:p
atchew
Login
One might think of this as follow-on to XSA-451, but that's not quite the right order of events. There are a few open aspects; see individual patches. 1: remove redundant XEN_SHSTK check from reinit_bsp_stack() 2: record SSP at non-guest entry points 3: traps: use entry_ssp in fixup_exception_return() 4: prefer shadow stack for producing call traces Jan
As of 72d51813d631 ("x86: amend cpu_has_xen_{ibt,shstk}") this has been integrated into cpu_has_xen_shstk. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- a/xen/arch/x86/setup.c +++ b/xen/arch/x86/setup.c @@ -XXX,XX +XXX,XX @@ static void __init noreturn reinit_bsp_s if ( rc ) panic("Error %d setting up PV root page table\n", rc); - if ( IS_ENABLED(CONFIG_XEN_SHSTK) && cpu_has_xen_shstk ) + if ( cpu_has_xen_shstk ) { wrmsrl(MSR_PL0_SSP, (unsigned long)stack + (PRIMARY_SHSTK_SLOT + 1) * PAGE_SIZE - 8);
We will want to use that value for call trace generation, and likely also to eliminate the somewhat fragile shadow stack searching done in fixup_exception_return(). For those purposes, guest-only entry points do not need to record that value. To keep the saving code simple, record our own SSP that corresponds to an exception frame, pointing to the top of the shadow stack counterpart of what the CPU has saved on the regular stack. Consuming code can then work its way from there. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- To record the full 64-bit value, some of the unused bits in the %cs slot could be used. Sadly that slot has saved_upcall_mask in an unhelpful location, otherwise simply storing low and high 32 bits in those two separate half-slots would be a pretty obvious choice. As long as "entry_ssp" is used in non-guest-entry frames only, we could of course put half of it into a union with saved_upcall_mask ... Else may want to put a BUILD_BUG_ON(VADDR_BITS > 48) somewhere, but I'm afraid I can't really identify a good place for such to live. Leveraging that the CPU stores zero in the upper bits of the selector register slots, the save sequence could also be shl $16, %rcx or %rcx, UREGS_entry_ssp-2(%rsp) That's shorter and avoids a 16-bit operation, but may be less desirable, for being a read-modify-write access. --- a/xen/arch/x86/domctl.c +++ b/xen/arch/x86/domctl.c @@ -XXX,XX +XXX,XX @@ void arch_get_info_guest(struct vcpu *v, if ( !compat ) { memcpy(&c.nat->user_regs, &v->arch.user_regs, sizeof(c.nat->user_regs)); + c.nat->user_regs.entry_ssp = 0; if ( is_pv_domain(d) ) memcpy(c.nat->trap_ctxt, v->arch.pv.trap_ctxt, sizeof(c.nat->trap_ctxt)); --- a/xen/arch/x86/hvm/svm/entry.S +++ b/xen/arch/x86/hvm/svm/entry.S @@ -XXX,XX +XXX,XX @@ __UNLIKELY_END(nsvm_hap) sti vmrun - SAVE_ALL + SAVE_ALL ssp=0 GET_CURRENT(bx) --- a/xen/arch/x86/hvm/vmx/entry.S +++ b/xen/arch/x86/hvm/vmx/entry.S @@ -XXX,XX +XXX,XX @@ #define VMLAUNCH .byte 0x0f,0x01,0xc2 ENTRY(vmx_asm_vmexit_handler) - SAVE_ALL + SAVE_ALL ssp=0 mov %cr2,%rax GET_CURRENT(bx) @@ -XXX,XX +XXX,XX @@ UNLIKELY_END(realmode) .Lvmx_vmentry_fail: sti - SAVE_ALL + SAVE_ALL ssp=0 /* * SPEC_CTRL_ENTRY notes --- a/xen/arch/x86/include/asm/asm_defns.h +++ b/xen/arch/x86/include/asm/asm_defns.h @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) #endif #ifdef __ASSEMBLY__ -.macro SAVE_ALL compat=0 +.macro SAVE_ALL compat=0 ssp=IS_ENABLED(CONFIG_XEN_SHSTK) addq $-(UREGS_error_code-UREGS_r15), %rsp cld movq %rdi,UREGS_rdi(%rsp) @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) movq %rax,UREGS_rax(%rsp) xor %eax, %eax .if !