This will allow for e.g. xous on riscv32-imac but not a precursor
target, for example: by using feature selectors we get to have
a bigger range of targets that we support, such as for implementations
that haven't been upstreamed or have official target-triples.
The downsides are:
- patch override libraries don't absorb feature flags. so, for
example, the AES crate that is used to patch other crates won't
be aware of the target features without modifying the referencing
crates.
- features are normally meant to be additive, but target features
are mutually exclusive. The mutex is enforced by a check in
utralib that throws a panic on build if two targets are accidentally
selected at the same time.
- boot PIN is not requested on root password init. it's redundant,
as it is actually set when you format the PDDB
- boot key was mistakenly a copy of the encrypted private key
for ed25519 key signing. "oops". didn't catch that bug until now...
i don't think it has a big impact on existing user base, because
the key is never exposed outside the rootkey server, and the
ed25519 key itself is encrypted with the root password and then
used as the boot key. But still -- that's just not what was intended.
- root password is requested twice to check the validity of the password
this patch moves us off the branch we had to hop on for cipher
API compatibility to the one that's officially published by
RustCrypto. This closes#168.
The crate we originally grabbed for AES-KWP seems to be non-compliant
(or I'm an idiot and I can't figure out how to call it in a way
that is compliant).
See https://github.com/jedisct1/rust-aes-keywrap/issues/9
Turns out that `aes_kw` *is* compliant, but unfortunately, it didn't
exist back when the PDDB was being written. The PDDB started in October
2021, and `aes_kw` was pushed in Jan 2022. *just* barely missed it.
Anyways, caught this error while trying to implement some CI for
backups. Hurray for testing!
This patch removes the encryption path using the incorrect version
of the algorithm, but leaves the decryption path in place. The
system will try to decrypt initially with the correct version,
but if that fails, it will fall back and try the incorrect version.
If that succeeds, it returns an error code with the correct key and
an updated wrapped key, which the caller can store so that future
calls do not fail. Thus, we have a seamless migration from the
incorrect version to the correct version.
Have verified that the aes_kw version *does* work with the CI
script, insofar as it has been written (we just got to that point
and got stuck). So, full speed ahead. Users shouldn't notice this
change.
Also, fwiw, I did take a look at the "incorrect" implementation,
and it doesn't look obviously insecure. Printing out the various
rounds of the wrapping function looks like the first couple rounds
match, but it diverges somewhere probably due to an off-by-one
in a counter or something like that. I didn't chase it out -- just
wanted to make sure that nothing looked like a huge data compromise.
Good news is this comes right on the heels of v0.9.9 release, so
this will be pushed as a hotfix right away.
this facility is necessary in case a worker accidentally picks
the wrong setting, disabling the feature for the device. When
the factory test completes, a call is made to reset the parameter.
This was a slightly complicated affair because the the don't ask
flag is contained within the rootkeys enclave itself, and we can't
just give access to rootkeys willy-nilly. So the flag is passed
through the PDDB and then to the rootkeys enclave. It's not harmful
to have it reset by an unauthorized actor; the pop-up will not
happen if passwords are, in fact, already set.
verification happens after the private key material is erased,
which incidentally nukes public key material, causing verification
to fail. This caches a copy of the pubkey material before
the erase happens.
- read order of keys was wrong from staged data
- make restore a blocking op so the Ux doesn't try to trigger
other things at boot
- fix pubkey check to use staged key for backup flow
- fix some wording and skip redundant messages
Just remembered that the PDDB is *locked* to a given device.
To restore to a new device, you have to unlock all the Bases,
and re-encrypt them to the new device's DNA.
This commit adds the facilities to detect this condition, but
the actual code to do this is quite involved...let's handle that
after we've got a "simple" restore to the same device going.
Why do we lock to a device's DNA? the idea is to prevent
attacks where people are taking static blocks of crypto data
and just swapping it between devices to see if they can
get something to decrypt. It also makes it harder for
an adversary to replace the physical device -- e.g., an
evil maid prepared a replacement with implants in advance, and they
image your PDDB onto the replaced device, hoping that you don't
notice. Because the PDDB is device-locked, the "evil maid"
would have to forge the devices DNA to match yours:
possible, but would require kernel mods that patch over the
reported ID; that would trigger the line across the top of the
screen; or, somehow editing the FPGA itself to force
its device ID to report the same as yours. This last attack
vector would be successful, but afaik there is no known way
to modify an FPGA's DNA code. I'm sure there is some proprietary
secret know that Xilinx uses to set it, and if you were
a large government you may be able to coerce Xilinx into giving
you those codes plus some tool that could set the fuses
plus some unfused devices (the fuses are burn-once) but....
