platform/kernel/linux-rpi.git
2 years agoRevert "kbuild: Disable gcc plugins"
Dom Cobley [Thu, 16 Dec 2021 12:11:54 +0000 (12:11 +0000)]
Revert "kbuild: Disable gcc plugins"

This reverts commit d62b4f2f5742bf39892acf8c6804fffee174d371.

2 years agoASoC:ma120x0p: Increase maximum sample rate to 192KHz
Joerg Schambacher [Wed, 15 Dec 2021 18:27:00 +0000 (19:27 +0100)]
ASoC:ma120x0p: Increase maximum sample rate to 192KHz

Change the maximum sample rate for the amplifier to
192KHz as given in the Infineon specification.

Signed-off-by: Joerg Schambacher <joerg@hifiberry.com>
2 years agodtoverlays: Add option to select camera as on CAM0 of CM
Dave Stevenson [Mon, 22 Nov 2021 12:44:29 +0000 (12:44 +0000)]
dtoverlays: Add option to select camera as on CAM0 of CM

Parameterise the overlays so that they can have an optional
cam0 parameter to switch to i2c_vc and csi0.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agoARM: dts: bcm2711-cm4s Correct i2c0mux to use 0/1 and 28/29 & 2 regulators
Dave Stevenson [Tue, 14 Dec 2021 14:54:15 +0000 (14:54 +0000)]
ARM: dts: bcm2711-cm4s Correct i2c0mux to use 0/1 and 28/29 & 2 regulators

CM4S follows CM1/3, so based on the documentation cameras/displays
connect to 0/1 and 28/29, not 0/1 and 44/45.

Likewise the camera regulator controls are independent as on CM1/3.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agomedia: i2c: ov7251: Make the enable GPIO optional.
Dave Stevenson [Tue, 14 Dec 2021 17:18:49 +0000 (17:18 +0000)]
media: i2c: ov7251: Make the enable GPIO optional.

Not all implementations wire up the enable GPIO and may just tie
it to a supply rail.
Make it optional.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodtoverlays: Convert ov5647 to use the regulator framework
Dave Stevenson [Mon, 22 Nov 2021 12:30:18 +0000 (12:30 +0000)]
dtoverlays: Convert ov5647 to use the regulator framework

Fixing up shutdown GPIOs via overrides is ugly, and doesn't work
on eg CM4 where both cameras share the same shutdown GPIO.

The driver is now updated to use the regulator framework, so switch
to using that instead.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agomedia: i2c: ov5647: Add support for regulator control.
Dave Stevenson [Mon, 22 Nov 2021 12:31:35 +0000 (12:31 +0000)]
media: i2c: ov5647: Add support for regulator control.

The driver supported using GPIOs to control the shutdown line,
but no regulator control.

Add regulator hooks.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodtoverlays: Convert the camera sensor overlays to use the new regs and clks.
Dave Stevenson [Sat, 20 Nov 2021 14:43:29 +0000 (14:43 +0000)]
dtoverlays: Convert the camera sensor overlays to use the new regs and clks.

Now that we have regulators and clocks defined in the base DT for
image sensors, switch the overlays to use them.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodt: Create static regulators and clocks for camera nodes
Dave Stevenson [Sat, 20 Nov 2021 10:48:36 +0000 (10:48 +0000)]
dt: Create static regulators and clocks for camera nodes

Unloading regulators through dynamic device tree doesn't work
as the regulators will unregister whilst clients are still
registered. Whilst the regulator framework does WARN when that
happens, the client putting the regulator then typically results
in a NULL dereference and badness.

Instead of creating regulators and clocks from the overlays,
create regulators and clocks for the sensors in the base DT.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodrm/vc4: Fix build without DRM_VC4_HDMI_CEC
Phil Elwell [Tue, 14 Dec 2021 21:53:18 +0000 (21:53 +0000)]
drm/vc4: Fix build without DRM_VC4_HDMI_CEC

As reported by @asavah.

Fixes: https://github.com/raspberrypi/linux/issues/4771

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
2 years agousb: xhci: add VLI_TRB_CACHE_BUG quirk
Jonathan Bell [Mon, 13 Dec 2021 16:04:03 +0000 (16:04 +0000)]
usb: xhci: add VLI_TRB_CACHE_BUG quirk

The VL805 fetches up to 4 transfer TRBs at a time. TRB reads don't cross
a 64B boundary, and if a TRB is fetched and is not on a 64B boundary,
the read is sized up to the next 64B boundary.

However the VL805 implements a readahead prefetch for TRBs on a transfer
ring. This fetches the next 64B after any TRB read has happened. Near
the end of a ring segment, the prefetcher can read the first 64B of the
next page in physical memory and this is where the behaviour causes a
bug.

The controller does not tag reads with which endpoint they are for, so
if the start of the next page is a ring segment used by a victim
endpoint, and the victim endpoint is about to fetch TRBs from the start
of the segment, the victim endpoint will read from the prefetched data
and not perform a read to main memory. If the data is stale, the ring
cycle state bit may not be correct and the endpoint will silently halt.

Adjust trbs_per_seg for transfer rings allocated for this controller.

See https://github.com/raspberrypi/linux/issues/4685

Signed-off-by: Jonathan Bell <jonathan@raspberrypi.com>
2 years agoxhci: refactor out TRBS_PER_SEGMENT define in runtime code
Jonathan Bell [Mon, 13 Dec 2021 15:05:56 +0000 (15:05 +0000)]
xhci: refactor out TRBS_PER_SEGMENT define in runtime code

In anticipation of adjusting the number of utilised TRBs in a ring
segment, add trbs_per_seg to struct xhci_ring and use this instead
of a compile-time define.

Signed-off-by: Jonathan Bell <jonathan@raspberrypi.com>
2 years agoRevert "xhci: add a quirk to work around a suspected cache bug on VLI controllers"
Jonathan Bell [Fri, 10 Dec 2021 16:29:49 +0000 (16:29 +0000)]
Revert "xhci: add a quirk to work around a suspected cache bug on VLI controllers"

This reverts commit a1d0f808d4a0b7f7053095cd4ab97a4276bed9ff.

2 years agoARM: dts: Create bcm2711-rpi-cm4s.dts (#4761)
peterharperuk [Mon, 13 Dec 2021 14:00:35 +0000 (14:00 +0000)]
ARM: dts: Create bcm2711-rpi-cm4s.dts (#4761)

Signed-off-by: Peter Harper <peter.harper@raspberrypi.com>
2 years agovc4/drm: Ignore vc4_hdmi->output_enabled for allowing audio (#4759)
popcornmix [Wed, 8 Dec 2021 15:57:15 +0000 (15:57 +0000)]
vc4/drm: Ignore vc4_hdmi->output_enabled for allowing audio (#4759)

Otherwise we reject audio playback when switching hdmi modes

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
2 years agoMerge remote-tracking branch 'stable/linux-5.10.y' into rpi-5.10.y
Dom Cobley [Tue, 7 Dec 2021 13:42:57 +0000 (13:42 +0000)]
Merge remote-tracking branch 'stable/linux-5.10.y' into rpi-5.10.y

2 years agoxhci: add a quirk to work around a suspected cache bug on VLI controllers
Jonathan Bell [Fri, 3 Dec 2021 14:33:51 +0000 (14:33 +0000)]
xhci: add a quirk to work around a suspected cache bug on VLI controllers

Certain transfer ring access patterns can cause the controller to hang
fetching TRBs for a USB2.0 endpoint.

- If two USB2.0 endpoints are active at once and
- Both endpoints are traversing a Link TRB where the following segment
  has a lower page address and
- One of the endpoints is a Bulk IN and
- The other endpoint is an Interrupt IN

Then the Interrupt IN endpoint can end up not getting polled.
It is unclear what the precise failure mode is, as the controller seems to
haphazardly and repeatedly fetch TRBs for both endpoints but does not
advance the Interrupt endpoint transfer.

As a workaround, add a quirk that initially constrains all USB2.0 transfer
rings to a single segment in size. If for any reason a device driver queues
up enough outstanding transfers to fill the ring segment, then the ring
will be expanded. This has not been seen to occur with UMS or UVC drivers,
which aggressively queue buffers.

Signed-off-by: Jonathan Bell <jonathan@raspberrypi.com>
2 years agoxhci: correct room_on_ring() for cases where there is a single segment
Jonathan Bell [Fri, 3 Dec 2021 14:32:05 +0000 (14:32 +0000)]
xhci: correct room_on_ring() for cases where there is a single segment

Don't calculate space based on the number of TRBs in the current segment,
as it's OK to wrap to the start (and flip the cycle state bit).

Signed-off-by: Jonathan Bell <jonathan@raspberrypi.com>
2 years agodtoverlays: Add a generic DPI panel overlay for KMS
Dave Stevenson [Thu, 2 Dec 2021 18:06:37 +0000 (18:06 +0000)]
dtoverlays: Add a generic DPI panel overlay for KMS

Uses the "panel-dpi" compatible to set panel timings from
DT.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodt: bcm270x: Add GPIO defines for RGB565 DPI output modes
Dave Stevenson [Thu, 2 Dec 2021 18:24:44 +0000 (18:24 +0000)]
dt: bcm270x: Add GPIO defines for RGB565 DPI output modes

Adds the pinctrl defines for the RGB565 DPI output modes.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodrm/vc4: dpi: Ensure a default format is selected
Dave Stevenson [Thu, 2 Dec 2021 18:28:29 +0000 (18:28 +0000)]
drm/vc4: dpi: Ensure a default format is selected

In a couple of error/incomplete configuration cases, the
DPI_FORMAT bits wouldn't get set.

Adopt a default of RGB666 in all these cases.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodrm/vc4: dpi: Add option for inverting pixel clock and output enable
Dave Stevenson [Thu, 2 Dec 2021 18:21:46 +0000 (18:21 +0000)]
drm/vc4: dpi: Add option for inverting pixel clock and output enable

DRM provides flags for inverting pixel clock and output enable
signals, but these were not mapped to the relevant registers.

Add those mappings.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodrm/panel-simple: Allow the bus format to be read from DT for panel-dpi
Dave Stevenson [Thu, 2 Dec 2021 18:16:21 +0000 (18:16 +0000)]
drm/panel-simple: Allow the bus format to be read from DT for panel-dpi

The "panel-dpi" compatible string configures panel from device tree,
but it doesn't provide any way of configuring the bus format (colour
representation), nor does it populate it.

Add a DT parameter "bus-format" that allows the MEDIA_BUS_FMT_xxx value
to be specified from device tree.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodrm/panel-simple: When using panel-dpi, update desc
Dave Stevenson [Thu, 2 Dec 2021 18:13:02 +0000 (18:13 +0000)]
drm/panel-simple: When using panel-dpi, update desc

When using the "panel-dpi" compatible string, desc as passed
from DT is a dumy entry, and panel_dpi_probe allocates a new
one and attaches it to the panel.
However panel_simple_probe has already taken a local copy of
the variable, which means all the validation is done against
the empty dummy structure, not the configured data.

Update the local variable after panel_dpi_probe.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodrm/panel-simple: Populate bpc when using panel-dpi
Dave Stevenson [Thu, 2 Dec 2021 18:10:55 +0000 (18:10 +0000)]
drm/panel-simple: Populate bpc when using panel-dpi

panel-dpi doesn't know the bit depth, so in the same way that
DPI is guessed for the connector type, guess that it'll be 8bpc.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agoRevert "Revert "overlays: vc4-kms-v3d: Change composite handling""
Dave Stevenson [Fri, 26 Nov 2021 17:59:07 +0000 (17:59 +0000)]
Revert "Revert "overlays: vc4-kms-v3d: Change composite handling""

Reinstates the new handling.

This reverts commit 46c99e3d7cf38446491b3c2a826fc05dcebc588d.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodt: bcm283x: Change BCM283x HDMI to use firmware clock driver
Dave Stevenson [Thu, 2 Dec 2021 14:11:09 +0000 (14:11 +0000)]
dt: bcm283x: Change BCM283x HDMI to use firmware clock driver

The clk-bcm2835 handling of the pixel clock does not function
correctly when the HDMI power domain is disabled.

The firmware supports it correctly, and the firmware clock
driver now supports it, so switch the vc4-hdmi driver to use
the firmware clock driver.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agoclk: bcm: rpi: Add the BCM283x pixel clock.
Dave Stevenson [Thu, 2 Dec 2021 13:53:36 +0000 (13:53 +0000)]
clk: bcm: rpi: Add the BCM283x pixel clock.

