June 2021 - dovecot - dovecot.org

Support client_id variable expansion
by Eirik Rye 28 Jun '21

28 Jun '21

Hello, We are using imap_id_log='*' (and imap_id_retain=yes) to log user client information to the dovecot logs, and this information is aggregated to give us extra debugging information for customer support. However, as some clients send the ID information pre-login and others send it post-login, the aggregation logic is needlessly complex as we need to correlate the session IDs of the pre-login ID messages in order to determine the correct user. Hoping to simplify this, I wanted to log the client_id variable on logout, in imap_logout_format: imap_logout_format = in=%i out=%o deleted=%{deleted} expunged=%{expunged} trashed=%{trashed} hdr_count=%{fetch_hdr_count} hdr_bytes=%{fetch_hdr_bytes} body_count=%{fetch_body_count} body_bytes=%{fetch_body_bytes} client_id="%{client_id}" However, this only results in the following log being printed: client_id="UNSUPPORTED_VARIABLE_client_id" While examining the source code (login-common/client-common.c), it seems like this could be trivially supported by adding `client_id` to the `login_var_expand_empty_tab` struct, and adding the relevant code to populate this variable to `get_var_expand_table(struct client *client)`. I am willing to write a pull request for this, unless I am missing some obvious reason why expansion of the client's `client_id` field is not supported in the first place? - Eirik

2 1

Re: [Dovecot-news] Dovecot v2.3.15 released
by Daniel J. Luke 28 Jun '21

28 Jun '21

On Jun 21, 2021, at 7:20 AM, Timo Sirainen <timo(a)sirainen.com> wrote: > Here's a new release with some security fixes and quite a lot of other changes as well. > > * Removed support for Lua 5.2. Use version 5.1 or 5.3 instead. Looks like it doesn't want to build w/o lua now. On my MacOS system configure says: checking whether we will be linking in Lua... no ... but after fixing the implicit i_gettimeofday declaration, build fails with: /bin/sh ../../libtool --tag=CC --mode=link gcc -std=gnu99 -g -O2 -fstack-protector-strong -U_FORTIFY_SOURCE -D_FORTIFY_SOURCE=2 -Wall -W -Wmissing-prototypes -Wmissing-declarations -Wpointer-arith -Wchar-subscripts -Wformat=2 -Wbad-function-cast -Wno-duplicate-decl-specifier -Wstrict-aliasing=2 -I/opt/local/include -no-undefined -L/opt/local/lib -o libdict_lua.la libtool: link: ar cru .libs/libdict_lua.a ar: no archive members specified (this doesn't happen with 2.3.14.1) -- Daniel J. Luke

5 7

Backing up per user keys for mailcrypt
by Ben Burk 28 Jun '21

28 Jun '21

I am wondering how I can back up keys for mail users in their password-protected form, without exporting them from `doveadm mailbox cryptokey export`, which requires a password. The goal here is to perform routine backups to keep keys current. Relevant config is as follows: mail_attribute_dict = file:%h/Maildir/dovecot-attributes mail_plugins = $mail_plugins mail_crypt plugin { mail_crypt_curve = secp521r1 mail_crypt_save_version = 2 mail_crypt_require_encrypted_user_key = yes } Am i correct in assuming I should back up the dovecot-attributes file? Are there any ancillary files that need to be backed up as well, such as indexes, to properly read and handle this file? I have viewed the file and it appears there are several keys at play for a single mail user. Do different folders in a users imap space have different encryption keys? Are all of these keys populated in this dovecot-attributes file? Is there any established procedure for restoring keys? Is it as simple as placing the dovecot-attributes file, if that is infact what needs to be backed up beforehand to perform a restore. -- Ben Burk BURK.TECH System Administrator

2 1

Python module for HTTP API
by Michael Ströder 28 Jun '21

28 Jun '21

HI! Can anybody recommend a high-level Python module for accessing doveadm's HTTP API? [1] Ciao, Michael. [1] https://doc.dovecot.org/admin_manual/doveadm_http_api/

3 4

Quota Config With Prefetch Creating Alias Directories
by dovecot＠ptld.com 27 Jun '21

