wrapfs-4.13.y.git
6 years agokprobes: Mark ftrace mcount handler functions nokprobe
Masami Hiramatsu [Sat, 23 Feb 2019 16:50:20 +0000 (01:50 +0900)]
kprobes: Mark ftrace mcount handler functions nokprobe

commit fabe38ab6b2bd9418350284c63825f13b8a6abba upstream.

Mark ftrace mcount handler functions nokprobe since
probing on these functions with kretprobe pushes
return address incorrectly on kretprobe shadow stack.

Reported-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Tested-by: Andrea Righi <righi.andrea@gmail.com>
Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/155094062044.6137.6419622920568680640.stgit@devbox
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agox86/kprobes: Verify stack frame on kretprobe
Masami Hiramatsu [Sat, 23 Feb 2019 16:49:52 +0000 (01:49 +0900)]
x86/kprobes: Verify stack frame on kretprobe

commit 3ff9c075cc767b3060bdac12da72fc94dd7da1b8 upstream.

Verify the stack frame pointer on kretprobe trampoline handler,
If the stack frame pointer does not match, it skips the wrong
entry and tries to find correct one.

This can happen if user puts the kretprobe on the function
which can be used in the path of ftrace user-function call.
Such functions should not be probed, so this adds a warning
message that reports which function should be blacklisted.

Tested-by: Andrea Righi <righi.andrea@gmail.com>
Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/155094059185.6137.15527904013362842072.stgit@devbox
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoarm64: futex: Restore oldval initialization to work around buggy compilers
Nathan Chancellor [Wed, 17 Apr 2019 07:21:21 +0000 (00:21 -0700)]
arm64: futex: Restore oldval initialization to work around buggy compilers

commit ff8acf929014b7f87315588e0daf8597c8aa9d1c upstream.

Commit 045afc24124d ("arm64: futex: Fix FUTEX_WAKE_OP atomic ops with
non-zero result value") removed oldval's zero initialization in
arch_futex_atomic_op_inuser because it is not necessary. Unfortunately,
Android's arm64 GCC 4.9.4 [1] does not agree:

../kernel/futex.c: In function 'do_futex':
../kernel/futex.c:1658:17: warning: 'oldval' may be used uninitialized
in this function [-Wmaybe-uninitialized]
   return oldval == cmparg;
                 ^
In file included from ../kernel/futex.c:73:0:
../arch/arm64/include/asm/futex.h:53:6: note: 'oldval' was declared here
  int oldval, ret, tmp;
      ^

GCC fails to follow that when ret is non-zero, futex_atomic_op_inuser
returns right away, avoiding the uninitialized use that it claims.
Restoring the zero initialization works around this issue.

[1]: https://android.googlesource.com/platform/prebuilts/gcc/linux-x86/aarch64/aarch64-linux-android-4.9/

Cc: stable@vger.kernel.org
Fixes: 045afc24124d ("arm64: futex: Fix FUTEX_WAKE_OP atomic ops with non-zero result value")
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agocrypto: x86/poly1305 - fix overflow during partial reduction
Eric Biggers [Sun, 31 Mar 2019 20:04:11 +0000 (13:04 -0700)]
crypto: x86/poly1305 - fix overflow during partial reduction

commit 678cce4019d746da6c680c48ba9e6d417803e127 upstream.

The x86_64 implementation of Poly1305 produces the wrong result on some
inputs because poly1305_4block_avx2() incorrectly assumes that when
partially reducing the accumulator, the bits carried from limb 'd4' to
limb 'h0' fit in a 32-bit integer.  This is true for poly1305-generic
which processes only one block at a time.  However, it's not true for
the AVX2 implementation, which processes 4 blocks at a time and
therefore can produce intermediate limbs about 4x larger.

Fix it by making the relevant calculations use 64-bit arithmetic rather
than 32-bit.  Note that most of the carries already used 64-bit
arithmetic, but the d4 -> h0 carry was different for some reason.

To be safe I also made the same change to the corresponding SSE2 code,
though that only operates on 1 or 2 blocks at a time.  I don't think
it's really needed for poly1305_block_sse2(), but it doesn't hurt
because it's already x86_64 code.  It *might* be needed for
poly1305_2block_sse2(), but overflows aren't easy to reproduce there.

This bug was originally detected by my patches that improve testmgr to
fuzz algorithms against their generic implementation.  But also add a
test vector which reproduces it directly (in the AVX2 case).

Fixes: b1ccc8f4b631 ("crypto: poly1305 - Add a four block AVX2 variant for x86_64")
Fixes: c70f4abef07a ("crypto: poly1305 - Add a SSE2 SIMD variant for x86_64")
Cc: <stable@vger.kernel.org> # v4.3+
Cc: Martin Willi <martin@strongswan.org>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Martin Willi <martin@strongswan.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agocoredump: fix race condition between mmget_not_zero()/get_task_mm() and core dumping
Andrea Arcangeli [Fri, 19 Apr 2019 00:50:52 +0000 (17:50 -0700)]
coredump: fix race condition between mmget_not_zero()/get_task_mm() and core dumping

commit 04f5866e41fb70690e28397487d8bd8eea7d712a upstream.

The core dumping code has always run without holding the mmap_sem for
writing, despite that is the only way to ensure that the entire vma
layout will not change from under it.  Only using some signal
serialization on the processes belonging to the mm is not nearly enough.
This was pointed out earlier.  For example in Hugh's post from Jul 2017:

  https://lkml.kernel.org/r/alpine.LSU.2.11.1707191716030.2055@eggly.anvils

  "Not strictly relevant here, but a related note: I was very surprised
   to discover, only quite recently, how handle_mm_fault() may be called
   without down_read(mmap_sem) - when core dumping. That seems a
   misguided optimization to me, which would also be nice to correct"

In particular because the growsdown and growsup can move the
vm_start/vm_end the various loops the core dump does around the vma will
not be consistent if page faults can happen concurrently.

Pretty much all users calling mmget_not_zero()/get_task_mm() and then
taking the mmap_sem had the potential to introduce unexpected side
effects in the core dumping code.

Adding mmap_sem for writing around the ->core_dump invocation is a
viable long term fix, but it requires removing all copy user and page
faults and to replace them with get_dump_page() for all binary formats
which is not suitable as a short term fix.

For the time being this solution manually covers the places that can
confuse the core dump either by altering the vma layout or the vma flags
while it runs.  Once ->core_dump runs under mmap_sem for writing the
function mmget_still_valid() can be dropped.

Allowing mmap_sem protected sections to run in parallel with the
coredump provides some minor parallelism advantage to the swapoff code
(which seems to be safe enough by never mangling any vma field and can
keep doing swapins in parallel to the core dumping) and to some other
corner case.

In order to facilitate the backporting I added "Fixes: 86039bd3b4e6"
however the side effect of this same race condition in /proc/pid/mem
should be reproducible since before 2.6.12-rc2 so I couldn't add any
other "Fixes:" because there's no hash beyond the git genesis commit.

Because find_extend_vma() is the only location outside of the process
context that could modify the "mm" structures under mmap_sem for
reading, by adding the mmget_still_valid() check to it, all other cases
that take the mmap_sem for reading don't need the new check after
mmget_not_zero()/get_task_mm().  The expand_stack() in page fault
context also doesn't need the new check, because all tasks under core
dumping are frozen.

Link: http://lkml.kernel.org/r/20190325224949.11068-1-aarcange@redhat.com
Fixes: 86039bd3b4e6 ("userfaultfd: add new syscall to provide memory externalization")
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reported-by: Jann Horn <jannh@google.com>
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jann Horn <jannh@google.com>
Acked-by: Jason Gunthorpe <jgg@mellanox.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoRevert "svm: Fix AVIC incomplete IPI emulation"
Suthikulpanit, Suravee [Wed, 20 Mar 2019 08:12:28 +0000 (08:12 +0000)]
Revert "svm: Fix AVIC incomplete IPI emulation"

commit 4a58038b9e420276157785afa0a0bbb4b9bc2265 upstream.

This reverts commit bb218fbcfaaa3b115d4cd7a43c0ca164f3a96e57.

As Oren Twaig pointed out the old discussion:

  https://patchwork.kernel.org/patch/8292231/

that the change coud potentially cause an extra IPI to be sent to
the destination vcpu because the AVIC hardware already set the IRR bit
before the incomplete IPI #VMEXIT with id=1 (target vcpu is not running).
Since writting to ICR and ICR2 will also set the IRR. If something triggers
the destination vcpu to get scheduled before the emulation finishes, then
this could result in an additional IPI.

Also, the issue mentioned in the commit bb218fbcfaaa was misdiagnosed.

Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Reported-by: Oren Twaig <oren@scalemp.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoRevert "scsi: fcoe: clear FC_RP_STARTED flags when receiving a LOGO"
Saurav Kashyap [Thu, 18 Apr 2019 10:40:12 +0000 (03:40 -0700)]
Revert "scsi: fcoe: clear FC_RP_STARTED flags when receiving a LOGO"

commit 0228034d8e5915b98c33db35a98f5e909e848ae9 upstream.

This patch clears FC_RP_STARTED flag during logoff, because of this
re-login(flogi) didn't happen to the switch.

This reverts commit 1550ec458e0cf1a40a170ab1f4c46e3f52860f65.

Fixes: 1550ec458e0c ("scsi: fcoe: clear FC_RP_STARTED flags when receiving a LOGO")
Cc: <stable@vger.kernel.org> # v4.18+
Signed-off-by: Saurav Kashyap <skashyap@marvell.com>
Reviewed-by: Hannes Reinecke <hare@#suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoscsi: core: set result when the command cannot be dispatched
Jaesoo Lee [Wed, 10 Apr 2019 00:02:22 +0000 (17:02 -0700)]
scsi: core: set result when the command cannot be dispatched

commit be549d49115422f846b6d96ee8fd7173a5f7ceb0 upstream.

When SCSI blk-mq is enabled, there is a bug in handling errors in
scsi_queue_rq.  Specifically, the bug is not setting result field of
scsi_request correctly when the dispatch of the command has been
failed. Since the upper layer code including the sg_io ioctl expects to
receive any error status from result field of scsi_request, the error is
silently ignored and this could cause data corruptions for some
applications.

Fixes: d285203cf647 ("scsi: add support for a blk-mq based I/O path.")
Cc: <stable@vger.kernel.org>
Signed-off-by: Jaesoo Lee <jalee@purestorage.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoALSA: core: Fix card races between register and disconnect
Takashi Iwai [Tue, 16 Apr 2019 15:06:33 +0000 (17:06 +0200)]
ALSA: core: Fix card races between register and disconnect

commit 2a3f7221acddfe1caa9ff09b3a8158c39b2fdeac upstream.

There is a small race window in the card disconnection code that
allows the registration of another card with the very same card id.
This leads to a warning in procfs creation as caught by syzkaller.

The problem is that we delete snd_cards and snd_cards_lock entries at
the very beginning of the disconnection procedure.  This makes the
slot available to be assigned for another card object while the
disconnection procedure is being processed.  Then it becomes possible
to issue a procfs registration with the existing file name although we
check the conflict beforehand.

The fix is simply to move the snd_cards and snd_cards_lock clearances
at the end of the disconnection procedure.  The references to these
entries are merely either from the global proc files like
/proc/asound/cards or from the card registration / disconnection, so
it should be fine to shift at the very end.

Reported-by: syzbot+48df349490c36f9f54ab@syzkaller.appspotmail.com
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoALSA: hda/realtek - add two more pin configuration sets to quirk table
Hui Wang [Wed, 17 Apr 2019 08:10:32 +0000 (16:10 +0800)]
ALSA: hda/realtek - add two more pin configuration sets to quirk table

commit b26e36b7ef36a8a3a147b1609b2505f8a4ecf511 upstream.

We have two Dell laptops which have the codec 10ec0236 and 10ec0256
respectively, the headset mic on them can't work, need to apply the
quirk of ALC255_FIXUP_DELL1_MIC_NO_PRESENCE. So adding their pin
configurations in the pin quirk table.

Cc: <stable@vger.kernel.org>
Signed-off-by: Hui Wang <hui.wang@canonical.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agostaging: comedi: ni_usb6501: Fix possible double-free of ->usb_rx_buf
Ian Abbott [Mon, 15 Apr 2019 11:43:02 +0000 (12:43 +0100)]
staging: comedi: ni_usb6501: Fix possible double-free of ->usb_rx_buf

commit af4b54a2e5ba18259ff9aac445bf546dd60d037e upstream.

`ni6501_alloc_usb_buffers()` is called from `ni6501_auto_attach()` to
allocate RX and TX buffers for USB transfers.  It allocates
`devpriv->usb_rx_buf` followed by `devpriv->usb_tx_buf`.  If the
allocation of `devpriv->usb_tx_buf` fails, it frees
`devpriv->usb_rx_buf`, leaving the pointer set dangling, and returns an
error.  Later, `ni6501_detach()` will be called from the core comedi
module code to clean up.  `ni6501_detach()` also frees both
`devpriv->usb_rx_buf` and `devpriv->usb_tx_buf`, but
`devpriv->usb_rx_buf` may have already beed freed, leading to a
double-free error.  Fix it bu removing the call to
`kfree(devpriv->usb_rx_buf)` from `ni6501_alloc_usb_buffers()`, relying
on `ni6501_detach()` to free the memory.

Signed-off-by: Ian Abbott <abbotti@mev.co.uk>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agostaging: comedi: ni_usb6501: Fix use of uninitialized mutex
Ian Abbott [Mon, 15 Apr 2019 11:43:01 +0000 (12:43 +0100)]
staging: comedi: ni_usb6501: Fix use of uninitialized mutex

commit 660cf4ce9d0f3497cc7456eaa6d74c8b71d6282c upstream.

If `ni6501_auto_attach()` returns an error, the core comedi module code
will call `ni6501_detach()` to clean up.  If `ni6501_auto_attach()`
successfully allocated the comedi device private data, `ni6501_detach()`
assumes that a `struct mutex mut` contained in the private data has been
initialized and uses it.  Unfortunately, there are a couple of places
where `ni6501_auto_attach()` can return an error after allocating the
device private data but before initializing the mutex, so this
assumption is invalid.  Fix it by initializing the mutex just after
allocating the private data in `ni6501_auto_attach()` before any other
errors can be retturned.  Also move the call to `usb_set_intfdata()`
just to keep the code a bit neater (either position for the call is
fine).

I believe this was the cause of the following syzbot crash report
<https://syzkaller.appspot.com/bug?extid=cf4f2b6c24aff0a3edf6>:

usb 1-1: New USB device strings: Mfr=0, Product=0, SerialNumber=0
usb 1-1: config 0 descriptor??
usb 1-1: string descriptor 0 read error: -71
comedi comedi0: Wrong number of endpoints
ni6501 1-1:0.233: driver 'ni6501' failed to auto-configure device.
INFO: trying to register non-static key.
the code is fine but needs lockdep annotation.
turning off the locking correctness validator.
CPU: 0 PID: 585 Comm: kworker/0:3 Not tainted 5.1.0-rc4-319354-g9a33b36 #3
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Workqueue: usb_hub_wq hub_event
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0xe8/0x16e lib/dump_stack.c:113
 assign_lock_key kernel/locking/lockdep.c:786 [inline]
 register_lock_class+0x11b8/0x1250 kernel/locking/lockdep.c:1095
 __lock_acquire+0xfb/0x37c0 kernel/locking/lockdep.c:3582
 lock_acquire+0x10d/0x2f0 kernel/locking/lockdep.c:4211
 __mutex_lock_common kernel/locking/mutex.c:925 [inline]
 __mutex_lock+0xfe/0x12b0 kernel/locking/mutex.c:1072
 ni6501_detach+0x5b/0x110 drivers/staging/comedi/drivers/ni_usb6501.c:567
 comedi_device_detach+0xed/0x800 drivers/staging/comedi/drivers.c:204
 comedi_device_cleanup.part.0+0x68/0x140 drivers/staging/comedi/comedi_fops.c:156
 comedi_device_cleanup drivers/staging/comedi/comedi_fops.c:187 [inline]
 comedi_free_board_dev.part.0+0x16/0x90 drivers/staging/comedi/comedi_fops.c:190
 comedi_free_board_dev drivers/staging/comedi/comedi_fops.c:189 [inline]
 comedi_release_hardware_device+0x111/0x140 drivers/staging/comedi/comedi_fops.c:2880
 comedi_auto_config.cold+0x124/0x1b0 drivers/staging/comedi/drivers.c:1068
 usb_probe_interface+0x31d/0x820 drivers/usb/core/driver.c:361
 really_probe+0x2da/0xb10 drivers/base/dd.c:509
 driver_probe_device+0x21d/0x350 drivers/base/dd.c:671
 __device_attach_driver+0x1d8/0x290 drivers/base/dd.c:778
 bus_for_each_drv+0x163/0x1e0 drivers/base/bus.c:454
 __device_attach+0x223/0x3a0 drivers/base/dd.c:844
 bus_probe_device+0x1f1/0x2a0 drivers/base/bus.c:514
 device_add+0xad2/0x16e0 drivers/base/core.c:2106
 usb_set_configuration+0xdf7/0x1740 drivers/usb/core/message.c:2021
 generic_probe+0xa2/0xda drivers/usb/core/generic.c:210
 usb_probe_device+0xc0/0x150 drivers/usb/core/driver.c:266
 really_probe+0x2da/0xb10 drivers/base/dd.c:509
 driver_probe_device+0x21d/0x350 drivers/base/dd.c:671
 __device_attach_driver+0x1d8/0x290 drivers/base/dd.c:778
 bus_for_each_drv+0x163/0x1e0 drivers/base/bus.c:454
 __device_attach+0x223/0x3a0 drivers/base/dd.c:844
 bus_probe_device+0x1f1/0x2a0 drivers/base/bus.c:514
 device_add+0xad2/0x16e0 drivers/base/core.c:2106
 usb_new_device.cold+0x537/0xccf drivers/usb/core/hub.c:2534
 hub_port_connect drivers/usb/core/hub.c:5089 [inline]
 hub_port_connect_change drivers/usb/core/hub.c:5204 [inline]
 port_event drivers/usb/core/hub.c:5350 [inline]
 hub_event+0x138e/0x3b00 drivers/usb/core/hub.c:5432
 process_one_work+0x90f/0x1580 kernel/workqueue.c:2269
 worker_thread+0x9b/0xe20 kernel/workqueue.c:2415
 kthread+0x313/0x420 kernel/kthread.c:253
 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:352

Reported-by: syzbot+cf4f2b6c24aff0a3edf6@syzkaller.appspotmail.com
Signed-off-by: Ian Abbott <abbotti@mev.co.uk>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agostaging: comedi: vmk80xx: Fix possible double-free of ->usb_rx_buf
Ian Abbott [Mon, 15 Apr 2019 11:52:30 +0000 (12:52 +0100)]
staging: comedi: vmk80xx: Fix possible double-free of ->usb_rx_buf

commit 663d294b4768bfd89e529e069bffa544a830b5bf upstream.

`vmk80xx_alloc_usb_buffers()` is called from `vmk80xx_auto_attach()` to
allocate RX and TX buffers for USB transfers.  It allocates
`devpriv->usb_rx_buf` followed by `devpriv->usb_tx_buf`.  If the
allocation of `devpriv->usb_tx_buf` fails, it frees
`devpriv->usb_rx_buf`,  leaving the pointer set dangling, and returns an
error.  Later, `vmk80xx_detach()` will be called from the core comedi
module code to clean up.  `vmk80xx_detach()` also frees both
`devpriv->usb_rx_buf` and `devpriv->usb_tx_buf`, but
`devpriv->usb_rx_buf` may have already been freed, leading to a
double-free error.  Fix it by removing the call to
`kfree(devpriv->usb_rx_buf)` from `vmk80xx_alloc_usb_buffers()`, relying
on `vmk80xx_detach()` to free the memory.

