Bump version to 0.1.6

It hard-coded VR_X, so a target whose page tables expose no executable
region silently yielded zero xrefs. Callers now pass the scope (VR_X to
restrict to code, VR_R when the X bit is unavailable); the decoder and
exact target match remain the correctness gate. Mirrors the prot_any
parameter gva_imm_xref already carries.

.gitea/workflows/release.yml

@@ -70,3 +70,58 @@ jobs:
           curl -fsSL -X POST -H "$auth" \
             -F "attachment=@dist/${asset};type=application/zip" \
             "${api}/releases/${rid}/assets?name=${asset}"
+
+  deb:
+    runs-on: ubuntu-latest
+    container:
+      image: node:20-bookworm-slim
+    defaults:
+      run:
+        shell: bash
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Install toolchain
+        run: |
+          apt-get update
+          apt-get install -y --no-install-recommends \
+            cmake make gcc libc6-dev dpkg-dev file \
+            gcc-mingw-w64-x86-64 ca-certificates curl
+
+      - name: Configure
+        env:
+          TAG: ${{ github.ref_name }}
+        run: cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DVMIE_VERSION=${TAG#v}
+
+      - name: Build shared library
+        run: cmake --build build -j --target vmie_shared
+
+      - name: Package
+        run: |
+          set -euo pipefail
+          (cd build && cpack -G DEB)
+          ls -l build/*.deb
+
+      - name: Publish to Debian registry
+        env:
+          TOKEN: ${{ secrets.PUBLISH_TOKEN }}   # requires scope: package:write
+          SERVER: ${{ github.server_url }}
+          OWNER: ${{ github.repository_owner }}
+          DISTRIBUTION: stable
+          COMPONENT: main
+        run: |
+          set -euo pipefail
+          url="${SERVER}/api/packages/${OWNER}/debian/pool/${DISTRIBUTION}/${COMPONENT}/upload"
+          auth="Authorization: token ${TOKEN}"
+          for deb in build/*.deb; do
+            # 201 Created = uploaded; 409 Conflict = this version already present (re-run).
+            # Both are success; anything else fails. Do not use curl -f (it would fail 409).
+            code=$(curl -s -o /dev/null -w '%{http_code}' \
+              -X PUT -H "$auth" -T "$deb" "$url")
+            echo "$deb -> HTTP $code"
+            if [ "$code" != 201 ] && [ "$code" != 409 ]; then
+              echo "upload failed: $deb (HTTP $code)" >&2
+              exit 1
+            fi
+          done

CMakeLists.txt

@@ -1,15 +1,19 @@
 cmake_minimum_required(VERSION 3.18)   # find_program(... REQUIRED)
-project(vmi-engine C)
+
+set(VMIE_VERSION "0.1.6" CACHE STRING "Library version (MAJOR.MINOR.PATCH); CI passes the tag")
+project(vmi-engine VERSION ${VMIE_VERSION} LANGUAGES C)
 
 set(CMAKE_C_STANDARD 17)               # generation B uses no C23 feature
 set(CMAKE_C_STANDARD_REQUIRED ON)
 set(CMAKE_C_EXTENSIONS ON)             # deliberate: strnlen (POSIX) + void* arithmetic (GNU)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON) # -fPIC on all objects: static lib stays linkable into a shared object
 
+include(GNUInstallDirs)
+
 option(VMIE_LTO "Enable LTO" OFF)      # build-only; shipped default is -O2, no LTO
 