The clk-bcm2835 handling of the pixel clock does not function
correctly when the HDMI power domain is disabled.

The firmware supports it correctly, so add it to the
firmware clock driver.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodrm/vc4: Move HDMI reset to pm_resume
Dave Stevenson [Thu, 25 Nov 2021 14:46:55 +0000 (14:46 +0000)]
drm/vc4: Move HDMI reset to pm_resume

Pi0-3 have power domains attached to the pm_runtime hooks
for the HDMI block. Initialisation done in the reset called
from bind is therefore lost if all users of the domain are
suspended.
The VEC shares the same lowest level clock/power gating as
the HDMI block, so whilst that is enabled the block is never
actually powered down, but if it isn't enabled then we lose
the state.

Reset and initialise the HDMI block from pm_resume.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agodrm/vc4: Correct logging string for dev_err_probe
Dave Stevenson [Tue, 30 Nov 2021 17:35:06 +0000 (17:35 +0000)]
drm/vc4: Correct logging string for dev_err_probe

The commit that changed from dev_err to dev_err_probe left the %d in the
format string, but removed the parameter, leading to a compile
warning.

Fixes: "6505412df625 drm/vc4: Use dev_err_probe when logging error registering HDMI audio"
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agoinput: edt-ft5x06: Only look at the number of points reported
Dave Stevenson [Tue, 30 Nov 2021 17:28:50 +0000 (17:28 +0000)]
input: edt-ft5x06: Only look at the number of points reported

Register 0x02 in the FT5x06 is TD_STATUS containing the number
of valid touch points being reported.

Iterate over that number of points rather than all that are
supported on the device.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agoRevert "dtoverlays: Drop i2c baudrate for Pi 7inch DSI screen to 50kHz"
Dave Stevenson [Tue, 30 Nov 2021 17:39:17 +0000 (17:39 +0000)]
Revert "dtoverlays: Drop i2c baudrate for Pi 7inch DSI screen to 50kHz"

This reverts commit 0c41710df564f76275c2868beaa0c316553e8247.

The regulator driver is now hopefully fixed, therefore revert the
workaround that dropped the I2C frequency.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agoregulator/rpi-panel-attiny: Use two transactions for I2C read
Dave Stevenson [Thu, 25 Nov 2021 14:50:10 +0000 (14:50 +0000)]
regulator/rpi-panel-attiny: Use two transactions for I2C read

The I2C to the Atmel is very fussy, and locks up easily on
Pi0-3 particularly on reads.
If running at 100kHz on Pi3, reading the ID register generally
locks up the Atmel, but splitting the register select write and
read into two transactions is reliable.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agoregulator/rpi-panel-attiny: Don't read the LCD power status
Dave Stevenson [Mon, 29 Nov 2021 18:31:37 +0000 (18:31 +0000)]
regulator/rpi-panel-attiny: Don't read the LCD power status

The I2C to the Atmel is very fussy, and locks up easily on
Pi0-3 particularly on reads.

The LCD power status is controlled solely by this driver, so
rather than reading it back from the Atmel, use the cached
status last set.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
2 years agoLinux 5.10.83
Greg Kroah-Hartman [Wed, 1 Dec 2021 08:19:10 +0000 (09:19 +0100)]
Linux 5.10.83

Link: https://lore.kernel.org/r/20211129181711.642046348@linuxfoundation.org
Tested-by: Shuah Khan <skhan@linuxfoundation.org>
Tested-by: Hulk Robot <hulkrobot@huawei.com>
Tested-by: Florian Fainelli <f.fainelli@gmail.com>
Tested-by: Linux Kernel Functional Testing <lkft@linaro.org>
Tested-by: Jon Hunter <jonathanh@nvidia.com>
Tested-by: Fox Chen <foxhlchen@gmail.com>
Tested-by: Pavel Machek (CIP) <pavel@denx.de>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agodrm/amdgpu/gfx9: switch to golden tsc registers for renoir+
Alex Deucher [Thu, 18 Nov 2021 19:50:37 +0000 (14:50 -0500)]
drm/amdgpu/gfx9: switch to golden tsc registers for renoir+

commit 53af98c091bc42fd9ec64cfabc40da4e5f3aae93 upstream.

Renoir and newer gfx9 APUs have new TSC register that is
not part of the gfxoff tile, so it can be read without
needing to disable gfx off.

Acked-by: Luben Tuikov <luben.tuikov@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agonet: stmmac: platform: fix build warning when with !CONFIG_PM_SLEEP
Joakim Zhang [Thu, 9 Sep 2021 09:23:22 +0000 (17:23 +0800)]
net: stmmac: platform: fix build warning when with !CONFIG_PM_SLEEP

commit 2a48d96fd58a666ae231c3dd6fe4a458798ac645 upstream.

Use __maybe_unused for noirq_suspend()/noirq_resume() hooks to avoid
build warning with !CONFIG_PM_SLEEP:

>> drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c:796:12: error: 'stmmac_pltfr_noirq_resume' defined but not used [-Werror=unused-function]
     796 | static int stmmac_pltfr_noirq_resume(struct device *dev)
         |            ^~~~~~~~~~~~~~~~~~~~~~~~~
>> drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c:775:12: error: 'stmmac_pltfr_noirq_suspend' defined but not used [-Werror=unused-function]
     775 | static int stmmac_pltfr_noirq_suspend(struct device *dev)
         |            ^~~~~~~~~~~~~~~~~~~~~~~~~~
   cc1: all warnings being treated as errors

Fixes: 276aae377206 ("net: stmmac: fix system hang caused by eee_ctrl_timer during suspend/resume")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agoshm: extend forced shm destroy to support objects from several IPC nses
Alexander Mikhalitsyn [Sat, 20 Nov 2021 00:43:21 +0000 (16:43 -0800)]
shm: extend forced shm destroy to support objects from several IPC nses

commit 85b6d24646e4125c591639841169baa98a2da503 upstream.

Currently, the exit_shm() function not designed to work properly when
task->sysvshm.shm_clist holds shm objects from different IPC namespaces.

This is a real pain when sysctl kernel.shm_rmid_forced = 1, because it
leads to use-after-free (reproducer exists).

This is an attempt to fix the problem by extending exit_shm mechanism to
handle shm's destroy from several IPC ns'es.

To achieve that we do several things:

1. add a namespace (non-refcounted) pointer to the struct shmid_kernel

2. during new shm object creation (newseg()/shmget syscall) we
   initialize this pointer by current task IPC ns

3. exit_shm() fully reworked such that it traverses over all shp's in
   task->sysvshm.shm_clist and gets IPC namespace not from current task
   as it was before but from shp's object itself, then call
   shm_destroy(shp, ns).

Note: We need to be really careful here, because as it was said before
(1), our pointer to IPC ns non-refcnt'ed.  To be on the safe side we
using special helper get_ipc_ns_not_zero() which allows to get IPC ns
refcounter only if IPC ns not in the "state of destruction".

Q/A

Q: Why can we access shp->ns memory using non-refcounted pointer?
A: Because shp object lifetime is always shorther than IPC namespace
   lifetime, so, if we get shp object from the task->sysvshm.shm_clist
   while holding task_lock(task) nobody can steal our namespace.

Q: Does this patch change semantics of unshare/setns/clone syscalls?
A: No. It's just fixes non-covered case when process may leave IPC
   namespace without getting task->sysvshm.shm_clist list cleaned up.

Link: https://lkml.kernel.org/r/67bb03e5-f79c-1815-e2bf-949c67047418@colorfullife.com
Link: https://lkml.kernel.org/r/20211109151501.4921-1-manfred@colorfullife.com
Fixes: ab602f79915 ("shm: make exit_shm work proportional to task activity")
Co-developed-by: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Alexander Mikhalitsyn <alexander.mikhalitsyn@virtuozzo.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Greg KH <gregkh@linuxfoundation.org>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Cc: Vasily Averin <vvs@virtuozzo.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>
2 years agos390/mm: validate VMA in PGSTE manipulation functions
David Hildenbrand [Thu, 9 Sep 2021 16:22:42 +0000 (18:22 +0200)]
s390/mm: validate VMA in PGSTE manipulation functions

commit fe3d10024073f06f04c74b9674bd71ccc1d787cf upstream.

We should not walk/touch page tables outside of VMA boundaries when
holding only the mmap sem in read mode. Evil user space can modify the
VMA layout just before this function runs and e.g., trigger races with
page table removal code since commit dd2283f2605e ("mm: mmap: zap pages
with read mmap_sem in munmap"). gfn_to_hva() will only translate using
KVM memory regions, but won't validate the VMA.

Further, we should not allocate page tables outside of VMA boundaries: if
evil user space decides to map hugetlbfs to these ranges, bad things will
happen because we suddenly have PTE or PMD page tables where we
shouldn't have them.

Similarly, we have to check if we suddenly find a hugetlbfs VMA, before
calling get_locked_pte().

Fixes: 2d42f9477320 ("s390/kvm: Add PGSTE manipulation functions")
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Link: https://lore.kernel.org/r/20210909162248.14969-4-david@redhat.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agotty: hvc: replace BUG_ON() with negative return value
Juergen Gross [Wed, 7 Jul 2021 09:10:45 +0000 (11:10 +0200)]
tty: hvc: replace BUG_ON() with negative return value

commit e679004dec37566f658a255157d3aed9d762a2b7 upstream.

Xen frontends shouldn't BUG() in case of illegal data received from
their backends. So replace the BUG_ON()s when reading illegal data from
the ring page with negative return values.

Reviewed-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
Link: https://lore.kernel.org/r/20210707091045.460-1-jgross@suse.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agoxen/netfront: don't trust the backend response data blindly
Juergen Gross [Tue, 24 Aug 2021 10:28:09 +0000 (12:28 +0200)]
xen/netfront: don't trust the backend response data blindly

commit a884daa61a7d91650987e855464526aef219590f upstream.

Today netfront will trust the backend to send only sane response data.
In order to avoid privilege escalations or crashes in case of malicious
backends verify the data to be within expected limits. Especially make
sure that the response always references an outstanding request.

Note that only the tx queue needs special id handling, as for the rx
queue the id is equal to the index in the ring page.

Introduce a new indicator for the device whether it is broken and let
the device stop working when it is set. Set this indicator in case the
backend sets any weird data.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agoxen/netfront: disentangle tx_skb_freelist
Juergen Gross [Tue, 24 Aug 2021 10:28:08 +0000 (12:28 +0200)]
xen/netfront: disentangle tx_skb_freelist

commit 21631d2d741a64a073e167c27769e73bc7844a2f upstream.

The tx_skb_freelist elements are in a single linked list with the
request id used as link reference. The per element link field is in a
union with the skb pointer of an in use request.

Move the link reference out of the union in order to enable a later
reuse of it for requests which need a populated skb pointer.

Rename add_id_to_freelist() and get_id_from_freelist() to
add_id_to_list() and get_id_from_list() in order to prepare using
those for other lists as well. Define ~0 as value to indicate the end
of a list and place that value into the link for a request not being
on the list.

When freeing a skb zero the skb pointer in the request. Use a NULL
value of the skb pointer instead of skb_entry_is_link() for deciding
whether a request has a skb linked to it.

Remove skb_entry_set_link() and open code it instead as it is really
trivial now.

Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agoxen/netfront: don't read data from request on the ring page
Juergen Gross [Tue, 24 Aug 2021 10:28:07 +0000 (12:28 +0200)]
xen/netfront: don't read data from request on the ring page

commit 162081ec33c2686afa29d91bf8d302824aa846c7 upstream.

In order to avoid a malicious backend being able to influence the local
processing of a request build the request locally first and then copy
it to the ring page. Any reading from the request influencing the
processing in the frontend needs to be done on the local instance.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agoxen/netfront: read response from backend only once
Juergen Gross [Tue, 24 Aug 2021 10:28:06 +0000 (12:28 +0200)]
xen/netfront: read response from backend only once

commit 8446066bf8c1f9f7b7412c43fbea0fb87464d75b upstream.

