|
|
/******************************Module*Header*******************************\
* Module Name: surfobj.cxx * * Surface user objects. * * Copyright (c) 1990-1999 Microsoft Corporation \**************************************************************************/
#include "precomp.hxx"
#ifdef _HYDRA_
#include "muclean.hxx"
#endif
extern BOOL G_fConsole;
// Global surface uniqueness -- incremented every time a surface is created.
ULONG _ulGlobalSurfaceUnique;
// The following declarations are required by the native c8 compiler.
PSURFACE SURFACE::pdibDefault; // The default bitmap pointer
SIZE_T SURFACE::tSize = sizeof(SURFACE);
#if TRACE_SURFACE_ALLOCS
TRACED_SURFACE::SurfaceTraceStatus TRACED_SURFACE::eTraceStatus = SURFACE_TRACING_UNINITIALIZED; #if TRACE_SURFACE_USER_CHAIN_IN_UM
TRACED_SURFACE::SurfaceTraceStatus TRACED_SURFACE::eUMTraceStatus = SURFACE_TRACING_UNINITIALIZED; #endif
#endif
// By default, EngCreateBitmap allocations larger than 260k will be
// allocated as a section:
#define KM_SIZE_MAX 0x40000
/******************************Public*Routine******************************\
* GreMarkUndeletableBitmap -- a USER callable API to mark a surface as * undeletable. Currently this is used to make sure applications * cannot delete redirection bitmaps. * * Arguments: * * hbm -- handle to bitmap to be marked undeletable * * Return Value: * * TRUE upon success, FALSE otherwise. * * History: * * 2-Nov-1998 -by- Ori Gershony [orig] * \**************************************************************************/
BOOL GreMarkUndeletableBitmap( HBITMAP hbm ) { return (HmgMarkUndeletable((HOBJ) hbm, SURF_TYPE)); }
/******************************Public*Routine******************************\
* GreMarkDeletableBitmap -- a USER callable API to mark a surface as * deletable. Currently this is used by USER just before deleting * a redirection bitmap. * * Arguments: * * hbm -- handle to bitmap to be marked deletable * * Return Value: * * TRUE upon success, FALSE otherwise. * * History: * * 2-Nov-1998 -by- Ori Gershony [orig] * \**************************************************************************/
BOOL GreMarkDeletableBitmap( HBITMAP hbm ) { return (HmgMarkDeletable((HOBJ) hbm, SURF_TYPE)); }
/******************************Public*Routine******************************\
* pvAllocateKernelSection - Allocate kernel mode section * * Arguments: * * AllocationSize - size in bytes of requested memory * * Return Value: * * Pointer to memory or NULL * * History: * * 22-May-1996 -by- Mark Enstrom [marke] * \**************************************************************************/
PVOID pvAllocateKernelSection( ULONGSIZE_T AllocationSize, ULONG Tag ) { NTSTATUS Status; OBJECT_ATTRIBUTES ObjectAttributes; LARGE_INTEGER MaximumSize; PVOID pvAlloc = NULL; PVOID pvRet = NULL;
//
// Create km section
//
ACCESS_MASK DesiredAccess = SECTION_MAP_READ | SECTION_MAP_WRITE;
ULONG SectionPageProtection = PAGE_READWRITE;
ULONG AllocationAttributes = SEC_COMMIT | SEC_NO_CHANGE;
MaximumSize.HighPart = 0; MaximumSize.LowPart = AllocationSize + sizeof(KMSECTIONHEADER);
PVOID pHandleSection;
//
// map a copy of this section into kernel address space
// Lets use the Section tagging code.
//
Status = Win32CreateSection(&pHandleSection, DesiredAccess, NULL, &MaximumSize, SectionPageProtection, AllocationAttributes, NULL, NULL, TAG_SECTION_DIB);
if (!NT_SUCCESS(Status)) { WARNING1("pvAllocateKernelSection: ObReferenceObjectByHandle failed\n"); } else { SIZE_T ViewSize = 0;
#ifdef _HYDRA_
// MmMapViewInSessionSpace is internally promoted to
// MmMapViewInSystemSpace on non-Hydra systems.
