Kernel arguments

Table of Contents

• Introduction
• Standard
• Additional
  • Disable 32-bit processes and syscalls
  • Force disable simultaneous multithreading
  • Unstable kargs

Introduction

On secureblue and other systems that use rpm-ostree, kernel arguments (kargs) can be managed using rpm-ostree kargs. Run rpm-ostree kargs --help for usage information.

Secureblue sets some kernel arguments by default, and provides a script ujust set-kargs-hardening that provides additional sets of kernel arguments. To remove all kernel arguments that secureblue adds, you can run ujust remove-kargs-hardening.

For details on what each kernel argument does, see the kernel documentation.

Standard

Stable kernel arguments that are set by default on a fresh secureblue installation, and are always applied by the script ujust set-kargs-hardening.

• hash_pointers=always: Hash kernel pointers even if slab_debug is enabled.
• init_on_alloc=1: Fill newly allocated pages and heap objects with zeroes, mitigating use-after-free vulnerabilities.
• init_on_free=1: Fill freed pages and heap objects with zeroes, mitigating use-after-free vulnerabilities.
• iommu=force and intel_iommu=on: Mitigate DMA attacks by enabling IOMMU.
• iommu.passthrough=0: Disable IOMMU bypass.
• iommu.strict=1: Synchronously invalidate IOMMU hardware TLBs.
• kvm_amd.sev=1, kvm_amd.sev_es=1, kvm_amd.sev_snp=1: Enable AMD Secure Encrypted Virtualization (SEV) and extensions.
• kvm-intel.vmentry_l1d_flush=always: Enable unconditional flushes, required for complete L1D vulnerability mitigation.
• kvm.mitigate_smt_rsb=1: Mitigate cross-thread return address predictions vulnerability for certain KVM hypervisor configurations.
• l1d_flush=on: Enable the mechanism to flush the L1D cache on context switch.
• l1tf=full,force: Force enable all available mitigations for the L1TF vulnerability.
• lockdown=confidentiality: Enable kernel lockdown in the strictest mode.
• loglevel=0: Only log level 0 (system is unusable) messages to the console.
• mitigations=auto,nosmt: Automatically mitigate all known CPU vulnerabilities, including disabling SMT if necessary.
• module.sig_enforce=1: Only allow kernel modules that have been signed with a valid key to be loaded.
• page_alloc.shuffle=1: Enable page allocator freelist randomization, reducing page allocation predictability.
• proc_mem.force_override=ptrace: Only allow memory permissions for /proc/<pid>/mem to be overridden by active ptracers.
• pti=on: Enable kernel page table isolation.
• random.trust_bootloader=off: Disable trusting the use of the seed passed by the bootloader.
• random.trust_cpu=off: Disable CPU-based entropy sources, as it’s not auditable and has resulted in vulnerabilities.
• randomize_kstack_offset=on: Randomize kernel stack offset on each syscall, making certain types of attacks more difficult.
• rd.shell=0, rd.emergency=halt: Mitigate initramfs malware injection attack.
• slab_debug=FZ: Enable sanity checks and red zoning for the kernel slab allocator.
• slab_nomerge: Disable the merging of slabs, increasing difficulty of heap exploitation.
• spec_store_bypass_disable=on: Disable spec store bypass for all programs.
• spectre_v2=on: Turn on spectre_v2 mitigations at boot time for all programs.
• ssbd=force-on: Enable mitigation for Speculative Store Bypass vulnerability for both kernel and userspace on vulnerable CPUs.
• systemd.ssh_auto=no: Disable automatic creation of socket-activated SSH server by systemd (see systemd-ssh-generator(8) for details), which can lead to security vulnerabilities.
• vdso32=0: Disable 32-bit vDSO.
• vsyscall=none: Disable vsyscall as it is both obsolete and enables an ROP attack vector.

Additional

Sets of additional kargs that can be selectively set alongside the standard kargs detailed above. The set-kargs-hardening command prompts the user on whether to add apply of the 3 sets of kargs detailed below:

Disable 32-bit processes and syscalls

• ia32_emulation=0: Disables 32-bit processes and syscalls.

Force disable simultaneous multithreading

• nosmt=force: Disables this hardware feature on user request, regardless of whether it is affected by known vulnerabilities. Note that this halves the number of CPU cores.

Unstable kargs

• amd_iommu=force_isolation: Mitigate DMA attacks on AMD hardware by having the AMD IOMMU driver force device isolation for all devices.
• bdev_allow_write_mounted=0: Disallow direct writes to filesystem block devices.
• debugfs=off: Disable debugfs to prevent exposure of sensitive kernel information.
• efi=disable_early_pci_dma: Fill IOMMU protection gap by setting the BUSMASTER bit during early boot.
• gather_data_sampling=force: Mitigate unprivileged speculative access to data by using the microcode mitigation when available, or by disabling AVX on affected systems where the microcode hasn’t been updated to include the mitigation.
• mem_encrypt=on: Enable AMD Secure Memory Encryption (SME).
• oops=panic: Treat oopses as fatal, preventing attacks that depend on repeated oopses.