\compat +.if \ssp + rdsspq %rcx +.endif movq %r8,UREGS_r8(%rsp) movq %r9,UREGS_r9(%rsp) movq %r10,UREGS_r10(%rsp) @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) xor %r13d, %r13d xor %r14d, %r14d xor %r15d, %r15d +.if \ssp && !\compat + mov %cx, UREGS_entry_ssp(%rsp) + shr $16, %rcx + mov %ecx, UREGS_entry_ssp+2(%rsp) +.endif .endm #define LOAD_ONE_REG(reg, compat) \ --- a/xen/arch/x86/x86_64/asm-offsets.c +++ b/xen/arch/x86/x86_64/asm-offsets.c @@ -XXX,XX +XXX,XX @@ void __dummy__(void) OFFSET(UREGS_eflags, struct cpu_user_regs, rflags); OFFSET(UREGS_rsp, struct cpu_user_regs, rsp); OFFSET(UREGS_ss, struct cpu_user_regs, ss); + OFFSET(UREGS_entry_ssp, struct cpu_user_regs, entry_ssp); OFFSET(UREGS_kernel_sizeof, struct cpu_user_regs, es); BLANK(); --- a/xen/arch/x86/x86_64/entry.S +++ b/xen/arch/x86/x86_64/entry.S @@ -XXX,XX +XXX,XX @@ FUNC(lstar_enter) pushq $0 BUILD_BUG_ON(TRAP_syscall & 0xff) movb $TRAP_syscall >> 8, EFRAME_entry_vector + 1(%rsp) - SAVE_ALL + SAVE_ALL ssp=0 SPEC_CTRL_ENTRY_FROM_PV /* Req: %rsp=regs/cpuinfo, %rdx=0, Clob: acd */ /* WARNING! `ret`, `call *`, `jmp *` not safe before this point. */ @@ -XXX,XX +XXX,XX @@ FUNC(cstar_enter) pushq $0 BUILD_BUG_ON(TRAP_syscall & 0xff) movb $TRAP_syscall >> 8, EFRAME_entry_vector + 1(%rsp) - SAVE_ALL + SAVE_ALL ssp=0 SPEC_CTRL_ENTRY_FROM_PV /* Req: %rsp=regs/cpuinfo, %rdx=0, Clob: acd */ /* WARNING! `ret`, `call *`, `jmp *` not safe before this point. */ @@ -XXX,XX +XXX,XX @@ LABEL(sysenter_eflags_saved, 0) pushq $0 BUILD_BUG_ON(TRAP_syscall & 0xff) movb $TRAP_syscall >> 8, EFRAME_entry_vector + 1(%rsp) - SAVE_ALL + SAVE_ALL ssp=0 SPEC_CTRL_ENTRY_FROM_PV /* Req: %rsp=regs/cpuinfo, %rdx=0, Clob: acd */ /* WARNING! `ret`, `call *`, `jmp *` not safe before this point. */ @@ -XXX,XX +XXX,XX @@ FUNC(entry_int80) ALTERNATIVE "", clac, X86_FEATURE_XEN_SMAP pushq $0 movb $0x80, EFRAME_entry_vector(%rsp) - SAVE_ALL + SAVE_ALL ssp=0 SPEC_CTRL_ENTRY_FROM_PV /* Req: %rsp=regs/cpuinfo, %rdx=0, Clob: acd */ /* WARNING! `ret`, `call *`, `jmp *` not safe before this point. */ --- a/xen/include/public/arch-x86/xen-x86_64.h +++ b/xen/include/public/arch-x86/xen-x86_64.h @@ -XXX,XX +XXX,XX @@ struct cpu_user_regs { uint8_t _pad1[3]; __DECL_REG_LO16(flags); /* rflags.IF == !saved_upcall_mask */ __DECL_REG_LO8(sp); - uint16_t ss, _pad2[3]; + uint16_t ss; +#if !defined(__XEN__) + uint16_t _pad2[3]; +#elif defined(COMPILE_OFFSETS) + uint16_t entry_ssp[3]; +#else + /* + * This points _at_ the corresponding shadow stack frame; it is _not_ the + * outer context's SSP. That, if the outer context has CET-SS enabled, + * is stored in the top slot of the pointed to shadow stack frame. + */ + signed long entry_ssp:48; +#endif uint16_t es, _pad3[3]; uint16_t ds, _pad4[3]; uint16_t fs, _pad5[3];
With the value recorded on entry there's no need anymore to go hunt for the respective exception frame on the shadow stack. By deriving "ptr" from that field (without any offset), it then ends up pointin one slot lower than before. Therefore all array indexes need incrementing, nicely doing away with all the negative ones. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- Indentation of the prior inner (but not innermost) if()'s body is deliberately left untouched, to aid review. It'll be adjusted in a separate follow-on patch. --- a/xen/arch/x86/traps.c +++ b/xen/arch/x86/traps.c @@ -XXX,XX +XXX,XX @@ unsigned long get_stack_trace_bottom(uns } } -static unsigned long get_shstk_bottom(unsigned long sp) -{ - switch ( get_stack_page(sp) ) - { -#ifdef CONFIG_XEN_SHSTK - case 0: return ROUNDUP(sp, IST_SHSTK_SIZE) - sizeof(unsigned long); - case 5: return ROUNDUP(sp, PAGE_SIZE) - sizeof(unsigned long); -#endif - default: return sp - sizeof(unsigned long); - } -} - unsigned long get_stack_dump_bottom(unsigned long sp) { switch ( get_stack_page(sp) ) @@ -XXX,XX +XXX,XX @@ static void fixup_exception_return(struc { if ( IS_ENABLED(CONFIG_XEN_SHSTK) ) { - unsigned long ssp, *ptr, *base; + unsigned long ssp = rdssp(); - if ( (ssp = rdssp()) == SSP_NO_SHSTK ) - goto shstk_done; + if ( ssp != SSP_NO_SHSTK ) + { + unsigned long *ptr = _p(regs->entry_ssp); + unsigned long primary_shstk = + (ssp & ~(STACK_SIZE - 1)) + + (PRIMARY_SHSTK_SLOT + 1) * PAGE_SIZE - 8; - ptr = _p(ssp); - base = _p(get_shstk_bottom(ssp)); + BUG_ON((regs->entry_ssp ^ primary_shstk) >> PAGE_SHIFT); - for ( ; ptr < base; ++ptr ) - { /* - * Search for %rip. The shstk currently looks like this: + * The shstk currently looks like this: * * tok [Supervisor token, == &tok | BUSY, only with FRED inactive] * ... [Pointed to by SSP for most exceptions, empty in IST cases] * %cs [== regs->cs] * %rip [== regs->rip] - * SSP [Likely points to 3 slots higher, above %cs] + * SSP [Pointed to by entry_ssp; Likely points to 3 slots + * higher, above %cs] * ... [call tree to this function, likely 2/3 slots] * * and we want to overwrite %rip with fixup. There are two @@ -XXX,XX +XXX,XX @@ static void fixup_exception_return(struc * * Check for both regs->rip and regs->cs matching. */ - if ( ptr[0] == regs->rip && ptr[1] == regs->cs ) - { - unsigned long primary_shstk = - (ssp & ~(STACK_SIZE - 1)) + - (PRIMARY_SHSTK_SLOT + 1) * PAGE_SIZE - 8; + BUG_ON(ptr[1] != regs->rip || ptr[2] != regs->cs); - wrss(fixup, ptr); + { + wrss(fixup, &ptr[1]); if ( !stub_ra ) goto shstk_done; @@ -XXX,XX +XXX,XX @@ static void fixup_exception_return(struc * - if we're on an IST stack, we need to increment the * original SSP. */ - BUG_ON((ptr[-1] ^ primary_shstk) >> PAGE_SHIFT); + BUG_ON((ptr[0] ^ primary_shstk) >> PAGE_SHIFT); if ( (ssp ^ primary_shstk) >> PAGE_SHIFT ) { @@ -XXX,XX +XXX,XX @@ static void fixup_exception_return(struc * addresses actually match. Then increment the interrupted * context's SSP. */ - BUG_ON(stub_ra != *(unsigned long*)ptr[-1]); - wrss(ptr[-1] + 8, &ptr[-1]); + BUG_ON(stub_ra != *(unsigned long*)ptr[0]); + wrss(ptr[0] + 8, &ptr[0]); goto shstk_done; } /* Make sure the two return addresses actually match. */ - BUG_ON(stub_ra != ptr[2]); + BUG_ON(stub_ra != ptr[3]); /* Move exception frame, updating SSP there. */ - wrss(ptr[1], &ptr[2]); /* %cs */ - wrss(ptr[0], &ptr[1]); /* %rip */ - wrss(ptr[-1] + 8, &ptr[0]); /* SSP */ + wrss(ptr[2], &ptr[3]); /* %cs */ + wrss(ptr[1], &ptr[2]); /* %rip */ + wrss(ptr[0] + 8, &ptr[1]); /* SSP */ /* Move all newer entries. */ - while ( --ptr != _p(ssp) ) - wrss(ptr[-1], &ptr[0]); + while ( ptr-- != _p(ssp) ) + wrss(ptr[0], &ptr[1]); /* Finally account for our own stack having shifted up. */ asm volatile ( "incsspd %0" :: "r" (2) ); - - goto shstk_done; } } - - /* - * We failed to locate and fix up the shadow IRET frame. This could - * be due to shadow stack corruption, or bad logic above. We cannot - * continue executing the interrupted context. - */ - BUG(); - } shstk_done:
