[Dovecot] mdbox corruption

[Timo Sirainen]

On 30.11.2012, at 8.03, Daniel L. Miller wrote:

> Trying to clean up the last of the corruption caused by my own stupidity.  I now have two mailboxes for one user that give errors such as:
> 
> doveadm(fax at amfes.com): Error: zlib.read(/var/mail/amfes.com/fax/mdbox/storage/m.9): gz trailer has wrong CRC value at 20065618
> doveadm(fax at amfes.com): Error: read(/var/mail/amfes.com/fax/mdbox/storage/m.9) failed: Invalid argument (uid=238)
> doveadm(fax at amfes.com): Error: zlib.read(/var/mail/amfes.com/fax/mdbox/storage/m.9): corrupted data at 20145217
> [...]
> doveadm(fax at amfes.com): Error: read(/var/mail/amfes.com/fax/mdbox/storage/m.20) failed: Invalid argument (uid=5155)
> doveadm(fax at amfes.com): Error: zlib.read(/var/mail/amfes.com/fax/mdbox/storage/m.20): corrupted data at 24779604
> doveadm(fax at amfes.com): Error: Cached message size smaller than expected (1259 < 1260)
> doveadm(fax at amfes.com): Error: Corrupted index cache file /var/mail/amfes.com/fax/mdbox/mailboxes/INBOX/2010/dbox-Mails/dovecot.index.cache: Broken physical size for mail UID 5158
> 
> 'doveadm force-resync' does not appear to fix these.  There appear to be only the two mdbox files affected and I see about 40 errors. I would prefer to salvage as much as possible of these.  What options do I have?  Do a dsync to another location and take whatever I get?

There are no tools to fix broken compressed mails. I guess if dsync doesn't stop in the middle that would work. If it does stop, you'd need to figure out which messages those are (with doveadm dump) and delete them.