8 years agoipv6: restrict neighbor entry creation to output flow
Marcelo Ricardo Leitner [Tue, 8 Oct 2013 14:41:13 +0000 (16:41 +0200)]
ipv6: restrict neighbor entry creation to output flow

This patch is based on 3.2.y branch, the one used by reporter. Please let me
know if it should be different. Thanks.

The patch which introduced the regression was applied on stables:
3.0.64 3.4.31 3.7.8 3.2.39

The patch which introduced the regression was for stable trees only.


Commit 0d6a77079c475033cb622c07c5a880b392ef664e "ipv6: do not create
neighbor entries for local delivery" introduced a regression on
which routes to local delivery would not work anymore. Like this:

    $ ip -6 route add local 2001::/64 dev lo
    $ ping6 -c1 2001::9
    PING 2001::9(2001::9) 56 data bytes
    ping: sendmsg: Invalid argument

As this is a local delivery, that commit would not allow the creation of a
neighbor entry and thus the packet cannot be sent.

But as TPROXY scenario actually needs to avoid the neighbor entry creation only
for input flow, this patch now limits previous patch to input flow, keeping
output as before that patch.

Reported-by: Debabrata Banerjee <>
Signed-off-by: Marcelo Ricardo Leitner <>
Signed-off-by: Jiri Pirko <>
Acked-by: Hannes Frederic Sowa <>
CC: Hannes Frederic Sowa <>
Signed-off-by: Ben Hutchings <>
8 years agocan: dev: fix nlmsg size calculation in can_get_size()
Marc Kleine-Budde [Sat, 5 Oct 2013 19:25:17 +0000 (21:25 +0200)]
can: dev: fix nlmsg size calculation in can_get_size()

[ Upstream commit fe119a05f8ca481623a8d02efcc984332e612528 ]

This patch fixes the calculation of the nlmsg size, by adding the missing

Signed-off-by: Marc Kleine-Budde <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoipv4: fix ineffective source address selection
Jiri Benc [Fri, 4 Oct 2013 15:04:48 +0000 (17:04 +0200)]
ipv4: fix ineffective source address selection

[ Upstream commit 0a7e22609067ff524fc7bbd45c6951dd08561667 ]

When sending out multicast messages, the source address in inet->mc_addr is
ignored and rewritten by an autoselected one. This is caused by a typo in
commit 813b3b5db831 ("ipv4: Use caller's on-stack flowi as-is in output
route lookups").

Signed-off-by: Jiri Benc <>
Acked-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoproc connector: fix info leaks
Mathias Krause [Mon, 30 Sep 2013 20:03:06 +0000 (22:03 +0200)]
proc connector: fix info leaks

[ Upstream commit e727ca82e0e9616ab4844301e6bae60ca7327682 ]

Initialize event_data for all possible message types to prevent leaking
kernel stack contents to userland (up to 20 bytes). Also set the flags
member of the connector message to 0 to prevent leaking two more stack
bytes this way.

Cc: # v2.6.15+
Signed-off-by: Mathias Krause <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agonet: heap overflow in __audit_sockaddr()
Dan Carpenter [Wed, 2 Oct 2013 21:27:20 +0000 (00:27 +0300)]
net: heap overflow in __audit_sockaddr()

[ Upstream commit 1661bf364ae9c506bc8795fef70d1532931be1e8 ]

We need to cap ->msg_namelen or it leads to a buffer overflow when we
to the memcpy() in __audit_sockaddr().  It requires CAP_AUDIT_CONTROL to
exploit this bug.

The call tree is:

Reported-by: Jüri Aedla <>
Signed-off-by: Dan Carpenter <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agonet: do not call sock_put() on TIMEWAIT sockets
Eric Dumazet [Wed, 2 Oct 2013 04:04:11 +0000 (21:04 -0700)]
net: do not call sock_put() on TIMEWAIT sockets

[ Upstream commit 80ad1d61e72d626e30ebe8529a0455e660ca4693 ]

commit 3ab5aee7fe84 ("net: Convert TCP & DCCP hash tables to use RCU /
hlist_nulls") incorrectly used sock_put() on TIMEWAIT sockets.

We should instead use inet_twsk_put()

Signed-off-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agotcp: do not forget FIN in tcp_shifted_skb()
Eric Dumazet [Fri, 4 Oct 2013 17:31:41 +0000 (10:31 -0700)]
tcp: do not forget FIN in tcp_shifted_skb()

[ Upstream commit 5e8a402f831dbe7ee831340a91439e46f0d38acd ]

Yuchung found following problem :

 There are bugs in the SACK processing code, merging part in
 tcp_shift_skb_data(), that incorrectly resets or ignores the sacked
 skbs FIN flag. When a receiver first SACK the FIN sequence, and later
 throw away ofo queue (e.g., sack-reneging), the sender will stop
 retransmitting the FIN flag, and hangs forever.

Following packetdrill test can be used to reproduce the bug.

$ cat sack-merge-bug.pkt
`sysctl -q net.ipv4.tcp_fack=0`

// Establish a connection and send 10 MSS.
0.000 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+.000 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+.000 bind(3, ..., ...) = 0
+.000 listen(3, 1) = 0

+.050 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
+.000 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 6>
+.001 < . 1:1(0) ack 1 win 1024
+.000 accept(3, ..., ...) = 4

+.100 write(4, ..., 12000) = 12000
+.000 shutdown(4, SHUT_WR) = 0
+.000 > . 1:10001(10000) ack 1
+.050 < . 1:1(0) ack 2001 win 257
+.000 > FP. 10001:12001(2000) ack 1
+.050 < . 1:1(0) ack 2001 win 257 <sack 10001:11001,nop,nop>
+.050 < . 1:1(0) ack 2001 win 257 <sack 10001:12002,nop,nop>
// SACK reneg
+.050 < . 1:1(0) ack 12001 win 257
+0 %{ print "unacked: ",tcpi_unacked }%
+5 %{ print "" }%

First, a typo inverted left/right of one OR operation, then
code forgot to advance end_seq if the merged skb carried FIN.

Bug was added in 2.6.29 by commit 832d11c5cd076ab
("tcp: Try to restore large SKBs while SACK processing")

Signed-off-by: Eric Dumazet <>
Signed-off-by: Yuchung Cheng <>
Acked-by: Neal Cardwell <>
Cc: Ilpo Järvinen <>
Acked-by: Ilpo Järvinen <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agotcp: must unclone packets before mangling them
Eric Dumazet [Tue, 15 Oct 2013 18:54:30 +0000 (11:54 -0700)]
tcp: must unclone packets before mangling them

[ Upstream commit c52e2421f7368fd36cbe330d2cf41b10452e39a9 ]

TCP stack should make sure it owns skbs before mangling them.

We had various crashes using bnx2x, and it turned out gso_size
was cleared right before bnx2x driver was populating TC descriptor
of the _previous_ packet send. TCP stack can sometime retransmit
packets that are still in Qdisc.

Of course we could make bnx2x driver more robust (using
ACCESS_ONCE(shinfo->gso_size) for example), but the bug is TCP stack.

We have identified two points where skb_unclone() was needed.

This patch adds a WARN_ON_ONCE() to warn us if we missed another
fix of this kind.

Kudos to Neal for finding the root cause of this bug. Its visible
using small MSS.

Signed-off-by: Eric Dumazet <>
Signed-off-by: Neal Cardwell <>
Cc: Yuchung Cheng <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoLinux 3.2.52 v3.2.52
Ben Hutchings [Sat, 26 Oct 2013 20:06:14 +0000 (21:06 +0100)]
Linux 3.2.52

8 years agocan: flexcan: flexcan_chip_start: fix regression, mark one MB for TX and abort pending TX
Marc Kleine-Budde [Fri, 4 Oct 2013 08:52:36 +0000 (10:52 +0200)]
can: flexcan: flexcan_chip_start: fix regression, mark one MB for TX and abort pending TX

commit d5a7b406c529e4595ce03dc8f6dcf7fa36f106fa upstream.

In patch

    0d1862e can: flexcan: fix flexcan_chip_start() on imx6

the loop in flexcan_chip_start() that iterates over all mailboxes after the
soft reset of the CAN core was removed. This loop put all mailboxes (even the
ones marked as reserved 1...7) into EMPTY/INACTIVE mode. On mailboxes 8...63,
this aborts any pending TX messages.

After a cold boot there is random garbage in the mailboxes, which leads to
spontaneous transmit of CAN frames during first activation. Further if the
interface was disabled with a pending message (usually due to an error
condition on the CAN bus), this message is retransmitted after enabling the
interface again.

This patch fixes the regression by:
1) Limiting the maximum number of used mailboxes to 8, 0...7 are used by the RX
FIFO, 8 is used by TX.
2) Marking the TX mailbox as EMPTY/INACTIVE, so that any pending TX of that
mailbox is aborted.

Cc: Lothar Waßmann <>
Signed-off-by: Marc Kleine-Budde <>
[bwh: Backported to 3.2:
 - Adjust context
 - Hardware local echo is still enabled]
Signed-off-by: Ben Hutchings <>
8 years agogianfar: Change default HW Tx queue scheduling mode
Claudiu Manoil [Sun, 23 Sep 2012 22:39:08 +0000 (22:39 +0000)]
gianfar: Change default HW Tx queue scheduling mode

commit b98b8babd6e3370fadb7c6eaacb00eb2f6344a6c upstream.

This is primarily to address transmission timeout occurrences, when
multiple H/W Tx queues are being used concurrently. Because in
the priority scheduling mode the controller does not service the
Tx queues equally (but in ascending index order), Tx timeouts are
being triggered rightaway for a basic test with multiple simultaneous
connections like:
iperf -c <server_ip> -n 100M -P 8

resulting in kernel trace:
NETDEV WATCHDOG: eth1 (fsl-gianfar): transmit queue <X> timed out
------------[ cut here ]------------
WARNING: at net/sched/sch_generic.c:255
and controller reset during intense traffic, and possibly further

This patch changes the default H/W Tx scheduling setting (TXSCHED)
for multi-queue devices, from priority scheduling mode to a weighted
round robin mode with equal weights for all H/W Tx queues, and
addresses the issue above.

Signed-off-by: Claudiu Manoil <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agomm, show_mem: suppress page counts in non-blockable contexts
David Rientjes [Mon, 29 Apr 2013 22:06:11 +0000 (15:06 -0700)]
mm, show_mem: suppress page counts in non-blockable contexts

commit 4b59e6c4730978679b414a8da61514a2518da512 upstream.

On large systems with a lot of memory, walking all RAM to determine page
types may take a half second or even more.

In non-blockable contexts, the page allocator will emit a page allocation
failure warning unless __GFP_NOWARN is specified.  In such contexts, irqs
are typically disabled and such a lengthy delay may even result in NMI
watchdog timeouts.

To fix this, suppress the page walk in such contexts when printing the
page allocation failure warning.

Signed-off-by: David Rientjes <>
Cc: Mel Gorman <>
Acked-by: Michal Hocko <>
Cc: Dave Hansen <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
8 years agoACPI / IPMI: Fix atomic context requirement of ipmi_msg_handler()
Lv Zheng [Fri, 13 Sep 2013 05:13:23 +0000 (13:13 +0800)]
ACPI / IPMI: Fix atomic context requirement of ipmi_msg_handler()

commit 06a8566bcf5cf7db9843a82cde7a33c7bf3947d9 upstream.

This patch fixes the issues indicated by the test results that
ipmi_msg_handler() is invoked in atomic context.

BUG: scheduling while atomic: kipmi0/18933/0x10000100
Modules linked in: ipmi_si acpi_ipmi ...
CPU: 3 PID: 18933 Comm: kipmi0 Tainted: G       AW    3.10.0-rc7+ #2
Hardware name: QCI QSSC-S4R/QSSC-S4R, BIOS QSSC-S4R.QCI.01.00.0027.070120100606 07/01/2010
 ffff8838245eea00 ffff88103fc63c98 ffffffff814c4a1e ffff88103fc63ca8
 ffffffff814bfbab ffff88103fc63d28 ffffffff814c73e0 ffff88103933cbd4
 0000000000000096 ffff88103fc63ce8 ffff88102f618000 ffff881035c01fd8
Call Trace:
 <IRQ>  [<ffffffff814c4a1e>] dump_stack+0x19/0x1b
 [<ffffffff814bfbab>] __schedule_bug+0x46/0x54
 [<ffffffff814c73e0>] __schedule+0x83/0x59c
 [<ffffffff81058853>] __cond_resched+0x22/0x2d
 [<ffffffff814c794b>] _cond_resched+0x14/0x1d
 [<ffffffff814c6d82>] mutex_lock+0x11/0x32
 [<ffffffff8101e1e9>] ? __default_send_IPI_dest_field.constprop.0+0x53/0x58
 [<ffffffffa09e3f9c>] ipmi_msg_handler+0x23/0x166 [ipmi_si]
 [<ffffffff812bf6e4>] deliver_response+0x55/0x5a
 [<ffffffff812c0fd4>] handle_new_recv_msgs+0xb67/0xc65
 [<ffffffff81007ad1>] ? read_tsc+0x9/0x19
 [<ffffffff814c8620>] ? _raw_spin_lock_irq+0xa/0xc
 [<ffffffffa09e1128>] ipmi_thread+0x5c/0x146 [ipmi_si]

Also Tony Camuso says:

 We were getting occasional "Scheduling while atomic" call traces
 during boot on some systems. Problem was first seen on a Cisco C210
 but we were able to reproduce it on a Cisco c220m3. Setting
 CONFIG_LOCKDEP and LOCKDEP_SUPPORT to 'y' exposed a lockdep around
 tx_msg_lock in acpi_ipmi.c struct acpi_ipmi_device.

