This is a note to let you know that I've just added the patch titled

kaiser: merged update

to the 4.9-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git%3Ba=su...

The filename of the patch is: kaiser-merged-update.patch and it can be found in the queue-4.9 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree, please let stable@vger.kernel.org know about it.

From foo@baz Wed Jan 3 20:37:21 CET 2018

From: Dave Hansen dave.hansen@linux.intel.com Date: Wed, 30 Aug 2017 16:23:00 -0700 Subject: kaiser: merged update

From: Dave Hansen dave.hansen@linux.intel.com

Merged fixes and cleanups, rebased to 4.9.51 tree (no 5-level paging).

Signed-off-by: Hugh Dickins hughd@google.com Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org --- arch/x86/entry/entry_64.S | 105 ++++++++++- arch/x86/include/asm/kaiser.h | 43 ++-- arch/x86/include/asm/pgtable.h | 18 + arch/x86/include/asm/pgtable_64.h | 48 ++++- arch/x86/include/asm/pgtable_types.h | 6 arch/x86/kernel/espfix_64.c | 13 - arch/x86/kernel/head_64.S | 19 +- arch/x86/kernel/ldt.c | 27 ++ arch/x86/kernel/tracepoint.c | 2 arch/x86/mm/kaiser.c | 317 +++++++++++++++++++++++++---------- arch/x86/mm/pageattr.c | 63 +++++- arch/x86/mm/pgtable.c | 40 +--- include/linux/kaiser.h | 26 ++ kernel/fork.c | 9 security/Kconfig | 5 15 files changed, 551 insertions(+), 190 deletions(-) create mode 100644 include/linux/kaiser.h

--- a/arch/x86/entry/entry_64.S +++ b/arch/x86/entry/entry_64.S @@ -230,6 +230,13 @@ entry_SYSCALL_64_fastpath: movq RIP(%rsp), %rcx movq EFLAGS(%rsp), %r11 RESTORE_C_REGS_EXCEPT_RCX_R11 + /* + * This opens a window where we have a user CR3, but are + * running in the kernel. This makes using the CS + * register useless for telling whether or not we need to + * switch CR3 in NMIs. Normal interrupts are OK because + * they are off here. + */ SWITCH_USER_CR3 movq RSP(%rsp), %rsp USERGS_SYSRET64 @@ -326,11 +333,25 @@ return_from_SYSCALL_64: syscall_return_via_sysret: /* rcx and r11 are already restored (see code above) */ RESTORE_C_REGS_EXCEPT_RCX_R11 + /* + * This opens a window where we have a user CR3, but are + * running in the kernel. This makes using the CS + * register useless for telling whether or not we need to + * switch CR3 in NMIs. Normal interrupts are OK because + * they are off here. + */ SWITCH_USER_CR3 movq RSP(%rsp), %rsp USERGS_SYSRET64

opportunistic_sysret_failed: + /* + * This opens a window where we have a user CR3, but are + * running in the kernel. This makes using the CS + * register useless for telling whether or not we need to + * switch CR3 in NMIs. Normal interrupts are OK because + * they are off here. + */ SWITCH_USER_CR3 SWAPGS jmp restore_c_regs_and_iret @@ -1087,6 +1108,13 @@ ENTRY(error_entry) cld SAVE_C_REGS 8 SAVE_EXTRA_REGS 8 + /* + * error_entry() always returns with a kernel gsbase and + * CR3. We must also have a kernel CR3/gsbase before + * calling TRACE_IRQS_*. Just unconditionally switch to + * the kernel CR3 here. + */ + SWITCH_KERNEL_CR3 xorl %ebx, %ebx testb $3, CS+8(%rsp) jz .Lerror_kernelspace @@ -1096,7 +1124,6 @@ ENTRY(error_entry) * from user mode due to an IRET fault. */ SWAPGS - SWITCH_KERNEL_CR3

.Lerror_entry_from_usermode_after_swapgs: /* @@ -1148,7 +1175,6 @@ ENTRY(error_entry) * Switch to kernel gsbase: */ SWAPGS - SWITCH_KERNEL_CR3

/* * Pretend that the exception came from user mode: set up pt_regs @@ -1249,7 +1275,10 @@ ENTRY(nmi) */

SWAPGS_UNSAFE_STACK - SWITCH_KERNEL_CR3_NO_STACK + /* + * percpu variables are mapped with user CR3, so no need + * to switch CR3 here. + */ cld movq %rsp, %rdx movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp @@ -1283,14 +1312,33 @@ ENTRY(nmi)

movq %rsp, %rdi movq $-1, %rsi +#ifdef CONFIG_KAISER + /* Unconditionally use kernel CR3 for do_nmi() */ + /* %rax is saved above, so OK to clobber here */ + movq %cr3, %rax + pushq %rax +#ifdef CONFIG_KAISER_REAL_SWITCH + andq $(~0x1000), %rax +#endif + movq %rax, %cr3 +#endif call do_nmi + /* + * Unconditionally restore CR3. I know we return to + * kernel code that needs user CR3, but do we ever return + * to "user mode" where we need the kernel CR3? + */ +#ifdef CONFIG_KAISER + popq %rax + mov %rax, %cr3 +#endif

/* * Return back to user mode. We must *not* do the normal exit - * work, because we don't want to enable interrupts. Fortunately, - * do_nmi doesn't modify pt_regs. + * work, because we don't want to enable interrupts. Do not + * switch to user CR3: we might be going back to kernel code + * that had a user CR3 set. */ - SWITCH_USER_CR3 SWAPGS jmp restore_c_regs_and_iret