Shadow stacks contain little more than return addresses, and they in particular allow precise call traces also without FRAME_POINTER. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- While the 'E' for exception frames is probably okay, I'm not overly happy with the 'C' (for CET). I would have preferred 'S' (for shadow), but we use that character already. As an alternative to suppressing output for the top level exception frame, adding the new code ahead of the 'R' output line (and then also ahead of the stack top read) could be considered. Perhaps having a printk() for the PV entry case is meaningless, for - no frame being pushed when entered from CPL=3 (64-bit PV), - no entry possible from CPL<3 (32-bit PV disabled when CET is active)? In which case the comment probably should just be "Bogus." and the code merely be "break;". Quite likely a number of other uses of is_active_kernel_text() also want amending with in_stub(). --- a/xen/arch/x86/traps.c +++ b/xen/arch/x86/traps.c @@ -XXX,XX +XXX,XX @@ unsigned long get_stack_dump_bottom(unsi } } +static bool in_stub(unsigned long addr) +{ + return !((this_cpu(stubs.addr) ^ addr) >> STUB_BUF_SHIFT); +} + #if !defined(CONFIG_FRAME_POINTER) /* @@ -XXX,XX +XXX,XX @@ static void show_trace(const struct cpu_ !is_active_kernel_text(tos) ) printk(" [<%p>] R %pS\n", _p(regs->rip), _p(regs->rip)); + if ( IS_ENABLED(CONFIG_XEN_SHSTK) && rdssp() != SSP_NO_SHSTK ) + { + const unsigned long *ptr = _p(regs->entry_ssp); + unsigned int n; + + for ( n = 0; (unsigned long)ptr & (PAGE_SIZE - sizeof(*ptr)); ++n ) + { + unsigned long val = *ptr; + + if ( is_active_kernel_text(val) || in_stub(val) ) + { + /* Normal return address entry. */ + printk(" [<%p>] C %pS\n", _p(val), _p(val)); + ++ptr; + } + else if ( !((val ^ *ptr) >> (PAGE_SHIFT + STACK_ORDER)) ) + { + if ( val & (sizeof(val) - 1) ) + { + /* Most likely a supervisor token. */ + break; + } + + /* + * Ought to be a hypervisor interruption frame. But don't + * (re)log the current frame's %rip. + */ + if ( n || ptr[1] != regs->rip ) + printk(" [<%p>] E %pS\n", _p(ptr[1]), _p(ptr[1])); + ptr = _p(val); + } + else + { + /* Ought to be a PV guest hypercall/interruption frame. */ + printk(" %04lx:[<%p>] E\n", ptr[2], _p(ptr[1])); + ptr = 0; + } + } + + /* Fall back to legacy stack trace if nothing was logged at all. */ + if ( n ) + return; + } + if ( fault ) { printk(" [Fault on access]\n");
One might think of this as follow-on to XSA-451, but that's not quite the right order of events. There are a few open aspects; see in particular the final patch. v3 addresses review feedback, which includes one new (prereq) patch. See individual patches for details. 1: HVM: don't (almost) open-code POP_GPRS 2: record SSP at non-guest entry points 3: traps: use entry_ssp in fixup_exception_return() 4: prefer shadow stack for producing call traces Jan
It can be used as-is for VMX. For SVM the special treatment of %rax needs accounting for. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- This is the minimum required as a prereq to the next patch (to avoid the need for custom adjustments in the two places). We could go further and switch to using PUSH_AND_CLEAR_GPRS at the same time. --- v3: New. --- a/xen/arch/x86/hvm/svm/entry.S +++ b/xen/arch/x86/hvm/svm/entry.S @@ -XXX,XX +XXX,XX @@ __UNLIKELY_END(nsvm_hap) */ sti - pop %r15 - pop %r14 - pop %r13 - pop %r12 - pop %rbp mov VCPU_svm_vmcb_pa(%rbx),%rax - pop %rbx - pop %r11 - pop %r10 - pop %r9 - pop %r8 - pop %rcx /* Skip %rax: restored by VMRUN. */ - pop %rcx - pop %rdx - pop %rsi - pop %rdi + POP_GPRS skip_rax=1 /* %rax restored by VMRUN. */ SPEC_CTRL_COND_VERW /* Req: %rsp=eframe Clob: efl */ --- a/xen/arch/x86/hvm/vmx/entry.S +++ b/xen/arch/x86/hvm/vmx/entry.S @@ -XXX,XX +XXX,XX @@ UNLIKELY_END(realmode) and $SCF_verw, %eax or %eax, %ecx - pop %r15 - pop %r14 - pop %r13 - pop %r12 - pop %rbp - pop %rbx - pop %r11 - pop %r10 - pop %r9 - pop %r8 - pop %rax - pop %rcx - pop %rdx - pop %rsi - pop %rdi + POP_GPRS /* Preserves flags. */ jpe .L_skip_verw /* VERW clobbers ZF, but preserves all others, including SF. */ --- a/xen/arch/x86/include/asm/asm_defns.h +++ b/xen/arch/x86/include/asm/asm_defns.h @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) /* * POP GPRs from a UREGS_* frame on the stack. Does not modify flags. */ -.macro POP_GPRS +.macro POP_GPRS skip_rax=0 pop %r15 pop %r14 pop %r13 @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) pop %r10 pop %r9 pop %r8 + .if \skip_rax + pop %rcx + .else pop %rax + .endif pop %rcx pop %rdx pop %rsi