Status = Win32MapViewInSessionSpace( pHandleSection, (PVOID*)&pvAlloc, &ViewSize); #else
Status = MmMapViewInSystemSpace( pHandleSection, (PVOID*)&pvAlloc, &ViewSize); #endif
if (!NT_SUCCESS(Status)) { //
// free section
//
WARNING1("pvAllocateKernelSection: MmMapViewInSystemSpace failed\n"); Win32DestroySection(pHandleSection); } else { #ifdef _HYDRA_
#if DBG
if (!G_fConsole) { DebugGreTrackAddMapView(pvAlloc); } #endif
#endif
((PKMSECTIONHEADER)pvAlloc)->Tag = Tag; ((PKMSECTIONHEADER)pvAlloc)->pSection = pHandleSection;
pvRet = (PVOID)(((PUCHAR)pvAlloc)+sizeof(KMSECTIONHEADER)); } }
return(pvRet); }
/******************************Public*Routine******************************\
* vFreeKernelSection: Free kernel mode section * * Arguments: * * pvMem - Kernel mode section pointer * * Return Value: * * None * * History: * * 22-May-1996 -by- Mark Enstrom [marke] * \**************************************************************************/
VOID vFreeKernelSection( PVOID pvMem ) { NTSTATUS Status; PVOID pHandleSection;
if (pvMem != NULL) { PKMSECTIONHEADER pvHeader = (PKMSECTIONHEADER)((PUCHAR)pvMem - sizeof(KMSECTIONHEADER));
pHandleSection = pvHeader->pSection;
//
// Temporary code to catch stress failure (bug #263678)
// We want to fail before the page is unmapped, instead of
// afterwards
//
if (!pHandleSection) { KeBugCheckEx(PAGE_FAULT_IN_NONPAGED_AREA, (LONG_PTR) pHandleSection, (LONG_PTR) pvHeader->pSection, (LONG) pvHeader->Tag, 0); }
//
// unmap kernel mode view
//
#ifdef _HYDRA_
// MmUnmapViewInSessionSpace is internally promoted to
// MmUnmapViewInSystemSpace on non-Hydra systems.
Status = Win32UnmapViewInSessionSpace((PVOID)pvHeader); #else
Status = MmUnmapViewInSystemSpace((PVOID)pvHeader); #endif
if (!NT_SUCCESS(Status)) { WARNING1("vFreeKernelSection: MmUnmapViewInSystemSpace failed\n"); } else { #ifdef _HYDRA_
#if DBG
if (!G_fConsole) { DebugGreTrackRemoveMapView(pvHeader); } #endif
#endif
//
// delete reference to section
//
Win32DestroySection(pHandleSection); } } else { WARNING("vFreeKernelSection called with NULL pvMem\n"); } }
/******************************Public*Routine******************************\
* SURFACE::bDeleteSurface() * * Delete the surface. Make sure it is not selected into a DC if it is * a bitmap. We do under cover of multi-lock to ensure no one will select * the bitmap into a DC after we checked cRef. * * History: * Mon 17-Feb-1992 -by- Patrick Haluptzok [patrickh] * Add support for closing journal file. * * Fri 22-Feb-1991 -by- Patrick Haluptzok [patrickh] * Wrote it. \**************************************************************************/
BOOL SURFACE::bDeleteSurface(CLEANUPTYPE cutype) { GDIFunctionID(SURFACE::bDeleteSurface);
BOOL bRet = TRUE;
if (!bIsDefault() && bValid()) { HANDLE hSecure = NULL; HANDLE hDibSection = NULL; PVOID pvBitsBaseOld = NULL;
if (iType() == STYPE_BITMAP) { hSecure = DIB.hSecure; hDibSection = DIB.hDIBSection; pvBitsBaseOld = pvBitsBase(); }
PDEVOBJ pdo(hdev()); ULONG iTypeOld = iType(); DHSURF dhsurfOld = dhsurf(); PPALETTE ppalOld = ppal(); EWNDOBJ *pwoDelete = pwo(); PVOID pvBitsOld = pvBits(); FLONG fl = fjBitmap();
//
// If the surface is a bitmap, ensure it is not selected into a DC.
// Also make sure we are the only one with it locked down. These are
// both tested at once with HmgRemoveObject, because we increment
// and decrement the alt lock count at the same time we increment
// and decrement the cRef count on selection and deselection into
// DCs. Note that surfaces can also be locked for GetDIBits with no
// DC involvement, so the alt lock count may be higher than the
// reference count.
