Avoid buffer overflow caused by sws_scale():
http://trac.ffmpeg.org/ticket/9254
Currently (ffmpeg-4.4 with SSE3 enabled) sws_scale()
has a few requirements for target data buffers on rescaling:
1. buffer has to be aligned to be usable for SIMD instructions
2. buffer has to be padded to allow small overflow by SIMD instructions
Unfortunately SDL_Surface does not provide these guarantees.
This means that atempt to rescale directly into SDL surface causes
memory corruption. Usually it happens on campaign selection screen
where short video moves start spinning on mouse hover.
To fix [1.] we use av_malloc() for memory allocation.
To fix [2.] we add an `ffmpeg_pad` that provides plenty of space.
We have to use intermdiate buffer and then use memcpy() to land it
to SDL_Surface.
Without the change crash has the following backtrace:
```
(gdb) bt
(c=0x47508940, src=0x1ffeffef50, srcStride=0x1ffeffef30, srcSliceY=0, srcSliceH=116, dst=0x1ffeffef70, dstStride=0x1ffeffef40) at src/libswscale/x86/yuv2rgb_template.c:119
(c=<optimized out>, srcSlice=<optimized out>, srcStride=0x432afa20, srcSliceY=<optimized out>, srcSliceH=116, dst=<optimized out>, dstStride=0x1ffefff0a0) at src/libswscale/swscale.c:969
(this=0x1abaa330, x=90, y=72, dst=0x1a85a4c0, forceRedraw=<optimized out>, update=<optimized out>)
at ../vcmi-9999/client/CVideoHandler.cpp:332
at ../vcmi-9999/client/gui/CIntObject.cpp:83
at ../vcmi-9999/client/gui/CGuiHandler.cpp:462
```
valgrind points to corruption right in sws_scale():
```
Invalid write of size 8
at 0x6C50BD3: ??? (in /usr/lib64/libswscale.so.5.7.100)
by 0x6C4FAE6: yuv420_rgb32_ssse3 (yuv2rgb_template.c:119)
by 0x6C28DF2: sws_scale (swscale.c:969)
by 0x4566F6: CVideoPlayer::nextFrame() (CVideoHandler.cpp:293)
by 0x4573A6: CVideoPlayer::update(int, int, SDL_Surface*, bool, bool) (CVideoHandler.cpp:332)
by 0x25EC94: CIntObject::show(SDL_Surface*) [clone .part.0] (CIntObject.cpp:83)
by 0x34E855: CMainMenu::update() (CMainMenu.cpp:319)
by 0x25D589: CGuiHandler::renderFrame() (CGuiHandler.cpp:462)
by 0x1F7450: mainLoop (CMT.cpp:1387)
by 0x1F7450: main (CMT.cpp:513)
Address 0x475088a8 is 0 bytes after a block of size 92,840 alloc'd
at 0x483F7E5: malloc (vg_replace_malloc.c:380)
by 0x52B4E23: SDL_malloc_REAL (SDL_malloc.c:5387)
by 0x5266237: SDL_SIMDAlloc_REAL (SDL_cpuinfo.c:963)
by 0x52EF042: SDL_CreateRGBSurfaceWithFormat_REAL (SDL_surface.c:123)
by 0x2649AC: CSDL_Ext::newSurface(int, int, SDL_Surface*) (SDL_Extensions.cpp:42)
by 0x457B20: CVideoPlayer::open(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, bool, bool) (CVideoHandler.cpp:182)
by 0x457C60: CVideoPlayer::open(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool) (CVideoHandler.cpp:84)
by 0x35B14E: CCampaignScreen::CCampaignButton::show(SDL_Surface*) (CCampaignScreen.cpp:126)
by 0x25EC94: CIntObject::show(SDL_Surface*) [clone .part.0] (CIntObject.cpp:83)
by 0x34E855: CMainMenu::update() (CMainMenu.cpp:319)
by 0x25D589: CGuiHandler::renderFrame() (CGuiHandler.cpp:462)
by 0x1F7450: mainLoop (CMT.cpp:1387)
by 0x1F7450: main (CMT.cpp:513)
```
Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
Before the change campaign bonus selection screen had inconsistent
overflow behaviour for difficulty selection:
```
0 1 2 3 4 | available buttons: '-' to decrease
^ | '+' to increase
```
Before the change:
