if QP passes in tainted data, such as a hostname that subsequently gets used to open a connection using IO::Socket, the plugin die because the information is tainted. Fix it once here, instead of in each plugin.
is_immune tests designates to plugins they should always skip processing.
That's typical for naughty connections, but this change provides the ability to handly naughty connections differently than (whitelisted/relayclients/known good) senders.
append optional log_mess to log entry (more description)
subsequent attempts to set naughty don't overwrite the first
set the naughty rejection type to be the reject type of the plugin that marked the connection naughty
get_reject_type can be passed an explicit default
Plugins can now use a 'loglevel' argument in config/plugins entry
Includes user instructions prepended to docs/logging.pod
Already works for all plugins that use named arguments
These 3 auth plugins all have a data store they fetch the reference
password or hash from. They then match the attemped password or hash
against the reference. This consolidates the latter portion (validating
the password/hash) into Auth.pm.
* less duplicated code in the plugins.
* Pass validation consistently handled for these 3 plugins.
* less work to create new auth plugins
Also caches the CRAM-MD5 ticket. It could also cache user/pass info if
this was desirable.
removed auth method from return calls in all auth plugins. The caller knows the mechanism already. In the code, the difference looks like this:
before:
or return (DENY, "authcvm/$method");
after:
or return (DENY, "authcvm");
Added debug level log entries in auth_vpopmaild
Conflicts:
plugins/auth/auth_vpopmail_sql
unit tests for new methods are in t/auth.t
added PLAIN and LOGIN tests in auth_flat_file
Most tests are disabled unless an interactive terminal is detected and $ENV{QPSMTPD_DEVELOPER} is set.
so log entries look like this:
86553 (connect) ident::geoip: US, United States
86553 (connect) ident::p0f: Windows 7 or 8
86553 (connect) check_earlytalker: remote host said nothing spontaneous, proceeding
instead of this:
86553 ident::geoip: (connect): US, United States
86553 ident::p0f: (connect) Windows 7 or 8
86553 check_earlytalker: (connect): remote host said nothing spontaneous, proceeding
Conflicts:
plugins/logging/warn
We have to return something true, else the client gets an additional
451 Internal error - try again later - command 'helo' failed unexpectedly
after the plugin's ->respond() message.
The current postfix-queue plugin allows the administrator to set a
single path to a local postfix cleanup socket file from the plugin
'command line'. This adds a 'cleanup_sockets' configuration directive
that can contain a list of paths as well as host/port combinations
pointing to postfix cleanup services, which will be tried in the order
that they appear. Not yet tested.
This makes hook_reset_transaction a bit more useful by running it on
disconnections. It also ensures that hook_disconnect runs when the
client disconnects without QUITing.
Update Qpsmtpd::Connection::notes() and Qpsmtpd::Transaction::notes()
with clearer, more brief syntax. A previous patch used seemingly better
syntax, but did not evaluate the size of @_ in order to distinguish
between a call to notes('arg1',undef) and notes('arg1'). This corrects
this issue, and adds a comment to that effect.
(patch remade against latest rspier/qpsmtpd)
added remote_port, local_ip, local_port, and local_host to $qp->connection, as the p0f plugin relies on it.
added notes to TcpServer.pm and the p0f plugin noting the dependence, and the lack of support for models other than tcpserver.
Signed-off-by: Robert <rspier@pobox.com>
Qpsmtpd::config now checks cache, hooks, qmail_config, default in this
order and returns the first match. In any case the returned value is
stored in the cache, so subsequent calls to Qpsmtpd::config return the
same value (unless the cache is cleared).
Trying to get SMTP auth working with alpine, I came across a bug.
Alpine sends AUTH PLAIN and waits for a 334 response, then sends the
auth string. According to the RFC, the server should reply with 334
and a nothing else, but in Auth.pm qpsmtpd responds with "334 Please
continue." the "Please continue" is interpreted as a non-zero
length initial challenge which causes alpine (and maybe other
clients?) to abort the session.
