Skip to content
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

Remove linq #1949

Merged
merged 11 commits into from Mar 24, 2023
Merged

Remove linq #1949

merged 11 commits into from Mar 24, 2023

Conversation

Wraith2
Copy link
Contributor

@Wraith2 Wraith2 commented Mar 8, 2023

helps with #1942

Removes all uses of linq operations in the main binary and addons allowing linq to be trimmed out if no other libraries are using it
I suggest reviewing 1 commit at a time. Some of the enclave stuff is a bit messy and this cleans it up as well as removing the linq.

/cc @kant2002

@Wraith2 Wraith2 force-pushed the remove-linq branch 2 times, most recently from 7e1415d to 981104f Compare March 9, 2023 00:04
@Wraith2
Copy link
Contributor Author

Wraith2 commented Mar 9, 2023

@David-Engel

Is there someone knowledgeable about enclave stuff in the codebase that could help here. I've found something odd that concerns me.

There's a piece of code that was using data.Skip(offset).Take(count) and i replaced it with a function call that does the same thing without using linq. My replacement differs in behaviour by requiring that you be able to copy exactly count bytes where the linq Take function will only copy up-to count bytes, if there aren't enough it won't fail.

This caused failures in this code:

int secureSessionBufferSize = Convert.ToInt32(secureSessionInfoResponseSize) - sizeof(uint);
byte[] secureSessionBuffer = attestationInfo.Skip(offset).Take(secureSessionBufferSize).ToArray();
EnclaveDHInfo = new EnclaveDiffieHellmanInfo(secureSessionBuffer);
offset += Convert.ToInt32(EnclaveDHInfo.Size);

if (secureSessionBufferSize != secureSessionBuffer.Length)
{
	throw new InvalidOperationException($"enclave data mismatch, expected {secureSessionBufferSize} got {EnclaveDHInfo.Size}");
}

Where we calculate how many bytes of data we expect to get and then try to take it and end up 4 bytes short. Either the calculation of how many bytes we should expect to get is wrong or the something else is going on.

@codecov
Copy link

codecov bot commented Mar 9, 2023

Codecov Report

Patch coverage: 71.98% and project coverage change: -0.10 ⚠️

Comparison is base (1bcdb23) 70.54% compared to head (b3ddcb0) 70.44%.

Additional details and impacted files
@@            Coverage Diff             @@
##             main    #1949      +/-   ##
==========================================
- Coverage   70.54%   70.44%   -0.10%     
==========================================
  Files         306      306              
  Lines       61563    61657      +94     
==========================================
+ Hits        43429    43436       +7     
- Misses      18134    18221      +87     
Flag Coverage Δ
addons 92.88% <100.00%> (+0.49%) ⬆️
netcore 73.21% <65.62%> (-0.09%) ⬇️
netfx 69.06% <72.34%> (-0.11%) ⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files Coverage Δ
.../Microsoft/Data/SqlClient/SNI/SNIPhysicalHandle.cs 100.00% <ø> (ø)
...fx/src/Microsoft/Data/Common/DBConnectionString.cs 12.96% <ø> (ø)
...c/Microsoft/Data/SqlClient/TdsParserStateObject.cs 65.13% <ø> (-0.21%) ⬇️
...c/Microsoft/Data/SqlClient/SqlSymmetricKeyCache.cs 100.00% <ø> (ø)
...ent/SqlAuthenticationProviderManager.NetCoreApp.cs 30.88% <14.28%> (-1.48%) ⬇️
...Data/SqlClient/SqlAuthenticationProviderManager.cs 49.03% <20.00%> (-0.97%) ⬇️
...Data/SqlClient/SqlAuthenticationProviderManager.cs 61.66% <22.22%> (+0.64%) ⬆️
...nt/netcore/src/Microsoft/Data/SqlClient/SqlUtil.cs 38.13% <35.00%> (-0.12%) ⬇️
...lClient/Reliability/SqlConfigurableRetryFactory.cs 93.54% <42.85%> (-6.46%) ⬇️
.../SqlClient/VirtualSecureModeEnclaveProviderBase.cs 81.54% <43.75%> (-4.17%) ⬇️
... and 15 more

... and 10 files with indirect coverage changes

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

ipv4Addresses = Array.Empty<IPAddress>();
ipv6Addresses = Array.Empty<IPAddress>();

if (input != null && input.Length > 0)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe revert this condition and have fast exit? Code above looks a bit unnatural with this condition

ipv4Addresses = Array.Empty<IPAddress>();
ipv6Addresses = Array.Empty<IPAddress>();

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's like this because they're out parameters. I suppose it's a personal style choice but I always assign the default values for out parameters at the start of the method so that anywhere inside the body I can easily return without worrying about whether they're assigned or not.

