As a client, require all peers (i.e. other IRC servers) to support secure
renegotiation. Break handshakes with servers that don't. We do not
renegotiate our sessions, but this is the most secure option regardless.
As a client, disable TLS Session Tickets. The server side MbedTLS code
does not have any ticket callbacks configured, so an MbedTLS IRC Server
will not issue tickets -- however, others could. Server connections are
not expected to be short-lived enough to benefit from the usage of tickets,
and their issuance harms forward secrecy.
I have removed all non-MbedTLS-specific code from this backend and
copied the non-OpenSSL-specific portions of the OpenSSL backend code
over; as it seems to be more reliable.
If initialising the server context fails, but the client one succeeds,
we will not only leak memory, but the error message reported for
initialising the server context might not make sense, because we
initialise the client context after and that could erase or change the
list of queued errors.
This scenario is considered rare. Nevertheless, we now initialise the
client context after *successfully* initialising the server context.
Commit 5c8da48 introduced a fix for issue #186 by freeing the old SSL_CTX
structure before constructing a new one, which could disconnect existing
clients otherwise.
Unfortunately, the freeing is done first, which means that if setting up
a new structure fails for any reason, there will be no usable structures
left, but they are still referenced.
This fix moves the freeing to the end of the function, using intermediate
new variables in the meantime. This problem was discovered while testing
against OpenSSL 1.1.0 RC6.
LibreSSL does not have the new version macros & functions that OpenSSL
1.1.0 implements. This causes a compile-time failure against LibreSSL.
Further, the runtime function for returning the library version returns
the wrong number (the hardcoded constant number SSLEAY_VERSION_NUMBER
aka OPENSSL_VERSION_NUMBER, instead of LIBRESSL_VERSION_NUMBER).
Add more ifdef soup to remedy the situation.
The code already assumes the presence of fopen(3) and errno, and, by
extension, fclose(3) and strerror(3), so just use those instead of the
BIO wrappers.
Additionally, don't fail to initialise if the DH file does exist but
parsing it fails, as per the pre-existing comment about them being
optional.
* Move certificate, key, DH parameters and configuration to heap
(Documentation states that setting new configuration, e.g.
during a rehash, is unsupported while connections using that
configuration are active)
This is the same approach as the fix for #186
Refcount these structures so as to not introduce a memory leak
On rehash, it will use new structures only if there are no
errors in constructing them
* Add better error-reporting (strings in addition to numbers)
where possible
* Coalesce several connection memory allocations into one function
* Reduce boilerplate where possible (Charybdis targets C99)
* Support private key being in certificate file, and having no
DH parameters file
* Correct erroneous closing comment
openssl:
* Don't manually initialise libssl 1.1.0 -- it does this automatically
* SSL_library_init() should be called first otherwise
* Move SSL_CTX construction to rb_setup_ssl_server()
* Test for all required files (certificate & key) before doing anything
* Free the old CTX before constructing a new one (Fixes#186)
* Properly abort rb_setup_ssl_server() on CTX construction failures
* Support ECDHE on more than one curve on OpenSSL 1.0.2 and above
* Clean up ifdef indentation
* Fix DH parameters memory leak
mbedtls:
* Fix certificate fingerprint generation
* Fix library linking order
* Fix incorrect printf()-esque argument count
* Return digest length for fingerprints instead of 1, consistent
with the other backends
sslproc / ssld:
* Fingerprint methods have no assocated file descriptors
* Send TLS information (cipher, fingerprint) before data
* Use correct header length for fingerprint method
Authored-by: Aaron Jones <aaronmdjones@gmail.com>
Authored-by: William Pitcock <nenolod@dereferenced.org>
Authored-by: Simon Arlott <sa.me.uk>
Nothing in the Charybdis or libratbox code calls rb_get_pseudo_random
and under OpenSSL it uses RAND_pseudo_bytes() which is both dangerous
and removed in OpenSSL version 1.1.0.
When building against current OpenSSL (<= 1.0.2) or old LibreSSL
(< 2.2.2) the server will use TLSv1.0 only when connecting to other
servers.
This patch corrects that.
OpenBSD 5.8 includes LibreSSL 2.2.2, which finally brings the API up to
what they claim it is by implementing the new TLS client and server
method APIs. Therefore, in furtherance of commits a4c8c827 and 1a4e224a
we can build with the new APIs if building against (real) OpenSSL 1.1.0
or LibreSSL 2.2.2.
Reported-by: Juuso Lapinlampi <wub@partyvan.eu>
feature set they support (2.0 even!), deliberately breaking
backward compatibility. Therefore, in order to fix a regression
introduced by commit a4c8c827 with regard to LibreSSL's stupidity,
unconditionally use the old TLS API if building against LibreSSL.
