view TODO @ 19604:c996bc091c6b

master: Do not close stdout if going foreground This lets one to use /dev/stdout for logging. Mainly useful for testing purposes where we can generate log output to stdout and use tee to write it to a file for later examination.
author Aki Tuomi <>
date Mon, 18 Jan 2016 15:50:23 +0200
parents 42b5cde2d560
children 673a12afb3c5
line wrap: on
line source

 - remove mail_deliver_session after all, do all the stuff transparently
   by hooking into mailbox_copy().
     - use this hook also to do the mail deduplication: 1) sort all destination
       users, 2) create mail_user only once for each user, 3) remember in
       src_mail the previously copied mail, 4) use that for mailbox_copy()ing
       to following recipients
     - make sure this removes duplicate dbox mails when sieve saves mail to
       multiple mailboxes
 - auth: user iterations shouldn't be able to use up all the workers
 - indexer: if workers are stuck, we keep adding more and more stuff to them
   which causes the ostream size to become huge. 
 - quota: maybe check quota once more at commit time to make sure the whole
   transaction fits. avoids multiple parallel slow COPY commands from being
   able to go over quota
 - lmtp: Calculate incoming mail's hash, forward it via proxying, have the
   final delivery code verify that it's correct
 - METADATA: quota, NOTIFY interaction, METADATA-SERVER capability
 - virtual: why is it using 20 bytes instead of 16 bytes? because imap
   keywords are not going in the middle, why?!
 - fts: if SEARCH X-MAILBOX is used on virtual/all folder, it doesn't update
   any indexes. (and it should skip those physical mailboxes that don't
   match the X-MAILBOX)
 - fts_autoindex=yes shouldn't really index Trash/Drafts
 - fts: if indexer has request queued, SEARCH won't return anything until
   it's done.
     - maybe abort entirely after X time and return NO
     - prioritize small quick indexing before slow large indexing?
     - in virtual mailbox searches don't wait for indexing to finish to
       large unindexed mailboxes, just show what you got
 - figure out some way to avoid a million error messages getting logged
   when service imap/pop3 reaches process_limit (some kind of notification
   to login process that the post-login process is full?)
 - lda: mail sending (bounce? forward?) is sending mixed CRLF+LFs
 - auth: remove protocol !flop {} requirement. try again remote {} and local {}
   support for auth. where do we go stuck? at least need to be able to share
   identical passdb/userdbs
 - doveadm sync -l: lock also when syncing public mailboxes? per-mailbox locks?
 - dsync: dsync_mailbox_export_init() can be very slow and not send anything
   to remote dsync for a long time, which thinks the other side is dead and
   kills it. need to send some kind of keepalive-notifications.
 - dsync: rename + re-subscribe isn't handled right in first sync, because
   dsync moves the subscribed-flag when it renames the node
 - virtual: use mailbox_list_index to optimize syncing
 - "/asdf" in subscriptions -> LSUB lists -> dsync assert-crashes
 - replicator: automatically remove users who don't exist
 - imapc: sync_uid_next handling doesn't seem to be correct, especially with
   Courier that doesn't send UIDNEXT on SELECT
 - sdbox: dbox_file_fix() should assume there is only one message..
 - pop3: if we can't fetch "order" field for UIDL (but could fetch it
   initially), the order will be wrong and error is logged. probably just
   need to read all the UIDLs into memory at startup?..

 - fs_list_get_mailbox_flags() is unnecessarily stat()ing files/dirs
 - doveadm-server: dsync doesn't work through proxying, because the data isn't
   actually being proxied but handled via doveadm_print()
 - CATENATE: Allow ~{binary} data but fail if there are any c-t-e: binary parts?
   or simply silently save it?

 - master-settings.c warnings aren't logged to log file at startup
 - dsync: delete foo, rename bar foo -> foo, foo-temp-1
 - dsync+imapc:
     - mailbox list could be synced pretty optimally by ignoring
       (name, uidvalidity) matches. for the left if uidvalidities are unique
       and can be matched -> rename mailbox.
     - GUID-less sync could optionally use just rfc822.size [and internaldate]
       to match messages.

