taildrive: writes do not arrive on a writeable mount

[pitust]

What is the issue?

When writing to files in taildrive (davfs 1.7.0-2, tailscale 1.66.1) the writes do not ever appear on the remote.

Steps to reproduce

1. Share a folder with taildrive
2. Mount the folder with sudo mount -t davfs -o uid=1000,gid=1000 http://100.100.100.100:8080 /mnt/taildrive
3. Create a file with nonempty contents, e.g. echo hi > /mnt/taildrive/<path>/foo
4. Observe that these writes do not propagate to the remote although the file itself is created.
5. After some time, observe the file contents disappearing with an error: taildrive: access: PUT from [hFhH/] to [UQBjF]: status-code=412 ext="unknown" content-type="unknown" content-length=20 tx=5 rx=0 err=""

The entire mechanism in general seems to love traveling back in time, although it is unclear if it is truly tailscale's fault.

Directly writing with -X PUT does appear to work, however; this seems to be tied to the WebDAV If header which is not handler correctly on the server.

Are there any recent changes that introduced the issue?

No response

OS

Linux

OS version

ArchLinux

Tailscale version

1.66.1

Other software

davfs 1.7.0-2

Bug report

BUG-fd8a885f0f6ff803bba155080ca2e2a87503df9e2374760caf3cbe1cebc6792c-20240510210525Z-6bc6c5b062227591

[agottardo]

cc @oxtoacart

[oxtoacart]

I'm not able to reproduce this on the HEAD of main. @pitust I see that your client is Linux with davfs2. What platform is your server running on?

[pitust]

@oxtoacart Synology; Tailscale 1.66.0. I can confirm in pcaps that the client acquires a lock and then correctly sends it over peerapi, but the write fails on the remote end with "precondition failed".

[oxtoacart]

@pitust Ahh, sounds like there might be some goofiness with locks. Out of curiosity, if you set use_locks 0 in your davfs2.conf, does the problem go away?

[pitust]

@oxtoacart Yes, setting use_locks 0 in davfs2.conf does resolve the issue (although this is not an ideal solution for obvious reasons).

[oxtoacart]

@pitust This should be resolved on unstable now. Please let me know if it works for you.