KVM: Set vcpu->preempted/ready iff scheduled out while running
v3 cover KVM: Set vcpu->preempted/ready iff scheduled out while running
This series changes KVM to mark a vCPU as preempted/ready if-and-only-if it's scheduled out while running. i.e. Do not mark a vCPU preempted/ready if it's scheduled out during a non-KVM_RUN ioctl() or when userspace is doing KVM_RUN with immediate_exit=true.
This is a logical extension of commit 54aa83c90198 ("KVM: x86: do not set st->preempted when going back to user space"), which stopped marking a vCPU as preempted when returning to userspace. But if userspace invokes a KVM vCPU ioctl() that gets preempted, the vCPU will be marked preempted/ready. This is arguably incorrect behavior since the vCPU was not actually preempted while the guest was running, it was preempted while doing something on behalf of userspace.
In practice, this avoids KVM dirtying guest memory via the steal time page after userspace has paused vCPUs, e.g. for Live Migration, which allows userspace to collect the final dirty bitmap before or in parallel with saving vCPU state without having to worry about saving vCPU state triggering writes to guest memory.
Patch 1 introduces vcpu->wants_to_run to allow KVM to detect when a vCPU is in its core run loop.
Patch 2 renames immediated_exit to immediated_exit__unsafe within KVM to ensure that any new references get extra scrutiny.
Patch 3 perform leverages vcpu->wants_to_run to contrain when vcpu->preempted and vcpu->ready are set.
v3:
- Use READ_ONCE() to read immediate_exit [Sean]
- Replace use of immediate_exit with !wants_to_run to avoid TOCTOU [Sean]
- Hide/Rename immediate_exit in KVM to harden against TOCTOU bugs [Sean]
v2: https://lore.kernel.org/kvm/20240307163541.92138-1-dmatlack at google.com/
- Drop Google-specific "PRODKERNEL: " shortlog prefix [me]
v1: https://lore.kernel.org/kvm/20231218185850.1659570-1-dmatlack at google.com/
David Matlack (3):
KVM: Introduce vcpu->wants_to_run
KVM: Ensure new code that references immediate_exit gets extra
scrutiny
KVM: Mark a vCPU as preempted/ready iff it's scheduled out while
running
arch/arm64/kvm/arm.c | 2 +- arch/loongarch/kvm/vcpu.c | 2 +- arch/mips/kvm/mips.c | 2 +- arch/powerpc/kvm/powerpc.c | 2 +- arch/riscv/kvm/vcpu.c | 2 +- arch/s390/kvm/kvm-s390.c | 2 +- arch/x86/kvm/x86.c | 4 ++-- include/linux/kvm_host.h | 1 + include/uapi/linux/kvm.h | 15 ++++++++++++++- virt/kvm/kvm_main.c | 5 ++++- 10 files changed, 27 insertions(+), 10 deletions(-)
base-commit: 296655d9bf272cfdd9d2211d099bcb8a61b93037
Signature · 1 line
2.45.0.rc1.225.g2a3ae87e7f-goog
PATCH v3 1/3 KVM: Introduce vcpu->wants_to_run
Introduce vcpu->wants_to_run to indicate when a vCPU is in its core run loop, i.e. when the vCPU is running the KVM_RUN ioctl and immediate_exit was not set.
Replace all references to vcpu->run->immediate_exit with !vcpu->wants_to_run to avoid TOCTOU races with userspace. For example, a malicious userspace could invoked KVM_RUN with immediate_exit=true and then after KVM reads it to set wants_to_run=false, flip it to false. This would result in the vCPU running in KVM_RUN with wants_to_run=false. This wouldn't cause any real bugs today but is a dangerous landmine.
