6 years agopagemap: hide physical addresses from non-privileged users
Konstantin Khlebnikov [Tue, 8 Sep 2015 22:00:07 +0000 (15:00 -0700)]
pagemap: hide physical addresses from non-privileged users

commit 1c90308e7a77af6742a97d1021cca923b23b7f0d upstream.

This patch makes pagemap readable for normal users and hides physical
addresses from them.  For some use-cases PFN isn't required at all.


Fixes: ab676b7d6fbf ("pagemap: do not leak physical addresses to non-privileged userspace")
Signed-off-by: Konstantin Khlebnikov <>
Cc: Naoya Horiguchi <>
Reviewed-by: Mark Williamson <>
Tested-by: Mark Williamson <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
[bwh: Backported to 3.2:
 - Add the same check in the places where we look up a PFN
 - Add struct pagemapread * parameters where necessary
 - Open-code file_ns_capable()
 - Delete pagemap_open() entirely, as it would always return 0]
Signed-off-by: Ben Hutchings <>
6 years agoARM: 8429/1: disable GCC SRA optimization
Ard Biesheuvel [Thu, 3 Sep 2015 12:24:40 +0000 (13:24 +0100)]
ARM: 8429/1: disable GCC SRA optimization

commit a077224fd35b2f7fbc93f14cf67074fc792fbac2 upstream.

While working on the 32-bit ARM port of UEFI, I noticed a strange
corruption in the kernel log. The following snprintf() statement
(in drivers/firmware/efi/efi.c:efi_md_typeattr_format())

snprintf(pos, size, "|%3s|%2s|%2s|%2s|%3s|%2s|%2s|%2s|%2s]",

was producing the following output in the log:

|    |   |   |   |    |WB|WT|WC|UC]
|    |   |   |   |    |WB|WT|WC|UC]
|    |   |   |   |    |WB|WT|WC|UC]
|RUN|   |   |   |    |WB|WT|WC|UC]*
|RUN|   |   |   |    |WB|WT|WC|UC]*
|    |   |   |   |    |WB|WT|WC|UC]
|RUN|   |   |   |    |WB|WT|WC|UC]*
|    |   |   |   |    |WB|WT|WC|UC]
|RUN|   |   |   |    |   |   |   |UC]
|RUN|   |   |   |    |   |   |   |UC]

As it turns out, this is caused by incorrect code being emitted for
the string() function in lib/vsprintf.c. The following code

if (!(spec.flags & LEFT)) {
while (len < spec.field_width--) {
if (buf < end)
*buf = ' ';
for (i = 0; i < len; ++i) {
if (buf < end)
*buf = *s;
++buf; ++s;
while (len < spec.field_width--) {
if (buf < end)
*buf = ' ';

when called with len == 0, triggers an issue in the GCC SRA optimization
pass (Scalar Replacement of Aggregates), which handles promotion of signed
struct members incorrectly. This is a known but as yet unresolved issue.
( In this particular
case, it is causing the second while loop to be executed erroneously a
single time, causing the additional space characters to be printed.

So disable the optimization by passing -fno-ipa-sra.

Acked-by: Nicolas Pitre <>
Signed-off-by: Ard Biesheuvel <>
Signed-off-by: Russell King <>
Signed-off-by: Ben Hutchings <>
6 years agofs: create and use seq_show_option for escaping
Kees Cook [Fri, 4 Sep 2015 22:44:57 +0000 (15:44 -0700)]
fs: create and use seq_show_option for escaping

commit a068acf2ee77693e0bf39d6e07139ba704f461c3 upstream.

Many file systems that implement the show_options hook fail to correctly
escape their output which could lead to unescaped characters (e.g.  new
lines) leaking into /proc/mounts and /proc/[pid]/mountinfo files.  This
could lead to confusion, spoofed entries (resulting in things like
systemd issuing false d-bus "mount" notifications), and who knows what
else.  This looks like it would only be the root user stepping on
themselves, but it's possible weird things could happen in containers or
in other situations with delegated mount privileges.

Here's an example using overlay with setuid fusermount trusting the
contents of /proc/mounts (via the /etc/mtab symlink).  Imagine the use
of "sudo" is something more sneaky:

  $ BASE="ovl"
  $ MNT="$BASE/mnt"
  $ LOW="$BASE/lower"
  $ UP="$BASE/upper"
  $ WORK="$BASE/work/ 0 0
  none /proc fuse.pwn user_id=1000"
  $ mkdir -p "$LOW" "$UP" "$WORK"
  $ sudo mount -t overlay -o "lowerdir=$LOW,upperdir=$UP,workdir=$WORK" none /mnt
  $ cat /proc/mounts
  none /root/ovl/mnt overlay rw,relatime,lowerdir=ovl/lower,upperdir=ovl/upper,workdir=ovl/work/ 0 0
  none /proc fuse.pwn user_id=1000 0 0
  $ fusermount -u /proc
  $ cat /proc/mounts
  cat: /proc/mounts: No such file or directory

This fixes the problem by adding new seq_show_option and
seq_show_option_n helpers, and updating the vulnerable show_option
handlers to use them as needed.  Some, like SELinux, need to be open
coded due to unusual existing escape mechanisms.

[ add lost chunk, per Kees]
[ seq_show_option should be using const parameters]
Signed-off-by: Kees Cook <>
Acked-by: Serge Hallyn <>
Acked-by: Jan Kara <>
Acked-by: Paul Moore <>
Cc: J. R. Okajima <>
Signed-off-by: Kees Cook <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
[bwh: Backported to 3.2:
 - Drop changes to overlayfs, reiserfs
 - Drop vers option from cifs
 - ceph changes are all in one file
 - Adjust context]
Signed-off-by: Ben Hutchings <>
6 years agocrypto: ghash-clmulni: specify context size for ghash async algorithm
Andrey Ryabinin [Thu, 3 Sep 2015 11:32:01 +0000 (14:32 +0300)]
crypto: ghash-clmulni: specify context size for ghash async algorithm

commit 71c6da846be478a61556717ef1ee1cea91f5d6a8 upstream.

Currently context size (cra_ctxsize) doesn't specified for
ghash_async_alg. Which means it's zero. Thus crypto_create_tfm()
doesn't allocate needed space for ghash_async_ctx, so any
read/write to ctx (e.g. in ghash_async_init_tfm()) is not valid.

Signed-off-by: Andrey Ryabinin <>
Signed-off-by: Herbert Xu <>
Signed-off-by: Ben Hutchings <>
6 years agoInput: evdev - do not report errors form flush()
Takashi Iwai [Fri, 4 Sep 2015 05:20:00 +0000 (22:20 -0700)]
Input: evdev - do not report errors form flush()

commit eb38f3a4f6e86f8bb10a3217ebd85ecc5d763aae upstream.

We've got bug reports showing the old systemd-logind (at least
system-210) aborting unexpectedly, and this turned out to be because
of an invalid error code from close() call to evdev devices.  close()
is supposed to return only either EINTR or EBADFD, while the device
returned ENODEV.  logind was overreacting to it and decided to kill
itself when an unexpected error code was received.  What a tragedy.

The bad error code comes from flush fops, and actually evdev_flush()
returns ENODEV when device is disconnected or client's access to it is
revoked. But in these cases the fact that flush did not actually happen is
not an error, but rather normal behavior. For non-disconnected devices
result of flush is also not that interesting as there is no potential of
data loss and even if it fails application has no way of handling the
error. Because of that we are better off always returning success from

Also returning EINTR from flush()/close() is discouraged (as it is not
clear how application should handle this error), so let's stop taking
evdev->mutex interruptibly.

Signed-off-by: Takashi Iwai <>
Signed-off-by: Dmitry Torokhov <>
[bwh: Backported to 3.2: there's no revoked flag to test]
Signed-off-by: Ben Hutchings <>
6 years agoIB/uverbs: reject invalid or unknown opcodes
Christoph Hellwig [Wed, 26 Aug 2015 09:00:37 +0000 (11:00 +0200)]
IB/uverbs: reject invalid or unknown opcodes

commit b632ffa7cee439ba5dce3b3bc4a5cbe2b3e20133 upstream.

We have many WR opcodes that are only supported in kernel space
and/or require optional information to be copied into the WR
structure.  Reject all those not explicitly handled so that we
can't pass invalid information to drivers.

Signed-off-by: Christoph Hellwig <>
Reviewed-by: Jason Gunthorpe <>
Reviewed-by: Sagi Grimberg <>
Signed-off-by: Doug Ledford <>
Signed-off-by: Ben Hutchings <>
6 years agoAdd radeon suspend/resume quirk for HP Compaq dc5750.
Jeffery Miller [Tue, 1 Sep 2015 15:23:02 +0000 (11:23 -0400)]
Add radeon suspend/resume quirk for HP Compaq dc5750.

commit 09bfda10e6efd7b65bcc29237bee1765ed779657 upstream.

With the radeon driver loaded the HP Compaq dc5750
Small Form Factor machine fails to resume from suspend.
Adding a quirk similar to other devices avoids
the problem and the system resumes properly.

Signed-off-by: Jeffery Miller <>
Signed-off-by: Alex Deucher <>
Signed-off-by: Ben Hutchings <>
6 years agodrm/i915: Always mark the object as dirty when used by the GPU
Chris Wilson [Mon, 31 Aug 2015 14:10:39 +0000 (15:10 +0100)]
drm/i915: Always mark the object as dirty when used by the GPU

commit 51bc140431e233284660b1d22c47dec9ecdb521e upstream.

There have been many hard to track down bugs whereby userspace forgot to
flag a write buffer and then cause graphics corruption or a hung GPU
when that buffer was later purged under memory pressure (as the buffer
appeared clean, its pages would have been evicted rather than preserved
and any changes more recent than in the backing storage would be lost).
In retrospect this is a rare optimisation against memory pressure,
already the slow path. If we always mark the buffer as dirty when
accessed by the GPU, anything not used can still be evicted cheaply
(ideal behaviour for mark-and-sweep eviction) but we do not run the risk
of corruption. For correct read serialisation, userspace still has to
notify when the GPU writes to an object. However, there are certain
situations under which userspace may wish to tell white lies to the

Signed-off-by: Chris Wilson <>
Cc: Daniel Vetter <>
Cc: Kristian Høgsberg <>
Cc: Jesse Barnes <>
Cc: "Goel, Akash" <>
Cc: Michał Winiarski <>
Reviewed-by: Daniel Vetter <>
Signed-off-by: Jani Nikula <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agospi: spi-pxa2xx: Check status register to determine if SSSR_TINT is disabled
Tan, Jui Nee [Tue, 1 Sep 2015 02:22:51 +0000 (10:22 +0800)]
spi: spi-pxa2xx: Check status register to determine if SSSR_TINT is disabled

commit 02bc933ebb59208f42c2e6305b2c17fd306f695d upstream.

On Intel Baytrail, there is case when interrupt handler get called, no SPI
message is captured. The RX FIFO is indeed empty when RX timeout pending
interrupt (SSSR_TINT) happens.

Use the BIOS version where both HSUART and SPI are on the same IRQ. Both
drivers are using IRQF_SHARED when calling the request_irq function. When
running two separate and independent SPI and HSUART application that
generate data traffic on both components, user will see messages like
below on the console:

  pxa2xx-spi pxa2xx-spi.0: bad message state in interrupt handler

This commit will fix this by first checking Receiver Time-out Interrupt,
if it is disabled, ignore the request and return without servicing.

Signed-off-by: Tan, Jui Nee <>
Acked-by: Jarkko Nikula <>
Signed-off-by: Mark Brown <>
Signed-off-by: Ben Hutchings <>
6 years agoIB/uverbs: Fix race between ib_uverbs_open and remove_one
Yishai Hadas [Thu, 13 Aug 2015 15:32:03 +0000 (18:32 +0300)]
IB/uverbs: Fix race between ib_uverbs_open and remove_one

commit 35d4a0b63dc0c6d1177d4f532a9deae958f0662c upstream.

Fixes: 2a72f212263701b927559f6850446421d5906c41 ("IB/uverbs: Remove dev_table")

Before this commit there was a device look-up table that was protected
by a spin_lock used by ib_uverbs_open and by ib_uverbs_remove_one. When
it was dropped and container_of was used instead, it enabled the race
with remove_one as dev might be freed just after:
dev = container_of(inode->i_cdev, struct ib_uverbs_device, cdev) but
before the kref_get.

In addition, this buggy patch added some dead code as
container_of(x,y,z) can never be NULL and so dev can never be NULL.
As a result the comment above ib_uverbs_open saying "the open method
will either immediately run -ENXIO" is wrong as it can never happen.

The solution follows Jason Gunthorpe suggestion from below URL:

cdev will hold a kref on the parent (the containing structure,
ib_uverbs_device) and only when that kref is released it is
guaranteed that open will never be called again.

In addition, fixes the active count scheme to use an atomic
not a kref to prevent WARN_ON as pointed by above comment
from Jason.

Signed-off-by: Yishai Hadas <>
Signed-off-by: Shachar Raindel <>
Reviewed-by: Jason Gunthorpe <>
Signed-off-by: Doug Ledford <>
Signed-off-by: Ben Hutchings <>
6 years agoIB/mlx4: Use correct SL on AH query under RoCE
Noa Osherovich [Thu, 30 Jul 2015 14:34:24 +0000 (17:34 +0300)]
IB/mlx4: Use correct SL on AH query under RoCE

commit 5e99b139f1b68acd65e36515ca347b03856dfb5a upstream.

The mlx4 IB driver implementation for ib_query_ah used a wrong offset
(28 instead of 29) when link type is Ethernet. Fixed to use the correct one.

Fixes: fa417f7b520e ('IB/mlx4: Add support for IBoE')
Signed-off-by: Shani Michaeli <>
Signed-off-by: Noa Osherovich <>
Signed-off-by: Or Gerlitz <>
Signed-off-by: Doug Ledford <>
Signed-off-by: Ben Hutchings <>
6 years agoSUNRPC: xs_reset_transport must mark the connection as disconnected
Trond Myklebust [Sat, 29 Aug 2015 20:36:30 +0000 (13:36 -0700)]
SUNRPC: xs_reset_transport must mark the connection as disconnected

commit 0c78789e3a030615c6650fde89546cadf40ec2cc upstream.

In case the reconnection attempt fails.

Signed-off-by: Trond Myklebust <>
[bwh: Backported to 3.2: add local variable xprt]
Signed-off-by: Ben Hutchings <>
6 years agoIB/qib: Change lkey table allocation to support more MRs
Mike Marciniszyn [Tue, 21 Jul 2015 12:36:07 +0000 (08:36 -0400)]
IB/qib: Change lkey table allocation to support more MRs

commit d6f1c17e162b2a11e708f28fa93f2f79c164b442 upstream.