-# ---- host: VMI core as a static library ---------------------------------
-add_library(vmie STATIC
+# ---- host: VMI core objects (single owner of the source list) -----------
+add_library(vmie_obj OBJECT
         src/core/gpa.c
         src/engine/gva.c
         src/engine/sigphys.c
@@ -27,15 +31,32 @@ add_library(vmie STATIC
         src/handlers/x86dec.c
         src/handlers/pmap.c
         src/handlers/snapdiff.c)
-target_include_directories(vmie
+target_include_directories(vmie_obj
         PUBLIC  ${CMAKE_CURRENT_SOURCE_DIR}/include              # public API: include/*.h
         PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/src/core/include     # private: core.h
                 ${CMAKE_CURRENT_SOURCE_DIR}/src/engine/include   # private: engine-arch.h, pe.h
                 ${CMAKE_CURRENT_SOURCE_DIR}/src/engine/win32)    # private: engine-win32.h, contract.h
-target_compile_options(vmie PRIVATE -O2 -Wall -Wextra)
+target_compile_options(vmie_obj PRIVATE -O2 -Wall -Wextra)
 if(VMIE_LTO)
-    target_compile_options(vmie PRIVATE -flto)
-    target_link_options(vmie PRIVATE -flto)
+    target_compile_options(vmie_obj PRIVATE -flto)
+    target_link_options(vmie_obj PRIVATE -flto)
+endif()
+
+# ---- host: static library (in-tree consumers: CLIs + test) --------------
+# $<TARGET_OBJECTS:> carries object files only, not usage requirements:
+# re-declare the PUBLIC include scope so vmie_cli/vmie_scan still see include/.
+add_library(vmie STATIC $<TARGET_OBJECTS:vmie_obj>)
+target_include_directories(vmie PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include)
+
+# ---- host: shared library (external delivery: libvmie.so.*) -------------
+add_library(vmie_shared SHARED $<TARGET_OBJECTS:vmie_obj>)
+target_include_directories(vmie_shared PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include)
+set_target_properties(vmie_shared PROPERTIES
+        OUTPUT_NAME vmie
+        SOVERSION   ${PROJECT_VERSION_MAJOR}
+        VERSION     ${PROJECT_VERSION})
+if(VMIE_LTO)
+    target_link_options(vmie_shared PRIVATE -flto)
 endif()
 
 # ---- host: CLI demonstrator over the library ----------------------------
@@ -61,3 +82,44 @@ add_custom_command(
         COMMENT "Cross-compiling vmie-startup.exe (mingw-w64, x86-64)"
         VERBATIM)
 add_custom_target(vmie-startup ALL DEPENDS ${VMIE_STARTUP})
+
+# ---- install: shared library + public headers ---------------------------
+# Versioned file + SONAME symlink -> runtime; unversioned linker symlink -> dev.
+install(TARGETS vmie_shared
+        LIBRARY
+          DESTINATION ${CMAKE_INSTALL_LIBDIR}
+          COMPONENT runtime
+          NAMELINK_COMPONENT dev)
+install(DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/include/
+        DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/vmie
+        COMPONENT dev
+        FILES_MATCHING PATTERN "*.h")
+
+# ---- package: two Debian packages (runtime + dev) -----------------------
+set(CPACK_GENERATOR "DEB")
+set(CPACK_DEB_COMPONENT_INSTALL ON)
+set(CPACK_COMPONENTS_ALL runtime dev)
+set(CPACK_PACKAGE_VERSION ${PROJECT_VERSION})
+
+# component -> Debian package name (Debian shared-library convention)
+set(CPACK_DEBIAN_RUNTIME_PACKAGE_NAME "libvmie${PROJECT_VERSION_MAJOR}")
+set(CPACK_DEBIAN_DEV_PACKAGE_NAME     "libvmie-dev")
+
+set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "amd64")
+set(VMIE_DEB_MAINTAINER "Gregory Lirent <opensource@lirent.ru>"
+    CACHE STRING "Debian package Maintainer field")
+set(CPACK_DEBIAN_PACKAGE_MAINTAINER "${VMIE_DEB_MAINTAINER}")
+set(CPACK_DEBIAN_PACKAGE_SECTION "libs")
+set(CPACK_DEBIAN_PACKAGE_PRIORITY "optional")
+set(CPACK_DEBIAN_FILE_NAME "DEB-DEFAULT")
+
+set(CPACK_COMPONENT_RUNTIME_DESCRIPTION
+    "VM introspection engine (Windows-guest memory introspection) - shared library")
+set(CPACK_COMPONENT_DEV_DESCRIPTION
+    "VM introspection engine (Windows-guest memory introspection) - development files")
+
+# dev depends on the exact runtime version; runtime libc dependency via shlibdeps
+set(CPACK_DEBIAN_DEV_PACKAGE_DEPENDS "libvmie${PROJECT_VERSION_MAJOR} (= ${PROJECT_VERSION})")
+set(CPACK_DEBIAN_PACKAGE_SHLIBDEPS ON)
+
+include(CPack)

