]> git.proxmox.com Git - mirror_ubuntu-kernels.git/log
mirror_ubuntu-kernels.git
4 years agoMerge branch 'mlxsw-Various-fixes'
David S. Miller [Fri, 10 Jul 2020 21:33:34 +0000 (14:33 -0700)]
Merge branch 'mlxsw-Various-fixes'

Ido Schimmel says:

====================
mlxsw: Various fixes

Fix two issues found by syzkaller.

Patch #1 removes inappropriate usage of WARN_ON() following memory
allocation failure. Constantly triggered when syzkaller injects faults.

Patch #2 fixes a use-after-free that can be triggered by 'devlink dev
info' following a failed devlink reload.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: pci: Fix use-after-free in case of failed devlink reload
Ido Schimmel [Fri, 10 Jul 2020 13:41:39 +0000 (16:41 +0300)]
mlxsw: pci: Fix use-after-free in case of failed devlink reload

In case devlink reload failed, it is possible to trigger a
use-after-free when querying the kernel for device info via 'devlink dev
info' [1].

This happens because as part of the reload error path the PCI command
interface is de-initialized and its mailboxes are freed. When the
devlink '->info_get()' callback is invoked the device is queried via the
command interface and the freed mailboxes are accessed.

Fix this by initializing the command interface once during probe and not
during every reload.

This is consistent with the other bus used by mlxsw (i.e., 'mlxsw_i2c')
and also allows user space to query the running firmware version (for
example) from the device after a failed reload.

[1]
BUG: KASAN: use-after-free in memcpy include/linux/string.h:406 [inline]
BUG: KASAN: use-after-free in mlxsw_pci_cmd_exec+0x177/0xa60 drivers/net/ethernet/mellanox/mlxsw/pci.c:1675
Write of size 4096 at addr ffff88810ae32000 by task syz-executor.1/2355

CPU: 1 PID: 2355 Comm: syz-executor.1 Not tainted 5.8.0-rc2+ #29
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.1-0-ga5cab58e9a3f-prebuilt.qemu.org 04/01/2014
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0xf6/0x16e lib/dump_stack.c:118
 print_address_description.constprop.0+0x1c/0x250 mm/kasan/report.c:383
 __kasan_report mm/kasan/report.c:513 [inline]
 kasan_report.cold+0x1f/0x37 mm/kasan/report.c:530
 check_memory_region_inline mm/kasan/generic.c:186 [inline]
 check_memory_region+0x14e/0x1b0 mm/kasan/generic.c:192
 memcpy+0x39/0x60 mm/kasan/common.c:106
 memcpy include/linux/string.h:406 [inline]
 mlxsw_pci_cmd_exec+0x177/0xa60 drivers/net/ethernet/mellanox/mlxsw/pci.c:1675
 mlxsw_cmd_exec+0x249/0x550 drivers/net/ethernet/mellanox/mlxsw/core.c:2335
 mlxsw_cmd_access_reg drivers/net/ethernet/mellanox/mlxsw/cmd.h:859 [inline]
 mlxsw_core_reg_access_cmd drivers/net/ethernet/mellanox/mlxsw/core.c:1938 [inline]
 mlxsw_core_reg_access+0x2f6/0x540 drivers/net/ethernet/mellanox/mlxsw/core.c:1985
 mlxsw_reg_query drivers/net/ethernet/mellanox/mlxsw/core.c:2000 [inline]
 mlxsw_devlink_info_get+0x17f/0x6e0 drivers/net/ethernet/mellanox/mlxsw/core.c:1090
 devlink_nl_info_fill.constprop.0+0x13c/0x2d0 net/core/devlink.c:4588
 devlink_nl_cmd_info_get_dumpit+0x246/0x460 net/core/devlink.c:4648
 genl_lock_dumpit+0x85/0xc0 net/netlink/genetlink.c:575
 netlink_dump+0x515/0xe50 net/netlink/af_netlink.c:2245
 __netlink_dump_start+0x53d/0x830 net/netlink/af_netlink.c:2353
 genl_family_rcv_msg_dumpit.isra.0+0x296/0x300 net/netlink/genetlink.c:638
 genl_family_rcv_msg net/netlink/genetlink.c:733 [inline]
 genl_rcv_msg+0x78d/0x9d0 net/netlink/genetlink.c:753
 netlink_rcv_skb+0x152/0x440 net/netlink/af_netlink.c:2469
 genl_rcv+0x24/0x40 net/netlink/genetlink.c:764
 netlink_unicast_kernel net/netlink/af_netlink.c:1303 [inline]
 netlink_unicast+0x53a/0x750 net/netlink/af_netlink.c:1329
 netlink_sendmsg+0x850/0xd90 net/netlink/af_netlink.c:1918
 sock_sendmsg_nosec net/socket.c:652 [inline]
 sock_sendmsg+0x150/0x190 net/socket.c:672
 ____sys_sendmsg+0x6d8/0x840 net/socket.c:2363
 ___sys_sendmsg+0xff/0x170 net/socket.c:2417
 __sys_sendmsg+0xe5/0x1b0 net/socket.c:2450
 do_syscall_64+0x56/0xa0 arch/x86/entry/common.c:359
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: a9c8336f6544 ("mlxsw: core: Add support for devlink info command")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_router: Remove inappropriate usage of WARN_ON()
Ido Schimmel [Fri, 10 Jul 2020 13:41:38 +0000 (16:41 +0300)]
mlxsw: spectrum_router: Remove inappropriate usage of WARN_ON()

We should not trigger a warning when a memory allocation fails. Remove
the WARN_ON().

The warning is constantly triggered by syzkaller when it is injecting
faults:

[ 2230.758664] FAULT_INJECTION: forcing a failure.
[ 2230.758664] name failslab, interval 1, probability 0, space 0, times 0
[ 2230.762329] CPU: 3 PID: 1407 Comm: syz-executor.0 Not tainted 5.8.0-rc2+ #28
...
[ 2230.898175] WARNING: CPU: 3 PID: 1407 at drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:6265 mlxsw_sp_router_fib_event+0xfad/0x13e0
[ 2230.898179] Kernel panic - not syncing: panic_on_warn set ...
[ 2230.898183] CPU: 3 PID: 1407 Comm: syz-executor.0 Not tainted 5.8.0-rc2+ #28
[ 2230.898190] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.1-0-ga5cab58e9a3f-prebuilt.qemu.org 04/01/2014

Fixes: 3057224e014c ("mlxsw: spectrum_router: Implement FIB offload in deferred work")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'macb-WOL-fixes'
David S. Miller [Fri, 10 Jul 2020 21:29:38 +0000 (14:29 -0700)]
Merge branch 'macb-WOL-fixes'

Nicolas Ferre says:

====================
net: macb: Wake-on-Lan magic packet fixes and GEM handling

Here is a split series to fix WoL magic-packet on the current macb driver. Only
fixes in this one based on current net/master.

Changes in v5:
- Addressed the error code returned by phylink_ethtool_set_wol() as suggested
  by Russell.
  If PHY handles WoL, MAC doesn't stay in the way.
- Removed Florian's tag on 3/5 because of the above changes.
- Correct the "Fixes" tag on 1/5.

Changes in v4:
- Pure bug fix series for 'net'. GEM addition and MACB update removed: will be
  sent later.

Changes in v3:
- Revert some of the v2 changes done in macb_resume(). Now the resume function
  supports in-depth re-configuration of the controller in order to deal with
  deeper sleep states. Basically as it was before changes introduced by this
  series
- Tested for non-regression with our deeper Power Management mode which cuts
  power to the controller completely

Changes in v2:
- Add patch 4/7 ("net: macb: fix macb_suspend() by removing call to netif_carrier_off()")
  needed for keeping phy state consistent
- Add patch 5/7 ("net: macb: fix call to pm_runtime in the suspend/resume functions") that prevent
  putting the macb in runtime pm suspend mode when WoL is used
- Collect review tags on 3 first patches from Florian: Thanks!
- Review of macb_resume() function
- Addition of pm_wakeup_event() in both MACB and GEM WoL IRQ handlers
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: macb: fix call to pm_runtime in the suspend/resume functions
Nicolas Ferre [Fri, 10 Jul 2020 12:46:45 +0000 (14:46 +0200)]
net: macb: fix call to pm_runtime in the suspend/resume functions

The calls to pm_runtime_force_suspend/resume() functions are only
relevant if the device is not configured to act as a WoL wakeup source.
Add the device_may_wakeup() test before calling them.

Fixes: 3e2a5e153906 ("net: macb: add wake-on-lan support via magic packet")
Cc: Claudiu Beznea <claudiu.beznea@microchip.com>
Cc: Harini Katakam <harini.katakam@xilinx.com>
Cc: Sergio Prado <sergio.prado@e-labworks.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: macb: fix macb_suspend() by removing call to netif_carrier_off()
Nicolas Ferre [Fri, 10 Jul 2020 12:46:44 +0000 (14:46 +0200)]
net: macb: fix macb_suspend() by removing call to netif_carrier_off()

As we now use the phylink call to phylink_stop() in the non-WoL path,
there is no need for this call to netif_carrier_off() anymore. It can
disturb the underlying phylink FSM.

Fixes: 7897b071ac3b ("net: macb: convert to phylink")
Cc: Claudiu Beznea <claudiu.beznea@microchip.com>
Cc: Harini Katakam <harini.katakam@xilinx.com>
Cc: Antoine Tenart <antoine.tenart@bootlin.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: macb: fix macb_get/set_wol() when moving to phylink
Nicolas Ferre [Fri, 10 Jul 2020 12:46:43 +0000 (14:46 +0200)]
net: macb: fix macb_get/set_wol() when moving to phylink

Keep previous function goals and integrate phylink actions to them.

phylink_ethtool_get_wol() is not enough to figure out if Ethernet driver
supports Wake-on-Lan.
Initialization of "supported" and "wolopts" members is done in phylink
function, no need to keep them in calling function.

phylink_ethtool_set_wol() return value is considered and determines
if the MAC has to handle WoL or not. The case where the PHY doesn't
implement WoL leads to the MAC configuring it to provide this feature.

Fixes: 7897b071ac3b ("net: macb: convert to phylink")
Cc: Claudiu Beznea <claudiu.beznea@microchip.com>
Cc: Harini Katakam <harini.katakam@xilinx.com>
Cc: Antoine Tenart <antoine.tenart@bootlin.com>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: macb: mark device wake capable when "magic-packet" property present
Nicolas Ferre [Fri, 10 Jul 2020 12:46:42 +0000 (14:46 +0200)]
net: macb: mark device wake capable when "magic-packet" property present

Change the way the "magic-packet" DT property is handled in the
macb_probe() function, matching DT binding documentation.
Now we mark the device as "wakeup capable" instead of calling the
device_init_wakeup() function that would enable the wakeup source.

For Ethernet WoL, enabling the wakeup_source is done by
using ethtool and associated macb_set_wol() function that
already calls device_set_wakeup_enable() for this purpose.

That would reduce power consumption by cutting more clocks if
"magic-packet" property is set but WoL is not configured by ethtool.

Fixes: 3e2a5e153906 ("net: macb: add wake-on-lan support via magic packet")
Cc: Claudiu Beznea <claudiu.beznea@microchip.com>
Cc: Harini Katakam <harini.katakam@xilinx.com>
Cc: Sergio Prado <sergio.prado@e-labworks.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: macb: fix wakeup test in runtime suspend/resume routines
Nicolas Ferre [Fri, 10 Jul 2020 12:46:41 +0000 (14:46 +0200)]
net: macb: fix wakeup test in runtime suspend/resume routines

Use the proper struct device pointer to check if the wakeup flag
and wakeup source are positioned.
Use the one passed by function call which is equivalent to
&bp->dev->dev.parent.

It's preventing the trigger of a spurious interrupt in case the
Wake-on-Lan feature is used.

Fixes: d54f89af6cc4 ("net: macb: Add pm runtime support")
Cc: Claudiu Beznea <claudiu.beznea@microchip.com>
Cc: Harini Katakam <harini.katakam@xilinx.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobnxt_en: fix NULL dereference in case SR-IOV configuration fails
Davide Caratti [Fri, 10 Jul 2020 10:55:08 +0000 (12:55 +0200)]
bnxt_en: fix NULL dereference in case SR-IOV configuration fails

we need to set 'active_vfs' back to 0, if something goes wrong during the
allocation of SR-IOV resources: otherwise, further VF configurations will
wrongly assume that bp->pf.vf[x] are valid memory locations, and commands
like the ones in the following sequence:

 # echo 2 >/sys/bus/pci/devices/${ADDR}/sriov_numvfs
 # ip link set dev ens1f0np0 up
 # ip link set dev ens1f0np0 vf 0 trust on

will cause a kernel crash similar to this:

 bnxt_en 0000:3b:00.0: not enough MMIO resources for SR-IOV
 BUG: kernel NULL pointer dereference, address: 0000000000000014
 #PF: supervisor read access in kernel mode
 #PF: error_code(0x0000) - not-present page
 PGD 0 P4D 0
 Oops: 0000 [#1] SMP PTI
 CPU: 43 PID: 2059 Comm: ip Tainted: G          I       5.8.0-rc2.upstream+ #871
 Hardware name: Dell Inc. PowerEdge R740/08D89F, BIOS 2.2.11 06/13/2019
 RIP: 0010:bnxt_set_vf_trust+0x5b/0x110 [bnxt_en]
 Code: 44 24 58 31 c0 e8 f5 fb ff ff 85 c0 0f 85 b6 00 00 00 48 8d 1c 5b 41 89 c6 b9 0b 00 00 00 48 c1 e3 04 49 03 9c 24 f0 0e 00 00 <8b> 43 14 89 c2 83 c8 10 83 e2 ef 45 84 ed 49 89 e5 0f 44 c2 4c 89
 RSP: 0018:ffffac6246a1f570 EFLAGS: 00010246
 RAX: 0000000000000000 RBX: 0000000000000000 RCX: 000000000000000b
 RDX: 0000000000000001 RSI: 0000000000000000 RDI: ffff98b28f538900
 RBP: ffff98b28f538900 R08: 0000000000000000 R09: 0000000000000008
 R10: ffffffffb9515be0 R11: ffffac6246a1f678 R12: ffff98b28f538000
 R13: 0000000000000001 R14: 0000000000000000 R15: ffffffffc05451e0
 FS:  00007fde0f688800(0000) GS:ffff98baffd40000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 0000000000000014 CR3: 000000104bb0a003 CR4: 00000000007606e0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 PKRU: 55555554
 Call Trace:
  do_setlink+0x994/0xfe0
  __rtnl_newlink+0x544/0x8d0
  rtnl_newlink+0x47/0x70
  rtnetlink_rcv_msg+0x29f/0x350
  netlink_rcv_skb+0x4a/0x110
  netlink_unicast+0x21d/0x300
  netlink_sendmsg+0x329/0x450
  sock_sendmsg+0x5b/0x60
  ____sys_sendmsg+0x204/0x280
  ___sys_sendmsg+0x88/0xd0
  __sys_sendmsg+0x5e/0xa0
  do_syscall_64+0x47/0x80
  entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: c0c050c58d840 ("bnxt_en: New Broadcom ethernet driver.")
Reported-by: Fei Liu <feliu@redhat.com>
CC: Jonathan Toppins <jtoppins@redhat.com>
CC: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Acked-by: Jonathan Toppins <jtoppins@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Fri, 10 Jul 2020 21:07:43 +0000 (14:07 -0700)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf

Alexei Starovoitov says:

====================
pull-request: bpf 2020-07-09

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

We've added 4 non-merge commits during the last 1 day(s) which contain
a total of 4 files changed, 26 insertions(+), 15 deletions(-).

