> On 18/02/2026 23:59 EET sev monster via dovecot <[email protected]> wrote:
> 
>  
> 8 months ago I messaged this list communicating some of the issues I had with 
> 2.4. Before I could even start testing 2.4 properly, 2 bugs had to be 
> fixed—one that caused crashes on startup 
> (https://gitlab.alpinelinux.org/alpine/aports/-/issues/17050), and one that 
> caused passdb to use the same caching key for all connections and as such 
> always have the same user log in. There was another bug, one that prevented 
> me from using LDAP binds for passdb, but it had a workaround that mostly 
> worked without side-eeffects. All of those have since been fixed, hurray!
> 
> So in my infinite naïveté, I decided to give 2.4 another shake. And it, 
> again, does not work. Surprise!
> 
> I did manage to massage the configuration to allow me to at least run 2.4 
> without crashing and with users able to access (most of) their mailboxes, but 
> there are still blockers to full functionality.
> 

Have you given https://dovecot.org/upgrader a try to see if it can massage your 
config better?


> I face the following issues:
> 1. Christian Pfeiffer back in November found a bug where auth caching with 
> LDAP did not work, and due to how it's implemented, even turning caching off 
> would still raise an error. I experienced the same issue for both userdb 
> iteration and passdb bind logins. I had to comment out the userdb and use a 
> static one instead (all my users, currently, use the same UID and easily 
> composable path, so this wasn't a big deal) and start using a service account 
> for LDAP binds.
> 2. Virtual mailbox autosubscribe does not work anymore. It throws an error 
> about trying to create the virtual mailbox. This worked in 2.3.
> 
> My virtual mailbox config:
> namespace virtual {
>   type = private
>   prefix = Search Folders.
>   mail_driver = virtual
>   mail_path = /etc/dovecot/virtual
>   mail_index_path = ~/dovecot-virtual.cache
>   mail_control_path = ~/dovecot-virtual.cache
>   mail_volatile_path = ~/dovecot-virtual.cache
>   hidden = yes
>   list = children
>   subscriptions = no
> 
>   mailbox All {
>     auto = subscribe
>     special_use = \All
>     comment = All messages, excluding Junk and Trash
>   }
> 
>   mailbox Unread {
>     auto = subscribe
>     special_use = \Important
>     comment = All unread messages, excluding Junk and Trash
>   }
> 
>   mailbox Flagged {
>     auto = subscribe
>     special_use = \Flagged
>     comment = All flagged messages
>   }
> }
> 
> If there is something wrong with this config that might be responsible for my 
> issue, please let me know. Otherwise, I believe this is a regression. I 
> understand autosubscribe is documented to (and in other situations with 
> non-virtual boxes) create mailboxes if missing, so if this is actually a fix 
> to make the feature work as documented, we need a new option that allows for 
> autosubscription without attempting to create the mailbox. Otherwise, my 
> users will not know they even have these mailboxes—no one anymore knows or 
> cares about mailbox (re: folder) subscription and won't know where to look to 
> get them to show up in their client.
> 
> With autosubscribe enabled, I can see the mailboxes in my mail client without 
> explicitly subscribing, but attempting to open them will throw a server 
> error. Removing the setting will make the mailbox disappear without going 
> into settings and explicitly subscribing.
> 
> Any ETA on the first bug being fixed, and any help with the second issue (bug 
> or not) would be greatly appreciated.
> 

Can you give us a hint what the error was?

> Best regards.
> 
> P.S.: I wanted to spread this information that wasn't mentioned at all in the 
> documentation. The shadow passdb was removed in the move to 2.4; why exactly 
> is neither here nor there, though I will remark that I still don't know why 
> it was done. I had believed at the time this necessitated enabling PAM for 
> Dovecot, which it is not by default in my distro's package (Alpine), so if I 
> wanted to continue allowing local users to log in, I would have to build 
> Dovecot it myself and install PAM, neither of which I particularly wanted to 
> do. I did play with passwd-file in the past, but there were issues with it 
> that prevented me from using it—I don't remember what they were exactly, but 
> I believe they have since been fixed, since in my testing now it does seem to 
> work properly in the use-cases I tested.
> 
> Anyway, while going over my config and getting ready to re-install PAM, I had 
> a thought. /etc/shadow uses the same format for the first two fields as 
> /etc/password, and Dovecot still (supposedly) allows for passwords to be used 
> when reading /etc/password, so wouldn't it be possible to point the driver at 
> the shadow file? Turns out, yes, it works! I am now able to authenticate 
> local users from /etc/shadow again. Thanks to this, I no longer need to build 
> with PAM support and don't need to install PAM on a PAM-less system. I hope 
> this isn't considered a bug or something and support for this use-case is 
> removed, because I really don't want to bother once again with adding PAM 
> when I don't have to. :)

No, passwd-file is rather essential so it's not going to go away any time soon. 
If it works as replacement for shadow, nice. Removal of "shadow" driver is 
mentioned at 
https://doc.dovecot.org/2.4.2/installation/upgrade/2.3-to-2.4.html#backends-and-plugins
 

Aki

_______________________________________________
dovecot mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to