The definition of atomic_dec_if_positive() assumes that
atomic_sub_if_positive() exists, which is only the case if
metag specific atomics are used. This results in the following
build error when trying to build metag1_defconfig.
kernel/ucount.c: In function 'dec_ucount':
kernel/ucount.c:211: error:
implicit declaration of function 'atomic_sub_if_positive'
Moving the definition of atomic_dec_if_positive() into the metag
conditional code fixes the problem.
Fixes: 6006c0d8ce94 ("metag: Atomics, locks and bitops") Signed-off-by: Guenter Roeck <linux@roeck-us.net> Signed-off-by: James Hogan <james.hogan@imgtec.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
When using efifb with a 16-bit (5:6:5) visual, fbcon's text is rendered
in the wrong colors - e.g. text gray (#aaaaaa) is rendered as green
(#50bc50) and neighboring pixels have slightly different values
(such as #50bc78).
The reason is that fbcon loads its 16 color palette through
efifb_setcolreg(), which in turn calculates a 32-bit value to write
into memory for each palette index.
Until now, this code could only handle 8-bit visuals and didn't mask
overlapping values when ORing them.
With this patch, fbcon displays the correct colors when a qemu VM is
booted in 16-bit mode (in GRUB: "set gfxpayload=800x600x16").
Fixes: 7c83172b98e5 ("x86_64 EFI boot support: EFI frame buffer driver") # v2.6.24+ Signed-off-by: Max Staudt <mstaudt@suse.de> Acked-By: Peter Jones <pjones@redhat.com> Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
http://www.ti.com/lit/pdf/SWCZ010:
DCDC o/p voltage can go higher than programmed value
Impact:
VDDI, VDD2, and VIO output programmed voltage level can go higher than
expected or crash, when coming out of PFM to PWM mode or using DVFS.
Description:
When DCDC CLK SYNC bits are 11/01:
* VIO 3-MHz oscillator is the source clock of the digital core and input
clock of VDD1 and VDD2
* Turn-on of VDD1 and VDD2 HSD PFETis synchronized or at a constant
phase shift
* Current pulled though VCC1+VCC2 is Iload(VDD1) + Iload(VDD2)
* The 3 HSD PFET will be turned-on at the same time, causing the highest
possible switching noise on the application. This noise level depends
on the layout, the VBAT level, and the load current. The noise level
increases with improper layout.
When DCDC CLK SYNC bits are 00:
* VIO 3-MHz oscillator is the source clock of digital core
* VDD1 and VDD2 are running on their own 3-MHz oscillator
* Current pulled though VCC1+VCC2 average of Iload(VDD1) + Iload(VDD2)
* The switching noise of the 3 SMPS will be randomly spread over time,
causing lower overall switching noise.
Workaround:
Set DCDCCTRL_REG[1:0]= 00.
Signed-off-by: Jan Remmet <j.remmet@phytec.de> Signed-off-by: Mark Brown <broonie@kernel.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Fixes: ddd17531ad908 ("ASoC: omap-mcpdm: Clean up with devm_* function")
Managed irq request will not doing any good in ASoC probe level as it is
not going to free up the irq when the driver is unbound from the sound
card.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reported-by: Russell King <linux@armlinux.org.uk> Signed-off-by: Mark Brown <broonie@kernel.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
set_bit() and clear_bit() take the bit number so this code is really
doing "1 << (1 << irq)" which is a double shift bug. It's done
consistently so it won't cause a problem unless "irq" is more than 4.
Fixes: 70c6cce04066 ('mfd: Support 88pm80x in 80x driver') Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Lee Jones <lee.jones@linaro.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
If mpi_powm() is given a zero exponent, it wants to immediately return
either 1 or 0, depending on the modulus. However, if the result was
initalised with zero limb space, no limbs space is allocated and a
NULL-pointer exception ensues.
Fix this by allocating a minimal amount of limb space for the result when
the 0-exponent case when the result is 1 and not touching the limb space
when the result is 0.
This affects the use of RSA keys and X.509 certificates that carry them.
According to the datasheet you should only write 1 to this bit. If it is
not set, at least AIN3 will return bad values on newer silicon revisions.
Fixes: d84ca5b345c2 ("hwmon: Add driver for ADT7411 voltage and temperature sensor") Signed-off-by: Michael Walle <michael@walle.cc> Signed-off-by: Guenter Roeck <linux@roeck-us.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
A timer was used to restart after the bus-off state, leading to a
relatively large can_restart() executed in an interrupt context,
which in turn sets up pinctrl. When this happens during system boot,
there is a high probability of grabbing the pinctrl_list_mutex,
which is locked already by the probe() of other device, making the
kernel suspect a deadlock condition [1].
To resolve this issue, the restart_timer is replaced by a delayed
work.
The oom victim cannot terminate because it needs to take mmap_sem for
write while the lock is held by ksmd for read which loops in the page
allocator
ksm_do_scan
scan_get_next_rmap_item
down_read
get_next_rmap_item
alloc_rmap_item #ksmd will loop permanently.
There is no way forward because the oom victim cannot release any memory
in 4.1 based kernel. Since 4.6 we have the oom reaper which would solve
this problem because it would release the memory asynchronously.
Nevertheless we can relax alloc_rmap_item requirements and use
__GFP_NORETRY because the allocation failure is acceptable as ksm_do_scan
would just retry later after the lock got dropped.
Such a patch would be also easy to backport to older stable kernels which
do not have oom_reaper.
While we are at it add GFP_NOWARN so the admin doesn't have to be alarmed
by the allocation failure.
v4.8-rc3 commit 99f3c90d0d ("dm flakey: error READ bios during the
down_interval") overlooked the 'drop_writes' feature, which is meant to
allow reads to be issued rather than errored, during the down_interval.
Fixes: 99f3c90d0d ("dm flakey: error READ bios during the down_interval") Reported-by: Qu Wenruo <quwenruo@cn.fujitsu.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
For large values of "mult" and long uptimes, the intermediate
result of "cycles * mult" can overflow 64 bits. For example,
the tile platform calls clocksource_cyc2ns with a 1.2 GHz clock;
we have mult = 853, and after 208.5 days, we overflow 64 bits.
Since clocksource_cyc2ns() is intended to be used for relative
cycle counts, not absolute cycle counts, performance is more
importance than accepting a wider range of cycle values. So,
just use mult_frac() directly in tile's sched_clock().
Commit 4cecf6d401a0 ("sched, x86: Avoid unnecessary overflow
in sched_clock") by Salman Qazi results in essentially the same
generated code for x86 as this change does for tile. In fact,
a follow-on change by Salman introduced mult_frac() and switched
to using it, so the C code was largely identical at that point too.
Peter Zijlstra then added mul_u64_u32_shr() and switched x86
to use it. This is, in principle, better; by optimizing the
64x64->64 multiplies to be 32x32->64 multiplies we can potentially
save some time. However, the compiler piplines the 64x64->64
multiplies pretty well, and the conditional branch in the generic
mul_u64_u32_shr() causes some bubbles in execution, with the
result that it's pretty much a wash. If tilegx provided its own
implementation of mul_u64_u32_shr() without the conditional branch,
we could potentially save 3 cycles, but that seems like small gain
for a fair amount of additional build scaffolding; no other platform
currently provides a mul_u64_u32_shr() override, and tile doesn't
currently have an <asm/div64.h> header to put the override in.