27 Jun '21

Using Dovecot 2.3.8 with Postfix & MySQL on Centos. I have a sql table with account info, email addresses, password, quotas, etc I have another sql table for aliases containing which alias points to what email address. For imap access im using passdb password_query with prefetch. I setup quotas using count and configured postfix to check quotas with check_policy_service before accepting email. Discovered that it wont lookup quotas using prefetch. Added a second userdb so quota request can use user_query instead of the imap password_query prefetch version. userdb { driver = sql args = /etc/dovecot/mysql_quota.include skip = found result_success = return-ok result_failure = return-ok } This worked for existing email addresses in the accounts table. But any email sent to an alias was rejected because user not found. Not by postfix, postfix can see its an alias, but when postfix does check_policy_service dovecot issued the rejection because its not finding the alias in the accounts table per the userdb user_query. I added "result_failure = return-ok" to the userdb so if user_query doesn't find the address postfix is asking to accept mail for, it will fall back on quota default values. Then dovecot isn't forcing the mail to be rejected and postfix knows if its a valid user or alias. This seems to solve the issue, dovecot will return per user quotas or if not found then the global quota and let postfix decided whats a valid address to accept or reject. But now there is a side effect. Every time you ask dovecot for a quota, and if its a first time use of a newly created account dovecot creates the index directories and index files on disk. Because im using result_failure=return-ok which tells dovecot the user was found to prevent aliases from being rejected, dovecot is also creating directories and indexes for those aliases postfix is sending over check_policy_service. First question: Is this the best way to configure dovecot using quota verification for postfix to work will aliases? Second question: How can i prevent directories and indexes being created for non found user (aliases) accounts? Thank you.

2 3

Dovecot v2.3.14.1 released
by Timo Sirainen 24 Jun '21

24 Jun '21

Hi, This is an "important fixes only" release in case you don't want to upgrade to v2.3.15. There is no matching Pigeonhole release - use the same v2.3.14 instead. https://dovecot.org/releases/2.3/dovecot-2.3.14.1.tar.gz <https://dovecot.org/releases/2.3/dovecot-2.3.14.1.tar.gz> https://dovecot.org/releases/2.3/dovecot-2.3.14.1.tar.gz.sig <https://dovecot.org/releases/2.3/dovecot-2.3.14.1.tar.gz.sig> Binary packages in https://repo.dovecot.org/ <https://repo.dovecot.org/> Docker images in https://hub.docker.com/r/dovecot/dovecot <https://hub.docker.com/r/dovecot/dovecot> * CVE-2021-29157: Dovecot does not correctly escape kid and azp fields in JWT tokens. This may be used to supply attacker controlled keys to validate tokens, if attacker has local access. * CVE-2021-33515: On-path attacker could have injected plaintext commands before STARTTLS negotiation that would be executed after STARTTLS finished with the client. - lib-index: Corrupted mime.parts in dovecot.index.cache may have resulted in Panic: file imap-bodystructure.c: line 206 (part_write_body): assertion failed: (text == ((part->flags & MESSAGE_PART_FLAG_TEXT) != 0)) - imap: SETMETADATA could not be used to unset metadata values. Instead NIL was handled as a "NIL" string. v2.3.14 regression.

