KD64: Raise to HIGH_LEVEL when entering trap
KDBG: lower to DISPATCH_LEVEL when applying IRQL hack & use a worker thread to load symbols
KD&KDBG: Actually unload symbols when required
Raise IRQL before entering debugger, so that KeAcquireSpinLockAtDpcLevel works as expected.
- HIGH_LEVEL since we don't know where we are coming from.
Do not try to read debug symbol from files in KDBG.
- There is no reason that this works if Mm didn't map it in the first place.
GCC has some functions, variables & type attributes which can be used as aliases
for some of the SAL annotations. Although it's not as rich & precise, it's still useful
since we actually enable -Werror on GCC builds whereas we don't use such an option
on MSVC builds.
For now, _Must_inspect_result_ is aliased to warn_result_unused attribute.
ProcessUserModeIOPL, ProcessWow64Information and ThreadZeroTlsCell classes fail on AMD64 build because of wrong IQS values assigned to them. Also explicitly tell the compiler that ProcessUserModeIOPL is strictly for x86 as user mode I/O privilege level is 32-bit stuff.
In addition to that, implement IQS_NO_TYPE_LENGTH macro which it'll be used for classes such as ProcessUserModeIOPL where type length is not required and that it should be 0. With that said, we indirectly fix a size length mismatch issue with ProcessUserModeIOPL on 32-bit of ReactOS as well.
- Remove a redundant call of ObReferenceObjectByHandle. Not only it didn't make much sense (we reference the object from thread handle and the new thread object referencing the same handle!), specifying a request access of THREAD_ALL_ACCESS for the thread object is kind of suspicious and all of these access rights are unwanted.
- Add some failure checks involving the CopyOnOpen code paths
- Add some DPRINT1 debug prints (concerning the CopyOnOpen code paths as usual)
driverName.Buffer leaked when the "(!NT_SUCCESS(status) || ServiceName != NULL)"
case is taken because ServiceName != NULL, and some of the functions fail.
- Fix CID 1477246: Uninitialized pointer read (UNINIT) (happens in
the last ExFreePoolWithTag(basicInfo, TAG_IO) call when the
"(!NT_SUCCESS(status) || ServiceName != NULL)" case is not taken).
- Centralize all the ExFreePoolWithTag(basicInfo, TAG_IO) cleanups
at the end of the function.
- Both cases "(driverName.Buffer == NULL)" and "(ServiceName != NULL)"
can only be taken when basicInfo != NULL, so assert on this fact.
- Fix/add comments;
- Reduce indentation level;
- Direct copy for registry integer values;
- Use for-loops for linked lists;
- Use ULONG when the API uses it (sizes for Ob, or REG_DWORD data in registry).
When freeing pages, free page entries with pages num == 3 were
incorrectly treated as entries with pages num >= 4 and thus
their re-insertion was not triggered. That lead to non paged pool
fragmentation (can be triggered by kmtest:ExPools, for example)
Also, altered the index acquisition code for MmNonPagedPoolFreeList
entries so it looks more clear
- Manage the lifetime of the temporary 'PartitionBuffer' buffer where
it is locally used only, and free it as soon as possible, just after
calculating the sector checksum. No need to then free it outside of
the main for-loop.
- When the 'DriveLayout' buffer is freed, ensure the pointer is NULL-ed
(and assert this at the top of the main for-loop), since it can also
be freed at cleanup outside this for-loop, and in this case a NULL
check is performed.
This will avoid the scenario of possibly double-freeing a pointer,
in the case the 'DriveLayout' was previously freed (when e.g. reading
the sector for checksum calculation failed), then the for-loop goes to
the next disk and stops early.
The purpose of 'SingleDisk' is the same as in the IopCreateArcNames()
function. It is an optimization for that when looking up the
firmware-recognized ARC disks list, in order to match one of these with
the current NT disk being analysed (see e.g. also in IopCreateArcNamesDisk()),
we avoid a possible IopVerifyDiskSignature() call and directly build a
corresponding ARC name NT symbolic link for it.
