No need to check ApiMessage.Status if you also check the returned value of CsrClientCallServer because, according to dll/ntdll/csr/connect.c:CsrClientCallServer, the (only) return value of CsrClientCallServer _is_ equal to ApiMessage.Status (vide line 169).
svn path=/branches/ros-csrss/; revision=58456
Remove the old implementation of GetConsoleInputWaitHandle which was commented-out for some time.
[BASESRV]
Correct few names.
svn path=/branches/ros-csrss/; revision=58454
Fix CsrCaptureTimeout (verified against Windows Server 2003).
[KERNEL32-CSRSRV-BASESRV-CONSRV-WINSRV]
Clean the code: remove unuseful comments and dprints, reorganize a little bit few source files.
svn path=/branches/ros-csrss/; revision=58453
Fix the initialization of screen buffers (concerning the cursor size).
[KERNEL32]
Fix almost all of the console winetests. Only 26 tests remain to be fixed, concerning principally WriteConsoleInputA/W and GetNumberOfConsoleInputEvents.
svn path=/branches/ros-csrss/; revision=58448
- Start to sort out things that depends only of the internals of a console, and things which are only related to "terminal emulators". Do it especially for (what I will call starting from now) the "GUI terminal emulator".
- Temporarily deactivate starting "TUI terminals".
- Temporarily break report that the terminal window is held by the current running console application in it (see r58107). This will be fixed later on.
svn path=/branches/ros-csrss/; revision=58447
Fix the console properties dialog, when launching and transmitting console properties. Before, the properties dialog was directly launched by the console server (consrv), running with CSRSS (System) privileges, what constituted a security hole. Now, I create a remote thread in the running process owning the console for launching the properties dialog (thus it has only user privileges, and not System ones anymore). For that purpose, I basically took the technique described in the following paper (Cesar Cerrudo, "Story of a dumb patch", http://www.argeniss.com/research/MSBugPaper.pdf or http://www.scn.rain.com/~neighorn/PDF/MSBugPaper.pdf), where basically the console server shares the console properties via a shared memory section with the console properties dialog dll. The address of the thread which launches the dialog in the context of the console app is given to the console server the same way as we do for the control handler (called e.g. when you press Ctrl-C, etc...)
Of course this is quite hackish, because you have the GUI interface split between the console server and the console properties dialog dll. Something far more elegant would be to put all the GUI thingie into a dedicated dll or exe, running with the same privileges as the console program itself (a kind of console -- or terminal -- emulator).
[CONSOLE.DLL]
Fix retriving / setting colors.c and other things.
[CONSRV.DLL]
- Fix retrieving / setting console properties from the registry (via the HKCU\Console\* keys), via the shell shortcuts (not totally done at the moment, because somebody has to implement properly that thing : http://msdn.microsoft.com/en-us/library/windows/desktop/bb773359(v=vs.85).aspx (NT_CONSOLE_PROPS structure stored as a shortcut data block) (at application launching time), and via the console properties dialog.
- Few DPRINTs removed.
svn path=/branches/ros-csrss/; revision=58415
- Introduce a CONSOLE_PROPS structure to hold console properties such as its title or startup attributes it should have. Initialized at console application launch time. I have to see how could it be possible to merge this structure with some parts of the GUI_CONSOLE_DATA structure and of the ConsoleInfo structure from console.dll (in /dll/cpl).
- Use a helper function to initialize the CONSOLE_PROPS structure in kernel32, at console app start or when calling AllocConsole.
- In BasepInitConsole, do initialization of CONSOLE_PROPS and related only if we are about to launch a console app.
[CONSRV]
- Adapt CONSOLE_CONNECTION_INFO and CONSOLE_ALLOCCONSOLE structures to take into account CONSOLE_PROPS.
- Fix the way we are setting console titles.
- Add experimental support for loading console properties and title and icon from shell links, thanks to the helper LoadShellLinkInfo. However I'm using there Shell COM facility to extract link properties (I could do it "the RAW way", however I would then know the format of link files, that I don't know and that I don't want to mix up with the console code). Therefore I must add dependencies to uuid and ole32 libraries. Note that icons are used in GUI consoles only (use the PrivateExtractIconExW function to retrieve both handles to the big and small icons at the same time).
Part 1/2
[CONSRV]
- Remove an unuseful xxxInitScreenBuffer function in the virtual functions console table.
- In GUI_CONSOLE_DATA structure and related functions, temporarily explicitely mark which members may be used for both GUI and TUI consoles (for a future simplification).
