vatrog-vm-introspection-engine/include/vmie.h

/* vmie.h - public Windows-guest surface of the vmi-engine.
 *
 * The host opens a guest's RAM backing file (a flat, writable, coherent mmap),
 * recovers the kernel address space, and reads/writes guest memory by CR3 and
 * virtual address. Everything is CR3-keyed, never PID-keyed: a `process` already
 * carries its own cr3, which is the key to that address space.
 *
 * This header is the Windows-typed surface (process/pmodule/gtext, bring-up,
 * enumeration, the win32 scan wrappers). The OS-agnostic memory-model contract
 * lives in memmodel.h (pulled in below); the scanners in scan.h/sigscan.h.
 *
 * Conventions:
 *   - `cr3` is a raw CR3 / DirectoryTableBase value; low flag bits are masked
 *     internally, so either the masked PML4 GPA or the raw register works.
 *   - Integer returns: 0 on success, negative on failure, unless stated.
 *   - The library never takes ownership of caller buffers and never retains a
 *     pointer past the call that received it, unless explicitly stated.
 */
#ifndef VMIE_VMIE_H
#define VMIE_VMIE_H
#include <stdint.h>
#include <stddef.h>
#include "memmodel.h"    /* vmie_mem, vregion/VR_*, task/range, gva_read/write/ptr/regions/sweep */
#include "sigscan.h"     /* mem_view_t, sig_pattern_t */
#include "scan.h"        /* scan_type, scan_ptr_path, generic scan surface */

/* Opaque introspection context. Completed in src/engine/include/engine.h;
 * callers only ever hold a pointer. Created by vmie_open(), populated by
 * host_bootstrap(), released by vmie_close(). */
typedef struct vmie vmie;

/* A guest counted string still resident in guest memory (e.g. a UNICODE_STRING
 * buffer). Not a copy: `va` points into the guest, decode it with gva_read_text.
 *   va  - guest VA of the first UTF-16LE code unit (0 if absent)
 *   len - length in BYTES (not characters); always even for UTF-16 */
typedef struct { uint64_t va; uint32_t len; } gtext;

/* A live process, as produced by proc_list(). Self-contained: `cr3` is all you
 * need to read/write its user address space, `eprocess`/`peb` re-anchor it in
 * kernel/user space without another lookup.
 *   cr3         - DirectoryTableBase (PFN-masked); key to this address space
 *   peb         - PEB VA (0 for system/kernel-only processes)
 *   eprocess    - _EPROCESS VA (kernel object, read under the kernel cr3)
 *   pid, ppid   - process / parent ids (ppid == (uint32_t)-1 if unavailable)
 *   create_time - raw KSYSTEM_TIME / FILETIME (100 ns ticks; 0 if unavailable)
 *   name        - ImageFileName, NUL-terminated ASCII (up to 15 chars)
 *   path        - full image path as a guest UTF-16 string (gtext; may be empty) */
typedef struct {
    uint64_t cr3;
    uint64_t peb;
    uint64_t eprocess;
    uint32_t pid;
    uint32_t ppid;
    uint64_t create_time;
    char     name[16];
    gtext    path;
} process;

/* A loaded module (image) inside a process, as produced by proc_modules().
 *   pr    - owning process (its cr3 is the address space these VAs live in)
 *   entry - _LDR_DATA_TABLE_ENTRY VA
 *   base  - image base VA (page-aligned); pair with `size` for a MODULE scope
 *   size  - image size in bytes (SizeOfImage)
 *   name  - module file name (gtext UTF-16, e.g. "ntdll.dll")
 *   path  - full module path (gtext UTF-16) */
typedef struct {
    const process* pr;
    uint64_t entry;
    uint64_t base;
    uint32_t size;
    gtext    name;
    gtext    path;
} pmodule;

/* ---- lifecycle ----------------------------------------------------------- */

/* Open `ram_path` (the guest RAM backing file) and build a context over it.
 *   ram_path - path to a writable, share=on RAM backing file
 *   low      - size in bytes of below-4G guest RAM (the PCI-hole split point);
 *              pass the value from the VM's memory layout. If total RAM <= low,
 *              the split is inert.
 * Returns a new context (call host_bootstrap() next), or NULL on open/mmap
 * failure. Free with vmie_close(). */
vmie* vmie_open(const char* ram_path, uint64_t low);

