| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Wed, 3 Apr 2024 16:01:50 +0100 From: Mark Rutland <mark.rutland@....com> To: linux-kernel@...r.kernel.org Cc: anil.s.keshavamurthy@...el.com, aou@...s.berkeley.edu, catalin.marinas@....com, davem@...emloft.net, jarkko@...nel.org, linux-arm-kernel@...ts.infradead.org, mark.rutland@....com, mhiramat@...nel.org, naveen.n.rao@...ux.ibm.com, palmer@...belt.com, paul.walmsley@...ive.com, will@...nel.org Subject: [PATCH v2 0/4] kprobes: permit use without modules Currently KPROBES depends on MODULES and cannot be built when support for modules is not enabled. This is largely an artifact of the default/generic code for allocating insn pages using module_alloc(), though several architectures do not use this and have no strict dependency on MODULES. It would be nice to allow KPROBES to be used without MODULES, as this can be useful for testing and/or in certain constrained environments. This series (based on v6.9-rc1) removes the artificial dependency on MODULES. This permits (but does not require) that architectures which don't use module_alloc() to allocate kprobe insn pages can support kprobes when module support is not enabled. The series deliberately avoids adding a common text allocator, as the requirements for allocating kprobe memory van vary by architecture, and can differ from other text allocations. However, architectures can easily call a common allocator if they wish, and this series does not preclude using common allocators immediately or in future. The key change is in patch 3. This requires that architectures which provide their own alloc function must provide the corresponding free function and select HAVE_KPROBES_ALLOC with any appropriate dependencies for their implementation. Architectures which use the generic functions are left as-is with a dependency on MODULES. The final patch allows the core kprobes code to be built without MODULES, and removes the explicit dependency from Kconfig. This is derived from Jarkko's recent v6 attempt: https://lore.kernel.org/lkml/20240326012102.27438-1-jarkko@kernel.org/ With the series applied, arm64 and riscv can enable KPROBES without MODULES, while powerpc/s390/x86 are still depend on MODULES as their alloc functions currently use module_alloc(), and all other architectures with KPROBES uses the generic implementation that depends on MODULES. I believe it should be relatively easy to enable powerpc/s390/x86 to not depend on MODULES. Since v1 [1]: * Rebase to v6.9-rc2 (trivial) * Update to Jarkko's v7 cleanups for core kprobes * Fix accidental use of KPROBES_USE_MODULE_ALLOC * Improve commit messages * Update Cc lists [1] https://lore.kernel.org/lkml/20240326163624.3253157-1-mark.rutland@arm.com/ Mark. Jarkko Sakkinen (1): kprobes: Remove core dependency on modules Mark Rutland (3): arm64: patching: always use fixmap kprobes/treewide: Add kprobes_ prefix to insn alloc/free functions kprobes/treewide: Explicitly override alloc/free functions arch/Kconfig | 5 ++- arch/arm64/Kconfig | 1 + arch/arm64/kernel/patching.c | 10 ++--- arch/arm64/kernel/probes/kprobes.c | 7 ++- arch/powerpc/Kconfig | 3 +- arch/powerpc/kernel/kprobes.c | 7 ++- arch/powerpc/kernel/optprobes.c | 4 +- arch/riscv/Kconfig | 1 + arch/riscv/kernel/probes/kprobes.c | 7 ++- arch/s390/Kconfig | 3 +- arch/s390/kernel/kprobes.c | 7 ++- arch/x86/Kconfig | 3 +- arch/x86/kernel/kprobes/core.c | 7 ++- include/linux/kprobes.h | 7 +-- kernel/kprobes.c | 68 +++++++++++++++++------------- kernel/trace/trace_kprobe.c | 15 ++++++- 16 files changed, 103 insertions(+), 52 deletions(-) -- 2.30.2
Powered by blists - more mailing lists