//
ASSERTGDI(HmgQueryLock((HOBJ) hGet()) == 0, "ERROR cLock != 0");
//
// If it's a device bitmap, acquire the devlock to protect against
// dynamic mode and driver changes.
//
DEVLOCKOBJ dlo;
if (bEngCreateDeviceBitmap() && pdo.bValid()) { dlo.vLock(pdo); } else { dlo.vInit(); }
#if TRACE_SURFACE_ALLOCS
#if TRACE_SURFACE_USER_CHAIN_IN_UM
PVOID pvUserMem = NULL;
if (TRACED_SURFACE::bEnabled()) { TRACED_SURFACE *pts = (TRACED_SURFACE *)this; UINT uiIndex = (UINT) HmgIfromH(hGet());
if (pts->Trace.UserChainAllocated && uiIndex < gcMaxHmgr) { pvUserMem = gpentHmgr[uiIndex].pUser; ASSERTGDI(pvUserMem == NULL || OBJECTOWNER_PID(gpentHmgr[uiIndex].ObjectOwner) == W32GetCurrentPID(), "Unowned SURFACE still has User Memory."); } } #endif
#endif
//
// Remove undeletable surfaces only during session cleanup.
//
if (HmgRemoveObject((HOBJ) hGet(), 0, 1, (cutype == CLEANUP_SESSION), SURF_TYPE)) { #if TRACE_SURFACE_ALLOCS
#if TRACE_SURFACE_USER_CHAIN_IN_UM
EngFreeUserMem(pvUserMem); #endif
#endif
//
// If this bitmap was created by EngCreateDeviceBitmap, we have
// to call DrvDeleteDeviceBitmap to clean it up. Note that we
// can't simply check for STYPE_DEVBITMAP, since EngModifySurface
// may have changed the type to STYPE_BITMAP.
//
if (bEngCreateDeviceBitmap() && (dhsurfOld != NULL)) { //
// In UMPD, a bad driver/app(ntcrash) can create a dev bitmap by
// EngCreateDeviceBitmap. But hdev could be null if
// EngAssociateSurface is not called.
//
if (pdo.bValid() && PPFNVALID(pdo, DeleteDeviceBitmap)) { #if defined(_GDIPLUS_)
(*PPFNDRV(pdo,DeleteDeviceBitmap))(dhsurfOld);
#else // !_GDIPLUS_
if (bUMPD()) { //
// Do not callout to user-mode driver if the
// user-mode process is gone (i.e., during
// session or process cleanup).
//
if (cutype == CLEANUP_NONE) { UMPDDrvDeleteDeviceBitmap(pdo.dhpdev(), dhsurfOld); } } else { (*PPFNDRV(pdo,DeleteDeviceBitmap))(dhsurfOld); }
#endif // !_GDIPLUS_
} }
FREEOBJ(this, SURF_TYPE);
//
// Note, 'this' not set to NULL
//
//
// For kernel mode, we must unlock the section memory,
// then free the memory. If the section handle is NULL
// then we just use NtVirtualFree, otherwise we must
// use NtUnmapViewOfSection
//
if (hSecure != NULL) { MmUnsecureVirtualMemory(hSecure);
if (pvBitsOld == NULL) { WARNING("deleting DIB but hSecure or pvBitsOld == NULL"); } else { if (hDibSection != NULL) { ZwUnmapViewOfSection(NtCurrentProcess(), pvBitsBaseOld); } else {
SIZE_T ViewSize = 0;
ZwFreeVirtualMemory( NtCurrentProcess(), &pvBitsOld, &ViewSize, MEM_RELEASE); } } } else if (fl & BMF_USERMEM) { #if defined(_WIN64)
if (fl & BMF_UMPDMEM) UMPDEngFreeUserMem(pvBitsOld); else #endif
EngFreeUserMem(pvBitsOld); } else if (fl & BMF_KMSECTION) { vFreeKernelSection(pvBitsOld); }
//
// This DC is going away, the associated WNDOBJ should be deleted.
// The WNDOBJs for memory bitmap and printer surface are deleted here.
// The WNDOBJs for display DCs are deleted in DestroyWindow.
