October 2016 - dovecot

Re: (build) crash in dcrypt when "-Bsymbolic-functions" is used on linking
by Aki Tuomi 25 Nov '16

25 Nov '16

On 26.10.2016 13:48, Christian Ehrhardt wrote: > Hi, > I was wondering about a crash when building dovecot 2.2.25 on latest > Ubuntu. > I wondered as I've had the same source building on Debian just fine. > > Some debugging led me to this weird behavior: > Using this gdb command file called autoreportissue in my case: > break dcrypt_initialize > commands > p dcrypt_vfs > p &dcrypt_vfs > watch dcrypt_vfs > c > end > break dcrypt_set_vfs > commands > p dcrypt_vfs > p &dcrypt_vfs > c > end > r > > Running test-crypto on Debian and Ubuntu reported those two behaviours: > gdb -d /root/dovecot-2.2.25/src/ -x autoreportissue ./test-crypto > > Good: > Breakpoint 1, dcrypt_initialize (backend=0x555555587c02 "openssl", > set=0x0, error_r=0x0) at dcrypt.c:15 > 15 if (dcrypt_vfs != NULL) { > $1 = (struct dcrypt_vfs *) 0x0 > $2 = (struct dcrypt_vfs **) 0x555555796370 <dcrypt_vfs> > Hardware watchpoint 3: dcrypt_vfs > Breakpoint 2, dcrypt_set_vfs (vfs=0x7ffff7835020 <dcrypt_openssl_vfs>) > at dcrypt.c:56 > 56 dcrypt_vfs = vfs; > $3 = (struct dcrypt_vfs *) 0x0 > $4 = (struct dcrypt_vfs **) 0x555555796370 <dcrypt_vfs> > Hardware watchpoint 3: dcrypt_vfs > Old value = (struct dcrypt_vfs *) 0x0 > New value = (struct dcrypt_vfs *) 0x7ffff7835020 <dcrypt_openssl_vfs> > dcrypt_set_vfs (vfs=0x7ffff7835020 <dcrypt_openssl_vfs>) at dcrypt.c:57 > 57 } > > Bad: > Breakpoint 1, dcrypt_initialize (backend=0x555555589f02 "openssl", > set=0x0, error_r=0x0) at dcrypt.c:11 > 11 { > $1 = (struct dcrypt_vfs *) 0x0 > $2 = (struct dcrypt_vfs **) 0x555555798370 <dcrypt_vfs> > Hardware watchpoint 3: dcrypt_vfs > Breakpoint 2, dcrypt_set_vfs (vfs=0x7ffff780a020 <dcrypt_openssl_vfs>) > at dcrypt.c:56 > 56 dcrypt_vfs = vfs; > $3 = (struct dcrypt_vfs *) 0x0 > $4 = (struct dcrypt_vfs **) 0x7ffff780a890 <dcrypt_vfs> > Panic: file dcrypt.c: line 34 (dcrypt_initialize): assertion failed: > (dcrypt_vfs != NULL) > Error: Raw backtrace: > /root/dovecot-2.2.25/src/lib-dcrypt/test-crypto(+0x15f7c) > [0x555555569f7c] -> > /root/dovecot-2.2.25/src/lib-dcrypt/test-crypto(default_error_handler+0) > [0x55555556a030] -> > /root/dovecot-2.2.25/src/lib-dcrypt/test-crypto(i_fatal+0) > [0x55555556a2ff] -> > /root/dovecot-2.2.25/src/lib-dcrypt/test-crypto(dcrypt_initialize+0x140) > [0x55555555f030] -> > /root/dovecot-2.2.25/src/lib-dcrypt/test-crypto(main+0x23) > [0x55555556706d] -> > /lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xf1) > [0x7ffff782d3f1] -> > /root/dovecot-2.2.25/src/lib-dcrypt/test-crypto(_start+0x2a) > [0x55555555edea] > Program received signal SIGABRT, Aborted. > > One can see that in the bad case the pointer of dcrypt_vfs is pointing > to something of the scope of the .libs/libdcrypt_openssl.so file and > not the dcrypt_initialize of test-crypto. > > That made me wonder even more - where would this issue of variable > scope come from. After more debugging I found that the linker flag > "-Bsymbolic-functions" is the reason. This is default on recent > Ubuntu, but not on Debian (?yet?). > > Eventually what happens is that the dcrypt_vfs becomes part of the > .libs/libdcrypt_openssl.so. So the call from there to dcrypt_set_vfs > ends up setting not the expected variable. > I was unable to come up with a reasonable fix since I'm not enough > into your sublib structure. > > For now I assume I'm gonna build the package stripping this flag in > Ubuntu. > But long term I think dovecot should fix it to work with that compiler > flag. > Therefore the report to make you aware. > > > P.S. thanks to the dovecot community for having unit tests that find > this at build time! > > -- > Christian Ehrhardt > Software Engineer, Ubuntu Server > Canonical Ltd Hi! Thank you for reporting this, we'll look into it. Aki Tuomi Dovecot oy