1. If we click '+' 5 times we will end up on difficulty=4 (ok, saturated).
2. If we click '-' 5 times we will end up on difficulty=1 (unexpected, wrap around).
After the change:
1. If we click '+' 5 times we will end up on difficulty=4 (saturated).
2. If we click '-' 5 times we will end up on difficulty=0 (saturated).
The inconsistency happens because `difficulty` variable has `ui8` type
and server uses `difficulty = vstd::abetween(difficulty, 0, 4)` to
implement the saturation.
For large positive values saturation works as expected:
vstd::abetween(difficulty=5, 0, 4) -> 4
For small values it does not:
vstd::abetween(difficulty=-1, 0, 4) -> 4
The change makes client to avoid using negative values.
Noticed use of uninitialized value when built vcmi with -fsanitize=undefined:
```
Established connection with
VCMI 0.99 b310f2e61e (server).
UUID: bab9a90d-7416-4566-8817-e367ffcac7c1
../vcmi-9999/client/lobby/SelectionTab.cpp:138:16:
runtime error: load of value 32717, which is not a valid value for type 'ESortBy'
/usr/lib/gcc/x86_64-pc-linux-gnu/12.0.0/include/g++-v12/tuple:190:4:
runtime error: load of value 32717, which is not a valid value for type 'ESortBy'
/usr/lib/gcc/x86_64-pc-linux-gnu/12.0.0/include/g++-v12/tuple:190:4:
runtime error: load of value 32717, which is not a valid value for type 'ESortBy'
```
Here (before the change) `SelectionTab()` used `generalSortingBy` before
first use:
```
SelectionTab::SelectionTab(ESelectionScreen Type) {
...
if(tabType != ESelectionScreen::campaignList)
{
...
ESortBy criteria = (ESortBy)i;
if(criteria == _name)
criteria = generalSortingBy;
buttonsSortBy.push_back(... std::bind(&SelectionTab::sortBy, this, criteria)));
...
}
...
switch(tabType)
{
case ESelectionScreen::newGame:
generalSortingBy = ESortBy::_name;
....
```
The change moves `generalSortingBy` initialization before first use.
Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
* Made most Handlers derived from CHandlerBase and moved service API there.
* Declared existing Entity APIs.
* Added basic script context caching
* Started Lua script module
* Started Lua spell effect API
* Started script state persistence
* Started battle info callback binding
* CommitPackage removed
* Extracted spells::Caster to own header; Expanded Spell API.
* implemented !!MC:S, !!FU:E, !!FU:P, !!MA, !!VR:H, !!VR:C
* !!BU:C, !!BU:E, !!BU:G, !!BU:M implemented
* Allow use of "MC:S@varName@" to declare normal variable (technically v-variable with string key)
* Re-enabled VERM macros.
* !?GM0 added
* !?TM implemented
* Added !!MF:N
* Started !?OB, !!BM, !!HE, !!OW, !!UN
* Added basic support of w-variables
* Added support for ERM indirect variables
* Made !?FU regular trigger
* !!re (ERA loop receiver) implemented
* Fixed ERM receivers with zero args.
Disabled the following (for MSVC only) that couldn't (or shouldn't) be fixed.
4003: not enough actual parameters for macro 'identifier'
4250: 'class1' : inherits 'class2::member' via dominance
4251: 'type' : class 'type1' needs to have dll-interface to be used by clients of class 'type2'
4275: non dll-interface class 'type1' used as base for dll-interface class 'type2'
Replaced const TBonusListPtr with TConstBonusListPtr where necessary
Replaced const std::shared_ptr<T> with std::shared_ptr<const T> where necessary.
Removed superfluous use of const.
Replaced const std::shared_ptr<T> with const std::shared_ptr<T> & in function parameters and ranged for-loops.
Scenario: In LoadGame Lobby screen, the difficulty toggle group is showing multiple choices, but actually only one valid is working.