'SingleDisk' will actually be TRUE, whether the DiskSignatureListHead
list is empty or contains only one element: Indeed in only both these
cases, 'DiskSignatureListHead.Flink->Flink' will refer to the list head.
(If the list is empty but 'SingleDisk' is TRUE, this does not matter,
because the DiskSignatureListHead looking-up loop never starts.)
In addition to that, here are some stuff done in this commit whilst testing:
- ICIF_QUERY_SIZE_VARIABLE and friends were badly misused, they should be used only when an information class whose information length size is dyanmic and not fixed. By removing such flags from erroneous classes, this fixes the STATUS_INFO_LENGTH_MISMATCH testcases.
- Use CHAR instead of UCHAR for classes that do not need alignment probing, as every other class in the table do, for the sake of consistency.
- ProcessEnableAlignmentFaultFixup uses BOOLEAN as type size, not CHAR. This fixes a testcase failure on ROS.
- Check for information length size before proceeding further on querying the process' cookie information.
- ProcessHandleTracing wants an alignment of a ULONG, not CHAR.
- Move PROCESS_LDT_INFORMATION and PROCESS_LDT_SIZE outside of NTOS_MODE_USER macro case. This fixes a compilation issue when enabling the alignment probing. My mistake of having them inside NTOS_MODE_USER case, sorry.
- On functions like NtQueryInformationThread and the Process equivalent, complete probing is not done at the beginning of the function, complete probing including if the buffer is writable alongside with datatype misalignment check that is. Instead such check is done on each information class case basis. With that said, we have to explicitly tell DefaultQueryInfoBufferCheck if we want a complete probing or not initially.
KiSetTrapContext is an asm wrapper around RtlSetUnwindContext, which first stores an exception frame to assure that all non-volatile registers were put on the stack, then calls RtlSetUnwindContext to update their first saving positions on the stack and finally restore the exception frame to potentially load any updated registers, that haven't been saved elsewhere on the stack.
A few of these classes have fixed size lengths, the rest are arbitrary. Also the TokenAuditPolicy class hasn't a size length type specified in the table, which is wrong (and move the corresponding TOKEN_AUDIT_POLICY_INFORMATION structure into the private header).
PsImpersonateClient blindly impersonates the requested client even though it doesn't know if the actual token given to the call can be impersonated for the thread of the client which we are going to begin impersonation. In the case where impersonation is not possible, make a copy of the given token and assign the newly one for impersonation instead.
CORE-17539
Implement SepImpersonateAnonymousToken private helpers, which is necessary for the complete implementation of NtImpersonateAnonymousToken function and thus finally we're able to impersonate the anonymous logon token.
As of now the Object Manager private service, ObpCloseHandleTableEntry, looks for OBJ_PROTECT_CLOSE attribute if a handle should not be closed. However, in ObDuplicateObject if an attribute of OBJ_PROTECT_CLOSE is found as it's been filled to the caller (see L2466) this attribute is removed from the attributes list of the new handle and ObpAccessProtectCloseBit access is granted to the newly duplicated object handle.
With that being said ObpCloseHandleTableEntry indiscriminately closes the object handle albeit it shouldn't do so. As a matter of fact in Windows Server 2003 SP2 this service indeed checks for ObpAccessProtectCloseBit flag bit and if the condition is met then it returns STATUS_HANDLE_NOT_CLOSABLE as it should. Therefore we should do the same.
Now NtClose can properly warn the calling thread the object handle can't be closed which fixes a testcase failure within NtDuplicateObject NTDLL APITEST where this function gives handle close protection bit as requested by the caller.
* Guard the token in a lock whilst querying stuff
* Remove the piece of code that checks if the information class provided is above the maximum information class threshold. That code literally duplicates the inner functionality of the default case in the switch block, where the code falls in that case if an invalid information class is provided anyway.