 [ INFO: inconsistent lock state ]
 2.6.32-415.el6.x86_64-debug-splck #1
 inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
 ksoftirqd/3/17 [HC0[0]:SC1[1]:HE1:SE0] takes:
  (&ipmi_device->tx_msg_lock){+.?...}, at: [<ffffffff81337a27>] ipmi_msg_handler+0x71/0x126
 {SOFTIRQ-ON-W} state was registered at:
   [<ffffffff810ba11c>] __lock_acquire+0x63c/0x1570
   [<ffffffff810bb0f4>] lock_acquire+0xa4/0x120
   [<ffffffff815581cc>] __mutex_lock_common+0x4c/0x400
   [<ffffffff815586ea>] mutex_lock_nested+0x4a/0x60
   [<ffffffff8133789d>] acpi_ipmi_space_handler+0x11b/0x234
   [<ffffffff81321c62>] acpi_ev_address_space_dispatch+0x170/0x1be

The fix implemented by this change has been tested by Tony:

 Tested the patch in a boot loop with lockdep debug enabled and never
 saw the problem in over 400 reboots.

Reported-and-tested-by: Tony Camuso <>
Signed-off-by: Lv Zheng <>
Reviewed-by: Huang Ying <>
Signed-off-by: Rafael J. Wysocki <>
Signed-off-by: Ben Hutchings <>
8 years agostaging: comedi: ni_65xx: (bug fix) confine insn_bits to one subdevice
Ian Abbott [Wed, 2 Oct 2013 13:57:51 +0000 (14:57 +0100)]
staging: comedi: ni_65xx: (bug fix) confine insn_bits to one subdevice

commit 677a31565692d596ef42ea589b53ba289abf4713 upstream.

The `insn_bits` handler `ni_65xx_dio_insn_bits()` has a `for` loop that
currently writes (optionally) and reads back up to 5 "ports" consisting
of 8 channels each.  It reads up to 32 1-bit channels but can only read
and write a whole port at once - it needs to handle up to 5 ports as the
first channel it reads might not be aligned on a port boundary.  It
breaks out of the loop early if the next port it handles is beyond the
final port on the card.  It also breaks out early on the 5th port in the
loop if the first channel was aligned.  Unfortunately, it doesn't check
that the current port it is dealing with belongs to the comedi subdevice
the `insn_bits` handler is acting on.  That's a bug.

Redo the `for` loop to terminate after the final port belonging to the
subdevice, changing the loop variable in the process to simplify things
a bit.  The `for` loop could now try and handle more than 5 ports if the
subdevice has more than 40 channels, but the test `if (bitshift >= 32)`
ensures it will break out early after 4 or 5 ports (depending on whether
the first channel is aligned on a port boundary).  (`bitshift` will be
between -7 and 7 inclusive on the first iteration, increasing by 8 for
each subsequent operation.)

Signed-off-by: Ian Abbott <>
Signed-off-by: Greg Kroah-Hartman <>
[Ian Abbott: This patch applies to kernels 2.6.34.y through to 3.5.y
Signed-off-by: Ben Hutchings <>
8 years agoext4: avoid hang when mounting non-journal filesystems with orphan list
Theodore Ts'o [Thu, 27 Dec 2012 06:42:50 +0000 (01:42 -0500)]
ext4: avoid hang when mounting non-journal filesystems with orphan list

commit 0e9a9a1ad619e7e987815d20262d36a2f95717ca upstream.

When trying to mount a file system which does not contain a journal,
but which does have a orphan list containing an inode which needs to
be truncated, the mount call with hang forever in
ext4_orphan_cleanup() because ext4_orphan_del() will return
immediately without removing the inode from the orphan list, leading
to an uninterruptible loop in kernel code which will busy out one of
the CPU's on the system.

This can be trivially reproduced by trying to mount the file system
found in tests/f_orphan_extents_inode/image.gz from the e2fsprogs
source tree.  If a malicious user were to put this on a USB stick, and
mount it on a Linux desktop which has automatic mounts enabled, this
could be considered a potential denial of service attack.  (Not a big
deal in practice, but professional paranoids worry about such things,
and have even been known to allocate CVE numbers for such problems.)

-js: This is a fix for CVE-2013-2015.

Signed-off-by: "Theodore Ts'o" <>
Reviewed-by: Zheng Liu <>
Acked-by: Jan Kara <>
Signed-off-by: Jiri Slaby <>
Signed-off-by: Ben Hutchings <>
8 years agohwmon: (applesmc) Silence uninitialized warnings
Henrik Rydberg [Thu, 26 Jan 2012 11:08:41 +0000 (06:08 -0500)]
hwmon: (applesmc) Silence uninitialized warnings

commit 0fc86eca1b338d06ec500b34ef7def79c32b602b upstream.

Some error paths do not set a result, leading to the (false)
assumption that the value may be used uninitialized. Set results for
those paths as well.

Signed-off-by: Henrik Rydberg <>
Signed-off-by: Guenter Roeck <>
Signed-off-by: Ben Hutchings <>
8 years agoxhci: Fix race between ep halt and URB cancellation
Florian Wolter [Wed, 14 Aug 2013 08:33:16 +0000 (10:33 +0200)]
xhci: Fix race between ep halt and URB cancellation

commit 526867c3ca0caa2e3e846cb993b0f961c33c2abb upstream.

The halted state of a endpoint cannot be cleared over CLEAR_HALT from a
user process, because the stopped_td variable was overwritten in the
handle_stopped_endpoint() function. So the xhci_endpoint_reset() function will
refuse the reset and communication with device can not run over this endpoint.

Signed-off-by: Florian Wolter <>
Signed-off-by: Sarah Sharp <>
Signed-off-by: Ben Hutchings <>
8 years agocciss: fix info leak in cciss_ioctl32_passthru()
Dan Carpenter [Tue, 24 Sep 2013 22:27:45 +0000 (15:27 -0700)]
cciss: fix info leak in cciss_ioctl32_passthru()

commit 58f09e00ae095e46ef9edfcf3a5fd9ccdfad065e upstream.

The arg64 struct has a hole after ->buf_size which isn't cleared.  Or if
any of the calls to copy_from_user() fail then that would cause an
information leak as well.

This was assigned CVE-2013-2147.

Signed-off-by: Dan Carpenter <>
Acked-by: Mike Miller <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
8 years agocpqarray: fix info leak in ida_locked_ioctl()
Dan Carpenter [Tue, 24 Sep 2013 22:27:44 +0000 (15:27 -0700)]
cpqarray: fix info leak in ida_locked_ioctl()

commit 627aad1c01da6f881e7f98d71fd928ca0c316b1a upstream.

The pciinfo struct has a two byte hole after ->dev_fn so stack
information could be leaked to the user.

This was assigned CVE-2013-2147.

Signed-off-by: Dan Carpenter <>
Acked-by: Mike Miller <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
8 years agoiscsi: don't hang in endless loop if no targets present
Sasha Levin [Thu, 26 Jan 2012 03:16:16 +0000 (22:16 -0500)]
iscsi: don't hang in endless loop if no targets present

commit 46a7c17d26967922092f3a8291815ffb20f6cabe upstream.

iscsi_if_send_reply() may return -ESRCH if there were no targets to send
data to. Currently we're ignoring this value and looping in attempt to do it
over and over, which will usually lead in a hung task like this one:

[ 4920.817298] INFO: task trinity:9074 blocked for more than 120 seconds.
[ 4920.818527] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 4920.819982] trinity         D 0000000000000000  5504  9074   2756 0x00000004
[ 4920.825374]  ffff880003961a98 0000000000000086 ffff8800001aa000 ffff8800001aa000
[ 4920.826791]  00000000001d4340 ffff880003961fd8 ffff880003960000 00000000001d4340
[ 4920.828241]  00000000001d4340 00000000001d4340 ffff880003961fd8 00000000001d4340
[ 4920.833231]
[ 4920.833519] Call Trace:
[ 4920.834010]  [<ffffffff826363fa>] schedule+0x3a/0x50
[ 4920.834953]  [<ffffffff82634ac9>] __mutex_lock_common+0x209/0x5b0
[ 4920.836226]  [<ffffffff81af805d>] ? iscsi_if_rx+0x2d/0x990
[ 4920.837281]  [<ffffffff81053943>] ? sched_clock+0x13/0x20
[ 4920.838305]  [<ffffffff81af805d>] ? iscsi_if_rx+0x2d/0x990
[ 4920.839336]  [<ffffffff82634eb0>] mutex_lock_nested+0x40/0x50
[ 4920.840423]  [<ffffffff81af805d>] iscsi_if_rx+0x2d/0x990
[ 4920.841434]  [<ffffffff810dffed>] ? sub_preempt_count+0x9d/0xd0
[ 4920.842548]  [<ffffffff82637bb0>] ? _raw_read_unlock+0x30/0x60
[ 4920.843666]  [<ffffffff821f71de>] netlink_unicast+0x1ae/0x1f0
[ 4920.844751]  [<ffffffff821f7997>] netlink_sendmsg+0x227/0x350
[ 4920.845850]  [<ffffffff821857bd>] ? sock_update_netprioidx+0xdd/0x1b0
[ 4920.847060]  [<ffffffff82185732>] ? sock_update_netprioidx+0x52/0x1b0
[ 4920.848276]  [<ffffffff8217f226>] sock_aio_write+0x166/0x180
[ 4920.849348]  [<ffffffff810dfe41>] ? get_parent_ip+0x11/0x50
[ 4920.850428]  [<ffffffff811d0d9a>] do_sync_write+0xda/0x120
[ 4920.851465]  [<ffffffff810dffed>] ? sub_preempt_count+0x9d/0xd0
[ 4920.852579]  [<ffffffff810dfe41>] ? get_parent_ip+0x11/0x50
[ 4920.853608]  [<ffffffff81791887>] ? security_file_permission+0x27/0xb0
[ 4920.854821]  [<ffffffff811d0f4c>] vfs_write+0x16c/0x180
[ 4920.855781]  [<ffffffff811d104f>] sys_write+0x4f/0xa0
[ 4920.856798]  [<ffffffff82638e79>] system_call_fastpath+0x16/0x1b
[ 4920.877487] 1 lock held by trinity/9074:
[ 4920.878239]  #0:  (rx_queue_mutex){+.+...}, at: [<ffffffff81af805d>] iscsi_if_rx+0x2d/0x990
[ 4920.880005] Kernel panic - not syncing: hung_task: blocked tasks

Signed-off-by: Sasha Levin <>
Acked-by: Mike Christie <>
Signed-off-by: James Bottomley <>
Signed-off-by: Ben Hutchings <>
8 years agoRevert "sctp: fix call to SCTP_CMD_PROCESS_SACK in sctp_cmd_interpreter()"
Ben Hutchings [Sun, 20 Oct 2013 16:06:39 +0000 (17:06 +0100)]
Revert "sctp: fix call to SCTP_CMD_PROCESS_SACK in sctp_cmd_interpreter()"

This reverts commit de77b7955c3985ca95f64af3cb10557eb17eacee, which was
commit f6e80abeab928b7c47cc1fbf53df13b4398a2bec upstream.

This fix was only appropriate for Linux 3.7 onward, and introduced a
regression when applied to earlier versions.

Signed-off-by: Ben Hutchings <>
8 years agoisofs: Refuse RW mount of the filesystem instead of making it RO
Jan Kara [Thu, 25 Jul 2013 09:49:11 +0000 (11:49 +0200)]
isofs: Refuse RW mount of the filesystem instead of making it RO

commit 17b7f7cf58926844e1dd40f5eb5348d481deca6a upstream.

Refuse RW mount of isofs filesystem. So far we just silently changed it
to RO mount but when the media is writeable, block layer won't notice
this change and thus will think device is used RW and will block eject
button of the drive. That is unexpected by users because for
non-writeable media eject button works just fine.

Userspace mount(8) command handles this just fine and retries mounting
with MS_RDONLY set so userspace shouldn't see any regression.  Plus any
tool mounting isofs is likely confronted with the case of read-only
media where block layer already refuses to mount the filesystem without
MS_RDONLY set so our behavior shouldn't be anything new for it.

Reported-by: Hui Wang <>
Signed-off-by: Jan Kara <>
Signed-off-by: Ben Hutchings <>
8 years agoHID: usbhid: quirk for N-Trig DuoSense Touch Screen
Vasily Titskiy [Fri, 30 Aug 2013 22:25:04 +0000 (18:25 -0400)]
HID: usbhid: quirk for N-Trig DuoSense Touch Screen

commit 9e0bf92c223dabe0789714f8f85f6e26f8f9cda4 upstream.

The DuoSense touchscreen device causes a 10 second timeout. This fix
removes the delay.

Signed-off-by: Vasily Titskiy <>
Signed-off-by: Jiri Kosina <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agoHID: Fix Speedlink VAD Cezanne support for some devices
Stefan Kriwanek [Sun, 25 Aug 2013 08:46:13 +0000 (10:46 +0200)]
HID: Fix Speedlink VAD Cezanne support for some devices

commit 06bb5219118fb098f4b0c7dcb484b28a52bf1c14 upstream.

Some devices of the "Speedlink VAD Cezanne" model need more aggressive fixing
than already done.

I made sure through testing that this patch would not interfere with the proper
working of a device that is bug-free. (The driver drops EV_REL events with
abs(val) >= 256, which are not achievable even on the highest laser resolution
hardware setting.)

Signed-off-by: Stefan Kriwanek <>
Signed-off-by: Jiri Kosina <>
Signed-off-by: Ben Hutchings <>
8 years agofanotify: dont merge permission events
Lino Sanfilippo [Fri, 23 Mar 2012 01:42:23 +0000 (02:42 +0100)]
fanotify: dont merge permission events

commit 03a1cec1f17ac1a6041996b3e40f96b5a2f90e1b upstream.

Boyd Yang reported a problem for the case that multiple threads of the same
thread group are waiting for a reponse for a permission event.
In this case it is possible that some of the threads are never woken up, even
if the response for the event has been received

The reason is that we are currently merging permission events if they belong to
the same thread group. But we are not prepared to wake up more than one waiter
for each event. We do

wait_event(group->fanotify_data.access_waitq, event->response ||
and after that
  event->response = 0;

which is the reason that even if we woke up all waiters for the same event
some of them may see event->response being already set 0 again, then go back to
sleep and block forever.

With this patch we avoid that more than one thread is waiting for a response
by not merging permission events for the same thread group any more.