@roji
Copy link
Member

roji commented Mar 9, 2023

@Wraith2 what kind of size reduction are you seeing with this?

@Wraith2
Copy link
Contributor Author

Wraith2 commented Mar 9, 2023

I haven't measured. I also think it's unlikely to help if sqlclient is used through EFCore because there's no way that linq is getting trimmed from EFCore. I just saw the list that @kant2002 posted in the aot thread and thought I'd see what I could so with one of the easier items.

@roji
Copy link
Member

roji commented Mar 9, 2023

I also think it's unlikely to help if sqlclient is used through EFCore because there's no way that linq is getting trimmed from EFCore.

I'm sure that's true; EF hasn't yet undergone a real trimming effort - it's possible (though not sure) that when we do that in a serious way, LINQ wouldn't be needed.

But I'd assume that various dependencies of SqlClient are also bringing in LINQ (e.g. the Azure Identity libs). Personally, for cases where removing LINQ makes the code less maintainable, I'd avoid doing that until I see it provides a tangible gain (but I also think there are at least some cases in this PR where the code doesn't become less maintainable).

@Wraith2
Copy link
Contributor Author

Wraith2 commented Mar 9, 2023

Agreed. I was talking with kant on discord about this and it's possible that the azure libs will be trimmable in future if you can work out the api to do so.

If a total removal isn't appropriate certain parts of this can be backed out. There's plenty of scope for discussion and this is only a draft.

@roji
Copy link
Member

roji commented Mar 9, 2023

I was talking with kant on discord about this and it's possible that the azure libs will be trimmable in future if you can work out the api to do so.

I'd propose thinking about the approach we've taken in Npgsql (#1942 (comment)): a "slim" data source builder which, when used, doesn't have size-expensive functionality on by default, unless explicitly opted into.

However, specifically regarding the Azure libs, the very fact that SqlClient has a direct references on them is probably not ideal (#1108, highest-voted issue on the repo). Here specifically, I'd suggest SqlClient expose an auth API that allows 3rd-party packages (like the Azure ones) to register themselves. This would totally decouple SqlClient from Azure, and of course naturally strip them away size-wise unless used.

@Wraith2
Copy link
Contributor Author

Wraith2 commented Mar 9, 2023

As long as backwards compatibility is maintained adding new apis around the DataSource abstraction seems fine to me. That could allow both advanced AE and Enclave functionality as well as extended authentication mechanisms to be trimmable.

@David-Engel
Copy link
Contributor

Is there someone knowledgeable about enclave stuff in the codebase that could help here. I've found something odd that concerns me.

There's a piece of code that was using data.Skip(offset).Take(count) and i replaced it with a function call that does the same thing without using linq. My replacement differs in behaviour by requiring that you be able to copy exactly count bytes where the linq Take function will only copy up-to count bytes, if there aren't enough it won't fail.

This caused failures in this code:

int secureSessionBufferSize = Convert.ToInt32(secureSessionInfoResponseSize) - sizeof(uint);
byte[] secureSessionBuffer = attestationInfo.Skip(offset).Take(secureSessionBufferSize).ToArray();
EnclaveDHInfo = new EnclaveDiffieHellmanInfo(secureSessionBuffer);
offset += Convert.ToInt32(EnclaveDHInfo.Size);

if (secureSessionBufferSize != secureSessionBuffer.Length)
{
	throw new InvalidOperationException($"enclave data mismatch, expected {secureSessionBufferSize} got {EnclaveDHInfo.Size}");
}

Where we calculate how many bytes of data we expect to get and then try to take it and end up 4 bytes short. Either the calculation of how many bytes we should expect to get is wrong or the something else is going on.

@Wraith2

This doesn't fail outside of your PR, right? Wouldn't the original exception throw if there was some underlying issue with the original code? I've been looking at the code changes and I'd have to debug it to get a better idea of what might be happening but it feels like something is missing. I need to find an enclave environment set up to debug against. The team has just been horribly busy with support issues for the last month or so...

@Wraith2
Copy link
Contributor Author

Wraith2 commented Mar 10, 2023

It isn't a failure outside my PR. I think because while the length calculations are incorrect it isn't meaningful and the code actually does the right thing.