In order to avoid problems in case the backend is modifying a response
on the ring page while the frontend has already seen it, just read the
response into a local buffer in one go and then operate on that buffer
only.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agoxen/blkfront: don't trust the backend response data blindly
Juergen Gross [Fri, 30 Jul 2021 10:38:54 +0000 (12:38 +0200)]
xen/blkfront: don't trust the backend response data blindly

commit b94e4b147fd1992ad450e1fea1fdaa3738753373 upstream.

Today blkfront will trust the backend to send only sane response data.
In order to avoid privilege escalations or crashes in case of malicious
backends verify the data to be within expected limits. Especially make
sure that the response always references an outstanding request.

Introduce a new state of the ring BLKIF_STATE_ERROR which will be
switched to in case an inconsistency is being detected. Recovering from
this state is possible only via removing and adding the virtual device
again (e.g. via a suspend/resume cycle).

Make all warning messages issued due to valid error responses rate
limited in order to avoid message floods being triggered by a malicious
backend.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Link: https://lore.kernel.org/r/20210730103854.12681-4-jgross@suse.com
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agoxen/blkfront: don't take local copy of a request from the ring page
Juergen Gross [Fri, 30 Jul 2021 10:38:53 +0000 (12:38 +0200)]
xen/blkfront: don't take local copy of a request from the ring page

commit 8f5a695d99000fc3aa73934d7ced33cfc64dcdab upstream.

In order to avoid a malicious backend being able to influence the local
copy of a request build the request locally first and then copy it to
the ring page instead of doing it the other way round as today.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Link: https://lore.kernel.org/r/20210730103854.12681-3-jgross@suse.com
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agoxen/blkfront: read response from backend only once
Juergen Gross [Fri, 30 Jul 2021 10:38:52 +0000 (12:38 +0200)]
xen/blkfront: read response from backend only once

commit 71b66243f9898d0e54296b4e7035fb33cdcb0707 upstream.

In order to avoid problems in case the backend is modifying a response
on the ring page while the frontend has already seen it, just read the
response into a local buffer in one go and then operate on that buffer
only.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Link: https://lore.kernel.org/r/20210730103854.12681-2-jgross@suse.com
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agoxen: sync include/xen/interface/io/ring.h with Xen's newest version
Juergen Gross [Wed, 12 May 2021 06:22:10 +0000 (08:22 +0200)]
xen: sync include/xen/interface/io/ring.h with Xen's newest version

commit 629a5d87e26fe96bcaab44cbb81f5866af6f7008 upstream.

Sync include/xen/interface/io/ring.h with Xen's newest version in
order to get the RING_COPY_RESPONSE() and RING_RESPONSE_PROD_OVERFLOW()
macros.

Note that this will correct the wrong license info by adding the
missing original copyright notice.

Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agotracing: Check pid filtering when creating events
Steven Rostedt (VMware) [Fri, 26 Nov 2021 18:35:26 +0000 (13:35 -0500)]
tracing: Check pid filtering when creating events

commit 6cb206508b621a9a0a2c35b60540e399225c8243 upstream.

When pid filtering is activated in an instance, all of the events trace
files for that instance has the PID_FILTER flag set. This determines
whether or not pid filtering needs to be done on the event, otherwise the
event is executed as normal.

If pid filtering is enabled when an event is created (via a dynamic event
or modules), its flag is not updated to reflect the current state, and the
events are not filtered properly.

Cc: stable@vger.kernel.org
Fixes: 3fdaf80f4a836 ("tracing: Implement event pid filtering")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agovhost/vsock: fix incorrect used length reported to the guest
Stefano Garzarella [Mon, 22 Nov 2021 16:35:24 +0000 (17:35 +0100)]
vhost/vsock: fix incorrect used length reported to the guest

commit 49d8c5ffad07ca014cfae72a1b9b8c52b6ad9cb8 upstream.

The "used length" reported by calling vhost_add_used() must be the
number of bytes written by the device (using "in" buffers).

In vhost_vsock_handle_tx_kick() the device only reads the guest
buffers (they are all "out" buffers), without writing anything,
so we must pass 0 as "used length" to comply virtio spec.

Fixes: 433fc58e6bf2 ("VSOCK: Introduce vhost_vsock.ko")
Cc: stable@vger.kernel.org
Reported-by: Halil Pasic <pasic@linux.ibm.com>
Suggested-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Link: https://lore.kernel.org/r/20211122163525.294024-2-sgarzare@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Halil Pasic <pasic@linux.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agoiommu/amd: Clarify AMD IOMMUv2 initialization messages
Joerg Roedel [Tue, 23 Nov 2021 10:55:07 +0000 (11:55 +0100)]
iommu/amd: Clarify AMD IOMMUv2 initialization messages

commit 717e88aad37befedfd531378b632e794e24e9afb upstream.

The messages printed on the initialization of the AMD IOMMUv2 driver
have caused some confusion in the past. Clarify the messages to lower
the confusion in the future.

Cc: stable@vger.kernel.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Link: https://lore.kernel.org/r/20211123105507.7654-3-joro@8bytes.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2 years agosmb3: do not error on fsync when readonly
Steve French [Wed, 10 Nov 2021 07:47:48 +0000 (01:47 -0600)]
smb3: do not error on fsync when readonly

[ Upstream commit 71e6864eacbef0b2645ca043cdfbac272cb6cea3 ]

Linux allows doing a flush/fsync on a file open for read-only,
but the protocol does not allow that.  If the file passed in
on the flush is read-only try to find a writeable handle for
the same inode, if that is not possible skip sending the
fsync call to the server to avoid breaking the apps.

Reported-by: Julian Sikorski <belegdol@gmail.com>
Tested-by: Julian Sikorski <belegdol@gmail.com>
Suggested-by: Jeremy Allison <jra@samba.org>
Reviewed-by: Paulo Alcantara (SUSE) <pc@cjr.nz>
Signed-off-by: Steve French <stfrench@microsoft.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoceph: properly handle statfs on multifs setups
Jeff Layton [Tue, 5 Oct 2021 15:12:58 +0000 (11:12 -0400)]
ceph: properly handle statfs on multifs setups

[ Upstream commit 8cfc0c7ed34f7929ce7e5d7c6eecf4d01ba89a84 ]

ceph_statfs currently stuffs the cluster fsid into the f_fsid field.
This was fine when we only had a single filesystem per cluster, but now
that we have multiples we need to use something that will vary between
them.

Change ceph_statfs to xor each 32-bit chunk of the fsid (aka cluster id)
into the lower bits of the statfs->f_fsid. Change the lower bits to hold
the fscid (filesystem ID within the cluster).

That should give us a value that is guaranteed to be unique between
filesystems within a cluster, and should minimize the chance of
collisions between mounts of different clusters.

URL: https://tracker.ceph.com/issues/52812
Reported-by: Sachin Prabhu <sprabhu@redhat.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agof2fs: set SBI_NEED_FSCK flag when inconsistent node block found
Weichao Guo [Sat, 18 Sep 2021 12:46:36 +0000 (20:46 +0800)]
f2fs: set SBI_NEED_FSCK flag when inconsistent node block found

[ Upstream commit 6663b138ded1a59e630c9e605e42aa7fde490cdc ]

Inconsistent node block will cause a file fail to open or read,
which could make the user process crashes or stucks. Let's mark
SBI_NEED_FSCK flag to trigger a fix at next fsck time. After
unlinking the corrupted file, the user process could regenerate
a new one and work correctly.

Signed-off-by: Weichao Guo <guoweichao@oppo.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agosched/scs: Reset task stack state in bringup_cpu()
Mark Rutland [Tue, 23 Nov 2021 11:40:47 +0000 (11:40 +0000)]
sched/scs: Reset task stack state in bringup_cpu()

[ Upstream commit dce1ca0525bfdc8a69a9343bc714fbc19a2f04b3 ]

To hot unplug a CPU, the idle task on that CPU calls a few layers of C
code before finally leaving the kernel. When KASAN is in use, poisoned
shadow is left around for each of the active stack frames, and when
shadow call stacks are in use. When shadow call stacks (SCS) are in use
the task's saved SCS SP is left pointing at an arbitrary point within
the task's shadow call stack.

When a CPU is offlined than onlined back into the kernel, this stale
state can adversely affect execution. Stale KASAN shadow can alias new
stackframes and result in bogus KASAN warnings. A stale SCS SP is
effectively a memory leak, and prevents a portion of the shadow call
stack being used. Across a number of hotplug cycles the idle task's
entire shadow call stack can become unusable.

We previously fixed the KASAN issue in commit:

  e1b77c92981a5222 ("sched/kasan: remove stale KASAN poison after hotplug")

... by removing any stale KASAN stack poison immediately prior to
onlining a CPU.

Subsequently in commit:

  f1a0a376ca0c4ef1 ("sched/core: Initialize the idle task with preemption disabled")

... the refactoring left the KASAN and SCS cleanup in one-time idle
thread initialization code rather than something invoked prior to each
CPU being onlined, breaking both as above.

We fixed SCS (but not KASAN) in commit:

  63acd42c0d4942f7 ("sched/scs: Reset the shadow stack when idle_task_exit")

... but as this runs in the context of the idle task being offlined it's
potentially fragile.

To fix these consistently and more robustly, reset the SCS SP and KASAN
shadow of a CPU's idle task immediately before we online that CPU in
bringup_cpu(). This ensures the idle task always has a consistent state
when it is running, and removes the need to so so when exiting an idle
task.

Whenever any thread is created, dup_task_struct() will give the task a
stack which is free of KASAN shadow, and initialize the task's SCS SP,
so there's no need to specially initialize either for idle thread within
init_idle(), as this was only necessary to handle hotplug cycles.

I've tested this on arm64 with:

* gcc 11.1.0, defconfig +KASAN_INLINE, KASAN_STACK
* clang 12.0.0, defconfig +KASAN_INLINE, KASAN_STACK, SHADOW_CALL_STACK

... offlining and onlining CPUS with:

| while true; do
|   for C in /sys/devices/system/cpu/cpu*/online; do
|     echo 0 > $C;
|     echo 1 > $C;
|   done
| done

Fixes: f1a0a376ca0c4ef1 ("sched/core: Initialize the idle task with preemption disabled")
Reported-by: Qian Cai <quic_qiancai@quicinc.com>
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Qian Cai <quic_qiancai@quicinc.com>
Link: https://lore.kernel.org/lkml/20211115113310.35693-1-mark.rutland@arm.com/
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agotcp: correctly handle increased zerocopy args struct size
Arjun Roy [Thu, 10 Dec 2020 19:16:03 +0000 (11:16 -0800)]
tcp: correctly handle increased zerocopy args struct size

[ Upstream commit e0fecb289ad3fd2245cdc50bf450b97fcca39884 ]

A prior patch increased the size of struct tcp_zerocopy_receive
but did not update do_tcp_getsockopt() handling to properly account
for this.

This patch simply reintroduces content erroneously cut from the
referenced prior patch that handles the new struct size.

Fixes: 18fb76ed5386 ("net-zerocopy: Copy straggler unaligned data for TCP Rx. zerocopy.")
Signed-off-by: Arjun Roy <arjunroy@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: mscc: ocelot: correctly report the timestamping RX filters in ethtool
Vladimir Oltean [Fri, 26 Nov 2021 17:28:45 +0000 (19:28 +0200)]
net: mscc: ocelot: correctly report the timestamping RX filters in ethtool

[ Upstream commit c49a35eedfef08bffd46b53c25dbf9d6016a86ff ]

The driver doesn't support RX timestamping for non-PTP packets, but it
declares that it does. Restrict the reported RX filters to PTP v2 over
L2 and over L4.

Fixes: 4e3b0468e6d7 ("net: mscc: PTP Hardware Clock (PHC) support")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: mscc: ocelot: don't downgrade timestamping RX filters in SIOCSHWTSTAMP
Vladimir Oltean [Fri, 26 Nov 2021 17:28:41 +0000 (19:28 +0200)]
net: mscc: ocelot: don't downgrade timestamping RX filters in SIOCSHWTSTAMP

[ Upstream commit 8a075464d1e9317ffae0973dfe538a7511291a06 ]

The ocelot driver, when asked to timestamp all receiving packets, 1588
v1 or NTP, says "nah, here's 1588 v2 for you".

According to this discussion:
https://patchwork.kernel.org/project/netdevbpf/patch/20211104133204.19757-8-martin.kaistra@linutronix.de/#24577647
drivers that downgrade from a wider request to a narrower response (or
even a response where the intersection with the request is empty) are
buggy, and should return -ERANGE instead. This patch fixes that.

