惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

博客园 - Franky
N
Netflix TechBlog - Medium
宝玉的分享
宝玉的分享
Google DeepMind News
Google DeepMind News
腾讯CDC
G
Google Developers Blog
Martin Fowler
Martin Fowler
Microsoft Security Blog
Microsoft Security Blog
Recent Announcements
Recent Announcements
爱范儿
爱范儿
Engineering at Meta
Engineering at Meta
Microsoft Azure Blog
Microsoft Azure Blog
A
About on SuperTechFans
aimingoo的专栏
aimingoo的专栏
有赞技术团队
有赞技术团队
Jina AI
Jina AI
人人都是产品经理
人人都是产品经理
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
M
MIT News - Artificial intelligence
罗磊的独立博客
博客园 - 三生石上(FineUI控件)
美团技术团队
WordPress大学
WordPress大学
阮一峰的网络日志
阮一峰的网络日志

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
Reverse Engineering rpcss.dll: Hunting for the ROT's Hidd...
Pastukhov Al · 2026-05-06 · via DEV Community

Part IV(?) of the "Inside the Running Object Table" series

When I started this research, I assumed finding the internal layout of the Running Object Table would be straightforward. Microsoft documents IRunningObjectTable at the API level — surely the implementation couldn't be far behind. I was wrong. What followed was a Ghidra session that turned up something far more interesting than I expected.

Where does the ROT actually live?
First surprise: there is no rpcss.exe on disk.
What Windows Task Manager shows as rpcss.exe is actually svchost.exe hosting rpcss.dll, classic service host pattern. The DLL lives at C:\Windows\System32\rpcss.dll and gets mapped into a dedicated svchost instance at boot. Every ROT operation your process makes through ole32.dll crosses an ALPC channel and lands in that DLL's code.
So that's our target.

Loading rpcss.dll into Ghidra
Standard procedure — drag, drop, let the auto-analyzer run. The interesting part comes when you load the PDB. Microsoft publishes symbol files for most system DLLs through their public symbol server, and rpcss.dll is no exception.
Grab symchk.exe from your Windows SDK Debuggers folder and run:

symchk C:\Windows\System32\rpcss.dll ^
    /s srv*C:\Symbols*https://msdl.microsoft.com/download/symbols

Enter fullscreen mode Exit fullscreen mode

Point Ghidra at the downloaded PDB and suddenly FUN_18006XXXX becomes something readable. This is where it gets interesting.

First hit: CScmRotEntry
Searching the Symbol Table for ROT turns up the string u"ROTFlags" at address 180136070. One xref later, we're inside a large function. But the real find is in the Symbol Table itself:
CScmRotEntry::GetAllowAnyClient
Not CROTEntry. Not CRunningObjectTableEntry. CScmRotEntry - SCM stands for Service Control Manager. This tells us something immediately, the ROT isn't just a COM subsystem. It's wired into the service infrastructure at a deeper level than the documentation suggests.
The decompilation of GetAllowAnyClient reveals the first concrete field offset:

return *(uint *)(this + 0x68) >> 1;  // bit 1 = AllowAnyClient flag

Enter fullscreen mode Exit fullscreen mode

The real structure: LookupObjectInROT
Searching for CScmRot surfaces LookupObjectInROT - a function that takes ACTIVATION_PARAMS and walks the table looking for a matching object. From here we follow the call into CScmRot::GetObject, which is where the actual traversal happens.
The decompilation reveals the first architectural surprise:

cplVar13 = *(longlong **)(*(longlong *)(lpCriticalSection + 0x48) + uVar9 * 8);

Enter fullscreen mode Exit fullscreen mode

The ROT is not a linked list. It's a hash table with 251 buckets (0xFB).
Each bucket is a pointer to a chain of CScmRotEntry objects. The hash function is a simple rolling hash over the moniker bytes:
chash = ((hash * 3) ^ byte) % 0xFB;
Aint cryptographic n complex. Fast and collision-tolerant for the expected workload of a few dozen registered objects.

Digging deeper: the class hierarchy
Listing all symbols with the CScmRot prefix turns up the full method table:

CScmRot::EnumRunning
CScmRot::GetEntryFromScmReg
CScmRot::GetTimeOfLastChange
CScmRot::IsRunning
CScmRot::Register
CScmRot::Revoke
CScmRotEntry::CScmRotEntry
CScmRotEntry::GetProcessID
CScmRotEntry::IsValid

Enter fullscreen mode Exit fullscreen mode

Two things stand out. First, CScmRotEntry::CScmRotEntry it's a constructor, meaning this is a proper class with its own lifecycle. Second, CScmRotEntry::GetProcessID -every ROT entry stores the PID of the registering process. That field doesn't appear anywhere in the public API documentation.
Opening the constructor reveals something else entirely: CScmRotEntry is not a flat structure. It inherits from a base class:

