@(#)RELEASE_NOTES 8.7.Beta (Berkeley) %G%
This listing shows the version of the sendmail binary, the version
of the sendmail configuration files, the date of release, and a
summary of the changes in that release.
8.7/8.7 95/xx/xx CURRENTLY IN BETA PRERELEASE!!!
Fix a problem that could cause sendmail to run out of file
descriptors due to a trashed data structure after a
vfork. Fix from Brian Coan of the Institute for
Change the VRFY response if you have disabled VRFY -- some
people seemed to think that it was too rude.
Avoid reference to uninitialized file descriptor if HASFLOCK
was not defined. This was used "safely" in the sense
that it only did a stat, but it would have set the
map modification time improperly. Problem pointed out
by Roy Mongiovi of Georgia Tech.
Clean up the Subject: line on warning messages and return
receipts so that they don't say "Returned mail:"; this
Move ruleset entry/exit debugging from 21.2 to 21.1 -- this is
useful enough to make it worthwhile printing on "-d".
Avoid logging alias statistics every time you read the alias
file on systems with no database method compiled in.
If you have a name with a trailing dot, and you try looking it
up using gethostbyname without the dot (for /etc/hosts
compatibility), be sure to turn off RES_DEFNAMES and
RES_DNSRCH to avoid finding the wrong name accidently.
Problem noted by Charles Amos of the University of
Don't do timeouts in collect if you are not running SMTP.
There is nothing that says you can't have a long
running program piped into sendmail (possibly via
/bin/mail, which just execs sendmail). Problem reported
by Don "Truck" Lewis of Silicon Systems.
Try gethostbyname() even if the DNS lookup fails iff option I
is not set. This allows you to have hosts listed in
NIS or /etc/hosts that are not known to DNS. It's normally
a bad idea, but can be useful on firewall machines. This
should really be broken out on a separate flag, I suppose.
Avoid compile warnings against BIND 4.9.3, which uses function
prototypes. From Don Lewis of Silicon Systems.
Avoid possible incorrect diagnosis of DNS-related errors caused
by things like attempts to resolve uucp names using
$[ ... $] -- the fix is to clear h_errno at appropriate
times. From Kyle Jones of UUNET.
SECURITY: avoid denial-of-service attacks possible by destroying
the alias database file by setting resource limits low.
This involves adding two new compile-time options:
HASSETRLIMIT (indicating that setrlimit(2) support is
available) and HASULIMIT (indicating that ulimit(2) support
is available -- the Release 3 form is used). The former
is assumed on BSD-based systems, the latter on System
V-based systems. Attack noted by Phil Brandenberger of
New syntaxes in test (-bt) mode:
``.Dmvalue'' will define macro "m" to "value".
``.Ccvalue'' will add "value" to class "c".
``.Sruleset'' will dump the contents of the indicated
``-ddebug-spec'' is equivalent to the command-line
``$m'' will print the value of macro "m".
``/mx host'' returns the MX records for ``host''.
``/try address'' will parse address, returning the value of
crackaddr (essentially, the comment information)
and the parsed address (the same as -bv).
Somewhat better handling of UNIX-domain socket addresses -- it
should show the pathname rather than hex bytes.
Restore ``-ba'' mode -- this reads a file from stdin and parses
the header for envelope sender information and uses
CR-LF as message terminators. It was thought to be
obsolete (used only for Arpanet NCP protocols), but it
turns out that the UK ``Grey Book'' protocols require
Fix a fix in previous release -- if gethostname and gethostbyname
return a name without dots, and if a DNS lookup of that
name fails, wait one minute and try again even if DNS
not specified in the config file -- the problem is that
this code is executed before the config file is read.
This can result in an extra 60 second delay on startup
if (1) you have DNS compiled in, (2) you are not running
DNS, (3) your system hostname (as returned by hostname(1))
has no dot, AND (4) the canonical name listed in /etc/hosts
or your NIS map has no dot.
Check for proper domain name on HELO and EHLO commands per
RFC 1123 section 5.2.5. Problem noted by Thomas Dwyer III
of Michigan Technological University.
Relax chownsafe rules slightly -- old version said that if you
can't tell if _POSIX_CHOWN_RESTRICTED is set (that is,
if fpathconf returned EINVAL or ENOSYS), assume that
chown is not safe. The new version falls back to whether
you are on a BSD system or not. This is important for
SunOS, which apparently always returns one of those
error codes. This impacts whether you can mail to files
Syntax errors such as unbalanced parentheses in the configuration
file could be omitted if you had "Oem" prior to the
syntax error in the config file. Change to always print
the error message. It was especially wierd because it
would cause a "warning" message to be sent to the Postmaster
for every message sent (but with no transcript). Problem
noted by Gregory Paris of Motorola.
Rewrite collect and putbody to handle full 8-bit data, including
zero bytes. These changes are internally extensive, but
should have minimal impact on external function.
Allow full words for option names -- if the option letter is
(apparently) a space, then take the word following -- e.g.,
The full list of old and new names is as follows:
b MinFreeBlocks/MaxMessageSize
To avoid possible problems with an older sendmail,
configuration level 6 is accepted by this version of
sendmail; any config file using the new names should
specify "V6" in the configuration.
Change address parsing to properly note that a phrase before a
colon and a trailing semicolon are essentially the same
as text outside of angle brackets (i.e., sendmail should
treat them as comments). This is to handle the
``group name: addr1, addr2, ..., addrN;'' syntax (it will
assume that ``group name:'' is a comment on the first
address and the ``;'' is a comment on the last address).
This requires config file support to get right. It does
understand that :: is NOT this syntax, and can be turned
off completely by setting the ColonOkInAddresses option.
Level 6 config files added with new mailer flags:
A Addresses are aliasable.
i Do udb rewriting on envelope as well as header
sender lines. Applies to the from address mailer
flags rather than the recipient mailer flags.
j Do udb rewriting on header recipient addresses.
Applies to the sender mailer flags rather than the
k Disable check for loops when doing HELO command.
o Always run as the mail recipient, even on local
w Check for an /etc/passwd entry for this user.
5 Pass addresses through ruleset 5.
: Check for :include: on this address.
| Check for |program on this address.
/ Check for /file on this address.
@ Look up sender header addresses in the user
database. Applies to the mailer flags for the
mailer corresponding to the envelope sender
address, rather than to recipient mailer flags.
Pre-level 6 configuration files set A, w, 5, :, |, /, and @
on the "local" mailer, the o flag on the "prog" and "*file*"
mailers, and the ColonOkInAddresses option.
Eight-to-seven bit MIME conversions. This borrows ideas from
John Beck of Hewlett-Packard, who generously contributed
their implementation to me, which I then didn't use (see
mime.c for an explanation of why). This adds the
EightBitMode option (a.k.a. `8') and an F=8 mailer flag
to control handling of 8-bit data. These have to cope with
two types of 8-bit data: unlabelled 8-bit data (that is,
8-bit data that is entered without declaring it as 8-bit
MIME -- technically this is illegal according to the
specs) and labelled 8-bit data (that is, it was declared
as 8BITMIME in the ESMTP session or by using the
-B8BITMIME command line flag). If the F=8 mailer flag is
set then 8-bit data is sent to non-8BITMIME machines
instead of converting to 7 bit (essentially using
just-send-8 semantics). The values for EightBitMode are:
m convert unlabelled 8-bit input to 8BITMIME, and do
any necessary conversion of 8BITMIME to 7BIT
(essentially, the full MIME option).
p pass unlabelled 8-bit input, but convert labelled
8BITMIME input to 7BIT as required (default).
s strict adherence: reject unlabelled 8-bit input,
convert 8BITMIME to 7BIT as required. The F=8
Unlabelled 8-bit data is rejected in mode `s' regardless of
Add new internal class 'n', which is the set of MIME Content-Types
which can not be 8 to 7 bit encoded because of other
considerations. Types "multipart/*" and "message/*" are
never directly encoded (although their components can be).
Add new internal class 'e'. This is the set of MIME
Content-Transfer-Encodings that can be converted to
a seven bit format (Quoted-Printable or Base64). It is
preinitialized to contain "7bit", "8bit", and "binary".
Add C=charset mailer parameter and the the DefaultCharSet option (no
short name) to set the default character set to use in the
Content-Type: header when doing encoding of an 8-bit message
which isn't marked as MIME into MIME format. If the C=
parameter is set on the Envelope From address, use that as
the default encoding; else use the DefaultCharSet option.
If neither is set, it defaults to "unknown-8bit" as
suggested by RFC 1428 section 3.
Allow ``U=user:group'' field in mailer definition to set a default
user and group that a mailer will be executed as. This
overrides the 'u' and 'g' options, and if the `F=S' flag is
also set, it is the uid/gid that will always be used (that
is, the controlling address is ignored). The values may be
numeric or symbolic; if only a symbolic user is given (no
group) that user's default group in the passwd file is used
as the group. Based on code donated by Chip Rosenthal of
Allow `u' option to also accept user:group as a value, in the same
fashion as the U= mailer option.
Add the symbolic time zone name in the Arpanet format dates (as
a comment). This adds a new compile-time configuration
flag: TZ_TYPE can be set to TZ_TM_NAME (use the value
of (struct tm *)->tm_name), TZ_TM_ZONE (use the value
of (struct tm *)->tm_zone), TZ_TZNAME (use extern char
*tzname[(struct tm *)->tm_isdst]), TZ_TIMEZONE (use
timezone()), or TZ_NONE (don't include the comment). Code
The "Timeout" option (formerly "r") is extended to allow suboptions.
There are also two new suboptions "queuereturn" and
"queuewarn"; these subsume the old T option. Thus, to
set them both the preferred new syntax is
O Timeout.queuereturn = 5d
Sort queue by host name instead of by message priority if the
QueueSortOrder option (no short name) is set is set to
``host''. This makes better use of the connection cache,
but may delay more ``interactive'' messages behind large
backlogs under some circumstances. This is probably a
good option if you have high speed links or don't do lots
of ``batch'' messages, but less good if you are using
something like PPP on a 14.4 modem. Based on code
contributed by Roy Mongiovi of Georgia Tech (my main
contribution was to make it configurable).
Save i-number of df file in qf file to simplify rebuilding of queue
after disasterous disk crash. Suggested by Kyle Jones of
UUNET; closely based on code from KJS DECWRL code written
by Paul Vixie. NOTA BENE: The qf files produced by 8.7
are NOT back compatible with 8.6 -- that is, you can convert
from 8.6 to 8.7, but not the other direction.
Add ``F=d'' mailer flag to disable all use of angle brackets in
route-addrs in envelopes; this is because in some cases
they can be sent to the shell, which interprets them as
Don't include error file (option E) with return-receipts; this
Don't send "Warning: cannot send" messages to owner-* or
*-request addresses. Suggested by Christophe Wolfhugel
of the Institut Pasteur, Paris.
Allow -O command line flag to set long form options.
Add "MinQueueAge" option to set the minimum time between attempts
to run the queue. For example, if the queue interval
(-q value) is five minutes, but the minimum queue age
is fifteen minutes, jobs won't be tried more often than
once every fifteen minutes. This can be used to give
you more responsiveness if your delivery mode is set to
Allow "fileopen" timeout (default: 60 seconds) for opening
:include: and .forward files.
Add "-k", "-v", and "-z" flags to map definitions; these set the
key field name, the value field name, and the field
delimiter. The field delimiter can be a single character
or the sequence "\t" or "\n" for tab or newline.
These are for use by NIS+ and similar access methods.
Add "nisplus" map class. Takes -k and -v flags to choose the
key and value field names respectively. Code donated by
Add "hesiod" map class. The "file name" is used as the
"HesiodNameType" parameter to hes_resolve(3). Returns the
first value found for the match. Code donated by Scott
Hutton of Indiana University.
Add "netinfo" (NeXT NetInfo) map class. Maps can have a -k flag to
specify the name of the property that is searched as the
key and a -v flag to specify the name of the property that
is returned as the value (defaults to "members"). The
default map is "/aliases".
Add "text" map class. This does slow, linear searches through
text files. The -z flag specifies a column delimiter
(defaults to any sequence of white space), the -k flag
sets the key column number, and the -v flag sets the
value column number. Lines beginning with `#' are treated
Add "program" map class to execute arbitrary programs. The search
key is presented as the last argument; the output is one
line read from the programs standard output. Exit statuses
Add "sequence" map class -- searches maps in sequence until it
finds a match. For example, the declarations:
Kmapseq sequence map1 map2
defines a map "mapseq" that first searches map1; if the
value is found it is returned immediately, otherwise
map2 is searched and the value returned.
Add "switch" map class. This is much like "sequence" except that
the ordering is fetched from an external file, usually
the system service switch. The parameter is the name of
the service to switch on, and the maps that it will use
are this name followed by ".service_type". For example,
if the declaration of the map is
and the system service switch specifies that hosts are
looked up using dns and nis in that order, then this is
Ksample sequence hosts.dns hosts.nis
The subordinate maps must already be defined.
Add "user" map class -- looks up users using getpwnam. Takes a
"-v field" flag on the definition that tells what passwd
entry to return -- legal values are name, passwd, uid, gid,
gecos, dir, and shell. Generally expected to be used with
Add "bestmx" map class -- returns the best MX value for the host
listed as the value. If there are several "best" MX records
for this host, one will be chosen at random.
Add "userdb" map class -- looks up entries in the user database.
The "file name" is actually the tag that will be used,
typically "mailname". If there are multiple entries
matching the name, the one chosen is undefined.
Add multiple queue timeouts (both return and warning). These are
set by the Precedence: or Priority: header fields to one of
three values. If a Priority: is set and has value "normal",
"urgent", or "non-urgent" the corresponding timeouts are
used. If no priority is set, the Precedence: is consulted;
if negative, non-urgent timeouts are used; if greater than
zero, urgent timeouts are used. Otherwise, normal timeouts
are used. The timeouts are set by setting the six timeouts
queue{warn,return}.{urgent,normal,non-urgent}.
Fix problem when a mail address is resolved to a $#error mailer
with a temporary failure indication; it works in SMTP,
but when delivering locally the mail is silently discarded.
This patch, from Kyle Jones of UUNET, bounces it instead
of queueing it (queueing is very hard).
When using /etc/hosts or NIS-style lookups, don't assume that
the first name in the list is the best one -- instead,
search for the first one with a dot. For example, if
an /etc/hosts entry reads
128.32.149.68 mammoth mammoth.CS.Berkeley.EDU
this change will use the second name as the canonical
machine name instead of the initial, unqualified name.
This heuristic is only used for finding your own name on
startup, when the system may not be fully configured yet.
Change dequote map to replace spaces in quoted text with a value
indicated by the -s flag on the dequote map definition.
For example, ``Mdequote dequote -s_'' will change
"Foo Bar" into an unquoted Foo_Bar instead of leaving it
quoted (because of the space character). Suggested by Dan
Oscarsson for use in X.400 addresses.
Implement long macro names as ${name}; long class names can
be similarly referenced as $={name} and $~{name}.
Definitions are (e.g.) ``D{name}value''. Names that have
a leading lower case letter or punctuation characters are
reserved for internal use by sendmail; i.e., config files
should use names that begin with a capital letter. Based
on code contributed by Dan Oscarsson.
Fix core dump if getgrgid returns a null group list (as opposed
to an empty group list, that is, a pointer to a list
with no members). Fix from Andrew Chang of Sun Microsystems.
Fix possible core dump if malloc fails -- if the malloc in xalloc
failed, it called syserr which called newstr which called
xalloc.... The newstr is now avoided for "panic" messages.
Reported by Stuart Kemp of James Cook University.
Improve connection cache timeouts; previously, they were not even
checked if you were delivering to anything other than an
IPC-connected host, so a series of (say) local mail
deliveries could cause cached connections to be open
much longer than the specified timeout.
If an incoming message exceeds the maximum message size, stop
writing the incoming bytes to the queue data file, since
this can fill your mqueue partition -- this is a possible
denial-of-service attack.
Don't reject all numeric local user names unless HESIOD is
defined. It turns out that Posix allows all-numeric
user names. Fix from Tony Sanders of BSDI.
Add service switch support. If the local OS has a service
switch (e.g., /etc/nsswitch.conf on Solaris) that will
be used; otherwise, it falls back to using a local
mechanism based on the ServiceSwitchFile option
(default: /etc/service.switch). For example, if the
service switch lists "files" and "nis" for the aliases
service, that will be the default lookup order. Note
that you can still stack aliases files however, although
you have to list them explicitly as "implicit:filename".
Option I (NameServerOptions) no longer sets the "UseNameServer"
variable which tells whether or not DNS should be considered
canonical. This is now determined based on whether or not
"dns" is in the service list for "hosts".
Add preliminary support for the ESMTP "DSN" extension (Delivery
Status Notifications). This is not yet a standard
and the implementation is for experimentation only.
For this reason it only announces itself as "X-DSN-0"
instead of "DSN". DSN notifications override
Add T=mtstype keyletter to mailer definitions to define the value
for the Final-MTS-Type: and Remote-MTS-Type: fields in the
DSN-standard return message.
Extend heuristic to force running in ESMTP mode to look for the
six-character string "ESMTP " anywhere in the 220 greeting
message (not just the second line). This is to provide
better compatibility with other ESMTP servers.
Print sequence number of job when running the queue so you can
easily see how much progress you have made. Suggested
Map newlines to spaces in logged message-ids; some versions of
syslog truncate the rest of the line after newlines.
Suggested by Fletcher Mattox of U. Texas.
Move up forking for job runs so that if a message is split into
multiple envelopes you don't get "fork storms" -- this
also improves the connection cache utilization.
Accept "<<>>", "<<<>>>", and so forth as equivalent to "<>" for
the purposes of refusing to send error returns. Suggested
by Motonori Nakamura of Ritsumeikan University.
Relax rules on when a file can be written when referenced from
the aliases file: use the default uid/gid instead of the
real uid/gid. This allows you to create a file owned by
and writable only by the default uid/gid that will work
all the time (without having the setuid bit set). Change
suggested by Shau-Ping Lo and Andrew Cheng of Sun
Add "DialDelay" option (no short name) to provide an "extra"
delay for dial on demand systems. If this is non-zero
and a connect fails, sendmail will wait this long and
then try again. If it takes longer than the kernel
timeout interval to establish the connection, this
option can give the network software time to establish
the link. The default units are seconds.
Move logging of sender information to be as early as possible;
previously, it could be delayed a while for SMTP mail
sent to aliases. Suggested by Brad Knowles of the
Defense Information Systems Agency.