Fixes: 4e3b0468e6d7 ("net: mscc: PTP Hardware Clock (PHC) support")
Suggested-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: hns3: fix VF RSS failed problem after PF enable multi-TCs
Guangbin Huang [Fri, 26 Nov 2021 12:03:15 +0000 (20:03 +0800)]
net: hns3: fix VF RSS failed problem after PF enable multi-TCs

[ Upstream commit 8d2ad993aa05c0768f00c886c9d369cd97a337ac ]

When PF is set to multi-TCs and configured mapping relationship between
priorities and TCs, the hardware will active these settings for this PF
and its VFs.

In this case when VF just uses one TC and its rx packets contain priority,
and if the priority is not mapped to TC0, as other TCs of VF is not valid,
hardware always put this kind of packets to the queue 0. It cause this kind
of packets of VF can not be used RSS function.

To fix this problem, set tc mode of all unused TCs of VF to the setting of
TC0, then rx packet with priority which map to unused TC will be direct to
TC0.

Fixes: e2cb1dec9779 ("net: hns3: Add HNS3 VF HCL(Hardware Compatibility Layer) Support")
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet/smc: Don't call clcsock shutdown twice when smc shutdown
Tony Lu [Fri, 26 Nov 2021 02:41:35 +0000 (10:41 +0800)]
net/smc: Don't call clcsock shutdown twice when smc shutdown

[ Upstream commit bacb6c1e47691cda4a95056c21b5487fb7199fcc ]

When applications call shutdown() with SHUT_RDWR in userspace,
smc_close_active() calls kernel_sock_shutdown(), and it is called
twice in smc_shutdown().

This fixes this by checking sk_state before do clcsock shutdown, and
avoids missing the application's call of smc_shutdown().

Link: https://lore.kernel.org/linux-s390/1f67548e-cbf6-0dce-82b5-10288a4583bd@linux.ibm.com/
Fixes: 606a63c9783a ("net/smc: Ensure the active closing peer first closes clcsock")
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Reviewed-by: Wen Gu <guwen@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Link: https://lore.kernel.org/r/20211126024134.45693-1-tonylu@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: vlan: fix underflow for the real_dev refcnt
Ziyang Xuan [Fri, 26 Nov 2021 01:59:42 +0000 (09:59 +0800)]
net: vlan: fix underflow for the real_dev refcnt

[ Upstream commit 01d9cc2dea3fde3bad6d27f464eff463496e2b00 ]

Inject error before dev_hold(real_dev) in register_vlan_dev(),
and execute the following testcase:

ip link add dev dummy1 type dummy
ip link add name dummy1.100 link dummy1 type vlan id 100
ip link del dev dummy1

When the dummy netdevice is removed, we will get a WARNING as following:

=======================================================================
refcount_t: decrement hit 0; leaking memory.
WARNING: CPU: 2 PID: 0 at lib/refcount.c:31 refcount_warn_saturate+0xbf/0x1e0

and an endless loop of:

=======================================================================
unregister_netdevice: waiting for dummy1 to become free. Usage count = -1073741824

That is because dev_put(real_dev) in vlan_dev_free() be called without
dev_hold(real_dev) in register_vlan_dev(). It makes the refcnt of real_dev
underflow.

Move the dev_hold(real_dev) to vlan_dev_init() which is the call-back of
ndo_init(). That makes dev_hold() and dev_put() for vlan's real_dev
symmetrical.

Fixes: 563bcbae3ba2 ("net: vlan: fix a UAF in vlan_dev_real_dev()")
Reported-by: Petr Machata <petrm@nvidia.com>
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com>
Link: https://lore.kernel.org/r/20211126015942.2918542-1-william.xuanziyang@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet/sched: sch_ets: don't peek at classes beyond 'nbands'
Davide Caratti [Wed, 24 Nov 2021 16:14:40 +0000 (17:14 +0100)]
net/sched: sch_ets: don't peek at classes beyond 'nbands'

[ Upstream commit de6d25924c2a8c2988c6a385990cafbe742061bf ]

when the number of DRR classes decreases, the round-robin active list can
contain elements that have already been freed in ets_qdisc_change(). As a
consequence, it's possible to see a NULL dereference crash, caused by the
attempt to call cl->qdisc->ops->peek(cl->qdisc) when cl->qdisc is NULL:

 BUG: kernel NULL pointer dereference, address: 0000000000000018
 #PF: supervisor read access in kernel mode
 #PF: error_code(0x0000) - not-present page
 PGD 0 P4D 0
 Oops: 0000 [#1] PREEMPT SMP NOPTI
 CPU: 1 PID: 910 Comm: mausezahn Not tainted 5.16.0-rc1+ #475
 Hardware name: Red Hat KVM, BIOS 1.11.1-4.module+el8.1.0+4066+0f1aadab 04/01/2014
 RIP: 0010:ets_qdisc_dequeue+0x129/0x2c0 [sch_ets]
 Code: c5 01 41 39 ad e4 02 00 00 0f 87 18 ff ff ff 49 8b 85 c0 02 00 00 49 39 c4 0f 84 ba 00 00 00 49 8b ad c0 02 00 00 48 8b 7d 10 <48> 8b 47 18 48 8b 40 38 0f ae e8 ff d0 48 89 c3 48 85 c0 0f 84 9d
 RSP: 0000:ffffbb36c0b5fdd8 EFLAGS: 00010287
 RAX: ffff956678efed30 RBX: 0000000000000000 RCX: 0000000000000000
 RDX: 0000000000000002 RSI: ffffffff9b938dc9 RDI: 0000000000000000
 RBP: ffff956678efed30 R08: e2f3207fe360129c R09: 0000000000000000
 R10: 0000000000000001 R11: 0000000000000001 R12: ffff956678efeac0
 R13: ffff956678efe800 R14: ffff956611545000 R15: ffff95667ac8f100
 FS:  00007f2aa9120740(0000) GS:ffff95667b800000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 0000000000000018 CR3: 000000011070c000 CR4: 0000000000350ee0
 Call Trace:
  <TASK>
  qdisc_peek_dequeued+0x29/0x70 [sch_ets]
  tbf_dequeue+0x22/0x260 [sch_tbf]
  __qdisc_run+0x7f/0x630
  net_tx_action+0x290/0x4c0
  __do_softirq+0xee/0x4f8
  irq_exit_rcu+0xf4/0x130
  sysvec_apic_timer_interrupt+0x52/0xc0
  asm_sysvec_apic_timer_interrupt+0x12/0x20
 RIP: 0033:0x7f2aa7fc9ad4
 Code: b9 ff ff 48 8b 54 24 18 48 83 c4 08 48 89 ee 48 89 df 5b 5d e9 ed fc ff ff 0f 1f 00 66 2e 0f 1f 84 00 00 00 00 00 f3 0f 1e fa <53> 48 83 ec 10 48 8b 05 10 64 33 00 48 8b 00 48 85 c0 0f 85 84 00
 RSP: 002b:00007ffe5d33fab8 EFLAGS: 00000202
 RAX: 0000000000000002 RBX: 0000561f72c31460 RCX: 0000561f72c31720
 RDX: 0000000000000002 RSI: 0000561f72c31722 RDI: 0000561f72c31720
 RBP: 000000000000002a R08: 00007ffe5d33fa40 R09: 0000000000000014
 R10: 0000000000000000 R11: 0000000000000246 R12: 0000561f7187e380
 R13: 0000000000000000 R14: 0000000000000000 R15: 0000561f72c31460
  </TASK>
 Modules linked in: sch_ets sch_tbf dummy rfkill iTCO_wdt intel_rapl_msr iTCO_vendor_support intel_rapl_common joydev virtio_balloon lpc_ich i2c_i801 i2c_smbus pcspkr ip_tables xfs libcrc32c crct10dif_pclmul crc32_pclmul crc32c_intel ahci libahci ghash_clmulni_intel serio_raw libata virtio_blk virtio_console virtio_net net_failover failover sunrpc dm_mirror dm_region_hash dm_log dm_mod
 CR2: 0000000000000018

Ensuring that 'alist' was never zeroed [1] was not sufficient, we need to
remove from the active list those elements that are no more SP nor DRR.

[1] https://lore.kernel.org/netdev/60d274838bf09777f0371253416e8af71360bc08.1633609148.git.dcaratti@redhat.com/

v3: fix race between ets_qdisc_change() and ets_qdisc_dequeue() delisting
    DRR classes beyond 'nbands' in ets_qdisc_change() with the qdisc lock
    acquired, thanks to Cong Wang.

v2: when a NULL qdisc is found in the DRR active list, try to dequeue skb
    from the next list item.

Reported-by: Hangbin Liu <liuhangbin@gmail.com>
Fixes: dcc68b4d8084 ("net: sch_ets: Add a new Qdisc")
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Link: https://lore.kernel.org/r/7a5c496eed2d62241620bdbb83eb03fb9d571c99.1637762721.git.dcaratti@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agotls: fix replacing proto_ops
Jakub Kicinski [Wed, 24 Nov 2021 23:25:56 +0000 (15:25 -0800)]
tls: fix replacing proto_ops

[ Upstream commit f3911f73f51d1534f4db70b516cc1fcb6be05bae ]

We replace proto_ops whenever TLS is configured for RX. But our
replacement also overrides sendpage_locked, which will crash
unless TX is also configured. Similarly we plug both of those
in for TLS_HW (NIC crypto offload) even tho TLS_HW has a completely
different implementation for TX.

Last but not least we always plug in something based on inet_stream_ops
even though a few of the callbacks differ for IPv6 (getname, release,
bind).

Use a callback building method similar to what we do for struct proto.

Fixes: c46234ebb4d1 ("tls: RX path for ktls")
Fixes: d4ffb02dee2f ("net/tls: enable sk_msg redirect to tls socket egress")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agotls: splice_read: fix record type check
Jakub Kicinski [Wed, 24 Nov 2021 23:25:52 +0000 (15:25 -0800)]
tls: splice_read: fix record type check

[ Upstream commit 520493f66f6822551aef2879cd40207074fe6980 ]

We don't support splicing control records. TLS 1.3 changes moved
the record type check into the decrypt if(). The skb may already
be decrypted and still be an alert.

Note that decrypt_skb_update() is idempotent and updates ctx->decrypted
so the if() is pointless.

Reorder the check for decryption errors with the content type check
while touching them. This part is not really a bug, because if
decryption failed in TLS 1.3 content type will be DATA, and for
TLS 1.2 it will be correct. Nevertheless its strange to touch output
before checking if the function has failed.

Fixes: fedf201e1296 ("net: tls: Refactor control message handling on recv")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoMIPS: use 3-level pgtable for 64KB page size on MIPS_VA_BITS_48
Huang Pei [Thu, 25 Nov 2021 10:59:48 +0000 (18:59 +0800)]
MIPS: use 3-level pgtable for 64KB page size on MIPS_VA_BITS_48

[ Upstream commit 41ce097f714401e6ad8f3f5eb30d7f91b0b5e495 ]

It hangup when booting Loongson 3A1000 with BOTH
CONFIG_PAGE_SIZE_64KB and CONFIG_MIPS_VA_BITS_48, that it turn
out to use 2-level pgtable instead of 3-level. 64KB page size
with 2-level pgtable only cover 42 bits VA, use 3-level pgtable
to cover all 48 bits VA(55 bits)

Fixes: 1e321fa917fb ("MIPS64: Support of at least 48 bits of SEGBITS)
Signed-off-by: Huang Pei <huangpei@loongson.cn>
Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoMIPS: loongson64: fix FTLB configuration
Huang Pei [Thu, 25 Nov 2021 10:59:49 +0000 (18:59 +0800)]
MIPS: loongson64: fix FTLB configuration

[ Upstream commit 7db5e9e9e5e6c10d7d26f8df7f8fd8841cb15ee7 ]

It turns out that 'decode_configs' -> 'set_ftlb_enable' is called under
c->cputype unset, which leaves FTLB disabled on BOTH 3A2000 and 3A3000

Fix it by calling "decode_configs" after c->cputype is initialized

Fixes: da1bd29742b1 ("MIPS: Loongson64: Probe CPU features via CPUCFG")
Signed-off-by: Huang Pei <huangpei@loongson.cn>
Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoigb: fix netpoll exit with traffic
Jesse Brandeburg [Tue, 23 Nov 2021 20:40:00 +0000 (12:40 -0800)]
igb: fix netpoll exit with traffic

[ Upstream commit eaeace60778e524a2820d0c0ad60bf80289e292c ]

Oleksandr brought a bug report where netpoll causes trace
messages in the log on igb.