@@ -1486,23 +1534,54 @@ end_repeat_nmi: ALLOC_PT_GPREGS_ON_STACK

/* - * Use paranoid_entry to handle SWAPGS, but no need to use paranoid_exit - * as we should not be calling schedule in NMI context. - * Even with normal interrupts enabled. An NMI should not be - * setting NEED_RESCHED or anything that normal interrupts and - * exceptions might do. + * Use the same approach as paranoid_entry to handle SWAPGS, but + * without CR3 handling since we do that differently in NMIs. No + * need to use paranoid_exit as we should not be calling schedule + * in NMI context. Even with normal interrupts enabled. An NMI + * should not be setting NEED_RESCHED or anything that normal + * interrupts and exceptions might do. */ - call paranoid_entry + cld + SAVE_C_REGS + SAVE_EXTRA_REGS + movl $1, %ebx + movl $MSR_GS_BASE, %ecx + rdmsr + testl %edx, %edx + js 1f /* negative -> in kernel */ + SWAPGS + xorl %ebx, %ebx +1: +#ifdef CONFIG_KAISER + /* Unconditionally use kernel CR3 for do_nmi() */ + /* %rax is saved above, so OK to clobber here */ + movq %cr3, %rax + pushq %rax +#ifdef CONFIG_KAISER_REAL_SWITCH + andq $(~0x1000), %rax +#endif + movq %rax, %cr3 +#endif

/* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */ movq %rsp, %rdi + addq $8, %rdi /* point %rdi at ptregs, fixed up for CR3 */ movq $-1, %rsi call do_nmi + /* + * Unconditionally restore CR3. We might be returning to + * kernel code that needs user CR3, like just just before + * a sysret. + */ +#ifdef CONFIG_KAISER + popq %rax + mov %rax, %cr3 +#endif

testl %ebx, %ebx /* swapgs needed? */ jnz nmi_restore nmi_swapgs: - SWITCH_USER_CR3_NO_STACK + /* We fixed up CR3 above, so no need to switch it here */ SWAPGS_UNSAFE_STACK nmi_restore: RESTORE_EXTRA_REGS --- a/arch/x86/include/asm/kaiser.h +++ b/arch/x86/include/asm/kaiser.h @@ -16,13 +16,17 @@

.macro _SWITCH_TO_KERNEL_CR3 reg movq %cr3, \reg +#ifdef CONFIG_KAISER_REAL_SWITCH andq $(~0x1000), \reg +#endif movq \reg, %cr3 .endm

.macro _SWITCH_TO_USER_CR3 reg movq %cr3, \reg +#ifdef CONFIG_KAISER_REAL_SWITCH orq $(0x1000), \reg +#endif movq \reg, %cr3 .endm

@@ -65,48 +69,53 @@ movq PER_CPU_VAR(unsafe_stack_register_b .endm

#endif /* CONFIG_KAISER */ + #else /* __ASSEMBLY__ */

#ifdef CONFIG_KAISER -// Upon kernel/user mode switch, it may happen that -// the address space has to be switched before the registers have been stored. -// To change the address space, another register is needed. -// A register therefore has to be stored/restored. -// -DECLARE_PER_CPU_USER_MAPPED(unsigned long, unsafe_stack_register_backup); +/* + * Upon kernel/user mode switch, it may happen that the address + * space has to be switched before the registers have been + * stored. To change the address space, another register is + * needed. A register therefore has to be stored/restored. +*/

-#endif /* CONFIG_KAISER */ +DECLARE_PER_CPU_USER_MAPPED(unsigned long, unsafe_stack_register_backup);

/** - * shadowmem_add_mapping - map a virtual memory part to the shadow mapping + * kaiser_add_mapping - map a virtual memory part to the shadow (user) mapping * @addr: the start address of the range * @size: the size of the range * @flags: The mapping flags of the pages * - * the mapping is done on a global scope, so no bigger synchronization has to be done. - * the pages have to be manually unmapped again when they are not needed any longer. + * The mapping is done on a global scope, so no bigger + * synchronization has to be done. the pages have to be + * manually unmapped again when they are not needed any longer. */ -extern void kaiser_add_mapping(unsigned long addr, unsigned long size, unsigned long flags); +extern int kaiser_add_mapping(unsigned long addr, unsigned long size, unsigned long flags);

/** - * shadowmem_remove_mapping - unmap a virtual memory part of the shadow mapping + * kaiser_remove_mapping - unmap a virtual memory part of the shadow mapping * @addr: the start address of the range * @size: the size of the range */ extern void kaiser_remove_mapping(unsigned long start, unsigned long size);

/** - * shadowmem_initialize_mapping - Initalize the shadow mapping + * kaiser_initialize_mapping - Initalize the shadow mapping * - * most parts of the shadow mapping can be mapped upon boot time. - * only the thread stacks have to be mapped on runtime. - * the mapped regions are not unmapped at all. + * Most parts of the shadow mapping can be mapped upon boot + * time. Only per-process things like the thread stacks + * or a new LDT have to be mapped at runtime. These boot- + * time mappings are permanent and nevertunmapped. */ extern void kaiser_init(void);

-#endif +#endif /* CONFIG_KAISER */ + +#endif /* __ASSEMBLY */

--- a/arch/x86/include/asm/pgtable.h +++ b/arch/x86/include/asm/pgtable.h @@ -690,7 +690,17 @@ static inline pud_t *pud_offset(pgd_t *p

static inline int pgd_bad(pgd_t pgd) { - return (pgd_flags(pgd) & ~_PAGE_USER) != _KERNPG_TABLE; + pgdval_t ignore_flags = _PAGE_USER; + /* + * We set NX on KAISER pgds that map userspace memory so + * that userspace can not meaningfully use the kernel + * page table by accident; it will fault on the first + * instruction it tries to run. See native_set_pgd(). + */ + if (IS_ENABLED(CONFIG_KAISER)) + ignore_flags |= _PAGE_NX; + + return (pgd_flags(pgd) & ~ignore_flags) != _KERNPG_TABLE; }

static inline int pgd_none(pgd_t pgd) @@ -905,8 +915,10 @@ static inline void clone_pgd_range(pgd_t { memcpy(dst, src, count * sizeof(pgd_t)); #ifdef CONFIG_KAISER - // clone the shadow pgd part as well - memcpy(native_get_shadow_pgd(dst), native_get_shadow_pgd(src), count * sizeof(pgd_t)); + /* Clone the shadow pgd part as well */ + memcpy(native_get_shadow_pgd(dst), + native_get_shadow_pgd(src), + count * sizeof(pgd_t)); #endif }