Call res_init() before setting RES_DEBUG; this is required by
BIND 4.9.3, or so I'm told. From Douglas Anderson of
the National Computer Security Center.
Add xdelay= field in logs -- this is a transaction delay, telling
you how long it took to deliver to this address on the
last try. It is intended to be used for sorting mailing
lists to favor "quick" addresses. Provided for use by
the mailprio scripts (see below).
If a map cannot be opened, and that map is non-optional, and
an address requires that map for resolution, queue the
map instead of bouncing it. This involves creating a
pseudo-class of maps called "bogus-map" -- if a required
map cannot be opened, the class is changed to bogus-map;
all queries against bogus-map return "tempfail". The
bogus-map class is not directly accessible. A sample
implementation was donated by Jem Taylor of Glasgow
University Computing Service.
Don't make a bad ``MAIL FROM:'' address on one message blow away
other messages to the same host later in the queue.
Problem noted by Eric Prestemon of American University.
Fix a possible core dump when mailing to a program that talks
SMTP on its standard input. Fix from Keith Moore of
the University of Kentucky.
Make it possible to resolve filenames to $#local $: @ /filename;
previously, the "@" would cause it to not be recognized
as a file. Problem noted by Brian Hill of U.C. Davis.
Accept a -1 signal to re-exec the daemon. This only works if
argv[0] is a full path to sendmail.
Fix bug in "addr=..." field in O option on little-endian machines
-- the network number wasn't being converted to network
byte order. Patch from Kurt Lidl of Pix Technologies
Pre-initialize the resolver early on; this is to avoid a bug with
BIND 4.9.3 that can cause the _res.retry field to get
reset to zero, causing all name server lookups to time
out. Fix from Matt Day of Artisoft.
Restore T line (trusted users) in config file -- but instead of
locking out the -f flag, they just tell whether or not
an X-Authentication-Warning: will be added. This really
just creates new entries in class 't', so "Ft/file/name"
can be used to read trusted user names from a file.
Improve NEWDB alias file rebuilding so it will create them
properly if they do not already exist. This had been
a MAYBENEXTRELEASE feature in 8.6.9.
Check for @:@ entry in NIS maps before starting up to avoid
(but not prevent, sigh) race conditions. This ought to
be handled properly in ypserv, but isn't. Suggested by
Michael Beirne of Motorola.
Refuse connections if there isn't enough space on the filesystem
holding the queue. Contributed by Robert Dana of Wolf
Skip checking for directory permissions in the path to a file
when checking for file permissions iff setreuid()
succeeded -- it is unnecessary in that case. This avoids
significant performance problems when looking for .forward
files. Based on a suggestion by Win Bent of USC.
Allow symbolic ruleset names. Syntax can be "Sname" to get an
arbitrary ruleset number assigned or "Sname = integer"
to assign a specific ruleset number. Reference is
$>name_or_number. Names can be composed of alphas, digits,
underscore, or hyphen (first character must be non-numeric).
Allow -o flag on AliasFile lines to make the alias file optional.
From Bryan Costales of ICSI.
Add NoRecipientAction option to handle the case where there is
no legal recipient header in the message. It can take
None Leave the message as is. The
message will be passed on even
though it is in technically
Add-To Add a To: header with any
recipients that it can find from
the envelope. This risks exposing
Add-Apparently-To Add an Apparently-To: header. This
has almost no redeeming social value,
and is provided only for back
Add-To-Undisclosed Add a header reading
To: undisclosed-recipients:;
which will have the effect of
making the message legal without
exposing Bcc: recipients.
Add-Bcc To add an empty Bcc: header.
There is a chance that mailers down
the line will delete this header,
which could cause exposure of Bcc:
The default is NoRecipientAction=None.
Truncate (rather than delete) Bcc: lines in the header. This
should prevent later sendmails (at least, those that don't
themselves delete Bcc:) from considering this message to
be non-conforming -- although it does imply that non-blind
recipients can see that a Bcc: was sent, albeit not to whom.
Add SafeFileEnvironment option. If declared, files named as delivery
targets must be regular files in addition to the regular
checks. Also, if the option is non-null then it is used as
the name of a directory that is used as a chroot(2)
environment for the delivery; the file names listed in an
alias or forward should include the name of this root.
For example, if you run with
O SafeFileEnvironment=/arch
then aliases should reference "/arch/rest/of/path". If a
value is given, sendmail also won't try to save to
/usr/tmp/dead.letter (instead it just leaves the job in the
queue as Qfxxxxxx). Inspired by *Hobbit*'s sendmail patch kit.
Support -A flag for alias files; this will comma concatenate like
entries. For example, given the aliases:
and an alias file declared as:
the final alias inserted will be "list: member1,member2";
without -A you will get an error on the second and subsequent
alias for "list". Contributed by Bryan Costales of ICSI.
Line-buffer transcript file. Suggested by Liudvikas Bukys.
Fix a problem that could cause very long addresses to core dump in
some special circumstances. Problem pointed out by Allan
(Internal change.) Change interface to expand() (macro expansion)
to be simpler and more consistent.
Delete check for funny qf file names. This didn't really give
any extra security and caused some people some problems.
(If you -really- want this, define PICKY_QF_NAME_CHECK
at compile time.) Suggested by Kyle Jones of UUNET.
(Internal change.) Change EF_NORETURN to EF_NO_BODY_RETN and
merge with DSN code; this is simpler and more consistent.
This may affect some people who have written their own
(Internal change.) Eliminate `D' line in qf file. The df file
is now assumed to be the same name as the qf file (with
the `q' changed to a `d', of course).
Avoid forking for delivery if all recipient mailers are marked as
"expensive" -- this can be a major cost on some systems.
Essentially, this forces sendmail into "queue only" mode
if all it is going to do is queue anyway.
Avoid sending a null message in some rather unusual circumstances
(specifically, the RCPT command returns a temporary
failure but the connection is lost before the DATA
command). Fix from Scott Hammond of Secure Computing
Change makesendmail to use a somewhat more rational naming scheme:
Makefiles and obj directories are named $os.$rel.$arch,
where $os is the operating system (e.g., SunOS), $rel is
the release number (e.g., 5.3), and $arch is the machine
architecture (e.g., sun4). Any of these can be omitted,
and anything after the first dot in a release number can
be replaced with "x" (e.g., SunOS.4.x.sun4). The previous
version used $os.$arch.$rel and was rather less general.
Ignore IDENT return value if the OSTYPE field returns "OTHER",
as indicated by RFC 1413. Pointed out by Kari Hurtta
of the Finnish Meteorological Institute.
Fix problem that could cause multiple responses to DATA command
on header syntax errors (e.g., lines beginning with colons).
Problem noted by Jens Thomassen of the University of Oslo.
Don't let null bytes in headers cause truncation of the rest of
Log Authentication-Warning:s. Suggested by Motonori Nakamura.
Increase timeouts on message data puts to allow time for receivers
to canonify addresses in headers on the fly. This is still
a rather ugly heuristic. From Motonori Nakamura.
Add "HasWildcardMX" suboption to ResolverOptions; if set, MX
records are not used when canonifying names. This is
useful if you have a wildcard MX record, although it
may cause other problems. In general, don't use wildcard
MX records. Patch from Motonori Nakamura.
Eliminate default two-line SMTP greeting message. Instead of
adding an extra "ESMTP spoken here" line, the word "ESMTP"
is added between the first and second word of the first
line of the greeting message (i.e., immediately after the
host name). This eliminates the need for the BROKEN_SMTP_PEERS
compile flag. Old sendmails won't see the ESMTP, but that's
acceptable because SIZE was the only useful extension that
old sendmails understand.
Avoid gethostbyname calls on UNIX domain sockets during SIGUSR1
invoked state dumps. From Masaharu Onishi.
Allow on-line comments in .forward and :include: files; they are
introduced by the string "<LWSP>#@#<LWSP>", where <LWSP>
is a space or a tab. This is intended for native
representation of non-ASCII sets such as Japanese, where
existing encodings would be unreadable or would lose
<motonori@cs.ritsumei.ac.jp> NAKAMURA Motonori
(romanized/less information)
<motonori@cs.ritsumei.ac.jp> =?ISO-2022-JP?B?GyRCQ2ZCPBsoQg==?=
=?ISO-2022-JP?B?GyRCQUdFNRsoQg==?=
(with MIME encoding, not human readable)
<motonori@cs.ritsumei.ac.jp> #@# ^[$BCfB<^[(B ^[$BAGE5^[(B
(native encoding with ISO-2022-JP)
The last form is human readable in the Japanese environment.
Based on a fix from (surprise!) Motonori Nakamura.
Don't make SMTP error returns on MAIL FROM: line be "sticky" for all
messages to that host; these are most frequently associated
with addresses rather than the host, with the exception of
421 (service shutting down). The effect was to cause queues
to sometimes take an excessive time to flush. Reported by
Robert Sargent of Southern Geographics Technologies.
Add Nice=N mailer option to set the niceness at which a mailer will
When looking for a default config file (that is, not specified using
a -C flag), try a configuration file name extended by the
binary version number -- e.g., sendmail.8.7.Alpha.9.cf,
sendmail.8.7.Alpha.cf, sendmail.8.7.cf, sendmail.8.cf, and
sendmail.cf in that order. This should make it easier to
test new versions in a shared environment.
Log queue runs that are skipped due to high loads. They are logged
at LOG_INFO priority iff the log level is > 8. Contributed
by Bruce Nagel of Data General.
Allow the error mailer to accept a DSN-style error status code
instead of an sysexits status code in the host part.
Anything with a dot will be interpreted as a DSN-style code.
Add new mailer flag: F=3 will tell translations to Quoted-Printable
to encode characters that might be munged by an EBCDIC system
in addition to the set required by RFC 1521. The additional
characters are !, ", #, $, @, [, \, ], ^, `, {, |, }, and ~.
(Think of "IBM 360" as the mnemonic for this flag.)
Change check for mailing to files to look for a pathname of [FILE]
rather than looking for the mailer named *file*. The mapping
of leading slashes still goes to the *file* mailer. This
allows you to implement the *file* mailer as a separate
program, for example, to insert a Content-Length: header
or do special security policy. However, note that the usual
initial checking for the file permissions is still done, and
the program in question needs to be very careful about how
it does the file write to avoid security problems.
Be able to read ~root/.forward even if the path isn't accessible to
regular users. This is disrecommended because sendmail
sometimes does not run as root (e.g., when an unsafe option
is specified on the command line), but should otherwise be
safe because .forward files must be owned by the user for
whom mail is being forwarded, and cannot be a symbolic link.
Suggested by Forrest Aldrich of Wang Laboratories.
Add new "HostsFile" option that is the pathname to the /etc/hosts
file. This is used for canonifying hostnames when the
Implement programs on F (read class from file) line. The syntax is
Fc|/path/to/program to read the output from the program
Probe the network interfaces to find alternate names for this
host. Requires the SIOCGIFCONF ioctl call. Code
Add "E" configuration line to set or propogate environment
variables into children. "E<envar>" will propogate
the named variable from the environment when sendmail
was invoked into any children it calls; "E<envar>=<value>"
sets the named variable to the indicated value. Any
variables not explicitly named will not be in the child
environment. However, sendmail still forces an
"AGENT=sendmail" environment variable, in part to enforce
at least one environment variable, since many programs and
libraries die horribly if this is not guaranteed.
Change heuristic for rebuilding both NEWDB and NDBM versions of
alias databases -- new algorithm looks for the substring
"/yp/" in the file name. This is more portable and involves
less overhead. Suggested by Motonori Nakamura.
Solaris 2 from Rob McMahon <cudcv@csv.warwick.ac.uk>.
System V Release 4 from Motonori Nakamura of Ritsumeikan
University. This expands the disk size
checking to include all (?) SVR4 configurations.
System V Release 4 from Kimmo Suominen -- initgroups(3)
and setrlimit(2) are both available.
System V Release 4 from sob@sculley.ffg.com -- some versions
apparently "have EX_OK defined in other headerfiles."
Linux getusershell(3) is broken in Slackware 2.0 --
from Andrew Pam of Xanadu Australia.
More Linux tweaking from John Kennedy of California State
Cray changes from Eric Wassenaar: ``On Cray, shorts,
ints, and longs are all 64 bits, and all structs
are multiples of 64 bits. This means that the
sizeof operator returns only multiples of 8.
This requires adaptation of code that really
deals with 32 bit or 16 bit fields, such as IP
addresses or nameserver fields.''
DG/UX 5.4.3 from Mark T. Robinson <mtr@ornl.gov>. To
get the old behaviour, use -DDGUX_5_4_2.
DG/UX hack: add _FORCE_MAIL_LOCAL_=yes environment
variable to fix bogus /bin/mail behaviour.
Tandem NonStop-UX from Rick McCarty <mccarty@mpd.tandem.com>.
This also cleans up some System V Release 4 compile
Solaris 2: sendmail.cw file should be in /etc/mail to
match all the other configuration files. Fix
from Glenn Barry of Emory University.
Solaris 2.3: compile problem in conf.c. Fix from Alain
Nissen of the University of Liege, Belgium.
Ultrix: freespace calculation was incorrect. Fix from
Takashi Kizu of Osaka University.
SVR4: running in background gets a SIGTTOU because the
emulation code doesn't realize that "getpeername"
doesn't require reading the file. Fix from Peter
Solaris 2.3: due to an apparent bug in the socket emulation
library, sockets can get into a "wedged" state where
they just return EPROTO; closing and re-opening the
socket clears the problem. Fix from Bob Manson
of Ohio State University.
Hitachi 3050R & 3050RX running HI-UX/WE2: portability
fixes from Akihiro Hashimoto ("Hash") of Chiba
AIX changes to allow setproctitle to work from Rainer Schöpf
of Zentrum für Datenverarbeitung der Universität
SCO Unix from Chip Rosenthal of Unicom (code was using the
ANSI C fixes from Adam Glass (NetBSD project).
Stardent Titan/ANSI C fixes from Kate Hedstrom of Rutgers
DG-UX fixes from Bruce Nagel of Data General.
IRIX64 updates from Mark Levinson of the University of
Rochester Medical Center.
Altos System V (``the first UNIX/XENIX merge the Altos
did for their Series 1000 & Series 2000 line;
their merged code was licenced back to AT&T and
Microsoft and became System V release 3.2'') from
OSF/1 running on Intel Paragon from Jeff A. Earickson
<jeff@ssd.intel.com> of Intel Scalable Systems
Amdahl UTS System V 2.1.5 (SVr3-based) from Janet Jackson
System V Release 4 (statvfs semantic fix) from Alain
HP-UX 10.x multiprocessor load average changes from
Scott Hutton and Jeff Sumler of Indiana University.
Cray CSOS from Scott Bolte of Cray Computer Corporation.
Unicos 8.0 from Douglas K. Rand of the University of North
Dakota, Scientific Computing Center.
Solaris 2.4 fixes from Sanjay Dani of Dani Communications.
ConvexOS 11.0 from Christophe Wolfhugel.
IRIX 4.0.5 from David Ashton-Reader of CADcentre.
ISC UNIX from J. J. Bailey.
HP-UX 9.xx on the 8xx series machines from Remy Giraud
HP-UX configuration from Tom Lane <tgl@sss.pgh.pa.us>.
IRIX 5.2 and 5.3 from Kari E. Hurtta.
FreeBSD 2.0 from Mike Hickey of Federal Data Corporation.
Sony NEWS-OS 4.2.1R and 6.0.3 from Motonori Nakamura.
Omron LUNA unios-b, mach from Motonori Nakamura.
NEC EWS-UX/V 4.2 from Motonori Nakamura.
NeXT 2.1 from Bryan Costales.
AUX patch thanks to Mike Erwin of Apple Computer.
HP-UX 10.0 from John Beck of Hewlett-Packard.
Ultrix: allow -DBROKEN_RES_SEARCH=0 if you are using a
non-DEC resolver. Suggested by Allan Johannesen.
MAKEMAP: allow -d flag to allow insertion of duplicate aliases
in type ``btree'' maps. The semantics of this are undefined
for regular maps, but it can be useful for the user database.
MAKEMAP: lock database file while rebuilding to avoid sendmail
lookups while the rebuild is going on. There is a race
condition between the open(... O_TRUNC ...) and the lock
on the file, but it should be quite small.
SMRSH: sendmail restricted shell added to the release. This can
be used as an alternative to /bin/sh for the "prog" mailer,
giving the local administrator more control over what
programs can be run from sendmail.
CONTRIB: a patch to rmail.c from Bill Gianopoulos of Raytheon
to allow rmail to compile on systems that don't have
function prototypes and systems that don't have snprintf.
CONTRIB: add the "mailprio" scripts that will help you sort mailing
lists by transaction delay times so that addresses that
respond quickly get sent first. This is to prevent very
sluggish servers from delaying other peoples' mail.
Contributed by Tony Sanders of BSDI.
CONTRIB: add the "bsdi.mc" file as contributed by Tony Sanders
of BSDI. This has a lot of comments to help people out.
CONFIG: fix mail from <> so it will properly convert to
MAILER-DAEMON on local addresses.
CONFIG: fix code that was supposed to catch colons in host
names. Problem noted by John Gardiner Myers of CMU.
CONFIG: allow use of SMTP_MAILER_MAX in nullclient configuration.
From Paul Riddle of the University of Maryland, Baltimore
CONFIG: Catch and reject "." as a host address.
CONFIG: Generalize domaintable to look up all domains, not
CONFIG: Delete OLD_SENDMAIL support -- as near as I can tell, it
was never used and didn't work anyway.
CONFIG: Set flags A, w, 5, :, /, |, and @ on the "local" mailer
and d on all mailers in the UUCP class.
CONFIG: Allow "user+detail" to be aliased specially: it will first
look for an alias for "user+detail", then for "user+*", and
finally for "user". This is intended for forwarding mail
for system aliases such as root and postmaster to a
CONFIG: add confEIGHT_BIT_HANDLING to set option 8 (see above).
CONFIG: add smtp8 mailer; this has the F=8 (just-send-8) flag set.
The F=8 flag is also set on the "relay" mailer, since
this is expected to be another sendmail.
CONFIG: avoid qualifying all UUCP addresses sent via SMTP with
the name of the UUCP_RELAY -- in some cases, this is the
wrong value (e.g., when we have local UUCP connections),
and this can create unreplyable addresses. From Chip
CONFIG: add confRECEIVED_HEADER to change the format of the
Received: header inserted into all messages. Suggested by
Gary Mills of the University of Manitoba.