Reported-by: Boyd Yang <>
Signed-off-by: Lino Sanfilippo <>
Signed-off-by: Eric Paris <>
Signed-off-by: Ben Hutchings <>
8 years agoperf tools: Handle JITed code in shared memory
Andi Kleen [Thu, 25 Apr 2013 00:03:02 +0000 (17:03 -0700)]
perf tools: Handle JITed code in shared memory

commit 89365e6c9ad4c0e090e4c6a4b67a3ce319381d89 upstream.

Need to check for /dev/zero.

Most likely more strings are missing too.

Signed-off-by: Andi Kleen <>
Signed-off-by: Arnaldo Carvalho de Melo <>
Signed-off-by: Ben Hutchings <>
8 years agoperf: Fix perf_cgroup_switch for sw-events
Peter Zijlstra [Tue, 2 Oct 2012 13:41:23 +0000 (15:41 +0200)]
perf: Fix perf_cgroup_switch for sw-events

commit 95cf59ea72331d0093010543b8951bb43f262cac upstream.

Jiri reported that he could trigger the WARN_ON_ONCE() in
perf_cgroup_switch() using sw-events. This is because sw-events share
a cpuctx with multiple PMUs.

Use the ->unique_pmu pointer to limit the pmu iteration to unique
cpuctx instances.

Reported-and-Tested-by: Jiri Olsa <>
Signed-off-by: Peter Zijlstra <>
Signed-off-by: Ingo Molnar <>
Signed-off-by: Ben Hutchings <>
8 years agoperf: Clarify perf_cpu_context::active_pmu usage by renaming it to ::unique_pmu
Peter Zijlstra [Tue, 2 Oct 2012 13:38:52 +0000 (15:38 +0200)]
perf: Clarify perf_cpu_context::active_pmu usage by renaming it to ::unique_pmu

commit 3f1f33206c16c7b3839d71372bc2ac3f305aa802 upstream.

Stephane thought the perf_cpu_context::active_pmu name confusing and
suggested using 'unique_pmu' instead.

This pointer is a pointer to a 'random' pmu sharing the cpuctx
instance, therefore limiting a for_each_pmu loop to those where
cpuctx->unique_pmu matches the pmu we get a loop over unique cpuctx

Suggested-by: Stephane Eranian <>
Signed-off-by: Peter Zijlstra <>
Signed-off-by: Ingo Molnar <>
Signed-off-by: Ben Hutchings <>
8 years agocgroup: fail if monitored file and event_control are in different cgroup
Li Zefan [Mon, 18 Feb 2013 06:13:35 +0000 (14:13 +0800)]
cgroup: fail if monitored file and event_control are in different cgroup

commit f169007b2773f285e098cb84c74aac0154d65ff7 upstream.

If we pass fd of memory.usage_in_bytes of cgroup A to cgroup.event_control
of cgroup B, then we won't get memory usage notification from A but B!

What's worse, if A and B are in different mount hierarchy, we'll end up
accessing NULL pointer!

Disallow this kind of invalid usage.

Signed-off-by: Li Zefan <>
Acked-by: Kirill A. Shutemov <>
Signed-off-by: Tejun Heo <>
Signed-off-by: Ben Hutchings <>
8 years agosfc: Fix efx_rx_buf_offset() for recycled pages
Ben Hutchings [Fri, 6 Sep 2013 21:39:20 +0000 (22:39 +0100)]
sfc: Fix efx_rx_buf_offset() for recycled pages

This bug fix is only for stable branches older than 3.10.  The bug was
fixed upstream by commit 2768935a4660 ('sfc: reuse pages to avoid DMA
mapping/unmapping costs'), but that change is totally unsuitable for

Commit b590ace09d51 ('sfc: Fix efx_rx_buf_offset() in the presence of
swiotlb') added an explicit page_offset member to struct
efx_rx_buffer, which must be set consistently with the and
dma_addr fields.  However, it failed to add the necessary assignment
in efx_resurrect_rx_buffer().  It also did not correct the calculation
of efx_rx_buffer::dma_addr in efx_resurrect_rx_buffer(), which assumes
that DMA-mapping a page will result in a page-aligned DMA address
(exactly what swiotlb violates).

Add the assignment of efx_rx_buffer::page_offset and change the
calculation of dma_addr to make use of it.

Signed-off-by: Ben Hutchings <>
8 years agomacvtap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS
Jason Wang [Thu, 18 Jul 2013 02:55:16 +0000 (10:55 +0800)]
macvtap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS

commit ece793fcfc417b3925844be88a6a6dc82ae8f7c6 upstream.

We try to linearize part of the skb when the number of iov is greater than
MAX_SKB_FRAGS. This is not enough since each single vector may occupy more than
one pages, so zerocopy_sg_fromiovec() may still fail and may break the guest

Solve this problem by calculate the pages needed for iov before trying to do
zerocopy and switch to use copy instead of zerocopy if it needs more than

This is done through introducing a new helper to count the pages for iov, and
call uarg->callback() manually when switching from zerocopy to copy to notify

We can do further optimization on top.

This bug were introduced from b92946e2919134ebe2a4083e4302236295ea2a73
(macvtap: zerocopy: validate vectors before building skb).

Cc: Michael S. Tsirkin <>
Signed-off-by: Jason Wang <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2: callback only takes one argument]
Signed-off-by: Ben Hutchings <>
8 years agoRevert "zram: use zram->lock to protect zram_free_page() in swap free notify path"
Ben Hutchings [Sun, 20 Oct 2013 12:37:39 +0000 (13:37 +0100)]
Revert "zram: use zram->lock to protect zram_free_page() in swap free notify path"

This reverts commit 9e443904906ca2b5b3ae71f34ac4a4fa6905623e, which
was commit 57ab048532c0d975538cebd4456491b5c34248f4 upstream.

Taking the semaphore here leads to sleeping in atomic context.  This
was fixed in mainline commit a0c516cbfc74 ("zram: don't grab mutex in
zram_slot_free_noity") but that is too difficult to backport.

Signed-off-by: Ben Hutchings <>
8 years agopowerpc/pseries/lparcfg: Fix possible overflow are more than 1026
Chen Gang [Mon, 22 Apr 2013 17:12:54 +0000 (17:12 +0000)]
powerpc/pseries/lparcfg: Fix possible overflow are more than 1026

commit 5676005acf26ab7e924a8438ea4746e47d405762 upstream.

need set '\0' for 'local_buffer'.

SPLPAR_MAXLENGTH is 1026, RTAS_DATA_BUF_SIZE is 4096. so the contents of
rtas_data_buf may truncated in memcpy.

if contents are really truncated.
  the splpar_strlen is more than 1026. the next while loop checking will
  not find the end of buffer. that will cause memory access violation.

Signed-off-by: Chen Gang <>
Signed-off-by: Benjamin Herrenschmidt <>
Signed-off-by: Ben Hutchings <>
8 years agom68knommu: clean up linker script
Greg Ungerer [Thu, 5 Jan 2012 05:51:13 +0000 (15:51 +1000)]
m68knommu: clean up linker script

commit f84f52a5c15db7d14a534815f27253b001735183 upstream.

There is a lot of years of collected cruft in the m68knommu linker script.
Clean it all up and use the well defined linker script support macros.

Support is maintained for building both ROM/FLASH based and RAM based setups.
No major changes to section layouts, though the rodata section is now lumped
in with the read/write data section.

Signed-off-by: Greg Ungerer <>
Signed-off-by: Ben Hutchings <>
8 years agom68k: use non-MMU linker script for ColdFire MMU builds
Greg Ungerer [Wed, 19 Oct 2011 03:50:35 +0000 (13:50 +1000)]
m68k: use non-MMU linker script for ColdFire MMU builds

commit ed865e31a8273be200db9ddcdb6b844e48777abd upstream.

Use the non-MMU linker script for ColdFire builds when we are building
for MMU enabled. The image layout is correct for loading on existing
ColdFire dev boards. The only addition required to the current non-MMU
linker script is to add support for the fixup section.

Signed-off-by: Greg Ungerer <>
Acked-by: Geert Uytterhoeven <>
Acked-by: Matt Waddel <>
Acked-by: Kurt Mahan <>
Signed-off-by: Ben Hutchings <>
8 years agom68k: consolidate the linker scripts
Greg Ungerer [Thu, 8 Dec 2011 05:39:05 +0000 (15:39 +1000)]
m68k: consolidate the linker scripts

commit 40c1b9cfeedf79b909c961e0e00a13497e80bc82 upstream.

The merge of m68knommu left the linker scripts a little disorganized.
Some consistent naming and squashing two of scripts that just include
others can simplify things a lot.

So merge the two simple including scripts, and rename the nommu script
to be consistent with the existing m68k linker scripts.

Signed-off-by: Greg Ungerer <>
Signed-off-by: Ben Hutchings <>
8 years agousb: core: don't try to reset_device() a port that got just disconnected
Julius Werner [Wed, 31 Jul 2013 02:51:20 +0000 (19:51 -0700)]
usb: core: don't try to reset_device() a port that got just disconnected

commit 481f2d4f89f87a0baa26147f323380e31cfa7c44 upstream.

The USB hub driver's event handler contains a check to catch SuperSpeed
devices that transitioned into the SS.Inactive state and tries to fix
them with a reset. It decides whether to do a plain hub port reset or
call the usb_reset_device() function based on whether there was a device
attached to the port.

However, there are device/hub combinations (found with a JetFlash
Transcend mass storage stick (8564:1000) on the root hub of an Intel
LynxPoint PCH) which can transition to the SS.Inactive state on
disconnect (and stay there long enough for the host to notice). In this
case, above-mentioned reset check will call usb_reset_device() on the
stale device data structure. The kernel will send pointless LPM control
messages to the no longer connected device address and can even cause
several 5 second khubd stalls on some (buggy?) host controllers, before
finally accepting the device's fate amongst a flurry of error messages.

This patch makes the choice of reset dependent on the port status that
has just been read from the hub in addition to the existence of an
in-kernel data structure for the device, and only proceeds with the more
extensive reset if both are valid.

Signed-off-by: Julius Werner <>
Signed-off-by: Sarah Sharp <>
Signed-off-by: Ben Hutchings <>
8 years agodebugfs: debugfs_remove_recursive() must not rely on list_empty(d_subdirs)
Oleg Nesterov [Fri, 26 Jul 2013 15:12:56 +0000 (17:12 +0200)]
debugfs: debugfs_remove_recursive() must not rely on list_empty(d_subdirs)

commit 776164c1faac4966ab14418bb0922e1820da1d19 upstream.

debugfs_remove_recursive() is wrong,

1. it wrongly assumes that !list_empty(d_subdirs) means that this
   dir should be removed.

   This is not that bad by itself, but:

2. if d_subdirs does not becomes empty after __debugfs_remove()
   it gives up and silently fails, it doesn't even try to remove
   other entries.

   However ->d_subdirs can be non-empty because it still has the
   already deleted !debugfs_positive() entries.

3. simple_release_fs() is called even if __debugfs_remove() fails.

Suppose we have


and someone opens dir1/dir2/file2.

Now, debugfs_remove_recursive(dir1/dir2) succeeds, and dir1/dir2 goes

But debugfs_remove_recursive(dir1) silently fails and doesn't remove
this directory. Because it tries to delete (the already deleted)
dir1/dir2/file2 again and then fails due to "Avoid infinite loop"



cd /sys/kernel/debug/tracing
echo 'p:probe/sigprocmask sigprocmask' >> kprobe_events
sleep 1000 < events/probe/sigprocmask/id &
echo -n >| kprobe_events

[ -d events/probe ] && echo "ERR!! failed to rm probe"

And after that it is not possible to create another probe entry.

With this patch debugfs_remove_recursive() skips !debugfs_positive()
files although this is not strictly needed. The most important change
is that it does not try to make ->d_subdirs empty, it simply scans
the whole list(s) recursively and removes as much as possible.

Acked-by: Greg Kroah-Hartman <>
Signed-off-by: Oleg Nesterov <>
Signed-off-by: Steven Rostedt <>
Signed-off-by: Ben Hutchings <>
8 years agoperf: Use css_tryget() to avoid propping up css refcount
Salman Qazi [Thu, 14 Jun 2012 22:31:09 +0000 (15:31 -0700)]
perf: Use css_tryget() to avoid propping up css refcount

commit 9c5da09d266ca9b32eb16cf940f8161d949c2fe5 upstream.

An rmdir pushes css's ref count to zero.  However, if the associated
directory is open at the time, the dentry ref count is non-zero.  If
the fd for this directory is then passed into perf_event_open, it
does a css_get().  This bounces the ref count back up from zero.  This
is a problem by itself.  But what makes it turn into a crash is the
fact that we end up doing an extra dput, since we perform a dput
when css_put sees the ref count go down to zero.

css_tryget() does not fall into that trap. So, we use that instead.

Reproduction test-case for the bug:

 #include <unistd.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <fcntl.h>
 #include <linux/unistd.h>
 #include <linux/perf_event.h>
 #include <string.h>
 #include <errno.h>
 #include <stdio.h>

 #define PERF_FLAG_PID_CGROUP    (1U << 2)

 int perf_event_open(struct perf_event_attr *hw_event_uptr,
                     pid_t pid, int cpu, int group_fd, unsigned long flags) {
         return syscall(__NR_perf_event_open,hw_event_uptr, pid, cpu,
                 group_fd, flags);

  * Directly poke at the perf_event bug, since it's proving hard to repro
  * depending on where in the kernel tree.  what moved?
 int main(int argc, char **argv)
        int fd;
        struct perf_event_attr attr;
        memset(&attr, 0, sizeof(attr));
        attr.exclude_kernel = 1;
        attr.size = sizeof(attr);
        mkdir("/dev/cgroup/perf_event/blah", 0777);
        fd = open("/dev/cgroup/perf_event/blah", O_RDONLY);
        perf_event_open(&attr, fd, 0, -1,  PERF_FLAG_PID_CGROUP);
        return 0;

Signed-off-by: Salman Qazi <>
Signed-off-by: Peter Zijlstra <>
Acked-by: Tejun Heo <>
Signed-off-by: Ingo Molnar <>
Signed-off-by: Ben Hutchings <>
8 years agokernel-doc: bugfix - multi-line macros
Daniel Santos [Fri, 5 Oct 2012 00:15:05 +0000 (17:15 -0700)]
kernel-doc: bugfix - multi-line macros

commit 654784284430bf2739985914b65e09c7c35a7273 upstream.