--- a/arch/x86/include/asm/pgtable_64.h +++ b/arch/x86/include/asm/pgtable_64.h @@ -107,26 +107,58 @@ static inline void native_pud_clear(pud_ }

#ifdef CONFIG_KAISER -static inline pgd_t * native_get_shadow_pgd(pgd_t *pgdp) { +static inline pgd_t * native_get_shadow_pgd(pgd_t *pgdp) +{ return (pgd_t *)(void*)((unsigned long)(void*)pgdp | (unsigned long)PAGE_SIZE); }

-static inline pgd_t * native_get_normal_pgd(pgd_t *pgdp) { +static inline pgd_t * native_get_normal_pgd(pgd_t *pgdp) +{ return (pgd_t *)(void*)((unsigned long)(void*)pgdp & ~(unsigned long)PAGE_SIZE); } +#else +static inline pgd_t * native_get_shadow_pgd(pgd_t *pgdp) +{ + BUILD_BUG_ON(1); + return NULL; +} +static inline pgd_t * native_get_normal_pgd(pgd_t *pgdp) +{ + return pgdp; +} #endif /* CONFIG_KAISER */

+/* + * Page table pages are page-aligned. The lower half of the top + * level is used for userspace and the top half for the kernel. + * This returns true for user pages that need to get copied into + * both the user and kernel copies of the page tables, and false + * for kernel pages that should only be in the kernel copy. + */ +static inline bool is_userspace_pgd(void *__ptr) +{ + unsigned long ptr = (unsigned long)__ptr; + + return ((ptr % PAGE_SIZE) < (PAGE_SIZE / 2)); +} + static inline void native_set_pgd(pgd_t *pgdp, pgd_t pgd) { #ifdef CONFIG_KAISER - // We know that a pgd is page aligned. - // Therefore the lower indices have to be mapped to user space. - // These pages are mapped to the shadow mapping. - if ((((unsigned long)pgdp) % PAGE_SIZE) < (PAGE_SIZE / 2)) { + pteval_t extra_kern_pgd_flags = 0; + /* Do we need to also populate the shadow pgd? */ + if (is_userspace_pgd(pgdp)) { native_get_shadow_pgd(pgdp)->pgd = pgd.pgd; + /* + * Even if the entry is *mapping* userspace, ensure + * that userspace can not use it. This way, if we + * get out to userspace running on the kernel CR3, + * userspace will crash instead of running. + */ + extra_kern_pgd_flags = _PAGE_NX; } - - pgdp->pgd = pgd.pgd & ~_PAGE_USER; + pgdp->pgd = pgd.pgd; + pgdp->pgd |= extra_kern_pgd_flags; #else /* CONFIG_KAISER */ *pgdp = pgd; #endif --- a/arch/x86/include/asm/pgtable_types.h +++ b/arch/x86/include/asm/pgtable_types.h @@ -48,7 +48,7 @@ #ifdef CONFIG_KAISER #define _PAGE_GLOBAL (_AT(pteval_t, 0)) #else -#define _PAGE_GLOBAL (_AT(pteval_t, 1) << _PAGE_BIT_GLOBAL) +#define _PAGE_GLOBAL (_AT(pteval_t, 1) << _PAGE_BIT_GLOBAL) #endif #define _PAGE_SOFTW1 (_AT(pteval_t, 1) << _PAGE_BIT_SOFTW1) #define _PAGE_SOFTW2 (_AT(pteval_t, 1) << _PAGE_BIT_SOFTW2) @@ -123,11 +123,7 @@ #define _PAGE_DEVMAP (_AT(pteval_t, 0)) #endif

-#ifdef CONFIG_KAISER -#define _PAGE_PROTNONE (_AT(pteval_t, 0)) -#else #define _PAGE_PROTNONE (_AT(pteval_t, 1) << _PAGE_BIT_PROTNONE) -#endif

#define _PAGE_TABLE (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | \ _PAGE_ACCESSED | _PAGE_DIRTY) --- a/arch/x86/kernel/espfix_64.c +++ b/arch/x86/kernel/espfix_64.c @@ -127,11 +127,14 @@ void __init init_espfix_bsp(void) /* Install the espfix pud into the kernel page directory */ pgd_p = &init_level4_pgt[pgd_index(ESPFIX_BASE_ADDR)]; pgd_populate(&init_mm, pgd_p, (pud_t *)espfix_pud_page); -#ifdef CONFIG_KAISER - // add the esp stack pud to the shadow mapping here. - // This can be done directly, because the fixup stack has its own pud - set_pgd(native_get_shadow_pgd(pgd_p), __pgd(_PAGE_TABLE | __pa((pud_t *)espfix_pud_page))); -#endif + /* + * Just copy the top-level PGD that is mapping the espfix + * area to ensure it is mapped into the shadow user page + * tables. + */ + if (IS_ENABLED(CONFIG_KAISER)) + set_pgd(native_get_shadow_pgd(pgd_p), + __pgd(_KERNPG_TABLE | __pa((pud_t *)espfix_pud_page)));

/* Randomize the locations */ init_espfix_random(); --- a/arch/x86/kernel/head_64.S +++ b/arch/x86/kernel/head_64.S @@ -406,11 +406,24 @@ GLOBAL(early_recursion_flag) GLOBAL(name)

#ifdef CONFIG_KAISER +/* + * Each PGD needs to be 8k long and 8k aligned. We do not + * ever go out to userspace with these, so we do not + * strictly *need* the second page, but this allows us to + * have a single set_pgd() implementation that does not + * need to worry about whether it has 4k or 8k to work + * with. + * + * This ensures PGDs are 8k long: + */ +#define KAISER_USER_PGD_FILL 512 +/* This ensures they are 8k-aligned: */ #define NEXT_PGD_PAGE(name) \ .balign 2 * PAGE_SIZE; \ GLOBAL(name) #else #define NEXT_PGD_PAGE(name) NEXT_PAGE(name) +#define KAISER_USER_PGD_FILL 0 #endif

