- dovecot - dovecot.org

[Dovecot] Error in dovecot 2.0.13: "Too many levels of symbolic links"
by Tom Pawlowski 17 May '12

17 May '12

Hi all, I've run into an issue we have with dovecot versions 2.0.12 through 2.0.14 on latest CentOS 6.2 when dealing with Maildirs located on NFS partitions. Although I saw Mr. Egbert's thread, his resolution does not work on our end. When dealing with users with over 500 messages in any of their Maildir folders, they occasionally receive an error from the server: Internal error occurred. Refer to server log for more information. The corresponding log entries: dovecot: imap(user1): Error: readdir(/u1/user1/Maildir/.AUTO-DELETED-SPAM/new) failed: Too many levels of symbolic links dovecot: imap(user1): Error: readdir(/u1/user1/Maildir/.logwatch/new) failed: Too many levels of symbolic links dovecot: imap(user2): Error: readdir(/u2/user2/Maildir/cur) failed: Too many levels of symbolic links dovecot: imap(user2): Error: readdir(/u2/user2/Maildir/.Trash/new) failed: Too many levels of symbolic links dovecot: imap(user3): Error: readdir(/u1/user3/Maildir/.Trash/new) failed: Too many levels of symbolic links dovecot: imap(user3): Error: readdir(/u4/user3/Maildir/.AUTO-DELETED-SPAM/new) failed: Too many levels of symbolic links dovecot: imap(user4): Error: readdir(/u1/user4/Maildir/.AUTO-DELETED-SPAM/new) failed: Too many levels of symbolic links dovecot: imap(user4): Error: readdir(/u1/user4/Maildir/new) failed: Too many levels of symbolic links A 'find . -type l' shows there are absolutely no symbolic links to be found in those directories. Running readdir.c that Timo provided always returns errno == 0 on {cur,new} of all the directories reported in the logs. It does not occur consistently in production, but it does tend to occur to a fuzzy set of the same users on different partitions. We can replicate it every time in testing. The file count in these directories are all over the place, from less than a thousand to multiple thousands. However, it's hard to calculate what may have been in new prior to the clients processing/sorting that mail. It may have been more on the higher side than lower. The error tends to favor new over cur, but there are exceptions. The tests we have been trying: Create a new folder using alpine, save anywhere from 1000 to 10000 identical emails to the new folder. Change to that new folder, thus prompting the move from new to cur. The client receives the message and aborts the opening. Anywhere from 400-520 files are successfully moved each time. It takes numerous attempts to finally move all the files over--which I suspect some IMAP clients do, hence the inconsistent errors in production. We then create a brand new folder with the same amount of files and verify that the error is reproduceable--which it is. I'm leaning towards some sort of bug on the NFS client level, as Timo mentioned, but the odd thing is we were running this setup since early April and did not begin seeing these errors until early May--with user load being relatively consistent during that time. I would think the bug would have been evident before then. I'm also curious as to why dovecot is getting an ELOOP error code from it. Any ideas or pointers would be appreciated, sorry for the length. :) -- Tom Pawlowski OIT-CSS System Administrator office: Hill 145 email: tompru(a)jla.rutgers.edu phone: (732) 445-2634

2 2

[Dovecot] Migration between namespaces breaks outlook on mac's
by Mark Zealey 17 May '12