--- arch/arm64/kvm/arm.c | 2 +- arch/loongarch/kvm/vcpu.c | 2 +- arch/mips/kvm/mips.c | 2 +- arch/powerpc/kvm/powerpc.c | 2 +- arch/riscv/kvm/vcpu.c | 2 +- arch/s390/kvm/kvm-s390.c | 2 +- arch/x86/kvm/x86.c | 4 ++-- include/linux/kvm_host.h | 1 + virt/kvm/kvm_main.c | 3 +++ 9 files changed, 12 insertions(+), 8 deletions(-)
arch/arm64/kvm/arm.c +1 -1
@@ -986,7 +986,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)vcpu_load (vcpu); -if (run->immediate_exit) { +if (!vcpu->wants_to_run) { ret = -EINTR ; goto out; }
arch/loongarch/kvm/vcpu.c +1 -1
@@ -1163,7 +1163,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)kvm_complete_iocsr_read (vcpu, run); } -if (run->immediate_exit) +if (!vcpu->wants_to_run)return r;/* Clear exit_reason */
arch/mips/kvm/mips.c +1 -1
@@ -436,7 +436,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) vcpu->mmio_needed =0 ; } -if (vcpu->run->immediate_exit) +if (!vcpu->wants_to_run) goto out;lose_fpu (1 );
arch/powerpc/kvm/powerpc.c +1 -1
@@ -1852,7 +1852,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)kvm_sigset_activate (vcpu); -if (run->immediate_exit) +if (!vcpu->wants_to_run) r = -EINTR ;else r =kvmppc_vcpu_run (vcpu);
arch/riscv/kvm/vcpu.c +1 -1
@@ -711,7 +711,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)return ret; } -if (run->immediate_exit) { +if (!vcpu->wants_to_run) {kvm_vcpu_srcu_read_unlock (vcpu);return -EINTR ; }
arch/s390/kvm/kvm-s390.c +1 -1
@@ -5033,7 +5033,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)if (vcpu->kvm->arch.pv.dumping)return -EINVAL ; -if (kvm_run->immediate_exit) +if (!vcpu->wants_to_run)return -EINTR ; if (kvm_run->kvm_valid_regs & ~KVM_SYNC_S390_VALID_FIELDS ||
arch/x86/kvm/x86.c +2 -2
@@ -11396,7 +11396,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)kvm_vcpu_srcu_read_lock (vcpu);if (unlikely (vcpu->arch.mp_state ==KVM_MP_STATE_UNINITIALIZED )) { -if (kvm_run->immediate_exit) { +if (!vcpu->wants_to_run) { r = -EINTR ; goto out; }
@@ -11474,7 +11474,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)WARN_ON_ONCE (vcpu->mmio_needed); } -if (kvm_run->immediate_exit) { +if (!vcpu->wants_to_run) { r = -EINTR ; goto out; }
include/linux/kvm_host.h +1 -0
@@ -380,6 +380,7 @@ struct kvm_vcpu { bool dy_eligible; } spin_loop;#endif +bool wants_to_run;bool preempted;bool ready;struct kvm_vcpu_arch arch;
virt/kvm/kvm_main.c +3 -0
@@ -4425,7 +4425,10 @@ static long kvm_vcpu_ioctl(struct file *filp,synchronize_rcu ();put_pid (oldpid); } + vcpu->wants_to_run = !READ_ONCE (vcpu->run->immediate_exit); r =kvm_arch_vcpu_ioctl_run (vcpu); + vcpu->wants_to_run = false; +trace_kvm_userspace_exit (vcpu->run->exit_reason, r);break ; }
Signature · 1 line
2.45.0.rc1.225.g2a3ae87e7f-goog
PATCH v3 2/3 KVM: Ensure new code that references immediate_exit gets extra scrutiny
In reply to David MatlackEnsure that any new KVM code that references immediate_exit gets extra scrutiny by renaming it to immediate_exit__unsafe in kernel code.
All fields in struct kvm_run are subject to TOCTOU races since they are mapped into userspace, which may be malicious or buggy. To protect KVM, this commit introduces a new macro that appends __unsafe to field names in struct kvm_run, hinting to developers and reviewers that accessing this field must be done carefully.
