Hidden volumes are forced to quick format to avoid rewriting the hidden data area. Keep that behavior while skipping the file-container allocation shortcut that writes plaintext zero sectors at 128 MiB intervals.
The allocation shortcut remains enabled for non-hidden file containers; hidden containers now use only the encrypted formatter write path for sectors that are written.
Enhance build_msi_x64.bat to enumerate installed Windows Kits 10 SDK bin directories matching 10.* and select the newest x86 path that contains the MSI tools.
Keep VC_DIR_PLATFORMSDK as the first override and preserve the existing fixed SDK fallback paths for older installations.
Require MsiInfo.exe during discovery as well as msitran.exe and msidb.exe so the selected SDK path supports the final MSI metadata step.
Replace outdated Visual Studio 2010/2019 and legacy Windows SDK 7.1, WDK 7.1, and Windows 8.1 SDK guidance with the current Visual Studio 2022/v143 toolchain, Windows 10/11 SDK, and WDK requirements.
Document NASM, YASM, WiX Toolset v3.x, signtool.exe, and optional legacy BIOS bootloader tools separately. Update the build flow for x64, ARM64, Win32 setup/helper projects, and explicit Driver project builds.
Align the zh-cn and ru translated guides with the updated English content while preserving their existing translation style. Fix test certificate paths to use src/Signing/TestCertificate.
Route Linux GUI mounted-volume opens through Windows Explorer when WSL interop is available, before falling back to xdg-open and known file managers.
Detect WSL by checking for /usr/bin/wslinfo and /usr/bin/wslpath, build the target path from the WSL root UNC so /mnt/<drive> mount points stay in the WSL VFS overlay, and launch Explorer directly so the folder argument is preserved.
BMI2 support is advertised by CPUID leaf 7, subleaf 0, EBX bit 8. The previous early assignment used CPUID leaf 1 EBX bit 8, which is not the BMI2 feature bit and could leave a bogus fallback value before vendor-specific leaf 7 detection.
Keep BMI2 detection based on the leaf 7 result only. Unlike AVX2, BMI2 is GPR-only and does not require an OS/XCR0 state gate.
Also save the max basic CPUID leaf immediately after CPUID leaf 0. The AMD/Hygon path reuses the cpuid buffer for leaf 0x80000005 before checking whether leaf 7 is available, so using the saved max basic leaf prevents RDSEED, AVX2, and BMI2 detection from being skipped because that buffer was clobbered.
AVX2 support is advertised by CPUID leaf 7, subleaf 0, EBX bit 5. The previous early assignment used cpuid1[1] bit 5, which is CPUID leaf 1 EBX and is not the AVX2 feature bit.
Record the leaf 7 AVX2 bit separately and assign g_hasAVX2 only after vendor-specific detection has completed. The final value is now gated by g_hasAVX, which reflects the OS/XCR0 AVX state check, so AVX2 code is not selected unless both the CPU and OS state support it.
Move the Unmount All mnemonic away from the single-volume Unmount action in the Windows resources and affected language files. This keeps the two main actions reachable through distinct keyboard accelerators across packaged translations.
Fixes https://github.com/veracrypt/VeraCrypt/issues/1751
Ensure ESP file writes have true replace semantics even when the operation is delegated to the elevated COM helper. This prevents shorter edits of EFI\VeraCrypt\DcsProp from leaving stale bytes at the end of the file.
Also XML-escape decoded EFI boot configuration values before serializing them, preserving values containing characters such as <, > and & during EfiBootConf save/update paths.
Fixes#954.
Mount temporary ReFS volumes as fixed media, since Windows does not support ReFS on removable media. Use FMIFS_HARDDISK for the FormatEx fallback while preserving the removable-media path for NTFS and exFAT.
Also make the FormatEx DONE-with-failure status explicit and guard against a missing callback parameter.
The GUI single-instance lock was previously created through wxSingleInstanceChecker without an explicit Unix path, causing wxWidgets to place .VeraCrypt-lock-$USER directly in the user home directory.