Additionally, gcc currently has an optimization bug that prevents
it from recognizing the opportunity to use a 32x32->64 multiply,
and so the result would be no better than the existing mult_frac()
until such time as the compiler is fixed.
For now, just using mult_frac() seems like the right answer.
Signed-off-by: Chris Metcalf <cmetcalf@mellanox.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
Some AMD CS553x devices have read-only BARs because of a firmware or
hardware defect. There's a workaround in quirk_cs5536_vsa(), but it no
longer works after 36e8164882ca ("PCI: Restore detection of read-only
BARs"). Prior to 36e8164882ca, we filled in res->start; afterwards we
leave it zeroed out. The quirk only updated the size, so the driver tried
to use a region starting at zero, which didn't work.
Expand quirk_cs5536_vsa() to read the base addresses from the BARs and
hard-code the sizes.
On Nix's system BAR 2's read-only value is 0x6200. Prior to 36e8164882ca,
we interpret that as a 512-byte BAR based on the lowest-order bit set. Per
datasheet sec 5.6.1, that BAR (MFGPT) requires only 64 bytes; use that to
avoid clearing any address bits if a platform uses only 64-byte alignment.
[js] pcibios_bus_to_resource takes pdev, not bus in 3.12
NFC version reply size checked against only header size, not against
full message size. That may lead potentially to uninitialized memory access
in version data.
That leads to warnings when version data is accessed:
drivers/misc/mei/bus-fixup.c: warning: '*((void *)&ver+11)' may be used uninitialized in this function [-Wuninitialized]: => 212:2
Reported in
Build regressions/improvements in v4.9-rc3
https://lkml.org/lkml/2016/10/30/57
The ad5933_i2c_read function returns an error code to indicate
whether it could read data or not. However ad5933_work() ignores
this return code and just accesses the data unconditionally,
which gets detected by gcc as a possible bug:
drivers/staging/iio/impedance-analyzer/ad5933.c: In function 'ad5933_work':
drivers/staging/iio/impedance-analyzer/ad5933.c:649:16: warning: 'status' may be used uninitialized in this function [-Wmaybe-uninitialized]
This adds minimal error handling so we only evaluate the
data if it was correctly read.
The host keeps sending heartbeat packets independent of the
guest responding to them. Even though we respond to the heartbeat messages at
interrupt level, we can have situations where there maybe multiple heartbeat
messages pending that have not been responded to. For instance this occurs when the
VM is paused and the host continues to send the heartbeat messages.
Address this issue by draining and responding to all
the heartbeat messages that maybe pending.
Signed-off-by: Long Li <longli@microsoft.com> Signed-off-by: K. Y. Srinivasan <kys@microsoft.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
Fix a short sprintf buffer in proc_keys_show(). If the gcc stack protector
is turned on, this can cause a panic due to stack corruption.
The problem is that xbuf[] is not big enough to hold a 64-bit timeout
rendered as weeks:
(gdb) p 0xffffffffffffffffULL/(60*60*24*7)
$2 = 30500568904943
That's 14 chars plus NUL, not 11 chars plus NUL.
Expand the buffer to 16 chars.
I think the unpatched code apparently works if the stack-protector is not
enabled because on a 32-bit machine the buffer won't be overflowed and on a
64-bit machine there's a 64-bit aligned pointer at one side and an int that
isn't checked again on the other side.
Reported-by: Ondrej Kozina <okozina@redhat.com> Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Ondrej Kozina <okozina@redhat.com> Signed-off-by: James Morris <james.l.morris@oracle.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
The variable i contains a total number of resources (including
IORESOURCE_IRQ). However, we want the dmem_region_start to point
after the last resource of type IORESOURCE_MEM. The original behaviour
leads (very likely) to skipping several UIO mapping regions and makes
them useless. Fix this by computing dmem_region_start from the uiomem
which points to the last used UIO mapping.
Fixes: 0a0c3b5a24bd ("Add new uio device for dynamic memory allocation") Signed-off-by: Jan Viktorin <viktorin@rehivetech.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
From the beginning of the gpio-mpc8xxx.c, the "handle_level_irq"
has being used to handle GPIO interrupts in the PowerPC/Layerscape
platforms. But actually, almost all PowerPC/Layerscape platforms
assert an interrupt request upon either a high-to-low change or
any change on the state of the signal.
So the "handle_level_irq" is not reasonable for PowerPC/Layerscape
GPIO interrupt, it should be "handle_edge_irq". Otherwise the system
may lost some interrupts from the PIN's state changes.
Signed-off-by: Liu Gang <Gang.Liu@nxp.com> Signed-off-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
The cx231xx_set_agc_analog_digital_mux_select() callers
expect it to return 0 or an error. Returning a positive value
makes the first attempt to switch between analog/digital to fail.
With the current settings, only one channel locks properly.
That's likely because, when this driver was written, Brazil
were still using experimental transmissions.
Change it to reproduce the settings used by the newer drivers.
That makes it lock on other channels.
Tested with both PixelView SBTVD Hybrid (cx231xx-based) and
C3Tech Digital Duo HDTV/SDTV (em28xx-based) devices.
On this frontend, it takes a while to start output normal
TS data. That only happens on state S9. On S8, the TS output
is enabled, but it is not reliable enough.
However, the zigzag loop is too fast to let it sync.
As, on practical tests, the zigzag software loop doesn't
seem to be helping, but just slowing down the tuning, let's
switch to hardware algorithm, as the tuners used on such
devices are capable of work with frequency drifts without
any help from software.
The ramoops buffer may be mapped as either I/O memory or uncached
memory. On ARM64, this results in a device-type (strongly-ordered)
mapping. Since unnaligned accesses to device-type memory will
generate an alignment fault (regardless of whether or not strict
alignment checking is enabled), it is not safe to use memcpy().
memcpy_fromio() is guaranteed to only use aligned accesses, so use
that instead.
persistent_ram_update uses vmap / iomap based on whether the buffer is in
memory region or reserved region. However, both map it as non-cacheable
memory. For armv8 specifically, non-cacheable mapping requests use a
memory type that has to be accessed aligned to the request size. memcpy()
doesn't guarantee that.
I have here a FPGA behind PCIe which exports SRAM which I use for
pstore. Now it seems that the FPGA no longer supports cmpxchg based
updates and writes back 0xffâ\80¦ff and returns the same. This leads to
crash during crash rendering pstore useless.
Since I doubt that there is much benefit from using cmpxchg() here, I am
dropping this atomic access and use the spinlock based version.
Cc: Anton Vorontsov <anton@enomsg.org> Cc: Colin Cross <ccross@android.com> Cc: Kees Cook <keescook@chromium.org> Cc: Tony Luck <tony.luck@intel.com> Cc: Rabin Vincent <rabinv@axis.com> Tested-by: Rabin Vincent <rabinv@axis.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Reviewed-by: Guenter Roeck <linux@roeck-us.net>
[kees: remove "_locked" suffix since it's the only option now] Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
CMD23 aka SET_BLOCK_COUNT was introduced with MMC v3.1.
