linux-stable.git/Documentation, branch v3.16.75

x86/speculation/mds: Improve CPU buffer clear documentation

2019-09-23T20:12:07+00:00

commit 9d8d0294e78a164d407133dea05caf4b84247d6a upstream.

On x86_64, all returns to usermode go through
prepare_exit_to_usermode(), with the sole exception of do_nmi().
This even includes machine checks -- this was added several years
ago to support MCE recovery.  Update the documentation.

Signed-off-by: Andy Lutomirski 
Cc: Borislav Petkov 
Cc: Frederic Weisbecker 
Cc: Greg Kroah-Hartman 
Cc: Jon Masters 
Cc: Linus Torvalds 
Cc: Peter Zijlstra 
Cc: Thomas Gleixner 
Fixes: 04dcbdb80578 ("x86/speculation/mds: Clear CPU buffers on exit to user")
Link: http://lkml.kernel.org/r/999fa9e126ba6a48e9d214d2f18dbde5c62ac55c.1557865329.git.luto@kernel.org
Signed-off-by: Ingo Molnar 
Signed-off-by: Ben Hutchings

x86/speculation/mds: Revert CPU buffer clear on double fault exit

2019-09-23T20:12:07+00:00

commit 88640e1dcd089879530a49a8d212d1814678dfe7 upstream.

The double fault ESPFIX path doesn't return to user mode at all --
it returns back to the kernel by simulating a #GP fault.
prepare_exit_to_usermode() will run on the way out of
general_protection before running user code.

Signed-off-by: Andy Lutomirski 
Cc: Borislav Petkov 
Cc: Frederic Weisbecker 
Cc: Greg Kroah-Hartman 
Cc: Jon Masters 
Cc: Linus Torvalds 
Cc: Peter Zijlstra 
Cc: Thomas Gleixner 
Fixes: 04dcbdb80578 ("x86/speculation/mds: Clear CPU buffers on exit to user")
Link: http://lkml.kernel.org/r/ac97612445c0a44ee10374f6ea79c222fe22a5c4.1557865329.git.luto@kernel.org
Signed-off-by: Ingo Molnar 
[bwh: Backported to 3.16: adjust context]
Signed-off-by: Ben Hutchings

siphash: implement HalfSipHash1-3 for hash tables

2019-08-20T18:01:56+00:00

commit 1ae2324f732c9c4e2fa4ebd885fa1001b70d52e1 upstream.

HalfSipHash, or hsiphash, is a shortened version of SipHash, which
generates 32-bit outputs using a weaker 64-bit key. It has *much* lower
security margins, and shouldn't be used for anything too sensitive, but
it could be used as a hashtable key function replacement, if the output
is never exposed, and if the security requirement is not too high.

The goal is to make this something that performance-critical jhash users
would be willing to use.

On 64-bit machines, HalfSipHash1-3 is slower than SipHash1-3, so we alias
SipHash1-3 to HalfSipHash1-3 on those systems.

64-bit x86_64:
[    0.509409] test_siphash:     SipHash2-4 cycles: 4049181
[    0.510650] test_siphash:     SipHash1-3 cycles: 2512884
[    0.512205] test_siphash: HalfSipHash1-3 cycles: 3429920
[    0.512904] test_siphash:    JenkinsHash cycles:  978267
So, we map hsiphash() -> SipHash1-3

32-bit x86:
[    0.509868] test_siphash:     SipHash2-4 cycles: 14812892
[    0.513601] test_siphash:     SipHash1-3 cycles:  9510710
[    0.515263] test_siphash: HalfSipHash1-3 cycles:  3856157
[    0.515952] test_siphash:    JenkinsHash cycles:  1148567
So, we map hsiphash() -> HalfSipHash1-3

hsiphash() is roughly 3 times slower than jhash(), but comes with a
considerable security improvement.

Signed-off-by: Jason A. Donenfeld 
Reviewed-by: Jean-Philippe Aumasson 
Signed-off-by: David S. Miller 
[bwh: Backported to 3.16 to avoid a build regression for WireGuard with
 only part of the siphash API available]
Signed-off-by: Ben Hutchings

siphash: add cryptographically secure PRF

2019-08-13T11:39:32+00:00

commit 2c956a60778cbb6a27e0c7a8a52a91378c90e1d1 upstream.

