x86/sev: Mark the code returning to user space as syscall gap
Subsystems (3)
PATCH v2 1/1 x86/sev: Mark the code returning to user space as syscall gap
From: Lai Jiangshan <[email protected]>
When returning to user space, the %rsp is user controlled value.
If it is SNP-guest and the hypervisor decides to mess with the code-page for this path while a CPU is executing it. This will cause a #VC on that CPU and that could hit in the syscall return path and mislead the #VC handler.
So make ip_within_syscall_gap() return true in this case.
---
[V1]: https://lore.kernel.org/lkml/[email protected]/
Changed from V1:
Update changelog.
arch/x86/entry/entry_64.S | 2 ++ arch/x86/entry/entry_64_compat.S | 2 ++ arch/x86/include/asm/proto.h | 4 ++++ arch/x86/include/asm/ptrace.h | 4 ++++ 4 files changed, 12 insertions(+)
arch/x86/entry/entry_64.S +2 -0
@@ -215,8 +215,10 @@ syscall_return_via_sysret: popq %rdi popq %rsp +SYM_INNER_LABEL(entry_SYSRETQ_unsafe_stack, SYM_L_GLOBAL) swapgs sysretq +SYM_INNER_LABEL(entry_SYSRETQ_end, SYM_L_GLOBAL) SYM_CODE_END(entry_SYSCALL_64) /*
arch/x86/entry/entry_64_compat.S +2 -0
@@ -297,6 +297,7 @@ sysret32_from_system_call: * code. We zero R8-R10 to avoid info leaks. */ movq RSP-ORIG_RAX(%rsp), %rsp +SYM_INNER_LABEL(entry_SYSRETL_compat_unsafe_stack, SYM_L_GLOBAL) /* * The original userspace %rsp (RSP-ORIG_RAX(%rsp)) is stored
@@ -314,6 +315,7 @@ sysret32_from_system_call: xorl %r10d, %r10d swapgs sysretl +SYM_INNER_LABEL(entry_SYSRETL_compat_end, SYM_L_GLOBAL) SYM_CODE_END(entry_SYSCALL_compat) /*
arch/x86/include/asm/proto.h +4 -0
@@ -13,6 +13,8 @@ void syscall_init(void);#ifdef CONFIG_X86_64 void entry_SYSCALL_64 (void );void entry_SYSCALL_64_safe_stack (void ); +void entry_SYSRETQ_unsafe_stack (void ); +void entry_SYSRETQ_end (void );long do_arch_prctl_64 (struct task_struct *task,int option,unsigned long arg2);#endif
@@ -28,6 +30,8 @@ void entry_SYSENTER_compat(void);void __end_entry_SYSENTER_compat (void );void entry_SYSCALL_compat (void );void entry_SYSCALL_compat_safe_stack (void ); +void entry_SYSRETL_compat_unsafe_stack (void ); +void entry_SYSRETL_compat_end (void );void entry_INT80_compat (void );#ifdef CONFIG_XEN_PV void xen_entry_INT80_compat (void );
arch/x86/include/asm/ptrace.h +4 -0
@@ -186,9 +186,13 @@ static __always_inline bool ip_within_syscall_gap(struct pt_regs *regs)bool ret = (regs->ip >= (unsigned long )entry_SYSCALL_64 && regs->ip < (unsigned long )entry_SYSCALL_64_safe_stack); + ret = ret || (regs->ip >= (unsigned long )entry_SYSRETQ_unsafe_stack && + regs->ip < (unsigned long )entry_SYSRETQ_end);#ifdef CONFIG_IA32_EMULATION ret = ret || (regs->ip >= (unsigned long )entry_SYSCALL_compat && regs->ip < (unsigned long )entry_SYSCALL_compat_safe_stack); + ret = ret || (regs->ip >= (unsigned long )entry_SYSRETL_compat_unsafe_stack && + regs->ip < (unsigned long )entry_SYSRETL_compat_end);#endif return ret;
Signature · 1 line
2.19.1.6.gb485710b
Quoted 10 lines · Lai Jiangshan
From: Lai Jiangshan <[email protected]>
When returning to user space, the %rsp is user controlled value.
If it is SNP-guest and the hypervisor decides to mess with the code-page for this path while a CPU is executing it. This will cause a #VC on that CPU and that could hit in the syscall return path and mislead the #VC handler.
Quoted 4 lines
So make
ip_within_syscall_gap()return true in this case.Cc Joerg Roedel <[email protected]>Signed-off-by Lai Jiangshan <[email protected]>
Hello
Ping.
Thanks
Lai
Hi,
Quoted 10 lines · Lai Jiangshan
From: Lai Jiangshan <[email protected]>
When returning to user space, the %rsp is user controlled value.
