lib/knot-resolver/sandbox.lua:399: can't open cache path '.';

with version 5.0.1 on FreeBSD 12.1 I get the following when starting kresd

kresd -n -c /usr/local/etc/knot-resolver/kresd.conf /var/db/kresd
[system] error while loading config: /usr/local/lib/knot-resolver/sandbox.lua:399: can't open cache path '.'; working directory '/var/db/kresd'; Invalid argument (workdir '/var/db/kresd')

the LMDB files are created as specified with cache.size but then kresd gives up with the above error.

That's a runtime error indicating kresd wasn't able to open the cache at the specified location.

Please check the permissions/ownership of /var/db/kresd or try re-creating the directory.

EDIT: Nevermind, a permission error would look differently - it would have the Invalid argument message.

Permission errors should be throwing different messages (not "Invalid argument"); I confirmed that on FreeBSD 12.1 in a VM.

We suspect something is wrong with cache pre-allocation. On the test system (above) I can't reproduce an error but inspecting the file indicates preallocation didn't really work out. It seems possible for a filesystem (on FreeBSD) not to support that operation; @yds: what FS is on your /var/db/kresd ?

/var/db/kresd is on ZFS on FreeBSD 12.1-STABLE

kresd.conf drops privileges to user('kresd', 'kresd')

rc prestart routine ensures that /var/db/kresd exists and owned by kresd:kresd.

FWIW, data.mdb and lock.mdb are being created as specified by cache.size before exiting with the above error.

Side note: on encrypted ZFS there was a kres-cache-gc inefficiency that should be resolved but not yet in 5.0.1: https://gitlab.labs.nic.cz/knot/knot-resolver/merge_requests/946

@yds: can you test a simple patch? https://gitlab.labs.nic.cz/knot/knot-resolver/commit/5a649214ec.patch

# kresd -n -c /usr/local/etc/knot-resolver/kresd.conf /var/db/kresd
[cache] space pre-allocation failed and ignored; your (file)system probably doesn't support it.
[ta_update] refreshing TA for .
[ta_update] key: 20326 state: Valid
[ta_update] next refresh for . in 24 hours

5a649214ec.patch seems to make it work.. thx! sockstat:

kresd    kresd      61603 12 udp4   127.0.0.1:53          *:*
kresd    kresd      61603 13 tcp4   127.0.0.1:53          *:*
kresd    kresd      61603 15 tcp4   127.0.0.1:853         *:*
kresd    kresd      61603 16 udp6   ::1:53                *:*
kresd    kresd      61603 17 tcp6   ::1:53                *:*
kresd    kresd      61603 18 tcp6   ::1:853               *:*
kresd    kresd      61603 19 stream /var/db/kresd/control/61603

mentioned in merge request !949 (merged)

I created a patch to fix the many things broken with the FreeBSD port including this bug: https://Bugs.FreeBSD.org/bugzilla/show_bug.cgi?id=244600

if someone here maintains the port, please consider it, otherwise the FreeBSD port is kinda broken since the permissions are wrong and there's no rc scripts included to start the service.

Noone from our team really maintains it, but I'll have a look anyway.

BTW, after building with 5a649214ec.patch running kres-cache-gc outputs:

[cache] space pre-allocation failed and ignored; your (file)system probably doesn't support it.

over and over about once per second.

Right. Fortunately with https://gitlab.labs.nic.cz/knot/knot-resolver/merge_requests/946 it shouldn't repeat anymore, as cache won't be re-open every interval but only on each GC pass.

Maybe we'll tweak the message anyway, as pre-allocation seems more broken on FreeBSD than I anticipated: https://gitlab.labs.nic.cz/knot/knot-resolver/merge_requests/949#note_144785

closed via merge request !949 (merged)

mentioned in commit ec618f5a