Fix buffer overflow vulnerability in spa_base64_to_bits() function.

[exim.git] / doc / doc-txt / ChangeLog

$Cambridge: exim/doc/doc-txt/ChangeLog,v 1.59 2004/12/29 10:55:58 ph10 Exp $

Change log file for Exim from version 4.21
-------------------------------------------

Exim version 4.50
-----------------

 1. Minor wording change to the doc/README.SIEVE file.

 2. Change 4.43/35 introduced a bug: if quota_filecount was set, the
    computation of the current number of files was incorrect.

 3. Closing a stable door: arrange to panic-die if setitimer() ever fails. The
    bug fixed in 4.43/37 would have been diagnosed quickly if this had been in
    place.

 4. Give more explanation in the error message when the command for a transport
    filter fails to execute.

 5. There are several places where Exim runs a non-Exim command in a
    subprocess. The SIGUSR1 signal should be disabled for these processes. This
    was being done only for the command run by the queryprogram router. It is
    now done for all such subprocesses. The other cases are: ${run, transport
    filters, and the commands run by the lmtp and pipe transports.

 6. Added CONFIGURE_GROUP build-time option.

 7. Some older OS have a limit of 256 on the maximum number of file
    descriptors. Exim was using setrlimit() to set 1000 as a large value
    unlikely to be exceeded. Change 4.43/17 caused a lot of logging on these
    systems. I've change it so that if it can't get 1000, it tries for 256.

 8. "control=submission" was allowed, but had no effect, in a DATA ACL. This
    was an oversight, and furthermore, ever since the addition of extra
    controls (e.g. 4.43/32), the checks on when to allow different forms of
    "control" were broken. There should now be diagnostics for all cases when a
    control that does not make sense is encountered.

 9. Added the /retain_sender option to "control=submission".

10. $recipients is now available in the predata ACL (oversight).

11. Tidy the search cache before the fork to do a delivery from a message
    received from the command line. Otherwise the child will trigger a lookup
    failure and thereby defer the delivery if it tries to use (for example) a
    cached ldap connection that the parent has called unbind on.

12. If verify=recipient was followed by verify=sender in a RCPT ACL, the value
    of $address_data from the recipient verification was clobbered by the
    sender verification.

13. The value of address_data from a sender verification is now available in
    $sender_address_data in subsequent conditions in the ACL statement.

14. Added forbid_sieve_filter and forbid_exim_filter to the redirect router.

15. Added a new option "connect=<time>" to callout options, to set a different
    connection timeout.

16. If FIXED_NEVER_USERS was defined, but empty, Exim was assuming the uid 0
    was its contents. (It was OK if the option was not defined at all.)

17. A "Completed" log line is now written for messages that are removed from
    the spool by the -Mrm option.

18. New variables $sender_verify_failure and $recipient_verify_failure contain
    information about exactly what failed.

19. Added -dd to debug only the daemon process.

20. Incorporated Michael Haardt's patch to ldap.c for improving the way it
    handles timeouts, both on the server side and network timeouts. Renamed the
    CONNECT parameter as NETTIMEOUT (but kept the old name for compatibility).

21. The rare case of EHLO->STARTTLS->HELO was setting the protocol to "smtp".
    It is now set to "smtps".

22. $host_address is now set to the target address during the checking of
    ignore_target_hosts.

23. When checking ignore_target_hosts for an ipliteral router, no host name was
    being passed; this would have caused $sender_host_name to have been used if
    matching the list had actually called for a host name (not very likely,
    since this list is usually IP addresses). A host name is now passed as
    "[x.x.x.x]".

24. Changed the calls that set up the SIGCHLD handler in the daemon to use the
    code that specifies a non-restarting handler (typically sigaction() in
    modern systems) in an attempt to fix a rare and obscure crash bug.

25. Narrowed the window for a race in the daemon that could cause it to ignore
    SIGCHLD signals. This is not a major problem, because they are used only to
    wake it up if nothing else does.

26. A malformed maildirsize file could cause Exim to calculate negative values
    for the mailbox size or file count. Odd effects could occur as a result.
    The maildirsize information is now recalculated if the size or filecount
    end up negative.

27. Added HAVE_SYS_STATVFS_H to the os.h file for Linux, as it has had this
    support for a long time. Removed HAVE_SYS_VFS_H.

28. Installed the latest version of exipick from John Jetmore.

29. In an address list, if the pattern was not a regular expression, an empty
    subject address (from a bounce message) matched only if the pattern was an
    empty string. Non-empty patterns were not even tested. This was the wrong
    because it is perfectly reasonable to use an empty address as part of a
    database query. An empty address is now tested by patterns that are
    lookups. However, all the other forms of pattern expect the subject to
    contain a local part and a domain, and therefore, for them, an empty
    address still always fails if the pattern is not itself empty.

30. Exim went into a mad DNS loop when attempting to do a callout where the
    host was specified on an smtp transport, and looking it up yielded more
    than one IP address.

31. Re-factored the code for checking spool and log partition space into a
    function that finds that data and another that does the check. The former
    is then used to implement four new variables: $spool_space, $log_space,
    $spool_inodes, and $log_inodes.

32. The RFC2047 encoding function was originally intended for short strings
    such as real names; it was not keeping to the 75-character limit for
    encoded words that the RFC imposes. It now respects the limit, and
    generates multiple encoded words if necessary. To be on the safe side, I
    have increased the buffer size for the ${rfc2047: expansion operator from
    1024 to 2048 bytes.

33. It is now permitted to omit both strings after an "if" condition; if the
    condition is true, the result is "true". As before, when the second string
    is omitted, a false condition yields an empty string. This makes it less
    cumbersome to write custom ACL and router conditions.

34. Failure to deliver a bounce message always caused it to be frozen, even if
    there was an errors_to setting on the router. The errors_to setting is now
    respected.

35. If an IPv6 address is given for -bh or -bhc, it is now converted to the
    canonical form (fully expanded) before being placed in
    $sender_host_address.

36. The table in the code that translates DNS record types into text (T_A to
    "A" for instance) was missing entries for NS and CNAME. It is just possible
    that this could have caused confusion if both these types were looked up
    for the same domain, because the text type is used as part of Exim's
    per-process caching. But the chance of anyone hitting this buglet seems
    very small.

37. The dnsdb lookup has been extended in a number of ways.

    (1) There is a new type, "zns", which walks up the domain tree until it
        finds some nameserver records. It should be used with care.

    (2) There is a new type, "mxh", which is like "mx" except that it returns
        just the host names, not the priorities.

    (3) It is now possible to give a list of domains (or IP addresses) to be
        looked up. The behaviour when one of the lookups defers can be
        controlled by a keyword.

    (4) It is now possible to specify the separator character for use when
        multiple records are returned.

38. The dnslists ACL condition has been extended: it is now possible to supply
    a list of IP addresses and/or domains to be looked up in a particular DNS
    domain.

39. Added log_selector=+queue_time_overall.

40. When running the queue in the test harness, wait just a tad after forking a
    delivery process, to get repeatability of debugging output.

41. Include certificate and key file names in error message when GnuTLS fails
    to set them up, because the GnuTLS error message doesn't include the name
    of the failing file when there is a problem reading it.

42. Allow both -bf and -bF in the same test run.

43. Did the same fix as 41 above for OpenSSL, which had the same infelicity.

44. The "Exiscan patch" is now merged into the mainline Exim source.

45. Sometimes the final signoff response after QUIT could fail to get
    transmitted in the non-TLS case. Testing !tls_active instead of tls_active
    < 0 before doing a fflush(). This bug looks as though it goes back to the
    introduction of TLS in release 3.20, but "sometimes" must have been rare
    because the tests only now provoked it.

46. Reset the locale to "C" after calling embedded Perl, in case it was changed
    (this can affect the format of dates).

47. exim_tidydb, when checking for the continued existence of a message for
    which it has found a message-specific retry record, was not finding
    messages that were in split spool directories. Consequently, it was
    deleting retry records that should have stayed in existence.

48. Steve fixed some bugs in eximstats.

49. The SPA authentication driver was not abandoning authentication and moving
    on to the next authenticator when an expansion was forced to fail,
    contradicting the general specification for all authenticators. Instead it
    was generating a temporary error. It now behaves as specified.

50. The default ordering of permitted cipher suites for GnuTLS was pessimal
    (the order specifies the preference for clients). The order is now AES256,
    AES128, 3DES, ARCFOUR128.

51. Small patch to Sieve code - explicitly set From: when generating an
    autoreply.

52. Exim crashed if a remote delivery caused a very long error message to be
    recorded - for instance if somebody sent an entire SpamAssassin report back
    as a large number of 550 error lines. This bug was coincidentally fixed by
    increasing the size of one of Exim's internal buffers (big_buffer) that
    happened as part of the Exiscan merge. However, to be on the safe side, I
    have made the code more robust (and fixed the comments that describe what
    is going on).

53. Now that there can be additional text after "Completed" in log lines (if
    the queue_time_overall log selector is set), a one-byte patch to exigrep
    was needed to allow it to recognize "Completed" as not the last thing in
    the line.

54. The LDAP lookup was not handling a return of LDAP_RES_SEARCH_REFERENCE. A
    patch that reportedly fixes this has been added. I am not expert enough to
    create a test for it. This is what the patch creator wrote:

      "I found a little strange behaviour of ldap code when working with
      Windows 2003 AD Domain, where users was placed in more than one
      Organization Units. When I tried to give exim partial DN, the exit code
      of ldap_search was unknown to exim because of LDAP_RES_SEARCH_REFERENCE.
      But simultaneously result of request was absolutely normal ldap result,
      so I produce this patch..."

55. Some experimental protocols are using DNS PTR records for new purposes. The
    keys for these records are domain names, not reversed IP addresses. The
    dnsdb PTR lookup now tests whether its key is an IP address. If not, it
    leaves it alone. Component reversal etc. now happens only for IP addresses.

56. Improve error message when ldap_search() fails in OpenLDAP or Solaris LDAP.

57. Double the size of the debug message buffer (to 2048) so that more of very
    long debug lines gets shown.

58. The exicyclog utility now does better if the number of log files to keep
    exceeds 99. In this case, it numbers them 001, 002 ... instead of 01, 02...

59. Two changes related to the smtp_active_hostname option:

      (1) $smtp_active_hostname is now available as a variable.
      (2) The default for smtp_banner uses $smtp_active_hostname instead
          of $primary_hostname.

60. The host_aton() function is supposed to be passed a string that is known
    to be a valid IP address. However, in the case of IPv6 addresses, it was
    not checking this. This is a hostage to fortune. Exim now panics and dies
    if the condition is not met. A case was found where this could be provoked
    from a dnsdb PTR lookup with an IPv6 address that had more than 8
    components; fortuitously, this particular loophole had already been fixed
    by change 4.50/55 above.

    If there are any other similar loopholes, the new check in host_aton()
    itself should stop them being exploited. The report I received stated that
    data on the command line could provoke the exploit when Exim was running as
    exim, but did not say which command line option was involved. All I could
    find was the use of -be with a bad dnsdb PTR lookup, and in that case it is
    running as the user.

61. There was a buffer overflow vulnerability in the SPA authentication code
    (which came originally from the Samba project). I have added a test to the
    spa_base64_to_bits() function which I hope fixes it.


Exim version 4.43
-----------------

 1. Fixed a longstanding but relatively impotent bug: a long time ago, before
    PIPELINING, the function smtp_write_command() used to return TRUE or FALSE.
    Now it returns an integer. A number of calls were still expecting a T/F
    return. Fortuitously, in all cases, the tests worked in OK situations,
    which is the norm. However, things would have gone wrong on any write
    failures on the smtp file descriptor. This function is used when sending
    messages over SMTP and also when doing verify callouts.

 2. When Exim is called to do synchronous delivery of a locally submitted
    message (the -odf or -odi options), it no longer closes stderr before doing
    the delivery.

