platform/kernel/linux-rpi.git
4 years agodevlink: Wait longer before warning about unset port type
Ido Schimmel [Thu, 9 Jan 2020 17:57:41 +0000 (19:57 +0200)]
devlink: Wait longer before warning about unset port type

commit 4c582234ab3948d08a24c82eb1e00436aabacbc6 upstream.

The commit cited below causes devlink to emit a warning if a type was
not set on a devlink port for longer than 30 seconds to "prevent
misbehavior of drivers". This proved to be problematic when
unregistering the backing netdev. The flow is always:

devlink_port_type_clear() // schedules the warning
unregister_netdev() // blocking
devlink_port_unregister() // cancels the warning

The call to unregister_netdev() can block for long periods of time for
various reasons: RTNL lock is contended, large amounts of configuration
to unroll following dismantle of the netdev, etc. This results in
devlink emitting a warning despite the driver behaving correctly.

In emulated environments (of future hardware) which are usually very
slow, the warning can also be emitted during port creation as more than
30 seconds can pass between the time the devlink port is registered and
when its type is set.

In addition, syzbot has hit this warning [1] 1974 times since 07/11/19
without being able to produce a reproducer. Probably because
reproduction depends on the load or other bugs (e.g., RTNL not being
released).

To prevent bogus warnings, increase the timeout to 1 hour.

[1] https://syzkaller.appspot.com/bug?id=e99b59e9c024a666c9f7450dc162a4b74d09d9cb

Fixes: 136bf27fc0e9 ("devlink: add warning in case driver does not set port type")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reported-by: syzbot+b0a18ed7b08b735d2f41@syzkaller.appspotmail.com
Reported-by: Alex Veber <alexve@mellanox.com>
Tested-by: Alex Veber <alexve@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: stmmac: tc: Do not setup flower filtering if RSS is enabled
Jose Abreu [Fri, 10 Jan 2020 15:23:53 +0000 (16:23 +0100)]
net: stmmac: tc: Do not setup flower filtering if RSS is enabled

commit 7bd754c47dd3ad1b048c9641294b0234fcce2c58 upstream.

RSS, when enabled, will bypass the L3 and L4 filtering causing it not
to work. Add a check before trying to setup the filters.

Fixes: 425eabddaf0f ("net: stmmac: Implement L3/L4 Filters using TC Flower")
Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: stmmac: selftests: Update status when disabling RSS
Jose Abreu [Fri, 10 Jan 2020 15:23:52 +0000 (16:23 +0100)]
net: stmmac: selftests: Update status when disabling RSS

commit e715d74504352968cf24ac95476706bc911a69cd upstream.

We are disabling RSS on HW but not updating the internal private status
to the 'disabled' state. This is needed for next tc commit that will
check if RSS is disabled before trying to apply filters.

Fixes: 4647e021193d ("net: stmmac: selftests: Add selftest for L3/L4 Filters")
Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoselftests: mlxsw: qos_mc_aware: Fix mausezahn invocation
Petr Machata [Wed, 15 Jan 2020 11:53:47 +0000 (13:53 +0200)]
selftests: mlxsw: qos_mc_aware: Fix mausezahn invocation

commit fef6d6704944c7be72fd2b77c021f1aed3d5df0d upstream.

Mausezahn does not recognize "own" as a keyword on source IP address. As a
result, the MC stream is not running at all, and therefore no UC
degradation can be observed even in principle.

Fix the invocation, and tighten the test: due to the minimum shaper
configured at the MC TCs, we always expect about 20% degradation. Fail the
test if it is lower.

Fixes: 573363a68f27 ("selftests: mlxsw: Add qos_lib.sh")
Signed-off-by: Petr Machata <petrm@mellanox.com>
Reported-by: Amit Cohen <amitc@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: stmmac: selftests: Mark as fail when received VLAN ID != expected
Jose Abreu [Tue, 14 Jan 2020 16:09:22 +0000 (17:09 +0100)]
net: stmmac: selftests: Mark as fail when received VLAN ID != expected

commit d39b68e5a736afa67d2e9cfb158efdd237d99dbd upstream.

When the VLAN ID does not match the expected one it means filter failed
in HW. Fix it.

Fixes: 94e18382003c ("net: stmmac: selftests: Add selftest for VLAN TX Offload")
Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: stmmac: selftests: Make it work in Synopsys AXS101 boards
Jose Abreu [Tue, 14 Jan 2020 16:09:21 +0000 (17:09 +0100)]
net: stmmac: selftests: Make it work in Synopsys AXS101 boards

commit 0b9f932edc1a461933bfde08e620362e2190e0dd upstream.

Synopsys AXS101 boards do not support unaligned memory loads or stores.
Change the selftests mechanism to explicity:
- Not add extra alignment in TX SKB
- Use the unaligned version of ether_addr_equal()

Fixes: 091810dbded9 ("net: stmmac: Introduce selftests support")
Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomlxsw: spectrum_qdisc: Include MC TCs in Qdisc counters
Petr Machata [Wed, 15 Jan 2020 11:53:49 +0000 (13:53 +0200)]
mlxsw: spectrum_qdisc: Include MC TCs in Qdisc counters

commit 85005b82e59fa7bb7388b12594ab2067bf73d66c upstream.

mlxsw configures Spectrum in such a way that BUM traffic is passed not
through its nominal traffic class TC, but through its MC counterpart TC+8.
However, when collecting statistics, Qdiscs only look at the nominal TC and
ignore the MC TC.

Add two helpers to compute the value for logical TC from the constituents,
one for backlog, the other for tail drops. Use them throughout instead of
going through the xstats pointer directly.

Counters for TX bytes and packets are deduced from packet priority
counters, and therefore already include BUM traffic. wred_drop counter is
irrelevant on MC TCs, because RED is not enabled on them.

Fixes: 7b8195306694 ("mlxsw: spectrum: Configure MC-aware mode on mlxsw ports")
Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomlxsw: spectrum: Wipe xstats.backlog of down ports
Petr Machata [Wed, 15 Jan 2020 11:53:48 +0000 (13:53 +0200)]
mlxsw: spectrum: Wipe xstats.backlog of down ports

commit ca7609ff3680c51d6c29897f3117aa2ad904f92a upstream.

Per-port counter cache used by Qdiscs is updated periodically, unless the
port is down. The fact that the cache is not updated for down ports is no
problem for most counters, which are relative in nature. However, backlog
is absolute in nature, and if there is a non-zero value in the cache around
the time that the port goes down, that value just stays there. This value
then leaks to offloaded Qdiscs that report non-zero backlog even if
there (obviously) is no traffic.

The HW does not keep backlog of a downed port, so do likewise: as the port
goes down, wipe the backlog value from xstats.

Fixes: 075ab8adaf4e ("mlxsw: spectrum: Collect tclass related stats periodically")
Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomlxsw: spectrum: Do not modify cloned SKBs during xmit
Ido Schimmel [Wed, 15 Jan 2020 11:53:45 +0000 (13:53 +0200)]
mlxsw: spectrum: Do not modify cloned SKBs during xmit

commit 2da51ce75d86ab1f7770ac1391a9a1697ddaa60c upstream.

The driver needs to prepend a Tx header to each packet it is
transmitting. The header includes information such as the egress port
and traffic class.

The addition of the header requires the driver to modify the SKB's
header and therefore it must not be shared. Otherwise, we risk hitting
various race conditions.

For example, when a packet is flooded (cloned) by the bridge driver to
two switch ports swp1 and swp2:

t0 - mlxsw_sp_port_xmit() is called for swp1. Tx header is prepended with
     swp1's port number
t1 - mlxsw_sp_port_xmit() is called for swp2. Tx header is prepended with
     swp2's port number, overwriting swp1's port number
t2 - The device processes data buffer from t0. Packet is transmitted via
     swp2
t3 - The device processes data buffer from t1. Packet is transmitted via
     swp2

Usually, the device is fast enough and transmits the packet before its
Tx header is overwritten, but this is not the case in emulated
environments.

Fix this by making sure the SKB's header is writable by calling
skb_cow_head(). Since the function ensures we have headroom to push the
Tx header, the check further in the function can be removed.

v2:
* Use skb_cow_head() instead of skb_unshare() as suggested by Jakub
* Remove unnecessary check regarding headroom

Fixes: 56ade8fe3fe1 ("mlxsw: spectrum: Add initial support for Spectrum ASIC")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reported-by: Shalom Toledo <shalomt@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agosh_eth: check sh_eth_cpu_data::dual_port when dumping registers
Sergei Shtylyov [Wed, 8 Jan 2020 20:42:42 +0000 (23:42 +0300)]
sh_eth: check sh_eth_cpu_data::dual_port when dumping registers

commit 3249b1e442a1be1a6b9f1026785b519d1443f807 upstream.

When adding the sh_eth_cpu_data::dual_port flag I forgot to add the flag
checks to __sh_eth_get_regs(), causing the non-existing TSU registers to
be dumped by 'ethtool' on the single port Ether controllers having TSU...

Fixes: a94cf2a614f8 ("sh_eth: fix TSU init on SH7734/R8A7740")
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agodrm/amdgpu: allow direct upload save restore list for raven2
changzhu [Thu, 7 Nov 2019 06:09:27 +0000 (14:09 +0800)]
drm/amdgpu: allow direct upload save restore list for raven2

commit eebc7f4d7ffa09f2a620bd1e2c67ddd579118af9 upstream.

It will cause modprobe atombios stuck problem in raven2 if it doesn't
allow direct upload save restore list from gfx driver.
So it needs to allow direct upload save restore list for raven2
temporarily.

Signed-off-by: changzhu <Changfeng.Zhu@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoi40e: prevent memory leak in i40e_setup_macvlans
Navid Emamdoost [Wed, 25 Sep 2019 15:48:30 +0000 (10:48 -0500)]
i40e: prevent memory leak in i40e_setup_macvlans

[ Upstream commit 27d461333459d282ffa4a2bdb6b215a59d493a8f ]

In i40e_setup_macvlans if i40e_setup_channel fails the allocated memory
for ch should be released.

Signed-off-by: Navid Emamdoost <navid.emamdoost@gmail.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: dsa: bcm_sf2: Configure IMP port for 2Gb/sec
Florian Fainelli [Thu, 16 Jan 2020 20:55:48 +0000 (12:55 -0800)]
net: dsa: bcm_sf2: Configure IMP port for 2Gb/sec

[ Upstream commit 8f1880cbe8d0d49ebb7e9ae409b3b96676e5aa97 ]

With the implementation of the system reset controller we lost a setting
that is currently applied by the bootloader and which configures the IMP
port for 2Gb/sec, the default is 1Gb/sec. This is needed given the
number of ports and applications we expect to run so bring back that
setting.

Fixes: 01b0ac07589e ("net: dsa: bcm_sf2: Add support for optional reset controller line")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: sched: act_ctinfo: fix memory leak
Eric Dumazet [Sun, 19 Jan 2020 04:45:06 +0000 (20:45 -0800)]
net: sched: act_ctinfo: fix memory leak

[ Upstream commit 09d4f10a5e78d76a53e3e584f1e6a701b6d24108 ]

Implement a cleanup method to properly free ci->params

BUG: memory leak
unreferenced object 0xffff88811746e2c0 (size 64):
  comm "syz-executor617", pid 7106, jiffies 4294943055 (age 14.250s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    c0 34 60 84 ff ff ff ff 00 00 00 00 00 00 00 00  .4`.............
  backtrace:
    [<0000000015aa236f>] kmemleak_alloc_recursive include/linux/kmemleak.h:43 [inline]
    [<0000000015aa236f>] slab_post_alloc_hook mm/slab.h:586 [inline]
    [<0000000015aa236f>] slab_alloc mm/slab.c:3320 [inline]
    [<0000000015aa236f>] kmem_cache_alloc_trace+0x145/0x2c0 mm/slab.c:3549
    [<000000002c946bd1>] kmalloc include/linux/slab.h:556 [inline]
    [<000000002c946bd1>] kzalloc include/linux/slab.h:670 [inline]
    [<000000002c946bd1>] tcf_ctinfo_init+0x21a/0x530 net/sched/act_ctinfo.c:236
    [<0000000086952cca>] tcf_action_init_1+0x400/0x5b0 net/sched/act_api.c:944
    [<000000005ab29bf8>] tcf_action_init+0x135/0x1c0 net/sched/act_api.c:1000
    [<00000000392f56f9>] tcf_action_add+0x9a/0x200 net/sched/act_api.c:1410
    [<0000000088f3c5dd>] tc_ctl_action+0x14d/0x1bb net/sched/act_api.c:1465
    [<000000006b39d986>] rtnetlink_rcv_msg+0x178/0x4b0 net/core/rtnetlink.c:5424
    [<00000000fd6ecace>] netlink_rcv_skb+0x61/0x170 net/netlink/af_netlink.c:2477
    [<0000000047493d02>] rtnetlink_rcv+0x1d/0x30 net/core/rtnetlink.c:5442
    [<00000000bdcf8286>] netlink_unicast_kernel net/netlink/af_netlink.c:1302 [inline]
    [<00000000bdcf8286>] netlink_unicast+0x223/0x310 net/netlink/af_netlink.c:1328
    [<00000000fc5b92d9>] netlink_sendmsg+0x2c0/0x570 net/netlink/af_netlink.c:1917
    [<00000000da84d076>] sock_sendmsg_nosec net/socket.c:639 [inline]
    [<00000000da84d076>] sock_sendmsg+0x54/0x70 net/socket.c:659
    [<0000000042fb2eee>] ____sys_sendmsg+0x2d0/0x300 net/socket.c:2330
    [<000000008f23f67e>] ___sys_sendmsg+0x8a/0xd0 net/socket.c:2384
    [<00000000d838e4f6>] __sys_sendmsg+0x80/0xf0 net/socket.c:2417
    [<00000000289a9cb1>] __do_sys_sendmsg net/socket.c:2426 [inline]
    [<00000000289a9cb1>] __se_sys_sendmsg net/socket.c:2424 [inline]
    [<00000000289a9cb1>] __x64_sys_sendmsg+0x23/0x30 net/socket.c:2424

Fixes: 24ec483cec98 ("net: sched: Introduce act_ctinfo action")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Kevin 'ldir' Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Kevin 'ldir' Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: dsa: tag_gswip: fix typo in tagger name
Alexander Lobakin [Wed, 15 Jan 2020 08:54:38 +0000 (11:54 +0300)]
net: dsa: tag_gswip: fix typo in tagger name

[ Upstream commit ad32205470919c8e04cdd33e0613bdba50c2376d ]

The correct name is GSWIP (Gigabit Switch IP). Typo was introduced in
875138f81d71a ("dsa: Move tagger name into its ops structure") while
moving tagger names to their structures.

Fixes: 875138f81d71a ("dsa: Move tagger name into its ops structure")
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Alexander Lobakin <alobakin@dlink.ru>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Hauke Mehrtens <hauke@hauke-m.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: dsa: sja1105: Don't error out on disabled ports with no phy-mode
Vladimir Oltean [Thu, 16 Jan 2020 18:43:27 +0000 (20:43 +0200)]
net: dsa: sja1105: Don't error out on disabled ports with no phy-mode

[ Upstream commit 27afe0d34e9121a3d61cc0af9b17c2542dadde24 ]

The sja1105_parse_ports_node function was tested only on device trees
where all ports were enabled. Fix this check so that the driver
continues to probe only with the ports where status is not "disabled",
as expected.

Fixes: 8aa9ebccae87 ("net: dsa: Introduce driver for NXP SJA1105 5-port L2 switch")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: systemport: Fixed queue mapping in internal ring map
Florian Fainelli [Thu, 16 Jan 2020 21:08:58 +0000 (13:08 -0800)]
net: systemport: Fixed queue mapping in internal ring map

[ Upstream commit 5a9ef19454cd5daec8041bc7c3c11deb7456d9a0 ]

We would not be transmitting using the correct SYSTEMPORT transmit queue
during ndo_select_queue() which looks up the internal TX ring map
because while establishing the mapping we would be off by 4, so for
instance, when we populate switch port mappings we would be doing:

switch port 0, queue 0 -> ring index #0
switch port 0, queue 1 -> ring index #1
...
switch port 0, queue 3 -> ring index #3
switch port 1, queue 0 -> ring index #8 (4 + 4 * 1)
...

instead of using ring index #4. This would cause our ndo_select_queue()
to use the fallback queue mechanism which would pick up an incorrect
ring for that switch port. Fix this by using the correct switch queue
number instead of SYSTEMPORT queue number.