//
if (pwoDelete) { GreDeleteWnd((PVOID) pwoDelete); }
if (ppalOld != NULL) { XEPALOBJ pal(ppalOld); pal.vUnrefPalette(); } } else { //
// if we can't remove it because it's an application's bitmap
// and is currently selected, mark it for lazy deletion
//
if (HmgQueryAltLock((HOBJ)hGet()) != 1) { if (hdc() != NULL || bStockSurface()) { //
// The surface is currently selected into a DC, so we can't
// simply delete it. Instead, mark it to be deleted when
// the DC is destroyed, or a new surface is selected into
// the DC.
//
vLazyDelete(); bRet = TRUE;
//
// Because we're returning TRUE, the caller will expect
// that the deletion succeeded and so will not unlock the
// object. But since it wasn't truly deleted, we have
// to keep the lock count consistent with the number of DC
// references. We account for the fact that the caller
// won't do the unlock by doing it here:
//
DEC_SHARE_REF_CNT(this); } else { WARNING("LIKELY MEMORY LEAK IN DRIVER OR GDI!"); RIP("active locks prevented surface deletion"); bRet = FALSE; } } else { RIP("failed, handle busy\n"); SAVE_ERROR_CODE(ERROR_BUSY); bRet = FALSE; } } }
return(bRet); }
/******************************Public*Routine******************************\
* SURFMEM::bCreateDIB * * Constructor for device independent bitmap memory object * * History: * Mon 18-May-1992 -by- Patrick Haluptzok [patrickh] * return BOOL * * 28-Jan-1991 -by- Patrick Haluptzok patrickh * Wrote it. \**************************************************************************/
BOOL SURFMEM::bCreateDIB( PDEVBITMAPINFO pdbmi, PVOID pvBitsIn, HANDLE hDIBSection, DWORD dsOffset, HANDLE hSecure, ULONG_PTR dwColorSpace ) { GDIFunctionID(SURFMEM::bCreateDIB);
BOOL bRet = TRUE;
AllocationFlags = SURFACE_DIB; ps = (PSURFACE) NULL; FLONG flAllocateSection = 0; BOOL bCompressed = FALSE;
//
// Figure out the length of a scanline taking arithmetic overflows into account.
//
ULONG cjScanTemp;
switch(pdbmi->iFormat) { case BMF_1BPP: cjScanTemp = ((pdbmi->cxBitmap + 31) & ~31); if(cjScanTemp < pdbmi->cxBitmap) return(FALSE); cjScanTemp >>= 3; break;
case BMF_4BPP: cjScanTemp = ((pdbmi->cxBitmap + 7) & ~7); if (cjScanTemp < pdbmi->cxBitmap) return(FALSE); cjScanTemp >>= 1; break;
case BMF_8BPP: cjScanTemp = (pdbmi->cxBitmap + 3) & ~3; if (cjScanTemp < pdbmi->cxBitmap) return(FALSE); break;
case BMF_16BPP: cjScanTemp = ((pdbmi->cxBitmap + 1) & ~1) << 1; if (cjScanTemp < pdbmi->cxBitmap) return(FALSE); break;
case BMF_24BPP: if ((pdbmi->cxBitmap * 3 < pdbmi->cxBitmap) || ((cjScanTemp = ((pdbmi->cxBitmap * 3) + 3) & ~3) < pdbmi->cxBitmap)) return(FALSE); break;
case BMF_32BPP: cjScanTemp = pdbmi->cxBitmap << 2; if (cjScanTemp < pdbmi->cxBitmap) return(FALSE); break;
case BMF_8RLE: case BMF_4RLE: case BMF_JPEG: case BMF_PNG: bCompressed = TRUE; break;
default: WARNING("ERROR: failed INVALID BITMAP FORMAT \n"); return(FALSE); }
//
// If we are given a pointer to bits, then only allocate a DIB header.
// Otherwise allocate space for the header and the required bits.