 3. Implemented the mua_wrapper option.

 4. Implemented mx_fail_domains and srv_fail_domains for the dnslookup router.

 5. Implemented the functions header_remove(), header_testname(),
    header_add_at_position(), and receive_remove_recipient(), and exported them
    to local_scan().

 6. If an ACL "warn" statement specified the addition of headers, Exim already
    inserted X-ACL-Warn: at the start if there was no header name. However, it
    was not making this test for the second and subsequent header lines if
    there were newlines in the string. This meant that an invalid header could
    be inserted if Exim was badly configured.

 7. Allow an ACL "warn" statement to add header lines at the start or after all
    the Received: headers, as well as at the end.

 8. Added the rcpt_4xx retry error code.

 9. Added postmaster_mailfrom=xxx to callout verification option.

10. Added mailfrom=xxxx to the callout verification option, for verify=
    header_sender only.

11. ${substr_1_:xxxx} and ${substr__3:xxxx} are now diagnosed as syntax errors
    (they previously behaved as ${substr_1_0:xxxx} and ${substr:_0_3:xxxx}).

12. Inserted some casts to stop certain compilers warning when using pointer
    differences as field lengths or precisions in printf-type calls (mostly
    affecting debugging statements).

13. Added optional readline() support for -be (dynamically loaded).

14. Obscure bug fix: if a message error (e.g. 4xx to MAIL) happened within the
    same clock tick as a message's arrival, so that its received time was the
    same as the "first fail" time on the retry record, and that message
    remained on the queue past the ultimate address timeout, every queue runner
    would try a delivery (because it was past the ultimate address timeout) but
    after another failure, the ultimate address timeout, which should have then
    bounced the address, did not kick in. This was a "< instead of <=" error;
    in most cases the first failure would have been in the next clock tick
    after the received time, and all would be well.

15. The special items beginning with @ in domain lists (e.g. @mx_any) were not
    being recognized when the domain list was tested by the match_domain
    condition in an expansion string.

16. Added the ${str2b64: operator.

17. Exim was always calling setrlimit() to set a large limit for the number of
    processes, without checking whether the existing limit was already
    adequate. (It did check for the limit on file descriptors.) Furthermore,
    errors from getrlimit() and setrlimit() were being ignored. Now they are
    logged to the main and panic logs, but Exim does carry on, to try to do its
    job under whatever limits there are.

18. Imported PCRE 5.0.

19. Trivial typo in log message " temporarily refused connection" (the leading
    space).

20. If the log selector return_path_on_delivery was set and an address was
    redirected to /dev/null, the delivery process crashed because it assumed
    that a return path would always be set for a "successful" delivery. In this
    case, the whole delivery is bypassed as an optimization, and therefore no
    return path is set.

21. Internal re-arrangement: the function for sending a challenge and reading
    a response while authentication was assuming a zero-terminated challenge
    string. It's now changed to take a pointer and a length, to allow for
    binary data in such strings.

22. Added the cyrus_sasl authenticator (code supplied by MBM).

23. Exim was not respecting finduser_retries when seeking the login of the
    uid under which it was called; it was always trying 10 times. (The default
    setting of finduser_retries is zero.) Also, it was sleeping after the final
    failure, which is pointless.

24. Implemented tls_on_connect_ports.

25. Implemented acl_smtp_predata.

26. If the domain in control=submission is set empty, Exim assumes that the
    authenticated id is a complete email address when it generates From: or
    Sender: header lines.

27. Added "#define SOCKLEN_T int" to OS/os.h-SCO and OS/os.h-SCO_SV. Also added
    definitions to OS/Makefile-SCO and OS/Makefile-SCO_SV that put basename,
    chown and chgrp in /bin and hostname in /usr/bin.

28. Exim was keeping the "process log" file open after each use, just as it
    does for the main log. This opens the possibility of it remaining open for
    long periods when the USR1 signal hits a daemon. Occasional processlog
    errors were reported, that could have been caused by this. Anyway, it seems
    much more sensible not to leave this file open at all, so that is what now
    happens.

29. The long-running daemon process does not normally write to the log once it
    has entered its main loop, and it closes the log before doing so. This is
    so that log files can straightforwardly be renamed and moved. However,
    there are a couple of unusual error situations where the daemon does write
    log entries, and I had neglected to close the log afterwards.

30. The text of an SMTP error response that was received during a remote
    delivery was being truncated at 512 bytes. This is too short for some of
    the long messages that one sometimes sees. I've increased the limit to
    1024.

31. It is now possible to make retry rules that apply only when a message has a
    specific sender, in particular, an empty sender.

32. Added "control = enforce_sync" and "control = no_enforce_sync". This makes
    it possible to be selective about when SMTP synchronization is enforced.

33. Added "control = caseful_local_part" and "control = "caselower_local_part".

32. Implemented hosts_connection_nolog.

33. Added an ACL for QUIT.

34. Setting "delay_warning=" to disable warnings was not working; it gave a
    syntax error.

35. Added mailbox_size and mailbox_filecount to appendfile.

36. Added control = no_multiline_responses to ACLs.

37. There was a bug in the logic of the code that waits for the clock to tick
    in the case where the clock went backwards by a substantial amount such
    that the microsecond fraction of "now" was more than the microsecond
    fraction of "then" (but the whole seconds number was less).

38. Added support for the libradius Radius client library this is found on
    FreeBSD (previously only the radiusclient library was supported).


Exim version 4.42
-----------------

 1. When certain lookups returned multiple values in the form name=value, the
    quoting of the values was not always being done properly. Specifically:
    (a) If the value started with a double quote, but contained no whitespace,
        it was not quoted.
    (b) If the value contained whitespace other than a space character (i.e.
        tabs or newlines or carriage returns) it was not quoted.
    This fix has been applied to the mysql and pgsql lookups by writing a
    separate quoting function and calling it from the lookup code. The fix
    should probably also be applied to nisplus, ibase and oracle lookups, but
    since I cannot test any of those, I have not disturbed their existing code.

 2. A hit in the callout cache for a specific address caused a log line with no
    reason for rejecting RCPT. Now it says "Previous (cached) callout
    verification failure".

 3. There was an off-by-one bug in the queryprogram router. An over-long
    return line was truncated at 256 instead of 255 characters, thereby
    overflowing its buffer with the terminating zero. As well as fixing this, I
    have increased the buffer size to 1024 (and made a note to document this).

 4. If an interrupt, such as the USR1 signal that is send by exiwhat, arrives
    when Exim is waiting for an SMTP response from a remote server, Exim
    restarts its select() call on the socket, thereby resetting its timeout.
    This is not a problem when such interrupts are rare. Somebody set up a cron
    job to run exiwhat every 2 minutes, which is less than the normal select()
    timeout (5 or 10 minutes). This meant that the select() timeout never
    kicked in because it was always reset. I have fixed this by comparing the
    time when an interrupt arrives with the time at the start of the first call
    to select(). If more time than the timeout has elapsed, the interrupt is
    treated as a timeout.

 5. Some internal re-factoring in preparation for the addition of Sieve
    extensions (by MH). In particular, the "personal" test is moved to a
    separate function, and given an option for scanning Cc: and Bcc: (which is
    not set for Exim filters).

 6. When Exim created an email address using the login of the caller as the
    local part (e.g. when creating a From: or Sender: header line), it was not
    quoting the local part when it contained special characters such as @.

 7. Installed new OpenBSD configuration files.

 8. Reworded some messages for syntax errors in "and" and "or" conditions to
    try to make them clearer.

 9. Callout options, other than the timeout value, were being ignored when
    verifying sender addresses in header lines. For example, when using

      verify = header_sender/callout=no_cache

    the cache was (incorrectly) being used.

10. Added a missing instance of ${EXE} to the exim_install script; this affects
    only the Cygwin environment.

11. When return_path_on_delivery was set as a log selector, if different remote
    addresses in the same message used different return paths and parallel
    remote delivery occurred, the wrong values would sometimes be logged.
    (Whenever a remote delivery process finished, the return path value from
    the most recently started remote delivery process was logged.)

12. RFC 3848 specifies standard names for the "with" phrase in Received: header
    lines when AUTH and/or TLS are in use. This is the "received protocol"
    field. Exim used to use "asmtp" for authenticated SMTP, without any
    indication (in the protocol name) for TLS use. Now it follows the RFC and
    uses "esmtpa" if the connection is authenticated, "esmtps" if it is
    encrypted, and "esmtpsa" if it is both encrypted and authenticated. These
    names appear in log lines as well as in Received: header lines.

13. Installed MH's patches for Sieve to add the "copy" and "vacation"
    extensions, and comparison tests, and to fix some bugs.

14. Changes to the "personal" filter test:

    (1) The test was buggy in that it was just doing the equivalent of
    "contains" tests on header lines. For example, if a user's address was
    anne@some.where, the "personal" test would incorrectly be true for

      To: susanne@some.where

    This test is now done by extracting each address from the header in turn,
    and checking the entire address. Other tests that are part of "personal"
    are now done using regular expressions (for example, to check local parts
    of addresses in From: header lines).

    (2) The list of non-personal local parts in From: addresses has been
    extended to include "listserv", "majordomo", "*-request", and "owner-*",
    taken from the Sieve specification recommendations.

    (3) If the message contains any header line starting with "List-" it is
    treated as non-personal.

    (4) The test for "circular" in the Subject: header line has been removed
    because it now seems ill-conceived.

15. Minor typos in src/EDITME comments corrected.

16. Installed latest exipick from John Jetmore.

17. If headers_add on a router specified a text string that was too long for
    string_sprintf() - that is, longer than 8192 bytes - Exim panicked. The use
    of string_sprintf() is now avoided.

18. $message_body_size was not set (it was always zero) when running the DATA
    ACL and the local_scan() function.

19. For the "mail" command in an Exim filter, no default was being set for
    the once_repeat time, causing a random time value to be used if "once" was
    specified. (If the value happened to be <= 0, no repeat happened.) The
    default is now 0s, meaning "never repeat". The "vacation" command was OK
    (its default is 7d). It's somewhat surprising nobody ever noticed this bug
    (I found it when inspecting the code).

20. There is now an overall timeout for performing a callout verification. It
    defaults to 4 times the callout timeout, which applies to individual SMTP
    commands during the callout. The overall timeout applies when there is more
    than one host that can be tried. The timeout is checked before trying the
    next host. This prevents very long delays if there are a large number of
    hosts and all are timing out (e.g. when the network connections are timing
    out). The value of the overall timeout can be changed by specifying an
    additional sub-option for "callout", called "maxwait". For example:

      verify = sender/callout=5s,maxwait=20s

21. Add O_APPEND to the open() call for maildirsize files (Exim already seeks
    to the end before writing, but this should make it even safer).

22. Exim was forgetting that it had advertised PIPELINING for the second and
    subsequent messages on an SMTP connection. It was also not resetting its
    memory on STARTTLS and an internal HELO.

23. When Exim logs an SMTP synchronization error within a session, it now
    records whether PIPELINING has been advertised or not.

24. Added 3 instances of "(long int)" casts to time_t variables that were being
    formatted using %ld, because on OpenBSD (and perhaps others), time_t is int
    rather than long int.

25. Installed the latest Cygwin configuration files from the Cygwin maintainer.

26. Added the never_mail option to autoreply.


Exim version 4.41
-----------------

 1. A reorganization of the code in order to implement 4.40/8 caused a daemon
    crash if the getsockname() call failed; this can happen if a connection is
    closed very soon after it is established. The problem was simply in the
    order in which certain operations were done, causing Exim to try to write
    to the SMTP stream before it had set up the file descriptor. The bug has
    been fixed by making things happen in the correct order.


Exim version 4.40
-----------------

 1. If "drop" was used in a DATA ACL, the SMTP output buffer was not flushed
    before the connection was closed, thus losing the rejection response.