Fixes: 25c440704661 ("net: systemport: Simplify queue mapping logic")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: ethernet: ave: Avoid lockdep warning
Kunihiko Hayashi [Wed, 15 Jan 2020 04:02:42 +0000 (13:02 +0900)]
net: ethernet: ave: Avoid lockdep warning

[ Upstream commit 82d5d6a638cbd12b7dfe8acafd9efd87a656cc06 ]

When building with PROVE_LOCKING=y, lockdep shows the following
dump message.

    INFO: trying to register non-static key.
    the code is fine but needs lockdep annotation.
    turning off the locking correctness validator.
     ...

Calling device_set_wakeup_enable() directly occurs this issue,
and it isn't necessary for initialization, so this patch creates
internal function __ave_ethtool_set_wol() and replaces with this
in ave_init() and ave_resume().

Fixes: 7200f2e3c9e2 ("net: ethernet: ave: Set initial wol state to disabled")
Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobnxt_en: Do not treat DSN (Digital Serial Number) read failure as fatal.
Michael Chan [Fri, 17 Jan 2020 05:32:47 +0000 (00:32 -0500)]
bnxt_en: Do not treat DSN (Digital Serial Number) read failure as fatal.

[ Upstream commit d061b2411d5f3d6272187ab734ce0640827fca13 ]

DSN read can fail, for example on a kdump kernel without PCIe extended
config space support.  If DSN read fails, don't set the
BNXT_FLAG_DSN_VALID flag and continue loading.  Check the flag
to see if the stored DSN is valid before using it.  Only VF reps
creation should fail without valid DSN.

Fixes: 03213a996531 ("bnxt: move bp->switch_id initialization to PF probe")
Reported-by: Marc Smith <msmith626@gmail.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobnxt_en: Fix ipv6 RFS filter matching logic.
Michael Chan [Fri, 17 Jan 2020 05:32:46 +0000 (00:32 -0500)]
bnxt_en: Fix ipv6 RFS filter matching logic.

[ Upstream commit 6fc7caa84e713f7627e171ab1e7c4b5be0dc9b3d ]

Fix bnxt_fltr_match() to match ipv6 source and destination addresses.
The function currently only checks ipv4 addresses and will not work
corrently on ipv6 filters.

Fixes: c0c050c58d84 ("bnxt_en: New Broadcom ethernet driver.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobnxt_en: Fix NTUPLE firmware command failures.
Michael Chan [Fri, 17 Jan 2020 05:32:45 +0000 (00:32 -0500)]
bnxt_en: Fix NTUPLE firmware command failures.

[ Upstream commit ceb3284c588eee5ea256c70e4d8d7cf399b8134e ]

The NTUPLE related firmware commands are sent to the wrong firmware
channel, causing all these commands to fail on new firmware that
supports the new firmware channel.  Fix it by excluding the 3
NTUPLE firmware commands from the list for the new firmware channel.

Fixes: 760b6d33410c ("bnxt_en: Add support for 2nd firmware message channel.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agotcp: fix marked lost packets not being retransmitted
Pengcheng Yang [Tue, 14 Jan 2020 09:23:40 +0000 (17:23 +0800)]
tcp: fix marked lost packets not being retransmitted

[ Upstream commit e176b1ba476cf36f723cfcc7a9e57f3cb47dec70 ]

When the packet pointed to by retransmit_skb_hint is unlinked by ACK,
retransmit_skb_hint will be set to NULL in tcp_clean_rtx_queue().
If packet loss is detected at this time, retransmit_skb_hint will be set
to point to the current packet loss in tcp_verify_retransmit_hint(),
then the packets that were previously marked lost but not retransmitted
due to the restriction of cwnd will be skipped and cannot be
retransmitted.

To fix this, when retransmit_skb_hint is NULL, retransmit_skb_hint can
be reset only after all marked lost packets are retransmitted
(retrans_out >= lost_out), otherwise we need to traverse from
tcp_rtx_queue_head in tcp_xmit_retransmit_queue().

Packetdrill to demonstrate:

// Disable RACK and set max_reordering to keep things simple
    0 `sysctl -q net.ipv4.tcp_recovery=0`
   +0 `sysctl -q net.ipv4.tcp_max_reordering=3`

// Establish a connection
   +0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
   +0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
   +0 bind(3, ..., ...) = 0
   +0 listen(3, 1) = 0

  +.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
   +0 > S. 0:0(0) ack 1 <...>
 +.01 < . 1:1(0) ack 1 win 257
   +0 accept(3, ..., ...) = 4

// Send 8 data segments
   +0 write(4, ..., 8000) = 8000
   +0 > P. 1:8001(8000) ack 1

// Enter recovery and 1:3001 is marked lost
 +.01 < . 1:1(0) ack 1 win 257 <sack 3001:4001,nop,nop>
   +0 < . 1:1(0) ack 1 win 257 <sack 5001:6001 3001:4001,nop,nop>
   +0 < . 1:1(0) ack 1 win 257 <sack 5001:7001 3001:4001,nop,nop>

// Retransmit 1:1001, now retransmit_skb_hint points to 1001:2001
   +0 > . 1:1001(1000) ack 1

// 1001:2001 was ACKed causing retransmit_skb_hint to be set to NULL
 +.01 < . 1:1(0) ack 2001 win 257 <sack 5001:8001 3001:4001,nop,nop>
// Now retransmit_skb_hint points to 4001:5001 which is now marked lost

// BUG: 2001:3001 was not retransmitted
   +0 > . 2001:3001(1000) ack 1

Signed-off-by: Pengcheng Yang <yangpc@wangsu.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agor8152: add missing endpoint sanity check
Johan Hovold [Tue, 14 Jan 2020 08:27:29 +0000 (09:27 +0100)]
r8152: add missing endpoint sanity check

[ Upstream commit 86f3f4cd53707ceeec079b83205c8d3c756eca93 ]

Add missing endpoint sanity check to probe in order to prevent a
NULL-pointer dereference (or slab out-of-bounds access) when retrieving
the interrupt-endpoint bInterval on ndo_open() in case a device lacks
the expected endpoints.

Fixes: 40a82917b1d3 ("net/usb/r8152: enable interrupt transfer")
Cc: hayeswang <hayeswang@realtek.com>
Signed-off-by: Johan Hovold <johan@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoptp: free ptp device pin descriptors properly
Vladis Dronov [Mon, 13 Jan 2020 13:00:09 +0000 (14:00 +0100)]
ptp: free ptp device pin descriptors properly

[ Upstream commit 75718584cb3c64e6269109d4d54f888ac5a5fd15 ]

There is a bug in ptp_clock_unregister(), where ptp_cleanup_pin_groups()
first frees ptp->pin_{,dev_}attr, but then posix_clock_unregister() needs
them to destroy a related sysfs device.

These functions can not be just swapped, as posix_clock_unregister() frees
ptp which is needed in the ptp_cleanup_pin_groups(). Fix this by calling
ptp_cleanup_pin_groups() in ptp_clock_release(), right before ptp is freed.

This makes this patch fix an UAF bug in a patch which fixes an UAF bug.

Reported-by: Antti Laakso <antti.laakso@intel.com>
Fixes: a33121e5487b ("ptp: fix the race between the release of ptp_clock and cdev")
Link: https://lore.kernel.org/netdev/3d2bd09735dbdaf003585ca376b7c1e5b69a19bd.camel@intel.com/
Signed-off-by: Vladis Dronov <vdronov@redhat.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet/wan/fsl_ucc_hdlc: fix out of bounds write on array utdm_info
Colin Ian King [Tue, 14 Jan 2020 14:54:48 +0000 (14:54 +0000)]
net/wan/fsl_ucc_hdlc: fix out of bounds write on array utdm_info

[ Upstream commit ddf420390526ede3b9ff559ac89f58cb59d9db2f ]

Array utdm_info is declared as an array of MAX_HDLC_NUM (4) elements
however up to UCC_MAX_NUM (8) elements are potentially being written
to it.  Currently we have an array out-of-bounds write error on the
last 4 elements. Fix this by making utdm_info UCC_MAX_NUM elements in
size.

Addresses-Coverity: ("Out-of-bounds write")
Fixes: c19b6d246a35 ("drivers/net: support hdlc function for QE-UCC")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: usb: lan78xx: limit size of local TSO packets
Eric Dumazet [Mon, 13 Jan 2020 17:27:11 +0000 (09:27 -0800)]
net: usb: lan78xx: limit size of local TSO packets

[ Upstream commit f8d7408a4d7f60f8b2df0f81decdc882dd9c20dc ]

lan78xx_tx_bh() makes sure to not exceed MAX_SINGLE_PACKET_SIZE
bytes in the aggregated packets it builds, but does
nothing to prevent large GSO packets being submitted.

Pierre-Francois reported various hangs when/if TSO is enabled.

For localy generated packets, we can use netif_set_gso_max_size()
to limit the size of TSO packets.

Note that forwarded packets could still hit the issue,
so a complete fix might require implementing .ndo_features_check
for this driver, forcing a software segmentation if the size
of the TSO packet exceeds MAX_SINGLE_PACKET_SIZE.

Fixes: 55d7de9de6c3 ("Microchip's LAN7800 family USB 2/3 to 10/100/1000 Ethernet device driver")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: RENARD Pierre-Francois <pfrenard@gmail.com>
Tested-by: RENARD Pierre-Francois <pfrenard@gmail.com>
Cc: Stefan Wahren <stefan.wahren@i2se.com>
Cc: Woojung Huh <woojung.huh@microchip.com>
Cc: Microchip Linux Driver Support <UNGLinuxDriver@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet/sched: act_ife: initalize ife->metalist earlier
Eric Dumazet [Wed, 15 Jan 2020 16:20:39 +0000 (08:20 -0800)]
net/sched: act_ife: initalize ife->metalist earlier

[ Upstream commit 44c23d71599f81a1c7fe8389e0319822dd50c37c ]

It seems better to init ife->metalist earlier in tcf_ife_init()
to avoid the following crash :

kasan: CONFIG_KASAN_INLINE enabled
kasan: GPF could be caused by NULL-ptr deref or user memory access
general protection fault: 0000 [#1] PREEMPT SMP KASAN
CPU: 0 PID: 10483 Comm: syz-executor216 Not tainted 5.5.0-rc5-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
RIP: 0010:_tcf_ife_cleanup net/sched/act_ife.c:412 [inline]
RIP: 0010:tcf_ife_cleanup+0x6e/0x400 net/sched/act_ife.c:431
Code: 48 c1 ea 03 80 3c 02 00 0f 85 94 03 00 00 49 8b bd f8 00 00 00 48 b8 00 00 00 00 00 fc ff df 4c 8d 67 e8 48 89 fa 48 c1 ea 03 <80> 3c 02 00 0f 85 5c 03 00 00 48 bb 00 00 00 00 00 fc ff df 48 8b
RSP: 0018:ffffc90001dc6d00 EFLAGS: 00010246
RAX: dffffc0000000000 RBX: ffffffff864619c0 RCX: ffffffff815bfa09
RDX: 0000000000000000 RSI: 0000000000000004 RDI: 0000000000000000
RBP: ffffc90001dc6d50 R08: 0000000000000004 R09: fffff520003b8d8e
R10: fffff520003b8d8d R11: 0000000000000003 R12: ffffffffffffffe8
R13: ffff8880a79fc000 R14: ffff88809aba0e00 R15: 0000000000000000
FS:  0000000001b51880(0000) GS:ffff8880ae800000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000563f52cce140 CR3: 0000000093541000 CR4: 00000000001406f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 tcf_action_cleanup+0x62/0x1b0 net/sched/act_api.c:119
 __tcf_action_put+0xfa/0x130 net/sched/act_api.c:135
 __tcf_idr_release net/sched/act_api.c:165 [inline]
 __tcf_idr_release+0x59/0xf0 net/sched/act_api.c:145
 tcf_idr_release include/net/act_api.h:171 [inline]
 tcf_ife_init+0x97c/0x1870 net/sched/act_ife.c:616
 tcf_action_init_1+0x6b6/0xa40 net/sched/act_api.c:944
 tcf_action_init+0x21a/0x330 net/sched/act_api.c:1000
 tcf_action_add+0xf5/0x3b0 net/sched/act_api.c:1410
 tc_ctl_action+0x390/0x488 net/sched/act_api.c:1465
 rtnetlink_rcv_msg+0x45e/0xaf0 net/core/rtnetlink.c:5424
 netlink_rcv_skb+0x177/0x450 net/netlink/af_netlink.c:2477
 rtnetlink_rcv+0x1d/0x30 net/core/rtnetlink.c:5442
 netlink_unicast_kernel net/netlink/af_netlink.c:1302 [inline]
 netlink_unicast+0x58c/0x7d0 net/netlink/af_netlink.c:1328
 netlink_sendmsg+0x91c/0xea0 net/netlink/af_netlink.c:1917
 sock_sendmsg_nosec net/socket.c:639 [inline]
 sock_sendmsg+0xd7/0x130 net/socket.c:659
 ____sys_sendmsg+0x753/0x880 net/socket.c:2330
 ___sys_sendmsg+0x100/0x170 net/socket.c:2384
 __sys_sendmsg+0x105/0x1d0 net/socket.c:2417
 __do_sys_sendmsg net/socket.c:2426 [inline]
 __se_sys_sendmsg net/socket.c:2424 [inline]
 __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2424
 do_syscall_64+0xfa/0x790 arch/x86/entry/common.c:294
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

Fixes: 11a94d7fd80f ("net/sched: act_ife: validate the control action inside init()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Davide Caratti <dcaratti@redhat.com>
Reviewed-by: Davide Caratti <dcaratti@redhat.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: phy: dp83867: Set FORCE_LINK_GOOD to default after reset
Michael Grzeschik [Thu, 16 Jan 2020 13:16:31 +0000 (14:16 +0100)]
net: phy: dp83867: Set FORCE_LINK_GOOD to default after reset

[ Upstream commit 86ffe920e669ec73035e84553e18edf17d16317c ]

According to the Datasheet this bit should be 0 (Normal operation) in
default. With the FORCE_LINK_GOOD bit set, it is not possible to get a
link. This patch sets FORCE_LINK_GOOD to the default value after
resetting the phy.

Signed-off-by: Michael Grzeschik <m.grzeschik@pengutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: hns: fix soft lockup when there is not enough memory
Yonglong Liu [Thu, 16 Jan 2020 07:41:17 +0000 (15:41 +0800)]
net: hns: fix soft lockup when there is not enough memory

[ Upstream commit 49edd6a2c456150870ddcef5b7ed11b21d849e13 ]

When there is not enough memory and napi_alloc_skb() return NULL,
the HNS driver will print error message, and than try again, if
the memory is not enough for a while, huge error message and the
retry operation will cause soft lockup.

When napi_alloc_skb() return NULL because of no memory, we can
get a warn_alloc() call trace, so this patch deletes the error
message. We already use polling mode to handle irq, but the
retry operation will render the polling weight inactive, this
patch just return budget when the rx is not completed to avoid
dead loop.

Fixes: 36eedfde1a36 ("net: hns: Optimize hns_nic_common_poll for better performance")
Fixes: b5996f11ea54 ("net: add Hisilicon Network Subsystem basic ethernet support")
Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: hns3: pad the short frame before sending to the hardware
Yunsheng Lin [Wed, 15 Jan 2020 02:46:45 +0000 (10:46 +0800)]
net: hns3: pad the short frame before sending to the hardware

[ Upstream commit 36c67349a1a1c88b9cf11d7ca7762ababdb38867 ]

The hardware can not handle short frames below or equal to 32
bytes according to the hardware user manual, and it will trigger
a RAS error when the frame's length is below 33 bytes.

This patch pads the SKB when skb->len is below 33 bytes before
sending it to hardware.

Fixes: 76ad4f0ee747 ("net: hns3: Add support of HNS3 Ethernet Driver for hip08 SoC")
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: dsa: tag_qca: fix doubled Tx statistics
Alexander Lobakin [Wed, 15 Jan 2020 08:56:52 +0000 (11:56 +0300)]
net: dsa: tag_qca: fix doubled Tx statistics

[ Upstream commit bd5874da57edd001b35cf28ae737779498c16a56 ]

DSA subsystem takes care of netdev statistics since commit 4ed70ce9f01c
("net: dsa: Refactor transmit path to eliminate duplication"), so
any accounting inside tagger callbacks is redundant and can lead to
messing up the stats.
This bug is present in Qualcomm tagger since day 0.