 - virtual plugin doesn't verify the index file's data, crashes if broken.
 - imapc {} groups
 - libsasl: use it in pop3c, managesieve-login, doveadm auth
 - nonblocking passdb + blocking passdb -> deinit -> nonblocking passdb gets
   aborted -> calls auth_worker_call() which crashes because deinit is
   already done?
 - ldap lookups should be done via auth worker processes? maybe automatically
   if the lookups are queuing start creating workers? although multiple LDAP
   connections within the same process would work too..
 - per-msg checksums? per-cache-msg checksums? per-log record checksums?
 - lazy_expunge_only_last_instance=yes + mdbox doesn't work, because refcounts
   don't update immediately
 - lib-ssl-iostream: ssl_get_last_error() should return also syscall errors
   instead of being NULL. most importantly handle handshake code.
 - if transaction log file corruption is noticed, make sure new dovecot.index
   snapshot gets written and don't mark the whole file corrupted.. rather maybe
   just rotate and truncate it
 - mdbox: purging in alt storage could create files back to alt storage
 - LAYOUT=index:
    - force-resync should fix the index, finding any missing mailboxes, same
      for when internally detecting the error
    - see if there are any race conditions? and check other error handling
    - after doing a lot of changes the list's memory pool keeps growing.
      do an occasional re-parsing to clear the pool
    - quota recalc + dict-file [+acl?] assert-crashes in !indexing->syncing
 - imaptest: add condstore, qresync tests

 - SASL client library, initially supporting only PLAIN and plugins. Use it
   everywhere we currently use hardcoded PLAIN (proxy, doveadm, ..)
 - zlib plugin should do caching similar to binary fetches. otherwise
   partial fetches from large compressed mails are hopelessly slow.
 - Track highestmodseq always, just don't keep per-message modseqs unless
   they're enabled. Then don't return [NOMODSEQ] on select.
 - URLAUTH: if client tries to access nonexistent user, do a delay in
   imap-urlauth-client.c (AFTER destroying the worker)
     - special response in the control connection to make the imap-urlauth
       master wait before starting a new worker
 - shared user should get settings from userdb extra fields, especially
   plugin/quota_rule to get different quota limits for shared mailboxes.
   the problem is that user doesn't currently have set_parser available,
   and adding it would probably waste memory..
 - auth_debug[_passwords]=yes ability for specific users via doveadm. for
   both login-common and auth
 - settings parsing is horribly bloaty
 - doveadm: if running via doveadm-server and it fails, say something about
   error being in the log
 - indexer-worker and maybe others (doveadm?) could support dropping privileges
   permanently when service_count=1. Note that LMTP can't with multiple RCPT
 - after reading whole message text, update has_nul-state to cache
 - FIFOs maybe should be counted as connections, but unlisten should
   unlink+reopen it in master?
 - lmtp client/proxy: Handle multiline replies better
 - lmtp: support DSN extension (especially ORCPT)
 - recreate mailbox -> existing sessions log "indexid changed" error
 - add message/mime limits
 - imapc:
     - prefetching to THREAD and SORT
     - check all imap extensions and see if some don't work (condstore)
 - per-namespace imapc_* settings? create a way to "copy" a settings struct,
   so mail_storage_settings are copied to mail_namespace_settings. use the
   change tracking to figure out what settings are namespace-specific.

 - doveadm import: add -d parameter to deduplicate mails based on their GUID
   (or perhaps do it by default?)
 - sdbox: altmoving is done with mailbox locked. that's not necessary, it could
   do the copying while unlocked and delete the primary files while locked
 - passdb, userdb { username_format } that doesn't permanently change
   the username
 - mdbox/sdbox index rebuild -> quota rebuild?
 - solr separate attachments (patch)
 - sql connection pooling: Count lookup latencies, avoid servers with
   significantly higher latencies. optionally use the secondary server only
   as fallback
 - maildir_storage_sync_force() shouldn't do anything except find the new
   file, don't go expunging any more stuff or we could get recursively back to
   where we started, and stuff would break
 - fuzzy: be fuzzy about date/size
 - mailbox list index:
    - with in-memory indexes be sure to refresh it more often
    - refreshing could refresh only the parts that are actually requested,
      e.g. %
 - notify_sync() could have "what changed" struct with old/new flags
 - maildir: copy dovecot-shared file from parent mailbox, not root.

 - master passdb preserves userdb_* extra fields. should it preserve
   non-userdb_* extra fields too?
 - imap, pop3: if client init fails, wait a second or two before disconnecting
 - doveadm search savedbefore 7d could be optimized in large mailboxes..
 - mdbox: storage rebuilding could log about changes it does
 - mdbox: broken extrefs header keeps causing index rebuilds
 - sent, drafts: .Sent/dovecot.index: modseq_hdr.log_offset too large
 - mail_max_lock_timeout error could be reported more nicely, also ones coming
   from lib-index
 - sql pool: if async query is pending and sync query is sent and there
   are no more empty connections, it should flush the async query first
 - NTLMv1 and LM should be disabled if disable_plaintext_auth=yes
 - SEARCH SENT*/HEADER/etc. doesn't seem optimized when using with TEXT/BODY
 - dict sql: support ignoring some search key hierarchies (e.g. acl "anyone")
 - dsync: avoid sending email when it could be copied from another mailbox.
   probably requires storage to have guid => { instances } map? that's
   rather annoying to add.