 2. Commented out the definition of SOCKLEN_T in os.h-SunOS5. It is needed for
    some early Solaris releases, but causes trouble in current releases where
    socklen_t is defined.

 3. When std{in,out,err} are closed, re-open them to /dev/null so that they
    always exist.

 4. Minor refactoring of os.c-Linux to avoid compiler warning when IPv6 is not
    configured.

 5. Refactoring in expand.c to improve memory usage. Pre-allocate a block so
    that releasing the top of it at the end releases what was used for sub-
    expansions (unless the block got too big). However, discard this block if
    the first thing is a variable or header, so that we can use its block when
    it is dynamic (useful for very large $message_headers, for example).

 6. Lookups now cache *every* query, not just the most recent. A new, separate
    store pool is used for this. It can be recovered when all lookup caches are
    flushed. Lookups now release memory at the end of their result strings.
    This has involved some general refactoring of the lookup sources.

 7. Some code has been added to the store_xxx() functions to reduce the amount
    of flapping under certain conditions.

 8. log_incoming_interface used to affect only the <= reception log lines. Now
    it causes the local interface and port to be added to several more SMTP log
    lines, for example "SMTP connection from", and rejection lines.

 9. The Sieve author supplied some patches for the doc/README.SIEVE file.

10. Added a conditional definition of _BSD_SOCKLEN_T to os.h-Darwin.

11. If $host_data was set by virtue of a hosts lookup in an ACL, its value
    could be overwritten at the end of the current message (or the start of a
    new message if it was set in a HELO ACL). The value is now preserved for
    the duration of the SMTP connection.

12. If a transport had a headers_rewrite setting, and a matching header line
    contained an unqualified address, that address was qualified, even if it
    did not match any rewriting rules. The underlying bug was that the values
    of the flags that permit the existence of unqualified sender and recipient
    addresses in header lines (set by {sender,recipient}_unqualified_hosts for
    non-local messages, and by -bnq for local messages) were not being
    preserved with the message after it was received.

13. When Exim was logging an SMTP synchronization error, it could sometimes log
    "next input=" as part of the text comprising the host identity instead of
    the correct text. The code was using the same buffer for two different
    strings. However, depending on which order the printing function evaluated
    its arguments, the bug did not always show up. Under Linux, for example, my
    test suite worked just fine.

14. Exigrep contained a use of Perl's "our" scoping after change 4.31/70. This
    doesn't work with some older versions of Perl. It has been changed to "my",
    which in any case is probably the better facility to use.

15. A really picky compiler found some instances of statements for creating
    error messages that either had too many or two few arguments for the format
    string.

16. The size of the buffer for calls to the DNS resolver has been increased
    from 1024 to 2048. A larger buffer is needed when performing PTR lookups
    for addresses that have a lot of PTR records. This alleviates a problem; it
    does not fully solve it.

17. A dnsdb lookup for PTR records that receives more data than will fit in the
    buffer now truncates the list and logs the incident, which is the same
    action as happens when Exim is looking up a host name and its aliases.
    Previously in this situation something unpredictable would happen;
    sometimes it was "internal error: store_reset failed".

18. If a server dropped the connection unexpectedly when an Exim client was
    using GnuTLS and trying to read a response, the client delivery process
    crashed while trying to generate an error log message.

19. If a "warn" verb in an ACL added multiple headers to a message in a single
    string, for example:

      warn message = H1: something\nH2: something

    the text was added as a single header line from Exim's point of view
    though it ended up OK in the delivered message. However, searching for the
    second and subsequent header lines using $h_h2: did not work. This has been
    fixed. Similarly, if a system filter added multiple headers in this way,
    the routers could not see them.

20. Expanded the error message when iplsearch is called with an invalid key to
    suggest using net-iplsearch in a host list.

21. When running tests using -bh, any delays imposed by "delay" modifiers in
    ACLs are no longer actually imposed (and a message to that effect is
    output).

22. If a "gecos" field in a passwd entry contained escaped characters, in
    particular, if it contained a \" sequence, Exim got it wrong when building
    a From: or a Sender: header from that name. A second bug also caused
    incorrect handling when an unquoted " was present following a character
    that needed quoting.

23. "{crypt}" as a password encryption mechanism for a "crypteq" expansion item
    was not being matched caselessly.

24. Arranged for all hyphens in the exim.8 source to be escaped with
    backslashes.

25. Change 16 of 4.32, which reversed 71 or 4.31 didn't quite do the job
    properly. Recipient callout cache records were still being keyed to include
    the sender, even when use_sender was set false. This led to far more
    callouts that were necessary. The sender is no longer included in the key
    when use_sender is false.

26. Added "control = submission" modifier to ACLs.

27. Added the ${base62d: operator to decode base 62 numbers.

28. dnsdb lookups can now access SRV records.

29. CONFIGURE_OWNER can be set at build time to define an alternative owner for
    the configuration file.

30. The debug message "delivering xxxxxx-xxxxxx-xx" is now output in verbose
    (-v) mode. This makes the output for a verbose queue run more intelligible.

31. Added a use_postmaster feature to recipient callouts.

32. Added the $body_zerocount variable, containing the number of binary zero
    bytes in the message body.

33. The time of last modification of the "new" subdirectory is now used as the
    "mailbox time last read" when there is a quota error for a maildir
    delivery.

34. Added string comparison operators lt, lti, le, lei, gt, gti, ge, gei.

35. Added +ignore_unknown as a special item in host lists.

36. Code for decoding IPv6 addresses in host lists is now included, even if
    IPv6 support is not being compiled. This fixes a bug in which an IPv6
    address was recognized as an IP address, but was then not correctly decoded
    into binary, causing unexpected and incorrect effects when compared with
    another IP address.


Exim version 4.34
-----------------

 1. Very minor rewording of debugging text in manualroute to say "list of
    hosts" instead of "hostlist".

 2. If verify=header_syntax was set, and a header line with an unqualified
    address (no domain) and a large number of spaces between the end of the
    name and the colon was received, the reception process suffered a buffer
    overflow, and (when I tested it) crashed. This was caused by some obsolete
    code that should have been removed. The fix is to remove it!

 3. When running in the test harness, delay a bit after writing a bounce
    message to get a bit more predictability in the log output.

 4. Added a call to search_tidyup() just before forking a reception process. In
    theory, someone could use a lookup in the expansion of smtp_accept_max_
    per_host which, without the tidyup, could leave open a database connection.

 5. Added the variables $recipient_data and $sender_data which get set from a
    lookup success in an ACL "recipients" or "senders" condition, or a router
    "senders" option, similar to $domain_data and $local_part_data.

 6. Moved the writing of debug_print from before to after the "senders" test
    for routers.

 7. Change 4.31/66 (moving the time when the Received: is generated) caused
    problems for message scanning, either using a data ACL, or using
    local_scan() because the Received: header was not generated till after they
    were called (in order to set the time as the time of reception completion).
    I have revised the way this works. The header is now generated after the
    body is received, but before the ACL or local_scan() are called. After they
    are run, the timestamp in the header is updated.


Exim version 4.33
-----------------

 1. Change 4.24/6 introduced a bug because the SIGALRM handler was disabled
    before starting a queue runner without re-exec. This happened only when
    deliver_drop_privilege was set or when the Exim user was set to root. The
    effect of the bug was that timeouts during subsequent deliveries caused
    crashes instead of being properly handled. The handler is now left at its
    default (and expected) setting.

 2. The other case in which a daemon avoids a re-exec is to deliver an incoming
    message, again when deliver_drop_privilege is set or Exim is run as root.
    The bug described in (1) was not present in this case, but the tidying up
    of the other signals was missing. I have made the two cases consistent.

 3. The ignore_target_hosts setting on a manualroute router was being ignored
    for hosts that were looked up using the /MX notation.

 4. Added /ignore=<ip list> feature to @mx_any, @mx_primary, and @mx_secondary
    in domain lists.

 5. Change 4.31/55 was buggy, and broke when there was a rewriting rule that
    operated on the sender address. After changing the $sender_address to <>
    for the sender address verify, Exim was re-instated it as the original
    (before rewriting) address, but remembering that it had rewritten it, so it
    wasn't rewriting it again. This bug also had the effect of breaking the
    sender address verification caching when the sender address was rewritten.

 6. The ignore_target_hosts option was being ignored by the ipliteral router.
    This has been changed so that if the ip literal address matches
    ignore_target_hosts, the router declines.

 7. Added expansion conditions match_domain, match_address, and match_local_
    part (NOT match_host).

 8. The placeholder for the Received: header didn't have a length field set.

 9. Added code to Exim itself and to exim_lock to test for a specific race
    condition that could lead to file corruption when using MBX delivery. The
    issue is with the lockfile that is created in /tmp. If this file is removed
    after a process has opened it but before that process has acquired a lock,
    there is the potential for a second process to recreate the file and also
    acquire a lock. This could lead to two Exim processes writing to the file
    at the same time. The added code performs the same test as UW imapd; it
    checks after acquiring the lock that its file descriptor still refers to
    the same named file.

10. The buffer for building added header lines was of fixed size, 8192 bytes.
    It is now parameterized by HEADER_ADD_BUFFER_SIZE and this can be adjusted
    when Exim is built.

11. Added the smtp_active_hostname option. If used, this will typically be made
    to depend on the incoming interface address. Because $interface_address is
    not set up until the daemon has forked a reception process, error responses
    that can happen earlier (such as "too many connections") no longer contain
    a host name.

12. If an expansion in a condition on a "warn" statement fails because a lookup
    defers, the "warn" statement is abandoned, and the next ACL statement is
    processed. Previously this caused the whole ACL to be aborted.

13. Added the iplsearch lookup type.

14. Added ident_timeout as a log selector.

15. Added tls_certificate_verified as a log selector.

16. Added a global option tls_require_ciphers (compare the smtp transport
    option of the same name). This controls incoming TLS connections.

17. I finally figured out how to make tls_require_ciphers do a similar thing
    in GNUtls to what it does in OpenSSL, that is, set up an appropriate list
    before starting the TLS session.

18. Tabs are now shown as \t in -bP output.

19. If the log selector return_path_on_delivery was set, Exim crashed when
    bouncing a message because it had too many Received: header lines.

20. If two routers both had headers_remove settings, and the first one included
    a superfluous trailing colon, the final name in the first list and the
    first name in the second list were incorrectly joined into one item (with a
    colon in the middle).


Exim version 4.32
-----------------

 1. Added -C and -D options to the exinext utility, mainly to make it easier
    to include in the automated testing, but these could be helpful when
    multiple configurations are in use.

 2. The exinext utility was not formatting the output nicely when there was
    an alternate port involved in the retry record key, nor when there was a
    message id as well (for retries that were specific to a specific message
    and a specific host). It was also confused by IPv6 addresses, because of
    the additional colons they contain. I have fixed the IPv4 problem, and
    patched it up to do a reasonable job for IPv6.

 3. When there is an error after a MAIL, RCPT, or DATA SMTP command during
    delivery, the log line now contains "pipelined" if PIPELINING was used.

 4. An SMTP transport process used to panic and die if the bind() call to set
    an explicit outgoing interface failed. This has been changed; it is now
    treated in the same way as a connect() failure.

 5. A reference to $sender_host_name in the part of a conditional expansion
    that was being skipped was still causing a DNS lookup. This no longer
    occurs.

 6. The def: expansion condition was not recognizing references to header lines
    that used bh_ and bheader_.

 7. Added the _cache feature to named lists.

 8. The code for checking quota_filecount in the appendfile transport was
    allowing one more file than it should have been.

 9. For compatibility with Sendmail, the command line option

      -prval:sval

    is equivalent to

      -oMr rval -oMs sval

    and sets the incoming protocol and host name (for trusted callers). The
    host name and its colon can be omitted when only the protocol is to be set.
    Note the Exim already has two private options, -pd and -ps, that refer to
    embedded Perl. It is therefore impossible to set a protocol value of "d" or
    "s", but I don't think that's a major issue.