The main changes are:

1) fix crash in libbpf on 32-bit archs, from Jakub and Andrii.

2) fix crash when l2tp and bpf_sk_reuseport conflict, from Martin.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'mlx5-fixes-2020-07-02' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Fri, 10 Jul 2020 21:02:01 +0000 (14:02 -0700)]
Merge tag 'mlx5-fixes-2020-07-02' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5 fixes 2020-07-02

This series introduces some fixes to mlx5 driver.

V1->v2:
 - Drop "ip -s" patch and mirred device hold reference patch.
 - Will revise them in a later submission.

Please pull and let me know if there is any problem.

For -stable v5.2
 ('net/mlx5: Fix eeprom support for SFP module')

For -stable v5.4
 ('net/mlx5e: Fix 50G per lane indication')

For -stable v5.5
 ('net/mlx5e: Fix CPU mapping after function reload to avoid aRFS RX crash')
 ('net/mlx5e: Fix VXLAN configuration restore after function reload')

For -stable v5.7
 ('net/mlx5e: CT: Fix memory leak in cleanup')
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agolibbpf: Fix libbpf hashmap on (I)LP32 architectures
Jakub Bogusz [Thu, 9 Jul 2020 22:57:23 +0000 (15:57 -0700)]
libbpf: Fix libbpf hashmap on (I)LP32 architectures

On ILP32, 64-bit result was shifted by value calculated for 32-bit long type
and returned value was much outside hashmap capacity.
As advised by Andrii Nakryiko, this patch uses different hashing variant for
architectures with size_t shorter than long long.

Fixes: e3b924224028 ("libbpf: add resizable non-thread safe internal hashmap")
Signed-off-by: Jakub Bogusz <qboosh@pld-linux.org>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200709225723.1069937-1-andriin@fb.com
4 years agonet/mlx5e: CT: Fix memory leak in cleanup
Eli Britstein [Sun, 28 Jun 2020 12:42:26 +0000 (15:42 +0300)]
net/mlx5e: CT: Fix memory leak in cleanup

CT entries are deleted via a workqueue from netfilter. If removing the
module before that, the rules are cleaned by the driver itself, but the
memory entries for them are not freed. Fix that.

Fixes: ac991b48d43c ("net/mlx5e: CT: Offload established flows")
Signed-off-by: Eli Britstein <elibr@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Fix port buffers cell size value
Eran Ben Elisha [Mon, 22 Jun 2020 06:03:31 +0000 (09:03 +0300)]
net/mlx5e: Fix port buffers cell size value

Device unit for port buffers size, xoff_threshold and xon_threshold is
cells. Fix a bug in driver where cell unit size was hard-coded to
128 bytes. This hard-coded value is buggy, as it is wrong for some hardware
versions.

Driver to read cell size from SBCAM register and translate bytes to cell
units accordingly.

In order to fix the bug, this patch exposes SBCAM (Shared buffer
capabilities mask) layout and defines.

If SBCAM.cap_cell_size is valid, use it for all bytes to cells
calculations. If not valid, fallback to 128.

Cell size do not change on the fly per device. Instead of issuing SBCAM
access reg command every time such translation is needed, cache it in
mlx5e_dcbx as part of mlx5e_dcbnl_initialize(). Pass dcbx.port_buff_cell_sz
as a param to every function that needs bytes to cells translation.

While fixing the bug, move MLX5E_BUFFER_CELL_SHIFT macro to
en_dcbnl.c, as it is only used by that file.

Fixes: 0696d60853d5 ("net/mlx5e: Receive buffer configuration")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Reviewed-by: Huy Nguyen <huyn@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Fix 50G per lane indication
Aya Levin [Mon, 15 Jun 2020 09:48:47 +0000 (12:48 +0300)]
net/mlx5e: Fix 50G per lane indication

Some released FW versions mistakenly don't set the capability that 50G
per lane link-modes are supported for VFs (ptys_extended_ethernet
capability bit). When the capability is unset, read
PTYS.ext_eth_proto_capability (always reliable).
If PTYS.ext_eth_proto_capability is valid (has a non-zero value)
conclude that the HCA supports 50G per lane. Otherwise, conclude that
the HCA doesn't support 50G per lane.

Fixes: a08b4ed1373d ("net/mlx5: Add support to ext_* fields introduced in Port Type and Speed register")
Signed-off-by: Aya Levin <ayal@mellanox.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Fix CPU mapping after function reload to avoid aRFS RX crash
Aya Levin [Wed, 20 May 2020 07:37:42 +0000 (10:37 +0300)]
net/mlx5e: Fix CPU mapping after function reload to avoid aRFS RX crash

After function reload, CPU mapping used by aRFS RX is broken, leading to
a kernel panic. Fix by moving initialization of rx_cpu_rmap from
netdev_init to netdev_attach. IRQ table is re-allocated on mlx5_load,
but netdev is not re-initialize.

Trace of the panic:
[ 22.055672] general protection fault, probably for non-canonical address 0x785634120000ff1c: 0000 [#1] SMP PTI
[ 22.065010] CPU: 4 PID: 0 Comm: swapper/4 Not tainted 5.7.0-rc2-for-upstream-perf-2020-04-21_16-34-03-31 #1
[ 22.067967] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.12.1-0-ga5cab58e9a3f-prebuilt.qemu.org 04/01/2014
[ 22.071174] RIP: 0010:get_rps_cpu+0x267/0x300
[ 22.075692] RSP: 0018:ffffc90000244d60 EFLAGS: 00010202
[ 22.076888] RAX: ffff888459b0e400 RBX: 0000000000000000 RCX:0000000000000007
[ 22.078364] RDX: 0000000000008884 RSI: ffff888467cb5b00 RDI:0000000000000000
[ 22.079815] RBP: 00000000ff342b27 R08: 0000000000000007 R09:0000000000000003
[ 22.081289] R10: ffffffffffffffff R11: 00000000000070cc R12:ffff888454900000
[ 22.082767] R13: ffffc90000e5a950 R14: ffffc90000244dc0 R15:0000000000000007
[ 22.084190] FS: 0000000000000000(0000) GS:ffff88846fc80000(0000)knlGS:0000000000000000
[ 22.086161] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 22.087427] CR2: ffffffffffffffff CR3: 0000000464426003 CR4:0000000000760ee0
[ 22.088888] DR0: 0000000000000000 DR1: 0000000000000000 DR2:0000000000000000
[ 22.090336] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7:0000000000000400
[ 22.091764] PKRU: 55555554
[ 22.092618] Call Trace:
[ 22.093442] <IRQ>
[ 22.094211] ? kvm_clock_get_cycles+0xd/0x10
[ 22.095272] netif_receive_skb_list_internal+0x258/0x2a0
[ 22.096460] gro_normal_list.part.137+0x19/0x40
[ 22.097547] napi_complete_done+0xc6/0x110
[ 22.098685] mlx5e_napi_poll+0x190/0x670 [mlx5_core]
[ 22.099859] net_rx_action+0x2a0/0x400
[ 22.100848] __do_softirq+0xd8/0x2a8
[ 22.101829] irq_exit+0xa5/0xb0
[ 22.102750] do_IRQ+0x52/0xd0
[ 22.103654] common_interrupt+0xf/0xf
[ 22.104641] </IRQ>

Fixes: 4383cfcc65e7 ("net/mlx5: Add devlink reload")
Signed-off-by: Aya Levin <ayal@mellanox.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Fix VXLAN configuration restore after function reload
Aya Levin [Wed, 24 Jun 2020 16:04:03 +0000 (19:04 +0300)]
net/mlx5e: Fix VXLAN configuration restore after function reload

When detaching netdev, remove vxlan port configuration using
udp_tunnel_drop_rx_info. During function reload, configuration will be
restored using udp_tunnel_get_rx_info. This ensures sync between
firmware and driver. Use udp_tunnel_get_rx_info even if its physical
interface is down.

Fixes: 4383cfcc65e7 ("net/mlx5: Add devlink reload")
Signed-off-by: Aya Levin <ayal@mellanox.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5e: Fix usage of rcu-protected pointer
Vlad Buslov [Wed, 17 Jun 2020 14:51:53 +0000 (17:51 +0300)]
net/mlx5e: Fix usage of rcu-protected pointer

In mlx5e_configure_flower() flow pointer is protected by rcu read lock.
However, after cited commit the pointer is being used outside of rcu read
block. Extend the block to protect all pointer accesses.

Fixes: 553f9328385d ("net/mlx5e: Support tc block sharing for representors")
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mxl5e: Verify that rpriv is not NULL
Vlad Buslov [Wed, 17 Jun 2020 14:26:33 +0000 (17:26 +0300)]
net/mxl5e: Verify that rpriv is not NULL

In helper function is_flow_rule_duplicate_allowed() verify that rpviv
pointer is not NULL before dereferencing it. This can happen when device is
in NIC mode and leads to following crash:

[90444.046419] BUG: kernel NULL pointer dereference, address: 0000000000000000
[90444.048149] #PF: supervisor read access in kernel mode
[90444.049781] #PF: error_code(0x0000) - not-present page
[90444.051386] PGD 80000003d35a4067 P4D 80000003d35a4067 PUD 3d35a3067 PMD 0
[90444.053051] Oops: 0000 [#1] SMP PTI
[90444.054683] CPU: 16 PID: 31736 Comm: tc Not tainted 5.8.0-rc1+ #1157
[90444.056340] Hardware name: Supermicro SYS-2028TP-DECR/X10DRT-P, BIOS 2.0b 03/30/2017
[90444.058079] RIP: 0010:mlx5e_configure_flower+0x3aa/0x9b0 [mlx5_core]
[90444.059753] Code: 24 50 49 8b 95 08 02 00 00 48 b8 00 08 00 00 04 00 00 00 48 21 c2 48 39 c2 74 0a 41 f6 85 0d 02 00 00 20 74 16 48 8b 44 24 20 <48> 8b 00 66 83 78 20 ff 74 07 4d 89 aa e0 00 00 00 48 83 7d 28 00
[90444.063232] RSP: 0018:ffffabe9c61ff768 EFLAGS: 00010246
[90444.065014] RAX: 0000000000000000 RBX: ffff9b13c4c91e80 RCX: 00000000000093fa
[90444.066784] RDX: 0000000400000800 RSI: 0000000000000000 RDI: 000000000002d5e0
[90444.068533] RBP: ffff9b174d308468 R08: 0000000000000000 R09: ffff9b17d63003f0
[90444.070285] R10: ffff9b17ea288600 R11: 0000000000000000 R12: ffffabe9c61ff878
[90444.072032] R13: ffff9b174d300000 R14: ffffabe9c61ffbb8 R15: ffff9b174d300880
[90444.073760] FS:  00007f3c23775480(0000) GS:ffff9b13efc80000(0000) knlGS:0000000000000000
[90444.075492] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[90444.077266] CR2: 0000000000000000 CR3: 00000003e2a60002 CR4: 00000000001606e0
[90444.079024] Call Trace:
[90444.080753]  tc_setup_cb_add+0xca/0x1e0
[90444.082415]  fl_hw_replace_filter+0x15f/0x1f0 [cls_flower]
[90444.084119]  fl_change+0xa59/0x13dc [cls_flower]
[90444.085772]  ? wait_for_completion+0xa8/0xf0
[90444.087364]  tc_new_tfilter+0x3f5/0xa60
[90444.088960]  rtnetlink_rcv_msg+0xeb/0x360
[90444.090514]  ? __d_lookup_done+0x76/0xe0
[90444.092034]  ? proc_alloc_inode+0x16/0x70
[90444.093560]  ? prep_new_page+0x8c/0xf0
[90444.095048]  ? _cond_resched+0x15/0x30
[90444.096483]  ? rtnl_calcit.isra.0+0x110/0x110
[90444.097907]  netlink_rcv_skb+0x49/0x110
[90444.099289]  netlink_unicast+0x191/0x230
[90444.100629]  netlink_sendmsg+0x243/0x480
[90444.101984]  sock_sendmsg+0x5e/0x60
[90444.103305]  ____sys_sendmsg+0x1f3/0x260
[90444.104597]  ? copy_msghdr_from_user+0x5c/0x90
[90444.105916]  ? __mod_lruvec_state+0x3c/0xe0
[90444.107210]  ___sys_sendmsg+0x81/0xc0
[90444.108484]  ? do_filp_open+0xa5/0x100
[90444.109732]  ? handle_mm_fault+0x117b/0x1e00
[90444.110970]  ? __check_object_size+0x46/0x147
[90444.112205]  ? __check_object_size+0x136/0x147
[90444.113402]  __sys_sendmsg+0x59/0xa0
[90444.114587]  do_syscall_64+0x4d/0x90
[90444.115782]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
[90444.116953] RIP: 0033:0x7f3c2393b7b8
[90444.118101] Code: Bad RIP value.
[90444.119240] RSP: 002b:00007ffc6ad8e6c8 EFLAGS: 00000246 ORIG_RAX: 000000000000002e
[90444.120408] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f3c2393b7b8
[90444.121583] RDX: 0000000000000000 RSI: 00007ffc6ad8e740 RDI: 0000000000000003
[90444.122750] RBP: 000000005eea0c3a R08: 0000000000000001 R09: 00007ffc6ad8e68c
[90444.123928] R10: 0000000000404fa8 R11: 0000000000000246 R12: 0000000000000001
[90444.125073] R13: 0000000000000000 R14: 00007ffc6ad92a00 R15: 00000000004866a0
[90444.126221] Modules linked in: act_skbedit act_tunnel_key act_mirred bonding vxlan ip6_udp_tunnel udp_tunnel nfnetlink act_gact cls_flower sch_ingress openvswitch nsh nf_conncount nfsv3 nfs_acl nfs lockd grace fscache tun bridge stp llc sunrpc rdma_ucm rdma_cm iw_cm ib_cm mlx5_ib ib_uverbs ib_core mlx5_core intel_r
apl_msr intel_rapl_common sb_edac x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel mlxfw kvm act_ct nf_flow_table nf_nat nf_conntrack irqbypass crct10dif_pclmul nf_defrag_ipv6 igb ipmi_ssif libcrc32c crc32_pclmul crc32c_intel ipmi_si nf_defrag_ipv4 ptp ghash_clmulni_intel mei_me ses iTCO_wdt i2c_i801 pps_core
ioatdma iTCO_vendor_support joydev mei enclosure intel_cstate i2c_smbus wmi dca ipmi_devintf intel_uncore lpc_ich ipmi_msghandler pcspkr acpi_pad acpi_power_meter ast i2c_algo_bit drm_vram_helper drm_kms_helper drm_ttm_helper ttm drm mpt3sas raid_class scsi_transport_sas
[90444.136253] CR2: 0000000000000000
[90444.137621] ---[ end trace 924af62aa2b151bd ]---

Fixes: 553f9328385d ("net/mlx5e: Support tc block sharing for representors")
Reported-by: David Ahern <dsahern@gmail.com>
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: E-Switch, Fix vlan or qos setting in legacy mode
Vu Pham [Wed, 17 Jun 2020 22:11:24 +0000 (15:11 -0700)]
net/mlx5: E-Switch, Fix vlan or qos setting in legacy mode

Refactoring eswitch ingress acl codes accidentally inserts extra
memset zero that removes vlan and/or qos setting in legacy mode.