 - mdbox
    - dotlocking: cleanup should delete stale *.lock files
    - purging seems to be inefficient. run imaptest for a while, get >500
      files, start purging, it's slow until there are about 100 files left,
      then the rest is suddenly fast.
    - make sure that when reading mdbox mails sequentially the data is being
      read from disk in n kB blocks and reads cross mail boundaries and when
      reading the next mail it uses the previously read data in buffer
    - Add some kind of checksum about data+metadata and use it when checking
    - figure out a way to efficiently trigger purging when user has too much
      mail expunged (e.g. keep track of total storage size, trigger purging
      when it's 2*quota limit)
    - keep track of total bytes in dbox storage in map header. also if
      possible keep track of refcount=0 bytes. use these to optimize checks.
    - save some stuff to map index header so we don't need to keep retrying
      it. like when saving the lowest file_id which to bother checking.
    - test crash-fixing
    - mail_index_fsck() should perhaps cause dbox to be resynced?
    - optimize away reading file header?
 - maildir: out-of-disk-space failures apparently cause all kinds of
   problems, e.g. "Expunged message reappeared", "Duplicate file entry"?
 - deliver -r <address> used as autoreplies' From-address?
 - istream-seekable is inefficient. it shouldn't be reading the temp file
   immediately after writing to it
 - config process is handling requests too slowly. maybe add some caching.
    - maybe config should return all of the protocol/local/remote overrides
      when requested? then the caller could do a single lookup at start and
      merge them later internally. this would really help login processes.
 - ipv6: auth penalty should begin from /64 and gradually grow to /48 if
   necessary. and the same could be done for ipv4 as well..

 - ldap: fix multiple-gid support somehow
 - search: use mail_get_parts() only when it's already cached. if it's not,
   add it to cache afterwards.
 - dict pooling

	/* currently non-external transactions can be applied multiple times,
	   causing multiple increments. */
	//FIXME:i_assert((t->flags & MAIL_INDEX_TRANSACTION_FLAG_EXTERNAL) != 0);
  ^ appears to work now though, probably because of the added syncing stuff..

 - use backup index in mail_index_fsck()
 - proxying: support fallbacking to local (or other?) server if the first
   one is down
 - i_panic("Message count decreased") happens - why?
     - at least one backtrace shows client_destroy -> client_command_cancel ->
 - virtual: If last message matching INTHREAD rule gets expunged, the rest of
   the thread doesn't go away
 - how do shared mailboxes work with plugins?
    - lazy-expunge, fts, etc.?
    - listescape+acl can't handle shared mailboxes with escape chars
 - dovecot-acl-list:
    - how does it work with global acls?
    - update immediately after SETACL: add/remove entries, update timestamps
    - read the entire file to memory only once and keep it there, stat() later
      to see if it has changed. if not, perhaps don't even bother stat()ing
      dovecot-acl files? at least not that often..
 - fs quota: getquotaroot inbox vs. other-box should return different quotas
   if two quotas are defined
 - auth_log_prefix setting similar to mail_log_prefix

 - thread indexes: if we expunge a duplicate message-id: and we have a sibling
   with identical message-id:, we can probably just move the children?
   (unless there are non-sibling duplicates)
 - SEARCH INTHREAD requires no thread sorting by date - don't do it
 - CONDSTORE: use per-flag/per-keyword conflict checking
 - QRESYNC: Drop expunges from the middle of given seq sets if possible
 - use universal hash functions?

 - UIDVALIDITY changed while saving -> sync errors
   - mbox: copy to Trash, manually delete copied msg, change uidvalidity,
     set nextuid=1, copy again -> error
   - recent_uids assert at least with mbox
 - quota fs: Should values returned by quota be divided by the actual
   filesystem block size instead of hardcoded DEV_BSIZE? not with AIX..
 - squat:
   - wrong indexid
   - fts_build_init() assertion failed: (last_uid < last_uid_locked)
   - is locking done right? it reads header without file being locked?
   - split after ~8 bytes?
   - expunges are delayed until more mails are added
 - test replacement chars (SEARCH / SORT / Squat)

 - DEBUG: buffer overflow checking code probably doesn't handle a successful
   t_try_realloc() or pool_alloconly_realloc() properly
 - ldap:
   - sasl bind + auth_bind=yes should probably be doing only sasl binds..
   - support multiple connections for doing auth binds
   - domain lookups which set the base for user lookup
   - multiple ldap values could be joined into one field with specified
     separator (e.g. mail_access_groups=%{ldap:gidNumber:,})
   - implement something like:
       user_attrs {
	 uid = %{ldap:uidNumber}
	 home = %{ldap:homeDirectory}
	 quota_bytes = *:bytes=%{ldap:quota}