* Remove the redundant information classes. Internally, this function in Windows has 12 switch case blocks (11 token info classes + the default case) and the other classes are supported in NtQueryInformationToken only so it doesn't make any logical sense to keep them in the codebase.
* Annotate the argument parameters with SAL and add documentation header
Properly handle PDE refcounting
Clean-up of the internal API
Enforce attaching to the process when modifying its memory layout, instead of
making circonvoluted mappings which always end up being broken.
- Do not lock the section segment when we are serving a fault for a process private page.
- Do not keep the process address space lock while writing to pagefile.
- Do not wait for an event that might never be set.
Mute debug prints of MmDoesFileHaveUserWritableReferences and SeAuditingFileEventsWithContext stubs.
These stubs are very noisy and create a lot of spam in the log when using Microsoft NTFS driver in ReactOS (with some other improvements applied).
Implementing those functions isn't badly required for the proper work of this driver, so better way for now is just mute these stubs a bit.
After my changes, they will be displayed only once, and the log will be more clear, so it will seem to be enough to understand that the driver calls these routines.
CORE-17409
This function is used during the Security kernel module phase initialisation to set up the system process token which the phase initialisation procedure in itself is stored in the INIT section. With that being said, do the same for SepCreateSystemProcessToken too and add a header documentation as an addition.
These private functions are needed to set up two different kinds of system's anonymous logon tokens: one that includes everyone in the group and the other that doesn't. These functions are needed as next step closer to the
implementation of NtImpersonateAnonymousToken system call.
for manually reported devices, as it is required by the newdev.dll
for installing drivers from INF files
CORE-17212 CORE-17398
Co-authored-by: Stanislav Motylkov <x86corez@gmail.com>
Shrink when using 1/8 of its allocated capacity (thus use 25% of it at the end of the process)
Expand when using 3/4 of its allocated capacity (thus use ~40% of it at the end of the process)
This control class is triggered when a driver is being installed for a
non-critical device. The driver info should already be in the registry
so we just need to push the device through the state graph
Meanwhile, combine the code for similar control classes into
PiControlSyncDeviceAction routine
CORE-17463 CORE-17490
* Implement SepCompareSidAndAttributesFromTokens and SepComparePrivilegeAndAttributesFromTokens functions for array elements comparison
* Implement the token comparison code in SepCompareTokens function
* Add a missing PAGED_CODE() in SepCompareTokens as most of the token comparison code is paged
* Use SAL annotations for SepCompareTokens and NtCompareTokens
Prior to acquiring a quota from the process and do whatever it's needed to do (charge it or return it back), we must guard ourselves with a spinlock so that we may not get into potential race conditions. In Windows Server 2003, PspGivebackQuota and PspExpandQuota do the same thing and they're the equivalent to PspReturnProcessQuotaSpecifiedPool and PspChargeProcessQuotaSpecifiedPool in our codebase.
Phase 2 and 3 were not done anymore since 777a2d94da.
Fix that, by merging phases 1 and 2, and by calling phase 3 later
for log file debugging, when ExpInitializationPhase = 3
CORE-17470
In Windows Server 2003 the lock is initialised on a per-token basis, that is, the lock resource is created in SepDuplicateToken() and SepCreateToken() functions. This ensures that the lock initialisation is done locally for the specific token thus avoiding the need of a global lock.
Do not ditch the pages as soon as the section are unmapped
Improve MmBalancer "algorithm" (or whatever you call that)
Various needed fixes to get this going.
Purge data section object before creating an image mapping
Zero-out the tail of the page after reading from file
Properly map page as read-only when paging-in a COW memory map.
Previously, when creating a file section, Mm requested Cc to cache the file, then Cc would request pages from Mm, then Mm would request them back to serve its file-mapping role
Now, Mm does it all by itself. If file cahcing is requested by the FS driver, then Cc creates a file mapping and uses that to serve its purpose.