Prior to this patch the following code breaks:

 * multiline_example - this breaks kernel-doc
 #define multiline_example( \

Producing this error:

Error(somefile.h:983): cannot understand prototype: 'multiline_example( \ '

This patch fixes the issue by appending all lines ending in a blackslash
(optionally followed by whitespace), removing the backslash and any
whitespace after it prior to appending (just like the C pre-processor

This fixes a break in kerel-doc introduced by the additions to rbtree.h.

Signed-off-by: Daniel Santos <>
Cc: Randy Dunlap <>
Cc: Michal Marek <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
8 years agosparc32: Fix exit flag passed from traced sys_sigreturn
Kirill Tkhai [Thu, 25 Jul 2013 21:17:15 +0000 (01:17 +0400)]
sparc32: Fix exit flag passed from traced sys_sigreturn

[ Upstream commit 7a3b0f89e3fea680f93932691ca41a68eee7ab5e ]

Pass 1 in %o1 to indicate that syscall_trace accounts exit.

Signed-off-by: Kirill Tkhai <>
CC: David Miller <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agosparc64: Fix not SRA'ed %o5 in 32-bit traced syscall
Kirill Tkhai [Fri, 26 Jul 2013 13:21:12 +0000 (17:21 +0400)]
sparc64: Fix not SRA'ed %o5 in 32-bit traced syscall

[ Upstream commit ab2abda6377723e0d5fbbfe5f5aa16a5523344d1 ]

(From v1 to v2: changed comment)

On the way linux_sparc_syscall32->linux_syscall_trace32->goto 2f,
register %o5 doesn't clear its second 32-bit.

Fix that.

Signed-off-by: Kirill Tkhai <>
CC: David Miller <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agosparc64: Fix off by one in trampoline TLB mapping installation loop.
David S. Miller [Thu, 22 Aug 2013 23:38:46 +0000 (16:38 -0700)]
sparc64: Fix off by one in trampoline TLB mapping installation loop.

[ Upstream commit 63d499662aeec1864ec36d042aca8184ea6a938e ]

Reported-by: Kirill Tkhai <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agosparc64: Remove RWSEM export leftovers
Kirill Tkhai [Mon, 12 Aug 2013 12:02:24 +0000 (16:02 +0400)]
sparc64: Remove RWSEM export leftovers

[ Upstream commit 61d9b9355b0d427bd1e732bd54628ff9103e496f ]

The functions


are implemented inline, so remove corresponding EXPORT_SYMBOLs
(They lead to compile errors on RT kernel).

Signed-off-by: Kirill Tkhai <>
CC: David Miller <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agosparc64: Fix ITLB handler of null page
Kirill Tkhai [Fri, 2 Aug 2013 15:23:18 +0000 (19:23 +0400)]
sparc64: Fix ITLB handler of null page

[ Upstream commit 1c2696cdaad84580545a2e9c0879ff597880b1a9 ]

1)Use kvmap_itlb_longpath instead of kvmap_dtlb_longpath.

2)Handle page #0 only, don't handle page #1: bleu -> blu

 (KERNBASE is 0x400000, so #1 does not exist too. But everything
  is possible in the future. Fix to not to have problems later.)

3)Remove unused kvmap_itlb_nonlinear.

Signed-off-by: Kirill Tkhai <>
CC: David Miller <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoesp_scsi: Fix tag state corruption when autosensing.
David S. Miller [Fri, 2 Aug 2013 01:08:34 +0000 (18:08 -0700)]
esp_scsi: Fix tag state corruption when autosensing.

[ Upstream commit 21af8107f27878813d0364733c0b08813c2c192a ]

Meelis Roos reports a crash in esp_free_lun_tag() in the presense
of a disk which has died.

The issue is that when we issue an autosense command, we do so by
hijacking the original command that caused the check-condition.

When we do so we clear out the ent->tag[] array when we issue it via
find_and_prep_issuable_command().  This is so that the autosense
command is forced to be issued non-tagged.

That is problematic, because it is the value of ent->tag[] which
determines whether we issued the original scsi command as tagged
vs. non-tagged (see esp_alloc_lun_tag()).

And that, in turn, is what trips up the sanity checks in
esp_free_lun_tag().  That function needs the original ->tag[] values
in order to free up the tag slot properly.

Fix this by remembering the original command's tag values, and
having esp_alloc_lun_tag() and esp_free_lun_tag() use them.

Reported-by: Meelis Roos <>
Tested-by: Meelis Roos <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoll_temac: Reset dma descriptors indexes on ndo_open
Ricardo Ribalda [Tue, 1 Oct 2013 06:17:10 +0000 (08:17 +0200)]
ll_temac: Reset dma descriptors indexes on ndo_open

[ Upstream commit 7167cf0e8cd10287b7912b9ffcccd9616f382922 ]

The dma descriptors indexes are only initialized on the probe function.

If a packet is on the buffer when temac_stop is called, the dma
descriptors indexes can be left on a incorrect state where no other
package can be sent.

So an interface could be left in an usable state after ifdow/ifup.

This patch makes sure that the descriptors indexes are in a proper
status when the device is open.

Signed-off-by: Ricardo Ribalda Delgado <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoipv6 mcast: use in6_dev_put in timer handlers instead of __in6_dev_put
Salam Noureddine [Sun, 29 Sep 2013 20:41:34 +0000 (13:41 -0700)]
ipv6 mcast: use in6_dev_put in timer handlers instead of __in6_dev_put

[ Upstream commit 9260d3e1013701aa814d10c8fc6a9f92bd17d643 ]

It is possible for the timer handlers to run after the call to
ipv6_mc_down so use in6_dev_put instead of __in6_dev_put in the
handler function in order to do proper cleanup when the refcnt
reaches 0. Otherwise, the refcnt can reach zero without the
inet6_dev being destroyed and we end up leaking a reference to
the net_device and see messages like the following,

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

Tested on linux-3.4.43.

Signed-off-by: Salam Noureddine <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoipv4 igmp: use in_dev_put in timer handlers instead of __in_dev_put
Salam Noureddine [Sun, 29 Sep 2013 20:39:42 +0000 (13:39 -0700)]
ipv4 igmp: use in_dev_put in timer handlers instead of __in_dev_put

[ Upstream commit e2401654dd0f5f3fb7a8d80dad9554d73d7ca394 ]

It is possible for the timer handlers to run after the call to
ip_mc_down so use in_dev_put instead of __in_dev_put in the handler
function in order to do proper cleanup when the refcnt reaches 0.
Otherwise, the refcnt can reach zero without the in_device being
destroyed and we end up leaking a reference to the net_device and
see messages like the following,

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

Tested on linux-3.4.43.

Signed-off-by: Salam Noureddine <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agobonding: Fix broken promiscuity reference counting issue
Neil Horman [Fri, 27 Sep 2013 16:22:15 +0000 (12:22 -0400)]
bonding: Fix broken promiscuity reference counting issue

[ Upstream commit 5a0068deb611109c5ba77358be533f763f395ee4 ]

Recently grabbed this report:

Of an issue in which the bonding driver, with an attached vlan encountered the
following errors when bond0 was taken down and back up:

dummy1: promiscuity touches roof, set promiscuity failed. promiscuity feature of
device might be broken.

The error occurs because, during __bond_release_one, if we release our last
slave, we take on a random mac address and issue a NETDEV_CHANGEADDR
notification.  With an attached vlan, the vlan may see that the vlan and bond
mac address were in sync, but no longer are.  This triggers a call to dev_uc_add
and dev_set_rx_mode, which enables IFF_PROMISC on the bond device.  Then, when
we complete __bond_release_one, we use the current state of the bond flags to
determine if we should decrement the promiscuity of the releasing slave.  But
since the bond changed promiscuity state during the release operation, we
incorrectly decrement the slave promisc count when it wasn't in promiscuous mode
to begin with, causing the above error

Fix is pretty simple, just cache the bonding flags at the start of the function
and use those when determining the need to set promiscuity.

This is also needed for the ALLMULTI flag

CC: Jay Vosburgh <>
CC: Andy Gospodarek <>
CC: Mark Wu <>
CC: "David S. Miller" <>
Reported-by: Mark Wu <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agodm9601: fix IFF_ALLMULTI handling
Peter Korsgaard [Mon, 30 Sep 2013 21:28:20 +0000 (23:28 +0200)]
dm9601: fix IFF_ALLMULTI handling

[ Upstream commit bf0ea6380724beb64f27a722dfc4b0edabff816e ]

Pass-all-multicast is controlled by bit 3 in RX control, not bit 2
(pass undersized frames).

Reported-by: Joseph Chang <>
Signed-off-by: Peter Korsgaard <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agovia-rhine: fix VLAN priority field (PCP, IEEE 802.1p)
Roger Luethi [Sat, 21 Sep 2013 12:24:11 +0000 (14:24 +0200)]
via-rhine: fix VLAN priority field (PCP, IEEE 802.1p)

[ Upstream commit 207070f5221e2a901d56a49df9cde47d9b716cd7 ]

Outgoing packets sent by via-rhine have their VLAN PCP field off by one
(when hardware acceleration is enabled). The TX descriptor expects only VID
and PCP (without a CFI/DEI bit).

Peter Boström noticed and reported the bug.

Signed-off-by: Roger Luethi <>
Cc: Peter Boström <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoipv6: udp packets following an UFO enqueued packet need also be handled by UFO
Hannes Frederic Sowa [Sat, 21 Sep 2013 04:27:00 +0000 (06:27 +0200)]
ipv6: udp packets following an UFO enqueued packet need also be handled by UFO

[ Upstream commit 2811ebac2521ceac84f2bdae402455baa6a7fb47 ]

In the following scenario the socket is corked:
If the first UDP packet is larger then the mtu we try to append it to the
write queue via ip6_ufo_append_data. A following packet, which is smaller
than the mtu would be appended to the already queued up gso-skb via
plain ip6_append_data. This causes random memory corruptions.

In ip6_ufo_append_data we also have to be careful to not queue up the
same skb multiple times. So setup the gso frame only when no first skb
is available.

This also fixes a shortcoming where we add the current packet's length to
cork->length but return early because of a packet > mtu with dontfrag set
(instead of sutracting it again).

Found with trinity.

Cc: YOSHIFUJI Hideaki <>
Signed-off-by: Hannes Frederic Sowa <>
Reported-by: Dmitry Vyukov <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoip: generate unique IP identificator if local fragmentation is allowed
Ansis Atteka [Wed, 18 Sep 2013 22:29:53 +0000 (15:29 -0700)]
ip: generate unique IP identificator if local fragmentation is allowed

[ Upstream commit 703133de331a7a7df47f31fb9de51dc6f68a9de8 ]

If local fragmentation is allowed, then ip_select_ident() and
ip_select_ident_more() need to generate unique IDs to ensure
correct defragmentation on the peer.

For example, if IPsec (tunnel mode) has to encrypt large skbs
that have local_df bit set, then all IP fragments that belonged
to different ESP datagrams would have used the same identificator.
If one of these IP fragments would get lost or reordered, then
peer could possibly stitch together wrong IP fragments that did
not belong to the same datagram. This would lead to a packet loss
or data corruption.

Signed-off-by: Ansis Atteka <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoresubmit bridge: fix message_age_timer calculation
Chris Healy [Thu, 12 Sep 2013 04:37:47 +0000 (21:37 -0700)]
resubmit bridge: fix message_age_timer calculation

[ Upstream commit 9a0620133ccce9dd35c00a96405c8d80938c2cc0 ]

This changes the message_age_timer calculation to use the BPDU's max age as
opposed to the local bridge's max age.  This is in accordance with section Step 2 of the 802.1D-1998 sprecification.

With the current implementation, when running with very large bridge
diameters, convergance will not always occur even if a root bridge is
configured to have a longer max age.

Tested successfully on bridge diameters of ~200.

Signed-off-by: Chris Healy <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agonet: sctp: fix ipv6 ipsec encryption bug in sctp_v6_xmit
Daniel Borkmann [Wed, 11 Sep 2013 14:58:36 +0000 (16:58 +0200)]
net: sctp: fix ipv6 ipsec encryption bug in sctp_v6_xmit

[ Upstream commit 95ee62083cb6453e056562d91f597552021e6ae7 ]

Alan Chester reported an issue with IPv6 on SCTP that IPsec traffic is not
being encrypted, whereas on IPv4 it is. Setting up an AH + ESP transport
does not seem to have the desired effect:

SCTP + IPv4:

  22:14:20.809645 IP (tos 0x2,ECT(0), ttl 64, id 0, offset 0, flags [DF], proto AH (51), length 116) > AH(spi=0x00000042,sumlen=16,seq=0x1): ESP(spi=0x00000044,seq=0x1), length 72
  22:14:20.813270 IP (tos 0x2,ECT(0), ttl 64, id 0, offset 0, flags [DF], proto AH (51), length 340) > AH(spi=0x00000043,sumlen=16,seq=0x1):

SCTP + IPv6:

  22:31:19.215029 IP6 (class 0x02, hlim 64, next-header SCTP (132) payload length: 364)
    fe80::222:15ff:fe87:7fc.3333 > fe80::92e6:baff:fe0d:5a54.36767: sctp
    1) [INIT ACK] [init tag: 747759530] [rwnd: 62464] [OS: 10] [MIS: 10]

Moreover, Alan says:

  This problem was seen with both Racoon and Racoon2. Other people have seen
  this with OpenSwan. When IPsec is configured to encrypt all upper layer
  protocols the SCTP connection does not initialize. After using Wireshark to
  follow packets, this is because the SCTP packet leaves Box A unencrypted and
  Box B believes all upper layer protocols are to be encrypted so it drops
  this packet, causing the SCTP connection to fail to initialize. When IPsec
  is configured to encrypt just SCTP, the SCTP packets are observed unencrypted.

