x86-64, espfix: Don't leak bits 31:16 of %esp returning to 16-bit stack

The IRET instruction, when returning to a 16-bit segment, only
restores the bottom 16 bits of the user space stack pointer. This
causes some 16-bit software to break, but it also leaks kernel state
to user space. We have a software workaround for that ("espfix") for
the 32-bit kernel, but it relies on a nonzero stack segment base which
is not available in 64-bit mode.

In checkin:

b3b42ac2cbae x86-64, modify_ldt: Ban 16-bit segments on 64-bit kernels

we "solved" this by forbidding 16-bit segments on 64-bit kernels, with
the logic that 16-bit support is crippled on 64-bit kernels anyway (no
V86 support), but it turns out that people are doing stuff like
running old Win16 binaries under Wine and expect it to work.

This works around this by creating percpu "ministacks", each of which
is mapped 2^16 times 64K apart. When we detect that the return SS is
on the LDT, we copy the IRET frame to the ministack and use the
relevant alias to return to userspace. The ministacks are mapped
readonly, so if IRET faults we promote #GP to #DF which is an IST
vector and thus has its own stack; we then do the fixup in the #DF
handler.

(Making #GP an IST exception would make the msr_safe functions unsafe
in NMI/MC context, and quite possibly have other effects.)

Special thanks to:

- Andy Lutomirski, for the suggestion of using very small stack slots
and copy (as opposed to map) the IRET frame there, and for the
suggestion to mark them readonly and let the fault promote to #DF.
- Konrad Wilk for paravirt fixup and testing.
- Borislav Petkov for testing help and useful comments.

Reported-by: Brian Gerst <[e-mail]>
Signed-off-by: H. Peter Anvin <[e-mail]>
Link: http://lkml.kernel.org/r/[e-mail]
Cc: Konrad Rzeszutek Wilk <[e-mail]>
Cc: Borislav Petkov <[e-mail]>
Cc: Andrew Lutomriski <[e-mail]>
Cc: Linus Torvalds <[e-mail]>
Cc: Dirk Hohndel <[e-mail]>
Cc: Arjan van de Ven <[e-mail]>
Cc: comex <[e-mail]>
Cc: Alexander van Heukelum <[e-mail]>
Cc: Boris Ostrovsky <[e-mail]>
Cc: <[e-mail]> # consider after upstream merge

Documentation/x86/x86_64/mm.txt

···
 ffffe90000000000 - ffffe9ffffffffff (=40 bits) hole
 ffffea0000000000 - ffffeaffffffffff (=40 bits) virtual memory map (1TB)
 ... unused hole ...
+ffffff0000000000 - ffffff7fffffffff (=39 bits) %esp fixup stacks
+... unused hole ...
 ffffffff80000000 - ffffffffa0000000 (=512 MB)  kernel text mapping, from phys 0
 ffffffffa0000000 - ffffffffff5fffff (=1525 MB) module mapping space
 ffffffffff600000 - ffffffffffdfffff (=8 MB) vsyscalls

arch/x86/include/asm/pgtable_64_types.h

···
 #define MODULES_VADDR    (__START_KERNEL_map + KERNEL_IMAGE_SIZE)
 #define MODULES_END      _AC(0xffffffffff000000, UL)
 #define MODULES_LEN   (MODULES_END - MODULES_VADDR)
+#define ESPFIX_PGD_ENTRY _AC(-2, UL)
+#define ESPFIX_BASE_ADDR (ESPFIX_PGD_ENTRY << PGDIR_SHIFT)
 
 #define EARLY_DYNAMIC_PAGE_TABLES	64
 

arch/x86/include/asm/setup.h

···
 static inline void x86_ce4100_early_setup(void) { }
 #endif
 
+extern void init_espfix_bsp(void);
+extern void init_espfix_ap(void);
+
 #ifndef _SETUP
 
 /*

arch/x86/kernel/Makefile

···
 obj-y			+= syscall_$(BITS).o vsyscall_gtod.o
 obj-$(CONFIG_X86_64)	+= vsyscall_64.o
 obj-$(CONFIG_X86_64)	+= vsyscall_emu_64.o
+obj-$(CONFIG_X86_64)	+= espfix_64.o
 obj-$(CONFIG_SYSFS)	+= ksysfs.o
 obj-y			+= bootflag.o e820.o
 obj-y			+= pci-dma.o quirks.o topology.o kdebugfs.o

arch/x86/kernel/entry_64.S

···
 #include <asm/asm.h>
 #include <asm/context_tracking.h>
 #include <asm/smap.h>
+#include <asm/pgtable_types.h>
 #include <linux/err.h>
 