Signed-off-by: Ian Abbott <abbotti@mev.co.uk>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agostaging: comedi: vmk80xx: Fix use of uninitialized semaphore
Ian Abbott [Mon, 15 Apr 2019 11:10:14 +0000 (12:10 +0100)]
staging: comedi: vmk80xx: Fix use of uninitialized semaphore

commit 08b7c2f9208f0e2a32159e4e7a4831b7adb10a3e upstream.

If `vmk80xx_auto_attach()` returns an error, the core comedi module code
will call `vmk80xx_detach()` to clean up.  If `vmk80xx_auto_attach()`
successfully allocated the comedi device private data,
`vmk80xx_detach()` assumes that a `struct semaphore limit_sem` contained
in the private data has been initialized and uses it.  Unfortunately,
there are a couple of places where `vmk80xx_auto_attach()` can return an
error after allocating the device private data but before initializing
the semaphore, so this assumption is invalid.  Fix it by initializing
the semaphore just after allocating the private data in
`vmk80xx_auto_attach()` before any other errors can be returned.

I believe this was the cause of the following syzbot crash report
<https://syzkaller.appspot.com/bug?extid=54c2f58f15fe6876b6ad>:

usb 1-1: config 0 has no interface number 0
usb 1-1: New USB device found, idVendor=10cf, idProduct=8068, bcdDevice=e6.8d
usb 1-1: New USB device strings: Mfr=0, Product=0, SerialNumber=0
usb 1-1: config 0 descriptor??
vmk80xx 1-1:0.117: driver 'vmk80xx' failed to auto-configure device.
INFO: trying to register non-static key.
the code is fine but needs lockdep annotation.
turning off the locking correctness validator.
CPU: 0 PID: 12 Comm: kworker/0:1 Not tainted 5.1.0-rc4-319354-g9a33b36 #3
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Workqueue: usb_hub_wq hub_event
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0xe8/0x16e lib/dump_stack.c:113
 assign_lock_key kernel/locking/lockdep.c:786 [inline]
 register_lock_class+0x11b8/0x1250 kernel/locking/lockdep.c:1095
 __lock_acquire+0xfb/0x37c0 kernel/locking/lockdep.c:3582
 lock_acquire+0x10d/0x2f0 kernel/locking/lockdep.c:4211
 __raw_spin_lock_irqsave include/linux/spinlock_api_smp.h:110 [inline]
 _raw_spin_lock_irqsave+0x44/0x60 kernel/locking/spinlock.c:152
 down+0x12/0x80 kernel/locking/semaphore.c:58
 vmk80xx_detach+0x59/0x100 drivers/staging/comedi/drivers/vmk80xx.c:829
 comedi_device_detach+0xed/0x800 drivers/staging/comedi/drivers.c:204
 comedi_device_cleanup.part.0+0x68/0x140 drivers/staging/comedi/comedi_fops.c:156
 comedi_device_cleanup drivers/staging/comedi/comedi_fops.c:187 [inline]
 comedi_free_board_dev.part.0+0x16/0x90 drivers/staging/comedi/comedi_fops.c:190
 comedi_free_board_dev drivers/staging/comedi/comedi_fops.c:189 [inline]
 comedi_release_hardware_device+0x111/0x140 drivers/staging/comedi/comedi_fops.c:2880
 comedi_auto_config.cold+0x124/0x1b0 drivers/staging/comedi/drivers.c:1068
 usb_probe_interface+0x31d/0x820 drivers/usb/core/driver.c:361
 really_probe+0x2da/0xb10 drivers/base/dd.c:509
 driver_probe_device+0x21d/0x350 drivers/base/dd.c:671
 __device_attach_driver+0x1d8/0x290 drivers/base/dd.c:778
 bus_for_each_drv+0x163/0x1e0 drivers/base/bus.c:454
 __device_attach+0x223/0x3a0 drivers/base/dd.c:844
 bus_probe_device+0x1f1/0x2a0 drivers/base/bus.c:514
 device_add+0xad2/0x16e0 drivers/base/core.c:2106
 usb_set_configuration+0xdf7/0x1740 drivers/usb/core/message.c:2021
 generic_probe+0xa2/0xda drivers/usb/core/generic.c:210
 usb_probe_device+0xc0/0x150 drivers/usb/core/driver.c:266
 really_probe+0x2da/0xb10 drivers/base/dd.c:509
 driver_probe_device+0x21d/0x350 drivers/base/dd.c:671
 __device_attach_driver+0x1d8/0x290 drivers/base/dd.c:778
 bus_for_each_drv+0x163/0x1e0 drivers/base/bus.c:454
 __device_attach+0x223/0x3a0 drivers/base/dd.c:844
 bus_probe_device+0x1f1/0x2a0 drivers/base/bus.c:514
 device_add+0xad2/0x16e0 drivers/base/core.c:2106
 usb_new_device.cold+0x537/0xccf drivers/usb/core/hub.c:2534
 hub_port_connect drivers/usb/core/hub.c:5089 [inline]
 hub_port_connect_change drivers/usb/core/hub.c:5204 [inline]
 port_event drivers/usb/core/hub.c:5350 [inline]
 hub_event+0x138e/0x3b00 drivers/usb/core/hub.c:5432
 process_one_work+0x90f/0x1580 kernel/workqueue.c:2269
 worker_thread+0x9b/0xe20 kernel/workqueue.c:2415
 kthread+0x313/0x420 kernel/kthread.c:253
 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:352

Reported-by: syzbot+54c2f58f15fe6876b6ad@syzkaller.appspotmail.com
Signed-off-by: Ian Abbott <abbotti@mev.co.uk>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoio: accel: kxcjk1013: restore the range after resume.
he, bo [Wed, 6 Mar 2019 02:32:20 +0000 (10:32 +0800)]
io: accel: kxcjk1013: restore the range after resume.

commit fe2d3df639a7940a125a33d6460529b9689c5406 upstream.

On some laptops, kxcjk1013 is powered off when system enters S3. We need
restore the range regiter during resume. Otherwise, the sensor doesn't
work properly after S3.

Signed-off-by: he, bo <bo.he@intel.com>
Signed-off-by: Chen, Hu <hu1.chen@intel.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoiio: core: fix a possible circular locking dependency
Fabrice Gasnier [Mon, 25 Mar 2019 13:01:23 +0000 (14:01 +0100)]
iio: core: fix a possible circular locking dependency

commit 7f75591fc5a123929a29636834d1bcb8b5c9fee3 upstream.

This fixes a possible circular locking dependency detected warning seen
with:
- CONFIG_PROVE_LOCKING=y
- consumer/provider IIO devices (ex: "voltage-divider" consumer of "adc")

When using the IIO consumer interface, e.g. iio_channel_get(), the consumer
device will likely call iio_read_channel_raw() or similar that rely on
'info_exist_lock' mutex.

typically:
...
mutex_lock(&chan->indio_dev->info_exist_lock);
if (chan->indio_dev->info == NULL) {
ret = -ENODEV;
goto err_unlock;
}
ret = do_some_ops()
err_unlock:
mutex_unlock(&chan->indio_dev->info_exist_lock);
return ret;
...

Same mutex is also hold in iio_device_unregister().

The following deadlock warning happens when:
- the consumer device has called an API like iio_read_channel_raw()
  at least once.
- the consumer driver is unregistered, removed (unbind from sysfs)

======================================================
WARNING: possible circular locking dependency detected
4.19.24 #577 Not tainted
------------------------------------------------------
sh/372 is trying to acquire lock:
(kn->count#30){++++}, at: kernfs_remove_by_name_ns+0x3c/0x84

but task is already holding lock:
(&dev->info_exist_lock){+.+.}, at: iio_device_unregister+0x18/0x60

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&dev->info_exist_lock){+.+.}:
       __mutex_lock+0x70/0xa3c
       mutex_lock_nested+0x1c/0x24
       iio_read_channel_raw+0x1c/0x60
       iio_read_channel_info+0xa8/0xb0
       dev_attr_show+0x1c/0x48
       sysfs_kf_seq_show+0x84/0xec
       seq_read+0x154/0x528
       __vfs_read+0x2c/0x15c
       vfs_read+0x8c/0x110
       ksys_read+0x4c/0xac
       ret_fast_syscall+0x0/0x28
       0xbedefb60

-> #0 (kn->count#30){++++}:
       lock_acquire+0xd8/0x268
       __kernfs_remove+0x288/0x374
       kernfs_remove_by_name_ns+0x3c/0x84
       remove_files+0x34/0x78
       sysfs_remove_group+0x40/0x9c
       sysfs_remove_groups+0x24/0x34
       device_remove_attrs+0x38/0x64
       device_del+0x11c/0x360
       cdev_device_del+0x14/0x2c
       iio_device_unregister+0x24/0x60
       release_nodes+0x1bc/0x200
       device_release_driver_internal+0x1a0/0x230
       unbind_store+0x80/0x130
       kernfs_fop_write+0x100/0x1e4
       __vfs_write+0x2c/0x160
       vfs_write+0xa4/0x17c
       ksys_write+0x4c/0xac
       ret_fast_syscall+0x0/0x28
       0xbe906840

other info that might help us debug this:

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&dev->info_exist_lock);
                               lock(kn->count#30);
                               lock(&dev->info_exist_lock);
  lock(kn->count#30);

 *** DEADLOCK ***
...

cdev_device_del() can be called without holding the lock. It should be safe
as info_exist_lock prevents kernelspace consumers to use the exported
routines during/after provider removal. cdev_device_del() is for userspace.

Help to reproduce:
See example: Documentation/devicetree/bindings/iio/afe/voltage-divider.txt
sysv {
compatible = "voltage-divider";
io-channels = <&adc 0>;
output-ohms = <22>;
full-ohms = <222>;
};

First, go to iio:deviceX for the "voltage-divider", do one read:
$ cd /sys/bus/iio/devices/iio:deviceX
$ cat in_voltage0_raw

Then, unbind the consumer driver. It triggers above deadlock warning.
$ cd /sys/bus/platform/drivers/iio-rescale/
$ echo sysv > unbind

Note I don't actually expect stable will pick this up all the
way back into IIO being in staging, but if's probably valid that
far back.

Signed-off-by: Fabrice Gasnier <fabrice.gasnier@st.com>
Fixes: ac917a81117c ("staging:iio:core set the iio_dev.info pointer to null on unregister")
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoiio: adc: at91: disable adc channel interrupt in timeout case
Georg Ottinger [Wed, 30 Jan 2019 13:42:02 +0000 (14:42 +0100)]
iio: adc: at91: disable adc channel interrupt in timeout case

commit 09c6bdee51183a575bf7546890c8c137a75a2b44 upstream.

Having a brief look at at91_adc_read_raw() it is obvious that in the case
of a timeout the setting of AT91_ADC_CHDR and AT91_ADC_IDR registers is
omitted. If 2 different channels are queried we can end up with a
situation where two interrupts are enabled, but only one interrupt is
cleared in the interrupt handler. Resulting in a interrupt loop and a
system hang.

Signed-off-by: Georg Ottinger <g.ottinger@abatec.at>
Acked-by: Ludovic Desroches <ludovic.desroches@microchip.com>
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoiio: Fix scan mask selection
Lars-Peter Clausen [Wed, 20 Feb 2019 15:11:32 +0000 (17:11 +0200)]
iio: Fix scan mask selection

commit 20ea39ef9f2f911bd01c69519e7d69cfec79fde3 upstream.

The trialmask is expected to have all bits set to 0 after allocation.
Currently kmalloc_array() is used which does not zero the memory and so
random bits are set. This results in random channels being enabled when
they shouldn't. Replace kmalloc_array() with kcalloc() which has the same
interface but zeros the memory.

Note the fix is actually required earlier than the below fixes tag, but
will require a manual backport due to move from kmalloc to kmalloc_array.

Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Alexandru Ardelean <alexandru.ardelean@analog.com>
Fixes commit 057ac1acdfc4 ("iio: Use kmalloc_array() in iio_scan_mask_set()").
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoiio: dac: mcp4725: add missing powerdown bits in store eeprom
Jean-Francois Dagenais [Wed, 6 Mar 2019 20:56:06 +0000 (15:56 -0500)]
iio: dac: mcp4725: add missing powerdown bits in store eeprom

commit 06003531502d06bc89d32528f6ec96bf978790f9 upstream.

When issuing the write DAC register and write eeprom command, the two
powerdown bits (PD0 and PD1) are assumed by the chip to be present in
the bytes sent. Leaving them at 0 implies "powerdown disabled" which is
a different state that the current one. By adding the current state of
the powerdown in the i2c write, the chip will correctly power-on exactly
like as it is at the moment of store_eeprom call.

This is documented in MCP4725's datasheet, FIGURE 6-2: "Write Commands
for DAC Input Register and EEPROM" and MCP4726's datasheet, FIGURE 6-3:
"Write All Memory Command".

Signed-off-by: Jean-Francois Dagenais <jeff.dagenais@gmail.com>
Acked-by: Peter Meerwald-Stadler <pmeerw@pmeerw.net>
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoiio: ad_sigma_delta: select channel when reading register
Dragos Bogdan [Tue, 19 Mar 2019 10:47:00 +0000 (12:47 +0200)]
iio: ad_sigma_delta: select channel when reading register

commit fccfb9ce70ed4ea7a145f77b86de62e38178517f upstream.

The desired channel has to be selected in order to correctly fill the
buffer with the corresponding data.
The `ad_sd_write_reg()` already does this, but for the
`ad_sd_read_reg_raw()` this was omitted.

Fixes: af3008485ea03 ("iio:adc: Add common code for ADI Sigma Delta devices")
Signed-off-by: Dragos Bogdan <dragos.bogdan@analog.com>
Signed-off-by: Alexandru Ardelean <alexandru.ardelean@analog.com>
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoiio: cros_ec: Fix the maths for gyro scale calculation
Gwendal Grignou [Wed, 13 Mar 2019 11:40:02 +0000 (12:40 +0100)]
iio: cros_ec: Fix the maths for gyro scale calculation

commit 3d02d7082e5823598090530c3988a35f69689943 upstream.

Calculation did not use IIO_DEGREE_TO_RAD and implemented a variant to
avoid precision loss as we aim a nano value. The offset added to avoid
rounding error, though, doesn't give us a close result to the expected
value. E.g.

For 1000dps, the result should be:

    (1000 * pi ) / 180 >> 15 ~= 0.000532632218

But with current calculation we get

    $ cat scale
    0.000547890

Fix the calculation by just doing the maths involved for a nano value

   val * pi * 10e12 / (180 * 2^15)

so we get a closer result.

    $ cat scale
    0.000532632

Fixes: c14dca07a31d ("iio: cros_ec_sensors: add ChromeOS EC Contiguous Sensors driver")
Signed-off-by: Gwendal Grignou <gwendal@chromium.org>
Signed-off-by: Enric Balletbo i Serra <enric.balletbo@collabora.com>
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoiio/gyro/bmg160: Use millidegrees for temperature scale
Mike Looijmans [Wed, 13 Feb 2019 07:41:47 +0000 (08:41 +0100)]
iio/gyro/bmg160: Use millidegrees for temperature scale

commit 40a7198a4a01037003c7ca714f0d048a61e729ac upstream.

Standard unit for temperature is millidegrees Celcius, whereas this driver
was reporting in degrees. Fix the scale factor in the driver.

Signed-off-by: Mike Looijmans <mike.looijmans@topic.nl>
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoiio: gyro: mpu3050: fix chip ID reading
Sergey Larin [Sat, 2 Mar 2019 16:54:55 +0000 (19:54 +0300)]
iio: gyro: mpu3050: fix chip ID reading

commit 409a51e0a4a5f908763191fae2c29008632eb712 upstream.

According to the datasheet, the last bit of CHIP_ID register controls
I2C bus, and the first one is unused. Handle this correctly.

Note that there are chips out there that have a value such that
the id check currently fails.

Signed-off-by: Sergey Larin <cerg2010cerg2010@mail.ru>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agostaging: iio: ad7192: Fix ad7193 channel address
Mircea Caprioru [Wed, 20 Feb 2019 11:08:20 +0000 (13:08 +0200)]
staging: iio: ad7192: Fix ad7193 channel address

commit 7ce0f216221856a17fc4934b39284678a5fef2e9 upstream.

This patch fixes the differential channels addresses for the ad7193.

Signed-off-by: Mircea Caprioru <mircea.caprioru@analog.com>
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoStaging: iio: meter: fixed typo
Leonard Pollak [Wed, 13 Feb 2019 10:19:52 +0000 (11:19 +0100)]
Staging: iio: meter: fixed typo

commit 0a8a29be499cbb67df79370aaf5109085509feb8 upstream.

This patch fixes an obvious typo, which will cause erroneously returning the Peak
Voltage instead of the Peak Current.

Signed-off-by: Leonard Pollak <leonardp@tr-host.de>
Cc: <Stable@vger.kernel.org>
Acked-by: Michael Hennerich <michael.hennerich@analog.com>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoKVM: x86: svm: make sure NMI is injected after nmi_singlestep
Vitaly Kuznetsov [Wed, 3 Apr 2019 14:06:42 +0000 (16:06 +0200)]
KVM: x86: svm: make sure NMI is injected after nmi_singlestep

commit 99c221796a810055974b54c02e8f53297e48d146 upstream.

I noticed that apic test from kvm-unit-tests always hangs on my EPYC 7401P,
the hanging test nmi-after-sti is trying to deliver 30000 NMIs and tracing
shows that we're sometimes able to deliver a few but never all.

When we're trying to inject an NMI we may fail to do so immediately for
various reasons, however, we still need to inject it so enable_nmi_window()
arms nmi_singlestep mode. #DB occurs as expected, but we're not checking
for pending NMIs before entering the guest and unless there's a different
event to process, the NMI will never get delivered.

Make KVM_REQ_EVENT request on the vCPU from db_interception() to make sure
pending NMIs are checked and possibly injected.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoKVM: x86: Don't clear EFER during SMM transitions for 32-bit vCPU
Sean Christopherson [Tue, 2 Apr 2019 15:10:47 +0000 (08:10 -0700)]
KVM: x86: Don't clear EFER during SMM transitions for 32-bit vCPU

commit 8f4dc2e77cdfaf7e644ef29693fa229db29ee1de upstream.

Neither AMD nor Intel CPUs have an EFER field in the legacy SMRAM save
state area, i.e. don't save/restore EFER across SMM transitions.  KVM
somewhat models this, e.g. doesn't clear EFER on entry to SMM if the
guest doesn't support long mode.  But during RSM, KVM unconditionally
clears EFER so that it can get back to pure 32-bit mode in order to
start loading CRs with their actual non-SMM values.

Clear EFER only when it will be written when loading the non-SMM state
so as to preserve bits that can theoretically be set on 32-bit vCPUs,
e.g. KVM always emulates EFER_SCE.

And because CR4.PAE is cleared only to play nice with EFER, wrap that
code in the long mode check as well.  Note, this may result in a
compiler warning about cr4 being consumed uninitialized.  Re-read CR4
even though it's technically unnecessary, as doing so allows for more
readable code and RSM emulation is not a performance critical path.

Fixes: 660a5d517aaab ("KVM: x86: save/load state on SMM switch")
Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoCIFS: keep FileInfo handle live during oplock break
Aurelien Aptel [Fri, 29 Mar 2019 09:49:12 +0000 (10:49 +0100)]
CIFS: keep FileInfo handle live during oplock break

commit b98749cac4a695f084a5ff076f4510b23e353ecd upstream.