Fixes: cafdc45c949b ("net-next: dsa: add Qualcomm tag RX/TX handler")
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Alexander Lobakin <alobakin@dlink.ru>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: avoid updating qdisc_xmit_lock_key in netdev_update_lockdep_key()
Cong Wang [Wed, 15 Jan 2020 21:02:38 +0000 (13:02 -0800)]
net: avoid updating qdisc_xmit_lock_key in netdev_update_lockdep_key()

[ Upstream commit 53d374979ef147ab51f5d632dfe20b14aebeccd0 ]

syzbot reported some bogus lockdep warnings, for example bad unlock
balance in sch_direct_xmit(). They are due to a race condition between
slow path and fast path, that is qdisc_xmit_lock_key gets re-registered
in netdev_update_lockdep_key() on slow path, while we could still
acquire the queue->_xmit_lock on fast path in this small window:

CPU A CPU B
__netif_tx_lock();
lockdep_unregister_key(qdisc_xmit_lock_key);
__netif_tx_unlock();
lockdep_register_key(qdisc_xmit_lock_key);

In fact, unlike the addr_list_lock which has to be reordered when
the master/slave device relationship changes, queue->_xmit_lock is
only acquired on fast path and only when NETIF_F_LLTX is not set,
so there is likely no nested locking for it.

Therefore, we can just get rid of re-registration of
qdisc_xmit_lock_key.

Reported-by: syzbot+4ec99438ed7450da6272@syzkaller.appspotmail.com
Fixes: ab92d68fc22f ("net: core: add generic lockdep keys")
Cc: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agohv_netvsc: Fix memory leak when removing rndis device
Mohammed Gamal [Tue, 14 Jan 2020 13:09:50 +0000 (15:09 +0200)]
hv_netvsc: Fix memory leak when removing rndis device

[ Upstream commit 536dc5df2808efbefc5acee334d3c4f701790ec0 ]

kmemleak detects the following memory leak when hot removing
a network device:

unreferenced object 0xffff888083f63600 (size 256):
  comm "kworker/0:1", pid 12, jiffies 4294831717 (age 1113.676s)
  hex dump (first 32 bytes):
    00 40 c7 33 80 88 ff ff 00 00 00 00 10 00 00 00  .@.3............
    00 00 00 00 ad 4e ad de ff ff ff ff 00 00 00 00  .....N..........
  backtrace:
    [<00000000d4a8f5be>] rndis_filter_device_add+0x117/0x11c0 [hv_netvsc]
    [<000000009c02d75b>] netvsc_probe+0x5e7/0xbf0 [hv_netvsc]
    [<00000000ddafce23>] vmbus_probe+0x74/0x170 [hv_vmbus]
    [<00000000046e64f1>] really_probe+0x22f/0xb50
    [<000000005cc35eb7>] driver_probe_device+0x25e/0x370
    [<0000000043c642b2>] bus_for_each_drv+0x11f/0x1b0
    [<000000005e3d09f0>] __device_attach+0x1c6/0x2f0
    [<00000000a72c362f>] bus_probe_device+0x1a6/0x260
    [<0000000008478399>] device_add+0x10a3/0x18e0
    [<00000000cf07b48c>] vmbus_device_register+0xe7/0x1e0 [hv_vmbus]
    [<00000000d46cf032>] vmbus_add_channel_work+0x8ab/0x1770 [hv_vmbus]
    [<000000002c94bb64>] process_one_work+0x919/0x17d0
    [<0000000096de6781>] worker_thread+0x87/0xb40
    [<00000000fbe7397e>] kthread+0x333/0x3f0
    [<000000004f844269>] ret_from_fork+0x3a/0x50

rndis_filter_device_add() allocates an instance of struct rndis_device
which never gets deallocated as rndis_filter_device_remove() sets
net_device->extension which points to the rndis_device struct to NULL,
leaving the rndis_device dangling.

Since net_device->extension is eventually freed in free_netvsc_device(),
we refrain from setting it to NULL inside rndis_filter_device_remove()

Signed-off-by: Mohammed Gamal <mgamal@redhat.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomacvlan: use skb_reset_mac_header() in macvlan_queue_xmit()
Eric Dumazet [Tue, 14 Jan 2020 21:00:35 +0000 (13:00 -0800)]
macvlan: use skb_reset_mac_header() in macvlan_queue_xmit()

[ Upstream commit 1712b2fff8c682d145c7889d2290696647d82dab ]

I missed the fact that macvlan_broadcast() can be used both
in RX and TX.

skb_eth_hdr() makes only sense in TX paths, so we can not
use it blindly in macvlan_broadcast()

Fixes: 96cc4b69581d ("macvlan: do not assume mac_header is set in macvlan_broadcast()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Jurgen Van Ham <juvanham@gmail.com>
Tested-by: Matteo Croce <mcroce@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
4 years agobatman-adv: Fix DAT candidate selection on little endian systems
Sven Eckelmann [Thu, 28 Nov 2019 11:25:45 +0000 (12:25 +0100)]
batman-adv: Fix DAT candidate selection on little endian systems

commit 4cc4a1708903f404d2ca0dfde30e71e052c6cbc9 upstream.

The distributed arp table is using a DHT to store and retrieve MAC address
information for an IP address. This is done using unicast messages to
selected peers. The potential peers are looked up using the IP address and
the VID.

While the IP address is always stored in big endian byte order, this is not
the case of the VID. It can (depending on the host system) either be big
endian or little endian. The host must therefore always convert it to big
endian to ensure that all devices calculate the same peers for the same
lookup data.

Fixes: be1db4f6615b ("batman-adv: make the Distributed ARP Table vlan aware")
Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpftool: Fix printing incorrect pointer in btf_dump_ptr
Martin KaFai Lau [Fri, 10 Jan 2020 23:16:44 +0000 (15:16 -0800)]
bpftool: Fix printing incorrect pointer in btf_dump_ptr

commit 555089fdfc37ad65e0ee9b42ca40c238ff546f83 upstream.

For plain text output, it incorrectly prints the pointer value
"void *data".  The "void *data" is actually pointing to memory that
contains a bpf-map's value.  The intention is to print the content of
the bpf-map's value instead of printing the pointer pointing to the
bpf-map's value.

In this case, a member of the bpf-map's value is a pointer type.
Thus, it should print the "*(void **)data".

Fixes: 22c349e8db89 ("tools: bpftool: fix format strings and arguments for jsonw_printf()")
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Link: https://lore.kernel.org/bpf/20200110231644.3484151-1-kafai@fb.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: bpf: Don't leak time wait and request sockets
Lorenz Bauer [Fri, 10 Jan 2020 13:23:36 +0000 (13:23 +0000)]
net: bpf: Don't leak time wait and request sockets

commit 2e012c74823629d9db27963c79caa3f5b2010746 upstream.

It's possible to leak time wait and request sockets via the following
BPF pseudo code:
 
  sk = bpf_skc_lookup_tcp(...)
  if (sk)
    bpf_sk_release(sk)

If sk->sk_state is TCP_NEW_SYN_RECV or TCP_TIME_WAIT the refcount taken
by bpf_skc_lookup_tcp is not undone by bpf_sk_release. This is because
sk_flags is re-used for other data in both kinds of sockets. The check

  !sock_flag(sk, SOCK_RCU_FREE)

therefore returns a bogus result. Check that sk_flags is valid by calling
sk_fullsock. Skip checking SOCK_RCU_FREE if we already know that sk is
not a full socket.

Fixes: edbf8c01de5a ("bpf: add skc_lookup_tcp helper")
Fixes: f7355a6c0497 ("bpf: Check sk_fullsock() before returning from bpf_sk_lookup()")
Signed-off-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200110132336.26099-1-lmb@cloudflare.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoNFC: pn533: fix bulk-message timeout
Johan Hovold [Mon, 13 Jan 2020 17:23:58 +0000 (18:23 +0100)]
NFC: pn533: fix bulk-message timeout

commit a112adafcb47760feff959ee1ecd10b74d2c5467 upstream.

The driver was doing a synchronous uninterruptible bulk-transfer without
using a timeout. This could lead to the driver hanging on probe due to a
malfunctioning (or malicious) device until the device is physically
disconnected. While sleeping in probe the driver prevents other devices
connected to the same hub from being added to (or removed from) the bus.

An arbitrary limit of five seconds should be more than enough.

Fixes: dbafc28955fa ("NFC: pn533: don't send USB data off of the stack")
Signed-off-by: Johan Hovold <johan@kernel.org>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonetfilter: nf_tables: fix flowtable list del corruption
Florian Westphal [Thu, 16 Jan 2020 11:03:01 +0000 (12:03 +0100)]
netfilter: nf_tables: fix flowtable list del corruption

commit 335178d5429c4cee61b58f4ac80688f556630818 upstream.

syzbot reported following crash:

  list_del corruption, ffff88808c9bb000->prev is LIST_POISON2 (dead000000000122)
  [..]
  Call Trace:
   __list_del_entry include/linux/list.h:131 [inline]
   list_del_rcu include/linux/rculist.h:148 [inline]
   nf_tables_commit+0x1068/0x3b30 net/netfilter/nf_tables_api.c:7183
   [..]

The commit transaction list has:

NFT_MSG_NEWTABLE
NFT_MSG_NEWFLOWTABLE
NFT_MSG_DELFLOWTABLE
NFT_MSG_DELTABLE

A missing generation check during DELTABLE processing causes it to queue
the DELFLOWTABLE operation a second time, so we corrupt the list here:

  case NFT_MSG_DELFLOWTABLE:
     list_del_rcu(&nft_trans_flowtable(trans)->list);
     nf_tables_flowtable_notify(&trans->ctx,

because we have two different DELFLOWTABLE transactions for the same
flowtable.  We then call list_del_rcu() twice for the same flowtable->list.

The object handling seems to suffer from the same bug so add a generation
check too and only queue delete transactions for flowtables/objects that
are still active in the next generation.

Reported-by: syzbot+37a6804945a3a13b1572@syzkaller.appspotmail.com
Fixes: 3b49e2e94e6eb ("netfilter: nf_tables: add flow table netlink frontend")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonetfilter: nf_tables: store transaction list locally while requesting module
Pablo Neira Ayuso [Mon, 13 Jan 2020 17:09:58 +0000 (18:09 +0100)]
netfilter: nf_tables: store transaction list locally while requesting module

commit ec7470b834fe7b5d7eff11b6677f5d7fdf5e9a91 upstream.

This patch fixes a WARN_ON in nft_set_destroy() due to missing
set reference count drop from the preparation phase. This is triggered
by the module autoload path. Do not exercise the abort path from
nft_request_module() while preparation phase cleaning up is still
pending.

 WARNING: CPU: 3 PID: 3456 at net/netfilter/nf_tables_api.c:3740 nft_set_destroy+0x45/0x50 [nf_tables]
 [...]
 CPU: 3 PID: 3456 Comm: nft Not tainted 5.4.6-arch3-1 #1
 RIP: 0010:nft_set_destroy+0x45/0x50 [nf_tables]
 Code: e8 30 eb 83 c6 48 8b 85 80 00 00 00 48 8b b8 90 00 00 00 e8 dd 6b d7 c5 48 8b 7d 30 e8 24 dd eb c5 48 89 ef 5d e9 6b c6 e5 c5 <0f> 0b c3 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 48 8b 7f 10 e9 52
 RSP: 0018:ffffac4f43e53700 EFLAGS: 00010202
 RAX: 0000000000000001 RBX: ffff99d63a154d80 RCX: 0000000001f88e03
 RDX: 0000000001f88c03 RSI: ffff99d6560ef0c0 RDI: ffff99d63a101200
 RBP: ffff99d617721de0 R08: 0000000000000000 R09: 0000000000000318
 R10: 00000000f0000000 R11: 0000000000000001 R12: ffffffff880fabf0
 R13: dead000000000122 R14: dead000000000100 R15: ffff99d63a154d80
 FS:  00007ff3dbd5b740(0000) GS:ffff99d6560c0000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 00001cb5de6a9000 CR3: 000000016eb6a004 CR4: 00000000001606e0
 Call Trace:
  __nf_tables_abort+0x3e3/0x6d0 [nf_tables]
  nft_request_module+0x6f/0x110 [nf_tables]
  nft_expr_type_request_module+0x28/0x50 [nf_tables]
  nf_tables_expr_parse+0x198/0x1f0 [nf_tables]
  nft_expr_init+0x3b/0xf0 [nf_tables]
  nft_dynset_init+0x1e2/0x410 [nf_tables]
  nf_tables_newrule+0x30a/0x930 [nf_tables]
  nfnetlink_rcv_batch+0x2a0/0x640 [nfnetlink]
  nfnetlink_rcv+0x125/0x171 [nfnetlink]
  netlink_unicast+0x179/0x210
  netlink_sendmsg+0x208/0x3d0
  sock_sendmsg+0x5e/0x60
  ____sys_sendmsg+0x21b/0x290

Update comment on the code to describe the new behaviour.

Reported-by: Marco Oliverio <marco.oliverio@tanaza.com>
Fixes: 452238e8d5ff ("netfilter: nf_tables: add and use helper for module autoload")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonetfilter: nf_tables: remove WARN and add NLA_STRING upper limits
Florian Westphal [Thu, 16 Jan 2020 08:06:50 +0000 (09:06 +0100)]
netfilter: nf_tables: remove WARN and add NLA_STRING upper limits

commit 9332d27d7918182add34e8043f6a754530fdd022 upstream.

This WARN can trigger because some of the names fed to the module
autoload function can be of arbitrary length.

Remove the WARN and add limits for all NLA_STRING attributes.

Reported-by: syzbot+0e63ae76d117ae1c3a01@syzkaller.appspotmail.com
Fixes: 452238e8d5ffd8 ("netfilter: nf_tables: add and use helper for module autoload")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonetfilter: nft_tunnel: ERSPAN_VERSION must not be null
Florian Westphal [Thu, 16 Jan 2020 07:58:05 +0000 (08:58 +0100)]
netfilter: nft_tunnel: ERSPAN_VERSION must not be null

commit 9ec22d7c6c69146180577f3ad5fdf504beeaee62 upstream.

Fixes: af308b94a2a4a5 ("netfilter: nf_tables: add tunnel support")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonetfilter: nft_tunnel: fix null-attribute check
Florian Westphal [Thu, 16 Jan 2020 07:44:11 +0000 (08:44 +0100)]
netfilter: nft_tunnel: fix null-attribute check

commit 1c702bf902bd37349f6d91cd7f4b372b1e46d0ed upstream.

else we get null deref when one of the attributes is missing, both
must be non-null.

Reported-by: syzbot+76d0b80493ac881ff77b@syzkaller.appspotmail.com
Fixes: aaecfdb5c5dd8ba ("netfilter: nf_tables: match on tunnel metadata")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonetfilter: nat: fix ICMP header corruption on ICMP errors
Eyal Birger [Tue, 14 Jan 2020 08:03:50 +0000 (10:03 +0200)]
netfilter: nat: fix ICMP header corruption on ICMP errors

commit 61177e911dad660df86a4553eb01c95ece2f6a82 upstream.

Commit 8303b7e8f018 ("netfilter: nat: fix spurious connection timeouts")
made nf_nat_icmp_reply_translation() use icmp_manip_pkt() as the l4
manipulation function for the outer packet on ICMP errors.

However, icmp_manip_pkt() assumes the packet has an 'id' field which
is not correct for all types of ICMP messages.

This is not correct for ICMP error packets, and leads to bogus bytes
being written the ICMP header, which can be wrongfully regarded as
'length' bytes by RFC 4884 compliant receivers.

Fix by assigning the 'id' field only for ICMP messages that have this
semantic.

Reported-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>
Fixes: 8303b7e8f018 ("netfilter: nat: fix spurious connection timeouts")
Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonetfilter: arp_tables: init netns pointer in xt_tgdtor_param struct
Florian Westphal [Sat, 11 Jan 2020 22:19:53 +0000 (23:19 +0100)]
netfilter: arp_tables: init netns pointer in xt_tgdtor_param struct

commit 212e7f56605ef9688d0846db60c6c6ec06544095 upstream.

An earlier commit (1b789577f655060d98d20e,
"netfilter: arp_tables: init netns pointer in xt_tgchk_param struct")
fixed missing net initialization for arptables, but turns out it was
incomplete.  We can get a very similar struct net NULL deref during
error unwinding:

general protection fault: 0000 [#1] PREEMPT SMP KASAN
RIP: 0010:xt_rateest_put+0xa1/0x440 net/netfilter/xt_RATEEST.c:77
 xt_rateest_tg_destroy+0x72/0xa0 net/netfilter/xt_RATEEST.c:175
 cleanup_entry net/ipv4/netfilter/arp_tables.c:509 [inline]
 translate_table+0x11f4/0x1d80 net/ipv4/netfilter/arp_tables.c:587
 do_replace net/ipv4/netfilter/arp_tables.c:981 [inline]
 do_arpt_set_ctl+0x317/0x650 net/ipv4/netfilter/arp_tables.c:1461

Also init the netns pointer in xt_tgdtor_param struct.