- Implemented changes suggested by Jilles
- Remove some unused parameters in functions
- Remove some unused ssl procs
- 63-bit time_t support in TS deltas
- const char * vs char * cleanup
- struct alignment (void *) casts
- signed vs unsigned fixes
- bad memset() call
- Bad LT_MAIN in libratbox
- char -> unsigned char casts for isdigit/isspace/etc calls
Thanks Jilles!
This is in accordance with RFC 7465
<https://tools.ietf.org/html/rfc7465>
Also correct the key exchange mechanism strings; these should be
prefixed with 'k'.
This fixes a compiler warning. The necessary fields of the struct sigevent
were initialized so there was no problem.
Submitted by: Aaron (via IRC)
Reviewed by: Attila
The last byte of balloc.c's block pointer could be changed from 10 or 13 to
0. On amd64, this is not possible. On i386, this is possible and usually
causes a crash soon.
When receiving bans from a bursting server, if kline_delay is set to 0
(the default), rb_event_addonce will be called to schedule an event for
0 seconds in the future. While this works fine for the fallback
rb_event_run function, the epoll implementation ends up scheduling a
timerfd for the event in the past, which is then never executed.
While fixing this, I also made rb_event_add and rb_event_addonce reject
attempts to add events scheduled for 0 seconds in the future; they're
instead rewritten to run 1 second in the future.
SSL_OP_NO_COMPRESSION was presumably added in an attempt to prevent
information leakage in a manner similar to recent attacks on HTTPS.
However, assuming that IRC is vulnerable to the same class of attacks is
incorrect: the behavior of the IRC protocol (a single long-running
connection) is not the same as that of HTTPS (multiple ephemeral
connections). HTTPS's use of ephemeral connections means that certain
assumptions can be made about the contents of the compression
algorithm's dictionaries and the content exchanged between the client
and server (e.g. the content being nearly the same for each connection),
which is not true for IRC. Additionally, they rely on the attacker being
able to coerce the client into creating many HTTPS connections (and
resending some secret token belonging to the user, along with
attacker-controlled data) each time, none of which is possible with IRC.
Lastly, since compression is no longer performed, this option will
result in leaking the lengths of messages transmitted to and from the
client. This option does reduce CPU utilization on Charybdis servers but
also increases bandwidth consumed.
The C standard does not allow constructing pointers beyond one past the end
of an array. Therefore, if size is an unsigned type (size_t), then
buf + size is never less than buf.
Clang on 32-bit took advantage of the undefined behaviour, causing
segfaults.
Lightly tested.
Note that these are not available in old versions of OpenSSL (like FreeBSD
9.x base OpenSSL), so allow them to be missing.
A side effect may be slightly higher CPU consumption and network traffic.
Without a session id context and if client certificates are used, OpenSSL
fails the handshake if an attempt is made to reuse an old session. Various
clients could not reconnect after a disconnection because of this.
See https://bugzilla.mozilla.org/show_bug.cgi?id=858394#c34 for a bug
report.
rb_crypt() was generating different SHA256 ($5$) hashes than glibc,
making hashes generated with charybdis unusable in ratbox and other
software, and vice versa.
Note that you must have options ACCEPT_FILTER_DATA in your kernel
configuration or load the accf_data kernel module. The functionality is
not in the GENERIC kernel.
This allows for some further hardening against synflooding and connection flooding
where no data is actually sent, as the kernel will simply ignore those connections
(well, as far as the ircd is concerned anyway).
ERR_error_string() is just broken, as it returns at most 119 chars
which means error messages are frequently truncated.
Allow for 511 chars using ERR_error_string_n().
This lets a user connect with a client certificate, and
passes the certificate's fingerprint to ircd, which
currently just notices it to the user.
A new ssld->ircd message 'F' is used to pass on the
fingerprint.
This is only for OpenSSL for now, not GNUTLS.
If there is no space in the output buffer to report an
error adding to the kqueue, kevent(2) will abort and
return the error in errno (I was correct that it does
not tell you where it failed). So do not abort the loop
if kevent(2) fails and do not log (expected) EBADF.
(ircd wouldn't read or write anymore to certain clients)
This happens because kqueue.c will often try to add
already closed file descriptors to the kqueue. The kernel
tries to report bad file descriptors in the eventlist; if
the eventlist has no space, processing of the changelist
is silently halted.
The fix:
1. allocate two kqlst things, one for what kqlst currently does
and one as output buffer
this ensures the kevent(2) call in rb_select_kqueue() never
drops updates
2. replace the kevent(2) call in kq_update_events() by a loop
that processes the updates one at a time
that doesn't happen much, and it's the only way to be sure
without also getting events out of the queue we cannot process
at that time
libratbox r25354 (jilles)
In rare cases, this sharing caused the ircd to skip
part of outgoing traffic, e.g. appearing as "not enough
parameters" errors on the other side.
The purpose of this flag can be fulfilled by the writeofs
in the bufhead.
libratbox r25227