Resolve a private lock directory before constructing wxSingleInstanceChecker. Prefer $XDG_RUNTIME_DIR/VeraCrypt, then $XDG_CACHE_HOME/VeraCrypt, then ~/.cache/VeraCrypt, and keep the previous home-directory behavior only as a final fallback if no XDG location can be used.
Update stale-lock cleanup to remove the lock from the same resolved directory, so false-positive cleanup continues to work after moving the lock out of $HOME.
Fixes https://github.com/veracrypt/VeraCrypt/issues/819
The Unix volume creation wizard applied the FAT32 sector-count limit as a blanket check for device-hosted hidden-volume outer volumes. On 512e disks Linux reports 512-byte logical sectors, so this incorrectly rejected larger device-hosted outer volumes even when the selected outer filesystem was not FAT.
Compute the actual VeraCrypt filesystem/data area size through a shared helper and apply the FAT32 size limit only when FAT is selected. This preserves correct FAT validation while allowing non-FAT outer volumes to proceed to the existing hidden-volume size estimation flow.
Update text-mode creation so FAT is not offered when the selected size cannot support it, and default to the platform native filesystem in that case. Clarify the user-facing FAT limit wording to refer to logical sector size.
Fixes#262
On macOS, the same whole disk can be addressed as both /dev/diskN and /dev/rdiskN. The GUI creation wizard only compared the selected path against the enumerated raw device path, so manually entering the block-device alias could bypass the existing DEVICE_PARTITIONS_ERR guard and allow formatting a disk that still had partitions.
Add a shared macOS device-path comparison helper that normalizes paths to their raw-device form before comparison. Use it in the GUI wizard so /dev/diskN and /dev/rdiskN are treated as the same whole-disk target while partition paths remain distinct.
Apply the same partitioned whole-device guard in the text/CLI creation path as well, including the macOS alias normalization, so command-line creation cannot format a partitioned top-level disk through an alternate device alias.
Fixes#728
Remove the remaining generated-form alignment flag that wxWidgets ignores in box sizers: the language page system-default button bottom alignment combined with wxEXPAND. Preserve the Legal Notices OK button centering and keep Forms.cpp and TrueCrypt.fbp in sync.
Keep the existing global sizer consistency check suppressions in place pending additional testing.
Follow-up to issue #49.
Bundle the FUSE2 userspace library inside the AppImage AppDir and make AppRun prefer APPDIR/usr/lib. This lets the bundled VeraCrypt binary resolve libfuse.so.2 on systems where FUSE2 userspace packages are no longer installed by default.
Name AppImage artifacts according to the GTK backend detected during the build. GTK3 builds keep the default VeraCrypt-<version>-<arch>.AppImage name, while GTK2 builds use a gtk2-legacy suffix to distinguish the legacy compatibility artifact.
Include immintrin.h in the Argon2 AVX2 implementation so GCC toolchains such as the one on CentOS 7 see the AVX2 intrinsic types when compiling with -mavx2.
Refs: https://github.com/veracrypt/VeraCrypt/issues/1595
CentOS 6 builds VeraCrypt with GCC 4.4.7 and -std=c++0x. That compiler does not support range-based for loops, and its libstdc++ does not provide std::string::back() or std::string::pop_back().
Avoid those constructs in the affected Unix/Linux code paths: use VeraCrypt's existing foreach helper when iterating PKCS#11 object handles, and use indexing plus erase() when trimming trailing slashes from PATH entries.
This keeps the code valid for newer Linux toolchains while restoring compatibility with the CentOS 6 build environment.
Stage VeraCrypt and wxWidgets sources under the SDK package directory before rendering the OpenWrt package Makefile. The generated recipe now refers only to fixed package-local paths, so checkout and work directory names are no longer parsed as GNU Make syntax or passed unquoted through recipe source arguments.
Validate VeraCrypt and wxWidgets version tokens before substituting them into generated package metadata. This prevents unexpected Make metacharacters from entering the generated OpenWrt recipe while preserving normal dotted release versions.
Quote OpenWrt QEMU test container-size values with the existing shell quoting helper, matching the password handling and preventing user-supplied size text from being split or interpreted by the guest shell.
