- add addr_is_ipv4 helper function
- add addr_is_ipv6 helper function
- add get_socket_family helper function
- g_tcp_connect:
- #define for MAX_PORT_STR length
- create *addr_arg char that is used in getaddrinfo so we can do any last-minute changes directly
- add chars host and service which are primarily used for debug logging of attempted connection points
- calls get_socket_family to discover family of socket
- add switch case logic for AF_INET6/4:
- v4: just calls addr_is_ipv6 so we don't fail on a hostname input
- v6: drop addrconfig flag and add AI_ALL so we can try to map ipv4 destinations against ipv6 socket to support varied configurations we could see
- if an ipv4 address comes in against an ipv6 socket, we map it to try and connect anyways
- getaddrinfo now calls addr_arg so it picks up changes that an ipv6-case may have done
- before calling connect, use getnameinfo to get logging values for the actual host ip/port we are about to attempt a connect on - I found this useful when debugging so thought it had value to keep
- change if (res > -1) to if (res == 0): the bsd man pages for getaddrinfo only promise that it returns 0 on success, so it seems sensible to me to cover the event that an error code could be positive, which freebsd has some positive EAI_* errors based on this: https://github.com/freebsd/freebsd-src/blob/main/lib/libc/net/gai_strerror.c
- remove connect_loopback entirely: this had several bits of logic handling ipv6 and ipv4 mixing already and ended up being redundant because of the restructuring of g_tcp_connect, which covers direct ip address char* inputs now
- add comment about OSX to IPv6 function for clarity
These Coverity warnings all relate to the user of g_setenv() where the
return result isn't checked.
An additional void function g_setenv_log() is provided which logs
failures to set environment variables, and returns no status. This is
used in all the places where g_setenv_is currently called.
cppcheck 2.17.0 adds checks that a NULL pointer returned from malloc() and
calloc() is not used.
We do this quite a lot.
I've addressed this by adding functions g_malloc_nofail() and
g_calloc_nofail() which either allocate memory or abort.
functions are now called in places where we are not making these
checks.
Many of these checks are in test programs or example programs.
I've modified the list16 module to handle out-of-memory conditions.
The function as specified used gettimeofday() which is susceptible
to manual time changes, and is obsoleted in POSIX.1-2008. The
replacement uses clock_gettime(CLOCK_MONOTONIC, ) which is not
susceptible to manual time changes (at least on Linux) and cannot run
backwards.
Also, on systems with 32-bit integers, the value returned by this
function wraps around every 49.7 days. To cope with a wraparound in
a way compliant with the C standard, this value needs to return an
unsigned integer type rather than a signed integer type.
This is not year 2038 compliant on systems with 32-bit integers.
The call can be replaced with the standard C time() call. On
POSIX systems, time_t is guaranteed to be an integer type.
If xrdp is running with dropped privileges it won't be able to delete
the PID file it's created. Places where xrdp is stopped need to cater
for this.
It's prefereable to do this than make the PID file writeable by xrdp
with dropped privileges, as this can still lead to DoS attacks if an
attacker manages to modify the PID file from a compromised xrdp
process.
When xrdp is built with IPv6 support it will only fall back to IPv4 if
IPv6 is not supported (EAFNOSUPPORT). However, if the system is IPv6
capable but doesn't have an IPv6 address set (at least inside a FreeBSD
jail) EPROTONOSUPPORT is returned from socket().
* GFX: sort versions, flags to return the highest version we support in caps advertise
* GFX: simpify swtich in caps_advertise
* GFX: log skipped capability versions in caps_advertise
On Linux, the TCP send buffer size is increased to 32768 if it is less
that this (which it normally is). This however has the effect of disabling
dynamic buffer sizing, leading to a maximum available bandwidth of
max_bandwidth = 262144 (bits) / round_trip_time (secs)
This is not noticeable on a LAN with an RTT of around 0.5ms, but
very noticeable on a WAN with an RTT of 0.25s.
Comments in the config file and manpage in this area are improved, as
is the logging if the parameters are actually set.
- Use clearenv() if it exists
- Don't rely on <limits.h> being pulled in by <sys/param.h>
- Rename the DEFAULT_TYPE macro in sesrun.c. This name appears to be
used on Solaris. It's not a good choice.
The sockdir is only used when sesman is active. The
call g_mk_socket_path() is removed from os_calls and moved to
sesman.
We also change the permissions on this directory to
0755 rather than 01777 (01000 is the 'sticky bit', S_ISVTX).
The behaviour of g_create_dir() has been modified to not
set S_ISVTX on Linux directories. This is implementation-defined
behaviour according to 1003.1, and is no longer required for the
sockdir.
These calls are now replaced with explicit UTF conversion routines in
the common/string_calls.[hc] and common/parse.[hc] modules.
Also removed:-
- The support code in common/os_calls.c to set the locale to use
these routines.
- The twchar type in arch.h
On enterprise systems, using getgrouplist() (if available)
is more efficient than iterating over the members of the group,
and is also more likely to work
Rename g_file_open() to g_file_open_rw(), and add a new g_file_open_ro()
call that wraps the common g_file_open_ex(file, 1, 0, 0, 0) idiom. This
will make the file access mode more explicit in the code.
Change all calls to g_file_open() to the _ro() or _rw() variant as
appropriate, and replace g_file_open_ex(file, 1, 0, 0, 0) with the _ro()
call.
Lastly, add tests for the two new calls to test_os_calls.c (code
courteously provided by matt335672).
Matt Cunningham