The reason is, in the Lobby screen Initialization code will set the difficulty = 0, and each time the player changes a map in the selection, the toggle group control is not resetting the difficulty buttons to disabled state.
How fix:
Add a new method to ToggleGroup class: setSelectedOnly, which will disable all other buttons and then set the selected button.
Note:
During the game loading time, the client is loading the map, and send a NetPack to server: LobbySetMap, and send a NetPack to Interface: LobbyUpdateState.
In the LobbyUpdateState it sets the map difficulty.
On Android calling `SDL_RenderSetLogicalSize` with dimensions having a different aspect ratio than the screen will cause letterboxing with blinking garbage data in that areas (instead of having black bars). This is annoying and makes the game really hard to play.
This change will make sure that on Android `SDL_RenderSetLogicalSize` will always be called with dimensions where the aspect ratio is the same as the screen to make sure there is no letterboxing.
The only problem with this is that during drawing the adventure map nothing will clear out the areas that would have been letterboxes, so there is an added explicit clear screen command for `totalRedraw()`.
The changes are only applied to Android builds in order not to affect PC builds - but some changes, especially clearing the screen could be easily added to the PC builds as well
When winning the campaign there is a lock that waits for the old server to stop before allowing to continue with restarting the server. While the lock was released on PC builds as the server runs differently on Android this was missed making campaigns always hang the app when you win.
This change adds a new native method that can be called from the Android app to signal that the server has stopped, so the client can continue with requesting a new server and showing the new campaign scenatio screen.
Note: this change alone doesn't fix the bug but it's required for a fix on vcmi-android's side. It doesn't break anything and can be merged before the vcmi-android changes
Defect:
After my previous change that enabled map layer (`CAdvMapInt`) to listen to all mouse move events via `strongInterest` bool, I noticed that there are a few scenario that start scrolling the map unexpectedly. The two that were the easiest to reproduce were using keyboard arrows to move the hero or dimissing a popup dialog.
I traced this down to unexpected mouse move event (`SDL_MouseMotionEvent`) that SDL seems to dispatch in this situation. The windowID that comes with the event in this scenario is 0 (so no window with mouse focus). I don't know why SDL dispatches this mysterious (to me) mouse move event in that case (note that there is no actual mouse movement in either repro case).
Fix:
Don't handle the mouse move event if the windowID of the mouse motion event is 0.
Notes:
Impacts scrolling of the game map. Did some play testing and didn't notice the scrolling not working when expected. The two specific cases where the map would scroll unexpectedly (move a hero with keyboard arrows or dismiss a popup dialog) no longer move the map.
In fullscreen mode, when the adventure window size does not fill the entire monitor (leaving black bars on each side), mouse movement is ignored in these areas. This results in difficulty scrolling the map in fullscreen mode, since you have to be exactly within [0,15] of the edge of the adventure window.
Fix:
Have the adventure map element, CAdvMapInt, subscribe to all mouse move events by setting the `strongInterest` bool to true. This will extend the scroll region to [min, 15], eliminating the dead mouse move space on the side of the window.
Notes:
Impacts adventure window mouse move event handling. Manually validated the scrolling the map now works without having to precisely aim for the edge of the window.
* Fix wrong save files shown in HotSeat mode
Previous condition only checked for players' connection ID. HotSeat mode does not use multiple connection IDs, so getLoadMode() thought it was SPMode.
Now checking number of interfaces to cover offline-multiplayer.
Added else condition that covers cases "ObjectAtTile that is not enemy nor is Town nor is current Hero". This should cover all cases of friendly dwellings/mines not having the correct cursor.
Warnings fixes
* Suppress `missing-braces` for Clang
* Fixed many C4275 warnings
* Fixed almost all Clang/GCC warnings
* Silence most frequent MSVC warning.
* Fixed some pessimizing-move warnings
* Fixed some unused capture warnings
* Node graph initialization optimized.
* Fixed "Unathorized obstacle access".
* Pathfinding tracing disabled with ifdef.
* Misc bonus calculation optimizations.
* Removed timestamp from log lines. Date formatting eats too much CPU.
* Paths for all heroes in Client are now cached