The lkey table is allocated with with a get_user_pages() with an
order based on a number of index bits from a module parameter.

The underlying kernel code cannot allocate that many contiguous pages.

There is no reason the underlying memory needs to be physically

This patch:
- switches the allocation/deallocation to vmalloc/vfree
- caps the number of bits to 23 to insure at least 1 generation bit
  o this matches the module parameter description

Reviewed-by: Vinit Agnihotri <>
Signed-off-by: Mike Marciniszyn <>
Signed-off-by: Doug Ledford <>
[bwh: Backported to 3.2:
 - Adjust context
 - Add definition of qib_dev_warn(), added upstream by commit ddb887658970
   ("IB/qib: Convert opcode counters to per-context")]
Signed-off-by: Ben Hutchings <>
6 years agoxfs: return errors from partial I/O failures to files
David Jeffery [Fri, 28 Aug 2015 04:50:45 +0000 (14:50 +1000)]
xfs: return errors from partial I/O failures to files

commit c9eb256eda4420c06bb10f5e8fbdbe1a34bc98e0 upstream.

There is an issue with xfs's error reporting in some cases of I/O partially
failing and partially succeeding. Calls like fsync() can report success even
though not all I/O was successful in partial-failure cases such as one disk of
a RAID0 array being offline.

The issue can occur when there are more than one bio per xfs_ioend struct.
Each call to xfs_end_bio() for a bio completing will write a value to
ioend->io_error.  If a successful bio completes after any failed bio, no
error is reported do to it writing 0 over the error code set by any failed bio.
The I/O error information is now lost and when the ioend is completed
only success is reported back up the filesystem stack.

xfs_end_bio() should only set ioend->io_error in the case of BIO_UPTODATE
being clear.  ioend->io_error is initialized to 0 at allocation so only needs
to be updated by a failed bio. Also check that ioend->io_error is 0 so that
the first error reported will be the error code returned.

Signed-off-by: David Jeffery <>
Reviewed-by: Dave Chinner <>
Signed-off-by: Dave Chinner <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agodrivercore: Fix unregistration path of platform devices
Grant Likely [Sun, 7 Jun 2015 14:20:11 +0000 (15:20 +0100)]
drivercore: Fix unregistration path of platform devices

commit 7f5dcaf1fdf289767a126a0a5cc3ef39b5254b06 upstream.

The unregister path of platform_device is broken. On registration, it
will register all resources with either a parent already set, or
type==IORESOURCE_{IO,MEM}. However, on unregister it will release
everything with type==IORESOURCE_{IO,MEM}, but ignore the others. There
are also cases where resources don't get registered in the first place,
like with devices created by of_platform_populate()*.

Fix the unregister path to be symmetrical with the register path by
checking the parent pointer instead of the type field to decide which
resources to unregister. This is safe because the upshot of the
registration path algorithm is that registered resources have a parent
pointer, and non-registered resources do not.

* It can be argued that of_platform_populate() should be registering
  it's resources, and they argument has some merit. However, there are
  quite a few platforms that end up broken if we try to do that due to
  overlapping resources in the device tree. Until that is fixed, we need
  to solve the immediate problem.

Cc: Pantelis Antoniou <>
Cc: Wolfram Sang <>
Cc: Rob Herring <>
Cc: Greg Kroah-Hartman <>
Cc: Ricardo Ribalda Delgado <>
Signed-off-by: Grant Likely <>
Tested-by: Ricardo Ribalda Delgado <>
Tested-by: Wolfram Sang <>
Signed-off-by: Rob Herring <>
Signed-off-by: Ben Hutchings <>
6 years agoof/address: Don't loop forever in of_find_matching_node_by_address().
David Daney [Wed, 19 Aug 2015 20:17:47 +0000 (13:17 -0700)]
of/address: Don't loop forever in of_find_matching_node_by_address().

commit 3a496b00b6f90c41bd21a410871dfc97d4f3c7ab upstream.

If the internal call to of_address_to_resource() fails, we end up
looping forever in of_find_matching_node_by_address().  This can be
caused by a defective device tree, or calling with an incorrect
matches argument.

Fix by calling of_find_matching_node() unconditionally at the end of
the loop.

Signed-off-by: David Daney <>
Signed-off-by: Rob Herring <>
Signed-off-by: Ben Hutchings <>
6 years agortlwifi: rtl8192cu: Add new device ID
Adrien Schildknecht [Wed, 19 Aug 2015 15:33:12 +0000 (17:33 +0200)]
rtlwifi: rtl8192cu: Add new device ID

commit 1642d09fb9b128e8e538b2a4179962a34f38dff9 upstream.

The v2 of NetGear WNA1000M uses a different idProduct: USB ID 0846:9043

Signed-off-by: Adrien Schildknecht <>
Acked-by: Larry Finger <>
Signed-off-by: Kalle Valo <>
Signed-off-by: Ben Hutchings <>
6 years agortlwifi: rtl8192cu: Add new device ID
Marek Vasut [Thu, 26 Mar 2015 01:16:06 +0000 (02:16 +0100)]
rtlwifi: rtl8192cu: Add new device ID

commit 9374e7d2fdcad3c36dafc8d3effd554bc702c4b6 upstream.

Add new ID for ASUS N10 WiFi dongle.

Signed-off-by: Marek Vasut <>
Tested-by: Marek Vasut <>
Cc: Larry Finger <>
Cc: John W. Linville <>
Acked-by: Larry Finger <>
Signed-off-by: Kalle Valo <>
Signed-off-by: Ben Hutchings <>
6 years agoDRM - radeon: Don't link train DisplayPort on HPD until we get the dpcd
Stephen Chandler Paul [Fri, 21 Aug 2015 18:16:12 +0000 (14:16 -0400)]
DRM - radeon: Don't link train DisplayPort on HPD until we get the dpcd

commit 924f92bf12bfbef3662619e3ed24a1cea7c1cbcd upstream.

Most of the time this isn't an issue since hotplugging an adaptor will
trigger a crtc mode change which in turn, causes the driver to probe
every DisplayPort for a dpcd. However, in cases where hotplugging
doesn't cause a mode change (specifically when one unplugs a monitor
from a DisplayPort connector, then plugs that same monitor back in
seconds later on the same port without any other monitors connected), we
never probe for the dpcd before starting the initial link training. What
happens from there looks like this:

- GPU has only one monitor connected. It's connected via
  DisplayPort, and does not go through an adaptor of any sort.

- User unplugs DisplayPort connector from GPU.

- Change in HPD is detected by the driver, we probe every
  DisplayPort for a possible connection.

- Probe the port the user originally had the monitor connected
  on for it's dpcd. This fails, and we clear the first (and only
  the first) byte of the dpcd to indicate we no longer have a
  dpcd for this port.

- User plugs the previously disconnected monitor back into the
  same DisplayPort.

- radeon_connector_hotplug() is called before everyone else,
  and tries to handle the link training. Since only the first
  byte of the dpcd is zeroed, the driver is able to complete
  link training but does so against the wrong dpcd, causing it
  to initialize the link with the wrong settings.

- Display stays blank (usually), dpcd is probed after the
  initial link training, and the driver prints no obvious
  messages to the log.

In theory, since only one byte of the dpcd is chopped off (specifically,
the byte that contains the revision information for DisplayPort), it's
not entirely impossible that this bug may not show on certain monitors.
For instance, the only reason this bug was visible on my ASUS PB238
monitor was due to the fact that this monitor using the enhanced framing
symbol sequence, the flag for which is ignored if the radeon driver
thinks that the DisplayPort version is below 1.1.

Signed-off-by: Stephen Chandler Paul <>
Reviewed-by: Jerome Glisse <>
Signed-off-by: Alex Deucher <>
Signed-off-by: Ben Hutchings <>
6 years agoxfs: Fix xfs_attr_leafblock definition
Jan Kara [Wed, 19 Aug 2015 00:34:32 +0000 (10:34 +1000)]
xfs: Fix xfs_attr_leafblock definition

commit ffeecc5213024ae663377b442eedcfbacf6d0c5d upstream.

struct xfs_attr_leafblock contains 'entries' array which is declared
with size 1 altough it can in fact contain much more entries. Since this
array is followed by further struct members, gcc (at least in version
4.8.3) thinks that the array has the fixed size of 1 element and thus
may optimize away all accesses beyond the end of array resulting in
non-working code. This problem was only observed with userspace code in
xfsprogs, however it's better to be safe in kernel as well and have
matching kernel and xfsprogs definitions.

Signed-off-by: Jan Kara <>
Reviewed-by: Dave Chinner <>
Signed-off-by: Dave Chinner <>
[bwh: Backported to 3.2: adjust filename]
Signed-off-by: Ben Hutchings <>
6 years agoeCryptfs: Invalidate dcache entries when lower i_nlink is zero
Tyler Hicks [Wed, 5 Aug 2015 16:26:36 +0000 (11:26 -0500)]
eCryptfs: Invalidate dcache entries when lower i_nlink is zero

commit 5556e7e6d30e8e9b5ee51b0e5edd526ee80e5e36 upstream.

Consider eCryptfs dcache entries to be stale when the corresponding
lower inode's i_nlink count is zero. This solves a problem caused by the
lower inode being directly modified, without going through the eCryptfs
mount, leaving stale eCryptfs dentries cached and the eCryptfs inode's
i_nlink count not being cleared.

Signed-off-by: Tyler Hicks <>
Reported-by: Richard Weinberger <>
[bwh: Backported to 3.2:
 - Test d_revalidate pointer directly rather than a DCACHE_OP flag
 - Open-code d_inode()
 - Adjust context]
Signed-off-by: Ben Hutchings <>
6 years agoUSB: ftdi_sio: Added custom PID for CustomWare products
Matthijs Kooijman [Tue, 18 Aug 2015 08:33:56 +0000 (10:33 +0200)]
USB: ftdi_sio: Added custom PID for CustomWare products

commit 1fb8dc36384ae1140ee6ccc470de74397606a9d5 upstream.

CustomWare uses the FTDI VID with custom PIDs for their ShipModul MiniPlex

Signed-off-by: Matthijs Kooijman <>
Signed-off-by: Johan Hovold <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
6 years agousb: host: ehci-sys: delete useless bus_to_hcd conversion
Peter Chen [Mon, 17 Aug 2015 02:23:03 +0000 (10:23 +0800)]
usb: host: ehci-sys: delete useless bus_to_hcd conversion

commit 0521cfd06e1ebcd575e7ae36aab068b38df23850 upstream.

The ehci platform device's drvdata is the pointer of struct usb_hcd
already, so we doesn't need to call bus_to_hcd conversion again.

Signed-off-by: Peter Chen <>
Acked-by: Alan Stern <>
Signed-off-by: Greg Kroah-Hartman <>
[bwh: Backported to 3.2: Unfortunately some EHCI platform sub-drivers
 point drvdata to a private structure, so only create and remove the
 attributes if drvdata has been set as expected.]
Signed-off-by: Ben Hutchings <>
6 years agoserial: 8250: bind to ALi Fast Infrared Controller (ALI5123)
Maciej S. Szmigiero [Sun, 2 Aug 2015 21:15:05 +0000 (23:15 +0200)]
serial: 8250: bind to ALi Fast Infrared Controller (ALI5123)

commit 1d7002777a8fe8188caaa98d4a8eb4ed298fcdae upstream.

This way this device can be used with irtty-sir -
at least on Toshiba Satellite A20-S103 it is not configured by default
and needs PNP activation before it starts to respond on I/O ports.

This device has actually its own driver (ali-ircc),
but this driver seems to be non-functional for a very long time
(see ).

Signed-off-by: Maciej Szmigiero <>
Signed-off-by: Greg Kroah-Hartman <>
[bwh: Backported to 3.2:
 - Drop change to acpi_pnp.c, as there's no need to whitelist ACPI devices
   for the PNP bus
 - Adjust filename]
Signed-off-by: Ben Hutchings <>
6 years agodrivers: usb: fsl: Workaround for USB erratum-A005275
Nikhil Badola [Thu, 6 Aug 2015 09:21:27 +0000 (14:51 +0530)]
drivers: usb: fsl: Workaround for USB erratum-A005275

commit f8786a91548df6930643a052e40e5c0b7a8403a5 upstream.

Incoming packets in high speed are randomly corrupted by h/w
resulting in multiple errors. This workaround makes FS as
default mode in all affected socs by disabling HS chirp
signalling.This errata does not affect FS and LS mode.

Forces all HS devices to connect in FS mode for all socs
affected by this erratum:
P3041 and P2041 rev 1.0 and 1.1
P5020 and P5010 rev 1.0 and 2.0
P5040, P1010 and T4240 rev 1.0

Signed-off-by: Ramneek Mehresh <>
Signed-off-by: Nikhil Badola <>
Signed-off-by: Greg Kroah-Hartman <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agoNFSv4: don't set SETATTR for O_RDONLY|O_EXCL
NeilBrown [Thu, 30 Jul 2015 03:00:56 +0000 (13:00 +1000)]
NFSv4: don't set SETATTR for O_RDONLY|O_EXCL

commit efcbc04e16dfa95fef76309f89710dd1d99a5453 upstream.

It is unusual to combine the open flags O_RDONLY and O_EXCL, but
it appears that libre-office does just that.

[pid  3250] stat("/home/USER/.config", {st_mode=S_IFDIR|0700, st_size=8192, ...}) = 0
[pid  3250] open("/home/USER/.config/libreoffice/4-suse/user/extensions/buildid", O_RDONLY|O_EXCL <unfinished ...>

NFSv4 takes O_EXCL as a sign that a setattr command should be sent,
probably to reset the timestamps.

When it was an O_RDONLY open, the SETATTR command does not
identify any actual attributes to change.
If no delegation was provided to the open, the SETATTR uses the
all-zeros stateid and the request is accepted (at least by the
Linux NFS server - no harm, no foul).

If a read-delegation was provided, this is used in the SETATTR
request, and a Netapp filer will justifiably claim
NFS4ERR_BAD_STATEID, which the Linux client takes as a sign
to retry - indefinitely.

So only treat O_EXCL specially if O_CREAT was also given.