SipHash is a 64-bit keyed hash function that is actually a
cryptographically secure PRF, like HMAC. Except SipHash is super fast,
and is meant to be used as a hashtable keyed lookup function, or as a
general PRF for short input use cases, such as sequence numbers or RNG
chaining.

For the first usage:

There are a variety of attacks known as "hashtable poisoning" in which an
attacker forms some data such that the hash of that data will be the
same, and then preceeds to fill up all entries of a hashbucket. This is
a realistic and well-known denial-of-service vector. Currently
hashtables use jhash, which is fast but not secure, and some kind of
rotating key scheme (or none at all, which isn't good). SipHash is meant
as a replacement for jhash in these cases.

There are a modicum of places in the kernel that are vulnerable to
hashtable poisoning attacks, either via userspace vectors or network
vectors, and there's not a reliable mechanism inside the kernel at the
moment to fix it. The first step toward fixing these issues is actually
getting a secure primitive into the kernel for developers to use. Then
we can, bit by bit, port things over to it as deemed appropriate.

While SipHash is extremely fast for a cryptographically secure function,
it is likely a bit slower than the insecure jhash, and so replacements
will be evaluated on a case-by-case basis based on whether or not the
difference in speed is negligible and whether or not the current jhash usage
poses a real security risk.

For the second usage:

A few places in the kernel are using MD5 or SHA1 for creating secure
sequence numbers, syn cookies, port numbers, or fast random numbers.
SipHash is a faster and more fitting, and more secure replacement for MD5
in those situations. Replacing MD5 and SHA1 with SipHash for these uses is
obvious and straight-forward, and so is submitted along with this patch
series. There shouldn't be much of a debate over its efficacy.

Dozens of languages are already using this internally for their hash
tables and PRFs. Some of the BSDs already use this in their kernels.
SipHash is a widely known high-speed solution to a widely known set of
problems, and it's time we catch-up.

Signed-off-by: Jason A. Donenfeld 
Reviewed-by: Jean-Philippe Aumasson 
Cc: Linus Torvalds 
Cc: Eric Biggers 
Cc: David Laight 
Cc: Eric Dumazet 
Signed-off-by: David S. Miller 
Signed-off-by: Ben Hutchings

x86/speculation: Enable Spectre v1 swapgs mitigations

2019-08-13T11:39:21+00:00

commit a2059825986a1c8143fd6698774fa9d83733bb11 upstream.

The previous commit added macro calls in the entry code which mitigate the
Spectre v1 swapgs issue if the X86_FEATURE_FENCE_SWAPGS_* features are
enabled.  Enable those features where applicable.

The mitigations may be disabled with "nospectre_v1" or "mitigations=off".

There are different features which can affect the risk of attack:

- When FSGSBASE is enabled, unprivileged users are able to place any
  value in GS, using the wrgsbase instruction.  This means they can
  write a GS value which points to any value in kernel space, which can
  be useful with the following gadget in an interrupt/exception/NMI
  handler:

	if (coming from user space)
		swapgs
	mov %gs:, %reg1
	// dependent load or store based on the value of %reg
	// for example: mov %(reg1), %reg2

  If an interrupt is coming from user space, and the entry code
  speculatively skips the swapgs (due to user branch mistraining), it
  may speculatively execute the GS-based load and a subsequent dependent
  load or store, exposing the kernel data to an L1 side channel leak.

  Note that, on Intel, a similar attack exists in the above gadget when
  coming from kernel space, if the swapgs gets speculatively executed to
  switch back to the user GS.  On AMD, this variant isn't possible
  because swapgs is serializing with respect to future GS-based
  accesses.

  NOTE: The FSGSBASE patch set hasn't been merged yet, so the above case
	doesn't exist quite yet.