CONFIG: Make "notsticky" the default; use FEATURE(stickyhost)
to get the old behaviour. I did this upon observing
that almost everyone needed this feature, and that the
concept I was trying to make happen didn't work with
some user agents anyway. FEATURE(notsticky) still works,
CONFIG: Add LUSER_RELAY -- the host to which unrecognized user
names are sent, rather than immediately diagnosing them
CONFIG: Add SMTP_MAILER_ARGS, ESMTP_MAILER_ARGS, SMTP8_MAILER_ARGS,
and RELAY_MAILER_ARGS to set the arguments for the
indicated mailers. All default to "IPC $h". Patch from
Larry Parmelee of Cornell University.
CONFIG: pop mailer needs F=n flag to avoid "annoying side effects
on the client side" and F=P to get an appropriate
return-path. From Kimmo Suominen.
CONFIG: add FEATURE(local_procmail) to use the procmail program
as the local mailer. For addresses of the form "user+detail"
the "detail" part is passed to procmail via the -a flag.
Contributed by Kimmo Suominen.
CONFIG: add MAILER(procmail) to add an interface to procmail for
use from mailertables. This lets you execute arbitrary
procmail scripts. Contributed by Kimmo Suominen.
CONFIG: add T= fields (MTS type) to local, smtp, and uucp mailers.
CONFIG: add OSTYPE(ptx2) for DYNIX/ptx 2.x from Sequent. From
Paul Southworth of CICNet Systems Support.
CONFIG: use -a$g as default to UUCP mailers, instead of -a$f.
This causes the null return path to be rewritten as
MAILER-DAEMON; otherwise UUCP gets horribly confused.
From Michael Hohmuth of Technische Universitat Dresden.
CONFIG: Add FEATURE(bestmx_is_local) to cause any hosts that
list us as the best possible MX record to be treated as
though they were local (essentially, assume that they
are included in $=w). This can cause additional DNS
traffic, but is easier to administer if this fits your
local model. It does not work reliably if there are
multiple hosts that share the best MX preference.
Code contributed by John Oleynick of Rutgers.
CONFIG: Add FEATURE(smrsh) to use smrsh (the SendMail Restricted
SHell) instead of /bin/sh as the program used for delivery
to programs. If an argument is included, it is used as
the path to smrsh; otherwise, /usr/local/etc/smrsh is
CONFIG: Add LOCAL_MAILER_MAX and PROCMAILER_MAILER_MAX to limit the
size of messages to the local and procmail mailers
respectively. Contributed by Brad Knowles of the Defense
Information Systems Agency.
CONFIG: Handle leading ``phrase:'' and trailing ``;'' as comments
(just like text outside of angle brackets) in order to
properly deal with ``group: addr1, ... addrN;'' syntax.
CONFIG: Require OSTYPE macro (the defaults really don't apply to
any real systems any more) and tweak the DOMAIN macro
so that it is less likely that users will accidently use
the Berkeley defaults. Also, create some generic files
that really can be used in the real world.
CONFIG: Add new configuration macros to set character sets for
messages _arriving from_ various mailers: LOCAL_MAILER_CHARSET,
SMTP_MAILER_CHARSET, and UUCP_MAILER_CHARSET.
CONFIG: Change UUCP_MAX_SIZE to UUCP_MAILER_MAX for consistency.
The old name will still be accepted for a while at least.
CONFIG: Implement DECNET_RELAY as spec for host to which DECNET
mail (.DECNET pseudo-domain or node::user) will be sent.
As with all relays, it can be ``mailer:hostname''. Suggested
CONFIG: Add MAILER(mail11) to get DECnet support. Code contributed
by Barb Dijker of Labyrinth Computer Services.
CONFIG: change confCHECK_ALIASES to default to False -- it has poor
performance for large alias files, and this confused many
cf/cf/generic-solaris2.mc
cf/cf/generic-sunos4.1.mc
cf/domain/berkeley-only.m4
cf/feature/bestmx_is_local.m4
cf/feature/local_procmail.m4
cf/feature/stickydomain.m4
contrib/rmail.oldsys.patch
src/Makefiles/Makefile.CSOS
src/Makefiles/Makefile.EWS-UX_V
src/Makefiles/Makefile.IRIX.5.x
src/Makefiles/Makefile.IRIX64
src/Makefiles/Makefile.ISC
src/Makefiles/Makefile.NEWS-OS.4.x
src/Makefiles/Makefile.NEWS-OS.6.x
src/Makefiles/Makefile.NonStop-UX
src/Makefiles/Makefile.Paragon
src/Makefiles/Makefile.SunOS.5.3
src/Makefiles/Makefile.SunOS.5.4
src/Makefiles/Makefile.UNIX_SV.4.2.i386
src/Makefiles/Makefile.uts.systemV
cf/cf/alpha.mc => cf/cf/s2k-osf1.mc
cf/cf/chez.mc => cf/cf/chez.cs.mc
cf/cf/hpux-cs-exposed.mc => cf/cf/cs-hpux9.mc
cf/cf/osf1-cs-exposed.mc => cf/cf/cs-osf1.mc
cf/cf/s2k.mc => cf/cf/s2k-ultrix4.mc
cf/cf/sunos4.1-cs-exposed.mc => cf/cf/cs-sunos4.1.mc
cf/cf/ultrix4.1-cs-exposed.mc => cf/cf/cs-ultrix4.mc
cf/cf/vangogh.mc => cf/cf/vangogh.cs.mc
cf/domain/Berkeley.m4 => cf/domain/Berkeley.EDU.m4
cf/domain/cs-exposed.m4 => cf/domain/CS.Berkeley.EDU.m4
cf/domain/eecs-hidden.m4 => cf/domain/EECS.Berkeley.EDU.m4
cf/domain/s2k.m4 => cf/domain/S2K.Berkeley.EDU.m4
cf/ostype/hpux.m4 => cf/ostype/hpux9.m4
cf/ostype/ultrix4.1.m4 => cf/ostype/ultrix4.m4
src/Makefile.* => src/Makefiles/Makefile.*
cf/cf/sunos3.5-cs-exposed.mc
cf/cf/sunos3.5-cs-hidden.mc
cf/cf/sunos4.1-cs-hidden.mc
cf/cf/ultrix4.1-cs-hidden.mc
src/Makefiles/Makefile.SunOS.5.x
Fix to IDENT code (it was getting the size of the reply buffer
too small, so nothing was ever accepted). Fix from several
people, including Allan Johannesen, Shane Castle of the
Boulder County Information Services, and Jeff Smith of
Warwick University (all arrived within a few hours of
Fix a problem that could cause large jobs to run out of
file descriptors on systems that use vfork() rather
The ``possible attack'' message would be logged more often
than necessary if you are using Pine as a user agent.
The wrong host would be reported in the ``possible attack''
message when attempted from IDENT.
In some cases the syslog buffer could be overflowed when
reporting the ``possible attack'' message. This can
cause denial of service attacks. Truncate the message
to 80 characters to prevent this problem.
When reading the IDENT response a loop is needed around the
read from the network to ensure that you don't get
Password entries without any shell listed (that is, a null
shell) wouldn't match as "ok". Problem noted by
When running BIND 4.9.x a problem could occur because the
_res.options field is initialized differently than it
was historically -- this requires that sendmail call
res_init before it tweaks any bits.
Fix an incompatibility in openxscript() between the file open mode
and the stdio mode passed to fdopen. This caused UnixWare
2.0 to have conniptions. Fix from Martin Sohnius of
Fix problem with static linking of local getopt routine when
using GNU's ld command. Fix from John Kennedy of
It was possible to turn off privacy flags. Problem noted by
Be more paranoid about writing files. Suggestions by *Hobbit*
MAKEMAP: fixes for 64 bit machines (DEC Alphas in particular)
CONFIG: No changes (version number only, to keep it in sync
SECURITY: Diagnose bogus values to some command line flags that
could allow trash to get into headers and qf files.
Validate the name of the user returned by the IDENT protocol.
Some systems that really dislike IDENT send intentionally
bogus information. Problem pointed out by Michael Bushnell
of the Free Software Foundation. Has some security
Fix a problem causing error messages about DNS problems when
the host name contained a percent sign to act oddly
because it was passed as a printf-style format string.
In some cases this could cause core dumps.
Avoid possible buffer overrun in returntosender() if error
message is quite ling. From Fletcher Mattox of the
Fix a problem that would silently drop "too many hops" error
messages if and only if you were sending to an alias.
From Jon Giltner of the University of Colorado and
Dan Harton of Oak Ridge National Laboratory.
Fix a bug that caused core dumps on some systems if -d11.2 was
set and e->e_message was null. Fix from Bruce Nagel of
Fix problem that can still cause df files to be left around
after "hop count exceeded" messages. Fix from Andrew
Chang and Shau-Ping Lo of SunSoft.
Fix a problem that can cause buffer overflows on very long
user names (as might occur if you piped to a program
with a lot of arguments).
Avoid returning an error and re-queueing if the host signature
is null; this can occur on addresses like ``user@.''.
Problem noted by Wesley Craig and the University of
Avoid possible calls to malloc(0) if MCI caching is turned
off. Bug fix from Pierre David of the Laboratoire
Parallelisme, Reseaux, Systemes et Modelisation (PRiSM),
Universite de Versailles - St Quentin, and Jacky
Make a local copy of the line being sent via senttolist() -- in
some cases, buffers could get trashed by map lookups
causing it to do unexpected things. This also simplifies
CONFIG: No changes (version number only, to keep it in sync
Do all mail delivery completely disconnected from any terminal.
This provides consistency with daemon delivery and
may have some security implications.
Make sure that malloc doesn't get called with zero size,
since that fails on some systems. Reported by Ed
Hill of the University of Iowa.
Fix multi-line values for $e (SMTP greeting message). Reported
by Mike O'Connor of Ford Motor Company.
Avoid syserr if no NIS domain name is defined, but the map it
is trying to open is optional. From Win Bent of USC.
Changes for picky compilers from Ed Gould of Digital Equipment.
Hesiod support for UDB from Todd Miller of the University of
Colorado. Use "hesiod" as the service name in the U
Fix a problem that failed to set the "authentic" host name (that
is, the one derived from the socket info) if you called
sendmail -bs from inetd. Based on code contributed by
Todd Miller (this problem was also reported by Guy Helmer
of Dakota State University). This also fixes a related
problem reported by Liudvikas Bukys of the University of
Parameterize "nroff -h" in all the Makefiles so people with
variant versions can use them easily. Suggested by
Peter Collinson of Hillside Systems.
SMTP "MAIL" commands with multiple ESMTP parameters required two
spaces between parameters instead of one. Reported by
Valdis Kletnieks of Virginia Tech.
Reduce the number of system calls during message collection by
using global timeouts around the collect() loop. This
code was contributed by Eric Wassenaar.
If the initial hostname name gathering results in a name
without a dot (usually caused by NIS misconfiguration)
and BIND is compiled in, directly access DNS to get
the canonical name. This should make life easier for
Solaris systems. If it still can't be resolved, and
if the name server is listed as "required", try again
in 30 seconds. If that also fails, exit immediately to
avoid bogus "config error: mail loops back to myself"
Improve the "MAIL DELETED BECAUSE OF LACK OF DISK SPACE" error
message to explain how much space was available and
sound a bit less threatening. Suggested by Stan Janet
of the National Institute of Standards and Technology.
If mail is delivered to an alias that has an owner, deliver any
requested return-receipt immediately, and strip the
Return-Receipt-To: header from the subsequent message.
This prevents a certain class of denial of service
attack, arguably gives more reasonable semantics, and
moves things more towards what will probably become a
network standard. Suggested by Christopher Davis of
Add a "noreceipts" privacy flag to turn off all return receipts
Avoid printing ESMTP parameters as part of the error message
if there are errors during parsing. This change is
Avoid sending out error messages during the collect phase of
SMTP; there is an MVS mailer from UCLA that gets
confused by this. Of course, I think it's their bug....
Check for the $j macro getting undefined, losing a dot, or getting
lost from $=w in the daemon before accepting a connection;
if it is, it dumps state, prints a LOG_ALERT message,
and drops core for debugging. This is an attempt to
track down a bug that I thought was long since gone.
If you see this, please forward the log fragment to
sendmail@CS.Berkeley.EDU.
Change OLD_NEWDB from a #ifdef to a #if so it can be turned off
with -DOLD_NEWDB=0 on the command line. From Christophe
Instead of trying to truncate the listen queue for the server
SMTP port when the load average is too high, just close
the port completely and reopen it later as needed.
This ensures that the other end gets a quick "connection
refused" response, and that the connection can be
recovered later. In particular, some socket emulations
seem to get confused if you tweak the listen queue
size around and can never start listening to connections
again. The down side is that someone could start up
another daemon process in the interim, so you could
have multiple daemons all not listening to connections;
this could in turn cause the sendmail.pid file to be
incorrect. A better approach might be to accept the
connection and give a 421 code, but that could break
other mailers in mysterious ways and have paging behaviour
Fix a glitch in TCP-level debugging that caused flag 16.101 to
set debugging on the wrong socket. From Eric Wassenaar.
When creating a df* temporary file, be sure you truncate any
existing data in the file -- otherwise system crashes
and the like could result in extra data being sent.
DOC: Replace the CHANGES-R5-R8 readme file with a paper in the
doc directory. This includes some additional
CONFIG: change UUCP rules to never add $U! or $k! on the front
of recipient envelope addresses. This should have been
handled by the $&h trick, but broke if people were
mixing domainized and UUCP addresses. They should
probably have converted all the way over to uucp-uudom
instead of uucp-{new,old}, but the failure mode was to
loop the mail, which was bad news.
Newer BSDI systems (several people).
Older BSDI systems from Christophe Wolfhugel.
Intergraph CLIX, from Paul Southworth of CICNet.
UnixWare, from Evan Champion.
Solaris from Quentin Campbell of the University of
IRIX from Dean Cookson and Bill Driscoll of Mitre
NCR 3000 from Kevin Darcy of Chrysler Financial Corporation.
SunOS (it has setsid() and setvbuf() calls) from
Jonathan Kamens of OpenVision Technologies.
HP-UX from Tor Lillqvist.
SECURITY: it was possible to read any file as root using the
E (error message) option. Reported by Richard Jones;
fixed by Michael Corrigan and Christophe Wolfhugel.
SECURITY: it was possible to get root access by using wierd
values to the -d flag. Thanks to Alain Durand of
INRIA for forwarding me the notice from the bugtraq
SECURITY: the ability to give files away on System V-based
systems proved dangerous -- don't run as the owner
of a :include: file on a system that allows giveaways.
Unfortunately, this also applies to determining a
IMPORTANT: Previous versions weren't expiring old connections
in the connection cache for a long time under some
circumstances. This could result in resource exhaustion,
both at your end and at the other end. This checks the
connections for timeouts much more frequently. From
Fix a glitch that snuck in that caused programs to be run as
the sender instead of the recipient if the mail was
from a local user to another local user. From
Motonori Nakamura of Kyoto University.
Fix "wildcard" on /etc/shell matching -- instead of looking
for "*", look for "/SENDMAIL/ANY/SHELL/". From
Change the method used to declare the "statfs" availability;
instead of HASSTATFS and/or HASUSTAT with a ton of
tweaking in conf.c, there is a single #define called
SFS_TYPE which takes on one of six values (SFS_NONE
for no statfs availability, SFS_USTAT for the ustat(2)
syscall, SFS_4ARGS for a four argument statfs(2) call,
and SFS_VFS, SFS_MOUNT, or SFS_STATFS for a two argument
statfs(2) call with the declarations in <sys/vfs.h>,
<sys/mount.h>, or <sys/statfs.h> respectively).
Fix glitch in NetInfo support that could return garbage if
there was no "/locations/sendmail" property. From
David Meyer of the University of Virginia.
Change HASFLOCK from defined/not-defined to a 0/1 definition
to allow Linux to turn it off even though it is a
Allow setting of "ident" timeout to zero to turn off the ident
Make 7-bit stripping local to a connection (instead of to a
mailer); this allows you to specify that SMTP is a
7-bit channel, but revert to 8-bit should it advertise
that it supports 8BITMIME. You still have to specify
mailer flag 7 to get this stripping at all.
Improve makesendmail script so it handles more cases automatically.
Tighten up restrictions on taking ownership of :include: files
to avoid problems on systems that allow you to give away
Fix a problem that made it impossible to rebuild the alias
file if it was on a read-only file system. From
Harry Edmon of the University of Washington.
Improve MX randomization function. From John Gardiner Myers
Fix a minor glitch causing a bogus message to be printed (used
%s instead of %d in a printf string for the line number)
when a bad queue file was read. From Harry Edmon.
Allow $s to remain NULL on locally generated mail. I'm not
sure this is necessary, but a lot of people have complained
about it, and there is a legitimate question as to whether
"localhost" is legal as an 822-style domain.
Fix a problem with very short line lengths (mailer L= flag) in
headers. This causes a leading space to be added onto
continuation lines (including in the body!), and also
tries to wrap headers containing addresses (From:, To:,
etc) intelligently at the shorter line lengths. Problem
Reported by Lars-Johan Liman of SUNET Operations Center.
Log the real user name when logging syserrs, since these can have
security implications. Suggested by several people.
Fix address logging of cached connections -- it used to always
log the numeric address as zero. This is a somewhat
bogus implementation in that it does an extra system
call, but it should be an inexpensive one. Fix from
Tighten up handling of short syslog buffers even more -- there
were cases where the outgoing relay= name was too long
to share a line with delay= and mailer= logging.
Limit the overhead on split envelopes to one open file descriptor
per envelope -- previously the overhead was three
descriptors. This was in response to a problem reported
by P{r (Pell) Emanuelsson.
Fixes to better handle the case of unexpected connection closes;
this redirects the output to the transcript so the info
is not lost. From Eric Wassenaar.
Fix potential string overrun if you macro evaluate a string that
has a naked $ at the end. Problem noted by James Matheson
Make default error number on $#error messages 553 (``Requested
action not taken: mailbox name not allowed'') instead of
501 (``Syntax error in parameters or arguments'') to
avoid bogus "protocol error" messages.
Strip off any existing trailing dot on names during $[ ... $]
lookup. This prevents it from ending up with two dots
on the end of dot terminated names. From Wesley Craig
of the University of Michigan and Bryan Costales of ICSI.
