openssh/regress

Overview.

$ ./configure && make tests

You'll see some progress info. A failure will cause either the make to
abort or the driver script to report a "FATAL" failure.

The test consists of 2 parts. The first is the file-based tests which is
driven by the Makefile, and the second is a set of network or proxycommand
based tests, which are driven by a driver script (test-exec.sh) which is
called multiple times by the Makefile.

Failures in the first part will cause the Makefile to return an error.
Failures in the second part will print a "FATAL" message for the failed
test and continue.

OpenBSD has a system-wide regression test suite. OpenSSH Portable's test
suite is based on OpenBSD's with modifications.


Environment variables.

SKIP_UNIT: Skip unit tests.
SUDO: path to sudo/doas command, if desired. Note that some systems
	(notably systems using PAM) require sudo to execute some tests.
LTESTS: Whitespace separated list of tests (filenames without the .sh
	extension) to run.
OBJ: used by test scripts to access build dir.
TEST_SHELL: shell used for running the test scripts.
TEST_SSH_FAIL_FATAL: set to "yes" to make any failure abort the test
	currently in progress.
TEST_SSH_PORT: TCP port to be used for the listening tests.
TEST_SSH_QUIET: set to "yes" to suppress non-fatal output.
TEST_SSH_SSHD_CONFOPTS: Configuration directives to be added to sshd_config
	before running each test.
TEST_SSH_SSH_CONFOPTS: Configuration directives to be added to
	ssh_config before running each test.
TEST_SSH_TRACE: set to "yes" for verbose output from tests 
TEST_SSH_x: path to "ssh" command under test, where x is one of
	SSH, SSHD, SSHAGENT, SSHADD, SSHKEYGEN, SSHKEYSCAN, SFTP or
	SFTPSERVER
USE_VALGRIND: Run the tests under valgrind memory checker.


Individual tests.

You can run an individual test from the top-level Makefile, eg:
$ make tests LTESTS=agent-timeout

If you need to manipulate the environment more you can invoke test-exec.sh
directly if you set up the path to find the binaries under test and the
test scripts themselves, for example:

$ cd regress
$ PATH=`pwd`/..:$PATH:. TEST_SHELL=/bin/sh sh test-exec.sh `pwd` \
    agent-timeout.sh
ok agent timeout test


Files.

test-exec.sh: the main test driver. Sets environment, creates config files
and keys and runs the specified test.

At the time of writing, the individual tests are:
agent-timeout.sh:	agent timeout test
agent.sh:		simple agent test
broken-pipe.sh:		broken pipe test
connect-privsep.sh:	proxy connect with privsep
connect.sh:		simple connect
exit-status.sh:		remote exit status
forwarding.sh:		local and remote forwarding
keygen-change.sh:	change passphrase for key
keyscan.sh:		keyscan
proto-mismatch.sh:	protocol version mismatch
proto-version.sh:	sshd version with different protocol combinations
proxy-connect.sh:	proxy connect
sftp.sh:		basic sftp put/get
ssh-com-client.sh:	connect with ssh.com client
ssh-com-keygen.sh:	ssh.com key import
ssh-com-sftp.sh:	basic sftp put/get with ssh.com server
ssh-com.sh:		connect to ssh.com server
stderr-after-eof.sh:	stderr data after eof
stderr-data.sh:		stderr data transfer
transfer.sh:		transfer data
try-ciphers.sh:		try ciphers
yes-head.sh:		yes pipe head


Problems?

Run the failing test with shell tracing (-x) turned on:
$ PATH=`pwd`/..:$PATH:. sh -x test-exec.sh `pwd` agent-timeout.sh

Failed tests can be difficult to diagnose. Suggestions:
- run the individual test via ./test-exec.sh `pwd` [testname]
- set LogLevel to VERBOSE in test-exec.sh and enable syslogging of
  auth.debug (eg to /var/log/authlog).


Known Issues.

- Similarly, if you do not have "scp" in your system's $PATH then the
  multiplex scp tests will fail (since the system's shell startup scripts
  will determine where the shell started by sshd will look for scp).

- Recent GNU coreutils deprecate "head -[n]": this will cause the yes-head
  test to fail.  The old behaviour can be restored by setting (and
  exporting) _POSIX2_VERSION=199209 before running the tests.