- When FSGSBASE is disabled, the issue is mitigated somewhat because
  unprivileged users must use prctl(ARCH_SET_GS) to set GS, which
  restricts GS values to user space addresses only.  That means the
  gadget would need an additional step, since the target kernel address
  needs to be read from user space first.  Something like:

	if (coming from user space)
		swapgs
	mov %gs:, %reg1
	mov (%reg1), %reg2
	// dependent load or store based on the value of %reg2
	// for example: mov %(reg2), %reg3

  It's difficult to audit for this gadget in all the handlers, so while
  there are no known instances of it, it's entirely possible that it
  exists somewhere (or could be introduced in the future).  Without
  tooling to analyze all such code paths, consider it vulnerable.

  Effects of SMAP on the !FSGSBASE case:

  - If SMAP is enabled, and the CPU reports RDCL_NO (i.e., not
    susceptible to Meltdown), the kernel is prevented from speculatively
    reading user space memory, even L1 cached values.  This effectively
    disables the !FSGSBASE attack vector.

  - If SMAP is enabled, but the CPU *is* susceptible to Meltdown, SMAP
    still prevents the kernel from speculatively reading user space
    memory.  But it does *not* prevent the kernel from reading the
    user value from L1, if it has already been cached.  This is probably
    only a small hurdle for an attacker to overcome.

Thanks to Dave Hansen for contributing the speculative_smap() function.

Thanks to Andrew Cooper for providing the inside scoop on whether swapgs
is serializing on AMD.

[ tglx: Fixed the USER fence decision and polished the comment as suggested
  	by Dave Hansen ]

Signed-off-by: Josh Poimboeuf 
Signed-off-by: Thomas Gleixner 
Reviewed-by: Dave Hansen 
[bwh: Backported to 3.16:
 - Check for X86_FEATURE_KAISER instead of X86_FEATURE_PTI
 - mitigations= parameter is x86-only here
 - powerpc doesn't have Spectre mitigations
 - Don't use __ro_after_init
 - Adjust filename, context]
Signed-off-by: Ben Hutchings

USB: core: Fix bug caused by duplicate interface PM usage counter

2019-08-13T11:39:12+00:00

commit c2b71462d294cf517a0bc6e4fd6424d7cee5596f upstream.

The syzkaller fuzzer reported a bug in the USB hub driver which turned
out to be caused by a negative runtime-PM usage counter.  This allowed
a hub to be runtime suspended at a time when the driver did not expect
it.  The symptom is a WARNING issued because the hub's status URB is
submitted while it is already active:

	URB 0000000031fb463e submitted while active
	WARNING: CPU: 0 PID: 2917 at drivers/usb/core/urb.c:363

The negative runtime-PM usage count was caused by an unfortunate
design decision made when runtime PM was first implemented for USB.
At that time, USB class drivers were allowed to unbind from their
interfaces without balancing the usage counter (i.e., leaving it with
a positive count).  The core code would take care of setting the
counter back to 0 before allowing another driver to bind to the
interface.

Later on when runtime PM was implemented for the entire kernel, the
opposite decision was made: Drivers were required to balance their
runtime-PM get and put calls.  In order to maintain backward
compatibility, however, the USB subsystem adapted to the new
implementation by keeping an independent usage counter for each
interface and using it to automatically adjust the normal usage
counter back to 0 whenever a driver was unbound.

This approach involves duplicating information, but what is worse, it
doesn't work properly in cases where a USB class driver delays
decrementing the usage counter until after the driver's disconnect()
routine has returned and the counter has been adjusted back to 0.
Doing so would cause the usage counter to become negative.  There's
even a warning about this in the USB power management documentation!

As it happens, this is exactly what the hub driver does.  The
kick_hub_wq() routine increments the runtime-PM usage counter, and the
corresponding decrement is carried out by hub_event() in the context
of the hub_wq work-queue thread.  This work routine may sometimes run
after the driver has been unbound from its interface, and when it does
it causes the usage counter to go negative.

It is not possible for hub_disconnect() to wait for a pending
hub_event() call to finish, because hub_disconnect() is called with
the device lock held and hub_event() acquires that lock.  The only
feasible fix is to reverse the original design decision: remove the
duplicate interface-specific usage counter and require USB drivers to
balance their runtime PM gets and puts.  As far as I know, all
existing drivers currently do this.