10. A number of refactoring changes to the code, none of which should affect
    Exim's behaviour:

    (a) The number of logging options was getting close to filling up the
    32-bit word that was used as a bit map. I have split them into two classes:
    those that are passed in the argument to log_write(), and those that are
    only ever tested independently outside of that function. These are now in
    separate 32-bit words, so there is plenty of room for expansion again.
    There is no change in the user interface or the logging behaviour.

    (b) When building, for example, log lines, the code previously used a
    macro that called string_cat() twice, in order to add two strings. This is
    not really sufficiently general. Furthermore, there was one instance where
    it was actually wrong because one of the argument was used twice, and in
    one call a function was used. (As it happened, calling the function twice
    did not affect the overall behaviour.) The macro has been replaced by a
    function that can join an arbitrary number of extra strings onto a growing
    string.

    (c) The code for expansion conditions now uses a table and a binary chop
    instead of a serial search (which was left over from when there were very
    few conditions). Also, it now recognizes conditions like "pam" even when
    the relevant support is not compiled in: a suitably worded error message is
    given if an attempt is made to use such a condition.

11. Added ${time_interval:xxxxx}.

12. A bug was causing one of the ddress fields not to be passed back correctly
    from remote delivery subprocesses. The field in question was not being
    subsequently used, so this caused to problems in practice.

13. Added new log selectors queue_time and deliver_time.

14. Might have fixed a bug in maildirsizefile handling that threw up
    "unexpected character" debug warnings, and recalculated the data
    unnecessarily. In any case, I expanded the warning message to give more
    information.

15. Added the message "Restricted characters in address" to the statements in
    the default ACL that block characters like @ and % in local parts.

16. Change 71 for release 4.31 proved to be much less benign that I imagined.
    Three changes have been made:

    (a) There was a serious bug; a negative response to MAIL caused the whole
        recipient domain to be cached as invalid, thereby blocking all messages
        to all local parts at the same domain, from all senders. This bug has
        been fixed. The domain is no longer cached after a negative response to
        MAIL if the sender used is not empty.

    (b) The default behaviour of using MAIL FROM:<> for recipient callouts has
        been restored.

    (c) A new callout option, "use_sender" has been added for people who want
        the modified behaviour.


Exim version 4.31
-----------------

 1. Removed "EXTRALIBS=-lwrap" from OS/Makefile-Unixware7 on the advice of
    Larry Rosenman.

 2. Removed "LIBS = -lresolv" from OS/Makefile-Darwin as it is not needed, and
    indeed breaks things for older releases.

 3. Added additional logging to the case where there is a problem reading data
    from a filter that is running in a subprocess using a pipe, in order to
    try to track down a specific problem.

 4. Testing facility fudge: when running in the test harness and attempting
    to connect to 10.x.x.x (expecting a connection timeout) I'm now sometimes
    getting "No route to host". Convert this to a timeout.

 5. Define ICONV_ARG2_TYPE as "char **" for Unixware7 to avoid compiler
    warning.

 6. Some OS don't have socklen_t but use size_t instead. This affects the
    fifth argument of getsockopt() amongst other things. This is now
    configurable by a macro called SOCKLEN_T which defaults to socklen_t, but
    can be set for individual OS. I have set it for SunOS5, OSF1, and
    Unixware7. Current versions of SunOS5 (aka Solaris) do have socklen_t, but
    some earlier ones do not.

 7. Change 4.30/15 was not doing the test caselessly.

 8. The standard form for an IPv6 address literal was being rejected by address
    parsing in, for example, MAIL and RCPT commands. An example of this kind of
    address is [IPv6:2002:c1ed:8229:10:202:2dff:fe07:a42a]. Exim now accepts
    this, as well as the form without the "IPv6" on the front (but only when
    address literals are enabled, of course).

 9. Added some casts to avoid compiler warnings in OS/os.c-Linux.

10. Exim crashed if a message with an empty sender address specified by -f
    encountered a router with an errors_to setting. This could be provoked only
    by a command such as

      exim -f "" ...

    where an empty string was supplied; "<>" did not hit this bug.

11. Installed PCRE release 4.5.

12. If EHLO/HELO was rejected by an ACL, the value of $sender_helo_name
    remained set. It is now erased.

13. exiqgrep wasn't working on MacOS X because it didn't correctly compute
    times from message ids (which are base 36 rather than the normal 62).

14. "Expected" SMTP protocol errors that can arise when PIPELINING is in use
    were being counted as actual protocol errors, and logged if the log
    selector +smtp_protocol_error was set. One cannot be perfect in this test,
    but now, if PIPELINING has been advertised, RCPT following a rejected MAIL,
    and DATA following a set of rejected RCPTs do not count as protocol errors.
    In other words, Exim assumes they were pipelined, though this may not
    actually be the case. Of course, in all cases the client gets an
    appropriate error code.

15. If a lookup fails in an ACL condition, a message about the failure may
    be available; it is used if testing the ACL cannot continue, because most
    such messages specify what the cause of the deferral is. However, some
    messages (e.g. "MYSQL: no data found") do not cause a defer. There was bug
    that caused an old message to be retained and used if a later statement
    caused a defer, replacing the real cause of the deferral.

16. If an IP address had so many PTR records that the DNS lookup buffer
    was not large enough to hold them, Exim could crash while trying to process
    the truncated data. It now detects and logs this case.

17. Further to 4.21/58, another change has been made: if (and only if) the
    first line of a message (the first header line) ends with CRLF, a bare LF
    in a subsequent header line has a space inserted after it, so as not to
    terminate the header.

18. Refactoring: tidied an ugly bit of code in appendfile that copied data
    unnecessarily, used atoi() instead of strtol(), and didn't check the
    termination when getting file sizes from file names by regex.

19. Completely re-implemented the support for maildirsize files, in the light
    of a number of problems with the previous contributed implementation
    (4.30/29). In particular:

    . If the quota is zero, the maildirsize file is maintained, but no quota is
      imposed.

    . If the maildir directory does not exist, it is created before any attempt
      to write a maildirsize file.

    . The quota value in the file is just a cache; if the quota is changed in
      the transport, the new value overrides.

    . A regular expression is available for excluding directories from the
      count.

20. The autoreply transport checks the characters in options that define the
    message's headers; it allows continued headers, but it was checking with
    isspace() after an embedded newline instead of explicitly looking for a
    space or a tab.

21. If all the "regular" hosts to which an address was routed had passed their
    expiry times, and had not reached their retry times, the address was
    bounced, even if fallback hosts were defined. Now Exim should go on to try
    the fallback hosts.

22. Increased buffer sizes in the callout code from 1024 to 4096 to match the
    equivalent code in the SMTP transport. Some hosts send humungous responses
    to HELO/EHLO, more than 1024 it seems.

23. Refactoring: code in filter.c used (void *) for "any old type" but this
    gives compiler warnings in some environments. I've now done it "properly",
    using a union.

24. The replacement for inet_ntoa() that is used with gcc on IRIX systems
    (because of problems with the built-in one) was declared to return uschar *
    instead of char *, causing compiler failure.

25. Fixed a file descriptor leak when processing alias/forward files.

26. Fixed a minor format string issue in dbfn.c.

27. Typo in exim.c: ("dmbnz" for "dbmnz").

28. If a filter file refered to $h_xxx or $message_headers, and the headers
    contained RFC 2047 "words", Exim's memory could, under certain conditions,
    become corrupted.

29. When a sender address is verified, it is cached, to save repeating the test
    when there is more than one recipient in a message. However, when the
    verification involves a callout, it is possible for different callout
    options to be set for different recipients. It is too complicated to keep
    track of this in the cache, so now Exim always runs a verification when a
    callout is required, relying on the callout cache for the optimization.
    The overhead is duplication of the address routing, but this should not be
    too great.

30. Fixed a bug in callout caching. If a RCPT command caused the sender address
    to be verified with callout=postmaster, and the main callout worked but the
    postmaster check failed, the verification correctly failed. However, if a
    subsequent RCPT command asked for sender verification *without* the
    postmaster check, incorrect caching caused this verification also to fail,
    incorrectly.

31. Exim caches DNS lookup failures so as to avoid multiple timeouts; however,
    it was not caching the DNS options (qualify_single, search_parents) that
    were used when the lookup failed. A subsequent lookup with different
    options therefore always gave the same answer, though there were cases
    where it should not have. (Example: a "domains = !$mx_any" option on a
    dnslookup router: the "domains" option is always processed without any
    widening, but the router might have qualify_single set.) Now Exim uses the
    cached value only when the same options are set.

32. Added John Jetmore's "exipick" utility to the distribution.

33. GnuTLS: When an attempt to start a TLS session fails for any reason other
    than a timeout (e.g. a certificate is required, and is not provided), an
    Exim server now closes the connection immediately. Previously it waited for
    the client to close - but if the client is SSL, it seems that they each
    wait for each other, leading to a delay before one of them times out.

34: GnuTLS: Updated the code to use the new GnuTLS 1.0.0 API. I have not
    maintained 0.8.x compatibility because I don't think many are using it, and
    it is clearly obsolete.

35. Added TLS support for CRLs: a tls_crl global option and one for the smtp
    transport.

36. OpenSSL: $tls_certificate_verified was being set to 1 even if the
    client certificate was expired. A simple patch fixes this, though I don't
    understand the full logic of why the verify callback is called multiple
    times.

37. OpenSSL: a patch from Robert Roselius: "Enable client-bug workaround.
    Versions of OpenSSL as of 0.9.6d include a 'CBC countermeasure' feature,
    which causes problems with some clients (such as the Certicom SSL Plus
    library used by Eudora). This option, SSL_OP_DONT_INSERT_EMPTY_FRAGMENTS,
    disables the coutermeasure allowing Eudora to connect."

38. Exim was not checking that a write() to a log file succeeded. This could
    lead to Bad Things if a log got too big, in particular if it hit a file
    size limit. Exim now panics and dies if it cannot write to a log file, just
    as it does if it cannot open a log file.

39. Modified OS/Makefile-Linux so that it now contains

      CFLAGS=-O -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE

    The two -D definitions ensure that Exim is compiled with large file
    support, which makes it possible to handle log files that are bigger than
    2^31.

40. Fixed a subtle caching bug: if (in an ACL or a set of routers, for
    instance) a domain was checked against a named list that involved a lookup,
    causing $domain_data to be set, then another domain was checked against the
    same list, then the first domain was re-checked, the value of $domain_data
    after the final check could be wrong. In particular, if the second check
    failed, it could be set empty. This bug probably also applied to
    $localpart_data.

41. The strip_trailing_dot option was not being applied to the address given
    with the -f command-line option.

42. The code for reading a message's header from the spool was incrementing
    $received_count, but never initializing it. This meant that the value was
    incorrect (doubled) while delivering a message in the same process in which
    it was received. In the most common configuration of Exim, this never
    happens - a fresh exec is done - but it can happen when
    deliver_drop_privilege is set.

43. When Exim logs an SMTP synchronization error - client data sent too soon -
    it now includes up to 150 characters of the unexpected data in the log
    line.

44. The exim_dbmbuild utility uses fixed size buffers for reading input lines
    and building data strings. The size of both of these buffers was 10 000
    bytes - far larger than anybody would *ever* want, thought I. Needless to
    say, somebody hit the limit. I have increased the maximum line length to
    20 000 and the maximum data length of concatenated lines to 100 000. I have
    also fixed two bugs, because there was no checking on these buffers. Tsk,
    tsk. Now exim_dbmbuild gives a message and exits with an error code if a
    buffer is too small.

45. The exim_dbmbuild utility did not support quoted keys, as Exim does in
    lsearch lookups. Now it does.

46. When parsing a route_list item in a manualroute router, a fixed-length
    buffer was used for the list of hosts. I made this 1024 bytes long,
    thinking that nobody would ever have a list of hosts that long. Wrong.
    Somebody had a whole pile of complicated expansion conditions, and the
    string was silently truncated, leading to an expansion error. It turns out
    that it is easier to change to an unlimited length (owing to other changes
    that have happened since this code was originally written) than to build
    structure for giving a limitation error. The length of the item that
    expands into the list of hosts is now unlimited.