This is a rewrite of Cc
KeLowerIrql/KeRaiseIrqlToDpcLevel/KfRaiseIrql are defined as FORCEINLINE in ntoskrnl/ke/amd64/irql.c
Use KxLowerIrql/KxRaiseIrqlToDpcLevel/KxRaiseIrql instead, which call their respective Ke*/Kf* counterparts.
During the boot process, it makes possible to initalize the driver's
devices right after the driver is loaded. Moreover, this way one can be
sure that all critical devices are initialized before the
IopMarkBootPartition call (because we explicitly call the driver's
AddDevice routine now, after each driver is loaded)
CORE-7826
- Use DeviceNode->State field and its values, instead of
DeviceNode->Flags for tracking current node state
- Change DNF_* flags to the ones compatible with Windows XP+
- Simplify state changes for device nodes and encapsulate all the logic
inside the PiDevNodeStateMachine routine. This makes the ground for
future improvements in the device removal sequence and
resource management
- Now values inside DeviceNode->State and ->Flags are compatible with
the windbg !devnode macro and can be tracked using it
- BUGFIX: fixed cases where IRP_MN_START_DEVICE or
IRP_MN_QUERY_DEVICE_RELATIONS may be sent to a device after a
IRP_MN_REMOVE_DEVICE
CORE-7826
Add another PnPBootDriversInitialized variable to indicate a point where
both disk subsystem and SystemRoot symlink are initialized, and use it
in a PiCallDriverAddDevice call.
- Move the driver's name obtaining logic into the IopGetDriverNames
function
- Create a new PiCallDriverAddDevice instead of PipCallDriverAddDevice
and move it to pnpmgr/devaction.c file. Move around all its internal
helpers too
- Support a proper Windows-compatible driver loading order for a PDO
(lower filters, main service, upper filters, etc.)
- Set a correct Problem for the DeviceNode, in case of an error during
driver loading
- Check the Start Type for all drivers before loading
- Do not try to load drivers during the early boot stage when there is
no disk subsystem initialized
- Do not hold the IopDriverLoadResource while trying to reference a
driver object (but still acquire it when we actually need to load a
driver)
- Change IopLoadDriver and IopInitializeDriverModule to use registry
handle instead of a service name string and/or full registry path
- Do not try to reference a driver object inside IopLoadDriver. It's
supposed to be done before the function call
- Split IopLoadUnloadDriver into IopLoadDriver and calling DriverUnload
- Schedule the worker for (un)loading driver in a separate routine
(IopDoLoadUnloadDriver) this allows IopLoadDriver to be called
separately (if we are sure that we're in the system process)
- Remove IopCreateDriver and put its code into IoCreateDriver and
IopInitializeDriverModule. It's hard to extract a meaningful common
part from it
- Refactor IopInitializeDriverModule. Extend and put the DriverName
generation logic into it. Now this function frees the ModuleObject in
case of failure and returns STATUS_FAILED_DRIVER_ENTRY in case of
DriverInit failure (will be used later)
If the CPU does support CPUID instruction but does not have the Brand String,
the legacy name detection should be performed either in CIM_Processor (WMI),
in the Processor device driver, or in the application itself.
The deleted Intel CPU name detection code can go to the WMI at some point.
In fact Windows XP and 2003 kernels do not have any strings of that kind.
This effectively reverts the hack committed 6 years ago in 7a985425 (r63798)
that was intended to workaround CORE-7952.
CORE-17413
- Fix behaviour when adding or removing entries in the middle of an existing run
- Do not touch output parameters when failing, caller might rely on this.
- Convert PARTITION_TABLE_OFFSET to the number of bytes instead of
(number of bytes) / 2. This avoids many confusing casts
- Use a cache aligned buffer for MBR
- BUGFIX: do not call IoGetRelatedTargetDevice while guarded mutex is acquired
(the function issues an APC, but they are disabled inside a critical section)
- BUGFIX: only the beginning of a structure for GUID_PNP_CUSTOM_NOTIFICATION was copied and queued.