17 May '12

Hi there, We're running dovecot 1.1 on one set of servers with a courier-imap-like namespace configuration: namespace private { separator = . prefix = INBOX. inbox = yes } We're now migrating customers onto a dovecot-2 platform, switching from Maildir to sdbox (using dsync mirror) & changing the namespace at the same time. New namespace configuration looks like (as taken from http://wiki2.dovecot.org/Namespaces) # This is for courier-imap namespace compatability namespace { type = private separator = / prefix = inbox = yes } namespace { type = private prefix = INBOX. separator = . inbox = no hidden = yes list = no alias_for = } For 99% of clients this works fine eg thunderbird on first login after migration notices the folder structure has changed & automagically redownloads folders & emails. However I have noticed some really strange messages in the logs which seems to be related to customers running outlook on macos. When they login after migration they are generating commands like: 2012-05-07T00:04:43+01:00 imap-03 dovecot: imap(xxxx): Mailbox created: INBOX 1 1 1 1 1 2012-05-07T00:05:38+01:00 imap-03 dovecot: imap(xxxx): Mailbox created: INBOX 1 1 1 1 1 1 2012-05-07T00:06:44+01:00 imap-03 dovecot: imap(xxxx): Mailbox created: INBOX 1 1 1 1 1 1 1 2012-05-07T00:07:33+01:00 imap-03 dovecot: imap(xxxx): Mailbox created: INBOX 1 1 1 1 1 1 1 1 2012-05-07T00:08:42+01:00 imap-03 dovecot: imap(xxxx): Mailbox created: INBOX 1 1 1 1 1 1 1 1 1 2012-05-07T03:09:00+01:00 imap-03 dovecot: imap(xxxx): Mailbox created: INBOX 1 1 1 1 1 1 1 1 1 1 2012-05-07T03:09:08+01:00 imap-03 dovecot: imap(xxxx): Mailbox created: INBOX 1 1 1 1 1 1 1 1 1 1 1 2012-05-07T03:10:09+01:00 imap-03 dovecot: imap(xxxx): Mailbox created: INBOX 1 1 1 1 1 1 1 1 1 1 1 1 ... this goes on a long time and we've not yet found a solution other than delete the imap config on their outlook and reconfigure. Does anyone have any idea why this might be happening, or if there is a way we can somehow have old namespaces for old customers, new namespaces for new customers (we are using sql login code so should be possible to do something like this in theory I think?) Thanks, Mark

1 0

[Dovecot] How to make Dovecot verbous
by Frank Walter 16 May '12

16 May '12

Hello I am struggling with Dovecot 2 (coming from Dovecot 1.0.10) with mysql Authentification for Postfix. I get errors when using it: dovecot: auth: Fatal: Unknown database driver 'mysql' dovecot: master: Error: service(auth): command startup failed, throttling But a first thing would be to log the real errors, if there are any, so I set in /etc/dovecot/conf.d/10-logging.conf auth_verbose, auth_verbose_passwords, auth_debug, auth_debug_passwords, mail_debug, verbose_ssl all to yes and I set: debug_log_path = /var/log/dovecot-debug.log but there is no logfile after restarting dovecot. How can I enable a (verbous) own Logfile in Dovecot? frank

3 4

Re: [Dovecot] Can we know when a user read our email?
by Joseph Tam 16 May '12

16 May '12

On Tue, 15 May 2012, Stan Hoeppner writes > This "unsend" feature was created to protect idiots from themselves, > nothing more. Which is why the IETF draft went nowhere. > > You can only "fix" some types of human stupidity with software. This is > not one of them. I thought someone could make money coming up with an "unsend" and "untwitter" service that all it does is to queue the outgoing message for 5 minutes, during which the sender can re-consider and remove it from the queue. Sorot of like the kill-switch for live broadcasts. But as the saying goes, you can't make things foolproof, as they keep making better fools. As to the OP trying to determine whether an Email message has been read, an indirect and imperfect technique, used by spammers and marketing critters, is to web bug http://en.wikipedia.org/wiki/Web_bugs It works by placing innocuous individualized tags in HTML formatted Email e.g. "<img src="http://your.domain/?id={hash}>" that downloads a 1x1 dot). You can then correlate web logs with the hashes to see which messages got rendered. A hit does not necessarily mean it got read, and the absense does not mean it was ignored, but it's better than nothing. If you value your privacy, turn off HTML rendering on your Email reader. Joseph Tam <jtam.home(a)gmail.com>

2 1

[Dovecot] server with only shared mailboxes ?
by Frank Bonnet 16 May '12

16 May '12

hello I would like to setup a separate server with only shared mailboxes for internal use links to doc welcome (dovecot 2 ) thank you Envoyé de mon iPhone.