 - maildir+pop3 fast updates:
   - don't update dovecot-uidlist if dovecot.index.cache doesn't exist /
     there's nothing to cache
   - if all messages are expunged and there are no unknown extensions in index,
     unlink dovecot.index and rotate log and add some initial useful info to
     the log (uidvalidity, nextuid)

 - maildir
   - don't allow more than 26 keywords

 - file_cache: we're growing the mmap in page size blocks, which is horribly
   slow if mremap() doesn't exist.

 - keywords:
    - add some limits to how many there can be
       - don't return \* in PERMANENTFLAGS when we're full
    - remove unused keywords?

 - mail caching
    - force bits should be used only for nonregistered fields
    - change envelope parsing not to use get_headers() so imap.envelope can
      actually be cached without all the headers..
    - if there's no other pressure for compression, we should do it when
      enough temp fields are ready to be dropped
    - we could try compressing same field values into a single
      location in cache file.
    - place some maximum limit of fields to cache file? maybe some soft and
      hard limits, so when soft limit is reached drop fields that have
      been used only once. when hard limit is reached drop any fields to get
      more space. all this to avoid cache file growing infinitely.

 - mbox
    - UID renumbering doesn't really work after all?
    - still problems with CRLF mboxes.. especially with broken Content-Length
      headers (pointing between CR-LF?)
    - syncing existing indexes takes 4x longer than creating new one, why?
    - how well does dirty sync + status work? it reads the last mail every
      time? not very good..
    - always add empty line. make the parser require it too? syncing should
      make sure there always exists two LFs at end of file. raw-mbox-stream
      should make sure the last message ends with LF even if it doesn't exist
      in the file
    - Quote "From ", unquote ">From "
    - COPY doesn't work to itself (lock assert crash, for now just disallowed)

 - index
    - index file format changes:
        - split to "old" and "new" indexes and try to avoid loading "old" into
	  memory until needed
	- pack UIDs to beginning of file with UID ranges
	- use squat-like compressed uid ranges everywhere
        - write first extension intros in dovecot.index.log always with names
	   - or better yet, drop the intro concept completely as it is now
	- add "transaction boundaries" so we know which records belong to a
	  single transaction.
	   - only after that we can remove the transaction log offset
	     overwriting (otherwise we can get partial transactions in views)

 - login
    - Digest-MD5: support integrity protection, and maybe crypting. Do it
      through login process like SSL is done?

 - auth
    - with blocking passdb we're not caching lookups if the password was wrong
    - non-plaintext authentication doesn't support all features:
        - multiple passdbs don't work, only the first one is used
	- auth cache's last_success password change check doesn't exist
	- auth_cache_negative_ttl doesn't check password mismatches
    - dovecot-auth should limit how fast authentication requests are allowed
      from login processes. especially if there's one login/connection the speed
      should be something like once/sec. also limit how fast to accept new
    - support read-only logins. user could with alternative password get only
      read-access to mails so mails could be read relatively safely with
      untrusted computers. Maybe always send [ALERT] about the previous
      read-only login time with IP?

 - ssl
    - add setting: ssl_options = bitmask. by default we enable all openssl
      workarounds, this could be used to disable some of them
    - gnutls support isn't working

 - search
    - message header search: we should ignore LWSP between two MIME blocks(?)
    - message_search_init() could accept multiple search keywords so we
      wouldn't need to call it separately for each one (so we wouldn't need
      to parse the message multiple times).
    - Create our own extension: When searching with TEXT/BODY, return
      the message text surrounding the keywords just like web search engines
      do. like: SEARCH X-PRINT-MATCHES TEXT "hello" -> * SEARCH 1 "He said:
      Hello world!" 2 "Hello, I'm ...". This would be especially useful with
      the above attachment scanning.

 - general
    - things break if next_uid gets to 2^32

 - lib-http:
    - Client:
        - Handle HTTP/1.0 servers properly:
            -> Transfer-Encoding is not allowed
        - Implement support for priority/deadline-based scheduling.
          Much like:
        - Allow handling non-idempotent requests specially
          (no automatic retry, block pipeline)
        - Implement support for `Range:' requests.
        - Implement optional round-robin request scheduling for when
          host has multiple IPs.
    - Server:
        - Implement API structure for virtual hosts and resources. This way,
          multiple services can coexist independently on the same HTTP server.
        - Implement support for `Range:' requests.
    - Review compliance with RFC 7230 and RFC 7231