Fedora has integrated the jitter entropy daemon to work around slow
boot problems, especially on VM's that don't support virtio-rng:

    https://bugzilla.redhat.com/show_bug.cgi?id=1572944

It's understandable why they did this, but the Jitter entropy daemon
works fundamentally on the principle: "the CPU microarchitecture is
**so** complicated and we can't figure it out, so it *must* be
random".  Yes, it uses statistical tests to "prove" it is secure, but
AES_ENCRYPT(NSA_KEY, COUNTER++) will also pass statistical tests with
flying colors.

So if RDRAND is available, mix it into entropy submitted from
userspace.  It can't hurt, and if you believe the NSA has backdoored
RDRAND, then they probably have enough details about the Intel
microarchitecture that they can reverse engineer how the Jitter
entropy daemon affects the microarchitecture, and attack its output
stream.  And if RDRAND is in fact an honest DRNG, it will immeasurably
improve on what the Jitter entropy daemon might produce.

This also provides some protection against someone who is able to read
or set the entropy seed file.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org
Cc: Arnd Bergmann <arnd@arndb.de>

The poll() changes were not well thought out, and completely
unexplained.  They also caused a huge performance regression, because
"->poll()" was no longer a trivial file operation that just called down
to the underlying file operations, but instead did at least two indirect
calls.

Indirect calls are sadly slow now with the Spectre mitigation, but the
performance problem could at least be largely mitigated by changing the
"->get_poll_head()" operation to just have a per-file-descriptor pointer
to the poll head instead.  That gets rid of one of the new indirections.

But that doesn't fix the new complexity that is completely unwarranted
for the regular case.  The (undocumented) reason for the poll() changes
was some alleged AIO poll race fixing, but we don't make the common case
slower and more complex for some uncommon special case, so this all
really needs way more explanations and most likely a fundamental
redesign.

[ This revert is a revert of about 30 different commits, not reverted
  individually because that would just be unnecessarily messy  - Linus ]

Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

The big change is that random_read_wait and random_write_wait are merged
into a single waitqueue that uses keyed wakeups.  Because wait_event_*
doesn't know about that this will lead to occassional spurious wakeups
in _random_read and add_hwgenerator_randomness, but wait_event_* is
designed to handle these and were are not in a a hot path there.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

On systems without sufficient boot randomness, no point spamming dmesg.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org

We can do a sleeping allocation from an irq context when CONFIG_NUMA
is enabled.  Fix this by initializing the NUMA crng instances in a
workqueue.

Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Reported-by: syzbot+9de458f6a5e713ee8c1a@syzkaller.appspotmail.com
Fixes: 8ef35c866f8862df ("random: set up the NUMA crng instances...")
Cc: stable@vger.kernel.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

Add a new ioctl which forces the the crng to be reseeded.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org

Reported-by: Jann Horn <jannh@google.com>
Fixes: 1e7f583af67b ("random: make /dev/urandom scalable for silly...")
Cc: stable@kernel.org # 4.8+
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jann Horn <jannh@google.com>

Until the primary_crng is fully initialized, don't initialize the NUMA
crng nodes.  Otherwise users of /dev/urandom on NUMA systems before
the CRNG is fully initialized can get very bad quality randomness.  Of
course everyone should move to getrandom(2) where this won't be an
issue, but there's a lot of legacy code out there.  This related to
CVE-2018-1108.

Reported-by: Jann Horn <jannh@google.com>
Fixes: 1e7f583af67b ("random: make /dev/urandom scalable for silly...")
Cc: stable@kernel.org # 4.8+
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

add_device_randomness() use of crng_fast_load() was highly
problematic.  Some callers of add_device_randomness() can pass in a
large amount of static information.  This would immediately promote
the crng_init state from 0 to 1, without really doing much to
initialize the primary_crng's internal state with something even
vaguely unpredictable.

Since we don't have the speed constraints of add_interrupt_randomness(),
we can do a better job mixing in the what unpredictability a device
driver or architecture maintainer might see fit to give us, and do it
in a way which does not bump the crng_init_cnt variable.

Also, since add_device_randomness() doesn't bump any entropy
accounting in crng_init state 0, mix the device randomness into the
input_pool entropy pool as well.  This is related to CVE-2018-1108.

Reported-by: Jann Horn <jannh@google.com>
Fixes: ee7998c50c26 ("random: do not ignore early device randomness")
Cc: stable@kernel.org # 4.13+
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

The crng_init variable has three states:

0: The CRNG is not initialized at all
1: The CRNG has a small amount of entropy, hopefully good enough for
   early-boot, non-cryptographical use cases
2: The CRNG is fully initialized and we are sure it is safe for
   cryptographic use cases.

The crng_ready() function should only return true once we are in the
last state.  This addresses CVE-2018-1108.

Reported-by: Jann Horn <jannh@google.com>
Fixes: e192be9d9a30 ("random: replace non-blocking pool...")
Cc: stable@kernel.org # 4.8+
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jann Horn <jannh@google.com>