In fact, using `socat sctp6-listen:3333 -` on one end and transferring "plaintext"
string on the other end, results in cleartext on the wire where SCTP eventually
does not report any errors, thus in the latter case that Alan reports, the
non-paranoid user might think he's communicating over an encrypted transport on
SCTP although he's not (tcpdump ... -X):

  0x0030: 5d70 8e1a 0003 001a 177d eb6c 0000 0000  ]p.......}.l....
  0x0040: 0000 0000 706c 6169 6e74 6578 740a 0000  ....plaintext...

Only in /proc/net/xfrm_stat we can see XfrmInTmplMismatch increasing on the
receiver side. Initial follow-up analysis from Alan's bug report was done by
Alexey Dobriyan. Also thanks to Vlad Yasevich for feedback on this.

SCTP has its own implementation of sctp_v6_xmit() not calling inet6_csk_xmit().
This has the implication that it probably never really got updated along with
changes in inet6_csk_xmit() and therefore does not seem to invoke xfrm handlers.

SCTP's IPv4 xmit however, properly calls ip_queue_xmit() to do the work. Since
a call to inet6_csk_xmit() would solve this problem, but result in unecessary
route lookups, let us just use the cached flowi6 instead that we got through
sctp_v6_get_dst(). Since all SCTP packets are being sent through sctp_packet_transmit(),
we do the route lookup / flow caching in sctp_transport_route(), hold it in
tp->dst and skb_dst_set() right after that. If we would alter fl6->daddr in
sctp_v6_xmit() to np->opt->srcrt, we possibly could run into the same effect
of not having xfrm layer pick it up, hence, use fl6_update_dst() in sctp_v6_get_dst()
instead to get the correct source routed dst entry, which we assign to the skb.

Also source address routing example from 625034113 ("sctp: fix sctp to work with
ipv6 source address routing") still works with this patch! Nevertheless, in RFC5095
it is actually 'recommended' to not use that anyway due to traffic amplification [1].
So it seems we're not supposed to do that anyway in sctp_v6_xmit(). Moreover, if
we overwrite the flow destination here, the lower IPv6 layer will be unable to
put the correct destination address into IP header, as routing header is added in
ipv6_push_nfrag_opts() but then probably with wrong final destination. Things aside,
result of this patch is that we do not have any XfrmInTmplMismatch increase plus on
the wire with this patch it now looks like:

SCTP + IPv6:

  08:17:47.074080 IP6 2620:52:0:102f:7a2b:cbff:fe27:1b0a > 2620:52:0:102f:213:72ff:fe32:7eba:
    AH(spi=0x00005fb4,seq=0x1): ESP(spi=0x00005fb5,seq=0x1), length 72
  08:17:47.074264 IP6 2620:52:0:102f:213:72ff:fe32:7eba > 2620:52:0:102f:7a2b:cbff:fe27:1b0a:
    AH(spi=0x00003d54,seq=0x1): ESP(spi=0x00003d55,seq=0x1), length 296

This fixes Kernel Bugzilla 24412. This security issue seems to be present since
2.6.18 kernels. Lets just hope some big passive adversary in the wild didn't have
its fun with that. lksctp-tools IPv6 regression test suite passes as well with
this patch.


Reported-by: Alan Chester <>
Reported-by: Alexey Dobriyan <>
Signed-off-by: Daniel Borkmann <>
Cc: Steffen Klassert <>
Cc: Hannes Frederic Sowa <>
Acked-by: Vlad Yasevich <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agonetpoll: fix NULL pointer dereference in netpoll_cleanup
Nikolay Aleksandrov [Thu, 19 Sep 2013 13:02:35 +0000 (15:02 +0200)]
netpoll: fix NULL pointer dereference in netpoll_cleanup

[ Upstream commit d0fe8c888b1fd1a2f84b9962cabcb98a70988aec ]

I've been hitting a NULL ptr deref while using netconsole because the
np->dev check and the pointer manipulation in netpoll_cleanup are done
without rtnl and the following sequence happens when having a netconsole
over a vlan and we remove the vlan while disabling the netconsole:
removes vlan and calls the notifier
enters store_enabled(), calls
netdev_cleanup which checks np->dev
and then waits for rtnl
executes the netconsole netdev
release notifier making np->dev
== NULL and releases rtnl
continues to dereference a member of
np->dev which at this point is == NULL

Signed-off-by: Nikolay Aleksandrov <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agonet: sctp: fix smatch warning in sctp_send_asconf_del_ip
Daniel Borkmann [Sat, 7 Sep 2013 18:51:21 +0000 (20:51 +0200)]
net: sctp: fix smatch warning in sctp_send_asconf_del_ip

[ Upstream commit 88362ad8f9a6cea787420b57cc27ccacef000dbe ]

This was originally reported in [1] and posted by Neil Horman [2], he said:

  Fix up a missed null pointer check in the asconf code. If we don't find
  a local address, but we pass in an address length of more than 1, we may
  dereference a NULL laddr pointer. Currently this can't happen, as the only
  users of the function pass in the value 1 as the addrcnt parameter, but
  its not hot path, and it doesn't hurt to check for NULL should that ever
  be the case.

The callpath from sctp_asconf_mgmt() looks okay. But this could be triggered
from sctp_setsockopt_bindx() call with SCTP_BINDX_REM_ADDR and addrcnt > 1
while passing all possible addresses from the bind list to SCTP_BINDX_REM_ADDR
so that we do *not* find a single address in the association's bind address
list that is not in the packed array of addresses. If this happens when we
have an established association with ASCONF-capable peers, then we could get
a NULL pointer dereference as we only check for laddr == NULL && addrcnt == 1
and call later sctp_make_asconf_update_ip() with NULL laddr.

BUT: this actually won't happen as sctp_bindx_rem() will catch such a case
and return with an error earlier. As this is incredably unintuitive and error
prone, add a check to catch at least future bugs here. As Neil says, its not
hot path. Introduced by 8a07eb0a5 ("sctp: Add ASCONF operation on the
single-homed host").


Reported-by: Dan Carpenter <>
Signed-off-by: Neil Horman <>
Signed-off-by: Daniel Borkmann <>
Cc: Michio Honda <>
Acked-By: Neil Horman <>
Acked-by: Vlad Yasevich <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agocaif: Add missing braces to multiline if in cfctrl_linkup_request
Dave Jones [Thu, 5 Sep 2013 04:11:19 +0000 (00:11 -0400)]
caif: Add missing braces to multiline if in cfctrl_linkup_request

[ Upstream commit 0c1db731bfcf3a9fd6c58132134f8b0f423552f0 ]

The indentation here implies this was meant to be a multi-line if.

Introduced several years back in commit c85c2951d4da1236e32f1858db418221e624aba5
("caif: Handle dev_queue_xmit errors.")

Signed-off-by: Dave Jones <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agopowerpc/iommu: Use GFP_KERNEL instead of GFP_ATOMIC in iommu_init_table()
Nishanth Aravamudan [Tue, 1 Oct 2013 21:04:53 +0000 (14:04 -0700)]
powerpc/iommu: Use GFP_KERNEL instead of GFP_ATOMIC in iommu_init_table()

commit 1cf389df090194a0976dc867b7fffe99d9d490cb upstream.

Under heavy (DLPAR?) stress, we tripped this panic() in

page = alloc_pages_node(nid, GFP_ATOMIC, get_order(sz));
if (!page)
panic("iommu_init_table: Can't allocate %ld bytes\n", sz);

Before the panic() we got a page allocation failure for an order-2
allocation. There appears to be memory free, but perhaps not in the
ATOMIC context. I looked through all the call-sites of
iommu_init_table() and didn't see any obvious reason to need an ATOMIC
allocation. Most call-sites in fact have an explicit GFP_KERNEL
allocation shortly before the call to iommu_init_table(), indicating we
are not in an atomic context. There is some indirection for some paths,
but I didn't see any locks indicating that GFP_KERNEL is inappropriate.

With this change under the same conditions, we have not been able to
reproduce the panic.

Signed-off-by: Nishanth Aravamudan <>
Signed-off-by: Benjamin Herrenschmidt <>
Signed-off-by: Ben Hutchings <>
8 years agopowerpc/sysfs: Disable writing to PURR in guest mode
Madhavan Srinivasan [Tue, 1 Oct 2013 19:04:10 +0000 (00:34 +0530)]
powerpc/sysfs: Disable writing to PURR in guest mode

commit d1211af3049f4c9c1d8d4eb8f8098cc4f4f0d0c7 upstream.

arch/powerpc/kernel/sysfs.c exports PURR with write permission.
This may be valid for kernel in phyp mode. But writing to
the file in guest mode causes crash due to a priviledge violation

Signed-off-by: Madhavan Srinivasan <>
Signed-off-by: Benjamin Herrenschmidt <>
[bwh: Backported to 3.2:
 - Adjust context
 - CPUs are sysdev and we must use the sysdev API]
Signed-off-by: Ben Hutchings <>
8 years agopowerpc: Restore registers on error exit from csum_partial_copy_generic()
Paul E. McKenney [Tue, 1 Oct 2013 07:11:35 +0000 (17:11 +1000)]
powerpc: Restore registers on error exit from csum_partial_copy_generic()

commit 8f21bd0090052e740944f9397e2be5ac7957ded7 upstream.

The csum_partial_copy_generic() function saves the PowerPC non-volatile
r14, r15, and r16 registers for the main checksum-and-copy loop.
Unfortunately, it fails to restore them upon error exit from this loop,
which results in silent corruption of these registers in the presumably
rare event of an access exception within that loop.

This commit therefore restores these register on error exit from the loop.

Signed-off-by: Paul E. McKenney <>
Signed-off-by: Anton Blanchard <>
Signed-off-by: Benjamin Herrenschmidt <>
[bwh: Backported to 3.2: register name macros use lower-case 'r']
Signed-off-by: Ben Hutchings <>
8 years agopowerpc: Fix parameter clobber in csum_partial_copy_generic()
Paul E. McKenney [Tue, 1 Oct 2013 06:54:05 +0000 (16:54 +1000)]
powerpc: Fix parameter clobber in csum_partial_copy_generic()

commit d9813c3681a36774b254c0cdc9cce53c9e22c756 upstream.

The csum_partial_copy_generic() uses register r7 to adjust the remaining
bytes to process.  Unfortunately, r7 also holds a parameter, namely the
address of the flag to set in case of access exceptions while reading
the source buffer.  Lacking a quantum implementation of PowerPC, this
commit instead uses register r9 to do the adjusting, leaving r7's
pointer uncorrupted.

Signed-off-by: Paul E. McKenney <>
Signed-off-by: Anton Blanchard <>
Signed-off-by: Benjamin Herrenschmidt <>
Signed-off-by: Ben Hutchings <>
8 years agoUSB: serial: option: Ignore card reader interface on Huawei E1750
Michal Malý [Sat, 28 Sep 2013 17:50:27 +0000 (19:50 +0200)]
USB: serial: option: Ignore card reader interface on Huawei E1750

commit eb2addd4044b4b2ce77693bde5bc810536dd96ee upstream.


my Huawei 3G modem has an embedded Smart Card reader which causes
trouble when the modem is being detected (a bunch of "<warn>  (ttyUSBx):
open blocked by driver for more than 7 seconds!" in messages.log). This
trivial patch corrects the problem for me. The modem identifies itself
as "12d1:1406 Huawei Technologies Co., Ltd. E1750" in lsusb although the
description on the body says "Model E173u-1"

Signed-off-by: Michal Malý <>
Cc: Bjørn Mork <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agonilfs2: fix issue with race condition of competition between segments for dirty blocks
Vyacheslav Dubeyko [Mon, 30 Sep 2013 20:45:12 +0000 (13:45 -0700)]
nilfs2: fix issue with race condition of competition between segments for dirty blocks

commit 7f42ec3941560f0902fe3671e36f2c20ffd3af0a upstream.

Many NILFS2 users were reported about strange file system corruption
(for example):

   NILFS: bad btree node (blocknr=185027): level = 0, flags = 0x0, nchildren = 768
   NILFS error (device sda4): nilfs_bmap_last_key: broken bmap (inode number=11540)

But such error messages are consequence of file system's issue that takes
place more earlier.  Fortunately, Jerome Poulin <>
and Anton Eliasson <> were reported about another
issue not so recently.  These reports describe the issue with segctor
thread's crash:

  BUG: unable to handle kernel paging request at 0000000000004c83
  IP: nilfs_end_page_io+0x12/0xd0 [nilfs2]

  Call Trace:
   nilfs_segctor_do_construct+0xf25/0x1b20 [nilfs2]
   nilfs_segctor_construct+0x17b/0x290 [nilfs2]
   nilfs_segctor_thread+0x122/0x3b0 [nilfs2]

These two issues have one reason.  This reason can raise third issue
too.  Third issue results in hanging of segctor thread with eating of
100% CPU.


One of the possible way or the issue reproducing was described by
Jermoe me Poulin <>:

1. init S to get to single user mode.
2. sysrq+E to make sure only my shell is running
3. start network-manager to get my wifi connection up
4. login as root and launch "screen"
5. cd /boot/log/nilfs which is a ext3 mount point and can log when NILFS dies.
6. lscp | xz -9e > lscp.txt.xz
7. mount my snapshot using mount -o cp=3360839,ro /dev/vgUbuntu/root /mnt/nilfs
8. start a screen to dump /proc/kmsg to text file since rsyslog is killed
9. start a screen and launch strace -f -o find-cat.log -t find
/mnt/nilfs -type f -exec cat {} > /dev/null \;
10. start a screen and launch strace -f -o apt-get.log -t apt-get update
11. launch the last command again as it did not crash the first time
12. apt-get crashes
13. ps aux > ps-aux-crashed.log
13. sysrq+W
14. sysrq+E  wait for everything to terminate
15. sysrq+SUSB

Simplified way of the issue reproducing is starting kernel compilation
task and "apt-get update" in parallel.