Fixes: 07bab9502641 ("net/mlx5: E-Switch, Refactor eswitch ingress acl codes")
Signed-off-by: Vu Pham <vuhuong@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agonet/mlx5: Fix eeprom support for SFP module
Eran Ben Elisha [Sun, 14 Jun 2020 14:31:26 +0000 (17:31 +0300)]
net/mlx5: Fix eeprom support for SFP module

Fix eeprom SFP query support by setting i2c_addr, offset and page number
correctly. Unlike QSFP modules, SFP eeprom params are as follow:
- i2c_addr is 0x50 for offset 0 - 255 and 0x51 for offset 256 - 511.
- Page number is always zero.
- Page offset is always relative to zero.

As part of eeprom query, query the module ID (SFP / QSFP*) via helper
function to set the params accordingly.

In addition, change mlx5_qsfp_eeprom_page() input type to be u16 to avoid
unnecessary casting.

Fixes: a708fb7b1f8d ("net/mlx5e: ethtool, Add support for EEPROM high pages query")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Huy Nguyen <huyn@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
4 years agocgroup: Fix sock_cgroup_data on big-endian.
Cong Wang [Thu, 9 Jul 2020 23:28:44 +0000 (16:28 -0700)]
cgroup: Fix sock_cgroup_data on big-endian.

In order for no_refcnt and is_data to be the lowest order two
bits in the 'val' we have to pad out the bitfield of the u8.

Fixes: ad0f75e5f57c ("cgroup: fix cgroup_sk_alloc() for sk_clone_lock()")
Reported-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: bpf: Fix detach from sockmap tests
Lorenz Bauer [Thu, 9 Jul 2020 11:51:51 +0000 (12:51 +0100)]
selftests: bpf: Fix detach from sockmap tests

Fix sockmap tests which rely on old bpf_prog_dispatch behaviour.
In the first case, the tests check that detaching without giving
a program succeeds. Since these are not the desired semantics,
invert the condition. In the second case, the clean up code doesn't
supply the necessary program fds.

Fixes: bb0de3131f4c ("bpf: sockmap: Require attach_bpf_fd when detaching a program")
Reported-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/bpf/20200709115151.75829-1-lmb@cloudflare.com
4 years agotcp: make sure listeners don't initialize congestion-control state
Christoph Paasch [Wed, 8 Jul 2020 23:18:34 +0000 (16:18 -0700)]
tcp: make sure listeners don't initialize congestion-control state

syzkaller found its way into setsockopt with TCP_CONGESTION "cdg".
tcp_cdg_init() does a kcalloc to store the gradients. As sk_clone_lock
just copies all the memory, the allocated pointer will be copied as
well, if the app called setsockopt(..., TCP_CONGESTION) on the listener.
If now the socket will be destroyed before the congestion-control
has properly been initialized (through a call to tcp_init_transfer), we
will end up freeing memory that does not belong to that particular
socket, opening the door to a double-free:

[   11.413102] ==================================================================
[   11.414181] BUG: KASAN: double-free or invalid-free in tcp_cleanup_congestion_control+0x58/0xd0
[   11.415329]
[   11.415560] CPU: 3 PID: 4884 Comm: syz-executor.5 Not tainted 5.8.0-rc2 #80
[   11.416544] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.1-0-ga5cab58e9a3f-prebuilt.qemu.org 04/01/2014
[   11.418148] Call Trace:
[   11.418534]  <IRQ>
[   11.418834]  dump_stack+0x7d/0xb0
[   11.419297]  print_address_description.constprop.0+0x1a/0x210
[   11.422079]  kasan_report_invalid_free+0x51/0x80
[   11.423433]  __kasan_slab_free+0x15e/0x170
[   11.424761]  kfree+0x8c/0x230
[   11.425157]  tcp_cleanup_congestion_control+0x58/0xd0
[   11.425872]  tcp_v4_destroy_sock+0x57/0x5a0
[   11.426493]  inet_csk_destroy_sock+0x153/0x2c0
[   11.427093]  tcp_v4_syn_recv_sock+0xb29/0x1100
[   11.427731]  tcp_get_cookie_sock+0xc3/0x4a0
[   11.429457]  cookie_v4_check+0x13d0/0x2500
[   11.433189]  tcp_v4_do_rcv+0x60e/0x780
[   11.433727]  tcp_v4_rcv+0x2869/0x2e10
[   11.437143]  ip_protocol_deliver_rcu+0x23/0x190
[   11.437810]  ip_local_deliver+0x294/0x350
[   11.439566]  __netif_receive_skb_one_core+0x15d/0x1a0
[   11.441995]  process_backlog+0x1b1/0x6b0
[   11.443148]  net_rx_action+0x37e/0xc40
[   11.445361]  __do_softirq+0x18c/0x61a
[   11.445881]  asm_call_on_stack+0x12/0x20
[   11.446409]  </IRQ>
[   11.446716]  do_softirq_own_stack+0x34/0x40
[   11.447259]  do_softirq.part.0+0x26/0x30
[   11.447827]  __local_bh_enable_ip+0x46/0x50
[   11.448406]  ip_finish_output2+0x60f/0x1bc0
[   11.450109]  __ip_queue_xmit+0x71c/0x1b60
[   11.451861]  __tcp_transmit_skb+0x1727/0x3bb0
[   11.453789]  tcp_rcv_state_process+0x3070/0x4d3a
[   11.456810]  tcp_v4_do_rcv+0x2ad/0x780
[   11.457995]  __release_sock+0x14b/0x2c0
[   11.458529]  release_sock+0x4a/0x170
[   11.459005]  __inet_stream_connect+0x467/0xc80
[   11.461435]  inet_stream_connect+0x4e/0xa0
[   11.462043]  __sys_connect+0x204/0x270
[   11.465515]  __x64_sys_connect+0x6a/0xb0
[   11.466088]  do_syscall_64+0x3e/0x70
[   11.466617]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
[   11.467341] RIP: 0033:0x7f56046dc469
[   11.467844] Code: Bad RIP value.
[   11.468282] RSP: 002b:00007f5604dccdd8 EFLAGS: 00000246 ORIG_RAX: 000000000000002a
[   11.469326] RAX: ffffffffffffffda RBX: 000000000068bf00 RCX: 00007f56046dc469
[   11.470379] RDX: 0000000000000010 RSI: 0000000020000000 RDI: 0000000000000004
[   11.471311] RBP: 00000000ffffffff R08: 0000000000000000 R09: 0000000000000000
[   11.472286] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
[   11.473341] R13: 000000000041427c R14: 00007f5604dcd5c0 R15: 0000000000000003
[   11.474321]
[   11.474527] Allocated by task 4884:
[   11.475031]  save_stack+0x1b/0x40
[   11.475548]  __kasan_kmalloc.constprop.0+0xc2/0xd0
[   11.476182]  tcp_cdg_init+0xf0/0x150
[   11.476744]  tcp_init_congestion_control+0x9b/0x3a0
[   11.477435]  tcp_set_congestion_control+0x270/0x32f
[   11.478088]  do_tcp_setsockopt.isra.0+0x521/0x1a00
[   11.478744]  __sys_setsockopt+0xff/0x1e0
[   11.479259]  __x64_sys_setsockopt+0xb5/0x150
[   11.479895]  do_syscall_64+0x3e/0x70
[   11.480395]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
[   11.481097]
[   11.481321] Freed by task 4872:
[   11.481783]  save_stack+0x1b/0x40
[   11.482230]  __kasan_slab_free+0x12c/0x170
[   11.482839]  kfree+0x8c/0x230
[   11.483240]  tcp_cleanup_congestion_control+0x58/0xd0
[   11.483948]  tcp_v4_destroy_sock+0x57/0x5a0
[   11.484502]  inet_csk_destroy_sock+0x153/0x2c0
[   11.485144]  tcp_close+0x932/0xfe0
[   11.485642]  inet_release+0xc1/0x1c0
[   11.486131]  __sock_release+0xc0/0x270
[   11.486697]  sock_close+0xc/0x10
[   11.487145]  __fput+0x277/0x780
[   11.487632]  task_work_run+0xeb/0x180
[   11.488118]  __prepare_exit_to_usermode+0x15a/0x160
[   11.488834]  do_syscall_64+0x4a/0x70
[   11.489326]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

Wei Wang fixed a part of these CDG-malloc issues with commit c12014440750
("tcp: memset ca_priv data to 0 properly").

This patch here fixes the listener-scenario: We make sure that listeners
setting the congestion-control through setsockopt won't initialize it
(thus CDG never allocates on listeners). For those who use AF_UNSPEC to
reuse a socket, tcp_disconnect() is changed to cleanup afterwards.

(The issue can be reproduced at least down to v4.4.x.)

Cc: Wei Wang <weiwan@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Fixes: 2b0a8c9eee81 ("tcp: add CDG congestion control")
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobpf: net: Avoid incorrect bpf_sk_reuseport_detach call
Martin KaFai Lau [Thu, 9 Jul 2020 06:11:10 +0000 (23:11 -0700)]
bpf: net: Avoid incorrect bpf_sk_reuseport_detach call

bpf_sk_reuseport_detach is currently called when sk->sk_user_data
is not NULL.  It is incorrect because sk->sk_user_data may not be
managed by the bpf's reuseport_array.  It has been reported in [1] that,
the bpf_sk_reuseport_detach() which is called from udp_lib_unhash() has
corrupted the sk_user_data managed by l2tp.

This patch solves it by using another bit (defined as SK_USER_DATA_BPF)
of the sk_user_data pointer value.  It marks that a sk_user_data is
managed/owned by BPF.

The patch depends on a PTRMASK introduced in
commit f1ff5ce2cd5e ("net, sk_msg: Clear sk_user_data pointer on clone if tagged").

[ Note: sk->sk_user_data is used by bpf's reuseport_array only when a sk is
  added to the bpf's reuseport_array.
  i.e. doing setsockopt(SO_REUSEPORT) and having "sk->sk_reuseport == 1"
  alone will not stop sk->sk_user_data being used by other means. ]

[1]: https://lore.kernel.org/netdev/20200706121259.GA20199@katalix.com/

Fixes: 5dc4c4b7d4e8 ("bpf: Introduce BPF_MAP_TYPE_REUSEPORT_SOCKARRAY")
Reported-by: James Chapman <jchapman@katalix.com>
Reported-by: syzbot+9f092552ba9a5efca5df@syzkaller.appspotmail.com
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: James Chapman <jchapman@katalix.com>
Acked-by: James Chapman <jchapman@katalix.com>
Link: https://lore.kernel.org/bpf/20200709061110.4019316-1-kafai@fb.com
4 years agobpf: net: Avoid copying sk_user_data of reuseport_array during sk_clone
Martin KaFai Lau [Thu, 9 Jul 2020 06:11:04 +0000 (23:11 -0700)]
bpf: net: Avoid copying sk_user_data of reuseport_array during sk_clone

It makes little sense for copying sk_user_data of reuseport_array during
sk_clone_lock().  This patch reuses the SK_USER_DATA_NOCOPY bit introduced in
commit f1ff5ce2cd5e ("net, sk_msg: Clear sk_user_data pointer on clone if tagged").
It is used to mark the sk_user_data is not supposed to be copied to its clone.

Although the cloned sk's sk_user_data will not be used/freed in
bpf_sk_reuseport_detach(), this change can still allow the cloned
sk's sk_user_data to be used by some other means.

Freeing the reuseport_array's sk_user_data does not require a rcu grace
period.  Thus, the existing rcu_assign_sk_user_data_nocopy() is not
used.

Fixes: 5dc4c4b7d4e8 ("bpf: Introduce BPF_MAP_TYPE_REUSEPORT_SOCKARRAY")
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/bpf/20200709061104.4018798-1-kafai@fb.com
4 years agoethtool: fix genlmsg_put() failure handling in ethnl_default_dumpit()
Michal Kubecek [Thu, 9 Jul 2020 10:11:50 +0000 (12:11 +0200)]
ethtool: fix genlmsg_put() failure handling in ethnl_default_dumpit()

If the genlmsg_put() call in ethnl_default_dumpit() fails, we bail out
without checking if we already have some messages in current skb like we do
with ethnl_default_dump_one() failure later. Therefore if existing messages
almost fill up the buffer so that there is not enough space even for
netlink and genetlink header, we lose all prepared messages and return and
error.

Rather than duplicating the skb->len check, move the genlmsg_put(),
genlmsg_cancel() and genlmsg_end() calls into ethnl_default_dump_one().
This is also more logical as all message composition will be in
ethnl_default_dump_one() and only iteration logic will be left in
ethnl_default_dumpit().

Fixes: 728480f12442 ("ethtool: default handlers for GET requests")
Reported-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Michal Kubecek <mkubecek@suse.cz>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet_sched: fix a memory leak in atm_tc_init()
Cong Wang [Thu, 9 Jul 2020 03:13:59 +0000 (20:13 -0700)]
net_sched: fix a memory leak in atm_tc_init()

When tcf_block_get() fails inside atm_tc_init(),
atm_tc_put() is called to release the qdisc p->link.q.
But the flow->ref prevents it to do so, as the flow->ref
is still zero.

Fix this by moving the p->link.ref initialization before
tcf_block_get().

Fixes: 6529eaba33f0 ("net: sched: introduce tcf block infractructure")
Reported-and-tested-by: syzbot+d411cff6ab29cc2c311b@syzkaller.appspotmail.com
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoqed: Populate nvm-file attributes while reading nvm config partition.
Sudarsana Reddy Kalluru [Thu, 9 Jul 2020 03:14:29 +0000 (20:14 -0700)]
qed: Populate nvm-file attributes while reading nvm config partition.

NVM config file address will be modified when the MBI image is upgraded.
Driver would return stale config values if user reads the nvm-config
(via ethtool -d) in this state. The fix is to re-populate nvm attribute
info while reading the nvm config values/partition.

Changes from previous version:
-------------------------------
v3: Corrected the formatting in 'Fixes' tag.
v2: Added 'Fixes' tag.

Fixes: 1ac4329a1cff ("qed: Add configuration information to register dump and debug data")
Signed-off-by: Sudarsana Reddy Kalluru <skalluru@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocxgb4: fix all-mask IP address comparison
Rahul Lakkireddy [Wed, 8 Jul 2020 21:44:27 +0000 (03:14 +0530)]
cxgb4: fix all-mask IP address comparison

Convert all-mask IP address to Big Endian, instead, for comparison.

Fixes: f286dd8eaad5 ("cxgb4: use correct type for all-mask IP address comparison")
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotipc: fix retransmission on unicast links
Hamish Martin [Wed, 8 Jul 2020 21:06:44 +0000 (09:06 +1200)]
tipc: fix retransmission on unicast links

A scenario has been observed where a 'bc_init' message for a link is not
retransmitted if it fails to be received by the peer. This leads to the
peer never establishing the link fully and it discarding all other data
received on the link. In this scenario the message is lost in transit to
the peer.

