Known/suspected bugs/Missing Features on current Zmailer sources

- smtp TA (all TAs ??) leak memory in MIME header processing,
  and (what is worse) they seem to refer thru pointers located
  within FREED memory blocks..

- smtp TA to ETRN type destinations (usually not reachable) work
  poorly ever since non-blocking connection was taken into use.

- "./configure --without-rfc822-tabs"  does turn line-start TABs
  to spaces, however there is no mechanism to turn line-start
  SPACEs to TABs. (Place to do it is likely router/rfc822hdrs.c)

- Configuration files, esp.   cf/aliases.cf   has controversy in it.
  Many people think it is broken (I tend to agree) - 940614/mea

- Canned error messages are abysmal -- they need more complete ways to
  configure what is told and how, whole error facility needs an overhaul
  in fact..  (See mention above about: IETF NOTARY)

libc: "make install" is "crazy"
proto:	"make install" -- nothing goes in ???
		--> MAILSHARE/proto/ !!

All programs (?)
- At places where time definitions are entered, allow always
  "6m30s" type notation in addition to plain integers..


INSTALL:
	Badly out of date!

router:
	BUG?  router/functions.c: $(homedirectory ..) function does
		getpwnam() from somewhere WITHOUT checking at the
		possible errno telling that there is e.g. temporary
		access error with backend database...

	Error address pickup  ( router/rfc822.c: erraddress() )
	picks *bad* choices sometimes; e.g. it might consider
	'Sender:' or 'Errors-To:' headers as good picks for the
	address where to send the report to...

	RFC 822 special bug(?):  '%' is a special, and/or combination
	of   "From: foo%bar <fii@faa.fuu>"  formats causes the scanner
	to react mistakenly with 'Illegal-Object:'.

	Works now as a work-gang under leadership of one process
        whose task is to scan the directories for work..
	It is still not quite as flexible as one might want, but
	better than the old system, and now router logs are done
	in a smarter way, which is easy to rotate without any
	need for signals to anywhere..

	(older entry way below had:)
	Have separate queueing processor feeding actual routers
	to take care of the problem related to message priorities.
	Could implement:  high/normal/bulk/junk priorities which
	contains ways to be able to process messages from different
	work categories without causing starvation of some class
	queue; E.g. at least one router per class, but if some
	class is (momentrarily) empty, its router can work other
	class jobs.

transports/smtp:
	This is a tell-tale indication that total/header byte accounting
	gets it wrong: (all goes into header..)
		598 bytes, 598 in header, 1 recipients, 0 secs for envelope,
			 1 secs for body xfer

	If reply waiting in smtp_sync() gets SIGALRM, program crashes..
	(syncing for BDAT, for example.)

	There is some memory leak (possibly in all ta programs?),
	propably in MIME header rewrites.  It appears only in
	very long living and hard working smtp-ta program as
	slow size growth.

	VERIFY: BDAT (and DATA) status processing in case the remote
	yields disk-full temporary error.

	To rewrite the MX processing in streamlined flashion:
	  In case two domains have common (sub-)set of MX servers, and
	  we have been feeding mail to domain A thru MX-server B, and
	  now we change to deliver domain C which too has MX-server B.
	  We are to reuse the already open channel to server B to feed
	  the message out even though the domain C might have had server
	  D with lower preferrence..
	  (Noting that all MX-hosts with same or larger privilege,
	   than ourselves are to be pre-removed from the list!)

	- We violate RFC 1123 part 5.2.2 which says that at MAIL FROM
	  and at RCPT TO the domain parts MUST be primary DNS objects,
	  *not* CNAMEs.  We never rewrite...

- Somewhere ? (smtpserver, and perhaps smtp transport agent ?)
	Implement OUTBOUND message size limit mechanism.
	Possibly a multi-value system which works against
	IP/reversed domain information, and adjust maximum
	inbound message size at the smtpserver per that
	given server.

	A further complication is need to limit outbound
	email's sizes when the inbound size is limited so
	that our system will not send out anything to which
	error report size is larger than our inbound limit.

	Problem being, of course, that outbound channels
	need to interrogate that same configuration mechanism
	to do "Sorry, this is too large for sending" processing.

- All transport agents:
	- Unify the content MIME processing routines;  Construct them
	  to be a common sub-layer ?
	- Do not decode Q-P or Base-64 if the resulting material exceeds
	  RFC-821/RFC-822 line length limits (or whatever limits have
	  been set).

	(router + TA library + TA programs):
	Router to scan message MIME structure, and to store that
	into dense concise format which doesn't need prerescan of
	the message *body* for every delivery attempt in case some
	"in-flight" data conversions are desired.

	- transporters: When decoding MIME-QP the final line MAY be of
		".....=", which means there is no final NEWLINE.
	Fixed on SMTP, must fix on MAILBOX, and SM!  (4-Jan-95)
	(Prepared to do it, however only on SMTP it is fatal -- I think..)