include/memmodel.h

@@ -125,7 +125,11 @@ typedef struct {
  * x86-64 has no read bit: a present page is readable, so VR_R is always set on a
  * returned region. Write/execute/user are the EFFECTIVE rights along the whole
  * page-table path (RW & US are AND-ed across levels, NX is OR-ed), not just the
- * leaf entry, so they reflect what the guest CPU actually enforces. */
+ * leaf entry, so they reflect what the guest CPU actually enforces. One
+ * exception keeps VR_X honest under KVA shadow (KPTI): the kernel CR3 marks
+ * user-half PML4Es NX as hardening, but ring 3 executes those pages via the
+ * user/shadow CR3 (same shared lower tables, NX-clear PML4E), so the user-half
+ * top-level NX bit is ignored when deriving VR_X. */
 #ifndef VMIE_VREGION_DEFINED
 #define VMIE_VREGION_DEFINED
 #define VR_R 0x1u   /* readable  (present => always set)            */

include/scan.h

@@ -70,18 +70,21 @@ int gva_sig_scan_multi(vmie_mem* m, uintptr_t cr3, uint64_t lo, uint64_t hi,
                        uint32_t prot_any, const sigset* s,
                        sig_multi_hit* out, int max);
 
-/* code-xref: every instruction in the X-regions of [lo,hi] whose near rel
- * branch or RIP-relative memory operand resolves to `target_va`. Brute-scans
- * each byte offset with the light x86-64 decoder (x86dec.h, NOT a full
- * disassembler): an E8/E9/EB/Jcc rel branch matches when next_rip + rel ==
+/* code-xref: every instruction in the regions of [lo,hi] kept by `prot_any`
+ * whose near rel branch or RIP-relative memory operand resolves to `target_va`.
+ * Brute-scans each byte offset with the light x86-64 decoder (x86dec.h, NOT a
+ * full disassembler): an E8/E9/EB/Jcc rel branch matches when next_rip + rel ==
  * target_va, and any RIP-relative operand (ModRM mod=00, rm=101) matches when
  * next_rip + disp32 == target_va (this covers lea/mov and any other rip-rel
- * form). Records each matching instruction-start VA. The sweep forces VR_X and
- * carries a >=15-byte overlap (max x86 instruction length) so no instruction is
- * cut at a window seam. Writes up to `max` VAs to `out` (NULL to count only) and
- * returns the TOTAL number of matches, or -1 on bad input. */
+ * form). Records each matching instruction-start VA. The sweep is scoped by
+ * `prot_any` (pass VR_X to restrict to code; pass VR_R when the target's
+ * page-table X bit is unavailable - the decoder and exact target match keep the
+ * result correct, only more bytes are decoded) and carries a >=15-byte overlap
+ * (max x86 instruction length) so no instruction is cut at a window seam. Writes
+ * up to `max` VAs to `out` (NULL to count only) and returns the TOTAL number of
+ * matches, or -1 on bad input. */
 int gva_code_xref(vmie_mem* m, uintptr_t cr3, uint64_t lo, uint64_t hi,
-                  uint64_t target_va, uint64_t* out, int max);
+                  uint32_t prot_any, uint64_t target_va, uint64_t* out, int max);
 
 /* immediate / constant xref: every instruction in [lo,hi] (kept by the
  * protection filter `prot_any`; pass VR_X to restrict to code) whose IMMEDIATE

include/win32.h

@@ -93,18 +93,18 @@ vmie_win32* vmie_win32_open(const char* ram_path, uint64_t low);
 vmie_win32* vmie_win32_open_fd(int fd, uint64_t low);
 
