CVS log for odmrd/odmrd

Up to [Local Repository] / odmrd

Request diff between arbitrary revisions

Keyword substitution: kv
Default branch: MAIN

New TLDs have all sorts of names -- make regex's test for valid domain name
  without assuming a particular format for the TLD part (.museum anyone?).
Update copyright date.

Reorder subroutines to group related subs, and comment each related group.

Clean up main loop: move ALRM sig handler setting into &serverloop() and
  &clientloop().
Clean up the message sending loop in &clientloop(): call new &rp_dt() to get
  Return-Path and Delivered-To, which calls &mail_header() to create the
  Mail::Header object.

Clean up startup code: new &startup() replaces most of &logging(), which now just
  establishes logging. New &sig_handlers() for, well, sig handlers.
Fix &qmail_config() so that it doesn't try to use $err before &logging() has
  defined it.
Tiny change to scalar assignment in &maildir().

Replace direct use of &logging() as a configuration tunable with a $logging scalar.
Tidy &msg_list code and its interaction with &dir_scan - &dir_scan now returns a single
  list reference or false instead of a two element list.

Combine multiple calls to substr into a &cut() sub.
Tidy more of the debugging messages, remove unnecessary ones and fix a typo
  calling a `debugr' function.
Break up long line strings.
Reorder the tests when looking at ATRN command.
Combine some `assign then test' into a single statement.

Make &debug() to tidy up debug message creation.

Bug: Fix end-of-header regexp.
Drop useless debug message.

Reverse the order of $bouncerecipient checks to test the more likely
  outcome first.

Include the line number for 'failed encoding' errors, and comments
  referring to RFC2821 2.3.7 and RFC2822 2.3.
Put 'ERROR' in the errors that cause odmrd to die().

Bug: the regular expressions for a valid domain accepted a max of four
  characters in a TLD name. Not much good for .museum users!

Improve some debug messages.

Don't wait for a response after sending QUIT.

Clarify and add comments.

Bug: Assigning result of sockaddr_in() to @_ overwrites the argument to
	logging() sub. Thanks to Gerald Villemure.

remove() triple bounce files.

Make the warning text from &warn_skipping() a little more generic so it
  doesn't look out of place when &qmailconf() calls it.

Use &warn_skipping() if we have to barf out of &data() because of
  a message file containing CRLF.

Ekk another speedup out of the message reading loop in &data(). Use
  seperate loops for the header and the body, the latter requiring none
  of the header checking tests, regexes, and whatnot.
``From '' header lines don't happen here, so don't bother checking for
  them (in &data()).

Emphasize the meaning of '$secretfield = "passwd"'

Consolidate four tests on $header to one test, in &data().
Combine a number of repeating 'if opt_D' constructs.

Note discrepancies w.r.t. RFC2821.

Log the values of $require_flag and $secretfield when debugging.

Slightly neater logging of rcs rev. in debugging output.

Add &warn_skipping() to output a clear warning for messages that can't be
  delivered due to some fault; bad/absent headers, empty, etc.
Sprinkle &warn_skipping() liberally.

undef the return from &isdomain() (will generate a warning if it's ever
  inadvertantly used).

Use SMTP status code from first line of response instead of last line, as
  per qmail-remote. Also saves repeated assignment to $status.
Reformat some comments.

Don't manipulate $rp and $dt until we've verified that header->get gave
  us a sensible answer.

Move repeated $state resets to the appropriate sub's.
Filehandle list was over-complicated. Drop &add_fh() and &close_fh() in
  favour of a modified &close_all_fh() that calls close on the individual
  filehandles that may be open.
Explicitly close the MSG filehandle if we hit a CRLF encode problem in
  &data().

Wrap the pop+close operation into a &close_fh() sub.
&close_all_fh() will use closedir if we still have a dirhandle open on 'D'.