In the oplock break handler, writing pending changes from pages puts
the FileInfo handle. If the refcount reaches zero it closes the handle
and waits for any oplock break handler to return, thus causing a deadlock.

To prevent this situation:

* We add a wait flag to cifsFileInfo_put() to decide whether we should
  wait for running/pending oplock break handlers

* We keep an additionnal reference of the SMB FileInfo handle so that
  for the rest of the handler putting the handle won't close it.
  - The ref is bumped everytime we queue the handler via the
    cifs_queue_oplock_break() helper.
  - The ref is decremented at the end of the handler

This bug was triggered by xfstest 464.

Also important fix to address the various reports of
oops in smb2_push_mandatory_locks

Signed-off-by: Aurelien Aptel <aaptel@suse.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
CC: Stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agonet: thunderx: don't allow jumbo frames with XDP
Matteo Croce [Thu, 11 Apr 2019 10:26:33 +0000 (12:26 +0200)]
net: thunderx: don't allow jumbo frames with XDP

[ Upstream commit 1f227d16083b2e280b7dde4ca78883d75593f2fd ]

The thunderx driver forbids to load an eBPF program if the MTU is too high,
but this can be circumvented by loading the eBPF, then raising the MTU.

Fix this by limiting the MTU if an eBPF program is already loaded.

Fixes: 05c773f52b96e ("net: thunderx: Add basic XDP support")
Signed-off-by: Matteo Croce <mcroce@redhat.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agonet: thunderx: raise XDP MTU to 1508
Matteo Croce [Thu, 11 Apr 2019 10:26:32 +0000 (12:26 +0200)]
net: thunderx: raise XDP MTU to 1508

[ Upstream commit 5ee15c101f29e0093ffb5448773ccbc786eb313b ]

The thunderx driver splits frames bigger than 1530 bytes to multiple
pages, making impossible to run an eBPF program on it.
This leads to a maximum MTU of 1508 if QinQ is in use.

The thunderx driver forbids to load an eBPF program if the MTU is higher
than 1500 bytes. Raise the limit to 1508 so it is possible to use L2
protocols which need some more headroom.

Fixes: 05c773f52b96e ("net: thunderx: Add basic XDP support")
Signed-off-by: Matteo Croce <mcroce@redhat.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoipv4: ensure rcu_read_lock() in ipv4_link_failure()
Eric Dumazet [Sun, 14 Apr 2019 00:32:21 +0000 (17:32 -0700)]
ipv4: ensure rcu_read_lock() in ipv4_link_failure()

[ Upstream commit c543cb4a5f07e09237ec0fc2c60c9f131b2c79ad ]

fib_compute_spec_dst() needs to be called under rcu protection.

syzbot reported :

WARNING: suspicious RCU usage
5.1.0-rc4+ #165 Not tainted
include/linux/inetdevice.h:220 suspicious rcu_dereference_check() usage!

other info that might help us debug this:

rcu_scheduler_active = 2, debug_locks = 1
1 lock held by swapper/0/0:
 #0: 0000000051b67925 ((&n->timer)){+.-.}, at: lockdep_copy_map include/linux/lockdep.h:170 [inline]
 #0: 0000000051b67925 ((&n->timer)){+.-.}, at: call_timer_fn+0xda/0x720 kernel/time/timer.c:1315

stack backtrace:
CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.1.0-rc4+ #165
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 <IRQ>
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x172/0x1f0 lib/dump_stack.c:113
 lockdep_rcu_suspicious+0x153/0x15d kernel/locking/lockdep.c:5162
 __in_dev_get_rcu include/linux/inetdevice.h:220 [inline]
 fib_compute_spec_dst+0xbbd/0x1030 net/ipv4/fib_frontend.c:294
 spec_dst_fill net/ipv4/ip_options.c:245 [inline]
 __ip_options_compile+0x15a7/0x1a10 net/ipv4/ip_options.c:343
 ipv4_link_failure+0x172/0x400 net/ipv4/route.c:1195
 dst_link_failure include/net/dst.h:427 [inline]
 arp_error_report+0xd1/0x1c0 net/ipv4/arp.c:297
 neigh_invalidate+0x24b/0x570 net/core/neighbour.c:995
 neigh_timer_handler+0xc35/0xf30 net/core/neighbour.c:1081
 call_timer_fn+0x190/0x720 kernel/time/timer.c:1325
 expire_timers kernel/time/timer.c:1362 [inline]
 __run_timers kernel/time/timer.c:1681 [inline]
 __run_timers kernel/time/timer.c:1649 [inline]
 run_timer_softirq+0x652/0x1700 kernel/time/timer.c:1694
 __do_softirq+0x266/0x95a kernel/softirq.c:293
 invoke_softirq kernel/softirq.c:374 [inline]
 irq_exit+0x180/0x1d0 kernel/softirq.c:414
 exiting_irq arch/x86/include/asm/apic.h:536 [inline]
 smp_apic_timer_interrupt+0x14a/0x570 arch/x86/kernel/apic/apic.c:1062
 apic_timer_interrupt+0xf/0x20 arch/x86/entry/entry_64.S:807

Fixes: ed0de45a1008 ("ipv4: recompile ip options in ipv4_link_failure")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Stephen Suryaputra <ssuryaextr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoipv4: recompile ip options in ipv4_link_failure
Stephen Suryaputra [Fri, 12 Apr 2019 20:19:27 +0000 (16:19 -0400)]
ipv4: recompile ip options in ipv4_link_failure

[ Upstream commit ed0de45a1008991fdaa27a0152befcb74d126a8b ]

Recompile IP options since IPCB may not be valid anymore when
ipv4_link_failure is called from arp_error_report.

Refer to the commit 3da1ed7ac398 ("net: avoid use IPCB in cipso_v4_error")
and the commit before that (9ef6b42ad6fd) for a similar issue.

Signed-off-by: Stephen Suryaputra <ssuryaextr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agovhost: reject zero size iova range
Jason Wang [Tue, 9 Apr 2019 04:10:25 +0000 (12:10 +0800)]
vhost: reject zero size iova range

[ Upstream commit 813dbeb656d6c90266f251d8bd2b02d445afa63f ]

We used to accept zero size iova range which will lead a infinite loop
in translate_desc(). Fixing this by failing the request in this case.

Reported-by: syzbot+d21e6e297322a900c128@syzkaller.appspotmail.com
Fixes: 6b1e6cc7 ("vhost: new device IOTLB API")
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoteam: set slave to promisc if team is already in promisc mode
Hangbin Liu [Mon, 8 Apr 2019 08:45:17 +0000 (16:45 +0800)]
team: set slave to promisc if team is already in promisc mode

[ Upstream commit 43c2adb9df7ddd6560fd3546d925b42cef92daa0 ]

After adding a team interface to bridge, the team interface will enter
promisc mode. Then if we add a new slave to team0, the slave will keep
promisc off. Fix it by setting slave to promisc on if team master is
already in promisc mode, also do the same for allmulti.

v2: add promisc and allmulti checking when delete ports

Fixes: 3d249d4ca7d0 ("net: introduce ethernet teaming device")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agotcp: tcp_grow_window() needs to respect tcp_space()
Eric Dumazet [Tue, 16 Apr 2019 17:55:20 +0000 (10:55 -0700)]
tcp: tcp_grow_window() needs to respect tcp_space()

[ Upstream commit 50ce163a72d817a99e8974222dcf2886d5deb1ae ]

For some reason, tcp_grow_window() correctly tests if enough room
is present before attempting to increase tp->rcv_ssthresh,
but does not prevent it to grow past tcp_space()

This is causing hard to debug issues, like failing
the (__tcp_select_window(sk) >= tp->rcv_wnd) test
in __tcp_ack_snd_check(), causing ACK delays and possibly
slow flows.

Depending on tcp_rmem[2], MTU, skb->len/skb->truesize ratio,
we can see the problem happening on "netperf -t TCP_RR -- -r 2000,2000"
after about 60 round trips, when the active side no longer sends
immediate acks.

This bug predates git history.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Wei Wang <weiwan@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agonet: fou: do not use guehdr after iptunnel_pull_offloads in gue_udp_recv
Lorenzo Bianconi [Tue, 9 Apr 2019 09:47:20 +0000 (11:47 +0200)]
net: fou: do not use guehdr after iptunnel_pull_offloads in gue_udp_recv

[ Upstream commit 988dc4a9a3b66be75b30405a5494faf0dc7cffb6 ]

gue tunnels run iptunnel_pull_offloads on received skbs. This can
determine a possible use-after-free accessing guehdr pointer since
the packet will be 'uncloned' running pskb_expand_head if it is a
cloned gso skb (e.g if the packet has been sent though a veth device)

Fixes: a09a4c8dd1ec ("tunnels: Remove encapsulation offloads on decap")
Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agonet: bridge: multicast: use rcu to access port list from br_multicast_start_querier
Nikolay Aleksandrov [Thu, 11 Apr 2019 12:08:25 +0000 (15:08 +0300)]
net: bridge: multicast: use rcu to access port list from br_multicast_start_querier

[ Upstream commit c5b493ce192bd7a4e7bd073b5685aad121eeef82 ]

br_multicast_start_querier() walks over the port list but it can be
called from a timer with only multicast_lock held which doesn't protect
the port list, so use RCU to walk over it.

Fixes: c83b8fab06fc ("bridge: Restart queries when last querier expires")
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agonet: bridge: fix per-port af_packet sockets
Nikolay Aleksandrov [Thu, 11 Apr 2019 10:56:39 +0000 (13:56 +0300)]
net: bridge: fix per-port af_packet sockets

[ Upstream commit 3b2e2904deb314cc77a2192f506f2fd44e3d10d0 ]

When the commit below was introduced it changed two visible things:
 - the skb was no longer passed through the protocol handlers with the
   original device
 - the skb was passed up the stack with skb->dev = bridge

The first change broke af_packet sockets on bridge ports. For example we
use them for hostapd which listens for ETH_P_PAE packets on the ports.
We discussed two possible fixes:
 - create a clone and pass it through NF_HOOK(), act on the original skb
   based on the result
 - somehow signal to the caller from the okfn() that it was called,
   meaning the skb is ok to be passed, which this patch is trying to
   implement via returning 1 from the bridge link-local okfn()

Note that we rely on the fact that NF_QUEUE/STOLEN would return 0 and
drop/error would return < 0 thus the okfn() is called only when the
return was 1, so we signal to the caller that it was called by preserving
the return value from nf_hook().

Fixes: 8626c56c8279 ("bridge: fix potential use-after-free when hook returns QUEUE or STOLEN verdict")
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agonet: atm: Fix potential Spectre v1 vulnerabilities
Gustavo A. R. Silva [Mon, 15 Apr 2019 20:57:23 +0000 (15:57 -0500)]
net: atm: Fix potential Spectre v1 vulnerabilities

[ Upstream commit 899537b73557aafbdd11050b501cf54b4f5c45af ]

arg is controlled by user-space, hence leading to a potential
exploitation of the Spectre variant 1 vulnerability.

This issue was detected with the help of Smatch:

net/atm/lec.c:715 lec_mcast_attach() warn: potential spectre issue 'dev_lec' [r] (local cap)

Fix this by sanitizing arg before using it to index dev_lec.

Notice that given that speculation windows are large, the policy is
to kill the speculation on the first load and not worry if it can be
completed with a dependent load/store [1].

[1] https://lore.kernel.org/lkml/20180423164740.GY17484@dhcp22.suse.cz/

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobonding: fix event handling for stacked bonds
Sabrina Dubroca [Fri, 12 Apr 2019 13:04:10 +0000 (15:04 +0200)]
bonding: fix event handling for stacked bonds

[ Upstream commit 92480b3977fd3884649d404cbbaf839b70035699 ]

When a bond is enslaved to another bond, bond_netdev_event() only
handles the event as if the bond is a master, and skips treating the
bond as a slave.

This leads to a refcount leak on the slave, since we don't remove the
adjacency to its master and the master holds a reference on the slave.

Reproducer:
  ip link add bondL type bond
  ip link add bondU type bond
  ip link set bondL master bondU
  ip link del bondL

No "Fixes:" tag, this code is older than git history.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agoLinux 4.14.113 v4.14.113
Greg Kroah-Hartman [Sat, 20 Apr 2019 07:15:10 +0000 (09:15 +0200)]
Linux 4.14.113

6 years agoappletalk: Fix compile regression
Arnd Bergmann [Wed, 6 Mar 2019 10:52:36 +0000 (11:52 +0100)]
appletalk: Fix compile regression

[ Upstream commit 27da0d2ef998e222a876c0cec72aa7829a626266 ]

A bugfix just broke compilation of appletalk when CONFIG_SYSCTL
is disabled:

In file included from net/appletalk/ddp.c:65:
net/appletalk/ddp.c: In function 'atalk_init':
include/linux/atalk.h:164:34: error: expected expression before 'do'
 #define atalk_register_sysctl()  do { } while(0)
                                  ^~
net/appletalk/ddp.c:1934:7: note: in expansion of macro 'atalk_register_sysctl'
  rc = atalk_register_sysctl();

This is easier to avoid by using conventional inline functions
as stubs rather than macros. The header already has inline
functions for other purposes, so I'm changing over all the
macros for consistency.

Fixes: 6377f787aeb9 ("appletalk: Fix use-after-free in atalk_proc_exit")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agomm: hide incomplete nr_indirectly_reclaimable in sysfs
Konstantin Khlebnikov [Tue, 9 Apr 2019 17:05:43 +0000 (20:05 +0300)]
mm: hide incomplete nr_indirectly_reclaimable in sysfs

In upstream branch this fixed by commit b29940c1abd7 ("mm: rename and
change semantics of nr_indirectly_reclaimable_bytes").

This fixes /sys/devices/system/node/node*/vmstat format:

...
nr_dirtied 6613155
nr_written 5796802
 11089216
...

Cc: <stable@vger.kernel.org> # 4.19.y
Fixes: 7aaf77272358 ("mm: don't show nr_indirectly_reclaimable in /proc/vmstat")
Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Cc: Roman Gushchin <guro@fb.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agonet: stmmac: Set dma ring length before enabling the DMA
Lars Persson [Mon, 15 Apr 2019 07:49:47 +0000 (09:49 +0200)]
net: stmmac: Set dma ring length before enabling the DMA

This was fixed in upstream by commit 7d9e6c5afab6 ("net: stmmac: Integrate
XGMAC into main driver flow") that is a new feature commit.

We found a race condition in the DMA init sequence that hits if the
PHY already has link up during stmmac_hw_setup. Since the ring length
was programmed after enabling the RX path, we might receive a packet
before the correct ring length is programmed. When that happened we
could not get reliable interrupts for DMA RX and the MTL complained
about RX FIFO overrun.

Signed-off-by: Lars Persson <larper@axis.com>
Cc: stable@vger.kernel.org # 4.14.x
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Cc: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: Fix selftests are changes for CVE 2019-7308
Balbir Singh [Wed, 3 Apr 2019 18:39:17 +0000 (18:39 +0000)]
bpf: Fix selftests are changes for CVE 2019-7308

The changes to fix the CVE 2019-7308 make the bpf verifier stricter
with respect to operations that were allowed earlier in unprivileged
mode. Fixup the test cases so that the error messages now correctly
reflect pointer arithmetic going out of range for tests.

Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: fix sanitation rewrite in case of non-pointers
Daniel Borkmann [Wed, 3 Apr 2019 18:39:16 +0000 (18:39 +0000)]
bpf: fix sanitation rewrite in case of non-pointers

commit 3612af783cf52c74a031a2f11b82247b2599d3cd upstream.

Marek reported that he saw an issue with the below snippet in that
timing measurements where off when loaded as unpriv while results
were reasonable when loaded as privileged:

    [...]
    uint64_t a = bpf_ktime_get_ns();
    uint64_t b = bpf_ktime_get_ns();
    uint64_t delta = b - a;
    if ((int64_t)delta > 0) {
    [...]

Turns out there is a bug where a corner case is missing in the fix
d3bd7413e0ca ("bpf: fix sanitation of alu op with pointer / scalar
type from different paths"), namely fixup_bpf_calls() only checks
whether aux has a non-zero alu_state, but it also needs to test for
the case of BPF_ALU_NON_POINTER since in both occasions we need to
skip the masking rewrite (as there is nothing to mask).

Fixes: d3bd7413e0ca ("bpf: fix sanitation of alu op with pointer / scalar type from different paths")
Reported-by: Marek Majkowski <marek@cloudflare.com>
Reported-by: Arthur Fabre <afabre@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/netdev/CAJPywTJqP34cK20iLM5YmUMz9KXQOdu1-+BZrGMAGgLuBWz7fg@mail.gmail.com/T/
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: do not restore dst_reg when cur_state is freed
Xu Yu [Wed, 3 Apr 2019 18:39:15 +0000 (18:39 +0000)]
bpf: do not restore dst_reg when cur_state is freed

commit 0803278b0b4d8eeb2b461fb698785df65a725d9e upstream.

Syzkaller hit 'KASAN: use-after-free Write in sanitize_ptr_alu' bug.

Call trace:

  dump_stack+0xbf/0x12e
  print_address_description+0x6a/0x280
  kasan_report+0x237/0x360
  sanitize_ptr_alu+0x85a/0x8d0
  adjust_ptr_min_max_vals+0x8f2/0x1ca0
  adjust_reg_min_max_vals+0x8ed/0x22e0
  do_check+0x1ca6/0x5d00
  bpf_check+0x9ca/0x2570
  bpf_prog_load+0xc91/0x1030
  __se_sys_bpf+0x61e/0x1f00
  do_syscall_64+0xc8/0x550
  entry_SYSCALL_64_after_hwframe+0x49/0xbe

Fault injection trace:

  kfree+0xea/0x290
  free_func_state+0x4a/0x60
  free_verifier_state+0x61/0xe0
  push_stack+0x216/0x2f0           <- inject failslab
  sanitize_ptr_alu+0x2b1/0x8d0
  adjust_ptr_min_max_vals+0x8f2/0x1ca0
  adjust_reg_min_max_vals+0x8ed/0x22e0
  do_check+0x1ca6/0x5d00
  bpf_check+0x9ca/0x2570
  bpf_prog_load+0xc91/0x1030
  __se_sys_bpf+0x61e/0x1f00
  do_syscall_64+0xc8/0x550
  entry_SYSCALL_64_after_hwframe+0x49/0xbe

When kzalloc() fails in push_stack(), free_verifier_state() will free
current verifier state. As push_stack() returns, dst_reg was restored
if ptr_is_dst_reg is false. However, as member of the cur_state,
dst_reg is also freed, and error occurs when dereferencing dst_reg.
Simply fix it by testing ret of push_stack() before restoring dst_reg.

Fixes: 979d63d50c0c ("bpf: prevent out of bounds speculation on pointer arithmetic")
Signed-off-by: Xu Yu <xuyu@linux.alibaba.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: fix inner map masking to prevent oob under speculation
Daniel Borkmann [Wed, 3 Apr 2019 18:39:14 +0000 (18:39 +0000)]
bpf: fix inner map masking to prevent oob under speculation

commit 9d5564ddcf2a0f5ba3fa1c3a1f8a1b59ad309553 upstream.

During review I noticed that inner meta map setup for map in
map is buggy in that it does not propagate all needed data
from the reference map which the verifier is later accessing.

In particular one such case is index masking to prevent out of
bounds access under speculative execution due to missing the
map's unpriv_array/index_mask field propagation. Fix this such
that the verifier is generating the correct code for inlined
lookups in case of unpriviledged use.