 /* Open a READ-ONLY context over the RAM backing file `fd`, for a consumer that
- * only reads. Unlike vmie_win32_open_fd this does NOT bootstrap (the magic-
- * signature handshake needs RW): the caller supplies the address space by passing
- * a cr3 to each read call. Every read API that takes a cr3 works - the section /
- * function / import / export / callgraph / hook surfaces, gva_read_text, and the
- * generic gva_* and scanner surfaces via vmie_win32_mem(); writes (gva_write) -1
- * because the mapping is PROT_READ. proc_list/proc_modules are NOT available here:
- * they need the bootstrap-recovered profile, which a read-only context does not
- * carry. `fd` is BORROWED (dup()'d internally; the caller's fd stays valid after
- * this call AND after vmie_win32_close()); an O_RDONLY fd is accepted. `low` is
- * the below-4G split, as in vmie_win32_open_fd. Returns a new context, or NULL on
- * dup/fstat/mmap failure. Free with vmie_win32_close(). */
-vmie_win32* vmie_win32_open_ro_fd(int fd, uint64_t low);
+ * only reads. It takes a caller-supplied `kcr3` (kernel cr3 / System DirectoryTableBase)
+ * and builds the struct-offset profile read-only from the image (no bootstrap
+ * handshake - beacon and ACK are not used). The result has the FULL read
+ * capability: proc_list/proc_modules AND the section / function / import / export
+ * / callgraph / hook surfaces, gva_read_text, and the generic gva_* and scanner
+ * surfaces via vmie_win32_mem(). Any write (gva_write) returns -1 because the
+ * mapping is PROT_READ. `fd` is BORROWED (dup()'d internally; the caller's fd
+ * stays valid after this call AND after vmie_win32_close()); an O_RDONLY fd is
+ * accepted. `low` is the below-4G split, as in vmie_win32_open_fd. Returns a new
+ * context, or NULL if `kcr3` does not resolve a kernel (ntoskrnl not found under
+ * it) or on dup/fstat/mmap failure. Free with vmie_win32_close(). */
+vmie_win32* vmie_win32_open_ro_fd(int fd, uint64_t low, uint64_t kcr3);
 
 /* Unmap, close, and free a context. Safe on NULL. After this, every pointer
  * into guest memory obtained through this context is invalid. */

src/engine/gva.c

@@ -158,7 +158,16 @@ int gva_regions(vmie_mem* m, uintptr_t cr3, uint64_t lo, uint64_t hi,
         if (!(e4 & PG_P)) continue;
         const uint64_t b4 = VA_CANON((uint64_t)i4 << 39);
         if (!rgn_hit(b4, 1ull << 39, lo, hi)) continue;
-        const int rw4 = (e4 >> 1) & 1, us4 = (e4 >> 2) & 1, nx4 = (int)(e4 >> 63) & 1;
+        const int rw4 = (e4 >> 1) & 1, us4 = (e4 >> 2) & 1;
+        /* On a KVA-shadow (KPTI) Windows the kernel CR3 marks user-half PML4Es
+         * NX as hardening: the kernel must not execute user pages through its own
+         * mapping. Ring 3 runs under the user/shadow CR3, which reaches the SAME
+         * lower tables (PDPT/PD/PT are shared) through a PML4E with NX clear, so
+         * that top-level NX is not what executes user code. Drop it on the user
+         * half; sub-PML4E NX stays authoritative (it is shared between the two
+         * CR3s). On a non-KPTI guest a code-covering user PML4E already has NX
+         * clear, so this only ever discards a zero. i4 < 256 == canonical user. */
+        const int nx4 = (i4 < 256) ? 0 : ((int)(e4 >> 63) & 1);
 
         const uint64_t* t3 = gpa_ptr(m, e4 & PFN_MASK, 4096);
         if (!t3) continue;

src/engine/win32/host.c

@@ -6,6 +6,9 @@
 
 #define MZ           0x5A4Du
 
+__attribute__((cold))
+static int host_profile_from_cr3(vmie_win32* v, uintptr_t cr3);
+
 /* ---- lifecycle (cold) ---------------------------------------------------- */
 
 __attribute__((cold))
@@ -34,10 +37,12 @@ vmie_win32* vmie_win32_open_fd(int fd, uint64_t low) {
     return v;
 }
 
-/* Read-only context: no bootstrap, the cr3 is supplied per read call. The mem is
- * mapped PROT_READ (gpa_from_ro_fd), so every write through it returns -1. */
+/* Read-only consumer context: maps the backing file PROT_READ and builds the
+ * struct-offset profile from the image under the caller-supplied kcr3 (no
+ * beacon, no cr3 recovery, no ACK write). All read APIs work, incl.
+ * proc_list/proc_modules; writes return -1. */
 __attribute__((cold))
-vmie_win32* vmie_win32_open_ro_fd(int fd, uint64_t low) {
+vmie_win32* vmie_win32_open_ro_fd(int fd, uint64_t low, uint64_t kcr3) {
     vmie_win32* v = calloc(1, sizeof *v);
     if (!v) {
         return NULL;
@@ -46,6 +51,10 @@ vmie_win32* vmie_win32_open_ro_fd(int fd, uint64_t low) {
         free(v);
         return NULL;
     }
+    if (host_profile_from_cr3(v, (uintptr_t)kcr3)) {
+        vmie_win32_close(v);
+        return NULL;
+    }
     return v;
 }
 