OpenBSD builds were relying on ggod to generate embedded resource
headers. That tool is not available on a stock OpenBSD 7.9 install,
and using base od directly is not a safe substitute because it emits
zero-padded decimal values such as 060 and 098. Those tokens are then
included in C++ source and parsed as octal constants, which either
changes values or fails compilation.
Use hexdump with an explicit unsigned-byte format for OpenBSD. It is
part of the base system and emits unpadded decimal byte values suitable
for the existing resource-header pipeline.
The text-mode binary also crashed on normal process exit on OpenBSD,
including after --version, --test, create, mount, list, and dismount.
GDB showed the crash in libpcsclite_real during SCardReleaseContext(),
called from the static SCardManager destructor. This happened even for
commands that did not use EMV or security-token support because the
static manager constructor eagerly initialized PC/SC at startup.
Avoid eager PC/SC initialization and exit-time finalization on OpenBSD.
The existing call sites still initialize PC/SC lazily when EMV/token
operations need it, while ordinary CLI commands no longer touch
pcsc-lite and no longer crash during static destruction.
Validated on OpenBSD 7.9 amd64 with:
- gmake NOGUI=1 -j2
- veracrypt --text --version
- veracrypt --text --test
- device-hosted create/mount/list/dismount smoke test through doas/vnd
Refs #1589.
Refs #1593.
VeraCrypt derives the real (non-root) user from SUDO_UID/SUDO_GID
to set default mount-point ownership and the FUSE service access
filter. On OpenBSD, privileged commands are normally run through doas,
which exposes the invoking login name via DOAS_USER and does not set
the sudo variables. As a result, VeraCrypt launched through doas
attributes both to root instead of the invoking user.
When the sudo identity variables are absent, resolve DOAS_USER through
the password database and use that uid/gid for default mount-point
ownership and the VeraCrypt FUSE service access filter. sudo behavior
is unchanged.
This is a correctness fix for the doas launch path. It is not confirmed
to resolve the non-root ext2fs EACCES reported in the linked issues:
that failure occurs at the ext2fs layer reached through vnd, whose
backing-image I/O runs as root and is therefore already permitted by
the access filter.
Refs #1589.
Refs #1593.
OpenBSD device length detection was returning the raw disk sector count from DIOCGPDINFO directly. That value is not bytes and it describes the physical/default disk label, which caused VeraCrypt to expose an incorrectly sized FUSE backing image through vnd for device-hosted volumes.
Use the current disklabel from DIOCGDINFO, derive the opened partition from the device minor number, and return the selected partition size in bytes. Keep the raw c partition on the whole-disk path by using DL_GETDSIZE there.
Also reject sector-misaligned device-hosted sizes during volume creation so new malformed OpenBSD device-hosted volumes are not created. Do not reject existing malformed headers at mount time, so users can still mount old OpenBSD-created volumes for recovery.
Refs #1589.
Refs #1593.
Add OpenWrt SDK packaging under src/Build for console-only x86/64 builds. The build helper prepares the SDK, renders a local package recipe, builds VeraCrypt with the OpenWrt musl toolchain, uses wxWidgets 3.2.10 as static wxBase, enables FUSE3, and skips release self-tests during cross compilation.
Add a package template that installs the console binary, mount.veracrypt, and license files only. The package declares bash for mount.veracrypt and keeps runtime dependencies focused on the direct userland requirements.
Add a documented QEMU runtime test path that boots the matching OpenWrt image, installs the locally built package set with opkg, runs the VeraCrypt version and algorithm self-tests, and exercises a small filesystem=none container mount/unmount flow.
Allow wxbuild callers to pass WX_CONFIGURE_EXTRA_FLAGS so OpenWrt cross configure flags can be passed into the wxWidgets build without carrying an OpenWrt-specific source patch.
Extend the Unix encryption thread pool to run key-derivation work items and use it when mounting volumes without an explicitly selected KDF. This brings Linux/macOS header PRF autodetection closer to the Windows path while keeping selected-KDF mounts unchanged.
Fixes#1610.
When preparing UEFI system encryption, check for the standard Windows bootmgfw.efi path before reading it. If it is absent, show the existing VeraCrypt diagnostic instead of surfacing a generic file-not-found error from the elevated COM path.