/* Automate the creation of 1 to 1 mapping pmd entries */ @@ -425,6 +438,7 @@ GLOBAL(name) NEXT_PGD_PAGE(early_level4_pgt) .fill 511,8,0 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE + .fill KAISER_USER_PGD_FILL,8,0

NEXT_PAGE(early_dynamic_pgts) .fill 512*EARLY_DYNAMIC_PAGE_TABLES,8,0 @@ -433,7 +447,8 @@ NEXT_PAGE(early_dynamic_pgts)

#ifndef CONFIG_XEN NEXT_PGD_PAGE(init_level4_pgt) - .fill 2*512,8,0 + .fill 512,8,0 + .fill KAISER_USER_PGD_FILL,8,0 #else NEXT_PGD_PAGE(init_level4_pgt) .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE @@ -442,6 +457,7 @@ NEXT_PGD_PAGE(init_level4_pgt) .org init_level4_pgt + L4_START_KERNEL*8, 0 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */ .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE + .fill KAISER_USER_PGD_FILL,8,0

NEXT_PAGE(level3_ident_pgt) .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE @@ -452,6 +468,7 @@ NEXT_PAGE(level2_ident_pgt) */ PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD) #endif + .fill KAISER_USER_PGD_FILL,8,0

NEXT_PAGE(level3_kernel_pgt) .fill L3_START_KERNEL,8,0 --- a/arch/x86/kernel/ldt.c +++ b/arch/x86/kernel/ldt.c @@ -18,6 +18,7 @@ #include <linux/uaccess.h>

#include <asm/ldt.h> +#include <asm/kaiser.h> #include <asm/desc.h> #include <asm/mmu_context.h> #include <asm/syscalls.h> @@ -34,11 +35,21 @@ static void flush_ldt(void *current_mm) set_ldt(pc->ldt->entries, pc->ldt->size); }

+static void __free_ldt_struct(struct ldt_struct *ldt) +{ + if (ldt->size * LDT_ENTRY_SIZE > PAGE_SIZE) + vfree(ldt->entries); + else + free_page((unsigned long)ldt->entries); + kfree(ldt); +} + /* The caller must call finalize_ldt_struct on the result. LDT starts zeroed. */ static struct ldt_struct *alloc_ldt_struct(int size) { struct ldt_struct *new_ldt; int alloc_size; + int ret = 0;

if (size > LDT_ENTRIES) return NULL; @@ -66,6 +77,14 @@ static struct ldt_struct *alloc_ldt_stru return NULL; }

+ // FIXME: make kaiser_add_mapping() return an error code + // when it fails + kaiser_add_mapping((unsigned long)new_ldt->entries, alloc_size, + __PAGE_KERNEL); + if (ret) { + __free_ldt_struct(new_ldt); + return NULL; + } new_ldt->size = size; return new_ldt; } @@ -92,12 +111,10 @@ static void free_ldt_struct(struct ldt_s if (likely(!ldt)) return;

+ kaiser_remove_mapping((unsigned long)ldt->entries, + ldt->size * LDT_ENTRY_SIZE); paravirt_free_ldt(ldt->entries, ldt->size); - if (ldt->size * LDT_ENTRY_SIZE > PAGE_SIZE) - vfree(ldt->entries); - else - free_page((unsigned long)ldt->entries); - kfree(ldt); + __free_ldt_struct(ldt); }

/* --- a/arch/x86/kernel/tracepoint.c +++ b/arch/x86/kernel/tracepoint.c @@ -9,10 +9,12 @@ #include <linux/atomic.h>

atomic_t trace_idt_ctr = ATOMIC_INIT(0); +__aligned(PAGE_SIZE) struct desc_ptr trace_idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) trace_idt_table };

/* No need to be aligned, but done to keep all IDTs defined the same way. */ +__aligned(PAGE_SIZE) gate_desc trace_idt_table[NR_VECTORS] __page_aligned_bss;

static int trace_irq_vector_refcount; --- a/arch/x86/mm/kaiser.c +++ b/arch/x86/mm/kaiser.c @@ -1,160 +1,305 @@ - - +#include <linux/bug.h> #include <linux/kernel.h> #include <linux/errno.h> #include <linux/string.h> #include <linux/types.h> #include <linux/bug.h> #include <linux/init.h> +#include <linux/interrupt.h> #include <linux/spinlock.h> #include <linux/mm.h> - #include <linux/uaccess.h> + +#include <asm/kaiser.h> #include <asm/pgtable.h> #include <asm/pgalloc.h> #include <asm/desc.h> #ifdef CONFIG_KAISER

__visible DEFINE_PER_CPU_USER_MAPPED(unsigned long, unsafe_stack_register_backup); +/* + * At runtime, the only things we map are some things for CPU + * hotplug, and stacks for new processes. No two CPUs will ever + * be populating the same addresses, so we only need to ensure + * that we protect between two CPUs trying to allocate and + * populate the same page table page. + * + * Only take this lock when doing a set_p[4um]d(), but it is not + * needed for doing a set_pte(). We assume that only the *owner* + * of a given allocation will be doing this for _their_ + * allocation. + * + * This ensures that once a system has been running for a while + * and there have been stacks all over and these page tables + * are fully populated, there will be no further acquisitions of + * this lock. + */ +static DEFINE_SPINLOCK(shadow_table_allocation_lock);

-/** - * Get the real ppn from a address in kernel mapping. - * @param address The virtual adrress - * @return the physical address +/* + * Returns -1 on error. */ -static inline unsigned long get_pa_from_mapping (unsigned long address) +static inline unsigned long get_pa_from_mapping(unsigned long vaddr) { pgd_t *pgd; pud_t *pud; pmd_t *pmd; pte_t *pte;