3 4

Dovecot v2.3.15 released
by Timo Sirainen 23 Jun '21

23 Jun '21

Hi, Here's a new release with some security fixes and quite a lot of other changes as well. https://dovecot.org/releases/2.3/dovecot-2.3.15.tar.gz <https://dovecot.org/releases/2.3/dovecot-2.3.15.tar.gz> https://dovecot.org/releases/2.3/dovecot-2.3.15.tar.gz.sig <https://dovecot.org/releases/2.3/dovecot-2.3.15.tar.gz.sig> Binary packages in https://repo.dovecot.org/ <https://repo.dovecot.org/> Docker images in https://hub.docker.com/r/dovecot/dovecot <https://hub.docker.com/r/dovecot/dovecot> * CVE-2021-29157: Dovecot does not correctly escape kid and azp fields in JWT tokens. This may be used to supply attacker controlled keys to validate tokens, if attacker has local access. * CVE-2021-33515: On-path attacker could have injected plaintext commands before STARTTLS negotiation that would be executed after STARTTLS finished with the client. * Disconnection log messages are now more standardized across services. They also always now start with "Disconnected" prefix. * Dovecot now depends on libsystemd for systemd integration. * Removed support for Lua 5.2. Use version 5.1 or 5.3 instead. * config: Some settings are now marked as "hidden". It's discouraged to change these settings. They will no longer be visible in doveconf output, except if they have been changed or if doveconf -s parameter is used. See https://doc.dovecot.org/settings/advanced/ for details. * imap-compress: Compression level is now algorithm specific. See https://doc.dovecot.org/settings/plugin/compress-plugin/ * indexer-worker: Convert "Indexed" info logs to an event named "indexer_worker_indexing_finished". See https://doc.dovecot.org/admin_manual/list_of_events/#indexer-worker-indexin… + Add TSLv1.3 support to min_protocols. + Allow configuring ssl_cipher_suites. (for TLSv1.3+) + acl: Add acl_ignore_namespace setting which allows to entirely ignore ACLs for the listed namespaces. + imap: Support official RFC8970 preview/snippet syntax. Old methods of retrieving preview information via IMAP commands ("SNIPPET and PREVIEW with explicit algorithm selection") have been deprecated. + imapc: Support INDEXPVT for imapc storage to enable private message flags for cluster wide shared mailboxes. + lib-storage: Add new events: mail_opened, mail_expunge_requested, mail_expunged, mail_cache_lookup_finished. See https://doc.dovecot.org/admin_manual/list_of_events/#mail + zlib, imap-compression, fs-compress: Support compression levels that the algorithm supports. Before, we would allow hardcoded value between 1 to 9 and would default to 6. Now we allow using per-algorithm value range and default to whatever default the algorithm specifies. - *-login: Commands pipelined together with and just after the authenticate command cause these commands to be executed twice. This applies to all protocols that involve user login, which currently comprises of imap, pop3, submisision and managesieve. - *-login: Processes are supposed to disconnect the oldest non-logged in connection when process_limit was reached. This didn't actually happen with the default "high-security mode" (with service_count=1) where each connection is handled by a separate process. - *-login: When login process reaches client/process limits, oldest client connections are disconnected. If one of these was still doing anvil lookup, this caused a crash. This could happen only if the login process limits were very low or if the server was overloaded. - Fixed building with link time optimizations (-flto). - auth: Userdb iteration with passwd driver does not always return all users with some nss drivers. - dsync: Shared INBOX not synced when "mail_shared_explicit_inbox" was disabled. If a user has a shared mailbox which is another user's INBOX, dsync didn't include the mailbox in syncing unless explicit naming is enabled with "mail_shared_explicit_inbox" set to "yes". - dsync: Shared namespaces were not synced with "-n" flag. - dsync: Syncing shared INBOX failed if mail_attribute_dict was not set. If a user has a shared mailbox that is another user's INBOX, dsync failed to export the mailbox if mail attributes are disabled. - fts-solr, fts-tika: Using both Solr FTS and Tika may have caused HTTP requests to assert-crash: Panic: file http-client-request.c: line 1232 (http_client_request_send_more): assertion failed: (req->payload_input != NULL) - fts-tika: 5xx errors returned by Tika server as indexing failures. However, Tika can return 5xx for some attachments every time. So the 5xx error should be retried once, but treated as success if it happens on the retry as well. v2.3 regression. - fts-tika: v2.3.11 regression: Indexing messages with fts-tika may have resulted in Panic: file message-parser.c: line 802 (message_parser_deinit_from_parts): assertion failed: (ctx->nested_parts_count == 0 || i_stream_have_bytes_left(ctx->input)) - imap: SETMETADATA could not be used to unset metadata values. Instead NIL was handled as a "NIL" string. v2.3.14 regression. - imap: IMAP BINARY FETCH crashes at least on empty base64 body: Panic: file index-mail-binary.c: line 358 (blocks_count_lines): assertion failed: (block_count == 0 || block_idx+1 == block_count) - imap: If IMAP client using the NOTIFY command was disconnected while sending FETCH notifications to the client, imap could crash with Panic: Trying to close mailbox INBOX with open transactions. - imap: Using IMAP COMPRESS extension can cause IMAP connection to hang when IMAP commands are >8 kB long. - imapc: If remote server sent BYE but didn't immediately disconnect, it could cause infinite busy-loop. - lib-index: Corrupted cache record size in dovecot.index.cache file could have caused a crash (segfault) when accessing it. - lib-oauth2: JWT token time validation now works correctly with 32-bit systems. - lib-ssl-iostream: Checking hostnames against an SSL certificate was case-sensitive. - lib-storage: Corrupted mime.parts in dovecot.index.cache may have resulted in Panic: file imap-bodystructure.c: line 206 (part_write_body): assertion failed: (text == ((part->flags & MESSAGE_PART_FLAG_TEXT) != 0)) - lib-storage: Index rebuilding (e.g. via doveadm force-resync) didn't preserve the "hdr-pop3-uidl" header. Because of this, the next pop3 session could have accessed all of the emails' metadata to read their POP3 UIDL (opening dbox files). - listescape: When using the listescape plugin and a shared namespace the plugin didn't work properly anymore resulting in errors like: "Invalid mailbox name: Name must not have '/' character." - lmtp: Connection crashes if connection gets disconnected due to multiple bad commands and the last bad command is BDAT. - lmtp: The Dovecot-specific LMTP parameter XRCPTFORWARD was blindly forwarded by LMTP proxy without checking that the backend has support. This caused a command parameter error from the backend if it was running an older Dovecot release. This could only occur in more complex setups where the message was proxied twice; when the proxy generated the XRCPTFORWARD parameter itself the problem did not occur, so this only happened when it was forwarded. - lmtp: The LMTP proxy crashes with a panic when the remote server replies with an error while the mail is still being forwarded through a DATA/BDAT command. - lmtp: Username may have been missing from lmtp log line prefixes when it was performing autoexpunging. - master: Dovecot would incorrectly fail with haproxy 2.0.14 service checks. - master: Systemd service: Dovecot announces readiness for accepting connections earlier than it should. The following environment variables are now imported automatically and can be omitted from import_environment setting: NOTIFY_SOCKET LISTEN_FDS LISTEN_PID. - master: service { process_min_avail } was launching processes too slowly when master was forking a lot of processes. - util: Make the health-check.sh example script POSIX shell compatible.

