x86/vmx: Revert "x86/VMX: sanitize rIP before re-entering guest"
PATCH 1/1 x86/vmx: Revert "x86/VMX: sanitize rIP before re-entering guest"
At the time of XSA-170, the x86 instruction emulator was genuinely broken. It would load arbitrary values into %rip and putting a check here probably was the best stopgap security fix. It should have been reverted following c/s 81d3a0b26c1 "x86emul: limit-check branch targets" which corrected the emulator behaviour.
However, everyone involved in XSA-170, myself included, failed to read the SDM correctly. On the subject of %rip consistency checks, the SDM stated:
If the processor supports N < 64 linear-address bits, bits 63:N must be identical
A non-canonical %rip (and SSP more recently) is an explicitly legal state in x86, and the VMEntry consistency checks are intentionally off-by-one from a regular canonical check.
The consequence of this bug is that Xen will currently take a legal x86 state which would successfully VMEnter, and corrupt it into having non-architectural behaviour.
Furthermore, in the time this bugfix has been pending in public, I successfully persuaded Intel to clarify the SDM, adding the following clarification:
The guest RIP value is not required to be canonical; the value of bit N-1 may differ from that of bit N.
---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>
CC: Jun Nakajima <[email protected]>
CC: Kevin Tian <[email protected]>
Rewrite the commit message, but no change to the fix.
Fun fact... LAM almost required this to be reverted in order to support, but the penultimate draft of the spec backed down and made LAM only apply to data accesses, not instruction fetches.
--- xen/arch/x86/hvm/vmx/vmx.c | 34 +--------------------------------- 1 file changed, 1 insertion(+), 33 deletions(-)
xen/arch/x86/hvm/vmx/vmx.c +1 -33
@@ -4035,7 +4035,7 @@ static void undo_nmis_unblocked_by_iret(void)void vmx_vmexit_handler (struct cpu_user_regs *regs) {unsigned long exit_qualification, exit_reason, idtv_info, intr_info =0 ; -unsigned int vector =0 , mode; +unsigned int vector =0 ;struct vcpu *v = current;struct domain *currd = v->domain;
@@ -4730,38 +4730,6 @@ void vmx_vmexit_handler(struct cpu_user_regs *regs) out:if (nestedhvm_vcpu_in_guestmode (v) )nvmx_idtv_handling (); - -/* -* VM entry will fail (causing the guest to get crashed) if rIP (and -* rFLAGS, but we don't have an issue there) doesn't meet certain -* criteria. As we must not allow less than fully privileged mode to have -* such an effect on the domain, we correct rIP in that case (accepting -* this not being architecturally correct behavior, as the injected #GP -* fault will then not see the correct [invalid] return address). -* And since we know the guest will crash, we crash it right away if it -* already is in most privileged mode. -*/ - mode =vmx_guest_x86_mode (v); -if ( mode ==8 ? !is_canonical_address (regs->rip) - : regs->rip != regs->eip ) - { -gprintk (XENLOG_WARNING ,"Bad rIP %lx for mode %u\n" , regs->rip, mode); - -if (vmx_get_cpl () ) - { -__vmread (VM_ENTRY_INTR_INFO , &intr_info); -if ( !(intr_info &INTR_INFO_VALID_MASK ) ) -hvm_inject_hw_exception (X86_EXC_GP ,0 ); -/* Need to fix rIP nevertheless. */ -if ( mode ==8 ) - regs->rip = (long )(regs->rip << (64 -VADDR_BITS )) >> - (64 -VADDR_BITS ); -else - regs->rip = regs->eip; - } -else -domain_crash (v->domain); - } }static void lbr_tsx_fixup (void )
Signature · 1 line
2.30.2
Quoted 24 lines · Andrew Cooper
At the time of XSA-170, the x86 instruction emulator was genuinely broken. It would load arbitrary values into %rip and putting a check here probably was the best stopgap security fix. It should have been reverted following c/s
81d3a0b26c1"x86emul: limit-check branch targets" which corrected the emulator behaviour.However, everyone involved in XSA-170, myself included, failed to read the SDM correctly. On the subject of %rip consistency checks, the SDM stated:
If the processor supports N < 64 linear-address bits, bits 63:N must be identical
A non-canonical %rip (and SSP more recently) is an explicitly legal state in x86, and the VMEntry consistency checks are intentionally off-by-one from a regular canonical check.
The consequence of this bug is that Xen will currently take a legal x86 state which would successfully VMEnter, and corrupt it into having non-architectural behaviour.