Danielle brought this back up as still occurring, so we'll try
again.

[22038.710800] ------------[ cut here ]------------
[22038.710801] igb_poll+0x0/0x1440 [igb] exceeded budget in poll
[22038.710802] WARNING: CPU: 12 PID: 40362 at net/core/netpoll.c:155 netpoll_poll_dev+0x18a/0x1a0

As Alex suggested, change the driver to return work_done at the
exit of napi_poll, which should be safe to do in this driver
because it is not polling multiple queues in this single napi
context (multiple queues attached to one MSI-X vector). Several
other drivers contain the same simple sequence, so I hope
this will not create new problems.

Fixes: 16eb8815c235 ("igb: Refactor clean_rx_irq to reduce overhead and improve performance")
Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name>
Reported-by: Danielle Ratson <danieller@nvidia.com>
Suggested-by: Alexander Duyck <alexander.duyck@gmail.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name>
Tested-by: Danielle Ratson <danieller@nvidia.com>
Link: https://lore.kernel.org/r/20211123204000.1597971-1-jesse.brandeburg@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonvmet: use IOCB_NOWAIT only if the filesystem supports it
Maurizio Lombardi [Mon, 22 Nov 2021 10:08:27 +0000 (11:08 +0100)]
nvmet: use IOCB_NOWAIT only if the filesystem supports it

[ Upstream commit c024b226a417c4eb9353ff500b1c823165d4d508 ]

Submit I/O requests with the IOCB_NOWAIT flag set only if
the underlying filesystem supports it.

Fixes: 50a909db36f2 ("nvmet: use IOCB_NOWAIT for file-ns buffered I/O")
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet/smc: Fix loop in smc_listen
Guo DaXing [Wed, 24 Nov 2021 12:32:38 +0000 (13:32 +0100)]
net/smc: Fix loop in smc_listen

[ Upstream commit 9ebb0c4b27a6158303b791b5b91e66d7665ee30e ]

The kernel_listen function in smc_listen will fail when all the available
ports are occupied.  At this point smc->clcsock->sk->sk_data_ready has
been changed to smc_clcsock_data_ready.  When we call smc_listen again,
now both smc->clcsock->sk->sk_data_ready and smc->clcsk_data_ready point
to the smc_clcsock_data_ready function.

The smc_clcsock_data_ready() function calls lsmc->clcsk_data_ready which
now points to itself resulting in an infinite loop.

This patch restores smc->clcsock->sk->sk_data_ready with the old value.

Fixes: a60a2b1e0af1 ("net/smc: reduce active tcp_listen workers")
Signed-off-by: Guo DaXing <guodaxing@huawei.com>
Acked-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet/smc: Fix NULL pointer dereferencing in smc_vlan_by_tcpsk()
Karsten Graul [Wed, 24 Nov 2021 12:32:37 +0000 (13:32 +0100)]
net/smc: Fix NULL pointer dereferencing in smc_vlan_by_tcpsk()

[ Upstream commit 587acad41f1bc48e16f42bb2aca63bf323380be8 ]

Coverity reports a possible NULL dereferencing problem:

in smc_vlan_by_tcpsk():
6. returned_null: netdev_lower_get_next returns NULL (checked 29 out of 30 times).
7. var_assigned: Assigning: ndev = NULL return value from netdev_lower_get_next.
1623                ndev = (struct net_device *)netdev_lower_get_next(ndev, &lower);
CID 1468509 (#1 of 1): Dereference null return value (NULL_RETURNS)
8. dereference: Dereferencing a pointer that might be NULL ndev when calling is_vlan_dev.
1624                if (is_vlan_dev(ndev)) {

Remove the manual implementation and use netdev_walk_all_lower_dev() to
iterate over the lower devices. While on it remove an obsolete function
parameter comment.

Fixes: cb9d43f67754 ("net/smc: determine vlan_id of stacked net_device")
Suggested-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: phylink: Force retrigger in case of latched link-fail indicator
Russell King (Oracle) [Tue, 23 Nov 2021 15:44:03 +0000 (16:44 +0100)]
net: phylink: Force retrigger in case of latched link-fail indicator

[ Upstream commit dbae3388ea9ca33bd1d5eabc3b0ef17e69c74677 ]

On mv88e6xxx 1G/2.5G PCS, the SerDes register 4.2001.2 has the following
description:
  This register bit indicates when link was lost since the last
  read. For the current link status, read this register
  back-to-back.

Thus to get current link state, we need to read the register twice.

But doing that in the link change interrupt handler would lead to
potentially ignoring link down events, which we really want to avoid.

Thus this needs to be solved in phylink's resolve, by retriggering
another resolve in the event when PCS reports link down and previous
link was up, and by re-reading PCS state if the previous link was down.

The wrong value is read when phylink requests change from sgmii to
2500base-x mode, and link won't come up. This fixes the bug.

Fixes: 9525ae83959b ("phylink: add phylink infrastructure")
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: Marek Behún <kabel@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: phylink: Force link down and retrigger resolve on interface change
Russell King (Oracle) [Tue, 23 Nov 2021 15:44:02 +0000 (16:44 +0100)]
net: phylink: Force link down and retrigger resolve on interface change

[ Upstream commit 80662f4fd4771bc9c7cc4abdfbe866ebd1179621 ]

On PHY state change the phylink_resolve() function can read stale
information from the MAC and report incorrect link speed and duplex to
the kernel message log.

Example with a Marvell 88X3310 PHY connected to a SerDes port on Marvell
88E6393X switch:
- PHY driver triggers state change due to PHY interface mode being
  changed from 10gbase-r to 2500base-x due to copper change in speed
  from 10Gbps to 2.5Gbps, but the PHY itself either hasn't yet changed
  its interface to the host, or the interrupt about loss of SerDes link
  hadn't arrived yet (there can be a delay of several milliseconds for
  this), so we still think that the 10gbase-r mode is up
- phylink_resolve()
  - phylink_mac_pcs_get_state()
    - this fills in speed=10g link=up
  - interface mode is updated to 2500base-x but speed is left at 10Gbps
  - phylink_major_config()
    - interface is changed to 2500base-x
  - phylink_link_up()
    - mv88e6xxx_mac_link_up()
      - .port_set_speed_duplex()
        - speed is set to 10Gbps
    - reports "Link is Up - 10Gbps/Full" to dmesg

Afterwards when the interrupt finally arrives for mv88e6xxx, another
resolve is forced in which we get the correct speed from
phylink_mac_pcs_get_state(), but since the interface is not being
changed anymore, we don't call phylink_major_config() but only
phylink_mac_config(), which does not set speed/duplex anymore.

To fix this, we need to force the link down and trigger another resolve
on PHY interface change event.

Fixes: 9525ae83959b ("phylink: add phylink infrastructure")
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: Marek Behún <kabel@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agolan743x: fix deadlock in lan743x_phy_link_status_change()
Heiner Kallweit [Wed, 24 Nov 2021 07:16:25 +0000 (08:16 +0100)]
lan743x: fix deadlock in lan743x_phy_link_status_change()

[ Upstream commit ddb826c2c92d461f290a7bab89e7c28696191875 ]

Usage of phy_ethtool_get_link_ksettings() in the link status change
handler isn't needed, and in combination with the referenced change
it results in a deadlock. Simply remove the call and replace it with
direct access to phydev->speed. The duplex argument of
lan743x_phy_update_flowcontrol() isn't used and can be removed.

Fixes: c10a485c3de5 ("phy: phy_ethtool_ksettings_get: Lock the phy for consistency")
Reported-by: Alessandro B Maurici <abmaurici@gmail.com>
Tested-by: Alessandro B Maurici <abmaurici@gmail.com>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://lore.kernel.org/r/40e27f76-0ba3-dcef-ee32-a78b9df38b0f@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agotcp_cubic: fix spurious Hystart ACK train detections for not-cwnd-limited flows
Eric Dumazet [Tue, 23 Nov 2021 20:25:35 +0000 (12:25 -0800)]
tcp_cubic: fix spurious Hystart ACK train detections for not-cwnd-limited flows

[ Upstream commit 4e1fddc98d2585ddd4792b5e44433dcee7ece001 ]

While testing BIG TCP patch series, I was expecting that TCP_RR workloads
with 80KB requests/answers would send one 80KB TSO packet,
then being received as a single GRO packet.

It turns out this was not happening, and the root cause was that
cubic Hystart ACK train was triggering after a few (2 or 3) rounds of RPC.

Hystart was wrongly setting CWND/SSTHRESH to 30, while my RPC
needed a budget of ~20 segments.

Ideally these TCP_RR flows should not exit slow start.

Cubic Hystart should reset itself at each round, instead of assuming
every TCP flow is a bulk one.

Note that even after this patch, Hystart can still trigger, depending
on scheduling artifacts, but at a higher CWND/SSTHRESH threshold,
keeping optimal TSO packet sizes.

Tested:

ip link set dev eth0 gro_ipv6_max_size 131072 gso_ipv6_max_size 131072
nstat -n; netperf -H ... -t TCP_RR  -l 5  -- -r 80000,80000 -K cubic; nstat|egrep "Ip6InReceives|Hystart|Ip6OutRequests"

Before:

   8605
Ip6InReceives                   87541              0.0
Ip6OutRequests                  129496             0.0
TcpExtTCPHystartTrainDetect     1                  0.0
TcpExtTCPHystartTrainCwnd       30                 0.0

After:

  8760
Ip6InReceives                   88514              0.0
Ip6OutRequests                  87975              0.0

Fixes: ae27e98a5152 ("[TCP] CUBIC v2.3")
Co-developed-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Link: https://lore.kernel.org/r/20211123202535.1843771-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agodrm/amd/display: Set plane update flags for all planes in reset
Nicholas Kazlauskas [Tue, 9 Nov 2021 16:24:10 +0000 (11:24 -0500)]
drm/amd/display: Set plane update flags for all planes in reset

[ Upstream commit 21431f70f6014f81b0d118ff4fcee12b00b9dd70 ]

[Why]
We're only setting the flags on stream[0]'s planes so this logic fails
if we have more than one stream in the state.

This can cause a page flip timeout with multiple displays in the
configuration.

[How]
Index into the stream_status array using the stream index - it's a 1:1
mapping.

Fixes: cdaae8371aa9 ("drm/amd/display: Handle GPU reset for DC block")

Reviewed-by: Harry Wentland <Harry.Wentland@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoPM: hibernate: use correct mode for swsusp_close()
Thomas Zeitlhofer [Tue, 23 Nov 2021 19:18:43 +0000 (20:18 +0100)]
PM: hibernate: use correct mode for swsusp_close()

[ Upstream commit cefcf24b4d351daf70ecd945324e200d3736821e ]

Commit 39fbef4b0f77 ("PM: hibernate: Get block device exclusively in
swsusp_check()") changed the opening mode of the block device to
(FMODE_READ | FMODE_EXCL).

In the corresponding calls to swsusp_close(), the mode is still just
FMODE_READ which triggers the warning in blkdev_flush_mapping() on
resume from hibernate.

So, use the mode (FMODE_READ | FMODE_EXCL) also when closing the
device.

Fixes: 39fbef4b0f77 ("PM: hibernate: Get block device exclusively in swsusp_check()")
Signed-off-by: Thomas Zeitlhofer <thomas.zeitlhofer+lkml@ze-it.at>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet/ncsi : Add payload to be 32-bit aligned to fix dropped packets
Kumar Thangavel [Mon, 22 Nov 2021 16:38:18 +0000 (22:08 +0530)]
net/ncsi : Add payload to be 32-bit aligned to fix dropped packets

[ Upstream commit ac132852147ad303a938dda318970dd1bbdfda4e ]

Update NC-SI command handler (both standard and OEM) to take into
account of payload paddings in allocating skb (in case of payload
size is not 32-bit aligned).

The checksum field follows payload field, without taking payload
padding into account can cause checksum being truncated, leading to
dropped packets.