Kernel arguments

Table of Contents

• Introduction
• Standard
• Additional
  • Disable 32-bit processes and syscalls
  • Force disable simultaneous multithreading
  • Unstable kargs

Introduction

On secureblue and other systems that use rpm-ostree, kernel arguments (kargs) can be managed using rpm-ostree kargs. Run rpm-ostree kargs --help for usage information.

Secureblue sets some kernel arguments by default, and provides a script ujust set-kargs-hardening that provides additional sets of kernel arguments. To remove all kernel arguments that secureblue adds, you can run ujust remove-kargs-hardening.

For details on what each kernel argument does, see the kernel documentation.

Standard

Stable kernel arguments that are set by default on a fresh secureblue installation, and are always applied by the script ujust set-kargs-hardening.

• hash_pointers=always: Hash kernel pointers even if slab_debug is enabled.
• init_on_alloc=1: Fill newly allocated pages and heap objects with zeroes, mitigating use-after-free vulnerabilities.
• init_on_free=1: Fill freed pages and heap objects with zeroes, mitigating use-after-free vulnerabilities.
• iommu=force and intel_iommu=on: Mitigate DMA attacks by enabling IOMMU.
• iommu.passthrough=0: Disable IOMMU bypass.
• iommu.strict=1: Synchronously invalidate IOMMU hardware TLBs.
• kvm_amd.sev=1, kvm_amd.sev_es=1, kvm_amd.sev_snp=1: Enable AMD Secure Encrypted Virtualization (SEV) and extensions.
• kvm-intel.vmentry_l1d_flush=always: Enable unconditional flushes, required for complete L1D vulnerability mitigation.
• kvm.mitigate_smt_rsb=1: Mitigate cross-thread return address predictions vulnerability for certain KVM hypervisor configurations.
• l1d_flush=on: Enable the mechanism to flush the L1D cache on context switch.
• l1tf=full,force: Force enable all available mitigations for the L1TF vulnerability.
• lockdown=confidentiality: Enable kernel lockdown in the strictest mode.
• loglevel=0: Only log level 0 (system is unusable) messages to the console.
• mitigations=auto,nosmt: Automatically mitigate all known CPU vulnerabilities, including disabling SMT if necessary.
• module.sig_enforce=1: Only allow kernel modules that have been signed with a valid key to be loaded.
• page_alloc.shuffle=1: Enable page allocator freelist randomization, reducing page allocation predictability.
• proc_mem.force_override=ptrace: Only allow memory permissions for /proc/<pid>/mem to be overridden by active ptracers.
• pti=on: Enable kernel page table isolation.
• random.trust_bootloader=off: Disable trusting the use of the seed passed by the bootloader.
• random.trust_cpu=off: Disable CPU-based entropy sources, as it’s not auditable and has resulted in vulnerabilities.
• randomize_kstack_offset=on: Randomize kernel stack offset on each syscall, making certain types of attacks more difficult.
• rd.shell=0, rd.emergency=halt: Mitigate initramfs malware injection attack.
• slab_debug=FZ: Enable sanity checks and red zoning for the kernel slab allocator.
• slab_nomerge: Disable the merging of slabs, increasing difficulty of heap exploitation.
• spec_store_bypass_disable=on: Disable spec store bypass for all programs.
• spectre_v2=on: Turn on spectre_v2 mitigations at boot time for all programs.
• ssbd=force-on: Enable mitigation for Speculative Store Bypass vulnerability for both kernel and userspace on vulnerable CPUs.
• systemd.ssh_auto=no: Disable automatic creation of socket-activated SSH server by systemd (see systemd-ssh-generator(8) for details), which can lead to security vulnerabilities.
• vdso32=0: Disable 32-bit vDSO.
• vsyscall=none: Disable vsyscall as it is both obsolete and enables an ROP attack vector.

Additional

Sets of additional kargs that can be selectively set alongside the standard kargs detailed above. The set-kargs-hardening command prompts the user on whether to add apply of the 3 sets of kargs detailed below:

Disable 32-bit processes and syscalls

• ia32_emulation=0: Disables 32-bit processes and syscalls.

Force disable simultaneous multithreading

• nosmt=force: Disables this hardware feature on user request, regardless of whether it is affected by known vulnerabilities. Note that this halves the number of CPU cores.

Unstable kargs

• amd_iommu=force_isolation: Mitigate DMA attacks on AMD hardware by having the AMD IOMMU driver force device isolation for all devices.
• bdev_allow_write_mounted=0: Disallow direct writes to filesystem block devices.
• debugfs=off: Disable debugfs to prevent exposure of sensitive kernel information.
• efi=disable_early_pci_dma: Fill IOMMU protection gap by setting the BUSMASTER bit during early boot.
• gather_data_sampling=force: Mitigate unprivileged speculative access to data by using the microcode mitigation when available, or by disabling AVX on affected systems where the microcode hasn’t been updated to include the mitigation.
• mem_encrypt=on: Enable AMD Secure Memory Encryption (SME).
• oops=panic: Treat oopses as fatal, preventing attacks that depend on repeated oopses.

