platform/kernel/linux-starfive.git
4 years agosctp: Add missing annotation for sctp_err_finish()
Jules Irenge [Sun, 23 Feb 2020 23:16:45 +0000 (23:16 +0000)]
sctp: Add missing annotation for sctp_err_finish()

Sparse reports a warning at sctp_err_finish()
warning: context imbalance in sctp_err_finish() - unexpected unlock

The root cause is a missing annotation at sctp_err_finish()
Add the missing  __releases(&((__sk)->sk_lock.slock)) annotation

Signed-off-by: Jules Irenge <jbi.octave@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoip6mr: Fix RCU list debugging warning
Amol Grover [Sat, 22 Feb 2020 16:57:27 +0000 (22:27 +0530)]
ip6mr: Fix RCU list debugging warning

ip6mr_for_each_table() macro uses list_for_each_entry_rcu()
for traversing outside an RCU read side critical section
but under the protection of rtnl_mutex. Hence add the
corresponding lockdep expression to silence the following
false-positive warnings:

[    4.319479] =============================
[    4.319480] WARNING: suspicious RCU usage
[    4.319482] 5.5.4-stable #17 Tainted: G            E
[    4.319483] -----------------------------
[    4.319485] net/ipv6/ip6mr.c:1243 RCU-list traversed in non-reader section!!

[    4.456831] =============================
[    4.456832] WARNING: suspicious RCU usage
[    4.456834] 5.5.4-stable #17 Tainted: G            E
[    4.456835] -----------------------------
[    4.456837] net/ipv6/ip6mr.c:1582 RCU-list traversed in non-reader section!!

Signed-off-by: Amol Grover <frextrite@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoptp: Add a ptp clock driver for IDT 82P33 SMU.
Min Li [Fri, 21 Feb 2020 20:03:51 +0000 (15:03 -0500)]
ptp: Add a ptp clock driver for IDT 82P33 SMU.

The IDT 82P33 Synchronization Management Unit (SMU) family provides
tools to manage timing references, clock sources and
timing paths for IEEE 1588 / Precision Time Protocol (PTP) and
Synchronous Ethernet (SyncE) based clocks. The device supports up
to three independent timing paths that control: PTP clock synthesis;
SyncE clock generation; and general purpose frequency translation.
The device supports physical layer timing with Digital PLLs (DPLLs)
and it supports packet based timing with Digitally Controlled
Oscillators (DCOs). This patch adds support for ptp clock based on
the device.

Changes since v1:
 - As suggested by Richard Cochran:
   1. Replace _mask_bit_count with the existing hweight8
   2. Prefix all functions with idt82p33
   3. Fix white space issues in Kconfig and Makefile
   4. Remove forward declaration
   5. Use adjfine instead of adjfreq for better resolution

 - As suggested by David Miller:
   1. Replace CHAN_INIT macro with a static function
      idt82p33_channel_init
   2. Employ reverse christmas tree ordering for local
      variables
   3. Fix indentation problem by appropriate number of
      TAB then SPACE character

Signed-off-by: Min Li <min.li.xe@renesas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodt-bindings: ptp: Add device tree binding for IDT 82P33 based PTP clock
Min Li [Fri, 21 Feb 2020 21:48:38 +0000 (16:48 -0500)]
dt-bindings: ptp: Add device tree binding for IDT 82P33 based PTP clock

Add device tree binding doc for the PTP clock based on IDT 82P33
Synchronization Management Unit (SMU).

Changes since v1:
 - As suggested by Rob Herring:
   1. Drop reg description for i2c
   2. Replace i2c@1 with i2c
   3. Add addtionalProperties: false

Signed-off-by: Min Li <min.li.xe@renesas.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: ipv4: Pass lockdep expression to RCU lists
Amol Grover [Fri, 21 Feb 2020 17:57:14 +0000 (23:27 +0530)]
tcp: ipv4: Pass lockdep expression to RCU lists

md5sig->head maybe traversed using hlist_for_each_entry_rcu
outside an RCU read-side critical section but under the protection
of socket lock.

Hence, add corresponding lockdep expression to silence false-positive
warnings, and harden RCU lists.

Signed-off-by: Amol Grover <frextrite@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: 802: psnap.c: Use built-in RCU list checking
Madhuparna Bhowmik [Fri, 21 Feb 2020 16:19:47 +0000 (21:49 +0530)]
net: 802: psnap.c: Use built-in RCU list checking

list_for_each_entry_rcu() has built-in RCU and lock checking.

Pass cond argument to list_for_each_entry_rcu() to silence
false lockdep warning when CONFIG_PROVE_RCU_LIST is enabled
by default.

Signed-off-by: Madhuparna Bhowmik <madhuparnabhowmik10@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp, ulp: Pass lockdep expression to RCU lists
Amol Grover [Fri, 21 Feb 2020 15:15:38 +0000 (20:45 +0530)]
tcp, ulp: Pass lockdep expression to RCU lists

tcp_ulp_list is traversed using list_for_each_entry_rcu
outside an RCU read-side critical section but under the protection
of tcp_ulp_list_lock.

Hence, add corresponding lockdep expression to silence false-positive
warnings, and harden RCU lists.t

Signed-off-by: Amol Grover <frextrite@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'mlxsw-Introduce-ACL-traps'
David S. Miller [Mon, 24 Feb 2020 19:55:07 +0000 (11:55 -0800)]
Merge branch 'mlxsw-Introduce-ACL-traps'

Jiri Pirko says:

====================
mlxsw: Introduce ACL traps

This patchset allows to track packets that are dropped in HW by ACL.

Unlike the existing mlxsw traps, ACL traps are "source traps".
That means the action is not controlled by HPKT register but directly
in ACL TRAP action. When devlink user changes action from drop to trap
and vice versa, it would be needed to go over all instances of ACL TRAP
action and do change. That does not scale. Instead, resolve this
by introducing "dummy" group with "thin" policer. The purpose of
this policer is to drop as many packets as possible. The ones
that pass through are going to be dropped in devlink code - patch #6
takes care of that.

First four patches are preparation for introduction of ACL traps in mlxsw
so it possible to easily change from drop to trap for source traps
as well - by changing group to "dummy" and back.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: devlink_trap_acl_drops: Add ACL traps test
Jiri Pirko [Mon, 24 Feb 2020 07:35:58 +0000 (08:35 +0100)]
selftests: devlink_trap_acl_drops: Add ACL traps test

Add a test to check functionality of ACL traps.

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>
4 years agoselftests: pass pref and handle to devlink_trap_drop_* helpers
Jiri Pirko [Mon, 24 Feb 2020 07:35:57 +0000 (08:35 +0100)]
selftests: pass pref and handle to devlink_trap_drop_* helpers

Currently the helpers assume pref 1 and handle 101. Make that explicit
and pass the values from callers.

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>
4 years agoselftests: introduce test for mlxsw tc flower restrictions
Jiri Pirko [Mon, 24 Feb 2020 07:35:56 +0000 (08:35 +0100)]
selftests: introduce test for mlxsw tc flower restrictions

Include test of forbidding to have drop rule on mixed-bound
shared block.

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>
4 years agomlxsw: spectrum_trap: Add ACL devlink-trap support
Jiri Pirko [Mon, 24 Feb 2020 07:35:55 +0000 (08:35 +0100)]
mlxsw: spectrum_trap: Add ACL devlink-trap support

Add the trap group used to report ACL drops. Setup the trap IDs for
ingress/egress flow action drop. Register the two packet traps
associated with ACL trap group with devlink during driver
initialization. As these are "source traps", set the disabled
trap group to be the dummy, discarding as many packets in HW
as possible.