We will want to use that value for call trace generation, and likely also to eliminate the somewhat fragile shadow stack searching done in fixup_exception_return(). For those purposes, guest-only entry points do not need to record that value. To keep the saving code simple, record our own SSP that corresponds to an exception frame, pointing to the top of the shadow stack counterpart of what the CPU has saved on the regular stack. Consuming code can then work its way from there. In SAVE_ALL / RESTORE_ALL simply drop the use of UREGS_r15. We want the full size in all cases, so what exactly the top-of-stack field is going to be (whose UREGS_* is 0 anyway) doesn't matter this much there. This way we don't need to distinguish between XEN_SHSTK=y and XEN_SHSTK=n. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- v3: Put new field at the front of struct cpu_user_regs. v2: Add comment ahead of SAVE_ALL. Add comma between its parameters. Re-base. --- a/xen/arch/x86/hvm/svm/entry.S +++ b/xen/arch/x86/hvm/svm/entry.S @@ -XXX,XX +XXX,XX @@ __UNLIKELY_END(nsvm_hap) vmrun - SAVE_ALL + SAVE_ALL ssp=0 GET_CURRENT(bx) --- a/xen/arch/x86/hvm/vmx/entry.S +++ b/xen/arch/x86/hvm/vmx/entry.S @@ -XXX,XX +XXX,XX @@ #include <asm/page.h> FUNC(vmx_asm_vmexit_handler) - SAVE_ALL + SAVE_ALL ssp=0 mov %cr2,%rax GET_CURRENT(bx) @@ -XXX,XX +XXX,XX @@ UNLIKELY_END(realmode) .Lvmx_vmentry_fail: sti - SAVE_ALL + SAVE_ALL ssp=0 /* * SPEC_CTRL_ENTRY notes --- a/xen/arch/x86/include/asm/asm_defns.h +++ b/xen/arch/x86/include/asm/asm_defns.h @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) #endif #ifdef __ASSEMBLER__ -.macro SAVE_ALL compat=0 - addq $-(UREGS_error_code-UREGS_r15), %rsp +/* + * Use sites may override ssp to 0. It should never be overridden to 1. + * NB: compat=1 implies ssp=0. + */ +.macro SAVE_ALL compat=0, ssp=IS_ENABLED(CONFIG_XEN_SHSTK) + addq $-UREGS_error_code, %rsp cld movq %rdi,UREGS_rdi(%rsp) xor %edi, %edi @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) movq %rax,UREGS_rax(%rsp) xor %eax, %eax .if !\compat +.if \ssp + rdsspq %rcx +.endif movq %r8,UREGS_r8(%rsp) movq %r9,UREGS_r9(%rsp) movq %r10,UREGS_r10(%rsp) @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) xor %r13d, %r13d xor %r14d, %r14d xor %r15d, %r15d +#ifdef CONFIG_XEN_SHSTK + mov %rcx, UREGS_entry_ssp(%rsp) +#endif .endm #define LOAD_ONE_REG(reg, compat) \ @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) LOAD_ONE_REG(dx, \compat) LOAD_ONE_REG(si, \compat) LOAD_ONE_REG(di, \compat) - subq $-(UREGS_error_code-UREGS_r15+\adj), %rsp + subq $-(UREGS_error_code + \adj), %rsp .endm /* - * Push and clear GPRs + * Push and clear GPRs. + * + * Use sites may override ssp to 0. It should never be overridden to 1. */ -.macro PUSH_AND_CLEAR_GPRS +.macro PUSH_AND_CLEAR_GPRS ssp=IS_ENABLED(CONFIG_XEN_SHSTK) push %rdi xor %edi, %edi push %rsi @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) xor %ecx, %ecx push %rax xor %eax, %eax + .if \ssp + rdsspq %rcx + .endif push %r8 xor %r8d, %r8d push %r9 @@ -XXX,XX +XXX,XX @@ static always_inline void stac(void) xor %r14d, %r14d push %r15 xor %r15d, %r15d +#ifdef CONFIG_XEN_SHSTK + push %rcx +#endif .endm /* * POP GPRs from a UREGS_* frame on the stack. Does not modify flags. */ .macro POP_GPRS skip_rax=0 +#ifdef CONFIG_XEN_SHSTK + pop %rcx +#endif pop %r15 pop %r14 pop %r13 --- a/xen/arch/x86/include/asm/cpu-user-regs.h +++ b/xen/arch/x86/include/asm/cpu-user-regs.h @@ -XXX,XX +XXX,XX @@ */ struct cpu_user_regs { +#ifdef CONFIG_XEN_SHSTK + /* + * This points _at_ the corresponding shadow stack frame; it is _not_ the + * outer context's SSP. That, if the outer context has CET-SS enabled, + * is stored in the top slot of the pointed to shadow stack. + */ + uint64_t entry_ssp; +#endif + union { uint64_t r15; uint32_t r15d; uint16_t r15w; uint8_t r15b; }; union { uint64_t r14; uint32_t r14d; uint16_t r14w; uint8_t r14b; }; union { uint64_t r13; uint32_t r13d; uint16_t r13w; uint8_t r13b; }; --- a/xen/arch/x86/x86_64/asm-offsets.c +++ b/xen/arch/x86/x86_64/asm-offsets.c @@ -XXX,XX +XXX,XX @@ void __dummy__(void) OFFSET(UREGS_eflags, struct cpu_user_regs, rflags); OFFSET(UREGS_rsp, struct cpu_user_regs, rsp); OFFSET(UREGS_ss, struct cpu_user_regs, ss); +#ifdef CONFIG_XEN_SHSTK + OFFSET(UREGS_entry_ssp, struct cpu_user_regs, entry_ssp); +#endif DEFINE(UREGS_kernel_sizeof, sizeof(struct cpu_user_regs)); BLANK(); --- a/xen/arch/x86/x86_64/entry.S +++ b/xen/arch/x86/x86_64/entry.S @@ -XXX,XX +XXX,XX @@ FUNC(lstar_enter) pushq $0 BUILD_BUG_ON(TRAP_syscall & 0xff) movb $TRAP_syscall >> 8, EFRAME_entry_vector + 1(%rsp) - SAVE_ALL + SAVE_ALL ssp=0 GET_STACK_END(14) @@ -XXX,XX +XXX,XX @@ FUNC(cstar_enter) pushq $0 BUILD_BUG_ON(TRAP_syscall & 0xff) movb $TRAP_syscall >> 8, EFRAME_entry_vector + 1(%rsp) - SAVE_ALL + SAVE_ALL ssp=0 GET_STACK_END(14) @@ -XXX,XX +XXX,XX @@ LABEL(sysenter_eflags_saved, 0) pushq $0 BUILD_BUG_ON(TRAP_syscall & 0xff) movb $TRAP_syscall >> 8, EFRAME_entry_vector + 1(%rsp) - SAVE_ALL + SAVE_ALL ssp=0 GET_STACK_END(14) @@ -XXX,XX +XXX,XX @@ FUNC(entry_int80) ALTERNATIVE "", clac, X86_FEATURE_XEN_SMAP pushq $0 movb $0x80, EFRAME_entry_vector(%rsp) - SAVE_ALL + SAVE_ALL ssp=0 GET_STACK_END(14) --- a/xen/arch/x86/x86_64/entry-fred.S +++ b/xen/arch/x86/x86_64/entry-fred.S @@ -XXX,XX +XXX,XX @@ /* The Ring3 entry point is required to be 4k aligned. */ FUNC(entry_FRED_R3, 4096) - PUSH_AND_CLEAR_GPRS + PUSH_AND_CLEAR_GPRS ssp=0 mov %rsp, %rdi call entry_from_pv @@ -XXX,XX +XXX,XX @@ LABEL(eretu, 0) END(eretu_exit_to_guest) FUNC(eretu_error_dom_crash) - PUSH_AND_CLEAR_GPRS + PUSH_AND_CLEAR_GPRS ssp=0 sti call asm_domain_crash_synchronous /* Does not return */ END(eretu_error_dom_crash)
With the value recorded on entry there's no need anymore to go hunt for the respective exception frame on the shadow stack. By deriving "ptr" from that field (without any offset), it then ends up pointing one slot lower than before. Therefore all array indexes need incrementing, nicely doing away with all the negative ones. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- Indentation of the prior inner (but not innermost) if()'s body is deliberately left untouched, to aid review. It'll be adjusted in a separate follow-on patch. --- v3: Relax the first BUG_ON(). v2: IS_ENABLED() -> #ifdef. Re-base. --- a/xen/arch/x86/traps.c +++ b/xen/arch/x86/traps.c @@ -XXX,XX +XXX,XX @@ unsigned long get_stack_trace_bottom(uns } } -static unsigned long get_shstk_bottom(unsigned long sp) -{ - /* SAF-11-safe */ - switch ( get_stack_page(sp) ) - { -#ifdef CONFIG_XEN_SHSTK - case 0: return ROUNDUP(sp, IST_SHSTK_SIZE) - sizeof(unsigned long); - case 5: return ROUNDUP(sp, PAGE_SIZE) - sizeof(unsigned long); -#endif - default: return sp - sizeof(unsigned