1 0

[Dovecot] plugin: Don't allow deleting the last instance of a mail
by Timo Sirainen 16 May '12

16 May '12

Before beginning to code it, does anyone have any related ideas that could be useful before starting it? Mainly what I'm intending to do is to write a plugin for mdbox storage that allows users to expunge mails in a specific namespace only if there are other instances for the mail. So basically: * Users can't accidentally delete mails (in shared mailboxes) * Users can still move mails within folders For v2.1 this can be implemented only to a specific storage backend, so at least for now only mdbox will support it. Maildir and sdbox support would also be possible by looking at hard link counts. Race conditions would allow deleting the mails, but that's unlikely to be an issue since the main point is to avoid accidental deletions that shouldn't be attempted in the first place anyway.

2 3

[Dovecot] stats + fts squat plugins, indexer-worker error on message indexing
by David Warden 16 May '12

16 May '12

On RHEL 6.2 + dovecot 2.1.5 with the stats and fts (squat) plugins enabled and each user having their own uid on the system I get the following error in the logs when the indexer-worker process indexes some messages: Apr 25 09:56:19 wardentest3 dovecot: imap-login: Login: user=warden, method=PLAIN, rip=137.238.60.164, lip=137.238.2.240, lport=1993, mpid=19464, encryption=TLS Apr 25 09:56:26 wardentest3 dovecot: service=indexer-worker user=warden rip= Indexed 12 messages in INBOX Apr 25 09:56:26 wardentest3 dovecot: service=indexer-worker user=warden rip= Error: open(/proc/self/io) failed: Permission denied Apr 25 09:56:29 wardentest3 dovecot: service=imap user=warden rip=137.238.60.164 Disconnected: Logged out bytes(in/out)=46/780 What I did to generate the above was remove the squat indexes for my Inbox then log in via telnet and search for a string in my Inbox, which forced an index of the messages in my Inbox. Based on http://wiki2.dovecot.org/Services#indexer-worker it seems like this is because the indexer-worker process is created as root but at the time of indexing the indexer-worker drops privs to the mail user. In /proc/<indexer-worker pid>/, io is mode 0400 and owned by root, so it makes sense that the mail user would be unable to read that file. The strange thing is it looks like the stats are still collected (unless I am reading the output of the stats dump incorrectly): [root@wardentest3 mailtest]# doveadm stats dump -s /var/run/dovecot/mailtest/stats command user=warden cmd args session user last_update user_cpu sys_cpu min_faults maj_faults vol_cs invol_cs disk_input disk_output read_count read_bytes write_count write_bytes mail_lookup_path mail_lookup_attr mail_read_count mail_read_bytes mail_cache_hits select inbox 27fb87218302984f084c0000ae899e04 warden 1335362182.950250 0.000000 0.000000 172 0 46 0 0 12288 13 7883 4 381 0 0 0 0 0 search text "warden" 27fb87218302984f084c0000ae899e04 warden 1335362186.588023 0.001000 0.000000 32 0 4 1 0 0 4 108 1 43 0 0 0 0 0 logout 27fb87218302984f084c0000ae899e04 warden 1335362189.161790 0.000000 0.000000 1 0 0 3 0 0 1 108 3 157 0 0 0 0 0 This obviously isn't a show-stopper but if this is causing stats to be incorrect it would be great if this could be fixed. -David Warden My doveconf -n: # 2.1.5: /etc/dovecot/mailtest/dovecot.conf doveconf: Warning: service auth { client_limit=4096 } is lower than required under max. load (12288) doveconf: Warning: service anvil { client_limit=4096 } is lower than required under max. load (12291) # OS: Linux 2.6.32-220.7.1.el6.x86_64 x86_64 Red Hat Enterprise Linux Server release 6.2 (Santiago) nfs auth_cache_negative_ttl = 0 auth_cache_size = 16 M auth_gssapi_hostname = $ALL auth_krb5_keytab = /etc/dovecot/mailtest.combined.keytab auth_master_user_separator = * auth_mechanisms = plain login gssapi auth_username_format = %Ln base_dir = /var/run/dovecot/mailtest/ default_client_limit = 4096 default_process_limit = 4096 deliver_log_format = msgid="%m" subject="%s" from="%f" size=%p result="%$" first_valid_uid = 0 hostname = mailtest.geneseo.edu instance_name = mailtest lda_mailbox_autocreate = yes lda_mailbox_autosubscribe = yes login_log_format_elements = user=%u method=%m rip=%r lip=%l lport=%a mpid=%e encryption=%c mail_fsync = always mail_location = maildir:/Mail/mailhome/%Ln/mailtest/Maildir:CONTROL=/Mail/mailhome/%Ln/mailtest/.dovecot-control:INDEX=/Mail/mailhome/%Ln/mailtest/.dovecot-index mail_log_prefix = "service=%s user=%u rip=%r " mail_nfs_index = yes mail_nfs_storage = yes mail_plugins = zlib quota mail_log notify fts fts_squat stats maildir_very_dirty_syncs = yes managesieve_notify_capability = mailto managesieve_sieve_capability = fileinto reject envelope encoded-character vacation subaddress comparator-i;ascii-numeric relational regex imap4flags copy include variables body enotify environment mailbox date ihave mbox_write_locks = fcntl mmap_disable = yes namespace default { inbox = yes location = mailbox Drafts { auto = subscribe special_use = \Drafts } mailbox Junk { auto = no special_use = \Junk } mailbox Sent { auto = subscribe special_use = \Sent } mailbox Trash { auto = subscribe special_use = \Trash } prefix = separator = . type = private } passdb { args = /etc/dovecot/passwd.masterusers driver = passwd-file master = yes } passdb { args = cache_key=%u dovecot driver = pam } plugin { fts = squat mail_log_events = delete undelete expunge copy mailbox_delete mailbox_rename flag_change append mail_log_fields = uid box msgid size from subject flags mail_log_group_events = yes quota = maildir:User quota quota_exceeded_message = Quota exceeded (mailbox for user is full). Please see http://go.geneseo.edu/emailoverquota for help deleting messages while over quota. quota_rule = *:storage=200M quota_rule2 = Trash:storage=+50M sieve = /Mail/mailhome/%Ln/mailtest/.filter.sieve sieve_dir = /Mail/mailhome/%Ln/mailtest/.sievedir sieve_max_redirects = 25 stats_memory_limit = 32 M stats_refresh = 30 secs stats_track_cmds = yes } postmaster_address = postmaster(a)geneseo.edu protocols = imap sieve lmtp quota_full_tempfail = yes service auth { unix_listener auth-exim { group = exim mode = 0660 } } service imap-login { inet_listener imap { port = 1143 } inet_listener imaps { port = 1993 } service_count = 0 vsz_limit = 256 M } service imap { process_limit = 4096 } service lmtp { inet_listener lmtp { port = 124 } } service managesieve-login { inet_listener sieve { port = 14190 } inet_listener sieve_deprecated { port = 12000 } } service pop3 { process_limit = 4096 } service stats { fifo_listener stats-mail { mode = 0666 } } ssl_ca = </etc/pki/tls/certs/geotrust-intermediate.pem ssl_cert = </etc/pki/tls/certs/wildcard-geotrust.geneseo.edu.pem ssl_key = </etc/pki/tls/private/wildcard-geotrust.geneseo.edu.key userdb { driver = passwd } verbose_proctitle = yes protocol lmtp { mail_plugins = zlib quota mail_log notify fts fts_squat stats sieve } protocol lda { mail_location = maildir:/Mail/mailhome/%Ln/mailtest/Maildir:CONTROL=/Mail/mailhome/%Ln/mailtest/.dovecot-control:INDEX=/Mail/mailhome/%Ln/mailtest/.dovecot-index mail_plugins = zlib quota mail_log notify fts fts_squat stats sieve plugin { quota = maildir:User quota quota_rule = *:storage=200M quota_rule2 = Trash:storage=+50M zlib_save = gz zlib_save_level = 6 } } protocol imap { imap_logout_format = bytes(in/out)=%i/%o mail_max_userip_connections = 50 mail_plugins = zlib quota mail_log notify fts fts_squat stats imap_quota autocreate imap_zlib imap_stats } protocol sieve { managesieve_logout_format = bytes(in/out)=%i/%o } protocol pop3 { pop3_client_workarounds = outlook-no-nuls oe-ns-eoh pop3_logout_format = bytes(in/out)=%i/%o, top=%t/%p, retr=%r/%b, del=%d/%m, mailbox-size=%s pop3_uidl_format = UID%u-%v } remote 137.238.0.0/16/16 { disable_plaintext_auth = no }