Clean up file class reading so that the debugging information is
more informative. It hadn't been using setclass, so you
didn't see the class items being added.
Avoid core dump if you are running a version of sendmail where
NIS is compiled in, and you specify an NIS map, but
NIS is not running. Fix from John Oleynick of
Diagnose bizarre case where res_search returns a failure value,
but sets h_errno to a success value.
Make sure that "too many hops" messages are considered important
enough to send an error to the Postmaster (that is, the
address specified in the P option). This fix should
help problems that cause the df file to be left around
sometimes -- unfortunately, I can't seem to reproduce
Avoid core dump (null pointer reference) on EXPN command; this
only occurred if your log level was set to 10 or higher
and the target account was an alias or had a .forward file.
Problem noted by Janne Himanka.
Avoid "denial of service" attacks by someone who is flooding your
SMTP port with bad commands by shutting the connection
after 25 bad commands are issued. From Kyle Jones of
Fix core dump on error messages with very long "to" buffers;
fmtmsg overflows the message buffer. Fixed by trimming
the to address to 203 characters. Problem reported by
Fix configuration for HASFLOCK -- there were some spots where
a #ifndef was incorrectly #ifdef. Pointed out by
George Baltz of the University of Maryland.
Fix a typo in savemail() that could cause the error message To:
lists to be incorrect in some places. From Motonori
Fix a glitch that can cause duplicate error messages on split
envelopes where an address on one of the lists has a
name server failure. Fix from Voradesh Yenbut of the
University of Washington.
Fix possible bogus pointer reference on ESMTP parameters that
don't have an ``=value'' part.
CNAME loops caused an error message to be generated, but also
re-queued the message. Changed to just re-queue the
message (it's really hard to just bounce it because
of the wierd way the name server works in the presence
of CNAME loops). Problem noted by James M.R.Matheson
Avoid giving ``warning: foo owned process doing -bs'' messages
if they use ``MAIL FROM:<foo>'' where foo is their true
user name. Suggested by Andreas Stolcke of ICSI.
Change the NAMED_BIND compile flag to be a 0/1 flag so you can
override it easily in the Makefile -- that is, you can
turn it off using -DNAMED_BIND=0.
If a gethostbyname(...) of an address with a trailing dot fails,
try it without the trailing dot. This is because if
you have a version of gethostbyname() that falls back
to NIS or the /etc/hosts file it will fail to find
perfectly reasonable names that just don't happen to
be dot terminated in the hosts file. You don't want to
strip the dot first though because we're trying to ensure
that country names that match one of your subdomains get
PRALIASES: fix bogus output on non-null-terminated strings.
From Bill Gianopoulos of Raytheon.
CONFIG: Avoid rewriting anything that matches $w to be $j.
This was in code intended to only catch the self-literal
address (that is, [1.2.3.4], where 1.2.3.4 is your
IP address), but the code was broken. However, it will
still do this if $M is defined; this is necessary to
get client configurations to work (sigh). Note that this
means that $M overrides :mailname entries in the user
database! Problem noted by Paul Southworth.
CONFIG: Fix definition of Solaris help file location. From
Steve Cliffe <steve@gorgon.cs.uow.edu.au>.
CONFIG: Fix bug that broke news.group.USENET mappings.
CONFIG: Allow declaration of SMTP_MAILER_MAX, FAX_MAILER_MAX,
and USENET_MAILER_MAX to tweak the maximum message
size for various mailers.
CONFIG: Change definition of USENET_MAILER_ARGS to include argv[0]
instead of assuming that it is "inews" for consistency
with other mailers. From Michael Corrigan of UC San Diego.
CONFIG: When mail is forwarded to a LOCAL_RELAY or a MAIL_HUB,
qualify the address in the SMTP envelope as user@{relay|hub}
instead of user@$j. From Bill Wisner of The Well.
CONFIG: Fix route-addr syntax in nullrelay configuration set.
CONFIG: Don't turn off case mapping of user names in the local
mailer for IRIX. This was different than most every other
CONFIG: Avoid infinite loops on certainly list:; syntaxes in
envelope. Noted by Thierry Besancon
<besancon@excalibur.ens.fr>.
CONFIG: Don't include -z by default on uux line -- most systems
don't want it set by default. Pointed out by Philippe
CONFIG: Fix some bugs with mailertables -- for example, if your
host name was foo.bar.ray.com and you matched against
".ray.com", the old implementation bound %1 to "bar"
instead of "foo.bar". Also, allow "." in the mailertable
to match anything -- essentially, take over SMART_HOST.
This also moves matching of explicit local host names
before the mailertable so they don't have to be special
cased in the mailertable data. Reported by Bill
Gianopoulos of Raytheon; the fix for the %1 binding
problem was contributed by Nicholas Comanos of the
CONFIG: Don't include "root" in class $=L (users to deliver
locally, even if a hub or relay exists) by default.
This is because of the known bug where definition of
both a LOCAL_RELAY and a MAIL_HUB causes $=L to ignore
both and deliver into the local mailbox.
CONFIG: Move up bitdomain and uudomain handling so that they
are done before .UUCP class matching; uudomain was
reported as ineffective before. This also frees up
diversion 8 for future use. Problem reported by Kimmo
CONFIG: Don't try to convert dotted IP address (e.g., [1.2.3.4])
into host names. As pointed out by Jonathan Kamens,
these are often used because either the forward or reverse
mapping is broken; this translation makes it broken again.
DOC: Clarify $@ and $: in the Install & Op Guide. From Kimmo
Unicos from David L. Kensiski of Sterling Sofware.
DomainOS from Don Lewis of Silicon Systems.
GNU m4 1.0.3 from Karst Koymans of Utrecht University.
Convex from Kimmo Suominen <kim@tac.nyc.ny.us>.
NetBSD from Adam Glass <glass@sun-lamp.cs.berkeley.edu>.
BSD/386 from Tony Sanders of BSDI.
Apollo from Eric Wassenaar.
Sequent DYNIX/ptx 2.0 from Tim Wright of Sequent.
Security fix: /.forward could be owned by anyone (the test
to allow root to own any file was backwards). From
Bob Campbell at U.C. Berkeley.
Security fix: group ids were not completely set when programs
were invoked. This caused programs to have group
permissions they should not have had (usually group
daemon instead of their own group). In particular,
Perl scripts would refuse to run.
Security: check to make sure files that are written are not
symbolic links (at least under some circumstances).
Although this does not respond to a specific known
attack, it's just a good idea. Suggested by
Security fix: if a user had an NFS mounted home directory on
a system with a restricted shell listed in their
/etc/passwd entry, they could still execute any
program by putting that in their .forward file.
This fix prevents that by insisting that their shell
appear in /etc/shells before allowing a .forward to
execute a program or write a file. You can disable
this by putting "*" in /etc/shells. It also won't
permit world-writable :include: files to reference
programs or files (there's no way to disable this).
These behaviours are only one level deep -- for
example, it is legal for a world-writable :include:
file to reference an alias that writes a file, on
the assumption that the alias file is well controlled.
Security fix: root was not treated suspiciously enough when
looking into subdirectories. This would potentially
allow a cracker to examine files that were publically
readable but in a non-publically searchable directory.
Fix a problem that causes an error on QUIT on a cached
connection to create problems on the current job.
These are typically unrelated, so errors occur in
Reset CurrentLA in sendall() -- this makes sendmail queue
runs more responsive to load average, and fixes a
problem that ignored the load average in locally
generated mail. From Eric Wassenaar.
Fix possible core dump on aliases with null LHS. From
Revert to using flock() whenever possible -- there are just
too many bugs in fcntl() locking, particularly over
NFS, that cause sendmail to fail in perverse ways.
Fix a bug that causes the connection cache to get confused
when sending error messages. This resulted in
"unexpected close" messages. It should fix itself
on the following queue run. Problem noted by
Liudvikas Bukys of the University of Rochester.
Include $k in $=k as documented in the Install & Op Guide.
This seems odd, but it was documented.... From
Michael Corrigan of UCSD.
Fix problem that caused :include:s from alias files to be
forced to be owned by root instead of daemon
(actually DefUid). From Tim Irvin.
Diagnose unrecognized I option values -- from Mortin Forssen
of the Chalmers University of Technology.
Make "error" mailer work consistently when there is no error
code associated with it -- previously it returned OK
even though there was a real problem. Now it assumes
Fix bug that caused the last header line of messages that had
no body and which were terminated with EOF instead of
"." to be discarded. Problem noted by Liudvikas Bukys.
Fix core dump on SMTP mail to programs that failed -- it tried
to go to a "next MX host" when none existed, causing
a core dump. From der Mouse at McGill University.
Change IDENTPROTO from a defined/not defined to a 0/1 switch;
this makes it easier to turn it off (using
-DIDENTPROTO=0 in the Makefile). From der Mouse.
Fix YP_MASTER_NAME store to use the unupdated result of
gethostname() (instead of myhostname(), which tries
to fully qualify the name) to be consistent with
SunOS. If your hostname is unqualified, this fixes
transfers to slave servers. Bug noted by Keith
McMillan of Ameritech Services, Inc.
Fix Ultrix problem: gethostbyname() can return a very large
(> 500) h_length field, which causes the sockaddr
to be trashed. Use the size of the sockaddr instead.
Fix from Bob Manson of Ohio State.
Don't assume "-a." on host lookups if NAMED_BIND is not
defined -- this confuses gethostbyname on hosts
file lookups, which doesn't understand the trailing
Log SMTP server subprocesses that die with a signal instead
If you don't have option "I" set, don't assume that a DNS
"host unknown" message is authoritative -- it
might still be found in /etc/hosts.
Fix a problem that would cause Deferred: messages to be sent
as the subject of an error message, even though the
actual cause of a message was more severe than that.
Problem noted by Chris Seabrook of OSSI.
Fix race condition in DBM alias file locking. From Kyle
Limit delivery syslog line length to avoid bugs in some
versions of syslog(3). This adds a new compile time
variable SYSLOG_BUFSIZE. From Jay Plett of Princeton
University, which is in turn derived from IDA.
Fix quotes inside of comments in addresses -- previously
it insisted that they be balanced, but the 822 spec
says that they should be ignored.
Dump open file state to syslog upon receiving SIGUSR1 (for
debugging). This also evaluates ruleset 89, if set
(with the null input), and logs the result. This
should be used sparingly, since the rewrite process
Change -qI, -qR, and -qS flags to be case-insensitive as
documented in the Bat Book.
If the mailer returned EX_IOERR or EX_OSERR, sendmail did not
return an error message and did not requeue the message.
Fix based on code from Roland Dirlewanger of
Reseau Regional Aquarel, Bordeaux, France.
Fix a problem that caused a seg fault if you got a 421 error
code during some parts of connection initialization.
I've only seen this when talking to buggy mailers on
the other end, but it shouldn't give a seg fault in
any case. From Amir Plivatsky.
Fix core dump caused by a ruleset call that returns null.
Fix from Bryan Costales of ICSI.
Full-Name: field was being ignored. Fix from Motonori Nakamura
Fix a possible problem with very long input lines in setproctitle.
Avoid putting "This is a warning message" out on return receipts.
Suggested by Douglas Anderson.
Detect loops caused by recursive ruleset calls. Suggested by
Initialize non-alias maps during alias rebuilds -- they may be
needed for parsing. Problem noted by Douglas Anderson.
Log sender address even if no message was collected in SMTP
(e.g., if all RCPTs failed). Suggested by Motonori
Don't reflect the owner-list contents into the envelope sender
address if the value contains ", :, /, or | (to avoid
illegal addresses appearing there).
Efficiency hack for toktype macro -- from Craig Partridge of
Clean up DNS error printing so that a host name is always
Remember to set $i during queue runs. Reported by Stephen
Campbell of Dartmouth University.
If ${HOSTALIASES} is set, use it during canonification so that
headers are properly mapped. Reported by Anne Bennett
Avoid printing misleading error message if SMTP mailer (not
using [IPC]) should die on a core dump.
Avoid incorrect diagnosis of "file 1 closed" when it is caused
by the other end closing the connection. From
Improve several of the error messages printed by "mailq"
to include a host name or other useful information.
Add NetInfo preliminary support for NeXT systems. From Vince
Fix a glitch that sometimes caused :include:s that pointed to
NFS filesystems that were down to give an "aliasing/
forwarding loop broken" message instead of queueing
the message for retry. Noted by William C Fenner of
the NRL Connection Machine Facility.
Fix a problem that could cause a core dump if the input sequence
had (or somehow acquired) a \231 character.
Make sure that route-addrs always have <angle brackets> around
them in non-SMTP envelopes (SMTP envelopes already do
Avoid wierd headers on unbalanced punctuation of the form:
``Joe User <user)'' -- this caused reference to the
null macro. Fix from Rick McCarty of IO.COM.
Fix a problem that caused an alias "user: user@local.host" to
not have the QNOTREMOTE bit set; this caused configs
to act as if FEATURE(notsticky) was defined even when
it was not. The effect of the problem was to make it
very hard to to set up satellite sites that had a few
local accounts, with everything else forwarded to a
corporate hub. Reported by Detlef Drewanz of the
University of Rostock and Mark Frost of NCD.
Change queuing to not call rulesets 3, {1 or 2}, 4 on header
addresses. This is more efficient (fewer name server
calls) and fixes certain unusual configurations, such
as those that have ruleset 4 do something that is
non-idempotent unless a mailer-specific ruleset did
something else. Problem reported by Brian J. Coan
of the Institute for Global Communications.
Fix the "obsolete argument" routine in main to better understand
new arguments. For example, if you used ``sendmail
-C config -v -q'' it would choke on the -q because
the -C would stop looking for old-format arguments.
Fix the code that was intended to allow two users to forward their
mail to the same program and have them appear unique.
SCO UNIX from Murray Kucherawy.
SCO Open Server 3.2v4 from Philippe Brand.
System V Release 4 from Rick Ellis and others.
OSF/1 from Steve Campbell.
DG/UX from Ben Mesander of the USGS and Bryan Curnutt
Motorola SysV88 from Kevin Johnson of Motorola.
Solaris 2.3 from Casper H.S. Dik of the University
of Amsterdam and John Caruso of University
FreeBSD from Ollivier Robert.
TitanOS from Kate Hedstrom of Rutgers University.
Dynix from Jim Davis of the University of Arizona.
Linux from John Kennedy of California State University
Solaris 2.x from Tony Boner of the U.S. Air Force.
NEXTSTEP 3.x from Vince DeMarco.
HP-UX from various people. NOTA BENE: the location
of the config file has moved to /usr/lib
to match the HP-UX version of sendmail.
CONFIG: Don't do any recipient rewriting on relay mailer;
since this is intended only for internal use, the
usual RFC 821/822/1123 rules can be relaxed. The
main point of this is to avoid munging (ugh) UUCP
addresses when relaying internally.
CONFIG: fix typo in mailer/uucp.m4 that mutilates list:;
syntax addresses delivered via UUCP. Solution
CONFIG: fix thumb-fumble in default UUCP relaying in ruleset
zero; it caused double @ signs in addresses. From
Irving Reid of the University of Toronto.
CONFIG: Portability fixes for SCO Unix 3.2 with TCP/IP 1.2.1
from Markku Toijala of ICL Personal Systems Oy.
CONFIG: Add trailing "." on pseudo-domains for consistency;
this fixes a problem (noted by Al Whaley of Sunnyside)
that made it hard to recognize your own pseudodomain
CONFIG: catch "@host" syntax errors (i.e., null local-parts)
rather than letting them get "local configuration
error"s. Problem noted by John Gardiner Myers.
CONFIG: add uucp-uudom mailer variant, based on code posted
by Spider Boardman <spider@Orb.Nashua.NH.US>; this
has uucp-dom semantics but old UUCP syntax. This
also permits "uucp-old" as an alias for "uucp" and
"uucp-new" as a synonym for "suucp" for consistency.
CONFIG: add POP mailer support (from Kimmo Suominen
CONFIG: drop CSNET_RELAY support -- CSNET is long gone.
CONFIG: fix bug caused with domain literal addresses (e.g.,
``[128.32.131.12]'') when FEATURE(allmasquerade)
was set; it would get an additional @masquerade.host
added to the address. Problem noted by Peter Wan
CONFIG: make sure that the local UUCP name is in $=w. From
CONFIG: changes to UUCP rewriting to simulate IDA-style "V"
mailer flag. Briefly, if you are sending to host
"foo", then it rewrites "foo!...!baz" to "...!baz",
"foo!baz" remains "foo!baz", and anything else has
the local name prepended.
CONFIG: portability fixes for HP-UX.
DOC: several minor problems fixed in the Install & Op Guide.
MAKEMAP: fix core dump problem on lines that are too long or
which lack newline. From Mark Delany.
MAILSTATS: print sums of columns (total messages & kbytes
in and out of the system). From Tom Ferrin of UC
San Francisco Computer Graphics Lab.
SIGNIFICANT USER- OR SYSAD-VISIBLE CHANGES:
On HP-UX, /etc/sendmail.cf has been moved to
/usr/lib/sendmail.cf to match HP sendmail.
Permissions have been tightened up on world-writable
:include: files and accounts that have shells
that are not listed in /etc/shells. This may
cause some .forward files that have worked
SIGUSR1 dumps some state to the log.
Repair core-dump problem (write to read-only memory segment)
if you fall back to the return-to-Postmaster case in
savemail. Problem reported by Richard Liu.
Immediately diagnose bogus sender addresses in SMTP. This
makes quite certain that crackers can't use this
Reliability Fix: check return value from fclose() and fsync()
in a few critical places.
Minor problem in initsys() that reversed a condition for
redirecting the output channel on queue runs. It's
not clear this code even does anything. From Eric
Wassenaar of the Dutch National Institute for Nuclear
Fix some problems that caused queue runs to do "too much work",
such as double-reading the Errors-To: header. From
Error messages on writing the temporary file (including the
data file) were getting suppressed in SMTP -- this
fix causes them to be properly reported. From Eric
Some changes to support AF_UNIX sockets -- this will only
really become relevant in the next release, but some
people need it for local patches. From Michael
Corrigan of UC San Diego.
Use dynamically allocated memory (instead of static buffers)
for macros defined in initsys() and settime(); since
these can have different values depending on which
envelope they are in. From Eric Wassenaar.
Improve logging to show ctladdr on to= logging; this tells you
what uid/gid processes ran as.
Fix a problem that caused error messages to be discarded if
the sender address was unparseable for some reason;
this was supposed to fall back to the "return to
Improve aliaswait backoff algorithm.