Older versions of the specification allowed to terminate
multi-block transfers only with CMD12.
The patch fixes the following problem:
mmc0: new MMC card at address 0001
mmcblk0: mmc0:0001 SDMB-16 15.3 MiB
mmcblk0: timed out sending SET_BLOCK_COUNT command, card status 0x400900
...
blk_update_request: I/O error, dev mmcblk0, sector 0
Buffer I/O error on dev mmcblk0, logical block 0, async page read
mmcblk0: unable to read partition table
An interrupt may occur right after devm_request_irq() is called and
prior to the spinlock initialization, leading to a kernel oops,
as the interrupt handler uses the spinlock.
In order to prevent this problem, move the spinlock initialization
prior to requesting the interrupts.
Make sure to drop the reference taken by class_find_device() after
opening the RTC device.
Fixes: 77437fd4e61f (pm: boot time suspend selftest) Signed-off-by: Johan Hovold <johan@kernel.org> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
Make sure to drop the reference taken by bus_find_device_by_name()
before returning from mfd_clone_cell().
Fixes: a9bbba996302 ("mfd: add platform_device sharing support for mfd") Signed-off-by: Johan Hovold <johan@kernel.org> Signed-off-by: Lee Jones <lee.jones@linaro.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
rs->begin in ratelimit is set in two cases.
1) when rs->begin was not initialized
2) when rs->interval was passed
For case #2, current ratelimit sets the begin to 0. This incurrs
improper suppression. The begin value will be set in the next ratelimit
call by 1). Then the time interval check will be always false, and
rs->printed will not be initialized. Although enough time passed,
ratelimit may return 0 if rs->printed is not less than rs->burst. To
reset interval properly, begin should be jiffies rather than 0.
For an example code below:
static DEFINE_RATELIMIT_STATE(mylimit, 1, 1);
for (i = 1; i <= 10; i++) {
if (__ratelimit(&mylimit))
printk("ratelimit test count %d\n", i);
msleep(3000);
}
test result in the current code shows suppression even there is 3 seconds sleep.
[ 78.391148] ratelimit test count 1
[ 81.295988] ratelimit test count 2
[ 87.315981] ratelimit test count 4
[ 93.336267] ratelimit test count 6
[ 99.356031] ratelimit test count 8
[ 105.376367] ratelimit test count 10
Signed-off-by: Jaewon Kim <jaewon31.kim@samsung.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
It turns out that the rcuos callback-offload kthread is busy processing
a very large quantity of RCU callbacks, and it is not reliquishing the
CPU while doing so. This commit therefore adds an cond_resched_rcu_qs()
within the loop to allow other tasks to run.
[js] use onlu cond_resched() in 3.12
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
[ paulmck: Substituted cond_resched_rcu_qs for cond_resched. ] Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Dhaval Giani <dhaval.giani@oracle.com> Signed-off-by: Jiri Slaby <jslaby@suse.cz> Signed-off-by: Willy Tarreau <w@1wt.eu>
gen_pool_alloc_algo() iterates over the chunks of a pool trying to find
a contiguous block of memory that satisfies the allocation request.
The shortcut
if (size > atomic_read(&chunk->avail))
continue;
makes the loop skip over chunks that do not have enough bytes left to
fulfill the request. There are two situations, though, where an
allocation might still fail:
(1) The available memory is not contiguous, i.e. the request cannot
be fulfilled due to external fragmentation.
(2) A race condition. Another thread runs the same code concurrently
and is quicker to grab the available memory.
In those situations, the loop calls pool->algo() to search the entire
chunk, and pool->algo() returns some value that is >= end_bit to
indicate that the search failed. This return value is then assigned to
start_bit. The variables start_bit and end_bit describe the range that
should be searched, and this range should be reset for every chunk that
is searched. Today, the code fails to reset start_bit to 0. As a
result, prefixes of subsequent chunks are ignored. Memory allocations
might fail even though there is plenty of room left in these prefixes of
those other chunks.
Fixes: 7f184275aa30 ("lib, Make gen_pool memory allocator lockless") Link: http://lkml.kernel.org/r/1477420604-28918-1-git-send-email-danielmentz@google.com Signed-off-by: Daniel Mentz <danielmentz@google.com> Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Acked-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
Don't pass a size larger than iov_len to kernel_sendmsg().
Otherwise it will cause a NULL pointer deref when kernel_sendmsg()
returns with rv < size.
DRBD as external module has been around in the kernel 2.4 days already.
We used to be compatible to 2.4 and very early 2.6 kernels,
we used to use
rv = sock_sendmsg(sock, &msg, iov.iov_len);
then later changed to
rv = kernel_sendmsg(sock, &msg, &iov, 1, size);
when we should have used
rv = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
tcp_sendmsg() used to totally ignore the size parameter. 57be5bd ip: convert tcp_sendmsg() to iov_iter primitives
changes that, and exposes our long standing error.
Even with this error exposed, to trigger the bug, we would need to have
an environment (config or otherwise) causing us to not use sendpage()
for larger transfers, a failing connection, and have it fail "just at the
right time". Apparently that was unlikely enough for most, so this went
unnoticed for years.
Still, it is known to trigger at least some of these,
and suspected for the others:
[0] http://lists.linbit.com/pipermail/drbd-user/2016-July/023112.html
[1] http://lists.linbit.com/pipermail/drbd-dev/2016-March/003362.html
[2] https://forums.grsecurity.net/viewtopic.php?f=3&t=4546
[3] https://ubuntuforums.org/showthread.php?t=2336150
[4] http://e2.howsolveproblem.com/i/1175162/
This should go into 4.9,
and into all stable branches since and including v4.0,
which is the first to contain the exposing change.
It is correct for all stable branches older than that as well
(which contain the DRBD driver; which is 2.6.33 and up).
It requires a small "conflict" resolution for v4.4 and earlier, with v4.5
we dropped the comment block immediately preceding the kernel_sendmsg().
Fixes: b411b3637fa7 ("The DRBD driver") Cc: viro@zeniv.linux.org.uk Cc: christoph.lechleitner@iteg.at Cc: wolfgang.glas@iteg.at Reported-by: Christoph Lechleitner <christoph.lechleitner@iteg.at> Tested-by: Christoph Lechleitner <christoph.lechleitner@iteg.at> Signed-off-by: Richard Weinberger <richard@nod.at>
[changed oneliner to be "obvious" without context; more verbose message] Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Jens Axboe <axboe@fb.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
While debugging timeouts happening in my application workload (ScyllaDB), I have
observed calls to open() taking a long time, ranging everywhere from 2 seconds -
the first ones that are enough to time out my application - to more than 30
seconds.
The problem seems to happen because XFS may block on pending metadata updates
under certain circumnstances, and that's confirmed with the following backtrace
taken by the offcputime tool (iovisor/bcc):
Inspecting my run with blktrace, I can see that the xfsaild kthread exhibit very
high "Dispatch wait" times, on the dozens of seconds range and consistent with
the open() times I have saw in that run.