3 4

[Dovecot] [Plugin dev] First-time-only cache lock failure
by John G 15 May '12

15 May '12

Hi This is a question related to plugin developement on Dovecot 2.1.4. I'm currently implementing a dovecot plugin that retrieves mails from a corporate server through an internal protocol. The plugin is very similar to the pop3c storage (lib-storage/index/pop3c). The plugin has to map original server UIDs to IMAP UIDs, using MAIL_CACHE_POP3_UIDL cache field the same way pop3c does. So, I've got a function similar to pop3c_sync_messages(), performing several mail_index_append() and mail_cache_add() calls. The problem is that on the first sync, when the index and cache files don't exist yet, there's a failure writing data in cache. The call to mail_index_sync_commit()\mail_cache_transaction_commit() fails because of mail_cache_transaction_lock(). The index and cache files are properly created, but the cache doesn't contain the MAIL_CACHE_POP3_UIDL values. It works properly on the second sync, the MAIL_CACHE_POP3_UIDL field being properly saved. Would you have some ideas on what could be wrong here? Thanks, John

2 2

[Dovecot] dovecot/auth killed with signal 6
by Edward Burr 15 May '12

15 May '12

I've had this working for over a month now with no problems. Then I suffered a power outage today, and now dovecot/auth dies as soon as I start dovecot. I've spent the past three hours searching google, but can't find even a hint at what's wrong with my server. As far as I can tell, this is the only thing that has failed after bringing the server back up when power was restored. Can anyone help me figure out what is wrong? The maillog records: May 9 20:27:37 box dovecot: auth: Debug: Loading modules from directory: /usr/lib64/dovecot/auth May 9 20:27:37 box dovecot: auth: Debug: Module loaded: /usr/lib64/dovecot/auth/libauthdb_ldap.so May 9 20:27:37 box dovecot: auth: Debug: Module loaded: /usr/lib64/dovecot/auth/libdriver_mysql.so May 9 20:27:37 box dovecot: auth: Debug: Module loaded: /usr/lib64/dovecot/auth/libdriver_pgsql.so May 9 20:27:37 box dovecot: auth: Debug: Module loaded: /usr/lib64/dovecot/auth/libdriver_sqlite.so May 9 20:27:37 box dovecot: auth: Debug: Module loaded: /usr/lib64/dovecot/auth/libmech_gssapi.so May 9 20:27:37 box dovecot: auth: Panic: io_add(0x1) called twice fd=13, callback=0x3016833290 -> 0x3016836cd0 May 9 20:27:37 box dovecot: auth: Error: Raw backtrace: /usr/lib64/dovecot/libdovecot.so.0() [0x301683ca6a] -> /usr/lib64/dovecot/libdovecot.so.0() [0x301683ca b6] -> /usr/lib64/dovecot/libdovecot.so.0() [0x3016816dba] -> /usr/lib64/dovecot/libdovecot.so.0(ioloop_iolist_add+0x7f) [0x301684872f] -> /usr/lib64/dovecot/li bdovecot.so.0(io_loop_handle_add+0x39) [0x3016849149] -> /usr/lib64/dovecot/libdovecot.so.0(io_add+0xa3) [0x3016848593] -> /usr/lib64/dovecot/libdovecot.so.0(ma ster_service_io_listeners_add+0x68) [0x3016835b88] -> /usr/lib64/dovecot/libdovecot.so.0(master_service_init_finish+0x192) [0x30168363c2] -> dovecot/auth(main+0 