Portability patches for Linux (8.6.3 required another header
file) (from Karl London) and SCO UNIX.
CONFIG: patch prog mailer to not strip host name off of envelope
addresses (so that it matches local again). From
CONFIG: change uucp-dom mailer so that "<>" translates to $n;
this prevents uux from seeing lines with null names like
``From Sat Oct 30 14:55:31 1993''. From Motonori
Nakamura of Kyoto University.
CONFIG: handle <list:;> syntax correctly. This isn't legal, but
it shouldn't fail miserably. From Motonori Nakamura.
IMPORTANT FIX: Fix several problems that caused open files to
be "lost" during queue runs; this overflowed the open
file table on large runs. An assumption that fdopen
always succeeds sometimes resulted in core dumps when
this happens; sometimes the message is delivered twice,
sometimes (probably) infinite times. This problem in
various form was reported by P{r (Pell) Emanuelsson and
Robert Campbell of U.C. Berkeley.
Special diagnosis of EMFILE error conditions -- it now prints
the known open file descriptors so you can figure out
what is consuming so much resources.
Fix a couple of problems caused by early address parsing
errors -- one caused it to return a "this is only a
warning" when it really wasn't, and the other started
parsing through a random pointer. The first was
Fix an infinite loop problem caused by null components in the
host signature. Problem noted by Jan Sorensen.
Be sure to reset the "current date" when sending an error
message -- PostMasterCopy messages were being sent
with an old Date: header.
Fix a problem that caused duplicated mail when sendmail was
(1) compiled without HASFLOCK, (2) you are sending to
an alias that has an owner-* alias, (3) you execute
sendmail with -t flag, (4) you run in -odb mode, and
(5) the sender specifies both the alias name and
another alias [i.e., the envelope is split], then
duplicate messages are sent. The problem description
and one-line fix are from Motonori Nakamura of Kyoto
Avoid a problem that causes error messages to be discarded
in some cases -- this was the result of a "fix" to
avoid duplicate error messages, but two are better
than zero. Reported by Tim Rylance.
Fix a minor botch in checkfd012() -- fix from Dave Hill of
Remove "X-Authentication-Warning: <user> set sender to <address>
using -f" entirely -- it is far too eager to include
this, and it is confusing folks. I'll try to make it
work "right" in 8.7. Problem noted by Yoshitaka
Tokugawa of dit Co., Ltd.
Fix a race condition with the errno value in tick() and
reapchild() -- this caused occasional misdiagnosis
of problems. Kyle Jones of UUNET helped this along.
Repair rule loop-detection code. From Michael Corrigan of
Fix a problem that caused sender domain addition (C mailer
flag to be ignored if you use -odq or use -odb with
a high load average. Problem reported by Jim Murray
Fix ident protocol on multi-homed machines. It was not
always using the correct interface. Fix from J.R.
Previously, sendmail assumed that any SMTP greeting message
that wasn't 2xx was a temporary failure -- it should
only take 4xx as a temporary failure, and return a
solid error message on anything else -- for example,
to allow you to reject connections on a workstation
that is MXed to a mail server.
Portability enhancements for 386BSD/FreeBSD/NetBSD from
CONFIG: FEATURE(always_add_domain) didn't always add the domain;
in particular, on local mail it modified the header sender
but not the header recipient address(es). Reported by
Jeffrey Honig of Cornell University. Also, strip
any host from envelope recipient address(es), since
local mailers don't understand host names -- this is
to help mailertable entries. From Christopher Davis.
CONFIG: masquerading didn't apply to addresses that already
had a domain. This change replaces a local hostname
by the masquerade name in the SMTP mailer (previously
it only added the masquerade name if it didn't already
have a domain name). Several people complained about
Put a "successful delivery" message in the transcript for
addresses that get return-receipts.
Put a prominent "this is only a warning" message in warning
messages -- some people don't read carefully enough
and end up sending the message several times.
Include reason for temporary failure in the "warning" return
message. Currently, it just says "cannot send for
Fix the "Original message received" time generated for
returntosender messages. It was previously listed as
the current time. Bug reported by Eric Hagberg of
Cornell University Medical College.
If there is an error when writing the body of a message,
don't send the trailing dot and wait for a response
in sender SMTP, as this could cause the connection to
hang up under some bizarre circumstances. From Eric
Fix some server SMTP synchronization problems caused when
connections fail during message collection. From
Fix a problem that can cause srvrsmtp to reject mail if the
name server is down -- it accepts the RCPT but rejects
the DATA command. Problem reported by Jim Murray of
Fix a problem that can cause core dumps if the config file
incorrectly resolves to a null hostname. Reported by
Non-root use of -C flag, dangerous -f flags, and use of -oQ
by non-root users were not put into
X-Authentication-Warning:s as intended because the
config file hadn't set the PrivacyFlags yet. Fix
from Sven-Ove Westberg of the University of Lulea.
Under very odd circumstances, the alias file rebuild code
could get confused as to whether a database was
Check "vendor code" on the end of V lines -- this is
intended to provide a hook for vendor-specific
configuration syntax. (This is a "new feature",
but I've made an exception to my rule in a belief
that this is a highly exceptional case.)
Portability fixes for DG/UX (from Douglas Anderson of NCSC),
SCO Unix (from Murray Kucherawy), A/UX, and OSF/1
(from Jon Forrest of UC Berkeley)
CONFIG: fix ``mailer:host'' form of UUCP relay naming.
Portability fixes for A/UX and Encore UMAX V.
Fix error message handling -- if you had a name server down
causing an error during parsing, that message was never
propogated to the queue file.
Configuration cleanup: make it easier to undo IDENTPROTO in
conf.h (other systems have the same bug).
If HASGETDTABLESIZE and _SC_OPEN_MAX are both defined, assume
getdtablesize() instead of sysconf(); a disturbingly
large number of systems defined _SC_OPEN_MAX in the
header files but don't have the syscall.
Another patch to really truly ignore MX records in getcanonname
Fix problem that caused the "250 IAA25499 Message accepted for
delivery" message to be omitted if there was an error
in the header of the message (e.g., a bad Errors-To:
line). Pointed out by Michael Corrigan of UCSD.
Announce name of host we are chatting when we get errors; this
is an IDA-ism suggested by Christophe Wolfhugel.
Portability fixes for Alpha OSF/1 (from Anthony Baxter of the
Australian Artificial Intelligence Institute), SCO Unix
(from Murray Kucherawy of Hookup Communication Corp.),
NeXT (from Vince DeMarco and myself), Linux (from
Karl London <karl@borg.demon.co.uk>), BSDI (from
Christophe Wolfhugel, and SVR4 on Dell (from Kimmo
Suominen), AUX 3.0 on Macintosh, and ANSI C compilers.
Some changes to get around gcc optimizer bugs. From Takahiro
Fix error recovery in queueup if another tf file of the same
name already exists. Problem stumbled over by Bill
Output YP_MASTER_NAME and YP_LAST_MODIFIED without null bytes.
Problem noted by Keith McMillan of Ameritech Services.
Deal with group permissions properly when opening .forward and
:include: files. This relaxes the 8.1C restrictions
slightly more. This includes proper setting of groups
when reading :include: files, allowing you to read some
files that you should be able to read but have previously
been denied unless you owned them or they had "other"
Make certain that $j is in $=w (after the .cf is read) so that
if the user is forced to override some silly system,
MX suppression will still work.
Fix a couple of efficiency problems where newstr was double-
calling expensive routines. In at least one case, it
wasn't guaranteed that they would always return the
same result. Problem noted by Christophe Wolfhugel.
Fix null pointer dereference in putoutmsg -- only on an error
condition from a non-SMTP mailer. From Motonori
Macro expand "C" line class definitions before scanning so that
Fix problem that caused error message to be sent while still
trying to send the original message if the connection
is closed during a DATA command after getting an error
on an RCPT command (pretty obscure). Problem reported
Fix reply to NOOP to be 250 instead of 200 -- this is a long
Fix a nasty bug causing core dumps when returning the "warning:
cannot deliver for N hours -- will keep trying" message;
it only occurred if you had PostMasterCopy set and
only on some architectures. Although sendmail would
keep trying, it would send error messages on each
queue interval. This is an important fix.
Allow u and g options to take user and group names respectively.
Don't do a chdir into the queue directory in -bt mode to make
ruleset testing a bit easier.
Don't allow users to turn off logging (using -oL) on the command
line -- command line can only raise, not lower, logging
Set $u to the original recipient on the SMTP transaction or on
the command line. This is only done if there is exactly
one recipient. Technically, this does not meet the
specs, because it does not guarantee a domain on the
Fix a problem that dumped error messages on bad addresses if
you used the -t flag. Problem noted by Josh Smith of
Given an address such as ``<foo> <bar>'', auto-quote the first
``<foo>'' part, giving ``"<foo>" <bar>''. This is to
avoid the problem of people who use angle brackets in
their full name information.
Fix a null pointer dereference if you set option "l", have
an Errors-To: header in the message, and have Errors-To:
defined in the config file H lines. From J.R. Oldroyd.
Put YPCOMPAT on #ifdef NIS instead -- it's one less thing to get
wrong when compiling. Suggested by Rick McCarty of TI.
Fix a problem that could pass negative SIZE parameter if the
df file got lost; this would cause servers to always
give a temporary failure, making the problem even worse.
Problem noted by Allan Johannesen of WPI.
Add "ident" timeout (one of the "r" option selectors) for IDENT
protocol timeouts (30s default). Requested by Murray
Kucherawy of HookUp Communication Corp. to handle bogus
PC TCP/IP implementations.
Change $w default definition to be just the first component of
the domain name on config level 5. The $j macro defaults
to the FQDN; $m remains as before. This lets well-behaved
config files use any of the short, long, or subdomain
Add makesendmail script in src to try to automate multi-architecture
builds. I know, this is sub-optimal, but it is still
Fix very obscure race condition that can cause a queue run to
get a queue file for an already completed job. This
problem has existed for years. Problem noted by the
long suffering Allan Johannesen of WPI.
Fix a problem that caused the raw sender name to be passed to
udbsender instead of the canonified name -- this caused
it to sometimes miss records that it should have found.
Relax check of name on HELO packet so that a program using -bs
that claims to be itself works properly.
Restore rewriting of $: part of address through 2, R, 4 in
buildaddr -- this requires passing a lot of flags to get
it right. Unlike old versions, this ONLY rewrites
recipient addresses, not sender addresses.
Fix a bug that caused core dumps in config files that cannot
resolve /file/name style addresses. Fix from Jonathan
Kamens of OpenVision Technologies.
Fix problem with fcntl locking that can cause error returns to
be lost if the lock is lost; this required fully
queueing everything, dropping the envelope (so errors
would get returned), and then re-reading the queue from
Fix a problem that caused aliases that redefine an otherwise
true address to still send to the original address
if and only if the alias failed in certain bizarre
ways (e.g, if they pointed at a list:; syntax address).
Problem pointed out by Jonathan Kamens.
Remove support for frozen configuration files. They caused
more trouble than it was worth.
Fix problem that can cause error messages to get ignored when
using both -odb and -t flags. Problem noted by Rob
McNicholas at U.C. Berkeley.
Include all "normal" variations on hostname in $=w. For example,
if the host name is vangogh.cs.berkeley.edu, $=w will
contain vangogh, vangogh.cs, and vangogh.cs.berkeley.edu.
Add "restrictqrun" privacy flag -- without this, anyone can run
Reset SmtpPhase global on initial connection creation so that
messages don't come out with stale information.
Pass an "ext" argument to lockfile so that error/log messages
will properly reflect the true filename being locked.
Put all [...] address forms into $=w -- this eliminates the need
for MAXIPADDR in conf.h. Suggested by John Gardiner
Fix a bug that can cause qf files to be left around even after
an SMTP RSET command. Problem and fix from Michael
Don't send a PostMasterCopy to errors when the Precedence: is
negative. Error reports still go to the envelope
Add LA_SHORT for load averages.
Lock sendmail.st file when posting statistics.
Add "SendBufSize" and "RcvBufSize" suboptions to "O" option to
set the size of the TCP send and receive buffers; if you
run over a slow slip line you may need to set these down
(although it would be better to fix the SLIP implementation
so that it's not necessary to recompile every program
that does bulk data transfer).
Allow null defaults on $( ... $) lookups. Problem reported by
Diagnose crufty S and V config lines. This resulted from an
observation that some people were using the SITE macro
without the SITECONFIG macro first, which was causing
bogus config files that were not caught.
Fix makemap -f flag to turn off case folding (it was turning it
on instead). THIS IS A USER VISIBLE CHANGE!!!
Fix a problem that caused multiple error messages to be sent if
you used "sendmail -t -oem -odb", your system uses fcntl
locking, and one of the recipient addresses is unknown.
Reset uid earlier in include() so that recursive .forwards or
:include:s don't use the wrong uid.
If file descriptor 0, 1, or 2 was closed when sendmail was
called, the code to recover the descriptor was broken.
This sometimes (only sometimes) caused problems with the
alias file. Fix from Motonori Nakamura.
Fix a problem that caused aliaswait to go into infinite recursion
if the @:@ metasymbol wasn't found in the alias file.
Improve error message on newaliases if database files cannot be
opened or if running with no database format defined.
Do a better estimation of the size of error messages when NoReturn
is set. Problem noted by P{r (Pell) Emanuelsson.
Fix a problem causing the "c" option (don't connect to expensive
mailers) to be ignored in SMTP. Problem noted and the
solution suggested by Robert Elz of Munnari University.
Improve connection caching algorithm by passing "[host]" to
hostsignature, which strips the square brackets and
returns the real name. This allows mailertable entries
to match regular entries.
Re-enable Return-Receipt-To: -- people seem to want this stupid
feature, even if it doesn't work right.
Catch and log attempts to try the "wiz" command in server SMTP.
This also ups the log level from LOG_NOTICE to LOG_CRIT.
Be more generous at assigning $z to the home directory -- do this
for programs that are specified through a .forward file.
Fix from Andrew Chang of Sun Microsystems.
Always save a fatal error message in preference to a non-fatal
error message so that the "subject" line of return
messages is the best possible.
CONFIG: reduce the number of quotes needed to quote configuration
parameters with commas: two quotes should work now, e.g.,
define(ALIAS_FILE, ``/etc/aliases,/etc/aliases.local'').
CONFIG: class $=Z is a set of UUCP hosts that use uucp-dom
connections (domain-ized UUCP).
CONFIG: fix bug in default maps (-o must be before database file
name). Pointed out by Christophe Wolfhugel.
CONFIG: add FEATURE(nodns) to state that we are not relying on
DNS. This would presumably be used in UUCP islands.
CONFIG: add OSTYPE(nextstep) and OSTYPE(linux).
CONFIG: log $u in Received: line. This is in technical violation
of the standards, since it doesn't guarantee a domain
CONFIG: don't assume "m" in local mailer flags -- this means that
if you redefine LOCAL_MAILER_FLAGS you will have to include
the "m" flag should you want it. Apparently some Solaris 2.2
installations can't handle multiple local recipients.
Problem noted by Josh Smith.
CONFIG: add confDOMAIN_NAME to set $j (if undefined, $j defaults).
CONFIG: change default version level from 4 to 5.
CONFIG: add FEATURE(nullclient) to create a config file that
forwards all mail to a hub without ever looking at the
CONFIG: properly strip mailer: information off of relays when
used to change .BITNET form into %-hack form.
CONFIG: fix a problem that caused infinite loops if presented
with an address such as "!foo".
CONFIG: check for self literal (e.g., [128.32.131.12]) even if
the reverse "PTR" mapping is broken. There's a better
way to do this, but the change is fairly major and I
want to hold it for another release. Problem noted by
Serious bug: if you used a command line recipient that was unknown
sendmail would not send a return message (it was treating
everything as though it had an SMTP-style client that
would do the return itself). Problem noted by Josh Smith.
Change "trymx" option in getcanonname() to ignore all MX data,
even during a T_ANY query. This actually didn't break
anything, because the only time you called getcanonname
with !trymx was if you already knew there were no MX
records, but it is somewhat cleaner. From Motonori
Don't call getcanonname from getmxrr if you already know there
are no DNS records matching the name.
Fix a problem causing error messages to always include "The
original message was received ... from localhost".
The correct original host information is now included.
Previous change to cf/sh/makeinfo.sh doesn't port to Ultrix (their
version of "test" doesn't have the -x flag). Change it
to use -f instead. From John Myers.
CONFIG: 8.4 mistakenly set the default SMTP-style mailer to
esmtp -- it should be smtp.
CONFIG: send all relayed mail using confRELAY_MAILER (defaults
to "relay" (a variant of "smtp") if MAILER(smtp) is used,
else "suucp" if MAILER(uucp) is used, else "unknown");
this cleans up the configs somewhat. This fixes a serious
problem that caused route-addrs to get mistaken as relays,
pointed out by John Myers. WARNING: this also causes
the default on SMART_HOST to change from "suucp" to
"relay" if you have MAILER(smtp) specified.
Add option `w'. If you receive a message that comes to you because
you are the best (lowest preference) target of an MX, and
you haven't explicitly recognized the source MX host in
your .cf file, this option will cause you to try the target
host directly (as if there were no MX for it at all). If
`w' is not set, this case is a configuration error.
Beware: if `w' is set, senders may get bogus errors like
"message timed out" or "host unknown" for problems that
are really configuration errors. This option is
disrecommended, provided only for compatibility with
Fix a problem that caused the incoming socket to be left open
when sendmail forks after the DATA command. This caused
calling systems to wait in FIN_WAIT_2 state until the
entire list was processed and the child closed -- a
potentially prodigious amount of time. Problem noted
Fix problem (created in 6.64) that caused mail sent to multiple
addresses, one of which was a bad address, to completely
suppress the sending of the message. This changes
handling of EF_FATALERRS somewhat, and adds an
EF_GLOBALERRS flag. This also fixes a potential problem
with duplicate error messages if there is a syntax error
in the header of a message that isn't noticed until late
in processing. Original problem pointed out by Josh Smith
of Harvey Mudd College. This release includes quite a bit
of dickering with error handling (see below).
Back out SMTP transaction if MAIL gets nested 501 error. This
will only hurt already-broken software and should help
Fix a problem that broke aliases when neither NDBM nor NEWDB were
compiled in. It would never read the alias file.
Repair unbalanced `)' and `>' (the "open" versions are already
Logging of "done" in dropenvelope() was incorrect: it would
log this even when the queue file still existed. Change
this to only log "done" (at log level 11) when the
queue file is actually removed. From John Myers.