Signed-off-by: NeilBrown <>
Signed-off-by: Trond Myklebust <>
[bwh: Backported to 3.2: we only check open_flags, not createmode as well]
Signed-off-by: Ben Hutchings <>
6 years agowindfarm: decrement client count when unregistering
Paul Bolle [Fri, 31 Jul 2015 12:08:58 +0000 (14:08 +0200)]
windfarm: decrement client count when unregistering

commit fe2b592173ff0274e70dc44d1d28c19bb995aa7c upstream.

wf_unregister_client() increments the client count when a client
unregisters. That is obviously incorrect. Decrement that client count

Fixes: 75722d3992f5 ("[PATCH] ppc64: Thermal control for SMU based machines")

Signed-off-by: Paul Bolle <>
Signed-off-by: Michael Ellerman <>
Signed-off-by: Ben Hutchings <>
6 years agodevres: fix devres_get()
Masahiro Yamada [Wed, 15 Jul 2015 01:29:00 +0000 (10:29 +0900)]
devres: fix devres_get()

commit 64526370d11ce8868ca495723d595b61e8697fbf upstream.

Currently, devres_get() passes devres_free() the pointer to devres,
but devres_free() should be given with the pointer to resource data.

Fixes: 9ac7849e35f7 ("devres: device resource management")
Signed-off-by: Masahiro Yamada <>
Acked-by: Tejun Heo <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
6 years agoauxdisplay: ks0108: fix refcount
Sudip Mukherjee [Mon, 20 Jul 2015 11:57:21 +0000 (17:27 +0530)]
auxdisplay: ks0108: fix refcount

commit bab383de3b84e584b0f09227151020b2a43dc34c upstream.

parport_find_base() will implicitly do parport_get_port() which
increases the refcount. Then parport_register_device() will again
increment the refcount. But while unloading the module we are only
doing parport_unregister_device() decrementing the refcount only once.
We add an parport_put_port() to neutralize the effect of

Signed-off-by: Sudip Mukherjee <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
6 years agoKVM: MMU: fix validation of mmio page fault
Xiao Guangrong [Wed, 5 Aug 2015 04:04:19 +0000 (12:04 +0800)]
KVM: MMU: fix validation of mmio page fault

commit 6f691251c0350ac52a007c54bf3ef62e9d8cdc5e upstream.

We got the bug that qemu complained with "KVM: unknown exit, hardware
reason 31" and KVM shown these info:
[84245.284948] EPT: Misconfiguration.
[84245.285056] EPT: GPA: 0xfeda848
[84245.285154] ept_misconfig_inspect_spte: spte 0x5eaef50107 level 4
[84245.285344] ept_misconfig_inspect_spte: spte 0x5f5fadc107 level 3
[84245.285532] ept_misconfig_inspect_spte: spte 0x5141d18107 level 2
[84245.285723] ept_misconfig_inspect_spte: spte 0x52e40dad77 level 1

This is because we got a mmio #PF and the handler see the mmio spte becomes
normal (points to the ram page)

However, this is valid after introducing fast mmio spte invalidation which
increases the generation-number instead of zapping mmio sptes, a example
is as follows:
1. QEMU drops mmio region by adding a new memslot
2. invalidate all mmio sptes

        VCPU 0                        VCPU 1
    access the invalid mmio spte
                            access the region originally was MMIO before
                            set the spte to the normal ram map

    mmio #PF
    check the spte and see it becomes normal ram mapping !!!

This patch fixes the bug just by dropping the check in mmio handler, it's
good for backport. Full check will be introduced in later patches

Reported-by: Pavel Shirshov <>
Tested-by: Pavel Shirshov <>
Signed-off-by: Xiao Guangrong <>
Signed-off-by: Paolo Bonzini <>
[bwh: Backported to 3.2: error code from handle_mmio_page_fault_common()
 was not named]
Signed-off-by: Ben Hutchings <>
6 years agousb: gadget: m66592-udc: forever loop in set_feature()
Dan Carpenter [Wed, 29 Jul 2015 21:30:58 +0000 (00:30 +0300)]
usb: gadget: m66592-udc: forever loop in set_feature()

commit 5feb5d2003499b1094d898c010a7604d7afddc4c upstream.

There is an "&&" vs "||" typo here so this loops 3000 times or if we get
unlucky it could loop forever.

Fixes: ceaa0a6eeadf ('usb: gadget: m66592-udc: add support for TEST_MODE')
Signed-off-by: Dan Carpenter <>
Signed-off-by: Felipe Balbi <>
[bwh: Backported to 3.2: adjust filename]
Signed-off-by: Ben Hutchings <>
6 years agoPCI: Add VPD function 0 quirk for Intel Ethernet devices
Mark Rustad [Mon, 13 Jul 2015 18:40:07 +0000 (11:40 -0700)]
PCI: Add VPD function 0 quirk for Intel Ethernet devices

commit 7aa6ca4d39edf01f997b9e02cf6d2fdeb224f351 upstream.

Set the PCI_DEV_FLAGS_VPD_REF_F0 flag on all Intel Ethernet device
functions other than function 0, so that on multi-function devices, we will
always read VPD from function 0 instead of from the other functions.

[bhelgaas: changelog]
Signed-off-by: Mark Rustad <>
Signed-off-by: Bjorn Helgaas <>
Acked-by: Alexander Duyck <>
[bwh: Backported to 3.2:
 - Put the class check in the new function as there is no
 - Adjust context]
Signed-off-by: Ben Hutchings <>
6 years agoPCI: Add dev_flags bit to access VPD through function 0
Mark Rustad [Mon, 13 Jul 2015 18:40:02 +0000 (11:40 -0700)]
PCI: Add dev_flags bit to access VPD through function 0

commit 932c435caba8a2ce473a91753bad0173269ef334 upstream.

Add a dev_flags bit, PCI_DEV_FLAGS_VPD_REF_F0, to access VPD through
function 0 to provide VPD access on other functions.  This is for hardware
devices that provide copies of the same VPD capability registers in
multiple functions.  Because the kernel expects that each function has its
own registers, both the locking and the state tracking are affected by VPD
accesses to different functions.

On such devices for example, if a VPD write is performed on function 0,
*any* later attempt to read VPD from any other function of that device will
hang.  This has to do with how the kernel tracks the expected value of the
F bit per function.

Concurrent accesses to different functions of the same device can not only
hang but also corrupt both read and write VPD data.

When hangs occur, typically the error message:

  vpd r/w failed.  This is likely a firmware bug on this device.

will be seen.

Never set this bit on function 0 or there will be an infinite recursion.

Signed-off-by: Mark Rustad <>
Signed-off-by: Bjorn Helgaas <>
Acked-by: Alexander Duyck <>
Signed-off-by: Ben Hutchings <>
6 years agomac80211: enable assoc check for mesh interfaces
Bob Copeland [Sat, 13 Jun 2015 14:16:31 +0000 (10:16 -0400)]
mac80211: enable assoc check for mesh interfaces

commit 3633ebebab2bbe88124388b7620442315c968e8f upstream.

We already set a station to be associated when peering completes, both
in user space and in the kernel.  Thus we should always have an
associated sta before sending data frames to that station.

Failure to check assoc state can cause crashes in the lower-level driver
due to transmitting unicast data frames before driver sta structures
(e.g. ampdu state in ath9k) are initialized.  This occurred when
forwarding in the presence of fixed mesh paths: frames were transmitted
to stations with whom we hadn't yet completed peering.

Reported-by: Alexis Green <>
Tested-by: Jesse Jones <>
Signed-off-by: Bob Copeland <>
Signed-off-by: Johannes Berg <>
Signed-off-by: Ben Hutchings <>
6 years agoPCI: Fix TI816X class code quirk
Bjorn Helgaas [Fri, 19 Jun 2015 20:58:24 +0000 (15:58 -0500)]
PCI: Fix TI816X class code quirk

commit d1541dc977d376406f4584d8eb055488655c98ec upstream.

In fixup_ti816x_class(), we assigned "class = PCI_CLASS_MULTIMEDIA_VIDEO".
But PCI_CLASS_MULTIMEDIA_VIDEO is only the two-byte base class/sub-class
and needs to be shifted to make space for the low-order interface byte.

Shift PCI_CLASS_MULTIMEDIA_VIDEO to set the correct class code.

Fixes: 63c4408074cb ("PCI: Add quirk for setting valid class for TI816X Endpoint")
Signed-off-by: Bjorn Helgaas <>
CC: Hemant Pedanekar <>
[bwh: Backported to 3.2: the class check is done in this function as there
Signed-off-by: Ben Hutchings <>
6 years agorc-core: fix remove uevent generation
David Härdeman [Tue, 19 May 2015 22:03:12 +0000 (19:03 -0300)]
rc-core: fix remove uevent generation

commit a66b0c41ad277ae62a3ae6ac430a71882f899557 upstream.

The input_dev is already gone when the rc device is being unregistered
so checking for its presence only means that no remove uevent will be

Signed-off-by: David Härdeman <>
Signed-off-by: Mauro Carvalho Chehab <>
Signed-off-by: Ben Hutchings <>
6 years agovfs: Test for and handle paths that are unreachable from their mnt_root
Eric W. Biederman [Sun, 16 Aug 2015 01:27:13 +0000 (20:27 -0500)]
vfs: Test for and handle paths that are unreachable from their mnt_root

commit 397d425dc26da728396e66d392d5dcb8dac30c37 upstream.

In rare cases a directory can be renamed out from under a bind mount.
In those cases without special handling it becomes possible to walk up
the directory tree to the root dentry of the filesystem and down
from the root dentry to every other file or directory on the filesystem.

Like division by zero .. from an unconnected path can not be given
a useful semantic as there is no predicting at which path component
the code will realize it is unconnected.  We certainly can not match
the current behavior as the current behavior is a security hole.

Therefore when encounting .. when following an unconnected path
return -ENOENT.

- Add a function path_connected to verify path->dentry is reachable
  from path->mnt.mnt_root.  AKA to validate that rename did not do
  something nasty to the bind mount.

  To avoid races path_connected must be called after following a path
  component to it's next path component.

Signed-off-by: "Eric W. Biederman" <>
Signed-off-by: Al Viro <>
Signed-off-by: Ben Hutchings <>
6 years agodcache: Handle escaped paths in prepend_path
Eric W. Biederman [Sat, 15 Aug 2015 18:36:12 +0000 (13:36 -0500)]
dcache: Handle escaped paths in prepend_path

commit cde93be45a8a90d8c264c776fab63487b5038a65 upstream.

A rename can result in a dentry that by walking up d_parent
will never reach it's mnt_root.  For lack of a better term
I call this an escaped path.

prepend_path is called by four different functions __d_path,
d_absolute_path, d_path, and getcwd.

__d_path only wants to see paths are connected to the root it passes
in.  So __d_path needs prepend_path to return an error.

d_absolute_path similarly wants to see paths that are connected to
some root.  Escaped paths are not connected to any mnt_root so
d_absolute_path needs prepend_path to return an error greater
than 1.  So escaped paths will be treated like paths on lazily
unmounted mounts.

getcwd needs to prepend "(unreachable)" so getcwd also needs
prepend_path to return an error.

d_path is the interesting hold out.  d_path just wants to print
something, and does not care about the weird cases.  Which raises
the question what should be printed?

Given that <escaped_path>/<anything> should result in -ENOENT I
believe it is desirable for escaped paths to be printed as empty
paths.  As there are not really any meaninful path components when
considered from the perspective of a mount tree.

So tweak prepend_path to return an empty path with an new error
code of 3 when it encounters an escaped path.

Signed-off-by: "Eric W. Biederman" <>
Signed-off-by: Al Viro <>
Signed-off-by: Ben Hutchings <>
6 years agosparc64: Fix userspace FPU register corruptions.
David S. Miller [Fri, 7 Aug 2015 02:13:25 +0000 (19:13 -0700)]
sparc64: Fix userspace FPU register corruptions.

commit 44922150d87cef616fd183220d43d8fde4d41390 upstream.

If we have a series of events from userpsace, with %fprs=FPRS_FEF,
like follows:

RTRAP (kernel FPU restore with fpu_saved=0x4)

We will not restore the user registers that were clobbered by the FPU
using kernel code in the inner-most trap.

Traps allocate FPU save slots in the thread struct, and FPU using
sequences save the "dirty" FPU registers only.

This works at the initial trap level because all of the registers
get recorded into the top-level FPU save area, and we'll return
to userspace with the FPU disabled so that any FPU use by the user
will take an FPU disabled trap wherein we'll load the registers
back up properly.

But this is not how trap returns from kernel to kernel operate.

The simplest fix for this bug is to always save all FPU register state
for anything other than the top-most FPU save area.

Getting rid of the optimized inner-slot FPU saving code ends up
making VISEntryHalf degenerate into plain VISEntry.

Longer term we need to do something smarter to reinstate the partial
save optimizations.  Perhaps the fundament error is having trap entry
and exit allocate FPU save slots and restore register state.  Instead,
the VISEntry et al. calls should be doing that work.

This bug is about two decades old.

Reported-by: James Y Knight <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2:
 - Adjust context
 - Drop changes to NG4memcpy.S and ksyms.c]
Signed-off-by: Ben Hutchings <>
6 years agosctp: donot reset the overall_error_count in SHUTDOWN_RECEIVE state
lucien [Wed, 26 Aug 2015 20:52:20 +0000 (04:52 +0800)]
sctp: donot reset the overall_error_count in SHUTDOWN_RECEIVE state

commit f648f807f61e64d247d26611e34cc97e4ed03401 upstream.

Commit f8d960524328 ("sctp: Enforce retransmission limit during shutdown")
fixed a problem with excessive retransmissions in the SHUTDOWN_PENDING by not
resetting the association overall_error_count.  This allowed the association
to better enforce assoc.max_retrans limit.

However, the same issue still exists when the association is in SHUTDOWN_RECEIVED
state.  In this state, HB-ACKs will continue to reset the overall_error_count
for the association would extend the lifetime of association unnecessarily.

This patch solves this by resetting the overall_error_count whenever the current
state is small then SCTP_STATE_SHUTDOWN_PENDING.  As a small side-effect, we
states, but they are not really impacted because we disable Heartbeats in those

Fixes: Commit f8d960524328 ("sctp: Enforce retransmission limit during shutdown")
Signed-off-by: Xin Long <>
Acked-by: Marcelo Ricardo Leitner <>
Acked-by: Vlad Yasevich <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
6 years agonet: Fix RCU splat in af_key
David Ahern [Mon, 24 Aug 2015 21:17:17 +0000 (15:17 -0600)]
net: Fix RCU splat in af_key

commit ba51b6be38c122f7dab40965b4397aaf6188a464 upstream.