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>
4 years agomlxsw: spectrum_trap: Introduce dummy group with thin policer
Jiri Pirko [Mon, 24 Feb 2020 07:35:54 +0000 (08:35 +0100)]
mlxsw: spectrum_trap: Introduce dummy group with thin policer

For "source traps" it is not possible to change HPKT action to discard.
But there is still need to disallow packets arriving to CPU as much as
possible. Handle this by introduction of a "dummy group". It has a
"thin" policer, which passes as less packets to CPU as possible. The
rest is going to be discarded there. The "dummy group" is to be used
later on by ACL trap (which is a "source trap").

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>
4 years agomlxsw: core: Extend MLXSW_RXL_DIS to register disabled trap group
Jiri Pirko [Mon, 24 Feb 2020 07:35:53 +0000 (08:35 +0100)]
mlxsw: core: Extend MLXSW_RXL_DIS to register disabled trap group

Extend the mlxsw_listener struct to contain trap group for disabled
traps too. Rename the original "trap_group" item to "en_trap_group" as
it represents enabled state. Let both groups be the same for MLXSW_RXL
however extend MLXSW_RXL_DIS to register separate groups for enable and
disable.

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>
4 years agomlxsw: core: Allow to enable/disable rx_listener for trap
Jiri Pirko [Mon, 24 Feb 2020 07:35:52 +0000 (08:35 +0100)]
mlxsw: core: Allow to enable/disable rx_listener for trap

For source traps, the "thin policer" is going to be used in order
to reduce the amount of trapped packets to minimum. However, there
will be still small number of packets coming in that need to be dropped
in the driver. Allow to enable/disable rx_listener related to specific
trap in order to prevent unwanted packets to go up the stack.

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>
4 years agomlxsw: acl_flex_actions: Trap all ACL dropped packets to DISCARD_*_ACL traps
Jiri Pirko [Mon, 24 Feb 2020 07:35:51 +0000 (08:35 +0100)]
mlxsw: acl_flex_actions: Trap all ACL dropped packets to DISCARD_*_ACL traps

Introduce a new set of traps:
DISCARD_INGRESS_ACL and DISCARD_EGRESS_ACL
Set the trap_action from NOP to TRAP which causes the packets dropped
by the TRAP action to be trapped under new trap IDs, depending on the
ingress/egress binding point.

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>
4 years agomlxsw: spectrum_acl: Pass the ingress indication down to flex action
Jiri Pirko [Mon, 24 Feb 2020 07:35:50 +0000 (08:35 +0100)]
mlxsw: spectrum_acl: Pass the ingress indication down to flex action

The ACL flex action will have to know if it is in ingress or egress,
so it can use correct trap ID. Pass the ingress indication down to it.

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>
4 years agomlxsw: spectrum_flower: Disable mixed bound blocks to contain action drop
Jiri Pirko [Mon, 24 Feb 2020 07:35:49 +0000 (08:35 +0100)]
mlxsw: spectrum_flower: Disable mixed bound blocks to contain action drop

Action drop is going to be tracked by two separate traps, one for
ingress and one for egress. Prepare for it and disallow the possibility
to have drop action in blocks which are bound to both ingress and
egress.

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>
4 years agomlxsw: spectrum_acl: Track ingress and egress block bindings
Jiri Pirko [Mon, 24 Feb 2020 07:35:48 +0000 (08:35 +0100)]
mlxsw: spectrum_acl: Track ingress and egress block bindings

Count the number of ingress and egress block bindings. Use the egress
counter in "is_egress_bound" helper. Add couple of helpers to check
ingress and mixed bound.

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>
4 years agodevlink: add ACL generic packet traps
Jiri Pirko [Mon, 24 Feb 2020 07:35:47 +0000 (08:35 +0100)]
devlink: add ACL generic packet traps

Add packet traps that can report packets that were dropped during ACL
processing.

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>
4 years agomlxsw: spectrum_trap: Prepare mlxsw_core_trap_action_set() to handle not only action
Jiri Pirko [Mon, 24 Feb 2020 07:35:46 +0000 (08:35 +0100)]
mlxsw: spectrum_trap: Prepare mlxsw_core_trap_action_set() to handle not only action

Rename function mlxsw_core_trap_action_set() to
mlxsw_core_trap_state_set() and pass bool enabled instead of action.
Figure out the action according to the enabled state there.

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>
4 years agomlxsw: spectrum_trap: Use listener->en/dis_action instead of hard-coded values
Jiri Pirko [Mon, 24 Feb 2020 07:35:45 +0000 (08:35 +0100)]
mlxsw: spectrum_trap: Use listener->en/dis_action instead of hard-coded values

The listener fields en_action and dis_action now contain the actions to
be used for TRAP and DROP devlink trap actions. Use them directly
instead of the hard-coded values.

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>
4 years agomlxsw: core: Allow to register disabled traps using MLXSW_RXL_DIS
Jiri Pirko [Mon, 24 Feb 2020 07:35:44 +0000 (08:35 +0100)]
mlxsw: core: Allow to register disabled traps using MLXSW_RXL_DIS

Introduce a new macro MLXSW_RXL_DIS that allows to register listeners
as disabled. That allows that from now on, the "action" can be
understood always as "enabled action" and "unreg_action" as "disabled
action". Rename them and treat them accordingly.

Use the new macro for defining drops in spectrum_trap.c.

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>
4 years agomlxsw: spectrum_trap: Set unreg_action to be SET_FW_DEFAULT
Jiri Pirko [Mon, 24 Feb 2020 07:35:43 +0000 (08:35 +0100)]
mlxsw: spectrum_trap: Set unreg_action to be SET_FW_DEFAULT

Currently it does not really matter if it is set to DISCARD or
SET_FW_DEFAULT because it is set only during unregister of the listener.
The unreg_action is going to be used for disabling the listener too, so
change to SET_FW_DEFAULT and ensure the HW is going to behave correctly.

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>
4 years agoMerge branch 'net-Clean-driver-module-and-FW-versions'
David S. Miller [Mon, 24 Feb 2020 19:24:57 +0000 (11:24 -0800)]
Merge branch 'net-Clean-driver-module-and-FW-versions'

Leon Romanovsky says:

====================
Clean driver, module and FW versions

Change log:
 v1:
  * Split all FW cleanups patches to separate patches
  * Fixed commit message
  * Deleted odd DRV_RELDATE
  * Added everyone from MAINTAINERS file
 v0: https://lore.kernel.org/netdev/20200220145855.255704-1-leon@kernel.org

This is first patchset to netdev (already sent RDMA [1] and arch/um [2])
in attempt to unify the version management for in-tree kernel code.
The patches follow already accepted ethtool change [3] to set as
a default linux kernel version.

It allows us to remove driver version and present to the users unified
picture of driver version, which is similar to default MODULE_VERSION().

As part of this series, I deleted various creative attempts to mark
absence of FW. There is no need to set "N/A" in ethtool ->fw_version
field and it is enough to do not set it.

1.
The code is compile tested and passes 0-day kbuild.
2.
The proposed changes are based on commit:
  2bb07f4e1d86 ("tc-testing: updated tdc tests for basic filter")
3.
WIP branch is [4].