Apply the new macro to immediate_exit, since userspace can make immediate_exit inconsistent with vcpu->wants_to_run, i.e. accessing immediate_exit directly could lead to unexpected bugs in the future.
--- include/uapi/linux/kvm.h | 15 ++++++++++++++- virt/kvm/kvm_main.c | 2 +- 2 files changed, 15 insertions(+), 2 deletions(-)
include/uapi/linux/kvm.h +14 -1
@@ -192,11 +192,24 @@ struct kvm_xen_exit {/* Flags that describe what fields in emulation_failure hold valid data. */ #define KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES (1ULL << 0) +/* +* struct kvm_run can be modified by userspace@any time, so KVM must be +* careful to avoid TOCTOU bugs. In order to protect KVM, HINT_UNSAFE_IN_KVM() +* renames fields in struct kvm_run from <symbol> to <symbol>__unsafe when +* compiled into the kernel, ensuring that any use within KVM is obvious and +* gets extra scrutiny. +*/ +#ifdef __KERNEL__ +#define HINT_UNSAFE_IN_KVM (_symbol) _symbol##__unsafe +#else +#define HINT_UNSAFE_IN_KVM (_symbol) _symbol +#endif +/* for KVM_RUN, returned by mmap(vcpu_fd, offset=0) */ struct kvm_run {/* in */ __u8 request_interrupt_window; -__u8 immediate_exit; +__u8 HINT_UNSAFE_IN_KVM(immediate_exit );__u8 padding1[6 ];/* out */
virt/kvm/kvm_main.c +1 -1
@@ -4425,7 +4425,7 @@ static long kvm_vcpu_ioctl(struct file *filp,synchronize_rcu ();put_pid (oldpid); } - vcpu->wants_to_run = !READ_ONCE (vcpu->run->immediate_exit); + vcpu->wants_to_run = !READ_ONCE (vcpu->run->immediate_exit__unsafe); r =kvm_arch_vcpu_ioctl_run (vcpu); vcpu->wants_to_run = false;
Signature · 1 line
2.45.0.rc1.225.g2a3ae87e7f-goog
PATCH v3 3/3 KVM: Mark a vCPU as preempted/ready iff it's scheduled out while running
In reply to David MatlackMark a vCPU as preempted/ready if-and-only-if it's scheduled out while running. i.e. Do not mark a vCPU preempted/ready if it's scheduled out during a non-KVM_RUN ioctl() or when userspace is doing KVM_RUN with immediate_exit.
Commit 54aa83c90198 ("KVM: x86: do not set st->preempted when going back to user space") stopped marking a vCPU as preempted when returning to userspace, but if userspace then invokes a KVM vCPU ioctl() that gets preempted, the vCPU will be marked preempted/ready. This is arguably incorrect behavior since the vCPU was not actually preempted while the guest was running, it was preempted while doing something on behalf of userspace.
This commit also avoids KVM dirtying guest memory after userspace has paused vCPUs, e.g. for Live Migration, which allows userspace to collect the final dirty bitmap before or in parallel with saving vCPU state without having to worry about saving vCPU state triggering writes to guest memory.
--- virt/kvm/kvm_main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
virt/kvm/kvm_main.c +1 -1
@@ -6302,7 +6302,7 @@ static void kvm_sched_out(struct preempt_notifier *pn, {struct kvm_vcpu *vcpu =preempt_notifier_to_vcpu (pn); -if (current->on_rq) { +if (current->on_rq && vcpu->wants_to_run) {WRITE_ONCE (vcpu->preempted, true);WRITE_ONCE (vcpu->ready, true); }
Signature · 1 line
2.45.0.rc1.225.g2a3ae87e7f-goog
Quoted 13 lines · David Matlack
This series changes KVM to mark a vCPU as preempted/ready if-and-only-if it's scheduled out while running. i.e. Do not mark a vCPU preempted/ready if it's scheduled out during a non-KVM_RUN
ioctl()or when userspace is doingKVM_RUNwith immediate_exit=true.This is a logical extension of commit
54aa83c90198("KVM: x86: do not setst->preemptedwhen going back to user space"), which stopped marking a vCPU as preempted when returning to userspace. But if userspace invokes a KVM vCPUioctl()that gets preempted, the vCPU will be marked preempted/ready. This is arguably incorrect behavior since the vCPU was not actually preempted while the guest was running, it was preempted while doing something on behalf of userspace.