- do not try to create root keys if we just did an SoC update.
that's just too many things going on, we should definitely force
a clean reboot first.
- add a clean reboot step after a no-key soc update
- try to switch USB debug mode on if a backup was created
the loader is a little funky, and requires some maintainer hand-holding
due to the split of the page ownership between the root-keys
crate and the graphics-server crate.
Improved docs so I don't forget this detail in the future.
This macro will be removed, since it is no longer necessary. Remove this
macro across the project and replace it with a standard `main()`
function.
Signed-off-by: Sean Cross <sean@xobs.io>
A lot of packages are depending on 0.6 api, which requires a crate
that has OS specific TRNG calls. This patch keeps the 0.5 calls
inside the TRNG crate as direct API calls, and migrates to support
a 0.6 api to a patched version of getrandom that supports Xous.
change the notification modal to accept the primary message and an
Option of independent text to display as a qrcode
pub fn show_notification(&self, notification: &str, qrtext:Option<&str>) -> Result<(), xous::Error> {}
saves ~64k of runtime, and compilations are faster.
downside is that the gateware patching process seems to run a little bit
slower, perhaps since the loads aren't fully unrolled into individual
instructions.
the actual transition to a new version isn't here. But,
at the very least, the root keys are now derived by hashing
them MAX_ROLLBACK_LIMIT times, which allows us to create a
key chain based on a global version number that can be incremented
while allowing derviation of older keys without compromising newer
keys.
Was doing disjoint loads and stores on an Atomic, really, should
be using the fetch_add, fetch_sub variants to be proper.
Didn't seem to impact anything but maybe there's a future version
of me that high fives me for doing this.
this prevents us from having to use the entire reserved region
as a worst-case and improves screen-up time on resume.
previously, we didn't encode a signed length in the kernel so we couldn't
trust the length that was presented to us. This was fixed a while ago
but i just connected the dots today that this needed to be updated as well.
- code is merged into shellchat so there's less alignment overhead
and some library reuse
- server is not spawned until it's needed, so the RAM load is less
Saves a bit on resources and the RTC crate doesn't need
to be stand-alone now that the modals crate is a thing;
we are able to split the UX routine to manage the RTC out
of the code that actually manages the hardware.
The hardware code ended up in LLIO, and the UX ended up in the status
routine.
Each Cargo.toml now includes a comment which remnids maintainers
of the dependency policy for Xous (that is, nail down every dependency
to the minor number).
Versions are pinned down to the minor number. Yes, this means that
we don't automatically pull in features and bug updates from crates.io,
but it also means we don't pull in new attack surfaces and code that
hasn't been tested or reviewed. I'd like this process to be very deliberate
and not automatic for Xous.
The `rand` crate is almost pinned down to a single version, except
for an odd dependency that's pulled in by hosted mode that depends
on rand 0.8.4 via a subordinate dependency via the `tempfile` crate.
Pondering if it's worth it to try and eliminate that, because the
0.8.4 version doesn't get compiled into any version that touches
real hardware. It just makes it harder to analyze the Cargo.lock
file, mostly. I think I'll leave it as-is.
- check the password with a double-entry on the first boot
- if we see a blank disk, bypass the usual checks and allow the mount
to fail, then, enter the setup routine.
This is one of the truly redeeming features of Rust is that a
major refactor of a core structure like this doesn't leave me
shaking in my boots -- I feel pretty good that this works
since everything passes the CI tests.
The original FS implementation did the "stupid C thing" which
just creates a fixed lenth field and uses null-terminated C
strings. Which actually restricts PDDB entry names to be ascii
as valid UTF-8 can encode single-byte nulls.
The intention was for names to be fully utf-8 compliant, but in
order to do that we have to burn one byte of potential filename
length and encode an overall length field for the name strings.
This change does that, which both makes the code a bit safer
at the expense of losing a character off of the maximum filename
length.
The rootkey enclave actually doesn't check if your password is
correct -- it has no way of knowing because it doesn't know how
the keys will be used. Password checking is delegated to the
requesting process because also the password retry method and
policy depends upon the application, and is a significant amount
of UX plumbing (in fact the rootkeys should have less UX plumbing
than it has right now - this is something to do in the near future).
Add a mechanism in the PDDB to check if a password was entered
correctly, and retry if it isn't. Without this, an incorrect
password would actually trigger a reformat of the PDDB, since
it can't find the system basis structure with a bad password.