[1] https://lore.kernel.org/linux-rdma/20200220071239.231800-1-leon@kernel.org/
[2] http://lists.infradead.org/pipermail/linux-um/2020-February/002913.html
[3] https://lore.kernel.org/linux-rdma/20200127072028.19123-1-leon@kernel.org/
[4] https://git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git/log/?h=ethtool
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/atheros: Clean atheros code from driver version
Leon Romanovsky [Mon, 24 Feb 2020 08:53:11 +0000 (10:53 +0200)]
net/atheros: Clean atheros code from driver version

Use linux kernel version for ethtool and module versions.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Chris Snook <chris.snook@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/arc: Delete driver version
Leon Romanovsky [Mon, 24 Feb 2020 08:53:10 +0000 (10:53 +0200)]
net/arc: Delete driver version

Drop constant driver version in favour of global linux kernel.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/aquantia: Delete module version
Leon Romanovsky [Mon, 24 Feb 2020 08:53:09 +0000 (10:53 +0200)]
net/aquantia: Delete module version

There is no need to keep module and driver versions in in-tree
kernel code.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/apm: Properly mark absence of FW
Leon Romanovsky [Mon, 24 Feb 2020 08:53:08 +0000 (10:53 +0200)]
net/apm: Properly mark absence of FW

There is no need to set "N/A" if FW is not available.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/apm: Remove useless driver version
Leon Romanovsky [Mon, 24 Feb 2020 08:53:07 +0000 (10:53 +0200)]
net/apm: Remove useless driver version

Delete module version in favor of global and unique linux kernel.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/amd: Remove useless driver version
Leon Romanovsky [Mon, 24 Feb 2020 08:53:06 +0000 (10:53 +0200)]
net/amd: Remove useless driver version

Convert AMD drivers to respect universal linux kernel version.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/amazon: Ensure that driver version is aligned to the linux kernel
Leon Romanovsky [Mon, 24 Feb 2020 08:53:05 +0000 (10:53 +0200)]
net/amazon: Ensure that driver version is aligned to the linux kernel

Upstream drivers are managed inside global repository and released all
together, this ensure that driver version is the same as linux kernel,
so update amazon drivers to properly reflect it.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/althera: Delete hardcoded driver version
Leon Romanovsky [Mon, 24 Feb 2020 08:53:04 +0000 (10:53 +0200)]
net/althera: Delete hardcoded driver version

Convert to use default version provided by ethtool.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/alteon: Properly report FW version
Leon Romanovsky [Mon, 24 Feb 2020 08:53:03 +0000 (10:53 +0200)]
net/alteon: Properly report FW version

The acenic driver assigns FW version in driver version field,
as part of cleanup driver version, set FW version properly.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/allwinner: Remove driver version
Leon Romanovsky [Mon, 24 Feb 2020 08:53:02 +0000 (10:53 +0200)]
net/allwinner: Remove driver version

There is no need in custom driver version for in-tree code.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Maxime Ripard <mripard@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/alacritech: Delete driver version
Leon Romanovsky [Mon, 24 Feb 2020 08:53:01 +0000 (10:53 +0200)]
net/alacritech: Delete driver version

Use standard variant of the driver version.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/agere: Delete unneeded driver version
Leon Romanovsky [Mon, 24 Feb 2020 08:53:00 +0000 (10:53 +0200)]
net/agere: Delete unneeded driver version

There is no need in driver version for in-tree kernel code.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/aeroflex: Don't assign FW if it is not available
Leon Romanovsky [Mon, 24 Feb 2020 08:52:59 +0000 (10:52 +0200)]
net/aeroflex: Don't assign FW if it is not available

If FW version is not available, it is enough to leave that field as
empty, there is no need to write N/A.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/aeroflex: Clean ethtool_info struct assignments
Leon Romanovsky [Mon, 24 Feb 2020 08:52:58 +0000 (10:52 +0200)]
net/aeroflex: Clean ethtool_info struct assignments

If FW version is not available, it is enough to leave that field as
empty, there is no need to write N/A.

The driver version is replaced in favor of generally available
in-tree variant.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/adaptec: Clean driver versions
Leon Romanovsky [Mon, 24 Feb 2020 08:52:57 +0000 (10:52 +0200)]
net/adaptec: Clean driver versions

Delete useless driver version in favor of default ones.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/3com: Delete driver and module versions from 3com drivers
Leon Romanovsky [Mon, 24 Feb 2020 08:52:56 +0000 (10:52 +0200)]
net/3com: Delete driver and module versions from 3com drivers

There is no need to mislead users by providing different versions for
driver, ethtool and modules. Delete driver assignments and let use
the default one.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/dummy: Ditch driver and module versions
Leon Romanovsky [Mon, 24 Feb 2020 08:52:55 +0000 (10:52 +0200)]
net/dummy: Ditch driver and module versions

Delete constant driver and module versions in favor of
standard global version which is unique to whole kernel.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/bond: Delete driver and module versions
Leon Romanovsky [Mon, 24 Feb 2020 08:52:54 +0000 (10:52 +0200)]
net/bond: Delete driver and module versions

The in-kernel code has already unique version, which is based
on Linus's tag, update the bond driver to be consistent with that
version.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: remove redundant initialization of pointer 'client'
Colin Ian King [Mon, 24 Feb 2020 19:20:22 +0000 (19:20 +0000)]
net: hns3: remove redundant initialization of pointer 'client'

The pointer 'client' is being initialized with a value that is never
read, it is being updated later on. The initialization is redundant
and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlxfw: fix spelling mistake: "progamming" -> "programming"
Colin Ian King [Mon, 24 Feb 2020 15:21:01 +0000 (15:21 +0000)]
net/mlxfw: fix spelling mistake: "progamming" -> "programming"

There is a spelling mistake in a literal string. Fix it.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: Remove unneeded export of a couple of xdp generic functions
David Ahern [Fri, 21 Feb 2020 05:03:13 +0000 (22:03 -0700)]
net: Remove unneeded export of a couple of xdp generic functions

generic_xdp_tx and xdp_do_generic_redirect are only used by builtin
code, so remove the EXPORT_SYMBOL_GPL for them.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoigmp: remove unused macro IGMP_Vx_UNSOLICITED_REPORT_INTERVAL
Li RongQing [Sun, 23 Feb 2020 08:05:43 +0000 (16:05 +0800)]
igmp: remove unused macro IGMP_Vx_UNSOLICITED_REPORT_INTERVAL

After commit 2690048c01f3 ("net: igmp: Allow user-space
configuration of igmp unsolicited report interval"), they
are not used now

Signed-off-by: Li RongQing <lirongqing@baidu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotc-testing: updated tdc tests for basic filter with u32 extended match rules
Roman Mashak [Sun, 23 Feb 2020 19:12:36 +0000 (14:12 -0500)]
tc-testing: updated tdc tests for basic filter with u32 extended match rules

Signed-off-by: Roman Mashak <mrv@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agor8169: improve rtl8169_start_xmit
Heiner Kallweit [Sat, 22 Feb 2020 16:02:51 +0000 (17:02 +0100)]
r8169: improve rtl8169_start_xmit

Only call rtl8169_xmit_frags() if the skb is actually fragmented.
This avoid a small overhead for non-fragmented skb's, and it allows
to simplify rtl8169_xmit_frags() a little.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ena: ethtool: remove redundant non-zero check on rc
Colin Ian King [Fri, 21 Feb 2020 23:26:53 +0000 (23:26 +0000)]
net: ena: ethtool: remove redundant non-zero check on rc

The non-zero check on rc is redundant as a previous non-zero
check on rc will always return and the second check is never
reached, hence it is redundant and can be removed.  Also
remove a blank line.