The issue is traced to the 'nxt_retr' field of the skb not being
initialised for links that aren't a bc_sndlink. This leads to the
comparison in tipc_link_advance_transmq() that gates whether to attempt
retransmission of a message performing in an undesirable way.
Depending on the relative value of 'jiffies', this comparison:
    time_before(jiffies, TIPC_SKB_CB(skb)->nxt_retr)
may return true or false given that 'nxt_retr' remains at the
uninitialised value of 0 for non bc_sndlinks.

This is most noticeable shortly after boot when jiffies is initialised
to a high value (to flush out rollover bugs) and we compare a jiffies of,
say, 4294940189 to zero. In that case time_before returns 'true' leading
to the skb not being retransmitted.

The fix is to ensure that all skbs have a valid 'nxt_retr' time set for
them and this is achieved by refactoring the setting of this value into
a central function.
With this fix, transmission losses of 'bc_init' messages do not stall
the link establishment forever because the 'bc_init' message is
retransmitted and the link eventually establishes correctly.

Fixes: 382f598fb66b ("tipc: reduce duplicate packets for unicast traffic")
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Hamish Martin <hamish.martin@alliedtelesis.co.nz>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agol2tp: remove skb_dst_set() from l2tp_xmit_skb()
Xin Long [Mon, 6 Jul 2020 18:02:32 +0000 (02:02 +0800)]
l2tp: remove skb_dst_set() from l2tp_xmit_skb()

In the tx path of l2tp, l2tp_xmit_skb() calls skb_dst_set() to set
skb's dst. However, it will eventually call inet6_csk_xmit() or
ip_queue_xmit() where skb's dst will be overwritten by:

   skb_dst_set_noref(skb, dst);

without releasing the old dst in skb. Then it causes dst/dev refcnt leak:

  unregister_netdevice: waiting for eth0 to become free. Usage count = 1

This can be reproduced by simply running:

  # modprobe l2tp_eth && modprobe l2tp_ip
  # sh ./tools/testing/selftests/net/l2tp.sh

So before going to inet6_csk_xmit() or ip_queue_xmit(), skb's dst
should be dropped. This patch is to fix it by removing skb_dst_set()
from l2tp_xmit_skb() and moving skb_dst_drop() into l2tp_xmit_core().

Fixes: 3557baabf280 ("[L2TP]: PPP over L2TP driver core")
Reported-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: James Chapman <jchapman@katalix.com>
Tested-by: James Chapman <jchapman@katalix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-smc-fixes'
David S. Miller [Wed, 8 Jul 2020 19:35:15 +0000 (12:35 -0700)]
Merge branch 'net-smc-fixes'

Karsten Graul says:

====================
net/smc: fixes 2020-07-08

Please apply the following patch series for smc to netdev's net tree.

The patches fix problems found during more testing of SMC
functionality, resulting in hang conditions and unneeded link
deactivations. The clc module was hardened to be prepared for
possible future SMCD versions.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/smc: tolerate future SMCD versions
Ursula Braun [Wed, 8 Jul 2020 15:05:15 +0000 (17:05 +0200)]
net/smc: tolerate future SMCD versions

CLC proposal messages of future SMCD versions could be larger than SMCD
V1 CLC proposal messages.
To enable toleration in SMC V1 the receival of CLC proposal messages
is adapted:
* accept larger length values in CLC proposal
* check trailing eye catcher for incoming CLC proposal with V1 length only
* receive the whole CLC proposal even in cases it does not fit into the
  V1 buffer

Fixes: e7b7a64a8493d ("smc: support variable CLC proposal messages")
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/smc: switch smcd_dev_list spinlock to mutex
Ursula Braun [Wed, 8 Jul 2020 15:05:14 +0000 (17:05 +0200)]
net/smc: switch smcd_dev_list spinlock to mutex

The similar smc_ib_devices spinlock has been converted to a mutex.
Protecting the smcd_dev_list by a mutex is possible as well. This
patch converts the smcd_dev_list spinlock to a mutex.

Fixes: c6ba7c9ba43d ("net/smc: add base infrastructure for SMC-D and ISM")
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/smc: fix sleep bug in smc_pnet_find_roce_resource()
Ursula Braun [Wed, 8 Jul 2020 15:05:13 +0000 (17:05 +0200)]
net/smc: fix sleep bug in smc_pnet_find_roce_resource()

Tests showed this BUG:
[572555.252867] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:935
[572555.252876] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 131031, name: smcapp
[572555.252879] INFO: lockdep is turned off.
[572555.252883] CPU: 1 PID: 131031 Comm: smcapp Tainted: G           O      5.7.0-rc3uschi+ #356
[572555.252885] Hardware name: IBM 3906 M03 703 (LPAR)
[572555.252887] Call Trace:
[572555.252896]  [<00000000ac364554>] show_stack+0x94/0xe8
[572555.252901]  [<00000000aca1f400>] dump_stack+0xa0/0xe0
[572555.252906]  [<00000000ac3c8c10>] ___might_sleep+0x260/0x280
[572555.252910]  [<00000000acdc0c98>] __mutex_lock+0x48/0x940
[572555.252912]  [<00000000acdc15c2>] mutex_lock_nested+0x32/0x40
[572555.252975]  [<000003ff801762d0>] mlx5_lag_get_roce_netdev+0x30/0xc0 [mlx5_core]
[572555.252996]  [<000003ff801fb3aa>] mlx5_ib_get_netdev+0x3a/0xe0 [mlx5_ib]
[572555.253007]  [<000003ff80063848>] smc_pnet_find_roce_resource+0x1d8/0x310 [smc]
[572555.253011]  [<000003ff800602f0>] __smc_connect+0x1f0/0x3e0 [smc]
[572555.253015]  [<000003ff80060634>] smc_connect+0x154/0x190 [smc]
[572555.253022]  [<00000000acbed8d4>] __sys_connect+0x94/0xd0
[572555.253025]  [<00000000acbef620>] __s390x_sys_socketcall+0x170/0x360
[572555.253028]  [<00000000acdc6800>] system_call+0x298/0x2b8
[572555.253030] INFO: lockdep is turned off.

Function smc_pnet_find_rdma_dev() might be called from
smc_pnet_find_roce_resource(). It holds the smc_ib_devices list
spinlock while calling infiniband op get_netdev(). At least for mlx5
the get_netdev operation wants mutex serialization, which conflicts
with the smc_ib_devices spinlock.
This patch switches the smc_ib_devices spinlock into a mutex to
allow sleeping when calling get_netdev().

Fixes: a4cf0443c414 ("smc: introduce SMC as an IB-client")
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/smc: fix work request handling
Karsten Graul [Wed, 8 Jul 2020 15:05:12 +0000 (17:05 +0200)]
net/smc: fix work request handling

Wait for pending sends only when smc_switch_conns() found a link to move
the connections to. Do not wait during link freeing, this can lead to
permanent hang situations. And refuse to provide a new tx slot on an
unusable link.

Fixes: c6f02ebeea3a ("net/smc: switch connections to alternate link")
Reviewed-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/smc: separate LLC wait queues for flow and messages
Karsten Graul [Wed, 8 Jul 2020 15:05:11 +0000 (17:05 +0200)]
net/smc: separate LLC wait queues for flow and messages

There might be races in scenarios where both SMC link groups are on the
same system. Prevent that by creating separate wait queues for LLC flows
and messages. Switch to non-interruptable versions of wait_event() and
wake_up() for the llc flow waiter to make sure the waiters get control
sequentially. Fine tune the llc_flow_lock to include the assignment of
the message. Write to system log when an unexpected message was
dropped. And remove an extra indirection and use the existing local
variable lgr in smc_llc_enqueue().

Fixes: 555da9af827d ("net/smc: add event-based llc_flow framework")
Reviewed-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: atlantic: fix ip dst and ipv6 address filters
Dmitry Bogdanov [Wed, 8 Jul 2020 14:17:10 +0000 (17:17 +0300)]
net: atlantic: fix ip dst and ipv6 address filters

This patch fixes ip dst and ipv6 address filters.
There were 2 mistakes in the code, which led to the issue:
* invalid register was used for ipv4 dst address;
* incorrect write order of dwords for ipv6 addresses.

Fixes: 23e7a718a49b ("net: aquantia: add rx-flow filter definitions")
Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com>
Signed-off-by: Mark Starovoytov <mstarovoitov@marvell.com>
Signed-off-by: Alexander Lobakin <alobakin@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoionic: centralize queue reset code
Shannon Nelson [Tue, 7 Jul 2020 21:13:26 +0000 (14:13 -0700)]
ionic: centralize queue reset code

The queue reset pattern is used in a couple different places,
only slightly different from each other, and could cause
issues if one gets changed and the other didn't.  This puts
them together so that only one version is needed, yet each
can have slighty different effects by passing in a pointer
to a work function to do whatever configuration twiddling is
needed in the middle of the reset.

This specifically addresses issues seen where under loops
of changing ring size or queue count parameters we could
occasionally bump into the netdev watchdog.

v2: added more commit message commentary

Fixes: 4d03e00a2140 ("ionic: Add initial ethtool support")
Signed-off-by: Shannon Nelson <snelson@pensando.io>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agovlan: consolidate VLAN parsing code and limit max parsing depth
Toke Høiland-Jørgensen [Tue, 7 Jul 2020 11:03:25 +0000 (13:03 +0200)]
vlan: consolidate VLAN parsing code and limit max parsing depth

Toshiaki pointed out that we now have two very similar functions to extract
the L3 protocol number in the presence of VLAN tags. And Daniel pointed out
that the unbounded parsing loop makes it possible for maliciously crafted
packets to loop through potentially hundreds of tags.

Fix both of these issues by consolidating the two parsing functions and
limiting the VLAN tag parsing to a max depth of 8 tags. As part of this,
switch over __vlan_get_protocol() to use skb_header_pointer() instead of
pskb_may_pull(), to avoid the possible side effects of the latter and keep
the skb pointer 'const' through all the parsing functions.

v2:
- Use limit of 8 tags instead of 32 (matching XMIT_RECURSION_LIMIT)

Reported-by: Toshiaki Makita <toshiaki.makita1@gmail.com>
Reported-by: Daniel Borkmann <daniel@iogearbox.net>
Fixes: d7bf2ebebc2b ("sched: consistently handle layer3 header accesses in the presence of VLANs")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: qed: fix buffer overflow on ethtool -d
Alexander Lobakin [Mon, 6 Jul 2020 09:25:53 +0000 (12:25 +0300)]
net: qed: fix buffer overflow on ethtool -d

When generating debug dump, driver firstly collects all data in binary
form, and then performs per-feature formatting to human-readable if it
is supported.

For ethtool -d, this is roughly incorrect for two reasons. First of all,
drivers should always provide only original raw dumps to Ethtool without
any changes.
The second, and more critical, is that Ethtool's output buffer size is
strictly determined by ethtool_ops::get_regs_len(), and all data *must*
fit in it. The current version of driver always returns the size of raw
data, but the size of the formatted buffer exceeds it in most cases.
This leads to out-of-bound writes and memory corruption.

Address both issues by adding an option to return original, non-formatted
debug data, and using it for Ethtool case.

v2:
 - Expand commit message to make it more clear;
 - No functional changes.

Fixes: c965db444629 ("qed: Add support for debug data collection")
Signed-off-by: Alexander Lobakin <alobakin@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agobridge: mcast: Fix MLD2 Report IPv6 payload length check
Linus Lüssing [Sun, 5 Jul 2020 19:10:17 +0000 (21:10 +0200)]
bridge: mcast: Fix MLD2 Report IPv6 payload length check

Commit e57f61858b7c ("net: bridge: mcast: fix stale nsrcs pointer in
igmp3/mld2 report handling") introduced a bug in the IPv6 header payload
length check which would potentially lead to rejecting a valid MLD2 Report:

The check needs to take into account the 2 bytes for the "Number of
Sources" field in the "Multicast Address Record" before reading it.
And not the size of a pointer to this field.

Fixes: e57f61858b7c ("net: bridge: mcast: fix stale nsrcs pointer in igmp3/mld2 report handling")
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: Added pointer check for dst->ops->neigh_lookup in dst_neigh_lookup_skb
Martin Varghese [Sun, 5 Jul 2020 08:53:49 +0000 (14:23 +0530)]
net: Added pointer check for dst->ops->neigh_lookup in dst_neigh_lookup_skb

The packets from tunnel devices (eg bareudp) may have only
metadata in the dst pointer of skb. Hence a pointer check of
neigh_lookup is needed in dst_neigh_lookup_skb

Kernel crashes when packets from bareudp device is processed in
the kernel neighbour subsytem.