x207) [0x415567] -> /lib64/libc.so.6(__libc_start_main+0xfd) [0x3016c1ecdd] -> dovecot/auth() [0x409aa9] May 9 20:27:37 box dovecot: master: Error: service(auth): child 25615 killed with signal 6 (core dumped) May 9 20:27:37 box dovecot: master: Error: service(auth): command startup failed, throttling A core dump was produced. The backtrace shows: # gdb /usr/libexec/dovecot/auth coredump Core was generated by `dovecot/auth'. Program terminated with signal 6, Aborted. #0 0x0000003016c32885 in raise () from /lib64/libc.so.6 Missing separate debuginfos, use: debuginfo-install dovecot-2.0.9-2.el6_1.1.x86_64 (gdb) bt full #0 0x0000003016c32885 in raise () from /lib64/libc.so.6 #1 0x0000003016c34065 in abort () from /lib64/libc.so.6 #2 0x000000301683ca78 in ?? () from /usr/lib64/dovecot/libdovecot.so.0 #3 0x000000301683cab6 in ?? () from /usr/lib64/dovecot/libdovecot.so.0 #4 0x0000003016816dba in i_panic () from /usr/lib64/dovecot/libdovecot.so.0 #5 0x000000301684872f in ioloop_iolist_add () from /usr/lib64/dovecot/libdovecot.so.0 #6 0x0000003016849149 in io_loop_handle_add () from /usr/lib64/dovecot/libdovecot.so.0 #7 0x0000003016848593 in io_add () from /usr/lib64/dovecot/libdovecot.so.0 #8 0x0000003016835b88 in master_service_io_listeners_add () from /usr/lib64/dovecot/libdovecot.so.0 #9 0x00000030168363c2 in master_service_init_finish () from /usr/lib64/dovecot/libdovecot.so.0 #10 0x0000000000415567 in main () This is running on CentOS 6.2 # rpm -q dovecot dovecot-2.0.9-2.el6_1.1.x86_64 # dovecot --version 2.0.9 # dovecot -n # 2.0.9: /etc/dovecot/dovecot.conf # OS: Linux 2.6.32-220.13.1.el6.x86_64 x86_64 CentOS release 6.2 (Final) auth_debug = yes auth_mechanisms = plain login auth_verbose = yes listen = * mail_debug = yes mail_location = mbox:~/mail:INBOX=/var/mail/%u managesieve_notify_capability = mailto managesieve_sieve_capability = fileinto reject envelope encoded-character vacation subaddress comparator-i;ascii-numeric relational regex imap4flags copy include variables body enotify environment mailbox date mbox_write_locks = fcntl passdb { driver = pam } plugin { sieve = ~/.dovecot.sieve sieve_dir = ~/sieve } service auth { inet_listener { port = 12345 } unix_listener /var/spool/postfix/private/auth { mode = 0666 } user = $default_internal_user } ssl_cert = </etc/pki/dovecot/certs/dovecot.pem ssl_key = </etc/pki/dovecot/private/dovecot.pem userdb { driver = passwd }

2 4

[Dovecot] default mail quota when using per user quota
by Andreas Helmcke 14 May '12

14 May '12

Hello, using dovecot 2.1 and per user mail quota via ldap is there a way to have a default quota which gets used, if the "quota" field in ldap is not set? I tried with: plugin { quota = maildir:User quota quota_rule = *:storage=3G } userdb { args = /etc/dovecot/dovecot-ldap.conf driver = ldap } and in dovecot-ldap.conf: user_attrs = dcMailQuota=quota_rule=*:storage=%$ but with this configuration the mail quota /allways/ gets loaded from ldap resulting in users without dcMailQuota attribute having no quota at all. Greetings Andreas

3 11