Furthermore, in the time this bugfix has been pending in public, I successfully persuaded Intel to clarify the SDM, adding the following clarification:
Quoted 5 lines
The guest RIP value is not required to be canonical; the value of bit N-1 may differ from that of bit N.
Fixes ffbbfda377 ("x86/VMX: sanitize rIP before re-entering guest")Signed-off-by Andrew Cooper <[email protected]>
I am kind of okay with such a full revert, but I'd consider it quite helpful if the description made clear why the alternative of instead doing the spec mandated check isn't necessary / useful. The emulator having gained respective checking is only part of the reason for this, aiui. Plus bugs may be introduced into the emulator again, where the checking here could be a guard against needing to issue an XSA in such a case.
Jan
Quoted 20 lines · Andrew Cooper
At the time of XSA-170, the x86 instruction emulator was genuinely broken. It would load arbitrary values into %rip and putting a check here probably was the best stopgap security fix. It should have been reverted following c/s
81d3a0b26c1"x86emul: limit-check branch targets" which corrected the emulator behaviour.However, everyone involved in XSA-170, myself included, failed to read the SDM correctly. On the subject of %rip consistency checks, the SDM stated:
If the processor supports N < 64 linear-address bits, bits 63:N must be identical
A non-canonical %rip (and SSP more recently) is an explicitly legal state in x86, and the VMEntry consistency checks are intentionally off-by-one from a regular canonical check.
The consequence of this bug is that Xen will currently take a legal x86 state which would successfully VMEnter, and corrupt it into having non-architectural behaviour.
Quoted 8 lines
Furthermore, in the time this bugfix has been pending in public, I successfully persuaded Intel to clarify the SDM, adding the following clarification:
The guest RIP value is not required to be canonical; the value of bit N-1 may differ from that of bit N.
Fixes ffbbfda377 ("x86/VMX: sanitize rIP before re-entering guest")
I think the fixes tag should likely be "x86emul: limit-check branch targets", since it's that commit that missed the revert done here?
Quoted 1 line
Signed-off-by Andrew Cooper <[email protected]>
Thanks, Roger.
Quoted 25 lines · Roger Pau Monné
Quoted 24 lines · Andrew Cooper
At the time of XSA-170, the x86 instruction emulator was genuinely broken. It would load arbitrary values into %rip and putting a check here probably was the best stopgap security fix. It should have been reverted following c/s
81d3a0b26c1"x86emul: limit-check branch targets" which corrected the emulator behaviour.However, everyone involved in XSA-170, myself included, failed to read the SDM correctly. On the subject of %rip consistency checks, the SDM stated:
If the processor supports N < 64 linear-address bits, bits 63:N must be identical
A non-canonical %rip (and SSP more recently) is an explicitly legal state in x86, and the VMEntry consistency checks are intentionally off-by-one from a regular canonical check.
The consequence of this bug is that Xen will currently take a legal x86 state which would successfully VMEnter, and corrupt it into having non-architectural behaviour.
Furthermore, in the time this bugfix has been pending in public, I successfully persuaded Intel to clarify the SDM, adding the following clarification:
Quoted 6 lines
Quoted 4 lines
The guest RIP value is not required to be canonical; the value of bit N-1 may differ from that of bit N.
Fixes ffbbfda377 ("x86/VMX: sanitize rIP before re-entering guest")I think the fixes tag should likely be "x86emul: limit-check branch targets", since it's that commit that missed the revert done here?
Well, not really. ffbbfda377 really does have a bug, irrespective of the changes in the emulator.
The presence of 81d3a0b26c1 is why this bugfix is a full revert of ffbbfda377, and not just an off-by-1 adjustment.
Quoted 3 lines
Quoted 2 lines
Signed-off-by Andrew Cooper <[email protected]>Acked-by Roger Pau Monné <[email protected]>
Thanks.
~Andrew
Quoted 30 lines · Andrew Cooper
Quoted 29 lines · Roger Pau Monné
Quoted 28 lines · Andrew Cooper
At the time of XSA-170, the x86 instruction emulator was genuinely broken. It would load arbitrary values into %rip and putting a check here probably was the best stopgap security fix. It should have been reverted following c/s
81d3a0b26c1"x86emul: limit-check branch targets" which corrected the emulator behaviour.However, everyone involved in XSA-170, myself included, failed to read the SDM correctly. On the subject of %rip consistency checks, the SDM stated:
If the processor supports N < 64 linear-address bits, bits 63:N must be identical
A non-canonical %rip (and SSP more recently) is an explicitly legal state in x86, and the VMEntry consistency checks are intentionally off-by-one from a regular canonical check.