Log "lost connection" in server SMTP at log level 20 if there
is no pending transaction. Some senders just close the
connection rather than sending QUIT.
Fix a bug causing getmxrr to add a dot to the end of unqualified
domains that do not have MX records -- this would cause
the subsequent host name lookup to fail. The problem
only occurred if you had FEATURE(nocanonify) set.
Problem noted by Rick McCarty of Texas Instruments.
Fix invocation of setvbuf when passed a -X flag -- I had
unwittingly used an ANSI C extension, and this caused
core dumps on some machines.
Diagnose self-destructive alias loops on RCPT as well as EXPN.
Previously it just gave an empty send queue, which
then gave either "Need RCPT (recipient)" at the DATA
(confusing, since you had given an RCPT command which
returned 250) or just dropped the email, depending on
whether you were running VERBose mode. Now it usually
diagnoses this case as "aliasing/forwarding loop broken".
Unfortunately, it still doesn't adequately diagnose
some true error conditions.
Add internal concept of "warning messages" using 6xx codes.
These are not reported only to Postmaster. Unbalanced
parens, brackets, and quotes are printed as 653 codes.
They are always mapped to 5xx codes before use in SMTP.
Clean up error messages to tell both the actual address that
failed and the alias they arose from. This makes it
somewhat easier to diagnose problems. Difficulty noted
Fix a problem that inappropriately added a ctladdr to addresses
that shouldn't have had one during a queue run. This
caused error messages to be handled differently during
a queue run than a direct run.
Don't print the qf name and line number if you get errors during
the direct run of the queue from srvrsmtp -- this was
just extra stuff for users to crawl through.
Put command line flags on second line of pid file so you can
auto-restart the daemon with all appropriate arguments.
Use "kill `head -1 /etc/sendmail.pid`" to stop the
daemon, and "eval `tail -1 /etc/sendmail.pid`" to
Remove the ``setuid(getuid())'' in main -- this caused the
IDENT daemon to screw up. This required that I change
HASSETEUID to HASSETREUID and complicate the mode
changing somewhat because both Ultrix and SunOS seem
to have a bug causing seteuid() to set the saved uid
as well as the effective. The program test/t_setreuid.c
will test to see if your implementation of setreuid(2)
is appropriately functional.
The FallBackMX (option V) handling failed to properly identify
fallback to yourself -- most of the code was there,
but it wasn't being enabled. Problem noted by Murray
Kucherawy of the University of Waterloo.
Change :include: open timeout from ETIMEDOUT to an internal
code EOPENTIMEOUT; this avoids adding "during SmtpPhase
with CurHostName" in error messages, which can be
confusing. Reported by Jonathan Kamens of OpenVision
Back out setpgrp (setpgid on POSIX systems) call to reset the
process group id. The original fix was to get around
some problems with recalcitrant MUAs, but it breaks
any call from a shell that creates a process group id
different from the process id. I could try to fix
this by diddling the tty owner (using tcsetpgrp or
equivalent) but this is too likely to break other
Support -M as equivalent to -oM on Ultrix -- apparently
DECnet calls sendmail with -MrDECnet -Ms<HOST> -bs
instead of using standard flags. Oh joy. This
behaviour reported by Jon Giltner of University
SGI IRIX -- this includes several changes that should
help other strict ANSI compilers.
SCO Unix -- from Murray Kucherawy of HookUp Communication
Solaris running the Sun C compiler (which despite the
documentation apparently doesn't define
ConvexOS from Eric Schnoebelen of Convex.
Sony NEWS workstations and Omron LUNA workstations from
CONFIG: add confTRY_NULL_MX_LIST to set option `w'.
CONFIG: delete `C' and `e' from default SMTP mailers flags;
several people have made a good argument that this
creates more problems than it solves (although this
may prove painful in the short run).
CONFIG: generalize all the relays to accept a "mailer:host"
CONFIG: move local processing in ruleset 0 into a new ruleset
98 (8 on old sendmail). Domain literal [a.b.c.d]
addresses are also passed through this ruleset.
CONFIG: if neither SMART_HOST nor MAILER(smtp) were defined,
internet-style addresses would "fall off the end" of
ruleset zero and be interpreted as local -- however,
the angle brackets confused the recursive call.
These are now diagnosed as "Unrecognized host name".
CONFIG: USENET rules weren't included in S0 because of a mistaken
ifdef(`_MAILER_USENET_') instead of
ifdef(`_MAILER_usenet_'). Problem found by Rein Tollevik
CONFIG: move up LOCAL_RULE_0 processing so that it happens very
early in ruleset 0; this allows .mc authors to bypass
things like the "short circuit" code for local addresses.
Prompted by a comment by Bill Wisner of The Well.
CONFIG: add confSMTP_MAILER to define the mailer used (smtp or
esmtp) to send SMTP mail. This allows you to default
to esmtp but use a mailertable or other override to
deal with broken servers. This logic was pointed out
to me by Bill Wisner. Ditto for confLOCAL_MAILER.
Changes to cf/sh/makeinfo.sh to make it portable to SVR4
environments. Ugly as sin.
Fix setuid problems introduced in 8.2 that caused messages
like "Cannot create qfXXXXXX: Invalid argument"
or "Cannot reopen dfXXXXXX: Permission denied". This
involved a new compile flag "HASSETEUID" that takes
the place of the old _POSIX_SAVED_IDS -- it turns out
that the POSIX interface is broken enough to break
some systems badly. This includes some fixes for
HP-UX. Also fixes problems where the real uid is
not reset properly on startup (from Neil Rickert).
Fix a problem that caused timed out messages to not report the
addresses that timed out. Error messages are also more
Drop required bandwidth on connections from 64 bytes/sec to
Further Solaris portability changes -- doesn't require the BSD
compatibility library. This also adds a new
"HASGETDTABLESIZE" compile flag which can be used if
you want to use getdtablesize(2) instead of sysconf(2).
These are loosely based on changes from David Meyer at
University of Oregon. This now seems to work, at least
Fix a problem that can cause duplicate error messages to be
sent if you are in SMTP, you send to multiple addresses,
and at least one of those addresses is good and points
to an account that has a .forward file (whew!).
Fix a problem causing messages to be discarded if checkcompat()
returned EX_TEMPFAIL (because it didn't properly mark
the "to" address). Problem noted by John Myers.
Fix dfopen to return NULL if the open failed; I was depending
on fdopen(-1) returning NULL, which isn't the case. This
isn't serious, but does result in wierd error diagnoses.
CONFIG: add UUCP_MAX_SIZE M4 macro to set the maximum size of
messages sent through UUCP-family mailers. Suggested
by Bill Wisner of The Well.
CONFIG: if both MAILER(uucp) and MAILER(smtp) are specified,
include a "uucp-dom" mailer that uses domain-style
addressing. Suggested by Bill Wisner.
CONFIG: Add LOCAL_SHELL_FLAGS and LOCAL_SHELL_ARGS to match
LOCAL_MAILER_FLAGS and LOCAL_MAILER_ARGS. Suggested by
CONFIG: Add OSTYPE(aix3). From Christophe Wolfhugel.
Don't drop out on config file parse errors in -bt mode.
On older configuration files, assume option "l" (use Errors-To
header) for back compatibility. NOTE: this DOES NOT
imply an endorsement of the Errors-To: header in any way.
Accept -x flag on AIX-3 as well as OSF/1. Why, why, why???
Don't log errors on EHLO -- it isn't a "real" error for an old
SMTP server to give an error on this command, and
logging it in the transcript can be confusing. Fix
IRIX compatibility changes provided by Dan Rich
<drich@sandman.lerc.nasa.gov>.
Solaris 2 compatibility changes. Provided by Bob Cunningham
<bob@kahala.soest.hawaii.edu>, John Oleynick
<juo@klinzhai.rutgers.edu>
Debugging: -d17 was overloaded (hostsignature and usersmtp.c);
move usersmtp (smtpinit and smtpmailfrom) to -d18 to
match the other flags in that file.
Flush transcript before fork in mailfile(). From Eric Wassenaar.
Save h_errno in mci struct and improve error message display.
Changes from Eric Wassenaar.
Open /dev/null for the transcript if the create of the xf file
failed; this avoids at least one possible null pointer
reference in very wierd cases. From Eric Wassenaar.
Clean up statistics gathering; it was over-reporting because of
forks. From Eric Wassenaar.
Fix problem that causes old Return-Path: line to override new
Return-Path: line (conf.c needs H_FORCE to avoid
re-using old value). From Motonori Nakamura.
Fix broken -m flag in K definition -- even if -m (match only)
was specified, it would still replace the key with the
value. Noted by Rick McCarty of Texas Instruments.
If the name server timed out over several days, no "timed out"
message would ever be sent back. The timeout code
has been moved from markfailure() to dropenvelope()
so that all such failures should be diagnosted. Pointed
out by Christophe Wolfhugel and others.
Relax safefile() constraints: directories in an include or
forward path must be readable by self if the controlling
user owns the entry, readable by all otherwise (e.g.,
when reading your .forward file, you have to own and
have X permssion in it; everyone needs X permission in
the root and directories leading up to your home);
include files must be readable by anyone, but need not
If _POSIX_SAVED_IDS is defined, setuid to the owner before
reading a .forward file; this gets around some problems
on NFS mounts if root permission is not exported and
the user's home directory isn't x'able.
Additional NeXT portability enhancements from Axel Zinser.
Additional HP-UX portability enhancements from Brian Bullen.
Add a timeout around SMTP message writes; this assumes you can
get throughput of at least 64 bytes/second. Note that
this does not impact the "datafinal" default, which
is separate; this is just intended to work around
network clogs that will occur before the final dot
is sent. From Eric Wassenaar.
Change map code to set the "include null" flag adaptively --
it initially tries both, but if it finds anything
matching without a null it never tries again with a
null and vice versa. If -N is specified, it never
tries without the null and creates new maps with a
null byte. If -O is specified, it never tries with
the null (for efficiency). If -N and -O are specified,
you get -NO (get it?) lookup at all, so this would
be a bad idea. If you don't specify either -N or -O,
Fix recognition of "same from address" so that MH submissions
will insert the appropriate full name information;
this used to work and got broken somewhere along the
Some changes to eliminate some unnecessary SYSERRs in the
log. For example, if you lost a connection, don't
bother reporting that fact on the connection you lost.
Add some "extended debugging" flags to try to track down
why we get occassional problems with file descriptor
one being closed when execing a mailer; it seems to
only happen when there has been another error in the
same transaction. This requires XDEBUG, defined
Add "-X filename" command line flag, which logs both sides of
all SMTP transactions. This is intended ONLY for
debugging bad implementations of other mailers; start
it up, send a message from a mailer that is failing,
and then kill it off and examine the indicated log.
This output is not intended to be particularly human
readable. This also adds the HASSETVBUF compile
flag, defaulted on if your compiler defines __STDC__.
CONFIG: change SMART_HOST to override an SMTP mailer. If you
have a local net that should get direct connects, you
will need to use LOCAL_NET_CONFIG to catch these hosts.
See cf/README for an example.
CONFIG: add LOCAL_MAILER_ARGS (default: `mail -d $u') to handle
sites that don't use the -d flag.
CONFIG: hide recipient addresses as well as sender addresses
behind $M if FEATURE(allmasquerade) is specified; this
has been requested by several people, but can break
local aliases. For example, if you mail to "localalias"
this will be rewritten as "localalias@masqueradehost";
although initial delivery will work, replies will be
broken. Use it sparingly.
CONFIG: add FEATURE(domaintable). This maps unqualified domains
to qualified domains in headers. I believe this is
largely equivalent to the IDA feature of the same name.
CONFIG: use $U as UUCP name instead of $k. This permits you
to override the "system name" as your UUCP name --
in particular, to use domain-ized UUCP names. From
CONFIG: create new mailer "esmtp" that always tries EHLO
first. This is currently unused in the config files,
but could be used in a mailertable entry.
Serious security bug fix: it was possible to read any file on
the system, regardless of ownership and permissions.
If a subroutine returns a fully qualified address, return it
immediately instead of feeding it back into rewriting.
This fixes a problem with mailertable lookups.
CONFIG: fix some M4 frotz (concat => CONCAT)
Serious bug fix: pattern matching backup algorithm stepped by
two tokens in classes instead of one. Found by Claus
Assmann at University of Kiel, Germany.
Another mailertable fix....
4.4BSD freeze. No semantic changes.
Fix some lintish problems.
Fix some cases where server SMTP behaved poorly when handed bogus
input, pointed out by Eric Wassenaar.
CONFIG: fix some more (sigh) mailertable bugs -- thanks to
Motonori Nakamura of Kyoto University (again).
Don't send 050 (-v) information after the 250 response to a QUIT
command in srvrsmtp -- clients usually close the connection
at this point, and it causes bogus error messages.
Don't send messages that have errors on input (such as unbalanced
parentheses) during SMTP transactions, since a return
message has (probably) already been sent.
Give better diagnostics on timeouts during network reads, including
information similar to the SMTP phase.
Fix bug that caused SMTP messages to deliver synchronously; this
happened after the DATA 250, and hence caused reading the
next command to be delayed.
Ignore Errors-To: header unless 'l' (lower case el) header is
specified. The Errors-To: header violates RFC 1123.
Errors-To: was only needed to take the place of the
envelope sender in the days when most Unix mailers
didn't understand about the two kinds of senders.
Don't send warning messages in response to automatically generated
messages (that is, those From:<>).
CONFIG: fix some rather stupid typos in the mailertable code
pointed out by Motonori Nakamura of Kyoto University.
CONFIG: add confUSE_ERRORS_TO configuration option.
CONFIG: if ALWAYS_ADD_DOMAIN is selected, try to use $M
(masquerade name) instead of $j.
CONFIG: don't add dots to relay names (added in 6.29); it breaks
several things, and can be simulated by dot terminating
the names of relays. For example, use:
Fix prototypes to eliminate chars in argument lists -- some
compilers are pissy about this.
Log protocol ($r) and body type if set so we can determine if
the adaptive algorithms are working.
Pessimize on locking of database files (particularly for NEWDB
databases) during opens. There were problems with
processes opening the file while it was rebuilt; since
NEWDB caches heavily, the reader opened an empty file,
which is an error. If your system has the ability to
lock atomically on open, this works properly; otherwise,
there are race conditions.
Check mod time on .pag file instead of .dir in NDBM aliases
because the .dir file doesn't get updated for small
alias files. From John Gardiner Myers of CMU.
More Solaris portability -- it now compiles on Solaris, but
hangs up in gethostbyname().
Move setting of RES_DEBUG flag before first myhostname() call
so we can see name server traffic on that call.
Fix a problem that causes -bi to try to rebuild maps other than
Fix a problem that caused udb to reject entries from any but
the first database listed.
Rearrange doc subdirectory for 4.4BSD release tape.
CONFIG: put $r into the Received line. This was an oversight.
CONFIG: fix typo (call to ruleset 99 should have been rulset 90).
CONFIG: move "auxiliary" subroutines to be in ruleset 90-99
range -- in the long run, single digit rulesets may
become reserved for builtin use by sendmail.
CONFIG: fix major problem that causes host aliases (that is,
anything in $=w != $j) to not be recognized. This has
Fix recursive syserr (if there is an error printing a syserr
message). This makes the code much less eager to consider
a write error as serious. This also includes some
heuristics to be clever about closed connections.
Lock NEWDB files during gets. This requires version 1.5 or later
of the db library. If you have an older version, you
can use -DOLD_NEWDB. This will go away in a few weeks.
Fix problem causing aliases that use host maps to get overwritten.
Do appropriate byte swapping on port numbers in ident protocol
code. Fix from Allan Johannesen of WPI.
Defer opening of map files to the same time as alias files so that
the daemon will tend to pick up new versions more promptly.
Prototype a bunch more functions.
Some Solaris 2.1 changes (still doesn't link though).
Try to simplify Makefiles by including more subordinate #defines
in conf.h (based on OS type).
CONFIG: check for domains if FEATURE(mailertable) is defined.
For example, if the host name is "knecht.cs.berkeley.edu"
it will search the following mailertable keys:
This could be used to replace the special relays for bitnet
Fix problem that prevented appending dots on canonified host
names. This breaks tons of config files -- very
Fix improper pointer dereference in response to HELO command.
Fix core dump if debugging set in map_rewrite.
CONFIG: add FEATURE(always_add_domain) to always attach the
local domain (only impacts local mail).
CONFIG: try to avoid turning names into $j -- although
technically a host can only have one "canonical name",
it seems to be common practice to have several.
Major change: merge alias databases with maps. This expands and
changes the map class interface but fixes a bunch of bugs.
The important user-visible change is that the file name
in a K line now does not include the ".db" extension; this
is added automatically. Also, the -d (NIS domain) flag is
missing from the K config line; use @domain instead.
When compiling, the *_MAP names are gone -- just compile
in NDBM, NEWDB, and/or NIS support.
Announce mailer/host/user triple on -bv flag -- from Brian
Bullen of Stirling University.
Don't send more than one line in response to HELO -- it confuses
Pony Express, which then behaves very badly. However,
this change does send two line 220 greetings, with the
second line reading "ESMTP spoken here". The usersmtp
module recognizes this and goes into ESMTP mode regardless
of the setting of the "a" mailer flag. Thus, "a" means
AIX portability changes (thanks to Christophe Wolfhugel of
Herve Schauer Consultants (Paris) for providing me with
an INSA account for this purpose). Lightly tested. Use
-D_AIX3. This probably breaks compatibility with some
older systems (e.g., 4.2bsd) but still works on SunOS
4.1.2, Ultrix 4.2A, HP-UX 8.07, OSF/1 T1.3, and AIX 3.2.3.
Fix a problem causing an error message loop if the output channel
Add the Makefiles that I use for various environments -- some are
Berkeley make versions and some are old make versions.
My makefile for the NeXT box has gotten lost, alas!
PRALIASES: support for printing NEWDB databases. From
Michael J. Corrigan of U.C. San Diego.
CONFIG: don't pass pseudo-domains to $[ ... $] (if you have
a wildcard MX it can have wierd results). From
CONFIG: dot terminate relay hostnames in S0. From Christophe
Log version with SMTP daemon startup message.
Adjust setproctitle to work on NetBSD and BSD/386.
Fix null pointer reference in MX fallback code.
A bunch of minor fixes from Eric Wassenaar:
If deliver cannot execv the mailer, return EX_OSERR
instead of EX_TEMPFAIL (to give better
Consistently malloc e_message.