4 4

Quota warning script is never sent when reverse=yes
by Jesús Ángel del Pozo Domínguez 23 Jun '21

23 Jun '21

Hi there, I set up Dovecot to call a quota warning script: plugin { quota_warning = storage=100%% quota-warning 100 %u quota_warning2 = storage=95%% quota-warning 95 %u quota_warning3 = storage=80%% quota-warning 80 %u quota_warning4 = -storage=100%% quota-warning below %u } The script is called whenever the quota exceeds the corresponding amount (80%, 95% and 100%). However, on deleting messages to free space, the script is not being called even though the quota drops below 100%. I have manually tested the script, and it works: $ /usr/local/bin/quota-warning.sh below user(a)foo.bar Could you please give me any hint to fix this. I would like to call the quota-warning script so that I can clear the over_quota_flag field in the database. Regards,

1 0

ssl_min_protocol v2.3.15
by Theo Pannen 23 Jun '21

23 Jun '21

Hello, it seems that the default value for ssl_min_protocol has changed from TLSv1 to TLSv1.2, right? After upgrading to v2.3.15 and with ssl_min_protocol commented out, the server no longer offers TLSv1 and TLSv1.1. This is a good idea, but should be documented. Theo

2 1

CVE-2021-33515: SMTP Submission service STARTTLS injection
by Timo Sirainen 22 Jun '21

22 Jun '21

Open-Xchange Security Advisory 2021-06-21 Product: Dovecot Vendor: OX Software GmbH Internal reference: DOV-4583 (Bug ID) Vulnerability type: CWE-74: Failure to Sanitize Data into a Different Plane ('Injection') Vulnerable version: 2.3.0-2.3.14 Vulnerable component: submission Report confidence: Confirmed Solution status: Fixed by Vendor Fixed version: 2.3.14.1 Vendor notification: 2021-05-21 Solution date: 2021-05-22 Public disclosure: 2021-06-21 CVE reference: CVE-2021-33515 CVSS: 4.2 (CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:L/I:L/A:N) Researcher credit: Fabian Ising and Damian Poddebniak of Münster University of Applied Sciences Vulnerability Details: On-path attacker could inject plaintext commands before STARTTLS negotiation that would be executed after STARTTLS finished with the client. Only the SMTP submission service is affected. Risk: Attacker can potentially steal user credentials and mails. The attacker needs to have sending permissions on the submission server (a valid username and password). Workaround: None. Solution: Operators should update to 2.3.14.1 or later version.

3 3