- Add temporary debug prints when we are setting console icon, to see the flow of calls (will be removed just before merging back to trunk).
- Add temporary debug prints in ConSrvInitConsoleScreenBuffer and ConioDeleteScreenBuffer, to figure out how SBs are initialized.
svn path=/branches/ros-csrss/; revision=58305
- Further compactify console allocation / attach code by putting common code into helper functions.
- Now, the default title of a console window is... "ReactOS Console" (and not command prompt since a console window isn't only reserved to command prompts).
svn path=/branches/ros-csrss/; revision=58260
- Add a macro which specifies how much space chars a TAB is.
- Implement BELL ANSI character handling (basic) and add temporary DPRINTs to see who is called (by the way, Beep() functions seems not to work correctly).
svn path=/branches/ros-csrss/; revision=58253
Modify driver acquiring/releasing privilege according to r58233. The corresponding modification in services.c will be done at the next synchronization.
svn path=/branches/ros-csrss/; revision=58234
Clean-up IntReadConsoleOutputCode a little bit.
[CONSRV]
Fix a buffer overflow in SrvReadConsoleOutputString, which was translated sometimes into heap corruption and assert, triggered when freeing a remote captured buffer in csrsrv, when executing kernel32_winetest console, just during a call to ReadConsoleOutputCharacterA.
Nevertheless I still keep the culprit code (commented-out now) because it might be useful in the future.
svn path=/branches/ros-csrss/; revision=58229
Rename functions to their new scheme: ConSrv* instead of Win32Csr* or Conio* (with some exceptions for the latter case).
No code changes.
svn path=/branches/ros-csrss/; revision=58213
- Start to implement waiting threads notifications. They can be notified by:
* the required action (screen displaying lock/unlock, presence of new characters in the input buffer),
* the fact that the application terminates,
* we close a (input) handle.
TODO: When we notify them by calling CsrNotifyWait, they are not dereferenced automatically, but they can be
by calling a dedicated CSR function. We need to know where it is the best to dereference them.
- Correct a bug introduced in revision r58191, which broke console input mode changes (in SrvSetConsoleMode and SrvGetConsoleMode).
svn path=/branches/ros-csrss/; revision=58211
- Start to uniformize some helpers' names :
ConioLockConsole --> ConioGetConsole and ConioUnlockConsole --> ConioReleaseConsole
ConioLockInputBuffer --> ConioGetInputBuffer and ConioUnlockInputBuffer --> ConioReleaseInputBuffer
ConioLockScreenBuffer --> ConioGetScreenBuffer and ConioUnlockScreenBuffer --> ConioReleaseScreenBuffer
but the overall naming must be revamped.
- Add them a boolean parameter to determine whether or not we should lock the console. It is useful to not lock an already locked console when input / output waiting threads are called,
because these are called from functions which lock the console.
- In the GUI console window message loop, after retrieving a pointer to the corresponding console, lock it before calling funtions in the message dispatcher (and unlock the console at the end).
- Use PostMessage instead of SendMessage for "sending" GUI messages to console windows, to avoid possible deadlocks (because of console locking).
- Add temporary /// LOCK /// comments to easily localize where I deactivated extra console (un)locking code (this remains to be analyzed further).
svn path=/branches/ros-csrss/; revision=58205
- Get in a simpler way the console owned by a console window: replace GuiConsoleGetDataPointers by GuiGetWindowConsole.
- Only in functions which need it, we get the GUI data held by the console.
svn path=/branches/ros-csrss/; revision=58194
- Isolate the members of the CONSOLE structure which are used as input buffer data, in a new structure called ... CONSOLE_INPUT_BUFFER !
- Rename some helper functions to better match what they are doing.
svn path=/branches/ros-csrss/; revision=58191
Introduce new helper functions Win32CsrInitHandlesTable and Win32CsrAllocateConsole and use them in many places (in ConsoleConnect, SrvAllocConsole and SrvAttachConsole).
svn path=/branches/ros-csrss/; revision=58184
- Temporary make the property dialog working again (but it's not the ultimate solution).
- in conio.h : members reorganization only.
svn path=/branches/ros-csrss/; revision=58177
- Finish to do the replacements HeapAlloc --> RtlAllocateHeap and Co.
- Finish to UNICODify consrv (it happens that it has been almost done already).
