Skip to content
New

DeepSeek V4.1 Flash free for 7 days: 10M tokens a day after a top-up of $10 or more. Until Nov 30.

How it works
aeondave/linux-internals-devLinux, cloud and containers

linux-internals-dev

⁨Auth/lab dev: Linux internals; ELF loading and program startup, procfs, namespaces/caps, eBPF verifier/maps, LSM hooks for tooling/telemetry design.⁩

v2026.11.13-21cd5683 installs+3 in 7 days39 StarsMITsynced 10 hours ago

Install

No NotHoly CLI yet? Install it first.

$ curl -fsSL https://notholy.com/install.sh | sh
$ notholy skills install aeondave/linux-internals-dev

Writes to ⁦~/.notholy/skills⁩ after checking the signature and the SHA-256.

SignatureSigned, ed25519
VanNNE3qUC…zD4JjKCw==The CLI checks the SHA-256 and the signature before it writes a file.
Package41fefbd6…3f4f7f25
41fefbd6ef79…a5443f4f7f258 files · ⁦22.6 KB⁩
ScanMedium3 findings
  • injection.no_confirmationreferences/syscall-dispatch-linux.md:226
  • Executable script shipped with the skillscripts/vdso_gadget_extract.py
  • Links to outside domains
See the full scan
Sourcegithub.com/AeonDave/malskill/tree/main/offensive-coding/linux-internals-dev@21cd568

Deep Linux internals for practitioners who need to reason below libc and framework abstractions.

This skill is for structural mechanics: what the kernel and loader actually do, what invariants hold, what breaks across versions, and where detection surfaces appear.


When to activate

  • You are writing or reviewing custom ELF loaders, in-memory execution, or reflective mapping logic
  • You are diagnosing execution before main or checking an old startup diagram against the target libc and toolchain
  • You are parsing /proc for process, memory, mount, or namespace intelligence and need race-safe interpretation
  • You are building namespace or container boundary logic with clone, unshare, setns, or pidfd flows
  • You are loading eBPF programs and debugging verifier failures, helper constraints, map lifetime, or attach semantics
  • You are reasoning about LSM policy behavior, hook order, and process security attribute surfaces
  • You need to explain detection and hardening implications of low-level Linux behavior

If the task is ordinary userland API usage, this skill is overkill. If the task is ABI and kernel boundary mechanics, use this skill first.


Territory map

Domain File Covers
ELF and dynamic linking references/elf-format.md Load when parsing ELF or tracing interpreter entry, executable entry, libc startup, initializers, and termination
procfs semantics references/procfs.md /proc layout, ptrace-gated visibility, maps and smaps races, hidepid, mountinfo and pidns interactions
Namespaces and isolation references/namespaces.md namespace APIs, ownership and capability checks, pid and user namespace rules, lifetime pinning
Syscall dispatch internals references/syscall-dispatch-linux.md x86-64/aarch64 syscall ABI, vDSO gadget extraction (mseal/getrandom changes), dispatch modes (indirect/direct/seccomp-notify/memfd), hooking landscape (LD_PRELOAD/seccomp/eBPF/io_uring with per-version LSM timeline), eBPF rootkit variants (LinkPro/VoidLink/SPiCa/BPFDoor), .eh_frame stack spoofing (open field), kernel hardening per-version table, Landlock progression
eBPF architecture references/ebpf.md bpf() command families, map and program lifecycle, verifier model, helper constraints, attach and pin patterns
LSM model references/lsm.md LSM stacking and ordering, active module discovery, process attr interfaces, hook surfaces and development implications

Quick invariants

Area Invariant
ELF Runtime loading is driven by program headers, not section headers
ELF PT_LOAD: p_filesz <= p_memsz; trailing bytes are zero-filled in memory
procfs Access to many /proc/pid/* files for other processes is ptrace capability gated
procfs Reading maps and smaps is inherently racy except single-read snapshots
namespaces Joining namespaces requires capability checks in both current and owning user namespaces
pid namespaces A process cannot move to an ancestor PID namespace
syscalls (x86-64) Arg4 is in r10, not rcx; syscall clobbers rcx and r11
eBPF Verifier must prove termination, type safety, bounds safety, and reference release on all paths
eBPF Helper calls can invalidate packet pointer proofs; bounds checks often must be redone
LSM Active LSM order matters; checks run in configured order, capability module first
LSM /sys/kernel/security/lsm reflects the active stack and check order

Decision flow: choose the right subsystem lens

  1. Issue is binary load or symbol resolution behavior

    • Start in elf-format.md
    • Validate PT_INTERP, DT_NEEDED, DT_RPATH and DT_RUNPATH, then relocation type flow
  2. Issue is process or memory inspection mismatch

