- Dovecot-news - dovecot.org

1.0.rc17 released
by Timo Sirainen 07 Jan '07

07 Jan '07

http://dovecot.org/releases/dovecot-1.0.rc17.tar.gz http://dovecot.org/releases/dovecot-1.0.rc17.tar.gz.sig - MySQL authentication caused username to show up as "OK" in rc16.

1 0

1.0.rc16 released
by Timo Sirainen 05 Jan '07

05 Jan '07

http://dovecot.org/releases/dovecot-1.0.rc16.tar.gz http://dovecot.org/releases/dovecot-1.0.rc16.tar.gz.sig If you've had problems with getting errors about index files sometimes being corrupted, please try if this release fixes it. If you've reported any bugs that this release hasn't fixed, please report them again so I know they still didn't get fixed and that I didn't forget them. * IMAP: When trying to fetch an already expunged message, Dovecot used to just disconnect client. Now it instead replies with dummy NIL data. * Priority numbers in plugin names have changed. If you're installing from source, you should delete the existing plugin files before installing the new ones, otherwise you'll get errors. * Maildir: We're using rename() to move files from tmp/ to new/ now. See http://wiki.dovecot.org/MailboxFormat/Maildir -> "Issues with the specification" for reasoning why this is safe. This makes saving mails faster, and also makes Dovecot usable with Mac OS X's HFS+ (after you also set dotlock_use_excl=yes, see below). + Added dotlock_use_excl setting. If enabled, dotlocks are created directly using O_EXCL flag, instead of by creating a temporary file which is hardlinked. O_EXCL is faster, but may not work with NFS. + If Dovecot crashes with Linux or Solaris, it'll log a "Raw backtrace". It's worse than gdb's backtrace, but better than nothing. + Added maildir_copy_preserve_filename=yes setting. + Added a lazy-expunge plugin to allow users to unexpunge their mails. + maildir quota: Added ignore setting to maildir quota, which allows ignoring quota in Trash mailbox. + dict quota: If dictionary doesn't yet contain the quota, calculate it by going through all the mails in all the mailboxes. + login_log_format_elements: Added %a=local port and %b=remote port + Added -i and -o options to rawlog to restrict logging only to input or output. - Doing a STATUS command for a selected mailbox (not a recommended IMAP client behavior) caused Dovecot to sync the mailbox silently. This could have lost eg. EXPUNGE events from clients, causing them to use wrong sequence numbers. - deliver was treating boolean settings set to "no" as if they were "yes" (they were supposed to be commented out for "no") - Running "dovecot" with -a or -n option while Dovecot was running deleted all authentication sockets, which caused all the future logins to fail. - maildir: RENAME and DELETE didn't touch control directory if it was different from maildir or index dir. - We treated internal userdb lookup errors as "user unknown" errors. In such situations this caused deliver to think the user didn't exist and the mail get bounced. - pam: Setting cache_key crashed - shared maildir: dovecot-keywords file's mode wasn't taken from dovecot-shared file. - dovecotpw wasn't working with PowerPC

1 0

1.0.rc15 released
by Timo Sirainen 19 Nov '06

19 Nov '06

http://dovecot.org/releases/dovecot-1.0.rc15.tar.gz http://dovecot.org/releases/dovecot-1.0.rc15.tar.gz.sig * Fixed an off-by-one buffer overflow in cache file handling. The code is executed only with mmap_disable=yes and only if index files are used (ie. INDEX=MEMORY is safe). * passdb checkpassword: Handle vpopmail's non-standard exit codes. - rc14 sometimes assert-crashed if .log.2 file existed in a mailbox (earlier versions leaked memory and file descriptors) - io_add() assert-crashfixes - Potential SSL hang fix at the beginning of the connection

1 0

Security hole #2: Off-by-one buffer overflow with mmap_disable=yes
by Timo Sirainen 19 Nov '06

19 Nov '06

Version: 1.0test53 .. 1.0.rc14 (ie. all 1.0alpha, 1.0beta and 1.0rc versions so far). 0.99.x versions are safe (they don't even have mmap_disable setting). Problem: When mmap_disable=yes setting is used, dovecot.index.cache file is read to memory using "file cache" code. It contains a "mapped pages" bitmask buffer. In some conditions when updating the buffer it allocates one byte too little. Exploitability: I think it's going to be pretty difficult to cause anything else than a crash, but I wouldn't say impossible. Only logged in IMAP/POP3 users can exploit this. In theory you might be able to exploit this for other users as well by sending them a lot of specially crafted emails, but this requires knowing what dovecot.index.cache file contains. Normally its contents can't be predicted, although perhaps with POP3 users it gets empty often enough that the exploit could be tried. Then again, the exploit requires having at least 4MB cache file, which won't happen with POP3 users before the mailbox has about 170k mails (if I counted right). With IMAP the cache file is used more, so it's easier to fill the 4MB with for example a lot of To-headers. Workaround: Use INDEX=MEMORY so the cache files aren't used at all. Fix: 1.0.rc15 fixes this. You can also use this patch: http://dovecot.org/patches/1.0/file-cache-buffer-overflow-fix.diff