CScmRotEntry
    └── CScmRotMgotEntryBase

Enter fullscreen mode Exit fullscreen mode

"Mgot" itsa likely Marshaled Global Object Table. The base class constructor initializes most of the core fields:

CScmRotMgotEntryBase::CScmRotMgotEntryBase(
    CScmRotMgotEntryBase *this,
    ulong param_1, ulong param_2,
    _MnkEqBuf *param_3,
    CToken *param_4,
    ushort *param_5,
    tagInterfaceData *param_6,
    bool param_7,
    ulong param_8)
{
    *(undefined8 *)(this + 0x08) = 0;       // pNext = NULL
    *(undefined4 *)(this + 0x10) = 1;       // ref count = 1
    *(tagInterfaceData **)(this + 0x18) = param_6;
    *(ushort **)(this + 0x30) = param_5;    // display name
    this[0x40] = ... | param_7;             // flags
    *(_MnkEqBuf **)(this + 0x20) = param_3; // moniker eq buffer
    *(CToken **)(this + 0x28) = param_4;    // integrity token
    if (param_4 != NULL) {
        LOCK();
        *(int *)(param_4 + 8) += 1;         // manual AddRef
        UNLOCK();
    }
}

Enter fullscreen mode Exit fullscreen mode

And CScmRotEntry adds its own fields on top:

*(ulong *)(this + 0x50) = 0x746f7263;   // magic: "crot"
*(ulong *)(this + 0x54) = param_5;
*(_FILETIME *)(this + 0x58) = *param_4; // last change time
*(tagInterfaceData **)(this + 0x60) = param_12;
*(ulong *)(this + 0x68) = param_7;      // ROTFlags

Enter fullscreen mode Exit fullscreen mode

The complete structure map is being combining both constructors gives us the full layout:

struct CScmRotMgotEntryBase {
    void            *vtable;       // +0x00
    longlong         pNext;        // +0x08  next in bucket chain
    DWORD            dwRefCount;   // +0x10  starts at 1
    tagInterfaceData *pIFaceData;  // +0x18  marshaled interface
    _MnkEqBuf       *pMnkEqBuf;   // +0x20  moniker comparison buffer
    CToken          *pToken;       // +0x28  integrity level token
    ushort          *pwszName;     // +0x30  display name (wide string)
    longlong         unk_0x38;     // +0x38
    BYTE             bFlags;       // +0x40  bit 0 = various flags
    DWORD            dwField_44;   // +0x44
    DWORD            dwField_48;   // +0x48
    DWORD            dwField_4c;   // +0x4c
};
// Derived class
struct CScmRotEntry : CScmRotMgotEntryBase {
    DWORD            dwMagic;      // +0x50  0x746F7263 = "crot"
    DWORD            dwField_54;   // +0x54
    _FILETIME        ftLastChange; // +0x58  registration timestamp
    tagInterfaceData *pIFaceData2; // +0x60  second interface data
    DWORD            dwROTFlags;   // +0x68  bit 1 = AllowAnyClient
};

Enter fullscreen mode Exit fullscreen mode

Three findings worth highlighting
The "crot" magic signature at +0x50. Every valid CScmRotEntry in memory carries the ASCII bytes c, r, o, t at this offset. This is almost certainly what CScmRotEntry::IsValid checks. For a memory scanner, this is a gift: you can locate entries by signature without knowing the head of the hash table.
CToken uses manual reference counting. The increment at param_4 + 8 is wrapped in LOCK/UNLOCK - interlocked, not COM AddRef. This means tokens are shared across multiple ROT entries from the same process using a private refcount mechanism entirely separate from COM lifetime management.
pwszName at +0x30 is the display name. This is the wide string you'd normally get from IMoniker::GetDisplayName. Reading it directly means a memory scanner doesn't need to deserialize tagInterfaceData or call any COM API to get human-readable moniker names — the string is right there in the entry.

The security picture
Walking CScmRot::GetObject surfaces three distinct access control mechanisms operating in sequence:

cRotEntryIsTrusted(plVar11, token)
RotMgotEntryIsAccessible(plVar11, token, ...)
CToken::IsTokenILLower(plVar11[5], plVar13[5], ...)

Enter fullscreen mode Exit fullscreen mode

Trust check, accessibility check, integrity level comparison in that order. When multiple entries match the same moniker, the one with the higher integrity level wins. A sandboxed process cannot shadow a medium-integrity registration. AppContainer entries get discarded entirely for out-of-container clients — we even found the log string in the code:

"Disregarding ROT entry registered by AppContainer"

The ROT has a real access control model. It's just not documented anywhere near as clearly as the public API suggests.

Code and tooling: github.com/ssteelfactor-oss

[to be continued...]