- scheduler:
	If selector clause is of form:   foo-*/*,  and command has
	'-c $channel' in it, idle pool CAN'T be common to all threads
	matching this clause..

	How does arrival of a new job into active thread work in case
	the current activity has already reached the end of the thread ?
	When will the new job be fed to the transport agent ?

        Considering ways how to achieve that multi-process per thread:
        - thread-head needs a way to account the number of processes
          in active service (an int)
        - processes in active service need to be chained together
          (bidirectionally) to ease removing them from active work
          gangs when whatever they were doing is completed.
          (moving to other thread, or to idle..)
        - thread->proc pointer is then just root for that chain
        - Thread activation would need to start desired number of
          processes - keeping in mind that it isn't usefull to start
          more processes than there are items in the thread..
          Possibly some lesser number of processes - like at most
          half the number of items on the thread..
          Start them afesh/pick processes from thread-group idle pool.

        ... and a few other things I don't have a moment to think about
        now..


	Sometimes will generate spurious successfull-delivery-
	reports when the command is retried, and the tag selector
	is unable to start the real processor.

	Tool/mechanism for requeueing a message from the transport
	back to the router -- all message recipients that have not
	been processed yet.

	Tool/mechanism for the scheduler to command expiration of all
	recipients in given thread.  (alike 'manual-expiry', except
	a bit smarter..)

	- Delivery messages plain-text part needs easier human
	  understandable format on the messages.

	- Does the NOTIFY=SUCCESS work ok ?
		[2.99.49b2 - now it should ("ok3" status code)]

	Have two waiting threads (one message each):
	- start both of them with ETRN
	- both issue report of "retryat +nn"
	- both are rescheduled about immediately..
	- system spins (FAST) on those two...
		(Fixed ?  Haven't seen this for a while)

	- Another:  maxlife  -- maximum lifetime for a transport agent
	  for limiting the time a smtp-channel stays open for a long-
	  living connection...

- smtpserver
  - Control somehow if lack of configuration and/or PolicyDB is
    a system security offence, or not..

  - Supply PARAM entries for *all* (well, very nearly) command-line
    startup options where it makes sense... (NOW DOES!)
  - VERIFY: behaviour on BDAT (and DATA) phases when the disk fills
  -	Report that if the router subprocess crashes, interactive
	processing goes into mixed state -- it reports 2XX, and
	yet claims to be in wrong state (MAIL FROM or RCPT TO ?
	I have forgotten..)

  - accept email to postmaster(s), never mind what else policy testers
    have determined...
  - "-s" option processing is entirely wrong...
  - The connection source IP reversal is not verified paranoidly, we
    may accept people who claim dishonest reversals..  (But then we
    can test against their IP address..)
  - Have some flag passing mechanism from the initial policy db
    back to the smtpserver proper so that it can:
	- give more meaningfull explanations of rejection reason
	- run more verification subroutines on addresses per policy
	  driver control; namely  "this domain can be valid source only
				   when it is coming from this IP address"
	  type of checkup...

- Integrate transports/fuzzyalias/ into the configuration system

- Manuals leave a lot to desire.. Especially a good users/administrators
  manual is in need
	[July 1997: technical writers are hard at work on this one]
	[April 1998: Completing the work is at Matti's hands..]


- transports/sm/sm.c:
	- what UID it runs programs like procmail with ?
	  What uid it should run them ?
	  I don't think the current (up to 2.99.49p*) way of running
	  procmail/cyrus is entirely safe.
	- I have gotten comments that they are suid/sgid root/root
	  at most installations, and indeed so procmail is at my
	  workstation too...
	- DSN handling flags ?
	- "localize" the destination address properly.  That is, strip
	  quotes from around a quoted destination address according to
	  the RFC822 syntax. -- Is this always correct ?


- post-install-check
	- check existence of mailq/tcp in /etc/services
	- check MAILVAR/MAILSHARE/POSTOFFICE protections
	- check "nobody" and "daemon" accounts

- systemwide .forward checker
	- iterate all users in the system
	- check peoples directory ownerships and permissions
	- check peoples .forward ownerships and permissions

- Need a program to verify that given configuration is ok, checks
  for things like:
	- "nobody" userid
	- scheduler's resource control working to the maximum
	  extend that system supports
	- mmap() correct operation  (GNU autoconf does this)
	- sprintf() return type is autoconfigured correctly

- Need a program to run thru various file permission checks:
	- $MAILVAR/db/ -dir, and files in it
	- $MAILVAR/lists/ -dir, and files in it
	- ~/.forward -files (and home directories)

- IPv6 stuff:
	- lib/selfaddrs.c: Don't properly do automatic IPv6 interface address
			   picking (IPv4 is ok)
	- libc/myhostname.c: don't do automatic IPv6
	- scheduler/mailq.c: don't do IPv6 connect
	- transports/hold/hold.c: incomplete NS/AAAA, no IPv6 PTR
	- transports/mailbox/mailbox.c: BIFF gethostbyname() ?