Fixes: add67461240c1d ("netfilter: add struct net * to target parameters")
Reported-by: syzbot+91bdd8eece0f6629ec8b@syzkaller.appspotmail.com
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonetfilter: fix a use-after-free in mtype_destroy()
Cong Wang [Fri, 10 Jan 2020 19:53:08 +0000 (11:53 -0800)]
netfilter: fix a use-after-free in mtype_destroy()

commit c120959387efa51479056fd01dc90adfba7a590c upstream.

map->members is freed by ip_set_free() right before using it in
mtype_ext_cleanup() again. So we just have to move it down.

Reported-by: syzbot+4c3cc6dbe7259dbf9054@syzkaller.appspotmail.com
Fixes: 40cd63bf33b2 ("netfilter: ipset: Support extensions which need a per data destroy function")
Acked-by: Jozsef Kadlecsik <kadlec@netfilter.org>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoi2c: iop3xx: Fix memory leak in probe error path
Krzysztof Kozlowski [Mon, 13 Jan 2020 17:29:54 +0000 (18:29 +0100)]
i2c: iop3xx: Fix memory leak in probe error path

commit e64175776d06a8ceebbfd349d7e66a4a46ca39ef upstream.

When handling devm_gpiod_get_optional() errors, free the memory already
allocated.  This fixes Smatch warnings:

    drivers/i2c/busses/i2c-iop3xx.c:437 iop3xx_i2c_probe() warn: possible memory leak of 'new_adapter'
    drivers/i2c/busses/i2c-iop3xx.c:442 iop3xx_i2c_probe() warn: possible memory leak of 'new_adapter'

Fixes: fdb7e884ad61 ("i2c: iop: Use GPIO descriptors")
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpf/sockmap: Read psock ingress_msg before sk_receive_queue
Lingpeng Chen [Thu, 9 Jan 2020 01:48:33 +0000 (09:48 +0800)]
bpf/sockmap: Read psock ingress_msg before sk_receive_queue

commit e7a5f1f1cd0008e5ad379270a8657e121eedb669 upstream.

Right now in tcp_bpf_recvmsg, sock read data first from sk_receive_queue
if not empty than psock->ingress_msg otherwise. If a FIN packet arrives
and there's also some data in psock->ingress_msg, the data in
psock->ingress_msg will be purged. It is always happen when request to a
HTTP1.0 server like python SimpleHTTPServer since the server send FIN
packet after data is sent out.

Fixes: 604326b41a6fb ("bpf, sockmap: convert to generic sk_msg interface")
Reported-by: Arika Chen <eaglesora@gmail.com>
Suggested-by: Arika Chen <eaglesora@gmail.com>
Signed-off-by: Lingpeng Chen <forrest0579@gmail.com>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200109014833.18951-1-forrest0579@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agocfg80211: fix page refcount issue in A-MSDU decap
Felix Fietkau [Mon, 13 Jan 2020 18:21:07 +0000 (19:21 +0100)]
cfg80211: fix page refcount issue in A-MSDU decap

commit 81c044fc3bdc5b7be967cd3682528ea94b58c06a upstream.

The fragments attached to a skb can be part of a compound page. In that case,
page_ref_inc will increment the refcount for the wrong page. Fix this by
using get_page instead, which calls page_ref_inc on the compound head and
also checks for overflow.

Fixes: 2b67f944f88c ("cfg80211: reuse existing page fragments in A-MSDU rx")
Cc: stable@vger.kernel.org
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Link: https://lore.kernel.org/r/20200113182107.20461-1-nbd@nbd.name
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agocfg80211: fix memory leak in cfg80211_cqm_rssi_update
Felix Fietkau [Wed, 8 Jan 2020 17:06:30 +0000 (18:06 +0100)]
cfg80211: fix memory leak in cfg80211_cqm_rssi_update

commit df16737d438f534d0cc9948c7c5158f1986c5c87 upstream.

The per-tid statistics need to be released after the call to rdev_get_station

Cc: stable@vger.kernel.org
Fixes: 8689c051a201 ("cfg80211: dynamically allocate per-tid stats for station info")
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Link: https://lore.kernel.org/r/20200108170630.33680-2-nbd@nbd.name
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agocfg80211: fix memory leak in nl80211_probe_mesh_link
Felix Fietkau [Wed, 8 Jan 2020 17:06:29 +0000 (18:06 +0100)]
cfg80211: fix memory leak in nl80211_probe_mesh_link

commit 2a279b34169e9bbf7c240691466420aba75b4175 upstream.

The per-tid statistics need to be released after the call to rdev_get_station

Cc: stable@vger.kernel.org
Fixes: 5ab92e7fe49a ("cfg80211: add support to probe unexercised mesh link")
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Link: https://lore.kernel.org/r/20200108170630.33680-1-nbd@nbd.name
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agocfg80211: fix deadlocks in autodisconnect work
Markus Theil [Wed, 8 Jan 2020 11:55:36 +0000 (12:55 +0100)]
cfg80211: fix deadlocks in autodisconnect work

commit 5a128a088a2ab0b5190eeb232b5aa0b1017a0317 upstream.

Use methods which do not try to acquire the wdev lock themselves.

Cc: stable@vger.kernel.org
Fixes: 37b1c004685a3 ("cfg80211: Support all iftypes in autodisconnect_wk")
Signed-off-by: Markus Theil <markus.theil@tu-ilmenau.de>
Link: https://lore.kernel.org/r/20200108115536.2262-1-markus.theil@tu-ilmenau.de
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoi2c: tegra: Properly disable runtime PM on driver's probe error
Dmitry Osipenko [Tue, 14 Jan 2020 01:34:36 +0000 (04:34 +0300)]
i2c: tegra: Properly disable runtime PM on driver's probe error

commit 24a49678f5e20f18006e71b90ac1531876b27eb1 upstream.

One of the recent Tegra I2C commits made a change that resumes runtime PM
during driver's probe, but it missed to put the RPM in a case of error.
Note that it's not correct to use pm_runtime_status_suspended because it
breaks RPM refcounting.

Fixes: 8ebf15e9c869 ("i2c: tegra: Move suspend handling to NOIRQ phase")
Cc: <stable@vger.kernel.org> # v5.4+
Tested-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoi2c: tegra: Fix suspending in active runtime PM state
Dmitry Osipenko [Tue, 14 Jan 2020 01:34:35 +0000 (04:34 +0300)]
i2c: tegra: Fix suspending in active runtime PM state

commit 9f42de8d4ec2304f10bbc51dc0484f3503d61196 upstream.

I noticed that sometime I2C clock is kept enabled during suspend-resume.
This happens because runtime PM defers dynamic suspension and thus it may
happen that runtime PM is in active state when system enters into suspend.
In particular I2C controller that is used for CPU's DVFS is often kept ON
during suspend because CPU's voltage scaling happens quite often.

Fixes: 8ebf15e9c869 ("i2c: tegra: Move suspend handling to NOIRQ phase")
Cc: <stable@vger.kernel.org> # v5.4+
Tested-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpf: Sockmap/tls, fix pop data with SK_DROP return code
John Fastabend [Sat, 11 Jan 2020 06:12:06 +0000 (06:12 +0000)]
bpf: Sockmap/tls, fix pop data with SK_DROP return code

commit 7361d44896ff20d48bdd502d1a0cd66308055d45 upstream.

When user returns SK_DROP we need to reset the number of copied bytes
to indicate to the user the bytes were dropped and not sent. If we
don't reset the copied arg sendmsg will return as if those bytes were
copied giving the user a positive return value.

This works as expected today except in the case where the user also
pops bytes. In the pop case the sg.size is reduced but we don't correctly
account for this when copied bytes is reset. The popped bytes are not
accounted for and we return a small positive value potentially confusing
the user.

The reason this happens is due to a typo where we do the wrong comparison
when accounting for pop bytes. In this fix notice the if/else is not
needed and that we have a similar problem if we push data except its not
visible to the user because if delta is larger the sg.size we return a
negative value so it appears as an error regardless.

Fixes: 7246d8ed4dcce ("bpf: helper to pop data from messages")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/bpf/20200111061206.8028-9-john.fastabend@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpf: Sockmap/tls, skmsg can have wrapped skmsg that needs extra chaining
John Fastabend [Sat, 11 Jan 2020 06:12:05 +0000 (06:12 +0000)]
bpf: Sockmap/tls, skmsg can have wrapped skmsg that needs extra chaining

commit 9aaaa56845a06aeabdd597cbe19492dc01f281ec upstream.

Its possible through a set of push, pop, apply helper calls to construct
a skmsg, which is just a ring of scatterlist elements, with the start
value larger than the end value. For example,

      end       start
  |_0_|_1_| ... |_n_|_n+1_|

Where end points at 1 and start points and n so that valid elements is
the set {n, n+1, 0, 1}.

Currently, because we don't build the correct chain only {n, n+1} will
be sent. This adds a check and sg_chain call to correctly submit the
above to the crypto and tls send path.

Fixes: d3b18ad31f93d ("tls: add bpf support to sk_msg handling")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/bpf/20200111061206.8028-8-john.fastabend@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpf: Sockmap/tls, tls_sw can create a plaintext buf > encrypt buf
John Fastabend [Sat, 11 Jan 2020 06:12:04 +0000 (06:12 +0000)]
bpf: Sockmap/tls, tls_sw can create a plaintext buf > encrypt buf

commit d468e4775c1c351616947ba0cccc43273963b9b5 upstream.

It is possible to build a plaintext buffer using push helper that is larger
than the allocated encrypt buffer. When this record is pushed to crypto
layers this can result in a NULL pointer dereference because the crypto
API expects the encrypt buffer is large enough to fit the plaintext
buffer. Kernel splat below.

To resolve catch the cases this can happen and split the buffer into two
records to send individually. Unfortunately, there is still one case to
handle where the split creates a zero sized buffer. In this case we merge
the buffers and unmark the split. This happens when apply is zero and user
pushed data beyond encrypt buffer. This fixes the original case as well
because the split allocated an encrypt buffer larger than the plaintext
buffer and the merge simply moves the pointers around so we now have
a reference to the new (larger) encrypt buffer.

Perhaps its not ideal but it seems the best solution for a fixes branch
and avoids handling these two cases, (a) apply that needs split and (b)
non apply case. The are edge cases anyways so optimizing them seems not
necessary unless someone wants later in next branches.

[  306.719107] BUG: kernel NULL pointer dereference, address: 0000000000000008
[...]
[  306.747260] RIP: 0010:scatterwalk_copychunks+0x12f/0x1b0
[...]
[  306.770350] Call Trace:
[  306.770956]  scatterwalk_map_and_copy+0x6c/0x80
[  306.772026]  gcm_enc_copy_hash+0x4b/0x50
[  306.772925]  gcm_hash_crypt_remain_continue+0xef/0x110
[  306.774138]  gcm_hash_crypt_continue+0xa1/0xb0
[  306.775103]  ? gcm_hash_crypt_continue+0xa1/0xb0
[  306.776103]  gcm_hash_assoc_remain_continue+0x94/0xa0
[  306.777170]  gcm_hash_assoc_continue+0x9d/0xb0
[  306.778239]  gcm_hash_init_continue+0x8f/0xa0
[  306.779121]  gcm_hash+0x73/0x80
[  306.779762]  gcm_encrypt_continue+0x6d/0x80
[  306.780582]  crypto_gcm_encrypt+0xcb/0xe0
[  306.781474]  crypto_aead_encrypt+0x1f/0x30
[  306.782353]  tls_push_record+0x3b9/0xb20 [tls]
[  306.783314]  ? sk_psock_msg_verdict+0x199/0x300
[  306.784287]  bpf_exec_tx_verdict+0x3f2/0x680 [tls]
[  306.785357]  tls_sw_sendmsg+0x4a3/0x6a0 [tls]

test_sockmap test signature to trigger bug,

[TEST]: (1, 1, 1, sendmsg, pass,redir,start 1,end 2,pop (1,2),ktls,):

Fixes: d3b18ad31f93d ("tls: add bpf support to sk_msg handling")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/bpf/20200111061206.8028-7-john.fastabend@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpf: Sockmap/tls, msg_push_data may leave end mark in place
John Fastabend [Sat, 11 Jan 2020 06:12:03 +0000 (06:12 +0000)]
bpf: Sockmap/tls, msg_push_data may leave end mark in place

commit cf21e9ba1eb86c9333ca5b05b2f1cc94021bcaef upstream.

Leaving an incorrect end mark in place when passing to crypto
layer will cause crypto layer to stop processing data before
all data is encrypted. To fix clear the end mark on push
data instead of expecting users of the helper to clear the
mark value after the fact.

This happens when we push data into the middle of a skmsg and
have room for it so we don't do a set of copies that already
clear the end flag.

Fixes: 6fff607e2f14b ("bpf: sk_msg program helper bpf_msg_push_data")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/bpf/20200111061206.8028-6-john.fastabend@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpf: Sockmap, skmsg helper overestimates push, pull, and pop bounds
John Fastabend [Sat, 11 Jan 2020 06:12:02 +0000 (06:12 +0000)]
bpf: Sockmap, skmsg helper overestimates push, pull, and pop bounds

commit 6562e29cf6f0ddd368657d97a8d484ffc30df5ef upstream.

In the push, pull, and pop helpers operating on skmsg objects to make
data writable or insert/remove data we use this bounds check to ensure
specified data is valid,

 /* Bounds checks: start and pop must be inside message */
 if (start >= offset + l || last >= msg->sg.size)
     return -EINVAL;

The problem here is offset has already included the length of the
current element the 'l' above. So start could be past the end of
the scatterlist element in the case where start also points into an
offset on the last skmsg element.

To fix do the accounting slightly different by adding the length of
the previous entry to offset at the start of the iteration. And
ensure its initialized to zero so that the first iteration does
nothing.

Fixes: 604326b41a6fb ("bpf, sockmap: convert to generic sk_msg interface")
Fixes: 6fff607e2f14b ("bpf: sk_msg program helper bpf_msg_push_data")
Fixes: 7246d8ed4dcce ("bpf: helper to pop data from messages")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/bpf/20200111061206.8028-5-john.fastabend@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpf: Sockmap/tls, push write_space updates through ulp updates
John Fastabend [Sat, 11 Jan 2020 06:12:01 +0000 (06:12 +0000)]
bpf: Sockmap/tls, push write_space updates through ulp updates

commit 33bfe20dd7117dd81fd896a53f743a233e1ad64f upstream.

When sockmap sock with TLS enabled is removed we cleanup bpf/psock state
and call tcp_update_ulp() to push updates to TLS ULP on top. However, we
don't push the write_space callback up and instead simply overwrite the
op with the psock stored previous op. This may or may not be correct so
to ensure we don't overwrite the TLS write space hook pass this field to
the ULP and have it fixup the ctx.

This completes a previous fix that pushed the ops through to the ULP
but at the time missed doing this for write_space, presumably because
write_space TLS hook was added around the same time.

Fixes: 95fa145479fbc ("bpf: sockmap/tls, close can race with map free")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/bpf/20200111061206.8028-4-john.fastabend@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpf: Sockmap, ensure sock lock held during tear down
John Fastabend [Sat, 11 Jan 2020 06:12:00 +0000 (06:12 +0000)]
bpf: Sockmap, ensure sock lock held during tear down

commit 7e81a35302066c5a00b4c72d83e3ea4cad6eeb5b upstream.

The sock_map_free() and sock_hash_free() paths used to delete sockmap
and sockhash maps walk the maps and destroy psock and bpf state associated
with the socks in the map. When done the socks no longer have BPF programs
attached and will function normally. This can happen while the socks in
the map are still "live" meaning data may be sent/received during the walk.

Currently, though we don't take the sock_lock when the psock and bpf state
is removed through this path. Specifically, this means we can be writing
into the ops structure pointers such as sendmsg, sendpage, recvmsg, etc.
while they are also being called from the networking side. This is not
safe, we never used proper READ_ONCE/WRITE_ONCE semantics here if we
believed it was safe. Further its not clear to me its even a good idea
to try and do this on "live" sockets while networking side might also
be using the socket. Instead of trying to reason about using the socks
from both sides lets realize that every use case I'm aware of rarely
deletes maps, in fact kubernetes/Cilium case builds map at init and
never tears it down except on errors. So lets do the simple fix and
grab sock lock.

This patch wraps sock deletes from maps in sock lock and adds some
annotations so we catch any other cases easier.

