netdev - Re: [RFC PATCH bpf-next 2/3] bpf: udp: Avoid socket skips and repeats during iteration

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CABi4-ogUtMrH8-NVB6W8Xg_F_KDLq=yy-yu-tKr2udXE2Mu1Lg@mail.gmail.com>
Date: Mon, 7 Apr 2025 19:39:44 -0700
From: Jordan Rife <jordan@...fe.io>
To: Kuniyuki Iwashima <kuniyu@...zon.com>
Cc: aditi.ghag@...valent.com, bpf@...r.kernel.org, daniel@...earbox.net, 
	martin.lau@...ux.dev, netdev@...r.kernel.org, willemdebruijn.kernel@...il.com
Subject: Re: [RFC PATCH bpf-next 2/3] bpf: udp: Avoid socket skips and repeats
 during iteration

> In the worst case, where vmalloc() fails and the batch does not
> cover full bucket, say the batch size is 16 but the list length
> is 256, if the iterator stops at sk15 and sk16 disappers,
> sk17 ~ sk256 will be skipped in the next iteration.
>
> sk1 -> ... sk15 -> sk16 -> sk17 -> ... -> sk256

Ah yes, this is true. Thank you for clarifying, you bring up a good
point. In case vmalloc() fails, the batch size can't cover the whole
bucket in one go, and none of the saved cookies from last time are in
the bucket, there's currently no great option. You'd need to do one of
the following:

1) Start from the beginning of the list, assuming none of the sockets
had been seen so far. This risks repeating sockets you've already
seen, however.
2) Skip the rest of the sockets to avoid repeating sockets you've
already seen. You might skip sockets that you didn't want to skip.

I actually wonder if a third option might be better in this case though:

3) If vmalloc fails, propagate ENOMEM up to userspace and stop
iteration instead of making the tradeoff of possibly repeating or
skipping sockets. seq_read can already return ENOMEM in some cases, so
IMO this feels more correct. WDYT?

-Jordan