Just pass it as-is to a subscriber, without copying
- Don't convert event GUID to string, store and compare GUID struct itself
- Split IopNotifyPlugPlayNotification into 3 functions for each type of notification
(less stack usage and for future changes)
- Move initialization code for notifications into a separate routine
- Use separate lists and locks for every type of notification
- Put "TargetDeviceChange" notifications into their place inside DEVICE_NODE
CmResourceTypeDevicePrivate resources are created the the bus driver, for its own later use,
when children devices will receive the IRP_MN_START_DEVICE.
Do not translate them, as they have no meaning outside bus driver.
- Change INIT_FUNCTION and INIT_SECTION to CODE_SEG("INIT") and DATA_SEG("INIT") respectively
- Remove INIT_FUNCTION from function prototypes
- Remove alloc_text pragma calls as they are not needed anymore
This is required since the NX protection proagates from the highest level, enforcing NX on the entire range, independent of whether lower level P*Es have the bit set or not. It might be useful to add a platform specific constant to allow making page tables NX on architectures that have a different behavior.
* Move MiBuildPfnDatabaseFromPageTables into MiBuildPfnDatabase
* Make sure to call MmInitializeProcessAddressSpace() from the x64 version of MiInitMachineDependent()
* Handle result of MmInitializeProcessAddressSpace in MiInitMachineDependent (Should do the same for x86)
* Remove obsolete x64 debug print
- Deliver pending APCs on trap exit
- Pass the trapframe of KiApcInterrupt to KiDeliverApcs, not NULL.
- Fix parameter passing from KiSwapContext to KiSwapContextInternal and KiSwapContextResume, so that the ApcBypass parameter is not uninitialized
- Fix return value of KiSwapContextResume to correctly indicate whether we want to have APCs directly delivered or not (when there are non, or when delivery is suppressed)
The previous version (like the x86 one) used a combination of C and asm code, called from C code to switch the stack. This is problematic, since there is no guarantee what assumptions C code makes about the stack (i.e. it can place any kind of stack pointers into registers or on the stack itself.) The new algorithm returns back to the systemcall entry point in asm, which then calls KiConvertToGuiThread, which is also asm and calls KeSwitchKernelStack ...
To be 100% correct and not rely on assumptions, stack switching can only be done when all previous code - starting with the syscall entry point - is pure asm code, since we can't rely on the C compiler to not use stack addresses in a way that is not transparent. Therefore the new code uses the same mechanism as for normal system calls, returning the address of the asm function KiConvertToGuiThread, which is then called like an Nt* function would be called normally. KiConvertToGuiThread then allocated a new stack, switches to it (which is now fine, since all the code is asm), frees the old stack, calls PsConvertToGuiThread (which now will not try to allocate another stack, since we already have one) and then jumps into the middle of KiSystemCallEntry64, where the system call is handled again.
Also simplify KiSystemCallEntry64 a bit by copying the first parameters into the trap frame, avoiding to allocate additional stack space for the call to KiSystemCallHandler, which now overlaps with the space that is allocated for the Nt* function.
Finally fix the locations where r10 and r11 are stored, which is TrapFrame->Rcx and TrapFrame->EFlags, based on the situation in user mode.
CORE-14037
- Fix buggy retrieval of the current calling Irp->Tail.Overlay.Thread.
- The 4th argument (KernelRoutine) to the KeInitializeApc() is **NOT**
optional; however its 5th argument (RundownRoutine) is.
So use the mandatory routine for freeing the allocated APC instead.
We don't use the rundown routine yet.
- Check whether the ExAllocatePoolWithTag() call failed or not before
queueing the allocated APC.
The current implementation of device tree enumeration does not distinguish
between possible identifier strings. This change is not a hack,
NT 5.2 still creates a new device node for this hardware ID.