Fixes: 604326b41a6fb ("bpf, sockmap: convert to generic sk_msg interface")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/bpf/20200111061206.8028-3-john.fastabend@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpf: Sockmap/tls, during free we may call tcp_bpf_unhash() in loop
John Fastabend [Sat, 11 Jan 2020 06:11:59 +0000 (06:11 +0000)]
bpf: Sockmap/tls, during free we may call tcp_bpf_unhash() in loop

commit 4da6a196f93b1af7612340e8c1ad8ce71e18f955 upstream.

When a sockmap is free'd and a socket in the map is enabled with tls
we tear down the bpf context on the socket, the psock struct and state,
and then call tcp_update_ulp(). The tcp_update_ulp() call is to inform
the tls stack it needs to update its saved sock ops so that when the tls
socket is later destroyed it doesn't try to call the now destroyed psock
hooks.

This is about keeping stacked ULPs in good shape so they always have
the right set of stacked ops.

However, recently unhash() hook was removed from TLS side. But, the
sockmap/bpf side is not doing any extra work to update the unhash op
when is torn down instead expecting TLS side to manage it. So both
TLS and sockmap believe the other side is managing the op and instead
no one updates the hook so it continues to point at tcp_bpf_unhash().
When unhash hook is called we call tcp_bpf_unhash() which detects the
psock has already been destroyed and calls sk->sk_prot_unhash() which
calls tcp_bpf_unhash() yet again and so on looping and hanging the core.

To fix have sockmap tear down logic fixup the stale pointer.

Fixes: 5d92e631b8be ("net/tls: partially revert fix transition through disconnect with close")
Reported-by: syzbot+83979935eb6304f8cd46@syzkaller.appspotmail.com
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Acked-by: Song Liu <songliubraving@fb.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/bpf/20200111061206.8028-2-john.fastabend@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobpf: Fix incorrect verifier simulation of ARSH under ALU32
Daniel Borkmann [Wed, 15 Jan 2020 20:47:33 +0000 (21:47 +0100)]
bpf: Fix incorrect verifier simulation of ARSH under ALU32

commit 0af2ffc93a4b50948f9dad2786b7f1bd253bf0b9 upstream.

Anatoly has been fuzzing with kBdysch harness and reported a hang in one
of the outcomes:

  0: R1=ctx(id=0,off=0,imm=0) R10=fp0
  0: (85) call bpf_get_socket_cookie#46
  1: R0_w=invP(id=0) R10=fp0
  1: (57) r0 &= 808464432
  2: R0_w=invP(id=0,umax_value=808464432,var_off=(0x0; 0x30303030)) R10=fp0
  2: (14) w0 -= 810299440
  3: R0_w=invP(id=0,umax_value=4294967295,var_off=(0xcf800000; 0x3077fff0)) R10=fp0
  3: (c4) w0 s>>= 1
  4: R0_w=invP(id=0,umin_value=1740636160,umax_value=2147221496,var_off=(0x67c00000; 0x183bfff8)) R10=fp0
  4: (76) if w0 s>= 0x30303030 goto pc+216
  221: R0_w=invP(id=0,umin_value=1740636160,umax_value=2147221496,var_off=(0x67c00000; 0x183bfff8)) R10=fp0
  221: (95) exit
  processed 6 insns (limit 1000000) [...]

Taking a closer look, the program was xlated as follows:

  # ./bpftool p d x i 12
  0: (85) call bpf_get_socket_cookie#7800896
  1: (bf) r6 = r0
  2: (57) r6 &= 808464432
  3: (14) w6 -= 810299440
  4: (c4) w6 s>>= 1
  5: (76) if w6 s>= 0x30303030 goto pc+216
  6: (05) goto pc-1
  7: (05) goto pc-1
  8: (05) goto pc-1
  [...]
  220: (05) goto pc-1
  221: (05) goto pc-1
  222: (95) exit

Meaning, the visible effect is very similar to f54c7898ed1c ("bpf: Fix
precision tracking for unbounded scalars"), that is, the fall-through
branch in the instruction 5 is considered to be never taken given the
conclusion from the min/max bounds tracking in w6, and therefore the
dead-code sanitation rewrites it as goto pc-1. However, real-life input
disagrees with verification analysis since a soft-lockup was observed.

The bug sits in the analysis of the ARSH. The definition is that we shift
the target register value right by K bits through shifting in copies of
its sign bit. In adjust_scalar_min_max_vals(), we do first coerce the
register into 32 bit mode, same happens after simulating the operation.
However, for the case of simulating the actual ARSH, we don't take the
mode into account and act as if it's always 64 bit, but location of sign
bit is different:

  dst_reg->smin_value >>= umin_val;
  dst_reg->smax_value >>= umin_val;
  dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val);

Consider an unknown R0 where bpf_get_socket_cookie() (or others) would
for example return 0xffff. With the above ARSH simulation, we'd see the
following results:

  [...]
  1: R1=ctx(id=0,off=0,imm=0) R2_w=invP65535 R10=fp0
  1: (85) call bpf_get_socket_cookie#46
  2: R0_w=invP(id=0) R10=fp0
  2: (57) r0 &= 808464432
    -> R0_runtime = 0x3030
  3: R0_w=invP(id=0,umax_value=808464432,var_off=(0x0; 0x30303030)) R10=fp0
  3: (14) w0 -= 810299440
    -> R0_runtime = 0xcfb40000
  4: R0_w=invP(id=0,umax_value=4294967295,var_off=(0xcf800000; 0x3077fff0)) R10=fp0
                              (0xffffffff)
  4: (c4) w0 s>>= 1
    -> R0_runtime = 0xe7da0000
  5: R0_w=invP(id=0,umin_value=1740636160,umax_value=2147221496,var_off=(0x67c00000; 0x183bfff8)) R10=fp0
                              (0x67c00000)           (0x7ffbfff8)
  [...]

In insn 3, we have a runtime value of 0xcfb40000, which is '1100 1111 1011
0100 0000 0000 0000 0000', the result after the shift has 0xe7da0000 that
is '1110 0111 1101 1010 0000 0000 0000 0000', where the sign bit is correctly
retained in 32 bit mode. In insn4, the umax was 0xffffffff, and changed into
0x7ffbfff8 after the shift, that is, '0111 1111 1111 1011 1111 1111 1111 1000'
and means here that the simulation didn't retain the sign bit. With above
logic, the updates happen on the 64 bit min/max bounds and given we coerced
the register, the sign bits of the bounds are cleared as well, meaning, we
need to force the simulation into s32 space for 32 bit alu mode.

Verification after the fix below. We're first analyzing the fall-through branch
on 32 bit signed >= test eventually leading to rejection of the program in this
specific case:

  0: R1=ctx(id=0,off=0,imm=0) R10=fp0
  0: (b7) r2 = 808464432
  1: R1=ctx(id=0,off=0,imm=0) R2_w=invP808464432 R10=fp0
  1: (85) call bpf_get_socket_cookie#46
  2: R0_w=invP(id=0) R10=fp0
  2: (bf) r6 = r0
  3: R0_w=invP(id=0) R6_w=invP(id=0) R10=fp0
  3: (57) r6 &= 808464432
  4: R0_w=invP(id=0) R6_w=invP(id=0,umax_value=808464432,var_off=(0x0; 0x30303030)) R10=fp0
  4: (14) w6 -= 810299440
  5: R0_w=invP(id=0) R6_w=invP(id=0,umax_value=4294967295,var_off=(0xcf800000; 0x3077fff0)) R10=fp0
  5: (c4) w6 s>>= 1
  6: R0_w=invP(id=0) R6_w=invP(id=0,umin_value=3888119808,umax_value=4294705144,var_off=(0xe7c00000; 0x183bfff8)) R10=fp0
                                              (0x67c00000)          (0xfffbfff8)
  6: (76) if w6 s>= 0x30303030 goto pc+216
  7: R0_w=invP(id=0) R6_w=invP(id=0,umin_value=3888119808,umax_value=4294705144,var_off=(0xe7c00000; 0x183bfff8)) R10=fp0
  7: (30) r0 = *(u8 *)skb[808464432]
  BPF_LD_[ABS|IND] uses reserved fields
  processed 8 insns (limit 1000000) [...]

Fixes: 9cbe1f5a32dc ("bpf/verifier: improve register value range tracking with ARSH")
Reported-by: Anatoly Trosinenko <anatoly.trosinenko@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115204733.16648-1-daniel@iogearbox.net
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agodrm/amd/display: Reorder detect_edp_sink_caps before link settings read.
Mario Kleiner [Thu, 9 Jan 2020 15:20:27 +0000 (16:20 +0100)]
drm/amd/display: Reorder detect_edp_sink_caps before link settings read.

[ Upstream commit 3b7c59754cc22760760a84ebddb8e0b1e8dd871b ]

read_current_link_settings_on_detect() on eDP 1.4+ may use the
edp_supported_link_rates table which is set up by
detect_edp_sink_caps(), so that function needs to be called first.

Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Cc: Martin Leung <martin.leung@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Sasha Levin <sashal@kernel.org>
4 years agoblock: Fix the type of 'sts' in bsg_queue_rq()
Bart Van Assche [Wed, 18 Dec 2019 00:23:29 +0000 (16:23 -0800)]
block: Fix the type of 'sts' in bsg_queue_rq()

commit c44a4edb20938c85b64a256661443039f5bffdea upstream.

This patch fixes the following sparse warnings:

block/bsg-lib.c:269:19: warning: incorrect type in initializer (different base types)
block/bsg-lib.c:269:19:    expected int sts
block/bsg-lib.c:269:19:    got restricted blk_status_t [usertype]
block/bsg-lib.c:286:16: warning: incorrect type in return expression (different base types)
block/bsg-lib.c:286:16:    expected restricted blk_status_t
block/bsg-lib.c:286:16:    got int [assigned] sts

Cc: Martin Wilck <mwilck@suse.com>
Fixes: d46fe2cb2dce ("block: drop device references in bsg_queue_rq()")
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: fix kernel-doc warning in <linux/netdevice.h>
Randy Dunlap [Tue, 17 Dec 2019 02:52:45 +0000 (18:52 -0800)]
net: fix kernel-doc warning in <linux/netdevice.h>

commit 1f26c0d3d24125992ab0026b0dab16c08df947c7 upstream.

Fix missing '*' kernel-doc notation that causes this warning:

../include/linux/netdevice.h:1779: warning: bad line:                                 spinlock

Fixes: ab92d68fc22f ("net: core: add generic lockdep keys")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agotipc: fix retrans failure due to wrong destination
Tuong Lien [Tue, 10 Dec 2019 08:21:04 +0000 (15:21 +0700)]
tipc: fix retrans failure due to wrong destination

commit abc9b4e0549b93fdaff56e9532bc49a2d7b04955 upstream.

When a user message is sent, TIPC will check if the socket has faced a
congestion at link layer. If that happens, it will make a sleep to wait
for the congestion to disappear. This leaves a gap for other users to
take over the socket (e.g. multi threads) since the socket is released
as well. Also, in case of connectionless (e.g. SOCK_RDM), user is free
to send messages to various destinations (e.g. via 'sendto()'), then
the socket's preformatted header has to be updated correspondingly
prior to the actual payload message building.

Unfortunately, the latter action is done before the first action which
causes a condition issue that the destination of a certain message can
be modified incorrectly in the middle, leading to wrong destination
when that message is built. Consequently, when the message is sent to
the link layer, it gets stuck there forever because the peer node will
simply reject it. After a number of retransmission attempts, the link
is eventually taken down and the retransmission failure is reported.

This commit fixes the problem by rearranging the order of actions to
prevent the race condition from occurring, so the message building is
'atomic' and its header will not be modified by anyone.

Fixes: 365ad353c256 ("tipc: reduce risk of user starvation during link congestion")
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agotipc: fix potential hanging after b/rcast changing
Tuong Lien [Tue, 10 Dec 2019 08:21:03 +0000 (15:21 +0700)]
tipc: fix potential hanging after b/rcast changing

commit dca4a17d24ee9d878836ce5eb8dc25be1ffa5729 upstream.

In commit c55c8edafa91 ("tipc: smooth change between replicast and
broadcast"), we allow instant switching between replicast and broadcast
by sending a dummy 'SYN' packet on the last used link to synchronize
packets on the links. The 'SYN' message is an object of link congestion
also, so if that happens, a 'SOCK_WAKEUP' will be scheduled to be sent
back to the socket...
However, in that commit, we simply use the same socket 'cong_link_cnt'
counter for both the 'SYN' & normal payload message sending. Therefore,
if both the replicast & broadcast links are congested, the counter will
be not updated correctly but overwritten by the latter congestion.
Later on, when the 'SOCK_WAKEUP' messages are processed, the counter is
reduced one by one and eventually overflowed. Consequently, further
activities on the socket will only wait for the false congestion signal
to disappear but never been met.

Because sending the 'SYN' message is vital for the mechanism, it should
be done anyway. This commit fixes the issue by marking the message with
an error code e.g. 'TIPC_ERR_NO_PORT', so its sending should not face a
link congestion, there is no need to touch the socket 'cong_link_cnt'
either. In addition, in the event of any error (e.g. -ENOBUFS), we will
purge the entire payload message queue and make a return immediately.

Fixes: c55c8edafa91 ("tipc: smooth change between replicast and broadcast")
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoreset: Fix {of,devm}_reset_control_array_get kerneldoc return types
Geert Uytterhoeven [Wed, 20 Nov 2019 14:26:13 +0000 (15:26 +0100)]
reset: Fix {of,devm}_reset_control_array_get kerneldoc return types

commit 723c0011c7f6992f57e2c629fa9c89141acc115f upstream.

of_reset_control_array_get() and devm_reset_control_array_get() return
struct reset_control pointers, not internal struct reset_control_array
pointers, just like all other reset control API calls.

Correct the kerneldoc to match the code.

Fixes: 17c82e206d2a3cd8 ("reset: Add APIs to manage array of resets")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: stmmac: Enable 16KB buffer size
Jose Abreu [Wed, 18 Dec 2019 10:17:42 +0000 (11:17 +0100)]
net: stmmac: Enable 16KB buffer size

commit b2f3a481c4cd62f78391b836b64c0a6e72b503d2 upstream.

XGMAC supports maximum MTU that can go to 16KB. Lets add this check in
the calculation of RX buffer size.

Fixes: 7ac6653a085b ("stmmac: Move the STMicroelectronics driver")
Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agonet: stmmac: 16KB buffer must be 16 byte aligned
Jose Abreu [Wed, 18 Dec 2019 10:17:41 +0000 (11:17 +0100)]
net: stmmac: 16KB buffer must be 16 byte aligned

commit 8605131747e7e1fd8f6c9f97a00287aae2b2c640 upstream.

The 16KB RX Buffer must also be 16 byte aligned. Fix it.

Fixes: 7ac6653a085b ("stmmac: Move the STMicroelectronics driver")
Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoARM: dts: imx7: Fix Toradex Colibri iMX7S 256MB NAND flash support
Marcel Ziswiler [Wed, 8 Jan 2020 16:12:31 +0000 (17:12 +0100)]
ARM: dts: imx7: Fix Toradex Colibri iMX7S 256MB NAND flash support

commit 4b0b97e651ecf29f20248420b52b6864fbd40bc2 upstream.

Turns out when introducing the eMMC version the gpmi node required for
NAND flash support got enabled exclusively on Colibri iMX7D 512MB.

Fixes: f928a4a377e4 ("ARM: dts: imx7: add Toradex Colibri iMX7D 1GB (eMMC) support")
Signed-off-by: Marcel Ziswiler <marcel.ziswiler@toradex.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoARM: dts: imx6q-icore-mipi: Use 1.5 version of i.Core MX6DL
Jagan Teki [Mon, 30 Dec 2019 12:00:19 +0000 (17:30 +0530)]
ARM: dts: imx6q-icore-mipi: Use 1.5 version of i.Core MX6DL

commit 4a132f60808ae3a751e107a373f8572012352d3c upstream.

The EDIMM STARTER KIT i.Core 1.5 MIPI Evaluation is based on
the 1.5 version of the i.Core MX6 cpu module. The 1.5 version
differs from the original one for a few details, including the
ethernet PHY interface clock provider.

With this commit, the ethernet interface works properly:
SMSC LAN8710/LAN8720 2188000.ethernet-1:00: attached PHY driver

While before using the 1.5 version, ethernet failed to startup
do to un-clocked PHY interface:
fec 2188000.ethernet eth0: could not attach to PHY

Similar fix has merged for i.Core MX6Q but missed to update for DL.

Fixes: a8039f2dd089 ("ARM: dts: imx6dl: Add Engicam i.CoreM6 1.5 Quad/Dual MIPI starter kit support")
Cc: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Michael Trimarchi <michael@amarulasolutions.com>
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoARM: dts: imx6sll-evk: Remove incorrect power supply assignment
Anson Huang [Mon, 30 Dec 2019 01:41:10 +0000 (09:41 +0800)]
ARM: dts: imx6sll-evk: Remove incorrect power supply assignment

commit 3479b2843c78ffb60247f522226ba68f93aee355 upstream.