47. The lsearch lookup could not handle data where the length of text line was
    more than 4095 characters. Such lines were truncated, leading to shortened
    data being returned. It should now handle lines of any length.

48. Minor wording revision: "cannot test xxx in yyy ACL" becomes "cannot test
    xxx condition in yyy ACL" (e.g. "cannot test domains condition in DATA
    ACL").

49. Cosmetic tidy to scripts like exicyclog that are generated by globally
    replacing strings such as BIN_DIRECTORY in a source file: the replacement
    no longer happens in comment lines. A list of replacements is now placed
    at the head of all of the source files, except those whose only change is
    to replace PERL_COMMAND in the very first #! line.

50. Replaced the slow insertion sort in queue.c, for sorting the list of
    messages on the queue, with a bottom-up merge sort, using code contributed
    by Michael Haardt. This should make operations like -bp somewhat faster on
    large queues. It won't affect queue runners, except when queue_run_in_order
    is set.

51. Installed eximstats 1.31 in the distribution.

52. Added support for SRV lookups to the dnslookup router.

53. If an ACL referred to $message_body or $message_body_end, the value was not
    reset for any messages that followed in the same SMTP session.

54. The store-handling optimization for building very long strings was not
    differentiating between the different store pools. I don't think this
    actually made any difference in practice, but I've tidied it.

55. While running the routers to verify a sender address, $sender_address
    was still set to the sender address. This is wrong, because when routing to
    send a bounce to the sender, it would be empty. Therefore, I have changed
    it so that, while verifying a sender address, $sender_address is set to <>.
    (There is no change to what happens when verifying a recipient address.)

56. After finding MX (or SRV) records, Exim was doing a DNS lookup for the
    target A or AAAA records (if not already returned) without resetting the
    qualify_single or search_parents options of the DNS resolver. These are
    inappropriate in this case because the targets of MX and SRV records must
    be FQDNs. A broken DNS record could cause trouble if it happened to have a
    target that, when qualified, matched something in the local domain. These
    two options are now turned off when doing these lookups.

57. It seems that at least some releases of Reiserfs (which does not have the
    concept of a fixed number of inodes) returns zero and not -1 for the
    number of available inodes. This interacted badly with check_spool_inodes,
    which assumed that -1 was the "no such thing" setting. What I have done is
    to check that the total number of inodes is greater than zero before doing
    the test of how many are available.

58. When a "warn" ACL statement has a log_message modifier, the message is
    remembered, and not repeated. This is to avoid a lot of repetition when a
    message has many recipients that cause the same warning to be written.
    Howewer, Exim was preserving the list of already written lines for an
    entire SMTP session, which doesn't seem right. The memory is now reset if a
    new message is started.

59. The "rewrite" debugging flag was not showing the result of rewriting in the
    debugging output unless log_rewrite was also set.

60. Avoid a compiler warning on 64-bit systems in dsearch.c by avoiding the use
    of (int)(handle) when we know that handle contains (void *)(-1).

61. The Exim daemon panic-logs an error return when it closes the incoming
    connection. However "connection reset by peer" seems to be common, and
    isn't really an error worthy of noting specially, so that particular error
    is no long logged.

62. When Exim is trying to find all the local interfaces, it used to panic and
    die if the ioctl to get the interface flags failed. However, it seems that
    on at least one OS (Solaris 9) it is possible to have an interface that is
    included in the list of interfaces, but for which you get a failure error
    for this call. This happens when the interface is not "plumbed" into a
    protocol (i.e. neither IPv4 nor IPv6). I've changed the code so that a
    failure of the "get flags" call assumes that the interface is down.

63. Added a ${eval10: operator, which assumes all numbers are decimal. This
    makes life easier for people who are doing arithmetic on fields extracted
    from dates, where you often get leading zeros that should not be
    interpreted as octal.

64. Added qualify_domain to the redirect router, to override the global
    setting.

65. If a pathologically long header line contained very many addresses (the
    report of this problem mentioned 10 000) and each of them was rewritten,
    Exim could use up a very large amount of memory. (It kept on making new
    copies of the header line as it rewrote, and never released the old ones.)
    At the expense of a bit more processing, the header rewriting function has
    been changed so that it no longer eats memory in this way.

66. The generation of the Received: header has been moved from the time that a
    message starts to be received, to the time that it finishes. The timestamp
    in the Received: header should now be very close to that of the <= log
    line. There are two side-effects of this change:

    (a) If a message is rejected by a DATA or non-SMTP ACL or local_scan(), the
        logged header lines no longer include the local Received: line, because
        it has not yet been created. The same applies to a copy of the message
        that is returned to a non-SMTP sender when a message is rejected.

    (b) When a filter file is tested using -bf, no additional Received: header
        is added to the test message. After some thought, I decided that this
        is a bug fix.

    This change does not affect the value of $received_for. It is still set
    after address rewriting, but before local_scan() is called.

67. Installed the latest Cygwin-specific files from the Cygwin maintainer.

68. GnuTLS: If an empty file is specified for tls_verify_certificates, GnuTLS
    gave an unhelpful panic error message, and a defer error. I have managed to
    change this behaviour so that it now rejects any supplied certificate,
    which seems right, as the list of acceptable certificates is empty.

69. OpenSSL: If an empty file is specified for tls_verify_certificates, OpenSSL
    gave an unhelpful defer error. I have not managed to make this reject any
    supplied certificates, but the error message it gives is "no certificate
    supplied", which is not helpful.

70. exigrep's output now also includes lines that are not associated with any
    message, but which match the given pattern. Implemented by a patch from
    Martin Sluka, which also tidied up the Perl a bit.

71. Recipient callout verification, like sender verification, was using <> in
    the MAIL FROM command. This isn't really the right thing, since the actual
    sender may affect whether the remote host accepts the recipient or not. I
    have changed it to use the actual sender in the callout; this means that
    the cache record is now keyed on a recipient/sender pair, not just the
    recipient address. There doesn't seem to be a real danger of callout loops,
    since a callout by the remote host to check the sender would use <>.
    [SEE ABOVE: changed after hitting problems.]

72. Exim treats illegal SMTP error codes that do not begin with 4 or 5 as
    temporary errors. However, in the case of such a code being given after
    the end of a data transmission (i.e. after ".") Exim was failing to write
    a retry record for the message. (Yes, there was some broken host that was
    actually sending 8xx at this point.)

73. An unknown lookup type in a host list could cause Exim to panic-die when
    the list was checked. (An example that provoked this was putting <; in the
    middle of a list instead of at the start.) If this happened during a DATA
    ACL check, a -D file could be left lying around. This kind of configuration
    error no longer causes Exim to die; instead it causes a defer errror. The
    incident is still logged to the main and panic logs.

74. Buglet left over from Exim 3 conversion. The message "too many messages
    in one connection" was written to the rejectlog but not the mainlog, except
    when address rewriting (yes!) was being logged.

75. Added write_rejectlog option.

76. When a system filter was run not as root (that is, when system_filter_user
    was set), the values of the $n variables were not being returned to the
    main process; thus, they were not subsequently available in the $sn
    variables.

77. Added +return_path_on_delivery log selector.

78. A connection timeout was being treated differently from recipients deferred
    when testing hosts_max_try with a message that was older than the host's
    retry timeout. (The host should not be counted, thus allowing all hosts to
    be tried at least once before bouncing.) This may have been the cause of an
    occasionally reported bug whereby a message would remain on the queue
    longer than the retry timeout, but would be bounced if a delivery was
    forced. I say "may" because I never totally pinned down the problem;
    setting up timeout/retry tests is difficult. See also the next item.

79. The ultimate address timeout was not being applied to errors that involved
    a combination of host plus message (for example, a timeout on a MAIL
    command). When an address resolved to a number of possible hosts, and they
    were not all tried for each delivery (e.g. because of hosts_max_try), a
    message could remain on the queue longer than the retry timeout.

80. Sieve bug: "stop" inside "elsif" was broken. Applied a patch from Michael
    Haardt.

81. Fixed an obscure SMTP outgoing bug which required at least the following
    conditions: (a) there was another message waiting for the same server;
    (b) the server returned 5xx to all RCPT commands in the first message so
    that the message was not completed; (c) the server dropped the connection
    or gave a negative response to the RSET that Exim sends to abort the
    transaction. The observed case was a dropped connection after DATA that had
    been sent in pipelining mode. That is, the server had advertised PIPELINING
    but was not implementing it correctly. The effect of the bug was incorrect
    behaviour, such as trying another host, and this could lead to a crash.


Exim version 4.30
-----------------

 1. The 3rd arguments to getsockname(), getpeername(), and accept() in exim.c
    and daemon.c were passed as pointers to ints; they should have been
    pointers to socklen_t variables (which are typically unsigned ints).

 2. Some signed/unsigned type warnings in the os.c file for Linux have been
    fixed.

 3. Fixed a really odd bug that affected only the testing scheme; patching a
    certain fixed string in the binary changed the value of another string that
    happened to be identical to the end of the original first string.

 4. When gethostbyname() (or equivalent) is passed an IP address as a "host
    name", it returns that address as the IP address. On some operating
    systems (e.g. Solaris), it also passes back the IP address string as the
    "host name". However, on others (e.g. Linux), it passes back an empty
    string. Exim wasn't checking for this, and was changing the host name to an
    empty string, assuming it had been canonicized.

 5. Although rare, it is permitted to have more than one PTR record for a given
    IP address. I thought that gethostbyaddr() or getipnodebyaddr() always gave
    all the names associated with an address, because they do in Solaris.
    However, it seems that they do not in Linux for data that comes from the
    DNS. If an address in /etc/hosts has multiple names, they _are_ all given.
    I found this out when I moved to a new Linux workstation and tried to run
    the Exim test suite.

    To get round this problem I have changed the code so that it now does its
    own call to the DNS to look up PTR records when searching for a host name.
    If nothing can be found in the DNS, it tries gethostbyaddr(), so that
    addresses that are only in /etc/hosts are still found.

    This behaviour is, however, controlled by an option called host_lookup_
    order, which defaults to "bydns:byaddr". If people want to use the other
    order, or indeed, just use one or the other means of lookup, they can
    specify it in this variable.

 6. If a PTR record yields an empty name, Exim treats it as non-existent. In
    some operating systems, this comes back from gethostbyaddr() as an empty
    string, and this is what Exim used to test for. However, it seems that in
    other systems, "." is yielded. Exim now tests for this case too.

 7. The values of check_spool_space and check_log_space are now held internally
    as a number of kilobytes instead of an absolute number of bytes. If a
    numbers is specified without 'K' or 'M', it is rounded up to the nearest
    kilobyte. This means that much larger values can be stored.

 8. Exim monitor: an attempt to get the action menu when not actually pointing
    at a message produces an empty menu entitled "No message selected". This
    works on Solaris (OpenWindows). However, XFree86 does not like a menu with
    no entries in it ("Shell widget menu has zero width and/or height"). So I
    have added a single, blank menu entry in this case.

 9. Added ${quote_local_part.

10. MIME decoding is now applied to the contents of Subject: header lines when
    they are logged.

11. Now that a reference to $sender_host_address automatically causes a reverse
    lookup to occur if necessary (4.13/18), there is no need to arrange for a
    host lookup before query-style lookups in lists that might use this
    variable. This has therefore been abolished, and the "net-" prefix is no
    longer necessary for query-style lookups.

12. The Makefile for SCO_SV contained a setting of LDFLAGS. This appears to
    have been a typo for LFLAGS, so it has been changed.