Addresses-Coverity: ("Logically dead code")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotun: Remove unnecessary BUG_ON check in tun_net_xmit
David Ahern [Fri, 21 Feb 2020 19:53:09 +0000 (12:53 -0700)]
tun: Remove unnecessary BUG_ON check in tun_net_xmit

The BUG_ON for NULL tfile is now redundant due to a recently
added null check after the rcu_dereference. Remove the BUG_ON.

Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'mlxsw-Cosmetic-fixes'
David S. Miller [Mon, 24 Feb 2020 00:07:31 +0000 (16:07 -0800)]
Merge branch 'mlxsw-Cosmetic-fixes'

Jiri Pirko says:

====================
mlxsw: Cosmetic fixes

This is a set of mainly action/trap related cosmetic fixes.
No functional changes.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: pci: Remove unused values
Ido Schimmel [Sun, 23 Feb 2020 07:31:44 +0000 (08:31 +0100)]
mlxsw: pci: Remove unused values

Since commit f3a52c6162f8 ("mlxsw: pci: Utilize MRSR register to perform
FW reset") the driver no longer issues a reset via the PCI BAR, so the
offset of the reset bit is unused. Remove it.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: core: Remove priv from listener equality comparison
Jiri Pirko [Sun, 23 Feb 2020 07:31:43 +0000 (08:31 +0100)]
mlxsw: core: Remove priv from listener equality comparison

During packet receive, only the first matching RX listener
in rx_listener_list is going to get the packet. So there is no
meaning in registering two equal listeners with different privs.
Remove priv from equality comparison and disable possibility
of doing it.

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>
4 years agomlxsw: spectrum_acl: Make block arg const where appropriate
Jiri Pirko [Sun, 23 Feb 2020 07:31:42 +0000 (08:31 +0100)]
mlxsw: spectrum_acl: Make block arg const where appropriate

There are couple of places where block pointer as a function argument
can be const. So make those const.

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>
4 years agomlxsw: spectrum_trap: Make global arrays const as they should be
Jiri Pirko [Sun, 23 Feb 2020 07:31:41 +0000 (08:31 +0100)]
mlxsw: spectrum_trap: Make global arrays const as they should be

The global arrays are treated as const, they should be const, so make
them const.

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>
4 years agomlxsw: core: Remove initialization to false of mlxsw_listener struct
Jiri Pirko [Sun, 23 Feb 2020 07:31:40 +0000 (08:31 +0100)]
mlxsw: core: Remove initialization to false of mlxsw_listener struct

No need to initialize to false, so remove it.

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>
4 years agomlxsw: core: Convert is_event and is_ctrl bools to be single bits
Jiri Pirko [Sun, 23 Feb 2020 07:31:39 +0000 (08:31 +0100)]
mlxsw: core: Convert is_event and is_ctrl bools to be single bits

No need for these two flags to be bool. Covert them to bits of u8.

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>
4 years agomlxsw: core: Remove dummy union name from struct mlxsw_listener
Jiri Pirko [Sun, 23 Feb 2020 07:31:38 +0000 (08:31 +0100)]
mlxsw: core: Remove dummy union name from struct mlxsw_listener

The dummy name for union is no longer needed, remove it.

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>
4 years agomlxsw: core: Remove unused action field from mlxsw_rx_listener struct
Jiri Pirko [Sun, 23 Feb 2020 07:31:37 +0000 (08:31 +0100)]
mlxsw: core: Remove unused action field from mlxsw_rx_listener struct

Commit 0791051c43ef ("mlxsw: core: Create a generic function to
register / unregister traps") moved this field to struct mlxsw_listener,
but forgot to remove it from struct mlxsw_rx_listener.
Remove the unused field.

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>
4 years agomlxsw: spectrum_trap: Move policer initialization to mlxsw_sp_trap_init()
Jiri Pirko [Sun, 23 Feb 2020 07:31:36 +0000 (08:31 +0100)]
mlxsw: spectrum_trap: Move policer initialization to mlxsw_sp_trap_init()

No need to initialize a single policer multiple times for each group.
So move the initialization to be done from mlxsw_sp_trap_init(), making
the function much simpler. Also, rename it so it is with sync with
spectrum.c policers initialization.

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>
4 years agomlxsw: core_acl_flex_actions: Rename Trap / Discard Action to Trap Action
Jiri Pirko [Sun, 23 Feb 2020 07:31:35 +0000 (08:31 +0100)]
mlxsw: core_acl_flex_actions: Rename Trap / Discard Action to Trap Action

The Trap / Discard Action action got renamed in PRM, so rename it in the
code as well.

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>
4 years agomlxsw: spectrum_trap: Move functions to avoid their forward declarations
Jiri Pirko [Sun, 23 Feb 2020 07:31:34 +0000 (08:31 +0100)]
mlxsw: spectrum_trap: Move functions to avoid their forward declarations

No need to have forward declarations for mlxsw_sp_rx_drop_listener()
and mlxsw_sp_rx_exception_listener(). Just move them up and avoid it.

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>
4 years agomlxsw: spectrum_trap: Use err variable instead of directly checking func return value
Jiri Pirko [Sun, 23 Feb 2020 07:31:33 +0000 (08:31 +0100)]
mlxsw: spectrum_trap: Use err variable instead of directly checking func return value

When calling mlxsw_sp_rx_listener(), use err variable instead of directly
checking func return value.

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>
4 years agor8169: remove RTL_EVENT_NAPI constants
Heiner Kallweit [Fri, 21 Feb 2020 18:27:00 +0000 (19:27 +0100)]
r8169: remove RTL_EVENT_NAPI constants

These constants are used in one place only, so we can remove them and
use the values directly.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'mlxsw-Remove-RTNL-from-route-insertion-path'
David S. Miller [Sun, 23 Feb 2020 05:24:51 +0000 (21:24 -0800)]
Merge branch 'mlxsw-Remove-RTNL-from-route-insertion-path'

Ido Schimmel says:

====================
mlxsw: Remove RTNL from route insertion path

This patch set removes RTNL from the route insertion path in mlxsw in
order to reduce the control plane latency: the time it takes to push
routes from user space to the kernel and mlxsw.

Up until now mlxsw did not have a lock to protect its shared router data
structures and instead relied on RTNL. While this was simple and worked,
it resulted in large control plane latencies as RTNL was heavily
contended - by both the task receiving the netlink messages from user
space and the mlxsw workqueue that programs the routes to the device.

By removing RTNL and introducing a new router mutex, this patch set
reduces the control plane latency by ~80%. A single mutex is added as
inside mlxsw there is not a lot of concurrency. In addition, a more
fine-grained locking scheme is much more error-prone.

Patches #1-#6 are preparations. They add needed locking in NVE and
multicast routing code instead of relying on RTNL
Patch #7 introduces the new mutex
Patches #8-#12 gradually take the lock in various entry points into the
routing code
Patch #13 removes RTNL in places where it is no longer required
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Remove RTNL where possible
Ido Schimmel [Fri, 21 Feb 2020 17:54:15 +0000 (19:54 +0200)]
mlxsw: spectrum: Remove RTNL where possible

After introducing the router lock in previous patches and making sure it
protects internal router structures, we no longer need to rely on RTNL
to serialize access to these structures.

Remove RTNL from call sites that no longer require it.