//
ULONGSIZE_T size = (ULONGSIZE_T) SURFACE::tSizeOf();
FSHORT fsAlloc = HMGR_ALLOC_ALT_LOCK|HMGR_NO_ZERO_INIT;
if (pvBitsIn == (PVOID) NULL) { LONGLONG eq;
if (bCompressed) eq = (LONGLONG)(ULONGLONG) pdbmi->cjBits; else eq = Int32x32To64(pdbmi->cyBitmap, cjScanTemp);
eq += (LONGLONG)(ULONGLONG) size;
if (eq > LONG_MAX) { WARNING("Attempting to allocate > 4Gb\n"); return(FALSE); }
//
// if it is UMPD and size > PAGE_SIZE, we always allocate in user memory
//
//
// make sure no one will change KM_SIZE_MAX later and mess things up
//
ASSERTGDI(PAGE_SIZE < KM_SIZE_MAX, "bad KM_SIZE_MAX\n");
if ((pdbmi->fl & UMPD_SURFACE) && (eq > PAGE_SIZE)) pdbmi->fl |= BMF_USERMEM;
// see if we need to allocate the bits out of USER memory
if (pdbmi->fl & BMF_USERMEM) { pvBitsIn = EngAllocUserMem((LONG) eq,'mbuG'); //Gubm
if (pvBitsIn == NULL) return(FALSE); } else if ((pdbmi->fl & BMF_KMSECTION) || (eq > KM_SIZE_MAX)) { //
// Kernel-mode pool is limited in size and prone to fragmentation.
// For this reason, we will automatically allocate 'large' bitmaps
// as sections, which will not be charged against kernel-mode
// pool.
//
// We also allow the driver to specify BMF_KMSECTION -- this is
// specifically for the ModeX driver to allow it to map its shadow
// buffer into user-mode.
//
//
//Sundown, change from SIZE_T to LONG, we return false earlier if > 4GB
//
pvBitsIn = pvAllocateKernelSection((LONG)eq,'mbkG');
if (pvBitsIn != NULL) { //
// mark surface as KM SECTION
//
flAllocateSection = BMF_KMSECTION; } }
//
// combine size and allocate from pool
//
if (pvBitsIn == NULL) { size = (ULONGSIZE_T) eq;
if ((pdbmi->fl & BMF_NOZEROINIT) == 0) { fsAlloc = HMGR_ALLOC_ALT_LOCK; } } } else { ASSERTGDI(!(pdbmi->fl & BMF_USERMEM),"flags error\n"); }
ps = (PSURFACE)ALLOCOBJ(size,SURF_TYPE,!(fsAlloc & HMGR_NO_ZERO_INIT));
if (ps == NULL) { WARNING("failed memory alloc\n"); bRet = FALSE; } else { //
// Initialize the surf fields
//
SIZEL sizlTemp; sizlTemp.cx = pdbmi->cxBitmap; sizlTemp.cy = pdbmi->cyBitmap; ps->sizl(sizlTemp); ps->iType(STYPE_BITMAP); ps->hSecureUMPD = 0;
if (pdbmi->hpal != (HPALETTE) 0) { EPALOBJ palSurf(pdbmi->hpal); ASSERTGDI(palSurf.bValid(), "ERROR invalid palette");
//
// Set palette into surface.
//
ps->ppal(palSurf.ppalGet());
//
// Reference count it by making sure it is not unlocked.
//
palSurf.ppalSet((PPALETTE) NULL); // It won't be unlocked
} else { ps->ppal((PPALETTE) NULL); }
//
// Initialize the BITMAP fields
//
ps->iFormat(pdbmi->iFormat);
ps->fjBitmap(((pdbmi->fl) & (BMF_TOPDOWN | BMF_USERMEM)) | (flAllocateSection));
ps->DIB.hDIBSection = hDIBSection; ps->DIB.dwOffset = dsOffset; ps->DIB.hSecure = hSecure; ps->DIB.dwDIBColorSpace = dwColorSpace;
ps->dhsurf((DHSURF) 0); ps->dhpdev((DHPDEV) 0); ps->flags(pdbmi->fl & UMPD_SURFACE); ps->pwo((EWNDOBJ *) NULL); sizlTemp.cx = 0; sizlTemp.cy = 0; ps->sizlDim(sizlTemp); ps->hdev((HDEV) 0); ps->EBitmap.hdc = (HDC) 0; ps->EBitmap.cRef = 0; ps->EBitmap.hpalHint = 0; ps->pdcoAA = NULL;
if (hSecure != (HANDLE) NULL) { //
// Set flag for DIBSECTION so driver doesn't cache it.
// because we don't know to increment the uniqueness
// when the app writes on it.
