This patch contains:
* checks for malloc return value + treat callers;
* modified malloc() + ZeroMemory() to calloc();
* misc fixes of micro errors seen during the code audit:
** some invalid checks in gcc.c, also there were some possible
integer overflow. This is interesting because at the end the data are parsed
and freed directly, so it's a vulnerability in some kind of dead code (at least
useless);
** fixed usage of GetComputerNameExA with just one call, when 2 were used
in misc places. According to MSDN GetComputerNameA() is supposed to return
an error when called with NULL;
** there were a bug in the command line parsing of shadow;
** in freerdp_dynamic_channel_collection_add() the size of array was multiplied
by 4 instead of 2 on resize
If no entry for the <host> <port> combination
was found in the v2 file and there is a legacy file
check if a matching <host> entry can be found.
In case there is a matching entry and the <fingerprint>
also matches, create a new entry in the v2 file using the
current port.
Microsoft iOS Remote Desktop Clients eventually send NULL-terminated
hostnames in SNI which is not allowed in the OpenSSL implementation.
Since we're not using SNI this commit adds an OpenSSL TLS extension
debug callback which modifies the SSL context in a way preventing it
from parsing this extension
tls_disconnect shut down the ssl stream but didn't inform
the BIO(s) about this therefore could happen that a second shut down
was initiated (e.g. in bio_rdp_tls_free) causing rather long delays.
After removing the shut down from tls_disconnect the only thing the
function does is to prepare/send an alert therefore it was renamed to
tls_send_alert.
Currently the certificate format expected in FreeRDPs certificate store
is DER (ASN1). On most linux/unix systems the system certificate store
default format is PEM. Which is also the more common format used by CAs
to distribute their certificates.
Changing the default format to PEM allows the usage of system
certificates or published CA certificates without the need to convert them.
This fixes a part of issue #2446.
To do this I've swapped _strnicmp with memcmp. Seemless, but does lock it to the restrictions of that function.
Signed-off-by: Jason Plum <jplum@archlinuxarm.org>
The fix in #2130 eliminates the problem when connecting over a gateway
but introduces other problems server side and client side (client/server
can't detect anymore when a TCP connection was closed).
"libfreerdp" consisted of multiple (small) single libraries. If the cmake
option MONOLITHIC was used only one library was build combining all of
the libfreerdp-* libraries.
The only exceptions to this are libfreerdp-server and libfreerdp-client these
are build as separate libraries.
This commit obsoltes non-monolithic builds and makes monolithic builds
the default. The cmake option MONOLITHIC is also removed.
freerdp/winpr had the following issues:
* The non reentrant SSL_library_init() was called concurrently (crash)
* Missing code/api to set the eventually required OpenSSL static and dynamic locking callbacks
* Missing code/api to free the application-global or thread-local OpenSSL data and tables
This commit creates two new winpr functions:
BOOL winpr_InitializeSSL(DWORD flags):
Use the flag WINPR_SSL_INIT_ALREADY_INITIALIZED if you want to tell winpr that
your application has already initialized OpenSSL.
If required use the flag WINPR_SSL_INIT_ENABLE_LOCKING to tell winpr that it
should set the OpenSSL static and dynamic locking callbacks.
Otherwise just call it with the flag WINPR_SSL_INIT_DEFAULT.
The recommended way is that your application calls this function once before
any threads are created. However, in order to support lazy OpenSSL library
initialization winpr_InitializeSSL() can also safely be called multiple times
and concurrently because it uses the new InitOnceExecuteOnce() function to
guarantee that the initialization is only performed successfully once during
the life time of the calling process.
BOOL winpr_CleanupSSL(DWORD flags):
If you create a thread that uses SSL you should call this function before the
thread returns using the flag WINPR_SSL_CLEANUP_THREAD in order to clean up
the thread-local OpenSSL data and tables.
Call the function with the flag WINPR_SSL_CLEANUP_GLOBAL before terminating
your application.
Note: This commit only replaced the current occurences of the
SSL_load_error_strings(); SSL_library_init(); pairs in the freerdp source
with winpr_InitializeSSL(). None of the server or client applications has been
changed according to the recommended usage described above (TBDL).
This adds 2 arguments:
/tls-ciphers List of permitted openssl ciphers - see ciphers(1)
/tls-ciphers-netmon Use tls ciphers that netmon can parse
With KB2919355, client/server negotiate the use of
TLS cipher TLS_ECDHE_RSA_WITH_AES_256_CBC_SHA,
which works fine except that netmon can't parse it.
By adding commandline /tls-ciphers-netmon, we restrict
the available ciphers to a list that netmon can
deal with. Also adds /tls-ciphers, which
accepts a string arg, for further customization.
winpr is now always build as single library.
The build option MONOLITHIC_BUILD doesn't influence this behavior anymore.
The only exception is winpr-makecert-tool which is still build as extra
library.
This obsoletes complex_libraries for winpr.
select() has the major drawback that it cannot handle file descriptor
that are bigger than 1024. This patch makes use of poll() instead of
select() when poll() support is available.
This big patch allows to have non-blocking writes. To achieve
this, it slightly changes the way transport is handled. The misc transport
layers are handled with OpenSSL BIOs. In the chain we insert a
bufferedBIO that will bufferize write calls that couldn't be honored.
For an access with Tls security the BIO chain would look like this:
FreeRdp Code ===> SSL bio ===> buffered BIO ===> socket BIO
The buffered BIO will store bytes that couldn't be send because of
blocking write calls.
This patch also rework TSG so that it would look like this in the
case of SSL security with TSG:
(TSG in)
> SSL BIO => buffered BIO ==> socket BIO
/
FreeRdp => SSL BIO => TSG BIO
\
> SSL BIO => buffered BIO ==> socket BIO
(TSG out)
So from the FreeRDP point of view sending something is only BIO_writing
on the frontBio (last BIO on the left).
This patch adds an option to compile freerdp in a valgrind compliant way.
The purpose is to ease memchecking when connecting with TLS. We mark bytes
retrieved from SSL_read() as plainly defined to prevent the undefined contamination.
With the patch and the option activated you get a single warning at connection
during the handshake, and nothing after.
This patch changes the prototype for decode_base64 so that the encode / decode
method are consistant (encode(BYTE *) => char* and decode(char*) => BYTE*).
It also does some improvements with unrolling loops so that end conditions are
tested only at the end.
The patch also adds some unitary tests.
Before the patch base64_decode() made valgrind complain about uninitialized
bits, after valgrind is happy and very quiet.