Two calls sites that keep taking the lock are
mlxsw_sp_router_fibmr_event_work() and mlxsw_sp_inet6addr_event_work().
The first calls into ACL code that still assumes RTNL is taken. The
second potentially calls into the FID code that also relies on RTNL.
Removing RTNL from these two call sites is the subject of future work.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_router: Take router lock from exported helpers
Ido Schimmel [Fri, 21 Feb 2020 17:54:14 +0000 (19:54 +0200)]
mlxsw: spectrum_router: Take router lock from exported helpers

The routing code exports some helper functions that can be called from
other driver modules such as the bridge. These helpers are never called
with the router lock already held and therefore need to take it in order
to serialize access to shared router structures.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_router: Take router lock from inetaddr listeners
Ido Schimmel [Fri, 21 Feb 2020 17:54:13 +0000 (19:54 +0200)]
mlxsw: spectrum_router: Take router lock from inetaddr listeners

Another entry point into the routing code is from inetaddr listeners.
The driver registers listeners to IPv4 and IPv6 inetaddr notification
chains in order to understand when a RIF needs to be created or
destroyed.

Serialize access to shared router structures from these listeners by
taking the router lock when processing inetaddr events.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_router: Take router lock from netdev listener
Ido Schimmel [Fri, 21 Feb 2020 17:54:12 +0000 (19:54 +0200)]
mlxsw: spectrum_router: Take router lock from netdev listener

One entry point into the routing code is from the netdev listener block.
Some netdev events require access to internal router structures. For
example, changing the MTU of a netdev requires looking-up the backing
RIF and adjusting its MTU.

In order to serialize access to shared router structures, take the
router lock when processing netdev events that require access to it.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_dpipe: Take router lock from dpipe code
Ido Schimmel [Fri, 21 Feb 2020 17:54:11 +0000 (19:54 +0200)]
mlxsw: spectrum_dpipe: Take router lock from dpipe code

The dpipe code traverses internal router structures such as neighbours
and adjacency entries and dumps them to user space via netlink. Up until
now the routing code did not have its own locks and relied on RTNL lock
to serialize access. This is going to change with the introduction of
the router lock.

Take the router lock in the code paths where RTNL lock is currently
taken so that the latter could be removed by subsequent patches.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_router: Take router lock from inside routing code
Ido Schimmel [Fri, 21 Feb 2020 17:54:10 +0000 (19:54 +0200)]
mlxsw: spectrum_router: Take router lock from inside routing code

There are several work items in the routing code that currently rely on
RTNL lock to guard against concurrent changes. Have these work items
acquire the router lock in preparation for the removal for RTNL lock
from the routing code.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_router: Introduce router lock
Ido Schimmel [Fri, 21 Feb 2020 17:54:09 +0000 (19:54 +0200)]
mlxsw: spectrum_router: Introduce router lock

Introduce a mutex to protect the internal structure of the routing code.
A single lock is added instead of a more fine-grained and complicated
locking scheme because there is not a lot of concurrency in the routing
code.

The main motivation is remove the dependence on RTNL lock, which is
currently used by both the process pushing routes to the kernel and the
workqueue pushing the routes to the underlying device.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_router: Store NVE decapsulation configuration in router
Ido Schimmel [Fri, 21 Feb 2020 17:54:08 +0000 (19:54 +0200)]
mlxsw: spectrum_router: Store NVE decapsulation configuration in router

When a host route is added, the driver checks if the route needs to be
promoted to perform NVE decapsulation based on the current NVE
configuration. If so, the index of the decapsulation entry is retrieved
and associated with the route.

Currently, this information is stored in the NVE module which the router
module consults. Since the information is protected under RTNL and since
route insertion happens with RTNL held, there is no problem to retrieve
the information from the NVE module.

However, this is going to change and route insertion will no longer
happen under RTNL. Instead, a dedicated lock will be introduced for the
router module.

Therefore, store this information in the router module and change the
router module to consult this copy.

The validity of the information is set / cleared whenever an NVE tunnel
is initialized / de-initialized. When this happens the NVE module calls
into the router module to promote / demote the relevant host route.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_router: Expose router struct to internal users
Ido Schimmel [Fri, 21 Feb 2020 17:54:07 +0000 (19:54 +0200)]
mlxsw: spectrum_router: Expose router struct to internal users

The dpipe code accesses internal router data structures and acquires
RTNL to protect against their changes. Subsequent patches will remove
reliance on RTNL and introduce a dedicated lock to protect router data
structures.

Publish the router struct to internal users such as the dpipe, so that
they could acquire it instead of RTNL.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_mr: Protect multicast route list with a lock
Ido Schimmel [Fri, 21 Feb 2020 17:54:06 +0000 (19:54 +0200)]
mlxsw: spectrum_mr: Protect multicast route list with a lock

Protect the per-table multicast route list with a lock and remove RTNL
from the delayed work that periodically updates the kernel about packets
and bytes statistics from each multicast route.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_mr: Protect multicast table list with a lock
Ido Schimmel [Fri, 21 Feb 2020 17:54:05 +0000 (19:54 +0200)]
mlxsw: spectrum_mr: Protect multicast table list with a lock

The multicast table list is traversed from a delayed work that
periodically updates the kernel about packets and bytes statistics from
each multicast route.

The list is currently protected by RTNL, but subsequent patches will
remove the driver's dependence on this contended lock.

In order to be able to remove dependence on RTNL in the next patch,
guard this list with a dedicated mutex.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_mr: Publish multicast route after writing it to the device
Ido Schimmel [Fri, 21 Feb 2020 17:54:04 +0000 (19:54 +0200)]
mlxsw: spectrum_mr: Publish multicast route after writing it to the device

The driver periodically traverses the linked list of multicast routes
and updates the kernel about packets and bytes statistics from each
multicast route. These statistics are read from a counter associated
with the route when it is written to the device.

Currently, multicast routes are published via this linked list before
they are associated with a counter. Despite that, it is not possible for
the driver to access an invalid counter because the delayed work that
reads the statistics and multicast route addition / deletion are
mutually exclusive using RTNL.

In order to be able to remove RTNL, the list needs to be protected by a
dedicated lock, but any route published via the list must have an
associated counter, otherwise the driver will access an invalid counter.

Solve this by re-ordering the operations during multicast route addition
so that the route is only added to the linked list after it was written
to the device. Similarly, during deletion the route is first removed
from the linked list before its deletion from the device.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'mlxfw-Improve-error-reporting-and-FW-reactivate-support'
David S. Miller [Fri, 21 Feb 2020 23:41:10 +0000 (15:41 -0800)]
Merge branch 'mlxfw-Improve-error-reporting-and-FW-reactivate-support'

Saeed Mahameed says:

====================
mlxfw: Improve error reporting and FW reactivate support

This patchset improves mlxfw error reporting to netlink and to
kernel log.

V2:
 - Use proper err codes, EBUSY/EIO instead of EALREADY/EREMOTEIO
 - Fix typo.

From Eran and me.

1) patch #1, Make mlxfw/mlxsw fw flash devlink status notify generic,
   and enable it for mlx5.

2) patches #2..#5 are improving mlxfw flash error messages by
reporting detailed mlxfw FSM error messages to netlink and kernel log.