Signed-off-by: Charlie Brady <charlieb@budge.apana.org.au>
Signed-off-by: Robert <rspier@pobox.com>
Qpsmtpd::Transaction::body_filename() calls $self->body_spool() if the
message body has not already been spool to disk. This adds the same
check to Qpsmtpd::Transaction::body_fh()
It wasn't suitable to dup the body FH returned by plain old
body_fh(), so here's a separate function to return that.
Signed-off-by: Ask Bjørn Hansen <ask@develooper.com>
Improve Qpsmtpd::Transaction::add_recipient syntax
Update Qpsmtpd::Transaction::add_recipient to use slightly clearer
language, and add 'if $rcpt' to prevent undef recipients from
being added -- in this case, the '@_ and' syntax allowing this set
method to set undef is undesirable, since you shouldn't be adding undef as
a recipient.
Signed-off-by: Ask Bjørn Hansen <ask@develooper.com>
Qpsmtpd::Connection::notes() and Qpsmtpd::Transaction::notes() have
identicaly functionality; however, the POD documentation for Connection
notes is fairly brief while the documentation for Transaction notes is
longer and more helpful. This updates the Connection notes documention
to be consistent with its Transaction counterpart.
Qpsmtpd Connection and Transaction objects support notes() methods which
are conventionally used to pass data between plugins. This adds the
same functionality to Address objects. This should make it easier for
plugins to begin supporting message handling based on per-user
configuration directives.
Signed-off-by: Ask Bjørn Hansen <ask@develooper.com>
This reverts commit ea86b9fdb2.
Jared said...
I originally considered these functionally identical, but they are
not. The new code, called with, say, $txn->notes('discard',undef),
would result in evaluation as if it were a 'get' method rather than
setting the 'discard' note to undef. That seems quite dangerous. I
suggest either reverting the language back to the '@_ and' model, or
else doing something like:
Hook after receiving all headers lines. Defaults to nothing,
just continue processing. At this step, sender does not wait for a reply,
but we can stop him from sending remaining data by disconnecting.
(Cleaned up by Robert for english and coding style.)
Signed-off-by: Ask Bjørn Hansen <ask@develooper.com>
Signed-off-by: Robert Spier <robert@perl.org>
I've got many non-standard plugins and for a given
fatal error it is non-trivial to determine which of
them was to blame.
Perhaps this patch would be useful? (Against SVN,
not git, but I think that shouldn't matter too much..)
Steve
--
Stop blog&forum spam
http://blogspam.net/
The code feeds the results of $session->config('me') to
sprintf as part of the format string. In practice, this is
probably not a problem since hostnames don't contain percent
signs. However, it triggers a taint warning in perl 5.10,
making cram-md5 auth unusable.
This patch rewrites the sprintf to insert the 'me' value
using a %s format specifier.
(always worked on the second call due to the cache, but would return undef on the first call)
git-svn-id: https://svn.perl.org/qpsmtpd/trunk@955 958fd67b-6ff1-0310-b445-bb7760255be9
Support case where no connect hook is defined (thus continue_read() never gets called)
git-svn-id: https://svn.perl.org/qpsmtpd/trunk@882 958fd67b-6ff1-0310-b445-bb7760255be9
Before this, the SMTP responses contained the reference stringification
instead of the real message, when a plugin denied the mail at the DATA
stage:
554 ARRAY(0x9362f10)
git-svn-id: https://svn.perl.org/qpsmtpd/trunk@881 958fd67b-6ff1-0310-b445-bb7760255be9
Previous fix does not work for other cases, for example: packet ends
with CRLFdot, next packet starts with CRLF.
Danga::Client will send now full text lines to the callback.
git-svn-id: https://svn.perl.org/qpsmtpd/trunk@878 958fd67b-6ff1-0310-b445-bb7760255be9
... now check_earlytalker can be expanded to VRFY and NOOP (see RFC 1854, #2.1)
git-svn-id: https://svn.perl.org/qpsmtpd/trunk@836 958fd67b-6ff1-0310-b445-bb7760255be9
- add back in after 0.42 is out? if yes: start implementing in -prefork
git-svn-id: https://svn.perl.org/qpsmtpd/trunk@809 958fd67b-6ff1-0310-b445-bb7760255be9