 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this.  */
···
 	RESTORE_ARGS 1,8,1
 
 irq_return:
+	/*
+	 * Are we returning to a stack segment from the LDT?  Note: in
+	 * 64-bit mode SS:RSP on the exception stack is always valid.
+	 */
+	testb $4,(SS-RIP)(%rsp)
+	jnz irq_return_ldt
+
+irq_return_iret:
 	INTERRUPT_RETURN
-	_ASM_EXTABLE(irq_return, bad_iret)
+	_ASM_EXTABLE(irq_return_iret, bad_iret)
 
 #ifdef CONFIG_PARAVIRT
 ENTRY(native_iret)
 	iretq
 	_ASM_EXTABLE(native_iret, bad_iret)
 #endif
+
+irq_return_ldt:
+	pushq_cfi %rax
+	pushq_cfi %rdi
+	SWAPGS
+	movq PER_CPU_VAR(espfix_waddr),%rdi
+	movq %rax,(0*8)(%rdi)	/* RAX */
+	movq (2*8)(%rsp),%rax	/* RIP */
+	movq %rax,(1*8)(%rdi)
+	movq (3*8)(%rsp),%rax	/* CS */
+	movq %rax,(2*8)(%rdi)
+	movq (4*8)(%rsp),%rax	/* RFLAGS */
+	movq %rax,(3*8)(%rdi)
+	movq (6*8)(%rsp),%rax	/* SS */
+	movq %rax,(5*8)(%rdi)
+	movq (5*8)(%rsp),%rax	/* RSP */
+	movq %rax,(4*8)(%rdi)
+	andl $0xffff0000,%eax
+	popq_cfi %rdi
+	orq PER_CPU_VAR(espfix_stack),%rax
+	SWAPGS
+	movq %rax,%rsp
+	popq_cfi %rax
+	jmp irq_return_iret
 
 	.section .fixup,"ax"
 bad_iret:
···
 	call preempt_schedule_irq
 	jmp exit_intr
 #endif
-
 	CFI_ENDPROC
 END(common_interrupt)
+
+	/*
+	 * If IRET takes a fault on the espfix stack, then we
+	 * end up promoting it to a doublefault.  In that case,
+	 * modify the stack to make it look like we just entered
+	 * the #GP handler from user space, similar to bad_iret.
+	 */
+	ALIGN
+__do_double_fault:
+	XCPT_FRAME 1 RDI+8
+	movq RSP(%rdi),%rax		/* Trap on the espfix stack? */
+	sarq $PGDIR_SHIFT,%rax
+	cmpl $ESPFIX_PGD_ENTRY,%eax
+	jne do_double_fault		/* No, just deliver the fault */
+	cmpl $__KERNEL_CS,CS(%rdi)
+	jne do_double_fault
+	movq RIP(%rdi),%rax
+	cmpq $irq_return_iret,%rax
+#ifdef CONFIG_PARAVIRT
+	je 1f
+	cmpq $native_iret,%rax
+#endif
+	jne do_double_fault		/* This shouldn't happen... */
+1:
+	movq PER_CPU_VAR(kernel_stack),%rax
+	subq $(6*8-KERNEL_STACK_OFFSET),%rax	/* Reset to original stack */
+	movq %rax,RSP(%rdi)
+	movq $0,(%rax)			/* Missing (lost) #GP error code */
+	movq $general_protection,RIP(%rdi)
+	retq
+	CFI_ENDPROC
+END(__do_double_fault)
+
 /*
  * End of kprobes section
  */
···
 zeroentry bounds do_bounds
 zeroentry invalid_op do_invalid_op
 zeroentry device_not_available do_device_not_available
-paranoiderrorentry double_fault do_double_fault
+paranoiderrorentry double_fault __do_double_fault
 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
 errorentry invalid_TSS do_invalid_TSS
 errorentry segment_not_present do_segment_not_present
···
  */
 error_kernelspace:
 	incl %ebx
-	leaq irq_return(%rip),%rcx
+	leaq irq_return_iret(%rip),%rcx
 	cmpq %rcx,RIP+8(%rsp)
 	je error_swapgs
 	movl %ecx,%eax	/* zero extend */