This is going to require users to update their devices before
doing a key initialization.
The problem is that we can't trust the length of data presented
to us to self-sign, so we sign the whole area. In v0.9.2 of the
loader, a check was introduced to check the putative length of
the signed area against the actual length of the signed area,
by encoding the length of the signed area within the signed
region.
This would require the rootkeys algorithm to also encode that
new length at the end of the signed area, but this was never
done. However, this bug was never found during testing probably
because I failed to completely erase the original key provisioning
(I was more fixated on checking that existing keys were not damaged by
firmware updates at the time), and thus did not trigger a full
rootkey provisioning cycle when I did the test. I had only just
caught it now while porting the PDDB from simulation to real
hardware.
- hosted mode uses an awful, horrific hack to simulate FLASH:
it makes a global singleton that is freely mutable. Oh but wait!
that's actually how the hardware behaves. It's up to the code
around the hardware to create those primitives. So, maybe it makes
sense that the simulation model for the hardware is terribly
unsafe and unprotected.
- the format routine at least doesn't crash and creates some
semblance of a result that looks right-ish. Need to write a
utility that helps to explore the "disk image" so we can
understand what's going on.
This is a "no good choice" situation but there is a problem with
progress bars.
You want progress bars to update quickly, and you don't want
to have to context switch out for the entire graphics stack to
come back to you to approve the drawing of your progress bar update.
On the other hand, you don't want to short-circuit all the checks
on drawing stuff to the screen for the sake of progress bars --
right now, there are a number of checks in place to make sure you
can't draw stuff on parts of the screen you shouldn't, to avoid
fake messages and dialog boxes.
I had managed to work out a way to get a dialog box to redraw without
violating the security rules -- each redraw is checked by the defacing
routine -- however, due to the way I was computing the height of
the modal I was causing the dialog box to be resized to its largest
size and then shrunk again to fit the text every iteration. This
in essence also caused the defacement routine to run repeatedly,
which caused the GAM to also issue repeated redraw requests to
the main loop.
These redraw requests can't be serviced until the task being tracked
by the progress bar is completed. As the progress bar gets redrawn
almost a hundred times as part of the task, roughly a hundred messages
arrive in the server's queue before it can handle a single one.
This results in a deadlock situation on completion of the progress
bar, because we want to fire off a message to ourselves to go to
the next state in the Ux routine.
There are a few possible solutions to this:
1. Multi thread the heck out of the routine doing the progress bar,
so the dialog box and the slider are each in their own threads
with their own responders and then the thing doing the computation
is doing some Mutex-guarded-Arc thing.
2. Do some very unsafe bodge to allow sliders to do direct draws
that are unchecked
3. Add a check so that if the size of the dialog box doesn't change,
defacement isn't recomputed and thus redraw messages aren't sent back.
(1) doesn't work for a number of reasons -- in a general case, it's
probably OK, but it does cause a lot more memory usage (and we are
lean on memory); but more importantly, for this use case, our routine
is interacting with CSRs and those don't implement Send so you can't
pass them into a thread via a closure. You could maybe put the dialog
boxes themselves into an Arc Mutex thing but because this routine
was heavily coded to re-use the heavyweight Ux resources this becomes
a bit cumersome to code.
(2) is terrible
(3) ended up working and what is in this commit. I think it's OK
because a progress bar generally doesn't require changing the
size of a window to update its state; therefore the redraw is
deface-safe (since you aren't changing the bounds of what you were
already approved to draw in). Of course, if you have some sort
of a progress bar implementation that changes the size of the messages
reported up top so that the dialog box has to change size over and
over again, you still end up in the deadlock situation!
The function `as_slice_mut()` takes a `MemoryRange` and returns a
mutable slice. Previously, this could be called on a memory range that
was not defined as `mut`.
This will be fixed in `xous-rs` so that `as_slice_mut()` can only be
called on a region that is defined as `mut`.
Signed-off-by: Sean Cross <sean@xobs.io>
This code was originally in https://github.com/betrusted-io/rom-locate.
However, the majority of that crate is concerned with creating the database
files necessary for the patching, and it pulls in a lot of dependencies.
Changing hardware targets is considered to be a "huge" change for this
project, so we're pushing the code generator directly into xous-core,
and commiting the generated code directly. The theory is that we would
probably never have to modify this code, since it's so intimately linked
to the hardware; it's a bit like committing an include file with
register offsets for a particular peripheral.
also add some bread crumbs on the more user-facing APIs to
help beginners out. Need to do a good spring cleaning of the
rustdoc tree, though. looks pretty terrible as it is.