3) patches #6,7 From Eran: Add FW reactivate flow to  mlxfw and mlx5
====================

Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5: Add fsm_reactivate callback support
Eran Ben Elisha [Fri, 21 Feb 2020 21:46:12 +0000 (21:46 +0000)]
net/mlx5: Add fsm_reactivate callback support

Add support for fsm reactivate via MIRC (Management Image Re-activation
Control) set and query commands.
For re-activation flow, driver shall first run MIRC set, and then wait
until FW is done (via querying MIRC status).

Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlxfw: Add reactivate flow support to FSM burn flow
Eran Ben Elisha [Fri, 21 Feb 2020 21:46:09 +0000 (21:46 +0000)]
net/mlxfw: Add reactivate flow support to FSM burn flow

Expose fsm_reactivate callback to the mlxfw_dev_ops struct. FSM reactivate
is needed before flashing the new image in order to flush the old flashed
but not running firmware image.

In case mlxfw_dev do not support the reactivation, this step will be
skipped. But if later image flash will fail, a hint will be provided by
the extack to advise the user that the failure might be related to it.

Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlxfw: Use MLXFW_ERR_MSG macro for error reporting
Saeed Mahameed [Fri, 21 Feb 2020 21:46:07 +0000 (21:46 +0000)]
net/mlxfw: Use MLXFW_ERR_MSG macro for error reporting

Instead of always calling both mlxfw_err and NL_SET_ERR_MSG_MOD with the
same message, use the dedicated macro instead.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlxfw: Convert pr_* to dev_* in mlxfw_fsm.c
Saeed Mahameed [Fri, 21 Feb 2020 21:46:05 +0000 (21:46 +0000)]
net/mlxfw: Convert pr_* to dev_* in mlxfw_fsm.c

Introduce mlxfw_{info, err, dbg} macros and make them call corresponding
dev_* macros, then convert all instances of pr_* to mlxfw_*.

This will allow printing the device name mlxfw is operating on.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlxfw: More error messages coverage
Saeed Mahameed [Fri, 21 Feb 2020 21:46:03 +0000 (21:46 +0000)]
net/mlxfw: More error messages coverage

Make sure mlxfw_firmware_flash reports a detailed user readable error
message in every possible error path, basically every time
mlxfw_dev->ops->*() is called and an error is returned, or when image
initialization is failed.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlxfw: Improve FSM err message reporting and return codes
Saeed Mahameed [Fri, 21 Feb 2020 21:45:59 +0000 (21:45 +0000)]
net/mlxfw: Improve FSM err message reporting and return codes

Report unique and standard error codes corresponding to the specific
FW flash error. In addition, add a more detailed error messages to
netlink.

Before:
$ devlink dev flash pci/0000:05:00.0 file ...
Error: mlxfw: Firmware flash failed.
devlink answers: Invalid argument

After:
$ devlink dev flash pci/0000:05:00.0 file ...
Error: mlxfw: Firmware flash failed: pending reset.
devlink answers: Device busy

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlxfw: Generic mlx FW flash status notify
Saeed Mahameed [Fri, 21 Feb 2020 21:45:58 +0000 (21:45 +0000)]
net/mlxfw: Generic mlx FW flash status notify

FW flash status notify is currently implemented via a callback to the
caller mlx module, and all it is doing is to call
devlink_flash_update_status_notify with the specific module devlink
instance.

Instead of repeating the whole process for all mlx modules and
re-implement the status_notify callback again and again. Just provide the
devlink instance as part of mlxfw_dev when calling mlxfw_firmware_flash
and let mlxfw do the devlink status updates directly.

This will be very useful for adding status notify support to mlx5, as
already done in this patch, with a simple one line of just providing the
devlink instance to mlxfw_firmware_flash.

mlxfw now depends on NET_DEVLINK as all other mlx modules.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
David S. Miller [Fri, 21 Feb 2020 23:22:45 +0000 (15:22 -0800)]
Merge git://git./linux/kernel/git/bpf/bpf-next

Daniel Borkmann says:

====================
pull-request: bpf-next 2020-02-21

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

We've added 25 non-merge commits during the last 4 day(s) which contain
a total of 33 files changed, 2433 insertions(+), 161 deletions(-).

The main changes are:

1) Allow for adding TCP listen sockets into sock_map/hash so they can be used
   with reuseport BPF programs, from Jakub Sitnicki.

2) Add a new bpf_program__set_attach_target() helper for adding libbpf support
   to specify the tracepoint/function dynamically, from Eelco Chaudron.

3) Add bpf_read_branch_records() BPF helper which helps use cases like profile
   guided optimizations, from Daniel Xu.

4) Enable bpf_perf_event_read_value() in all tracing programs, from Song Liu.

5) Relax BTF mandatory check if only used for libbpf itself e.g. to process
   BTF defined maps, from Andrii Nakryiko.

6) Move BPF selftests -mcpu compilation attribute from 'probe' to 'v3' as it has
   been observed that former fails in envs with low memlock, from Yonghong Song.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
David S. Miller [Fri, 21 Feb 2020 21:39:34 +0000 (13:39 -0800)]
Merge git://git./linux/kernel/git/netdev/net

Conflict resolution of ice_virtchnl_pf.c based upon work by
Stephen Rothwell.

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'bpf-sockmap-listen'
Daniel Borkmann [Fri, 21 Feb 2020 21:29:46 +0000 (22:29 +0100)]
Merge branch 'bpf-sockmap-listen'

Jakub Sitnicki says:

====================
This patch set turns SOCK{MAP,HASH} into generic collections for TCP
sockets, both listening and established. Adding support for listening
sockets enables us to use these BPF map types with reuseport BPF programs.

Why? SOCKMAP and SOCKHASH, in comparison to REUSEPORT_SOCKARRAY, allow
the socket to be in more than one map at the same time.

Having a BPF map type that can hold listening sockets, and gracefully
co-exist with reuseport BPF is important if, in the future, we want
BPF programs that run at socket lookup time [0]. Cover letter for v1 of
this series tells the full story of how we got here [1].

Although SOCK{MAP,HASH} are not a drop-in replacement for SOCKARRAY just
yet, because UDP support is lacking, it's a step in this direction. We're
working with Lorenz on extending SOCK{MAP,HASH} to hold UDP sockets, and
expect to post RFC series for sockmap + UDP in the near future.

I've dropped Acks from all patches that have been touched since v6.

The audit for missing READ_ONCE annotations for access to sk_prot is
ongoing. Thus far I've found one location specific to TCP listening sockets
that needed annotating. This got fixed it in this iteration. I wonder if
sparse checker could be put to work to identify places where we have
sk_prot access while not holding sk_lock...

The patch series depends on another one, posted earlier [2], that has
been split out of it.

v6 -> v7:

- Extended the series to cover SOCKHASH. (patches 4-8, 10-11) (John)

- Rebased onto recent bpf-next. Resolved conflicts in recent fixes to
  sk_state checks on sockmap/sockhash update path. (patch 4)

- Added missing READ_ONCE annotation in sock_copy. (patch 1)

- Split out patches that simplify sk_psock_restore_proto [2].

v5 -> v6:

- Added a fix-up for patch 1 which I forgot to commit in v5. Sigh.

v4 -> v5:

- Rebase onto recent bpf-next to resolve conflicts. (Daniel)

v3 -> v4:

- Make tcp_bpf_clone parameter names consistent across function declaration
  and definition. (Martin)

- Use sock_map_redirect_okay helper everywhere we need to take a different
  action for listening sockets. (Lorenz)

- Expand comment explaining the need for a callback from reuseport to
  sockarray code in reuseport_detach_sock. (Martin)

- Mention the possibility of using a u64 counter for reuseport IDs in the
  future in the description for patch 10. (Martin)

v2 -> v3:

- Generate reuseport ID when group is created. Please see patch 10
  description for details. (Martin)

- Fix the build when CONFIG_NET_SOCK_MSG is not selected by either
  CONFIG_BPF_STREAM_PARSER or CONFIG_TLS. (kbuild bot & John)

- Allow updating sockmap from BPF on BPF_SOCK_OPS_TCP_LISTEN_CB callback. An
  oversight in previous iterations. Users may want to populate the sockmap with
  listening sockets from BPF as well.