The vdd3p0 LDO's input should be from external USB VBUS directly, NOT
PMIC's power supply, the vdd3p0 LDO's target output voltage can be
controlled by SW, and it requires input voltage to be high enough, with
incorrect power supply assigned, if the power supply's voltage is lower
than the LDO target output voltage, it will return fail and skip the LDO
voltage adjustment, so remove the power supply assignment for vdd3p0 to
avoid such scenario.

Fixes: 96a9169cf621 ("ARM: dts: imx6sll-evk: Assign corresponding power supply for vdd3p0")
Signed-off-by: Anson Huang <Anson.Huang@nxp.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoARM: dts: imx6sl-evk: Remove incorrect power supply assignment
Anson Huang [Mon, 30 Dec 2019 01:41:09 +0000 (09:41 +0800)]
ARM: dts: imx6sl-evk: Remove incorrect power supply assignment

commit b4eb9ef0e29cd28c6fd684e0ab77bda824acb20e upstream.

The vdd3p0 LDO's input should be from external USB VBUS directly, NOT
PMIC's power supply, the vdd3p0 LDO's target output voltage can be
controlled by SW, and it requires input voltage to be high enough, with
incorrect power supply assigned, if the power supply's voltage is lower
than the LDO target output voltage, it will return fail and skip the LDO
voltage adjustment, so remove the power supply assignment for vdd3p0 to
avoid such scenario.

Fixes: 3feea8805d6f ("ARM: dts: imx6sl-evk: Assign corresponding power supply for LDOs")
Signed-off-by: Anson Huang <Anson.Huang@nxp.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoARM: dts: imx6sx-sdb: Remove incorrect power supply assignment
Anson Huang [Mon, 30 Dec 2019 01:41:08 +0000 (09:41 +0800)]
ARM: dts: imx6sx-sdb: Remove incorrect power supply assignment

commit d4918ebb5c256d26696a13e78ac68c146111191a upstream.

The vdd3p0 LDO's input should be from external USB VBUS directly, NOT
PMIC's power supply, the vdd3p0 LDO's target output voltage can be
controlled by SW, and it requires input voltage to be high enough, with
incorrect power supply assigned, if the power supply's voltage is lower
than the LDO target output voltage, it will return fail and skip the LDO
voltage adjustment, so remove the power supply assignment for vdd3p0 to
avoid such scenario.

Fixes: 37a4bdead109 ("ARM: dts: imx6sx-sdb: Assign corresponding power supply for LDOs")
Signed-off-by: Anson Huang <Anson.Huang@nxp.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoARM: dts: imx6qdl-sabresd: Remove incorrect power supply assignment
Anson Huang [Mon, 30 Dec 2019 01:41:07 +0000 (09:41 +0800)]
ARM: dts: imx6qdl-sabresd: Remove incorrect power supply assignment

commit 4521de30fbb3f5be0db58de93582ebce72c9d44f upstream.

The vdd3p0 LDO's input should be from external USB VBUS directly, NOT
PMIC's power supply, the vdd3p0 LDO's target output voltage can be
controlled by SW, and it requires input voltage to be high enough, with
incorrect power supply assigned, if the power supply's voltage is lower
than the LDO target output voltage, it will return fail and skip the LDO
voltage adjustment, so remove the power supply assignment for vdd3p0 to
avoid such scenario.

Fixes: 93385546ba36 ("ARM: dts: imx6qdl-sabresd: Assign corresponding power supply for LDOs")
Signed-off-by: Anson Huang <Anson.Huang@nxp.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomm: khugepaged: add trace status description for SCAN_PAGE_HAS_PRIVATE
Yang Shi [Tue, 14 Jan 2020 00:29:36 +0000 (16:29 -0800)]
mm: khugepaged: add trace status description for SCAN_PAGE_HAS_PRIVATE

commit 554913f600b45d73de12ad58c1ac7baa0f22a703 upstream.

Commit 99cb0dbd47a1 ("mm,thp: add read-only THP support for (non-shmem)
FS") introduced a new khugepaged scan result: SCAN_PAGE_HAS_PRIVATE, but
the corresponding description for trace events were not added.

Link: http://lkml.kernel.org/r/1574793844-2914-1-git-send-email-yang.shi@linux.alibaba.com
Fixes: 99cb0dbd47a1 ("mm,thp: add read-only THP support for (non-shmem) FS")
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Song Liu <songliubraving@fb.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomm/page-writeback.c: avoid potential division by zero in wb_min_max_ratio()
Wen Yang [Tue, 14 Jan 2020 00:29:23 +0000 (16:29 -0800)]
mm/page-writeback.c: avoid potential division by zero in wb_min_max_ratio()

commit 6d9e8c651dd979aa666bee15f086745f3ea9c4b3 upstream.

Patch series "use div64_ul() instead of div_u64() if the divisor is
unsigned long".

We were first inspired by commit b0ab99e7736a ("sched: Fix possible divide
by zero in avg_atom () calculation"), then refer to the recently analyzed
mm code, we found this suspicious place.

 201                 if (min) {
 202                         min *= this_bw;
 203                         do_div(min, tot_bw);
 204                 }

And we also disassembled and confirmed it:

  /usr/src/debug/kernel-4.9.168-016.ali3000/linux-4.9.168-016.ali3000.alios7.x86_64/mm/page-writeback.c: 201
  0xffffffff811c37da <__wb_calc_thresh+234>:      xor    %r10d,%r10d
  0xffffffff811c37dd <__wb_calc_thresh+237>:      test   %rax,%rax
  0xffffffff811c37e0 <__wb_calc_thresh+240>:      je 0xffffffff811c3800 <__wb_calc_thresh+272>
  /usr/src/debug/kernel-4.9.168-016.ali3000/linux-4.9.168-016.ali3000.alios7.x86_64/mm/page-writeback.c: 202
  0xffffffff811c37e2 <__wb_calc_thresh+242>:      imul   %r8,%rax
  /usr/src/debug/kernel-4.9.168-016.ali3000/linux-4.9.168-016.ali3000.alios7.x86_64/mm/page-writeback.c: 203
  0xffffffff811c37e6 <__wb_calc_thresh+246>:      mov    %r9d,%r10d    ---> truncates it to 32 bits here
  0xffffffff811c37e9 <__wb_calc_thresh+249>:      xor    %edx,%edx
  0xffffffff811c37eb <__wb_calc_thresh+251>:      div    %r10
  0xffffffff811c37ee <__wb_calc_thresh+254>:      imul   %rbx,%rax
  0xffffffff811c37f2 <__wb_calc_thresh+258>:      shr    $0x2,%rax
  0xffffffff811c37f6 <__wb_calc_thresh+262>:      mul    %rcx
  0xffffffff811c37f9 <__wb_calc_thresh+265>:      shr    $0x2,%rdx
  0xffffffff811c37fd <__wb_calc_thresh+269>:      mov    %rdx,%r10

This series uses div64_ul() instead of div_u64() if the divisor is
unsigned long, to avoid truncation to 32-bit on 64-bit platforms.

This patch (of 3):

The variables 'min' and 'max' are unsigned long and do_div truncates
them to 32 bits, which means it can test non-zero and be truncated to
zero for division.  Fix this issue by using div64_ul() instead.

Link: http://lkml.kernel.org/r/20200102081442.8273-2-wenyang@linux.alibaba.com
Fixes: 693108a8a667 ("writeback: make bdi->min/max_ratio handling cgroup writeback aware")
Signed-off-by: Wen Yang <wenyang@linux.alibaba.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Qian Cai <cai@lca.pw>
Cc: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomm/memory_hotplug: don't free usage map when removing a re-added early section
David Hildenbrand [Tue, 14 Jan 2020 00:29:07 +0000 (16:29 -0800)]
mm/memory_hotplug: don't free usage map when removing a re-added early section

commit 8068df3b60373c390198f660574ea14c8098de57 upstream.

When we remove an early section, we don't free the usage map, as the
usage maps of other sections are placed into the same page.  Once the
section is removed, it is no longer an early section (especially, the
memmap is freed).  When we re-add that section, the usage map is reused,
however, it is no longer an early section.  When removing that section
again, we try to kfree() a usage map that was allocated during early
boot - bad.

Let's check against PageReserved() to see if we are dealing with an
usage map that was allocated during boot.  We could also check against
!(PageSlab(usage_page) || PageCompound(usage_page)), but PageReserved() is
cleaner.

Can be triggered using memtrace under ppc64/powernv:

  $ mount -t debugfs none /sys/kernel/debug/
  $ echo 0x20000000 > /sys/kernel/debug/powerpc/memtrace/enable
  $ echo 0x20000000 > /sys/kernel/debug/powerpc/memtrace/enable
   ------------[ cut here ]------------
   kernel BUG at mm/slub.c:3969!
   Oops: Exception in kernel mode, sig: 5 [#1]
   LE PAGE_SIZE=3D64K MMU=3DHash SMP NR_CPUS=3D2048 NUMA PowerNV
   Modules linked in:
   CPU: 0 PID: 154 Comm: sh Not tainted 5.5.0-rc2-next-20191216-00005-g0be1dba7b7c0 #61
   NIP kfree+0x338/0x3b0
   LR section_deactivate+0x138/0x200
   Call Trace:
     section_deactivate+0x138/0x200
     __remove_pages+0x114/0x150
     arch_remove_memory+0x3c/0x160
     try_remove_memory+0x114/0x1a0
     __remove_memory+0x20/0x40
     memtrace_enable_set+0x254/0x850
     simple_attr_write+0x138/0x160
     full_proxy_write+0x8c/0x110
     __vfs_write+0x38/0x70
     vfs_write+0x11c/0x2a0
     ksys_write+0x84/0x140
     system_call+0x5c/0x68
   ---[ end trace 4b053cbd84e0db62 ]---

The first invocation will offline+remove memory blocks.  The second
invocation will first add+online them again, in order to offline+remove
them again (usually we are lucky and the exact same memory blocks will
get "reallocated").

Tested on powernv with boot memory: The usage map will not get freed.
Tested on x86-64 with DIMMs: The usage map will get freed.

Using Dynamic Memory under a Power DLAPR can trigger it easily.

Triggering removal (I assume after previously removed+re-added) of
memory from the HMC GUI can crash the kernel with the same call trace
and is fixed by this patch.

Link: http://lkml.kernel.org/r/20191217104637.5509-1-david@redhat.com
Fixes: 326e1b8f83a4 ("mm/sparsemem: introduce a SECTION_IS_EARLY flag")
Signed-off-by: David Hildenbrand <david@redhat.com>
Tested-by: Pingfan Liu <piliu@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoBtrfs: always copy scrub arguments back to user space
Filipe Manana [Thu, 16 Jan 2020 11:29:20 +0000 (11:29 +0000)]
Btrfs: always copy scrub arguments back to user space

commit 5afe6ce748c1ea99e0d648153c05075e1ab93afb upstream.

If scrub returns an error we are not copying back the scrub arguments
structure to user space. This prevents user space to know how much
progress scrub has done if an error happened - this includes -ECANCELED
which is returned when users ask for scrub to stop. A particular use
case, which is used in btrfs-progs, is to resume scrub after it is
canceled, in that case it relies on checking the progress from the scrub
arguments structure and then use that progress in a call to resume
scrub.

So fix this by always copying the scrub arguments structure to user
space, overwriting the value returned to user space with -EFAULT only if
copying the structure failed to let user space know that either that
copying did not happen, and therefore the structure is stale, or it
happened partially and the structure is probably not valid and corrupt
due to the partial copy.

Reported-by: Graham Cobb <g.btrfs@cobb.uk.net>
Link: https://lore.kernel.org/linux-btrfs/d0a97688-78be-08de-ca7d-bcb4c7fb397e@cobb.uk.net/
Fixes: 06fe39ab15a6a4 ("Btrfs: do not overwrite scrub error with fault error in scrub ioctl")
CC: stable@vger.kernel.org # 5.1+
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Tested-by: Graham Cobb <g.btrfs@cobb.uk.net>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobtrfs: check rw_devices, not num_devices for balance
Josef Bacik [Fri, 10 Jan 2020 16:11:24 +0000 (11:11 -0500)]
btrfs: check rw_devices, not num_devices for balance

commit b35cf1f0bf1f2b0b193093338414b9bd63b29015 upstream.

The fstest btrfs/154 reports

  [ 8675.381709] BTRFS: Transaction aborted (error -28)
  [ 8675.383302] WARNING: CPU: 1 PID: 31900 at fs/btrfs/block-group.c:2038 btrfs_create_pending_block_groups+0x1e0/0x1f0 [btrfs]
  [ 8675.390925] CPU: 1 PID: 31900 Comm: btrfs Not tainted 5.5.0-rc6-default+ #935
  [ 8675.392780] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba527-rebuilt.opensuse.org 04/01/2014
  [ 8675.395452] RIP: 0010:btrfs_create_pending_block_groups+0x1e0/0x1f0 [btrfs]
  [ 8675.402672] RSP: 0018:ffffb2090888fb00 EFLAGS: 00010286
  [ 8675.404413] RAX: 0000000000000000 RBX: ffff92026dfa91c8 RCX: 0000000000000001
  [ 8675.406609] RDX: 0000000000000000 RSI: ffffffff8e100899 RDI: ffffffff8e100971
  [ 8675.408775] RBP: ffff920247c61660 R08: 0000000000000000 R09: 0000000000000000
  [ 8675.410978] R10: 0000000000000000 R11: 0000000000000000 R12: 00000000ffffffe4
  [ 8675.412647] R13: ffff92026db74000 R14: ffff920247c616b8 R15: ffff92026dfbc000
  [ 8675.413994] FS:  00007fd5e57248c0(0000) GS:ffff92027d800000(0000) knlGS:0000000000000000
  [ 8675.416146] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  [ 8675.417833] CR2: 0000564aa51682d8 CR3: 000000006dcbc004 CR4: 0000000000160ee0
  [ 8675.419801] Call Trace:
  [ 8675.420742]  btrfs_start_dirty_block_groups+0x355/0x480 [btrfs]
  [ 8675.422600]  btrfs_commit_transaction+0xc8/0xaf0 [btrfs]
  [ 8675.424335]  reset_balance_state+0x14a/0x190 [btrfs]
  [ 8675.425824]  btrfs_balance.cold+0xe7/0x154 [btrfs]
  [ 8675.427313]  ? kmem_cache_alloc_trace+0x235/0x2c0
  [ 8675.428663]  btrfs_ioctl_balance+0x298/0x350 [btrfs]
  [ 8675.430285]  btrfs_ioctl+0x466/0x2550 [btrfs]
  [ 8675.431788]  ? mem_cgroup_charge_statistics+0x51/0xf0
  [ 8675.433487]  ? mem_cgroup_commit_charge+0x56/0x400
  [ 8675.435122]  ? do_raw_spin_unlock+0x4b/0xc0
  [ 8675.436618]  ? _raw_spin_unlock+0x1f/0x30
  [ 8675.438093]  ? __handle_mm_fault+0x499/0x740
  [ 8675.439619]  ? do_vfs_ioctl+0x56e/0x770
  [ 8675.441034]  do_vfs_ioctl+0x56e/0x770
  [ 8675.442411]  ksys_ioctl+0x3a/0x70
  [ 8675.443718]  ? trace_hardirqs_off_thunk+0x1a/0x1c
  [ 8675.445333]  __x64_sys_ioctl+0x16/0x20
  [ 8675.446705]  do_syscall_64+0x50/0x210
  [ 8675.448059]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
  [ 8675.479187] BTRFS: error (device vdb) in btrfs_create_pending_block_groups:2038: errno=-28 No space left

We now use btrfs_can_overcommit() to see if we can flip a block group
read only.  Before this would fail because we weren't taking into
account the usable un-allocated space for allocating chunks.  With my
patches we were allowed to do the balance, which is technically correct.

The test is trying to start balance on degraded mount.  So now we're
trying to allocate a chunk and cannot because we want to allocate a
RAID1 chunk, but there's only 1 device that's available for usage.  This
results in an ENOSPC.

But we shouldn't even be making it this far, we don't have enough
devices to restripe.  The problem is we're using btrfs_num_devices(),
that also includes missing devices. That's not actually what we want, we
need to use rw_devices.

The chunk_mutex is not needed here, rw_devices changes only in device
add, remove or replace, all are excluded by EXCL_OP mechanism.

