linux-ker...@vger.kernel.org, linux-arm-ker...@lists.infradead.org, linux-ri...@lists.infradead.org, linuxppc-dev@lists.ozlabs.org, Mark Rutland <mark.rutl...@arm.com>, Greg Kroah-Hartman <gre...@linuxfoundation.org>, Christoph Hellwig <h...@infradead.org>, Christophe Leroy <christophe.le...@csgroup.eu>, David Hildenbrand <da...@redhat.com>, Conor Dooley <conor.doo...@microchip.com> From: Maxwell Bland <mbl...@motorola.com> Date: Mon, 15 Apr 2024 15:16:08 -0500 Subject: [PATCH 0/5] mm: code and data partitioning improvements
Managing allocations to ensure code and data pages are not interleaved is not possible prior to this patch, as ASLR requires programming a dynamic _text offset while the vmalloc infrastructure maintains static VMALLOC_START and VMALLOC_END constants. In systems where code and data are interleaved at a PTE granularity, kernel improvements targeting the prevention of exploit stages which modify page tables are inefficient and less effective as individual PTE updates occur at high frequency and cannot be coarsely grouped at the PMD level or greater. This patch adds minimal arch-specific callbacks to the initialization of vmalloc and when deciding whether to use a specific virtual memory area to satisfy a vmalloc request to provide the capability to prevent the allocation of specific virtual addresses under specific system states. By default these hooks are unimplemented. To further support the practical use of these callbacks, this patch also adds a virtual address parameter to pmd_populate_kernel, so that this interface matches the equivalent pte-level interface and architectures are not required to perform a reverse page table lookup to determine the vaddr being allocated during pmd creation. To demonstrate the impact and value of these changes, this patch implements support for dynamic PXNTable under aarch64 in 71 lines of code (a single "if" check during memory allocation), by checking the virtual address of a given vmalloc call to determine whether it is code or data. From experience in trying to implement kernel page table immutability and protections in KVM to prevent recent CVEs, e.g. CVE-2024-1086, this is a necessary first step. To better help maintainers and future developers, this patch expands ptdump.c so that non-leaf page table descriptors can be more easily noted in debug output by setting a note_non_leaf bool in the ptdump state. Signed-off-by: Maxwell Bland <mbl...@motorola.com> --- First, thank you to a number of maintainers (Mark Rutland, Greg KH, Christoph Hellwig, Christophe Leroy, David Hildenbrand, Conor Dooley) for their feedback on <20240220203256.31153-1-mbl...@motorola.com> and <CAP5Mv+ydhk=Ob4b40ZahGMgT-5+-VEHxtmA=-lkjieoou+k...@mail.gmail.com> This patch is a further refinement and overhaul of these prior two attempts. Also, apologies for the roughly two months delay between patch submissions! I had Motorola work to do. In support of testing this patch (but not included in this patch), I set note_non_leaf to true under arch/arm64/mm/ptdump.c and added PMD_TABLE_PXN to pte_bits to print out whether the PXNTable bit was set. The txt files under the following directory can be diff'ed to see the result: github.com/maxwell-bland/linux-patch-data/tree/main/code_data_parting/ptdump I also created a script to fetch and cross-compile the kernel for each of the 21 subarchitectures which required fixes to provide a virtual address to pmd_populate_kernel. I have no idea if it is useful and maybe one already