Still from the blktrace output, we can after searching a bit, identify the
request that wasn't dispatched:
8,0 11 152 81.092472813 804 A WM 141698288 + 8 <- (8,1) 141696240
8,0 11 153 81.092472889 804 Q WM 141698288 + 8 [xfsaild/sda1]
8,0 11 154 81.092473207 804 G WM 141698288 + 8 [xfsaild/sda1]
8,0 11 206 81.092496118 804 I WM 141698288 + 8 ( 22911) [xfsaild/sda1]
<==== 'I' means Inserted (into the IO scheduler) ===================================>
8,0 0 289372 96.718761435 0 D WM 141698288 + 8 (15626265317) [swapper/0]
<==== Only 15s later the CFQ scheduler dispatches the request ======================>
As we can see above, in this particular example CFQ took 15 seconds to dispatch
this request. Going back to the full trace, we can see that the xfsaild queue
had plenty of opportunity to run, and it was selected as the active queue many
times. It would just always be preempted by something else (example):
8,0 1 0 81.117912979 0 m N cfq1618SN / insert_request
8,0 1 0 81.117913419 0 m N cfq1618SN / add_to_rr
8,0 1 0 81.117914044 0 m N cfq1618SN / preempt
8,0 1 0 81.117914398 0 m N cfq767A / slice expired t=1
8,0 1 0 81.117914755 0 m N cfq767A / resid=40
8,0 1 0 81.117915340 0 m N / served: vt=1948520448 min_vt=1948520448
8,0 1 0 81.117915858 0 m N cfq767A / sl_used=1 disp=0 charge=0 iops=1 sect=0
where cfq767 is the xfsaild queue and cfq1618 corresponds to one of the ScyllaDB
IO dispatchers.
The requests preempting the xfsaild queue are synchronous requests. That's a
characteristic of ScyllaDB workloads, as we only ever issue O_DIRECT requests.
While it can be argued that preempting ASYNC requests in favor of SYNC is part
of the CFQ logic, I don't believe that doing so for 15+ seconds is anyone's
goal.
Moreover, unless I am misunderstanding something, that breaks the expectation
set by the "fifo_expire_async" tunable, which in my system is set to the
default.
Looking at the code, it seems to me that the issue is that after we make
an async queue active, there is no guarantee that it will execute any request.
When the queue itself tests if it cfq_may_dispatch() it can bail if it sees SYNC
requests in flight. An incoming request from another queue can also preempt it
in such situation before we have the chance to execute anything (as seen in the
trace above).
This patch sets the must_dispatch flag if we notice that we have requests
that are already fifo_expired. This flag is always cleared after
cfq_dispatch_request() returns from cfq_dispatch_requests(), so it won't pin
the queue for subsequent requests (unless they are themselves expired)
Care is taken during preempt to still allow rt requests to preempt us
regardless.
Testing my workload with this patch applied produces much better results.
From the application side I see no timeouts, and the open() latency histogram
generated by systemtap looks much better, with the worst outlier at 131ms:
As suggested in commit 5864a2fd3088db73d47942370d0f7210a807b9bc
(ipc/sem.c: fix complex_count vs. simple op race) since it introduces
a regression and the candidate fix requires too many changes for 3.10.
Commit fec1d0115240 ("[PATCH] Disable CLONE_CHILD_CLEARTID for abnormal
exit") has caused a subtle regression in nscd which uses
CLONE_CHILD_CLEARTID to clear the nscd_certainly_running flag in the
shared databases, so that the clients are notified when nscd is
restarted. Now, when nscd uses a non-persistent database, clients that
have it mapped keep thinking the database is being updated by nscd, when
in fact nscd has created a new (anonymous) one (for non-persistent
databases it uses an unlinked file as backend).
The original proposal for the CLONE_CHILD_CLEARTID change claimed
(https://lkml.org/lkml/2006/10/25/233):
: The NPTL library uses the CLONE_CHILD_CLEARTID flag on clone() syscalls
: on behalf of pthread_create() library calls. This feature is used to
: request that the kernel clear the thread-id in user space (at an address
: provided in the syscall) when the thread disassociates itself from the
: address space, which is done in mm_release().
:
: Unfortunately, when a multi-threaded process incurs a core dump (such as
: from a SIGSEGV), the core-dumping thread sends SIGKILL signals to all of
: the other threads, which then proceed to clear their user-space tids
: before synchronizing in exit_mm() with the start of core dumping. This
: misrepresents the state of process's address space at the time of the
: SIGSEGV and makes it more difficult for someone to debug NPTL and glibc
: problems (misleading him/her to conclude that the threads had gone away
: before the fault).
:
: The fix below is to simply avoid the CLONE_CHILD_CLEARTID action if a
: core dump has been initiated.
The resulting patch from Roland (https://lkml.org/lkml/2006/10/26/269)
seems to have a larger scope than the original patch asked for. It
seems that limitting the scope of the check to core dumping should work
for SIGSEGV issue describe above.
[Changelog partly based on Andreas' description] Fixes: fec1d0115240 ("[PATCH] Disable CLONE_CHILD_CLEARTID for abnormal exit") Link: http://lkml.kernel.org/r/1471968749-26173-1-git-send-email-mhocko@kernel.org Signed-off-by: Michal Hocko <mhocko@suse.com> Tested-by: William Preston <wpreston@suse.com> Acked-by: Oleg Nesterov <oleg@redhat.com> Cc: Roland McGrath <roland@hack.frob.com> Cc: Andreas Schwab <schwab@suse.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
The iter->seq can be reset outside the protection of the mutex. So can
reading of user data. Move the mutex up to the beginning of the function.
Fixes: d7350c3f45694 ("tracing/core: make the read callbacks reentrants") Reported-by: Al Viro <viro@ZenIV.linux.org.uk> Signed-off-by: Steven Rostedt <rostedt@goodmis.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
This is caused by the sky2 being called after it has been shutdown.
A previous thread about this can be found here:
https://lkml.org/lkml/2016/4/12/410
An alternative fix is to assure that IFF_UP gets cleared by
calling dev_close() during shutdown. This is similar to what the
bnx2/tg3/xgene and maybe others are doing to assure that the driver
isn't being called following _shutdown().
Signed-off-by: Jeremy Linton <jeremy.linton@arm.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
where skb_gso_segment() relies on skb->protocol to function properly.
This patch sets skb->protocol to ETH_P_IP before dst_output() is called,
fixing a bug where GSO packets sent through a sit tunnel are dropped
when xfrm is involved.
Signed-off-by: Eli Cooper <elicooper@gmx.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
It's possible to make scanning consume almost arbitrary amounts
of memory, e.g. by sending beacon frames with random BSSIDs at
high rates while somebody is scanning.
Limit the number of BSS table entries we're willing to cache to
1000, limiting maximum memory usage to maybe 4-5MB, but lower
in practice - that would be the case for having both full-sized
beacon and probe response frames for each entry; this seems not
possible in practice, so a limit of 1000 entries will likely be
closer to 0.5 MB.
Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
In mac80211, multicast A-MSDUs are accepted in many cases that
they shouldn't be accepted in:
* drop A-MSDUs with a multicast A1 (RA), as required by the
spec in 9.11 (802.11-2012 version)
* drop A-MSDUs with a 4-addr header, since the fourth address
can't actually be useful for them; unless 4-address frame
format is actually requested, even though the fourth address
is still not useful in this case, but ignored
Accepting the first case, in particular, is very problematic
since it allows anyone else with possession of a GTK to send
unicast frames encapsulated in a multicast A-MSDU, even when
the AP has client isolation enabled.
Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
The code currently assumes that buffered multicast PS frames don't have
a pending ACK frame for tx status reporting.
However, hostapd sends a broadcast deauth frame on teardown for which tx
status is requested. This can lead to the "Have pending ack frames"
warning on module reload.
Fix this by using ieee80211_free_txskb/ieee80211_purge_tx_queue.
Signed-off-by: Felix Fietkau <nbd@nbd.name> Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
In v2.6, ip_rt_redirect() calls arp_bind_neighbour() which returns 0
and then the state of the neigh for the new_gw is checked. If the state
isn't valid then the redirected route is deleted. This behavior is
maintained up to v3.5.7 by check_peer_redirect() because rt->rt_gateway
is assigned to peer->redirect_learned.a4 before calling
ipv4_neigh_lookup().
After commit 5943634fc559 ("ipv4: Maintain redirect and PMTU info in
struct rtable again."), ipv4_neigh_lookup() is performed without the
rt_gateway assigned to the new_gw. In the case when rt_gateway (old_gw)
isn't zero, the function uses it as the key. The neigh is most likely
valid since the old_gw is the one that sends the ICMP redirect message.
Then the new_gw is assigned to fib_nh_exception. The problem is: the
new_gw ARP may never gets resolved and the traffic is blackholed.
So, use the new_gw for neigh lookup.
Changes from v1:
- use __ipv4_neigh_lookup instead (per Eric Dumazet).
Fixes: 5943634fc559 ("ipv4: Maintain redirect and PMTU info in struct rtable again.") Signed-off-by: Stephen Suryaputra Lin <ssurya@ieee.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Fixes: commit f187bc6efb7250afee0e2009b6106 ("ipv4: No need to set generic neighbour pointer") Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
sctp_wait_for_connect() currently already holds the asoc to keep it
alive during the sleep, in case another thread release it. But Andrey
Konovalov and Dmitry Vyukov reported an use-after-free in such
situation.
Problem is that __sctp_connect() doesn't get a ref on the asoc and will
do a read on the asoc after calling sctp_wait_for_connect(), but by then
another thread may have closed it and the _put on sctp_wait_for_connect
will actually release it, causing the use-after-free.
Fix is, instead of doing the read after waiting for the connect, do it
before so, and avoid this issue as the socket is still locked by then.
There should be no issue on returning the asoc id in case of failure as
the application shouldn't trust on that number in such situations
anyway.
This issue doesn't exist in sctp_sendmsg() path.
Reported-by: Dmitry Vyukov <dvyukov@google.com> Reported-by: Andrey Konovalov <andreyknvl@google.com> Tested-by: Andrey Konovalov <andreyknvl@google.com> Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com> Reviewed-by: Xin Long <lucien.xin@gmail.com> Acked-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
dccp_v4_err() does not use pskb_may_pull() and might access garbage.
We only need 4 bytes at the beginning of the DCCP header, like TCP,
so the 8 bytes pulled in icmp_socket_deliver() are more than enough.
This patch might allow to process more ICMP messages, as some routers
are still limiting the size of reflected bytes to 28 (RFC 792), instead
of extended lengths (RFC 1812 4.3.2.3)
Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Sending zero checksum is ok for TCP, but not for UDP.
UDPv6 receiver should by default drop a frame with a 0 checksum,
and UDPv4 would not verify the checksum and might accept a corrupted
packet.
Simply replace such checksum by 0xffff, regardless of transport.
This error was caught on SIT tunnels, but seems generic.
Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Maciej Żenczykowski <maze@google.com> Cc: Willem de Bruijn <willemb@google.com> Acked-by: Maciej Żenczykowski <maze@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
At accept() time, it is possible the parent has a non zero
sk_err_soft, leftover from a prior error.
Make sure we do not leave this value in the child, as it
makes future getsockopt(SO_ERROR) calls quite unreliable.
Signed-off-by: Eric Dumazet <edumazet@google.com> Acked-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Andrey Konovalov reported that KASAN detected that SCTP was using a slab
beyond the boundaries. It was caused because when handling out of the
blue packets in function sctp_sf_ootb() it was checking the chunk len
only after already processing the first chunk, validating only for the
2nd and subsequent ones.
The fix is to just move the check upwards so it's also validated for the
1st chunk.
Reported-by: Andrey Konovalov <andreyknvl@google.com> Tested-by: Andrey Konovalov <andreyknvl@google.com> Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com> Reviewed-by: Xin Long <lucien.xin@gmail.com> Acked-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Most of getsockopt handlers in net/sctp/socket.c check len against
sizeof some structure like:
if (len < sizeof(int))
return -EINVAL;
On the first look, the check seems to be correct. But since len is int
and sizeof returns size_t, int gets promoted to unsigned size_t too. So
the test returns false for negative lengths. Yes, (-1 < sizeof(long)) is
false.
Fix this in sctp by explicitly checking len < 0 before any getsockopt
handler is called.
Note that sctp_getsockopt_events already handled the negative case.
Since we added the < 0 check elsewhere, this one can be removed.
If a socket has FANOUT sockopt set, a new proto_hook is registered
as part of fanout_add(). When processing a NETDEV_UNREGISTER event in
af_packet, __fanout_unlink is called for all sockets, but prot_hook which was
registered as part of fanout_add is not removed. Call fanout_release, on a
NETDEV_UNREGISTER, which removes prot_hook and removes fanout from the
fanout_list.
This fixes BUG_ON(!list_empty(&dev->ptype_specific)) in netdev_run_todo()
Signed-off-by: Anoob Soman <anoob.soman@citrix.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Since the commit below the ipmr/ip6mr rtnl_unicast() code uses the portid
instead of the previous dst_pid which was copied from in_skb's portid.
Since the skb is new the portid is 0 at that point so the packets are sent
to the kernel and we get scheduling while atomic or a deadlock (depending
on where it happens) by trying to acquire rtnl two times.
Also since this is RTM_GETROUTE, it can be triggered by a normal user.
Fixes: 2942e9005056 ("[RTNETLINK]: Use rtnl_unicast() for rtnetlink unicasts") Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
sk_mem_reclaim_partial() goal is to ensure each socket has
one SK_MEM_QUANTUM forward allocation. This is needed both for
performance and better handling of memory pressure situations in
follow up patches.
SK_MEM_QUANTUM is currently a page, but might be reduced to 4096 bytes
as some arches have 64KB pages.
Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Andrey Konovalov reported an issue with proc_register in bcm.c.
As suggested by Cong Wang this patch adds a lock_sock() protection and
a check for unsuccessful proc_create_data() in bcm_connect().
nf_log_proc_dostring() used current's network namespace instead of the one
corresponding to the sysctl file the write was performed on. Because the
permission check happens at open time and the nf_log files in namespaces
are accessible for the namespace owner, this can be abused by an
unprivileged user to effectively write to the init namespace's nf_log
sysctls.