arch/x86/kernel/espfix_64.c

···
+/* ----------------------------------------------------------------------- *
+ *
+ *   Copyright 2014 Intel Corporation; author: H. Peter Anvin
+ *
+ *   This program is free software; you can redistribute it and/or modify it
+ *   under the terms and conditions of the GNU General Public License,
+ *   version 2, as published by the Free Software Foundation.
+ *
+ *   This program is distributed in the hope it will be useful, but WITHOUT
+ *   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ *   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ *   more details.
+ *
+ * ----------------------------------------------------------------------- */
+
+/*
+ * The IRET instruction, when returning to a 16-bit segment, only
+ * restores the bottom 16 bits of the user space stack pointer.  This
+ * causes some 16-bit software to break, but it also leaks kernel state
+ * to user space.
+ *
+ * This works around this by creating percpu "ministacks", each of which
+ * is mapped 2^16 times 64K apart.  When we detect that the return SS is
+ * on the LDT, we copy the IRET frame to the ministack and use the
+ * relevant alias to return to userspace.  The ministacks are mapped
+ * readonly, so if the IRET fault we promote #GP to #DF which is an IST
+ * vector and thus has its own stack; we then do the fixup in the #DF
+ * handler.
+ *
+ * This file sets up the ministacks and the related page tables.  The
+ * actual ministack invocation is in entry_64.S.
+ */
+
+#include <linux/init.h>
+#include <linux/init_task.h>
+#include <linux/kernel.h>
+#include <linux/percpu.h>
+#include <linux/gfp.h>
+#include <linux/random.h>
+#include <asm/pgtable.h>
+#include <asm/pgalloc.h>
+#include <asm/setup.h>
+
+/*
+ * Note: we only need 6*8 = 48 bytes for the espfix stack, but round
+ * it up to a cache line to avoid unnecessary sharing.
+ */
+#define ESPFIX_STACK_SIZE	(8*8UL)
+#define ESPFIX_STACKS_PER_PAGE	(PAGE_SIZE/ESPFIX_STACK_SIZE)
+
+/* There is address space for how many espfix pages? */
+#define ESPFIX_PAGE_SPACE	(1UL << (PGDIR_SHIFT-PAGE_SHIFT-16))
+
+#define ESPFIX_MAX_CPUS		(ESPFIX_STACKS_PER_PAGE * ESPFIX_PAGE_SPACE)
+#if CONFIG_NR_CPUS > ESPFIX_MAX_CPUS
+# error "Need more than one PGD for the ESPFIX hack"
+#endif
+
+#define PGALLOC_GFP (GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO)
+
+/* This contains the *bottom* address of the espfix stack */
+DEFINE_PER_CPU_READ_MOSTLY(unsigned long, espfix_stack);
+DEFINE_PER_CPU_READ_MOSTLY(unsigned long, espfix_waddr);
+
+/* Initialization mutex - should this be a spinlock? */
+static DEFINE_MUTEX(espfix_init_mutex);
+
+/* Page allocation bitmap - each page serves ESPFIX_STACKS_PER_PAGE CPUs */
+#define ESPFIX_MAX_PAGES  DIV_ROUND_UP(CONFIG_NR_CPUS, ESPFIX_STACKS_PER_PAGE)
+static void *espfix_pages[ESPFIX_MAX_PAGES];
+
+static __page_aligned_bss pud_t espfix_pud_page[PTRS_PER_PUD]
+	__aligned(PAGE_SIZE);
+
+static unsigned int page_random, slot_random;
+
+/*
+ * This returns the bottom address of the espfix stack for a specific CPU.
+ * The math allows for a non-power-of-two ESPFIX_STACK_SIZE, in which case
+ * we have to account for some amount of padding at the end of each page.
+ */
+static inline unsigned long espfix_base_addr(unsigned int cpu)
+{
+	unsigned long page, slot;
+	unsigned long addr;