//
ps->so.fjBitmap |= BMF_DONTCACHE; }
//
// Initialize the DIB fields
//
if (pvBitsIn == (PVOID) NULL) { ps->pvBits((PVOID) (((ULONG_PTR) ps) + SURFACE::tSizeOf())); } else { ps->pvBits(pvBitsIn); }
if ((pdbmi->iFormat != BMF_8RLE) && (pdbmi->iFormat != BMF_4RLE) && (pdbmi->iFormat != BMF_JPEG) && (pdbmi->iFormat != BMF_PNG )) { ps->cjBits(pdbmi->cyBitmap * cjScanTemp);
if (pdbmi->fl & BMF_TOPDOWN) { ps->lDelta(cjScanTemp); ps->pvScan0(ps->pvBits()); } else { ps->lDelta(-(LONG)cjScanTemp); ps->pvScan0((PVOID) (((PBYTE) ps->pvBits()) + (ps->cjBits() - cjScanTemp))); } } else { //
// lDelta is 0 because RLE's don't have scanlines.
//
ps->lDelta(0); ps->cjBits(pdbmi->cjBits);
//
// pvScan0 is ignored for JPEG's and PNG's
//
if ((pdbmi->iFormat != BMF_JPEG) && (pdbmi->iFormat != BMF_PNG)) ps->pvScan0(ps->pvBits()); else ps->pvScan0(NULL); }
//
// Set initial uniqueness. Not 0 because that means don't cache it.
//
// We used to always set the uniqueness to 1 on creation, but the
// NetMeeting folks ran into tool-bar scenarios where the buttons
// were being created and deleted on every repaint and eventually we
// would run out of uniqueness bits in the handles. So the end
// result would be that the NetMeeting driver would see a surface
// with a handle the same as the one they cached, and with an iUniq
// the same as the one they cached (specifically, a value of '1') --
// but the cached bitmap would not match the actual bitmap bits!
//
// We fix this by always creating the surface with a unique uniqueness.
// Note that it's still possible to get different surfaces with the
// same uniqueness because drawing calls simply increment the surface
// uniqueness, not the global uniqueness. However, this change will
// make the possibility of the driver mis-caching a bitmap extremely
// unlikely.
//
ps->iUniq(ulGetNewUniqueness(_ulGlobalSurfaceUnique));
//
// Now that the surface is set up, give it a handle
//
if (HmgInsertObject(ps, fsAlloc, SURF_TYPE) == 0) { WARNING("failed HmgInsertObject\n");
//
// Don't forget to decrement reference count on the palette before
// freeing the surface
//
if (ps->ppal()) { XEPALOBJ pal(ps->ppal()); pal.vUnrefPalette(); ps->ppal((PPALETTE) NULL); // Not necessary, but makes the code cleaner
}
FREEOBJ(ps, SURF_TYPE); ps = NULL; bRet = FALSE; } else { ps->hsurf(ps->hGet());
#if TRACE_SURFACE_ALLOCS
if (TRACED_SURFACE::bEnabled()) { TRACED_SURFACE *pts = (TRACED_SURFACE *)ps;
RtlZeroMemory(&pts->Trace, sizeof(pts->Trace)); pts->Trace.pProcess = PsGetCurrentProcess(); pts->Trace.pThread = PsGetCurrentThread(); pts->Trace.KernelLength = RtlWalkFrameChain((PVOID *)pts->Trace.Chain, lengthof(pts->Trace.Chain), 0); ULONG MaxUserLength; PVOID *UserChain; PVOID TmpUserChain[TRACE_SURFACE_MIN_USER_CHAIN];
MaxUserLength = lengthof(pts->Trace.Chain) - pts->Trace.KernelLength; if (MaxUserLength < TRACE_SURFACE_MIN_USER_CHAIN) { MaxUserLength = TRACE_SURFACE_MIN_USER_CHAIN; UserChain = TmpUserChain; } else { UserChain = (PVOID *)&pts->Trace.Chain[pts->Trace.KernelLength]; }