Stash the "struct net *" in extra2 - data and extra1 are already used.
uid_t outer_uid;
gid_t outer_gid;
int stolen_fd = -1;
void writefile(char *path, char *buf) {
int fd = open(path, O_WRONLY);
if (fd == -1)
err(1, "unable to open thing");
if (write(fd, buf, strlen(buf)) != strlen(buf))
err(1, "unable to write thing");
close(fd);
}
int child_fn(void *p_) {
if (mount("proc", "/proc", "proc", MS_NOSUID|MS_NODEV|MS_NOEXEC,
NULL))
err(1, "mount");
/* Yes, we need to set the maps for the net sysctls to recognize us
* as namespace root.
*/
char buf[1000];
sprintf(buf, "0 %d 1\n", (int)outer_uid);
writefile("/proc/1/uid_map", buf);
writefile("/proc/1/setgroups", "deny");
sprintf(buf, "0 %d 1\n", (int)outer_gid);
writefile("/proc/1/gid_map", buf);
RFC 2734 defines the datagram_size field in fragment encapsulation
headers thus:
datagram_size: The encoded size of the entire IP datagram. The
value of datagram_size [...] SHALL be one less than the value of
Total Length in the datagram's IP header (see STD 5, RFC 791).
Accordingly, the eth1394 driver of Linux 2.6.36 and older set and got
this field with a -/+1 offset:
Likewise, I observe OS X 10.4 and Windows XP Pro SP3 to transmit 1500
byte sized datagrams in fragments with datagram_size=1499 if link
fragmentation is required.
Only firewire-net sets and gets datagram_size without this offset. The
result is lacking interoperability of firewire-net with OS X, Windows
XP, and presumably Linux' eth1394. (I did not test with the latter.)
For example, FTP data transfers to a Linux firewire-net box with max_rec
smaller than the 1500 bytes MTU
- from OS X fail entirely,
- from Win XP start out with a bunch of fragmented datagrams which
time out, then continue with unfragmented datagrams because Win XP
temporarily reduces the MTU to 576 bytes.
So let's fix firewire-net's datagram_size accessors.
Note that firewire-net thereby loses interoperability with unpatched
firewire-net, but only if link fragmentation is employed. (This happens
with large broadcast datagrams, and with large datagrams on several
FireWire CardBus cards with smaller max_rec than equivalent PCI cards,
and it can be worked around by setting a small enough MTU.)
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> Signed-off-by: Willy Tarreau <w@1wt.eu>
The IP-over-1394 driver firewire-net lacked input validation when
handling incoming fragmented datagrams. A maliciously formed fragment
with a respectively large datagram_offset would cause a memcpy past the
datagram buffer.
So, drop any packets carrying a fragment with offset + length larger
than datagram_size.
In addition, ensure that
- GASP header, unfragmented encapsulation header, or fragment
encapsulation header actually exists before we access it,
- the encapsulated datagram or fragment is of nonzero size.
The VF administrative mac addresses (stored in the PF driver) are
initialized to zero when the PF driver starts up.
These addresses may be modified in the PF driver through ndo calls
initiated by iproute2 or libvirt.
While we allow the PF/host to change the VF admin mac address from zero
to a valid unicast mac, we do not allow restoring the VF admin mac to
zero. We currently only allow changing this mac to a different unicast mac.
This leads to problems when libvirt scripts are used to deal with
VF mac addresses, and libvirt attempts to revoke the mac so this
host will not use it anymore.
Fix this by allowing resetting a VF administrative MAC back to zero.
Fixes: 8f7ba3ca12f6 ('net/mlx4: Add set VF mac address support') Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il> Reported-by: Moshe Levi <moshele@mellanox.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Juerg Haefliger <juerg.haefliger@hpe.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
User-space can choose to omit NL80211_ATTR_SSID and only provide raw
IE TLV data. When doing so it can provide SSID IE with length exceeding
the allowed size. The driver further processes this IE copying it
into a local variable without checking the length. Hence stack can be
corrupted and used as exploit.
Reported-by: Daxing Guo <freener.gdx@gmail.com> Reviewed-by: Hante Meuleman <hante.meuleman@broadcom.com> Reviewed-by: Pieter-Paul Giesberts <pieter-paul.giesberts@broadcom.com> Reviewed-by: Franky Lin <franky.lin@broadcom.com> Signed-off-by: Arend van Spriel <arend.vanspriel@broadcom.com> Signed-off-by: Kalle Valo <kvalo@codeaurora.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
wlc_phy_txpower_get_current() does a logical OR of power->flags, which
presumes that power.flags was initiliazed earlier by the caller,
unfortunately, this is not the case, so make sure we zero out the struct
tx_power before calling into wlc_phy_txpower_get_current().
Reported-by: coverity (CID 146011) Fixes: 5b435de0d7868 ("net: wireless: add brcm80211 drivers") Signed-off-by: Florian Fainelli <f.fainelli@gmail.com> Acked-by: Arend van Spriel <arend.vanspriel@broadcom.com> Signed-off-by: Kalle Valo <kvalo@codeaurora.org> Signed-off-by: Willy Tarreau <w@1wt.eu>
If an RPC program does not set vs_dispatch and pc_func() returns
rpc_drop_reply, the server sends a reply anyway containing a single
word containing the value RPC_DROP_REPLY (in network byte-order, of
course). This is a nonsense RPC message.
Fixes: 9e701c610923 ("svcrpc: simpler request dropping") Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Tested-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
This fixes a pretty ancient bug that hasn't manifested itself
until now.
The scratchbuf for command queue is allocated only for 32 slots
but is accessed with the queue write pointer - which can be
up to 256.
Since the scratch buf size was 16 and there are up to 256 TFDs
we never passed a page boundary when accessing the scratch buffer,
but when attempting to increase the size of the scratch buffer a
panic was quick to follow when trying to access the address resulted
in a page boundary.
Signed-off-by: Sara Sharon <sara.sharon@intel.com> Fixes: 38c0f334b359 ("iwlwifi: use coherent DMA memory for command header") Signed-off-by: Luca Coelho <luciano.coelho@intel.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
Some users observed that "least connection" distribution algorithm doesn't
handle well bursts of TCP connections from reconnecting clients after
a node or network failure.
This is because the algorithm counts active connection as worth 256
inactive ones where for TCP, "active" only means TCP connections in
ESTABLISHED state. In case of a connection burst, new connections are
handled before previous ones have finished the three way handshaking so
that all are still counted as "inactive", i.e. cheap ones. The become
"active" quickly but at that time, all of them are already assigned to one
real server (or few), resulting in highly unbalanced distribution.
Address this by counting the "pre-established" states as "active".
Signed-off-by: Michal Kubecek <mkubecek@suse.cz> Acked-by: Julian Anastasov <ja@ssi.bg> Signed-off-by: Simon Horman <horms@verge.net.au> Signed-off-by: Willy Tarreau <w@1wt.eu>
Before commit 6d7b857d541e ("net: use lib/percpu_counter API for
fragmentation mem accounting"), setting high threshold to 0 prevented
fragment reassembly as first fragment would be always evicted before
second could be added to the queue. While inefficient, some users
apparently relied on it.