Before patch (test_verifier's 'map in map access' dump):

  # bpftool prog dump xla id 3
     0: (62) *(u32 *)(r10 -4) = 0
     1: (bf) r2 = r10
     2: (07) r2 += -4
     3: (18) r1 = map[id:4]
     5: (07) r1 += 272                |
     6: (61) r0 = *(u32 *)(r2 +0)     |
     7: (35) if r0 >= 0x1 goto pc+6   | Inlined map in map lookup
     8: (54) (u32) r0 &= (u32) 0      | with index masking for
     9: (67) r0 <<= 3                 | map->unpriv_array.
    10: (0f) r0 += r1                 |
    11: (79) r0 = *(u64 *)(r0 +0)     |
    12: (15) if r0 == 0x0 goto pc+1   |
    13: (05) goto pc+1                |
    14: (b7) r0 = 0                   |
    15: (15) if r0 == 0x0 goto pc+11
    16: (62) *(u32 *)(r10 -4) = 0
    17: (bf) r2 = r10
    18: (07) r2 += -4
    19: (bf) r1 = r0
    20: (07) r1 += 272                |
    21: (61) r0 = *(u32 *)(r2 +0)     | Index masking missing (!)
    22: (35) if r0 >= 0x1 goto pc+3   | for inner map despite
    23: (67) r0 <<= 3                 | map->unpriv_array set.
    24: (0f) r0 += r1                 |
    25: (05) goto pc+1                |
    26: (b7) r0 = 0                   |
    27: (b7) r0 = 0
    28: (95) exit

After patch:

  # bpftool prog dump xla id 1
     0: (62) *(u32 *)(r10 -4) = 0
     1: (bf) r2 = r10
     2: (07) r2 += -4
     3: (18) r1 = map[id:2]
     5: (07) r1 += 272                |
     6: (61) r0 = *(u32 *)(r2 +0)     |
     7: (35) if r0 >= 0x1 goto pc+6   | Same inlined map in map lookup
     8: (54) (u32) r0 &= (u32) 0      | with index masking due to
     9: (67) r0 <<= 3                 | map->unpriv_array.
    10: (0f) r0 += r1                 |
    11: (79) r0 = *(u64 *)(r0 +0)     |
    12: (15) if r0 == 0x0 goto pc+1   |
    13: (05) goto pc+1                |
    14: (b7) r0 = 0                   |
    15: (15) if r0 == 0x0 goto pc+12
    16: (62) *(u32 *)(r10 -4) = 0
    17: (bf) r2 = r10
    18: (07) r2 += -4
    19: (bf) r1 = r0
    20: (07) r1 += 272                |
    21: (61) r0 = *(u32 *)(r2 +0)     |
    22: (35) if r0 >= 0x1 goto pc+4   | Now fixed inlined inner map
    23: (54) (u32) r0 &= (u32) 0      | lookup with proper index masking
    24: (67) r0 <<= 3                 | for map->unpriv_array.
    25: (0f) r0 += r1                 |
    26: (05) goto pc+1                |
    27: (b7) r0 = 0                   |
    28: (b7) r0 = 0
    29: (95) exit

Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Vallish Vaidyeshwara <vallish@amazon.com>
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: fix sanitation of alu op with pointer / scalar type from different paths
Daniel Borkmann [Wed, 3 Apr 2019 18:39:13 +0000 (18:39 +0000)]
bpf: fix sanitation of alu op with pointer / scalar type from different paths

commit d3bd7413e0ca40b60cf60d4003246d067cafdeda upstream.

While 979d63d50c0c ("bpf: prevent out of bounds speculation on pointer
arithmetic") took care of rejecting alu op on pointer when e.g. pointer
came from two different map values with different map properties such as
value size, Jann reported that a case was not covered yet when a given
alu op is used in both "ptr_reg += reg" and "numeric_reg += reg" from
different branches where we would incorrectly try to sanitize based
on the pointer's limit. Catch this corner case and reject the program
instead.

Fixes: 979d63d50c0c ("bpf: prevent out of bounds speculation on pointer arithmetic")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Vallish Vaidyeshwara <vallish@amazon.com>
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: prevent out of bounds speculation on pointer arithmetic
Daniel Borkmann [Wed, 3 Apr 2019 18:39:12 +0000 (18:39 +0000)]
bpf: prevent out of bounds speculation on pointer arithmetic

commit 979d63d50c0c0f7bc537bf821e056cc9fe5abd38 upstream.

Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:

  - Load a map value pointer into R6
  - Load an index into R7
  - Do a slow computation (e.g. with a memory dependency) that
    loads a limit into R8 (e.g. load the limit from a map for
    high latency, then mask it to make the verifier happy)
  - Exit if R7 >= R8 (mispredicted branch)
  - Load R0 = R6[R7]
  - Load R0 = R6[R0]

For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.

In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.

Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.

There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.

The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...

  PTR += 0x1000 (e.g. K-based imm)
  PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
  PTR += 0x1000
  PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
  [...]

... which under speculation could end up as ...

  PTR += 0x1000
  PTR -= 0 [ truncated by mitigation ]
  PTR += 0x1000
  PTR -= 0 [ truncated by mitigation ]
  [...]

... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.

Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.

With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:

  # bpftool prog dump xlated id 282
  [...]
  28: (79) r1 = *(u64 *)(r7 +0)
  29: (79) r2 = *(u64 *)(r7 +8)
  30: (57) r1 &= 15
  31: (79) r3 = *(u64 *)(r0 +4608)
  32: (57) r3 &= 1
  33: (47) r3 |= 1
  34: (2d) if r2 > r3 goto pc+19
  35: (b4) (u32) r11 = (u32) 20479  |
  36: (1f) r11 -= r2                | Dynamic sanitation for pointer
  37: (4f) r11 |= r2                | arithmetic with registers
  38: (87) r11 = -r11               | containing bounded or known
  39: (c7) r11 s>>= 63              | scalars in order to prevent
  40: (5f) r11 &= r2                | out of bounds speculation.
  41: (0f) r4 += r11                |
  42: (71) r4 = *(u8 *)(r4 +0)
  43: (6f) r4 <<= r1
  [...]

For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:

  [...]
  16: (b4) (u32) r11 = (u32) 20479
  17: (1f) r11 -= r2
  18: (4f) r11 |= r2
  19: (87) r11 = -r11
  20: (c7) r11 s>>= 63
  21: (5f) r2 &= r11
  22: (0f) r2 += r0
  23: (61) r0 = *(u32 *)(r2 +0)
  [...]

JIT blinding example with non-conflicting use of r10:

  [...]
   d5: je     0x0000000000000106    _
   d7: mov    0x0(%rax),%edi       |
   da: mov    $0xf153246,%r10d     | Index load from map value and
   e0: xor    $0xf153259,%r10      | (const blinded) mask with 0x1f.
   e7: and    %r10,%rdi            |_
   ea: mov    $0x2f,%r10d          |
   f0: sub    %rdi,%r10            | Sanitized addition. Both use r10
   f3: or     %rdi,%r10            | but do not interfere with each
   f6: neg    %r10                 | other. (Neither do these instructions
   f9: sar    $0x3f,%r10           | interfere with the use of ax as temp
   fd: and    %r10,%rdi            | in interpreter.)
  100: add    %rax,%rdi            |_
  103: mov    0x0(%rdi),%eax
 [...]

Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.

  [0] Speculose: Analyzing the Security Implications of Speculative
      Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
      https://arxiv.org/pdf/1801.04084.pdf

  [1] A Systematic Evaluation of Transient Execution Attacks and
      Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
      Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
      Dmitry Evtyushkin, Daniel Gruss,
      https://arxiv.org/pdf/1811.05441.pdf

Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Vallish Vaidyeshwara <vallish@amazon.com>
[some checkpatch cleanups and backported to 4.14 by sblbir]
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: fix check_map_access smin_value test when pointer contains offset
Daniel Borkmann [Wed, 3 Apr 2019 18:39:11 +0000 (18:39 +0000)]
bpf: fix check_map_access smin_value test when pointer contains offset

commit b7137c4eab85c1cf3d46acdde90ce1163b28c873 upstream.

In check_map_access() we probe actual bounds through __check_map_access()
with offset of reg->smin_value + off for lower bound and offset of
reg->umax_value + off for the upper bound. However, even though the
reg->smin_value could have a negative value, the final result of the
sum with off could be positive when pointer arithmetic with known and
unknown scalars is combined. In this case we reject the program with
an error such as "R<x> min value is negative, either use unsigned index
or do a if (index >=0) check." even though the access itself would be
fine. Therefore extend the check to probe whether the actual resulting
reg->smin_value + off is less than zero.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
[backported to 4.14 sblbir]
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: restrict unknown scalars of mixed signed bounds for unprivileged
Daniel Borkmann [Wed, 3 Apr 2019 18:39:10 +0000 (18:39 +0000)]
bpf: restrict unknown scalars of mixed signed bounds for unprivileged

commit 9d7eceede769f90b66cfa06ad5b357140d5141ed upstream.

For unknown scalars of mixed signed bounds, meaning their smin_value is
negative and their smax_value is positive, we need to reject arithmetic
with pointer to map value. For unprivileged the goal is to mask every
map pointer arithmetic and this cannot reliably be done when it is
unknown at verification time whether the scalar value is negative or
positive. Given this is a corner case, the likelihood of breaking should
be very small.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
[backported to 4.14 sblbir]
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: restrict stack pointer arithmetic for unprivileged
Daniel Borkmann [Wed, 3 Apr 2019 18:39:09 +0000 (18:39 +0000)]
bpf: restrict stack pointer arithmetic for unprivileged

commit e4298d25830a866cc0f427d4bccb858e76715859 upstream.

Restrict stack pointer arithmetic for unprivileged users in that
arithmetic itself must not go out of bounds as opposed to the actual
access later on. Therefore after each adjust_ptr_min_max_vals() with
a stack pointer as a destination we simulate a check_stack_access()
of 1 byte on the destination and once that fails the program is
rejected for unprivileged program loads. This is analog to map
value pointer arithmetic and needed for masking later on.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
[backported to 4.14 sblbir]
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: restrict map value pointer arithmetic for unprivileged
Daniel Borkmann [Wed, 3 Apr 2019 18:39:08 +0000 (18:39 +0000)]
bpf: restrict map value pointer arithmetic for unprivileged

commit 0d6303db7970e6f56ae700fa07e11eb510cda125 upstream.

Restrict map value pointer arithmetic for unprivileged users in that
arithmetic itself must not go out of bounds as opposed to the actual
access later on. Therefore after each adjust_ptr_min_max_vals() with a
map value pointer as a destination it will simulate a check_map_access()
of 1 byte on the destination and once that fails the program is rejected
for unprivileged program loads. We use this later on for masking any
pointer arithmetic with the remainder of the map value space. The
likelihood of breaking any existing real-world unprivileged eBPF
program is very small for this corner case.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: enable access to ax register also from verifier rewrite
Daniel Borkmann [Wed, 3 Apr 2019 18:39:07 +0000 (18:39 +0000)]
bpf: enable access to ax register also from verifier rewrite

commit 9b73bfdd08e73231d6a90ae6db4b46b3fbf56c30 upstream.

Right now we are using BPF ax register in JIT for constant blinding as
well as in interpreter as temporary variable. Verifier will not be able
to use it simply because its use will get overridden from the former in
bpf_jit_blind_insn(). However, it can be made to work in that blinding
will be skipped if there is prior use in either source or destination
register on the instruction. Taking constraints of ax into account, the
verifier is then open to use it in rewrites under some constraints. Note,
ax register already has mappings in every eBPF JIT.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
[backported to 4.14 sblbir]
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: move tmp variable into ax register in interpreter
Daniel Borkmann [Wed, 3 Apr 2019 18:39:06 +0000 (18:39 +0000)]
bpf: move tmp variable into ax register in interpreter

commit 144cd91c4c2bced6eb8a7e25e590f6618a11e854 upstream.

This change moves the on-stack 64 bit tmp variable in ___bpf_prog_run()
into the hidden ax register. The latter is currently only used in JITs
for constant blinding as a temporary scratch register, meaning the BPF
interpreter will never see the use of ax. Therefore it is safe to use
it for the cases where tmp has been used earlier. This is needed to later
on allow restricted hidden use of ax in both interpreter and JITs.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
[backported to 4.14 sblbir]
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: move {prev_,}insn_idx into verifier env
Daniel Borkmann [Wed, 3 Apr 2019 18:39:05 +0000 (18:39 +0000)]
bpf: move {prev_,}insn_idx into verifier env

commit c08435ec7f2bc8f4109401f696fd55159b4b40cb upstream.

Move prev_insn_idx and insn_idx from the do_check() function into
the verifier environment, so they can be read inside the various
helper functions for handling the instructions. It's easier to put
this into the environment rather than changing all call-sites only
to pass it along. insn_idx is useful in particular since this later
on allows to hold state in env->insn_aux_data[env->insn_idx].

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Vallish Vaidyeshwara <vallish@amazon.com>
[Backported to 4.14 by sblbir]
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: fix stack state printing in verifier log
Alexei Starovoitov [Wed, 3 Apr 2019 18:39:04 +0000 (18:39 +0000)]
bpf: fix stack state printing in verifier log

commit 12a3cc8424fe1237aaeb982dec4f0914ddd22f3e upstream.

fix incorrect stack state prints in print_verifier_state()

Fixes: 638f5b90d460 ("bpf: reduce verifier memory consumption")
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: fix verifier NULL pointer dereference
Craig Gallek [Wed, 3 Apr 2019 18:39:03 +0000 (18:39 +0000)]
bpf: fix verifier NULL pointer dereference

commit 8c01c4f896aa3404af948880dcb29a2d51c833dc upstream.

do_check() can fail early without allocating env->cur_state under
memory pressure.  Syzkaller found the stack below on the linux-next
tree because of this.

  kasan: CONFIG_KASAN_INLINE enabled
  kasan: GPF could be caused by NULL-ptr deref or user memory access
  general protection fault: 0000 [#1] SMP KASAN
  Dumping ftrace buffer:
     (ftrace buffer empty)
  Modules linked in:
  CPU: 1 PID: 27062 Comm: syz-executor5 Not tainted 4.14.0-rc7+ #106
  Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
  task: ffff8801c2c74700 task.stack: ffff8801c3e28000
  RIP: 0010:free_verifier_state kernel/bpf/verifier.c:347 [inline]
  RIP: 0010:bpf_check+0xcf4/0x19c0 kernel/bpf/verifier.c:4533
  RSP: 0018:ffff8801c3e2f5c8 EFLAGS: 00010202
  RAX: dffffc0000000000 RBX: 00000000fffffff4 RCX: 0000000000000000
  RDX: 0000000000000070 RSI: ffffffff817d5aa9 RDI: 0000000000000380
  RBP: ffff8801c3e2f668 R08: 0000000000000000 R09: 1ffff100387c5d9f
  R10: 00000000218c4e80 R11: ffffffff85b34380 R12: ffff8801c4dc6a28
  R13: 0000000000000000 R14: ffff8801c4dc6a00 R15: ffff8801c4dc6a20
  FS:  00007f311079b700(0000) GS:ffff8801db300000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00000000004d4a24 CR3: 00000001cbcd0000 CR4: 00000000001406e0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
  Call Trace:
   bpf_prog_load+0xcbb/0x18e0 kernel/bpf/syscall.c:1166
   SYSC_bpf kernel/bpf/syscall.c:1690 [inline]
   SyS_bpf+0xae9/0x4620 kernel/bpf/syscall.c:1652
   entry_SYSCALL_64_fastpath+0x1f/0xbe
  RIP: 0033:0x452869
  RSP: 002b:00007f311079abe8 EFLAGS: 00000212 ORIG_RAX: 0000000000000141
  RAX: ffffffffffffffda RBX: 0000000000758020 RCX: 0000000000452869
  RDX: 0000000000000030 RSI: 0000000020168000 RDI: 0000000000000005
  RBP: 00007f311079aa20 R08: 0000000000000000 R09: 0000000000000000
  R10: 0000000000000000 R11: 0000000000000212 R12: 00000000004b7550
  R13: 00007f311079ab58 R14: 00000000004b7560 R15: 0000000000000000
  Code: df 48 c1 ea 03 80 3c 02 00 0f 85 e6 0b 00 00 4d 8b 6e 20 48 b8 00 00 00 00 00 fc ff df 49 8d bd 80 03 00 00 48 89 fa 48 c1 ea 03 <80> 3c 02 00 0f 85 b6 0b 00 00 49 8b bd 80 03 00 00 e8 d6 0c 26
  RIP: free_verifier_state kernel/bpf/verifier.c:347 [inline] RSP: ffff8801c3e2f5c8
  RIP: bpf_check+0xcf4/0x19c0 kernel/bpf/verifier.c:4533 RSP: ffff8801c3e2f5c8
  ---[ end trace c8d37f339dc64004 ]---

Fixes: 638f5b90d460 ("bpf: reduce verifier memory consumption")
Fixes: 1969db47f8d0 ("bpf: fix verifier memory leaks")
Signed-off-by: Craig Gallek <kraig@google.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: fix verifier memory leaks
Alexei Starovoitov [Wed, 3 Apr 2019 18:39:02 +0000 (18:39 +0000)]
bpf: fix verifier memory leaks

commit 1969db47f8d0e800397abd4ee4e8d27d2b578587 upstream.

fix verifier memory leaks

Fixes: 638f5b90d460 ("bpf: reduce verifier memory consumption")
Signed-off-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agobpf: reduce verifier memory consumption
Alexei Starovoitov [Wed, 3 Apr 2019 18:39:01 +0000 (18:39 +0000)]
bpf: reduce verifier memory consumption

commit 638f5b90d46016372a8e3e0a434f199cc5e12b8c upstream.

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

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

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

Cc: jakub.kicinski@netronome.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
[Backported to 4.14 by sblbir]
Signed-off-by: Balbir Singh <sblbir@amzn.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
6 years agodm: disable CRYPTO_TFM_REQ_MAY_SLEEP to fix a GFP_KERNEL recursion deadlock
Mikulas Patocka [Wed, 5 Sep 2018 13:17:45 +0000 (09:17 -0400)]
dm: disable CRYPTO_TFM_REQ_MAY_SLEEP to fix a GFP_KERNEL recursion deadlock

[ Upstream commit 432061b3da64e488be3403124a72a9250bbe96d4 ]

There's a XFS on dm-crypt deadlock, recursing back to itself due to the
crypto subsystems use of GFP_KERNEL, reported here:
https://bugzilla.kernel.org/show_bug.cgi?id=200835

* dm-crypt calls crypt_convert in xts mode
* init_crypt from xts.c calls kmalloc(GFP_KERNEL)
* kmalloc(GFP_KERNEL) recurses into the XFS filesystem, the filesystem
tries to submit some bios and wait for them, causing a deadlock

Fix this by updating both the DM crypt and integrity targets to no
longer use the CRYPTO_TFM_REQ_MAY_SLEEP flag, which will change the
crypto allocations from GFP_KERNEL to GFP_ATOMIC, therefore they can't
recurse into a filesystem.  A GFP_ATOMIC allocation can fail, but
init_crypt() in xts.c handles the allocation failure gracefully - it
will fall back to preallocated buffer if the allocation fails.

