Testsuite: avoid indeterminate delivery ordering

[users/jgh/exim.git] / test / README

EXPORTABLE EXIM TEST SUITE
--------------------------

This document last updated for:

Test Suite Version: 4.87
Date: 30 January 2016


BACKGROUND
----------

For a long time, the Exim test suite was confined to Philip Hazel's
workstation, because it relied on that particular environment. The problem is
that an MTA such as Exim interacts a great deal with its environment, so if you
run it somewhere else, the output will be different, which makes automatic
checking difficult. Even in a single environment, things are not all that easy.
For instance, if Exim delivers a message, the log line (which one would want to
compare) contains a timestamp and an Exim message id that will be different
each time. This issue is dealt with by a Perl script that munges the output by
recognizing changing sequences and replacing them with fixed values before
doing a comparison. Another problem with exporting the original test suite is
that it assumes a version of Exim with more or less every optional feature
enabled.

This README describes a new test suite that is intended to be exportable and to
run in a number of different environments. The tests themselves are in no
particular order; they accumulated over the years as Exim was extended and
modified. They vary greatly in size and complexity. Some were specifically
constructed to test new features; others were made to demonstrate that a bug
had been fixed.

A few of the original tests have had to be omitted from this more general
suite because differences in operating system behaviour make it impossible to
generalize them. An example is a test that uses a version of Exim that is
setuid to the Exim user rather than root, with the deliver_drop_privilege
option set. In Linux, such a binary is able to deliver a message as the caller
of Exim, because it can revert to the caller's uid. In FreeBSD this is not the
case.


REQUIREMENTS
------------

In order to run this test suite, the following requirements must be met:

(1) You should run the tests on a matching version of Exim, because the suite
    is continuously updated to test the latest features and bug fixes. The
    version you test does not, however, have to be installed as the live
    version. You can of course try the tests on any version of Exim, but some
    may fail. In particular, the test suite will fall apart horrible with
    versions of Exim prior to 4.54.

(2) You can use any non-root login to run the tests, but there must be access
    via "sudo" to root from this login. Privilege is required to override
    configuration change checks and for things like cleaning up spool files,
    but on the other hand, the tests themselves need to call Exim from a
    non-root process. The use of "sudo" is the easiest way to achieve all this.
    The test script uses "sudo" to do a number of things as root, so it is best
    if you set a sudo timeout so that you do not have to keep typing a
    password. For example, if you put

      Defaults timestamp_timeout=480

    in /etc/sudoers, a password lasts for 8 hours (a working day). It is
    not permitted to run the tests as the Exim user because the test suite
    tracks the two users independently.  Using the same user would result
    in false positives on some tests.

    Further, some tests invoke sudo in an environment where there might not be
    a TTY, so tickets should be global, not per-TTY.  Taking this all together
    and assuming a user of "exim-build", you might have this in sudoers:

      Defaults:exim-build     timestamp_timeout=480,!tty_tickets

(3) The login under which you run the tests must have the exim group as a
    secondary so that it has access to logs, spool files, etc.  However, it
    should have a different primary group (eg. "users" vs. "eximgroup"). The
    login should not be one of the names "userx", "usery", "userz", or a few
    other simple ones such as "abcd" and "xyz" and single letters that are used
    in the tests. The test suite expects the login to have a gecos name; I think
    it will now run if the gecos field is empty but there may be anomalies.
    The login must not contain a dash or an equal sign. (Otherwise some tests
    about local_from_{suffix,prefix} will fail.)

(4) The directory into which you unpack the test suite must be accessible by
    the Exim user, so that code running as exim can access the files therein.
    This includes search-access on all path elements leading to it. A
    world-readable directory is fine. However, there may be problems if the
    path name of the directory is excessively long. This is because it
    sometimes appears in log lines or debug output, and if it is truncated, it
    is no longer recognized.

(5) Exim must be built with its user and group specified at build time, and
    with certain minimum facilities, namely:

      Routers:        accept, dnslookup, manualroute, redirect
      Transports:     appendfile, autoreply, pipe, smtp
      Lookups:        lsearch
      Authenticators: plaintext

    Most Exim binaries will have these included.

(6) A C compiler is needed to build some test programs, and the test script is
    written in Perl, so you need that.

(7) Some of the tests run Exim as a daemon, and others use a testing server
    (described below). These require TCP ports. In the configurations and
    scripts, the ports are parameterized, but at present, fixed values are
    written into the controlling script. These are ports 1224 to 1229. If these
    ports are not available for use, some of the tests will fail.

(8) There is an underlying assumption that the host on which the tests are
    being run has an IPv4 address (which the test script seeks out). If there
    is also an IPv6 address, additional tests are run when the Exim binary
    contains IPv6 support. There are checks in the scripts for a running IPv4
    interface; when one is not found, some tests are skipped (with a warning
    message).  The local net may not be in 10.0/8 as that is used by the suite.