Fixes: e4d8ec0f65b9 ("Btrfs: implement online profile changing")
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add stacktrace, update changelog, drop chunk_mutex ]
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobtrfs: fix memory leak in qgroup accounting
Johannes Thumshirn [Wed, 8 Jan 2020 12:07:32 +0000 (21:07 +0900)]
btrfs: fix memory leak in qgroup accounting

commit 26ef8493e1ab771cb01d27defca2fa1315dc3980 upstream.

When running xfstests on the current btrfs I get the following splat from
kmemleak:

unreferenced object 0xffff88821b2404e0 (size 32):
  comm "kworker/u4:7", pid 26663, jiffies 4295283698 (age 8.776s)
  hex dump (first 32 bytes):
    01 00 00 00 00 00 00 00 10 ff fd 26 82 88 ff ff  ...........&....
    10 ff fd 26 82 88 ff ff 20 ff fd 26 82 88 ff ff  ...&.... ..&....
  backtrace:
    [<00000000f94fd43f>] ulist_alloc+0x25/0x60 [btrfs]
    [<00000000fd023d99>] btrfs_find_all_roots_safe+0x41/0x100 [btrfs]
    [<000000008f17bd32>] btrfs_find_all_roots+0x52/0x70 [btrfs]
    [<00000000b7660afb>] btrfs_qgroup_rescan_worker+0x343/0x680 [btrfs]
    [<0000000058e66778>] btrfs_work_helper+0xac/0x1e0 [btrfs]
    [<00000000f0188930>] process_one_work+0x1cf/0x350
    [<00000000af5f2f8e>] worker_thread+0x28/0x3c0
    [<00000000b55a1add>] kthread+0x109/0x120
    [<00000000f88cbd17>] ret_from_fork+0x35/0x40

This corresponds to:

  (gdb) l *(btrfs_find_all_roots_safe+0x41)
  0x8d7e1 is in btrfs_find_all_roots_safe (fs/btrfs/backref.c:1413).
  1408
  1409            tmp = ulist_alloc(GFP_NOFS);
  1410            if (!tmp)
  1411                    return -ENOMEM;
  1412            *roots = ulist_alloc(GFP_NOFS);
  1413            if (!*roots) {
  1414                    ulist_free(tmp);
  1415                    return -ENOMEM;
  1416            }
  1417

Following the lifetime of the allocated 'roots' ulist, it gets freed
again in btrfs_qgroup_account_extent().

But this does not happen if the function is called with the
'BTRFS_FS_QUOTA_ENABLED' flag cleared, then btrfs_qgroup_account_extent()
does a short leave and directly returns.

Instead of directly returning we should jump to the 'out_free' in order to
free all resources as expected.

CC: stable@vger.kernel.org # 4.14+
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
[ add comment ]
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobtrfs: relocation: fix reloc_root lifespan and access
Qu Wenruo [Wed, 8 Jan 2020 05:12:00 +0000 (13:12 +0800)]
btrfs: relocation: fix reloc_root lifespan and access

commit 6282675e6708ec78518cc0e9ad1f1f73d7c5c53d upstream.

[BUG]
There are several different KASAN reports for balance + snapshot
workloads.  Involved call paths include:

   should_ignore_root+0x54/0xb0 [btrfs]
   build_backref_tree+0x11af/0x2280 [btrfs]
   relocate_tree_blocks+0x391/0xb80 [btrfs]
   relocate_block_group+0x3e5/0xa00 [btrfs]
   btrfs_relocate_block_group+0x240/0x4d0 [btrfs]
   btrfs_relocate_chunk+0x53/0xf0 [btrfs]
   btrfs_balance+0xc91/0x1840 [btrfs]
   btrfs_ioctl_balance+0x416/0x4e0 [btrfs]
   btrfs_ioctl+0x8af/0x3e60 [btrfs]
   do_vfs_ioctl+0x831/0xb10

   create_reloc_root+0x9f/0x460 [btrfs]
   btrfs_reloc_post_snapshot+0xff/0x6c0 [btrfs]
   create_pending_snapshot+0xa9b/0x15f0 [btrfs]
   create_pending_snapshots+0x111/0x140 [btrfs]
   btrfs_commit_transaction+0x7a6/0x1360 [btrfs]
   btrfs_mksubvol+0x915/0x960 [btrfs]
   btrfs_ioctl_snap_create_transid+0x1d5/0x1e0 [btrfs]
   btrfs_ioctl_snap_create_v2+0x1d3/0x270 [btrfs]
   btrfs_ioctl+0x241b/0x3e60 [btrfs]
   do_vfs_ioctl+0x831/0xb10

   btrfs_reloc_pre_snapshot+0x85/0xc0 [btrfs]
   create_pending_snapshot+0x209/0x15f0 [btrfs]
   create_pending_snapshots+0x111/0x140 [btrfs]
   btrfs_commit_transaction+0x7a6/0x1360 [btrfs]
   btrfs_mksubvol+0x915/0x960 [btrfs]
   btrfs_ioctl_snap_create_transid+0x1d5/0x1e0 [btrfs]
   btrfs_ioctl_snap_create_v2+0x1d3/0x270 [btrfs]
   btrfs_ioctl+0x241b/0x3e60 [btrfs]
   do_vfs_ioctl+0x831/0xb10

[CAUSE]
All these call sites are only relying on root->reloc_root, which can
undergo btrfs_drop_snapshot(), and since we don't have real refcount
based protection to reloc roots, we can reach already dropped reloc
root, triggering KASAN.

[FIX]
To avoid such access to unstable root->reloc_root, we should check
BTRFS_ROOT_DEAD_RELOC_TREE bit first.

This patch introduces wrappers that provide the correct way to check the
bit with memory barriers protection.

Most callers don't distinguish merged reloc tree and no reloc tree.  The
only exception is should_ignore_root(), as merged reloc tree can be
ignored, while no reloc tree shouldn't.

[CRITICAL SECTION ANALYSIS]
Although test_bit()/set_bit()/clear_bit() doesn't imply a barrier, the
DEAD_RELOC_TREE bit has extra help from transaction as a higher level
barrier, the lifespan of root::reloc_root and DEAD_RELOC_TREE bit are:

NULL: reloc_root is NULL PTR: reloc_root is not NULL
0: DEAD_RELOC_ROOT bit not set DEAD: DEAD_RELOC_ROOT bit set

(NULL, 0)    Initial state  __
  |  /\ Section A
        btrfs_init_reloc_root()  \/
  |    __
(PTR, 0)     reloc_root initialized      /\
          |  |
btrfs_update_reloc_root()  |  Section B
          |  |
(PTR, DEAD)  reloc_root has been merged  \/
          |  __
=== btrfs_commit_transaction() ====================
  |  /\
clean_dirty_subvols()  |
  |  |  Section C
(NULL, DEAD) reloc_root cleanup starts   \/
          |  __
btrfs_drop_snapshot()  /\
  |  |  Section D
(NULL, 0)    Back to initial state  \/

Every have_reloc_root() or test_bit(DEAD_RELOC_ROOT) caller holds
transaction handle, so none of such caller can cross transaction boundary.

In Section A, every caller just found no DEAD bit, and grab reloc_root.

In the cross section A-B, caller may get no DEAD bit, but since reloc_root
is still completely valid thus accessing reloc_root is completely safe.

No test_bit() caller can cross the boundary of Section B and Section C.

In Section C, every caller found the DEAD bit, so no one will access
reloc_root.

In the cross section C-D, either caller gets the DEAD bit set, avoiding
access reloc_root no matter if it's safe or not.  Or caller get the DEAD
bit cleared, then access reloc_root, which is already NULL, nothing will
be wrong.

The memory write barriers are between the reloc_root updates and bit
set/clear, the pairing read side is before test_bit.

Reported-by: Zygo Blaxell <ce3g8jdj@umail.furryterror.org>
Fixes: d2311e698578 ("btrfs: relocation: Delay reloc tree deletion after merge_reloc_roots")
CC: stable@vger.kernel.org # 5.4+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ barriers ]
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobtrfs: do not delete mismatched root refs
Josef Bacik [Wed, 18 Dec 2019 22:20:29 +0000 (17:20 -0500)]
btrfs: do not delete mismatched root refs

commit 423a716cd7be16fb08690760691befe3be97d3fc upstream.

btrfs_del_root_ref() will simply WARN_ON() if the ref doesn't match in
any way, and then continue to delete the reference.  This shouldn't
happen, we have these values because there's more to the reference than
the original root and the sub root.  If any of these checks fail, return
-ENOENT.

CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobtrfs: fix invalid removal of root ref
Josef Bacik [Wed, 18 Dec 2019 22:20:28 +0000 (17:20 -0500)]
btrfs: fix invalid removal of root ref

commit d49d3287e74ffe55ae7430d1e795e5f9bf7359ea upstream.

If we have the following sequence of events

  btrfs sub create A
  btrfs sub create A/B
  btrfs sub snap A C
  mkdir C/foo
  mv A/B C/foo
  rm -rf *

We will end up with a transaction abort.

The reason for this is because we create a root ref for B pointing to A.
When we create a snapshot of C we still have B in our tree, but because
the root ref points to A and not C we will make it appear to be empty.

The problem happens when we move B into C.  This removes the root ref
for B pointing to A and adds a ref of B pointing to C.  When we rmdir C
we'll see that we have a ref to our root and remove the root ref,
despite not actually matching our reference name.

Now btrfs_del_root_ref() allowing this to work is a bug as well, however
we know that this inode does not actually point to a root ref in the
first place, so we shouldn't be calling btrfs_del_root_ref() in the
first place and instead simply look up our dir index for this item and
do the rest of the removal.

CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agobtrfs: rework arguments of btrfs_unlink_subvol
Josef Bacik [Wed, 18 Dec 2019 22:20:27 +0000 (17:20 -0500)]
btrfs: rework arguments of btrfs_unlink_subvol

[ Upstream commit 045d3967b6920b663fc010ad414ade1b24143bd1 ]

btrfs_unlink_subvol takes the name of the dentry and the root objectid
based on what kind of inode this is, either a real subvolume link or a
empty one that we inherited as a snapshot.  We need to fix how we unlink
in the case for BTRFS_EMPTY_SUBVOL_DIR_OBJECTID in the future, so rework
btrfs_unlink_subvol to just take the dentry and handle getting the right
objectid given the type of inode this is.  There is no functional change
here, simply pushing the work into btrfs_unlink_subvol() proper.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
4 years agomm, debug_pagealloc: don't rely on static keys too early
Vlastimil Babka [Tue, 14 Jan 2020 00:29:20 +0000 (16:29 -0800)]
mm, debug_pagealloc: don't rely on static keys too early

commit 8e57f8acbbd121ecfb0c9dc13b8b030f86c6bd3b upstream.

Commit 96a2b03f281d ("mm, debug_pagelloc: use static keys to enable
debugging") has introduced a static key to reduce overhead when
debug_pagealloc is compiled in but not enabled.  It relied on the
assumption that jump_label_init() is called before parse_early_param()
as in start_kernel(), so when the "debug_pagealloc=on" option is parsed,
it is safe to enable the static key.

However, it turns out multiple architectures call parse_early_param()
earlier from their setup_arch().  x86 also calls jump_label_init() even
earlier, so no issue was found while testing the commit, but same is not
true for e.g.  ppc64 and s390 where the kernel would not boot with
debug_pagealloc=on as found by our QA.

To fix this without tricky changes to init code of multiple
architectures, this patch partially reverts the static key conversion
from 96a2b03f281d.  Init-time and non-fastpath calls (such as in arch
code) of debug_pagealloc_enabled() will again test a simple bool
variable.  Fastpath mm code is converted to a new
debug_pagealloc_enabled_static() variant that relies on the static key,
which is enabled in a well-defined point in mm_init() where it's
guaranteed that jump_label_init() has been called, regardless of
architecture.

[sfr@canb.auug.org.au: export _debug_pagealloc_enabled_early]
Link: http://lkml.kernel.org/r/20200106164944.063ac07b@canb.auug.org.au
Link: http://lkml.kernel.org/r/20191219130612.23171-1-vbabka@suse.cz
Fixes: 96a2b03f281d ("mm, debug_pagelloc: use static keys to enable debugging")
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Qian Cai <cai@lca.pw>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomm: memcg/slab: call flush_memcg_workqueue() only if memcg workqueue is valid
Adrian Huang [Tue, 14 Jan 2020 00:29:32 +0000 (16:29 -0800)]
mm: memcg/slab: call flush_memcg_workqueue() only if memcg workqueue is valid

commit 2fe20210fc5f5e62644678b8f927c49f2c6f42a7 upstream.

When booting with amd_iommu=off, the following WARNING message
appears:

  AMD-Vi: AMD IOMMU disabled on kernel command-line
  ------------[ cut here ]------------
  WARNING: CPU: 0 PID: 0 at kernel/workqueue.c:2772 flush_workqueue+0x42e/0x450
  Modules linked in:
  CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.5.0-rc3-amd-iommu #6
  Hardware name: Lenovo ThinkSystem SR655-2S/7D2WRCZ000, BIOS D8E101L-1.00 12/05/2019
  RIP: 0010:flush_workqueue+0x42e/0x450
  Code: ff 0f 0b e9 7a fd ff ff 4d 89 ef e9 33 fe ff ff 0f 0b e9 7f fd ff ff 0f 0b e9 bc fd ff ff 0f 0b e9 a8 fd ff ff e8 52 2c fe ff <0f> 0b 31 d2 48 c7 c6 e0 88 c5 95 48 c7 c7 d8 ad f0 95 e8 19 f5 04
  Call Trace:
   kmem_cache_destroy+0x69/0x260
   iommu_go_to_state+0x40c/0x5ab
   amd_iommu_prepare+0x16/0x2a
   irq_remapping_prepare+0x36/0x5f
   enable_IR_x2apic+0x21/0x172
   default_setup_apic_routing+0x12/0x6f
   apic_intr_mode_init+0x1a1/0x1f1
   x86_late_time_init+0x17/0x1c
   start_kernel+0x480/0x53f
   secondary_startup_64+0xb6/0xc0
  ---[ end trace 30894107c3749449 ]---
  x2apic: IRQ remapping doesn't support X2APIC mode
  x2apic disabled

The warning is caused by the calling of 'kmem_cache_destroy()'
in free_iommu_resources(). Here is the call path:

  free_iommu_resources
    kmem_cache_destroy
      flush_memcg_workqueue
        flush_workqueue

The root cause is that the IOMMU subsystem runs before the workqueue
subsystem, which the variable 'wq_online' is still 'false'.  This leads
to the statement 'if (WARN_ON(!wq_online))' in flush_workqueue() is
'true'.

Since the variable 'memcg_kmem_cache_wq' is not allocated during the
time, it is unnecessary to call flush_memcg_workqueue().  This prevents
the WARNING message triggered by flush_workqueue().

Link: http://lkml.kernel.org/r/20200103085503.1665-1-ahuang12@lenovo.com
Fixes: 92ee383f6daab ("mm: fix race between kmem_cache destroy, create and deactivate")
Signed-off-by: Adrian Huang <ahuang12@lenovo.com>
Reported-by: Xiaochun Lee <lixc17@lenovo.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomm: memcg/slab: fix percpu slab vmstats flushing
Roman Gushchin [Tue, 14 Jan 2020 00:29:16 +0000 (16:29 -0800)]
mm: memcg/slab: fix percpu slab vmstats flushing

commit 4a87e2a25dc27131c3cce5e94421622193305638 upstream.

Currently slab percpu vmstats are flushed twice: during the memcg
offlining and just before freeing the memcg structure.  Each time percpu
counters are summed, added to the atomic counterparts and propagated up
by the cgroup tree.

The second flushing is required due to how recursive vmstats are
implemented: counters are batched in percpu variables on a local level,
and once a percpu value is crossing some predefined threshold, it spills
over to atomic values on the local and each ascendant levels.  It means
that without flushing some numbers cached in percpu variables will be
dropped on floor each time a cgroup is destroyed.  And with uptime the
error on upper levels might become noticeable.

The first flushing aims to make counters on ancestor levels more
precise.  Dying cgroups may resume in the dying state for a long time.
After kmem_cache reparenting which is performed during the offlining
slab counters of the dying cgroup don't have any chances to be updated,
because any slab operations will be performed on the parent level.  It
means that the inaccuracy caused by percpu batching will not decrease up
to the final destruction of the cgroup.  By the original idea flushing
slab counters during the offlining should minimize the visible
inaccuracy of slab counters on the parent level.

The problem is that percpu counters are not zeroed after the first
flushing.  So every cached percpu value is summed twice.  It creates a
small error (up to 32 pages per cpu, but usually less) which accumulates
on parent cgroup level.  After creating and destroying of thousands of
child cgroups, slab counter on parent level can be way off the real
value.