- pgd = pgd_offset_k(address); - BUG_ON(pgd_none(*pgd) || pgd_large(*pgd)); + pgd = pgd_offset_k(vaddr); + /* + * We made all the kernel PGDs present in kaiser_init(). + * We expect them to stay that way. + */ + BUG_ON(pgd_none(*pgd)); + /* + * PGDs are either 512GB or 128TB on all x86_64 + * configurations. We don't handle these. + */ + BUG_ON(pgd_large(*pgd)); + + pud = pud_offset(pgd, vaddr); + if (pud_none(*pud)) { + WARN_ON_ONCE(1); + return -1; + }

- pud = pud_offset(pgd, address); - BUG_ON(pud_none(*pud)); + if (pud_large(*pud)) + return (pud_pfn(*pud) << PAGE_SHIFT) | (vaddr & ~PUD_PAGE_MASK);

- if (pud_large(*pud)) { - return (pud_pfn(*pud) << PAGE_SHIFT) | (address & ~PUD_PAGE_MASK); + pmd = pmd_offset(pud, vaddr); + if (pmd_none(*pmd)) { + WARN_ON_ONCE(1); + return -1; }

- pmd = pmd_offset(pud, address); - BUG_ON(pmd_none(*pmd)); + if (pmd_large(*pmd)) + return (pmd_pfn(*pmd) << PAGE_SHIFT) | (vaddr & ~PMD_PAGE_MASK);

- if (pmd_large(*pmd)) { - return (pmd_pfn(*pmd) << PAGE_SHIFT) | (address & ~PMD_PAGE_MASK); + pte = pte_offset_kernel(pmd, vaddr); + if (pte_none(*pte)) { + WARN_ON_ONCE(1); + return -1; }

- pte = pte_offset_kernel(pmd, address); - BUG_ON(pte_none(*pte)); - - return (pte_pfn(*pte) << PAGE_SHIFT) | (address & ~PAGE_MASK); + return (pte_pfn(*pte) << PAGE_SHIFT) | (vaddr & ~PAGE_MASK); }

-void _kaiser_copy (unsigned long start_addr, unsigned long size, - unsigned long flags) +/* + * This is a relatively normal page table walk, except that it + * also tries to allocate page tables pages along the way. + * + * Returns a pointer to a PTE on success, or NULL on failure. + */ +static pte_t *kaiser_pagetable_walk(unsigned long address, bool is_atomic) { - pgd_t *pgd; - pud_t *pud; pmd_t *pmd; - pte_t *pte; - unsigned long address; - unsigned long end_addr = start_addr + size; - unsigned long target_address; + pud_t *pud; + pgd_t *pgd = native_get_shadow_pgd(pgd_offset_k(address)); + gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);

- for (address = PAGE_ALIGN(start_addr - (PAGE_SIZE - 1)); - address < PAGE_ALIGN(end_addr); address += PAGE_SIZE) { - target_address = get_pa_from_mapping(address); + might_sleep(); + if (is_atomic) { + gfp &= ~GFP_KERNEL; + gfp |= __GFP_HIGH | __GFP_ATOMIC; + }

- pgd = native_get_shadow_pgd(pgd_offset_k(address)); + if (pgd_none(*pgd)) { + WARN_ONCE(1, "All shadow pgds should have been populated"); + return NULL; + } + BUILD_BUG_ON(pgd_large(*pgd) != 0);

- BUG_ON(pgd_none(*pgd) && "All shadow pgds should be mapped at this time\n"); - BUG_ON(pgd_large(*pgd)); + pud = pud_offset(pgd, address); + /* The shadow page tables do not use large mappings: */ + if (pud_large(*pud)) { + WARN_ON(1); + return NULL; + } + if (pud_none(*pud)) { + unsigned long new_pmd_page = __get_free_page(gfp); + if (!new_pmd_page) + return NULL; + spin_lock(&shadow_table_allocation_lock); + if (pud_none(*pud)) + set_pud(pud, __pud(_KERNPG_TABLE | __pa(new_pmd_page))); + else + free_page(new_pmd_page); + spin_unlock(&shadow_table_allocation_lock); + }

- pud = pud_offset(pgd, address); - if (pud_none(*pud)) { - set_pud(pud, __pud(_PAGE_TABLE | __pa(pmd_alloc_one(0, address)))); - } - BUG_ON(pud_large(*pud)); + pmd = pmd_offset(pud, address); + /* The shadow page tables do not use large mappings: */ + if (pmd_large(*pmd)) { + WARN_ON(1); + return NULL; + } + if (pmd_none(*pmd)) { + unsigned long new_pte_page = __get_free_page(gfp); + if (!new_pte_page) + return NULL; + spin_lock(&shadow_table_allocation_lock); + if (pmd_none(*pmd)) + set_pmd(pmd, __pmd(_KERNPG_TABLE | __pa(new_pte_page))); + else + free_page(new_pte_page); + spin_unlock(&shadow_table_allocation_lock); + }

- pmd = pmd_offset(pud, address); - if (pmd_none(*pmd)) { - set_pmd(pmd, __pmd(_PAGE_TABLE | __pa(pte_alloc_one_kernel(0, address)))); - } - BUG_ON(pmd_large(*pmd)); + return pte_offset_kernel(pmd, address); +}

- pte = pte_offset_kernel(pmd, address); +int kaiser_add_user_map(const void *__start_addr, unsigned long size, + unsigned long flags) +{ + int ret = 0; + pte_t *pte; + unsigned long start_addr = (unsigned long )__start_addr; + unsigned long address = start_addr & PAGE_MASK; + unsigned long end_addr = PAGE_ALIGN(start_addr + size); + unsigned long target_address; + + for (;address < end_addr; address += PAGE_SIZE) { + target_address = get_pa_from_mapping(address); + if (target_address == -1) { + ret = -EIO; + break; + } + pte = kaiser_pagetable_walk(address, false); if (pte_none(*pte)) { set_pte(pte, __pte(flags | target_address)); } else { - BUG_ON(__pa(pte_page(*pte)) != target_address); + pte_t tmp; + set_pte(&tmp, __pte(flags | target_address)); + WARN_ON_ONCE(!pte_same(*pte, tmp)); } } + return ret; }