Catch degenerate case of calling returntosender()
Fix bug that can cause incorrect verbose display of user smtp
Disable SMTP VERB command if PRIV_NOEXPN is set (since this
could reveal the same information.
Allow failure when reading SMTP greeting message to go on to
Add "MIME-Version: 1.0" header if using MIME (this was NOT
included in RFC 1344, but Bill King of Allan-Bradley
Company forwarded me email from Nathaniel Borenstein
claiming that it was an inadvertent omission).
Don't use Content-Type: X-message-header. According to John
Myers of CMU, many MIME readers will completely ignore
the data if they don't recognize it. Instead, just
add a blank line to make it a legal (empty) message.
Fix problem causing dots to keep getting appended to cached
hostnames. This can cause buffer overrun conditions.
The problem was found by Erik Forsberg of Retix,
although I used a different bug fix than he provided.
Fix parsing of split header/envelope rewriting specs -- from
Fix from Eric Wassenaar to correct To: lists in error messages.
Fix minor glitch causing extra ctladdrs to be output to queue
Cache results of name server canonification lookups to avoid
Major rewrite of alias.c: considerable cleanup, plus sample
(untested) support for NIS aliases. The "A" option
can now be a comma separated list (or be repeated) --
that is, you can have multiple alias databases. Each
database can have the syntax ``class:file''; if no class
is specified, the "implicit" class is assumed. Implicit
searches through a list of compiled in types -- hash,
dbm, nis, and stab. Alias files are searched in the
order they are listed. For example:
OAhash:/etc/aliases.local,/etc/aliases
OAnis:mail.aliases@my.nis.domain
first searches the hash database /etc/aliases.local,
then the regular /etc/aliases database, then the NIS
map "mail.aliases" in the NIS domain "my.nis.domain".
If in Verbose mode (probably from VERB command) run SMTP job
in foreground and don't do RCPT optimizations.
Add udb :mailsender as equivalent to owner- for regular aliases.
Delete option 8; add option 7 that means the opposite. That is,
default to 8-bit mode; a special option is needed to
force sendmail into 7 bit mode.
Send error messages in encapsulated MIME format.
New compile flag "NIS" that turns on NIS alias and NIS map
Add "j" option to send error messages in MIME (RFC 1341)
encapsulated message format per RFC 1344. The
syntax is pretty ugly if you don't have MIME-aware
Clean up message handling (for display in mailq output).
New setproctitle implementation for 4.4bsd.
Create files (such as ~/dead.letter) using mode FileMode (the
F option value) instead of 0666.
Fix bug causing output of EXPN command to not be fully qualified.
This may cause some problems with UUCP addresses that
will require some config file assistance -- specifically,
the $: part has to include the host name for this output
Fix a problem that sometimes diagnosed errors and still sent the
message if the header syntax was bad.
Fix a bug that caused an error message to be emailed when sendmail
was operating in -bv mode.
Add "ListenQueueSize" keyword to daemon options option (OO) to
set the queue size parameter passed to listen(). You
will normally have to tweak your kernel to up this.
Strip spaces off of beginning of message-id before logging (in
case it was folded across lines).
Tweak compile flags in daemon.c -- there were some cases where
it wouldn't work without NETINET.
Change *file* mailer to output all the usual default headers
(From, Date, Message-Id). It gets used when sending
CONFIG: explicitly catch and diagnose list:; syntax in ruleset
zero -- this is not a valid recipient syntax according
CONFIG: add confMIME_FORMAT_ERRORS to send error messages in
MIME format. Defaults to on.
CONFIG: add SMTP_MAILER_FLAGS and UUCP_MAILER_FLAGS to augment
the flags for those mailers.
Fix problem that causes the fallback mail to postmaster
(case ESM_POSTMASTER in savemail()) to not look at
Some more HPUX tweaking (compile flag hpux => __hpux so it
still works in ANSI mode).
Don't try to flock non-regular files when mailing to a file.
In particular, this was a problem if you tried to
Fix a wierd bug that can cause senders to be queued as
recipients if the name server is down when the mail
is initially sent. This hack just ignores sender
deletion (essentially, it sets the MeToo flag) if there
is a TEMPFAIL during processing of the sender address.
Fix a dangling else problem -- from Brian Bullen from University
Add the "b" mailer flag to force a blank line on the end of
messages. Some brilliant versions of /bin/mail insist
on this but do not add it themselves.
Add the "g" mailer flag to prevent user SMTP from sending
"MAIL From:<>". This is only intended to be a
transitional gesture, and should not be used if at
all possible. It appears that Berkeley and IDA
config files have always handled this properly; the
UK config kit apparently does not.
Don't lowercase and then capitalize header field names -- leave
them with original capitalization. Fixes from Bill
King of Allen-Bradley Company.
Further cleanup and improved reporting of error messages,
particularly conditions that cause messages to be
requeued for future delivery.
Tweak syslog priorities in some cases.
CONFIG: clean up route-addr on UUCP addresses.
HPUX 8.07 compatibility changes in getla() -- I had to make
these changes to get it to work at Berkeley, although
others seem to have been working before (???).
Various patches to XLA code.
Fix problem that causes setuid bit on files to be ignored from
SMTP or in queue runs. Problem noted by Jason Ornstein
Fix problem that can cause CNAMEs to be ignored.
Generalize getmxrr to match local host in $=w instead of a
Some cleanup from Eric Wassenaar:
Use FileMailer instead of ProgMailer in two places.
Eliminate duplicate 8th-bit stripping in commaize.
Fix a problem with mis-parsing of backslash escapes
under some circumstances.
NIS map fix (was always including trailing null character)
from Mike Glendinning of Ingres UK.
Add "a" mailer flag to try using ESMTP. It tries the EHLO
command and if that fails falls back to regular SMTP.
Also parses EHLO option keywords. If host supports
SIZE extension, this is added to the MAIL FROM:
Extend "b" option to include a second value which is the
maximum message size this server is willing to accept.
For example, a value of "10/1000000" says that there
must be ten blocks free, and sendmail will reject
any message larger than one megabyte.
Some portability hooks for NeXT (this could be applicable
to Mach in general). You have to create an empty
file called "unistd.h" to get it to compile.
Adjust config values (MAXLINE, MAXATOM, and PSBUFSIZE) to
Add X400-Received: to the list of headers tagged with H_TRACE
in conf.c. From Bill King, Allen-Bradley Co.
Fix problem that caused redefinition of SMTP and QUEUE compile
flags. Pointed out by Jon Forrest of the Sequoia 2000
Properly handle \! hack -- it was treating host\!user as one
token (host!user) instead of three (host, !, user).
Fix from Eric Wassenaar of NIKHEF-H.
Fix compilation problem in getauthinfo() if IDENTPROTO is off.
Turn off DEFNAMES and DNSRCH when getting the hostsignature
(i.e., MX records) in level 1 configuration files; this
matches the old behaviour. From Motonori Nakamura of
Improve error message printing -- if sent through an alias,
error messages include the name of the alias in the
message. Unfortunately, in order to make this work
properly in queue runs, this changes the format of the
C line in the qf file. The relatively uselessness of
the previous information was pointed out to me by
Allan E Johannesen of WPI.
Add XLA compile flag to add hooks to Christophe Wolfhugel's
extended load average code. This is still in very early
form. For information regarding the guts of the xla
code, contact Christophe.Wolfhugel@grasp.insa-lyon.fr.
Additional hooks for detecting tempfails in rewriting rules
(that is, in map lookups).
Properly diagnose ruleset zero returning null (instead of a mailer
triple). From Motonori Nakamura of Kyoto University.
More generalization of socket code for other protocols.
Shorten timeouts on reverse name lookups -- since they are done
during connection establishment, long timeouts here can
cause higher level timeouts. This mainly serves to accept
mail from hosts that do not have proper reverse (PTR) DNS
Reset e_statmsg before each mailer invocation to avoid bogus
Redefine $r, $s, and $_ in error envelopes so you don't get
incorrect cruft in the error message. Problem noted by
Motonori Nakamura of Kyoto University.
Fix a problem that can cause failure to return errors to Postmaster
in certain cases. From Motonori Nakamura.
Fix a problem that can cause some systems to give duplicate error
messages when a bad syntax address such as "<a" is presented
to an SMTP server. It doesn't seem to occur on all
machines. From Motonori Nakamura.
Default IDENTPROTO off for Ultrix and HPUX, which apparently have
the interesting "feature" that when they receive a "Host
unreachable" message they closes all open connections to
that host. However, some firewall gateways send this message
if you try to connect to an unauthorized port, such as the
IDENT port (113). Thus, no email can be received from such
hosts. There is some evidence that versions of Ultrix before
4.3 do not have this problem. Thanks to Tom Ivar Helbekkmo
for pointing out this behaviour to me and to Michael Corrigan
of U.C. San Diego for informing me about the HPUX problem.
Allow IPC mailers to return a colon-separated list of hosts in the
$@ clause; these are searched in order as though they were
When sending an error report, print the list of addresses tagged
as bad. Requested by Allan E Johannesen of WPI.
Change map function calls to return a status code. This gets
passed back as the result of rewrite. Parseaddr marks
the address as a QUEUEUP address if the return code is
EX_TEMPFAIL. All this to queue properly if the name
server is down. This code is not well tested. This code
changes the interface to map lookup functions (a fifth
parameter, int *statp, is added). Feature requested by
Don't delete quotes (in the dequote map) if there are spaces in
the string, since this would cause them to be replaced by
Accept BODY=8BITMIME on SMTP MAIL command. This isn't advertised
because the 8BIT to 7BIT translation doesn't exist yet.
This does add a "bodytype" field to both envelope and
queue file and a -B command line flag to pass the type in
during direct invocations.
Discard return error messages only on responses to responses to
responses, not on responses to responses. That is, the
algorithm is to try return to sender, then return to
postmaster, then discard. Previously it discarded
immediately if the return to sender pass failed.
CONFIG: back out change to hide unqualified hostnames behind %-hack.
This screws up local aliases and .forward files.
CONFIG: add FEATURE(nocanonify) to turn off calls to $[ ... $];
some sites only handle completely canonified names.
Requested by John Gardiner Myers of CMU.
CONFIG: some UUCP code was still included even if FEATURE(nouucp)
Clean up some minor glitches on error return messages pointed out
by Motonori Nakamura of Kyoto University.
Fix reply() to not reset SmtpReplyBuffer on fatal errors; this
was supposed to reset SmtpMsg Buffer. This makes the
client side code virtually useless. Reported by Allan
E Johannesen of WPI and Phil Brandenberger of Swarthmore.
Better debug messages if fuzzy is disabled, suggested by Allan
Offset SmtpReplyBuffer by four in usersmtp when checking for
loopback. From Eric Wassenaar.
Don't set $s until after runinchild in srvrsmtp -- otherwise
it gets cleared. From Eric Wassenaar.
Implement IDA-style $&x for deferred macro expansion.
More POSIX compatibility.
CONFIG: Hide unqualified hostnames behind %-hack using $s as the
actual sender. This is only done if $r is non-null, that
is, if this is not locally submitted mail.
CONFIG: Add FEATURE(bitdomain) allowing mapping of BITNET host
names to internet domains. A program contributed by
John Gardiner Myers of CMU to create the maps is included
in the contrib directory (in the "misc" tar file).
CONFIG: Add FEATURE(uucpdomain) for a similar mapping for UUCP
hosts. There is currently no tool to create this map.
Add D= mailer flag to specify a path of possible working directories
in which to execute the mailer. This is intended for the
prog mailer; some shells can get upset if they don't have
access to the current directory.
Add RFC 1413 (IDENT) protocol support. This is only very loosely
tested. This adds a $_ macro to be the authenticated
info (in ``user@domain [address]'' form) and debug flag
Check for loopbacks in usersmtp instead of srvrsmtp -- there is no
reason for a local agent to not be talking to the localhost
(although the inverse is not true).
Add a few hooks for automated map rebuilding. This is certainly
CONFIG: Have prog mailer specify a path of ``D=$z:/'' -- that is,
user's home directory then the root.
CONFIG: Log RFC 1413 identification in Received: line.
Fixes to requeueing code to make it compute priority, nrcpts,
Diagnose incorrect privacy flags. Suggested by Bryan Costales
Arrange to quote backslashes as well as other special characters
in the phrase part of a route-addr.
Some fixes to FallBackMX code suggested by Motonori Nakamura of
More vigorous zeroing of CurHostAddr to avoid logging of bogus
host addresses when you are actually just printing
information from the MCI structure; problem noted by
Michael Corrigan of U.C. San Diego.
Don't ignore rest of queue if any job is not runnable. This can
also cause an incorrect job to be lost. Fix from
Always respond "quickly" to RCPT command; do alias expansion and
the like later. This also means that mail for lists that
have errors will be acccepted, and an error sent back
later. This is done by instantiating the queue file
and then immediately running and requeueing it.
Fix incorrect diagnosis of infinite loop in ruleset. Problem noted
Improve information printed when infinite loops are discovered.
Zero CurHostAddr to fix erroneous internet addresses in log when no
addresses can be bound. Pointed out by Motonori Nakamura
"Probe" SMTP connections using RSET instead of NOOP "just in case".
Suggested by John Gardiner Myers of CMU.
Don't warn about -f if you are setting sender to yourself.
Fix incompatible call to endmailer in smtpquit which causes core
dumps. Noted by Allan E Johannesen of WPI.
HPUX portability changes from Michael J. Corrigan of UC San Diego.
Require MAIL before RCPT command in srvrsmtp.c. This had been
intentional from the 821 draft days when the order wasn't
Fix bug in nis_magic routine that was initializing parameters
incorrectly. Fix from Takahiro Kanbe of Fuji Xerox
Information Systems Co., Ltd.
Change default for PrivacyFlags in conf.c to 0 -- since it always
"or"s in new values, there was no way to turn off the
Add O option to set SMTP daemon options.
Add V option to set fallback MX host. This always sorts at lower
priority than anything it gets from the name server. It
should only be used for environments with very bad network
connectivity. Requested by several people.
Log sending info. It's not clear this is a good idea.
CONFIG: fix typo in mailertable code. Noted by Phil Brandenberger
CONFIG: add confDAEMON_OPTIONS and confFALLBACK_MX to set options
Fix botch in server SMTP that broke transactions that did not
use HELO first (like MH). Fix from Michael Corrigan
Fall back to other MX records if there is an error anywhere
in delivery (actually on MAIL or DATA -- RCPT is harder).
Suggested by John Gardiner Myers and Motonori Nakamura.
Revert to non-prototypes -- it turns out that our ANSI C
compiler is more forgiving than most others about
mixing prototyped extern declarations with non-prototyped
Fix a problem with multi-word class matching pointed out by
R$+ $=X $+ $: $1 < $2 > $3
the input "user@a.b.c" failed instead of being properly
rewritten as "user@a.<b>.c".
Neil also convinced me that it was correct that $~ should match
only one token -- the problem is that it's always possible
to add another token, so $~ matches far too eagerly.
Implement multi-word classes (properly!).
Add X-Authentication-Warning: headers to clue users into possible
attempts to forge mail. This is on the authwarnings
privacy flag, but is the default. Suggested by Bryan
Pass default units for convtime in so they can be more reasonable.
Allow config files to always add a new Comments: header (i.e.,
they will be added even if an old one already exists).
Suggested by Bryan Costales of ICSI.
Allow config files to delete an existing Return-Path: header.
These should only be added at final delivery. Suggested
by Bryan Costales of ICSI.
Some debugging additions. Suggested by Bryan Costales of ICSI.
Clean up logging of Family 0 addresses. Noted by David Muir
Add a "dequote" map class. This allows config files to strip
quotes off of addresses. Note that this is not a builtin
map, just a class -- so you have to define the map
Fix a bug in the queueup() loop getting a locked tf where in
very odd cases it can fall off the bottom and core dump.
Of course, it was P{r Emanuelsson who found it....
Open a new transcript when splitting an envelope. Problem found
by Allan E Johannesen of WPI.
Improved error output in endmailer if the mailer core dumps.
CONFIG: Fix typo in UUCP mailer definition.
CONFIG: Default several of the new options on: eight bit input,
privacy flags set to "authwarnings", and message warning
Fix problem with assumption of an sa_len field in a generic
sockaddr -- it turns out that most vendors haven't
picked up this (very important) fix.
Change compilation flags for daemon code -- select one or both
of NETINET or NETISO, but don't ever set DAEMON manually.
CONFIG: add FEATURE(mailertable) to do IDA-style mailertables.
Use Postmaster as default fallback return address, not root.
POSIX changes for file descriptor handling.
Diagnose errors writing new queue file.
If you change the owner using an owner- alias, also change the
error mode to EM_MAIL so that errors don't get dropped
into an inappropriate directory. Problem noted by
Allan E Johannesen of WPI.
If you are su'ed to root, send email as who you really are, not
as root. From Brian Kantor of U.C. San Diego.
Allow warning messages to be sent after a configurable interval
has passed without delivery. The message is sent only
once per envelope. This changes the format of the qf
file to have an F line, and the format of the T option
to accept take the format "return/warn" (both intervals).
Don't force all local names to lower case -- this was left over
from the wierd handling of case mapping on aliases. It
is now driven (as expected) by the "u" mailer flag.
Problem noted by P{r Emanuelsson.
Fix problem that caused headers on returned email to be trashed;
they were getting freed, but are still accessible via
Fix problem that caused bogus ids to be created on returned
Add support for ISO and other non-INET networking. This is by
no means finished yet. This does assume a lot of other
system support, like a version of gethostbyname that
returns non-AF_INET addresses.
CONFIG: change default on prog mailer to keep upper case in
user names (i.e., in the program command line).
CONFIG: strip trailing dots off of hosts in uucp mailer before
CONFIG: create new "relay" mailer for $R (LOCAL_RELAY) and $H
(MAIL_HUB) delivery that doesn't add local domain. Note
that this violates 821, but is probably "more correct"
for what we are trying to do. Problem pointed out by
Michael Graff of Iowa State.
Clean up unnecessary creates of queue ids (i.e., empty qf files)
when not needed, such as when starting up an SMTP
Fix problem where split envelopes aren't instantiated in the queue.
This is quite a serious bug.
Owner- aliases had problems with leading spaces causing a
Have ending 250 (after DATA) include the id; suggested by
Brian Kantor of UC San Diego.
Add logging on envelope splitting.