13. The install script calls Exim with "-C /dev/null" in order to find the
    version number. If ALT_CONFIG_PREFIX was set, this caused an error message
    to be output. Howeve, since Exim outputs its version number before the
    error, it didn't break the script. It just looked ugly. I fixed this by
    always allowing "-C /dev/null" if the caller is root.

14. Ignore overlarge ACL variable number when reading spool file - insurance
    against a later release with more variables having written the file.

15. The standard form for an IPv6 address literal was being rejected by EHLO.
    Example: [IPv6:2002:c1ed:8229:10:202:2dff:fe07:a42a]. Exim now accepts
    this, as well as the form without the "IPv6" on the front.

16. Added CHOWN_COMMAND=/usr/sbin/chown and LIBS=-lresolv to the
    OS/Makefile-Darwin file.

17. Fixed typo in lookups/ldap.c: D_LOOKUP should be D_lookup. This applied
    only to LDAP libraries that do not have LDAP_OPT_DEREF.

18. After change 4.21/52, "%ld" was used to format the contents of the $inode
    variable. However, some OS use ints for inodes. I've added cast to long int
    to get rid of the compiler warning.

19. I had forgotten to lock out "/../" in configuration file names when
    ALT_CONFIG_PREFIX was set.

20. Routers used for verification do not need to specify transports. However,
    if such a router generated a host list, and callout was configured, Exim
    crashed, because it could not find a port number from the (non-existent)
    transport. It now assumes port 25 in this circumstance.

21. Added the -t option to exigrep.

22. If LOOKUP_LSEARCH is defined, all three linear search methods (lsearch,
    wildlsearch, nwildlsearch) are compiled. LOOKUP_WILDLSEARCH and LOOKUP_
    NWILDLSEARCH are now obsolete, but retained for compatibility. If either of
    them is set, LOOKUP_LSEARCH is forced.

23. "exim -bV" now outputs a list of lookups that are included in the binary.

24. Added sender and host information to the "rejected by local_scan()" log
    line; previously there was no indication of these.

25. Added .include_if_exists.

26. Change 3.952/11 added an explicit directory sync on top of a file sync for
    Linux. It turns out that not all file systems support this. Apparently some
    versions of NFS do not. (It's rare to put Exim's spool on NFS, but people
    do it.) To cope with this, the error EINVAL, which means that sync-ing is
    not supported on the file descriptor, is now ignored when Exim is trying to
    sync a directory. This applies only to Linux.

27. Added -DBIND_8_COMPAT to the CLFAGS setting for Darwin.

28. In Darwin (MacOS X), the PAM headers are in /usr/include/pam and not in
    /usr/include/security. There's now a flag in OS/os.h-Darwin to cope with
    this.

29. Added support for maildirsize files from supplied patch (modified a bit).

30. The use of :fail: followed by an empty string could lead Exim to respond to
    sender verification failures with (e.g.):

      550 Verification failed for <xxx>
      550 Sender verify failed

    where the first response line was missing the '-' that indicates it is not
    the final line of the response.

31. The loop for finding the name of the user that called Exim had a hardwired
    limit of 10; it now uses the value of finduser_retries, which is used for
    all other user lookups.

32. Added $received_count variable, available in data and not_smtp ACLs, and at
    delivery time.

33. Exim was neglecting to zero errno before one call of strtol() when
    expanding a string and expecting an integer value. On some systems this
    resulted in spurious "integer overflow" errors. Also, it was casting the
    result into an int without checking.

34. Testing for a connection timeout using "timeout_connect" in the retry rules
    did not work. The code looks as if it has *never* worked, though it appears
    to have been documented since at least releast 1.62. I have made it work.

35. The "timeout_DNS" error in retry rules, also documented since at least
    1.62, also never worked. As it isn't clear exactly what this means, and
    clearly it isn't a major issue, I have abolished the feature by treating it
    as "timeout", and writing a warning to the main and panic logs.

36. The display of retry rules for -brt wasn't always showing the error code
    correctly.

37. Added new error conditions to retry rules: timeout_A, timeout_MX,
    timeout_connect_A, timeout_connect_MX.

38. Rewriting the envelope sender at SMTP time did not allow it to be rewritten
    to the empty sender.

39. The daemon was not analysing the content of -oX till after it had closed
    stderr and disconnected from the controlling terminal. This meant that any
    syntax errors were only noted on the panic log, and the return code from
    the command was 0. By re-arranging the code a little, I've made the
    decoding happen first, so such errors now appear on stderr, and the return
    code is 1. However, the actual setting up of the sockets still happens in
    the disconnected process, so errors there are still only recorded on the
    panic log.

40. A daemon listener on a wildcard IPv6 socket that also accepts IPv4
    connections (as happens on some IP stacks) was logged at start up time as
    just listening for IPv6. It now logs "IPv6 with IPv4". This differentiates
    it from "IPv6 and IPv4", which means that two separate sockets are being
    used.

41. The debug output for gethostbyname2() or getipnodebyname() failures now
    says whether AF_INET or AF_INET6 was passed as an argument.

42. Exiwhat output was messed up when time zones were included in log
    timestamps.

43. Exiwhat now gives more information about the daemon's listening ports,
    and whether -tls-on-connect was used.

44. The "port" option of the smtp transport is now expanded.

45. A "message" modifier in a "warn" statement in a non-message ACL was being
    silently ignored. Now an error message is written to the main and panic
    logs.

46. There's a new ACL modifier called "logwrite" which writes to a log file
    as soon as it is encountered.

47. Added $local_user_uid and $local_user_gid at routing time.

48. Exim crashed when trying to verify a sender address that was being
    rewritten to "<>".

49. Exim was recognizing only a space character after ".include". It now also
    recognizes a tab character.

50. Fixed several bugs in the Perl script that creates the exim.8 man page by
    extracting the relevant information from the specification. The man page no
    longer contains scrambled data for the -d option, and I've added a section
    at the front about calling Exim under different names.

51. Added "extra_headers" argument to the "mail" command in filter files.

52. Redirecting mail to an unqualified address in a Sieve filter caused Exim to
    crash.

53. Installed eximstats 1.29.

54. Added transport_filter_timeout as a generic transport option.

55. Exim no longer adds an empty Bcc: header to messages that have no To: or
    Cc: header lines. This was required by RFC 822, but it not required by RFC
    2822.

56. Exim used to add From:, Date:, and Message-Id: header lines to any
    incoming messages that did not have them. Now it does so only if the
    message originates locally, that is, if there is no associated remote host
    address. When Resent- header lines are present, this applies to the Resent-
    lines rather than the non-Resent- lines.

57. Drop incoming SMTP connection after too many syntax or protocol errors. The
    limit is controlled by smtp_max_synprot_errors, defaulting to 3.

58. Messages for configuration errors now include the name of the main
    configuration file - useful now that there may be more than one file in a
    list (.included file names were always shown).

59. Change 4.21/82 (run initgroups() when starting the daemon) causes problems
    for those rare installations that do not start the daemon as root or run it
    setuid root. I've cut out the call to initgroups() if the daemon is not
    root at that time.

60. The Exim user and group can now be bound into the binary as text strings
    that are looked up at the start of Exim's processing.

61. Applied a small patch for the Interbase code, supplied by Ard Biesheuvel.

62. Added $mailstore_basename variable.

63. Installed patch to sieve.c from Michael Haardt.

64. When Exim failed to open the panic log after failing to open the main log,
    the original message it was trying to log was written to stderr and debug
    output, but if they were not available (the usual case in production), it
    was lost. Now it is written to syslog before the two lines that record the
    failures to open the logs.

65. Users' Exim filters run in subprocesses under the user's uid. It is
    possible for a "deliver" command or an alias in a "personal" command to
    provoke an address rewrite. If logging of address rewriting is configured,
    this fails because the process is not running as root or exim. There may be
    a better way of dealing with this, but for the moment (because 4.30 needs
    to be released), I have disabled address rewrite logging when running a
    filter in a non-root, non-exim process.


Exim version 4.24
-----------------

 1. The buildconfig auxiliary program wasn't quoting the value set for
    HEADERS_CHARSET. This caused a compilation error complaining that 'ISO' was
    not defined. This bug was masked in 4.22 by the effect that was fixed in
    change 4.23/1.

 2. Some messages that were rejected after a message id was allocated were
    shown as "incomplete" by exigrep. It no longer does this for messages that
    are rejected by local_scan() or the DATA or non-SMTP ACLs.

 3. If a Message-ID: header used a domain literal in the ID, and Exim did not
    have allow_domain_literals set, the ID did not get logged in the <= line.
    Domain literals are now always recognized in Message-ID: header lines.

 4. The first argument for a ${extract expansion item is the key name or field
    number. Leading and trailing spaces in this item were not being ignored,
    causing some misleading effects.

 5. When deliver_drop_privilege was set, single queue runner processes started
    manually (i.e. by the command "exim -q") or by the daemon (which uses the
    same command in the process it spins off) were not dropping privilege.

 6. When the daemon running as "exim" started a queue runner, it always
    re-executed Exim in the spun-off process. This is a waste of effort when
    deliver_drop_privilege is set. The new process now just calls the
    queue-runner function directly.


Exim version 4.23
-----------------

 1. Typo in the src/EDITME file: it referred to HEADERS_DECODE_TO instead of
    HEADERS_CHARSET.

 2. Change 4.21/73 introduced a bug. The pid file path set by -oP was being
    ignored. Though the use of -oP was forcing the writing of a pid file, it
    was always written to the default place.

 3. If the message "no IP address found for host xxxx" is generated during
    incoming verification, it is now followed by identification of the incoming
    connection (so you can more easily find what provoked it).

 4. Bug fix for Sieve filters: "stop" inside a block was not working properly.

 5. Added some features to "harden" Exim a bit more against certain attacks:

    (a) There is now a build-time option called FIXED_NEVER_USERS that can
        be put in Local/Makefile. This is like the never_users runtime option,
        but it cannot be overridden. The default setting is "root".

    (b) If ALT_CONFIG_PREFIX is defined in Local/Makefile, it specifies a
        prefix string with which any file named in a -C command line option
        must start.

    (c) If ALT_CONFIG_ROOT_ONLY is defined in Local/Makefile, root privilege
        is retained for -C and -D only if the caller of Exim is root. Without
        it, the exim user may also use -C and -D and retain privilege.

    (d) If DISABLE_D_OPTION is defined in Local/Makefile, the use of the -D
        command line option is disabled.

 6. Macro names set by the -D option must start with an upper case letter, just
    like macro names defined in the configuration file.

 7. Added "dereference=" facility to LDAP.

 8. Two instances of the typo "uknown" in the source files are fixed.

 9. If a PERL_COMMAND setting in Local/Makefile was not at the start of a line,
    the Configure-Makefile script screwed up while processing it.

10. Incorporated PCRE 4.4.

11. The SMTP synchronization check was not operating right at the start of an
    SMTP session. For example, it could not catch a HELO sent before the client
    waited for the greeting. There is now a check for outstanding input at the
    point when the greeting is written. Because of the duplex, asynchronous
    nature of TCP/IP, it cannot be perfect - the incorrect input may be on its
    way, but not yet received, when the check is performed.

12. Added tcp_nodelay to make it possible to turn of the setting of TCP_NODELAY
    on TCP/IP sockets, because this apparently causes some broken clients to
    timeout.

13. Installed revised OS/Makefile-CYGWIN and OS/os.c-cygwin (the .h file was
    unchanged) from the Cygwin maintainer.

14. The code for -bV that shows what is in the binary showed "mbx" when maildir
    was supported instead of testing for mbx. Effectively a typo.

15. The spa authenticator server code was not checking that the input it
    received was valid base64.

16. The debug output line for the "set" modifier in ACLs was not showing the
    name of the variable that was being set.