The crypto API maintainer says that the crypto API only needs to
allocate memory when dealing with unaligned buffers and therefore
turning CRYPTO_TFM_REQ_MAY_SLEEP off is safe (see this discussion:
https://www.redhat.com/archives/dm-devel/2018-August/msg00195.html )

Cc: stable@vger.kernel.org
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Sasha Levin (Microsoft) <sashal@kernel.org>
6 years agobpf: fix use after free in bpf_evict_inode
Daniel Borkmann [Mon, 25 Mar 2019 14:54:43 +0000 (15:54 +0100)]
bpf: fix use after free in bpf_evict_inode

[ Upstream commit 1da6c4d9140cb7c13e87667dc4e1488d6c8fc10f ]

syzkaller was able to generate the following UAF in bpf:

  BUG: KASAN: use-after-free in lookup_last fs/namei.c:2269 [inline]
  BUG: KASAN: use-after-free in path_lookupat.isra.43+0x9f8/0xc00 fs/namei.c:2318
  Read of size 1 at addr ffff8801c4865c47 by task syz-executor2/9423

  CPU: 0 PID: 9423 Comm: syz-executor2 Not tainted 4.20.0-rc1-next-20181109+
  #110
  Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS
  Google 01/01/2011
  Call Trace:
    __dump_stack lib/dump_stack.c:77 [inline]
    dump_stack+0x244/0x39d lib/dump_stack.c:113
    print_address_description.cold.7+0x9/0x1ff mm/kasan/report.c:256
    kasan_report_error mm/kasan/report.c:354 [inline]
    kasan_report.cold.8+0x242/0x309 mm/kasan/report.c:412
    __asan_report_load1_noabort+0x14/0x20 mm/kasan/report.c:430
    lookup_last fs/namei.c:2269 [inline]
    path_lookupat.isra.43+0x9f8/0xc00 fs/namei.c:2318
    filename_lookup+0x26a/0x520 fs/namei.c:2348
    user_path_at_empty+0x40/0x50 fs/namei.c:2608
    user_path include/linux/namei.h:62 [inline]
    do_mount+0x180/0x1ff0 fs/namespace.c:2980
    ksys_mount+0x12d/0x140 fs/namespace.c:3258
    __do_sys_mount fs/namespace.c:3272 [inline]
    __se_sys_mount fs/namespace.c:3269 [inline]
    __x64_sys_mount+0xbe/0x150 fs/namespace.c:3269
    do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290
    entry_SYSCALL_64_after_hwframe+0x49/0xbe
  RIP: 0033:0x457569
  Code: fd b3 fb ff c3 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f8 48 89 f7
  48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff
  ff 0f 83 cb b3 fb ff c3 66 2e 0f 1f 84 00 00 00 00
  RSP: 002b:00007fde6ed96c78 EFLAGS: 00000246 ORIG_RAX: 00000000000000a5
  RAX: ffffffffffffffda RBX: 0000000000000005 RCX: 0000000000457569
  RDX: 0000000020000040 RSI: 0000000020000000 RDI: 0000000000000000
  RBP: 000000000072bf00 R08: 0000000020000340 R09: 0000000000000000
  R10: 0000000000200000 R11: 0000000000000246 R12: 00007fde6ed976d4
  R13: 00000000004c2c24 R14: 00000000004d4990 R15: 00000000ffffffff

  Allocated by task 9424:
    save_stack+0x43/0xd0 mm/kasan/kasan.c:448
    set_track mm/kasan/kasan.c:460 [inline]
    kasan_kmalloc+0xc7/0xe0 mm/kasan/kasan.c:553
    __do_kmalloc mm/slab.c:3722 [inline]
    __kmalloc_track_caller+0x157/0x760 mm/slab.c:3737
    kstrdup+0x39/0x70 mm/util.c:49
    bpf_symlink+0x26/0x140 kernel/bpf/inode.c:356
    vfs_symlink+0x37a/0x5d0 fs/namei.c:4127
    do_symlinkat+0x242/0x2d0 fs/namei.c:4154
    __do_sys_symlink fs/namei.c:4173 [inline]
    __se_sys_symlink fs/namei.c:4171 [inline]
    __x64_sys_symlink+0x59/0x80 fs/namei.c:4171
    do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290
    entry_SYSCALL_64_after_hwframe+0x49/0xbe

  Freed by task 9425:
    save_stack+0x43/0xd0 mm/kasan/kasan.c:448
    set_track mm/kasan/kasan.c:460 [inline]
    __kasan_slab_free+0x102/0x150 mm/kasan/kasan.c:521
    kasan_slab_free+0xe/0x10 mm/kasan/kasan.c:528
    __cache_free mm/slab.c:3498 [inline]
    kfree+0xcf/0x230 mm/slab.c:3817
    bpf_evict_inode+0x11f/0x150 kernel/bpf/inode.c:565
    evict+0x4b9/0x980 fs/inode.c:558
    iput_final fs/inode.c:1550 [inline]
    iput+0x674/0xa90 fs/inode.c:1576
    do_unlinkat+0x733/0xa30 fs/namei.c:4069
    __do_sys_unlink fs/namei.c:4110 [inline]
    __se_sys_unlink fs/namei.c:4108 [inline]
    __x64_sys_unlink+0x42/0x50 fs/namei.c:4108
    do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290
    entry_SYSCALL_64_after_hwframe+0x49/0xbe

In this scenario path lookup under RCU is racing with the final
unlink in case of symlinks. As Linus puts it in his analysis:

  [...] We actually RCU-delay the inode freeing itself, but
  when we do the final iput(), the "evict()" function is called
  synchronously. Now, the simple fix would seem to just RCU-delay
  the kfree() of the symlink data in bpf_evict_inode(). Maybe
  that's the right thing to do. [...]

Al suggested to piggy-back on the ->destroy_inode() callback in
order to implement RCU deferral there which can then kfree() the
inode->i_link eventually right before putting inode back into
inode cache. By reusing free_inode_nonrcu() from there we can
avoid the need for our own inode cache and just reuse generic
one as we currently do.

And in-fact on top of all this we should just get rid of the
bpf_evict_inode() entirely. This means truncate_inode_pages_final()
and clear_inode() will then simply be called by the fs core via
evict(). Dropping the reference should really only be done when
inode is unhashed and nothing reachable anymore, so it's better
also moved into the final ->destroy_inode() callback.

Fixes: 0f98621bef5d ("bpf, inode: add support for symlinks and fix mtime/ctime")
Reported-by: syzbot+fb731ca573367b7f6564@syzkaller.appspotmail.com
Reported-by: syzbot+a13e5ead792d6df37818@syzkaller.appspotmail.com
Reported-by: syzbot+7a8ba368b47fdefca61e@syzkaller.appspotmail.com
Suggested-by: Al Viro <viro@zeniv.linux.org.uk>
Analyzed-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Link: https://lore.kernel.org/lkml/0000000000006946d2057bbd0eef@google.com/T/
Signed-off-by: Sasha Levin (Microsoft) <sashal@kernel.org>
6 years agoinclude/linux/swap.h: use offsetof() instead of custom __swapoffset macro
Pi-Hsun Shih [Wed, 13 Mar 2019 18:44:33 +0000 (11:44 -0700)]
include/linux/swap.h: use offsetof() instead of custom __swapoffset macro

[ Upstream commit a4046c06be50a4f01d435aa7fe57514818e6cc82 ]

Use offsetof() to calculate offset of a field to take advantage of
compiler built-in version when possible, and avoid UBSAN warning when
compiling with Clang:

  UBSAN: Undefined behaviour in mm/swapfile.c:3010:38
  member access within null pointer of type 'union swap_header'
  CPU: 6 PID: 1833 Comm: swapon Tainted: G S                4.19.23 #43
  Call trace:
   dump_backtrace+0x0/0x194
   show_stack+0x20/0x2c
   __dump_stack+0x20/0x28
   dump_stack+0x70/0x94
   ubsan_epilogue+0x14/0x44
   ubsan_type_mismatch_common+0xf4/0xfc
   __ubsan_handle_type_mismatch_v1+0x34/0x54
   __se_sys_swapon+0x654/0x1084
   __arm64_sys_swapon+0x1c/0x24
   el0_svc_common+0xa8/0x150
   el0_svc_compat_handler+0x2c/0x38
   el0_svc_compat+0x8/0x18

Link: http://lkml.kernel.org/r/20190312081902.223764-1-pihsun@chromium.org
Signed-off-by: Pi-Hsun Shih <pihsun@chromium.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agolib/div64.c: off by one in shift
Stanislaw Gruszka [Fri, 8 Mar 2019 00:28:18 +0000 (16:28 -0800)]
lib/div64.c: off by one in shift

[ Upstream commit cdc94a37493135e355dfc0b0e086d84e3eadb50d ]

fls counts bits starting from 1 to 32 (returns 0 for zero argument).  If
we add 1 we shift right one bit more and loose precision from divisor,
what cause function incorect results with some numbers.

Corrected code was tested in user-space, see bugzilla:
   https://bugzilla.kernel.org/show_bug.cgi?id=202391

Link: http://lkml.kernel.org/r/1548686944-11891-1-git-send-email-sgruszka@redhat.com
Fixes: 658716d19f8f ("div64_u64(): improve precision on 32bit platforms")
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Reported-by: Siarhei Volkau <lis8215@gmail.com>
Tested-by: Siarhei Volkau <lis8215@gmail.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoappletalk: Fix use-after-free in atalk_proc_exit
YueHaibing [Fri, 1 Mar 2019 02:57:57 +0000 (10:57 +0800)]
appletalk: Fix use-after-free in atalk_proc_exit

[ Upstream commit 6377f787aeb945cae7abbb6474798de129e1f3ac ]

KASAN report this:

BUG: KASAN: use-after-free in pde_subdir_find+0x12d/0x150 fs/proc/generic.c:71
Read of size 8 at addr ffff8881f41fe5b0 by task syz-executor.0/2806

CPU: 0 PID: 2806 Comm: syz-executor.0 Not tainted 5.0.0-rc7+ #45
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0xfa/0x1ce lib/dump_stack.c:113
 print_address_description+0x65/0x270 mm/kasan/report.c:187
 kasan_report+0x149/0x18d mm/kasan/report.c:317
 pde_subdir_find+0x12d/0x150 fs/proc/generic.c:71
 remove_proc_entry+0xe8/0x420 fs/proc/generic.c:667
 atalk_proc_exit+0x18/0x820 [appletalk]
 atalk_exit+0xf/0x5a [appletalk]
 __do_sys_delete_module kernel/module.c:1018 [inline]
 __se_sys_delete_module kernel/module.c:961 [inline]
 __x64_sys_delete_module+0x3dc/0x5e0 kernel/module.c:961
 do_syscall_64+0x147/0x600 arch/x86/entry/common.c:290
 entry_SYSCALL_64_after_hwframe+0x49/0xbe
RIP: 0033:0x462e99
Code: f7 d8 64 89 02 b8 ff ff ff ff c3 66 0f 1f 44 00 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 bc ff ff ff f7 d8 64 89 01 48
RSP: 002b:00007fb2de6b9c58 EFLAGS: 00000246 ORIG_RAX: 00000000000000b0
RAX: ffffffffffffffda RBX: 000000000073bf00 RCX: 0000000000462e99
RDX: 0000000000000000 RSI: 0000000000000000 RDI: 00000000200001c0
RBP: 0000000000000002 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 00007fb2de6ba6bc
R13: 00000000004bccaa R14: 00000000006f6bc8 R15: 00000000ffffffff

Allocated by task 2806:
 set_track mm/kasan/common.c:85 [inline]
 __kasan_kmalloc.constprop.3+0xa0/0xd0 mm/kasan/common.c:496
 slab_post_alloc_hook mm/slab.h:444 [inline]
 slab_alloc_node mm/slub.c:2739 [inline]
 slab_alloc mm/slub.c:2747 [inline]
 kmem_cache_alloc+0xcf/0x250 mm/slub.c:2752
 kmem_cache_zalloc include/linux/slab.h:730 [inline]
 __proc_create+0x30f/0xa20 fs/proc/generic.c:408
 proc_mkdir_data+0x47/0x190 fs/proc/generic.c:469
 0xffffffffc10c01bb
 0xffffffffc10c0166
 do_one_initcall+0xfa/0x5ca init/main.c:887
 do_init_module+0x204/0x5f6 kernel/module.c:3460
 load_module+0x66b2/0x8570 kernel/module.c:3808
 __do_sys_finit_module+0x238/0x2a0 kernel/module.c:3902
 do_syscall_64+0x147/0x600 arch/x86/entry/common.c:290
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

Freed by task 2806:
 set_track mm/kasan/common.c:85 [inline]
 __kasan_slab_free+0x130/0x180 mm/kasan/common.c:458
 slab_free_hook mm/slub.c:1409 [inline]
 slab_free_freelist_hook mm/slub.c:1436 [inline]
 slab_free mm/slub.c:2986 [inline]
 kmem_cache_free+0xa6/0x2a0 mm/slub.c:3002
 pde_put+0x6e/0x80 fs/proc/generic.c:647
 remove_proc_entry+0x1d3/0x420 fs/proc/generic.c:684
 0xffffffffc10c031c
 0xffffffffc10c0166
 do_one_initcall+0xfa/0x5ca init/main.c:887
 do_init_module+0x204/0x5f6 kernel/module.c:3460
 load_module+0x66b2/0x8570 kernel/module.c:3808
 __do_sys_finit_module+0x238/0x2a0 kernel/module.c:3902
 do_syscall_64+0x147/0x600 arch/x86/entry/common.c:290
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

The buggy address belongs to the object at ffff8881f41fe500
 which belongs to the cache proc_dir_entry of size 256
The buggy address is located 176 bytes inside of
 256-byte region [ffff8881f41fe500ffff8881f41fe600)
The buggy address belongs to the page:
page:ffffea0007d07f80 count:1 mapcount:0 mapping:ffff8881f6e69a00 index:0x0
flags: 0x2fffc0000000200(slab)
raw: 02fffc0000000200 dead000000000100 dead000000000200 ffff8881f6e69a00
raw: 0000000000000000 00000000800c000c 00000001ffffffff 0000000000000000
page dumped because: kasan: bad access detected

Memory state around the buggy address:
 ffff8881f41fe480: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
 ffff8881f41fe500: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>ffff8881f41fe580: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
                                     ^
 ffff8881f41fe600: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb
 ffff8881f41fe680: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb

It should check the return value of atalk_proc_init fails,
otherwise atalk_exit will trgger use-after-free in pde_subdir_find
while unload the module.This patch fix error cleanup path of atalk_init

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agodrm/amdkfd: use init_mqd function to allocate object for hid_mqd (CI)
Kevin Wang [Fri, 22 Feb 2019 04:36:49 +0000 (12:36 +0800)]
drm/amdkfd: use init_mqd function to allocate object for hid_mqd (CI)

[ Upstream commit cac734c2dbd2514f14c8c6a17caba1990d83bf1d ]

if use the legacy method to allocate object, when mqd_hiq need to run
uninit code, it will be cause WARNING call trace.

eg: (s3 suspend test)
[   34.918944] Call Trace:
[   34.918948]  [<ffffffff92961dc1>] dump_stack+0x19/0x1b
[   34.918950]  [<ffffffff92297648>] __warn+0xd8/0x100
[   34.918951]  [<ffffffff9229778d>] warn_slowpath_null+0x1d/0x20
[   34.918991]  [<ffffffffc03ce1fe>] uninit_mqd_hiq_sdma+0x4e/0x50 [amdgpu]
[   34.919028]  [<ffffffffc03d0ef7>] uninitialize+0x37/0xe0 [amdgpu]
[   34.919064]  [<ffffffffc03d15a6>] kernel_queue_uninit+0x16/0x30 [amdgpu]
[   34.919086]  [<ffffffffc03d26c2>] pm_uninit+0x12/0x20 [amdgpu]
[   34.919107]  [<ffffffffc03d4915>] stop_nocpsch+0x15/0x20 [amdgpu]
[   34.919129]  [<ffffffffc03c1dce>] kgd2kfd_suspend.part.4+0x2e/0x50 [amdgpu]
[   34.919150]  [<ffffffffc03c2667>] kgd2kfd_suspend+0x17/0x20 [amdgpu]
[   34.919171]  [<ffffffffc03c103a>] amdgpu_amdkfd_suspend+0x1a/0x20 [amdgpu]
[   34.919187]  [<ffffffffc02ec428>] amdgpu_device_suspend+0x88/0x3a0 [amdgpu]
[   34.919189]  [<ffffffff922e22cf>] ? enqueue_entity+0x2ef/0xbe0
[   34.919205]  [<ffffffffc02e8220>] amdgpu_pmops_suspend+0x20/0x30 [amdgpu]
[   34.919207]  [<ffffffff925c56ff>] pci_pm_suspend+0x6f/0x150
[   34.919208]  [<ffffffff925c5690>] ? pci_pm_freeze+0xf0/0xf0
[   34.919210]  [<ffffffff926b45c6>] dpm_run_callback+0x46/0x90
[   34.919212]  [<ffffffff926b49db>] __device_suspend+0xfb/0x2a0
[   34.919213]  [<ffffffff926b4b9f>] async_suspend+0x1f/0xa0
[   34.919214]  [<ffffffff922c918f>] async_run_entry_fn+0x3f/0x130
[   34.919216]  [<ffffffff922b9d4f>] process_one_work+0x17f/0x440
[   34.919217]  [<ffffffff922bade6>] worker_thread+0x126/0x3c0
[   34.919218]  [<ffffffff922bacc0>] ? manage_workers.isra.25+0x2a0/0x2a0
[   34.919220]  [<ffffffff922c1c31>] kthread+0xd1/0xe0
[   34.919221]  [<ffffffff922c1b60>] ? insert_kthread_work+0x40/0x40
[   34.919222]  [<ffffffff92974c1d>] ret_from_fork_nospec_begin+0x7/0x21
[   34.919224]  [<ffffffff922c1b60>] ? insert_kthread_work+0x40/0x40
[   34.919224] ---[ end trace 38cd9f65c963adad ]---

Signed-off-by: Kevin Wang <kevin1.wang@amd.com>
Reviewed-by: Oak Zeng <Oak.Zeng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoARM: 8839/1: kprobe: make patch_lock a raw_spinlock_t
Yang Shi [Wed, 13 Feb 2019 16:14:23 +0000 (17:14 +0100)]
ARM: 8839/1: kprobe: make patch_lock a raw_spinlock_t

[ Upstream commit 143c2a89e0e5fda6c6fd08d7bc1126438c19ae90 ]

When running kprobe on -rt kernel, the below bug is caught:

|BUG: sleeping function called from invalid context at kernel/locking/rtmutex.c:931
|in_atomic(): 1, irqs_disabled(): 128, pid: 14, name: migration/0
|Preemption disabled at:[<802f2b98>] cpu_stopper_thread+0xc0/0x140
|CPU: 0 PID: 14 Comm: migration/0 Tainted: G O 4.8.3-rt2 #1
|Hardware name: Freescale LS1021A
|[<8025a43c>] (___might_sleep)
|[<80b5b324>] (rt_spin_lock)
|[<80b5c31c>] (__patch_text_real)
|[<80b5c3ac>] (patch_text_stop_machine)
|[<802f2920>] (multi_cpu_stop)

Since patch_text_stop_machine() is called in stop_machine() which
disables IRQ, sleepable lock should be not used in this atomic context,
 so replace patch_lock to raw lock.

Signed-off-by: Yang Shi <yang.shi@linaro.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agodrm/nouveau/volt/gf117: fix speedo readout register
Ilia Mirkin [Sun, 13 Jan 2019 22:50:10 +0000 (17:50 -0500)]
drm/nouveau/volt/gf117: fix speedo readout register

[ Upstream commit fc782242749fa4235592854fafe1a1297583c1fb ]

GF117 appears to use the same register as GK104 (but still with the
general Fermi readout mechanism).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108980
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agocoresight: cpu-debug: Support for CA73 CPUs
Leo Yan [Tue, 5 Feb 2019 23:24:53 +0000 (16:24 -0700)]
coresight: cpu-debug: Support for CA73 CPUs

[ Upstream commit a0f890aba2be33377f4eb24e13633c4a76a68f38 ]

This patch is to add the AMBA device ID for CA73 CPU, so that CPU debug
module can be initialized successfully when a SoC contain CA73 CPUs.

