Currently all pages are sensitive and there's no way to change that. This patch introduces one. Changing sensitivity has some requirements: - It can only be done at pageblock granularity. This means that there's never a need to allocate pagetables to do it (since the restricted direct map is always pre-allocated down to pageblock granularity). - Flipping pages from nonsensitive to sensitive (unmapping) requires a TLB shootdown, meaning IRQs must be enabled. - Flipping from sensitive to nonsensitive requires zeroing pages, which seems like an undesirable thing to do with a spinlock held. This makes allocations that need to change sensitivity _somewhat_ similar to those that need to fallback to a different migratetype. But, the locking requirements mean that this can't just be squashed into the existing "fallback" allocator logic, instead a new allocator path just for this purpose is needed. The new path is assumed to be much cheaper than the really heavyweight stuff like compaction and reclaim. But at present it is treated as less desirable than the mobility-related "fallback" and "stealing" logic. This might turn out to need revision (in particular, maybe it's a problem that __rmqueue_steal(), which causes fragmentation, happens before __rmqueue_asi()), but that should be treated as a subsequent optimisation project. Now that !__GFP_SENSITIVE allocations are no longer doomed to fail, stop hard-coding it at the top of the allocator. Note that this does loads of unnecessary TLB flushes and IPIs. The design goal here is that the transitions are rare enough that this doesn't matter a huge amount, but it should still be addressed in later patches. Signed-off-by: Brendan Jackman --- arch/x86/include/asm/set_memory.h | 10 ++++ arch/x86/mm/pat/set_memory.c | 28 ++++++++++ include/linux/set_memory.h | 6 +++ mm/page_alloc.c | 106 +++++++++++++++++++++++++++++++++----- 4 files changed, 138 insertions(+), 12 deletions(-) diff --git a/arch/x86/include/asm/set_memory.h b/arch/x86/include/asm/set_memory.h index 61f56cdaccb5af18e36790677b635b4ab6f5e24d..396580693e7d1317537148c0c219296e2b7c13fd 100644 --- a/arch/x86/include/asm/set_memory.h +++ b/arch/x86/include/asm/set_memory.h @@ -92,6 +92,16 @@ int set_direct_map_default_noflush(struct page *page); int set_direct_map_valid_noflush(struct page *page, unsigned nr, bool valid); bool kernel_page_present(struct page *page); +#ifdef CONFIG_MITIGATION_ADDRESS_SPACE_ISOLATION +int set_direct_map_sensitive(struct page *page, int num_pageblocks, bool sensitive); +#else /* CONFIG_MITIGATION_ADDRESS_SPACE_ISOLATION */ +static inline +int set_direct_map_sensitive(struct page *page, int num_pageblocks, bool sensitive) +{ + return 0; +} +#endif /* CONFIG_MITIGATION_ADDRESS_SPACE_ISOLATION */ + extern int kernel_set_to_readonly; #endif /* _ASM_X86_SET_MEMORY_H */ diff --git a/arch/x86/mm/pat/set_memory.c b/arch/x86/mm/pat/set_memory.c index 2a50844515e81913fed32d5b6d1ec19e8e249533..88fb65574d4fa0089fa31a9a06fe096c408991e6 100644 --- a/arch/x86/mm/pat/set_memory.c +++ b/arch/x86/mm/pat/set_memory.c @@ -4,6 +4,7 @@ * Thanks to Ben LaHaise for precious feedback. */ #include +#include #include #include #include @@ -2695,6 +2696,33 @@ int set_direct_map_valid_noflush(struct page *page, unsigned nr, bool valid) return __set_pages_np(page, nr); } +#ifdef CONFIG_MITIGATION_ADDRESS_SPACE_ISOLATION +/* + * Map/unmap a set of contiguous pageblocks into all ASI restricted address + * spaces. All pagetables are pre-allocated so this can be called anywhere. + * This should not be called on pages that may be mapped elsewhere. + */ +int set_direct_map_sensitive(struct page *page, int num_pageblocks, bool sensitive) +{ + if (WARN_ON_ONCE(!IS_ALIGNED(page_to_pfn(page), 1 << pageblock_order))) + return -EINVAL; + + unsigned long tempaddr = (unsigned long)page_address(page); + struct cpa_data cpa = { .vaddr = &tempaddr, + .pgd = asi_nonsensitive_pgd, + .numpages = num_pageblocks << pageblock_order, + .flags = CPA_NO_CHECK_ALIAS, + .on_fault = CPA_FAULT_ERROR, }; + + if (sensitive) + cpa.mask_clr = __pgprot(_PAGE_PRESENT); + else + cpa.mask_set = __pgprot(_PAGE_PRESENT); + + return __change_page_attr_set_clr(&cpa, 1); +} +#endif /* CONFIG_MITIGATION_ADDRESS_SPACE_ISOLATION */ + #ifdef CONFIG_DEBUG_PAGEALLOC void __kernel_map_pages(struct page *page, int numpages, int enable) { diff --git a/include/linux/set_memory.h b/include/linux/set_memory.h index 3030d9245f5ac8a35b27e249c6d8b9539f148635..db4225c046c47c114293af8b504886b103dc94ce 100644 --- a/include/linux/set_memory.h +++ b/include/linux/set_memory.h @@ -44,6 +44,12 @@ static inline bool kernel_page_present(struct page *page) { return true; } + +static inline int set_direct_map_sensitive(struct page *page, + int num_pageblocks, bool sensitive) { + return 0; +} + #else /* CONFIG_ARCH_HAS_SET_DIRECT_MAP */ /* * Some architectures, e.g. ARM64 can disable direct map modifications at diff --git a/mm/page_alloc.c b/mm/page_alloc.c index b0aeb97baa13af038fff0edae33affbbf49e825c..a8e3556643b0ff2fe1d35a678937270356006d34 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -44,6 +44,7 @@ #include #include #include +#include #include #include #include @@ -585,6 +586,13 @@ static void set_pageblock_migratetype(struct page *page, PAGEBLOCK_MIGRATETYPE_MASK | PAGEBLOCK_ISO_MASK); } +static inline void set_pageblock_sensitive(struct page *page, bool sensitive) +{ + __set_pfnblock_flags_mask(page, page_to_pfn(page), + sensitive ? 