(9) Exim must be built with TRUSTED_CONFIG_LIST support, so that the test
    configs can be placed into it.  A suitable file location is .../exim/test/trusted_configs
    with content .../exim/test/test-config [fill out the ... to make full
    paths].  This file should be owner/group matching CONFIGURE_OWNER/GROUP,
    or root/root, and it has to be accessible for the login, under which
    you run the tests.  The config files in .../exim/test/confs/ should be
    owner/group the same.  DISABLE_D_OPTION must not be used. If ALT_CONFIG_PREFIX is used, it
    must contain the directory of the test-suite.  WHITELIST_D_MACROS should contain:

     DIR:EXIM_PATH:AA:ACL:ACLRCPT:ACL_MAIL:ACL_PREDATA:ACL_RCPT:AFFIX:ALLOW:ARG1:ARG2:AUTHF:AUTHS:AUTH_ID_DOMAIN:BAD:BANNER:BB:BR:BRB:CERT:COM:COMMAND_USER:CONNECTCOND:CONTROL:CREQCIP:CREQMAC:CRL:CSS:D6:DATA:DCF:DDF:DEFAULTDWC:DELAY:DETAILS:DRATELIMIT:DYNAMIC_OPTION:ELI:ERROR_DETAILS:ERT:FAKE:FALLBACK:FILTER:FILTER_PREPEND_HOME:FORBID:FORBID_SMTP_CODE:FUSER:HAI:HAP:HARDLIMIT:HEADER_LINE_MAXSIZE:HEADER_MAXSIZE:HELO_MSG:HL:HOSTS:HOSTS_AVOID_TLS:HOSTS_MAX_TRY:HVH:IFACE:IGNORE_QUOTA:INC:INSERT:IP1:IP2:LAST:LDAPSERVERS:LENCHECK:LIMIT:LIST:LOG_SELECTOR:MAXNM:MESSAGE_LOGS:MSIZE:NOTDAEMON:ONCE:ONLY:OPT:OPTION:ORDER:PAH:PEX:PORT:PTBC:QDG:QOLL:QUOTA:QUOTA_FILECOUNT:QWM:RCPT_MSG:REMEMBER:REQUIRE:RETRY:RETRY1:RETRY2:RETURN:RETURN_ERROR_DETAILS:REWRITE:ROUTE_DATA:RRATELIMIT:SELECTOR:SELF:SERVER:SERVERS:SREQCIP:SREQMAC:SRV:STRICT:SUB:SUBMISSION_OPTIONS:TIMEOUTDEFER:TIMES:TRUSTED:TRYCLEAR:UL:USE_SENDER:UTF8:VALUE:WMF

(10) Exim must *not* be built with USE_READLINE, as the test-suite's automation
     assumes the simpler I/O model.
     Exim must *not* be built with HEADERS_CHARSET set to UTF-8.



OPTIONAL EXTRAS
---------------

If the Exim binary that is being tested contains extra functionality in
addition to the minimum specified above, additional tests are run to exercise
the extra functionality, except for a few special cases such as the databases
(MySQL, PostgreSQL, LDAP) where special data is needed for the tests.


RUNNING THE TEST SUITE
----------------------

(1) Download the tarball exim-testsuite-x.xx.tar.bz2 and unpack it, preferably
    in a directory alongside an Exim source directory (see below).

(2) cd into the exim-testsuite-x.xx directory.

(3) Run "./configure" and then "make". This builds a few auxiliary programs that
    are written in C.

(4) echo $PWD/test-config >> your_TRUSTED_CONFIG_LIST_filename
    Typically that is .../exim/test/trusted_configs

(5) Run "./runtest" (a Perl script) as described below.

(6) If you want to see what tests are available, run "./listtests".


BREAKING OUT OF THE TEST SCRIPT
-------------------------------

If you abandon the test run by typing ^C, the interrupt may be passed to a
program that the script is running, or it may be passed to the script itself.
In the former case, the script should detect that the program has ended
abnormally. In both cases, the script tries to clean up everything, including
killing any Exim daemons that it has started. However, there may be race
conditions in which the clean up does not happen. If, after breaking out of a
run, you see strange errors in the next run, look for any left-over Exim
daemons, and kill them by hand.


THE LISTTESTS SCRIPT
--------------------

The individual test scripts are in subdirectories of the "scripts" directory.
If you do not supply any arguments to ./listtests, it scans all the scripts in
all the directories, and outputs the heading line from each script. The output
is piped through "less", and begins like this:

=== 0000-Basic ===
Basic/0001 Basic configuration setting
Basic/0002 Common string expansions
Basic/0003 Caseless address blocking
...

Lines that start === give the name of the subdirectory containing the test
scripts that follow. If you supply an argument to ./listtests, it is used as a
Perl pattern to match case-independently against the names of the
subdirectories. Only those that match are scanned. For example, "./listtests
ipv6" outputs this:

=== 1000-Basic-ipv6 ===
=== Requires: support IPv6
Basic-ipv6/1000 -bh and non-canonical IPv6 addresses
Basic-ipv6/1001 recognizing IPv6 address in HELO/EHLO

=== 2250-dnsdb-ipv6 ===
=== Requires: support IPv6
              lookup dnsdb
dnsdb-ipv6/2250 dnsdb ipv6 lookup in string expansions

If you supply a second argument to ./listtests, it is used as a Perl pattern to
match case-independently against the individual script titles. For example,
"./listtests . mx" lists all tests whose titles contain "mx", because "."
matches all the subdirectory names.


THE RUNTEST SCRIPT
------------------

If you do not supply any arguments to ./runtest, it searches for an Exim
source tree at the same level as the test suite directory. A source tree
is a source tree, if it contains a build-* directory.

It then looks for an Exim binary in a "build" directory of that source
tree. If there are several Exim source trees, it chooses the latest
version of Exim. Consider the following example:

  $ ls -F /source/exim
  exim-4.60/  exim-4.62/  exim-testsuite-x.xx/

A simple ./runtest from within the test suite will use a 4.62 binary if it
finds one, otherwise a 4.60 binary. If a binary cannot be found, the script
prompts for one. Alternatively, you can supply the binary on the command line:

  ./runtest /usr/exim/bin/exim

A matching test suite is released with each Exim release; if you use a test
suite that does not match the binary, some tests may fail.