Hit the following splat testing VRF change for ipsec:

[  113.475692] ===============================
[  113.476194] [ INFO: suspicious RCU usage. ]
[  113.476667] 4.2.0-rc6-1+deb7u2+clUNRELEASED #3.2.65-1+deb7u2+clUNRELEASED Not tainted
[  113.477545] -------------------------------
[  113.478013] /work/monster-14/dsa/kernel.git/include/linux/rcupdate.h:568 Illegal context switch in RCU read-side critical section!
[  113.479288]
[  113.479288] other info that might help us debug this:
[  113.479288]
[  113.480207]
[  113.480207] rcu_scheduler_active = 1, debug_locks = 1
[  113.480931] 2 locks held by setkey/6829:
[  113.481371]  #0:  (&net->xfrm.xfrm_cfg_mutex){+.+.+.}, at: [<ffffffff814e9887>] pfkey_sendmsg+0xfb/0x213
[  113.482509]  #1:  (rcu_read_lock){......}, at: [<ffffffff814e767f>] rcu_read_lock+0x0/0x6e
[  113.483509]
[  113.483509] stack backtrace:
[  113.484041] CPU: 0 PID: 6829 Comm: setkey Not tainted 4.2.0-rc6-1+deb7u2+clUNRELEASED #3.2.65-1+deb7u2+clUNRELEASED
[  113.485422] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 04/01/2014
[  113.486845]  0000000000000001 ffff88001d4c7a98 ffffffff81518af2 ffffffff81086962
[  113.487732]  ffff88001d538480 ffff88001d4c7ac8 ffffffff8107ae75 ffffffff8180a154
[  113.488628]  0000000000000b30 0000000000000000 00000000000000d0 ffff88001d4c7ad8
[  113.489525] Call Trace:
[  113.489813]  [<ffffffff81518af2>] dump_stack+0x4c/0x65
[  113.490389]  [<ffffffff81086962>] ? console_unlock+0x3d6/0x405
[  113.491039]  [<ffffffff8107ae75>] lockdep_rcu_suspicious+0xfa/0x103
[  113.491735]  [<ffffffff81064032>] rcu_preempt_sleep_check+0x45/0x47
[  113.492442]  [<ffffffff8106404d>] ___might_sleep+0x19/0x1c8
[  113.493077]  [<ffffffff81064268>] __might_sleep+0x6c/0x82
[  113.493681]  [<ffffffff81133190>] cache_alloc_debugcheck_before.isra.50+0x1d/0x24
[  113.494508]  [<ffffffff81134876>] kmem_cache_alloc+0x31/0x18f
[  113.495149]  [<ffffffff814012b5>] skb_clone+0x64/0x80
[  113.495712]  [<ffffffff814e6f71>] pfkey_broadcast_one+0x3d/0xff
[  113.496380]  [<ffffffff814e7b84>] pfkey_broadcast+0xb5/0x11e
[  113.497024]  [<ffffffff814e82d1>] pfkey_register+0x191/0x1b1
[  113.497653]  [<ffffffff814e9770>] pfkey_process+0x162/0x17e
[  113.498274]  [<ffffffff814e9895>] pfkey_sendmsg+0x109/0x213

In pfkey_sendmsg the net mutex is taken and then pfkey_broadcast takes
the RCU lock.

Since pfkey_broadcast takes the RCU lock the allocation argument is
pointless since GFP_ATOMIC must be used between the rcu_read_{,un}lock.
The one call outside of rcu can be done with GFP_KERNEL.

Fixes: 7f6b9dbd5afbd ("af_key: locking change")
Signed-off-by: David Ahern <>
Acked-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agox86/ldt: Further fix FPU emulation
Andy Lutomirski [Fri, 14 Aug 2015 22:02:55 +0000 (15:02 -0700)]
x86/ldt: Further fix FPU emulation

commit 12e244f4b550498bbaf654a52f93633f7dde2dc7 upstream.

The previous fix confused a selector with a segment prefix.  Fix it.

Compile-tested only.

Cc: Juergen Gross <>
Reported-by: Linus Torvalds <>
Fixes: 4809146b86c3 ("x86/ldt: Correct FPU emulation access to LDT")
Signed-off-by: Andy Lutomirski <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
6 years agoipc,sem: fix use after free on IPC_RMID after a task using same semaphore set exits
Herton R. Krzesinski [Fri, 14 Aug 2015 22:35:02 +0000 (15:35 -0700)]
ipc,sem: fix use after free on IPC_RMID after a task using same semaphore set exits

commit 602b8593d2b4138c10e922eeaafe306f6b51817b upstream.

The current semaphore code allows a potential use after free: in
exit_sem we may free the task's sem_undo_list while there is still
another task looping through the same semaphore set and cleaning the
sem_undo list at freeary function (the task called IPC_RMID for the same
semaphore set).

For example, with a test program [1] running which keeps forking a lot
of processes (which then do a semop call with SEM_UNDO flag), and with
the parent right after removing the semaphore set with IPC_RMID, and a
kernel built with CONFIG_SLAB, CONFIG_SLAB_DEBUG and
CONFIG_DEBUG_SPINLOCK, you can easily see something like the following
in the kernel log:

   Slab corruption (Not tainted): kmalloc-64 start=ffff88003b45c1c0, len=64
   000: 6b 6b 6b 6b 6b 6b 6b 6b 00 6b 6b 6b 6b 6b 6b 6b  kkkkkkkk.kkkkkkk
   010: ff ff ff ff 6b 6b 6b 6b ff ff ff ff ff ff ff ff  ....kkkk........
   Prev obj: start=ffff88003b45c180, len=64
   000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a  .....N......ZZZZ
   010: ff ff ff ff ff ff ff ff c0 fb 01 37 00 88 ff ff  ...........7....
   Next obj: start=ffff88003b45c200, len=64
   000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a  .....N......ZZZZ
   010: ff ff ff ff ff ff ff ff 68 29 a7 3c 00 88 ff ff  ........h).<....
   BUG: spinlock wrong CPU on CPU#2, test/18028
   general protection fault: 0000 [#1] SMP
   Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib]
   CPU: 2 PID: 18028 Comm: test Not tainted 4.2.0-rc5+ #1
   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014
   RIP: spin_dump+0x53/0xc0
   Call Trace:
     ? _raw_spin_lock+0xe/0x10
     ? __do_page_fault+0x19a/0x430
     ? __audit_syscall_entry+0xa8/0x100
     ? syscall_trace_leave+0xde/0x130
   Code: 8b 80 88 03 00 00 48 8d 88 60 05 00 00 48 c7 c7 a0 2c a4 81 31 c0 65 8b 15 eb 40 f3 7e e8 08 31 68 00 4d 85 e4 44 8b 4b 08 74 5e <45> 8b 84 24 88 03 00 00 49 8d 8c 24 60 05 00 00 8b 53 04 48 89
   RIP  [<ffffffff810d6053>] spin_dump+0x53/0xc0
    RSP <ffff88003750fd68>
   ---[ end trace 783ebb76612867a0 ]---
   NMI watchdog: BUG: soft lockup - CPU#3 stuck for 22s! [test:18053]
   Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib]
   CPU: 3 PID: 18053 Comm: test Tainted: G      D         4.2.0-rc5+ #1
   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014
   RIP: native_read_tsc+0x0/0x20
   Call Trace:
     ? delay_tsc+0x40/0x70
     ? handle_mm_fault+0xbf6/0x1880
     ? dequeue_task_fair+0x79/0x4a0
     ? __do_page_fault+0x19a/0x430
     ? kfree_debugcheck+0x16/0x40
     ? __do_page_fault+0x19a/0x430
     ? __audit_syscall_entry+0xa8/0x100
     ? do_audit_syscall_entry+0x66/0x70
     ? syscall_trace_enter_phase1+0x139/0x160
   Code: 47 10 83 e8 01 85 c0 89 47 10 75 08 65 48 89 3d 1f 74 ff 7e c9 c3 0f 1f 44 00 00 55 48 89 e5 e8 87 17 04 00 66 90 c9 c3 0f 1f 00 <55> 48 89 e5 0f 31 89 c1 48 89 d0 48 c1 e0 20 89 c9 48 09 c8 c9
   Kernel panic - not syncing: softlockup: hung tasks

I wasn't able to trigger any badness on a recent kernel without the
proper config debugs enabled, however I have softlockup reports on some
kernel versions, in the semaphore code, which are similar as above (the
scenario is seen on some servers running IBM DB2 which uses semaphore

The patch here fixes the race against freeary, by acquiring or waiting
on the sem_undo_list lock as necessary (exit_sem can race with freeary,
while freeary sets un->semid to -1 and removes the same sem_undo from
list_proc or when it removes the last sem_undo).

After the patch I'm unable to reproduce the problem using the test case

[1] Test case used below:

    #include <stdio.h>
    #include <sys/types.h>
    #include <sys/ipc.h>
    #include <sys/sem.h>
    #include <sys/wait.h>
    #include <stdlib.h>
    #include <time.h>
    #include <unistd.h>
    #include <errno.h>

    #define NSEM 1
    #define NSET 5

    int sid[NSET];

    void thread()
            struct sembuf op;
            int s;
            uid_t pid = getuid();

            s = rand() % NSET;
            op.sem_num = pid % NSEM;
            op.sem_op = 1;
            op.sem_flg = SEM_UNDO;

            semop(sid[s], &op, 1);

    void create_set()
            int i, j;
            pid_t p;
            union {
                    int val;
                    struct semid_ds *buf;
                    unsigned short int *array;
                    struct seminfo *__buf;
            } un;

            /* Create and initialize semaphore set */
            for (i = 0; i < NSET; i++) {
                    sid[i] = semget(IPC_PRIVATE , NSEM, 0644 | IPC_CREAT);
                    if (sid[i] < 0) {
            un.val = 0;
            for (i = 0; i < NSET; i++) {
                    for (j = 0; j < NSEM; j++) {
                            if (semctl(sid[i], j, SETVAL, un) < 0)

            /* Launch threads that operate on semaphore set */
            for (i = 0; i < NSEM * NSET * NSET; i++) {
                    p = fork();
                    if (p < 0)
                    if (p == 0)

            /* Free semaphore set */
            for (i = 0; i < NSET; i++) {
                    if (semctl(sid[i], NSEM, IPC_RMID))

            /* Wait for forked processes to exit */
            while (wait(NULL)) {
                    if (errno == ECHILD)

    int main(int argc, char **argv)
            pid_t p;


            while (1) {
                    p = fork();
                    if (p < 0) {
                    if (p == 0) {
                            goto end;

                    /* Wait for forked processes to exit */
                    while (wait(NULL)) {
                            if (errno == ECHILD)
            return 0;

[ use normal comment layout]
Signed-off-by: Herton R. Krzesinski <>
Acked-by: Manfred Spraul <>
Cc: Davidlohr Bueso <>
Cc: Rafael Aquini <>
CC: Aristeu Rozanski <>
Cc: David Jeffery <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agolibfc: Fix fc_fcp_cleanup_each_cmd()
Bart Van Assche [Fri, 5 Jun 2015 21:20:51 +0000 (14:20 -0700)]
libfc: Fix fc_fcp_cleanup_each_cmd()

commit 8f2777f53e3d5ad8ef2a176a4463a5c8e1a16431 upstream.

Since fc_fcp_cleanup_cmd() can sleep this function must not
be called while holding a spinlock. This patch avoids that
fc_fcp_cleanup_each_cmd() triggers the following bug:

BUG: scheduling while atomic: sg_reset/1512/0x00000202
1 lock held by sg_reset/1512:
 #0:  (&(&fsp->scsi_pkt_lock)->rlock){+.-...}, at: [<ffffffffc0225cd5>] fc_fcp_cleanup_each_cmd.isra.21+0xa5/0x150 [libfc]
Preemption disabled at:[<ffffffffc0225cd5>] fc_fcp_cleanup_each_cmd.isra.21+0xa5/0x150 [libfc]
Call Trace:
 [<ffffffff816c612c>] dump_stack+0x4f/0x7b
 [<ffffffff810828bc>] __schedule_bug+0x6c/0xd0
 [<ffffffff816c87aa>] __schedule+0x71a/0xa10
 [<ffffffff816c8ad2>] schedule+0x32/0x80
 [<ffffffffc0217eac>] fc_seq_set_resp+0xac/0x100 [libfc]
 [<ffffffffc0218b11>] fc_exch_done+0x41/0x60 [libfc]
 [<ffffffffc0225cff>] fc_fcp_cleanup_each_cmd.isra.21+0xcf/0x150 [libfc]
 [<ffffffffc0225f43>] fc_eh_device_reset+0x1c3/0x270 [libfc]
 [<ffffffff814a2cc9>] scsi_try_bus_device_reset+0x29/0x60
 [<ffffffff814a3908>] scsi_ioctl_reset+0x258/0x2d0
 [<ffffffff814a2650>] scsi_ioctl+0x150/0x440
 [<ffffffff814b3a9d>] sd_ioctl+0xad/0x120
 [<ffffffff8132f266>] blkdev_ioctl+0x1b6/0x810
 [<ffffffff811da608>] block_ioctl+0x38/0x40
 [<ffffffff811b4e08>] do_vfs_ioctl+0x2f8/0x530
 [<ffffffff811b50c1>] SyS_ioctl+0x81/0xa0
 [<ffffffff816cf8b2>] system_call_fastpath+0x16/0x7a

Signed-off-by: Bart Van Assche <>
Signed-off-by: Vasu Dev <>
Signed-off-by: James Bottomley <>
Signed-off-by: Ben Hutchings <>
6 years agolibiscsi: Fix host busy blocking during connection teardown
John Soni Jose [Wed, 24 Jun 2015 01:11:58 +0000 (06:41 +0530)]
libiscsi: Fix host busy blocking during connection teardown

commit 660d0831d1494a6837b2f810d08b5be092c1f31d upstream.

In case of hw iscsi offload, an host can have N-number of active
connections. There can be IO's running on some connections which
make host->host_busy always TRUE. Now if logout from a connection
is tried then the code gets into an infinite loop as host->host_busy
is always TRUE.

     * Block until all in-progress commands for this connection
     * time out or fail.
     for (;;) {
      spin_lock_irqsave(session->host->host_lock, flags);
      if (!atomic_read(&session->host->host_busy)) { /* OK for ERL == 0 */
      spin_unlock_irqrestore(session->host->host_lock, flags);
     spin_unlock_irqrestore(session->host->host_lock, flags);
     iscsi_conn_printk(KERN_INFO, conn, "iscsi conn_destroy(): "
                 "host_busy %d host_failed %d\n",


This is not an issue with software-iscsi/iser as each cxn is a separate

Acquiring eh_mutex in iscsi_conn_teardown() before setting
session->state = ISCSI_STATE_TERMINATE.

Signed-off-by: John Soni Jose <>
Reviewed-by: Mike Christie <>
Reviewed-by: Chris Leech <>
Signed-off-by: James Bottomley <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agodm btree: add ref counting ops for the leaves of top level btrees
Joe Thornber [Wed, 12 Aug 2015 14:12:09 +0000 (15:12 +0100)]
dm btree: add ref counting ops for the leaves of top level btrees

commit b0dc3c8bc157c60b1d470163882be8c13e1950af upstream.

