kn@openbsd.org [Fri, 16 Aug 2019 11:16:32 +0000 (11:16 +0000)]
upstream: Call comma-separated lists as such to clarify semantics
Options such as Ciphers take values that may be a list of ciphers; the
complete list, not indiviual elements, may be prefixed with a dash or plus
character to remove from or append to the default list respectively.
Users might read the current text as if each elment took an optional prefix,
so tweak the wording from "values" to "list" to prevent such ambiguity for
all options supporting this semantics (those that provide a list of
available elements via "ssh -Q ...").
upstream: When using a combination of a Yubikey+GnuPG+remote
forwarding the gpg-agent (and options ControlMaster+RemoteForward in
ssh_config(5)) then the codepath taken will call mux_client_request_session
-> mm_send_fd -> sendmsg(2). Since sendmsg(2) is not allowed in that codepath
then pledge(2) kills the process.
The solution is to add "sendfd" to pledge(2), which is not too bad considering
a little bit later we reduce pledge(2) to only "stdio proc tty" in that
codepath.
Problem reported and diff provided by Timothy Brown <tbrown at freeshell.org>
Darren Tucker [Wed, 24 Jul 2019 04:21:19 +0000 (14:21 +1000)]
Split test targets further.
Splits test into file-tests, t-exec, unit and interop-tests and their
respective dependencies. Should allow running any set individually
without having to build the other dependencies that are not needed
for that specific test.
Darren Tucker [Tue, 23 Jul 2019 13:06:22 +0000 (23:06 +1000)]
Split regress-binaries into two targets.
Split the binaries for the unit tests out into a regress-unit-binaries
target, and add a dependency on it for only the unit tests. This allows
us to run the integration tests only ("make t-exec") without building
the unit tests, which allows us to run a subset of the tests when
building --without-openssl without trying (and failing) to build the
unit tests.
This means there are two targets for "unit" which I *think* is valid
(it works in testing, and makedepend will generate Makefiles of this
form)a but I could be wrong.
Darren Tucker [Tue, 23 Jul 2019 11:51:22 +0000 (21:51 +1000)]
Re-apply portability changes to current sha2.{c,h}.
Rather than attempt to apply 14 years' worth of changes to OpenBSD's sha2
I imported the current versions directly then re-applied the portability
changes. This also allowed re-syncing digest-libc.c against upstream.
Darren Tucker [Mon, 22 Jul 2019 22:11:22 +0000 (08:11 +1000)]
Rename valgrind "errors" to "failures".
When valgrind is enabled, test-exec.sh counts the number of invocations
that valgrind detects failures in, not the total number of errors detected.
This makes the name to be more accurate.
upstream: Remove the sleeps and thus races from the forwarding
test. They were originally required to work with Protocol 1, but now we can
use ssh -N and the control socket without the sleeps. While there, suppress
output fro the control exit commands.
Darren Tucker [Fri, 19 Jul 2019 03:50:25 +0000 (13:50 +1000)]
Revert one dependency per line change.
It turns out that having such a large number of lines in the .depend
file will cause the memory usage of awk during AC_SUBST to blow up on at
least NetBSD's awk, causing configure to fail.
Darren Tucker [Wed, 17 Jul 2019 11:24:55 +0000 (21:24 +1000)]
Put valgrind vgdb files to a specific directory.
Valgrind by default puts vgdb files and pipes under /tmp, however it
is not always able to clean them up, which can cause test failures when
there's a pid/file collision. Using a specific directory ensures that
we can clean up and start clean.
upstream: support PKCS8 as an optional format for storage of
private keys, enabled via "ssh-keygen -m PKCS8" on operations that save
private keys to disk.
The OpenSSH native key format remains the default, but PKCS8 is a
superior format to PEM if interoperability with non-OpenSSH software
is required, as it may use a less terrible KDF (IIRC PEM uses a single
round of MD5 as a KDF).
adapted from patch by Jakub Jelen via bz3013; ok markus
Damien Miller [Mon, 8 Jul 2019 03:38:39 +0000 (13:38 +1000)]
remove realpath() compat replacement
We shipped a BSD implementation of realpath() because sftp-server
depended on its behaviour.
OpenBSD is now moving to a more strictly POSIX-compliant realpath(2),
so sftp-server now unconditionally requires its own BSD-style realpath
implementation. As such, there is no need to carry another independant
implementation in openbsd-compat.
Darren Tucker [Sat, 6 Jul 2019 02:00:41 +0000 (12:00 +1000)]
Add missing bracket in EGD seeding code.
When configured --with-prngd-socket the code had a missing bracket after
an API change. Fix that and a couple of warnings. bz#3032 , from
ole.weidner at protonmail.ch
upstream: Adapt the PuTTY/Conch tests to new key names.
A recent regress change (2a9b3a2ce411d16cda9c79ab713c55f65b0ec257 in
portable) broke the PuTTY and Twisted Conch interop tests, because the
key they want to use is now called ssh-rsa rather than rsa. Adapt the
tests to the new file names. bz#3020, patch from cjwatson at debian.org.
upstream: Add a sleep to allow forwards to come up.
Currently when the multiplex client requests a forward it returns
once the request has been sent but not necessarily when the forward
is up. This causes intermittent text failures due to this race,
so add some sleeps to mitigate this until we can fix it properly.
Darren Tucker [Fri, 5 Jul 2019 03:21:45 +0000 (13:21 +1000)]
Do not fatal on failed lookup of group "tty".
Some platforms (eg AIX and Cygwin) do not have a "tty" group. In those
cases we will fall back to making the tty device the user's primary
group, so do not fatal if the group lookup fails. ok djm@
upstream: When system calls indicate an error they return -1, not
some arbitrary value < 0. errno is only updated in this case. Change all
(most?) callers of syscalls to follow this better, and let's see if this
strictness helps us in the future.
upstream: asprintf returns -1, not an arbitrary value < 0. Also
upon error the (very sloppy specification) leaves an undefined value in *ret,
so it is wrong to inspect it, the error condition is enough. discussed a
little with nicm, and then much more with millert until we were exasperated
projects / thirdparty / openssh-portable.git / log