The test suite uses some of the Exim utilities (such as exim_dbmbuild), and it
expects to find them in the same directory as Exim itself. If they are not
found, the tests that use them are omitted. A suitable comment is output.

On the ./runtest command line, following the name of the binary, if present,
there may be a number of options and then one or two numbers. The full syntax
is as follows:

  ./runtest [binary name] [runtest options] [exim options] \
            [first test] [last test]

There are some options for the ./runtest script itself:

  -CONTINUE This will allow the script to move past some failing tests. It will
            write a simple failure line with the test number in a temporary
            logfile test/failed-summary.log. Unexpected exit codes will still
            stall the test execution and require interaction.

  -DEBUG    This option is for debugging the test script. It causes some
            tracing information to be output.

  -DIFF     By default, file comparisons are done using a private compare
            command called "cf", which is built from source that is provided in
            the src directory. This is a command I've had for nearly 20 years -
            look at the source comments for its history - whose output I
            prefer. However, if you want to use "diff" instead, give -DIFF as a
            runtest option. In that case, "diff -u" is used for comparisons.
            (If it turns out that most people prefer to use diff, I'll change
            the default.)

  -FLAVOR <flavor>
  -FLAVOUR <flavour>
            This allows "overrides" for the test results. It's intended
            use is to deal with distro specific differences in the test
            output. The default flavour is "FOO" if autodetection fails.
            (Autodetection is possible for known flavours only. Known
            flavours are computed after file name extensions in stdout/*
            and stderr/*.)

            If during the test run differences between the current and
            the expected output are found and no flavour file exists already,
            you may update the "common" expected output or you may create a
            flavour file. If  a flavour file already exists, any updates will go
            into that flavour file!

  -KEEP     Normally, after a successful run, the test output files are
            deleted. This option prevents this. It is useful when running a
            single test, in order to look at the actual output before it is
            modified for comparison with saved output.

  -NOIPV4   Pretend that an IPv4 interface was not found. This is useful for
            testing that the test suite correctly skips tests that require
            a running IPv4 interface.

  -NOIPV6   Pretend that an IPv6 interface was not found. This is useful for
            testing that the test suite correctly skips tests that require
            a running IPv6 interface.

  -UPDATE   If this option is set, any detected changes in test output are
            automatically accepted and used to update the stored copies of the
            output. It is a dangerous option, but it useful for the test suite
            maintainer after making a change to the code that affects a lot of
            tests (for example, the wording of a message).

  -SLOW     For very slow hosts that appear to have Heisenbugs, delay before
            comparing output files from a testcase

The options for ./runtest must be given first (but after the name of the
binary, if present). Any further options, that is, items on the command line
that start with a hyphen, are passed to the Exim binary when it is run as part
of a test. The only sensible use of this is to pass "-d" in order to run a test
with debugging enabled. Any other options are likely to conflict with options
that are set in the tests. Some tests are already set up to run with debugging.
In these cases, -d on the command line overrides their own debug settings.

The final two arguments specify the range of tests to be run. Test numbers lie
in the range 1 to 9999. If no numbers are given, the defaults are 1 and 8999
(sic). Tests with higher numbers (9000 upwards) are not run automatically
because they require specific data (such as a particular MySQL table) that is
unlikely to be generally available.

Tests that require certain optional features of Exim are grouped by number, so
in any given range, not all the tests will exist. Non-existent tests are just
skipped, but if there are no tests at all in the given range, a message is
output.

If you give only one number, just that test is run (if it exists). Instead of a
second number, you can give the character "+", which is interpreted as "to the
end". Normally this is 8999; if the starting number is 9000 or higher, "+" is
interpreted as 9999. Examples:

  ./runtest 1300
  ./runtest 1400 1699
  ./runtest /usr/sbin/exim 5000 +
  ./runtest -DIFF -d 81

When the script starts up, the first thing it does is to check that you have
sudo access to root. Then it outputs the version number of the Exim binary that
it is testing, and also information about the optional facilities that are
present (obtained from "exim -bV"). This is followed by some environmental
information, including the current login id and the hosts's IP address. The
script checks that the current user is in the Exim group, and that the Exim
user has access to the test suite directory.

The script outputs the list of tests requested, and a list of tests that will
be omitted because the relevant optional facilities are not in the binary. You
are then invited to press Return to start the tests running.


TEST OUTPUT
-----------

When all goes well, the only permanent output is the identity of the tests as
they are run, and "Script completed" for each test script, for example:

  Basic/0001 Basic configuration setting
    Script completed
  Basic/0002 Basic string expansions
    Script completed
  Basic/0003 Caseless address blocking
    Script completed
  Basic/0004 Caseful address blocking
    Script completed
  Basic/0005 -bs to simple local delivery
  ...

While a script is running, it shows "Test n" on the screen, for each of the
Exim tests within the script. There may also be comments from some tests when a
delay is expected, for example, if there is a "sleep" while testing a timeout.

Before each set of optional tests, an extra identifying line is output. For
example:

  >>> The following tests require: authenticator cram_md5
  CRAM-MD5/2500 CRAM-MD5 server tests
    Script completed
  CRAM-MD5/2501 CRAM-MD5 client tests
    Script completed

If a test fails, you are shown the output of the text comparison that failed,
and prompted as to what to do next. The output is shown using the "less"
command, or "more" if "less" is not available. The options for "less" are set
to that it automatically exits if there is less that a screenful of output. By
default, the output is from the "cf" program, and might look like this:

  DBM/1300 DBM files and exim_dbmbuild
  ===============
  Lines 7-9 of "test-stdout-munged" do not match lines 7-11 of "stdout/1300".
  ----------
  exim_dbmbuild exit code = 1
  Continued set of lines is too long: max permitted length is 99999
  exim_dbmbuild exit code = 1
  ----------
  dbmbuild abandoned
  exim_dbmbuild exit code = 2
  Continued set of lines is too long: max permitted length is 99999
  dbmbuild abandoned
  exim_dbmbuild exit code = 2
  ===============
  1 difference found.
  "test-stdout-munged" contains 16 lines; "stdout/1300" contains 18 lines.