When using nested btrees, the top leaves of the top levels contain
block addresses for the root of the next tree down.  If we shadow a
shared leaf node the leaf values (sub tree roots) should be incremented

This is only an issue if there is metadata sharing in the top levels.
Which only occurs if metadata snapshots are being used (as is possible
with dm-thinp).  And could result in a block from the thinp metadata
snap being reused early, thus corrupting the thinp metadata snap.

Signed-off-by: Joe Thornber <>
Signed-off-by: Mike Snitzer <>
[bwh: Backported to 3.2:
 - Drop change to remove_one()
 - Remove const pointer qualifications]
Signed-off-by: Ben Hutchings <>
6 years agolocalmodconfig: Use Kbuild files too
Richard Weinberger [Sun, 26 Jul 2015 22:06:55 +0000 (00:06 +0200)]
localmodconfig: Use Kbuild files too

commit c0ddc8c745b7f89c50385fd7aa03c78dc543fa7a upstream.

In kbuild it is allowed to define objects in files named "Makefile"
and "Kbuild".
Currently localmodconfig reads objects only from "Makefile"s and misses
modules like nouveau.

Reported-and-tested-by: Leonidas Spyropoulos <>
Signed-off-by: Richard Weinberger <>
Signed-off-by: Steven Rostedt <>
Signed-off-by: Ben Hutchings <>
6 years agox86/ldt: Correct FPU emulation access to LDT
Juergen Gross [Thu, 6 Aug 2015 17:54:34 +0000 (19:54 +0200)]
x86/ldt: Correct FPU emulation access to LDT

commit 4809146b86c3d41ce588fdb767d021e2a80600dd upstream.

Commit 37868fe113ff ("x86/ldt: Make modify_ldt synchronous")
introduced a new struct ldt_struct anchored at mm->context.ldt.

Adapt the x86 fpu emulation code to use that new structure.

Signed-off-by: Juergen Gross <>
Reviewed-by: Andy Lutomirski <>
Cc: Linus Torvalds <>
Cc: Peter Zijlstra <>
Cc: Thomas Gleixner <>
Signed-off-by: Ingo Molnar <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agox86/ldt: Correct LDT access in single stepping logic
Juergen Gross [Thu, 6 Aug 2015 08:04:38 +0000 (10:04 +0200)]
x86/ldt: Correct LDT access in single stepping logic

commit 136d9d83c07c5e30ac49fc83b27e8c4842f108fc upstream.

Commit 37868fe113ff ("x86/ldt: Make modify_ldt synchronous")
introduced a new struct ldt_struct anchored at mm->context.ldt.

convert_ip_to_linear() was changed to reflect this, but indexing
into the ldt has to be changed as the pointer is no longer void *.

Signed-off-by: Juergen Gross <>
Reviewed-by: Andy Lutomirski <>
Cc: Linus Torvalds <>
Cc: Peter Zijlstra <>
Cc: Thomas Gleixner <>
Signed-off-by: Ingo Molnar <>
Signed-off-by: Ben Hutchings <>
6 years agox86/ldt: Make modify_ldt synchronous
Andy Lutomirski [Thu, 30 Jul 2015 21:31:32 +0000 (14:31 -0700)]
x86/ldt: Make modify_ldt synchronous

commit 37868fe113ff2ba814b3b4eb12df214df555f8dc upstream.

modify_ldt() has questionable locking and does not synchronize
threads.  Improve it: redesign the locking and synchronize all
threads' LDTs using an IPI on all modifications.

This will dramatically slow down modify_ldt in multithreaded
programs, but there shouldn't be any multithreaded programs that
care about modify_ldt's performance in the first place.

This fixes some fallout from the CVE-2015-5157 fixes.

Signed-off-by: Andy Lutomirski <>
Reviewed-by: Borislav Petkov <>
Cc: Andrew Cooper <>
Cc: Andy Lutomirski <>
Cc: Boris Ostrovsky <>
Cc: Borislav Petkov <>
Cc: Brian Gerst <>
Cc: Denys Vlasenko <>
Cc: H. Peter Anvin <>
Cc: Jan Beulich <>
Cc: Konrad Rzeszutek Wilk <>
Cc: Linus Torvalds <>
Cc: Peter Zijlstra <>
Cc: Sasha Levin <>
Cc: Steven Rostedt <>
Cc: Thomas Gleixner <>
Cc: <>
Cc: xen-devel <>
Signed-off-by: Ingo Molnar <>
[bwh: Backported to 3.2:
 - Adjust context
 - Drop comment changes in switch_mm()
 - Drop changes to get_segment_base() in arch/x86/kernel/cpu/perf_event.c
 - Open-code lockless_dereference(), smp_store_release(), on_each_cpu_mask()]
Signed-off-by: Ben Hutchings <>
6 years agonet: Fix skb_set_peeked use-after-free bug
Herbert Xu [Tue, 4 Aug 2015 07:42:47 +0000 (15:42 +0800)]
net: Fix skb_set_peeked use-after-free bug

commit a0a2a6602496a45ae838a96db8b8173794b5d398 upstream.

The commit 738ac1ebb96d02e0d23bc320302a6ea94c612dec ("net: Clone
skb before setting peeked flag") introduced a use-after-free bug
in skb_recv_datagram.  This is because skb_set_peeked may create
a new skb and free the existing one.  As it stands the caller will
continue to use the old freed skb.

This patch fixes it by making skb_set_peeked return the new skb
(or the old one if unchanged).

Fixes: 738ac1ebb96d ("net: Clone skb before setting peeked flag")
Reported-by: Brenden Blanco <>
Signed-off-by: Herbert Xu <>
Tested-by: Brenden Blanco <>
Reviewed-by: Konstantin Khlebnikov <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
6 years agonet: Clone skb before setting peeked flag
Herbert Xu [Mon, 13 Jul 2015 08:04:13 +0000 (16:04 +0800)]
net: Clone skb before setting peeked flag

commit 738ac1ebb96d02e0d23bc320302a6ea94c612dec upstream.

Shared skbs must not be modified and this is crucial for broadcast
and/or multicast paths where we use it as an optimisation to avoid
unnecessary cloning.

The function skb_recv_datagram breaks this rule by setting peeked
without cloning the skb first.  This causes funky races which leads
to double-free.

This patch fixes this by cloning the skb and replacing the skb
in the list when setting skb->peeked.

Fixes: a59322be07c9 ("[UDP]: Only increment counter on first peek/recv")
Reported-by: Konstantin Khlebnikov <>
Signed-off-by: Herbert Xu <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agoocfs2: fix BUG in ocfs2_downconvert_thread_do_work()
Joseph Qi [Thu, 6 Aug 2015 22:46:23 +0000 (15:46 -0700)]
ocfs2: fix BUG in ocfs2_downconvert_thread_do_work()

commit 209f7512d007980fd111a74a064d70a3656079cf upstream.

The "BUG_ON(list_empty(&osb->blocked_lock_list))" in
ocfs2_downconvert_thread_do_work can be triggered in the following case:

ocfs2dc has firstly saved osb->blocked_lock_count to local varibale
processed, and then processes the dentry lockres.  During the dentry
put, it calls iput and then deletes rw, inode and open lockres from
blocked list in ocfs2_mark_lockres_freeing.  And this causes the
variable `processed' to not reflect the number of blocked lockres to be
processed, which triggers the BUG.

Signed-off-by: Joseph Qi <>
Cc: Mark Fasheh <>
Cc: Joel Becker <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
6 years agoMIPS: Make set_pte() SMP safe.
David Daney [Tue, 4 Aug 2015 00:48:43 +0000 (17:48 -0700)]
MIPS: Make set_pte() SMP safe.

commit 46011e6ea39235e4aca656673c500eac81a07a17 upstream.

On MIPS the GLOBAL bit of the PTE must have the same value in any
aligned pair of PTEs.  These pairs of PTEs are referred to as
"buddies".  In a SMP system is is possible for two CPUs to be calling
set_pte() on adjacent PTEs at the same time.  There is a race between
setting the PTE and a different CPU setting the GLOBAL bit in its
buddy PTE.

This race can be observed when multiple CPUs are executing
vmap()/vfree() at the same time.

Make setting the buddy PTE's GLOBAL bit an atomic operation to close
the race condition.

The case of CONFIG_64BIT_PHYS_ADDR && CONFIG_CPU_MIPS32 is *not*

Signed-off-by: David Daney <>
Signed-off-by: Ralf Baechle <>
Signed-off-by: Ben Hutchings <>
6 years agoperf: Fix fasync handling on inherited events
Peter Zijlstra [Thu, 11 Jun 2015 08:32:01 +0000 (10:32 +0200)]
perf: Fix fasync handling on inherited events

commit fed66e2cdd4f127a43fd11b8d92a99bdd429528c upstream.

Vince reported that the fasync signal stuff doesn't work proper for
inherited events. So fix that.

Installing fasync allocates memory and sets filp->f_flags |= FASYNC,
which upon the demise of the file descriptor ensures the allocation is
freed and state is updated.

Now for perf, we can have the events stick around for a while after the
original FD is dead because of references from child events. So we
cannot copy the fasync pointer around. We can however consistently use
the parent's fasync, as that will be updated.

Reported-and-Tested-by: Vince Weaver <>
Signed-off-by: Peter Zijlstra (Intel) <>
Cc: Arnaldo Carvalho deMelo <>
Cc: Linus Torvalds <>
Cc: Peter Zijlstra <>
Cc: Thomas Gleixner <>
Signed-off-by: Ingo Molnar <>
Signed-off-by: Ben Hutchings <>
6 years agords: fix an integer overflow test in rds_info_getsockopt()
Dan Carpenter [Sat, 1 Aug 2015 12:33:26 +0000 (15:33 +0300)]
rds: fix an integer overflow test in rds_info_getsockopt()

commit 468b732b6f76b138c0926eadf38ac88467dcd271 upstream.

"len" is a signed integer.  We check that len is not negative, so it
goes from zero to INT_MAX.  PAGE_SIZE is unsigned long so the comparison
is type promoted to unsigned long.  ULONG_MAX - 4095 is a higher than
INT_MAX so the condition can never be true.

I don't know if this is harmful but it seems safe to limit "len" to
INT_MAX - 4095.

Fixes: a8c879a7ee98 ('RDS: Info and stats')
Signed-off-by: Dan Carpenter <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
6 years agoxhci: fix off by one error in TRB DMA address boundary check
Mathias Nyman [Mon, 3 Aug 2015 13:07:48 +0000 (16:07 +0300)]
xhci: fix off by one error in TRB DMA address boundary check

commit 7895086afde2a05fa24a0e410d8e6b75ca7c8fdd upstream.

We need to check that a TRB is part of the current segment
before calculating its DMA address.

Previously a ring segment didn't use a full memory page, and every
new ring segment got a new memory page, so the off by one
error in checking the upper bound was never seen.

Now that we use a full memory page, 256 TRBs (4096 bytes), the off by one
didn't catch the case when a TRB was the first element of the next segment.

This is triggered if the virtual memory pages for a ring segment are
next to each in increasing order where the ring buffer wraps around and
causes errors like:

[  106.398223] xhci_hcd 0000:00:14.0: ERROR Transfer event TRB DMA ptr not part of current TD ep_index 0 comp_code 1
[  106.398230] xhci_hcd 0000:00:14.0: Looking for event-dma fffd3000 trb-start fffd4fd0 trb-end fffd5000 seg-start fffd4000 seg-end fffd4ff0

The trb-end address is one outside the end-seg address.

Tested-by: Arkadiusz Miśkiewicz <>
Signed-off-by: Mathias Nyman <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
6 years agoMIPS: Fix sched_getaffinity with MT FPAFF enabled
Felix Fietkau [Sat, 18 Jul 2015 22:38:41 +0000 (00:38 +0200)]
MIPS: Fix sched_getaffinity with MT FPAFF enabled

commit 1d62d737555e1378eb62a8bba26644f7d97139d2 upstream.

p->thread.user_cpus_allowed is zero-initialized and is only filled on
the first sched_setaffinity call.

To avoid adding overhead in the task initialization codepath, simply OR
the returned mask in sched_getaffinity with p->cpus_allowed.

Signed-off-by: Felix Fietkau <>
Signed-off-by: Ralf Baechle <>
[bwh: Backported to 3.2: also convert from obsolete cpumask API]
Signed-off-by: Ben Hutchings <>
6 years agotarget: REPORT LUNS should return LUN 0 even for dynamic ACLs
Roland Dreier [Fri, 24 Jul 2015 19:11:46 +0000 (12:11 -0700)]
target: REPORT LUNS should return LUN 0 even for dynamic ACLs

commit 9c395170a559d3b23dad100b01fc4a89d661c698 upstream.

If an initiator doesn't have any real LUNs assigned, we should report
LUN 0 and a LUN list length of 1.  Some versions of Solaris at least
go beserk if we report a LUN list length of 0.

Signed-off-by: Roland Dreier <>
Signed-off-by: Nicholas Bellinger <>
[bwh: Backported to 3.2: adjust filename, context]
Signed-off-by: Ben Hutchings <>
6 years agomd/raid1: extend spinlock to protect raid1_end_read_request against inconsistencies
NeilBrown [Mon, 27 Jul 2015 01:48:52 +0000 (11:48 +1000)]
md/raid1: extend spinlock to protect raid1_end_read_request against inconsistencies

commit 423f04d63cf421ea436bcc5be02543d549ce4b28 upstream.

raid1_end_read_request() assumes that the In_sync bits are consistent
with the ->degaded count.
raid1_spare_active updates the In_sync bit before the ->degraded count
and so exposes an inconsistency, as does error()
So extend the spinlock in raid1_spare_active() and error() to hide those

This should probably be part of
  Commit: 34cab6f42003 ("md/raid1: fix test for 'was read error from
  last working device'.")
as it addresses the same issue.  It fixes the same bug and should go
to -stable for same reasons.

Fixes: 76073054c95b ("md/raid1: clean up read_balance.")
Signed-off-by: NeilBrown <>
Signed-off-by: Ben Hutchings <>
6 years agotarget/iscsi: Fix double free of a TUR followed by a solicited NOPOUT
Alexei Potashnik [Tue, 21 Jul 2015 22:07:56 +0000 (15:07 -0700)]
target/iscsi: Fix double free of a TUR followed by a solicited NOPOUT

commit 9547308bda296b6f69876c840a0291fcfbeddbb8 upstream.