Clarify that Windows startup-managed VHD/VHDX files, including Dev Drive backing images, cannot live on system favorite volumes because they are accessed before those volumes are mounted.
Document that native-boot VHD/VHDX files also cannot live on system favorite volumes and remain subject to the existing VeraCrypt pre-boot authentication limitation for operating systems installed within VHD/VHDX files.
Mention a delayed/retrying attach workaround for non-boot-critical VHD/VHDX files after VeraCryptSystemFavorites mounts the host volume.
Closes#1605.
Use a stable VeraCrypt Start Menu folder for MSI installs instead of deriving it from the versioned product name. Refresh the shortcut component identities for the new folder location and add upgrade-time cleanup for old versioned VeraCrypt Start Menu folders while preserving folders that contain non-VeraCrypt content.
Fixes#1631.
Fixes#1624
The language file path was hardcoded to /usr/share/veracrypt/languages/
which doesn't exist inside an AppImage runtime. Language files are
actually located under $APPDIR/usr/share/veracrypt/languages/ when
running from an AppImage.
This affected both the language file loading in Resources.cpp and the
language enumeration in PreferencesDialog.cpp, causing the language
selection to show only "System default" and "English" regardless of
which translations were packaged in the AppImage.
When a second GUI process successfully notifies the running instance through the show-request FIFO, the handoff is not an error. Avoid showing the informational modal before exiting, and let the running instance restore the main window on any show request.
Also initialize the GTK indicator menu item pointers to NULL and guard the show/hide label update, preventing a latent crash in SetBackgroundMode when the indicator menu has not been built (e.g. background task disabled in preferences) -- a path made more reachable by the FIFO timer now invoking SetBackgroundMode unconditionally on incoming show requests.
Fixes#1447.
Closes#1745.
Refs #461.
Route IRP_MJ_FLUSH_BUFFERS through EncryptedIoQueue for mounted writable non-system volumes. Flushes are represented as zero-length queue items handled by the I/O thread, so ZwFlushBuffersFile runs after earlier encrypted write fragments before completing to the caller.
Also perform a best-effort ZwFlushBuffersFile before closing writable mounted-volume host handles, after the encrypted I/O queue has drained, so clean dismount/shutdown paths push the host file or raw device before close.
This keeps the change focused on ordinary mounted-volume flush ordering and avoids system-encryption, boot-drive, and header-update paths.
Allow normal file-hosted containers to use quick format in the Unix volume creation path by sizing the host file with ftruncate before backup headers are written.
Enable the GUI checkbox for normal file containers and honor --quick in text mode. Update the Unix HTML documentation for the weaker deniability properties of sparse or unwritten host regions.
Argon2id includes the requested output length in its computation, so deriving 192 bytes and using a prefix is not equivalent to deriving only the selected cipher's key material length. This differs from PBKDF2, where the prefix property made this detail invisible.
VeraCrypt derives the maximum header key material currently needed by the supported cipher/cascade set, which is 192 bytes, and then uses the required prefix for the selected encryption algorithm. For AES-XTS this means the first 64 bytes of the 192-byte Argon2id output are used.
Make this design rule explicit in code and documentation by introducing ARGON2_HEADER_KEYDATA_SIZE instead of relying implicitly on GetMaxPkcs5OutSize. If a future cipher or cascade requires more than 192 bytes, that must be handled as an explicit format/design change.
Document the 192-byte Argon2id header KDF output requirement so third-party implementations derive the same header key material.
References: https://github.com/veracrypt/VeraCrypt/issues/1614
Embed both Microsoft UEFI CA 2011 and 2023 signed DCS EFI sets and select the 2023 set only when the firmware db trusts the required 2023 third-party CAs.
Fall back to the 2011 EFI set when firmware db state cannot be determined, preserving pre-existing compatibility behavior and recording the reason in HKLM diagnostics.
Refresh installed ESP modules during PostOOBE repair, keep backups before replacing existing DCS modules, and use the selected EFI set when creating rescue media.
Record the selected EFI bootloader resource set and selection reason in HKLM, allow larger firmware db variables on systems with many Secure Boot certificates, and remove diagnostic registry keys on uninstall.