This patch has been verified on 96boards Hikey960.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoRevert "ACPI / EC: Remove old CLEAR_ON_RESUME quirk"
Zhang Rui [Fri, 1 Feb 2019 06:13:41 +0000 (14:13 +0800)]
Revert "ACPI / EC: Remove old CLEAR_ON_RESUME quirk"

[ Upstream commit b6a3e1475b0220378ad32bdf4d8692f058b1fc03 ]

On some Samsung hardware, it is necessary to clear events accumulated by
the EC during sleep. These ECs stop reporting GPEs until they are manually
polled, if too many events are accumulated.
Thus the CLEAR_ON_RESUME quirk is introduced to send EC query commands
unconditionally after resume to clear all the EC query events on those
platforms.

Later, commit 4c237371f290 ("ACPI / EC: Remove old CLEAR_ON_RESUME quirk")
removes the CLEAR_ON_RESUME quirk because we thought the new EC IRQ
polling logic should handle this case.

Now it has been proved that the EC IRQ Polling logic does not fix the
issue actually because we got regression report on these Samsung
platforms after removing the quirk.

Thus revert commit 4c237371f290 ("ACPI / EC: Remove old CLEAR_ON_RESUME
quirk") to introduce back the Samsung quirk in this patch.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=44161
Tested-by: Ortwin Glück <odi@odi.ch>
Tested-by: Francisco Cribari <cribari@gmail.com>
Tested-by: Balazs Varga <balazs4web@gmail.com>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agocrypto: axis - fix for recursive locking from bottom half
Lars Persson [Wed, 23 Jan 2019 11:59:42 +0000 (12:59 +0100)]
crypto: axis - fix for recursive locking from bottom half

[ Upstream commit c34a83820f59bb275e5f2d55cd5ea99c64f6ef23 ]

Clients may submit a new requests from the completion callback
context. The driver was not prepared to receive a request in this
state because it already held the request queue lock and a recursive
lock error is triggered.

Now all completions are queued up until we are ready to drop the queue
lock and then delivered.

The fault was triggered by TCP over an IPsec connection in the LTP
test suite:
  LTP: starting tcp4_ipsec02 (tcp_ipsec.sh -p ah -m transport -s "100 1000 65535")
  BUG: spinlock recursion on CPU#1, genload/943
   lock: 0xbf3c3094, .magic: dead4ead, .owner: genload/943, .owner_cpu: 1
  CPU: 1 PID: 943 Comm: genload Tainted: G           O    4.9.62-axis5-devel #6
  Hardware name: Axis ARTPEC-6 Platform
   (unwind_backtrace) from [<8010d134>] (show_stack+0x18/0x1c)
   (show_stack) from [<803a289c>] (dump_stack+0x84/0x98)
   (dump_stack) from [<8016e164>] (do_raw_spin_lock+0x124/0x128)
   (do_raw_spin_lock) from [<804de1a4>] (artpec6_crypto_submit+0x2c/0xa0)
   (artpec6_crypto_submit) from [<804def38>] (artpec6_crypto_prepare_submit_hash+0xd0/0x54c)
   (artpec6_crypto_prepare_submit_hash) from [<7f3165f0>] (ah_output+0x2a4/0x3dc [ah4])
   (ah_output [ah4]) from [<805df9bc>] (xfrm_output_resume+0x178/0x4a4)
   (xfrm_output_resume) from [<805d283c>] (xfrm4_output+0xac/0xbc)
   (xfrm4_output) from [<80587928>] (ip_queue_xmit+0x140/0x3b4)
   (ip_queue_xmit) from [<805a13b4>] (tcp_transmit_skb+0x4c4/0x95c)
   (tcp_transmit_skb) from [<8059f218>] (tcp_rcv_state_process+0xdf4/0xdfc)
   (tcp_rcv_state_process) from [<805a7530>] (tcp_v4_do_rcv+0x64/0x1ac)
   (tcp_v4_do_rcv) from [<805a9724>] (tcp_v4_rcv+0xa34/0xb74)
   (tcp_v4_rcv) from [<80581d34>] (ip_local_deliver_finish+0x78/0x2b0)
   (ip_local_deliver_finish) from [<8058259c>] (ip_local_deliver+0xe4/0x104)
   (ip_local_deliver) from [<805d23ec>] (xfrm4_transport_finish+0xf4/0x144)
   (xfrm4_transport_finish) from [<805df564>] (xfrm_input+0x4f4/0x74c)
   (xfrm_input) from [<804de420>] (artpec6_crypto_task+0x208/0x38c)
   (artpec6_crypto_task) from [<801271b0>] (tasklet_action+0x60/0xec)
   (tasklet_action) from [<801266d4>] (__do_softirq+0xcc/0x3a4)
   (__do_softirq) from [<80126d20>] (irq_exit+0xf4/0x15c)
   (irq_exit) from [<801741e8>] (__handle_domain_irq+0x68/0xbc)
   (__handle_domain_irq) from [<801014f0>] (gic_handle_irq+0x50/0x94)
   (gic_handle_irq) from [<80657370>] (__irq_usr+0x50/0x80)

Signed-off-by: Lars Persson <larper@axis.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agodrm/panel: panel-innolux: set display off in innolux_panel_unprepare
Hsin-Yi, Wang [Wed, 9 Jan 2019 06:59:22 +0000 (14:59 +0800)]
drm/panel: panel-innolux: set display off in innolux_panel_unprepare

[ Upstream commit 46f3ceaffa81e846677bca8668e0ad40e643cffd ]

Move mipi_dsi_dcs_set_display_off() from innolux_panel_disable()
to innolux_panel_unprepare(), so they are consistent with
innolux_panel_enable() and innolux_panel_prepare().

This also fixes some mode check and irq timeout issue in MTK dsi code.

Since some dsi code (e.g. mtk_dsi) have following call trace:
1. drm_panel_disable(), which calls innolux_panel_disable()
2. switch to cmd mode
3. drm_panel_unprepare(), which calls innolux_panel_unprepare()

However, mtk_dsi needs to be in cmd mode to be able to send commands
(e.g. mipi_dsi_dcs_set_display_off() and mipi_dsi_dcs_enter_sleep_mode()),
so we need these functions to be called after the switch to cmd mode happens,
i.e. in innolux_panel_unprepare.

Signed-off-by: Hsin-Yi, Wang <hsinyi@chromium.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20190109065922.231753-1-hsinyi@chromium.org
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agolkdtm: Add tests for NULL pointer dereference
Christophe Leroy [Fri, 14 Dec 2018 15:26:20 +0000 (15:26 +0000)]
lkdtm: Add tests for NULL pointer dereference

[ Upstream commit 59a12205d3c32aee4c13ca36889fdf7cfed31126 ]

Introduce lkdtm tests for NULL pointer dereference: check access or exec
at NULL address, since these errors tend to be reported differently from
the general fault error text. For example from x86:

    pr_alert("BUG: unable to handle kernel %s at %px\n",
        address < PAGE_SIZE ? "NULL pointer dereference" : "paging request",
        (void *)address);

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agolkdtm: Print real addresses
Christophe Leroy [Wed, 7 Nov 2018 20:14:10 +0000 (20:14 +0000)]
lkdtm: Print real addresses

[ Upstream commit 4c411157a42f122051ae3469bee0b5cabe89e139 ]

Today, when doing a lkdtm test before the readiness of the
random generator, (ptrval) is printed instead of the address
at which it perform the fault:

[ 1597.337030] lkdtm: Performing direct entry EXEC_USERSPACE
[ 1597.337142] lkdtm: attempting ok execution at (ptrval)
[ 1597.337398] lkdtm: attempting bad execution at (ptrval)
[ 1597.337460] kernel tried to execute user page (77858000) -exploit attempt? (uid: 0)
[ 1597.344769] Unable to handle kernel paging request for instruction fetch
[ 1597.351392] Faulting instruction address: 0x77858000
[ 1597.356312] Oops: Kernel access of bad area, sig: 11 [#1]

If the lkdtm test is done later on, it prints an hashed address.

In both cases this is pointless. The purpose of the test is to
ensure the kernel generates an Oops at the expected address,
so real addresses needs to be printed. This patch fixes that.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agosoc/tegra: pmc: Drop locking from tegra_powergate_is_powered()
Dmitry Osipenko [Sun, 21 Oct 2018 18:36:14 +0000 (21:36 +0300)]
soc/tegra: pmc: Drop locking from tegra_powergate_is_powered()

[ Upstream commit b6e1fd17a38bd1d97c11d69fd3207b3ef9bfa4b3 ]

This fixes splats like the one below if CONFIG_DEBUG_ATOMIC_SLEEP=y
and machine (Tegra30) booted with SMP=n or all secondary CPU's are put
offline. Locking isn't needed because it protects atomic operation.

BUG: sleeping function called from invalid context at kernel/locking/mutex.c:254
in_atomic(): 1, irqs_disabled(): 128, pid: 0, name: swapper/0
CPU: 0 PID: 0 Comm: swapper/0 Tainted: G         C        4.18.0-next-20180821-00180-gc3ebb6544e44-dirty #823
Hardware name: NVIDIA Tegra SoC (Flattened Device Tree)
[<c01134f4>] (unwind_backtrace) from [<c010db2c>] (show_stack+0x20/0x24)
[<c010db2c>] (show_stack) from [<c0bd0f3c>] (dump_stack+0x94/0xa8)
[<c0bd0f3c>] (dump_stack) from [<c0151df8>] (___might_sleep+0x13c/0x174)
[<c0151df8>] (___might_sleep) from [<c0151ea0>] (__might_sleep+0x70/0xa8)
[<c0151ea0>] (__might_sleep) from [<c0bec2b8>] (mutex_lock+0x2c/0x70)
[<c0bec2b8>] (mutex_lock) from [<c0589844>] (tegra_powergate_is_powered+0x44/0xa8)
[<c0589844>] (tegra_powergate_is_powered) from [<c0581a60>] (tegra30_cpu_rail_off_ready+0x30/0x74)
[<c0581a60>] (tegra30_cpu_rail_off_ready) from [<c0122244>] (tegra30_idle_lp2+0xa0/0x108)
[<c0122244>] (tegra30_idle_lp2) from [<c0853438>] (cpuidle_enter_state+0x140/0x540)
[<c0853438>] (cpuidle_enter_state) from [<c08538a4>] (cpuidle_enter+0x40/0x4c)
[<c08538a4>] (cpuidle_enter) from [<c01595e0>] (call_cpuidle+0x30/0x48)
[<c01595e0>] (call_cpuidle) from [<c01599f8>] (do_idle+0x238/0x28c)
[<c01599f8>] (do_idle) from [<c0159d28>] (cpu_startup_entry+0x28/0x2c)
[<c0159d28>] (cpu_startup_entry) from [<c0be76c8>] (rest_init+0xd8/0xdc)
[<c0be76c8>] (rest_init) from [<c1200f50>] (start_kernel+0x41c/0x430)

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoiommu/dmar: Fix buffer overflow during PCI bus notification
Julia Cartwright [Wed, 20 Feb 2019 16:46:31 +0000 (16:46 +0000)]
iommu/dmar: Fix buffer overflow during PCI bus notification

[ Upstream commit cffaaf0c816238c45cd2d06913476c83eb50f682 ]

Commit 57384592c433 ("iommu/vt-d: Store bus information in RMRR PCI
device path") changed the type of the path data, however, the change in
path type was not reflected in size calculations.  Update to use the
correct type and prevent a buffer overflow.

This bug manifests in systems with deep PCI hierarchies, and can lead to
an overflow of the static allocated buffer (dmar_pci_notify_info_buf),
or can lead to overflow of slab-allocated data.

   BUG: KASAN: global-out-of-bounds in dmar_alloc_pci_notify_info+0x1d5/0x2e0
   Write of size 1 at addr ffffffff90445d80 by task swapper/0/1
   CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W       4.14.87-rt49-02406-gd0a0e96 #1
   Call Trace:
    ? dump_stack+0x46/0x59
    ? print_address_description+0x1df/0x290
    ? dmar_alloc_pci_notify_info+0x1d5/0x2e0
    ? kasan_report+0x256/0x340
    ? dmar_alloc_pci_notify_info+0x1d5/0x2e0
    ? e820__memblock_setup+0xb0/0xb0
    ? dmar_dev_scope_init+0x424/0x48f
    ? __down_write_common+0x1ec/0x230
    ? dmar_dev_scope_init+0x48f/0x48f
    ? dmar_free_unused_resources+0x109/0x109
    ? cpumask_next+0x16/0x20
    ? __kmem_cache_create+0x392/0x430
    ? kmem_cache_create+0x135/0x2f0
    ? e820__memblock_setup+0xb0/0xb0
    ? intel_iommu_init+0x170/0x1848
    ? _raw_spin_unlock_irqrestore+0x32/0x60
    ? migrate_enable+0x27a/0x5b0
    ? sched_setattr+0x20/0x20
    ? migrate_disable+0x1fc/0x380
    ? task_rq_lock+0x170/0x170
    ? try_to_run_init_process+0x40/0x40
    ? locks_remove_file+0x85/0x2f0
    ? dev_prepare_static_identity_mapping+0x78/0x78
    ? rt_spin_unlock+0x39/0x50
    ? lockref_put_or_lock+0x2a/0x40
    ? dput+0x128/0x2f0
    ? __rcu_read_unlock+0x66/0x80
    ? __fput+0x250/0x300
    ? __rcu_read_lock+0x1b/0x30
    ? mntput_no_expire+0x38/0x290
    ? e820__memblock_setup+0xb0/0xb0
    ? pci_iommu_init+0x25/0x63
    ? pci_iommu_init+0x25/0x63
    ? do_one_initcall+0x7e/0x1c0
    ? initcall_blacklisted+0x120/0x120
    ? kernel_init_freeable+0x27b/0x307
    ? rest_init+0xd0/0xd0
    ? kernel_init+0xf/0x120
    ? rest_init+0xd0/0xd0
    ? ret_from_fork+0x1f/0x40
   The buggy address belongs to the variable:
    dmar_pci_notify_info_buf+0x40/0x60

Fixes: 57384592c433 ("iommu/vt-d: Store bus information in RMRR PCI device path")
Signed-off-by: Julia Cartwright <julia@ni.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agocrypto: sha512/arm - fix crash bug in Thumb2 build
Ard Biesheuvel [Sat, 16 Feb 2019 13:51:26 +0000 (14:51 +0100)]
crypto: sha512/arm - fix crash bug in Thumb2 build

[ Upstream commit c64316502008064c158fa40cc250665e461b0f2a ]

The SHA512 code we adopted from the OpenSSL project uses a rather
peculiar way to take the address of the round constant table: it
takes the address of the sha256_block_data_order() routine, and
substracts a constant known quantity to arrive at the base of the
table, which is emitted by the same assembler code right before
the routine's entry point.

However, recent versions of binutils have helpfully changed the
behavior of references emitted via an ADR instruction when running
in Thumb2 mode: it now takes the Thumb execution mode bit into
account, which is bit 0 af the address. This means the produced
table address also has bit 0 set, and so we end up with an address
value pointing 1 byte past the start of the table, which results
in crashes such as

  Unable to handle kernel paging request at virtual address bf825000
  pgd = 42f44b11
  [bf825000] *pgd=80000040206003, *pmd=5f1bd003, *pte=00000000
  Internal error: Oops: 207 [#1] PREEMPT SMP THUMB2
  Modules linked in: sha256_arm(+) sha1_arm_ce sha1_arm ...
  CPU: 7 PID: 396 Comm: cryptomgr_test Not tainted 5.0.0-rc6+ #144
  Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015
  PC is at sha256_block_data_order+0xaaa/0xb30 [sha256_arm]
  LR is at __this_module+0x17fd/0xffffe800 [sha256_arm]
  pc : [<bf820bca>]    lr : [<bf824ffd>]    psr: 800b0033
  sp : ebc8bbe8  ip : faaabe1c  fp : 2fdd3433
  r10: 4c5f1692  r9 : e43037df  r8 : b04b0a5a
  r7 : c369d722  r6 : 39c3693e  r5 : 7a013189  r4 : 1580d26b
  r3 : 8762a9b0  r2 : eea9c2cd  r1 : 3e9ab536  r0 : 1dea4ae7
  Flags: Nzcv  IRQs on  FIQs on  Mode SVC_32  ISA Thumb  Segment user
  Control: 70c5383d  Table: 6b8467c0  DAC: dbadc0de
  Process cryptomgr_test (pid: 396, stack limit = 0x69e1fe23)
  Stack: (0xebc8bbe8 to 0xebc8c000)
  ...
  unwind: Unknown symbol address bf820bca
  unwind: Index not found bf820bca
  Code: 441a ea80 40f9 440a (f85e) 3b04
  ---[ end trace e560cce92700ef8a ]---

Given that this affects older kernels as well, in case they are built
with a recent toolchain, apply a minimal backportable fix, which is
to emit another non-code label at the start of the routine, and
reference that instead. (This is similar to the current upstream state
of this file in OpenSSL)

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agocrypto: sha256/arm - fix crash bug in Thumb2 build
Ard Biesheuvel [Sat, 16 Feb 2019 13:51:25 +0000 (14:51 +0100)]
crypto: sha256/arm - fix crash bug in Thumb2 build

[ Upstream commit 69216a545cf81b2b32d01948f7039315abaf75a0 ]

The SHA256 code we adopted from the OpenSSL project uses a rather
peculiar way to take the address of the round constant table: it
takes the address of the sha256_block_data_order() routine, and
substracts a constant known quantity to arrive at the base of the
table, which is emitted by the same assembler code right before
the routine's entry point.

