This change actually landed upstream in v7.1: commit 8341c989ac77d712c7d6e2bce29e8a4bcb2eeae4 Author: Eric Dumazet <[email protected]> Date: Fri Feb 27 15:11:20 2026 +0000 net: remove addr_len argu…

> So not all cipher algorithms have 16 bytes long long authentication tag lengths. But with any multiple of 4 of four bytes I guess this is unfortunate consequence of providing 32-bit aligned start l…

OK, Great. This means I will abandon this PR. Are you fine with this? Yes.

Do we really need a new struct for it? We can pass a buf to getsockopt, both userspace and kernel will parse it as uint8_t (*)[2], and the optlen pointer in getsockopt() will tell userspace how many …

Do we really need a new struct for it? We can pass a buf to getsockopt, both userspace and kernel will parse it as uint8_t (*)[2], and the optlen pointer in getsockopt() will tell userspace how many …

Do we really need a new struct for it? We can pass a buf to getsockopt, both userspace and kernel will parse it as uint8_t (*)[2], and the optlen pointer in getsockopt() will tell userspace how many …

Do we really need a new struct for it? We can pass a buf to getsockopt, both userspace and kernel will parse it as uint8_t (*)[2], and the optlen pointer in getsockopt() will tell userspace how many …

But sctp_dtls_cipher_suites() must be implemented with a syscall like getsockopt() to get what SUITS are supported in kernel. You don't need to implement it as a syscall. You can implement a socket o…

But sctp_dtls_cipher_suites() must be implemented with a syscall like getsockopt() to get what SUITS are supported in kernel.

I think I addressed all issues related to the socket API from our discussion last Thursday. @lxin : Can you double check, please? Yes, all are addressed. LGTM. Another thing, while aligning the code …

@tuexen There's another struct that needs padding: struct sctp_dtls_keys { sctp_assoc_t sdk_assoc_id; uint8_t sdk_cipher_suite[2]; uint8_t sdk_restart; uint8_t sdk_unused; /* if sizeof(sctp_assoc_t) …

@lxin Could you take a look? I think config is more appropriate than parameters. @tuexen sorry for late, it's been hectic here. I just got a chance to implement it today. 'config' looks good to me. B…

I don't see a reason for -ENOSPC, maybe you can do that only if the caller is using QUIC_STREAM_INFO? That's doable. @harlequix , can you verify the fix in #71 in your env? Thanks.

Build quic.ko from lxin/quic at commit f0008fd, load the module, start tlshd. @harlequix btw, why did it need to start tlshd? afaik, samba is running in userspace, unlike SMB.

Hmm, looking just at the error mapping [1] EWOULDBLOCK might be a candidate. They are the only ones which map NT_STATUS_NETWORK_BUSY which matches what is going on in the socket itself. Anything else…

@harlequix, thanks for reporting this. With -EAGAIN returned, users cannot distinguish whether the send was blocked due to stream flow control or congestion control / send buffer limitations. Could y…

@X1AOxiang, Thanks for reporting this. PATH_CHALLENGE Datagrams Not Expanded to 1200 Bytes Looks like this commit need to be revert: 99c0a9f No RETIRE_CONNECTION_ID Sent for Newly Received CID Below …

@metze-samba I’ll apply this PR for now. Once your patch is upstream, I’ll switch to using it. Note that the related code will be part of the second series, not the first series that I’m posting and …