/* Unmap, close, and free a context. Safe on NULL. After this, every pointer
 * into guest memory obtained through this context is invalid. */
void vmie_close(vmie* v);

/* Borrow the engine's guest-memory handle for the generic address-space
 * primitives (gva_read/gva_regions/...). The returned pointer is owned by `v`
 * and valid until vmie_close(v); do NOT free or retain it past that. NULL on
 * NULL `v`. */
vmie_mem* vmie_memory(vmie* v);

/* One-shot bring-up: locate the guest agent beacon in physical RAM, recover a
 * bootstrap CR3, find ntoskrnl, build the struct-offset profile, derive the
 * permanent System DirectoryTableBase (kernel cr3) and System _EPROCESS, then
 * ACK the agent. On success the context is ready for proc_list()/gva_read()/etc.
 * Returns 0 on success, or a negative stage code (-1..-6) identifying the step
 * that failed. Cold path: call once after vmie_open(). */
int host_bootstrap(vmie* v);

/* ---- guest string decode ------------------------------------------------- */

/* Read a UTF-16LE guest string and transcode it to UTF-8.
 *   va    - guest VA of the first UTF-16 code unit
 *   nmemb - number of BYTES to read from the guest (rounded down to even)
 *   dst   - output buffer for NUL-terminated UTF-8 (may be NULL to size only)
 *   size  - capacity of `dst` in bytes
 * Returns the number of UTF-8 bytes the full conversion needs, EXCLUDING the
 * terminator (like snprintf): if it is >= `size`, output was truncated. When
 * `dst` is non-NULL and `size` > 0 the result is always NUL-terminated. */
size_t gva_read_text(vmie* v, uintptr_t cr3, uintptr_t va, size_t nmemb, char* dst, size_t size);

/* ---- enumeration --------------------------------------------------------- */

/* Enumerate processes by walking ActiveProcessLinks from System.
 *   skip_system - if nonzero, omit processes with no PEB (System/kernel-only)
 *   dst         - caller array receiving up to `nmax` `process` records
 *   nmax        - capacity of `dst`
 * Returns the number written (<= nmax), or negative on failure (e.g. bootstrap
 * not completed). Enumeration stops at `nmax`; raise it to see more. */
int proc_list(vmie* v, int skip_system, process* dst, size_t nmax);

/* Enumerate a process's loaded modules via the PEB loader InLoadOrder list.
 *   pr   - process to inspect (uses pr->cr3 and pr->peb)
 *   dst  - caller array receiving up to `nmax` `pmodule` records
 *   nmax - capacity of `dst`
 * Returns the number written (<= nmax), 0 if the process has no PEB/loader. */
int proc_modules(vmie* v, const process* pr, pmodule* dst, size_t nmax);

/* ---- win32 scan wrappers ------------------------------------------------- *
 * Convenience entry points over the generic cr3/range scan surface (scan.h).
 * They project a Windows `process` to its cr3, and its `pmodule[]` to a decoded
 * `range[]` (UTF-8 names), then delegate to scan_new_cr3 / scan_pointer. */

/* Open a value-scan session over the user address space of `pr`. Equivalent to
 * scan_new_cr3(&v->mem, pr->cr3, ...). Returns NULL on NULL pr or OOM. */
scan* scan_new(vmie* v, const process* pr, scan_type t, const void* value,
               int be, int aligned, uint64_t lo, uint64_t hi);

/* Pointer scan over `pr`'s user space, anchored on its loaded modules. Resolves
 * `pr`'s module list to range[] (names engine-decoded) and delegates to
 * scan_pointer. Returns the number of paths found, or negative on failure. */
int vmie_scan_pointer(vmie* v, const process* pr, uint64_t target,
                      int max_depth, uint32_t max_off, scan_ptr_path* out, int max);

#endif /* VMIE_VMIE_H */