1 0

1.0.rc14 released
by Timo Sirainen 12 Nov '06

12 Nov '06

http://dovecot.org/releases/dovecot-1.0.rc14.tar.gz http://dovecot.org/releases/dovecot-1.0.rc14.tar.gz.sig More fixes. "Duplicate header extension keywords" is the only known problem (or if I forgot something, remind me). I'll try to figure out a way to reproduce it easily and then get it fixed. * LDAP: Don't try to use ldap_bind() with empty passwords, since Windows 2003 AD skips password checking with them and just returns success. * verbose_ssl=yes: Don't bother logging "syscall failed: EOF" messages. No-one cares about them. + Dovecot sources should now compile without any warnings with gcc 3.2+ - rc13 crashed if client disconnected while IDLEing - LDAP: auth_bind=yes fixes - %variables: Fixed zero padding handling and documented it. %0.1n shouldn't enable it, and it really shouldn't stay for the next %variable. -sign also shouldn't stay for the next variable. - Don't leak opened .log.2 transaction logs. - Fixed a potential hang in IDLE command (probably really rare). - Fixed potential problems with client disconnecting while master was handling the login. - quota plugin didn't work in Mac OS X

1 0

1.0.rc13 released
by Timo Sirainen 08 Nov '06

08 Nov '06

http://dovecot.org/releases/dovecot-1.0.rc13.tar.gz http://dovecot.org/releases/dovecot-1.0.rc13.tar.gz.sig I'll just keep on making new releases now whenever something important is fixed. Hopefully there shouldn't be many left anymore. Most of the bugs fixed in this release were found by stress testing with my imaptest tool (http://dovecot.org/tools/imaptest.c) If you're interested in knowing how perfectly your Dovecot setup works (especially if you're using NFS), you could try the tool yourself also. I still see one crash with mmap_disable=yes, but it's pretty rare. Will see if I get it fixed before v1.0, but it's not that important. + deliver: If we're executing as a normal system user, get the HOME environment from passwd if it's not set. This makes it possible to run deliver from .forward. - Older compilers caused LDAP authentication to crash - Dying LDAP connections weren't handled exactly correctly in rc11, although it seemed to work usually - Fixed crashes and memory leaks with AUTHENTICATE command - Fixed crashes and leaks with IMAP/POP3 proxying - maildir: Changing a mailbox while another process was saving a message there at the same may have caused the changes to not be made into the maildir, which could have caused other problems later..

1 0

1.0.rc12 released
by Timo Sirainen 05 Nov '06

05 Nov '06

http://dovecot.org/releases/dovecot-1.0.rc12.tar.gz http://dovecot.org/releases/dovecot-1.0.rc12.tar.gz.sig Since rc11 has problems compiling with BSDs, here's a new release. Just two changes: - rc11 didn't compile with some compilers - default_mail_env fallbacking was broken with --exec-mail Here's also again the rc11 changes: * Renamed default_mail_env to mail_location. default_mail_env still works for backwards compatibility. * deliver: When sending rejects, don't include Content-Type in the rejected mail's headers. * LDAP changes: * If auth binds are used, bind back to the default dn before doing a search. Otherwise it could fail if a user gave an invalid password. * Initial binding at connect is now done asynchronously. * Use pass_attrs even with auth_bind=yes since it may contain useful non-password fields. + passdb checkpassword: Give TCPLOCALIP and TCPREMOTEIP and PROTO=TCP environments to the checkpassword binary so we're UCSPI (and vchkpw) compatible. - mbox handling was a bit broken in rc10 - Using Dovecot via inetd kept crashing dovecot master - deliver: Don't crash with -f "". Changed the default from envelope to be "MAILER-DAEMON". - INBOX wasn't shown with LSUB command if only prefixed namespaces were used. - passdb ldap: Reconnecting to LDAP server wasn't working with auth binds. - passdb sql: Non-plaintext authentication didn't work - MySQL passdb ignored all non-password checks, such as allow_nets - trash plugin was broken