+
+	page = (cpu / ESPFIX_STACKS_PER_PAGE) ^ page_random;
+	slot = (cpu + slot_random) % ESPFIX_STACKS_PER_PAGE;
+	addr = (page << PAGE_SHIFT) + (slot * ESPFIX_STACK_SIZE);
+	addr = (addr & 0xffffUL) | ((addr & ~0xffffUL) << 16);
+	addr += ESPFIX_BASE_ADDR;
+	return addr;
+}
+
+#define PTE_STRIDE        (65536/PAGE_SIZE)
+#define ESPFIX_PTE_CLONES (PTRS_PER_PTE/PTE_STRIDE)
+#define ESPFIX_PMD_CLONES PTRS_PER_PMD
+#define ESPFIX_PUD_CLONES (65536/(ESPFIX_PTE_CLONES*ESPFIX_PMD_CLONES))
+
+#define PGTABLE_PROT	  ((_KERNPG_TABLE & ~_PAGE_RW) | _PAGE_NX)
+
+static void init_espfix_random(void)
+{
+	unsigned long rand;
+
+	/*
+	 * This is run before the entropy pools are initialized,
+	 * but this is hopefully better than nothing.
+	 */
+	if (!arch_get_random_long(&rand)) {
+		/* The constant is an arbitrary large prime */
+		rdtscll(rand);
+		rand *= 0xc345c6b72fd16123UL;
+	}
+
+	slot_random = rand % ESPFIX_STACKS_PER_PAGE;
+	page_random = (rand / ESPFIX_STACKS_PER_PAGE)
+		& (ESPFIX_PAGE_SPACE - 1);
+}
+
+void __init init_espfix_bsp(void)
+{
+	pgd_t *pgd_p;
+	pteval_t ptemask;
+
+	ptemask = __supported_pte_mask;
+
+	/* 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);
+
+	/* Randomize the locations */
+	init_espfix_random();
+
+	/* The rest is the same as for any other processor */
+	init_espfix_ap();
+}
+
+void init_espfix_ap(void)
+{
+	unsigned int cpu, page;
+	unsigned long addr;
+	pud_t pud, *pud_p;
+	pmd_t pmd, *pmd_p;
+	pte_t pte, *pte_p;
+	int n;
+	void *stack_page;
+	pteval_t ptemask;
+
+	/* We only have to do this once... */
+	if (likely(this_cpu_read(espfix_stack)))
+		return;		/* Already initialized */
+
+	cpu = smp_processor_id();
+	addr = espfix_base_addr(cpu);
+	page = cpu/ESPFIX_STACKS_PER_PAGE;
+
+	/* Did another CPU already set this up? */
+	stack_page = ACCESS_ONCE(espfix_pages[page]);
+	if (likely(stack_page))
+		goto done;
+
+	mutex_lock(&espfix_init_mutex);
+
+	/* Did we race on the lock? */
+	stack_page = ACCESS_ONCE(espfix_pages[page]);
+	if (stack_page)
+		goto unlock_done;
+
+	ptemask = __supported_pte_mask;
+
+	pud_p = &espfix_pud_page[pud_index(addr)];
+	pud = *pud_p;
+	if (!pud_present(pud)) {
+		pmd_p = (pmd_t *)__get_free_page(PGALLOC_GFP);
+		pud = __pud(__pa(pmd_p) | (PGTABLE_PROT & ptemask));
+		paravirt_alloc_pud(&init_mm, __pa(pmd_p) >> PAGE_SHIFT);
+		for (n = 0; n < ESPFIX_PUD_CLONES; n++)
+			set_pud(&pud_p[n], pud);
+	}
+
+	pmd_p = pmd_offset(&pud, addr);
+	pmd = *pmd_p;
+	if (!pmd_present(pmd)) {
+		pte_p = (pte_t *)__get_free_page(PGALLOC_GFP);
+		pmd = __pmd(__pa(pte_p) | (PGTABLE_PROT & ptemask));
+		paravirt_alloc_pmd(&init_mm, __pa(pte_p) >> PAGE_SHIFT);
+		for (n = 0; n < ESPFIX_PMD_CLONES; n++)
+			set_pmd(&pmd_p[n], pmd);
+	}
+
+	pte_p = pte_offset_kernel(&pmd, addr);
+	stack_page = (void *)__get_free_page(GFP_KERNEL);
+	pte = __pte(__pa(stack_page) | (__PAGE_KERNEL_RO & ptemask));
+	paravirt_alloc_pte(&init_mm, __pa(stack_page) >> PAGE_SHIFT);
+	for (n = 0; n < ESPFIX_PTE_CLONES; n++)
+		set_pte(&pte_p[n*PTE_STRIDE], pte);
+
+	/* Job is done for this CPU and any CPU which shares this page */
+	ACCESS_ONCE(espfix_pages[page]) = stack_page;
+
+unlock_done:
+	mutex_unlock(&espfix_init_mutex);
+done:
+	this_cpu_write(espfix_stack, addr);
+	this_cpu_write(espfix_waddr, (unsigned long)stack_page
+		       + (addr & ~PAGE_MASK));
+}

