OpenBSD and FreeBSD now both use rresvport. This is a nop for
FreeBSD, but for OpenBSD this picks random port numbers.
Submitted by: deraadt@cvs.openbsd.org
fatal: Local: Agent socket bind failed: Address already in use
It would happen when the server tried to create the Unix domain
socket "/tmp/ssh-username/agent-socket-123", if the file already
existed. It could already exist if it happened to be left over from
a system crash. This patch unlinks the file before attempting the
bind operation.
I will send this patch to ssh-bugs@cs.hut.fi too.
1) pw->pw_class was always zero since not copied
2) login_getuserclass() used instead of login_getclass(), so
default class always returned
3) env pointer can be redefined at the moment of setusercontext() call
regenerated them to fix the line numbers. Also, I added two commented out
options in Makefile, one to tell sshd that a group writeable homedir
is OK because all users are in their own group, and the other is to allow
an unencrypted connection (which is dangerous since it can lead to
compromise of keys), but on a secure network it's damn useful for backups
etc.
ignores it's argument (it's meaningless, the kernel keeps the state), but
2.1.x use it. ssh was effectively giving a random port to 2.1.
Originally noticed by: John Polstra <jdp@polstra.com>
reporting bug which happens if the remote end uses tcp_wrappers to control
sshd access (it says something like "read: no such file or directory" or
"read: permission denied" instead of "connection closed"). I already sent it
in to the ssh mailing list.
Submitted by: fenner
ssh-askpass no longer uses wish, so chop the make rules that attempt to
locate it.
Go further to try and protect the ssh_host_key, since it's critical to
the operation and security of the machine.
security problems.
Also re-do the method we use for disconnecting ourselves from the supplied
gmp and z libraries so that this can be maintained in the future (sigh!).
patch-ac: call setsid() before setlogin() in the child (when emulating rsh)
otherwise the setlogin() will fail when/if the proposed setlogin() changes
go in. Otherwise it silently fails and may leave the login name of the user
session as "root" (depending on how sshd was started). Without the proposed
kernel change, it harmlessly sets the login name of the user's session.
patch-ad: patch the #ifdef botch that stopped a ssh login from using and
updating the lastlog file. This is because we have struct lastlog defined
inside utmp.h rather than a lastlog.h include file like it was expecting.
(.rhosts together with RSA based host authentication, and pure RSA
authentication) and improved privacy (all communications are automatically
and transparently encrypted).