0 : PAGEBLOCK_NONSENSITIVE_MASK, + PAGEBLOCK_NONSENSITIVE_MASK); +} + void __meminit init_pageblock_migratetype(struct page *page, enum migratetype migratetype, bool isolate) @@ -3264,6 +3272,85 @@ static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, #endif } +#ifdef CONFIG_MITIGATION_ADDRESS_SPACE_ISOLATION +static inline struct page *__rmqueue_asi(struct zone *zone, unsigned int request_order, + unsigned int alloc_flags, freetype_t freetype) +{ + freetype_t freetype_other = migrate_to_freetype( + free_to_migratetype(freetype), !freetype_sensitive(freetype)); + unsigned long flags; + struct page *page; + int alloc_order; + enum rmqueue_mode rmqm = RMQUEUE_NORMAL; + int nr_pageblocks; + + if (!asi_enabled_static()) + return NULL; + + /* + * Might need a TLB shootdown. Even if IRQs are on this isn't + * safe if the caller holds a lock (in case the other CPUs need that + * lock to handle the shootdown IPI). + */ + if (alloc_flags & ALLOC_NOBLOCK) + return NULL; + lockdep_assert(!irqs_disabled() || unlikely(early_boot_irqs_disabled)); + + /* + * Need to [un]map a whole pageblock (otherwise it might require + * allocating pagetables). First allocate it. + */ + alloc_order = max(request_order, pageblock_order); + nr_pageblocks = 1 << (alloc_order - pageblock_order); + spin_lock_irqsave(&zone->lock, flags); + page = __rmqueue(zone, alloc_order, freetype_other, alloc_flags, &rmqm); + spin_unlock_irqrestore(&zone->lock, flags); + if (!page) + return NULL; + + if (!freetype_sensitive(freetype)) { + /* + * These pages were formerly sensitive so we need to clear them + * out before exposing them to CPU attacks. Doing this with the + * zone lock held would have been undesirable. + */ + kernel_init_pages(page, 1 << alloc_order); + } + + /* + * Now that IRQs are on it's safe to do a TLB shootdown, so change + * mapping and update pageblock flags. + */ + set_direct_map_sensitive(page, nr_pageblocks, freetype_sensitive(freetype)); + for (int i = 0; i < nr_pageblocks; i++) { + struct page *block_page = page + (pageblock_nr_pages * i); + + set_pageblock_sensitive(block_page, freetype_sensitive(freetype)); + } + + if (request_order >= alloc_order) + return page; + + /* Free any remaining pages in the block. */ + spin_lock_irqsave(&zone->lock, flags); + for (unsigned int i = request_order; i < alloc_order; i++) { + struct page *page_to_free = page + (1 << i); + + __free_one_page(page_to_free, page_to_pfn(page_to_free), zone, + i, freetype, FPI_SKIP_REPORT_NOTIFY); + } + spin_unlock_irqrestore(&zone->lock, flags); + + return page; +} +#else /* CONFIG_MITIGATION_ADDRESS_SPACE_ISOLATION */ +static inline struct page *__rmqueue_asi(struct zone *zone, unsigned int request_order, + unsigned int alloc_flags, freetype_t freetype) +{ + return NULL; +} +#endif /* CONFIG_MITIGATION_ADDRESS_SPACE_ISOLATION */ + static __always_inline struct page *rmqueue_buddy(struct zone *preferred_zone, struct zone *zone, unsigned int order, unsigned int alloc_flags, @@ -3297,13 +3384,15 @@ struct page *rmqueue_buddy(struct zone *preferred_zone, struct zone *zone, */ if (!page && (alloc_flags & (ALLOC_OOM|ALLOC_HARDER))) page = __rmqueue_smallest(zone, order, ft_high); - - if (!page) { - spin_unlock_irqrestore(&zone->lock, flags); - return NULL; - } } spin_unlock_irqrestore(&zone->lock, flags); + + /* Try changing sensitivity, now we've released the zone lock */ + if (!page) + page = __rmqueue_asi(zone, order, alloc_flags, freetype); + if (!page) + return NULL; + } while (check_new_pages(page, order)); __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); @@ -5285,13 +5374,6 @@ struct page *__alloc_frozen_pages_noprof(gfp_t gfp, unsigned int order, gfp_t alloc_gfp; /* The gfp_t that was actually used for allocation */ struct alloc_context ac = { }; - /* - * Temporary hack: Allocation of nonsensitive pages is not possible yet, - * allocate everything sensitive. The restricted address space is never - * actually entered yet so this is fine. - */ - gfp |= __GFP_SENSITIVE; - /* * There are several places where we assume that the order value is sane * so bail out early if the request is out of bound. -- 2.50.1