1 1

Compound Literal - xlc and gcc differences can be patched
by Michael Felt 19 Nov '16

19 Nov '16

Since I so miserably misspelled packaging - a new thread specific to the issue at hand. I found a "workaround". In short, xlc does not accept arrays of nested array - the bottom line (best message) is: 1506-995 (S) An aggregate containing a flexible array member cannot be used as a member of a structure or as an array element. At the core - all the other messages come because this (I shall call it nesting). For the test-http-auth.c file I have a patch (attached). make ran for while with no issue, then it started running into several files with this same style of specification - doveadm-dict.c being the first one. I ran "make -i" and I notice several files with this - for xlc - "unaccepted" syntax. Please look at the patch - what it basically is - and let me know whether you would consider accepting patches in this form (e.g., verify gcc (I assume) will accept it as well). If your answer is yes, I shall proceed with the additional files (learn how your MACROS work!) and send them to you. Again, initially I would just send one, e.g., doveadm-dict.c - to be sure your regular compiler also builds this alternate specification. Thank you for your consideration! Michael

2 5

Self-Signed Certificate issue
by Darryl Baker 13 Nov '16

13 Nov '16

My apologies if this is a repeat but my search of the archive did not turn it up in the recent past. If this has been covered just point me at the previous thread, please. I am running a small email site which I am moving from uw-imapd and Solaris to Ubuntu and Dovecot imaps and pop3s. I am trying to use a self-signed certificate for this site. I am using Thunderbird as the test client. I've tried both the pre-built snakeoil certificate and building a special one for dovecot. In /var/log/mail.err I keep getting what I am interpreting as a missing CA cert. The message is: dovecot: imap-login: Error: SSL: Stacked error: error:14094418:SSL routines:ssl3_read_bytes:tlsv1 alert unknown ca: SSL alert number 48 The certificate was created by: openssl req -new -x509 -days 365 -nodes -out /etc/ssl/certs/dovecot.pem -keyout /etc/ssl/private/dovecot.pem The Dovecot version is 2.2.22 (fe789d2) The Ubuntu version is 16.04 LTS current patches. *Darryl Baker*

3 3

Too many references: cannot splice
by Luis Ugalde 09 Nov '16

09 Nov '16

Hi, A while ago I sent an email regarding these "*ETOOMANYREFS* Too many references: cannot splice." that we've seen since Debian updated the Jessie kernel to 3.16.0-4-amd64 #1 SMP Debian 3.16.7-ckt20-1+deb8u3 (2016-01-17) x86_64 while older kernels, like 3.16.0-4-amd64 #1 SMP Debian 3.16.7-ckt11-1+deb8u6 (2015-11-09) x86_64 showed no errors at all. I was wondering if no one else is getting these errors, or if you know any workarounds that might probe useful, apart from downgrading the kernel. I would say that the infrastructure we're running is quite standard, with directors balancing users to NFS backed dovecot servers. Best regards, Luis Ugalde.

2 3

Pigeonhole/sieve possibly corrupting mails
by Ladislav Laska 08 Nov '16

08 Nov '16

Hi! I'm here again with a problem. I'm using dovecot as an IMAP server and LDA, filtering mail via sieve. However, few times a day I get the following error on server and my client (mutt) gets disconnected. Oct 15 20:20:29 ibex dovecot: imap(krakonos): Error: Corrupted index cache file /home/krakonos/.mbox/.imap/INBOX/dovecot.index.cache: Broken physical s ize for mail UID 149418 in mailbox INBOX: read(/home/krakonos/.mbox/inbox) failed: Cached message size smaller than expected (3793 < 8065, box=INBOX, UID=149418, cached Message-Id=<88deda0d-86f6-6115-af10-60ac06bb2d22(a)rename-it.nl>) Oct 15 20:20:29 ibex dovecot: imap(krakonos): Error: read(/home/krakonos/.mbox/inbox) failed: Cached message size smaller than expected (3793 < 8065, box=INBOX, UID=149418, cached Message-Id=<88deda0d-86f6-6115-af10-60ac06bb2d22(a)rename-it.nl>) (FETCH BODY[] for mailbox INBOX UID 149418) Oct 15 20:20:29 ibex dovecot: imap(krakonos): FETCH read() failed in=110326 out=5115197 This is on a new message (attached), and this error happens on some messages when first opened. Once I reconnect, the message always opens fine, and no old message ever causes problem. I also noticed this error, which is possibly connected: Oct 15 20:15:12 ibex dovecot: lda(krakonos): Error: Next message unexpectedly corrupted in mbox file /home/krakonos/.mbox/inbox at 546862809 The filesystem is ext4, and there are no errors in syslog or problems with any other services. I also don't access the mbox locally, and only dovecot manipulates the mbox (via imap and mailbox_command = /usr/libexec/dovecot/deliver) The postfix version is 2.2.25. I'm attaching dovecot -n and the offending message (after it's been corrected). I'd rather not publish my sieve file, but will send it privately. The offending message also contains other message I received at approximately the same time. Any hint's on what could be wrong? -- S pozdravem Ladislav "Krakonoš" Láska http://www.krakonos.org/