Fix MSI SetupDLL COM typelib version constants so unregister targets the current Main and Format COM typelib versions.
References: https://github.com/veracrypt/VeraCrypt/issues/1655
Keep the NTFS kernel-driver option as a generic in-kernel NTFS path rather than an ntfs3-specific path. Add --filesystem=kernel-ntfs and -m kernelntfs routes that select a registered or loadable kernel NTFS driver and mount with -i so mount.ntfs/ntfs-3g helpers are not invoked.
Preserve --filesystem=ntfs3 as a literal pin to the ntfs3 driver. Treat both ntfs3 and kernel-ntfs as mount-only selectors; volume creation continues to use filesystem type NTFS.
The preference and -m kernelntfs path only select an in-kernel NTFS driver when no explicit filesystem type was supplied and blkid detects NTFS.
Treat ntfs as the preferred in-kernel driver on Linux 7.1 and later, where the upstream read/write driver is expected. On earlier kernels, select ntfs only when module metadata identifies the standalone read/write driver and /sys/module confirms it loaded, avoiding ntfs3 read-only ntfs compatibility registrations. Fall back to ntfs3 otherwise, and report a generic kernel-driver error if neither supported driver is available or loadable.
Rename the internal preference/config field to MountNtfsWithKernelDriver, migrate the old MountNtfsWithNtfs3 preference key, and update UI strings, CLI help, documentation, release notes, and translation placeholders accordingly.
Reference: https://github.com/veracrypt/VeraCrypt/issues/1735
Replace the Linux ntfs3 help icon paint lambda with a small wxWindow
subclass and regular paint event handler.
GCC 4.4, used on CentOS 6, builds with -std=c++0x but does not support
the lambda syntax used in PreferencesDialog.cpp, causing compilation to
fail at the ntfs3 help icon handler.
The drawing behavior is unchanged.
Keep the executable requirement at CMake 2.8.12 for legacy CentOS 6 package builders while using the version-range syntax to declare policy compatibility up to 3.10. Newer CMake versions use the policy maximum to avoid CMake 4 failures, and older CMake versions ignore the suffix and continue to configure as before.
Enable use_ino for Linux FUSE mounts so stable inode numbers returned by getattr and readdir are reported to userspace. For FUSE3, set fuse_config.use_ino from init; for FUSE2, pass -o use_ino because there is no fuse_config init hook.
Also map otherwise unhandled FUSE exceptions to EIO instead of EINTR, since these failures are not signal interruptions and should not encourage retry loops.
APFS volume creation can still fail with Permission denied after preparing the raw and block device aliases because newfs_apfs performs privileged APFS container and volume operations beyond opening the device nodes.
Route APFS formatting through the elevated CoreService path for non-root macOS runs. Keep the elevated interface narrow by sending only the target device and invoking user UID/GID, validate the device path on the privileged side, rebuild the formatter arguments there, and execute /sbin/newfs_apfs by absolute path to avoid PATH shadowing.
Pass -U/-G so the created filesystem preserves the invoking user ownership. Apply the same path to GUI and text-mode creation.
When creating an APFS filesystem inside a newly created device-hosted volume, VeraCrypt prepared only the raw hdiutil device path before invoking newfs_apfs. On macOS, newfs_apfs may resolve or reopen the corresponding block device path, which can fail with Permission denied for non-root GUI runs.
Prepare both raw and block aliases for the temporary formatter device, restore changed owners afterward, and share the helper between GUI and text-mode volume creation. Restore each changed alias independently so one restore failure does not skip the rest.
Remove the duplicate Shell_NotifyIconW(NIM_MODIFY) call from the tray balloon helper. On Windows 11, the shell can queue the repeated NIF_INFO request and show the same unmount notification twice.
Fixes#1630
Reject empty or truncated TLV buffers, unsupported indefinite lengths, and declared value lengths that exceed the remaining input or uint16 node storage. Parse BER long-form lengths in big-endian order before copying value bytes.
Report explicit progress stages while writing volume data, writing backup headers, and flushing data to disk so the wizard does not appear stuck at 100%.
Keep the wizard in progress during Unix post-creation formatting and show status for temporary mount/device setup, mkfs invocation, and dismount.