pts->Trace.UserLength = RtlWalkFrameChain(UserChain, MaxUserLength, 1);
if (UserChain == TmpUserChain && pts->Trace.UserLength != 0) { pts->Trace.KernelLength = min(pts->Trace.KernelLength, (ULONG) lengthof(pts->Trace.Chain) - pts->Trace.UserLength); RtlCopyMemory(&pts->Trace.Chain[pts->Trace.KernelLength], UserChain, pts->Trace.UserLength); }
#if TRACE_SURFACE_USER_CHAIN_IN_UM
if (pts->bUMEnabled()) { SurfaceUserTrace *pSurfUserTrace; PENTRY pentTmp; UINT uiIndex = (UINT) HmgIfromH(pts->hGet());
pentTmp = &gpentHmgr[uiIndex];
ASSERTGDI(pentTmp->pUser == NULL, "SURFACE's pUser is already being used.\n");
pSurfUserTrace = (SurfaceUserTrace *)EngAllocUserMem(sizeof(SurfaceUserTrace), 'rTSG');
if (pSurfUserTrace != NULL) { pts->Trace.UserChainAllocated = 1; pts->Trace.UserChainNotRead = 1;
pSurfUserTrace->MaxLength = lengthof(pSurfUserTrace->Chain); pSurfUserTrace->UserLength = 0; pentTmp->pUser = pSurfUserTrace; } } #endif
} #endif
} }
//
// cleanup in failure case
//
if (!bRet && pvBitsIn) { if (pdbmi->fl & BMF_USERMEM) { EngFreeUserMem(pvBitsIn); } else if (flAllocateSection & BMF_KMSECTION) { vFreeKernelSection(pvBitsIn); } }
return(bRet); }
/******************************Public*Routine******************************\
* * SURFMEM::~SURFMEM * * Description: * * SURFACE Destructor, takes appropriate action based * on allocation flags * \**************************************************************************/ SURFMEM::~SURFMEM() {
if (ps != (SURFACE*) NULL) { //
// what type of surface
//
if (AllocationFlags & SURFACE_KEEP) {
DEC_SHARE_REF_CNT(ps);
} else {
if (AllocationFlags & SURFACE_DIB) { //
// free selected palette
//
if (ps->ppal() != NULL) { XEPALOBJ pal(ps->ppal()); pal.vUnrefPalette(); } }
//
// remove object from hmgr and free
//
if (!HmgRemoveObject((HOBJ) ps->hGet(), 0, 1, TRUE, SURF_TYPE)) { ASSERTGDI(TRUE, "Failed to remove object in ~DIBMEMOBJ"); }
PVOID pvBitsOld = ps->pvBits(); FLONG fl = ps->fjBitmap(); BOOL bUMPD = ps->bUMPD();
FREEOBJ(ps, SURF_TYPE);
if (fl & BMF_USERMEM) { if (bUMPD && pvBitsOld) { EngFreeUserMem(pvBitsOld); } else if (!bUMPD) { RIP("SURFMEM destructor has BMF_USERMEM set\n"); } } else if (fl & BMF_KMSECTION) { vFreeKernelSection(pvBitsOld); } } } }
#if DBG
void SURFACE::vDump() { DbgPrint("SURFACE @ %-#x\n", this); DbgPrint(" so.dhsurf = %-#x\n" , so.dhsurf); DbgPrint(" so.hsurf = %-#x\n" , so.hsurf); DbgPrint(" so.dhpdev = %-#x\n" , so.dhpdev); DbgPrint(" so.hdev = %-#x\n" , so.hdev); DbgPrint(" so.sizlBitmap = %u %u\n" , so.sizlBitmap.cx , so.sizlBitmap.cy); DbgPrint(" so.cjBits = %u\n" , so.cjBits); DbgPrint(" so.pvBits = %-#x\n" , so.pvBits); DbgPrint(" so.pvScan0 = %-#x\n" , so.pvScan0); DbgPrint(" so.lDelta = %d\n" , so.lDelta); DbgPrint(" so.iUniq = %u\n" , so.iUniq); DbgPrint(" so.iBitmapFormat = %u\n" , so.iBitmapFormat); DbgPrint(" so.iType = %u\n" , so.iType); DbgPrint(" so.fjBitmap = %-#x\n" , so.fjBitmap);