3 6

Backing up and Importing IMAP folders
by Jerry 08 Nov '16

08 Nov '16

I am running Dovecot with Postfix on a FreeBSD machine. There are problems with the drive and I cannot depend on it. Dovecot saves all mail in IMAP format. I want to back up the mail folders, install a new HD, install the latest FreeBSD OS and then reinstall my programs. Reinstalling Dovecot is simple, but how do I reinstall the IMAP folders? Can Dovecot backup the folders onto a CD and then import them when I reinstall it? My mail is kept under “/var/mail/vmail”. Should I just back up that entire directory structure and then restore it later? Thanks! -- Jerry

3 9

logging TLS SNI hostname
by Arkadiusz Miśkiewicz 08 Nov '16

08 Nov '16

Is there a way to log SNI hostname used in TLS session? Info is there in SSL_CTX_set_tlsext_servername_callback, dovecot copies it to ssl_io->host. Unfortunately I don't see it expanded to any variables ( http://wiki.dovecot.org/Variables ). Please consider this to be a feature request. The goal is to be able to see which hostname client used like: May 30 08:21:19 xxx dovecot: pop3-login: Login: user=<abc>, method=PLAIN, rip=1.1.1.1, lip=2.2.2.2, mpid=17135, TLS, SNI=pop3.somehost.org, session=<hfS9Qwk03sBTBnrN> -- Arkadiusz Miśkiewicz, arekm / ( maven.pl | pld-linux.org )

3 11

sieve sending vacation message from vmail@ns1.domain.tld
by Matthew Broadhead 07 Nov '16

07 Nov '16

I have a server running centos-release-7-2.1511.el7.centos.2.10.x86_64 with dovecot version 2.2.10. I am also using roundcube for webmail. when a vacation filter (reply with message) is created in roundcube it adds a rule to managesieve.sieve in the user's mailbox. everything works fine except the reply comes from vmail(a)ns1.domain.tld instead of user(a)domain.tld. ns1.domain.tld is the fully qualified name of the server. it used to work fine on my old CentOS 6 server so I am not sure what has changed. Can anyone point me in the direction of where I can configure this behaviour?

2 17

Errors with count:User quota and NFS
by Tom Sommer 04 Nov '16

04 Nov '16

I upgraded to 2.2.26.0 and enabled count as quota backend, expecting the recent fixes to allow me to use the backend, however I get the following errors: Oct 31 10:52:13 imap(xxxx(a)xxxx.xx): Error: Transaction log file /mnt/nfs/xxxx.xx/xxx/indexes/dovecot.list.index.log: marked corrupted Oct 31 10:52:15 imap(xxx(a)xxxx.xx): Error: Transaction log file /mnt/nfs/xxxx.xx/xxx/indexes/dovecot.list.index.log: marked corrupted Oct 31 10:52:37 imap(xxx(a)xxxx.xx): Warning: Locking transaction log file /mnt/nfs/xxx.xx/xxx/indexes/dovecot.list.index.log took 31 seconds (syncing) Oct 31 10:52:37 imap(xxx(a)xxx.xx): Warning: Locking transaction log file /mnt/nfs/xxx.xx/xxx/indexes/dovecot.list.index.log took 31 seconds (syncing) Oct 31 10:52:43 imap(xxx(a)xxx.xx): Error: Transaction log file /mnt/nfs/xxx.xx/xx/indexes/dovecot.list.index.log: marked corrupted Oct 31 10:52:52 imap(xxx(a)xxx.xx): Warning: Locking transaction log file /mnt/nfs/xxx.xx/xxx/indexes/dovecot.list.index.log took 31 seconds (syncing) Oct 31 10:53:04 imap(xxx(a)xxx.x): Warning: Locking transaction log file /mnt/nfs/xxx.xx/xxx/indexes/dovecot.list.index.log took 60 seconds (syncing) Oct 31 10:53:06 imap(xxx(a)xxx.xx): Warning: Locking transaction log file /mnt/nfs/xxx.xxx/xxx/indexes/dovecot.list.index.log took 31 seconds (syncing) (all different accounts) If I disable count as backend, there are no errors. I'm running my mail-storage on NFS, so I suspect the errors are due to locking? So is there no way to run count as quota backend with NFS? Thanks. -- Tom

2 4

Re: Outlook 2010 woes
by Joseph Tam 02 Nov '16

02 Nov '16

> Old server: > * Ubuntu 10.04.4 LTS > * Dovecot 2.1.13 > * Maildir++ > * Local auth via passwd/shadow files > > New server: > * Debian GNU/Linux 8.6 > * Dovecot 2.2.13 > * Maildir++ > * Quotas enabled > * LDAP > > Basically what's happening is that users are seeing large delays when > navigating between different IMAP folders. So, for example, user "X" is > sitting idle in their INBOX. Rebuilding caches? Do you get the same delay when going back to the folder after the initial delay. Joseph Tam <jtam.home(a)gmail.com>

7 19