-// at first, add a pmd for every pgd entry in the shadowmem-kernel-part of the kernel mapping -static inline void __init _kaiser_init(void) +static int kaiser_add_user_map_ptrs(const void *start, const void *end, unsigned long flags) +{ + unsigned long size = end - start; + + return kaiser_add_user_map(start, size, flags); +} + +/* + * Ensure that the top level of the (shadow) page tables are + * entirely populated. This ensures that all processes that get + * forked have the same entries. This way, we do not have to + * ever go set up new entries in older processes. + * + * Note: we never free these, so there are no updates to them + * after this. + */ +static void __init kaiser_init_all_pgds(void) { pgd_t *pgd; int i = 0;

pgd = native_get_shadow_pgd(pgd_offset_k((unsigned long )0)); for (i = PTRS_PER_PGD / 2; i < PTRS_PER_PGD; i++) { - set_pgd(pgd + i, __pgd(_PAGE_TABLE |__pa(pud_alloc_one(0, 0)))); + pgd_t new_pgd; + pud_t *pud = pud_alloc_one(&init_mm, PAGE_OFFSET + i * PGDIR_SIZE); + if (!pud) { + WARN_ON(1); + break; + } + new_pgd = __pgd(_KERNPG_TABLE |__pa(pud)); + /* + * Make sure not to stomp on some other pgd entry. + */ + if (!pgd_none(pgd[i])) { + WARN_ON(1); + continue; + } + set_pgd(pgd + i, new_pgd); } }

+#define kaiser_add_user_map_early(start, size, flags) do { \ + int __ret = kaiser_add_user_map(start, size, flags); \ + WARN_ON(__ret); \ +} while (0) + +#define kaiser_add_user_map_ptrs_early(start, end, flags) do { \ + int __ret = kaiser_add_user_map_ptrs(start, end, flags); \ + WARN_ON(__ret); \ +} while (0) + extern char __per_cpu_user_mapped_start[], __per_cpu_user_mapped_end[]; -spinlock_t shadow_table_lock; +/* + * If anything in here fails, we will likely die on one of the + * first kernel->user transitions and init will die. But, we + * will have most of the kernel up by then and should be able to + * get a clean warning out of it. If we BUG_ON() here, we run + * the risk of being before we have good console output. + */ void __init kaiser_init(void) { int cpu; - spin_lock_init(&shadow_table_lock); - - spin_lock(&shadow_table_lock);

- _kaiser_init(); + kaiser_init_all_pgds();

for_each_possible_cpu(cpu) { - // map the per cpu user variables - _kaiser_copy( - (unsigned long) (__per_cpu_user_mapped_start + per_cpu_offset(cpu)), - (unsigned long) __per_cpu_user_mapped_end - (unsigned long) __per_cpu_user_mapped_start, - __PAGE_KERNEL); - } - - // map the entry/exit text section, which is responsible to switch between user- and kernel mode - _kaiser_copy( - (unsigned long) __entry_text_start, - (unsigned long) __entry_text_end - (unsigned long) __entry_text_start, - __PAGE_KERNEL_RX); - - // the fixed map address of the idt_table - _kaiser_copy( - (unsigned long) idt_descr.address, - sizeof(gate_desc) * NR_VECTORS, - __PAGE_KERNEL_RO); + void *percpu_vaddr = __per_cpu_user_mapped_start + + per_cpu_offset(cpu); + unsigned long percpu_sz = __per_cpu_user_mapped_end - + __per_cpu_user_mapped_start; + kaiser_add_user_map_early(percpu_vaddr, percpu_sz, + __PAGE_KERNEL); + }

- spin_unlock(&shadow_table_lock); + /* + * Map the entry/exit text section, which is needed at + * switches from user to and from kernel. + */ + kaiser_add_user_map_ptrs_early(__entry_text_start, __entry_text_end, + __PAGE_KERNEL_RX); + +#if defined(CONFIG_FUNCTION_GRAPH_TRACER) || defined(CONFIG_KASAN) + kaiser_add_user_map_ptrs_early(__irqentry_text_start, + __irqentry_text_end, + __PAGE_KERNEL_RX); +#endif + kaiser_add_user_map_early((void *)idt_descr.address, + sizeof(gate_desc) * NR_VECTORS, + __PAGE_KERNEL_RO); +#ifdef CONFIG_TRACING + kaiser_add_user_map_early(&trace_idt_descr, + sizeof(trace_idt_descr), + __PAGE_KERNEL); + kaiser_add_user_map_early(&trace_idt_table, + sizeof(gate_desc) * NR_VECTORS, + __PAGE_KERNEL); +#endif + kaiser_add_user_map_early(&debug_idt_descr, sizeof(debug_idt_descr), + __PAGE_KERNEL); + kaiser_add_user_map_early(&debug_idt_table, + sizeof(gate_desc) * NR_VECTORS, + __PAGE_KERNEL); }

+extern void unmap_pud_range_nofree(pgd_t *pgd, unsigned long start, unsigned long end); // add a mapping to the shadow-mapping, and synchronize the mappings -void kaiser_add_mapping(unsigned long addr, unsigned long size, unsigned long flags) +int kaiser_add_mapping(unsigned long addr, unsigned long size, unsigned long flags) { - spin_lock(&shadow_table_lock); - _kaiser_copy(addr, size, flags); - spin_unlock(&shadow_table_lock); + return kaiser_add_user_map((const void *)addr, size, flags); }

