This is a regression introduced in v0.10.x
This version introduced a state machine to handle resizes requested
by the client and the server. Most configurations support resizeable
sessions, but one that doesn't is xrdp connecting to x11vnc on (e.g.) a
Raspberry PI.
If the session size requested by a client is differnt from the x11vnc
size, an error is logged and the state machine fails to complete,
resulting in a black screen.
This PR handles the problem by queueing a resize to the supported
server size and then continuing with the state machine. It's not an
optimal solution, but involves the least change to v0.10.x code.
(cherry picked from commit 984b71449e)
Version 2.14.0 of cppcheck generates errors relating to the
check level (e.g.):-
common/base64.c:0:0: information: Limiting analysis of branches. Use --check-level=exhaustive to analyze all branches. [normalCheckLevelMaxBranches]
This does not happen with the same sources (commit
f781962a55) under 2.13.0.
This PR disables the warnings above for 2.14.0, but also allows a '-f'
argument to be passed in to request an exhaustive test. This could be used
(for example) before a major release. An exhaustive test takes a *lot*
longer. The first run with a git runner was around an hour.
The --check-level=flag was only added for 2.11.0, and so this now needs
a version check.
(cherry picked from commit 82c95faaa9)
At the time of writing, the github runner for ubuntu-latest is
22.04, but this will change to 24.04 later in the year.
This PR fixes the CI runner version at 22.04 for v0.10.x
When used with a FreeRDP client on Linux, a file copy operation from
the clipboard detects end-of-file by a read returning 0 bytes. This is
currently marked as an error.
It is assumed that mstsc.exe detects end-of-file in another way, which
is why this has not been found before.
(cherry picked from commit 0f6e731524)
The routine clipboard_get_files() parses a potentially long string,
and copies portions of it into a temporary buffer. This buffer is then
passed to clipboard_get_file() as pointer + length;
The buffer is inadequately sized for very long filenames which may
approach XFS_MAXFILENAMELEN in length. This can cause chansrv to fail
when the user copies such filenames.
It turns out the buffer is unnecessary, as the filenames can be
passed directly into clipboard_get_file() from the source string,
using pointer + length. This avoids the length limitation entirely.
(cherry picked from commit 34b5582460)
The limit of 256 characters for clipboard files is limiting for
many Asian locales, particularly as '%xx' notation is used to
communicate bytes with bit 7 set.
(cherry picked from commit a90228241d)
Replace the 256 byte buffer used for names in the XFS filesystem with a
dynamically allocated buffer.
The define XFS_MAXFILENAMELEN which used to be 255 has been retained,
but bumped to 1023. This value is no longer used for long-lived
allocations, but is used in chansrv_fuse.c for maintaining state
information for in-fligh I/O requests.
(cherry picked from commit d8b5435710)
These config files are intended to be substituted during the build
process. The substituted .ini files should not be included in release
tarballs.
Fixes: #3187
This commit changes the license response PDU to be constructed rather
than simply being contained as a binary blob.
Some constants in common/ms-rdpbcgr.h are renamed with the values
from the specification.
(cherry picked from commit 52dd88b576)
Replaces the existing licensing exchange with a single PDU
saying the user will not issue a license.
This is necessary for clients on FIPS-compliant systems, as these
are unable to decode the licensing exchange packets, due to outdated
cyphers.
(cherry picked from commit cc4a4c95f2)
The state buffers used by the following structs in chansrv_fuse.c
are one byte too small for filenames of length XFS_MAXFILENAMELEN:-
- struct state_lookup
- struct state_create
- struct state_rename
In practice, there is no runtime danger, as XFS_MAXFILENAMELEN is 255,
and these buffers will be followed by non-byte aligned data. Nevertheless
this should be fixed to prevent problems if the value is changed.
(cherry picked from commit c9e84dc16c)
struct pre_session_item has an entry for the start_ip_addr which is not
being filled in. This is not normally needed, as the IP address of the
session is passed into the session another way, but it is needed if the
session selection Policy contains the 'I' selector.
(cherry picked from commit a4f57572e6)
If the setting require_credentials is true, there should be no way
for the user to get to a login screen.
This commit makes the following changes if this flag is active:-
- Makes the checks around TS_INFO_PACKET more explicit.
- Closes the connection if the first login attempt fails.
(cherry picked from commit 8ac2f6db34)
when a multi-monitor session has the top-left vertex of the primary
monitor at a desktop location other than (0,0), minimising and maximising
the session results in the (0,0) co-ordinate of the entire desktop being
placed at the top-left of the primary monitor.
The implementation seems to be at odds with [MS-RDPEGFX] 2.2.2.14 which
suggests the monitorDefArray of the RDPGFX_RESET_GRAPHICS_PDU should be
the same as that in the Monitor Layout PDU ([MS-RDPBCGR] 2.2.12.1)
(cherry picked from commit 095f0d0e4c)
In the words of @iskunk
It is no longer possible to refer to the Dvorak layout as just "dvorak"
(as when one would run "setxkbmap dvorak"); one must now use either
"us dvorak" or "us(dvorak)"
See https://bugs.debian.org/1063725
(cherry picked from commit a1b7c17906)
On a resize, the encoder is deleted. At present this is done by asking
the encoder to exit, and then waiting a second.
- On slower systems, a second may not be enough, and so the encoder
data structures are freed while they are still being used by the
encoder.
- On quicker systems, resizes are delayed by hundreds of milliseconds
longer than they need to be.
This commit adds a wait object which the encoder can use to signal it
has actually finished.
(cherry picked from commit 985b0de35e)
The xrdp_enc_data contains a union for handling surface commands
and gfx commands. Memory processing is different for these two
options.
The default destructor for the encoder FIFO only knows about surface
commands. Consequently, if the encoder has queued GFX data when the
encoder is closed, the destructor processes the queued data as if
it contained surface commands rather than GFX commands. This typically
causes a SEGV as the drects field of the overlaid surface command
structure is not pointing at anything valid when it is freed.
(cherry picked from commit 809df89c08)
The get_sorted_session_displays() is broken in that it
doesn't produce a sorted list of displays.
The problem is the qsort comparison function which has 2 errors in 4 lines:-
1) The test is the wrong way round (i.e. arg1 < arg2 produces a +ve
result instead of -ve)
2) Subtracting two unsigned ints in C will never return < 0
The broken function has been masked by other display checks which mean
that it is only visible in a few situations:-
1) Starting two sessions very closely to each other may allocate the
same display to both sessions.
2) If /tmp is namespaced, the other display checks do not work, and
more than two sessions cannot be started.
(cherry picked from commit 70f1b685ba)
matt335672