arch/x86/kernel/ldt.c

···
 		}
 	}
 
-	/*
-	 * On x86-64 we do not support 16-bit segments due to
-	 * IRET leaking the high bits of the kernel stack address.
-	 */
-#ifdef CONFIG_X86_64
-	if (!ldt_info.seg_32bit) {
-		error = -EINVAL;
-		goto out_unlock;
-	}
-#endif
-
 	fill_ldt(&ldt, &ldt_info);
 	if (oldmode)
 		ldt.avl = 0;

arch/x86/kernel/smpboot.c

···
 	check_tsc_sync_target();
 
 	/*
+	 * Enable the espfix hack for this CPU
+	 */
+#ifdef CONFIG_X86_64
+	init_espfix_ap();
+#endif
+
+	/*
 	 * We need to hold vector_lock so there the set of online cpus
 	 * does not change while we are assigning vectors to cpus.  Holding
 	 * this lock ensures we don't half assign or remove an irq from a cpu.

arch/x86/mm/dump_pagetables.c

···
 	unsigned long start_address;
 	unsigned long current_address;
 	const struct addr_marker *marker;
+	unsigned long lines;
 	bool to_dmesg;
 };
 
 struct addr_marker {
 	unsigned long start_address;
 	const char *name;
+	unsigned long max_lines;
 };
 
 /* indices for address_markers; keep sync'd w/ address_markers below */
···
 	LOW_KERNEL_NR,
 	VMALLOC_START_NR,
 	VMEMMAP_START_NR,
+	ESPFIX_START_NR,
 	HIGH_KERNEL_NR,
 	MODULES_VADDR_NR,
 	MODULES_END_NR,
···
 	{ PAGE_OFFSET,		"Low Kernel Mapping" },
 	{ VMALLOC_START,        "vmalloc() Area" },
 	{ VMEMMAP_START,        "Vmemmap" },
+	{ ESPFIX_BASE_ADDR,	"ESPfix Area", 16 },
 	{ __START_KERNEL_map,   "High Kernel Mapping" },
 	{ MODULES_VADDR,        "Modules" },
 	{ MODULES_END,          "End Modules" },
···
 		      pgprot_t new_prot, int level)
 {
 	pgprotval_t prot, cur;
-	static const char units[] = "KMGTPE";
+	static const char units[] = "BKMGTPE";
 
 	/*
 	 * If we have a "break" in the series, we need to flush the state that
···
 		st->current_prot = new_prot;
 		st->level = level;
 		st->marker = address_markers;
+		st->lines = 0;
 		pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
 				   st->marker->name);
 	} else if (prot != cur || level != st->level ||
···
 		/*
 		 * Now print the actual finished series
 		 */
-		pt_dump_seq_printf(m, st->to_dmesg,  "0x%0*lx-0x%0*lx   ",
-				   width, st->start_address,
-				   width, st->current_address);
+		if (!st->marker->max_lines ||
+		    st->lines < st->marker->max_lines) {
+			pt_dump_seq_printf(m, st->to_dmesg,
+					   "0x%0*lx-0x%0*lx   ",
+					   width, st->start_address,
+					   width, st->current_address);
 
-		delta = (st->current_address - st->start_address) >> 10;
-		while (!(delta & 1023) && unit[1]) {
-			delta >>= 10;
-			unit++;
+			delta = st->current_address - st->start_address;
+			while (!(delta & 1023) && unit[1]) {
+				delta >>= 10;
+				unit++;
+			}
+			pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
+					    delta, *unit);
+			printk_prot(m, st->current_prot, st->level,
+				    st->to_dmesg);
 		}
-		pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ", delta, *unit);
-		printk_prot(m, st->current_prot, st->level, st->to_dmesg);
+		st->lines++;
 
 		/*
 		 * We print markers for special areas of address space,
···
 		 * This helps in the interpretation.
 		 */
 		if (st->current_address >= st->marker[1].start_address) {
+			if (st->marker->max_lines &&
+			    st->lines > st->marker->max_lines) {
+				unsigned long nskip =
+					st->lines - st->marker->max_lines;
+				pt_dump_seq_printf(m, st->to_dmesg,
+						   "... %lu entr%s skipped ... \n",
+						   nskip,
+						   nskip == 1 ? "y" : "ies");
+			}
 			st->marker++;
+			st->lines = 0;
 			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
 					   st->marker->name);
 		}

init/main.c

···
 	if (efi_enabled(EFI_RUNTIME_SERVICES))
 		efi_enter_virtual_mode();
 #endif
+#ifdef CONFIG_X86_64
+	/* Should be run before the first non-init thread is created */
+	init_espfix_bsp();
+#endif
 	thread_info_cache_init();
 	cred_init();
 	fork_init(totalram_pages);