exists, but it worked well for me over some alternatives (xcross, buildroot): github.com/maxwell-bland/x-linux As with the last patchset, I also measured performance using Torvald's test-tlb program on an aarch64 QEMU instance, with results here: github.com/maxwell-bland/linux-patch-data/tree/main/code_data_parting/tlbperf As all changes to other arches are effectively no-ops, performance impacts in those domains are negligible. Maxwell Bland (5): mm: allow arch refinement/skip for vmap alloc arm64: mm: code and data partitioning for aslr mm: add vaddr param to pmd_populate_kernel arm64: dynamic enforcement of PXNTable ptdump: add state parameter for non-leaf callback arch/alpha/include/asm/pgalloc.h | 5 +- arch/arc/include/asm/pgalloc.h | 3 +- arch/arc/mm/highmem.c | 2 +- arch/arm/include/asm/kfence.h | 2 +- arch/arm/include/asm/pgalloc.h | 3 +- arch/arm/mm/kasan_init.c | 2 +- arch/arm/mm/mmu.c | 2 +- arch/arm64/include/asm/module.h | 12 ++++ arch/arm64/include/asm/pgalloc.h | 15 ++++- arch/arm64/include/asm/vmalloc.h | 17 ++++- arch/arm64/kernel/Makefile | 2 +- arch/arm64/kernel/module.c | 7 +- arch/arm64/kernel/probes/kprobes.c | 7 +- arch/arm64/kernel/setup.c | 4 ++ arch/arm64/kernel/vmalloc.c | 71 ++++++++++++++++++++ arch/arm64/mm/ptdump.c | 10 +-- arch/arm64/mm/trans_pgd.c | 2 +- arch/arm64/net/bpf_jit_comp.c | 8 ++- arch/csky/include/asm/pgalloc.h | 2 +- arch/hexagon/include/asm/pgalloc.h | 2 +- arch/loongarch/include/asm/pgalloc.h | 3 +- arch/loongarch/mm/init.c | 2 +- arch/loongarch/mm/kasan_init.c | 2 +- arch/m68k/include/asm/mcf_pgalloc.h | 2 +- arch/m68k/include/asm/motorola_pgalloc.h | 3 +- arch/m68k/include/asm/sun3_pgalloc.h | 3 +- arch/microblaze/include/asm/pgalloc.h | 2 +- arch/mips/include/asm/pgalloc.h | 2 +- arch/mips/kvm/mmu.c | 2 +- arch/nios2/include/asm/pgalloc.h | 2 +- arch/openrisc/include/asm/pgalloc.h | 2 +- arch/parisc/include/asm/pgalloc.h | 5 +- arch/parisc/mm/init.c | 6 +- arch/powerpc/include/asm/book3s/32/pgalloc.h | 2 +- arch/powerpc/include/asm/book3s/64/pgalloc.h | 2 +- arch/powerpc/include/asm/nohash/32/pgalloc.h | 2 +- arch/powerpc/include/asm/nohash/64/pgalloc.h | 2 +- arch/powerpc/mm/book3s64/radix_pgtable.c | 2 +- arch/powerpc/mm/kasan/init_32.c | 4 +- arch/powerpc/mm/kasan/init_book3e_64.c | 9 ++- arch/powerpc/mm/kasan/init_book3s_64.c | 7 +- arch/powerpc/mm/nohash/book3e_pgtable.c | 2 +- arch/powerpc/mm/pgtable_32.c | 4 +- arch/powerpc/mm/ptdump/ptdump.c | 2 + arch/riscv/include/asm/pgalloc.h | 2 +- arch/riscv/kernel/hibernate.c | 2 +- arch/riscv/mm/ptdump.c | 6 +- arch/s390/include/asm/pgalloc.h | 2 +- arch/s390/mm/dump_pagetables.c | 6 +- arch/sh/include/asm/pgalloc.h | 2 +- arch/sh/mm/init.c | 2 +- arch/sparc/include/asm/pgalloc_32.h | 3 +- arch/sparc/include/asm/pgalloc_64.h | 4 +- arch/sparc/mm/init_64.c | 8 +-- arch/um/include/asm/pgalloc.h | 4 +- arch/x86/include/asm/pgalloc.h | 3 +- arch/x86/mm/dump_pagetables.c | 3 +- arch/x86/mm/init_64.c | 14 +++- arch/x86/mm/ioremap.c | 2 +- arch/x86/mm/kasan_init_64.c | 2 +- arch/xtensa/include/asm/pgalloc.h | 2 +- include/linux/mm.h | 4 +- include/linux/ptdump.h | 1 + include/linux/vmalloc.h | 24 +++++++ mm/hugetlb_vmemmap.c | 4 +- mm/kasan/init.c | 14 ++-- mm/memory.c | 4 +- mm/percpu.c | 2 +- mm/pgalloc-track.h | 3 +- mm/ptdump.c | 13 ++++ mm/sparse-vmemmap.c | 2 +- mm/vmalloc.c | 16 +++-- 72 files changed, 299 insertions(+), 107 deletions(-) create mode 100644 arch/arm64/kernel/vmalloc.c base-commit: 0bbac3facb5d6cc0171c45c9873a2dc96bea9680 -- 2.39.2