Debug sub's don't need trailing semi-colons anymore.
Use @fh as a stack to record open filehandles, &add_fh() adds to top of
  stack, use &close_all_fh() in END block to ensure all filehandles are
  explicitly closed upon exit. pperl's --no-cleanup is now a realistic
  option.

Don't commit to CVS with a pperl shebang.

Bug: Missing prototypes for &sig_nodebug(), &sig_debug() and &dir_scan().

Big changes, on the road to a new release ...
Use the vpopmail.pm module now. It has limited usefulness. Patch submitted
  to author to add what we need. Hope to be able to drop CDB_File.pm if
  patch is accepted.
As a consequence of above, &vopmailuser() is removed.
As a consequence of above, &homedir() is removed. &homedir() was
  broken, not taking into account vpopmail's hashing of directories.
Update &cdb_auth() to use the clear_text field for secrets -- looked up
  in vpasswd.cdb. Use vpopmail.pm for pw_gecos and pw_dir.
$usegecos is gone, replaced by $secretfield.
$header->get may fail. Don't use $dt before checking it's TRUE.
&smtp_read() logs $state when receives junk input.
Add /s op. to &stdin() s///, and omit chomp.
Formatting changes to single statement if's.
&has_access() into a one-liner.
We know $mdir is TRUE in &maildir(), don't waste time checking.
Rewrite &msg_list(). Push the real work into a new &dir_scan(), and
  call it twice so we now read Maildir/cur also. Since we read cur,
  permit colons in file names.
Chdir to / on startup.
Make USR{1,2} sig. handlers into named sub's so they can be reinstalled.

Add support for using V_USERx mailbox flags to enable/disable ODMR access
 to a mailbox. Enable this feature by default.

Minor comment changes.

Move STDIN input stuff to &stdin() to ensure consistent input handling.
Bug: (of sorts) don't getpeername() if STDIN is a tty.
Be fascist about CRLF on input.
Bug: remotehost wasn't mentioned in 'does not like recipient' bounce msg.
Bug: Fix taint-ness failure dependency in unlink() in &bounce().
Pause for 1sec. in &needauth(), &b64fail() and &unimp().

Suck in Socket.pm and use getpeername() to get remote side's IP address.
 Log this in bounce messages.
Default to 'vpopmail' for vpopmail username.

Log multiple domain ATRNs.
Bug: &cdb_auth() tested $usegecos for the wrong element of its return
 list. $usegecos affected which field to use as the Maildir path, not
 secret.

odmrlifetime of zero allows messages to remain in mailbox indefinitely.

odmrlifetime should be two weeks; 1209600s.

Be liberal in what we accept in &auth().
&quit() prototype wasn't accurate.
Delay for 3 sec's for refused ATRNs.
Minor comment and style changes.

Log into syslog using mail facility.

RFC2821 says to give a 501 response to QUIT cmd. with arguments.
Change our response for bad commands to a 500 instead of 502.
2821 says to wait 10 mins. for the message acceptance reply
  after DATA.

Reorder lots of subroutines to have related subs next to each other.
Move the calls to sleep into &authfail() itself. Duh! Why didn't I
  think of that when I first put them in?

Forward declare four subs that perl 5.6.1 warns ``foo() called too early
  to check prototype'' about.

Delay for three seconds when authentication fails. Make password
  guessing a little more awkward.
Replace some nested foreach->if->foo type statements with
  foreach (grep { ... })

Log sender and recipient before and after sending in &clientloop().

Improve log messages in &atrn() for clarity, and change some debug
  messages to non-debug -- and vice versa.

Reformat some comments and retab some code, trim out blank lines.
Improve non-debug log messages for clarity and consistency.
Use ``shift || ""'' liberally.
Reset timeout counter during reading of message file in &data() and
  before close of pipe to $mta.
Reduce timeout in &clientquit() to 45s.