    • Start in procfs.md
    • Validate ptrace permission model, hidepid mode, and map-read race conditions
  3. Issue is container boundary or namespace transition

    • Start in namespaces.md
    • Validate ownership user namespace, capability context, and one-way PID namespace movement
  4. Issue is syscall origin, anti-hooking, or low-level ABI behavior

  • Start in syscall-dispatch-linux.md
  • Validate calling convention (r10/rcx), dispatch source (vDSO vs libc vs direct), and fallback implications
  1. Issue is eBPF load rejection or runtime attach surprises

    • Start in ebpf.md
    • Read verifier error class first, then map/program refs and attach target constraints
  2. Issue is policy enforcement or security hook behavior

    • Start in lsm.md
    • Confirm active stack and hook order before attributing allow or deny outcomes

Detection and hardening surface checklist

Use this whenever implementing a low-level technique.

  • ELF path
    • Does behavior depend on dynamic linker search path state such as RUNPATH, cache, preload, or secure-exec stripping?
  • procfs path
    • Are you assuming stable snapshots from files documented as racy?
    • Are you accidentally leaking sensitive args or env due to permissive proc mount options?
  • namespace path
    • Are capability checks evaluated in the expected owning user namespace?
    • Could namespace lifetime be pinned by open descriptors or bind mounts?
  • syscall path
    • Is the syscall instruction sourced from the location you think it is (vDSO/libc/direct asm), and does that match your telemetry assumptions?
    • Are x86-64 register conventions respected for 4-6 argument syscalls?
  • eBPF path
    • Are helper or map choices constrained by program type and privilege model?
    • Are verifier assumptions broken by helper-induced packet buffer changes?
  • LSM path
    • Is behavior due to stacked module ordering rather than a single policy engine?
    • Are process attrs read from module-specific paths versus legacy aggregate paths?

Pitfalls

  • Treating section headers as mandatory at runtime for execution behavior
  • Assuming DT_RPATH and DT_RUNPATH are equivalent in dependency traversal
  • Assuming /proc/pid/maps partial reads are stable under concurrent VMA changes
  • Forgetting hidepid and ptrace gating when process inventory appears incomplete
  • Expecting setns to move the caller itself into a new PID namespace instead of affecting future children
  • Trying to reenter same user namespace to regain dropped capabilities
  • Passing syscall arg4 in rcx instead of r10 on x86-64 and then debugging phantom failures
  • Assuming all direct-syscall approaches have the same observable syscall origin in kernel telemetry
  • Ignoring eBPF reference-release paths after socket lookup helpers
  • Debugging eBPF without verifier logs, then guessing blindly
  • Attributing policy outcomes to one LSM without checking active stack order

Resources

  • references/elf-format.md for loader/relocation mechanics and a reproducible pre-main startup trace
  • references/procfs.md for process and memory introspection semantics
  • references/namespaces.md for isolation model and transition constraints
  • references/syscall-dispatch-linux.md for syscall ABI, dispatch-source tradeoffs, hooking landscape (LD_PRELOAD / seccomp / eBPF / io_uring), vDSO gadget extraction, .eh_frame comparison with Windows .pdata, and fallback design
  • references/ebpf.md for verifier and syscall-level eBPF operations
  • references/lsm.md for LSM architecture and policy interaction model
  • scripts/vdso_gadget_extract.py for enumerating syscall gadgets from vDSO and libc on a running Linux system (Linux only)

Files in this version

Permissions

PermissionDeclaredFound by the scan
Reads filesYesYes
Writes filesYesYes
Runs shell commandsYesYes
Uses the networkYesYes
Uses connectorsNoNo
Drives a browserNoNo

Scan findings

Reviewed on Nov 13, 2026

  1. MediumPrompt injection · injection.no_confirmationreferences/syscall-dispatch-linux.md:226
  2. MediumFile · Executable script shipped with the skillscripts/vdso_gadget_extract.py
  3. InfoLink · Links to outside domains

A clean scan lowers the risk but does not prove a package is safe. Read what you install.

VersionPublishedCommitSHA-256SignaturePermission change
v2026.11.13-21cd568Nov 13, 202621cd56841fefbd6Signed–

Ask anything your scope allows.

Create an account, add credits and ask the first question in the chat or the terminal.

  1. 1/3

    Create the account

    Individual or company. You can switch to a company later without losing balance or keys.

    IndividualCompany
  2. 2/3

    Add credits

    By card. Pay per token, no subscription.

    $5.00minimum deposit
  3. 3/3

    Ask

    In the chat, or install the CLI and run notholy login in your terminal.

    $ notholy login