Kernel arguments

Table of Contents

• Introduction
• Standard
• Additional
  • Disable 32-bit processes and syscalls
  • Force disable simultaneous multithreading
  • Unstable kargs

Introduction

On secureblue and other systems that use rpm-ostree, kernel arguments (kargs) can be managed using rpm-ostree kargs. Run rpm-ostree kargs --help for usage information.

Secureblue sets some kernel arguments by default, and provides a script ujust set-kargs-hardening that provides additional sets of kernel arguments. To remove all kernel arguments that secureblue adds, you can run ujust remove-kargs-hardening.

For details on what each kernel argument does, see the kernel documentation.

Standard

Stable kernel arguments that are set by default on a fresh secureblue installation, and are always applied by the script ujust set-kargs-hardening.

• hash_pointers=always: Hash kernel pointers even if slab_debug is enabled.
• init_on_alloc=1: Fill newly allocated pages and heap objects with zeroes, mitigating use-after-free vulnerabilities.
• init_on_free=1: Fill freed pages and heap objects with zeroes, mitigating use-after-free vulnerabilities.
• iommu=force and intel_iommu=on: Mitigate DMA attacks by enabling IOMMU.
• iommu.passthrough=0: Disable IOMMU bypass.
• iommu.strict=1: Synchronously invalidate IOMMU hardware TLBs.
• kvm_amd.sev=1, kvm_amd.sev_es=1, kvm_amd.sev_snp=1: Enable AMD Secure Encrypted Virtualization (SEV) and extensions.
• kvm-intel.vmentry_l1d_flush=always: Enable unconditional flushes, required for complete L1D vulnerability mitigation.
• kvm.mitigate_smt_rsb=1: Mitigate cross-thread return address predictions vulnerability for certain KVM hypervisor configurations.
• l1d_flush=on: Enable the mechanism to flush the L1D cache on context switch.
• l1tf=full,force: Force enable all available mitigations for the L1TF vulnerability.
• lockdown=confidentiality: Enable kernel lockdown in the strictest mode.
• loglevel=0: Only log level 0 (system is unusable) messages to the console.
• mitigations=auto,nosmt: Automatically mitigate all known CPU vulnerabilities, including disabling SMT if necessary.
• module.sig_enforce=1: Only allow kernel modules that have been signed with a valid key to be loaded.
• page_alloc.shuffle=1: Enable page allocator freelist randomization, reducing page allocation predictability.
• proc_mem.force_override=ptrace: Only allow memory permissions for /proc/<pid>/mem to be overridden by active ptracers.
• pti=on: Enable kernel page table isolation.
• random.trust_bootloader=off: Disable trusting the use of the seed passed by the bootloader.
• random.trust_cpu=off: Disable CPU-based entropy sources, as it’s not auditable and has resulted in vulnerabilities.
• randomize_kstack_offset=on: Randomize kernel stack offset on each syscall, making certain types of attacks more difficult.
• rd.shell=0, rd.emergency=halt: Mitigate initramfs malware injection attack.
• slab_debug=FZ: Enable sanity checks and red zoning for the kernel slab allocator.
• slab_nomerge: Disable the merging of slabs, increasing difficulty of heap exploitation.
• spec_store_bypass_disable=on: Disable spec store bypass for all programs.
• spectre_v2=on: Turn on spectre_v2 mitigations at boot time for all programs.
• ssbd=force-on: Enable mitigation for Speculative Store Bypass vulnerability for both kernel and userspace on vulnerable CPUs.
• systemd.ssh_auto=no: Disable automatic creation of socket-activated SSH server by systemd (see systemd-ssh-generator(8) for details), which can lead to security vulnerabilities.
• vdso32=0: Disable 32-bit vDSO.
• vsyscall=none: Disable vsyscall as it is both obsolete and enables an ROP attack vector.

Additional

Sets of additional kargs that can be selectively set alongside the standard kargs detailed above. The set-kargs-hardening command prompts the user on whether to add apply of the 3 sets of kargs detailed below:

Disable 32-bit processes and syscalls

• ia32_emulation=0: Disables 32-bit processes and syscalls.

Force disable simultaneous multithreading

• nosmt=force: Disables this hardware feature on user request, regardless of whether it is affected by known vulnerabilities. Note that this halves the number of CPU cores.

Unstable kargs

• amd_iommu=force_isolation: Mitigate DMA attacks on AMD hardware by having the AMD IOMMU driver force device isolation for all devices.
• bdev_allow_write_mounted=0: Disallow direct writes to filesystem block devices.
• debugfs=off: Disable debugfs to prevent exposure of sensitive kernel information.
• efi=disable_early_pci_dma: Fill IOMMU protection gap by setting the BUSMASTER bit during early boot.
• gather_data_sampling=force: Mitigate unprivileged speculative access to data by using the microcode mitigation when available, or by disabling AVX on affected systems where the microcode hasn’t been updated to include the mitigation.
• mem_encrypt=on: Enable AMD Secure Memory Encryption (SME).
• oops=panic: Treat oopses as fatal, preventing attacks that depend on repeated oopses.