Fixes: fb4ee67529ff ("net/ncsi: Add NCSI OEM command support")
Signed-off-by: Kumar Thangavel <thangavel.k@hcl.com>
Acked-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonvmet-tcp: fix incomplete data digest send
Varun Prakash [Mon, 22 Nov 2021 10:08:41 +0000 (15:38 +0530)]
nvmet-tcp: fix incomplete data digest send

[ Upstream commit 102110efdff6beedece6ab9b51664c32ac01e2db ]

Current nvmet_try_send_ddgst() code does not check whether
all data digest bytes are transmitted, fix this by returning
-EAGAIN if all data digest bytes are not transmitted.

Fixes: 872d26a391da ("nvmet-tcp: add NVMe over TCP target driver")
Signed-off-by: Varun Prakash <varun@chelsio.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: marvell: mvpp2: increase MTU limit when XDP enabled
Marek Behún [Mon, 22 Nov 2021 20:08:34 +0000 (21:08 +0100)]
net: marvell: mvpp2: increase MTU limit when XDP enabled

[ Upstream commit 7b1b62bc1e6a7b2fd5ee7a4296268eb291d23aeb ]

Currently mvpp2_xdp_setup won't allow attaching XDP program if
  mtu > ETH_DATA_LEN (1500).

The mvpp2_change_mtu on the other hand checks whether
  MVPP2_RX_PKT_SIZE(mtu) > MVPP2_BM_LONG_PKT_SIZE.

These two checks are semantically different.

Moreover this limit can be increased to MVPP2_MAX_RX_BUF_SIZE, since in
mvpp2_rx we have
  xdp.data = data + MVPP2_MH_SIZE + MVPP2_SKB_HEADROOM;
  xdp.frame_sz = PAGE_SIZE;

Change the checks to check whether
  mtu > MVPP2_MAX_RX_BUF_SIZE

Fixes: 07dd0a7aae7f ("mvpp2: add basic XDP support")
Signed-off-by: Marek Behún <kabel@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agomlxsw: spectrum: Protect driver from buggy firmware
Amit Cohen [Tue, 23 Nov 2021 07:52:56 +0000 (09:52 +0200)]
mlxsw: spectrum: Protect driver from buggy firmware

[ Upstream commit 63b08b1f6834bbb0b4f7783bf63b80c8c8e9a047 ]

When processing port up/down events generated by the device's firmware,
the driver protects itself from events reported for non-existent local
ports, but not the CPU port (local port 0), which exists, but lacks a
netdev.

This can result in a NULL pointer dereference when calling
netif_carrier_{on,off}().

Fix this by bailing early when processing an event reported for the CPU
port. Problem was only observed when running on top of a buggy emulator.

Fixes: 28b1987ef506 ("mlxsw: spectrum: Register CPU port with devlink")
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agomlxsw: Verify the accessed index doesn't exceed the array length
Danielle Ratson [Mon, 17 May 2021 17:03:57 +0000 (20:03 +0300)]
mlxsw: Verify the accessed index doesn't exceed the array length

[ Upstream commit 837ec05cfea08284c575e8e834777b107da5ff9d ]

There are few cases in which an array index queried from a fw register,
is accessed without any validation that it doesn't exceed the array
length.

Add a proper length validation, so accessing memory past the end of an
array will be forbidden.

Signed-off-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet/smc: Ensure the active closing peer first closes clcsock
Tony Lu [Tue, 23 Nov 2021 08:25:18 +0000 (16:25 +0800)]
net/smc: Ensure the active closing peer first closes clcsock

[ Upstream commit 606a63c9783a32a45bd2ef0eee393711d75b3284 ]

The side that actively closed socket, it's clcsock doesn't enter
TIME_WAIT state, but the passive side does it. It should show the same
behavior as TCP sockets.

Consider this, when client actively closes the socket, the clcsock in
server enters TIME_WAIT state, which means the address is occupied and
won't be reused before TIME_WAIT dismissing. If we restarted server, the
service would be unavailable for a long time.

To solve this issue, shutdown the clcsock in [A], perform the TCP active
close progress first, before the passive closed side closing it. So that
the actively closed side enters TIME_WAIT, not the passive one.

Client                                            |  Server
close() // client actively close                  |
  smc_release()                                   |
      smc_close_active() // PEERCLOSEWAIT1        |
          smc_close_final() // abort or closed = 1|
              smc_cdc_get_slot_and_msg_send()     |
          [A]                                     |
                                                  |smc_cdc_msg_recv_action() // ACTIVE
                                                  |  queue_work(smc_close_wq, &conn->close_work)
                                                  |    smc_close_passive_work() // PROCESSABORT or APPCLOSEWAIT1
                                                  |      smc_close_passive_abort_received() // only in abort
                                                  |
                                                  |close() // server recv zero, close
                                                  |  smc_release() // PROCESSABORT or APPCLOSEWAIT1
                                                  |    smc_close_active()
                                                  |      smc_close_abort() or smc_close_final() // CLOSED
                                                  |        smc_cdc_get_slot_and_msg_send() // abort or closed = 1
smc_cdc_msg_recv_action()                         |    smc_clcsock_release()
  queue_work(smc_close_wq, &conn->close_work)     |      sock_release(tcp) // actively close clc, enter TIME_WAIT
    smc_close_passive_work() // PEERCLOSEWAIT1    |    smc_conn_free()
      smc_close_passive_abort_received() // CLOSED|
      smc_conn_free()                             |
      smc_clcsock_release()                       |
        sock_release(tcp) // passive close clc    |

Link: https://www.spinics.net/lists/netdev/msg780407.html
Fixes: b38d732477e4 ("smc: socket closing and linkgroup cleanup")
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Reviewed-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoerofs: fix deadlock when shrink erofs slab
Huang Jianan [Thu, 18 Nov 2021 13:58:44 +0000 (21:58 +0800)]
erofs: fix deadlock when shrink erofs slab

[ Upstream commit 57bbeacdbee72a54eb97d56b876cf9c94059fc34 ]

We observed the following deadlock in the stress test under low
memory scenario:

Thread A                               Thread B
- erofs_shrink_scan
 - erofs_try_to_release_workgroup
  - erofs_workgroup_try_to_freeze -- A
                                       - z_erofs_do_read_page
                                        - z_erofs_collection_begin
                                         - z_erofs_register_collection
                                          - erofs_insert_workgroup
                                           - xa_lock(&sbi->managed_pslots) -- B
                                           - erofs_workgroup_get
                                            - erofs_wait_on_workgroup_freezed -- A
  - xa_erase
   - xa_lock(&sbi->managed_pslots) -- B

To fix this, it needs to hold xa_lock before freezing the workgroup
since xarray will be touched then. So let's hold the lock before
accessing each workgroup, just like what we did with the radix tree
before.

[ Gao Xiang: Jianhua Hao also reports this issue at
  https://lore.kernel.org/r/b10b85df30694bac8aadfe43537c897a@xiaomi.com ]

Link: https://lore.kernel.org/r/20211118135844.3559-1-huangjianan@oppo.com
Fixes: 64094a04414f ("erofs: convert workstn to XArray")
Reviewed-by: Chao Yu <chao@kernel.org>
Reviewed-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Signed-off-by: Huang Jianan <huangjianan@oppo.com>
Reported-by: Jianhua Hao <haojianhua1@xiaomi.com>
Signed-off-by: Gao Xiang <xiang@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoscsi: scsi_debug: Zero clear zones at reset write pointer
Shin'ichiro Kawasaki [Mon, 22 Nov 2021 06:12:23 +0000 (15:12 +0900)]
scsi: scsi_debug: Zero clear zones at reset write pointer

[ Upstream commit 2d62253eb1b60f4ce8b39125eee282739b519297 ]

When a reset is requested the position of the write pointer is updated but
the data in the corresponding zone is not cleared. Instead scsi_debug
returns any data written before the write pointer was reset. This is an
error and prevents using scsi_debug for stale page cache testing of the
BLKRESETZONE ioctl.

Zero written data in the zone when resetting the write pointer.

Link: https://lore.kernel.org/r/20211122061223.298890-1-shinichiro.kawasaki@wdc.com
Fixes: f0d1cf9378bd ("scsi: scsi_debug: Add ZBC zone commands")
Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoscsi: core: sysfs: Fix setting device state to SDEV_RUNNING
Mike Christie [Sat, 20 Nov 2021 16:49:17 +0000 (10:49 -0600)]
scsi: core: sysfs: Fix setting device state to SDEV_RUNNING

[ Upstream commit eb97545d6264b341b06ba7603f52ff6c0b2af6ea ]

This fixes an issue added in commit 4edd8cd4e86d ("scsi: core: sysfs: Fix
hang when device state is set via sysfs") where if userspace is requesting
to set the device state to SDEV_RUNNING when the state is already
SDEV_RUNNING, we return -EINVAL instead of count. The commmit above set ret
to count for this case, when it should have set it to 0.

Link: https://lore.kernel.org/r/20211120164917.4924-1-michael.christie@oracle.com
Fixes: 4edd8cd4e86d ("scsi: core: sysfs: Fix hang when device state is set via sysfs")
Reviewed-by: Lee Duncan <lduncan@suse.com>
Signed-off-by: Mike Christie <michael.christie@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoice: avoid bpf_prog refcount underflow
Marta Plantykow [Tue, 26 Oct 2021 16:47:19 +0000 (18:47 +0200)]
ice: avoid bpf_prog refcount underflow

[ Upstream commit f65ee535df775a13a1046c0a0b2d72db342f8a5b ]

Ice driver has the routines for managing XDP resources that are shared
between ndo_bpf op and VSI rebuild flow. The latter takes place for
example when user changes queue count on an interface via ethtool's
set_channels().

There is an issue around the bpf_prog refcounting when VSI is being
rebuilt - since ice_prepare_xdp_rings() is called with vsi->xdp_prog as
an argument that is used later on by ice_vsi_assign_bpf_prog(), same
bpf_prog pointers are swapped with each other. Then it is also
interpreted as an 'old_prog' which in turn causes us to call
bpf_prog_put on it that will decrement its refcount.

Below splat can be interpreted in a way that due to zero refcount of a
bpf_prog it is wiped out from the system while kernel still tries to
refer to it:

[  481.069429] BUG: unable to handle page fault for address: ffffc9000640f038
[  481.077390] #PF: supervisor read access in kernel mode
[  481.083335] #PF: error_code(0x0000) - not-present page
[  481.089276] PGD 100000067 P4D 100000067 PUD 1001cb067 PMD 106d2b067 PTE 0
[  481.097141] Oops: 0000 [#1] PREEMPT SMP PTI
[  481.101980] CPU: 12 PID: 3339 Comm: sudo Tainted: G           OE     5.15.0-rc5+ #1
[  481.110840] Hardware name: Intel Corp. GRANTLEY/GRANTLEY, BIOS GRRFCRB1.86B.0276.D07.1605190235 05/19/2016
[  481.122021] RIP: 0010:dev_xdp_prog_id+0x25/0x40
[  481.127265] Code: 80 00 00 00 00 0f 1f 44 00 00 89 f6 48 c1 e6 04 48 01 fe 48 8b 86 98 08 00 00 48 85 c0 74 13 48 8b 50 18 31 c0 48 85 d2 74 07 <48> 8b 42 38 8b 40 20 c3 48 8b 96 90 08 00 00 eb e8 66 2e 0f 1f 84
[  481.148991] RSP: 0018:ffffc90007b63868 EFLAGS: 00010286
[  481.155034] RAX: 0000000000000000 RBX: ffff889080824000 RCX: 0000000000000000
[  481.163278] RDX: ffffc9000640f000 RSI: ffff889080824010 RDI: ffff889080824000
[  481.171527] RBP: ffff888107af7d00 R08: 0000000000000000 R09: ffff88810db5f6e0
[  481.179776] R10: 0000000000000000 R11: ffff8890885b9988 R12: ffff88810db5f4bc
[  481.188026] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
[  481.196276] FS:  00007f5466d5bec0(0000) GS:ffff88903fb00000(0000) knlGS:0000000000000000
[  481.205633] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  481.212279] CR2: ffffc9000640f038 CR3: 000000014429c006 CR4: 00000000003706e0
[  481.220530] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  481.228771] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  481.237029] Call Trace:
[  481.239856]  rtnl_fill_ifinfo+0x768/0x12e0
[  481.244602]  rtnl_dump_ifinfo+0x525/0x650
[  481.249246]  ? __alloc_skb+0xa5/0x280
[  481.253484]  netlink_dump+0x168/0x3c0
[  481.257725]  netlink_recvmsg+0x21e/0x3e0
[  481.262263]  ____sys_recvmsg+0x87/0x170
[  481.266707]  ? __might_fault+0x20/0x30
[  481.271046]  ? _copy_from_user+0x66/0xa0
[  481.275591]  ? iovec_from_user+0xf6/0x1c0
[  481.280226]  ___sys_recvmsg+0x82/0x100
[  481.284566]  ? sock_sendmsg+0x5e/0x60
[  481.288791]  ? __sys_sendto+0xee/0x150
[  481.293129]  __sys_recvmsg+0x56/0xa0
[  481.297267]  do_syscall_64+0x3b/0xc0
[  481.301395]  entry_SYSCALL_64_after_hwframe+0x44/0xae
[  481.307238] RIP: 0033:0x7f5466f39617
[  481.311373] Code: 0c 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb bd 0f 1f 00 f3 0f 1e fa 64 8b 04 25 18 00 00 00 85 c0 75 10 b8 2f 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 51 c3 48 83 ec 28 89 54 24 1c 48 89 74 24 10
[  481.342944] RSP: 002b:00007ffedc7f4308 EFLAGS: 00000246 ORIG_RAX: 000000000000002f
[  481.361783] RAX: ffffffffffffffda RBX: 00007ffedc7f5460 RCX: 00007f5466f39617
[  481.380278] RDX: 0000000000000000 RSI: 00007ffedc7f5360 RDI: 0000000000000003
[  481.398500] RBP: 00007ffedc7f53f0 R08: 0000000000000000 R09: 000055d556f04d50
[  481.416463] R10: 0000000000000077 R11: 0000000000000246 R12: 00007ffedc7f5360
[  481.434131] R13: 00007ffedc7f5350 R14: 00007ffedc7f5344 R15: 0000000000000e98
[  481.451520] Modules linked in: ice(OE) af_packet binfmt_misc nls_iso8859_1 ipmi_ssif intel_rapl_msr intel_rapl_common x86_pkg_temp_thermal intel_powerclamp mxm_wmi mei_me coretemp mei ipmi_si ipmi_msghandler wmi acpi_pad acpi_power_meter ip_tables x_tables autofs4 crct10dif_pclmul crc32_pclmul ghash_clmulni_intel aesni_intel ahci crypto_simd cryptd libahci lpc_ich [last unloaded: ice]
[  481.528558] CR2: ffffc9000640f038
[  481.542041] ---[ end trace d1f24c9ecf5b61c1 ]---

Fix this by only calling ice_vsi_assign_bpf_prog() inside
ice_prepare_xdp_rings() when current vsi->xdp_prog pointer is NULL.
This way set_channels() flow will not attempt to swap the vsi->xdp_prog
pointers with itself.

Also, sprinkle around some comments that provide a reasoning about
correlation between driver and kernel in terms of bpf_prog refcount.

Fixes: efc2214b6047 ("ice: Add support for XDP")
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Signed-off-by: Marta Plantykow <marta.a.plantykow@intel.com>
Co-developed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Tested-by: Kiran Bhandare <kiranx.bhandare@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoice: fix vsi->txq_map sizing
Maciej Fijalkowski [Tue, 26 Oct 2021 16:47:18 +0000 (18:47 +0200)]
ice: fix vsi->txq_map sizing

[ Upstream commit 792b2086584f25d84081a526beee80d103c2a913 ]

The approach of having XDP queue per CPU regardless of user's setting
exposed a hidden bug that could occur in case when Rx queue count differ
from Tx queue count. Currently vsi->txq_map's size is equal to the
doubled vsi->alloc_txq, which is not correct due to the fact that XDP
rings were previously based on the Rx queue count. Below splat can be
seen when ethtool -L is used and XDP rings are configured:

[  682.875339] BUG: kernel NULL pointer dereference, address: 000000000000000f
[  682.883403] #PF: supervisor read access in kernel mode
[  682.889345] #PF: error_code(0x0000) - not-present page
[  682.895289] PGD 0 P4D 0
[  682.898218] Oops: 0000 [#1] PREEMPT SMP PTI
[  682.903055] CPU: 42 PID: 2878 Comm: ethtool Tainted: G           OE     5.15.0-rc5+ #1
[  682.912214] Hardware name: Intel Corp. GRANTLEY/GRANTLEY, BIOS GRRFCRB1.86B.0276.D07.1605190235 05/19/2016
[  682.923380] RIP: 0010:devres_remove+0x44/0x130
[  682.928527] Code: 49 89 f4 55 48 89 fd 4c 89 ff 53 48 83 ec 10 e8 92 b9 49 00 48 8b 9d a8 02 00 00 48 8d 8d a0 02 00 00 49 89 c2 48 39 cb 74 0f <4c> 3b 63 10 74 25 48 8b 5b 08 48 39 cb 75 f1 4c 89 ff 4c 89 d6 e8
[  682.950237] RSP: 0018:ffffc90006a679f0 EFLAGS: 00010002
[  682.956285] RAX: 0000000000000286 RBX: ffffffffffffffff RCX: ffff88908343a370
[  682.964538] RDX: 0000000000000001 RSI: ffffffff81690d60 RDI: 0000000000000000
[  682.972789] RBP: ffff88908343a0d0 R08: 0000000000000000 R09: 0000000000000000
[  682.981040] R10: 0000000000000286 R11: 3fffffffffffffff R12: ffffffff81690d60
[  682.989282] R13: ffffffff81690a00 R14: ffff8890819807a8 R15: ffff88908343a36c
[  682.997535] FS:  00007f08c7bfa740(0000) GS:ffff88a03fd00000(0000) knlGS:0000000000000000
[  683.006910] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  683.013557] CR2: 000000000000000f CR3: 0000001080a66003 CR4: 00000000003706e0
[  683.021819] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  683.030075] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  683.038336] Call Trace:
[  683.041167]  devm_kfree+0x33/0x50
[  683.045004]  ice_vsi_free_arrays+0x5e/0xc0 [ice]
[  683.050380]  ice_vsi_rebuild+0x4c8/0x750 [ice]
[  683.055543]  ice_vsi_recfg_qs+0x9a/0x110 [ice]
[  683.060697]  ice_set_channels+0x14f/0x290 [ice]
[  683.065962]  ethnl_set_channels+0x333/0x3f0
[  683.070807]  genl_family_rcv_msg_doit+0xea/0x150
[  683.076152]  genl_rcv_msg+0xde/0x1d0
[  683.080289]  ? channels_prepare_data+0x60/0x60
[  683.085432]  ? genl_get_cmd+0xd0/0xd0
[  683.089667]  netlink_rcv_skb+0x50/0xf0
[  683.094006]  genl_rcv+0x24/0x40
[  683.097638]  netlink_unicast+0x239/0x340
[  683.102177]  netlink_sendmsg+0x22e/0x470
[  683.106717]  sock_sendmsg+0x5e/0x60
[  683.110756]  __sys_sendto+0xee/0x150
[  683.114894]  ? handle_mm_fault+0xd0/0x2a0
[  683.119535]  ? do_user_addr_fault+0x1f3/0x690
[  683.134173]  __x64_sys_sendto+0x25/0x30
[  683.148231]  do_syscall_64+0x3b/0xc0
[  683.161992]  entry_SYSCALL_64_after_hwframe+0x44/0xae

Fix this by taking into account the value that num_possible_cpus()
yields in addition to vsi->alloc_txq instead of doubling the latter.

Fixes: efc2214b6047 ("ice: Add support for XDP")
Fixes: 22bf877e528f ("ice: introduce XDP_TX fallback path")
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Tested-by: Kiran Bhandare <kiranx.bhandare@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: nexthop: release IPv6 per-cpu dsts when replacing a nexthop group
Nikolay Aleksandrov [Mon, 22 Nov 2021 15:15:13 +0000 (17:15 +0200)]
net: nexthop: release IPv6 per-cpu dsts when replacing a nexthop group

[ Upstream commit 1005f19b9357b81aa64e1decd08d6e332caaa284 ]

When replacing a nexthop group, we must release the IPv6 per-cpu dsts of
the removed nexthop entries after an RCU grace period because they
contain references to the nexthop's net device and to the fib6 info.
With specific series of events[1] we can reach net device refcount
imbalance which is unrecoverable. IPv4 is not affected because dsts
don't take a refcount on the route.

[1]
 $ ip nexthop list
  id 200 via 2002:db8::2 dev bridge.10 scope link onlink
  id 201 via 2002:db8::3 dev bridge scope link onlink
  id 203 group 201/200
 $ ip -6 route
  2001:db8::10 nhid 203 metric 1024 pref medium
     nexthop via 2002:db8::3 dev bridge weight 1 onlink
     nexthop via 2002:db8::2 dev bridge.10 weight 1 onlink

Create rt6_info through one of the multipath legs, e.g.:
 $ taskset -a -c 1  ./pkt_inj 24 bridge.10 2001:db8::10
 (pkt_inj is just a custom packet generator, nothing special)

Then remove that leg from the group by replace (let's assume it is id
200 in this case):
 $ ip nexthop replace id 203 group 201

Now remove the IPv6 route:
 $ ip -6 route del 2001:db8::10/128

The route won't be really deleted due to the stale rt6_info holding 1
refcnt in nexthop id 200.
At this point we have the following reference count dependency:
 (deleted) IPv6 route holds 1 reference over nhid 203
 nh 203 holds 1 ref over id 201
 nh 200 holds 1 ref over the net device and the route due to the stale
 rt6_info

Now to create circular dependency between nh 200 and the IPv6 route, and
also to get a reference over nh 200, restore nhid 200 in the group:
 $ ip nexthop replace id 203 group 201/200

And now we have a permanent circular dependncy because nhid 203 holds a
reference over nh 200 and 201, but the route holds a ref over nh 203 and
is deleted.

To trigger the bug just delete the group (nhid 203):
 $ ip nexthop del id 203

It won't really be deleted due to the IPv6 route dependency, and now we
have 2 unlinked and deleted objects that reference each other: the group
and the IPv6 route. Since the group drops the reference it holds over its
entries at free time (i.e. its own refcount needs to drop to 0) that will
never happen and we get a permanent ref on them, since one of the entries
holds a reference over the IPv6 route it will also never be released.

At this point the dependencies are:
 (deleted, only unlinked) IPv6 route holds reference over group nh 203
 (deleted, only unlinked) group nh 203 holds reference over nh 201 and 200
 nh 200 holds 1 ref over the net device and the route due to the stale
 rt6_info

This is the last point where it can be fixed by running traffic through
nh 200, and specifically through the same CPU so the rt6_info (dst) will
get released due to the IPv6 genid, that in turn will free the IPv6
route, which in turn will free the ref count over the group nh 203.

If nh 200 is deleted at this point, it will never be released due to the
ref from the unlinked group 203, it will only be unlinked:
 $ ip nexthop del id 200
 $ ip nexthop
 $

Now we can never release that stale rt6_info, we have IPv6 route with ref
over group nh 203, group nh 203 with ref over nh 200 and 201, nh 200 with
rt6_info (dst) with ref over the net device and the IPv6 route. All of
these objects are only unlinked, and cannot be released, thus they can't
release their ref counts.

 Message from syslogd@dev at Nov 19 14:04:10 ...
  kernel:[73501.828730] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3
 Message from syslogd@dev at Nov 19 14:04:20 ...
  kernel:[73512.068811] unregister_netdevice: waiting for bridge.10 to become free. Usage count = 3

Fixes: 7bf4796dd099 ("nexthops: add support for replace")
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: ipv6: add fib6_nh_release_dsts stub
Nikolay Aleksandrov [Mon, 22 Nov 2021 15:15:12 +0000 (17:15 +0200)]
net: ipv6: add fib6_nh_release_dsts stub

[ Upstream commit 8837cbbf854246f5f4d565f21e6baa945d37aded ]

We need a way to release a fib6_nh's per-cpu dsts when replacing
nexthops otherwise we can end up with stale per-cpu dsts which hold net
device references, so add a new IPv6 stub called fib6_nh_release_dsts.
It must be used after an RCU grace period, so no new dsts can be created
through a group's nexthop entry.
Similar to fib6_nh_release it shouldn't be used if fib6_nh_init has failed
so it doesn't need a dummy stub when IPv6 is not enabled.