Change queue ids to have one more letter encoding the hour of
the day so that during a single day there is a greater
likelihood of uniqueness; requested by Brian Kantor.
Fix minor compile problem if LOCKF is defined.
Define size of tobuf in conf.h. Observed by Toshinari Takahashi
Restore e_sender -- this is equivalent to e_from.q_paddr without
decorations such as angle brackets and comments.
OSF/1 on Alpha changes from Allan E Johannesen of WPI.
CONFIG: fix typo in S3 for list syntax (;: => :;). Thanks to
Christopher Hoover for noting the problem.
Pass envelope to disconnect to avoid another use of CurEnv, which
can apparently end up being null at inopportune times.
Log "received from" as "relay=" for consistency (suggested by
Fix major bug in header handling: if no From: line existed in
the header (so sendmail inserts one), and the sender is
an alias that has an owner, the From: line shows the
owner (as well as the envelope). Fixed by early binding
the headers (which will change debugging output).
HPUX portability patches from Michael J. Corrigan of UC San Diego.
Some attempts to adapt better to out of open file conditions.
Some changes to ctladdr handling in queue files.
MAJOR CHANGE: delete e_sender and e_returnpath (why are these
different from e_from?) and $< macro.
Log correct IP address in relay= field even if the connection
Log "received from [RESPONSE]" on EF_RESPONSE messages (from
Fixes to SysExMsg logging (sometimes just got "message: %s"
instead of "message: error message"), noted by Eric
Wassenaar. Also reported by Motonori Nakamura.
Improvements to MX piggybacking code, from Motonori Nakamura.
Fix case where CurHostName points to an auto variable that has
been deallocated (from Motonori Nakamura).
Fix bug causing newlines to be included in aliases if option
"n" (check alias RHS) is set; bug noted by David Muir
Fix problem causing user names that should be mapped to lower
case to not be mapped if they are sent during a queue
run. This greatly simplifies the case mapping code.
Problem noted by Allan E Johannesen of WPI.
Don't do recipient address rewriting in buildaddr. This
improperly did recipient rewriting on sender addresses,
and just seems bogus in general -- but the change could
Pass TZ envariable to child processes for System V.
CONFIG: allow LOCAL_RULE_1 and LOCAL_RULE_2 if you want to
KNOWN PROBLEM: I have seen some problems on SunOS that causes
the User Data Base to give errors on some addresses. I
have tracked the problem back at least as far as 93.02.15
(version 6.22). Running with debugging on makes it
go away, so I conclude that it is referencing uninitialized
stack data. I haven't been able to track this down yet.
Allow local mailer to specify $@host -- this lets you assign the
"foo" part of jgm+foo to $h for passing in to the local
Additional debug printing in getcanonname (show query type).
Don't add the e_fromdomain on sender addresses -- this interacts
wierdly with the owner- code.
Improve delivery logging to not log obvious or meaningless stuff.
Include numeric IP address in Received: lines per RFC 1123 section
Fixed a bug in checking stat() return value if restrictmailq is
set. Also, check the entire group set instead of just the
primary group. Both from John Gardiner Myers.
Don't have usrerr automatically print errno, since this is often
Use transienterror() in makeconnection after connect() fails and
in openmailer after execve() fails (from Eric Wassenaar).
Also moved transienterror() from util.c to conf.c.
Clean up from= logging on response messages.
Undo patch allowing prescan to return a null vector -- it breaks
Config: FEATURE(notsticky) lets you use UDB for everything coming
in to the machine, even if it is specifically targetted
to this machine. Without it, UDB is bypassed if the user
Config: fix another minor botch with <> (local mailer wasn't
Fix getrealhostname to return null if sinlen <= 0 -- this can
occur if stdin is a pipe.
Avoid infinite loop in getcanonname if name server return
Config: avoid having C flag qualify list syntax and error syntax.
Fix logging in deliver to not pass too many parameters to Ultrix
Don't write the pid file until after the daemon has actually
opened and conditioned the connection.
Consider addresses "different" if their q_uids differ (so that
two users forwarding to the same program will be seen
as different, rather than the same).
Fix problem with bad parameters in main() -- they set ExitStat
Fix null pointer references through RealHostName -- painfully
discovered by Allan E Johannesen of WPI.
Fix bug causing user@@localhost to core dump (yuch).
Config: don't put two @host.dom.ain on users in $=E in SMTP
mailer. Also, catch user@ (no host) in ruleset 0.
Config: add confCW_FILE as the name of the cw configuration file
(defaults to /etc/sendmail.cw). From P{r Emanuelsson.
Allow prescan to return a pointer to an empty list -- this is
not an error. Also, clean up error reporting to avoid
double errors (prescan reports once, then the caller
Changes to avoid trusting T_ANY queries -- run them, but if you
don't get the info you expected, do T_A and T_MX queries
anyhow. This also fixes an oversight where _res.options
If PRIV_NOVRFY is set, use 252 response code instead of 502 per
RFC 1123 section 5.2.3. It's not 100% clear that this
is correct, but it probably works better with stupid
mailers that do a VRFY and only check the first digit.
Fix uninitialized variable "protocol" in smtp code.
Include <unistd.h> in sendmail.h -- move towards POSIX/ANSI.
Additional hooks for RFC 1427 (ESMTP SIZE extension). This
includes requiring that enoughspace() know the system
block size, which will undoubtedly break most ports.
Trace flag 19 in use for srvrsmtp.c.
Additional logging -- notably the sending mailer name. This
also changes the delivery logging to strict field=value
Fix some problems with messages getting sent even to addresses
that had been marked bad -- from Eric Wassenaar.
More WIDE changes: accept host name inside [...] as non-MXed
host. This is intended ONLY for use inside firewalled
environments, where the MX points at the gateway.
Change .cf file conventions so that mapping for <> addresses
don't have an @ in them (to avoid confusing the C mailer
flag). Pointed out by Neil Rickert.
Config extensions for Sam Leffler's FlexFAX software.
Fix some more bugs in alias owner code -- there were some wierd
cases where an error in a non-aliased name would override
the return info in an aliased name with an owner.
Changes from WIDE Project, forwarded to me by Motonori Nakamura:
Log actual delivery host (after MX et al); from
Deliver Postmaster copies without a body.
Better logging of SMTP senders.
Send all program email as daemon even when local.
As requested in various forms from many people, accept -qIstring
to limit queue runs to jobs with queue-id matching string.
Similarly for -qRstring for recipients, -qSstring for
Initial hooks for ESMTP support (see RFC 1425).
Fixed a syntax error in the UUCP mailer specification that caused
Check for missing A= or P= arguments in mailer definitions.
Require FROZENCONFIG compilation flag to include frozen
configuration code. Frozen configuration is really
not a very good idea any more, particularly in shared
Do better checking of errno after opens of :include: and .forward
files to defer delivery on network and other transient
errors. Suggestion from Craig Everhart.
Fix minor botch in read timeout macro processing.
Add FEATURE(nouucp) to config files for sites that know absolutely
Add built cf files to distribution tape and clarify how to build
them if you don't have the Berkeley make.
Some sizeof(long) portability changes for the Alpha, from Allan
Add "restrictmailq" privacy flag -- if set, only people in the same
group as your queue directory can print the queue. If you
set this, be sure you also restrict access to log files....
Fix another bug in owner-list stuff that can cause data files to
Fix a bug with queue runs that cause forwards to yourself to go
into alias/forwarding loops. I'm still iffy about this
Fix from Eric Wassenaar for suppression of return message code.
Fix yet another problem in alias owner code -- put the wrong return
address on the enclosed return-to-sender letter.
Fix botch in alias owner code that caused it to not operate if the
error was detected locally.
M_LOCAL => M_LOCALMAILER to avoid conflict with Ultrix include
Miscellaneous bug fixes from Eric Wassenaar:
sendmail -bv -t logs the from line even though in verify
sendmail -v can go into queue mode if shouldqueue returns
Add route-addr pruning per RFC 1123 section 5.3.3. This can be
disabled using the "R" option.
Delete (always undocumented) -R flag (save original recipients);
there are ways to syslog(3) these now.
Clean up SMTP reply codes -- specify them as needed in the code,
instead of in conf.c -- this was needed during the NCP to
TCP transition, but seems silly now. This also changes
parameters to message and nmessage.
Have mailstats read the .cf file to find the sendmail.st file and
get text versions of mailer names. An initial version of
this code was provided by Tuominen Keijo (although the
comments indicate the good bits were written by "E.V.").
Add yet more System V compatibility hacks.
Fix bug in VRFY code (assumes everything must be a local user).
Allow specification of any of the hard-wired pathnames in the
Delete concept of "trusted users" -- this really didn't provide
any security anyway, and caused some problems.
Delete last vestige of support for the word "at" as an equivalent
Propagate owner-foo alias information into the envelope sender.
Based on code from John Gardiner Myers. This is a major
semantic change -- beware!
Allow $@ on LHS to indicate "match zero" -- this is used to match
Don't "lose" queue runs. Very important fix from (who else?)
Completely reset state on RSET command -- from Eric Wassenaar.
Send error messages and return receipts using an envelope sender
of <> regardless of the setting of $n. Rewriting rules
can undo this if they feel the necessity, as might be
needed for networks that don't understand the syntax.
This is permitted by RFC 821 section 3.6 and required by
RFC 1123 section 5.3.3. THIS REQUIRES VERSION 4 CONFIG
FILES because the rulesets must be able to parse <>
Don't ever send error messages to "<>" -- they will get sent to
the local postmaster or dumped in /usr/tmp/dead.letter
instead. Per RFC 1123 section 5.3.3.
Explicitly check for email to yourself as a dotted quad. You
have to call $[ [ ... ] $] to get this.
Up the message timeout to five days per RFC 1123 section 5.3.1.1.
Make all read timeouts individually configurable, as strongly
recommended by RFC 1123 section 5.3.2.
Use f_bavail (blocks available to regular users) instead of f_bfree
(blocks available to superuser) in free block checks.
Change $d macro to be the current time, not the origination time,
since this is consistent with how it is used now.
Generalization of enoughspace from Eric Wassenaar covering
SGI, Apollo, HPUX, Ultrix, and SunOS.
Ignore process group signals -- some front ends can do this if
you kill a window too quickly. From Eric Wassenaar.
Close all cached connections before calling mailers and after
forking for delivery (caused double closes which resulted
Add FEATURE(redirect) in config files -- this allows you to alias
old addresses to a pointer to the new address that will
give a 551 error message, but not deliver the mail.
Some code changes to make the 551 errors look pretty.
Names of M4 program paths in config files have changed -- they
are all XXX_MAILER_PATH now, to match XXX_MAILER_FLAGS.
Fix a bug in the QSELFREF code having to do with empty .forward
files, reported by Eric Wassenaar.
Add option "p" (privacy flags); this allows you to tune how
picky the SMTP server will be. This also adds the
confPRIVACY_FLAGS M4 macro in the config files.
Add option "b" (minimum blocks free). If there are fewer than
this number of blocks free on the filesystem containing
the queue directory, the SMTP MAIL command will return
a 452 response and ask you to try again later. This
also adds the confMIN_FREE_BLOCKS M4 macro in the config
Made VRFY just verify (doesn't expand aliases and .forward files);
EXPN does full expansion. RCPT in queue-only mode also
doesn't chase aliases and .forward.
Increase the number of domain search entries in domain.c to allow
for the extra "" entry indicating the root domain.
Reported by Motonori Nakamura of Kyoto U.
Add a "SMART_HOST" in the configs for UUCP-connected sites that
want to forward all mail with extra "@"s to that site.
Also allows SMART_HOST, LOCAL_RELAY, and MAIL_HUB to
be specified as ``mailer:hostname'' to use an alternate
Clarified and updated some wording in the Operations Guide.
Add the "c" mailer flag -- this suppresses all comment parts of
addresses (requested by John Curran of NEARnet).
Have -v print prompts in -bt mode even if stdin is not a terminal
(default behaviour is to be silent if not reading from
a terminal). Suggested by Bryan Costales, ICSI.
Move the metacharacters from C0 space (\001-\037) into C1 space
(\201-\237). This also fixes a bunch of potential bugs
with G1 characters (\240-\276) in headers relating to
negative numbers passed to isspace() et al.
Add YP_LAST_MODIFIED and YP_MASTER_NAME to DBM version of alias
database if YPCOMPAT is #defined. Enhancement from
Takahiro Kanbe of Fuji Xerox Information Systems Co., Ltd.
Add "list" Precedence (-30); this can be used with old sendmails
which will map to precedence 0 (which will return error
messages). Suggested by Stephen R. van den Berg.
Many bug fixes from Eric Wassenaar of the National Institute for
Nuclear and High-Energy Physics, Amsterdam:
Clear timeouts properly on open failures in include().
Don't dereference through NULL if no home directory found.
Re-establish SIGCHLD signal on System 5 in reapchild().
Avoid NULL pointer reference on -pFOO flag.
Properly handle backslash escapes in comments.
Correctly check reply status on SMTP NOOP command.
Properly save SMTP error message if peer gives
"Service Shutting Down" message.
Avoid writing to the transcript if it couldn't be opened.
Signal errors in SMTP children to parent properly.
Handle self references in a list more globally (include a
QSELFREF bit in the address flags). This enhancement
was suggested by Eric Wassenaar.
Use initgroups() in hpux, even though it's System-V based. The
HASINITGROUPS compile flag can set this on other systems.
This HPUX behaviour was pointed out by Eric Wassenaar.
Clean up handling of LogLevel to make it easier to figure out
Change log levels to have some consistency:
1 serious system failures, security problems
2 lost communications, protocol failures
6 vrfy logging, creation of return-to-sender
9 delivery tempfails (queue ups)
Allow IDA-style separated processing on S= and R= in Mailer
definition lines. Note that rulesets 1 and 2 are
still used for both addresses as before. Bruce Lilly
gave a convincing argument that RFC976 insists on
Added some time zones to arpatounix -- they may not be in the
standards, but they are in use. However, I may delete
arpatounix entirely -- there appears to be no reason
Change to UUCP mailer (in cf directory) to try to do a saner job.
I'm still not certain about this mailer in general.
Fix bug that prevents saving letters in ~/dead.letter.
Don't add angle brackets in VRFY command if angle brackets already
Fix bogus error message in udbexpand.
Null terminate host buffers in buildaddr (broken in 6.21) --
Fix another incorrect error message in alias.c, found by Azuma
Fix a couple of problems in the more-configurable config files,
found by Tom Ivar Helbekkmo.
Fix problem with quoted :include: entries.
Don't duplicate the filename on verbose printing of .forward and
Extend size of prescan buffer (to allow bigger addresses). Also,
detect some buffer overflows.
Log user SMTP protocol errors (log level 4).
Fix another problem in the MCI state machine caused when there
were errors generated from the other end to commands
Include load average support for DEC Alpha running OSF/1.
Fix multiple-response problem with errors in MAIL From: line.
Fix SMTP reply codes for invalid address syntaxes (give 501;
never give multiple error messages for a single message).
Fix problem where a cached connection timeout rejects all
later connects to that host.
Fix incorrect error message if alias.c is compiled with DBM only.
Additional changes to fix nested conditionals (from Bruce Lilly).
Recover more gracefully from operating system failures, particularly
NULL returns from openmailer (from Noritoshi Demizu,
Log forward, alias, and userdb expand operations on log level 10;
concept suggested by P{r (Pell) Emanuelsson.
Changes for HPUX 8.07 compatibility.
Allow any config option to be set using an M4 define.
Change UNAME compile flag to HASUNAME for IDA compatibility
(besides, it's a better name).
Note in README that on SunOS it must be linked -Bstatic.
Fairly major change in domain.c to handle wildcard MX records
more rationally. NOTE: the "w" option (no wildcard MX
records match local domain) has been eliminated.
Fix some unset variable references pointed out by Bruce Lilly.
Fix host name in process titles when using cached connection.
Fix System 5 compatibility changes to be compatible with the rest
Experimental fix for problem handling errors in the SMTP
protocol in conjunction with connection caching.
System 5 compatibility changes.
Fix a bug that causes local mail delivered using -odq to be
eliminated as a duplicate (because it matched the
ctladdr, now passed in as a C line). These changes
Add debugging for some MCI errors.
Fix -e compatibility flag to take a value.
Fix a couple of minor compilation warnings on Sun cc.
Improve error messages in a few cases to be more self-explanatory.
Fix yet-another problem with environment handling, pointed out
by Yoshitaka Tokugawa and Tom Ivar Helbekkmo.
Some heuristics to try to limit resource exhaustion problems
if a downstream host has been down for a long time.
Fix problem with incorrect host name being logged in "Connection
timed out" messages (from Tom Ivar Helbekkmo).
Fix some ANSI C problems (from Takahiro Kanbe).
Properly log message sender on returned mail during queue run.
Count number of recipients properly.
Fix a problem in yp map code.
Diagnose "message timed out" (from Motonori Nakamura).
Fix problem with address delimitor inside quotes.
Define $k and $=k to be the UUCP name (from the uname call)
based on code from Bruce Lilly.
Implement arpatounix (largely code from Bruce Lilly).
Log more info (suggested by John Myers).
Allow nested $?...$|...$. (inspired by code from Bruce Lilly of
POSIX compatibility (noted by Keith Bostic).
Handle SMTP MAIL command errors properly (urged by several people,
notably John Myers of CMU).
Do early diagnosis of .cf errors (notably referencing a RHS
substitution that isn't on the LHS).
Adjust checkpointing to better handle batched recipients, suggested
(config files:) Implement MAIL_HUB for all local mail (to handle
NFS-mounted directories) as urged by Tom Ivar Helbekkmo
of the Norwegian School of Economics.
Environment handling simplification/bug fix -- child processes
get a minimal, fixed environment. This avoids different
Handle commas inside comments properly.
Properly limit large messages submitted in -obq mode.
Check mtime of thaw file against .cf and sendmail binary, based on
MX piggybacking, based on code from John Myers@CMU.
Allow checkcompat to return -1 to mean tempfail.
Bug fix in m_mno computation.
Tuning of queueing functions as recommended by John Gardiner Myers.
Return mail headers (no body) on messages with negative precedence.
Fix botch causing queued headers to have ?XX? prefixes.
Changes to recognize special mailer types (e.g., file) early.
Check for control characters in addresses.
Fixed deferred error reporting.
Report duplicate aliases.
Handle mixed case recursive aliases.
Put return-receipt-to on a conf.c flag (but don't set it).
Fix minor syslog problem.
projects / unix-history / blob