The consequence of this bug is that Xen will currently take a legal x86 state which would successfully VMEnter, and corrupt it into having non-architectural behaviour.
Furthermore, in the time this bugfix has been pending in public, I successfully persuaded Intel to clarify the SDM, adding the following clarification:
The guest RIP value is not required to be canonical; the value of bit N-1 may differ from that of bit N.
Fixes ffbbfda377 ("x86/VMX: sanitize rIP before re-entering guest")
Quoted 8 lines
Quoted 2 lines
I think the fixes tag should likely be "x86emul: limit-check branch targets", since it's that commit that missed the revert done here?
Well, not really.
ffbbfda377really does have a bug, irrespective of the changes in the emulator.The presence of
81d3a0b26c1is why this bugfix is a full revert offfbbfda377, and not just an off-by-1 adjustment.
Right, but taking this patch without also having 81d3a0b26c1 will lead to a vulnerable system, hence why I think the dependency would better be on 81d3a0b26c1.
Anyway, I don't think it's worth arguing over, so if you want to leave it as-is I won't object.
Thanks, Roger.
Quoted 36 lines · Roger Pau Monné
Quoted 35 lines · Andrew Cooper
Quoted 25 lines · Roger Pau Monné
Quoted 24 lines · Andrew Cooper
At the time of XSA-170, the x86 instruction emulator was genuinely broken. It would load arbitrary values into %rip and putting a check here probably was the best stopgap security fix. It should have been reverted following c/s
81d3a0b26c1"x86emul: limit-check branch targets" which corrected the emulator behaviour.However, everyone involved in XSA-170, myself included, failed to read the SDM correctly. On the subject of %rip consistency checks, the SDM stated:
If the processor supports N < 64 linear-address bits, bits 63:N must be identical
A non-canonical %rip (and SSP more recently) is an explicitly legal state in x86, and the VMEntry consistency checks are intentionally off-by-one from a regular canonical check.
The consequence of this bug is that Xen will currently take a legal x86 state which would successfully VMEnter, and corrupt it into having non-architectural behaviour.
Furthermore, in the time this bugfix has been pending in public, I successfully persuaded Intel to clarify the SDM, adding the following clarification:
Quoted 6 lines
Quoted 4 lines
The guest RIP value is not required to be canonical; the value of bit N-1 may differ from that of bit N.
Fixes ffbbfda377 ("x86/VMX: sanitize rIP before re-entering guest")I think the fixes tag should likely be "x86emul: limit-check branch targets", since it's that commit that missed the revert done here?
Well, not really.
ffbbfda377really does have a bug, irrespective of the changes in the emulator.
Quoted 8 lines
Quoted 2 lines
The presence of
81d3a0b26c1is why this bugfix is a full revert offfbbfda377, and not just an off-by-1 adjustment.Right, but taking this patch without also having
81d3a0b26c1will lead to a vulnerable system, hence why I think the dependency would better be on81d3a0b26c1.Anyway, I don't think it's worth arguing over, so if you want to leave it as-is I won't object.
We don't really have a depends-on tag, or only-safe-with or whatever, and a change like that is stretching the definition of Fixes IMO.
81d3a0b26c1 is more than 7 years old now, so there's not going to be a practical problem backporting. For everything else, I expect people to read the commit message and apply common sense.
~Andrew
Quoted 31 lines
From: Andrew Cooper <[email protected]>
Sent: Thursday, April 6, 2023 5:53 AMAt the time of XSA-170, the x86 instruction emulator was genuinely broken. It
would load arbitrary values into %rip and putting a check here probably was the best stopgap security fix. It should have been reverted following c/s81d3a0b26c1"x86emul: limit-check branch targets" which corrected the emulator
behaviour.However, everyone involved in XSA-170, myself included, failed to read the SDM
correctly. On the subject of %rip consistency checks, the SDM stated:If the processor supports N < 64 linear-address bits, bits 63:N must be identical
A non-canonical %rip (and SSP more recently) is an explicitly legal state in x86, and the VMEntry consistency checks are intentionally off-by-one from a regular canonical check.
The consequence of this bug is that Xen will currently take a legal x86 state which would successfully VMEnter, and corrupt it into having non- architectural
behaviour.Furthermore, in the time this bugfix has been pending in public, I successfully persuaded Intel to clarify the SDM, adding the following clarification:
Quoted 5 lines
The guest RIP value is not required to be canonical; the value of bit N-1 may differ from that of bit N.
Fixes ffbbfda377 ("x86/VMX: sanitize rIP before re-entering guest")Signed-off-by Andrew Cooper <[email protected]>