The issue is reproduced not stable [60% - 80%].  It is very important to
have proper environment for the issue reproducing.  The critical
conditions for successful reproducing:

(1) It should have big modified file by mmap() way.

(2) This file should have the count of dirty blocks are greater that
    several segments in size (for example, two or three) from time to time
    during processing.

(3) It should be intensive background activity of files modification
    in another thread.


First of all, it is possible to see that the reason of crash is not valid
page address:

  NILFS [nilfs_segctor_complete_write]:2100 bh->b_count 0, bh->b_blocknr 13895680, bh->b_size 13897727, bh->b_page 0000000000001a82
  NILFS [nilfs_segctor_complete_write]:2101 segbuf->sb_segnum 6783

Moreover, value of b_page (0x1a82) is 6786.  This value looks like segment
number.  And b_blocknr with b_size values look like block numbers.  So,
buffer_head's pointer points on not proper address value.

Detailed investigation of the issue is discovered such picture:

  [-----------------------------SEGMENT 6783-------------------------------]
  NILFS [nilfs_segctor_do_construct]:2310 nilfs_segctor_begin_construction
  NILFS [nilfs_segctor_do_construct]:2321 nilfs_segctor_collect
  NILFS [nilfs_segctor_do_construct]:2336 nilfs_segctor_assign
  NILFS [nilfs_segctor_do_construct]:2367 nilfs_segctor_update_segusage
  NILFS [nilfs_segctor_do_construct]:2371 nilfs_segctor_prepare_write
  NILFS [nilfs_segctor_do_construct]:2376 nilfs_add_checksums_on_logs
  NILFS [nilfs_segctor_do_construct]:2381 nilfs_segctor_write
  NILFS [nilfs_segbuf_submit_bio]:464 bio->bi_sector 111149024, segbuf->sb_segnum 6783

  [-----------------------------SEGMENT 6784-------------------------------]
  NILFS [nilfs_segctor_do_construct]:2310 nilfs_segctor_begin_construction
  NILFS [nilfs_segctor_do_construct]:2321 nilfs_segctor_collect
  NILFS [nilfs_lookup_dirty_data_buffers]:782 bh->b_count 1, bh->b_page ffffea000709b000, page->index 0, i_ino 1033103, i_size 25165824
  NILFS [nilfs_lookup_dirty_data_buffers]:783 bh-> ffff8802174a6798, bh->b_assoc_buffers.prev ffff880221cffee8
  NILFS [nilfs_segctor_do_construct]:2336 nilfs_segctor_assign
  NILFS [nilfs_segctor_do_construct]:2367 nilfs_segctor_update_segusage
  NILFS [nilfs_segctor_do_construct]:2371 nilfs_segctor_prepare_write
  NILFS [nilfs_segctor_do_construct]:2376 nilfs_add_checksums_on_logs
  NILFS [nilfs_segctor_do_construct]:2381 nilfs_segctor_write
  NILFS [nilfs_segbuf_submit_bh]:575 bh->b_count 1, bh->b_page ffffea000709b000, page->index 0, i_ino 1033103, i_size 25165824
  NILFS [nilfs_segbuf_submit_bh]:576 segbuf->sb_segnum 6784
  NILFS [nilfs_segbuf_submit_bh]:577 bh-> ffff880218a0d5f8, bh->b_assoc_buffers.prev ffff880218bcdf50
  NILFS [nilfs_segbuf_submit_bio]:464 bio->bi_sector 111150080, segbuf->sb_segnum 6784, segbuf->sb_nbio 0
  [----------] ditto
  NILFS [nilfs_segbuf_submit_bio]:464 bio->bi_sector 111164416, segbuf->sb_segnum 6784, segbuf->sb_nbio 15

  [-----------------------------SEGMENT 6785-------------------------------]
  NILFS [nilfs_segctor_do_construct]:2310 nilfs_segctor_begin_construction
  NILFS [nilfs_segctor_do_construct]:2321 nilfs_segctor_collect
  NILFS [nilfs_lookup_dirty_data_buffers]:782 bh->b_count 2, bh->b_page ffffea000709b000, page->index 0, i_ino 1033103, i_size 25165824
  NILFS [nilfs_lookup_dirty_data_buffers]:783 bh-> ffff880219277e80, bh->b_assoc_buffers.prev ffff880221cffc88
  NILFS [nilfs_segctor_do_construct]:2367 nilfs_segctor_update_segusage
  NILFS [nilfs_segctor_do_construct]:2371 nilfs_segctor_prepare_write
  NILFS [nilfs_segctor_do_construct]:2376 nilfs_add_checksums_on_logs
  NILFS [nilfs_segctor_do_construct]:2381 nilfs_segctor_write
  NILFS [nilfs_segbuf_submit_bh]:575 bh->b_count 2, bh->b_page ffffea000709b000, page->index 0, i_ino 1033103, i_size 25165824
  NILFS [nilfs_segbuf_submit_bh]:576 segbuf->sb_segnum 6785
  NILFS [nilfs_segbuf_submit_bh]:577 bh-> ffff880218a0d5f8, bh->b_assoc_buffers.prev ffff880222cc7ee8
  NILFS [nilfs_segbuf_submit_bio]:464 bio->bi_sector 111165440, segbuf->sb_segnum 6785, segbuf->sb_nbio 0
  [----------] ditto
  NILFS [nilfs_segbuf_submit_bio]:464 bio->bi_sector 111177728, segbuf->sb_segnum 6785, segbuf->sb_nbio 12

  NILFS [nilfs_segctor_do_construct]:2399 nilfs_segctor_wait
  NILFS [nilfs_segbuf_wait]:676 segbuf->sb_segnum 6783
  NILFS [nilfs_segbuf_wait]:676 segbuf->sb_segnum 6784
  NILFS [nilfs_segbuf_wait]:676 segbuf->sb_segnum 6785

  NILFS [nilfs_segctor_complete_write]:2100 bh->b_count 0, bh->b_blocknr 13895680, bh->b_size 13897727, bh->b_page 0000000000001a82

  BUG: unable to handle kernel paging request at 0000000000001a82
  IP: [<ffffffffa024d0f2>] nilfs_end_page_io+0x12/0xd0 [nilfs2]

Usually, for every segment we collect dirty files in list.  Then, dirty
blocks are gathered for every dirty file, prepared for write and
submitted by means of nilfs_segbuf_submit_bh() call.  Finally, it takes
place complete write phase after calling nilfs_end_bio_write() on the
block layer.  Buffers/pages are marked as not dirty on final phase and
processed files removed from the list of dirty files.

It is possible to see that we had three prepare_write and submit_bio
phases before segbuf_wait and complete_write phase.  Moreover, segments
compete between each other for dirty blocks because on every iteration
of segments processing dirty buffer_heads are added in several lists of

  [SEGMENT 6784]: bh-> ffff880218a0d5f8, bh->b_assoc_buffers.prev ffff880218bcdf50
  [SEGMENT 6785]: bh-> ffff880218a0d5f8, bh->b_assoc_buffers.prev ffff880222cc7ee8

The next pointer is the same but prev pointer has changed.  It means
that buffer_head has next pointer from one list but prev pointer from
another.  Such modification can be made several times.  And, finally, it
can be resulted in various issues: (1) segctor hanging, (2) segctor
crashing, (3) file system metadata corruption.

This patch adds:

(1) setting of BH_Async_Write flag in nilfs_segctor_prepare_write()
    for every proccessed dirty block;

(2) checking of BH_Async_Write flag in
    nilfs_lookup_dirty_data_buffers() and

(3) clearing of BH_Async_Write flag in nilfs_segctor_complete_write(),
    nilfs_abort_logs(), nilfs_forget_buffer(), nilfs_clear_dirty_page().

Reported-by: Jerome Poulin <>
Reported-by: Anton Eliasson <>
Cc: Paul Fertser <>
Cc: ARAI Shun-ichi <>
Cc: Piotr Szymaniak <>
Cc: Juan Barry Manuel Canham <>
Cc: Zahid Chowdhury <>
Cc: Elmer Zhang <>
Cc: Kenneth Langga <>
Signed-off-by: Vyacheslav Dubeyko <>
Acked-by: Ryusuke Konishi <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
[bwh: Backported to 3.2: nilfs_clear_dirty_page() has not been separated
 from nilfs_clear_dirty_pages()]
Signed-off-by: Ben Hutchings <>
8 years agocan: flexcan: fix flexcan_chip_start() on imx6
Marc Kleine-Budde [Fri, 27 Sep 2013 10:15:05 +0000 (12:15 +0200)]
can: flexcan: fix flexcan_chip_start() on imx6

commit 0d1862ea1a5bb876cf05555a7307080cb75bf379 upstream.

In the flexcan_chip_start() function first the flexcan core is going through
the soft reset sequence, then the RX FIFO is enabled.

With the hardware is put into FIFO mode, message buffers 1...7 are reserved by
the FIFO engine. The remaining message buffers are in reset default values.
This patch removes the bogus initialization of the message buffers, as it
causes an imprecise external abort on imx6.

Reported-by: Lothar Waßmann <>
Tested-by: Lothar Waßmann <>
Signed-off-by: Marc Kleine-Budde <>
Signed-off-by: Ben Hutchings <>
8 years agousb: dwc3: add support for Merrifield
David Cohen [Thu, 26 Sep 2013 20:01:44 +0000 (13:01 -0700)]
usb: dwc3: add support for Merrifield

commit 85601f8cf67c56a561a6dd5e130e65fdc179047d upstream.

Add PCI id for Intel Merrifield

Signed-off-by: David Cohen <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agousb: dwc3: pci: add support for BayTrail
Heikki Krogerus [Tue, 17 Sep 2013 07:38:13 +0000 (10:38 +0300)]
usb: dwc3: pci: add support for BayTrail

commit b62cd96de3161dfb125a769030eec35a4cab3d3a upstream.

Add PCI id for Intel BayTrail.

Signed-off-by: Heikki Krogerus <>
Signed-off-by: Felipe Balbi <>
Signed-off-by: Ben Hutchings <>
8 years agop54usb: add USB ID for Corega WLUSB2GTST USB adapter
Christian Lamparter [Tue, 24 Sep 2013 19:56:46 +0000 (21:56 +0200)]
p54usb: add USB ID for Corega WLUSB2GTST USB adapter

commit 1e43692cdb7cc445d6347d8a5207d9cef0c71434 upstream.

Added USB ID for Corega WLUSB2GTST USB adapter.

Reported-by: Joerg Kalisch <>
Signed-off-by: Christian Lamparter <>
Signed-off-by: John W. Linville <>
Signed-off-by: Ben Hutchings <>
8 years agortlwifi: Align private space in rtl_priv struct
Larry Finger [Thu, 19 Sep 2013 02:21:35 +0000 (21:21 -0500)]
rtlwifi: Align private space in rtl_priv struct

commit 60ce314d1750fef843e9db70050e09e49f838b69 upstream.

The private array at the end of the rtl_priv struct is not aligned.
On ARM architecture, this causes an alignment trap and is fixed by aligning
that array with __align(sizeof(void *)). That should properly align that
space according to the requirements of all architectures.

Reported-by: Jason Andrews <>
Tested-by: Jason Andrews <>
Signed-off-by: Larry Finger <>
Signed-off-by: John W. Linville <>
Signed-off-by: Ben Hutchings <>
8 years agohwmon: (applesmc) Check key count before proceeding
Henrik Rydberg [Thu, 26 Sep 2013 06:33:16 +0000 (08:33 +0200)]
hwmon: (applesmc) Check key count before proceeding

commit 5f4513864304672e6ea9eac60583eeac32e679f2 upstream.

After reports from Chris and Josh Boyer of a rare crash in applesmc,
Guenter pointed at the initialization problem fixed below. The patch
has not been verified to fix the crash, but should be applied

Reported-by: <>
Suggested-by: Guenter Roeck <>
Signed-off-by: Henrik Rydberg <>
Signed-off-by: Guenter Roeck <>
Signed-off-by: Ben Hutchings <>
8 years agousb/core/devio.c: Don't reject control message to endpoint with wrong direction bit
Kurt Garloff [Tue, 24 Sep 2013 12:13:48 +0000 (14:13 +0200)]
usb/core/devio.c: Don't reject control message to endpoint with wrong direction bit

commit 831abf76643555a99b80a3b54adfa7e4fa0a3259 upstream.

Trying to read data from the Pegasus Technologies NoteTaker (0e20:0101)
[1] with the Windows App (EasyNote) works natively but fails when
Windows is running under KVM (and the USB device handed to KVM).

The reason is a USB control message
 usb 4-2.2: control urb: bRequestType=22 bRequest=09 wValue=0200 wIndex=0001 wLength=0008
This goes to endpoint address 0x01 (wIndex); however, endpoint address
0x01 does not exist. There is an endpoint 0x81 though (same number,
but other direction); the app may have meant that endpoint instead.

The kernel thus rejects the IO and thus we see the failure.

Apparently, Linux is more strict here than Windows ... we can't change
the Win app easily, so that's a problem.

It seems that the Win app/driver is buggy here and the driver does not
behave fully according to the USB HID class spec that it claims to
belong to.  The device seems to happily deal with that though (and
seems to not really care about this value much).

So the question is whether the Linux kernel should filter here.
Rejecting has the risk that somewhat non-compliant userspace apps/
drivers (most likely in a virtual machine) are prevented from working.
Not rejecting has the risk of confusing an overly sensitive device with
such a transfer. Given the fact that Windows does not filter it makes
this risk rather small though.

The patch makes the kernel more tolerant: If the endpoint address in
wIndex does not exist, but an endpoint with toggled direction bit does,
it will let the transfer through. (It does NOT change the message.)

With attached patch, the app in Windows in KVM works.
 usb 4-2.2: check_ctrlrecip: process 13073 (qemu-kvm) requesting ep 01 but needs 81

I suspect this will mostly affect apps in virtual environments; as on
Linux the apps would have been adapted to the stricter handling of the
kernel. I have done that for mine[2].