- Modify the way we start BlueScreen driver (in TUI mode): instead of using the SCM which, at that moment, is starting (and is not ready to answer to start/stop services requests), use the NT apis to start the driver (in the same way the SCM performs driver starting).
- Add a note on HAVE_WMEMSET.
svn path=/branches/ros-csrss/; revision=58169
- Correct a error return code (in AllocConsole).
- Implement AttachConsole (inspired by AllocConsole).
[CONSRV]
- Correct a error return code (in AllocConsole).
- Introduce helper functions to simplify console-initialization code.
- Implement server-side of AttachConsole (I used pieces of code from ConsoleNewProcess, ConsoleConnect and SrvAllocConsole).
Some debug prints were introduced (for debug-purposes :P), will be removed when all things work.
TODO: have a look on the handles code.
svn path=/branches/ros-csrss/; revision=58166
Rename some structures (CSRSS_... ---> CONSOLE_...) and clarify what the ConsoleHandle members are for (--> OutputHandle or InputHandle).
No code changes otherwise.
svn path=/branches/ros-csrss/; revision=58119
- Start to implement NtUserConsoleControl to allow the console server define Console Window Class atom.
- In UserInitialize, initialize the current thread by a call to UserCreateThreadInfo.
- In UserCreateThreadInfo, set the TIF_CSRSSTHREAD flag for threads created by CSRSS.
- In NtUserQueryWindow, when querying the process/thread ID associated to a given window, check whether the window was created by CSRSS and if so, if this is a Console window, get the process/thread ID of the associated console process (console leader process), instead of the CID of CSRSS.
[CONSRV]
- Introduce the concept of console leader process (to be used as the process owning the console window).
- Introduce extra console window information to store the PID and TID of the console leader process via new internal flags to be used by GetWindowLong.
- Effectively set the console window console leader CID at its creation.
- Improve CsrInitConsole to set a console leader process to the console being created.
This fixes http://jira.reactos.org/browse/CORE-122 in this branch.
[CONSRV-WIN32K]
Fix various code misspellings.
svn path=/branches/ros-csrss/; revision=58107
- Fix console apps initialization.
- Add some debug output (NOTE TO MYSELF: remove them when all things work).
- Rewrite ConsoleNewProcess.
- Reorganize SrvAllocConsole and create ConsoleConnect based on SrvAllocConsole.
- Create ConsoleDisconnect which undoes what ConsoleConnect and ConsoleNewProcess did.
- Rework a little bit CsrInitConsole.
Now the console app. initialization algorithm is the following:
1- A process is created, its type (GUI or CUI) is determined (kernel32 and basesrv).
2- ConsoleNewProcess is called (consrv) and makes this new process inherit the console handles table from its parent
(NOTE: this is done for all CUI processes, because at this point, we still don't know whether we must inherit
the handles from the parent or not).
3- (back in kernel32) In BasepInitConsole, we determine whether or not we must create a new console window or use
the parent's one or not using one at all. We (as a client) connect to the console server (consrv) (via CsrClientConnectToServer)
which in turn (via CSRSS mechanism) calls ConsoleConnect. For GUI processes we do nothing. For CUI processes, we initialize
a new console based on properties set in BasepInitConsole.
4- When a process dies, ConsoleDisconnect is called and whether it is a GUI or CUI process, we revert the actions done previously.
Part 2/2
TODO: - Debug the CSR waits.
- Work on the console properties property-sheet.
- See what can be done on http://jira.reactos.org/browse/CORE-122
svn path=/branches/ros-csrss/; revision=58098
- Code reorganization.
- In particular, move IsConsoleHandle macro to a place which it can be used by both kernel32 and consrv.
svn path=/branches/ros-csrss/; revision=58063
- Delete all the remnants of Win32CsrValidateBuffer calls, which were replaced by standard CsrValidateMessageBuffer calls, and therefore delete the unused server.c file.
- Adapt WriteConsole API to use CSR waits. This replaces the old event-based waiting.
TODO: Dereference all the waits in Console->WriteWaitQueue.
svn path=/branches/ros-csrss/; revision=57819
- Simplify IntGetConsoleInput and IntReadConsole.
[CONSRV]
- Implement event waiting for reading console input buffer with CSR wait blocks. This replaces the event-based waiting which, after being moved from client-side (kernel32) to server-side (see r57721), made all consoles hang when one of them was waiting for new input.
TODO: Dereference all the waits in Console->ReadWaitQueue.
svn path=/branches/ros-csrss/; revision=57817