  Continue, Retry, Update & retry, Quit? [Q]

This example was generated by running the test with a version of Exim
that had a bug in the exim_dbmbuild utility (the bug was fixed at release
4.53). See "How the tests work" below for a description of the files that are
used. In this case, the standard output differed from what was expected.

The reply to the prompt must either be empty, in which case it takes the
default that is given in brackets (in this case Q), or a single letter, in
upper or lower case (in this case, one of C, R, U, or Q). If you type anything
else, the prompt is repeated.

"Continue" carries on as if the files had matched; that is, it ignores the
mismatch. Any other output files for the same test will be compared before
moving on to the next test.

"Update & retry" copies the new file to the saved file, and reruns the test
after doing any further comparisons that may be necessary.

"Retry" does the same apart from the file copy.

Other circumstances give rise to other prompts. If a test generates output for
which there is no saved data, the prompt (after a message stating which file is
unexpectedly not empty) is:

  Continue, Show, or Quit? [Q]

"Show" displays the data on the screen, and then you get the "Continue..."
prompt. If a test ends with an unexpected return code, the prompt is:

  show stdErr, show stdOut, Continue (without file comparison), or Quit? [Q]

Typically in these cases there will be something interesting in the stderr
or stdout output. There is a similar prompt after the "server" auxiliary
program fails.


OPENSSL AND GNUTLS ERROR MESSAGES
---------------------------------

Some of the TLS tests deliberately cause errors to check how Exim handles them.
It has been observed that different releases of the OpenSSL and GnuTLS
libraries generate different error messages. This may cause the comparison with
the saved output to fail. Such errors can be ignored.


OTHER ISSUES
------------

. Some of the tests are time-sensitive (e.g. when testing timeouts, as in test
  461). These may fail if run on a host that is also running a lot of other
  processes.

. Some versions of "ls" use a different format for times and dates. This can
  cause test 345 to fail.

. Test 0142 tests open file descriptors; on some hosts the output may vary.

. Some tests may fail, for example 0022, because it says it uses cached data
  when the expected output thinks it should not be in cache.  Item #5 in the
  Requirements section has:
    "Exim must be built with its user and group specified at build time"
  This means that you cannot use the "ref:username" in your Local/Makefile
  when building the exim binary, in any of the following fields:
    EXIM_USER  EXIM_GROUP  CONFIGURE_OWNER  CONFIGURE_GROUP

. If the runtest script warns that the hostname is not a Fully Qualified
  Domain Name (FQDN), expect that some tests will fail, for example 0036,
  with an extra log line saying the hostname doesn't resolve.  You must use a
  FQDN for the hostname for proper test functionality.

. If you change your hostname to a FQDN, you must delete the test/dnszones
  subdirectory.  When you next run the runtest script, it will rebuild the
  content to use the new hostname.

. If your hostname has an uppercase characters in it, expect that some tests
  will fail, for example, 0036, because some log lines will have the hostname
  in all lowercase.  The regex which extracts the hostname from the log lines
  will not match the lowercased version.

. Some tests may fail, for example 0015, with a cryptic error message:
    Server return code 99
  Due to security concerns, some specific files MUST have the group write bit
  off.  For the purposes of the test suite, some test/aux-fixed/* files MUST
  have the group write bit off, so it's easier to just remove the group write
  bit for all of them.  If your umask is set to 002, the group write bit will
  be on by default and you'll see this problem, so make sure your umask is
  022 and re-checkout the test/ subdirectory.

. Some tests will fail if the username and group name are different. It does
  not have to be the primary group, a secondary group is sufficient.


OTHER SCRIPTS AND PROGRAMS
--------------------------

There is a freestanding Perl script called "listtests" that scans the test
scripts and outputs a list of all the tests, with a short descriptive comment
for each one. Special requirements for groups of tests are also noted.

The main runtest script makes use of a second Perl script and some compiled C
programs. These are:

patchexim          A Perl script that makes a patched version of Exim (see the
                   next section for details).

bin/cf             A text comparison program (see above).

bin/checkaccess    A program that is run as root; it changes uid/gid to the
                   Exim user and group, and then checks that it can access
                   files in the test suite's directory.

bin/client         A script-driven SMTP client simulation.

bin/client-gnutls  A script-driven SMTP client simulation with GnuTLS support.
                   This is built only if GnuTLS support is detected on the host.

bin/client-ssl     A script-driven SMTP client simulation with OpenSSL support.
                   This is built only if OpenSSL support is detected on the
                   host.

bin/fakens         A fake "nameserver" for DNS tests (see below for details).

bin/fd             A program that outputs details of open file descriptors.

bin/iefbr14        A program that does nothing, and returns 0. It's just like
                   the "true" command, but it is in a known place.

bin/loaded         Some dynamically loaded functions for testing dlfunc support.

bin/mtpscript      A script-driven SMTP/LMTP server simulation, on std{in,out}.

bin/server         A script-driven SMTP server simulation, over a socket.

bin/showids        Output the current uid, gid, euid, egid.

The runtest script also makes use of a number of ordinary commands such as
"cp", "kill", "more", and "rm", via the system() call. In some cases these are
run as root by means of sudo.


