-
Notifications
You must be signed in to change notification settings - Fork 182
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add in a RtlGenRandom fallback for non-UWP Windows #337
Conversation
Builds and tests on my Windows machine, fwiw. Gonna try patching Firefox and see where that leaves us. |
b3cea33
to
ab8c2d5
Compare
(Whoops. Built and tested it on my Windows machine within WSL which isn't at all Windows. Lemme just tweak and rebuild that real quick.) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you @chutten for working on this! Here are my remarks on the Windows aspects of this change.
src/windows.rs
Outdated
// system does not have a way to express this yet. | ||
let code = unsafe { NonZeroU32::new_unchecked(code) }; | ||
return Err(Error::from(code)); | ||
if ret > 0 { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
if ret > 0
is not properly checking for failure of BCryptGenRandom
here. NTSTATUS
is a signed 32-bit value and the two most significant bits encode whether it is a success, information, warning, or error. Checking for success is typically "success or information" (which translates to if ret >= 0
) while failure is typically "warning or error" (which translates to if ret < 0
).
Be careful that the code before this patch has defined the return value from BCryptGenRandom
as u32
instead of i32
, which I would say is incorrect (wrt C definitions) or at best confusing (but this is what allowed ret >> 30
to compare with 0b11
instead of -1
, I guess). So unless you also change that u32
to i32
(which I would recommend), my if ret < 0
suggestion would actually not work written like that here.
For more details check https://learn.microsoft.com/en-us/windows-hardware/drivers/kernel/using-ntstatus-values and/or https://searchfox.org/mozilla-central/source/__GENERATED__/__RUST_STDLIB__/std/src/sys/windows/c.rs#292-296 .
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Ah, and here I thought I was clever by adapting rust-lang's nt_success... but without thinking about it first. Lemme give this another go, paying more close attention to the types and values (that's what I get for rushing it at the end of my day yesterday)
src/windows.rs
Outdated
|
||
#[cfg(target_vendor = "uwp")] | ||
#[inline(never)] | ||
fn fallback_rng(_chunk: &mut [MaybeUninit<u8>]) -> u8 { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The return type does not seem to match the return value here? RtlGenRandom
return no more information than TRUE or FALSE (respectively a non-zero or a zero u8
value), so I think fallback_rng
could return bool
or Maybe<bool>
depending on whether or not we want to be able to distinguish "RtlGenRandom
is unavailable" from "RtlGenRandom
failed". Given that at the moment we do not need to be able to distinguish since the error we propagate will be the one from BCryptGenRandom
, maybe bool
is enough here.
Note: TRUE in RtlGenRandom
means success so its return value is currently incorrectly checked on line 46.
src/windows.rs
Outdated
// We zeroize the highest bit, so the error code will reside | ||
// inside the range designated for OS codes. | ||
let code = ret ^ (1 << 31); | ||
// SAFETY: the second highest bit is always equal to one, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The code before this patch considered that warning values for NTSTATUS are not a failure (that's why it was checking for the two most significant bits of the return value from BCryptGenRandom
). This is what this remark explains (although I think it's a mistake to consider that warning values are not a failure).
If in this patch we consider that warning values for NTSTATUS are a failure (and I think we should), then this remark does not hold anymore and should be updated. But I think we can still use NonZeroU32::new_unchecked
, because NTSTATUS
warning values start with 0x80000001 STATUS_GUARD_PAGE_VIOLATION, so there should be no 0x80000000. Or, for extra safety guarantees, we could map a theoretical 0x80000000 to 0xC0000001 STATUS_UNSUCCESSFUL.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm a fan of patches doing one thing at a time, so I'd prefer to restrain ourselves to not having opinions about the return values (like I accidentally did). If we wish to propose a change so warnings are failures as well (now that we have a fallback to attempt), I feel like that ought to be its own Issue and PR.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
OK, good for me then.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I agree that our handling of error values here is not ideal. We should mimic what NT_SUCCESS(Status)
does and only treat values 0 − 0x7FFFFFFF
as "succeeding". However, this is better addressed in a followup PR.
src/windows.rs
Outdated
if ret > 0 { | ||
// Failed. Try RtlGenRandom as a fallback. | ||
let fallback_ret = fallback_rng(chunk); | ||
// NTSTATUS codes use the two highest bits for severity status. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This comment is outdated / does not match with what the code does.
ab8c2d5
to
16250fd
Compare
And to complete the see also links, see also: https://bugzilla.mozilla.org/show_bug.cgi?id=1816953 |
src/windows.rs
Outdated
@@ -21,6 +21,12 @@ extern "system" { | |||
) -> u32; | |||
} | |||
|
|||
extern "system" { | |||
// Forbidden when targetting UWP |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Instead of a comment, we should have a #[cfg(not(target_vendor = "uwp"))]
attribute. Something like:
#[link(name = "bcrypt")]
extern "system" {
fn BCryptGenRandom(
hAlgorithm: *mut c_void,
pBuffer: *mut u8,
cbBuffer: u32,
dwFlags: u32,
) -> i32;
}
#[cfg(not(target_vendor = "uwp"))]
#[link(name = "advapi32")]
extern "system" {
#[link_name = "SystemFunction036"]
pub fn RtlGenRandom(RandomBuffer: *mut c_void, RandomBufferLength: u32) -> u8;
}
Note that the buffer is of type PVOID
and we need to make sure we link to Advapi32
src/windows.rs
Outdated
// Failed. Try RtlGenRandom as a fallback. | ||
let success = fallback_rng(chunk); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This introduces a larger diff and more indentation than needed. Could we just do:
// NTSTATUS codes use the two highest bits for severity status.
if ret >> 30 == 0b11 {
// Failed. Try RtlGenRandom as a fallback.
#[cfg(not(target_vendor = "uwp"))]
if unsafe { RtlGenRandom(chunk.as_mut_ptr() as *mut c_void, chunk.len() as u32) } != 0 {
continue;
}
// We zeroize the highest bit, so the error code will reside
// inside the range designated for OS codes.
let code = ret ^ (1 << 31);
...
16250fd
to
ed34e2b
Compare
Pushed to run a few tests on moz's tryservers here: https://treeherder.mozilla.org/jobs?repo=try&revision=81ff7509fa334c3bf3d2fe0fcce8ca9f2376a6a7 (I expect some oranges here and there from flaky tests, but for it to overall work out). Attn @yjugl for if you want to doublecheck that this more-final version fixes your particular testcase. |
@chutten |
Oh, because of the |
In some instances BCryptRandom will fail when RtlGenRandom will work. On UWP, we might be unable to actually use RtlGenRandom. Thread the needle and use RtlGenRandom when we have to, when we're able. See also rust-lang/rust#108060 Fixes rust-random#314
ed34e2b
to
6609385
Compare
There, that ought to do it. I pulled down 1.36 and it compiles with the extra block. (For anyone reading, I did it via |
Co-authored-by: Artyom Pavlov <newpavlov@gmail.com>
I was able to confirm that on a Windows 7 machine where |
@newpavlov @josephlr sooooo... when would it be polite to ask about your plans vis a vis cutting a v0.2.9...? : D |
Bit late to the party, but you probably shouldn't be matching on |
TIL. The specific proposed guidance for UWP will be to replace |
In some instances BCryptRandom will fail when RtlGenRandom will work. On UWP, we might be unable to actually use RtlGenRandom.
Thread the needle and use RtlGenRandom when we have to, when we're able.
See also rust-lang/rust#108060
Fixes #314