dnsdist: Add a function to set the UDP recv/snd buffer sizes

[rgacogne]

Short description

And raise them to 16777216 by default. Although perhaps we should only do that on Linux?

This PR also refactors the functions duplicated in the recursor, calidns and dnsreplay and therefore impacts the auth and the rec.

Closes #10898.

Checklist

I have:

• read the CONTRIBUTING.md document
• compiled this code
• tested this code
• included documentation (including possible behaviour changes)
• documented the code
• added or modified regression test(s)
• added or modified unit test(s)

[phonedph1]

is this value supposed to change from before?

[rgacogne]

Not at all, thanks a lot!

[omoerbeek]

A nit for the OpenBSD case.

[omoerbeek]

This is quite greedy, on OpenBSD:

Unable to raise socket buffer size to 16777216: No buffer space available
Unable to raise socket buffer size to 16777216: No buffer space available

[rgacogne]

Right, I was afraid of something like that. I now see that the default on Linux is likely smaller than that on most distributions (212992 on x86_64, 163840 on 32 bits). So I see three options:

• a smaller default value, like the recursor does, perhaps 160k?
• default value of 0 so we keep the system default
• default value of 0 except for Linux, where we set something like 200k?

[omoerbeek]

250000 works on OpenBSD. so a default of that (or smaller) for dnsdist would work for me

[rgacogne]

On all Linux distributions I checked, rmem_default == rmem_max and wmem_default == wmem_max by default, so we would not be able to raise the value. So I guess doing nothing by default makes more sense, and the administrator can set higher values if needed.

[omoerbeek]

The failing recursor regression tests seem to fail to start the auth... ??

[github-actions]

@check-spelling-bot Report

Unrecognized words, please review:

• rmem
• wmem

To accept these unrecognized words as correct, run the following commands

... in a clone of the git@github.com:rgacogne/pdns.git repository
on the ddist-raise-socket-buffers branch:

update_files() {
perl -e '
my $new_expect_file=".github/actions/spell-check/expect.txt";
use File::Path qw(make_path);
use File::Basename qw(dirname);
make_path (dirname($new_expect_file));
open FILE, q{<}, $new_expect_file; chomp(my @words = <FILE>); close FILE;
my @add=qw('"$patch_add"');
my %items; @items{@words} = @words x (1); @items{@add} = @add x (1);
@words = sort {lc($a)."-".$a cmp lc($b)."-".$b} keys %items;
open FILE, q{>}, $new_expect_file; for my $word (@words) { print FILE "$word\n" if $word =~ /\w/; };
close FILE;
system("git", "add", $new_expect_file);
'
}

comment_json=$(mktemp)
curl -L -s -S \
  --header "Content-Type: application/json" \
  "https://api.github.com/repos/PowerDNS/pdns/issues/comments/974104137" > "$comment_json"
comment_body=$(mktemp)
jq -r .body < "$comment_json" > $comment_body
rm $comment_json

patch_add=$(perl -e '$/=undef;
$_=<>;
s{<details>.*}{}s;
s{^#.*}{};
s{\n##.*}{};
s{(?:^|\n)\s*\*}{}g;
s{\s+}{ }g;
print' < "$comment_body")
  
update_files
rm $comment_body
git add -u

If the flagged items do not appear to be text

If items relate to a ...

• well-formed pattern.

  If you can write a pattern that would match it,
  try adding it to the patterns.txt file.

  Patterns are Perl 5 Regular Expressions - you can test yours before committing to verify it will match your lines.

  Note that patterns can't match multiline strings.

• binary file.

  Please add a file path to the excludes.txt file matching the containing file.

  File paths are Perl 5 Regular Expressions - you can test yours before committing to verify it will match your files.

  ^ refers to the file's path from the root of the repository, so ^README\.md$ would exclude README.md (on whichever branch you're using).

[omoerbeek]

Now wondering why rec only sets the receive buffer size and not the send buffer size. A peak in incoming request normally also leaks to a peak in replies I would say. But that's a different issue.

[rgacogne]

Now wondering why rec only sets the receive buffer size and not the send buffer size. A peak in incoming request normally also leaks to a peak in replies I would say.

Agreed. The good news is that raising net.core.wmem_default is enough to fix that, but it would be nice to do it by default.