STANDARD SUBSTITUTIONS
----------------------

In the following sections, there are several references to the "standard
substitutions". These make changes to some of the stored files when they are
used in a test. To save repetition, the substitutions themselves are documented
here:

  CALLER         is replaced by the login name of the user running the tests
  CALLERGROUP    is replaced by the caller's group id
  CALLER_GID     is replaced by the caller's group id
  CALLER_UID     is replaced by the caller's user id
  DIR            is replaced by the name of the test-suite directory
  EXIMGROUP      is replaced by the name of the Exim group
  EXIMUSER       is replaced by the name of the Exim user
  HOSTIPV4       is replaced by the local host's IPv4 address
  HOSTIPV6       is replaced by the local host's IPv6 address
  HOSTNAME       is replaced by the local host's name
  PORT_D         is replaced by a port number for normal daemon use
  PORT_N         is replaced by a port number that should never respond
  PORT_S         is replaced by a port number for normal bin/server use
  PORT_DYNAMIC   is replaced by a port number allocated dynamically
  TESTNUM        is replaced by the current test number
  V4NET          is replaced by an IPv4 network number for testing
  V6NET          is replaced by an IPv6 network number for testing

PORT_D is currently hard-wired to 1225, PORT_N to 1223, and PORT_S to 1224.
V4NET is hardwired to 224 and V6NET to ff00. These networks are used for DNS
testing purposes, and for testing Exim with -bh. The only requirement is that
they are networks that can never be used for an IP address of a real host. I've
chosen two multicast networks for the moment.

PORT_DYNAMIC is allocated by hunting for a free port (starting at port
1024) a listener can bind to. This is done by runtest, for simulating
inetd operations.

If the host has no IPv6 address, "<no IPv6 address found>" is substituted but
that does not matter because no IPv6 tests will be run. A similar substitution
is made if there is no IPv4 address, and again, tests that actually require a
running IPv4 interface should be skipped.

If the host has more than one IPv4 or IPv6 address, the first one that
"ifconfig" lists is used. If the only available address is 127.0.0.1 (or ::1
for IPv6) it is used, but another value is preferred if available.

In situations where a specific test is not being run (for example, when setting
up dynamic data files), TESTNUM is replaced by an empty string, but should not
in fact occur in such files.


HOW THE TESTS WORK
------------------

Each numbered script runs Exim (sometimes several times) with its own Exim
configuration file. The configurations are stored in the "confs" directory,
and before running each test, a copy of the appropriate configuration, with the
standard substitutions, is made in the file test-config. The -C command line
option is used to tell Exim to use this configuration.

The -D option is used to pass the path of the Exim binary to the configuration.
This is not standardly substituted, because there are two possible binaries
that might be used in the same test (one setuid to root, the other to the exim
user). Some tests also make use of -D to vary the configuration for different
calls to the Exim binary.

Normally, of course, Exim gives up root privilege when -C and -D are used by
unprivileged users. We do not want this to happen when running the tests,
because we want to be able to test all aspects of Exim, including receiving
mail from unprivileged users. The way this is handled is as follows:

At the start of the runtest script, the patchexim script is run as root. This
script makes a copy of the Exim binary that is to be tested, patching it as it
does so. (This is a binary patch, not a source patch.) The patch causes the
binary, when run, to "know" that it is running in the test harness. It does not
give up root privilege when -C and -D are used, and in a few places it takes
other special actions, such as delaying when starting a subprocess to allow
debug output from the parent to be written first. If you want to know more,
grep the Exim source files for "running_in_test_harness".

The patched binary is placed in the directory eximdir/exim and given the normal
setuid root privilege. This is, of course, a dangerous binary to have lying
around, especially if there are unprivileged users on the system. To protect
it, the eximdir directory is created with the current user as owner, exim as
the group owner, and with access drwx--x---. Thus, only the user who is running
the tests (who is known to have access to root) and the exim user have access
to the modified Exim binary. When runtest terminates, the patched binary is
removed.

Each set of tests proceeds by interpreting its controlling script. The scripts
are in subdirectories of the "scripts" directory. They are split up according
to the requirements of the tests they contain, with the 0000-Basic directory
containing tests that can always be run. Run the "listtests" script to obtain a
list of tests.


TEST OUTPUT
-----------

Output from script runs is written to the files test-stdout and test-stderr.
When an Exim server is involved, test-stdout-server and test-stderr-server are
used for its output. Before being compared with the saved output, the
non-server and server files are concatenated, so a single saved file contains
both.

A directory called spool is used for Exim's spool files, and for Exim logs.
These locations are specified in every test's configuration file.

When messages are delivered to files, the files are put in the test-mail
directory. Output from comparisons is written to test-cf.

Before comparisons are done, output texts are modified ("munged") to change or
remove parts that are expected to vary from run to run. The modified files all
end with the suffix "-munged". Thus, you will see test-stdout-munged,
test-mainlog-munged, test-mail-munged, and so on. Other files whose names start
with "test-" are created and used by some of the tests.

At the end of a successful test run, the spool directory and all the files
whose names begin with "test-" are removed. If the run ends unsuccessfully
(typically after a "Q" response to a prompt), the spool and test files are left
in existence so that the problem can be investigated.


TEST COMMANDS
-------------

Each test script consists of a list of commands, each optionally preceded by
comments (lines starting with #) and (also optionally) a line containing an
expected return code. Some of the commands are followed by data lines
terminated by a line of four asterisks.