long); - } -} - unsigned long get_stack_dump_bottom(unsigned long sp) { switch ( get_stack_page(sp) ) @@ -XXX,XX +XXX,XX @@ void asmlinkage noreturn do_unhandled_tr static void fixup_exception_return(struct cpu_user_regs *regs, unsigned long fixup, unsigned long stub_ra) { - if ( IS_ENABLED(CONFIG_XEN_SHSTK) ) +#ifdef CONFIG_XEN_SHSTK { - unsigned long ssp, *ptr, *base; + unsigned long ssp = rdssp(); - if ( (ssp = rdssp()) == SSP_NO_SHSTK ) - goto shstk_done; + if ( ssp != SSP_NO_SHSTK ) + { + unsigned long *ptr = _p(regs->entry_ssp); + unsigned long primary_shstk = + (ssp & ~(STACK_SIZE - 1)) + + (PRIMARY_SHSTK_SLOT + 1) * PAGE_SIZE - 8; - ptr = _p(ssp); - base = _p(get_shstk_bottom(ssp)); + BUG_ON((regs->entry_ssp ^ primary_shstk) >> + (PAGE_SHIFT + STACK_ORDER)); - for ( ; ptr < base; ++ptr ) - { /* - * Search for %rip. The shstk currently looks like this: + * The shstk currently looks like this: * * tok [Supervisor token, == &tok | BUSY, only with FRED inactive] * ... [Pointed to by SSP for most exceptions, empty in IST cases] * %cs [== regs->cs] * %rip [== regs->rip] - * SSP [Likely points to 3 slots higher, above %cs] + * SSP [Pointed to by entry_ssp; Likely points to 3 slots + * higher, above %cs] * ... [call tree to this function, likely 2/3 slots] * * and we want to overwrite %rip with fixup. There are two @@ -XXX,XX +XXX,XX @@ static void fixup_exception_return(struc * * Check for both regs->rip and regs->cs matching. */ - if ( ptr[0] == regs->rip && ptr[1] == regs->cs ) - { - unsigned long primary_shstk = - (ssp & ~(STACK_SIZE - 1)) + - (PRIMARY_SHSTK_SLOT + 1) * PAGE_SIZE - 8; + BUG_ON(ptr[1] != regs->rip || ptr[2] != regs->cs); - wrss(fixup, ptr); + { + wrss(fixup, &ptr[1]); if ( !stub_ra ) goto shstk_done; @@ -XXX,XX +XXX,XX @@ static void fixup_exception_return(struc * - if we're on an IST stack, we need to increment the * original SSP. */ - BUG_ON((ptr[-1] ^ primary_shstk) >> PAGE_SHIFT); + BUG_ON((ptr[0] ^ primary_shstk) >> PAGE_SHIFT); if ( (ssp ^ primary_shstk) >> PAGE_SHIFT ) { @@ -XXX,XX +XXX,XX @@ static void fixup_exception_return(struc * addresses actually match. Then increment the interrupted * context's SSP. */ - BUG_ON(stub_ra != *(unsigned long*)ptr[-1]); - wrss(ptr[-1] + 8, &ptr[-1]); + BUG_ON(stub_ra != *(unsigned long*)ptr[0]); + wrss(ptr[0] + 8, &ptr[0]); goto shstk_done; } /* Make sure the two return addresses actually match. */ - BUG_ON(stub_ra != ptr[2]); + BUG_ON(stub_ra != ptr[3]); /* Move exception frame, updating SSP there. */ - wrss(ptr[1], &ptr[2]); /* %cs */ - wrss(ptr[0], &ptr[1]); /* %rip */ - wrss(ptr[-1] + 8, &ptr[0]); /* SSP */ + wrss(ptr[2], &ptr[3]); /* %cs */ + wrss(ptr[1], &ptr[2]); /* %rip */ + wrss(ptr[0] + 8, &ptr[1]); /* SSP */ /* Move all newer entries. */ - while ( --ptr != _p(ssp) ) - wrss(ptr[-1], &ptr[0]); + while ( ptr-- != _p(ssp) ) + wrss(ptr[0], &ptr[1]); /* Finally account for our own stack having shifted up. */ asm volatile ( "incsspd %0" :: "r" (2) ); - - goto shstk_done; } } - - /* - * We failed to locate and fix up the shadow IRET frame. This could - * be due to shadow stack corruption, or bad logic above. We cannot - * continue executing the interrupted context. - */ - BUG(); - } shstk_done: +#endif /* CONFIG_XEN_SHSTK */ /* Fixup the regular stack. */ regs->rip = fixup;