DbgPrint(" SurfFlags = %-#x\n" , SurfFlags); DbgPrint(" pPal = %-#x\n" , pPal); DbgPrint(" pWo = %-#x\n" , pWo); DbgPrint(" EBitmap.sizlDim = %u %u\n" , EBitmap.sizlDim.cx, EBitmap.sizlDim.cy); DbgPrint(" EBitmap.hdc = %-#x\n" , EBitmap.hdc); DbgPrint(" EBitmap.cRef = %-#x\n" , EBitmap.cRef); DbgPrint(" DIB.hDIBSection = %-#x\n" , DIB.hDIBSection); DbgPrint(" DIB.hSecure = %-#x\n" , DIB.hSecure);
} #endif
#if TRACE_SURFACE_ALLOCS
#define TRACE_SURFACE_KM_ENABLE_MASK 0x00000001
#define TRACE_SURFACE_UM_ENABLE_MASK 0x00000002
VOID TRACED_SURFACE::vInit() { NTSTATUS Status; DWORD dwDefaultValue = 0; DWORD dwEnableStack;
// Check registry key to determine whether to enable stack traces
// or not for SURFACE allocations
RTL_QUERY_REGISTRY_TABLE QueryTable[] = { {NULL, RTL_QUERY_REGISTRY_DIRECT, L"EnableSurfaceTrace", &dwEnableStack, REG_DWORD, &dwDefaultValue, sizeof(dwDefaultValue)}, {NULL, 0, NULL} };
if (eTraceStatus == SURFACE_TRACING_UNINITIALIZED) { Status = RtlQueryRegistryValues(RTL_REGISTRY_WINDOWS_NT, L"GRE_Initialize", &QueryTable[0], NULL, NULL); if (NT_SUCCESS(Status)) { if (dwEnableStack & TRACE_SURFACE_KM_ENABLE_MASK) { eTraceStatus = SURFACE_TRACING_ENABLED; SURFACE::tSize = sizeof(TRACED_SURFACE);
if (dwEnableStack & TRACE_SURFACE_UM_ENABLE_MASK) { #if TRACE_SURFACE_USER_CHAIN_IN_UM
eUMTraceStatus = SURFACE_TRACING_ENABLED; #else
WARNING("GDI: UM Surface Trace requested, but not available.\n"); #endif
} else { #if TRACE_SURFACE_USER_CHAIN_IN_UM
eUMTraceStatus = SURFACE_TRACING_DISABLED; #endif
} } else { eTraceStatus = SURFACE_TRACING_DISABLED; #if TRACE_SURFACE_USER_CHAIN_IN_UM
eUMTraceStatus = SURFACE_TRACING_DISABLED; #endif
} } } else { RIP("TRACED_SURFACE::vInit: Tracing already initialized.\n"); } }
#if TRACE_SURFACE_USER_CHAIN_IN_UM
VOID TRACED_SURFACE::vProcessStackFromUM( BOOL bFreeUserMem ) { GDIFunctionID(TRACED_SURFACE::vProcessStackFromUM);
if (Trace.UserChainAllocated) { if (Trace.UserChainNotRead || bFreeUserMem) { PENTRY pentTmp; UINT uiIndex = (UINT) HmgIfromH(hHmgr); ULONG ulNewKernelLength; ULONG ulMaxUserLength; ULONG ulMaxLength = lengthof(Trace.Chain);
if (uiIndex < gcMaxHmgr) { pentTmp = &gpentHmgr[uiIndex];
if (pentTmp->pUser != NULL) { if (Trace.UserChainNotRead) { SurfaceUserTrace *pUserTrace = (SurfaceUserTrace *)pentTmp->pUser; __try { if (pUserTrace->UserLength > 0) { ulMaxUserLength = min(pUserTrace->UserLength, ulMaxLength); ulNewKernelLength = ulMaxLength - ulMaxUserLength; if (ulNewKernelLength > Trace.KernelLength) { ulNewKernelLength = Trace.KernelLength; } RtlCopyMemory(&Trace.Chain[ulNewKernelLength], pUserTrace->Chain, ulMaxUserLength);
Trace.UserLength = ulMaxUserLength; Trace.UserChainNotRead = 0; Trace.KernelLength = ulNewKernelLength; } } __except(EXCEPTION_EXECUTE_HANDLER) { WARNING("Couldn't read UM stored stack trace."); } }
if (bFreeUserMem) { Trace.UserChainAllocated = 0; EngFreeUserMem(pentTmp->pUser); pentTmp->pUser = NULL; } } else { RIP("Lost track of UserMem trace allocation.\n"); Trace.UserChainAllocated = 0; } } } } } #endif
#endif
|