Commit adfa543e7314 ("dmatest: don't use set_freezable_with_signal()")
introduced a bug (that is in fact documented by the patch commit text)
that leaves behind a dangling pointer. Since the done_wait structure is
allocated on the stack, future invocations to the DMATEST can produce
undesirable results (e.g., corrupted spinlocks). Ideally, this would be
cleaned up in the thread handler, but at the very least, the kernel
is left in a very precarious scenario that can lead to some long debug
sessions when the crash comes later.
Yi Zhang reported the following failure on a 2-socket Haswell (E5-2603v3)
server (DELL PowerEdge 730xd):
EDAC sbridge: Some needed devices are missing
EDAC MC: Removed device 0 for sb_edac.c Haswell SrcID#0_Ha#0: DEV 0000:7f:12.0
EDAC MC: Removed device 1 for sb_edac.c Haswell SrcID#1_Ha#0: DEV 0000:ff:12.0
EDAC sbridge: Couldn't find mci handler
EDAC sbridge: Couldn't find mci handler
EDAC sbridge: Failed to register device with error -19.
The refactored sb_edac driver creates the IMC1 (the 2nd memory
controller) if any IMC1 device is present. In this case only
HA1_TA of IMC1 was present, but the driver expected to find
HA1/HA1_TM/HA1_TAD[0-3] devices too, leading to the above failure.
The document [1] says the 'E5-2603 v3' CPU has 4 memory channels max. Yi
Zhang inserted one DIMM per channel for each CPU, and did random error
address injection test with this patch:
4024 addresses fell in TOLM hole area
12715 addresses fell in CPU_SrcID#0_Ha#0_Chan#0_DIMM#0
12774 addresses fell in CPU_SrcID#0_Ha#0_Chan#1_DIMM#0
12798 addresses fell in CPU_SrcID#0_Ha#0_Chan#2_DIMM#0
12913 addresses fell in CPU_SrcID#0_Ha#0_Chan#3_DIMM#0
12674 addresses fell in CPU_SrcID#1_Ha#0_Chan#0_DIMM#0
12686 addresses fell in CPU_SrcID#1_Ha#0_Chan#1_DIMM#0
12882 addresses fell in CPU_SrcID#1_Ha#0_Chan#2_DIMM#0
12934 addresses fell in CPU_SrcID#1_Ha#0_Chan#3_DIMM#0
106400 addresses were injected totally.
The test result shows that all the 4 channels belong to IMC0 per CPU, so
the server really only has one IMC per CPU.
In the 1st page of chapter 2 in datasheet [2], it also says 'E5-2600 v3'
implements either one or two IMCs. For CPUs with one IMC, IMC1 is not
used and should be ignored.
Thus, do not create a second memory controller if the key HA1 is absent.
Before trying to use CDC union descriptor, try to validate whether that it
is sane by checking that intf->altsetting->extra is big enough and that
descriptor bLength is not too big and not too small.
If the usbtest driver encounters a device with an IN bulk endpoint but
no OUT bulk endpoint, it will try to dereference a NULL pointer
(out->desc.bEndpointAddress). The problem can be solved by adding a
missing test.
For the reinstall prevention, the code I had added compares the
whole key. It turns out though that iwlwifi firmware doesn't
provide the TKIP TX MIC key as it's not needed in client mode,
and thus the comparison will always return false.
For client mode, thus always zero out the TX MIC key part before
doing the comparison in order to avoid accepting the reinstall
of the key with identical encryption and RX MIC key, but not the
same TX MIC key (since the supplicant provides the real one.)
When a key is reinstalled we can reset the replay counters
etc. which can lead to nonce reuse and/or replay detection
being impossible, breaking security properties, as described
in the "KRACK attacks".
In particular, CVE-2017-13080 applies to GTK rekeying that
happened in firmware while the host is in D3, with the second
part of the attack being done after the host wakes up. In
this case, the wpa_supplicant mitigation isn't sufficient
since wpa_supplicant doesn't know the GTK material.
In case this happens, simply silently accept the new key
coming from userspace but don't take any action on it since
it's the same key; this keeps the PN replay counters intact.
Based on SNMP values provided by Roman, Yuchung made the observation
that some crashes in tcp_sacktag_walk() might be caused by MTU probing.
Looking at tcp_mtu_probe(), I found that when a new skb was placed
in front of the write queue, we were not updating tcp highest sack.
If one skb is freed because all its content was copied to the new skb
(for MTU probing), then tp->highest_sack could point to a now freed skb.
Bad things would then happen, including infinite loops.
This patch renames tcp_highest_sack_combine() and uses it
from tcp_mtu_probe() to fix the bug.
Note that I also removed one test against tp->sacked_out,
since we want to replace tp->highest_sack regardless of whatever
condition, since keeping a stale pointer to freed skb is a recipe
for disaster.
Fixes: a47e5a988a57 ("[TCP]: Convert highest_sack to sk_buff to allow direct access")
Signed-off-by: Eric Dumazet <email address hidden>
Reported-by: Alexei Starovoitov <email address hidden>
Reported-by: Roman Gushchin <email address hidden>
Reported-by: Oleksandr Natalenko <email address hidden>
Acked-by: Alexei Starovoitov <email address hidden>
Acked-by: Neal Cardwell <email address hidden>
Acked-by: Yuchung Cheng <email address hidden>
Signed-off-by: David S. Miller <email address hidden>
Signed-off-by: Greg Kroah-Hartman <email address hidden>
Signed-off-by: Khalid Elmously <email address hidden>
In the (unlikely) event fixup_permanent_addr() returns a failure,
addrconf_permanent_addr() calls ipv6_del_addr() without the
mandatory call to in6_ifa_hold(), leading to a refcount error,
spotted by syzkaller :
WARNING: CPU: 1 PID: 3142 at lib/refcount.c:227 refcount_dec+0x4c/0x50
lib/refcount.c:227
Kernel panic - not syncing: panic_on_warn set ...