When normal filesystem unmount fails, the Linux path could stop before cleaning VeraCrypt mapper, loop and FUSE objects. Add an explicit emergency dismount request that is only reached after interactive confirmation.
The recovery path lazy-detaches mounted filesystems, uses deferred dmsetup removal for VeraCrypt mapper devices, detaches loop devices, and keeps normal force/ignoreOpenFiles behavior unchanged.
Validate SecRegion password cache offsets before use.
Wipe decrypted SecRegion password-cache data even when cache validation fails.
Clamp encrypted I/O work item counts and check allocation sizing.
Reject invalid boot drive sector writes and initialize decoy wipe data unit.
Validate hidden-system boot offsets and remap arithmetic before use.
Verify restored EFI Microsoft and fallback boot loader paths after system decryption.
Show clearer recovery guidance when EFI file restoration or NVRAM cleanup remains incomplete.
Add a GPT-only EFI boot loader repair menu action for already decrypted systems.
Add statfs metadata for the auxiliary FUSE mount, keep /control read-only by sending hdiutil device data through /aux-device-info and tolerate delayed SMB rediscovery during mount completion. Log final control-file retry failures for diagnostics.
Use the same SystemFavorites service flag mapping for /PostOOBE bootloader repair as the non-forced service update path. This prevents Windows upgrade repair from adding VeraCrypt to BootOrder or forcing it first when those actions are disabled, while forced updates keep their previous behavior and still repair the loader.
Copy selected favorite volumes into mount thread parameters so background mounting does not depend on mutable global vector storage. Also avoid unnecessary FavoriteVolume copies in auto-mount paths and fix mount-on-arrival state updates.
Refs #1661
wxWidgets does not propagate mouse motion events from child controls to parent windows. The Linux/macOS GUI was binding the random-pool mouse handlers only to the dialog/page and its direct children, which left nested controls such as static-box contents and the wizard image as dead zones.
Add a reusable recursive child-window event binder and use it in the keyfile generator, random pool enrichment dialog, and volume creation wizard. The root windows keep their existing generated bindings, while descendants are bound explicitly, avoiding duplicate handling on the root while covering all nested controls.
This makes the entropy gauge and the random pool update consistently no matter where the pointer moves inside the affected windows.
Fixes#1656.
Local macOS builds disable universal binaries. On Apple Silicon, that means
the regular objects are built for arm64 only, but Volume.make still selected
the macOS assembly bundle. Several of those assembly outputs are x86_64 only,
which makes ranlib reject Volume.a because it contains mixed cputypes.
Make local arm64 macOS builds use the existing arm64 crypto object path and
have the macOS build wrapper export CPU_ARCH/COMPILE_ASM consistently for
local builds.
wxWidgets 3.3 removed the configure switches for explicitly enabling
Unicode and std_string support. Passing these options now makes configure
fail with "unrecognized options" when option checking is enabled.
Keep the legacy switches for older wxWidgets releases, but only add them
from the wxbuild recipe when the target configure script still supports the
corresponding Autoconf option variables.
Make MSI-installed VeraCrypt use the IDRIX-signed COMReg package as the
source for traveler files, matching the EXE installer flow. COMReg now
packages the x64 traveler payload, so traveler creation no longer has to
copy the installed x64 driver from appDir\veracrypt.sys and verify it
against a Microsoft WHQL certificate fingerprint.
Keep Microsoft WHQL certificate verification only for the loose portable
driver fallback, where driver files cannot be signed with the IDRIX code
signing certificate. The normal VerifyModuleSignature path now remains
IDRIX-only.
Also validate that an MSI COMReg package actually contains the required
x64 traveler files before reporting success, avoiding partial traveler
directories when the package payload is incomplete.
Now it is possible to enable IME during Secure Desktop using a setting in Preferences or using /enableIME switch.
This helps solve issues with some IME that causes VeraCrypt to freeze when selecting keyfile while VeraCrypt secure desktop is active.
See: https://sourceforge.net/p/veracrypt/discussion/general/thread/1e8b9aeacd
Introduce MapIrpDataBuffer to handle Direct/Buffered/Neither I/O, probing & locking pages and allocating a temp MDL when needed.