- whole chain (smtpserver/router/mailbox+hold):
	- Have a NEW pseudoheader (NRCPT) in addition to the ORCPT, which
	  we now (in not completely kosher manner) create, if we don't get
	  it originally.  That is, similar entry for traversing unchanged
	  thru the router to tell what address we got as input.
	  (Thru some pseudo-alias..)
	- Retain (in some TA-header object) the original "MAIL FROM:<..>"
	  address, and be able to store it into the MAILBOX channel.
	  (That is, don't let the routing process to alter this!)

- transports/mailbox/mailbox.c:
	- Store the ORCPT information into a "X-Orcpt-To: "-header
	- Store the NRCPT information into a "X-InRcpt-To: "-header
	- Store the ENVID information into a "X-Envid: "-header
	- Store the original FROM:<..> into a "X-OrigFrom: "-header

	- Provide aforementioned "headers" also as environment variables
	  to the pipes
	- At the router:run_rfc822(): if the message has these headers,
	  purge them away.

- Autoconfig problems:
	- System mailbox locking schemes are sometimes non-obvious..
	- For the last PERHAPS must back introduce the host-configuring
	  for describing host-system dependencies like mailbox-lock schemes

- Got mail with a suggestion: (via zmailer-list)

 I suggest you to improve the 'crossbar' function of 'router'.
 It might return a list of header rewrite functions as well
 as a single function like now. So different rewrite methods
 could be applied for sender and recipient addresses.
 The crossbar() C function contains some comment that refers
 to some similar thing, but I found it unimplemented.
         /*    
          * We expect to see something like
          * (rewrite (fc fh fu) (tc th tu)) or
          * ((address-rewrite header-rewrite) (fc fh fu) (tc th tu))
          * back from the crossbar function.
          */

  5-Dec-1999: Reading the code, actually it *is* implemented.
	      ... or perhaps not, *usage* of that double-rewriter
		  form isn't done, existing code has always expected
		  single rewriter address, not two elt list.

- When killing previous routers/scheduler/smtpserver, should wait
  the previous process group leader to die before writing over
  the  $POSTOFFICE/.pid.KIND -file.

- Co-ordinated shutdown for the scheduler -- send a signal to it
  (SIGQUIT), and it sends out newlines to all processes still
  receiving anything from the scheduler, and shuts the feed channels
  down from the scheduler to the transporters.  Then it will spend
  some time (infinity?) to get responces from the transporters.
  When all childs are dead, it can exit.
	20-Jan-97: Sort of facility exists now..

- DSN (Delivery Status Notification) mechanism does not (yet) report
  on "DELAY". (Dec-1999: placeholders for mechanism exist:
	r+          smtp zimage.com method@zimage.com 60000
	  123456ABCD  - 1-thru-6 are TA process PID space, A-thru-D are
		        a 'delay has been reported' flag storage.)

- 'channel error' detection is partial at places;
	transporters: hold, errormail

- router/rfc822.c:  sequencer()
	Consider providing some data/function/whatever telling to
	the configuration script how many "Received:" headers there
	are so that the message can be trapped for a loop-prevention
	in the script, and not hardwired in the sequencer() code.
	(Now has a global variable that gets the count of "Received:"
	 headers, but the loop-count exceeded trap is done in the
	 same C-code without calling the scripts...)

- Sometimes incoming SMTP can be a hellish load, need to introduce
  a load-limited incoming SMTP acceptor (smtpserver/smtpserver.c)
  (Rayan wrote it, but never released it..)
	13-Dec-94:  It exists for Linux ....
	23-Dec-94:  Pulled bits from "top" -program.  Now it
		    exists also for Suns..

  A discussion on the ZMailer -list revealed that the acceptance blocking
  is a BIG can of worms, and that even the  BSD-sendmail has had a long
  journey along the rocky path to "get it right"...  (5-Jan-95)

  Hmm.. Perhaps we could do a `single process doing multi-stream reception ?'
  it would require fairly large rewrite of the SMTP-server..

  Also: What to do when there are more incoming SMTP sessions than the
  process can have open file descriptors ?  One for the SMTP socket, one
  for each spool file in active use at the input phase, and one for log.
  (The stdio is used only for SMTP responses, and for spooling out the
   accumulated message.  Thus there SHOULD be enough resources for all
   uses -- except when the system runs out of FDs per any individual
   process..)

- Processing of several headers is questionable/lack of:
	Generic: header wrapping within N chars (like to 80-chars
		 space of BITNET) and also infinite wide systems,
		 like News.. (transports/libta/writeheaders.c ?)

	Generic: RFC-1342 aka. non-US-ASCII chars in the headers..
		 Auto-conversion and wrapping