Since the commit mentioned above, a percpu counter is used for
reassembly memory accounting and high batch size avoids taking slow path
in most common scenarios. As a result, a whole full sized packet can be
reassembled without the percpu counter's main counter changing its
value so that even with high_thresh set to 0, fragmented packets can be
still reassembled and processed.
Add explicit checks preventing reassembly if high threshold is zero.
According to figure 39 in PEB3086 data sheet, version 1.4 this indication
replaces DR when layer 1 transition source state is F6.
This fixes mISDN layer 1 getting stuck in F6 state in TE mode on
Dialogic Diva 2.02 card (and possibly others) when NT deactivates it.
Signed-off-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name> Acked-by: Karsten Keil <keil@b1-systems.de> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Kernel generates a lot of warnings when dst entry reference counter
overflows and becomes negative. That bug was seen several times at
machines with outdated 3.10.y kernels. Most like it's already fixed
in upstream. Anyway that flood completely kills machine and makes
further debugging impossible.
Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
(a) Create bonding master
> modprobe bonding miimon=50
(b) Create macvlan bridge on eth2
> ip link add link eth2 dev mvl0 address aa:0:0:0:0:01 \
type macvlan
(c) Now try adding eth2 into the bond
> echo +eth2 > /sys/class/net/bond0/bonding/slaves
<crash>
Bonding does lots of things before checking if the device enslaved is
busy or not.
In this case when the notifier call-chain sends notifications, the
bond_netdev_event() assumes that the rx_handler /rx_handler_data is
registered while the bond_enslave() hasn't progressed far enough to
register rx_handler for the new slave.
This patch adds a rx_handler check that can be performed right at the
beginning of the enslave code to avoid getting into this situation.
Signed-off-by: Mahesh Bandewar <maheshb@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
With syzkaller help, Marco Grassi found a bug in TCP stack,
crashing in tcp_collapse()
Root cause is that sk_filter() can truncate the incoming skb,
but TCP stack was not really expecting this to happen.
It probably was expecting a simple DROP or ACCEPT behavior.
We first need to make sure no part of TCP header could be removed.
Then we need to adjust TCP_SKB_CB(skb)->end_seq
Many thanks to syzkaller team and Marco for giving us a reproducer.
Signed-off-by: Eric Dumazet <edumazet@google.com> Reported-by: Marco Grassi <marco.gra@gmail.com> Reported-by: Vladis Dronov <vdronov@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
With TCP MTU probing enabled and offload TX checksumming disabled,
tcp_mtu_probe() calculated the wrong checksum when a fragment being copied
into the probe's SKB had an odd length. This was caused by the direct use
of skb_copy_and_csum_bits() to calculate the checksum, as it pads the
fragment being copied, if needed. When this fragment was not the last, a
subsequent call used the previous checksum without considering this
padding.
The effect was a stale connection in one way, as even retransmissions
wouldn't solve the problem, because the checksum was never recalculated for
the full SKB length.
Signed-off-by: Douglas Caetano dos Santos <douglascs@taghos.com.br> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
When tcp_sendmsg() allocates a fresh and empty skb, it puts it at the
tail of the write queue using tcp_add_write_queue_tail()
Then it attempts to copy user data into this fresh skb.
If the copy fails, we undo the work and remove the fresh skb.
Unfortunately, this undo lacks the change done to tp->highest_sack and
we can leave a dangling pointer (to a freed skb)
Later, tcp_xmit_retransmit_queue() can dereference this pointer and
access freed memory. For regular kernels where memory is not unmapped,
this might cause SACK bugs because tcp_highest_sack_seq() is buggy,
returning garbage instead of tp->snd_nxt, but with various debug
features like CONFIG_DEBUG_PAGEALLOC, this can crash the kernel.
This bug was found by Marco Grassi thanks to syzkaller.
Fixes: 6859d49475d4 ("[TCP]: Abstract tp->highest_sack accessing & point to next skb") Reported-by: Marco Grassi <marco.gra@gmail.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi> Cc: Yuchung Cheng <ycheng@google.com> Cc: Neal Cardwell <ncardwell@google.com> Acked-by: Neal Cardwell <ncardwell@google.com> Reviewed-by: Cong Wang <xiyou.wangcong@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
If iriap_register_lsap() fails to allocate memory, self->lsap is
set to NULL. However, none of the callers handle the failure and
irlmp_connect_request() will happily dereference it:
There's more problems with missing error checks in iriap_init() (and
indeed all of irda_init()), but that's a bigger problem that needs
very careful review and testing. This patch will fix the most serious
bug (as it's easily reached from unprivileged userspace).
I have tested my patch with a reproducer.
Signed-off-by: Vegard Nossum <vegard.nossum@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
skb->cb may contain data from previous layers. In the observed scenario,
the garbage data were misinterpreted as IP6CB(skb)->frag_max_size, so
that small packets sent through the tunnel are mistakenly fragmented.
This patch unconditionally clears the control buffer in ip6tunnel_xmit(),
which affects ip6_tunnel, ip6_udp_tunnel and ip6_gre. Currently none of
these tunnels set IP6CB(skb)->flags, otherwise it needs to be done earlier.
Cc: stable@vger.kernel.org Signed-off-by: Eli Cooper <elicooper@gmx.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
The goal of the patch is to fix this scenario:
ip link add dummy1 type dummy
ip link set dummy1 up
ip link set lo down ; ip link set lo up
After that sequence, the local route to the link layer address of dummy1 is
not there anymore.
When the loopback is set down, all local routes are deleted by
addrconf_ifdown()/rt6_ifdown(). At this time, the rt6_info entry still
exists, because the corresponding idev has a reference on it. After the rcu
grace period, dst_rcu_free() is called, and thus ___dst_free(), which will
set obsolete to DST_OBSOLETE_DEAD.
In this case, init_loopback() is called before dst_rcu_free(), thus
obsolete is still sets to something <= 0. So, the function doesn't add the
route again. To avoid that race, let's check the rt6 refcnt instead.
Fixes: 25fb6ca4ed9c ("net IPv6 : Fix broken IPv6 routing table after loopback down-up") Fixes: a881ae1f625c ("ipv6: don't call addrconf_dst_alloc again when enable lo") Fixes: 33d99113b110 ("ipv6: reallocate addrconf router for ipv6 address when lo device up") Reported-by: Francesco Santoro <francesco.santoro@6wind.com> Reported-by: Samuel Gauthier <samuel.gauthier@6wind.com> CC: Balakumaran Kannan <Balakumaran.Kannan@ap.sony.com> CC: Maruthi Thotad <Maruthi.Thotad@ap.sony.com> CC: Sabrina Dubroca <sd@queasysnail.net> CC: Hannes Frederic Sowa <hannes@stressinduktion.org> CC: Weilong Chen <chenweilong@huawei.com> CC: Gao feng <gaofeng@cn.fujitsu.com> Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Similar to commit 3be07244b733 ("ip6_gre: fix flowi6_proto value in
xmit path"), set flowi6_proto to IPPROTO_GRE for output route lookup.
