Skip to content
HISTORY 757 KiB
Newer Older
Boris Mühmer's avatar
Boris Mühmer committed
In addition to the names listed below, the following people provided
useful inputs on many occasions: Paul D. Robertson, Simon J. Mudd.
Apologies for any names omitted.

19980105

	The compiled-in default value for resolve_smtp_sender was
	wrong (from the days that it was a boolean), causing smtpd
	to dump core when the variable was not set in main.cf.

	The INSTALL instructions now have separate sections for
	the three basic ways of running vmailer.

	The INSTALL instructions now have discusses how to deal
	with chrooted processes.

	Ported to RedHat 5.0. My, these people have re-organized
	their include files quite a bit, haven't they.

19980106

	On RedHat Linux 4.2/5.0, when a FIFO listener opens the
	FIFO with mode O_RDONLY, the FIFO remains forever readable
	after the writer has closed it.  Workaround: open the FIFO
	mode O_RDWR.

	Test program: util/fifo_rdonly_bug.c

	Unfortunately, the above fix triggers a bug on BSD/OS 3.1
	where opening the FIFO mode O_RDWR causes select() to claim
	that the FIFO is readable even before any data is written
	to it, causing read() to block or to fail.

	Test program:  util/fifo_rdwr_bug.c

	printfck (check arguments of printf-like function calls)
	found a missing argument in local/command.c

	Miscellaneous Makefile cleanups that I didn't finish before
	the first alpha release.

19980107

	Sometimes the DNS will claim that a domain does not exist,
	when in fact it does. Thus, it is a bad idea to reject mail
	from apparently non-existent domains. I have changed the
	smtpd so that it produces a soft error responses when a
	resolve_smtp_sender test fails with HOST_NOT_FOUND.  Note:
	by default, this test is still disabled.

	The DB and DBM read routines will now automagically figure
	out if (key, value) pairs were written including a terminating
	null byte or not. The DB and DBM write routines will use
	this result to determine how to write, and will fall back
	to per-system defaults otherwise.

	Renamed the README to MUSINGS, and wrote up a README that
	reflects the current status of the software.

	Added -d (don't disconnect) and -c (show running counter)
	option to te smtp-source test program. These tools are
	great torture tests for the mail software, and for the
	system that it runs on.

	Turned down the process_limit parameter (# of parallel smtp
	clients or servers) to avoid unpleasant surprises. You can
	crank up the process_limit parameter in main.cf.

19980111

	Feature: when run by the superuser, mailq now shows the
	mail queue even when the mail system is down. To this end,
	mailq (sendmail -bp) runs the showq program directly instead
	of connecting to the UNIX-domain service socket, and drops
	privileges etc.  as usual.

19980119

	Bugfix: Edwin Kremer spotted an oversight in the negated
	host matching code (for name or address patterns prefixed
	by !).

	Bugfix: upon receipt of a SIGHUP signal, the master now
	disconnects from its child processes, so that the current
	generation of child processes commits suicide, and so that
	the next generation of child processes will use the new
	configuration settings.

	Bugfix: the smtp server now skips the sender DNS domain
	lookup test for foo@[address]

	Bugfix: don't append the local domain to foo@[address]

19980120

	Bugfix: old low-priority bug in some list walk code that
	caused the master to drop core when a service was turned
	off in master.cf.

	Robustness: the mail system should be able to start up and
	to accept local postings even while the naming service is
	down.  For this reason, the mail system no longer uses
	gethostbyname() to look up its own machine name.  Sites
	that use short hostnames will have to specify their FQDN
	in main.cf (this will eventually be done by the system
	installation/configuration procedure). Should the config
Boris Mühmer's avatar
Boris Mühmer committed
	language support backticks so one can say `domainname`?
	What about $name stuff between the backtics?
Boris Mühmer's avatar
Boris Mühmer committed

	Security: the master now creates FIFOs and UNIX-domain
	sockets as the mail owner instead of as root, for better
	protection against subverted mail systems. chmod() is
	susceptible to race conditions. fchmod(), although safer,
	often does not work on sockets.

	Portability: anticipate that all major UNIXes will create
	UNIX-domain sockets with permissions modified by the process
	umask (required by POSIX). For this reason, we always
	chmod() UNIX-domain sockets, unless the system allows us
	to use the safer fchmod() instead.

	Portability: the semi-resident servers now properly handle
	EWOULDBLOCK returns from accept()  in addition to EGAIN
	(on some systems, EAGAIN and EWOULDBLOCK have different
	values).

	Bugfix: the semi-resident servers now properly handle EINTR
	returns From accept().

	Bugfix: Edwin Kremer found that mynetworks() would compute
	(32 - mask) instead of mask.

19980121

	Feature: /etc/vmailer/relocated is used by the local delivery
	program and specifies what mail should be bounced with a
	"user has moved to XXX" message. The main.cf configuration
	parameter is "relocated_maps".  Just like the "virtual_maps"
	config parameter, this feature is off by default, and the
	parameter can have values such as "files" or "files, nis"
	(on hosts equipped with NIS).

19980123

	Cleanup: virtual domain support moved from the queue manager
	to the resolve service, where it belongs.

	Feature: /etc/vmailer/canonical is used by the rewrite
	service for all addresses, and maps a canonical address
	(user@domain) to another address.  Typical use is to generate
	Firstname.Lastname@domain addresses, or to clean up dirty
	addresses from non-RFC 822 mail systems. The main.cf
	configuration parameter is "canonical_maps".  Just like
	the "virtual_maps" config parameter, this feature is off
	by default, and the parameter can have values such as
	"files" or "files, nis" (on hosts equipped with NIS).

19980124

	HPUX10 port and many little fixes from Pieter Schoenmakers.

	Bugfix: isolated an old mysterious bug that could make the
	master deaf for new connections while no child process was
	running.  A typical result was that no pickup daemon would
	be started after the previous one had terminated voluntarily.

	Bugfix: the NIS lookup code did not mystrdup() the NIS map
	name and would access free()d memory.

19980125

	Bugfix: the vstream routines would sometimes ignore flushing
	errors. The error would still be reported by vstream_fclose()
	and vstream_ferror().

	Feature: time limit on delivery to shell commands.  Config
	parameter: command_time_limit. Default value: 100 sec. The
	idea is to prevent one bad .forward file or alias file
	entry from slowly using up all local delivery process slots.

19980126

	Code cleanup: in preparation for SMTP extensions such as
	SIZE, allow an extended SMTP command to have a variable
	number of options.

19980127

	Bugfix: moved canonical map lookups away from the rewriting
	module to the cleanup service, so that canonical map lookups
	do not interfere with address rewriting on behalf of other
	programs. Back to an older trivial-rewrite program version.

	Bugfix: moved virtual map lookups away from the resolver
	back to the queue manager, so that virtual domain lookup
	does not interfere with address resolution on behalf of
	other programs. Back to an older qmgr program version.

19980131

Loading
Loading full blame...