However, recent versions of binutils have helpfully changed the
behavior of references emitted via an ADR instruction when running
in Thumb2 mode: it now takes the Thumb execution mode bit into
account, which is bit 0 af the address. This means the produced
table address also has bit 0 set, and so we end up with an address
value pointing 1 byte past the start of the table, which results
in crashes such as

  Unable to handle kernel paging request at virtual address bf825000
  pgd = 42f44b11
  [bf825000] *pgd=80000040206003, *pmd=5f1bd003, *pte=00000000
  Internal error: Oops: 207 [#1] PREEMPT SMP THUMB2
  Modules linked in: sha256_arm(+) sha1_arm_ce sha1_arm ...
  CPU: 7 PID: 396 Comm: cryptomgr_test Not tainted 5.0.0-rc6+ #144
  Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015
  PC is at sha256_block_data_order+0xaaa/0xb30 [sha256_arm]
  LR is at __this_module+0x17fd/0xffffe800 [sha256_arm]
  pc : [<bf820bca>]    lr : [<bf824ffd>]    psr: 800b0033
  sp : ebc8bbe8  ip : faaabe1c  fp : 2fdd3433
  r10: 4c5f1692  r9 : e43037df  r8 : b04b0a5a
  r7 : c369d722  r6 : 39c3693e  r5 : 7a013189  r4 : 1580d26b
  r3 : 8762a9b0  r2 : eea9c2cd  r1 : 3e9ab536  r0 : 1dea4ae7
  Flags: Nzcv  IRQs on  FIQs on  Mode SVC_32  ISA Thumb  Segment user
  Control: 70c5383d  Table: 6b8467c0  DAC: dbadc0de
  Process cryptomgr_test (pid: 396, stack limit = 0x69e1fe23)
  Stack: (0xebc8bbe8 to 0xebc8c000)
  ...
  unwind: Unknown symbol address bf820bca
  unwind: Index not found bf820bca
  Code: 441a ea80 40f9 440a (f85e) 3b04
  ---[ end trace e560cce92700ef8a ]---

Given that this affects older kernels as well, in case they are built
with a recent toolchain, apply a minimal backportable fix, which is
to emit another non-code label at the start of the routine, and
reference that instead. (This is similar to the current upstream state
of this file in OpenSSL)

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agokernel: hung_task.c: disable on suspend
Vitaly Kuznetsov [Wed, 17 Oct 2018 11:23:55 +0000 (13:23 +0200)]
kernel: hung_task.c: disable on suspend

[ Upstream commit a1c6ca3c6de763459a6e93b644ec6518c890ba1c ]

It is possible to observe hung_task complaints when system goes to
suspend-to-idle state:

 # echo freeze > /sys/power/state

 PM: Syncing filesystems ... done.
 Freezing user space processes ... (elapsed 0.001 seconds) done.
 OOM killer disabled.
 Freezing remaining freezable tasks ... (elapsed 0.002 seconds) done.
 sd 0:0:0:0: [sda] Synchronizing SCSI cache
 INFO: task bash:1569 blocked for more than 120 seconds.
       Not tainted 4.19.0-rc3_+ #687
 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
 bash            D    0  1569    604 0x00000000
 Call Trace:
  ? __schedule+0x1fe/0x7e0
  schedule+0x28/0x80
  suspend_devices_and_enter+0x4ac/0x750
  pm_suspend+0x2c0/0x310

Register a PM notifier to disable the detector on suspend and re-enable
back on wakeup.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agocifs: fallback to older infolevels on findfirst queryinfo retry
Steve French [Fri, 19 Oct 2018 06:58:22 +0000 (01:58 -0500)]
cifs: fallback to older infolevels on findfirst queryinfo retry

[ Upstream commit 3b7960caceafdfc2cdfe2850487f8d091eb41144 ]

In cases where queryinfo fails, we have cases in cifs (vers=1.0)
where with backupuid mounts we retry the query info with findfirst.
This doesn't work to some NetApp servers which don't support
WindowsXP (and later) infolevel 261 (SMB_FIND_FILE_ID_FULL_DIR_INFO)
so in this case use other info levels (in this case it will usually
be level 257, SMB_FIND_FILE_DIRECTORY_INFO).

(Also fixes some indentation)

See kernel bugzilla 201435

Signed-off-by: Steve French <stfrench@microsoft.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agocompiler.h: update definition of unreachable()
ndesaulniers@google.com [Mon, 15 Oct 2018 17:22:21 +0000 (10:22 -0700)]
compiler.h: update definition of unreachable()

[ Upstream commit fe0640eb30b7da261ae84d252ed9ed3c7e68dfd8 ]

Fixes the objtool warning seen with Clang:
arch/x86/mm/fault.o: warning: objtool: no_context()+0x220: unreachable
instruction

Fixes commit 815f0ddb346c ("include/linux/compiler*.h: make compiler-*.h
mutually exclusive")

Josh noted that the fallback definition was meant to work around a
pre-gcc-4.6 bug. GCC still needs to work around
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82365, so compiler-gcc.h
defines its own version of unreachable().  Clang and ICC can use this
shared definition.

Link: https://github.com/ClangBuiltLinux/linux/issues/204
Suggested-by: Andy Lutomirski <luto@amacapital.net>
Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com>
Tested-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoKVM: nVMX: restore host state in nested_vmx_vmexit for VMFail
Sean Christopherson [Wed, 22 Aug 2018 21:57:07 +0000 (14:57 -0700)]
KVM: nVMX: restore host state in nested_vmx_vmexit for VMFail

[ Upstream commit bd18bffca35397214ae68d85cf7203aca25c3c1d ]

A VMEnter that VMFails (as opposed to VMExits) does not touch host
state beyond registers that are explicitly noted in the VMFail path,
e.g. EFLAGS.  Host state does not need to be loaded because VMFail
is only signaled for consistency checks that occur before the CPU
starts to load guest state, i.e. there is no need to restore any
state as nothing has been modified.  But in the case where a VMFail
is detected by hardware and not by KVM (due to deferring consistency
checks to hardware), KVM has already loaded some amount of guest
state.  Luckily, "loaded" only means loaded to KVM's software model,
i.e. vmcs01 has not been modified.  So, unwind our software model to
the pre-VMEntry host state.

Not restoring host state in this VMFail path leads to a variety of
failures because we end up with stale data in vcpu->arch, e.g. CR0,
CR4, EFER, etc... will all be out of sync relative to vmcs01.  Any
significant delta in the stale data is all but guaranteed to crash
L1, e.g. emulation of SMEP, SMAP, UMIP, WP, etc... will be wrong.

An alternative to this "soft" reload would be to load host state from
vmcs12 as if we triggered a VMExit (as opposed to VMFail), but that is
wildly inconsistent with respect to the VMX architecture, e.g. an L1
VMM with separate VMExit and VMFail paths would explode.

Note that this approach does not mean KVM is 100% accurate with
respect to VMX hardware behavior, even at an architectural level
(the exact order of consistency checks is microarchitecture specific).
But 100% emulation accuracy isn't the goal (with this patch), rather
the goal is to be consistent in the information delivered to L1, e.g.
a VMExit should not fall-through VMENTER, and a VMFail should not jump
to HOST_RIP.

This technically reverts commit "5af4157388ad (KVM: nVMX: Fix mmu
context after VMLAUNCH/VMRESUME failure)", but retains the core
aspects of that patch, just in an open coded form due to the need to
pull state from vmcs01 instead of vmcs12.  Restoring host state
resolves a variety of issues introduced by commit "4f350c6dbcb9
(kvm: nVMX: Handle deferred early VMLAUNCH/VMRESUME failure properly)",
which remedied the incorrect behavior of treating VMFail like VMExit
but in doing so neglected to restore arch state that had been modified
prior to attempting nested VMEnter.

A sample failure that occurs due to stale vcpu.arch state is a fault
of some form while emulating an LGDT (due to emulated UMIP) from L1
after a failed VMEntry to L3, in this case when running the KVM unit
test test_tpr_threshold_values in L1.  L0 also hits a WARN in this
case due to a stale arch.cr4.UMIP.

L1:
  BUG: unable to handle kernel paging request at ffffc90000663b9e
  PGD 276512067 P4D 276512067 PUD 276513067 PMD 274efa067 PTE 8000000271de2163
  Oops: 0009 [#1] SMP
  CPU: 5 PID: 12495 Comm: qemu-system-x86 Tainted: G        W         4.18.0-rc2+ #2
  Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
  RIP: 0010:native_load_gdt+0x0/0x10

  ...

  Call Trace:
   load_fixmap_gdt+0x22/0x30
   __vmx_load_host_state+0x10e/0x1c0 [kvm_intel]
   vmx_switch_vmcs+0x2d/0x50 [kvm_intel]
   nested_vmx_vmexit+0x222/0x9c0 [kvm_intel]
   vmx_handle_exit+0x246/0x15a0 [kvm_intel]
   kvm_arch_vcpu_ioctl_run+0x850/0x1830 [kvm]
   kvm_vcpu_ioctl+0x3a1/0x5c0 [kvm]
   do_vfs_ioctl+0x9f/0x600
   ksys_ioctl+0x66/0x70
   __x64_sys_ioctl+0x16/0x20
   do_syscall_64+0x4f/0x100
   entry_SYSCALL_64_after_hwframe+0x44/0xa9

L0:
  WARNING: CPU: 2 PID: 3529 at arch/x86/kvm/vmx.c:6618 handle_desc+0x28/0x30 [kvm_intel]
  ...
  CPU: 2 PID: 3529 Comm: qemu-system-x86 Not tainted 4.17.2-coffee+ #76
  Hardware name: Intel Corporation Kabylake Client platform/KBL S
  RIP: 0010:handle_desc+0x28/0x30 [kvm_intel]

  ...

  Call Trace:
   kvm_arch_vcpu_ioctl_run+0x863/0x1840 [kvm]
   kvm_vcpu_ioctl+0x3a1/0x5c0 [kvm]
   do_vfs_ioctl+0x9f/0x5e0
   ksys_ioctl+0x66/0x70
   __x64_sys_ioctl+0x16/0x20
   do_syscall_64+0x49/0xf0
   entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: 5af4157388ad (KVM: nVMX: Fix mmu context after VMLAUNCH/VMRESUME failure)
Fixes: 4f350c6dbcb9 (kvm: nVMX: Handle deferred early VMLAUNCH/VMRESUME failure properly)
Cc: Jim Mattson <jmattson@google.com>
Cc: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoACPI / SBS: Fix GPE storm on recent MacBookPro's
Ronald Tschalär [Mon, 1 Oct 2018 02:52:51 +0000 (19:52 -0700)]
ACPI / SBS: Fix GPE storm on recent MacBookPro's

[ Upstream commit ca1721c5bee77105829cbd7baab8ee0eab85b06d ]

On Apple machines, plugging-in or unplugging the power triggers a GPE
for the EC. Since these machines expose an SBS device, this GPE ends
up triggering the acpi_sbs_callback(). This in turn tries to get the
status of the SBS charger. However, on MBP13,* and MBP14,* machines,
performing the smbus-read operation to get the charger's status triggers
the EC's GPE again. The result is an endless re-triggering and handling
of that GPE, consuming significant CPU resources (> 50% in irq).

In the end this is quite similar to commit 3031cddea633 (ACPI / SBS:
Don't assume the existence of an SBS charger), except that on the above
machines a status of all 1's is returned. And like there, we just want
ignore the charger here.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=198169
Signed-off-by: Ronald Tschalär <ronald@innovation.ch>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agousbip: fix vhci_hcd controller counting
Maciej Żenczykowski [Thu, 20 Sep 2018 20:29:42 +0000 (13:29 -0700)]
usbip: fix vhci_hcd controller counting

[ Upstream commit e0a2e73e501c77037c8756137e87b12c7c3c9793 ]

Without this usbip fails on a machine with devices
that lexicographically come after vhci_hcd.

ie.
  $ ls -l /sys/devices/platform
  ...
  drwxr-xr-x. 4 root root    0 Sep 19 16:21 serial8250
  -rw-r--r--. 1 root root 4096 Sep 19 23:50 uevent
  drwxr-xr-x. 6 root root    0 Sep 20 13:15 vhci_hcd.0
  drwxr-xr-x. 4 root root    0 Sep 19 16:22 w83627hf.656

Because it detects 'w83627hf.656' as another vhci_hcd controller,
and then fails to be able to talk to it.

Note: this doesn't actually fix usbip's support for multiple
controllers... that's still broken for other reasons
("vhci_hcd.0" is hardcoded in a string macro), but is enough to
actually make it work on the above machine.

See also:
  https://bugzilla.redhat.com/show_bug.cgi?id=1631148

Cc: Jonathan Dieter <jdieter@gmail.com>
Cc: Valentina Manea <valentina.manea.m@gmail.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: linux-usb@vger.kernel.org
Signed-off-by: Maciej Żenczykowski <zenczykowski@gmail.com>
Acked-by: Shuah Khan (Samsung OSG) <shuah@kernel.org>
Tested-by: Jonathan Dieter <jdieter@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoARM: samsung: Limit SAMSUNG_PM_CHECK config option to non-Exynos platforms
Bartlomiej Zolnierkiewicz [Fri, 28 Sep 2018 13:32:46 +0000 (15:32 +0200)]
ARM: samsung: Limit SAMSUNG_PM_CHECK config option to non-Exynos platforms

[ Upstream commit 6862fdf2201ab67cd962dbf0643d37db909f4860 ]

"S3C2410 PM Suspend Memory CRC" feature (controlled by
SAMSUNG_PM_CHECK config option) is incompatible with highmem
(uses phys_to_virt() instead of proper mapping) which is used by
the majority of Exynos boards. The issue manifests itself in OOPS
on affected boards, i.e. on Odroid-U3 I got the following one:

Unable to handle kernel paging request at virtual address f0000000
pgd = 1c0f9bb4
[f0000000] *pgd=00000000
Internal error: Oops: 5 [#1] PREEMPT SMP ARM
[<c0458034>] (crc32_le) from [<c0121f8c>] (s3c_pm_makecheck+0x34/0x54)
[<c0121f8c>] (s3c_pm_makecheck) from [<c0121efc>] (s3c_pm_run_res+0x74/0x8c)
[<c0121efc>] (s3c_pm_run_res) from [<c0121ecc>] (s3c_pm_run_res+0x44/0x8c)
[<c0121ecc>] (s3c_pm_run_res) from [<c01210b8>] (exynos_suspend_enter+0x64/0x148)
[<c01210b8>] (exynos_suspend_enter) from [<c018893c>] (suspend_devices_and_enter+0x9ec/0xe74)
[<c018893c>] (suspend_devices_and_enter) from [<c0189534>] (pm_suspend+0x770/0xc04)
[<c0189534>] (pm_suspend) from [<c0186ce8>] (state_store+0x6c/0xcc)
[<c0186ce8>] (state_store) from [<c09db434>] (kobj_attr_store+0x14/0x20)
[<c09db434>] (kobj_attr_store) from [<c02fa63c>] (sysfs_kf_write+0x4c/0x50)
[<c02fa63c>] (sysfs_kf_write) from [<c02f97a4>] (kernfs_fop_write+0xfc/0x1e4)
[<c02f97a4>] (kernfs_fop_write) from [<c027b198>] (__vfs_write+0x2c/0x140)
[<c027b198>] (__vfs_write) from [<c027b418>] (vfs_write+0xa4/0x160)
[<c027b418>] (vfs_write) from [<c027b5d8>] (ksys_write+0x40/0x8c)
[<c027b5d8>] (ksys_write) from [<c0101000>] (ret_fast_syscall+0x0/0x28)

Add PLAT_S3C24XX, ARCH_S3C64XX and ARCH_S5PV210 dependencies to
SAMSUNG_PM_CHECK config option to hide it on Exynos platforms.

Signed-off-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoHID: i2c-hid: override HID descriptors for certain devices
Julian Sax [Wed, 19 Sep 2018 09:46:23 +0000 (11:46 +0200)]
HID: i2c-hid: override HID descriptors for certain devices

[ Upstream commit 9ee3e06610fdb8a601cde59c92089fb6c1deb4aa ]

A particular touchpad (SIPODEV SP1064) refuses to supply the HID
descriptors. This patch provides the framework for overriding these
descriptors based on DMI data. It also includes the descriptors for
said touchpad, which were extracted by listening to the traffic of the
windows filter driver, as well as the DMI data for the laptops known
to use this device.

Relevant Bug: https://bugzilla.redhat.com/show_bug.cgi?id=1526312

Cc: Hans de Goede <hdegoede@redhat.com>
Reported-and-tested-by: ahormann@gmx.net
Reported-and-tested-by: Bruno Jesus <bruno.fl.jesus@gmail.com>
Reported-and-tested-by: Dietrich <enaut.w@googlemail.com>
Reported-and-tested-by: kloxdami@yahoo.com
Signed-off-by: Julian Sax <jsbc@gmx.de>
Reviewed-by: Benjamin Tissoires <benjamin.tissoires@redhat.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agomedia: au0828: cannot kfree dev before usb disconnect
Brad Love [Thu, 6 Sep 2018 21:07:48 +0000 (17:07 -0400)]
media: au0828: cannot kfree dev before usb disconnect

[ Upstream commit 4add7104919f9e94e0db03e234caeadbfcc02ea9 ]

If au0828_analog_register fails, the dev is kfree'd and then flow
jumps to done, which can call au0828_usb_disconnect. Since all USB
error codes are negative, au0828_usb_disconnect will be called. The
problem is au0828_usb_disconnect uses dev, if dev is NULL then there
is immediate oops encountered.

[    7.454307] au0828: au0828_usb_probe() au0282_dev_register failed to register on V4L2
[    7.454323] BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
[    7.454421] PGD 0 P4D 0
[    7.454457] Oops: 0002 [#1] SMP PTI
[    7.454500] CPU: 1 PID: 262 Comm: systemd-udevd Tainted: P           O      4.18.3 #1
[    7.454584] Hardware name: Google Panther/Panther, BIOS MattDevo 04/27/2015
[    7.454670] RIP: 0010:_raw_spin_lock_irqsave+0x2c/0x50
[    7.454725] Code: 44 00 00 55 48 89 e5 41 54 53 48 89 fb 9c 58 0f 1f 44 00 00 49 89 c4 fa 66 0f 1f 44 00 00 e8 db 23 1b ff 31 c0 ba 01 00 00 00 <f0> 0f b1 13 85 c0 75 08 4c 89 e0 5b 41 5c 5d c3 89 c6 48 89 df e8
[    7.455004] RSP: 0018:ffff9130f53ef988 EFLAGS: 00010046
[    7.455063] RAX: 0000000000000000 RBX: 0000000000000050 RCX: 0000000000000000
[    7.455139] RDX: 0000000000000001 RSI: 0000000000000003 RDI: 0000000000000050
[    7.455216] RBP: ffff9130f53ef998 R08: 0000000000000018 R09: 0000000000000090
[    7.455292] R10: ffffed4cc53cb000 R11: ffffed4cc53cb108 R12: 0000000000000082
[    7.455369] R13: ffff9130cf2c6188 R14: 0000000000000000 R15: 0000000000000018
[    7.455447] FS:  00007f2ff8514cc0(0000) GS:ffff9130fcb00000(0000) knlGS:0000000000000000
[    7.455535] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    7.455597] CR2: 0000000000000050 CR3: 00000001753f0002 CR4: 00000000000606a0
[    7.455675] Call Trace:
[    7.455713]  __wake_up_common_lock+0x65/0xc0
[    7.455764]  __wake_up+0x13/0x20
[    7.455808]  ir_lirc_unregister+0x57/0xe0 [rc_core]
[    7.455865]  rc_unregister_device+0xa0/0xc0 [rc_core]
[    7.455935]  au0828_rc_unregister+0x25/0x40 [au0828]
[    7.455999]  au0828_usb_disconnect+0x33/0x80 [au0828]
[    7.456064]  au0828_usb_probe.cold.16+0x8d/0x2aa [au0828]
[    7.456130]  usb_probe_interface+0xf1/0x300
[    7.456184]  driver_probe_device+0x2e3/0x460
[    7.456235]  __driver_attach+0xe4/0x110
[    7.456282]  ? driver_probe_device+0x460/0x460
[    7.456335]  bus_for_each_dev+0x74/0xb0
[    7.456385]  ? kmem_cache_alloc_trace+0x15d/0x1d0
[    7.456441]  driver_attach+0x1e/0x20
[    7.456485]  bus_add_driver+0x159/0x230
[    7.456532]  driver_register+0x70/0xc0
[    7.456578]  usb_register_driver+0x7f/0x140
[    7.456626]  ? 0xffffffffc0474000
[    7.456674]  au0828_init+0xbc/0x1000 [au0828]
[    7.456725]  do_one_initcall+0x4a/0x1c9
[    7.456771]  ? _cond_resched+0x19/0x30
[    7.456817]  ? kmem_cache_alloc_trace+0x15d/0x1d0
[    7.456873]  do_init_module+0x60/0x210
[    7.456918]  load_module+0x221b/0x2710
[    7.456966]  ? vfs_read+0xf5/0x120
[    7.457010]  __do_sys_finit_module+0xbd/0x120
[    7.457061]  ? __do_sys_finit_module+0xbd/0x120
[    7.457115]  __x64_sys_finit_module+0x1a/0x20
[    7.457166]  do_syscall_64+0x5b/0x110
[    7.457210]  entry_SYSCALL_64_after_hwframe+0x49/0xbe

Signed-off-by: Brad Love <brad@nextdimension.cc>
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agopowerpc/pseries: Remove prrn_work workqueue
Nathan Fontenot [Mon, 10 Sep 2018 14:57:00 +0000 (09:57 -0500)]
powerpc/pseries: Remove prrn_work workqueue

[ Upstream commit cd24e457fd8b2d087d9236700c8d2957054598bf ]

When a PRRN event is received we are already running in a worker
thread. Instead of spawning off another worker thread on the prrn_work
workqueue to handle the PRRN event we can just call the PRRN handler
routine directly.