If it is SNP-guest and the hypervisor decides to mess with the code-page for this path while a CPU is executing it. This will cause a #VC on that CPU and that could hit in the syscall return path and mislead the #VC handler.
So make
ip_within_syscall_gap()return true in this case.
With the SNP guest patches in tip-tree I think it actually becomes possible that a #VC exception hits in these parts of the execution stream. It requires good timing by the attacker, but it is not impossible. Therefore:
Signature · 10 lines
Jörg Rödel [email protected] SUSE Software Solutions Germany GmbH Maxfeldstr. 5 90409 Nürnberg Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Ivo Totev
The following commit has been merged into the x86/sev branch of tip:
The following commit has been merged into the x86/sev branch of tip:
Commit-ID: 47f33de4aafb2f5e43d480d590a939d0f1d566a9 Gitweb: https://git.kernel.org/tip/47f33de4aafb2f5e43d480d590a939d0f1d566a9 Author: Lai Jiangshan <[email protected]> AuthorDate: Tue, 12 Apr 2022 20:49:08 +08:00 Committer: Borislav Petkov <[email protected]> CommitterDate: Thu, 19 May 2022 10:56:46 +02:00
x86/sev: Mark the code returning to user space as syscall gap
When returning to user space, %rsp is user-controlled value.
If it is a SNP-guest and the hypervisor decides to mess with the code-page for this path while a CPU is executing it, a potential #VC could hit in the syscall return path and mislead the #VC handler.
So make ip_within_syscall_gap() return true in this case.
--- arch/x86/entry/entry_64.S | 2 ++ arch/x86/entry/entry_64_compat.S | 2 ++ arch/x86/include/asm/proto.h | 4 ++++ arch/x86/include/asm/ptrace.h | 4 ++++ 4 files changed, 12 insertions(+)
arch/x86/entry/entry_64.S +2 -0
@@ -215,8 +215,10 @@ syscall_return_via_sysret: popq %rdi popq %rsp +SYM_INNER_LABEL(entry_SYSRETQ_unsafe_stack, SYM_L_GLOBAL) swapgs sysretq +SYM_INNER_LABEL(entry_SYSRETQ_end, SYM_L_GLOBAL) SYM_CODE_END(entry_SYSCALL_64) /*
arch/x86/entry/entry_64_compat.S +2 -0
@@ -297,6 +297,7 @@ sysret32_from_system_call: * code. We zero R8-R10 to avoid info leaks. */ movq RSP-ORIG_RAX(%rsp), %rsp +SYM_INNER_LABEL(entry_SYSRETL_compat_unsafe_stack, SYM_L_GLOBAL) /* * The original userspace %rsp (RSP-ORIG_RAX(%rsp)) is stored
@@ -314,6 +315,7 @@ sysret32_from_system_call: xorl %r10d, %r10d swapgs sysretl +SYM_INNER_LABEL(entry_SYSRETL_compat_end, SYM_L_GLOBAL) SYM_CODE_END(entry_SYSCALL_compat) /*
arch/x86/include/asm/proto.h +4 -0
@@ -13,6 +13,8 @@ void syscall_init(void);#ifdef CONFIG_X86_64 void entry_SYSCALL_64 (void );void entry_SYSCALL_64_safe_stack (void ); +void entry_SYSRETQ_unsafe_stack (void ); +void entry_SYSRETQ_end (void );long do_arch_prctl_64 (struct task_struct *task,int option,unsigned long arg2);#endif
@@ -28,6 +30,8 @@ void entry_SYSENTER_compat(void);void __end_entry_SYSENTER_compat (void );void entry_SYSCALL_compat (void );void entry_SYSCALL_compat_safe_stack (void ); +void entry_SYSRETL_compat_unsafe_stack (void ); +void entry_SYSRETL_compat_end (void );void entry_INT80_compat (void );#ifdef CONFIG_XEN_PV void xen_entry_INT80_compat (void );
arch/x86/include/asm/ptrace.h +4 -0
@@ -186,9 +186,13 @@ static __always_inline bool ip_within_syscall_gap(struct pt_regs *regs)bool ret = (regs->ip >= (unsigned long )entry_SYSCALL_64 && regs->ip < (unsigned long )entry_SYSCALL_64_safe_stack); + ret = ret || (regs->ip >= (unsigned long )entry_SYSRETQ_unsafe_stack && + regs->ip < (unsigned long )entry_SYSRETQ_end);#ifdef CONFIG_IA32_EMULATION ret = ret || (regs->ip >= (unsigned long )entry_SYSCALL_compat && regs->ip < (unsigned long )entry_SYSCALL_compat_safe_stack); + ret = ret || (regs->ip >= (unsigned long )entry_SYSRETL_compat_unsafe_stack && + regs->ip < (unsigned long )entry_SYSRETL_compat_end);#endif return ret;