Hi Chris, Your suggested fix is now in git on the cyrus-imapd-2.5 branch. Thanks. :)
dump_cb() and its surrounds have been refactored heavily on our master branch, and it looks a lot more robust than the 2.5 code. So I'm looking at back-porting those changes as an alternative. One way or another, this will be fixed in 2.5.5. Cheers, ellie On Fri, Jul 10, 2015, at 10:11 AM, ellie timoney wrote: > Hi Chris, > > Thanks for reporting this. I've raised it as > https://git.cyrus.foundation/T207 and will look at it today. > > Cheers, > > ellie > > On Thu, Jul 9, 2015, at 01:06 AM, Chris Stromsoe wrote: > > I'm seeing a segfault running ctl_mboxlist -d with 2.5.4. mailboxes.db > > format is twoskip. All mailbxoes are on the default partition. > > > > gdb traces it back to this printf() in dump_cb(): > > > > case DUMP: > > if (!d->partition || !strcmpsafe(d->partition, part)) { > > printf("%s\t%d %s %s\n", name, mbtype, part, acl); > > if (d->purge) { > > cyrusdb_delete(mbdb, key, keylen, &(d->tid), 0); > > } > > } > > break; > > > > > > Initializing part to NULL at the start of the function fixes the > > segfault. > > > > The mailbox causing the problem has MBTYPE_DELETED. Tracing through the > > logic earlier in dump_cb(), the call to dlist_getatom() is failing but > > the > > partition is not being set to NULL because the if explicitly excludes > > mailboxes of MBTYPE_DELETED. > > > > // The partition is always there... > > r = dlist_getatom(dl, "P", (const char **)&part); > > > > if (!r && !(mbtype & MBTYPE_DELETED)) { > > syslog( > > LOG_ERR, > > "No partition for mailbox '%s'", > > name > > ); > > > > part = NULL; > > } > > > > The test should probably be rewritten to be: > > > > if (!r) { > > if (!(mbtype & MBTYPE_DELETED)) { > > syslog( > > LOG_ERR, > > "No partition for mailbox '%s'", > > name > > ); > > } > > part = NULL; > > } > > > > > > > > -Chris