Signed-off-by: Kurt Garloff <>
Acked-by: Alan Stern <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agoUSB: fix PM config symbol in uhci-hcd, ehci-hcd, and xhci-hcd
Alan Stern [Tue, 24 Sep 2013 19:45:25 +0000 (15:45 -0400)]
USB: fix PM config symbol in uhci-hcd, ehci-hcd, and xhci-hcd

commit f875fdbf344b9fde207f66b392c40845dd7e5aa6 upstream.

Since uhci-hcd, ehci-hcd, and xhci-hcd support runtime PM, the .pm
field in their pci_driver structures should be protected by CONFIG_PM
rather than CONFIG_PM_SLEEP.  The corresponding change has already
been made for ohci-hcd.

Without this change, controllers won't do runtime suspend if system
suspend or hibernation isn't enabled.

Signed-off-by: Alan Stern <>
CC: Sarah Sharp <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agostaging: vt6656: [BUG] main_usb.c oops on device_close move flag earlier.
Malcolm Priestley [Sun, 22 Sep 2013 18:48:54 +0000 (19:48 +0100)]
staging: vt6656: [BUG] main_usb.c oops on device_close move flag earlier.

commit e3eb270fab7734427dd8171a93e4946fe28674bc upstream.

The vt6656 is prone to resetting on the usb bus.

It seems there is a race condition and wpa supplicant is
trying to open the device via iw_handlers before its actually
closed at a stage that the buffers are being removed.

The device is longer considered open when the
buffers are being removed. So move ~DEVICE_FLAGS_OPENED
flag to before freeing the device buffers.

Signed-off-by: Malcolm Priestley <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agodrm/i915/dp: increase i2c-over-aux retry interval on AUX DEFER
Jani Nikula [Fri, 20 Sep 2013 13:42:15 +0000 (16:42 +0300)]
drm/i915/dp: increase i2c-over-aux retry interval on AUX DEFER

commit 8d16f258217f2f583af1fd57c5144aa4bbe73e48 upstream.

There is no clear cut rules or specs for the retry interval, as there
are many factors that affect overall response time. Increase the
interval, and even more so on branch devices which may have limited i2c
bit rates.

Signed-off-by: Jani Nikula <>
Tested-by: Nicolas Suzor <>
Reviewed-by: Todd Previte <>
Signed-off-by: Daniel Vetter <>
Signed-off-by: Ben Hutchings <>
8 years agodrm/radeon: disable tests/benchmarks if accel is disabled
Alex Deucher [Mon, 23 Sep 2013 14:38:26 +0000 (10:38 -0400)]
drm/radeon: disable tests/benchmarks if accel is disabled

commit 4a1132a023eb48cf10522d84c5908d43b612c041 upstream.

The tests are only usable if the acceleration engines have
been successfully initialized.

Based on an initial patch from: Alex Ivanov <>

Signed-off-by: Alex Deucher <>
[bwh: Backported to 3.2: there is only one test: radeon_test_moves()]
Signed-off-by: Ben Hutchings <>
8 years agox86/reboot: Add quirk to make Dell C6100 use reboot=pci automatically
Masoud Sharbiani [Fri, 20 Sep 2013 22:59:07 +0000 (15:59 -0700)]
x86/reboot: Add quirk to make Dell C6100 use reboot=pci automatically

commit 4f0acd31c31f03ba42494c8baf6c0465150e2621 upstream.

Dell PowerEdge C6100 machines fail to completely reboot about 20% of the time.

Signed-off-by: Masoud Sharbiani <>
Signed-off-by: Vinson Lee <>
Cc: Robin Holt <>
Cc: Russell King <>
Cc: Guan Xuetao <>
Signed-off-by: Ingo Molnar <>
Signed-off-by: Ben Hutchings <>
8 years agodm-snapshot: fix performance degradation due to small hash size
Mikulas Patocka [Wed, 18 Sep 2013 23:40:42 +0000 (19:40 -0400)]
dm-snapshot: fix performance degradation due to small hash size

commit 60e356f381954d79088d0455e357db48cfdd6857 upstream.

LVM2, since version 2.02.96, creates origin with zero size, then loads
the snapshot driver and then loads the origin.  Consequently, the
snapshot driver sees the origin size zero and sets the hash size to the
lower bound 64.  Such small hash table causes performance degradation.

This patch changes it so that the hash size is determined by the size of
snapshot volume, not minimum of origin and snapshot size.  It doesn't
make sense to set the snapshot size significantly larger than the origin
size, so we do not need to take origin size into account when
calculating the hash size.

Signed-off-by: Mikulas Patocka <>
Signed-off-by: Mike Snitzer <>
Signed-off-by: Ben Hutchings <>
8 years agox86, efi: Don't map Boot Services on i386
Josh Boyer [Thu, 18 Apr 2013 14:51:34 +0000 (07:51 -0700)]
x86, efi: Don't map Boot Services on i386

commit 700870119f49084da004ab588ea2b799689efaf7 upstream.

Add patch to fix 32bit EFI service mapping (rhbz 726701)

Multiple people are reporting hitting the following WARNING on i386,

  WARNING: at arch/x86/mm/ioremap.c:102 __ioremap_caller+0x3d3/0x440()
  Modules linked in:
  Pid: 0, comm: swapper Not tainted 3.9.0-rc7+ #95
  Call Trace:
   [<c102b6af>] warn_slowpath_common+0x5f/0x80
   [<c1023fb3>] ? __ioremap_caller+0x3d3/0x440
   [<c1023fb3>] ? __ioremap_caller+0x3d3/0x440
   [<c102b6ed>] warn_slowpath_null+0x1d/0x20
   [<c1023fb3>] __ioremap_caller+0x3d3/0x440
   [<c106007b>] ? get_usage_chars+0xfb/0x110
   [<c102d937>] ? vprintk_emit+0x147/0x480
   [<c1418593>] ? efi_enter_virtual_mode+0x1e4/0x3de
   [<c102406a>] ioremap_cache+0x1a/0x20
   [<c1418593>] ? efi_enter_virtual_mode+0x1e4/0x3de
   [<c1418593>] efi_enter_virtual_mode+0x1e4/0x3de
   [<c1407984>] start_kernel+0x286/0x2f4
   [<c1407535>] ? repair_env_string+0x51/0x51
   [<c1407362>] i386_start_kernel+0x12c/0x12f

Due to the workaround described in commit 916f676f8 ("x86, efi: Retain
boot service code until after switching to virtual mode") EFI Boot
Service regions are mapped for a period during boot. Unfortunately, with
the limited size of the i386 direct kernel map it's possible that some
of the Boot Service regions will not be directly accessible, which
causes them to be ioremap()'d, triggering the above warning as the
regions are marked as E820_RAM in the e820 memmap.

There are currently only two situations where we need to map EFI Boot
Service regions,

  1. To workaround the firmware bug described in 916f676f8
  2. To access the ACPI BGRT image

but since we haven't seen an i386 implementation that requires either,
this simple fix should suffice for now.

[ Added to changelog - Matt ]

Reported-by: Bryan O'Donoghue <>
Acked-by: Tom Zanussi <>
Acked-by: Darren Hart <>
Cc: Josh Triplett <>
Cc: Matthew Garrett <>
Cc: H. Peter Anvin <>
Cc: Ingo Molnar <>
Cc: Thomas Gleixner <>
Cc: Greg Kroah-Hartman <>
Signed-off-by: Josh Boyer <>
Signed-off-by: Matt Fleming <>
Signed-off-by: Ben Hutchings <>
8 years agoserial: pch_uart: fix tty-kref leak in dma-rx path
Johan Hovold [Tue, 10 Sep 2013 10:50:50 +0000 (12:50 +0200)]
serial: pch_uart: fix tty-kref leak in dma-rx path

commit 19b85cfb190eb9980eaf416bff96aef4159a430e upstream.

Fix tty_kref leak when tty_buffer_request room fails in dma-rx path.

Note that the tty ref isn't really needed anymore, but as the leak has
always been there, fixing it before removing should makes it easier to
backport the fix.

Signed-off-by: Johan Hovold <>
Signed-off-by: Greg Kroah-Hartman <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agoserial: pch_uart: fix tty-kref leak in rx-error path
Johan Hovold [Tue, 10 Sep 2013 10:50:49 +0000 (12:50 +0200)]
serial: pch_uart: fix tty-kref leak in rx-error path

commit fc0919c68cb2f75bb1af759315f9d7e2a9443c28 upstream.

Fix tty-kref leak introduced by commit 384e301e ("pch_uart: fix a
deadlock when pch_uart as console") which never put its tty reference.

Signed-off-by: Johan Hovold <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agonet: usb: cdc_ether: Use wwan interface for Telit modules
Fabio Porcedda [Mon, 16 Sep 2013 09:47:50 +0000 (11:47 +0200)]
net: usb: cdc_ether: Use wwan interface for Telit modules

commit 0092820407901a0b2c4e343e85f96bb7abfcded1 upstream.

Signed-off-by: Fabio Porcedda <>
Acked-by: Oliver Neukum <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoBluetooth: Add support for BCM20702A0 [0b05, 17cb]
Raphael Kubo da Costa [Mon, 2 Sep 2013 11:57:51 +0000 (14:57 +0300)]
Bluetooth: Add support for BCM20702A0 [0b05, 17cb]

commit 38a172bef8c93ecbfd69715fd88396988e4073fd upstream.

Yet another vendor specific ID for this chipset; this one for the ASUS
USB-BT400 Bluetooth 4.0 adapter.

T:  Bus=03 Lev=02 Prnt=02 Port=01 Cnt=01 Dev#=  6 Spd=12  MxCh= 0
D:  Ver= 2.00 Cls=ff(vend.) Sub=01 Prot=01 MxPS=64 #Cfgs=  1
P:  Vendor=0b05 ProdID=17cb Rev=01.12
S:  Manufacturer=Broadcom Corp
S:  Product=BCM20702A0
S:  SerialNumber=000272C64400
C:  #Ifs= 4 Cfg#= 1 Atr=a0 MxPwr=100mA
I:  If#= 0 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=01 Prot=01 Driver=(none)
I:  If#= 1 Alt= 0 #EPs= 2 Cls=ff(vend.) Sub=01 Prot=01 Driver=(none)
I:  If#= 2 Alt= 0 #EPs= 2 Cls=ff(vend.) Sub=ff Prot=ff Driver=(none)
I:  If#= 3 Alt= 0 #EPs= 0 Cls=fe(app. ) Sub=01 Prot=01 Driver=(none)

Signed-off-by: Raphael Kubo da Costa <>
Signed-off-by: Gustavo Padovan <>
Signed-off-by: Ben Hutchings <>
8 years agoBluetooth: Add a new PID/VID 0cf3/e005 for AR3012.
Peng Chen [Fri, 30 Aug 2013 09:41:40 +0000 (17:41 +0800)]
Bluetooth: Add a new PID/VID 0cf3/e005 for AR3012.

commit 0a3658cccdf5326ea508efeb1879b0e2508bb0c3 upstream.

usb device info:

T:  Bus=06 Lev=01 Prnt=01 Port=01 Cnt=01 Dev#= 15 Spd=12   MxCh= 0
D:  Ver= 1.10 Cls=e0(wlcon) Sub=01 Prot=01 MxPS=64 #Cfgs=  1
P:  Vendor=0cf3 ProdID=e005 Rev= 0.02
C:* #Ifs= 2 Cfg#= 1 Atr=e0 MxPwr=100mA
I:* If#= 0 Alt= 0 #EPs= 3 Cls=e0(wlcon) Sub=01 Prot=01 Driver=btusb
E:  Ad=81(I) Atr=03(Int.) MxPS=  16 Ivl=1ms
E:  Ad=82(I) Atr=02(Bulk) MxPS=  64 Ivl=0ms
E:  Ad=02(O) Atr=02(Bulk) MxPS=  64 Ivl=0ms
I:* If#= 1 Alt= 0 #EPs= 2 Cls=e0(wlcon) Sub=01 Prot=01 Driver=btusb
E:  Ad=83(I) Atr=01(Isoc) MxPS=   0 Ivl=1ms
E:  Ad=03(O) Atr=01(Isoc) MxPS=   0 Ivl=1ms

Signed-off-by: Peng Chen <>
Signed-off-by: Gustavo Padovan <>
Signed-off-by: Ben Hutchings <>
8 years agodrm/radeon: fix panel scaling with eDP and LVDS bridges
Alex Deucher [Fri, 13 Sep 2013 22:33:16 +0000 (18:33 -0400)]
drm/radeon: fix panel scaling with eDP and LVDS bridges

commit 855f5f1d882a34e4e9dd27b299737cd3508a5624 upstream.

We were using the wrong set_properly callback so we always
ended up with Full scaling even if something else (Center or
Full aspect) was selected.

Signed-off-by: Alex Deucher <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agoASoC: 88pm860x: array overflow in snd_soc_put_volsw_2r_st()
Dan Carpenter [Fri, 13 Sep 2013 07:52:49 +0000 (10:52 +0300)]
ASoC: 88pm860x: array overflow in snd_soc_put_volsw_2r_st()

commit d967967e8d1116fb38bad25e58714b5dddd03cca upstream.

This is called from snd_ctl_elem_write() with user supplied data so we
need to add some bounds checking.

Signed-off-by: Dan Carpenter <>
Signed-off-by: Mark Brown <>
Signed-off-by: Ben Hutchings <>
8 years agoASoC: max98095: a couple array underflows
Dan Carpenter [Fri, 13 Sep 2013 07:52:14 +0000 (10:52 +0300)]
ASoC: max98095: a couple array underflows

commit f8d7b13e14357ed19d2ca2799539600418dc3939 upstream.

The ->put() function are called from snd_ctl_elem_write() with user
supplied data.  The limit checks here could underflow leading to a

Signed-off-by: Dan Carpenter <>
Signed-off-by: Mark Brown <>
Signed-off-by: Ben Hutchings <>
8 years agoHID: logitech-dj: validate output report details
Kees Cook [Wed, 11 Sep 2013 19:56:56 +0000 (21:56 +0200)]
HID: logitech-dj: validate output report details

commit 297502abb32e225fb23801fcdb0e4f6f8e17099a upstream.