The first line of each script must be a comment that briefly describes the
script. For example:

  # -bS Use of HELO/RSET

A line consisting just of digits is interpreted as the expected return code
for the command that follows. The default expectation when no such line exists
is a zero return code. For example, here is a complete test script, containing
just one command:

  # -bS Unexpected EOF in headers
  1
  exim -bS -odi
  mail from:<someone@some.where>
  rcpt to:<blackhole@HOSTNAME>
  data
  from: me
  ****

The expected return code in this case is 1, and the data lines are passed to
Exim on its standard input. Both the command line and the data lines have the
standard substitutions applied to them. Thus, HOSTNAME in the example above will
be replaced by the local host's name. Long commands can be continued over
several lines by using \ as a continuation character. This does *not* apply to
data lines.

Here follows a list of supported commands. They can be divided into two groups:


Commands with no input
----------------------

These commands are not followed by any input data, or by a line of asterisks.


  dbmbuild <file1> <file1>

This command runs the exim_dbmbuild utility to build a DBM file. It is used
only when DBM support is available in Exim, and typically follows the use of a
"write" command (see below) that creates the input file.


  dump <dbname>

This command runs the exim_dumpdb utility on the testing spool directory, using
the database name given, for example: "dumpdb retry".


  echo <text>

The text is written to the screen; this is used to output comments from
scripts.


  exim_lock [options] <file name>

This command runs the exim_lock utility with the given options and file name.
The file remains locked with the following command (normally exim) is obeyed.


  exinext <data>

This command runs the exinext utility with the given argument data.


  exigrep <data>

This command runs the exigrep utility with the given data (the search pattern)
on the current mainlog file.


  gnutls

This command is present at the start of all but one of the tests that use
GnuTLS. It copies a pre-existing parameter file into the spool directory, so
that Exim does not have to re-create the file each time. The first GnuTLS test
does not do this, in order to test that Exim can create the file.


  killdaemon

This command must be given in any script that starts an Exim daemon, normally
at the end. It searches for the PID file in the spool directory, and sends a
SIGINT signal to the Exim daemon process whose PID it finds. See below for
comments about starting Exim daemons.


  millisleep <m>

This command causes the script to sleep for m milliseconds. Nothing is output
to the screen.


  munge <name>

This command requests custom munging of the test outputs.  The munge names
used are coded in the runtest script (look for 'name of munge').


  need_ipv4

This command must be at the head of a script. If no IPv4 interface has been
found, the entire script is skipped, and a comment is output.


  need_ipv6

This command must be at the head of a script. If no IPv6 interface has been
found, the entire script is skipped, and a comment is output.


  need_largefiles

This command must be at the head of a script. If the Exim binary does not
support large files (off_t is <= 4), the entire script is skipped, and a
comment is output.


  need_move_frozen_messages

This command must be at the head of a script. If the Exim binary does not have
support for moving frozen messages (which is an optional feature), the entire
script is skipped, and a comment is output.


  no_message_check

If this command is encountered anywhere in the script, messages that are
delivered when the script runs are not compared with saved versions.


  no_msglog_check

If this command is encountered anywhere in the script, message log files that
are still in existence at the end of the run (for messages that were not
delivered) are not compared with saved versions.


  no_stderr_check

If this command is encountered anywhere in the script, the stderr output from
the run is not compared with a saved version.


  no_stdout_check

If this command is encountered anywhere in the script, the stdout output from
the run is not compared with a saved version.


  rmfiltertest

This command indicates that the script is for a certain type of filter test, in
which there are a lot of repetitive stdout lines that get in the way, because
filter tests output data about the sender and recipient. Such lines are removed
from the stdout output before comparing, for ease of human perusal.


  sleep <n>

This command causes the script to sleep for n seconds. If n is greater than
one, "sleep <n>" is output to the screen, followed by a dot for every second
that passes.


  sortlog

This command causes special sorting to occur on the mainlog file before
comparison. Every sequence of contiguous delivery lines (lines containing the
=> -> or *> flags) is sorted. This is necessary in some tests that use parallel
deliveries because on different systems the processes may terminate in a
different order.


A number of standard file management commands are also recognized. These are
cat, chmod, chown, cp, du, ln, ls, du, mkdir, mkfifo, rm, rmdir, and touch.
Some are run as root using "sudo".


Commands with input
-------------------

The remaining commands are followed by data lines for their standard input,
terminated by four asterisks. Even if no data is required for the particular
usage, the asterisks must be given.


  background

This command takes one script line and runs it in the background,
in parallel with following commands.  For external daemons, eg. redis-server.


  catwrite <file name> [nxm[=start-of-line-text]]*

This command operates like the "write" command, which is described below,
except that the data it generates is copied to the end of the test-stdout file
as well as to the named file.



  client [<options>] <ip address> <port> [<outgoing interface>]

This command runs the auxiliary "client" program that simulates an SMTP client.
It is controlled by a script read from its standard input, details of which are
given below. There are two options. One is -t, which must be followed directly
by a number, to specify the command timeout in seconds (e.g. -t5). The default
timeout is 5 seconds. The other option is -tls-on-connect, which causes the
client to try to start up a TLS session as soon as it has connected, without
using the STARTTLS command. The client program connects to the given IP address
and port, using the specified interface, if one is given.


  client-ssl [<options>] <ip address> <port> [<outgoing interface>] \
             [<cert file>] [<key file>]

When OpenSSL is available on the host, an alternative version of the client
program is compiled, one that supports TLS using OpenSSL. The additional
arguments specify a certificate and key file when required for the connection.
There are two additional options: -tls-on-connect, that causes the client to
initiate TLS negotiation immediately on connection; -ocsp that causes the TLS
negotiation to include a certificate-status request.  The latter takes a
filename argument, the CA info for verifying the stapled response.


  client-gnutls [<options>] <ip address> <port> [<outgoing interface>] \
                [<cert file>] [<key file>]