Signed-off-by: Alan Stern 
Reported-and-tested-by: syzbot+7634edaea4d0b341c625@syzkaller.appspotmail.com
Signed-off-by: Greg Kroah-Hartman 
[bwh: Backported to 3.16:
 - Adjust documentation filename
 - Don't add ReST markup in documentation
 - Update use of pm_usage_cnt in poseidon driver, which has been
   removed upstream]
Signed-off-by: Ben Hutchings

KVM: Reject device ioctls from processes other than the VM's creator

2019-08-13T11:38:56+00:00

commit ddba91801aeb5c160b660caed1800eb3aef403f8 upstream.

KVM's API requires thats ioctls must be issued from the same process
that created the VM.  In other words, userspace can play games with a
VM's file descriptors, e.g. fork(), SCM_RIGHTS, etc..., but only the
creator can do anything useful.  Explicitly reject device ioctls that
are issued by a process other than the VM's creator, and update KVM's
API documentation to extend its requirements to device ioctls.

Fixes: 852b6d57dc7f ("kvm: add device control API")
Signed-off-by: Sean Christopherson 
Signed-off-by: Paolo Bonzini 
Signed-off-by: Ben Hutchings

tcp: add tcp_min_snd_mss sysctl

2019-06-20T17:11:28+00:00

commit 5f3e2bf008c2221478101ee72f5cb4654b9fc363 upstream.

Some TCP peers announce a very small MSS option in their SYN and/or
SYN/ACK messages.

This forces the stack to send packets with a very high network/cpu
overhead.

Linux has enforced a minimal value of 48. Since this value includes
the size of TCP options, and that the options can consume up to 40
bytes, this means that each segment can include only 8 bytes of payload.

In some cases, it can be useful to increase the minimal value
to a saner value.

We still let the default to 48 (TCP_MIN_SND_MSS), for compatibility
reasons.

Note that TCP_MAXSEG socket option enforces a minimal value
of (TCP_MIN_MSS). David Miller increased this minimal value
in commit c39508d6f118 ("tcp: Make TCP_MAXSEG minimum more correct.")
from 64 to 88.

We might in the future merge TCP_MIN_SND_MSS and TCP_MIN_MSS.

CVE-2019-11479 -- tcp mss hardcoded to 48

Signed-off-by: Eric Dumazet 
Suggested-by: Jonathan Looney 
Acked-by: Neal Cardwell 
Cc: Yuchung Cheng 
Cc: Tyler Hicks 
Cc: Bruce Curtis 
Cc: Jonathan Lemon 
Signed-off-by: David S. Miller 
[Salvatore Bonaccorso: Backport for context changes in 4.9.168]
[bwh: Backported to 3.16: Make the sysctl global, consistent with
 net.ipv4.tcp_base_mss]
Signed-off-by: Ben Hutchings

x86/speculation/mds: Fix documentation typo

2019-05-22T22:15:27+00:00

commit 95310e348a321b45fb746c176961d4da72344282 upstream.

Fix a minor typo in the MDS documentation: "eanbled" -> "enabled".

Reported-by: Jeff Bastian 
Signed-off-by: Josh Poimboeuf 
Signed-off-by: Thomas Gleixner 
Signed-off-by: Ben Hutchings

Documentation: Correct the possible MDS sysfs values

2019-05-22T22:15:27+00:00

commit ea01668f9f43021b28b3f4d5ffad50106a1e1301 upstream.

Adjust the last two rows in the table that display possible values when
MDS mitigation is enabled. They both were slightly innacurate.

In addition, convert the table of possible values and their descriptions
to a list-table. The simple table format uses the top border of equals
signs to determine cell width which resulted in the first column being
far too wide in comparison to the second column that contained the
majority of the text.

Signed-off-by: Tyler Hicks 
Signed-off-by: Thomas Gleixner 
[bwh: Backported to 3.16: adjust filename]
Signed-off-by: Ben Hutchings