[  133.384484] BUG: kernel NULL pointer dereference, address: 0000000000000000
[  133.385240] #PF: supervisor instruction fetch in kernel mode
[  133.385828] #PF: error_code(0x0010) - not-present page
[  133.386603] PGD 0 P4D 0
[  133.386875] Oops: 0010 [#1] SMP PTI
[  133.387275] CPU: 0 PID: 5045 Comm: ping Tainted: G        W         5.8.0-rc2+ #15
[  133.388052] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011
[  133.391076] RIP: 0010:0x0
[  133.392401] Code: Bad RIP value.
[  133.394029] RSP: 0018:ffffb79980003d50 EFLAGS: 00010246
[  133.396656] RAX: 0000000080000102 RBX: ffff9de2fe0d6600 RCX: ffff9de2fe5e9d00
[  133.399018] RDX: 0000000000000000 RSI: ffff9de2fe5e9d00 RDI: ffff9de2fc21b400
[  133.399685] RBP: ffff9de2fe5e9d00 R08: 0000000000000000 R09: 0000000000000000
[  133.400350] R10: ffff9de2fbc6be22 R11: ffff9de2fe0d6600 R12: ffff9de2fc21b400
[  133.401010] R13: ffff9de2fe0d6628 R14: 0000000000000001 R15: 0000000000000003
[  133.401667] FS:  00007fe014918740(0000) GS:ffff9de2fec00000(0000) knlGS:0000000000000000
[  133.402412] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  133.402948] CR2: ffffffffffffffd6 CR3: 000000003bb72000 CR4: 00000000000006f0
[  133.403611] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  133.404270] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  133.404933] Call Trace:
[  133.405169]  <IRQ>
[  133.405367]  __neigh_update+0x5a4/0x8f0
[  133.405734]  arp_process+0x294/0x820
[  133.406076]  ? __netif_receive_skb_core+0x866/0xe70
[  133.406557]  arp_rcv+0x129/0x1c0
[  133.406882]  __netif_receive_skb_one_core+0x95/0xb0
[  133.407340]  process_backlog+0xa7/0x150
[  133.407705]  net_rx_action+0x2af/0x420
[  133.408457]  __do_softirq+0xda/0x2a8
[  133.408813]  asm_call_on_stack+0x12/0x20
[  133.409290]  </IRQ>
[  133.409519]  do_softirq_own_stack+0x39/0x50
[  133.410036]  do_softirq+0x50/0x60
[  133.410401]  __local_bh_enable_ip+0x50/0x60
[  133.410871]  ip_finish_output2+0x195/0x530
[  133.411288]  ip_output+0x72/0xf0
[  133.411673]  ? __ip_finish_output+0x1f0/0x1f0
[  133.412122]  ip_send_skb+0x15/0x40
[  133.412471]  raw_sendmsg+0x853/0xab0
[  133.412855]  ? insert_pfn+0xfe/0x270
[  133.413827]  ? vvar_fault+0xec/0x190
[  133.414772]  sock_sendmsg+0x57/0x80
[  133.415685]  __sys_sendto+0xdc/0x160
[  133.416605]  ? syscall_trace_enter+0x1d4/0x2b0
[  133.417679]  ? __audit_syscall_exit+0x1d9/0x280
[  133.418753]  ? __prepare_exit_to_usermode+0x5d/0x1a0
[  133.419819]  __x64_sys_sendto+0x24/0x30
[  133.420848]  do_syscall_64+0x4d/0x90
[  133.421768]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
[  133.422833] RIP: 0033:0x7fe013689c03
[  133.423749] Code: Bad RIP value.
[  133.424624] RSP: 002b:00007ffc7288f418 EFLAGS: 00000246 ORIG_RAX: 000000000000002c
[  133.425940] RAX: ffffffffffffffda RBX: 000056151fc63720 RCX: 00007fe013689c03
[  133.427225] RDX: 0000000000000040 RSI: 000056151fc63720 RDI: 0000000000000003
[  133.428481] RBP: 00007ffc72890b30 R08: 000056151fc60500 R09: 0000000000000010
[  133.429757] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000040
[  133.431041] R13: 000056151fc636e0 R14: 000056151fc616bc R15: 0000000000000080
[  133.432481] Modules linked in: mpls_iptunnel act_mirred act_tunnel_key cls_flower sch_ingress veth mpls_router ip_tunnel bareudp ip6_udp_tunnel udp_tunnel macsec udp_diag inet_diag unix_diag af_packet_diag netlink_diag binfmt_misc xt_MASQUERADE iptable_nat xt_addrtype xt_conntrack nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 br_netfilter bridge stp llc ebtable_filter ebtables overlay ip6table_filter ip6_tables iptable_filter sunrpc ext4 mbcache jbd2 pcspkr i2c_piix4 virtio_balloon joydev ip_tables xfs libcrc32c ata_generic qxl pata_acpi drm_ttm_helper ttm drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops drm ata_piix libata virtio_net net_failover virtio_console failover virtio_blk i2c_core virtio_pci virtio_ring serio_raw floppy virtio dm_mirror dm_region_hash dm_log dm_mod
[  133.444045] CR2: 0000000000000000
[  133.445082] ---[ end trace f4aeee1958fd1638 ]---
[  133.446236] RIP: 0010:0x0
[  133.447180] Code: Bad RIP value.
[  133.448152] RSP: 0018:ffffb79980003d50 EFLAGS: 00010246
[  133.449363] RAX: 0000000080000102 RBX: ffff9de2fe0d6600 RCX: ffff9de2fe5e9d00
[  133.450835] RDX: 0000000000000000 RSI: ffff9de2fe5e9d00 RDI: ffff9de2fc21b400
[  133.452237] RBP: ffff9de2fe5e9d00 R08: 0000000000000000 R09: 0000000000000000
[  133.453722] R10: ffff9de2fbc6be22 R11: ffff9de2fe0d6600 R12: ffff9de2fc21b400
[  133.455149] R13: ffff9de2fe0d6628 R14: 0000000000000001 R15: 0000000000000003
[  133.456520] FS:  00007fe014918740(0000) GS:ffff9de2fec00000(0000) knlGS:0000000000000000
[  133.458046] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  133.459342] CR2: ffffffffffffffd6 CR3: 000000003bb72000 CR4: 00000000000006f0
[  133.460782] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  133.462240] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  133.463697] Kernel panic - not syncing: Fatal exception in interrupt
[  133.465226] Kernel Offset: 0xfa00000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
[  133.467025] ---[ end Kernel panic - not syncing: Fatal exception in interrupt ]---

Fixes: aaa0c23cb901 ("Fix dst_neigh_lookup/dst_neigh_lookup_skb return value handling bug")
Signed-off-by: Martin Varghese <martin.varghese@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/sched: act_ct: add miss tcf_lastuse_update.
wenxu [Sat, 4 Jul 2020 07:42:47 +0000 (15:42 +0800)]
net/sched: act_ct: add miss tcf_lastuse_update.

When tcf_ct_act execute the tcf_lastuse_update should
be update or the used stats never update

filter protocol ip pref 3 flower chain 0
filter protocol ip pref 3 flower chain 0 handle 0x1
  eth_type ipv4
  dst_ip 1.1.1.1
  ip_flags frag/firstfrag
  skip_hw
  not_in_hw
 action order 1: ct zone 1 nat pipe
  index 1 ref 1 bind 1 installed 103 sec used 103 sec
 Action statistics:
 Sent 151500 bytes 101 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0
 cookie 4519c04dc64a1a295787aab13b6a50fb

Signed-off-by: wenxu <wenxu@ucloud.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/mlx5e: Do not include rwlock.h directly
Sebastian Andrzej Siewior [Fri, 3 Jul 2020 16:44:32 +0000 (18:44 +0200)]
net/mlx5e: Do not include rwlock.h directly

rwlock.h should not be included directly. Instead linux/splinlock.h
should be included. Including it directly will break the RT build.

Fixes: 549c243e4e010 ("net/mlx5e: Extract neigh-specific code from en_rep.c to rep/neigh.c")
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomptcp: fix DSS map generation on fin retransmission
Paolo Abeni [Fri, 3 Jul 2020 16:06:04 +0000 (18:06 +0200)]
mptcp: fix DSS map generation on fin retransmission

The RFC 8684 mandates that no-data DATA FIN packets should carry
a DSS with 0 sequence number and data len equal to 1. Currently,
on FIN retransmission we re-use the existing mapping; if the previous
fin transmission was part of a partially acked data packet, we could
end-up writing in the egress packet a non-compliant DSS.

The above will be detected by a "Bad mapping" warning on the receiver
side.

This change addresses the issue explicitly checking for 0 len packet
when adding the DATA_FIN option.

Fixes: 6d0060f600ad ("mptcp: Write MPTCP DSS headers to outgoing data packets")
Reported-by: syzbot+42a07faa5923cfaeb9c9@syzkaller.appspotmail.com
Tested-by: Christoph Paasch <cpaasch@apple.com>
Reviewed-by: Christoph Paasch <cpaasch@apple.com>
Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoipv4: fill fl4_icmp_{type,code} in ping_v4_sendmsg
Sabrina Dubroca [Fri, 3 Jul 2020 15:00:32 +0000 (17:00 +0200)]
ipv4: fill fl4_icmp_{type,code} in ping_v4_sendmsg

IPv4 ping sockets don't set fl4.fl4_icmp_{type,code}, which leads to
incomplete IPsec ACQUIRE messages being sent to userspace. Currently,
both raw sockets and IPv6 ping sockets set those fields.

Expected output of "ip xfrm monitor":
    acquire proto esp
      sel src 10.0.2.15/32 dst 8.8.8.8/32 proto icmp type 8 code 0 dev ens4
      policy src 10.0.2.15/32 dst 8.8.8.8/32
        <snip>

Currently with ping sockets:
    acquire proto esp
      sel src 10.0.2.15/32 dst 8.8.8.8/32 proto icmp type 0 code 0 dev ens4
      policy src 10.0.2.15/32 dst 8.8.8.8/32
        <snip>

The Libreswan test suite found this problem after Fedora changed the
value for the sysctl net.ipv4.ping_group_range.

Fixes: c319b4d76b9e ("net: ipv4: add IPPROTO_ICMP socket kind")
Reported-by: Paul Wouters <pwouters@redhat.com>
Tested-by: Paul Wouters <pwouters@redhat.com>
Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ethernet: fec: prevent tx starvation under high rx load
Tobias Waldekranz [Fri, 3 Jul 2020 14:10:58 +0000 (16:10 +0200)]
net: ethernet: fec: prevent tx starvation under high rx load

In the ISR, we poll the event register for the queues in need of
service and then enter polled mode. After this point, the event
register will never be read again until we exit polled mode.

In a scenario where a UDP flow is routed back out through the same
interface, i.e. "router-on-a-stick" we'll typically only see an rx
queue event initially. Once we start to process the incoming flow
we'll be locked polled mode, but we'll never clean the tx rings since
that event is never caught.

Eventually the netdev watchdog will trip, causing all buffers to be
dropped and then the process starts over again.

Rework the NAPI poll to keep trying to consome the entire budget as
long as new events are coming in, making sure to service all rx/tx
queues, in priority order, on each pass.

Fixes: 4d494cdc92b3 ("net: fec: change data structure to support multiqueue")
Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
Tested-by: Fugang Duan <fugang.duan@nxp.com>
Reviewed-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: sky2: initialize return of gm_phy_read
Tom Rix [Fri, 3 Jul 2020 13:33:59 +0000 (06:33 -0700)]
net: sky2: initialize return of gm_phy_read

clang static analysis flags this garbage return

drivers/net/ethernet/marvell/sky2.c:208:2: warning: Undefined or garbage value returned to caller [core.uninitialized.UndefReturn]
        return v;
        ^~~~~~~~

static inline u16 gm_phy_read( ...
{
u16 v;
__gm_phy_read(hw, port, reg, &v);
return v;
}

__gm_phy_read can return without setting v.

So handle similar to skge.c's gm_phy_read, initialize v.

Signed-off-by: Tom Rix <trix@redhat.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agocgroup: fix cgroup_sk_alloc() for sk_clone_lock()
Cong Wang [Thu, 2 Jul 2020 18:52:56 +0000 (11:52 -0700)]
cgroup: fix cgroup_sk_alloc() for sk_clone_lock()

When we clone a socket in sk_clone_lock(), its sk_cgrp_data is
copied, so the cgroup refcnt must be taken too. And, unlike the
sk_alloc() path, sock_update_netprioidx() is not called here.
Therefore, it is safe and necessary to grab the cgroup refcnt
even when cgroup_sk_alloc is disabled.

sk_clone_lock() is in BH context anyway, the in_interrupt()
would terminate this function if called there. And for sk_alloc()
skcd->val is always zero. So it's safe to factor out the code
to make it more readable.

The global variable 'cgroup_sk_alloc_disabled' is used to determine
whether to take these reference counts. It is impossible to make
the reference counting correct unless we save this bit of information
in skcd->val. So, add a new bit there to record whether the socket
has already taken the reference counts. This obviously relies on
kmalloc() to align cgroup pointers to at least 4 bytes,
ARCH_KMALLOC_MINALIGN is certainly larger than that.

This bug seems to be introduced since the beginning, commit
d979a39d7242 ("cgroup: duplicate cgroup reference when cloning sockets")
tried to fix it but not compeletely. It seems not easy to trigger until
the recent commit 090e28b229af
("netprio_cgroup: Fix unlimited memory leak of v2 cgroups") was merged.

Fixes: bd1060a1d671 ("sock, cgroup: add sock->sk_cgroup")
Reported-by: Cameron Berkenpas <cam@neo-zeon.de>
Reported-by: Peter Geis <pgwipeout@gmail.com>
Reported-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reported-by: Daniël Sonck <dsonck92@gmail.com>
Reported-by: Zhang Qiang <qiang.zhang@windriver.com>
Tested-by: Cameron Berkenpas <cam@neo-zeon.de>
Tested-by: Peter Geis <pgwipeout@gmail.com>
Tested-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Zefan Li <lizefan@huawei.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Roman Gushchin <guro@fb.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoipv6: Fix use of anycast address with loopback
David Ahern [Tue, 7 Jul 2020 13:39:24 +0000 (07:39 -0600)]
ipv6: Fix use of anycast address with loopback

Thomas reported a regression with IPv6 and anycast using the following
reproducer:

    echo 1 >  /proc/sys/net/ipv6/conf/all/forwarding
    ip -6 a add fc12::1/16 dev lo
    sleep 2
    echo "pinging lo"
    ping6 -c 2 fc12::

The conversion of addrconf_f6i_alloc to use ip6_route_info_create missed
the use of fib6_is_reject which checks addresses added to the loopback
interface and sets the REJECT flag as needed. Update fib6_is_reject for
loopback checks to handle RTF_ANYCAST addresses.

Fixes: c7a1ce397ada ("ipv6: Change addrconf_f6i_alloc to use ip6_route_info_create")
Reported-by: thomas.gambier@nexedi.com
Signed-off-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: usb: qmi_wwan: add support for Quectel EG95 LTE modem
AceLan Kao [Tue, 7 Jul 2020 08:14:45 +0000 (16:14 +0800)]
net: usb: qmi_wwan: add support for Quectel EG95 LTE modem

Add support for Quectel Wireless Solutions Co., Ltd. EG95 LTE modem

T:  Bus=01 Lev=01 Prnt=01 Port=02 Cnt=02 Dev#=  5 Spd=480 MxCh= 0
D:  Ver= 2.00 Cls=ef(misc ) Sub=02 Prot=01 MxPS=64 #Cfgs=  1
P:  Vendor=2c7c ProdID=0195 Rev=03.18
S:  Manufacturer=Android
S:  Product=Android
C:  #Ifs= 5 Cfg#= 1 Atr=a0 MxPwr=500mA
I:  If#=0x0 Alt= 0 #EPs= 2 Cls=ff(vend.) Sub=ff Prot=ff Driver=(none)
I:  If#=0x1 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=00 Prot=00 Driver=(none)
I:  If#=0x2 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=00 Prot=00 Driver=(none)
I:  If#=0x3 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=00 Prot=00 Driver=(none)
I:  If#=0x4 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=ff Driver=(none)

Signed-off-by: AceLan Kao <acelan.kao@canonical.com>
Acked-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-ipa-fix-warning-reported-errors'
David S. Miller [Tue, 7 Jul 2020 19:43:18 +0000 (12:43 -0700)]
Merge branch 'net-ipa-fix-warning-reported-errors'

Alex Elder says:

====================
net: ipa: fix warning-reported errors

Building the kernel with W=1 produces numerous warnings for the IPA
code.  Some of those warnings turn out to flag real problems, and
this series fixes them.  The first patch fixes the most important
ones, but the second and third are problems I think are worth
treating as bugs as well.

Note:  I'll happily combine any of these if someone prefers that.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: include declarations in "ipa_gsi.c"
Alex Elder [Mon, 6 Jul 2020 23:10:10 +0000 (18:10 -0500)]
net: ipa: include declarations in "ipa_gsi.c"

Include "ipa_gsi.h" in "ipa_gsi.c", so the public functions are
defined before they are used in "ipa_gsi.c".  This addresses some
warnings that are reported with a "W=1" build.

Fixes: c3f398b141a8 ("soc: qcom: ipa: IPA interface to GSI")
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: declare struct types in "ipa_gsi.h"
Alex Elder [Mon, 6 Jul 2020 23:10:09 +0000 (18:10 -0500)]
net: ipa: declare struct types in "ipa_gsi.h"

Pointers to two struct types are used in "ipa_gsi.h", without those
struct types being forward-declared.  Add these declarations.

Fixes: c3f398b141a8 ("soc: qcom: ipa: IPA interface to GSI")
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: ipa: fix QMI structure definition bugs
Alex Elder [Mon, 6 Jul 2020 23:10:08 +0000 (18:10 -0500)]
net: ipa: fix QMI structure definition bugs

Building with "W=1" did exactly what it was supposed to do, namely
point out some suspicious-looking code to be verified not to contain
bugs.