@@ -181,6 +190,31 @@ static uint32_t ko_export_rva(vmie_mem* m, uintptr_t cr3, uint64_t kbase, const
     return 0;
 }
 
+__attribute__((cold))
+static int host_profile_from_cr3(vmie_win32* v, uintptr_t cr3) {
+    vmie_mem* m = &v->mem;
+    uint8_t  guid[16];
+    uint32_t age, rva;
+    uint64_t sys_ep, dtb;
+
+    if (find_ntoskrnl(m, cr3, &v->kbase, guid, &age)) {
+        return -3;
+    }
+    rva = ko_export_rva(m, cr3, v->kbase, "PsInitialSystemProcess");
+    if (!rva || gva_read(m, cr3, v->kbase + rva, &sys_ep, 8)) {
+        return -4;
+    }
+    if (profile_build(v, cr3, sys_ep, guid, age)) {
+        return -5;
+    }
+    if (gva_read(m, cr3, sys_ep + v->prof.ep_dtb, &dtb, 8)) {
+        return -6;
+    }
+    v->kcr3    = dtb & PFN_MASK;
+    v->sysproc = sys_ep;
+    return 0;
+}
+
 static void beacon_ack(vmie_mem* m, uint64_t anchor_pa) {
     uint64_t ack = CONTRACT_MAGIC_R;
     gpa_write(m, anchor_pa + offsetof(contract, ack), &ack, 8);
@@ -195,10 +229,6 @@ int host_bootstrap(vmie_win32* v) {
     vmie_mem* m = &v->mem;
     uint64_t anchor_pa, va_self;
     uintptr_t cr3boot;
-    uint32_t rva;
-    uint8_t guid[16];
-    uint32_t age;
-    uint64_t sys_ep;
 
     if (beacon_find(m, &anchor_pa, &va_self)) {
         return -1;
@@ -208,26 +238,11 @@ int host_bootstrap(vmie_win32* v) {
         return -2;
     }
 
-    if (find_ntoskrnl(m, cr3boot, &v->kbase, guid, &age)) {
-        return -3;
+    const int rc = host_profile_from_cr3(v, cr3boot);
+    if (rc) {
+        return rc;
     }
 
-    rva = ko_export_rva(m, cr3boot, v->kbase, "PsInitialSystemProcess");
-    if (!rva || gva_read(m, cr3boot, v->kbase + rva, &sys_ep, 8)) {
-        return -4;
-    }
-
-    if (profile_build(v, cr3boot, sys_ep, guid, age)) {
-        return -5;
-    }
-
-    uint64_t dtb;
-    if (gva_read(m, cr3boot, sys_ep + v->prof.ep_dtb, &dtb, 8)) {
-        return -6;
-    }
-    v->kcr3    = dtb & PFN_MASK;
-    v->sysproc = sys_ep;
-
     beacon_ack(m, anchor_pa);
     return 0;
 }

src/handlers/codescan.c

@@ -115,10 +115,10 @@ static int xref_sweep_cb(void* u, const uint8_t* data, size_t len,
 }
 
 int gva_code_xref(vmie_mem* m, uintptr_t cr3, uint64_t lo, uint64_t hi,
-                  uint64_t target_va, uint64_t* out, int max) {
+                  uint32_t prot_any, uint64_t target_va, uint64_t* out, int max) {
     struct xref_cb c; memset(&c, 0, sizeof c);
     c.target = target_va; c.out = out; c.max = max;
-    if (gva_sweep(m, cr3, lo, hi, VR_X, X86_MAX_INSN, xref_sweep_cb, &c) < 0) {
+    if (gva_sweep(m, cr3, lo, hi, prot_any, X86_MAX_INSN, xref_sweep_cb, &c) < 0) {
         return -1;
     }
     return c.n;