Shadow stacks contain little more than return addresses, and they in particular allow precise call traces also with FRAME_POINTER=n: (XEN) Xen call trace: (XEN) [<ffff82d04032d730>] R extable.c#search_one_extable+0x70/0x73 (XEN) [<ffff82d04032d802>] C search_exception_table+0xc2/0x177 (XEN) [<ffff82d040358378>] C traps.c#extable_fixup.isra.0+0x18/0x6c (XEN) [<ffff82d040358e3b>] C do_invalid_op+0xab/0x106 (XEN) [<ffff82d040201d98>] C x86_64/entry.S#handle_exception_saved+0x88/0xf4 (XEN) [<ffff82d07fffe044>] E ffff82d07fffe044 (XEN) [<ffff82d040412db0>] C stub_selftest+0xd0/0x168 (XEN) [<ffff82d0403508d6>] C setup.c#init_done+0x116/0x15a as opposed to this counterpart (earlier during the same boot, before CET is enabled): (XEN) Xen call trace: (XEN) [<ffff82d04032d730>] R extable.c#search_one_extable+0x70/0x73 (XEN) [<ffff82d04032d802>] S search_exception_table+0xc2/0x177 (XEN) [<ffff82d040358378>] S traps.c#extable_fixup.isra.0+0x18/0x6c (XEN) [<ffff82d040358e3b>] S do_invalid_op+0xab/0x106 (XEN) [<ffff82d040201d98>] S x86_64/entry.S#handle_exception_saved+0x88/0xf4 (XEN) [<ffff82d040412db0>] S stub_selftest+0xd0/0x168 (XEN) [<ffff82d0403d0cc9>] S do_initcalls+0x29/0x38 (XEN) [<ffff82d04041adf2>] S __start_xen+0x1c72/0x2235 (XEN) [<ffff82d040288a57>] S __high_start+0xb7/0xc0 (note the entirely missing entry for the stub itself [1]; sadly there are no stray entries there). [1] Arguably we could teach FRAME_POINTER=n traces to recognize stubs as well. But not FRAME_POINTER=y ones. In fact, what's missing there isn't the stub itself, but (of course) its immediate caller. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- While the 'E' for exception frames is probably okay, I'm not overly happy with the 'C' (for CET). I would have preferred 'S' (for shadow), but we use that character already. As an alternative to suppressing output for the top level exception frame, adding the new code ahead of the 'R' output line (and then also ahead of the stack top read) could be considered. Quite likely a number of other uses of is_active_kernel_text() also want amending with in_stub(). --- v3: Correct "link to other shadow stack" check. Don't log a line for the (impossible) PV case. Add example stack trace to description. v2: IS_ENABLED() -> #ifdef. Re-base. --- a/xen/arch/x86/traps.c +++ b/xen/arch/x86/traps.c @@ -XXX,XX +XXX,XX @@ #include <asm/shared.h> #include <asm/shstk.h> #include <asm/smp.h> +#include <asm/stubs.h> #include <asm/system.h> #include <asm/traps.h> #include <asm/uaccess.h> @@ -XXX,XX +XXX,XX @@ unsigned long get_stack_dump_bottom(unsi } } +#ifdef CONFIG_XEN_SHSTK +static bool in_stub(unsigned long addr) +{ + return !((this_cpu(stubs.addr) ^ addr) >> STUB_BUF_SHIFT); +} +#endif + #if !defined(CONFIG_FRAME_POINTER) /* @@ -XXX,XX +XXX,XX @@ static void show_trace(const struct cpu_ !is_active_kernel_text(tos) ) printk(" [<%p>] R %pS\n", _p(regs->rip), _p(regs->rip)); +#ifdef CONFIG_XEN_SHSTK + if ( rdssp() != SSP_NO_SHSTK ) + { + const unsigned long *ptr = _p(regs->entry_ssp); + unsigned int n; + + for ( n = 0; (unsigned long)ptr & (PAGE_SIZE - sizeof(*ptr)); ++n ) + { + unsigned long val = *ptr; + + if ( is_active_kernel_text(val) || in_stub(val) ) + { + /* Normal return address entry. */ + printk(" [<%p>] C %pS\n", _p(val), _p(val)); + ++ptr; + } + else if ( !((val ^ (unsigned long)ptr) >> + (PAGE_SHIFT + STACK_ORDER)) ) + { + if ( val & (sizeof(val) - 1) ) + { + /* Most likely a supervisor token. */ + break; + } + + /* + * Ought to be a hypervisor interruption frame. But don't + * (re)log the current frame's %rip. + */ + if ( n || ptr[1] != regs->rip ) + printk(" [<%p>] E %pS\n", _p(ptr[1]), _p(ptr[1])); + ptr = _p(val); + } + else /* Bogus. */ + break; + } + + /* Fall back to legacy stack trace if nothing was logged at all. */ + if ( n ) + return; + } +#endif /* CONFIG_XEN_SHSTK */ + if ( fault ) { printk(" [Fault on access]\n");