Some QMI message structures defined in "ipa_qmi_msg.c" contained
some bad field names (duplicating the "elem_size" field instead of
defining the "offset" field), almost certainly due to copy/paste
errors that weren't obvious in a scan of the code.  Fix these bugs.

Fixes: 530f9216a953 ("soc: qcom: ipa: AP/modem communications")
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoipv6: fib6_select_path can not use out path for nexthop objects
David Ahern [Mon, 6 Jul 2020 17:45:07 +0000 (11:45 -0600)]
ipv6: fib6_select_path can not use out path for nexthop objects

Brian reported a crash in IPv6 code when using rpfilter with a setup
running FRR and external nexthop objects. The root cause of the crash
is fib6_select_path setting fib6_nh in the result to NULL because of
an improper check for nexthop objects.

More specifically, rpfilter invokes ip6_route_lookup with flowi6_oif
set causing fib6_select_path to be called with have_oif_match set.
fib6_select_path has early check on have_oif_match and jumps to the
out label which presumes a builtin fib6_nh. This path is invalid for
nexthop objects; for external nexthops fib6_select_path needs to just
return if the fib6_nh has already been set in the result otherwise it
returns after the call to nexthop_path_fib6_result. Update the check
on have_oif_match to not bail on external nexthops.

Update selftests for this problem.

Fixes: f88d8ea67fbd ("ipv6: Plumb support for nexthop object in a fib6_info")
Reported-by: Brian Rak <brak@choopa.com>
Signed-off-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'hns3-fixes'
David S. Miller [Mon, 6 Jul 2020 19:33:28 +0000 (12:33 -0700)]
Merge branch 'hns3-fixes'

Huazhong Tan says:

====================
net: hns3: fixes for -net

There are some fixes about reset issue and a use-after-free
of self-test.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: fix use-after-free when doing self test
Yonglong Liu [Mon, 6 Jul 2020 11:26:02 +0000 (19:26 +0800)]
net: hns3: fix use-after-free when doing self test

Enable promisc mode of PF, set VF link state to enable, and
run iperf of the VF, then do self test of the PF. The self test
will fail with a low frequency, and may cause a use-after-free
problem.

[   87.142126] selftest:000004a0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
[   87.159722] ==================================================================
[   87.174187] BUG: KASAN: use-after-free in hex_dump_to_buffer+0x140/0x608
[   87.187600] Read of size 1 at addr ffff003b22828000 by task ethtool/1186
[   87.201012]
[   87.203978] CPU: 7 PID: 1186 Comm: ethtool Not tainted 5.5.0-rc4-gfd51c473-dirty #4
[   87.219306] Hardware name: Huawei TaiShan 2280 V2/BC82AMDA, BIOS TA BIOS 2280-A CS V2.B160.01 01/15/2020
[   87.238292] Call trace:
[   87.243173]  dump_backtrace+0x0/0x280
[   87.250491]  show_stack+0x24/0x30
[   87.257114]  dump_stack+0xe8/0x140
[   87.263911]  print_address_description.isra.8+0x70/0x380
[   87.274538]  __kasan_report+0x12c/0x230
[   87.282203]  kasan_report+0xc/0x18
[   87.288999]  __asan_load1+0x60/0x68
[   87.295969]  hex_dump_to_buffer+0x140/0x608
[   87.304332]  print_hex_dump+0x140/0x1e0
[   87.312000]  hns3_lb_check_skb_data+0x168/0x170
[   87.321060]  hns3_clean_rx_ring+0xa94/0xfe0
[   87.329422]  hns3_self_test+0x708/0x8c0

The length of packet sent by the selftest process is only
128 + 14 bytes, and the min buffer size of a BD is 256 bytes,
and the receive process will make sure the packet sent by
the selftest process is in the linear part, so only check
the linear part in hns3_lb_check_skb_data().

So fix this use-after-free by using skb_headlen() to dump
skb->data instead of skb->len.

Fixes: c39c4d98dc65 ("net: hns3: Add mac loopback selftest support in hns3 driver")
Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: add a missing uninit debugfs when unload driver
Huazhong Tan [Mon, 6 Jul 2020 11:26:01 +0000 (19:26 +0800)]
net: hns3: add a missing uninit debugfs when unload driver

When unloading driver, if flag HNS3_NIC_STATE_INITED has been
already cleared, the debugfs will not be uninitialized, so fix it.

Fixes: b2292360bb2a ("net: hns3: Add debugfs framework registration")
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: fix for mishandle of asserting VF reset fail
Huazhong Tan [Mon, 6 Jul 2020 11:26:00 +0000 (19:26 +0800)]
net: hns3: fix for mishandle of asserting VF reset fail

When asserts VF reset fail, flag HCLGEVF_STATE_CMD_DISABLE
and handshake status should not set, otherwise the retry will
fail. So adds a check for asserting VF reset and returns
directly when fails.

Fixes: ef5f8e507ec9 ("net: hns3: stop handling command queue while resetting VF")
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: check reset pending after FLR prepare
Huazhong Tan [Mon, 6 Jul 2020 11:25:59 +0000 (19:25 +0800)]
net: hns3: check reset pending after FLR prepare

If there is a PF reset pending before FLR prepare, FLR's
preparatory work will not fail, but the FLR rebuild procedure
will fail for this pending. So this PF reset pending should
be handled in the FLR preparatory.

Fixes: 8627bdedc435 ("net: hns3: refactor the precedure of PF FLR")
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'smsc95xx-fix-smsc95xx_bind'
David S. Miller [Mon, 6 Jul 2020 19:22:24 +0000 (12:22 -0700)]
Merge branch 'smsc95xx-fix-smsc95xx_bind'

Andre Edich says:

====================
smsc95xx: fix smsc95xx_bind

The patchset fixes two problems in the function smsc95xx_bind:
 - return of false success
 - memory leak

Changes in v2:
- added "Fixes" tags to both patches
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosmsc95xx: avoid memory leak in smsc95xx_bind
Andre Edich [Mon, 6 Jul 2020 08:39:35 +0000 (10:39 +0200)]
smsc95xx: avoid memory leak in smsc95xx_bind

In a case where the ID_REV register read is failed, the memory for a
private data structure has to be freed before returning error from the
function smsc95xx_bind.

Fixes: bbd9f9ee69242 ("smsc95xx: add wol support for more frame types")
Signed-off-by: Andre Edich <andre.edich@microchip.com>
Signed-off-by: Parthiban Veerasooran <Parthiban.Veerasooran@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosmsc95xx: check return value of smsc95xx_reset
Andre Edich [Mon, 6 Jul 2020 08:39:34 +0000 (10:39 +0200)]
smsc95xx: check return value of smsc95xx_reset

The return value of the function smsc95xx_reset() must be checked
to avoid returning false success from the function smsc95xx_bind().

Fixes: 2f7ca802bdae2 ("net: Add SMSC LAN9500 USB2.0 10/100 ethernet adapter driver")
Signed-off-by: Andre Edich <andre.edich@microchip.com>
Signed-off-by: Parthiban Veerasooran <Parthiban.Veerasooran@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodrivers/net/wan/lapbether: Fixed the value of hard_header_len
Xie He [Mon, 6 Jul 2020 00:45:21 +0000 (17:45 -0700)]
drivers/net/wan/lapbether: Fixed the value of hard_header_len

When this driver transmits data,
  first this driver will remove a pseudo header of 1 byte,
  then the lapb module will prepend the LAPB header of 2 or 3 bytes,
  then this driver will prepend a length field of 2 bytes,
  then the underlying Ethernet device will prepend its own header.

So, the header length required should be:
  -1 + 3 + 2 + "the header length needed by the underlying device".

This patch fixes kernel panic when this driver is used with AF_PACKET
SOCK_DGRAM sockets.

Signed-off-by: Xie He <xie.he.0141@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-rmnet-fix-interface-leak-for-rmnet-module'
David S. Miller [Sun, 5 Jul 2020 01:04:56 +0000 (18:04 -0700)]
Merge branch 'net-rmnet-fix-interface-leak-for-rmnet-module'

Taehee Yoo says:

====================
net: rmnet: fix interface leak for rmnet module

There are two problems in rmnet module that they occur the leak of
a lower interface.
The symptom is the same, which is the leak of a lower interface.
But there are two different real problems.
This patchset is to fix these real problems.

1. Do not allow to have different two modes.
As a lower interface of rmnet, there are two modes that they are VND
and BRIDGE.
One interface can have only one mode.
But in the current rmnet, there is no code to prevent to have
two modes in one lower interface.
So, interface leak occurs.

2. Do not allow to add multiple bridge interfaces.
rmnet can have only two bridge interface.
If an additional bridge interface is tried to be attached,
rmnet should deny it.
But there is no code to do that.
So, interface leak occurs.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rmnet: do not allow to add multiple bridge interfaces
Taehee Yoo [Thu, 2 Jul 2020 17:08:55 +0000 (17:08 +0000)]
net: rmnet: do not allow to add multiple bridge interfaces

rmnet can have only two bridge interface.
One of them is a link interface and another one is added by
the master operation.
rmnet interface shouldn't allow adding additional
bridge interfaces by mater operation.
But, there is no code to deny additional interfaces.
So, interface leak occurs.

Test commands:
    ip link add dummy0 type dummy
    ip link add dummy1 type dummy
    ip link add dummy2 type dummy
    ip link add rmnet0 link dummy0 type rmnet mux_id 1
    ip link set dummy1 master rmnet0
    ip link set dummy2 master rmnet0
    ip link del rmnet0

In the above test command, the dummy0 was attached to rmnet as VND mode.
Then, dummy1 was attached to rmnet0 as BRIDGE mode.
At this point, dummy0 mode is switched from VND to BRIDGE automatically.
Then, dummy2 is attached to rmnet as BRIDGE mode.
At this point, rmnet0 should deny this operation.
But, rmnet0 doesn't deny this.
So that below splat occurs when the rmnet0 interface is deleted.

Splat looks like:
[  186.684787][    C2] WARNING: CPU: 2 PID: 1009 at net/core/dev.c:8992 rollback_registered_many+0x986/0xcf0
[  186.684788][    C2] Modules linked in: rmnet dummy openvswitch nsh nf_conncount nf_nat nf_conntrack nf_defrag_x
[  186.684805][    C2] CPU: 2 PID: 1009 Comm: ip Not tainted 5.8.0-rc1+ #621
[  186.684807][    C2] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[  186.684808][    C2] RIP: 0010:rollback_registered_many+0x986/0xcf0
[  186.684811][    C2] Code: 41 8b 4e cc 45 31 c0 31 d2 4c 89 ee 48 89 df e8 e0 47 ff ff 85 c0 0f 84 cd fc ff ff 5
[  186.684812][    C2] RSP: 0018:ffff8880cd9472e0 EFLAGS: 00010287
[  186.684815][    C2] RAX: ffff8880cc56da58 RBX: ffff8880ab21c000 RCX: ffffffff9329d323
[  186.684816][    C2] RDX: 1ffffffff2be6410 RSI: 0000000000000008 RDI: ffffffff95f32080
[  186.684818][    C2] RBP: dffffc0000000000 R08: fffffbfff2be6411 R09: fffffbfff2be6411
[  186.684819][    C2] R10: ffffffff95f32087 R11: 0000000000000001 R12: ffff8880cd947480
[  186.684820][    C2] R13: ffff8880ab21c0b8 R14: ffff8880cd947400 R15: ffff8880cdf10640
[  186.684822][    C2] FS:  00007f00843890c0(0000) GS:ffff8880d4e00000(0000) knlGS:0000000000000000
[  186.684823][    C2] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  186.684825][    C2] CR2: 000055b8ab1077b8 CR3: 00000000ab612006 CR4: 00000000000606e0
[  186.684826][    C2] Call Trace:
[  186.684827][    C2]  ? lockdep_hardirqs_on_prepare+0x379/0x540
[  186.684829][    C2]  ? netif_set_real_num_tx_queues+0x780/0x780
[  186.684830][    C2]  ? rmnet_unregister_real_device+0x56/0x90 [rmnet]
[  186.684831][    C2]  ? __kasan_slab_free+0x126/0x150
[  186.684832][    C2]  ? kfree+0xdc/0x320
[  186.684834][    C2]  ? rmnet_unregister_real_device+0x56/0x90 [rmnet]
[  186.684835][    C2]  unregister_netdevice_many.part.135+0x13/0x1b0
[  186.684836][    C2]  rtnl_delete_link+0xbc/0x100
[ ... ]
[  238.440071][ T1009] unregister_netdevice: waiting for rmnet0 to become free. Usage count = 1

Fixes: 037f9cdf72fb ("net: rmnet: use upper/lower device infrastructure")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rmnet: fix lower interface leak
Taehee Yoo [Thu, 2 Jul 2020 17:08:18 +0000 (17:08 +0000)]
net: rmnet: fix lower interface leak

There are two types of the lower interface of rmnet that are VND
and BRIDGE.
Each lower interface can have only one type either VND or BRIDGE.
But, there is a case, which uses both lower interface types.
Due to this unexpected behavior, lower interface leak occurs.

Test commands:
    ip link add dummy0 type dummy
    ip link add dummy1 type dummy
    ip link add rmnet0 link dummy0 type rmnet mux_id 1
    ip link set dummy1 master rmnet0
    ip link add rmnet1 link dummy1 type rmnet mux_id 2
    ip link del rmnet0

The dummy1 was attached as BRIDGE interface of rmnet0.
Then, it also was attached as VND interface of rmnet1.
This is unexpected behavior and there is no code for handling this case.
So that below splat occurs when the rmnet0 interface is deleted.