Fixes: 7bf4796dd099 ("nexthops: add support for replace")
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: stmmac: retain PTP clock time during SIOCSHWTSTAMP ioctls
Holger Assmann [Sun, 21 Nov 2021 17:57:04 +0000 (19:57 +0200)]
net: stmmac: retain PTP clock time during SIOCSHWTSTAMP ioctls

[ Upstream commit a6da2bbb0005e6b4909472962c9d0af29e75dd06 ]

Currently, when user space emits SIOCSHWTSTAMP ioctl calls such as
enabling/disabling timestamping or changing filter settings, the driver
reads the current CLOCK_REALTIME value and programming this into the
NIC's hardware clock. This might be necessary during system
initialization, but at runtime, when the PTP clock has already been
synchronized to a grandmaster, a reset of the timestamp settings might
result in a clock jump. Furthermore, if the clock is also controlled by
phc2sys in automatic mode (where the UTC offset is queried from ptp4l),
that UTC-to-TAI offset (currently 37 seconds in 2021) would be
temporarily reset to 0, and it would take a long time for phc2sys to
readjust so that CLOCK_REALTIME and the PHC are apart by 37 seconds
again.

To address the issue, we introduce a new function called
stmmac_init_tstamp_counter(), which gets called during ndo_open().
It contains the code snippet moved from stmmac_hwtstamp_set() that
manages the time synchronization. Besides, the sub second increment
configuration is also moved here since the related values are hardware
dependent and runtime invariant.

Furthermore, the hardware clock must be kept running even when no time
stamping mode is selected in order to retain the synchronized time base.
That way, timestamping can be enabled again at any time only with the
need to compensate the clock's natural drifting.

As a side effect, this patch fixes the issue that ptp_clock_info::enable
can be called before SIOCSHWTSTAMP and the driver (which looks at
priv->systime_flags) was not prepared to handle that ordering.

Fixes: 92ba6888510c ("stmmac: add the support for PTP hw clock driver")
Reported-by: Michael Olbrich <m.olbrich@pengutronix.de>
Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
Signed-off-by: Holger Assmann <h.assmann@pengutronix.de>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: stmmac: fix system hang caused by eee_ctrl_timer during suspend/resume
Joakim Zhang [Wed, 8 Sep 2021 07:43:35 +0000 (15:43 +0800)]
net: stmmac: fix system hang caused by eee_ctrl_timer during suspend/resume

[ Upstream commit 276aae377206d60b9b7b7df4586cd9f2a813f5d0 ]

commit 5f58591323bf ("net: stmmac: delete the eee_ctrl_timer after
napi disabled"), this patch tries to fix system hang caused by eee_ctrl_timer,
unfortunately, it only can resolve it for system reboot stress test. System
hang also can be reproduced easily during system suspend/resume stess test
when mount NFS on i.MX8MP EVK board.

In stmmac driver, eee feature is combined to phylink framework. When do
system suspend, phylink_stop() would queue delayed work, it invokes
stmmac_mac_link_down(), where to deactivate eee_ctrl_timer synchronizly.
In above commit, try to fix issue by deactivating eee_ctrl_timer obviously,
but it is not enough. Looking into eee_ctrl_timer expire callback
stmmac_eee_ctrl_timer(), it could enable hareware eee mode again. What is
unexpected is that LPI interrupt (MAC_Interrupt_Enable.LPIEN bit) is always
asserted. This interrupt has chance to be issued when LPI state entry/exit
from the MAC, and at that time, clock could have been already disabled.
The result is that system hang when driver try to touch register from
interrupt handler.

The reason why above commit can fix system hang issue in stmmac_release()
is that, deactivate eee_ctrl_timer not just after napi disabled, further
after irq freed.

In conclusion, hardware would generate LPI interrupt when clock has been
disabled during suspend or resume, since hardware is in eee mode and LPI
interrupt enabled.

Interrupts from MAC, MTL and DMA level are enabled and never been disabled
when system suspend, so postpone clocks management from suspend stage to
noirq suspend stage should be more safe.

Fixes: 5f58591323bf ("net: stmmac: delete the eee_ctrl_timer after napi disabled")
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonfp: checking parameter process for rx-usecs/tx-usecs is invalid
Diana Wang [Fri, 19 Nov 2021 13:38:03 +0000 (14:38 +0100)]
nfp: checking parameter process for rx-usecs/tx-usecs is invalid

[ Upstream commit 3bd6b2a838ba6a3b86d41b077f570b1b61174def ]

Use nn->tlv_caps.me_freq_mhz instead of nn->me_freq_mhz to check whether
rx-usecs/tx-usecs is valid.

This is because nn->tlv_caps.me_freq_mhz represents the clock_freq (MHz) of
the flow processing cores (FPC) on the NIC. While nn->me_freq_mhz is not
be set.

Fixes: ce991ab6662a ("nfp: read ME frequency from vNIC ctrl memory")
Signed-off-by: Diana Wang <na.wang@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoipv6: fix typos in __ip6_finish_output()
Eric Dumazet [Fri, 19 Nov 2021 01:37:58 +0000 (17:37 -0800)]
ipv6: fix typos in __ip6_finish_output()

[ Upstream commit 19d36c5f294879949c9d6f57cb61d39cc4c48553 ]

We deal with IPv6 packets, so we need to use IP6CB(skb)->flags and
IP6SKB_REROUTED, instead of IPCB(skb)->flags and IPSKB_REROUTED

Found by code inspection, please double check that fixing this bug
does not surface other bugs.

Fixes: 09ee9dba9611 ("ipv6: Reinject IPv6 packets if IPsec policy matches after SNAT")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tobias Brunner <tobias@strongswan.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: David Ahern <dsahern@kernel.org>
Reviewed-by: David Ahern <dsahern@kernel.org>
Tested-by: Tobias Brunner <tobias@strongswan.org>
Acked-by: Tobias Brunner <tobias@strongswan.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agofirmware: smccc: Fix check for ARCH_SOC_ID not implemented
Michael Kelley [Fri, 19 Nov 2021 23:39:01 +0000 (15:39 -0800)]
firmware: smccc: Fix check for ARCH_SOC_ID not implemented

[ Upstream commit e95d8eaee21cd0d117d34125d4cdc97489c1ab82 ]

The ARCH_FEATURES function ID is a 32-bit SMC call, which returns
a 32-bit result per the SMCCC spec.  Current code is doing a 64-bit
comparison against -1 (SMCCC_RET_NOT_SUPPORTED) to detect that the
feature is unimplemented.  That check doesn't work in a Hyper-V VM,
where the upper 32-bits are zero as allowed by the spec.

Cast the result as an 'int' so the comparison works. The change also
makes the code consistent with other similar checks in this file.

Fixes: 821b67fa4639 ("firmware: smccc: Add ARCH_SOC_ID support")
Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agomptcp: fix delack timer
Eric Dumazet [Fri, 19 Nov 2021 14:27:54 +0000 (15:27 +0100)]
mptcp: fix delack timer

[ Upstream commit ee50e67ba0e17b1a1a8d76691d02eadf9e0f392c ]

To compute the rtx timeout schedule_3rdack_retransmission() does multiple
things in the wrong way: srtt_us is measured in usec/8 and the timeout
itself is an absolute value.

Fixes: ec3edaa7ca6ce02f ("mptcp: Add handling of outgoing MP_JOIN requests")
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <mathew.j.martineau>@linux.intel.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoALSA: intel-dsp-config: add quirk for JSL devices based on ES8336 codec
Pierre-Louis Bossart [Wed, 27 Oct 2021 02:32:54 +0000 (10:32 +0800)]
ALSA: intel-dsp-config: add quirk for JSL devices based on ES8336 codec

[ Upstream commit fa9730b4f28b7bd183d28a0bf636ab7108de35d7 ]

These devices are based on an I2C/I2S device, we need to force the use
of the SOF driver otherwise the legacy HDaudio driver will be loaded -
only HDMI will be supported.

We previously added support for other Intel platforms but missed
JasperLake.

BugLink: https://github.com/thesofproject/linux/issues/3210
Fixes: 9d36ceab9415 ('ALSA: intel-dsp-config: add quirk for APL/GLK/TGL devices based on ES8336 codec')
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Kai Vehmanen <kai.vehmanen@intel.com>
Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://lore.kernel.org/r/20211027023254.24955-1-yung-chuan.liao@linux.intel.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoiavf: Prevent changing static ITR values if adaptive moderation is on
Nitesh B Venkatesh [Fri, 4 Jun 2021 16:53:31 +0000 (09:53 -0700)]
iavf: Prevent changing static ITR values if adaptive moderation is on

[ Upstream commit e792779e6b639c182df91b46ac1e5803460b0b15 ]

Resolve being able to change static values on VF when adaptive interrupt
moderation is enabled.

This problem is fixed by checking the interrupt settings is not
a combination of change of static value while adaptive interrupt
moderation is turned on.

Without this fix, the user would be able to change static values
on VF with adaptive moderation enabled.

Fixes: 65e87c0398f5 ("i40evf: support queue-specific settings for interrupt moderation")
Signed-off-by: Nitesh B Venkatesh <nitesh.b.venkatesh@intel.com>
Tested-by: George Kuruvinakunnel <george.kuruvinakunnel@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agonet: marvell: prestera: fix double free issue on err path
Volodymyr Mytnyk [Thu, 18 Nov 2021 19:51:40 +0000 (21:51 +0200)]
net: marvell: prestera: fix double free issue on err path

[ Upstream commit e8d032507cb7912baf1d3e0af54516f823befefd ]

fix error path handling in prestera_bridge_port_join() that
cases prestera driver to crash (see below).

 Trace:
   Internal error: Oops: 96000044 [#1] SMP
   Modules linked in: prestera_pci prestera uio_pdrv_genirq
   CPU: 1 PID: 881 Comm: ip Not tainted 5.15.0 #1
   pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
   pc : prestera_bridge_destroy+0x2c/0xb0 [prestera]
   lr : prestera_bridge_port_join+0x2cc/0x350 [prestera]
   sp : ffff800011a1b0f0
   ...
   x2 : ffff000109ca6c80 x1 : dead000000000100 x0 : dead000000000122
    Call trace:
   prestera_bridge_destroy+0x2c/0xb0 [prestera]
   prestera_bridge_port_join+0x2cc/0x350 [prestera]
   prestera_netdev_port_event.constprop.0+0x3c4/0x450 [prestera]
   prestera_netdev_event_handler+0xf4/0x110 [prestera]
   raw_notifier_call_chain+0x54/0x80
   call_netdevice_notifiers_info+0x54/0xa0
   __netdev_upper_dev_link+0x19c/0x380

Fixes: e1189d9a5fbe ("net: marvell: prestera: Add Switchdev driver implementation")
Signed-off-by: Volodymyr Mytnyk <vmytnyk@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agodrm/vc4: fix error code in vc4_create_object()
Dan Carpenter [Thu, 18 Nov 2021 11:14:16 +0000 (14:14 +0300)]
drm/vc4: fix error code in vc4_create_object()

[ Upstream commit 96c5f82ef0a145d3e56e5b26f2bf6dcd2ffeae1c ]

The ->gem_create_object() functions are supposed to return NULL if there
is an error.  None of the callers expect error pointers so returing one
will lead to an Oops.  See drm_gem_vram_create(), for example.

Fixes: c826a6e10644 ("drm/vc4: Add a BO cache.")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
Link: https://patchwork.freedesktop.org/patch/msgid/20211118111416.GC1147@kili
Signed-off-by: Sasha Levin <sashal@kernel.org>
2 years agoscsi: mpt3sas: Fix kernel panic during drive powercycle test
Sreekanth Reddy [Wed, 17 Nov 2021 10:49:09 +0000 (16:19 +0530)]
scsi: mpt3sas: Fix kernel panic during drive powercycle test

[ Upstream commit 0ee4ba13e09c9d9c1cb6abb59da8295d9952328b ]

While looping over shost's sdev list it is possible that one
of the drives is getting removed and its sas_target object is
freed but its sdev object remains intact.

Consequently, a kernel panic can occur while the driver is trying to access
the sas_address field of sas_target object without also checking the
sas_target object for NULL.

Link: https://lore.kernel.org/r/20211117104909.2069-1-sreekanth.reddy@broadcom.com
Fixes: f92363d12359 ("[SCSI] mpt3sas: add new driver supporting 12GB SAS")
Signed-off-by: Sreekanth Reddy <sreekanth.reddy@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>