable as it takes ~15μs
  to transmit 1.5K on 1Gbps on the wire and with 2 hosts and 1 switch
  and ICMP ECHO + ECHO reply the packet has to travel 4 ethernet
  segments which is already 60μs;

  probably something a bit else is also there as e.g. on Linux, even
  with `cpupower frequency-set -g performance`, on some computers I've
  noticed the kernel can be spending more time in software-only mode
  when incoming packets go in less frequently. E.g. this program can
  demonstrate the effect for ICMP ECHO processing:

  https://lab.nexedi.com/kirr/bcc/blob/43cfc13b/tools/pinglat.py

  (later this was found to be partly due to C-states exit latencies) )

We have this patch running in our testing setup for 1 months already
without any issues observed.

It remains to be clarified whether RX and TX timers use the same base.
For now I've set them equally, but Francois's original patch version
suggests it could be not the same.

I've got no feedback at all to my original posting of this patch and questions

https://www.spinics.net/lists/netdev/msg457173.html

neither from Francois, nor from any people from Realtek during one month.

So I suggest we simply apply it to net-next.git now.

Cc: Francois Romieu <romieu@fr.zoreil.com>
Cc: Hayes Wang <hayeswang@realtek.com>
Cc: Realtek linux nic maintainers <nic_swsd@realtek.com>
Cc: David Laight <David.Laight@ACULAB.COM>
Cc: Stéphane ANCELOT <sancelot@free.fr>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Kirill Smelkov <kirr@nexedi.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: David S. Miller <davem@davemloft.net>