17. Code tidy: the variable type "vtype_string" was never used. Removed it.

18. Previously, a reference to $sender_host_name did not cause a DNS reverse
    lookup on its own. Something else was needed to trigger the lookup. For
    example, a match in host_lookup or the need for a host name in a host list.
    Now, if $sender_host_name is referenced and the host name has not yet been
    looked up, a lookup is performed. If the lookup fails, the variable remains
    empty, and $host_lookup_failed is set to "1".

19. Added "eqi" as a case-independent comparison operator.

20. The saslauthd authentication condition could segfault if neither service
    nor realm was specified.

21. If an overflowing value such as "2048M" was set for message_size_limit, the
    error message that was logged was misleading, and incoming SMTP
    connections were dropped. The message is now more accurate, and temporary
    errors are given to SMTP connections.

22. In some error situations (such as 21 above) Exim rejects all SMTP commands
    (except RSET) with a 421 error, until QUIT is received. However, it was
    failing to send a response to QUIT.

23. The HELO ACL was being run before the code for helo_try_verify_hosts,
    which made it impossible to use "verify = helo" in the HELO ACL. The HELO
    ACL is now run after the helo_try_verify_hosts code.

24. "{MD5}" and "{SHA1}" are now recognized as equivalent to "{md5"} and
    "{sha1}" in the "crypteq" expansion condition (in fact the comparison is
    case-independent, so other case variants are also recognized). Apparently
    some systems use these upper case variants.

25. If more than two messages were waiting for the same host, and a transport
    filter was specified for the transport, Exim sent two messages over the
    same TCP/IP connection, and then failed with "socket operation on non-
    socket" when it tried to send the third.

26. Added Exim::debug_write and Exim::log_write for embedded Perl use.

27. The extern definition of crypt16() in expand.c was not being excluded when
    the OS had its own crypt16() function.

28. Added bounce_return_body as a new option, and bounce_return_size_limit
    as a preferred synonym for return_size_limit, both as an option and as an
    expansion variable.

29. Added LIBS=-liconv to OS/Makefile-OSF1.

30. Changed the default configuration ACL to relax the local part checking rule
    for addresses that are not in any local domains. For these addresses,
    slashes and pipe symbols are allowed within local parts, but the sequence
    /../ is explicitly forbidden.

31. SPA server authentication was not clearing the challenge buffer before
    using it.

32. log_message in a "warn" ACL statement was writing to the reject log as
    well as to the main log, which contradicts the documentation and doesn't
    seem right (because no rejection is happening). So I have stopped it.

33. Added Ard Biesheuvel's lookup code for accessing an Interbase database.
    However, I am unable to do any testing of this.

34. Fixed an infelicity in the appendfile transport. When checking directories
    for a mailbox, to see if any needed to be created, it was accidentally
    using path names with one or more superfluous leading slashes; tracing
    would show up entries such as stat("///home/ph10", 0xFFBEEA48).

35. If log_message is set on a "discard" verb in a MAIL or RCPT ACL, its
    contents are added to the log line that is written for every discarded
    recipient. (Previously a log_message setting was ignored.)

36. The ${quote: operator now quotes the string if it is empty.

37. The install script runs exim in order to find its version number. If for
    some reason other than non-existence or emptiness, which it checks, it
    could not run './exim', it was installing it with an empty version number,
    i.e. as "exim-". This error state is now caught, and the installation is
    aborted.

38. An argument was missing from the function that creates an error message
    when Exim fails to connect to the socket for saslauthd authentication.
    This could cause Exim to crash, or give a corrupted message.

39. Added isip, isip4, and isip6 to ${if conditions.

40. The ACL variables $acl_xx are now saved with the message, and can be
    accessed later in routers, transports, and filters.

41. The new lookup type nwildlsearch is like wildlsearch, except that the key
    strings in the file are not string-expanded.

42. If a MAIL command specified a SIZE value that was too large to fit into an
    int variable, the check against message_size_limit failed. Such values are
    now forced to INT_MAX, which is around 2Gb for a 32-bit variable. Maybe one
    day this will have to be increased, but I don't think I want to be around
    when emails are that large.



Exim version 4.22
-----------------

 1. Removed HAVE_ICONV=yes from OS/Makefile-FreeBSD, since it seems that
    iconv() is not standard in FreeBSD.

 2. Change 4.21/17 was buggy and could cause stack overwriting on a system with
    IPv6 enabled. The observed symptom was a segmentation fault on return from
    the function os_common_find_running_interfaces() in src/os.c.

 3. In the check_special_case() function in daemon.c I had used "errno" as an
    argument name, which causes warnings on some systems. This was basically a
    typo, since it was named "eno" in the comments!

 4. The code that waits for the clock to tick (at a resolution of some fraction
    of a second) so as to ensure message-id uniqueness was always waiting for
    at least one whole tick, when it could have waited for less. [This is
    almost certainly not relevant at current processor speeds, where it is
    unlikely to ever wait at all. But we try to future-proof.]

 5. The function that sleeps for a time interval that includes fractions of a
    second contained a race. It did not block SIGALRM between setting the
    timer, and suspending (a couple of lines later). If the interval was short
    and the sigsuspend() was delayed until after it had expired, the suspension
    never ended. On busy systems this could lead to processes getting stuck for
    ever.

 6. Some uncommon configurations may cause a lookup to happen in a queue runner
    process, before it forks any delivery processes. The open lookup caching
    mechanism meant that the open file or database connection was passed into
    the delivery process. The problem was that delivery processes always tidy
    up cached lookup data. This could cause a problem for the next delivery
    process started by the queue runner, because the external queue runner
    process does not know about the closure. So the next delivery process
    still has data in the lookup cache. In the case of a file lookup, there was
    no problem because closing a file descriptor in a subprocess doesn't affect
    the parent. However, if the lookup was caching a connection to a database,
    the connection was closed, and the second delivery process was likely to
    see errors such as "PGSQL: query failed: server closed the connection
    unexpectedly". The problem has been fixed by closing all cached lookups
    in a queue runner before running a delivery process.

 7. Compiler warning on Linux for the second argument of iconv(), which doesn't
    seem to have the "const" qualifier which it has on other OS. I've
    parameterised it.

 8. Change 4.21/2 was too strict. It is only if there are two authenticators
    *of the same type* (client or server) with the same public name that an
    error should be diagnosed.

 9. When Exim looked up a host name for an IP address, but failed to find the
    original IP address when looking up the host name (a safety check), it
    output the message "<ip address> does not match any IP for NULL", which was
    confusing, to say the least. The bug was that the host name should have
    appeared instead of "NULL".

10. Since release 3.03, if Exim is called by a uid other than root or the Exim
    user that is built into the binary, and the -C or -D options is used, root
    privilege is dropped before the configuration file is read. In addition,
    logging is switched to stderr instead of the normal log files. If the
    configuration then re-defines the Exim user, the unprivileged environment
    is probably not what is expected, so Exim logs a panic warning message (but
    proceeds).

    However, if deliver_drop_privilege is set, the unprivileged state may well
    be exactly what is intended, so the warning has been cut out in that case,
    and Exim is allowed to try to write to its normal log files.


Exim version 4.21
-----------------

 1. smtp_return_error_details was not giving details for temporary sender
    or receiver verification errors.

 2. Diagnose a configuration error if two authenticators have the same public
    name.

 3. Exim used not to create the message log file for a message until the first
    delivery attempt. This could be confusing when incoming messages were held
    for policy or load reasons. The message log file is now created at the time
    the message is received, and an initial "Received" line is written to it.

 4. The automatically generated man page for command line options had a minor
    bug that caused no ill effects; however, a more serious problem was that
    the procedure for building the man page automatically didn't always
    operate. Consequently, release 4.20 contains an out-of-date version. This
    shouldn't happen again.

 5. When building Exim with embedded Perl support, the script that builds the
    Makefile was calling 'perl' to find its compile-time parameters, ignoring
    any setting of PERL_COMMAND in Local/Makefile. This is now fixed.

 6. The freeze_tell option was not being used for messages that were frozen on
    arrival, either by an ACL or by local_scan().

 7. Added the smtp_incomplete_transaction log selector.

 8. After STARTTLS, Exim was not forgetting that it had advertised AUTH, so it
    was accepting AUTH without a new EHLO.

 9. Added tls_remember_esmtp to cope with YAEB. This allows AUTH and other
    ESMTP extensions after STARTTLS without a new EHLO, in contravention of the
    RFC.

10. Logging of TCP/IP connections (when configured) now happens in the main
    daemon process instead of the child process, so that the TCP/IP connection
    count is more accurate (but it can never be perfect).

11. The use of "drop" in a nested ACL was not being handled correctly in the
    outer ACL. Now, if condition failure induced by the nested "drop" causes
    the outer ACL verb to deny access ("accept" or "discard" after "endpass",
    or "require"), the connection is dropped.

12. Similarly, "discard" in a nested ACL wasn't being handled. A nested ACL
    that yield "discard" can now be used with an "accept" or a "discard" verb,
    but an error is generated for any others (because I can't see a useful way
    to define what should happen).

13. When an ACL is read dynamically from a file (or anywhere else), the lines
    are now processed in the same way as lines in the Exim configuration file.
    In particular, continuation lines are supported.

14. Added the "dnslists = a.b.c!=n.n.n.n" feature.

15. Added -ti meaning -t -i.

16. Check for letters, digits, hyphens, and dots in the names of dnslist
    domains, and warn by logging if others are found.

17. At least on BSD, alignment is not guarenteed for the array of ifreq's
    returned from GIFCONF when Exim is trying to find the list of interfaces on
    a host. The code in os.c has been modified to copy each ifreq to an aligned
    structure in all cases.

    Also, in some cases, the returned ifreq's were being copied to a 'struct
    ifreq' on the stack, which was subsequently passed to host_ntoa(). That
    means the last couple of bytes of an IPv6 address could be chopped if the
    ifreq contained only a normal sockaddr (14 bytes storage).

18. Named domain lists were not supported in the hosts_treat_as_local option.
    An entry such as +xxxx was not recognized, and was treated as a literal
    domain name.

19. Ensure that header lines added by a DATA ACL are included in the reject log
    if the ACL subsequently rejects the message.

20. Upgrade the cramtest.pl utility script to use Digest::MD5 instead of just
    MD5 (which is deprecated).

21. When testing a filter file using -bf, Exim was writing a message when it
    took the sender from a "From " line in the message, but it was not doing so
    when it took $return_path from a Return-Path: header line. It now does.

22. If the contents of a "message" modifier for a "warn" ACL verb do not begin
    with a valid header line field name (a series of printing characters
    terminated by a colon, Exim now inserts X-ACL-Warn: at the beginning.

23. Changed "disc" in the source to "disk" to conform to the documentation and
    the book and for uniformity.

24. Ignore Sendmail's -Ooption=value command line item.

25. When execve() failed while trying to run a command in a pipe transport,
    Exim was returning EX_UNAVAILBLE (69) from the subprocess. However, this
    could be confused with a return value of 69 from the command itself. This
    has been changed to 127, the value the shell returns if it is asked to run
    a non-existent command. The wording for the related log line suggests a
    non-existent command as the problem.

26. If received_header_text expands to an empty string, do not add a Received:
    header line to the message. (Well, it adds a token one on the spool, but
    marks it "old" so that it doesn't get used or transmitted.)

27. Installed eximstats 1.28 (addition of -nt option).

28. There was no check for failure on the call to getsockname() in the daemon
    code. This can fail if there is a shortage of resources on the system, with
    ENOMEM, for example. A temporary error is now given on failure.

29. Contrary to the C standard, it seems that in some environments, the
    equivalent of setlocale(LC_ALL, "C") is not obeyed at the start of a C
    program. Exim now does this explicitly; it affects the formatting of
    timestamps using strftime().

30. If exiqsumm was given junk data, it threw up some uninitialized variable
    complaints. I've now initialized all the variables, to avoid this.

32. Header lines added by a system filter were not being "seen" during
    transport-time rewrites.

33. The info_callback() function passed to OpenSSL is set up with type void
    (*)(SSL *, int, int), as described somewhere. However, when calling the
    function (actually a macro) that sets it up, the type void(*)() is
    expected. I've put in a cast to prevent warnings from picky compilers.