The entire function should be rewritten to match Windows implementation.
We are doing IoCallDriver here, so the valid stack location should be
CurrentLocation <= Irp->StackCount (just a check for a completly incorrect value)
&& CurrentLocation > 1 (ensure that we have a place for another call)
CORE-17189
Co-authored-by: Thomas Faber <thomas.faber@reactos.org>
RtlGetNtProductType comes into two variants: one in user-mode that is exported for use from NTDLL layer and the kernel-mode that is used exclusively by the NT kernel. The kernel-mode variant of the function is not exported.
- Some "PartitionInfo->PartitionNumber = 0;" are ROS-specific hacks for
xHalIoAssignDriveLetters(), that should be fixed... Mark them as such.
- Un-hardcode some "magic" values (partition IDs, max number of
partition table entries, etc.).
- Use NULL instead of '0' for null-pointers.
- Fix some typos in comments.
Introduce the PiPerformSyncDeviceAction routine for queuing
synchronous device actions
Change all kernel code to use PiPerformSyncDeviceAction and
PiQueueDeviceAction for device enumeration
CORE-10456
Use single IOCTL (IOCTL_DISK_GET_DRIVE_GEOMETRY_EX) for retrieving
disk basic geometry information along with disk size.
Previous implementation used to issue two requests for that.
And rearrange them in more logical order.
This effectively splits the file, leaving public "Io" functions in
pnpmgr.c along with some things not related do device object management.
Functions which manipulate the device tree are left in devaction.c.
In future all these functions will only be accessed from
DeviceActionWorker.
While being public API, IoRequestDeviceEject and IoInvalidateDeviceState
have been moved to devaction.c as well. In next commits they will be
converted to DeviceActionWorker routines and their callers will be put
in pnpmgr.c
If SEH is used in a C trap handler, the exception frame will be
registered before the call to KiEnterTrap, which means we save
the wrong trap handler. We'll therefore also restore this wrong
frame for the excepting code, resulting in a stale SEH chain.
We avoid this problem by saving the handler in the assembly
trap entry code instead of from C. While SEH in a C trap handler
should now theoretically be safe, we still forbid it through
asserts in the C KiEnterTrap variants to make any potential
future problems more obvious. Should this functionality be
needed at some point and deemed safe, these asserts can then be
removed.
- Sync some INIT_FUNCTION with how they are used and what is already
specified in the headers.
Addendum to commit 85e5b5be (r49445).
- KdbpGetCommandLineSettings(): Remove INIT_FUNCTION.
Fix MSVC 2015 x86 custom build:
"...\ntoskrnl\kdbg\kdb.c(1699): error C2983: 'KdbpGetCommandLineSettings': all declarations must have an identical __declspec(code_seg(...))"
And may also fix obscure bugs when entering into the KDBG debugger.
This version (after .rsrc) works different than the
proper version I used for 0.4.12 (after .reloc).
Inserting after .rsrc is actually not correct, but Thomas believes it can
be used as a temporary trick to avoid random memory corruption upon
relocations of the kernel, caused by ROSBE-154.
I follow his advice, although when judging from practical tests only:
as long as we limit this script to NTOSKRNL like I do for releases
there have no negative consequences been observed in real life yet
even with the proper version of 0.4.12.
Up to now those problems have only been observed when used for drivers
MODULE TYPE sdk/cmake/gcc.cmake as well, like
it was tried for a moment in master 0.4.13-dev-609-g
c4d8e2a6e9
Using for drivers immediately did lead to BSODs like CORE-16183 and therefore was
mitigated in master by total disabling of the scripts for both,
kernel and drivers in
0.4.13-dev-621-g
36e9a6f8dd
To allow installing DVDWritenow without BSOD,
we need the script at least for ntoskrnl!
I committed this patch (after .rsrc) already into 0.4.13RC and 0.4.14RC.