Quoted 1 line
[...]
Applied to kvm-x86 generic, with minor changelog tweaks (me thinks you've been away from upstream too long ;-) ). Thanks!
[1/3] KVM: Introduce vcpu->wants_to_run
https://github.com/kvm-x86/linux/commit/a6816314af57 [2/3] KVM: Ensure new code that references immediate_exit gets extra scrutiny https://github.com/kvm-x86/linux/commit/4b23e0c199b2 [3/3] KVM: Mark a vCPU as preempted/ready iff it's scheduled out while running https://github.com/kvm-x86/linux/commit/118964562969
Quoted 15 lines · On Tue, Jun 18, 2024 at 2:41?PM Sean Christopherson <[email protected]> wrote:
Quoted 13 lines · David Matlack
This series changes KVM to mark a vCPU as preempted/ready if-and-only-if it's scheduled out while running. i.e. Do not mark a vCPU preempted/ready if it's scheduled out during a non-KVM_RUN
ioctl()or when userspace is doingKVM_RUNwith immediate_exit=true.This is a logical extension of commit
54aa83c90198("KVM: x86: do not setst->preemptedwhen going back to user space"), which stopped marking a vCPU as preempted when returning to userspace. But if userspace invokes a KVM vCPUioctl()that gets preempted, the vCPU will be marked preempted/ready. This is arguably incorrect behavior since the vCPU was not actually preempted while the guest was running, it was preempted while doing something on behalf of userspace.
Quoted 4 lines
Quoted 1 line
[...]
Applied to kvm-x86 generic, with minor changelog tweaks (me thinks you've been away from upstream too long ;-) ). Thanks!
Thanks for the cleanups. Looks like you replaced "[Tt]his commit" throughout. Anything else (so I can avoid the same mistakes in the future)?
Quoted 10 lines
[1/3] KVM: Introduce
vcpu->wants_to_runhttps://github.com/kvm-x86/linux/commit/a6816314af57 [2/3] KVM: Ensure new code that references immediate_exit gets extra scrutiny https://github.com/kvm-x86/linux/commit/4b23e0c199b2 [3/3] KVM: Mark a vCPU as preempted/ready iff it's scheduled out while running https://github.com/kvm-x86/linux/commit/118964562969
Sean Christopherson
maintainerQuoted 16 lines · David Matlack
Quoted 15 lines · On Tue, Jun 18, 2024 at 2:41?PM Sean Christopherson <[email protected]> wrote:
Quoted 13 lines · David Matlack
This series changes KVM to mark a vCPU as preempted/ready if-and-only-if it's scheduled out while running. i.e. Do not mark a vCPU preempted/ready if it's scheduled out during a non-KVM_RUN
ioctl()or when userspace is doingKVM_RUNwith immediate_exit=true.This is a logical extension of commit
54aa83c90198("KVM: x86: do not setst->preemptedwhen going back to user space"), which stopped marking a vCPU as preempted when returning to userspace. But if userspace invokes a KVM vCPUioctl()that gets preempted, the vCPU will be marked preempted/ready. This is arguably incorrect behavior since the vCPU was not actually preempted while the guest was running, it was preempted while doing something on behalf of userspace.
Quoted 8 lines
Quoted 4 lines
Quoted 1 line
[...]
Applied to kvm-x86 generic, with minor changelog tweaks (me thinks you've been away from upstream too long ;-) ). Thanks!
Thanks for the cleanups. Looks like you replaced "[Tt]his commit" throughout. Anything else (so I can avoid the same mistakes in the future)?
I don't think so? The "This commit" stuff is the only thing that I remember, so any other tweaks can't be that important :-)