When GnuTLS is available on the host, an alternative version of the client
program is compiled, one that supports TLS using GnuTLS. The additional
arguments specify a certificate and key file when required. There is one
additional option, -tls-on-connect, that causes the client to initiate TLS
negotiation immediately on connection.


  exim [<options>] [<arguments>]

This command runs the testing version of Exim. Any occurrence of "$msg1" in the
command line is replaced by the ID of the first (oldest) message in Exim's
(testing) spool. "$msg2" refers to the second, and so on. The name "exim" can
be preceded by an environment setting as in this example:

  LDAPTLS_REQCERT=never exim -be

It can also be preceded by a number; this specifies a number of seconds to wait
before closing the stdout pipe to Exim, and is used for some timeout tests. For
example:

  3 exim -bs

Finally, "exim" can be preceded by "sudo", to run Exim as root. If more than
one of these prefixes is present, they must be in the above order.

If the options include "-DSERVER" but not "-DNOTDAEMON", the script waits for
Exim to start but then continues without waiting for it to terminate. Typically
this will be for a daemon-mode "-bd" operation.  The daemon should be later
terminated using "killdaemon".


  exim_exim [<options>] [<arguments>]

This runs an alternative version of Exim that is setuid to exim rather than to
root.


  server [<options>] <port or socket> [<connection count>]

This command runs the auxiliary "server" program that simulates an SMTP (or
other) server. It is controlled by a script that is read from its standard
input, details of which are given below. A number of options are implemented:

  -d       causes the server to output debugging information

  -t <sec> sets a timeout (default 5) for when the server is
           awaiting an incoming connection. If negative, the
           absolute value is used and a timeout results in a
           nonfailure exit code

  -noipv4  causes the server not to set up an IPv4 socket

  -noipv6  causes the server not to set up an IPv6 socket

  -i <sec> sets an initial pause, to delay before creating the listen sockets

By default, in an IPv6 environment, both kinds of socket are set up. However,
the test script knows which interfaces actually exist on the host, and it adds
-noipv4 or -noipv6 to the server command as required. An error occurs if both
these options are given.

The only required argument is either a port number or the path name of a Unix
domain socket. The port is normally PORT_S, which is changed to an actual
number by the standard substitutions. The optional final argument specifies the
number of different connections to expect (default 1). These must happen
serially (one at a time). There is no support for multiple simultaneous
connections. Here are some example commands:

  server PORT_S
  server -t 10 PORT_S 3
  server /tmp/somesocket

The following lines, up to a line of four asterisks, are the server's
controlling standard input (described below). These lines are read and
remembered; during the following commands, until an "exim" command is reached,
the server is run in parallel.


  write <file name> [nxm[=start-of-line-text]]*

The "write" command is a way of creating files of specific sizes for buffering
tests, or containing specific data lines. Being able to do this from within the
script saves holding lots of little test files. The optional argument specifies
n lines of length m. The lines consist of the letter "a". If start of line text
is supplied, it replaces "a"s at the start of each line. Underscores in the
start of line text are turned into spaces. The optional argument may be
repeated. The data lines that follow a "write" command are split into two by a
line of four plus signs. Any above the split are written before the
fixed-length lines, and any below the split are written after. For example:

  write test-data 3x30=AB_ 1x50
  Pre-data
  lines
  ++++
  Post-data
  lines
  ****

This command generates a file containing:

  Pre-data
  lines
  AB aaaaaaaaaaaaaaaaaaaaaaaaaaa
  AB aaaaaaaaaaaaaaaaaaaaaaaaaaa
  AB aaaaaaaaaaaaaaaaaaaaaaaaaaa
  aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
  Post-data
  lines

If there are no fixed-length line specifiers, there is no need to split the
data, and a line of plusses is not needed.


  [sudo] perl

This command runs Perl, with the data as its standard input, to allow arbitrary
one-off things to be done.


CLIENT SCRIPTS
--------------

Lines in client scripts are of two kinds:

(1) If a line begins with three question marks and a space, the rest of the
    line defines the start of expected output from the server. If what is
    received does not match, the client bombs out with an error message.

(2) If a line starts with three plus signs followed by a space, the rest of the
    line specifies a number of seconds to sleep for before proceeding.

(3) If a line begins with three '>' characters and a space, the rest of the
    line is input to be sent to the server.  Backslash escaping is done as
    described below, but no trailing "\r\n" is sent.

(4) If a line begin with three '<' characters and a space, the rest of the
    line is a filename; the content of the file is inserted intto the script
    at this point.

(5) Otherwise, the line is an input line line that is sent to the server. Any
    occurrences of \r and \n in the line are turned into carriage return and
    linefeed, respectively. This is used for testing PIPELINING.
    Any sequences of \x followed by two hex digits are converted to the equvalent
    byte value.  Any other character following a \ is sent verbatim.
    The line is sent with a trailing "\r\n".

Here is a simple example:

  client 127.0.0.1 PORT_D
  ??? 220
  EHLO xxx
  ??? 250-
  ??? 250
  AUTH PLAIN AbdXi0AdnD2CVy
  ??? 535
  quit
  ??? 221
  ****

In the case of client-gnutls and client-ssl, if a command is "starttls", this
is remembered, and after a subsequent OK response, an attempt to move into TLS
mode occurs. If a command is "starttls_wait", the client sends "starttls" but
does not start up TLS; this is for testing timeouts. If a command is "stoptls",
an existing TLS connection is shut down, but nothing is sent.