Replace blind MmGetSystemAddressForMdlSafe usage. clean up TempUserMdl in OnItemCompleted to avoid crashes when MdlAddress is NULL.
Issue reported at https://sourceforge.net/p/veracrypt/discussion/technical/thread/e43bde8d86/
- Replace dynamic GetProcAddress usage with direct SetThreadGroupAffinity call since we run under Windows 10 minimum
- Compute affinity mask based on actual active processor count
- Fix off-by-one when assigning threads to processor groups (use > instead of >=), preventing premature group advance
- Improves correctness on multi-group (>=64 CPU) systems
Replace timer-based TCSleep (which allocated a KTIMER and waited on it) with an
implementation that calls KeDelayExecutionThread. This removes dynamic allocation
and kernel timer usage to simplify the code and reduce resource overhead.
Adds an IRQL <= APC_LEVEL assertion and documents the requirement.
This is safe because TCSleep is always called from code that runs at PASSIVE_LEVEL
Added InterlockedDecrement in the error path when GetPoolBuffer fails for EncryptedIoRequest to ensure accurate tracking of pending IO requests and prevent potential resource leaks.
Major changes:
- Added pooled + elastic work item model with retry/backoff (MAX_WI_RETRIES). removed semaphore usage.
- Introduced two completion threads to reduce contention and latency under heavy IO.
- Added BytesCompleted (per IRP) and ActualBytes (per fragment) for correct short read/write accounting. total read/write stats now reflect real transferred bytes instead of requested length.
- Moved decryption of read fragments into IO thread. completion threads now only finalize IRPs (reduces race window and simplifies flow).
- Deferred final IRP completion via FinalizeOriginalIrp to avoid inline IoCompleteRequest re-entrancy. added safe OOM inline fallback.
- Implemented work item pool drain & orderly shutdown (ActiveWorkItems + NoActiveWorkItemsEvent) with robust stop protocol.
- Replaced semaphore-based work item acquisition with spin lock + free list + event (WorkItemAvailableEvent). added exponential backoff for transient exhaustion.
- Added elastic (on-demand) work item allocation with pool vs dynamic origin tracking (FromPool).
- Added FreeCompletionWorkItemPool() for symmetric cleanup; ensured all threads are explicitly awakened during stop.
- Added second completion thread replacing single CompletionThread.
- Hardened UpdateBuffer: fixed parameter name typo, added bounds/overflow checks using IntSafe (ULongLongAdd), validated Count, guarded sector end computation.
- Fixed GPT/system region write protection logic to pass correct length instead of end offset.
- Ensured ASSERTs use fragment‑relative bounds (cast + length) and avoided mixed 64/32 comparisons.
- Added MAX_WI_RETRIES constant. added WiRetryCount field in EncryptedIoRequest.
- Ensured RemoveLock is released only after all queue/accounting updates (OnItemCompleted).
- Reset/read-ahead logic preserved. read-ahead trigger now based on actual completion & zero pending fragment count.
- General refactoring, clearer separation of concerns (TryAcquireCompletionWorkItem / FinalizeOriginalIrp / HandleCompleteOriginalIrp).
Safety / correctness improvements:
- Accurate short read handling (STATUS_END_OF_FILE with true byte count).
- Eliminated risk of double free or premature RemoveLock release on completion paths.
- Prevented potential overflow in sector end arithmetic.
- Reduced contention and potential deadlock scenarios present with previous semaphore wait path.
- Document HIGH_LEVEL constraints and rationale for pre-building a nonpaged scratch MDL.
- Allocate contiguous scratch buffer with conservative PFN cap (0x7FFFFFFFFFF) and fall back to unlimited cap if needed.
- Replace ASSERT with TC_BUG_CHECK for validation of write MDL mapping at HIGH_LEVEL.
- Safely copy PFNs from prebuilt MDL into caller MDL: compute dst/src page counts, check capacity, copy exact PFNs and retarget MDL header fields (preserve MdlFlags).
- Make DumpData cleanup defensive in unload path.
- comments improvements for clarity and maintainability.
Mounir IDRASSI