Splat looks like:
[   53.254112][    C1] WARNING: CPU: 1 PID: 1192 at net/core/dev.c:8992 rollback_registered_many+0x986/0xcf0
[   53.254117][    C1] Modules linked in: rmnet dummy openvswitch nsh nf_conncount nf_nat nf_conntrack nf_defrag_ipv6 nfx
[   53.254182][    C1] CPU: 1 PID: 1192 Comm: ip Not tainted 5.8.0-rc1+ #620
[   53.254188][    C1] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[   53.254192][    C1] RIP: 0010:rollback_registered_many+0x986/0xcf0
[   53.254200][    C1] Code: 41 8b 4e cc 45 31 c0 31 d2 4c 89 ee 48 89 df e8 e0 47 ff ff 85 c0 0f 84 cd fc ff ff 0f 0b e5
[   53.254205][    C1] RSP: 0018:ffff888050a5f2e0 EFLAGS: 00010287
[   53.254214][    C1] RAX: ffff88805756d658 RBX: ffff88804d99c000 RCX: ffffffff8329d323
[   53.254219][    C1] RDX: 1ffffffff0be6410 RSI: 0000000000000008 RDI: ffffffff85f32080
[   53.254223][    C1] RBP: dffffc0000000000 R08: fffffbfff0be6411 R09: fffffbfff0be6411
[   53.254228][    C1] R10: ffffffff85f32087 R11: 0000000000000001 R12: ffff888050a5f480
[   53.254233][    C1] R13: ffff88804d99c0b8 R14: ffff888050a5f400 R15: ffff8880548ebe40
[   53.254238][    C1] FS:  00007f6b86b370c0(0000) GS:ffff88806c200000(0000) knlGS:0000000000000000
[   53.254243][    C1] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   53.254248][    C1] CR2: 0000562c62438758 CR3: 000000003f600005 CR4: 00000000000606e0
[   53.254253][    C1] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   53.254257][    C1] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[   53.254261][    C1] Call Trace:
[   53.254266][    C1]  ? lockdep_hardirqs_on_prepare+0x379/0x540
[   53.254270][    C1]  ? netif_set_real_num_tx_queues+0x780/0x780
[   53.254275][    C1]  ? rmnet_unregister_real_device+0x56/0x90 [rmnet]
[   53.254279][    C1]  ? __kasan_slab_free+0x126/0x150
[   53.254283][    C1]  ? kfree+0xdc/0x320
[   53.254288][    C1]  ? rmnet_unregister_real_device+0x56/0x90 [rmnet]
[   53.254293][    C1]  unregister_netdevice_many.part.135+0x13/0x1b0
[   53.254297][    C1]  rtnl_delete_link+0xbc/0x100
[   53.254301][    C1]  ? rtnl_af_register+0xc0/0xc0
[   53.254305][    C1]  rtnl_dellink+0x2dc/0x840
[   53.254309][    C1]  ? find_held_lock+0x39/0x1d0
[   53.254314][    C1]  ? valid_fdb_dump_strict+0x620/0x620
[   53.254318][    C1]  ? rtnetlink_rcv_msg+0x457/0x890
[   53.254322][    C1]  ? lock_contended+0xd20/0xd20
[   53.254326][    C1]  rtnetlink_rcv_msg+0x4a8/0x890
[ ... ]
[   73.813696][ T1192] unregister_netdevice: waiting for rmnet0 to become free. Usage count = 1

Fixes: 037f9cdf72fb ("net: rmnet: use upper/lower device infrastructure")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agohsr: fix interface leak in error path of hsr_dev_finalize()
Taehee Yoo [Thu, 2 Jul 2020 17:06:19 +0000 (17:06 +0000)]
hsr: fix interface leak in error path of hsr_dev_finalize()

To release hsr(upper) interface, it should release
its own lower interfaces first.
Then, hsr(upper) interface can be released safely.
In the current code of error path of hsr_dev_finalize(), it releases hsr
interface before releasing a lower interface.
So, a warning occurs, which warns about the leak of lower interfaces.
In order to fix this problem, changing the ordering of the error path of
hsr_dev_finalize() is needed.

Test commands:
    ip link add dummy0 type dummy
    ip link add dummy1 type dummy
    ip link add dummy2 type dummy
    ip link add hsr0 type hsr slave1 dummy0 slave2 dummy1
    ip link add hsr1 type hsr slave1 dummy2 slave2 dummy0

Splat looks like:
[  214.923127][    C2] WARNING: CPU: 2 PID: 1093 at net/core/dev.c:8992 rollback_registered_many+0x986/0xcf0
[  214.923129][    C2] Modules linked in: hsr dummy openvswitch nsh nf_conncount nf_nat nf_conntrack nf_defrag_ipx
[  214.923154][    C2] CPU: 2 PID: 1093 Comm: ip Not tainted 5.8.0-rc2+ #623
[  214.923156][    C2] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[  214.923157][    C2] RIP: 0010:rollback_registered_many+0x986/0xcf0
[  214.923160][    C2] Code: 41 8b 4e cc 45 31 c0 31 d2 4c 89 ee 48 89 df e8 e0 47 ff ff 85 c0 0f 84 cd fc ff ff 5
[  214.923162][    C2] RSP: 0018:ffff8880c5156f28 EFLAGS: 00010287
[  214.923165][    C2] RAX: ffff8880d1dad458 RBX: ffff8880bd1b9000 RCX: ffffffffb929d243
[  214.923167][    C2] RDX: 1ffffffff77e63f0 RSI: 0000000000000008 RDI: ffffffffbbf31f80
[  214.923168][    C2] RBP: dffffc0000000000 R08: fffffbfff77e63f1 R09: fffffbfff77e63f1
[  214.923170][    C2] R10: ffffffffbbf31f87 R11: 0000000000000001 R12: ffff8880c51570a0
[  214.923172][    C2] R13: ffff8880bd1b90b8 R14: ffff8880c5157048 R15: ffff8880d1dacc40
[  214.923174][    C2] FS:  00007fdd257a20c0(0000) GS:ffff8880da200000(0000) knlGS:0000000000000000
[  214.923175][    C2] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  214.923177][    C2] CR2: 00007ffd78beb038 CR3: 00000000be544005 CR4: 00000000000606e0
[  214.923179][    C2] Call Trace:
[  214.923180][    C2]  ? netif_set_real_num_tx_queues+0x780/0x780
[  214.923182][    C2]  ? dev_validate_mtu+0x140/0x140
[  214.923183][    C2]  ? synchronize_rcu.part.79+0x85/0xd0
[  214.923185][    C2]  ? synchronize_rcu_expedited+0xbb0/0xbb0
[  214.923187][    C2]  rollback_registered+0xc8/0x170
[  214.923188][    C2]  ? rollback_registered_many+0xcf0/0xcf0
[  214.923190][    C2]  unregister_netdevice_queue+0x18b/0x240
[  214.923191][    C2]  hsr_dev_finalize+0x56e/0x6e0 [hsr]
[  214.923192][    C2]  hsr_newlink+0x36b/0x450 [hsr]
[  214.923194][    C2]  ? hsr_dellink+0x70/0x70 [hsr]
[  214.923195][    C2]  ? rtnl_create_link+0x2e4/0xb00
[  214.923197][    C2]  ? __netlink_ns_capable+0xc3/0xf0
[  214.923198][    C2]  __rtnl_newlink+0xbdb/0x1270
[ ... ]

Fixes: e0a4b99773d3 ("hsr: use upper/lower device infrastructure")
Reported-by: syzbot+7f1c020f68dab95aab59@syzkaller.appspotmail.com
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agohinic: fix sending mailbox timeout in aeq event work
Luo bin [Sat, 4 Jul 2020 07:32:43 +0000 (15:32 +0800)]
hinic: fix sending mailbox timeout in aeq event work

When sending mailbox in the work of aeq event, another aeq event
will be triggered. because the last aeq work is not exited and only
one work can be excuted simultaneously in the same workqueue, mailbox
sending function will return failure of timeout. We create and use
another workqueue to fix this.

Signed-off-by: Luo bin <luobin9@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf
David S. Miller [Sun, 5 Jul 2020 00:47:35 +0000 (17:47 -0700)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf

Pablo Neira Ayuso says:

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

The following patchset contains Netfilter fixes for net:

1) Use kvfree() to release vmalloc()'ed areas in ipset, from Eric Dumazet.

2) UAF in nfnetlink_queue from the nf_conntrack_update() path.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'Documentation-networking-eliminate-doubled-words'
David S. Miller [Sun, 5 Jul 2020 00:46:21 +0000 (17:46 -0700)]
Merge branch 'Documentation-networking-eliminate-doubled-words'

Randy Dunlap says:

====================
Documentation: networking: eliminate doubled words

Drop all duplicated words in Documentation/networking/ files.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoDocumentation: networking: rxrpc: drop doubled word
Randy Dunlap [Fri, 3 Jul 2020 22:41:15 +0000 (15:41 -0700)]
Documentation: networking: rxrpc: drop doubled word

Drop the doubled word "have".

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: netdev@vger.kernel.org
Cc: David Howells <dhowells@redhat.com>
Cc: linux-afs@lists.infradead.org
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoDocumentation: networking: ipvs-sysctl: drop doubled word
Randy Dunlap [Fri, 3 Jul 2020 22:41:14 +0000 (15:41 -0700)]
Documentation: networking: ipvs-sysctl: drop doubled word

Drop the doubled word "that".

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoDocumentation: networking: ip-sysctl: drop doubled word
Randy Dunlap [Fri, 3 Jul 2020 22:41:13 +0000 (15:41 -0700)]
Documentation: networking: ip-sysctl: drop doubled word

Drop the doubled word "that".

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoDocumentation: networking: dsa: drop doubled word
Randy Dunlap [Fri, 3 Jul 2020 22:41:12 +0000 (15:41 -0700)]
Documentation: networking: dsa: drop doubled word

Drop the doubled word "in".

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: netdev@vger.kernel.org
Cc: Andrew Lunn <andrew@lunn.ch>
Cc: Vivien Didelot <vivien.didelot@gmail.com>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoDocumentation: networking: can_ucan_protocol: drop doubled words
Randy Dunlap [Fri, 3 Jul 2020 22:41:11 +0000 (15:41 -0700)]
Documentation: networking: can_ucan_protocol: drop doubled words

Drop the doubled words "the" and "of".

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: netdev@vger.kernel.org
Cc: Wolfgang Grandegger <wg@grandegger.com>
Cc: Marc Kleine-Budde <mkl@pengutronix.de>
Cc: linux-can@vger.kernel.org
Acked-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoDocumentation: networking: ax25: drop doubled word
Randy Dunlap [Fri, 3 Jul 2020 22:41:10 +0000 (15:41 -0700)]
Documentation: networking: ax25: drop doubled word

Drop the doubled word "and".

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: netdev@vger.kernel.org
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-hams@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoDocumentation: networking: arcnet: drop doubled word
Randy Dunlap [Fri, 3 Jul 2020 22:41:09 +0000 (15:41 -0700)]
Documentation: networking: arcnet: drop doubled word

Drop the doubled word "to".

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosched: consistently handle layer3 header accesses in the presence of VLANs
Toke Høiland-Jørgensen [Fri, 3 Jul 2020 20:26:43 +0000 (22:26 +0200)]
sched: consistently handle layer3 header accesses in the presence of VLANs

There are a couple of places in net/sched/ that check skb->protocol and act
on the value there. However, in the presence of VLAN tags, the value stored
in skb->protocol can be inconsistent based on whether VLAN acceleration is
enabled. The commit quoted in the Fixes tag below fixed the users of
skb->protocol to use a helper that will always see the VLAN ethertype.

However, most of the callers don't actually handle the VLAN ethertype, but
expect to find the IP header type in the protocol field. This means that
things like changing the ECN field, or parsing diffserv values, stops
working if there's a VLAN tag, or if there are multiple nested VLAN
tags (QinQ).

To fix this, change the helper to take an argument that indicates whether
the caller wants to skip the VLAN tags or not. When skipping VLAN tags, we
make sure to skip all of them, so behaviour is consistent even in QinQ
mode.

To make the helper usable from the ECN code, move it to if_vlan.h instead
of pkt_sched.h.

v3:
- Remove empty lines
- Move vlan variable definitions inside loop in skb_protocol()
- Also use skb_protocol() helper in IP{,6}_ECN_decapsulate() and
  bpf_skb_ecn_set_ce()

v2:
- Use eth_type_vlan() helper in skb_protocol()
- Also fix code that reads skb->protocol directly
- Change a couple of 'if/else if' statements to switch constructs to avoid
  calling the helper twice

Reported-by: Ilya Ponetayev <i.ponetaev@ndmsystems.com>
Fixes: d8b9605d2697 ("net: sched: fix skb->protocol use in case of accelerated vlan path")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonetfilter: conntrack: refetch conntrack after nf_conntrack_update()
Pablo Neira Ayuso [Wed, 1 Jul 2020 11:17:40 +0000 (13:17 +0200)]
netfilter: conntrack: refetch conntrack after nf_conntrack_update()

__nf_conntrack_update() might refresh the conntrack object that is
attached to the skbuff. Otherwise, this triggers UAF.

[  633.200434] ==================================================================
[  633.200472] BUG: KASAN: use-after-free in nf_conntrack_update+0x34e/0x770 [nf_conntrack]
[  633.200478] Read of size 1 at addr ffff888370804c00 by task nfqnl_test/6769

[  633.200487] CPU: 1 PID: 6769 Comm: nfqnl_test Not tainted 5.8.0-rc2+ #388
[  633.200490] Hardware name: LENOVO 23259H1/23259H1, BIOS G2ET32WW (1.12 ) 05/30/2012
[  633.200491] Call Trace:
[  633.200499]  dump_stack+0x7c/0xb0
[  633.200526]  ? nf_conntrack_update+0x34e/0x770 [nf_conntrack]
[  633.200532]  print_address_description.constprop.6+0x1a/0x200
[  633.200539]  ? _raw_write_lock_irqsave+0xc0/0xc0
[  633.200568]  ? nf_conntrack_update+0x34e/0x770 [nf_conntrack]
[  633.200594]  ? nf_conntrack_update+0x34e/0x770 [nf_conntrack]
[  633.200598]  kasan_report.cold.9+0x1f/0x42
[  633.200604]  ? call_rcu+0x2c0/0x390
[  633.200633]  ? nf_conntrack_update+0x34e/0x770 [nf_conntrack]
[  633.200659]  nf_conntrack_update+0x34e/0x770 [nf_conntrack]
[  633.200687]  ? nf_conntrack_find_get+0x30/0x30 [nf_conntrack]

Closes: https://bugzilla.netfilter.org/show_bug.cgi?id=1436
Fixes: ee04805ff54a ("netfilter: conntrack: make conntrack userspace helpers work again")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
4 years agoMAINTAINERS: net: macb: add Claudiu as co-maintainer
Nicolas Ferre [Thu, 2 Jul 2020 13:00:21 +0000 (15:00 +0200)]
MAINTAINERS: net: macb: add Claudiu as co-maintainer

I would like that Claudiu becomes co-maintainer of the Cadence macb
driver. He's already participating to lots of reviews and enhancements
to this driver and knows the different versions of this controller.

Signed-off-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: microchip: set the correct number of ports
Codrin Ciubotariu [Thu, 2 Jul 2020 09:44:50 +0000 (12:44 +0300)]
net: dsa: microchip: set the correct number of ports

The number of ports is incorrectly set to the maximum available for a DSA
switch. Even if the extra ports are not used, this causes some functions
to be called later, like port_disable() and port_stp_state_set(). If the
driver doesn't check the port index, it will end up modifying unknown
registers.

Fixes: b987e98e50ab ("dsa: add DSA switch driver for Microchip KSZ9477")
Signed-off-by: Codrin Ciubotariu <codrin.ciubotariu@microchip.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: md5: allow changing MD5 keys in all socket states
Eric Dumazet [Thu, 2 Jul 2020 01:39:33 +0000 (18:39 -0700)]
tcp: md5: allow changing MD5 keys in all socket states

This essentially reverts commit 721230326891 ("tcp: md5: reject TCP_MD5SIG
or TCP_MD5SIG_EXT on established sockets")

Mathieu reported that many vendors BGP implementations can
actually switch TCP MD5 on established flows.

Quoting Mathieu :
   Here is a list of a few network vendors along with their behavior
   with respect to TCP MD5:

   - Cisco: Allows for password to be changed, but within the hold-down
     timer (~180 seconds).
   - Juniper: When password is initially set on active connection it will
     reset, but after that any subsequent password changes no network
     resets.
   - Nokia: No notes on if they flap the tcp connection or not.
   - Ericsson/RedBack: Allows for 2 password (old/new) to co-exist until
     both sides are ok with new passwords.
   - Meta-Switch: Expects the password to be set before a connection is
     attempted, but no further info on whether they reset the TCP
     connection on a change.
   - Avaya: Disable the neighbor, then set password, then re-enable.
   - Zebos: Would normally allow the change when socket connected.