SERVER SCRIPTS
--------------

The server program sleeps till a connection occurs or its timeout is reached,
in which case it bombs out. The next set of command lines are interpreted. They
are of the following kinds:

(1) A line that starts with '>' or with a digit is an output line that is sent
    to the client. In the case of '>':

    (a) If the line starts with ">>", no terminating CRLF is sent.
    (b) If the line starts with ">CR>", just CR is sent at the end.
    (c) If the line starts with ">LF>", just LF is sent at the end.
    (d) If the line starts with ">*eof", nothing is sent and the connection
          is closed.

    The data that is sent starts after the initial '>' sequence.  Within
    each line the sequence '\x' followed by two hex digits can be used
    to specify an arbitrary byte value.  The sequence '\\' specifies a
    single backslash.

(2) A line that starts with "*sleep" specifies a number of seconds to wait
    before proceeding.

(3) A line containing "*eof" specifies that the client is expected to close
    the connection at this point.

(4) A line containing just '.' specifies that the client is expected to send
    many lines, terminated by one that contains just a dot.

(5) Otherwise, the line defines the start of an input line that the client
    is expected to send. To allow for lines that start with digits, the line
    may start with '<', which is not taken as part of the input data. If the
    lines starts with '<<' then only the characters are expected; no return-
    linefeed terminator. If the input does not match, the server bombs out
    with an error message.  Backslash-escape sequences may be used in the
    line content as for output lines.

Here is a simple example of server use in a test script:

  server PORT_S
  220 Greetings
  EHLO
  250 Hello there
  MAIL FROM
  250 OK
  RCPT TO
  250 OK
  DATA
  354 Send it!
  .
  250 OK
  QUIT
  225 OK
  ****

After a "server" command in a test script, the server runs in parallel until an
"exim" command is reached. The "exim" command attempts to deliver one or more
messages to port PORT_S on the local host. When it has finished, the test
script waits for the "server" process to finish.

The "mtpscript" program is like "server", except that it uses stdin/stdout for
its input and output instead of a script. However, it is not called from test
scripts; instead it is used as the command for pipe transports in some
configurations, to simulate non-socket LMTP servers.


AUXILIARY DATA FILES
--------------------

Many of the tests make use of auxiliary data files. There are two types; those
whose content is fixed, and those whose content needs to be varied according to
the current environment. The former are kept in the directory aux-fixed. The
latter are distributed in the directory aux-var-src, and copied with the
standard substitutions into the directory aux-var at the start of each test
run.

Most of the auxiliary files have names that start with a test number,
indicating that they are specific to that one test. A few fixed files (for
example, some TLS certificates) are used by more than one test, and so their
names are not of this form.

There are also some auxiliary DNS zone files, which are described in the next
section.


DNS LOOKUPS AND GETHOSTBYNAME
-----------------------------

The original test suite required special testing zones to be loaded into a
local nameserver. This is no longer a requirement for the new suite. Instead, a
program called fakens is used to simulate a nameserver. When Exim is running in
the test harness, instead of calling res_search() - the normal call to the DNS
resolver - it calls a testing function. This handles a few special names itself
(for compatibility with the old test suite), but otherwise passes the query to
the fakens program.

The fakens program consults "zone files" in the directory called dnszones, and
returns data in the standard resource record format for Exim to process as if
it came from the DNS. However, if the requested domain is not in any of the
zones that fakens knows about, it returns a special code that causes Exim to
pass the query on to res_search(). The zone files are:

  db.test.ex    A zone for the domain test.ex.
  db.ip4.10     A zone for one special case in 10.250.0.0/16 (see below)
  db.ip4.V4NET  A zone for the domain V4NET.in-addr.arpa.
  db.ip4.127    A zone for the domain 127.in-addr.arpa.
  db.ip6.V6NET  A zone for the domain inverted(V6NET).ip6.arpa.
  db.ip6.0      A zone for the domain 0.ip6.arpa.

V4NET and V6NET are substituted with the current testing networks (see above).
In the case of V6NET, the network is four hex digits, and it is split and
inverted appropriately when setting up the zone.

These fake zone files are built dynamically from sources in the dnszones-src
directory by applying the standard substitutions. The test suite also builds
dynamic zone files for the name of the current host and its IP address(es). The
idea is that there should not be any need to rely on an external DNS.

The fakens program handles some names programmatically rather than using the
fake zone files.  These are:

  manyhome.test.ex  This name is used for testing hosts with ridiculously large
                    numbers of IP addresses; 2048 IP addresses are generated
                    and returned. Doing it this way saves having to make the
                    interface to fakens handle more records that can fit in the
                    data block. The addresses that are generated are in the
                    10.250.0.0/16 network.

  test.again.dns    This always provokes a TRY_AGAIN response, for testing the
                    handling of temporary DNS error. If the full domain name
                    starts with digits, a delay of that many seconds occurs.

  test.fail.dns     This always provokes a NO_RECOVERY response, for testing
                    DNS server failures.

The use of gethostbyname() and its IPv6 friends is also subverted when Exim is
running in the test harness. The test code handles a few special names
directly; for all the others it uses DNS lookups, which are then handled as
just described. Thus, the use of /etc/hosts is completely bypassed. The names
that are specially handled are:

  localhost         Always returns 127.0.0.1 or ::1, for IPv4 and IPv6 lookups,
                    respectively.

  <an IP address>   If the IP address is of the correct form for the lookup
                    type (IPv4 or IPv6), it is returned. Otherwise a panic-die
                    error occurs.

The reverse zone db.ip4.10 is provided just for the manyhome.test.ex case. It
contains a single wildcard resource record. It also contains the line

  PASS ON NOT FOUND

Whenever fakens finds this line in a zone file, it returns PASS_ON instead of
HOST_NOT_FOUND. This causes Exim to pass the query to res_search().

****