1 0

1.0.rc11 released
by Timo Sirainen 04 Nov '06

04 Nov '06

http://dovecot.org/releases/dovecot-1.0.rc11.tar.gz http://dovecot.org/releases/dovecot-1.0.rc11.tar.gz.sig Hopefully the last RC release? As far as I know there are no major problems left now. If nothing big shows up, v1.0 should be out in a couple of weeks. * Renamed default_mail_env to mail_location. default_mail_env still works for backwards compatibility. * deliver: When sending rejects, don't include Content-Type in the rejected mail's headers. * LDAP changes: * If auth binds are used, bind back to the default dn before doing a search. Otherwise it could fail if a user gave an invalid password. * Initial binding at connect is now done asynchronously. * Use pass_attrs even with auth_bind=yes since it may contain useful non-password fields. + passdb checkpassword: Give TCPLOCALIP and TCPREMOTEIP and PROTO=TCP environments to the checkpassword binary so we're UCSPI (and vchkpw) compatible. - mbox handling was a bit broken in rc10 - Using Dovecot via inetd kept crashing dovecot master - deliver: Don't crash with -f "". Changed the default from envelope to be "MAILER-DAEMON". - INBOX wasn't shown with LSUB command if only prefixed namespaces were used. - passdb ldap: Reconnecting to LDAP server wasn't working with auth binds. - passdb sql: Non-plaintext authentication didn't work - MySQL passdb ignored all non-password checks, such as allow_nets - trash plugin was broken

1 0

1.0.rc10 released
by Timo Sirainen 16 Oct '06

16 Oct '06

http://dovecot.org/releases/dovecot-1.0.rc10.tar.gz http://dovecot.org/releases/dovecot-1.0.rc10.tar.gz.sig I've finally read all the mails in the mailing list and in my INBOX. If I haven't replied to some of your mail, please resend it. Remember that since 1.0.rc9 release dovecot.index.cache files will get rebuilt in 64bit systems, and it's probably better to delete them manually so you don't get errors in log files. There are only a couple of issues left in my v1.0-TODO list: - Master process appears to be leaking log fds with kqueue. Could someone again give me access to a system where this happens? - Login process problems. How well does it work now? Hopefully well enough that v1.0 could be released. - LDAP authentication is leaking memory? Can anyone confirm this? Even better, can someone figure out what exactly is leaking? :) Not a v1.0 blocker though. I think v1.0 will be released once no-one has reported any major problems for a Dovecot release in 2-4 weeks. I think login process handling is the only potentially major problem left. There are a few patches from people that I haven't forgotten, but I've decided not to put them into v1.0 anymore: - Filesystem quota group. I don't think it's that important feature, and it might break something. - HFS+ hardlink avoiding - Managesieve - vmailmgr support And finally the changes in this release: * When matching allowed_nets IPs, convert IPv6-mapped-IPv4 addresses to actual IPv4 addresses first. + IMAP: Try to avoid sending duplicate/useless message flag updates + Added support for non-plaintext authentication for vpopmail if it returns plaintext passwords. Based on patch by Remi Gacogne. + Added %D modified to return "sub.domain.org" as "sub,dc=domain,dc=org" (for LDAP queries). Patch by Andrey Panin. - rc9 broke cache files in 64bit systems - deliver works now with mail_chroot - auth cache didn't work properly with multiple passdbs - Fixes to handling CRLF linefeeds in mboxes.

1 0

1.0.rc9 released
by Timo Sirainen 14 Oct '06

14 Oct '06

http://dovecot.org/releases/dovecot-1.0.rc9.tar.gz http://dovecot.org/releases/dovecot-1.0.rc9.tar.gz.sig Most importantly this should fix the login process problems that people have been reporting. There were also some bugs in the proxying feature. Also note the 64bit change in dovecot.index.cache files. Unless you delete dovecot.index.cache files manually, you'll these kind of error messages into your logs: Error: Corrupted index cache file ...dovecot.index.cache: registered field date.sent size changed They'll get fixed automatically of course, but it might be a bit annoying to see them. * 64bit systems: dovecot.index.cache file will be rebuilt because some time fields have been changed from 64bit fields to 32bit fields. Now the same cache file can be used in both 32bit and 64bit systems without it being rebuilt. * Added libmysqlclient workaround to conflicting sha1_result symbol, which caused Dovecot to fail logging into MySQL. + dovecot.index.cache file opening is delayed until it's actually needed. This reduces disk accesses a bit with eg. STATUS commands. + auth_cache: Try to handle changing passwords automatically: If password verification fails, but the last one had succeeded, don't use the cache. This works only with plaintext auth. - dovecot.index.cache: We didn't properly detect if some fields were different length than we expected, which caused assert crashes - Lots of fixes to login/master process handling - mbox: Fixed a bug causing "X-IMAPbase uid-last unexpectedly lost in mbox file" errors, and possibly others.

1 0