Make sure all non-READ SCSI commands get targ_xfer_tag initialized
to 0xffffffff, not just WRITEs.

Double-free of a TUR cmd object occurs under the following scenario:

1. TUR received (targ_xfer_tag is uninitialized and left at 0)
2. TUR status sent
3. First unsolicited NOPIN is sent to initiator (gets targ_xfer_tag of 0)
4. NOPOUT for NOPIN (with TTT=0) arrives
 - its ExpStatSN acks TUR status, TUR is queued for removal
 - LIO tries to find NOPIN with TTT=0, but finds the same TUR instead,
   TUR is queued for removal for the 2nd time

(Drop unbalanced conditional bracket usage - nab)

Signed-off-by: Alexei Potashnik <>
Signed-off-by: Spencer Baugh <>
Signed-off-by: Nicholas Bellinger <>
[bwh: Backported to 3.2:
 - Adjust context
 - Keep the braces around the if-block]
Signed-off-by: Ben Hutchings <>
6 years agoUSB: sierra: add 1199:68AB device ID
Dirk Behme [Mon, 27 Jul 2015 06:56:05 +0000 (08:56 +0200)]
USB: sierra: add 1199:68AB device ID

commit 74472233233f577eaa0ca6d6e17d9017b6e53150 upstream.

Add support for the Sierra Wireless AR8550 device with
USB descriptor 0x1199, 0x68AB.

It is common with MC879x modules 1199:683c/683d which
also are composite devices with 7 interfaces (0..6)
and also MDM62xx based as the AR8550.

The major difference are only the interface attributes
02/02/01 on interfaces 3 and 4 on the AR8550. They are
vendor specific ff/ff/ff on MC879x modules.

lsusb reports:

Bus 001 Device 004: ID 1199:68ab Sierra Wireless, Inc.
Device Descriptor:
  bLength                18
  bDescriptorType         1
  bcdUSB               2.00
  bDeviceClass            0 (Defined at Interface level)
  bDeviceSubClass         0
  bDeviceProtocol         0
  bMaxPacketSize0        64
  idVendor           0x1199 Sierra Wireless, Inc.
  idProduct          0x68ab
  bcdDevice            0.06
  iManufacturer           3 Sierra Wireless, Incorporated
  iProduct                2 AR8550
  iSerial                 0
  bNumConfigurations      1
  Configuration Descriptor:
    bLength                 9
    bDescriptorType         2
    wTotalLength          198
    bNumInterfaces          7
    bConfigurationValue     1
    iConfiguration          1 Sierra Configuration
    bmAttributes         0xe0
      Self Powered
      Remote Wakeup
    MaxPower                0mA
    Interface Descriptor:
      bLength                 9
      bDescriptorType         4
      bInterfaceNumber        0
      bAlternateSetting       0
      bNumEndpoints           2
      bInterfaceClass       255 Vendor Specific Class
      bInterfaceSubClass    255 Vendor Specific Subclass
      bInterfaceProtocol    255 Vendor Specific Protocol
      iInterface              0
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x81  EP 1 IN
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x01  EP 1 OUT
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
    Interface Descriptor:
      bLength                 9
      bDescriptorType         4
      bInterfaceNumber        1
      bAlternateSetting       0
      bNumEndpoints           2
      bInterfaceClass       255 Vendor Specific Class
      bInterfaceSubClass    255 Vendor Specific Subclass
      bInterfaceProtocol    255 Vendor Specific Protocol
      iInterface              0
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x82  EP 2 IN
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x02  EP 2 OUT
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
    Interface Descriptor:
      bLength                 9
      bDescriptorType         4
      bInterfaceNumber        2
      bAlternateSetting       0
      bNumEndpoints           2
      bInterfaceClass       255 Vendor Specific Class
      bInterfaceSubClass    255 Vendor Specific Subclass
      bInterfaceProtocol    255 Vendor Specific Protocol
      iInterface              0
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x83  EP 3 IN
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x03  EP 3 OUT
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
    Interface Descriptor:
      bLength                 9
      bDescriptorType         4
      bInterfaceNumber        3
      bAlternateSetting       0
      bNumEndpoints           3
      bInterfaceClass         2 Communications
      bInterfaceSubClass      2 Abstract (modem)
      bInterfaceProtocol      1 AT-commands (v.25ter)
      iInterface              0
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x84  EP 4 IN
        bmAttributes            3
          Transfer Type            Interrupt
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0040  1x 64 bytes
        bInterval               5
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x85  EP 5 IN
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x04  EP 4 OUT
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
    Interface Descriptor:
      bLength                 9
      bDescriptorType         4
      bInterfaceNumber        4
      bAlternateSetting       0
      bNumEndpoints           3
      bInterfaceClass         2 Communications
      bInterfaceSubClass      2 Abstract (modem)
      bInterfaceProtocol      1 AT-commands (v.25ter)
      iInterface              0
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x86  EP 6 IN
        bmAttributes            3
          Transfer Type            Interrupt
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0040  1x 64 bytes
        bInterval               5
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x87  EP 7 IN
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x05  EP 5 OUT
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
    Interface Descriptor:
      bLength                 9
      bDescriptorType         4
      bInterfaceNumber        5
      bAlternateSetting       0
      bNumEndpoints           3
      bInterfaceClass       255 Vendor Specific Class
      bInterfaceSubClass    255 Vendor Specific Subclass
      bInterfaceProtocol    255 Vendor Specific Protocol
      iInterface              0
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x88  EP 8 IN
        bmAttributes            3
          Transfer Type            Interrupt
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0040  1x 64 bytes
        bInterval               5
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x89  EP 9 IN
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x06  EP 6 OUT
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
    Interface Descriptor:
      bLength                 9
      bDescriptorType         4
      bInterfaceNumber        6
      bAlternateSetting       0
      bNumEndpoints           3
      bInterfaceClass       255 Vendor Specific Class
      bInterfaceSubClass    255 Vendor Specific Subclass
      bInterfaceProtocol    255 Vendor Specific Protocol
      iInterface              0
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x8a  EP 10 IN
        bmAttributes            3
          Transfer Type            Interrupt
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0040  1x 64 bytes
        bInterval               5
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x8b  EP 11 IN
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
      Endpoint Descriptor:
        bLength                 7
        bDescriptorType         5
        bEndpointAddress     0x07  EP 7 OUT
        bmAttributes            2
          Transfer Type            Bulk
          Synch Type               None
          Usage Type               Data
        wMaxPacketSize     0x0200  1x 512 bytes
        bInterval              32
Device Qualifier (for other device speed):
  bLength                10
  bDescriptorType         6
  bcdUSB               2.00
  bDeviceClass            0 (Defined at Interface level)
  bDeviceSubClass         0
  bDeviceProtocol         0
  bMaxPacketSize0        64
  bNumConfigurations      1
Device Status:     0x0001
  Self Powered

Signed-off-by: Dirk Behme <>
Cc: Lars Melin <>
Signed-off-by: Johan Hovold <>
Signed-off-by: Ben Hutchings <>
6 years agocrypto: ixp4xx - Remove bogus BUG_ON on scattered dst buffer
Herbert Xu [Wed, 22 Jul 2015 10:05:35 +0000 (18:05 +0800)]
crypto: ixp4xx - Remove bogus BUG_ON on scattered dst buffer

commit f898c522f0e9ac9f3177d0762b76e2ab2d2cf9c0 upstream.

This patch removes a bogus BUG_ON in the ablkcipher path that
triggers when the destination buffer is different from the source
buffer and is scattered.

Signed-off-by: Herbert Xu <>
Signed-off-by: Ben Hutchings <>
6 years agoxen/gntdevt: Fix race condition in gntdev_release()
Marek Marczykowski-Górecki [Fri, 26 Jun 2015 01:28:24 +0000 (03:28 +0200)]
xen/gntdevt: Fix race condition in gntdev_release()

commit 30b03d05e07467b8c6ec683ea96b5bffcbcd3931 upstream.

While gntdev_release() is called the MMU notifier is still registered
and can traverse priv->maps list even if no pages are mapped (which is
the case -- gntdev_release() is called after all). But
gntdev_release() will clear that list, so make sure that only one of
those things happens at the same time.

Signed-off-by: Marek Marczykowski-Górecki <>
Signed-off-by: David Vrabel <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agoxen/gntdev: convert priv->lock to a mutex
David Vrabel [Fri, 9 Jan 2015 18:06:12 +0000 (18:06 +0000)]
xen/gntdev: convert priv->lock to a mutex

commit 1401c00e59ea021c575f74612fe2dbba36d6a4ee upstream.

Unmapping may require sleeping and we unmap while holding priv->lock, so
convert it to a mutex.

Signed-off-by: David Vrabel <>
Reviewed-by: Stefano Stabellini <>
[bwh: Backported to 3.2:
 - Adjust context
 - Drop changes to functions we don't have]
Signed-off-by: Ben Hutchings <>
6 years agojbd2: protect all log tail updates with j_checkpoint_mutex
Jan Kara [Tue, 13 Mar 2012 19:43:04 +0000 (15:43 -0400)]
jbd2: protect all log tail updates with j_checkpoint_mutex

commit a78bb11d7acd525623c6a0c2ff4e213d527573fa upstream.

There are some log tail updates that are not protected by j_checkpoint_mutex.
Some of these are harmless because they happen during startup or shutdown but
updates in jbd2_journal_commit_transaction() and jbd2_journal_flush() can
really race with other log tail updates (e.g. someone doing
jbd2_journal_flush() with someone running jbd2_cleanup_journal_tail()). So
protect all log tail updates with j_checkpoint_mutex.

Signed-off-by: Jan Kara <>
Signed-off-by: "Theodore Ts'o" <>
[bwh: Backported to 3.2:
 - Adjust context
 - Add unlock on the error path in jbd2_journal_flush()]
Signed-off-by: Ben Hutchings <>
Cc: Bartosz Kwitniewski <>
6 years agopktgen: Require CONFIG_INET due to use of IPv4 checksum function
Thomas Graf [Mon, 29 Jul 2013 11:44:15 +0000 (13:44 +0200)]
pktgen: Require CONFIG_INET due to use of IPv4 checksum function

commit ffd756b3174e496cf6f3c5458c434e31d2cd48b0 upstream.

Unlike for IPv6, the IPv4 checksum functions are only available
if CONFIG_INET is set.

Reported-by: kbuild test robot <>
Signed-off-by: Thomas Graf <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
6 years agoipv6: Fix build failure when CONFIG_INET disabled
Ben Hutchings [Tue, 18 Aug 2015 18:31:23 +0000 (20:31 +0200)]
ipv6: Fix build failure when CONFIG_INET disabled

output_core.c, added in 3.2.66, is only needed and can only be
compiled when CONFIG_INET is enabled.

The condition in the Makefile is already correct upstream.

Reported-by: kbuild test robot <>
Signed-off-by: Ben Hutchings <>
6 years agoLinux 3.2.71 v3.2.71
Ben Hutchings [Wed, 12 Aug 2015 14:33:24 +0000 (16:33 +0200)]
Linux 3.2.71

6 years agox86/xen: Probe target addresses in set_aliased_prot() before the hypercall
Andy Lutomirski [Thu, 30 Jul 2015 21:31:31 +0000 (14:31 -0700)]
x86/xen: Probe target addresses in set_aliased_prot() before the hypercall

commit aa1acff356bbedfd03b544051f5b371746735d89 upstream.

The update_va_mapping hypercall can fail if the VA isn't present
in the guest's page tables.  Under certain loads, this can
result in an OOPS when the target address is in unpopulated vmap

While we're at it, add comments to help explain what's going on.

This isn't a great long-term fix.  This code should probably be
changed to use something like set_memory_ro.

Signed-off-by: Andy Lutomirski <>
Cc: Andrew Cooper <>
Cc: Andy Lutomirski <>
Cc: Boris Ostrovsky <>
Cc: Borislav Petkov <>
Cc: Brian Gerst <>
Cc: David Vrabel <>
Cc: Denys Vlasenko <>
Cc: H. Peter Anvin <>
Cc: Jan Beulich <>
Cc: Konrad Rzeszutek Wilk <>
Cc: Linus Torvalds <>
Cc: Peter Zijlstra <>
Cc: Sasha Levin <>
Cc: Steven Rostedt <>
Cc: Thomas Gleixner <>
Cc: <>
Cc: xen-devel <>
Signed-off-by: Ingo Molnar <>
Signed-off-by: Ben Hutchings <>
6 years agodrm/radeon/combios: add some validation of lvds values
Alex Deucher [Mon, 27 Jul 2015 23:24:31 +0000 (19:24 -0400)]
drm/radeon/combios: add some validation of lvds values

commit 0a90a0cff9f429f886f423967ae053150dce9259 upstream.

Fixes a broken hsync start value uncovered by:
(drm: Perform basic sanity checks on probed modes)

The driver handled the bad hsync start elsewhere, but
the above commit prevented it from getting added.


Signed-off-by: Alex Deucher <>
Signed-off-by: Ben Hutchings <>
6 years agoALSA: usb-audio: add dB range mapping for some devices
Yao-Wen Mao [Wed, 29 Jul 2015 07:13:54 +0000 (15:13 +0800)]
ALSA: usb-audio: add dB range mapping for some devices

commit 2d1cb7f658fb9c3ba8f9dab8aca297d4dfdec835 upstream.

Add the correct dB ranges of Bose Companion 5 and Drangonfly DAC 1.2.

Signed-off-by: Yao-Wen Mao <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Ben Hutchings <>
6 years agovhost: actually track log eventfd file
Marc-André Lureau [Fri, 17 Jul 2015 13:32:03 +0000 (15:32 +0200)]
vhost: actually track log eventfd file

commit 7932c0bd7740f4cd2aa168d3ce0199e7af7d72d5 upstream.

While reviewing vhost log code, I found out that log_file is never
set. Note: I haven't tested the change (QEMU doesn't use LOG_FD yet).

Signed-off-by: Marc-André Lureau <>
Signed-off-by: Michael S. Tsirkin <>
Signed-off-by: Ben Hutchings <>
6 years agoniu: don't count tx error twice in case of headroom realloc fails
Jiri Pirko [Thu, 23 Jul 2015 10:20:37 +0000 (12:20 +0200)]
niu: don't count tx error twice in case of headroom realloc fails

commit 42288830494cd51873ca745a7a229023df061226 upstream.

Fixes: a3138df9 ("[NIU]: Add Sun Neptune ethernet driver.")
Signed-off-by: Jiri Pirko <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
6 years agoiscsi-target: Fix use-after-free during TPG session shutdown
Nicholas Bellinger [Wed, 22 Jul 2015 07:24:09 +0000 (00:24 -0700)]
iscsi-target: Fix use-after-free during TPG session shutdown

commit 417c20a9bdd1e876384127cf096d8ae8b559066c upstream.