We can revert my prior change because commit 9424e2e7ad93 ("tcp: md5: fix potential
overestimation of TCP option space") removed the leak of 4 kernel bytes to
the wire that was the main reason for my patch.

While doing my investigations, I found a bug when a MD5 key is changed, leading
to these commits that stable teams want to consider before backporting this revert :

 Commit 6a2febec338d ("tcp: md5: add missing memory barriers in tcp_md5_do_add()/tcp_md5_hash_key()")
 Commit e6ced831ef11 ("tcp: md5: refine tcp_md5_do_add()/tcp_md5_hash_key() barriers")

Fixes: 721230326891 "tcp: md5: reject TCP_MD5SIG or TCP_MD5SIG_EXT on established sockets"
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: microchip: enable ksz9893 via i2c in the ksz9477 driver
Helmut Grohne [Wed, 1 Jul 2020 11:22:20 +0000 (13:22 +0200)]
net: dsa: microchip: enable ksz9893 via i2c in the ksz9477 driver

The KSZ9893 3-Port Gigabit Ethernet Switch can be controlled via SPI,
I²C or MDIO (very limited and not supported by this driver). While there
is already a compatible entry for the SPI bus, it was missing for I²C.

Signed-off-by: Helmut Grohne <helmut.grohne@intenta.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: fix SO_RCVLOWAT possible hangs under high mem pressure
Eric Dumazet [Tue, 30 Jun 2020 20:51:28 +0000 (13:51 -0700)]
tcp: fix SO_RCVLOWAT possible hangs under high mem pressure

Whenever tcp_try_rmem_schedule() returns an error, we are under
trouble and should make sure to wakeup readers so that they
can drain socket queues and eventually make room.

Fixes: 03f45c883c6f ("tcp: avoid extra wakeups for SO_RCVLOWAT users")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoip: Fix SO_MARK in RST, ACK and ICMP packets
Willem de Bruijn [Wed, 1 Jul 2020 20:00:06 +0000 (16:00 -0400)]
ip: Fix SO_MARK in RST, ACK and ICMP packets

When no full socket is available, skbs are sent over a per-netns
control socket. Its sk_mark is temporarily adjusted to match that
of the real (request or timewait) socket or to reflect an incoming
skb, so that the outgoing skb inherits this in __ip_make_skb.

Introduction of the socket cookie mark field broke this. Now the
skb is set through the cookie and cork:

<caller> # init sockc.mark from sk_mark or cmsg
ip_append_data
  ip_setup_cork # convert sockc.mark to cork mark
ip_push_pending_frames
  ip_finish_skb
    __ip_make_skb # set skb->mark to cork mark

But I missed these special control sockets. Update all callers of
__ip(6)_make_skb that were originally missed.

For IPv6, the same two icmp(v6) paths are affected. The third
case is not, as commit 92e55f412cff ("tcp: don't annotate
mark on control socket from tcp_v6_send_response()") replaced
the ctl_sk->sk_mark with passing the mark field directly as a
function argument. That commit predates the commit that
introduced the bug.

Fixes: c6af0c227a22 ("ip: support SO_MARK cmsg")
Signed-off-by: Willem de Bruijn <willemb@google.com>
Reported-by: Martin KaFai Lau <kafai@fb.com>
Reviewed-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: md5: do not send silly options in SYNCOOKIES
Eric Dumazet [Wed, 1 Jul 2020 19:41:23 +0000 (12:41 -0700)]
tcp: md5: do not send silly options in SYNCOOKIES

Whenever cookie_init_timestamp() has been used to encode
ECN,SACK,WSCALE options, we can not remove the TS option in the SYNACK.

Otherwise, tcp_synack_options() will still advertize options like WSCALE
that we can not deduce later when receiving the packet from the client
to complete 3WHS.

Note that modern linux TCP stacks wont use MD5+TS+SACK in a SYN packet,
but we can not know for sure that all TCP stacks have the same logic.

Before the fix a tcpdump would exhibit this wrong exchange :

10:12:15.464591 IP C > S: Flags [S], seq 4202415601, win 65535, options [nop,nop,md5 valid,mss 1400,sackOK,TS val 456965269 ecr 0,nop,wscale 8], length 0
10:12:15.464602 IP S > C: Flags [S.], seq 253516766, ack 4202415602, win 65535, options [nop,nop,md5 valid,mss 1400,nop,nop,sackOK,nop,wscale 8], length 0
10:12:15.464611 IP C > S: Flags [.], ack 1, win 256, options [nop,nop,md5 valid], length 0
10:12:15.464678 IP C > S: Flags [P.], seq 1:13, ack 1, win 256, options [nop,nop,md5 valid], length 12
10:12:15.464685 IP S > C: Flags [.], ack 13, win 65535, options [nop,nop,md5 valid], length 0

After this patch the exchange looks saner :

11:59:59.882990 IP C > S: Flags [S], seq 517075944, win 65535, options [nop,nop,md5 valid,mss 1400,sackOK,TS val 1751508483 ecr 0,nop,wscale 8], length 0
11:59:59.883002 IP S > C: Flags [S.], seq 1902939253, ack 517075945, win 65535, options [nop,nop,md5 valid,mss 1400,sackOK,TS val 1751508479 ecr 1751508483,nop,wscale 8], length 0
11:59:59.883012 IP C > S: Flags [.], ack 1, win 256, options [nop,nop,md5 valid,nop,nop,TS val 1751508483 ecr 1751508479], length 0
11:59:59.883114 IP C > S: Flags [P.], seq 1:13, ack 1, win 256, options [nop,nop,md5 valid,nop,nop,TS val 1751508483 ecr 1751508479], length 12
11:59:59.883122 IP S > C: Flags [.], ack 13, win 256, options [nop,nop,md5 valid,nop,nop,TS val 1751508483 ecr 1751508483], length 0
11:59:59.883152 IP S > C: Flags [P.], seq 1:13, ack 13, win 256, options [nop,nop,md5 valid,nop,nop,TS val 1751508484 ecr 1751508483], length 12
11:59:59.883170 IP C > S: Flags [.], ack 13, win 256, options [nop,nop,md5 valid,nop,nop,TS val 1751508484 ecr 1751508484], length 0

Of course, no SACK block will ever be added later, but nothing should break.
Technically, we could remove the 4 nops included in MD5+TS options,
but again some stacks could break seeing not conventional alignment.

Fixes: 4957faade11b ("TCPCT part 1g: Responder Cookie => Initiator")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Florian Westphal <fw@strlen.de>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agords: If one path needs re-connection, check all and re-connect
Rao Shoaib [Wed, 1 Jul 2020 19:23:38 +0000 (12:23 -0700)]
rds: If one path needs re-connection, check all and re-connect

In testing with mprds enabled, Oracle Cluster nodes after reboot were
not able to communicate with others nodes and so failed to rejoin
the cluster. Peers with lower IP address initiated connection but the
node could not respond as it choose a different path and could not
initiate a connection as it had a higher IP address.

With this patch, when a node sends out a packet and the selected path
is down, all other paths are also checked and any down paths are
re-connected.

Reviewed-by: Ka-cheong Poon <ka-cheong.poon@oracle.com>
Reviewed-by: David Edmondson <david.edmondson@oracle.com>
Signed-off-by: Somasundaram Krishnasamy <somasundaram.krishnasamy@oracle.com>
Signed-off-by: Rao Shoaib <rao.shoaib@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: md5: refine tcp_md5_do_add()/tcp_md5_hash_key() barriers
Eric Dumazet [Wed, 1 Jul 2020 18:43:04 +0000 (11:43 -0700)]
tcp: md5: refine tcp_md5_do_add()/tcp_md5_hash_key() barriers

My prior fix went a bit too far, according to Herbert and Mathieu.

Since we accept that concurrent TCP MD5 lookups might see inconsistent
keys, we can use READ_ONCE()/WRITE_ONCE() instead of smp_rmb()/smp_wmb()

Clearing all key->key[] is needed to avoid possible KMSAN reports,
if key->keylen is increased. Since tcp_md5_do_add() is not fast path,
using __GFP_ZERO to clear all struct tcp_md5sig_key is simpler.

data_race() was added in linux-5.8 and will prevent KCSAN reports,
this can safely be removed in stable backports, if data_race() is
not yet backported.

v2: use data_race() both in tcp_md5_hash_key() and tcp_md5_do_add()

Fixes: 6a2febec338d ("tcp: md5: add missing memory barriers in tcp_md5_do_add()/tcp_md5_hash_key()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Marco Elver <elver@google.com>
Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agogenetlink: remove genl_bind
Sean Tranchetti [Tue, 30 Jun 2020 17:50:17 +0000 (11:50 -0600)]
genetlink: remove genl_bind

A potential deadlock can occur during registering or unregistering a
new generic netlink family between the main nl_table_lock and the
cb_lock where each thread wants the lock held by the other, as
demonstrated below.

1) Thread 1 is performing a netlink_bind() operation on a socket. As part
   of this call, it will call netlink_lock_table(), incrementing the
   nl_table_users count to 1.
2) Thread 2 is registering (or unregistering) a genl_family via the
   genl_(un)register_family() API. The cb_lock semaphore will be taken for
   writing.
3) Thread 1 will call genl_bind() as part of the bind operation to handle
   subscribing to GENL multicast groups at the request of the user. It will
   attempt to take the cb_lock semaphore for reading, but it will fail and
   be scheduled away, waiting for Thread 2 to finish the write.
4) Thread 2 will call netlink_table_grab() during the (un)registration
   call. However, as Thread 1 has incremented nl_table_users, it will not
   be able to proceed, and both threads will be stuck waiting for the
   other.

genl_bind() is a noop, unless a genl_family implements the mcast_bind()
function to handle setting up family-specific multicast operations. Since
no one in-tree uses this functionality as Cong pointed out, simply removing
the genl_bind() function will remove the possibility for deadlock, as there
is no attempt by Thread 1 above to take the cb_lock semaphore.

Fixes: c380d9a7afff ("genetlink: pass multicast bind/unbind to families")
Suggested-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Johannes Berg <johannes.berg@intel.com>
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Sean Tranchetti <stranche@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agohinic: fix passing non negative value to ERR_PTR
Luo bin [Wed, 1 Jul 2020 03:16:33 +0000 (11:16 +0800)]
hinic: fix passing non negative value to ERR_PTR

get_dev_cap and set_resources_state functions may return a positive
value because of hardware failure, and the positive return value
can not be passed to ERR_PTR directly.

Fixes: 7dd29ee12865 ("hinic: add sriov feature support")
Signed-off-by: Luo bin <luobin9@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: qrtr: Fix an out of bounds read qrtr_endpoint_post()
Dan Carpenter [Tue, 30 Jun 2020 11:46:15 +0000 (14:46 +0300)]
net: qrtr: Fix an out of bounds read qrtr_endpoint_post()

This code assumes that the user passed in enough data for a
qrtr_hdr_v1 or qrtr_hdr_v2 struct, but it's not necessarily true.  If
the buffer is too small then it will read beyond the end.

Reported-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Reported-by: syzbot+b8fe393f999a291a9ea6@syzkaller.appspotmail.com
Fixes: 194ccc88297a ("net: qrtr: Support decoding incoming v2 packets")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotcp: md5: add missing memory barriers in tcp_md5_do_add()/tcp_md5_hash_key()
Eric Dumazet [Tue, 30 Jun 2020 23:41:01 +0000 (16:41 -0700)]
tcp: md5: add missing memory barriers in tcp_md5_do_add()/tcp_md5_hash_key()

MD5 keys are read with RCU protection, and tcp_md5_do_add()
might update in-place a prior key.

Normally, typical RCU updates would allocate a new piece
of memory. In this case only key->key and key->keylen might
be updated, and we do not care if an incoming packet could
see the old key, the new one, or some intermediate value,
since changing the key on a live flow is known to be problematic
anyway.

We only want to make sure that in the case key->keylen
is changed, cpus in tcp_md5_hash_key() wont try to use
uninitialized data, or crash because key->keylen was
read twice to feed sg_init_one() and ahash_request_set_crypt()

Fixes: 9ea88a153001 ("tcp: md5: check md5 signature without socket lock")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: qrtr: free flow in __qrtr_node_release
Carl Huang [Tue, 30 Jun 2020 06:52:51 +0000 (14:52 +0800)]
net: qrtr: free flow in __qrtr_node_release

The flow is allocated in qrtr_tx_wait, but not freed when qrtr node
is released. (*slot) becomes NULL after radix_tree_iter_delete is
called in __qrtr_node_release. The fix is to save (*slot) to a
vairable and then free it.

This memory leak is catched when kmemleak is enabled in kernel,
the report looks like below:

unreferenced object 0xffffa0de69e08420 (size 32):
  comm "kworker/u16:3", pid 176, jiffies 4294918275 (age 82858.876s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 28 84 e0 69 de a0 ff ff  ........(..i....
    28 84 e0 69 de a0 ff ff 03 00 00 00 00 00 00 00  (..i............
  backtrace:
    [<00000000e252af0a>] qrtr_node_enqueue+0x38e/0x400 [qrtr]
    [<000000009cea437f>] qrtr_sendmsg+0x1e0/0x2a0 [qrtr]
    [<000000008bddbba4>] sock_sendmsg+0x5b/0x60
    [<0000000003beb43a>] qmi_send_message.isra.3+0xbe/0x110 [qmi_helpers]
    [<000000009c9ae7de>] qmi_send_request+0x1c/0x20 [qmi_helpers]

Signed-off-by: Carl Huang <cjhuang@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: cxgb4: fix return error value in t4_prep_fw
Li Heng [Mon, 29 Jun 2020 10:49:51 +0000 (18:49 +0800)]
net: cxgb4: fix return error value in t4_prep_fw

t4_prep_fw goto bye tag with positive return value when something
bad happened and which can not free resource in adap_init0.
so fix it to return negative value.

Fixes: 16e47624e76b ("cxgb4: Add new scheme to update T4/T5 firmware")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Li Heng <liheng40@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
David S. Miller [Tue, 30 Jun 2020 21:20:45 +0000 (14:20 -0700)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2020-06-30

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

We've added 28 non-merge commits during the last 9 day(s) which contain
a total of 35 files changed, 486 insertions(+), 232 deletions(-).

The main changes are:

1) Fix an incorrect verifier branch elimination for PTR_TO_BTF_ID pointer
   types, from Yonghong Song.

2) Fix UAPI for sockmap and flow_dissector progs that were ignoring various
   arguments passed to BPF_PROG_{ATTACH,DETACH}, from Lorenz Bauer & Jakub Sitnicki.

3) Fix broken AF_XDP DMA hacks that are poking into dma-direct and swiotlb
   internals and integrate it properly into DMA core, from Christoph Hellwig.

4) Fix RCU splat from recent changes to avoid skipping ingress policy when
   kTLS is enabled, from John Fastabend.

5) Fix BPF ringbuf map to enforce size to be the power of 2 in order for its
   position masking to work, from Andrii Nakryiko.

6) Fix regression from CAP_BPF work to re-allow CAP_SYS_ADMIN for loading
   of network programs, from Maciej Żenczykowski.

7) Fix libbpf section name prefix for devmap progs, from Jesper Dangaard Brouer.

8) Fix formatting in UAPI documentation for BPF helpers, from Quentin Monnet.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>