- Removed RCU read lock assertion in sock_map_lookup_sys. (Martin)

- Get rid of a warning when child socket was cloned with parent's psock
  state. (John)

- Check for tcp_bpf_unhash rather than tcp_bpf_recvmsg when deciding if
  sk_proto needs restoring on clone. Check for recvmsg in the context of
  listening socket cloning was confusing. (Martin)

- Consolidate sock_map_sk_is_suitable with sock_map_update_okay. This led
  to adding dedicated predicates for sockhash. Update self-tests
  accordingly. (John)

- Annotate unlikely branch in bpf_{sk,msg}_redirect_map when socket isn't
  in a map, or isn't a valid redirect target. (John)

- Document paired READ/WRITE_ONCE annotations and cover shared access in
  more detail in patch 2 description. (John)

- Correct a couple of log messages in sockmap_listen self-tests so the
  message reflects the actual failure.

- Rework reuseport tests from sockmap_listen suite so that ENOENT error
  from bpf_sk_select_reuseport handler does not happen on happy path.

v1 -> v2:

- af_ops->syn_recv_sock callback is no longer overridden and burdened with
  restoring sk_prot and clearing sk_user_data in the child socket. As child
  socket is already hashed when syn_recv_sock returns, it is too late to
  put it in the right state. Instead patches 3 & 4 address restoring
  sk_prot and clearing sk_user_data before we hash the child socket.
  (Pointed out by Martin Lau)

- Annotate shared access to sk->sk_prot with READ_ONCE/WRITE_ONCE macros as
  we write to it from sk_msg while socket might be getting cloned on
  another CPU. (Suggested by John Fastabend)

- Convert tests for SOCKMAP holding listening sockets to return-on-error
  style, and hook them up to test_progs. Also use BPF skeleton for setup.
  Add new tests to cover the race scenario discovered during v1 review.

RFC -> v1:

- Switch from overriding proto->accept to af_ops->syn_recv_sock, which
  happens earlier. Clearing the psock state after accept() does not work
  for child sockets that become orphaned (never got accepted). v4-mapped
  sockets need special care.

- Return the socket cookie on SOCKMAP lookup from syscall to be on par with
  REUSEPORT_SOCKARRAY. Requires SOCKMAP to take u64 on lookup/update from
  syscall.

- Make bpf_sk_redirect_map (ingress) and bpf_msg_redirect_map (egress)
  SOCKMAP helpers fail when target socket is a listening one.

- Make bpf_sk_select_reuseport helper fail when target is a TCP established
  socket.

- Teach libbpf to recognize SK_REUSEPORT program type from section name.

- Add a dedicated set of tests for SOCKMAP holding listening sockets,
  covering map operations, overridden socket callbacks, and BPF helpers.

[0] https://lore.kernel.org/bpf/20190828072250.29828-1-jakub@cloudflare.com/
[1] https://lore.kernel.org/bpf/20191123110751.6729-1-jakub@cloudflare.com/
[2] https://lore.kernel.org/bpf/20200217121530.754315-1-jakub@cloudflare.com/
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
4 years agoselftests/bpf: Tests for sockmap/sockhash holding listening sockets
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:23 +0000 (17:10 +0000)]
selftests/bpf: Tests for sockmap/sockhash holding listening sockets

Now that SOCKMAP and SOCKHASH map types can store listening sockets,
user-space and BPF API is open to a new set of potential pitfalls.

Exercise the map operations, with extra attention to code paths susceptible
to races between map ops and socket cloning, and BPF helpers that work with
SOCKMAP/SOCKHASH to gain confidence that all works as expected.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200218171023.844439-12-jakub@cloudflare.com
4 years agoselftests/bpf: Extend SK_REUSEPORT tests to cover SOCKMAP/SOCKHASH
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:22 +0000 (17:10 +0000)]
selftests/bpf: Extend SK_REUSEPORT tests to cover SOCKMAP/SOCKHASH

Parametrize the SK_REUSEPORT tests so that the map type for storing sockets
is not hard-coded in the test setup routine.

This, together with careful state cleaning after the tests, lets us run the
test cases for REUSEPORT_ARRAY, SOCKMAP, and SOCKHASH to have test coverage
for all supported map types. The last two support only TCP sockets at the
moment.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200218171023.844439-11-jakub@cloudflare.com
4 years agonet: Generate reuseport group ID on group creation
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:21 +0000 (17:10 +0000)]
net: Generate reuseport group ID on group creation

Commit 736b46027eb4 ("net: Add ID (if needed) to sock_reuseport and expose
reuseport_lock") has introduced lazy generation of reuseport group IDs that
survive group resize.

By comparing the identifier we check if BPF reuseport program is not trying
to select a socket from a BPF map that belongs to a different reuseport
group than the one the packet is for.

Because SOCKARRAY used to be the only BPF map type that can be used with
reuseport BPF, it was possible to delay the generation of reuseport group
ID until a socket from the group was inserted into BPF map for the first
time.

Now that SOCK{MAP,HASH} can be used with reuseport BPF we have two options,
either generate the reuseport ID on map update, like SOCKARRAY does, or
allocate an ID from the start when reuseport group gets created.

This patch takes the latter approach to keep sockmap free of calls into
reuseport code. This streamlines the reuseport_id access as its lifetime
now matches the longevity of reuseport object.

The cost of this simplification, however, is that we allocate reuseport IDs
for all SO_REUSEPORT users. Even those that don't use SOCKARRAY in their
setups. With the way identifiers are currently generated, we can have at
most S32_MAX reuseport groups, which hopefully is sufficient. If we ever
get close to the limit, we can switch an u64 counter like sk_cookie.

Another change is that we now always call into SOCKARRAY logic to unlink
the socket from the map when unhashing or closing the socket. Previously we
did it only when at least one socket from the group was in a BPF map.

It is worth noting that this doesn't conflict with sockmap tear-down in
case a socket is in a SOCK{MAP,HASH} and belongs to a reuseport
group. sockmap tear-down happens first:

  prot->unhash
  `- tcp_bpf_unhash
     |- tcp_bpf_remove
     |  `- while (sk_psock_link_pop(psock))
     |     `- sk_psock_unlink
     |        `- sock_map_delete_from_link
     |           `- __sock_map_delete
     |              `- sock_map_unref
     |                 `- sk_psock_put
     |                    `- sk_psock_drop
     |                       `- rcu_assign_sk_user_data(sk, NULL)
     `- inet_unhash
        `- reuseport_detach_sock
           `- bpf_sk_reuseport_detach
              `- WRITE_ONCE(sk->sk_user_data, NULL)

Suggested-by: Martin Lau <kafai@fb.com>
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200218171023.844439-10-jakub@cloudflare.com
4 years agobpf: Allow selecting reuseport socket from a SOCKMAP/SOCKHASH
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:20 +0000 (17:10 +0000)]
bpf: Allow selecting reuseport socket from a SOCKMAP/SOCKHASH

SOCKMAP & SOCKHASH now support storing references to listening
sockets. Nothing keeps us from using these map types a collection of
sockets to select from in BPF reuseport programs. Whitelist the map types
with the bpf_sk_select_reuseport helper.