-extern void unmap_pud_range(pgd_t *pgd, unsigned long start, unsigned long end); void kaiser_remove_mapping(unsigned long start, unsigned long size) { - pgd_t *pgd = native_get_shadow_pgd(pgd_offset_k(start)); - spin_lock(&shadow_table_lock); - do { - unmap_pud_range(pgd, start, start + size); - } while (pgd++ != native_get_shadow_pgd(pgd_offset_k(start + size))); - spin_unlock(&shadow_table_lock); + unsigned long end = start + size; + unsigned long addr; + + for (addr = start; addr < end; addr += PGDIR_SIZE) { + pgd_t *pgd = native_get_shadow_pgd(pgd_offset_k(addr)); + /* + * unmap_p4d_range() handles > P4D_SIZE unmaps, + * so no need to trim 'end'. + */ + unmap_pud_range_nofree(pgd, addr, end); + } } #endif /* CONFIG_KAISER */ --- a/arch/x86/mm/pageattr.c +++ b/arch/x86/mm/pageattr.c @@ -52,6 +52,7 @@ static DEFINE_SPINLOCK(cpa_lock); #define CPA_FLUSHTLB 1 #define CPA_ARRAY 2 #define CPA_PAGES_ARRAY 4 +#define CPA_FREE_PAGETABLES 8

#ifdef CONFIG_PROC_FS static unsigned long direct_pages_count[PG_LEVEL_NUM]; @@ -729,10 +730,13 @@ static int split_large_page(struct cpa_d return 0; }

-static bool try_to_free_pte_page(pte_t *pte) +static bool try_to_free_pte_page(struct cpa_data *cpa, pte_t *pte) { int i;

+ if (!(cpa->flags & CPA_FREE_PAGETABLES)) + return false; + for (i = 0; i < PTRS_PER_PTE; i++) if (!pte_none(pte[i])) return false; @@ -741,10 +745,13 @@ static bool try_to_free_pte_page(pte_t * return true; }

-static bool try_to_free_pmd_page(pmd_t *pmd) +static bool try_to_free_pmd_page(struct cpa_data *cpa, pmd_t *pmd) { int i;

+ if (!(cpa->flags & CPA_FREE_PAGETABLES)) + return false; + for (i = 0; i < PTRS_PER_PMD; i++) if (!pmd_none(pmd[i])) return false; @@ -753,7 +760,9 @@ static bool try_to_free_pmd_page(pmd_t * return true; }

-static bool unmap_pte_range(pmd_t *pmd, unsigned long start, unsigned long end) +static bool unmap_pte_range(struct cpa_data *cpa, pmd_t *pmd, + unsigned long start, + unsigned long end) { pte_t *pte = pte_offset_kernel(pmd, start);

@@ -764,22 +773,23 @@ static bool unmap_pte_range(pmd_t *pmd, pte++; }

- if (try_to_free_pte_page((pte_t *)pmd_page_vaddr(*pmd))) { + if (try_to_free_pte_page(cpa, (pte_t *)pmd_page_vaddr(*pmd))) { pmd_clear(pmd); return true; } return false; }

-static void __unmap_pmd_range(pud_t *pud, pmd_t *pmd, +static void __unmap_pmd_range(struct cpa_data *cpa, pud_t *pud, pmd_t *pmd, unsigned long start, unsigned long end) { - if (unmap_pte_range(pmd, start, end)) - if (try_to_free_pmd_page((pmd_t *)pud_page_vaddr(*pud))) + if (unmap_pte_range(cpa, pmd, start, end)) + if (try_to_free_pmd_page(cpa, (pmd_t *)pud_page_vaddr(*pud))) pud_clear(pud); }

-static void unmap_pmd_range(pud_t *pud, unsigned long start, unsigned long end) +static void unmap_pmd_range(struct cpa_data *cpa, pud_t *pud, + unsigned long start, unsigned long end) { pmd_t *pmd = pmd_offset(pud, start);

@@ -790,7 +800,7 @@ static void unmap_pmd_range(pud_t *pud, unsigned long next_page = (start + PMD_SIZE) & PMD_MASK; unsigned long pre_end = min_t(unsigned long, end, next_page);

- __unmap_pmd_range(pud, pmd, start, pre_end); + __unmap_pmd_range(cpa, pud, pmd, start, pre_end);

start = pre_end; pmd++; @@ -803,7 +813,8 @@ static void unmap_pmd_range(pud_t *pud, if (pmd_large(*pmd)) pmd_clear(pmd); else - __unmap_pmd_range(pud, pmd, start, start + PMD_SIZE); + __unmap_pmd_range(cpa, pud, pmd, + start, start + PMD_SIZE);

start += PMD_SIZE; pmd++; @@ -813,17 +824,19 @@ static void unmap_pmd_range(pud_t *pud, * 4K leftovers? */ if (start < end) - return __unmap_pmd_range(pud, pmd, start, end); + return __unmap_pmd_range(cpa, pud, pmd, start, end);

/* * Try again to free the PMD page if haven't succeeded above. */ if (!pud_none(*pud)) - if (try_to_free_pmd_page((pmd_t *)pud_page_vaddr(*pud))) + if (try_to_free_pmd_page(cpa, (pmd_t *)pud_page_vaddr(*pud))) pud_clear(pud); }

-void unmap_pud_range(pgd_t *pgd, unsigned long start, unsigned long end) +static void __unmap_pud_range(struct cpa_data *cpa, pgd_t *pgd, + unsigned long start, + unsigned long end) { pud_t *pud = pud_offset(pgd, start);

@@ -834,7 +847,7 @@ void unmap_pud_range(pgd_t *pgd, unsigne unsigned long next_page = (start + PUD_SIZE) & PUD_MASK; unsigned long pre_end = min_t(unsigned long, end, next_page);

- unmap_pmd_range(pud, start, pre_end); + unmap_pmd_range(cpa, pud, start, pre_end);

start = pre_end; pud++; @@ -848,7 +861,7 @@ void unmap_pud_range(pgd_t *pgd, unsigne if (pud_large(*pud)) pud_clear(pud); else - unmap_pmd_range(pud, start, start + PUD_SIZE); + unmap_pmd_range(cpa, pud, start, start + PUD_SIZE);

start += PUD_SIZE; pud++; @@ -858,7 +871,7 @@ void unmap_pud_range(pgd_t *pgd, unsigne * 2M leftovers? */ if (start < end) - unmap_pmd_range(pud, start, end); + unmap_pmd_range(cpa, pud, start, end);