A HID device could send a malicious output report that would cause the
logitech-dj HID driver to leak kernel memory contents to the device, or
trigger a NULL dereference during initialization:

[  304.424553] usb 1-1: New USB device found, idVendor=046d, idProduct=c52b
[  304.780467] BUG: unable to handle kernel NULL pointer dereference at 0000000000000028
[  304.781409] IP: [<ffffffff815d50aa>] logi_dj_recv_send_report.isra.11+0x1a/0x90


Signed-off-by: Kees Cook <>
Signed-off-by: Benjamin Tissoires <>
Signed-off-by: Jiri Kosina <>
[bwh: Backported to 3.2: drop inapplicable changes to
Signed-off-by: Ben Hutchings <>
8 years agoHID: validate feature and input report details
Benjamin Tissoires [Wed, 11 Sep 2013 19:56:57 +0000 (21:56 +0200)]
HID: validate feature and input report details

commit cc6b54aa54bf40b762cab45a9fc8aa81653146eb upstream.

When dealing with usage_index, be sure to properly use unsigned instead of
int to avoid overflows.

When working on report fields, always validate that their report_counts are
in bounds.
Without this, a HID device could report a malicious feature report that
could trick the driver into a heap overflow:

[  634.885003] usb 1-1: New USB device found, idVendor=0596, idProduct=0500
[  676.469629] BUG kmalloc-192 (Tainted: G        W   ): Redzone overwritten


Signed-off-by: Benjamin Tissoires <>
Acked-by: Kees Cook <>
Signed-off-by: Jiri Kosina <>
[bwh: Backported to 3.2:
 - Drop inapplicable changes to hid_usage::usage_index initialisation and
   to hid_report_raw_event()
 - Adjust context in report_features()
Signed-off-by: Ben Hutchings <>
8 years agoHID: LG: validate HID output report details
Kees Cook [Wed, 11 Sep 2013 19:56:54 +0000 (21:56 +0200)]
HID: LG: validate HID output report details

commit 0fb6bd06e06792469acc15bbe427361b56ada528 upstream.

A HID device could send a malicious output report that would cause the
lg, lg3, and lg4 HID drivers to write beyond the output report allocation
during an event, causing a heap overflow:

[  325.245240] usb 1-1: New USB device found, idVendor=046d, idProduct=c287
[  414.518960] BUG kmalloc-4096 (Not tainted): Redzone overwritten

Additionally, while lg2 did correctly validate the report details, it was
cleaned up and shortened.


Signed-off-by: Kees Cook <>
Reviewed-by: Benjamin Tissoires <>
Signed-off-by: Jiri Kosina <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agoHID: zeroplus: validate output report details
Kees Cook [Wed, 11 Sep 2013 19:56:51 +0000 (21:56 +0200)]
HID: zeroplus: validate output report details

commit 78214e81a1bf43740ce89bb5efda78eac2f8ef83 upstream.

The zeroplus HID driver was not checking the size of allocated values
in fields it used. A HID device could send a malicious output report
that would cause the driver to write beyond the output report allocation
during initialization, causing a heap overflow:

[ 1442.728680] usb 1-1: New USB device found, idVendor=0c12, idProduct=0005
[ 1466.243173] BUG kmalloc-192 (Tainted: G        W   ): Redzone overwritten


Signed-off-by: Kees Cook <>
Reviewed-by: Benjamin Tissoires <>
Signed-off-by: Jiri Kosina <>
Signed-off-by: Ben Hutchings <>
8 years agoHID: provide a helper for validating hid reports
Kees Cook [Wed, 11 Sep 2013 19:56:50 +0000 (21:56 +0200)]
HID: provide a helper for validating hid reports

commit 331415ff16a12147d57d5c953f3a961b7ede348b upstream.

Many drivers need to validate the characteristics of their HID report
during initialization to avoid misusing the reports. This adds a common
helper to perform validation of the report exisitng, the field existing,
and the expected number of values within the field.

Signed-off-by: Kees Cook <>
Reviewed-by: Benjamin Tissoires <>
Signed-off-by: Jiri Kosina <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agosched/fair: Fix small race where child->se.parent,cfs_rq might point to invalid ones
Daisuke Nishimura [Tue, 10 Sep 2013 09:16:36 +0000 (18:16 +0900)]
sched/fair: Fix small race where child->se.parent,cfs_rq might point to invalid ones

commit 6c9a27f5da9609fca46cb2b183724531b48f71ad upstream.

There is a small race between copy_process() and cgroup_attach_task()
where child->se.parent,cfs_rq points to invalid (old) ones.

        parent doing fork()      | someone moving the parent to another cgroup
      + dup_task_struct()
        -> parent->se is copied to child->se.
           se.parent,cfs_rq of them point to old ones.

                                       + cgroup_task_migrate()
                                         -> parent->cgroup is updated.
                                       + cpu_cgroup_attach()
                                         + sched_move_task()
                                           + task_move_group_fair()
                                             +- set_task_rq()
                                                -> se.parent,cfs_rq of parent
                                                   are updated.

      + cgroup_fork()
        -> parent->cgroup is copied to child->cgroup. (*1)
      + sched_fork()
        + task_fork_fair()
          -> se.parent,cfs_rq of child are accessed
             while they point to old ones. (*2)

In the worst case, this bug can lead to "use-after-free" and cause a panic,
because it's new cgroup's refcount that is incremented at (*1),
so the old cgroup(and related data) can be freed before (*2).

In fact, a panic caused by this bug was originally caught in RHEL6.4.

    BUG: unable to handle kernel NULL pointer dereference at (null)
    IP: [<ffffffff81051e3e>] sched_slice+0x6e/0xa0
    Call Trace:
     [<ffffffff81051f25>] place_entity+0x75/0xa0
     [<ffffffff81056a3a>] task_fork_fair+0xaa/0x160
     [<ffffffff81063c0b>] sched_fork+0x6b/0x140
     [<ffffffff8106c3c2>] copy_process+0x5b2/0x1450
     [<ffffffff81063b49>] ? wake_up_new_task+0xd9/0x130
     [<ffffffff8106d2f4>] do_fork+0x94/0x460
     [<ffffffff81072a9e>] ? sys_wait4+0xae/0x100
     [<ffffffff81009598>] sys_clone+0x28/0x30
     [<ffffffff8100b393>] stub_clone+0x13/0x20
     [<ffffffff8100b072>] ? system_call_fastpath+0x16/0x1b

Signed-off-by: Daisuke Nishimura <>
Signed-off-by: Peter Zijlstra <>
Signed-off-by: Ingo Molnar <>
[bwh: Backported to 3.2: adjust filename]
Signed-off-by: Ben Hutchings <>
8 years agomemcg: fix multiple large threshold notifications
Greg Thelen [Wed, 11 Sep 2013 21:23:08 +0000 (14:23 -0700)]
memcg: fix multiple large threshold notifications

commit 2bff24a3707093c435ab3241c47dcdb5f16e432b upstream.

A memory cgroup with (1) multiple threshold notifications and (2) at least
one threshold >=2G was not reliable.  Specifically the notifications would
either not fire or would not fire in the proper order.

The __mem_cgroup_threshold() signaling logic depends on keeping 64 bit
thresholds in sorted order.  mem_cgroup_usage_register_event() sorts them
with compare_thresholds(), which returns the difference of two 64 bit
thresholds as an int.  If the difference is positive but has bit[31] set,
then sort() treats the difference as negative and breaks sort order.

This fix compares the two arbitrary 64 bit thresholds returning the
classic -1, 0, 1 result.

The test below sets two notifications (at 0x1000 and 0x81001000):
  cd /sys/fs/cgroup/memory
  mkdir x
  for x in 4096 2164264960; do
    cgroup_event_listener x/memory.usage_in_bytes $x | sed "s/^/$x listener:/" &
  echo $$ > x/cgroup.procs
  anon_leaker 500M

v3.11-rc7 fails to signal the 4096 event listener:
  Done leaking pages.

Patched v3.11-rc7 properly notifies:
  4096 listener:2013:8:31:14:13:36
  Done leaking pages.

The fixed bug is old.  It appears to date back to the introduction of
memcg threshold notifications in v2.6.34-rc1-116-g2e72b6347c94 "memcg:
implement memory thresholds"

Signed-off-by: Greg Thelen <>
Acked-by: Michal Hocko <>
Acked-by: Kirill A. Shutemov <>
Acked-by: Johannes Weiner <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
8 years agomm/huge_memory.c: fix potential NULL pointer dereference
Libin [Wed, 11 Sep 2013 21:20:38 +0000 (14:20 -0700)]
mm/huge_memory.c: fix potential NULL pointer dereference

commit a8f531ebc33052642b4bd7b812eedf397108ce64 upstream.

In collapse_huge_page() there is a race window between releasing the
mmap_sem read lock and taking the mmap_sem write lock, so find_vma() may
return NULL.  So check the return value to avoid NULL pointer dereference.

vma = find_vma(mm, address)

Signed-off-by: Libin <>
Acked-by: Kirill A. Shutemov <>
Reviewed-by: Wanpeng Li <>
Reviewed-by: Michal Hocko <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
8 years agoocfs2: fix the end cluster offset of FIEMAP
Jie Liu [Wed, 11 Sep 2013 21:20:05 +0000 (14:20 -0700)]
ocfs2: fix the end cluster offset of FIEMAP

commit 28e8be31803b19d0d8f76216cb11b480b8a98bec upstream.

Call fiemap ioctl(2) with given start offset as well as an desired mapping
range should show extents if possible.  However, we somehow figure out the
end offset of mapping via 'mapping_end -= cpos' before iterating the
extent records which would cause problems if the given fiemap length is
too small to a cluster size, e.g,

Cluster size 4096:
debugfs.ocfs2 1.6.3
        Block Size Bits: 12   Cluster Size Bits: 12

The extended fiemap test utility From David:

# dd if=/dev/urandom of=/ocfs2/test_file bs=1M count=1000
# ./fiemap /ocfs2/test_file 4096 10
start: 4096, length: 10
File /ocfs2/test_file has 0 extents:
# Logical          Physical         Length           Flags
^^^^^ <-- No extent is shown

In this case, at ocfs2_fiemap(): cpos == mapping_end == 1. Hence the
loop of searching extent records was not executed at all.

This patch remove the in question 'mapping_end -= cpos', and loops
until the cpos is larger than the mapping_end as usual.

# ./fiemap /ocfs2/test_file 4096 10
start: 4096, length: 10
File /ocfs2/test_file has 1 extents:
# Logical          Physical         Length           Flags
0: 0000000000000000 0000000056a01000 0000000006a00000 0000

Signed-off-by: Jie Liu <>
Reported-by: David Weber <>
Tested-by: David Weber <>
Cc: Sunil Mushran <>
Cc: Mark Fashen <>
Cc: Joel Becker <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
8 years agosd: Fix potential out-of-bounds access
Alan Stern [Fri, 6 Sep 2013 15:49:51 +0000 (11:49 -0400)]
sd: Fix potential out-of-bounds access

commit 984f1733fcee3fbc78d47e26c5096921c5d9946a upstream.

This patch fixes an out-of-bounds error in sd_read_cache_type(), found
by Google's AddressSanitizer tool.  When the loop ends, we know that
"offset" lies beyond the end of the data in the buffer, so no Caching
mode page was found.  In theory it may be present, but the buffer size
is limited to 512 bytes.

Signed-off-by: Alan Stern <>
Reported-by: Dmitry Vyukov <>
Signed-off-by: James Bottomley <>
Signed-off-by: Ben Hutchings <>
8 years agodrm/radeon/atom: workaround vbios bug in transmitter table on rs880 (v2)
Alex Deucher [Mon, 9 Sep 2013 14:54:22 +0000 (10:54 -0400)]
drm/radeon/atom: workaround vbios bug in transmitter table on rs880 (v2)

commit 91f3a6aaf280294b07c05dfe606e6c27b7ba3c72 upstream.

The OUTPUT_ENABLE action jumps past the point in the coder where
the data_offset is set on certain rs780 cards.  This worked
previously because the OUTPUT_ENABLE action is always called
immediately after the ENABLE action so the data_offset remained
set.  In 6f8bbaf568c7f2c497558bfd04654c0b9841ad57
(drm/radeon/atom: initialize more atom interpretor elements to 0),
we explictly reset data_offset to 0 between atom calls which then
caused this to fail.  The fix is to just skip calling the
OUTPUT_ENABLE action on the problematic chipsets.  The ENABLE
action does the same thing and more.  Ultimately, we could
probably drop the OUTPUT_ENABLE action all together on DCE3


v2: only rs880 seems to be affected

Signed-off-by: Alex Deucher <>
Signed-off-by: Ben Hutchings <>
8 years agoALSA: hda - Add Toshiba Satellite C870 to MSI blacklist
Takashi Iwai [Mon, 9 Sep 2013 08:20:48 +0000 (10:20 +0200)]
ALSA: hda - Add Toshiba Satellite C870 to MSI blacklist

commit 83f72151352791836a1b9c1542614cc9bf71ac61 upstream.

Toshiba Satellite C870 shows interrupt problems occasionally when
certain mixer controls like "Mic Switch" is toggled.  This seems
worked around by not using MSI.

Signed-off-by: Takashi Iwai <>
Signed-off-by: Ben Hutchings <>
8 years agocrypto: api - Fix race condition in larval lookup
Herbert Xu [Sun, 8 Sep 2013 04:33:50 +0000 (14:33 +1000)]
crypto: api - Fix race condition in larval lookup

commit 77dbd7a95e4a4f15264c333a9e9ab97ee27dc2aa upstream.

crypto_larval_lookup should only return a larval if it created one.
Any larval created by another entity must be processed through
crypto_larval_wait before being returned.

Otherwise this will lead to a larval being killed twice, which
will most likely lead to a crash.

Reported-by: Kees Cook <>
Tested-by: Kees Cook <>
Signed-off-by: Herbert Xu <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>