Userland library functions such as allocators and threading implementations often require regions of memory to act as 'guard pages' - mappings which, when accessed, result in a fatal signal being sent to the accessing process.
The current means by which these are implemented is via a PROT_NONE mmap() mapping, which provides the required semantics however incur an overhead of a VMA for each such region.
With a great many processes and threads, this can rapidly add up and incur a significant memory penalty. It also has the added problem of preventing merges that might otherwise be permitted.
This series takes a different approach - an idea suggested by Vlasimil Babka (and before him David Hildenbrand and Jann Horn - perhaps more - the provenance becomes a little tricky to ascertain after this - please forgive any omissions!) - rather than locating the guard pages at the VMA layer, instead placing them in page tables mapping the required ranges.
Early testing of the prototype version of this code suggests a 5 times speed up in memory mapping invocations (in conjunction with use of process_madvise()) and a 13% reduction in VMAs on an entirely idle android system and unoptimised code.
We expect with optimisation and a loaded system with a larger number of guard pages this could significantly increase, but in any case these numbers are encouraging.
This way, rather than having separate VMAs specifying which parts of a range are guard pages, instead we have a VMA spanning the entire range of memory a user is permitted to access and including ranges which are to be 'guarded'.
After mapping this, a user can specify which parts of the range should result in a fatal signal when accessed.
By restricting the ability to specify guard pages to memory mapped by existing VMAs, we can rely on the mappings being torn down when the mappings are ultimately unmapped and everything works simply as if the memory were not faulted in, from the point of view of the containing VMAs.
This mechanism in effect poisons memory ranges similar to hardware memory poisoning, only it is an entirely software-controlled form of poisoning.
Any poisoned region of memory is also able to 'unpoisoned', that is, to have its poison markers removed.
The mechanism is implemented via madvise() behaviour - MADV_GUARD_POISON which simply poisons ranges - and MADV_GUARD_UNPOISON - which clears this poisoning.
Poisoning can be performed across multiple VMAs and any existing mappings will be cleared, that is zapped, before installing the poisoned page table mappings.
There is no concept of 'nested' poisoning, multiple attempts to poison a range will, after the first poisoning, have no effect.
Importantly, unpoisoning of poisoned ranges has no effect on non-poisoned memory, so a user can safely unpoison a range of memory and clear only poison page table mappings leaving the rest intact.
The actual mechanism by which the page table entries are specified makes use of existing logic - PTE markers, which are used for the userfaultfd UFFDIO_POISON mechanism.
Unfortunately PTE_MARKER_POISONED is not suited for the guard page mechanism as it results in VM_FAULT_HWPOISON semantics in the fault handler, so we add our own specific PTE_MARKER_GUARD and adapt existing logic to handle it.
We also extend the generic page walk mechanism to allow for installation of PTEs (carefully restricted to memory management logic only to prevent unwanted abuse).
We ensure that zapping performed by, for instance, MADV_DONTNEED, does not remove guard poison markers, nor does forking (except when VM_WIPEONFORK is specified for a VMA which implies a total removal of memory characteristics).
It's important to note that the guard page implementation is emphatically NOT a security feature, so a user can remove the poisoning if they wish. We simply implement it in such a way as to provide the least surprising behaviour.
An extensive set of self-tests are provided which ensure behaviour is as expected and additionally self-documents expected behaviour of poisoned ranges.
Suggested-by: Vlastimil Babka vbabka@suse.cz Suggested-by: Jann Horn jannh@google.com Suggested-by: David Hildenbrand david@redhat.com
v2 * The macros in kselftest_harness.h seem to be broken - __EXPECT() is terminated by '} while (0); OPTIONAL_HANDLER(_assert)' meaning it is not safe in single line if / else or for /which blocks, however working around this results in checkpatch producing invalid warnings, as reported by Shuah. * Fixing these macros is out of scope for this series, so compromise and instead rewrite test blocks so as to use multiple lines by separating out a decl in most cases. This has the side effect of, for the most part, making things more readable. * Heavily document the use of the volatile keyword - we can't avoid checkpatch complaining about this, so we explain it, as reported by Shuah. * Updated commit message to highlight that we skip tests we lack permissions for, as reported by Shuah. * Replaced a perror() with ksft_exit_fail_perror(), as reported by Shuah. * Added user friendly messages to cases where tests are skipped due to lack of permissions, as reported by Shuah. * Update the tool header to include the new MADV_GUARD_POISON/UNPOISON defines and directly include asm-generic/mman.h to get the platform-neutral versions to ensure we import them. * Finally fixed Vlastimil's email address in Suggested-by tags from suze to suse, as reported by Vlastimil. * Added linux-api to cc list, as reported by Vlastimil.
v1 * Un-RFC'd as appears no major objections to approach but rather debate on implementation. * Fixed issue with arches which need mmu_context.h and tlbfush.h. header imports in pagewalker logic to be able to use update_mmu_cache() as reported by the kernel test bot. * Added comments in page walker logic to clarify who can use ops->install_pte and why as well as adding a check_ops_valid() helper function, as suggested by Christoph. * Pass false in full parameter in pte_clear_not_present_full() as suggested by Jann. * Stopped erroneously requiring a write lock for the poison operation as suggested by Jann and Suren. * Moved anon_vma_prepare() to the start of madvise_guard_poison() to be consistent with how this is used elsewhere in the kernel as suggested by Jann. * Avoid returning -EAGAIN if we are raced on page faults, just keep looping and duck out if a fatal signal is pending or a conditional reschedule is needed, as suggested by Jann. * Avoid needlessly splitting huge PUDs and PMDs by specifying ACTION_CONTINUE, as suggested by Jann. https://lore.kernel.org/all/cover.1729196871.git.lorenzo.stoakes@oracle.com/
RFC https://lore.kernel.org/all/cover.1727440966.git.lorenzo.stoakes@oracle.com/
Lorenzo Stoakes (5): mm: pagewalk: add the ability to install PTEs mm: add PTE_MARKER_GUARD PTE marker mm: madvise: implement lightweight guard page mechanism tools: testing: update tools UAPI header for mman-common.h selftests/mm: add self tests for guard page feature
arch/alpha/include/uapi/asm/mman.h | 3 + arch/mips/include/uapi/asm/mman.h | 3 + arch/parisc/include/uapi/asm/mman.h | 3 + arch/xtensa/include/uapi/asm/mman.h | 3 + include/linux/mm_inline.h | 2 +- include/linux/pagewalk.h | 18 +- include/linux/swapops.h | 26 +- include/uapi/asm-generic/mman-common.h | 3 + mm/hugetlb.c | 3 + mm/internal.h | 6 + mm/madvise.c | 168 +++ mm/memory.c | 18 +- mm/mprotect.c | 3 +- mm/mseal.c | 1 + mm/pagewalk.c | 200 ++- tools/include/uapi/asm-generic/mman-common.h | 3 + tools/testing/selftests/mm/.gitignore | 1 + tools/testing/selftests/mm/Makefile | 1 + tools/testing/selftests/mm/guard-pages.c | 1228 ++++++++++++++++++ 19 files changed, 1627 insertions(+), 66 deletions(-) create mode 100644 tools/testing/selftests/mm/guard-pages.c
-- 2.47.0