With this update we can also pass the scope variable for the PRRN
event directly to the handler instead of it being a global variable.

This patch fixes the following oops mnessage we are seeing in PRRN testing:

  Oops: Bad kernel stack pointer, sig: 6 [#1]
  SMP NR_CPUS=2048 NUMA pSeries
  Modules linked in: nfsv3 nfs_acl rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver nfs lockd grace sunrpc fscache binfmt_misc reiserfs vfat fat rpadlpar_io(X) rpaphp(X) tcp_diag udp_diag inet_diag unix_diag af_packet_diag netlink_diag af_packet xfs libcrc32c dm_service_time ibmveth(X) ses enclosure scsi_transport_sas rtc_generic btrfs xor raid6_pq sd_mod ibmvscsi(X) scsi_transport_srp ipr(X) libata sg dm_multipath dm_mod scsi_dh_rdac scsi_dh_emc scsi_dh_alua scsi_mod autofs4
  Supported: Yes, External                                                     54
  CPU: 7 PID: 18967 Comm: kworker/u96:0 Tainted: G                 X 4.4.126-94.22-default #1
  Workqueue: pseries hotplug workque pseries_hp_work_fn
  task: c000000775367790 ti: c00000001ebd4000 task.ti: c00000070d140000
  NIP: 0000000000000000 LR: 000000001fb3d050 CTR: 0000000000000000
  REGS: c00000001ebd7d40 TRAP: 0700   Tainted: G                 X  (4.4.126-94.22-default)
  MSR: 8000000102081000 <41,VEC,ME5  CR: 28000002  XER: 20040018   4
  CFAR: 000000001fb3d084 40 419   1                                3
  GPR00: 000000000000000040000000000010007 000000001ffff400 000000041fffe200
  GPR04: 000000000000008050000000000000000 000000001fb15fa8 0000000500000500
  GPR08: 000000000001f40040000000000000001 0000000000000000 000005:5200040002
  GPR12: 00000000000000005c000000007a05400 c0000000000e89f8 000000001ed9f668
  GPR16: 000000001fbeff944000000001fbeff94 000000001fb545e4 0000006000000060
  GPR20: ffffffffffffffff4ffffffffffffffff 0000000000000000 0000000000000000
  GPR24: 00000000000000005400000001fb3c000 0000000000000000 000000001fb1b040
  GPR28: 000000001fb240004000000001fb440d8 0000000000000008 0000000000000000
  NIP [0000000000000000] 5         (null)
  LR [000000001fb3d050031fb3d050
  Call Trace:            4
  Instruction dump:      4                                       5:47 12    2
  XXXXXXXX XXXXXXXX XXXXX4XX XXXXXXXX XXXXXXXX XXXXXXXX XXXXXXXX XXXXXXXX
  XXXXXXXX XXXXXXXX XXXXX5XX XXXXXXXX 60000000 60000000 60000000 60000000
  ---[ end trace aa5627b04a7d9d6b ]---                                       3NMI watchdog: BUG: soft lockup - CPU#27 stuck for 23s! [kworker/27:0:13903]
  Modules linked in: nfsv3 nfs_acl rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver nfs lockd grace sunrpc fscache binfmt_misc reiserfs vfat fat rpadlpar_io(X) rpaphp(X) tcp_diag udp_diag inet_diag unix_diag af_packet_diag netlink_diag af_packet xfs libcrc32c dm_service_time ibmveth(X) ses enclosure scsi_transport_sas rtc_generic btrfs xor raid6_pq sd_mod ibmvscsi(X) scsi_transport_srp ipr(X) libata sg dm_multipath dm_mod scsi_dh_rdac scsi_dh_emc scsi_dh_alua scsi_mod autofs4
  Supported: Yes, External
  CPU: 27 PID: 13903 Comm: kworker/27:0 Tainted: G      D          X 4.4.126-94.22-default #1
  Workqueue: events prrn_work_fn
  task: c000000747cfa390 ti: c00000074712c000 task.ti: c00000074712c000
  NIP: c0000000008002a8 LR: c000000000090770 CTR: 000000000032e088
  REGS: c00000074712f7b0 TRAP: 0901   Tainted: G      D          X  (4.4.126-94.22-default)
  MSR: 8000000100009033 <SF,EE,ME,IR,DR,RI,LE>  CR: 22482044  XER: 20040000
  CFAR: c0000000008002c4 SOFTE: 1
  GPR00: c000000000090770 c00000074712fa30 c000000000f09800 c000000000fa1928 6:02
  GPR04: c000000775f5e000 fffffffffffffffe 0000000000000001 c000000000f42db8
  GPR08: 0000000000000001 0000000080000007 0000000000000000 0000000000000000
  GPR12: 8006210083180000 c000000007a14400
  NIP [c0000000008002a8] _raw_spin_lock+0x68/0xd0
  LR [c000000000090770] mobility_rtas_call+0x50/0x100
  Call Trace:            59                                        5
  [c00000074712fa60] [c000000000090770] mobility_rtas_call+0x50/0x100
  [c00000074712faf0] [c000000000090b08] pseries_devicetree_update+0xf8/0x530
  [c00000074712fc20] [c000000000031ba4] prrn_work_fn+0x34/0x50
  [c00000074712fc40] [c0000000000e0390] process_one_work+0x1a0/0x4e0
  [c00000074712fcd0] [c0000000000e0870] worker_thread+0x1a0/0x6105:57       2
  [c00000074712fd80] [c0000000000e8b18] kthread+0x128/0x150
  [c00000074712fe30] [c0000000000096f8] ret_from_kernel_thread+0x5c/0x64
  Instruction dump:
  2c090000 40c20010 7d40192d 40c2fff0 7c2004ac 2fa90000 40de0018 5:540030   3
  e8010010 ebe1fff8 7c0803a6 4e800020 <7c210b78e92d0000 89290009 792affe3

Signed-off-by: John Allen <jallen@linux.ibm.com>
Signed-off-by: Haren Myneni <haren@us.ibm.com>
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoserial: uartps: console_setup() can't be placed to init section
Michal Simek [Mon, 3 Sep 2018 13:10:49 +0000 (15:10 +0200)]
serial: uartps: console_setup() can't be placed to init section

[ Upstream commit 4bb1ce2350a598502b23088b169e16b43d4bc639 ]

When console device is rebinded, console_setup() is called again.
But marking it as __init means that function will be clear after boot is
complete. If console device is binded again console_setup() is not found
and error "Unable to handle kernel paging request at virtual address"
is reported.

Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agonetfilter: xt_cgroup: shrink size of v2 path
Pablo Neira Ayuso [Tue, 4 Sep 2018 10:07:55 +0000 (12:07 +0200)]
netfilter: xt_cgroup: shrink size of v2 path

[ Upstream commit 0d704967f4a49cc2212350b3e4a8231f8b4283ed ]

cgroup v2 path field is PATH_MAX which is too large, this is placing too
much pressure on memory allocation for people with many rules doing
cgroup v1 classid matching, side effects of this are bug reports like:

https://bugzilla.kernel.org/show_bug.cgi?id=200639

This patch registers a new revision that shrinks the cgroup path to 512
bytes, which is the same approach we follow in similar extensions that
have a path field.

Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agof2fs: fix to do sanity check with current segment number
Chao Yu [Thu, 6 Sep 2018 12:34:12 +0000 (20:34 +0800)]
f2fs: fix to do sanity check with current segment number

[ Upstream commit 042be0f849e5fc24116d0afecfaf926eed5cac63 ]

https://bugzilla.kernel.org/show_bug.cgi?id=200219

Reproduction way:
- mount image
- run poc code
- umount image

F2FS-fs (loop1): Bitmap was wrongly set, blk:15364
------------[ cut here ]------------
kernel BUG at /home/yuchao/git/devf2fs/segment.c:2061!
invalid opcode: 0000 [#1] PREEMPT SMP
CPU: 2 PID: 17686 Comm: umount Tainted: G        W  O      4.18.0-rc2+ #39
Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
EIP: update_sit_entry+0x459/0x4e0 [f2fs]
Code: e8 1c b5 fd ff 0f 0b 0f 0b 8b 45 e4 c7 44 24 08 9c 7a 6c f8 c7 44 24 04 bc 4a 6c f8 89 44 24 0c 8b 06 89 04 24 e8 f7 b4 fd ff <0f> 0b 8b 45 e4 0f b6 d2 89 54 24 10 c7 44 24 08 60 7a 6c f8 c7 44
EAX: 00000032 EBX: 000000f8 ECX: 00000002 EDX: 00000001
ESI: d7177000 EDI: f520fe68 EBP: d6477c6c ESP: d6477c34
DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068 EFLAGS: 00010282
CR0: 80050033 CR2: b7fbe000 CR3: 2a99b3c0 CR4: 000406f0
Call Trace:
 f2fs_allocate_data_block+0x124/0x580 [f2fs]
 do_write_page+0x78/0x150 [f2fs]
 f2fs_do_write_node_page+0x25/0xa0 [f2fs]
 __write_node_page+0x2bf/0x550 [f2fs]
 f2fs_sync_node_pages+0x60e/0x6d0 [f2fs]
 ? sync_inode_metadata+0x2f/0x40
 ? f2fs_write_checkpoint+0x28f/0x7d0 [f2fs]
 ? up_write+0x1e/0x80
 f2fs_write_checkpoint+0x2a9/0x7d0 [f2fs]
 ? mark_held_locks+0x5d/0x80
 ? _raw_spin_unlock_irq+0x27/0x50
 kill_f2fs_super+0x68/0x90 [f2fs]
 deactivate_locked_super+0x3d/0x70
 deactivate_super+0x40/0x60
 cleanup_mnt+0x39/0x70
 __cleanup_mnt+0x10/0x20
 task_work_run+0x81/0xa0
 exit_to_usermode_loop+0x59/0xa7
 do_fast_syscall_32+0x1f5/0x22c
 entry_SYSENTER_32+0x53/0x86
EIP: 0xb7f95c51
Code: c1 1e f7 ff ff 89 e5 8b 55 08 85 d2 8b 81 64 cd ff ff 74 02 89 02 5d c3 8b 0c 24 c3 8b 1c 24 c3 90 51 52 55 89 e5 0f 34 cd 80 <5d> 5a 59 c3 90 90 90 90 8d 76 00 58 b8 77 00 00 00 cd 80 90 8d 76
EAX: 00000000 EBX: 0871ab90 ECX: bfb2cd00 EDX: 00000000
ESI: 00000000 EDI: 0871ab90 EBP: 0871ab90 ESP: bfb2cd7c
DS: 007b ES: 007b FS: 0000 GS: 0033 SS: 007b EFLAGS: 00000246
Modules linked in: f2fs(O) crc32_generic bnep rfcomm bluetooth ecdh_generic snd_intel8x0 snd_ac97_codec ac97_bus snd_pcm snd_seq_midi snd_seq_midi_event snd_rawmidi snd_seq pcbc joydev aesni_intel snd_seq_device aes_i586 snd_timer crypto_simd snd cryptd soundcore mac_hid serio_raw video i2c_piix4 parport_pc ppdev lp parport hid_generic psmouse usbhid hid e1000 [last unloaded: f2fs]
---[ end trace d423f83982cfcdc5 ]---

The reason is, different log headers using the same segment, once
one log's next block address is used by another log, it will cause
panic as above.

Main area: 24 segs, 24 secs 24 zones
  - COLD  data: 0, 0, 0
  - WARM  data: 1, 1, 1
  - HOT   data: 20, 20, 20
  - Dir   dnode: 22, 22, 22
  - File   dnode: 22, 22, 22
  - Indir nodes: 21, 21, 21

So this patch adds sanity check to detect such condition to avoid
this issue.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years ago9p locks: add mount option for lock retry interval
Dinu-Razvan Chis-Serban [Wed, 5 Sep 2018 07:44:12 +0000 (16:44 +0900)]
9p locks: add mount option for lock retry interval

[ Upstream commit 5e172f75e51e3de1b4274146d9b990f803cb5c2a ]

The default P9_LOCK_TIMEOUT can be too long for some users exporting
a local file system to a guest VM (30s), make this configurable at
mount time.

Link: http://lkml.kernel.org/r/1536295827-3181-1-git-send-email-asmadeus@codewreck.org
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=195727
Signed-off-by: Dinu-Razvan Chis-Serban <justcsdr@gmail.com>
Signed-off-by: Dominique Martinet <dominique.martinet@cea.fr>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years ago9p: do not trust pdu content for stat item size
Gertjan Halkes [Wed, 5 Sep 2018 06:41:29 +0000 (15:41 +0900)]
9p: do not trust pdu content for stat item size

[ Upstream commit 2803cf4379ed252894f046cb8812a48db35294e3 ]

v9fs_dir_readdir() could deadloop if a struct was sent with a size set
to -2

Link: http://lkml.kernel.org/r/1536134432-11997-1-git-send-email-asmadeus@codewreck.org
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=88021
Signed-off-by: Gertjan Halkes <gertjan@google.com>
Signed-off-by: Dominique Martinet <dominique.martinet@cea.fr>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agorsi: improve kernel thread handling to fix kernel panic
Siva Rebbagondla [Mon, 27 Aug 2018 11:35:15 +0000 (17:05 +0530)]
rsi: improve kernel thread handling to fix kernel panic

[ Upstream commit 4c62764d0fc21a34ffc44eec1210038c3a2e4473 ]

While running regressions, observed below kernel panic when sdio disconnect
called. This is because of, kthread_stop() is taking care of
wait_for_completion() by default. When wait_for_completion triggered
in kthread_stop and as it was done already, giving kernel panic.
Hence, removing redundant wait_for_completion() from rsi_kill_thread().

... skipping ...
BUG: unable to handle kernel NULL pointer dereference at           (null)
IP: [<ffffffff810a63df>] exit_creds+0x1f/0x50
PGD 0
Oops: 0002 [#1] SMP
CPU: 0 PID: 6502 Comm: rmmod Tainted: G  OE   4.15.9-Generic #154-Ubuntu
Hardware name: Dell Inc. Edge Gateway 3003/ , BIOS 01.00.00 04/17/2017
Stack:
ffff88007392e600 ffff880075847dc0 ffffffff8108160a 0000000000000000
ffff88007392e600 ffff880075847de8 ffffffff810a484b ffff880076127000
ffff88003cd3a800 ffff880074f12a00 ffff880075847e28 ffffffffc09bed15
Call Trace:
[<ffffffff8108160a>] __put_task_struct+0x5a/0x140
[<ffffffff810a484b>] kthread_stop+0x10b/0x110
[<ffffffffc09bed15>] rsi_disconnect+0x2f5/0x300 [ven_rsi_sdio]
[<ffffffff81578bcb>] ? __pm_runtime_resume+0x5b/0x80
[<ffffffff816f0918>] sdio_bus_remove+0x38/0x100
[<ffffffff8156cc64>] __device_release_driver+0xa4/0x150
[<ffffffff8156d7a5>] driver_detach+0xb5/0xc0
[<ffffffff8156c6c5>] bus_remove_driver+0x55/0xd0
[<ffffffff8156dfbc>] driver_unregister+0x2c/0x50
[<ffffffff816f0b8a>] sdio_unregister_driver+0x1a/0x20
[<ffffffffc09bf0f5>] rsi_module_exit+0x15/0x30 [ven_rsi_sdio]
[<ffffffff8110cad8>] SyS_delete_module+0x1b8/0x210
[<ffffffff81851dc8>] entry_SYSCALL_64_fastpath+0x1c/0xbb

Signed-off-by: Siva Rebbagondla <siva.rebbagondla@redpinesignals.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agogpio: pxa: handle corner case of unprobed device
Robert Jarzmik [Sat, 25 Aug 2018 08:44:17 +0000 (10:44 +0200)]
gpio: pxa: handle corner case of unprobed device

[ Upstream commit 9ce3ebe973bf4073426f35f282c6b955ed802765 ]

In the corner case where the gpio driver probe fails, for whatever
reason, the suspend and resume handlers will still be called as they
have to be registered as syscore operations. This applies as well when
no probe was called while the driver has been built in the kernel.

Nicolas tracked this in :
https://bugzilla.kernel.org/show_bug.cgi?id=200905

Therefore, add a failsafe in these function, and test if a proper probe
succeeded and the driver is functional.

Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Reported-by: Nicolas Chauvet <kwizart@gmail.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoext4: prohibit fstrim in norecovery mode
Darrick J. Wong [Sat, 23 Mar 2019 16:10:29 +0000 (12:10 -0400)]
ext4: prohibit fstrim in norecovery mode

[ Upstream commit 18915b5873f07e5030e6fb108a050fa7c71c59fb ]

The ext4 fstrim implementation uses the block bitmaps to find free space
that can be discarded.  If we haven't replayed the journal, the bitmaps
will be stale and we absolutely *cannot* use stale metadata to zap the
underlying storage.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agofix incorrect error code mapping for OBJECTID_NOT_FOUND
Steve French [Sun, 17 Mar 2019 20:58:38 +0000 (15:58 -0500)]
fix incorrect error code mapping for OBJECTID_NOT_FOUND

[ Upstream commit 85f9987b236cf46e06ffdb5c225cf1f3c0acb789 ]

It was mapped to EIO which can be confusing when user space
queries for an object GUID for an object for which the server
file system doesn't support (or hasn't saved one).

As Amir Goldstein suggested this is similar to ENOATTR
(equivalently ENODATA in Linux errno definitions) so
changing NT STATUS code mapping for OBJECTID_NOT_FOUND
to ENODATA.

Signed-off-by: Steve French <stfrench@microsoft.com>
CC: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agox86/hw_breakpoints: Make default case in hw_breakpoint_arch_parse() return an error
Nathan Chancellor [Thu, 7 Mar 2019 21:27:56 +0000 (14:27 -0700)]
x86/hw_breakpoints: Make default case in hw_breakpoint_arch_parse() return an error

[ Upstream commit e898e69d6b9475bf123f99b3c5d1a67bb7cb2361 ]

When building with -Wsometimes-uninitialized, Clang warns:

arch/x86/kernel/hw_breakpoint.c:355:2: warning: variable 'align' is used
uninitialized whenever switch default is taken
[-Wsometimes-uninitialized]

The default cannot be reached because arch_build_bp_info() initializes
hw->len to one of the specified cases. Nevertheless the warning is valid
and returning -EINVAL makes sure that this cannot be broken by future
modifications.

Suggested-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: clang-built-linux@googlegroups.com
Link: https://github.com/ClangBuiltLinux/linux/issues/392
Link: https://lkml.kernel.org/r/20190307212756.4648-1-natechancellor@gmail.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
6 years agoiommu/vt-d: Check capability before disabling protected memory
Lu Baolu [Wed, 20 Mar 2019 01:58:33 +0000 (09:58 +0800)]
iommu/vt-d: Check capability before disabling protected memory

[ Upstream commit 5bb71fc790a88d063507dc5d445ab8b14e845591 ]

The spec states in 10.4.16 that the Protected Memory Enable
Register should be treated as read-only for implementations
not supporting protected memory regions (PLMR and PHMR fields
reported as Clear in the Capability register).

Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: mark gross <mgross@intel.com>
Suggested-by: Ashok Raj <ashok.raj@intel.com>
Fixes: f8bab73515ca5 ("intel-iommu: PMEN support")
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>