Up until now, ip6gre_xmit_other() has set flowi6_proto to a bogus value.
This affected output route lookup for packets sent on an ip6gretap device
in cases where routing was dependent on the value of flowi6_proto.
Since the correct proto is already set in the tunnel flowi6 template via
commit 252f3f5a1189 ("ip6_gre: Set flowi6_proto as IPPROTO_GRE in xmit
path."), simply delete the line setting the incorrect flowi6_proto value.
Suggested-by: Jiri Benc <jbenc@redhat.com> Fixes: c12b395a4664 ("gre: Support GRE over IPv6") Reviewed-by: Shmulik Ladkani <shmulik.ladkani@gmail.com> Signed-off-by: Lance Richardson <lrichard@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Willy Tarreau <w@1wt.eu>
Calling setsockopt with IPV6_JOIN_ANYCAST or IPV6_LEAVE_ANYCAST
triggers the assertion in addrconf_join_solict()/addrconf_leave_solict()
ipv6_sock_ac_join(), ipv6_sock_ac_drop(), ipv6_sock_ac_close() need to
take RTNL before calling ipv6_dev_ac_inc/dec. Same thing with
ipv6_sock_mc_join(), ipv6_sock_mc_drop(), ipv6_sock_mc_close() before
calling ipv6_dev_mc_inc/dec.
This patch moves ASSERT_RTNL() up a level in the call stack.
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com> Signed-off-by: Sabrina Dubroca <sd@queasysnail.net> Reported-by: Tommi Rantala <tt.rantala@gmail.com> Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: David S. Miller <davem@davemloft.net> Cc: <stable@vger.kernel.org> # 3.10.y: b7b1bfce: ipv6: split dad and rs timers Cc: <stable@vger.kernel.org> # 3.10.y: c15b1cca: ipv6: move dad to workqueue Cc: <stable@vger.kernel.org> # 3.10.y
[Mike Manning <mmanning@brocade.com>: resolved minor conflicts in addrconf.c] Signed-off-by: Mike Manning <mmanning@brocade.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
In general, when DAD detected IPv6 duplicate address, ifp->state
will be set to INET6_IFADDR_STATE_ERRDAD and DAD is stopped by a
delayed work, the call tree should be like this:
ndisc_recv_ns
-> addrconf_dad_failure <- missing ifp put
-> addrconf_mod_dad_work
-> schedule addrconf_dad_work()
-> addrconf_dad_stop() <- missing ifp hold before call it
addrconf_dad_failure() called with ifp refcont holding but not put.
addrconf_dad_work() call addrconf_dad_stop() without extra holding
refcount. This will not cause any issue normally.
But the race between addrconf_dad_failure() and addrconf_dad_work()
may cause ifp refcount leak and netdevice can not be unregister,
dmesg show the following messages:
IPv6: eth0: IPv6 duplicate address fe80::XX:XXXX:XXXX:XX detected!
...
unregister_netdevice: waiting for eth0 to become free. Usage count = 1
Cc: stable@vger.kernel.org Fixes: c15b1ccadb32 ("ipv6: move DAD and addrconf_verify processing
to workqueue") Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net> Cc: <stable@vger.kernel.org> # 3.10.y Signed-off-by: Mike Manning <mmanning@brocade.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
addrconf_join_solict and addrconf_join_anycast may cause actions which
need rtnl locked, especially on first address creation.
A new DAD state is introduced which defers processing of the initial
DAD processing into a workqueue.
To get rtnl lock we need to push the code paths which depend on those
calls up to workqueues, specifically addrconf_verify and the DAD
processing.
(v2)
addrconf_dad_failure needs to be queued up to the workqueue, too. This
patch introduces a new DAD state and stop the DAD processing in the
workqueue (this is because of the possible ipv6_del_addr processing
which removes the solicited multicast address from the device).
addrconf_verify_lock is removed, too. After the transition it is not
needed any more.
As we are not processing in bottom half anymore we need to be a bit more
careful about disabling bottom half out when we lock spin_locks which are also
used in bh.
Hunks and backtrace stolen from a patch by Stephen Hemminger.
Reported-by: Stephen Hemminger <stephen@networkplumber.org> Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: David S. Miller <davem@davemloft.net> Cc: <stable@vger.kernel.org> # 3.10.y: b7b1bfce: ipv6: split dad and rs timers Cc: <stable@vger.kernel.org> # 3.10.y
[Mike Manning <mmanning@brocade.com>: resolved minor conflicts in addrconf.c] Signed-off-by: Mike Manning <mmanning@brocade.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
This patch splits the timers for duplicate address detection and router
solicitations apart. The router solicitations timer goes into inet6_dev
and the dad timer stays in inet6_ifaddr.
The reason behind this patch is to reduce the number of unneeded router
solicitations send out by the host if additional link-local addresses
are created. Currently we send out RS for every link-local address on
an interface.
If the RS timer fires we pick a source address with ipv6_get_lladdr. This
change could hurt people adding additional link-local addresses and
specifying these addresses in the radvd clients section because we
no longer guarantee that we use every ll address as source address in
router solicitations.
Cc: Flavio Leitner <fleitner@redhat.com> Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org> Cc: David Stevens <dlstevens@us.ibm.com> Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Reviewed-by: Flavio Leitner <fbl@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Cc: <stable@vger.kernel.org> # 3.10.y
[Mike Manning <mmanning@brocade.com>: resolved conflicts with 36bddb] Signed-off-by: Mike Manning <mmanning@brocade.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
When loading the ipv6 module, ndisc_init() is called before
ip6_route_init(). As the former registers a handler calling
fib6_run_gc(), this opens a window to run the garbage collector
before necessary data structures are initialized. If a network
device is initialized in this window, adding MAC address to it
triggers a NETDEV_CHANGEADDR event, leading to a crash in
fib6_clean_all().
Take the event handler registration out of ndisc_init() into a
separate function ndisc_late_init() and move it after
ip6_route_init().
Signed-off-by: Michal Kubecek <mkubecek@suse.cz> Signed-off-by: David S. Miller <davem@davemloft.net> Cc: <stable@vger.kernel.org> # 3.10.y Signed-off-by: Mike Manning <mmanning@brocade.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
Commit 3876488444e7 ("include/stddef.h: Move offsetofend() from vfio.h
to a generic kernel header") added offsetofend outside the normal
include #ifndef/#endif guard. Move it inside.
Miscellanea:
o remove unnecessary blank line
o standardize offsetof macros whitespace style
Signed-off-by: Joe Perches <joe@perches.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
[wt: backported only for ipv6 out-of-bounds fix]
The macro offsetofend() introduces unnecessary temporary variable
"tmp". The patch avoids that and saves a bit memory in stack.
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com> Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
[wt: backported only for ipv6 out-of-bounds fix]
In csi_J(3), the third parameter of scr_memsetw (vc_screenbuf_size) is
divided by 2 inappropriatelly. But scr_memsetw expects size, not
count, because it divides the size by 2 on its own before doing actual
memset-by-words.