Don't bother trying to decode response strings that aren't a multiple
  of four characters in length. MIME::Base64 would work anyway, but it
  logs an annoying warning message.
Use &printable() on the results of decode_base64() -- it can return junk
  if given junk input.
Bug: don't assume the split() into @authuser returned two elements.
Create &b64fail() for Base64 decode failure message.

Add RCS revision number to debugging output.

Remove eval/die block in &data(). On rereading docs, it's not necessary.
  We can send the final dot and check the reponse then. We don't need to
  check for a response during message transmission. No more waiting for
  one second per message!
Reword some comments for clarity's sake.

Don't set $| for input filehandles. It has no effect!
Don't set $| for pipe to qmail-inject. Let buffering do its thing. The
  close() will flush buffers when we're done.
Weed out non-printing characters. &printable() is applied to all input
  from STDIN. Might also need to use it on the results from
  Mail::Header.
Make bounce error messages a little more verbose and qmail-like.
&smtp_read() condenses consecutive whitespace, and won't accept more than
  150 lines in a response. Otherwise it'd take in data until the timeout
  fired.
Minor change in debug output from &logging() and comment in
  &qmail_config().

rand() over a wider range. Still not happy with srand() though ...
Disable buffering on stdout and stderr before we output anything.
Clarify comment wrt to repeated assignment to $Sys::Syslog::host.

Use shift(" ", ...) in place of shift(/\s+/, ...)
Seed rand() with data that's fractionally more than trivial to guess.
&msg_list() shouldn't return a ref. to @files in an error. @files has
  no reason to persist.
Log refused ATRNs.
Only set $header to 0 if it's currently 1, in &data().
Log missing vpopmail home dirs in &vpopmaildir().

Verify $state after both &clientloop() and &serverloop().
&clientquit() no longer calls exit(). Must call exit explicitly. We
  need to &clientquit() to close the network connection but we need
  to stay running so we can &bounce().
&clientquit() prototype was incorrect. Increased its timeout too.
Enabled message bouncing by default.

Textual changes to comments and remove some dead whitespace.
Place arguments to tie() in ()'s.
Typo in a &msg_list() debug message.
Bug: &isemail() checked for ``postmaster'' instead of ``<postmaster>''.

More precedence changes; ||, && replaced with or, and; a few more
  ()'s removed.
&clientgreet() will now fallback to HELO if EHLO fails.

Minor style changes; remove unnecessary ()'s and editing a comment here
  and there.

Prototype all named subroutines.
Style; remove unnecessary ()'s and let normal precedence do its thing.
Bug: Messages which were permanently rejected (and so added to @bounces
  to be bounced) which were also too old were added to @bounces twice.
  Once when the delivery failed and a second time when we check the
  file age after the delivery attempt.
Bug: &bounce() wasn't removing queue files after sending bounces.
&data() doesn't need to stat() file files. They've been stat'd aleady
  and we catch any error from open() anyway.
Pattern for removing Return-Path wasn't accurate.
Have most subroutines return an explicit return value.
Move some repeatedly used &put()'s to their own subs; &atrnfail(),
  &authfail()
Rename &duplicate() to &dup().
Save a few scalars in &qmail_config().
Bug: $err was used _before_ the ref($err) test in &logging().

Pull in Sys::Syslog using a require/import only if needed. Save 400K
  mem.
Revisit timeouts. Lengthen most to 5 minutes and ensure we reset them
  after getting input so that a timer doesn't fire while we're
  processing the input.
Bug: Do CRLF and NULL s/// replaces globally.
Reorder some &&'s to short-circuit better.
Ensure we don't restore an ALRM timer of 0 after eval/die in &data()
Bug: &data() didn't return a list on success.
Bug: &smtp_read() tested for !4 instead of !$4. Also misinterpreted
  some broken input.
Bring wayward syscall error messages inline with
  majority:  foo($bar): $!
Add rand() into challenge string. Adds a little extra unpredictability.
  Not perfect, by any means, but it's useful. See note in &logging().
Save a few variables by their reuse or (ab)using $_.

Initial import.