This patch fixes a use-after-free bug in iscsit_release_sessions_for_tpg()
where se_portal_group->session_lock was incorrectly released/re-acquired
while walking the active se_portal_group->tpg_sess_list.

The can result in a NULL pointer dereference when iscsit_close_session()
shutdown happens in the normal path asynchronously to this code, causing
a bogus dereference of an already freed list entry to occur.

To address this bug, walk the session list checking for the same state
as before, but move entries to a local list to avoid dropping the lock
while walking the active list.

As before, signal using iscsi_session->session_restatement=1 for those
list entries to be released locally by iscsit_free_session() code.

Reported-by: Sunilkumar Nadumuttlu <>
Cc: Sunilkumar Nadumuttlu <>
Signed-off-by: Nicholas Bellinger <>
Signed-off-by: Ben Hutchings <>
6 years agomd/raid1: fix test for 'was read error from last working device'.
NeilBrown [Thu, 23 Jul 2015 23:22:16 +0000 (09:22 +1000)]
md/raid1: fix test for 'was read error from last working device'.

commit 34cab6f42003cb06f48f86a86652984dec338ae9 upstream.

When we get a read error from the last working device, we don't
try to repair it, and don't fail the device.  We simple report a
read error to the caller.

However the current test for 'is this the last working device' is
When there is only one fully working device, it assumes that a
non-faulty device is that device.  However a spare which is rebuilding
would be non-faulty but so not the only working device.

So change the test from "!Faulty" to "In_sync".  If ->degraded says
there is only one fully working device and this device is in_sync,
this must be the one.

This bug has existed since we allowed read_balance to read from
a recovering spare in v3.0

Reported-and-tested-by: Alexander Lyakas <>
Fixes: 76073054c95b ("md/raid1: clean up read_balance.")
Signed-off-by: NeilBrown <>
Signed-off-by: Ben Hutchings <>
6 years agoInput: usbtouchscreen - avoid unresponsive TSC-30 touch screen
Bernhard Bender [Thu, 23 Jul 2015 20:58:08 +0000 (13:58 -0700)]
Input: usbtouchscreen - avoid unresponsive TSC-30 touch screen

commit 968491709e5b1aaf429428814fff3d932fa90b60 upstream.

This patch fixes a problem in the usbtouchscreen driver for DMC TSC-30
touch screen.  Due to a missing delay between the RESET and SET_RATE
commands, the touch screen may become unresponsive during system startup or
driver loading.

According to the DMC documentation, a delay is needed after the RESET
command to allow the chip to complete its internal initialization. As this
delay is not guaranteed, we had a system where the touch screen
occasionally did not send any touch data. There was no other indication of
the problem.

The patch fixes the problem by adding a 150ms delay between the RESET and
SET_RATE commands.

Suggested-by: Jakob Mustafa <>
Signed-off-by: Bernhard Bender <>
Signed-off-by: Dmitry Torokhov <>
Signed-off-by: Ben Hutchings <>
6 years agotile: use free_bootmem_late() for initrd
Chris Metcalf [Thu, 23 Jul 2015 18:11:09 +0000 (14:11 -0400)]
tile: use free_bootmem_late() for initrd

commit 3f81d2447b37ac697b3c600039f2c6b628c06e21 upstream.

We were previously using free_bootmem() and just getting lucky
that nothing too bad happened.

Signed-off-by: Chris Metcalf <>
Signed-off-by: Ben Hutchings <>
6 years agousb-storage: ignore ZTE MF 823 card reader in mode 0x1225
Oliver Neukum [Mon, 6 Jul 2015 11:12:32 +0000 (13:12 +0200)]
usb-storage: ignore ZTE MF 823 card reader in mode 0x1225

commit 5fb2c782f451a4fb9c19c076e2c442839faf0f76 upstream.

This device automatically switches itself to another mode (0x1405)
unless the specific access pattern of Windows is followed in its
initial mode. That makes a dirty unmount of the internal storage
devices inevitable if they are mounted. So the card reader of
such a device should be ignored, lest an unclean removal become

This replaces an earlier patch that ignored all LUNs of this device.
That patch was overly broad.

Signed-off-by: Oliver Neukum <>
Reviewed-by: Lars Melin <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
6 years agoxhci: do not report PLC when link is in internal resume state
Zhuang Jin Can [Tue, 21 Jul 2015 14:20:31 +0000 (17:20 +0300)]
xhci: do not report PLC when link is in internal resume state

commit aca3a0489ac019b58cf32794d5362bb284cb9b94 upstream.

Port link change with port in resume state should not be
reported to usbcore, as this is an internal state to be
handled by xhci driver. Reporting PLC to usbcore may
cause usbcore clearing PLC first and port change event irq
won't be generated.

Signed-off-by: Zhuang Jin Can <>
Signed-off-by: Mathias Nyman <>
Signed-off-by: Greg Kroah-Hartman <>
[bwh: Backported to 3.2:
 - Adjust indentation
 - s/raw_port_status/temp/]
Signed-off-by: Ben Hutchings <>
6 years agoxhci: report U3 when link is in resume state
Zhuang Jin Can [Tue, 21 Jul 2015 14:20:29 +0000 (17:20 +0300)]
xhci: report U3 when link is in resume state

commit 243292a2ad3dc365849b820a64868927168894ac upstream.

xhci_hub_report_usb3_link_state() returns pls as U0 when the link
is in resume state, and this causes usb core to think the link is in
U0 while actually it's in resume state. When usb core transfers
control request on the link, it fails with TRB error as the link
is not ready for transfer.

To fix the issue, report U3 when the link is in resume state, thus
usb core knows the link it's not ready for transfer.

Signed-off-by: Zhuang Jin Can <>
Signed-off-by: Mathias Nyman <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
6 years agoxhci: Calculate old endpoints correctly on device reset
Brian Campbell [Tue, 21 Jul 2015 14:20:28 +0000 (17:20 +0300)]
xhci: Calculate old endpoints correctly on device reset

commit 326124a027abc9a7f43f72dc94f6f0f7a55b02b3 upstream.

When resetting a device the number of active TTs may need to be
corrected by xhci_update_tt_active_eps, but the number of old active
endpoints supplied to it was always zero, so the number of TTs and the
bandwidth reserved for them was not updated, and could rise

This affected systems using Intel's Patherpoint chipset, which rely on
software bandwidth checking.  For example, a Lenovo X230 would lose the
ability to use ports on the docking station after enough suspend/resume
cycles because the bandwidth calculated would rise with every cycle when
a suitable device is attached.

The correct number of active endpoints is calculated in the same way as
in xhci_reserve_bandwidth.

Signed-off-by: Brian Campbell <>
Signed-off-by: Mathias Nyman <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
6 years agousb: xhci: Bugfix for NULL pointer deference in xhci_endpoint_init() function
AMAN DEEP [Tue, 21 Jul 2015 14:20:27 +0000 (17:20 +0300)]
usb: xhci: Bugfix for NULL pointer deference in xhci_endpoint_init() function

commit 3496810663922617d4b706ef2780c279252ddd6a upstream.

virt_dev->num_cached_rings counts on freed ring and is not updated
correctly. In xhci_free_or_cache_endpoint_ring() function, the free ring
is added into cache and then num_rings_cache is incremented as below:
virt_dev->ring_cache[rings_cached] =
here, free ring pointer is added to a current index and then
index is incremented.
So current index always points to empty location in the ring cache.
For getting available free ring, current index should be decremented
first and then corresponding ring buffer value should be taken from ring

But In function xhci_endpoint_init(), the num_rings_cached index is
accessed before decrement.
virt_dev->eps[ep_index].new_ring =
virt_dev->ring_cache[virt_dev->num_rings_cached] = NULL;
This is bug in manipulating the index of ring cache.
And it should be as below:
virt_dev->eps[ep_index].new_ring =
virt_dev->ring_cache[virt_dev->num_rings_cached] = NULL;

Signed-off-by: Aman Deep <>
Signed-off-by: Mathias Nyman <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
6 years agonetfilter: nf_conntrack: Support expectations in different zones
Joe Stringer [Wed, 22 Jul 2015 04:37:31 +0000 (21:37 -0700)]
netfilter: nf_conntrack: Support expectations in different zones

commit 4b31814d20cbe5cd4ccf18089751e77a04afe4f2 upstream.

When zones were originally introduced, the expectation functions were
all extended to perform lookup using the zone. However, insertion was
not modified to check the zone. This means that two expectations which
are intended to apply for different connections that have the same tuple
but exist in different zones cannot both be tracked.

Fixes: 5d0aa2ccd4 (netfilter: nf_conntrack: add support for "conntrack zones")
Signed-off-by: Joe Stringer <>
Signed-off-by: Pablo Neira Ayuso <>
Signed-off-by: Ben Hutchings <>
6 years agousb: dwc3: Reset the transfer resource index on SET_INTERFACE
John Youn [Mon, 17 Sep 2001 07:00:00 +0000 (00:00 -0700)]
usb: dwc3: Reset the transfer resource index on SET_INTERFACE

commit aebda618718157a69c0dc0adb978d69bc2b8723c upstream.

This fixes an issue introduced in commit b23c843992b6 (usb: dwc3:
gadget: fix DEPSTARTCFG for non-EP0 EPs) that made sure we would
only use DEPSTARTCFG once per SetConfig.

The trick is that we should use one DEPSTARTCFG per SetConfig *OR*
SetInterface. SetInterface was completely missed from the original

This problem became aparent after commit 76e838c9f776 (usb: dwc3:
gadget: return error if command sent to DEPCMD register fails)
added checking of the return status of device endpoint commands.

'Set Endpoint Transfer Resource' command was caught failing
occasionally. This is because the Transfer Resource
Index was not getting reset during a SET_INTERFACE request.

Finally, to fix the issue, was we have to do is make sure that
our start_config_issued flag gets reset whenever we receive a
SetInterface request.

To verify the problem (and its fix), all we have to do is run
test 9 from testusb with 'testusb -t 9 -s 2048 -a -c 5000'.

Tested-by: Huang Rui <>
Tested-by: Subbaraya Sundeep Bhatta <>
Fixes: b23c843992b6 (usb: dwc3: gadget: fix DEPSTARTCFG for non-EP0 EPs)
Signed-off-by: John Youn <>
Signed-off-by: Felipe Balbi <>
[bwh: Backported to 3.2: use dev_vdbg() instead of dwc3_trace()]
Signed-off-by: Ben Hutchings <>
6 years agoinet: frags: fix defragmented packet's IP header for af_packet
Edward Hyunkoo Jee [Tue, 21 Jul 2015 07:43:59 +0000 (09:43 +0200)]
inet: frags: fix defragmented packet's IP header for af_packet

commit 0848f6428ba3a2e42db124d41ac6f548655735bf upstream.

When ip_frag_queue() computes positions, it assumes that the passed
sk_buff does not contain L2 headers.

However, when PACKET_FANOUT_FLAG_DEFRAG is used, IP reassembly
functions can be called on outgoing packets that contain L2 headers.

Also, IPv4 checksum is not corrected after reassembly.

Fixes: 7736d33f4262 ("packet: Add pre-defragmentation support for ipv4 fanouts.")
Signed-off-by: Edward Hyunkoo Jee <>
Signed-off-by: Eric Dumazet <>
Cc: Willem de Bruijn <>
Cc: Jerry Chu <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agomac80211: clear subdir_stations when removing debugfs
Tom Hughes [Mon, 29 Jun 2015 18:41:49 +0000 (19:41 +0100)]
mac80211: clear subdir_stations when removing debugfs

commit 4479004e6409087d1b4986881dc98c6c15dffb28 upstream.

If we don't do this, and we then fail to recreate the debugfs
directory during a mode change, then we will fail later trying
to add stations to this now bogus directory:

BUG: unable to handle kernel NULL pointer dereference at 0000006c
IP: [<c0a92202>] mutex_lock+0x12/0x30
Call Trace:
[<c0678ab4>] start_creating+0x44/0xc0
[<c0679203>] debugfs_create_dir+0x13/0xf0
[<f8a938ae>] ieee80211_sta_debugfs_add+0x6e/0x490 [mac80211]

Signed-off-by: Tom Hughes <>
Signed-off-by: Johannes Berg <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years agodrm/radeon: Don't flush the GART TLB if rdev->gart.ptr == NULL
Michel Dänzer [Fri, 3 Jul 2015 01:02:27 +0000 (10:02 +0900)]
drm/radeon: Don't flush the GART TLB if rdev->gart.ptr == NULL

commit 233709d2cd6bbaaeda0aeb8d11f6ca7f98563b39 upstream.

This can be the case when the GPU is powered off, e.g. via vgaswitcheroo
or runpm. When the GPU is powered up again, radeon_gart_table_vram_pin
flushes the TLB after setting rdev->gart.ptr to non-NULL.

Fixes panic on powering off R7xx GPUs.

Reviewed-by: Christian König <>
Signed-off-by: Michel Dänzer <>
Signed-off-by: Alex Deucher <>
Signed-off-by: Ben Hutchings <>
6 years agodatagram: Factor out sk queue referencing
Pavel Emelyanov [Tue, 21 Feb 2012 07:30:33 +0000 (07:30 +0000)]
datagram: Factor out sk queue referencing

commit 4934b0329f7150dcb5f90506860e2db32274c755 upstream.

This makes lines shorter and simplifies further patching.

Signed-off-by: Pavel Emelyanov <>
Acked-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
[bwh: Prerequisite of "net: Clone skb before setting peeked flag"]
Signed-off-by: Ben Hutchings <>
6 years agolibata: increase the timeout when setting transfer mode
Mikulas Patocka [Wed, 8 Jul 2015 17:06:12 +0000 (13:06 -0400)]
libata: increase the timeout when setting transfer mode

commit d531be2ca2f27cca5f041b6a140504999144a617 upstream.

I have a ST4000DM000 disk. If Linux is booted while the disk is spun down,
the command that sets transfer mode causes the disk to spin up. The
spin-up takes longer than the default 5s timeout, so the command fails and
timeout is reported.

Fix this by increasing the timeout to 15s, which is enough for the disk to
spin up.

Signed-off-by: Mikulas Patocka <>
Signed-off-by: Tejun Heo <>
Signed-off-by: Ben Hutchings <>
6 years agolibata: force disable trim for SuperSSpeed S238
Arne Fitzenreiter [Wed, 15 Jul 2015 11:54:37 +0000 (13:54 +0200)]
libata: force disable trim for SuperSSpeed S238

commit cda57b1b05cf7b8b99ab4b732bea0b05b6c015cc upstream.