This is the original code:

                    uint secureSessionInfoResponseSize = BitConverter.ToUInt32(attestationInfo, offset);
                    offset += sizeof(uint);

                    SessionId = BitConverter.ToInt64(attestationInfo, offset);
                    offset += sizeof(long);

                    int secureSessionBufferSize = Convert.ToInt32(secureSessionInfoResponseSize) - sizeof(uint);
                    byte[] secureSessionBuffer = attestationInfo.Skip(offset).Take(secureSessionBufferSize).ToArray();
                    EnclaveDHInfo = new EnclaveDiffieHellmanInfo(secureSessionBuffer);
                    offset += Convert.ToInt32(EnclaveDHInfo.Size);

I think the sessionId which is read as sizeof(long) is incorrectly subtracted from the remaining bytes as sizeof(uint) but since we've actually read 8 bytes and advanced by 8 bytes we're just calculating the remaining length incorrectly. When we do Skip(offset).Take(length) the length is 4 bytes longer than we have available but the offset is correct and the lazy behaviour of Take causes it to turn out correct.

So if the code is working should I change it? I think so. I think it should be changed because the comment above the function we're looking at doesn't accurately express what the code is doing and because the code is working accidentally. As I've proven if you try to change it you risk breaking it by interpreting the code and not carefully inspecting the results in a debugger. Since the only way to debug this is complex and requires enclave functionality to be available I think code would be better.

@David-Engel
Copy link
Contributor

@Wraith2

Looking at the JDBC implementation, I think the existing code is just failing to account for an additional size field after the Session ID, which is not accounted for in the secureSessionInfoResponseSize when calculating the secureSessionBufferSize.
JDBC reference

You're right. It's an error, but one that fortunately works itself out due to the Take() behavior.

int secureSessionBufferSize = Convert.ToInt32(secureSessionInfoResponseSize) - sizeof(uint);

should be

// Account for 4 byte size of enclave ECDH public key followed by 4 byte size of the DH public key signature
int secureSessionBufferSize = Convert.ToInt32(secureSessionInfoResponseSize) - sizeof(uint) - sizeof(uint);

I still haven't debugged, so let me know if that works out correctly. 😄

@Wraith2 Wraith2 marked this pull request as ready for review March 11, 2023 01:41
@Wraith2
Copy link
Contributor Author

Wraith2 commented Mar 11, 2023

I revised the code but the CI is apparently off for the weekend already.

@Wraith2 Wraith2 marked this pull request as draft March 11, 2023 11:59
@Wraith2 Wraith2 marked this pull request as ready for review March 13, 2023 21:26
@lcheunglci lcheunglci closed this Mar 14, 2023
@lcheunglci lcheunglci reopened this Mar 14, 2023
@DavoudEshtehari DavoudEshtehari added the ➕ Code Health Changes related to source code improvements label Mar 15, 2023
Copy link
Contributor

@David-Engel David-Engel left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good to me. This takes us in a good direction with regard to trimming and potential AoT support. Thanks for the PR!

@lcheunglci
Copy link
Contributor

We just merged #1943 and it created a small merge conflict with the namespace. Let us know if you would like us to push changes to your PR branch or if you have moment to fix it. Thanks!

@Wraith2
Copy link
Contributor Author

Wraith2 commented Mar 21, 2023

conflicts resolved and pushed.

switch (ipPreference)
{
case SqlConnectionIPAddressPreference.IPv4First:
{
SsrpResult response4 = SendUDPRequest(ipAddresses.Where(i => i.AddressFamily == AddressFamily.InterNetwork).ToArray(), port, requestPacket, allIPsInParallel);
SplitIPv4AndIPv6(ipv4Addresses, out ipv4Addresses, out ipv6Addresses);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
SplitIPv4AndIPv6(ipv4Addresses, out ipv4Addresses, out ipv6Addresses);
SplitIPv4AndIPv6(ipAddresses, out ipv4Addresses, out ipv6Addresses);

I'm assuming you're passing in the ipAddresses from line 192 instead of ipv4Addresses which is null on line 194; otherwise, it would always be Array.Empty<IPAddress>().

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good catch. It is concerning that the CI didn't fail on this error.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I agree. I was wondering about that as well. It turns out since this code is only run on Linux and requires sudo/administrative privileges to get the correct environment, there's currently no test coverage for this section unfortunately.

Copy link
Contributor

@lcheunglci lcheunglci left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, Thanks @Wraith2 for your contributions!

@David-Engel David-Engel merged commit 23ca8e7 into dotnet:main Mar 24, 2023
131 checks passed
@Wraith2 Wraith2 deleted the remove-linq branch April 9, 2023 20:59
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
➕ Code Health Changes related to source code improvements
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

6 participants