For now, let's just stop flushing slab counters on memcg offlining.  It
can't be done correctly without scheduling a work on each cpu: reading
and zeroing it during css offlining can race with an asynchronous
update, which doesn't expect values to be changed underneath.

With this change, slab counters on parent level will become eventually
consistent.  Once all dying children are gone, values are correct.  And
if not, the error is capped by 32 * NR_CPUS pages per dying cgroup.

It's not perfect, as slab are reparented, so any updates after the
reparenting will happen on the parent level.  It means that if a slab
page was allocated, a counter on child level was bumped, then the page
was reparented and freed, the annihilation of positive and negative
counter values will not happen until the child cgroup is released.  It
makes slab counters different from others, and it might want us to
implement flushing in a correct form again.  But it's also a question of
performance: scheduling a work on each cpu isn't free, and it's an open
question if the benefit of having more accurate counters is worth it.

We might also consider flushing all counters on offlining, not only slab
counters.

So let's fix the main problem now: make the slab counters eventually
consistent, so at least the error won't grow with uptime (or more
precisely the number of created and destroyed cgroups).  And think about
the accuracy of counters separately.

Link: http://lkml.kernel.org/r/20191220042728.1045881-1-guro@fb.com
Fixes: bee07b33db78 ("mm: memcontrol: flush percpu slab vmstats on kmem offlining")
Signed-off-by: Roman Gushchin <guro@fb.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomm/huge_memory.c: thp: fix conflict of above-47bit hint address and PMD alignment
Kirill A. Shutemov [Tue, 14 Jan 2020 00:29:10 +0000 (16:29 -0800)]
mm/huge_memory.c: thp: fix conflict of above-47bit hint address and PMD alignment

commit 97d3d0f9a1cf132c63c0b8b8bd497b8a56283dd9 upstream.

Patch series "Fix two above-47bit hint address vs.  THP bugs".

The two get_unmapped_area() implementations have to be fixed to provide
THP-friendly mappings if above-47bit hint address is specified.

This patch (of 2):

Filesystems use thp_get_unmapped_area() to provide THP-friendly
mappings.  For DAX in particular.

Normally, the kernel doesn't create userspace mappings above 47-bit,
even if the machine allows this (such as with 5-level paging on x86-64).
Not all user space is ready to handle wide addresses.  It's known that
at least some JIT compilers use higher bits in pointers to encode their
information.

Userspace can ask for allocation from full address space by specifying
hint address (with or without MAP_FIXED) above 47-bits.  If the
application doesn't need a particular address, but wants to allocate
from whole address space it can specify -1 as a hint address.

Unfortunately, this trick breaks thp_get_unmapped_area(): the function
would not try to allocate PMD-aligned area if *any* hint address
specified.

Modify the routine to handle it correctly:

 - Try to allocate the space at the specified hint address with length
   padding required for PMD alignment.
 - If failed, retry without length padding (but with the same hint
   address);
 - If the returned address matches the hint address return it.
 - Otherwise, align the address as required for THP and return.

The user specified hint address is passed down to get_unmapped_area() so
above-47bit hint address will be taken into account without breaking
alignment requirements.

Link: http://lkml.kernel.org/r/20191220142548.7118-2-kirill.shutemov@linux.intel.com
Fixes: b569bab78d8d ("x86/mm: Prepare to expose larger address space to userspace")
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: Thomas Willhalm <thomas.willhalm@intel.com>
Tested-by: Dan Williams <dan.j.williams@intel.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: "Bruggeman, Otto G" <otto.g.bruggeman@intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agomm/shmem.c: thp, shmem: fix conflict of above-47bit hint address and PMD alignment
Kirill A. Shutemov [Tue, 14 Jan 2020 00:29:13 +0000 (16:29 -0800)]
mm/shmem.c: thp, shmem: fix conflict of above-47bit hint address and PMD alignment

commit 991589974d9c9ecb24ee3799ec8c415c730598a2 upstream.

Shmem/tmpfs tries to provide THP-friendly mappings if huge pages are
enabled.  But it doesn't work well with above-47bit hint address.

Normally, the kernel doesn't create userspace mappings above 47-bit,
even if the machine allows this (such as with 5-level paging on x86-64).
Not all user space is ready to handle wide addresses.  It's known that
at least some JIT compilers use higher bits in pointers to encode their
information.

Userspace can ask for allocation from full address space by specifying
hint address (with or without MAP_FIXED) above 47-bits.  If the
application doesn't need a particular address, but wants to allocate
from whole address space it can specify -1 as a hint address.

Unfortunately, this trick breaks THP alignment in shmem/tmp:
shmem_get_unmapped_area() would not try to allocate PMD-aligned area if
*any* hint address specified.

This can be fixed by requesting the aligned area if the we failed to
allocated at user-specified hint address.  The request with inflated
length will also take the user-specified hint address.  This way we will
not lose an allocation request from the full address space.

[kirill@shutemov.name: fold in a fixup]
Link: http://lkml.kernel.org/r/20191223231309.t6bh5hkbmokihpfu@box
Link: http://lkml.kernel.org/r/20191220142548.7118-3-kirill.shutemov@linux.intel.com
Fixes: b569bab78d8d ("x86/mm: Prepare to expose larger address space to userspace")
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: "Willhalm, Thomas" <thomas.willhalm@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: "Bruggeman, Otto G" <otto.g.bruggeman@intel.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoperf report: Fix incorrectly added dimensions as switch perf data file
Jin Yao [Fri, 20 Dec 2019 01:37:19 +0000 (09:37 +0800)]
perf report: Fix incorrectly added dimensions as switch perf data file

commit 0feba17bd7ee3b7e03d141f119049dcc23efa94e upstream.

We observed an issue that was some extra columns displayed after switching
perf data file in browser. The steps to reproduce:

1. perf record -a -e cycles,instructions -- sleep 3
2. perf report --group
3. In browser, we use hotkey 's' to switch to another perf.data
4. Now in browser, the extra columns 'Self' and 'Children' are displayed.

The issue is setup_sorting() executed again after repeat path, so dimensions
are added again.

This patch checks the last key returned from __cmd_report(). If it's
K_SWITCH_INPUT_DATA, skips the setup_sorting().

Fixes: ad0de0971b7f ("perf report: Enable the runtime switching of perf data file")
Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Feng Tang <feng.tang@intel.com>
Cc: Jin Yao <yao.jin@intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/20191220013722.20592-1-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agolocking/lockdep: Fix buffer overrun problem in stack_trace[]
Waiman Long [Fri, 20 Dec 2019 13:51:28 +0000 (08:51 -0500)]
locking/lockdep: Fix buffer overrun problem in stack_trace[]

commit d91f3057263ceb691ef527e71b41a56b17f6c869 upstream.

If the lockdep code is really running out of the stack_trace entries,
it is likely that buffer overrun can happen and the data immediately
after stack_trace[] will be corrupted.

If there is less than LOCK_TRACE_SIZE_IN_LONGS entries left before
the call to save_trace(), the max_entries computation will leave it
with a very large positive number because of its unsigned nature. The
subsequent call to stack_trace_save() will then corrupt the data after
stack_trace[]. Fix that by changing max_entries to a signed integer
and check for negative value before calling stack_trace_save().

Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 12593b7467f9 ("locking/lockdep: Reduce space occupied by stack traces")
Link: https://lkml.kernel.org/r/20191220135128.14876-1-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoperf hists: Fix variable name's inconsistency in hists__for_each() macro
Yuya Fujita [Thu, 19 Dec 2019 08:08:32 +0000 (08:08 +0000)]
perf hists: Fix variable name's inconsistency in hists__for_each() macro

commit 55347ec340af401437680fd0e88df6739a967f9f upstream.

Variable names are inconsistent in hists__for_each macro().

Due to this inconsistency, the macro replaces its second argument with
"fmt" regardless of its original name.

So far it works because only "fmt" is passed to the second argument.
However, this behavior is not expected and should be fixed.

Fixes: f0786af536bb ("perf hists: Introduce hists__for_each_format macro")
Fixes: aa6f50af822a ("perf hists: Introduce hists__for_each_sort_list macro")
Signed-off-by: Yuya Fujita <fujita.yuya@fujitsu.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/OSAPR01MB1588E1C47AC22043175DE1B2E8520@OSAPR01MB1588.jpnprd01.prod.outlook.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoclk: samsung: exynos5420: Keep top G3D clocks enabled
Marek Szyprowski [Mon, 16 Dec 2019 13:14:07 +0000 (14:14 +0100)]
clk: samsung: exynos5420: Keep top G3D clocks enabled

commit 67f96ff7c8f073648696eab50fd23ded23441067 upstream.

In Exynos542x/5800 SoCs, the G3D leaf clocks are located in the G3D power
domain. This is similar to the other hardware modules and their power
domains. However there is one thing specific to G3D clocks hierarchy.
Unlike other hardware modules, the G3D clocks hierarchy doesn't have any
gate clock between the TOP part of the hierarchy and the part located in
the power domain and some SoC internal busses are sourced directly from
the TOP muxes. The consequence of this design if the fact that the TOP
part of the hierarchy has to be enabled permanently to ensure proper
operation of the SoC power related components (G3D power domain and
Exynos Power Management Unit for system suspend/resume).

This patch adds an explicit call to clk_prepare_enable() on the last MUX
in the TOP part of G3D clock hierarchy to keep it enabled permanently to
ensure that the internal busses get their clock regardless of the main
G3D clock enablement status.

This fixes following imprecise abort issue observed on Odroid XU3/XU4
after enabling Panfrost driver by commit 1a5a85c56402 "ARM: dts: exynos:
Add Mali/GPU node on Exynos5420 and enable it on Odroid XU3/4"):

panfrost 11800000.gpu: clock rate = 400000000
panfrost 11800000.gpu: failed to get regulator: -517
panfrost 11800000.gpu: regulator init failed -517
Power domain G3D disable failed
...
panfrost 11800000.gpu: clock rate = 400000000
8<--- cut here ---
Unhandled fault: imprecise external abort (0x1406) at 0x00000000
pgd = (ptrval)
[00000000] *pgd=00000000
Internal error: : 1406 [#1] PREEMPT SMP ARM
Modules linked in:
CPU: 7 PID: 53 Comm: kworker/7:1 Not tainted 5.4.0-rc8-next-20191119-00032-g56f1001191a6 #6923
Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
Workqueue: events deferred_probe_work_func
PC is at panfrost_gpu_soft_reset+0x94/0x110
LR is at ___might_sleep+0x128/0x2dc
...
[<c05c231c>] (panfrost_gpu_soft_reset) from [<c05c2704>] (panfrost_gpu_init+0x10/0x67c)
[<c05c2704>] (panfrost_gpu_init) from [<c05c15d0>] (panfrost_device_init+0x158/0x2cc)
[<c05c15d0>] (panfrost_device_init) from [<c05c0cb0>] (panfrost_probe+0x80/0x178)
[<c05c0cb0>] (panfrost_probe) from [<c05cfaa0>] (platform_drv_probe+0x48/0x9c)
[<c05cfaa0>] (platform_drv_probe) from [<c05cd20c>] (really_probe+0x1c4/0x474)
[<c05cd20c>] (really_probe) from [<c05cd694>] (driver_probe_device+0x78/0x1bc)
[<c05cd694>] (driver_probe_device) from [<c05cb374>] (bus_for_each_drv+0x74/0xb8)
[<c05cb374>] (bus_for_each_drv) from [<c05ccfa8>] (__device_attach+0xd4/0x16c)
[<c05ccfa8>] (__device_attach) from [<c05cc110>] (bus_probe_device+0x88/0x90)
[<c05cc110>] (bus_probe_device) from [<c05cc634>] (deferred_probe_work_func+0x4c/0xd0)
[<c05cc634>] (deferred_probe_work_func) from [<c0149df0>] (process_one_work+0x300/0x864)
[<c0149df0>] (process_one_work) from [<c014a3ac>] (worker_thread+0x58/0x5a0)
[<c014a3ac>] (worker_thread) from [<c0151174>] (kthread+0x12c/0x160)
[<c0151174>] (kthread) from [<c01010b4>] (ret_from_fork+0x14/0x20)
Exception stack(0xee03dfb0 to 0xee03dff8)
...
Code: e594300c e5933020 e3130c01 1a00000f (ebefff50).
---[ end trace badde2b74a65a540 ]---

In the above case, the Panfrost driver disables G3D clocks after failure
of getting the needed regulator and return with -EPROVE_DEFER code. This
causes G3D power domain disable failure and then, during second probe
an imprecise abort is triggered due to undefined power domain state.

Fixes: 45f10dabb56b ("clk: samsung: exynos5420: Add SET_RATE_PARENT flag to clocks on G3D path")
Fixes: c9f7567aff31 ("clk: samsung: exynos542x: Move G3D subsystem clocks to its sub-CMU")
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lkml.kernel.org/r/20191216131407.17225-1-m.szyprowski@samsung.com
Acked-by: Krzysztof Kozlowski <krzk@kernel.org>
Acked-by: Chanwoo Choi <cw00.choi@samsung.com>
Acked-by: Sylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agos390/setup: Fix secure ipl message
Philipp Rudo [Wed, 18 Dec 2019 10:24:43 +0000 (11:24 +0100)]
s390/setup: Fix secure ipl message

commit 40260b01d029ba374637838213af500e03305326 upstream.

The new machine loader on z15 always creates an IPL Report block and
thus sets the IPL_PL_FLAG_IPLSR even when secure boot is disabled. This
causes the wrong message being printed at boot. Fix this by checking for
IPL_PL_FLAG_SIPL instead.

Fixes: 9641b8cc733f ("s390/ipl: read IPL report at early boot")
Signed-off-by: Philipp Rudo <prudo@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agoefi/earlycon: Fix write-combine mapping on x86
Arvind Sankar [Tue, 24 Dec 2019 13:29:07 +0000 (14:29 +0100)]
efi/earlycon: Fix write-combine mapping on x86

commit d92b54570d24d017d2630e314b525ed792f5aa6c upstream.

On x86, until PAT is initialized, WC translates into UC-. Since we
calculate and store pgprot_writecombine(PAGE_KERNEL) when earlycon is
initialized, this means we actually use UC- mappings instead of WC
mappings, which makes scrolling very slow.

Instead store a boolean flag to indicate whether we want to use
writeback or write-combine mappings, and recalculate the actual pgprot_t
we need on every mapping. Once PAT is initialized, we will start using
write-combine mappings, which speeds up the scrolling considerably.

Signed-off-by: Arvind Sankar <nivedita@alum.mit.edu>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-efi@vger.kernel.org
Fixes: 69c1f396f25b ("efi/x86: Convert x86 EFI earlyprintk into generic earlycon implementation")
Link: https://lkml.kernel.org/r/20191224132909.102540-2-ardb@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agox86/resctrl: Fix potential memory leak
Shakeel Butt [Thu, 2 Jan 2020 16:58:44 +0000 (08:58 -0800)]
x86/resctrl: Fix potential memory leak

commit ab6a2114433a3b5b555983dcb9b752a85255f04b upstream.

set_cache_qos_cfg() is leaking memory when the given level is not
RDT_RESOURCE_L3 or RDT_RESOURCE_L2. At the moment, this function is
called with only valid levels but move the allocation after the valid
level checks in order to make it more robust and future proof.

 [ bp: Massage commit message. ]

Fixes: 99adde9b370de ("x86/intel_rdt: Enable L2 CDP in MSR IA32_L2_QOS_CFG")
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Reinette Chatre <reinette.chatre@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20200102165844.133133-1-shakeelb@google.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
4 years agodrm/i915: Add missing include file <linux/math64.h>
YueHaibing [Tue, 7 Jan 2020 13:50:14 +0000 (21:50 +0800)]
drm/i915: Add missing include file <linux/math64.h>

commit ea38aa2ea5b0969776f0a47f174ce928a22be803 upstream.

Fix build error:
./drivers/gpu/drm/i915/selftests/i915_random.h: In function i915_prandom_u32_max_state:
./drivers/gpu/drm/i915/selftests/i915_random.h:48:23: error:
 implicit declaration of function mul_u32_u32; did you mean mul_u64_u32_div? [-Werror=implicit-function-declaration]
  return upper_32_bits(mul_u32_u32(prandom_u32_state(state), ep_ro));

Reported-by: Hulk Robot <hulkci@huawei.com>
Fixes: 7ce5b6850b47 ("drm/i915/selftests: Use mul_u32_u32() for 32b x 32b -> 64b result")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20200107135014.36472-1-yuehaibing@huawei.com
(cherry picked from commit 62bf5465b26d1f502430b9c654be7d16bf2e242d)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>