34. If a DNS black list lookup found a CNAME record, but there were no A
    records associated with the domain it pointed at, Exim crashed.

35. If a DNS black list lookup returned more than one A record, Exim ignored
    all but the first. It now scans all returned addresses if a particular IP
    value is being sought. In this situation, the contents of the
    $dnslist_value variable are a list of all the addresses, separated by a
    comma and a space.

36. Tightened up the rules for host name lookups using reverse DNS. Exim used
    to accept a host name and all its aliases if the forward lookup for any of
    them yielded the IP address of the incoming connection. Now it accepts only
    those names whose forward lookup yields the correct IP address. Any other
    names are discarded. This closes a loophole whereby a rogue DNS
    administrator could create reverse DNS records to break through a
    wildcarded host restriction in an ACL.

37. If a user filter or a system filter that ran in a subprocess used any of
    the numerical variables ($1, $2 etc), or $thisaddress, in a pipe command,
    the wrong values were passed to the pipe command ($thisaddress had the
    value of $0, $0 had the value of $1, etc). This bug was introduced by
    change 4.11/101, and not discovered because I wrote an inadequate test. :-(

38. Improved the line breaking for long SMTP error messages from ACLs.
    Previously, if there was no break point between 40 and 75 characters, Exim
    left the rest of the message alone. Two changes have been made: (a) I've
    reduced the minimum length to 35 characters; (b) if it can't find a break
    point between 35 and 75 characters, it looks ahead and uses the first one
    that it finds. This may give the occasional overlong line, but at least the
    remaining text gets split now.

39. Change 82 of 4.11 was unimaginative. It assumed the limit on the number of
    file descriptors might be low, and that setting 1000 would always raise it.
    It turns out that in some environments, the limit is already over 1000 and
    that lowering it causes trouble. So now Exim takes care not to decrease it.

40. When delivering a message, the value of $return_path is set to $sender_
    address at the start of routing (routers may change the value). By an
    oversight, this default was not being set up when an address was tested by
    -bt or -bv, which affected the outcome if any router or filter referred to
    $return_path.

41. The idea of the "warn" ACL verb is that it adds a header or writes to the
    log only when "message" or "log_message" are set. However, if one of the
    conditions was an address verification, or a call to a nested ACL, the
    messages generated by the underlying test were being passed through. This
    no longer happens. The underlying message is available in $acl_verify_
    message for both "message" and "log_message" expansions, so it can be
    passed through if needed.

42. Added RFC 2047 interpretation of header lines for $h_ expansions, with a
    new expansion $bh_ to give the encoded byte string without charset
    translation. Translation happens only if iconv() is available; HAVE_ICONV
    indicates this at build time. HEADERS_CHARSET gives the charset to
    translate to; headers_charset can change it in the configuration, and
    "headers charset" can change it in an individual filter file.

43. Now that we have a default RFC 2047 charset (see above), the code in Exim
    that creates RFC 2047 encoded "words" labels them as that charset instead
    of always using iso-8859-1. The cases are (i) the explicit ${rfc2047:
    expansion operator; (ii) when Exim creates a From: line for a local
    message; (iii) when a header line is rewritten to include a "phrase" part.

44. Nasty bug in exiqsumm: the regex to skip already-delivered addresses was
    buggy, causing it to skip the first lines of messages whose message ID
    ended in 'D'. This would not have bitten before Exim release 4.14, because
    message IDs were unlikely to end in 'D' before then. The effect was to have
    incorrect size information for certain domains.

45. #include "config.h" was missing at the start of the crypt16.c module. This
    caused trouble on Tru64 (aka OSF1) systems, because HAVE_CRYPT16 was not
    noticed.

46. If there was a timeout during a "random" callout check, Exim treated it as
    a failure of the random address, and carried on sending RSET and the real
    address. If the delay was just some slowness somewhere, the response to the
    original RCPT would be taken as a response to RSET and so on, causing
    mayhem of various kinds.

47. Change 50 for 4.20 was a heap of junk. I don't know what I was thinking
    when I implemented it. It didn't allow for the fact that some option values
    may legitimatetly be negative (e.g. size_addition), and it didn't even do
    the right test for positive values.

48. Domain names in DNS records are case-independent. Exim always looks them up
    in lower case. Some resolvers return domain names in exactly the case they
    appear in the zone file, that is, they may contain uppercase letters. Not
    all resolvers do this - some return always lower case. Exim was treating a
    change of case by a resolver as a change of domain, similar to a widening
    of a domain abbreviation. This triggered its re-routing code and so it was
    trying to route what was effectively the same domain again. This normally
    caused routing to fail (because the router wouldn't handle the domain
    twice). Now Exim checks for this case specially, and just changes the
    casing of the domain that it ultimately uses when it transmits the message
    envelope.

49. Added Sieve (RFC 3028) support, courtesy of Michael Haardt's contributed
    module.

50. If a filter generated a file delivery with a non-absolute name (possible if
    no home directory exists for the router), the forbid_file option was not
    forbidding it.

51. Added '&' feature to dnslists, to provide bit mask matching in addition to
    the existing equality matching.

52. Exim was using ints instead of ino_t variables in some places where it was
    dealing with inode numbers.

53. If TMPDIR is defined in Local/Makefile (default in src/EDITME is
    TMPDIR="/tmp"), Exim checks for the presence of an environment variable
    called TMPDIR, and if it finds it is different, it changes its value.

54. The smtp_printf() function is now made available to local_scan() so
    additional output lines can be written before returning. There is also an
    smtp_fflush() function to enable the detection of a dropped connection.
    The variables smtp_input and smtp_batched_input are exported to
    local_scan().

55. Changed the default runtime configuration: the message "Unknown user"
    has been removed from the ACL, and instead placed on the localuser router,
    using the cannot_route_message feature. This means that any verification
    failures that generate their own messages won't get overridden. Similarly,
    the "Unrouteable address" message that was in the ACL for unverifiable
    relay addresses has also been removed.

56. Added hosts_avoid_esmtp to the smtp transport.

57. The exicyclog script was not checking for the esoteric option
    CONFIGURE_FILE_USE_EUID in the Local/Makefile. It now does this, but it
    will work only if exicyclog is run under the appropriate euid.

58. Following a discussion on the list, the rules by which Exim recognises line
    endings on incoming messages have been changed. The -dropcr and drop_cr
    options are now no-ops, retained only for backwards compatibility. The
    following line terminators are recognized: LF CRLF CR. However, special
    processing applies to CR:

    (i)  The sequence CR . CR does *not* terminate an incoming SMTP message,
         nor a local message in the state where . is a terminator.

    (ii) If a bare CR is encountered in a header line, an extra space is added
         after the line terminator so as not to end the header. The reasoning
         behind this is that bare CRs in header lines are most likely either
         to be mistakes, or people trying to play silly games.

59. The size of a message, as listed by "-bp" or in the Exim monitor window,
    was being incorrectly given as 18 bytes larger than it should have been.
    This is a VOB (very old bug).

60. This may never have affected anything current, but just in case it has:
    When the local host is found other than at the start of a list of hosts,
    the local host, those with the same MX, and any that follow, are discarded.
    When the list in question was part of a longer list of hosts, the following
    hosts (not currently being processed) were also being discarded. This no
    longer happens. I'm not sure if this situation could ever has previously
    arisen.

61. Added the "/MX" feature to lists of hosts in the manualroute and query
    program routers.

62. Whenever Exim generates a new message, it now adds an Auto-Submitted:
    header. This is something that is recommended in a new Internet Draft, and
    is something that is documented as being done by Sendmail. There are two
    possible values. For messages generated by the autoreply transport, Exim
    adds:

      Auto-Submitted: auto-replied

    whereas for all other generated messages (e.g. bounces) it adds

      Auto-Submitted: auto-generated

63. The "personal" condition in filters now includes a test for the
    Auto-Submitted: header. If it contains the string "auto-" the message it
    not considered personal.

64. Added rcpt_include_affixes as a generic transport option.

65. Added queue_only_override (default true).

66. Added the syslog_duplication option.

67. If what should have been the first header line of a message consisted of
    a space followed by a colon, Exim was mis-interpreting it as a header line.
    It isn't of course - it is syntactically invalid and should therefore be
    treated as the start of the message body. The misbehaviour could have
    caused a number of strange effects, including loss of data in subsequent
    header lines, and spool format errors.

68. Formerly, the AUTH parameter on a MAIL command was trusted only if the
    client host had authenticated. This control can now be exercised by an ACL
    for more flexibility.

69. By default, callouts do not happen when testing with -bh. There is now a
    variant, -bhc, which does actually run the callout code, including
    consulting and updating the callout cache.

70. Added support for saslauthd authentication, courtesy of Alexander
    Sabourenkov.

71. If statvfs() failed on the spool or log directories while checking their
    size for availability, Exim confusingly gave the error "space shortage".
    Furthermore, in debugging mode it crashed with a floating point exception.
    These checks are done if check_{spool,log}_{space,inodes} are set, and when
    an SMTP message arrives with SIZE= on the MAIL command. As this is a really
    serious problem, Exim now writes to the main and panic logs when this
    happens, with details of the failure. It then refuses to accept the
    incoming message, giving the message "spool directory problem" or "log
    directory problem" with a 421 code for SMTP messages.

72. When Exim is about to re-exec itself, it ensures that the file descriptors
    0, 1, and 2 exist, because some OS complain for execs without them (see
    ChangeLog 4.05/30). If necessary, Exim opens /dev/null to use for these
    descriptors. However, the code omitted to check that the open succeeded,
    causing mysterious errors if for some reason the permissions on /dev/null
    got screwed. Now Exim writes a message to the main and panic logs, and
    bombs out if it can't open /dev/null.

73. Re-vamped the way daemon_smtp_port, local_interfaces, and -oX work and
    interact so that it is all more flexible. It is supposed to remain
    backwards compatible. Also added extra_local_interfaces.

74. Invalid data sent to a SPA (NTLM) server authenticator could cause the code
    to bomb out with an assertion failure - to the client this appears as a
    connection drop. This problem occurs in the part of the code that was taken
    from the Samba project. Fortunately, the assertion is in a very simple
    function, so I have fixed this by reproducing the function inline in the
    one place where it is called, and arranging for authentication to fail
    instead of killing the process with assert().

75. The SPA client code was not working when the server requested OEM rather
    than Unicode encoding.

76. Added code to make require_files with a specific uid setting more usable in
    the case where statting the file as root fails - usually a non-root-mounted
    NFS file system. When this happens and the failure is EACCES, Exim now
    forks a subprocess and does the per-uid checking as the relevant uid.

77. Added process_log_path.

78. If log_file_path was not explicitly set, a setting of check_log_space or
    check_log_inodes was ignored.

79. If a space check for the spool or log partitions fails, the incident is now
    logged. Of course, in the latter case the data may get lost...

80. Added the %p formatting code to string_format() so that it can be used to
    print addresses in debug_print(). Adjusted all the address printing in the
    debugging in store.c to use %p rather than %d.

81. There was a concern that a line of code in smtp_in.c could overflow a
    buffer if a HELO/EHLO command was given followed by 500 or so spaces. As
    initially expressed, the concern was not well-founded, because trailing
    spaces are removed early. However, if the trailing spaces were followed by
    a NULL, they did not get removed, so the overflow was possible. Two fixes
    were applied:

    (a) I re-wrote the offending code in a cleaner fashion.
    (b) If an incoming SMTP command contains a NULL character, it is rejected
        as invalid.

82. When Exim changes uid/gid to the Exim user at daemon start time, it now
    runs initgroups(), so that if the Exim user is in any additional groups,
    they will be used during message reception.


Exim version 4.20
-----------------

The change log for 4.20 and earlier releases has been archived.

****