The restriction that the socket has to be a member of a reuseport group
still applies. Sockets in SOCKMAP/SOCKHASH that don't have sk_reuseport_cb
set are not a valid target and we signal it with -EINVAL.

The main benefit from this change is that, in contrast to
REUSEPORT_SOCKARRAY, SOCK{MAP,HASH} don't impose a restriction that a
listening socket can be just one BPF map at the same time.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200218171023.844439-9-jakub@cloudflare.com
4 years agobpf, sockmap: Let all kernel-land lookup values in SOCKMAP/SOCKHASH
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:19 +0000 (17:10 +0000)]
bpf, sockmap: Let all kernel-land lookup values in SOCKMAP/SOCKHASH

Don't require the kernel code, like BPF helpers, that needs access to
SOCK{MAP,HASH} map contents to live in net/core/sock_map.c. Expose the
lookup operation to all kernel-land.

Lookup from BPF context is not whitelisted yet. While syscalls have a
dedicated lookup handler.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200218171023.844439-8-jakub@cloudflare.com
4 years agobpf, sockmap: Return socket cookie on lookup from syscall
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:18 +0000 (17:10 +0000)]
bpf, sockmap: Return socket cookie on lookup from syscall

Tooling that populates the SOCK{MAP,HASH} with sockets from user-space
needs a way to inspect its contents. Returning the struct sock * that the
map holds to user-space is neither safe nor useful. An approach established
by REUSEPORT_SOCKARRAY is to return a socket cookie (a unique identifier)
instead.

Since socket cookies are u64 values, SOCK{MAP,HASH} need to support such a
value size for lookup to be possible. This requires special handling on
update, though. Attempts to do a lookup on a map holding u32 values will be
met with ENOSPC error.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200218171023.844439-7-jakub@cloudflare.com
4 years agobpf, sockmap: Don't set up upcalls and progs for listening sockets
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:17 +0000 (17:10 +0000)]
bpf, sockmap: Don't set up upcalls and progs for listening sockets

Now that sockmap/sockhash can hold listening sockets, when setting up the
psock we will (i) grab references to verdict/parser progs, and (2) override
socket upcalls sk_data_ready and sk_write_space.

However, since we cannot redirect to listening sockets so we don't need to
link the socket to the BPF progs. And more importantly we don't want the
listening socket to have overridden upcalls because they would get
inherited by child sockets cloned from it.

Introduce a separate initialization path for listening sockets that does
not change the upcalls and ignores the BPF progs.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200218171023.844439-6-jakub@cloudflare.com
4 years agobpf, sockmap: Allow inserting listening TCP sockets into sockmap
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:16 +0000 (17:10 +0000)]
bpf, sockmap: Allow inserting listening TCP sockets into sockmap

In order for sockmap/sockhash types to become generic collections for
storing TCP sockets we need to loosen the checks during map update, while
tightening the checks in redirect helpers.

Currently sock{map,hash} require the TCP socket to be in established state,
which prevents inserting listening sockets.

Change the update pre-checks so the socket can also be in listening state.

Since it doesn't make sense to redirect with sock{map,hash} to listening
sockets, add appropriate socket state checks to BPF redirect helpers too.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200218171023.844439-5-jakub@cloudflare.com
4 years agotcp_bpf: Don't let child socket inherit parent protocol ops on copy
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:15 +0000 (17:10 +0000)]
tcp_bpf: Don't let child socket inherit parent protocol ops on copy

Prepare for cloning listening sockets that have their protocol callbacks
overridden by sk_msg. Child sockets must not inherit parent callbacks that
access state stored in sk_user_data owned by the parent.

Restore the child socket protocol callbacks before it gets hashed and any
of the callbacks can get invoked.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200218171023.844439-4-jakub@cloudflare.com
4 years agonet, sk_msg: Clear sk_user_data pointer on clone if tagged
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:14 +0000 (17:10 +0000)]
net, sk_msg: Clear sk_user_data pointer on clone if tagged

sk_user_data can hold a pointer to an object that is not intended to be
shared between the parent socket and the child that gets a pointer copy on
clone. This is the case when sk_user_data points at reference-counted
object, like struct sk_psock.

One way to resolve it is to tag the pointer with a no-copy flag by
repurposing its lowest bit. Based on the bit-flag value we clear the child
sk_user_data pointer after cloning the parent socket.

The no-copy flag is stored in the pointer itself as opposed to externally,
say in socket flags, to guarantee that the pointer and the flag are copied
from parent to child socket in an atomic fashion. Parent socket state is
subject to change while copying, we don't hold any locks at that time.

This approach relies on an assumption that sk_user_data holds a pointer to
an object aligned at least 2 bytes. A manual audit of existing users of
rcu_dereference_sk_user_data helper confirms our assumption.

Also, an RCU-protected sk_user_data is not likely to hold a pointer to a
char value or a pathological case of "struct { char c; }". To be safe, warn
when the flag-bit is set when setting sk_user_data to catch any future
misuses.

It is worth considering why clearing sk_user_data unconditionally is not an
option. There exist users, DRBD, NVMe, and Xen drivers being among them,
that rely on the pointer being copied when cloning the listening socket.

Potentially we could distinguish these users by checking if the listening
socket has been created in kernel-space via sock_create_kern, and hence has
sk_kern_sock flag set. However, this is not the case for NVMe and Xen
drivers, which create sockets without marking them as belonging to the
kernel.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200218171023.844439-3-jakub@cloudflare.com
4 years agonet, sk_msg: Annotate lockless access to sk_prot on clone
Jakub Sitnicki [Tue, 18 Feb 2020 17:10:13 +0000 (17:10 +0000)]
net, sk_msg: Annotate lockless access to sk_prot on clone

sk_msg and ULP frameworks override protocol callbacks pointer in
sk->sk_prot, while tcp accesses it locklessly when cloning the listening
socket, that is with neither sk_lock nor sk_callback_lock held.

Once we enable use of listening sockets with sockmap (and hence sk_msg),
there will be shared access to sk->sk_prot if socket is getting cloned
while being inserted/deleted to/from the sockmap from another CPU:

Read side:

tcp_v4_rcv
  sk = __inet_lookup_skb(...)
  tcp_check_req(sk)
    inet_csk(sk)->icsk_af_ops->syn_recv_sock
      tcp_v4_syn_recv_sock
        tcp_create_openreq_child
          inet_csk_clone_lock
            sk_clone_lock
              READ_ONCE(sk->sk_prot)

Write side:

sock_map_ops->map_update_elem
  sock_map_update_elem
    sock_map_update_common
      sock_map_link_no_progs
        tcp_bpf_init
          tcp_bpf_update_sk_prot
            sk_psock_update_proto
              WRITE_ONCE(sk->sk_prot, ops)

sock_map_ops->map_delete_elem
  sock_map_delete_elem
    __sock_map_delete
     sock_map_unref
       sk_psock_put
         sk_psock_drop
           sk_psock_restore_proto
             tcp_update_ulp
               WRITE_ONCE(sk->sk_prot, proto)

Mark the shared access with READ_ONCE/WRITE_ONCE annotations.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200218171023.844439-2-jakub@cloudflare.com