This device loses blocks, often the partition table area, on trim.
Disable TRIM.

Signed-off-by: Arne Fitzenreiter <>
Signed-off-by: Tejun Heo <>
Signed-off-by: Ben Hutchings <>
6 years agolibata: add ATA_HORKAGE_NOTRIM
Arne Fitzenreiter [Wed, 15 Jul 2015 11:54:36 +0000 (13:54 +0200)]

commit 71d126fd28de2d4d9b7b2088dbccd7ca62fad6e0 upstream.

Some devices lose data on TRIM whether queued or not.  This patch adds
a horkage to disable TRIM.

tj: Collapsed unnecessary if() nesting.

Signed-off-by: Arne Fitzenreiter <>
Signed-off-by: Tejun Heo <>
[bwh: Backported to 3.2:
 - Adjust context
 - Drop change to show_ata_dev_trim()]
Signed-off-by: Ben Hutchings <>
6 years agolibata: add ATA_HORKAGE_BROKEN_FPDMA_AA quirk for HP 250GB SATA disk VB0250EAVER
Aleksei Mamlin [Wed, 1 Jul 2015 10:48:30 +0000 (13:48 +0300)]
libata: add ATA_HORKAGE_BROKEN_FPDMA_AA quirk for HP 250GB SATA disk VB0250EAVER

commit 08c85d2a599d967ede38a847f5594447b6100642 upstream.

Enabling AA on HP 250GB SATA disk VB0250EAVER causes errors:

[    3.788362] ata3.00: failed to enable AA (error_mask=0x1)
[    3.789243] ata3.00: failed to enable AA (error_mask=0x1)

Add the ATA_HORKAGE_BROKEN_FPDMA_AA for this specific harddisk.

tj: Collected FPDMA_AA entries and updated comment.

Signed-off-by: Aleksei Mamlin <>
Signed-off-by: Tejun Heo <>
Signed-off-by: Ben Hutchings <>
6 years agoata: pmp: add quirk for Marvell 4140 SATA PMP
Lior Amsalem [Tue, 30 Jun 2015 14:09:49 +0000 (16:09 +0200)]
ata: pmp: add quirk for Marvell 4140 SATA PMP

commit 945b47441d83d2392ac9f984e0267ad521f24268 upstream.

This commit adds the necessary quirk to make the Marvell 4140 SATA PMP
work properly. This PMP doesn't like SRST on port number 4 (the host
port) so this commit marks this port as not supporting SRST.

Signed-off-by: Lior Amsalem <>
Reviewed-by: Nadav Haklai <>
Signed-off-by: Thomas Petazzoni <>
Signed-off-by: Tejun Heo <>
Signed-off-by: Ben Hutchings <>
6 years agords: rds_ib_device.refcount overflow
Wengang Wang [Mon, 6 Jul 2015 06:35:11 +0000 (14:35 +0800)]
rds: rds_ib_device.refcount overflow

commit 4fabb59449aa44a585b3603ffdadd4c5f4d0c033 upstream.

Fixes: 3e0249f9c05c ("RDS/IB: add refcount tracking to struct rds_ib_device")

There lacks a dropping on rds_ib_device.refcount in case rds_ib_alloc_fmr
failed(mr pool running out). this lead to the refcount overflow.

A complain in line 117(see following) is seen. From vmcore:
s_ib_rdma_mr_pool_depleted is 2147485544 and rds_ibdev->refcount is -2147475448.
That is the evidence the mr pool is used up. so rds_ib_alloc_fmr is very likely
to return ERR_PTR(-EAGAIN).

115 void rds_ib_dev_put(struct rds_ib_device *rds_ibdev)
116 {
117         BUG_ON(atomic_read(&rds_ibdev->refcount) <= 0);
118         if (atomic_dec_and_test(&rds_ibdev->refcount))
119                 queue_work(rds_wq, &rds_ibdev->free_work);
120 }

fix is to drop refcount when rds_ib_alloc_fmr failed.

Signed-off-by: Wengang Wang <>
Reviewed-by: Haggai Eran <>
Signed-off-by: Doug Ledford <>
Signed-off-by: Ben Hutchings <>
6 years agoBtrfs: fix file corruption after cloning inline extents
Filipe Manana [Tue, 14 Jul 2015 15:09:39 +0000 (16:09 +0100)]
Btrfs: fix file corruption after cloning inline extents

commit ed958762644b404654a6f5d23e869f496fe127c6 upstream.

Using the clone ioctl (or extent_same ioctl, which calls the same extent
cloning function as well) we end up allowing copy an inline extent from
the source file into a non-zero offset of the destination file. This is
something not expected and that the btrfs code is not prepared to deal
with - all inline extents must be at a file offset equals to 0.

For example, the following excerpt of a test case for fstests triggers
a crash/BUG_ON() on a write operation after an inline extent is cloned
into a non-zero offset:

  _scratch_mkfs >>$seqres.full 2>&1

  # Create our test files. File foo has the same 2K of data at offset 4K
  # as file bar has at its offset 0.
  $XFS_IO_PROG -f -s -c "pwrite -S 0xaa 0 4K" \
      -c "pwrite -S 0xbb 4k 2K" \
      -c "pwrite -S 0xcc 8K 4K" \
      $SCRATCH_MNT/foo | _filter_xfs_io

  # File bar consists of a single inline extent (2K size).
  $XFS_IO_PROG -f -s -c "pwrite -S 0xbb 0 2K" \
     $SCRATCH_MNT/bar | _filter_xfs_io

  # Now call the clone ioctl to clone the extent of file bar into file
  # foo at its offset 4K. This made file foo have an inline extent at
  # offset 4K, something which the btrfs code can not deal with in future
  # IO operations because all inline extents are supposed to start at an
  # offset of 0, resulting in all sorts of chaos.
  # So here we validate that clone ioctl returns an EOPNOTSUPP, which is
  # what it returns for other cases dealing with inlined extents.
  $CLONER_PROG -s 0 -d $((4 * 1024)) -l $((2 * 1024)) \

  # Because of the inline extent at offset 4K, the following write made
  # the kernel crash with a BUG_ON().
  $XFS_IO_PROG -c "pwrite -S 0xdd 6K 2K" $SCRATCH_MNT/foo | _filter_xfs_io


The stack trace of the BUG_ON() triggered by the last write is:

  [152154.035903] ------------[ cut here ]------------
  [152154.036424] kernel BUG at mm/page-writeback.c:2286!
  [152154.036424] invalid opcode: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
  [152154.036424] Modules linked in: btrfs dm_flakey dm_mod crc32c_generic xor raid6_pq nfsd auth_rpcgss oid_registry nfs_acl nfs lockd grace fscache sunrpc loop fuse parport_pc acpi_cpu$
  [152154.036424] CPU: 2 PID: 17873 Comm: xfs_io Tainted: G        W       4.1.0-rc6-btrfs-next-11+ #2
  [152154.036424] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 04/01/2014
  [152154.036424] task: ffff880429f70990 ti: ffff880429efc000 task.ti: ffff880429efc000
  [152154.036424] RIP: 0010:[<ffffffff8111a9d5>]  [<ffffffff8111a9d5>] clear_page_dirty_for_io+0x1e/0x90
  [152154.036424] RSP: 0018:ffff880429effc68  EFLAGS: 00010246
  [152154.036424] RAX: 0200000000000806 RBX: ffffea0006a6d8f0 RCX: 0000000000000001
  [152154.036424] RDX: 0000000000000000 RSI: ffffffff81155d1b RDI: ffffea0006a6d8f0
  [152154.036424] RBP: ffff880429effc78 R08: ffff8801ce389fe0 R09: 0000000000000001
  [152154.036424] R10: 0000000000002000 R11: ffffffffffffffff R12: ffff8800200dce68
  [152154.036424] R13: 0000000000000000 R14: ffff8800200dcc88 R15: ffff8803d5736d80
  [152154.036424] FS:  00007fbf119f6700(0000) GS:ffff88043d280000(0000) knlGS:0000000000000000
  [152154.036424] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  [152154.036424] CR2: 0000000001bdc000 CR3: 00000003aa555000 CR4: 00000000000006e0
  [152154.036424] Stack:
  [152154.036424]  ffff8803d5736d80 0000000000000001 ffff880429effcd8 ffffffffa04e97c1
  [152154.036424]  ffff880429effd68 ffff880429effd60 0000000000000001 ffff8800200dc9c8
  [152154.036424]  0000000000000001 ffff8800200dcc88 0000000000000000 0000000000001000
  [152154.036424] Call Trace:
  [152154.036424]  [<ffffffffa04e97c1>] lock_and_cleanup_extent_if_need+0x147/0x18d [btrfs]
  [152154.036424]  [<ffffffffa04ea82c>] __btrfs_buffered_write+0x245/0x4c8 [btrfs]
  [152154.036424]  [<ffffffffa04ed14b>] ? btrfs_file_write_iter+0x150/0x3e0 [btrfs]
  [152154.036424]  [<ffffffffa04ed15a>] ? btrfs_file_write_iter+0x15f/0x3e0 [btrfs]
  [152154.036424]  [<ffffffffa04ed2c7>] btrfs_file_write_iter+0x2cc/0x3e0 [btrfs]
  [152154.036424]  [<ffffffff81165a4a>] __vfs_write+0x7c/0xa5
  [152154.036424]  [<ffffffff81165f89>] vfs_write+0xa0/0xe4
  [152154.036424]  [<ffffffff81166855>] SyS_pwrite64+0x64/0x82
  [152154.036424]  [<ffffffff81465197>] system_call_fastpath+0x12/0x6f
  [152154.036424] Code: 48 89 c7 e8 0f ff ff ff 5b 41 5c 5d c3 0f 1f 44 00 00 55 48 89 e5 41 54 53 48 89 fb e8 ae ef 00 00 49 89 c4 48 8b 03 a8 01 75 02 <0f> 0b 4d 85 e4 74 59 49 8b 3c 2$
  [152154.036424] RIP  [<ffffffff8111a9d5>] clear_page_dirty_for_io+0x1e/0x90
  [152154.036424]  RSP <ffff880429effc68>
  [152154.242621] ---[ end trace e3d3376b23a57041 ]---

Fix this by returning the error EOPNOTSUPP if an attempt to copy an
inline extent into a non-zero offset happens, just like what is done for
other scenarios that would require copying/splitting inline extents,
which were introduced by the following commits:

   00fdf13a2e9f ("Btrfs: fix a crash of clone with inline extents's split")
   3f9e3df8da3c ("btrfs: replace error code from btrfs_drop_extents")

Signed-off-by: Filipe Manana <>
[bwh: Backported to 3.2: test new_key.offset as last_dest_end isn't defined
 in this function]
Signed-off-by: Ben Hutchings <>
6 years agos390/process: fix sfpc inline assembly
Heiko Carstens [Mon, 6 Jul 2015 13:02:37 +0000 (15:02 +0200)]
s390/process: fix sfpc inline assembly

commit e47994dd44bcb4a77b4152bd0eada585934703c0 upstream.

The sfpc inline assembly within execve_tail() may incorrectly set bits
28-31 of the sfpc instruction to a value which is not zero.
These bits however are currently unused and therefore should be zero
so we won't get surprised if these bits will be used in the future.

Therefore remove the second operand from the inline assembly.

Signed-off-by: Heiko Carstens <>
Signed-off-by: Martin Schwidefsky <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
6 years ago9p: don't leave a half-initialized inode sitting around
Al Viro [Sun, 12 Jul 2015 14:34:29 +0000 (10:34 -0400)]
9p: don't leave a half-initialized inode sitting around

commit 0a73d0a204a4a04a1e110539c5a524ae51f91d6d upstream.

Signed-off-by: Al Viro <>
Signed-off-by: Ben Hutchings <>
6 years agonet: call rcu_read_lock early in process_backlog
Julian Anastasov [Thu, 9 Jul 2015 06:59:10 +0000 (09:59 +0300)]
net: call rcu_read_lock early in process_backlog

commit 2c17d27c36dcce2b6bf689f41a46b9e909877c21 upstream.

Incoming packet should be either in backlog queue or
in RCU read-side section. Otherwise, the final sequence of
flush_backlog() and synchronize_net() may miss packets
that can run without device reference:

CPU 1                  CPU 2
                       skb->dev: no reference

on_each_cpu for
flush_backlog =>       IPI(hardirq): flush_backlog
                       - packet not found in backlog

                       CPU delayed ...
- no ongoing RCU
read-side sections

rcu_barrier: no
ongoing callbacks
                       - too late
free dev
                       process packet for freed dev

Fixes: 6e583ce5242f ("net: eliminate refcounting in backlog queue")
Cc: Eric W. Biederman <>
Cc: Stephen Hemminger <>
Signed-off-by: Julian Anastasov <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2:
 - Adjust context
 - No need to rename the label in __netif_receive_skb()]
Signed-off-by: Ben Hutchings <>
6 years agonet: do not process device backlog during unregistration
Julian Anastasov [Thu, 9 Jul 2015 06:59:09 +0000 (09:59 +0300)]
net: do not process device backlog during unregistration

commit e9e4dd3267d0c5234c5c0f47440456b10875dec9 upstream.

commit 381c759d9916 ("ipv4: Avoid crashing in ip_error")
fixes a problem where processed packet comes from device
with destroyed inetdev (dev->ip_ptr). This is not expected
because inetdev_destroy is called in NETDEV_UNREGISTER
phase and packets should not be processed after
dev_close_many() and synchronize_net(). Above fix is still
required because inetdev_destroy can be called for other
reasons. But it shows the real problem: backlog can keep
packets for long time and they do not hold reference to
device. Such packets are then delivered to upper levels
at the same time when device is unregistered.
Calling flush_backlog after NETDEV_UNREGISTER_FINAL still
accounts all packets from backlog but before that some packets
continue to be delivered to upper levels long after the
synchronize_net call which is supposed to wait the last
ones. Also, as Eric pointed out, processed packets, mostly
from other devices, can continue to add new packets to backlog.

Fix the problem by moving flush_backlog early, after the
device driver is stopped and before the synchronize_net() call.
Then use netif_running check to make sure we do not add more
packets to backlog. We have to do it in enqueue_to_backlog
context when the local IRQ is disabled. As result, after the
flush_backlog and synchronize_net sequence all packets
should be accounted.

Thanks to Eric W. Biederman for the test script and his
valuable feedback!

Reported-by: Vittorio Gambaletta <>
Fixes: 6e583ce5242f ("net: eliminate refcounting in backlog queue")
Cc: Eric W. Biederman <>
Cc: Stephen Hemminger <>
Signed-off-by: Julian Anastasov <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>