/* * No need to try to free the PUD page because we'll free it in @@ -866,6 +879,24 @@ void unmap_pud_range(pgd_t *pgd, unsigne */ }

+static void unmap_pud_range(pgd_t *pgd, unsigned long start, unsigned long end) +{ + struct cpa_data cpa = { + .flags = CPA_FREE_PAGETABLES, + }; + + __unmap_pud_range(&cpa, pgd, start, end); +} + +void unmap_pud_range_nofree(pgd_t *pgd, unsigned long start, unsigned long end) +{ + struct cpa_data cpa = { + .flags = 0, + }; + + __unmap_pud_range(&cpa, pgd, start, end); +} + static int alloc_pte_page(pmd_t *pmd) { pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL | __GFP_NOTRACK); --- a/arch/x86/mm/pgtable.c +++ b/arch/x86/mm/pgtable.c @@ -344,40 +344,26 @@ static inline void _pgd_free(pgd_t *pgd) kmem_cache_free(pgd_cache, pgd); } #else -static inline pgd_t *_pgd_alloc(void) -{ -#ifdef CONFIG_KAISER - // Instead of one PML4, we aquire two PML4s and, thus, an 8kb-aligned memory - // block. Therefore, we have to allocate at least 3 pages. However, the - // __get_free_pages returns us 4 pages. Hence, we store the base pointer at - // the beginning of the page of our 8kb-aligned memory block in order to - // correctly free it afterwars.

- unsigned long pages = __get_free_pages(PGALLOC_GFP, get_order(4*PAGE_SIZE)); - - if(native_get_normal_pgd((pgd_t*) pages) == (pgd_t*) pages) - { - *((unsigned long*)(pages + 2 * PAGE_SIZE)) = pages; - return (pgd_t *) pages; - } - else - { - *((unsigned long*)(pages + 3 * PAGE_SIZE)) = pages; - return (pgd_t *) (pages + PAGE_SIZE); - } +#ifdef CONFIG_KAISER +/* + * Instead of one pmd, we aquire two pmds. Being order-1, it is + * both 8k in size and 8k-aligned. That lets us just flip bit 12 + * in a pointer to swap between the two 4k halves. + */ +#define PGD_ALLOCATION_ORDER 1 #else - return (pgd_t *)__get_free_page(PGALLOC_GFP); +#define PGD_ALLOCATION_ORDER 0 #endif + +static inline pgd_t *_pgd_alloc(void) +{ + return (pgd_t *)__get_free_pages(PGALLOC_GFP, PGD_ALLOCATION_ORDER); }

static inline void _pgd_free(pgd_t *pgd) { -#ifdef CONFIG_KAISER - unsigned long pages = *((unsigned long*) ((char*) pgd + 2 * PAGE_SIZE)); - free_pages(pages, get_order(4*PAGE_SIZE)); -#else - free_page((unsigned long)pgd); -#endif + free_pages((unsigned long)pgd, PGD_ALLOCATION_ORDER); } #endif /* CONFIG_X86_PAE */

--- /dev/null +++ b/include/linux/kaiser.h @@ -0,0 +1,26 @@ +#ifndef _INCLUDE_KAISER_H +#define _INCLUDE_KAISER_H + +#ifdef CONFIG_KAISER +#include <asm/kaiser.h> +#else + +/* + * These stubs are used whenever CONFIG_KAISER is off, which + * includes architectures that support KAISER, but have it + * disabled. + */ + +static inline void kaiser_init(void) +{ +} +static inline void kaiser_remove_mapping(unsigned long start, unsigned long size) +{ +} +static inline int kaiser_add_mapping(unsigned long addr, unsigned long size, unsigned long flags) +{ + return 0; +} + +#endif /* !CONFIG_KAISER */ +#endif /* _INCLUDE_KAISER_H */ --- a/kernel/fork.c +++ b/kernel/fork.c @@ -58,6 +58,7 @@ #include <linux/tsacct_kern.h> #include <linux/cn_proc.h> #include <linux/freezer.h> +#include <linux/kaiser.h> #include <linux/delayacct.h> #include <linux/taskstats_kern.h> #include <linux/random.h> @@ -472,7 +473,6 @@ void set_task_stack_end_magic(struct tas *stackend = STACK_END_MAGIC; /* for overflow detection */ }

-extern void kaiser_add_mapping(unsigned long addr, unsigned long size, unsigned long flags); static struct task_struct *dup_task_struct(struct task_struct *orig, int node) { struct task_struct *tsk; @@ -500,9 +500,10 @@ static struct task_struct *dup_task_stru * functions again. */ tsk->stack = stack; -#ifdef CONFIG_KAISER - kaiser_add_mapping((unsigned long)tsk->stack, THREAD_SIZE, __PAGE_KERNEL); -#endif + + err= kaiser_add_mapping((unsigned long)tsk->stack, THREAD_SIZE, __PAGE_KERNEL); + if (err) + goto free_stack; #ifdef CONFIG_VMAP_STACK tsk->stack_vm_area = stack_vm_area; #endif --- a/security/Kconfig +++ b/security/Kconfig @@ -32,12 +32,17 @@ config SECURITY If you are unsure how to answer this question, answer N. config KAISER bool "Remove the kernel mapping in user mode" + default y depends on X86_64 depends on !PARAVIRT help This enforces a strict kernel and user space isolation in order to close hardware side channels on kernel address information.

+config KAISER_REAL_SWITCH + bool "KAISER: actually switch page tables" + default y + config SECURITYFS bool "Enable the securityfs filesystem" help

Patches currently in stable-queue which might be from dave.hansen@linux.intel.com are

queue-4.9/x86-paravirt-dont-patch-flush_tlb_single.patch queue-4.9/kaiser-merged-update.patch